From ee7960bd8e54e4ca68e51ab11a423a6a9d05df5b Mon Sep 17 00:00:00 2001
From: Vladimir Mandic <mandic00@live.com>
Date: Fri, 16 Oct 2020 15:04:51 -0400
Subject: [PATCH] updated examples plus  bugfixes

---
 README.md                      |  82 ++++++++++++++++---------
 assets/sample1.jpg             | Bin 0 -> 144516 bytes
 assets/sample2.jpg             | Bin 0 -> 181758 bytes
 assets/sample3.jpg             | Bin 0 -> 221049 bytes
 assets/sample4.jpg             | Bin 0 -> 210824 bytes
 assets/sample5.jpg             | Bin 0 -> 166400 bytes
 assets/sample6.jpg             | Bin 0 -> 302445 bytes
 assets/screenshot1.jpg         | Bin 0 -> 369127 bytes
 assets/screenshot2.jpg         | Bin 0 -> 228796 bytes
 demo/demo-esm.html             |  21 +++----
 demo/demo-esm.js               | 105 +++++++++++++++++++++++----------
 demo/sample-image.jpg          | Bin 102644 -> 0 bytes
 demo/sample-video.jpg          | Bin 84394 -> 0 bytes
 dist/human-nobundle.cjs        |  26 ++++----
 dist/human-nobundle.cjs.map    |   4 +-
 dist/human.cjs                 |  94 ++++++++++++++---------------
 dist/human.cjs.map             |   4 +-
 dist/human.esm-nobundle.js     |   2 +-
 dist/human.esm-nobundle.js.map |   4 +-
 dist/human.esm.js              |  94 ++++++++++++++---------------
 dist/human.esm.js.map          |   4 +-
 dist/human.js                  |  94 ++++++++++++++---------------
 dist/human.js.map              |   4 +-
 src/config.js                  |  76 ++++++++++++++----------
 src/index.js                   |  13 ++--
 src/ssrnet/ssrnet.js           |  13 ++--
 26 files changed, 367 insertions(+), 273 deletions(-)
 create mode 100644 assets/sample1.jpg
 create mode 100644 assets/sample2.jpg
 create mode 100644 assets/sample3.jpg
 create mode 100644 assets/sample4.jpg
 create mode 100644 assets/sample5.jpg
 create mode 100644 assets/sample6.jpg
 create mode 100644 assets/screenshot1.jpg
 create mode 100644 assets/screenshot2.jpg
 delete mode 100644 demo/sample-image.jpg
 delete mode 100644 demo/sample-video.jpg

diff --git a/README.md b/README.md
index 050e6b8f..9be06d20 100644
--- a/README.md
+++ b/README.md
@@ -16,11 +16,13 @@ Compatible with Browser, WebWorker and NodeJS execution!
 
 <hr>
 
-**Example using static image:**  
-![Example Using Image](demo/sample-image.jpg)
+## Examples
 
-**Example using webcam:**  
-![Example Using WebCam](demo/sample-video.jpg)
+**Using static images:**  
+![Example Using Image](assets/screenshot1.jpg)
+
+**Using webcam:**  
+![Example Using WebCam](assets/screenshot2.jpg)
 
 <hr>
 
@@ -211,59 +213,85 @@ Below is output of `human.defaults` object
 Any property can be overriden by passing user object during `human.detect()`  
 Note that user object and default configuration are merged using deep-merge, so you do not need to redefine entire configuration  
 
+Configurtion object is large, but typically you only need to modify few values:
+
+- `enabled`: Choose which models to use
+- `skipFrames`: Must be set to 0 for static images
+- `modelPath`: Update as needed to reflect your application's relative path
+
+
 ```js
-human.defaults = {
-  console: true,            // enable debugging output to console
-  backend: 'webgl',         // select tfjs backend to use
+export default {
+  backend: 'webgl',          // select tfjs backend to use
+  console: true,             // enable debugging output to console
   face: {
-    enabled: true,          // controls if specified modul is enabled (note: module is not loaded until it is required)
+    enabled: true,           // controls if specified modul is enabled
+                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion
+                             // note: module is not loaded until it is required
     detector: {
-      modelPath: '../models/blazeface/tfhub/model.json', // can be 'tfhub', 'front' or 'back'
-      inputSize: 128,       // 128 for tfhub and front models, 256 for back
-      maxFaces: 10,         // how many faces are we trying to analyze. limiting number in busy scenes will result in higher performance
-      skipFrames: 10,       // how many frames to skip before re-running bounding box detection
-      minConfidence: 0.5,   // threshold for discarding a prediction
-      iouThreshold: 0.3,    // threshold for deciding whether boxes overlap too much in non-maximum suppression
-      scoreThreshold: 0.7,  // threshold for deciding when to remove boxes based on score in non-maximum suppression
+      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.
+                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.
+      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'
+      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance
+      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector
+                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis
+                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)
+      minConfidence: 0.5,    // threshold for discarding a prediction
+      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression
+      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression
     },
     mesh: {
       enabled: true,
       modelPath: '../models/facemesh/model.json',
+      inputSize: 192,        // fixed value
     },
     iris: {
       enabled: true,
       modelPath: '../models/iris/model.json',
+      enlargeFactor: 2.3,    // empiric tuning
+      inputSize: 64,         // fixed value
     },
     age: {
       enabled: true,
       modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'
-      skipFrames: 10,       // how many frames to skip before re-running bounding box detection
+                                                         // which determines training set for model
+      inputSize: 64,         // fixed value
+      skipFrames: 10,        // how many frames to go without re-running the detector
     },
     gender: {
       enabled: true,
-      modelPath: '../models/ssrnet-gender/imdb/model.json', // can be 'imdb' or 'wiki'
+      minConfidence: 0.8,    // threshold for discarding a prediction
+      modelPath: '../models/ssrnet-gender/imdb/model.json',
     },
     emotion: {
       enabled: true,
-      minConfidence: 0.5,   // threshold for discarding a prediction
-      skipFrames: 10,       // how many frames to skip before re-running bounding box detection
-      useGrayscale: true,   // convert color input to grayscale before processing or use single channels when color input is not supported
+      inputSize: 64,         // fixed value
+      minConfidence: 0.5,    // threshold for discarding a prediction
+      skipFrames: 10,        // how many frames to go without re-running the detector
+      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel
       modelPath: '../models/emotion/model.json',
     },
   },
   body: {
     enabled: true,
     modelPath: '../models/posenet/model.json',
-    maxDetections: 5,       // how many faces are we trying to analyze. limiting number in busy scenes will result in higher performance  
-    scoreThreshold: 0.7,    // threshold for deciding when to remove boxes based on score in non-maximum suppression
-    nmsRadius: 20,          // radius for deciding points are too close in non-maximum suppression
+    inputResolution: 257,    // fixed value
+    outputStride: 16,        // fixed value
+    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance
+    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression
+    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression
   },
   hand: {
     enabled: true,
-    skipFrames: 10,         // how many frames to skip before re-running bounding box detection
-    minConfidence: 0.5,     // threshold for discarding a prediction
-    iouThreshold: 0.3,      // threshold for deciding whether boxes overlap too much in non-maximum suppression
-    scoreThreshold: 0.7,    // threshold for deciding when to remove boxes based on score in non-maximum suppression
+    inputSize: 256,          // fixed value
+    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector
+                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis
+                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)
+    minConfidence: 0.5,      // threshold for discarding a prediction
+    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression
+    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression
+    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace
+    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance
     detector: {
       anchors: '../models/handdetect/anchors.json',
       modelPath: '../models/handdetect/model.json',
diff --git a/assets/sample1.jpg b/assets/sample1.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..82997d3ce77f12b2f369e7e0fbd94b6ca461e145
GIT binary patch
literal 144516
zcmbTe2|!a<_b+-f0%4FLj3NdEOi+|SP(-l-42B32!lbEMW|OEC5hq%0KtPG83;`jc
z1_(1)2e1yMMHFWg!ceQ$&k;pCI8j@*?OXc<(0=#b|GV$fda^m$IoW5gwSMcjhQmKM
z{<({p3W9=zFam*qJ%|6Wf9{c9@-|76F)TC`bHgx9A2T3Kz=$xSz`qQFDLyuUv7BIs
zpCiCnL6|r`E+Q~75}YHTnFKaI-xPjV5j^oR9mb8@r5G8XgJIw8a2Xqu?BE<05w>_K
zh6RTN^4w>KdZIIEWpKUG*VH&^O8ShHq;v*nhUW}#%+rHC+n3Gu^<XpDp1$5}UvK!(
zlq}WQiinb|{o~`*9ZsXGgX`^>sv8@egR#elRqyZ|Tam%J0u0lGZzg+qdU+I(#_oW)
zsen8_&MzWhG~Eh`>LTL!chwh>#?K!vB9D(x6j8?KbQI}*_}(J@(J@@C$Y68~dsjpo
zpMOzJ7(EZ`uQC}w-%~YVe0-zIbaW2Lp<1_o0uo6z0lyazHil`&u7k$J57*R?M#tzp
zdHkApHI&iwv*A4Tqd8-125D=?){Mrq(K%@BiC+gKPJE)_jA;Z(AHD+ppyedOKXdp;
z(@E3A*Ne^Y^zdcR@@4ysO8DRBu%y@mY(fARv%UKkz5esh(`)c??4N)B`5j#cE_+mW
zX`(Cooy1eW1f|8sUf~?lO8BRH|HG?p9q=B$o**8>+k4F#Mi6_34@2w^r!l1cXw76I
z+#8J0k+>KRFhW)WTTRM`BYr*xo!9LnI<J8l=(Fe_hEd1%m-H=2X6&=)uu1r54PpLm
zd>)3mz$^NUh3@sS7Q-xCV7z(sypQhnHmEDP4#wyQ%}3WnuTw`~8R!-E*ayr7Z7ux#
zbKhu5aA4qPLQz;ya0m~;0K<&<A<~r3iLg6sQqt2z3j-KSmWmmA*I-=+uxd7j#U-pu
z4UGtjL{|vr1v1vbBsBi;^ZTIgF8kd1LI&f*`TwJlSwd=BIxJNPv%M0N)+N9=8^)_Q
zrl+Fw&tXiDUxmg*&?}6d1`C9-4H~c1ji=*tbYm_WC$32WJtLwurzWmRMB{TX-mzhQ
z5{yY9FwWf|P1*?Kei%EiUcW{P<3C_bUy~HK4))dvolj3nkiggj#ztw8qCgnW1E@4w
z`O!H3qj7rDCRiti1%95oIZe7!lFnczuo&z)b9@;5q>ZbS($n3AaS5y9(h?bgpRY-c
zO983s?iqawGe>*NfK8soo-=EfyXOqp{_&sx@y+pB|MLLJRyX5SI4(1S)zuH@K797W
zxzDRG%x47lrr^W5_;L*UCI`c8ZhkoDtj4ej*%)^G`bYO;h4e*|o}TJEbLPg48)rz9
z5@w(c9sl|NoG?D~e?R!>`(~i;8{a#|f~4fQ^{dkvXj2nDU;X*|G{(BrxP&Bz`+qC(
z|MdqyTGmJFm@Z06PD)Ekfv<`NEt96K1aYS%O4FsEr!b@`|2ql)Uw-VPHK3tRuVG-J
z`5U8uHUl%gPRB^UT*b&1h8Stt0eFWnF1HYaXbf3i%;d*9y@xTpk01YaAnt)*iR+{*
z8R&GND3Xz|K5YXUgH50V1(0|GM#pS12W$$)#N07YY&JF*<6=BafQ4a^*ivi-mVimH
zHCP(95&II$#B#BnSOHdosj!2X25Z0$VMnoVuy3(5*m>*<)`JaTx3Ig|BkURQg5R-s
zKwk6+#smw3Enzajg)oiaN$?@eC-4X%1QB5=A(pU`kV05b_>!=Vu#-?kP!TE#4TK|v
zlY}#bON3s+kA!=KVZuwoTOx^QM5GfZ5nYKhi9SSs;v!-saXE1%aV;@}m`&VGEF)GC
zn~BGXXNXsb1H`+;XT%Y3mWCuNk|W8T<U<N1g^|RhWKtSwD@j2rB~_6Qkxr2=korh>
zNk5bRBvZ*2WCq!tJeSNTN0AfAYsp*5`D7Kjo_w5qj@(DSM}AKJn_^6{r%a>xQi3T<
zC@U!&DY=wVN-d?0(nT4dJfi$grBbb^Q>nA5eCkrFjQRyNpSqv=HMN7<OMO5c(bLnj
z)tjd0rx&Idr<bmmt5>eqtoOa1R_~GCANq#+ll49IgY?DvDf(IZrTUHf-|1`hAM3v}
zFg0*7@HGfCNHo}Nu*X1SaLVAi!6SorG&9;%njbBaCZlE2%4mmZU9{V@SB8d$PKI*~
z!wsc|a>Fvi!-f|P?-~ASWNzeU6lk=}XuZ*HqdKDwqaTf485<kB8gq@s#_7hpjT?;5
z8s9Pg!^Fa5hDoqVqRCe#`%I3TXia{dU@*aDg8zgS6EY^0PB=E<`h=fN4NP5415M*h
zx0>!VJ!$%b>4=$y8QUz(Y>ipIS)<t{vnS?y=C0;E^F;G(^J??6<_|1L7E>$&EfOrY
zS!gV}EFRIR^r`e<x{SV)-a@}de{N}R$+3*K++?}Wvfc8o70JrgD%fhZ)gG&(R)bb=
zttVOgTPIoPSs$|2TEDTGXv4KhvQgN4Z8Ko=)|O$*vt4CdX#0)rPj+NGH@k4Vjdlm@
zy6k?QNS`=&V!}kl#G?~$+7s>F>_zrp*jL+MwI7+}FlphW)Jf%&&QAK-!ODT_AahVU
zeCIGc*<$ki$<oQn$=^?Y%CKbkGgdRo80Q$z9qk?Yj_VyO9j`gQbDHWD<&@?0wbM^i
zXj6QqBu`OJ>74T1d9rh;^Ow#|&Np2QTzp(4E_+=rxV&+l>Kfy^-Sr#SCsS>v22ahH
z+BEex)0p`Qb1k!q+0UY~d|0bk`&m71L^p3YiQ7K6>(hwSW=)e$+dr*$I%WEt=_%8z
zrvKn><nHgj!M(}--VCc5Av3aOd^6+Mna(rCGYe;4nEAJdx5sJ^jmIsvIa|QaVxMBa
z^mOw~@Z9Iw?`7;2<n@);39lENX`Cd^0nQC?3-3_x?cSZ<f6babD{a=HS;MoPXUER2
znEitf-ACw??{jeuWlq4HujYI^=TF~RzUjV4eSe$lK6lmJ#<@@CO`VrCPc!d<pOasl
zU!~vO`HcCo^VRe3e&YB^+$U9^+~-c=CUR@JkNsKxGXG}(UjsY>(gNB7-URvvZVBvI
zKw7YHfnvcmo;fdyr{diSatc}*)D-l5q4&Zs7IyF{{6+ji{tv;Ef)j!pf}aa!3$_T(
zEizgZu}HP(eu!I0TF9wTVyGarIP~^nm&GZIzX`*_g2Rf#eiBX<t`(jRr-lo|_k|CM
z*rG2*7a}YoK8t9K_&qWpa!=%~C}!09sLtpK(aWM6qTj?Uh$)J>yJY5)ElaK}wOcA(
zdP1xxjuzL7M?U3!TJq_mW!}qjmklnTx_r~}ODpVFtX|Rn*@VyHK06jmiH(VEjC~&$
z7FQKF5+5925&t|PFkx@PvqWy9GI2P`FR3W$ak5`>QSy_Oek+St4og0fluCY<21-@Z
z7c#!=pzO`6uvK-d{#hNf`s+3NYhu@&OfgScm2&QLhtD^C-kUltH8=I{THm!LYoDhr
zN~>K*ShsxLiF7)UsjKUm>vPuM+c1B_z72nDjNW)`li8+}O;<O&ZO+^LI3p;d<_q!{
z314)6>G);lm-n{#Z>ii$*c!jJ^Q$ReWq<WZzEEDDNy}W7*`4K)Rh0F5Tg<jo*^{#6
z*$;9S<}~G+<fi5Ruzl|KgFDDOq&u$VdF7So{jEq;T;A!ivvlYC{Dl0=yV$$RcKx$E
zdH1zFv-TV)pcbSQ3>JP;SYKpTlu`7cIJEe<lA%;6UzdDVa;cP4dT=jo@A|!W%R<Ud
zlslIfl)qO=R0I10_I+JpUy)bwW`E*-?E&tALkA}v+<EY=TB^QLxv=thm1|XLHMM#}
z^^hi7bD_qkrlHoZc4zHhbt!ds>qYhF8fG^%G}<@rX(BeQZyIi1*4)#wpyk9N_d}Im
zTYbIrFm`zT;ipGpj|?7NbhPuB&#^<TF0B=97Hx{-gyWl!KmSJZ&D|4APxPJ?oa{U`
z_f*^I8K>*Mb^LZ;yJh>H?`Yp;fA`P#8Q+g|q;~v#X62a&ou75yI=kd-|GDsU*SZ#U
zT|6Ij{_F++3*TS#yZG%T-%BSi`&|C!%B(BLuX<l?>-O$$yEg0E@$0j%pXiy>bE<b<
z?|0fyv}gJj^quby?!Pjyc%XMMYVgJn%YL|fBk{(QA6NhQ;^u~%?{3L&Q*JAMGX1Ia
zj{Tjg|FHh^_1#%_+wU#7clCb6{aX(b9z1)v?&14K*+a%drH>~+Zg}GP<lEt(;hv|T
zK7I5o_1U|hbAK`WrQ%oSudTmvf4lm8$@51q(q8=YGXItBtGd^m*PSE6kvqSy`TgA+
z#UHkR)c@)8=Y_X1Zy&$g_+IaQ*<Y-`PW`>;@7w>Zfk^BY7Kj-e&<tt%#)dRQ6BFYJ
z77R-Zb90NS_72vJ87#KP43_(JFJHlYuUSE}r@M1w{DVRkiy}mx^TlyuVXPor6oxKB
zFflQ)Ftc#Av~&&gcJ~hZ|NZk%2WF;E!Sv)Lf+I#WBaqAp|D3}n1E-+?eSjatRnQNC
zNFr0Hdit<vn9v644g!fpB$Fr<GVmF~7I+^cn^DXi+5S|E@HjoEwRF$S!b<%q0Y}bQ
zih7?rd&Q?^8PKe(ZEWpaT&FTwZXEAfvwi0H1}@+QE#wCaA|j)rV}NKSBqk-Vlt^Xk
z(${a;xM_37w(OkT?K|=mMa9aJ(!FKns;X*DO>JF$L*vn7t!>A@IdSq-=h<^z=Pz8m
zbXnWiKQQ>ijUR8`yZ_+fqoK!7hM&H8`Ret^?{EJ2^DVkBi0yT^HGXIRbzf$1UqmvQ
zMAk$1MIdfOH*Q9zII^kc{^5FYYb~5SGxg~Kg_TFn7)<dJJ-3Wc>!n#abMCpkK(~h9
z+5g#%W&K~>+4zlpxUZX-5eb5AGm;th3HB->*k3JD1X>mJ5W~6H-GO?k`IS8Fei5Te
zP7Cm<r7V4#e=wuLU{{T0fJ3duUV)|7B^T(2x=Ju=MJrZYU?L$VHfU>hnRpv184N$E
zp15)XQA3nZ+Rf4k*%~1Wd*9a}7iDq{!>GlC4z0Ii;hg%V^={77rRAFL0+H5+7S5+s
z6&Ua^V$I^#Go{r{tdP#YVW~M?iG_0wgSiwswkxWpR$lEF{#AQ)p1C5HE+p_UdISbb
zWPW@ofT`gjL<PoFj~&Eg6ABD@9%%O1tK3R~u_OeR1TFwWrrLiI`ld%ssR|qGswTsw
z%I)DHWZgGp=C~7SOoB1b#e)@kr*XDe!_TT<$zO=v*%mHKcW0I+G#PMveTzk`7RTt+
zA4@jyo6lBzyqEb;-=;bG)qkQ+osax<Px&5;tG;1V{~B3rJ)2Z>?)x_bF+WUxsUC{|
ztMS3#TV4LazN>#1dg1;1MgL%1BHliJBRpA_dw<<OnENAmtD#mJ_WpSB-*2m*?Kx_a
z%(>+F%=zG(0e0De<_j-0FG_s%yllR@?maO<d#u#0>FKniQ(e_*tMpvo#6}f+iv<A;
z*$%DgSpk!HHA?Cmy*g1TvQdf%Of|Ye0VZH#8a1&fGl#A(VA?2!#Be^b2=*3DQ1_6G
zIei2c(`<imw1f`($Cto)V)Ge($hX@RsT#gWNYpwg3DL63sdpL!5?HeAIiY-GjBiYp
zn=81Ob{p5^ryvQDf89HY#jflY1_u)i=q7Ce&Xx>y!jYhxm{gp?#OmBJ-@QI`tn+qX
zNrQ)rIB1xkXl{4tXJ*oZ{NownisD3iYc9PNOIC2{LZYI?x5rhyyLqah+SWXrUoff(
z<_avR@ns(|`>&@vN17Jo=|nE4vE<=Mxac5f)*|H*A-x$tZ3t&@ok9n9-9)5(Jj}%%
z-XZpKvEk7-#-_T{_0vY0Qnj%w6wU6zf%W)SGO-$+=8?e~I_hQzM&4=>ISW#2$vcs*
z)F53kZ+0OqxFRBPFKk;(l}}PF>mjI?^%1;+Bm`ql4=K3*rraqbXP6KzIP_Zo(1@8j
z(AZ6j<R-JZ#MH2n%#g8wbs#};(NIVT#$T1WuvXovT)B<3wL`!(+xM~5D8EIk>1@yY
zuj?X<;8XkzA!X^QC<EUCN8_sc0JI*5iG%x32ni}LK8ceow~}vG8zjsdI5Etwd|czt
z(8#m3G_|w1oQKV$EfW&W!aW`<u(&!$wH~{Rt&-bTsET|;3=R2zv5Xp6jaMaZPPEhp
z?|@!@9FIgtb}lD~ri#f|)($9#BjYdhj!O=-5SLv30PpV+vGW>w2eRQ<om6mP;fAb}
z*r*JJ1l(lDMT@n<#4?bxs0WD{iS5HL<8$4K4(+nroHFb9rfDK!tB@!{$Tucd$6?{6
z5L!ff<C?(+^Nxrd4`*=+E7Lb>$1owzty388*1wZGfqpN9((R7TkaT0VRdQ36l+z*g
zo+u)^dXrSx*Dlc+<d}@Sdeh5kBN|G5<K;+pkw|2N<iY?2W$BD<)V!*|Mlco&l<k~u
zXpDLEg?gAP$jRYgZmn{$G3G+YOxjXtHsxzFB6bBk!mVwu!J<_4mj)a3lPm*h1U2O+
zfnDbAp9Cfn)1Q--H0}%*lgNKNaB-Mj6{uAgXo*ygd;ojx9mgbwFX9?|TZo9WG|{z;
z<V&>&BsPt1k@t&<NyEWY1%a&YsNkWjDrTKYiR+Hk8jTRF8>F|0OO-8n#w}0rJzNqr
zy-2ZOCz!^lo%Cs&u7uFMHa5<D)7U#j()d3hvIfZ|0*r(_Y9~6FE(KH60r$TU8Q#R<
znI_f&x|nJD`dRcn&3R;U{4N~2#7rH816WNEWC>y;%3efX2Yjvqd>&~ilDi5HD~%49
z_-bw4`sICuyM38zO7{$@JUJiC&Wo!b>d3)5WwSVhiFTRYSWLlylxeoS(N!H7I)_D&
zxwV!UNw&~);Z||DM5ri05=7t4r7Jto5Uh~|LIlIqaImwXwM3C{if>VK4@(N~^?-U|
zz5^$i@ky;i`$c?ER>$ACmbH!MTb^nhPiVZk^!Q9FvkRk-Xvj@r^p0hs_yQCE?oW3Y
z+c|%JI_e>(z;gLvX=P)!U{`d43@dYBg*xVie?pZzt_z9aZqMoFo0rjI=%jAPqLvBT
zg3Ifo5(YCXxiuPbLS{8*XGnmg+wrzgDzO1f*X(H_CS81bIm+h9Ymu|gH44zeQ%Y8>
zP<B-Mg>P+>6t+g;K>0CRj$<RTM}`&b0y4xgn}L&y_+!R8hMukCpMf(%CP4z`Oorj>
zrh?x!=a=gwh&;{*Uk$2SR7S6D-2PjX6E~s@nVy80-k>eGVi4g@a)}U;LK1R(?8q~u
z>$rv>JgU?ygzQ?N!s0mRIr6+lJ+;1bqt-5-Aw$NOAtLrKchjcVCpo^>xw-<13ql1E
zf=dcWbnI1R1E=_i2$&c2knB`axl%aAQ&G*Wad?fjHyC4d8a8*Jey;Ew&W!l2O02a{
zE1wn`sd*NeJ)52#h1~{%Gl7`JB(S;a1v^CwY?Hl^$FQU<^$*@PA*Cc*%#DsK=3A>O
z&D9Mt3asQa`t$zW8Jp^zebo9(h7Q^#XIhDTw_^13fsVDTp#qB<^YjMW#VbZWYiFyg
zIrN-v*Qe9x=ADvDjoF<C+?Gm~iR=O~H_3L6LFt1g!^n8>3`n&A`p~vJf)C_|A8Z$H
zt!*)%rBMHyCE=tAs1lB_7!$WiaN!ZSh(H?vq`_~ykG?-fplZ}44ldGGop!Ve8DIor
zEMt9Bfi__#FQNr3666W*X6XnR)v?~W)=4=a9BEvjV0D(6y9-(go`9rA``wb%iA^FJ
z)h8l{ZKx7s;QX*|TQ#{G9AIeBdHEC>GRBN7LHx9~pnm-%KVL2JHh4f={(8Dx0K7y)
zwEGxFb#{rn#!lQsL8%J*h+cy#38vxbcWW9sR=_}cfU8_mi#RLO^XZnG;vyd0L0~r)
zy{{b<pU-7Fl^Hwj)0UY~0g!}D(-HB0kX5RU<~P2cdZ@X@l`}29QJb$N8dvmXMd@$7
z9j=vIb1IMU6W;WD+V_ycH`<+~$C@bhOUureUA`u@Xkkou!38Nkl+j0a<S6NMrI8?<
z+w0WV73gjl#G*4TIq3$a)eVF7T}2;g4049UQICM~ev+a72g~>$gs6i}B&7c$zZ%@(
zfs82{BvG9Ti~^|+RS{Y72iC0=lJf=+1qSDzp0v2lWz?vF@8URaliPGDB&5bK%1>6v
zVBH@TVkD5WVPuoyLW#{ifiZDXkW}89EH~;^VF6MN2+^a594hp~xFe)Kz|&bkV*FTy
z6uZ_oqso4;Mh&=J+!}E{1NufrY@bgC4Z>|yM`HRysCH{GYEh<bYm~}Bz`|N{%Q_VS
znA>f-!Cvc{xt8P{ddK`$@P^T<`iq+z1L8@GYvH^508P~f#J;Q^QvbRzYNnxTZVx#x
zqGgety-bf@VD><jY8hZnEG*5;ANJ=EY`?U#9@04Vvj+-1m^Rt{Gj2x^k5y=E7}IMI
zq81HAKMOHcp3n$S)E08jbura?E%WS3CELSA>`n+-NYgK9>jc$MY;g+EX+C%e!o(W?
znF0yTV!&WN;xm?9KuI3J&-ek68VEcDEcqy1fp;d4SsN}e@R6X1xEw~^Ba@D`{@Vx%
zJR&WeYrb&s(AnVplOG8XzKVg{BN8Bwb$TEW2o1AN5zB9L>mfHp)HoK`gtQu!NbmNL
zx`2BtV%=>a29TR-?J#MrH@0BFu!lTB2!+fI^{PxGKXCti!@{VN`Xmz#Kn7wm0$5sR
z9!>;ALIy+-3M?d?9|zz%DnkmGv#}%Y5N8$D_ppFMJF4S&M9Q-m#xKEnAVT+1I^zLg
z>hf-Y9}>&CE0Y`4_F$}MT=TCvFZh%?TD+BVz1l!jx1waaBtl!6!PnpA6Qv)%*uIC|
z+-A6zYwFMzPw95FE{Kgd#=W?|@pPMqqpK#6Z^+V?ycO9lmepTSQFG{bnJhzvzB+-g
zA!eG;Ux=;*4CQN?sorjO1h7DD`A+AmLPwlY;vmWy*eq2uuSYt+8vvgO!6q<E`8X+M
zj2V;wYzmJkK1jB`9s(pVJCGh8S;D>m$}S2SCq1Qd^D)69i}V*TZJpO|poJJ7dEJoX
zAJF^n5O!6hc|-v7dej#Kjx?~q?JWpKBcFi<idw(|H>vd$sT`way&(a>cVXxIxW=6G
zYW<s)GA#4y={_@3WGFY+?#nGw!Unse?QXV#^<aH?fG2=2kJlL@Sf4J0kmKwIq&ruO
zoFdUHlmjdW!iY!+{^WGTQ!}xqf>ag?BWfC)VMn%w(2i|vv{>I5kQlFw!dlDJ<hpGN
ztk5`E#hoGD7@tYAtEl62l)J4JozL+$RPgm*EK*d`K*DFXTZ+Zl4ij1b6uSPg({$~7
zp{KR%(0RFC=%PsscNb)S5q9+%J)C(FU6h?D^3)KuLxuc+{(bd^yQ3Q(a0qE$(dOG`
z&BrMznAcdoU>GRr>qhhSO$(0vTX-LW31nGTV==}D@u3kA6`fTkBWfz@!?AGepc5)c
zkb;;&iQCV9jZ3T(8cH?sz1+@^ujk$PZR<JFfkq^_&ayl!jTkF;OV@a&H$-O^wLpvv
z{tv^neS~!g8Hg649IP({sA%L&?hdg!BNmS$u8laS%THP^73T_{vOc1rRNc+$Ol>?~
zA7!Y<Sdf&pLlDRX<g_zb#HIP-f}^#|9!_dtd*1A845kTU=@7PyQ%V9-+V~+mL<%(!
zJ0nD7#L6_5{sg-AH7HPIiiCJIxh^6YGCUR{t`|MNo<djmU}2Uk0=Z_ap_ZUc>cBS<
zOL{zK4VCe**_N={wn@`oYdrIRdfM>JKir;mzc4m*HU|S%Bo2@o#c|4OUFXQIM#@iG
z3tIOx9rx6Xau-|!fxEn^r5L=8(gV9I9VIWgZ$tD3=Q#L5CmuWpgrcW$Is?wGfa30f
zXu1S<iXCDYX7muWW7rI*0aX#(0nb-B(scOuhS=r1YsG~-c_d`jD1yYlVzBY0&?*0~
zr)^?W2q$)A^fz%W+(VSPi%UFJRDSINCgHv^g0OB8BEq(sT7pUd$Z5w70_1eEs&^>c
z;Q%HW*DwWu^GH<Au<-IE1e;tJ9ZN$o1B<24V&&t!1;<RJy-1*hQ1D96_|hE$CTRvD
zqI9fk8?rnkKzbPp1RH~mL=l*}ha!al{?ve@r@eNl-d`oA((myDOKT#yg!%)SyC%sJ
z)rR4(n^?hF_Oe+Td46vsM8*1#EyY1Ya+`LU)`V&LbeWJ%>!bCM1bx+X`O$5P$%P`%
zicUy#8Sbds%dxW6XfRf1=1$Pg+oJfg?e$G&CQ|*=rcEq=qb=?q`#ig*E1`sdts?<p
zkf%$~MS?ovd?Yy7T6bMqW!x2h$mEH+U<W`&rMiFtV{+h=I!eO*;0+;LM8k^aw`JIs
zQZ7DStJTD7g&rW(@qCUk1T<oDxT^6@Qvf}pi|F7MRqGiAk%R@SLD}QUs{?{wM~ek3
zZ%szxR_4m>q`3t<5N75))vI-~u2;#Y=p>ikG)-HnbO+uEK%m1HBk+8<yZdlzlz77;
zq-y<c$680G+US(qE@CFvgl<Q5q3eea9`e@~TeyEMSdX@hWel{lviZ-(f^tjxIeK!L
zi5|VO0rIa54yG9@li1u|C#sWS$0msdA*Qdw{TvJJSUbex*oBUbS2wFEyIWi^AIQKE
z6jC~-WeMX8t{7^!4~r;s8mqTFHQb?0&g!bscvNNXRE5AKL}r0$W<%bkq&9gxSO8JD
zU{g%ozHMiznPH=zP8a>*o(|jt@?9St#+f9-&M{*_X&1ElI%=W=QeEx^H={9n1IeiA
zTX-vSYY<OqcMGEbFESI0$(JFI@FNmh4$)a$D9taIC27c4BHY<U?tCVpb3fN`(WrM+
zNlbIg`Vm<2^~IGL@r;a|dO`_&17k*4v~^7&Q#^{6h>(07?Tw0VOoU9VwhhpdX{xk~
zO%_Gj<uKI&JrWzG5{Zl}1vi&oCqiznpoeUpKQMcbUxV8AVeiAf6WcCnUWQs@FY8}6
z+@_e>r3$bOm!<hy!n{+35p=@5*a0OqSN|(n-mp==Vp)V~e7?|=<rSrP*k|S}lU+7p
z68c-Q_yPr$qlmKC64%o|5fr$Exa~8M%$SlYIio(pb9c;Ts^fr?_=pIdHM_HxruiOL
zUj5O9o~L6UNC<WVE5s6-myTv8>1gICIT@1?64ZaETtEZCnIg%K2Lm6om=B4ykEF-h
zB75yU9TEs}BG9ZHK*@2UvI+vD9@p>#U~mg|)9-Q3-TAHOL)tuh$d^Pdn4t9q_X2G_
z#;mE;>4h>kQ^!%NGPnT)hKQrMbB!C~af+GK{9`~bkP%0qmUs;Sr8t!Ult!@L(X=%F
ziUg?wh7!!pt}3%tMyc`?kYZFr%0=k7Q2dRtqvum3I>3L(udw%eo9*FX2cjH4Yx9^e
z?dBf00=<%7f**Y$Bs>!R8Tl*m&-4RP*wm86Oa79WIn@X#Bzm8%gm9X9i`Ox$_xS#f
zUB%v$OP=x0`6YDmtx7tCt*52(HR=|UMbexIx_lGe<9>mqCZylhLBcYADS0WG)b(;p
z{mU&4OPaP9FD&^*aOUdNl3x}U^R{1gh-sUtlo%?Uyq%`D5l<$2H}pq1>cT&iT0vqn
z1Ln|C`ANq0;@_gGf)F8$)upUImT6U)VK7RmeozK{1eC^wE*}#sOoHwEUyO1LRlz7Z
z%W<PTQR2~HBkF;G47Yj2Ulp<5jjioZESRBDrw6JQ(sX%j4CE^GOl*jjJBtg1?9K`%
zK`PY+F)ogfIAhlnX90XV%5t9Q<ocVa3p`-t{bNb66tpxZo($CqH@m6;GN0pFwK1z)
zB_t;oWg752yqHbi=A{KR<Rmk(XVL1)yuHOW^EHt22NG>TkCz*TXL3;n&8B({Hjy`x
zhfS1jP+%Z4xi$TS8dD$uoYh18MRSX*AFYZM665#MvAe8W671HZ<8q56rWo76($|oa
zSo0dG4pCD}LP(dZuc%2^ey_g#yT*Eg+n#ARGjGp3HSZMh-lK@Dc=<`U7Cwb&r3k=T
z1S+9GZ78O+j6SvbPdwOz)Cm|9<G<vh;~Y9X9hK&o!y1>ORZ$suN1eD@!KaMFsO}_+
z^(=UUUk8fLFWX&fiGp>Pk4RC+s@IK2!3-+~WUc&ri|GWX{+%HPMKxmTViff445J%u
zs!{6!odr06WlPlocDr=aa*|~}F*EE(R#>H>Tx{%U{?Q*778Hc=a0o0Bf5qj8gFk`~
z+#HSB^@QA_vg|7PxUe`VyLP9!Ly~y(^%gfl5qP{&y#nRZ1-U5l8Ozi|u0mf5J~$Kt
z+~=&$0uy{sC0&>+#YRo(s3yvdtMX~MOpU8{RmqK8*U+)kJc5L<7BY`xu@vNm)MW1!
zf><G>zHq}4WvXJCfs2mHZS!$g^zi}qZiexUJ}x056xaB%vJm0}0D8e==}iXEeStW3
z)P!`g!MI;TX_6W`fwNrp-l^Emy(YvrlJ1T(I9&_@cIRO?5)FRMVju%#0tsG3Bh4X(
zGXmEJp&POw+^gY+Okf!kfr&r@EU%vETJHc>l&v+rGcqrrz<Nr<1wpP5l8;KI9B#Cm
z{t+I6G=W*d8pnJc$PT1ZyKfvC7x5`{J1mfgk?u5eq3@}BSk#YdK%?LQR0wWrRQ2#*
z47eJSyr#>;j^_cn^5Yd0hCRiHhAqQcy88zT2e%9Jb#9^+7KWn-90VvM|FydZGi>S0
zEl*qd3a<!FdWUkMze3I^niuIaJ`p~GBuH2*x?uGbJ6!NZ#QQz64NebM+-bbz(8p7R
z>@F%84XYTyGVpH#t$$?O5bY^3{mrec_rCn31Dodv9!c5S@nHqy5D0lnGTbVZKP9IW
z-y1K|2?{tJ%m+ms^+^$gucKVmA=Tb7$cMcM(L;f@8%XO?DOeDnbzknl>XciRA_V~{
zi0sHvslkE<EQI8l!Zs{m#6swpRj1Gx9_c>|^aDa`yEs{Z%`(wuWk<M+&nzkS6t^WL
zqHGse9y<@OvbCa(M=Ap;G295Gt>@_bnIsQwlxim=+qa>(+;SerU~knEH(3$By^up0
z3qtWs!f5Ii!bdJJ2`+0b_ncEPT_3BXs3}B1yZ8d6(709UOjm+-f7A|<+Kg%eZ0q)#
zn7!>Biow(DgOGRhwY=0+uRNbejL4$J#+=|%!sEK3R``+2u~G_250<HVvu3cW60Z<_
zH=GeKk-vNudFXz}HI}#fvxs>kn&$2Wqfr$A54wRa%!O&+Ln%w>7BBg$vB>ero#pH*
zy|%fuN5Kg<W%kjlTM;=#8$D*RW1xs?epx_TcvzUVSSLz>n*cXauu^~&xIAdyo~Wc+
zSK$Hmm?Zg-GU;y^l_anc6I8;h(IrL+#>i_y!ZNuU*ngw78t9~q%xJjZV4F6-#HSwc
zk6o)SJi=A{=IS<OcKck~{pQ;P_5a~;42rj@#DzI7BvcH_Ez3R9QD6QdfavA!@Ok@O
zf>W@nFw}83grm50?Z8Q}PeE#QR#{e&+{@S{ydEl#JrV-&7fR>^F}Yj65s8;ej{rB!
zz!4A+gGLj$VoYF+&8lD#7*%B*eNv1NHVJG71eYyz#FO@;&L8#wK?wvaJ!4|U1q`xv
zC=ml!98F1<{-@<LE;~e0KoiGColKeLcS>ujRDQhSRr0Cwq$aWDM4r#D3%+qB?p-6{
zmo;vzdsz81HT+qqcVWHGGk^h&62bj8SA^7T!>*-<(fK#bf^NpU-L5|`%d^~f`$S$a
zU#-&{omfAx5*$+XkwlF@+g^i=6C_F+#m_MZXx1L3Dl3Of1x0KMD0<FM7Ns_b2D=Pt
zbL#Q*52MQMS^XXBN8O}(5sK&JHpn(Zh3!}X3>>Rlmf*2Rh#Qrjo0-40M3fL<!Mr>|
zeIRz`PTN0_L7gW`;5~U8SyJu~B5`rdalwKfD;lQ7HyD(tKFj!g|8G(8DWT+(-kT?S
zd-9yb5W=AS0`qi)gej*_x`V|ec^1|8L!lCd>%gFDcS6PO4_O8>TY#fyqjvTY+5*#d
zR8#v(3(QTFt>>v_63mh@jPrp&u5nf7*f!yxK);B_$~4H1khVjv8R0C4GI96F0lZk?
zYjvin_j`IrnfH|W6t?o$iQbV~)0@B?Vdv#K?Eo4PaKpQalvP!+(TP_wr|v~`Fg+jh
zF5l?-D=T=HyC7(vP=enb*x+dTi@<tO;f+Z;&aM%!;EBBwj)bOAJXp8Cv0O0HeCN>u
zL0FAb)|i{eohgoNA57RsMBF5M@5l1swGN9EML8d%2#rac-FPg*f*Cfbwv-g!uV7^Y
zchULD5?MI+f&!&E!$b6-;>1LiC`T{8grGtAE1RX(KY-$PsM-e5ju<bpUF#Mu^w4yZ
zZ2oeq_=_2koZc>T>os-z>uEcMWo-RmSQMQEcLlgFvZF4XDya11xQY_GXSFhp=!tuC
zt9VXp7YU*6jjTtB<(5eZwMAth<Rq1#+Ft_LZ-)@ZZAnItt0Yic*T4WVj8TwL1VYG~
zxigrXqv2r6>K29zz11iX@+VmudWRTt9J%s|@Fu<iI3SLCDuP5;j2pLGC73LPcSe7{
z)5Uk@&F62_l5hPr-2A$uh1DG(MKlYVTt<^DtwK`Is~B}LQDrg1x%HKI)Wsu4MTt=c
ze`?MbF40Do0tDbr-enX`U6}Xf?%Y3`dM|Z}y%G<^*56s|Ca?BNZc=rZ0x$0@`jEfw
z_P}?3Oz9w>4fHbJ?)wR{5XtcpMuSc3xDx@i3FaDQG|kMAgj`ouB5@u-#ds^Z3lwyr
z4i6m^mW*0<CwPwfZs?*=`>UXk+wI7aOKXd{M&8J1!)FVSL^Y0-CAq~2hiA$%b6i<Q
zt7r|LLn;feI#Hb}t91c8=C`OPEWYIX$^IbK92el+0AlBZ{&L%O0lvQKCuvn1;}}KK
zIc*L-uR$h2K&^AMyST=hT1PNg@UwO~?uf&HMno2_SBLcEHq(&7>ipwaCQ1V+7Y8cQ
z3Xv$Lgb)|()hIB8gV72iOk1Hh2?>?2OWx!19QDxuOYYJIgC>(jc%tm!)H_WJvLYL$
z^1u4aYsqPLxgXC)QX*+v7IxHTUn*(1)v#mLY)i3-9h|vDD*}3<lf_sqK`kV|coETL
z7MYjurf=!7NGIDlmV{@*%K^HI%$WZ~sW#ohz*L_B6OqsD+s+}xJG5;z3(j({;p5J9
z?Bd9r_3t!xYhlNSo(UC47b-T@k_I7<T`0pdYF2V*I(Sp6(hd_!*Fjed)R3SG)EBHK
zyOv3HofQ^Pe=wWPIUm6fa9(`@OKv%9_@YpB(~fuX{^x-a`uk2aR}@z5w7xXb?GW%*
zbztqwe2HE|{ffSg3#%*U&|cm+V&mgfzo_n1+P4a=x33w~bo2CyHJ{4Un^r4#_P(wg
z#ElvFSOjs7Z`8!yuY1aQZ~In>3}=nhOb}e-Imz5;z%KDyg7hpV)o{+a&~w3>k8yzk
zK)L{IV*bo_?k;^<M+0M=G(w{w2pciPaFo?|r)L9yUgIMqZ>w<!z_-J~kpqUt3Lt_U
zEe^{#H5_xhs3iin@1p}0*ZR%-xWWU+9-^9(RSmVwP$iYcvUf`nqyI1y{I`9b>hP{E
zj;p|K?|Qb;%6jM3+JptpsE-Pi0>=lUR!AIz_zP+Z#sXuJ%kFM+A91QT;hE@smLd?6
z%aHdN)nLT=0GSXdqu@k@n(db6etkALYdZ3^<NBam0th@bRR8+EmLRuFki0LOfQqTn
zH4!0+(8Pw#v)jbM?k%n*4guyZ>mfoT4KcYug-JPD2_YYH%;ZG71Ok^<dVS_k5z$qB
z%5{$2Npo(T=zP@rO`aZ&b=T>&_{{Nc5r6s9x#8c6Y(AeTt?78J_CFaW4%*|#B)SG!
znmJTunzbJlcv9_}Y>dJ#woI9kEpTr1V4d||uoDHP5CL^0yl(V*Q=5Hx!r(q}!P2bO
z%HgFKHK9Czf7l=;h)iGR1$8h;8yN#|2S&?4l&#3}4Z?Q*CY%!BFp#V%1mwZDb5v9~
zZIXv`yQ&JZwsDPNllZ<s(HxX#ximN4Fa(4oS|B%9rK47;n-qaJ0+e2qVZ^B_NFNWu
zAar*qA&=&8TExzhI%=1P^?D9$)W7{6xN|I0apyJm{B_N+;;ijc&on8@FOScH2`E;F
zPK)DiJX!ptQ(67Ll2`v&-|fmqRYgQUbaX<uf{t;S7f|Lv>Bf8O3}sob)J*{8jWN~1
z^@p4m6@0oc0PH%PrnBoDeydTTj8{u;60z_acPy$;xzB*!40;C(_1mzpn*mHh&6ZG)
z%qO7?OQ+2fF1Mvjbos1YX|eQLbg#8d@m)rT?OTy~*fPQruW--yOLg^u+JY`NNvmFe
z^FJP(tj_G~(=I;Y1-vjfbey-IuD`Qjb9YE{kmc#u9BgE|PYTcHyfL=$(JxzBi97fj
zhbom>aDR2|ZsqzGqptH57dG52$o~ghOHtxAX{Bd%*L6Yc_M2@dXH8Yivpa2CU&t?2
zWaa>8K}O;T?qwF3hyc67iBf@xs_@`|^#XzX_L`-Z&O4!ej`kXv0PcHH{D|uTCd?3$
zq*u6?77D}|z`Q<%?~4T8S<pk0XdERV`Pfm(f5^3=xCnFtPvU7H7Xb=T?Q?gio=sew
z5mm*%E{2-h$9@t?1SyX#G^IiN3+LbO;VubogIPyX#LjqX2-wB>7%0SJ>7%h!1(NS#
zae%x3;MY9SWuB&iWi+l0x{%GOsImKw%Ix55aeh=0NY}%G^(+KhaomZ{mQFqFyM1SB
zNtrcNXlpqM^(-b<t%3#xY^zdxwhfzIw8?VbtghYSs>K)UoA#P>_8bp2xnY}-x$L)y
z$QaML`wv&Hdn?*G?d1Fy`hNAj#;%tlLJHEf7GduQE7JTMRv!xQWLdOKbA5DnhHPOx
zHLrU0*8=*4**+KS-}Ov0aK;Yk+16|Fn_l!pSS+aDw|Bm>p33)=5H{3*RhA!NBfmgD
z$F7cfs0gGHI=ZYacQ|-{FUZMt>vODZjWP$F$c{`oS?gqBIs0<m+6hZuL@>Kt&u`bF
zzDqp*LM<sIsWIvR8iRh6Gr<I3ezi&HD)GPBtbl%?Oi&V(8fAdJ0kM{#G#)AuiwII<
z%*99O>8~w?(4tmu8Yu_A2eF+awEd|A5CC-OT*r9eMYPx*8aSz6#1}d&e73kF3~&Q3
zjiW<2^k5og$yv*d%-brS+&UFi>)5>Ov<>HSCI0>Yb%4)|{{OKBf0SlId3Y|aSm-3`
zul$!U!8M$g*5h>5PxgL!qYa>enD<}bVHT`y49URZAjC3J0k9908?`Od$oyIjwd?{Z
z43ef&e^69kJVlueuwT4^g=zcUjvZjeuK4D-&86XI(a*8Y{0Wbjz1*1G8|_rmnMxh<
zqfDO_#{P-@KwJ4l6vL@J&OdL-ZOEeKc8loD*EKgL$O?B!=>4H<6vT6D^y1yVE&bzR
z_*zcDnNk&<{V>8gYA2RcTc)^M<jGQ;owRUWH?P7j-hrmt7*^w`e|ds`l<SW5pIxq4
z>TbEqva$?%_BBqK28vLWBy2_MuPf9+@C^;t&_6_RVR&|1CTHTPsT2|;s^0Y*s63(y
zwdzdVsPTLjm^9v3iXtX3Gm#c5tt2Ur<T8n=rWS{C<dk{fefM&(<RYX*5b)u(5M*tL
zH@F}zvfcYbQr>Q_3+tXm24!s9CU&Wu-h1@BpLW#xq+A$mr7S4qLnvSi>2Dvp32=Dd
z9;_o%c=VMjhS*#%m+jv1rhe<y6<q6rxxH7XSu9$8<j|1M#b1_+7QQ+#>+976{<rO%
ze172*yep^d)*IY6?tpMF^x@F@Z%DzF1Yr&yuzzU%8?Uz_&4I>M#Ck%A6|DrBUAtu<
zSu7|j#C2Tf<j~~_76Ws@eJ~ddmPb4C=0y-X57d_B(fbt=B9HT2n!l-z1iTP-=x41W
zb}saiTY&T9x=OvcCJUy^-kX}#shfv>UKXDC<!uvs<mXf8PI;(0*K=%@#QO2pB?FH_
zOx&EiotpCkw7)bkg{>zv9>jbhyII{Y3i5?Q%K?utrz(%P5djj~=2ZnU<x$fg{iIb(
z$wkH5DvNcw^o!1<KlT=ih}>s|+&pNG<#E2qT}UmUCyERnIJ8-=&M#|?Fjb<&2~k0X
zO~I`aZn`gAg{^UcfQq1$G;x47^5|GDz<K&t8JS)Iv{pBLoGc*^G3GY#;BQn#*zLnS
zjL$@Y4E))YdFeQ;<F-h_p=B(Rmf-zw;oqOSVHAF_Y-GMtOHCG1Bz(`ub>ZSuTP~Da
zU3hopZU2?Hs-b(qLq>O_W0(dP?^my6Xje<ZJXBKUwoJi;!&%AG-7d3zehG++4@r5Z
zEzHGk>J1jh7EV?gCcZzvR}!92E4>h3bM?l>XwszagllJ~&MW@3Zvj7h#)F&&o5mRf
zk4~%li`+nW|6jES>G4?D@Ii|}G5$3h`OA+w3`ci`b%kqH+4LgPgQsjj1oLCJXNToI
zt?2J5KR8pIv}8LJl}KGcRCo>aKGPQ>T7rYIj~&&=4yfgB>8&FjwZvplB6w7YVba<-
zr{fvddd<C&rX8nqeVK<g&sAvx=l%J)oA0reYmN=2gf8l(w<Rmw1(amT_RK{J9u1Wb
zx?FDtnzpciPy|jp^&&q?y7B8kg(Rd|d-SEO;;Q{)`qdu|E)Mf!HgYamk{mC{4p&d|
zb?BF8Oc1PHrobxGDoQ8siu7B>Z#Swr0iK4gJO{Z+98^$A3iN=m3zSBHJUewMG@8N*
zU~C*E7E%3-ACx_bBA(<156p72{wqhI^A<qMIYtmwkS+)U`DD&zh$?`*h@h(kSt#HN
z3MYVS;vEs7H6gS&8;SaVe8L`Ha<!|hbdGDIQ!KO5^khR^!3D<JKYBWPKF#s*+x1EM
z%I~J_Z?bbVjD0ROd|xgr753a!PZ=yYP+MSp{Wr{V^Mtjp)eGKj{cJdtw)9B<ThWQ1
z`4K5wz8ILeFK04yUihRXenvlLeZxo`nDy|1<c)W)#VlI_f6ohVV+mh$7mt|`#{J8S
zj72^ldnuh)8l8$iX@NcxGO99ogwl<_3k3zxg4=$L@0YvTI9-lxbNSHHgDi+cpqB__
zl90-At;qp$a2E<(dfOoThqNZa-k3QgxBv8THQ+Mp?-zPq(h&2DbDD&_>+K;whV%OU
z^EcGjRg}K0e0Jx^^aI)-R$hJICOO~czL&lvT(QZLy6uR7;<(#zF;PxrX+$<<nrSL(
zQ16?1mz_=pUY<kln!?D-)ZtLuqkfuM69ixF-L^-ZXq@hJ(AGslOd6TI%jbDNfzoLb
zmaz2NzF^Dn#9%Y+U~U}hJ0{bGcxjc3`vxP<*$}D<bjA*6BSN(h7$p!1V#UFXf?fJ^
zqmF2y!a1J(8%1m`PV~Vp;acs(opJ|XksTL>qJ+9_T@d!j!xx1BG=Z|^Ad^IiR`6VK
z)pF;Zc2)@;wN&R<wwmex+1?hY>M86zaOB94+`(ytbI-aR!%F^K_quMn#y&cY=PGcX
z^fdhHg6L4ITyBkWPoU`N`+?~RmHEd~Mux@gK)Jy$-HQ_Vn;&QP{n_8xWS{ulFO+}v
zS;K*l#?*t?j@_|3lW;J0+7d}%zpRV(a96c9_sC{y(EO4whb`hi`T6v&+pmAGn>2%K
z(tqiV_lg0-l0x^L3pR!rkPc6h*~TZQSjWEc7R1CKG04yIpYp2qesjQoDO#%%VmUmB
zL5B?=v7PZm9Ks<4(fA(VG1<SMY(#A+jQXR|#z`A7O#e81!h{l{t=lL(Nxb8sdkFZq
zB6eOR{`ANhbwXwxDMXJH7I6kp30E|&&2#qP{vR^=2Gt?y&tjaG(Jn6@$|D~=^4IOa
z>9Vetx>C>EM<6OM`EjdHyR2hbdh;!Berlb9XF`Ocnga!Dp7F)NU&D_VoIhN@L*rR;
zb*Y}>LZMMwV#EUe@{<oY)@2?rCeyBzTxbcfUImYLSTHV#V;uJLJf?hJ&i8w|t8>!H
zNztm?dyhnywt3N5kcLKgFF+v;?%2+Ta5~_LDuIBTY$rG-#P!O>1*1QF*Pku%?}<9>
zqk)GSvUU65PBKS%3`j_T$L~a-Whw;b^hll1MJ+J$o@hs9c)LOGo0J7DxG1IIt003;
ze-MbHck!0BL*EU2!faGnu!=*wm*mAQ?^bAcV;0Urq9^kge$*CL<_ex|eej&FK$f}V
zZI6tB6Z*`E_9TvIvLrirs^#3NSN<4ue=45-DX*hx@n$UDlH|~6J>>UQNp$(0=$qfC
z3o~~uTK3n|a(Cf++uDa#E>=(4O;-+wN^D;H0F+s3jw+8Yxc<6+1*_-ow8>u%u$;t6
zOA}S|N@MhH6-H|IU3I7%{<LeoV~J*`?AKdA460{@#ZF87m3wnEqS67?zcl89zTu-*
z9r%GJj_W4y|FLJ0#-sO)9?A(fglN4<tk=0{gU26~foxK)pk_8?DdD-vTG|mPNkLf(
z%8oQhusoOE?X$Tn?dV~5^c{N%R2zICEhB#&UypF2|6(OCq&(QlX~mF+qWOG=;~#@b
zc2<KsO@~W=m8~?N`O~ugWv!V><w1hPjL4?VlV2o?B(0YcRI}eT8!v8~F3C`AqUZWR
zpXavEIq`A}`yL(%YP>Cuh5XFzg3Im4pIz{)ryf!W4+y;ZF$Zg?3H?-Z#3V0mL1xDN
zQh&KI+tk6B=l1K=p8N<4ug-(#MA7w^a^NvE&^HKubqP!e)7cHVWd_0U2yKCFa#3?X
zu1@-pL5L{+P3ZLAa&3kvaVO6ac>}y5kbWMXs08MTl5NQEA)UL&y+q+?3A;BnUi(9>
zq5bF`=PNYY;^M@kKaEBCoQMz11=N+M%5JguS@!5eccMj%89O(#SX<BA88%Wec+@mT
zA+h<v(%>0a@yg!YpWzZLE*06*x2In8t|Z+4bEJ8w)y?Hul0zJ+J@nC|jzcR?Hcp>O
z!XoyTzFps7^J?z{8~Vdtk*Q3AVdKGwQ0b1}6x;=UMPChWC1%b(Sxe0G>Ms;g^}gdV
z%pL^U-yH~Uv3%=xu)c9|r)=oU<CQi;PFr*9LnSqN0Sk}OhO@7V-8|R$OjRYV);+VM
zQzFndq)a%vk7Juo>rjdS->0#|Xt@n-)VQiaQ)COn{cOzrVbS@WED<dbRRzW;g8wAz
z{3j^e06f`8$DoX6lnPytct-E8^hX43D<OHS1*cGbvfBBzI;rdj!(tKhN_41lC;Qwh
z)B5=<hT7J&L_6PI_iX7U70<%TU{Ez<%HPlLm*nz`sw>upsCpC0E$)ij-6hV)i{~k{
zeJLTAqyB#0SlamT!oq}L)*toL*^3caK<3>>i(G7{TYG<8F(Zvdk6Y$?;f|glqtJ;e
z4hl&t(&xOSZf(_1zHKT|Ch=>XNcp5+!e=wLH0-GJ{N>Bt+f_RhTG!PAw>%o+UW8y>
zz7;5$P5)Vl0&5}JXPdh(Qr3w-ccz34@z-Lm<kR1qc4~_mVwOJQ{`dyMGbIr7!0&N*
zZ5et_@4N=p(-S}wlBazXYqBat%lvYxcbYE+9wlRh3sAQcgi!v-5hGEd{zqHqRc@AR
z3s$*4?XFzCJ95C)sO1}byTXJO2J~O2nKPp7eH;cn_DMV2tS9BX3n_o1UwvKgw`ruX
zvS`PO+$U=VJ5C3AhGHHplgQmA-Y0LlJszxD<X^p_->9_VC3d|f?N}ls;Z{SFS<u?|
zPlNbZt24X(olj>}2+6KNpP%yZ3k|rotpPH1cOI1Z@2T-i-1ja^ZdSYgUGL-J#Y>}3
zB?_Im!A}<U^bT42#PS?>oJqWY>^DX^ftmJQc67mpj6&z6OW0T62AD-H-(YX~$*%QX
z<(|H7*}yxrAH&@^)_p*||1L8_`T=|o6su7wZ2XA_*kI`GO9hv^WE2U(VDStZ+CO-J
zC(uQ7u!iHh^3t0I%95Nq8Wo%_mdd%uoj|QHX751-h7Hn=7u?Q@b}2UG*Z67rPnVvs
zW~+M>T0U(%E)5HDPyA6j(ds$;t<9!Q%TF}#^A1W@chL)%{k%6~n$^u4-D}7zU6&p0
zPuOtA<x@TTH!Y_Z6(5;*_RV&^!CFab$%Pe+BCn^u(aS4_Cdz$dUj?t8eNx+TuixxW
zbG-VvO--#Mj}G0>_X~`jRM7=J=WXh5B(CFEJH2I)qPH~fOJX0`UPI;|UF_Rm{%4L;
zz~b#U`d2S}9KwrD(`=;X2se$;&qbYFUf=R$t<a-u8NcAN;3yGVy}+K~nI}Y4G-|3U
ze*`@$103JzY-po5^mtg;Osa=>$Tcfk30wcuwj3USfgFWS?NEm%Y5>BciW-RIA(4p?
z^TKZBE_l9lUX<P(S?MbCO_b$}m6r;B*7or4;g1R#!cphtKmz1$TL)+%ed)%E&xj6r
zGjp0*#mpKJ(ZxQ$?-B_f!SkrFD@og&!n7GIm>CiLdqs~E_s@m|{hvP7)0P>|BiAR*
z>GEfrU2?c~TZIiCFSnX>HI@0rlfVS$J+h(HHL5!oPp>AcgAeZgrhT#2>Tc?^S>Y>!
zQ#X7q&e1ISbKS{yCZD%-<>m=fNj<{MTk6^Y`Tod%FtaYcZ>lVa@3lL0YH8hJ4e8se
z_CM+<Horo7VO!%vyMV>>Q(Dg@{d6l$Y)|@Wm0R-Xr=8M=m#6>Kn_ozAk)fwEaW&Ot
z;y?5$(GwsfK%o@=RKR~BEq)p&jOg&&hxXTq!2)Z?Ut;bzB=E8!ovKR`kM^r;5T-8P
zs)SH>>K%<`EE9n$pxZroj^3Vi7y0f?bv$~i-l#vXW4Z(rF7L41Vl9)qIbY_To|`Z&
zCGO?mw@^{0B)LlIOY<Xs^z>=&HhOoXT-g#x4xjgx@3!4-jW===pPHR@KF!#{n$%^t
zrN8jR@roDkr)pc8s-6FOkYRnNY~KB&zd5pRZx27yJUDdp#_hgkWy*Ka1>P<uM5T}b
zX)l8j2%l+z+RQ?B{dJVe`SXiGw&)nqPvXj}og>^L^Y7)W<pt0GBRKTSg$JPqg){HW
z&&?eEjhidjSGhJa@AAO8awCm*Wf|nl5P?CKjT6jOk;!HHC^Jewmu$Bt{1|Y>*4oTn
zP@ZSSv|yioGaM6BvxtGRW5|Jx#)lqwdJ<f{s0Z>d)Hh)4eBTHMIqRxyU)sR*n{%xy
z?W1b$mz&>SrrqWbfia|FB@k1i@2zEU%GIR#%JT&?Q<=nyAIk0~4kV;MuA@6<@wKem
z=Y}e=BOOm!JJwFzR6}~yjrq7*uYJ#qJ-3c~@xY9qLl=>;LodJG@J3C&*tJ;C(rU<6
z?%wgT<X0ZM@9LMVOMge|SL|_PXnuXasqy(0yK8pBrGAt!f7O3(4&DpcQ9<alsj<sX
znZLYa%be!@uAlij6$`UdHQRo5J90Yz;I~^(H12Z!#zN)%>FK0rC--QTPuI>gF#BXT
z^LO`O?T3wgn;Z%WfAsB*y#AlSJqe*b%e#It3}10G_|2P{gS7?|zwGr&xG{8CX}{5(
zbBR#6kZkcpI?PP5U*k<NziS46+N0jSunP<SZU^NRUDb6`X?f#Z@w4TTZ9!hQ>$&Is
zYX=`9H2G+o5wPOeJsKZT-i~+cd|(qF_yO8poPMFOTSEw2+)7BJwOb2HIZFko|BkD>
z7_Rz(T4>>`zYc{2|G!5VjRa{Y6f#rQdbCu;A!vZxcO^DzP<}lVnRM~Rk}aDx=R;RI
zP+#>AJaP2As24YQthisZYg2)foU(M+D%aJ8zKSAEqDURry)x8r!PSdCe!OK%J%ijk
z7v~3<w^VG4T0Zrsp-ZpWTh0ys(DV3s?O(;k`EMg%o;&({mG#Nqp7zz1rj#bWOyB|u
z$8)ZdZg{qaOE1@u7dF&-J1Jt84IQ1ab=l+$Gf!vx3$MKyt1%`c<>w+lUQ9~lyNCd@
zts{(2*bcdw`WH$a+7^)$zZh6hZZt(D=)W2QdxjGWT}TBjIkLD~k`(Z|mKs+(V0Rc|
zsYTrLD&VW_65?Oq%TBesEd{!k^)WTGLMcZ0D+=fVco*ms${{)6`hGnMyApX!^7E%V
z(winVRL?PXt*FfkF5S;HSj06!874A4f?#le-)TQ4p~xf3S#?d7_MT>X`QhC@gZpI`
zL8;Y~q}-e59;nDuj9!TvSK2<25MCNzE&N(MmEfW0&%GG6=;X@b=<-+D;nzNK!<wAF
zd|%Nw=t5dY!&V=i8_yw6kR5SgY<C|luuQHvIn~=v{{A%w{^rFe-V0J_$D>si(=~Y>
zmnMe~N9>57k{J?U6LYw>Y+;RX+a<9yn}*G6OkXp7A^Y@%^Tolx8tjeh*Q{_Ap6y@X
z^kUf=pYuOY-a4(cW}<9Y_1q_y<cSaM6`Nwu{zrRwx<c|zP>e10<hy6wlT)AztKD|#
zjcbzbB)7G}%Y#0O^a00#WsT*O7VRD_;j`93vJJ7`e^;$*2@7R-;u<vw!u}y+0OjV=
zUqi-_(e&aq0C-NF)~0p*@jujR9~eS=Ily{rC*kQm3yy@AmJV^MnnHiRNsa!{0O+y0
zhuyiCLvUVp*plV2y2GSoda{!JrJ5ivJwpC0qwClb_eGMnI4kv!PV2>M>DfWj{K^xY
zp7Q0t6n^(|TU1JHen;;wG})hdj{*WGmwYoEZ7~E7P~6C+EcVf`Zyi}oErUc&JTwm&
z)E4s-6JLcS@RR0lJJZ~Bb<Je4ndc;H=EC~QYfMIL?2`DZPuBmno3$hIUH^rDFmeC6
z3H@eK4{Z`cQ(OanlD0K^#@hu#rHg_58!Ab#j4|&<XsZD|Dk`(hEvg^L!%H5y+qhXd
zuqQ1~U%qZgWaD&VECJFgB-{jDZADRPX(nu8A9R1e2*1Ro4-KiqUqv2EI+DWfX!w7~
zde5*X^X+Rmg)W^0LqMdH04gO&*U&?x1QLn?l_o+Eq$nb`M2dj4(1d^(Ae7J)rCL$x
ziqb_y2c;+?SO68voaYyG&YA!Fei%lF(d#1jz1!Mrt({2>I_y`mAETBhwu=rJ>EcT~
zltBnn2}7r6jIbcw<J;|#j^c46axxjQxHUZGI9u31Ep7htJw7I0YVAN=w#L-iXRhN6
ziM04AEv!Q}IfFi^w?0vxS+Pg%zVxmY5}!R}W`&!Ji<ghMovtL0`E1Vzr@XkJIG+Gf
z@ozuag@P3b5wPcfih0HsL<uT34HhI*cTVviXuTNGr6+6PKvC*ahoCrJ5bS<zDPcB<
zzU}!77xeU&#kaKg(VV;QTT2^gO?O6}RClFa3Vaj#@ieWZf)<#T_t>Q(Krm$UHdlsv
zyQq7LhhllX=RvKA&}IW@(}pMwME?JI^DYVw(`5uMHvgX8{>i}LbNnayf?xfs<pBQ!
z?}g?Ow#Xy(kbUQy_g?^vM;-NDS%7r#&mRX_T=gg$yqR+EvQi~LLrfCE^0vTW$5Jc?
zN1#Ebg*y*`U$_#4OLw2Rt+kY3b|=79h1tj#J3H39sBuL4k$;sXDcH->O^OwGO(UZq
zfNd3Kh`5@f$J)+kP&3QF<lLH^Q%vF?EI=f@&gy(YR1oo3^X#n(Bte{TEL68_r7Ra(
z7m_e6=RC8_>bn^(G0xJk$xqMTmFws%F=NSp)XeL2h;z*DAETWn&Jw$Fd*q51@ri!9
z-zn0WEY_4GqX#!RCZ~&=^i}5xK6o<$k@rsGM#OXh_YD-8hq$EbXDJ&*W+IUfQC<((
z-@4>UuGj-8vkd<K^=%uViT@BUqBm%<=nz}wZ1FsWOW3q3Om)_&F`c2q-f0vL*&y&R
z7Rt6srfu{Kqo7s6pUYzT8fGFrPPC*2F>>X;wo>uJ#)ytsdaI$o*{NveDu?eTVu#P@
zHdIgt4)Qd)Od;wtw|rCLY9)N>s!!uj?0s{SU6ph+<yJh?&151Wlg$?Bq~Gyodg>!i
zm9y}k+4}22#+BapE57QYwqRiPE{0dD81Bm+xUJZSCrSJUADr>d5;T9|or%U>w(?=8
z$W}oKZUU~X9AvO%svbUGfy56Xs16WRV06h!anE>sc*Fw%`c`CTdW13wb^Hl!DCM5K
zH`C`>S#Fb2e3%^T(w!$oH6=+`)#QCrUl!dk$i6Ob8w_BC-zkr`wgXxiUM)A+lk*OA
ze!H65*>ILHo39+8jA{RUFaL@7jfUa>eA%W8)HYyaj#N3u-gNa#?a-)R*?scce<e<k
zVJcgRkaNWH$d_Cpo<Btv5Pbt}xZ4PHw2YC7C}U&@0L6}&u_6Z6sYRrlv7R@drzLJp
zKe*G!H8w)et(zuyV_P@vKKfi=?nvv&5%mi<aw0hJy<JvMSXSu9a1)ifGX5<Xlo`Xl
z#C6e8h!tTrU+qH^PtUM*lNsD|Iv^A>D;Y}N)Hb(On>3iyZX9ReMjEe_App`@PpUQc
zg#xzDd%WAeUZ_xJ??=xma)UGU`=F(J(L9yeE%}R<0>6LmH0D2${Nh1a>nHL1<-WoK
zbnQoNUH0Dg0oZVL+=yhZ+wa9Ocy-x>d|^NYAHu81z6USUh~+^*&k1~)0A2&gov`iC
z-DelvdIJ`)LjPuDfK2?G6bW!G3_HLH!vV}OedpY6N){Or?G}8&HA9OS6nMmSo*`jP
z&d7y;q4_wC2clUzYBuYt6n@#EalP_|Y6J!~YO8R}P^sfB2Ie)9d9%yumOmo{(xPjR
z3p;2$DRCzlqU?QsU1FuONVg;FS=lRp_2kYdTa)7kpN(M*$zGHZ9LCzD$QglAeA{;9
zEV+aT$rmx6*(ynzXMgV0k{HjMf06vkSuhR)F^d?;d8=hYQF^R@8Uwr5{kXG)-<+O`
zR~UiaWMm0FrfD81pr5Er2^?&P3pCiRPqR20jwTh)Da!Y-_pp$Udd>s7=n&4KW{9ck
z?8A9}K~;hJ+S*f}E>|~D2cH&<OTQ(=Je2d06RfVYa31qt&l9(xg_xO%hjMcXf~mxJ
z`qKZ<8F@o7G<rn@<vR%C7{osLT+Sc~J}%=TDKiv}m);b#3|MT97I-%Q&Vx5(T$>yI
znqdn);!|v|)K~Znx*cFDr#JckX%cvGu@M{_U$7x`Fnt3h6jui1p7`ed8Kvp6vX=<I
zbIm_NVOR~jm1h?{4&L;?;((RTF$nCEBR}Gu!1j>)j8tsLAS~5&JPZ9rcQs(bqjOVY
z-v_V0ccBOeH}yS+w#k!Ko?VuK{WG`CBG%Y^`Vv&9^$(V+n+{d%GG6K9@J;&{s=j9A
z&c#@0)wVy0c7$MQq0a@u^lb&;;=!p>EgE0wg@7bcy}_}n=$$OX)y4lnjNj)ZEq$Q+
zjv5B}>>HHn(q-lAC5)ivvk+J0MM9<5+7U7;UEDdX&^v`++1Uxjrl)D6myaW@<jl@2
z37DpCO(9Q_l!qghEo^l^pQEc3pSqjw@qzjvz$APdZVb2u;nn}l5dd;Y=V7=!Is?eo
zd3I&4_}?--VLSkHTGmdRx&`Wux=HfN-R}Wf7BpN<JXB%>a&w&oqhMK<v`8p`5)UE9
z5D4eo@4~9trJ5|h9=j-7|0)}3Fa&vH2~2V%E~o&}z=3mXj$p$tVXWn|Z#yXlIRVRn
zuGrUk7e`m{oscx&*Y(~>nn`NEFt9^>ms#au{l^75IPjT#BVWa)ux-A%J9STE(6Zh_
znJ+sIP%DmACCK^ceks+PsQ6<4I3kkxWmx9E9;|RDGOS_iHwmOd{s{Jsk^~j@!2f0E
z+!u!19=Sk;tn#O~VOR%!`@9P|i;kiLeU-`?5^~jZKCQQLFx9;26%S?`A3E6ay3AOv
zT&-FT&=|r2_g5MuSEoL5o-vNq)>mG9&dBX0{ZmAgP{VQYy{gWk<t3L|c0v|OG9}H|
z3^o8u*!-WD5!=JOOoW8-#jpjnWVy0ZQ}c8+uZ!j)1@;qeF!!`CfMBe=FiaP#HXYe-
z>v|6X70xP|FKTKv!9i|De!ae@nXrHt4L$w95Rqk?VS?yLDBThZ=m&pZbRGovFCO>5
zLx2Njyf~5DTdY9?VhS1%5OTaaz40ZK#D797u_unBa^dtlB3l+%ndBF2az=={0)V5e
z7=wC0nzf@)zj}?+@iuH`5(ex8Hu=nd+N?kl$6zI-XDTp-H$}1=oD|aHKDh}#iGv*P
zcd54^<DgphZj2^3Ir)bRq}j&#tVF%#?-T>I$OeZl9KV5S=8cZOaYM0oXBOZuxyn0T
zkdMO4&w@Wm{AjQ5OSdfdoqCu*R$nRa&>f{6H0ksS7WcJ`nyNJ^E!(AgwfZ0+%03+A
z-}KC&J5*(%-PKa``lc|-fLBogD=T%!*O~8QDK3v_3AWf@!!naa+Vh0#3;2|M48LXU
z=3jf+d>7~4ENtBtsQAbJBh)%(E_g4Mz^B1sLr6en40dSNe;SvKPyj%^GU&8(BZ8%(
z{%StpoTrmynrg+HN`8wU=->h5Gw9$0rSkE36o9=~fe_7dGa};gGpryCCJ(y`MVwEq
z85qFrQA-$WULXj%0nC%vz4NJiv=`g8Z3&hucwTX~V33d!6<3nmtp_JgiCZ{1t4<nx
zAMA)Yf_EaO#mp#>V8uDlKXc2%ZacwsxGy@YD@E)K29&fR))5BOooS2b;s%9IKdM(0
zcb0G^*`I8@ap`>H68@pxoJhEpzM1<8LwIpo<Ro3ab@WDrkVWKV{;YbBT1MPfD&L`X
zGw*D@CC9d?zp$F9x7chF27YL3A`zhyxQ6RUd@lIcI>k{xx@5FdS8OVytTJsT)nC(Y
zzQJm}#_C{J2=K2ULLe1DmgayLz($>Fv9d!2EVzwz7v>!cr~K?>EKb(yWA>ZurV@mb
zdqD97{mVuR!|P_WiqR0~q^jVKtm23hK<;PNDRN0=Yoe06P#ly&)`^89AnlvRSw1Ab
z!$wE>231{0Nf6gys3lu4y((ij+O5Vc9=|E^odY6~vT9D^lfKuc2t_&*L{`~C1YGVP
zvfjo}`42JUe{Vkq!j-9JZyJoL9a^mh$R8LS|EUKX$<gozi9bXuvCB=$_Rbp6+yjxd
zGstBMBt2W+*jT~#AFsT4Bx%L#=z^<JpkaG6t^J|m+F`Lr{Qieih<+iA&9_y05?oy%
zF^Hrcw9?2(H8*Ji%CS-?Z8Ge=L;$rB;+fH;tJckL^}!)6O{|+E!cRD#xGsi&s~~BM
z6D=F>?FlU_i5uTl8-~yfYx8}R6F=OXa>*$*;-yQd%DtF&*KQZui*s|OTARYtZ)_z-
z*^J~pyskB%Xiq&@8@BEgMDvPSAT`k`X=!(8*uY7QR9^JX1qs8;X7dMY8pSU5yhuLj
zp0=F!W0c|0^TD56vD;|I(I)!xRK5QG#zEy*p9P0Y64t?<a_S?k#l$c|#R~0LF9f95
z8><i=Ar5%Mu>1g=jRxTWXfcMo;=v;n0f$*ZD&*!7&k@k*X9A*iDsOiKCR265Vy07h
z%5c#8^SrA7&C(XGwgoGuq~hn;#V<~c1aND;yWF*qGC6bPv0Y1$^T0fKc@)3|iok*<
zFpT}fgOaxIg5obs0G{|j)|*=`g+4U(0<*81>h6w$UAk_5_W;ov{m6jf^fH<P|0=NM
zUDqwzRDK7Cum0B29K-!~S%b=rwc)el%@vKO3=~Up#iVF>W$nFYz5WJW@WA9pw<Qoy
zO}>+qYb0pR;j^dGoW)=|1pI!jmN6ZN*%5p%prjh?&}3H!zC5$0Cy9KW#a(4vX?+YC
zDwfWCUb<SzH*t^q>J*yea7<q1G6npIn|UqdrC>-mU{{MB^W0Yl6}Y~9c<0Z-*-WCN
z6h7M+mz%`P9YCtYYlAis1(tK67`$GI6G6W-D64N+SAg)O*~`774<W>~d+B|pLRC-I
zf*)e_#A(M{6E;LpQ;rMM19QMLr3hL2gIlM*55ilz?Fc`+u;*1(Q8>sw`HZBDVySq%
z(7@t&!P}~aVA!pz;^kHK6sCv*84B=TpfVWR;FzoP)DZtC3J?$qca+5cR_pyS(cr61
z-vu_F{y#+oxN3N0SNj|XnLu`tpapbK5LY4~)NK!j+8UGeECKB6)sk>?Wzb_b^qOw@
z?SuMd!S`=UiZsLKuLt$!dk7U8UWh#7F-KnJT7=kChTONt+#7H%%i+V;92f30o*qov
z32(%2saf63An?6T;=kLJw1uX5xC9G%`<Ecncs@&7;$~{ChWkRPz6n%|_;hzokN$OI
z&3RPs64mB$Ww%a>dAc%-tl9?GoU9AUF$-t(smH;oM$xE{(xHj|{ko{nTK2mrjGKp8
z_m+>=r|fsRXy|ok#z8FE8|gfMMgrIQs8G#Bqqh7AOZV(;$48?9_fLuxx-55D+QzIm
zAmYfJBe|E!j@Fv<hLwafM+Ju&e1Xe`=Jrb8MIZrZ>y#)uy4?+G1RzNnWbisDP-=Me
z0kBk%CxcAL8(-y)<^_ScN&HP6$~wkTS-MNRyOrpev<sLI{NXAz@er-e^}m-vj13Gy
z3|Gd;-X@SG)og%3(A{zvc%)`{GdYIJm3){LQ(HTn)#Ek{1@bNtei-IuZ-k_xcfGso
z<C#IXgDTr9k58R>yWrt!l}VFFu~$C)<Tm^y(P`y+U${2AFPSB#xV^~KMC#klt(vp8
z%Oa92LW+c^X9xK&oOF7l<|{IivJ+x{)nbLfNqgZUku9X=VX{?^?O3usPt{IBiKJXz
z;18+FJ7}y@Afxq04%b(D<W`m7f^7$TZtU?MnYGFm?{-`t>a>8qMeykw6I`c((x)Cn
zoyH(Nttfjc#Q)A2^79Jfc>%7_O_@&~{I1GGexBEpo}C_EOcOWB2O(!VlJswVVLF?j
zj=1hUE}PwPLhgX81NdN)K%&`D^}{NXMF-&%KgZ0m*-(ytCvk7V%-n(Ty6r<5^9+#=
zMTctjtni)+2!5db`k-6<ub~{g2=mcE&gU!j4ozQ!Gz!FBGYWT8vjlc`3%(-o39o(a
z+Ww$K980eg$v9CNE(i=s7kdXZgw)On0aXLP$DHX`0e0;AJ)cY>o|3<%9h+@omYNEf
z$Qx=LUiSO{N0vzbu78>{P3mbl0F3?@dSo<Rdk;nPEiJ3i;l{8#y{QR8t_BN|4g5XP
zM8FGzk-Kt8HuA03&$EUdu0x@67JE|I#joqTKGz>orRb9mTp>QVi(f&)bB2?bsp>8H
z-lw!nY-#tRu0bROa<(+-977`FwQOMqXlsQ8bk(kk<|oSq)gL_l<r~W@U)r0(u3SDs
z9JewZ)n_7x>C-!?cQ=LOoiz0QTpt(C)3wvCP*X#XU1VlfxW%Z~78*>{som(-mXisu
zZ94~#^GKQMtCepFIOJSd8-1>?E9k4Cx8%)i$ZF*w1U>acOlcjLqiH_42h;y;K>zuh
zcFTnUPb*W_=VJaZFpl6{-9r_@ud_eMsE_=5e4Hg}MnH<p0zM(V-F);@%VxzU$k12r
zmGK6(-ZYk9y=W(Bzjz-Vuc1%|#!SG;Tugmqp5fFO@<@42qC}wPJI=O!cxJ5h|3nr@
zyS&KSfSLag1Hg{f72hWtPY^%{VNhf>a}vLN(xUrFevJ)UQamp*j*@E%M2+W(K!kV+
z$`<W&UXZ&BCEanD*p1HaV_H?e&bGVPVg8i^H|R2f<>VXUy*wbNYTYX5&6NgJE^Int
zc69o**3pr<Be$Hk`{#OV?$b*r;D&EIheS(cT^GfoX8mfk6Q-WYZMC)v2!4FC#ou_~
zpaayS#&MwZ<EbjX2wGH|2}0;>-mUQc<<+GEAM);sY=;+$1~9vz-+E_HlBE=wYURfB
zH4bw42#li>QWuduI;3o(DNpOf#6GWv{L&&!1*UxrhGA8dKT7ORDpZ&ge`W$>sb75U
zcaO~U82sdsU>xnT1z5|0z8JV2+rkw11sExokaIw0HRY{sg6$&57MfUKy##>nkOzXo
z>6#T5&KRhRn)}-hlFKPK*!?uEv>z;}dK+8dQP(b8ljy&OOKldXz5=TF4(G8TUCC2y
ziD<l7w(l<Y`2rNcj-{-{g{zXIhmwLJdW72&iv;V8?gBBL7ct-`l50IHj)CdR4-N<B
zEaHXi?+BGY;+r0Z_%=lJh)sMS+gm#O<;WXHkqTllnuo3M_T?M@1_*gDc)b5qt1UFI
z?}Pl`{*<zJ|EH8evGu_1chBChNP;8hMQ|<9SMY;~1(25l(1+<+*2eUeTyoM7dd9Gc
zWRNG)AANWi3a12j>eD7rNPWW;{s&7wA`jiAX?kypqiv33e`=(^sLr#y*ZX1bV*A$6
z2wM$zqte<#J6Jn&2An_X;@}?GsY&6@%z;A>{r|cjy;fN$hO}*J89tvbwAK^%G<DuZ
z=TwzO(k3gf+~tR#dsw3_Kgx3Cf^r)?7?S$Yj(RhB#vDGk&MiWIlmpJBSEHh-Artux
z<{ce}6x<poZw*D}ZIX7znjT}@ZhB^?D;7lf4+J}av&P&ij3$Tl7yxJLlFs^pkQ=8V
zRotbey=Y0;@x~`KVRN!1Ar>k2Fsf>hq^xWbK*p{E3Cv@~6f<;ncCXin0ktp@bjQE<
z!!Y2M#Y4@&06@etE;>+X09I)F?ZeEFUHajXa2l&G2in1nG4;>l@i&R^A8ay-@jKR7
zfPauwn)}HLZW_~9d0(H_i}R(nm{Rl;E5NSZ7A9R@_{GjiH28@>(xI{p5D4DUL=={;
zMt-9x$>1d@L1rZP=b^9&AGCa<t0&HXq^AlQBahoZCT=&B5~&1dcR4bllNHzA?+#P;
z8HOe)#riQ_&E?}nQ)^SoI73Uu=BZEG>5BNu<ZLa}Rhn(4YDROA^^6rqGLHPnJ4H2q
z0WWppO|IYNdYex(^J#x01{3T|mW7sB4yfoXA+{~n-P<hoOEp?AEvQ~vFH00^C5ncy
zTnZ826=|G~oEt0|q*|c+L4UCW&9@{Hb432CpsHWYFA*bw9U?dgpp3<{AuOwoV>nU2
zJV6g6Uuz_v8UX`~fAMDd${Y8WcF`+nPt@c_eb_;R#3`WN+86K=mXq}|2|q)i%31ME
z$B~$0guV}WN!8`37hZsSn=K`AwT~_2&%Csfk*9VMa;Uf&;;FA^fqgd{jhB@UD_@@H
z$S9ki+l_`W50S9XAB&HqjX)EhorL3=!mGPhA#F;5BIzR*jRc{In9BE}dN1jc<=f%D
z@dS4xzaGKc5Nn)baYi@y8nO}|%nLa%5o|o?|GBk;(1W`5jMI1A^9%qJR$|qLQ0ou!
z1oE|09Hf2+@p0P+N<#mkdUdZ&pmhvp^H&AY5$OF%j0{_yDUbSGYc61fv0pcD@BQJw
zYb~ZLrsK{Uvj*q8e@)xSkMtGiji~g0e7e@;g1rH;ia+E*)SHh*<Hz-Up2y3;2xW^-
ze~Ev3Af8lY7kLqzn{6)r>Y7k}tRdeqXpf=5l22o9*1-r{n?df9Wx<fpDk4eu)-*|9
z_gSv{TQS{j*n!<RX+1jbK`JDTIXEnh6$;EiM(svFYR`%G3&z{%^4pphc@(6D7Pjhr
zsWIpm@pV@VdgfZN;<-B3`(%B-##?5~0QP&ciOY-c9vTUv-+~lZ=gbcc7YHk_+j_RE
zihvjU#hm~TK^YN9bX4>n<^EI<Wll!Tv>+ae`Nt@L@Y|pjL>U&#$dBD9YCM*BE(rpR
z?8E<R4z?f8QQz<u+h9Tu^HLs<nFaoi4elT)Z+-xlU<Tvp0O0mFB63Yig$q9`H1C0}
z)~F4o#B$#e0r;Z;P-*ua^1_*Ms3Ed^0rss))yYv=t7k#AF<&Y^F|A|Jo`&^x9o+)j
zR@xC}9$`gYva)(^HU3QNjw%L$AKZgc+>CqbjDa;Z3J9h?$=}9_A6%`mxmMmaUW)Sh
zK31-g;5j*=8UM11{A<Qqt=`<PNHczsl!S?xNe#AgtWAH|TT|&Pj>_BBG*~EOWNdLd
zBq5$cFjf`ZQnRamKv$8&?2@*NDHd;ny5?qJ+gEFCE^a##XnSw{{+i-g72l^v?RHP9
zfk7?sdrZ*C!3jAIo8?*IA7kJaYtoSkJ>#`|q74)SjlWbx)!`j19DvS}M>_%U9H4*m
ziWnCW$HEiKStry)AWq--SbMjA4gG-c>FjUJ$2sK4@%?-h1xBZwr32e{2}UoHQuOfB
zc6-}ZINwQp$=;1zL5pC?k2h}~s}h<xJ(5X-7+p)-6)akRZ<YklE0pMux|38P+OUfp
z|27zU4O@L;zY7uKGwZ968Upzk3{60-3(o!Rj-YS7)&3O$-8)16>U}vbC6{{-aJv5m
zTKvy7bHjK9av;8WAHcL_>P-QE<M%V)c<{sk;2IigI&<4i23t~@+6ZC`IL^QZF;Gjc
zUb`iGlk8A_xb}YJL_r<R!bv5G@SG!J5gPvNj4e{V{$a;qVp=QTC<_^2Ry==w4dY`X
zLhz8Q*vzdPd+j;2BQfz_YB=)A0pjBL;*wPI5Cd6mGAA$EJ~ERn^Xt{it8*El$7iL>
z+v=V*o9w)>^LV05YuuQFhwW5TYW6$+8WVZgIibxT<{NAx@k7455g+vFSI4u<LiF&3
zG7bZ}4^wp<s3+qG-5qD=3+TlnPi|d?tl$eTnE$0`Qj@6>8kpvd7Op>S^vetN?tbM-
zakKn+^JcBq{_~TKIOU7K{BZ0~nN(k$^UWT`KOeIABo&>!1h<2|-K*;G0I=2&NS%1Z
z`-t!xm;@pnfEZ!Az)S`Jwg-VtL<$gtg5Q9ToFN<b!y^rpu)RDp#@}xWFQ)$V6u;R@
zR4njgs?@h-4@1G5lsU|GjvHj;CO>rVmFA`Zw`!4Wjvq*CfS)RuT7EQINimm$^Mtgv
z)~ltL@H+(NGf!L-9bc=1!&3_l2&#6tjxM4sdrZRB4N3^0SNQ|<0E(A}67Dl3O;?!i
z^J%6m-$&0pLzKqD`t;gVv*OTgg@nu#kL_KEEtk!Gdq!g_d*$@ZbnR6u+*aa$rmqhs
zbhigTIgmT%Knm~w+nwN;SR}|~5baJNt&j!T(0Nv;xB$jV?ZSuniKvjS3xVIkwV2QH
z6RXw2d`BVU$@xNEDBX8j_$B!hOlfTLaPuhqRnYy`D*>K+4go~4a{$q)5h9^N*V|4G
zd~|0GW$rtg%NI~GI?1DR0o7Q=fj9_ISHulN-&tt69{`p#n(!*5Li$QX+@QX!Izw^w
zOB5d73Zo<RbOd6l_CbfIXDotYuy*sKTq#`;ZcY<b8WM4kJgba(j<k}(xnRlCuZqyB
zh7VLoE;NH+C}R_y#}?3y(6^tlGU;|v9VNqJ=N}uah~$<r^OWKzv!s&AjrW$Za`x68
zOL)m^#WQJ?*yknFU)VzGc=40y#A^bknQ^{$o)*}DBOiY#DG;7;e+Vo|Ke|u6Lk*dZ
za5oko|HK(zr3J)2|9XQNu;wjX+~c7HK*1s4UWylYA8<j>E+GnrD#9c!hwE|@HQJxw
zG$i(>Ze_hX(>zSAjI_5ZZ#VC0wk#3iq>FI#vPQ3Wy?L;#CDv*uwpM$jzd?w5^tQu$
zwdWd+Vc$J!-4wzzSYkx&icK=-2yZ4YMT@N4yulSamQC$T+jHTBl7`LWzliz$Kec^J
zB&bF5K4a1Xasp+?ANrE;@OX0hFj`Gw&mBhcFvW4bRG4|4vi)0)61~L*mzb!GtjZPV
z4>+(VE$wU7mcAVQgIPJG<@r`SwI2B~`{u(TZ{*K37k|+*%W-@*U?S0bm)JR&;e38{
z#7MGs$<g}KowN6Ar?%b*7La{2Zx}}C&#~qsj2e9}cM{HW^_JuRU$e+_@pP($I`HUd
z{hUf(uebqt@vFnOd9NYM!C;opv)BJW+ri*JMQnpk7{Vs~2{d3N6fMe<)JJB{*IupU
z->J6u_u(keDN1D^#-lGSpJ~>|iuuE)DQfdfyM{txRpUkP!OB5xqxL4*#6-kLtgJD)
zOF5Go1N^n{zzBhlq_UHGMQh3Iyz^_I*|t&$r}niu^^?*R?-DsS%wkk6l2#(@cu4hA
zZICTjh-r;QGGi;$UfxZ|ENhs`@uf7k&!i^k`r8Sq)Y(KyP!DU0U0t=%%1X?Vbh46o
zVkH$UdIesxS}g#q>>YJ)tqyn`gRIr2==fD(<p~yIqy!{w{L?GrBZU?ixEnU0@VMQx
zEj^Jg048w(x>|!Z)aUz_3-Oc?1m^H*thhlJm~VMB6X5zS^e(vj9o`9j+z?0G;|mLh
zdD>ZyAUz!O*@2Dsa=x=@-7bo&N#@?np81gM!wNf;$S|El?5d?F2qAaNsgVwN8(Z+4
zg)Qc=qWD=6K8H6F%j98OzhSyM)P?B!`3KlmxT=(#06LpdZa;wQ6Xu*H`a?eTTL;7p
zV_?Uafv7VQ^?U(yXY2#;V%YB3o+EKJ3oE_~hbkbtFwfAnC6dv@&n1`p>8Rf5&ugn|
z)lV$A^vQo$w2fyBd?G2pvH=8I@1H>1NFd-{`ENOMsNE=MT8d5Xz$KYSVQS~9?i+>r
zsi;l6PxuL^t<_a+PwbM-^(6;}Ys<JE?hg03QDnOwP5Gi=PrSuy=(kndOFgLdyoZ>L
z6@2A3;J>;Br>!*$aSA-(^=3IEq{_eUkgd)*dA+e%>gmf?+zW65{a{;X>_X0X)71U4
zcU~p$%$l)qZRlufXM1{S)-{=X75@IKjVE31_Af>E)}y@7c&&09Jr%s?tF(vFf#(jy
zng2!9aqlj~-h3;A`q1T`y+3Y5nYo_ZrwOFsPLF;Rr5&#;<h=YvD@$hX8^NUrngmdN
zE=U)}c#2UyE@Pgs2W(ypxq2wR*_}=(R;t@0r`Pcvof$LJg;U;l_1l-~?cx+Nep}-#
zy1fLeYr{T>jr`G~II66F8`h%#&(vF1AnX*x7LWnw!~Pw6fO^M6n}iw#X+sw(H1?>M
zF;#|mOL*{K{~_jVQ0z8RA#M;Or!ojM<GGM-<i4jGbpFheyn~xz^sO7f7Kl65!mE|l
zhmiVM$=+wUtEDF(h*VQ=p%%7rJzw3xkW$q!=9n?sT8#{t(pwyc84`$hkPxIiz@j;g
z$i6K<JPwX}w^M^1``a#P8h2~)t8I<+9PBEyDGCkv=Ca^;lk~7;bz5h@R7;;l&E6!G
zu>$&<-IUU#(BH~kWU;1C?o#@FW`o8*X=;upI3@MRn{g9;4|$l;#W&}7HB&^>IV+P5
zDuIWTR%~5W;(?yc1ie!<&9YuQbnN0Rn8!*hHK;1N+h8g;TwX{EL=FV@mLGpe;Mm!1
zu%JN$?c<+W0YU0mv#fQNIux6A_EI4%e-VhF>TS2fZ%_zmIPBE*WvyO@d#lvoA*4=|
zYkSNa3kxl#=C0f!9Y7Hh``~Cjws7y;TMRV90(<ZD0EF1K9P$AEEm+DNe@p?L3oX`V
z3!-}5r&FyS2|Ak%!cp35uo?$p>k&Yp?azp#lvx!UDGl;nNN;x&DJd(@gCd~&EsNI%
ztd(Bgyb0LO5EsQuV!COHaU(Ek-3<=gpEuyYEh9g$Q-s_Vq}>GDU`;C38i3It_I@`z
zbdmJ8FwP6HLoQ$#O9K6(EnM~@Fg+z1loY$aUHYi2ZSt_xHT|JTtz^gSXtG&J^MxR;
zv<$=5grgs39v@qF<7VV%@<}Ajiai=-%j|NfNX^zH(Tb9iXQQxJFKku#JI|;%W0#-X
z09o7V&=Fk|)PP%#ZCd7wUR$@kl-%w1is_N!r;4)M7gKY7n65C?Zf3>wP#caux?Qx-
zN^D!_y=$Q%0oo{7j(?-fEal)QykL>DH2Lbt=F<Wsex-J|(972DRDUOZdGtzFdyT%2
z!YnIB{aWrYusmCn#>OmAQVzxojP)(LSqG0OFcT3CX2Q`b0+N-&A3RiEZT*Pc;kPDv
z`Qi$fG_SljErZ-;b65LUNU6x1`2l;QsY<05bN89A*gcVD!GBLzv8Us$1l&~xvRm|i
zf5b!xW&oo190_+w^@Xmu3_x(r8sB}l2LpR}YsXR0{{49zQvTx2<(omB+q36?NAdp1
zXn3I2#`DM+1sx?jU_5uDqZFz{a^QI(GawIEf^092AgMGz^b>Q_sK`w&%DDaROOfn4
zv`e!wa%w@DRKrgc3knBU3&fN&s>HFt%cEmZA0HAd3D0QyI~alk3415W)~zkxM1`*3
z;QLOft|&41x!M`SkitIXi^e3Rv7ef|<D&}GrS<bOWcwe_{RL@Njmo+!Iz#8(jSX2-
zCUQ0OLc^v4bqQ-1RlZckX~X9&!Q`|NNkU9bb<MDNlvbIB@_Nqk$a|GQMGt7YEJ*8-
zxKCME@(;{WnZ)ddw-Zu7xydK#{gxlR$X70^D!-5A2wm9KB((r8AJ9d>MJ<@ZNLfc#
zxB>}cVrU-9U<nX8%N^WvjdD18WUm~p&iTtiDrBgsIjr=1b*)S1A#;hiq~NsL;zlJ1
zq;4Grrz8g$q2a|NQYyXnG-l9&*Bxx(yqP1!vs;Q%Atc?ndCmbEA0ab-<P$fK@eYuP
zvZUjr9Bv&2+Vl7yx5)1Qfh2P2!|}U!%n?Wd@!#BFTbASlk8+1?FaF8_9E^N@lMn6%
zVW*gYqan9`VhhGsPJg$EfM>WFnRm(IUufe-%fj!1gali9sxFXl&OHA(-l7fc1U5zw
zBqA5;7M^zrFzkf$f)&P+eaPp&59+;3D8>wUpBu>BZp_WCGwz=2N)JF)gg_KL<ckjL
z5WSXZR|T}Z2D=Y81M}Z;wfc$hv2pLa{n*>7jDub>zzsRaN_bjSUn_x$w7{0?3nV(s
z%;8F`GZN5f?4Gu{Bo|_yQuut{`NtQ8!=0SwAZ@sW7u|Kfin#=eC|Qk*5vVm~4s)e@
zua4eK26jA^D(n+-H_-lvgYL%-T2=S4?T!uH+tw^N<BoC|jCYdm(b=IIAkbhRtg`Ph
zzO=DobW8b*_Fkdg)cR<y88(A|@=lX_q>+DpYJS_pD5sDsQhIB&!eva_V@qzpYtu*9
z<mBI|<U%6uS6idzk}Dtn-M|oQeH-a{Z`~Ffj`O;Z`xCn`=!WlK5jEQK&h*(~|NFdU
zG7QiV#nhElvbp^PT!&3|r9_vWXo%KsBNl`iI6y!%3Y*A=B9{Mk;<+GstR8j4E$|wZ
zYLmE8;x@XcjoOvl1WMk9A`|WeV`Ngv1RkF{jc4}84{9gqurN3wZfvY8Wn%OibMwnn
z9)IV_O76r0aeVk;%<D_3k0yZ78*dTvI;#uVL@NPL6cKDkTqIsN8!J~0O_(pZpM?l7
z?N#$|>cpQz1-h&3%`^|%iGALz|LjX6nR-<3-M0QifAeXX38+|AmA=0GOp|+%tmk99
z&|8-x_(6;U&7i-wWZmc|%5K3|ZR<?E6vxU}GFx%9AqTd+8d)c+$@z?_5OSw~pbZu&
zN~*)WgbocE>i@KJ)?W-l^FB~_SLd>J+!w<GZkc?>-_iaUuu})SM}{+-&my*J8(b|t
zDvnY~S5JmS(H@DSUT1AmS?e3{#J~g5_;fQ(;|}vU_mpxmIseGi{NNV7D;#%Hs>z2B
zdozRsD9Kqxge7O8xDiPtPA71UE%<)ZT9xVym%T}<_jWSZT7m{C6tJ(ZlGe*8Sryu(
z`^Mz^=sCw<%NS^Tmoj9<QDFTLB2D4wJY6M9;w9>9^}fCGm=LIrLDGE!R>QJaP=&6|
z$fl&BGjn?_fJANI{W7I^N6T^E&j@G)=XeO$#zvLRzrWdE3$=N=mN_aWllrH){np)q
z#FFZq3v~`py569bRp?Pt<3viCVHC_DgUrl-&A4JWoa<6CH$_L%F6B3d&-ZzyUw@fA
zvmMA#+qg~s%*XteHbJb@{Ccv!=V|dTTk7Wf#ZMsjiDb;Jo*K&y*Anx39qnG}IH*BB
z`lbR0r&yW?ch{-;9^*_5Z@sw%o2dlXp%JzEVhj?i(~4@|Tl>Ndv1!l&n93>fJP&(#
z1x}QnHY{CjQt2SG=PA-TWXofdE{Kk7*ByVYj)aV4G5FP|8aoQl)w|}unmeqCQwdyR
zuU*%<6L>yGi_`cj6IJ7wZ#o#3W60^3o%$8M_hDpb-L4PgmGHT0OwAj$%W<dn3D>Ck
z&Gl97{(j4!uG)&?J300($rdXntGx9lQa>|4TfqH{<(KN(gN-@BT7cK{6bPLFaz!IU
zV8Ip)Yve=6o^lsTb<o<OJAa9ur#7(HZa{<I6tffTSvLI6lz@+Le0}hP;fm}&E;#GF
zvGCkTcD!iYh%R2{18M`_u!ZI5@nE+<=j#9n=W4ltPc^dFKZA77<YVPirKbdPCCK}1
zuL`5!&5r3x`_pF0$<>CCkaZ3`cN7Q-LV$lf=pacUe-VLk8<;$Rsi@S|C)6jm5~3)i
zghwZPk85tRJ*bOWuU83B=LSXInUebEPT1w0Sx<>kU*5lU+cQ)L^X&Jr9G$bf%rdYY
zj;CB-CE1gyBD%9-W-fAF(h83AR@bi0pl~any9u>QSLC<+-BZok)+9B0DD@g)A{yYq
zWJ$MKu#A*S1;}hLgiC$JlEAiaNrzYibqXl@)4s>5Bq6Ssqw(FofvhNOKo`VW2oG6&
zSs~2KA@Hdxz{&R`jKNJoj+#hapnrsz()NjEi4_^@@Qo8FxS{B{8OlgcHVZJ+YY*?7
z9rGy?%~lfPV9bIF>U6JEm*;e#DwPwxhgp<2C7>(s{ZWq48CISi<gZ@7*YTsWN+vKW
zSaK~!HNzHb?t2cK9KNu5`3yzp)+>fvFbqE@?n2Co4L^^7$`>mrpaXnAmVe~%N!#J4
zOb}zatKoN6Atro5Idcc!<luW(;+I@a|Mo9PUa;Ryg-B3|K<fma_{Q24OlY!ow&O_x
z2^KZ8$H3L&J-nSDkLvTUGDi#;zh22#qLo!(eRw%8k}CNu!H4N;^4uoWIVUb(dV2@d
zIaon-zOj9f+tP#zP$^0y+u$txr%u!gF>JAV;^rcWkK=~%cFvJ*4-^+HYO*fOMdW5Z
zoZH<@8shS`cO6K-ht!$;E}c}_HlQq3&`MHr9lwzJ;9Bi)c?kYp2-G8V`j%2(O=4y7
z=Lid};YP73mRXeP=shXZ6>C1jIY=bQ!Fa8bkr#N`;vMYgi25}XjY&>w-;LY;1hSKM
z_R9p1?+kK_b4l~D9btlx9%0+Ard@O$sKb{FasEbbdN*wP_mbXPMc>pG=Pn10?4U<=
zF3nyoawb2yx07!_tK2;nhIE@(5FNdi`myQ*;^f$RY%47v=b<W6R-r0&Xp&vVMEa|*
zW3z1pAKBlw=-$loRnyM#tbeNgZL~MIMEyQx=!#1mLC(#xC1!s&+g!n8F+c;s^R@!C
z0zhPkY8m+q^6!;F?rW<|@QluJs+C+dA1g@L&_`qCe=`uasq2cE4XOl77y`dk$9xxC
z2W3$Fcsi=z6CY2m5B{=^wVW9499^;L9XM!22Rl3eREQq%w#la>NH4&NxQJ)W+ciz+
z=U_}r=&lIuA;8f8+$Lu2xqW;0?fT(fLBpd8sa>`5nnyL`VnTrR!18NGs{HuH0XUb#
zcjN^6n4ja1D8;1OT%QMLUP8GtB%kS_{_XzvU)cQ=s0tUXGTi1B!l+clpdPpqZeC<T
z`rFRWt28#KLX!=H9}`sMOSMkA8}SG7#TdHkA6!`xDw4C<EmC1B(PZ>v%2hIG`BH^|
zd-n99HtM*xLTIaqV1?tQ$H}&8zM&qG%rD6h)NW&B($Z(<qPOJ>g*A7Mr>5K*ZZPs`
zMR4GbE#E;0w*fZ%^98K=d(7Rgs#l|=;#2bkMh74!_9hlIut4lB%5bO^=`R<;z;1n3
z><r4DRC%v&;E=9L@z;&d5U07T_N#q0_Xw&m0xB_QNrM<;gwM9fUA-<ZE1Mh19f<b$
zT$gKavYRdk^?dx~{CZBR{;6UTJ2b*XgO$ZX4`CIe-8mP4tCSMG6j3u9nT=(r>`_@Y
zG~KWJ>O4ZEb5ZN=izD~X83i7#R2EpBF+r+KuTFm_K{_+Utmnhd#B%tOr`J?V?yT{5
znt4YZWeXkud6$Sk!4AA>J$>}Z8<MfsoOgAla{3cu@FHGhDlN17f8KH%<G$8da@?>l
z*e&>EusMz!fHw$spp)2GrdJ2DMV`o;bRZfhRxVaxy={*9U42~pdAmJKX_`{yUu9f(
zpzz&E<B=QK!p+A8E<QON&@5~IbZ3kFEtb0vE9iK;-)7jQVv4TzLnT4v-Z}NFTEjAd
zjj%@+p>9Gk8ZOk};C$E5?sk>%()QTQ>Uy&<OQG2nYg^lkGwpWET7LQg^-ig6PundA
z$3nKBua}hxlsNY5B~5jH&uErbHi+#wd~ILV2b+aGk?-xkDjlyG-p2q$W1o8U4;A>Q
z>fDL^Zowd9h0#)x^kD^X#`*6xyk5^<X%n@vE-4LrRuumejq2=!dRLrj_zrHD2JFfs
zJTgua5e>vw{wDg-lb0fnV_?@0GVdwMM{*A-@UH>ipX9<belA&O-YAZ#Vz2&vjAqz%
zo5U1pnEzTu+(m>aZM8VS*ma=0B3;IA_fePI<`I%V1{2(GZoM_7CAi3Z*_zSq5?M8#
zT>zMpxE8Dt&lN!mINO2?kAXg^^(v$U3NwOc+N5HG^xq{0SHb0`p96gFY`XX;utxy`
zu|=xxaNxDBe{M$oS^fe1o%g(cFDgPf2*_F$xVQi==Rr_q;Gzy*2si@XXMQKCZV&j;
zU9)`VaGV=vaM(j5<pLv{_Gr9O(e-1<0kTb#QbrKtYIIK|wdLtLToXq$61s^NRyxK<
zp^w`=ctA0jIuVfS;3K+zOc-(M)|ltrOWy?QZIXKTI|$FERCT@4?GW>@q-lF!G>7e~
zY*HSnP?AHPL2f5-QXUsE@}k7_PW=d4PdUiF-QRp6^6@osF|&G0YeFTHCE#{5Gkq*2
zt{`hnMU8Cb;X%!?edu1U;Ud3vyTMzUx7m><8ie^-+(`}B@R%{LwX@q~$^Br{iKx$n
z&dX==G~S*=ogMdX>^-K+nh$cy)(ziT?2(&RTVJ@BVLqZAT>zBk8}IW!U*Z_&f>CqR
zu9x>WL1ukg%#ImqM7_LES8Ur#jhM+H2x_BK9YC*-fLO~D9z@oM#@f(?7F{(5hcP1g
zN-?#1&k57ej_UBPim1_@s2^jY@B0Px_L`upC&X^du&XC0w!T6%$jX25HKiO8h{M#h
z?^FssaF`;!*X$^gXw#OCJ^`6saSTO#7{5|{QM#gw4QEgkw>6g?H=sQbM5;)hggaWl
zkW4F`JT0cN9$@TuxeC`wXcd{#Rr)?!`YYme7femiJf3xqbh{f1h5DbLtjeGqdl3Dz
z3u3KvQeHLBP-^-Y2K9N}C!;K7YaGV!Nl>NA)`0Q&!S92x9pJ(Q45a#v)VI-6Lp?xM
z&w*Q^Tenye3O2IBzlyr1My_p1Q&Hnc(t{hrt(MQ-?TP7cY^KK#-O3H_?rnH#+lCm=
zx;i`hoZNrKwcm+3Ex!6nNx|J~_Hg#@f{(_vL<S8d;;dCa(2*XuV&&=kf?0CiAiL)-
z(N#mj*(of*tO$Lu{jk}1IT(TsZKA~WC{w<Je1Vk1LJM8G$nbF0A9YS<Kim$!VpREu
zV@B1pW-7cgoc=oT_G*KdgX(?`C20XNo$8@A>k?~0Yr0tYHkja)HyQQfoys$FR;j4w
zaMig<2ZEZfo1EWEd7Jmt#w*8UqpVgd^ZIfcS)aqJQwD;AAmH?~f|l34aDEN`Ue(F=
z9z)fg#>kv)(%rF;W8B`?6~Hj3CjrstsqbH!YY;@uKiILAXpLo3758uZ4`kZVCQo^*
zo}tdrTr3A(KQmKh7{|9BqL-O+HfHKLTa^6;2c^yiGTmMDGHH`$sE3ia0hQL$X&b{)
zXy3aXMgl<mz0SSP1crg#9KN8kW3LMkbNidUGH8P-<#ODP4z|du+fIN#fuoe5l)Fjk
zScEz&=IZaYK&owwxEoD3?_0e2c7tyXeuEo64*G9XMj~LoHPJVoHt?!Oxb4OX_2>Kw
z`93Ofzmb!=DMS%BQ(ETTZYJXyW*?;-t!J(<R&;VHJ;lzVYiB~?V8~aqtW(dnq=%FT
zdgAg5hc=gL(&{79-r^BV$MP>GLJvjBqI$Lucb>`fROt#i;+y=DJ*?4^8i#I{?VE3;
ziG^#J7keb2%UZ47X8d0-?{l&wRD^c=rHNcX`q8_c?Y9yOw|?Su?-^&=Dk<j~VUdZ|
z_8mt$##91>!X65&miAQ9%9Y38d8>k1Ohs7EVnBo9TBV>R85q6ja#TK&F#7oD^QeTl
zr-pS(n@1qGooNf<wr3L_`Ql#}K3&`=s8qoF&|n}1{<r%9dPYR;LvX#Kn_t;@KiIN5
zptrPJ&@2_5u@Fa_P`XLYr_!K4j)B!nsrIu5RFs1M^&@1j`b{1!KQdlx1G(2;hay{P
zXsR`d#3p}Y1v`ocAfD(bjtYqaW9x3qj9loZ2PL-dEcjETL-vpzWpVFU$(0^~mJ52>
zm>bdObZHj#KP_sl_@$X&?^1J#_2Bv9iHhu;C1PFtj|Yp6HmQ$H$V}n^RdFnIlX!$F
z#Rg%55WS=(zO(qy`sm?1xlnYru)xK-z_=(Zy(9Wwxy4(_G`&?0xT7I8#p4-+pL9cN
z56|btHSDK_)+7ayvhg1LHqz#yDs(5Yi8jl}KGm@hk&HbFX~*gcS1BPq63*SHTGLhi
zqFsyEqAhzou9$sBjc9$u-ickSO>;4=-*X#c623+g+7uO>pWH_awv`nmo{aP|uyHT(
zL+5KW&M_-b<#dHPTR6^tmyW1GP+A)0!jbldj)ZshWx^=PQN-89y+!HTO!uISg4DT?
zor5q05|OoB6^Zi*p0)0%JaG*>iw?}#Eco)$8@;tk<xP!i6>@Ar(Gog~Wfr1lp^HIp
zQ<Lm4B3P=_?0Z+!PPp;IZR(1u^!#TB)GVvUr=+>Y&TT3@7}2L#Za(rw+KyjIGh?Dv
zCqnC}jSqkIVa^5A{ZY-Nx_%A1?BTK5w0y{DW7+&;Ey<gg4f*MgMl-he{mzQuU4^NM
zyCXBd*n(3}{<CA5+UJCp@SUwLh4k918iF_mOjPwhklRi@D7Hwe^)jo{4yw4;6r-O%
z<ARthv*GDheJO>F?&c#m0JRzZ1{rw*slyM#TVLNo%gCGzE3T|*rE*kaZK=XcJYwO`
zcny$WB@UZc<RiXt`T|!e*}U-@i1GfM64{9FKMlMpIT*U=0`70~T6td5Rzim~@$Sl|
z)n3kj^~s&HTV1HM!=%~Gq_4|!;?Xy-zJOO<y~<FU>fIDzm=dLu@rvcu8h`nsaTgAL
zc>8QiZh*R3>SCChG}|}UFwcE-yy)!#8#TA4SY!&9xYoa&-`_{3Z-A;^m@VlIJ#Mn(
zp*$QOWUUPKykdAw@1Uw^e591Dt+AR2Gm!P-`gs@p{f4Kq-RYHOYU^IFYApjbQHBZS
zL1Dr^t#HEwiLnDg1KX*tmf(E*)!w>vOYva~C)^%A1<7wnHHRCPB?)8$9dGLkY@zAY
z#%MEJ_=&Tz#S%2RNb7BedvfJ{lHlXb$~RZ+T&OuVNso4)srR|lQ1?B7;jEWBTrz*`
zj?7M*?A^C}D+A_Uco)Z8zPAez`7r|BIRyOc82?QK)5dK7(PYlCMS`#aBbDOQ;-ssi
zH1k(8F`kwk%Y`+y$8fj3(!7EwYIN`!Bs`~=r`|qt*xA6##OsABQB9}~@pC-rTt)~$
z@aagEv?y$;?%|9kz#pW48ZNj`SJsMV5>5baZB-=1@9T-WzEZ(Cb9D7JwX}ulgFhaq
z)V}Xgxgj%nuQa3h;6;^EEzZxXtyF*1aqRSP%7L4JI<Mk24>!A3HCeUS60&Q{>0s!7
z<+Se3%(#R`K|VJq<c;<^h4JKtU#ft;)=bH3Rx!8${NOfZL~^G8bs-)K97A~uXI`1l
zrz2^PEJAd5Jn&Sl#!9DBd0eqS-}J9}bA@=;a=@l=ajt3i+-z#A%n}M&yr=&~2k`&6
z&fP`G^!k!oWnSNwG3H7w{#O0Nn3KFo=4A!L=rVP1Wp71Tu*9RWDp94VP~raf!x&}(
zG_mHm(I(Vl4k>Tq`8B;|^$>5qPmXxAr*0m0Mr}XmTaNAtcvNS)TAj1q>=G<(o6WCK
z17Ads&miA6F{3^WhLA7rbTNo_Kfk~Hn0J#h-(L)8yGG1OybpMXJ==Kt?fhoU-HGXA
zxU)famrqpBi@TYKRa||j`x+?9I}ZB#{21%nxhLDs=c|Epgx5%EJZ&z?p(a1IXTQKj
z^Y)V7@>s}4Snb%h4B~=THbKiKt&!b(Bfry5+;l{*xvlzkkKgkYV>!c^$*Gy>k6L>P
zC$F3t+qZsLgn4mWK(}BtRZ8<iBw2+SGAQ(pFYWbpOq7j-!tsh57epN1W0kPBiV=hl
zRL9FV6sJWRk<%B>(K@+J62J}(y#(~J^I3x(_WA#Te4x<I(D*Ek%ot0Q8}Z)OK%;J8
z?mHrZv`GtaHHdT+iI%2S{}?0K{)oF(ERqlBrV|3gNWI`2WQ9@7C|bhalt>RsBwlWn
zbQma(c-SZa%K5IY)niOW9P_=(c-J(6o@7Hx0@{X+^~66D0=ft<K?Di}-r(?iW~eLz
zw*)I0a>pMW{)<}uM%&UZDQ|0`Ot`}K$Zul~<<BBI)&>mQUUqk7uvOCl-+T*g)8fgK
z7gtnIbJqCCFrlUC^8)sQZ?&Td2P3SapmlE=Ibz>tdMl20EO^Nywl956dwq3BnO;0q
z#eQL2!)R}i%k5r`mKSj9HB7f9vMdM~f90(;En=T%nbB#%+k$RPZ07`^i|2|ef=iOZ
zXtvS*7xZ>M$+C#8mAy+tkp!P2^^a52LmJyd5ggb9)>x8iQ&3~@GkyQIN_2KNdSKKJ
z@hezHch;v-rRti~XK_;zAzYe#IO6yu&HJP+OTR&oDn+OTV&bR(q^Q=XzEr#9J3Y&9
zt`0<b=4`&5H@ZdxDa8F{ir9*K(lfH*rtZ7<qp9Tub^Q*Uj%8Y(47`zjd4C8XHvD6Q
z1>p6?ZzNqo1s2og;S(y}g3|bfd!Nn^A=1UYE5y6fm4ti`ohr$z5^IA5T`Nv!U-;O`
z7T3BzeY8%ovNsF8zbVO>eBCjCu~wDD77SmV&jQ;StPmfG@1W%e%l#(vbmgNauWMas
z>IxaBL-s#HW1NEVD%XQr0yRa%!sJl>>M{IrrAR*d)>lZ7iDN`ga}!Sko=KNh_HAI<
zV30*Sp(1X??p?;f#Mqh4tAD+~P0yxZ+hLKeob1p&Y`o-Jjn%BiL5Me98mkwQNg-x~
z5IeCym>#bHL5$sbAVILT$1o2g80B@D8{f7{V>wL>7c0z`-N=SBNDj)Hu(F=J`-zF$
zDIQse>h&l)G%jWMsGGe!d?hksbH$;?Lneli`<xvQhKEEo&=ENdY79vd+GS#<nvQN{
zTjn^2SbWsJ#6o=7!AG?E_3p9g<4x_e<5$y}#h0Ez=eYKMRLQ+92?H3CaYIADlmZb_
zHy`FCTU3F6K@;O+li(dyHSqDQzI@{z685+a4=cTycs7|5#>L<5#$0VkKH%AUU_JQm
zCC`idTlZ(ODg^p~>z`4w!)u&I^#J$9lb_!9%^~QRse#cuT^>3om?^tfQ?DrB7uVkN
zO#G4wB|B|T#B)Bma3`Hktzr)s@`9fo4!ai>t#?8gH|3GMo?_N_%_h?@NKJv#qIS3j
z+w2pAyrkpjGhd3huUmTEGvHK)3oJ<}BgAr=tm=GRaWpA-LgUR;oPR?HH0yF4IBqm_
zAeZIkFKEc0Z7se4sXk^ni4u{W{-R}5kalRee(Rd`WHs|@wa}TSYqNPzDgYaN#XSK>
zPmJ(zeM>XX0!Bi7)<%ErdR-l#q%{6<qTgDA8e;A)Z8KjQ;3Ra;(!FuTU1|ICvAG4e
z4yqqVFNqDV+;V=5R)b3$k3z`RU63GG#|a+x1BxWD4cka57=Six%o$C<EbO<b&A;Xi
zw;(#&8QTst?qrt%_+~(05*=azw-WG=T?k|t+y|hesgTm@2Hi2%i%2I#Mm*tsSoVJ)
z%>(2y|7S`}TVmWr_2pda*71@+AlWHAbARIDit+2#)i$ffjyjq_tZIzm{lN!@Os;&j
zDs!)C8ZjzH=B+&~*iP;Gl<fnvu>JHeoP8b(v7(VnsV}jpfyOIQPenKV`fy6bYWid6
zRzHfh&U3q!OU&x;C1ck^{Tl(~bG^9E2RVQ#g&eJgTGa>9Y}TYWZfIJ?yX6$nTqmkb
z8`q+}9x#^(8I|Q(+wCnHUE-BDi6^jvIkswdcbkvM>RfuSi$$wcRzxYa9N*$SN1cag
zW`$BuDvXu1@M{<UKd#O@F6p*^`{35xLb>;X)I@Qnh8xX^xX{X-V45pSEyElsxmSt{
z6%}xl14m_9xknm~($q@LvYjKZ=JkB7`~E$@=lO?T`Ujf8d49%mybr@cqty{ji794i
z0BSschKr(9z%z&ue!yc`r)Tj89lqbIQ6#xc*hwckiEOX>KX^d6&VVb<0i@ugRmF;H
z5ntaJyRbi8k489kyL|Zh_T)n!G7nJ@KI{JIQ7X=9PM)R_*$5=f76C!xRd-wvt5py;
zF=(MF4N<sQW+16xR`lrz<jQ)u91~HudQEU$lP4m|y`kNf8!mu+e_Xe5rmb4zH&@5$
zbgLJXvw=Pqo;pSL(Kgg=5cgtJd*dD-`RPcYZIvfiojfT3ik{Mo<B|_Fku3sUN2GHW
zMH|lr=eKnm$)@LtDzSffj~|Sepah=UFdw#0OG9wNq&JAd+xyn~{bOSPfsYv?+gU0p
zywel~Sj+%k;Qqa158OX~T#oilX`{5lHwp3Khes!O<nu>gxLLe#xt*?6NB1liu}dCh
zc!~vP&EP7}Dbrlr9DWXK9v{*^>~=rm<Ethgzj#Gvpmk>aJFm$dxcZ52MtZo7vl_DT
zbkc391c|X0WEsJG<WzB3jE-Q@KiE4IKClc|IpGjDv|GHtS>iD?NRS|g8ro~5Q8?o3
za!dT+MT-NjpUVP2I9j&{>xpJ9PJDXs(%LO?;rS=~IYvXthE{l9XY;am?wX^KspE-!
zHox=KXK!G{7ge1HM;B963$+l@aaLb&x-u-e2QAHn6i`4%#Os+CHwmc`dx6S%g50?J
z7**l#<orN5?S6+L3>m)BX%<vwTkljVP7g%Z<cU9+OMHCQkS^iwLoTvM_fxplPGJaR
z!+D&lx;E4PQZi}YoidJrijSkBI%jHLI+|OReE9LwUC-uQ#$HW?Pyk22IjMA8mb>07
zy=I70W4Ks$7JO2-r+dCxMP4E6=k_I7PEr&qqtOHiAp9B@DluaD1o?2j`HBJb04bEG
zEE>*MiX8AJ2DCQXI|PseG2rx?!g_6(;Wvg?8nB>p+XMsNpEE%J__yf5uz4lz;?1XB
zMuSOkJD{ep|DUSk?<n$5^(j`5$Oc(_vS5N7(y9AlM1KwmE(rj)-V2Z{g~?fcY{JL%
z<OzP(aph~xr^O2)kenM%zCJd`79h9MXeG-T??{so;bC<k>zvwF@2AM&M1J)apLQMl
z_fK+EGC|T{T*AeR?C-pEFVy)&9IOk(c(HWgfHy&0F?C?<kUsqFKB4YfljoIy)I4tc
z#q!>4D`f=C$Ygfi{|aq=KK-<L{WG35e%pzJ`R1tgCnn60&mPuPhZxHps_ThK3rEE~
z)*Q#r8Y-8{@XAH({ISDT#w{ENANKS)B&WSrY)Hq9w9&rLNG=t`NGlyRC(s7hzfq66
z#8En9z861PjNp-rGK7{6b041X${aN~A2kkuATe!Eb*}1JWdw=VaHsDwCY<6K7D}d}
z`(btif9_le0LyjFx9k>&q#Hw(H<}q~;-650YklyfFYQTh0C8D2zdNstqhrWFs|`L-
znMueQQU`bn_d4WFS0x}0G7lg{0>=rgA0z%5j^bpO?U~Y(Un_vsdJ7D`M?_VSr^>xu
z_s~$P|D}ngjC)=7RbFm6H$Uv5^-fM6C@na;@SH~f`BN*%5UM3Za!5a#tCz{n*1{0D
zbDmysQ!N@6z;En+;M&D6$@UuI66X&tRRBiMZmWOJ=v3o2ul1%DXWXpMbP}q!c~_Mn
z+0N7T+N|>*>89|Aq5+@4)K&TzrJ_X{L0adR%vXkA>Vsum7O3pw8G?~|#Unb?k#q=E
zpbwwn|KAQ~zm4JM>kv@o{D-iL)yY@??->G9mw>D2pB`pwk9fbgsbVP<4V8Hl_|4sH
zSWr-U*SFdtdt#*+RXMEM%Hy5p!{kt)`(@qEfiJvP^3ZNmGsO>fY)d{EN>MIVT%V6^
zFOyPjJwO)?8nLjv-RBmkF)){5%~5q2^Wd2u|HDEzHNcu=EM@rlO3O~OQ;`jDsXFk~
zR~a6P&EgDRddo7iI;OtclzNt4^g3L`!mi5ldH}&>e$%VpB9y|;mi66*Eo}-Hhd>Rz
zth{o>&6CoNzFx%?J>60Ca+dnmEM>*ob-wA&z+OMiG-0>MJ>%r5k8|;Y*w-x>53}=!
zGQYNt2iY{`I}9VGgJ=lUSu!<v{Pef9L#;x{bhlHud?O^7AkAJl-+*V!O>hYQp|8na
z?eE!=e&}<vN-y;|>~~8aUE$cR@3Rm`-ht<7F0)d)zmc|)iTqO|0``p>_w3uP**Bol
zS6{CKdik1{ZJ#`_yA;01-BSjv($^o!9Qv1U31E5vOxZkFwBmBUqiUyoV2mqMTY?m*
zi4e?9@yPpPmlb7#JrfG<g-UIspr=}RBJmkaDJ(#Y#{%}qKeUKHW5@zoW+S(~gvL#E
zy7FJ|z~8ZaDW?P#O#Xv=1HLU_wF73K?gSfSi~h!c038ce8^~3&6rZ)40m%S~!Q3~O
znEB^L>>Sx`tsC!#uN`|&y7MmxsbSir);ryWX!+pqL@H`4#mD*}lBZ6bQU1o^5JvSi
zTh|#eKC|L-r1hpT*j0q=5b9z`TgO={D0VQ9LVHv5?iuIsFs^kd+P}G<!<L#pqS)%?
zlNquSKNxFJa|Ktaa>pDS<uMDM_TXCUNOm1razT2;xFTp8{(6wcbtG@;R#w|ojECmc
zPvuWt%p+8eXk;bwMfVW>JypxlC#aR5Mvv78y*pF`xi5C{`C!g0YgYAx7&M(L@jRn|
zKL$7ODiAF2FzS<&$FpJgCS@KOk2?RVP?XI>u5zqGr2Y?<-;8i9@CO6_PBsAY{qnKc
zB;;yJ);5~gB@J?RbdSQ#=9PA`p&9-HDw78CydCNEc(IbbxDa|IAit6Qb&6-*3(unl
z)}_zd=;j+WMTU|W0E~?roj($}i{y_#=RPhC_V)ZvU&P2(PWmdv80j9gqZVx~&3^rU
zv2cz2rZ1}{dM824>`Rl#!n0@5_w8sgcONp-3SPPdXmh%dL&_e&R2c}dpy{hR&8}Yx
zM_VX8F5HAi^hoj5CVTK|8&cf?aevd>bAGJ9J+~#C>ohVDVx)9TXO?6#`1{;+Oy0<C
zALNsh2J9n1O{#TLM@a8vu>3lP&5UQ$p1A&<&(A2@EO_$pv2h?_2b%r@WU^ZOW~ek3
zx!<PzJvUu|x#vT><QgG!pk1l;|Ep|v0o$jjA?w@j%2s#MYSHgSOy72$*LSq?Xr$-T
zLum?jgPuLt3Hgq1hf{Mq>L$LIMfb*iZ6GfkRvK(Fi|^&(a)vgB3uhY`;L6&pl73D0
z{CezAa50u{zjlVM)KO9Bnp^_%NEP#~q|r=`7(G<3;_Rvy5@})Pg1&+e+6!kFMW*L7
z<62;PTvUuUAVGlU#xqg_ZvGseXCQjTuETXz$#&0Tv#SX?w?x|}mIvw}{SH75k+!22
zhyoyj-3nuN#3AMS^)~K7b)pwiBP%Ct@G7^~Uai(XF&#cxC3wB#+HC8I@&bILo(K91
zV^z&9PA^;g;zCWSXzj&RavD*<1(6u#NY(Kw4->@CROF+BF3_=lF6=2HPTc1wLfKC_
zpWX3jQRm)jjsF-XlR5n}cc)g%A6gbTy=tCoKXu2b2WcB+nx-7@=Qi^xc@rmk^caGs
zVpNoCjuN=`NvPh7155_B96EA@Rr)U|GEtI4C+0`p4&%C<wx^8!{7pscBf4^FoIf_0
zy25=nQz(JwXj~~#U%=r8N=OHLW+z>{PmkK5$bDwU{Qf%!-U9o|AV{gij+;K8E`Tff
zo9d`?^^-@D%>VQwe->3hDH4*$)RsJ1cnPGQ5P60LUfs8C0r7Ca1WrB12}|L{22=vV
z1{-zIAdnta2O%QtNFWQiDi;w1$3pt^u!ppHUA~M^b>>UaMA&aJ(&KHYCqF@U<4b~I
zvXa$OrK3?VJy;>XCZ5>x&i(0?LP?dtLPvTqr`~rAo0#K^8IgPj?t@u?U>_k|i(g$i
z5X$@{b~th9Rz(yh#U)hfm^4;vwv?q1Li3s}8QLg2^1Z`c)!%odZl!s#VY3<{ocWk{
zo5%Evlh;4vlED&-4GlNhA+|bltlrlegA17|xwxJt@!8_#Ox_0wJrDP(>A8W);su^f
zM}F*lYOZ&DW4hZ1G#ApZ%BTUtR+?VH=Ku&MAp=M3ji<pWn1LQ6B?=|rb*M7ya6uFa
zWmL|DSLvd}hV-+S-=dxc5<vBcfd^GB+aZOL$d*%)d(cKHPm}yRcJmExF5|KSXdwUz
zf~zE3LiB{-R`nIDsi_OsXoSR3P@d^wL@?fqsCxq8VWYWBp7-GFk4L$F!d(dspCPP~
zuD7P(pr_}!4D~8C-?vMvtF|KAG;_PtKgMKJ+22{uw3G>3Z#yJ?*<s*r^OA2lwxLsV
z_nvx`LML5UY#u)i(;_rCrMP%wVn>=cJ+Hu5M+>xX6pr->8cSMndPUuDrS%DV_Baul
zo(oeqe`5zkM6b$LNXF8O_aHniBgl2XMzv~!pXvJYv_GEsa^FM!D1;-*A;QjoKS%bj
z2IkLsVMLe(W}`^OH*gjf@>c)>_?tnagAR3}Q}tG5Z(kLCQq#H9(t8tT$W{?+<5gx{
z+?=t{KMr_<%cJDoNVD&U#ur>)StkLhrp`P}yr56i2M^`zYN4Cnrru#zsO^i-025N`
zMvJ19xT9Ufy*ohC!A(n*qpwtNJjHKV$P7bu;;gMtl*{6VYOwB2p>3nh-mLtfJZ?Rk
zx4)$#qx-APEj)RV#S58<bi1gn26sXp^iE(<w7u`6RT6vvlGcV2P$v6Ep@h&2xA>}j
z&@T+D{G4>ySjex@u9!p-_Jc|8gRqz!!}>sm_`TDDUYh8xk4DPmD2}fU(e_SG?-7oJ
z99icYDBp$>Fn&I<Ppff^ciY&52I5q9)w}~g1FF?r(E(s*2@3lnRYHx8sJFALxO^Yl
z1bQL4QMJ(~Cj@^`DfInLf=BQbIWleea_cc%)RMEm(HJRTFywL|X*&N!aK?_F_cL81
z3Y%E(+qiGoX-;5l{CYks7)`M{JI2@0-zk4Nt}|IwH~5UM;I8;}ZmupHi;ka6krKMR
z#0K9fol2s~SmgDUV&eJoPw3#i+J5cPNj>Ubqrb@YN!9Nf>?{XtlUknu7z)XL0HlBd
zjJEyGQ}p3pll`BOXuiV!^}q(qCh$ns4(OqZP0*mWcyPlvK>hYot<NEs45}2w&k}*j
zctbvlx4lRcqy>ut;H!fmae%ejLmiaocLk0Sv<FdG#e~mKQxyjKakj+5lyJ9Bsf@@U
zMLM`~ea{ZL`FwJrFe|nt-~z2`BkyTkwTZJvib>7=7Ih|%#Mi^IpWMb4d`Azlv+2NM
zyh}KwvHa_iq2Ti5I%I!X?MP2l{&C@Kt~o`zDR0n=T*CPX3ds*at_h(BQ!OT4tMCFO
zPTR;R8IE=m|Ktwzz;h}56~^nR4YH@=S1rr)awX47+2R>b67tkB?1k?f<)hn;HZ-<G
zhn1@cwsP{{=Sta6`*SrRRCP(T%;HhgO={N9hk8Nj+^+U@g$`V8LkN5EHEuG8>e3EG
z{E^NvxSy>xf)kh%b9|+P6-^SzmXg2H;E}zbP<))nMWosRJuAqmaeazTYn2OieH+Tw
z`C0>Ee`9N!XS)4jGH$wJUKCe89Kez+^$1BBG=-xs0qJ-*&Rr;3YxQB>{J6`j5<xA*
zJupQV!Up*Z#~<0XJUe>fLy;~K*XmqDKA70?*vYKeH@+jXcM#dEGkKT1r7tdq3u}DH
zb>PF$QMaYz(Jf!=q@zm1GX$3V4j*%3hL!!CAAMbrICJuCkE!>}P%;FJ0sDq8mn%p|
zxvRdjfI&~von?KL`miT!3lQhfH!Hp}J_}42n_@qQF{Gs1moB~qS-ypL+*rTBP|mpY
z{x?@Wzu?j@0k*?Ss-p<fHQAzm_cL+?%q4a3n*?^pSB@Z#^ANxQ`#)v?Stdc!L4YUG
zz4Py)1<==0DS^`gC8rk=-hIxLhYz2=>@5SyyotFFbq;)Vf4GYI0}@fsLp|U8vml!A
za;f6M4O1ugX<xTBAY-c`QgPE?`U2yIJcyi8Tds8Nwqz8LaO6f?00Vn)r120t@}W}l
z2@Q0c4G3Cn+mgMhsilGMYfo_7xRXAwbK<H@lj)?C-HB3zgwE|Y5z%W{-jD!0<?rM-
zS!Z&*?M9c05T;ccXBx;L4i6(Tc$-|Qq2RpVkm~#v@xigJ1-@<mpVyKBQupy~X_o!a
z)2!o+B22p#G3t2MoOJDKk&6NkgryFqM!CkCU?dVwRjnBOe0#{5+#Z~lKVK=yFNO}Z
z*g(Nfgo4L<8%0PULRJ-FuK8qO#4uZ7f`DY#y(q>{?qpw<o>kj<HgYOWv8KrQ*d0Q}
zQ{C?Q3Ij-b%>)THT~XkSv1{Q~Ud|%dODt^xu#IQ*;TE&FZy7*&Yc9kyyix^FdlgNB
z5+rM1pPhV)nDj;GmtCORUm7A*tvV$>v2rNgy|lnZQBIk}fPtxGh5_cIStsZ$G6D%I
z&uvi+1?}g)RY;RS1tgo7*7Bky$p9pEf1&d0Ei0v-KizAeYlo7cPJ$E&VWuD(7S>Tk
zTL}+WEvZb^?nRbQg5aZ*-kIB|LxDzgf(@X;_~RA&tC{`Z9#=q92X&H`3|}%Mq@S#9
z4Dmf@fuGi~>rFe=dwsn7yiA%y<vLzQF<^0QWZ<LHdE7#$`uJYD<C%ciJeZe@EE=j3
z)%R%d=KI-0yjve3#QJ+&70QMTb~Tq-WjlR#QWNJ+ll*vOSS`XsyNGG<L2X+};I>}a
z#A<uUV^2$O#z8`QsEre_uPia9^#16flwh+0C(b_TXjL+c3^1Ot(#hmlt|%ns0ps>x
zVU~&=(ckFaNv@W{<2hm^8jlepqm2DUw63EXAYK3zcZE5jV4g_>u2gxBoShHDv<6V*
z@1ar~fF%Toyq(^__{~QfyM6gUph<Ma&$>R4(`0AhXU27Il}l*|;le^E%PJV6cDHO1
z)pEgyG?Z9w8L$-YcGx3a()R)(Uy7VxkW0wHJVU#s*ny{3n#;czzH>D{vn`gKy!6E4
z>EuVeRc1t~>r{;Eo%gRn(~4*X9g*sJW{Z4wfDy*W#APU3Mb^r{RBIlZ$$0A@c!2^E
z%$jq})O<R?oE#tuqo?fOzE=>j;?f3QQ4K@I&ybhC4G&L2olbU6KX!l1CDmcU74$ZL
zgKHOBO#o||cAUU|HDW|)+$aTqY$5Tba9!eQ;U9}`T`>Oq{83(2=fLsEsvtXgiIMx?
z=d`jvp!hs{DEwbgeA&50J4l0jFHr(qz~%9up(sEIFoiFf61*mP2rO(?(Vs$A1)f!v
zfQcAk6&Y>R+DCHFH!Z<_EHgxBzkb<XuceT_EV2EttBrS8GMO`d<SShSvk!5|35p&n
zVA>BC!(#NGy3?@gOL#Pz6n{V;E_&Y24J`rBQin+Tv9~xIJ*ee#5IA9C&0$_i)I*Xh
zZ&gH0vD)xmu`u=8mPa?^J`e*}I<A?t+znBChqF0|sQ8vKaqV{KJ0Mjhi0cwP6ij6W
z&J)>D9NP}u$xeF|e^+!*zL$%lf0hfNO#ouqQ&0jCV48ru=3>{6IOmV#{Ohlp>w9aT
zGR7y-O&Oc~XAYDJ7`yjV*@nsnO`iZeYtciIwVke3__XgBzib|{GZ)N?LzceyzMm5!
z{{Hg2vpLBPGbSVcmXLMzm%`@5sN??m+T#ZG8k*Z{zsS?cK^M6gk;f)sdz1qnjVX@B
zTOZKi3#q@X(ZIqBcHwf%=_R415mQHB-u&td&u`fdAD*XAu#HG0tqC5kRn^EAT;RGC
zns;#^L9$KcF)*a=2hQpJ&?O*kn=Qc}fR_VH^FN0GFyKs>FOmV+My3)Jb2~D~;wf#>
z9+X|bM_a^caA+Rsvtfl)rXt&rW%~hToL=dGj!taQ8S|f1oX|fq+W(n_{#nP9Kya^T
zCfN2DyJ`R@yg<D~-|}(#VWpVrvXd(wncMl(-{;c{a!n|OTP+bQ*R@IW!&t3Ub(0~G
z8&`LT6uvBGw>83JfprZZ4@rUdlg@OcOJO$}9;#2&JK?V#5!XZ8L^MXQ^mo?JGa`NL
zeZ-I9-eeVLsmM6KQ=f3nnwy+Ut!i|D=CO7r7pMqSBN$aGdE*#T(fbtj!mgeGD~-97
zKn_h0)a4wP?c)^fdIkJCRs#2%4A*35OY=|u;G?*s?*=RN;bk%t^9+pk^E`84z9Gq=
zO_GBPTk4c6dW=(BDtXrRJ^=8tR#gZ0V}PU^IDd?70*LoL)8v$v?mY3JTQ`&=HITbt
zX-0(8$VmX?OWc!bI=S@99GaEA))uaVaEl>L+!T7}WEUBU_d2g>^I~SfGR8%xV?iUk
zWYhk$MpR`TD6VNkbts5hU@dH3Sh&Kp>ageO3TQrlS)iS<eh^UjEohdVUu+uTD_Ly3
zX)<rp9bnll?NKfga~j>n*`5B)_T35optx2->-$B%kq$ZJ&xCo^RLqW^y;DOf(xafx
zwb0?r)a>;=C&XgA=Ap{xt|yRPMuZ4`cQdQhnB!s*Daw%#ccJvMldp6SUV`hr(w#58
zraz{fwzgy9&MU?rh2pu>dX)9>cRNw~alZVHo@@dT#|OMg#79Q{Mh%Pi0Hx&)G(rT^
ze^0>2KEHspUq}egOh>sXJs$U;^!Sxpf}&y%6>dli!fbovi=FUWWxb2%CDY?CK1`nW
z$hQ?@Mm4-#%D~K3tmmt4(1RoEphu!h1}(RxXVy;PYV8EH+jCZvGlH#O8#OQ&0rxnE
zQAFP71-j+X!*mGfGril0(S69|gVSt(bMvb$9mg;nYD-?{V%_X)w;xi|FMAYg(SP}%
zA>;hbOI6Xt6gyMfdeRPykgA5ux0Rjkk3-cNQ8)rg_}387cN|F9?l$n!W=(UPou0L=
zmse|#Y!|+q;r_&-CD?RQsm=YmfD`*+QhutqsHF0@4wUFKtyiy^Vye2RVIS4O<)4lo
ztBzZWU+IYgl7_>tFt{v=*<1?J;$hz8E?qk5lUgr5P}w{$Oe<t?pmEwQeKnt7pi3*E
zt<iF#BEf^2+A=DKqw~J}H7T!o|G{j~h6N0)Tj4{#Xd8Pu4Wtt_6lT1BX;;RRtCeG<
zQC`%?{wSRx>Q6aESW$IIOY+ZMm4VHOY-pRBx9}>x`(vN~GpzjcKqmb;6Oz&9fEWt~
z2K1LU?ETb{pl{SkHa2{H8%lUo72Yv9g$v~&pGwFYCC3RbFHj`?wD5SK3*mI$A6zH!
z-Gtj{?$cj^bLKA;O;i5A)K37B1spz<7aN6=8g)T#vsv?M782*eWgc97w=I~?z>R1s
z4;y~+s49UndkH>GK3QD+Nu5m%i%z`<QZTsScc$9b|AI87&NhTO85VOx1+vZnuk~aZ
ze>hoK+d@2HHW?cWa_33<ZmD{d1z$5+s=(4wzC~_n@{&VM6BlZD^F%86S{TPZKFCQS
zg+<E_xwXD>{Y1{9FO#MR9DDPQ^m(aUUEQMgH`0WP9o;aV`HP)71ARlsv!5lu>}@!`
z(wskLjRr1NkZXHh@>3Eh!W5tC1~#M)se=I^%p}M-{S7BbOB+U=237R1HONF?y3oMp
z+ha%*e~sh`hJZtsdz2PC!yC4K%p%WnHJ^~Y0v?+_XQoX%CYgiUP&9Zntel0-U*<B6
zuNi<`o8L}Z>8uHV&{oh^s|gPZDbr9WA{Dt7MunKG)dwz}8i&*eIB{D5rJtTod7Ip`
zkEIts+Bg%g*!_-P(75!@syS$ZH@}kcD*MKwDsPs{a=xplantd0SF~yHpT}9f${!~S
zv=seAL;7jzHx7r;N8%||)X&*aSPJ~@fgjhud#F%`beEo3%j6Jl7a#*i{f^FsIfVl=
zucsr@TwM`Hj6AtJS6dKbi)N#U89nGkV$j7nDrXe2KPSCNpnaONmN(lgl29&G`hA*+
zg29L`f&w6*JOI(5c%8h@MR!j)j@^7wY;v6B3BA_lPNi7Y{z0x;%Xh(k5Z@AHU-y<+
z%r0F)&s|SS@Gzspd_ja4&ACUQW`yh_tLE{}$2jHarO5hsD6yKSWsHx5*~$n|e@#7B
zc4q*!ts83ym9+yH!oVB5NYog>Tj%%p2a>8-lOLTDj>K5Tv4=dm?~tsyikIpgJZnoG
z=1&UVNyhTOmkHywe|a~fKEB1OO)63RAab)O$I<`jA?3Ew76mDV!@H9Bo=LA^`Kf7_
z$wDTc2!02M04#Lvep(P!2G#6Aum++SvX=@#DaBAcEzJ8ndN+8V3%<>~E_yqwcMyJK
z%rV3jV!ILOG;+UzqszfS+Y}z3c3eXCy`TFple2LV4@qTb^m?QIIr5&VNgueJ7BDGn
z5xr4g@oV-c9#QacB4}Rk{y+ygatQ-(QyM!p)YuWuN;+4=9wOiN6?#%oDFDykcEBG`
z-`fP6RKGnY87mj*!ZxPEv^Yz4_}+r$R5I4xMpv)MwP4^LI(-j49UQ#9A|J{!Rqjtr
zB)%rMqv<hKoxo6bf2UovZ>IwM!eXwN2uKmn--Ck6(RS(W^i`l?a3ovcc`OJ95$bHz
zDN+MVl#o}t#F_zz?gt6tL7kldX$ZsiQ_l~Z1qkp%{{z1HGt>Y3RQunt2ZUQx#Mn#Y
z3b&G2w{gD7I<IzC-a}I6dtAs^PUkhHaDDW*!O#T@chsOjwvrtr8AG~4Bwe{sJo|l`
zD~L30H?lV0m_0X)tC3gYCfhi$BqVEuJo2IdY7fkgs-th>j`vYxOFb^^7lYLrF8ya0
zicWV?qU}(UNgWCb4imbIjd>SSf6Vp`1+R1<4!_ZLpT5{=Iv6gKwIDdbdcR6DXVMh1
zK~gxPBvEFY&sP+XEbyOZN(msi9+FR7g@D45B9rXmj@(kF+bF@SS0*=|-EKMi1K{y!
zYpbIh`xL?0@U~?7z%Wx46PVH=+fby@@{}c^WAo2z<f{{neG0V$x}D*VqI~8Lf%`Yw
z$Zx~A&J392c<wi6yJEgc)cVV<DE4k5hd~8kGiIksP_W^05w$ZZJ#}XE$AVEyeAc()
zT?vr_>CTUs0c~F106tD84CXM;H$=s=n&k%r#;smdB)TaN`_5)HDKMr`BfzDG?7#9e
zY4dJzO88IESG~u!fmN;Rln}*BZ|gfmS@Ofu^FzuvGF}!c<*jy%y|2hMCk4p<M9U;?
z-lvEB3-YxMMDu0KK7r$VM%jp31Gj4qFU2xCTBR$_QMlm0u`Fs_1DkN@Vkj8MwYX2=
zhR-sX8dNmrQ7Gp%QUL;`v*eD46*vHJKF0{iG}rMH$n$Dd#r|hGdR*tP$~J}^5`WVu
zx=xt5<Dp+G?wk%;jT&8Q=7K$F&d0yeiKlSpcl0*|Lh_zIQ4@!X<j$9-K84*s#41f0
zlc8VCM~y5aSapk+U|>ZnuWc4{C(|!IvfIJx#HMNL<db=pdfeu8s#BFewD=s&>AWu5
zD}S_E;}OXY_d(+Ex7_RgX{NBDZYylcwcBsiP8MZFZRIqSl_D1PN!;xFF$bl5FmeAh
zkOK?>&>oSX(w&AqDrgz52=MOaiz|A%<Y%1&>UyOM!O1I)Aq!#ExN8To89%92I~#`5
zE{Yep4(S~=b2BT}pjzql%kH){(1Z&=xoM_flL;)&hU-857sNMmleuAE;IVu-G0HBh
zy;?&KcfV|}vu=)-{!)<_y*=&ro_E`<tc6GN^$}IL_VZhQF#ZCiEpoJ6DTZy}>R#YG
zD>aY{NA_Nr*^8HUJ9;il7~P<wTF~2Ku1>doMW&xv=mdjw)xSShSXPpA!wKt?{3QPF
zC5eC~l24tD0C#*)@E`TCioD9T6wg%?J_(I-WGYKJf?y;B0Ry=*P42Z`wI#`Ef1*Q)
zuE$Xa+$;eN{*Fa6U;Z`VivKek_}_xe{RUxQ3&TdTWUHBNiq$JO&ANXtZap`~di<K~
z^(XqJwi3fFp*~@QUT2-hYKG~Zu8p!CVep>=DHzooAy+@ijn$M;#u0>)=t~Q>EULlH
zxctk;9)cb=-?QLGQQzhcR3<Dc5ufG$K+Z-S*3~~N++4vUo*U!3_%`b_Sm`M5>%|)d
zBOV58Dznm9#$i<iUko{Nl_*2T6Hr#4+(*aV6~(DPnNBvYO1jcuN8YlqBc<;>0ivkv
zZiw}@bK=V+J0~Uf>DgV0YGba8vv~1M(!!aLfvrolEqn%t*=&PIP7#l@+G4KSV)9kM
z<DgKxD)i^8Ys9<g9jf-vJA7IkAtvnZ>xQdz{)}C!Dp4&PAsd<zIXqvQ7WB>B^lb;-
z<35G!F^coVO|RvQq-DBkdN)$|+3+;~EMsJt2p13uFf6`meq1G4DVk*MMOWmxzl4JP
zGTt<K+14{<?|6bO?zDTjs*7Y<PFU%(Nbx9*${b+;yktj!u0YDybhTY(M5`!XSaV&h
z@BhT2QngR|G<{1P`efggg|MofU(G6f8edR%e$w1ZRI>SlxfS-6^t7VXO5id3*M5cp
zmlyv9NxrFTn*#io65Rpt;+xOnfi$Y1`8NaPy)-}LgHR{;iS<1-0DgXy;=1f4;MF66
z5P1?I^b*bGker8R@;X;561BiU0#riS;BDZn2;`8R71tet`C446mZ6(XS@-ips%)8u
zz*e%eUj$S!vWkuI?wm`OP$Q*5kau>Jhp?M+k*k9_dRd)a^1Jk1)P22xp&38^WjwHW
zO!F|jrWZpldS^QNY#LDDjZrXjqM9bsd>{umyFriv@-z|@UVp&B^1*6R1s^CYi|mrp
z#}47cUa_lRB<8Q=UCU`#4m*f8w}WWvZto@jm=CWkB}6pO9?zCcUGUz@&N5%k=ySH!
zt?h+8F_%ci_sl4AY}mXFvwSFVTf@;b!G)b}tu4!`bm_A?o6~GMQJ57+;TQx8QE;On
z?Gy==)BI5NspD~&NR7_;2aky0F*sidMp2EAQ!Dxbkk%`AOSQOseb7=Jvi7Crwx3dm
zi!r%OKrH#1ZQ#-@=goV#iMG#oXEmL&yMf3hbnwDcWwocChEBfK7axo0Nbvr^<Zrrp
z%dH$^HRoqVO7GFw{hj+13TwW>zEoE_%4^ssp<sE1$<;VZO$3mn@kqxvNiTA#Okyuk
zKw{VfQh2#Z`D<%x(8R#YQB!(lm<8K^5-$F;Yv!260>10NaY?|1R$|7CQ?e@<wQDG$
zqFHlN-H`r91Pw65*Oma;kelcL(glE&NM$EZdZppDwQS%~7$CC|xb6Q-Z~b%s|M>zq
zboOIUUBJ%LWQU&OU3E(<IT_=5SK_}QFdfr21C5nGo_0(fxqC?Mw|AlAR{M2h6(65Q
zBc4NF=+~WYnUtaE8<_;1z#;PUVFa>vh3A&Q<TuvF-NEQ*(G$jDz86_qbMN6VI&wH4
zP-=hiEALMCfu%t7WGKHyp&i*$m}&Kigy$`$mK5t8QtNYN56V+K7?|w)YxyPlQd@y=
z5;M|MF=D_&%P!Ih%<)?<$qYSJkL^}CXgl7&T35JXNWVGa5RbfK+=)7~QDp1Dx%Fo2
z4G4G`WE0ZL)gm4muN#Me;<4GX5E7Sr1&4=OdPjM|B}QJxw%r`DfwS>j03=T<Qpl;H
z^y=s~ThHc5gX5tx;o^in3SX=X*vVFX;iLe_spF8}(RpS?yo;klwo;S6X6;FXSUuB;
zyACJPUt)^K6QtADKOtKR)20aB{^BALGS)_+nK`fB8b<FzjhbGN))Fh#BL_|VrqhS)
z#NPp7-a1y?khCG2g!2vEY`i;9f9_Vn@>;=4({m%G8U3bVxJhIdUDNhpgRZAhuNc!}
zh$zeloR;wC`U=sBe6OdAvH<xLa1?k(58nG;SM`g)?D-eu7*j>bWpyDAez{79R9k*`
zw;d_A`-soTG%T>*o5aPGirn5tvn_o}E;OM}y>!TqIxJ)lY7_rvU$r!wpOq!)5L9c-
z?=G)EDjfStVITWS-Rh$g?|L%Xj4fr_vy0NDU=}VMBqp51VFJK%;GKK2K2a82Hd$J}
zGI#ho2W;*V-9*(-zEznS%i4V#rbAaaK=O6_lia~u2TUaRqLWbvNI;I_0XHDB9@pMp
z@YaPU)Z3a7s{N}PFK1@V3OTzHBCt}e*ZGXEK*-W#XJIfu-#%SK`x4_Mt*67{RJ*UY
zz`b{zK;aEK)Dk@VoxlCM#tGY5%dM;`W-;yfkA7QReQ#^PlQzYr8SRk=q=L9-;;ysv
zjg$ElIsj;^;MA7ohCFI_;Q*L(ZfFYI6&NTLD=HYuCT726ZI^w3cai#?m1%&ES)|rH
z+IgjRwwKCv#b2}klY=SY095cB)HcQ^m>60s=jN(Vi?2A~SJ%rJY>)ER89o~BSliGk
z!@E1FP`Gke-fVZO<6?ipOh)O0%a{Mbe!X~5jYTU^I|GDPajZeWCa2dMho0;F!t-~)
zMj#{3;(Jt`C!fhv6<hzj<@7aBwLUOB;97CIVm3zG4|`ssWi@DGt>0{CY?fI2C;WZC
zHQ4VDz<N-w`+r6VKoMV?ENVO7YLx&Wi7e(M2o_>3df(mHnjcUY76Bwz2MZQPytW#}
z2YcH~<zQ6Il|F%L98iaw|H-Y}pBDU$YyqD6zj_DY-@h5g*cYro>nm5#V>d?>+crEp
z%lUbohB}x##+3JfS=d&*vA$ng8jE~Xuk4`JFOAXLz<WiY(Ldi6rgB5+ryG?oBU57B
z&-fcXw+iLxe}RqBvO(By1flu9V6NbA9zv!8Jn$+Mr(4OICXrB}()M|{y7If=l@dEw
zu1#_9&&`jT`Dyf(JME>jAIz}<N-<(jC6^B?Ro-q3uNRp}yDG84oa-Kug4=#h-gHw&
z+6(%S5ryO;?Pq8YTHZ6K11PRVeK*jv<LyL&&NCGDaH&f)&@(ia4>Z<yW@Jo=tz-My
zd<#!t0^h~3-dPupj_yZvHy@$us-gk=(ag^mXdfOe$g2KmAgYg*D=XIRpP>cP@hN(B
z(lZw9m;RaBv@1Rs`jB4hqpRwbdkQy6{*G^N9!Oa%ur`7&{k+y04a~Y8zIyhH5V9j}
zfZWN9bT~cVW!{R9h#mHSF`E%jlTywLzQ7POa!ZFiR#x@5+emThH}y^+71@2V%LeDU
zN!nW~;9=|`L|M*b5O6u8OW|^TU?hb79qDb|=Zd-Wn(_nevvX)q0|~ihRVG65uF)-&
zDW^x#$8uncc-AiY_vfZy^JJ9CHW_6dzU3P;Q}0!9Oe+BJdBx<|t~q75sWBC<yM$2r
z;5Mpb`iU-V@kvEzb|IiMb-^I5F?|;>@X-g=;9u;cclh(W=z7J6C6vF<ScGJE2B_O4
z%pLtv4QQza)5D<whnV}GLAF3-xbigH8q>%V0%|1Tk^N^iE-_wJmOpx!btJ}IC(hUh
zT`*b%|MCv^#{CK09K!bA<92t>3y7cYJKwoS74v)CYQPjzbjB?Kaso$EPx$@D?_B-X
z=(k+e4zpX@uJ(t6u(vC0giPi`o%OEW1B@Qd2Xvd_X<DfxJ)5kW;_l-+O(yx{&H$VM
z^49Mta`sedA?hm3l&4V=h>~R&&FKV{(x}obHeS!3`q*C&suLVSuf=h3ilp4uQxTP&
zgIAYX%WHf~yU7VV1_NT$7qs3b=gifMo$U?W96YU9>(z6{F9FLFuRMRI?Lb7}rss!$
zT+u<2Sv@Yy9Nz6i$5t3Q12>g)$W{Emi~3V{30ItiURsD4R~HnUeP>C8SUP60*q(o$
zzj$YCl_5p4wyWFLw3B{+#+@s$;(Nd?#h?+@6Ir%vt{$}i|I@CiX?GLwLHsG>&8<O3
zLV6LpgQ}_~<OsZ|rrs?#w$9q7IR)*;eSiZ{07y_dXvN0$svKiXcos$iUIha%T>A^h
zzu2VCKl4EVs`KY8Q33qke_8}CujDOle1w9xOS(jJRo3DWOI07X>>`d!@FVTxzvtY?
zF+pt$4$mL#COmO6;+*}hr(smqb{2#Jf!vwmv_Q)?*IO(J5noJ5X}WH*`>UP<vvbO-
zW8p_9;u@WCRZ>yn%WAmBngg!go}~P|W}TyA-tD*19PLV+j}BG6p7qGfE}mf6e;m5B
zN{J19ba8OXG}}^cDX+RdG(<7ac)pl2_Q`!c($16-ggn8`RRQOd0h9xp|M^e+dH2m4
z2zBY^boLKaZ<xq3y`l%^YWVl)E&m+7%?An--4-Rx`hDZP8UV{^lh<Mw6od=iSvZ!6
zOjVh%u;vhSJbe=$iO&L1OJh0-tITk8*l7<V8M;ob`?+K1;+0vYU4lkura8;Q&Y$N@
z5A$jjgBGnc+{E!SFKfL?d}7VJM4Z>vC;pt7Yx6EK_0fZ}B61s%q7!A|8%BrSPkj)P
z#m;9I@W$DmgvZ|I?o++;Dz#L0SN<r&Yq~-HL8+4Z%51BvI%M?%l}n59%^`vOC@MYM
zxizf4V9yCxEkXzo2{<*>gaQ?K?znKT;Uu7+j;+18`JJ!TQ>bdE|C&mL5X$<b>#Zg2
zPiUUDYeZA)6T_7UUl>8a9&X*T4b{9O8_CxxNs-HmiF5h$lt9|weBbMTGk$?!VXi-A
zVfuQq#3YuPa<P2M-0=Wi^ZhEa|D8AgafB+FO9g5HC_;e^BOe$3DTUMhBi<1RXO8Mw
zEtNEmC|8-uElQJdJf|P`dH4{JS1Zgs&#M?JepRZ@RHz=uOba+Mw!oEKCYj2@JwKb{
zGPdl-VmKc@_DSw@^Xb0aKv-(T`U=%mA!H>KaCu8h`(aEM>})hU1Z{y>ViuJh#5Z?X
zX*~@4wJ=(GKqF|*{k3XYzII~{u=`<cF9Klm-a!P&uznI~fBAtM@q??WypsRm;mgB4
zvOA@ZU3NDrgzO744{1*Mqm)PA!@wb8qh?p?A8GEL2iTX_BtwVx^x~#>%^KQDUJYE_
zipT?_hl7%K`D+v;h5)+RBg6-Ei2Kq?P^<ZBlhEB_bpA}_2U3+%zu9d6HEE`M;U~Z8
z<U;nXnYv?rIFxrD%XU8R;60){?}hvf2Czloa7-(ey-J=aPDu6{7|k0W@z{m(w&ede
z1eNcvxPuJ`cf6AFXH4$`c(l1;O(IN1Zy2&D)da|Z^z$f+a!229NPPehGCP6PWOurI
z6q*-{9k<KMlZ;G2nd0i*P=bGVjGaJ{@IMx+|LRudfu93Btx~ujVfE*O+R*uOS%oD+
z=u4Lbb{E36&}%tYs@GE%^UZJaxsn_>mEFWiD9<=3M?o6Hs>`*f??C_hT&#D5lxDwa
zlU1|jdCd(etMPnt&ACJk5|v%rnPs`lNR9Y7QqkO;QS!n2gqKDdmyz6s1X6)lf43kB
z?f<;uOa4Fe^N8Lulh8bYf#%JuM%r!zG91e2;Q!Vr&+?Df2B1AG(ea*!_c;dA;7@J}
zwL^AFG7M`Yn)-$M1_hvv*&mbqb(>-$b+KCY-<*tRP$6iY=H=_8rpL~GIN{V-C|ueK
zK;p`aE1z$Ai-<t&G$=gox}bK2(^99nNS4TA^T5o1`zU}VYko<WiJH`gJx1|S=_5^#
z9e5UbHqwz!c_scSr0FYz#yz9W!CL8UnUVcUC%Ir*hRu_FFXT+4e7;Ut`AL=)i$0(S
zIxplf%P8@5@uS(+uQS^Z<)bp)ZHypK(tKzTo_bTVU>qMK{N*FSofQyzej#vpY~8Ka
z!{ga)Lm+kez29D=HW#H<cTmUUfh1_~-q8kkQ{y?M!V8C5=UNvp^Nt?NDETr{l&Hy5
zHyZMBz&#IB+`SAvh70@2lI<77XWS4TFZ|%C_$^$BDKB5ybzJK8O3R{wN}?o}WgG(W
z5vFiP6XYEi?0#<3BfL>m^ElfyvNK&C`M*hBt|&-^*X>lO@hT5KQ^G)}EJ%%MF^9Gc
zZ2*%aB`BPOT8kiG%qn21VF7hR-G+%rYgv8YeZescJ_()o=MFw@VStF?O<HX_S2ei%
z*cB3gpe}Y5<F9*04Terh@duK$4p;otpYe4F@;mXYx}Jc)6?W>BZNz)U++zqrYU_Jh
z7IyB%`;py_)8UEA2aog$ubDB&SC+oD#}h8d5CUb+B%VW5{Db_|q?nyJOVSB*8<L8A
z3c}hOQg}V6?DIq!67QfkL{b;U1(0b&h@txl+O9Vz<;AV7)?bhuLX&T?c1^HF_IFg|
zf6NjqorHg|%FIy3O33geAB;05%A-a>X%z4;FeI}}v=TA82Vd;bC6Kn#RdYzznQpb)
z7>SaHJi1FW#qnIU{NPg+zhLuAlu3}P&ae!jVyR8$enWD>jipB0-)%`udGEsR`GO<6
zMc`1mB9C~Zo^wA56d2U?-i7gZgV>%E9hWZX-8rul9CGZo>#upm+INFzGDQEsX1|E3
zdjCH@27osSIoBk@QYn~L7ybf34TVQ~8IOvlGrp2fG?em*QUT>L+YB85^WOi0292q2
zBc@qwRt3#U)5zm@iL;>3F~`yW>jU%WfdBijuR~CQfQkcnfz`KNsIN>}loX(y`{Art
zc1tyfaI-xGkd%>gFL34Y1k-}b*;{yd+tUWORHGks1l<8ztAkPRaO!3HX6hDszOQm`
z5Mpm-X$LDE%`wi!b@rB4^tx6@9Zh%cG<vp|%wCn??w8~+*)!0e2EHY^a9G?R`gjm}
z!xc$Xe`P0gsmWZ;`yh(=z>|Y^tPC;=BM2M17eDfLT7JGF!;tXbT7ddQ%X$<_Bs#OG
zl__I<jg2kg<Kg0IE}Wp)Df`zIt&d7A9^Z!R-5q&Ill_O**7-vczx2rd;V{#fmIq8h
z_YaKy{YME=sjREPG)!mn>z5TW+kt*j!?lw=<L(AO`uRE)dPWD)by^IdCs#0jb-NJ9
z7pWb}rWTOHUGJ!+oMq66B%*yGa7P2;L^@HoK|F6ZwjVb3%uCK5wAg+w-~Fon(~K0-
z>#EmQ`YMG|@D6MVN>GJken6;n7)dp}PAQ9;)L&hbR#Tpj(_n;8w3PX-{itoR6dk&!
z4#J$VqRb>u%CnQ|64!?%&MQQ%ko$Jn%X{XCm%6^0cHk1?4~uIJmh~#nZ|A~QBnc0J
zz^1uJ-V%f>=cise0$va~ofp5|Jq!e5-yiMbj2jLt@VOTG@}u($cr+VD{3jt(U0yb}
zz~BaCXuVB_mtqxFBkn)DK39LD%W`R&(=I+D0c<B-A1qK!09C!}4nK>_^x=l^-wUy7
z(~(_<v*G41J%3n*2e*i23TvCvca`?pZGS~4`&FZjp&JS+%&ri#NMG^29?rPHuqp-w
zTx36p8;D&`9Rhj*z&QbrE|1en=k$Rpb&-zx&`(C9XzjY))*N1?uHe)xb}wvUR^c1G
z9n89Ik2F$?Hb3)aFkofWEb^o6&6d`X`FeH{LjKx0uz^@k6lyX|z~a&Dpjh-oQru4C
zJA=ibTb)tQ+IYApUaUpcE<32sQtJXM=I0d4FABzW+mB{D@+_NbPPaEDW7PX6wc%su
zkMGpG4uX!0za|$VL1}F(Wfe1Q=6g2y#A&cOl1B$0jX?1UXoZTqdHMaou<5VXoRMY)
zS}6yy8gT`gM=`WfI-5mYiffIkJr`l<xuS|!{#{?b*C8~Erw_inG^*(-X$<KmFP0ts
zR&Yj!6XX%BGF*=yy<*lmV1^xM|3C^lz>InmDjq&C47GR>oGBAyA*SktsTBW~Oy$fp
z7HZMgy-GR2*~5_d7i53Tp2lbGHG=n%%l5o|l)C(O?$7xlce`*QndXdJ5auS^`$7JD
z!2<<8$NxP)kf0*KuU<^JFa#4Fg;}%70zzZGlA2fwPnX*jH{lnZJruUQ3K#a{Gr$(r
zah8k1UANjyEh-)5DB3L~qnxE?DI5D{?7!W@-?77=*#oHPPrtBVOa!$pg&PKvx;mN!
zDYJG`HlEV$ey{aRVsdfilB%Yy+%2d*_W9!E_U7+bY!DUXM^#_FlRmN0m5g-S2e!I=
z)RyLM=p1l2Qm7ET!g%snq^sfF;1Bcy$BmneY$fB=VbJHWpysUdp0@UjD`5!3Z<lmG
zdHCA~g?EUF3Ka#h#eBdXKNw`dZ^bH%)dmfL<<>kjH5aG{)VEg5B+5VjMuL?|A*F*6
z1^|@R49GjyL{jbeXNW<8m6XVP_G8o!wfgl3Zg<%7rA+fR#zcu1ju$6c^o|-fE4Icu
zQqx=JIs|I5J41v!G-%JdY;lyMu0-d2irDnbpm*M%QORGe#c>q0(J0ci$uL=S#Arl%
zPF^C~p!dwVO{`C*5yL{vP?>ze@KiLy!QzG%B~C9=S~+zW058Ny|6HkU@E8NM3@EFs
z8^2(WqB^Vk-mtYS{_M}SZ`ltv>GnT%U6pm=?CTR8qMwbL37U@dz5|Z(F!zPN!@8Ib
z(Nm;uf%(CQQI?`X^o!+fx1m8Nt;LsoG@oKSF@Dz*a}wryHkRO$Fr8q+<<h3TQlS>*
zR?fqd9*Z2x+JSElS60O{VFwV3i0Qc2TgaAAh^Kvxn_osHLd>6lR$HZF4zbq<Voq*b
zb)s`nbns-*kW$y2D$UofOP@e%YO{M2{Bdiwz#?8cNl$=se<_LiGzWo=xFx<>L9U8>
zZ`j|!X@~vXG>!F^vU{&C!iV<Gd|$`@r75D{T=Xl2ucN~6G@T~>s;7Y?d7Q${guNqM
zs<!7*m(Dls@e~51Qi20d=R#I=sr0YIZeOVPK~xR!TTX>+-3NCh4sBC^V2?fj1zRAk
zm0T$Qc<XoUXY)<_eoY8Jpp#zTp=n8^i_IrPTd7+$i^k;1M})jo>Yxj;;#@;YFHJ&Q
zypa!>hJ&f!F^3N9QaIf4QnODisiA`-CH~1zU96aUPGt2t2s{hya;n)vdx9$w`v}&g
zi@tr}Y(gH8w2d(F``YZ&QZRBfXv>jTG<35|zw~rpo&_x}uX<IF<DdX-WD93}R5<C>
z$C=RKN;(od)9Dg)`c9q=Wy{g+3*UmWzlOO+V~bXO%g30BZF}3&fwlQ&t16~-!7lPB
zrzL8vX>{}xCK-roN_Gqxwg;V6UY{wB#wV*w&@sGndD^D$aIy8SZ?FLi#$HD+U-)&f
z@%LyE&7XhZ(&5IZuDQdCy#d`pXB*H>1qHM{T<g~evv=$O8vpZj!LOy1f+H`3G+HDc
z(jiMv0EPAYi4Xo`_$A3a2}NKQ6_kfYmQJU%z)4p!<6f+$zD!fdDGewCO%&Y%Hwt}!
zkf=^K*gKXM(}7hxMW;{uVZ`?e$6Vjyp-E@HKlT$TtoSkeN}Cg8G*fTEC2_S6bhuUO
zJTusVS3(v@VME%g<G)lzEn)nek1WaIPTNp;GrtkavwTHO;k7V`g7OKGs)lkcS-oMG
zr>$wdncg;5V+p&A4JMFW{Qt02c|+;?0IzaiMGYkUMz@$~vYm-;v9E(<%AhE`>J)Y+
zV9k=~sssj7k`^cm&jLenp+P}+e*l71jlyVDxFA--?WB9D2Q`OwgME;tqU>ZC&;PpA
z0OKQYY5ncW_AfbgFy9qze7Q{0S`yclLLuW|xyb5zrBinfmtkpp^j9!+<~eiSa7)L6
z(-G9^xvP`W@;cG+$gNTFiBvF}uY<r{oZ*)+ry3Vn;1m`*G=wQ0+9<YJ<+gMvSkh_(
zE<6X)RjtN8Q-7b#4&H?+P!+<#>URDd>YC^gkEBvM?$wr0-1I{M@lBN2;@iw{3El{U
ze?b?>RnNSO>t!#TxZ3l<YpinDAloN2gQ@9QpF7^Y{G{Y-bKJJ?tEKMQxd$e+%<-}%
zN6NWLuk?@TNU+K!0sU6<a6O)9QbqKps43~{sAO8y&5w3}Y0G~`!jBh;>P!015gpTh
z?&u<pI2lKbu;ClyuME$y_Bc8FP`>vuq#}QtwWmo;i-Sk*G_JH7F0$BB!n;{NF#R#r
z1k^F{=(;{L^E;BD8?o$G;)19RIzaazLZm&IY4f{$&J)LQnN77X-cN!JWI}bKyvnm(
zOD2tqwC__nc>~T}KTr|Z)K$_NIY|zGV~-;v20Tes$1_?AwH_)lLkbJq^+>uKPkizv
z)0)3AVe7w@PISJZNiHaq%(IrwHXc34v4Xr@1I@P$K<=Y?lTBU&F7Weg74L4{ng94v
zQA~oXNDBt$pDN6?AG=qcm(s1uq%w7BJL)I(pVnoZGa<)X?s>}`rb|D3KeTz`b}VGm
zC}ek%O(ClCDPz!X1h6kFNGM#sLzsKa9+2~Z2zQE~wBV5#)RxY&b~19YmRHFXcSuVp
zB+dRXo!4D2fFYJ@;&P>%)G^d^t1U9-P>`H%;8tT8X)S3nM`y#~yB4R_6LbRVo`>>`
z{R5${@!Whx=UTon-%}!6<M;ms0hINyrpJ}fq~6w+m|A#EO$o^^xp0o`qJ6}b<!pe}
z@NMZ)Rqe=2aGYdNQL)Ownhf~$q+<IJQF~uOZu?eZ;PU8Qj47NiG#LgoOG?TBJ1Ec>
zC>44r0W)^M$D^(PofAnl?uDo$Tg2FF$GSH!Q6*7R{wMg<U5VP9Isv!t)~Qse9T;uC
zgp0F#eBv$!#;mbcy@h_Fa$gR2spZw3w5;(vfv>*mvCH4O7axe=eza2T<}}wDyP&LX
z_E>#!m_^p53kC;#tT8Neu|8US;M-i{;X}OvI(W|8&`%nYT!A@<T0ErrLOYEv5hR`&
z;<)r|;O-ERZXhYtcMI^=jrP-Drcve9e}>sd#uVqVdT$5TcBseSi@W>FF>E+ITu9^r
z@9*sx@{~jl_Id~PEKf{=Xl}#*Vd_1Anq0$f?KkvZ1Jb(?K#&qTXy~Cs=mC)~p-D%I
z8mbyY2Zcxp5JIR50*Z*D^lF5FNU<Ot!7W8LsC%E+{e9>E&y31A!wf?R_xn8SUh7)S
zoi?4%d&eB6pWmi6xb{2_u7#>jyQmpCSm%aas~fZ7yZCwXDq9I=!K!Rb{a*H^IlQpd
z6N8uYp&NVQl?qadZsIOS0$*pE#4Z{9CS8GjwCr|{W+o(e#tP+X*?*2&%(4;TWr{{I
z3>uqMUY4LqO<++%C+}LcXQSRXq;NN+iZ>jQme^C+1_^wY2ja7OzBtwfVU7vgS|d&O
zTG6gqmO_v3(qA$h77W{vzV0%OG4fF<V$;4>u;1^S%XAYYPRN;q*x#Q~X3;A0jCXE)
zI-T)qq%z>6YMNetR9I+~2~N)YxbdNSIeYl0XKS%PN5<q8AG}&<5r?`!h^@6!CiMT!
zAljH)QT{WQK?uopbZ1@aGfBu#PwR$YfJ}Zokrh1W{mIoopqxi=o_Erm;dvcvg#e(v
zU>|}e;&l-mDqLUdk5w5wi?*DU7UXR>AxY~visW;NO;f?U<o|O5E&iiCIgM+W=WKxL
zfoFhU{_BEPX`fcuBX7Uw*qcIR?P|J1xf$?(v$9{wO+MZ&E16E|dJ>0nXJ8hSV%S5O
z4mZbsxE0sA+_361V~sW`IPyL`=f2zwK7vxGjE9$)W$6a`;((6KGL7HSQ?U*=mB-!i
z0+C+~MSNj;V$&Q2bNP^y6n^XM3x#Ls7xKs1Ih%fulTH5tJrr5Wp597(h{p<zVzdLV
zn3Jn*uGlFq<3lY`Me)#yA&HLa_eo4xd7EyR@3P!2nD>{2_aEl;&-B_L?#6X8f?_jh
zu0d56Lkg0Y1dc-Q>GpZNGYnAptC5#Qv_eB)Y?ghdY;ng9a#>Ofxx1ew{%jT^qv~Ym
zogc_7E*y<C_f7f%hwU#>8NRFo$I$A$vW%$^2>zv}qL|2=Af*9`no>^iVH2Y%%jBF6
zqqktE_H*~UR_XG-YCbEf8Lym&yG*Ee8>Ixvd*XCQKFR|U9WNKO+q9XyB#L2rJu)k;
z<`>%QklBWk&YpZz)9P3Jty&vhx;as8g1^ckMeY&jj`I(qUX;`gb;9MJ<$h2rNGneu
zW775_pCA8hnA!29CEN|{47lz9{p=Hk#nVvYjUn{<D~LvqIlTb6y3+5L;P;&?Yd4(}
zHO8i{M=8xzsft~xOn;(&&NGL?9oaG6r|OW*U^39fn<U@lT~TIyd}ozPFO|2>z*Qo*
zky3;ZSVb92<r;;mWC)M^QUvRK2cr6SkipOT5=b~!gS3{Dc$Tj!v@t~FKOJyc!t7CA
zSFkSa+znDpn672};-t!PE~6HavAswA>gVGvw~2dE_@0i3ziFRt#P!|zwmc<+{N%If
zu5bWvd>`JiPXMUtuDMc|Dqa3L8&&@=@X)U3-fQP)9~7mHUg<$IuRIUk$mi}o#|h^y
z&0X7TK?V=DP|H$IV;4G9K7UxuGj@)!(D-qEcr=f=9d=>eHCKjsU^<&MQ5a)Ip5>F)
z72sD7NV8=g>eMANLW9$CRGc2Keh#7Yx+dYyMl1|#m_8(R8K3)mdCLy0P0;?b;1aQA
z*IhJj%S9%j?*&^-CV!=en*ADW{z>B>q25!tV$l-V(WcyO^Ha&qSz7N<;4HZ+Nl&Sb
zr$RHg<c;5x2xPpG{1?lz)O8NANCt^2H0}fSgRP)?fBcVXjR57L^;3);!#|h$re;6U
zeuQ-GO;|98s-i3s+B8lg8!wzR%L}b!ss&^<4(t_e*|I{NM+X5ZVR&<tnqrz!#p5?J
zjp;FAt&e6_+H%m2A_3F{^s`Q*EYYLj{G8>xW0lbn1A7w1yQPjoxq8fV&lA~OgO8WC
z9EbuZF&EJs+fT>kfgv3>*bJrhQ+nk2=+Cdb<+-Wr#k^C|g(;_QQ+AlM{wlK^(NmT;
zO3>SL%7e;M{dv#6Q0gom*SeQ+Hq{o-o|6z3()j%DNTT!J!;Vky9{MQ#di#Dmb}}p`
zDn^$)R-OM&^DM8)1z%1NcE>SOx_|zvq}{%lL@2l?ZtR?AjayDyuPe6SwFBX)&10%W
z@y6a6NiFDgjC9+EiqxrFOa5|IsE`b%kN^Br@}u8#iX4!}c07`0Pf<<I@0nqWo6*QX
z$J~k^E9CEQsNd7vsnfo9IdouTb;fLcNA)*jxI2APRmOIF6Ten2H#8PjJdsNI^v=cB
zW8J{EbW`7jAM(7wEpvp@$W#1Bcn4?ce=B+w0r_fU5}-KNK`-apAfQ+{TB5j!VbPxU
z$BKZRrcqnqy*o_HWD>!9pLE{VEc|ep)&~KxGeiPH2OgYu2P{p)F<Ix~9z%=M?iKvq
zU^r&NKe3KIWGN9Ljer;v4Hy8jO6dPD5X~Y4qQi~8N+|!;j{~y4ReR&t8!wO-Uzo3e
zxcY!ljmiuGv}dJ{H`uunl0Xj>Ag#bNu|)_#41B~+J0x=Uh+g~bdqwF$0*l{mJpM8L
zx-vYX*Q2~d%;hqwIe1igd?M$bfVii$IiRX!hDvs191rtc8!i#$-&v?Is`VNXv_P%K
zFo#wKrJNbtjV>Hi0z9M0ore!ZlRoo3ndsX$QVO+x`%J0)`i`crml=~`=kRn*k$qQj
z>f7_52V)iCaZkxtmO^aOeN!b(Jqu0a1HTv)|Ilms>||r%ryr?bW1L#}>;9u#{{kUV
zA1jnS&NccBT~_0w-1w{U@9Qjo|F%{wJ&RQ8r)cqdyJZJNJHNMEwHG=x1hcYhzGEz#
znAtLuw|{mbD$LbHj+C@#YmmKIH{_64JKmM$mQXq5eXQrZjNj7#bWjCO`7lO`X`#~l
zt=B`hxix^=MgoHid%C$ID#elhi|%^)Ml}>zF{MPVP?lBYFMNUbJ^b$(h#EGiPtCKo
z^Ty0myGqV}%h*DV9GJy4JZlT{(hNLQUh=X1rR|dJ8m;)v%i=9su+Go4Fp4R3_g;n3
zmoK7c7}i_YF15X3TNrW4wQPYesfmd^Zd8HIBT+UDbU*gOh#jt30z6U5_j%8^`zgdC
zTr$$U)9I%hmR>NyyfW0aSoagystvwAk&^Jj?U0*uF+J@t=sl;c(S|DqB=#*3LhhY+
zyUo$s)DC-Xl1{Al_`xl60lQa)kp@-MSe3?b1TofiL4H&twFAP>ut~2a%eb|+drH-i
z$q;j{MR`LIBI60W+yy)nc!Z-8r?vK-TUFA9&Z8S?w&r?{{XB9vz7OAcP2df1oH<Qx
z*g^0bK0HsKB@S%a=So*SAONLGxmYj`BL3g9k;7%xk@0q&i!?3vm+4&R^e<4reKop?
zfsP&Ml<@i|HDAde87MM2i<igG@N(o;GO{S9>3>J14Fb}-WI<wJ;5}UdIVVA1W@i)n
znzvtY%a5f-Tkgbs<@$E-NwMCgyA>ClHs3&f{{=Yt+ddRhmTD=KF7j3XiU~T4<?xBj
zN`wI2;>i^T-^Am_;tsw`RvEkCo=LgavpJ!IJ6oY($0OHz^R9buSCJj$!35!IY0<)c
zgkx$;!Kd8%x%^sahoxNugH7tt;;P^1VYAy+6?X3+liXXq$h}ziCl)Jfw>LT6hmvQd
z%wj4zO)7453|Z^ei4lNAV=ED1F_=J99=wntL0^mTevx(g=CzNBMos3?Jq_nl-pOk;
zmVlSNEpo@xqbaKqW=Zv3Pxpl-M}C1v2VysQHx)lcJ>)m&EkGq4JKc@GnHcpx$x9-=
zsqjQn^G5vP;g$*8TN6iQvI9I~A31iTvli3xMwmUukalEb9&r1$$#ZmwSaNOklc%_C
z9_ov=Z`&Q~-s-=VSHD9y?EgxsA<obmZib(a7@_AI(x|5;yYt^<zAumX^sYYpK|}WA
z5As@qt7@b3wBL=M1w1D{&3FMykTlz`G69izt+RrzF)y<$)QE&^4UH4MhKM0At#hQy
zcwqiQ2|E7w4+lS7F-O^%Sh@=(4Iw>N5;fWI3sJTORn{ntj2&bE%Sn#g{l=dJTjqzd
zhs`}l0~sGO<xP3}Q)X{%nKC$39i|>rKp^d?D3L61f+m27=-6}v?a-FtN!^?1?K^il
z_MXIk)Z)0Mdto-jlywr^RsVN=|I;W1YCUwJ)<urcYAX^0u0#sW8eZT$2X0Tyh#+sl
z7R5UVSj<W$1J_6(c|tmoBED74p<=i6r-2WGiAD6o%ZLn<pbm-xJI_&2kpeNC={ZvI
zzkqhSnMTUY+x=uJ!#3@&!_4z#mSqU$vTqir2+_?07;vDNItFlLbc6Y~Mx2Ql0mE9h
z8?7AUJwxC2MuE`7V$BK)uv7>#1fmFzC5OPC)>s;4<a9cPGt#GnkxW3bdJIuR!&184
zRGm|$3U_jsw}s6_?p3xicBdJq``n_=c?yjvXSh{nwsk-%nbH2|1~Hnx#Zo%-AG%cH
zeul+!Cdy>|p7<BgpNiSc@e_|6zuJf%6!u8|f@K!4w8vlaxmUXO;>gT@<w4MxwVj;z
zTx#*HafJ;dfO2SaJ7Uflo~DTpE_r}m^j?4T@^NHQ{f0BC?c!2_*JhZ4U`}z7gt_@S
zE18mb4(<RejsY4xo!I+y?AXDfF0lGqRgAjZ<*8@Ys(~>sdCJrm_jxW;aEA;JY9?w=
zW(}G=RW66uO8uS_=j1)UlyyESR=_jbu7zW=M`5xtmg<U~eE^Gsf1L)y^xM9w+a~Xc
z9NjO!{Jv0vc{D2Z4d<_XMbjd7k?HZbYu$c!5QeJc`~X8|Ud+YWqAj;wQ9ZCnYxI8j
z{I3Q9s}U$bB}ujzau*dKIPtsTF78%+(L1(mmKoMyCV%12n?%5)Hpre6vv8T|wGIB1
zX0K!Z+AV0gZR9!DxOu1PvdS@}gpx5g)c?5<{<Bfjt@g#WeUOP+^DHthR@I9=85Pn$
zu$X(1*R|$J@X)`&SoPFUq(w_JBZBRzK>bx-F~XOk7d7UcAj&(4w<pb0=Lsa{{i&<@
zC7k_8zlLzoxlpF7+EE5<U#*8_gKlFTr;=Sl6ECEIaPGwyG&XMe>QzTNJJo^t-jo@Y
zAKBv)#RHz2z+yXuJgOEwWpC5~;kU~?acACgExTJpVA2!@3-iK#L%=nQ>00))aAEcR
zhvd*Qrj+5rC=<(ZHeZ)}{`(!RI{Z{uw{u@T-^hcw0DTuq5|K4aKc9^gyx{~?T7(Eq
z8)Ueh+e6F4Hjb~pn?I#8j(@CN%Ah5?>3zW=O<Xjr=3QArbJxkn2yXFEyLl)u6uDkj
zF$}^wkWILI3j5_cb?~oyn!yX7FU7pP-Spi%Le%QSVe69fam!iS?sPNz?tz(*_Q5<#
zHp%1MO}2s0$r=m0!vGu3L(o3i&DZIf+F~nvP#)lZF<!JR@uJ?YRgQ*QTEz{Y%Fdpk
zeapf{%T76+wi_u`?XSezd0bEE4c}?C?zDfL?C|AL1H>6+wI75`=0v@ms&er}J5RE}
zc+VODu^@kBv?YgbeD@6S#CY`7t)t_4k4$a%oUh(1dU~u8BTQG+W#gl`^2qwj=TEL&
zdM1?IMVvuKs6b#ZujRIQ)Nvb%Wn6a6AT>BUc@><Ay=*Zf+l#w$&!i_A;@QqwShCYG
zmrq$kIeq0ViV2y2-qD_Cj&l_%c@(?jVJTJabgP(Dr4vP3L>DbwvI)>F)IQA}a0kid
zOJ;@DXN3Z}9YXgHlHJxCUj%tMPECvXrnu{^uxsU7E*@Xr5E#T>B{*Go(zMhZ>-Dr4
zo5Qd3-byYI#e=9qeU3Ua3r)}Vaki+8EtloLPp1QTVxrx@ei;>}1j&5%o{0XcRPlD4
zN52S%6t2F?C2f*?)Rd36`WqEsGIjH&(CN(g{b^xtOG==)D3wYleR5NzjzUJr6!OCw
zR$a-T6ziqTvm~DlnkZQy``0M1PC$-ia8Td;YBHfodi`0A??}I2`VS*as~WP2S6oH?
zzpy*l1pFVJg0;bfM&M#aDNPHE3Me67&{(1bWN<({Sl)-kKtyTRx3Hh5OV_bVWAhZ&
zNfZdY&iXH|0t)$cP);t;eqheDz(Vx1A(%?E;1a%oc0Tz!dl)Y)93>q6A&vRLH_Hd#
z!2VwF1JdNABdG&O0{`(~i6r}K)8l9LN%GQgXJ8i`vLSE)B?4GGA=vCZXVNsq;)f+~
z!7D-RZ9?Ya!UU`AL~6iXGomdkl7)k<%ZOz#@B{fyh+rgBg;CC{CA`Kh>W^mUpeQdf
z*M=tkk#U}oX=JQymO!m|Rvq2~!?e;nX;y!oH#(~`IQ+QKMlmWOcmAtVcg(O{5&CR^
z_MTc%B}B{r!`i2kz&r|h_LV^>zwQ25VE$f!?S>)Vv4c60QY$`a%_=rDQNM%m87$TE
z8jsES`Hni*OMY_o0-xWS&P(f|xM#K3TsJ4hm@bVOGe2#vaIoX2KPSt_RXf+7+3iXC
zGLd`zm78=LgUHLEgxsp2=b|-6-0`gq$}L<TJq8+`N3{z!kM?gBUv+Y2sXr4R6WILK
ztL;H(@kJ;1%uW@E%%}q=(*F#(3<fbAGklW_R7Wjd<;_d3x;gRD+FzMz0B0o9v@MnN
z`Bt<Y1Z~_0e@k;^>E^p<Qd8GuAd2|~+Ny$>IdQ?-zxwNw0qv`XiCcCHudzbk3lmkC
zRiFCVa-7@`FD2su)BnX(2>SLEs<ML4n@g5zuYToRfgM@2-8Ytd%)EZ5ZP%%p;rb8x
zEF(_V6`v7h`zh+GCfk&)Aej`64fRX&)S!OFg<UY96+1sGmnpPYF7JJ&G1}=M&<Wd#
z6?s_g`7tmsx3?i;hRv%tPDoJxM8;*$!p@wRK30MT^xR36h@QI>42v>?(Y;jt6l1mJ
z<|A+?o&Nqm=R+CHCK7$)o7gpPf5*b;dc1e#A{b`Sj+VyYbS)1ikOU>s3~_EOI=&K7
zoWY^)3KL}(+K*$u*>e2-it~jn)>7iI@R8qsD?ns$6wpW}xWFO$u_%cStnKtV_Dnpu
zR}O&JU0X#7<r&@;d>sR<G|5JKgB35iR<V6#*c-IJ932^4Pjf_D5g63k0@9{Q=6$2Y
zfLFCGzy(G08IJnDTEbEal!6?L2IHJce$Pc%q_z%ak#>c5x>IJ?hCcPJIT(L4yI1Wa
zm^#uUW^S<oRuoaTozMeO?=O_3y-^5^$+@xr;m(Fah#|N2ll7$Xzl=%40)94)KZMD>
zCwFcH(-kU~-A^ul{EK35WM|HBO)t`lY!$)*I^qDTMs0Z~vY4fqElk9hMMq7dbAFST
z<?HhIaWS^|%c>$VC!#3#;IxkgT}|7G7lYapC9nVE`cLOt&|mS&Lq=SL2f`O5pMte=
z{Mf|5SdVJmI-c5WiTnEEW3=(O(<X|0Pcl6A<Go~ph(P8z!;tUW=DTS-M~!bj?^GFx
z7FS<P(JCAm(vtgGuzPONY~gn|ZhG0t3OS_2$pjPK<&ME-`LK$)nq1x9Z5eY|(r6Xn
zc)0c@(q7rU|B1r+@L|1^dk;mqQsMUt9-2*d*$o2$JDv-xDa~K}^-Me7rKR6|dD|x7
zc15(}>CB3u|FghYzN6}PBCls#&8kpHmZ^fXXZx^q^cnF%SD;H#{T`nhi&kyuQ*T^0
z+_t_+GS+kM<I5hoTpchM5h;0}&44Xc;rNqFd;U0i@GEE<{(8d+8RUN>G}k%r9#-VS
z7NxaVBOe3F?GMaMXo{-X%Cl|=m-DhOJ))HQH8NckM&qxqaKh!Cdhx>1GqSkabAfu1
z^Q{ykj-YP$i`v(>t!{q+3+^any8jl@XH@|354iq+U1yAPf}xXJq$C)19qDjUMCL0v
zYrxtR$+A!b?@PEw`Q!Ppw=;J^g4?mT2qrL0$pi!AqOUl0Es-RoY<&bty$-`hrB9g7
zA|{FnmJ6nF(~-a^SD3aE^L-`)zwr9o^|$`!2O$R`kNPKTCu?4FYd53)t!6#Go6d(t
z-~TGI`{PiIt;J+sVkNN#ETIFFhm6o$Euff8fpCJ8S0|*OX!z%l8<$+pYs%rz*WB8;
zVPsr4&102%+e`SknLwLX{oZ}q*)@CrlZCP40N&YYzwtw-hWS7b&ljz{rb{Q&Ect8v
z54-E_E@gsUUmC;oEnyxQPF0DUB`I9%9iacHTy|peN?SS%Y33@l-kn~3sl?vm8il;5
zTGN@ctYFJlgTV$H;(xSQg{!Q4Tk%z;y2TXrz9j_&1;`(m1TyY=$^Im_mq}<V%9Qxc
z&rz$?@t-3c`z@du1qA9u8-}vvsV#7KaM><SJCW)`4hYwYHwvq4)~~H8{#FayK20Q$
zuUSrA_)B~TSQV3~X)SKWZx2naXZ61JUKBMbWYWud6X4mdXbJXD4tma)8IXo*Bf@bV
zMb3eOeBQLz)@iq=UAWKps)=r-yxnkSFWIXy@Q7!sEZmdK0xV^BwFG9ax`g^#ED_0#
ztHGsryXu+BuN6VD?d{BVIj<yJnDGsp=D;~4>$WeQe4IHrz<BtJ<0EMbUEOL*RDX42
z|6D0OTLk!ap-Te?F-+Ick+H{X?oRs4iI-!Y0(`#04(OrKsr9!d<M+BY%PZ|dWC3;0
z$7TlC4JIdT%vk~@IvEZ!Q68V1e7oO-ie%U++!tPH;K$m<z1FsbSIkEQ%W6dmyfL~(
zM=3Puwpjo1$ypjCfLU+&f4-@Y+qoq%cCkdAtK))OkZitg9fyaYVG)Acu0hzp>?%6l
zsw{g}JY||rrPOeyZxn!l(735r7StrNWl0o_-hW|dth3PKqPjPCOQ7KSgO|#`HX*Hp
z6mI}@7!9ApxKJ6sS7J_}Snj#J766z@BQn!ZUZjvvvO!DHj+X}f7A!5unPX$!4Zxr}
zpHu@k9aea*)GNAg3vsJvg+%mKG#&)0d?UO3Os46I^;hDUTI#5!QBli-q3_2`(~fT?
ze@!S)DwreHJD}{GT|H_SZb^LT9QU3$=)Q6>E5-<}>(<=dFUv&yc+PS)0B6p=F=dI4
z9L`@GC25%DSVo4%K?k(mW$QF2<iG|x&vDuof{#DVsR+U6#WAMb-If<rtfZ!)X(RB@
zPUP1QxohPSZzec>uVtgI<nsPvuNs)svs3jp-M5%2`5H)+rsb;pq@9_(Dr~rB5jH>k
zyNIm^|MXe*ykZG<bkOu>q8v}{q5@$eImI37m!k7ngWdDSa<IfTEa$9)mcVYQe(IwI
zqRH~DWXcOquew)a6A7AHv?%>{!SF!bvxwT6&TE2->~v9KZ<}0gmb~DTOW(kd7?zlL
z)QGN;{Cc7}#iDiK;TrljxG-5vd}Ltqh&+_kdh*M5Ev1MpU-jtJ4J9JivBWCguz!Bi
zv;D0=zhCsOBRNLNW4u5yVGGUHBijd_>lx~U8ir9yoHFN-tfvHH+VlPOfV>u)kj2)#
zpE7mD*-Hl$?w}^lV+;qj90?Z1uidQrPK0<VVa}e^Ixh;5n7nrmBppL-ET$}5oaalr
zBVtX@YcT?1{<X??E`Y8B)NfYQzrgbP<0~=FF_oP&4qz9K+j7-&#e=*L(VNNv8cK>-
zkd5;Hb=N}FcHCqHWEJSj8KuEEwJ_Mg&NV5I0m{JGuxzXozb(*dwgHFG|D;`O<nd2K
z+~NQSWx#8vwP}c%|NS8PKO-}rN(3#lJU9$UD5zsc++wU7q~}0^rP9dICNc0?AEBaQ
z^50yfhyN<sj7D&b$`!``Cm~uP*o{F+w~WXLZr11;{|H=O+S;$rN_6rkdrY%8_`4;M
zco^4`E+@zL@GNE<JI_o%R^fi0ZRLkCX=ND}b5U27rl+C;OtnNmDlM4{pUlU8RTeik
zjF<0BZ|^Q&Ud~xQZ^y#$n4@jeNEu_H3^pn)@V3KtbR_H=Ty=v#xofkk)_!lnv~X-2
z*^iB^&s{t=N26aosE&4R{VM&;R{vlg$Nxwxw&KN%e5W4p{KktClV`Q6L9zFu*Zkw)
zvTziujDNEMj@)XUpmVtF8L7@WB<M5z*wrX>c=YB4^?jYo0p8!ei`t%78L9@N_#Dui
zdP7&BpJ*(V3Jv|V_Ix}DC=go52Bu7^ez7(6zUPe^nD-_Cfp~j6S+UoeAn-{K?fT`y
zVP9KhzwC}%eV+OreF2iiw1T*?F5o9QA$&*#%0;fl=Z8lcaRTtvZd4NVJtXK&9pGM&
zI)dXrY}8fG$6)uGoGLp&$Oau4mLW2=uYi@c-fcZ<+H)7Be=7yH;_g1y33$o@VGwyN
z_Gyk&oVD;9gUgL5W`>oLTQQ2Q(^_(m=pKFR6f7<AIdd|XpXaPNUTp=LN+A?~$~QQF
z&|-T_>Kqtf)p$5^VVk3k$noQ<z7HeXoPJF*N1Zc&Lw8e2ao#Y^EuX942R7pgI?iqg
zgWakh`|9+r1z}GG&CIdb6haZ5+&Kq0#O}fQA_o(jb*w_cn|zrgJuQ0m#oYcnKo#M%
zW2dLU{GG_OAzxk&icr@T`=%lNm3H7#parA@py&tbg+%!{W_vcxx(sMB9nz(roCh_m
zn4$axAfWPfrVyM3B*UqPfb1+nEC-mSprCa)ZZWtO>o?yKW{e5_q9fqjV`D-pe7=7s
zwh+UXfs(sjhB17JHiZ(0t?}Qdu>$7Sy3he*>Gj%4fPIdP1~f2a>^aYWftJ?&kC=3|
z<Z!6}%QVqr9g;O-aZzA$r%JWvWE1g^OVG`4e24JsY;K!D-ML!AOMULwGQvYMh=6-_
zZZavrJyz4JGdCsX=Z}#LuzGm5s@BJvIxbd5r!^8Vf9hiTp=M_D>|Z5CuDzu!;VZ$5
zLoU%L?`-%Vn>2sna*344uIca-h2<~27k)Om*@<TuCM+8@F)dg=aJ*9|*A{)rBRPSz
z8piUor6L5NTtqiDXrGfi9!wqLi&&9)K6U3|+^4;hlDu9^N%YZ`@0vM#w)!+`)dPXz
zE|xuSGl-^R+o(Kfk#=V<J#ehoZScjsr!08<Xjr}8WL0dg%+Hqbkx<yTn??G?D9vp#
z<@KQjb1QaaokWt$c`!q0Pe#!dy%0MkJXDThJt79sK0W$T|7aN?F{N(x<m=;F=)V?6
zeqm||w`zQj6LNtHQR+sda3QvV+Y&O|(H`M1dz3Fh@nlzV%>IilcCM1{(hs#pTOEC1
zqWce1?oUPx)I0s${J;k)2z&?uwM$(QAOXWEB?<x;fB-EMFM_W;hEq+*9bS34la3!u
zNfQATT@?WNXL1cLS}ZXAY4oMqfx-Q!L}rNnAe~@gT8@0F9bj?feMHV2s3Rl)+j4y#
zWF`O+PyqBtfo3TXzz8CSKpw@JbN<V)fQ7Drm9_`8&2(zc2Zs)<<#;8hXZ~aSC;lBO
zE$c2iwV|UwF;C@|<nLsV9D=?YvZ#I}TpxBffPfLfc+StgC<uCbXI1PbON;z2pVULy
zBsC>Olyd2o-mZnka~gx<2s3oqVmV^M@7k_c%Q?KnlPjk6UE#GCX2V5Z?L~Hp>DB!p
z(R-TG2?dJu(>kJJRh}?En|<xw3~IC<{qcckJ$TGDp5?kVE2QRjCdKP<J~r0X;PAWN
z7$lmj(N<|E-6(}2XW)P`-G-6|4peAIMF<4zk_MiW&CM*tax05D!hX-#{tMi@XliRx
zA6nqD0pg$ULbkWNR%M@(7`UDv*m;T=2A2xnKfd$SSv{}#`z8kX^fOy5wZEFQWIPF&
zo)^QlsEv$}8IU1CAl1t6&ui{ave&25&vT@>sA_}G+z=wW;PF*varfyC8Jeu@`iK^n
zkq!Q<&F?yqT#=?XjgddgX%y(XV`0)+y@FgLMeM;0VQwg~TZwDEylL?}99!<kA8)qL
zOAq;V1qvW~EK+Dqqd+`-KpO>F=w)<O`ME>@P)oG!SMGlinTG0~sP9e+fx#Ks3AUFN
zmWB^<4cGu*?=%!;v2Pc_VkK9EWMVdA1hg((8Lf*=%l*9xNiTL}7<T|$L%4^Mol97E
zk&J<ZI#8gdoEyrKZF;~q({WRKxi)eoQ&Yy{SGO>CfC(-zZ%`|DfkE9;uffGp{e3-{
zYJ@?|4d6gWF^kinVkk)SG=Oq&TetoWfoy_6>HQhH+AfiNN9l1?27)PRrKlHh2#wZw
z7azE|E9HCh7S#Xda?#hxE&hDOs8-gBZreF$NWUsK%`GXeJ6ij?Q|)`^?A^>KoCdzP
zPq~w~^>!}Jcl*>=zj69GB(PVK7<+Bw&3jhE*hy0X3T|j+QZlLVQe#RXvtE;H5f10Y
zUbgUxG^x@7!Zxnm-gva-RiCS*&KkQMFSdJ~%fbD@<crkGGif%pW`Qp_b9sV$$iFLv
zU&_Yy_EQqrfab|a$9o<}dJl`oXRh6TjLuuuEfY-ubNR9YTTc?A+MS#iZ#L7By+FM`
zAal@tNSWU$@bdDPg~r!UPic?fC7F)6z=;HfhSh1I_Cr_6CFqX}-VjD939k9!>DGDi
zd#k6k;@h<bd&@m>O{fbr1uBXT8!=&Ls(1q0AYBgrcBZa_&|J@KS6V`&YUXH_HKlM8
zCYp+%^_m--B#^DUZD$Z8Z%!YD2?i}HzG=B?I~HVV*UoF(njz(!VcpM(^94g9|Nq&D
zA4ia#M2?%*L9?a7QBw|s&@!nE;j^MzfLi$sE4WPtdn#$cRqgpqV<yGG@*nizX9qNs
zJ|P4`2JM>Q2hB-C5C;a2BpN@<?*ze3I&e<T(5!}Pb=aSg0=}E*2;gP)Pr{1<4qi_|
zL=`{;X`t^1fzYk*E5LkHllT78R_=hk*VHSi+}SAZV#r+nTT$(qbY*3?(k$Nes3kC2
zTV#Zr6I@>jvR}4~=Y{3a<pNOG<aw8S_XYNdcYg<h#JBqB=EY29-rx(bYt+KK?PMy8
zH&uPCkjwb=(9X5d?rsdX@oU`j`D=qO<Kka5#a_*_JhAjww*ELEn+Es&*-)p1`RWqt
z73;R>+*t*Y15R}WqSOpCJgij>9w<a>B9j_~t0-G-OVvu+qx0`H;SL;GV!pT!#Sg1<
zJdT9zwA!#AI<nIsl*JYRS-&PA6>56sy4i+2mPaIC0tR_cSruN5CkEQaP7^N9KKg~Z
z9b)M6R%Zjj$47tVU5kTgEP}h}U5mw-7>Z7JJw%jwk5IA3QTx>BHw=Osj^c<h5Ghvc
z4?IBdJPuNK3=|wxlL=I^yD04KNbjlfVx?vfKtiF%btrB$5U*ih3<Ax8%~?-u@Pbj8
zRk#$AqupMTsz?l(^vjkhg0fubIP?Y&7yU}Qk~*Qus!Rrn@N9GVg9jmoY>CG_@nrhr
zfZSQwNx<@leB4^EVs1Gd4C17D@32l-kI0<SWyp0nbiJv{hfz(WKG1%+;gbvE&%Ugd
zexYvvAG<F^5`($L^z~4A1%>HS?%NSL<utj~bu5%E3JOum)kG=nNwEP{p+kCxr;ftq
zyXJDIY(Pm%h5E)b^UA5DU(aZP=2PzzB@v3iYqld{=!1hrD94@c50$@GG8uGs?xtHN
zx88yeT&Okdq!X*DqB^QEfp1$(2*ex;fX5gFn`mA#lz);=PRx<Zj-H89783%wMpv7#
zQYsL1Q456r8aohuVPXUZ+isL7b-jm^?UP+dz+RQwsq2*C-$|_tgqeh3qI7oh>1^0f
zzr>L5(a&REq7Nvv)h5;lRv~XE7MI@CF9mBDi<`$K=<jyEW-L9q`fd8fU4GLA)oW$;
zI>!Z+c1x)--K*7&YFjJ9zuu@l;_#r~4trxav{ib~T7GxY;+uCY@|WTTS}XIJsKJf=
z%ATA!5%Gsbkb)uhgMS3xm(f;VFLu8ZXot^>MKQ)oD0B<&zuFqSHn010_lF7PP_K5n
zvO<}%F6q3;yU}7Z9~N^^Q2s-J`$OmeEj5#oB7+Kk(_%xH<Q11@z;@f}g?9oo6cRLo
zVG&0_3&lEZgk>d*=vvJZywWz6gps(VUjsjBGhLf`z4S`w$3$upPRymb@&5u!h4P=X
z01JDeS+A^x_fV>am#~hq;@8ma!$F0obFTx|WeWaTV_|`=%S)8pr$~FU`c#C4rj*(E
zmzPtnhGOyVCK^03=K7b9vMEgeNm`&f@n6L>!z|n=)WXZB7$uZlh-Gs^aDx&S!#UdI
zd!jMYTusCr%2A;7`oIcHaz>Kqk(4*kq3j}iXT)%Z4W3Hpk4N!jy4bxUa|B~doq2j1
zbikC)vm9D>_FgR<DoxfwgSWA>8erH0Ovpf|LBlxgkI65)m<<5ak2t^&N?gH;)f37=
z$Kld6O^f)Z#X*L_(A{8ZJMZ=Fv)&c`EuLBSU}p;=+gsnFZdD56y}UZK^66z{dI!0)
zyTX!sZ{euLn3s4ig<y3)vi|2Zj}hZ#rQVT^gf?DOQm{NBGqP{X#Y#Kr#ENwCAdBJ4
zbUxcHF}3=yZeKao=y{#@3ze>&Z{58;znW)#uV{%fTI=@iGRkDB`f@b-(WO_q6DY+W
ztw}S{`m*rRJI%Xp1KtCdn5TM;w*9~E=_xmfjVq#zw-ukZ_TwX<ET(^2gBcB-pP+;U
z$wi1hkr)Dl0em`yPPURpO~(03Z7UA!hU0)@?X~0jL50W_tcuy^Ap@F5utxNBK1;&L
zbBR#hrJ+Fw2U7%}wriq-)R6%9gUa<%Zg+)@52e=v5L||hPke^{R2zIi{%RDK{YM(X
zdD+w$#RWEeiv1vW0&3GBS%<?MK(g^>jFJm-i-9yNL)vIRh%`|EzMs%yZT5H3Od8P9
zrDY{YkiZMRZVV25S;8@zPLrzt0wov-*M*y*o>7c|zWVJG5`9MQA@F4Y!FD^S6d!FH
z5Gb3rWrwF9PG&#{DglG&eX;ijj3*Wv7>+uH@vI+|uA-S<bv}D-nIR*MP&&0}1eenm
zsxeDA6o^9}jAn_s4)ZRM^}C5o!>rNkQ;V`kMh??ti>Gmm!RahNN0!uU?TP2K-NqJM
zg*PI<=nDASiCx=|@V89~{HH@FN@;0i&qiDCM`Hvsz`?rkZPy44vXC~OvFkG1O_pAJ
zJSa3sC%|NnnMfk~w}5qH;3|q!*Ac)xr-FrA-YaQb%LE9~7ftYH1MXD8m)1)K)^?D^
z5ViUQkm!U4$~jt<ZA^Q$p8yiemXwyw(51^RE{Xcz@pQLfORb6%`T@cZMI1xFe~qhu
ztPy`na~JUj0E2D!=epm^1fE5{n!GOWb>~M%VM?`%B&==RP*;7de=~GDEU_)^AX{SP
z*|uZmu}KMhoP7n1p5e6Q%jA2aWp*6}A9?7KE!#OYUhQ2uSifAPH+-BEhYdH*O@AIf
z{iK7^FdP^rvFgsxLd0+j3A_$XWLjB0C>JR|7h#Rso7irz7F!{qX?^h*G;cAu^k(1n
z3I57fz@*pgNh}069?B8tK9sW+<ebNM@8BZp!HzL~oe4C_a335tG36_*n*RKU4FP8A
z>K{N@rzk68EXi<2vQA1|xc>(9@4Sf+p7-`*$s*z7C6jQ*6xps%Ahz{>FW*x1eAG{x
z?g^5FZ4&BNYt*r29L;&A)>7SF+@*O%lxxOu{h3pK!>gqfXDcvk19o=*&$jY^wwCwE
znX5E>S1$|>9&{!F;W{QP^oam{Cjff&-~&dOzbC;UU9BvXBSuW5LZpb!cAEbil8N7`
zjSRQ2gTZkU<VsFDs0f-t>zNA|rdmS=qvOmINs><^PO4&{8E_)iu@^&flvkT}md*mZ
z_&y`j^Qm+o<{T4^;H-}WPdsT>b67e}I~gp*!A%CM6?eCnG>~%G5`>!UZenfHtl|1D
zGNSW+`tIG#uDo+eUyU$lV;;1nUB7m$GWt3f>nc6&w&m%u9=Z3WPq<!?e#h4A$6V~C
zCu3?adK=fK%EWuiqCV<=^X%RU46UwvZ!UX9T9%=0>ymtxpBOKr61(Jcj7v$$ExUNZ
z1IzR%+kk+dZsLO_gSDyK(c6m76^8-k`ZauOdGIR)+JJKg&Dk9mgLnBULbtXP;j_oV
zEaP(A5~9`QX(6{Pd@O+9inn;%@U}E@S9E#gtys050J_UdXgs>Ulc`MBPV1Q?y(M(F
zTbSs-TN@#rz2FGr;awmY(_#5T1}(mD4yuWM!2WUF`Eqd@{N2t4qYI4UGG1?1<5R^z
z><&butv!S24Ykyuf_Y0WC{=q4bp$6W5m0jfd_rXX1Su#!UBCUkGC7V!zv|pkfX}n`
z|NK(K*Z5qJm-2MEo)cn-GM;ZKAbneHD8@3G!7W~CKqLY<dwv=VM5E5j;($b{=()<f
zX`2RBXePO7-v}Qi*i1@bIX9ZYb1W&%ohTqw6w3p8M8|~otq$uPnZZ3D;`y|;F!rBa
z068FMc%;ReszjBETU6rB7wrnA6BfE63mr1~v<J6X^cG>jM0&D)!gC$92%A&u-k?5K
zuyCcWt)5Gv-C;R!pJ@S1)lbUapl+Zu*?=;OGlaIFAg#kv<4kdnLnx#zQJh>7*oc=a
z8(#Cgjl*8@Ip`_Q>$!Q<fIT~L;C-L%c~s9WHlr-GS+HKgxQH!liJCr2?FZO8-=CLE
z=~tj{QO6Im?mvP9PgGw1)d<$Jm}v6Qj>biRww(XevCsBUU>o!(6^j7zpd`nEa)g-D
zfrkU(09zGMKoX&!;Tc{h6la5V&3&2<_8lx0$&QH@ug$iVY83<tA_@A}xt*<KdB9t;
z;?Me@<SsooHVfqD^{@vT_2U;Y;$^dThtTpzinQ6V9_6sX-z&Q}rj+k<-zgC(|32Sf
zu`@C9z@*#sXJO&_cTA1iW!JbqK7L`@d0k8Fx)SbKb0ciUw08P@;128MhXQq1YAt!W
zV@#hAH<hc-l)f&HU`~K9b!kx<E*Epw6h9rdKG_T2TkBC>&Q0{8A)NWNV!Foz2}eoS
z%VoKS;mh-Tzh-E(eEo7=(kuSq+)kzTL%H>~U`l~ntmBvgSBb3h^RlxkE^R=;U&++I
z)GN5+>mQVasE7ZEJlZWeWh=}d2-l!ZSLII&d)54jQMf(|HP<cLL9kx{oEHjigp3>#
z&COm<aaXs%*BpBFIN4{IN=_!FlqdrOr?`5`V41+M(_Ci;ZORM!kg6@5!Ugs(&4}Ym
zpBkwT!to;2|LZ9JA6xNRJ8raXP$%dfF1U^905Yeq5}EE(E>U>Jl?cApB9&yU<XmkK
zk>ymRKqP;Zo{Gi_p=;&S^NUfi1dK7!wBI?6SkN5G45yX}Lrck0U}qUgsBTau-w83y
zg-KXXQ^$9V1wJB?rNB{SJeQASI7^3xI=3NM|Mc4d2S{*EsTUzw`k_R61i#q_FEFj3
z?6c@>2?;uJco=$TXg$B=F5e|{5n=X>L#>SXxPuW7LMhm`tC|%_YrqG0Y6y0~ceU!T
zY8Yq!6R&w+NtL%pA6AR3Wg`O0{2os~!tOYXJ+Er<v-7c-Nw`f=itJX|R3;&Sn%d)Z
z0<{R9CxRqtq=UQnfXkbGMQ2|!Pq&dQ?u?DpAzN?TpNbtxM_3eDp@rF4>n7D%tq`1{
z`ZxeXg0uLo)ZHxsrTXcWd{(R?VjasQMvI9fGOg4W`<L?mIhSzmAh?!ujhlGw3cexO
z)eg>F6T?x2G7wxE<-Rl5H-T@2lA$&5hshctdYp05y7JoViRiZ!z`0;%$Em8|`#b?C
zC|0Ael^7;~_SZt!zbq6d2JMY6{7qzeus#)Ss{|spcynQ(+}dhqRXN^AWaNN+DYAK4
zUU?lREWxfGyoL0rWTx+=lQVsiyIvs5fCC<F09?fx-&&;8J{94@eFf=hRJxD8`jkV8
zBpAvgnkYy3=&zFw?g$y;0+>QxKD()p#}4@Vr}3FT(;oG1319N*TL>M<ST`~Uf+zw?
zAl)?2+t-m%4_Y3hTg?<Z2zcc3^rzN`2OKpOEbHPK4WxW<7jMrX%;8lR7g}>Su0G3d
zBHr7(v}X?nQed-8Km#qIfez6z2Zi{!jx=g;Xi4XYP4A8nSuQsS@Ca#{@D3-pT@_}3
z3I>&QDM;`p=R64LgE@OPZz_En$mdbLR@({4!zKYZc%RUALh!cJ910NG89NS6h*6+E
zjw~e}YEEP={%#K{Qg7(@>?;Q2PVL(ZEu^37IrT;Ecu^V;D|B9(B6X<qP90a>xNg<5
zMM>rAyW@-MlVkMHPd3ukmWzov6%Fl$iW|iRtcw<M%im{zqR{Zo^VNd=no)K}Rd{@J
zg}XBnN{YC*7uLgUx!_Ro6~CdZES9s>a_&JX&039B>%C&ELHUH!90^^(Zp|qVN0I^r
zWGchJPVPQk==&+zV+yQ>V#60|K59PP!htMERMm9jy<?pXO60#Xw6u+#QE;`W6>`y@
zh^f*`dI5>pIqaFI&|km$!BBdC+-2)r&7Tt~jxxe$_-6lpiU<iXef+9!${mhvMe+yC
zHEvsm<(VjZL|#$B&qbDPhSsiX#mmY=Zr6RRtf3$fMvDKNGeChN36h!i7j?*s8D~qw
zX3nsK+awj+^R`T9$VY#ObvnEb?nKtN?TUjTbb?}FLX*7*r{f5Rcj_2UtFRL^k^VDS
zqHPqkn~c0gz~iaXJt2#>CcIWS^FM=u2>uNycELK@i0rtItb^Np1^_j~;{?tGc2M&J
z@eVU{B&RGQT^iW`n?V39SDlT&hCn|3eEST`(=98!@?Z|jGm`~?*q7-1JHe9Y{lFh!
zH3INuTW?T)Oy+GDPEyNe3Yd4-nrNNAXisjv_fEb(z}@6>*_3j;-4*8*)yn=sT^A8{
zmv7-R&rI`ml4tBE>@JBdFPGnIHKlWJYdo4g6CYdXw3fhzt}!XZ^5iNqTLUJF<TQ@Z
zIA-oVG!s~@1ddPsNpl6TM}j%caU_XMZ`rF2=B#i4N*&4ST!0V^D9TM1Nq1RTih~1F
zv|@lr8?hoXTkhLyQfJ@KGYf-+S-OBQ$-PID=25Ib+6a-+;;gaJ=}d(I4rtJh&EEah
zAREM){+<adw9}c(33-f7KBdxoMq%`zz}N3MyZJ2y8)WD@#?{LA;s}w>p|>Nh4gq;n
zZw;VgzKAz0-b_}X0pAx@>O%F3qa?K(XLjs6Y{%dHMU`P6TGr%sY#5O#<$MEwv4TD8
zDmRm!I+RAx4>Sp8&s(YA(N^YsrM$mU&c=A%_N{2EUcFTbf^}ySF!6MV3`KSwCA`iH
zWzx3EP(C=eG<H6zbg|xtRDE*TLEM)0R6YAq+HkHotP=!eGmN8Jq|Vag{D>?h;Vl;h
zF#?lWr@z)1iov<q-_&V^f>`n;l&GGOD27PIM;;D|hd`jt6krl8w_+;i1XMw%G&^{G
zp@(+>Gq(QbfW`Ysiw}n^!OV5Y8mt5i-sgl;0c<-ufH?Tmpp#kyuaj8|d(}<M{L6rr
z$<S9GyGL(Kz5d$Kj+GB|%jtEm3-jXUaEW}Pg(u7k)^tB9`h97Pe{;#@i{8k!shbNA
zl{~BumvdZ^y+fmNn19+shQGL;rG2e^^G(DAd|=m|W3(=Op1IwGmRGHQ2=7MTaN6v|
zecfAJ)%)?##QRa69EaRk>*6Y$6z##sK1YHAiS=B?rBnrPLghvDlzC13(ZwtclS}Ei
zk+vc~{*avy6?B|6*$mB$a$>74ZAF&<G~)`A1n7m9ke!<QFkyy@P{|b{GuL7FClR0`
z$&}D&|Ne31p|twU&1MdR*%6kI%@j^9Kb-j*hNDGqVC4FC*TYu=9Xb}M_hsRUX3?|u
zp@OUQmtdoK&@pqbl=Xk>JHV|pU)E?7FCG;jKc!gL01SoVI{^#ygTfF1x((PDXHw~{
z>Oh^=Bd4(sY_Qcq-v-TfcHqN)=`F!rY7*?i18I^f(&}LBh>R>H4bt&Lmz*IiB=Bip
zx`Ic+)Xf1{eTMBl**J=~GtU@|hv;Zor>-WEo!3G~r*55&6&r?DP#`=(kN;RFVSoaT
znGXDgh5@IbPEu#+NXOM#Tepz3e>Hml{Eo9ipo_r!7ev@0U75=6^NTmk-w@Zk3$F&$
zb>Od?{JNCdwhH?lN=zN%;efpR@Jv=zS;*Qw=z+nH$KguZ&NR2>)|RW<^x`?+oKR<G
z&)8q*Wbo}*GM1qInqyj72P<E6c;q%+`}A64ogRf*j;wKpS)=F2nj>4Cw#?(iKxh&u
zN}Gsah)lg|V!CihhzQyg+$^xJ*M<Uq3Pj2jvyYvch@K(Uf)ii}{ElJ-v6d1mpeFtH
zNr3e*Uh*|o1~i^lN?hJSa7Q98WH&G6yj4CQfMDrX(p)I-PrY){8ez_?=j_Z?YGkga
ze5sl_YWG)TtNNQzAeJwBQ?dKa<tZ!_pEyF=n#1zz)tN8N?;`l6J}kLJ+9%|pfudlS
zwy%nJ3;&@3dB=~GqN%*|u?RY`HY03(|5(vCr2i<@<;N?N9!HtX=OZ7{cXh*YzS2VQ
zMph#Gy~U_l8rwbi(ont#+WMUPP4zF(-$7|*uDJ-%b~tC(FI(=SK?yHdz$#rG50K1J
zj)<hM8p2A7g;p8xTxZlZJLc}5zYzzC(M^w+&9g4$B_M3?7SSNis%OZG*sn7zATH}u
znv1GFRRiMYYbb)i#eP?$fUOR!(ywVyfLK7KjjC_bAR75eP<{j}!5e(#U`$wvKX!3P
z2*H~QmedlNDwX8H=4_LdN+w0@HVpEb<YS=-QJQJ91rDq^0S;oWxrNg^32*$<Mtu1t
z>dd1Wo6euD+;gO#mp%$fmbGTxxsU4ZNvqttH8G~7(fV;m&R<F4c5w%P(V0-<YaH&%
zYeJtHZ|-_|$~45S5(d$<cecu5BNjQI_gi>e5K?pv%*KRY8g6L5D+P@GawTfZ!+f(f
zBYctxHvWkzE{DP)99KS+XA`T7u0##mgJ49ouelo7T*L<6XbNylc9P1j!58(0Lqlxi
zhGE$)Xvuu)4-xsRJ$9y($gy|y$8D#<wX;0P=MspK7Qbg2_nk%NnG!7aIL*d0FQT_C
z%;KDRKs54yd@Yb9K}69_w^%>_qqAu`GT73lLxhU|`0X+$rQv<n)(m#MP0{;`>bTd{
zeh3aNVJI|{L~?X4jztLaxHNX$5ywTsAYf@K9Z!oi+82)VgF-W)#W*%=fB!#b3BJLp
zP;Ihwb{Yuv)q@U*-{{~rMPLt)>+Fkx`XZc#Q5IliA^2aV3pZQ@u${5{gZt{hhXCGJ
z2)#^oKSENU=JQvQ$xhGOoDyi+UW;XKCU*1eXD{8|D6*65Vp?1(U|n=z#&dNH$XZGn
z;;)CCyz2^~yLL6w@1m@3)7!8otjDun+v-j*d5lPoV9j?r$)f_^Jx;GY&&M}f-M5v`
zSheK1nvJ@<JifTU<Z9%A>MEp%-hA$_f9fO>^VB+Yh?mq_R15=4s?s<~aNr#!2k<m4
zVY(0J0P7Qh>{|a2I+=kP43?{7OU;A#16Iy#+`r_7chTRD^y>TT`Kp$1fT}htucK-J
zZlG8UXF?zduuk`upVSc)IG|Pz6WUlb5F<A_0JE5vb%qU6+-QVb{z*)v#PeC=V%O7Q
zc%P@thb-i_ZZnmn&JYNa>J>NVPiH)ZB*1v1E|~x*nE~f}9O)0%6zLx(=U61cA&OU*
z^I|PNKEMI^+*s6r%^G;7HgXTN<wnOpo$w^7i^W{0xJe^H#Fjbc!U<UdgY6zUb^fg6
z#e_Ylj_QNGqHpo&DsY!E2IEegcVT_w`CQkUckS*x^=G#B6*QOgW!sNYoY~r6Fk45}
z$E^BNnWS_T<ff%YM>k;*k22AP<^i>*MArT+N%iXBE52=t0SIuwkB{<CR7^7*<*&3D
z2SNP#HYSEqJS(NGoM-OSoFpd_Ryj`b(~!sKd{ejb;x10N?WtFCL$lqR1UCo}fykS4
z1Yfd@UaZp}gp{6G^-8nyEwOkaqff_56pZdn{ze+gqWNu729e0f{O2k8-`YRZ5k?2L
zvN=lCl(JjHAfKGyfhWHbJAEu4-jV(m`242_qo|znV58rx=adH<;3|aXhG*+xLV!B7
zZHV`$nd4-)7#04-0k4h7^a?#W-*~N}r5cY~o+0*lE#cHI2WxYl5zkVMO2xSW((`@J
z3+dY-U24r>`^By1yMpFhDA);v#F)~S^YokJ21fkVj*Ik0UY(-QE0P9syP|JV)d({j
z6_7W{a-k%Y7DZDt`)#g%;Y2{nI6?jKj%&QUj4NGN_U4p@>4o=+Epk;+{?BZ0Q^??V
z^M9||C@SMy9J>oXr?OTBF+>1qFyJpVRhnD~eg;c&d{aRjX$Wa4zz$X!;KJCLZf}wt
z{p*Yo-~cD00`Gzyb<*lqID{xh`)zcNvYn!B2nLO@W|9`7Ou{)#Fq*d6x+0)82^Xo3
zV7CEbo|%cY4}CgVOlJr#KRW=f?rNs>Nx<qqi#*8u<pZTba3O-E0c=rD1STWs;|Qn1
zWVY$fX`B3aV%&Q)dpA|}ZnrIeulqc|-wt|Be@)#oJYKUj-m$iS%hEKmSU9E_&`)W;
zJE??hf15hK<@q{+Iv3Sl86(QUGN{-0Mr>}KBlzu>wLl7+KS$b@MF>x^BjnFlXR*c*
z86rz~iLCw?Rjd%N^QVC=*VXV!eP>7+$PpcagiBhds(%|jB(jh5cqAq&&=H^F@!a+u
z5N7z1Z8aPi^7b!IZ$Yrj+;Z%1G0rG=WGJZSELtRptvyBnsSW7I^rdeevjI90A70cn
z*Q@^@s@^;v%KeQSzh|)xV;!<*Z)Qk_Arg&!?86{QC0k}fmQu8wL8NKOo`y7*!PrU>
z+K$Q=$u=aZWGmV&Ny|CUXFA{K^?QDQ%yeFHdd+fQ_vgCa*Lz_d<A^p&tMz>4HfxlL
zw7NhyG9QSvF4ykP`GCBZMDhY6{(d7GBej~?2L&XWcZ2ltyjFJhwmuj~Oo^l+<zkcl
z5({fbcjL*utheoz@YWyqUcWLm<#hV5b3%>H#AV#opmmPJaf^hW2BT3(M1(j~8Ha@p
zlYQIkhG*Pj$>_bYI!3}8&4=3lQRWg}eTWCnaQg@%lRmz$rF*@{{EsVgBzLQ{SBCu3
z=0dL08}xm@(pU;16BNP%bk(%~*Xt=Y`dUez(5^%F=l{B!G>QlozIj2)?C0fsfnU8{
zYsl|!nUCvQ<p-xk`sQ4a^`9I4_)uY^^$bR0o7+~U<x#~?N#pW8uS_-UykC?k3;VV0
zx<~AwWr+4x-AJM!?SlpDuk1K1{KSjc^=k0>u#APmtY}R)L({fkU|8OtrV#-|<!&R}
zU68gQZ#Nulz`%GX!c1b4c2apPFm`V>j1Z{CZNqaJiaqkaS&KLrcG2A`Lh+3YSkHq=
z7mC{}AFSAF^m3%!F~`s(2&#C8%`&Rq+V5x3lz0F2<)<R_%9UJ?9aNW@$kXjpX6pB9
zPf(4e=x#EJ&j`P9{e7u>6n3(hM?=;R5z}psTKihv^ff6Myyq>a>!>tI`+1VmeNN*+
z?XwD5Tv@c#LEaOclVklMDbH8SQgr2&|Ap#_jI0k~)ZcG>j+tD(U(jInWFh4GMy6%L
zZsFUkjCUMCKHeW`C$M$Izz!cXWmkrX=@pZ@fO5V(S=3z<sQPdWKqiShR+r5dcAu<>
zC5PGr&wpVV_ex6h8nNSwB%U;8kn<dn47gpI>DU)!q`td8sk@-6pb8m@SE#jBav-1p
zeO^PW=rpjYa0M)RQI1^_>v*L|DoK0-mE$N3Ft8nhd!{b9%~^U=1b2w4H6k>PKY227
zOH7S*^+;aAZspyFY(fsIwIMSHmm>I|tOuhLelJY#w%VvXdWvB2A@}vu+xs+apYzrh
zn(LY{twg^|DMp7&rnrgKPCIfd4ks}mwBKaE%$lDaZ>_dw^;PaDzn@rs3%q@xk`i-(
zgYFis$q{C=h7e_+sT2a1BRUs}I;ZiC2!Tzh>eZxck7Q7gS_FJje!f^mjsaqBPQ4ZA
zvQKn;qS=0Af&KqQ821~d{v61ckaZwjZP4m^`Z)k9xI~Xa<oP~NDC3CUhWR1jbstHR
z(P6?@B(4(FZjuyKQ%Gt(ru!#8Bv2=YgvPpUexV)C6lH8qB86=i=6dBLMv4$D$NLLq
z9Fc9RY(l9!WG=z8F3>7rpdlB>1MhGmueB!`ifwSV$^tRnqqE+1gxhQ(AvMm>icP1p
zz<`N3rdK>(#5lH4FX=;p&8L>6dOJ2M$}eOOoY=N2#_Iu+<%_P4eek)<XxpOw=}Or8
zEp}dKu0pFAV+Tc>O>Na!#|)F?9L2w0solZ87g_&Ww>;{f)dtez&KT0`ZJ$Fw4UC%V
zgf>0qh-Yidnl(DtElkEM>=O+L)^}BLdV%?ws52j;pSwebB(sm_zu|}ck+pU{cAZDa
z=3hTvDdhY#)S~KIK=5X=17!bbp%Qd()z<=`lkl^~qy5oJKn$7Ds#UP*h!?nvgGIC?
zFlc}Se%zvaI7e97C0Js`DalS1_mH8A2l7cwc#%jzpx{zc;ts%Nv)NF(-$Qn?1(8vt
z>;{fg%-JCVC`%3XYT-uv6K|PM!*lc`>jpOmBm-Zk*1SEF^HbZM;_e(s|L9YdHWMk+
z6XLNWqVs1~Yp_aQ3jW8hCyL(LS_`u`zjEzXf{u0t`+r@};7K3DFg0%f3tf!E=l%=T
znR-nW>PyJi5wagmQ65O@r?QzJVZOxd=lgfvxmCZn1W1V8sKE}&UaL7gH0$;jv9<c2
zE5+$QP~)Ww@g_Po4<0mD28%T4f@uIgA~5*^1<J+1tb{8=$bbU`3J9cHi2}+Yh_3;@
zd%GL<rZWr@s>~~mwQy~KeLE^zN!a0?>L!xTlV6l2tCuKecL7X6GpjsX%n>JHi#Mxc
zq%8?I$QVT9Vo7O$nGNZ)4C(i2j3dOd!C2BcI8d>miIdQ&kS(UVkgi$;1bRx4^pOtM
z`mrR@6IEHwu(E?;Z-0G$C^g=yabAm31Ied3@0)m464~f)An$BC_{#0saq|PW%`v=&
z=p7ebrK4b9tP4h;>-C2ou<_j;ir-skQyqTYOjVM0O_tmigf=2vYmR@Cea1JdgQuB}
zK1wYZ9`H5`_6kr+134O?H~MpME1$nT1n-O)GmB8XT+YzQE<8mgcR}V`wOMA95ZnBV
zlROV4#FCoP2Ez09psTm{UwHQ3%RVOV<Pf5Fs&l=DYt9iZ-yi?p4+AHCa)-8do-Cpg
zS{oqZ9BZvc2PpJ1)~XKye=wbk29-Zh<YJ^$5_J7qq0$FBA(`-FhOy0lSHkHe?<w@R
z>C%(l2yH~dC|}h#yO^rDP$WGSQzK-z2AJlfS?)l$o6);giO@ko%@y+&@4a|RN4Ss!
zR>clCQ{?2hR%zTH(nmSTJY9zRZq)^{PH$1!7r2lUtAx5kIar-z^njN)ud=t*Mc9>f
zXwh%0oUz<HoM=-bKBHrDyX4w_G2i^TQxA*!3snNX>Y%S@S+{J@u-(ui*}=^@P6mk<
z)mn2aBMd!?q{-+=eSQ5j*4%4?VjQZ5pTgx0K!+#Yg>ETjMB`mE<AnAx(v=q7^Fs7y
zz0c3-nQEwyiF`|@wqsJT_($P?=h_q+*sPhl?0!Wv(&m-FaVKNu|AqW%n;IzRhDj3I
z3k%sSCX~Ta_FcllJCx*APGthX^1*Mk;fEpkoOc~n`ol1(t8zd?CQHbEfFvqO>zoFG
z2B{vHbaQGNgyktPbbz((4<IcO+6kOoWSL0J=m<mq>4`2g!t#jXr`72=;6G0WqJH2_
zSgK{Ra<uy1!)xLLhjioZ`<7h>*NYJ&6|V;|^@wqA5v_WgW50|U{hysXk8e2{dhBKN
zqb=XvjbBb$>}-b#R~$7Vf5~r(U-8q;IwfJw3(;61hlB@q_XD+Yna}asjH228n(FQ@
z)vGR%KxGifWFEY2@WY4In7Yk-B6FnoBq&-Y;N~<C)4QkOF24_({rnvtGSXb#Uh?EF
zi0fT79S;@;O@%-JA%tOo*)<=8Q{x?hahQO51IRE@Mkpc35KDyW#v5#tD5A|fr~%o%
zdmGFc2rr{9h>*G6oTFIAd-G1`1U=hS^CH?)n9GWx@8hQ{Re)Z1%EyhbDcxRk(X?DT
z_!35xf+u0&7EguoZ3{(`9o#Osgx4wuxrG@tH-zjn!htDuD?|h&9KrDhsR5c+dVwgQ
z7>~E`HB#hlRhTzSFC6@})qh9S&11wI`KPtH4<E~3-@;<^qO^4H$mMBo9%(e9`ClTw
z>^)n&H}<WFY0HuNN7oIQyB7!K&Y3+B<F?0|>f7!bPxco7NgWBRPt=L5JTEl=_pfiB
zPcODkX5HNO)-@;VUHZQLBStY@gPg{M85<%1N2cPLlO+LD4=eK~vRLC)HbANoo-FnX
z&=UdHjNi^ddMI-T5E4V4fRWd}%M>>^3itoY|F-Cnxt(C;Nr&3uSZ(b^s;5z!>({m~
zDcMR3b|r+T?&L->NP-F{b>gfB2+v~S3<wh0>~b^``Dh|ITY)JCER1S6L8u_x3tk>Z
zXPRMq?tiHSn38SG*`!hbsGf0#4!UcL?#gGfcx<1sW1^|1E=PNWfz4QV+pCHy^kM}U
z$se+a^*RRTIS>4?j1r)H{S|b`n*Z|d*8#{QDfSg7{DC<~c=KffS)|vAi`dc~5<rWp
z7yU?*4O~X1mc6GOvcicTRyDQva5s{7mJi;JJXixCF`Uz$yhiK05O=2ozx$r3;(!ON
zFw9&nEnG8b>KF#bi{SLMKhyY;P<uPG|1rw8Q(7|q6~n|#dV;I76>{Ee8Yb$sqDnFL
zY{9;f2EnO#72XaZR#H=!s0fi#Gtl+p?N45w7qL)m-TubrGx!$qlvka9#|=yI*{y0H
zK0)U^F8uR-%f@h!c0qJiz6ir{umhHm>ch<}k83n~l!Et1^R#zq^5FUB-V!g*`HE&M
z!#K!;?u#Uu)&No>M?6QO!OW+ZgL-n-8}$AU9o1LMJrY6t9JxxI=63`HK!^E3-mo`x
zoYWH`2o8Xz8Ne8f1=}f*7S&}W0azLx!z3ZiN=*}YJqDh9+{&dwoYXON7f@GeMCfuw
z0#0K4S2qK9XR=K$?Ey_9R*|Ra_G@6sTk$g?rxotF!#Bh5*JJr`O5!xX<)|3@T4wP%
zE%w!=r<UWwDP`z<;`bAlr}2*G_jr5`*<*a@Y2-IkFFARO;zcT~AmqJI=Z>-EyGKRj
zP0$Z6O>#eZPgKaA>Y1!+imF{m*YT(}uiCuUo&C@wP>Md#zESlTm*yF7aoI{aMZdoD
zf2wPNvoN+2=7LM!btZzR2ufcUve->*P#1!1PgGVxV5!rr3{tkhA-z>HKa(7~)r|u$
z;=|;q5z@gWLU8rwwE@kG!h%B@?RvX_vE6%9`6K0kitwsznLKA7&4(<C=1eITE`ITz
zmj<Nz52hn9xG-4=$ZQDQ%(!@xEZDR<KxSa9^&+7`(_U$Xg*8_IJ&Q<^+qHuvIqieA
z4h9;b_&h#~AGzxl5h|17RSQ{HF)-V#+$|&2;v%HMSPaYs0{-;Di(ph0W+>Axk|b1>
z&nyWmRc<AV7B_%hIJtc%s`>9bfs&OJ<ggJdbS!aUMaSUfCa#8-$DN6|Bvc_GeZ5L3
z-|Nc2cj|#M?FJ2v@E#BT^=nT}_(MW3zXwWIuhvY3hU@|e+^JgU8b!HFM;<ctw;Q;d
zv;a1nQFX5|X9&7m!xKeaCPKabm*YR~p0O;r*QJNtf|6RGTm5{V@lFV$h7{@*yNUGt
zbN%#T=a46%Nkc+TI~Ph)Ey<sO3;A|<-QR7$B4C7NmUrz}fTJYZTH*DUB4A&QLJ1br
zt2JMIub+%12Rp9=!vI@;h6zcb+~97OLx5}yYZstjuP<*fK3X}1_GljxX+fCP$IsQM
zzHJCy68|ULc69)bx9D*Z++Dw>W|cYdc=p*_KqT*LfE((T?$=}qG9-O+_6+F%^TR!r
zUotuoHgWW)j#4R2^#``gs^NpkRA$pZqm(&K-=G7xQyTOtc7}*)Qb1Bb*3$Bx+5UE2
z4~Ou5t?Y-kah?Yn5c>S4-;c;B<I^kdRt}W&(hK=u`c!6U({08XhF$AIU@C;p=gEs}
za72rOhIR}I6?KGh3D=dVZo+jEZ`z2zzLF$%4hd<<XAS51t`rVH<Z=Kxy&(~yr6kZ0
zb(y1rFvmCuMJ}9y0_n%Jnz<+rTiDO@@=_8=cY^e2m0%mvwLO=<hn!$ZZSxv%wzGLl
z{1O*y2lt=ZHhx7r1a(hHGn+6)Kk54t`>paDl73m@Ac>cZ6gkkj+mdLf`XOevitan(
z_G1LON3|tvO7D54_3Dh7!=~5f3*4<4ies|6Ilo_=rcY}W9Pwgaj`qgh_>l8^z?p8M
zLir|Za80r1=FicqF62l`+;r(!UQDb(DxW<fXp;9f$d8BO0<&=cML7rk7r7Mk-c4)~
zoVQ?pd;oBMLXT%eTTPpM*acN)qob55jggNE+A7SDR;Tl=YTz@gmC^a>AmI%x*+xm}
zMpeOpCw(FSIuQ_qVFU()Ue|=wxz&9esryou!XMgNQBgY>Dln$zW?1;$fx@cvnIcKF
z3mT3fdtl+#0RBy8Lb0G-xBJsQ?@t9Sw2Od-Ru8%t5=W#*rEkJPubDG){eD<~l-E-s
zdv5ToN`9zJ-K{4d2Qfcg6AeQ<6!$Oi?&vq^CLC7Nvbw5g)8)l}7;$rX?xbJr)p%#>
zhusz;NM+cd3%@sD$B-t_f(xj2NH8fVyFezOb_swr?1*dU1b+flJFtI*N)3dzdg~w8
zW5~=MGFL+3&Xybnr0-na$B*a6GDbyqUFL{1QH)Q(>s(ECp^TatXRlNqTCOwe%*z3!
zs3Xwv0Ta#h@FJPgz0qg9evIlNaaMOILf(~IjhF!aE}`D;1gLP%*9X;>p&{2fNN*om
z{Kuxg#*<7p)Vh1C_{3Vx`iIOWK8p&!2kwjJ2LtePJr6c+4++uI_r`p7%lrmbrJr78
z)XcAbB#B)F$vXg@?MU0HP6Ml;5R&w#w^=P_pN$Z|>>3ctQ(xaxa+pZN$x|Urs72)<
z;F6w&$yw}#!!XeEeuJ>Tw0$wZo)aaWkLq|=#Hxk^9kXuEi)uPD{FbY7oFNuYGCmVR
zefZU(uT$;bUf|MZP(QLu>ERDaA-9z5$M8lxKjDitERVLXO3@5Z9J^k*Io4&XeEwCi
z)}!WH<$j3%rCfo>6wF#Vxd_{NiqMsz%AN(TwpP(uYAW**kmQe7A*0;YXs|dChJ=C6
zh8HaYlg>as5k}ma>eoYrsb<vU+g~cJVPIND2fHkZ9qHkZCvg|ZVt%FIu07EOh(SIV
zNK(RuMdGgD`V%8Ua_)NBqTQS;_XD6%t*QY?bB65>83E;)TLk59=xBZt|CTr0kw$16
zQ7nt?kS7v0MGNV5d*WmayWYR5ZC|Z~V;fY=Zy7F3voAN5@3x%(hDm(2H+JMklS4^5
zHS5>tv+wS%%DRW}HnqoR(QT4wnW@NY&>QyEZPL>T@}U>PjneC#`c5lzGo@cjkZDoh
zL0qQi!L=<lu>L$-VcgH+d+&|EM73Mbnl@PVmUbULR@qnqC!GnTigH3C00<1aVFAjG
zU#mi{`mg|3hkQQIPdkfyeXCZGPnu0Ya>4usQ<n&Od8ePYaBlOTXbUM*x*Tbf92X7F
zN-s9j44`f@Y+3*WEpWZaA|JsB+*V1HI0%T=uryv;jj{^c#O)y4gp~HuaS4f~CoCat
zU((`}SyX8bV)BxDK2%lI%|L<y%<T`*Wk-_G0zq9Pfb7*a09I~~#u>2P^#=cp9RGVt
zWQCV^qupzF_>@tP>>CVfa4dTH%%IZj_oi1(={NY3(Y;T6bnhSOyM0(uUNcZKcDL_4
zcT?W6B=3AbRc7Zzo%I#W-H%H3cN*~I<z8cbvd4?dPdAx7cezrJD{r-U2;_tTpi1wk
zMMzIiuU4dvFyx5&;FZ8QB2;(C{)3bUzfTyeRH42K-HCsdnGa1%64)mY?ZkAI_J=Hf
z&cmmUj!>1>2xmV~b!j^7D$vk(c|Nx)zh~~B&u`s?4V}9p9xQJ`u%X!462uzuVOMc5
zOBuS_FH^PmvvPyUXLAssu{+^$$2(lnLbmML-!F|+C8)4_lIU%8Nbzg()Y!xchsutI
z_!Z~xKC)EUm6`$2%4}JrKGKT?D@es(UGDaeWGTq$qt^NB`}#m*i|*-X+QtC1;QatD
zl9fP|Xc*!%W*0D`3Db#H01ygcEL3`tAz3jIuG?m5vSq9rGS_ZvTaRyTs%1L|t>TBT
zF2!@F+Jj-!6s<N3PSZz{lv~b#yK?A|P;ZAkHV7vX*JcP=)PZ@7Dt;p`B|>TcA)UFg
zz2{zPU4NeiLS<Bc(-eQ@!zgRjoZc$U`ZnU~P6}dCbcUfxk9bc~sMv`r5-bWcUgGO3
z&`e=1RpnNdPzt=&C5@LXhocs7S5C|}XlPqkl{CZftJRXt&`g9NWd+C-a1V>b*#H&=
z_ffNKG@WM>_7(UG*c&zf8iEsQ9&<!5^LlZFa_<vmhmE^uw|e7e*(yo~UN?bv+-><F
z&u^KoAr4~2@oah{ErV@T8%IAeXYBq#$WB>z8^bf6Vj`F4ZG1xYqw3h4Wr4dr({AF^
zQ%mKPcb|H#+RG67-Da90qb=I1ojv0<b`9STH271`9%*n@*1WsRd>wQB>Pyzpx6R&x
zv!<xiv(5<Bzs7=V=weU*h3F%tbFLw{<SydwJsGnPN8{N@Fn|gsQo)l2PzyCTr&~}>
zjUCwT$p^K%Jb283>pBQbbGkOxbiaRcRS~8ty*7tAW|`sQ0^CGwBj7W;FkBwSzPmy%
z8LbD%XMlB*!gV28r;<9iU>-)m&FBZ|!HNNFcZ3`O=fI;VmDmnWRY1g-6V#>mY1{?}
zsaLUu4HXi)EI~so*ocUc!JQ0f(qb^m5Uq-W$Y%wO)f|~iG0K^SeRQtNv{Vr0nP|dx
zm#zhDdQLd5a`8xmOUho%{OLz$6r0|cjOPbDc7Smf-$mqGoKuR+tGcu4tNp{OebEzZ
z61z|Xqv6bQy?y(K)=r(l{0n__HQqH2tP-uCI>c9O^JR|&CaT^~V)i#ovja+2X!qdE
zatCmb7DdR2>bj!*=QO6-!emuBW<rFFk_GV%2}L{0%8r1`jprukVyz3XUay(ZzPk4=
z+9LNfRIfKC@k;|FIcM}3s<l@#>O?QAQV1?`9U8vQ{kWY&TE%mv?#4w@tW#r$5)n!q
z5lbYrZ7a0{Zn!Wv1k*0;C)B9u#sUzf3MUgsgXteSBJWdTjbq3aKK|%h@|FqhkQi$#
zCh)#-MD>1l{@`y}0Abvk`y=4~U;f7ZgFj_<Np}0E*^uuPp^2l<!>?#%XY?N(*t1U$
z8;v~fka6U;lj;u&VgJL|Ph%Bkm@&o??KjC~yIF!6a`XH`*%Dtg@y&^vo2fuFo7s&<
z#YDi8R=I;MkcIytko^jvYDJM;i%J7CYqVx&=Jm5u-IqTB;>U8+v?_|lM(SE5?gR$h
z$sV$491(-u-U$-@gTf}0f1xCXVNX!hN}I$?zBkOdrfA5x`7N`?9nzdb9sWHD)jOGL
z#o#i$O~~dP5pRG|i?0b##b6UEjw2GzM2c{|fXfFb80M!%hjNOU8AF<2Zxe-J2RSBI
zZwau|<A5=|8mNladC+5y2tY#wX}ivD2x2}XGt>L&+bv+`;QnluBgr(?ysUvS!Wg`c
z2k<Hf&5vDBEX5zYQ5$;kWRN7zOz2%s@PVXjJ1;*Tu$>CCni@@2AB(iV(A(uHiIJmp
zr#NoVF3)7kk~~%d6yBPvugT(M?M+O4<jjt%^Rs&u7r^ZxSX%VGj|S^kA3o-ct<{qM
zu>8AusTPC3miS&nAum3$Gp*M&RAYN<TW1MU)gY_<d}Qb$RppCE2D*+FW#uJ%0S4gz
z$`iZ`@F@g!^VjSay^T9#$IiN>cNF+tVdenu49wYZhJOF%7B4owtx4WLXv8l?wlp)9
zq~Pa)a~)9}Ssmd+XF&pi5ZdD<Nz~mcAX`>=2IxYS1BwPo8P9oVM`0#^GL8W<RmgyF
zFNGkJ26G9D=G0L9dy+&j4SbSf=Uk3Kpdi5jph<j05@W^yYhJ+uKZF_q>vuejg_FO(
z_gH$)3E7kGWL|A>9?*OG)MqXMHHY4$1x8fRf`T)(8Hz?Um}Yc45k(t&7R)Olpzu!m
zb&uxyHhCl&)%*i8>2d^u<hI9LaZ=B<{pelSys@xqXqlGQ!{Z>0)%jv%KrKRaKJ|or
z&z1&)@7maPiB_pFpYbaKzKLSVJSZ~4P*bP<?2dqk?vH1E^^=hlUV{AGN=oy4ydERC
z8rmm%6+h8rjj^yxk}?Tdo=KG+9yCDCN$wy*CATnsEVGe~JrktLj_XG)pJyAFPcIwu
zQy;ZH|H<4n?X#-)xe0%Q{^9pR=3mM;H>VenSE>hZFb&Uz4~P;km=k_-29~DA@}yNY
z5R+B&x*_`B$+MlqDJa$7xae!a_eOWq7s5Blt49KY{BA^LcqPB8mXHeMKRc^?Yicfq
zmOSV(#g=hiNJh=V>)Hn(C|xmHD#T420|6?z&;u*%2IE}>ERO#vCPI;!@RpsC>kbD{
zKdxhjVX*gwYfKDhcD#9dG%Ej8*TP@qZqly^rO=1jHJb8-iBh^RkZ{7aTNg=e4caS5
zgAdNccZ<-snqn>jJ7v0?1J896T43nQ6%JUr6QlsHD)_R`FJ5^E_aE@5g8d0aQ^%Y&
zrVacjJgH3KT?t7;V0gK)xv(gMPVN%m6kwilFjB|+c4=%L&szj6_8o%rRScjv8keu9
z*L526VWFis763R*jfRlfnuivtQNQO8ZQ(^E22Ok3D4&<g+R1LN&)IfQkEl%udpTL>
zy6!Hj6nKn>@okb4fph&fNHrm1lKv>$n2AbcPRyB>vctdiEimV`ciuTouW0Ra;+3Iu
zoO7C)H+qb!M~X7OrXzj&;&v&!MSuhPqy0XGyor#BUsV<Rn;ChV@7A`4D)z3{X3h17
z$ec%~dfF5YBq&N`&u1F{XQm~9ZvawD5$-8q`<|P;(QaR-Zb|Ln!6GWb`x19{Lmp%|
z7$jnblvaAO3IbZ})1Xn<o3F0-lhc;~RVzFjEQm*2De!2n6j|gm_caN<RiNjH2K(aw
zRkb(}>rnvZ73P5;iUtDa-$5=*fZGfK+(saXG6R_L3E*!7XhlHIhbU}#vBx!L3SGBE
zbMf1)>O1j1BP7K=<3+|2y_vpzze{N$H)&sHLuAy8#n>W(Mt{<nc!TL1o!Y{HjKO{G
zx@aP<sHaSr!8sa?riYq~Z0^eOcnsFOf^E_;U^Ojp416;<p-0<^P#mp;0&oC|#unyb
z3Ieupg)IzlW@zLz4Wzo~ORNAUZ(s?2EG=!&LnNLg*8{JeTR!dXSHjA^dIhp=N?qCR
zl5RJMf#%<cVfaGDa0c;U_i*Xh!J<~^XeBbL0E6qB`?g3zFgCraI4)We=jj5KZQFb@
zm#$H{d;-%UqQ}iL%S*$+q806_nmeaM$6sBMX+QmO9rkNLPk6a;IQa!LcC(>7`4MrX
zofCBCqhqJme1`gJN1fHfDaVk^ILoF*e@C;?BYu8L=QD{~3l*q0+qVAvJuMZ${N$6J
z9vD;=OH%Yc6bl5oispcqv$uuBhmKV!Q9<L%p&-Vq4%{J0Zpq3oj+e3M!7v(9VBe|)
z`!>~D|7qWhH{$*Aq)pxnM7JHEv^lCM$G@+1k40U0XI?(rHc0$SPL|eh(*8nuNcc*@
z*%hafi=8u7Cl6(0?>1kp)X#3Q|Ai7h^b{D>F;L79Qih2~0AK=Af$D;xoy26QpA)Xv
z<cQf+1Iqe8pYXumyGxD6hd84(P?hrlL581Ugp)X)6a$2cn2sE@3l0MPHyA^=f&C?d
zCN{&+4^&Sn7JUFfodITYu^vMb*U^C3YL{k9<Gf{e6`uD}>xF!i;oa@oz|&yAud1%8
z@dqKuzj&ThUOTwVQ5z7C36;sBNBR;ZwZ&YOdgZl6W)6D%bsq=Y85E@4DmK|(Kj*De
zaMtT&Q?0B8I*YF(vHK^7eZ6A+@M_fq>9<mTA#ddkBGT7xba?l?ShQ1I|5=q|p_MJW
z-OH!3Z3<u3E<4ei9k%#eyYZ1M!rAT5^#`tZKyY-yA`<tw@09V!2&L)0Aql^6{0g32
z?|MV8u&;28Bbe?W2cnR3heiEcMf!-A3X|ZYNdjM24d)<e8XX|u?4fXp$iU$kMS@Wf
zJivZ~832G*Mod-HM>iq$)el3OCZM}C=7P9(NiMkQF{Igc;Fqv4CxA$!37LXI6K*Vy
zxVX7kPiiNs<*yUBhCZBAc~NEQC%ax9685Qk!RKH}?V7>iguVeSw^?>=)+CC(bigaw
zWjGy7dQ)Y`t8^-`Mu)X#2Hr#(mn4l3U0Dq`Ip;$Xmc$*V%7DD92!UL{Hr97yJ`)U4
z$98hy3`LrNAB;j`#&=R-)eTCOMu5Hu6h{;o^<gS4VUeWJ@c3m(P(qzM)?p*d^B`F?
zz%N_22ZfhEVA}RZ<h>i*4>7-g%r{*U3p;+QttTK?9?W{TvW%cQawjXsY7sJio9s@3
zbO12;+JJu3%yaV<$VTJALb*i<*j^|^HMnKw##Hd?zI4|VQ3&tJp>*z5tBHSXZ(lK%
z!s~hy>=5?X4m79<k$YVY6Q`+e7cS<{4GA4DjQmwJI8tu3v*hXAztBxLk&O_<jVylp
znBf~j+w_pIE#by(vl~~2)rC*`>d(%*g;0S8TPs5;l6?u}k4MP(KBZj+QP<uak*rl1
z<^nT_BPJb(e+b?MS%_}39!nP0>5WtlGeX6&*wB?d(kDlhT!BTR*C$#$IqZd++1~1D
z)#DBGH*Q`(XWXmwQ;694+E%4*mVl}?%Q~lMin7eI(f-O2Id8mix_V-rd)Hq=J-A0F
zaq+3o2tv{$pqSLLTDx2_bk@-kvpFDc6gCZ7MOTu{-!AD4jBmyO<W^N;!Lj#V@hzel
zVcJ2>2{R-{w(Aw8OSWJu;mrN{5&<~IYNYaI4;u6ZZ0|0wl`|0F`qHdYhIsnX**~dc
z4;#V!{?vaoo+PxxR5uN=?W``mW{$rYSp&<SxgI&J<<~~eleuQy;^C~c5#o%_NzUuq
z*Fv%mnM@bewVbQH<n*>nW=gU58T%4u;+&gVmwUz++)>!_H4UcJ6PS8_x_?HPLjVFs
z0%fW3ciuXa!Iv%h?cg1NQ18LFL|J5bhd6|+XYM(F-Naic@t<(w_cnNP?sgZtYH+sf
zJ)b)Vimk^eg2waXlgy7^cVd|z2ulC|s@yRooHFp)TkQ9HYp)A$e~u*1j%QKYz;oWy
zihRT%cjetYv{JKDb6unqC-G`EE?G4fj;~{}n-K&@Tt=W5Xiy>)!IBdkdWsw&AaoHN
ztX+bFXSxVZSSGYTe<YD31csXCA&nRApxt4F@`3TzR?zgSN5I4chCP1_kI;G*dJGkp
zdl9gWkYDzwSnnq$LlnY`zP1Ms)Ry>3X;1cT25o<=XDrXOiEb6eKlL?rwN9*&bY|J8
z{-t4eZFO89{afZ(#t-JneG{&a$ajD6<DBK6T)xD6Kxe9p!Ge=e13&z}L{AoZJR5Ss
zp9h!>7JU7!iZwM#28S@$dQJ@?aKO`IE3}os%ZSnhWeQ#*1-4rn&7dGs3#QfS{@TW7
zd$}62Z;~xHePY@hJPJlZdqO7L5PEeOUyEo@u94zk2?LVXG6VbV%&a%>cLyH{fU+Mq
zhwx0r|5hASTZH*J8cuknas7*CQ4w%lPV$_L!;K?R9B|}lD`lRGoT(~s>CVyEm|FeD
z345;+cCw3PT*rFmB)hKG2AWopes;R2o41nsR2FP=CfrEm2_L4)sv9ZaOnA$M^h5fs
zSMk}8jv!G%UO!cj(=76Co~j;ce_t4yYoGl&I+>S|?Bt>ojmxM{<QxZ^cV-f6I;&YC
ze#YTZ%xU^Aoh(nHs4w+)y%Xb+6&fMWTvcVK<=-N4?B$NBPx%~$yi=T9pdI`RfSj==
z@e$ZIM}=cV3bXrPM@Gyp&uln(SdN{#+4_Vw1@)ch1K_H61egkAWfF<~msGAKY^(D{
z+nS}$y)58_A1u!n?U9G6T*YW;=cT_`v{k2N<a?D+V6T9$>#P=s1FvEY38jH(dKo$2
z<D3wO9iTQ-Nhk|qdNLZp&PEBFQ!_Y7Oj{M4nd1T2chTmcg~_jC$YIhMDsJ&<dJNp5
zEkt0^=EaPmM8F*`n$krpJpB32g*p0xjF?oe?efi5y_@8^3loR#nWkwnFZi{7bT3aw
zIGOsL2_tHRl<%)9tzaXak6Kgk@`{lZ<M=+RT(%Qe!@zE`QDft{`dIN+rIC~s1?v{c
z8UKX_S^xbc`Dl~nU~Ybf%6T+d663>QZI@$ubz>F3{cP@fPZq1(3CFh#tC)Ob)_iii
zWi0aQ?L7yhM>BV17!PbPow|RJ8o`se2cW^4IJ#>{Tn#X$F+zv)$OxU|6|o#d>{><a
zT9w9B#1yC;qJXSs3vOx-2rwu7skQ(238?V_gJ=trgq8k+o8Ta8CRRwO)O#)hG*mWK
z;MFJ^0cUIiL))mp#c$FVz|*PERH@GyoX|rm*I%Ly$M>JpVyG%^zH3D5NCJF}!4{0D
z1^Af3D+CUUWZ?N{FL~S)o_rC3Af9DIE&))02}#_Y4u$=B7g<1GSJB|nkj4@6>ksm+
z`U|gmD32VLu;n6k_f|;1$L94oL*Y&S>Dlk7!Ga67k|ZA%WxPNUUXPBHrs|XAG$(K~
zagJlEAoE16?wQV-soZ@>5f_qCXbEcgR&!poGng5}+#M1gH>;LVqilR4PtSI6#2+hp
z=deyy=vB9}<{NdCKBSH_Oo_cNDV?XD#2wymEldKW<Cl1okARO5KPaz+%)-Lb4dW6k
zpQm;i$#+A8E{AVD$$Q1XRTO=wLQdb53mh9pYr=~%zP)j`+b&L$x%QHnsx&}Ao+j@~
zbKWdP>KRm0JUCK!E-W&12>DnqiGS9_2*HIcqtQAc>y;F(`X#%oZu|parUt}PYg5As
z42&Cn0YWm|Er6(z^_3jr)_0CSEIYN&dLn;i(W7Jkj^Ihd_tnDGVG+INqrCET_vP~P
z<nBabS9{fu^1eGEjzK5c3#N(k)ta74i;FJf$og=0Ku%6CUEch06(gf0H$sh;aV1e;
z7Q7V)NQ`9%Q-L1>k0UNhw>`nJQ(>mPQ&I!I8IlqDizIjo)+$08=82K;@&vmGPS7YX
zBhZ^^r&=wq@j$R;g5xS@7>e$8F%xcJ<w|#f@S$O95y}{sJ=cZYp4cY8D$7r^>2}}p
zb%s;k5t#E1tDo1wQw+g0Gjfuc4cI5?Y`BA>yn)zl5*BJ-2(m;mq{E1!rG|TH<u>!z
zd(O0I>)oP)bn~0*?si)-N{PYv8Zali8rUbr`^^B)3cD@)ZiE-#&hDw1HZ^aQ#7QI=
zSCy$MliXe}E#dNxqCXCpe{!AZ%AV#M9#?%vH&XFXWd^JK={N;Xi2QJ_Z{<WMrnv*u
z2AH`1>;nFjfM9(F@JMk71PTlLe+x%&f&o%1;Dp8iVT=DaOjr%fICcPxwdY5!fa~6q
z3#kFB9GLF}Zqdq+W6i|?NZ-!mh8!=~^G>NqgWG&q9H+7K?*hz*5ko#H=N0c@QF3aY
zeS%mt&4Ih=oMB84*QXE{wKwuY!5UZT<pj^RZ`SeI&+0G4VwX}e`B3q;buPySz%HOv
z!HdJyCwyfi9X!0=lX_ywU@!+70Dz{BV?bOz1dd!Gu1yJe32s3v9$UMGAnR=b@%J|%
z@O64Vom&O*b1!>eY=(|k5<V4O++bI0zUY1bVlr9!7;k7A6(saH8t<c$xbvzDbRsW4
z?elBQjbGOoSomK5br*)D^9<lA9u6NNAZMR3?JREC=N9H|*_pSy`90<Nl`plpWOK^r
zvzz}~Rc|>RwQhWkQ~@P<2hm>iG<o3pA-i01E>2?gNoYKddjf2$eNO$L!gMSGVI!QY
z>&ZUHh%h_KL21z>^WTSsBZ)mnTq5iv-EFLaDUIXbN11M?@;e-MaXqYq6e&W@FLeB^
z=q1A+tt>E%$ekt1EedvwlT_Fzv!>OWS*RB$nm!$OV$t{9%PM#X3ry1mpp(2K?SBc}
z*fMcPz)kxECY`p(Cbqh%Ez7kxe5HLOe{;$RK`dmincBAwRqQORTkcSH{PfZH%ZQQM
z0K%~?W6#8<81c;k^M+SXbzTf`3nhkEA`!7&HpWHEXd)6rl`%NfVo?JyW|An~0t?HZ
z=ZIzVXS1==&+!U?m}8h;2x3}&Qn{dC1mrvH9FTDD^qw7ID4(dt4Zt@uk0gOJtdOyp
zZ5!(*{Fa!WjjnA)U+MvLKu2&FW!jM*A5@$hecT^Q>9fDr<e6#wye3Rx*yI^kJKNe<
z9-(Cg#tN$>zWI}UnBI=DV$+Qq5pc1u3l_Ca%X6bL#@h$F_TY8)C-IxgmOP<^I_THs
zny~Wv+D)-)OBMD``Xdd-{;Si;yURb&{fG8T1pQ#vl{#n^23sgr2WNMeo~0&TOd^pJ
zv$74}O`N3KhD$s`rcQDi`SDG_-I)Lb)^&f*oO}s3R7xChlf4;vqmA1jCi_P(_5YO<
zPCzda5}XGQnGhId|EoCw7yj2N3jUQSbBZRP3Tqg-OcoQ`A-LcPU=500{O^C02yx>;
z#~MVB+$}cCbQ+l8_ie`;cE&!m?h8M&CrhDW@w)y%dwRkucS{9c-zTc}>&Ru(8*Y_G
zBBm(`+X>s6Z=HL4RyE02`pI4M#*DsqZ}TI2WJP-`fA~rRpUpq7^FV{+)DRrzIsk0}
z_MhN<YeqqVqn<>lF9y(%KQ>2k5K(xRDx2ik19IZsH@|$m=x$fIl_cf$)YCLU8lKEA
zeVE-@GOfDu&eHE_X5m_yXqXEJkOTs^?Z2P@vGbM*N$PB{P3u?FaEE>Kw}&8B$g#g%
zT1)~1YmTc>j$+bv{4({1L|!fWeR|O+PE(f@8DIt-Xlm8lMaWM{^8G%8H&>(kD`*`!
z5OBRh)m!lwg+LN@;yU&KE1lI;b`atId%|k%msrDvRSk-8c;oSHs*s7l&rs3mpfKlB
z4jbucj)7%67%5tR<!NMV2mKx;Da<}}P8}Nk6#P<serd-DywJ?K?Py(%&4V?L&|h~}
z<@fECl~9I^8_ChISyR7=CmxiQtkn+MB3al2zzGgA*Q{vqY&%uv9tJ|M_2@k_!>~4U
ziiw*F@xboA$Cf|lOW#E>pRyNDwmxBbc)A3WeKR!PBFC;+!spC<q)!eR4`xI^d?A{l
z0>s;M6LTxxqApS{BvHo?{Ov4d5A5D^wk`6us@1~Wr(kmCc0t6iUiTnl;F8%?GFGri
zLg$tP0{-=dVj$dmPy<UEQHChA8HPHzFKQFde<qz5(!@wma2?VSu;-P;4mWRQl2M7o
zn3+W|xOYx;aF9>2UhPe!#}`#4$8_jB#v_||DJ@bT!q;j-?C+ePrXCRQc<)U}xsoLW
zIV)~l94QxHkt0bVn630X!NC!5R?lZ2FBaULa!&KXfQya>(;aS5^RD9y{jZJC*geL>
z$o|{L4pe8+NaadBGHECCCgKolsQY}QbhE~~Jeir3+dXxXKfV5e20c>@@jA3niEQ7X
z4WAxY=MO19(@11lE#2sRaj))%rRs+ama4a}C<x#BbD`pTjU(V6T)W&9t`{z($#luL
z{|BuICM`i%CMYu|QV~ESMF9O^U=@Q|0+N{EFY<#K5@1#XsotOf!NDcWka4mgei(wa
zdA}ugnqI^Vk7rAPRuHsy|M%=vhB7nF81haHmJvWzANe35H)e-V)1X15o9C@Ohb_Ut
zqlz|GaBif&tHJ1BXpiD9mx0lJ;=Y@``AH4+sgkJQ5xFCMAHvN7e<~Nd=)QEu7#`(m
z3RPGZ-n^~q*L2C4kRPd3IpC!=EO^~n6Bs5iY&MDj(6X~&X3Pi9f>=loD2wkcJ>1rZ
zg(Zx9tc2Yv&tnETuBJ9X<hP>GXP1h|J!>2>-&p27F$a$mYs8|y-Mck%c^hLM&7D|z
z{lDsF7%HWXBE;1|)z*2Ef;?ZQm6<Jkd>TZcNrakf#P8;$I~VE0J>ncRwmo?tVP;#t
z_nad17utBF=kP|J+A`rW5q3@aSBu3PM?5Llw1X#vP{+lxA=~?F5`c+*(8YW|E*aA?
z8+{sgQC)tYllHDmD2N`XD*w5}Z_iunHQNx<3@7_nn{~73#cOZKCqN2S$QJFknjaLQ
zUR83j|MCvscOE`<$0~&^DxPanM_a6?{)NOiVxBSd3d3han8HluTXyK+-7Fg8k%ukp
zF-K@yo_8nrB1U4gHB&3F)SdiH@==6&0F=?kJQLzHxzR~?w}kC<TnZ*AnMMTcFB>fE
z_%Y1chuPMslhAps{T%o@U7;=vgZ9R`_69IfSM0-O1jOCKF}~1d*(Aj(BaY}raJl7b
zulP1|kg+*l6oe|30N{S0P#if%gpQeNF`?6{Lol|-Dol%g8p6C80cOcT$WgS0R>Af3
z+98++5gHwERi2I}77Z<3c2p=UJTD*1=QFa#B+5@kQugu|DccR4d9dAF{Bi``cUG_H
zbyBwYkZ<p<*@;p3Sw^&Z%?g8_qYWAqLYGA7^(WG4XV#LmR5VnXPsue+7az{>24qCy
zhNVLz=B!7l_Fd}5q)da2oY*PfO^T5|7}FD`2PNNqC84^@h>QZyevb>h9~5nc55SA-
zo^9@p(xly?351h5*B|$OcB|F-zbDnz%>lv}hLwxd^-K2_vcB3hrXXIi1g^ResGi_I
zCj}@bkh&-il!+Q$5CuNNq8NeZ3xO#DkN?U`R8V??-)Dgz=Yc=jQz*!@VaxuMWXT6T
zbR!z;@Ub$Q=xNY40>$@%Btec6*qiBUazqCJP0Fj7qgRSmGy##g;-#yiF;n-@<^<sV
zCv5YM_Mqr!2<fr)u-VJ|hi6AkrIMWeRU!jif&q$wa{zG`p)+T<G<89iOEfg3s|#%d
z#Un74sN@K**PUMntt41%80q++S4`nrXH+n$4MxHira{><CubOPPn8A(l(+{k-UjJ7
zq@)qRfe@Vq#tbBP$mtH6>UgJpDFXf^Mm2>#tl|?30{qr(=Ii9fukGpkT-0!B!Ro+u
z4qE>tc{*_0ledK-eD)RXx78R%5wwXN2PY3`77Ah2&jAJ^JI0z1Da@3`m!>eZCQpuJ
zLsiJ_n~!V1_qx0*?mo8$!5%X&T9$OKg+i%vN+q(EKC&grZg$e4B|p56cR^ZTI;o2B
zBvfTxDCh=sUNL~KMb0h7|Hh6i2VbYjsUacOCnb4_4%OLvM*8Nnrm**20w?WU;^T!z
zhZaz`ql_J+a!=c0ns64YWw1ozyN2%fi?^<R+NQ&=eL~)=Z_N2ijx`3;F?+WBIN5OY
zpKn6WCqw>H)#bJuSTgvHDeoSIX=ZuQ2w>;K0gKU+t$xwXLxg6kba_^OM-n@UO)?(=
zS|_p*SfkASeAv)^f+Moa<t$(^bvd5kKtbShMdfJZJWJwrp(kY9iuWY(s&O2l0W2Ld
z0Zo0sr!Yh9)fEz~^1U092pFKFw><F5kk@#LV?0T;La)*~B~MvRv41Qp1qk(p_tZX#
zkXwjuG2nz7ZV)H4B~z|QgFUV%#6L}gmYT)<W8F|&OKC;5_L*w#<PxT${Csb2!=zhp
z|GNnKJB9fTC{*WoZSO`@!yau%zxIs!7isugvAc(If8RV+>>1+L=H|Rs->0dqN_`V)
zhDPYbrgk~~@?pfSV^-gjSOeZ#H=@<v6Blkzf0k2<hH_jI>vU`hpsf>hdP3IL&FD?q
zJAIzdw9mJPDeH3>n702F$$!5g3<O2O046lV3e*cg0TB47mXLqy3DgqEeUJk+C3pya
zOmM_PH51=BNZ-MLJT4rCW1YV}g0fY|0o)Y>2}&SB;GJI81nS6I@yQMRHt$$FWaLYt
zpWg#*M66=H_lF*4MpWF}dlwQC+~b}{3%SQuId3NXdNgv*(h_aEG;b#4F^*J*>P+Oo
z^Z>Bm|4oDeJ5W%@X!a}-`oN+p&VjujV5Nb)ADRhC2swl}IaaW1D`a_0g(nfRU0yme
z-px+zF{osD?W_3|L<2v%Kc0C~yZ1W;M(D-V<nN;*TY#&Dx%KLF4}d`xqVcOJ?ZDOm
zsCg9cXk<*c8Z<az-JBlif4RU;-BK!Gti=m?0=)fUMzoSCK`o6e4*rR;*a3uYz%e69
zoP!3!jo5LpDyIvw)0l$3)jlor+C4-OA=KPojQ7pE+CVTiKC@=CWtn3a(mRN0!*saC
zp*8n=Vi@jseF{G|-JC2^M#aQW@J-5`B!qo{)4NYj-*&Yw!eo&=AaJHZ7!-762enl_
zm2J26`JRoZ66X-FZWAwfq`amFMn%RmxiQDz`K)}H&Tfv~7@TMkUifB{pILA^%}imb
zSM+Ye!)oI5p!=cLk<7YJ{BuVSxs;ndICiA0W~KD3PezveY$;mi*DnepZ08K;PbE<_
z9kv=lwD(;aIskd?R+2|USaZ{Gg<e1({1-|9%T3f_6m-aL{1Oqma4Dvuy|Hv$-d~>l
zanF-B`6F9X?vtj+6fOYD*&cV9qU3K?Ma4}Ga+;kqiRv5|w3l`=$RO!KorX)ZT_7}Z
zm4h-?Z3k3cT!#V?X85w=ReoxhE0+CI3^~JH?$|}Be>Gq3fWOO(Uhd|y;zp9mNO}@5
zI?evo%;$*AvTytgEp>e3prYIs{mv_IYH*fbNEc;>ZS<DQrFE$xcoiz)n!aznt7Ga5
z=kGBJu0VrN@M_$>NUBWmCawBjI|=@c9C67t*%DI?gQh9Ys3KseOQL~;kk;0Kot;_w
zGWn_de7+eM&PFtdDP)0JeizE^&-o*a>mSe7IvVM7r%vyJ{_tLDtU!VLziCWxP6>fF
z4hWCHb_~Q)gAek*zEA+!0f_=QcYv)xeGJ-BI0E(?tCuVs@siX#u>hRm`OQ#xAwyz;
z34tRCTMl+-2Z0DUS4nw`<Xst(R(STymr@qDQpc+bv^PN@p*F}Ts_f)a<Yc!jA(`32
za57CEVy--YWVDg*7uH?)b*?kc3LSVTs<*_@>u9Ksxd#74S9H)=DoACq`J=H$DFSNK
zJB|?O@gVb$0EiK2u0d@@bV0i08zD8Ip}*0Xs7hrueG7oH{6s$SeL&O*j4CvhN`x6;
zJ3sChiKDOJbxsUI$0~&*8``pW_>g9b#9o!wfm^=|m-=a$?laYsoJST5u;Seh?^vzY
zseI$qcRyd^RSRj_qMOjQm%M)eB#N|;^-)n?V-2_~B>tlB2m9EWs{(lkt_rN~0Oh+U
zlzCFlciwyt?2`#emHYh3T#}F1w9#M#0bL|XTC06x><(8TtjK&HR#q4sXQ&xRos!It
zLLx}rLK_)Xw`_s`Z-8sgeA5r#+-ec48W|5a;n{h;vT)!iz9rQiaJ6xLtdkUEZ$ZMV
zhkHn2flOil!}`sNO7JqXR*_dl3R^1=Zx<$;T=d$#g|Bya-}R}Czn||!MIoG*4h_(1
zQWC@0{d!HW8tm4Xag6*q(zbujQDm*EXm0;yZj8|7bLd`Pd5}unKU&!`kH;(CyNl8)
z0bqoIDk*4)Ty=#c>R>y|4l?7WF(Q>o{%8Ek-bKJ{Di3x+jt+nxTKaIC@uH2Y<l*X!
zAQVVuu{&rO@>;3>QlRo#{{hDEyxho@hmYLsk6|7um-6{%XY9Yq^b)6TtjW`+2I^#a
zNa{@=JypQwC64Mu;)pOZSbBOpCa|IWvL$!@T^f==WpKUTmxO84Sw5`i-%t{qgOezr
zO;jTMX9}#`A*&z2?h1e$jb{R&#+S6xmQaSOcQppeZS<h;3<kWIzlG=_=cSng4X%4J
z`vT0&KT;j=ebz_r;!8Vokv%S#Jt#iL+1$96K$fXyp3|sseijZwdCpWKFX0xR=S=yj
zTnpKUd6gcoO_<3GI~<3I((8lFhHoUNPuey^BUk{5-@?6EiX>DqZ#aQ_KMe31g#j}Z
z13E+SU%&`b5FEb=Sn-3jp#eDHKqjYyvq2cxcnN+@K&yxbk8r@A4A12M`Sa`oh~Gvs
zDT=?xwtV4;M8%Gq{o?)YMe@(QJtEmG8UQ9RP@y|OZqcw{7W+fT0=g+FuS;H^{j$&<
zh$<2<Ogq64<I~<4Ul}<3pi*dWoqd4^sa`ze7tiI8c?0i$y6+}$ygukm1Ww;5TqA)b
z!R<+cfc_o{%;*I?Ofck?7on|aHfxpOP9a_jHrL<`#rs5V83MlY-<1}<`YoN8-$E3m
zUFurpL_DB-O#3Fq(R@t6s@9iUxRc)>BdlO5P8PkU$`FpDTLwRwm>WV&F3n6%Z^T`P
z-YU)9TK#z1W8zc(wx70F81I6Q9g}1!dD`M5-!gO0U6a=UFiUM>7i0}Ce^G&}01<|K
z-v;y9CN5-o+M*2B-)k30_~7p7u^GU*gpEuBH7*(>YA>X#@x!;6{OGgVJH>sr6xK@D
zZdNETbd!-Y#Or5#dE?|pL6Q?n*5xZyQ^>!+x;gB57BfH%jt3O_Mei<!Ih|FY#y53q
z#KIkdEb6|sF6h4CQ%vlBI_q9kd{AYF%yW@2*q)$N9n-$^=8{6t?<!=0RX6?YGe5bK
zB9{AWHQW#Qrxc>?-);BBKlI4UcisK2+ZYo2&l)+f+(zVg(7@(hHAF_{b<d}5#Obm<
zr{BRXcCa&(Mx|K<_dgb}r-^GE5gZQ~@0_Iq2}F1r@Ytg5)h9b94IxNf!VK+bG=509
z>Y~D)%3u0Xz?fp-;Pn~fN?)a@r_S<11#Dv};@k{lGr&I8rIRR@@~^Z^y8MnUG#Z@O
zGA4}VJqi2l8yAVa62B7boi1UF-Xq~yyV43#v=9Iz@bXlzVyNATeZccFw&8_ooeF8d
zn&t%0Y9)7R<^m?k5aIwAPtklz^YMl);hx6elEoH^{WR2YA{fU03N{S`X-+=p#rfXL
zbhWc%)~A-X?3=;HVM&jJ^S>xpdyA;y1C&Gf{J3pNxhdATZm^?EqBv;i+2n_<)%O>l
z&hjvC2>Q^QvzLY$X=#Gp%53g%2iO`?-wjCr!DZRuHq3_4F48iv92-z(k3eBb46`8<
z!6_yX>s+1zE5J{GekSO@1ofpSs4;<NbeN3<wI*RJD9y`%YEptfK4jHdEccaIe4}mT
zPE$)%RhJ^NadwHmOk$docA?(JMJ}?ub(SyvtXQsUsW-Ha;QxcsGh@AIFob(-qJ68c
zyC8C|z%q@dv~GOG^ftZ}+u&wL6!a+I5QQ(~GL+qQA=fB`r-s8XclUUD)McBf7qy8k
zO)fhE&#Ic6FM1l6d=e>HC_2BJdRsN|DyKR#VES0lU#P2xWPewvnqIj0?JvA!Lqd}L
z<}~#Ik>=2-nwPD@!PX_d!Cgg#7J58~t>^QLr_j2_4u8s664SF2pc$@UVFgls!CwK;
zrwZ^H(17~-XNV2_%EZF(<PLD#w6>Lk7hwPog;tqIlBB?tb}@KLorjYjta5~-&7UUn
z8rlzn%%K<Ifs*{fhFicu;l5(g_L>^tWJ+J$kw2aVSyov2JljW$RqG5UKRxZl)w`Xk
z0ITLz5~jgda>C+Izd-~1&~(rHR<&oZ5|>Zn%3yuljZ3Har}uuX(#*YD#J_GBq$8OR
zy~xKd{bpA+{R;(phLAR0JBZG0H?O~P$N#EJP=Aher`#g5Aud**A`~2VdU1SN-h7o~
zf5Y4gpPz~@?PjR33&O(OU?iE@EyU^C9WBsK;=T#~@b8kf{EH8hb89YqKNQf4JgyO)
z)+!Z33T(U0?b~uGF71-yBUExw)9v5@J2C4i>WiS3TN<}VbkA<eI8PlrOyr@{d$}>d
ze9ecQQJ+(~%E-1jcur{@vtZyK`UO~-VbY7PCKa6nW2BmZ)w(uEdGvp%dhfU-*zbLu
z;ZDtgg)_G}bKuUxjpjf=Q?nj-LTS#1jR!}T<ieGb8Y&_<D$7~<WVtm8k<7|^%vEa3
z%67efH}?MgKHtwjq`d4U$bIf}&ULPHU1-SjGHo>~<T#Mu_ZatKNY)+7BLa^Dpa3&h
zAp=Pbjoxc|;a@s0Ab5AmvDXfH@F`|A9;4b36em?vI^>xbfRVsh@5g2=yNCrI8)2&-
zbi4OB<oCMN>t0EvIPP12wb?w|&>Yut;44&2LEua15S-PX`oBtvKHUWoTvXVUyTTcG
zs|Qw`DuxwvM<FC602dn0X@t2$?ESgBP0=Fz*iE{A=e;n4!(aw3M@@9RFoTEUX!zpG
z=3oj?RQ807>4)(8l5g8Jwv7qUc==PD(4sqDV$4BJ6<t&!=>wz8iA>W!#A!Rn9%-F`
z0r}EMVfS?~m0d9jt;-SKSH)~MuuTW5NYH$E0Q=7pGzA4>t{qZYmmzG$0O}yDIKi+}
z_%q<a9Gp~eVlle!Q7~c`5Yh@NJ3ma?df#&3GrOi~<O$-*!}0=l*4rWfdCbE;U<LJ8
z>w%F&XD?r^p4(OJP4Dfoi;37>mhaxCf2)i3&z@gKEiDC&(lVN*Dl`4nN5<~fUp~_|
z@I@-X$7QKF<BTapnE{w`DETY(L$F(Cm%TW?YQ;t@$^EQr&tq?5?njLx`1t}jRd5g2
z_kLzu=#l=XVid*0?D_@>q8=e(_s9Dykpa<Yzk6I*R>NBj$Pkbm0j*QTTLjTAO=2=P
z-k0!!0gw_^_2t1)u$0TSq3Gr&4}2#`jslcU$@(^^S^aA~w2O5*0ge)K6F<=`i)k8n
zfgPi_n_OXDuW(S<)4k3T-%CnSpV1C3@Jw0`g0fM^0458|@)H9!g-`($C=zX|Ng?We
zi!68jiQ$QxZ%x^Da*pBrv1(FYnv$b^WqCf;(%^?aWfME(b(icB>4zt_8XJMIvGJ0w
zY&kwOtoa2kQ;c!Ooe=B^iM?-FOep%<S0JeDPhd#eMyIivBdgCTNbxJ)C3@ZN;C5**
z;bPlb;cD`yjN)qZ)<ZFVCNB=Z7_>hm^-;Rz?5eMY)C0DQ`GC(q^VVudgI-V9*yQ%Q
z?{Y53RBt`ia@P$k!u|*1-DveK|KnxVUQOk^RMrR8X#r)}=VV{hRKbom9|^UOzWZe?
z6li6_p{XONP5{cU#)GTjweBz^Ijumj-$);chUkPp%mBcm>mWo&dNMNaR#SZsLzeZ-
zhM|$!c=<@GCMdKDLX&`+mJ%yVjoLI$4L4u&Md!G@l5ycJaV$eo{@d;+1;krlGmo4H
z!JJ(u6d+REd-?O=U+(J0_7~(5p$r>O*mYE*GS@a9;==%|1br@sr4Xq!h<*Bm9!!T3
zC;|8jZv{CN1(2kri)82M{k_=1&9sq52s4oc>6NkfM*@hGH*Ia^Dpj6S0rek-BPh`f
zxfHKWrQrT5zYAD5CEE`@lnjryPKs`K@o?Zrw(Fv_0TuKdpkmaO4I}z;aW`iJUyxZl
zG&){In6EWmq@4*Za70{pPf`Wp1lUL5nBh03z;dY|k_;)}sohvG$d(qu=F;He|MMkE
z1n6n=|GoJ;(BX~$jk*$t<vXr>x(m0rdfuF&25G064bGgt?TvHaVq}RhbTWM5w9BD9
zjgUY52HxO|GI}h=ywUv3Xten8j!9$UajO|oGg7gB97v>~CoBRY78!KN)h<g<=owkd
z<19_hl&P|oz_(!?Z68Ncs)D*Z3df(eM=6yq9yuuS>cGr$Sjo1<5dPlX@<S5Md2iin
z<IjyIP36_hwcHEi0}yB9hLAD@)aWCu5_&kG3II~4uuG-50&bc|9P-J=^5B0v*X~dD
z4LA>9h)vb`9yaV=(wy?<j-Ws=eS2D9xctEsg7N3lp-YtQtmrPyntCPIT5yhv1U+*U
zGt#3EAwKGU#9xc+m+ww3IkMf_)UuxPOH>TRYGky|(-YB<Sx~4;kZqJ15>;v6gHEjt
zRSow_$cC#^gVWeI98OHijni@^m-nEv&PJ>lW*m9wI;iV&Y(ZsoI{t0Q=J(c+LuOK(
zZK+eIIa@}R&T3L!W85R_H79F)*WUitTlFcviAvbh>oS}d35*cDUnXZQW2BW5^qxur
zJsRkEY(uCRYBeYVLFuMI2@HlSciZF9<VM5q1i1Ex(bo+-eX^r?4+D}P57Zk>Wt246
z7swbYY3eeO<_;w|+f9RzJv;bPwTgu|1I(8yBn*8u6_~CVSsiP6RbL49tade63I<wG
z&GMaqhYluW-W`pe|178_8am~4If%>>%QCw0c;q67$b*mNDB<HAsbYjw?K(8%uA$UV
z=68akWy6o5I_n$u+2~0hF+LAqY_MldZ}v^E1~HT)eGh2Q**^mF12(Em>=<(hkUCd^
zmVTrPmIYS9(#%#&;CB}A;~06K;knJ;p!Vd0V+5q}jngN;0)L@FO*<y6xgdwKuoHDP
zOmeA8HN&fRhW3}W<tKKVYjAW*fZZTo5vd=&L$GX;k~qyJxXnbDGd^^L%xZHig0u5f
zNKNvpQgNZ-cBY&86`RPHEUZ&HP8?Grdnz41H(tNG-(q?r0qOn!_y<tp6AEPt5q)7G
zfh2XeSg3#Cje@RJt_JDgdJV0}7B_qJ4f8PXu+I}latlaC1vcvt?6N=J>l?=k>vYH$
zsa_yn%W;*hHnvGe)3KoiZZqxqv6ov2X>V{r><>oG79_(;5V#v1mY%&k#AtEv(S}{8
z=e$<pRP=WDynyIbcf^`T2L70}eSw>;^p4-MCp7QZ*OBikN1wK^UR=8!5HK<|M6HX)
zyOT7>d~O#ic?8er{hIac2-~}IK)1_H??Q(@d1T{FY$*lhT)>79#RmMtg!>};Pz<pN
zEC4V5JJ%3ps6YZe5;9vf<%fdT-(a>xD0h8dXRE^w0Ij55C)AhKbCC=Q&Iz38xW`!h
z1MTJ+v&WlzA-b_ZWL~-X#*uoxygQ|fo1gy`3V9uGHg_v~_a@lpfSEuFImA^6sL5`W
z<~bDY1g{pZ>+c<7fbqNZ6dCg$NSwgn5x*k)(o&e<d%&+%+4LhUvFeP#n`sgKQmxIQ
zue(D$9l!1hg`6sRtFe1+csyVQdC&!!$6;E@727p<KF#}OSZr~Bv$|pS5@l-#YC#?&
zqwXGOKFDTBX?`S_!quO*^mp@N3*mDu42B!+c&u|TMS0=j(_Zf>;}kVBD-f^TnHilP
z$Lquc(GkO;VlFMj1(h26ti)mQsS@b0&K2!CS~nW1q-UY%k=$M^`>Bsw1K1@bBzh6u
zS4z-TUxp0YS!-AE;A~)2EUj`fz5oSF5R^im2>66|)`o|ON3ExrEy`E%u|>mDG@#1X
zE+@gXY<a&B9VjW6<s<QVVa>nNCiJFl1MCFyFxSI9wW{COoCmIS7`V$Qj7?*J!EN(v
z)qcy_e~h$1xj1oZ&8_|*SRz@Y;`*qNGf%}TqMNX;NBPYk-6_XTr$5Zg!H|L_z9Lk@
z)j2?Qg)RVq7;LGARf2pSOCQCHM2GXUM$mo86wmiOI4wUUe{HI^wl1=1wai+Cois_u
z9%}i$E^oIZh@BbtcHgo4r`xaW{4EGJ+mmdV!*<0oOwG3<rVxFn+N2gV)+0Rw_jkOW
zU2zY1_&O=@e2V3(cl({EO@T5Pm{@HTc^huer9zdi{|%~*q$CV88li_FS?@cn-!LSz
z`@lZr6I+#+@Pdv_Dsj|%-i8%YV!qzZ!yWqZ-nQz6-neHM=~<V}`un412)Rg(1Mza;
z1=a4R%l56>$1C!OCWt8@s5XF1V(u_-_^aZ#9C|5*LY|!He6;P1A?~<33Tp1ik7u}i
zdh4!lHAPVq2D#N65F8jYQV$x0#eZW01Kt3dPk@rZm~2C=X(Lpsm`53sE}Ug*;yJK5
z{-$cN@CEkq>0Xp_GW16VJ@Y7JTd0dM>f3IOPkE>RN|~yVp8&E!8J)fUmfvWX0x(L@
zL97Zzy$bY39Z{#OOYh7jI2H`ul-Y&P`EHJ*{1WhZ6R9eR);e3$z_cd}EbED->0j&c
zdDHh9p=qA5GiWx|T2_*4A${oTpjWuh9-UQc_(6|<_Pv6CZF>0T#<-;4FV888`J&90
z$Jg`HK3iWpn*8zhuA_WPw8#Qpv5euO2;kEvK#!!VyH#??bc$m2AT-??t}F7LIC1Ot
z-5#dsxy9~RI=a7%BD)q+lmp%xzJi^idn_Hyc798iEFas|o5<hx*&UkP>?{7B^OIC_
z=<jDth|l)dWSetYcSgX7U0W}>`2e-Rl?$=V@^PKF){&-$V>}FTAF<*$)&@Q7SRzd%
z><ClDEgVRCNrDe+RLp~HF$nQ%q!KWadqX4NWUUgMs5y)4b{zLeZc#QL1PvT-3eyh_
zb=PXK3XDB_)EP5Uf0i_=B-V5%USfTY{*byCsXW56POYk!u@rU@?*0jv23=w49fv&b
z`nsKW)hnTsJA5-2{)sgwdwrNVi^w5<H2jCOKc{J6CgOqtJ)gP6(USh?wCq;%sjtYZ
zW<5dKR^KR?mvCnZ=;Beq!gpN}4<5&W<$cG&`|)qBZIF{(+g=N)^njn}oK5`3yy_n2
z9myr79V$nkY)%hSABhehs?|}7qqV=StH`NoFYN;@IMx%<7lB;%Nghb6o71XegM;>G
z5T^3qGA2@Z#DqLuFKRd<Qs|?wL52bp%%MvI<UrqH#6>fUURn$;gJKvQ9(`poWaiUx
z`@nOm$k`8M%d@MYIt5ML7kwd${K_-Afll3xvoze}$KIw|n1-iVUhI*Y_A|qm4X5`%
z^mH+lO(&We(YS&C;vEdDUI-g7aNVF*y>~`sCV?$>6O;Q0>4m9ht#ps9f`bmlojL8i
z@$qfDiXg-3qvgu+)0wiU>>DEUg1FR(kJ2jpss#%{Il#x_i4g9T1RB18;~N%x@i|2{
zjs+7kk3s840U2~d)E9iQQK@b`z=*DQmi?cP@|H64EFCqDabDmC6u&tCd<A>8AfG4u
z(3_~YM#bpJwEZ%IIa~^j1V}rS_#&hE&@}EYMY9Z|b>VH-vJV2%GiVkEO~Lo?C>qhw
zH|p<ZQI^|#Pus64J(e{1a#LkST4Kartf7`=7RS{(-wgr~bG)T_zK7!UgbvsL2J7-$
z542QW=2}=3S6{bPORL@W$WVQRDR$eD@3YVv(t9gMBI~Af4F!-9xij}4ITkus&9O(!
zX9#jLb6Lal!bWyGIdL1{(R%*A8n~MEqYMGXWKH8QJMt{kv^z}RzMv{t<T$OlK;IuO
zj~`FD&ZFVEA|Jgv4=x(7ptg=&v8j%&e|gK~)EQ#=)Z?A=<#n^piv|q_#Wkfl=%;Yr
zT(z2q)rBAJVPf`hE%H_{`bD&$<P7_X^E8b74M9{EMgdOqi2$mOhK=h8N)(8KG14bU
zb`s9~4mdw#0!D^=I2dL+Rd-^m%=oDl%t4(1NhTu4h+gSyp;$XtO)V~p=W68<XQC~o
zQ(JuF*HpH1n$HgS{8rTRN9P$cmUI$I$HD=k#U35phf#tshq6uaxi<Y?wMDloS-Ix@
zbQ6&!XZ+K!%fAYFf_BOLTV2yfhrjsZOgzP^(!_G!qI1|T0T@}40`fh@a2`*)|E20#
zH1#1|kBeeN0H%?i@C<HXcud$lHQ)=_c1+_0b~7#Vj<XbfT$2HjEL>=WE0l#E&{JlJ
zuxND16{>fesHt%0OBx?pB>n=-f5B*2xLzV6q*nk#RX|d)aED8SoFJrNL<%#|;7+?Q
z8tM3iLHB*r73&bGHpB_fUe|Q>cCF+8wrL_p!%~%uQ&{!av@|u^FFJp2wpSx)v?-7S
zAc(4b&WouJGTkdypIBuNT)z_Qy*vh@i^#SYI>~(dmu5s4{&&L3vFrI`uU)<^dsVG7
z_Nb6hnqbzpL4!<QZ#aAl1zWrxUnaTU`p`OmAD8nQYtLTn^x7&Sbx$ySYm!<#V!Wey
z==gsrh=b4QM&|&62i^r>Z|M7qKNpf{z#j^vR8!3jB_VLPFok>*;h0>=;Ob^FL^$%B
za3a40p57V6tn|&JA+P3te^H9r`ARd2pHY%|dW{+Q-1hq?rs#2x(*dgi*qg^#+Hwuu
zh_f9KJ)bMnV7xY;^~Klv0tp<mx}r_HeMH5T8XFJVZizP3oBzGcke>`8mv=SU*zNhf
z8n|*Xn8Bc8&{bv!EoVRU{9a>;IypN-*155}#(tayl+_tC8@rdug1uLr8SagO&LJ=Y
zOQ+wf@zU^pJ9r%fbFu&~U)q)^7)4`ez@$DM0Z6bSQPdMB+oFIm<lyhCK^Kj#$qZ8=
z%{ypK@fa&GR9K>zpoYD#OKbI*7)rNm=&Cv+*0UWJyf6x<@{r?)#h+SN8=W~^x`vUT
z19i`3$wAZkhyYTWGL{~pzLKrE>X+aSQx4Lx4}}D&5vA=Jq6A<$p-oW%svdH5KS7_k
z7*nih6E**lHU-%zLx9sTlDtRGF~4h1mnlwnC+vMnKPGk!m@j#C77c5EB-AZcoJVCc
zr;TkwVcqN}MPt|KFRd>;OWeu}YX7a{a8Y@#tU0AB_r_e&B(xnTa`p0F>xg`P%Pofp
zK1?2bcam545=H$M&81|DV%l)$aYs+n2Z*1iHx`hc<5>V*1DlC~c8pC|y=YnfPqXkt
zUNeO2*vJXtodZfNeek&>s$W6p8#X;JlF`nff!%%^irj<`==g=$FM2>91@2%x5Lk-S
z90V-~Ug%s8rYS<w4>*@#pI1n|y$hZ}K^oqG_8Ovsg1`-w;t@s}|Mt4&ziLhJQUQw5
zoFrU2B9Nw59F-ApV2YzmtX@#k-kxlaws7880&P$7co&-07MF7Y-s`N(__XS)qf(~8
z%k(_L_U>5$V8KIs=1`e-35gXItLlr?rd&;9GLU$l=EjRj&rHeSS2;(}FvG+8=BJFE
zt<52#R-ddD&RQguoex+kf0;AWI1|Wz`{v)bW4EiLXD6EI(b>bAKgRx6X-Eq%hK*C=
z&Xcl>;|6FudZZQ<oD9o&DgQIZhRl_K6$d11gT${1E&u@i0dYY&K6{i2?+my`ktsdI
zfXA4Di5?-^g&9kcgU$$Q-<zW#|CWp=`2~Fj$D(uTGe^GbDxATHx!W)gJqDb)7Y7ij
zup*^;%)PL@??hY7%LWU%y@mg_dL2!OieK633b=crVIRSaZ=UwXN93vZ7K)nO#5Ias
zmRA-=HYnOYEYmo2{I46G(CA+)vfM`>8L{iEn^7s02a6W3|D6vdkF1u4_X||YTj0Q^
zoPO^Xp55Z8BIdHZgxS-*9C9J4X=tye5>V9J(tS_MqAJyT%lz{Ud9?*<)Q8*NBG!8m
z^I0RW;-G*aIqUCz3>n6Qdw6lJ3w$9)iFdEq9l$kXq)hIF-8`)nluA5Y#IxXZf8k81
zebT4s2kSZ^1T|r!zg}Qu^!H!P_ZWM|g_ytGbpr8s-!kOcm2Rbk)9oe7k{Sm>A!pJ3
zQx%X#jZ%r2-HJeVtMn5a2+yCA!2<iIVKjhCwO0KGfo(8q2Os7q$i|77n3LA3fDT&g
z51k*`---T_FM86*&LJvPcYQ<1LTH-`|4{U*Ki3%>rE-P~H}+06F1vl|i{GhL^4^Mv
z)0Yh&ww)VNM)M6R{TP7X%;L+L7THU{B_b@A(hk*b*x`v4{jhvQ1+wFY6i(z&hcXx!
z;m<H6qmVlSrz^}M>&&5CWph&ZAQl|>jbeYJ7(oEkJ?=gUGAU#x^?|3DXb7GF0>nia
zI7;j+VQ%_=$~17**8LyW(f<&Vgd`+)$WA@gYbS^UFTB{a$=%q3jIVF}it%<xPA6;S
ziV0fF<OGFe`y+7rju%zcpmm3D0SO&1xi<iqVj{U74|LLYl}XvChh8hbrt_^ZG%+l7
z)wk-s?=^Rfs2biOKB7@0GZXXL%4aCZ&1`x$%35+~ho_a4tFwuBsED2rQV_jJQjph5
zyT90{q++d|jS9J1Pxs87|ItT{yecjd!Abfxg_eLu{B$Fnw_kbvw~C2l{-9XK1{4o#
zvg(A4GNaRWS<>9c7LRm4FKt&a-Ds)I{xBScY{lXV3DTPMN33Mh?oKcPx4^>V&7MGL
zN;=<|yUy&nS*e%QmIwj3Qv*x$0s6xX4i?+uJ`Bq!pD)SqI&LEe^3|afW&O(JP08Yj
z0peffm{Y61HX{`Y&a0S-(0QMD*6zP6Ph3d~S&&Qgkby}B0^PocDFR}q1?J2XQPDW)
zO;J?9GDcR)k?O($21|9CZ?dKR8A`2<q#NgGA+d1;K#uhmOLM=S47JzctT%O8d@HhX
zvSXvhvu}#KJBAflO;72)wv``oAM#p%P0BKcdV2u1yq4ZUkG?EWPtkE$hD#cHHk$|k
zETSCctJ`&zA$I+Ux^Pd8-8?Y{0~2iB%8RvqO^c-(z+E_(`s*R=(&yX<9y-gIvkim%
zqQECD`S^9rzd7HjCNg2E>80CgDPoep_1MqqqAD)wJf~>WoL2)RWRqpVDh1v|Ye!OG
zbTW5z*-JV%--cb(2#lI1z0b;Ewpf9p4kLfHK`3lq3aMTeKvaYbc2-p!%vKpO2&Tf@
z$<MCZclkHCeKu3Qz;A$EvsS12q+p|XfoFh|Z(XUN)ME*(jVhU4I^^dc5!P}xrDSz4
z-@M7DG>GLwz;exz<JfvSqzKPPZctKd>L_{v1S5Wwnj-oT6ZV>Ai4*bb*kp>!QEGru
z^t{`)iIu3B7{6>4r_*u)#}(YFAY_D)SrNEc0d^-mNTm=wD7A#87wDcPe84xx6QSoU
zyb8cZ5)%rlG|+W}LTrNp`@h9YJOnsnf+7a|3ksQvhC#S1FQ(1^!kIqyiE6p8$dUVk
zFX*AjP>#GoL-vDSqqjTENF}QOx&$c0x98Du^-ALSAeeZAE#%W&1-*NnFcW*J1M*g8
zcWnEM_~0q=Hm;Reg_p!JW%e}fK{_*qs7yVWN3fqBlL7iEU}gg$)bAV0XgFe${fQ%`
z?bfmxonvxTW>Z4gHg<FDTjSEFMxsvVnDebe4ZH>uM}IZJ3G6b_6=DCXPvuA)*bD<`
zhQb9P#D4RT;|=<?m_0v!{vcYn-v7CSUudiyo&Vq&to~rqPkoN%0T0WEaUQ%IpSe~d
zN6_~}TmohsC#Vt#sUu*e1tWy>VR!;m#%t>)tfXxK9TGM{cioEq_@n*#37$5y<z8}1
z)}*z%(3qit0Z}Qp9Tv+d$+dXUfcc;d>|{0ty*o4LWnt2dkZ3e|v&8JWF^x$mm44Jw
zq4+2ZA2$7T=EG`^&b#em*;Wd>B}Mz)x8C>E>2Ue{(Je{W=XQV(ZKCZRK8Jw`EmDD`
zH+Mb5|1H&~ls|JjS^PVV8ouwWOV8<;)V$NNSNdy18!f?l`+LK%BilU>s_DqPHGe{w
zNAQ_-hI-N4Ec)Jg#h;Sti^=VeU&SaHXxpFHQWn=GzEzmh#)?GId}7t#URsQqyKj^$
zUc2gS3fzYQC>TpzM^xmcJ;R(e_HZz4uspy*;NX(tj2Ti3@`E#sBI<EEvt}gVI6xzZ
z4X78j5C=*y@>W|`U24gYYHjE7t4fw=%*iJ1*GP(~yLI<!g4cf_Cy52`%SJjP-Xc}t
z_^kyKk3{J<0C)^kZp}M@x^uRTA;x6>yz)&ju?`3hH{-hw1<FtAXG=#2O<94N>8ck-
zap-Xv9$W${&KOy7?IIqemjpIt!CV`v%@jvN<Z<2oI5=-1vg^}-AYZ?P=DrTI-QM8W
zQDUv`%d~V;vINUjvpiofS2clj5>}cU<lFi3U+T;JRQ~i!PItw6T|2nI0h^a<9tRcR
zk)_}$mhwcXcUd0HA@5dyu_7qzDgvr#>Nf5pL7hzUaib;?v<CGpaSt`^IY^w){|5w}
zP@6=7iUB<u=yo>RDKFrq0dg5A5J5oVMAMD&E=9t$CtA0GF3^D}2FMpM{E{#l!G8v%
z)g>W$2^=g$&Ka4`im<jpoj_9)yXtyaJ(|bdttR?yhtyS=hk3mQi{)SY(EEbP@k&*j
z<1C(Uf2$E^$7e=T*!_Vq=8;wJfWpwM4$s{(A8t$S=!6RhJ+js3s-#~wGT79TpaQAp
zePZ+Us61-Dn5Q#?2B;jnePAta=9TMAu3;;7CMI9m5#z*GoFS>Ld29}3&{nrc`3<Gx
zv>Hh5LV!<<qEL1m0eQawgyXRiY6Us87F|bl?e*uhsft<O?PH$WnRNeTD}|bA^n`4>
z%29g*|Gio&%3HsBX$bCpM!$_g*#VM&CWaqNQ2=E*tARif)!e#4%Syp)F^#b7{?z>C
z^WU!^>_sH7L_cwE<W_~RP2iEHJSZ+uNG}A`L_orX?Cp?0^&ZS^^X9YoEw$n$c{tI7
z1t}SWL^L#VYOyw{Cg=t-HY9MLAE}P!_wGw*HogFAXWgtk@+8r-S5y7--UB~)9{zlK
z>MA&q<x)FK!-6z!oT>m{Z~7`YAAjif5sBfSV&~7~?jiqr2!CaG2Z$`m8o4LNL1$a7
zJ9#>lkFNCFcb)0pc8uD}*!gtI<yaca;`_vj69Y74{$1weyPpqAi?)5^iGE33o23rk
z&UUvQ<>mcCscL5idVDK2|76W5|I8n*rg|}G9M)d%fruO@nZVnVd~s)03t-Uu+L?4v
zaR#|$fSgxie~)iZY;K93zw8&MbZ$xBr7@Dd&3o(=zsnEIMnq1k<X-DrqUii}!Gy#t
z2D;JfxcZdSMCHv1q7W&`Wqga<x<19k{lKkt7|S6gqb%jCikYRd=zxyQ$1Al;!PnaK
zBt3x2t8_&7$Ns+^PAWunc7;O9m)DRu3p)E4qDV-g2YZlWEFK(e@ji2hSe&s1RY0f=
zWRo9g`OzR!5lwM0pC$UdHbr~C$_-=a?}AI-e`3*x)fQWS9RoB?&g0?uG1m-c6yDOt
z8t9`8O>JAm9n`+T`DuXuty$NS!J>wnfo@bxgdoWPx}jJhb5Bi!83K3-#SIETSlp7n
zzeuv1fSiQsaM6=;)rK!#!2;8ZoOZja|C9MC2i>c`5^iU0rzT|P0Unwx=t04W1-n(k
zQv)hAFq+*^stK&?&<!P_h01imkEE_hOgKpWLuLvC=kouMn*Mz2Bgp0bGDMu)#~F3r
zo%C9AHjM*ZC$>KV_BRfwNqpJ!k%}TwmzSO7fQS3N`~93q9#0OK@@k#!lq@xeul^$0
zq~bzninFL$S1rZ&m#JcRZT6N|JmThaw5IfK{c@;IruV#r4x4Km>#uH9Vn3rkWEoip
zFr{3N{y4@xAae4RcN6-wckdGmN;>g+Q(kHh{PNT&OEIT!sWQ8)vbHg#W#YQs18b#c
z7oT=w#sr2Zrc?`WCfA$2piR*)6K6j)TN&Vy1aBZg;o$+U9u*XHvo;MJo@xEE-P`#Y
zb(pWK{2Fy%{WJo7#W8Wc@d}sEd`5n4$DaK#kd8mu&)DRC7k8bPypR^<uWh>m+2!r7
zOK(3b5t$(M5x1MnQ0V!NV>)yPkEu<}G~wyEg7m4grMB;xEEtpvj1~bc!w4H`*$pd^
z=LJ8>#$nmkCtgS)>wgS2Y_7de3FenZj4DNn9+KK5oyvm8F@-3LeeiatSJ1JaORYnm
z7kbY1z44LabITpmiMmTgl0hx+G@h+J+#G;WxzJ62rbp=0oXWNob4!}Rg~a$`RWdf+
z$m2tRijTDKbUhdB_4>aHwD$FU{dqv^*L<QATd5%6yoFwmL2C$~`C%RDwOmDzZ=AEc
zlc5VQvuhDame8D~%2~w@y<ru`&guf+A@`N*a`%`?1%?QmkC^K9cLbBdO{ZTM_p;>^
zrrw*}DTH=wvjpN5)cLQzkl4{4UD+%>`|s3gFaF+reLq{;`}(UN|2wWO>a7%o5f63I
zjMmH8b;i>|%5H^}u_-ScM}c<pG{+soQ~b@#l%s&UtGG2-Y`Bg%bez-av-6ccMcMe?
zRt1lwx5hi;`eFT<hEPa+zs#ey`jX%8S(c<K>l>Trng)0oxxg+da0d^bZ;ODGS=z)w
z2V(1W#3)c_hA83%W|OlT4lcn2&FM;t-yB6d;$!R$N3C;;%K=X3a~3-|<sei!ds?*$
z)d+VuF<L2DU#+~Tz!_df#f>{c(s&x-s~_pDeq57gLLApzLuLmK_J-?0phF@;Nl2!J
zl@~o+8C`phA*nRdm`j4ZvnjNsY_=>p1<@iW0FEHQiCrVBZzCY@EDJ5mattcpVNUYk
z51ZV{NCo(Ax+|FZ;#GuTh%lS}Pmc(d5hheoZ(#=AfU`j?3!wzG5Q^Jqoj?pDgaHm7
z{PSM;YSWa%sswGUc~>)|*VRW1dYG0kzmgB74v~#U4tl)_*s4$WuqNB|AMHRL4)7Zy
zyK#nH^8`-|+3W+SwniMi<n8<QHTdxpe)B4rWBT=`B}5t*Uw7EqxxtPVt8c4D$SfT1
zHKyOSRo151H5I2d<#Kbq<oOTPGTbf7<A=viEia@CtgdaEaZ8XCH?UUJ(MJ}@Sk;hb
zCMK_l{AK)W@_W#IotmBZWTLDOB+tIhJAV_dY$`YMspaw3r(zx1m-8A`EvSR(#9Wkt
zNF+Ps_01k7dE_ca9IS^xD&`x|nJ8#r{*Dw3@6)~+jZbTE`I<75l?K^pEUoKH$~vc9
zF^`^e{|WKpOsN+9Ej6#eA(dI#;T~;q4o;7KmtIf1hW%F>$~mVSTjGei+1HSK)9(dO
z09|!^Tc~l%o{YA~9=-)_Fc;fqxRmqS4*zn0w7Zmo1QT&jh4lD5m{cfRUuT8BM^Y;#
zCT?}_-W{28-?*eEs0RN!*;nFF>cI?pi0l?N800&HJT3g>;H4Q;WlVztL0=4lay-BD
zEnJ%}j)pcK);efVcvb1BZ_&k*8UjSV#KH*2m1ptvIwlGuD?<;(h>}>AanM$IwMX_x
zZL5<P3n8;O>9bO#sPivH#WmM3`A<jgebzO937Km<J=S|>tE%D%%GL0WM%f|XrMpnz
zg=6=fCu$E>nC}gd-=??EQ!GPCS9GuOVA!=AX8#b+7F6+|=-3?@U(Wg`1uDFFK|ua!
z?0`6EN?maB-FCcZR!T&-SVXOeeEH;KTaD7b5V;HDB2xMH{luLy=WS>oyQv}f_rfUh
zqW%gWF)H$ggZ^sy$|DdyY}3<kQZJ4@z_~}Y*XD|6#)-59h7&*f6E1x@W3Ij(>h3B`
zNPsP_21cARMnatoLnICk+BFA|t@u1RmIZJ`S;JW{Q&$dmTCM90bE|(+7t?pGLwm|Z
zaI6uE$GuJh!r}$sa?7#E$9$5kxbfg$@B2y*{a@I>?3Tv{wPbXoYf;cUN&cHKn=%2L
zNo5cEZfTBfj4BFhgMZBsc|u~UKz!)?hCrq(0dCfSs_U!T1|F2reU@O`1VT<EW9&m4
zfT6n5iaWXZY!-+(0B8!SOd#?$a6<?Fe{(63F~tOgV1R21A25F+vHefV{hzP|vB{Xr
zs9p6Z42kN;L%#F#q&&_=U7D&S<{5NHq1l#k`i><G<mjrmyYo}^oG$g{9d{j6Ok{lz
zdGM1)#L<2IGlaH#i%0Bi>+v}{SC+92`aanjR->Gxj12t!l@oH~e2&*tlYu@e#x!II
zF)O)@nMlynf}npUupMW61?OgzSRC=JZ-F{B=~!vIZ*?!9G`8P5rkYju*3ddFGJcGU
zy%Jvf#zgFdO3d4;;wzO~#h<j@Ja5{d8|j;{)`^T{7`dk!q*OO=_L6#^PMoo0xW#KY
z5HOlNk#ml;L1Qxqs#Iszk-SfhEwMSn$WAQVd1QVyXkJRfplbd5RPuq^vxR1hMU+ns
zMoB})Iuou*ms*`sPxj6PY8#!@ixs^=@HV$TP&27fY2Pfdo6Wp`%aK1|g)$l=SGpM0
z-S;7GH>QgERHLFNg4FuV`=y$-<97<?Mg|5GEX&yorXemdrRjUzk-=II80-_7_HXdZ
ztkTQJl)S{#8?NSBWd!uxH_VH%-?LL7QM$L>g_u>Cp1lsZlW0hpNkUV@K0n~5b&xH-
z814>+xB+<|F<WOPqU@~qyGa4;s3fEtKQ;`ZPEJz4IeCH=T#VA;)*aT`XLB0G7~+zm
zPSFUcX5xv-Gcd#~b)m!fXy=VY=O=!|?>zk=7q6wR)NX|9OY?D!8Ubwk5@+V=m4E%5
z+_kP%dgt;AJb`%r=uKJ2_2w<xizd1R1d*L9Jg>7DsFV8v@4VZV@Sb!kwC-8Qz)|_v
zC1=%F%J_#$Go29<WbW1QJSrivdm!pkx`=h8^il8m<se&Eg`eH}6-unC8&yP94_PE`
z@PnIgU{udX%&1)S!x06FSzQ(3oR+Vaa+jw;1HR05JDe`&g4|ZNtn4k<Ph0g7Eqy6I
zRsTnm6VBuHBl-ZDRAhwo396c>;tUptb_*9i5ZO35h+Lp`ATx>ja=BP5Td9hx>D^0q
zZ>g-)4NX+ilSMxPut7E-ImM=DAnA{uyM{0x1@6jTs+r5$9fTzStI9-&2^wI?GeiUE
zQh4M=)Z;f;c#Qs7IJyDd3Ndd5j7Kd1bb_d}0>3b}Vv`r(+3%)fP-NUEs8KKW3O2fF
zUD$$xCUT?1-5?@@5*MIALRbjuSKO{^2#huUn;17zW_<@>NdTto|1$&@x_3c0b;I$-
z_*h<KiLXAbUd7K>FW`ac-J09?GlrAH=JFqkheidLJgFekCClrO(>0}w8KXn?jj6F5
zZ-3{1WBw*5Kee;hJ!N#9)}A=&lZfkih7G&o=dQ=;>N^)*-fS_-4GpquCcaeNO-v~)
zuzblo+E~1GF;qt3T2G{0$T3df@fFe7$N&1CV5jo(YcrGrN)9HH24}{49UTijUa}k8
z$uq?}>vFuQ<a?RH`(!<J1jO<$ANvy3=6RnQV|2qOv&>F)s?u^xfRkjaHRH#K>-G+x
zJ!To(k1^NYpm~G-O0_hj@tdu7lsIK0_ETq%2<pY^{r8u3jRp`YcxXYHDcwb_{%$m$
zfja(*r0VLKJQH6d$uYM2$5Gnp7+YT%Jz>3b3&y*!)%MsY>xM#mE+nZYhe^oMj6Pah
zjc7=<iM_MUG-wB1J7(YWAc#H3WzDz5S7I2}Job1Nm)EwvR$bjUWau!-JihRFJ>kX&
z0h=v*cBfaY?iY?Yti?5i0|->0;K5;Eff2iBXmylq4_NMs-0_j1Q*=`gH~#bqMdkg0
zif<Zak{;qT3`E3SBrtdO$=KFz#wHuynJ_wa_T-E-&hjQ<Qb)vMn>Qx7VCC3%tp6i=
zk1TVuqM6em+~>C1dF>n*3H19oR2pFK#+=#VeV3xE@Uqo&M^c#nW1{(QqG;aveBIh-
z2~W?~t_HP4Tl6oCNL~BRGghqhY_(>JH)v`Zme3En$-t1j1vw*8F?&l#g35M}rx$2}
z?=<cI)krBg^-8%+`%wIRYjD<B_r<a|BV3V^i^!*K8fE+UYUYVyxLeh8dg*SK9}4d`
zV<%8|VW-<iqZ}AJ*G(j=PVhu~?vG?r+3l1zBsc5~_8n%+qa$``tC6N-BWQ_x-I3Au
z$fh@#u@7c`K{Uzz5+e*427+AvACc}wJM<C@4rC!~?{HkDAe+&HKmi-f*@QMMpg+A|
z3MCI5?s49yklNDN!<5|85Xz-%JNv_cIu?tDjSRpHspPX=EYfPry~ns4h3-!mK=W>)
zx$oN94izc<h?lg$Y*rxjFW^LIYV?bU99skcjDXl3y59%rzK24hqyUNof)R|tL5dV=
z4?u8QFH-zIyn?!@(1?6xBWk6GD0XmggPpYznHvEsM2t#7!W4Bdc5$JG|9K8XGIk@U
z{uyBZNw9#}ZZ74L1aR{7gg0JgsxuV_dbPyq9uFce26w?4g4I5VI!>SQ!>22lwx7KU
zYuGZ791O)T6`YG;c0}z9JRvw4n@!+B8@M7f>7O1f;!#f=KH~1*z)I?u&^_vGmG=Ge
zd=aQv-Z^-5mFtl{)mW`r>oEsm6hF;CIAfL=5}x`ERwU28+HbC^!TGFW(@U0AZm+*f
zDIqgzp0>}1764{k$`EFCWtw)Q!auxYLo%Dr*lD--jE>RlAfN2mEQ;;(ux$Zej!H%+
z*V+Qg=62RKQ0Garv^_m{C(mD3H_koY_n;Ls?o)0>YLEum;U;t}<(q(+-+R2J=4FHb
z3pc-CKlqXyX6(yY?`xaM1pU*m8X~{FZdWU&J){i!8T2h?{4}(yMi}oS_uW%{JHN(W
zI2`ogYOTl6^|HA@+So51#2P;GvevrEvk;%WbWPW{@8qzx{`?)WhU(Od0Xk*VR}30F
zWYBqy`itzJllLfQhStdfL7MTnmI0!lSl~5I-vjl}J&J`exYg3z7mYkuLmci=W?t{_
zr(>0?dPxIwY*#Pv>6gMz#8UKJ#w=AV)*<{02AOOBf&9Dc7a4NULl&F*9qyO$9_(1Q
z4V{koI;p;OmmyFJQY#?N^0B+O-@Ba;*WOgP9MtuV$^?!yk<c>s<vmV43@M*G!Rh#1
zFwIoo)w{cQ*DXWn-(E>e4ml&k-U4Zq=^<>RwdNmSzmq-#W(Qy21vZoy$s2ERBB$<0
z$)j9I&-Yo##I4+#I8cQ{>u&E_BwrI-U#P+b;8(`&qgR(h6pZ&ipbeq)=%2d|nuOax
zmHr|mH{Eo8*$d^39*C-{nlvo(zKuTirTNJ6M=3GA@x{)9hbny70A}g(017s2$GQ*$
zjHfp>KL0jk#9>4yoNbzn6!EsSV~A8qukmYH*?(q#4PZ+r0tv<{ggZbG0~j<i#K8tT
z*e@2=#wroO-pD&_rsS0|V}Duc7weq<DG;pXxj{@NX+WGQ8j`2FVxy3gGH4yukRPKV
z7Y{aFx|`xZCwCd86kQ;e83(&w)u9hKNz~-CQ6G4suXIp72_0H!s$z);{{$W?MVp>f
z)oF#PfR&Nb913zq5dCLwTgc-8IoAjDpkTyFjAap-L%SDwI)m8F#KB$ha2RM!2SKn3
zQ|=#mtv|ggfF6zhH+03*L4GX)+!#O+0CByMooiG0Osp3}viwQuQ}7B}$?4RLxfq5R
z^bycH^|dG{aC-A5Xv)0Fk=U6QvHUE>LQ+FOS(t}qba5;p_9H(zPyYjX^2<p1*1n>I
zm+$BYdzNqEioDYkT!Krgvb;v$1f|%Ai8-!$Gc5YhO6nmF2B!r-Rk+pfgPgw-fX%%8
z?S?%Ge!gLT4)=1oZshyQl7y%Fdy{Hk8<9V_pD$ZHg_2S}5nEsT=-_DPqD1qAc`?{_
zuyi#49_MNjqiVcc?UG^8Pn6N_vgrc%m)i9@X?S~Phe}@HRq~T!4^MPugE+4sILNuo
zZfBWGa45L3&%cVG)_<nS=@Oe>bSHEifHb6F#kpCSxekL-`L%#=jZ<~Q(GBAfXFg*0
z@+-5Mw)gBg_{bSW;(`04lkbT249<c5h){Sq-{JB8pm731I`3f(U_FGmi{P*2sbujp
z!y1lzicG3zkJvZa8=dsY!=9L|So7(xyk^EKr)dSX>`A-(12@18I{7R-H6ZwCLUi|$
zo$W=J-B9{$A@wzk2OZrZRkZhBzqSFG97;#W{1|iBd=P=zQd^XF;8GO9tL?o@8AD=Q
zVQ6^Km6KF4^MB^KUaBdNsF2=G-w7`v9$xP-%CgBr7crSsNIG_{b0KmwyH!>7n{LTC
z?Rrg44w~y$QP2@-&)H(~DDiGjS=*cE?0-7s?64KT{7^~*`x30a%td^%j&+<ic(Lu0
z>m7!oPIX(zJHzc%F@H&=WDB}mInC%VvkYOlo)Zrhqv9!&gSvapEB9APNtrSy#w-h3
zjLI9xy0{g4^*FKEisvSlAO2}CdEP#8{6LW-J-T!`$)`B3GY8ITZmVwcg&XyZm-q)@
zH<?p=Rd^!Z^VW(VQ1X{-O>uN&A0cx=4X^CyT0Y$1Pimam_a8{pekb1m)9Hd0{=pC%
zpKfbWnBNN^!<Y{6ql(yiZO|t8h%WDbLQ(Ryqhf>yeWO;r&D{PDIW@f;{2(RxkUcq&
z9aBs_O;(Nz<B_BzSS47Ar_?-poQKWjztiiVu=fymD$@U!OJmW3u+=DFQ;^X91X7O5
zJIf3FIv@Oq>JwKcf9Y+36FIu`0Z{i_74iCV4xt=Rc+QMcS(2R)_?q5)KT16!uiQoT
zUyl!5$gST&t(6|s0)z{1!pR$ei13#gWwTkK8AYxZaTj<Yf!7#_hFZ!K>ZN?*Dt)~&
z@D@pcTOjWXT4Zn+{28DM+gdR&QUr6*01P6o)$32^`+wS79Ke?){Q^zmU;^BCYCG)&
z9u}eQIF5kiXp#*m8UpCW<fiN<0V(0#eRC(@t=0?4N8GD)jVZ}w@v;iO_jkJA$@69M
z7to&)>{&5eIW|A>$IEOSj-Qbx+a4<9u=18ZDPvgSjs^k&+uubTUD;vt)*E?5myl%}
zDdXyydSH6r?^}2)X(BzPPEu8k=(ZNBn?x*oyeXN~J;z=;7!a8^@>SjO;a!mkjJi=Y
zH}394vR=4up_0VDMmqM@L5{`wWSF5&lxn!(;_x%l1#`VZomOJWqLVpEhkP@P+-;h}
zjCu0-O;ECv$3J|hWnRyIRAg5B<<Pi+taOc2MlkQr#|Zg%xHK;5bau*KPwn!&*@p(X
zMYB`1Qx9wvT`xQ@xEpgey?m#iFTsBK%CWMuBjoE+rKv%CWw`p}`yK-v9kr#5Edl?8
ziw6I7^Aq#_jM-c){TQO4=g=Of;|#?bqe25tq|NM)$@4Iy&3+fZ1O)la5(q#+v!435
zyfQ@P==1#!la?wThK3Txp58uV{teRZ4SfgCIyFlc1k(fLq`Vmbr{F~8r;j9}V1lDX
zce@K9{+w0cQyzFfCGi*Zs2=Gq{O&KAr=>(awh2ya7v(#8F{+RFQSUiu<bUJ)v^8%t
z&DF@i!;nfgUuTDA<Z~2t*D&XCHWtszKSZ>n(9#d;GC-~()TgNC?60HPwl?fNm!<Wx
z`xEn4>C>lM2sy6|x1N&DNWA~}i0%&oqdpDEps8rAkZO@?A2Acj#utpQxn6R`Huqe|
zj9Zw={B;g5pWsjdBT0BD?GK#GsNb2Ubjcx&Wyv_=Ic9-3_I~o?{$ED3(w}#>4<@&c
z%MDo%$pGiO+o4d}8QocYsZk<{I#5)2DI+L4o<oIXkb{SOHO<_N7NXD+MOi-KHdT{w
zhol_qFg`Mq>Fv5@e0TZV$fWECp(}6V@l$eC2SRij>R{*G3Cr1&quk`^0J#+#YLpZ%
zF%E`HHf%AZJQ0PM0x9C!MmQaVII^*J2B7zJ%!>!T2d{V!ow(>4F-w@b9#a?L$BEHI
z^Z@cG*^&eqvS%q$BhA-sxTBwy+&YafZ<FMj_BojyYtkj5XLurqr^>b()RaO=NNvGY
zIJ!@<ZxtK!GD&6aTj0@~Mek)rjb&)c!Rz`J(dYr&B&X%0Dv;zJaqY&C59U9q5LcC&
zpo@!q=#CpZZisEa@W%53R`f9zUm+92R!7OphY?61rlne`l^QuB%Ya`Q5K$;YKSGiQ
zc^*wAZLp0tuB{CM!A6)%p@W!b^!@;sFo1A5|1YG4gJsZrM2OI2yEQzRSOO?OjD(~x
zYwl7kqQhfMj*m7T*2Sf|Pe+}ap=Z;h{gb-6*m{6d9n~OysTaZ9UX7+kM7da1z)l&i
z<UJF$vJ9S0p0>82H0rK1ANN|v4@mC5O00%czObJi=j_d{$&THz&iZEmiu^Ww<hhu5
z0qG1<)|pj#@x1BN!D~CN-a$V;x6XN1z5Q8ZMBA9BlFrqLCl}rBHcZvmm!28&3@vxx
zqdX$UdQy!mE<NO;JiqIN&J*U*kgbjAU_|?3ObFc6B&FmI<YDkqX;n<b{VZki_bf4>
z(N%UAzj)N-j=B2EXk0^>g;G_C&znZOG0_lhi>i8<(aoL5tdj(HuiR+m+33h-X2@+j
zy53~*Iq4a^zUkLwAvuGiO*RnZa*XcVcF#JI#xRT{`guR+S`6nwlef~p3ETIEv08j9
zWNY~ee$&Zmu49!vbEaj~CEiX1H{wz<=zjnAUSp+8x^jl+X;Up97rXnpCZ}Fm$rt$Q
z!6HKG0aD0I-pCupDW46Q>)c2cxid_b5cy{U1}7BI-DKJ?#z8SUEK<nKx;;Z|p{D)p
zlOhPGC>$rkK)Cd?n_317{6it}!n+-SB)V2!n-cQRS?dihJTel=4o{bYT=CyyjM<`d
z!AZtnS=xRT+wPdK*_*C~*+M7@ntOvef3<ZlagOPNP78`#^Mzha>^*3c(uE#F-Gt5X
zRKck;IUc@OL?HV#m*L**cy#eSBb|oxu4V4B!Mvl2Ifz0z`iD9V&$)Bv5YQu{-2M6H
z{{Ww_umU3+|5Bb<8b7=$&}yn^NHCE%S2Xb`<ZjzutzmL%K8E`7N@c>z3F;hk-?eMs
zRgw(YLx?AUYYuaB;tUX=+4xkj2o023!Jef$z<NT#6a~K53z_>*M^i^$mrcPyW{t5;
z@R~oE>bPrd1ZNyHds1zd;9~z{=vneoGkvvS*mx#y&8vG-E`h7cMrx48F$&v_{2M~9
z%@MGYW|A|fjM{dVLZCH5xmBPRCfYCvu}8KByMVo#c6>Y{M>VdG3C#f7kWpr21MhIR
z(E!W2iX0Vh2#}+NYg~(Dg`lqH26<QhMa{|Cp~hb->YK!};Mk2`8GML(ARsr0etHl*
zb0h7B3NhzD{cor@20&L}Fa+Lm@Ya-|QbVld_ar|+UEXp5KxImy@5{Q9olv4ZdH&Xx
zW@5fp$KG1H7LzsF<xPj7gn?=&jwPpg<RXNcz$43Yv>NP|v6>G;kD8Yz7`l76#kzXs
z5v<Y`_9?2$+=^}fR`AneF6W?sL`nuWl=(Q1WD!tFVpohcI^P7svd?GNo7yf}INiLi
zc(cf;5A_nsK0$9iQ)YA7)0_3i50uF+Tg1^Y&(Q0o#kR_t4@(n?#P+m^MV=&A6H!nb
z3~RK;V{f+N%?CVp=Shzp=ahY%svK2Dn@v`=+UrHz85vrqq!W^k>Teb8)+DBQTxPN%
zKZtr!O^0jr_HmVm$u!+(Gg8;7ilh%E!#dl$8&ae%owB(UfSbCZ-sBz7PJKfv`OxWZ
zI~F*in-!QsCnENZR0p3jWIR;khLpHx9=9gz;l;tGQFK~FaZk8`n!yrjN2g|un2}j}
z*o`=MZKKXpq&Df^z)yAp1x~HQkORG7UaGo@AXup)hL44;dW(zGc^@i~s=hm|@5gDG
zV)GGlcdy(jYd@k<YN60(%)FF7QVW+d6ix+z!ZHEOJ*W_HWB_a`@rlKUx`$?@WX7p%
zH-ajQq_vFJeV8sBD^}I2tJDIKEW|5AC*Saw=ErH&OuNCqE`pu)8q1fD&%nKxQl}>O
z+UU7B{lE`vh59i4uEjQbaN#lC1>~P_v_jn7p5}8^L<@_dokr~w?~k#vqn`~I9Mo7$
ze~5@3vu6V%d41}ELc0ZdT&C*+7!Gyar13QGUa89?x-C)k3bXYYQROUcH>FrD%+@!L
z^sg}8WJ%{zwkxTkHjSX6F3Wy=_gLG|a*N)TIQz8kye^fQ*S3+m8iM{&j$GbtLyd=n
z&-1&k6@G3ipF2@R8}uZCNhLx}r4)3#-hgEEZlk^JsiPpk-jNCl{+IW*o(H{T(<oDJ
zeH*F~8gh-d=Ep$=I5RD6*nsydu-YkzdSyg2c^T5>s5rw{S+)1L%5?*gqRbE#tHzAB
z4$Y0v6d10x<=<h7JQBI~R$T^jiw9o=f5vv9B4!xBIUb1DAk7=lIa(X42zpv!lQ8`N
z+zp<HiIEjB|1noY&rUNW16E?v%SD<ETD_6!f>OscpsLJw*8*j6Fir*FXF^O0|7Xb^
zc19V*=6{Dke*zRVyTYbe_`!9A(I=hHFyX*ANder)5G}Ppb@$A@Md@_0E=!$#+!8)B
zcGx?1Z9?9Hi{2b6{d^%8=qH!j!57)IUKlYbqFBsu$m*G|`~5C;iO(N-)FQJj?8O+b
zlpCz}u}*Scs33KbSyHv;3s-m7=e|(YVt(|Ea>h@e;d+qodnLeuLap|I_~Kj*b;(cE
zb>`Wju%vU}S{|ogYPF*|<;3O%BBi#{E1dz!<I|QdH5rJY`ds2t#k?}>T+<Z=1$Y`O
z-k8t+I4XaRlk-??Gr{ZP#3^Ofp|Q9<#@JWa&&%Jv+Vn=5KpScD)Y~_zeNs)O`ht#j
zi7n%us{FfRtLJ5JcMSGtd?QXL#G3`}dG;}C2X5_)lBZKo39^OwkbPh{9q;~PC3zhG
z2`{EjKKZlbyi>*4%2L|B+P~sf{W?W{4z<5Z9Q+t$vK4OaJ?G1PZRqw$(b!`Eqxgws
zrS`BUwOx>qOpf!l3}4qeeCu?3sdC#v=U^H8v&pDP1<zfNYY*CFOQSZ-eZjsD;)Rbm
zJ_GR3K<A@f4`w~d6FIL{T;12Wdv2V+r@1}UH8zFgVbm`|AQgi#Z684}J9(?+1Mmm-
zmiN5_($^M?XCBYj1prE|Oj0!=X)7g4?Cdyyyb(DXcl@un-*I?D2yr%3CKyO1RMu|x
zp6FCZ=}w-d-b9XYI_siuZc0SZbAp3qiN6D{v2={So5_UFTnQ!QQ!N|!ljWG7L<Epj
zMqVBjg26T#6JX=jS2w4(*po&4GF!p2I&ASq+aWl00^9KP4ohD(Sk~9#@eO%ErmZ-B
zz-d+e20uzQ$W&~(=mcM7X61n$oL^e2{*5CySuz~Du+EmYeN**w=AoA51#*6l3{G-4
zR)izBCS>yVf!v=~2gEyi7%-qxh)c)AqjvuXQkWD<3La01d+jZX1=}PDyr2|S4|Jkd
ztN;_AO}0GLb(k<1cGPV=N-kiZr+MiR0;&vFpC-YIm)l^lydZFTe(R1}Xs?XE1vGAY
z(2}+yqtmzcv%l0y#yBrdySyT^+9$#bfCU%_TO%?-PwQNC{XK^0X1XmO@~_wJ=M^%h
zxE_W?upkzg-wDHwF;{a1D_OSUtZx_``R$)i$J<+=O9bimU^828qe<!Ot-5<SP_EV)
zVI?CeobJH#^~l%&3_zctGJnb&NC2qSg<*+gGB$OlF>fC!k}`4!yPu{It(JI?zbxak
z>Ud#W8V=vE@U_>^ig>;^%8sREzf3nXbY?)Z=v@*cORN(rywK^Z-Lq0ji4bn|5!;w#
zXJiuHBm2x;%j}YX#j{zVquL&!pNgusEm;(kTDPeYXIAp?ZobOu9%M<mf-<j4L8x5O
zDZ`+xDSfT7`e)A{c$R5_{}1Fqr<K=L_jB=?+pP`8%JjP={a$@-d3*+Kc*w<J@;WgR
z<pfMSBi?}!jQszYIuCHF`~UwRd+$RWlFhL)jy<v+$3DkAC_Cqn?9jNa#K~T9LdH3V
zV?`(<CD|+EkR(Y6MWxi;y#H_Y{ax4px>A>}E8%#5-mmd|J|D%TA)9BuEYR0a5CSLD
zl5a2*@lW&`*?JdA<|A|IK1LJlE50*MI;T5X$e}gI%({mxc$%?>-y7yyZ_?=rO0{`2
z*c4knTgl6NOm;6ZjDm|UdXH~LV=SI|i_@Ns*(_Uxci}@s5WI`zE)L1JZmXBugp7-C
zx}T+*h4Ws}v@N+jOtE<>EO<^VX18O%CuenOW1fCL42>v^N-PZH;QQ{1xrdl(lxa$Z
z&@#(kS=nns-;GwF+E!R87!5ynhP_VG*a^+x$w*I{WNS@nN>OYTeX<=B^dm?tM}RtN
z&*p9tY@pW9gloq!^R6x;X(9WOcdA*|SRvZ!d82DZly2n9#XA)+ZN=BjL;*9NM4#?$
zT!tCzTN8hM7$ojX)u~Pf*&Nf6=9tANTVA!?QE>Iai~Ne@#~(tJM@qNM9j`Uh2d4>A
z&o8Igvo@Z!?a+~z-_Koa;jrr8mTJ$@VhE|R^;ij@FMcMkcYCMOTh3#Ip*ixNHu-rv
z{XOb?G#Y{0-QXY(k7(!_544DeRCmT^(*LA%Co73je~i3EDe!hd8}x*$EDG;J1Yzv{
za=Fc?v(y16tex3<v9ngjOn-qbOD4u}4q47Q$8gH&gaQL`B)p?^P%G5l=PA-?hbug`
zTD#>UF_b=DP>QEScFgem)04MXMvMm1B#J@Ltf_!G-NcLrgC@mt802O>1?DS{MgIwa
zO$J$fILL#~XQh&~W2RLWJMtDaR#puJnvL%SRS#sLg*(#b=(Y9gmu8Z(a|QDoZ7LeL
zaE|Lzl;MaS;P!ttgK<RAiGoHEd~{^7zX(i+5oTBGEPL<1(+z2^XTRX>ODkR!%-7b+
z*Tm^BaY5MK?J6gM(kHL1E{ZFeS(2gs7<)P3;E-Uq@(0w`$toKYZhq%2XJ3Iy(kBIo
z2F7+PN!S4lm;Oi(nNFg>2G{2^q|&c;5&=m_&jUug_lV6xBBV@)y0C-LaC%k;mP&t@
zw#=6WtN)Zle>c2G9ZS-m8<)NiK^MI6VW2Z_69?(oL?q>Dsp68w-rtcsZ|VVD&LD08
zk#yFG$sQU&t*%gW_iv9*wqL#shZ$Kn1sLX|(@e*F6+KD*B{hhQK6w4}4QPR8vvXLs
zuKT7S+>}sTl>8cc{kWXBC$t>q8wM+E4&w%SSO5*=%C8uvw)dI3M@S{M7g*dGy_v4)
zu{r)4%9}@ubPXbT33B`EHE3{dU`%@?moF=@UR>~gQk=8Q;s*UJ{giC0T_<WZfoyBt
z%(+qoHU0H|yuH_mNzhN*>Y^4C=J?Kwv}&OYA9vckHg$WU=Ikx~$gLGzRwBX0`vTM(
zAi28rTdY^1Yu9L3VviXKkC<XLteyUWHOd5v7dxNTRxO-E`i^Qz`Sl!9rV~#ImXcBP
zUd5!CF;9eRWGeC+mrKGAWX02nQA*9Ha+eBj<HND}d4z4csOFGtcgL+Rvl%s!(|G<#
z#T=VDx&iZ_7PPkV4T4=mRIgm#@9kfTnQ|F7G7_uJU8juKg-OOCZ&z<x8%>FvtF6$H
zW7R4%svV9Offn+42^%h?uP>HnX1wqZ)lbugVt~hw!%__j=pv}DZ9S@+$eDdW$My4z
zbH7j&X{UXUV=QadS1K^*RS_^g$|Xq#k&b|aq&?z;Vr3FabNQxLk3$z=WlEO03!(d?
zvnO^HLF{nGubXv!x`U$mLkz<He$Nqeth)-l#f#hHXB^Fe`UN?y?-sNU)Nr#|{q5zh
z8MTH^oDb<=3)X>k-sj+66>cprMro=@^RY*tPCP?!A!xkl1+xlcrx%O&zpVLnHJ|^{
zm_W3CC0w-g5AD=ko5-ln%Trh;#$0yXcDJ9MMbxEfyrmhFnoMrG;O(>Nr^_hGJi?w4
z@y#Kn3Bn8r0jb73*&(}8tl=9ny~b@G7I5G@l>_WULMD2?!J#?rj2{&BHjupSypUac
zWR921A}nl#Zs7a(MQvNLLZG-)>iTFZB4{p+U^jt7mmvk3OLLuSOi+>&NT$&Sx3|-g
zi?SAxeh&BW<GW5$aoSOjyW=Q$Q-y;FhFoYJlnKD12bl1VE@3>5McpV}q?J%y#na4E
zy-Kqwg<vDBN12C^kK^d1XcMEPaq%W=Z11^x`}1GDl(zBMuK6~9>-IhorA^yko4;$Q
z_5wqLN%~)6$;S}^af$g?S&H<;i^>UiEaiyy>fl(yn{=z#H+X@Z<2~5Wg0^)?0_3v!
z9+_!C9S$lm4$u~ZjBH_l-~zybJbJVAA-e)JeN3U2#sfT7_`8BISSi&)>2BF#`Hzb2
zzfXWa3P%Z{!oI&dC!XxPVEljs%K<wMxq>`)F4bLM%)=Y(yyAVKhJ3w5yJb-xOkpy*
z*Y_OO+TwVgrH})*DT9`yj()b_z`mTV&XY(8lztH6=<IT>x7wpJfx2gPMl0-7c$oQI
z<r{iZNUmNT1VS3mL<f9wk)c(zEd^+4>NQx!3ePDbtrVv{xg$Thd?1J@MO||BeA<js
zMF#q?HXKk_FJcrGlB%D`BdV>*2H6-N%Vi_@xOJyAa!Af2BMV50R2`dxzZz7xo;3<t
zn^R8i*ITbyR<ej%AcSe>qM%%lr79-7(Sf4`7Sxp?X|nm*F#jfIqq;glX!4b!c{!3h
zTzv8~LQyR|(lt8MlInfaFKAm4E~N;Ur!Jr?UPqM^bi`oStDAYGY|(=btdG%mSy0w?
zKCdHq2xi~(<O(dN6Fs8rn9LQFhpw61Wj-NlOVc-v9MmnQE82SH#)&&0pK?`nlqyi|
z?KNYXk`!*xd*|7H|KbI)&5MNRnB$YC^McBJ6&LYpKFq9dT~GWrsy8{Ee*C+a+AdQd
z1{iq#K|@5Q&}V&VbX#mzN}@CFcS=9MyuH^EPG@|dIG<{tqNd@QyGW!TUpCRVI0ZY?
z*8k6D!8{-1a!^Mt6Km7;Cp=vfUR~$EI6~i*dJlBnEnrbn6f{2cxoYkvT%Gn0_TITD
z(M>~`9bW=cOYtMxX`dBD*mz~@!=uIyv{+%9@@<vkjrqHZkjZOZ4KjK5zg{_V=Dz$?
z#iTB|mrD-ce#&wDUWwxiL^4b+#*gN9?nm}Ds8H5Yz?4r8=-D-BIKo55pjOC~Zexm7
zR$M`VZcDCJmwJEODW>?wS*yqR(Wl%^x>ETSE`j+@!M<-VJgQZ%Zn0VmBsTuIt8&ht
z(QtlC_g?hxv3i9K(f22s83?ShZi0ye(_U2OA%n~KnifQi8&-uRA;N2LL9V4jW85^A
z{WpCFI@C|^K^i*5O<C9OM6(P`_P+gKb}s5<#G3;ZE~%+*MVs%mhtKjEpNr<BZ7(QH
z)lN2HwYi)eA1Hk3;k6WE5ZXBybPJ8u;ta44jZe_=c{78C=_ge9Wfdf}T<|JeUC4N1
z)ZlzFpcGN;i@HX@`dSU#M1oN&$-fk$TUVHmq{BOO^qVPa9085~=j)&91FV!<l@M9K
zqqf{n-zB9zs2Y$NNO%*o1<}tGCZ#dME|0C&(v6M9&n4-CI$}Ix^B#+o01iwD1WY5r
z3=2RnZHi9SZxxaycwuF#o>s!)P{^-5+sX5r_o{cyW!iFA<u-wCn}NRZ*ZjY*P^FX3
zCTXB2fzBz=OWmK3JbG08h3Iztc@7jEJr4+g4*`sB$)Ms-!yu1hg7Zx>`98YwqwE4l
z4<)2X0X<+TWhtS#6N<JV|H;upK|rYwIsls^C5bWkSU?%(k>XEDLe!M1y~PB=@~=v_
zS}G^9`vy*KSt(zn$)_ThN+hqY;KpBS+e=%#L&=^l#|);<pxd5`RA2a*_YZ@!;mrq>
zw5#YuP6_c&qdRL0n(0W|wf&hbk{P@3s}Q)zCv|ST)aRz~2CvAg+sqAvD0cvD{)o0x
z`<D_pCL<_=C@tC8swuWoRK?G725>n$7$?@$WQc@tR2Cici*tfHp)Ck{;)imm4|tIU
zuB20)r{b(7<`9xmI+yW=7hP-FdkIh?dtL%_*)64&Vs(c%tyKNjt&M6wniTa+<a1H{
zZJLWRkhQt=!v@_#TBYWYVV<=it^I)kbYJSx3X7mkI$N{Uey+TLV~HK4F50xnT2W{h
zCoa95HW70(d$b#R_z$+0aj#aK&~$mnL*vsjX;sO#nw<F-RSuUUYKFfvMVTN*1H-zg
z?^=kfrccHb32fTGJ!?7drzoj@mxhUPbD>0~?{&xAjeoR8AcGMQJv(^iD9n{*l(D!+
zx#JbfJ=blK=kk<MEYy8`-b7Ne?R|De@w1)cepP1g5$9Q&)g-&`rtfMLO8EYR+U)d^
zvz7AE_$YpD{;j#Yxs|%!+r>P7Zp&b%A6`vAZ$D~gU4hw=#z4{+(*N+S(?k&C3nzwM
z1^TpQ+mq_$gj(5-wsMVphm8(|h`?gYPP;%%!#~#v!qUMWqhfBc!~1D-Dgw`~v{T*r
zLImWT3m`4G9NPE>n;sWvVw)wSO7go&ZL0d15(yqL*yQzm$5@r#<nmNk@`K`GO@i-d
z%@($>t#(;-w4F_HqL%FQb9agWhZv2U+1TDQWvDhgQ2bu*o)Exp8c0au)@$@)Sh`dB
z`PVB;4XZK^@%t_xLD}l7qRVb}+^ztT=SPD<Mra6yvrN_BM3=iS?rCq$=9Z@2{(6-w
zT17|Y;wqVWy@!WWvA)FZoszoHo1)2$FMCd<CBDuXZ#pGO(fZl7%hTf%$(4!~&a8=;
zZQ(^RKt$a{tizKA)Iyl^cra*gT#|nt7&-VsmrYbu7`Q=pr1s}0|9-*FU<bTzTKmUP
zErTaaZIOQV0Z3MFn(7jTb8JC&JEfgPYo2Mp{8M7?o^xBH#Gm~E@E#zG_0d6+-zY?L
z0o^K?zYvL#;Qu2hay!w9skrA}k$-&Lt@pAHC9~t*_sUNIIGx5a>edzM-1jH7cuE}2
zZ{ZmcDLm59QVIjlDpcQzF@8w)sCWQ1K1llhyOjDf@&@lK2zP+Q*!d9NqD_AV`oNMp
zYifX;TOBP62*=OPxsgDqEYts2HXx5g4u8ceF(3!GyxBWvP5g6GspANiu;o2qLPD>Q
z;+^doqWx1(llBfqTy68>s~KOUaYs)X+NO@6Y67$?X64ISNvk|qH#lu;#pyptM&sgt
zkQkU<$$p5%OIfYbU=nZDoZfm$z8U5dtmO@K7M;kqRivdMAfIB=JcaZY(NGXk-@w$q
zwMkjIB|_=Ratr>V6alNbS?I#*_b9t-!v$rJ@DgJ~%$tl!`?~MAMTjKSpLLawagT7O
zq8-$UamV+RSl{$%#yQ|>^?TEQVzP=RX+1pI$!{IiQ@?r;N@7M<GdXJETeTW|Lx>>f
zuIv}K#9HFUhv*@rq8z;Vn@cBpGB$qPU(n+fw{&r^LwCdKwA8ZL!`_fNavts3%OA%{
zdxoS1xgA#&%19kT>ou5tuflYE!ZgyW6WJ->`H)@b@mYO{03%`Mx>UPKOp2nMH;gd^
zcf*UP=uPTeM{qU>t%#)tXfR%JFHNT|fKd!kgJ63po}v?3;(Q4K_FTuqSo5nagq@e`
zSdQb<L^JE?Ot|SQ2#l(5bvs5%mnZLt1e1p^0rFa_La#|RXVw3T%i;y>jjqSljc#9q
z)K1hjBWJ0ckNPEOYg?r1wRs?y5~1A3Kjy5O(P$!Gavc9jX}m5{o$mt)6Z>eYjS!ER
z7|+9}4IaHN{?CgWulHBk5tqU%j;;2_7bS$|q85I#)_sujh3|HjmECCQ(x^Y)P>O!R
z#NglAl%?+~vUfZ>A8RRJRiF{<bH~*PYfrowJrRj=Zf`puaC1sw!zIt7gK=dT@>}tJ
zt<T+$sZRw%{rZwxo$W+PhU_5)gAsc_7I4f=l03}iTo7H5$q?$IF<3x#cd~>;FW<iA
zN*fD(oam~eB9j7_$`jPvKo|6MllNP^c3hW=yU<3VBHWCLv@7@k5#6b1PO=S4GuofH
ztDq&1ub~G65?0;1R3gL*Iu3D)MArr<ov<+(&IjrZQ4|+MpP1;xymj+Gh<Ky>BA~C$
zE>-Be)U*dxkxA;n*UBoRg|+IJBMPX6Q$ER_M_toeEY6>Nu|(Ryz%x+)J7$UHF5@_j
zeAOuM4+89g!!TTmoDH*7uH8SZf;1{H6(!*W$IGb_Kb)ji^s_!jIJ|O4|6YyEti`;~
zg1HI5zw><TH|5;JcMlz>YYOel5)QudxFWojF$K>SdQJ?dToWVs`Lj6SHIDvAQ1F90
zpWW^cvF3<cz(&*r8gG!8Gz;RIeP`&ScR<eo4E)KEQL5{d9BXSXrZJeDl_v(oG*5tm
z5d?jqj%FZ73$i1mBQV+hH8_F?K~ZnU4iKxhzIU|7aT}n6Mk0e>CodOBrjL<Azg%;8
zsj5hQYbHl|WsR*>1Jq&`Y5svuNnvN3>|-vk@N6=Y4JL`U#At0xCON0lyx~Rzeq8lf
z_LKr*OpMD$^e3$hdGBE|+6s^IDxq61?iOVJ#a_@dPt0-s7VUxx)ay51uNj@j)!aNb
zc?Qb?KdkcwQZRLvBz*jqK_tN*HrIGIltWS`n$Y}^rBepy7I#5x8a2;ZYQ$wAOsf|i
zq8q8M3pf>WP5O0Pqm-G6RH3yi`wcg~XKe+RZuxhjlg$JZLae(k!_9fHjbeEG#;h?9
zwo6#8X|sD9Xi6;=BVC?w;%kC3w;w~w3B~Ra>o+xPS_}b#GvP)hTpBg|6?s9>aDMiY
zmbF+=bD2zjiUyOrOsx~4soQ7f(D#kJAj6v82rA@F{k*%V19Ms4hY%2t!^zl4uTNO*
zC<4rj5v*NE%5UY9rKeycMj<AHV5kQG>C;THi9TKTI$asI!tK#81YQ5vI<P=0wnWw>
zi81N2GXj<p2xPIcgrFDBU0tnE<hx<o9eT|WR>m0tzGO<C3kdd*0;$^Y*K-|v?}bdW
z*1o&6QM}zp*Juqif-Ez5l`EE$BySKJUK?FykV}qGypobEq#WOc4sGb@{bnN9v8vEg
za#CgB`<zXTpaYkwkVa56;@hECV`zw=so#`Z&_mhCxvG*EUMUKK(*}du>vQjam52yV
z;M>1dJ3BHI`Uo!O--y9s%+lQMO#N<-c3+#@&-f4Ga!SB^%PKqP2I2y8P7%UmlF(9u
zxyLZ$DyI#;gVSDQFB7OQCx&-Z1YeTO4Qfqb8d$1gTH%re!>{;;V_zDjM5H0Qzi?`f
z7=BMP?nIhXqpgC3z?7egPxg9Ki|r-u53Jitm9%O$xsR)9duqh{C^Hgsr#ckPXU96F
zXD@L)ix#F9B3uMqMEs4xw@d~}ARx*L4pJ{>Xsj>ht-mWSEHue>-;Yz4MS(pxNYrpS
zk7U17iNE@;S5m%vy+sKf_*q?egBTuKzjYiiLK*+Av$z<bE)6*K8mADvISkh^2;Gp0
zFqo(J^te0wYt}h98S;h<Tz)b(Wmn$vxvLz){S6DO+enBhg-%;s`ZL<uU)2~OkG%SV
zWA91zC^eC@Q8n-9>-^nLvvb<9)3ZHSB|677tUlg7j=E)T*T`E|i(myFNx;wnZ~yZd
zJ5M300JdUDG$gtUCm7QnICd%}fj_}7Z^nbm)Jyq0mi?T@w3qS-Z-&t?WZu`E$LQXQ
z8@M!dlyZYSUvmX#Yyp@U{1LCINFMpHz+~j_O3wetRRAQajcUGDj)``2m9ND%=0<iy
zRKl&5-jJ=B?HR{j6tMonkrFYo*+Y_5`5&@Im9UaYaP^uoUF>ZTA(MnGdZQ(Mmr1fX
zQ1X+jgXtp*!7pphKbgbv`g?W}WCz}33i-W6vv@seB?=z~n$*;Tj}F<4c?%+|!<W*O
z*1l-5`LeB3C8urJ9xmUQG@>pIxrF+e7p@jDRI*jg@ka%u@+MHmZ+2s?G2XLn-fgG3
zaXPoKA?t<y*QCk12ADR($*Arp*fO>uTb*k8$hGe@%+T;ADTi>KC~M{cV<5Ia>6Wrd
zoognITPJ2JRm3L}PNGeAtud}!XKto0W;BXE#XHYjmjMT!9~OlbcDjja3y3soU~}L>
zrNu5@*<tXzCP6a?_8Y8tSmizC5sBDGZ(YxhUV_<1;}5ndr5O=y6j!lx+fXOT{`E@k
z5V{tFg%^tF3ciKK<L$U`d4@rbk=`5BJ8oypQPKel4Pe+Oq+B@JlkI5t)Sd4ayt>A3
z%C6&4K{K+Z^jfk5l=ZZNE)YE&VH`6?jb956ynNo-x02ubh$WXS-#Y6>MvAFDc1arO
z+Cj3hNE1x3_a4MgpV$&?WEzahss(pKQ5nu2#U*)F1!8$p%bNfp8W&5$z*cE#jj(X9
zXH;xqXq$=F{koY;J^b@cx|tKJRO3#XRc*+SR%_+sFSM`@{-`VJPP4<0>qlGYw=WI%
zC>qRNn_RAY%-<>#quZP<zV7k3J|IP@eDrCj@>BITdA+U93Qfr=-5hgTKr))U*rK3n
zV7|Y~ybVYvfnE1JN@s?_=7|=AE{k|a|C>%WG{@aD%HrR&!UuMkdXb~d`fiHYo8)T7
zr!`jJ^7vf#_oR9=zMW;6Bim?ecj$HVj!0r?VoqFmUt<6c9mo?y)0-k6BZf7hMQIAU
zRM$HSbTk;Vq?k*fe=;i<e8)ue3xg2#b7SZqii1Zk`RKp{<e`hB33LlTObePwV<F=l
zqxEym4T(%K!W=esOjE!)3yuVolL<vBgvl_M<2Z37iX_|`0R4KYMms5`?0vXrhgD|9
zEOpSEVi#v1!Z|ghoC{f-Vz*~zn6Ey#jfm$oC(K=Qy?Zm0`|<CY-<_(ma*nbc?raqe
zZ+G>YCbD5VTXM_)po(L3sAbW0-jBxYA1~nZN7okWqObruRGJ22n2&PR*(1CGrTn;O
zj=mk+0709OpFV2~mb3~SCd{)o$tL{@h>yY~`fvWZ%4o>J5EI}6DS=Q@rz1hk|Ca85
ztRR2SIJoGK1_ZJobdl;meVKaf_#<lf5EJLJXBTyWo+pt-kQTZTTtl(A2he0TT6Z=9
zG<Osb+u!J7bENP4=i3~U%BJUExE1$r-rkOx$Qh%ZCWhFzcIFXoIUQVdf-=%uQ*vnK
z!l5Ix+jV)oYgng}V5IzyPT4wjK6|zz)u&&x6l=K;kr?n5mSt4Y#-E))iIC-kb#IgD
zNuPyMWwO*70(o(@xqPAq1Xu;}tZq${Rl<6qcI`+0XoqPlE}Lgj_uCfY%(&3Hc&o{f
zDI+f^yXs)0ENW6X+O((QwLib+T(Pk?CfGTyc+9nLJ31c-$6d(r#beTa9G}Es?K$7G
z^s=eb1C}B`2J)gq)-5i>|HONP6c>i?1mqH_DrY!JU-wBFd+WZ^F-&gIaAX#(`cAQ+
z8M^}g){XxlC#XJBVs!S<yUNycby$B{Y(B04$<nOrU&<{^mb8Z;>Z&Ox5*aP5xcuYO
z2W@$loi2@ieNY6=@u9^+$CJdGm*xrd8gR}K?+@{ahSNf$qg5PVzF)8ZwYCD>lC-<s
zIerIf@*vVasE?F2;|@(XRdEOYej5?0Cee<GzAj^($QLK46|8BK>}P-enGbX31@TDN
zV^uLKWlw3YxP4dCtupzOx4S9@ZecgHI6_`$_3S@c?6`R;6J`Xc3$DMk(Z`cNuiuf_
zRv!6oXTExO^T&d%->5H@H*j9297DtK?y4gp1Y8v}(XWbi<)h1+OZo)J&_dzl1E)SD
zscbN$^>%n?T2BSvGhQL;yM)s3{xt?gU(`EE&JjM$8VOCWyoO$g?Rn-xJI!E^2|1vP
zc0h0jNkUmsPLU=8N`aVqI(Qa<t2?n+2djtQLbdjVR_w>9TxUZQ(w%m3B(-{yp%ZiL
z<2}Yj#`h1*Lh-iLY%@jd(kKY#H9;g{)lJxJO$O_zReTbNHM|w${E=uA<XSQ5!Wb10
z9{{4WPCAdvVAOhN2%_Hn;>vre+6nk0MkmG(J`)_wIaZj%=FrO_cZK$4^ZtpC7)Wn1
zAhtaYy}<p$t83|N_VnV3*PdT{Q1>(}Oiv9e6eakSq21I~W4iriMn3T@)@x{#t4Yk5
zlp=g*=sISgGYl!PVsw(6#gJX6v=}z;9wNz%c5@r@(z?u!Q3IV&loX#>j7?`ursA!j
zFbEC{yec<sy!gS8?N7i6h>ZGEkb{^Xpdn}IJkrI0?HwED*8(KAsw>+_(YTI2@0q@W
z>v~E49XE5J26ir{WS#9QB|}G*u^^wamiJr3=}ncCwLt&s4CofQ9obm27<EioxZ&6-
z%{Pl_<3-9DZ9`^L=?xkIB_{C*wa--{=9R^WOYD|4zESW|V=P+$ZgfzKNh%@Te@Z3)
z8?s7#=puP=nA#H+=|U*DkBqdj+d78W2!pX43$`QW6Esr^g)`S@x!<~OT{mJuNQDl6
zRN}S9YPux6PNq6%Aj+C`s0w-C=TpDaDo|#c49*)cQh<?1jl_%Y`c=8pH2zoQS@B5Q
zhl{BOORk9(!aB?;RfjB0xTg53<?!)AzdoGHt1e1(^&DmMNu=v{w5hY_3JP8Z=9s_*
zk1^-|OxH^qmASX%sO0ok{mQ#jWJl-E>LGtG(wjjmXIFb%4GBR~%dlcDbpaI=r8JXC
z7sLAw4YY*^a2`Zb=2iFmCSNcNGu72xVgh37l=Z6WNE911$vbti{EpzOq?peL$OEk2
zq)O_qd2El{4Vl|pj>X5@OtPQeQ+U{`Hxl3)T;X@3PZS4ry_ivcsfEk!?AY4e@v&;X
zMs}P!|7}#hB`b%zD+y65Ky55J*z*o<kP0hb5sPhTpLWE!WrPjWf>(Mi{qX;M+{f3=
z%wCc^iSMlYr-fm$osag6=kwBA?3oAn+Sbjkg+{*Hf))krc}Kkfi5_Nd!lfnZ#E2GG
z(qG&9hs<E<o?>0R#Zd?<>-Mf-8FGMbdq|t{NQ8s7YsbBdUUB}UHnMOh)xpYgmrXJ(
zC-j&wV&~miXS@Qr+3s<t`ZYw7spAxcfe3C5sRHn9N<~vgt60!Z48j<eRVdxOSC-jB
zRB_@{gOzQ|G-^q8QPRA<WU`?Pzjy~-Zv8|XD(5iE!9E6q75kK!X^Z9k^TW$Q1goi<
z<jmC~@eoENmm$oY7sOQ6Byo2f#a<mlRLw~4F8J#RSFxE+!RgY48BF-J?x=sW!8nhd
zM_fwox`?66)P(Pv3N{9`tUS+dTJ`@Yw!8Xmm-o~3PYyI@(Xad`#;f{fFR1R-SiYM;
zw{!%k?IT7q3T~-=Zb+Bn7WQ56uxHzxOQX^~g70=fKEQFDU;*Yt4>E=!H*u_0M@U)|
zHt1m@vt2VfeacpN7Y;dEvyIi<mHF$>I`VD+A1LSSA4lceqe&SE`YI}d)HcH}=^&uj
z_|<KM$EKK?#|CEH1?4T%TeoVZ=ZI=Vm?Z?G(S~1$#5(2RMQY3()bK*Wg38OpnGhC1
z(Hy;ML8A;pd7x_?OJt+9CLqJ<iRE-W(r|t3QbSq*KR877Vg#>+DGQNFss?!mPAD0T
zt>(5hQ~H>xc+*wt>43#!E}JYM%furH&Y!Dn1P0l{S+IrUhf>wO;uQ)n*c*=36dqR3
zbd9Z3wpxy56ZTkTXgqAk73U3ii_W$7>ur%3vIru)W~r&@+%U9GhVIT>+2{@-&>pWK
zZ&>Jt;qD5@Y2t;#C(}bu@3=femGjwairF8te-S2!+g)nVtC`3cf16%^)~s@!^p<#E
z)qM(ps$8OKE`4;}@D`i1Di%srp!oiIxGB_CT0%lX?wsr4sO*e(F3{V1?ziaxxD;7$
zw*Q>fn7#Dxk`heOJ`$o{;ziKweC31+do@`My1GH87;kE4-cKH)hgqrt4$y+V<<Dm0
zxc?x&K)kh<F@a8YauSDgN${5wCi@j+h>(|apOd<XYtc(?YXeCv^o9IVUA;N08V72E
z6EDvgtqkV&E;(sHK_Q~C;lxybOArSm1zp9LaRUf;tRId}gP4B?I;-qf^t#!l5|h)3
z%NVb;pYyNZy|oJ;TbT{>BleOnJtcHe`Dh-W5<8nuzSMGFlI^n6#H12xtgEQN>i|sy
z$Zuv?_BK76eeephP1E2rv_-4CShZTSFn!sG$)ey1BH6jL@V)oGeS5V`wk}C1a?9$a
z<X&YlBQa}oB|jHcXY9jRJgie3Pmb*T)mZu9Zm3tvwaP>1eT(t+lBP*)C`b`dFsi3p
z12<^X(FzJa4wRVS41svNhAr3cGc2~eg8@5^r#q*o(>ze6G4(+bq$BC}eGtj~CRT*Q
z&SYD$IMzCV=tY3zGeN5d##m&4tA=u#F9K5<F&>ew^1liZ@F-Mqrd7bu%te3{2TGAW
zLE<M5Czp}eaW06GGWw%b*7uyw)3zeVN*xY=h0sHui2A0mHrt=_?CO+YS?Qx4aJihj
zsHT6DkMg#4&-y@h#MOu^csG%8@tsB6MqE%!Ls7BTn9BDAqVCc_<<$Dvu%&x{i-vF)
z6wTWW*7V@>I6AJSAP`Ru4vwr;P6dcbAB2^O7t*`pBaU(mis<q!!+?P8ua*>yv%%`>
zujCXwWsX+rRdfzqh@21wbUanEy7>5nBAa1s-M27D{8W3JQg$FDa4t-kQl_ea%Zj6v
zestIhFEeT#!9V01k_-MUf=JZ`C``q$P&QBYEBko4tu&X4nfh*wtDLvG7IN6ujA5*b
z^K*lyWyU-`$r&AXSI$zG>g1oIp9;a<#Di&$sbKij#d%Ysj99);!S$CObRxUR6ZMZ%
zmxxYALS=>7t0|xMZ0;FtpHK=Yx=8F4abmK_S!V3UEa$aYn$q0zky;`mjn-V2I1Npq
z*Zrg=_Z0ZbNXc@x5`j)CIZX5y&Xi(%L5P%*htou1k>$qmh-XqZ&~lfRJt8K%%zkfF
z{+yW$b;G~I?m*tLxL`zXwKGgNUo8?nP1LZDy3^F=`_Y-3N6Q-VdP2%Fa<#w;&BaR9
z@`u4eXgj!Q+2ODEXBfXnhxMAjcgTNoOM)evmhR%0YJa2_JuVuS?=`d4Q@<%NP;lD{
zD#$;;Qv@sHQZFDBb7|NV(HdQ(Ol?h)XkqXSKar-r^U^lAcFa~nqtg*lJc4DW<%hM<
zB;xX-?F|lC;d`IsIc0+Noc)^L=Y%$YEuas6D+TH4h4(vYR`m8<ta~A;`%Y2N*<@_q
zxKn{SnP1h?Ui1XkTxD1@2ftxU5Ed#RGm}_S9beexw~Pon#L2)yILF4g@)z6B*In?b
zBWvWB4tIQpPCeH9&1cTkJ%?L)$-Yo{<LCr;OJ%Vq9^f<O8R*O5U<NWH3Fm?!XpSa=
z8$U=-wd%bMKb5{_OuFb|zwNCf%9VC|@}bo4&RB^UpOS5sAJ)ymHTW0975cBf`Jd8v
zd`Hc7!=9eHrdFE%q;5(!Xm?ls`t-=8{x8KsTeHBH7XSIHA%W2JB$%Z(xM>MsEK5Hx
zf=e1uw?VQr1~enNLm=dg;F#|-nk3T$_Nc>_=&IIYiTVIDz!nJ*Uhqn%V<T~HoxMEm
zNUaZT7Hc|+qDqqm1!GYXxUE2^2xDRpgAe45996?deI5~>bVPriOEtO1T*thX1O%W$
zC~%mmHRbTFB)Pzp4L=%42Y2-SxLIRxtu7IgqLe9_r_rv&z1|~wN^sShr_ez8<E{5u
z`95<)ibcix*InCHHSV+{{BJrg2y9Bh9nWK@Xfbgah4;j>^xf%apA~NSPjL(^`Ta}t
ztWK($zbYY=_0a@^0ApqVt2(-Yv%p#!bOHR#RKW%;y-5g3Q;M*Z562xPIR81cN0yC0
zftUaL8bFZ@-yt$<P%<J-BUbKY9cxAgA8(XviKFCmAh{5y`p{NUc^}qq#7`M;lnFfl
z<W`cObx*R#lA|~N3E6dH27wfQ4HZE4ZtEN7_cz?rTd_n-c?o_gTqfrjCs0NSw2n4$
z-MWOpSV87R)#v8jiD<)%9$Eg$v8^vUUVxmeSuQxnYZF)P=&v-}L+y~SuZzYzZ)Gx2
zF>|FdBiLGzRpp2X+%t-9Z_+TZujbDp^YoyH1(XqV-q!Nl_pZ-O5Yn)T`jzNKlB8$C
zTbBTbsb_imi7v{ldv=8UHa^1{SpDlQTE5q6lZEM!t$xVE_#5;$7}--F401={NVTyd
z&3td{8=qx`=x`37V!x&$vag_7<b8R=(7v+cL!*4B`Y<YdGRLEgr@lVKN=s_}-2fr*
z?*V4A%`Hi;UOXGNdzniw=j}CPN)q&WnUr#$>5eN>Vr`CD7CI<66LOF%Ve-<os-Bo%
z&IwBtg9XTgv|#(ALp)(;kx~iIP`CHXx3jQrc3JS(*lDldbxUrZX7(UmZIjBkp9*ve
zkl$p{qSozbtnePBWd7V{NEDX3mnDOOkPBQbS0({=F=?!Cp@;nN*P$o(yLX<MGGD8d
zbwuh^7?g(<eRWCwm?7n}M94$7X~mzH6?`2!KHyD!Xlt}FzHHZ<WCU~UNaE=IC%P20
z=Gb}cix~ml(Kp>goI+b)3N+&#tD4O;wi*n$y>g!TVTRQdGi0EzkW3^5>k;T1f2=A{
zQ_W+kva~6vl71v0O@09{GpDLhacRGy4QW(=Hg4@&#*Y{>)e=|T%Mdm0eGx|elKDE^
zC;4j=d9n;A=Wu5vrZ-Ud%5ZxpQ+#!dme;O)XO`hyDz=ovl0h^FUe~P7q~`@q%7dW-
z_^yJSMBZX4Er}D3_e8oKn>j<apYG4y#=g<#0G(}#8e*8lc+TbI%RuhS)RBmbVz0$<
zowTSNXh%N^bhMU7#~93L5KiirpkO|BbY_kkLFNz&AZ-94hgm9ES%IiZEr{W%K81oL
zSsOxSXSmXB%=dYE8>xF;XG$TfHSPP#8`riR%Pw0=ND(e%-eF^Ow;}&~on7+2xPQ#p
zE`0er=S+9&-M);8;PL)L?3<H_uU9jj;&+V{Y$hD4yUp_=aHgPYt_Qv&81_$)P!R;2
zS$cm0MAMPWgkK5s7%Y^=Nq}Du=EVPJfbIa#A23Be`V9!tlfd7@G>74wcEJ?KA}4Ph
zXrHEhr%mU+*z9PvfXdY=T|MaJd;b?@Q&^r9OIz_x&z#f6LMb&^WQk?}Y1}-PDK{>m
zsxhbadNeH8>8g<TxkpvEI{m9$pqooPcyl*-Ytz&rdrd1BUTY|n&xaJ6gv-^9dqHGu
zzL;MYJv0CM60P0BD5FlE!%WUgF#IObR}cE^srf7EqVUrfU2UXA#^*D2GwP^pqhx;Z
z!?R59kS&Ta5%zZZPpxW}u0*30xhDw1+?3KuM^k%g#XB=ut@}|So~*{jak_oCm0L1(
zgx)*%6e#EieyfdYyp0%uS6<4ax3m{4eQ&N<)HIg%V5_o)^}TaIk5)ePT4k|_zrJfN
z;ryyh=96g0eAf{poTkT*nRZ3PAbTlII3@bou*L6?pvSIp!|~O}+)$|3A^)qNjjXlQ
zPsplY5ALO$)-(mA-76sH2EP{W{fU@<pytsktYu_9Q)lTcRX;(8ZB5Vhb-b@fn#jD*
zy}@lj5bSOc(I*yjA#5~=6<ivgim&^0HNMaD_9h>abd9(G?bg|W!wYG!SH+j&rZ*op
z+sH#U|CJj&BBA?bDj?dOf6&csCR)tgxQ<vxW__s6&ZI;bj$OSC_07on@=FPvzR>iE
zwh=N~=Bn3X1)yA$PJ>d~tZoFFM6ept)-%RKX763f$Ep*Bg`ji$B>M2JN?5X*T;n({
zqgn?XXy~|A!doBlteY_(E<1z-<ZLbGoVxU8uH#~%%U$R_@l?4c6=&(Ix**^V&5N;z
zs5cy|3gR+*^X?-jXF;M{)Y#dmNU4Odi-d9e)02xzwmG3XN&6{R41z1VpKn&6DyKV|
z&oNtl@?O4cvwE$Xd`$c48Rb#IK3sH6-rfSA?OB)_T=gX?XtS+yCh1Wrx4OO6K=}5i
z4@wWMW=ayQ&giZpGE3lj5RiFGdU47vJHJ0b>ztxABQGtZ+dnVMFJ5b%ods6fs;%zn
zlvt_>HGr3P4#X9Z19zWfsPEz)N7@E(&I04dpPu)q8vV<dr6)L>A}>SJp-z8j8h=|Q
zXmBz+mjMD&1z6ZD%yldPclo;Xc4Fu8e!p}?f8ZCY+Df@j=h@L*2ESw|-9<;*R%f2@
z$}gom^S2zH?y(zWoVT|7z!fu-MvOi7?C15AUNx<A;<gE8GWK_PMi-?Ur3HOW1)n&p
zo?~1nU71gA8U$n|$k7`FM2a1y`TZ%O>C>r#R5i`|MzE1!Vm_(@K@9;!ZqPqC;NVj5
zj5^9^EJ{K-=~V^v55cRtG8fB_A+r4qRRNFQ6oci2c1#C*we8|B=q8r9m``xg(3ct#
z5Z5gh=NS|%-(z;2$M}`6?GyP~WC(Zf4KvGC_ubc=s2r%sh+5qAd9#>3vRH97^3oif
zGwo~5K$I^q=~j9qFD85EPWmR&X1((gNbBCgvpUPETAn#h1C&L?n{Mo@JG2Ev&aNo?
z%eaPzt#!P&+XS)>*Q+H6GoNJS?2Q5?=2*w-qWn(wor>R0AP3yhI^V?>5l^2_k~s)L
zmMxs)E1Xir=UC|)bJ;b)3iTBn-(|<AUdA6w%b|HUbolIN{Jdk&#9H;px;`o2(S3-a
zO9X7>0(IhSg`c3@wN6IIUy@P9r8aAdgd;~;>&w`GDVU;-teove|2z~6)C=8^-xLDm
zX9Hd)rqmQi#=M@*mQd;yyhLb+j%n-_-fouD5ZN2uOc3-);X%v`D0@mIq$je^mAGa`
zq;4TJYTKf1i;6=6!yT2}dbAK4f#JpaM3RkymZIJ^JfrNGkpteh3L_IwpF&r~`b>SU
z)N0fBVwMw&@|X<&ozdr34qR5Mfg1jLIy~g4sij>)cj!3rReUh!_gI5F%;m{fQrvgq
zl<VEi@$4^}XXk@+bwwJQ1#YjRmExz6TSYJ!_mQ_oMnOvo#_XLA&UY!{lZ*4O?-1JE
z^^RJEnj8y(15I!AlScT=8X2xA!le$S?0C8t%_66aLP>KZ_f}WT_=i6SJdf2AcAMeo
zc*|$G43tSzE!B=8{+w>BLk_ga&FIMRPr_%tU-!9Gj-TukT@x?c@w9=9T5SgghZwty
z2U^b)Wo@L-XulZC?fOh^zE)*^g%nmnhq0?1Rr_=<$9;Y-H!~qOScB~Xk6hS~;9B3s
zPv>OzjNl2W>RP(@d+&}|ZBcVN8`OldFmxm*cm`&avJN1wXn>IbiyoW^h0i2O%(an+
zxbugn$~Q+sy*X>D54)kI>@VQlMq;#!P~!3Yr_8P4)}Y-N4eF5_<Vd>>XoTP<{9pDK
zbS(tV4hBwKVB8JXFO;K9fFnU41i&{Tb6dd48G(mH;PV5eN(Dxu?={;6<peFS@x3>J
zzPYDfF}rd5pk{ezJz^Y{zv4y<MHRJNJ9aiE_HN=X-fbuJpN?W-_Yo~e%)I8|Z=vIF
z_3uXYES-B%%EOr7=7z~f^8E+t36-*3{txovR59I&A9V*+zNW{m6j+Svc{|Fg0&LOg
z6v_TKeX2N4palUYYY2waz#B`h`a|jhP2Zow@$YX(nFf+qCBf`V_viKf5E!48gT3J)
z3Cmuf&n4Nlj@PM*)fTnXEN>57Dj@fqD_}r>g+qFLN{8GNbhJZSwe5qShx7heTx^!=
zOa7K_X-s->P1{S3h|92>2ec<1%aurr{9+P=%hk8$4-%aiG?Dgdf?mwG%|^oUfi4$L
z6bGc*hb&#59STm2d2pQhRy6VIqYZYFVowdfrcFmoGmm<fdaAvGB}87TTzC#1qHCn2
zrd!pxjUd|7CYP%F@E$K$(_N02>0Y8GM;V9Fz4vUN&8b0{|I3U&dD1A-ED*)U8KOB?
z9Y6l1!5Z=EobxfMA}!}>ThTSzG@Xc`@=je7>VSdhOYvVMr)^+_51G?KPhI6zdqko{
zwSrCs>ubxq)vWgG8EM$(!^*|m(1o+!FUJkS5lPi2F^3d}oTC1AuCX)4-vnh^l^c8X
zE3S2U4m-8$;9DEJ%FgT)m^1te+<H9)^e{fmuVg|MV@^#IVPfe`@CGv(pvZTt^=2p8
zsD)u3$&=}2T;=-=CX6W+{Q=CUYo%&s0K#YUa4<wB?^8;Y62Ae(d?<QcCu}FHsRf6{
z@A1DXxjociEx$Sx?QqjebfD&9;kP`LfeU(RNMVgVZ5d9Qe48y6<>D5z9Mkrhxku8y
z`-#5Wf*=tt@E#6rBQAf{XiJ2(EvMUH@r08PO<*Oz8qG9^IN92Vtfr8x*MBZZ^{L5l
z0dR?>O%H6m*9cc#tKPHF$T8AcC|zu1=_F5mzN0EWI$BHDuzrAV_AS}GO0{In1!^OJ
z?!l2}By;dAx{PxonHLy?{ii3>J{y@*2IW%I={+31qdk_Vw_jEj4u9kU2R-Y{=bGh1
z27lLnBV^6brnnx_gkbeIUOz`K8@OKdN>4F;!ctHA^~_tS{$BKExxvZtM72G;N_Ca4
zK#AfV_gb0$g`6FMr-GGC=%q1Ge;!fkpo1*H^ULOdAut<?&4kc_;SzJcTg<E9C7maI
zy(;J{6<vj!M7sSTQlw!;6z_Uz1da}xO5fe=*&dr~>VD(|8<Mp3VWzWG`vY(QYaB4{
zTn58)cnF*u+5sk4tT^xk0+PaxPClrL8EX;!-m~DI^#&Mt+nyw@1u5fs6QP>Y7$Zf4
z>wc*+J_ZCx3LK>-kq4fbht7unt7~s&h<0rI5|E96pZI#eTRlqJ-mzk`48J|8bqQ}*
zWyw+B%z3BdI?T1)J~{Hqmz{XAZv>}U<eSugp=hkR@|%Q+KnHcT3}F!GDFqO-ZEGS3
zS4;yhK+&Hz`;Q2$_Q)f0^w}3dU|4f740I9EGh)&`NT?#EjUe8^EUTI9n1xc2;qv(G
zZakBd-FlY53`R{Z8Fd#QTHQ?tNDCz;S2gp2xLN#mM~b3nI1KrvAkvQgGwGvAy1<lT
zxf5f?4+mV(jR@d$*_fZ<525o!Qk>{!7D$(=HpP4s+ALz2pHnY(l`qWQ>h;q!SG7dA
zbXr~z+bVO*)yGf5-76KK7X<Uf7ep<rQPvjJLvuAtXjW3H&D}B;4novNyZUAuH@gc|
z#tGU1UO9HS#;f3RmEd}<#>N~ns=l=HoQ)VE-j|iKn37ifPH*p3;r<7F@3Z``4mc~H
z5qEZS6LL|bjs@U=BV8u_jvW^+;LM_#u!od2+w?;sT}wBo70es&@Ldx;u1PetWW1_C
zVf>JTpZE{bcnW1l!pHlt-7-#EK$mrQNw*=c(#EeuJ7XYJ|I*<VaRmk|DH*%ZQx56!
zD>=rl$61<fLb5E-h7#Xy0wDtNqO!!m`>#(<U_;_m18_sg>>+2_4AHGOecsc$U<@7c
zy4kM$0GDhSOdbQlGix;dkr#<GX_J?u__R_BjDtj4;3oXOv?<ugE7hWJSG-}VhHD=}
zCB2JJyKH(dly-{by=v6M*v)CdPClk8<(9e@6WJ2iyO%?D?A2xNq&xQuqRvSslD!oX
z7CRJo?OeVMG_OfZ(SFppwvxTm(AV8et#iGupR1E4>&s`DqW*(up|VW`7=Xgr(z6FP
z(PIb;365~PFpy#&@_c8zO6&2bU$geT+=_34>J%xJ$yQ8QRGZP*90kg+i=F`&AU&TQ
zf;Btaz1ughExeR8d{0y5nG8lGQc8TJzTvC(C#?4N*is^FsIjp&w(pwP8F{!qr#K0t
z`mUeZ!#L2mqW<5!FM-<kR5;eksk@37OQkO9AMfo+L}MEnX0xw))S4>1ooh7P;~dtU
z*RAU;6~T`0^?Z@ms^r3FiLcKEARuXIe}KVK;6)z|xv|Wkgf&y#zs1q?04Ih%Z{k&<
ztQ}vfoSVFvke#zid%9c#*PDJ5ruf56`{c4mC&GUbSsOQ@pPFkoH<=Ah*M4=)8NHA#
zE%RAQuGHSf0)4CKsY<=}+k0P{;o)a7{$1qeNwYtV@c(tifJt7i0E1$ZNlxI3m8Ji`
zX9JM=gOxVmNsvK2Qa5-5u)E^XZG3!-NIZb@#<09OGl=oetEJ~=t>V66fB%^F50A}~
zhFxb{VGRq%+l$rBD7%Rtu_IrvN<1vT0FA2tq_KbFg3P`8+cQtie)xDIM-k0_KJ5kF
znB$MDjJV9Z`|z0pGKY2H>q!7wjDW1+24~bYiB32=H$wH%tq;)307Cd@ba=F{6Z#Y0
zJq4PTB6?W{HY+%6$rUQj#pJ<vL2bw*P<LlYZVETl86`IjPYRr|7`gg|Jfz`b6L6+z
z%{el!Q-{`VHrwqX7!F@OL)I+EnJGxj)1?<8#{SIaO$n6?7mO8P@?BX9ZwNK-=++BF
zMSA7P=*TIxOexT=f<bkCn65hDbxVh4g-7o=!$^?}*sFRjhZH;4)wFHs2WH1+c&?Fs
z@Uz_kc4w*hq&+9?nM8l$Sy?=uGu%>_2UW}1VsMs9<!6#Ev}q5)I7Z&Fy6)7*K$d}%
z(mug;h_x&dEA~br*L17uZB}3Z+=6a)8%;UFdj=7Qywf~-ly4~&8L=4uwXSdfV8cgF
z*CZoWYum3<Ec$HT+MyS!gFuEaB>)pqld1-6y=|;pO{c!Me<q>zdDzTgdK;$>rZpDa
zY%kx9T4ff_e-^poe_e$7V^ME$14c+$LeEHBJ!=zp^mE8a`$zo=Bdz4Rnf+S#H>WF|
zSdFL0L`T^B6cknnmPD2D{yQINuUzCD`j{2pDWf(Pq&!bD6oR>Dq{(&L{<`2&t^m6i
zOEXeAl*TXaV{3#pIXhH;B5ic;6zf8P#degUWtYgWb(#kU8{*{5dgVAN!nB^A$6!2+
z&rQ0^e74P5Hx`lA<VD!B-1L%aYW~4<UrE*LCY<mU`I<<EGk)|k^%x)(!X5C!xj<J3
zyZfTiRKeHqbHhP7n!Z+N3S&j<Z>1FXp!wVaJP!t@Z46ZmmT=yeoBSN4bCoYHY;>}k
zdkYlKTFcOyJsYg8d7tBxI3wrCro^S85+{E$*y+<35g26=<xUJ2oD!WFPr8P0WJ(Rb
zjO`)~mrxA9$XZw6pU*$9uX*Vd`)i?pwSxPi+X)Vx6;<WblNI<2l6yPO%RUJHu)_UH
zR>URMS3M^#oD<a!khIxIgdued?axa*=60)S8oJC{-K{=$$2pG+mI8UJae1*eMpPtm
ziP7TMM^6d2jG=uM1qwaP05F#L?3ia1c-JoD29gPIIOJb}OASiZt#<_+@?OjSE4Ieq
z=~|rcytJBU2?d;h(_U^$5-f6O{viiHKmEQS{u23MQ;+dK2v2JZ@}6vzpXm04!Oxvq
zj~mlnX*c4+=|)V6=@C;gH;!r6Bs}>9Nn~Hte2{fd6rXZ2w-fJmZYDwb5pVJ%%1T(y
z$L81P-qnFvK7{U*T8Z+a@zzJjO<UYv-E(-e+EPGMR8JwSjABvDU}aW>p#$=O*`op;
zooX_|9)TeIReu#6360?jyxl<aeT3tM)BtFdEs%|ap(5DtvE~1O1zu0hpG%VI><W=}
zH&5Y;X`MK}!XW(SGqqqB;fs$;Uyjc0S|QY#(XW<1KVPf&-bXJ?lf48Ht$*!U!DW04
zzGN`{FAYabj1ZA0-T_&&R1E<nMw7qgg%f=oD2I-oC+rR#5U|@z0^eq5!0sqOJdsNQ
zGV3}2{N`&4)k9Ky)mry2k2V^ZHf<al;H+kl=`QtV3nKx<rpvFEs<?3lj>2#E>?@)=
zkK1qEwwsD@t@re}Fcp$;+jKQyAAaf8inDK5x(SYh;4j*fO=HyoyL|{;?iS;)Jackx
zZ;vK~MbK$RZpgF{^O3-wv1SeZ2^=BwiOi8GE@>Bd{rIOR(0L6i^HNnn)szA{k$3^>
zYcIkeObylP-kJN?LaK_WUcR2DIO`xA;R1`07OdW)q^&TWER7^s!n3CnXUy{F%2Zz<
zM<Ji9F2hUP`U7dnx&V>Ym=RDfW5&g*I52t^5jd~6Z*+pW&XXn4>O)DYcdDJ+BMRKe
zx@#QPY$~Y3uAm8mGC9=P3ot+0k+roS9v8LVm;RlQ?|GmH!(8kpnPE$<^X-N-O1r8~
zBIx3@e!YN4jCM=2<E@knZ&tLLsP#=bETyY+`dlR1b+(XSXzL{1$b03LqWD{F!Ffse
zFigxX&76-wGH1rOv#_(hw_}^xpe1dk(W?szD8)es1M%U*Cb{#1W2Z;%Z2r>p)Oj!3
zxi59R<El};rf|ULD#Y~}AVn@YrKQV>_B9HM-Np}IP>$D9Rk)x`N~U_uBTL2$Y3e6d
zU0D}9_poepg>-MouB9_fQdjU`&Loug(ILEuPC#OJ<p;)ie6C_TuBv9pBEBbub}m^*
z=W8~vePpD91e5ncq}gS@vzBu*;tR^I)J$Cj;8Dp;E2x-oJwYrY=tOj@PBqi>O`Uxy
z6P?UqJiMzI)aJZ0b<pzeMAXlTmx6vZ_a0=7riP`~3l^LSBED=;>Yu|JKmAq{zJ9&K
z?BBwygBFFIV1^Ryf~CQi2?-dd=PPs)o;|6Hm5W@%c;|C@$cM4ibI+&l240D@#d(Mi
zpT1*yK#LPB&N_Nd$l-8Fsm&l0g@oug17{19T;)GEN554e9;lL2+<jK$yrzMPX|LQJ
z;)AP~3oTYrPuq;Q7g{8j|NSl4@C8-qwbHnE@JgnUo3w}`da-G1em(H&Tgx*QvWP!~
z-fXTZzMp}OnCSF4(6kvIl8>4ji3&J3XC1CQFOjn`?=MtxqjgGjED~-%>2mgs5tG9B
zWTt^euU8Lt)$tO^;Yp`UWg1vrdS1uHF|RC5T!gLiv=*L&j-ZsB*cF7)!O6_HKbJeW
z>W#qr1UhbLO};txNGr<t$Cd!%IgZK{z$`rSkAmR<a43OvAPo8}-~vfW<A#8r$x0-y
z@PKH?m2E~B8=9xLt5g=G7?DUQ%_^B{ixv5d-|e}3*{vu&=Yosy{qu=rd$6Ymq@P*@
zC)iXn5`6w&Pv0Gu^!omfTgj1%g?kTB%Yo%A7l>PM<f_~XspTG>%8`rQD@8LzKyam|
z<;XHqbCx2SmX<qrr_y%k{N8px-`_u8?~6<D4}9+DzF+ritS3nl1e}1%1c=c917%?N
z03LwEZ+@)bAq@^tr|>(6K)+3h0>q~5><DY(RC7fX^LCIS6Vbf-RPV%bW|1L%K#!r2
z&keIpf(za$5ItkBdqt)MC*@oizG6!-Q#CQI(*gEsvf`sL6rI$;k4}^Ice^i@tlMSh
z;#=yiATwl)izx=!6_L0Y3g){>YV009YMpxSQ*1C8n}%Oi2#i-33qo&V$2W61SW_(e
zj=dsTseC0HoRcJVK)FTi64kHKL;2N6G)pP~Uf>s!iMi(B&uHdl^VGlJL&}Yp+l)~M
zVv%SDsF%c4(CQtcf1Q9Caqlo!opn1FQTP~RVe1Oae5@{Jq#Pp2I+x|%s4_pLAIg6-
z3e5-Ma62B08=c9Y>Q?3<q+rsQQ*CZ7hy!>F>{+9h#y~H%_*Z&?l|}U`j=8uJNvwsh
zpf3j@ZjPPU-*E<V)z|ZnYrcr8N^<pB3N(27oi1x({(^(w;Q5PpRF5=;WV_rrJw?kl
zIpWfKXi@U4TS;->rI_~`l@gdZU(H(_^hW3l#q~Bz?WDacCidQ%DO@01$$&+d(bA>F
z#3^Almy1#MkvyT!!Y{TPTjPiHvf_@$%p?6zX1?1|C~vO_U-f43pBI$o{iUW0+DM#g
zPER^M;>^rb16O2YB&Kcx-inji&Tps~W$T|TKs(()Q}la9YUHH!Mn%^Z?IZX}<DXfR
zr$X3Y%fOwi5;;%Y$ov}!CW$dZ$1>zY0xjhqxH=jx?m9f{WcZZY-nxmNXfWM}l&0NL
z&xYrjVJ&sNphfg`DhD%3Msy~nEeRl9dirQh8BNGHT{?+LlYVBLBfc{`M6Vinb|sk#
zRxHbKTq+QZz&!bZSR6VfI{B*R8vdH?u|1voFPU0H*MEGWp%fmqYcrk+0x^f%K5#GZ
z_O`d)HF8q2czbiXRzClqdE=;$5B<7wANodwP8zI6(;L}xR#E$Po^x4cam?Et#{0|l
z={$`g548Ot`+$Iq4GFBt45!$Dq$*|hWI@PBbl^8t_b^*I<t^ym2^9_7W}Xx_dPv(3
z(FUCs&SSqeN5w`0F=ZSqjv4Zrdvo|@iMcR}cYm$^fX8hojK4q+Z8gTILa|3OfBM0R
zRxdy}|N54-;#k}ePsAI-MBAnM%!Sb1ns(E{8+rglXc*zn%9syS8T)Ov%S4^I^mbu5
z>on}U%U5WF<k--FJNvKO&l_D;b{W<o0sh>n12O!6N&>*T{R0iMQGvQqOir>$Q=>Sw
zIkBM+rKjc?I}XI*PJ`=@zviDHCt!I17_@_7b0p4iq(Y$}5)@6eA0i6mup&GXd!Cd;
zdJC1*>(z)mh-NwceB~iLa<6MrzJ|lqpldO<4l>i9_FoGDBnkP;6145i2-*Pc{V;U)
zKW>m;VUplJ$%%a#AQj}{?c0(7yA05PBoNtwa%86-O&&I2gl^T?Z0(3AL8Y9J1SO9~
zxcp*$>MJ@$RhKuOcb56fDZlPjYpa3|P*&|~m`%?`^QYwTrzQA$hW$cwaZuk}m_%d7
zi9`v4`xmJ%2!(()g#j^?nJHjZSLQHJxCH^Z-tyqQT*ORLdDP(n;i^+fGRFcz<-E;w
zn~ye-8^=9SBhu~%7m<t=`)AxVt1#YeUzwvlT&JOo&ik~Q=8k2uKOuxb7_!UVr=-BM
zofwX9%_!s5>hgb$q@(l|5lZ%ndAH5G`H7$bkM%;iO~*Irjg*N?Cy?i2<R*~rIRHK%
z1j`k`HWUMt+aO~bm|YUV%7RXm54Q(hotC21v57%igohya7;V+#N61n>)0stz2wgbl
zQ3Lv%m{@;2h-vq>!DPnF+9iTZKZeFtyo6M`8_iVP*nhdOnL3?Z_sS|f_w6c^J8GrT
z2#_L~0=8h!a`8T7k5BJnV8^mJtD<k7<n6IXYE)fXy1uapF2zW%J<7r>sfN6?v6l$D
zeZfOj5KeAW@wy|J=(=`!p3o-u;5+Sy%ZqmgB)+BLMc1}PALpZm<7<Cw-?d4z)t(<L
zgA!jKJO8Yup*8MJ6w3`e|7s`qVbRyWVo8rKe;Ohel)HI(c|g;QJnOA%haL?+rFZO|
z3nVIjhEYU+O!_E!uzkSRbG$7{w`p77v`e8(L(Xn<IZ2+n!xf@xQ_B9(rZ)15nuX57
zKlrL+>t)CN^>lvjQ{-Dket|`xOTSlFgPTbdq%za!SIY;J`~Qete;$9wc!n*{l7lLo
zSQ<Jy3izyQh6LE9(4=8WA5kz~939ZtC1w`FGW{QltBO_8E^2pL`}>cqlaal-mtcy&
z56XQEi}z^*f3?;Y#HaE+p<4C*$~i6Z*XXg4yE+SXM>5CwPc-HMfF{;QbIfmxEdu&I
z<1M5#R5f|A=%LNEYZldFt!3d4WYVjk94F${iHc-VgRF{Zz-!>98%+*Mh?l3c@eU{h
zB2ZH<AXn24P|#w46k2f8=)_F=U1-6P6)BdZUri@NBKOc{0%~&USg7^d$q39SPBO2H
zje4cU%?roTP}pr#Kl{UVsOh3J>r0I_Ke)ofY|ylxiPBBJu)JZ?jIOc06hEq3X4vDD
zrs(BWk@mB8%`}lkp932~%T}4f=M_XDe*XoN|3O5xYPFUDm8}}5K9;HG2?$SZ0J0%4
zrhot!Y+#A_9}IbjF%Upp0OUZvW6FcbXaspr1#8U<#v8j0Qjd1DYV=s?2}PDZH-4n`
z<CSZ8l$6zPB9_w+U!V1~4q7XH6XJTfWI2?sX(=6c5k3HG5DWezhZg`T1Iy#XfhY+G
zFoL6}jEVL{W(K0c;W}mj@lcG1v_FxSb6PK0GBH=cLVriC1s`h}y1Gx>2_g-KJ-nj*
zb8_vo-PkGrb~R5nL0+@vI|S4)?;M9pmA1kzBKeEDA;eEWYg&=JL0N6v&T>@yNE^#?
zb%Gr`0kS`p53aMfVdS#JpsmsG8Ov3syVzJ{h(gO*&jML)Y6Y{TTy<<AWUgvj)%l#5
z{{cP7R5y3<=7A2qhsl8!z(+)R<hngY7=e+DpX30do4C{zDk{zrup(94-*x8H-)DsG
zR7+NyC-So{%QmVLS+#owP%C_JF>Zb(`X^Z7vJF^k*OVn8VMbe>$Smm}rt;%6I}wro
zjqdYJ?V8fX)cDzkt<!Fq$Fh6`qhR}pv6{V6bJ`kW?w^oxrJVOli#sl4`sb8lX9*41
zcbwACw)E&&Qq$d|vJ2xbtQQM<JTv*a+}Tq{gR&>M-uPMn0JqAQB0nu-ZDaJ56`svG
zLfBm6&&C5%goioYI{`n48m^=1YU`MFIUhp~55vJBrR@H?2Ej-!gi46VyykRA{<Xbq
zlWlX=v3MHim(<%pHn9AUDd9w}6^=%0jmMt57XVdVf%RIa3PlwhB3=^z7%&KLIL58D
zF<to(v4ISj)99=n>bQ<O>huUD67J2yEZE)Q1AI(Re$s|Jw*;ScVnd|#JJPCJ&frff
z#J?Qt{Ndcu<epu%|Harzfif+1_Bvgl!Xf2etxYLM1=Xo*Bg8_-``3DdnVkS{hL}D@
zexOu&Na{<SsP4$O)gtQ>kwK9@lkxP*ec1&@^!ZQk67q^Z*kAU0``kRCu;HYyr#PTG
zXzPvP&QAvtg4j>dWy-{A3r-am5Yku9loUnaG-M46HxhMbbkFJ4lJYOUWE)&QH!)4~
zx`vGG$i4g?duKmTFAT$PuY4bR;<ol_F&*TW1mg$ach`Oc-mU={Nw3m8gPb2O3TNRT
zBCqJCLw1g{YsWPo@F>)s4={fJx~%HCVItAyzFbW;iSPJ2B)bEV6b9k9Ib2DA0D>Uf
z`h)0$H~7Z~qO<hy5aic7CgEv?0w5904zHX(0;na(emPre4rBJ`#n$d9wV{b7x4>1k
zAM`L)7vcRVA!pA<kMzN7s?Nk&Q=~kJ6F?S#NOoYAz`-_C8Nx;`{a<GQY#RW<3mGU1
zz$kI_(4a~S3<I?l_x`bE76H68gNgw*lhy}|uA`uBQ}#+Ng@-$;pYs)?eF<5i2><ga
z-_GcLJwTQ}18hS=MrY1_C0yG+RT+B}>k61;nIFQ4G!FXsVYoWrRtn54KvUI8W3xs3
ztrIwbVFMJD62dlgTF(>7D0r1|-pW7Q#jc%qt57~M=#oXc7eNrWta2F-zA5!Zra8=*
zS&UG@Vee|75`tOSgy7fZLnlk^Py2IBAp}&6Q;tLBrfyn^sk*=(Xuo17zYu~b8WP5C
zE4V(3sbiv?embMD=YggFm^Yv@IYPhz`nmx)&0e13kz;-!c4{%cml9{DEXkKf`iQ>9
zT^U5d>YEIxxEiRJas))!K*3{s0^c{C91^E&o#9W8!0#qzSIVddBw+oER2%iPv3lvf
ze59&!lG47SF|%&He9TcI_nAs9-}ZR0N@7{a1;rN3P+P$y_H>poBq_=z;foU1h;V(t
z_jQpWY>s>?8z8Xq{&2Q*p_GefnbLNERD=j?0y`l;f|-tUc-kgoul;Q~sjyj5rr7Sf
zvuS6~s2H6VdCR74FzXdn{o3RKDcE#zCq++5=p?bRVbKF*<l`cLpS_TtPyJvj@%nsc
zajCJ|)+L{#NlN}=$_m%BhUq#T-wmB~T$j`<XOtG=^33e{vZqeuwD<B~w_1FK5r6KG
zUZ$|d_}Q0fkUu4U&CAcfM?pSlp4D$Mhl#H%^45kzDkPAni%c%*{N%F!fF03Zx;tE9
zQg7Ni#^yY3t#l5~E->wgcWKREf+|wA-h3W7E}bTTFyR|;xE<KF%s4vn`)RkcM7gp@
zo&Mtkty7w3HX1n@KI$^?v{x495;`wWx{*A6MnC4l17+HH30QYlfn~F5*(#5T?&|_{
z6Vq%x7+3x>%1go>KI>E6^3`<H0noT1ZkMXmnh__)n#omaYzgFKULdl7O_X>KFfR{W
z@U9zf8@!k2>8=Nz2n)GV+pL#oQr?=L`j$DfYy0NG%zney_Ns-4YO2yRMR_iWA38hE
zM~2XH;*iMNZ;UTXfoWvtk$oQemwJcQR)aUI9!AyAz>V?CMmLLPOP(XNg)clfp;&Nr
z+Qi23gt+eX#+GVJ2GXOf4CsbAQU`3mrh!n;xGT`?K#UkbnpXo1oz}Y~UEknSU*D&D
zpVoLb8q`Q?@TgOM4w=I0sL+)t+<JW)x7oZ4+NO^%IEK@H*7r7*LDA{Ajom0&hnyh9
z1RGNR5Kx@Zg6y$CBN$WpzGAG}f2X(bf5*D`&$0f4?g4KBb_jr>2{#CE5f*He*1-zu
z4jC{?pIID~<-I<Jl+tQ6(OQ3@@kP&so6yDl`n4f3V%AQr4uCEJqp74xJ1+8n-19%5
zbJ!-tP_)KD6Y?QIh2gg4aVEA(`Lol|+oYYFC-5qNG11Hzjm$DpW{K}hrRmDy^M!Eu
zUC$!@P3();5g0df(IO&r3KjE30TE@l1WqiPo$x*_+u1U%@-N7<LVB`jO9wTyObQp^
zeP6fN&m+Bs%$JcUPa8D$`P6=lDTEjb0oBrJpx&w2^(3j5A1r!G0ic{Oere<DjWrUP
zaKJ^PDID&r;E+&f1ImFCfJN43y-ajadtIf1ncUJplSmA>xsp7flw1CicpMq&fYX7$
zG-$Qlg1sl&ZBm;na8yS;ms&UP@w=Y!&K5xtX~RXJ6&gNfHEQO5Fg4?K-*O5Y6p@?W
zWH~9=E)3lq_tf7Su=1rcjK?JErK)De;=@=>yhHA#`fNYA)U;w@(~_eN)FR@$*pbS1
zF1tvWN@>g+agxEc!I3s+11?}ikuMP&IbyP*w&^^gt1t|Io*A!lzuijHhf=v04o8i7
z+61K8UOwWeL(a7(2^P3&`U>9z3EEp<>MR~v0qu)%2~|Yc2`$s=R=`Hi>~V6co@*!H
z#H<49dvQLWRGNLXr^&m!;bM6&Dz726Q2he!FPYT#wQ=p+M=ys~m7Im}`hyZ#nj$d>
zeukmdtY7P8Vdzl_vK^K2avtTMpT2Q*N(7bts}Wsl_bTwV(6}j6XlePD0Q<b*Pu3Is
zOHc2bj+Myl4n1)5LJ2LuJXW#5hBXAY$2|MGnQ}_g5Io^fqQMnjg4zaTed2dNF`G=`
z$+v4ZvCN~=Eln%KyDq8)l+dDY*n$YvG&iQKp;@`t-pSgYkPWZ9V+8;!nQSzB+5bE|
z+DF}42P0RTK)%Hbz}frD0IaMSOdl0xCPAn9Gw?;5XyFQF;`wznG-I#PMfqXrvHnb?
zuiAl?W#D8^u2oJl)=Z4tDoOup`6d`Mbsw9$A#&bk^Ga2CpyLVC>!1DwVVviyK^MG^
zx?fu~ny!tuwU9xUSk`|HfckFWi|bzEx85Zxx*RC^a@(Tcbl}HlR#ldqn>~efX2*fZ
zd!TTG?3|awBTR*2=;#5!#~CX8NBeSQG_Nr_;v+Z@|DyVeN=lQJ`+XN{dwgN$MU!;A
zinWj_7rR!wYat)s+2?QXs~-lSjYc40%MBJ$)<n)j;S6y0fL<92+~3ed<{>QDMt~e%
zs(|i|*9P$B-vyp~0e#&6?plDKe0VUkowR{H0vBVK`suC-gRN60Tyq89opYWb43X0X
zMOC`tj$Z>CGS{V7XqS#`yNb92On<=D;;?ToJ9O9nvm_O22g(e<^*nT*^P4>MM1qBg
zE2)YzfN^-U+tU0Rz#38MmHLPkEQ0{C49%damJc@X6Fu3m+_)T~@XPZds1Px$qa5St
z9xkq_`6y?&Lko?3y=@FGAPwlzdPM}#2F4QzJIz$uVgmi9r4-gKGMz$RsHnsmUxL1G
zJg!_XyhpY8?997o(^x<-$4M6?0t;hFMa5LR_0&RJOak?JqqdR>ugn3?!3wNyAd@bQ
z^OfKVC%J^-7VN=DR^MP^qquD8xkeM0mgTy2t<>D?8>KuK?DYrTgP~ujQB-I9qs4D*
z)C}em&Zkh^22^kkZUM<px6uJFEo`1U6Px?~1?8YtiIiMCvuon!SIX+XSa5~!RwE&<
z_<2={z@38BkV|s!tBS`vr3kok2g6%T%1wf}%VPz*lO+0i9&aq_K#mA9rP;-y@VPkK
z`49$DHxtNnUBpe^eS2>PFn2l^*P*XE;o9R}@-XGBSrPU$YlvK~klQ|ibE~qu9+ku=
zh4sl|ueg;|yO&yYZfK*1YNK}Pvz$!v^w0hu&KwHc=kkpP>pG;|F!{$dP}a1Yvx5eq
z)4M0mrEGff;PerFMt7XTBjF*k_!bhGrL&aCdK{xAIzEn6s*IuN>MTd@xH^h?=Xv%y
z?98bvz~pirACVc3wTjnXsAV_6P;25Efew`{8M~xKh8VdpC|px8<ce-3%To9?(NBb$
z9c6l}MGAa1npF6wV=t3Dc8BYt8_1=Z+t;P(@g{KQYkmh_NTG*zWDxQtEb|?{pXs|5
zA5~5fcik|tKSAxcR={}=N6DEc3LbO=Tp7<h<xdZ4#pr7-fqZ&?i1Co1;hP?{2^^iZ
z5p1_E+k}65BmuhO6QaM1<|Q)Tz3|O|10?Y~`O8gX?W_}QOP{~F(qpt$hp*Jgklf(4
zQNdgjfg@f`M+QDUw6Yj5Ef|#2d-P55euRb1;yu5s-4;>DiHrxBuNhzKavG=J47Yb$
zJcYc?U2J#LNbnQ?+aNvMs-@Q0>wK$zz<Q&CxSQNDR{DlS=YsjgQ@caP(RG$+fb>B_
zFWDf0tY5js39jVNa(a;ZT!83LCHW<rK|rrPeU5y6KGUU^>vxeF=)~mwau9)T@E}Wf
zB(ks$SzpV*ZsHE-vCwY<cd4$|{LjV+U}<xLWTfOzT34n($J2&+VXvni?jsI+%S=%>
zd0?-B1sVWAQpXlT0?uSV!io0^6>=!d!<h0}+nEu8_8SO3K3o(0Il}<a37G9QIe;h`
z0OW^kf!<E8j)4WyK=j8_Wtz88;&|~nmW!k~Pp#eyAnX>fwDI{9c9=r&KaM!GJmylh
z;ntK60rn7Phkq8ZyGkMWK(OK(pyThU4)f8J%R)(Mx%R4_0U`>lr}%&uJ9;JtWs5x2
zlVYmp(``2~RXuoeq~Fo`h*_rpYx;7CsR0A@8>X{~+Oe=7SEOdm6osf3*p1liNC>l}
zT)fbvKFocHris5QkzMU!Xomi{(r~K<)@%M@&<IMvh4m#vGI7eQ+&1zqp2w4nQCmSs
zScRTBTFa#rO`1fuP}nM`v^UfY-XAgI?n@+|PjXUr>{yeMC==nbFbdg76Y)>Qa*tb2
z9*<}h!pu~w^5nwFzULw=DY4@YxdH;m9jN<$ZN8mQ`fCvZr9R3@hNL{tOxbv{OS=L5
z3?|*;by_G88*>wWTK+qAm1+2vo+oaM{Z(@h@gksbx-d(>molu!zrEfyf$BThgFpTe
zVNlntHI7b|-fY3n7mFQG{CEf9n3>ltY~rDU#YOx;-UUok5<ck?2?!<CYOVfhc0zb&
zr$2Hs3}?jLUNFovvt`(glT)ltzUA%;D6T-fZ&yLs234Gk*uwW6sN@gbF9%H5J_Kr;
zx(Z5XK={|^q*{iQm<c+;l9G4vaVi6up>vrhuBfAwK-v7n7ufEJ8SPfqn`~K4YiYUw
z%Z*;@C~kyXYcWj|7_f*sF>6ivA~8QI0<2lf)GIec<0%V7(5ZA90-Q+TJXbg@<N&Xn
zX}TL7;K~pmktXcQ2#ryk3rqw;f&h6|)q<K5L(@w9nKK7=?wi<k<7NtHMLaz)`!uIh
zdZx%w__!Bg!HNz4wDt9yCId(wIJuW%Kk_mh`d*1`rr!H}vyAseHL0NL4cyUoQ+jx~
z&g@UpSG@^$?Mo)(({&_+P|r-)jzKPua-FHqrd)a{Rc*XYI}m0DrxORjzfgbU8}*6j
z^|Mcm!mbufY@Dw0)Slu$0)Tx>H)<=vO*Sp6qb(Yy;{Kpxu54AZCY;1P?+JZg!M7m1
zaSgVYc3(0qK}jHc?V5mXLilaWZ}PhjH$<i`d}`73m=7!Z-i#3XT2$ahHMn`Y{gQO%
z)v{Mbp;Hb*22D|nVOsX=BCOZ>D3IYTfg|xPu0>dm{=C172DdGZb3njCKnM#=03>G&
zbOJ}Uw5rdP+%aD?xhJ2IvUu0bb#*49ANQR__pUV(zSi{5`^?FtFOYrahhT$}FL_bK
zxNPe-iuWzcw1kqmm#@UaG&ISwO19tK!o!GaN~c2n;bzER^~9cs+IGEyMy}YzS>i3*
zlFR|ES7w>O)qRMx0fpg+={DM*PV%>hfZDb!{;z%LKRoCFxrp=6f#>_fhdzY<xB<I>
zNmJ0aBS|1ktXa~8O|1D-f6e+8xF$n~h;pD_&k;{o1fpO@w(aEy9yIlnCJqTyKttlp
z2H2GRdBi$tM<Ch%KNIs1Uo3q3wa!XU<>NQrkK-mt9bq*oLa~c3i`*!T6jSNHJbY$a
z+EagfPZiPj!UrVk=*lTk`htjYlW?Axvbk$F8D%A$)4NZxs9Z09c{URjO7#6PrIk;<
z+c6=m@*qFUO3!a2{duFc<CS)$pt`pYHWO&s)Xs|t(zHmHlPIAmcm+GRfk|7Tck^y&
zquzP5<~J}BBiwL9im$73GwBC&RoHG=(4k!?yV1gszN4r{;Uu7Dnw#bZO#cNXRyrj1
zaiD(1)oHJ{=-4>27k5ZI*xTn;U9Ha!e#s<7!pV{6;w-K>wwnN%hZS12+^ZhKx%)e6
z&;jJj3tHA<WnUl*9jntap(U$m;Gdq>_QcHjQiUNwEyX&j3NJTpEv0*B_U~z(j{RyP
zwxq24KDu|MQ4wcUo%Iyqa3jWU@Shd<q}>`oZ`M3>w&$A@!qMmSc}*i1gO2AUD(f29
zrLA9j`+EO}`+1)j{wT5l!{KK=IxxgxGeDJ8mzxt488LAM#$V;Az%&lyKR_-}!^Xe6
z4DTZ{H^?|IJotfxy7Qon<Ah72o}_BZTKSXsC}P>~e~e}dN+xyOG)T+odI8`wyDqGX
zuueqPGq18u6{K~y-iy$*mpyD7;tG%9rprl3iN1ibOZB-eF4w|om$mr1TPheR<R0X-
z4eLF86r;9YCUI<TxxHFejl|bC;WrL4wo1kh$R_#)yulUngU!|L`)2UZ-aAC~30hOe
z{hQxpJ=!#v>pKbNihgjbb)0RRpkNT{mFZesud58%^5JBK6`2a7%juCW6AGqtVtF?b
zbiBoKhB%n_={vV*MTt2X&L#0hg75_8f{D@apRc+KhUnNZ34O{)`iHaEp)Z#REn#ut
zY$)@(#9VT5z&-l>`8N4SEue%4?;~Zdz>gUGV#GdN2<IiwH=OHI2^2n``@(?dXCI}q
z_?+29&Zs9@{KZyx;cLCK{ZSs~MBS^o?04+UY=M=zog#!y?_px0@6QJ$Fk%Vq&kz8~
z)7nA0#?!?_+(`e;<su1$922@kjr!(xQ{vhqcO$klwASK1JUIkK-~oY)!NWv3V`l+5
z=r*x7Dx`>C>-`Kd)(^kY|7dD0LrGwCOWnB1@Rf@aB{xA`fI5JYCmde-z<M5xLHjYC
zv`k=JAj8u9KHvqwjR2MjOb3T!#GmdSz~cc_@K0gj2i$34k!m1m9E&aJt%a1%uKrNn
zu5sh^xO`;&Jm#fm42KB80-!}=<0!xoz(LnD2Qa-u))3HIZ~-|~)`xirhc6|U%4es|
zKNH_{@O6%=RyXzGFrK>IsB>(Uw8*?o581C)w>-_fjc|V)se|K&)f7^i20+-|<#Hky
zREah6MAI_2%XUFgp|GHu{|+u|V!EcVwqGTIx;UjRnH^14$Y;v({Ftu4t;j1NZ!)oc
zvBTb}%b;uu6{%2ruenZHl>R;8fRQZStbQsnI@;7;9#POKb*Jvw;<O^Mxt!z*6$Bh#
zM^59jq!G}6+HMLSs5(PupNkz?ty~eq7^pw)50-vi<{n$GA8;Te`4~5+$xM$nTMz^<
zY*MunmJ5CLyYXtEt5e0FP0f_{HDYO%f;7s$hcIS%5JxkGG)1{xaj@Y}d+1^(2&*jY
z=!2>0_!mo*xJ=+UjkpW424NCC>3{V{%+y?4kSKqtJ{t%TPt()_NA9a6llRIxEAN^)
zZ3$GAQdzYvo~7mT=eY|hZ&dK^lXSym`#TRB6$n}e>T=Q^Em-4QxMM6?gAZsC-QnQS
zLTkOCBL+$aXW=a6*h|fu?}cIM!fkOT<8F4P>qQbLEcNU2*Se(y_m?`TEumLnwMAay
zBK~U)zuiUgvxk>_Z<gwoftVKV_y<_S`5dRmagA!lL6kUU51vM}RN@5aLPa-``|J%d
z{yh6|m3_H)Md$@}|2rjs!O`<^JJg-JU@-=G5p3$=3Rwn`xcAnnD2^U&9s4aS&zzwH
z64GGUKN*A0l9Ckv^8SO&dJ)1x;|t<$S2pLxN4qm!oCaCA{z#2}{rV#=)@oFm9N-=-
zT;PPJB@0)}Ew|p&R85$h=HmelbA_jbGtR_jof+{jNTcEtL+<!d{Xi7Y=1hxqP0*Ca
zF#aJrAS)Jq=2O){<<Mk{8>jHLgDsu><5)vecGy~{k>oz5%kp#V$f<{}5fi`hMP17E
z+Cf>=`!78dR)yy=Qcpr2@_ftSZVb0I*u~qwV%Lf{^XO}Y<__qCK>%!=BnlZM@r7Bb
z)D}dE{#N5Stxx4Hu=olXMB!*HBnx$0JJXIxBi5)NsohY2C;&>)Om&wOMa_xn(*xxc
z%8*)6{3!^V<cWBN`+NICv{AL!tQFPck%5vmGFJ`cLJcbu2~8+<hEaXxuDgv5AUu>K
zC#S;>HGT4F>`BLpowUgLN2~!QvNnLc!6=6|4*)Ocf2}>h0s8X<Q)3&QDW2e*SgjZG
zyr>FEGVh>J0GV=M@_+Y#8kzooziOk#_J?Tn-%0Mq;hwzDkC=5Co;xlU>t5;G%)?%9
zgO`gQT~ZhaefESlJ|}ou`OU(vI*BCmM-jr8VU>|#+D|u^=x`k=BmW-9Y_*2fM94qF
zxdt?>Mv7<7)#;kt+J4nU?<pNN8>9kr;%I@~kD2U#ggNDQ*O72Vy^1SQusReL3R_&J
zqYJ!%p>Vt^?Ko+N`5uv5Jb2b7tLyOC7K?CUUM$??ha;n^xKs$W*q3I@77HzA1=rh8
z8I~Z!EOn)Pd|U)qbK9+@+G``vcvdnFBa^ekcO3EyUe~&pB!uH7P2et_ZZ%oo9VT7p
z9ITjy(lMlzqF5m@D^<NLv0MZFn(Kxbe})TlRYxR&#=p*x67yP@CIsA}t9bE$`UH&5
z4Lx+-OLW34!}8X)>5@{~XP3BY3Kk|2RnL3e_&wxP@@>dd)EbsoA5kpTH)+RS^&~c#
z{7>nb;4{+>8|zE|v~?AKQ6Q%vF{K<cU8B6l9Y<yKDTB9s6IANo@~hw=!E;@VIR7L8
zL)*5Xp-kB*?3}g8n-OrijGn9XN6h#QdHq{pwe?o`Gvq|k@4tL|o^#XEU&AW2I%Ni_
zSD(!?A2jD&2?+h^T=}r#eo()(_S-hhnsRq8;_ZQ@$PEFPC*ltg)aI2=^P|;VXmjC-
za|WHFZw43Y?!OjV(kCys08(RsP<QvSL%TV|?in4ibk=y%*E_CKd#1Y3++~?)UL<17
zs-=bmxD{3$9R>Sb@+2^`qbVwOFz5tl!g`f}mNXHC;7uEHa!jm&x8>zhn1TJbujH&e
z$a|%6%|05fMF3JwBQ*xxjx`h_W!iR<O>b&EZmYOGlO%zrHFK8{9)ggPXQ_b$j1E63
z?u$b+&bQSOHxu&jf2SI6CwQ#v84(y?%AdTixgJwofoicXw7PL6<3qU2!R4{^AG)1q
z<9W<p{au%RuLe>RhOur;jp9gG{N~6_oegiyKEX>obIxP~EvfG5-KG<IqkAItH7&=u
zVi5VWke@(r^%&ifyY=1dp951akEA6VE-fjp6KWve0rPz}ljiUEIjm37Jk}qD$r|qE
zt|~2jXog)aXz*Fx$vX4&G?jhO;+@!;Ny!F{vL(NTaUENeCj(Xnmw5tXhaP3SzOkB<
zd1vErPUnj^t+DO9XZc+fQlWbVU~fdNu0C{`hav$^fG-deAuQS7#Oq5O?Pfk5wRpN;
z4+IX6oYo0SK|4sJ>6faF%Jh8xfbX#Y-w$92n^LqxNWQj*qDBDvJ2Z}lW!n9($NvwV
zKkW7owG2Q(7!W4~-nyPb^&dwIx6n60)Lp>f_z&ebS&&lq=N<=sh!D6BSwYbNRS3Av
zLFezc&07rV@2I2s!F}O7Z^1*1E4|uv4)fE!olj!qVViEj($7~QWt5oeSxPtzd{y|Y
zN>MxJxn^H!SpmEXY24ZH+5{ZghPr)F6N^-rh{_*{#OrP03q~<N`<C+w0bl`rBC6iE
z<^#B6w=7M;Q~BZ9?uNN1GsKWS3thr#S3V&(#?$5MJH^m$`T}}m9r`JjPPC!qdQM+%
zUiiL^vZ7+fBcdF6nsCQ?=)nXhK1yiQvtR~o7xed8L%vF6ugw1bbIvJ$a?kQ#D^#vl
zb=`@MI;$wmkfR*GA8~&0{#Udu+bg2^`-ft|m20z!kp{K<q)qDIIld+I`?Jp4jUPEm
zk9=bMV0oxEaz}O?Abtfnx#%gxCvOf|4q0K$L~FxPQ^iie<tH~Tdn!ba6;RbKtqTi>
zw26rMEu2o>;Um#Y_urL@OWa>ID9P)Q?@6CDQ<|wc&3D&xTfV?1n_w4G+(oFN&*#aP
z;62cTIm42=z4fsl#V=MtVgO#n%(EutdGlJ{9=L=Z*}VoQiO(Bs<eQld%_!YtY3e~W
z8s2M+InFdtpCxf<%SDe17uiWn<a-BI$;O1<86&DV7dS0^4u5BJ*~hDFa&!=mEbPq5
zxJ<hZ;rC5s<DrwxGYh(%J*S1t?A+w9dPmbGblPzRN+yu03aUrBr$gJuaUy7@{X)q#
zF7|+8fYf^|0r1S{Ni`!`in#bB-~$G|$V?cYf2ijUBBQFKDGi0LBS>nZw6PzKxX8U7
zXXe*e?f2qx{sr}5bVk(JRZ5Y@>&ekpp5`s*Fdf0S`~W|64+iIRO4xTdG11Vs$1s{+
z&ts14xep!i^tcnB207|zuXS$q&@o{isf7+3FG}CK=SmBlIvrA}Qz?uIDnDm;KL$$F
zj^cg#kCu@H;qAwo{Djna>7Rt|xuQ$;f-IkN?CUaS+XS!J&ta25u)q@=YNx<P)a$XU
zF))+wbv~|TdF;htAAQXeK%Qol>-+>)cT69+qh{^M7Vvf4m+oS_L#~>DNkzE#INaQ^
zoNT>H$X=|%yJb|P>JqiyY~UYlS+6=>a`vm|cd8Y$f@CdLT=LhuCJQjQ%J)37zO6UZ
zxMY@i-o?5zVK)!m?byq1_R?B|lH`bUO4ue+UgB8CWJ#-USnb4gpPK@vvs|z8Z`nJ7
zCP*`{+9;I<=yJC6;aZ-ki)IJ(NnjYzrUM=Jf6QNPTLPyF?204aM`a2%wQ0&dwTYTC
zJ<>ecLI=&_{<?3<QyTJV93+E5JjReXMi9IS+i0F6HxhBTc*Mzz2J$b(88Qp5iJy_Z
z$Je0j#Y9uYK@EEnKj13MW=-G*e9b|S^kl9>f}a*pDr`WotIR(zq94$6cy<Sd1jx_-
z)zh;TutOkp>#;9_xHx^>c+AR;{u)21n{IPUkMec&X6LJe#K{Ces(dIhQPupT=pdeU
zbzvAyHy*(*fEjxZb~2zL(Bw+9hPgTE^wvVy9RpHW0XoXgIjzo0y%dOZ73WT}yTX1a
z*=$leH7L>LLmwM51v#t8{?6p1`&3GRp0)u+7e|8a*DDpoOT$g)UD~3VCHWN$9c%J{
z!2?Ah&zuY@=4L8SGX?YoA?$}Vzba9s@BT$LM`b4(Lwlq2>6*I`Rn79kBbN704&tF=
z$^EY=xWZ#vovL1#dv*L5Di5=av0?or+Xu@H%wQGL-kVAJxv7v>%DIkIL^j-hM9G-B
zV#1Bj72eQhO>G$B&PJnpp$L{)CD9@RTYicMpR=+siN`!z8J83;^Nxf}F~zl+cP3wg
zMIZMS2X3`7t1}SZbI)rO;!*JwOSu)uUgS?BTXWZJi3kUMU@d`lgO6#!C2C&M8|BF5
z?9-W*WGUGtpt6L?hE+@-+q~;(;U7XO98X$^Y@bB-Qbx>e<r_x7N66?7GyIXZP9DwN
zQJ$@`3c}#uE29#nA%u2~qiy%ne%YI!Q>(%qgM6y~7nCgPSto>;RT#4(baHi%4=q`_
zAqu-CW6cfgI$gqg<Eq^m>kqPuPN3#Itx+4ZS6xM-YSk5m5@UbGHN}Rfj?1)FT{(cb
zbA4v};2?zN$rvyihdj+Ft7Xvd`hIZ9%unJ#uMpaTMsm;KOSx!@ZnHk%u43K@JxQXE
zxsaY5f<s0ChTzT4CZ;E#C9&yP)v2P<eD1Y3Mf;6Izj2M}!*Pkutl&~WdCge`!R|!8
zE#vjSK3I{(gCTP273)9acR|et6F^jDR>YW(e@pjie0_xGyy|*X>STt6yZslv>~q-8
zD%94_4T-<vu!E7UTE8yose~|39?9ssrJdh;&vj?oZoWp!9$d8hv_4Yas@kNYSLt<=
zX*l>{;=LPqpUB}6S!Q}sZt_)IOsKLz*YQ=Ym2G5Q!a4r>ud<38_?lKx9lh0mK{jp&
z7cLpNH4NnOIr7<=@6z-@rn}>o-0#_w5PlaxVnNy!NlzBUH0XQhIk9^U_Wdt#ITB|*
zEgbrTg!c3-;5rA8h8R@)bh(W;2j=4zj2tkL7C*xJdu%kRciMfe=p^zADbloFJ?iMT
z!avbOk%0OH(7&M4`tMT%2Tu~*Xk(NK$)A_MZ?!$6d);_xW~DBxBM~Ec`NKcO&lp~L
z%V$fZ8oY;a)0!EvzZkXd?FW}fh^fbQB!9PD+IUx@wS8>iRF59F)Q~79*S}pp=m#qj
zH~He{j86HMI=WuuwO<Yl^s{Y-`RZqToT~YwE?<u9X+F)dU;$!4o9-xtterg}Hw&{_
zAz1M9W7@E6r*hkToY-s*gTf;=O}S(}eE@^qvtaHJ5CjYofZPOzJvLL)As0vrOYo7V
oQ`ielX)zSezwdEnE;<K<R;jTY0s@HWKkJ4+WyolP=ij;i2cHTVDgXcg

literal 0
HcmV?d00001

diff --git a/assets/sample2.jpg b/assets/sample2.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b3a1ee089fb52658730d4d28a63afc510755098b
GIT binary patch
literal 181758
zcmeFZ1y~eq+c!QNuylj8OE=Pu((KYKy`a*tbclqCgd#2>Ei56ibV!G&Hz=_*EU5w_
z9f}2V+u!1y@B7sKJl}g9@A3bR|KQC0_MCIBIpd1+ysjC)FMMAC7>o>%1^@^I0^X4R
zfbTCL@AM-uo&aEK3Wx&$KmkyK=m9VZk&*T&5W^2lNy0H8&L1=o31dNAzvB#$C;%bR
zK*yXQ*&qB2q~AD@+z-q`!ZkNA0Q3hB06a_-`VB)kNi@L9+{^(0kS4nNQp%=s$CP8q
zNag){4syW+T#yd%xByp>mXlTl<YZ)()MQlDWaZ(cqax{_0ssia;eL|{>>XG5I}S?Z
z|3kV&kzX(espv$RKhfoWlZXRI0{|H5GE_!RUM3Cln+D*BH0bX*IRgYx{URZ_I0O9q
zv#K&6ztbBtpugkJ46@&OdNN?YK9^+Q42oYdsjv*nUopvc8C1XX6N*8<(n<CxrTd*e
zSxWyqUMOYwm51bqvR~u}9lN9q^g|btvjG7ATRF!V{Ks?gkUwGQ?`5vw$$q6PllaO1
z#PcVa@xRG@Tqf0@c;tSRL-IIy8?W$Njet;)*p9bjaokD!aV-P=*_*Ozva)Ilvj4@y
zztI2>=QMy`N9(uyzJFgOx$4*rKQ<BoHUCmHSeQhig6J2&BUM`9w=;=Gvh4Bx%k@9b
z0Mju)Y5#FH(1$B3`uoETWTjQ$Xl)W1IIjK2GDE>XB>o{c*zEYqFTN+48j?)X{`e}`
z?wJ0I9gpdFk|mDsJ>CI;{5ShT?vp%o+`^7|`T(9E_fnJa;txIm5F(wA?_vJsyhQ{6
z)_M|-{FScq%XyJx185})A8*I}$7LR$MSeZQkIy6<tEkD4EO~qv@O_oW9;vH)-U4NA
zfHcu3bp`-HgD}Aa1cFJ{@ejBVf-=;B+dH7)uyK-Hlq9KT0l>vAG|1G-!1}lZq`odZ
zl*Dw5|JXjv{i5Zx6v6@y|AYRQ5TjdA$OV$979?(YcaKmv623;le&H8_j_Gemn8nri
z7=uam3SbE#2}r{1$Jpx^Ect`y7p!%R-Tebd^$dJ0bCA2g`!Vh#;lzt!9wZDgA>o9J
z7>{rgo*`jDzc7Cc3I9yOEdCxYp(NYV9Mdm&xOtPX3<=YOSfg}FScBA6(s=y|yZ#AZ
z@Q5JE2>`l*L6IRCFYgO*Q8zKTtg5OC9N`h}=W*eJl!c3%uS<wKTsP1^$R&W})n9Zz
zz6)SFwk@1wWF=WuB_%02X_EPW-~Q>v-?{!dIQH!?9PcfE_zc80`UmZgd;g#XmI8pv
zA;~srf6!d>0HEVK0I)y(gC<xE0QA=Ypnc*``mi0>7w-!fg49l(3J(vL#(20%9~<=d
z?O!td&iRkQpPnau{Jh_72iNoPbP4mj06#XWTcBTHSO`2c$i>YAF7@9l@qc{dpTzo;
z9FiyxPmd6f0MesuNwo|U;6-wGfIH>_CNKbw3HV1I{ts{alMKi3mwHVC%=m8r%Xw*l
zVS)vKycz|d%+vtHsfct1^t<0oC~bk`mS=Zz>z8^@!ldgzj{oHVzD4>84#jxEkIA|y
zYq(oj$i-t!Y7@r?GEz@U53m3n052c_hyqf89H0!S16qJSU<8;0)_?<W4sZj!0e>I_
z2nQ|$u|NWlL>hLp0UU4#zysAl1JDd~0QZ4jU;r2aCV^St8L$Fu0DGjk-~;fP6uiJd
zv>;{>2k0b72qXcL1F3*cgY-crAQZ>}<P7ox1%SdpmqGEMBv1wj2Py$ogPK5HpkB}r
zXbSWM^a8X4dI$OhhJb0nEMOk6F!&T$1*{D=23v#Ag1x}O;3)7la4I+#TnerOw}X4Z
zBj8!^3V08ENE$4uA#4ynh!jKxq6;yHpdp@+5J)rx3(0|$LK-0VAcK%;$O_~o<Rg?E
z$_#}=rJ(9i1k?uV1`UQrLzAI6Xce>_+6SG6zJR`gej}qL<0g|JQzJu?*^_yZg_9+a
z<&ag7wUPCc&5~`9eIO?%XCoIOS0+c0JCOU3Um{N?FC@Q9-a|e`zD9lsgTXjp5-?4e
zIm`ui0hR#EgVn))f)QaGu%9WYDNa(zQ5aC5DFP_sC~_!jC>~M}DYhs+Q!-ErQL0gz
zQ@T?|Qr@D(Q{JPTpxmJROvOkgLZwM%P31!sOO;F2K-EvRNcEnYnwp<lmD-XTLmfk%
zOWjCKpngI9k%oyzoJN<%i6)FDm8O!Whvo^*ds<prVOlL(H0=f2RN89V$F$3|KhrVO
zNz)<e-07~;<<qs(5$X2nDd~miwdv2%N73idx6n_}zht0f5N6P2aAAmM$Y<zcm}fX-
zWM-6QG-vc@OlGWM9AeyNf-wm*=`*=AU1KU^dd#%O3}F^v)@62Mj%UU*_cL#>kh6%e
zAX$7^l340l##!F5GO;SK+OkHl=CeLvU15W;39}*D{Mc@>HM7mJeLBH&Li>criJK=H
zP7qHVv2(F&v3sy%+3&K?vVY=$bLeyUa-?%~a4c~`ImJ0GIm0=NIQuzYa<Oo!bGdP0
zxth72a)Y_Wxl!DgxXZXlxes}Gc?@}ic=C82^StC`<JIEz;mzWG$h&ir`Q+)7n3Gv2
ze>%AfXN7CS{ouLqKKL6xZaxHG7+(qBINxV}5q=x~IR3l*O9E5^DgvGYSpts)-Uyx)
zG!?uoSS$Ech*C&J$Xn>P(4f$fu!yjo@D1S(;cXFi5u`|zNUg}CD6Qxj(O}V1(HSvv
zF%>aiu|ly)aj>|exVLz|_=E&lLP-K6Q7ADbNhYZ(86a6IIWI*cr7d+)s#fZSG@G=E
zbewdD^!_QqQ|MFarv^`blTnoMlflb8lVy@Ml8uwSC;LuLT+U4{Uv5U8R^CAVs(h#X
zTLlRP4}~Iy1x02>Q^gyKj}*TuDJz92H7M;U3o1J+7bwrGu&7w5B&!fq$y9Yzud3cx
z{ivp-c0sLK?X|j;y03bT`nHCMhKB}TV@;D^(?zpHbLBMrwDalW(<^8A&bXW@J+rDM
zpyjSrp|zzgrtPC$r@gNuqZ6XjrgNmLrh7%VM-QTBsE5@X*JskV(Z}g88}J)=8PpoQ
zF;q0XWY~isLl`5{5%WkMq#Lpt`Nl}u=!#LFF^#d6G0u3^MBF6A<en+m)W|f`bkR)6
zEWoV8956?kXPPfrh*$($bX$^JT3F^=ZlYvSmr;XO%vR^EYOFq3>sa5ier6+T6K3<s
zmfqIMw%Yc{PR}mGZpHqT{T2HW2Tli!Lnj)BwnbN<4;}R#vmG~_6rB>B=FW<ojW|1W
zj`N(~xd-Rz&%2y&aVB%NbFOjz;$rSn>T>9cbS-dw<EHC&+ilNX%RS3|$3xR2!(+=+
z(=)?!+e_0c(`(23jCYRrON=fChk5IR@VVo2<ZJF*>HFQ!&hM^2g}<|ZR{&FhZ$Mul
zZ(u~=RFFhaLeNUET5xvon-JrWicnDK+0f1lETmv+G)y$?df1DLr!VGT{5jk<yd{D$
zA|PTkQatiz<W`hHRQV<7CAUkDF7sWEz5L>e_LY)oP_%3GqpJc}uU*}UF^s8-rHb{9
z9gCBR%ZNLOw~N1bjptg-wYBSp*J~5#5<(K@Z>Zn6lL$@3B#z&dznORQ8`d2=oFtQ!
zlk_FoEqORaHYGRZd#Y#Z_${SdMQP+|0cmsTXVR-O7&D?W)-p{q+q2+V*sO!>^Vvf=
z3ORRfQ{4``y^?E^+nFbrmzMVh=Z%}q*U7(Iz+G^&;Hc2OkXWQu)NqIAPSTxE#hBuS
z62p@AQsL5^GV-#EWt(_g{9w6Cd36P6MN-Ar%7Dt1DpXZpwQ_ZJ4R_70T5xSx?M|Ij
z-DJI9eP@GIL&;sXyGe~eV_4&ElXKHtvvKpI7L}HUR-x8{Hs&^LJE%Rf{Y{5=$4aL|
z=Tw(b*Q0yt_u9IpyDRVW-Oqo(`ry_>s)yGee*Y=zr^BA0o|nB|y=#xoKYI4q{_#wo
zW#4$eaX(?eVBqnf_TWzhO~U;lwV|$ImEn#NrIGeg#nHAg#j&<=rSbL&<%!No)yaEP
z8dDF6XNbMideZ|l$eEE@v)QRRo4JK~r}>oy_l4~zeox*$z4-L=vzSG)MeGv8QqD5>
za_Mui=XY0>Rvx_2doj9dwfbz$ZEbHobp6Z5wN2X1oUM~v)!TC0_je3-CU+foH}-<|
zKEF(O#rUdVUv$6qwbtv=H}-Ef-iEyW{x11F$NS0yg@Z?j7Kh6p{6Bm?!v4(hbJa(c
zkAt7=K5cyt{{s7x`&I1gy>G_f7Qg#{|Ni|MpbOAaQc+V;&{9)T)6vn=Gs9V#nV6VG
zxOq>&rNv}rq{XBp<<*Q%%PScuOG;_kX&acBp{!7Hr_nBG3uhxsl=*QXAUZlaW=3XV
zR#stiMJYw|e|!7h129sM*}-BUAU*)h2!b$zzV`tqNihu>DSRMp$Fa)s1_DE%WaKak
zlF%fEHqvwl2m%2^A!KAwQal5?Lb?t>8OfOVWVOkeEnQ&z!7OsI=_M2bI!(Q-sHr!C
z@~$CqlvHdd*f}_bghfQf#1#~klvPyKboKNN3=v2pD{C8DJ5r$K=I-I?<&E(Py%2UW
zJR&kG{@V3~8;Liu8JStxIk$83aHVDV@`}o;>YC=3*0%PJ&aQip9{2SR3=)QhiPJN)
zbMp&Np1xRJTi@8++TPiH`|kb0;fJH2KYlvai<H~{qSo)4{WrZBNqT{yPzV%utQQCz
zeylhnl#EZ7oJrdf<`T@zFBeO}qLW_I)JrKKk9x!E8Zt%2CaCa2=<Ts;KQ#MmipBj$
zn*FZWA9_6nXdt9XB}ND%a0WQo6$MITsE!ZtE)Mi);DfU)?fkmL3ZpKth*6#|CCg1R
z*4#%}K(rYA{~!Lb4JbA_j7HL2#sc;7z%x5&1fo#fy-pA%SlYsFsYS*D{D1$@074=g
z*|6SVtTT%RNFVqgJ0P_Hg;pS0t}QR<nx-#u*Cav8D7*Dq5~WqW3xC=t3EAuaFPIOo
zM1V6&b*a$7u2kYZ|Ff+rZ5q_oufjdqgRT$dHSj5Z**6@~Tg}$to>^|zmW<Wit{eCt
z^Va{M<Nt%A;8aF8X+$R!#bR1*AN}aH{>Kh@1rSXtTZv;EZJe)tZ>vO3c03K1g3xd9
zSlpD+cEZO8`$WAInNf)z*ZqIUpa6MBOIjZuwpI5j=95IR_OCt>kg?1pRwJuL{i{Hq
zaHb0wC(KM-sx@cuEI)OQmHyzcl9VS=H_jot6o1xirXWyYSM(m8t;YFh%1G@mw8q?T
zkw@H|t}TLiDDO2+sMLSlfUnKSn=tVDls5DX-Mhax=3yMA>&|&lMVuTqnJ0#U)X3Zj
z_19G!)ZVNYCMDft&64S@8RM{W=s)1%ZWEt^|KE1MOfYoHR?5{x=EGu)Q_OQckb2e~
zEe!1!m(X;B>mjt?CX?Uby_QR_=(>eyvr6k%QGV!Iti3ayd+Le#r^i3fsLFLQiu=R_
zt(`Gy;`w^NAmHhho$4cg>*qSd_Zo*A71~O&<Nh}XRbbjzY+pv9|IGa=$5<)WN%)UQ
zpbFoq2P>EgE)NwyvJiX)+ZK<b3zuDg=PSHF#?;J@+pyEX)_!P#!;3`y@Ch6}e0J4}
z?y}r;t=K01_GKUItWP-1%@VIp=iKFxE4%mYJcSo|VtEhkgb7pa^RxCsJ&R}Vt6`T?
z^>P$5lxXVYC|=vP$$5vdQ2hIyq8LX^3CL`98?K_585Us{+3=KwkSSM@ms{jhTYx=q
zt#e=ajC@KNpStAR)YxsD7z@L#4e0FU-U}?bn0}l8{u+wEQ{k}2Rj{<!<v-|+j>kZp
z&ahg#7Hx)p3bpKx&AB)1M*>v%g!H`dD*|HgCd>}8M?}iTeN(+B_xUx<9Lv1kR(u?d
zl7%g|bd|?n@)B$}e4t{~_u^(hTnNcQw=&4`k-p%*(%h@pD5%jp0srV=D~XmXkMH>C
za^E()0m7!E**?`)fyVV!hDwE4#w0Fnd#(kM*W2zaZe{sB!n*%&4V55#RR;T<3M-9&
zUadkKxkG6O11aLs6?$Yqxu->xubc@Fpt=gO7VI9rldqWRIdxNe>$wuB)Hp6_XiVFb
zL)lL6s8XUa<%+3AdI0;Zz3dr9wC=l%s1JA)6Ed>riw|G$d$*u3JH_rN!Y|Pa*YsX&
zMua?p+|?9M&nhZE)1ffm?<(M}f!jNfvh<&FP1{t`ydu7;-*919S5#aZR>IOc^zV)n
z#RXx-6z+u#VXQi<M;Z`EV&z3+Pn-TBk$TJYBwk-)C6|CO!QL}h-w*M)|LpYZvDWy{
z4|aVFaY~#Qz0PBTIv8PA1=yx@g0qDw)1ArkrV#QB_D^E}K_EDYQ}{ey_O>EFS8F0y
z>|N7B%<jdWwcx7f)*ajddy7^ZCnZEv&by_VS<gBg+PVUMpfs^AeU|EWr9Ioo5~?Mw
zUV#TF$BL&0$hwWw1L@b(Haex-&>pCdJyii4<*xUqnjdZu@K{?_4pmkmwvsDru&O=O
zE5d3lm7J)gg*&eph@WuxTIO9hL~og2>Z}0o%gj?{){al)*u7YQL+8aZ2A`#dmSFE*
zk&(eYGr7icRw1{Nw(f)-h~COpoB3JF#dV+YUVZSt>w#a>lK;!wDl(D90_lxyWZj-C
zGlDeT`raT*Q?*Ckxhodre)npWv{d>%m(Fvj@rnzMxiMuJJ-@#dZ!sd=JEmQY7L(sp
zSB9@prDEPuj<E@4QAcLPMAXi$T`sw`gm%@N>cZTdu(BEoYKosQ%1GDWM@ok1+hUp~
z6p@t82_+XC;^GX*tUhC~&wL8x!%oj)I35hYMGC*zINJ5CAto7eR)-8O$K0H;_sMW{
zp?H2Wc;}%0^QEc)s|#lIRr}(1gN>AgOXL2nw2P*4X8mQc4Mvf}ajx4%o1=?$dFR<&
z?fqom3K>MsaeaN%u9wVT8ZDm2>Uo(eRaw0WEmQ(SmYic>G}wwK-n-71#dz+tG+W_#
zq9z%>KP2bh9lNcSl`WnQK8ys@M;{P7*>Cgmdxqo$gJEwKE$^VZ)1u#bLY`s6+Xt{h
zD(4If<k=va$jv&Z#i)~W6&vN%_Gh;mNPSzUJ7>;F6Enh)JQo>Yj?TD76TS0=)+lL@
zm`3@Y4<5}@%Rek*#u)=DvKcfxq!u1h1|cZp;<|e`?{KlMylVy4gy}V~A<d1w+ZYgY
zNw0;0A)NSZSt2gu>jqx-b?kdFc7?|xJldyQW4y2S%#@mrwDl;^@aGs~lphW9<~78d
z&t+IJt*jg@eNzm|E==Cl&Eo>%<mj|Z+f{Y`S4M1%!lRj<IIl;zgt3-jqequ;))sOP
z#t7FKWV=~Md~dgOJ?i4v>8E{if~-0(P$T7z%zGmbH&IlCVtSK!!K1;+*~pV3<Y~hp
zi-1zq>xl>2q0$U)aXxjr?reQ+gp*6G^_}{Zt{53T%aJyuNT!9UDqT|MgfLD!<iu?O
z=#07QNf}MC1v}XWkyG;R5yGWQcXHdW)Fhr-IHkToL$%nNRadl`$!)5#-x#UNI~y?m
zF(+8iKTZwhm~&|8f(f~8LCNrbF^Usc_wQvsH6Iu{wDfH=jbQtV7j(KE(^jo-n`WQ6
zA@H>3K}%|lZw;&U&0s!n`-@e1^udnROa{X~FkQ4TKYjC%d=Mq8ntiM95v>(&!ALOu
zq_mbHyW%{iC9-`llh7dP9>_8VcPhO<ZSimRGxBYww(2b6GQ{+rgHQ9P1Qvbl$dS3(
zN3})-qJ(j(7b4dn_n=bwc5ig>n6aq)J4)GSJmRf4G*-*AnYaT~&b8KTIASzra7+Po
zL26SS1s4=z!f&UKAvqQ<b{J`?S;nSMKd=g*lhtD558AaE6nLP{?b5{(lgyB85RJAH
z@d0kGA5_$BL=bvQdl1s(y0UBK`f}3W0a<#@ZSX10nYA<5Pd$ykV<ah&n!4@f8+6I5
z+}`1|AFnu?(|s|Sk(gQLVSZ*D@A)vV=0O+ZSWeP+K)WHG=PZPe%v?tz!f1P|mP<WJ
z-y!YanzFF=jEHycoT%^yeZD<ob$Q`Ca9^|$me0PsdxjI!6~i4_{c*}=AtFP5YUZ25
ziuShl@2SRrl9@u--&$l^lSNLjvfdo)r#WltQ?Ixf{D#gHF?z+L?`^G^$_JH(=rw0h
zxaweabvmAc%|?=Z5w#vj>eW4^{R=i*ye=bxE=|Ui$@_b-Ys@GIrj)gz_;M6xX_VLE
ziFc|Nuk|^vaZkBi;6pk!Z}F&B3Fo$M#KO3(8W|PxD+VI^wZoMN3p8(3He)$lCOBTt
zRVD`vUY@0x(%pa$mfNS=?uO7Bb4p6}I?>D~N;_}?M2@L#&X&ir?@t_g#eY*=7}(>7
z8PHUw>0^lNNV?K=zJDwHH1(c(_%7a59Yp;^>pSom8{67xuA}eE`A)Ahy^vR6xd5lx
zsOp02xOrPUgbg*kM9wB_S~MrX@lmU8)<MZ8eRx^Ceu|3Bf7>q-{rYabfXcq#)H#Q$
zf6L>x(bZM;rIEpD9)ZLf@t47NULt}IuS|?TySdR~``Gog3UHxftS<EE#Pw5@<3Y2f
zE%kj4R%60~27-Ja8hc-2M!Q>5>EBm*l$2^*ew=)<nhO6I1{&1eLs>kg<UUa3|7qmL
zAP+6yEvES1io6pl8u18_dyjY|c`e~0L7s3}3_X=vC<a|%7r_FUeDc!wLwo&f43E#$
z>u!a%W|}?f63xA`_!dPu8B>1CNb&@Kq?+?)?%)}jy&e7QK{VVho&4ok*J6S7Tjz~T
z$wsD(bUtBoWn8lBo?uduN(G$r#I`dnGO4<s%RX5pwXn?2rl#A=C7>uY;*HP3;D6On
z^Xbg5Wms(;e@<JG8B>Vs3-Pfu<b4a*Sc8UI8a6m~XI-wnNLVer;aWb|JfQS!AY!#8
z<7;+aVUp~YqO`3?6*h3Gdu`9ew;J}A_mXGHPzv82?=4bW&#P#AUP2`vfac4zOu4#I
z5uKH`mv9m%cJd8AUw4w&`qnu9=EZy8fwtb0DFJ9-rQ3pH!pfTYC0~5)hvda{o|YM4
zJ^#am2O3B$cC)@_)2U&7QSf8m^a<3`Snq%`py_*O#xvb~|G5ILSt>x6XGD#CT~$$M
zD?u?|(WozCEk@pw=63H6%2M0}H5a<)(C`93M!|vJxKSkEZ!;qVh*`~Jy%Wwi3#$jI
z4%L0m)7X2}C2|Mt&7qDUzjEEiRzzeuHzDBT$IX_8#n#*jmc9YLi5Z&_9qNFCV*Nb(
zb*oj^6PX1>4}Rxefgz3Q`f}iQ01T}wMT^Y7k!3*Uzw=NKH;8e&r}1Uv&W3g2wL|qJ
z>d%*cW^)XOG7|R1XK6<NdiqH&IQ7$#i+RPgV+?6l??*gP<#YRHASK7gTRz0B$cnW5
z$qnN?A<0X)XH$3ldUy|aTu*elsq6<@o<mkyc>gxz{ue@9(cSsgf%%|K!xyl_1~i7_
zG2vCtpcPu*fwMPEzK&YqoqCjOtB7mP{^Lfx1HoKW)ingUTDLRrAxLG;n~<R0Jeo1s
zWey{a7z?MJa%oQQ9Fkhx>8T0LDrM@ugX$(sdfS#pk$7uIGJEKJ1;*Yq)v;NNb`Jj;
z5g>2$*u+Y8mzFO-VvkVTBQrFQ3(TMKs%N84=ZNT+8F^ZGvvV+QvkaqtR3VY#fZR6o
zvG{2EQwoq9`^lZSKd7s9<EC6tgd)F80B4$0jw-KG&zaaAJLfF=ysM0sg4Pa!l;Qy>
z3}btvoyJ9<(aGhhHhRe<%*1f?j%IE2J@|gAPkL({woK+P4A6pf$bR?pdtl9}XECEH
zOGToj8LPpJ!*Y)7Aw@IaKAx7Jtrv;?1~b?kr?|?wk}FyrR%S0;Vcni<A}7&$ozaiq
zR&%(j61dlK*_9Jy%~!fn92}vy>J=JZ#G_>IcupX8qDS~t1Rzh-2HRT1-^l-L35w_(
zI)B%QJY`H7oih)@pE`MDzr{Lj|Gf7SuWMlby`s{#<OUBE+bmmEjoe>w^xv?6#nfyr
zDthZXd|W{_NUe(eoMIC*@bLacf?(_`YoX4?F2D7e(K*r5qjJX#SS8E*;H^`m!0EP3
z$pD?|ubZZdw?+ofZ=qyUzXRJZ<7=rZV6U|_^%eN9ALJEeYl_FI;x-jUR5W+fX`W@6
zi-Rx4_Iv3lb>x3BET>>Mds)4Zqd-*B;cCYu-raPG)!fDJ-3(EkFoN=n5nkaQfM1=9
z`fLTZ+bX95&dil(T}$r)A<Jza1<k}{$Z$c|c(eV|+#nT>^*|V>`h)r!1@{D2QkwJD
zti*%QDXb>xVSH9VgAb%gpM~SEu8pt{C(}b`bC+$lO~b}VvmB85KVLTzB=1@@P*<>~
z5iO<)pEsrP8sYvlFBK-4X@u4>jZ#4fb|r5TCT+{({nO1K1UkMD7C}^s(@I5b_G%YH
zSz7;UG7cj6cIfQB_dX-eEoGe+8xV5IcEXf)wGON1o8Vxd9Q|f?WMIiNcyb0b^xS~i
zr|PvH<!$K2;!HDD4ug%?N7n_LVH!nQ2{w`W;pehB3U5V=Ha75H^iEIW+bWkJBcQI_
zS4I~K55J$wPRN7j@*YqIk4KcdHk>MuKdkup_<;o3KwhLGN2bPwF?GtOy5)nINr^<2
zYMXZp#nRA~NiLPm7|v1aH4W>Vr=ZG~lSD=rA$CxW^L`!G$;Wf$B^Bq|x@@^_-+E1N
z)cCW<@cTjjX75?`>-Sf_?Mw*l7ah)e-@Emc`Q`hQ&*6JtU+kf+q86~&?b*^c)kps-
z)e)nlq7(rvV9K?-M3Vt7w|3dt8&Qwmjj#6SX3mpR6;xxhy!hJwETxA<(EdP-*pQ_2
z7yR0CZEefRG|$`<#mUH&o;WfmL>3T?aAl93P<SpHd!3ynMrrMV>;{GWIEJQS-a$i^
zCF!EYJJP6E?y{I2nE3fwqk8oi8v9Ooow4W%9_5agc*h<(%_?;N{Zpg6Tr2TNb@goD
zvU=j7`#mn9N2kwF%`~LUjyo7;FNhYW{e{ltzp&5(x?(G{wu;`mI@h_Lb!a;%Pj`2{
zEN#U?#Mdr~eVm`Mb<Y^iU_gjQ1ru9}HVNgDbz^OrD%%USGHg9Oum|qLA}a8e_jAQd
z8!Mp+JFie+qZe+Tb(^`@UFn6ZDf8x;yl|$fu>AW6HSmsL>7xZTDmTW$vT{H+M(EON
ze$)#->yzbxRn~^~PQPNpI!xq?SmFiWoI!Lq7h^gWFkSpKdA?d9L5DtfPyl-Y|2Dn{
zFO#YxA3qpHY_tkuJEM!!3N(&r91zg$VWn-T&|4F{6@}<7%t^2Lx=`V4{dMl3{^?Yn
zsl#=an}5><{k?Gt45Hwvek8kNyDmZ9h<e4+AiGeG2O%&Z)V|cNZ&7zlEw98+w9?m*
ztl2@nIul=A=S|Nh8Y1olS=ACedl49b3jG(?l52OFuM|DP7}{9fs#mFPP3TZ@aFmNP
z_rdO>8r#iTYn44TPn$R*)N_RIj!F9~eRzfvD0txNzJzNFWM`E4plM!$M{2|v603*!
z7K4!xiSA@K;S%Z42_yG}m(hy+L$UG5ST1KJQ+S9NO%`&I^Xy<lbZHEk+rm1EO=S$;
z89>-W(3zv1MNd~Gzrz5<lk_XzV6BjctC>7eMCwHI(=)`y5O3<#RM$?ckF({OX|OX!
zreJqMpUef3@T*2s9i33G3bk8C;|QYUg^}-ol-rl!r}YkJWinLVRC46gzs(v+zN(jY
z>+V37X~^tsUlI4&Rz<foT584p^LE$R_5cC+mug()IQ)QL!eYBsw>(5PCx~f_{;RT|
ztT|bQ3m3Q#>uU4t3#h#Hm1v$Unx1QpuYjp;=5pnH$O{vw5F;805O%N`*c7?=*vx%0
zRfq}!1$zMI+)E81vq=xX@`M2eu1o9M-pgFBHEA9Q7WCpR!vXbO@n0GqFC60E7b2vJ
z)KgnuQr<54DN8ryNk6)McFmJoW#H0`f;Zmx*`hS0?<1^Ctl_Ty40MX(Z1Y;hCG5Sx
zGFF?g&~w<y&A(LD{#5;r7w5|OYjcNcwOy$*h6{gUw(k(k)`+{3;?Ec+8*t9X@4Ct-
z8!?`7gp=FTrkqWJpI-x*(2cIExo(CS4DeNwwaKi8uNLQM3xH+Y47E{Y82oS=z0X8i
zS|tu|EWhsBC$?uN@O8!+GMC41pyh`{nZ2BZIKvA%>g|5IGEF9eb=H+#i8{dRrl-z;
zq*_42ksLd=1u>yTtsu(FQJ#7C*Y>pfZ3P;u+(Y^x2ug-6Dth5|e?dq4n_Hn#L^plM
zq`Bw4dvp1;GqIflFswcg-R(#L1M)A;^S-4%kK7r?gB~a2@UM%JXSuR0T>n~9j1>im
z7Ly%!SD@4`*8E8SP9`#cBU@Kb;~vW{R>i8m+OzMnSM^iVPj8ps6~mG7Rm6}<!T*}=
zYohK25*rLZL%%uA>aOZK7HIYi3ba|>^M6y+RnO5a?x;i<KB~s~Ii{MIR5RY;-{)^Y
zw=YmBE8OHO(jnIr&uR$8K?%m8JEkH9oRrL`Tf6UtYEkty?wP)f#LM<pEv4!MqG|L4
z1&BjDME8iLIDI*Wn_XFfJJk>kSiYy(ysoP04c2!R>+jx7b8@30GN#FECB9AlX}mlE
zRmbwJHEyF-g~O`fJWyGKvFK2WEw9X9JUw-=T%tNQ{_f&8n|wus{To$@!ZvQ7yUe&2
zKF!;pdzaW}(2Jrq^8effvTcP9&bD#nT%Y8xhCkbLsvQ$dMr>bAoI&FH9T+o;zw~XU
zJYV9#TcX|bhG4)Y8{78YkExha0}XXUAX>N(<UOv<(zBz`m4k@!i$B0Ww7{?3*|VG}
z<Ehm5Yh1UO-@clps3S2HQeR}gW+Rvf?3zLsV<r}?%I-<L+b?}=5}EBDm}hw_$}j<@
zVk`Q*kT2QT*1>wi=&KK9{S)p6B0efrj79PUGrEJuuq|Nc-tfLn?#qBtHPn^iB=vd;
zYi1y$-LdJgKDDiF+Rjj8ZTJzE@}H~x+f{rqIOn)J$s<f<I1#TcjTW4!t{5Q(X9R6v
zMcHF#7ZKdA0wk`arRN1p&D$^XE>3f<hQ8|*4Cil3cW%Vr5F!XHSGprIp*gJioZ?6s
z3oQ+ss~YVH1$)!vNF=M7ZVd8Eb*UqNN7871(X8aCNe^Aj#R`5zugZjxAv_pC&Ihb5
zS2=ZEs?dzAkl}D6j3G|2vjFStD`Ay66J~KS0)huU<paQLETA)3bY+)K1FI|P=r&5&
zR#lYj(760SjFw?vY9gwLZ~Q8n12H$<pvOcfCr@a|Wn{!|E^e@g(7&sGkYsqad^ss8
zj3#mrdh{LGoBcGEi~P$Y;BVa$!?Y1(4(F^Zg%6>To4bT&_X0vBN(;@Bb5bt6fDsFw
zUKUs27^uWBe=*FonYpQiDn0bzF9_GgcB}YaSI5Q@4(mR+%BkDdEoB+|yS+ws3p_yb
zM_E7Kf2%EIoFMq#=4{I$-k9TF8V<_)Evu*%ScyI-CPrgI{6HSVsiOJT+;)ov2)*Ou
zw1NekZW*OJ@p7ios9L23o4*+3Rs_Tg-YyqWoDnP5MLu?WhxZhyhcgReT|_$2Kod=Q
z)Mojj9Zf}fRubEhpqn6eE-oya3TY|m%O=bHbfT&2<}LDWf4a}Fl)hP~<Oy(}d@ZFC
zU@|*Zt?&I$`!5HzxvKW+^?~oeg8=+h3j9*KrJWa_;`q_pozO9zY3Iv>f%Em+wM@$6
zM%9D0)+3n{ds>MV>2`~sXH^Az1g&($y5-xE)dFtGUJqoH4mO>O^I$x+1{=B+s}a?^
z@^l}V*jHRz@G06W1=Gyp1a@VPep}hmQSjE=3ItI{O!CR;xDw^lGd<9?2DQ&KP3J}V
z&I>%R=PT2pbWMYB$9zgX46u^xpb|9e95YtXb8#+Mk)Hduv+|AgoG+VGtH7C;xQwJI
z+lIN2YuSmtYl=H+N|ihZ@Bpr!ojp6)Bo@27S;*=M`U~@UNYaa8$MfOIM6gI2T{g$<
z<#NKDEW1a=@IoT4bC=eB>ve871Fg$q33My{EYrdmYcn81td+4-6-<rncO4N-Wwk)K
znm^DER+5u1uqJ!yh2)bec&pP19Dwrxcb<Duvq%q`J`3nvrH_1=O27C3NJL|j-}2qy
zdmF>R+;=K?IsPqGZq7u#@ElE3*$hc{XX16iHBx(9kv%MK5ZH|27uhiyWpQHxX5ctC
z#v(aU4soV*XWj=sCju}uIE}*zMGex|Uzdn(2FX<6bdmKV9Lp+SGM=C!U~OIk_W3x|
zA*?QkpG1geN*G&5lDJDEor)s<>{K=)M;9`QMSE)h1uOlTg_Y)JS8Pqv45gtFI*oy7
z1+eS_qWHgJYny=76@9YDp&7HzvIut`dUQ``VAfl-{t_3Wc(8=^Bu=GxG#f#vwrQF`
zL+!djsfJPbFDRk(c-%RC=nb?esE4+7aqA6U_9_*qAAkn;%d7a1tH_cU(DP;i4|DO~
zE`azOUG~MUt!$O^F_AA3R<IGO9Iv(FlP`SrCXCPVL~#NsvAS4HFYf`c5<>1{?*U&q
zP{^Hp_u1S`e70?H6twc8%DZ*s^U+{Dp<asrZT%*Sezf3LuvJsd%>GKY*L5M?ca3AG
z+f<8U|7?=~7AaWI1oYRPIVHVB7NYD5bA5l>{%|z@WQtx<yHXr+)24r{RMjRg_tuN#
zt5%kR33gvc9oDK<iz%nvlJd@9`W9D;JAky&vw)~tknb5}_kAApO{G#oKU*={jY>v*
zwEH-d*yDW2AA7<-z`;z7YN<JM^oC^LXOvr)?blD&>pTTFam*6K<2T?3RlF|>gGSo(
zgpdNz40?3K6yre!IMSV_z%_Sd8iv!=EfbUhB?{LMR+&TbfrJST1)P*Tiq?$7T543H
zXjI)cOM*-E37<fFWZhE*lgpp{PKezGR8K}+MJGJ^w1#|Oziw-x3gw;2<TAu#Zohs0
zv@X%$WeAy5{kU!2OBM_Gf63u~f5r8;VXicW3F|EM<E<opfz=9=&f4a>4A<KG3$Go@
zT+g%u2AGZx7Z;JI_o@drL%q%@u4Y8m)Q52~(Mh1O1K!xP*Q3!@tl``r>Ns@Wa}?RP
zPR|yGiZfkIrA6cwZhE{?t?pP4F%T8RB9((^_2Y3s8?#H$8O~7Vc^r_6v0vi`S-9UB
zNTW-YdZ|)7;?uceVKLX<tHNzX<Fr=%$qH1LyQ@-h#w-g&3An$<CKw2H6_Y2BmNLtm
zex8Kfd^D{+fhIR+<QY5LVQ(l&{{oqnqHcSwk%GOWvtBeS)g0WMUS}Lcut|_GTYl9{
z5uh(R+$$7{;_862`gGn70Zk~*WO(<Nj)uG3;4fz`d5v1tzj2f0K;dOPG5Z<|>bLRj
z4`cmZBo<?ZC;?vR!+*5H7o@%{<}b{k`W8@Ci{ZyGWpW=_vnd2r8<eeLU*W}jUTM<W
zC{Gd%RHKXETBgaTqdA)O0%qZt!^Zd+MnBd}{EXsDKlnt96BV>TXj~{?nFLLsooG<(
z1(1H;YZr>4u#S2t4etz1K#*7{0#@dA*N7`s6lAe-SROV3pX6f}sMp}eK~=a9`ss(?
z(CGv4pkbCGANj`?;JYuzPA#djVD)pG#Dgt)riNajPzwHDOUVVFW|ubkn?(scJ{+ZL
z-vK%wsx^Y!<B*@w+diuY@>OkA;)#FPGNM4$R<`qXd}vH*47xzNyY58ZO}~rat~_Df
z&O=-OpVF##{XVntwdr$}_w*x<E*TIKW_gz|6tRVO8Zg66{qILZf*)aB`EoWk&S&|f
z>j$phznTpo_yefJ%LB8VEud*la-coO52Lm;7Jaszw@)S8h+QOw)59U{Yh!i+5#&`F
zC$tMv+qjsg*(;!uS`{mbd?lUF4$DDWWU)i|5xuG@u@|WlW+N8%-)y5bZzz(2=PrNo
zVe$c3syMO~>Bcx(eAh<gXH)kP{d%U=D{|`wNYUt>mxkrxw(raPB=~u0eNC4*Y;K{@
z;-mj!IXokV%INrgVM8Mz)v(_R=M36x&g{C+ct>jRftIHf@tXaLDs<+=VqfUlEECU=
zqYOXc=H-pp6<p0-6a)3U2WD3}eRp}|S>t&Kyn9zQW-%VnJLn8YIFvd4X@R?-Yb@Bn
zT%_nw`Sv81+?+-4fK@Q<)%j2aRWqLk@4-zr$BJOB6aFqA2Yl8<S+IKkmNEwwa%|^^
zOTxxv#E0?GXT@nQ_u-%l1(x%j6%dE+C=~b-C%Gv{&lau&2#5eYGsRBS3p4&?iGJ%u
z&fQptPvJ*M8$`IE`hydoCkHC}o#vamhn_|#Hc_wJnJQoKvZQ<7>>QdhtCAKqmtYc+
z6*YbB=hsG$Sk9&|-+r~Q7jK*fOVK%f@Zu^g99-Ue{YDaj9!aKQ&bYnzamHTvF*yg6
z^Iy%6h>69J)}J_vp`<=8sO0w*h<`d~Kq2f?Er9ZTgJu=>!;ar?Pu-MjDE`Em^>V~Z
ztZMr;7u&7$9S7+QgUEA*weI`wz4nZm{U)6P&}oLjbE-AXeM{90gzRo*bfnb`CncHc
zYjbvuNpKpyHUh@3f(Og2gc{!(j=gy+N*4jdU~W{h055QvRmITlD?59H!6?9mpzwwM
zO|cebzlyS(7EQM5@&mE_4!0&2U_+Iqtmuq5xzzn4>u&QUjVt2oF^um+=+QZ7z7!m#
z9G!L1R-2!rk&qTlk*4!@>&K@YjdT>M=mToi{{9-HGl`*xl^|=h6As(*kNW(L%f{7B
zY0n-#p`4#S-AZ~vf}mjgZX}P|=u*K8?Qg9bBb+Z-=v`1&a7kk0Ab(Ngi!O~xzUJnE
zp1`3Gq?P*?ymJfFKn$FT<M{@Rr{?fMrxr(-4{*s}e{oQ_r98KVclZq2)qT!O?(L^}
zLqCWfNOd)VKWiPf)@IomI&LPaHLU-39_Q9(Dd<%%+r`ea4NAUp6@5!nl#}xep<JSS
z9!I4YZ8w+A<uc@F-8@xaxma6RD@<=mYT@CvXZF3El+7;9$r+@3+-JY-yWU^K9)^BK
zJcVYXSoaTDtDGwo!O#WY_q%|U4o1b}wlkUtQmpX*78U&YC)m;idYtCvg5cJtguH~E
z+(HW8yS4;~)Qx=*yDd4ZAdzZ{0vo2W0X96ge!^2bXR}(Hl8;m`wTp;w+v>fF(<&{S
ziOVCrHm4$~LC{%)+!@a84~|DrJc*NL!(;is1B!fX8O+Hx1gUUd3{5vrBe!c(t`@L{
zwWDTDDRwLDDmS)JcTKu}^P0BxCv3{NDH$a;@vZ`m8%-F0X<9f-Pqj46$;ON4y@{Tl
zSG{7Y%AN@YLACM}d$C6@p&D8qv4kZjcaGx4*pj{^>|EQka&6EIaJOA_EcbruBbauw
zUN|AI0~30m+<=?g4(Spis3rfN>Ox_;Yh{gOcY?=<YK{KGhIqDwyqMP}zOj2+_BR`D
z&y^};)v-<lPH&e7=h1hSimT)h_yw%IFJA~G#_)k<=DFLYql387Bp&nlHje$N`}30W
zslN49yE3J@|7-~ZWpzB98XZr0W?A&0|HAu!0^C#}mhxy*>+~9jX{MGooml~QW<7Rp
z>`?bK$|+4|X1YnrY2c2j_|MxuwNH(CgXA5ZhLV>_tKxWzy#m(P+B7#yZFd1hj^xyY
zSemEW6S~IZGTxxIJ%k}z3xXs4Cnc-6ksU2sSW_AN7Mfi6bk>`QVls`Dc%1?ZK6ehT
z8{~=O2DE?}>Fo-ZFoiz8wYZ3?XAOo532I8T&vCYd9b#-sXaTcb{i22;clOFxF}buS
z){s{yw2|xWfCxpJDbDDCN6m-8`?Gfv?m|wbndyeK=4=?-w-lGQ{-YIz>guKX^0J3P
zeJ;KeuTMvE6l8W6Q1K2lqB-Q3{L~p{#ErSz5I1ca#f;M)W`R%`bB*L9MI*&adGg~^
zpQa(U6dBNwkmb0#klL<s`vcyr1gm?lQ8vPw38r!mK7{D^xNSZm{#UP#b7p7;xy0%4
z2^b4ptUa`6HP`IYECps<3?H0^iDmv!^^rVnvc@pj3>=q6yCg<)sb!MXBssgag@djn
zDz^EGcK0xmrAP7}GKq10I3aB}oEv9&6+Ib$Bm&8O)^X%3l~<!?Djh26!v19hJ*?8u
zzMrWoST8Ikgyf(4$DaXF>dP=?#8P8@Rao+ykWO((@w#beTiQa#l;n>a!qtc;3Vcn>
z)Faj>>`E0M;fUfdO*0o%BSzZ1e5PXy=G`wCNN&D9Wn)pbK4&GHee<v?ze^xPvJuKX
z8y~pQu>UpKA@Qz6ZE&NsIFlS5Eu3O)gx-i4N85SGLUBJ0)++$pu`TNekP8y$lHz~l
zcm&}+cnf&&b&W}+K1R{TL#;#f>`0Xo=Gt;pv+kN3bFgB?{;d_n^;O^Y;ucF!t%yjy
zU0`%JLQnF_t{f&NW_>z_Fpf!1KOcbXS1f4SL%77r__&`eg!8nniKJUOzFV-GsKCgj
z-}Mll|9I(;wc9N$bqhEC!86It_*H>B&TSWD5Q?3DnmKHB$L86hjjCekr1BnWeO(Gm
z^r@_NC9})4mzX|-#@OGMD6PYdt~+y99Ze6-5(u^OaLh@~h@;J-RAVqbq#mqE-E+gg
zQDtCLgmo!3`;P7DSbf>Q)>@&JRP5CH|9Y;pYST{&;+XfYL(yi;SdEdIGZpX3XtUwO
zjivO6EOUUtZIoi8SXb*g2;)Dsg5=2c-4XT7uJs%=@9Bb#3(~|AX)a%=$}>tO&544b
zR;2vk?1Swd@-(1~+X9M`rqfr7VWG0@2Si^al9q%t&E1H(&jky;8Qwkni7F`=7O2qU
zcP%~r>9xzLtuZK>ClHo*+AJ&zg~M8#cQ9fp?-$+})(b0m%PesmumZ54i7k|Ois<>v
zLYn%RZDo`;Jo)SP1ng<aw)ZW)l)m)ZTdN$Ewv{q}|Lqp;jznqB)k7B(Rx{x6bVb*d
zbK80q%O~>)AM?L_s8n?G{vynv*^-GW9w*I3WEd|L&9aPZZDFA(`0xUUVl?}l5acJ9
z9W8$E@#(HE&Fg|0q<qPgFWW01NbRV`?+Ga|%*`vQy!%>y9=jl|<2=om&Apx`p8Sy~
zSc{dt;IQK4aPGCGg%2gSI3PH3y)Ol+nl<thXg6Jizyq$-2nDQO`Y`BYLPRaH>ES-P
zkArvUGs+VHWYDNGPbvU(@{tdRp*|)=BpGpocDb!SfF>#7#SKdbdA+fE>6Jlo#<OUd
z3p8q+PxJ4ah}Q?G{T59WPlEsMdR0aYjb%>-)Lh6eQbJ7LTd!WQ&CEgGAwOHtr(Q57
z#1zGsB^{^~AMdanYAPIB<m)Y<G)547SO4~vy;^r`*Fi(&o3v3?Dg4zJKF%$9kz9sr
z_DxjI=?2DhS7>EML_a#*tQPckDvJ6XUMZ-bUk+E0l9GG8QD1wBb+5hDB-l=>yj|t#
z8S$%WKGMD}tB_Pox$M&zPHmBL!yw>XR-f0sI5nH_bCETIzBDI&Zl;pn3?%<}&n6Nk
zlG@r+vqfIpwbq6byw)Z^Ps|-i_Ig@CrzsZGHN*l|w~+1qYHEK=(3whCkQ+>IUak<&
zyl*vKZY;)`(;`qxAg7O->4?@otjqJhM@)CPor#o;cNtK_W+_<*jimIwx2->z&r^;;
zC}I}!otp1y3uO0u`q)Nmla@R>hhL&n(la{*-x|rsC)+0QGXYyW=80Z7VE?jE=*A;*
zl?n8IaW_qwa7#{dd5xrG&C;$}tFW>5KWVe1o|BY<k-`})U_^?ATmM4xdN&ADTGpjw
zWm8m=Un*STiFcK)DvEP}oGQ1y9Y2O5rJg2g*&3Yv58^Q*9IW0$Srols+}rh_AfIde
zNPcb^8d$xF$jx~s+J^_g;PsPPKDTHpX^Ck=cCeC4XV)~^0nuGih5?=?ODv!jYW(~s
z1sc>9(Qn+OU%`acf)%-~&SE{EMPi4>G~mQ&>0HK$a8J}`uZ$g_g|L^=V6eUJaICID
zVQcs5<1kk4c=y~_DDNb;Oom?bt3Je(WMxXxZPF-96uDA-_Dbo5i>mC=hkBzaKNm%e
zOlcn9-%Jg$p8K@n?=P5rOM9@Zk~Q4fkSmN&FY>v1{!is0cA)}<{v_Tg$o#HP&A$I+
zUEToV>BDTf^%!dg2{}qjEmFLpDUQ#tywVh(nibOor3QL7Cah5V)?~N70}U|(i=))M
zl?PSJS~0Iw6$Wh6ZjYNQa>j7RIaF7Mi`?~j_)_Pf0%>^%z*P#w7hr+9g)w=SK0bl6
zV8mju7Ol!_Y+`coa$MRnjMm87nNb@Nco~Ji#pNDK%E0RbSP3tq+Z10_6-jN&>`j16
za%>y(M=~*lGvgip(kw3?e6|e{<H{jkA=Xd$>>lLbQKPM_nZ_SBDca1AQAvar{5l%}
z?U(xVx2m3`<&7UR9H8<hi~HGf9{sHA8MR0u_wn3XR>9kfXP48Si5p8xsBK2mD+bH^
z4Hfx6*(`57sl4W>7jMt2T9esVLz6)V^Vz*hNz;FSCXa3*UFkEn{6pbaN7#;i6KFk5
zA&zMYBwU;^!{&Y`H#)40EauGPk26NLoOTMYUOrmGPNs6x>3c4XF*OPIE~gZu*jX!B
zzhJHthQn)GQsqtb#ZG97w<ekBUtG!nR?9@698~btU8A}yDxg}-Vc>$R`#B)nTTg_y
zjiUx6zlho@2e{u8^GW7SS9@<;sse#2<PJ>+K<0|9*Ld`e#i+fPtp_#k20PCE%+5i0
z<<LCkALAC^+mCwE{340>eyj!7!?&<j&HYe=YUjdH5&<SquH3lRrEr^3I^<E;SilFR
zv+2%72{pKIsucUJqi_{fg_MM~V2&-;u2K2LnW`Z5X~ww-Oz(}mQM(yBKyP(hFPmFu
z*p_0oM~EQ1+nBMIQ)%o!9CZJK23#E!9cs!-JB+fqSmn8WvpN?KF;^HU&e0pOSvT#8
zi>ydYH`3DX-4(ANrerv%H82^($<@c2LIjqpbU7|@rj)HMOp+O#q7*VW$__A_+7#>O
z%LvpdFALBVN37HK0}`HPOJiV~&(opX7=TEjVRi5B1AvEqE>7J7V%{eG`u5#(qF_jN
zCUmAaVKo;fK+MqYgzO;P3}8{|r`?iQV^h*xIxZ@5S3)^;^=n18kt?%F1z=hMJvrr(
zge~m6ft;QaF2Tv{_9HtNX(jof$ARdVHDyhO`c;<Jo`25Newo*{=X|;BKXY+7+WI4|
z4S$!vtc$RB<mb*{5ZqD4nzD|^JhU8AELd9i9=s)acNC`CWGpz;K~*D;5D>z1E&89g
zLjJV*jJ;<c1LXA%31@PRuST$dScqSNm*Q~zdeas$mlD2Ewp^-Xeok|nS+RA6HFUXd
z{AZN<hYD>ow^q(dKGFmll&l+P5o$uNEP^vh?-gMIh;s6lnKPVJI=opiZ~Mvk3<Y?$
zsh<<evu+qDr6;epzB<+XLU3(zraA52^~UeOt=#UAo~vq($a}nl^`67Zbs{{U?2OCL
z=2qyK(^{Rtu+6T9!Q((Yw5-J7>woWSQS#$eNy>g;GnY&;pH(V_|3uVX!Q8rh<v8&X
zY4us8&=|7XVg0CvIAY2uj4JXAuglNt+Otg*7q*`r(XpxZ&cH~PanI7aZXIzM8f{;t
zl-v->#%E%2%Nn!;&8||0yr#YjPjWa`n=>1QF8cQS#A=Dt&)6icKuTi|>ZE;LhYhh=
zQVKK_tNHDD{Qeb<6@1AmHT!u`-h|k!vs6vwr4e66!2;x$T+<%Mj-6mUXaAY1T0BT!
zLD+MGSB0bBDm`kp6hS$G#%Rpa1V(4wXu!nesg7eP#{fjZ8m{>U2I@?7lj>PhsW#VC
zIo&3S*A;ei?pddL+MI?2dOu$G9M)eeSoh2mm#6(2Y?0*YS)8mxpio&YPW^_M4>0Gc
znUW0Ml^l6(E!-)mWG11P=!bY`T{O$AWN(wE>GU@J<8-fPkXYjO<<tGD`@VH?C_@E4
zenJ{u#T}lloF_1rf&j}%#cTkQX(fD@V^>@i4epCdrvIX4a}B5bxsI04K8+#D(nR}j
z#upX~(uBhC@cUyJ&a!}1{p<A@R-=_Kn>cZ<ti>5xi!s8}e5{Teq9S_?+HPPyW6mBJ
z80DQ{lOy?N=@Ms|0`0ikY?-j9DLZuCc~)5`C7oEM7qIoNf)W-*F6XDL=#;#fMnF96
zOP<bf<th>d&x9tNiz4R0?u^GW?pI2LsNx}5;w@b*>SQctTNH%<O!OixZ~B$Nc|*ji
zmJfO|Yc`XktrKKZ5cgeUli-UK+dh=!hQeu+pDzxO`|4;vWC$IMGHKPimY^6te6>Jq
zTi;lU)Aqy}ub<8vp2?0nTUUC?d;MgeXiSZtvA#p%-vrOJ>3zX1O+A*GPAM~6M=c#>
z<A%hJ(pJCDfKWNr47*^5=`J~*vFU?4-VMU_4K0ItC%3s+_(_?qJw)vF^&&0yj5FDe
z*5dcCL}y*o*^-KVnkN`*^~pl6@!R`%SF@3YW#QUSa}$m8*t@@h)1y86W#wJ_{LwfI
zN>v4>VLKSN4e67SYaN198LBKj%I=}P6&YH^P~kF9=z9~q!dt^mJc@-mRc$)E^GHz=
zZ_HWnMvaEp_-G%UV_8|T^nVz8&!{HXc3m{|-V>$wmH<i#9W?aNAqk<YR7pStf*>dw
z5D-EORS1ZbP(o8Wh|CE{m(ZIu6%YgjrV3)6-+AY^_8Mcav)0-B{9%m1;7^kGey;m!
zj|wdz^cVL+Ol7bwz2-<47H1JSZkZ=|gCXxPmGZgW4aQQczi7_PYh1KsT-nsd6&|mq
zq~<X-fW77~K}VyJEbBHr=NN+i+gBoe)}IsPr)_%dK1Kl!;9LmSJ?=m!G~33V;8N-s
z%Ws`eEvS2?$N)_5sjf%=0zGxl9m_3QkT&fLd$F+U5-3j$KkyjL-ZYJH1YrZqm5OC1
zZ|`|tHPM3hWyxpS+lTxliq!g3I>_g*sZ_RgvR6W;oFB2wKj)9d#D}{V7;!p5E8IRh
zVQaIgXkoMf7sekY!d21LFH&nKjS_w-T<h<|s?69>kJ(Po-Jl4;UU3+otEGo5PwLm(
z<&==sFXeMp^LQ}IvKx`?<`}479$4L-X53~#4}KRvr4G9j53f1mUr-@Dlme8PtiNyd
zEcG?ia<3$?JVPw$IiSOf3$&d!&^e<y+SCtMT0$FJzbI{CQv+1}lEvrD16Jcb<aAV@
z$R295OMk(p`o(L*uh*D;n-iG(pi}e8)h9DTrP^Yp7nUO^I!N)U@kx$@nq?#X2C1?0
zGwYg>zEC(z;i)zH+ZXJ4&ykIt@<9$)PqTT<Sc7ToCv>3f0A5}96X#>4jA54~2A_9%
zC$f*=|0R6Q{BKSx&xyer!QtQ$!sd+nkE7^+c=-SG50EDj_XsY6<=qZewW+m%TRb}1
zs!$EOtZ*VG0GBiBt;s%&XG@cdr=w(w9(8=zS9#!mo*kV78-sPBXY0CP^W01UF$Ycs
z+n*4l);kFdRK};^-@DyA@eDW399=&++JM@+Zuf(SMlahSY(UkLQZ=kZIX}B~#K>0I
z2*D|CZqJUR^28YGD2`&({kevIG#j5o4l~9>t;C2wrr{Q1KJ}KbUm{)|580{-2uvEo
z?nDj4+=FHcg}5W0`f6I?k6%Eezh)Sf>%$^0Jp{gRHe!i?3q_Eo{}Jf_E49x38-Me{
z#4s<{r*Ezbbv^8Z*7K)L6#O(iK3k99qaajor+aKiQ=bVWx^pLA=30uRqs^1Rm|dnS
zS&ny3y|J^epqF7Jg!!OlP4_VnSb9ldY+Rk5W|o2p_0$^gb124mg07+Yi3OWADR<`w
z>7HJnP?J>dodkE&w<RvR+tbVr>Lyb8R$a6`J2rl?Y-<V#CwE{kTZFtR{be@&Gl+^}
zI<(wc<Ft?+YWv1*5I*GL!p8mN4(+TqfTw_wwmVe9_u_Jo>vZGq^7Inq0~lTed81PN
zhOujVsGD87i}iBfHs)MrJ+kX_6UguLpFAIfD;nQ5PdVekE+QEOX970SAP?my&SQ`Z
zsvvU!bH9QpM$tt-<K$u9Oarx{7^>=YVhdi8WTNXn+mgmwZ&y%TUGx>`vg1<<*ce!L
zaX^fe1U_%}w9Z`QOmIQ*eb&<|l;}lV6lw#>m~h@Wxbk_ctr$O;XVBG7=BQgsS1p8)
zYu#CzP$wAX-7_nu*(7xVX>`5H{h3H;zN(eAkEDhtnOnIj<|tSqjfoT{TgJj#P+Oqq
z<c2nHPCpeB^_9b?&kAp}i*X_}l1yMb@%igoD-{|)<`}a00il|okPTq&#)o<k)8Nn-
zF4c-8%>hD*OSEZCX}hm5hG@K}3_3t5iSLllZHA<uYDeg->3pRH$leVg9J;&$M_93z
z=iOYj-d(M~UUF@oFzUwyVad;!$%BMekjYjMq+Pv{`_D?YcGd6{wH0?uV`I99Oi|0F
z&(_Dz3;<Fx=S5Gu{Jy{x9#NlZxG?0sY<Kk0hy6i1XU`I8CHxsV5hdWHp)-4fCVYds
ziey$nV#z$Ppfw9=FmkJW2>A%5{4PU<!2qvy(Mmf&JJprO`UnP`-LL<a!TK%whQ2_(
zzlNVd{-^I5(7nNVAQVvGqb^vYe9eGF{$K0Rzjvbl@khgBhk*FZqth_gDA+}`ptl9#
zGr>+Ys}{Vj<>2dfyhV60<mQU~5DZme8LG(Uh9){82cU^)0hU3}r7GX`{$yF22Hf8p
z;r#5fO^y7GjY9ri9%sGXXdgRvlS4|F@E(G9L_~x-B*T9Dhf`x_HBr^mHUU%GZunP(
zfha<sH|};-9@A_*S#SJkccTJWq+LQys=_DZaKr_p-!PP>7LvUrj0cw8e~fq~;WrZ7
z<dyz&TJsY4FTHwfh%R4j0ZQuPkGWi89CW%vPpQVnt_|lw|AQ4hH~ai5>&AE2H`mGW
zT{qa4x@ej2jh)wKJjCDqw?XH#PvB_cyv5<Q-8W{aVXV@uVmjP5N$~Do8<4&Y$pGJw
z6H4bkYtn&+Ua?@;0JfrHSJm4eq3j$-vFPxuPgI#|b+_|MLAwV|jv9n!y{k5$U;)jL
zk;xZ~h6_%YLiJ<|QyI5NIhT8X(NY%Mq^)vMLfIy&H4%jni>{lotd=j8o_hGFWhN8~
zJY|#2qXI`H$Vd98sQpG84;RtOX!aUYKm1%0D5^5h3Zft*62>4NSd*|747w9d3lI$(
z72Db8^YE`-?Br5)<HQTk-~(84_T|!lzr$`_dyA}LzBOc~Cc*mT(zlm{5`y==C*ixR
zP-83Bj_MPwy0&gSsg<E|F&Bc60l?k^`lWxKd+sUtJ9}av#goJ@aI}-K5XHb$vRTFH
z=;4Tvnk}TecfeC#@1KtyYYnPsYuV~W7vOXKF*%O0jq;@@Giv$k`!BMVAjo7)ftQO&
z$rns2OlRld*M6{sfoJys_f|K=IZLaB^<z^sCG;`m@pEFOUY%&#_s{(LPVA6X=M^1G
z^+i{G&mmC$Zeze3+jgDpMfQcF)K`nu6pp+*DF4S6Ig#{WnZ71mDE{G56%Ue)wQAk2
z;6l)WU3N8@<SRT-XaQ0u;lRo2jh354m1lIzS9a{nn2_!=c1g`xjHc(d_%w@H@s-}L
zhOui#yKH+N77r}^Bn?PDUfdM_jMcp%<<T-O{#(Ph>z_$&__f<Y&*h3|dC($?MV`gZ
zr_W3uvHj@Kmv$-cbA<3kEDoc0^pmaYLFzA|-yQKmmidj3p0sB%^*S#iaoj<dEcQsX
zM@r44b)?+CK#g`=#wr+uU-shjKe<<KdT&Q8%sg~05{xp)`NNH|M9laAUyIbT4CVrZ
zrg0ic5NTmxEHK@Gmz6O??V-Scu<tKpR!CBgD!U$I5s3JQ5(5+}q79S?JoW#!oY}aP
zvCaJdIMU-cgNiEFDz&SWj(NnRz5<S2kIuH3)dH9bu1Dt4Bm053`c>!7)PE0;Hbot|
zt40f56s+08p0mIwr~38NSraU??QWHsUnfoVJ`8*7M11)o3-=4(kec__C1t8eov=~1
z4^2s3DuuF%-zsCvv|{^Ay(0Vo>6hp;$)I1OkTm$hvbeXFxykV3BlV0Sj*%LCsjj=+
zNH|hedc4zv+;K0Dfh!F85ziPi1k#hNpz1*>*GwH2H><72w~TS1k-Y~<11IKBB~4bd
z^?%7rpTM~THAZ4%^{oM~N5$ZXybe$=l0?)q!sMkbMkpH+To09kgTDosS}ux{l0A$a
zW^Z+OWV(z@cGK$3kJ3e3IiV^41q)P`I>OG~H?dcFrvK<}>eczpmW%#wG7)#um;BpO
z8+b2W=aJ8SHC?LZl5<m|GxFrB&8GjO#{T<7T}%`#xs@=DVi+|!K=Ee}d!?7ovsbAm
zb!&S)u5%<Hxb0uB-NB@P>YS`R`)7+$*885dSe-`ky*P3K%=s*<i;2V`^W4a{);jZq
zJ0<Suc6rfNBQ&FISPa!Sd<x7UM4)SAFqvgP$CR5nd5HEsZ4?+a>6dzJ3l(e$^~y{r
zyC#s#K@*9Zd@Iwhggfx~P+_j##VqyQ-WJN!i55}+BfNxw8n|1e^Yx_Rb<oNjb-AyU
zp()q4OT)hPvPxtl)YzKxT(|PZz!OicLO)5Ig*ua+5?ET~Pu6O5wgZyQ*@-FKp_<8~
zLXpNGprkF!xB&EEtG_s_5pwTf;=<US<_2b)UsS*rc+dfGkk_(P_oDP<TE{iVrcpj`
zEF}KT{5<zBNnC{u>Q6ML`o2v$3srF4PJqE^DXtQS3uY{k(ov~S4=G9Czm)XY;oXhK
zqSl1lkg10L=^v!JEb|14`qO!uua!ts?C;d$p}@2NWbw{UXXFrQw@OhyvcQiM_Dy?A
z&UuyAH_QJPH$UdP^9VlXJL_ARs4}EZ66TgI1k_IL=NR4~1Y(=PIfi%!SVQtoWh?bQ
z5)d36l$o<Ke<6#WssQhS^VuX^=5QpDDFG(PU_`(o7O&di%L9e!8iI9sMx<7u2;LZ)
z-vy>*@alq;%7g4u3*;~^>;3wdwZ-)TR{PhqfQpi@qX%6AJ=L)F$Hn}0-{e-vucA9S
z0(ghylt{bB0k2r6AmKvFNI=_A#~Tso`1BFjcHM)L()<sXBsxDI=4GdI9=ufZ)G8{E
zxQ9>CfV`L{znu?l5VtuGFPn#}Pl3B>&tAd<sZ4NyY3GF+&N65q5hrE9oOq3x+$W93
z2T_3y9!L=g{^t^(QwX4bP~88~MerZE-{ls&c}D~Lo{PkArm4NY`NW;TD$M64rYaoN
zt4C0E5Iwlr62#W<YWnS5$KYB#DaY+El7W*2qo{=jjX?+_2(p6rjr^G?)WtcGA0-}D
zXWo)DPTRvg+!)f}9M>QvrVEB67{l<E(w}3eHIMRYLtSqREA>Wdy;hZX#J|-udg_Nd
zgOjFX%zM)48BybxK9`;*-N(J}Tu~cl`^0IMGhOYeSMI1-J8G?mii2?&@*B)CFwizf
z){T`Jdq-t{ggeIfI}ZadTqS<jX^oiBrKN=ahCurJcTdWEp=<FyXVv<koowWjs1NhA
zmDjd01*txS9{oL!+Mi^hH&wQ35^s_(+AfDEDe$U<dEJ>yn;s8#(v?Qrul!i>%Fc-T
zqh08VU!@cCb=2xcgXexY(zF}&V@}{MzonJz|5_?kF1r?on5xGZEzG%+<ybyFyBM;)
zq~GAc#lQAoI(Et4Du^PN+gmuMqt*A8VI^nz-j|y`ULlr+H)w65L1>_O27!#n-1<Qw
z;dD#&ocqemx%G~nG}e~)l^KM$GUw@QKNtYUuE0SoJ(}jnf@EWZGx&(@3fIRcjF9}v
zwu%TjO2e-HW_-4kiF5XJW*aE+NSu`ZX^|-Eb~a@C=1wmYzYdFC%XKg~b@(+apGPtR
zvjtIYGuZHX-1r9gKsy-`)L}$}U+C?|$2!|cQw@=`T0hI;@(77Acbxpocg^mt=H4B5
zl_h$lY}V)AZ-BRtq#7|&FZldwTFKVJvW>2Go%(9N=Pzr+C~?esl%$oEGFdgcd-^9A
zW8%Fa)>h$tD1&$gDNVZq<n~Gm|EvxiZsS<SZm-_%|Jqu>aGp<5MUlQqYI%iuUqGF%
zm(Raq4Vx+U=Y3r_YfbXc-*AXsh*8`k`9fbQ^;&^1hO1+kx94n*`l42N$pteH!@r1b
zjJkYnI+MgyWCk7`h5w=Ur|JRt2C;^+yW|48){S<<ervelXpVZZz@#A=_|850yet|;
z@-@8_{^W{E<(wSz>7I=zM`)zZtN4{h&VEW}6=+sSoW+P2nlpk<(sX-gOY+aar)MKZ
z3+F_LWANmFH~4AKoXQ=aJ-0^&ADfT}7L@r~?Jbp5gM!O8kfW*)me?r9HJv3Zu@cgB
zvYhB=>q{qvGXbvf%{-D!wsBq3yh;8uC^%g#k28GFWr;0&eldmqfO|_~|L}2b;4aBm
z;!X(L1RG=hQ*L7@qgvK;_LHi@^NCp@^Lgl`U)?j!-(2kl1p7;(XaVN6<Q}i41wEzN
zj`hmc&R20?kW>OCt!*}AaHr1ei!jsYMp;{EflA*fm7Cp&u+e6IzODA0XZz^hDC9$j
z!1t(e<N6-nM%Ia%uGP|ZPou>{4o)5;?EvMxB)YP8u7ym5^19PBYY^Q&S)}sTIsR@c
zYd+C?zGx>uS%gXlMT4>O|3UYvffT@?e;5Zlz}Ch;+o&!N#{A!N6LL-4usi8aY=#8M
zC2bpz&hB#ezt6zKlLoVlJ(HhG?X7vv=m|ItEeDQQF6rBt6x<HQLR!|x$+oYIQ?vG(
zNyDqc+eR+3?zsdZ-p4H0^JTq$x(M31MjOL4wT6`$q346Dp{Hg3e~sQAHW6R4s+<hW
zH_s?th8KkgzK9nL193gEsl3HLk#<10uF$Vf`n=;Wk)bQTP}uBgh;QAcFl!lwFE)$}
zJV0<n^i<Ma6Ulv1Wn7QZnF}2Hzj&ktZinoG9~pt}IBAMSjojl=tr4F_DRdgf|8U}?
zlL|Z}#O{9k+Gyy`C5=^r#m={6&wsmj9~P$dc~t#gZ+#^W`2lC;;}~(i<j1_4K9J-t
zmceD6WvEZY;S()CKiDY8i8x^9kUVrQwCd(c;K7-7Q$n&0Clm-vGgT2*N>*#+OC6b(
zvGdqUOaH8PYhw+s(5L!1dEc1BLs#z)*~o#vW$nE>)9Vsk3!0qNY8A3NV_mn1%8ZJH
z63$RUd^{p$_wEHYv@<Np>kQ73-Ki}W8&xes{f$TNX#C!|*T#T{E-VLx#nnxRS(@yb
z$lHB*5fM}Ce>Xjhg-LGg5P2K0I#V;}hx@?{0=q^S2WYyFUb!b*W**6+)=Mb;3tH1(
z(DF9*m4!ZkeFE%|`g1AxFTC5Zw^?=xCMs8Wn<14sQd{*pI^c4Vu#yOE!)Uu!&uFtX
zFeQf}J+NK1dikf7@HWa}qOxW#kek&&bbbo5^xR5*p@{k6PL-MK4R%RKh==7)inJWt
zPJMGEc!xcP3L%dKCWI-ATrtaaAZf;g=HGbDUQC2m)hiB@hw(`pts#rb3Dz1e3I%2B
zDiBG#)4I>rrluivpE0W68@Y>^b6oVPyDRY!YOcioi=5bLX6MTBZ%#2yJ&0Lmm_pv&
zk!;G|%i-hV=Eqh6epD~zzPhvAl(J}J5IwSrmb$Y-e`)k_B1U_1&AQ;gD`UMQa%BY)
zEvg#<74+;35qRRUA4xXq1*F5}+N03^X_x$+XMu@C^uVlqm`4;e=Os9@TAIfs)>rO&
zEyV_u5S*9Fr3GRk#1?dXE<p00Uo3H=P9zDgjk+-S+6Q|E#S?>*N1j!i!u{IXi9Sx@
z8>xz^2Fg@-j^|hAo9Vp|dztKdL`$I`^{6Q6@cufjM%nDNJ5pi#4PVZhu_^w4HEpU~
z*^tGs&*v_qWyov)0@d9z4%UzN5?YL}nAUe3cp~X0Fee8lDI;}WNnTD|vZ^i!>h~zX
z*!!C&-V6zRc1wjOd6#Un+lkL}cCL24SHj?|?~Yg@c*weQnXX#*jWod|ie0%LuOFD3
zw7%=&T6<))bYFPz8lTMxR0VCRxmIuNVjLVUT}3yu0O6zd$DKuex}2YvS3J0N;p04<
zp$8mVb?tl@b={`#!oDsFz(K%kLyAY=f6^iUcIE+eCx<m65ngZSUs(nD-^JqpubyMS
ztI8X1JSMX78JwteV*vJ`7+*|~E=7S{JOe$88_dRO-zkNqyC&?MWt`8dTUpg)cH+51
z^F5pq5UO1|soa3y9wnqzou%K<jY-uY0kr#hwkj*xe5G_wW{Y%+De(l%?HhL-yEI$%
zI^xAAGnnCHg<ZtfY(<1Ze)#W|<Qqbz3ze82X1?A<+P^^jb-+6zT0ho5T5T=nf8U^$
zV|*9Reowbj!CBv$U8xszHKV1=DBR_ZOoj}ptKyyyoe64{{iM-aV|*vR!c%ddgU*d4
z&#Lie{1stkGu?ho)9PM3cUz=~e_5IIuf9*7Q-u7<SQ0YA6*fl5Os;Q#(;JjsEWC8(
zB>8xwZqV!Iiq=Mam?0$!GA3I{C4C6PY`fm7Sxf7I=iQQHJ23MQinE1dz+fll^R4V^
zMPow|Ph}`kQNL6wNVD%{3#xP<g>L+EJx}{{qVi8P+WWva+Xg&ADvKCyb7c+Wye-a+
z1=-3d$Dcpe-E$tj7s5NAIGQ?TjN@WhEF}GUC3AVe05!H}DkP)74^uSyv{-8vV}y=b
zAF*L}>m1eB_8&*AR1;+bmB>LnKtL`IXQe^{`2a0Xm7#iNtUAci+SLA?`1G4qItfW@
zCU3vEHij}kQF$w5otN~SSa-DNQpw`RkGHuvyy1JphmP$c(nf_Lau@bG1H`T5xW8U%
zpIf!bE%VA)5VRP?@qiQkNX<;-w@>BT0ZN7UQJ5uLZzTS;lf^BQ4dvasb{X=oc~Wp_
zvfNp?(`CgHbM?6WPJ!=1r7z8?pR6gPel$_?a6^o~wv05V82`ZhFn+AnJmm_o7?qwD
z2Y#|}52$PUerysDBR_<Vu18MV#Xli6eRkn@&2fO0WD<}T*K-#uEpYOjWw{6mY$}y`
zaPFdUQe;|^Fl5hCJlk2HdUPaKFXs6~FHtJ0?gQ|DMn5s+7wLDfst<RJdBPrZr`33b
z?E)Ip9ud@3Z+heF+g&|c!OP5D6Fw8HlRwsOI^U|%2FiXVub5btUG*DUMN6@tAHEA{
zD80^q%*(c=w{{vQF^0T<+U|Z+JOO!y$f?@)x(Vsad9nA(pjKIXOY#B;p>=?=zMuyt
z)));2){e=@Z*Rtv+xrAV(sl@|5-NW79Rc2-{hJTWV2Wi<=L!d?Z6-=KRhP|HvV%r@
zKM>qQF5#(WYOl%u{X;ia*LsUNe1^kG)QRn;2EPq&P}GA0zo~X}guJA(sG`HU#Jf^e
zGWnsCi&MR!K})t0MN$3IC$xa_xi>E|#q=c;iVafb#=m4+%ceID>-hTs^#IN-IfUu6
zMgdjw#(PoeCmE>dYlAujZ0hx^?>z6X)e9_94q`b^HXGS1OFi>;^|if7^Za?qxn`pF
zVs)aT*F+!Px-JiY`R@s&u4cK91@Tyh+q^vg&+Oqj38g)gIE9SM1pVXRKSu&V|E-$>
zE0=`^I}S^fmw@6~|GNp`|3yfelyMt&au4p*_JBC#eT&!TcTUxuhhD1v!C5=(NN{Tn
zj*gL=VEO=?)rwHI4ktt|bAPnI`7L5`Q%Pa0UwlR_HM3#A(u*@M+>O#qu%3<Q@pI~;
zGs*_hu?6B^>^O5)Dx-E<Zr+Hp?mw{C^iW|RGBdg@0ZYWl@r>ZTuE!%FtM79co7}Ba
z$-C*(D1*K37VIG@c~xQlCq}IDy!VYxjjirFHnAj>M#0uqX$-INk2&eYlC4Htrfq}F
zdG^P8Aq(z(f<P_|RC;cliQ+zD!zwM3F8;~3Q$rn|>?F)%tMubJYpQAr&mO9K@Q=1M
z^Azyk+P-pal7O-A_00RVAxDp;7`_4)Gc~Z{3kDKXNjRl$$^1mS%`Q;ukmtbq>w{dn
zJNcdd<w58Gq0}Da-dU`a>-{h(^vfUX#ZP536XJMtNE2lhMD^3XG)?nMAHGwX<^0-u
zZuvc_ib7*kgZ*@|A;a~@xn9X-KWp?)oA6T`%8{Js@7560hTPkv6psk$iwL7GRrlu%
zF0D(aWp<Ww%%u3$=y7HFe6akzQ}M^@u*6Z%_>p(c*FkggIJRmW;t(MuEmx(?W<&w8
zpHRoBjLB21rR;LDg*-ma&wyGTP=;g3JORq~-APkp3cTh=;RM~Tso`eV8<NegVornK
z0Ur9UjvzO(W=%Ow*&o%ye<v23&=xJ=r{^;aj2>!6#r>*a+)`E4_7~K+l`VHFfqzmr
z3Lej`dBcclwk0rSVeWlDQ`N0=5jTnACry+KpEMc-C2?g-rpr5itwWZ(lpxmVdJReC
z(w7Q!(4v_nbeSfc)a=}fvCdTfw_2-n1B}h_a&cEUqv=>&kB#4;akR8$IcbG2Hi~F+
zNY*3&J{g0Jey?j$zwE(k*8Z3%aZokgR^sc9?R;rQalIpFkU6@yU<zUbQ<rd)@D2v0
zc7pEbB9=kw>30k0UDd^$`OQ-;q9d)ZA$gDT&axp;%kD$CT@Hp&ifo+XS^pmW<a*%u
zt2K+c%eeih=E!=eTdO6K7iw1sQ0z9iSPr3EBGuB`LLD=?cDX_>3yaI(G#;hTO#vI5
zKelq*v-a$T2v=TOyFL4Im%ir-P4Z$GG9-%6YeLH`to{7lo+m98>?luq{$YqVyBHKj
zx*z+ic~|jyP`+t8r(JE9F-#w3Yk?zRc8@6*MXqN0Kd_$H?`tEI;jMD-tyxo_)~vQ^
z3^^+VxxFZo&-K6?h%(zy^DnK}X)fpS4wFZ3S|Vf{_8NZ{INgR9A(i|Dge73~jOqL2
zQ~0?~iRc@tx|}r^usRYvC~$UZlcNcCQ}W*QOLF)Vk6HQ3-s$qshfC?#u=ORJrydJ(
zrFeS2{Pw&(203nq7#6apD-!KycUps}7jRm1i)1LgnC9lYlQG}G&_e;B59=n5u#szK
z%t?9&R5No}fYbn69H0cDfeHx#040_OF}m#8I;jI`h5sQg`M>?npWdMem#HlMv(vqw
z<uPAf307e^Maj$fuwOEX_Qh>+NhOPSrEb$Q`_H#-L%#Nvxlyl#2N4TUuZLoyi<G<0
z(hNXLRpSkQggQ)qJl)pTL6OQ#m%%f`DL6R0ALgbVyMz|zV%fR~(92S;1r-wHkM)Dw
z#OM9U%p23(@vgblE*XZ4XZpQ0*Jyw9CGDgvXn!$c#d24Fn8g;auJr*X@?5|7p52C}
z)<*r8Y)pDUmz~a2kB4I2C)Fn2f4g4F^L%oWF3A99?@;H3sZ=pDZ4lY@#0Y5n2kbt8
zhqiKCnwJOm1EpN4>1%2s$Ez$VsI!8P7%Z9oXCGS#_k6!C;PlF(p5qq{^JVe5{_1am
z$BxeFKQNjz34`^5>sB5Q=v%BEsp*MtC`><Hia5o006bmBrTzyW8)akqqEVxzrXzT9
zvtXp0iOkms$u>dm;&V-z^5%Eng#BGKgeJdbIPk7F59-j+)=}O53K{qQwZtTR;B`x4
zTXX)cycp}fkiZ5HMql~pJtxLV#sQ6mtUZ@yej+FHXw{86wnhy$efXgLSu$Q%czaGE
zf4yb>QrpWZ_?URJ<R9%Ygh?6O$7P%ghY=(2fp^LihAiX_aSUF+8i({Gwd^?5**DAY
zS24+i>R0te_8%Z-m@bG>a0e7t#c<QV*7v|^P95y>XY~UXFs9EBNp@5$1=J8FP6`RA
z42uPXk5<L8^Ljv|0-7Iu=_L8a-Xmsps}HdR&$8oz=FyMAP*K_2`4!hJB4b1T1-dYM
zTehR@hk#3vXVmvv$kOySxP&wH2_+n%6-~!h7@;ZC92`|>%4kp_H>YI?)_%F2CW6nP
zw1mtmCvDVQs|Ef;`&d(&4()l!J@dTZ>&$grqt0sOLtKhlT06C`Se=zwxJx~7`Qy_T
ztt{|sTGhE*1p%W?br2UP@HNg8npw`{%$S~0GQ||10()cK@;a-XaF$|Gck!%STl<nw
zEOM*8nq>yG=N@1!%tYEU^FzvnlQKqjO=Ma7>vcYZ^X1VAyaa?~YV{gdF(%h2n{G?E
z&r|z*>)H>j4*aHV-oHTaWZIOzjl6RZJ<PrT%FcP*3V3r4J<(@@!qtzl^Y>o`{MuXg
z8qD6uPJ}65oopP@96J-10y@FQMzmQ5LQyQBh$T_G7H`defjIQnU9Yt#j~mHmtkVs1
zvSHo#EF495dqHCZ+=1nfD^hJo`!Uh*K?0?qNq!>g1@E!jemQk?yR@3d&|3ZbsiAME
z0b%9&9sjTgG257yO*c8lirgyeAeziDK5ry_R+avwWRXzIR$iUzFX*RK<@$)!hIn=L
zv}E!5R?rASSwQ|apk5;H-)1;Q`g#MUalJkBD<*w68)XZ&R}y~x!baX`e#vOUmX5aQ
zrGE(UtiSQ{n3u=-F<Ac3BT7(&hu4UgGLVy|Rb4o<dyuhEp-6d=V|b`BbHhqDt@y+I
zTb<ygj?Ta<<LF{*k<a@!2Apme3kaDf`AF;6qK+JTR)w95h4@L63_H3aBPtNs@ZuDK
z4NoKb9=xY6M4c{Db&Y%Gc|Oqt$OK?TKD4=lfb<3+cB=t>H#Q(Z{L1@h*6}+j|9i^)
z->*Y)xkw#I1VbRSM;8HnXIUws*=}DhPKJ6#b)}u-ai|2Vx2GDQgiK@ubKx8*_HYV^
z@NnU9A@`6KEEJ6pJ`eNjTAhR{nle7ClP-@cl$QLpaDFRupin?OhZjEx3I+2>2#ZCw
znJrKnhJW5j)#hFoPdX`$`<3{ap|3_pKK;jn99y^A=y>=P<cbWSx930s`oHT&6A$CX
z3xp>ggbVL4FA`-?F3lT_ygo34X_iRC=V&^U%9<jt1%i;N4COXUKB?!ddUQ*6y7<{L
zBu3MTfO=*A>94n?ERh9+w*GPl!z%jc(rJ})5v6Y77zts2h7xSECPn5IFZ;;i;}X&7
zu}rr4IiNe9QOz_Ko0RskCnf6`ZF)ztGGw)}@}>(!E$~cq8P<F7G4VHaB)lYVD{|FJ
z9;0g&G*xx0H^WY4cp}95tSPK|?~|A2n)y=UipKl}*kPRjr}C6dcD6U5bP8lHFo=Xq
zb2o-XeA_H(U`#Z|<=8RESlN)Qn9058Nfqwy@%_ehWExRZzuH=}<>UY>Xa5q;;45KV
zE%zSw$7!5uWeVs#JYlntpZidJF<|hjj+Koo<AX9bczi_7@*G3`h(wYcU#y?BTrGT>
z9nQlcdn_VhVS*UQ<IN+PzaFt^7J7)P>6ImK6J{Pt{0lUk--jlfUG&J#LChR|?+2Yh
z{QIRacYot}&PlUq{8sT@x&J8mWazpal0qhrG8}HSt6ufvrab4J1utyUO<&hQo4#hs
zcztpy>aHX4Pv66L3v8x-gxlueZ_fpd9k^M4K(XG%1k_XP7L_w7<<SkR+9VZg1#C26
zUu6R9aHKX{>F8R(%n1<K3s<bFt+7BBEaz4k=SSPf#zsEt;8dVZArIT;L)W~nBG-h(
z>J2FOK{yCp@BY|3zCw!D&rxC)AO{&kxhLm&PdWC(-Jd_4cxUC~KEM2-x=qjC_Tuev
z>ps1@#B16SGMcKJ72(GwavD!OV*ULT<0T#@U*yaCt2F!z=hNE8q4g~=HWcr+Iy<2F
zFj{&o>goJ^TA7#MPPWx-{^+ygk4l$gZ_Z|YRE>OeQt?)5xxv8xJ85ptw_qV5M<kA>
zZn<V9C;LR!|2jk=PszKxp{JpQ0+QwV8pRnl@@JJ(scpRxq_9V5wn+8i&!`WKoP0Lr
zgge@=zo%ZBpHcP4YB?~hk`|h0*{^v87S2q2yly~T<9uFsE*Ea9Z6foJZdOfm0AfGN
z2HV+~8dSsme7G%i^s;E9dc8Rh(Pv4V;K`|(2O_>D!(JSI$vjC$Mof3a!R*x!L!3}+
z?XjTk)1iPKzQ23fX`$Y;RR6+S`Q%29kcs|;tX{JygP@sPe*Wf^OMP6`;n;T_$k7k%
zZ~{JRroRsVd7(h@k04iH?!UDiY<G?ABOP8;9}<FQ(*17DH_A`EEzo16&hGf#R(F7X
zxEu6-@M8F5-_OK3k6x`lIrspOdr+j<o0$pQE5_l4hxkZgvDs-9@A+f^wjo&ewQTxg
z?IF9|KvFah4XQx_*%oGMH9(7-0-z01&0OzL>&N+jOp$U>Oy4>&ihvq=igdf-MOE4~
zmwh?pG|mWPPC#B`0fW%?nx<qDhbW++Xz(b(hYt`5WD;m}p}g;lg=_k8JI@;0RO@E4
z{}l%f%yoT{&02HL7>T;|mW)@YWkDMkAE*7EvA%OT*EbGC({?n=L%F{yrS57QH1lA5
z$JV}QhwnLACBX%C!as*&{3kaIrdbnJCE><l8mOnvh!m=KV;2E6Y&m^LV3sJy6ih*_
zJwOb5(R8+-I%DF2R5?oS80&kvNBS1kL-gKX+Ocb2#%O`rs&ARHN5uswA0jIY)z^ov
z%WgU?a+6<?!%rJ%N?~S92^Kh>^c}<(Tbb{GQeizAY<4ba75BxSKktyF^$VqrxPR%-
zB)blErZzmbC+lY6-#tNHss94ChC&Ep@^9}?=}he{oz%F8o)2dg{!XkeEFGy|N%t7M
z{2HQWwsx)`qwTI^w$y>A{jyM-#JTROEo3Ru);cgxhMy}mSE`3j`h_2CG;fOJXZ40R
z3wc>}bH$JKjcmDYGRR2|Py_E~JU?>uaUNgxa$?cZzt;=x`8#AzzW4F7dZEb;3s;l;
zU45lqlmdDDt#$lLCGXKs*IrF~6<fbI@{8+_3&Px-mW!IXIqS5nBS^H8$Xdqd(m`kH
z+g}@c+MEg{%tZMo#{9OVF;?-}E4BVSnez~_du`lF58=b0T`&(fH)e;jW%SexF*-VQ
zNnwgLo)m$b(dVI<S>!}>BEtLWvPrHl50xMhM)6}b5uDtYOMBmY>kf;N-PRn^k#zGk
zz^yVo@)Enyn(*>T6)be3rnyn*2ythAJb3;(6ZYB&DKQf_Xv#qMNEGU;GEPpL83qSD
zpEdm9D!z>an|o9SHi+qxWTOE&&Y2KY0R)=q|KK$Y?`aUd<s3eqQaeUAbirNSC?rJ~
zyZYslSn)v=Pw#P&hvO<83F~T`><@$?tCmL1e-cEWdh*Fu`m@zG=4z?Hc{A-C<3Hfi
z&$t$ie)D)Ad9=c8ymt|OyaX3T(@;<2qyVQ6@%>i`wYHhXb!`vQqnpeWO-+FNyk_IM
zjUsCl7`0DSo+zF&T!y|~3Lrj^LC|eQuGqc^)2njtyrnJcAKx>Zi%2PaJFB{1QwwIF
zX?qk6q1JaB<Lx=mXK!ie%iau$E@K)q410l5a^PJ}Baiel29fL`9PhHe%??$m1}8bh
z;m9*lT+(HX8m<*Y<!ZRyc?o4akAN8Esmx|o0V$#smN%j=Le;F~qtvfmFsV+wZlkVO
zm-7W{#%^`Y5-juGf>Dl;%44GUy=PWLbNAIR<+-UJR+4StP94PuQ%Xz_PMpdwTS{rL
z#HWGoKSGlZJe9Y6KF1$>AblMQsgEarVDcd3+_hR$HV^s*+rYH_(8WZ}yL3v^KHELm
z+&pTT7WVTt<U%^(o=Vfh;jGL2zZauO_e|`y_oXgb;*yDSiiI9k>7lrudL0cd%a0h1
zOg;y`W5mlSX!KSi=)w;sC${RDD?X+(NMJ}Yk~|QALF}K6vRVdMDyiVUeCR2+&;OjF
z<tEty9d7eU9~i1_95spjV-#F0tC`ke`VUIkA&TzP%+zk_k?k8uo@TIz1G7*bnqi{q
zqb&j-)P;+JSlGbqPosqkSS>|H8w|011&VKp;r1Lwt$*w_FF_p82N+qFpKFT+#pjC^
zWBSU1=KUTH1qSQhd3_C~e(6+pb$VQwjg%+(Z4z4a$hV&uzDVm{Ro;x}%L~+6sl5z9
z9`_Ev0Am&pyCX3BjnO`%z>18|dVo=p{B|2{8^OaQ|MOEz2{&mffD0&ZFelJ$#r(kF
zhf;sK9<uI=)B^MBIq?^-JuaxzEZj6k{My%ss%C05T$}ygT&OFv-X7__;&s=R@62Hq
z-a7W?rg1+O4#GO7<C+4vhdKrymK%?Kg)~2qDcJW+wVn7l5u0;=3LM`($)ahiKGq&@
zY9hg|-HD4*%(>hK>tPkXzbUnaC>^Qs5MMF3I<v5MI%9bj*%0TJ?AP^)^xQMAQ^P+*
zg<oSC8!-3IjrS#_twXJ_(+k8GODFR0W}c4rb?mDXPrRgtsYMCqH8lj3=A&2dU(}EJ
zk*dPJN$wC;m_E-!S%&e6UV(%6%4U^0sb9fk0q0q*^~A9m`5nl<?-`mRe+Kv?8gI*T
zs_2X<)z*9`KW$`EDHpqx|0J>~d!&~|<cn2&slfsi;g#8r<g8(nM(6I@MOQnGZ2L#m
zipL{{w*`<IbH7Nemg7MvL>Y^|F#xU~3^A#i_m9AT@rN%`%cG8jN3@N*r6GPJA@j|T
zHua=4RV<-rVOyVSvx7QPyRJ6|c_5=QV>}X_J8>7@*O#Tc*0bGK498$|1u}FYK}1el
z5Sw<@O}>Ke(Dr<HoNvWLYb772f+msH3gwK_3Lr70LsEH>&(tnw`p;OR364tapAT6~
z8EZKY5M4z@9@cuBbz=(ZFr(6nMqJ0m7=yY5%a|JWFE&FGzmsx2JnL#jhTQPS2j!#d
z>%%C~GFd;m<xV4*#QpNnZ&{S4|9$Nz7Hv?L$@np}7uXHk@C{Gy?0F@t25PnqUv)+#
zha3#)qq#=QOh=R+DY`E9{JwCjcW!2EY?wz;v;8prZ7x^*M+?^u&H9gzO`ia+aK#L_
z{4o7$ueJWr?(Du_K}@_~+<3AXKOH#jy=(G+1O4G-J?9w^wbdTFT2l4{Ly;95<W`ZG
zR{4Y&b2fn}JMSSh2Gnr_unM<Nl7{FGUn0W7avq!_Z(R2xxr!o7eW+g-aSA6kS(t^f
z&e^k16!OOO)e#7@jr=2?pg_rmN|EijAch>N;t~v6s+a=0d!6f>Z-z0kTgn4CCR-G|
z%SumneKJ{mr|yZ$Tb0alMc~$EctR}@YoY=X!pls?b-gbaMhN=7+<TsKrR~9{Ku&DN
zd-;Bb<UN_6Pp3bQtb-S|_a!;RL^CoB+0#pLiSrui0C@5r)`c@U4J+fl7v#>zU3iOI
zyC#|^>T7IM@M2(~fXt4A0>wChYagMY0Q4#pCgc>d;|!;nn<+Wv>HnoI&e8&;Xd#va
zQrtH^0nMJw{_`Uy4^$c;6AQw{;$qE&y-0{{3a1Rkez@Kc)9Tj9c{IN^t&X-uFjP~R
z6-25PMQWF)WfkG$zML9}gdDQKXoSj>E@$4d!BSxt87-M&t<HAWk_GlAt%R7SG&`5%
zx$uMZX*lM2kFqrz<yH1)TV1bdZv2-qZP^kM2V8B~7xWeKIKpNEAdAyjY)8IfBP*Fn
z%+`T#UZ@<nR>&>&GD@kB{(=Fwj1d)b^Xcm1@tJKrg3wXPcGR{Hokh0Q|5lCVy*6;_
z=rHVeCI^sjvCVDLm%xh!zP>Fi&KT41O7}>7SkTFlsRWCy3F#F0-L#gTDKE8{!kiws
zn`r_G{PfCmg?L*a`v+xRSd+zw!9iqd%Pwb7DZ8}2JY;0TzBig1&9rcmYF8j|w9g9L
zy9TwtGg0`_OukQ?bL7vs<192C=b$qmzBc(pOkG8E>*z@=?PJhHT8r+*GRT$Jx|85&
z?eV_m$TCJY5$$tjrmKGoQ`sl6D(ud)bq;TI3}%(njGP`^FovZn`E}nut(nU2<TN}j
zE6`k&4%jVOvZz_DXw)&!D2|kl9C8P7P9WK9zivON0%5tHb)D2bdcXtt<~fxox=}oV
z6Vz}DRkN6pLn_1IV(;Feolm%km!7y}yZeZ^lcqGX<{IB{VIQ$CT3bNOo=IFE1?}1-
z52YX8$9L4+roxexN&spmRcfmS6c&qX=+)1>&C0J6LH@~HCIH+$@xe_cS%SwTXt6z_
zLBA1xmuL|yeEqiDPE)%b$5>67DncT-s!-09o!O5Nw;!1k1sOuQ&abe?Qwpa@??XNy
zlQznt8(z1VUs&MMhU=IfyKKkK$U!w(@6KP?Xb-dlD6Y|OWjPI5E();>+^M40jG5Nj
zZ@%3dF!a23oH6`dS<Mp`i|%fksOhz|kEBNM(6zJ1@x-PhHndAs@oXEM)LlCDckUB>
z#?+ez#mTlzk@?>hA(_#l)%(^>==_9CWsdG%8t5cVjjrPaabU9nRs7O1X-XTQJba|(
zsqrq^^}vj&PPe1Sk@A!MqlM3Y=gh<(-YptZ8^c<t_>X=fN|pn3Wy#_U%a0!;z6bsb
z6tq($SJ9pmCDN2;-}1R{bSGJ%yCt=abFTHWnn|5vk*=Dbcl9C=l=U;h)Z2*|-7=_!
zkNO^hlM4mf&3{ABktSp6W+kDTGcK(zC4poC$dGHv-L~oU?!Ijo>+@3iD1{tPMcB$`
z%h);ovD7axXiOOcpD_Jg1D0?~MH#HdFqt+OE2{>xbszVCNM!2;s*3D2k6_Q_AmA60
z(2C=7rco=Ay6B4@edCqW6^E<ZHSf09BOQIbmjhx`zJv>&i7f5DhX}RD27O{r3oIOd
zoM`FTs7^_C8o(tk+A-5b8eW@FOv9bj)kuhqoiyU%-`{CWXgfY^#f!}$;4*<^pzF*T
z!DWpE=yYHPr2wQ;axo={!%ti7y4o<%t5*pJ6+di*j{*j#aYIzhe4nFEmL`!;liUl^
zjgt(J<~d2KWvHT6)iUVL*-mRz?Q!RUihu7_{odo@(RA{}-=SE@vf`5fuaUHmbHOT5
zH$7IfQ^<;=PD?B10!=;eVK{pV#SYEiRS5<q7urk9`nko<tT+#8cdHIN0O8n9=5u}Z
z0-`i_YG!qZ(eZ^>az>fTqFVlnW?l1q2PDw>rw~oD1@ogDywQC4%IJy2_>K5;<vR?N
z$tMpl<*|ocRiV+n&C=fz>@149vw&vv`}tYBNG^p=Q%xApx+OG&??Fk{aB-9#`oIK9
zUAr^OG-YEUI~i?hTzB{Y@aAAfXBECrq8011lbK6{h0JdFe~CQx+ap!Tt9KSj%{*%U
zS|KBwS<X3pF39T0|Gi`R5mYTdwEQy%qsxw_<5g$R%5&{1MuEDpz63rvRIc5bB?Uqa
zD-(2N>^!WnRWBn49>rZMec}N+Yzuzjxn<nd$P^JX<?$i!!ckz#6~6~SIM%OcZ9&@~
zqYJaShj!PMd)gjgoFS5LAov@qy2W{{6sGrMNj<^Iop?aU%oj7uoKoQ^za(u-%VQI>
zSWf2=xS%8ZT0P1>6X4l3?<jwlY@c1RN@9GVFB<*;$Oj|Hkj<i2FgHzoRp;^xSmTdP
z1^Gxg399*?{n)*gyTnDN=<{{@R(UzE{E)^Y&IU60?j6yIYfky0Z9D>+TNJy^!XdO7
zF{aGVO3DMJcT&cBI5w&cznJb%LZp#O!fyYV|5R&?cQ+qU*FeKcIX{t-Q5VeLT;Qk=
z)5rGfvz8ZRr0HvdwCfrEmKyH~Yxxl1eMP?CT92s_=ZTQ@o;s_qXlQ64Sn%as8K`e{
zWwDMlC>N8}iA~*j7P6uozfuuM4KhD;(pr#pXX?cUHVUK{32B>8VFCdt)Yga~wHRZf
z`zI}!;BARKN`?0m7ykwNT3K<3?mKIQE`E(R7InKlPwh?=U1b_m`c#ko@vNxh!Bp=J
zaqo><7tQ6*8>)YFa=-kmTC7Rk{?>YE%*M-d_g#hFIu3d2Hcm7Ob~Mq}c$Q*3<hxFH
z(ENC#w?tB3ug*#=gP@gh#Hl~54)t~N+C~Yek{CAYCI};VP?hvo+6_f@hZbl*ibvz|
zw#x5veC}XhsDzc*f_vnh4t*?oAGuTURV-ulU!Z1%5XN%TlT&ze(EF=OlxU^AUP!`4
z8ASbGFWYfC9D?~no^5YumAhg*M*qA@$+nPQwYPgpxi+>L+m>O^Kk~J@U0oKRV1}f+
z5?rX-3KeX%A~AGnplg^mc@IqrvKucx&UT?X;b&A!1(Nn0u?*cnm?tFh{OC#To05Nl
zY*@MGz{|r*TZ`nGS==03y7D3RYP_5ARA~BB=HUm<&12G0Wi!pa=dv?T_4H&ZYQ~&`
zhOCA?F8cnjZtZ#VTWPyP9=F_X{1=F)T!gUp$$Q!2xHMGp%9cUkWi{8Rv8VQO<Np3%
zkoG#vA;7>22};d?Q^wK}x2qZoU&06Q!oy$7O75V9fd!@J;m8%7BHL4EF9l*%88bW&
zgihjSp;8%9qSs`obf-kRVh{sIHHFHQJy2?I`!(?lxZO4RZ4`J;-x!84gBtH80|6w{
z9bj<5d&O#nZ&t=h^JzYc1Bn5{i;Lrr+&GX@k-er9V@v|Yzl;^$E@pp>5P*L>w%5IM
zLNdEFueI1<RJoXIK{cI4>t64bonDl!xBEMf0gU?l3nWC8As(JMEdz9rYj8ZL*fWZy
z&u$iSk#J>Z_U6N6{_Y0j8-3|gX7-`B8V|l?4b#?iC&IC^_M6@*tPrbRY*CtYRy&iJ
z><upl7k@B=sJIYT=qM|eLv2Mo`G<>GlY4iWMbA|VgUbaB2M2i2Wm2+f&`i>h#OxLF
zT{?HRHPll4trK&$?vbTqY(^G68!`RujUQp-@g+OyD6|Q(Vm%hkN<dbOmz2>Ci-Qw@
zmTZdoYtRd?SQmHjwd=jUK|y;h50lETIRZL)$KbYL*u1%LL6(T7yN#`dl3cW(d#Ac}
zaM&n;eEU-6V!LI8MSf56TjoD3ec1$MUh+L{wCUg<w|afE{N`8r@9e+6{l-twxq>+w
z`N!>8ekl5T@!j1ikw!X~ZhqFuF&j1O`5tc1u(Bt^Q#V?N@X4k;cdSwD5lEH5KKgx!
z;g@W6<@9Xndus>&-8scpCFzkKU@I1k6-8FH2?E+8^L2sd(*_Rl)Gsvc{Ch(?N*m-E
z8+4LvT?ZF8O4;D2)63dw+4s*R-Q*)H>;(U`VfKC|rM~AaQTrCHRa7bZ2^H^T1I~md
z`o`ynE)icg*DXPJ%wJ`$p`c*U8qf7LMR~c!BFm7y_Wnakf`gb`bG7C+U$50{c~915
zb@8)m7Fnx%c&+}ennosZbyf@GaFU<U^}Fahn?werl@TYxy^~zq|KYw(M`NbRD9z8J
zcHQk-3DXbEXcgba=qW$~4~lo}Kd<B5X$`12EP1ToSRbu~ldy<%iY4Jc>8d%K-qe<F
z`3h-w51ot+?@`g36FEq2rO7JBrczo1Y5c7iyT4DS`2p?l6elv4|M>f6`SnJDT)C)P
z_uf6S=^?36E_RwlzqB>9TABI}FBN^~FvU37WW9OFeCjHP%IvezydM9BTF>zt!+BR~
z2+>#rE@%OUS#lY^N_Mt}><w)cZQ4i72Q_3p&oQ$$W*+|-E;~oP`~sqCrBwIyQLVl)
z&vmI6!H-KK)||9XYAiv1Qy!)Y*%536Qk7=OX&Gl$$sDMEr%=K%@fKsl*A_`W6>n09
zh1psgk8>WC+a;6AWrx^etm@R0C7;jp$KLpyGeTlbAIrU|4a$2Ej~J`$dI<`)6fuz?
z*bB~F(;H%*lTWs+`z&flHIvz_*WbzZw=<acM=?S{Qb|tTh9>aoqQaNqiWPc_je>@T
znD`B{zz|4|5*%60c!5WknwOkms6SyRES`DV(W2U%e8s{~o3{7vdX4hhR+XjWWcUVk
zFC=5C2A>s<-Ts70E@Iy24R_BaHj+NXPr2`7*Xg%PrMyh}YU#1Ums3jvWqu~l=|a4-
zs$!8}mIrQE7gWn?;u3dS$ulDG*|^|y>gyE1heOphrfH4V`tg^9A0Qpu&4i0~GjS2&
zLrBRHi->c|Ra8b%8xVAWcrJ&`fy5Ji;Inx01DL@xs;2VZ)FOFw1r>`4$ZE8`rnv6}
z@a+5c^B`bifn;Jse;^cp*q?xIQr{rB{>V7t;Q~gNYz}`feAF0M;Wk`Qf$QA^&xkzP
zs0=QNu2$jiU6!ynJ*&^;F=TH#d%0MDsf^DQmS?)nQ!80se`WZGVON(=ua#_GK~n#|
zfpteh9IWLka>8_A`4rId$!mtZ7*gu%!Z03dQ%~9k2c!%jIru4d(=t3V)iN3mTy|y&
zsE=i)0;G&2X*qn-2gVmQkjaA!qKLAS?YYd&z^itYdYOrmg_Kg3S3)a|gFp>w<Vz6t
zZY2d_*&9O}5#B6J)0u>%jCj;U-iwI)Am1HmLKHYyslUSQfJ>qNGWP0_W`Q$==zJ)N
zm4sD|ZW6_~c)Er(rr?a{^+jl2$w{sAfOH5B-=G+LLRLinOn4&j{ow)XRqCHLzh;Pz
z&ZOXB5eY-u=;pIK#WEc&!c{1?#i~W0iD1<@KE3i#K0n<EA_vclNa=<8go|o@@-b=C
zi?#G3-~yy{f&Th)eIlA~B5yTvr#1y#&jp>l#kNxg+CH^WeSoUbR0->^;J(AQ*Y5Ae
z_NMtvT*JM4xy_C7X<y`?;K!SW^Uk2&%>p{cDi<FiG=qy;tp(JFPl00dE*4DY`9>4Q
z0)4d+o1sUS{H`10z$CPe^|8}MHDvkn?h-`kx|}!wSr7bDVxy}a`^2NlOmI;%dLxh-
zcRUG*L}jbl*9R|eUlDd`&$}f7V`q{cQv5^$IzXhP7gWhZp0#k21!J}Tc<IAzm#622
z0kcl}tGt6D6!~s<wZ`HsGqVu-9M5nZE#7ofOYC~PB~{>o9fKvWA$r(ktBOfpSguiS
zBtDzYB+_Y=HJtbKsfQ^z)+J^2JyQv$xnkhbTz`GIb$&k$dT#SBFMbWUSO=(_3|%M2
zQL#g<-y?{x8QZEq9=SAS{68d}ha;Q)_xEG(Dq`<YgjS8%)Fw6&d(<W|YSnCe+a)y;
ztHfv!d$iOndQ*D@wTljWmDZ}3w)b;=f6pIro$GwgIq%o2K>S_cv~hI6duTgw#V=Ow
z+u%^HXZjm?`+^E5KQDb@#5(G7KohgTHHLkSyuDx$kWqw5E`p5-hhgfa9u5gXM~yZY
z>(LP@QnJDaX0&B^zHYG<i%<Kc@NSd4?CeS!e0;r&X>RF>m+0va4&5GQ8s@vOXYDLp
zX*;bhl(?PV+iI5q(NRyot&QN`l0vPr#-pk6XF=+S(Do$f<SNh>-hWaR5LfE8yB~sJ
zMTZh`oeH>#JALU&;JVNOaM~UAN&g`myc`*cLY}A$n^6iwLYz>ppC0Djpf6r3+h2P2
zFd#c~6Ss%<ayIGP#Y9LNDBn@1_u%>?(JqSDftB;w#HO|wLE;NLBxIr*-WB;!EG9gi
z5Li&*n+uROKh0Pv6l>80Z_JNz?Ucos616TU=5q}UQ1oAbD(;XeC3n;jIRt~0akH~n
zTFlnc`eUKs-wQe{(<)Ut!+s!|#C_8zyh4^Rrw+~TA?Md`WX4(E?l){@;t{V+mxLW6
zp`&$x&Jy=IR4jlz%+g3=gr*X86G7z;DK!9aUZrqIM}lm<R344!WB=E))OWC(Z87$D
zKJ0khV?9%1=T_JgNzKSwtPo#}|KHZF2B>l3uvx&RA-|Z8C#B270Z9=Zaq3APW7!Q2
zKe5XAuk(4MmgV`_zByDJ(4WyDJczX3Z#q&Rou*&W*Q^lrHe@t;+BiDM<)gknx!g9|
z$yumX<lqzh^gOZ`KKM}^4Kkbm2$Z_Y&szU$$%Jn9id}(p{x{_U23*)1=ZqGxnD;x^
zv)=m5=|<2$D>>ingkOKV{UVO&I#k)#A;KH>bHTqCmU|~)Eo)ulKM;@K_|-vJ&RUN&
z!qVocWz0PDBbt~>fvV@10d_f76zTZ7tLOh7I!Y9-HHwv$Ubwt8+F~;wY=PdDQ&Av$
zBd+ov=)#(XyHkU#T&)VAO^%Kabix?h-wO4ZGu)2*4-}2||A&?R!FlRE*KM^Y>^H<P
z4{AG5u~Zw!@%<P4Lk`0+_@R29>To!Q23^!9ESpPDDJ?N1!_<0lAHXliqu_*W#Tq<z
z)Uhl(4(dEAf*5HFy=>vcHh;5Wk!vla!_?e6QP8=w$|weeT0p;-f#Rnrx=~|lunDJ)
z->G&jgbkW-_X86o!<{O~V)0;KXpB<KU2R6*#RSu_s1STd&}T6T@^PifU^H~tx)P+z
zJ%aW>dx54-vi%iRo4c)bzL9`AHZ;$>c}|qmQ)a*IyF;8XIL;5Pd#qgRojwq(z3qhc
zFF6iqmjdP}Baa01Q3f2bwrOvJ<e-ehZ_<!?)+vIGiqyYSHZx_N^u~3A>?Wy}c%R<?
zCk7t3$|aH?WagK?eCQ?jvCgw(p&nLhIIg(}52h*4I?)_2$UKCn{>hg;!uPz-YwA9i
z6$#c7tJ(eIu5lpwH3ORal3Msdh(pD8mdI?E>rS@R;5{GlC0Jh<F00XqV$Geo*WIO9
zo~3UZj4=p1C))f%fu&`bbOyIVNaQdNuH>*a#BkXOZU|)drwF{3BKf#B+kTp6GNaSl
zG$nQD{1q#&5T={@?fVPSxn{A9_J6@iQo$CzDE(+~iD09NmOUerZKt5(_a|*4`u@3}
zik{lkS;KF|{S)O2l+wc&7uxlz)~6eVdP1jKy9%OeW8C{%tZ$-IwfQhp4<(k5OzBuU
zK^L{0zpTl|-{#$Zl7!$g5*WzkM{86l!{W>el)oXQ0)GSY^5$3KwIuuWvw5CuT#yo_
zFA%_G>~xbuo9O?%e;N0pGH_UL2D-OVcMI~dp(oBlA=<U9<*Ndo06-D|o>+1zv0NOO
z^tYxJ7cyw*k!=6(@k>JLLVQRDld_A2qO2N7T=OUF*!hG-t4Z(E@itG#*Q^e*yW9K5
zSNZu*@1isK`bvkpndI!Z^u?;m-tyQ+C_&Qbx3y~N-NGB*&u+<w=;CqWBaTNE)_Kdr
z4ME-q%GqXpK+0O0o?=R(Xuf}M+#)h(F+P;^>2+5Tj(_(Xn*v_zV{5$WxKqaP6KgHL
z>qNNz&*8nr`l*Uipl1{S6b|@ARGp6Ku)nAFyyN0Ok_LZPOaEjrqDqPp)S!iA1u9l@
zXdlKy1bbrfT2K*1J8!9o!R#K-z`UG`cUpF~fTh0o&<C8Ag^jcC;&Y3n=O|{~*@6rk
zE($OkhK{8Ola?VS$DsfaIxN?ap)8{Wdigf*snGL2_bsI9J=Z+AzMY1j^_dN=^(BxV
zymtf=H!m!=G?>+p;JS`hWn8EQ=Z+qz)c7=elN8S)kDdQWuJy2CNR8&p#wU{F!99_6
zh{p$=$4u!)5k|#lpUS&Lciup}mC9V%KS~H(tk|bJy`1g6&1pS77?Ad%PVVtx&!&ja
zMg@sc3tlDLu!Ruu98ywv##OkHz8K(+H=UPFHMejy{i}|YEvsIeG4mOY)n^|E>S%l_
zN<s^~u1g{4w3K81XA-)z=1KS~*ciH``nLJui_c$aenxKovh$XN>f)VnV%FET^(<@u
z#AtpV`-2{O+s2r&{n6m$UPw1rP4R<Q;(;VszGd0EJ1pve8Pd=nxg)LJ<deaI5}S^}
z8X?Gkm9_$lgxSC6DaKmDfCS*t8n7y!s`}N+sJ^0mw-+Qs`er6?2{X6^Mb_{f#rV_;
zXI$HwlN~xn<Y%;+gwwi~#lC2S`e85%kA`R|-Ka)D$b0`xQTwx*aIiXG*4jQ2Fr!;8
z)Y*1%vPeX>dsEp4YexL({|`ieW>VeRQL@o#YXtt1=mK8xF@s?qsC~&c)0sP0*=Q4T
zE5%MP{hFmVm<7bWnmisn_BkXWH<~tK!e?!D0KP=BV5)!HP$Wlv<6C7!RR55tb(miO
zJWse`c$kMHo|(Luqac~Qv;sV+imPGaMc#$ex)Q?r2|fa~A@Q#SEF1@(?*#C)irJN9
zp)>5n4^8z4UHb00=+p6o>~L>S&KvesZlgf`D-BJGb2l%>)ECs=GRd;-H>MWE&R-Uf
z%*=>LTIJKUnFYo2P&!c3OxIK^I#}##v4<rXpvZ>5#ZaeO9x2yLjnmWX$)^h}aH4eu
zsx{Dh$|8$7P@XI2f`ABxE>V+^2U#1g+D*|x5BSr6EltK@HKPR|I6hAA4jwmJU0op(
z<|}M|4UqAsC+^#SOA1RIalTr7C6B*3k;Y^SqUPXZ`kWV#?%-*f%4AbGJ;vqZd1S3C
z+%>-vO~d`}<tcb3%j&)Re;|FtKizhfK;CB-ahZ<E9>R>~MReDE@x_EiriEWheoogh
zBs5f9f_^kbIH!iRD~@o(Mt_OJ()g8s0jmM>Z(MHrV#<YI)vWv8oxq!40+whxm)RQr
z8W!Sb>7lPww88J<V8WuF><?A=sVHA%&(b}eW)*x_V^O}U1kJryQoEL&{zmlKq$%%e
z?`_3PtD8?kFp(65xvIRB52_n4Wf+M(DOA2{vN;YN!k*YSf94i(ev^gI`0HouCInEq
zvLbxfHlynT_GnP11@0_UGH(O4{?wp~%KQ4`!Ml^hLoL=DJ;;frUo&zKHwEU?>xH(1
zAEU^)pYj`^7#;W?63GLkGNA|XEj}Qa>cK4+U0-9wO;b9IgLeQB^Bhv({^@qYjN@IF
z?NPoEd^caJohUxvPtGWM6BADap6(Ei90dj=9jy-x8amTCwzwiM(?4jr1s~Hie1#8?
zW7E+3HVvbFc{?t$ZYWU(9Jl<c#63x97(LkXt2u8IPrnzoWcwCbO_4tT%H_`sE|`Xn
z+;*g@hLWwJB^`I;-F9L295a4gc6JwKPQ3t=c34!iUi6S-(3Uxh>)!`!N7+(xJKqb<
zqZ^%6oSWoSlz2+ap{(k^#|>g_#GUU&k5-xR^s!IJ=yUfIMReAf^7vQQ89edSUd|H3
zu?xYH*tF-zUl29@hqDFzicjlkv>(Vn+APyJEX%Zr$#+~}liKaod??K5iO*kn<;U{a
z2ZWTLq{nGZ#*QUG#~O@|`Ij61DUmJvF_SM6sqmUS*-EPr8U#(>NPs^B-rQ~QVZ1xR
z)1ucFy9;BSe)EVKJ=O0y*DE8~!9_BJq}vE{p9&Vtj+Cm@-K=};$7e|{(Apr^`Qsbh
zOVU_Vq1kKf?(t_qnIRKc%C=L%R%=)PCO>mcqncA*rxx=k{tyIdzgvy*MYa~dWbW@U
zlJc?N2jstvQm1iQ#(M?`&%mgXCvD4)GN*<>;T5!g#?rz!sXE7<97DtPF6_{XV;5YI
z7zwOqi&IkA%*5L~hK1Cv7Rz$Ei8a9VQ2EGmT>FKij=!S(irKADutc?`ER$j!z0v)N
zw4zQvYvZqmg0=Z5UVhb_1K@>k2e?xXoz@IojDZD`Ta+gzI-73E>wxP?d6!STtguo{
z?1R_1s%PrMG4X%=U8jyeTm4#pAo(R?`hH`W*+bZ~J(q;_M6MlPXB*rt4-%DJ>?}9!
zH~4Mwk^@xL{TT8A#)?I}&t<jyGOGB-`CO=m-zz31t;ib3ufC12xYSSA^K$nub{h5U
zq%<yTYXfF?EsG}F(wmQCnHx1y+EhtuMk)Pz*|^((tIPjL4xR?tjx|NA6%F|zQZ^I0
z)iGgRAeke}VVp)!cj}O1*y&DXo~J6RRB==s(}w<mCyJ@9Ms1H=rkE*@PfF^XypeK^
zQZsyGy;Z;3e4K&%#k2m~aNX&M-_ia0`oXWg#OnEorh@6N3ID6&9gAM4mRDW9Io>^T
z@R92}GgK^_|JQJiw-gJ<Pf(N*o3A4N12N97{nl96Nbj+BGB*4!b<<);*6K+8>3DZ>
z#yp#ovqJA+AYW?m_y8p``e?}ISjt|0-<TH`xAJ{x#R*NR`MXu8;?TvzWpHHY%c8Yo
zQNIg^r^M%n@wM;`f@z8_#(v6=+{&CpiGgkuBv%3Pq&G-2px)O`{RA`a1PPK#*3PG>
ztiuv4wr!Ja0`g%$=Y0yEwU2eZu_+v0&Vt2N%C6G!ZJqjS{5=B6JpSs_P{XwoCz^Ub
z3T$b2s>JRMwP#Z~cK#}djmdq+KCYeNlK~F)OAG30$*hkhC^UveA;rNUNPM|KHu7*^
z8=mPpTn;<}boqMtkcFefaf6f>V(;W#*wZVP68qVw<)i4L={kxvf4_e6H_r&%|1eI5
z7l*J4EN30s!qSRtU)gDh24JrHxdIrkY_VRXn-dc|4Xa9p3{ZeYRT+>ju#*m-=ENtl
z=dPcuQeYlp(q4W+9?&vsJU+nP^;|q}WAGt7mYidJe>asBV26{d>N<QYhGkw?u$$=&
zLcZN*?o`Mh*ls8cg@m2v)vR-?5Vc+l&^0jRO(-p4OhY8%9DQh-Ya{U4!&55v-L3j6
ztGF(zsD3TUovf65CNPo10+(+}ehrKVDnU$&LwllWcKJv0pC1?72<hW-Gh^O)9TZql
zt$GhV*0eU_#a&4AUA`B&2w0kFMyCQ_frhH36>I$4#+4Vi_$o^4^kSa6!?*I>M?kxV
z?!4Do`_$B<fEhTxQ7uMrL*86%5DjR`2sGqsz@$}~9}4;tpCW5z6l9l25`)30T@%@T
zpf?N^Zt>8DWa&CaH0h+!R8kEt+p%fm;oi#qNAUUBd*<?%-JTdzuiF<4y|5&pjNkT@
z_!>)3^cmkenT~Nsd&zO2lS>oeUOC*~Y<z&PZxE5~qakS$wdKJ;0rx3}>pNv(K$s0?
zSu>B1ePhMp1Pc$mvk9BGeY$N)dmAL9B^0%6)asx*vpSC<buaL^QdnY@LL02RNGyX6
zY2uBt99HC+l1S0gR>oY}VM%BTuaNaE9J%}-DAH}WhDdd--D!O{*iImqaY*#7VXyXY
z=QCuIDTX=5vS;q*#;{ydMhXq)pf$tr$WkIKjfTl6^RBgvC`t1%eBhy^>_?o|)1Ik{
zZSK115aF+W<jL<$<h`@NgRK<Y;0gU2k5yYkV_FY+kJ{-7T$4i_fmNzMJjmQy?Xg8n
z&x=SrmLKB@AVu43@;YIC<6V#1Rp&#R%w$BRR?2eSE!g3VUA%X3Is#F3qS5@dBU!`o
z87kB2h#x!g*ztb{)oh+c(@Gjr3NV>Ssui(H-d~ANhxba!(fB(|89w4k^gj$U>&w@a
zWp-^de>f5o0_)Q6oAxvx78z^Y&=s+6iw9U0zKX)Gu|Rtg(KV8~XMZMNdY;_RzURE@
z$ylg?7K<^}Ng;aTr+_=;FR^nBp0y3j0}s{tYy&)a<VRxiWkL6f-zWA%$}-67Rmi~2
z_wvvhVyD6M!-k8migP0+E;fyvw;7C~j7*C0REC<eEXWRuqTW<$x2UTn`6$J)5#tMw
zZ){?gLR)8A0pU&2<woTtxrz8}F9*fxfC%T>DH?;TCG}^fEpO5zEst~M)&@iVLK=qZ
zQmp!a`*X6{$Z6O5cCt(K($rI!<Tkmde2<yq`R<!&Ce%3){dYm{^6mLozQMrQZ&tih
zs&lWJ^p>ix7Z_YaGUkUusk?=~quD}@szhQ3uK2Es(tIW(`c7kCR7M7CKkx>nEGlHs
zR9jy(NLo1Yz7q`{LvH6N;A)H1S|9A(bw0DcH9jVw{bkmh==f?YXXT}l5VFSe!0Y2A
zo%+0^^VbK)!e4Hnc@;a}1R90De@VGuu)vj)>CMc`C|A1iLf#;RNc430X>II3(&_7)
zTHN3w`_;c;Zara*w$THQjj+OeK!*Aa2XC9se5$Ce`K0G9Btx?H+qH^RzRQbh{xrBm
zPjG%G$9p}o=RdXlCH_-W`b%CO+YG#i08}Xq=U9;GK;Qla?!Lw|^UUduc8Yh%-<h2p
zJ83z;uj*=C^R1h~$NOR%AnPIYoyPGd1OP{SS2k!~XFjRl(VY`vqs)E%5l#6#;H@=Q
zu!g$^Jp>gmzn!x292OT0`|fv>FoTNc1u8uKX)$mc_%;yxTF{s}tj_~bri<0VgEbBY
zwP;j^95Q4D9{aJBp*SaIOTs%XHOxT1&H^gmkb;dMnTHA}1EfbBKfw2F-23&jO1QoH
zvWMsG+VGRyF?qT%50pp%oQnK*sW}b0jb*<W$TcYOng5QgDh?$0xdTBtrr8ASQa`_>
zj4a$^^R!ko1qW{q7=#xz7Tl7B(`iBv?I%~nw%nxyC(T2-g8kJ82U}TBZvPX0l{wr3
z;}2T>H%yEo@$_2IxVo;-OBa%UGkBAq$vG0-E-N;5ob(ta;*rj6PTQ0era>Y}{&S~i
z$nr#XVDO)UMkh6?2W_{1$enfiC>{K3Of+rR5I3U=yvft@5o)p_?{rgBN2;V}1C{ny
zWaz#QK)F5pgSf^_xhL27&FV++n<>g-Y=y)Hg&TOw640W`d7fFjVpR;q$b8*Ux!su%
zYx|+IZyL4n9{Ov)exqAc{m5G*Tb$m~u5~pkD@Zs*2k=A)zj}^LBO6FOXMdq^p2Nzi
z4Et&#jmq~fR?9Yf7}0!`j__)ktuln22pAe`S@k*#H+i_3?j9mYZo*kov-K^T7zQb2
zq1`r>Tgx8^l-W9@7qdB(>rY+J2X!H!k+Z6rX=6~-?n}ogj~h%B{2h=SMinp_qzmwJ
zSa*i+!^s$4N{+C5KrM8q8-<geO&LH_X*Y2C*LKoH7U;Q_pb?C|2x8GCD{*E%-FOT)
zGiOlVy;In^$t9MfD<XcqhP0N;8KD5vVHH#jK0Ig@aB7s}w(nqPQ<M-&Fw!_6zK)#}
zZE{#!_v6Ui6$gZ)8?Pv8NRFAGgFt+iRLNPh)$l#CVbw;By3uM#p_tc-M=@5iVn0a$
znwq}<ZS>GeAu(~3Ky#;kUY$q9Rc(f%f2X+d%#w0qwCZ1T2^Km_6>$Bx*|@t7Z`Tr+
z+6*5(F@P}+g|+(NawPd850OdtWMoO|H$G>`GUW<ZNt!@c=qSpzEJd1L^_UbSLU%jQ
z>$~e1m_Ae*UJ&%gKYTL_pH?Q&mO$feVzAH*rr1>LdQ81yJ=k-5Yes=%^18%ji^|j4
zMkw3-Ltdu*EM9Wa;3(G#Lctz>Lz1y?-W)~dsEZ6D(pQ6LCXZcO(C>r}P^sKHr!h9X
zUC${gka3gh#hI-qvo*SX)vq>x`$_1iToCyvNJ1<?g?;w$_GE=nJkBWhS@49YG;Q^#
z<h)0FKZBBxJi-Ch(9`BGKVYs+pA2=)I107nr}+;YsF$f@7f;xPtVhoqX@e9crmb9l
zMtWv<p&pQ_Usf|<pL}!tmfjSiDDVIQ<uN_4hio(ye8WfQq|eLca5<@OS@V6_Xz>OK
z;iG=nam{S2>|fcW21-|NNZ4O*yw1(G^X5@}E6`K$`-{-;L&sGBAMNeqxxdm`w~;-2
zYxqu)E&fgq!#jz4?_S<1HhOO;&^7=acf!{Y{j6N{{{K%Ri5#h9)_xn)c`Q=&HaFY9
zw9Dm}cx6npZQ3&l)$bw2NN%B1Df(1g-%05L1U{-u;XFy?l%smsnYA7KBp|X@z=1!h
zuusEZ(b4OHnC3Is=Qj3BqAbtFbeqzV7x4e*`JS0l;h+6+KW;$fqN*?r;CaaHs?P@s
ztCML_8y)ypd>>A@N=!Q@<2VSLK8Z!-v~&T^X+k-XC3VBHt^!1lF#<K|R)PT&h}up8
zyH7x(!(Y$UQRCC<8Y0Q#!G4gy=cql@WvF{FlL^n-@S7ZZGso}en%&W>KeGglGpk;%
z-*tW~Px4wbNuO`-du~!{<h@*s-_uL@q)ss%6C7a&sq1cj5mqa!hTN!AEW7P{tCP4I
zP}47JweT(51j<*cQ@m8vp?=~L^|){%^&^uwKa{3+r$Z%~)&d=)FM@5ZkkyDIQ5X__
z%6cbS_ys|t3hrrzBqrLn-M{IC0_HxEw~7CRJaPVmmiF22G97q2>|Ky{$A0T1W#1-!
zNj0U7cL@7|Uo#2eYbsQ}s@jZ|HD>bB`i|<Gp-4mkXqqrf0DLN#L-IPjlgcajk-%uD
zDZY;M8Jn)OWK1&wX`rvF$Lmd*ob2PQEDtCcE0$+(aa~g0_#^r3%T7Eleyra3iK$P)
z4ZfZFkKsE7sqVMzSsZ1v$F((3sZ(z1?<8gJx%shp2|Vp$hgqC{nW0pw;G{06k-V47
zL~9@uW;F8$#pqg*XrPLaps7@=1~W0IVm})%TYiGO$!X?!%iHIlUr$2^A7HDi+fW}3
z#BK65FKS2cy<JGFlO*wwxJvn+3P#=d_iwWW+XIHXyv=PRHC1tw3$Wj1__asX#Xiki
z{ryw1OM3k8lNJhNmf<4l!YX&>BN`_hF430!1Ya)qev)5@eEvx7{;p!k3oCpv9`Um<
zd%01zzk#>Ugw648Qq3cMx+%{q`f8!^n`-l|e&eSR<mIkh^tM9bpCb?LJ5rU9@DRz7
zxIb#her=T@Z7)_a=Z&US!xq^CeUPaqW%y2cY^Y%y$U3{fV3*J4!K>x~vKP+4_&klx
zJX%ODP%VD{$Wd{{u#jsNYo<8X-0NFSi)7Yzgx;jrs0)%}Ud3d~4uO0R7?RQ3MDET9
zjbbRKv<lEgO_T7hPzbyw{k_{|O~ly{$RTBw6mVlR3#j5{k^{JY0<Bk)+|~&uF-Xmt
zOg;;3qkc9U&kRsiG1!}b<Rr*+<`QJa4gs%l9RCu#q<%MU5j&DAl4$CV(T7Cw`fXzV
zr8rWdX*7b*rkxTn?SkP^RT);i_tfTHQP%y+RWnR?yL<IWhj6Zv5!eUb=qpQI;V?@z
zHtkyu6s6RDCI+5U!G0`gHLW()J{oix^XZAfI4GtOUIC_kBEN{Y)2Pi>C>O|M)2ePG
z7x53VG72F2g5w%~4A3Z;5q@=n`%vv8#_n0vaRSsd;5pzy3W@P4(qeNm0)9Eq6NR$S
z)Shsr+hzsh>0UOp;g3qoi4?4s!mlsE6V_9--%Ry?R>!L3!>=L*%HQ~_)f|^G^D>9*
zBZdpTcu-<QFXph@fu~LA;;W2Rz8laCDICMtf1tFqs)=ZkyWCYMkD6$&FVJcRLzOdB
zZtG2h8W~9v(bX$4lZtY5IX#f8*3dXnm`NK7tX55bHcUK~Q{smbI7Bz`1;3x=3|X=V
zN#7jqcd=m4{K^-K#-=c`oX$<f%#r(zfOoid@%l_7Ft?zQMhZ5ysOmmhW%tY|;&2A4
zaUu7M+XA;pw(J>2;w-V{HEUKU&um+sSk?rVoR*02cA9XwL!{yiRnoOIADAF-k885i
z?s}^5&)24j7p{4q;kyk6fm?ixiAQi-FK&^{izVtQ?!g$?i;v2+I;J04re}CQ8aY{W
zywHgX_{Q}ctGOY0isC#5otn8@vzV*Bh7V=^9sIX%lIE<|<NNmS2D!S5PhFMIs@i=p
zw-?*L=SSx)ztEC0W~N)c`$^|X+mL!d>2TeP(DNsP?dj<T72g|aVS65Y0FK=z_Y!6a
ze<h;7M|GFIP{#*vCN}I?-p!a$0(V1IB~hSxo7^BxYGv|8Z?bM1?b~0<uBYw_TpSJ{
zJ;vp#IDu@`vtdKWceSFt)0@`9oA_;SVW(WF*)j(P!?>8>=!WH)cYhA#g*2dGna*ig
zQ9Kwd{0Qii0lzvh9UxJ#g}{9IW~fJ=!k#WX012h$>_q&{+YcV$Z<PWYvxD8p>=$B|
zSMal&09{GsYEi6`drPXAr)tSXo6?#`FHlC~fG;H=qV^6yQ+ev`8D`T3ld*1}lvYW#
zf)2!Ucb*GR%w8j?y+uzTk7It;Nz|sZNTp#+Y0VUp>Rs>!;fE^WDDt}PcDiXTN}708
zqd-@k`uD^bQ}3te7%J=2y_<;K>h^W(s%>tZeEh)bzUa!k>%zAu><Kl~$-r{@O-;St
zdeb*~;|Z<#mljtV#KzKXxL`@R?S1AZW15~f7?Cxy&bd5fI|O!(myOajZWAJj=DfOG
zk=zPib;`(pWyty;s6|CH!+UBgfKc(Op*wdMQcFWS%t=Og!p)0CzgK*lB3jDp<h1YY
zU_Jivn)vZ7?8xbj4PKi&Smkf{s@N5$V-Z4Y-Ixg^jf%9SaOtc`Bj}k;`SaS2CI6A2
z5YW+s)U%udH(auTo>Z^*?KZQdTL=BM3LK{wXdO5+hvV-|KFg{^h(O}ghq;h@hPsOd
zX`ud^Amou)0R|IsQUBwpQ<_Q6__g$hOv9M8zbC*sXH2Q&ph29-8?_f=WqiSYJm4!t
z#1(>vPnG_G_FOL{?&@#&$zkzEeA!9*MdOh?YaihxQwoaj3p{F0(M^EgE2#^V`+a;W
zHnl#p+vtIsp9!b^8yYTxRa=POIL&&&b%*<UU>48I&dZSJHu|Zpux7aIlAwJ++vwiK
zwjA%WI5~#B@f9TbH!Y2r`VTmi`nTHR-fsR#6KeMl6HS{%sTu3X!aJA_LpuAVpWOVz
zQu^QXm{TVUez!TNH!|_SYyP%zX=<3IdRG1ivK#VpWiD6@<uf*gbF&8`DG_R!{&Fdh
z?UAnf?%<GptqC&f_r^EbX5({sgFk}UA&x>Wb#W9+nXOGGVwofw^lLuJ#CYx71s;{!
z+7O&YQkWJ-75mGvOT|F!f5o3%Px)e}p`4DBI$>OXAU#1wAMxSo5F^7%j8wYp0EMme
z7fU&sO)apfIc=Ss^qdy^R)3{V)U<`8$1QVa$6TR%xu7z*3W;&Zf|~%6XNPTtEr}w?
z-@T1i$MaJ9wh5P=6nnf}jhi?pI^J1*yR>5A_%@vyT7gtZ-GvZ*oaa2b?yJ+A;&C*_
zI;tB94lFvTx3;mC@OtvhwfTB3ekxg3%F1PEs!MtLjrl<d@E3`jU!W0lx6&{=vMcc{
zua2x*5Da;hpocY%(sEQnLeJxExCB}mwde5e*UMFutb9ta<5tibyNF;xJ*%Lc%t({0
z)?}9*qrO30y-zJGynqHm<XTfEfkc0wQW-wT%Z=gjkRB)0id4$Yo|>p!8a%OHK#|RT
zxIYvgSK;3MveJ1JLM-bhJF;mC{Oepp@FQ~a@kA|i=5b$H-wxWgnn8wIV6_Ou)XfZ{
z`!E%HwNJ!Vb!M#S4psFgyO<{*K(H@486z-G)OORqWDQaQ{$#N=6@lH!*7MRW_1mw6
z=E&1=Uuh>pooddjMriYGC~n0r2mz5xh;tme%$fCuAEl9?$iDSmp+tCDgUYZ9IAB_t
zNb#zd>j~_HQxi~-J#XB(6)~z(&&1Oc>k_J&2Wj$N`v8mf9&>G@q{6j8!5I@)KROdU
z?=3i*k#B6cVSOVjkzVB&>=W&8>3VMS7}*#GQw0vJBbtv;bibu+o^5~NnmWBbv>rwF
zYeEU2kggZqI^@!HEc8Tcr1w6La5B5`o|+Q+6H@|P_1b66de@wXnutJD;p7~3m}zO&
zZCNS*9_g=2kI@HWfs<nmp$yF=D>qR<<nFvtkdjZ3)$r?Nx?}VvZ&>WWvmNTpFmD%v
zrGF@#imJljT*0iH=%FhrC|K92k7*PN!#HXFU3aY%^O}s!0|{CoMfZpF5%k2cMWXKM
zk_eB$q^gY#uK<!FTi&51H5Z4CuEOgJ6L;I8#LY@7CLfKO&ue&U-qSF+$|!+SyrjIk
zHU9XRl?w@Yl|Ys>b$eF<G3Lcs3Eq?EB3<=27D(a|4{w!GC}@lL5a`3skVNN1ZNH|2
z)>pzP&@|=j9e&2jSGZxi09(bf9i8QzL~~tz4Hfn4Ig!W9|A9&ZN%uIpYotx5x?X^a
z7p9qpMLQ!jA7#>g1z2}-5vKc<4-@V*{}o^C>UI3Wu9ktsf$v)vNB4Zt3(1&_0@-H&
zM4W~T1kO6u-w0Mx@Co{ei<szNx${yU_MGW+nA?-QX0WPFLKb8eJYp&_y=z8#&B^iE
zz2%^ALx!VnB4+ZQ53J!s2u9zqQ}h&78bsxXyPav0Pyt%d0VchRsn6_7xh7k@$L{U4
zg6KQJ@=&s>S)t(aI8K;P9M=Lb4=A|^w0N)FB;RT#)Q=@g>QN#-(|e^mc8(i8EAq<A
zW^|rT>zfu&$X&ig*BZ?yapzUWYa?&?PKK<7b}`g?h8mJlCxB^4NEx=8{E=Uldy@8E
z8+svApPb{wumrbnQ|$iw&{U`Fa0zJTwW#{J+3#1A$7fFYQNU7?1<qv$Vi6J=hK(vw
z8}S6nMEvT^=P=llUMZEYC<=sMD#awXW$a?W;?uFS^@AiphKhfFKJi&a=imFfb~l!*
zR?~<7Xk<VzcCtk(Ej@j|zx>05eETfMOTh5Hf#9ZN-xR!BGTJ^P#Zx5Fu2o`%Nsoqx
z=7JQi_byg`@%;FXWek0BY3ufhJXN#p*JDd5j^BUJ|48XK2d_<b0pSzfg0b_u=6Wf3
zufC1}Pg#W`d~L+R?pdk7N+Vg2<(HiX{T$F{rI+75Qlzob$OM3(&JVR{q_`9KOS{@Y
z4n-CuT-xRiI<FPYMeGzLGa<?dxDT~%*Ym>QHrZ2!Q2GBr4Ws<~^6=EgE1Ojx5^dR%
z6}eLUMMK)A<hhJ!tFRBnr&h6F&vW4P(liP$qE^YO;--ysl{C;+>9IDfe7lL-cZWb4
z-q^5tWce3Q=~miOksWB?kg!V+OEJQepY8W@?T@R=S{}CJc3Hc?TfbhZ&EDaZZR_Vh
zSPO!)!xxxyj(zI7Z|oFF1fA@g1LQn7@H8fDV4pHZR`wBaA;!>DM`$V&$fr=Guadmr
zDVfQ@-A;neV<>S3qIftY#H~I9X`J~5q=lRj<R$NJO1}O49Ip!r2$kei4q}qUPW*j6
zTUt#4i4*M~)*7vbuE}bR+xbPZ(y&{8kE(*#EY})+Go?|QEP>*fKT)g{JRAAr;ybR`
zCD=&LE}l=#4Dc-9Fljnxy0M0}_4=VVO+OgkAhRx>FBsh+lsEh+>vF95G3z^<Fn7~Y
zt@Og<j=Xx_@BNI)ZvIrORZNK+Ij$%YT*J;*j394qakKTF?DNI@8R&KRweYWP<UOQq
z7l3Od+a!n5>P&Zze!20tjmMD1Zh|e7BkopP%eBt$59DnD=Ub_azdz+FN@cqY6f|Ro
z@>#2KIy?|cDdhX&;s_E|4p-uh-`2oukwD!^6HkJuNx64{Qygrmln*cJ7hAmZ%w^2P
z<Rg~!O<t#n(^OY_{7lrzF$+DY_7r%V$G5VK!tKqaR$I6%4@JDZsU}_K4DU6psV&{`
zA$6yf*=a9Uu9z7WL_azILU;(Xr(r5jCr8D7se^8=;K}HSJ8_JbW*~7i;IsKyr|`r~
zh=tf+bL#~K-j1ONmbb^v@W0e4&rK#GE-pP088_VhpkSk=$w_FbQs2*O(y8K%&o;ao
zFNwANy}uXLWB}9TKQVo1CHZPSi~6-Uq{!P_Evro-ej4z@RCZRe?Q848cQQ*PxWYGK
zNu|Sa?9qES{AYf1mOaj9K4KsaHfcsMf6sC()L;^7Q#HCTRxj_hq^2YihN=68pmK1m
ze+&FAwfeIvZ#&r0%m5|zR^7JgNaseg!j530b352&OKw<DY%OGL>pgUW4gZYe!wV_c
z>@R<kb8BZTij$9J4M7<+^v{nPds*K3Cl5*J*Xk|NA!lm~V;{L^-?v(Kkbkm*{5knN
z+vf3aCoW~5D=W-ldC$A83!BbCMA=^`7_o#3ZSCS2*t8f`s*b0*>vW*5A6HiyrzNXm
zZV>0z-GvZOg}c}*Qv(CgI`PObYad)98!VD2JZ}jy69!0g3OXb$_U#8#HB$u{V;ZqD
zB6Qg9!1G$4tj8l0b1=r4l-p#OoFHkMAZY5G_9=6v7uxd=l%~1rR##TA+)kaQ29ybV
z5hMw}+Nja(7sZDUTp|a3O%?IkADc6Qk>s6W21bDehpQWarbxNE9u!5OfrnXM0HTwa
z(s=+r1m5LC;BfnJH<gRZF#?f>(xc&nGmI@ocWk3ve{hd9;JjLMzjx-^Cba??CnhIr
z&nv>SZkGJ%;yT^mM~PFBkeE#c!dwQB{CmiaAFm%smT%~v7G`QDy)d2?eJzI)#rRm7
z&qndrKN!d^+-B)S+)PjgE`1HdN@L~czf!LvuaLwA&9s6$n7NyRFN}xhc`v(uu$RNC
z+T^99#RzF{eXJdwbPeRR$7;C3EZ-Li^widI(qXk5c3LTYS{`lD=P%_7{ZNQ6^H`MQ
zFt`vx4v>W%pcq@9?5>+|TEQ8hL-x9#h&U4j5b>?wnjiOLk;In@_L{MdJ3jMv`qZU{
zyB@e8k{~vpbZriesG*Pv(s-D|+|cxC)u7$*oNZEB&PyPBZZWC7=dT20;;0a8a?GQ_
z222(V&3vblX<`nX$M~E*eGn{j7@L(J&l$d6j0u@8)d%2&rjO@H4rv`?%xGMhi=iXr
zai%<+h^`3A(#$hs*s0}(^s6}AsEnvv`vJS>>fPD0)Tnz^&!dKfLFiyK9Uo6IlPm~E
zm=krL{R2(Qqz1C(oUphrGw|eQqiz^2ieO3Q@ZAfHQ9`irzavjtjKxw*@JLRE*i3l0
zm{HL|*P?dv!rJ3*3y;qrYexyx&IY1p0~pdnp)*rGhn*^0s>l2B5Ud?vh$#*8&l$*7
zR<4an+pje6pzVY1X~)!bw?s3tifZ)PN=ZXxb>%}v)g&cUuKNq)Lg@<AH`Mwn9!&Gd
zTvMFH5ny4n6b^;b5oq}?>So6<AX3j8IQDfr%~{ivjf=v|Tw!KoV$uip=J+NC7jm+F
zysf`yO-AQw>`8wW#HZb)KJQdS;%VYHyCs7(8Jkp%eMPFh#%$OM;l!3ji*G)jZ=%zN
zE^r`m1m?_AqR}l9+D5C!o7uac?}+hIXQi-S8(E2ydhYz_srYPRw+5NOJY#rZ$Ere*
z*sdf>4%b|HXzg$_Q%UHXi{4Ux%vAA70z=;Zr`Nd;JNYQy{fIQJ{%PU(?cAjD&0Kl5
zsV}e7q2A(i^1nr5A8sWHkc~F-v00|ZrlmiG-PG{A33}Vw6G5Qr*Vym-Vok0TV}py^
zt>h$IJ)6bf2Rl>U9g=@6#C0^Jl?nW)9UZ>I*&Witrj7OF>xeJ)u;2rbt|?522Q8JD
zdI&#e9PSKg-f3xi@gU4KPt<GDA9xA3y|`<gV_aKESKSyYTdqI$4dZp#N~7{6(_hR{
z(j}pTi(lT#sQ&<MtmpMTWVWc(4@&nb{Y!|}@GgkQhWk*9-#791y$9l)9VXYQR}{Yu
zx<7b>+``$}mOt^Zmz<e3Mfe#CiI>$iJ*_2DnkLQ?awCSQAl+Wp@&L5CHE8)j@d%BT
z4|vG6Th>4_msgnm4;0R~n2j1t)gLzB{E$H41}<jVjT9JX62{)CsWJE3I#X|jN*d4)
zhSE{SSld%97D|P-VsDhcKuz3?MJRMd$INBQF4iQ_xK%@N-b`}<j>F}1e?WH?_HKK$
zW^!1^6|iK=|B#^(i(kBXFo~XTPaXD{94bunTsy-bs%zG+j`286t@>ESSs1Vr^HXFZ
ze2uy>E(_QZ3V{Cn5DZG1Ga^tcu`}>c>KGiO^G;Thn2jmDby>Do<IqF7uAV(l_arEc
zM80?wjHNAU?8nF4^JLGe<@+4}m!x>B3n%kyd85u$Mil|gOVQn$Pm9~A!L7&T(3I7V
zA7=^V4*E`Z{Y5UHs#-z}L|-=Q4G3Y}4AVpK4?$3stGOq+DHMq&Llpe0u)64eqRcR}
z#-3&}mGJvgGT3*;c0)i5NDNw;*se2{Zyx6T?QiC@61-O1z%%~=e8H7GX-Z=$C)tDx
z_QV=ylWmm5AETA;%2Nhbc9OhWTEpJR1w^glR+@yyRWsFRmWA+pk2uk%AGw1g&YLj)
z$S~xG(h8oh$;b%sNI)2}&4<d!Da=;k&}Wm~a?ucxQI06B3tAcKJ&a`$p8N9^A6JY|
zyE#&gph9$Vle{&O(s#MSrl1}!T~~|LsXd8a${I$MBGTXpSiyTw73FhqGfQ_RoC7&u
zw*UcWLo04a0Sl>uS}j2AQfq7Jjw+K&{yC~Jxi%chHk%_Y7F))M_M@X@fxkB64L!3m
zmj7wZg3#+fCmB4kj3!d4b@g;UyU|-m<Xnw2il$j?ggEiu=?M+Z$nrEjP4~|vl+F*@
zJ`~!be-z5HSn$kPm)_uz1ah9sWcZugA!+OVqyVr+k$tTu8_~5eWy58K!2OCJO{628
zBHGpFb6wh=Z?F7A-@!d#BVyk&-X1%zq|S$Fcl6>NvDP(Ovkl-XI@3bqm~AhrsmBPo
z>snp6IX3<`C2tBLi6;0m+5TE>0`{W;DmA^`v832iXtd@-Gn~HVo@3q(iJL8F+*h`y
zPj`Q#m~J|=tkgyRAP|Cj!pxmBCCI&0HmU%>4eggkWM|^f8kQFl!u!YtGDw@qivku?
z1EQ|%)|E}{5GA?l3?xxtev!M1QZhKIUdH^#snp<*nJm9~Uc*Q*WtfZu&7o6sC*(DH
zvc+wl<B~`bndPW+PUeUTW>(01{15+|Y*IdOi7D;jEbpTKBjuk0z9bD<%PHv~oTApM
z-OR&!p=O9#Yzun_w4;l?1+UTauEvvlL=+LjSjsEmwrybHf!3i$JCV`u<VyRIafeyL
z9_?V6{A^OVym;!a&}?}dCH->Lv)_MBvqSdx=kSq|8~=g42Lb0$-Tl>PQ5{W&=$p=s
zf!xJ>H}aqz`NDa>p>i_`J9?Tj^|njzgSHE~Vt8(I-Y)bg{=FNbapr_)T8kJQp3V&_
zU68ZW3jDD4a&-dw?p5PpE{j6|RLFyFBZr%y`{;G?PH88<nJ6UzqoZ$`S|}G)`;Msq
zCQ8XqS0(R(W%32)GCk{fI~Kkrp}XuZ-XD!OJR`Q#DuMvAZ1{ZM9;-wp0W|M}CnL^B
zg1FF|o${<9DGFJ!OF6D3BHUbOe9d)1H0tvULQlcl%0MS-Tn&3J60R498Oz7mCGOht
z59y37wLx&)-1|Rg!;jzqyBRn`FOw~;YVT32Fsi>V>gI84$6!QT>-~Ni<aN7A6!pAz
zoMcZ{DOytFwm%C2AK_n;FH6#yPcj%sbH*1{gE<Oaj+P%PIBY<O?_~qa&%_xV4fv0e
z_arx(So1JTsSs|G7yL-Q2>iQu-z22b3RvS*fazRCN-JiG-l-o&xpfow5t50Eg%b4A
zOg7@<&xOAG-I24<{8Ul;Lzx<QG70Q7N~LuNd4BwvLV3&07JF4n<++Bw3@deVJ2#i|
zd$(GAR{xmAw9CUt^hFHnUS3&hZw*2>K^YRK`k2P)@u7d{YW%wW1^WVX%rJVg^%&mz
zMH1wwI|PE^b^OT16&Uc8;`$A%o24|i8}W4ia6b3W<ExVNS8jKT1J*`PHbZh=6#iE{
zUimkf+R|!ozQJVu@bh#ke4RdSOXz+UbT1cd^`J+&LRlPBuA_|`aqRFF%A>z<Q>lXV
zoP3Q$sGdsJe8Fj@S#I&oxdeaUX9DRA(16!YzL`=@(QF!N4Z^<8a(&UJF0MW!_(g;m
zjcwp`hW@^S?x6>twNb?IA-qLQ;2|14`NuR9qgS@cuY=OdxS%+a$70p{TeLB+hW*91
z)t@CGH%s2(b$x)VFMX9=KAcI!qiZ+gxXp}D<+W0Dg*p@u<=)ljZ}iF!@E|^{K&28f
zd)4IuTlKveI{`K(X)?jk#<YYU+<nm;`;I$BUt?|FJC-1pUn*&qAi@TG7vGqA8Qqnw
z{ZTi#p6m<Z_T&mVuj^nazOuDmHb5rsf`bO*ncKqT^<0aO^kJ)tAJvYJj50V_UW*7U
z+}5eg$N}z^(Ry$)biOFaQp~Ia10rPG57~=KT5C;S=XCN6y|v1uf^2w=UDkS7zcAM4
z^c1BJsiH)XVJh#As{*I#r)Fz59Q?9?`2;h)QYo@M6=JG)gV{0lAwj>$cbB?B^&Ao>
z-WFK@BBm{ZLa)&%o{>?fAnghv1<odVw`mMXWvtCY>3ENX&J`-yo;qXv78%><hy_8T
z3S|bd9p7j@&xva>7d75#J>fgWJcJYtdwf)2#V2es^Ut{f>svr4s$lt)x|+Jm<o|k9
z>M(<-K7uY2@9Wuk>TiyoO5h~hd#!czpr+5zhUX&Dz$HgH_p~W}%?J~`+^m|L=&COx
zz{_M^5uuq3f{JhxgA8a#3HrR9F<8kggZ?@lfyTZ{4>~YeZNTR)@I-vqjgTQ(ZW1=k
z%zX(wXs7ZK5#v!n7Ga&InHPw1+<GsAK{!WTl+|a*S3a}Kc*XVQdY%+4o9!>AmfAt~
z_M4_L#-xwejV5vy>s<Xrx;-V`SEH}aR@Hwje0GZA?zL;1(VlqyT}8|@VZS1D>7dU`
zCr3j{ZE$$`x<_zoPcmqZFMb6(y2{SWID3<7tQNEp$1`E-eNNC-v>prxtqV5Z#WqA+
zX*oxJcp88iIAM(rE|0<cv4|!6349LW<5{{UXBh=^h6tJ_HPMqx<_)Kz$^HS9%W~;$
z?H%&bK8Fv=sy=BybeBBpyumDlQH}RmdF-vz8a(<b*}&iUF6#kDkBgJ+i0i@~vEd9}
z3$AxtbP(@*g-zw}j_%$We2+!`%_}Yo)Q`=NjPHj9(UjG6CUcWxbF}Ci&yd+h$yEiI
z%2n7}&LWZe)tanTE_pm-zRUzk`_(NDwwsd`gt02+X!taFf_W9@Jc+ynP6->XigT}#
zHe%Un&HOT%mKkd1{;A#Qr8zD=z03Sq&BlRdvAmX2nY+^2P86g2lz)GULT@}|w#6Hg
zH~MGtpW_XJ$krIi2<{7>`&|QN(sV>!h$TGZP|S)9#u?E0Di6gB=T+-h>>tvT?XVOw
z*Ai+7p)P4vO4XI#cbd>iQ&0xm+tLPurMaB|^#l~Zn3$+F>g8B3$!x5zpXO8W26Ur`
z{)7KNV(O9K9Ov|G^Y8}Luf^uwR2Kv2Sv00&rkqMxmc#f$^$q4IPIE9gEH}64c=2P&
zHBLfFORyPTa@r@;?>##u{KeUfV@cq?S8pEHGa8?f9Xw^vsEKl<+hy_%gWi7PepmUD
zgyOcJ3fYlW*C!5Dd>mm6gO$yNjP|qRic6)Uy6cNiZQbJ}{sW2C;G#&veY=eyGoNc|
zt3O;k=bf297e+$x<KB9C`pUMR<fx<>iZC<J%v>@yo~%Z4uEAsl?YR73hTbsRwwQAL
z%zIx4G>uCoUJ*Qd$-UNTkIb^~Okg}%?{QLwN^#og{wWizCHPHuI-6=8;+^)ULsk;f
z3k|qp=Xev)X`IUaL#liu0G1foGfyO)-Bi60tn@VzVD1GeicHbfiaJ;Mk#}FuNs!%O
zqTH>NVbj<nQGj`J-Qs|2hg`B{=1K^5J(I~Loz8PL{dgZ>81Vuiai=X`Zs(&1mEpG*
z?-*`7k3d2$YW!A&MO(Pz9&{fe+h{K($CjwGp~;0F%F~+Cl2?x7??E4&!yHdLnu~lb
z)<54E;kh+f$MA}>*bbpH9FKIP|FG{@k!gOsjq>h4MDn<tvLP2!S~(iC?+U#@4-_2#
z%e9+mr;jEPWk)7{E<P$bjPMw#{Yd*`qrMj?vc7`hf7vNF6CL0NF7LjZtMcf%oHfN_
zc4G6NaT;FQyB0NDwwAAmVJ{VK6Ybk`jd<^c0Xb5kOi6pTx%gUgWIe4P+btp_o_rMf
zWe)GEI>+K@x6%B~iE;-QB5%wY@S_&E;ze<owth{}fTjZLA$xq@vh8VwrtBfWbBrBR
zFP=-C(m0qL6LFr;qy>}FxvH@93V)TjSz~;vx=<NBmRS|grWM+=@a^Bg_|-NcNvtko
zu>QS=<8+`e7e8uwI8=gr1kjT~qR!_ULlA8Gjj(;|m5Ju>(ct$XF?ax@k!=S`U#v>@
zFUZ&bRLv+^by~e|CF<z$D{;1D2a}RCBzI*gD25Xm%ssh9*?f3QUBj}VySBeJdx*)x
zmZ1GF4S4$rDjQZDV{J495{3<<DP=anYhQfgxJ7=q;Y6PN++8I41@`}<vZ1*{f;{u`
zAhWt_>;~N*(g1$!R&bh}wYhlef|OATC|AQ_DkfX2D#@mnjHXcr7U*>8Mj!rD0+mW~
z@2~ch0$O{Alg^pyeN(bmRH~oJy~UeJGOCAnS|qdOhftDI(z9gL&9ND*hKwQUkT^**
zjCn-*)vN@f;$LI(_)#tb4VekXgZq}w+6+aA0hHFV^{5pD=t@zJl+xodjUY|mMhy}E
z{nY0V*hizv5f8F&eU&$3+-rUIkMxvm=0Z7`OzS<<;25LnaWIWo=N}8}{J&v*0k-TK
z-3$8cA=n|F^gxnptmfK-l0}81&e_;p)ShaO;(wrS8Ip`Eg+X$C`QP_*yB4L28NH%9
z|4y}J&!dXl>%mHPh3&~!ob_av_0``=tvGwE5vUjw`V+X?;^NQxRO=a6yR|4B+7F^n
zog<?G%L0H55#=fla%{@esSu;mb27%{Eb?+M@kVmqS2z58MrI4e&Gcg+WF~OXe4rU<
z&gS^v1>09^+-c$0$i-XzQ~q;Rt+GCk5*uxvT$Jd*jbL?Q&MjmU<{C1Zc^>2XDocO(
z*p_iNO9jJd6J6$gQx(ne1neB1{Md~qhTgFJr;@&tNl1;e_&2U6Uj+shJA8AyJ*vz{
z-?MPgw+Vb*M0f*BvEoG!Fr(!NIS!ca1t0*GsDQrCA>Pw8lT-1mA&#PTm%7dIR4Vlu
zwEwuerao~`-~MA&raw~5h>mZgPA{}mqaGUc(9nI>?GB=t;s`N}xxN-?-Ic&Fv@<`j
z)|z^1+Tk9MZ7MV9MKMp`+mT)>Io8>}*J75UDYS*hUtQ5cJjlxlWA_po13Y1C&pjvF
zp6$#(*y?;jrfjvRX2Z8$UnV8OaXfHQaYE3ZnUU9GwoW;JsrSAqf7dK|saTp>fo(Yi
zCU0;sTE<-`HxHTxG*lpWG)kGE7g{V_7f~V4!Cyl@n{bJ-(06=qb56)NiO)oF8=@(-
z2AGfRul}8g1cY95Jn}EuI9D~C6cmZob9q*HYPU5;8G#GlaFS(8$4CBsk>`}PRU+DA
zzqAGO^o1SdCWb)P?%QvphtW2H@8YCJRWV7sV9@(Llr~ot889lOU=Fdc6~X9=+c5T6
zC8+)&BG{vqFX|(_N=BoEf>GeqZ&*6@1+LFySC@?$(jgB61uUky`vn)DadMz(n*`$q
z<DxdwV86uM3K+Mfqkc4s7$A7@dIDS-u7ApNA8)To)%Ucgv6D;CaQ*(-R3}I`ZF4<h
za)Xu9s*k)JZ`?06I$EYyLR`>OXw84B&s6B92=V#R;PwBQdhbB0|M&m@9LL_{oU->`
zIUKSb``8?Nm8>{M_KLC+$2?XX(!p_TMYfb=Z^uYTLN=9AuTp)!&-d@I-=F?JuE+Jf
zuKRVny}NadKkcblPZ+!Y=vIK*zhC^9jY^(txYnfOa2u0P@0H8by}x(f8I8*con3Cj
zi3h!>AY@?2YRNq{J${jmrLuD>ChV-?8<k5lf`+{DgTyKgvr6vCa^T>9&*jGgBL%Z=
zy~k|_N_sQ~X(LHffbD?RJxu4gBngKMknwsV;Ys#WA2w~JrFVjluM2X|md1~W)x)4h
z){-aoJ_SJ#-h7DF3=M*IcJa9bOi-ZYq9vj@UvKBk!bK?8=9k?>DFiw$kpVg~wFM@d
zSviY?ud9-n;J?u<#ffrje?N^hBg3Y8w({7f=qG1Ei{BNuCqs*ZnS`pEa0qtTp|q3t
z>IWiP_LOPH9}LQ}Mk9iJ*DEoxto$sWU7E8wgaq@fe>R{;kRIW^1%~2C%cWoSeUIzv
zB*WeyDi57yV+uCPMG3{4CRQxEB15U%A#t8wDYiq&rNXDvswzxy`e}0s2!i%bL5D&=
z0E5Ck!y+Avx|K_XrallylEZu9IUS4v`BK+h3r;=a>;!>V{CaI1m7v(w{N|6kw?f%)
zcIr`=9PhCoq*UK&{!tRka>@0RmE%1u-uCJZ##5>6YbZYY+X9<|Zi<C%iaAxd+4ZSE
z)-;YaU8e)h>Azr^>qI9TE5e;4BS|o_MuegyeLZdgj-<n?%n9i&bk0Fg#8oL;P;C;D
zU3xK*-b2QN{&S?rkLPT#LkReny@7F%Z+DV$htn)zP8|F18Cp)_gHWL4)xXOO8cN>|
z?3}#csQOya2Z7R%Kf>FpEh*-6E}rs{4(Dzcxi6Z&R1zcARlRbc4D8a9`=DV|)Qfi@
z<L;(O;Q$TCy$O45g8hH=qwG}?6lpD~^BC%N3}z~+*_<f!sLwXcAGAE}SRjHZNzMAx
zQ(I9e$o*0JrU-P%zzRBY2TP|OCo|7#TL)%>*C7=auirkj2w<91diH8&b)wX2?^ctF
zf)8E)VR<>fp^tSw`$Kx&<sMP^=9TT+G35d&nD2N};xVTD?icM50uix?*s!3gr11XP
z(m1FsaC0HEuW0zyUCsXkv0V_lpYEJdSvP>J6+jJ!ean&jq;xM808eI-SXQrF&w4f>
z2V8%mDy=s;OfBFU7Qw)$u!}G44zI10V*k+nJbYzWFsjaX-Zx?Rt>a-#UCOl3gpq5-
z(shKKd|9ZVL&R#_8D=RhJ99oX?4Q?-k8U9yCagR6zR{2y+-F~SDSH~2g>xJKYjxx^
zQJ#oNak+8`lFK+t`n7ho_eTb0fNg722**t$&QDx<2P~kpO3!K}e_MmgFEp2brbQaI
z7ZFNZp-<_KG2Sg{r5|X1>Cy=PtbW+<>9kzHcc>(Pe~u!UD2HT{S6^UEj3&^EJ(2d8
zGg{wGx-OUIEsyy&;?F*=lIvnl?<o>7wt4Q<I~4RKP4h&{k6Y68=HUzY2oH73VQBJg
zzJ0Fgbs;rXUhbxMT+~2>oBG{}_1#^>kX!2RjX2bLuy7)~HT`A|1}YH+yxu!*Ettlv
zyU>B$emT$$4A5^%nA)pqhNHam#vpeZ-ks5eqlnTT&-pNZw<qt&E7Aav(m6AD?9IXy
zjWoxNDptPh&USgbH1A}7*EeQG>~h~Ft{8(8p`b6il+iGkVf|lwyuqtF(ia@-dv}!s
zwMU!u-Ze~lX}{C?8+YW*AGAvN$Ikf=lDmkV!MFL?ZR~E|`<P_E6Sxd_DJ{Ay(5GC^
z`<S!2IU>UPL&2nwtNyI`OWB-YP62&O1|zrByg%54aOX3D(%gQ#tQFm##Gj-$vY#G#
zk9)T=gQ^~ID3^DN&c5QKJ+R~4{Jz@b(}RjE`@5*6zcb$`{+}+NzAw?&DX5dU^M}m%
z&#0#DwaRRpUVK~JZ#37utEDJXcTDSWqSHc<-)mjnJ}Q;5NX!DFZ#X?f!}x)(!Do^8
z63SgCuJ+tHjRpEvJ*6B=U3o|B^&kN8GXc(<8H$IUgL(KcwI%Ln;kshoF<i=y){Dg-
zqbjLpbuEOQQ>V^aqdhHD83DtPf%#J1!gqrCQ;xM})nVLdv#`P0?_Ip-*p4-&+_hD2
z+)nTXfzD$~Jx)b@o#j@0VuvcbVZ~Ns<B6)4AJ?5XaVa`U4Q(<a_tD&<Ip&PMwKO6K
zs5%$-O&YBuPZjnN&d6rL$xOB<6>*EUgbx*$fz|SnTA0@sKAyufYA0rI5G9YP5t{>H
z(gvtzO9upPp0bvemY-NIU<U$ZPdw0*#c^n-Ay78iwC~$$&V(MAbC%`!*X*<1ihY2_
zq#c*<puO}<zJywK-MSQGf9YdVXSG3uqln{xw-~5`Oa{{#@B?%GmpHVML1z0me}Z6l
zI;(C2m1#Z`{F-%MHD7}X53usI({BH!6Kzgk5UE->>f*_}zSZPn+%wt_is+3yq39Go
zaO$5hEaIy5!ct2DzOnXga!ESAtTXZYJw%@$Yxgh{pn>4cG2|ptkn46!*s<2J1cPYy
z^v!6kB;{d$<VOtIbp6c2%^=$PTjG8DAj|JTvKvO4+gMcf>~7(t5^7JhyqsJwLKPc4
z*eQDSN29vrdvw@M)Ag($El;UsDlCsTq{j+sZYYao2Yq^H+Vbu>6pV0`&&H)VvY$AG
zzpeKxH3*3(pVkTbt0Bn#v|id&cKLd)(2@)eRf;rCVB|1LQ5CgSDx2B`<LR~+FO_>5
z1ltFpv4GTZagI@$Y8K8-*2pM%BGt`3*3A?1+_G$Ly!8C8vZW0cn+ltCF;pHG%5{05
z@gTB5iq#a(TVw>!v(m*U=P6v&0lc*2`RAXfxh)(ny{ruaCc6mFtx#JpJIwA!GH^43
z4NvNuOv7I99$B}x_-phPDj`xA(|yBrL!>~TlycfXZY8bT2g9vMiP0DkNh5v0j%664
zufRf6W8LFU*lyNWPFci`jEg(O(r7hYHn$N|sX+85Fsb0oY*x;(ts*-Sflxa=PZM$Q
zu+Oc|a6E(B$*qUdD_KuxLS&w4wn~)0H&;>H>rN#{9@!z|=8u+>s2{4UM*{qR;X=Nu
zylnRyL)WkAd;R+%aM}fYu(HNr2R&}8rB_VpH2ZF^)CiCMc+`|470&j5prj;=>mk7d
zc%_Rwww`leJBC*c{~YATvTBaKc`CGWr^f+35k_U_`dtKOt$Tu}7!v>ci#N{JbICl6
zDuh;|j=-(XpavP?vzl1P_U59d3uMp*$W%yj=yKg5j;0L|TOnr>cBbZT{zNr1oYAu<
zfTtA{03Own8gj?Jl8uo{^{f5Ebf9JW*ceab4y^AuQ=fx;=rXp)JubzAipoUzJ6udX
z^uqBW1RANgJJyDvn+Rqmp`^W)#qSf;uxz2K>DP;S*94@L0V>>j+uo#>Ayw7C!Y}gI
z_^zD7c^*pl*AMkp$x59~fW2qG{eHgZ4qe)uyY@<=DF|z!eb710su3NzUDuLnBCszx
z+Q|OUWk2N=BPqwrotg>#`AKfM!XKp0m0V<;SM0<>dHjE)=arFkf!X%=o<K)-+steg
z8Cry!@;<$WH34FpV-4#(S=z))<){AQ^10p0HY=Tv)h2-DZZ&+wA63l8&vRh1rkv@m
zXcu^M(U$nDT6>9oB@Ng<Whxg*d>Rh73d@_+@BIAsH?}J}Lxl4}83O0w*QL3-kf!<E
zLy6`RtJ&#|s1%#jv!WsQW}1*=-U(6iZzbLkGgfD3xZwh3lOMNhygad<#EgZt^BAt_
zWwlxGM9ek|K8p&YMesNw-OEn#(Z#OK!s3Q-J3h6B?>9s*c2mTS??(w))c?%YJ#Y#h
zUVq7Mq73*y3?VQ<2;j%8fXz}gk8){s6U`)?zQHX&3=4tXqUr#k3CP%bi9vpT7z6^9
z+rno)Bd>WBNGnaN_<f6(gRpZ&J#8kFA{lG{H?wk(bgVUv5A#~a8LYG69FFfwi7(78
z4IiSFRlexcXZ(WI8*0>r^%dGf1g0O&V7R3erF44LM&ppRexi$QDS{U@Okb^qKG$!~
z2_25qFpWal$D!W~EE$>L2vy#NB$AnKFW%B0`z{>cd}HDlOphZ*xMUWIjb_2OFuT-b
zvojk>Pw5B7DvPjnyJYc)qC7*%DdZ$}*T1JFb)Wv;`9T23eYX&81a7I}*LdkX$Byi3
zZv8z>6;$P&bNvDg>SdXiq?5h?Nz(3*7bH12eXZPBbC5dYCe9@5Qx}rpPzX^j;Ff)E
z8UlrCKnEpR3AX7xe&AH=Yf1Fil3y)=Mx;ndOsa?yK=01=70FDUy+Sa0(4kXAno4kl
zWqi&IJ41A9meY5<h}59k^z8mW5XUF3;77eG_G+r9{-YuHIsJ6TVSS(b72MYos_g$R
z-c35h>>MY_yzI)N0)dmzDbwL)l~=7crKNF<bMOQ8XZ>&M!)S>Ln>KHt%tQ6^8cUA~
z73%VZQSO6zIO{RN(gEs5iK$;iGdT*V_ad@5@CO08<4Xu%E-7`-7)E!mD;3yf?L^x-
z60Fg4ApdMV3MCthDPLgv>E-4PalS5-A9|}uE6e+?srq+2pCeoA<i70=LCw(NFqGHh
zI9x)2I9l+@li)<pdXTO#(1#ZMS}J3pyK%bAS@;sZY3-tAWwuZ`e{q>Ccyn8RgOnc|
zhLP5g;Sp{{j3_6v6<Elj^qmJ_5V%CMAm=O=Gv>#ou8l#ikWzW3RZT`L;3TSZ8>PGb
zRKgOf`~K`Vlw=2qeJie_A7msjJ=}>yImEfN4FH<ktxQ>ZJB|`d*1=4@3Qd`XUJuV-
zI;DKidAqkf0h4|(YI=xmQ`<Qm3dEXvQvb6Hz9i_`CT^<{?yM2E5*5|Stlt@MI4%;@
zBbM?36tr|2NB;A%!5L*|Lbr)?_-s=6XJROBaf~c}VToR%RujS5FNvg<Dit3~(@R~S
z%1+6%47av4cJWj{p?gVsR}s_f(Q{z!d&4vYLy1xkHOJ|X*q!5$Ge)Qha}k|blOxCH
zoLZ$KR|I?7p@)6;dk0Ae4jDo=XRL8cfp>D_Jj?=-@xHaCF%r!nz1j|4>Ei<JcF@TO
zw*^jH3Pm+wGTflMO9UwpsN1$mFEy><p{B1cck{-#<C!fJ8ck>HB0h-3_t!%+zIl0P
zMse~EZU1suBBtGc*3LPIvZJ}R*89MyS5hlbMZe{sX#(F&kWv@e0Dt29*U|rZ`WDX>
zrM-|()}8k{%@JagDt)Dsc6Vcd0DHdG=ZL*SoxhPipMORBJM3vbdRrUheB-lGZwK($
zjLCegplyo(clEv5t99Pkv;R*SakR!E=}r9O_lTAtPhTCu-}p+U2j8%zZQShQpX(Xo
zyMYSXu4PH9tj;&EsNy@ZaAWKpx$X2iB%~#l3A)s|exu)g#*u~$WZml@`)DcQ%!137
zW`Fug2F=d$(;gfdmBs&1JP|E9C+G1!Y8pq}nNMPxTgs`%drHsJ+4if?GmYNeYbDHM
z(x%y!yqQ^=5w63>lJduzb??FqZF}<sw=W3h#0hUT5p0gX1MMWoLJg(TH3vm_v?e|a
zfGC(^Eiw0Qd^td7s*EvsJEVqwA0v)?zjh+79iW9=G@m!?d1+CB^!Dy;OvAtZg`3l^
zeSuv&CCq_xyZYU;Q%IbV1n)2d@Q<3TLQc)yq)OBUNs%mxT#E~nj;XO%d#jA{TkPys
zU5&GALdC}|*W<ytA)5)9<|!~(1fq4IptSR$i($FS;9S5@nGyI8uEc8XBvmZgeKR>h
zFQoxDqV<lo(ByY-2gYmIJ?2cL$4zf@l7!y=S4+oaBgf37t8BtPq@{#^F7<I3?$k0=
z4}Fi(^=xX!4I326(pB$VG#Wo*KvA29E?3d)ie3;Ewh?U8f?|n(F5Zup__2Q4stG+?
zU_FXe46#rFj!{B^yb>1_{I+gv8>Yt@LWQDcwWTP01-O`GLviMGLhfX+yIiWOJ?DX~
zT;lsQ9UDg>HwQ!EG3ZG*Nnsd~HAhibqnT4ki)`O~-422c3qgkv3zDqsWZy*g0#yQt
z!wQ5WCvj>^gcWBB{G9fS30|9A*jC*+!~CWtFN(Rks)^|gl3r~@`}_C{h!)`MAbgZT
zqV?=&Q=zv~YF*`O8VpiSa~`xeXg@-}E)s0Yw&6VzNqF~rF_i9oVl|##Zj>K-4}^bc
zN-cySFJkrX4WkM?7E=rc-EvrzT>?h*Q*6Do(Onhi%5F)VjQHZL9TO4n6KD^;6uZGF
z{<(-JUaV**FQ>>duwZ-|t%Uo8b|g3fJ*ZUk|7vFU#IqW|#PM!{`^p6&pz-HM#UqM}
zdz<)5g&OPRM&59fa!1iLHLGUS>!kuIHpSEUXyJM#s-WtAT~vg1JDJi@b~F*b+%NG?
zUD#ixB4U8&2`_Lwq$NcRp!9PyD_Cyezz<gUA-FB{MYh^Fyu~sO%*vVlL^C!j@2U~e
zBI2dx^we5nZ;x#0UT)x7U;7cv@E%feO1dWL)#2V3^@V@BiMB9+<%}(u0@pLqo-=wj
zOsn$k2NJywxtmlztpB=qirxL%5!(`07vFxtXAwXf>22!HxEt6Z9^eIM(Y^qVc$F3k
z(4z7?RJ=d7lDIpyxtDIlrhRBVG?J-$@n3AHIQMC!?nmT!laI3;TSM*AQ>G`07iFQP
zDH9266dlcf)T&*x#IkhKXV06;%RW9Lf4KJWW&0P-Yk6LO-`WR9Xz_0fC{D0d9Cqu}
zSZ3a-6%UKsOAgX&Ukx}B^d1>~zU_A3KU_%T&g|lH-qYpk&=+#;p9>M@YRBph$!oGw
zy}p`J)3zSzw|wvYunWCxID5&KZ$-`<8=;v1k%)4kQr1TE<B0LQ&eflT9CS?Ir`svg
z;B4CBtU)chZ)=0UB6&Az*({7h_SebN!+ngG039Ul>UD?9gYa2QyRHBaO1sF}c<TNF
z*z5m5Zt&q-f-u2zBsNC{?*$zBk(07)m-Eilf#M7Vxgm<WKkBPOVkDqoxZ9#-YUFNG
zdNraP2%XjG9+}on3)%&8ka32Oue=$)Ofhc|-Wyh99G=+xatw=gl=&%kydI(4K_@1s
zyZ$NNOm0!3+d*3CUC1`T<ZxOLpL_nqk$e{9^zS>(w~EaxM|aDkN~t5yzn4?$uM3!l
zG=Bx1=WMV2a(J9J?&>Tg!XuD$9B11bm(ghOQQvlSylKbR?v*Bff>-Tc99e@8+NpFm
z;XR|2T#ol|b}H`*(oRFP9?O-{%UabZN47VsIH*)fr@Jo#)3kCYq^DTQ?_<kK#~BO0
zvi)aIw*b$hF`dy<61!cZq6SlluM50vnh@;PO?*i2E%#DnISFTJ((oT+=-eE2MM3;>
zi%8zT<9|hq!_s4fWR@CQx=?(wS)^A*TFHXUt1t=Bj-AEs-f6kD{mAE8m#jg$$#A5K
zVTgnBu?+KZ>3aCUzNBdLj$*3=rB{t<F|h^NAXuTQP9Z(Gdg@}4?i!P23kxo)=Z$BU
zc4a*m;A0ZP=A}wu1H?dj^T`UvFbGnE-cu-FfyNR;mD)4H?4J}Srz3LXeTU`pXAVd+
zgr_|zjSIR!J39HB%+2X}XISD1ZWanf42I|VGoqPoX;!dahUxBwOy7%?>x0)8Dwh2B
z{T`Keb}muIEt%+3#JxZaRBE!z1gHC0T$9}On1WA3jM6Gu{gleL$uEUFqyKSWEx5%(
z@1_qY_WxdV3Za29F%9K4#$vk>?^W{>nW-odW^Cg}iBeAI7W4znfZ)x7eQtB{@49`J
z3)?e#3`O`f+zG3TPby=WHe-P8Xw_d=oso9RIC&htN|c)x<dGUxBZ<RNRI@cwM99hG
zDSzI7-+&#{`eMlxFdtxa!1N;Xwc~&xtNvju%<XCWTq0=k$Hgp$Acayx%wjy{{10B=
zNpklobfT#8$VOFdHA#Eb_-4~le7B+~Nxfq+A_*P{H_P#3jR4FkcrR{GPK%qsu$nqC
z%MUy#yO>G)70JRz|6qW|)Km2nhuQ97=%BhhRb*UTY!t5USh|<p`@hbk$yU}qYJE4<
zTg{~*uAa}0R_d3vGB^Rqk$aODto$b;L(oI&GccK!jS;*MTue^R>me2BqWlqqOR_|T
z4kPbj;YZFB_@bxcZYoOp!Umn+2)m?wWHlzT&X>`^oM~uo$1WdWsyCPkLgy7_5RdRB
zj?r;M-<coz{S<@6iSz;1iCK$5bIa+BjV+`$d9~>{L?;|qs<5-f^$12Qtph}XLdQIq
zOlyHN1EH*te?^81e+8+yB$BYnxKA;*y4PBpPqzMDrrB8Ci~Cd&++hXKC=FGDMiGMn
zxI%*yY@<j4TA$4?4^S5{VrcS{acuc`?^8+1`iJqw82~G2zawu$xR>dwJ(qv0Jxe}y
zk}W1Lh(Cf-;D<|#;TD7V)zrS$d!30g(r<9wC)=0hTN=Q*%nZwsBV}R^bGUQ+^0t0T
zlG$$31`cKPjPLMT@7lN~nyW0jW$j%NOHV)5osUOP09X6*?ck1fsY~ykC`BUfW0w-g
zGnd@*-3}$_Wv$1RwW5=p@6+T_p4NQ@nTJ;2(rP2Z=6|8?cfraJC4=oezO<4LUPIa+
zrEHk`i2P@4de9t^t^J<gzTG3|zZPeU*>LZ|s2j-*3Iuw6N$B%CVV%mgqj4E6p&K=0
zWW>aoJ8o{t{|3);TbS>BveC7d^}V$PT;mVkn2zN}JE;C>!DHqX%6?{AH$L)*-ITo4
zq#t|zY_>3m<*J&Z-JlJ3N96pc@j1ynYUlE+d494*PgflNV{59h@6<X!uyifHP(jRq
zem0l<f-&Y>x3f(PW&53ztNo@G6g$$ti;Di<RyOHLTD-KefGBcq{>3!CUaC^^4pR~F
zI+UjUcc7CErxRdSU%DK(NmkoR6f6(5x6k<v3a4m3_D|hLFa|JYEt(s}kRaUUwNNld
ze4X0D^x%k`DD%pP78GheEtI)XV6_n=Qb$yB*tn90&FGRoX8QgrqImWrhEO$v3xCMl
zAV;EV&1(taIp6AIun6!Cy6iJbRx?_bS4{mI&`|hgvSn!8?cGw7*qr!H{QcNf?>`DO
z8cJ(_WWwJ$vbnjPW+!nN(EOo$_<N*1vokqvWY1!U&B{<_xu9qYd$YE{IiD3*e5u@g
z!t)f772!GM-b?<VV#Xf39+0niQ}+4q>YSnFpz*(#F#dFNLrq%<#eWw~Ji<d+1B|vO
zu7bJxroZ!-Q_1Jk4!H2X4Uz%Z6Ncmm9qhb5sz2(u1St%p<CenuoUqh}l3Mre(IIi+
zNcQSePR#s52bbK64Bt44xp?D?7;;RPn&dU=LJ;!$1jrz%M&Qb~oGB_KA?ll&sX2*D
zQ7=6e6}1Z?o|!hgCvvZHnrL95as#|rt}^eSYvW)5aih#d95C73=()uiUF$m8ejJKL
zx;ZdPS;B`YxQ$Mic;uy$*tV)XTx;M=aG%)vlau7E?E+**IhE?I_D^ydz^#VlV=Y6C
z<6kG}4<<ApOb)DE1s{BywwPw;N9;dVMKpVYe!>5nBcidwXMbL0T<6~+W(|Id9MHb3
z@~PYvesId9+`LOuGNNT@KZK~`vtt=-M0q&8OMXc2=Uo*Ll_ih-O>U?y-mpQ3jz~`v
z9Q;1%sL<8LKb;<{_xq@~Q-R!=lBVtmT<p$?@m)?X0|<*S#=5cEkUdQsFMPzhGo3^m
z9xT#~@Di)r9EFK*zW~ZOx9diM_Ql{T+xWEOgAT_0zO0IHONa;-$I-$D&uPbo0lsD)
zp>YB*SS0+jyGfD9zRH;Fcjp{9FUSdzcKZIs8}F%9qgsZaODW<c2{Swr`W9!e#xiX|
z8{x%ERvXx=qdyHF7DW4}&&_+ug@nnpc4pGUou{ttrrKoL^@UpdfoumZNEB=q!b#Lp
zeqbp$GF9*gan(QFqO^u>PRPKsf&+Ri6lylCz=3!l(qM9e%%L5-r@ZSm`&DT=I%=Es
zIk`T4Thl&#?@$apF`}T|k)F86-{Hr+jcmWREDfIw3%FzG1q3V%)f|37K)}bJ%9(6Z
z{CUA@s*1nMyDv9nO|&rai^VC3RI(J4c(oUbXfSO9C)#ku(;EV0GEblG;E)ij<(8I@
zeMvI`Dp{h5Jz57f1AP9hj(9hrw+)QLIa1YX3|)S^_?<@$21@<JmUCncfdcP`G+G>#
zF`%}eWegDBPPz3+F3a0@*7-cL);hV`8y)JF&&k|=yqy>z1x@P`?|sM+6FyaV+x_`i
ztz>Oay(``j6Zvusw5m>DbyDEGl)zN{j>%}GXN48@p$^F9=*lv9*vbBDr_K1Z(b(sO
zXE0#?(C{-l^Dgu5m$6?CsV&FOQ|@lcQ;$B0<*L=ui9nReIKc;niPk~`iRz0ucMI0P
z>n(5eEeD))T^2<)Eu9{AOgrZeJqdU>%=6E)a={#9!SRek&GsL*^36um+Ea(!{b%sb
z=RX~@(AfYyG^}EK4X=L~pr@a-Xp^gp%={gUyZ_*G(55HLcmCu)rW*qTu>9ZVKUZZc
znuFc1?atLE+!}ShOLG1o$vW6z9B)5>JaBtJ`j>dCZ1-wDv@YG}XWc5sqOi&_-&5U=
zI^p91G+hjMO|pl_=`CZRg^^N|`V~F>Q&}tPnCGkBu@sCHqaa652V;(C-?ZMg73a5<
z1=osopUK70YyAyi5wqnpyT4@YQM_5Ar}Wc0+&)-Lkw%LO?v`oU8PK@0Wtn<@Ec5(z
zmwmFBP4d+BZ-g|zk?K%}lrD-Eg8&tTlP%6t1Oa*~sU-=KadMxp<ST>aJbbTSzQ@&o
zoXU^pupVO$&==Re@w=`2tIalMtAxq+@$l^~jRLN#S_EjV^32=8Lv=S%!VOjaqb~h4
zmITjwNS`%FdAFkQ5(?Thy=b&H<pI}Gyw#n%(2{Oi9`b%JoHu4<Jn;@nKyyi}FiZR9
zmHepihh{E2QqX?RBixfmh9qcUFI>S?JAU8HI1C0Lj46V5UjSLR-WO-{RpPycI|~aI
zv4t{*jE{F8{qrl2V_wRBc=yI5<rw*9Gp*FeF|!(4#z!ue<Kjsg6`6MsH-Ch#54_FS
zEyw2jbqZkye-vdMd)}&$5wTHy<fw^1tUG~J;mS+wMHT@=Rr2u9gpHmSjpJay8*IG4
z@#DQZ{TDSh`w;yo7^iCC)v{cAAztoNATe%nJ<Ux*(6j0R=Yp?q(B@FdF{X^K8IP+n
z4B_M-DURGxd{#%G{gHP2!R0MfLOqzf{vgXf-D@05OYXXxTO?}#-)UL*G(sl#d@_#w
z>v4xd_lWEgSEg1hL7sx`z={pLoa`!XO5p|yEtms(5PL;+H0yXJBN1$0O~n%sMN}N)
zaxqR<JwQ=Y;#elh&vsRqjS&Ny%bHZxdx-Jt6MR<69C^%eyr+0F#b|hR<BdnZ?0m98
zbTsHVe4#$pqVO?o&w-WGp>q{~tL+eQEWU;$y?{c~-JzFFS7jc|`~L@$CQb;NxW_R1
z&mRV6cZ~C2RxNUQ+^#NHu>Fc=yC=D0YdTQM)sb?w_%4Y1EXR1;MI_Eoujf?C*{+L5
zIHj{>cSU=jB5!jp(vTjPyi#{>Ti|X`HBR5izN*}*cur=Z7E+^gEIo?ISN&pwNHg>Q
z-}v}V9+9>C$i-=|2vIT~luW{O72ltmXHQ?g)nStscAtK_6k+%hk8|7|HWBWNUymY<
zj#nl|*UWxK4_vQB<lmh9iZV<`I%&5*ULjM^Xl~jaBmz<kLz=fwYHoF@+9XS*CmehE
z`3Qc;kLtwWiDsfSS!h%Yf8ju47LH6x)B0G)R#}<d<;|bgrxEeX@UDXE6EEmtfiAKZ
zj0A%gOy|H@xVfHZa{D+_dk{hIq-G84$%tkH<n&mObVZFsk&nzCr!6u^LpO73(QKR0
z^08b$aAp*2hbPgVXO5fyWH+^p?yw@R@s$0yP&ea>?@x`M8-XEh24Eh{G7vjE{ES0E
zoEl87rq$?NOth>2`drjvH<g@lUK*BOpSv1Uj1~mwLY_DS*&*^>zJrV$cCUUvry1T@
z@AK8d;<PZA>n!w`zBjzfzOpq+#i~$))N;IEn%c<j4-)Fl4P{hubVSjHQ+qc70wqg=
zQ|$t!=eVL%c9nC()Sm0*jG&c*wPe(Tl^m_78rE%5+XdtvQ3Wo<7rjqWtvJ-dAGOa5
zk;3b4Bw^b;mbhba?+nV&M3@l>re#r;b#PmPtFEyk_;}HaJ7e`F8w8&<XTsCr=sPN&
zXuu+_z9{>PFfXiDle5Z!AcoGxy_buPc(0_?GbWLXS;ikSHrxlBe09m^oi9LwH}M*C
zAf*$NPk4|U**A${;J^~F7x(W&9Id`xo?PY1aSHtt-=sfoe@3<H8g##UPhpRyCT#in
zy(kGo_1hR@n|nc2ZB$rgYb`2Mmmc|h-Q<gRVvrXAkA8Y(8p7>TGHsZ=W4`1P^X`n;
z7>A95O`tPuze>O9R&DAx{W_WT3v0Y>PrG3-;gFFO8b|aH*G_BNXSlxRa<=(zcgZ~a
z^uD95gRx??_x0%fjrFq98TU2SfBP48>nCnMYXj(iN>A*7l$INxT}r$`fZu#BgY#NN
zY!`nH2)5munP>UsORcj3l(d?R+y(D*2cx|wT9`9Cc-5eav~`#yCz}NVY=#!tU;p6|
z!2gB9yhSIA%>tZ=QdIv5@IQvqAa%`IPUG5vAngvGdU{$YZ7NGy1(i$3y=TvFNm7{n
z^B$3bC(>{cysA}eA`hjK7&h$^{5UtHCIK;|b;7=Fjxo?*GnC)CUEA!+sItTAkc1nj
z(D}Hz6rGV*FpU}k@F+Kt59##{yK@9R;L`yI(PUSHZqY=oIBVkOX{PC(e$ev-A~9OE
z)GnptDEJ?bg{}D~zZJj73Kb<^ha;uVn;F+*-Dd*y-@N)Ze32ooaWFg(`hId&|KhpP
z`RCr&b3(}X-W@jQfXI+~9!$=I&EQ`F+<zXIOKJ)ZN4$@Elr34BEePsxTsZ5}it!A<
z?*34|8DA41{`GH2%}oL$4e#Q^v(>i1V)IdKHV-=lmF6eot6*rfSH8rG?&qnLRd2rA
zN=>Alyc3Ehmb!6MGrSA66wy+f1{twNA2|qSJwUKewl-Tf>WU_M5o*EaG*r{_t01^8
z2$<<LEBDn*`J_v4%V0a}Km;lu>P9nGiCRROI4$%-bt{j3liMP>Bz1)bqlau@<MQ0Y
zENmx23dTvo(f4GjAo3rn%otU<f}JjjZ?C}=q~2n<{M=tHB$MTshvlW9sdsETfNe^{
zS%Os6yk6wn^=f?6x288YjRczgStZv;rFBk!W8-gz8+A+6S_jzatuH3~;EZC#D~@Kl
z5HonMs4LI(MdSQEi4LXSk~hQCH*OX3g~}t#yyQPUf>$awZbsI6sRe}b*h`uRy``zK
zk{uKy;%1hMl_W}cB!bku;6^<=6&-%NU;4UHKw{W&m@%W7c0<uoL~|+nQR0Y<N={Vz
z)LlS54#t{rwdpjQW_Bu>4hrvwmfPe$S+2=~`r>{?+MtEEnd+pq)17`2=s!&UK6h|_
zZ8ew#Mg7Xq!c6h?kc5Y-BMctEa&JuhCyeJ{#2X!8E19g#BANqaJm>Qb84Z2_Q5+QN
z$UUK(-(@+Bs>jFB1Y2Oh3Dj7I&b(&OO{izdqhU@Gs?Tp&Q6Kmt6_{6p*3s5rr`Hl?
zljIM^j;f?opBNX$T|E}@PEC{09|>ZMyw`vEU$VWua<uwvrr&0k#%D=oSCL|Dmx(Q;
zADU;OD9#%u5{S^pH4sjW1R%7E*228JB-$`=-wF0(rKx=f&#@l(NitK3Kbzlcg}JXq
z1&B~97*{Or3K`M3_$5^iUUDhBY{h?*%AeJ(>YIepyP?6MCY_myX^Q?N5epP!<IX$o
z;cnReTpSl_s>ib9sp?3G`tnP`y7dLx;jl5*hMT<L=}2wfZ|82kvtdL7HPFu{9})@Y
zc-LRZEG9KHk-!M3;>Pm>B#6QH*jv>Qsqd8}GqRB<C<EEWK|eMge4|n*a^C(dL6V1E
zNw$`@T695nS6So7t8UdT<*TYXj*3fiQBg*>wQ(f@smi;DRj@Jz-GZZHwh|!&)z2vs
zp;4E$(x<8;CM4>Xr=O1|fAEM%Kijx79EG<^Z8}kT^=>*Z?xIEg^5k@AzsBqYwZDZ`
zeL_nB?@L>(wCGo6-!PZ=)8lk0!?$My`eXDKk1d8iTF4oulw{#Ermry+`RYiunB6Li
z7*vRHA?N>0Z~FU=qhO=f^^N@7=4#682{Q@EeVLl-ry9(io&yyA@&bGOr3Am`kPms_
z1(;Y9Q^X*2#Eg;bW$Y8cuGc#BN;IJ74Sn$%-Swvu8+db;p+%4jm~Uwiob*QIdoxI5
z5nj2E@j+nurC1O{_HVNKUnY&nh+jUOe+vO3ZAvA$sP)O5J;eOa9JOX#3QwXd;=>ss
zoSX!sk$`D3+U3A!q2R#y4>a9PJ*q{I1Lr_h6@L=_vzjGb?vj%-j{U7HCBamZM@aB?
zjTz&tzynFHARlH$W|kWB{2a;@;k@AhryZ`0MC~HTTVa7-)Fn~T54?sYcL?-tcVT9X
z(p&ikM?R?vJaX?^im%da*KuD#7zN+Z>stTNz9d_NyY?mO^tRm6FXlIYF>eAT=1Id9
zSB-QJAH{nAk-2MIhIWgK@P1iNod;>pQEg&K9@x2*PzK6On>4n~*GoZrTUpe}NGu_y
zC2G3K`D1~d1Xs{rhj~fGkZg#df_MMjFgo7NaGSk*E_6QoZkz*`#JqBQRz0B4h#hOH
z$kILb2Qy^X9_V$_)Cd&}tvyy2gwi5=l*4Fe`>%^VVbj8RbRSr-kAlsRk0omu%xR<-
z=M5=KoyV<>DWedyN!vW5YM{$nG-%C222Xu+l9LFBEHn$UiZG`=p$5!icQ^2@N)>Uo
zdfwx#w?I|?*C@xo<mw*Tp{`Jz1+uUD9u&no!WfB~a*_a!BWm{v6pN&SV)b3BF$@M)
z_B8~eU2(%H{SwhgSGYJ~V7mfCllsQ!U>cJlyONf{^$f5)_Avwryz;<gA8;V%UkpH$
z844<>{hwY86ibMP+q#Z8EgVKRv)$ua5xlBuW_<xjQcAP>IJ?lWUU}l?BD>PqdO5jO
zUY%W+#sTpwG!M`{^P|hj=j}E@9Bw8_<-RTO-6(i{)s3<U+15nWx{W!0&{pqL&X@UU
zDYvy_A!C<DQPo44;)vr91;X$g)p|Y(BPFMH;ZmN8qQ!>jrd7@cmbiF$+7o3)y4s)>
zUGrWc5j*~JGMj@?@gL04x;(Wynhlj@U&2fa9dZf1>K*D9nN@4h^OV}0`-mh7Pq!mC
zx9gj)P_&P07>h0C2JF1D8XYcQ{G#%Gd(F=FfV*<Qc5sejD-Y|T(9B6N2i1_wu4UPZ
z!cm5BI$+2^>)%ZPVZX%RRuGmnwkfbn?E5^X=Tz;*1uNZVAiQToGDk7u144OMsB%E#
z;{6P$&TBv5#&eJMj!f?}i{s@`GD^CJav%z<Ar)dyQDv*@K_S-AU%C(YRzsewR)(8X
zr=9wps{!P*M$ZwH(fv?|V$BMfn?0#rY~RY6JFefdpj4!z<-?S{_=vO1xJ&XZ6pLDK
zeJiyl?aEmx6}7ldwJJGKwds9``<wx3_ON`l)vu}a)12tD-l8a@9zxqv%2inIf%i-X
z#!_KpY>P2o@20~Bx4xiKc@rOE`_iN+b*(|50g9dm%rBc)_PDZ8`K{On^AE~J#&b(D
zLL~gi3ySNQW)U<6yT1~WZEOXGb^z#RIL2)3(ECVO)RnBnPye6krvWIL7rx?QVR2;s
zjvnFrHv2RPP;FP7+OQzA6-I~I4@~zP{b-E9#H3-;`zB8m-g@b^+Ls_O7}Z(kT^^gy
zax1wf6xNflavEV%`!M#YXs9TFRs;c(;BYr%vul8q9MPeGZ_ip!E3lZhPY=}dZyye2
zQovJeJ1trvPi=cca4$IqV1u6~7R;vr5&`yOsl$C?l9se)euzvs`<G{2#1O{RIka?-
zxW218tooPHe|sK5{nAeJbK5^P9_Dz>l1W83C(_?U3H@I7V!!?Iqv$J7-IPmMEUgTB
zLjPqBfICG?;;f+ULy94c@><#6>n%a3(`$)pQi|U3m(xaI#()TqfrQ`BBmQh9u_h&z
z+in3ap+q?&;KrrFtwk&_0s-b;e0;jV<vc2Zn5jz2Ag2t2B((h1@t|AHrU`Z6{Sb`D
z4p95-aPK*_q=%v{x+)BXah}&RTIU3sT~k4+Qj`@%<(Yc2jIOel3BzvfFY*tqv8J`(
z;Z19c?Kv;aJ38A0M>%hN%oSapU0t%3HAENXec{2G!1E$_5{|SFS6AKsbl<aaba-&+
z7vcbyZ_n_e9C%y!gU?VGq3y8rPd41SetpF@QFXU&dTzT}WgG`d_*z4&eaG5Y-LiwS
zHc<=UuYPF7=Gzt{#=H%d69xkdUmTdz@`WkcctI13Ro4DuytV4Vo`~^jKZ6opvRFY~
z?N&YeFa%NQpB}?mOHT<|%_#&C&|8U&XxMHZ^{RXO9Ic@ch@Oc0%thCyib&;HT%z?}
z5PM~0!ZH_!PoCIt{&Mp=r2J3gzNDw}e2CP4O&|P1pqtXBP<l9Wu7p%kj+SCn=CO7W
z_QEzxZ6)K-x1`I?Vw;62*st?ne7L1r$QwV>p(j(C`C=4P8m!PWs!girvGpkvJNK6T
zOXJeR{J!wFZ$1Abxy<=B8PawZx~)onakL+8^j^FJ$-Hg<{EOHRY`w5oQ&u0Y@Mo3S
z1dB_7QM>%6F$h)`swVpAtKAfP54~~so+f_nW#^UF>p$_+<X%a+;4sgp)t;!&BhHT{
zk}L`-xmv1cgO3R!ioo}9!$RH1fhON}YI}o^H!Z0gUp?7fRd<BWVKdH^okF|8POT}Q
z9cvN)r47xxEy6QdKS_<%$<G|XpJ8y6bSGoA=2EmRNM3wGx5j@Y;Xf%qW2VEq1ipV8
zIG;vyK$)dcOIaxtCE1w_K{IrGGL^}Ktc-T!c|SRPucAe`ru{$g4r&*PWCYqi?iaDs
zq)!0JiAyDBNp%x+u~PYiSno&FK%Xc^!EGom31%_<-0|}!_orf|5%vaQ^-Sos9~gBi
z*gMR^-&z~I<<KU;3KX4}8y@~GwIenDTsKL9;7Rf^Xrt1AmEJJ~LD8JH77!6H>waOJ
z^SS+}HwdhZHh_qpjq@<Ds^RHd;84DBVA+-0=3TDbjIVlGysNmLRLE>nCl;bH=R09*
zl8;ory^I&Cc>V%?0$T{T2{ja?mjqQMN?~kHa{>6s6RG8XMwz)8Bm33g4sMgfQT<sT
zYzq1#__@EG-%?-yNFl`crr)hYK)QRgWoDl;S)mRD3Zb&@F3FE=K}8naG^FF~FtP|A
z|A_zJC>JeE$FtEews9$F<Wdj`5yYM2TRthpEyxLzTH|Dc%xyX)3d8@6wV*oV3f!T~
zPlU|@twD+d%A+DJIrKkkioG17&9jk2$fzj!t_YyL(q}B_M&-3?C-<Ik3<j_HydL1w
z!wz=XTf~;3YiIuC%q-9FI1TD0Apwy3tCL$z$6k-*;ItMB9cY72h4KZHM$~Ytl8Lz^
zd)@+UgxjA$&0*64F&}xqch*chD>ob(mOov-eDS2$%YMV)E(DLs*=kpkUF+n8Y!IKu
z0S-xKg@1cB(LYq)+?^Nd+|;<CV>@l0mb~H{;+mN0AmmmVbvP>i(uaWOkn0o9Eq&ng
zM;#zR6!QK6bl78vFNnk<muKdBCjXj~0NdKaU%uHo0<-#osoM)aDQu7EkLLLl&QJ1N
z4~m}31m{_sDCMrFm}i}7`BAKP-NWSgFoFp_zS}jFU*hKAsnOF$MobDET!_a7$aft#
zkB|(x&k_BJz$M`{ip8AFdGXMtAiF&+BkGR6)Cm&rts|w(*zT5s#y$8A|LySfbtg_<
z@!;4|xe!NmejvbRoYZR|U~TBkFPIYSVj%GyNOz`{Hna+Tm#cS+&8R2=*vs^}|JZ4P
zyztEgunxIsZN4y{7vDFr!e**-QVRX>XD!Fz4ol;+Z<R9eQ-klQ%=67FY~{`PrH!4j
zmXEPD>3km!P)TIZhVeyfVLw-x|2MS;14ZfPs8aJeVLabE7>7KQIHb5Xix-Q$m4vwn
z7^ZCVfqtX15vjj#%i3K-k;sRIBk8H2iIT%`YDrS+ApW{|J#9fn;kdhS*0>NDIo=>z
z8^m#YN&*+kd4+xWG_Z<_ldnN^0?SV&38P{L&baRw3a26l{WwPeUM|t#(giTa*WWND
z{l#p-=W><4Xf<$EXSMpqZ`U<jGd(^Y`qTU@(EhRRqipMZs~Q(Xl7_&`wRA7Hmyxw=
z)!Q`n4aN!VW@V*9y4(e$Znl9f-gt}V3hDWks-EB=sa1kK!>iy%ykc#S%>rPquhQ@N
zaf81D*U`>!mEFK*rCi2(*V78qK5o+<j5$yt#b%D5jl6zcR4q4Ks*nRerm2atad6bk
za_rZLFw8I6*3wvFNCN_22d=y*W3Z4&PtyUke9^*BSjC1>cC^09^~tQu=Q{XuW4CY9
zUm0o$v(x6~wAu_c0PV~O8d-5a%!~)dIM;-E9F=*8{vl`^hW22XB`aVn)T(rnx(xYq
z*1B6ymoL3n3yYdznl8Y5s%tQWfkf$-47PJkr%#R1lGk@aPccii5Q@AlcT0@9HRf@W
z<uK%buSV)H0_*AJM>3<=H_xjoNybugPKyISTPe#$iPdfy{U81DJ$t*N_Cw|e%CY~Z
zYqgn~6=YJw*SD9;k^6r{f}M%wIkgmhl4;kKIo7)iz-HEPNuojxffBKR94lXxk0GLE
zv>YE=7Ky>uFSiKc)7GqA{R-u_Y_(M;S3haUo!;FBoHXM9U$Ltd{nEGO%v*86)z$~e
z^S4oqZ7|?}ZWtU0wFU#_(;6afr6}An?Jmw&-23ghLplx@);-o#Kevay=;eB8hMdFy
zyQdy04AV*o=ruqI(_&$E9^hM1;*L+G5y@>*>lPJ7&t!Jn%Gv`1ph*9YDeDnsh7Jmu
zu_gQQdTy>~E?z1$P$>5FBjD9Rlb23?_TNOr;r-oLs<U48<{y6kl9%ta6k=?-v{KX#
z?|x%R1^G2c!7^Cx2iD<M6$K)+Q1At;p1A)+Ju}TtMxef~d=d5`B(tDR<D-aP$3k8Z
zk{s>jN)LK2j}CvLq(QcE(BFp&F;9#8XSC9hI2>?=B?~4FzJpjwArfGaTwFjhVlwu*
z)pR)4sk@z$J{v2~vdsjL=z(qft2DKrMBM{Bp~ICs*p9r(@^d8H+S;$*sqlxfBM;TT
z8eqvpkCgU@&%W3Q&vVzYTV;Xt+5V%iP@wd<?+^$CG~YAU0{n>z*rTCVkWk)_e;n&*
znL7G7Y<>poIXZ2tq~rrn6g1+i-Dx@-?f}V~Lwt1i<4Q#l{Wl%%kAdO+V-kT<YDdl!
z5P%vb`cfEoolxh<11x0j6pgxw?rO*G=1uI;99U?Ez>y;r#vMlfiBU%uYttarg%C+j
zj{;K=Q0wqEqa=UEXfn}gmO5|p-ZNS0tVMmTVKX`IPDz-qAR~Sw`<heov3tKFKr8os
zA*UO3uRIpM_jlfreRawB&;8+N=w!lNfb5-|A^YEqwx-4V7Q%CxffgvaC5$Oso)`7Y
z8?t8%qqB9Drc+_r^Mywp={368xdG8_Y{Q=}a;a>*MEO#en$3v?``B+=_vlLgc6Of^
zAK+ZIh3iTtSq8fJF7_w_Ln}y{97RYoSGOH#0AR~Vg5INA@QFFqu)HLzXBHY9B1xj)
zHuW>tQ?i6WV5nk7y?S3&NV8A=YS9R_0e#HR2GR92I5!oqCkcWb$BW8KVT|M?33u>%
z{)|8j2B>J7cHTg%#(;k=G)*PkG>RqeQ8$CxQ4WESMO!zY41u_I{c8;9ul-}z>JCrJ
zpmgUqRSyFU{4EY`m7Wy9>qVEw_FR32vKX7!=4{PRh~wLIl9!Fvu2pakkaCmqUDr7F
zad9f!yweAm1_e@TVeSn`eM4;^t;vzoOS`PSlJ?;o>HDyA?}0-L7l6IwDC?>c?B={i
z`<T(lC0mO=MTVv1ef$HNC{5{S@4`;U%lo7tZ(Opy{sh?O1&y89(TCe;rD7dtN9N)j
zL{H8WWYS2RuI&PX^<#tH%Z%Qk+r`9!u72+c@w}oA2no|}HiF^I7#KWXZZwZYfSr7v
z@dX%t3~3+Ie<;85eWPvhF0en5a~&>N&@Sq!UG<{Ga+dZ{>LMFWyIqp$_Z0wU!Trd$
zM6^7i9=Vlbph<6o90tPEz_J8a<4-CnDeU*p7lYEen0jYg2q#Kn&NCCV{pm)aTRl1q
z`n&FvH_+0>gYq%SyVvtvO@xqYV_dsE{&upDU@upABOCQJp0l{7BmR5R!zJEj?Vz~6
z#7-{-LZiX{lAN}jbBfoMZ|NVIWV7}(d|f6I9SXkMF^1cv=~=nl0hAmaq+mq_w-jvm
zKaUwp4F4_%aI>E@>*8OaCB+4=$v85yK(ie|9d_UTua*G?4VKb0==o#{@&&X1PQ!9&
z(u+BS*BgDX>iTK6U<mnVq+e<_U+yKMWoD4Gy6cfw7L`3G1);m_#dCS>^F9?NQr?nh
zxbtGshbnI1@Y56ME64oU92}19$H$3i&QV@FMQuqrwvjXwU?3>KcF3PF)}H$Bke^ut
zzg&1v$S%rAMe9IaAZ}lggNEd9D9CzwFwN%&npKv)b~f|bQ*Y`%y+t-Aa`omT69(m<
zxB3EJbamxW(s3tNCsM^09bTFEw`tK^{|G&o1zR>@>0SJaxroy)1}Z{`sikW1L2Q*l
z6)-L!m89|e(-LS`>^OkKwqq$K{)BqVyRV{khKByZ96JM=d5<&t7M|p~fM{+>8|3X3
z;HhuD2<0Q2WWSBw?d8!(JbPJ?lIt?#Gt|c+i@T|&xmBBZnqhKLhcH;-tZ@TP80|dZ
zw9Cq>OivsQyZQ0ne!?MAm1uqJiFbuyJ$Y*(`0AKXQUvCE4!dslN`tx0ft_k!deD1f
zR?3!a=idJ!#^q5uVKEJ@IS&d43U5wH?cq?~q8@fZ%SqXxXP2a$!ht(SmbN+llHhmz
z?Pak9&mf5oRe6V`G}Qq_ZBBclj%i#z?GbW3-^A_<3+#?)ELV8aS8Op_IVfk(aP)!2
zJ>!KhEwN<=tu&vMz4tg_)ai!i_##`1B5e0JxCO?XsJ8?Ox1MIxJ6kj<MnWE{&(iA(
z{%97SA~lVXRFCXOr7dhSuPI^MvWtcloq)4kRg#GMnbymc4?V<NM$%3!hI>a2Rh$TL
zFx<h5*VdrTzBb7YbqSE+!2`Cf=DJuo7RgDV2f+dJ6dYw-gXUm!6mW@#G^0hwUv&q3
z_%{vyyP~7SeA(=Sd0_;lw@7w*La~Y7dg+rxB*H_D{6VFp0_2c5-a<HiWzpP^ivyoD
z4X+}O)J-neGd7)MCMGn@jm)jvS61fFfT+}h|1IX}%B}qdFXHF9k)3CTey`nMe3XdR
zY81Y!eJ5e4_0U2kl8iu17Rz2QaOYN)DqmjA8JD9K4J@Z`*<%dzD?lWR+~qv7J1u))
zEM8sT?ocC68QGjI45tS&qbm-Duf@V;Z+}}gtjT`<{3d^LyZQYG_MeSmpP2WO(u?qq
z@8mP#zAxt^z&E+)c4e&2ZwlAX9C6>TC{kMS8|e$=s35hK!^0E;l09y}L~Glw*7RsN
z`ZYO?g)Web!sP}ageGT$uJ|^>43z>)iJU&`OBHGwmwG9YW<ef`;5f2PhpJXXs@Esf
zt)SP1V<L*I)xh#I*#Cz+1LN}FS778;O&eF3a((NtpfoO{#J)QXL_|1e`2wY%iFH$Q
z7;>!u3i8auQi>t4-q{FSR%`!!IWeFoCKzjM(_#I6z*ke!xsQa;jyK!{@b!TrG!8|d
ztZ`4R+`{dTy_7x)gDWsYamwYMs`kiz=`;Gb_<S^wwZ|eUu+&0zC0BEl@|f%OQmp;R
z3#ffT3#Je6M{9vtpxDh*lDVQMN6y+<|5T{>;%}>EB;E5XZq9q?qe(BG?P&4;SCwEx
zbfI)Px_J*xLxVI3YA>tZGQUa)^l5%Q!R~u><t&4FksuwHqag=d6JgILLqtgfhlz7M
zTwd&^+W({J-2<8K-~aK=`Fz~UG3Udaa@a(uId9BqLyWpHX<L$0a_F$-5Ni%|iiL)m
zITfKux+UjYG$TqSL?|6lcgOqty5FDQ?+?Sjuh(<e^SU0_<G@i4f8UWMh)rYb+UmCS
z`Rle`cG?glyVugmeHt}DHQLx4HTXz-Db!hdvk<=`Z-H{G@k7jb*ycZOc8Wc4%3^k@
zOJ9iT^EV!8@o@N%I>{hjLCLQ{=i1%VITjYa>z!&T%goIk-k#Y9p5>Jn-lsnk)9wB}
zR^3JX@{^AUvciJHo78Tbt8#Rb&`fr2EavImHa**sU3QC2q`Q?UL<wlDKwhrM9K?x@
zFKGwQwbgii(8@>jY^A-g>OEAZo_wPO+@H?>Dc2=~wApRY!h>Fa%92h%yg=1n8l}kh
z3UvW!Oo2y+?RQtjj-Jt}e?j>t^ia1PGAi|!h?=M^v5RE@jg%E@kSET<0blAyxi!%C
z5k-k6l_g4(%mLVj$=-c_+FOOse*TRY%U)2g<p8_WlZQmHccTV=PKi5kw-X8|Y7O!6
zO_R7V^MbUvz`KWagzXU!S36CUJc}M)Z|h>bk4{0b>PKAr088QpLK!fZ*U>WYNK9pk
zE$hcr(|0&l2_HR=m4dL;AO2RJ;X80FNErT}K?9{L*L0H?A~H*`6CQ2XmfU}vrpx4x
z_~8dLt_|Ih$KR@Bbu31)Cm}K}Bsw{in9XyUk$I}nyVMk}Z4}SRV_Z*|nn|?m(aI+#
zb$~`p++sD<qKkJMEdY0+EvKn+4Mt$yw@vIo^^w&UMLJ(-!OISD+#PFB+$^93Nu=L^
zJ!}(#V5h-=M-E64xB+vTl;F&8oJ$lnG`(y-#<pFnvM0&`7iG3yeo{(C?qq<wx0>s)
zxwEsowhUhxImCnWRdSgB;U@tIVO_#=R+5XVFCnmMJ2)q}bP1UdT!+(llj}w)nzt1l
zorDdZXmdY%w)D$~e?cjk+X&9stlF*a{;7vspllV6wZ|_d4`S~#za*DI<0bo-VG;wL
zVbdp0&gHX(nXUXddmhXt)h+8z+^%|!T!w%-M|Syga8b!LSS3U|SVg#-<QJS672QU_
zPFQK!FB?{kLf_S|ksZcwS@*rJxpduD2xLN^NL-X*J}SFopYrY?w&tOS(@;%xR!uee
zYUsgd*R7a)(043+^m)DGt!!)a*D$-5;D8U<m55ohI@uR17U?lHl*F~Rn1S5e-<r3Z
zUhRc|V`j~gUX_2nXyPc>HEzvN3tlT8xxKTg5y9XEn7zJ3G1dqYU;-5|SNw!$E5?q8
zd&#%?sVo&ejj_T4DGiDA>CeN$YQZHCikDZElvfl0Xu*mhI0Y<Dp=%Z{v3!pNGgP7O
zBF=g^pRR9;I}|>Xtp64vU0zJG`sRHCTKmS~FWa*{$G-<IYT-}4lTPZ%ix>4ZRSqDs
zoPn2F3Yre+U`6le*8)@$-3z<ItCwiE)Y^2`+TbGkySv=ZkF+Rw@-e($0=_0J(U}`#
zvp{nJ=`>&2WsTE^jXCJeuHe--|6gAO@duH+zC;AMAEjQmS#3tmBhLKgvgAVIv<ulH
zln*rMC^`h4AV<*nfYtaVlA5mg1=1;jjnb&py^?hw-6u)GJqPaGG;;8qsDJ9|#nG`>
zfxap$_OzKP3`-&K8w&l(u4NgPQT{#!tSZZ2$jA-+us9j@Y~Wz&>kC#YKw9<^$H#nd
zHTRRgZG_VY!G(!l{h5|i)=ypwNeC_Xt~`@7><kWYh@bjP*?ot;-x*5{M{CL!G&JjD
zo-)~au{IuLz6}7#+XT{INSWkCx{G8(h71r~g=rKKoFD%2$_-oms8mhJO5yfy^QCXp
zemRG;2<3fgH!^kXe!)mG;=)BRqHy!4a~Wo9B^s!W-o=b>KaRM@{oNNO{#k1+bgYNo
znTH9{D?+WZEt?naOo%N?tgrvE=fd5y07XO*)>sx%bl}-+5wylSzjQI3Kyx)*$Wa*<
z);&kiUPEBt<u~pt1wT1f6|Urtt=bCtJU9D@P!m?#c<Mc9-Fq+8?P~f%YAD}g)}C#(
z*AOGst^M$nlRL<5LD;`%be2lH@bOK+G6N&r=!}ww%V3G`0IVLx+Yc1!0+ACmr*~j0
z-eE^RIB@%&$;#m*qkcmy{QX{=n+>O(i-S%|#+_2y@#cjDY+^&Ub~w~pN$#V-TTj)Q
z+4-(;H;aTIwiBrq_2jOdYI7Pd^vM8rttbgoJoxsiZdHFGDYpU?m^npzLi~E@b$Yo-
z^TD9pXK+fd+G60CNl~SB!Okwe4h?D)<>;h*_7VkvToS0XbO<sMVNyi<pp!DVqe<r9
z*jv*zG@;h^FKE7AarE)csq~<u7Bg4UilaMLN?fJYJ1m~kDuQF#4u?v+av)f+jgMi`
zLwG4Dr$Qi;m+v8v$>urhu#UsR^k+QnH-mG6N`LL9Kgjt5L=P+QT-F<s)HHOz?zXO~
z=i_49)%5fg<iyfitW1XfzUR;nt1f*MV87w1+2RH3P~egF67HM4%L(nII2i{S5M$qY
zjk#0jZ@K%HZ=}-&3phr)Laq=OYC5J|U~?2+xf<LT<q0z?!Iqo36wJIEC5$(>>H6Ij
zD%kNE08K^3yc{o|>(7`yzMU2TU1u_xZp5+EWq$ogF?p}WBA!a==*jY*ya4FLELMd}
zy~Y@5vDRc%orPwaH4Soy^>3sQ8eGBcnSHb>_wt1!pNAzdMU-4Wq^U{p0p))|ozJ=D
zq`CQI{WEQwAwpEDtIpM@Ryf_j+)^3bU7V^+OW&s^-E|rg?^Prv?JB@E$;iU#{=x}p
z8VD0Wi;{A&!Iaym)Kz4{ar&}W`BoZ6d2hFO%P!kXQ}NS(&+nvs#RR43FCC8x`B8Fo
z?Q)y@?P`V5#}5tj-UpsNpb!yw>6V!>52RX01+xK4zgm71SUJ!f;2Q6Wbmat=Zu6V(
z+pGg~*ihgJK-u{mt?M~_vJ!<aCpb+M-+#XyG-l5l&vD@IgB7QLfcEwT083%ee;;6t
zLTc#tcu5%Us57J?{f2M)tj*b&Of<)et0=9OcMT#zotsHht9S~KsoH=vKIdjvsP}Cv
z=JY|$-@_Fs?IKArZ?GV^D1R5iuJ45$61jh|U*)fq1ibAZkdYR#8j^0|6Su*Rdj2jy
z!%vhHA2EI(_mN^6EUS*`UUjg`acBpWbJd|vS!Q?3%G#gbXua5{!aL#9>Sy6*qW9M^
zPA_>j;Avbk|6%V*bDA06!)&F)2@QW6e#HF6*%L7nV;77$sF)H%rrT}RdIU-Jwjq&+
z)oJIcHUl4!A#=#MQc<xwL_(77I3?_307A6TK8OFjIdLKjbn&mvBYp|qgz5_i11~zi
zx?`hZzwW$_9MCxHI32$D8L7FJqiePPbptHQj`>B=`f1gc;eY6H%`$?1B~P>HyO!4L
z!%rFjse7n&Ve;X}0AB5efxU!5RM>1`A9sCcfDU7FFDzbS;XqP>lS%fpsEF*s=P{OC
zhW|O<D8K3Ok6Ri4S9Vn1HoK2$0{nL8pCy4OR0toDGIY?3p2jWxHMoq{Rt;IM5iTR|
z3$ww9tvxRQB>4RnHknM21D~$~q;P>MzBZ=tw87&o9YqR+ZbK}hYuBK&V7~2zD{vPK
z1H^SkgUWa3CS;SElq4Ig=HoF^Vs33{SWEDKN-9F7Hh)1#sCq3%Ck0-9r9D&8aZs5V
z9#bh&bVzPG*y?(Nkb&=D-Q6p8323UCg%zz+?Ke?qg+Qx5@|>w`5a1^ri24Lu!~rOf
zKNv-tb5SF~-|Nt4g4597j}(*U+-`ReN?z>3@7zJ30y8G3(x+t~dq)?4S)A1V5&Yp^
z<Po8UTUWgO#5xHO+%`&tr^>d18H0^kBESOA+K@ZvZ*A1fuZqdCF6A`eF>I5erGD=2
zc+hkzY;~X+3HWtv#ijq=$Q4453e;lSRlEtH8d&pVCo*Fm^|U+oCL&_N_-M;))4no;
zgf<I(5xKHjkv;)b7#Ly?lqRcvlgcW*M|wK;CEo6kfT5IMHgEIN^Z4G!XVL2cFDAB+
zzvxe%wj+;v8ivq|lJJ)M$9wNGcb*NVcK^}QJ9KJx?^c^<hQ;F_JCGk$tLIF6A0eX2
z*tDx6a+$Ozo!>lY%Ae{X<)qce;^a3B)YH4#=jvI|@FuCyc2R<E(K`@<!M~FEL=u{5
zOqE-;*O1D~OS9t&eP}&AM1Ish3<!Uui!Y`$Zu(B2`WC^My;&=>Zhq=3mC2_as$f6<
zX2w3Ft<q#{g=R+Rft(?3p6Ni{VGv>RkS^L@xWBLud<u-lIM33eb3I6GoIxHyu66bC
z)EO+nlss8A+#JFrUBCW4c%;veXQw@?FIDGzY<9ncKh5lfna!~>GL_)(AcL*+H>ahX
zv>qt|Cg8eiCdkVx0?y3)ukQHFGEmtY_eyCdC~|9cYBSR1Wh&=-I_U3)Uk!Ph<n2S`
z(5!FTaX#7*Rm%}rPNmU}fUs-X|5tE8SgV0ru-B6BAlGZR^IV6CL2w)?>zRXbBtjjb
zqu>xICy>Z~-h$eha0j{qjT#uU{_Mkp#C!ds&#W3BzkYM{6G%o*G;Cf^vfriU7<o^*
z_uY-;!YSqPA97vAy$%&ej=?OvNS{vp?maxng+IIMd1hg2@n4X<^LMF{Zr`AZf(<z2
z{CXUUvD6?*Tke>e`_$7Iq%c%hal=3dfmXeJ0AAVRaq0>aGovuqTMr)s@uwBAQ#wTW
zkW~{5lyeQ4MW7a+<el-(Ot=VdtWETsb`0+N2BFC%CirNc#ENH3xeJ4z7s)=UO&Vv(
zyYi<WU7Az;X8uTV%l9IyO*-MW$i7P*F$|^ol+MDhd5sJ53mw8`I?d9qPcy-^^Sapp
z)F4SaDdR7)pBMcSyjt~joOJ?`lB6?=nX#82LjioNg@#{8$^Hj{w&2IVATBVR$X`3u
z1(sVBJ2<`MkT0wW1N-$KaYl|Fmd3n;QL?affTf1im;%|8u_T!UBfwcI0VzJ4$uCm!
zu;I69zq8^QVr^Uh9IL5l>kE3GIonab%H<a}pu{0LM(M67Hlo=GUWv*pFxhM0WNv*g
zrv)|7ww>T#2nraH+-b3nJ~g1yFzI5fjr&qG*j67WL@_vaFpzy5^`0msn`P@{@)(Aa
z&b=pRavTrP!h@QGruE|?ES$o7Y*h0QklJSq&Z!t_(h1KiyY`v;=Sg$=xzT<H<;15#
za8NSz$zK!CuKpDN(euv_KEodo`$i&D=%(?otGSY}A5@0G<)OsRzx(f6q$m{#Ksohf
zF}E3IH8Ho{V8U#cHNDzII5<ato+u#H)`47SwA~Ia9A64OQk4++zM|~;hq3h%OwSg^
zwk6RSGTmN6lbeVQS&1B24669^&B<gEe`Tuj$-3ljljGO<j(Ab^YFKzOG1ZCupd!p8
z;oIc8%S<KjFP9>7hqvFM4XTPGP|@V+h2`N>QE!VY_0{hNy=W34>mB3t8}A%SzcYd+
z*ZA}@ddx%bm_?U3HJ5*=kd4>5ZLx5~-XX5&Ixgf0!9C-awO9h}Z;t!h0sfsg;jvxT
z9#6sq(*}<p_$RmOV+~bu_Iow^zVBFJ4wqVup#}@@%^<o|zF%V&S$O7rq?i=crgmK7
zk{Y-*BRW-;;JA~`e33+t)}D1d{KR?Le{gT1DF(x?hWY~mbgZ$f4;%cP;h{z?_ZM72
zPJQ-mGR7MN^Cc-4i1RY$cq{O6%IYY<fOaNq-m#CQ<!DmkBU!btr1Fey(p*XKJ+zLb
z+Sis1XWT=DwC9X-@t6AE<$$;~Z-=fm=)_80%B}p~Sj*Ga2mJh}eL$AG90bU1Mj8uK
z{O``$XRPY$I^KNigd<K>JuqTQ^d<bwwe=%mOw3?LzNz<}emz6P-Gl%oa-LHPZEnGs
zB}4-mvp~!?$BLz9srOP!{WwN5fS2HP3@d!1tmx#uWS9TKfpLn*Jes!zdzHk<<a1BN
z`{Cq5&riSW0GnbSg)jU5yg0eg|FiA2XfH{bFBek|d9Lxh$nRIt<L{EL-iO`4YoIfH
zS8MCcuZ=9ramQ;_<pe=VNq{}K{%*d3Ca_cqwzRZw@v6A7H)V?#o+mU-X^0HR1EZv?
zk8YTn4;;XJhj=;&H!B+tQn4yfIxI*7(+XW(Ly#PbQgIQ01XE1ON)pLG^Ux!6<G)QG
zv(LXDttsnv%4JBdt-EaFl^j$)dL8x>Q(1=pblaaok;|bt?Rn)?G}*_^HCm=$#NWM=
zCH%ZSrbDHEuY4hr>~VC=`Q1Y?Hrw;$pPRDZN*8Bn?vf}X2zb;U>A~jPpzRZd<ykgY
zdrvc2go?0+IJ&BPT9Y8rSn=%PyHI>U)tv?_NM|_$u2&`Hx(D}QhkIG5Vwf-t{~-m6
zJKvNMJq6xn#S>Mq6iK%5WiO;L_$62P4qnV#OSl2=OjiQPQXQmAff+G(%qM+M!qn=W
z<=oNJG)CSl`jj<HU3fD<83<dXn*eZ}byxS%?;2RmiH||AYq~vRZyu1Uf1ZnuAJ{G+
z>X#L~6)G2du(poe8{B|npv0l#Vw`NE(EEVTxYjeMaz_I@^8nqY>f2394<T7Ng{-K0
zf{$*T2gxcZGTg<fsbw2gf18whe&A5RXhTvtta3j0a7Mq|BWl0Opnx(P+vTWsVar6-
z<#c@t8R<_Nq-eTZDqTuI<}mZQs_F^;y~DyfNkIl=g^6OdOil9C@ZqG-jV?BQ&zrz|
zn##MDX#;49b0x2uAT<r2oCAhczW1<Ga2lD+J1P^?z3We-8&b8rPQjnd-AXl>ui9Ss
z@V>2HOfEL`4l5asdpjUpu$vSSP{6s%ZS!+b*7KNER?CUw6zgwqUD@Yxl8@JNvOjv;
zFU@!?FI|d*J4ca9c&Z9P<n@LW?X+5-4Ub61ue9#Mpp|N?+fnpw{4)`EwSjvPfVHqW
z6g@D2fLkqB%_^Bgx7aT&>B$fU&{}uYh?Vl=^N;PI$qIW5Ky1;BD@7eecCDsO-n~2$
z#I5qLI8SJ2?d$X9ks6nc7tWhJw{d7mk-1Xf-14+QG<r~<5`gMp10+f@J`GeRi;>l-
zT`I+z;qf3@_>?da;4T0b<xy%Rj<-?~Gn#G$ExYmP?RsYt{RXUU=!xIM<ugNt&uc%6
zygL3A^Qo?D!?|hk(K%<cp+sNPHeXX%97*%C^02GkGD7Bfg5lb5cmcII(}SoGQ%q4I
z+rnk?Huk`Ca{mHafK=A;BZ#BlS1ZQCX!I%#wTSTPTqWC#**wti=jD-3BL#;_|Mb7J
zj`pJ`&DJ_5{|`bL`TrplPrVP&>ozOQSvAvq-KxT%>dsJfxh0_SXDD`XkM^&Fg@JA3
z5*0e%EFPQ8At~L<$AsG2`jLq<x*0E1zW)4%4L@9NCm|<xs7wU3*3?|O6|P;^g*bKU
zi?g2LZSy6GduMK-`#QtQO9L~ZdvuqcBDYlQu1GEW=rp5pMwqUFxdf6V$5o9a*{qY3
z%6_?t0jyZ49N`BI5Kxyt;O@#6CUt=sOL-VPWaUik{?(I{e$tcpS2aVZ)4ru|KLn%d
z#c2UFG4U^Lnqp2rE=5PjL7XUC0!0XG*V^Py2*7B$Q(#)ntg|Rtn>pOnlz}x@puXk3
zfBmEttD*+ns<6v9;b2L7acE}q?Sb!(-O>qF3oD6!>0JH6qV(B!!@!=pQ*{NWf6F%t
zB=8k4Pg+N+{tFsrcE}%-)mBOxT@F~Q<yiT>T*pj8K*9@8>HB1bWpEu#5YjW<KLIJt
z1pjIR$<&Zu5>?^hpU7Xhb9PSTP}JWs3oxUavi`Vu(c`*LYZb?u$Q}kdcC2VDD2=-g
z$w9d6(A-1QxGE7?)uaDRB7|lBz{iiH=dN<}POYR*p(9n4cYmAs^o8lwvfL}Kb+icI
z6e}9v(>idE2IA*gnw%;l=2%P|Ekos~6NdxLTwrwMM;}SX)k!b<ZgYq3KG?P2Vm=N^
zI*-n8)sz~^IpBnjyhXKo2M!A;U&(n_g#)$9t7MPsxmhfV>ZJ3^4_tI2yTuBqCexI!
z9324;#9@lSnFL_+a14}5V@1o6TN{}~V(YHru>Zq0w1h*2ycbd8hS)!q=kt>=wjING
z)i}ws#<+EyoIF1#1s7ctHN(}%4SPFDVLE9p{l-@uB8Ef89fRf~gT7zu#uKhhC0uVF
z#m%|#Ck}iPJsUZ*Qn+8`yY_suv%#eu$lSl6cqd-E_#||`(^O++wdT6MxoMD7R?;1Q
z^gZ@((}Obai#;5fZ<=~hy^Vu9d5sC%K*W7-j70`>jZr62k`dE7F{pAY9rzDbtG1_T
z>@_~=_zP)%r;z_1zz~BLI_bXdHGu6YQS8uH4ozD~Ym4f#DO!ZBj{<xqn~qOU*vA7-
z+nihGoHuQDtqR5l6Y?xoA<!n{Q;56%);2D8;FL;POjpHI?M5m0S`tYn9EaingEbte
z1v9#CJRSJBI+Fw+R~hMj4dA<eFRJ2VaRn=#%FaS|`bnF!xAwEQx2ih@vsu;GvzNc@
zOr6Ky3BFanntSq&Cb?{=N1T}hca0iBs}}lR>Q*Y6H?FG?)d7M?0N5#y=dzQN8$J{2
zBW#MEtxYHBkn@_Tr+u~V?<;G%W2D<qF_{pW`i=YN9bDB~{Ks^Hi~jC&)1-(xiWFu_
zD3j@72N`HcFihq7mu((ld*$IhxOuIru4ryv(|<?@Pe?nZc?&qqM_&LMm2NSHX<s6$
z`GE4dH{#2f=f*Y9byqZkd2L)3{iFcHqwizGTnkVww4PSHQAKK_6c9YbQ>`290Vam`
zrnw|pfUpOj_OVu1Vv>sHN~ba9@aeNS1=jRT6IC0b&5C*ogmO$HWq}{S5Q%<Vx`Cah
z<(OH>LOis8Z(4VK&~<@08YlRaupWpah+Wl4W<{iG1b@udd40C#7Zv>;o^tqnKu7-e
zo<A3tgF8ZO0zI`K<o9jqZIAvFSjTWn98n>FPW;_qw!hbY!wya<W)i_B2i|hQm$U6O
zMU&Fse9s1|!P^JpYY1GU;T;dm747bjO#2VRR(s@~eJ+4hSyFL5rwM-aSdl4_#BH9<
z?l8!qD#?c>yzUNAro__VAlBLXer>5^)_!l1AfEANCi0QL+(}>k$0rJAlwi{0TW^oP
zl)RB6wp>tl&3x~2oTy(F%@qRVWz>{{)ukc<Zzpn&m#g~;Rjx(?s_tG*4l8n&3b`Ss
zLW^+pM>}d?T|Y8=sFV`#n)w0iGXLEN<ZFjG*z2z%vsAISZB0f@_9%|YcGaSV2i2)W
z)yGH!1Qw>tg%2)(7N4(^8?qk@qtE}m7Zl>^8?wziAXcfnLc^>~iEJd+b!xhu{qSds
z=XND8Ho5)`K`$gX45~g(4c+6a@uK0m$*;7fB4t`CPzy>2TzyzB2#*me!=Ag3TDH>6
zPx6tuu^>1#?EgS4JSg|7(%+CNg%9JAAHrWg#uVGx545=I<>x+V`E-9*^@C3}qyptY
zOFrEtsNQ**Zx^kkt=r}fDxziAz7z`!aiLO~Xhqu64Fg*R;3z@U=d&U!zTU)16&hKd
zy#hD8njH4}Zv=Mf=;J4L!qlTzk{)hYi!6wxmw2=o7rczZy<jFrZ6w~g5fXcB*Qx(T
zlz&pe?n9>u*R2<nD^o#-yd#otq^apHpcK&v5&H3xxnL!jyh5?19d`vLu_3=4nrX~S
z>UonfEv!s031pa;?bmuM9VAHph}S#Nt6NA8;K9Ak@lU=^vzrq#j}>yl+h*>1HQGa=
z3ho=k20Pm0DLO5`Nd|xe*w(3&(1Uz}+ZA7^L9Kx0Z$$!BLzc8tKuxId36R0?l`*bK
zz#!o<3VrlX@4uj@%ljJP%e`g~a@NMDEBC5jd)1jf&J>i~|MX|CH(g5bni|s2Qxv(C
z5PGNhP5eQ@Ek@h}7q=EX3?=D+odWy7L4)K2EQ#<Rfo3ECj1DfHvhMp6Buyh|$8qY(
z5y$wSEK2ufsm&9=Cc<n~w_Npv>^A+o^$gwo;FqaBO6f159qj=038yoOqVG&pR|Wvm
zpja$FC{(SvUeMhEw&y6BD&`Xa4+mXwJMPhZ&=JNhpCc)|9EsKs*S)B(g&1tFBR0A9
zF&_&|q@`0T@BD4gn%9K3-JVf}Hi@L_))ucEa~}5U7Q&n{H-*O6RcFnW##HC^0JcBw
zz@Me#v8N2I|2JrnI1G%doICh4QhDS>a{_Xzw}EZd1#FUhc%Xz=2P*L=iGeFDvxl40
z-;WqCY#&wl2lHyFz|$-Kt7@}CJuSTE*OaB6)5(@k^sv7!vUspqc=FR5N(+RJ!>oTn
zY8|u(%sSO)`LC*WnvRq~a!h6DfsXZYYrk)PW?M&`m7&Rdf#_8i;Ik|--xMTjGkmtk
zN@4f8J?D;=Yt{9p=9{Bpw>>b0!uCzIi_?n4c5#HVa@8tDr|8`LFOuXRHBf;s`A|kO
zOI&#rSa%@7ACbEskfSjr-5_JTdR3OtRq@#(&^yva8nwJ_()W8?i@RvrPU}jbM^|KM
zM-@T0J2u)}ZME}wy|2xbGENA&iIfNA?MNB;LloPCyqAsDj^rIFFPhl0B#Qw#IQCOr
zlmd4<*Ey_W%BzJ%g~MvM+ZpIv#rlddwaGxxpvbuYR+FxQ@4^aBQTMpI=|Ia=_`Y3K
zd06NO471$M!`l$lJWiA8?E&QolF|q~>-VtHCPrrxSJ;4ev&zvc06bVm)1&isohKpu
zt^OnMq5W0tqzB(Pd5{_jwkUTlVkS)RHhPQ034>}52Txpr(dSfKqfaSjf3jj;LfbIM
z%iSBT(%6SCL%T8&+8EKp-Bwbv=9WOCO#P2cjvUk2=DCef%icCm`aqB~>PdVqSpKoe
z8K<<(Knh*MMs2iD>U-p7V9trVMdVa<6_KTDwhg_fW8TpDX|i)Mb!+Dk2H7=+ljQzA
zwK>9Zu#hdz4t&oi2+v_MScA#c^XSqKO;y=Q5Byy3Afox>SxdGfJudg}Gsc&mMeR!u
zPJMjsNOy|c=irh2yV*zASglUJ1btq~KvT(7S#NnqT8WT55#^?^cHgwXUC-&sduE70
z3g{`cw9RarHoX{G;U&CCT?T&ZLzD|-*4sBa3XOnCc4U@ol2aT!p=gD97s7qton}ju
z+<o8%Aair$u*t?bopV9F&YFWa0GoSiTneK~P?lG#9=IR*R5~zi8142Mfj?GS@zjKw
z+5~r*&IVZ+55o;DY4lHIh75_L23R6bqdBOo>Ckp#z+Ajcp|f4eUjLc$lBUo86~)2C
zyw|hM(qYAq3B{mBjF8EPQQdC+56R+M^H0y4y1%9khB0-kzAPIduIbktA8c~k_QxJD
z_neSy;R5aGrE-9qCCL&)xUEvOlYkdfG^R0=W~_R2!$>1YhEhYiT*3t1X}oO_I6H8a
zu7(-OU@&+dFVp$f(Pkj{2z>lMWMU3LCV;W!12r&CpLfy+?BUV;WJWoi30S}yc1Bn-
z^YV2RtX5mb{#i#ttDF1&ZnLWE>kcma%+p=+4?J)c?BTV4^QdNik()*o<J7zwV?;F6
zeSrOV%eTz9$z+MoZx4sT<}4rqz?e&d(U!0PlFRgx1nB<?(-=dl1ljN_M|km^hFVt}
z)77mshT#%Z0+(^cL_tF<7QrRReT0^m>V4ligz68Y?akWX5P~ibO#HepVC^S8y6l-S
zEG#?@xNMPbzr1BXcahOlKCE|M{73Dd@jqp@rhAAv6TQdMv@p{V`YdrS+Zs>c>fQHD
z2W1^7pZlZy?0KX4m(NH=V}ol>cd_U}8q(GNrQ(ef#|Wz;n!|{m9~wIgL7^8u568xg
zG<_KMkyuCOEf5;q7LqqHpn{b^v$Q3O%G#lw;qf2-!a4H30P9gqBmgPN^e2KZbTu+i
zqMA*I%RnY;@M?9YSm>Rf%d}Trs)dKjs%W`4D0Z<4!L5GfZ_$f8+~f6u$tq2ZJ~gDH
z1}={W0vIl?dI;%l6x-$DY|D5ia$@VCK*ktc=Xl+#L$=fAriK<7LsCl^fFg$-Sx^H;
zyQ9FPQpVk1&J*%0f$_Xjzxv3lVtf0YQ2?+s*E`X)h#RT^6q317--{1`qDla2BiiY|
za*_>fnewV?VC*0MI=$?dqwpb*2Vs{)gNrpZYKKASQ&vCQ1VNQWXjSnBZ#83=e<qa2
z8hgiplA+X8D*gaI8k<=IJmt$I*WZNn?l{i@A-1IZ!HVp9_`s?EdS0Mb6m@XJr>rIA
z$bHks+0BN{Am>Sh<H<hlGfpN&iV_ogIU2MF){&JC%E3j_)?)cyyP9MWa`6-Cn)5~&
z%^&pSr0;Ze-V>Wp)JOKNroPTU(2(b2>J*Y;%J9)WZ*q_ohMW-8h-8~K?_+J^jP<-z
z_EBp4<C8Ok?^EYK;7pG#>@CuM*NxzQ_iVb+HL260J2d@0zUW|7@3Rsfqi<z4$=dwI
zpx)8PCe`m7UX#DxAZOPZVA^1(?n$jOi0SP?3o|mER~LiL*o2q2qqVC9mb><B@AP8U
zqYUhm7r~WgY_Vl5mC6ecri^bgygW5wx$*0?JB<s7?uEnxOQPa>IW~`lhOAh&q%O51
zbQAo$LAl3RfD*)8I_XH!ueuU7eH>P*#QC2FmbjEG)_|^C9|_FelEWS&m1e+JVOoJt
zZg{(r(g81SmvSO{EuBzq*-U(G6=)Epc{zPz?ZtK*|CYAZGiH9C6L68ql{`Ga8l38V
z>+2pjZlo&KQU12x_UpO*;!Nhqn2p&DF;we3Y9ZT}*4`y^1|s3|U-_^CUXX~ne=pm5
z@1fnJ?^J~aa*63lf!2mRTZ&o8=YyH}e|FiAPxR~)M+o0(&@*%yHLM^Nr~m%H9t2SU
zW0~*_{L6mKeh$uk8t|NDWZ1hxv@H^J04Rk9|35hM)&V@|2Crrpe`NI%L-;;9`{6x)
zn&l<*1`(#7eypZ_#M=I2SimUN*}CkJM43q&`hb1_tq5xZAQItT#?pCA7^SpSY7r1?
zX0hSp08Ysiu_}ONH|0jAN0NmW8=+*QvDyN&c1}5a3JClbV-9`x*~F7D-1zNt_3N*y
zg65nGosx?E@b0RPN6>lN-<IgxNU``dc(Xk#iuKy$X48#r51;b3KX0^ZxeV0vzy98H
z@vi5FNtCG}F{gd(0hjQ5p=T2BY6Op>C>9}HT#a6?oc-$)3wc9waYYUN1H3^L>RS7z
zFW1dGnRUL_lT{mc#Pxd5G9&0{VN#fC2a7(FTj&VT(8bjg$+}K`Q$i@q3UTOP^xwup
z*7Js?XMm)sC^}9B{Lx)ov7h_P82rz`8dDau-*=>#0y(qpJd#LMjd}5r8g~1y<bpY%
z0r9r+knKHG^;BUwenal0HBUu<f<BXF4+k6DYfN`Or^tNeFp(ZFO$>KgY?pbiK7?k9
zGT8$Gy#jbex6KD68ecFm+>)Ffw!Kk<PDinx;unEvs2}{jw-Duvad-i&*h!uo+j}Hf
z6F8FPdT%`d%vFl7Z?%%ZzuHZt&tbLJFgdnNPjZw3?YNLyIxWXz&L$J7XqL+>N`t0p
zz@yb%VMpc7A5v$pRozGmMa>$hw|?C)M!OY>iM7Y%@~3O(rB)KtfNT)tIC0oY+TBv!
z$B?4p<F*491_oJS?R4sTsmqLVax+r*dU~~_l1ah+nFOJi?}guw_^u>{pXWn*bh|cC
z`Fi@)%a{mSHKBko{yZbFlblg}o-q?fd?LE%$!CSMVQ;}!V{bilg1U(_7%T~$`<OY^
zeuJIh7ruNG-=p^0WkcR#e;XX#jlLNs^S;ms+H72}YT&qwXpvAH1`Gx^3Pww890GcE
zt-+f?@68_-@umX(LREInUXFLC&>p|7V6HXw5`#~hr!~KOn`hLJIM$`VP3eIQ#aj48
zk0x2KRiYsbz&uJaddyO?dArx~1T(sWK@ZHlo*qqu%}qkaMc>{Pu(pRmXOBVdZO8s9
zw~zi4#Fr~8!I#%$oep|cq6R#Q9&Su-NFfg}94&7sy<_B6=8h(;lX3%4J^^}Y4g&5C
z>=4L#1W6zw+S%|}L(7Y~+Jy(#Z)JubYiOF}V$O{Y%Fzx=@vSyokIQ86Ge@BjvICQ1
z^^Q03Am-qkwk;_2>kW?jOu5x|QYcSWM)5pn(fHj#fC`&|)NI)k?wxDmJS-~h4qQvh
z$lMnVZUV4$;C@uJ8Ce`IL;c+>HfGPWq94wOfBT4$eDC=K`L!*45eu-x2>mV#d}~6&
z4%@WQE4rr7Jr)oF3903r7-*&mq?DbojLkZ{8@H^wm$TX%>Wb!+_z$_QwxE1KM^c)f
z5Y<d4=9-_(v51XkmmQ9LRGZ>={l4x{Sn-9Z`Tn%~J`nV{Ql{-VvI#V-6CMwF<9uXK
zCQvWDgx;WmUoNREq>q7+0s!SmM8bjOL66Cg{%8ZNODuJDgmnLc!%T?t*TVGfQ-@IE
z{f1p4yZh^7*s`J5SO5T&W7Hl0XMf-JuCh>7tKRA5|Lni`*M~5-?eE*OlENzGoJgLU
z#^BV3fN@szZNdRP&}!cYD`aPE<~R8-tt<FBLHOyyKcc@)Qa*&<nyP621}Zr6eOZhh
z-Ep9T`uD*UST06`Zz+@Z0SS%C&OD9l`tkKwbDxh9__8iVuRH>>+M6Hn_-f%h*l5Cz
zwM7m@fqd=fEL%c6y;B}=?WowMS*;@#CKBh?{Tag`_e@_+V+BILg9=0@&S5%jg9y%u
zp?*sP?h`8{c&%RphyXR&{gb?8?aa0gk<-w^ziJ<#{}7<KO!ot>T|WX9t8e;;#@N0$
z**OXWTfm%`D0_;gM50f$c??Lo>xx5#`Nq`dJrJ;#hd96c!`vzC!Kf109<XquSNN6+
zi0?}97X3LSKT;pBa@-j~%&`=%sn~6p=U<V`bq*T);5BfN;B*UCr}9n`Bpl4|%2P7f
zfb~3RQ@kcuTDi6l2s5+X{EqVj;w>Zpl=6uRgBQy@cEDI<>%4|mK(G$uDuC$j@>RYM
zB=3rKJ{_CTSV^Bvgy%tqPOUbDcB=mi(obwG2|+#8442FnYSZSb{K8HNMHK;58Ov!f
zoor1M+WV22%{9*65JR{Ecw7HEOYzXxq~O@)*B0AAkbP)*e6!o-w?<;U`N1<Ewp2$=
z7BV+XU8GE{^lHjd=F6@Y2rAnSG(}p_Q4c{S{FZw>ot>}HJr0ZGk>N90)IwjAFo&oe
z^I>JV-8jXrq}i&_Tj@$ZI@&VRs$*U|#@;yIP^}qm)hvCrOW%I5Wcchu!f0isW6V;-
zm8pH?H3iS^)Pbh+p?Q@Fuu9s^HXZO@^AXXcz~j1Ew^aWk+CjbzY6pZr?g}z;+@=Me
zvk;9Ld<*lymx0)Zs8n2F_*<ZaMAPJ8i&2%j1y1XWpD`|=q*16YEh6rpn2%$EU~6RX
z;<F2FWvvZ8%k86xqwCfTVS#*Tr<>ySilBcAow1L?%BKP!p&_H#86}ZvNtP>msh$v+
z7mXoX&vCquJG9K55_H`HiT5Y|Xw0kZ)MPzBeD5S`h}T)gBpD{YA1V=)Fmq*sas#|P
zp6+t2*nc2=5{UAqGZP<tM<g4x{Q*ti?0OMjB&N3-=X(;17`PEkyloiZ$FXF?LI1rV
zMRm2jd|}-r{|>guet#!no%HOSq;*X6iB}aQ>S+vinmK*u#mIj+h>Pn!!C>1*)NuZn
z7u`B&&@f!@nQ>eGB^rouTQs1?YnCZQzZs;T&gTpd(TbRF2?~N!*ddn>K`*&Nt@%N0
zvKX!Sa08Bsy%4BMnDP&}=r^4^RQzS+jt(6Jg5<?^<>8E3IjB9=fanB91hH9Xg?%R5
z!125TaM2;dBEXp;3Ei_wmB0hZCO3e365vR|p^ER7P+v8BCwvSnQc(5AT{Xu@Vh<&f
z&qO#Rv)nh+&dLNekSDxxsqY=mLQc?%bM9S%oYJc@xis|b>dT4gg-1G$y=F-t@%#<9
ziP0Jxue#ed5V{(WR1U&T!rx4-7Cpjax9*N<%ATE-{*#+{upaz9IBB7~<BYeKLGS24
zg@3y7fPVn&28-ruB4&w7`sRo4l89s=rAR$MEP%Zo_t>FE5KRQv@2|ZeCZ7JbTlvc9
zf3uD1RqY@{d&eZalJ*WXYCpeXkGQWfP>@y*Bixpy5|E0tvw$lWY%!Tr#Xy3`>L_4(
zGQ4MEm9kf5AAs+$&vL%8#D!`8Brx1nN$6nk%WFlD3<0W#`Si<WX!ym}9|DLkC37_3
zh~9<li)yY$^2~TRsB8KI5*YaaXQi_|<PJ?cu3K6P=`nZ{@BOn~<aq8DpvH7(r)We5
z243@)J9<BpCZ<i_Z&JUBO<JdQc`Fll*m^r0WOk}O*Hh}eGe_rX)#2H;rtjy615weL
z45rGzpo33U+cyqzm<P8vF-1nd0QcZ89G(18A*LThg1++$mjA?N6A3-;T*b@}UDi7p
zosNa=2}XLDG>pwX7@ZPWF9p{i_cTl`zSdi?+#G1tCl(Z2QE}i<dMX4E9gtXJYLOga
z6@lj~D#Hl~5H9L=5sbLK+z?x80(IylQSo8!{XOqbhUkWtoi|nd)l^>Mo+LkSQE>R~
zXg7W#CwfEe-0j5n$S?b{lACAw(DVKIhOH#=rnIZXF;yPItArPmu5f(JGT!|dbtL|z
zZHn{Ny`+XQ@dm}1bw;OLxBBe_UP?fDqZn*YY^5ju{KM;(p!8=g;#)Ol-!!Q*5kP89
z@P*nl*<Xs@7UFR!9<Bts=f2g=z__)AA(^FK6C&JIQFW|od=melJeU2vz9A@Pe6=&e
z?br6JDp~?cV(mOFK7Gbb|F$2#F7#j(AVA89snqN6-5vFc5_0NoQWi--D!*`WKiIP&
zfmp|%TgtFo5qBuBL*`v)Qa4ml)E@$+Kmf{Dzwo82?GM5KQP|a~1V4yQN0K))>ObNh
zQ(32c)tG91koyf$nRIjgb9{`<d2s~Uywbkj<TP}+SSPz8ojKY2*?c;)Bj}{Nvh7yn
zv15-Czcc1mo31};by=fXJ786=d*OCjQC1oRd(L@4?WVdJyu&7)UM{@-o}VGJW4sq6
zliY$bEfthX;UfU*f(j{zq6e&6fVRcw6-DPlxAe*mz;mCi)2$eLp<Cxf6vnwo_s{p6
zx_G$!CR)}kqYLU!IP)3TRKl^UVgTQ?4VZQOi9+2E7lUIbkHU)}>~u<%U}m992WVX8
zh79YIhEcRFa@S=DI5@aDv1eFRx2hoYD{I_Ilo&#KAz!g<MiRCqToG#3#p;wa3ut!D
z28ED*fTY5<<QW1+GMo~OSUG2l<r2r!A%jTir2qPrfUX+O6^ZP(OD*0AtWf<J{@NuU
zt76@l#z0Uq%XI{n=rFC&a<Eo}*)Q@)_bm!h#b{NH#8)#PcPbBGeH}CHSo(AO^G?8Q
zk7ZKv$@G)1_Wf>$1b<f(Q8H&({zBB@{mx8RcryH@;YZxj-^EJ?(9&=4FeTSN4d(+3
zKebdhtD^9(sAwC{esWi$N*Lk#)s7vOv@=yztTYMmwA*7JJ01S3))R-M-C>=t9KCHy
zyBqZ^@n2AXV2=OY4Kl<2P+2tXUP7?0ohDHzVbWLdrALGo5lG;<!zS@aM5-oi%sOoy
zBR=uB$tsf-1%$~}`HO;#kL}1AgPe?w-hP9FqW?gyCCpNw>9~?9KKOHS&uVw;%6h^5
zGJu;AU*(=TTYteIU^BGbw+w!Q1R%m>%x55Bo@uO-4?y2=mtAR~b-63>K}lSr)lB$R
z(EY8yE|S&_W#J+li#rWZ$WP6}r7S@lS12;VlT3Z~F@qLO_jiGPWOk~4Ck>TjE^Css
z3MaZeVr?jgN@w3C@T?^K{=x41JKstMfJ29cR7mV8(Tsx1=IlNO4Z}VJ6Zi89t#CYJ
z))f`rm*?hMRVB62HRs4xOqdF`fvFMgnkcQ_dm%YgSq_t+;$z7K<>!0+r^d_UssdbS
zO!WS#nX`w=P|_*rC#2JFUup;NZx;-8%Hyt{+u)`h>kjFN{F~2x%x@RG>G8Yumqp+w
zN$&n+)!LHB-M`=ajOleBXoeRod7iMQ=!<B1rQQm7WGfpr|IWGCQ_}A$!Z83EyLu~j
zBh$j8_ahF<uIvhxsR$@{3_`PkCD#nPYOd|_bKbU82V~HDK)sH(tK+2PXwP{4T3Ba(
zD&C~JPBq5bHr|0>byH90*;Uo=3ZD+nVs2nEnQPpw2YXIEI}>7e?BpIi471Y94OzD5
z5<qs1hsQWn$cd699ATYbsXy>1c~Q)sqqE%DZLC#hC0!$GqHzEY){e+{cIu7yP3zG|
zMg~6Kig!n9ll&*#<bzfngJWk6h%c0AFFXHwc_~Anrn}exaOK$M1)K`ivp+tRmAj%e
zM?c*}JbwMpjSb9|Gm1BF;_xGyp;?XR>9CI$7HI1U-gR6+^UeDwah7KTp;)x{Ss-Om
z=1Dq#mnCvIr9i$-hqCTLRJq=uqmh7=tnO}mlly4Hy)Y7Io>Q=)GaTwXoTOF^IJB_g
z1CiY46kS39(5LtS@+F{8;R@+JV-moCR41&>&6CFih2JuO-WL2J0f327kL=oPb*6KP
zS`x~y7hraj29Lb)Ng3c!p7fTq#qbPLaKT90<9SX@J1{~5T9Xhf`hSNCMf-vXEPE$Z
zHAxLLDC&##4cTF_N$YnnOuj-tx_)ft76W90jxEEogE(ky$d9PK@ECxRU&#_N_j!rT
zGRFXstLCU<6|(y=`;jvHKVnE8#)`_1KKeV}?;1gVDe><xYO@5v9zwvn?GRN8u$4%=
zxkOJK$cX7&n*k^p&BB|3MIr2OC1YttzeSJoyA5wwNvf2dzGnmMqwluQ(l?h%t#ib+
z>4FC><s*0Ou0vm4srH_5{o-P~_PRe(i@Z*|WfS*4&Lac6MqjGPJdC$_-3_wvARWz_
z24YA%YdxzueJz;WBLFy6v)VCtj<8=>?ogEY^7QwatXS5MK*(h&Sz(2lG+i0_Hf@tX
zUzNYmZ#IO4&O8GxeiP?%45@{}km|5k6}kI`zGtA9HJ*vX9zRJ3e1e1;CLQi~qpn|L
z-f|a-Y0O=e0(EG?zx})yLxJ5sn-FlokOKB~<$=&JeJT2NfC)@}P_Hp1L_beTdU9};
zqOik`lCxt8JMweBvbCRh_8Tt(#IaA7?!*ww?SqR_<?uL9V+|>-)d<hF7@}|1PE(1~
zx=f`!`NBmt5comk9+hvbs69N7dBE>#56^8S;d#**zK$&B@=MD5?2G-5QlPu}I57^T
zOimV~br|1eh>|11$09C+ro$5tn?ze5vKOkk{1(_*8oE9&`w>)cYY6QAuCHNS&geY{
z(DD)P9Dp)Q0!y(IWDw2udP7hVCEnc-t7m4zn5_zX3pm{9cDG79{V2BSTD_aPr@fNc
zbBQg~X^f9#iIo6QJKeB?C}eBTeNJU`y9s2ncDc_h#`d+;)c0c*)8xzI@Z8F${7^|P
z<K>0f4rR-;X#Gae3bO;+NBz0feEC)DxKvbUUZz9GS7-y_hNtv*6dm{K%G|8MW?tmI
z@WG`6PWdO`d#w(9BpCR4K<Sk)WC80NTayW|FlfoarBKf%5td+Kb|s0jT%5YoY+XfL
zuX}uQQwS1O{qBIMwexCUU~6ipUV*dTn*#Cv@bc6*m!tN*Y+g3Eh<ZGF)v#`49So&c
zs3}A$zpAu}^p*8iMNb+5JG(x5piGhB(*Q3(A1<?MbjGXl1Ev_6j@y3SW+f#Dd(g|#
zd694HIzj{oW9S9l{8a3_hIx98T8ZL(2_wKBKtridM2}Sb>b?fg!)hP<5O@~BAF^c$
z)v75Adn90vJN))~K4`t&BYvxZ)6*X|^y~or>4Uq*_wO<JiW`r~esxxkeBY&^njIzl
zZEI0G-AeuDO3;)jNu^^GWj}e7Um^V+OA|xzv^8QV5QJtD%~j?6dQt%q+%}1rPW6>Z
zD?Ym(;FMEv*NP(Q+x>*>_`m6O8k|!{Kq^{s$Jru&0RS0zM%rIkf?I)8l|Hk{uQy_X
z`cJ5){<B4%%i53kP+mv{dpi4%4Q3j;g-oL-$<a0a?y+%D-=)}SsUFjI8w-PGxp~Aw
zS&>doNkd}b|Js&zf|?&x__m2#5Tk?HVDlx7C8g}p7wcSn-qh1)`rJLV_M_|k5?G&J
zjNM>CnI+y$UfLHjWT$3>LHZbya$N*`fdt_5xidflFjx?Y40J43H5JDwfIz%k9X*PV
zaQQe}XXckqP)S15)x^{(JDjeG{&lMF$=&b|_6RTU@o){zUMO0kb>6YqN4F3&LwU0~
z3|eL&)t~QUAKA0LPyQvMKPmC{*J8|QR`~CeH35RINynnjv8`X_`X&CrI&etJsS}7E
zHwIGo(i$RPRPj}(iH!J)8zV2{FSXaKjuY)~-I1l{PssaiultFJMTpsKgV$@wUTzfr
zlrWm3jFNbvRpzWKCR|v%1<NHZ^gfJER|_mH2onY@L+@AouJIU&mp`1HaM{S{8Ivf5
z`=;A;=U7>Eho@L0eVG;!TKF3By-@2i9PC!(V^a)B=&UkvZgmt*J4wpHnF~IaniMe`
zp0+AWbaY=SK`#>t*38#|NRS#cXrO?xMyOL0NMq(2*7-ZhnsHO6cd5<vcxr3q@gZkw
zFcc|NG%5!hp<bzh8LM3D!=ZUPjBZ!ml_ca?zT0nmwzZ^Fvw^nVp&ZJ7kBtGv0<s&K
zts4|4>Q(r}2dix4p#rx_8n#yH&}dvD@n$P!%e)mW7S0%b4`jW8)u|iMzcmns@j3;6
zw9Z8_&o|||q}lzz;VWbjs!J5TNCtg(i8&|!nI#bdOqT&Y{NuWZ_y6rh^O!I8)7%Jo
z#5^c`-?_$5we25I=%f=*G13#Nt->{ywVV=`T&jo5L~@v79;7y;!W}%oPE}6wAIBU|
zK`7+|Y-X&`%-A73C$)9{1K;>7ykgQwtSis#QSoPjm|c$3&zU6s!ttn!YExq=A~Z=o
zIJZr6qFjZ|iz4nZ-Is)QKJR$$w86!@X|16zl$N^tCtuM`GlHnG+}g~9kj`ek8CMOd
zv*pf94NO_nqLYV?n}l2l>fhT04O*NyM0(jZZFu_Jk>is}ZaQQ3Vx?mtU&tjkQOZ+N
z$=X}0Z1kz)!t~FW>YV50)?9o05ppsniH90GJL=VBzxahHRJkZ<!Z4O<4s{I~uT~e>
z!6kf?oE`mz!Ny5bYSVlR$2OO}xP=&s6IB1TiXNZ2TAQa}6CFe<oB(LWLxo;PpEVTP
zL7xIzQ{v@4px}MoH;0c`)H<y0>;T*i*u>bf#TKWhbljnw@R8qqz=;0L%RciH>nAX$
zU>|iwn#eWQbb%IO37eS6zUM{TNQv`EU~mB0*B>_5$-Q{cb(#snoM=nAu=mQT+|#&F
z!Q9CtHo)L86%H6Ir)C4@0k&KwRnf_VB$MB4$db7Bsy?n%CjWAJ_KyslkhncU(g^6%
zg;~=}lCf1Ojb2EEkt6NPVU0j1qD|B%7~n&i{ubKXK3=8o#vEd<D5c$CD&DU%Mq4b;
z*71@bMHR&`eG&vP2_GFqrjTuH5J?NFjb#P|hTQvj%c=y)bI`PN-|l&|kJH6>lyZqJ
zHHN(RM|H7iAR9C@3czT|QEGT5Z~*uF>__HY<-z}kfF2rHjz6c$0BsCLqQ)!lI()>E
zNs2^{z$qt~*;|2U;C!LhJLjSkf()zfr`ffjB>eo6&|B$260TlWyDZVgL<fJX0%LH{
zgrzp4;AB|veWmq1pIa->+Q&N(W*b=cU+uV`9DuaZnhLh26H%pKT!@+OFVuKQ@Xs?F
zldn9T`u3-U#UlYKgS55sHJ3SkEyq)InNzhrQm_3lUjIjMNnh*|Q6)<d4IFwNd|FvS
zzVsVsF(Lw|=ofta3Z}9UH=L05tJi6xh=!99W^!L=&}yy5H|!B6DI8sXJnQ;k@n3yN
z$oM8|;gd1g^CG$Y5G0DCY8q#MxXebc+fupP__q%Twp!Q*GU|q{GmtspO|QcTLH-#L
z)0rvuEb&w3V;`_?Lx-Ve#6FE4yx{h!F%wlH|8f~LBGE)jv%uKN+F0ch-QOzf+(;%$
z4B;VwxEg|&YA?WKsCNUAQB?WDx7w?%`jsQ(qP()-CP4@yhG{cmBUru8jJCCxQ;;hA
z%-G??%wt`D{tIFu3@-rgFy5-)FL0`vY-BIne0rNBj0XCVxKx&_k5vJQPOkwHb=CJ-
zAr<XE`gZ{Rh*3#q;EBX^HOvXu1~;XbAkCIzI^kayo+@)2qgYXD07canXh-t@^G|w+
zfrHfr7-LmsfgFi`V-4>5rxQLUQh-_5flj`D%!_w6?4l~!EL4xeR}y8(F?scoV+d=F
zBG*JGk`1bSK{FE$IKP4rnGU+gZo8<+ByIjblFmJx>Hq)#uNiZi^OOlWg;6p_BQ)n>
zn6pN`g|w|wqlDgSOT=<M#X^gjLslt@N^g~8HHSGQNrfDeB&1aD&+qyE{{GZ;bzLsk
z>-l^>4)^=*9;|UYEc(uo04vRyXYs~o=mi08uLOY)G9?UKZ}2~N?#jf>n3%zvt>W>*
zGg=xOdQKJPb#vpjFTH;c-9bg?re<vHQPsa>>j#TSzk6zmpPL#MQ0Sj`hY@(#8jtF5
znB<Y-`R+#5s175%hX2R|vW|lq_oiDl<oco;ckhnc+`$fYXiZ>s!o1JQu=d6-p_8)q
z-!eC_&pASlu9xP?Z<fc?paAubx!+E=WCN|fzeIM#o_`p2_C;Ii7;tTH#$|fCVm=@F
z``3^xYar+%a#vYFSwO6Ng*8{Iz0Vz81N%Y7Cann#<WJI=k%+6CWEh(?4L}qYi=wL>
zC-~ARhrx!(c8hHTdf0^(%v!^zW0tM=nzb9`++-*S9nj5W)o<lWjaS7;XzS5a9}K|d
zoc|<)>l%<@%(Nrj#fn>emkFh!yF5m!TA%TC<kEVb^%!5(x0hc?ws6=9vLmEsOAQ6U
z7KcuH`zlY7oM<WBb|+L2KS(zdzqq<OJaq8}#rlBip))y@J;C~2xMtja)fBXN&S>p~
zH>78&Yy1rPDy!YNymvX&>YlqyRu%Z|AJnnWcsR;f5K?rresIO#dF>`SZsBj`!##&&
zhhWtUsq>0MJ6xmI|N94!w<Z_8QTuuxTsT9rFjAC{arXK_&*p<^b=nxu_>Psp@!%M~
zKnJORpW+?&D#s&y_p}SN+lc_ApqkV*5SkqjTX9mMSzG5^;k!8^yo`*W7C`7Hn*i`c
zkk^ne3;1sGIBp(;t-3c@`Sh0RqjHndFQ?YfR)62ku6Ot!w6EOc@8_;b+4gN<_IEG1
zqN$Y|>0UcFT@4w~c;6(%BEDLX{PQY?d(HoLvwE|8=jjNOf}rPl1)UVJrnhqbF$Zgl
zZRL%l&^W0(kA4NjB`J|)1len3v1wJ^%oPH%;7zp)CmgS3)R#2N+>-K&EiO6iD`Ki>
zRWzO9N=Vva+}a>pjqL(F(&~Leo%)X5`Pk4c`f6Yb_v+rhhpOR8>K6v8a;i~COspgY
zqUqWCdt;D?R8>w=)pgGKc%-&1pd9}#+QuL`oylG|evGMf&-)Z`&V{n0H~{3k6lxj2
z99x?=4zlhPH!u5V@!dc`-Aj713(ixY61Lh+r#fzm_UwCJl2!l=oCXQCz5nS5N^8-w
zd2AM#&F7gJ+^~&$$=w_Vb#pI6J%>K%F$*-(rAb5SJZzRP2a!HrLIagTGaOCAA+sI~
zM>BQ)jtiXO19#r#Ge2+BxZ14+ht63W-2T!Py=!c=uqV&`frvf)vLq^sldo5nOpVT)
z7;y>mEeJZ^5&nL+{d&LNJ&7mH!I{z5PsiJNMf4SV${d?nue*Fi;9dz?_)Zvnb-}Gm
zHVP*V-ehj4{I#=Jx|_ay*_Of<9Zq<q96o&0CQvPJZhNYJPPgVlx-MQLu(rxH?vJtS
z0T-DzY+#JW5wYCo-Y$gMl5|iQl-K3kMgx>+3V5Igbl%$debHVa$hPdEWV}M6sc^y9
zFV`=h&gl0x;jXeY)YD9>&>6Gb!r%P6hdb`_GF|5;O9h`UAEG<nU*68Nvs>5^PdlrZ
zlJ8m^l-IknTYtdJBo5Z_b780rcMrqbrt-rtg1J>|IF1b+ESbu!FvDb$hQKC}#1N`+
z4-H5EHypzNDdeeeG%BeqxN@QQ&ZOmc)gmu;_AG~;(BLKMJ6_Vk12=4d7eQ%{s){<n
zwd*<^fvw;3oiA|)Zz+CBRHAsldH+xV?tKMuHg9*NL*Lwt|0CgLzwn++-YSQdcQt!h
zx&5St2g1wTCuG8oCJQ^T#mapBz@DX-8WZVEt<sxdFygtZrG+M|Wq+`FNb>Z&^cr^V
z(Kl)F-;)(Z<fHgw8gt`C&s*c)?N7a}$+QGZc@J(C9LOngy|N5a1QGC?4q9(#xocIV
z-uXS8KEPIWXje6Y$U1REO0IN^zb9F2E5DepBRk|d1mk-gN)pGl@X;9N(|<vJi<ugL
z>K<ev&5`STGz3QeyX$H*-^+XLg^6FOeH2~ad%j|e!1?f*)?GgWc3eJf8KgMSHs5Ks
zB{Qy~iAh=!Zufv-Lepd;XkV)4Kh+R=y4rcH&kNj)F$vu$rb3u}AvYOIkVHj}GPi0l
zf^jWR3YA{rN_{83G_`#1k~ax-Q|&noJMa$kuc<NaXO&h9U+3z*7Q3srMtMTG8l@^O
zUqV~4c~88WzY21tz<sOIg|kCD%K9^6_JpOgepTTca9@lE(yKKY0qHMBr1K+Z;i`wW
zYN(6vXPeQhyEjF4D<ml?bbBWp-wENrN|<^Nkr4NrEi?vn_6{|6;~<zrpOIplD+cQ6
ziX%vw%bD)YB#YfT_bzq!@BsWu#qXiJX=)3^DCUKy%?)beoz866wj2L0Bm*MfQS*Us
zXZ_Rui<bE$^#xd2*$5n)O>uITo4g)VCNfIB<K;R$kgj=~_8hQ-bkWAW{cRy`=o`*=
zUY;o0N4dK1q2XvlZ-Lu}3(1YNIpFRQmoavx&-j#D2F$M3G4fS`V@0!bwovcd3Ky~c
zBay}w#gPoR_b~wi$a;etMT!MduNjeAcQ|(>AY!9&cLWVv?o#=?YP?A?*kocv1?Kr*
z@j}HY0jnDZ*bRu0pwSTb8m$7okjBUn9PB{9#*ps5$69Ochc^{}{;O=F8PH<vAUrCj
zA9Qv<Oh4P`n`7)}`=Yv%%-3oMi!3X<ADTaw<nn}it=yP=%clnCnw)TV+$$b4gOvB`
z{>AqH9TnXmeWf{~uOvH2Xy=*~z^M#xStOmms>?xV)ILR)iromEdpVvkfjy9$q-A%S
z)|lUUCvrRuVh_xj?343t>__w6aWY{{lV~YNLtIHujDhUa(X$mldFCp;y?i2`FA-}!
zRo2wA1J#y_l`Zpcn^%N;_slX?^v28Y520Xo4MJIY#sSTWsv;QYI=7Kua04c^q9KO%
zH&#xTeJalEU`kqc(KoujuS{$&RNlG}=tU2xNL8I<tAQ|4r4l$NJzogC#GjE2hxR!X
zRN77XO8`gyV%v3^sT03ou#-&X%=ZjJ2M_;O5<xS`FFBDb{W;Qn^68DYsY!MZmRz((
ziX1Bs__o)`^bDLvAw6g~2vAnZ;v}%UGdI{AJ`SL!sXcs&cu?qBPvhDzZqDAK{bTU`
zdpobUxO2YBG&X67mjl0LX@E?vdebk!<xr#}B5{h2D<rqj-RZdbnY|}?*S0?sQ`sB|
zUO}z8|C>=0W~PZ09hdxi%f0Ec`yJ_Kxrk3*-;y~H!fY!;H7Mjr`NrfyFU)7yxa50G
zoK*uQ1AU*hmF~<<Huv(aFl+r8yxpVN>hJ+aZn3xIxB8=}8OPpSFDl$T!gl5qAJn&g
zed^-*CX;irJ$$k+3&eUH*|w$ph0L^VK>bLG@}=r`gp2RI{{QHdseZ5|YJ(3aRTMtk
z-6p1X7&o*Fi!?twgnZo8e&~v*<ZwG*{J5p-P!1(tNp{jWoBr@!?X;zbc3}SRilPHI
z^<(B1f^sju?YI8DAic(i%&bx8PPK;(atPupt|t{hiRQ{+VxTAmiljGSVcc^8#~RYX
z^RYs0J-Y;<#bmBb73VJjq`l}3A7>(_`A7e9$<f#FEAGmj_;PIMn1}KEg|H)@h#B>W
zvc2%dYkei@AJIK?l0S#GcuU0qjA6Uh+S7A%D68ZMs+3FG!*ED${DaBwGxWRCH@qVv
z*0>?@*^xGShF+ok+331CBrQG9M>r(#-im}~Sz8IxS_upLm#;*4K`xWI{;<hbu7)Cw
zx#t!=nyQG5P7y+g3ye;&t+$3yrOO1F^R|P(ZCAj8c(yR^4T$wSNfUy=djH~1I>jEl
zRu}2|w(g~#SVLMvmXdB<Z_O<fG6>gqX%YB_!UwQ&&H!z7;L}5nlb&tB@g*-frf=-f
zI42+0NwJ7}{yQmP(8>o0SWTUOs2)8p(;Q7I--cH`lsIm;kbmT$#$U(Rj|FyWzquhV
zWQ(_}OJL=B=l^X6o0)=r(*!vK!20<wdInI&ybN(^uDuma#e?vKOPUyL`K~+V%}H4H
z$FrQuLqw{W1&QvWYgH?h(7CkeWoM-Jly;FtlsZtqHO2@H-=lg!2sfU<2hL1Y-9MH4
z!CgO=Ufi*c6Z?7lxJDTx>}|)lCnm0l39STyHi>;lsP`=_m#*2d4o;#5!VOU*J%KTw
zI*@gy_7DAVu=mAL^S$AoD*0o}K~=pmUQvDNTc*QpM$$`8>vCVK%;{A~IG9{GS$WYs
zbtcg%ySb=^$4wDEsB2T$t=u-1h&7*U3y!HMYV3M`N6X7)<l$+qYjau1fjNzUp|HPO
z4i+LSg~OcGz9c^wHk*#I=%cY<8Y=|bRni3)+=pdR$tC!DwI-I3`nzr)n0(J=Zjz<@
zs4xfpRZ&CeL`k1|uJGvDNrc?~U;Fh0PEOluKUKJ1ow@$3$9gjux7~2EOf|P;4UYo#
zK%;c#;C=A2u!j;BIS^j2TAK@9^B_%YVwjiec-;b>jB8z~zZaOkZ}>E*gvX7pxfHQ_
zS_1#Uq%HlFWtLaBvS_VF=H-8v3ysf#A_D-E5*~1z2>KCvxbrMGj-Y5s7+d%;q@<!z
z7SZ&1sv~kAg;aX%pu+uBo{aECKq+>|ujaq<<Hj$9ynjM9mBVU^@q8J|#tH-sx_$8g
ze?~#OmO`NCs#X^UTJt2Ub@yMhmb5Rry88u3+Jty9Gkb0EXx-8_to0^&<ca2%@|$i}
zTb!EPOkyx2X%51;=T?VnVj0>Cp0feWiX%x*2ADA?QqwcTHpooq+tdVsivi2mYU*&C
zJJM!Z1Fx$I*E?FYy+d-E+%|Kx?<>Pg+vIVL%>N+8-=BD#j_}{DmUP|Lz;#;4Z#GnW
z@Y|73T>@=gbE5kTzqGjFA;ph<NozH8IOr>w4Ud;V_D!X;Ooc_yHz=mf9H7jOW?!Q?
zsTf?rNm)3N#Z-i@9V}3kM7!h4lQB}hY&#)*OO6M0RHH1|=epCwXqy19*Xlz*vb$8r
zxk&+2t=`2tGXcAb4jVpp`}uobadwVfXIwrVbLLd{#y^{DPh{>~XzHRn>F(;s31{{+
zW95T{4$1g5;aJ5Z8EP>jyOS5aZjjn#x9BG|Gq;bze|N4VL-*I~#h?#iw{bAL_BYa<
z<P#4-Im+ZvBDVKgbCIhAxO{TG?deW4G3!N#?_9@5u>HLVy!fKRGOfJ>+G=)n3f_ra
ziS3j?+dsgPFJ+fy`>mCS%`5mvKpU^Xj#p=G3(6Mv40^@s<>&_V;cEaO5mI-g!Y+Qp
zyp@2k^@bq1gw?MK6E7e}9Wyq-4Om#M-zM}YW#}((nxcA(oq`Viy;1LcXNHt;1AA{)
zHJa3{G%GNH-77T+Ts+S&-Z$BEW5DY8*x6a8mORa(-=I{laMfLB4$wwkuB=Vvav++0
z;Zy2=n^`b<owlPf2wi)c-NR+Ww&+zGNX;+3`l!m-%ASpW)~SO7+2zM*DA3SQcOh}3
zJahBJ+L8frZw7fU*&G2h(>Pd8tvBU3vuu~uQy){(!*H$UdJ6h}-9V_#4CH{|j-2Xv
z<fMOoXVp5dykxRw*G^@}S?jHmlU!@WeREZJ)o2Mf>qxe1?ry?$Z%N&T^o&KIz>jaw
zH;Jn-o{5inR>1DPTdChG=C;A9ey6N(g_9ECyjr{<s+_r&6scFHeDO@f{vWw{2T(ct
zP^nohiyqyt+K!zxFMLbWCbjV0x!)o`9>n@?F)YTVIlm0W>3#tM<i3{rZ)=(u7Oa~>
z#q~xOu67dJXz6D>0=bTP1Ot|=>q)=>*I?V6#xaw6UJN$+%ksNlQpKbx?jL&>JnC<}
zGblkMW31P?4lQ3Ti0&F>Sh3zsOkDP(0hnvtC=)qdZFs&K+96pDS6Og>CxEtt$FK=$
z)T8~LEb0CMZ+3^|=HGdG1qloBiLLa;OmQ2j+z740X9T5Y=eI^^S(%I$z#|?eB~!%^
zmMUdKVrb)sWh;a<=Z7By;rU0Cb&P>$HbYsw+wPN9yw+1*PsUY7r?w^__-Q4ni=U;X
zzdp+}mNxA6Kfr%w(lq_8HuGP8$<5$=KQ~IlH-J2S>2|xmAQsPciLWKj<#EC-@_N86
zM;kkO0;bfTZwJgkX{CdqpRhz&P#?Yh!S+ej%UNkUEmNGBZogrDliar7bo(E_KQ%mf
zBt7cDsfNX=UJJW;DuOQg$}K9=CpWA^PVoHP=kukte+9Kh4#?aL*s6Ej_2WO8In}Nk
zuXdV6)kJO8u^=XzzTO-{m#jw{PL@cEih}K^uqVlwcEE;|%}(;qf{a&2%N~hQVX>gc
zPYO4xvR(qFb_-E1tP0%9HgTN>zzN!^F>u~ke3yZb0McU=kWP#;^*+%UM1s-WEmYcW
zitq1Q7*|DGYyE9P(mN@&>=gF%yBl7$UrrI5eg|6*bky<=v`_12yZzjZW-6MQaM%tH
z1yEoS`s&SXT)ThNFUMgzm~~x=+{pcH#s*2^vO3~%`zasC%y!a~g#l*_Le-38_X%ul
zhI_Jo9tQMJ_F-5!Cx)7UsQc3mm6m?u>WLSVbuX(AiE&g+Sc@JYlvc}n(|`J496`KF
zt@hF?SNxFsgumh=Sf_6G8gwl|w62DeLQSItJyT3|1;OrVf`WgAbWKW^!Uv*MtEA<*
zK_GB!rAv7{lnvTBxP>e#Aatv+NJ|7{kGcN>GV*k-+IOy!61w)PKsUWCz!bxc_P)L`
zJgjsd3c4J*wbdv_f|AOt-*H8k-W>DVjy`zR6w_YkT%xt69Q~J7wB$tKujrS@3FhG^
zn?$sHJDYzCU)ioKlEnM=mG7|wLo_e|0+CpjnmQS5BlNbEKI7S|THEZ`s&DI^Q*Bp0
z?xy;Zj^V4jfJ}h8|GP1aXqMZ6*9TZN!M0`OOiVH;Ihx&?9;%{{9c9K3V_w+P8~0g-
zvKB&}J2?-If67-ZlP|UWM`^-ltuFuekj1(u%*dQO*K(8cWw)uLnF~$wbVO9Tgq^nn
znvdsfd6_iAYqjo*+uy5oF@9(F-rc(Vt#^-iJFt7JsCi9&-T7XgS|?+@45%4F<!pe%
zZCVL=#rT}NLK#XbOGv#QD8Efr=ZWe-V#9@~ev<zZGwhm&IR4%=abf_@=yLH=J_MQq
z?^t>hzeJI9*gRytCWYkG?M*vN)=5}$8)eFbk;UF`&gSM#GU4pKelQ38)P=J$;ie4N
zFE(Ic`WM|Pa{$&p@1@gcL6Bh#h>~r{V%rZ=trs~6O+qhN4><Ks<^JLzOc@KpJqtZt
zXtu}|@aEqnBq^Bo;y<<ueso**?|PA8RKw0s<qlx^oZ<i2UW6Xb_5WL+a+o@HpFkiL
zOFZpH>l_=pU)`ZKqbv!x9eu$Vs`fO`)B2_$(7S!ESzoc}Np-it+)4LB9Lce1&Q?A3
zqh?(Hx0}<BD!Xgud`so2id~V}Bhvb(Z3o2VU)zt!X0GHKDIMsh=WPE2mYlAz+HFGu
zNJ+T!JYy```01xTPXAJJ?F&PK^+OlC7qGZL8_ItcwyjonSmE9@e(GN3Sv{I7Af$Yr
zww@~J>2A}>!mhaXdmt=M92ibb;6}y|-e*GxT4qF3tjj4s2T2l}WSe!aq*#(xAs1{n
z{GUvtTKSMs4_876feD61#o8Gze5GoI2Pq_35D-+YR=#*&%5#Oqz|d@(nTkM-2m3=i
z*TNG!BQ&MLB4!V7ve%O3?e4LYT_QuHPB@RkQ9+0)C9jyrZ|BAS*}P};_Q6@U69`AM
z=nO76nOSRn@S%?`tFYYlvsgg(yBGh1UOA-wT_*QFE^i`fK=)Cavt(@1de6;soBsMW
zvw(c={JbvD&%m1z6m9O?8DOnHRkB48X^mc@to)s~y5OFLj@xdv)cP}y9CqHAO1ii{
zm8;>er2R3@&!sO;2q_C36YPhfXRoU9Ko{n&Rs}}viYYg#IH`T*_}Cf_lsSj{QUm7O
zW1GXw*<3m8;Q{+mQ0Wd1?6;p^9lh0MPd>VZep$*VsldK44ZY59?1Y?aOzDne^A^OY
zC}r!~o>?G+a7+T7M@-VNLZ=iuukPe0di|a{L2;M)L}}b)gY!9hl=;l@;9WV-ecJqV
zuE2g$cc=2*uO++=ABpV?>+WU{%tx26u)W^(fx0@I`Wr08X%`2uyiNaY9D>=Lttf((
zx&y+t3cRd#Y8%$o6+a8Ts<bOaweJ{+pmTCQ?rW)=w@nNDmyoA)15iY8GWrE@Z9s4s
zMzXTcvVyd8r23Na*I|8cS+mMgeMjvbW4%tOpygA)YgruFl2@C7a?yDvZ{IX~Gj&Uh
zm}WxPQ4>E={)hd!;hRAoLD5-ZGjpt^kJy$j%#GoomnHKr8k74yEK;7-9=Wc1JsO<^
z_kVbV$=Kg?T#@{`*H(sOO`NJO-k*_SX^pRgWhO=3UEXSKaC2MP$anLc=Uu$anTQ9z
zv)KNu5<6NRQb%a+;>#)^$shP?B=(ol`$%@)9lcH{FeWbW-7Qr6A&rxC<ETa#J4#6g
ztbO*1eWBFB!uSR7VM?;ap$(xXm^W#NzHItI{nl02r<hx-cl%D3z;K1Y6E8ZgSqnR_
z<W4{&%x2EA(y2J8^hwoCVAayH@daDZQIH^Z@6*9Rhlv46?wG4?s^WPW7R4Qi<UKIM
zC?l>&t*{FRt^pDdF=N5jGtbfy+zO{G;T1I2K}Y%){fZ`7O|=}WcR4t-Z1^rF23#p6
zx;&NzF|^{WzrIw@Q~|2vd3C=4o9%UG3pQP2gMar(9e6=@Z4UeG+wMz9SDS8j?bh2@
zZKj)FHIoSnp|1Dp8Lvi}I-p^hg5}_(2Ab|m#L~ZBL~2A_P(x`IK21SgG^?R!-l8CI
z++|YK+(&MtjL7_76!dg2_)M4wtpO<qvArCQYnN;@;z}4~ZVq4_4BNJ6iJ??n;|K`G
zcn+Xc*iIE*tA=zoMKkIIo$t8EQzg<B-Zn!lu56r<=PAjX76Cru)7<OKl4*{EpIF7C
zh|DAr48P#vEpG$^nDbi>Ia8_&<9<_pwn;&j54H_{`is{f*I_$E<Vr=OvqB#ok^D3C
z?|DjetOWs?bW5j;iuMSXkLwJaT$!md+f-Lnjk&q~MR-)S)|@4CX?1mx((&BUts>5^
zb6`YKPoZ#x6h44Rt-nTpi21hkx^aL(tTKJKtJv-nrhlIY&(kE|2Ul5=5pdYc9n$zk
z7SA_oukUN=$HCMVJrclQ!g|rO+lLaNZqc*Gt&oFs8?D#tR7P1+(=B{0Yp!^|bzR=$
zFcGYq<E)W6VDXnvVV#VRYOCx81OPHoFH_AV+ia&ryz5eYrCh3*8X_u!9!ZS!`H#bC
zSQ{z;`U%4HzEml6R-ML!96J9|(2Gdv7>^pu2!r1X`7fvwjPaYH+Y2;T1JVzVoVX&u
zl?^FT7XO6cLV8&#<4l}Wlplgmd_Up)n5kmT!-KdOnuZ8wHL!)nQsbAM<|2&V8s{WU
z-p!%Qn^?EWWK>?LFi|1no)M{8^;>Y4oxMn)Wy#8N_JXT*tHPTh!V>%D(gLAp^Y)Fw
zKQH4urI4E5FS|ZAeyccFl%rc@yBgXZiB^hvGR0>*^32Q31X4D3yZx1o+~v3b9DPdE
zpTk=UEu)WbJ$3lgs4N4W{15(%$(8(lrjBi$VOlATbfv;h@$jJ|r>wM6IJY(At1NfE
z)p`CCwf4!n_VAWU+sweSnE+)ynVQ&AO3RnH(XkgTIRF*$>f@7}7ZdMHlzeNe&4T<3
zNcoSx#Z!tWE$%Y_-?f?L>)FFS9BXD17Z^6PLUMBd>E$*G;##;<7u;_894uV<IRv5p
z$u?5t`oRdPOQI`W?xB%3niph{uu+ZIWg<m$x=fYk-OXb1O^@H!s+D%rfkE1kVgX{^
zZNFj6!OC9xzp;{J47HyQIc2T}uZ27kx?+dn9G~#L^9WR6hiV@+L&AuM9Zq9q0dgMX
zI<k!{z6;1P^-d(E&Wjl=1illPne-&B`oem-ayO`NBOsl=y)x&EF#8z)7#APb^~$a)
z+*gEsp(R%K$C#-1#_*(zU3mNwA;UH+MN;7UOhe|ng;}#}=;TATiha}5q_P2>hRGt?
z0r6uD4sDK)VJY=$3`zaAAoRSekZ=W7I7XM8jmoM#RjDTGcX&n7`Mh?VVmtfL0$W<`
zS(_`Ua$Q|%Qe8ISWngf!F1=bLbWL*{jnu}&l0ZeBJ;fqL_^qyUA%=%rH5``bdaen2
zJXST%k{&NyAwZjc;OEvQcaGE6IzXH(nX|Nf@Uy@e>~)mrKbRfz6EfVSRrgFM_gB~J
zV2nk#(cGt*V&@8>+Bb)tmJe1}Pk;8E{#Ds17#P3e_i;z~u^pDFa5YB0+Iyvxz}%v3
zKE9ohrO?5U#6kF#tk?V9_jN=ZncUOkZg*6tHFAm4(M4!^rZUdI^k=`NbCzx0GP!=s
zIXQ?n?n*#^!fX~K#{ZBk=J|Rk1}WOrH*|~7a#XZ6hOU4CxX}j}8`JaZ2#;7FOJtQ+
z6HF!>v+a!lE7!$vI{DjQG2Va^caJH7`u%G_=h^IUx|O&(1XP*D=EUz@=__i}x9F-T
zX~<^kk8rP3Ru>PQOlX*)skkS}oEha_m*(v|#X&eIASc&%X{!Sa1)=BKS}i5d8N5C-
z^Kp(U*wZS%+HFR#!kBwz#WVHE#!>c7%9M<|EKaLRZbx*T;tTrkIBy|V36C_R-w`WN
zAPpe$|FNUJsa1TROci2f)UVh#0l2iZ3tVx<mjmT@*{vQCJ&;n7vDhDgV^RH6qe9ZE
zB8k+DdINB_org)oY~5(E1dZ~3TfY^o=rx9sG^DE<Y3+v99Wkqlrh$waR&@FW1S9O;
z_2}<?M8C*?1FQE4(m57k@|C66JSKYXivQYXC^$Yw*E{JJ)Aq@5TV<uNXToc{;eFNK
zPmIJfeNqhzSGyToKl)~`r3uOh2{oJ|Z*MR6hLEoc^t^;czpfW%E0Z_fR)dZ?y$0Za
zlY-TVvNP6;#{yc&s3W3UB60SCZsO1cN2L`rgH$42$bKq{)Lit?4+o#so$y`3dU^m{
z_YI~$V$rKVBIQFrq=xJ060xN<<n8D#Xuep^N6#FJI6Q}L{mBI!-<BS(_`-TMNSct?
zyT%otP06N0`vh=ZChQf*f0U_WJGe}cQm6x=u&>%AQwY-8KXBz><kQ`z2Co5u?Qs=H
zPdsyfuu_tJq3|BO=~8p(!J<CuCH85Z%Rr3%l$J$MD)4QFfCc#M05*$k2RV0Z1kx1r
z=fL6K<3~R#_4luqN{uh!Muani;~iY6O~P}=*``*|;I47I%T3ygs4M%Ny_0#UIrCn;
zj9f$H?ds>LQJ6;7tBU5(Usg(n+NZ0El&7+d?h9}gXM<)VrRM~|c#&SV#xJONi(z)D
zX$mb{P*B%2zV?mXJbK%D?K`_#aJ%&UraIjhC+jXX{`@g*c|$9-Z1#fM>KdvGZA&IR
z%cdJFf^}cQGRanC<UyMOQ=<w@i!E?q8sMi2B(MENM;IPca`SHCW3k|<N{D9<cWESX
z{Il1j9A33)+NeX;OCEM0t!EB(#P$>pik}&*SVO`(#lNGZNF#rC#aTRmc2OfxN3QjF
zU&rQ1<Y)ox5t8hhuuSfPL@V#u2-M{70kP?wHY*2<J}f+1GFO=}m5U6Hq=$0&xj_c)
zv>bxczP8}fmAH4aPzs0^5%hc9!nSwnRAw%*uj^PFTtq&9T%Zw@bd54@pgiGUl$GTJ
zb>0S=H~sVGc|ZNHJN`Avs`3+WSF5*qkyfs!nW~dH7rq&YX>7>^ao&VDZ{L28sQlYM
zsjg4Bhi|v~c2lwsKOK-D<&9U_`Jg)B)#4E)7lAVN{QxY*gB!j|kg-w3Y@0=Mo$-D2
z5FZrm5K5~FKTbpb?Uj-jdT*!4ZcAlTgGZ)uO~dOFS^WYlr*D2}lv6(4pBU+dy;KbW
zkB6U8YCk2l&t}LmT8n!=d9gF-V9}kir=dBjy2?iSjoV;B(~(b9iD{w1LLzTSF>5#=
zrgd{bymc>Z*(Z2`10~SozpQoOmAr*wTdtrN2(p<XB{t+};j5L#;@Cumcy+fOb2T%O
zZu2_SXjC)21-Dik0$5Qb`WDL_l?M3bz_-81<fn1IcO!r~fP7aSV)a|9*PRi|?Tram
z1csq2(b8r-MzCsZ)aRpPf9cPw-j?$`Lp8`;GyEw^!U}w!{}u8xrqfljI{LQB0nA&C
zrI6#EsMHXDWg9Qk`6a!0x4){**WAvxR%yF$U)+DJ@4?y94pm4?ir3R6^_y1SL-Vt)
z)9W+#WZ&JXT{q%tj3y=3^{d)?T`;Gkk+(&16(bJ4i3xKY-9y0<+Xh!9x;=z>u_J?;
z+0tJ;rMfVEuAi!~5h><T)2gSK8EmNGQ|JQmS*E{O^{eOHBl;sG#(KS?R@v#=9zK|}
z8tI>$<-s*+%#{j;8%`>WF%G!tnWwn3Oyu^if|?|tY_+S^T7!5;u6LU5J-xvdx1%DM
zDt2C`M6ijxl;=)>thuwSOKM9*@3RB2f^0A7%o0I5uL2{D=R$uhIP;H8uUXtPK73pA
z)B)q?>g|KdGVwH|1<?zx4$zvgZR%M$C=lY{oXACJ85`nzxia6>Z|Kad0*R?!*=T>7
z&Z^~hi>9L-#sulJ>d(48N1EwG(T`~~|C<+N-6yH4^f8ey)|tzz7*UGb`Cdjmam5hi
zwJ9I<SK@6Bonf%Uv2BU`RltjUHZ;p`+~6WKzAunsesUwXX%9!u0sjY}EWO-ZC0sV^
zaUBvh3(7#XsZ-pA^)hh~_kLWY0!r;n1yPBsgl_JH${XB}h?S8!M4X#sr6(+Nysd#m
zXrl^+B>Aa;zgvHsVH+HEcE!~W^w!c#p5shdd(}P<$mSFey*VJ3@3I5FRZo<g?;-{<
zD8+7D4!42uxuCc!0PH9!@Px^!>TN9;Wi1n)l?@Jqtv}v=eImhX-tMZS&&#Z#@w@eH
zLmobpHpgG{ZpLvje3sJfT7%0;8sX-@pPw(`ng+W)pVPfw_yvzDC0@6Fp}6;Sz4>kE
zmp|d<-q&!C<oer-6-D*f$eGBGxo1ZTx9uuC>#w9?s7Zf&=i?7vSC&*Zs<>hQm@q-O
z?2f#varg?m|JvUE(K_keI_tGMgqv1juRi<?YVcVk3fU?fbFh^k7A^zv&lZrnLV@P<
zE7g!4q2yzmIk2@1cwNU<2n&RZl1NNP9zpT;96MG_4hqdFMlC#f1G3z(_mkcXy_2!~
z6cylA;y%@ok~ex62PehGmx<KIHvN0?Q@I^>m`l~#It=pdP6@^PpnI9(sF*E##}L=1
zs=Dc~)$H0+e9)jwy{^3~y4m%bQi+b(19*JPB&XE}EYMYcEk#}Rp+hHyAQAFCu_nz9
z!+cMBQVTod`>+kdaUy_vSpdaQv_ra86OVF2uewFJ)c6@ia#Dq!7)9$^06f^GxEJ7E
zKzT2!isnk~2kXNon6&N4<x}kwE|9}{Q`B^3u=N?kWpfh`YDV_`2T|dlP9CiqTGy{I
z%LJ*8d7`O{&6>4uczIjJTF=gZeGMs4PE}l}YjtH&&|3swDooREF5KOY9owj-F7nnL
zC(3#`7f;vi%`?W>xv82zOW}B3)g^Od6KbTPY1`3bYOfvyOmR$O6Y3f>c|Nmu4U;%h
z{qJyk(hrNixP}$%Z?KOzT4{*d&_9TvneAut9THU{HmYOJ@k`R5XVY$qVGKnabdl}V
ztJSWq2p2!a5>%EDi?mhuUQ^y}S2v9Z+@|hoJXzZ|8AAX?z1DE+6iYhutcZg!jB{V&
zK)!jyeiy(I=lmCkAR*lI4|w*EprTnU{G7l5h~*`Lq1GpCAQ6-ac>K*<&!q1SUe%=V
z&%%Z}MGbCYv+22h17-lmfINhI`v1#(a|7D|Hvwdpw^my=AmL_PHgW`jC-ROMmPI`H
z2l?#d>!ucXV*92{?H7X>`7Kux7bwbRt%w}6A+cl3Qe3A_mYAlFEPBZ|q}Ft@2{!vN
z=N9VXN~l^^?~>Op5OMNr$WuW8;c{Ip44CR(X1RL@%+1{dmW}_SU`mT5{9B?k``kgB
zprV5J5`0fw?ga}U`?pl{C~wJWUnPQ4g&i#mprUKCD@z=@#mVGL>=P?wF{Mu2DhHwS
zyB3EeW%|3LkkGUMp2tR@xmr}{QzCD3K*!C4K!prZaDtT0VD4U#TKPPvKjnQW|A?2|
zf4v~Y>7o_vN95gY{UVWrL)AptHJ)L1QyQG@=#GQh+7=_9?KyvBL~q;e?H_Txo9xSX
z-L9L6;cHUDY8MW9+0oJ?{r9NOoJPT}p5$9glRn3I1r;B(y5&EaM-*V6t~b2lk7q49
zZX0_7GOjj7$de4th?=>(^>{PwHcG*P9qwFa_Cz3y>T}mzFmtJM9oxPT^3`akJI})P
ztyM?I9~1S(ZT+Jhgky?+Knk>8uYaFjYRhtYxp#XFg<x(aGg(;h*cD4U9C2QU=#8MI
zAMsyy@>jgfhJcArQ_f^sLx~_~Xec0h2kZSL>u%LN@1Y2q_lYA&Im+0!ce(}Y7?pdj
zYBx5cIv!hystdE(h@Va}cLe)63aD4_k?RJz?ecaDNvC;O1&^!;erVa@T?F81dd~2Y
zE{W3|YV1!sfu6nRiaf}#sA5a$!?l`s9&<wa1Bv6t4aNs1h=R2m>V$~@RO4%LQlvP7
z23c$>yPO72ejI2>K`5V6{ZajQhBq;0u4{nCCmZ&1-kFW@(gv^sI|GV#9GOV>(21`N
znPTM?Z}72jPB>u+Fc!v{Y6pUg@nV7<&rAOG=j=qpR2#{LnBSYl3+e>O*{&!cj3}=P
ze@@hGw%Bv*v`g`B)5x?1VKn{lncSB4y3AwG3m%yN8Vra|MF?K2%*r!_?t?DBdLG)#
z4pReS6YK2o1=6|kyrbN`0p?%wWG5_6;av6eTKbIlYrNX!R&)1#qh*quC84)rB7?EB
z`4JV5dk>N13!S^7kFNx3C>89l+Hs@O5V@h%k!}38OyY3_zFy~wM^;Pym!=MjFPS6T
zO7FJg22lI&@4<a}C#pq6qOv#<dfgx09kk^DpiImKUjrA4ZQWO7`hg&2ZC$NR*5>u}
z{}?a>tGrq+^o`0DpYcok&V^^r_?1<-a}gCbQ((}a<ALW!x=K~;D~HgZTnu6pa8Q;|
z^CuVPE_dMCrJ>^&&;1_u33vUSY+Wb1jdL!ovTyqD;nx9uKG1T2*h(=2%WeL^$q<Eu
z`MK8Ld{>Q4XWnI68#bRQ#!4NJxOmDp5N4jJ4`<RU2{xR8JY>BO#aA1Mg`=+GW#|d%
zfxU}_G@*8QVwKM^B2{Vj_4>a7qcJ0m>6vHQF|kH4g}Ox|RjuzQzY+MwDw;%R)e){q
z*OF-8K}a8%c$4g#n6Z2IQCjKm%!KCXTPi0dDOapJEe!NvnZrA+XIV;lq;8Naa%fS+
zae#vpc1R5b9mTlw*OoisDi>TN7YH3)G==}v*C>d$GK4DW*JUxkgAMs2;iBf*!7{B!
z17^z{g-;l$eBKb~wyOIn5wPLF+>SE~1Y}%6#jxnCpQ7*Y1te*@S}a%6&kOn=v{?(R
z3j&!=4I5=CTP1>4>W3xIrSBdu-AW#r3Gx=hg!{)nd4HOkO(CS2)&7-Dl`Wjn{xSj6
zDysNY{d&5++gYf~juYlPiSgns^O6Bu7%y;WLq4`Ba0g);$Woozkp3D<v(Chsvt(U-
zSLk^2F{09*HUpf^=l?;KR(2*<6Nj^n(2W|Q5jvM(?l)uS&#hl-lR_&a)#I0M;d*h$
zMg?!X(6Kfw0@@GuFuNITjYhTST<@nGn)ymg$G*coRXDfSZo%>-Aohw>`f<=jhL2Nj
zz<^Fo(L$!bn;c<_;x`GZp|SklfX2$qPeGA+qcyjG&4HR_K@f^%H>01-j7F#Dely-<
zeNw@J7?~=`aZ+5Xo%rR4hHKFc0iZ}2bma?&O6u5`P_|JGRzKZ)H6#VLGSV8TYaU6+
zJ-7~vLyLO8m64mJb8(PMjPz`^Z*~$pLjMgD|9*{|CW(UB6S)c8vQs>GVTyMbfe!^0
zApSpjv3QBXITlLRn&qExtZKcb{=3;swJ*j!8zUe_nbOg6p1V{hRszd_h;qC#8fYEq
zd(L}+D?;cfOj#z8o<DpjXD~Q$x=y#{>2Katv+qFPX5p<E*?Sfh#^uzaJh^mg|Azpb
zAV)ziHL>*H4V=u3D~sqja>Hqr?K*r;KdkhQixcNk;{t*j^12WBt?wpn)xPug1%tu}
zHI^?bE3MDUJB+|D1?iy|_M7k57|eer;#LG)Fo*S8GOX!pdt$eq5SXj>7+T-Fc=APT
zROTaQP<<XU)XhJ@6L~mU=%#yIynr2HF<X*6Krehp;Yq|kTJ@O+>RPG(q{yTwf7r!{
zt0I<Jr=5=gtKjEtU2|%Ykos@H<3cDI;8VIuerL<T?AoXs1%AdZ6A*Suzx=|<u(DHC
z`jI&3NKTStaxEm3yhl|lr!IgY!t^Td2_H&acF&Y=g2c{1CJzULw<u1F?$q5LyxD9X
zgh(&IwQ?h(b2$jykUNOZEWHbcGr&R*=KESU>-C~$Gvl7*Pd6$Es4nV!{5$HfV)zHw
z=>rvFrV13%UgZtlZ~vqtXF1)s{R4Df$R6)STy4&1N0eM@NchMS?ta;=DS<^sK^n?*
zPi-d^g4RhN)Wla4e<h1$_p;MC2>6$^#<-C_aMaT{Ae&0NkkAM4>IuAKr@2wb?C4et
z-4sV9FG_0qA+~gM^~18$H)>;zGfQgNkqQ(konzl?$Kr@Qv&?K-c_WdB%TzylihDin
zY9?cnm1bt)USXtmanqkDgv%F}Z-VS2!n3+fGWh%aUNEOZmATSoe(I3LS`|(zwk%In
zGY57gZS()0lEN>uX<BP(h8`HG6!&40iJbAnfX#mU&_FZ1=ABr_Pp@ZZgDXQ{-r*kB
zzJS+ltZi;>-cz0N&~JE47v~msyp`*E?6XI%aRW0_@76KDomFv0+CSN9=g+pqZN#0(
zEm{~aBlDiOOtD_6-LQz_{<}hdp>X9Z7q(S~c;jByw`)mGPRrh7()|ZV<gcLf!?Eo+
z*!`X7=Y0%6{0KH=tP&ZEjtEzLSWY$c_q1NMxmV~8{P!E*J`*q7V+m^ccW@VL>*sD!
z!m&TzYc;(LpKj&D&&2bh`++@ssF<mdAZe)v;s^sS(1I^@olK@Gsy^wW)ueHy{cA02
zl-_HAJ3;W(KE}<-=k~rw?Y!r1F43{v!1-6=z-6N|Ggoc)Kd1qfws-Ub@EX*XP2~|U
z{u=hCf&Ci2`i0WAWQ#!MsK{(O!6dRgN33E?>CSj#LYtg;aReOo?a_!7!+WYCwcCm0
z(o$0NWL8j*@XdCCECo*%M!EUOxE`*Qj7?c)9%-wqMVba#_j#D}AGPReHCr<5Udamr
zGGwFY#<K7Pg+E|oljB6(N)C8#5pq@DHsBZyTe;QpKtH9JV$w6Re>~6Gl)fbfY^y-3
zLmf-D*>BiT$%W}U9jF>lsz_7+7VG7)!&u7T0g33Y52&mxNji;hpaYPdL|>vr-O?L*
zjQUV~Uc6JMWiXCo=W=7`vl9;mVQf|Z_6D9XzwtOE#_L%}+&*k~$u;qYZa`+N#gAcg
zba@<rH_1yJA$61~dmjza2;5I`==azo?(tURSn*(*r}`a($+z^3U0sVqcE;uFJty)K
z`FeP{CExy6n)Mf5mYjg`=-pJO4j-=Yfy5<|Y0qwA($Dz#d!xCMNA6Mo1GH8Qmv%c4
zWC1)#!)CU?Hjg0b<KeVC36wdYTXo=B_bJdJ1=OWcCghEYHMbyi;7Sop<ls-TnBR79
z4D<3v_h_4>$ANpk=O6A57-QgytDQh!a1Ov^5IU7${9hGGlIq8GaixUL-E1gT7+93X
z2rzHB)&FC7r}=*-jJ7+^j%_Gxl)O(iIlvFjZ}xQ5ni1&MCy!@@Vf&tuoU^cS7~Qfk
zt=A(QQ<SrpZg|xeM)JyctdHEW5E-SwBIH=;i&XlK`s?s!{NMyc;%x3U5|~GFXb`xr
z>=JcLyxpf*nYRA!h60_eeyw#h8c18WCshY^8SUU$-8m^i>v*jZsE1aL<hY_29K?58
zF9==R&*9{hlt{Wfp4l%y4YsV>m_g~gQC7zK8W#`;hKBeO;=jn))@2q5&E2TV9D*r$
zI0zw#nOFs$Gp9~8j_EX1tx1J}1gs4ztap^(=Z2Dk8M1-dSY^@8fgtTZWIdG&vt@J|
zFxI(nJg%@kKH+A;Gs?T%7e`~?J+#Hx{j)h@9_)1a?q1@AS&yZWC9AD{t^U5=N8J*(
z%CvT%b=-KtF11bGo5(S|!R?ZQf^JH!K>Ap?wV%;JHT9?hL-Xi{+c4}R;gZpx{%Wt7
z^GS#VA>66i;6p%Qxu#ilqY!Rodcfc7@N(v>EpKz`u1lwH#q<2HXsvY|4ZYWiejj-B
z*S)+B+J@UAQKtQnyPcPdjvY1tjwWu)vH&rM-e-Mt=;1mVL(D0rva8UoObZA@ehob+
z%dTW8M7BEE)d>JX#@@cDO8U<`Yal03I>2Ijds(&59&#@{wdE3v;t^|NMNP3?@vvXv
zVSU6nlsqTt+6j@mPUQcAPAxsw8<i{>x}j*}lk~%jloI8efT5NIa-rdzA{X&um1dv&
zdTgE!p;L{o8!~2c#4-kLRiU6&*nZEdr}bd7%JnDhJzQs(xF>wOhFDW66wG&R!iqdC
zN@Y*kXXO!8cIGQQYWn&<n+CJv#N^&HiB^3&YAF}K-`DS!*d0qCSyNXV1fmoWN2WSs
z%Lpv5Lpd*z8)r+DtU+<mXC$`uv`h$aKE9nmSMRFW8`Lyd_8PMFn_CyWOq*kq26Qak
z2^aEN(w;;K)C}I%&vb9`C@Gunl0;gods$)Kvm6xBI)`qU$>x1=|6XY`eEZQ|D?PUp
z+=eLDUGF#EZRbeN$@DWK(>KdHsu~TQBiXOC3*1}CKE>`0t!pnt9X}P=E%UJ>XT};%
z{i^;rb>eA#KeG0nxs`p<18a&}lo2*Fx~$;1#)vZhtzr$WGmv&?tllvy{q?f%^Efai
z2#d}wyG9S+lWh-TiQ9p~SG=PJVvMrn5#?n=FjpUx*m+|pY_1xk%YtlIyg`{IUahuB
zkOa02J*_FG41O_lK<tPfp8tmmQ=t2||7UFUp?e#t`2trQq)|4_AzdTMKz=6^7QDcs
zrZKZe^6#lOS)c9mJ$9&|z<A>w>#D}DtSdE)xDg_dCFnfd{-4e&+_ihOGV&)`B2g}&
zWHmsleh$@$$qyi#hfJ09eR&;uuQ_}7ms_|x{j=Xp{LO<)j<-|7%d#&M!u)}YBTOcf
z^!Y6&<Bq%jVQ-0`tPL#}td+-MV{D42w0J&MX--3iCLnd#+0{vC128=fNyBx0=OR{q
z_$O1<CbDTYYM`BojNJagOFQx_Pdr(=2GhgW6*iUN1)+&}&4ntA?x(!$rWSsbYdbK)
ziWwj(_26x~xgtAS+rpMKv$@fdghf%Fqte7|!B@~p=*NTRg0^oriZRJl`RSYM2DDa3
za9S6biUUbVGIsA-UxJhm4i4@F0mi<K3L}3r2I;$?t|yB3Zd3sqk&bT40*7FzIsYD%
zWNSQqe@KYO_JRCRO@m8QDu^-DJH5YmEgZ<)Ds3$3Dz5(A_b%?TVZSQn{Ys~Ui)^;R
zUmNsCDbSoKjbd%*eZqW!*=L-xJ-<%5L(@oYQbN2~J1qZfU@=;aD`_PjP%OBb@(u;v
ze@SXPdc^he-}Jp2WO2@ukYn%A{q&CkViIP@63KbRx;IQ_NdvmmVSlp%t|g@=g3?NQ
z+l)6|u{++U+6)bJV!qow5Kz)!W-N^f^B4<U=0NG+sPp^E#S{d__Y`n2<LSJyp^%7N
znUYTG8%2L0O6OU9bia7b$v32IVt2JC`kxpAUV#61DO#vlCBUDy=8?PY3q>?J<StU=
ze@s(DKn4r2-14V?I#}7Tu{4|ZD_uZB>-#|`>7v`6<{@7#gtqc-hzp*W>>``+5gp~@
znzh&wDeC3n0Cq#0krK}Dj(U@Uy!qY%vC~V$fd72Cz?yf|Xsxz988DXrl!Y5jHBD;-
zsZ5NDe=;T<wQy*lZ?z?N{U1L}#2r*ETL~=r4w?l9`C&v!Pg!Oh`C)e2MLiS0)}Mb@
z@1s2LR~?jyI<%)&zNz6^GnF{O2FymPP)x={9KcmYDSj_5?+EIxQ?1|s%G;$xr#M<x
z<5>4t(A#Ti%Qwvq#KRZKuSb?pTWTC4KX@kG-tc)eI-#_m!XK;r>v?6jrAObniwnkD
zY~us(zoXE|i1PN@5LX}Lea49N3VhC2nH!cxO{FhAfBK=XGPtij++PD&rup>K`PhiR
z12N65rkt~C8ytxl>(^JbM)Gzv3|yPk+dUk#l7Ck1x#`~Y?4ZKO$@FwZGrM`NxOS|I
zj$Q>YoRz*>;cHiO9oKA!?k*9aWAQa$Y*_Zo4+zIESh#|@u#69`2`bg3#Y%OW$?sG5
zmH>JJQ2s#F{*OfQX;JQq5S#Yj?whGn(<Yu7%~TKo{w7)Cvee7_jRrP(2UB9Kx+I$q
z1+)}zsit%cDu@ABypI?Vd1rDW{l&~_StMs~`WBqFNzY`nNATRwKv>APM9aG$-kdI8
zeaZON-9H_0f7{sn$Ar7X>Pkhfj-zQ&u?IEBl&(F`4f7n{AwD7Z@-sG5Mvo>v6GsZ(
z;vPNEVE+wE??rVozI5XOI39R{6Km$Y1Vk0XPt=(W9+(U2^wSZz3OpY~#&t{&`-cGO
zC|+pU%8gV&O-k|L3Sn#)JYBrk_i@tT{VSt&O8TIZfO9ydDeBKemEpAXj8x|rLU-aJ
zL=`Se@fpc>gGavT@8f~c0m^DHVZigiy{OXLslyf57%T&4#!!3UybQoCfUC9>KA9_-
zbg+^sp%!B^l6k>`kT$1Db668&ZX*)GU~WN(2983lwIQHN_zI$It0g(mN$X#af*uzL
z`Xio%T)8|TqEEXWt@%{E(Fos`xfw8Yp?mxeb<n)1UtT^@^;(_qW2gnZ&$@2sj6t%<
zB8ii6Nvt8)kC8(7H7s=>nyzTJS6Cp3zuOtfiLQ8fi@45}-CCECu2-Yl$WyLw0V?fx
zN;lseDp*yYiwK*>L01ILVkrUU4t<H>F>YNBsyr~kfp++H-x90*`xG}yc@Lc`UO`oB
zv0JbpNi2|dEsB`v_1Q87Ef8BV%RHGlLl(~iPH$0l(HT<$55Dzc=fS2E#RVAU2Vd?V
z3Ou%NdR`_!>&Ll2)UFBG{yT#PE@&Qk6!&sAw}Zd0Stqv{RF??nw<f#EhRmeEj*QQ~
z8bd32loGSoJ$9d5A|9zn-FX=c56X>-zTgh&^0Tu1l)_J|8sPh@$;`9EEWoLp48c_#
zoCIM&`X^~m9{T^Mh(q(>lq$Hn3@N($&fk?cxi@Js$2BtuQ{}|QYzEnSs=AO37F&7|
zyr!VM`iJT=+VP9NSkM+Z&yC&)>(O(K49MHrUizZP%Q@+lquHT#-fV3yh!D+ZNuZ=a
zq(d13kS({VN2i`Sd!eq`&g-<5J?dVnQ`o)f$-4#f+37EesK=hD7L~ja_sm-gvLBwi
zNIIula<z3^uf-Z7FQT;7mMqaiHryFY_t)0MAOYVovl-fB%>!Rqz==)>sQ%fvEQ=mJ
zQ&8uGD#TfxSD^H61nmA0VDR<!KE?{gSP*l>*!{DA!LzL<<vDo)2BO`E#MOE?gV+uG
zukr`IxTO}YQuSYmwwXa`$7mox$pOfVJjftIuPknu3@cdh2-zF}KA49YQ!LqQPYW#}
zqxUnSFOrJ{1cAE=up!xn0Q?9(`agDcOmz~BC~h0U@d2`UuYD8*kUKDuvjsR<-G})O
z4yW;7CC-<sd6SBpv$w|zWr|Fc2eqr0L9+oYw3cSTVdSsQQfwp4*1>{;r|O6ke=sa9
z^W?Y#F_-DLY3=tX3krf2i#u<bn#Fn3Mz&H_qa&S;R1Pc~l92xTRMVtgyO%G88!5IF
z^;?=R2e$LY4s}nen(r-|D#@|j9OMi*`FJ=cw)1qPdHu)8W_Gl4s+x$=jT-7Rf=MrX
zSucb0-)U9=SVY4^7K98`K|iJW1Q5}CDGNOT)}&78XO?GRPl=R?QZyenZDKn+t7p}F
zDL&onuW;}Ok9<k9j(;I7nWg}axEii(0QQbW1dFO2k{ks1jJZ&m$PZ@Wz!kS$b)!R)
zk(_exfHNrQ5t%X6CZ>{`T(_44*_|&khy!2&`)frgNTmW}HxJFV8NOmeu(|PTEs+{i
zOv8H*ipL3GTm2Lw57=Lu^40zP0NU<L9(hiBxNE{6cdP+hk=Rfk{}Fl4Pvb(I?{Tl)
z*ZUOaoYtml2~{TE2U~a&B@fmYf-F*TQiq?sOc3rLoRVV{jDZ56SbPR~T41N0av>39
zPXw;)%>33-o*L!S{+lXRz4)a-6C3De@c&pk^Ju8s|Np;dhOuNf%API8QdveJYV2d0
zv5ciiWs8|fvP2oF8OkytV>gMx7+WeuY0(E6OEs1;+>*LQSyL(MPRsqf?(g}XbD#dY
z&#6<c_jO&b>-l^<=!TgkeJ7nTJ?q^V=q04K^E<UDw~&FAk;dAh5Zx<;5a2vi1;xEf
zV_ts&<?afY<HLX$^bT6#y+hNB7)hP)TytakIe`}y=gT+51Y-%iaI1dn!~gv3uk&zQ
zKTf;l`Uwz`4|ky^=9tdv<BH%pafs3)WVv{G{|^OIUcb?TeSbNau+Zu)BT5G=|LSqt
zIl|ZuCaO8kw_hLOX!b}0sbcQAfh%yGt#bqK<!kA}A}}B8ak3kmz>;DA1T5i3H%YKd
z=Brp5<VIz`++uFX^KxP}OV32!XqbIaen;Eb3xxvBxBNBIGuJ?=`KE!_RH9<waLb`2
zp5>!^?m;<~;Ol2}`huy3_ru*rC66yRGezYN(rUk!td7(occ6MvMPT7WLT{aVJ*#|_
zQfdRjNlR(dUq&aS{)uS{?(=eg_RXIcWe=8y&E{1?lTxR(Q7QiXcB69cFH?ONd~-}X
zU5fK8y#^r4P7Gs>Wisd;QnJQ0_?(AfAtpnu)|P9|yDdOyxJ|i2L9t-;V3F=)Fa6q8
zx7cVbr1Vo?{pTj7f8w7V{}Jix7lr=doR(5l_qoVr-d3`y;2$@~KQ|qNOkR&(tDUSj
zx_r9yBYiyJDDuyK`x?;f3^2>Y<5~=>&Tg08GPmS|QAYo$4cT8@Ln&_?Rh6xP9ZY+y
zNWfYgWP!-{Ij&KKH+*5@>ID&$fIrtqRscIq?b#R+#HsA4-9O1gNNtX&=SvkH9Iivi
z#d+k>ydlSRE(f7nL*MgwkHs0m82pjsn+MAuPswJF-WA?3!uJZFcGZWKNrHobvnpse
zuYyqw(CL(AX>Kxa>C^mmRdz(5Z7KjJZ^r&ASjQt5@UZb$6bf>KMZ@?MrHEa6&fhP)
zeUeM&`{>;mu=Qa_mx`@B|M1;$J8;J<kg_|thvu!bO=-6V^&I!?vT;`6j`MS;4quJU
ztTVTNk@IfMtCOs=I+YK0x{A>;4E^*RUst<LCUl5mHv*@PNzMrwxm*ws2mdFSqRJg7
zY&3@3I8X|nfBT}CGv77?ysoMqTl3S<XH<ESPS3p`3e`<-&)0<JNletSb#rX~g;sp!
z@vhbNW0+^sN^EfTGPd(JSK}^%oZ$9}!WM+W#%v)EG%zUO;eosf`kVqoBKY8n8in-j
z153DM`w?AP0Gq;GcMpP`nes*`D6+5)FjP2**4aD>BpszKT6}eqq>c$Gx;$VIS^F?4
z#$$-)SHw)$ufjWjr=U#aONEu3%y&r~mxh0$Wkhekp9RJ)Y`^tI+*>J*W6JY|H@7Hv
zK70b)6P1m6rC^you@-^!8Fi?3@6L?|oDE0nM%V7bAfE(wNWh~~qvywn*oH#@toDjE
zRTC?LM_vlfYrFIngsrRu=4?o!wO;PrCu}C&f_P%!J4f=fql#G++j%G2tYU0db%6WK
zHJ5+tom&5MpCx_eyW#Bz(Jym;jbD%(zqV7kkSQh5&56mVwil`cw}g;=+(A|^qZL^V
z8mX=GcWzxBW(o}xCK(sn*7)R*{)R^HI7hR5>0)1In0Ax9b3j;r4zkIm<j~1H*?rF)
zntFGe@^R|)?Q}|=%`9DcUQ5<CGISzeDVg~cNOz(DEhzr5w5a_mqAqH0b?o%xVsjP7
z?PYaF{)473>LnzTXH;$`(=Dsn)IV86$S8jKKlSjvofe9?8;eef_I}2-`d{@cs_-58
z8zz4*uAbh0P2+UXV82bvm}S9TX~LKRn&aq!%`~y9D)#tR-`=)tIcmNg59ePYTF$@k
z;hq|fH@!RZ%U0{2!F9*ev~2u9NAJ?N9v0n+Zt@5uBK8K5N3*}BAr-{6$JJP6U&t!`
zNX(-K*z;lh1q)oLD}-c|Ri$BE%r1an3R5>m*?3cAyeXjYTMMUe3&C6#2d$QLq6KjJ
zvelH2R4><57xYPpVel8F51kC5&Y$`#b9KjiZ`}=*syz&`i<rB{+T)|rH>WDT@*b*i
zN7v`Y++(el0FCB$%#otY_q20TP3C+bR2hZ6xUjY3gm|6LDl_0Sx1U>6cr$maHsSOy
z{cYIT=RQ?3Z_YgXb0#%RcY~63es9VvG2z9Yz2|AGzZN%p#W#}kwADNNKaXWzjv3xD
z;<J$@G0wx;?@SQAs44rSQ7%7h-Y>kIbwwhx;E?+H164nwjSoHL7WjX?8gz@A^)0&e
zkK+4&v($6EJz33Tan=43gp1!!bj&bg$94{DpcmwP?ohnL=d-THQf=0=g7Z<#^KYL8
zdda&j{c^U}tW-Nblh3>wP5E%fpiTI$HQsUPTnjF0%Be%ICVpO|uYFWj6&I~Qosu)P
z92DPaA$vL~+*>S~YRJ>uqgNeHaFLq?G93wPoU84T36A~dnBx8Fu0!8D-hDQ$nLLU$
zlb8#ttKJ>I`!4U}k&Q+%@7>3!*{H|*3}*s8ww!}C)QH_$c08|)=OMQ$5Z1ft#G+wm
z3+AlJ15P2*A;}qW6ZxMdTf}sX+SN8Nbb?S5gpju5PX{4^RU91o_KnGrf%Ffw%X2H!
zYW3ML`bt;hwloVu$&pKQb;mES<*knvT^aqKb%bUfH3#?irwBh88?NLnods(IO>Vus
zMz!mY|Ms9l7b72nK`L5x3^1y&U0(pX2C25ewI6|N&K4SYn;VHp$}4;wx%DW}`4Dq3
zmRiM1;PJ>^ZdBZ4R1182C55a|zT&}#oDv)XucRjLH~7L6)xL8eO>fD0Iy`}|30RBN
zAZOR+q(7P>;X-7JsEC=%H<YttJFCb+_5VWKP>{rrCfqymE{@aDy@HDID_j1V6*IU0
z&J+8t)O_>T+C<})AElRUI4(Vrh2lfRJ)T7{w_^{uryl{%UgAxw%HPXh1{9%>G-)Wu
zc`$N0xq5=ZvV5DmY`DKefN+VP-r#N3=Fv4OzH@$#AmYUx<cJWihTWS7PN$f6d6cJ&
zLfn?yU(zdUs#G<y#YiV~ca_~ZJbcV#DDN?@cT>9Lkgq4mX}$&N!8W`c)zjktTFybP
zx^<2XM<IaVIDX#GQ;r5o#tthcE6C`G36^EKr2U$gN_EznuMu7<;<><+1M|BySQJ4j
z7LW}5%&yz{XUz`>Rc}Q#OZ$Su60Gj<8B9bTfC}~$&4LD167OAJM!wRk+)o9jL10+v
zuXbaLF{vetY{v$ed)uZj)_S8PXMY2#YYi8u6k@>S4W?+z17?&kO|E>4{lUBqa!J7d
z9=Y60G&0CJWc^<`<OtA<>EQ8VJJlS-p~Pbvt(U{U&U+^u`I1^dfT^zq?wfwSQ^IX&
z+a?G-T<*PLH+w}nuFERkqy7>Hmt-7=3(FHXvgL}ku91|ny-09hCyaS;5SSRg?le)v
z5#AnCxXt^d#`S;tqW-i%ciLODjl1v4lhQ;Tl(qPh%wC@kl4?1qTEVuJCB1{Fw(FDO
z#4opB9dX;6`yy&*w7+`lm3t&tiT0gCRdMQ?`&9dFs@;|suW_F%pvm3ZMgKx(0fglv
zlHt4`i94>mzlz@jbEO|mD-Ed&Q9fb)Hqa(S*8D{BDSx;3KI}^}NHI+|_eDgz>FsOw
zpH0a0=qUER$P#YaXhb-p!Y}M)09GwdAnhExd^klV^b6bXu7983c3SX@cJX*GGda=3
zmCE8BGUn+%)(<!R^$gN5<bhXHH~W4-Bt@I~k_Ofdsuea1^qnbLmvf!w%a~vG2GyI+
zwn#Rtf>?mFT7kdgpkY~Fr8s%c&;Msly>^`DJ<|k>k>;&G*f(zH0)eGsm6`A{%VY+0
zq`A+Karm(`8(%C#F2(QNhhs&(ZY?ECUde&y;Edp}@v???(ybX6fBPl!;679Nk`yjr
zb;!}iypF|TWmhJ^M}LX{3Oop0_iRn5Ku4P8rN%pG=9V(P2;1OA>Aj1pOb96DEU382
zT!5)4x9I2;pBtv#c)I<vVUTWY0-?`XU3Z@^ZlE8<)ZY|_PcgFG>+zJiHw*^IP`g5g
z$*4DE)$KHk6`YvCH*lrRsT9db85Xi3i^_G*S8VoQtR;6(G)#`XG{KPNjKI%q3ccZ)
ztSbDmB5)u??k|GE9gM*R5lSJY_hT}Mh%hWMxZ<+~iDWWRUnMNF3t=KnZfW{W+SqdS
zTBG-(7Bws*%jB-#c|3k@GGms$8$*98@Og9ASCO`L#!v99iiY*RxKZ}Wr2>5Ze)<#{
z-smYi)f-O(frz2$f1#i}f1`R7Ossax<yF7G{{l}t3P*Uxd-9V72$gt;#v_Yp_@04!
z_kJMmiBjWLb*5wcSOU|K84~<Muuh}yCpLlWd`7Jb!3s*U_>nXtS6}G;+;tI2JeCs&
z80Z6*#=jd>pT4Q3Iy>(hQ;k}})c_;UcdnRO94l~}{m}IH7QGV|OyiY_HVcEG@_k?y
z;inR2A!IAsiRh7yN@#*m^|#3-+rX0gybv<{L?WhxgoZuwH|&u|(PJb*s=ZOBGqTKw
zExcM6>L3U^9F!k$B9G^QFnvb7k;w#$o!9t6bj1WW&A!pcM($nX6;3rO^?h7N{q^_l
zVl|s)CyN;S`Gk46LsRQ8A)tYuy-`!H3mzwmX(JL|yx{G3O!APXp0BTHQ`JQ;qmKp0
z`2rS)d8p@Ch})}MD9qnakzpmH)`O5Zq_OH(ERrG%SU6lR%!7#(BPsKttTJ{CQFq#s
zEJHmU9Gl?wFGOQad;cj~6=lYq9eq+)RQ71ml4NYLcjQ?9N8)MK@^5z)ET(SqqtpvK
zr?773C!5&cV-f^EClH~h$Gzcp>S^zHO&GyMH%&y*ZHAl6q`Rl5A!_tr`uoCfM`1{4
zd=M5Dr8xpSh?SzdWB%UhMdcpUDVYA)lP7Buv>KhaKG16KBiW6>%&FhoD_6vnZR_BO
z>9PHM6*Ua)eDSkoWx_UeKSt?r?ib^CV@~?uk@CNVYep$>>M!)(7w32B*`2)x>h)G3
zYkt|L7CV}K-LERE1xF#^P5ZJxoJxLiVOhBK`op**<#!IJ{|hNLJ}!13!VU5;zJm?#
zU0F+#EPsVFoQMBHXM7d(=W0eG&A^_XiH@nJAX=lt==Q*CX0@T$5Ef;h$180&SGeG(
znZ$+f4WjhF#r6TPA0DFE@W8O_l5etpUJ;#wc?w=E>#OxUEw<{d9ICNFy?A#dYbGJU
zJ8#2%-wqJhlNNK_ZSDUTyCg_Aah<35hSj7W^H@6ccpu9d|EFy_jFCN&zv*qx$E0I>
zbw*Mjj=;ljx>5F#AnpnlW()csRY73>EiqB3wK<}9c(8r}P6utS7Pwr;cOU?&a>n-4
zAZNg`Rjo-E(&fsyAW6RIOP@&*!Vf=6rH)nPwJ}wnKin^mzib;;&&2LGckFJ&irWGw
zIQQ=r)ubBz3?9r>i^s|N^>V&!>&sSc@qpnWv|_w+h0M6n{7Dbp>Xl(MP2Hn49VcZ5
z#w-sLo6*0*F))jK@fBu>%)gLTY-i^pTqDEJx3F-@&vQx60z0_XUasiIlQc$-DVS$~
z9>~*z1M2=OSmiUp>L*h_Fp+x??25x7=ed@@+`orcQfA)9Q&1Uwdj+d++H2p1$W@NZ
zaF1R-H{W7}FJRfNbJ)%rtHfxMv7&nJ15}2~#?g)ytfj=K!^2+(`rou==kT@HKif82
zhKmvyzG%a8vy%PoCmb59mFw`FaXt6hQne;nlC)E0g}2T--x3lU?&F{xUrB+Ugw3^B
z;2)O-m0&2?4xfO+x`ct(%8iYV4d~!8)iHP*uW2BTL}td{52`giFG7)64>m<dY2&&U
zH&W)yJG7}RGr!;p_I2;XW+%hLvlpK*FVxgL7+t-o@FP6@8$W1A1%Ld}s~4j<bH`uH
z+Qn`}Bn7SDZNrh>xo6!a>A-92(V$@cGa=L<%$40{*Si}7F4FSD|BN_LGTx_BBz;Wx
zItGLdt8un?_@V{6%yTMgpvhZ{figfA@FANBX*}|a5oMUOt#F|%++n}l>K<GHt{>s~
z8V6a6uaQW~sYd0vv+|Y<L$ua&*0%e-1^VKl{VE5_cRrh)z^Wz=+$=kg)Mqt59jjh+
z@qFJiJNB8et2?sw>w=HmeITEpEew{;Y-m2RQrzM>*(okVeoZ?16cYisM{kPMp<rP;
zU}YF*U+J#n=?K&~Vl{MS%%u_C{|MH-ee;DaV$6C8X4xKB3mE`s;2FBiS4VE}JhrIc
z{Io=rDvk)Tq*mmX%->+O@_s;{Dh_#2TaUbInyZLxOEYx&n@oyr_Ii<-Z_^?(x$YiQ
zuY$gUD4=Aq(cNlwiVYXqE)lls*xGV?lR08LjQwi?Sbd;a-SV^%s_Uz#v2hN15e+R(
z6*h99nawef8a{o&xa9*H#-)+ek(f<3VS}%u?v~7}EP|fJS11x(vxWuT&8?d93xw=#
z`Z~;3)ww#KVLdG6wGF3W5A>V*_=-nK*7153DXp(0az=$VuQohyT4+Yt^os(kw$IMj
z%ag2H#s#CuPjA4+d7UE3g<YiqIE%&vc!*UG{tn=D(6>795+f&%`n875luH?*(MDkF
zTs~Z#EI=r0j*y{M`~h{ydG+8w-tdb-vBMWcbhgz%KkfS6ycD6TanH;b6BX4m8cEG|
z`{O01yG^t-)G(){Ob|=wLwlUgp|X;BV&!uxau?#T+jjnl^*q-+^XEh16hm!pf7}3%
z5Pwdfo4rLbUtY!DtFRyryT#y92P(C?7|^8^_T2@~c$Yu_^s?nh4a;_O2fI-+(U+75
z8j==lz@I$EcpCz!4)f#5p-iNkr>vdz^AM8SDK|)KDZ)k-#dP@=_xbPycMH2D@Ph8;
z*&8(?i<;mNU!rafrpGgxOVX!?t2^t4gJL@#R=i;2)?XJP=i$(Nxa>vzt#({<WpN~-
z*n7ho9VBJozhE6^Q30w@%eIZ~DrDBmC$CLkn4?-P=y%X*!c!>d6bdEKQTN^79w|I<
zEzza&@{Y!ubU5#r>Gf-Re%9zlbfN|5)V1jH{fdM;Q9K9gCBScY-5Wi2j!)WA%qx{3
z*IhAc0ai|wZ+G`nz@@;fsFuUXhAU;-y7Y2Z*Rt&mU*$?C@KBE{ao<3B<B9-hqj}GL
zlB>>C{}Oi0?qpPni<tjEZ{uyP%d72z78<S5V?7V=%DjtY1m9`BtSKF<O2rZ$`z;Be
zJ#O$CSvJ;KFaI{GD`u0}lP?)`?IB(EpN1){oiJSgtGF&mAzSc{)4VLJbB?ahh%$C9
z9~IG^)_c}7F#qe^`vQ}J4E9~UYyC{;V=^!Qn%#uSj4rgQ%6Ge_P+_<~?O{pZHmMWD
zOUis7^3H()K=8b~WjS<u@ZI0}cIs_+A`aQJ#--%{KKQVvkLmQOPSzD{`J{}R?l`Nd
z3fZk&@&q=(yiq5xK#~PIGD2lwUM7qwTmlX|0@&k9csjD4hq`#UTq70bduV0Q&kO__
z7UfFW09K|oj!<Vh9vQu^;$)3U@v?q#Yhe4(z23NbG3L}3Cj;^7V_fOu?K*Fho2{ac
zw>|jTj5jELIClHyf$x(u6~1$U2eQ~=9(Rpynm8)W6=Efd(mQR$v=$X+^f!3s*lBMG
zc2rf6)jwls=o@Dpr~<?)i!xc@<!xA1i;{UD7omJBh=Tacg{R8u4xv?rk|-e=>i*Bz
z>fVA_ArQ*todaea8d-Y3TxR6PPwijDgzC=SYO#q@<zf5yF7LuvI7VlxiJ|6tV`Kw>
zPgx0BI6_#NyD%-Is%~!uXc~XTc@e<QoVN}@0brL+aFgPMk${EQ7L|#Cp0s7q<tU4(
zuHY`6x+s7&*SW(W`F_!k5KD$Qn;^Gs4?x56OYY}KU_<9n?-!U@T!ndaNQ~+VE)1=r
z0R7LXPRDThaaEn$5i@xwv>Gbi;69B3w6l6fY=sR>SFEinFEJFm?bKDnn`wkIFlXDM
zg?3k6F2Y+_suKqcP5qdrIsi6z5ShfR_?O>yM8V=T;t|ZCaJ|BPM3)<`I@^-`#x4}e
zuxP^j`WV<&+ob?q)df^WwrD1j1v?AEzrx%ky7+O1M^_(p%WfV1;UH&C3NOFU+HHJ?
zZbnTQMdtGIqY7_E@hy#L_vrxadP{BVr9q!oUw!FX%fwJ-ajd)6q5<oWu-?U{$)&(#
z&pB&zw<f1;;YoFqqF!{sdfgPTGz60_b0CB#0!2Sfdlb#xWfABVjw@mC(MHzkdt8=$
zjT=?DW11u4fkglOksvf-{JK>lk7;Vbdvq$8$vtS?F#V%h5iug&=1B2lbv4#zI-GRL
zj=qz%tsMs;%XRgf(QrT6t3mNxlSeKdAR!sHa!;X6HcnbB@#o~H;O?0*mjOK5z<3md
z7>~C=d4GBK=X1}hRBI#zvP*TI;yC#*>l9xL_xNmOIB1Bi5HcV5o~w=BXjEMfq4wPF
zM>2}v#S(R>kdx^3|H(RfN8mA-OKhwKSIf%BN?BN_rg8eu$_0Prty<2~B^FOY>zgC1
zbCpe_Iz3o8uEr7%kHmEFG3^lMKIZkC^$f%1kx|`(l*fIxTEkSbS9s0Oq^gjg5oMUF
z-q&$~+b-yzP52g}k+QYrRyZ6rq3!syS7OOC`O?~-Wv6;!@zchu=K~`HYz2X7P63>k
zma*%f*VPSetap8)neAh0yeUoGx`nA;iaC5%Q@u3NRBnz@_nK+8O;!{m*s9X|+HKj)
zH2YzG`_?VFhHN6d<Mm6_i!p_3t^8ddd{uh?X69bAA2DTtxxFW)0Q-r>@UII&vAKFV
zxx*apWtoB|Mgh$qHv06mL7bSUxrq<L;>Wd(meIoP-_CUp>WCPjUigSs7gk<c>5vx0
zBl4{VCw(s!Y^Nrvbe|Veeg-FoV>6Q@u6BCjQiJFQis|@Ko-Ie-YQLy_HmyssinhNN
z(-mAnQf4B7rT}mPHQAJzRAcB;PDCQmVMx>mou0!X-jul?4?<epu^R&cl0eJtrAq;!
zG38Azcb(KEG{+}j>1!8_Ug}RjcoiQRra8;GK>Nle0cUB)oQh%A)&B`u4;H=vEqQK+
z8__(Ej6NBEj#t=6k)eDCF2*qJ|Bwl2&@-AZtj9y1zofx#EV;z?BtxJ?1+{@$Gp-)<
zq)1YVZxlQNTVXnbCqKh9GBkjRa%g~RLGN{NkwViphTc0^9GX2~@~16BNtQd;c_?3D
zj>W{<?oc%mO=uhe*Y$s)a}57?Ak%OtoQYZkfiCA1P*aA*BffwU!0^Y(g?7k$dunz=
z@D~a+9-TnkDS&LfsFep|*1K+RJAXpnX2~ytcmphYjGTllsQ{cJED988>NnjiQzGf=
zTsAN?4vw){&V?f*VChkNiIun+Z;Zn=2RpxS&n*7C9{<zlN}jqo`ibNd=ZY8CWfFBY
zNveGfQ(Uea(Wx(gptAWvH3CiDon=+|&yCxjC(nII7>;{2WIXFjc)^O?6b;5t`zl&a
zYj2nwovY?~*d4teO8R{hYxA3$pV{(iqpSW(-=9UE?%&bTzeh=1`1-}&9Ep}^+8vEz
z4JjiChVO__X&<}9BKzwBEmJ2jEc1<Qb-8hl>nsQn_(WRq39`BjYh^anqT5usH)xC7
zdBMcp_c270hK>=EQ+(xoJ1)b_$A5ATu@dB}FgQ+)&<w~DTpdCK<*|A{0yAF7#_riP
zWcT>xM6iS%ji}n<5J!nlNs@*KfnEpRTiji1|GrhC9|=^r(+P+)*(iOU;uG%dg)WiX
z+Y(PM^N?<{Qg-%6?a*moN5mz;v?%R;D}Z&D5ZK`0>m<4cLicJY=%4D6q+Xx9+e3rQ
z$q6>dS?~0p&4b*RKDp0Zm$t_K3q5yO7|)}685(Ripzcq*8@+gmr25Q3+|d9}YdCN@
zQs9F>x#g^_#jaE=->2#zq`W0RRd$=6U*S0FXas%MwZGGR$S$%x_mWlIMBxh_Y(BE|
zA_@`$$?vatGVC+oY`M>R;26=1l5AjZ*%f=Fu{y@+cF*NI!39La5v0irS2wk(dUl1b
z)B`VlHquWN6{s&!d+QBv>Upt>Vd+k@C+#{H?8b~OUj1I`{)+u*<FM}nW%8`-Z0W`G
zqahtnqFU@7z!BlK+n?7<hj)p0)F%G_kH|LVM8gk0p~3$djKur?lP>NuTqcZwZ%f$`
zVOiG8PO8yqe3<D$9j)bx;fi#>O94hZKLx3WeJB)|KGHI&?QqCVNHje|@Oc>U>4^Rw
zw8xhCd7i>eTWef~gNl35xrkK)>($p0(kJ^k!Xa%bzRs>ANQV5vM4kV@3bwyylP!&m
z79w6W!Xz+TwT38e-~juy1@vwp%on?{157#;OOZpXX2P%G#QI!Sp-8$$pB0lJ<865$
zm`Uciw6ZF_f?CzO|ASi+3;$yNPjO&{&=avT9CQ}-;*!m$iR!pj9GTJaFC?9vi7Bs9
z`!rP>UIiHV8r`ZqrM(erxau5BO2-0dBI7^}sUj@@Pmv5@4sVWFSaow9!1U}v7b-JH
z9Wu7+KR!g4YrDP(+0?E9^cYNVAQ3J{6DDA#D$mu{w?L{kLuBw~nZ5=t1~B`dKK+23
z&PQ)>;f))!Z7tYycPp=VQe*;)B!>3+#iLJXlHhIh-S#5};&?EBXc2(~!H_v;MBI6T
zL6DHM5Egy?XGA>f6<Mu0YvyUr21i=JsHEQ29I@cc7TOCHqL8<a<US{Ck@~XnX@b+?
znJ~pD?tThmR@~PV!Y$8kuXELb0B-@7SG3v*dYnY{b>0A$Bl)#((ZuCq52KL(RlE4Y
z4o!J{M$qow2YPLKsw#+`I=_~I+FTMG9eP1Nw1B1Iw;Sl&koz~pt`B!&WS)s8^P69_
z_ynH?=LdlUe^bct!>)4IF&ii)FZFj{1N3sF$*z~YO-^{)q6W1UeDC=~Vp>bD&<fGr
zv!ziJ`O+$vcLk>!Te9L`rc9*Gp!uOzWJ$r0@w1bL<SRbEr}DK;@Ctx23I!C6+g@w2
zWL3q!+yrOdW7K?0<(x#WAM|d#M$M8jRgV`M;BKc?b+yRwHHC91I`M%R$?QT6qp<Jt
zIW5zDexrJiUbI35-@kJmhVbW(Ny)^^D8%{rS^?QdW-tG+ip|X{1WWsiR?6TEb4%_f
zr6*@{{+2UYe&<9{lHR?Xwvq-Uj+=7ggGmnVagO$S&3392_zfxVv7l{V;p;q4H^|jK
zv_-und@_Jvl4Fy?8~%y-MLp6Qo4c_mlxgD|O>U~at)k2KXxzeP<Q~LiBh+(qoiQ#^
z?A%gJ#-WlOiMmA!sVC)r5+a_zQqFrsskLK0;ki!{V*I;F@;ajD@si6XLfelAhL5c$
zcjc*HPn%7quLCs#&+FM0@xeoWTPJ+ozakTR*51?!QP0IvlTkrE(W8~GYly#__nfQ~
zV|;JQxpcR5mq+u@@4Kw{y5)QOL!#pfJ;;-uxOr~d(L)>c89b(QZ)Mx+Y#IOWB~N*R
z`0IoJ15)Sov-L=h(~qKLK<nG~Vjg^^#GhWQGyF{6lCex&Pn(V5zWxNHijEHCExE&;
zVsrnUhi<8;MXzIe-9DH|(8#VH5i9-`Ux|{@xT8HeITexjXV0CQljJCgQxqbe#Kl~3
zFE^gqNsb&057a2+<67T7NC1NE4y-LTxDS}ot3}%4w!xeJl1YcZ)uS<V+hOZHz=8TM
zj!o0iEW~Pur7j{+xh{2cLOL9%lqVIjl0@Altn2|iLHjisCdtSAL1h?ul7~_kRYzAV
zd?f)>w`KY(C}q$z6|V-hagn_26CGL2BbDM-Y9+oC6_m0}S~f8EAhXn{dr;-%t6Ix@
z0f*e9Mn4*qiFhnBW%wqAOY^qiCh(1pwaGXy8_U~&#Y*KQ0(8M?!lv*H2g*b1gSf)T
zg*NC?SYbP~V;31(`sC>eG|TyG{WMPzRCTOKyK)_hn#dDDNdg@X0tGUh3z%9qc6V@6
z^Ea=rr@(>!@PiqIPj6~6GG8&+O?u4_upO>TMM&H;*~j>@V9T>)art5(#!&J24@#y;
zlNI3qgg}!<Ive+EnLLQWagdv%RYHbDtjTX)(i(Q_we^M*3!S>f^0K7ej$p3g2*ehZ
zZBV+fPD`R4!bzy0(ksbu%eXE~^CsJQ_p65B$Tl4{Yo(p)3O7wY`0Jb1?ZcpSzw<;n
z&OUov0G%%t6bLh%HjV)&5Ef8BKbC^#!U>g!lk+8WU9%P2uF@jpRRy^8iL`@?w|?*s
z``e&K$}M*gnsfY>?4?1@s?#`*9j3Xhl7i)0#1Z=r%Z;0HkJ3Z@uXjoJM2<S2jjHof
zDxA0znRTZYf%-7xR~oh_hLR`hYBp{n0d_Ci)7{$5nca3!Z#mBE*f=>YLd)cw2v8Z^
z;!ot8qx*<Hh&4V6;v}$0H-)yUoQQ^Ud%rR`hz=f_O1!?X`A(zLHRSJqVokQS#ZUfG
z##$#-o*_@yrl}^P%6P8XGsGd&y4zQy!!-k}_w4)p{2*Hq-!R$CnofUH6j*$2`EVfE
zk#acW6|>1#Jmwp?y5~0#Z)yroy+*^$B7hq5BqDh5&i?df8YzbFdUSH}luq7DIA`I(
zMl*h+_+w)H%fsi+3qSI<s`gcfdlY4<ekE9o=VHdN0~OLOIvlm^=B4bU%Y{<ri>Vjd
z2R-a3?v&M$Za!8xd6*jS7jRZ{_uwaH{;@Stz+>K7gCmc>*QfIGe5NxfSBwW+UQ5qy
zT^jg76BdjRCORG_ha0!S0<HQh(>RPUQm1&3592Skvt@h0l+S^z3L$pV#}U_75Y>Xe
z1uFV)8@t9-XGHI*)zaz}>hx`iyXLUzsjx+bM~h$4(BF7<y6l^BPCJ+=MAG48a6iBN
zfLDl>F%s791Ucd9amsNv*)_^wqv39)9Xo`&^sOoE%{w^F|8}eKhm$UCI|@v1eQv>N
z{R>_4FX%s2G*{u*tf)Va8I_6esr{r*YLHjKm7rfOk&w`tCWX%)s|QQgxw?K9%f|m)
z3<{CF;bmC&a+xCYWya0wV1J9gw<;f)_&|;?d<iRsOIX25#(~}>P<xFzcA@Cb1diB{
z4G(ORvQ{XmJPDOh4HJhB-lgZ}drW@q%(Ei&o#5L0aW=<QZ(D18u@fq8k)!G*a|McO
zHp?c4wyM^KSrq6iL4QPuK0Qa&2SKAgq$_5`!njSw>42S`w}m{W2rBCbac#uaAZI3I
zmq+Pbgl53Q8nEZupRMnWo%YuAUBQ9!^kY&K6GBSWBEaoKYV|)}q112-_$qwoXp&$r
zKbr6hXWu4_fMplFHrR0LT7o4LYZrQZq20S2H<SV_I`=xrlQh4C-#z=r)Cg-6g&06K
zr{#O@^A>ADzRfERS5@Pj2z;d?4iC-od6O@<r6TOsOLO1$Mn+ihZo+sHMuv$sS3S!~
zz=deY4&#|V88#fr?*5>1$lxqTDn)t8`-xlXqMAIdK(|!8h(4OxLVmSdMG51&*wxGu
z$y}%OFc_FAXCNte^}qqF7bij{&j0UXBg{y7cI>OVFsj|6JLbM7SNg1Wn}S1%bTL{n
z(o!Pmc8(1Xyt3zw>)87%0ZJ)jzARIY(#3?5T+twExSOhQqA3t}uVOc5X3H5t!oXs7
z<d!7++s(~e_XX^{=BEE`wl<vUtQR?3odg&*u9TpL7t>XEuDI0IAl?U47iWDW7kmIb
zsOUld5mzXn;ThqAl_$J`5huX4ZjK;&82^o1-`&z@29%l(2NjH&*uTkO@37^Nd)8$0
z-|Pz4)Yeb|r#bjk>H0rsaq=x^@^$>-w;MJbH0@J+wlv~Bs_@x~;=a%QYE0bTcbrZS
zmq<S}`xUVDJ*MARL=#R`#0S+zocQDy2K_lR5sp;~dNmlQqj~w`NzT1&U2K-Qc+}pJ
z#aivUe9DYSKXhj7F2TvlZ<nZSR0<=cY~52p*Q2!<1wk$78^kE$!O1IV_=)Yte~ROh
zB7G0JYv^HOd>AtIV!qCbN73(SN@m<MjV00Fc~Z<^?O*<{E!Tsdc>M4ZBT;A0jN(ca
zGjbUyd#|dZh=4tc1_48d;|`u*3v_-Mo<4aNb3<Zh+`izWnB0Hl9eoWtyd6x~##@zQ
zY^^Py$=NoxdK5@l>lBJbj97FpbvI_kQoTjcqq}d4_BQf#rBIv^-2^xCf2@~03UqBv
zSeOdg*lyIxf|$7#8^#GCV(wD`oh~TnKBm`Ezwiy`G%ZXg>6h`hShe@vi*{q!lU)2`
zt26Z(+42Oy9RNS*myc0h&&K#u3i@xcX=^00q756c><O)LU)3-4E)q#Q@-(2j>9n%U
zOmCao7sFKXIOi17$J2QD?9DmPADtA&JnEuJVaVeCqLe%|%+f+wdJ7c$*17iDXjpxV
z*mq7i?F&&&46tv9&V3>-gXOo5Nd)BqBUFzqA08qn?j<!~l8SaRbD{pSF$imT=Gtx_
zM@r{gA?+iCmLq!G;G7unc1>t=d8(q;m)wYPKAgaJOa3Hekv)sXRsf{15{SD0>kYK)
zqGX&srs&Eon(?v-<q9U@OB`rgXe|P4lgCv&B+2oH{c0TSU%+0`dgzM-Q%>FpED!Y4
zka2ESp#9B-F?~R4v0szapFfi)$~azmbj?)0ulinUVZ^#VnEa(ZK4<N8Vx|$5kr8z7
z<+JclGJXX&9R(KUpStzfZju5xJjIWt8;;#JbN?gFCpIZZP8?yyjczw5DSTLHF#$>R
z5L^??8gtPUj#i6%bw1>K(+8kr4i9M%=%{XBG7N2J%idgHgxjAVH#;A}gsy*X9GGvs
zZ)?cN?QZ0(4AZbzJ}>CA$L>3Ti?qv8&vxp87VKcrSU320*VSV=tN$P9FV|@S=;I|;
z{O3bj3sOz;%QOC7k#bnv(y#L186Tb#66dJ@mp+o`%ZD2}#hQe?;Q<ryhGjcO&wu~=
z49!>JGey*ElOp`a?W0!~+Z*>0qYNo3jq#>uFHHOYbINJs-Cvk@iDQ{0AqU*6^~3g9
zBiIqeubU*IeVt5kuc<`>F#a!!wAJ+A8r{mLW`XCYX~l&ClTq1HAEz}^LyJ$L0lc!A
zz?$UmN|A<Q)#`woH1M+1r>|?mH0I6v3zW6q{`P~$TV+eDw&>?H&fbaq+%X-vLx&cY
zR#Nkd(n6k$NnPH|kzS}=Xmp!q^0!P!-l>3lX_DL$b_Q11!0*pgNt;tPY^g0<e4bbG
z>LN+De4iKC8M&eXZ^179cHbq}{*#L_M|xXz{AX_%-#GH%mCA`QEY-gnp>6!yI4isz
zD{xlVOX6iTngzcDGA}Vt>Z8hagsF#r@K?W_m%}H&!9=fJ{%8JkMy__^!5z~YxyAPl
zH+t&?f%(;0i^P4&8=U5n)!iR=4u7opZR%t1YIl`w;<*<ln7zi|{va2%$zl(&H)lVd
zwwrhQn8HRyBUFA2CwMB|S-0Lv^)d0PY_}ZRry1wLF+7s<-@AHANXjhl=>xszj0vt%
zBwgEJ@LRnU5Ee*CpdNk2%E`MehoP)@Eo0RL8Vw5N!+QDeM&Rq188WPE^dTcFsBCx%
zt9aKhIt5@qTD6uR=Zbvc;sGVzO_m<~Q&W>|Cx>940`jjMkIG4%uQQfhZ#JNi5ADF#
zZ^_@kyzm(u&8IC6Bvn0h9$|s2K=$JeYs052epz9wS^`_(pms8NCQP#sGUR0BsYxJ<
zD-g4Wk~4#Bw5UYBjsA)&WTbUj4Cv}d3YUnW1vocBKgWgj(*RB#9(j$GGh*J~xfkGR
z>u<WZn2#80JruR(o4Z<U)wiFsj#1A0P0e|C)LB@6ESZ;Z2W#x*MT4?4s1nCOyFdnT
z3?M(YZl>_sUzjV-hiRK*Uf|3h6bU+geT8bAu!Ln}yYCzXkEpxQiBKv;K}A)$km^T@
zWcMr`j+TIfD~g@}&mP9s!a5@wL<#;%>;ou>##(^9HME*2)szo%k_l&n$B8TV9hNMX
zuyU(zmNpm}_8z2OrtPvZ{ns=JK6S8IbC&%<BRyxtvReFv%s%z|Cchp44FfU~Z$8f!
zNh+RIAGw$9<GlAhc9i_T=Y7zPSB;jFNV<uT>5_4-zAr^Re5XOk&#_;RwVu`?b_7V<
ziHK?7j8a?Ok9VTToQ@vTxJibUZ8U}*OGiOJj_BWVFlh$%SGjf!CgaLTVNlWnbI2|0
zfm*W<J1skdGS{Me1(0F-tThxU06vFZm|)~Pc>gaoUo(!~@sV)Pl8>_mU=dTTyPe_y
zQt*gqvVf7iZrdn|EH_{}n7D!x@rin{)BNbFH%$<jZ|{iA`6<d(e$_R$2Rqxoy6VK>
zCyZZjvs>)hEG5~?iQ~oNuVX1WAnY~8$sbcZDmcXIT&ls}aUhj<F3%pc`nVmMZQgY-
zE~s`-P==l-@tA8rUWgtdltp#o)7q>I;8k6yUQ9)$iOcaIC|O53yGod^s5RV$GLuR1
z7HdY@<%pUXUatjuto>{E->Wv=GJO;Z!&r{I)G5ltHbl5%FZc#}I(ROts5~GX{K0G$
z3%lD}t0stKaR<Z8<Tm2>32N0qe6qol&5$T*bmMmXT>n8(SGP7S@j>Y)Sn$tGrT&j4
zDfWKYjJl)x%=qi)E2PM~cG=48vNP3ESCJ(R1$$HwKr8*Ikf<i`h@h_#F_PJWfcz1e
zTe?kmKrTTh4SQ4izRt$Y1HNs0N~6nu$eW}#Q=@yDgzxu@iW`T}b>LhcCBcj>P6JPc
zp>dA{HaPN@JSElyH}FuWhx3H%@UgbvPg^+>$bM(9z-!C^_=yM6Vca9>EBL1mOkT^x
z(|q|u=LDuOoogcI#i~!PblR||($yd{U6<O4iXABW@xAQWe08kG7h`7z7P;m*y>A5e
zm^OBB1b%wpJ8$bc@Pnmyx(FK<DU*`gq&Rkl8^U1hHgn>8rN8j@0`j4M0R?4lglv>N
zTl67bz_;mb9;8(x7Y~7?<hIXDDeb$oo%ox$8Gl{NP?p!zvrR4*-BQbxw}&3^aGae7
zfB+b!I8mhejU$mq!C_$U+x)Iwskn+i4tCyKn9dFkvOr>i5H4B+`?Tl+hA#E1Q65fL
z?$p(4#r7`T25kY)Vh#Cc^svNwW4-bn#bJ9`Z(91?+>BXzAB?O|nB*nlc@<=Zr(+Y%
z55MBbH|Nwg8ipIexKN*`NfVV^vgL)pNy}iZdJH9lqd+_6HwX+RaqYd~A(d^HX6Z0c
zECd@p=8P&QOxg}42Fm)9Yb83N9kxWc*)4l!5EOI3#K8h9HHJ|fN*U2LC-)I`q$NCy
zAUQBqm|)NiMhmgh^0uD{`7_|cUhs6+TVRcOS6MDVs6C1WY_5C%WO&B$5yr1|Ugw*j
z$EAMcLWjSM9>`KBy+*!eU=hP#R*{Xkc(Sqv9<6Y_@j3Q+QbF!yjESh+x8E5%ATQz2
zXt&OV{|iZUT0fFTS_~nL>43Asq!+IGUhatj&}pZ4>e&xRbWwhzGV0E*1Iacm8VSO?
zteV|M!W(_bH=eo7jD5v%1V}~>HT%JA{=hXRpi4h4E^d4R=h40fIVaeD@D5n@?H9x(
z6r$&tNiWWDbzi|&JrzfESF|0P{Kb3?SgDuofz&|Y;AmKud3n7)Tu6sxy-p!K-=>_k
zwoE!?V)C(Xqy1qGT>{_Hc<g+Y)q^E|e6tb^6neu|MbTtM1Un7DN6yBKlMbze5l&!L
zi%<FHM5}K37u%$+#_GH_x9A7z>&=ojVE2z`77%LT4QZYbuaamX=d#fVN@h@o{RZY#
zPU1X+sU_j)wuu#ABUmk5e#b^;hr9h0YDy<<97BQEWZFHND850Tt&X{J*YehCc%`MP
z%(wl;&w_hCzX%T<+ql8r>l+`SC$ln!c@kX-Qr?rN^S@)7KK2?LHoRClZ(IL^@-5qa
z=<l<Whlr|OHz^Lpr534D<~eIa?wY8!4r91eji;;5p{b545=N9&_UxLSt-VkLJ1TJ8
z0X-A?9|E&;?1JIl_A8-rCp+Y?+og>2UzQIZ)oh(*PQQ~zLp5dMC_0)GT43BgMJLdn
z^;8&baL#Jq&HKB>Au5=OB;2VDpZ`ty`PWYoJ+`!}(<{)cWh=~u3*?^vLiU@ZA>tkS
zAWXN?xOFDx-DrrxPp7%G19q_$n(-#dE(ZkPe^YuwD?a|F9F|B<pD2Byx+2&Lv!9l0
z%kWv@?}Nt7HRst;TA#_cWi63nqKNN?*XbEYF1hN*s$92zZZpU5%U59>NU={piY#n+
z<fa!sutadH{#sQWW(<FFA(B&&$g~(VV8=@G7|LAwXL0<Egr0G|K(rZcEMrbIu6OwJ
z4I1f#E$`qCKSaweBw1mtjr7q6E3GM4q$cHHe0g{m>apK-SobRXfG-w?+TO#+H5y}w
z$AS+3Ufb<>aUHFkP11ItAs7S5rMNt@Lgp=i7Na_=r#5oq7Rj)BW_)FlQsC8Mn`Jp_
zukFAJo9mJ*87m$b#0}N+2b8q|g}nAHSMkm+@&Keu2bqE_x~BzED}f0*L$^?4eH0WJ
zcuyhsGM&MhzYFC0dO<jeiPY`VTH2pRm$`7}k~jS9e4Yat=0FBv{rgm(zc7-OR$V~U
z;biBHT|h~Cc2Xoc;@r7k#*oiPGBkW3SJPKju1FfIy`E<&g^|hFGx?^k#X@x$RPbfH
z2bxO^EblL2d)^~-_ZMSBiO{YO6-nGdj##&<9{V*}VVbm#l7s~IuvIMIRz{8b!AeU1
z=>`~rzW&Y$QAg^Z;r@L)Z=SBp`a|{W?m{vO-D=r8Vz<HRrcrDUw8HJ2cfMUy|Jak%
zsQQI`@{xNIo#5nuJoll0k%Re3Lz1C0xGCwS+m8?t`r$t9Kq2V;fFu1Oh21OCNPL<S
zaNDx6)mxkD8gCj7@4?G@=E>?R?~=tgp%p$qsBik|t1#0x1EMB6$Nz==9bC<|ja?xR
zRHU2qzKQqO@APuyIt2+6#l?9Q_PahFb8Uof`rZz5A}eaBUk2b032U)jLcxs|e}fnh
z&#O1xXaZ;`)5GQWoURO|Qm642e2Vg(!;ZcUd}_{16Y1-&KDs&U^-bP-#HQLp0tK|?
z70W%A++UC(K6*G*0HL%0{-i0*a)~>mxhYirD|!1zV~kmXHV(1kh$}vHIsnP~;tmOo
z9q_9Niv?zVLY=#fnWuW;A@yrl9$NB*m6l#&sV0pK$5%|(y{Y)GtGVY8*1K}aF)@GX
z=kYRIO%oosW1ApV_|04R_fdo-gJVS-nnLvLT4q6`B$}tAkLdxV9z8I2%W`!B{H2?-
zoH(3X5WdmSm84tDoom^f-o50PZF?#}^y>9JiGgWm-I4k;@q>Ykng&(^6Aytda~}8r
z<cp&qb#a7DrD*j7P-q^v$m*lG40X!F;#_;vJNK>1Nd$#=q*f2pyxR8Lv9LFdD*lDu
zV5Ht@sg7(j*cN|v&bI!_lH&?C_?+m3_WIj~!F0YUB<uQPmeDXkz72bF;X%WnS4Pw%
z=x6kGE8Jg}nqQJ@p6W&J90XKM<@7P8n4z1=$-o>^h6J6$5OwQ6rp_cb>uO~g>9Gv1
zoUZg}3^33l!VR_Y-v#zWrMzihm)7TYl|^rNYe_b{w9Y!wq`eQ1>0D6#UI&fA{d2T=
zMI2y!)!rl5c6r2GqBQPJq0L64o$K5Z-VTJI-9X(AcIlhId;~6&BAm^aTjA?2@e%5A
zZ>~+z71HEDQSVY%{#Otgv1B;z$npvwat*xn*@R`*sW{tG8{w}nrF8XJ5ogRwp{m90
z^^&ZD-PetVDlL=18l4y~B-;%DhRGR)M15f2?!Rf_feX{l#8f?F<YE-DegC(}^HLiY
z_l;OM2z2+iK^5}_^lfxUOy(<)-yGIsiV+p{nPouER1hm`7>ib@AOQhr*OI%|5>ynI
z!-nMBjqu@ebDhRZES#8F1x>ox=i2*aBEn=SSM!k<wZu~Tlo8umJz4Wx<4&#x&OBwr
za!%ppQhMMRPjileRTxWGa)qP5j<r~cfxnA4cBE4n@ypPBsRa1o+4|*~VDU`n9Zr<4
zsGJdt!)yL8^E|-g>8_7N{FOztn)#1I>>H~k24B#y@^?bpS^AFqD|`=3lp6n_y5w;g
ztDxI&>DEfd+w=`m#@m1gg)pkT18<Dl2}T_j;Ew-=BUvVrO96-&4QTVwfjGk@kMAur
z;oez^^@?xWEHt?c0pX4UO{f@YOp_UJJ3pd#My(O$tvi$8g<x09%~Tq9dFxKjSBN7p
zH^Qc1`s}~5&)vk|{IFWnq7-X<Qa@CRr~Du2OLa^csD6Z<V#Pa7(;D@|t#Ku!^3Cx*
zJFc<E)$SagxY1_=BuZB#P5iZ1Tp|XG!s>ijd>?}*jvP5^3s%9yI7VErkqVW2%Zb2y
zBuB#R-1quRdUVvSM(&SoHECqkx1CQGyK0{!O5?paT#GFxzo)`IupiwHSFTHJDO|<o
z{P+lkncBtQc>mDP%C+^fePv>|)R>LR_~M=mE=RI{<PzSb=PkHO{s%o>aM}^Xa>vLN
zQ)bqPlCgDs3n^Ukf6vFE*goLaVY=+HnQtBx=Bq@gkbth$#AC3b&(e0je1`*R(ktfN
zyEE1T%VHZ4n2ZkfIT_dHUoNiINcX{l6%BsbWy6J&Mqo@&(2JJ_u`?N(x2w5akvHrh
zGYm0RxPp~i8RQO!V!N&eS&7uKJ0%dE=7>*tdGIg>JCJLHhb-;9ZT5>`<K(Xycd+re
zl|xfen&?@UnQL+&ZEQRuI)W`_aOA855wFwlTu78E3Cs4AnXQAl)40x>YT~edYR%!_
z)1)*?R??A)!g&WGBt~NsWD*GO>~t2}$~*DO4<CRmL*RgpYG9S<V1+{nORK+e!oDq2
z>i=d!CjpWZT*rT4uHOdjSt6*PI^%3GN*-$6D*^erZ$J$5*riSX$suY^gr^)?U}bJV
z8zbtt>;`2<2^F(ZMe}ZfTe2Z|a#LL)*ef!n)l)V}inwt1CE_w!HC_T!0hXjt6T~*q
zn*iz~)GmUmh6?M!vzt0h02$$f0YP{%mU@u}Sqb*Gz)@lO-tfGNJY#cNNP@X93}op6
zRI`;6%6j$rzd8J~{nSS=Kih#Jw3^e}{evv;KF<qn8TjNS(EM}cL@d9n7ZaT>(O!og
z&9onL<T!m?Nav$LQ-2%kU>?)>O%kAs1rz7e5B_h8!9?8$(;a$MCtt_l83&3dpDs?e
z;Y66dv?nR9V5MsvXwu-gPgc{s6Ui9@Uy<|{nFO5F>Oc1Wf=7QL3yo3loR#NVmkve0
zx9PT4QnwC~faY)N2qFZEc79d49!`K*I`Cf!Tj9Yc?*o_-=Ao<iko|Qbtwxm_pvBz)
zuaNhozQ~mJ!Dcoqt~btgykg*7Cu7`ftxp;H=BSI|xxSxmc)ByBbsh-7Ze1%4@xl{}
zRgzA2Yl|PxWX_e`_?ULbkh4ZQ+-I$Zp=Yx>NlnTpDB4Z<btQk+U4H{w+d<=jAJ)=o
zo<wra%Sa2sVc-;3wzWK=={F!B)D}6&@~}~{>0ySc8Fcvv<hSzl$hA9%BvmN-fFWR@
z$e5^r{&V75v7w{oD}1!ye?Q=3Ix1T{aFT4tzpT}+gWX86y-RJry5OBS0K4@*di*pt
z!LM@WP9?a3sS^y3a0K7otzW&oIawsPQSppXQ{mTh+fLCg*2I+~C2mW-WaHaN2T;gU
zx?IaW4Y_(*$Jw`xP_O`e<4(*$<AdKD>0b%hITOdllNF+!y$Ka(>~kjaBNKv9(xV$K
z1hE>1_)cgC_zwk<M&RJMiPmTSx7!vaazgPEE#^w4MqdCt#s{0ef?ekzIsv^_%3Vm8
z_5p;L+D1;O`?2d}Md7)W(=T5o*-bpP4==<KyxQGtpsJp$ZH3?6wM9y3w(pp=-s9^Y
zdgW%tXU|g09Y2oaGVEa@cD{;4Ua`y7HxhVI2ty=e;A90*Hg)f`^xBK+n6nRvA}88n
z&d$>n)^J`2f*{J0$15;_M%A`JJ28cS9{`>_+g0!l5?f4gw4cwHBx1f|WiZ~j_cosb
zMt;$OZ>2m&nN?Ywv~{kIWWhS&NyXamB6NqtnW3tKt5}7OKBCj_CCkw*34Du8hh?Hd
z?&hXngyx91P}cf?LCzXu4F|4cenv7fBwBPqP{wNsOtUEuKyv})(K9F(EG#nB82m^_
z=6LH4b~oc~1dyRbixEvsq6GmmF!MPfBhZDS3_37<1LASQn^Mi$e5HETk5s3mn`CsI
zaX*G{|JxY)^#x$2`}-zm#=x$w8(P@iu(pb|kQ|(e!%Mscwe3jmk;oVrXN%&}4>cp?
zjXLJ;H)@y++upcakc_{Fff>#bD;B5n0|XC>3TPvi=H@g~9vP0URkN4Qeb*1C>zXC7
zDF&5AP-1nX3|r35fAN|Yk{al<l`NCx(UcXW2HCE7Tg8t))iFI(d-`I}-N>x#ILzl%
zlY?3~KwR&u2_wU|IW$Shfx<E2x#;+NsX&*6qY9HfDRk7^zf|19yZVD-HRG_E;v4nR
z16<v1oPC_H=^aq>5gjr}Z1#D*(?aRWWj$V?A26<&$bNuYjvS7aP^oy6Au6d`4^C<v
zOAr3A?9w&n(Ws`Op8bjmF<!mlupW*XH0#ItpP+`B;QwSa;eIr`+_|e+Nf~8;ao1tT
z0DV4Kk0xIO6g^fM0Xc(bxhDtt*r;8x^h!~$?atRjM_X3|Lhrd~?%v-93xUu>#xg`v
zM5PFN@Ab*|iisQ=zO6z!>&5Z^$I-dRGu{4w{Id;1&ZeBtr>WE!Io6zKhS`vo+bzVV
zlXKEtZBAtkb1D`ZwmDadO65*6bim9~xg{ZzqywGr`}@24{Z~CyQ=jW|y$`R~^OdLb
zg9QJcx+@C(GXB5M|AjJwuX+92W3?Q3;6SR>XoOkfUal;@u^47GzGK@1lV5aYZI}Fu
z$uRSQJtutjWA<@8<Vt)8&KhbOr#npaZg1W5_~s^(bUETeeOz-m!<?EVH0e86xxoja
zOBaHuLEjf>rM#9qoiK9HdN@_)VTwv>$ZREJt!;%2+p-6-5S2LBt?{hStV2tO0s_8s
zG}t6OnFJTpe#7&Bn;^w7A7~qOmm=i)IbdCgIYqUzhIDv8(|Cw3W1h!LZj@AMyNf0Z
zS_A~;7<MucclKC=y8+_^q5nlYFXQwkjkm(g=eW;o6hEcn&na|daZgn?@OU^2vwCUk
zW_2;eq7)rV9xMZGa*Ma!L6v`-ANDnCxH$<_WHEA|gQn<ba6$BAAbS1Sw5t9XYv{cu
z2W=0ye0Ir!<{V(u{x1!@I7Tw-84z%a0u0Oe1h8zXx&tc-Jesu+?5%w);QGQg?*QAe
zb+DjL)>jX&Kgey+*%a$;Lps^{IO2Wh01Ads0_iF-Wn0y%v+231z^wAQwsT_c@+%K7
zl~cNh(Mojvbm4zs;{A2mZw}WFGIA<;G6R@0f84<01m`k<QlM3U6D0@l!GqQXI=|k)
zD3=l+a%Hw<9hT>LepvwX>a+&H7|p-}Yc`@C4#GL~DfvjMmzMUm6)p3{M3UWE{p%`d
zBo(&{+05s~Wcwv@G<-ck*hmkZt&(YA4Rik09GbPp^Zr=mq?KriE|NuTX{Z$(Ydm)d
zxYNir1v3k0GtW8(Lu5F8lNCaM6Mvp$mfY1RHAp`b<pOMdpJ*v80XfkyN+|@NC4u*Q
z2_kbfUf9G@x3(P%30pm+dcFr~daXNr^Ya&Xw!r4Yr|cUTKHeqxF1^K`i16ku8E-wG
z@nzzG6_~j}_wwdsur0QXiF$NvPCv<}S?*SB$LQ8R5Y=Iv*^ETVXH{oW2GH`E^K+;f
z7qQ7b|3-*j>L#~B@mtuac!w)jloNErhV^IJUUkdOwTIs=3)>3voD2^*CZ5j04~tAn
z>_b(g^4b2bikDWFi9WltS~q_Y{x&d5S{^>rX)RDXz3LFC{p$eO-?-F1aZB`{Nv4`P
z?!}=}SxRkF-u9tqWOJ<S^e=jZ&z*riQ;f0DBXh@L5{7R4n!xkZ2{B!%B6f0T21*zj
zRSt^^=+Z|3wVttU<Ij?cKf`A^nxeQ(3pHOa20g)K&S{h6taUV}n38(?(oCM`k=P(G
z6tqURyMhs28ZsN!_H)^l9%C{WGN>e{?C-PHQC&YXq+0jQzBiklxwz{!E$Qx?&Ekmp
z0t!CvY_OK60S+9jN?6k*X8^O-G|A*Rc$n~I_t>=MaAP*#9LD*;bzj!(t~^Xm<W!Bm
z@Qp2Jod${KhuPW^K0tS<i)7AZje(RlE@(~}8^O&ff-KQlVkm2t4Ki>4r7h#`m*A|h
zy>iDxCK(%<<KQful9ZL;PODB7foUrLVqclg18}A0-~iLgX@P5xLnb1rGHKj{zOG@r
zra`Euu@cw&Dz3?e`^S*A+MLZ7eX%+39I;fBoNw8z_5jt<jQiT^t%$yl>euTQydZk!
z1Su2!Y7^FYH@f1ED%46f0d(H!EnUGN53=fB_02><m7|2Hf5A~b>@uFp{7rm%#{btE
za{q0)l}QcU*#ipLYk@q^9E#_&Xms7ir#Y-^>OY?7;6Yc}NGZu7-YYghw@;B|Y@LRh
z49n2!Y6`0%CBy#py7cy8S@vL)u-28VJw?d^+vz|PFeMs5D??UMcPn=20Nj>a672^G
zK^w^u!$FvG776x!Oe%pry4NiIC<1{O;3`CZZV<o$!8QxROoe<GZ?}?MsAhR>GAgw5
z?F9+b8#WSnhhvBR(v3gMQ+oM!mv$K1#Gic>Ssf=vq<&&~y4&Lh;%iUyP!I+v*}Dg+
zkm|H(L(Make8(oF_(^&Cn`GRBza?j!=2>~G3c!DD5wp_U+N=`>&68p5SN7R9ZCA@U
zPADbsD7%+U!a&P$=5(c6LN%BEQwk}PfKfWiI?Q~Hx1q!vzqEU6gYhTj1WQ`UNeek)
zoDx6wB4PL(`wvf1!x2a0<f+h%8%pvz_v+r}#70e?__lDlA*1b)H)9M?z6XFb9H<9+
zTdw{6?18(@Y=HB3qcl;^(jxEPrSjB{&W0QhGv=2@mW*3MyHoNviMpxu>^HHgkB-O<
zHi(i3Dd6^0>i$b#r~TmNH6^6o30_M3aIH3*D_02j^K|&|oYlZs+|VB_w|n5m@lEE5
zGGP^O`GvJTHXOUo!lIs*3x9tdJkm}_#SdO`l|C3naL076Q;T*kv~6bgvB6I;KmGRX
zMM#8a4jWldcJ>E38V{QVHS}QAs?Js$TMCLc9gbY#DV!R6rOyfOZn_x4);E#)N!Gr%
zW&GbK##!8|t-IT<YQ94&l|Amd@j4@CdAfDDwYTbiJj3RC^Fm6M2phTSZ4gcCL1+%e
zG}{`ElS6~Bjlak`7+n9fYL=im;Og@;%jA$?f`wEmt~PaV$49V><7{b&Z>s}a>!yl8
zLO!#!o!~-&(S#;GV8&fJz$_^X!SH2N5|LP?b_?7r1LOd?jH_?>NV@9<(Ep+*WlEQj
zS~#PL>%MkV$)X)PGYeIwT7w4kn><hTNsidK9$=?`H>8~;*fa}DhcblQXF=Y0k~?j%
z9kL24M*y(>;G)OiYutSU(KZA~w_$&vXBC6p$QY1Cp2XHD!!-pqN;gN_;Vx(Wcl1{3
z#|nau8lp8sq2}>5ui2@%JRd|mo}farDAKTQERZFe6l)sxRa?e}7NGEhJx)8CTfB=Y
zs~}_4<Q}8(u+YCZzVxbVzxE)N?!=3!<_4`nrv;Xv6vnXZi+;~lPu>&)nWr*X`c#jN
zYGa|(?!#uMO7L>(lK=teuk$w9Q~j9Jy?4Hej}ydCbcSvi5w6s<fVGbcC4yw?*krDs
zhOMG8^X!pzz*}<QR7s%w8*94I!8HHk<vHRqxS{&p>375!)d@~*D#&P=E_wml7Y1#4
zAQ>m`>q-J@8TVo;@iR`=2)u=x8=U7FQSv9Z)0P0;+l4D*C<TQHvd0Rgq^d3O83k4A
zhfUAI$U7YAN>>{dL|#^a*7l}sVJX~}bZ9gSYID*S-heZ7PbP+ar@<n!jtcdGbhOg_
zjlPipZnR*C{!4;<2zuTlvEw~Eu_GJEI(3J~26l!mPSk994pLoB%IE)D6Z%%YE`~D{
z1ii51VX325?b{JU;KElTq5T@OybMb=F7GyGq-kxfKt-}NoJD(!++waC>?{~_T5E>S
z)}3QEsL1c*4&52Qez!_>)|#kmHNNs9!akMyusXD(?wQC0G`B&CR$!w;^<CLBnR`3N
z*zf>;E51xDa!;iUcoa|V_H^cy*H@|_SXnGj!`2}O(_(_Bg-7iZyQX28Vgv~9K?<OV
z>v=+su~Fo1e>zkMXZE(W?95+_mEJkaIuVy9D<ej1pzP@h1o5z8FPi0<{{a5j49Tvo
za5ZfHupJoijn<NfnIV^>)^GU+o>NzFZnX|`nQhm(nt9*Jc_x(g_O5yC)=)1C@38L+
zt{2u&m$PyhJ{a`^%kp0>o<F#m-bg!4>=Mzt)z!LRt?XN9%GO^!@*cQ>t<2b}h4?Q^
z;bzyfx7~8)+a>KbpnDEvEq>->UdG=za)+@vQTA@*#m}N$Ht_qRiknN?sJk(Ybq_8w
z`lr$#*ZiI*i$CJpFm=iQIOyy)*o~6kH!8rCb0NBK+e+XAcC&lS9`7GGxhL|g)RtpS
zCHw&OC*=-h4&JTgc9XJpybDFGTOb4_u1V7~khBW&uwfgxmlhb^e++v{l?u9>p3~Ho
z9>N!zyJvIdGI)ZO(T?HR?9G{#kCJ~;m@^b}ucAz0h5KoO47TVP-}pClmt<^)Du?b;
zF#wO;^ZR7O$G(X%$770)>qaBHkf0#|aVxl0TrOZa+@(UW^#(VPxjYyGf*4?aW?Wgs
zCim3oB2`c<stlBA9}&PEI&WXu>AQ2bi+vx#^2Yo%F$XWkjGm;a!Rhz08ow?4p+|)Q
z2$#!V!w#3#{isJ(e1}x{Ec=^Ey3F;8HwxJAtkAriJ+dECHfK!k+miTvB(Y(0iBrnk
zJyAZkHtZmz0Wt`r;Jw>ye}|Ynyu*^Ay6k7lP_bT}WGGcHa8qk(kokC4Ef`RpswV5}
zXYe4lE)k$EHA~UT#(I+%aM{)Qxu&1=2O8X!_ni6hMBVi9p)%%3qGX5DFmQwKyC?vb
z$5^GGp&B_r<#p-UT#=LT${H91t25awI-=c_bAOdMSjQ(x+>Z%MzL7A0OHUh0l@d;Q
z;7`axCT_DFAOU|mosGfW*XEm{ps}azK>fiZK|RN28`&`2=>-;;_Za>Xo4AnIc+t+w
z{x7>AeZ{{{WiprWb=P8#5H5AEFX0uA)^tA6IUI=C;@SDVC^Q+T^(Tw{83)N-WPg;3
z;Z3qhS<?MUK{OaN3gBhA3*~SXYuH;}dUH8Iy*B-mJgjlU3m>QRD3uVo(mTsh;{;kG
zCCoF&KrbV!&+S0@?d+`b-#D9N&@vTrJmy*%7Js#X9n)BNz>#brPk%*6*^c9PWAj~{
z`aZKbAMtW!#MouPb{nK-`_2LRcKGfe6FYg|jl;!u7sPeszGMF^?=`E@u$viUQ?y?N
z>$5z~|4ig0FR?r=A{8s-vGOG+=x;V`S9Io^)sJuaRr%><=BL2K`q}RJ_ao1I$IRj9
zO1%7q_9U4@R)XzaaFY!87GxrBcgau1b~A45X8bdQR+ojjmYiX_OoC;&rz^KYJ1cyE
zc+GHD4ybvxC=0+hW>KsZx7vYR4)#rQyr*N;EyS%~>Ap}Ukaw4*=0BV^41j}Fz3?qZ
zx<c*^!xt=maytG8F~peVj9@&zS#sz_lT47~_SZgU)V`PZJTgoU81G#AWuv)aTL^v&
z48Ihlmj~N-uaa`3b}ip+7XZAeb@PAY1-(C-rD;3tnT>M;%9*P5wih8@-u=l%U0Y~@
z8+{yKfk=g5EhUM?nZiAU2OwOMEHEvQN`ay=gy`W90{*y#g^>3^eJoC`dzP0f0#U9T
zKujjOt&@=-IwXo8aV&KqG0Rt-uP64sIS?1?K3GJ5Yg3@_u55|H$y3~2C_4Bb6tzit
z(59gv^<l7`=O9ngep$mXO9AU6G&KhKV-XDoIDgCcM3}P@<|v@QvcZ%?5%RRd_~+m~
zwN`=t5VmtI9Rws!GLmsB`s@5TeKDL_`%JL>lnbSK9ek8CeM|nLvF1ef9$;k`Xv{UX
zn;w)S9|M`qZsLt<kGfI@S^`fZU(AyfEyDj!txgW(Cd586q@OD>Jodfll&(Fq3_N-}
zZ(cg@UG?`vgyGu6d?4+xk8NU-N4uso)k$-mzdFfV(I_(nbP2<?M&OfL{J@`OAXWmw
z6m3SLJ_={3+^H>x-2=f#FG((n*05a!&rSwigcZXn<!f~sk-0e~E8GrscvHc~dQpV$
z_m+~}v&J6@&TBlTu#OWukzMjfNA-`;gNcD^ryHGwMiV78_Sa#DTdDZEMu?&c`tX&&
zoi%v~&AS9&loI_%$TUN(m=GjYg1}yGwzwCZR+2SDz0$zMsbJHWa14?=4f^$FeU_og
zQZ=IDHjFqm&nYTcEwL0$ZZ7q<yuGtN%NhK-M7TNXCYEjbMh^bmw2(X+DVB|Yq>;>p
z2C>l1K}j`W(7GBig}eUDf{sj-9T7mri0XBLei{n;kQ%IQK+`TXaYx$)PlGX-mDOYg
zV+Bwb%>;lo3%Oelyp@Z{H+W_q_NER}u`z$d+R8uS)cgb!ah+CBu1cub&GONJ@B#!E
z=UOB!&)-Jy1aTsP-YZ1;KW;yv??sLMUYoWZxc&n3rJWZ`iBSx5a&9LEI4gBGd=_Gs
zgHa3)jJz!<&_o5fQ@ZZN_cxxaYj3xFlwrvMNzw1lHtG$%EKI^sn5;!tSzw;+$8hU2
zc=83pb3E^$XD#og`Qatq^H26fs&tC&?cUGZaFpY(m8Mod^$sh^eQu6l9l-47F-iLi
zL(GLCH63hjt+vAbU6xJxQTVc=9lBSTx0Jx*I-=ZHbvl;Z7NYXJ{#6`)qwSE5PsIPA
z+bp?KgR2;EC{xqGx;xx+1;G>Oha_EjE01g%Ms(<TH+F6Bb0wJww@?T`j>EFY<qZS;
zmeTlMu8ij!=h<tkF^`E2&HMp&kHMYWjh<Oan|WTM_($GD$>9yoZO`ksY$a4!XrdPc
zK%11r2k->C0#B|Cv<zn=q(NZd=deK+Dmd++gF@x2bHT*rfJMTeqEGpQUc>TAxATn>
zjwDAMnz*r7o0#&+MVfOtgy@xLyGu6)JYQY_@Sm5IDO3mf?d32WG6W1^Sf0v|%+Xa}
zr%^0}Kmqvn9Jdf)HLf<q5FoIwnn~xuzEhwjLh@4-181lDjjXlE1lceKXkG#r0tNXi
zt_pzcB?&@uaGIX5N?l2@fjw7da9(9in5O*zm5W&oJNc&NQzG&kZG+bS+sA?_EQbEN
zki`3pA9NN$Py#D5Iu-DtUyN+#N&J6q7s)(|jk4=SgvXqs@>3t^JIysM6dx`x;t7ES
z7&^4=H4fQja0~^UN?cD@5DF<tQHG$9U%={=;MCxRV$erUmKU-h`WL<vh`S`Xuw)``
z9Na#7NqH|jdTzSu92H!=<%VlxqjtUe$w9rVM5dm=%XnrsyCiaMDbS32j&^1QnYGXC
zDB!W)To-o8aRzkhUlu@yOf@ACgpG6QBA}|2iea4NA?rZo{k7q;0S#t{obwF^+0k;P
zOcni$genz?hs>D<7714g5Mu$16AOxRk-(NGIjY6ZjX9%Ol%LJ7J7LMwocoQ0OtZ_T
z41J3sk?+bd1iN5heQyb;N<@&AFQl@d4Yp2A)o~wjB6TBA+F*2NZ@lv$ftZ8_E>Pin
zK4cdp>(q=`0Q4R`wxT^qmm=pmE#U@j3@cmJJ;u#;N-v;I*GMbc?uq@$MpV%?u(8Yy
ztvjWdeY)i{nK>Y9QnchJVdkM23UlqhSK1&bz4x+NMoZ}qcc5-)O?pNd>TOGKjtS11
zcGV8m^Vw6d1%xAl`kw^M^D`3Iz3yf$@R*v`wo5o?fBaE*t6D?l-o&jTQ$-$(v01lx
zuijYE`K{KC6Y;b0+#vLs0&aaO7ux5b*JYyS8g|T(SrJ!eux$6~jV*3@iR$r3nN{r8
zv-syAFgaIqlrYtQt(&Bj(Wv}s2M>%SuYbt<*<g8Ta6|fEing00Ym^kezLxo+SYdVD
z9bwdWu3=15;nh3#DaEWQuW3u8Jl&!<oWLXRcPW+_%DtJxcIm#hkk3u>Xgubf)=>FZ
zUB2O3d`(~9!g=TV*(<qZL-ZLs*ykgH3Hwm}CbQuGZU;Gs8`sh1)I!PeFVz2}Y8Lij
z7urGMox=O#aAB)P&TvFQp%;1+fIywrvd{ojCm}p1fE!o9^`xmu=(*z{4N)Bn(@bwV
zv%t_!bOf9lKCe!pNq8ne&+I5FKb2Y@ul-!!d1SNBX{YwFQRB!3!Qf8m14DO_z6OSf
z&=#0|m<99CIh3TEC-cv;F9wvvQ(20bx7KQgNO-pMFl4+ueg3-y8Uz-A-oQKLlfY_r
z)Tt4iY=c1bd!Ow}3EhH0fiYwaeF87wo#lBROHV<x`_WZ^8V@O333$mQ%C4*g1xJz)
zW7oU(?9C04E}|Gk0e3#wY{Qm<)x37xqx9kJ)u6DQ`_SOFHp?(5DGIdu>oYNi_2&^P
zMATnvKV&m^;GJ8J$F|s$I}XMO!?BjpV}pn1iljB&3sqv{1)x1VT_Q3zs)3MY!}>D-
zM+qGX8Mb<bLoN<m1%KqhvrmCJP>`~T!5f(Ua-b8~aRD%Mps~hzu%lJK2NK#m*1?6*
z+0G7Y>PalgePojyG;EJi_MK(cZ#1mf09MdgFV#mA4Q@G1R+;zqkE|m%`IyG;nei0-
zu&k14*+*WC?qSe~>#q$X)QaeLNU${_=v`AK*&s8+QiaCDfi<BWsG#P=A*);&R5VCr
z>H$r~56J>Y6x_2E<Yq|I{(vaTlxw-QvVyN;l;B@dAyk6EeeCS1JSZan;X>rGr0!pv
zJ(U&z3psbzh<e^>FlCNFe!kH!|L~bNupbno`#W+<SjC4B!L7u0+nlV^l?7k52myLc
zR}DAKO<uA;`@HI5dBZJJx`jvQYe23&V)gCT6OLtYJ9Zv?Qs$l_IsdiRFSkM^H4-@M
zM3(eMv~T<0=TVtM55Bgw2WiUQlZkaucu^jLn%BF)egd~`ztxs$WqZPvGV!t~&<~Sg
z8G(v5ZTC~bbcQ;TeR5BK26r`ys$XA|oh8v;N4@phgstL1T%rU9Y`IRTUiOXGq@ic1
zeoDo6CkF+fe4z9kYVvE}&;9)l&|o!~e7wjN9#%mYLT}MO!BFzGb0ZoHGj?u1G<R$v
zVos*)M`%aHyUw2uH}Xg-$1hSPKE0RX-@nfg?TI*{1JLc3Ab;5#^pNgv;EsuT|Da$|
zeqUQ8L#|@E{jbG7y>7Kn&OP2!;2xOf5%G|iWOB;pkNFQ>aT+28!0f!M?QOJUr&PWJ
zHR2)9M7HBQ+)mr}|A#%e@nDzM&8V~C;f5}u-5cwJIEppVI`g63sCu`(h{^Ns9=-6f
zP3cy<xL@bt{P7%*qO{j>zm&<G?&AuJ#EZ*jqc<A`kAsq{jl4vgv<e*a-=z9V&{JDX
z%*mQT^PMIOmC^^=KG<=!NxIuO%-;W2m1#~$py4Z#rToJBnQ8P!oIEsfy=K?N!|ngi
zmXi|{T<NPTFO(lL4z1^P{Okx07XoABb0I#jS<uXoNc(^<&(T5C0rD^(yF8t=OIP67
zNY`1|zpD^?VZ*Zq+n&Vkfw|(urfCzeuQ9P<L)&P$$)B|)Orpz+!LVn$6#qKo-e7hp
z<&Dvu#t*2A|J*qC^U>O*^*pY{uIKg77pl`XrLHXATh-U0xi9*kzjSJBvaB+lS+hKE
zU=6JfBTGP*lgt|LcY!WH#5V&0hv++N3rvO4EL90ZEgh_JU3GrZ-J4){=HMK_8i+hD
z`7tNht~57Yo&<E((z1Z3UB)WaVtK#@L=qwuW7XLI(v{&58b3b=ux;!UGzzFQp99fX
zJfNx=lN1K5p?g2Xz{jBA)NmINj;b2t63U2(omCCvEGQ{DX_rgw3l=d~Lyjhwo{U#Y
z?4|S%Ug6zqC>=nnV2D%lUs08-pnG-AqWA;+QM=3x7*aQ~?8Pw*we^Lv8w8xy&hj1_
z46Oj17GRKry$I6DKpkAv$v6bh_8T~vc<c1n=&zuCjM+F!?N(oJ@zvO3HAlYWO0fgb
zmuEa0^V>`)(VcGZ?R_70f0<0xPAx*`#$@MThi19S_&R>-09{W^LZk-H#4}G*i6o5t
zuO;wrrJLd`vH(>~BkHs>2+;;y8JqQHX&ecDBdYTvRR;B{wow3C7#s%!565*mjtW~X
z=~%PA`iPIL#$Lut^voB*t?N6Z){)x3no;aeeEsMJLAHec&%xx<K&WCQC-6ht)mNAF
zXlq(yliyB`yfnJ4<ky}eZTnL2f{<Nwb!sdiqT!WU$hT0Yd+UayJ8s^lD^fJ#4(8l!
zipzKWIRw@$dNrmabF1FnKf)be=EXQ~$55z!oed3`W>32(hhzSDbN<|*somUU!*$7f
ztApFET^DXW!FQ!46RPXq4c`uQ*^vG!<;AlnqR@~*g6QE~qG<1<H(TeNt&SetZJ(a<
z)b3p3&H6nW?2jz*wQF0Pk(1Rmr%^PEtBr?yhMo7xeAv8kcnk3gbHpZ{t$3llKr5L`
zw$|n;Ce3n0!!jT}9F!ai{1<qct$|<@Q}9n$*g4_{be*00>O@0~&DfUFzujQVp~-vA
z6yEPZh3P2X+1_eZLZs2=mIEXl6$<$hP|*bFZi)u52Wa>12KwEEz!6wUx0U2-)4Mi0
zVaHL&F-K`tn+oSQVF#E)9Ie$g9TOdoWI=f^C9OjsWc-iUi&|ADGR%hbLC}m=2HIWe
ze9XH&1?G~C_b%l|_md;mS}_)+O{N7MY~60HyX^t7A+~-<1glFwqHU@W6@cCpWYg!j
zLYgNR*ojyiTboE$$dws%vS%zUlB^zD<*5B=CB$FxZkSv%M(vn}-(y?94#B2io&Gl`
z>udUhN(ZeUUcs{euArCNJI!*q^ipwt<r8nVC8zpd5Dgwj*#P|MYVP>GoAEw+%lmKU
zm7I>MINxyle$tmuafg~>eLrA%uezJ|bJcF{mkG;rkX30!5Iyq+mX*0%=ogksF^k?h
zG0^Is8`Rh3ZpZ41>pur>&WU#)EGRJd+GN;vN)R8peDqp^#>$^x4)~z_?Ck|NxnAch
zF1V*8y7wD)dBU@Pb;i{)NG5W)We=-vAeKm$N@M|bd`?oPFVXHcPv>c%8Qtr%=iR^p
zoV{KTx@DGXkQ7Rv<@_XpxasP2o{X6(=W-Jd=ZA&ipwUKxC$L2o9peFivhf=4WM^*_
z-A|6O$$cQz&)kp><yOZtfu?MQE0;A6bDk~kV<;tXFuq^|v9pSAwZ@aX5mgRW%3lN{
zfySC{3xJQ*Ab_lwq!RXQCG0D&3$k`v`}NlQP@4)3(KJ5qcRf-Kua&y_87bDR<O>g;
zSeU!C<oC4QSn~!U=pq3M79hl8V=Oq>9;4sL+5oHwq%`bQKX)MAc3Q&Ipn+q2Kvqz$
z&-OzvT>9trRn?Q_;wm>c!-{nLY_s;zl|tYiC-!I13$vYh>Pp#l2mb>m4O6!w8%Hp<
z-={yWAd*#bf&(VI_BcMdsK;7+T*47=H<V7zGS37lBQNWV5yyTn_)z)@VEABwnE;Sg
z$a%Jryxkg>b&-Gp8_aK#1qMp#lRe%a1==MN8hZ8h^n^e$I4IQvA04a5R;dt{nsff@
z`B9wZV>(d7bN^@cV9SZk?H!VtE}R)VLHg8T&x=nDmkM2N<LQG(m=gXcTw8fyWX?m>
zxHc-n7Tctymkjum5z;MvyiQ*#n+19D)}bfI+RZp=(0EXPC-<8DHj`S}t@@t916K~O
zyOxwczw+<q`Wk+c^z7M=%IUO~&l>&Gl0Af9Hq}`ZPw{6r%3au!IqkC6n&fu<t=4$v
zy1@Zwf4kBSy1C!L)W6C>&C`IXz_CTokhME0uvFh;{Othoxo-ds&{9AwR5r=b-OGw~
zbkPNSxP`jI!BiS?pLSE^%tE`y<DU(;yhaIayFa&=uy3ffn?xm^iO~`T-Ay8y$O{lw
zTqtFIp^<&Ue+jSB;3IjA3$*+1!<u1**!O~;)}tL6>7H&)=UAovmX%b7*;Z}=m#O~S
z5DnYVv%BP9n$CrzX(+XxlU*fdn+|rEb0_5=64}>P<??(>ZnlS9u)UWU?9{G(2)Dwm
zB8Yr8)!i687{+I0h1>c*w8$-L^V^sy^4f2Gu7UBjdBQ?#6ZXo=P5wVD8qpIC+9@=;
z)B^h?GF;aOQ?5SLn5<>dMgFHDrueNgHZo9EQ+e)ZA2&nmtOAx|8^Yupe`zINP%_UM
zwtV-bURRf1-jC(RUKU`~HCZX{aIGSr)TfsYt;cZ^<OIyS8{}}urM2Kqu(4h5wJ3@$
zwJWhOXo3VI=+gafbzj$1p>GeL|JjmtA~5x9l=-bPS1rwFN)si;DSCFchcDUc**zTm
zI1`qVJO$j*f+DBc3#F-LBDk6od6uzBdszai_~d)P-h2Vq_|UMiu^6+f-H70&_FKk+
zoT!xdb8mM2jN5RL5c!oS`;|WcR>;g{|EIqrNCv@(IpJ@#Dwr3mSQ~&yC{|Fl7#@>t
zUz9>;<VGj^?u;xRU_j`X_gC=<CPB3evI-G_38SL3cEb`8@!?0v2#6LfP93)D0MP;N
z)^lQAC@80W2gpMEK2}V{nS22QZfl-bp78?nw7az>V^fa~aChX_WJK3^{N&HxqRkii
zC^ed9TNY6`Sz{No;<M+|vD1w}v;O<oLD9)zF>MWj4HyG{0<ACzNnk(#@C*pXG8efn
zz(eJpxq=S@-w??U1&xES4y&ldewA=<cP3~^X@J62NcA1D57ZvIp*_c{NwQ%_+@&61
zSi`N1@^FJdMX7e<H3*V@qU`=7G;kb@@W&;USs!TD#G}6%&*L8d213BKf$JR)gb<C`
za;3x8JkJk+-N}W@l@vYk1J<yI&wq^HVk2;hE+2<2t<_SjAB<8V(o_m<6_4Cz+3qEc
z&n|J0?vuK}ZhpY6x1T_G9qoU*>!{>fvuy-davEDSFfB{c@<nBZeH#l+yA;uIm3U0Z
zttUwm9TNdh$-#4l{OEUV#rr3YHuFA(wr}MC@h=q*HSHS?xi#MY9=U2J*Rr*-&Yo3z
z4Q*R|NHfmw_A=$Whc6sLc2O?GCvpo=CCKHF3+eCeXvEVRX7KzeSprXXw#w`Z+8%db
zpaX!>V;~S1+YNpIdSh4*T>*r?Jd+*_4kpBy#g-Hk*dKsjTPHE!YnW<L^+{<c2wk_!
z=7~1OG33gcD8$%o5d}Q##nd4CXC!!&`3&6uQlrP%L)n|%k#|2duk4Xll%&mNq;(9k
zDpfXDE-BKnkR;KL2Di#$PBP5mn+(1!Fg8E3)~ZQAI)HlOqw&1L?X>eqVn<4l>>*9d
zVC8GO%)<p;Rcmz*d+TYUL*s14Zhh@REjIs<_PJ91FRkc<CpPae+pp1|!AluvjChg-
zBfogqbs81~XJV4^^a3`h8(DGrbo*ga?l6{nR4WHNv9tVr;F)3HnH_R9{h_7$j`J3K
zr4;gdz0=J|<}XifhNe%4dCH~vXumGZd99W>r4h6)FkNC_3{-77pxjsIOqFrp>(u6c
zB6&*+OiV@u1Y8*I@~Zy0`v)r@>wj1jcW;S`skPC;R~w>f@Sc_VjUmU%ZvN>p_0+5-
zGk@WY_RLPA-xaru&FyA-Z@p{ShrYi}IoY|dV39o`jNJ>R)=2Yw#V2U6?wKDHu#Vp}
z$u^=wNaLU6b>O0o|1H?JI^21B95Daiog#P!OE~wxU7_0(Wd;*fJa~vwOaM?2>q*%2
zB-lTeIbbdUECNB+Fc)!fB1HU+4ZoX~&|{DVj;bH_^mXO!nlfpE-kF?X$fyvpj7Od#
z*bf?wFx2LcVt<#-`sodZoxKVmC8`<Bmrpc-`6I`9$rm$4g2(P$%LQPS0Yimlf4pZF
zDG-9QWNc4Ze+4n@AV5`CXO^kr>6PGcng8Nl*`w4?V#`kSNgckGfemXDJGKWYCc%yR
zIH})Aun~;^H!=KVc??4n@f4emVi1#S%lV4X5fm7T(_S@+Ang_Wi8nA;O7bu~X3b%X
zpW&MPoqieLb*(MI)$z?b_`90>64Ona?v=*w%5%tro0NTSicDsc_t#q*<oP}ahg`j5
zU9`P=olojh-d=m0L>eC@(0_v;%L{S+rWe>?<1R26{F4fK?#$<aDFeW!@py>4*oJ9_
zIpx{Mph@{zhTb()0~mq)$pijbwL9wTe9&My{?{1|Mu;I5rfmZ$$+$7`o2=wp=@o-R
zciYoDcwveS&~A&}$0FX>x9gp&*KSp$n^y8c4hQfnvAI5`j!sRHl8)V*yPFJJsJgM}
zK0Z&ti^%~VMma;e0QS~e(<C!h(t1mMl50}$H~=;ji4B&yAW=---k%$*Bmuq50;ugS
z_OTZjx`ErfFiP=jO$DnZM7gr15ZBZPfOCf|haYQMu}#^GuQwm+C`~13cI_!MG;4UY
zD6sSS^li!iwK5PTrC3gPsj+X%k^(dt$gJq}TZY+JY(~@Utw@bsX?RA)r9F?TYDmR4
z=BP$iHZZ?fGR!RNkAl6_FbQnufGTqjPQTXLGk*V*m*#S958@Yds3M&Q9Gg4sy{%ua
z@0l-3T#X?+Zf)yDn0ChG6jkK%k7%Epy>8XDx9F+C0V^|F|CJJ(FF;Db@Ax)bt{8(-
zq{dz0yN#<2G!0`Es>$({xmJ?j-*%v7Wj@Fq1iv2Vvo9N6zbB+2Q!mZR@I?#lviiXD
zoqk_ecW;WJTJEb|yHBmx9x!ddKG3@@$Ig*b)GF`IUURtlwNFnR!#R_t^SAHWqnjsh
z-20vwA!fW{er>nDV1pqIoV9s!u{H6-l6cN!ao?xLb*#nF=)*&=E*}xzseI&?SNUHB
z^tD}rG1A?C241~aX+wq(J`%PdW2lUKu8-hY!?crE{Pcjx14aOvi*h)qpOs*eGBAf$
zaV5_)f#&6DK8RtdTI0naihjrQ$OUKFe;335ok?c>31(q_hzT~h(MnjvSa989r(ZO|
z6o5?;+IHh|J2b}xF)pe8EQm4~f)m1uC)lYgxDDblHgG3o{h&m|Gt@2;j>^P33t{8q
zpBg|GeA+-D+6l0<e5~ip#jvNW1INXf#_7uYs9Xz_{O`Di&H+Oym`wtW8=BGMyh7ED
zgDp(I*Yev%l^N<@o+xX?4j6b?<_rS+aXM-m;0wolP{4B=R0Bk4zw=NRC2VfA$y-@u
z&akmw(L%Wx4y)}<09VTF1(r;2dgDM6&HTDTW*^90SP-6df6JbaEGQ<1RX(fj@>Qw{
z&Qj1{$j1M3_x~O38|PT;Y@0nTX4sLc6@2H8%c%Q5NTmCtT_E6Kyo3uHdyv#ffCzS(
zJQ=IvEMnYoML82dQCEzj!M*jR3{BjIb^eSP#-R1C@w_rV2INRM8%kY6ZUV9LVXadE
zXQrmM9H_m?e^K#}b-UwDM)>lf!yU#51oyV?Zo~JBbyquF*p!w6rE&lIUO<T*izzcp
zC3TrpO!E#W?!nU=5hPM)dHz-%z@i32Il!Q{k)J1-Zn_b!pOn~k?)~3u<E9~-F(y}1
zly^Ak>&C}y{K|hplpTh|aFA%ojW*udnb!fv5_=_;qeO3?Z^BKpm1p?nVf2K14*Za<
z6;YA3oh3u3n-{t|%Jx@^9TjI-<-VA4flLSCW5A8qLgRmmq!*F^Brb239v|vsae=Y9
zS-gh52B+sQ760eD=Sf9HA=>x-!W$wPY~NhGocNYZWy{I|#&3FeGYoOJ$8+@oPce05
zQ%~%R=~6=7uLIzv@OY8aF+l(LPiNsMLHu~4bNpitJ6zOL>#{Wgt=ngr>zE+9<a4lf
zvU}=M+JNJh7RU+pq4TW%#`3N3O(=>QS&WarK=h&lg6}@C5L2j@Y!u?F_@hza{sL~T
zeBIbE6ZIRWt%36UYS+Cx3ewG$3t(~U?`_`VOD|Dm!b?ub(_)3rTAIh7fNc)@y0V@3
z*441fzTf15OLbs8?$Z&+vciE|O(PWdRPsGl(VpvnDQOM|=$bs~io0^7?^R>+r{dtO
z+4r{Q-0;%f+syP!6K6WR?N7*uM$e3YGiddc_1w)lrAT@KrtF)X&DWalx#vN)_7i1G
z)hbfx1kD{R!R7s^!TWH~*9M>_b+>i~B>!cP1%_VPn#dp@&_rLhBMp+lrAr0J^WK@Y
z#)L+&2>&cO4-$UoC5E5^PC#<^jfUW0OT7&{rI1l0!3pT2TaOCv>4S0PMS-n7P7Vy>
zfZ|Lp7~69V)a$AKB}1zKQ5xKZC?-Vf%$^5F-FZ}C3)Ti`gQh`kP^j~OV<|V){~k|H
zJ;xu917UQ2iE_57fLH|^dB@EF9}+^|AXF&jye~x;l&0(~Po1dj)djP3ZF3;$Jn_cu
zCmAM31N3q~9aDzxv^2UF&8VT7IROt8B^jW|Ke#S%!FwiSjc+CvTcTcLHTK?cepox(
zm|vyTS)CfbZ{KvY{wFJ$ED6X+DB*|Nfk<4nUD*4B$WQ#?5Or+_Jyk-fI9G|HEUN_c
z2Ep!N{(xV`<5xgJkH;Dcs-DdToVC+MboVzfi2q-?JjmY-TWmKrFXG>GM>C>-eDuZO
z7>hkRmJGGBaRYQ}t(1ZM#QQH4inW5@x-1{@?{fa1f`pKtZ4<0JCC`1k-JaxWjTBuU
zRHW7^5W!|Xq+29wD(9f7ON<L9U0J9r&oA*Kl;)wx+VLE&9F~ha222&r9z0ndfE>wr
znJeDsC{s$uf8L>ZJ5@v}XkPd0^6bwU62^;}SLpY8=x0d3lKt<3hAE1>UubG?n`n7{
zn(KM6;5C4~ZM+E0Ot8$a?=Jo8$?nRIc8!i|zTZudtrKz~ZsM~Yp0ByLsVcH)Z9DvL
zsiaZDk-N|2$`Dt&d{Z(-aF;+M{N8BG3XjSk|AUkd=E1uKwh5`1)L(}y5=!&A^jUC7
z8)EVD1}#XC7Pq_PdrR4r4B211PNDMXjeJkfv=)y&t!bT@!>=Fhb)VPRoH8*8?N8jl
zSxb9Ln*R(o)GF>ndPJ_&g!pt46**fyc)&p2+3%zn34~pNJa`!visw{Q{(tMGu7mAM
z#^t^M5)(~qmC&RcE1|)3n)8`T4tlAfRa^4Chr{${4vk&xYneZ9a?AgzQTo5i?)h@c
zi8YTWP4}-g{SkDm)wtNb`O@us22JWCkqG656DkK9uT=Xha#G2?cU&muWkZDm=RUSX
zE@k(5yT~BPP|D*|CSgwwgvJkkouPQRV~DbEda@+vVs3jtC#+%pN%GE?+>{17koE66
z$|c5v`!YJdtok#Mgws>Osf+473Z}~7R{YF`f>eSD&ITZ>HjW$%DMEyxR$M=nlQV20
z1O6oEMim-tMp9M_vNftD2%LPd0expah&4y$V4(&l{*P3mtd7Xw8&7tzQJ)NcgYPa~
zdoT$n7lF`TtBYBl*07rm&%nv976bYQ(({-|dkNgG0dJKd!}~1P8Q|GpEGnM{6S4=|
zPQnr(l8FZTbrGDg@%+URhQZL_r^foYnQ<LEukX?&qyQdg4#+o44}=ir>F>V|TJrw}
z`q!ZHj8S0BMgkOgraoC62(4YX#JS4^850rm10V!RNm)mEGQF!@_isFz)j)UPKB>$;
z??&=Zd~wBRFH!|0mu~p#gXb>_$j<W-?07yjZminw`9G+_v;?78$OMVD-oC!UR?B!l
ztN-7*h17PUuybA8f2foz9U&a~(+k%>n={Q&$)$jTdV8jOv-^^gi#{1_HsJ^Yuo|(I
zDF2oBhKvig!ZnfD8f%Cbmx6{)hQ{8d0@Y6%>+eRbH<K1O^F=77<cKHF-qhWWzoit~
zch&cCNRSY&%O01pO3d~emeqRR-BiFIFlnj`wd3@mL5rF@ti#9r9w{FMS74bFPxb}C
zMe6HeXXr{%^_7os>UIcesK#EioD$->OCjfm&%dv=x!rnMMmPSD`uV}9lFI#GTf95k
z(Sy`NVQ(uwe~6o0Ee+i1QXLMCrsrqogC@h2U@f}ZyyxRU$EF8YRy=q<?K}S&t%fHx
zw5Ji6BnacBtGVCA<C~N%v;Ub*w!a(f54rQg_L~}bSlRdRbc_A5)R5cVZ7$^aD=sp;
zakX)xB}e^S(;rRn+dbi>I9Cb0Vz{rRy0)SqJ!!mSLrqt*cOXZ8BxBC<^7Ext6Q!3G
zv==1`e_Z%-4hC(pJK*c&`Q_3v!lJiX%;V&fVRW+%q=N+z*a0iby6RIjH??aEP2`2Y
zQznKX6L1v48Rd_|?dbVcX5gQR=<z#|BOYY^V8>}{1HxK#o%)p(($$=5z`qE`cR&1)
z7_m)p*7<*sOVe$Qv4d+v&T^Ia9j@BN+4&UtqH|0DMS<BOy8g$v8SagS$XA{qvj9<7
z$fOp1Jo+5pDrJ?}7a__@Mlu`j5~R%jkAZW4)6}wlQm$^;-wpgV4|?kg+_=%(Kb?@3
zNo9Z;@2+hE?An?LGMQ_KRMw!A`nvM#ATU}m*aKO4c#N*|V$__l<6x1yHv5}wg#o^D
z35Xm{o~YK2{vn3dx+|N23)G(VComb*d~f*Mg#^2pC=ag$CXWFSk8B{a7GYpI&Zsma
zP-|+qO0th1<sp{*0Y)%51&smy>9%BGp+s@*r%14GAz(giKfH~?l0%|S1Cc7-z>lCO
zlVlAydn#dhZp^lZ44QFr3<Ifj?8^#qf7AD7#+*Rr>cyxY>wy$Bn<&u$m9u=77!0-7
zOtWb)jL0ApNW~W?FMyqw3oAHQQYr5uQSs*+n>Ds0fr>nyokfCY@ZI02O_FHpZr)#(
zcH5V-kB^(2E91Do1y820h82OcsBehoXz2mmr*{yNdg&X5MI2MI($d*k0a&4PtiscO
zNvsbaPXm3N9JUx>ItO-CkNsMLYEE`m8_FWoK=KG;f{C>1vI6twU(I9el4*t&!ieDU
z&KfpbG&QcKf9gkb1EBw$8Ubj)dtdlp2C=Xs0cVC!Y0Q0V9p6fcOKiB(X7j#?>uM)q
zaYD_)ofLt_h+0W!3^+u2(ySTdHW9tjL*#)olib7rFJ()*5^<JGW!#iJ6TaVBE_*-S
z^#lly$f4XPYk!amIm)J$pfBj-75R9W8-J-&s4xG*SnO!~z9J|t*UO{Y(HhKijVhym
zgsM8aex{1Fuj9rwLppOOnua^_uHorb*Fl8{OvanFoMqVI3+FT9gKfPY7AauuN*ap=
z#!>f6kB8cobTsafzni&yewQF|8)2dBw#zHSh@xRjI%=y4&kL}8`WPk|EuY*?wD@KE
zPd1z_(SCpx+Z!Hx%%6Xi8dCq>*mTjM-Y&kUs4_^`V^nbXpGg^m&CZYNx}4H&&HrGu
zqE*}D4wty_44T2KA6_jB-^hy5%s}smXJ{q~%ZOg7c(AR&_R$#<dO7ez1lsv(63xNE
zE0oF;C7|)dP66vpa2A~>=iq&xTaTX#qsJ;KlP%b+g@D-nN0wP(Yv-KH-4b%BK2;Bt
zV-g*$COb{4sP!A!*ImN@C>Nrce6yC$N;|}4d#MioU1l@loOJ8oqZto=S(St#r=8o?
zFChG<7AkXf^#A|7tP=FV02AY~h>e~BPr2vRc#r~UV2(=?+&gd^f<WbmP`$vyfdSFZ
z@`M{e*z^jY>BN&+{{qgj9QK2P0a^h}A?ZKFK~6j<y-^pkJ08PGon)x|5@%u|#UKw6
zWPNWX;MWp6=(IrQzj6kVn~{RS$!Z&?35~64%X@B$z?^s0zch;l$y+ry2x&0Y9_ueW
z+0mQ4WWaUw|MdpgndMHFYl6Xt>Mla?XBPCwi`yhnBKrmt$<bdysR}OK-wwtSKVoR}
z{S2XO(}ORqie9iuCJL$DfI{i;ovc8JUHQ~!91LF6fKnok4gLve1jiLYmK9XB>N_<K
zbE2%^K(^|r6M!+AIK4eS_O#IhIRB8742@Dwftk7PExvZ>QA%jR{>oy))r%e?>^{@|
zRuh5;4!$v8g;y4(5?#|WR@;4dA4{s)zK|P1K;$Vo*7VPJvh@+veUYs4z12Xg31-1R
zSkS>6wVTY9;FNcEs;}T>mWB;vA(`<xC$}}8!q6JO7u6eOZQw#OerL@?lm!$e<Ds?Q
zM$r-TwlGce^e;SRP}VK&{*$ixNTxvd>;8cQH*=$Ry9dUllO^VF(X*vnGEp~$CWRpm
z2=F<6yZmkJY<x|G>HzCJyd7vT!Q-arr9I5rhlO59L0`GEQ;v_wiQ$J2TVfx;8s+$;
zZXuA@VHrMfv-IKJX)WHJl<rPey2<!Os_3>PzSz>N`#=O?z`pQ{y@8I3w%UoMuB*<K
z<rHHAHAayDQB!WCzx$(L0%6ocEBSX3M%3lB<%1U;PLDfO?2bF)|HlQ)C0T#gf=fTJ
zwVME#+^ANRp9o=mv_<kaTD~sT33se*?}f6>A3;7F%aj;!fPPpXl8;&+qyLC$e7u`$
zwb?lL@Nng!z8UM*5K)9<x<$-GU#J-_(Vpq7qqF9+C9_pKlt_SM$p-OKa_2ybT@`>r
zD}y8&2+O5o<>`rRZj5#y7U;94Su#3&%SvT)VBY0}!8`VINcFh7o?@&Lxb&iE_q0fE
z=T^&ioo*+`UGtn=vXyh$l}vUCe^7Z<*r5FhQ&itwwd<%YGM4B&QI%YA&*2KdtORJ-
zbqS^(Ssl*5nzhzi)c+`1^)E^=>H(s6Ub4Q5+b~%(mPOG%E%aCbDY-|++d&&5Ct-rh
zuz{@`0s_{I8^#5YOgs~(zzPO3QbbWtY9PpkUkACayX*tqm^fv3dM5bts==UCf3P0g
z<d)C_OnZu|$pKH>Avxv%O<OTi5en*gGnWqO|JQj_d!6S=9*6!1nX=;&dLYZB6V}Kp
zJlT=$XjLK$3i`I}2lCEU7d`%^fv@;44N4eikyF8X2U$%5myk*&RwFqFr!;Iu#W`h>
zLV@N}5mnnM2o_+9p*e0=G6-;t*?tNUUXnafVxf9aOh0Etp!Y5f!<E+g@9c33lPt*H
zev<w5Uu$h60oc*Vm~xUukP(%rIn9&>qh|X-fz?wm$wGDjT9v$C7P1|#73^*^?`+QS
zGv!*Gi#~j;waEiD#xCHjHOq!n&N_{t+X0Wk9S5Me=cvX<vI;wv$+R;M?)5GOE|%fz
zazFly&iyPxs2BGDJhmOW8+e_Qxt?IXa(4pUy)ImGCTM4MoSGfgCm4oC`npLERz;W%
zw_kp2Na7*mLaI}&p|`06=&^y3XFldXU|ighCF&p2+lO9Vd3-4)uEf4`cS0l0C8K!L
zh3&g9wz{4QCY*Z+-T*PTQ}t!E40Bjq3^U?UYQ3H&SPPVReGDLa$2aUF#;Otunw8bJ
z5Bv6U2Gm2k{PBe?j#!rW!Q_w*K!r$T1L|h7|1E}E^%xh*;wBxYE0sGqIpE*Brt5&6
zsEH7vjLD~;VV0TJ$IVu4)v|4yz4>RAT=LVNeJ_?SZ`glwA9eh^Z!CF}v_3Q^8{O6Y
zP)sJ_flH2sW_G-Kw5xOa{jPMmz@uWHE*wH@c+Ya|HB|%K2uP9IKP4~h;CXpG5TW%+
zraGBg7qK@JaX3kG=%q_Ci77PP<IQKw?A27D=C-!lSNR?rsPEV}5`tqJw<J2vcA1^A
z236;~x48!k#%$6e$<`djOc0XzyCI~a&k($b@_-c{&jI?o9=IAfe-J1;#Xvn{$yBoB
z($CNVeEN(hSSWaEf>nbFRu^)m(A36Ph}d2>3+OL>r=ILN4}@o=l#fnnPNrRlww0#o
zol$XNO>B9H-v$OeX_t<Y*sVtptop1Cu{#IhH)c3jwCf5~|Gx1uzQ4%NVN}h-cuWr6
zs=P!TjAdQJ;CQuJCqp5n#fkII*j~V<n9gSwJiUQBos)w@z5*RQ!r%=!jEz`*pgvHz
zC`4mK(6%t)CnvCHxWl4@2v+A6k70yf_H&^8hm>`A2xk=B9;djov+kA{am~-2rk&~m
zj#2h`0f@)t#gU8`-984e*Mx9*7Ri6u7~wbmnFqyvqJdt`e`SILVosnPN3jXuQeE*d
z2jJR8)3SDG+XNfb0Rg%}cO<a4xol~+VRaG>)~BY7?_hr}0?eRIJEg}W#m1QEF9P-I
z44&5*-rd(m8teW>y;=@=a4a;!D1;T1Jc5Q^exiNhdBzvzaG}V#NlrR1Ca#zOs#<tj
zgVQSDLce6m$d&yq3jr#|sYFPDkDTDZM2?k}ywBnLoQ!{v3ik7ozV(VC{f6c~wD^*(
z3a_!Mf}ON{y$0QzReK;W3l9rRvT^L&M6*@WFzG&v?!*t#Sz-o)DmG16HVOa$>)5#U
z0a=8m+Yd_6aS$EPm4&P%<v?fo=8p82CQjMfc@-a8RNikHEP%v_s~$_?hYr5K9H($d
zJ$3mR`S5-J8*yoqM{@gKt<G;bb+hruQ=dcojvTxlou_tOaj|@=sFD&A72mmc%U0c2
zpdR_L+Z}!Tsd+_1wCAUejrVum?f%jp(^hoM_4?ovA?$*gR%iCLi~UD}UzwxU`78rG
zoiQH5g{gOWOHpSEw@cXv6n`I62hTxn>6uC?IE4j;zQZ_$+bkT=tVMzyGdw3)j-!g4
zsMRg;x^aSjlx!dZiUqkY>^Q8=W=ACOvTmcY@XQsDEst-HOK^n3_Fv1oG^(Y|qnu~p
zvpYvFH0vdL`gK`u_L9qH$)M(aGZY)lKcKVS5%(vfuXg;oJ;v`l_xfr0&4kSlu1Ru_
z#c<-wbxT-#Spe8YWCq+~szvUEwN<@b;3-n3<fWwtG!|Y9In3(#OV$N;3m1fU9_tJm
zR!E;e;}e|P-UieOtkbu(KsI62$InW3YuT`3EJ6U6z9cz-6_;m%?ezrb(Ij_V*;t$2
z1|m0ppH6>a8A}FxkxB<+R&4ILIUPF%;#R@H6$xra0vP*&w^1NaNGllF*aI@o?(>{X
zu>2{X3A40xoa#B?9Y1{h*aS^3*n#}s#M~zdps%&wVJ{T?xkZup{`k{?Lx~;6pp`?*
z#5{O-n9NT)P|=fnKk3bJUG9<a(_zP(D^i((tKTLToiv#uonJH<%uL`E3*?H6AQN$#
zS@=K5iEr2mb?XQgfvzA+a&aFK@k|nzj8n`S0}Dv8%3FDz`FLi*cS`CYkOR7B^06Rd
zr@Ab#0)H@>7#Gn{Em1o`149a>A4T-%?`bfPxruV1d1Q>xe!YQ4{7~QmSf3%E)E66%
z1XH6rp&j}vslLD$rM<{iSYRODTI(x`;R#s7KuBk3A!i)bOXQNk00jcAXf0TpMoYl6
zjWY+)^0T8MTwK9~+Xp$Du)>a-J0#5X-l7J9o<6AI;3(f0BAAg_Y~r{pIiRb^2`it{
zfT&JiqWIpN|A^D+Pw3g9?ckGVwS-6LOB}6Xvf7}~wnn%C-!tfVbEY{!T7ge;b_WHs
zRju`|d6J2;2|#Hy04v{)E`ugD#o8^&qgWHWXD}8>WASwwu_-Apm?IutXpCgVZz(sT
zs=$OY!GPHG+P|E&`k-ml_^0%mw>LfaB%n#-XD{Hyzx_z%EX~OafQTTl+6fj+>bMO>
z&W+YGpK%%t7ex|AJPsj|*hU#e83giDa*+C}F>IRQqP*$8YU=glQ+#*Nqlxbh9Ww4>
z`^gYkuCT~1tH<w(ZjoN^-e1^daVKhjKmF~LRFvrKu7fQ2gO8zJLJi#unqwd-W#3q$
z(&+<^%H19YI_Ajlx^Ibd*Yo~$=6Xt8H^w`guUmF$3SxJiSvuW^NB5O@<!!|%B^Ea6
zngPeYkGf`{DQDmYfeCPW9ZeBZc3z=3FrH(C@eD-|+y=j>HAYAfa++BzSFLQUuj2ZD
zG`)K~)Bpef|J>%395Xe?atI@3OwO9~FwBMul~b}QC5NJyS944YIVBbvW)7uN6qVA!
zsdAbnl}dOKQVAV&cz^Fczsv8R{&KlwE_<AA_uK7y6Qq77hbC8xC7<Rvkfe!26ft8V
z1^V|+zfklWe`Qg4=}OFsuWpF;mEgC#?@Gly`DRL2-RX90r^RL!qYe!m-+@%Whvl6g
zdE?7e^5mgAclN!FbKLRh^X}*8w|GQ2aCTb8`TzTB!s_vBXK4c5Lldc@2XZs`Kl!<t
z2V$D0u5ab9&37H=C>VV@z~JexX}`4MR~#1fsiq+y-kA|&Qs3OXs}Rv2Z-_BF7E@<n
zunt8__Q0r$x>eS#$w#^ApI*-v(STJdHwvR(F@@D>si8NtpsZFeDti!{|Nno?h+rF{
z{2Bt?2L%RILqtt)We6p`oh3IUjTiyM7QKDOcM|wPHG1Q<J$_jubL$KGMp&ia6b1f_
zGVRisKd+9S%e6NP&v(QkVxs1b;%*P8=B4&`*#-V>6YI(M{_P;v#Q!s)|8$ga@*MH3
z+pKxoL<KN55K94cj~&ROnW-;H9bT;Nf`P^t7taW-Eco^yY)U*>L<+LLiXwU<li#>;
z0Et4TvJikdS)KUC)2#(bUt%gNJehr0p5}I;O3E^)sdJ~+e^bSFy&9-0E+4U|Rsx?u
z_uQO&>464BN$mU;4~#5jkd4STYz!PgTvTJ+3jnnmH{GV|DT=7oYL!ldEn#orXlS`m
z3MFoVElZf^glxcMu_fl%2!XSw2`Q^fg5^H*f;gYr!WHjWKLvt9n$b2Ttkf;l=&K|n
zW|*TEJ5ToO1UOwr%F@vro;v&{Cg8!8f}<{D%kASaSP2JoJceyu4NQ-ipYzVH^;<lb
zj%?6mDOilHdL!Jd#s-jRmw%U#@<R2}0klGg?y5-&COU=cKj%nkf7Kbbg~$nkdIVyF
z$7AH;>Owcs9nZ^Um-AGg7mwuSG(DU2&k~MJi}>OiU)F>aXL2h3x!75NSDaL1t~==u
z>7~~q<&3l(#(FbAi=<^G0UGv~ylYlYlN7^ZzE{n10AtAs42pwv<2=lG1Ah%cXn5T=
zn-%(l>Cy#?=oax9l2Qj+4NLimMz3m@Imz#rS%w|sQw~om=9e$mKRu@`NPp8;MrR-Q
zpK@^z5#DUs;60sJ>f052U`KjMx$fy#pEOZB2sn&T_RHmy_y=p6K7Fieh|{fjGI%vu
z;o+Sy#EhTlm!D&>CMxz(h#g#(?FFl|_R;~OBh!rMt+rVsug!ibBU4)wrafVlgxX?(
zl2V5YNjz~CGrE@JO_F`{*-aV&n-W6`K0q=z1Rhlm3OmY7>sYTsB}JU)-c|<p5{#P`
zwy+(t*pKcu8@ZnQPu6Yt!iP|rjT=uHJvpWx85YwN_g8HoO(W`)aD#QF`rE17)v7Up
zar=9<ZWmgFSVwza?!{o`NmBc3M?#Y2dT?6RL`NP~Gy9js3#BUo)D}aH?4vQhqIK(Z
z1Wi*MQt!ZvfJhkiB0j$7mWb8r)g7KCb-s~ArnRB9B#-=_W=K;|x1<een~*(>R<Fbo
z45%{~Irv<~6b%patedNkQFp4wYCa&-e&jP{TQ6GKpW|Pg*2k!B<)g|YuoB+G=i892
z49fx3YN-G`*hAxLvk~X1OXL`!Xw(=S$T=t7)^zLW&uF}DjiA9;*)qw*jXiiwPT4_N
z{=Bwv_~^0l!R*{Wh3R;y(_o4)*>!92aw+P`?}zo4M91gjeRk)?MFjHfkJbxFx2hl+
zM9&4HjI@zV6F;|eVzDM3hL^w%zfDEQsSbdv=WSu^6bsc}brX+d<d(9<h!@@9a;O3K
z2(hh+F@7N`1JxJ<a&_1y;Q?rNLCb?BkRa{`s;!=8DnL{jF9EdkaVkYzz*?8HnhY}G
zSi-iT`h{`)_rHY_hs7-*@{is)?Q8=8oPKXN8DlFtq!(~-_<`1BF_nv|hM#4%{IS75
z61RY-=6^?8G_Q1Qb@4}<RbuAoS`#&2c^ZGiMb#s0T-@xjt}Yik(6SaT{@!{JZ8WbQ
ztr4&+9`$SHj6PqlYvVdzNJT3%ut?&d5Jo9090D79lE^3Gxb`y8gaB2JVoLqk=s6PW
z;*$%I%PDiIhF@H!jIjpx4~q|gd@Y;#kN{N8Zi_>$bAF*?7|QAOTIaB1oR;)`74?MB
zZ-c^^pLBT@084hyG6<D-S;uO3t%Ou_Ds{f|m7Sz__%l=?6@2Pd?jnFwW%uH1;3+cH
zh8b!@z*}fZMK25&Qbo86Y_a2z(`5EGTz(O=#qYzO3pseRAh+q5$`8?u4_7zI?zmbb
zxFsK%Pn$YsrKRp0vZpD8g>uz?OY&2CbJ4SKaLWasTfgGYmZ(TSjuMHtx&N>=oiP*}
zS<L=2DYcOwS*=<>Ra9>0X(>76-x*LQKXM$qKn@u%Dqg)U!yTD$o2CRZqCB}$w{~yt
zpub&Dtuim{5X^v33UH=uR^kCB0@Gkl#jGB#840T;;PJ<&3RnH&Trh2{Dj<i9zDA%L
zyqnRQ(X-6bh0A%@TgI<!qxlnbGCMTrdp2Y=9r<y$g_8WY!E3RB>0YFReV;HgH$m}v
z5SjH!ef@c?1i@JkG2mq97iVC_o1H_os?ipI{-sMOeclzn{i^k^?!MXRK?TFz!z2Eg
zT9~$o%3;Gd7dm#YOZM`%7qcz&m4FZBb2@a@`Li0=8-;T6i~>qHAJ(G+tjEq`I7R)z
zA>8s%`0y<_Z%dc?`HXxGusgt3UZunpM|Cpb>Iv*O9#;gsqyGnyBh&&WFt4TldNa~J
z-=O|AC@(c<{)0iyHK-jh%)4Ua72}X3F0s6=KBU@bJm0A?p1zwglD}qs{WETzcs0R<
zeLJlyTKhmf=gLmSk+U`*9&3qjqA82udNF`|#LT%I93Uhuu<=OhoC^T*k$;4zlwkB6
zKLg)Wr4FriaJZsO%3FEe@EDazq#c_E3XLp{AqxUzxA_dcy&c}A=Oa||*nF2jaOiAI
z*S&>(tn21LO0P1Xho@{wbr-Cx_LV`71Ek;nVv!)4y6kXqYM$Uc7{EG{B0j@QWVsiS
ziU;guT(ueie#~dwfe3r{Va^{0^676AO!2Zz*lUQ$Lnm8{-70lFE&I3sj1}w*4RUGa
zu5l*xC&oVq_M1Zr=7Xq)(>y8dD~gzK+$Mo5ipW09V(DQnS|@vScn7~=(-0T}_2J=_
zOtrw4#w3$aV+%bz_(}Wllp|ox5d$|GhVA(qRCD6}YD7Gz>Fp?N9Un|wp<(3@-in55
zt&wGhY<Zn8lNi$EQ=6N5)<2AV!d`OH=VV*vf3Ec|)@P`D*5iV6m=9Y30dB7fb<PEB
zC(RqnV?I(Q{nhS7$FS&;A(b}Gf&Qo_!AVe09}js(;EEr7=jp=yMg}aOh~<DDDyCm?
zOQ@Fa7w#GYyp7QcLk&jTW^Q}*RT%9$|KF!+uHsvUzmM`kl%`7QpJ^M9VE=vlRR&MJ
z3eT^mR5m*I=I}XT8F)1_zBEy1^SO5~H+Gl)ojEH$u#%*IVqV{%d9!L&`9#H1Lq)ws
z&8H`Sp8YV?zHxY~y=;YI9CLH)-T2M^NeGU+K$7RI-h+d^^T-tvfK2P&%MGT{40@<W
zJY!>g8YITkUD(KBH$3z`78OgHtJU(*-)Be=ZIG1S3d)$Itp5?B)DesEECG%Vj*t+A
zIoFF%Sh&PkAg{^W_!i(SU@rR2^1!$Bc2RP+%<je)ukIL}h$BDu5%tBP`IoPJ^fiA?
zYtNFTW^f0N)16--dzub)KqAa-k}hWQM7LSXu{~+%C%w_kc-dXzY7k!Y_oLj%9UeF2
zCXF7b`W;#_G>nSuY&K@6+WQ6yS%Bvz)u4P%AM|$Tgs80!Dbt2xC=RX=@po7YsThhx
zNnP3kDPui<J$EP0zbbr=;-FNQ-pbvn5QnPEv>Y5}p*#U91L(H`7%HPGS#kh_%QjVs
zTV=6A35Vi;|Ad!_Hk~h!7>Juq(!<m=Ue9?YmwBH)x6>D|e2{#}Yc0DnK#H!MtT<W!
z)sUcCet*Nc<AIo8eUW<8)-QbG&1)72vitn)qhHk~>w(2C#_oe<2umg@!s3G{204mH
z*@!!kfx4@f^a1%3Ar4-4EsG_db^%G;Q<x<dfn)t`3eT1BvKAarizan`uj%|BV#EzU
z!_^b7z5FxuH2{Fu^x>^yx%SF#BDinWP+q>0HJA_PJ-5IKDAT^bmyy6n#F;cbxlop+
z_Yk;y8lX%Ik?(gqyw?N<nYKS61tk8TOErRNLt;k2iW|c|X0t$&?tABk>Vj=7{sf(+
z{*otJjM8Zec3v3}%{8x<JP|{7XsxP7tF`=!<5|Nd&4Z@DBVAZjvYV&n^Yugl_~xyb
zEV$(_q1P~hdxDBNsk|=)_MKM+tBW%!q8AbQ-Bi*?;$n0-UJVFddzlFRL)fyPM5J}<
z!V(YVR#VQ%<ZOR{zKY-M`#(f#NYiO9LGip;&DRIN8CB~wGQ-bNCBxXx=JQkwrIi7i
z!M%how=0u?n#?@C;*+q<^H=tiCc=m_iNPy-^8r=EfueLkS|Nes`4E#?9=}}saDgN<
zU1Qo>`N3pw?qXDq`tPIv7FKWdJzAxHAhfx;nLVt?7LiE3cFC=|=U&a;Cy!GF<)X%r
z?j*5j<@jWgg-dnfVqEsrr~XV;*NwjyPkuPz+cDA?J>{qKi@wL`SXQUYp_g@ng3;Pb
z1_}-4bku4v6!au8+Y%q*u|~o}<S<!kn=83tvm?m9dEU*4^TSFe$=S8n4E_w&;I}4*
zv!teqSlCra^j3x)R~+Nc5fc%_5!rINmT(wJfnA3q;{3;i+Oal}qDejKL44eIy>rlT
zY}Y{SrS5yr&j)aRuB%@eV%0ld<AWke)_P-48o?DyhTK#UD@#97q#4Y)c)U}8{D77f
z)pu=+l%G-ig}Gfmqi-)h49N1C)py^FSv>+n*#6*Lj{zfqJ;<_5xRwSp85Vp^V__#W
z>T!aJkBvB?rf7IvHa_5Mq~D<EB#V7w*uivRT2QKoa?;~tSk>x6mOq0Va^O|bDiGL}
zD`KibHQr$Ax7#Q}DL|H(tdM_PQOFpMJcR7MoS?(wMbwYc+v~-wCFAk=MByR8rBls&
zjVU|iNa{k@9P0A7D@;3<!SRDfpZ~~Ewl{DfSjKWC5plu!6c-RevP!8(VahnF4Cw-M
z9^iu4l!C^(1Rf+1`c0Orh#A}VrsM(xf*0BaXWwvPSWE|I+iSCu^|DOYsyVip_Yu^$
z+P@G?ZxG*6CX(C>*~7iURtTi^Zi9L;OerV>;53+?@f5G)qLV*Ew}3^bo|@M|Dq0|4
z-UOV;hshL->T*@x7!@tgmgusWBL@$?Af_!Za~)heG^Vi#YCs5B5u`1ubitpn712pn
zkZHsJToIs>@R#-E1V+$(6AI%M!&3UO=pREEg<Hfje9<;fAvj*q<#)p`WC5o#(ffBT
zbO&#S!_7(Oq$T3HL)o?r0yN9LxL!17h!YVAg<cV=7}A>0PKcVv=1sD|kCI!`-Qm`z
zpr90Ztefv{=<<;%JbFED(pO7K;gqx+sT1xn<CknSD)0RD-P1EkBRaa2GmTxR0B>&F
z)cKiQaep{>X;u#rncy5AX317>;)-@gfFFa*EmGX?-G{}8>$1fC#o`imGd%q!s5Gm~
z`zH4IJJ(kDAmStg&e5p@6s;$x(L)O~>KD9lo!A8m20SP73e7u1DGu-FcCIT{Ftn!2
zeWd#rOWEje$Fd#6iy2bEd^n}<EZ3?-G49v6+8L~d(%6vOB1zu-ll}%JYzAr6XMZcq
zhN0X|?&B&^;O-y^vGyjVSNCF6L5mOOc#QCjbcZi+#0VEmRm`jhG<6PUVR{dJ@Uz<4
z7rkhzhKh>ycDOSlSy*}1ld<cVFuOQbQ8SAzM~D}gBjIIfC1OXOhiiRZbd_;DmCH=)
zisT&h-Y9j~=U2~+$GYyrK6kbKwq*JRMe3iS`=8hlqJ*yTrtRKTjqOo$W~M1s5hd&y
zYhZQ2(=FfVCL?QfK*y-r*}r)8c;ahrk80{9qEt)?*=VKPAkSdw3lxVs65JtYJh`IP
z+@SxH)QK<va#*2J32QnWJ9D21ZqYea2Fp_{(<K9I-HV$hm#$=NZm+<)vXAgF?bXB{
zWu^{SVsE9^aLML?rzb7nlK+R?ZGP-#M2_N)4L9{A-pM2UWYn0}<reGa4a8NqcR?}^
z)HG4#*avXLh2n^15xCK?pk?f-vki!Np6icAl7x&!)qSFrv9uYee%$X6tlXH5Q>CH_
zJ^XTTuEcP>$q~1L+HFVUE**v52{ztiLo<C7Oi+qT+0q12HGC#M0-63=)dtnP-{vUd
zF`dV&C#eK4H}C!^Xn{(kf@tX$r~uh>>%p9)x-6@)SS|p*7VSPHK}p4m-~bE;?T;o@
z<Rnrxp72sRjgktaUh~j)Ro`aZGEYOe>fHR)lm<F!#(EowSS0W<yEws730TzGxdNB>
zltz;u-rd%)Dxw9cD^zqi+!5uz&yZ#P5t>Vb1`bI9YfzpGATZ))84|#tBfG8)Y|>`K
zcmg5YEVu&S+5~D)0FjvDQe*h)N>{sB$(C}>^%Uh6yH^CCEZ(LqZLM|rl&pUXyH#B~
zcK&-;fR{6~s24<+E|Lf3)PF+j*y`z&0E6epww)JV<dH*YR1KFlRkj$et}1K0G_tR;
zvFMYx;@$nl%eLo6XOy+KyJ;U$Km1vKXlA$1xhB&Zb$$!>ivcFMozCLBt2r|=DFZN|
zzNxR9OGn=!>u*mK@i^Rb>5@Hqf1M~b74$z;MitY+(HYF06-~oDWAXYvX?MKbwO%8E
zo`XkJVTR+|bC?0qV`@zFh&7Pzg|LLa0Lpc(6N0@9pI)peIelvc4l`p%l!B4KcrH|5
zlw9*<ptfHWf2G^AQFHnGm76c)I}vO~V0NY$=W7r>Bp}z|Q~s42H#q{c?dkk@ExA6d
z=hC<TA+sIY+bPY?GqwkZLLBu*OvBfYMv!`59_9IYj2@3Fn+<PLRp;d$spN~@FjbYd
zB0h$N(<3E_;%ii|Y97Id-9k-88t>GBsQK`*59K9X>DK<AB63aNSoqlJRs-QIMDU<M
z=afvFK1mNXMrX}aFxXkd+X=pK3Z%!puJ_=`Y%{v(k3_0V+s`NWV(_E3;$oKqG+OVF
zPgz(G2DQF?{a7p8M+!00f09#(CF9%asE6z|%FOkjDH7#X+`P*MS;O;x`~R73Q<>VB
z&5(Y^JQ)6cDzEY9-*b(bWsNRMy@*;ub+3_vE_QUeW0Tah^-v5<=F;+<VTHuaxB@F>
zWiV^?JMQNF8d&q(avM>MasqaE3|>xZo;D0GY{qHsq3fxFX3MPURPbl)KdX6T`i#;h
z8v6dG#Z`Vx4>6Vx^%o)=1HB<I-C$+Yf<<`KUn+|``qgT|^ts_kb@#^z<dcc>@vdnK
z#(MBIsHQQ_M-v5dCEtHD)_@6ojauyr2ca4>dN%fU1EBsfbCMvRB2GlbZ`Q4y;;&5}
z&OJp<gyU+hBqN$4wqc3?$y^oWB=zIgL$3YA_NQ7epy)~u#LKdkf8kPTHv<?$w!QgG
zFI?p!P~z3d0$jSRR;e=IH4Bq-Kn*!T1L7g3dNZO?g-ydPSCs-Hp;v@To}ATR!W`)!
z7`rGT-&K7txh{Tn9p)J4#ct9nUwXGqoY*aY+h=Qpvcdfs--}oMoikZ5*@3=|NJ)ER
zn-920OS}&ZDf6YN+-Q>ca1je=QzHUmYqOHC2G09?@e=mJs&w!%QLKdjAUI0vbi$`M
zy?fEEDtB6+M;gd<m`jF)rdyV10H{z0v^0qN%EXZ5nn|B?_^hoj)p{-_4*1I%`_`~4
z?KuqAu&IKo051<SA_49cJCvrZ5ln62`5WZ*8nN>;#yP*?Rx?Aj*<|qKi2j9}D#I*Q
zeyNO}+_J?AT{u-Nv${WppKsF;chkc)g|p;1YY()+05nFhS|N>lJkAa2g?#odK~U_?
z?N1M*Yv<mL4nHP|U~^w?GFKGor`X#)^<5&ipHy`Fvc%WC8yYiQ=&%-=ZhhS!3}<K`
zFqxmax=U?gDotn4YoDu{eID!lMfO%}{}WBixa`rDQ6t88mlj6V%SEj|lHBR;bd4;-
z)htsH--NE%UDBp+3_{0{>+0R^;mH3tGmvSkdtw!mEbC3v(=3)nmqE4;THH8#D1HGb
zZ<BGgFKcln<KUrj?)Cleh$s?1`gFMIA5*cP+GVF~M6J-HvxP_Wy*o5deXEf~<y4kv
z`ByMS?1fK4&;|;R{qBVsm}B&g{Cm!O^z>|wcZTYMhi+p-&JoRmB|fUpIB_rs&x&2*
z7m&run_xnn<jR|ZP<+HjH^hA286Kr;f6|tKz+JjvH$#%{#H%cDg9j!!sGh0ZfiI6$
zhnw#y%Lz-!Cu+5{eRqe7sGE?({~45fJYLa>&mV}Z%A(t&W87b}gLIiDWV^10ZEiuj
z`<S5LiHAaZ5wUGj01Yz6P23uOnfA}Fci=g{=vvM|%367|kY%*L=yvo<1lorpT<Gej
z42X7DSnp@P+Ju!TM$CyVAk^6_ynwrgITq>EDao}^gEaGBpHYx29oWGue|_5f`z!_2
zUGdx;SdGMz&!trK>U~aD<d@(Wq$WSVI(rn*1y^CD_mif;L3WgBj16p>P+hKs3u{gj
zqTnJFAjDT9UE9Rjt;}h2V$c4&=oZ~Jjm(^#SN$WNI9-a&LEmJ)9dp6)$(^5lOP4z^
zPRBsaxK!{quToQ&6TAaZ!Rr#k1TkzeU2IvMlqZy&VN1BuYAS9QeD271<)(hNwUk^c
zv+Yz-iAxb}SMxd=DS7?FVYlr&x7OB-Unhnndt!S+200;^d3Q=)gSP@n?v8K5{@0{I
z&+&OVwcfJ_UzpWKjudtv{RfBKCJyYYEQ*wPc{{I{dDV@qq$y-Qm&5ZpL8kS_cxn0T
zP&13iZYHE_<X)={gqX**+Oz&ndAIyS%kfgc2iZVmHQ;fkq?>^Q8e?N|H5s-`?MIk-
z9*DW}aW0x|FKI$*R|Yr$3a*p5G~vNVSXLJ2g2&mUw~Kv)(BmGP`>jDyM6b>vr2HH$
zw5VRrK}WeM)B<d@hTgZWpG)k&rgF*Nwos}#w*Guzs)I=A-t8TiEDhTaW<(|Fi5;5G
zdTzyUx2Q2Zoin|yezt-CMkiEOW${I|JLBreC?ZuJkZ9`0AB{Tmt<K8CL&tAVcl%i?
z(@rPRXZ_X{vuH}Lp@$)b63nz0cH;t?iEva&9pJD3&dtkW8VlPQwoAj-fcqgnggH+L
zOI6fb5;?A@nM?tj5GP0{q92RA0*>=zq5#;<y`7@=cW$Y@flJy79rwElg}G_z>-rAE
ziI94F>gh@S*}FFCPbK=lx7MFn6BSj%%=Ia<j)B0ID9l2m&aH+22GVK@V%)oFUQr81
z-N67K7@N-QFsYc8k+x6tn>J4sq}bpyWo_#2y)8syZJRbiWhqWonF1<Aehk?O(Vccn
zjNyps{;T(Si7WaWe_{g1#pJmT0T@W2ET(~|M@$UoZtCf7h}`h(1TZ28gHKue9BiGZ
zdHY4A%AC7M+z(P#<_dV~93>fe*0HK)^g7tL3!B<e>V+84D#r`1!YlYel{Q?_SjV|`
z<)u_7%m$?yPpaKy99KMjCd(%??Hin0PK%*x%#TJWw#aT@g#H7YCbHo123}1aX7?#{
zl+sy2v8RkELcAClLk_^U+qZ`jAHH1T?w48=q<+-XbC}iZZvdwNJL>oZ2U+v2BD0G!
zfbme82VC!B0c85ANQ`S~&=hNZE}WI{)9S5Udm#7bxDcf*y<xqc7|+iY-HcXD6n=P3
z?jUS<f3Wjco3Z|5^Zm#or8g&kv|_Yr(mIaVeQhdA|IRgTME(3lx8&Da7Ce6~n>Bn@
zNvCXHFGk^Y!)i6Ah;2#}F<U>$ZTjiyfA@|5`>-H>r>gIpvo1r#AH+G|ls8E)?=&`u
zP<Cn^J7#6LKZHVWq-3mS-`HU}9pY>eLj9d)#P?jw+-|cMn=Rvbf!@!U$7-Z7|8xyJ
zU%S>Ks)m~BWfMvMsM|WL92;)1YhW4Ov~IQ9d)aZV9bY2@3jUzvgn>=rFtCo#*RBSw
z6xY%W<!zihbGx-?s21OOqE;CXmHYfe`#cE7K`~1;V)G<f=^2_BcMa-$4W_&h`MAOn
zhq){Rg5Kb8&^gZkG?knk>J4B9&G-g2HCw*|p|G(kQD(OjLnb9(MY&7Slb+%s<wUL#
z3s;&6SlP9KrXBPpaVVhRgaPOKO}WlfPs2rI17DcM&t9L`E%1AOI<55FOSxj;voQVh
z{>v)&(0q!i_HFqP5tI|~_=WVLFm2C@#@)6`&v}Pj^AGebbVO7gAMK+xeN0p}KAh$2
zE{I>4+ue31p!ED1!!rTe?HkYC*Pl#%e|s@t@CU`<AE&v1pYz2;6XRb8QUiv|wF1*y
zy?(44s`GI7(T@v^J)z?FJ$KaQo)SiudH&fIzsn!{F7U6U_Vzotlu$rdBsLrGdiZwf
z;V1;1gcoWd=+t>QQG~$ZcrsX{Qz<571P{3j8E>x+BUSu(vTS9sv`#)t!JMayMFa`_
zX`s8p#1G>-4zN;5K@Ypinl;UdC^~Rsf5$Opa!2W=ld-;JO(uW@Yl-UkhsEjGrgf(2
zj(`Uri>h8rVzDzk$5|G#asdZ#FAfB`zL?uJc{DPjO=@$?XpAca1Wrpy+T5Wc62J92
zz5Ee}&LfPPx5=m9<^s{?YOs4}^l)<aG*(>(l7c8zW$zIxs@xV(I%1q1Wf)ABKot_Y
zs8?4;?thM^J#o&{W;5!Dj+8Ry-A1uK=Lq}ofAR`Ltmp)v(?JB+Qo1&LqpQJ%%{Ko`
z5uL!MR<E8d_S4Yzf(8M8=wa$`a?aWrn*MaY9@^)OW<%HEJGB@=@_NChsu?<BBm)#(
znL<GVL=PMJ`L>1aQJ=l~vR1`Oa|`}P2Z~s}MLTbc9_~V$hcj7(gYac-KJ<!;#kj&m
ztiZK@6BSJZD37*`Pyt$}fc#a9GWjuh@k#EQC72i@9y-w!OdA%#WPPv972os)v<*br
z&Q7DdrcUL~far*frw#MXo%h{;n3?<kEAbM`yIF3>O}X2Uv+U>6$H<nmK;moNrGC*2
zr9!*>&l1@(g+pXCTMQe{+Qv|hy?yT0BV7Az?*{`x)4c6kmHjEAud3=kbqR+R_c>H%
z#kMA4CG-F9_~g6G4d73oVS7{Z-qi6+v=0j|kzwwaEv)c)VOak*I8(g*Ay0VLV?d`Z
z+Gg0-OCz%3`jEy=Ux`$<>TUE{x+j5st(jaRi(RQAIEfXfVYAG+66#fThN47)vKBLW
zW__l1t$$t^<Orv7yHz<U_IeoOZyxMKzoV|D7yToOcDWMTYfh;PQJS}^ZT7v*?ZK=F
zlkW^*lBXN2tf}Z}fYVBNVN0+qc=~}HrBelg6*-G74_y8=sJL<FX}XVkN`byk*)9Kj
z4W0Vbkzr<a{4ejbFlzf1DR^Os-DjP@eZ#9e?gqXYgnakV0?$_k8!EJ)M1Sp`Sh##9
zJ*<>Gc@2vVVsc)+8coL}r~)$R%bU4(yA2<G*_D_4s_Dq}R>__5+n<J9@_Z>)QIzA}
zadA86<;RCDbNGAxajLDIDmu6EzjYLCwS-HT@@#J3E=|#&>}mA~Iq%?V%NM(Y=?&Iq
zwq+*_Z|`nj7Q0#9nC(4QDmZ1{W!ji*+BSOg^KCQzr0m1Hj-N7jZ(NGr-KW|uP*nBc
zm4$hHXg|x`x4rdpY@qjMJgj13vl);8F0?VKYvmpaJBjTKB6w)`G_k1oC-=`G){=Yi
ztd^x}u0yRZww<Acg^q1yuLcb!!XTzM4zqADsAu8&lZP+$=r$ledB0YC@%t>6E;*i4
zx;|09Jo@3Q*!8rtpV->wociN+K2yBwBnFV3$BRm~ICvX%pp~+QijpOk{ph_&*zC2R
z!+rN3TPXGzy&_vQY+5odpO9_k4yQ?fUd5R4g5s;7Gx+YsL<ac8`}Fjk5Zw5M)!H+0
zU^!FudVnmUG|dH`HBn9vSBx$Ip@YLwvc4sv8UqXaVLo+T;sD81EHk*@E&KXi87E{X
z$H=ZzKP~S%*xxq}pAGdP$uv<B(&`g#6BpRO&}((IK}Qc2O$K$6mR#N>k%xSy>F1xI
zdTZ7&OSC4VNoNR;#YA_JWN}=__G8sH<~$AF_c6MAG$+W}izsl0buQA6M6<o<oFHNg
zfBiC7taQ0gnsA{Rqa5&-qtb@h<%{2M&`>YLZuB>!qSXswzFwkhJqVhtb=;gwAb>z#
zw5fi>A{YIwq9OZw;s8hXN&-w?RZ9~GHchvwL9j{i4fZlUhytZTN>dNrap(Oh>*w9Y
zYgE8TyqjL0NZmEyDs?uedCzMA1To}H!h{K(3~Yva)94BfchOCHldQiUuC2}<K92gC
z_t~ZgV};@S-aauSeEw(<cU-YYH$JY5Jd!A6>)~c;5!!A*kEoFh31Y~)h2^%^-!Mmn
z*T=Ke+@?3`fAvCSAoVkNjJ);mhM00+uP?&D=LTgW(e*5(j>Kb!)$98z))R6UJQbY}
zdb=?)5grq{@`uq<hnHaRQ^3;$Ag?<@S=#Qk>P#IFb*8ulZ5X3wSEJW<b3P7ZlFf~>
zN9_zf^TT3!1B06P6*XIcE!d-Y){bV7$0m_+Cw!7mF%h%qRcu-)hXO<wjt~(f&Fa8w
z<)-V@C~L<l&WEZ%Lkj!!(HB>kQ?n@<R>RIWqW|-^I?!y|x-C@v!Juhp)BY%X{Nc_v
z-1hZXP85h%sGjXO@W$v>%nXfg-A_&WF33HZ`66;`cT7!T|C^XSCVOZ4J$`lheeR%h
zJeyNp><?@kkAG9++8-V08%z7)$-FN=dR+g%>`RWWRsq9z*0u8^Ew&{TUA&$3v{QW3
zlIb+3?e=TtI#C2~p&0vJli#s*gjhV@Y$V&Dr=6sSep6IoTJ=x%xle7EDw~8_x4jBk
zZ#wqV)I1nZD-P~Z(j~svRXkn00b-6ikhUU7gFA-MtG7C-e_JStpci|sLt!4+g#1f{
zX=yrOdV+`?5BRGH1rey~wwdZSu*1Q<owow@Nq;jgJ^6Eo6X9fWQB|hdcSL<TqKE_^
zF(*|o600=Rv2G-K^tE4i-ML1Zw;rg{XkUvO45_$Tg&md$uO^ggM!nk!4vBugy!k2f
zi^P>R7|Bs~tqo=BKo)1;-bgyZh+sEfEMZd%fk4a$3x2r@gc;KY2o_ugP*`+KoGrM)
zrAZ7BTHGi>1E>xi6CiYoZr&t{`|Q~%i#5)>8uhDU?{A-)n~~ciGW(;B^Yh9JrLr4`
znjJLJP(o6tt<`MrlOybj@GN|v+Ri86Io_eM*PQt9VGx@zPfCqq*bfjqEvldiOBG3+
zx<E?DaPsKD>RY4e^6xq0+9hrno5d;FfL&9hH6!4AcM;cZbC`4f<bjyxM0*ziAq2#_
zNK?e<rj_7uSA&uAGn<qC;h&CkLwA2+T!wY2dyi>70p&aDrGuo$y||tUaK)=mvt{8T
zBr#_PP^}0xNE2_8w3KZ?HudxMRH<lGYA<9~>|TMq-^$~|XqAbVaD!xU5<HX5xemfy
zJ2j?|Hn!-o=6vWswc~k{7(Tur<c3A<p4%%_W1WL^S8j!sAS8Fq$RHV^ed7kXXuZ8j
zR5N-vD}P1ivBt$2*E^b(%{FQzDmGRP|K#WqgTfidZZV*7lVmE3(<@VUwbeNplP<b}
zKk)I+yla-jo)xj0^wDpaHV0F-)=<;p-93R>PyU@PXYc529*o|7V^;-umJ+3g>@<C`
z3()o*W?ASf*Tu%?>+x;Snoy|?c2xelXfhA(+X6#5o1OW25y`Eq!})+VyKQa`7Ee1E
zQC_#O*Is^_^C?dYpgX_wbjWImaN<8c`Csu4|MOZ->*!VO?;eT^st%!K^&b1gocs9X
z>x`Lv(%|j-rlZayaa!feu6%coD9PuVx7tK<-nhv&7;VtCab}n#gP&!Y2g4vF(yU?U
zxgSK-_VaqXRdn+sD^5Ss>%c3nn}<q`YyLO3u$|kdTF<T<olonO-4Z=~Me^h6kGsBT
z@5)Ia&yvLMpnqU%S{y<){4>0_#_aqdtpk-D@6o!036Ipi-cX-5eR&*v@tw@`7dz4>
zYg%8p0RU}sWms(5WPJOMps17VR{0N+i6?Nv(br!(gt<&N_PZ^m@_SFKd^&Z}jqcCh
z=3u+9@<M)G;cv-{{3g1ZN7l8Z823Cuv(W}jvz0g&<WF0?x|vLt4;!a-%uATouAL_S
zH?CI2nmtu|urKpY#j*Gr42=*&PGYc%t5hYCKp7!4ouEL0YXKVN`od^Z>f7@$EEFhE
zz$I%qw`v(E;)<=G=7gH7F_KkwD5`9aK%4z*^?Y~OWC#a+`D#C%O6Y=eSxuXy$diwD
zD#<EoA(N7agAM!Yb^G+&O~q0l)MRiS+0XTx(V;hh%b(#*%H+-XZsr*`rLvz252bxE
zL*%<Ftc?9FihjDbe~9*Y6Ht6+iT6M)fsZ2fGZ9&}+D-^2<qcoR0tlQOo{&ruGqGEV
zAw7N|&N1g=YMON|c<8ulnD4(QpY#{Q5!bme9h|fCo&|EQd--QXWW>%!%s?;R4ve^>
zyahk4yErVqmYpI^lUK@;Y3gk;b<Bv#bJ0vj*;sm~h{MQq&YdKYFvandMsU66I*IXI
z$GVNRm)1l*GZ8>lS6!WGBaZ8EH!JqD*RF-(<aM=L!V^BMplGt)PnznqkdDeZ-lM1@
z=Y$xQeof#L9`G@^u~>fNcyd<HrxWzG%V=N%HE1a#YbMx-?<XR`s|Po?7q(jDT~t$8
zC@g;Kg<e2)Nyu6cb_Mrt0<Y$_9sXiwnyEQL39G?=P$0fkcNj?xb|pH<38)&*iL?W-
z4|tPA!ShJNfA)~LmOn;ajGCcraqR!xLA(_sVBxyhp27^>;hehyWW_s+YSe-I){A%@
zUfXZ%q+AqkjvTw{I<V9<bnqb}mHqTq-Q`)EEe!O>hNfuGTR$@<cBvSiQL2hVMUM%l
ziqpl|A*+K#2iMxNS`oI{P81d<)}DfqQOZ`ns%*>s;Tz#~%Iv1!C|ZPGo%it?d2*qo
zKLF~S%zf8&(Z)+OJeVTRPDBA-O0y~$7W^aj)y=?4c%C5Z?^KEDAk<+Um-hyz^GsBA
z4(ML-*mo;!oq6TBGf!^Jcckju<;mXhGix_AyJ>ZV?@EZ18jf}|`XId_;*f6Z*@Vha
zLsVCDSkKx=Ixpo?Y$A6(nYMoaglI+?y3w&44HekNFfB6yxn-HJsT*baoUto$K*#vi
zp|i`T_Yc^P8F}C}D|_{(UeD1R+D*%{x4Kx5ZFRk3w&z0cZHz?+!-MJMQZf8v@3tE=
zYi6gT8{*HutqeG)Ca>^xxV2yY&B5!z7w<*;dYtGkxgS@Wa`A%c$Z@56j29mw`6t)(
zd&OKWjP}tn?+uDNdQbO=;lsq{qZ18>y+X^YzMok!zzWB7e?6bcyQ!)B>U5O+NYCwf
z?b4u~DpwoL?%TPo_3jU~@MOfiGp)!RJ82oX2(o9~oX25*#5cDJZBqKJ;mr=}x>{Em
zRUeN%8^psImxfYLys)^|`n@J<0^3easOH*_4W%R4)@f*W<f37fD?sZRm}u!}-Ge?(
zUnAtwXZR#hBTWW#`^jOJ(w$*WZ%$Ktic6P&gVx1BI@wVsCfVCeS23(Hi`iEa^DXkj
z2yh%2nV;q_>JxTiCncW^%7|U$h}gK+`3xy&_Ywy*MDbWwdWUZz4c(M-iQFopOo~x%
zZcLe+$2m{>ZhK^UchBINQLAVdw-AgQVTK~IJbtzY4m}$N!VPZ!podUV5BPbcM=Jto
zLH&*{NG6HH)jDA+nVlr6Lx4%X-HJfjaJQ`r!!-7mwT%Am2+Cr|))O?nP@@j#$FbM}
z2qzs#1<sMUbq@ZrO9M4?*9wijX2VeRwu4hFnbcI5gHW$Cf&;juymn+!-5`OqzM(>D
z!n8hbtC)Eg6Nk$5{x3(EreP<flT|0dJttm+J)laL8+cQv)B?c&fG@SB(``m7cjNgA
zqtW1@h~f?%Qy_H~Sx=v39!4a^OfQtT;j7zLqb}oRtg3tyvg59*R7yon-cr?$bVSXP
z256Db9XPK2l9FbDIn=R-gqG`pwY?LDRBecocrA}Bs-+gqfj9XH8sObl=Se4lxTV*D
zY~GeAf^8jVshf-}xRJ$*CCDARFJSXnp~Z2LoI+@9-k5H61yXl+hn^$W112}VWaKvL
zzTUsDq=k}`I(++Ckl(sv;k8%w-gLO)Mv7~<1g;2DNv9yW1%{heJ|ev}LSE&bK5tQ!
z=E{5N?R=2@V(8Q1Wxa&&rkCpxOm%p#=BXS!(?b+pW7iQ#&i#Uiw$$}o@egZ@%xO?a
zNcTYF7~q5kLyxHDhWf9P`Gb3A@4(TX2e!MgL!YUh#u<6e|3v@S=aMLFK2>45g`w=-
zC1~hU3-8S{zJmWzw`IFPkt?QHzcb#uk7xXssC&rq`SYBpZ=LSWx=LRD^(xg-Wqsk7
zKk5Y~(h7V?m=0oTI7>74`A!=~<!*m_2eiRH=fY?2J}uW6rdVBg{F_UUNwFN(;h7oB
zWx0MZ$Zv@;9gZoF_8-zbxD<)~db!fBQ;t4Ty10A29sVczir3?>soLIX^U2}D^LK%J
z#FtY&Vt!?)eeicd&%5M+{`bZ;t;25r`uAOI3pyD6uWs(o#+r#;OpB@88jbxsGH-<a
z-6OE*7dM?*FZelJ-~OeY9_ej(x>;gSL%-fKac61dskmRVcAmpwZx>9u+>T(^ZA0j%
zpta2^j%NYGX7!9R6=S0y@>FuwF|T_HTVI$5uMD%x4LEMkKAgFYHEXao)yNSmmqrW$
z=z40Hjo{5!6@Rcwt<V%G5-;nresCRSrmD8FgS2*t>r`$lUy=y_s4i`Gtb&i(J&WI1
zIz*QIYU^c+F{&qf<Tch7FZ6nAnb8iEvez7(kG1Nz9W?%TKhCsO<$$7W47qJhj3=EO
z`l5kyH;Iwog=R#^RTh5HL*LD)?@k!!g-Pje2o%Vd)%%}06UsVPquCmcJ52GrmZ<bJ
zI$~37@9s?}qTQ2U6-$W6Z~|pOuUEO)V0tZ)YSb0uN<#7e#!5v+zo;dKurM9Ewo6DN
z7z|+LgRDTtOBIdIM|1ACnAzy>CI^ml0<di5t}LH}klajGk@s%wa*N2znrH*YjB$X9
zduc3jos~t^En+9GR~<05#tGHFJ=`@DR9Vdi=r3Pu^u_fOUe^L_z!&13l&0DCSk-U5
z;CEPw)18SvhjK_AM#>}Vot%J;{y~oW$`hxmLcl0oc%XnIW_Fbs1$;Z*p$3gU5Ozfl
z&bM0-FQ;t&d2e|WF8KoY5+NY!zjU<z0pL9lpU9~<v8GrGzsIvm!5`nxg^dH=4};}N
z;C#0dM0KE1y-M>W@#<}0=g)_PdI<bOgok|CAe`eYy<Q_ps6`U0e&+_kplzO%k|UHd
zwi-y_A}0p(2_zYj=YKQ@uP)=bO0PV3tw?&jlvb}eq}-)ki6}zSKcG5@X)iXl;*-vh
znebTKi;&H+)pw{wm0Cy>FkSwkiZa3wRd9Jf8XYxk(k-&SXh(a>-X4@5bUM`TA)5o;
z>E3WDTvjp-c2SJh3}sbAOL=h%m%jN72cM6nY8qScIuIf|yF)W?ZEXElnxSI(rTcBJ
z5vS|5I8IVgOr=xZ4C6e%Hzu1826VO8?ROl^P@P~mHU9FWyp$P<c+GZDzybL55rv6d
zik5tmz6&iKW0_qQ;F@deAh;r1an4D|lWnkyPgH)1Gcug%FwMbdW;g@Ldj4(Qx+i@J
z4LfA-ZPlj4VcmmrrrktEWR_1i7F_cB@8J*Y@oP&XcmLRH&lNR<RvH&gS37f^#ksYw
z8-kdAW!IV?&JHPk=0zQm5k2h04E6osXUF;Ud*jCQH74>;n<MEV8m+A+3+(=ulZC@N
zH5<mal`jxxc59I5_Q;P+pMOlQVRX;!kvZM)Q(xnPb&`Nl5OeiT$$^{eu4)u2FRq<Y
z4(P%C4>`_0eAMr;gZM3M8(zwLh!Z$~mPr(BWQypTg9%GLAJ@HqEH&WU&XnU{y`EVP
z$E`n4Ox^OyRR5M6aKZgcr4egY`VCS#mWs+_Gnj~d0<iqnX8G)2BxSz*qA#A$(p)rV
zWc~OJ-+%XBBZ5UQK_Y-Sd{n&J2q*B7nc(nP={%)cf455V8OXa8J+5jq_Bbp1Ho<H!
zedD?RA!YGL$hc6`b;L7<+ce9~d4w5S2DEBfCd>G{%#hdu<@%`tr8RtknDg8@^DA_}
zn_su~ix?c$A9G|oXjLoqG`OO{&d(~CC)of4VN#NT*^!Qsn`k7K*vP=K>{<$W45p<r
z;<Zp^1D<qQ^}^Ee<ct%c9Fj}~%4@;{>fthMJe3>J(WL@qb(L}tV&j_^Y$u6b-#&Za
zhz4uP(}_86lLAgiC~hi01y|V0C7IV?<^;v<3rkgCBWNSfdnw_@SvpQ#rtsffD?OlA
zw<|ZABVub|@|9B(whS-e=fA?skIxBmrF5-@TN0*~_!t#IMd$;#_8Ff<fPIikdJMr+
z22m=CAv;Ivkn|$%heRbwa`AUEl^%Nfzk{v)!?_zE3NpmG-uSIUM@vH8GeWnNQ<<g<
zr?jjp5YQHQx&oFgAf`dw-?5j02mo(8Jwr;vVn|5UYy{G~84+V6R^DvHAFNd1I{d)3
z19cDpd&#FPCfIxVP@-4Z@{KF%Jx>BWp{jQdphbPHUPQmQF?#*_LMvT%+naCxC2f*y
z%i@{q_skVfS7;O7Cw%q{W{Z%XiCRFUr>k7lIRGn-5L>xo=U-FrwlT0-y>81PnTxcD
zo9S|+ZUpVpKX5O4UL%<@P+3Sd-qmFq>=a0u+JOyeG7g~)DFD`06kapZ+mo|{Dd2!&
zLp3XSI9z?-L6>Ic1p)A@Hp_D|g1;(LDvBaoFCBfg_1u;gPP9w1-+KKt_E8LukNblk
z!Dc@3IjHZCH{Q-CjOTWj*S%@K@!0#67)L?)-<6<<&0$fg)s8ZRfU0vJMBC~vFZ*dH
zzqZjAH+5Z;d(X?XnyVAF`;X|l5%19%-#WFH%hEZA{R_0WHF&%aIB7<ga*;6c|5LD|
z>6plv&pMkBX80^UE%~&K=#j5CXO!*~a6VKvS<E(uSWiTGByLaNkXQX!mS)VHzQ)~M
zvN9YuT0vbI_KNX|R!;jHH&M_{Pxf8hBQX{A@yG+NYAoM<yJ6g~Ov8UtPvP48nccB(
z47Mkh3Q&X`7(Oe^IFABSbw=oY7AU6#$oNv&MXy3dDvAp5Dqo&$-qXHZSGX8BucEo1
zd7PMGy&eAmE_)BsRcc}A1yY$AE$TZTqg1a`AO@j0!<`JJmU)U8G1&$CNQc3?aH>JK
z*rty4$TIa+72~uTGhbm-!YLASLqS$}ZIN@Ihlx%~fC7<9E~LwAdwM^@KTsT<Ip!XH
zNR{=~g~$u^@Hr@K7p}MQAU8=ICw_^LTP7Oy5)}6<`p)i>m-pyEH1<9(yS)2P<-ql`
z<+e^lcMJN1SI*LA_Z90jzm0kS7?P{9=p#6<f)4U#6Znn&?@66W42W;9D>&Pi%=*T&
zE&G5<x--~cjZyf*{bB(@mZ_|j0VIZnmsMj(q59pfWv7Gpw`eNwLg@(suJfq=cCbMg
zxE4@~G>C-~dxek&Tl$Sp`~pcAstloYs2yg@RI%eVsjssnk#o_>b2PkG8(Jr;bh&2#
z*+9t<sfl;Q|Na%S@&eSC%}U_H0%a{;P;Qx?<E6M(q1&2{${Lv}DNoYFm8juOE_SA*
zs<=Fh_AnOxRiz7ag}&Xbnt67i$@vUa6wW03c{ovAD~l@fK54<r*LGRcVvIq^<6LL`
zd6x+3Qcs~e3<LVPJR-|8(ff~{-2HSNJ)}x?00m(w^LbdW5wN>qTP(O3Y|sES*whi=
z6kd|>7uUi{S#Sr?XPJ8o5p`L?c_nOnpd=yn9hODt5Iq7(msZ~Eq^scIIr0g$a_p){
zw*7k_q~6E?|0&RQv6(5+e<Au~v*-q4Fw?@4F7IaVALjHaVa0c#7@^^Y$LqZCyA{R#
z%z00o3dONz=wijuudemSOw#E6Zbkk^4NsZ3Yca(d7CYvZ=cuAoQOX&I4nsSKbaSf6
zpbRsvLGysdkIKu>pg!SYQHiENS%k<wgn)0H9C2m*Qk$0J&iqR@nu98_wY{#P$`z&m
z6qY!0j)qMu4)}M?)1FWUvyJrkG-<mB`Eh)_gZTtj?QSP`L*qLQ_Rlw0Zm7MGM>0{_
zb?a?;$nw77Ysuec0vtliPb5R8+-RwRW}*>OdP3*{&4G?T8ychz`PuRdZ%TYRw1q|2
z$(88o4eF&^9CG?T8Y;Mba?R_Xepd;qt?es*`9ECGK0f3QPvf1(=5}##mP<cQ1?g^y
z4ji<Jtml09WS*r3m~M+vM&&;fO$^+{#@cm?T8Kz4l8v1sb?F!O-?^D6+4t&vlFYWv
z#Ra2NWj}V>J=3ria>rTNTg*N0+>kw^f_N-zFTr&PY@LCa?{zgGp1u}==8_29UAjrX
zn^?Ao<-y-+e+)bV6aCitUvLQ_Eo-*fIkG3PV<TMgDBc#6;be(@IhI3fVeY!PLo&EH
zrhp=`_n%|+pQH1HC=Xzv-QGJ@X|`K6_FGM`d`W+>>s0GU5X&d~-9k-ro|DB6NB!g%
zg^oVgAn-~<I7u`4or890+7OIjL=PX2|FR`R%)DRo)V1-rJX@#T&X;@F0BCB8r7cY}
zplj!O>}W=BD)x01x27q{qf3@~n{)-7oy5T{lgnou4#jrt{4dHWJf=L-N8=?A2T1h2
zEge>Te5NIvl9EUzWD!1lZ2AtZk8Gr8hV<y?RL({JAPwM=xeznJD-7T8IYGfa=(>8F
zVlvE9c6sk5o!rAS+ScnF*26ufU@_OBG-1Krs=?#h{jMWTb3|09(dTxSN1$FGwV8=u
z6co$6IstQ_7B1!%x}YXe#ActM+FoLkBjTUAE#6vqVQ18_NxIwxSIQ*aHn<z^!Bq?W
zibe-(r^~0YD9xTuOJS5S%|4*~oa(8ZQo51?!8*B}C-Dpv@^AHD`%Gpi$EGK7-?^3B
z`;Lxs7zb&n1;?=9F2$pI@k>hFwff~v%OnPB6K=*6n-=5Jpp&HkL}ql+6$nb%YPBfK
z7-xor$p8m9OT#(PzPVS{P23;xEW-;CZj5VFc=8z*p3vn6sQf@Ep-zF2;%rlK7yKD^
zigPF-$>RP-dI&KTa5oKBSma}91N<SSqSXtzR1NzHjz?Q9Ci3CH`R7&9gQ$?myAHd(
zj!s;AUE&yBq>G)miIIqT-dd|Sfa|a`L8J+a9}E6zz|Rc$e$^v&YmfQjCKTSKB*<s|
zC9)~^cA_f_s-@(e$4T7D7Y0{nwL>H>Oh>iwimI2k|6{}W7D?>>IkNv=^G$>85vK&~
z{8Jd1ome}t%Oc4o<hZT+-3r*ko?tgv_P{!T)>3Ix6`Jy<#0FjB8K(rM7@X!F&OjLS
znx=Yv#}g|13XLPHZtfLDUdD-u+9n6uGtyv89tu6j5a!z^sJ{>3C{1R19Y%k&R^O!|
zW<zu_-{4g<Ka|>VhaGG&8(E>Xo^acTa$;z6o_gZ&aMAYHXJa_D5v|r(GHJ#4#F^Av
zXWgE3o+n5wMqRTi-I>HZbC{v(kX?^|w#oFJA(i|C?lVs#X`frymZ#iEoVugY*kjnc
zL;AtgY~w=ieZrL$oYY76?Lsx156t%)!u;+hhlu9B*fzd~ZLcX_G)KLq@8wYLzqd-9
zEqqKU6@(<OO=ip3O;;OXh6|74rR@6fF>~NunoON{FO7$ygp-mCs5{YB!>5f=iMwy)
zcIjVhh${pJa%aS_)jETTHu6UGb0uMCbSc7$KOKmcdk&yXW4TH1Ntyeo>4`1%0;cz}
zbgq@Jwn6SqAVMqE_WsT?$NgqrHpna4eAJfdEp_&J3m?;w_nEeqB#ynZUe)kMC&7L4
z1PxbCo{gsVAe|)Qz?1on&CWZ;tNl0U-4d$`?e)H}E<eS-C|NARoQvkAIJ5WXC8m=4
zCcIjAkGW$1y!Y38dv3j<f>?*SGLoyD=YbAmkJOn(lQK)6A5$F=G4xjG)j%w2YJrWR
z7wNYz^2VX6X-}u<%hk9qNA(9-V==BTnyG8wvAt7YgN^KUC%y%~+Hz;P{Bltv<>S?=
z@sk4tLa!>wP@Rdizrg*=v72RB0OMtc8>JlTj88$jG$xfKVKIwu=+(%!T>`p{=(eZ`
z6s!}B4IU7K=Am!Pe$a8;i%-g$CURij9p8V25f54w5mayWHa$5EJMkt7D{)Isx|4ZQ
zn`PHakT*UjB0hd8<&6pHwESzF_49V$DjG>?i7UQ^`9CDCZPp@GpA^bLvd^nYn+j#%
zQr)TOYrSpSPJ2>aPcQvuhQ-`+?JKXR0+!Ym*ShzoqD2HsiRO)h8pm!xt?7A|(!U8%
zlI1K~`rKbEYl_efMwhJEOL~OSNy!}-f#*5rgls~yEMWLwSEXLaB7DL>M$M5PcO?eG
z2ST|{(r;$P0#VT_zPbw>GCtv~whyRyND(3Gg-CI)=&H5S`v*;)B=&acf)iqPn5fuJ
zL|7+cTwrh9#td8}b%@_&z-`{ehA{#gmqXq51<)9Wj@w#~QpS#Gb<OQLdc{B~ilN4~
zOx+NBI<df?y0ybs4e4;DT<`t5u>1Sb_eU(tMF|@WtK=PPtt3m%D)BY5g|DfL&czo$
zOworOE<ViLv-Y5Qeat;8LFnnexm)*~Zpe*Ju&0Q%<-cgog+<fRy{2DAYZi7btqpsP
zLtkcha4Omb0^IBo?pi`@T~U)WZKJynsj%?zLw4A;%48IbfV0z`g&CRZtM`Mwd(W+v
zg2a;{h3O=1>Ob~g+Fh@;N-FDxN7Q{i{^<hYb-dm>IZmjuQ>6|VdmG%HDwUKkS=1RC
zZ-;#ct+Z3K^-s8*%JBw`*3xgr7Slw*@wq*E(_shyJ*Pt7oQ*8?MlmAP&4uMw!lyx}
zuZ7(F+CAc;7FmA(P`0kMkIT`w9ZBmqHZk$#?4yy_5@h1D3fzLx>*J*}M0`xT*KZFw
zu2|MgymxAIjBd%1&!v+>7xBg`O-FCUJmhVDmX>$tu;QEl$I`jSGu{9H|2<<vj+q+G
zS!_hclw-|#)|}>0p^_z=)8!DtrMAdnIiJ!Ji#azXgi5KLawRiMCrQX5>4dt{;rhM$
ze1HG+Pq$k))84Q5^YwT>9{0z~OJ&nPJk!5AN7Z6?G{}1gx_9rrbke?}DTZ2B>$(`a
z)BnhV-?2YVJwTftB&b`DLpmIXeoY0AXPjsp#)x&(Vj^vXU{5m*bNfkG(bfHarlVjQ
zKGjxu#bqnJf^GB3Qtx_*V#oYc&R%I+7!o7qW+*YlnSbYwT0va+8lL&%QEci`vsRSt
ziRYxZjkA>EAFpfon{>P68Xq3{K7O}tyITLAptnjt_RO8|WMmkY->i}Gig~vAgX5bF
zm!1Q8Uw6zn^=}dXCoH(Q{FNIHG?#5xM1Svh(fvE2E`9&We&6^(JAdtiRFC9yplnMT
zs|jR%aNmsCS49qrSlpeVVpF!)P~EVZO?2@uV>>aY)~25|VoUvbvV$n8A_Q2=6{6>S
z<cSJHVYhaJm-0YTGaZHzkSvB~y8eR{4sHv-YL3t@uDaZ8Zz}Iz;U%EyH3YBOP(dLx
zbO$WpA!irz(O)=b`9NxFBSI1=RZ0J=u|o3(PD>-86o5vA6X91;p%Vp?dw`dq>N4*y
zK4hkLQ<y5DDUAh11$g2tCmL-Vi(|e;O?MGwPQq=*pTteD<OWiNrk{b`4F#{)t2s){
zxW<i6ZBUFF`;^*DQSh$CD@P&BSUy#jT;lbJ=C5!3cU1Lopbl}Ts<6EFPQSMr;1j}p
zOW?)5BhG-YIK(Ud)bxH%V5(;`)?N|JD$~pq`7YLYV#Q8)RrZ7(^CyxJR^w;BFrdqp
z3$@0*NECC<T6p}*(a|?f5Drn0#v$%JMTfvmXF9(Y3sk#JoyCJ;fg1OjP~D}s#h`%i
z%fpp^3m2lC;s$Kr+t3D3RnP{n-AxOHJjEh)mYgNI5}Q_BLGGif77ZHy2xr|OJ&U@6
z(*_U&J^?^g@q`|0Pcf$JlCT1%12!l=l@cW26IFX#U{P}v0~-;ux{5H?>e6qSHf*4s
z9ibVG=m6gR9<J1H^>C`$GO14YdQkk8HQF9-_VZITVr~eb=+2h%95-@Y?zwLjD;nMb
z(Ja(H^dcALiu$_6<^HE|;|~<LYo5OBjA4yYZ<adJzvs<?(@DDJ?@_+{NM}xcemZ!c
zU7*_gTpsxc2JV?uFW27Np+UYBHg-{i*sLiFy0xg@x?WCVn+pWukzVFdmTGg4Yqeny
zaV=MIlefIE&_Qb1_k)FlYtG$jEIyxqVoVOl&nla1&tP_C)X>QNdiDd=dc{t4c3OZw
z)lNM|VPx={bHj$8&|YuCoS1L2E;P!cS148D_=iaI$96Dy>vg25((dXa8f_F&azm9c
z_qWza>_+b31s&`Xr~LB6MfyqFrXAPSxiN=oD;Sug&+U%S6}$4CZy3CDQj_w9Wm5Lr
zUI|RvRcrQ7jA!Xw;HBy~T5Emc{7v`Z`SUt1saqaPMeOIfYUVxJkI9R(SF)51^?qrs
z8@h9RPw7baj=oNP!hvmF%;RT}J>uD9aDha~sttlc68jY;BRR609V{1fkI*#L&WxG;
z%makc-I(C=B;J5q)Sie9-y7u;7*QLkz~)2j7EZIyuX*!}*f!-GF`N!H?$FHT)@R+y
zS5*eh8Ai+%5@YifU<SPc7oN-2x(|Cs{jM$wZ1Mb(Ws8#u%5yTz>)ZU?LGyc!%9Z6@
zWjP9E|Fb8!&C{X#LSn*RGHnu-b*)OWzdV++4g01;v$xbK-R<0)w&ATCC`}3pRd$bp
z^)CDOWurnY&DFe$rLn-83Jwh{syF%0eaYqJL+<$vsnrVT++{#+7|1Ot*7RXDxx~Er
zOy!j@OU!M|@zNg3N*v<;!tVLzTPXr<Qc`o{H%6ATMU!f6!>x{!dAn(b>vcNWh5oy+
z55zI5^r00%Jk+Z&E2JRdz*Wmh0`QEa0bKg&jU~W$#^+(tCGAk9h*nv`lv-jt5M^}M
zxE!FGjq0(oh^X|1tALzr2~auqGGA*z0rq(y)~T<39<Mb~M>f{)Oh(wq-Hsz$%(LW1
z5yjx`T;=U`2dk(unjC7$bre|2`d_<d$O+!2sc(}%&`7?!yH-rove}c|E0eH7RNZy{
zo?NsUHSkqZS1`ntmU(Vzz73P?DZcYxAKmFVwlv!AI3(v$>(AEar9JI|8#`8&9?UuR
z!&AIM-W}IFEmZg<`ZRgGpkF3=Jnb_{ZZF0w-`mf##+E@`Krtm(NOq59A*@%wG0>2Z
zK`dc~LPZc2wybF5-KTGBQ3vnSu^I0>Tq1Ka|Hb)XHM+tZ1&die_MCb~K_7mtw_m>m
zU)c_2z?NL+PA9Mw3Ho)oJDzGlgbF=BqLr_{2E9tBg4HespmOB^??youpIJ|U$_EJ-
zRwxdc+!2Ecbq0(O$u8M_nTu==nx3f*RN{!ot8XKg<^A)jo8MR=xiLYV9E&)9e{QlO
zp6?iUqu{ynBS6WzS|g8NC|Moyay|2G0;w$i1(GN99UYjN4LqZL(s}Z<^y2gDCwc5b
z@ju5-q<eHhF^3XkAvI@q=vL+Yt(~%cTLw?<S@jh?^PF7VQYjmE@pA8Y8&M7;%6?d<
zEsQn@X?Db~`6?hlRvcTW2^hx&f%dKj4yzj3^d-L0NO`V7gYBGkwlR~hol&{&w|E?c
zl`=A3Ybp!-CVh>-DFUDwb%wcm)o~==5C%va?e^=NM5g0fG&jA159Yc1yF5P^B#lT?
z<~}M7Un<EnXo(rGjW^0;MB(#r7e;$b9W+nyZ4JIypa%6-=9C<W7h*SlZ@&?or;;@Q
zYhhe>rA-yt7n`q1;bV1vuP59I9k%Ui3>l=Ed3M3iL`%f_UwT~?o$v)o*6iqVTMDl4
zF^=;uFLKave)IH<oKyJKY~zWYM!iuMdMvZ;Jce#);q|^b(XkC6^$uaE$5~PwDK50j
z+Ur(@Kn(7DsvYWvgtQ5!4D%!aF^8%#5@U#ck5i~LLb+hw+C^>2XN&oc_5*o4eud4M
zEOdnEI35OB?0$c?Nx11bSF~-@Q8hRzp_g$Fi;WJXuAnl~=yfG1_BCPq{j`p)n-m@@
znA<PAvEwZc@mG9!4!MsG8hwb+orJxx&Rb8IS~1>_gbjv`r&7KOBFaa(O^Imf)Miq#
zj^$(ji+(}(S;RDzYNj-f=|$Y+?LXPm{*Si$B^N&2>L0^<1oH4nb8eBCg*jxin@{4u
zCcq4ONUBV+-)n<WSjB@cs^2kS$0ZgxT|npBrkZ_Ipi}-dQq29h*V(AM{sXp2%>goA
zOZ_|5vPVo{#s5;j-CHK6eXgs-q9;J}+wnhUwnAe6H?Yiu%IjudFcsMNC88vf_O!zo
zq?7Dz8+m{)N%*y=ovM%tMtW$bB#499d4z9!QhlwrMA|#&3?1EFEBlNit~V_;+o5$T
z2K*a$lNkWPamCh3`na1ICT)Rln0TMXIaa*;)#Fng)WaS?J;|ikb2SCpLEQtgHVZ7^
z=)u3Pb)M>J5JQ?P6>^+5zqsZXs#6i<vq)D7L0>guc^iHPQqS;zkU8wyNuE~!uchH%
z0ID-Xq3x};*bAi6w>Q4pGGYdmo$^34wCCNkVj67w3?`;^q4SjJ?K!=I&5}v?8q<Rk
z5KMKQY401-kcl-gb`5GXC2WG&hpjn?Z5h#Wa@fZ%fj<PWySDTxqyrK0xhiba`ob5k
zLoA+63lhK&L}J$a%~Wi@avewub-3EN0l*yM(b$Z`KRxj#xLTLP;A3~&9Ow8kB^<~o
zj<}7O0r7P%M6#T(r9xa6-#2X+U6HBJ`L17jQ~V;@sW;HKe;T?W`L2}ahCSLk^~&Ck
zRbiyNE5|c7(Sjb;hE;rtpJ$={1HXTk(*MC{zx#Xgc#(nA!MV{>;TJydcwhEoYtDLI
z=jLZ`j~*K3d~A1oUVq;@a>VMQgzZQW<H_SF3)N@>#1)NU3Mo$Qjb(g=LE)3lsFFSu
zP`tt7S0tI;q{pS7eRoTAee58x4uIRp1Q5<pCQnkT)}0F;u5BICnYvlN?QK)uK-}at
zS@9>Eod-a+n`i%In<{|&sn#6r*Q06ymU-c1ZtiYA+;SyMEPp=1HXO4y8n&1*X!8x7
z#8W%j?^T|`3Al#p8`un({shUM8cJc9mrcF2eDZtFi7)#Q=>C09`j>~olFyA=H{o9j
z+YE|YWhW2g;#llpXmY&s1Od9Z_;3h-4LlaeNQ^*C5G@ay)!8P1iwxJuH3rd)xN-hA
ztir4IbQ*Y}$^WT{Ju!fD#-Xy6!}=H_`;51D6<=!>4Cb9_Evz=Ge$pUpGuq;ddxOQE
z^zh%s60***@kBuvPn&nSHf3CT^&?Tn@Vkh@6?PHq2wE%|82ZYlWd!T?vsh?3?+H((
zWDJ#fIOxGb^he(hM>L;0QjQHDT5QHkLv@<9K+gu{iiAIov8@q1bDNn-4x;^Tk+Ih`
zk7=d}YgsYJkjk22MAB7{%>YIGy?r85#LUSL^K~zQ9XFBKSBX^^AN8mG7*}-Om<$ZK
z9ojYf?5W0%NF`H}q~=V;fdQGs6|&q^#S^G|S;%B$C2IT%WBjqVDvDH(0_c<_54@&(
zu=*f0*TRth8<^F#bK(k#QiFus>VI^7sICIkJWJ{otn<WJReUfg!}bw3ZH$8kIVo7b
z3cLp{*2{_d4|1m~Tx8jJqLYO%vwjsI^?uWsD34jD5X(mnbso)Ymu%}w@o`I|I6A_}
zJ+@9_dV%tv8h!G>O&*;h+6i<<vGz3e$tMZRBih+BnsOaG#dGUeWk-vAB8TDfF`T4Z
z^7<cZG!tJEH#cTD{~9m+y>2<iTHl~53~WE;C$5p~8*C*MbDR7z-kG3RfYyu~M0MG+
znh0AXEkI+Kq9$d$)Z|nIAXDReQSYY#sJdjn1k`)YZLE?#;Cf7%g>lgE-KQZesKk$G
zqB>OwLYF#OXj?amTQU8|!!ak^qSeT})ss~Bfgrg%v~hr8v1l!Tdz>ay;6q%y^*RMa
zDlW&3`<rk!6${Jhn)8kU!cH_J4$K<3hwHUNpa{<?YNl3Bx^Su>FY|%kLDeHWEj;&H
zy-eDT8@i|Fv=4{UDc7q}*xO&Y;rC*i8Mg2EsruhnW*QjjM!l3@(+sg^PoHzHCg0N9
zOO|pm3*A}V)HgS+0_}C^PdK~BUy*7^?s~Y^V%--T_4Bh^lyL?Djy8YIco50c8FNzJ
zh1tEcvHtkAT^Ly`0eUYI#IE%nCYShDA9&|~>>P)D>?~+I#41Q{t2NOI%Wf3YD*E63
z;g9XNS}QC>J3Wja_nY4c(J6k;>gy8$#Wz~4RYqfb3B1ofr2ApJc*Bm7p52W`3a#@s
zain(Q6BTnE9lZ-21XUEQ^zxc*cv<{2@`hjgRED|aLs0u<Ip1Nbupv|e%;xEF`3^CK
zm{D@o%6_rEJ4YfjZ_pJH^g2z=bwUr20s_5Wur}qe4@00<2a7nb-VR0f!{{*kue_Hd
zkTo5yQ>#x#sLZnM9@7=L(gnbtjjl32iBeWCfo(I!+U4`%jzd<bE1r~4W|d{S&Ijdp
zi(MV48CTB9^$OFSUGl3TkCbIUv1}Cc#Q(tJ2FXskgv-1a4AUai_H6*6c-f|nG<Lk;
zKyEdl*jsWXj~%x1wG^3X<~(tH>|&c5ar9!W4BV&g?@!6><-EZiYudbK(=9omp)*Jz
z0ijVPphAx;Ud@E>-?6jud9QBC8V~9ESL}6eh>dCorc{%1{h0Ry87gcym{~o=(@<}S
zo3El~=n9|!;=4dwza=HG^TkwbUJqk1+8MV{@vXN1zynl`#fTWHYN9xe609KS&ZvLp
z36CL8{}70+ukqIU;R=G}XgbWc!4?cI1N23pv_IwRSprS{FHq6~9ZU8E%|dx3Lf1N1
z;&7>Uj@$wpp<)Qib|8Q$Bc{Eq#V&L}=z94;kY#aZbaN>RPs$&{!A}$X92g%ctG)Fp
zY$;5aT2dn_Uawr9_H4d+2(Y<^h4TI_3p*Gxg;Ilo-VYQ9;)wr9X?ZbE5<Z&Q0JRil
zi;yM7YQ)|>EJ{ZlO#J;R3UwzU>Ssau5{Kxnv-I{_u}&+ujPjSgJRo=!Y3tQ0&}@8F
z<-eR(O)LeqyON;TJ0ClXU~PJ<_<iiHhC2>=OOZKI9|1lWJ%6=6E7b|EG^{dW(+{Tf
z*I2moy=OB_xQ6!mVemP!P#a!v8_--tIdf^P_De~n2LT%0fgB8z#@f*M(C#2;&YOG&
zm&Ks0!PS<oaOD7ngW<L8c66#-R+DX4@tFv-{p>cXhGS0sl<_9r<B6p)8PCC@z9en@
zbECb|I3Yd$l-#dJaSM%_pl<5v-JpQPtNxVmxbp5U9q+8hH)y3pj%J233wckr@A46C
z`#V&b{WN%DMn&q!6*E#et-*zQxnxeYv0+!R&I=YM_sy3(QSaW|^<Q~YZetKv!DMI+
zT}?Drc&c=D&r-}spU$oJIXali(y3?D0Z#cJ<lmC(`W<4Y3De<lFS~j;1$n%_^KFA7
zw^Xy)u~pRmP#Tv7s=2+0XVNH=oXcTEki66QBM-Rx@(8(#FhWU&){<}ik91Ew2jBls
ze)2MZbl89!>K!}8)4Re-AD7|{gPtLID%e0|Z-Dfzt>ei?Y<5?j85NcZ9B`t`Us}pp
z_kgyj^z8DaQ2ml27<9nG$3$UV-e0`0A6`r;7$nGtf!-)jki_bT;$E=iAg{cWTqNi*
zw+BK@Oy%0`KT_`vj(yVd2sm~h5+(%@!cvb3@QR==xt<Tl;K?Q@xzJUbeSHtyvB}oz
zN!NZ&pX)P+&)O-+BMhZftLGfpovMD6u`SA0J`WY6&YUUQt{}H|=KzZBa1dIdXi~9^
z{_UCy?4w-qs8XeTGLg8<sVDrc%G+mpwHR-r;lH5!W6=DmY#uo+dBck>>SrtFN6A_b
zN1gH>1c36qHBU`IN&x&?kO2Yi{Gx&_kx_G<N89I;{njtNkJBHHPBn8-Nzy~`Hit69
zTrs*M7JCWIvfW^wJ17B>_)8wINN~b9aE;S^4WEgBvNBn=pLD_Ph(rN{f<+yMro$Fp
zCeGSW7h1tfwlaB*8xO)JOO_xSR+t9JTpS3xw?Od3Lu#L09N3`e)&<;EPAs6ti4Q7Y
zfzA-<$12G=1z079v`dt}k3+qdgZ`A`Ow5@IN%&<h&2^Tqa&G}8`IqqgpZM&w5X99_
zOy_V8gVSKWPO)O18Jq)KZ!DO-+%Kkwgeu#+A1BpmjwVMR1wV0e0aDGtN#`!A&srzH
z)=vK^L8e*jYttMRy=HTwo@pCLe7L_}@4UqlQGWkV1LDtk?Sgcrqc`VkSA7Q`A?T2_
zN@iIfZHi|$xl_)O2_WxG&ju^$Kqxt$D}d%JAT<|6L8gonJ5hrIP_cQWy3cILp7>5K
zwJ{c)=d9Ms3xtcK0P?aMCSq@z?SWvu3pUiPMuJelgPNkezj)@8En{LV`(?y=Pa={!
z&*DsMZ0{Jb58v|UZhB?a<EBIOo2A!uofcev{rmXMZ-PY5SDD)u2B`zM?x@STV(0^F
z4|$Uh_z=`-qq<ui0G-ljZGsC5I}6h&;<~~uV2!oyb?xr73TcQUPi7n;2NA#^G`h+Y
zQ(UG15Y5=NusOF%2JG&empu03z#L3-aMu5@X8FsP&6nqH;5Ic5Z{7Z_$n|k}{rCsR
ziN)S8NJ9BT<GfE$ONwt+$)0LnZEG|4^@aDofp2HOd#p8g%BZ=CS@yVZwi_f*jYUn&
zy5|@6ASN`(n=u!*k3F<q&Yy7E@ROxZS}7}}a6Y`>J1f-qhWygl*y?GN=Kk1F$u7<v
z!NkZO?AU7#(e=rNVPiYujZ7W>S2o5Jf@I{lqLRpnL_b4rS=Y;I-tHI2an52F{UOkw
ziE<%B=g1bfz%2jcNSDEMsAsDi&BtEF?N9T|$p(L3JHA_alP6U(Ik$-nrg<TH{Nq7Y
zQIgwihfktj0D4dRg@qS4Ec;gawJbs0bf#dHm`*6K+b7RQf*1HKUyTlP`HO54_mU%y
z{rVH<j%?*3QyI>Ct%5A%7#97I3Mk_Nmh;FD3~LbCR2oqc7r?KX;D#kkP}0fypx`b~
zG<~L!<SA($rA}-N(Oz?be7jmFS_`imKVS>8j<+6f-T46F>~`=O62(@`taYY?xbfMu
z+J!31t{kEYusM!<bxst34v4~7IoxG4Lr?mQ^QwxSZQDrfs!aD$0cW|k^`TFWbS2@i
zDa^bz+hP`iQ6Frrih0}i9!mCB((NAO!T+XUR38<o1V>n1(`LyWbrlv!IYHY|kQYg;
z3X{7HLA67{T4@d0(M3VKSNZ0D3+XDFf1e<%M*x*-?nGl@BkZVur93L#*bRI<j~>UL
z@m!a^{2hm*4V+geuIC>W`8CkJ^6#I=BnwR7ASSW^N&z#~QgJ<Cy)RXXM5Pz9A)<2<
z@V#I{3!QDP1b7qry{}+{R+LoJ0Mhz@xl8B)hmeIDeY6J>PxeHDAmt}Bj=kj!i;ete
zh-V2%5~>!2MBr@JJ`d`(<pA-%&RQTgtHyZM1!~_<l`s>*Z7@i5ZW)-+E8BZK$l^HL
zHg}Bob~ZZMht;-3am<o`<25XYRdL&h^8`?)K`14@KF>LZIGB@I4+kfKI7s*+ZGmrg
zSH%K5dT1$!L$uhvK;4le$AFncB#vxj&CL{YFJ>gompojXVJZyU!Py?x7_hW6uaEeI
zCtk=$$=FP&HOLLtLdFXgze~8cK;fyDpRGaPmbgdvg0s4Tgwx%*i~r-RUFC2b#m#-w
z2~DT&Uv7|3x9CGu_Mv-Q$4)j~L@!u^-+}-2@sB%ozpQ8#IsXTR+UyVE(XE~|WmZvt
zSiY}1z1oz6$yxCAo@|SgeV6v@a-CCh)#?On_0*wo{Mw4~3j0N1(&qo5QCwHQ{+3Qc
z`mEO>RLfdQ^IsR<?6`Dk&rzQo;%<*#AtL>SX(~=SM-Kmym{I!9WUm#&3$%Y9wmx3I
zAL!4rnqM(qot<Qeam2yxB@1Lm?&Cq00TeB|)DrT;#7)*L(2X-Ctc+#SrgM7Eb_3Sq
zQ%Sn$#89>CpyniIoFfoM{p`dSf9YK14MbenjkO#(9!5J65r)V)wCD4csBiER8;h?k
zt@^Ija_P!P@1b6Zo&7Rz#x%H_;b6Wdi0U=u@bp_dPZT*vKkJWGy@lw}v9-7ywJEq^
zsXx&w<@oRxH5uA>AD?FpThhGS`h)4ljNznyfU)Ud{NoLGGC1m!-d`_WeZUJm(HDSw
zQcPN&9ITzpYLr#DYkug5kl=JM<;i}4UT{7WL!wDFge;yafg#Fc5?=c+uYi#PoF(lz
z$!0TECuh+)$79~o?4qoph6*Z9xw1N{gspuPl!@mq8kBDQ#p)VG8QrTGXQHRQ<@cVA
zip!m62XC858|90s#*vcLa77P#z>VcFFpU)cF>xA$t`m!6<C%ak|IR*j5!OTh2+9;d
zLlWQuCVMm2z~72ljki*eIo=`=rCquGwZ>wGmA&K_WMR=5QvFl^DW_X;wDzbD{d|yR
z*pv1{ec&7!BM5@1pep%~Su01h^M7@890vuEyp^;MQ+a!!nxS#l3n&%In3v(#cg)?8
zRvH&jr^0KCL#|n8y-R+&ucjr_lpo(%$%%Y5wi>z}wYnPL^z=qmXiQo`ILY<!P`1my
z>s&~+ZSyz14%6*alN1)^U^*p0Kc-Imm}oNkL84%$HoSa3aG}bX>ttC&jjcg6U01dd
z!peJVESHIAa$<-Q+aVMc=F7gO)4;#DZ4^(WjhGO;R=tf#*?x82Bq>i3zy$_^8qdK9
z=^;_Pc^3MOR@GBsh3`F`i$%^!o7$AB`~&A=M(FqlpKE`{5D{)WF~AlKP9|?y{Z)*#
zM*|naNRu6uL_l*qF*o+F5FI)Js-a+_pz(mWYQX3kT15`m2qR*{qUXk8B+Yz(Newb$
zH>lSR1>f&W^bxVG4curtI@;r3;WsWal)d+SErf~Ztq+30zf}Ed#c`G#iaW!SmhU8c
z_HN9Vmrp-i9$%v3)Ytkb_2x2zDqQLOOWi8{PT+)*sl4soJpfb~^&Z><A#*E6_#b3o
zuU2V^s4$yU701LBy6dYgibdvDtkn$1QRzDi_5EY-Kj?te9YPC*B>|P-4!QGr+l%}*
zPz&=l;ngTPy3@8FvW)y>SueSVYF{CZus2?r)*tzvAK|_Isj<wdK#N>kTe{de_YW_5
z-}AfbcN`BE!%e2Zh`Fx}@A==U&p05X*!i=)BvOO9eCS5;r*9r@p5I!&QS+X^=<qjs
z^j7cd?txNquk;&UoAcjzvCFJYnpZD9`FjI1HR|;->drx@?^3_F54bG&w5)|&4<#HT
z7AyXYs!3eL9qD#8Z#}m&BPw!Fr18dq%%7{p3*NuN^)`mz!mL{Ij(JUlyWb6r=gmHw
zQohQ|DHHOI?^%Vs`|uxB=`DN3wt3e_57CatdYuucPs*?_s&+i9+r55oR9NuQX6B6Z
zn|pKfXSP)(2Z`^pqHc~G;v#M^OhGPj+tY^m!VElUn)bH`d7)T&?c&QkczP{kDjvdh
z_0EYl;spVVZR0iqLi}H<*FPMVns3|rQqjbmqiWeG7dYGLzi6||kA5tx4e8w#_u>j=
ztUI0QfSGW*G{e{!CM|@Wh)7mdxVdd(WbSkj<4q}A@gCB$Mdbo7_I>PNF=b0%TApE>
zP$S_YvZYtn$?aU;+sW~^!UJU>r=9|Yl(hXj29pu=T1pSdi6aT`s)18S$`u4O`iyiu
zwe=d*3&!A_mS0P%+E?nDZC$IiZ5i??nQ+xJ4ig2+YhU|U;fLZ|huK{!B4Z<5v#A0*
z`>Y-n&EQdYWH$j;E8FRygd+o$^`bgvM60k$c=>_@ZbYl1e<mr;WYNqwn?LsWo17V7
z<>vjuW&I@t8HmX3;@M7OTRB+hC*wX-j^P5g;3DO23%|{j!w40?qASt6%yAb%nO7;o
zQ$En$^Rw=U+nLESg?}(R?bBeiM=FgtyVKGCvPSjlFp6vn?{NB=>Z^_?mD5LN0+~Ob
zv@&{I5+arC{43n>xpn4Rh33v{7f5HD^elMreL>!do?;)?TH^Y){wVLwe0%uLe0$Bd
zq2YGE(#1sG%JA?9h5b7FbhA{RZi{<yGwkmj^Yx!~qPz~=yplv=RrsCgG;cV1O>?2+
zz>Nnu+hy*;mKooiNbh2w=jURbOb=*8l^}GwxegQFc~bQmGIaD{V5NbnNeoH+qbu<Y
z*i!Of9&G1ambb8c+m=pyyNcDN{ncf~iXj@H@7I#2eazhCZy#p&EIThCowaYMS9%8m
zQFEpJd7e^-?aeCwv9{1M-QpMR<bGY33Qn?yzqkb^C!t=(Q*6}^u)s5y+-z}7p+}e*
zK)<f8xN;CGYc=n%Fg$>LT75+UF1K-tv_}hAIHbKOk0VNvrBnv}?YG9P*CcrK;8_bC
z>o0V4NgQwpdq1|Nqhta@YmH<66+m)(YaxK&7nTHh!ZXPUL=2k<Eli|Q96Eze@D?g@
zMK&4(wTInP#uc%9I%Mitr4<Y2TR~XYg0}1q!XmO3#{^;v6><zBST-O2Nu8U$`iA7d
z+9-A(^i286bsj9~ULtF&S^IDxyR}+Hfmjx6z{$p$djbW0+>uO0i{ub=*e-VY!4ch%
zF_W&3uGed}d^P=Oj*fg8a@*;d%;xnAwy*jJ5-UQlO?xW__Ia46cnEjKBcF#yeNms2
zL(aMxX1A9<f3f%<BzLSpmy>(Hx{|D~zq)j&?&6zQhGeg=;XYG)bAKPNZ!JE*`*QOw
z_)ogNNW`9vTf`1tU7{2MY?8$qsMUt6^wtql^6c{)>&iK1=Caa{Kh38&m5zSS!D;0C
zKf3r}f7t`8`0s}QK{io{w?8+=6GH5HmusH&-yLTw{i^5vv-Qf?XCWK2|3UnN^{#n;
z->6DN)%>UjDMu%cl{a<2li@{`$ln237xqdw>9}9&VPs$K%I|T$XE%soNii*ZQIF(K
z0uBwR#9S9e)YM^NXNd$MR{$r^bM}LRj%{#}w<V~&*Az`WH#4@HXS#&ToY@y`_i#Y6
z<+Oj-v5NZ^vz)VCWBC<IoX+|qwx#zU^_UH<H^ZCCPPVN49MYvqFI*+hh7|I;f`MsF
zS96q7V2j%i;EnK!o3Etq&KyCn4PAK`;gd*U)gNyqpi~q6q1cJDm+%}Zs;~BeH|}*L
zAea<^Ra=(7GuP3>(^iJwvJPKsBsjT-waxR!6uxo4V9^5&#}?^gm&%-$M&c(z4gRPm
zG$$+Xrbp>|H8D%XF%@*3YSGv>kud|szV4$}NXlVnH=>-Uo@~om=OOoAadf<wNgf%n
zHKf0%8TBoosseo#w0IVEVw)L%%m8pc927)MfT|jnv>ZtJY<f`wijKH40>}v%hG97z
z4=~q%Ngi*lh#3)6x2OkfKGaMM=y$WL2qiF&CGhaQ-|Egs<WCfm?uP2qCOWhiC1_sT
zdOPJ3<@pf-s`H_a`x@L0C<4$s9|mPN?c$kkZ8)$B0-=<>{=v9+`>TbM*M7{!1jawQ
zGS{Zw@x*@De#`LT5B-PhT9!Hoh8IqMs+9c7Kh|=Eh765sWS{(?^Lm?@toQA}nwXdD
z`<yl_y$?B-iuQ#CuJhIMZwGvnIZKwq3%YM}(R*j@w1hb{0NBf!1((%r0Q`UkFq_w5
zvSOS6ceps2Xl%Q_>#Ys@uEb|gzybdEFCE(e_ASi2WD}idLb}_Q%^|ju^%uqv2TP>=
zlyiRYBT>qo-}Cvj$wO|@b3>29idtQj9}a3?6s?conHDPFY5=HmbLM(&zZgcI&~1B-
z4h=AN6_3`MD5)${j;5^;A>>`cC43t3feLQ`EC*=Z5kL>L0c_{1e}y6RNCViF5lC|#
zWT0iM3(^SORZ5U#HB%;q3hpk{ufVpwXDt!rgsrJ!aYSj$J?&gTCHwexAGmaWtJTkI
z@6*yz2N>uReYPiUX6igHHq>boV+Lb9hs{0`l_ys#f0v>cu7){mZCF{Wk0TC(yqzXy
zZXRS2#|Mur0La4cWZ^W>ZxZ3%{n&eOOP<b2$a>Y82)>b9b+mJU`Wv-g^9ab}cjy?;
z5cD7)lAM0{@9H6fWWFTTRI=Z3WO3cpTtj*gZ~C%Z{Oc8GLCBdmIu+$L`m$P7B5KyK
z7#l!q8N0o^#vog#{K(!PZiBy8f18iFK-u=Wg_?^^CK4y^>9*Sp3rj9lo_%>x`Sjb%
zdvgQ#4}JS;*;piMuti4cl~kI#*(J&>aDOgt2uy(2WZQwy1;`Fq*JbE(`A+|Ue8-HR
zD}qQKo>s9r^<8nR(JrxFdsEPBS6==H$um0p+HVYBKYeO%>0P|jK(T*$=N{lDu=^qL
z>C)bX+@Z&(Rp-sErDhU#ZyYO|{hjPE{JX+Cyp<Pr#M#r$#Qh@Qm{$C&Z2E~|^S<VU
zzJlxdo)(QenvU7Z;dXZdIX5_&y$KN4RAZB+87g2Erli(*MGikvkoDNfWR)mmo7BkX
zk={zIvWWlacvtS$lahwNX!0-|#TD(}F;L3Px18mRk0Xq~!oymjA3|e@>%dxeM{<H}
zNssaCEf*h*-h7xz<$Kw(^X7;4wATo)5!5)*HUjv)1@oGaZwyu{W2T|Z#=8+nv36i)
z5i38-BS^BztVNnrV$&C}AY~W%{<%8=dn<ar`jn5rG`Dl>{hKe%zOWlqM9+i{{jM+C
z@v`j`7*JuP(ojwfsn%a>I0pO~JY^}6!F3|eunMn@I8IVkE;jYwF?zNX_+z%%-+_kv
zV4C){;>m^Fq3ZH<JwQNHx=ZM3fL2_DObD0SpuiL6fG=uytu2GuV=KE86ODzn%fu6)
z+wnCVarFzefMhZqi~9%2w4Vpb-h(s>eEGZAWE?4YVZDwv|CJk3e(@;chS9NA6pb@n
zEqHN}z#km(_lzdCxT?kM95`7Ive2?7+F`l5n%bwrwg3h}o`F<|;^pIu5`-4niaYfu
zLQAu4{2X&)!*3c44~oBfH4tAr+hgh#Jn<pm_@jvE6!lWVi=69?GQFt=NxmQa%gPK?
zHenX{yAy3mP1~=%s*G{%$0O*|o|-?g%OvqyWD6aw5#~TdzWu*Bg$wy)JwR(S9R`pN
zX;j{o9FqZ7F@G|xIcCLcUyN>rnbN<4EjqgDw>jZ^BfMcqLIqf~hHmONW=JOEk(`s7
zpRh2-r2GI&4!QS8m=x$AJ$2?H*zS#!24o@qFI<6%_zB>WfcCti_d~l;>aviYJCOj?
zbRJdO9_|E*UE<K}6KX+F!CG({R6*D$djh}S2DpMeHtV<6zw3`sd2hvmEdb*7*V=b<
zg;1Lqk0bZXnV8?H@9w9%iD?vA_Wo>Akx!->7Ef^EL??Jqb%nd7Z5Nj-Te<d$4of*+
z@<&qNh$p;XAL81lk>ypm&b{)*6IB_5$!T^;a1Q0#3K53n>k08dSC!DRfw7_8vjqrZ
zGw0OJ!p`%Ck$dWj_d6GyH?-Q9YH#Utl-0aXH`W-9K(=1t(<0ZM0&)#rB+70%S`e!u
zhj+bopj~5m%Ln}Pjo+!^2S)fg2GpEyMlZuBn(SZQ+urs~H|ofv(`Tbz9JY(TxH5Ne
z>l3u<%ZueRrCU7&?Ha|8<vT2_8s^x3+wl%&sbl|gq62@;f*5&+E?ii~Yd1oxmt=sw
z+1WFBjI+<y&>K&5_tD5QZ?ny};ez+mkFGUQGICyy$}Bz2ITjv+s{eP!+GFWJ=P4Bt
z|K6@d*XXL%$MFlRsP>Qnck7Dvua`Ds!u(I%;Qv}zIr7LCM+iCbD%6T>Z5&N+y&Y-f
zb<w=7?^&6iJ*Xl^Y^n$@=$pM7#ux+w$C^He0`_uOxa%|JV=$s(`>Cp2=dsxeV4~Ih
zf`t?Z0|oF{7w~P6wA%-^Mg9pMVL=vg0aKWZ;s2l)qr>!Dy(Sk+KD&RMxJJNOQ2|N$
z<@Ip)!4KnbJKRj;=0SG6sXcfwunHyB<(urdj;6jz*gyFYY~%5)#_fJZV8E!v>alWY
zoD6rAKmsfIgTC=J;8<*Rh0MDMh6^<>+*|Q~JcoWFxDaG#{E3vXzeI=a@X4Y1$YF7c
zCJir_NFf1PeT62r{juMq-VW~<Jn4OW?9g{^#W%i8#-F9{9}p&2Vj7@(C7@)1)@w3g
z(TL?AF6prxF@c<8zJsa?Od`nK%`;JQlVfXKqBC~z4~py!=t6qUhpf(DMfF;(*RYCk
zGVZT%F8&CcT@b{%TQ@0wK3s7%#%uct2C`8YaMk>j))_-~z@=8>fiSNTf}U1gD}Xh&
zHaz6PWu4bx{^SP}QAG6vHgawZ*TarJ3E3l~dNZj8pwQ@0DCi{rtwDxK2@3%EKgSb<
z1bb^xy?7RhA&Fh*ilq_z-{I<kXB_y}fc#37fb)U-$s?>lJ&r?zo}ug2pL1zRg9Hsw
zHk4UUCdOI;vQOF_2_r2zAZL1m${fvw4QhU^)~4P$PaLx<pCA(os3!_oon@ko%^DD~
ze`3)!Y+#k%iQv%D6>B`;NQ$!Zv?Bm){5a_z7Xq#ul7w0eh$-%Q1zbgh0W~-A&?zP?
zZ6dG7O>C<go_G(WI!+sy0aSAL;0+*&GN6y#n!TlX<PmZ-{3!u?+vs&m9hY^|_GClV
z9l97|&L@dW%d8J`0LyC(lJMppR|2Wex6tq+X2M@Ks5K6|e&gEDa=Em+y{gbVnJ{aw
zwn{_Of$Lb*eQ1e-ctZqzC<X{18G@uf&r2W>clI45*PGuh-?!5rp-Vjs__RY0UfTFw
zs?VCwI@@iZ+OO9_Ep}c9c7F?q1Y5IFoyhcb9(-gFN%PhHM@F+Nki@><bfXk~+Ef_I
z`k@@Av@Q3WUf#C@ZJH$N{9f+|r-H8rEKkHL6rI0O{F`I!vfw{Z?;P{w!FqOQr1tHj
zWxfR(Gl(l<xHgHp89@7N7O}|vfm?iN%wEnZhWXL@=n)AfHxVs_4JJn&y0h53P7rF{
zdT-$@Z}hN!TH39oG@}jXg-`t0?oGO|CyVBM<yIg9De^P=56U1cU+KymqW$!`RR7Z;
z{bF9je0;bRAtfQ<&KbH3=>tOLPG!gF_ZLpnKXU)l`n@#!rFJ{5NogVX*}|t^W1VIJ
zu9(mNLD!OgHtUh5^l;kB(aDOQmxfFHMWIhS#6wF_LrUV_Bq&7yR{@|p4wTIr!IuEC
zqpj^yH`fjb@DlokVAq1uNEVJQeEEosoTQfZ8$IS@)Yq{QPc6Bh$v9>WpYcqij7cb}
z1hPeRz4Wnje>4k8vV@5`j!(>*YIdFY<Fyou+I)P9nq8^>T6=Q|pyZC70rC%K)0QHu
z6D>D~esX<@K*(toRd<+3JCMZn*{ptvYEz4*kJNu0H%n)+on&JAi9f-wxAO9-61G;@
zLI?eZN0(ej^knw?P^!|BKiY)%KYJ&Z3q_maQath44S|-E1$fz|{p;zDV}{x;g0JWY
z9w%@*@qck5YXBR(*AEB8Ymi^;)MrBpQniCRzynYjL!!fyxTy@8J(mnf<;`=^PTGYN
z45jZA@8VYFI<9K?zu~lBNxb(euv6I`Z;#Mg^6c)%E8@Wezg5)@-_4ayG~E%?hM(TQ
zwJ}gdKd#a&z%!stb#$M~+iXIqFizU)7{`o5RJ$nv`{MQal7u=~`z#?H?1L0yT*4p^
zJ}Ui>O?uBOU|06k1b#@+t_mpKZ}Qe;<<%MoS(KxdCAm2uRC*?61Oq@odqk@C;J+x^
zLIs$XYLWGPyqqnr2aNatV85F8TLi^`8j4jg;oLWwTBF%i-YJnGrY_zCXs}a07*9OD
zUkgZNGX<YTU?Ih}2Wyh(6yWt`T2lcA#ELe>5|%^17;*oCYEj_ym?)9M!lJE%^Zxxv
zM3`89oGMKP1f?9B4Hd8lC5%&L@p6b#cX^a#L4}d?;HNs%UA8xULYcr`sZ!H5%SP<H
zhWgMAx||?f%ok7wwj{jtnlFaZ2#a0<3G5|Kw1Ixt!{vizj^Cs$K6;rye9y?rZ9yb+
zvK~EO_%6U_c=w9|ma5y!9ybaf2*mO#`r_1%n<RWE$3ZjFI%fD&YS@8^jq&nV<#K}@
z1pURy>=~^nh5VR+8yt}m>$>mWe8QD*m9^s)`@<gaNn`h-y9#|oB}2{?L8ncJleGT&
z8gTXBC*>W7+{e$Am%TxLmM^_~zdh$l2(_rPXWzB(8|U|GmfX|-7J~kj>^qlTCEjKE
zK`xyB@W7jj?OTr@%p<rHAFi7lhi$v1DzK09BHcT+I-M-7jWgA-kkdKr5jVlhaqrvQ
z@6r?CYD790x#-W+S1!EP`)+Q<vLDxrsgORl_{b=r-{1LsY53VPyPt}G5B@aFd~!SK
z^ofHJPa<n<5SY$|7nK8r^{EFRP6j@V(b;E{t5bPpL9g1=wHHaa&^Z2Y*i-7kw3fAe
zOxhIxWRb-|pXsfKd43)<yZ3#IS|IqQP?^ML%RZfn0zO2@vx=`Xt&9Oxx=6m`!x0no
z!@(G@NHFIu6LkoZf#7QjB}VvMYu)b$N8h`gs}@<>TedxF&bROHZ!HO7<g3(B?E-08
z#?r#uIs{v_K(!kE#sg7)h8-+TCZa%pOMyUpcN5sf(_7}7T}SqrI~}xI14ltwmYd0E
z63w4m@QE&#Bc9qdbMux?I-+S?p*PK6r=KDl7G$&?B8;mRl7J)gOW46`t(avmpby&G
zKVUm4^w^6s9PhW8XC*q^j#u`Y+}=7m?iLnjQhh;d<j-4D8UW*9C+Eh9(zc80BIv_+
zE={a_Q^gP9&*)us$5bf&3DrzeX(&n0yMd~n_;D1(l@rx;4<-j;ZAXwGr3HJ81N=|T
z6LqT`2iOX^e)(Tdvrzl)FCTW{zXyDO8lZFx9vENBOAsVxM=ALcXD%yisWw0KafM@i
zoEnq|=F36xGg@_J+UqTD<Ef$HmSDGds$cd$zEh6DW$mUib5tq>N_;1P(=rht%9}i*
z))cIO$v$0iL<@k1&1zBU0N^2(cN>&M{^584ILn_u;H>(u8{_D1{T_XXk09AFJ5j8$
z+Y$w7gi%15k%qq)s7_IrZ*jB0#B(LYSR-PfN<9x8fQfU$1+_&OZzH&PPY)KIN(YBr
z%GkGje+QX09%#S@_ww(a@Mu|mcfck_sX_n>_Mj03^q8kTYTzz~N#t>f$tP-;CzvRX
z*%GmTML^BJ&qSr-To(ZGHyRgfMsZ-R)q)pA2ma;WvnFVdku<jm5ZOk~?h}A~qVDe#
zI57WPVAPYn2i;4cCWZ?VU8fcIksV`@tSZj9&^=#cS}bhak<)6kN|vJa&Hx`&E5^to
z;1*~$v5YAH;vD94ZS=@n;zPeaNUBkS`|dOuDvcD$IaxS27qUpS^P$`k%LDLU6tg+*
zFHDo2r=faEl2)*{jvAhF?8b{HD{Zu-AZzo;gC`rM#Nx53F51OTIKqitoZCtb_~H7`
zd~A#FMgil1a4G%9*VX@^D|RCeY84X|OW!;cet(GFYruO?DGGbkA=i--l`!=(d+U$S
zpH3SHZ%jmdIO$3}{bBF$Y5y_8V{%DFynnS{(VNH63){xAUJv7E8-AT${+P1Z<T5X>
zJ&R&hH?JpOG}MaoJmeAimrB_?pIjt$h2<*yF4q5U{*K5_m-@!;qXRC(cdNVuk2lYx
z9`W^a)yMapK4ZrE4|>{Ue%3bXk)Ec;M<sW*%5h!CoabK~ckNpH50XAKnsM<PJ!$@C
zfSZoTxM+)*i+V>z?U?F)wF900>}utF$vbDd4^dYi$6jt5!Cn@Xl{qJWIaM;76XO?$
z*mD@i1IlmY=o?bXIu`+aA20KIfIiLq+Q1Gaz*NW#&pqwXWCDQ5+s{|Xqhz=FJ1!g-
zCMWRQi*?M?I|+`9-y2KcIoqfT#{)FSiDBDG23fSTHolVnScURx(A8jq!P;UKm7aiW
z05E1%f@bCR+rFW|QdyOp!*M`Ak*R-6fL(%(d^K!tjo>zA%z4oKx6E9@DK|2KEdIx&
zMR5LQa4Uv7qPWQ?Bo&8pyni`{6q3;3CZW9xBibsdo6};Uoo-foDS`B$-drsrwx<sq
z2mk-kF?QPyVMpLofI6Du4YR8qB|}~ll@_4wv14Z~UUHISH^tmqYdz@+Aqk)JzK=i}
z4zBr+nM%vf)}FB8b*_A}lTz;wU2*l&=LcU&U3O6$G#@=gzc6n1A=f&|ML8)jxYb#x
zy5c+eMB({ihb6DweDPeGu<6yLq$WdZ0Eyb3c(jw2bL&EFDNo<h+@)Fdz2$*RsXzm@
zW#B;8-@;pUD;hI4o<@|m60T!ZA+Ogqzfp>nSe0$Y0NfcOeyw(e1pn@7!4)U`-%0Sl
z<3KD8WQYa$-nrm3kRjbKim&;%?KJ>)UafP&6hB?(LTmjyL0Nb7khwk@cqjH$P=TE~
zor9cU%Jn!I)4>T8%*xa&R=Wi<2aEIpaLehT1Af2ah<RoMR~%%r*8%bvZNUR<13Nn|
zyrFjV4F{oz9l$C8*C(9g4}mUMR^CV9`T7awR-HmNzET7oF9Br+((;EzHLGBFd{Izf
z?J1t5vczF5QJ~OkwYJW5v<!869R<Au0Rfk;ne%|{{^l{B3~`;W2Mj$ZHP?RLx_od8
zrA$pq+4LDR*;`Sw!wE=+S%oej-}pHvm~(QCOWOw?hKaiKKXK4F)<%B8ZFOL(j=60C
z`_ZmEe!ig3>~juAw7K=Xuv{^fZT9V~#etVA(-^^4)rcvx@<Sy*d<0=HN-w<BG{vqj
z6AuscSFwZs-qD@iv*9;qRS{d~b#K-ZQCM9Ra&YIPanr4Xpv2|7TJ$^qzurb$A}il1
z{mzVC{tu$Op7$5z2Xtkgd+7(A#suaMw^KjF`Bqo9PTq}d3OMvx;y(UE_XGL+HFuw0
zAv^&<;x4DZJ}m4w4f7LPT850!3ZDF(x+A^rl9bcZJSq0Hap};Ju>GAM4!@f}=<&m?
zcSE}Dhnh)UP4i@3z_o|Xe*PhEmtis#C$|wV;j@#rN}G>Z{0E&-on#yBxd&zF4t->2
ze27HXHXdH~Lu+FEWWV0Xu6S6yC@ZCe`)U%^ksWI7&03ync7C4}3|K!O0uCea6BSZi
z`}O6ioMR_IUxYm(r5FqX_m<prD#5egq2eiL$3>Wa9%>k#6NsVeOuunyk{bf~2%%!w
zW10+if&D6DHbK&IBvjYmxPQ0B^po_0<cW~YnM2F|RsnPCbs3RbL&}=X&Jod#Tgo%3
zo|^qyJeVs534GXMTFF<p%h37T_k$<e>u^}zd~I3Qh;uUF`Yds-;_s_MfC+R<f9ht6
zlxcO-wqm1P`Vp3&SF?9TiK-O1R%>zT+3&UjZ^M>4(00CCSv-yH;1r6e{wr*gquT@s
z#mRvCK|){|oFA&V6u{3*;@cNpfB{R#pHW&G3rq0=_TZ2E1mm#e*%n4`G6#yDoAb^9
zrMb+tTAlplU{GO`19Ak=tmb&84v0?WmAnUCgfW#OIY8)>N2P#$XArH}^&`gtQ?J}s
zd#LmXwz7hq-CW}z_2^Y+w`Sf+{(<8%xA$8)KX!b1)_(q+^qi5`T~zPc$Ke+zuG$RX
zbwa$aPBd{uoP39>(uWpn4a?kK+XirXpC|$S@<0LuOIKzlhFUy8hlXmG!E1EWcfrXG
zfc4WL=Vn+7WE`Avk~l!zWc=Spg8n!8NEm|<(!C98-XHgW!ZO(!RI~lfAqN>Fz5Ats
zgM4Gv!#hs`vd#&u5P$>VM9mr(6gd5CK*Ko+U|$P=TB{CB1jx(toESC&UaPOu1MMno
z1pyMF#Tt(b{{#r30K;ywXYDu>?sCLm923k0834FCz*nI#aya78MRbomBJTnK8x=dJ
z0ying^CXa0)bZ}##LQz07ImPeXlX#!>?$}hENjllnxJ50D`3@>>z*&E`lkniKCjh&
zL-knzmm(5gug$a$YLK_HduAo$w_U~*IBcCuK=Zev1_$X)VKBLpm2+~J-tFrC*_d;-
zCCernmoZ%l<AGdvWnIZGXAfY=)92$JgCN6si-o;sE#z0(P;$zA=}E4X8Qa<El;mTu
z+v+bev1mFP74h0+c-Xc<oXi<{6pOgvJ7oFT(0{zo?~8)2U+!J9td6y6g6dGQXReM`
z$v0(@iz&&nZr8^~wSg)|?A|2R;%9l1SFB%4*_mz0NX)4^%$Upwc{u%}^zlFCub=90
za{tyi5%8^u>mN*>=bz7v*zsw+&`?^C@-hYB5=gykGnd~*qN;Nu9r{cLtER+KVOOm4
z6*C?S4s3Y0Zmyyk%z9VOYN~NJe7%&cL$1f!T`&!pmJ$841uN$3Ets!Vnqq3?B-Q{N
zT!eQt@Cs=mm)%~z2KGesGDSy^iGH}@O<I7!`+bn`s$C}BrO?YJB$*0BDxaE|TLI+=
zg_mUqkY4+B%GJ}QN2<?BVP`r{G$Jpyt2$UoZumT?^*5uPwPj?*Zx`E2C}EQA87z$o
z@r&+Zv>`4fdt`qRtGaFV|K8~tO@FwIT}xgaT7});n!X_=>2PB)w|g5;-RdhLFsfgO
z2=AMe?ob)xQ&g}7o=~Ag6#*KUk1hlMk$;Vj@^-~*=9}_xDE0@dCkbkFBSn)sp!hdG
zn71o^XIQ0~Ga-03P%WZ(`5gOWNBrNSo(oaI7qt5$*%nWc*!0enF)fWSSg)Ux1Dc<M
zz;$_$(6@W71_gELckI*wLKorArA1iE7!m33UXcjilES=8I2rqxr(YT)YoD8hQFte|
zv>iL%!9TvVbill^HR%YUorTZ1Trgkoi0`-hUa;bw8CCh@p@&v*62Gz8@Pa70)MJmR
ze3l~^vOb<OCxAWeKyzu#pLC2yGGyC;U@CFz^MG>#_FZslNt=~Rcrdg?%$P~z*l#Y$
z@r1b&0Ym5i27D@vuD&*4oVo<=;0v|Zd%oAfie53CV>po4xbE17KrCoMyFx-tU0$VD
ze~}IYk|ab)W}8@WB~<D`XBP4fb{)jSXDiprX9<u{E}5zbnikxmY6vpUqkqp|%%j6h
z;`V(6Xrdw}oCxqr0MMz2iv=?hxR3`~ER!X{K5z_h=b}WL0igtW$Fdfw0$71_stIrM
zaL~&0$Cjwf96)c;oSwbhu*CdW|D|?M36wBL{l#G39uv)j{<NLose@L@@|e#QmjmOo
zOO)^&>7iq8F$K}5$=NM(Bou6$J0n*oZlYXf$HEfHF|!3vl-AbQ9q)r%-s71%rP<8#
zEdr{*Wx<K{qCy9L^$k&4A-)!ZgD~Z6hu%+h;LMv3<?^PSARX7r)8w3=lP2Pz`|->!
z8}`{NIuV7SneEz3w}{`zHY@LW(#DwC*8I28qkzG0=lSrS+?RV5UOa|2N78;8^@rOv
zI-3n&)9LZ7tkLVq&GW%-_*!07Rw$xh8*d#AJEeUfn{?-ROYE|5_sw84W{R{Q(y&yV
z>iTemQpQ|gw$ojrpL<cM?YlxkN~o8tbbTalVp;(e+3p4F;YT7&tQh>q|CJmyxn3ux
zj2uTz1jd4Ewx%~Z&x|;*^0P7FhbOF44X3)}D^HwC#20$5;5)R+YA&k16UEV#cNv+A
z61O!UaVp<bAhk>tJiejR+b9w^=%N6b;2l;Wqd)-UF0E?K0E_ur3o#a`Fi1v^!Sf8R
zbnF;YuHszz73QY6Nj0moJ;QmR`Ehh63U&dx@g;SccWv6&B{rnBS@AL6Z`Q`IPUeD}
zH7ea`J6vJra#6t;3?&46#q9q5pyveHG@2tcQmhO!1FaKrnkNsYI>kN(O8c;jgTGy&
zmx6Fw(H{*QO3bnUdk)1)Ji(RqM15T4B|4g`F;G^%>Eph2aF=sGQmzW^S_K_BeHN<7
zt7OPpm;=i8fPZP}1pv`vQUrLjQCkK;cm(LBTL27|S7XR7h=1!~5@y2GpJS9udadbm
zdGu76ove{>R2@m1ftiLmvAYM}?ar7<h^t6YKhqXi6l-|mN)DLE)W7ov1LG*5@t!Nu
z4+qAoQ-l&3j|t%O_modQ*$-iDW5`KB_sZPXy+Vo+vScj?IVN=2o)V2@BC164{}=@T
zRDeLWLx3Ow`VHm0IfqMny+({vfk;1S>jY{5{$EY!9@NyChVk!^kQfMHYit9An<z<i
zi(FDoxOtF@;T{q#vs6(Dl|WktL=j`L2^XblhykLN43|`r0IqCp3s%&)h=xm7AO=K5
ziY`!ThE_x_S}%29r@Q|$!%QY~@_p}l-skx}$RF~U$pC6R*iHvJ{Y7U@$$bM5{#L17
zQjs8XCUWK;hq|k~x~UbpE_$#p00=$mQHY5DM~80D@$C4_YOBmmdUL262&On?23%Gq
z`h@&gYFgbh9dU9*byVikRSUnZ8u|hU)|Zz(>$2<-%QajAf5Xh{L9np<!r_o@NP#jF
zfsYt!!R{$SCt-EbV`5J&7y}sf6B>cf;pf^Nx#1El-xi0v|1xMxuHu{b68KDcbXOA3
zNVt44nX>vI2lFnWEgqZKk<0PlD-)k|MhqVfC2(qzp^dpaH{|$Fr>xY6GT$s0@j~;{
z+9wjs@m1DFi}zrw|Ew6NTv(u$!q~5~mefyky7?iZa^h|aG3cXpQ!1DC$gg^x_tbPQ
z_{nJF*>z!EM~ZbPzBNw7?D&&&J#r}~$Z^};=$WM{LD0tnQwzLdJmxh*yFyM6C&S_m
zNUdj}VA^1;$uoO^Yd`VsOpMG0&NU~$1Qeij%L(Go;@24YqcRru^A=QKr@qd<q}=wI
zCxj!hU3)%cYPaum;T%n1Tt5)LE`{H*-E_R($xj%1@8Fp1mcC_8m7>tdZxeXyb$f8y
z^)OLd&1&atGDQ7=M;6x~$Jvs#w#@{;qMa2`lyvliTWJse{MADGl>D@0X6Kqp?itmT
zse&)v-uv8P_-gfgeD*R?4DMv)bM-<WZnIVoxnRN^zwJZyu+wa$q;x1@Rf+3akMyzC
zSJG8Z+dTFCl8@ciiBH9+SHWiIuZNDEk0Z`+3KBQ-U$hQTXN4z?K2e87ax>cGI^F8)
zup9fX);g^qdhN1Sfa`O8&K59*p&bpev845Y#FC=O;d%4>Ez}<}{hcYV(aeXp^kYix
z0N?tzyTy2l_PfgtH^gZ;w)%m?F9(X^eh*$L{M=xhO|fqY+ROMpCp_tWx~)GOzc6TE
zx`Kx+21Yc-TpZj`HXA&nu@RRo1le?J7!~ro&svI_>D}H3jMXr56c~EL3LwR!8rH~y
z|F@Hd#Z{LAe~;D)YP=QRK?OzZ$Z3enPsUXx1bEah<e)j6v@c{nQHX<id4CgOfGwNt
zw4i=FYNj^o4-2LWj_lbX2w>iRjc)OK&glvHx5sW)*%RC^DxZTsSF`v~4ubYEtF|cg
z&Ea7J+tr)AJ|egJvyiV!0;c*ktndoNsD~NZ@wR~m^}j&$bXq^6#!k!Ljkuz@R8EIa
zwDCtN1<ZNt>5bp?6HwkPw0^>uy7LYPnew#-P)Um-itDNeY0K`9<|jJj7M|@)_TVzX
zyZ5-{oPkb5rUd^Yh#+42&iM6H-Lmm`#&80AQBT~PB;C6dv4E^SVGm`L#g`Ino7>LT
zdcd_NNgXV9^(2-DKzQB`wYX%9J>8!eejb%ETDvLz#NtD`&5x3;r}uhlI*%Vrv-i;v
zbex)=Sdw3`RqZmMCa2Zef&$Gjh-D7JeEVn%>nWUQU_r)QELON2iLcmt2b4uS)%R4F
zj`fi=&O9^n=Na^NN^xa0cyW5~29JJI_`^MI)T)eiqEx=PqvFd>P&i_T8V+3a%Dx#T
zplU>AL0f2T5SgV573pGgP@<;K8#?WRD$_QvZ94%M%*Wg_-)pNsz!S4K*uYH%N(bYE
zaSc|Nlh2&o=^^h@G(U>FbF1`L?~I6)*zT?Ok#Nwcc%k3t^rhhh*Z@h(THXQp@BP+Q
zl`Eto2`l1+DEZip^WIeLC%K0iSA`UEs}JD32KIVnIR#P{Ppf4@0k$<3#w5E}?7L*6
zJppCi^(qU%RBfgUy3HzX6TlW2`W3NI^_w53d>hseNnH^9Ku*LL0=wzaYDgtuJv4%r
z1o)m$ydwGO-MQC1{*gGEQxgs#VfT0DivzVC(`7raKVG-ajUICTY~~F?Ei*cz{g`jH
zW$vRbe|L=;E2nCmNn9fzLK2<;gkafl@{!9hE;;#0kVkmdR^TZ#Y4^cu!;`IlVF|3u
zugan<fTbyFMDtdq84-oz$2xHmO;bwy|58VN%>>oQ;08#>WuRI+vpbdZ@#)b02OtVa
za*`W>ZW;5E{o)wt(@TcI^OEIy)j~#r(Jph|_#}%O)z=sXW#22pGOb#7U+Al;@rvCH
zPwt9pv(M^<ml8C<B^4J!5$K=$_v<#pJE0p&8((N_h$Yb%2iVeF7Pet}CPu|t?!47c
zdDsnwFocte-QG0)?m2u}(P6>XG@}~kKrRP)4afl*0!dolEWXTzGutJXMVM@fY}txC
z`{4+IkG{z*M0JRU8tXM3p}8Rf-{w1k)>_nzY_UlO<n7z;0~9|;h4MzU#Ljoak#_gc
zeVC5)NIX)aMJtnjG^{nCO+WQDl^>dNKq5MTZA0>Kr%#=SM*)pQT$)Ls7R=+<8Lv^9
zceqWalMk9g^B`%}sST1~6fP9H5jhZsAb&9e3w}?`_+UzblU2je^G%(P<q;=ji~zD<
z9UKX?-P1JW*;cntf(0rPuG9h#;GO}9der!QGXViF2VbO#A%xW+rA*Ux>H8m8S1!?x
zMho7m6>X@PbTy-c^<ElNm0e6{YwtidR#hT&*7Ut|5}^8W`!{{&aQc?Z%_&FPYWS_p
z)Nj{Bft^yiJq2)>?zC&O{ESZv&ur8mdJlIUX?nQyPu11$N!cgi8Yy%!q?JD(F;Z){
z6|2kMXz<;T<iaO(*(X!I)#=|bOCWUqYAyMxUbu?eztI-C`Jlcsj%_CJzIt}~l98aS
zU)!?Ma-KKUj#c*Rd$KdImo4GV%NtT5{5W15a7gb1Lroptnb(lL%Yb-!^=*wjt`Q~i
znjDC%h~dSnJBH9;L|a`Ftb7ji`*_y$Bf>~+IEW1eeR`Wk5HKJex^NG$3DMlJJbp&T
zM3YK*;i3Ab)V;ry!4U4a|L>sR$d2P*vr~dI@@kz=HGj!$Po@e2?+D9!X{hvor$!78
zkt^Lw=NKLF*CLw>Nk+YiJIcpI=ZYlQHF8AT<4q1E;?isa^Ss0^ct(-KTHf}bo~V8T
zjanQ4()5JP{f{^L5eaf=pyAMy$7K$wD6kujl{Xtwxq^3E#&AyJG&1G!>62Q4bRI7o
zAmzo<wMY%^siXx+V8r;&#NZBFy(*ZAb-&U}AH1+aNdh4lDupA54F01iaZH%W1dUP5
zm?%+@yt%Y=IkspB+bsx$x-#Ww@N74YsGyR}@iu*SYy5Mm08>&0_IB^)q|E93rBIe`
z_xqCZ#6SY_zrU+Xttxy8kaiY5oP~H~ziI$GVoVzRHwA}AnIjt3Gz`9U=rikpds>7%
Wn@XYq&{psWDnEHWpIgYU@BSA;GF)>2

literal 0
HcmV?d00001

diff --git a/assets/sample3.jpg b/assets/sample3.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..6cfacc59a8fd4f8d44699c5e402f0dfc263ea453
GIT binary patch
literal 221049
zcmbTd2Ut@}_cywe0HN0e0|X2$5D+8~kgB1F4hdaEA+!WVLI*`@B1C$VCQ=e0G(kbY
zcCgS?S}3BT2LY*K1FT2A;hgt9=iKjkzVCkbKRkQSZ?9Q1v&!trn%Tpz*<bU3sI8Ta
z6#xQ(fcKm~;MXh2N6R=$5CAwh0BQgL@B(}w5dh3V+?*=`B>E@j<KQHa^q({k2h%|^
zf8zp>DgfcoK)akE?4SIioab_o_MaHW!A<8W0N0;90B|Qm`M($<&7lD#Cr2*;u(3C{
z)HvXvy-V4ZjMLtK@^C*&Sd?a1U=#|csjaCCXlr5h46)jVSbY>0Yp88th|>cA#gy{@
zN(2r`x%4*<&rtY}bQ~EoRQ|#|cn;S3N9qji|4PJz&jSD`r)MrLZ5^#V$bU5e$K`ST
zjk5|s0RLYSf~yO_fA`f`0QsBVUcmJ??kV8@o2Rb;`cJ<Fy#K+R#tQiUgE_V<;QyO{
zs2cPiI>#Qh!hh4pYeoLXv$dlC;o<05_m}*jU6<5>{?vuzYye>Vx1C)K{-<9C<X@QU
z?>6Td-2b5;;PCVOi|1c5GyW^{Zkzo7;?e%I9gfGrs|?(KM+AhI!?v4t#S!9M;+%{B
zZ*LwzY3mqr9Ev;ekB9$71Gt%|0Z@81AT|FD5HZ95`ZfKMQyuvA>({5<b~w#m{i|s(
zdiNP(S^beSXo3Ic9NO-v<XnH1|C9eE2dC^$d8{l^y1FM$pscW(`Y3Nx4jI@T{kt-A
zfp_1FKgT%$cH}^gX1i|*$JCH4&V9EO?6FJ#%Z|Hr28Uy}?(PKuJpZ*X<QB&>|E>D~
zko~i+0Ed6>Pd)%p=H$C|B>u`**Z=@;<>2`L(Dnby=QzWatAT@e(=PvRo4dKne{$4r
z&att+p%#|&#&Dj1UkiesHs<C>orz9XHujdB$p8QZ3HFq*lVFZ@PJ~585UtHno?hN4
z=qN`nK91B_0PrJ6hC7g~Tz6Zru{1|Ta+r4UKk3uVUs_IT5S&q{f9U^LNQ@jF5ycVJ
znZvCU5Ex13;4}_C9vc<DOMlP7F#ph94Cah002aX!kb@DsIQTET?@yk;Fn$*YoCpix
z@a)PQ9&jRH7eC<OjF{*^4u;rsaC!_SFqVU-I9Ta;^a%<Ff8k)*i9o+dj%@{Z=~02?
z5DwPjV8IAiqB#c}bEZnc;D2HNf8nUWIF6hEV16<@K7tY)5`|JFV^CNF1AP=BF!p$0
zRFsCZA34-7A^>H6@<g~_7{{xB>AYJD5Z|>eieqFwtbv}MhPEch{C}taW8vRi|7Wo4
z+rK!rT>kVKNb>PNw14XULpxau0Qx%|+vNR2^DhH{?sEWuc=iuXsTu%8(g5K4*uV6V
z+#N3=QBmQB`}fDj#%fXm$(p+c{X6}CGW^Z?zX$*7Pjk1wzio%I2n_OzJ|2bIH7fb!
z@srUJsK{_Xav)0Me;vgC<BI<h>tAy0BL)TqMg)d&x^m}?GD=u5$K7E8lqkx{Fcc;1
zf9K)<VYPqBunYeh*Bl_h_zA#{Y67BTFaWao7~ql+03be9oD$IAezWIu2X@~)kG(5@
zje8E}l>d4Ef8D^joJVjZB^b3!HYd8G$k7ooyO{G%?B2LJGqDH&1Ec^sKoL*{GyrYj
z0AK{*0ZYIZZ~|NbFTfWd10lc(AOeU5P6Nq6I&cBV1B!ui;4;7fSU?+a73c<T0e67`
z;1MtmOasq>dEhm$!HEk#0pB^n3lt;-k^o78_JWi_>L6{9KIkCG5@Zh|g1kV-K*69e
zP&DW?C>3-8Q~)Xm)qq%_4p1-XE@&7u0eT901zH1r1bqWTz=B{HSQd;1?+5FHO~H0x
zSMU*VFqj5T0H=X-z@^|?a0~c4_%8Sncp5wp-T?1#ek=tbk`Q@_21FlX4sn8bLxLa?
zkTVcEqy$n6X@lH=3_>O$^N=@?uUtG_5?m-Q4K5=t0+$;XnTy7ChAWG!oU4)RI@bfP
zNv>C1@40?*3vna4)wvD1ZMZ$TgSlh5)45By>$$tQ`?;sNUvq!r;o*_wQQ<kjL*Vh^
zq4J#K$>O=hbA_jmXM$&uX9o&}N<r13CQv7+A2bS@4lRSWK!1a>p|7D|cm;U(@@n&1
z@p|)y@uu*W@HX+@;brr#@P6kL<x}P}<a6Q+;EU(W<zw*O;2Yz6&G(&Oj9-P{gx{5)
z%Ad?%%HPJ{&p*e%B_JT6AYdTiB0v#H5-1gD7Z?(FCGb^HTu@EWT+l}_S};ejL9kEo
zso<87kPuo3FXSy0C6pt?61p$+Qs|4Ygs`Ttjc|bQS>X!d>%wf|w<3HZ$|9yBz9I=C
zB_h{E#zfwT@`<8F%|-o0&xlrt_KH3c-4T-z!-_eHoe;|sYZ4n4TNQ_jqs1-71H{wB
z>%{MiFG@fp6eY|h$P%d%42gb;*DxNK3d{yZg<XKP!bV~5;o@){+#MbVuYli%&r3oi
z(ULZj$0c(muS(8He%m9v$8=BNp3FUMd)RwEBV-VGL?D8WxPq8Qe3L>+SxSXU<x6!-
zy^!XTR+Dy-j+L&G?w5We1Cue5A<NKZuF5<^f{|)SBJvcn4*3|lBP%CsEgLRdCVOA@
zjhv($UXChPBzH${ZLh@MgL^4^i}wDucO3;snWB!PN>L9`@8yy51o>$B8u?NA?+Pjk
zZVD+1R}@|-@+;~q1}PRP-cx+9v{%VN>9kU_(lcc~Wqsuk<%`OL%Ae6HXb<#xbT@ic
z1)*Z2lAzM8GN&q}dPtR~TB|yR;lb!*LNS*x<7!|vU9}Lk3biqHu)3Z)Mg5Zc#6IqQ
z2K&PH)$V(uA*f-h5u?$p@k&!t(_S-0vs?4+ex?21`}6k??*FN!t94w9q4gXqj<v<6
zU~gbQYO86JwJWrzbcA%Qbk6GZ=zPGb;{tJ2xLI8ZT?gIsy7zQ{=pE3D&}-9MJD_yn
z*n!FePxN8>&iYyULk8RiW(H>sZW(+v)H94SylVK)NW&=9sL5#6Sj9Ndm|?tVqF~}@
zQe!fI5Owg_!RmwahvX0W9jZOFfLFu^;Op@#rWjMIX^ZJwGcB_Svo5pG=7#2p=6x0r
z3u_Cy#i*sYrJH5B<x49Ct6-~UtM}Hr)~Brd2;2laLO$V%jjRpXhGp~K_JD1o?E^bO
zJCa?w-GaTEeT4lD2e5;!L!rZ*qq1X|W49CFWaCum^uk%indW@cg~!F&rNU*Ih$WsT
z4w58DM@dbjPp)RJxvtOMRNbQ8?zxM&`?#~*KYLhs6nM;g?)OaeeB>qVMe*wKhI+eu
z*L&|Awme*X__dF&PrA>{5tSoxM}~c+eUJOzJ}Pq5@94E-+{ZkQH68oI&&jXWZ^z%p
zztaCb*_?clyb*v8C<<5$GzlySTnRD>DhOH)HVG~aUJE%CQWElpVooWie4rAjm#Lpa
zokANze;xNYe&q!3iDM^v!^Fcv!ycTJI~jL!B3wN@J$#;KNGqnjkFblVj|4>?iR_7j
zae}GG(W=qsqF==vjH!tE66+p&Elw;hEbei<T6|{wN`h4a^Ay)9@~L~L<xeM{ewApN
zSaSw+#{bN{vx;Za&c05vPHIf%PYz8UNzqCvNZC&HNWGCJo0gQec+UD<bGmSPMEaBS
zM&~bQaAi<3Ml*FX%QAn`1LzMgXk93|@JALo>tQxFyEOY(PEgKhu3m0c9#39a-c0_X
z{Kf*Yf`o#_LWjcZMW`Zr(RT6C;^7ip$>odu7o#uEm)e*1lqr?vmHkm3Qa)W_R&k{g
zS(#b+`BK0ob``#=?Xv9U3zxrDQ>tfctZS~<qH9a)c<N&6mKpAhL8d;FRWDtCq5elh
zSi^iHvGD=x0E^XxY|3p0H%B+GwfMA*w_3FJv}v@}T#>wTp&e+CZeQ;>)-iL{?&`g3
z`q$bzl{+iDB)aI=LD%E2zwZv|p6~JMndr6cy?4XtM%PWvn+><*Z&loe-_E_me<$tE
zuip}W+vyAMdviDV?&7_p_nzPPyg&88<-usbUH{O4)xiBh)4|_{Oona^8xHqA)PLCh
zNbk}0$GVTZMs!EIM)gLoj~y8688;ZeF=0G$hkb~BchX{VV9I9d(X`|A#Ejd_>=U0S
z^Rof7t51(V{qQX2+4tv3bKG<E7osmpULs%C{*L+m%DmqE?N=7B9xsp<o-dLYH<luo
z{&=0XEVNv*vUi2Gs=az^&1!9Y{qXwhjqr`{Z_+o#HY?w%zU_R6fA{#k=lj<mB0l{3
zn6)Lf)v%4*zPIDN^YYV)Pv1Y&zes&){Hp(T@SDfCmG7~CK>sNHf%$Rcr`^xFUne*#
zvFCs}AjHQnz|Sirz%L*yEF>a<f=h^tOQ;~__MkK|SS?MA#y%ZG+k-lKRtNTJ;5|&O
z>>Y_DqV_>=KX2z_wk|}c-9|vd!om__5@<La?WC)r>-7IS{pth6c)6{hNf3}c02Tv5
z#6Z6u0DCzx4L2uz;H2GHWjBGq5H4;WC@)874nr4by8{G)fVm*t++3V^29(Gt2e`zz
z#pSW4JQ6N`Pz4%HJ2}6GSJABFE}S^=UP;G4B887%at}gES{bdPic!Pq>K)KGFf_NY
zw6Z4H*pgh`+&wsf7C9g=C^&>djf{$piH(a-NKHGJem*0UUQk$6Tyn9rth~04!K`m+
zWHnvA*4cHvyQlZYz55UP2L^|RAF?N>re~haK7ICTVR7m8^2+Mk`iGBO+dH2=fBE`t
zS1-=m{x7xu*6e@LON^rzn2QU-1>MyP1diQRT#Sob9?K(c>H_tnNhoM1^TN#XYdY@o
zDe4g4!~G*B_$8HauarORs`jU5|96U|{EsyITd{xidIktWIE!sDh!}7PSe^=kk5$)8
zbkk==&s~jB(^Jvi$FKsuv+JJRTTJM`%eC@b7ndBjwpCu_SXfV;+;*Rt$XDF^JKMa)
zJbWw&dBZi6e4j=bd|8_2Qq?JE#0DFHp63{}G$(%_O_<VNaA@dOa=*OjAs*Z#GLs_Y
zRZ3=;-<Q>|<`Ju^l&0uQ;Xo3-t8tBSf2c{-O0%&tHbb(ko*z_O<o!Rsv7?N~u^zYO
z7k^8@6NGt8mnN}2Q6uu{ysI{KO8JGu4DR;_?r9$^)+~Ll%j>x!bI$&{ztPS5m+%`j
zm%4efN$BHzws8d2tu45h;O?(KVl_vnVFX!qCOY2oSKpkmL6)EEi$7@EPf_<w*<&SQ
zlj=#}x;tjet8fibda!761W;?Ed|7C_OS@0!V`vGQU<87a5F>-|BhR{$wGoKvMkKx7
zR`WuNWNpdoj+w3t0#X;XugTdo{Z5HS2!;ljOKb|pGq)<9;5?x%S=7g2gchnazDU~V
z`x6~qr1Yio(FfPjmh!lA(F)c9V!Ka1Vyb%TYT+JSR$Oz=LZD4kQP=7qna46qx&tmg
zOPlMKYI>IZ<58kI;w|-)LD2Wj9cHcpknkk_aZG~R**Xzsi1r6x6Q|5eF0<#7>Z5Q9
zp21zJ%wtF)ed+uho+9aE#=6F0w)f5NyMBPH6q-Rh0t|(>(+PtKHwOk{$kLt}sC?J{
z;p-KuXGp|5%1XgRjb>@lbus;~J-Pw*X#vtZ7+0dzfl#ZA*M+eVrZxRpUxq+&A^qw_
zm7bCKhjk6;Z?eLdrY9%}?G()Px|ywu1M-!nvtaRctC2O&Et{SKhj@duic0?zrc}3>
zrmNROtIF69uey09tM^COp<VI<Z*JM6KDD|)@ZYA|--n!nU0{(gZC)}8fd<uCGnScI
zq;qIA0ZWewl~TkCxE`NIW@&1b2StR|R^8X`@<k+4FAe_~dv)afQ-ob)N$Jm>!P~mI
zkJz_iK^Ylu95i)OxaD(wZ*AG$mt94?bQF?xatr}!JDTV?Y1(dAn;(dLSycH8kPHz$
zud8%AYQx^5?={8R+?DVL5a~opWtW~${RONyMOF?S?8HjG`USXSi61=<78A@5n^1aS
z0~UiqZc7HnH0MVaT$g4UCkGhEcoJxKGkN?XlQ2fuMh+&alHZCemkANg3RYZm;6E#k
zw~<nQf{Z-kSoyd;b@qu#;)V3ruVdg;zNhhRQi=g@sE5}9Kt?zE<h-I$c3@M`+y)Ez
zq;XD@T3PrO(X!GZUOc2zZpgn9!0Vk9w{Y!Lwm{{$4pG}piJ!09+)g^hN~P<heTiFm
zY-}CW0DcW@L7wcg-=nB2ZprgIH6oQhQX|}Ntx)(ZL&M_Xh(y2#d|K7Ojmh!hf<egH
zsHrM*G7~w(NSBF)=24~O!}=y27eaytk@+_1Z$-MW{E8-&bw(A4(eFQ{dTP4Ud!<56
zwb+A6XbQblp=2yyZ+qU4W#=;cI{I{(45B5C)B~Bhlb+j`6Wq)GLX(L{c@Udk7B4Io
zMrW@bR}W^_BrK<NW=mPEu$wyit_0=Y3wZcZGhzCGZ(OSd$a?YvYmW&=FrUX!%?D&t
zlw+myy?D^a)gk3Y$C~d={qf4}?RruEt$cckGZ{tZBA&VuMghVPm!Ex<OZAs|e#P6Y
zI>A{U56n-M%36P{mtQ<L29vvuEoa!t^$hi^`u^;E0MJscv_-zK!75z*4{t4-KQAxw
zv1tLl7}Re?K~3GngN#qfh3F6J1|6s9LvVw`vO-uPMdNRih(89H*`ghYh2{d&rFqwD
z7AS*JgMMS-J1=*X>DJvicEgnfyiq79pKhkVocCa4{}o%nH?>|BvNw~#{kA>J_1T$;
zQ^N6I&u_6x9t*GU*mzDSs@)j4U+8;$<mFN$_1N^|&l*A+Hsv8TZjnSax0*sV5$`fH
z7MI&>a!yTYK|Dijg=v@FcXS`WS#2TR9Pv<1yS9KX0ZO)pm5T$=R@?`YfBJIPH|IP-
zf1vR)yv2sB&stEZ5Z{LxYFv_nhXkEo0hSvN>_5VrZj+VVZ@@bOR>7-11i1a2XSm8>
zw4AQgeL-(X_di%>?uiEVE;DQR`{XFSjf8*<D@%JJJMV0MOslsk742TM+|)DN_N7%2
z@%z5^A=na8M4>WoQp3{ioheq=56)_yU1s91Ot$j!yvjc^01d7^JJ|Tm?X$5*K4~G?
z^^ADS6?)or3bUpHpxk98)v#W%aC>B{wr@5a2fuAlA0Ie>zM;81-9;_TuUyI@h&vvB
z4u3bZb+Mx;vzmx-?>rmj5hfFf$sN&*YrDr^aJiDrx}w=Y>NxJzeVOccL;6MwA^`cs
zSD5)~=o_-oJ+Ct@N!wAq>!!H#?VCiV=m#(1Kc1Y+U~Q`PYUGsjKU~_kBcVs>ds@8I
z#I90rLfhBL%do2H6XFC#E)v^V4x>l3WRU0y%e@S0=0&DPkY`Z#oQuC_&F$6wE{|A(
zp$9SN);<>KB%<GxGg^$wZj{)Gv=VPBnSb#(u4kn^_c~Cj?9Of0(FWUX!3o}tmz5&6
z=4Pp5OhU|$5Y>}kNH!}>l<3Y16HT)>&o$m*z<7_+9PQgM8I`AE8m>565|wiodziko
zeyj*EpD=A>{SJ`p#O@)#0vci#=`}*X0ArF@yMA>@ipN4TR{4gj_W0w-cLYbO6{LzM
zWPS^OfAtVQ)EkI#K-0!P)<dK+tA!~blN;xIEbk8MzMOr!Kd(o&xe)dypQ+1=-bWgD
zeH2jF6@LAAUj4YUhNYcs>YI6&CH|5&b>3sLjUSs21xX_OmMhyA8lHm9(-ciG_vA@$
z=<UA%RsXyPnD409qJ5o{z;;vCaZMt+TdwFnu;%S)fXC|)_u*Y}l1P8p9Zy!frwgJX
z-n^%Yuy21}Lzl@hmuDBC7(=r=#VUC{t7>(ir@Z5sO}C}%q-RaTP&IXvrks@8P`@>s
z7Xo>vz;*+(XOw26<)T7J+9akysv8S>eHz1^)hzqTN?fV1=SjwOj;g{3NPT6aIwQeN
z*?sTTpEbN!EL#6VyLqW_J(THeqFzwzX#eaGuNK_b4bh^^MbvKVp*9WNV4YUdhU&W9
zn8_mG6RM{7*%fBFUd<0w>|4lRG+_0{81RnzA-J-l8gt?eHhk~v=RFR-gF)oaB~q=V
z<^|cv#hE1JEsRPlM*nk@=*wA#QaK(?SkXnDZad2_nh6?`U3(A?f`_CNcz8r?F&TK;
z6Q;cJLY^S4a>i{)ff!los=eUUpy@We>CJa}piSv<9Ij_*x#h8${@s4F7YXC?Sp=l`
z!D2|&sgD%xd@-m#L;VXqJtS=}X2ie|JED+O`JK2Qxt!O(rG>SrTxmrKPl1y2@d8gi
zG@zYc5d`o*H`+|7q3B<L-TBK4aP6nbFX!HOELLIMynai+H7qNOe_&083;91Zzg2Gn
z!lm;;-<NxIcp&MIrd%IpCGdBu=g!{mLg@O#6y>JT0}v(-jF_FY-@oQONXwN;9-mUS
z*|czbFng~B!_Y;i0lHjRRtVR}x$TAw<hW7B%)2$0#Io5%xJ|*+eF>+kx&m=q4(;aU
zBjZ8MwrSy(6tS`=>N6}vra@bpxg5{b5<N=`oh(^n(C-5g@N}J<QM4IM$S&_tb$Hsq
z`dlM!q>x_$!wS8mJQ}g%gpbeiY|QA#H3?VN9P+bINZS|{ysiOBZMbO3s<RZ)`;J4s
zFMd8jQo;>p*D>91{sM^8<=*ALn^#u6D;JGCVioLFcksF8H@9Ks&sQ{E!t7U_%&y4}
zB$YLFl)lr<t6_}a55wSfQN|K>sYGxR)za*g^FSt7Fq=Wh^y?4$B$X(CZy|M0*Fguq
zWrb6xKQ)eYo2Yw-k7r`tk#fa>R@^Zz?mBrA^aJjH^yLc(jg%)a<3CPcYY5PnZ)gl$
zC9vp}7g7%BL#L|~7oF{v54h__s}OBR@>q<AS1ktHs60tel*@xHyd>~6=S~&fPM#r6
zMYZ0RwOsf#1Wk(EBI&GoXH*GO7eu@>_lQR!i3`0?n_x&{s<Rfl)Coy4Bg^Q|Po1t>
zZcOJ2h{jdu5QqK0p4sNA-WUA)<w}u+hL}SUXNsQ+rxOi=6-WLEIeYWSa)3|j#us8%
z;UtqAwRZ)MI&?`s>Fp_^)<)Jm8&nhbsX9xH?hgyTVrw)mj~#0@p=1;rCKH9RoE5{}
zZR5$J5BB*#PL(&SqQfZ3G$^>zDiM>BRVhcwirm>WmsTI5@&Fws>rx@X-a0u{TW}nL
zai|rLT9n0@{&BFg?_he?<-T$t%tj@r=vCIva-O7yes9L9Cv}t+E&5`Alx{<BqFNyQ
z!N~Y#CmTz|GKeRtO|R!Map~zr)fz`JF|wT#*Uy7yeT&j&BHWd0_!uKWg!8ZgmnXXB
zqWs<^sQ&%7`FwephYO|REFOjrNlIagZu-u(RJh-&&n;RCl3bpaT%IY2ub#Y@OeD%)
z&*4T3n$0pqymYiJ^AaZ1Q^l5(^nc*r#vI(Su|`G1H_cyJW2fW7H=P-u7j9_WpXq7B
zG2WhlSKl0LHOC;<)m4-R0)4e!BvC8G=xfaP1w*nuO}(S>t%$r-gB4`?P4E)LMT4I6
zy3?DMCFb+is)1YV7hp|vz-EqG+kf-8HD(+{sg{@}BHrH?L4iztM5v<BSLVoIT{Ee|
zo87p1Kvs(ajAQ#v$VwmRTy{u50xJjWo<SC-B>vz`H`r%IAGR4>CkjMln>FA42sGDJ
zdPYVXZjXfQhtpu^gP{#DiI1O}b;z<G6a4EozB&^I4jlko?J=CWW;Aj{H}?vaGugb@
zv?V}txE1osaBu5mS%RRxwL-hgWCG7N)BR1{H)qfw0<QKkGK&a4tvNc&I?fYTYn=Fs
zq)}S79{#NPd5vN<>)h2Wns4a2(IQ)o@f~o*C!A0pV=YEX?Kx_b$a}G#`-C*s;`#%<
ziTrU_D;PA{?M-c3X|3)WodhN`?NU@jvQ6h*FCa`B!}01C)!~>5C{xoyi5s>9Zg%1q
zPI6PNA+(TTRo-lxnMEM5=1AD`il`l)K^c)!KqSA}I8!H2*C_Bg(U4+i;i)(L6H1b>
zMdZwhPop%aMnq)$9JTpi9K2S(_)=zl&YkTCN4o6uJpzc5z3XA2vKfSfo%gyj9=P3Q
zS>jN*$Kvtz@^=%dp6i|>epN+DZkBunt%@G5E)|L@JhhQk#Z9P;<^0H-t!SJ>&y<JL
z_d2iH2_5d76&3oPEcc~Ll9$jZz-{Dd8@fyPu!!Y|W5)-ZnwpoM3{~!6luRcn#a$?=
z1{34D1yZ6dw=m2)={-KhWVxbO2q^u@(^d2%toEW1iQAlgW7SGLyK2R_!d^-O&X>~^
z#Xhr&5yH?Zh?dPDx3?CXF6ko}dp2+Q2C-gUm^oAga_kR<1w81N_n8dgeX6Es>O4;j
z&`fD2U`MPqvJIs?u5KlJvBhSt)5k}OmW-#Aw!C?-okYn}mm>DJT1?RP?W1G7EqB^t
zD6|6;w_MNl4RJ!D>f?TS3GDck3;E$&+;YdNJ@!2Q%oK-Mo{K#A3n+t{tQW66P+2<?
zz8W;@n)Ntt5Vx|}8k=W^H|W<FHe@5pzTO~xS;O<k$%bqW!_qMWA~3cNJDIQW_nC{W
zSYGI&EwoJtD}H5>$ZE3axPJxqs=U^wPzu`Js<$R)^<je!L?9ank<Rhorg<p`r)6lr
zfG3QBi~v;XrZaXRTumy(Pla8m$h0w~TW$Q&2yntDc5?mzs2(seT*%74eaIUK!{*bS
z0sVKI1Odsq0|Ln}Z{8G0xjygTOXFWI`mD)(9-Qp%zII!k?tH-f&g|Kn4x!h#;YFRQ
zA-yVHglm^=$1Y`@EkDKYB5%^PXw*^jBBZU(bwBQ0#wrHxe;`J#SYT)&$MtT%SXouw
zck>xyij{+B#N&ZVuMM=kK$jdjR#&KC7*c={C?{B)R0Pe`d1aqOG)qJTl-;HVmfa4O
zraz9AQzpNxw?66JKV0u)Ewy*1roiUVZR^hb6YweuXo<~N(c44{@(9SpjG-HZ`n_Xu
zsXaF{>#s>m1qiR0P|^A46YebwWo?zp)mQ9$<Gj^>f;%R&_k7G#f3<eR(KX)(F*=4C
zVt1T%4)a(Il6-yTe}1UA|4K3V<U8pU8$=Y@YjQg6x|)voQ>Ah;a{G#vk(B78dkf7`
z=n82G1A~#H*NyozkGsa=$j;QjT*cP#;dN?!r}bx(p85i-aQ!QF`aInv{8H+qECq-f
zh;I;g!gFiM+=MpZW|Ve}Cqde#J1oAhrenA?g)QR1ENr~=W1^}?PU6>Qiw=)M+r%4_
zN_g5VolhwtC~xRnN~Bl!EgR>DrpL}i2#2}?*km^`LO54Buj1{HAg%iNdSnhyh?p4k
ziY*x_haW)`6;-b5ICmK3Slf=2%SYyCopG(|F(#@WZCtba8OMNpqvNr7LI21>Sd>&h
zVBln+tE~m)$&(N(9W|PKBXeRZ_(u(ADytf7dv<v{<VPRNAzUM;@eSzSj;>m}@wDYS
z+VJ9rcMb1HKub+xMA#IjHZSL(U-@4BfI`awxI(+EgR(rf<!k*?!bJFNj?DzrrMyAu
zRjzle{LTD#a-Hdxvi%s_zVtoN+YS;}8e7`D(ZY9XQ}>aZ`Af#X=8bh(8EUZ>doyzH
zU<SMkmNSi4Z{GMB_2g!^>EjDuXD;sP%yd6tYb=_6O;hv23QWEs@}N7qSx(&ot<|a(
zL9@3U@~Kyw&1=8gmQYdNzp9Ajrx>(Y@#{GC)$sbztoo>FGBjHZu<p|D<FYR=ilI9x
zp=b7jlBI6OJH&Kj44i+X*lHg0+v=usNKRt1AO)nH9_lA6eEDai+YWqCW5mTzh_hHk
zY)++LEkuAE!R-M#;3EE{eHORe&UtS($nqVJgQd+BOz^rH==yOVLA)qbuy>)qd9AHT
zbDLY}-B1_aPwZ?{6GO>vT#Rf|YDSr?yU{khy8X+M%7f-k5;@ye%&c-bCDO}_;cZ24
z@c|?m*i*t4Wl~uZRJTxf)!$ck)WNxs&NdQ#m=oW4A;GNx)VoOUK%UQ`n516-jjCA;
z8?;d*Gqo)>BfWsMDa+paw4tZj2Wa)?FmbBe8a^)z&&cdHA%bOIgYLF`U;JKo#Cg@u
zAjkNAuD6GQ-1MBS;$p{O7jy&P7ml?x*s1nU-e!(5QNn*L<k7igXm{%ruP3f$2q$7(
z14v@1QOv6zB%OgKXqjNFOgHQcoWmPtSVKz>p~iA$^$P3ZMdE5{*q#j+!r&Ee`;d&v
zESmHDr<SL+h`7Ply)$_a)_$j((d?nYza{XFz1YC>Lc6_*oLS1z={P?=emP0gEBRBA
zOHS95=U_#tHtKSohUML6+qgD^iU;Ay$%o6mR3lg)8uN<gsVBcdgt+bTtLdA(+t%o0
zZZSxNj~p&nTMFxfu$t_f^I<#nh-QOpdL|02TetZ8Z`-urD;ltOKVFiCB;|JDY`8A1
z3(dBbzA@c*+`-GYc<gXQ-iN!*yieg{ii9wmN;)}fsvI6@o5A1JkycVS<zH{LSl)QG
zdPj+iDB(I{`!TlBN%uVI(pxJlHO!e;t!t+VCFB?6L)q4Q9V2}o&ZfrNeuc`aTMYX;
zDt^g&W+ZTvGt}gzm=08NRMgupIrTf}0FoZJ8)ue*?=hh+A;rAL{Ofv|pdr?M{~q*?
zt}$Ej`z`;RIxaH(8!0JQHqtKZ3@%AVe}~~N%U2J~<zRYD8Uf=dCREl52QY7A?ruFx
zmZ*bFGWYumdCTWOp5>@z0*O5g6ACuJ>gV~F*V_7YBIkZa3o<mW3`cglS8o(LD&);$
zZK@I@z3Z@siIj#0V`naVs53XV2$^hClp!0FpPV(x3;pdVHQieR(f~_316LG2UsW_8
zS;<E9uCdQS)-g#lcM*uX=O3$~!ccTh7H2#C*`<j}wwGTRY^A61l=vqT{dEzcmc#h(
zJ&iJMWjFHF$GWoFi34XtY@J><i>*gC4p`b<D*Y<6PKb`b-J~-0cJG&t?V2ZV<c4?1
zhdLW%598{HJ&K?tMw8w_yJ*Cd$Q@0}qqE_MvZk)9IA|rBEjE-=Sdmx*ZE1sZZzeFe
zDr$$QQf(>_4RvYc>{l^RPu`9{<<6;CGHFb{v1Y;lRJrzGo`cbKBhQF;JM}lOR@Ik2
z*ecDbFWi?<XokKQHke=)0rj5df1siM8J+J_I5p{SBaUg1t8{jeh-%tQnT6L%AYN!;
zI#O?xUH}pLJspe3oY{)sawu4OUk$~yK24JXshT!}A=%@Qeq;V;yrcFp7~WPkD*GI^
zX;^{mn0tlu#lkk6Do|7EV6K7Oo_*oqbW6uO`L|<dW7NJqQR>E7tZ~*k56%0H;8{uG
z-f`Y{J9wUz=zolKb2NTj;%(7S;7)MP&aV*bX}r|oHH8rn2yFW3J{u8qkiMZUpWiMN
zG=`CC301E%y5pZ#cs;&RI!JeD;)U!vui01@_w$*E6!M<UFpB%@&yg-~*n2QBn9M=l
zz@xfd2Wm7#gzU-QdN}vuY*Oj_P|1-5sRP!_P2z=zt;uL2xzNG*`?NU7(hkutC`Q?%
z(2z~@xrQ?zV%%-Pn+dB)j@i;A5LOOqHM$m_XC9Mm6q5=2wAh3B%HCq!i?5SDH|<&C
z9b@09883;nnE$qA`^VPCkq6mGV$9OG76jJ2SV-{Cqm6lEt3p7u%AM-%5irDI#bi~q
zS}vt10XZo;|8&EyQjA_BUKfCjZeU4ma+YRW#c0-@)30VqFRY%8F=Ah)`g1ncCD$sK
z({YyhBU(St?*24p+XwZ_XYeV$EWE8Ho1eoi`*i~kT{J~q!2#^J>c~_&aQhhOruocG
zgnovHMx>`igZI@->IFd)$}S_eh90v1=RuAFKG_-07zbkB1K9hJNJ`G)eNr-G!zgF*
zFw^4Rjn(N}cJ|p9%Po(3vj}{yWac4w@dGWUZb356Y&bQakU+9Jp5o-4fikg-YR%YZ
zy%(<U!@u14_EGz!ll9$}ay8|1$LX<LuAWn^2c(*wHm&-Z3_=`*=pN!qBIbu+UpsNj
z3h26#pt3@-bW4xRZ>!W34oOSQ4)hEkCt5Uw4_%>%hCLNTHOHZE(SMiG&FPD4d|o+C
zoFgjM>&7bZUz4D>epx7#!Wn<?sByFmc6kby9YGX=FVnY{-j3wz;LScfe|c$&cGNz8
zDG}-|Y(fDllIGbGqY;>-USujA&v!Bbz@OU7PM5tlleDM7(RwB+@hm@(e>7x_nxJ*4
z-#s&=^72L0a>bKWcl_<%YW6X$t;=g(-R3_~(j!IZ?>Rk##F(gC$X!~m#pEI<avyf;
zfDIRQ1*-0qDwsSq2~m8l(uUEVaCyGZNq^J9-{g>YIuSfoSyikkfq<&OdQDh@w1K%_
zfVX}AT^#W8c5lP6%BY1Tm|!_n#UungXDwFwjc5k+IL^`2(Fw3cJ+1jIdc7;C^2uez
zYS_|L@=bSb4dD?DQ?qDRjqeK6N0QNA_^j6X#cIjCK{{1W*-$ziZHH|B;|NuXIX4TX
zCvs^8u+JGRu@O567QDBgC>7`d(e<SbP-p3{dRciXSTi2o<PDeM&vC6{+-Gc-8J5Ef
z72sp%64?3qgj&g_w_0=Vjp^8P2T`CALB7Gq)CZv(K0899p8(gOjzECt5?wS6caJ)L
zPSDcCk&>)U&$m43AVoNXVbBhXay}Z*)+^|ZQBUgb>r_RTMg0PL4OkkAmoKI2D#%$<
z6>Hm%zTm&r>=Tht&akg#!Uby<s???4aM3aI9vXLR@Xc63{<Zs|3QCzpq>V`{iwH4#
z%b@O<%et$LJgIVecEg+hUYUk&4tI3vRmR6-(vQoohN@ku5u33*&^QDi8l6(+$v2aB
zp57-QI+7K1rC&4T!m6Fct81pqo#N_4YT1Lp%<(5R`Lxg1Twf2o3=JvMW~5fB8P-}t
z+4UJqTjsMNx5dS|y>oPY+ycq2nZ}o?PAboOzdn48_;AWCqSSaMuXsW{qe91tFhcTo
zQPL`O9v0v)Rx-f1JapZ%%^uTjv8P{jD04kEtzn&!6Sr_v{)T+zywu2kF4_@BO={|(
zLhfi@AMNX8yW!q@$rtXzP7K-tS~?^0Yxtx^*;N(UEqqcQ54WmfPG?^@#F*x2VzQQD
zx;ax%RLkFYHcYK|u%chvj&<+jSwkYtRsOt^t2S+>P+FmJ2%;-EA!xi@dY>tuA(B8^
zv*(s8V<_GRFj7mGTje@|mtWsG617#O&f4!?vm3O1gObYPOc}NT4|{Ztqla}u3af`%
z)1m3t_QiR}`Ksqvx2KU*z4<QKj|v6q$v>pJ>E`O#Gc<7Ym&IX6cyO&K?=?4T+A9Zr
zi2J1hO$^gIfs&Q|WZW$lP78{#5EnM6)l#oHjSh`?9Ji|G&!b*`xIC}DUk5G(yk8=g
zWxzaR74jcNH{P7;4AvN=N2~JzL)9yANirhR;gBt%L`Bzk>6Js2^G^kd*}=;t^Al1_
zu~$M;>pId(2dKXn(?6<(qQgp3y|6RY#|)a7z1oUo7lVx>Lyw!4O|!CM83PV`1l4T?
zX`b$1>~daz8H6q9`0xytTIcb8W*)lOm1w%cYcO_OiyS8#X_M7?osm_wT|XW}#Q1oK
zTos!NaMsA&w7fU3VcohnkYw3+^lC6Y;_3V^KxhwWVZ>1z|FiRXvu?WR21==_*e0`<
zw?eV7ys%p~)HUcnPm=zoIVer|l8)DB{MS@uq1O<@r^KYdGHdU}Dz)e}M{w!S1iA|+
zcoB|&=T>^`b3M-D>=09v^C`ie;Z7@I^cjrt)D+=z$JxTSrm~!0IITH;tLjWge5Kh|
zpWv*`8uABiov3)#RZQRO^WJ#x3vZFzjNBg)pT0JE=9-Z%9o)}!56AY$zgb1-H-HK~
zhdOO?FjRhIlEjEZr6zZ%P-cL|a<R3pxb)smX}ae}b@lbbIp;jSQY6Z!9^}#3l|u5l
z)uKW~Db+&7R^6{~yt|@QO@LZrqZ!(ri=?!-n3|!!sL^kGE2<`2qH$<BKOG}qJU1OY
z8tQ6}h~ID|TEqITNmAKbZPfI9rLTlW`j+R_bKoM(c8wOeM^??Gx7o};t)XdwB`==r
z1f6<fA;^&V;6PEgEA-c1s0)JWNlJv{qFKiosc_$|A2uuv<W}PPY4Sek?c+A=Sy6fL
z6^M`a+zVkLddwfL*Ky~>T_>)en$ej(At}j#54@~v?_s$#H;AYcGF$ml#{;g|*qIB6
z83om?hwf>{?2-2Yxx0D<D%Vubc0O4<Ty`~kXyIB=Tf)skrv-IvV`f`SjC^k$j?T|*
zj@Xv+^VYfoE{atynV>p!U9}J$>be}KrQ@O$Qg#6=lT9j`>5wn>4ER1tFsLVx(GAJk
zQ_S`W7JpQ8hYEaUdTNc1{@5>{<?h4Gzd>_T$}R6{7E=<Cq&^NmMXb6{M~l+f<)8X;
ziZk6YBcYqFOWl}}XyQe4sr#&pha^DnT)Pxbrw=j{L|-6VR!o;K{^l6knGcIy$Y~|`
zCq+<bU)YF-`nY~g{V{N<5CfT$Fpg4`>r6$Cqc+~znq&h65#+?6uUZ=991FHR#>Am*
z4TBrtM}Ho66ZesNZks6Nq96=fnmYO1czl4~;^-j1*cvn<dN2NS<7@&OA?k#v7Ni5d
zMGY@YK^v)q>^wOvejuOjdPu8UNMqkwW-c?vSyOSZ?NZwguHJ@TF7Hz~`%JlUrlqk>
zdLSblHm_cFU+kRD3HwZccw<J(Qm2yJLX~vyGifDKc?_l>ChrZN845@uZgN&M&g@RM
z+PI*fGohzdOkC}u{&Lw~p(E<AwAX#Jbw=z-^p#MOzg|w+>qxsIvv{k#{(d9w^wPCj
zl<MoA76Woae3bsVb01ef4kkE4+H2+QUhiijGr9Bb5Wl&H)z`S@dlTs^X62vlF^)c4
zc02-E-#+-3jYh{<#GAG3rG8}IG%R^QjB090m}bw71}H7Inyu*0W(^^U$N0^~^l9EP
zw~{$iiEU<WYL_i{{F@hYo|UXY`|<csS6Ifc8t&-7p;RYpBSmf#2TtJYy(~hLXzp_3
zS3I2h#xwi{@Ydg6!|I1=kB8~aye?#3<I!HGULNKb;3G0lY=7{v+V`q6ceQaZW1z(e
zrl-lfCDQ=UkE;$eosX(DN^3pA&uf)HzZeq_zp6vt+F)Ot`&l%bF}&%#&o^;he$^rH
zKn-DX`h<m*o=XxBw=2;ohvIS(7Y2uysUCWGQGa{khFHEt&lqkX5P~k9*z<B!2%F*a
zf#{7oTY75p?8R0))reNzBC_!^!~fX)LXjTt8Xcgmm7fgjOp_Y))~V{LhDL4U$V~Z?
zn!WNrFp8XL<~Q#WXthOTegAeN)VI>3YGm-lhza>q3xi00NOMqpR$bGJ8b@?&c!+7c
zB3+Z*BXM-$Vq^n-jOFB?)k&X4wC2#BUvq?*tV*d+?Ssk@5r)J}>`c=qMA;b1<@HJz
z=TyxC?h2o7led^ZgpcG1sVd&b=PKrKCd6EQ2wR?f-gu@l-6}C#EeX?(_F18q*|a5`
ztC=rkS8>ioT%Ay35?Ibl-e#SrU9F7W%#JPKvWQ{*QHpvzcqqD#a=!I=)*HEm<PmH`
zLwJW+53s}h{kMFZcA4LxT$8uvqHsQ3Tj#8!z9wh$GW_)>?WFDtx-ZFmBIfRMFBszY
zfXy_R&1O6V)pt%y?hAe<90!kiF$`|M@;+t%KqQI2U>@Q_ci8DtVF?LunXCyG{|=2r
zrzRQ*J5c=g`1{e@g<?pv56zbA^7~rt%gmPXkzUtyBd7U84;Pm;p{e%<iPGLlQ*^WC
zN;N5qjvUJnvpA;gaA41fU6p!V-Vmc@<8HBT<%P^tgZt$T@*@=4%m7FmB@W8ewp5av
z_MdZo#95QGc<0?_&%^t5nv~{=@Ro-P(M|0AxCJ}jLF4J!B3FAF8eHyPdtV_R1h-v~
zPjwUOSFRa3l654{NV<k~UH;4|k5;Fs_%|!Ams^eJPd%Y1NNiXWu>}8w2+I0dZ@f`A
zGL;awib<N=qK7340h*CEo&L%E_)R<xnE%;%^L$u(X{sWcVLG|pWEQXpa&UJxUUany
zCst82r7DBzna`PMoLRyTs@}#Z&3S~&B#z-X^s^`(C!83v6upTT7?wKNnKlU+sPI(e
za5klT5evnqg7+2!)R6lJ;$@-mr}uOah)tik>=8=N^AWFP9iPGk8HG3Dh3;Gy9!eLn
z{_y?-`Yq+F*p{i0g=-<{+=JAtu3&kg5mjojK(fMZqsI+zPQhuXA95Aj3dSOpV0VtS
z)|T011TglyrdtIxUb@>Vu3+g^k)zgL$a90-agn%ld#v^F)?L|9Da(3l=}DgKKIbH}
zdwKbeeYp0<M}m`pkgXCd>B#oYe)%Nok!?>XPs<PcS32IrFXIjGE4OEVf9KeZIF_hB
z9*kwr9z2Ge-o6WeJ>_0#xY$07zuP39ItF*`R;X29DkNB)@}Oi&u<FU2<#|Tx@^a^<
z70{yohqA>)F4LxbtL%nxt1#zSU8JCJ@s3Jm#*qGS{oY%QZ{Mam%B{=_t@=J%>A2%H
zrgIk^q^>&Lif!Qby+p}yu8vyBI>tuKMY7A3;#Zh}!s)9i+#a`31&OPNQ+4vdPH@yB
zGlX-_P>l2W&;wnX-Yqn~<<R*H$jkEDS-4u&uNds(bFbm$Rd_hG)lp_%_Q>#_wVCw!
z1nG6GJZ<gkly;V|HaE5=<Gu>)$#LWd;v<U_>lo;5bJd8*DdMG`zA;-odScX&eSFbm
z6;OmA>Vub8Re||a6IECO6f|b)h!{frpa@>$0bgck`~qf#<438Ua_0h=+dVzQ{W_2C
zuKAt{dcmidQ8iG1ICi^5MVVIfdr;&bxXS@~oY>47>mfbk^kdSw!V9*-)0veg;_7!7
zjxw{h&6MuXnCNC&eCR;*Tc$sC+x*=GlLCd-w1#Ct<M*#p39tMOU;~*mkU3{R+@lT|
zi&G>E6?UnJ*mu_pVsA0^3Q_j_HUNZU9EE6^1Xs0N+bn-&*P%6sxS>tU0r&hkG|T3G
z<Y5IVsfFUmDoRENV#Ld0B+?C+solS-7Dew>QS=-ny5&A&@X`B+rcoH@8`F~sbV70;
zxQXzk?OQ@!RUjour2+2qaCYOcXOi&C&aBNbT$0~h{3r%=`gE(A1VSnAfYMLgFF^O^
zq?xFWL;>~jQ@w^YQy1*m%T1@;MtBGMMeXV8g-%f<8Q!lkTG|TK3n`)+dt(#_q+9!}
zxXDLfiSK_9$GW@FbHQK7HE_e5P=fxJpr!lVE>enrkf*YhzGfk*SRw1+={4AT0%E{F
znPfN}A7-*~xU?&xXfmjY@U?}>V!$(`niN7MkEh{k3o#<F9e9OmiD8$Fe}RN$&&ztD
z{!mFfbL6yo-N*7WA>@y--*p|2QrSl059gMvs=Hysaub_y-Rm}uCH>ZsL`c6+c5{w3
zQCY~=xAwk#Ow-%h4F?6!{vw{{_QygNi&M0~ym>kw{L(|j8@0-}dgVP~;s%H7N4EAR
zj82WcEfrq4O+8gN8pJ;zPC3i}MhVK~N+2RYW3Bp2Y>>4r-@9i?Qn}?hm<*k-L~%oY
z9^e~s`&~a+*cx^llO!>An{Yl)<duW?BmOg_)X6Rn()ZtOui8}$Mw+hbBbt=5o6xY*
z2?3-u1-a5?jx+O5p*=4*myY{`)c~x_B*At*F$Q11!-GEe)o(qKn-UcH1+-PxL6n?%
zen0PuTQO<wBJ^c>2+MyTn7S@KMBQl$*KhMWPR&*=s8_|Z>#^e}@@$J;ik3TybOkcV
zDqXmHiRvnn=m#IORlCs_uSK92`&;)vRsAG%+=za{QNOYL#UR<P#AS|>UphzP^?O#1
zU@sB**5B|Y_sG?jK7n@B=gnn2e!!M;@|j^NgqbRq=ragoue_g|xoBs_vWy?cGD9zp
zyWOx{zjtvnKbH2D9W~oKF4OZ1xO2#max3>*?&w9uy@rUB65sEYKFO@AWs&@sB{nk-
z(F->-H(XgATLPc@2cfw8XP~Wzf{^zvd?+V13QwI<(xm<NOn<xS^F;@35mPtLnU6xu
z^_GBZfxUmM_xFD7II?d<#ywyY5Lb8S)?a#Z$s@8&MkIQnaZb3Cb=s8Ve>I8H2RhQA
zUjXZOjg?v`1dX>Xx1Ngo>EhFKzJ`hOv1C3kUkI7{IF$7Dym=1BX0g&^N{%Ka+p}Cc
z3%Tr#KU#QMU4PSKVTH~))MITN>9EF57J+8XI}_gFA0QLhhZIu@yi=#|n^14Bfap7$
zW;{WduNj|NH);JLnf`;5UtOyU?agIU*>&9~x+laoID6Xp%*Gd;ITKoa-i_F9gEiZ%
zc=EBvw}ZjY@+vE(SFQZX$k=*AuwhU#yv#^gjb|@bw#lnj8X>}zxpVS0@u6J9WU93z
z=j#Cqc+v4?^4g$6a<-qnUv4W(P`DE`$V4G^vV@C1ydwl=B*^j5Y?I1Bh=oG2wW_@t
zHVx>vaq##zd6f&dJ_|HgXyhxoHyODxx{sQg)t364&AXFwap<V2w;Hd9XkS%I-Ki*(
zbB6-*d|E!SW=R%6Ov?)xu~2qtKNI)-N_E{+i_`aot|suChgGFqyoT3$uHr+?a28G}
zuRkCDv7~fD<DkVJgul0`_>_LlT(QR8P2vU9oTA>Nrh1qzxnhZD-$~cZdd04PVjQ;r
z#CXeBkeHY3?7-(TRO9<C=QW=Pgc)r2Is1gT)SlY{!`n?_F}&;BE*`Q$C%C*kZ#TKw
z&KDYr%Sl#p+nD<{4<9DYmBE&&4Jrz^`A;FQsSbsQQ=qo9`5-BVaTXgi{vaS+1l)OM
z1V$;O+gnPdY$dxrFHCk#GZ1fa6nUCsf`MhlBDb6_yc2Z4XQ+;Qd~AALzm3ntv8NT7
z3ZoJAJjf#skS$y4YN73(g@)ggeSFU4Ut2Cc3}@<Z_y<_9BWD;nK2dF}Rs(`0d6jam
z`s;-JdnCs#2Z}f%K6}G{A2m}UsL`v*i;!7f>JaqE)K<t2BjSA~TwTw(=7W<>d%}30
zWNl@_bp8Rj&1`<F2C=sS*s@vIk&JmH;ijZ`n|fV7pFf`$HVn@syK9V3+4(Q_)`&J^
zd<r;=i+g$HNt_8Ta+uCWN2p1CAX>(F)HH|Rkhkq`eD@=gBrqSk$98~D_I}fCn?gYh
zIOzAIRWt<oegTWCRiTGm`!b%?iDV*=-$PceY%@*b){Q(|P?)nruyiR{?+OkGh9Ed+
zPb2ZXjymrgkP(!m-l5M7NJ=E0Uki$x`!oWE?X}fT#3UsqOpdSQJBWMj4c~715Iowf
zbm!o^nfwZ*o4A$D*x7;7CI3Qc`7^gMcYZjS<ID-~s-`9h{%1^BCaQq(=Y_n=Gp@kp
zY}sp-lfY==S+$5AgX&<*SEZkg^sq-N$*<un{(EnQI{E$dbvWE@@8A2BRVTVe+C;P}
zbShuvb8pfPn2GQ{H!QAJIQ*%;R?rDuDPLDqV40CXdSKREEO=Cv#5tSJnBSuidhC?z
z{tyxBK$MG3G5YE@-0Cn<$S>$pSxm23H%=gdavp=~?yJA0-wI1DYT^8E1gVyG*yJ1x
z<77*>SFJTFdMI(VMAs`js!r@&HW2(|CY?7<DV9lo$sn6HAfsWxt9Pa_LDQ5t=JPdt
zdXa%_!uYD0G{X*aD@;aLw_UlfKjcA9`a><4dy9I1{IqWV6QT2jTkg@z0M2}Y?@iD<
z#d?ZhYx$-@#!_XKkmXI|1%^P3PgfXTzl}Qdtc>D*>3Z*!g9ti1omMrqJZKiWnyVk;
zl07%2P40L{5O`P6AvQ_xmU`AVDPLGJXXu|#t8|us@w0QGGHV>$vyCfVMCXZp^f^;3
z`|Q0XuVBWL&kX)+S)4<${cg{UwyZe&R`zCc_088xmu;YK22zcA6Sx7dpt*!QI`kM`
z1d1L$iO*NaB4R(aiUA1W+iH4Yoqfr}=o;FV>(vQiKI0{#C{*g3YmvS<w5H#vupF<J
zXLzRb%?JEI<qA5#cU1@_F*kllo)ed?Q><x8oTv<#U%Wg}Re3N`$PO}jTAx-w_;a<D
zbHGhveEN8%=yE;DLsBqi{eYW~yJ(lircHZ?fC)~zVlKpI(gs(7n7eFCmLSox{3#XL
zdk|U6wZYVq+3^<x%v>e5#D0t7KoXs;TGEekxypY+f>!0MtMC9ywN*rt5lt7d;QCAp
zO@ALj39o|O(g`u`n}Twb36Toetn^I#+d()B{+q0N3iJ8+T29(8AgFJl++U%+SnmQ!
z+5_euqq!*k|4?+Efo%0{7*A|!i@iq?qZJZ0YZF@}M$y);gsN2~O6`#vF-wdZ31XF^
zsA^TkYONSmR28LGmsPEL-n`$EoP0PR&bj~hxUS#bXiY(LP-HS6R&=^K(!H;tFEWp)
z+pOK9cFKADG=#>7gw;vSRD53WBruq|b|I}md^S9o6_`H8z0<5qOK+`?9LpIyIZnAM
z`BKB3JK|XOM5ft=etFNG1IADw<rhibr+|ZuY6Ptc2wzmZY7MNOKj0>{F=KOWBL`n+
zj|1lG69yh<I#s)Nq}H2WcIdMfk;dGnMVQ5S%r$sO&^;m=GG9yKf&Tt(``4mRm-mCx
zbslHOy|rwFNA&ZC8Mv$16n>01cBfIAzxrZsW9Do^y<CVk3`<`~vnPZs_X;Qtuzks$
z%^){BFrqO9>w3<)8o9`~qvZ9(jxcE%1g8=z9qSzIqAKIXGHitl_$f>hjF#@KnzW}(
zH|?Y1zHauh+_&I^FpD`T*BfG~lC<{X?qUJYwjofQgIWA5u!mv&bjkubu15bp9;~$%
zFFItW_~+ht!^B4<H7-o97GTD-2sjx1-QjhTf`{*1-gO}v!oMgnr#$lqR%pf;R9m(D
z5IFgmLd&9J8Kt+nV6vh)BnHnJq2i78KIuxD21v=|Y#hQ`fnKk>QHvM9mE!Y}PgD2(
zb>1J!6U{$0FfjsjPU(;MGpd5GDj6?3H76S>qX1{ivsdZ`)?f2$G6p0M;lajJ(*4Nz
z24g|GOtvak92(&;ZJDb>?<Akh9o3{{I<HdeGYqP5Z3#vqmbB_G-T;&8OU#j}bv3t0
z5r;Cm*$OyPwEP_ObN5zT-Y>p<o6AYpndl!e%;G`~r~0<J^mojqHY*wl0KD`P{Eax)
z0W#J@HpVgNdcyePffUD!g}zd;7<a{!x5HLFx>D}t98W#jrn5qn6Wz<Ly)J4sm2Qqe
zX}(6=ueW}}&F+?ciN+oh?Ox0#RC%Yx1x7NiI`iMtOr>;+A=?n%Em40$K1dWCB$b8U
z_^_3_C(j)2FnR#q0<rhZuUzWuZXDC`IjStOYRsQdYS(~%V;y*6Wt$;zd+1ne>-N=+
zr-MrtP#|K<G@0k^J(y0L605kGl<36gBg)0qet81Du>E(t7?Co*aaz5X8oT}GWfW-e
zN@l1NN%Do=*({{ydgAwZyt46*VoK@D=`v@I-kN}B;jGttV4{C7&pWkv+MGw+7GS*L
zlPQvDlRHl;N?jwEgF*ClK~uI;e4omYNxnYNR4q@OgTbF!K$`5*e7=|@)==^LVp-3Q
zJ(cKJ|12A@Yn>?OGmn=j{5<m?fVaMX7dJJ^O!y4@`Va!q&vNzn<21C^E1t9s4|l3h
zEt_G>w?<_B>LU*rU~JX8U%GT&F5I^+OnEbQTw5GV4r3fy1a8}?#UNY1sV(JLp=rwZ
zQ;V=GN#cxh)~RI7>##G~1ZNbxbM8f4D54DM&B!kWTBB22BQKYM$;H!Yp=VBomp_ry
z*|qf8Wb9XPL7drRmkXSUp}5WD-~dvUnjJVhv7?O7*m1U5Cc$a(#SZS>c9PKC4I_dS
z3KyQ~e1EtP3p&>J>#-fO!{|8^ag3ML6oDNx_yFK)l4GM{C6C;lB(rUKdRztR5W=-G
zZY^-1()@0E_S$~{GXJg45dDd{))^;~RuyaELLO$kr`nx6z&whq;gOQ1iqLiLm?w^Y
zH7g@v-+S1&{!vC59+8d4-7aP?+4$AU2=d>svI$6@(YW?>${$zc$dj$~B9F(C<7x6i
z;deqy7gw*|{4+>U-yQAX+LqyqCeD7Q1y|AH2_OMf!BIna#Oj>PHMT3`<2D=&c?$B)
zuNSLh1Zktz=WHE?qp~DkZ^9hC@|adb7){%JOIjA!M2xV^pw-97ff$rWwN0pl2iHeJ
z$Z!10FeG$*6jnlRQ#8Yd{LU8eiM7D$Zt+h^$@%AYA6GI8?T^zZq5-$WBx-ZEYyhE?
zSWebTrfjh83_xFtxIYoV?Ss2nM{6Q+h1=yk#u8<M^`86EZ#J3=#yDzfHOi%D(neg(
z)*>=NnYp%FhxPvHUV0)XOX5oAP+qk81A4=u{LBQ4|Bt@_o(Y!o2`m6+0%BPID}CFK
zT)L%-o|ZnyjoUY~N@o^ITv7FFPyO<Vi9~gFCxuG>FL>`pR`Wjzf?jy<rnI$=hfGO#
zwRr5-S}SqaXTK@4+$i02*tm+&8)HbUDZXE6hXFI0@!m5ttsl*|{jl#buSxd;#oG++
z<kL|2+#q>{smp5fUhB(y{N>S?Hnp@Hl7lGV>!62N<OMpdag4vp0}s^dIf#R+61<~L
zgp9rHlu;zAhh}w(D(ZRT`!7xJ8*1&^n9Q{!>^|B$UMvvHvr{@RU=1&uv;Pdbr=xp!
z)#*@(pEA2GT`C02jj?+?XmhN1R3qks$PkSWw?P3L4CmSaT-b&9Y-QnZw>cgfQa5YY
z^`f;)#m&JB!$JOi`SEByq-cC7HU5&x<92xqnGVBKbA5d@YhEpIRI9dwRW?}9F$i8H
zc?q;oZg1e$Q9q&fvn&>TgK{zK#j~QXOSp|cC=14Ha2_mv!m$31P8?`TnCsU_IId6!
zJ6G*`hSw&ELr&*dOook|zI-nDJKI)+daB#cpP;qH-x%`hE4#*G&zI*{vfm8^8;IB?
zKA(PPy0utYyxz>tbNzIEzO?f>stDjih9<A)%;SgJBswIGhDp-#BYTYWVr2kDVW~rk
zxVj<`eul1;=-Guc-=ipPNE#iI>1d)>E4*)?l)l2i8=Z(~TAotHtxD~Y>43gWZxi#3
zq($vT$i=ouDnGw1+gcE)l<^W`S<^8E)Fdc)TNJ(E!qJg}h6)P3wNXZT&+rA*7J`86
z@!pN0&)X~3RR%C)g!&{pbn?`o5&;j>O84nR^aOL*9E(6&+>@?_ZFGUEHBG~GXI8!a
zZW;!7>y*;zwYi0^wbwc><iq$nJyzmzRFQ<niHfGL38Jb-QWV8Tfw3zJbWb;feqKL3
z)+Y4;x8e1!Kb<Xg+}f&R!X6Sl-n7~LK(nsZn;2=69r4rN(vzUB`zF@pSa$ELm?B^X
zW6-<YnGgSuBi`LNwlE7mPS@PZEOU7ihnn9Ai%}~E1$`J$<&6bT<+bX?2#Y1WJQycR
z#Nc?&2<0M%7D!M|(&~obCZ!CiyoNq`sNv~8Vzb?^Dje<(!pYoQC)8<FL|4i9j`8sh
z{zV07Z+2*?no_V%3&as}E<%CJ${j~@x;8wWF?|XgC0pok=jlgLZgx7tY57U$@=Qil
zW2kBv{G&ScH0m_PLaFwun|Kr37hwySN3ET$R=x~;y%8+gX^MkCcx3U&k=`oT!T#NC
zH<Ji1pQ7%~FO8Q`QWyUpP2_#DqmM|tD_wGSiFg8&u+U+RUV`C;eOm=)5<pRlBD|+z
z$eL%y@Jl@a*cUUC0~%zA7}h3=InX8QAm1kqVelFDEEXiq1>T`%?zesaz@l}{2s!iO
zSenQ+mFJW8xVeS)uERKw@Gdp!B!Q^LJfC{FmXIY|c37FMRSP`Z*7>Ih)Hw$=zJ4uP
zNT>7U!iazj6ZXAp-s;^aSD&7g^yz0XtQ^UVyNbv-Jzgohwcf0pO^L3%+UaU#J?$mJ
zZ55*37<i>mZqF$#BX)bu=A>bh6Y+W58rh>MG>Pb_I8rUbzfpx+VJ_)%@rfX%q2Dec
zIvj;EnUtD7D!M1+MuuRL3}DmZnH6iot=bJ55Cs^T_XD#+9f(l@W-JD(AGb=I$4_gb
zBA^@kX;vnn4MJEV%eZvjmJFUMR7PFJ1sJ1{5=u2aDh+BkE0ldK)5QmojYMNfo?f=K
z=4^rP`7IOWN!RzpZ<1~eG5PwqFHsZdS1A$2xmT_pmb^h_PPNM&H@mX-kNar22Dx+K
zLNJLra%ulU8190$0nYu~H>S5TgSILPhUgh+wkHgGsU9D-qvp;>@-dM;p!m***}2E6
z>2E?XaznXtn(u_v!-lIG6W=@7+)}vOSz3lAn5%!ix@B7|g+?V>U@DJyCGJ5Ks4W_f
zZo*@#`c|Y@Q5Ie{&jBs5W&!DtPbGFcQ9i$&s??2=CeUwZ`W%wIb{$3%%GiI{S~J|f
zT2_=h98&&#)<uwNrK!Mrwkr~M-Zb~p$rq_Dylbnl^4x}{{P``COP<Xi5uvF$K(ih|
zfLN`O`Dgb3xr>JOI#WRyn%%~~0svxn2)7#+y9)(8oIqo<`_O7)8Y62dTO{~f2v*Vx
zWuS?5<k`Lf@`QH$aXt)YZE+<;s3;>(oJa9f0>l-?%?nIB!9UFgUcYb<bKDe93$D4V
zi+lYX-%VWDGyerp;!`}T4+BfT3^*dHXWp$Tj#Rc)q&V^2pDQaL_MUZ|%T?7H5usOw
zj@phm166g$*PGdlycGy!59`;Pq~0vqOOX?p%yL_(yc*8+x7Sf6M=$tC4Ph3yp52L?
zLI+p{GdG(aB$*OX-XD1Kp;O@=RJY}?xYlMWO{QA?#@}x#Bv5t2tyii;+poKlO)#~A
z?(h!f;%g?Igp$iAWI!`3_YW+pjSMY;w<due-8zvc1tS+m46vm5xSvtq-P@rN&yr4z
z>-1bM&}t&?5lmC21}Z&QP8L?fewWX;h^ph9-PSsCUFk&zt~+3GB>r_>(shW2PXei#
zM|O603^T0~1<k}{D33b%es>&77pZ}<chE!Up{Xvp?N0u+8M#h9VyGBVm@l8Xoto0D
z?@vMr=phSw@DEA=^16Y_?n{ea!1waCVr<nGml1ByGWweu{;z&tvtxA6QqC={zTPQH
z;sCn&HvEmQ6xKsKBfzlGs=r`OglCT^3DxakK8JA2*bB?gJx-<gV$zg{J>HYJ(%1bR
zw7-HOW%FyGRb=#JGeB{D+i1O6iY|JU3u})B==_MP`P8Y#HJ_WR;{fdcbC383$?Zv9
z*RKbLu9Ur~&~cLH?KzExy8oIBNOfmkBEhnCop?<@(ll<{6l*Yt0+nDmY(q$#8eowr
z&F<|cLd>l_fkq5d5Tfq}_3lX}eMN14-Dewykzu=nFXs_JYSeumZkgv?u=t(8)Vjgj
z90~qmA2G{jSzfZPMF`guQb@nEAOGXXXi|puI8ui4^%1U^{#ucie}M)KL7*G+XS#3l
zr--K{<Xgb&O0Q>G9e_$?3LTpwQ0S+V|F&%~li8}uGTTH%Sll)x!jzI3@aZ>-kWyKg
z^XJ)F!?%pqSHAe8tensG?qK^LJUJLUc%)C!SkS*Wj<Ko1Ble%IOE}DMD)I{7X66Ti
z{w;cpv^c|`IVO$a^7kAg>GJWXuGK?!zt6sEI%07iug3=evR70S|BCXAf_mQM;`@5+
zj*NG+l>>BhFM3z*5|*aRQEEP;(48}BuJljIRp(0W1;Z;Xrokf@*<$AJAAjA?v(@U=
zXP6s#`Br#c{?T4Y?n=j&?=R_B-bwoZ0dx?4%vGd!Z*uvn<?bssvdTPhdqlQ2a{FvS
zYttu0CbbyoR5(1>>lV|JLjDRbJMY{Rqj#^s&e&NV)-wFyzFm9^H=XGXoph|m^Pd#g
zeVTN87MxV}t!WV~c=)$_E+KO5xgyLRBFt{7-(8O?Abk%)*3U*PUz}|Qs>-`(1Dj$#
zVgUUVwzJCf2rJwNR;QSDM!8$w`)8(vydZw@-_6sx4=?i)fWt`V_&kk@IYpRhl0-Pg
zbU3I!>hAz4OI)J%!J0Eu)KDOYs3L1q$X|X3XWEeJxTQv>qmJ6R@#iktPnzCyjkwHK
zF@PV6k2$Pr>tdh^3owhLqiEw)2q$oz1lUSc6#B+YJ9oB@pnz`e^N3XdG~R3!bXL&^
zaU8N0W-9pc5e@t*A^12rk99}cTe+<_-{X$)JZ1K|&0W%!z+DuPK5kwgZ8lbM*fa3v
z<<!eg(T~+9b?@jdbPT&WJ-_MZ@j7&ELt>4{n{;M!P*Z#fpL&<Mh@bsYK+kDU)d=Pz
zTi#02)U5oj{_Yi7Gw#0*M>YFWCsk@~AnHa4=4r3Ms)*-X$aoi-oH8}+X61Z=c`>aS
z@&t6RYQ$Oec<!j~LZ|*t@W6oZ0ksmJKiX)BDr)F%G*#C@%;N=BGsNbT?S|TOJ9rdT
z(ETFg1fmwWE`?%Z#(^b6c1@kDV}xCTHE<LDr;X&Gu(u%`367?pH0|909onQ2Hg1eL
zGRE7^xhe#;)@0(ciRoo9xifB=#Tjs1p>(1BZw<jZY-PyhC+c`h9rv+Bme@I5pyUi2
zIMEfLplzK!uXQ(=hh26X?jJ`R7Z#lGsy0l$wORzyH#_Hs+UX`KOZM6EPh~D3Q)or@
z`(~@!X5Qf>`0SwJ>1JcMLZ>`Sdeve5*cFmVK!-xn!dG+_U&T=2)X0q$lWAL*ym{wy
zdgFj1wT1CnX`%q%);E#+UrW_ZTTB0@{c616rFPDTR7yvy`?fujjF?RvKrYsmi(E@T
z=1ZWq2UiDV3wbxF_ww1)_hAqV&dGeH#9X)I#)NB53-#a7Z!(>TlyD928{j5JcDAQ;
zKJ!G@HJN5<p-P{2IJ{N{%+her;^gvSo%rj$&=&i$^P@*JB1Vn<n%2A56e%HCH&BY5
zEjR@0K%W5krRe=Y31EI_KL6AaXh9}UC;}#5xj}}|@yX{M_bQE#M)s{<k<<hVqb26x
zHZe5Sj6}h+;TEM__Gdasv3M?-raT#RQxq-Wn}!l>j)h?tS*J2_#@gTE-Z?kG8p9xV
z^tjY+**OT{8&0_67*)w26#@OCw?S_xel=mxWTi{Uh2K7XQIs>oJkXe*e^T={?;e_>
zJWTNmx90cFq_^hYEcJDF+_NT>`pG2ShCb!2O<?-{FW|sOoTA31X@IcYGN;v^e@){$
z-rN=X2KHe1>+xU(G!w&pE`2)PeVyZcD(&y?>GZ!Eu9No%bt9B_shT9_Xu*T}C5JA9
zU(=WZp`K~_l_iO%s;2LYUL8y;ExU#f2HsfAS#!H}oKpQCK<KP~10OzU(5cG&^SmI<
zsZ;zy9O_?L(xP;o%QMN9GqS<6=04G1wPg5>IgaziWZ(3K8M_`XUrgMP@o~Muv-N%9
zvu%Sehr^6lI&qYTi|Kb?(=2l|5l&Vot!<*0wq?I?b?t8Y(jZ4!wDswX-HX!^GrF9{
z(}zpy!gou`et=>bjgH$DIOaQ0r<UP&LKz+xlRrO);9Z(ETLe`(Zfr<Mg1+xJwmz8u
zd02mpkkt4lJ%5RH(<6rj=|s1`DVyH&7JHYxbC2OaKr@D^=4$9>*Ot>rQc2n|t=W1o
zcktK^{fb6(d24>Q9cFrI_`)`P)A?-s>fEJIM_H%SWAXMKFYPC0FQf{KPAGmcR<l67
z-nHA!zwR%k2Gp(%Efev)_Uu1^lhj%V{(Q}VB*kB4v7>>o4B+}WLXl70pj&kPqkV>6
z^@zl;6{g=o?BqS}la(BLG-9_(Ky$9+_VrNN1`&PWbV|2mpWeOt48T#Nunf<_>`o{u
zgC5bPuJHe0D!RD4_cY5QqAAKwcH^+AKFG;-c#ZHz=D7`y`UG}8ffh;=%%kfPvP0pX
z=X1i^>ooiFmfjI*Qp`~(nHN~|#{sgEMlF}n;iCZb6H|qxsC;n73A@m7ZF?U=ZoWA-
z!RYU@KVG~p?bG>1L;C!cCcSLYM6*!q$UFSsDzA~YU(AmE%00=l2AgqyynK?Sd;ir6
z?24q~a1r9nzfpg%^4Cy{d^nE#I-L0vO!!5?1D<A;TVolr&kJ)t!Th!T42A-)^boxS
zr;4M~8bypVv3Y@-<0!`kz4q@mkWQ&qYSwo<Gwn@+u>#LlJ~ek)=B&EOls=qL=p|qM
z#Wgg3ttx;+kk7<<d9yZK0_#xRWG6#WhmVLW%`oBe%wv`48FEfggkGZmP-j!)9M>0P
zx3S@5gF<^9{<pdA@IHrgr$|M(7;b!2BZ@Emr`?icfT9qI#+1sOyH?{fg&Gd+sw#&S
zJcXRv8b**uKSwp6PQP3GcKH}}UMPpPd6NeYul(YdPzPsZw$FQ{cIx`zmgSDp3xZze
z2NlMvAse0q?Hkr?t<X*S?`$69S!Pun8Ips_28b>?6(-Qyk*J%{kidlQ>@ogIS<m+M
z%RX`uQ|5_k7ZMy#Lm#j6zv-EaVGoS5zs9Js_w4jPfUAKHHt$Fw(@0Xn^K56p!|>Iw
z=?6MRrTkRM<6W8?RaNkSc-wgC#X}2~UcPaO4ciB|{{!d?R0l30=bg@W3gn(M>%qt8
ziZ@0&M3<!Joti<u3X17>_R`KeK#v!lZ+_y(s4G-R!K^QTk1I%X>5!z@f?3C8?Yhw4
z(4y|0CyleERdsHi`dMnb&aXZi6q-ErLBazAoojvz%{=H*O*?Cc_LV9xw)8K4X%#a^
zKVE8mFez)%=5zegB(;zE>=%^w%W-Nx@y#O3-LJi7e`YvezB(9;i-?iG9JrxyCI^s^
zkt~<~c_ADBg!b;Y9%rI|g8FLQ?0>`S4Ej6%r4{$ph}`{~KP338;b3Jn>}m+O0ZpGD
zkfJlYn|)-OH`uQkXtjc%_#r&%2@nHu%<LVHr{*K2v5=K~x{fDK1S4a^QEzNVj%BlX
z@%|z85GdN`*qhJ&irOHGvo|TYbCh$<Eq4`nh(5EBU3feCHDLQaDS|~~3CdYhfd&vC
zlsSMZW&vNa%2wI#|0r;<x0_`gQNxB?$tuj_xm-_I6VWXsuB#nVBFrRkR^oRlfc=&Y
z!w{`$uq|3B$L7slMM4~KW1VO)gA?%ao9+hFrz5r2)cV9ePWUQH!_?H4<}S<4pi!bo
z+i<UMF(T;gG6%K&?{JNAHIXRdSDh(3Reffg#}n~7eLRNAd~SPzxVphR92rs@^96K0
zwX_2w7CcUES0Ss)nGjfZuH$MeEA^EHol0r|jojfb_Sbj}k_Tc=Pj0=#c+FnYIPZ9W
zvppkgp|Re2#r$l?B<}CkzqfX+KUDVIupYHsGp`k`8~%hLF)lEP)}Ei9ZP$Lzah89q
z1gtgR?bSZ1W2I3y=%RaSCi-^`Q=>oZ5|EcL_|(&$op7Uk{p|&l3%N5S34h(0R@PTj
zGgoq{#mptcm@AweEDq<XkD|pK0Pfk<)sfm0wt$g`o@cY9=O>^+V8-De)aV|Ewa7&d
zJlJTIVrkqqDO_80)Bu=3_b?oteFetEflAxT90BaHG)eZyoGrL_nts$tWfwJG^i>yz
ze!UI^tZ@D2@n@#ktNW-WiHDgjoPgPi!_Lr+_l#NYyQ$qjp5AIDGSo-r5{d!lv1mMH
zp-Ig=V4<dyPF4vr#0ZJMKwz?%{RRFdR(V!WFL3a^%ro5KlC8ri?MQ`ra_TtAm<H?4
zs&fmwx6p3%%H2mah{@__WeFxmX1B3a3QC@+_ueuW^zav<$wgb6IF!o&x(*OZ$V>mE
z6vfX9;<(U%d{nbb<Iri<Nc4dLpK%`8P`FR4qVOJ?{YDI1)Cp1D^=q(bE)}1>^d^9_
z<hwI5hr>hTW9jf`Be7^$m{t_oKldWHgBQtbIngy$S^X<Wz9Uh+ek{y7;jsS7BR%J5
z>OG`@G<sMa``zTOE2)%@Oz%f;`88Ga+H8D5u8h_pXZZ<^EYdexl=3dNS-t-E)5iDN
z@FfGC+iV-2ZsawyvmGN?_c;XNoo`&SPgBt-Y4&tTMpIziVJ2}OeGSq@II9A5aoRV%
zqw%!3vC{DX?M_2U%r|U!O{7ndFa~MR-<_rK;bpmhiW{*RVSBuR&&K})sQmup$btQ6
z!9&m%|Lz_A*z3uf(5tPIdi!+l!|ccHLHG9~gcg77@Ez4pym-)K(I(l6pLqiLT5~`;
zU)I_H+AmZdKNY$=zXz;ex}M+arjHZsz<=8R&<A2ezpvfT8A~(OUf>8@J-Innu$vDo
zfAfp9XHt@-vp=MTP2Y8~DE|*YawH~|Wc=$4{kWqIE19%}*X`lM)n%JLa-Cw(buQhq
z|H7S=5DK)vkhI0dhJV(Uj*6XaJSNwTa`%GlD#Pa!?soj7SzNgO2Ktce$Cn<j8I)f|
zv1fj?eEi~p4kGM7z{)@LH5O#Zrxj@8=huIS%Htm_F0~P_(7fl<o+2}^dTL9SW@mux
zgD1p;SXxaT-Mlb!e`^v7NOi<QYQJmm!s(;Ob>C_28gg6;e@#5#0LUL$zl;}D5JTRR
zbOT(G=&N5DisuVu%G%#i1on>V5hB+O-$bJCi5TizQf$f_Tt{7+^?YX&ymfk+%uJ6(
zpwbgc%f4$pazPpQkk@vYoJty6EHTr(Zy`>FG~1&V0gW*<{!T%>c(B|>Z4u2^%???Q
zaki@ueEE1VTtBm4ORD1VcXc!;za^|Ul6SUSl_82f^IHsCZ=lbn7SmMjnFTC)2OUEg
z5+Ph5Ew#!kTFSPwqxiNh)%kdZh~lVuNE3~bIpZwbHMKQyy{NW(dLxeAH^EKM_=E6Y
z2~ueFcv-dLAL?k$b)DNvHR3F!{i%QX%Fj|Fx!53%&&t&d6L~}DvXIRJm;J|bx5I+p
zW_6PYBW<FO1L)(G=1f0ua`3-15*59cLnUlFGu&3vS4NElcnJ+fS2|Uj;?GhaqyMlG
zO;mlzTD9E_xA={fdX^~W_9PQk4*5rqZDuod*6H$h0%-18qhlhyt-`E#@zCmF?fzyw
zzeTvcgVwMcVCOJS`X&kePZ{@s=Y^(9zp^>!+_*;6BcE-nf67-=_%G^B3%dE<y+m=O
zDh5WhJa7vxjCd($W}m#JU~q+G%xNjU+ug4*tgm<}wBisjzSSw(ZaDZc|6jl|kk-CU
z+gi-}7)=4l+e>tmKw0QL*Kt>lC#{ImSMCu%ej-5tKABVQ)Ofn+UAU_!%^RB<EDpZF
zSzO7?R1-C!J4N4>T_~0YO~}-%&^f7OKWyNx{M^G-Ym@XIR6MnIUeJ4fbJx*<*tTci
zr|>QY$({C~>ZEr6j<KRp?@jRoqZMX@LV&_JLH%R;85po<K)aYAy+M}2GkswG!S~xS
zsGuS(`-KCe$63EbN!_a(ft4d(TjFW~H#pQTc21lEN18>$?iY$bcVx1%1CMQl+oZxT
z(iG>pizTwGlj3Ly!HgrA3-sii$f;o{bW+UN@RWTHsAs}Vvj2o5N;fgPHAX~zpC@JU
z+Q#`kb$!Ja+jG?wTG=!#bN|35lD-+^Z$3$_@n@h>vLqhpXvyG*Rwb}gBUM-jbrsM2
z`O@qX;*g?5Z5gu=OiGK(t{ki-X~lFIK`jF=l@UxptLsoK*F4de91khmI3jtK`OL@b
z+F$6_iPKn~Pi_LjDEc2d;y_m~MK^|4C|R5K&s^)SWxPgczswnq>lU^dD^{DtW;2}*
z1)6AJrhX_}$@VuX^w6FG<xuOF#NGsK)+ZbT;zf5v^1ipS<bK$#?`_RqZI>Oc*zaI8
zyThV=oNbd2mi`Wdz7!vhR;pNTYU<3|gT)+*q2|6pGwc(wm_dF2pu404S1iF$q4?w1
z6<d0~=zYr<2?T(b%51RI-AZEtOlF=v(#9xHdmN1Qmc8XuhP1re!D2g}ZOdkFWg@`}
zT8kn&Oq@|S<3&D|n-Bh(V$b|D1!SsEG%7<T3`0{HHnU>`fZ>byJndS#TPxY+V0J`4
z^%0200WwP5wqcmSPUp<W^Wlta7$?Q3p@zv)KAj4nc&(a{Ax41k_Dbdy9pCqH=^zKX
zJKN^pF`l9a6yZvG8cv|h0Z8yT2E{u~c4~#geQ}&huD_xU#z>jPTyFSqDy{G~P#Z@t
z)duRMm`-Ku%@t(#_>rHQzP#CKA$K3Oz?L^WZZ7+}p}Wam?5>|I`6Y?-*LJfcXo1^O
zLf1KVC+^G3LI-UA#j3aYshG*$jLrLuG_=c9<Skf>8#RJ8Se`3{XsRFc-n1x@@0p^v
z>5(j9pzp>lzqT1Gu|~w8`(WHxf6}#rFaTmtnZuT)c2)o;4VNdEW+(*PdidNn{or-g
zC@<hj5EHQaJ@7Gl3zWG1hTR8rYO4LMxT2#mYPY|pfLU23?Ib?VNg)>#t*BwOA0z`-
zT^%yc8Ea%!xfdPAXyDe}qn{CJ7SkvHadwp1WP;Jl!@U?TwQ-BV#ZM&L>fHcQwL#hP
zB;3AZJfeo_+SP)2eaJZCBHGU0V4_-qT2&N}bxhdCnec&L@J$-Q8HW^i6om$~`$04L
z^9kv!-{Elj1Va{kEcS=X?tHQMjXbUo?{PEyZw%5`L<PLoIy_s@rGwAQ72^u&mEhWS
zrrMPiqy>s4+{gVJVe{zO8z4=iKg}<Z2Gj7Zm=idY2<#qMPuf@ww}bTTq~eD__`7aK
zVs*bcl;cyib%L~N-r^s3KTx`7csXBDo6Yi%W4hn)vO%Gsq0$@hwy0#Yvim!xNXfBB
z>fDnuH^1@xM!p;epYz$5uprP7J?S&@C(mxJ0f?lPJ~1XiX8S!uNE!h_Tt4_Ea$`~I
z^0C>dSQSx6e`@Qt#G<sn8k%oQ*OnPx+P#?*$lVS?Yeb%-J`+cm5&I6VQB=u3t$=TF
z1$|G&GXzbKVTx1si217bfnL8Z{7bZttC;=1r}<SWws-H6<Hu(yLVw@pe|Lj?E!t|b
zK6U0-vbtjuKyi1Ze#HY5c+XLFCh`9P0Fp+Cw`wQw(l_&44Q#j0WLxP^Fg*?UwfNtW
zpIXKOqvAMw(cw;lVA0=klHb@K{$;zUr|L*X1YdQ~(F;72{Y5+<OXuo!1FN-iXhP=9
z6lXloS3%daWD~YCIa5gmx60!9x63x7ce52IAGiUEKBiR|>3O%vy>tbrRTgP2(=W5>
z?d7~+R1cC2p#Pl$ovgV3j*V_0`ai(_RIs@=KGQ*$Ly*NFv&<2;9H7yiZ^NcOx}_~8
z5sP->oUZ7Y%5aFgtAwj!`53!HdjD~VmUt>An)-L_KbXxS@hpce6z8wh;2)25NYhB7
zTg{M(=-q9{F(Es!(LTXMtY=)VGTQ>$|F8@hK5i}aLu)M?NEWJ$gm%*;P=;sZ{=0`n
zwctK%Bt=0g6P1|<DLNvHaU22{qHdn+xo6&O%xn8Z2}LbW-&bs7)@rQRBEzR4XH*`V
zRc-9RyFD1rL7*}rHNFXPht4;cU%fw{EGm$^)+xR3@oRaJF*DSVUsGw-%zQW3Eb+^Q
zjlpY8@<V&`tqaEM^|hMe6o&-#SKS#puM>=*a236fp+2~{K@O2g{Sg3|gm$D}WWWJ}
zqt~)K2CqmV?YOW3Ht55BIJ|*-AcoDRjLMxn-AkX1W)Hbq@Ne#-bO&buz>ARasu0TV
zwF_shh41~HeryE*>TgFw`|m-i0l*94mfb(lfbYeJpjEN#k~2l%fVM#|f!&dvqxNB?
z5Q3&f@1(U|G~3Ska5g*(Q)(>BKhT=3`ik7WuE)>{_x}J9Q@{-?ESPdRSK-7~5b(>_
zU;xd2xgs^5Z|r_Jfwujks*-`{%3bJ7WCTSJXk%Len%=N{54z5n@>j2TRRj?z{k8Uy
zx_I$VKd>0M*7~;G{nmccMB+Q|OcV|r|9P>=m>FBbUYvcHu1%Z7@eN%K#l7$Rs;omY
zoJnmqXAC{kyjcH9sKMLJ(#&WDFW}v%Oik%AAvfEYMPst&+0=$CiK%|g$w7lpfsW0j
zj$7dXcw=}BhmnlMbJulg+6o%}($M>MVV_dC)0dQ43HECgK|x)lZ>(XnirHB1X`Wcc
z5mvRe4!R%}?V7EF@=>*94-u*a)M?i;gu!V9W2y?}<<6>u^sZdVH_U7hC2So8^w#4a
z3ghs@XaaA(%OO{nHBn#@BKKRQ*-j@yQJu-GnRk8MsprxK=X??lk%YWCq06##y}4tf
zlYNH!*#(2fAG9a2UJ}b}MGjN?+i$4&VL*uGwjn6qK_ARg>7#+O`@I$~GAs_h+A(zH
zclSQhko3iA$Wp74^Sjkd>E4b#2`yjt;fapRZNiY7K+Vijw$h%nDcdNe-OQomSvN@3
zW9B$Wzl!ggqiXG<eyPGDA5n|+Yx}SCU!y#E(0hd^nFgQjcsi!)lm|C2Dku;#!Sc^P
z<UXldBp&3Qtv!ED8W>S$44ZSw4Rzt(=DwvRSqR-$!*Y#1ZU1&tiF@1OL9XS$N5Q&|
z?VPB~$3gVJ1I{P9&iCaO!m2)b(mueNy!JY)XI0Y2GjDzbU4Wi{JtP0Z4U*A9LW@dH
zotY0<$gO)B0*eMljt`^$13U-Ur{G`RvNy5`^g9HdG8rGg{i-C`b<;}Ql5g+rww3B5
zYI(B$ZI>w|rSh=h3vgh8bM~`<E~-0)!oy{S9jxoT$dlJZlY=+_0<00sWyp4CsT$A}
z6C*Ui1kNL>SbyEjpDH5d3C?70mQ$x0=FD}FB6=h7ED}d53ru|)1rN6=44g@N4s^+)
z3B=JY`C>K=BROBX0N{&*$<LOFTA+CKkZ5RmG{eTxxEedHYdNOSsmnXcPj|d+2lzSb
zU!M>bP2*7wv2GT#Y*k=7r%(0KYUVMLkDegiieBv`ki^G|1~>3p^{(f4ePz^p_LMY_
ziDR;!=o(t^lFpOfU#U3l-?w2&-(9Q@;@c(^K~|@8X%!|}=sUGQ#0127UQl^L6BkAA
z4k>Qq3@s(uX&RkVt_d9n=Ift5O4WOI3G+pqQ7x8@&jMcZObuCB^@@l&B9t~Nt`Q++
z;dTQ(H(V!ZkogU1TU3O+xi8MvG`D=9Dam`9^J*uV;TCSd8tGaze+fLG6nP;5R}=6x
z0A2Eju(sIqRI}{QOgiraF4%=^OsW)_Pj7}tT1%pnFaIlA%I85dlsrk-gkM|^u?6rL
zGI4@wcH@F99Dpu=N(B2_HUepq=bnCVVJkSd?F?+t{>@;kG!E5Aj7h(D96Nb-U{DC@
z_?tj0l^155WMmYcmWdUQoGJAxUov74Y4IJ@U^oKx!o?L8oj^ZcmXL*d-Ss%ZuSJE#
z(H@>YK@qh=u7{v-sn*nalO@oXwnDN_n_=Y#ne9A2$1Lzq8m*3eJVLU3_^)UUzulrc
zy^oKPH$M(yhFx#|fF*Py)Ww3N;t74XYBOAeVKAkzW`^9}z-mLoi!_8}hr8FDauKp1
zOB>-Ca7=U7ID<vuo&NzUtdahgP^N#Jy3P(H3!NTW(OTXWKEJ9g2elC{f7Jm|9uA8^
zLVpXNmI^8f|9cELXEX5vo{PNnSC$e`+xo^<qEao{Z=iH(;qaonqT%1ff)9fR@lLhy
z;m(}DiH)Oyl75nf@`ElhsPomc292Os%T-TPV~ZmZr;h(hCEp<5IaDwG%b`}N=J@?n
z@)%hASJDT53zU3=c$XIDlq31Z>B&@(Hp28ql|ohO4t-gutr*+Bx>#LBRO6KC<XGmc
z<$_;T(3iM<sm^?Z$o4Jc9zEx<;8A|k9r{dPXD20jk85#qQW1@CPUZ68z*uI2522QY
ztT=9xDPm|ZOu6Q~mTSC87{>*%QCnf=YXLKPVQ5t=c=d_d=!O1<{N>ekZCFEg<p-<4
z{2ACy_Why>Wz#1AkE6zI$Dm&IP{ZMPPzd<7<!P35!q&F!C1zoY8s9sGZ#hH7r@?Cp
z94xUKcBhZxwLtwv;K(BY^4$}8^AIplj;HJTMvT*=tPdgPX12rLT(SW|9?lX^6iOm_
zuu?;THC#>aFxz~Z*XTI4Wy<+9E-S#6vk{aDeI*6$S{c2c@Tk=-Q0f)2ztogNP^x%b
z!3!Tr5<RGuex-PP7tWwCyIGp~7z+^n30g@l9-MrFZTI;PAc=U;x01hL?=u(C$;8;d
z?`Bm)?cs%f&6r{<H*BzE!2|93YxB6N)ir)}sqttjBx}<Wm<Rp|++s1vu6M>|m3|fw
zw(x~`r?ihn9ZUEs_i;){ZbL%jIu@?U8RG)GJ}%w5|7Jd4cK&HkfXNZjz>P^7R7?i4
z()X#2^w9j)kPA<mAtWQ`4_A1CX(KLp7vw)c6iRK=<mHUecRRs0+Bc*NXN1-b;|J*u
zcVv$E?+L0)#VqvloX)C7rS;h^T}Z!_KTmWogi7@&d4S19o`I%=U_9!!8o%l?%Q;oB
zPd|7V<LuL;V_Si7!lo&!&juU3lc~x3{HW6!e`@s<qf*{*c|TD%t~1m{y>N!UN#gCF
zS^FlneW$(ev!x@+yEgVhXWOq9Th1SsU$pNP3A*{LO&neOWfDsQIW4r=QF>r^@;cR>
zFX>eHeh|C$*e&K3plJU_#=9#*25vXX?RRaMX*MfeoTt1#C;G(}T|A!_l-pbl%x-6o
z0fplNY{~^CHp@SWv2~d6)gf~&6t->fQfmnW2JNSU*`02yJi-dl;%UOLd}q_E^<~=X
zhHt2hNs>8y3+Brn<e7lr$%hloKYY`M&U<7Rq+Zn+`EG}&nPm+>Ov;pG9ukN2*f|S`
zo*QdKPmSNm=l{>BnCpcqk7zP>9PorV`D7l?D=kNe$$b2yl5==~UTyzS{<(SAXv;PI
zYk^)rm)4qyu2q`qG^2|f#Kx4@lmkhazFAKWitO>&Zd?C0t6C>hmSDAK$1rWo{nL%R
zl5zo2O??9D#o&Q%sYWIu>YbqvZzvNK<O|H<+G41n9z7>uW8L;csL{evv+H3LgxYrd
zWmV;WfW6YtGZ=_)z5ATaZnvIls$kQ_oV@v1qLk|-U^D-Mj91-K!U>~G#TXA2a_Z2<
zlv@s~&ist2S@NlD24x5zhs8KN8^p6TY|=YZ=d{L>+*+5+4dR`}t;{+TASY&*3m$(2
z^*FRhXFPbWBrNL@J8k#IrYfufxYl|{X61~nTbD<s?n91;qp1I3v-Tj}!XJ^s+pfp3
zOI7fw&Y(k>%wktn)AE>csn6E{fx0Q`me5gG@apU~J?9)pL5cdnN}*IrFYaOCyW{}(
zsabpWw6+EtyMUDDvg3NU*Gp~9s3tEk3-n~tYO1x<H`u7!@wq}N61$=Ybn$T;Q>1S6
zqMJHvjC=UxF@kBZ()4it@n^P7Gtlb~i*GGVk|ceOYf5mdD*qac(EFCQ@jnBzXGJkD
zBN=ROJM^_+X3LXJ=S%i2k5!I|A57*4CFaz(cegzbO26uI{-yhaOi9F|uD7|~Y<)tK
zZo6p(yTGAorR!|hcKa#sn@%U)1F0CXrXF6A9GN~x3v=~>=Y2}kHnOnE>mN8U<7Vwf
zi$sBp_xzOa{9|bkeZ^DsjB7fTJt8Hz*={g8ItvX9)Z7AnbCZhXr<Vu-T!SE0<~(X_
zLhu!ZV!!sqF}@y%lJH0i+FpxEZ|W6~xgkw6b=EJCmXSXvFc*$q6+7ENK24dWj;!v(
zMmqf(5=Wvjqe^rgzCjJl`9|}A5ZG-rPA3(8t(zV+rE^QZ-gHzH&$^1LmTJsHvuD*_
z7&vgyn>e2x4U)+!>6uX4gEP~Zh)<J4Fz>=>Fvdm2!Wv#Wqb>e~dh|rg_<Q|bH;pg3
z{IU_=%Zv34(yzrEgj`q?V;iXtpKB4?v~sXzcPTD-ujfuInd=FJK=yM`e86tzIf_Vo
zU()USq-g6M=moV`#i+E76@6^|C<xGGDJF8Kt`i_7-Vo?*Hs~n<B=#T1%ADM8NZNLj
zbpX8QKZ3QlLGP`eG={Dutvu525=rvijnqfS%_|<)dt{<t+;l@fXRfN6odOBHfk20H
z8(C5QV<pb$WqRC~qcs1*jApRz$OJJc{QF4`==X?z=yiYdmXgO1xc)B5{B}*g$I)|9
z<U*UUk24Qzi~G)Pulwy5MFBLW=KUI@%;}IXQbjbh58#P54B<N(fV^@sX6D8o|1NG}
zr2f6$k0_FR&90M|YiM~+`0333%_C@<ufE^a+9a?(2hfYXZUqlVr=+PQBL_D^EwcwR
z!i+ZNl6yRLXWQy?uBc$Hl(*00m*H;V_P9tnr2J5Gdy&}O@1elQ<h~9$ii!G96Nb#C
z&H=*P+k;2V+k1ZcLlFJ$hF`rBh2&D@F9&6WLi%GUt3ZuvLOw2EVI|jR-CM}H;C5lt
z4!_L-a{H7JhYh7*6|~ai@iZgfEpZO5t8<fEHq0R|I3LcKwPKFf!Zosbj(}Y|bLl>G
z6=YmqOB}5pK059+Dgpe4#zr(q$IG#4<WlHpwB}gMg`6Y*d=U+ni#g)T&XZSZHL~iM
z?|y&%sJJD+DUXKlpBdqHYA!sZRE~Z=@gtcXo5<X$8=xnK1nQN47{;U&WWVfG@%E)2
z?bGBQK{Khg(%C}H&B9yqc#0T1bq!A^R-yEkekDOXEvbig*zP23_T1Q;*X5;mR7Qk|
zza)&?efE}Dpw>uL&}v$uq}3C}J65Gxdk~;PRbwl89T)Jhh~G{J`HWEI!A3|my;@K#
z?-zLCJU!f_N=_mmeMUm0OuIpGyCdNom)JSYsmyT1i{(_bHJ!>|%uwr|e44*VbVpjs
z53HxO>62~agF8<BrtC~_S(z+W$BJ`dDy-67q+*3QOX@jGlBlFH_yx$d{YFbpMfWUq
z&w|^H7t$STq7E!udmICAe`vT<%cM=Q*D&pUcIB@c=WLA)axqP<T?|S&Bn#RaEw+dG
zSmw2qRTTWJt0UQwFyK~uEA-3^mC%h*SU^%Mp22mI%Pzi&bIh1~AM&Z$i0cWR_EugU
z_Es1fAKPjy+tsm(#Fl~9$Nol(6?mD70@p|tc}A-yaIM3?^?7`EV|R<Rm#v&gX?&UE
zPisR?uzDxER}cD%s%`)E0m5}mK9A_g%j)V&FRMH44TU!q>Pfr`&GPDIcaO+LW=H(v
zF`$K7)Db9}yrbCNqEJ>*P4G8bMs~NDzi0jT*>!Pw?Dp6bq#frk!<95~-MAmKp4pw-
zdGJgVc9(0!>r!39uvBTy)bf(r%RZ8WV6(HEEBvEx<z3W4Hz_;dPq#Bq_*E@AtX6|(
z4{J*^53Ao*r8=Usq#KOFm<#@(Vn$(+7jn$}wS>vDau1?5rapw4hApS)MerL(FB*qo
z5Kon~Cf$5^(~XZ{mFap<vDxU+JKYCVg6{&pb|zj8ob1yK)ny0de+65mRGP5J{hTLw
zPO8oqr?04`mV+J)zL6+kaZ#<%LHTNQjh8vFvok*TM0^3nI4JYV?AQ68K#NDY%l@Lq
z>@g-PgMUMdU>Z&LAfWKf>`qaPM%f&6)HAC@{q!57w|PpGEQ3p(adla6$Xt%HLVBTr
zn|rr$gHPke{JVFRZ#(`e$yiR9(HLzlXp1;odWqbkjUV~}x%m6_8auIQc25O*P#O)7
zF{y{Nz<SfK#OrS109*>$M;Gh=1Hf<SQavM|$MCSg?u|3JRIJpYoReE2s_${r$%Fx$
zIT`9<ZI9I5df{udy%W|K_-La>)wcm!bz<Jc>hdAAz{0r+zXzl{CTox;uWqT$a*?uw
z`Cc22Sg>UY_NWD)3H0b(sb8s}r0Y5jZhK6I-);Q^W^prz4!IQkymqhd3k)IeE|d>Z
z!SpO+-&?7PFgL@i6Y_wROA|-3g3R-oNN%4#1uP(>jx)wV-Sh@#tOYwu<A0%WSu{r2
zQ~A1qt6nSQJ>%u2PBFHrOqVOjFt@}T1q->4upzZi5Axk-7RF+-{MH)bYT<6&LjW3o
zS!mKqNM_)PIJn<0o`~@C`Hrl6iyO21LUBDv6=Lu&>tlkS?apW$9v#+5e4uRz?&FNX
z&ag-%5dco7SpR}9aT@AX8<du`7!R<}L*5HFEbi7f>aPplVpGbMNem@K+GB~<Ac%a0
z2DLb4m|9(SzfPR2W*#ETQAY13V4{|ZY98|)r>!)b%QLmmu2oylq>9LBkFFJ>@>lS0
z0YPFkk`4O<D}d{lf%W;^?DfTbp#cbm{)xVlxfc5YA%r1<2nmB&E>A(;2p=&O<W3!d
zl9hZlEwIJxnd=0=z8=5uLjU1-q~tBS9@WS_55LN4Zt#jj-PmyG3cQZR3?@K%r)Avn
zU0Pph4r;F<8-<Fjf03+Dwjzk!6WkloMHa!t8&JG-PDXZ$N(Bxr(E8fK8;5(TzJXs8
zN89GRi~a?|y4PoE&p?gujN)ELGNs`bl~$Q}aU@lK#)Z%|yPD8I7IYmrW1m)Vs2kA`
zX$jKh;AvMBo4cjtj@xLMduzx-T2Y1$N#h-(6UzrBJZnoB<p<Te_2D&GEDkSnJ=RSA
zP@E#ua=Eyox<c+I!jX3C7%Wb2L!6?VGE}lod)u?ZDmrHPhyIcD%@m*K%pyX(|1}sO
z;CEys2Mu(ZEqg?74jQB*IjaOuQlQ1`I+-_V0z9(Wzt2woB3b7%IkS!r(9(LyF9O7e
zF%^YM%aE^1kxFmQ-a}vd(16h5sltz`#qZTQXIpf6Vta-$fjhiFDp^_!FMjRy5Cmb_
zIHKd!V2S@`3PZd8Jll!-zUismi1N1X@d}m?K3tcp>kA-LBKU_O`LnYs)a*m3qbyqP
zs8*OLz9V8*`4?65yZ*1or%y0W9_0dbZ|UE|*an|7|GGTc4P()gS1-}E(e^(H9`*Or
z@04CX8FzjW88BgosVp2^%Tx2EsgclqG?+=4M4<892bV-Mn*@R6<{k9wkzIsWPS!%#
z4P|4-Nw*;~(YkF&Y0Ok#x%}y#!KlzS;)QXtfYPWCQ!zQ8cE~x0@~+;6IbI95B$)9N
z@6TG~+lnx~n|V=o#E-_k-ayVvF+vt-3=!k$jl<gHsT8Z@dR+c4%38ctakReVX$Uyo
zf3j3X!%i<}8_twM(Y^Hq!~#jMkp;2XSJo;0p?bnO!8>Sn4*~%UiAO`#cTxy?cVUy5
zKmMcEL&XYtfMzx7m-5Vpp`YrG0v|{E8~HBnN$*s$9lC4tf;~*@D`}dy_GR{?N~U;w
zfPbCqoM1%vWE~d3-m_j=U-oSa%5FrFrQzdPD<Br+T4@MQ8wBX4*!pYe{KMSnkJL=!
zVQW>MDBALfR9QN17P9Bh%tD~;L8Dy!Qe#bVm<qOPfiHu4j%{nSK=w8yuwXWO`>}RE
ztdHYeLy1}RYqIPOtczDlR;*&&5AUwx_ZRQ{fAk{-3+3XUIMH<H92c%+9EW8AoN00F
zO7lEzg0G+1<Dzi((}70wknUnCv;YAWB~|J06k*bEpBf)iS*wOml1U^(i@kVFMJHM%
zW1leQq$^_1K=w;QHdhW?jgJ!)IW@~78fM&m^c-rx*IS8z{I1uLbUua+Jc|*!NbU^M
z;;$^HHAI4sB^_*{YDdj&f1YyqXLt7LXKr;9v$`AdvI9*t4S|hQmq8V;#otQNSb)sm
z8Lj+_pgx8u<h%RcZ9UZ%#ckpmcJxwV>VflGDZ$uqI2ZFuJb#Xuh)w?cuKNyv=xsPq
zuY%~HwgZ_!vkc##l2(V{^uZZvTZTeH;@jzQQ8jWWv^@>4$vgeo9i!s&RmEqusVr}O
z4n(Hg?tJu-1?AebNe>%iG`DwAaH<wFjxK-pC*11`Z6OgHF*QYRAT}Fz$#9EcV(KbW
zAd>s=_c^o@D`Fz1y%XcOMTZ~UI3RfB1WiFKZXCX{p>l5s%MyyqBsy63&$w3yjbvqd
zW!C3=d}`uI7G77LG%FBlx@$L_MtPm|cO2gCx-eODK`gL=9_FtGm96s@rDF;V29*q+
znHr>{EOSbv<K7FOOIuyyn?-qm@+G<~cddf?W)bh8kQPMD>}719K(^N_SxMJI;xlio
zi~JeE+Igrjb+asugWxBAsgTT&fXK0;p3mAF2Qms|^olIf<hX2vJI6yai5;h|rTCvs
z<gc|Au*3&nN~-r+!_|!%CBKO{&$IjoD7uUB4lTK`>l<O!EZ+sz1AtflJ}iobm!fie
z<V*k69+lQby4iBMpVGF96)NO5h&h5~{BA+NSK#*`CBxW6OSv1V$(J0<y273~*K-^u
z|Cnn?R48EFNc&BWn+BiDI#8imP%V4ub6GpS(SPgL+Rf}_O1>BjqOXiS$fQ11&Skq$
z(S$Y(KQk{%QGD3>>IGWZUEEeTry$|&A41B*KZFiHIg5je%|||xG?J)kaDd`K0uSD3
z+i{8RCV|<<dr``;nWdUPUscZ4wJ1rb&WlG~)uqr}>6)1W;(77vZCHQ}@#hxIF6sv-
z#~nTaRJQR`9(RE{DgTZPm6>#Q)Ej=&EKemtdO{J@B}9o?sA-&s2TH|U3$Sv)Sh3MK
za0vu*%zzj&^?xLtdpwi>|Nn;^bG8W~r?80}Msv!Une%22Lt&0{$S8$(Mb2}avpM86
zwmC!&Q7Or}CdVaIaxA2TN{9FRdwqVtKlYD5cDuQ5yI$AxdOjb|`+c=!MC|VYi$e5C
z<Cw#Ky!b_uo<HrIFtbNf?YJ`Ds%q#4vCFK5;82^}go||r6(VR{4p6^wdSym0WB)0%
zXQ`U^QAMUmvUI>B#;|DrToewmkiv6!!-u$08sdst@pC~24HlMY<#@i7As|sBtWuIN
zrNC4Jg`Z@?3lf|;Vzruh{)Y1Ku{kV;f@=IL1yw<BDcGGtQ*pU6`i{G;T}C_cF0O*0
z?C0~w7JNM7S`8~6!50<OtMp$VMA@9H$6x%k7IZRigz>lZ#&T;CF5lAA{}{jiYtFL3
zR%uF}XES*OQR6R`Th!Yqw{84M=Pm(s1*P+ImhDz!i^oA!&D;xLo6zt?b1n4h<JWzx
zEN9YFNC49%Ko$md0_Pi(TH=~DF+pL`i`7t%eP>VWjb7b}_o6|!Tdz7VkI*gdwIUy%
z*JTBSeKNfQ>3SPh(6jk<p_;EZo%C#vU<mK4F$4OkT(0F0Sq@79N3Fiun=ZG_Tt0sx
z&AoJfG7qyB!Z{S>Z|2Fsw4dU{PMfdopiU778f~Lb`x43Sbv-U~ZaT%x4`5FuIF`<m
zQ&SFY-<LRSN@+}4px$778xbN%lr%alTU5g&PHjQz;w8K3_adgSho3x;)N_qtk4H)2
z$QO@|!!A&-IN_mt@+;U-CqU~l!uiEr{grG{_4C|~d>z|tsburdk3?b5q}@c9r^Dl7
zu1<0q=oFoxd>0fR#ZWAsJsYnvkNsZ%+N796>|)|((bc^W3BE}EA2nv597iS3q-ycb
zZRS@$hYx`g4s5#N%7~cP$C%P@45cW9WFzXL6AHxiE7gr3ScTaU@6TSP<v6FjI+J4<
zCHV`@miiw^%G?&5m=`6mW1$z5ErJXPmhIvfe2=m?B@HMD4l2<SeXB)8q~F}FGsOu0
zuK=eVs0@kk@<m5Tp`&e_>;|`a#-1%eN5Y!pl+Fx=Ro@oV&~gw-FRF~x^@j&-J6Kq3
zsCX<~RW21Y$(U+OE*7zU`bdguct80-imD7X^Jsj{Z_q)LT~B^}#E6zR?TCE&Q(u<~
zjyq~O?uKXE*+L8_>NcV?{2^kNoToU!!P{%%4&LRzM=II&;J=+=iEn1f_ShO1(D-l=
z+zd`aKs?`5LNf}uUc|W9Es-)K;daIBwZ)?Z3s~PP3+&l3BT8?<h$PP%1>nqYXP#f|
zF$GpoT!Bm9TQXk?ev~Rtexv~_+5^?{mY<4_j$>pj<>B6jJE13Bx4W}*R&9GKxzuzl
z(>a+&>dI(oFlKNrM{vm@poXW~Xx6B5zA;cqH4&Ww`s}zKowdPce`A!2_VZWH=0m;~
za8;9v9R<0i*pOb(#-BCd?JeyIyf7|wVXHfNA7^Qw(XK#WUQxdyrW{y0^jK=XtH{?a
zBQ13FJbRPqY;OxRKBq!gJx4*DrbL5V@`v^{prt%qo<h#@$yN%6Y8B8{Y42~&I-UXe
zDXGxYl)CXrwU&XBi(%&PlQ8mi_*qg2p&(0dYqK?s=zWwU7h>gx?t4fPqP?pSal}!L
zdXL>w5Ur@6<C_g99VoRQbeu#;AzkSwnm8E}{yAMdK1Jb7O^NBb2tfk*oY|FEa5#I_
zu>Yxa$VnSggPMqDV-l17O#5m@`;WRoe!eiaPpF}6xy}4mH4qy~0)WbGC-8N0Nbj3w
zVJw+v>{*aMoXaB<rlVY4^=in;x`$dpiE1MG3&0!9I+O#z<5q51uhE$o7RyqYKrT^>
z{ERv(Q3Loh?trhs2mF4`2!^oTR)9_3&Fdne7ewAHNn2hlmNU_fyBE`OO1^nou(%Lz
zxkj}*kFp~glbTvR)9CpR)X$?N_B~5n9wNB<vhMTe9JP^7cS=<h^VEjXUcR15|Dxvm
zEv%$xb84Ye0C%Z)@`tXBoxRy&9hoNS7`wvqZY|aA;%`tempFLWk|p<ICheJmUgt9e
zxNhN>>w$KGkle4J(%rso=z-vK*(DCD7A9;S>LWT`%f9c9cT_vo*AjaU=9(1h5s^i@
z1k2<nrJ*$m0f=(<6yJFd?pJO-%3c)KUrry3VH<SWDxqtXBtb5|LCCpkfrmya=E5F<
zqSY*eM%Je0%hI=OcM*G@m?V<b#=a-zRO@R<$YD!hU){pD`yfnZG!!q9SFN^yuxP>&
z!}qfRxUux?&FHuqfBxY1VaZ>7Tq+W4-O`bVHRfjs;K~Dg^YyK6jVJ?*aQPRM37@||
z$@Eq@U!85z*fZLO$yrcP$KMZQ&5NOb-$;D1jnu;tnWoQW9Go*Uw<JC}oEy3}jO62v
zJUki+Ve4HU1#6B<HfZaZ9(_S;-Q-l2(KeXUN|;I7F&$Oi<Nko=7kdRxln|^x#<0~`
zcHE8ZI+6@0fj<6$Y3ctu@cAd1vPs|9@%TqxLbACdiB#mG0hPcg-BVrdY<nnGU8|lw
zZn(*K_-Ec8In(~TUbmBHwR0P{(A5tmBwDx_EAvdSY<`_~S2p|N$exh%s-}r;$BlC*
zIKZC6&9W`N(6jxNzNGGn945#>k{i`?f8`oIKD#)Osk>yX+AboDJY*F7-KJ%d3pg*_
z0CH4iK_jq3cdq+I*QCz1Y*uYiD)_@t4zAUX0`P9h(&(Z28<#hS!=|dKA14VDEbR*!
zpV#`?Se+eBnjhqvcsbPs*M{nf?2SV`{&_xBS9-(vW4pHw#hWjn!y3v99>d%5;WZiN
za<O1?yZW$m&dFQzw={|i*@<S2@(r%q>heV!sQReh<|OR}W{KOSV#m<R<U$`U?Lezd
zr}&xfxAxX|v>@7jwTWMcW|MR|dFD?P@h34992C;T8%r8Mt(0mVihYBJ%SaetFRKOr
z55DF5;)PU*WLqbDNNR}Lk5FVte)eqa${g6^!5Q7KrK^xv9y46o*g&<ishd_RsBr$B
z(wzsb&WpB3@M#yz7vkLTuQkqu{I5_#{-`rFv6spY$Oa_<gItXDGkf#!#`}Drq&1pS
z)(oAuco7!)QhJp!l(d?EuvmT$&zE->MdmqO>7Tz>V}T@Cuo?$6Zmbq7$#+shIpD2L
z%OK^dYvC&1O}Ny0ZibddS4ofwRCbkSSO3(0^I^S@gh8M_JYB|vukBY1N8)N1)<xq)
z1pYxwCCRC?=JzCmf$mfHumD@!)vi6qh^nD9$e*n4U4t+G&@~q`PRCii2P_teh*$ky
zGsh2c<>9)03ugb)T*vh9>F-^uTj!wk%LFiaAGvDJcxCQlP2yXIh_2tyid*o<SZI!N
zb)Kt@eIB~mw?vgx-wvxwN_7|$2MrWNacb6{$R_%nQXXk5Xu28B#*CR?o-q{vaH%-Z
zt-XWxwX3(uPe7r0B9y;2`i2oJRA0F#sVGZ)wac%zSSItI_5_y=A(WfMtZ@Ac*&>Q0
ze+H~x`7}VWDQW%Lyf1?Op<yT%3BE$_<t~F>nBu^p?!nW!at6QaXk1pI1f1YthDPvz
zp*L{|!d4N(K_Q6VVYl+$dRlgKXCZdqM{8lkEBjV4&f*SXnqREdCnE|L;L7_Iri;jT
zmC|u!BV)9Q!!Zx&bw;!Ya#Ia>!;VYS1Lx(|$##V!s%&<_vw*wlc#0h@{k?`mQkrZ!
zx?##?bMY^|A6QYWLV$yx#AC$3vP0;*t#CU?G*_tU3!zW;1%96w>q-;jrF?@X-fOM4
zE!>>7jl^hNP}HYMuhI+5;}4b=E4if)(3&Gc6E@%)*(XIjdMm$YJ!Ibk;-ZzUKl|38
z%?5lrN)wn~vyC_T0<0!Z05K<sr~zfSK{aiDqS*t+FTpI8WQztq8fm`&6cF5hZFl|!
zW>q(R&_IXjmBSFRte{IWQvG?QI`<sS=-d2Eaw4N`wc@Olo=TFJ=3&dPk{;k$fNST=
zNtM$+B4=56SK#3v=44leyl2+u+580wQxJYmf74IchGkrSiuXNK6)&94Kn%TQ9LkUH
z;dISkpSQICU-6Xkcdy&BKoPre86qmBey*I(;mU6v3F~4G!*vYe?`V~bweoCV$}_?i
z>$e0%i1Q0rM#HQZ-mYJ(XTtebGxJ)F162A21hiJoVXg&*>e^;!p%oCt@SHZHbN?nM
z+#-PMu)1hAdZZ;~0EufcjcatFN5;FQ_^2ss6YcEDJdN5jcr{daRrF)2sA-ayq*(2V
zd0rdbj<u3z+b1qH|3wW>O}MBm%<NRRGJD28${nI8VP(EbMz*ciM1}HaCNVXW?YULy
zt8~28@bB&lnsiXHNd|q_SO1vFAzS%6bka#!A3_yoBLu*q1C1qgN#oveNC!cU($ONo
z!<cZ_qW!uTz2rRWZ|${F0Z|wM$FOMAY{O?2rs_)Q{5V{Vcx4w)EOY=Xf0Cqy^*5(j
zfvjJ+Xk(BXmg6qkfCQR<(U4@G+|4e5nRB(*vM<pjFJml(P?kwsx~GE**^S>mnk~Ac
z$tibkaG@0;Q8rG?pGxP~hj^|Cj48R+1bRmNje}U*A(nQ$%<`u%*|9xH)WV*fEuH0h
zkey8xHQ7BJy)shqUE<M`i^X*T3#aVQ|6u#WsCs*^@NStj#@-D|!w*O{l{H{r&_Rc<
zBpD)-D{g7o7rc+gAUZ$I4XaX6MjhyyXZQ50GP-#;+Z5S6?~j&)q_K9HgX<+{s(AUl
zz?c42TJNQKdGg(|m|g64-3xx1pX|MsKT`NR!pZrC(?VWlu8*I0gsRh1R9m2@>ADbM
zMtDQPajoRW+=nSmK&~rLHIz^vE;5HMX4Pc5k;r=3q~<%$UZjc0-7fFUZw9NaYWu8?
zpp>MK=N(sc?~;Hx9>>Lvhx7cE`(C4XsMz3wZtY1@IYe}(cUa@Yj^mnedt)i&TtKv@
z0t30?oH9&uB&?%d$06blo9MW@8s7kL^_6~*E>*VsM6OVLgL1I<1Lm-{o*auc_3|nI
zV?n~=3y&@w{_$d(C!(RGHs)gN^1Re*)WUcU*--#nnultzI4!x^RVR%ucRtHrZ&t3R
zV@+7P2k}b>nnt=d<m4srdB+6zp19`+$-~gCeC{pN=ZS2M>j$PF{ms0cPo^A#|IVdX
zPO)?RoJn5lfhNvYP8q9;f&GilxB!NyG_Bmj!yh~L<e^Gebaa0Rhqft4CAL}d+X~vN
z=*7@J(|Smj6Qh(;eWPI+*`Q_i<qv(OFb@Yp$77-ei4+NeXK-MIq6ji3I*qdO`9Py^
z06NRwcCqVo-<MZZ5S-8L(rW%NX4aSxI4J<S^3&J<<7*eCGTj&MtGZ3?AF1Qbw&!KS
zw8u|8Ih3?EnBt5J1YvpC$KcQE^@5ph>P&5iFm@DaF}bGYSC05B)quth{Hg<XlEovu
zldo-7#wRax8qiM0HWz<-cdx+$!hcS2nACbMfouER38Z#hbNdjlB;D6ZA!&eCK@uMG
zp1-b(EU{dk^NV9BmB5wed+U#Ag;MGYv-_UQcqHRnu7Im`u#-cgTgtWbmVM2z?(7HZ
zTEj3E$8L^3F&mdMCj3O8$5q<+wNLf|<d?RK?~~{*3luk~_5_!%TneScBZuH%9p1T^
zVrS_Pm^&}#^|{%;g8w7YO?^q-Ucl?td_5$<#rG<;QLeFDnKO>d%B9zSv+_Du;Id6(
zJPSLi-$v^cZ?Ot7NjPe|GtcH?aU!JQA50HcPPo;2ka|ZxK6U$~9MMI@RHV2OT;nU=
zm<@nX)m2yTcb_{z-5t{3^+O!z!KYb|1f!Q_ZnT~-QHKifIOsa*Drzh?;(8)x+55tb
z9ehJ9yZzkSkyAv7_3pRO+_SEZn`BCPNKgoqlO$k3=l7l=`KmVY`QwLGfv((KH!^?t
z0(ke7NGhO9X;C9<yE+zmXjr<-g!0sIY0}<=)_IDZwH&vw%=UBckXxnJNC7c-;2skp
zD1Rg}310bA`39_L)q85i;nx}efe3lGM#nC8)wL-Pn*W>--f=F-tNgn?qjX<V#M@U-
z08=hyLOg2q|0SZ(r!U>+pxL|zCA|^0IM*qtR7&_ISnO!~L2_Bj$!{30Kz|5xD#|)D
z+iE>y)Z)3&DZ4)JLC%>?ViH0WwX2VGetN9+y;fNUS_yj!kH255v;5~8ykJ{vz+ewv
z6gjei*IV->_quesg}-k34-{ASJLo}<p`^Zd%WTK?&3)S~eGW)Kfuc_Q-_TLH=;bN)
zE6&d;@jb-$>JUi-(VVfdFKDAnm%r2->Du9u9>K6R70S$=HKl+#+iTJe63Q35<1gUf
z^_4%iJxR9RoRMG^VdDZ-sJsMw5=PC~vc1EIHU2Sx)v&spYXdO~v57C4Kn)!uc96D5
z4bJ5kyJnnP*(Kf>Z=cF8WR;b53_s|{IRZHc;#BOHc=Mav;m)ZY?-H+Ht*zymAYvPX
za8X?7*mKKs?IRwnSLOmdwYr)auMMxm+qH#ZT|?RXj)s{<l^cCc8ez2(nd<1Wd4&Ky
z4ewwoVdSC<M*CC28hd?85aHz%E0Ksn9EWjOjFo*~p;#I>FByH9h>8#1as2+tbjT!=
zodnY6jx6G+F00|XJG+3ET@NOKbe7j$?v^5qZUGLbeiN8Z5|9HP<@e=Be+Q70xnqEE
zhpo|8#l?9HVdTtp_3jNy_1MNkX|$CTcN2S!?@dCcBHsJ}iNCW(DO255W!F0$TK<J(
z&n;+19nlA9`v!b2W7{EYHutEi-1opBwpXZp4P$`#`NcuK8rXutUNJ)}`mQ7)1pvbS
zg<qa+cY=>|qUV0*gzhR#xZV9;LOXY(5B?$AK~qwxUQ(IyIEkExGo3gW<8q`>uH2vg
z7Shda<}%mdb*X5(gVz?x-+KLDqs89M9f|@-9}ZyE&y}j-2v5~*Tyjd2T!U{Bu~Vge
z(!uKPAGm>9<g6O-Z#5(~%N;Vl4CBs{#hVygROc*W9pm>LeY~Xs?&-)0)ul+DrqB&3
zGjhT9A&Cb#k-4eVQB+rBN6AKBBe}Ym|Afef6K%5&YC2N(m!2bdg99$WuQ%CRm3{`*
z2caYUp?YWyt1FbQd^Tr~|3GqR32rL8T)`Ov9WFcX;vd4S;*&M2dv7=@?bgq$r`~K0
z8>$;T{C@`xKhGt`OGJ}M`D;ZvmimM*<zuo=fkKzlJIY6OZgzSBxiOjKjS-WY=a})$
zkgAQpGlnaBLzfuTaab}PK8aSC2~-%&RZ~H)P@sRzt4B`WEt{$Nv(vSuZV6kwTMBM~
zb5lHCtD{E<i-~GzE{Qf+MTP03(|lDPuF5`1%bd(02jyce(^WM;G@}y(&v^TW%0ye`
za@}<t6t9FPc72#(3`#>+DV80kO>Ayt^;Bh8i&`2KDCfP+`)K9R1$Vh?Ta@f<bo;Et
zjtgLUU-Zi}Ry|989N^0T_`tRfd8UmME-lSRF7?-cW*mkd1Eo+VEG$n=5c|314m=*D
zD@7m(49l1II1kAeS!0Sv&CTboD?yIESUpAR44v$I^^*D_$OHH2;1U@|)uB54Vqo)^
zr3ER4-C61E=vn&n60{*^?NNcwofsc+>kH>jE+`!b=gA%k3};6ROd&x=2bk-k4|Fy=
zt1qJVw*_OIqlHjidq~j))~j@d^HAeX7nk2H#7p}N)F-aaAITb9zMy-?RY5+?pBsMM
zvfUNuQnCw~5j+c%jvM~N5BdBCw!?sp=>5g&TGy?94r%YCB$t9oV3+~%yE4-9T#TI;
z;2nzUMW~-(NF33xoQoZZ;`z6Ga_fu7TMG-`vX&GnU?w7$Oq#D(3Rzs1c@D@7SytCZ
z<F{xnS7b6Z6e1oKQD`2yjYWFklw}7M+<o)P-WpyDespBR07p<2#e%~D&J)_8T*gcB
zo?_qoQGcD@<SR_B?lyiHxmo8W5b6ER3`Zq$nXNWzyll*VoySeZb=i|~;~y>E8XH{r
zU3Vl}{6h{;E18n&i9%deyF*9T-kIGwd}>q)fx>)3c7Ghh66wJeAbZpy`lqh0ma1)i
z21@(vt}S{q>_SHq#vd}!)Vtc@C4S?0GKQ7HL4L3MxjJgUGJBtN>0J@O>FDg&n!7lV
zJwL|D>1z$=$c@o%`QJ?(i;2PooEpeKA#6?it3~|yDU28dpk;6#0#7jH8hH5-Jg&@5
z${FQKwBkkFCX`IcEC5<Ha7gTUBNCdCm|RGSx`g<I<W97~*i~L{<PcJ>xFZBF<Kk^f
z`h}KgYC>$5CVw(L*(JDI2nW^to+*9B1uQX`6l><j&)^}B#COyI?sxgjR{~lW(jG#K
z()HVLcSWei84q0^Hq^ZR;xbO`Ood<Yg^e3tKogK2U6yFaRRyLB(Ow|JXq|5fsBl;t
z3X8LhA#f2}VcUP8(F%8|w~neA*T4xgTQjtvoXlhWR^b7@+<Hawf$N6PTOIeQ65&Rv
zF8=}CB;I0Svw0l-;55wGwQRM^Qd^{IUR#`tRxqfg++wNYiEQw6CZenY(yUk5uz^dH
z(Q3$~yHQl+$C#-Za{!d6^hmfFr3U3MVuupAIFgZWiKl`>-`U}3;oc1?B5}cr^uuOn
z%U3U4b54nvO1qLcIfVuq<eX1vaAK-+>$SHcT4#^y&Qugf-hOBmHkE9>aJbs#|H%{V
ziNl=$vuf|i0i>hV6di*Y*8R3P<%bFuk|A1WBNuZywWN4ws@yUHfJCF6lw5pf8bQ+1
z5`%1TUPc3aQ2>zu#9$F;)bZPKn^h(muy9|k<EK*b3Hv^yT0uJQifAkQwDesSS{+>e
zp;X#Ph&i%(O6!Yvr*3Af9(}%lr|iCA`rw)_QhMC2N5k->&OV9ZMd=C^W*SK~XsuSl
z@&%a)tcN)%sxiMpT8N4iKwIh|(DO>lU`)pab^vde8udSLZGM|I#+Y#`3%qz@%#-M>
zsZd*&zliy4S}xWhNvmXke63*|`h<%BcY9mt%vrX~HPGsH;h5;@0L~2d86&=iOkUQ|
zP9R;quG~1Rd^J-00zQ}UBrx?y?aDRCa;bI9^3yAHSiXj)OYit+bj6iU$DjxPinre8
z#4w|uDu+}}y&0T*3J<$Wc=@pB+wsX!T65zsK4i_{MKgb)NNCOYskizTU%a$=J!KLD
z^Y5LhaxZZf)Z@B>*0N>09BTC1>_ZP<RDu-cDbI&}8Ozt?=Vw$lTN0YfRg^l86ZqD%
z<EsRUe7?^I`!C;}=&=sZ?hHQp3!N$#_;8kOW&Karx8!E$6O4}HXDd@8wu99G3!IDx
zI@}bVkq?h4B@@WyPSZr+ef0YY7p*18#1VP=zP=rm2#p*mxG)A`<wjD=r~&J4#nm;=
z+50XC@7{?~i{P8m+mH<SPQ^93`~H2`vh=#A%au}f@(Ch0s>WLj`m$n0?&ZhNUVNuB
zSm;z#?LQPV_E&iB=@Vn&Iqkx5=g$8?G2lm$CXX=0H(5PvXRd=&HHcF#QC%5I7f#N(
zn%ku#Mqp&Yjj|6_9fjzd1-HMTH<4{_x~;(+U$FW^`?p0HS@z{tb$!*|f7iTlUL85t
zrDtjV)0m$*#T*gtRn`Lj-wpV)l73HAPsO<XD=NO(WkIU4vb2$x#u}!ccHmd+$coaD
zZ=^wHdbAR<iLUEsOmvauLfst9;jp?Dw`-DA<8;Ka6L`t2Tp-~4dHOx!)hZQKZt9WI
zb?M)$DU(RZXIY{$h?|VOiKxLlN$=Z@7#!AsXzSID6dPk8WBTYTROnU+NcIg=;;06M
zm+iV~I4ity?8qI@{ceTQV0TI>Z_5-U^2MI~ldD4IBrB;d<!J(?<b|vsf^Kzce68F1
ze-jslNKV*O33%d-oaNxcpXg)lwIW~v6!aV35Y@Ef_*-dI!unHgQDXcX`#8SP7l9@L
zaL`lVyN!#NcaLvBLgJrx27hF$2`&85MC1N`^~EHt`A>^Y6S&QgUj?4rEeAdyA2%#t
zQ7vKe>CVIP*XO+te&)z`c-9}%%3*vG?%kHt!L0(`JZIBGm={5@(lgyDI^g6&dx0=P
zP%BM{y>lC_uEvjffUOZ2EWF09-95sr=UdBo?w^s^MB;Q7|J(p|OKF)mwjRAR+6qcy
z(M64km2=z6M9-JBT;;m>BJGxXUDTzcI(af<eZJVCxzN>C$ycgRl!bwclcB?E3N2=;
z46%L6j6$Hj=&+LoGAm7?OR0~2tJT{QP{5sCvg`Wg3Yuw}sib-39tsX;eEzEa7KRa4
z>m`1+%0eopcJ{9D-TFh!nd@Pj*gCWQeU7o@t`wbs?z5QNx(CL;OOz{@;2@K?4Y5#*
z2$P{FZ>>NRqBaNF*~!1=?0DVBRAh3(Zt}n*#u3K9S=qW-YPF&97<w(I7R@&!0D>Y8
zE9n>zS)Cf(xR06qxv&2QUcu#Z>J3bp4Qs#(<#PgQ<%EC{@@X4N_OPe^DPQrG>F<8d
z>w|+g%i_`h`)uM*?LG$*mn*T<Sh(Q?=xU3MpLy^UtLv7$6TZ@@qlw1e&r#U{T3&HG
zdd!EHgro-he`+p_J8AQV)puNe52`T4i4t5arQrsTTMTEiN@*KbY(>4}@HxQL6_T+Y
zJ{QMCtMfs39Qbd*R|r$-L6rj7RN`O>UBP3ciPV*ef7XQQP?&a*_r`f+8|?Ve5TFCA
z<^ZLfJ0)MB(e3D}sH<&ZpGY#lLAMBuJ3Z$rAn&$Dwk@#!4`lMPu3ldzOe;f>2e>V(
z`c$X@-$MP=*sverMfa<EKFVcbMtDr;;KWi<zH#XErQ00UHWoXVN<OaRw9b;9_SRP}
zp!(|H7b~-3m!A%X8IT{_C3kAP%~RdO0JNt_`_R79XxWR!MOdb!!+KY*zS^;?=#ujC
zlB3XX71C>0vy-E84_=7!N%;$DUA)EfWyf)ahdRcQ<V6WrmiY71_j}8B|2bE2YLkDo
z$NIcy;BJGPt&|=%_LT)C@5EiyeeS55LYGeQotdON<LyM&o!vNK8;4`HaPH`no2#7U
zTc#z+s<+}gTRUOswgv~=XgrfrZ@p;hI8SdA9y8wIuGhuYOa1#kyL!G%aI$#T$<<Fc
z$aoVWv0AZJyzW+_u+bc0w&rW5(DV6P`}566!{2VYP!)QEO2kR(YM=KVZ9*!uakkfu
z7&}F8-o~mD!c;k!$%}1H7T|6JwJ<CnqM07~r1(Y9o0+$ObFA|o_57Prs=K`Ohjk&8
zbt0ZB$To?so)f;oATM!77O6u>-A5Y3qDeDNBK%itt=A%-Uy{!)u851|F+%_JsHQ!m
z7#Bzjs@XV|g7{enEEOX-xf@)LBdt0byfz`v1URm}Y<9EJZ76+czWh3?hR&*!mzSk4
z|DCmvj@4@XU%ITu9q4EQ4RV*!S2N|JXS*-IBeBee1C5!Xy9Q<{@E7t2Qq0}Uz_!Wo
zxnxpx^7f5!iP5L<qe&uw>djqiBF^ORlekW70cfOz7}NAPN2j;SRo~-;zDob*JNK;1
zaJTc!anSRxO=pYN*IpR?#zx=wLAKS8+$93&xk+Op>grDQwA8&|jgiR5r(C#7iP^h~
zi$$T=jn15#{f^!}-(u1Gf$A+wi&Fh}qE`k?x;5#O`gz67xS{7yFU@(QU!M+FQetCj
zTckATTy*rJRKJ#uP{?|zDQGwsezBLKgvU!`PJIz_LYYQ-c`rQ+f0Qq3Lp363iG3T+
z8)WF%hAVeVpHKSTK?=KEnJ=2|d`|yLrVJ<aY2d$7rFv+~h+9Kj(lY|4TOij1{>z`a
zjkP`_uQV8K=K6wtO1YKqpY`?W+(E5aBH<h#iAe6`Ndd%H1e|~I>vJ}mJN`V{A?DR8
zJZ(5Q3dVk{dzMqahDUjes&s>oXqv&4TRA|U{GQj-?N^3x4PsF(+f_LF&o6Sf8s-k3
zChypOCj;LcOz7m!0=?EA-PCm!6TXXcSZEf3^E=`CPvJ^9sT>j5qRM@Hb45MEa!@B3
zSz3vaMq8+jzoXmzXtuOWQ+Y>k2;wt_rPy5=gVYMQG{U{?{nK=T#2X{KxpxISBq@e_
zf@q=^MlvsGc{~!JyD)Lc#LtwpJ#?nrWj~bvTAP}7%gIQOuTE=j;tI+A97sN~xq_Qo
zj>793ecu_q9#Z1?1I<yYYLZR(48Hxyp`R0>;iS-jZ|7D4XKx#jvF#-__Z5khT(4^;
z2Dvxsk*6cw^`;S98?R_I+#K{72T4O*MhbRKqFL2cQU<Ce+Dy7$rrPUaG`-^x%H66w
zICNh1h*5|H_R;x|7o?bd3P6P_-z{exxj3OmLZ7&7YYiRo`)p1*rTSd+2y)uqI3z+7
z=~sR;^u%CiprGhIT2ox}j`N6oap;ryUvr6ZJHUlaJpvO9bHa+YD9>1D1rfU4XEMtB
z1#e60Y744O>yLk)e8U;Jv@DlbpjQ2`k9xu1soUo72ixysoj-<SzUepyoL}<)_bc^x
z01z`pv~QW8?)@7oese345U#6a_bA(y5!J0cmdVtki>Cs5J0yPC2n(27KDcGdP+VN|
z%l4fh$P$Fq4MxW~=pb-lcl9$`<Z4L%V&I6n!QSjjtY`fROnad!Dn%;&1j}vi(ML1b
zS`VRt7KL1GG*^pz*~9?`0yk33+|@QaPU+Qd%W8Xof{YBg_W057`3Rix*ZsL@m}0}I
zx_@-c1lp(d*i~%w8^HIA4_V!YBhRM4sGu21<#Hv2B_)%qRl{ng=e_J?)^CK1kEj`7
zzK5mViNr@1zkmcyckVg&rcXrUP0r)7K0jjhU*VRb?ZnF|ygv%e8*R2+l5!0WW*UEW
zQ1rwvkE+TE-sz&4HmAz5efDs459)BUI-S*0#kleBwaX85Z9PTb4!xopCThpe3crpR
zl}Xjip5j0*&BfRUjSoAEs&G`RhQ>*6W^s<Ms6RTJ0SQN$N#KVqa1^VLm+wE6%Z-C{
zf@_Pa8(!;TgRYu=40>N+J2vb;S)*qIX;sf=L|WxL$Q%!c2O%Pq0{F`PPyCZ|V4D>5
zZaX0|3fUSK;iy(D(((N2W=G|DhPqv?Qhs)zqm`Y1ZO|`P^!1FHxsSezA<dzr42ZTP
z6w%6Wp+G?b)FersVdwI}o-qAL;Qib$U-l>jF|4#JYT=oRe(2XVQ+T3Ff2TS?xr#81
zdnh}qy*!t@Mu>57h!(UCJ1gMO_wv&j4TsCKOUi9wZ<U2G(Xz$9Zdw}n%T)Bk-;L?L
zYaKZ%8lg6*PhOJzM5e1>A@hXxaKWEHtUA&}0%dmLEU^9?avvh(|BI42lWBndDC2b6
zU)upx@Wv!HL4{p5Uiy3E!4io}@~~OYGH&g*5yq=fJtqk+$JT|;%9z{oZF_Q~va%)S
zQtgV|K*Oo+$0xuEuBos{)L@l!BFF`~*jZhJy@?-tv3;-iIRk0PXEWa*$6n7)UG=m(
zxoN%lxF5s;zD+>cAb>w9F~ub{sA0d8ji1=I{2UmC<m~NTU5ML-K^?{O!DSoifb?pp
zpMZh`k1=U%zUi`HD8m#;OInFX^P>#;N;EYIrYCA~MK6J-5!gQQuz~Uc$X1cu1Osej
z22oK4vt@WvL8>JDs|n0k;j0s&UhD6gEU?x6{1wJjE-HlW%_j9<Dbui@SF$#YbFTf+
z7QO*_g;|AXO*LHa_=KX3G}Nvt*9?wQH3B#U>bnu1z2KNyPjUqm22~GmBvQP`&W$Lg
z3JMYW!0U&#sV`uAHu0FJ?`q;R(~Uow#&e+iw=WbqQGE-pO9NFXD2zB$K1rlZofQ8g
z?Urp`yLY0Y(p6}@hTB@dxr1s+vYaSy9GB%p?z&V)FhqL<!#v8zGL_iU2t%!9&miU&
zHHFj=E4h~Sv@8V|k_9kEF%&odhWU`ucS6wk)2_IB1~-NjJ5`_Lx%`M-*onQ-<89ts
zxtqYbSLLVp0RDm7D4U{AEZD^!f>UUF(EbxC?`rn#`W%!-{h|OG$AfOEw4~%62mftt
z?hVvS!f;T5Inn&it$o@&B_utO8&$0`@Qi`e4roiEoXro2B0RBKzEyRx=xwHz{zh2T
zOWr_k?xh`<=d@w92y4ErpSB<9-<C^uN4qy?lk;zG>(H(}qmB(ni#2N`%sMiJ>?yZi
zdJkQp{}#c%cV521#n5c;%UE1*XMEGMj=?pk<#R7a!<OA!5`CTL2TFfCdPf0EErtJq
zVw=dKOA?F4uc(^um#gPds_y=$m+}S+KbJ~QhHRL{o{8xF_elO}-|k4M!pV-t@9Y18
zBpXjR*H15|@}$0ztp55xC@j_ek#PVa){Iu;UHGEsVpn0LNYxOfckvfzmns|6zI)D0
z;TnBF`F^>JOnddun!&;Qu%Y}Vcn*%W=!{TNzTb=~HLpKOaS090dk7nt#bZBM#k}nd
zD9{PW<_4_W2K-p@jIjpxR&j)2=n7UzYqg{cnZ{n_mNJS-)wbnCU_}v_RG?1|u50DZ
zvcvjY)n$pYmt^sF@X%!GJXC{ui&0d${4ze@o$A~tZ>HbU)Xv%@MldbfFD0<->*mWn
zP52%}ufR0aF7)|uEllvIB+UJ&SJj%E(+w4>9Vcl0S;^&LsA-{Hv+18NRqAGr{NTzK
zEBY8e=Qi`X3}Ct3ov1hwW~FGR{Ww@r_tqYKPMz1Dw|+f=v{}%J*7fpueh@G+JKF?j
zWhll>ka0ZSzV$kCU&L<1I`RHDu1hDd(6N2MTz8i^z`@@6MI&yZ@)yg+J<zt;NxHhL
z;(=^$tNrMf4AgA!;25~uJ~NefuQr~DQ+LF26plA4=m(B+v?_%a_T7i&XTkmhX^uK=
zp}%}8L=7w3K?dAbd(NG=vH9XL-*h#?F}_5jD!?M?Zu}5Mhe9C?yDGfN!}*!VOD&A)
z75$pqk4vs?80ZsQ5#C|d7o0t#O+2KZG2ck5A0y#+)qHqPXjiLAHbYLuxUnL+b!(+7
z-3?-6bqgMz*>u^&9!P%6(Z=xCvkJ*5CG82&j^}lqf0m^bo29Grdwy{Ad;dmqw2Edx
zL#$sXUT2hE@OsUjDqscu7-TzcyzPOR3VM(ry%QZ|J?u{QxVBlKn!VM3X*Q3Ex_uI6
zSn>5LB+$x_!#1aY_V(ti#WHbW%<n+Tz#XNn5T63Y51-IJYT%csB)YSd0co&q3?W{w
zgj+2U%4*lX1`!PXh&?)ASBW+dg)W(0#*FZ6WzQDntQj}ILD+C|6-ta5O=)gD0L|t1
zyh*I(Els{j&<|97?AB<XF&Q0G=^S{m3-n&+F2<#6Jth~bsg&!mlOMWXeRMY4#1Y!3
zk~EWeHl|GLBjMp0Dvu<Llp8o^gkeuf-rA(5|GhE5L`o&xXw<X8O%<qcgwiVVsCH0s
zZY?O?Uvhme8zA-9e7)U@)A1FO=9dajRx>4)JXK#gRPGA=59FB#y!)u93l<&yn$pTm
zxDBh-u*?~ax5pdeDK0bleA-b&%x!KI7vbmie@*7%-6Qvt$01DYqIJ>y*q?52t@!;u
zYC0!)C3z;P0M=>(!nE`I8Z%m7!01f*{_LFEbtS&1S59?ff^5C}r7z6`vMpzyWFX)Z
z=byCDe5*I<N%6}ILRj_y>p!Hx5QX{n<Fqg|e6@B@9DGN>@^1*+*XVhwbR4c+YpBeI
zl<Sw|68-W~VI?bknKi8yAn}apAsPsR51k9<COr@nLyu&BTZtKlWlFE`&lNe6ld=Uq
z|HAe@zkcavhvzs&$zx%c>q8f21}?LsH_@T3aNf-6U44?dpC9m(st?`&<Eqb@L5TvU
zdM@u8H$2Dz0*(h{gQG)8h#T1HqX{mm*o!XtYo2f5vJ4wllsq5Oj$85c2nNlG0(?R*
zLB;6MXMkDTDx<<kDq`p)tn!ESDrF(PUPq&@K$+jn*ZM;4f1t~_R~1g&+Lkr^-?&{|
zt1=7vf`BTp=c)Gjw-@VYD!wZ?sHkzK{Rax>KWxeZ)u)2{7W%Ah10T_t*aSIPy*#3L
zJvoH&4LTg2sa6+BcqH&-TSDAzN_Or`hY*lxN#3>lWcyv~lhyMgJWdGZ*DYJ*{ZCEA
zO?ui~Ji7PT^0(?C6(IJIyq`PEs@qUeJ@qrxv2IbP`#Jq;XIxCnhkaDy#9TZjL4V%R
zE!>GO9mJwxtFmeasYAC>v3v7m>yiqo%AkiQLq@nEqZrQ$8~<-|P1*u!T`p5;N1A6H
z3J}J&Ol!PSP^}D4W;%Rv8m!usgcZiqbQyBMF)FVWx?^8p<_IwdWBxaHq(VYURH=-!
zSytLs;NBW?OBAdt$8;KpA-kUS??cqR7K1$=00tw!=`l0+DNyaSZ)X(y;bT6>Efg7M
zguYHr_r;p0n%pxT36!g_MJ-Q9!rwQ;PrmFO*zP5T$KW!~ADq9`oohW~LyZUjD^}t~
zmh_YmcDSe_`W7k|tZR?#+uSYtA=!F56!mbTJ{ick<D`b}(M>y+uVkGLI8Jr_=WuI@
zV*8c@-qp+3vnh>ly?lLOV8u-*o-*p6^LFxmv6C*6|DVyyUq2Xk?khXF2))}(latkQ
zMRKthbqvd$uGS7sSIL_w69frX{rz<RA~ac!<NS^r51Z*{S7c)0z-o09TtLY2aZszJ
zbY&9BzL%o6yYAP=^!$w=Qnvcbr;4Xflne+9nC`C`VO64W9gbMJ{Y6TiLiK;3Do^q?
z<;e@K&x#p1)n<d3(aX4*ne!g^$QP8Z>WaB%)a^2RsjBf`35%Cf&u9%t!y_9I+q^r-
zYn(^a3F)!Sd!w!9ubB%{ZRXx1y0Z^j&94J;u8hdFX$jfl^CerXs=+}c4KBiCk_|3&
z*}>yp_7h`=u*IdEdJ<M9-;CfH7HNRBZRP!_rfaCZPGb6Rb$fPpg)AvoGJuK3bv-{c
zs2PT!-_Z!Swx84Xy7Jkw@Z(uEh317sZ4g6eX}7Uqb$8s#%?U@D0DW77G*l-}_X{vm
zrCkS?1?NdkH?5d{?@SMCOt}e9gK2%E88j{Ix07XGr?U&JMDVvA)n>I+{JzcI6vCSh
zjZstX8U@7cP=w2GtP2u&^+!yk=EieItS-pb%nILvoOr%rvIQ_Ych2;_Icg3N?B>jl
ztjwM@licH!lim8k)vOqQeaD>3>5PgD{r-v&&Nmoo5=SXm_78j&Z%UZa9B<{3yCq;6
z9ZxF3os)kVHtZyH2>=2OYLDdqS|fD@4G5uZKE~J=p|k!q;U>)8@@9q4gI24B;3P#N
z6J44q#42QKYSeLfB?IM;E_^RTFkg1&x|hBb@lDL9=m#lXcAn|Jsv$Ub@B0xzi#U?W
z3HCouB&b!V6%j6tWN~YzW7P&M9R8ih|3Ul-{gzdp9pRaNItHuqB<^M^pM2~^X4VCT
zp4P%77xjtES}DF;)21Vz!M<qr5}L{MVxRuSeMVxrWTUK_nfPa|UbmM<G#cI_P<lR8
z5L&CnYW+PK7Ret05nHAe#xiDe1<b6i#-8a`so)ls?Nig4=n?Jy9GA0MA(Br%>ehqV
zZXLvY)Tb3{sNQcc%wcdp!loqt;Ja1n=1@WojhMHQ6B?l$ocl-a4?{M!75UGIiH?|)
zXre)qvoouAbf4%v9^1Ro%RurK5}ZdKUogR?8Lp`)B97pQJ1Vx%ugbYLYySwJTc{mb
z7kc#P)2EAjbu?b1sn8eM>ziew+=t55_u1a-+LE3>D7fe?<-C+GC)HqQ!IyaBOseYl
zKMX0@+Whvls8Ds@g45^J6v(+kE?6|jX|NIfs5aBEKuO+7+O<*6k%;jXn~*I^lebDt
zLk_!E8(jlW>F8aPW(BMX_=srZ<$wUNSIxjGJ@b@{l}QM_eF(-0r`6zjWQzvt8XBxk
z4;zc=s(Lf20LMk$*vi&oX-{hND0@%0^-RV@&j^ei*Yr`p?V#wY&6c!VkzMiTeVVr|
z&+rB5{T<IDLg;7G`<`2B1%P=+CTnb7-b6k0RILC3zdTxzN26)@;O$+NBD2Tareg}R
zx4SU0I$qWSwS7vSqz&&V!UO_76Po+Ovadn6PL9)n&ZC1qt0k(AzzBbE1kO@*)n*_L
zR7Xow2ao%c&><rHIary$p&lVgWmxF)d5kMry}a~FZj~V`V9$03qr{UQ$3dnjY&;vq
zJ*%7UZ~A~#xO@AcC#;ly&zz+9x)pjW%v$C~oe#T;&h2jo)%`hvj+&#h-T(hEEutE2
zHr(G_k-oawnt6&_MQPuy1WikkW*ZJv!kv4i&|gQM6q216=zdzZLLO-OZk_2vS>ah5
zkl;81YBP4#apSPlJ``jAn&EYgzyE<yg;Cp1kpAps#nYXM26KKdq`y=BxE#%ti^u~2
z&{@QBt=eF=7c+|E(#_=H=r?~e32ICk9ZbHCD(;o-Hirm5Jxz4;@6z8=rapo|qIY8}
z@e!VBuO(LRM_<SezeECQCApPoARy%X7MDN?SkH)<DeO?WLZ_!_ysJ})>($l^S|Rt&
zh1VyM(kE5p$&N!kr~+<j>Dj4NvWGwmEL^c+_HCr|naQ>a`+d@f(uccOqJHxj)2-6G
zoiwY`1x$<?R<lXls+(8*y(~t`o)p7ExBD6WtC2<?w9ihz<z#3sQu*WRi85*$d+#|A
zjr}acw_p*Ic(v&1TP}``J<4a_CeGTMes5XWyD%nObd(}E*K%k5V;SAFRY7^wzckP%
zvo}xHD(Kx1d!dVqfR0>JHDRx;{#A3yupPc82SL|!n@UR)jnC@I>Ha5WS#P@MuB!J}
zy`8%_j{$z3!?qfg6WO^%FF9CA*;gJzr?V%ZyrmUNQ86xvk6{>PW#fdIGgRkf(41Iu
z`&>tDk0QZ@)e<nLJ*0BCRc@P@jOKWy*^PCC)~H;Lw-w?B57Z*$X(^j4!A<>gy1-I=
z=5ZLBZ{}OHfve5C+LBDXzIVLqzI1i5qQA(2l)l_*t7L~hcqSTJK=NJPzo~<3ssxI-
zH{6bmrB8wNPsM(hioZb#XuO4xF%<C`2rBX((Qoihs&ZJE>p!(WY8zQy=;YW|qud+v
z#!h?frs#T*UR)<k{(xGxKO_JznaNG6|BBe5`&{@-PJVLT>{^<kYatfuKHaf~ru@@w
zS<>?CEoJ$E5#PvV5#ssLO4qVREgpCoJh6#R|E+UVja>Le`qyltR?$f73Ge&f?X6{{
zC!OEbTLeN-ct-_%%fG?Hk%LcFY1M0)2ps9!%3KNFC$EeZ#{YyfLMGee2|B6d*{@2z
z>UT(+lcF6*BdWeRVNu>ut-sjZ>rxFhoZ8gH37JDg+VNN8RN3fzB5j=N!jYEWf(zzx
zp3cz{Qp|0fnk4UfHH=qdUMTj8*&<!O{B?YXqh9BQ8pe6X{_h!`H{hV|9Alp76OX!0
z9+${zEoa}IXc<L3;&>6{D(#z7muXeoc+nPp0))Jq#MCqFGf3x$#i8swWj1dNd)uxJ
zEEA-l;)+#5Z<SkGk{->-WENJJLFCMB0QcX}!bJ3A%-s4nO+vw0(w`Xn_ayXkh|kJ_
zlc&~_LT&EQ#CE*2YXx>h=+X#fZu~8e^F6MlfjX97LJ|OsgL>eN`k9|ye3tgm?)t;x
zJ+(K({Tz8zbW5WVh}P9$1-0-J+3s%kRF~?mp)A{(yA&&)(3Bb9v~KHA?fkSt$XVXp
zP$GH+!57uwy!kt=umAEFG2Hae?B;O1igM#;)}u+kslWR;^52PxS{FbHu{c(DR4+W=
z7p5|y-$r;;VS%^4Fh=+M7gyHL)@omoRcsw(lIs7~+g3+ZPz^=+Rg!%yPre_gi}H#)
zrNZ?PK$S1-DW&#Js<}=6CW0srP8UjgQl=*)Fb9}0sXG+!S)@6M?5HaId7-H}h$5y0
z{xvN;n0?@uVkCtN&>~v7u+K`it0Jwg$2aR>LNr_{4)wa$amn&r4Ar!k^rn-=kye;k
zp&wz;q70@j8<(j7USWDVG{4!G68Prt&iQ`g`*(UDgB?+tzgP{yH|er-TT==#zY3Dn
z)S{ZGc<(gsP-)T^Dxls=C|XKJdlpc0b&_z}E|E=FN>mR>%o1CkW$MiXoZiQddy!$1
zl!2==iQ2#}-PyU#&<ySkaLvQqaBhyIVVKNl-bhzR+4x<RkWWY_#3!p1KH@9f$$;EB
zHigRbM~$6)7Q&SQTl=(U%gaZN3NHz!&CXt>Mx-l#)Yy<=f2~Sv@8`r_HD1Fik(?~z
z(r?%CD=sksnWCbwK^yc)y2P;NppZmk8T?GJY*9|QA*sz@8mZFmj@DF@+Ob;{l5Eay
z3q&luEP3<6-yKf&7A4Am8urO1-)p>dlk%B2T(LT=FVm8oF0+&S_~trOY^u^!Yo@wA
z4>fM-t|Xj?ipq$`(!_{^T6P<xdCQ8db~G5gY`%U)VmRosHM`VEnE9;cSEKhMJ?cX2
zjXqpi+c%m5TzllLq1I5IDEL=GA=_ZJn&r6v1r&2mXdL5`tE_mfRX|BhS3lgl$3XR$
zYw;U_+_*@IdPuC2>aZc(iXay;*-qBzS+%hmI<vEKR_IcV5S2rUILj*0$#l!tJ~qNg
z`&oy%`i~q*$2Q2x!ZL+(b(~+kjM#;0)u}1wzhUBl#ntdBi$mp#sktOMpmDn#Lt^yD
zV@y#gvw1)gNp+d+5qRup`OTj-0)hI#Me34ELnY-w_`7UH23K3GUht9IP5zm^&DRw)
z@-H|SSilGWyr8(iVpOh44hJ!vx6NLmEI9tPLRXXv8t~bNWRUYvnpVKV**#rl>j~dW
z(-jeiLHz>4aDr1!8}7YGtELD$VOfjc4LQ}WDDlZcj?vzRE$7xPk9{oddc+%!M?KY|
zA=$fh&U>_lMZV$U3{-ysRFHEOp$m=5)7H~giBx*hceh{er&N!bB4P?gGuIg+wSMD}
z2U36PIunXAnj~6wNWCPBT@#kO6_*~Vd=DWO%lN@fD##A2u@i<B3Z1L`_O7^22QSWD
zwg-%g{{s!)TeP?%&hssy<FlianNs*OHY}I5GBxxLKR=s?UN{+*@UedKn0w1LaK3IU
z)WjQY+fiI{v=_kFXl_iH&VFG3FWi_bW8IBIa_zOQOHnk$^v`iEn`449@!J#L?*Ja@
zVcRA-=F7XD9Lp~Gq0DmuZ_6clcG;S2H&q9FWvh#+v&t;ku=*hKdgIDXHS`<u@$a0E
zgrG+dE+0&96x%OsMuy4Y0E>TKe2vdwo;Gw7=!gZ1*L(av!)o+t-f<U|SoyRt`&@nx
zHqC#F`CMX^F@9QJkw#TQ+z0u1fj7uS-vk0eoTO`xnt9KjI$L+m_xXu^>ls7r9=@Fu
z-oWOdu6r{395>46jf6sN$HGK^*JaJ#-|4i&w#d&&C1Eq}{!j-}hzG>a+41CqPwp#9
zN1H80K8{7(7Q(Xv%36(O1eCd$Ez-{pJpv@at@!=}MZGo#JQJ-Uf2^^wT395Z<Wa8Q
zegL8sacd*cxtBC&-3QG2eh0K7%pk9fnRVVdWAsax3p!MvJuT$7ay}@Ie~s0x8TKlB
zMRZcu=R&mBWfa1^J?pb+k}U`P`Fd=v(?z8iykk|FU$ezE<uI8{@V;Y1n&dD-*ANt!
zcnyB%)2W}Sx}sjv=hr?;U|7FQzVJ;e>NytH=L(qUGuqRkP042GZq&?L<4?Le&I77h
zoHe7MN5|dIyVmeelXB;q2SkdOsB6h`VWIQIg-LGR>=zlY?b}@csy-O_ryH+m8R_xT
z`fWe7^!&pQWdL0$c5bN_?N>G)+1LC%r6R_%hi4D}o^1T6@}GfyPyT;4VAj%t-Ai`J
z%7f)sMZ!Z;1<U!pz&f^@xq7Ok^i5bCc;l1|r^UIB$P|7u(?k37S~p&uiLZlRTJG>^
z&b2~kbFR-@d5WpTA`mJZaKpIJ)a_-dM?C3>apO=6-yfgMl$ueF7rWAwkrLaXP4{r1
zP`Q1lOj|!wDYvhCBQc~m-CZNvS}qp?A{;LSWk;=LyYBF`#ZDHB24=^*GAhxkH%y@)
zti(Y}*3*uIy^liqX4u$+>Pwcj51UQ}*%wcj$M`qZ9eBD+;Vybw3tYWSfAJzH@KJxK
zM5VCVYQNCn^|}LF)6w&ii!mOr-E&4BogTmp`-ysdK&e^UiIb+x9DqpP>BE*601JL3
zu$~{P{jY8_K?qVblJ#PyQpYo~u98zr2>ih{1%Vk372!B+>9?q*(Om8C=(oz8CCckY
z{~t-`9?$guzyBe}9Kt5&Gua$6wnEP5^I>C{L(Xg@%yFfN#B!Qb%pr#ja|)3|DajnF
zIgTWikU~141J(Qc`uu+X`-9zXyF8wc$93H=;z`ZCU;kepy`wcDwbw|XuUs2db07}}
z-c9X;Nj3K?3YYeJjau$)xHXRB^t#Mdn97=V>W+KR#9jMTSdMt?jJu9>22&Mb6K4H`
z<7+Fx`MQe`m7s<K{>}97+CN(GDSskN9jm@VpX=>IOkD?#R`^1~b=u{T=u)n#s|MXX
z8`1|Y8^P@knnE|p46oeP4<vGz_#f7dWnhqeRIQ)~9wikphEK^v+UvL=`DX=Hfz6xg
z7~P?8tRyQdAgD4wnJ)gWYOO{ed(3a{`)Sb~{{o>umzw4VTB9Q!Jxq{3+26bspMkzP
z1v6a9m|*W)eggI@7lS3Q^a^2Tm?zs@kOM%t47D<+vTumqb6)443rmCIFE97a^$!$$
zk=n82P--j&TFGbTe-mF9d?wu(u+5Jpd%rkNX%?057BBwR{!8d7)FpQKwxP$R42GGt
z<8ZV-^8B;MZ$^U;+J!kQ&*7qPE7*^nt2t=I9@qQ&To<irkCk{~FtU-^)|zLwF`1;0
z%1yLaY0%o-TKX)XzQmUv7W9}J#hyH|g{9RAR3^S?RvHoh<$9Rre<$OhZ9wMfrbXGW
zn&03@k&YY9K#gAh2A#mce_`^FMJmf%;Qrv?jsh<&+iQb!e{$9T8_AZLofv;P(Aa{k
z)l(?iY8OkiG#<j&^{6}y;gGFpL7jvq9#8O1XeBoat;pp$k+dwU^&kPtsl)Zdr*Fx9
zEfkRQN_}BcgD(;d>|fvf!$92H&K1BaD-D#Lc_L_J<qL$%0mG-LC2G`0iM;F9Dp=9B
z(e~8&W2w1Zu#3mI1sOH-+3|)nt0v62#A=Dzdb5ZKlvg*l#ztw)a6=@5%;<%h_Aeay
zt{aUTFXSZP{ra?@oOavUZ{l(<oh@o(>WUpZ5xd{+Kp1X!B*#L3Q(d#lom6>35p!xd
zDp2?o!L%Lca7Npe=>L1yCV;-?gtJ-GY3<4z!!0qBq&Er+C_N`u`KCe@KH+^5CLtyF
zpwW0Une%z-rFY|V7I1g}&{0`&ho!@U<AxXr<A+z;RmokN3~%!%^ikscp7dtZ4Ycpy
zHT8y)@$#8IR$tOYtEMtR)rz0`S@UiR>Jo-xEAGB0X%*p>a~DNw&PxZ6D{JB)bj^yN
zBHbU3Ena%Mae!>kPdre1Ua?BK684iGZ28r1T9i<Q8LFX5vHhYC)O%XDId|wZ(jb=b
z-;H-YlvFL9;CcAmr{5Rs@Ht_r(-z)E_a^n}g^Cg?L4raMHcXTl1JFx@GskuV)9qj>
z7t_=w#PXP$U`q$UQQ+t7M=}*hjz7p7gjil*WgBQ$S31hrufHUCuKd>N_bWtpTV#dK
z_0oaI(~ws{BjKge=;D|+20xp3f5({E&y?V-Yr8-6x=5zvhD39QF@@Fj%x>Ga?S)r;
z^ImE#@pt-U$5eGz_YrZBWq&`&a&|c5f@jOyVZv7Xu*Yn9w``@9i)T=5v^*`|F(;6M
z(hHhb_jx#wh1Z<Q9|M7yk)vrr1KUH++io+RH!+}f(&$nQM7}X6Xm0+5n<F<L-1G)4
zcOmm;+#Hd&dx+V=CXa!9x*)t_fay%KMq4}AMf72IHN4(rm|*X&MsD78yOg-9vrjV$
zm=*A@$!;>7f860!G|u5bSrE~RV0w(I;&`b`6D!}m`ySBSnz|yM+x{cY_4#v#c-$mC
zK<(C#lEkVw#ZIgOc}-=7Vf1xh*Ybk7?k|U!gd`XjS}?g?Q`tl3eb-<$^x(BXJ_hWO
ztwtUGDlNmhagArk2W%@O{mnWqd3VQj?8B*a+ba9ULR|;>Mw}ei388j;t~bwG7RW;c
zUVtaL495PN8Ox6PUD|3Pv!i6D9XmywJA0lc4;JqH!E~2NrKcm3o9yQGSJ52CAHH02
z$D1TKA<`C|T~nbS8u1fKSMYb&hb%OOZhWpHvL<5QunL#je-;pNe}EAMpkd#uP&s9?
z#Q|v^f=Sd%TSrI}-Aka{i-0?*0QgTKYhqyT46}v_9;hAluqE032P&92BKY5s_&buS
zPPw;Pa!6`t`erYx%~VNB(OgO0Kigi_J*NBQ`-t~RBXy-Qjkb3~u3W~v?^^3F+a#YJ
z!~4!THmXK6DmK#o1IaS?J5=3{77u6Aa{=vFY_`t2{bX??<N`z}u5fZ1avR=~Mpd~Z
z<WneD;yxXde&#cb_Z6w=`IZgSrSGmzUnQI!KI8G$93_*~+-CRQ=PNgHwB{Z4VnvQB
z1vV;`MEvH_QGre`-fuL%F7E)>{O#e>7ZkgNwS4L$rb&ZUnyuB>04@<CV1FbE-g|K2
zu~!ZL>xI{X_)}64?zo7RJNN<q@&2}uy!r^U)9n#fYKnKeLSAKDg;;6utrxN#Kw?Fq
z56py~ySzhLPF4j_67LF!hH~74dE7FByOem!?8&ZOgL;AX77NwGPXU3=Ooov@)zkMU
zSNTm>=Izpav${2j-$d7inc};g&ide0_{^e{SouNFkd<DC_5MZ@<czSIzHdIxBfUkT
zKn|NX&yApf{VHy!^5?pbFVm+mK0zugmC2{-WV4D8p*A@ooE$=W@e^a)7y9)cbCYIu
zBw77Uq*RbXqonDXa#j{%-bwLVSs{aqiggUG{7e~H;1<BT{mO0m6jW$17Y<n_6pf^;
zbtoXKNR}0?KO(0hp`YB7khT7>{Yc;3EoFu~ODcq`D(P74a?t?FiZCtup!@^mt<}d+
z54>N71B&f(4XpRzfvdK=3~^!X!hMOSXDoSk<))=%ACOpg2i^=w3$xH<IPuoP-Zj{6
zTV+2m`1Dg^Q<tl);^$Zuf(-a?$V(pSyA{q}90$!0pNZGwpS!V$X5zbO4d_R8UH2Q6
z5i4<=){FA|cfc#rD(WJy9F&E64>3J)+^%MHTa_S3Ruq0MSRr#>AR>3-WgM(z9TJfh
z?o?dm!CUu+U~m5}j?=YCV7WqbG<j?#s4JR5Bj|5FzpaVG3f(g|JKFJ`Pqdztq$FB!
zj<Y;c!{m1!Bz8C)jSca)rtms7HMc+0Jxa5QwfrIN=;y=}uGH1GbGsS>SXg$O{`Vs9
zu8p#_5bYaM>EO>?YcI<43KF>H%uPU*iSKjNw!N%}t~itptF}CYj?J1|g!pu?>^Bt*
zQM#=N40z7i?C>EPG%W3^yM$tUz7o9GST?d&n8|HKbkAIs1mzZEDIwvY^*7fV2IjCG
zpmv^d$heXBTY;NAut1cLET^WMf;0G&JPq)NZL4+q#vt*hs53eFGBJ_HKPGIcY?gi%
zXCB^AYQ{QUgbRn)EEGh>A13yKm(;U1g6?5(4N%nr&z|b=bQO!z7@k&z)@!G0%U4<V
zTP-I$qZ^Q($GLF4v=}7<Bm8xZfaMQOaDyB&;_)zqqSn*XYsm*4Nf)F+Y%Q+qVX_?~
z_uE?6y%tvbCGyoB?8NIzHS<QdTsnS5#dX#<o7vgrel-Op9kQaw^)~y|l9lsKuxtS7
z<l>NYNNNO!mCGba`rilcE+!y@Cnjns_0~0b2@{z3tU$%+hLE!ru7(>zt{qAFS2E88
z$vMO7^i&H)M)5WhvZcVND2|jo@Q1Rs%fX;at;+cpk2hN$?KhKcDRKwlh6ERXF9OqE
zeDmEPN^<tZSaE(y!Ne+Bxj|kraH1Wl$ML3)q5!lWyTYsm=9-`V1hPS)3KdfZv4^OH
zEx`KFV6^SA%4h03H+_gNAnq%heB9daz`vrPw~=<QJp-cE!4X(AgPZ1h!hz{J;G2OG
z3Krb5DYvq&@6pj%PTgMWLzeU*a#92wC{|n%Ol{aU_nF=Ojx!ez+7B8tC+=^Su(@--
z5*qSxOP%!|8p{G2M4JGpiP(!Ti_f`Xu<-s>r{&h?M3&fuQZ8KU1IQ!t+3WRZ1(~wd
z+l%$qzIX1Y;k9d_n&Ss85*+s@5=VKPWly5_$Te#eb_wUlnZB16FHx`O5MI3L%_+%g
z5clP$*c2qWVRFP3pM->yswKwY6bUztap)I}$j9776Z^^01)V=9L(&V?t>3IlhekrJ
z$Fd)EHwb2;7j6XGD+cLg$-vsWx<$}5{CYy%Y5Ix6tHMYf-?3^PTy4nh3+s)Xh_{L2
z4W*n*t?{GZrr&ug1`WrO^ifNmD$7mn3OTw((3gDcEcq>ztS1*O_Xa^dfKqy%<-Qcs
zasbnYNb$B71rN?UOYK;waM>QZ`sBm3&IyG12HrT2zrJS8otyFaGt8f+$$wOt>~SUe
zYwPT-UeldDG4ZH^2e8^DZEUKyqNpPt6pd>nTR16D;ZQ3D5;szj)s-LkeNmMTxc0E;
zg-iW%b>g7vox^3KOn)<mW!C9fEGD%TR(Vo}nQY*(#EdgyPO6_n|1MN(?2}x1T(hE$
ztS~AQd=HuBePvn_Eo}U@^l{s3JKZy@I_k;M!A2xg`1pigYa+R?Vv`MP_c0Vgr)H*m
zY7icfHp4@l%Gp*uIv;q3$zei0Uh-I(@8-U!ZwQIe_bb#zOQda3MjBXej+=82x*Op{
zeu)mZ>GvWuRJLm#cr<rFjuyY-DjdvFEo-G`N8<#o{K$rk(Gr7s2ZC#9d1Zg*{G|mE
zg_c`<u!aE<-BYUzHuy$6>mbXT1#HAkICotKPI5GT_4ow6UG@7Zj6lYgmz`uo95gef
zFo~}GDpq1W!xx1~I%CaG-*{H=MRI9=`m=45=CyMrcNA8}^Lak(c#kY~cN%dc7o4%;
zQwF6WT`Pim&F^}@R~AJ&2LZvMhprjvZyE+to8iL@yHKvU(Lw?#r@WYn66J(wucqR%
z1NFC|-1-NIQFxB#7|av|e5=@mhLlp5P4$g?M2N219?#gwl$Cj<t#pf*r2<n~D}neq
zPpNaWfiMq4lx<Wpt<|>jrv3lQQe&;o@vUFiQtRwcql;!(Wxg~pvP&E3D(bSUFrby~
zYj7(SM&`cIl$mipsqsM4wUi2N9o_k(5+6E8=6$aRnc?b4xzrt|l-|VQ1V8S8NrMWw
z3J;<CeMvcB(nL8j8uq-=8_hU}m}R5HAV$>?tGGsm;raA-#nG{e<hIH;*e?1&jGA`4
zk#^ASp!Z#!)iC?d`Qu6}YON<qB+GOkhnSoLontv`6Am%&g}9AO{>EqWe-J;M(H=MY
zxJ*2nQT#4fLPQrF-p%F5G*6O$G*TMQs0!JW%&PvX1zSa4EL+{|NuQ9(3yF|9Ldll!
zn=Q6zKbaHX*F(bJTjR3p2tQj?9xw5L*;*xjbK(AoGTfA;!;MQlzQdHU${C*z=-~Jd
zhI6KRf7hAfka~qYy;C$qZdip1G`R$mo62c{uuS{5Ty>6>YcWT6`pHy9DJfNLR*ymX
zc{dx!pYl*StIz&3DVBZM`w^J>WC5YW{!i!>^YJUb7}e&^#wsUmg(A9dynd}CV*s!U
zQEn{%Pv!C>F&uN<HSb67OVD*uUePoG=T*-keZa~yzCzFqKxfx%=<PU5C$zBdW<TS{
zumX0+V6v-Oj(Tms4<-BgEG`fp-fL&=LUdPRPsyh<&E&b;?ZW*2zWvjk`_AS(aVlwj
zWkDhLc)raCC)?$uF<?uw>JxJ2n+T~vscJD&=Ch02j^()w*oro{M9Yrw#mZxUBO@x{
zJJByUSkFLH@M6yx+mv3T)Kq9ewAmNbOWy{=_NE~x5%v@b>Qk%hzlnT($Cl*X5m-cE
z33(c+ebs^#V_vLkr~3pwt;BWDmeF1MT>N)gv%yoT3`VF*7?AI4_IY?!vM0SD^iQ)z
zX!%zzgWEK3l0}s?rfg6*v~cN*^U+9QRtTw7me$T^{<Lz;v)%go4FT7e5m&wrZAWaN
zS1@rA8d&t0mAYDB>5c<IvwX@18{?Y~yAW5|x#%}JPxel7LFkC(bdJV;#v$o$M8Td1
zK1&m!uhisp4p}N-W0sVkkc-={Ws=lVq?W!k$E&%b9?Lw$7=kAZ2j#c0<y5>5q>`-F
zAc`al>9_HVVlBoN*twOb9b>{qI9+Cfc+m)uB0Uj^_d5RO4)(15R3pQNfaTuu431VO
zSF7Dh2s>|k9G$<OzZrQCuUHFgC_;3w4M&n4KP-%U>#K8OEUrA3i>HaDpFHkTeUmKj
zbuS<PK3DnPwo?iDE`Kt66HqYLdE<;y&{-ue`qLH4eq~Q|tohke(ws4@YEML4Al4Zx
zV}C|>L)6$-+QX!WDJ7c%f!kF~_BhD4O&o>#-f3=+IOSbv$8iNR>wG3ee`Tik{<Oej
zNE4W}PI)<cbERBM&<fhNG)hIJnd5u@kRoG^ax05-7cb{AeRF3qakj@~31%w0Klwuv
zIuh|D9b*8EeqTrsUV!~TZaJeNZD@K{uN_hFULiKaH{tA=oV?4qDpFb6_#ZW<<`K6>
zj>mRS<nkTvBX50BE6h5??3ZN`Lc$VcoRIIWKdSo($KK4^Yymzake*&G-;MMPr=967
z;3CAA=q!W}l49x|9l9!1i9r-(Qii{nj&<ob$~ws_bBK@Qt@Vr2l+)OU_jdfmZp%G|
zGoY7MOa3Y071A*x1Cr;+I+7K@3MC{BwMbIVXRr2r>NG;qU{80Zcb3;zQ!Mi_k6U>n
zVcE?$6}iouhT@io9+gShfAboeP32CujnZg&=>#m4ItHw)tWw0`tLU^kOY^l{k0mqP
z2=+5u9*q(Zoh7O#2c=6U1g!yZo<ca)YPtDh8iD2Cb*!kwDf_4QO_1S6&vMWb?V5%`
z*JxdBz=QU!K)z>xk4KS;0zKGJHt=Zd@tA=5x>Ai@I|)K?p3r<#KD=6hwV3aKvL)&s
zburKV;#)_yz(WGua+6x+;l-|#Z?brrPM(yHZ%0DN&1>8Q$MtVi5SPz#OXu(DMjXG}
z&nzK`R`NCuXsO9tVE_BRQ3<Q?J;O!WBOCZ)u|%W~+Fsfgu?;v2!aM+CqJ{L*1XF<#
zk-M~PEdzHfs`?MK<BqwOI$L0$1$dp|&H!IeNR{r(@u&s|uQzVYJptwSA4-8}za!2`
zL^_SXs(<|Ps=+>D`sMtVZ|_4T`&2XFl+szz=H#ZWT3u`^+#kG5Ur|4iOwD{rj%&JV
zA9ZFLx{B$WnV8FPiXnI^-c4v*j*2b4X&1f!A4s9uHFU4;$`q@9W!`;hVB(@?LE&5b
zb*+<8&y7CzRs(-hCMo=ztJ|YGpBx`$Pg@O^U`c~u<lQq*e~Ja;*1ELXzm5h~5BS}P
zx!b34@aV#Y>(Z#miaZ^hyi^J#gZ^CpDSfk~w8Aw>d*oKot_3G|?8dzH^~AzywLF`5
z*dL2q$IPfQ=TOCxP=ANbF1e9&nE3CUtML$5^APxO=tcN1nd?BsTw#k&YHNsTGxf-$
zxo&Gpt+>lI?-?7HUh^A`gJ#EtdQm|kXQ&5!u0l*!<Qfp$Z)<w5*Lq@QB9rUfb>1b(
zfhNuoy@OVZW3kNWUTY!K(3Socl)0I_yAwprvcvV=3&nP2a_3PK5?V5z<@R=u&gM_M
z9?=%|*(<LkU!Q(9m{j&yvF-by_Oh-^K1PpT+^wm&5hZB!u~g7#M-{b__FccRp_{^>
z)IUFY?9Ads>&M>`;8Vdc^0r~~6&q}rJ*8=7GEwGd7E!XU+50a|TW}8Q(7$#F4(_bc
zUvwJQepm3bh4tfM$TicLSf+hhW0~-UX9<8~TGw(m_5JWJhN_xvI`str-n1x3M^F*c
zZVbi3uo7VuoZy1C{omX8@xl~%vF9hS6TBADQ7jT+I{~SEphg`#MI5-`?Y<PdvXJ%_
zwkEk*_Q6~^@s)&XceIp-lin@%yHKPkJ8U*MDc1ThKy?jg?Sc<byBW*CVa*{{>esCH
zouM7-^G{lpiW;c`9;zE_eROqoVP}Jty7=dM*ye`V-#dBF(3qQMW@+hL09<P{^Jou?
z3O91Mk?D$XAdlfAg8I8ho9!$X@0UojHc(D4B@18k@(upn@xja$)o%6-&0<)ySx!&5
z8j%40h}RYeT15I27|4EFSHY#K&F&2%y+U(Bh&fJ7Z4^W0CL^bnA`1PMcjLHaOQqEw
z+iU-JB22s#+dT^nk=X3j2!DIjH+4yJvpG-EiayF$qiur}D|`F#PR&dDk<vEG`)vM@
ziGA(&wSd>|x1SEy1}io&D}(|m9)z0SkTUn(?Be-e*V2rKo7Aka5YcIA*GJecL02KH
zfR1Tg3{=oXqL$55hR7-xN3e;pZR5~)5LqUA8+_SCUPWr`3+2x)y>JQWciX3#WG>zs
z#dY9AIwEB<pP^4f0LG3yVl|~*M5Sx3tYwrF+y);hqh({UWY7+}juEa?Z{K*gH==ar
zv+p(a8mk7;Cn<dkD(MAJy&xSgt%N)Y5`y4f+w?d!CI|skuUm9UAw5eL<3bgM7aQ}t
zr8J;x)$ZA1W#u;jIZShNZZQvut5H|o`GO&=^p~b@_I!%icE5)_@eSn}C3{1i4?9Cv
ztIPdbS^Dan+Bm?UlV9t#Ixa5z0A2oPBt6U6AKE{0%EX@SR<P1DaSrh<XQS!ncDD>A
z*?}((xs+~B4pNp2UVPThTC3F!yxs>qaNBsdtq^HKcngu1iCBdXK=oIq5V3H7#}phb
zXg6)2&9Amuo)_Wr`PLnKf+*7~TP{C^X&FP9MpkNyyn1H_(XQjoV7q@f`=`~?;>6eF
zMoLIYhhCvD7XF<Cl&&KztLej+-1+^_c}QI!ErHi-{rX~?3TG;iU=hD9Y(tq)hx0bm
z-#qD9pR7J}8~i3+Y9cmdZ8jCU#v-~ODH{iltPs)H`}bSiuRs5P@zk1^_BGH!9DDe=
zOI)Xx)g4@Vq|6xMGN05dg@-{~C>|SHe4K)>+lhVJcZRWP|3j~afTpMFg|Dp*%MLwj
z{BwPQYNVQu+__5DoqT9zN?Trj-xD7Ggt=1j_U6;^ABtIb^rV~u{Ds~~dhlicH6o~7
z9Lb1izQ(#hK5yS?>Nl95!FnMTH->Z2H>7x+MN)eP{KFKITTW`31EmH$3uW>9Cpz|i
z=3yyep&JyJ7%SdNKKz5&&r!WNnM_7$jud#+JqXDSN+n=b9jwahzhGu#-A)m_)P*=m
zegnI<U5yG`j_@_Q4i>B_Wkk=uSQhbe*U|i1uwmGpWj`(juRW~Q(iHyv*Gnuy^Q(To
zI7JnG-TaDr)niFq-H1~hhTojcdB!E0V!ev=`Ks73K(KHM;JHhYGn{PX1=TK^QL!YK
zvbL;gZ06X^OC|0?*aNa`3ryMFTRLh-RWSJg$QTL6PD3<QJ8rUN>qAaIQ(Pn`&1@Fk
zYXHGwaNQkdQ!zwVFD<L4PiN`%V?7J`_~~Q$vFHcS*$Z`qxR;Aah>Oin&V2yQ70sD@
zOr^qBL=^1bTRX<U9;ETj)}xhbYhZllVTHYR=Fzmb_D}ljTHNcO^#3?&>`W*ajI@tL
z!z#tYiihGQL>~XVjyt_V4?dB){Rls#Zcqu4Vd~wY`C4^g;uUUP=moV6xFBPf`USv^
z+OKgGVWUTN&HtuwD;Tgve$U@@oUew%v6O&FbCo7YqL8Jx4D74~avL?XPL@&|NBX-n
z4T^Z`b8k*gG-KsoUndwbxymc9?7H7_7U>GoZ_RBBS6gxK4#^(VM>9-56(O#8OJH}s
z@<WP0#<dyTypX^O!3K|w?n7EH$PRPU-0|vO(Us1dnK1mgnf*^uGm1Gb8`sh&z;I6j
z=iY3nB8^wRFD$?kmKf2N?PY6CB<(r%ct8@9DdZ>o{jYOL12Wb9(k+yY;&5LP-d%-P
zgW4QI_H|%u{ht_}r3-kcgd9;-IySHCId@&#;c<FaQTfFM7uZX#@*95@>Z3WN=|4M0
z3EkARoO0nA78!ORO{%rL+v#<*=f3i1zz}UZaoB;I7y#~>$|?SO)IZQ{S8ZP9d7x<s
z6tlxn|0z++CEK7{)gZ$uFIP}Zv$!2>`s|Pxn)b!dsy@(W$nl>j{@g8=`&0E}2_o(G
zz4ctS(2-tC_!_C6vmWrckg@<(tjqS0W({KqP$90`p-qD^AF=!^^Barx;^W9AI-MwC
zo49=zwW<2}A;q6Ip@A5(o^v1QTFR@b52i5h=bFOYYb6XQ$w<4x|2=|NEYh3!s2p_s
ztTH;jt+IGe5^&8??Vq&cm-0D}$36>o;szh?l4dpo9We4t@MrF<*L5dFVhk2+q@}}p
zeZY#r(fV7?O1Go;$N1qC^B7&zlYw&g#WzbfBp1Zt6_57;*MH%(>{|WR=dO>zkKS*j
zMtTT>6Ok4591{#g64Q2cyxqOz@}DH?7@rBOP^hf%si#b=TfFbl0?$ImU$3<?>#>v%
z5Wi!daVg`(+&jMRI+GU;3t<yrMX$5q8<ZZ46aV7#<|%dH=z+r2NPF2}^e!n2s|+H$
zc>LMsvslpIy!-j93R9slpGYv{{!V4zD>RTSGdBAh!C_&2e*f3M`XM~SWOjbb%Vdok
zwK(h0gSiKJ!hVarWrW{dSEyvJXeMIw%b3<x$rqg*&mc^p?64w1C_oxj#;UQYG36Bo
zOa?B}Qg*-p2ck}Xjx6gNjdY2-XIzpN)P2U_Hh+}@WKNWE(c+!{`vJ_xpT^G5<cfUm
z9W*+w-lZ|){^quNqOn+G&Fb@^V8UYs@#Tj0I7W_gO=;Cn1L!Bu^3@{`3APl|Pf*2Y
zGeWnFA}W6TjbP8eJZVIqO4Q;;V--%^sioH!Af1*(24X$gsqEcnEQXuX9XY@Nv|Q-)
z@~r>{{P_DrREh_#XlDGT5e6`@SidVIlR^%VpPKf9C3Y~`PsyH_8dTe!o_+!`bWdI;
zkgiRxDb;<%`68U|OQ2nsW;K$7$#2?e;fgc%6~2e=#{?IopL8MnSK#RWP*)o~(8zI|
z_oPz0UFTW+rJ~oQdGohK<Ay$<@;d!@y6ggf&*z)41I4amCvc6y%8uf;`a6zM^v4^L
zvpe)6%nV&e8nW$4zHOg3w!$&<6|dNfyu{DTG%Rc-z1$A#fjxh7fU@w^%&|<Ns&<Fg
z={o6CUEknv{H{(;61BWuf3iA}gzt-m*b3Its3Aqyr=ob+I1sD$NYg<)cyq1i!j1nu
zf1eOgY?v8OrInA+H)olo>EL3lP0*Se_?x>|1r@0zzJj(<O&6>6_4qgr>4l^SyL~jK
zSP|j+UW*W*9M#H+l!J*lB4vmpNzejUA$+N@iyL+fnp3O{{Rle`<;F)th5P44YFHa-
zM3m>lj7y>#0r7RMEmG3+E73SzjA>SjZxX~!)!J!XqV0a>O7f0}W-jKvYe)MGtF}`K
z^x~ZLHw}hz?N{Dbu9Pfz4PM%*Bqz9RRj6Fr{d}ZHkksM;-mHx3sNx-@*JydazEURE
zch_rZ%KxVcj=zz7m54EW?A*I_PyaCQmhS2Dp@#>?;N=GijR=b*)en44G*5E+ZyUs{
zhDoLx9ZZ-OZ#nTvZlogPK4MfRnX9-<kE0hPQP&C30<&t)xgN_Kx#STexXzSE#xtI=
z`5E<(q`SF&jCa*)renHnY`>w}vFv8f*q0~9QGhJb>e71LwZZ5X@kx=U<76+VlFU~m
z(ngp8vn2P#@b=Au2UNMm!4iGnMS`NmmOfn`%0mwpx6&qMaq!LC&w)57WGmhb6|{4s
zDLTE@;&D@k=LuP8q`MveYjxF)KVzN1oG8+Hv~VYQjvlQqiJQCa_jPGAXssB6IpJnM
zoLs#V^zK*NZ(a#@K#j5#R{!;UT7Lu{JKc|{0H&*xn(6S|AK)Kb<B%43FqdG&DR%ma
zQ~vc0g_j@1*Ik1?`g8f3x|5*iT<xqNc!W?0v*v?QpL{Ta>J-k{0H*`0w2C~`H{6gf
z#_`oH<DSv;ng9M5{y1QwmxQ^EnkNpL)yi0*Oy6;^0z6lrYTO%<b;|KaqbxoLx{}+;
zj2P*^o_fPR3#-cAflnTDvd|Pc#t8M#bxgHV=zfItT_0i>gfs1LV=WNz`KKE`m}u+Y
z@#!-8IJw{U=>6(oSwy>~Rz#x$p10sB&LOa&`k_Q6vZ?8Ip(nx^PRUNDF%ybZBbf4{
zhzhEXRPj;cp(jW4z0)!_!T)>rVlR~zsFnyuDxqm#-&*)7Z28Roa!!f%=s#-g>PW_;
zfeZ+4l~+%kd0|WkM9*3e6>uuCfSqM48&Lh7BqpP-|M{XzP%Ha?pz9cp_<re)IV*~k
zC7#o5u#mvi%TUToY)r*VC`$k<BVlC+vZs<SUxm(EPKi4S<{-;y$QPgn?r&i;pS#Xr
zmlmtON9RokDBkk<oH}05<X%5FYY*KBYdI&3q}w0xbq3CrKQj2na^>N!g9M#g5B_en
zqYuvEH%oCYMV#JQ>?YgxXpht-f1B%8!*f}Xm)jfKn$1?{Mg45&;gj=ni|yBV_zff@
zDmeJ}o2MJ@tXqgkgLduNWb|nP!3)_)gYNFtgVqcmU1&!$X-&5eM)W{%t}AqU#YFU0
zxFuO$>YsM|L)Q|VQMVcp3M#q7K&R-pL}kifp1*<Q@6Uf_y?kx-)vhh3@W}-uPQ<a?
zE4~=9uAl1IOBIfc$B1`x_vCBX_W0&<0b{e8+u$9a8qRPdV1-n5igd4?5oa`&8SZED
z9A8Gjw&y7t;aE>kRH5ytIy(4>V#6854F`<7r`54gw|pvcIX@A>iW7vo1l60rIXFfR
zC;Rr@MfT1Nw*lqzn>}dN(TN3E5svO#mSfsPdc*V7F~NFaP&*h3n>azC$Hf-A=a6s3
z*&r*`<Tv7A6$Pk>%l9Or!rI$nS1;Gjh6XiUD#|V_C<$O`v~_^<4@!V4p07!ozH8^q
zUw`vmdCn}GMl!RB^{Wn%&|`GNbgY=>;G1_tt)%)hIa9vP<i0}@d`l?g+nYhLG0U<q
zU3YgL(?es(w@=gH6`zR3lL*ER6RRt0u(5~YU)^lx=rXx3zY*TfF12zvYszD{Pdl6e
zaS>1)`EXOgICFw)I+-bd2GJ`UMD|1Y0ftof?Yb@;(Y+R9`%xBV)#_+@t%mc0h<MMc
zx;cA^t3m`lf|JJ8O}0p9V6#l0aCK57Rn>SVTIL&r0(ZSuV#AREXLod3Ilhy2*5%|6
z-Q5#ICa$P4LZ46YqlD2h{c}AdJ=fL&)tMPt1?(QmCuKRcV55K)9n2F0{^kkhxNgvE
z+0FZJtL^6h=Af&<9JGBU*5{1!1@L<R`z~t9pBI-C^~(pSOmq-n5u<f>tx>X!S*vmC
z#*T%~+L0j$xbfCB_Pm8-&;iG>Z)U>`j?+E_pc+cA$>jBBkM-#|d9LIs9PfHMCR<c*
zW8P~c7%Tw3eKWnLw6RKloMskrzWo>xB_1w=r~Wav!ImK`x#@J9*BdCcCjHIo2Y-vB
zFU>38=hiO3hSad$3}?ISHOT&=t96AL2QPQ5#xr;vW3sh>*1qaGs;_1QHqH(~t%_x<
z&>sX*XFHYimKY|{+O7Lcr_blSN)@*#-6Sw7R9SEpgWQX*R4O3aMx_PIE#8_-7vxx^
zx-sMth;HN8;`@RAcLNCc+xE}?`5&i8BEou3^q^0;i^d~powGTK14SKPU+V`NzJYOd
zYt?=soGS8s_7;k_|9W?Dqs51=wi7S___Y)`BwLT$ynC267(Cb4@bxamr!-s++E!Vj
zT*C>wZkdp#00h%G&@M5dP8j)FF(Y;Ips&8{0R%$5X!jbTI`v$Tt^O(d^yXQsk1zwt
zB0u9doMCpi%C&TX`f@GhrE=gC^G?yhFOCi|)yCP|YS-#0qop>$g50Wz&6(kPgU=Uz
zqQ8mFk&~g;n5hMA>or%0yi|>}7URzVt6CPi$o~4fqgR%5cgtK@`?w;2np_i4+@zJ{
z)R)ml`d~4UswB(o_e9pV*^YOQ>(^FWdB2yNH~BX1>m7~$#d*f8O5#Lg-y-F(R0a5!
zy{=QO@-g8F;5P~OE@Zr3TJR2f^4WjiS42L>;)`_n7oqb+(nq+J(qhkmX2eqzT?r)F
zIHPa0wKDZ&J-y|rDPBd%w%$zOG}(JVhw+_L)|8eAH%xber_WV|j83eaP(SWv0_~vq
zR5*@F>^gt(IpM;Mov~mzi2qq#XqKaeYm#oVX44M|n1zREUnV}f;#SOmAbp8uGiDUI
z4d@7VR&9CGKY~Al&mxJbth{WU0EyGtlNrvv3s&MR@Vs5eL^)glJ^*m@L{33YTKLAT
zyivNA;n0~#g+J}p(Y#>iSK)XCZm)h1HjY@@&TE-1R-&r5i*VD>16%Id#X9dLnvVql
z@GWv~`9>dTBVWLdy(9|gh0iKkIM4z;sS0tI(_H}(g|JhNW1TAr>)bb=$wXF{{5zP<
zlDLXpZ~3C+_34a=#a;4T@BVtV!MkzS0U-JzX+gNnFDIWM(;{(;@1*QOi+ql@ka5o>
z^_1Z^OiqaMW+8DM8Bt_fo7BY5MhOyk(c`*RJ6wB=n`{jjuRy%h_6Ga_&9A6x$O$f3
zxM_Tj75RcDf2OBy>Qd04-ouZKKu<z<7K2K+3d&phR)9@bZ09Kn*l$YrJE3U4b_>#;
z`xy)zNBVnE5n2*dt&>9`Ykm`yt&>CAeybajH=H^*t;!QPW(Pf{;USJ0z6%OZpE++@
z(H=6fYmRYG)$3gbjZ3G|ajGWZ{^s}cem(c0&Ay&OT9n|HAAzxr27&pskzw5n;>{63
zYh~rv!S4VUTny6IBAv#>1_nLY{Z1M0H_>kk{U)i{uZL%K-4X{ct};Vntn40m`i8Xy
z%v5giml&kgpM6f%!8q6XHsx<yT$n{x>jFPQRE@|t8MMZ&B-IARkGX)U{}616&o?!W
zSnuN^Ekk~a2Ye|uyV0QoUBQUYoHZ2!-wrG3Ft1UIN>n%!eNbD<z{0?6TPWB0{e~S=
z9(>ix1m0_Ac6`X_(k9u;A?>WBFaiI`g}p*!AnNU)%$AB_6~U$UL~*gTHQw5*SMM@k
zUrp*q({YXKmsq!Pcn$TOA6e0VTvemw%4Y-mYob3>=PwT%1f`Qs$tTkG{yM25#wOJ`
z^da`E@4IpQ*`VYt=WzhF6>tTqSU^*_)iZnVyJ5nL`XXZeU-oVL$v+_nnzPt!EHtf`
ze@Ng|r;_6PE5w?apn*A7gPj3-hbA&ol+LN5K%V~B_JXao%3n{mm+7-+-Sag2O|NRC
zZZP%h3!qu8@gAT{H9n=ATi!|PlQeC!FdqDin~RN7B_kh1EM0%DzEGl@y?^SdFWb}d
zg<x+)lJzbf479>7E1EX!c%7a3Xr21(@;IgbQSzg9{5^$|4(iG;ZX~iLfN#D|9TQ>Q
z8w&U+$?||E30tP2(^^c`z}t0voN8b%i|c9-^OInpPq<y1;oW=j76h7kPa|QC?prI6
zF*ARYb*szCuRe8Pt;0$!FvB^*t_Uo!cltcu^tr*}r84`X#EVmno<Hfa$BQ+}cTlY)
z-#pGqwk8he5t*vF6A=LF@i}O9K7+4pJn-o&t8>Ypd-=<5@Sk-gS%Y)7E!_b#SC$87
zGH9kEGz#`XKCD(=)OiB|OJ71i=4e+%XdN^F?&_4AJ7%T8(3+=nr`>X{02?bBqNAtO
z9<akTNSAON=c+FDDRg0*WZ-P(>T<8>)7QkCgRLgK7AB^w0+?u>&r(104jL8OQD+@X
zoecNeQNUY_m|1mC9jfWzVjdFnfbXt!?N7Q}R8m*8B2VjazxcJ;%q4hpj^bc_2%O3|
z`<#17f86vY%UZv=0Q;_H^COW`$Y;k9%yLf)8pMr0jY@#9@^O6lGcDFp*%ZeGsztC}
z%U)bBPGYj6Tj4cBUr~mEwbn_siXd_HRLL%Y`0seRCng#I|6K>_qT>b-i>`H4LpIBU
z9Jkq$(5vty=w|oDtW81g=jzv-H>td*KHDLCsam@_F(x^&UYK7_3Ioi{mx$|G?6P=)
zjjmHeQQ3-m5)fXKv)CR7$z<?Gi)H?li#_=psZLmJg|!B>N`Wpq4@FEGZWi)f>1jx3
z1%ro)-yiP3->sUFFJx~3FDNClZJA0DJt2d|oFEKDDKTL-G5Z@~%QpIUuQ2sy!m*=Z
z68SW?*WFF9RfBPvZL@tmF^{#@n<sYf)N%(62(DgkLTt}kPENZvAmVNy8Pro?FJDGV
zIFG>sZ%4&otd0C7b5Ka(pW21~3+3N8SWlI5ZwAS8Iz&Iu<x1|rlYQHkby)r6wF^r-
zSe_WTf5=Z*$h9i9oJwh62bH6NwZM#Ovo3f~P_3PGk<+TKVR;K@*^*o(Z>HVfy8INE
zZTOVGgur%e%C-e%>9s*By;PEQJkH7cnh4cnc^#oNA%pj&glKQ(O47uJ&&>9C#IbD{
zOg|$=YB{ePA8a*1I+y-^8&~RUH!&L1H62xwEfEm{xpLmuJ-R!O3|G1bTO7L>`#5cv
zq4yjd4}J_6cq}&pwR$CI?<BZ^a`qt==mYpBkolXnLULcbJbUb8Bqr#%s9m0Luo2$l
zsD}hwNxx6AojlUGR$hUh+0<ff8Es{EIHyp@WAN5a9GSoK9HY72OnIVHZP2!=$}^yF
z#;IkYC|5DjDI=)iys6-xppahU@1{CLme%)Jr>NqhSX&P9Ke+94(x_lCzqD*D>NE$*
zLJPNXym|QGj@P_?j|(m2A3Q<J4O^tCsm(`nWAF=C82mZV>~BM{l1?|-r$%LtkdCRS
za41y*^U`rom;Dbwl1l{H=q7g>8%KG*v~?LleBrX(TXz>&ANgbJ3_c(>5ocI|tthS1
z6H>yzx%b;NgAA`9NJ~=gOQKeu!N5d!beAaGD@bmVfrRyk)=P^~-&1gXhBfLD72Vm&
zkAQGUmB)nZ%8+F2Tq6f*;<XU?pv6KKx<==Tsr6701ow;rL8D`u0Jcy?0{lx)u{N|9
z@hbaglMEhx7m{0X>rQiH+jY-VwD@>k4eADHc*Fe}K{VG`cW3TyENp{By0WuLeU+be
zlXIM|*l2yw(wS2t`qRloL=cFOrPY071yw1x;y)H9cyr6((ry3eSYqTCE9f=Sqk@tA
zkwFoA=wK!N%_5z$jp*v_Gnv7#S1!Hk0=lZ(0Y%^t{cWBv9<rXV1|XHJU`lhn@`3!x
ze3#csJ5I$Y&}zu~5cXOKL5CL^Cd1bJ=$`nud#x`OE?INC7tb{qv<;;-yj0NdR$U0E
zE3i}u86=vMXyW;N)H;Q^@Y~H35{A~{27g*e4|@BBkClD7mp_<kCjqEfKNFj~sDrRh
zcp=hj-lTdjf9!=S+F1+kqf(n)W1oBWe`mQU<fLb7%vp&J%83VU;riabJ;f!h!Yi!-
zp0rOMJ$luV-NX7aIPq97y=^?L4jVWQ@q%-;_#8`876QOYd)w-|bRqi+HIk}jEt-*F
zCO&!7&ZmKktAR)jn-HHTV3k9bMBPjjJ5<p`h&QlrjbAb~!f$@+2zPl?WN}`+<(*?u
zfG2<T2GJFhlOl@0e$ZmTm@bwadz91K@;e}0CIr<(4&^svqWgic3L}P6@q%B>uc4_o
zo@g@}_~P2zxvuuF6bk}3%CFn*gdF4d0w*Y3d962V)FF;Ls*9^ZnY`Fz2{*`ZtT}mn
z92jz+vD{SI*I&?FGu_7VZZ-93Qg6u)Y20AYoBf9+pszXx=o_*Fjq-6^=g>C57gFlb
z(W(Tqf_X*PJ1~F_X?W+)S~>i%op!^TLo;7zA+O1nih5nx@)N$MPje$zhWVGYbVE0L
z-MXZIo_^WIO_`B_h4da9r2(->QIY!6`)FHKq7_sL2c_d1%8-dVAG+it>ObpIkK#F6
zNlzo@s>hH2a{0#h2zSp{m>IW-$ypG1Mv&98+DGa*>-qzW!-iAdY+q1Zbjz-H^x;Kv
z2ttZ}+(Vx;ORTw~MR|-Yfmm5|tfewdxo`wiE-S5w&J08}_!GMb^Z}-D322zp(}vj9
z6>D6p-Ymd6k9s9hj3V|t7jNsBcV;EXFtlIG|5HG1O3HfQ5$J2m{hs_MQsHZ|Ols?b
zkGM^c-jvsmPk1--VuiVMDh@zF{R)L&<6fkmJhq_fU>YsORq>(9E!NAUG<&mns(!um
zCS3mmO`-fh&;i{ykwZz2=(lVCE?`eUd(So~zA0mu0Wa-vHtt0hMfa8;B{jOZE&7h7
z;=adkIa(fbn;!xKqqWBSR6Z{GX`=rM3B<deyXQ2OOx51_i(6W~gl0>nbIQvUj8y6E
zQ^!ts(m3X5xvsh>wc&lL&O*+AAX~@(K*^r1Tj6JL`)&W0RX+PMwIWd#h=Im5EXlb<
z+xqfb;%&|lPL`p2X$z=6ofk-EIs6oeS?>FD#bj}y^|I|^&1P_Tl4_}SN*go3_=(LX
zl9kxn@IOU6*6Zx%0k7-)dpP5A-%Ysll#v176PC|a<Hp8@nYkj?)2eljH!Xf$;(^{a
zgt|}Y>Yp8vvRiv+^w;`wO0w&I(^ePX1Gp2zpsH&e%m2iuxJ~S>SoP5cDEC*VmzA-k
z<c*_tj94iJ;ba8deUWxNG5Wx<BXeoncfRW^Skg5@{8RHbuE^KC;TFfYrZC1Oi1LHX
z8bQ353O^fm$`uDRNS|is;i96+-Z~`Dm^iPbwwGMI*f)Y7ryNjfDj^q|$iU9Mbxhak
znnHY;Ze~!^9z=79HM|SRBx0r_L5uK&pVg{UFu2snS$=9dmNIyY&src<rFl2ra5r!m
zsMCOYVU+kiI`IK<y6CfUMap~<c%A;XB~`UGngz+cNZMAhVHlL_I!a0^vkp<zy+*`<
z$$qAzVu2acMzDsZoVT$_DH)=m@fVO|ixJiv>#y`7^)q=-T`s(Wu=Ony(T<gTipC|2
zH@*Vk);r1}Zo*GIcwY1d*g#x7Kakv3I<ic~n-YMg$VwtBJFUJhH@1mpArWS9S7HO>
z*yH}t*Vf5{>t5DaYv9p0w+E%s7z@jgGQ_7qg4fDS`e&WTF0S~4yYJaLKB+&LBEIDf
z_&xv;J$Fale{g9pDR`CT=Mzb92M>7qE;{pWM<!-B6&v!Ur!ztk7r3mu1J;@_%ovuR
zb8~mZc0#={7%%|5^Aol)EbmE;bPpLtO4Y!~{ONgTd-1vA6^%%gPK8wWSY_pHd|FDe
z$O0tvYJz0o>aB;9Vhgvwh@yE!6vnYrh}~8L8HbPAn<ZwVwjXFgzP_pTI$jBFu|gpp
zI@Q_QD}i>)mn0LLcceqV0gam<z4Pr-*o>Uw3p+FHFcPaD_S@8>^5lc?$12n14W%XP
z)IrYN{tInq6s_V?aSt$5kkMc0C&FB_#VzqEoZM!BJ_F8YW5|cP;%3UbPaR%%0wGPA
zhUrVL3QuYt(ey&*+45ul#5avpmb-F0G^}5^s4%>ib-^XjvF^$qpUb@wJuTR~+D*CW
z8%L~z!MsCc{5ZNSq9iIh69twnZP5-cK0eo`TtP2F<j8MUlO`#3?XLRAW<|og{BdOp
zlbc;%5}eZZ(y>uN_bX!kjaWw((xPvNE-)H+OW=v<0<5&C*{(1D&V8wDsg1+{b;6Uq
z=O-hxRQtaPcSr0tXhrJA{V60ZIBD@1Qoe0#I%j!1bnT#NO9ySyGdYVJMAo%c;(~Qh
z=c5SOo_Rf>r`S;gF(pI{EFsRVaQ(hk{2UyY>w*AWS2x<VTqVQZ-W2b`&dt=bOVMPu
z1eUKRqu+mWNEE+Vc=bf%)rj%=m<Dprr0^NK_&-?g_Uz%j`-WDWZfo-g#Mt5ud~|gd
zb?He*|Hodb$h(Okw@;6|OzuTzum46n)01sbCVWN|NVc5oeG{T}Gxq!C;7f92bTj#`
z+iUQf>-lvBdu>T09q?)5zSx-CQkJ%ZbmC9C=?C#fxYU55dkWB@DEHk;-`Ykd;zScW
zCFKJcRQp4&{ZuRfn33SPvtu3mf?|67g&u8MnqbQ9OM;VuW*az(ST!%<{~>pw-Q=H9
z7{}27BK}nKGh^`HBDHX1tvuW7&{hBNKahc`p5W^(;=K5cH-!g}sc<1Id>oPH%B2SI
zu4mZZf(IlRAiYcIZ0eY*?i6lV_Vu>cI3ZV>;KD7=R48i~Jazt#{ehG6=A?0H@jpeC
z&*zyEz}g>2aDsE;^@YG^cp6T>X|nK6Pk`Ak@LmUlP(kUrC6}}Mog&u0jSmqPu?_Z@
zDyY|Ro@sDzDH*w}f8RQ&`+&g0%Qi$j?C?~cC==oxYh!G79@*DcM6+Rw!nj{@?vh|>
zwav_O(RQfg%R+i{sGD*qVS=7>7ydj;U-Ilk+tJ(7n?J_!Z_bquW$&*K<sfMqZBL#R
z>^J`8`DF1!zk)DVf94kN1f-=%fsx(qBt<A~KknyLlCHqj&OoS5GAJ{z$JJK%&{-30
zRZ-Ej*lpZ)g!ngGY{y9dVxih3(*A<e37H3!=J}``=dq~6ywN79PNNp>Vp{?e_q1%x
zGpbLf?Y=(bk_+l)7eHJ;ImlFOW7;kSbJYk6TE$8g3@40Z>tX>vO@sU*fH@6w@9cg~
zAa9}NhbMy6fwD8G$JiohK3DBEv^FSHeK~72oGZO92aZ`?_!upfTSdb3GsRC?OSy10
zD#iTV_390fFFZVeAo$O$&Xo;Mx~KoVD(qMoPWJB79E0j_(&``zal9LD3>_|m&uy&>
z1nA`xHsHmLXF*+77k~KdBc&SO@F@1^s!uf3z(fFQ_KDj4=4x$dIx29n<pY4G2>XfD
zAo8`hN>j=|ds-;(G1Y#wgZiR$BCWMJEyNGWFfTDBq=9VdO-<pqxxm#I6zLTeDU)^I
zJs`Z;se4wj-7sHW=P6-)FO1q7hJK+q*nEacfj4(l)D{J}r)BN>vs5XDlaCvrc`kS>
zl<^lZfHI#Q^V(n)sd5HtdOt{nI^LH`!WXGdbl7#=aP+@UMXDiPb96&m@Mb<%(%*wN
z3n{G<D}hK+N>6sN5~4JMk%xG`vCshJyYT#m1}r4UHK>l^zu8sXdwmLa%81F|U-hG%
zc<QRE>V$uPTY+I#2zT8b<F*0nR7u;-n2=7wV}?b$h@0nIBxNZq{wku?4>GLY(0l--
zt42CLE}qjB+t`GBhI<!@-lr@}N!W&WesMAk>Xp}hQ)3+LXfcR(e^1Z88cjqy>!wpc
zzJH;D=CM_-aK#VHHBTtVO!UN?f^!!@Qyq44_C;_12Ck*c$d-ZpVLrR)Pyd0GEeIap
zS9;ZzG(UHQpDVuy6OHpA6~)0mSrngs2^{JT(wrrGvu+#YO-?}OKm7<3ZlAq2sbNz^
z8WNaHc)Wutv=h=~I!NsS97!{=;3!saM%aUTz2O#7<CO)o3YL#V=!2g@Os~P~_bXVn
zE8JN6B(lyhPsAqbE8<wtV0oi^pWi24(%hBLS>T4LJJJ7v%J2QNocs5a)5dHEi}`U)
zuEIUw1=;@sv!+)M3l+t4{FQK_g{CAtLB}m!myU0y*tzZ_DR<@F?-#n)e`A=0rR5x0
zI}b2y{&o*dL19rfv1l*rj|+iFmB7dCO2?z6M~rK^uP)YKuuVkFaVx5_zP)L<U{_$T
z8RVP&@2yh&Yuo5(Au9yu7?Lq^z*Qc-p)?1$c_nK6shj2VlQnooki@AteI{Hvh)A|g
zlq+(<-VYb5WQE<W?1sTU&5GIOyQ8s_R|nN)3I3KYke2*f_c5069IbnRZ_oP<4SmVJ
z-@)9M|BlF?!HoYv4<flNUdRyqs}l9aMq|4;JBgH7y)j+KAkku{NhR`cSN1d^Gt>2U
zcJ^4C9aFC4tpnNyFwC+Q!RKy8jM;$sb;o74YOlnDKFzp*LFK^sYm~CZ0D1QB?_&1r
z2K7s>#?N08WsznsQ_-}2@Tk?4wTa~Uc6qrMK#|rr#~?YP%N17Esd<u~5pfQ*L$=}A
zG8L3dB!PU_)2(CYs6iV)n>YYq0eralMJ*k?-fKI#`$`l#PCU?*xEm}tclWBK%IApa
zh6G$J*R~(4gZ*Ul#uc@Gh)*%f3YFF;<KUxcURKVf#sKNK3`S-yzW9sW;d?EB`y5*J
zA>G;bb-#P?J|w+s_E&<p*9{bWTfOO@<ls47iD>f!j6MCDl$-#=TjzqaGk1Gf1xDwK
z*4^P%zULHYok*_FQ{}CS)e{@G0=jYZWBbThDK)%nDOw{yw7~vDM|iNr9=w1SZzf@T
z+XXaYpWoPfAqW(0gmd^mj?OcX?e_2Du}h62YQ-LvC^ce_+F}y~5v>s;R>dk&tvynE
zOHm_65Uo90t(I7=(U?`-ic+)H(w4UO{pA0=&5PtklJoms-|L*u!BVbTr8YoYj-0tW
zSm`b1*;;=*ta!rQo0(NSW+JLO9YYolN?PSLNf)_%GH9EwR_f?Y8$bU^L`qw!-tfqH
zUFd{(|Dn)*eSCBY{q93;1)--vZWKW3x8QB9{#n1OJwIqEi}#xSj>sS8?4CeWR!&Jt
zIx~Xs#TL4nP43wlKk{Ju$Fzb}OUHs-+9d<OyoLAPG-rIXGT1n6V(bYpn5n4&v3D+%
zSNn9&shfAUMXi&x`W%n8`yQg})a_l^AH8m*kJy(t)9O&N@r6RQ#_ypc4B{J^Ysk#$
zs@!Rg<0M1Qevu$`uD*=&19&RXq+EF!ctzeUqn|e`T=kceN^{>Tr|P*As!reG6`g^a
zb43o9oiS2M-6gznrA9p3%i1X9KS03!$T0?%e1l01k^RRfh>R)1qVaaKcK<M2*&pN2
zzB?5!7Ml1KXbse&5Z4KaNt35-u?m4w#4i;ozK8`8`hr6^G9IPX3p4~+COBTq15G9U
z3ZNf6ptqW)VHM69X+hL1GHFWAgEzI2k-nvBTh6uEOxu1~GH<f8&5#cvyYkyEa589J
z?cr=c4POUbAexj*K5a}R!?m_4YcS+PUMz9mWQQc!Bcs+Q$QLk`|Loq2({{yJwRrZs
z`Dh`dK#qJ+TT^}6W`Fi=)T+&3`ws=T1wxh$sipHOShwf89Wqo|NiSD=sVekV;+tZ?
zx5n=c`?&+A4871aeDsjokksOUh@*VHxJXb{<4Cqor`F!KnExmafz=BSm#@7vY82@i
za6p*NAB4b0S#EX1jAyid5DhP8C0uioY424tZ3Ez#0;NzGF8zLN8LOvEpi!W<2v~+Y
z{{e25rG-HjisH(i0%wJLeBShWYcEC=f6w$Pu*1!AD71IomIEPBC5KDGMf3t&d(K$s
zREs>k^h(p1K~VjAd!qyO8AQ>86Jod0@*zU~b4!mw6Zxym2iM;wG>|U($otJU_nw_2
zez(%brp9)>n|GmJt+4!Fk`=WHW$?{x`HM3vY8u;K)B~5;CRFM8{wqOPfj%prtdZz4
zo{ikIQLnCwKx(2=;gn}`KuaZU7lV`#`sdG4vFNCGr7FTvV&$QTPA@U#>~yI{>hp`V
z30?u!m9oY4+IHSd?oQO-6-U2Id(LksK1meXB0%_=erjoDH6HuOWWdjq()oJr0R+t<
zuVwE}9VoQn4ia)$e%^Q8T@2ZMwM2&&TUkr138(J0Zx!*WU`}#BiWU2VRpz<nIt=uW
zz+DFVpPP1k!)s~(*&HTc8-R<H0GXtp7J(4*=DX#4dDg`+-+Ac38bMF2*?%wD6Zsr;
zvffE*4_kfj#Do!4Mm55=9pI@%1{p)A?c9~->zsEeLD@#u5AOHB*DeTJ%o(fdkq|kV
z#;kpVq(PEw6wM_Qu2h^!Q4@12<p5Gyy9AHtJRIWe4tsy)=(vEN1ozG41k!~EX3eE&
z1c_adrWpVmz+OAD9ux^3dZM}6_BrML&!$O&=<9vc!2ACJY+n?0a0X^d9*qMzad=~C
z6lc*lI^d|%R=p~T2#7Zop@;+H;PG;F4^Z&48DN^ouX+%b&Q#Nr>I3uvos#GdA!aso
zBM%RQ=1z_O1MvL%x47bkwAJeDlwx24QtnDx+5AvyI`r4)rw_67#nqQ-VL8-01NOpW
z!pH6>zy1X7k_Xfp`;Fy!=$-iWRA-&+q}tERUzBgQt83rj==n5O(^m812xH~vqcARg
zGu!#T4KE0WjM)@13rwA^bv;7v77Yl$*0cQdrYdl7NNPuybYPziD-}@g6+cz;SQ5;B
z$+jqozD}u?!{u~O2lo8tbP=}Lb2r@gRd|RC-&`a=--J%nLjT*em6XMk4$ez_&QCz!
zE8>G<jQg`1v8G#0f>=lLam=c`NvoZ5xR$Yx$iA<y;~aCXb>BWJ&6+UCI^|u}@ME?M
zoYT)Y2Q=w{bo~!=V|n-6Hym3(nSIkf=vBc@vf3R@El&==IZm&yYg&xyjtHhLpB;gp
z2e#ttDYI)}O=;)lhVgMV$Lv`!zf{@9LQ3D6W}B$HYS0>Q>rhVD*Y}@I1Ijt*>~Wjb
z1#4nYp(~7&<Y1TThBWjRnKirMdg1^`h=v?Ef2zTG24CRm5_+yt!{?_Vji0D}bCR_|
z2m(m&=DSUBaQVJIp%x44X|}q>vu<f!%zdMKRik>XuYZ&97SJ3%Yr(-mK%X!O`hHpL
zwB1^Mljj35WNB{#wLG}W)kn@}PcjkQ^<PrE<yw3{<W0^m=&>X-8gygV+V4Ct4_`=V
zs@we&_>O&{^LKG&E2xi@JEEWFsyqMCoi3acxA1uYYd>n;4Z)`|?!obtESF&;Q+uy8
z$!6KpACg6Pn+vR-{$}eY)yeigGrz?p*~-d-n(up-t$deZy{ef5bGb%2Ixlb}zNJ|;
zxRht|JDkqF!E^S|(Iz14J$o{qIBATA7eKBP)b|j4EwDN*<&UI~EmpMQ__!|Ws_f>~
zhB=w14edh>8ToG8%m%DOEpmn+@2-LlrAFxwAkE;N+Wu#0YnNdg-qmZMHd4z(IK}=b
z_k8iHm32^3*Li(J2ixRGb3BeM8})2sm%d$Ywp%k#Nr1wSwh;H#ykB-k9L+eAq`^@c
z*1Nm+)ZL6&e!jHaeSE?Ly?t4$+%1=-x`~q`N{ZH=Lx}smAi)MWqOIwgO&%upGCD3N
zX#3zjRRw?v-g%c|gW(lSvWYoa_lPG}`T6(&V|?Ob(bk>Q28oIa_VB>JCD9d8&{sl>
zcna){#97rAfk&+nn$}QvmqlV6jh2;eW(U+Rx49>+n`*Hy;G~1u3+{JQH;+i%1!CK2
zy7)YUmhoGGLa-NpBSl_%d*wrILLFT>1A!-E!rvf&{KZ{vs=Wz_%K=wo-RE!uQ+zuQ
zuW2?te`5J1rAu>H#Qf4Yu=b{-)3(32YeLqx^RkTe60^*0$BrC4WoA%NbfHourC3HJ
zhl~1rplL5NlFoKT9>D8OJ)~`51t`va8CNh+Xv76k9IW=(Gr=}u>JEQQ*+kr!RUqDA
zv#mAAmEamZ3pM}(KH|+NN)Wr*MUk!h-|^J6ov3Ij7I|kytpg7~Ca^#>AFn!uP5NPh
z>UuA<Jqr<~t|<H{AcmN?BtcaZCUlS~0I3Q*N#Sys))W3ez289ogdO3kRhM>-Ts`il
zVKiH;RIhB4S7FZSKXU|Tu;3mwK}&EaFa5qQ_3SpS5;Y><A`fdY^nE>XsT-ikLd~V-
zUCuf6=>`bL92ow^Byx=E{S91CdG;xfla4w4X~}l<e*k&d%@t^eXgp0%$58fQ_Pd6L
znj2_)Wi3tOMpY5OXU{+~?=r1~P4(9-{gaQDc7ew1SOU^OY!_OcwN)>Xi(;74dtjMg
zVEkCk@ZU4s{gepL$ClH#d3aZ;Q;L`j`+zC?q_X+hjO7U1EJTiEzb^!70JNvGA^^~j
zBbT_y55~?f7WsP3Xi2G0TKOF~p<FGm#s61QH1BaxQ~nFX<G{VcxApzN6v(3U+BDSw
z+N`oiwk9>ivR+;%JXOhltHcq}70Dy9HrLe1t}!{!p1Q&M^Zu7RfoV?bFE+dJGUea>
zwFFl+_0KBr*W{fb6(&Bb&j)vboP|jbHCXB~Hnal<tXO1X&n}6-?;h5}U4Us`%+FJ`
z>WuAKPjrob1?U}CkHbdMPmojp)}KhSpLIe+9uA?Z-k?8U5Eg`7MYFvMjf_;3+a&yP
zL9-K$C3ia1l+;037yT4-KQA(|lw*uxvF>H3dV3P8ME_Kv>!%!7SJQQ=yj+w}iOyak
zts%|4y&+#i*2Ca<b$I7LfL~ARjbzAi=>2XhhelF6s2??dGIfAD6OS@2x$5=zBm8oL
z4%sURIP{N+=Guk5qM;tFrd=AD^GPJSj}f<^^Kz}gi5cneF8k0J^tE?=WB#N@cl1?y
zeZYj#ctw&KhZVg)-y^wa73}cdw}yK0LF$+8$54zK8S_7oBgU!HCoe9kc_l=RA6Nvo
z`s;~|Pe|eYSP$%>+7&dkMu7Fqy$W`Q8C+7#QnZL=;VObu@Xm6(6=sfg>v;35k-i(R
zn$M~`*PJOo<@7zh*~pJv^LNenfMxgpl{K7@hp0yNwwyS%7d}B22dOi@f#WB#!akci
zlzBDhlNH9x&=Jlu89b(Nzfsie1jCoC{F%T`O!znP>sowm*{I^0ShVyRDL+Sr<BlzB
zR^@FeLNS_tKnhlFU9M<A^67V6?EYa*=@^fqtu(K2u&T<A6G6`AVruGLGBD*FQ$3*y
zBa_5VIG9Yky|+<dOa=1D7|T1G2KsBDo%5MB)ckE?W%^{*HIEA2EfLB79GdGLu$j_Z
z&!wAu$GRN|`)=pO6FbjWitA;3R#G*`yi&G(baC@=Pc$9sa6b2%*mPOj-TwfPe1g5=
zxRr*ZbA!(gJMY+_?4_5D4AS`BrUTohvMz`3-;ya_{{e0|G%hPy-D~`ZY{sRf*Fq8i
zY@bbCv!%_!dH0sDk=;xI;jyR9Mgp-ovx`}JWxALH1Kn8$;k5eEPoFV#kSE9IKygMI
z4PJRP!f^a2MaAS+<#(hu?s(#~CP(MdqJwXcsSm|}ejoqSJY%L_(==yc$<|EO1php;
zF_rAD0t-V%E6HoAy0+x?tzuWIBa#zudAQt-c)_;9=)Xf25XMia5Hlyn8yJ$^9gEA0
zN|Nzue8Q%C9(k2H7_%*QYq{mXH6uP{BSJeD2kOk@T^HZO6VQn>+Gc-m{rNzniMhM}
zFQ=a%p;EPk6%aTu|2l_D*;kBm9fMPF7Q(lnP~9|0EM@XGHx-!#bRM$t#&|^#Q17tJ
z7*&JG-fL_@%=fy^207j~rz!O}mx7io5*-r0*!OMWO)raacrcf+X7ResFH@^_CON?%
z7r|M{Cu$HXoh#aj2V4`%qPitR`2hOaX5Rz2aOvd8VblO9UpS6w?h<O!;v--&H^H=B
z04sg*gHoK4NY4GP@-f>YK-HSu(b#{0w#0N15_Sd(=(?gQnMV<<YlH%@k4FmuKEO-(
z`{gY!L|*k{tE%VXSnru|C;rGBs{_F&mIm!JPWGybDz&L}Ll?Yw>!@nan-FVzx(@=x
z%{*oUkpPxn9Kc~F@OBeXId)9U(z*gpXU?)%q5P~7nkM&(l;^L1CL;-Zxu}4<o&FzS
zTlBHz<O7<N&cCXsI7lX1;fWH_vv9Hm$G5%xvX;J9SRHWR00<ZG%X$|kC~<BRwsJz*
zyV&X9Kh6dpJG{0H&$?>6ajJj%DbYm`eeEx8PX!rYUVTSrVVUnC{NvC5{|hVQNXL>E
zZe&&SXH#<!tmLgp%V5GaYCf$c+KyqpLk+C|_dHk4Pm6D!`6FrBCv%Kr{H%GVw*rwb
zQ>5rvk<(onNtd<R(v;^euGIY4vN6`Dz{6b4zVvz$_PO<5$ciepbs1$Y>@_;)VY3zx
zft-PdICz!wZAM+|YhEk}<PLGiptH?wo_nVpg5^JB_9pW(Ii`2+>Q?x*eo$IJ0a!+B
zwl?c=kXo*WY)PEBUp+Z-z3o`9a9MMxY1PF;-ACH&pXjGi+|mVyPTnkGknQo$>MQ}u
z16KiYPC~<!s8BYv<`LcjJrRKZfLb$F;*uWK^=$4(<&xcd)!C@94~}<DefpL{1RXmZ
zKZh-*TJ`b|{IM*382pVPI7@XYheMyrI{zhG0AKE2{_PNR`Tgw(+#-fP!7Ds||AOcK
zSBj5=NMt3&XNAD}if>F6Z6+KM2BWwqzBd!_>janZB}X>Gtjv=US??`fZ%cWn{~P(s
zt|;894(A|IXDO(AdVcXkvelrk^;n0e;&`WOm;tKH#{4IoPDF37?GdN>7HSUq@UxQo
zdd2*~4Unk3{Bn3UMa3^+{@7gZV=`QErQX|P{F5o*;%@7+@)uLX_l8cJ-IqNS9V&<?
z8V+Nk3;E)ZuvqejuL_M&x7k`Ch;x?xeL7oSAMlTif8b+iQOVY4|6LZ~nSOUQf5>m=
zU(K!)6wB^M^u8n}KjqC94_RToY^a9@$-cem1Hs>du{}*dvKG&!@Kjrf_?q&mmoW*l
ziS$+VaFU6pDX6-9w4YR?n*7SZ(mlbf729FN`zcZ5<b<HeV_&fRvNX>y4T%)ry(Jhi
z)PhlqMpyn28tCGTE)i@j&mVPK<fuItLX#%_@;C$S-$QPYE#-Ym+#i}UDr4kivVABn
z-fpxZnXQ!^D?2Y9)pplDAD^e~wCO=2`X;-yZljgOZk=w7urHO(cyaL@d$Iz_KXF^M
z%4!o(!t$TJANh9*uhmPGgvwq{WyuZ9ER|M|A&zE?$Y0a^Vjt9YmndJQrIzBnU-B=5
zNGh~&&%+O(35^W&77<e*biT7_HWD|x{vkIIb-XV}C-7%kcW#=+5|H{AYlNn)i#xMS
ze>u8jIWlWTg61q{7R#m5*M5@Yl^;af;RGF4*XPnF3Un506@z5YbR-Ag{Z_WT+LMy?
zi5qK$|4=CyuKYTg+ELLKa%nkn3h`hN#mCu{Q6cn<;YTZ8fixw>Mr95MI;KX|F$>fI
zM+E9^yLrI;UV0FX*K6E=KN7wNNx5eA(%o_rTc5JS@0#jS>YGh60=vhIEsA+77PP<N
zoV3w#Wxa#JFBs%+`J0E9(KmiT?LD%%+>@7VNFQMfsG4UdMd~{eUYB`0dH^+4$WKKp
z1Btt8)QDnfy$E~dKYKvh2YIL$lxN!byL^N;_IKr<SiW15$;oH*bX;JPPVh*oi+Ke+
zlqx<c*!(nO{Ho0$NxMpZIc=;ejZ>_u=$kn}s5WJWtf->z17q~U8^r2i0GsO#3D`Yq
zBAwPBIg#lIgOSlV#{`xJ44_Wyz*c{rU3m=XLz_V(FW}J^_L8hKZxDa<E!Vkour`M>
z{bG{Gp15NuEIcC7+E{CgaC2G?use4%`<fJ)f$p@f>J(53HV`={(u>AL5?OMR5`gez
zyul68u9uUEX<9o)@rUzSxU)Sfahh(7?>f-}aGprJ%=?G)J8gdbf6-m~U<OIiWV46l
zUq?A0#Os^sOo0FgLJ7TJ7FA&BcSGM`)W}`wLv}l|k4q#<L$xm8dF9(g)|<oB5(z1o
zv)QH3fY$76vIq}rPr1hU8g|yjVxFtsL%#vOnmDQ=Pk%w*(f);i&!9dMx84FRqzs`D
z0n0hJ4V_Ub=g$^0a*sZ^2PbU&1m=qCk{F$(<?}2*<mGI3Dsrw~csFsYHT_-r3$btt
zA?+iViC?$Bfomzkl`pd~f>6J$@DT*-JROb7(M+1skX2)Jb6ujQ+baQDKR{sws@@FY
zeIkJGGUB%3vF2utmq+AU7TRQ3&20;?&>IfBr8J!HyuM)P<pQ*lD0qbme|YZbVw065
zr1q!9>5%s3CUeCsAV*5|E!v~&u`+}apG9PF1Q;|uQcrp+3T`u7UAjJXWdyX@DmS#T
zXnIvjizHUGu}~{0bans_RTxefUQd&D>&*mAdgAc?JYOSR)m_Q9Vb)Q0-muF>*eX+3
zHw|_T{j`4f`316~xwL;1(LNp`x@Bi*M#I{S6WKk+7OThyrV|0BTDKkCT*f~P$q%Ui
z2N<)#@RKULz*C~*nY#Xct&g6E3Y!OhPAr<-@FCx<CSQ^$At}9?GxGg?i>GP-=7F4h
zR`q;~6Vq?3%O>A-bZw&{Ax#Bs?07vm4iS<FdA-2*R(d4n0zLrpC@(m_)h(!bfQXW?
zvoy2;+5uBxGivqhlf#01bB6o=aE1NM8VIYd)3MXo)ZMw|Ez*52pZ8f<=bw^qBit$P
z1-?z%5S`~`7&($UW${)iAF>!6;C1)+wFBOL!r#F@-NQfMY`u)2p*kJ>oT}<aCe7$b
zKBz2$<XL-Baq&bEe2Z@u`?NMmAcY#*3`-l|4&^Yd0ILxlrX2Rs>r+27j!=I`o@FvF
zDJtIjX4Zc%eoXPOVM`1d#B;T;eD4yG0l_jOs7dmb)a%>ocF(=?+BND3uF{&_pUiMJ
zWyOk^*1iiYd)G;nb6s9k_%rHWE}X+k@wd4ZQhGBL*j`n+_gr*=>M3##<fVChp0)w5
z;Y*A&etkitnCLr8E#|Wd($c54WH_#>i0*PlvI&qb;O{!7dV7xfhIrx!YMpn)7<58L
z1Rz!)yqw*RhWi>vgbyDt63y`w$I-*y_xwsj;LNF)t$N6pjBR`@vaj?tPs<mOy381+
zkO|-9dET6S1HYAbEgLrgV?D6Tl&AI0(qL07W9ur}M(dk#!Y+?85Pq`)$@~RzO={l#
z@4|UXp%YgKEnzWg3jY(pSWO^7o&NVu&wqeYzo0E`O)S2G8}<49!(!ri1FRfajc0W3
ze?{Z%&ZrVrXovP5p*2X+Uz;e<C-ZBAChtFE*dwjJM}F~;Efsftz0*nhN^o5LkYiU;
zG%jDT3#;<1=SegjVy`Wu*_0Jobhe-)Bl&-{_$V()r#pig!Lj4HNyBH%RW2vZ;7hUk
zJB6D67NsdhES70j8~-{o7d$$E8EqEor`H4IHj6|b6iei==+4Fm2<)rQ+tcHcjkygn
zd<Abo;+cBZvo0LtSVN`L|CQza4=`vPj+=A8MEhExgLamHARL^%VpiFZxF0~*U#F=>
zWU!dfg!(h&fD?GfkYT!KM244@n*NQzG=ZPMk9qOUZZO=BUBX?jf7WTS6CYjD7sBaF
z_KM|e)Q-}T%(zER-ROr_PAR~?79%!r0(_Pp?(~!mVy5=*UYmx8{ZARzS2N5VEdz?e
z<+CeR-3V*X3hZ>?95_cd=gJU3RpMA)6&}wehLHv>Azy~YEzGG@ZHGU>)x};a(9NmM
zta(+=N$<Z{zJow$n>|zI?z%5q7qzySAm{OVQd2<0u({{^p^m_|P?wd0XtB}h{XJ4P
zXJ_#1$CkaldZtzi>`Da~Mq6!1LvNGp88{^Bp@@z8D~S(bU_l#WUm~glN$Vwqr)`xC
z%~Si^Qqabd-1Z$iEoBHvbOTVY#dV5jQ<y_>Ca_yt(cx8q^6WO76r0a_7JP8BJME@r
zl5y&*4WY>Z@7F5YyL`U!91%s9$#aDAqVp(B!8j1~&FOET6Puj`tR_ieG%@U5f__@t
z$ooG*SCuK6KS|2pJgMOWe0VCmh7<(TY$-zAj^dj>^LyhRh7(=e8%`eIB&KOrk76Fh
z%kg-t<i*w_O(Q|9A`mbO4L^S{dm)4=^kH@{H)+j^NX~K&YI^>f?JJf#u$Znh-^|~n
z%WbCF($DC~RkWacZvkfgjPCVf{K?z2)P7T5I|qtAKiS;!Me%i3M?x>pV)$}EKaX=)
zmWz(!$ZKVMS^y2|3cK#PM(J_fvyY8S4ee1+p2eWpg>v)O7hoL@NAM=HceW%f=gqrp
zR~{X9UJs?B9#9Q`i^;=0Z34!?)<UbT=#f&0xxF&OVvz|;;G&XJrPEDMAhkOj33^xj
zUO2~qG<&bRNMF|3KY#}=w7TV00avC#jnQa=YEDDjs6zg?(Yn5nb^=vyr|GTm9$>TG
zpZ0_2+u0P^nX%&#MPm3yaiU!X(R|a9+~Hb$(evpDn#%{w8F0aQT2Pn(v6SZ&_&vY4
zkxr4lw37Yj5r1AZ%0*9N;=s`~@V&D(chp(+jlj>!yr2u}84cs^nrZGxEr`fbCDd@!
zf@v=f<F%j(-uI^M$gua9XK;Yj8G*l;kg;n(I?Ob)D&8A6#|#$J7*zU(t(cRJ*#L0K
z_A1<WMHJE&bqJ7|JRCwfjvn)#q=j^l(y|JC0r?FCQ+@YzwG3$q7edn^2X^V*BV&i=
z@W8h$gdMYnlz7Ok+Mdj!G<5_)dV+EnjvC)-I@E#ow_qh;NMst``d-MC{*FxzN9=NU
zh|OLHQ<-q<?ytZHy0|5#DZ-U$Jq-KS;=|SJW})U}=e_|zRE_-80W*=qN6UHP5MlXT
zMaR!U?dGfP)Bg-cRQEiC*X1Mh=NkvH?;9dLlxZim2I2Mt)09K{y*R?{?gss@)h;Id
z{8mptwn#jVEA)W6II_)Dle4>N?N?{B-j2A8e{1WDY&av3>fZ$(tu(mviCo@tHbME7
zMG+5K+aCD6f6^rsmA@?H|KkmB0w-iaBF^C7g}hT^s~^8hrk5xF&;Tx>iqSS(zULh`
zpP__lMVV};;^8%wpy);rjl0Lerg>9?S~gTTyO}qv{~V64;E}shz4b&3$c2S5UF+wQ
zR{kPkrh=&tyXyK@#YYQB*;|4VI-kyddbKe(cmnhGpM~#=|ExAjEL=}VMP1l4@^Te^
zJ+OP1dl3fu7rH)jCZ_p;WUZ+?yXvz}ST^Rm=deUg%m|Am>KCzx#{8&jq$cpbq$c4K
z7ZU^?XshA@EV^?UuQq5OKEW`kD`adurnfiw;umcm)McQseBfiR4KRaIa4xC;0k$aj
z|94GfoxSRSUmUrNw+9m$66z@d`I6pQ_VnIaZ~#RZ<LcE@2yA6Q-D~Jui?`^P5H$R`
z+1}g(P1}jY+JM-Z@7@fHz~f&VQMU3k5dzXbTC;8RuBzu8rZdvbtq*)PFOj$Oux<`j
zH&S*A&5;Ve4|KW|>OE;V(M;RXkMDy7BKI*J$_`5eMg+$Bg}Gkf!xgV@FS-M)E7Rub
zJD;Ec=7G!#iEj;XSR{DXX}s^ywwXgyoJ8wTy|niMYwYDk#LZzB^Y}yf>jRly&n3XJ
z&Z-eR2E&rokreiT)RHnMyrK^CN*MCq>!)9<=ChtV?s$1iLR8&NO3Zfo;D-g}B*Ltb
zo7LmuKFKblLTZ0G`W<I+upGqHdRa1Aq@$yq;cMfKxB4!8A(@hE14`RHf!oG}a@w?x
zf$2K7QzDWDoU}I<bW#JBLHAxx)|c|n-V49itv!&BFHukB1fEzt!*1r0=VLwoLipM1
z5~n!m<C$X*bNEJSs718OF$nJIGbxD?3mZv(Nro$T!d`vv8Ss89o#9JYmOp>L>vbPr
zE!vf=ys=;^riPR$l@hR=8l)X}C*$Ls9<DT9UVQHT<c4Jl<GT|VFFkoJ;KQd2BW=-c
za=K0`RdNiXr}Sr50-HVV{MJD^XQHYuyoTf5a-bz&?Yk~QByUqx6sZ*dIaP-jLMmXe
zbDoMaQ#Ti}9C**eNMFilu5c{DN1L%qM)Ouv@B2@=-eNLUY;84uClO&5OU_Xrku4h~
zNr9F78wyv*0zEp%`1;6m(1L=R(%SOQmrvJ6Xaxon?hGr>@j2gue^ir!8Tl_lZLU}Q
z<C8``t5`|ZAqC*u)AiXxJk=#P)o+eC%la|0heNEm!(v|@tTd(?2Je0+jy~j%2<diB
z{Hc~RreI6x0It{i2h>jm`^nO-&5{bjv}r|5lkw9r&EG1g@ZP+A^cCAjY)Z-o9JU-<
zzZs?g2D`gsf%h0g^+<HO6@LRzd8$rySv?jFImmM9htn({-J}0NXf>jtp{}__h>ECE
z$L?TPWC3l(-p(ONcCubgD|C2v#8-<HymdHDfX`&jIiVwNnYzTP&f+^jOm699zmodW
z&26WB`IwPWlL9k#zs_Z^PLIC(O`lp%ffhO4lC+3dWM9Qe-rphk7lBq^UT1N$^Mod~
zNBMVz3gTRtA=<9%kM?~Un)(&Fx>;OByI=1K#Tk-YMeCr$Ed`p)x#$}KCKnZB)y*BZ
z-u^{)r$RF#f^%oW^+PehXWtSCbKX{9nUywrLY_rj4qdBIrc2(lCm*uBD<og2>2Hb<
z$6_^zQPu9Z9`FjG-f_Bdpv&){Yu4L7U!3jkQq7n8yADa2NuCO{lpl|<|2Z>gCD#d<
z3O4jcXYt?`qhx$wW*=osU}|DN++;g`^xSw~pqO9Pm|O4m(A@OE(zz9%ExC;h_W5Q0
z2?3wtuWm(sfI7e?PdX}_&BefNOtH7*R}yz>`2k8lzrcMi&me2EuC*ys3-y>J631{F
z(ew$tFSzWGs#Q9(5)ngR;AWpt+%Tr%tnB`5C`d?Pj1%*oDIUYFI$q7AN?_ECJ0spN
z8ZlPX9e@`0+8fEmT_)+xj*|7`sOk}k`Gg>^D1l*uo?eqc3OcIcube<i{s%^?KNPUK
zwVq4?Fnnz&nWAT!3HB}momG49u2%`LQ-wqu>n*hori{Tnba;UZUi(S0%i;3Wdl2*G
z(ml5~n@0JwXoq}v)LRtZ{x!jyKddf>wxzsQ24+usn40_+Smk*}3KvP9678xAzo72W
zu^$}Rs6J^T?a*r}@?&=TMSN3A-HzOx;)EPm-(ZXeo~|kntzaF6Q@&sNHq7&^SM;Uj
z6rphME+Y!v@5yC6?_&3(n~fRs8WEgnwuSu6z1U)-{z+;v)AEy%y2`fZbexm1Vmy<S
zgRPXHAH}J!Z!#~LVKD)$>F1SmR>O`qFHtj+s7PRK!)=C7$9}PqP`~w5%lM2ekZsbd
z_P<~*lSPd<1v;yog>|UAQ^-It2t?kH>?#i{+O)>Mjja~?Z05a_pBCVmQI4sHT=y7K
z&<Us!4@P;{;c8)aYJ%iVD~ZCYVubT*BXqkhleVt^M@m;{sV7f6C#bmgU9b<?Et_Ie
zJUb`9IwCFo^T__XfR?U<=Zzs}C#+-ZS^~x0#Ux6>VA4)a$1DApi`tkB3H1qCH1Kp~
z$Vrm$g2=`HGmw2i83Fo1bZy?QVr6c~F=NrASY5vAkW!H&+;>**RW*p|R4BjYnqwTr
zK-LJe=hq9gGV@(22!9!Cr)C?G_u}jwN0x6tr_$*n2$2G8Y<~Cc&tgV;f0P+d$CsOn
zLIJ@|OnSJTrprINV*w;jYfXx4>eC=C4t}3J=hQOJv*{^CUDOS-Idn?|qcAl3tn7H4
zDB#O3kA1bTwph0k;gp@}5NOc-*}inZ$AOE!s>K8AuUihm7n=t6XsvHG9T+5s7#*zC
zx(gX&>5u9Kr?dmd@ziG`78Qtec^`|PO%uC>t}OIu;T~Ml6TKj)+|eVic>XnoTG$l|
z!R`>Wp7p*F#X0kwk8eIt?>{7`F(l^{%V&`#@;yaOwZge<S!f4azjeQ?0&UStXN?x!
znZCMCSJUL^5@-*2H#Q%DccBnuy<ZcR!{odr0{Pg^vl^nK8%<^fvR9k}&%_cNB#pw-
zKXkj?Mtbwv?(S5?%`)IbF}4~J27GHd-{;6SR}IIBj5$MuLU9E+UG~q|J(vz|j{CZC
zWCGht&+qY@Y-~9@f1j|)8$lihMm5%W*#W->#h8qSKNW?jEOeMD8bviV+ypoLxHM=@
zd85?==|g^VzNS&LH)*MZ2l_Q;CAu4Pgt!&94)9Ba-S5^Y?-_6g=Fe!q_?i|T(^Te|
z<#a18yi}PXYhH(-IwLtCMVeAztt(R^Y5bm_lo>VN&?<7a30!OQ6PV`5e>=%7uIs=L
zsD7lrTD`OZa;k&b9lKOQ+IIHK#i@?Ji^Zvhzq>`>6(7!uC<A`ic0PfAO~`}n&EJ)c
zJXJxe2;Z?>p9^ssCE`~)yCPMIZBvn*){%`5p{H%aoz3H6UAK?tQadi#)Hs-aYhSPh
z8T9GeGly(5u+VqB{<GXRaKpHyW8RI9Ey8Usiu~5WBS-a%%^xrj+4KA0i$_S2=g3RT
zosmdwIeruIAN5xc_E0O2)YC7~TnE>ssD#{b)L2G(y_^B$92TziMM%-dE-Yxe^hq#o
z=?O;;P{sCEmAibB_Ck>=>C3;s)t|IV0M;2p3j>xVyy^jgQ85M@mo-a&dMl=0tX=}o
zwU43njXfC@=WGNLBXWvHv@G_fq(tTDWj_V#@V~!DyqN1r=~(^p(lNjN)t*U(D2ANi
zBN_(Qxq@NG{9Z3g!B(7B`$Sazp@wq+02gcqTGfm*x(-z%j&v+r6h$pvDqV!thYG*E
zu^Q>P`KAbJO;ILYy|=d0)$1X-^*EN)Q2OQX%T892CfOw~?QwIqwJ*B9c&l+8$(&Ww
zAlcaW8Bq|g`E1Z8)`znJ^K#{S+}w`r+nZfti)9xNY1%2Dm$h&Aci(IN@$Ur<uSwP4
zrkTIzg}<4v*B)xwCw=V|ZqdK6D=XRdu0;|IU8gfy%4wXlzR)wAjxttIF8W_64K0|q
zG<j2f^(XB`<-QGNH4#f#z9A?a^HGPurn)0sKjvk&*0P5rN>==};>9<OMS$Rf3SBy@
zer+=;C{8^&?!tJv6?-wVr^mex(iyG04t<OIILyoa%6A97DX;brQ(d8Chs|W#O`hWY
z%`-KHRvCRCi^P@yr3fVBpDvn&-uc-iML*zBF08(C)q24q;8LFFtf?QK7sze<T9}Z1
zPGnS|n2h!&UQgZU{}i|)-axY}IKSaPkI|xpPh2gp7e+gMHpfz=6LZ0P<23VoL>0%i
zE7n!Ysl<sH+R-n2+SDWS)dpeveW^?&kF4H4D{mih)-*p<e1m$P*yr_&dYvAuD^z@i
zZpw;};=D2(Vhsz%%^DvoRW`Dan!k@+mOmX+lQKh(?OvmG1$=RO9}dJJPE_(n;qIDE
zt5-Mz%4LM{X-qN~i2o^OQFDar&*a@k?hf{diN@;lRQqgq9Y0@2L=sYouaF=H7VlPr
zxO3a@z%x}|WB#>hEk|4#F*TL;r?FmI?fWuk4DGtJ4k535`nLKshdt-iH)bOh(w_`k
z*5UW-<hf<SO_4E{M=3e6BRG;Qk$yliB0LQ|*@Z#?@!6@usfS8<%Xd;$b`GjGvScYi
z6&0-Yv>ji-ZlP?B{thv;D8;E_v%OnTAFk@Q^ov93AHkV5|Hrt-$ARCu99WZ=s0~b&
z7!K~ldB4&FI24xM1+V+ls3$;`W4JdYH@VX$onRJjs|Rs|%~bY|kKpEStH;wa1{6RQ
zYdwQj%aucP;=wx6FbSLyOrB;k!Z}{~H3}pEuANi^{sXiEu;xE=0_gj98dAdK=&g?f
zV-A-72h7-8Nw|L>oKf{Y;+g=t#rf|>L%sv<(QG7N6ekmFVB~G1sKCN72HAoC^)5^D
zmtMmM$hP>XHGRc^{OUVl$BqS(xmBYHL!3o5z>>7iLT`S(R3((senYoIf8`rxtHOMQ
z!~21EYG|JaF?n3L=%9*uAN&(>_Df{Xw+;BoQ4NZlOAA_ZE#t-wYU`?*y1&T5@LS(6
z->Rzk;(B>j08-oGHa6S2mfw;6b668&YUv~}{9&0{mUrh9abY_kwe#{Ch`e&xH$*e)
zeC`@z2cNbPUa~1So(L;90{?bT`S?**<W|7TkJ46x<S4Rf<&`mYUag)lpI_?^F((y!
zybe?M>yZ6DI>7x9na90MWx7^<G0XG_n+X&;ynb`x39U}95{M(Pfj8S>JqkB0(P=ia
znA9vg0W+Hp9_6g1@}9dJycFE5rUNJ?4frSY=VogSk#qI&uDPXw#BPHgMqA7hU`$KN
z-rd!;2SivZ?=2><|AyzHHuKoH>cPP|Wz|MGeZbB4+nhy!=J}ND{wXC1pKm{!$T=VC
z_0Jou6#B(N{Y3V@RGl|4#%rI|YtgEs1g3R<60hZ2GGF_QVP#}6U`>R@&d4S(P#;yv
z(P30`1_S7vwq7D{=}-mE^a4%p7!%UqG6+qEdd-W|CSQ8d^z~<A9B&?)3AQ$~z~6pX
z88>JqNB;f_F(?kDl_aOM(I(it<q~Jak@Q6Zv$^aUV}Yrp4K2GW(-e+-Cmp(*X!+4w
z)nD~=Gxu=s!=NG>)Ed60T1AF)$llt?@%-eStzx3J=5|qVxN%k=*HO@3$y2AyNr`EJ
zcxFNBbmE!R9cqHEQpVm!{M{nV!p3)HJA=E@6>)4>#=B67@}F#0qF8+5woH6F?!5w|
zQ`s?1S=zs2DOQ5f@1jXQ_G@nv0^F08k3oeq$hxpch1a@~5}Cvb>o*?k%(V?WML!gn
z8%a|}>e7Y;7e=<FLaV|llEo#}tDy_NvhZcT;)ngwtk^8)cXIC+dmAmfe6;pe!xZfF
zAgf58yYkbTn=P$nnre5vY9n!Z8hpiRI?!Uom0^c=;_UdX8XtFKZ@0@8bMnUOh~n@V
zih1a|({oT9f(My8m4O0X!7VKRdxz{-{182<)CYYGW>g+Oh7e@fk{MXOstGH*&~BY%
z;e6;}&t1L{vDrmlvy4ia84k+_mWjv_JJI`JJ~@}$9&Z+!%F>3KT|##vJ|AYiRj>Th
zNB$S}%G8%EUW#r-<E=HYE1Sw_M|49h8;fO^-#cNW7sbe!1U?5`p%3H}Gu8^jb4>nH
zb2*&ydQ*F7K<xRRGu@kT&NIpQj|hvoy|kOx{i$7J#i-nfsHcaN#lp84a8J>DKGeL`
zu_MbNvnA|L$-1DXT&yh^zpbEq?N_RJe~PF8qU3U|;P{PxJLbA(v%SxHrnZ&xwFH?$
zagr0;L1M)OJ(U~3zH=4ObzJPAQk5hMQ73z6Nskh;2sBZoMg5h!k><KY4M9?EmZM)>
zN3eHY-h6Gj*~XGqvt5r-e=GLJT6@n`{j;|i-;Li3_hHP>x84~K<Y<yu7<pj<Ai(?s
zq$%gL6O}uG8N)6qt*=V4(K$+H{ALG?;!K6d>=JDMQA0{&^eY3Mgk2t!;a4P9hxzo?
zN#hL`ZjY{;Y9yJNA6T=7^Tiy%h+BBTofE(OF~;1<1o8cN$cR<yu4WCLoB(YNIfJC>
z;`e&Pi76#TH%0f8o$e!MzZU%L)ty7VW-6hNY~wzg2Vx5vx8s}5K^#Mc@`N%~Hq&Xv
zb2DWhip=TKAPiTgxQo+fYn2Iw6R2?c$WyI`74O!-K_<)RP7|Nyj&-0PT~#C5HJjgR
zk-_Co;`yp_B%{sl{_>Va(;lAPK^t{ehcCBvvP#tuxGuih^y=66sBnTZjS47S@9_M0
zTH$Ja`}jW**fDbGo<BGFj>|_gD6Mfo{XEcSVLEsi>Q~k(0bLAcGd2r3eM|#ijj0XF
z)D+u{+Wm}aZAnauA!n6)45sQiJ>E08s!&v3+K4;|?sGt$<>2*f0GOdgx!^FUUsE2m
zh<)tjUg8mehKP?%PTo=4CF9=VI+KLi?-zaAV%%(f8me-0oT&9-%+f#=K(ho7_$BZ*
z>gj!+5S#`l5Kr>)hB<5TU_kZm$r<q)0iZ3~Ro|Y#1-#_3T&)LCJRZu2xhH`Ws5kH0
zE1J}aU@}Yia#v!t_7D{(Rb7{|s%*9AYz?LW*FbYzx-<@z>X>W}7misrc7qzRcRmNW
zZa@SA%Jrs1NB~}nVt5RcJxom!<fL3{suxSqRiYRxu%+V<pF<uUO$R|9afAj>1Z0im
zTVLh&KgLrgN+&Z*CU07nA;<o%s0qRDzMqXm;$7!Cndp2}J}ZU=HaE@i1xy?Vd}tbK
z!6xq6PhF)HIluvOFcz#8#g!dPiNH0>Ti#WNxB{ip#5Xc8NADq(R#LQ_ga>WF$`n*D
z2(7eHc;NUvk)J`ZNv)m!%|FnOU_DzVsFi6@DPX6iqn!DH<yx?z#g^RI>};Qc{iLG#
zzb!dumK*+LnZEarAls*V_OQp`W?726&!@FEnrZ5@gf6_?X=2#B-k2htt>~6RyceR?
z)<kAs-$DJ#p7_jRBF4xUX1>@CYnZwueQw6#<`%QP6faXVqiz@#s~hwvX83W0UHi)t
z^xq4{`y0+sx$&pp(rNe4$%Nge?5(HUxXN3#!!jIZG^EU<EFb&!nlZx$qa*uvie?L&
zb!G<ibzp`>5!$c$s;o7`L~DItw$5E}BGTR}4{WPGkNJGdvADU{+D9UN%L>bx6t>E1
zTg)o4bi$W)>X$zs4j3kA05xUXeG6LqHT4303q>LvJ*5$jvvdC<eh<MtCG7Q&Wd9(n
zy-5>|SCkit)&ZAr`IaAHqZ{Ci+&8%_oREq(zIFo9N)qEDDBHAZ`~15z(Adv;WT+z9
zDnyf|K)!ZGu#w~)%OK<J5&o-@ZoCa*x(^w--8&F4L~$2NzF54M@hm&u%sQ&cwg&vK
zqTIG&xW<dZ-#fKI@9j<q_nJz^SN4tn_|<CgJh-<YoymK|wU)bUH2EowbqZ;up`be)
zth;_`4^C=5A}#SB*P4>H+A>V;%34<9T&%f^%4KP07TjWq%3x|kJJ~xYc5>T&jI3ON
zn7nqq!3_O3P-Xj+%%fGWtOaFiRzc&twX@htG;3@O!NhdQl%vId;oa|w9=W8pV8{0&
zuBLT<!84D&Y(!I*xK}jeH%wt#n}tFma?E*gLIvrdf4j!?v}YNrH?~h3tp7#-b4aF}
z#cW5wVw;V4vlk?EJ)I?ZUJw4o_{Z0Q7I+Y#!m6&%OV?x*M0a~)8WNN%X6eCi{2qZ{
zpBP+8Mih$L*8mPr#_Oab&gZU7)9=)0an#Mv(f`HYov~c(K=rFVznXHUE-3<;Trdqp
zRV8<>l*P9=GW9hFnzj<}Flyr|m6w#wz%|NdOllQu>na^u^)?e3GE=&+YV=c5c)w+#
zk?=Tg3qH-pdqtmYztF3mEvUlX5?}vgh?tA)D!*KEfTSpTum=t2FGpGwcr-iNsTkA^
z@)Q^PIk?O&37A~cVM{V+Hqevb3DlYBT*P<CaNso-eK<BtV;Zh<SJ1}6?Q@Q+&p7hm
zytP98t2XHQQjWZT=xZ0@#tCRdFKIP}LeX=83F3oiFhAeV<7$TTwI!pftHEQRTKh+l
z-Kz2$>$+G+GeabxGx+Acm&pC#QF|9XADrnsTjq{->*#<fWm?-CAbcemT9Sn-h(Acm
zkB}mrsqH(iRj|6bvEO+%7CQfFi)AoSN%zDeXN=}gNx@EOfOZnD+>b0fII?Ei649#m
z7YPt37u?0GuH+^6iT_)aTsSdB&p@*)f*a^hbwPN?Bo*8&K>v$mOWNNfWtGXcgFJI&
z*Xf9nInqeUj02z{1*B>x!hCRNH97ov0G;Td=`|>T_n&Ur*lI@oKY06?YVRzeJD*@a
zMJIdk8v6Qd9!0MEXR<-+aK0Ev{E@EYgK(M@0(-StZmNlUxvDz6HUMuLnhwFYL;}0M
zysF*sS?fh6$!<p=6|$;!Dil)rv|q@pyFsRMPEW$YD5l>q;%^&$Pl36@F}3U%bODOf
zES5Rtj6Nt1Hh+f!b4+7n<eOKMLJwCU<A^-DAwG}Hz{{}A517qIJ3jeKi(PnCUyHSe
z=pAB`t<or&XEE8|KDN0zMN94a<uK4#)@f6y_qheday6cP*RewA2_&1y+3l?w%gZoF
zSna)2Bo-)r<vMuEP7E2)<7P(DHXo-1#YtL5G@NO*mV45|_u&79_8vw@j6aqK;a_wl
zziJzB>`*G`zb1XMPIP6}8YiQuvP!|-Kyfado75Kmwe8_&q$FvQkng^pkt~B~3?;)Q
zvixW{?)_GNala|Eu|=Yb6m%6dD+zw#q}9AQeha>q>MQewF7jSQ0TV09Q26LOuovqh
zU9~xw-QqTqaY;@3vrI<TBI4<>-w2X_<)2i{D@9Wu=KvI|lY|yXD9)^{Pg4zlJ+(HU
zOVGEm1>}i;1Rd*R_yWI9^t`!CLv7uBJGp%Y^Xwbe?WuHD0z|73a5PTL^aY0wB-{ke
zW9~KApb@Altv#gNYTD!yKbDwk-59ouPd($iQ=H7B&As#=;0aQ$yvLz$pu#h`pKDFB
z*#av8qCFL+ir3Ac8E?60bRE;>H^t}86&szNY^JR2%Kd+jCe2t-^kd)nLVVwOj{X>+
zHSqLz!SBX#xHo0DVB{X$4fyA)B<AsltG4$8Bpr;Bg$X=1W<%e4_U(S}5Y#R!Y9Jdo
zcK=|$wMR(Cv~`g*)ya=Bmq-QwSm!6=wI0pKR&Y@}1058?JKnguUQtSnAFjYR>KBJ{
z`aC~>uyXr5!J%RxIY}Z&Saa-8{CW1@1%^Kusb^h1sMzEObK;1}Z)Twr`7*oZ8U$a=
zrQNUOIQWuMd2MzNZFlqontUBBKypZE>#Kd#uFdTzk9cPm?lVkh)yKA?tko!yEhn=7
zyfAbF@ilP_@x;XBRw0&F%xo_HN)I#Soyf)l-RNq&Ds2yOph4t~-RqspHXJimRNFoT
zE3PN?wParAiCFAc)~V}c8>BvQwA3#ikIJEiQ%ji(h4|Wgq2?31(Ih#z;R9<&kvN*Y
z7Z9Z7G;|EEOVy%<tCi_~PO;?gSsOhW)MbAvG6m=M1(?b<zkMHEwO}N~$z+{;xzK<;
zqHqbCl5lAXASr}y>;{<C?Y0;nb7GC?%)xkpk7hbFO>WGAyk-ZmND>$<N4!-p5R_)0
z$H;XZ-Vhs2v%pEkAQ;{nb#KJkeL|FPwIv?`3H%Ow6CYp}Hg4*g?DE}wy!Ppi&HKuK
zCnRql3C5Y){5w}Tz-<nJW<xI?{%zX&6`=j-Cy;9Tn^9lSq(>D|;#*qtO6|EgF~{<2
z<e4e!F$(6u2_WVOXQmkj3r)4ClEOzJ*1mt_{e7gT!-ueEZTA^VKE8@@b$++p?Dzw}
z>rm&;Xua7jfZm8C*R=0$zRVc&znq9TbQH|{moEo$uj`Ry)*Yy=>seEp)$aO>s6jjH
zLV^~NZt4aU@Sc_8w_ez_U>+6+I!CQ>kt3t6zL7&*d?4<Gz~}JIr|RtR%>;JG+sCua
zzF&-;t{_f(l&iKQ9v3a5#t8FoTgR#(ut#^0fM=Mr%AlsskERg|8blPZzx542w7=TW
zR7vZ|l>U7$-|qJlX!88UQoT88SmBabn`+hRhOJ(U7aKF(wKlNw0p3_B%pAJPyJx_{
z(hDuW_8)+34M@zfs20O9%ZS+8Su!%L7xtf48;-FMxK67Lv^l|s;@)L!oMt*S=cM?>
zEc+nmOc&SBK$qWLmr|;fKR_2LO2Sp(#se6mz;Oi3oD6`5%`5+?W=PU`M@>xFZNzb*
zIOhSpcrK9>m{&d_-<CtxK}nnwH#h80qV;7tht@t5V=eqdm~0z%uVgQuGEY;RoJ!`f
zWUyy=;j><w7Z%%)Fgd7s;7oquM@Ec#*km}0J12^u-@tCmIc(#vUj>?gHziEdy{rw`
z8ou&gqfgq*M2II>b&==jlZ~zkhy*?xcp^<uz!npWG<5OlTDqb^Cavg^qEI`mbw&!v
zpB4!3u4tM;HoxZ)vNmL|m!9~HTUZywZHpGA!TWkYd{;oPfzbCONBl+4)kuNosexrC
zy@D-l%fNRgNtV%vv&&QlXa9%Tagyy@p#ec4yR8k<HN%Urt};6@@=S=`IfLSS3^&>E
z^v=Tvpn1lgX*PedBy^=>_veomnpJ(=ny*5aGo~a+0#vXvjIA^zF>$?G!vb@s<kO61
z>yt06WiHxe%G-K(xvHc}fJ{-tUB5>vgVR-`7~TMu?jYjtrK+w%H4|R3N3xIEQ@E4e
zV11=qA!~uyJo!o?6;mzX1gE^kJ!yq;q&53yYRGDI8zL}cxwtUd(Y(UR%Bw)xXV$wW
zM)&P83n9fy$jjL}%*J;n;7B6uoaP=2VfL3cR%lTd9HSXotb@#SFA#VW=#Ax{wC<Nv
zHBb*IvVmxo`AC;ZSb%gC((9vEf69X_6<URSi^kVMf6NvpKEEJbtqY`p@08C7IG^%)
z)a&h2Ww&jIzFUV7Bl(PkBy$I{9EAj4raZlH2ov5Gt#xUHjHiF<FB0>?dkYLNti2%H
z^(IXJVIDfRLwepSfNi8tWH5=$kR80a9B6d?ug8fp+qH2CaPHX$xMyxDh()7IP7qE2
zU&^QG5P|oCDw4V7F-@;_@ib5nn%5+8ZxgQVQgf-ISa$Y{=A1Qg@dUPAQ8wI9qQ6+E
z=x_n?1FLD+GJEhcMf_=P1;i{p(T?P4b_REJ%Jf6Dez-2H^-79w^zG_~%T<ZrWA{bk
zh<eNc)3MC`If>rq9V!uyV4Z=TLLup~@qk;!GFgnCRYge#EmQOYT3Tbk4o)Uz?z!)T
z8o;!{43Z%6-M<?uYaw;iJTq-$pJMo(6LMEmbH>uZ2)C1e(n|mUKbk=_bZ3<x&fjgx
zh}j|VGR$x05xf~cHHS5ZZxR^bAF3UwJ)Xa7yREN#WVTb>vj-4Q(Nk<=JL3(SoJyvS
zI5u!YdF}(x>;fh~^UEtGL;uIoc}KIoxPLtMs4dYT)GA`t-XnHwVm8%Ys|c!AspyDJ
zii%ZY)QAv7Yu2b<v}T(kVpLJJTSXOhyYBb*xxc@h;~eM2IXs`|^E}V{^^)_3y?5TB
zLSwCwvmJ$ZjLHwaid&R5H$ma#ilPF8rq%O)-xaTjCz5mplXJTolr9DYX~dPpemipb
zQ?BBYyR+vF&aZ#qv@|lVUGLVL;WBX9C#6mj=V1gpZv6~3{F1HhzzjPs^t?f}NFGCP
zP?V=E=W95)?RU&?K=A?^6BN+lRb{J0$VHm`h^N|+EtYP@nQR+&e9X0^q+lar*voAT
z##d@x(^$6a&h309zjU<|G{FHWzKbJQBydIe^F`v$S_{VS)OIz6IK<jvQS<V(0JO^K
zUJ=uy#8pb;`m=_%qBE(Ux5{c!S_E_%Wa?Vc)=(H|dNv~j!dQA(<ZW`5_9+5g+paBw
z50J3dfD)@m3WJ{??EFdk6Hc#TtXFNtU$|6Lf6ZIITPogcoLq2M3o-irj5H>yPhy5`
z6R{fu8hOUtp@Rk^EPQ`HaUL#DOPg^!e&0jjgp#5iKHym|q0Pqu+21VF-dI@r!$bXQ
z(RiKrnit&M4WJ{;$`hlQiYlX7knCfs&Lc1mW1vl%X1)6<$(5Nu%wUYF&bL@Ug|-wf
z%u`f&(#68GZ!b<xqy)-)jZ1cE-`8bV5=CAs@QZI98@Aoj$^9#>@gR<tLyz{|u8lU=
z#`(+F<hQ+Lst&%sEIe_BD#u_9CA~Vgy)-_DTb@fEZ<bMLTxLY2Q|=V*S>ygv{5kC{
zvFm5EpLZ%4#v9Pd;-mludlQ-}ht`RR$#<lY%RG1d3IS+aV=QA9neoa-ReGG<o)HNA
zY;OJ~5Pd`=Xa8ouU3G7GOkebPOaES!xsn;TpT8q1NUYcI-2=&ta<<-kTJEVMGOv~s
zJenA#)4%DG7_bX-ko)6Qv_g%lh%_6nS+Qpi&|AsTWIe5{LLlG=1i}@PrSi`9h%t6f
zT3x30I$4_J9*trabTZ}!z3DH!wD0q?1+URN-s07|PPi<WyCG^|(nICdxff&qhnFY4
zcLr@`b$#(xGyL03pogv(O@-quRi_AVJp{))3g(9`IYOg-_hK#lHeh0U3430vX(VeU
ziaS2pdDi!vO%`;kX%ouaQSve<je0>EN(;8*X%qQHZqk<1Gtg-f)kHMf(u+o46xT2(
zT1qI1J_1c^$?TX$a*o#@)8~z=2AVxWlC<Q{TnoJP1M$_qg$3arEtUyl=+Y}NcY(6O
zu`pLHM>hXFa(PuK(>W%BYe&Ita+;}PDy(<CA@O@7a&%_jk@K90K_I>Rgcr6luF%73
zVR-4RNy(x3yt7`&4UxrsB@jrWscSPm3o@QpP?5PQLphM{Td)$h2(5$?P%PI5M0NAT
zn%yvn*3M`resHu$z)1awj5YHkaLO>9&a_dlDkT%^Nott33kwRrf3)^;zGp&zWzX01
zU)jFQVXIds4Lp#u7T#f*gkPQ|8&n6lQ1xW=YPkYSva6fu|3KcS#*Cn#=%vz3HE^uk
z_`9E;(i)U~6w^{^`r(Mn1s0RA{4!XJ-H6rV7+A=G525BlCp@IG2tR?6%D3DA_IBnu
zdTn4bVpyYRZm?Q$qRGOKgZ3Dkz9=%nd@(&~y`^ClO4i)({gHm-%CkwO!x}r^o~pTE
z-FjNU7o?MXSnrTkP@_cj<plWCkLlO7{yul9-6FgHK%xu#_RG?pR!t=gW3z-jx0~dn
z|H%JE9_B2_^9Q$aqObRzp!z0bG~>Fxo#slXUv=%e^nk{m^UsP$XXyzZ?gjcSMtk4y
z&RG6Kd>gZ6dyPA@vvPHub|pB0t+#nT|K4%KIp3$(j`r3SLXE`OiB7lbUS2}Ke0tOW
zSnMg=ZT-HAQ*jd|^{o_^*K7^mz(;d%G`GWp@~#y>Y`jo3BQbmhn!mZiANDEY-(9K_
z0XcoGsEbe?j0eFV(vyqSTy?`5v+zrN<8SRZ^#oyFI`kXzov7YIFSzhqUIQ0wEw72n
zy??|$Q!^)(t6F!gRUc>vbf{=JAlOtjpg)QK1JxLOe`3(*>pSt-ccIRbGU6W7fipbD
zdl*1lQ`n>)59hDvzZ%g4gVMg*V0sMz<+E(weKdP4z>b3sPUw@5XSF#so!Ol@C_6+}
zj=UnEg#uNFDB#&N&P|!qY!ZPX4J%Cr9GYi{hwcys1T-3lhSp2@S@po02{071Jj!tP
zZZ0s{BwE-;VsIHbe$oG+Kp-MNOpIO8_&Z5~Dvdk9qJE+RC82y}5N(tnzC?Q@;bl-<
z>&u~o23eV>)UU<oDf3#xhEGYyYb=M{JAK-HVv+lqM^ud``?g1TwzvHR31<dC;nQsn
zfB2kNBeqt4j%tfiGTZO>pZ(!Bx%lIi;ctg&)cs+rfucD@&8xPHfw$AXqqF`=@zVEH
z(Eb69?Xa0oqCARwtZ2f9UbekMoH&Scf6!frJ*^+n&h^PmcI}^2TrQbxH`BRh-Xj_I
zMTS}TtU*d)m7tO@Fc4})GUlzN2u(<5hMf-_K`d9~!Mpr@13`6#uDx>6C{~$-+!yZu
z11(U}YCU&SSCh-IX`rp57{j@6_IJ|mzpuhi&bxam`6H)o_X_RBS+p}U6O(OxX05*I
zaq2ffVzOmQf^L@AU-3R>y8-o8t<lzm{ttw2Q5`y9`sp>9t@h@2rBRA$ZRLQ^9g^FZ
zrHeJBS-bTvn}}x3{GoK@0N^=`gIU{HyUB8r)koxU{*7edBs$wjx=tp>!Hciu7x`Sp
zysLb``$29H_Iqol&9hDR?eb+e?+}YYMQ$T^&O?t$E1sBlCKPvh`)xw4c2Q#Xw~eST
zN%R{Utq572HxULw*JX9|3zZ4B>MeTrFi>t<dq&(2y~vm8^0@!^yjY@}X;_4ain+8u
zO)itS3OD!attAicJNaNLjPb_ec2mM5!QnE(;FLKKS3Xzz?0go}oxs`owgWY^_vZXp
z8>eE25Xv?Y-6@B;iFMq@!O9fC-~ApdeM_2Vz209)4759RwB(vNYZ$vgJ9cOKhGbsj
z>l|S{t>)#GA9`{yD0^o2T5P}eoDeGl!~*hG4M_+j7e%x>yr4|}h?ozZ&VHXG*8JP%
z8LrGQj$(~v#OpU57(VKXm&FIBtGcDa;22It4sA}f8xr{Q7e*E+G#$W;!9jbCS6m7u
z5VYt>#FXjW%fHr*LVwE{Oxef4?m7iTC`uzK%bSdLf-!G?(t7!&1xll0_Bd2@L?-9i
z6Ywk9uVE~JXkPLG0VP@5B*2qOL;B>yYcypNg6K?Dih`F`93_ihz7F~qJQ(%639jjr
z<O*6b{P&hHN>rzBhl(}^v70wpwYBRqq)v61a~Xr}I1sOW!wnYe@`NGQ`L33pQNXhi
zDr=*+-jKVyt@diD9exgA3y>7U@<L03r0~?q2hj;O9&L7!V|}w$y?d9xRgvULt=Vpt
zJQS&49?jaHX4;iL!fgXWIH&An-e4;yd7GT!Q;o|NG2!=HUW{G%eoJ{dXUe!bBi1S^
zP{n1(O@8xKHz&{zt;my}H=fxYZog`#B*lO2<P$YAX5(V0;+dqA6NaPo{rl=an1$vq
zpL${7nkl(l9@NQ`s>q#2DTzNA1`g01bh^sGv0iGVd+t_1XBV`@v%KIGrp|Wt>Nwd~
z0v@hqu1HcR+XJipG&=7hiHG-wM0P|_HKe6{H~B66^dgl9!_KS2m$crku5(iJYx1+|
zi``aie<iqE1vPJd&cP9IO1raYp<E+^E__)i2V9=x`g`;NB@<`S1NVC;m!y)sHXlq>
ze?K8<khb$d0P-}G#$eh<NV%#Hl7`rmlMU7?==`HNf#jv4oNq?c$kmm`G0|!^VCczr
z@(O+eRhX|eSO#QT7++K4$Mv#zSK(4pX1Z8LP*ufU46Zy&n~Ogd$*N+vU1+v};DDXg
z%YH)e4o*R<qk%*Xbu=2O84B$b1ZWxycq0^a>(JRf&z!_EF+O2dfAb$~OiomE@@0=}
zpWK$|eb4VME{8o^`Vh%wRvA#+qMJobklV_&ne8>hHVwB6IF9|jhYZ#p3?bfiVs_s{
zFac&SsiL2=%t<~W{Y7$0Z|4}8pN?LscU~F)UEQ*ytSmO0zCz>K!KuR$EvSCOZb^~+
z`APXZuQV^%x9rSQm8EqphMW5X>VYfJI=1Lkq$3fc7L(-JLmkz#dsvU*T%L@4rPtB=
zk@=qKaFY>EAM6l47GTU1d4IBo(0-Q<Ia(whGspC3XYwddYtJs_Tw$@oeEZNhGH+fH
zi%x)_hx(zHIVm(b6N1Lvd5YZR7MMJ`${y+L<ZD-W;dTI3)jZ5;`%*AOTa9V&s>cj!
zv0@zbS@AL>p|$8bMP@l^JEKT4I`y++C??n8IZO}8k?Y|zvI?FS=H$VJ>s?S-A<3fA
zW;qN*7!@F^gF5{_mU`LaAA1|$+zR5(lefPLT+9bsKDKGeck~tM&g9%xemc`0sHv+N
z>Mwj57>!P_v-wQAD6%#h&(@hhy|rmFq3CgO(Y3kLAzuOBSXCO96UA)m|FDK^_^KC}
z&?ZyzXE-BtxhPpbVu`f&tfQ(faDG8drL{w#xkSI=ub&=ALwQ|_P6Z($E8uyQ_qKtA
zIW-_)yd7>GYJSCaAa8VeLR$2TNqBzJJ2i56c6sWwSLX(!%rqUdqlTS}zqj3ud=vQv
zB{IHUnhLyx;4bnA@>8vRPHRB4SKCCUumCwLjzZ_)C3T}|jiz%j2Ek_5zRkMeN*-J<
zh|70BZqKT{%ZI_Up>o&h@}*Q?)s^LguHv*rQ*>H)Sbo;kCaFYft-MdpS0&FU=!`MQ
z)v^Cl!Fi!*H;;fZZZy|ZHzWc+oD9T?kp;SqNSZtb?8XQ@mLecNE8|N27qMa+)FX+a
ze}z<{E9j{wExAlDzCsLO{3zhCr%Fcm*g3iv_r*FmGz)q`{U>*vu?PU24Kgqov0eq9
zrG?P#lvtW;Dp||i1e2BXjFwEV&ypmUCcjtBp?_z_!9bfo;~F&`TtS=PA(EO1lIB+5
z#u0^a%d|K&!>i{`MLgJsYcy2sB8z@Qw`D*foO#@mCypwP`3jnXq`m9py^?tBHSc^|
zyHH2PRv{3jvEKU8Kbi88GLl;(mwO%RPpmUZ=It{Lw`#p%|3vL!02{yHj*fBUlljh^
zTZ~UdoqeiuETD{QN$&Lz12)(DA)KVdsq+fTK8-=4?^B(T+m-zbmoC104{vM@#w;zh
z$f~~fObMfRXA->XC~Cp#F@qP-jAtNu<u`AXu1(w8h>18`ofL0NnBz4Z*=qCIz)U=2
z_?L|GEmOwFo`yDCXYp3vxbD~A@J;1~E&+?+fa*34po`RSqn!bW3(0-_-(W`32Jur$
zpGIMa;Fgr9QRXt4?BMERDRceDLkKv`{Dc+}bOUql*c*T=F6jJ~wcGHuFre~sks@G4
zP=%6Z#l9ovTbcH+B{uO~Q{POzmdlbCxn%&_5hdqUnu&dqN)R2@vU@g5v=qi`jBA4o
zaaws@f_tBk9D#d;`4l6TAJZEACUA{ZBD+F2Adm?a;J;@L!$u=gf4U{_&6lRh)J5DB
z-9tP`5Q_8}05x>m^SI(3=*yR4w@Z0t2UstR=ChlbQDyn`!Jt4KuRhzDHV*;>emP4A
zcy4qMVmp6C=23$i;-y4VFwST@pKsJCu7O9g2!~<|4<$8Mc-=A;9E<rqc({uZk@ugT
zhnMIN?$nCqQiB!J0{?M)o9>UAy#HhSL$R`tFVnv*^qq%MZCBUg%rpu3xcGf}U3`C5
z57GEr^27PoEC1Yjz+#;2{+=4L@4iUaejmKwDX=uM^}gEkAc<JYGw<CyUx0u56~5-h
zShp$_3-p>0UC#c|^o_?ol)K#&ICh8$7ePya6Aw|)jkls5`_|bJWiY9wlroc<96#9K
z-<nbFT3zp~V>I8y>`!QaoK;9hxehx#4udMS`X)OMD;a0l497KG{5y930o!=wGP8hJ
z9N~;2u6p@(@nTY3^Dd(-8q2Rrn6Fh5&vA8wskflcdvcgqU$xF~Zmj!CE?3gMpqeJ6
z@OtQb(~}0@QL|g{-{H5N%12~ERn+7LiLoaTtce=iUWplcUd^e{)Wf+uTJfXphTDW#
z<FF$2)Zn0820wT8OrPe99OmIGDwj#h`lATrt|BpMFV7Wd6SH%9tb*nvYZTsqy>#Nm
z6z$_JT)JXAD!;f4ga`%u%&xDPB%ts3t1_7IWdrsp+8(&BYo8H~F$k*ov?^targ1r@
z!N7Pvo%CAej-;&)O*L1>*TP2B>DsfN@uBtJpTwHuHqWUr@@&d&%5@>@o~p3lcH8ny
zn_eiW5-!SG`t1SR_2@h+lh8sX&7?W2Y#t8TC)&$|e8U7Pp-`6F=ZIbwZ5;pgc>zp@
zr_ihTMDC-`xgWQThQCBZ&6Iv}r*Ez?<+I(YzaA)_AG9M=Xza<!MJ_x<q=Ty;4+9VQ
z&zQ_sHV_l$vq>n^W?+LmlG1aG;C{uD{^`o9MB8X&J**1sb?a(vxV;|X#Tav)DlJBq
zO6WL70w<6K3Odj|v{i|VsE)qQXaCjAi=K^ZvCfe9BGEKH@JG*Bk@VUWv`CICvbH3@
z^avR$mrF^%O^I0Iq#AAt`pS3Q+`MKt#zF8cPB8q~RuMZ(l^QkKlS!9&sx-M-Y&xex
zay0dn+QG-ir1#_BOyeZp=h_K9Rv|cAC%k=0vz?9wWMTLyUZ$8>VgPfrW9(aw9Y>3%
zll6N4&VbU4rUMge<5o~L>!U3M2R19p0Pyv^tJQgpbw^ya9$c@KjmSBoU1n3)({-6A
zTW`<_R=WeA4aCnfzX9|(kGayv9K_+^nM;<sM)PT?EpyuS*dU)g-#Gnd!fV1jG5?D`
zQ0R%BqnUVu(@l1`G#Z?LXVAkbAno^a?;x47htG;lSVi*cGDhHCgua2BvoVq~dIMbm
zgpcX<x3DFd>S2(w!u7U_&y(bhHn%r>u!w`GCQYXhh3hdwFVa;tKOP2YNeah<le}#O
zh8I$^inl1AP+r*zXkTxG0HCWA0D^H$8a9f|TtCHz7}(j2k0x2#BY!@cp3>hxEUC0@
zq<rxifl{CWg<zb9o6}jSn8u}hrxwro!do8?$4<qFR#7Gtpfm$1$JT!o>a}iaPw2)0
zyNOT!7gS>y|Kw%ek?W~vCI!%E0Y{+Vn;)^IS#goYBYSS`A0CsVucgcP=1FEm8=R5c
z^7+@5NgN?8az{wcLpo0IMdZocMW4LR41>+(9td<aQ5)3XHgsrt!4es0E;^zXWdz`W
z(*)p2d7Yr%9ipMD^Z-pw!_nfNK_h8v1oS%Q$xq{JU24lxdo)ffIlmAXc2?b-%&hmx
zh9edUN;dy5G{g6T5p#E!X~+(`=lsj4Satdm+Np>j@+Kz_!<?_NPXnpF)eJrNW12bb
zH)i!(;wL;pepLgpcJz=cB~kGhyL(3~vVCIw@UmBYs*-pm{?m-$WU9+m(~G;cJvt$_
zN^&C6|B*e7zGVNIv9O$u>duLWz^0|#2g%y8eeJm0V2szcqr~&1`Gqq_)N|U&IiZgT
z8mw)*-AL%^TIW}#4WTPAHp9MvoGU`5E6T`PB(ZtI8F}3xQHKfI8jogE&}S!ea%gj9
zy+I5g9lGj|nK<zjsB=Yz2LE`fh#15&_$1ezvS)QLLa%2QTi^%Sxx`X=i@7Tn?uz3z
zy(VEKLc!f;-3KyLMIVzVXrW`kEhj3pvE6CeeU^aoS(wzdb+IqdS{93zcpa_h9Cl+Z
zFT6>kx2cjK=hT%vav3;9i!+zIt{3f~ra1;-(IYOJk5S#VdX=QyMvNH(E8~x;TJdmB
zOqk;wCq_q9dqV8qr<N<^V>LcQ?qLct<#?;xbiMb?Lfm*hv`nne`h}0dL3mpy>!`9}
z0fSDQ?zDvz-q(VUvZy7pHxRd$Mgy(d?|oDFv_!?wW!|usZEl!Q9Q;*jHq3BlX8XrL
zw)RF~PGC=Ka76ai*9zA-!tnYpb)(&*OHyt?w1py~tb-@jCTlWI58b4GJO!9)&tT!t
zOyFW4Ot7A$pm$eWzP>|+>Skh^x8+ArK6jb{A=&yq;z!D;?CkdaUFF(>=Q%sI2B)t=
zO<c5xmrq+e1hwpXqW2JK6QfvHMG!}q@t-A-KVVD2NwS$ieSsm-h<TNBC?R86Rq8@W
z#Dce-&`x1t-f(|tpiixb8;)JW_q2*4;d84lz+j^BU9()ocy^s>u@#|>oYz$=ET6;p
z96nQhV$dYLHk+PcT@hoMt`CF?r)w+N+zLes2r~Gn>Jr0zIwXVOB-2r-mm)=o_5F~o
z;yC@I$-wz<`>m$8jjDVuHB<H4ISK0>Oj_g3C5wKsRR$CSGSLivR&Vg_sgi>%i*f~H
z=HYT7l6xwbT=QM0R{G(tBFb)@EX`$n?07+o$?+V)wo}Bk^0T~wOldmk`zIFz1-Ug>
zR<#6m2@2=nh?LlE<5>6iYR+LLYp?%~^E4<=z;;^*dRLD)rqW3yNA<cY3|io&Y?otM
zwD5#>Ij2F39L@D4Kxz3rETBW%aR63Lb%KpJ8+mkCY{qPQOui$Rez9-Rky&t*7%HG5
zYx^l$_;xvNcs!oItt)|Nv%D^+4_AP^y<K~9K5v*~E9(P(hpMWQE7eC_vhK;46W|+u
zCS1+@TrBTf$Q@6D6tsQ>!8ZaWON~IQS-_>$YTgxhx~4wcQVo;Rz6;x3c-yDQF{wy7
zVl@@Q!a;5bHcPz~?b4a_JJC{&wkxfZ_qo!}lzb~fbS!KP^mVLiO|~n<&CrM&yDQ41
z%!370m#p2)VF#iXN+Z@LRNz0<msYC|Cy#1JA^!Moz<tvni*YdTdh765fDJ#EXKU%=
z2>n^HfXsM$y5|G~!YR^`{}wIpyf71E7nMp9i!?YyvcQgs3D@8}8ml?{XpsA3-}O8)
z4&oh6Q_1eUf4VCu>AZRpc#bhEa9}ot#k;Zu<5Zvi4`iEjJWmC)OT9Y6BRr#A8XL2R
zCLmVH2IL*8be7#1IPI98oL%{n=xMw``ALTYdiv~oEdz_PHM;19H-^!6kN={Va%*Co
zKZ9JMu?{^4KiYs#BE@q`;&_He7_qRJgg1OG`}18{=>rSQR4xGlppiZrT92qHGDW*O
zo1)osdZ|JfrsyiHXV!*bBR2@*?CqtUqjG=*)Xx}^v1DiFXL!v*5&Sz@4H6?F+#@?8
zoo2D4O=KMxQ@2J;T%KgvmQiu<#(2Iyr3JYu4wBme%aP!ZAQo3WK6b<Dv{#!I+|Rf<
zz2gmZYA@y`Uco37ddNXJj0@)}1^BJo6OCm`5uW3siP~*;TN6(!SWq)|pV2g@^Y(A%
zgdkJkX;NI1MX#oikw^bneY|CWZ)Qt1mhnKtEWXq6k<mj>wR<s6KCa*>czdLowj`WM
z*ux3IE1ypc%5qYlmKk`{FRR}&`24`cDbTgC_{m}^Z_wzjD=C$Cao!@Syw8vYr`*uZ
zglNS+bjK!%_%LSzY;t6AX|{k^DWA1&Tbk}JgYrGhrc0%Po9XB28C4W6vYxLkeH}`;
z`^mB*()U8xP@R<vIK_PI+Iouu%qs>iB75_rajLmfgGsZzTg=Me>5T;4an3{6my{SO
z(dgBpf(o&jh=8-Fvr;3LDrpVu4{~J<53IYp5+CDmZBNvR!F|sQ)-4NmZpD9`|1$i<
zzx93jQrX+^>Q`D*0}Af{y)p4w?`Z3pD1y$purTM3;q>MB3{m`*ebETMkikI-CF812
z+^0JxAzQ;C(Z^;k^*1y|q8SN=MxUCk1vMIp1>&OSYo5}AL~}p|)^4RwVJzG@Uz0Zd
z6|=SwKzO^ZFBU}&@YmLN9|`pv&T&P!mlO(oRg_{ZlTsxaqmYWS=WfDkVU=&L6s2hL
z#xM*qb(!<p018xzZ*0@cPv6MNBEh1dgGbQo#zT(PLRNbKO@`9)?>|EMP*UQ5o-_+s
zzmPs1aR{`D8{@I9Y5Y0gNP8#+i(aZCDNbQ)c(4WY611LUr$~7hjg3@#GtO?KUYF&E
z1g3RSR7Xk&Tv&+Q;b6vzv0tgb_RKP8ZYT_{ar|N?d7=p?^iHPfN<lX6Ek8sh%wU`X
z7b;>)tbeHS%*4pUArM*O9ln9LO_Rjohlny<GX9K!YBi5FMu|_#4v6R9u4pw6yc4)A
zSUD%5dB&b|kC<6Qy82r2)c=A^SA$x*HbC1~TDWcib=H$(uCl%b%Q3uFW2>VQdSiut
z`SOHORVx0Y%Inq7!RtUgY~p*X{X;L>d8>qOyENCk&2F}kkI&*0;v^i*Wo8g5c)yFR
zbWwEdP;0D*Nwjm-N3*nhh4enC%YXHZ?*f-@D9veji0C-YM9hdCVX@&o)MNx6EClJ=
zYM;SZU_a+0R-80#`0j>Tt^IP<ibX<4;s|$_wkSsu0{_@)*@SVQ2d>4rh+OOWfSLO4
z3}N5*?(Y9UcKQDUbtZXJU1Hx_2Q+ZV1rW<v2`5y<z1UBVvc~RAT2&nOkz4>79y-1=
zAVJ~zYcwCnXS;(8{9BB3=(8d;^|5`F?BDFkys9bcq46sdBO1K_c7Dt`LS*^{8qh2?
zHZPz-eTZM&0FLT>aIQxjbDDIy-S}zas<GrESw`1=I#5S+Eyh|VZDJD@Sanws%9L-e
z&GY>bC^ScQ6&J!le+6J{Ws)-z#|P`~@jsczfybTn5+VTc!}vZTLW~v3f&Fwxaj_Rt
zd^5RNVrDgC4VCJvfatB$<_Sl#d6zgb9^hq}E7-JiZGocIjhAtZ0(U)kEA*%56Xz96
zY{nqsVso;(<35b5P7Y#XW6agDg(FrOdatM4kd5a57QcOc(dXDa2TvQ17ly4dge?nu
zZZ+kt2SqD>7^l3K@3x6}=&wr5Lv~Ds^$Q^;3hB)>Ts)(#wjax#a(y?9TV?oNP6*XS
zJg^>c#5|bpF4f-eLPEFx3cH%oSABA^TCf`^v4FIvpAdZ6bFjgk>E$ORCRv#1Uu}sb
z69dGYYz}duf)=PC(>XzjW;jGB{l{=XyKEFJMKjhX`{2gzH_k2JuVQ`U@0B<Yedk2T
zC*KuvJBG+EHwNsk@plPWBj4WMwlM#ZIbhR{`{gU>HEy}$BVazhZ~vSJ7kycCMwKku
z(>#`Px0&kz4S}%~gl+P`;zxoe<=)OtInPP?c@y&?!HY5Qef@hyv(nT^n=xaC<1bq(
zicwg6R8rTyc-iJ&BIl*Nm-L=D>Ef|8tcPqUtV?1Os?YqW;aV|;1xOOTZw+O-JqE6J
zWtp1nk-+rG=ktw$KaSD!M~wf#?Up)0Q_o|)PWCJl<JwMrzl8;TY4z~dLVnrE4aiZ-
z0m1p@dt4B?O<Eq=ff_VkI)yW51}$UF0g-tInRWKkD(LyTdMeW$UaQ%!SumaNKGZr6
zE3`~*U_tv1)x<gw)hrS4&u|Sndp0#?d0K6n4ADA7{s!1)W2u^;%++yQR@}SeN_jBK
z)PKBH^0vLkcm0;NpWL<c8yvo#S?$lGo$|ATjCmM=CF`VGq<P2&0Ln1{OMfwMBp}Bt
z3T0WA>L7hufyb%DYZ?Sc%i?n<z(nqL`ZL@m@mybT)L#Kbc~_qr%?>UZ9;4LX>YHG4
z^fZ9+jS4vFdt)~1Pgk}Fw^NVt&tF+X;)Zo|-vk{hgReg8wyXENfTj_JSvU5a6>c)u
zKLTsLi8!sE9DAZuu7f15+De#1Zux2gr=up5PenMSpN5fUnBUo3TJfG^sX9rnbislu
zLvCXKzIof7sR`Zd7O!2?>m?Dk?4EPlX#eJ&QAvSo*qa!Zt8t~|P6aQ?E*Zkr@wJlm
zDvW4Ko->;~=jL<sW>v=Mk`T87Yu<KQO|_S4FJ#E`-~A5jqq@${3yM_G?I+@t<?ldM
zNj^u^LilI_4O1COKDjvZ%9al8^UY(iJ_l4-gbOJKw_wio1LN@zK6BXVzt)Ni_~+E}
zJm8-UW#QZX_m~*^tLr^)-sMPCuSt?uT-}kjkj8x_?5*|WO7<dCYn{JEe^|$O{F7H2
zVMl|oog>0=8TkM~q%cK~rC7h!+dX%FKN~8l|HO;e_t*`Y$A{(i8y|-}9WgQD#%$mN
z1di-}8%u^V8SD_@pq?pbMTLrE#8}8|y=?R#NrUS<$>66e$P_p8!(AYw2u2K$^9^L7
zT1SCQJ8|6~zVUMy4AtA3`bY^|2(uT8HeUWOGvs>$oC^asLavIl`tfT2KcWA6;IyiW
zl8EdV?Bnknf`V<aM(#{Tr_~|z`FuNII1t|rgfeyDKYVSzPkKnb?!6Y6w{wS|`H{|W
zT&^0M4vFLu7$bj&eye&#nGYydzOLmkHhr%n8&tpynuJK3h0f1dwae>>>x2f6mo;gI
zL1%ah0ty8GCJsl`j^_l_Dd|P~Qa__134Hw8U-Eu4+A{=8^tVY!y;}`5e*vfPtrynZ
zy)gW$W}kEo)GxSsa@jAAN+^d<zQ^9ux%|z0nwSmF(3|9Et|P=kb!0gvZMYc(B$QbR
zt)|i|JXhq0k(nrtMd6fv+HDngh-E^Z57?0+xn1pA%o=WO%%uWG$4)CF%mPIJ2zNnY
zQzedzeJ9lgy1sFZE721zo4$^(SG%$mUP}*cKLtN1dE{7eYlj-&XFa)sV80jP<r!F0
zuRyt*Dm_lAEmEf>lg6IAjT@}}oWHwGJkSE-Xa(L{jxDmc7a|B%aJ~HrgZFviRmeI1
zXZKf?`7%$bvTr5siw$3|9^R-ofTpIecU%`v*{Nrp_^bqC%^YwL4=ycJQ&*~LNR%k9
zDE!;VcGAQptf#qO^jT11NnOm@c^4IGi7~O3D_CdG`ZCOEr7)15nuDEYjsr_7f?&re
z#SMinDK_!9dN(A=MLgIeB<t8m=kL`48<HnEF`-NvqinZ`ypIT>a?zRh?iJ^Tgf4#I
zd!kHC+reYjyy1&Bm^4orpJjJ3ijb`}T3A@Smntk><9h5B*os$qn<UH~^~6E}mdWkr
z@vaicmZ277@JiMsV9{d_pHnTV^XDWJ-#XG2{TnNyea7E`cU(RGD5nfCJ-O)i=X2Hp
zaO8hgO!p-p{K!n-gtI9aSz<H1G41Ygl#@VFteO>H+<-{EJ=?B7Kxx(gAV>V1#R5aZ
zaz1_u%vuNzp4A_j44wppEjvZ8<-=C-uT-zTa%Gxm&|}fL$hizkqNAAgCXC(JbNYku
z7+)tMGDSyp&y!sECW3FYZ-<03pqOHGHTA!Zw#bF(<mMHbTFAKa|5cChd1me=i^R7{
zR9BJfJze!hb&AT`t(w}^hjTnjTgcJN0Z)sH5|q?qumD^y#Nc0JSM6T2zehs$3GKBb
z|Gv2s^-g>0OYh&=*4$0PwIf#z#=gdy&wlt-+T`w-=~Uyj6tv50PmbG4ZkkDP5((&&
zIwpA6^ssj})WQ2}m+X0c@*HnHv8d?O-<CTg4EK|w>Q;O2CS1+W$oR0u8Z%cJmYvib
zm3HmRVE5Z~+VXZkFZ+kPPk|lVk*|JP3_ZALBe7ET>8MH5S7Yyjym?U)jmhsnz&=po
zKs0G%S2m-<q1HfDgbv05u#Sa(QqG!xCiBV7EUV{^T&G74kdMI?U)!juVakqv(^)Cf
zAK3O<_06v*&#|6drl+jApzW&1w_UYr>E34!SK9u}_!j<G+>r1vd!bYW7ji|Bx!+Q(
zW*EO_Y|$UQ!C1J(pS3~M@Yzf_YaBAr-|lzJ3<@-%R6O?m@6R&|Q~lGOiA(C(Rw5CM
z1CH;28CvzmQ4Q{=n594kO1O$#npzQ@yrBI`amc`=j~)epxC%5L{yU%<ea!QCK3~La
zTk@`!lg&~^Rj0H@>q}9Z&1fm_)E*M+st%Pbir@Kxf->K(U&%>Ea4j@!*kkET1`|><
zKV|{UHqpg6VGoghTF-6G0tWmBVy>qCNUJ~Y`1fC53f|@z2>Gwj?H(rJYKc?69Ox|i
zJMt+hqBxzoicLE??;OyN<sc(m;n7#*f`|XV9Ro(|#BRJaLCeE=F5*pAm!({3OVAe-
z+|G=aGv7tLw=1>@7#kc=-Y}2Q^71N^y^8jLCJxxFarGG9u1McJ)#a~8RFgrw%RaH5
zoI})4%`$SfQfv8Ip1zWoGnxBDMjgrTg5?-7yv>DxRf=$qXY^ht+n+@W^bzQXCB#ln
zL-$~1*Dv&?l&sWT<A>Fb&*9V0#D5Yi)MfGSIOD@^FM5te1y$2`=AYRFMMisWh6M!i
z*lCwlvzLrv)2m8LR|i7%s|o!{46PVtg9u;8{uk(WwE)v#C%>v1Il-K}))BjB8LQWS
zG+l23Cm8;7Aab3D{CF$Fz`d(kUp7(GIv8e#c9(|KutxrX<;;nfE`<BtYEcriqJ~j<
zN?8eqR|G2>W&aM_E+g47p-gz~B$+nx6GisEOhXXiHtl>qp9$D$L%P%vF|@=i`jhX&
z?87_1+9IjH8WV5EQz@5jPu8{T@0Dq~^pbGi4bvMJ&*V#ys(4FD4)$`!G|P<L9!h-(
zcqjtdA}k;p(Xuz%#+^5zJjYJ6tth?U_Z}bcfD!lE1(~0OZMFW{lPJA)u*{~@=Ow>^
zTraiO;E|~d>vRxBhRw*@k;u}YCYs%Hxc6!UU5mu-g#$d20;DTuxM|ZTs-6;oVP-U6
zo_GXC!UyN?%wH;57j}{_tQ@UXC+1EKSY9tBu98afi@n&U%zZ6;6VDYEIEVT5l`-ye
zxeK+9{WfXpkhO+;edcvzeK;|uU#oT2fxKSsHj3TOUL4*3>2a!DW8V83J8NK~>|rjW
zhGl$J+u5o912FLm67>Wg%q~@PNAEj3f3f3U!QB0Yf-WBz)LQ7X;1N3B@=xB6bj5_*
zhoj_AxIQ%Hg(=P&pHbxwCQ{l}YIl;SJ<`%dO)vw3n5&CaX-zW;quEC+2RX!H$6dTU
z{(K<}HlO~Ya`4hfjAv5`uw!&7oz8nF;HOBLQ_k)_Gi0C^Z)cDMce<2|1FX{ms-z`8
z`Nm7i{AGW?XrSaldYU@?bC+n>Llb4BN3@KqWA@2E?+bd!RNS>3MC(sf&UgLKvE~IW
z=J&IR#W9B7=ouJorDDD<rZ8;Csw@xk`fj+874bH4M-=hP?o8)ZJ@s9k1T18GNz)^E
z;Jj_;FVk|?+nZ9fzz2_d2=$!57PQx$YRQnDN)<Gw+T13T$-JJ-OxIzCU%gPYaH&!6
zP~#9yU&i-~i=qGBGA-_au6LAdjrpITpq&{wp=RH^vlc8O_fH$Z!pT?L7Pm^aZ%j|<
z*k$?#r??KP&l!i0SBKcvKU|?kNUitm^v+3Sv#Clald4${ErIcB)0k3t`jOK9o&SNl
zt}^b)=3RmP>*9{AQ<(4PNta5VeTc?z99Of-<_%yP3KzbTnT2kz7nCyXS)_jGl45o~
zbU_0sT|CIVYT4}{F_3d6Y&(w+p5^R9f7Gpo;J3`HwAOWTr%9<=X*KkoY&TPWSj(}T
zliV(poiF88phDL+-I;9dIqncG&l^W0`Arm=VmRx-I40;$JqLKunUPjCi~~n6Jhs1W
zP|YD}EoERFREXpizamN=jbV$!nxwqfmx}eBYr+KGaIQQJvAH5&<SqTqKiZi&O=Ha+
zM-9l1h^ilnAQW;dTV)H(+5z7Pgfdi?y(R?Sozt-y({X&0Vl&uu{=B&Pc5hg}m}<^e
zF{#4utaZ%!q3$HQr^zg?58M8x3MC!D0jcmAf1d&RqOAWVbVBeUHeVn7g|%_gR=%zf
zKs`{c#+{vPHed@@j*GOxoP?$`DMDU&kR_auJd1BuyL`0NL2wgXz|{?y)GFaQ*Yn4>
z|5};Hh>nv7tKI>7jOOuojh{oiRYp&*OLk@nMe4`^>#a76pwB1kr|nv@@n=m3E1}FX
zbK-V~hzyYl(kDmr`8-7?abijUNT+Bu{neYEzvep*l9!y7%S>$M(>-pMLt<ERB*REn
z&Z$nl*T;;#Em>2nc=&jOpXux(-|(YTY?M8Q1jaW{<m)h|*R1_zJ!~uVzEPo9;5o-R
z3nmupF5m1Vw?1aNIl2)pgAD@ZPMh6cCKK>Z*#$~|^mf7V8lg!s>gRWtpZ`6Qqj>hj
zuq2^u)sFA-+wf<+65<9;0};XY4t8VqPamW+_d5za$=7g%D(L@rL^McZk<S;tH2M4%
zdB-Fa8Zy8^L`WPj0j#3P_3|nWN4|$IUH7xMWecK|lMGswTEB(dy%;L}YSu-ktU937
z2A-9}aPmkSCHgm8<;o;i%nHhIKP_Ic8V<tE4z<?Ho);XPO^2;-!r-x}ZCDBwNv<ES
z7yQ_`a@jy6A9y*c-DndZ2L};{=WZ>ovFr2wFzMqKj)_O?L*irfv?h$}H;uBJUWqM@
zDJLXIh)R9E-RuIblC+V)iaBzp3Ff4t_E*lS?^RjLqg2qB48}kn-phdtI?6FsBMQ6l
zYOlr&WOW<JlRB(;``1`K>S8{#-FjsO4>w?O+a|uTgYE+3U4S66xs8K3UokEALv98t
zEVasn#>l(qD3PzQ)cX(ul~Q%kwwEfRtv;B~qL)1_5=VTyIxy<U!B%v~mhDyjW36lF
z+@K%ti*>gBTqsij{3ROgZAxPm9yI_u>K%p;m_LFu^m%6IViK#lEZkG3rQ>JqL%KB1
zbysw6TPCC38`lo9Dm;qIj4+PwN{KC054-EvwKPAb99>J(;Eu)|MA;Lach_Z~RUO<V
zP~>e!liH{Q=t&bZ%PXPkuhcc|{#`Kk)~-KPEQapBD{C;{=B!oFbWq9(eCGbUnZAK&
zVmIV|4L(KE@zePz{HCr1GE|_b2vl34h|ouw*#~?MSp|xl>Ydmq(=Muv(wCc<mx7Iu
zCl~DBULmSx&LEjn9a)<m*=}j$MheS#d50I`;tfO}$7qJt@z+bs7~ff>)N;t~A>KZ6
zywDG6tz?US9ouBY5AFoS%{bu!**{B`O5W$!ahAfnb1M$(`de)@jom0XyWL6aOX&~k
zvbVlh2PU%2mrXU$OVi}UEbc|Pg7m??Z=-I50Wm;G8x176>^43Ek~qTavq(5;a3z1T
z=YOHdG>IHTM>NPj!6L4tmUG!&GxBsBrd!JD4{hg_jRLDu7C=gUgy4h4lgx5K@-Et6
zz5_7~qnU~wI*#Xbjb2Ct$;U2Ehca1I1Xltmh%Z}}8Pm~wSK`qY<a0ca-F-YMH+V6_
zKg*-g>2AsNCqn}6gRSS4E2FKS!UtV_{@Y;Eg@Jcc?s>(xj%ekeoJEw_j5MbUL{z86
zO(mCD9Zz*DmkD(12*|Kknw(NjjLv-%<`d^8FSeEa9kLUkq*y+zcj;x_yRq@CU}`~I
z>)aam`cBO9J8Rp3+*efD8^B>G&n>FRtF{fIB{7th$JRdn-_zQQ8DCeF6n>GrvU!s4
zohpBPwAyI}UXF?}^b|-y+SZ52<Y&DruPSR6ez9IhFMt+UIU2y`EnD9?-){2D95oH}
znkKl0X3u;L8`h`Fuud{g?_=L_6{u8oZYV;Uw4^ZuN+S7u<c#50m!h<i-G0<|e9vk%
zzcV#lA-+nie6_CjFTDN_+&-R7U<z{0y?<9(^tFSSoO@|xWHHb5mZ6vG!;)=BJj$YB
zzSYNO)^u76tyhCbFE09j_h9)OJKNqHoMk9{((;7yF8-40W=xbbe?8mS6Nv8uf*`6~
zA*p1LFTkmj^T0}NBgM$GQSL3BA3K^P@*wH=GuF66_pxI=i9L3=kV)=dc2bTQLG!Oi
zU9uH?m|A|bBOB7}igD<Ne#_3-1;&DzEF|qQ`c&~!G%PmZx=+RrEw4v~V__@#8OQ`l
zIAU-OxHvz7$mArUK!HH(4I_dKBq%_##;t;f4+BaI8MuTFG~CxPZq{E|>J(UD?Fy-b
z1Z9S3cU@$M=Q?tZH!e|x#%c0O1Q|gjhYr$_U~T{0+et4F&QH@9TPQ8(W3S?MDD?rd
z^OpHOu3?C<#LP*3F)jOzr;$yMo=q>=puJh9u`@}#Df4p`Rdp>>aj$IbZ9nZ@j=6Bn
z=_cp5I%4d0kP6o)Ix%qr5^Tw-benq4fEDjubHI-+i(pMao=<?CHcd~o;XVVQ-#%3_
z<xUIkye|7qp;v%#`^sX|3e$+ZCf2KwR%pTeZP{Hen9%!MDO<ACdRG%eD!BUzCuv9z
z0Sv)~OZnZm4-x$4ok6E3l)4>J2vWt7yOj;u3Qn5y#qb1A>=Ax{a&sVzT`Lv}CfL;1
zX9tLm(Ys~R6k*=w=msqlZuKpz$%-Qs>0Z~{NMki#HAUzFgD;fX8lp5Uy}@1qGEo%x
z5`brgHpf@1vc3+qSKz>_wMY%%NR2HGZIDjSnGyN513%MjNX1CD$Ad+tpYt5qk<z_B
zoxcQN2cqw%18VhR7iyjpqo@Dk$9(v-r>kensU$v^d+VP@sv3qBlJAad!AscT{p1x#
zKp+$z6h;2qo;CQxh5QALIIhx!?e5wTk6rD<({S_zy}e|+*`LvggL750(%k#Z;5AHu
zZ*`BC|17*WB1X$}o$|K}8j-Q)@iph^7!?$^6tk{F^@MrDo*9<B+^uHDgFs}Yel_=?
zx5PRj<$s_<?;B=Wn^j{flZXj<LX<dhPgC-Pt|31<I7t$N6W)cV9SzEq@(u#CGj5Q^
z`&M4*@4qQHJ}l~+EgGTFZ4G_}3r?P%X2vO+B!WSy+r}un<Wmycd*_Jj5=S0GKQPic
zH!VUHhylWKZu`F*nG8;W1TIKx=}{}cJ&0q+*3nhN{QK`lNivQPzW8&Rl~5B8Vva?3
z8wP*T%K7Ex85UDpC*BdiM02y5WBE$utOoHZr=TLf;8o>r8Okd;p@1z78w#P%eH+r^
z#`^$|UQ(pBVR$iMWKl*JvP6Se8%&kSib!9o32v4ql#m2oU*cBh$G#;Mc`mNE(!?FK
z(@otbKD%I_Mhz&BP!ts6<E=0Igxht+ddQWP{z9d&YnVr75{m<5OLQo*Z-E75d506H
zBTI$5W5kuzk<jeGZk&>+){Lz@P|?E1+7b=qiHSNj7ZXRpb38XhO`L)!tx+F+ljmFD
zt9n=zaFj%Q!5ef{YT3ynT}U4kGDKy~jnb#sWF-_C;v+9(zGzs#WZHB2Wp~o#iIzzJ
z<$!f}dM%mqxI19WD)=)0oxAX~HsgXY*!yLSG~8TAPa>WDk@k-k5njg8h<*MKE4gRP
zr5jI4ThWVYwpA`*%{v21;&559pAyu*u9I_D^UE_YBpqyb7M}6QUbxeBmx*9^trN3V
zm&QKca#&N}rE<4o(xJxhekGT*d8V1p`Ex=Xt1rl9awNc6SMR_qH5a*fe0eH>aO1rZ
zsM%}3u+YaOmsjz{_0#%b?m_M}2Ppk|$7<uAVdlHIcR_E(#l4q_bgpiK>Ze#*B<X`L
zSJQ&iw=~v`0ql&Im%g~rvcKK$rokWXa;HsxS&X1f)bUb2pKF5qVlq8{RQNMS;}aq?
zP`eC}+c53f7)~>}Bpbg@`^(+#i%##N`Ec^C(qxr#dR2ks^Vt#U2TOJS2p*YQe<!XS
znL<cVmN&hKo8$n~Z@{l!vXocVk9B(b^vH$12I${nZNhvr@~dZ)>!TOrLsj$aYS-Hc
zls&?G=6r!~_fmSVJSdB-zob<n|EylJWxLrl>mdYIfnOecsK1_Dl~E=GG0{NgO!q`r
z;Jug=DzCfu2q5=d5oM2-YUi!%5b-Qs)^YuQUKd_D1wYeHf+#rEi?cE->42V|QUl&B
zZtIVe-13~za(Y4-J<#%ZGD~bB(G~a=BlS0ko+eXh5P{v<?h`8Y*e}*2!}#egyInmf
z&pEUgBV;X@0ODS1Z(C&1pZM4F^|kuqJiI`EF9qfY;0T(1c@C_as@*$~)ihNpk)m7G
zTx>`<?juTZgtd*)LzYHRAd^*!NXHKF99HzJdRSqH!DpKdvd@96;+lYsa%tg9uNmL*
zkDFZ5Ti~_3TJzc6T~_q_*NN&^##sGjT$}Vk@sz#CvC+}i(&wx-0G~Qj)Cd^VOGe&9
zGl_cBz7dD=IeM78>)R@wXe%aBwO!}y1665WFDaoQlJ+2`$=f#7f}^d#IY_(U?$W-k
ztkw>U*f_SDFI@)-et(#3DQ5UU@S7^8alL1C#~V?n4P+;>qR0;PQq{?STFpkXD^K6K
ztQ#n_O+Fny+4Kn0+;|VPV1k<=IrvD!9U{B*da3v<F;KAY$Z5)?QJ}o%jPz->^c>FM
z(B}Tn6Hxqi&QESu>C*=IDHxQ*S2V9r0@c;aZh)C0r?=<-s`rv(+Zd43$YjukXW2{B
zPU@b?5<y&&41O$#>8F_1-aPJngLa^TaX|8e$MsN!S-?>bJ<*9=nzSp8hiDey`^CJ2
zx2mcdIbP!zQkE};tu-5Uv{pp;<|#cT%-c)|H?0E?YS%U3#Fk*yt<Mc?-*d{r+~_$O
zXi-KdURVHF2>{RJvZnK4GotMxVcEpO0M#X*N6dB^=bxT7U-xv@z$w#SdF~XKjPEvx
zOTYiBq!;0>tZSWk0pGg8+@v%d@gNC4PAORX9|%CVh+IzK@wjqCCo;WOY&cY?M@uK)
z$AYHNOY$*dcR`!8qS0gSUUv5TxDB0(FLpYG%QKKW&k|#&J^yiK9aXeYYkXzXegKl2
zL&~^!2gu8G1yZbQ{WO)%yLk+zrNG&O(dV_xNE-c(wDTZvCl6P3ENWD7y#oA<UMMDM
z!!D99t=($Gx2Od7-U4=PO-tgGl)#7UrYYU=Eun(!5F1d_%^KoW<f8Bx)q%IfG+6Ez
zw|St=MJ5}-Xrbd$@Mro&N^i%4-JIulURQv{h3!M9f{eW4uxZI~_!pE))%C%B?`jzL
zuyyTB{wM;Ca^OjwXhe)}ve}wA#r&(<&l?SC(zUXTeHeQc4B5_6GP2q%J;;BIO<60s
z$1nQC4Hi+Nk;Z~glsj4~6a({c8v-!e1yHop%Hy%6d`|Y;bx*W84ZtCf!$9h>4j+gh
z=eZ>0gE!Cl8LCz*wXaj26BtsXO6Uy}R%FpnYd$g&6v&-3Yt_~tYcpBHK6w>*@T7v7
z_R;c|mMQMV(U$oPTW&P1y?}>)40KSPwWV1HYPHXDp0|P^v@q=QBhfrq6ZYqPXWg5F
zK(9MZqZ$Er2HM<v`eVVRuXdJ*!qArzl%tO?uCbvmKKxd6w2Wm~#dtP`TJaY}@SO&O
zYFVUY6QVBz?$H?$C6_zF1+gzi^mk)@PPKO}qssCG_dKIemQ`IXNWaCOs7R_Hi^JNi
zorxWpneRpC7n~M?>oGd&LORPu4k>rWPJ{EwB(m~6%X~fZwJvBvM|H6?xA~lEYAl~q
zSS{NVT@v6$o^FuzDX86KR5M*BCf6<HpL>XQqp?UhyCH;_oXOI!QQYwt^T1y<6_F)_
z+)-e<aNwAPQs}1az?JOs>qRBSU-xLizl6h$inE4f@H^TfxKJiWZKEYc%h5vEiRK_C
zQkQ>(m?^w>&)p|1HpaATz$7OuGS1848-B)Z#Gmml0#2fV^u;V4+1u6|uRboCe#X&O
z__%ZiXxIAP#^3yyQmFU1Nzc#6t+!kuY;%huKS%J7caixaqi2SFoS|SG+HF4hoZFiy
z_*1K(_1?dP#<q=eSVrP!|H-@jB}T26*EKbpU8``lpNC7_1Uq}1#sXcQ)ffVX!lcg{
zct7n_yl)6-7Ie7hyu@d8-yUQmn;=19`Z+BfV}xQMAb3|QQf>98!Vppl+IHjGOp#OU
ztAEvdd^TDyfR>Gc@Mf2&8k~Tm6QuC9t~nZ!Sv&^#YadftcnG_9DR0<000Y|?8f)`*
z@tTwT;9UExPrk3$%0QzpNI8${&{12V5<0>u(L@K1>`o13vI$DpeO*J1xuYO<Tn~X&
zV;G@4u+9FzFXzEswN_b}|Ei0aGn>UedPHCxfN+G7znF{b)w_fXHFM+#eTPhWIax0P
zWJEzAfnN4alQ;3RVfixoxGj<&BJL|-pc8{f%DfM-*|w~jlWTNxQJsFTZ20uEl87Hk
z2t(q{J3e=b1Fx4?*5J?GYUKX1(21+y(hD|YbqRi*PDhfWR8kT{bJF0w$Ap$VTPZQ!
zUv8^ejhvQZT)`bRTv-k1SylL?jprR6SEMe_=3AxcoQtu>*rFa-{~u519?o>*|NqUL
z&xa|8kux*oFy)-Hne*m2Dk0gZFo{X`<h+=(m_x(nOgToSBy*~XIV8zl$RVLbrS9(g
z_x^nU`2CaX%5`y>x7S{W=kxI-Tj9b|+x(>~L&_pg;iQl=7y}~;ubc91HWbv?0CGc(
zA2XPpt>Tl`1!U(9Db(V!sS||kXPR7O-M2%^S%p^N+pgEu-V0B;oWw1P*HqpZeZUOd
zpEnpAZ?-SkP(aB02fag(8ju3zt?~L??QnDNA1n>EGnSTw1I@6&3Qkuwt0p-1KhWv6
zh@qvhc|-Uy&)tTrkf-Q0;TE+QcZwg3i|cp{B<{?;I2+_<zM(h#I$WCY=0(6~3dFqm
z`1nn?>vd)KtE_HSf;Baf87IHSxPwxxhcz|w55Lo`#WtLJl*EfF<<2+JK}Czyq;)bV
zndeG`{s*Cg`L5<aGX(AZ)5oe2`MZkUUs0hcD}`h^2V@)91<I)z>GZcdE}c#K_bcGG
zGkvWlAUHlKUa~Mnh~i}p4n`on=2dzoYW~)w@A&cen@gB#wRDv~d|ds>jnknz!)dAR
zc`JP}^i^?X)dF!6ezV6Gdg*5qaTA^udBV$0t%XaB6jszC(!b}0;@_y(&PJeZ>RFmR
zVs@U>5wqkII$}e0s;7p9E3dWRtbfN-a{Z0tiyx^WO_OWmsW|Iw38`xBQ^BGLZ|kYV
zS4`l225bZA(Tf02m>0P}+n&FbPdAdF^vFR*44i2;QYs(sBQCjjgRaJ~FP!tu&E0a`
zH!g_g)wb>#ikEC0WU`fS(|90I;(RF?X4;VArfl$&$yJpe^WK|`MV0<fv)H5}59862
zMyvIjLqEdi{7c7unOHBAZ1bFw#IbrmwYl=z1)R1cXeOmTjm1C?BWZA`R^!JlYgxIJ
zBex6B*{@dvXBI}1s*LX}Nz=`IdEapZ#~&Ux_Y_26rR$y;Z?iciP?r49dbY=lvsr6K
z8m1a8y-w%wT=ILqx}9O3JSqiq3u5`n2nP@8u^(ooLmUj)!9-3x)%e_pFZCH*QnZb0
zAAYY`I>cj_)=Nhfl$4}*lRwqL^e5e}tCm$(o4m(*xZ`tjRtsjJLouJVWJb1Gr`5t}
zplHo|B-hF>c=9Qa4`^UD+jPL#Dxh(LD7GkzDX#0|KzHN`zX(g#;)=95x?;Z4k|VK$
z3iUW=!d6BOEAaIwDg5^2s!pUss=J5vk~d%daDad-zr9BGe`oMiB>9!3KqyOH_q{>y
zFU{X?F^`Dkf44NBAq-COQ@yRwLUsK_(W&*G=;3Y57sd`Y%K*k8AC}}O_<wpU!zeEo
zoSNOSeA={%W07kp5ar~e65klm;_adm-WV{Tt@cdle?iPmsk9MNu9&2pvjQcgGsZ!~
z4>l{w#ioqZkL2qGI;Bg8UTVYhZ@QKXPk>rq6sw8XHlxZq2N*d*j_)4&WO$G!fb|_&
z3|&Cv6oSGROp&)%IZfu0bHkDrxvx{)5e4js9_50vA92t;8He@TJQI5wORE%$RAv7b
z>SEt`gg<Sf-#jRZrpBM^_Xokp=?;Nm0!y<8S7<}sTzuaf-7c!+bCPD<326D{O1QAs
zo(mN(Orjoq3<5mPnrLmbwBHQVIR_&f%Nb?Me^k@Fu<grPo)dc^KDHZ)a$JusP)5Z}
zv>EHJTv}>;9B0B*Z`8frk?ydmEuTVjiIL74o$ovhJ6iqA+NOSRqzct13zP6&HFb@b
z)<IRjyWZ<z<9o&QcJQik#Zu4xO>akvxi7QQO(*H0)?sTs(m(V;%3$N$^+}Ron()Y1
zkymfJ<87Y1zjc}!i2fYSfQ%B)Q31@h$@ZCq?8H#df^VV5^EpD={MO*LnL&UE=%}@l
zp(ZSP0ICB6C|07ifMX&Hq<BE2S;K>^x}*PWf&uaTuS&sBaSM+hU4w8s+7CLWGjVtB
zP@>yI?-$@#@>#WHNeYO^1Tdl8fr2n{!Rk<cEjmy055Ig|4ZiZ^1EQP8y8>Niiz(YF
zs4zB^{Mlo6-X_G|e`EO({CNq?_|oLa!RH9$CScCGGj|_OPWyv5`Z*&mHc#7r-9fke
zVEX>B0d3N3-^r-_qW#nk74bCG%5lVkT7sA2MU7ZpG-*#n7=`U-shZd_0DZS_yqE!k
z(@mKzwf*db&fCV-@~q8}59tsK4i&`$q3UF0m=<dk$85>r1`DF^;L!U+W1VFo+wWDE
z*HJPD(gSkA(Fp3}dk6(+-No2oGw&<}<O0CgAH-0swE&skN~Lx-jK>0t@>A4yXhCI+
z0{`*(QVj6v7KxbWZCVW%PNBNL#9-9~DF3uM1o<4RpFT!L7^bthQbeH{Wg>J8m;fhv
z@{c_|18)tCsb~%PKSGaLq`^Mrv6-(J@k$Ei6pbfNZwP;*Flh$8*3VE4yo+ly&MpE{
zjqLm1NsChd$#=8UyQ61NgloRnY|y3U#$XpZ8Kd0{KmL_@2F>)+??Uvnj{m5;fLT;F
zYE(Mh^SO4J_kmf97uoXDX1ibitR5cEZ-+LU;`&1nn2Fw|#DjZM?}`0=vakJNQYFVE
z`)yRG@V>lsK-r%3GBqUVD_hjjERG&#PhF|V2PE$L9j}xym!Ke69AoD-Eg1Bjxq6QP
zGE5xZcGIjb=Jy1~rnG%6u*n%t{v{wuy(9~19u@}_6mZqizgx{!GMrDK<Skw}xopJk
zBq2n6W07iAPZnz+1i@mgS*sCSeo>zn!+Gv-)w)ltqCan5Zj*+B5RNir2<1a%d%UB1
z-)hq|LdrXPItI8oOu&GuqofGPraF}KptU_H_HCf>xCr8nYwbJIX5Hs%?N8b6cp3Q&
z2q=fwQ=UEQ%%ntbv0T90Y^$zN2mHTwhl~u%9oi7?k-V-no7=4gP4eB6>100l3D1k3
zgi}76s=)Puqk;Ckp|OoSOt7G!^Ujs-)e|qeV_fa0hozmMf6Ci9z><<*(2$V(viuAV
zBZtVs*_D1s;ur`&6Tjf#?9g*`L3OcXPu>t@84;XGzVT6$bC0!Hs70eiS}VnP-fvyQ
zsA?+U`>hUn_${~Id*^t$Lo?9L1h>AXsP0r*i>c$rs*1eYb5$KQ*Mfb~CkGkqzaG*B
zsooQ%Uf+C1e2pSn_!AjQx0=o74u`1lC7CPr1vEc!x+8(glv#M1t1&d9)}ip6YL2OW
zALHDivePG6{8#Ipm`yS1a{Fi3PYJyDS$hcIhm2^ewPQ{?9*am(@Qx*zjx3NBduUQ7
zdou30h>#QU78~0^t0zhaT+Mor(Q`k`mMJlwjqj8W*>SSRo;*$0BUO6mM4y3=eygX(
zMu1*DCE3lBMN_E&R#kQ--xI=r)Ca664!|KJM!FpMXX4o#h+cim7TTjZg@kL&ac~n)
za9kb~NT8dx=1~V}G3&v!bKU%K18V^XeE`N?&~z!7ejUmGeAf8TZF({1?w+5e<?sW^
zt0gZ^khU8<O3qgLG=DSK%^G}V+He`*AV`^fGW}j(l-Fh;YLmh?zIz=V0%o64>;_dG
zAsf=!ASc<kN<Uik`h&iCXJ7QQ%MonSTAY0*;-o#8%IRGxQA?<YL7w{>AIG8$2L|7$
zk*SB)`7&!nS|da?w2}kHsqi(fGYm@PE86sxCj%+v$LpOqWV^Qz2=b%~lg2u>GoxoT
zBONMV31&M?rnTm7%h<<g?aklQTIo~U-M!`zAlcYJDY)%yJ(OZC2L0-$)b}ZRou(Gu
z(B_k^=EXwQcB9uX=Xq6ke?i!1MMblu?M0#8{-t{OXU#b7(My!e(Q{7r_H45b9W6GO
z@H7pC)&jAel+*x5L@q0*D27vhxu)6bTjd<wJOK9Xa+j~XcTPy_WovV6=exY}PR&B&
zU2Un@)B(E-FEjqM@sXmRc-a=&_GSE|Slp6eKC{6U_9*92h4F3W%1SrJji6#`Q8yzE
z(YbV=(Kgks|JL?D(2%xj{cvEk7cE|)Kr?@OzdPgoNfB1~baQE12dXo91{)O>SD#;@
zj4!m<Rgw!k9fMj!Kqz?Ghqpt|c+GK^ZClvkXR>4}frxeIu9JApj_H;A7Eb$(S8W2Z
zH@|Fj3>p<~So*J*{cXe*Ic=IPY64rGzhwh#vN<wCXVCZ!Iz7~^VENX~Vtz9qwNRWy
zEfzLBBWQGl1&e9Y_^rA@ysJIRKj=0^&bLHBTHAZ54b>^YXoFwsT07uvS=CDl<9vd|
z%yQzkT_HJMhX*^>pmnn8IPkolNLN}Tn;g&`OYeqUG|w_=r_9f4fdp6D%pU>o46{>I
zSV$}5+Qb`9ftk8{i}SI}6f}?jDA<FmN!ehi8WB-<$^<jtBo540>RE0Exe*0A*}3*R
zI2BnTW7{3<Deh#vLx{nkH^s;~MJdRu(l?6XV73QQAh5cVd}~f7^d9J7)wXcuooS=Q
zkYb4(!LWQ+!=AIED^4-~uZaGiO=9M0ZZ`JWIWfMfdHrol?e^u$Ec@N#ZG>>^&$ya>
zvij1j;j~m~(MNZ2Z~IpNCav=G`lZGN1|kNK9$}#ivD*hNSthlrYW1`!{xIXSZ2OF?
zGER7x4v$3)W+W3t@OCEz!d0$$^6=89T(N_6zS`r_Ewl4;k8QPXQCwXU*K@FHf(G1;
zjV_h%XZl2Kotvz>+d)n@=Y03oORQCyZlj^f?;~$ys;_FME_vU>_xOy+qCVWIX<C;~
z8xtr%m4T9`3<rnm-E@LGG`}?kdwj3IWhnlyv9I|)XQyD&e4|HQu{P>sPt(6M(Zf{I
zzu}z~4bp(?mEt7g7N&o|QWJD#tdbE}!%03OOKFRxoV{38d6Ee_Rdk{X1V;kpXYUoV
z?hBn8vF)5A2%4F$eP?>JuRUK_s~!Dn**xN8oEIC_Q_!K*kcy%+|9A0#M}Xl(SKIX>
zB&b-5>ZWgWO(a~%LA&8b^{xI2w&TiSUfIuc6J~yHa)03~OFsQ6#{N=Fj(|`}el`;)
zq09j{kmIN&e5|g8=xTR1#pGOZb01v*^n)dAvjy;Xp3Y}X8dzt2y(Q*O_>@D}XcEXr
zS%p#>;=pS@{LQ4f7W_Y5Jr7T8VAbO=ONAePz^a{_M^ib)o(m%zZKmQKt4y3k4G-N@
zf#v%+-~97AP}@aC@9?TO+%BbSKHWjY1#f1J-)z>6QCbO~agiCm_A`iQ%eVVy&I4Ih
za%?30kwew7Hwwa$Sa6V&^%wXosGb5s19}<g2RCK023?G8RCE*6hv5>|Xn@g;gocj_
zCgU8Vf-ouW!%JQ+4%vO4aDB!4dWvzWMbk!2yIIHV9D2;L_X4LZ^uny4(QU^9T1Y4~
zgFy#&F0yrqZ8w4YI&}kD1$;56Z09KB_SN&~OK)wq9P9P$L}ujf8PELUELD`?!Y!Q}
zUpBN8Qk0+I6}IYE6}B3eYBIiv9z4A)v+Xkmaj&5s++9?=`<UoW=lTy6@!-cNhuMlp
zUK7X{o`lE|w+OpN`_(qV?bw-qyk_=qinDG_8Icl?v0G?GKFEDw16Ld^z!(qrZfA#v
z?)F%;%VljGR?JPVjkQ*pdtM;D36Z`-;Yd8}FMa_3gnj(lZG05nl&L_()hJd_2ES@c
zi9tGAoBly@k`&`&#RgO*Mm=sKIHu{7yw2-)EVBW0o2c09#Ucn&7!NIHad8<<VM&K7
zWvur(A^C^v`QW;vDM-@cI(6VDbXW$$<ZBoJfrxC?ybbGPQ38?*kBdBEN4bOBE{*)7
zZwF44=u06pi?*x<&j?!Pf&}#w*4q~roXwjMX-$>=KF<(UK2>ykkl++m2h2*Gxqc~3
zy5K_S3e&rl?WK=P=I?MY5$uML8^FE=I?>QC|Gx&QOV_IB8q33cvoh@2p5{K1wrsDS
zi~80)CVG(uh1^X(F_<^bf_Q4QwjxV+$XJK4sKu2<FM<|$p<=bCd5--D3PAhumSwp7
zi1F5#u=@CZ9=pUbmsGpu03c6C`H3tW`vMfSw)!y91Lm6_hH|`$POAwm;<OcJ^!txC
zHQqu#ejw!e1~2hS#Xh64xx4e?qwh^x9nnLG*uS)2IvhQWjR-!U4(w3Sy*wtl6q0uo
zVHlt;K6+{_DJLV=R7>yH+UP;08tX9Jv$4%9UE$e)i{vsxe?-gIZBX(rFpwA+@HYME
zerz(~yFT1Nu!&Y4a#$B*uwoI{$jlas<&E@F5Z-GNaKLh?2#pBij-~g>1lbChsPn-N
zX2dVzk%Wt2x9lRs@#a2h7F^M&ooQ(6ud3+Bsi8(a;A@bM3_16)Bvb<$*J6IFokzp4
zn<E-axJ~weg&j8AkB!c%81!jcrOn4Rf=${#UD<+RaryUGX{F*^G0QzlF{>2Y1O`ZJ
z90?LF4y_)D09nK{cdYrN%`rJ4<l{hJ^al^_sms%o7;*7kH)%J!D`hEoUQmz}*Qk?^
z{olQc!mA7YpQBY19U{zGCoA{&Qx=WXTA_mLJ-V^Z`=x8Oq#wpT+BYUY5A~e$x!9lc
z(pkq@?T`T&Bw;G=McuTv*dXN#M>DA12=R~SRtlYD0OO5gS(Ej%6*6QLLUB;+r-u$g
zcyoy3^~#+Z+JWH{?`Ld7(2wVwXQP^J;rh=Qd`x|fzaLZaF!3ddow#!NV4F~2G~*cX
z+x80^e<OL$Eq$TNd_HjB+1N+njlQq__X)9cx9W>`(pH=5K4;&sH@GJqc5=cXC5Uq=
z*|;Wh*H%CByh~;j_?2&Kq`CbjWp)-_e61a8v&u9}NmmC!vEuLXt4xge2et*oJHOq+
zdkm0qWQ7hk+(S>R$b3p@L9sh#Ypb%NjL@PE%Df5n>-m!Eel8y_H6lHJT$I1F9W-|%
zm)_ob-p$r%{_X}#(=O<rAGq5kHLcAgtRTtTq&*`$M;qVH^s{V^GB5veI>aKM<ZD4&
z&F1Ir$+ZfjCbA#U*Jjs>02HN5v*8zcnzW@WtG6(z^wpR7#}Ye`>$pif`&J&`yIeJ3
z+kjg?n(tlP^Q~PN!>!GV#X4aFM<A-F754GDF`JPN;_~0`Vo{gWzQlMGUx%FOtFI_D
zdFN5AXL(Z0b+)Oo^H6i!dB)XG&#t8)r{`QIPF3cw`L1F`R+HSZUwf|i+xUA>7iEfI
zvq>);&GXM~_1@V`khNqHk1wC-?=ExZ*SnR4UQU~^^a=`uXzm@qYir=?B6&2CnnIgb
zGkVGnNk|I^2Vj2|Fuq%?I6Oc7fN;T7;Xjb#;a26K*p|~O8zl0&2xZMn7N<H!@L6nZ
zxpG@3A}eEgAi`3wDXv5g@QiR3nXE!V87V-H<ge-1OpxY$>I?<c&lGuWav;rJCuMTx
zq_L&-g>xp1X0VJH6S?oG1v-(g1$hm$Pa{Q4<b?WP{co2ax+LjW@dC5;?&mlx^$s#5
zQ=xfil(HVRnXSUK(9?FYYh-x>%Oe&BJXFa8oXpcLd^?OnD7ZPvVFo)Q3s6zI5~hXI
zUk%u(hEzx?ixj!P4M2&1#Fb;ZE-Jqdq1+lgX!Llw&{(9c)g>7mUd_g-g)a=@1=yaH
zE@6G9MGDFu^WFzOQJs1hH<?7UZ%~!~JwSLVfY1w*{WS~q-!D!XMR!R|+$Hqwy7nI5
z_W4)=)Ag<@Fz^d;fS4{H1U+QQs(sPE9Cmj1B}>Fe7A8KB9ze0UF;f3=p@p;@mcpmx
z9invCfU&^Q#nn3|(K?@B#str&xxV}KxX>r4Lb__7U=g-9+d+LxJ6q2z&TN(|@dHw@
z{cT1e$i%FMLZbj$o@C7)MnLtm>{GP0@xM1wf|lT>&uw#SlU6<q;l*rgnWj_gbxms1
z^ri&q*X|`HZ_+l}Z%Jym9%(8~dj)Wxh9~CCL@6Zw#xe_5efB;3Cq?<tAI$_pZ#i2V
z0)e2qs&+-CKvJEbnpWm5`oGfhfg^tp@hJ5>Zo4mwP7)XU-?S}Hh6^*`7TDWTx^Mdg
z9?izs&+NdGS2<NZLYoePiD@0&ziEmF`3sx)LkTC05B!@b;_kztUhwz!W-+_c-j6rw
z@lCTAq9Yl@jz|p1LKp+4reVPmpCdp!<8=)rYh{a>j=MNe^a@!>AJWe-YPwHWiySp+
z>1!9T^l;Q3GeGmGWKt67JWTopK$%Q*0I6$4i2H~*d1p)*6$(|~_T&F%F7>~4b@9MJ
z>AoB1wCsD5Pt^LORYjk$#Ll9kM<BRUO}!Eu31mk?QqXlV<8&^kMSfn|?WHM|i-WI_
zPtm*z_zA=0MlP>oq)W8(x*eR`)aQyu%e7gbi*qOp7qxuKt@C9>7EZlb4I}dlYhW#A
z8!O_WH3{&<l!68#M)&JIq#43%K7C_Hw5BFK6jOKPc4^D+jzdP@8LbUjV~-E~_Y(a5
zw8e}sI-(R>3InUYER@wBVXXR%r>#>|eH2;*Uxi&&cp-$L40hU};baUgM6$>P396$d
zY<Z_y5<Imz>Umz6#{a2_3#)6pfqG}Xe%HI8LNI6wRXGEsF%4@}RO9v4uv`AWc85!4
zVI`&?1`yS?rx=b!_f0ffvnjdHT5{_mL9=Jw1zNn$?`~rwr9x)?wElb=7yP=?Y*sjz
zqu}IHUY_Ij3_$yDqx!Q0{^>PoQ#F>?2-QORP&{<`P1!CN2|o<41jn!UkPzinFoeQm
z8*AC7kaw!rSh1*<;}#_XRZH_17oxs;X3XVOkPmI0o*B=$3U+!Fv1}(ju(VAS<oX8c
zC5|i;2)(LWvtGu#THRe<uWUrk$Q=bNy#~yH01q6f{7TvN7JULn(^rGZGFW=uG5<0O
zd<Ajyt>1$_S5w*cM7xqw#;v3zmsB7C;PObi{5!Z^iYIeo_g3_i#zl76_VZIK%_qb^
zc5FzsUktV+zzti*@40@Id-6U!H76QR4yHD!<bC)j##&^ANHbNv%OP{cOkqx746KK@
zZMJA2L%t4%DIXG4#sYAccH4_&;!jU-<pvWN|Aa`+tk?53eUP2POZBVR@2NGiqp+Fe
z41s2Fztv+-tX1*wNtwfzMQ^<NiW)^m(R2r^oAJOr$;BZ<+5sk7a!uv?Ju~-TRWS=y
zsw+%^fG%$nJuMbLodxm{j)uP1o+^6}AN0)@XF)&Bs~I@x0%Z@+n9nM89H|NQ%~FnU
z-eYUHaARqe>X(#<3w-ZV#(|r0ns;Z`8q&7hwDm_pl6`lAw8Yx{*sy^k@Y4TK<;KeP
zwi5#HbNYS#bPt+8r{_th@3Jq4MCjUyzcgvN$g2-1I+ZOXwNVRt&h=&G!rxZY*J#&z
zI?h-OBwK0?Ea1jU6HB$m)j$ZPqdT#8iV!juCrKKk-0mG4_XOc*`&Lr|g64;L#g~4~
zSQL8nAgHXd(8b!yRh1_Fh&Ju}&D2)Hu0zIA*vzp1w)1^d0;N*g!7e;oGU{yI6=$pN
zTIJVjv!`&jN5NkyUV^^_E2z(hAw4E`z{&fGpXD}YLZKjqN}{>C4#-u$_wJMsJZ2eL
zC*)G?XS(fC;y+R0Nz)iR%48Z}3pjD2KiI10op0^!PDlOpx#5Zc{hZ>}rJ1!^=-Z>w
zJi<zgRcNjF{kQQKGZWXHMLmA{w74WHJRW~`-k}}qdtOcODz4BD8bTJQ+*(;Sooc0s
z;=$?HL)9N>G4o0B&58ku={Vl=o`CBGo%>pqV*T=uul{BKRl3YS+jsl(j@3GSa)<i)
z@YS#b2$=Kb>He0c?M0QL!k%>pbe>NUJJtfJoKiA)MuX-CLP|Yam>pb&eX|y#RgYV|
zD3Q^ml}CA>cJkx;;DF|1Mk`VP(7va^P^A_)f*!2(zj;*4!c$e=cmCLt3|(n_>6hP0
zWS&m?yW3LRXS<az#z()tYt*T_hem{gZR(UnBf#dwLFOxfAy*OWU`ZVwr1OK8j*L%z
zd2+z-VWgSPTwd6+9z`gwG*Da>ee<U;Ysmb5S|Va*@Vk#0(yrn45Ucp+e2K;U>?*sa
zvWn6*s!}OYu<dhOBHzIe5m*rSUA0JfK>Jbq-%OCew#Ow4(Sr<+E05Rj!L%>-u`Wy3
zR94qtewpF4O0Rz#jIxVUsTn})1;!NYxt2*iV2b=^IwTEuME6+Ot*`f2Z`=^-b$+-h
zb+`1vamsqzS9$vaH+O_#Cea-VqGP@{+vl>>^<gH^Lnuc=g;FHb7SauTEQs{B15|w4
z9K-x>`keUab+!xxMfU`Tb~)xPSZ<2{6CF(4jMAsdc}ZM_^~ltr)^12(;@|84fl8y2
zPT1k7ESb9acGA?x^2cXaV`X9S4NeBEix;GvqDiUjKsRst$HtGmt`CPVfwR_I$77)P
zFGJKD=dN5zD$jK7&C^x6XG!!^>PoYDC~mNW-YMTlls_Oc-0h+P4(S!OGn<q2n{jP(
zg|9=HlQqrHpA{W6I|V+n`~3l6D%82_RC#`AWoe3nYF-kBYorMkSu_1;|1nE#9oRRV
za8P)WjO9^1B(Ub79x&Tc;R~-w!OF4>t`s3Bz>#9Khvo~Cvj&eTxVcEr7VOJ#2=CA4
z`2bJaUuJJR=G|XD3QJkO-{f+CH%_*7;A#aa>xN_D!ncY-?)?qAMoT}N6!9?xvb~&I
zok<CXYhKmJCyuH@K?Vv<{i8S5$--Zk^Mk)yvyfH5TO+87NV5Y#*NWxH(0;lt*kx)8
z;s{^z<f$_G55&~7#NWAb7A1DC&BJrkZqtVG>Vr4eiF4iMqiAktVK*Zl2LN?ZgqpGF
z6<yU0@F^+q!d2by`gik$4YoPS4{|BUSsI8b*T~|pm+;Qe(Bh)Wd~lbJaE)#NCQ~hD
zo0kelZd!6!qb?L!lS#>C=X$9LFHJu%R?9<u#qzW=CQ#APzSb99@U}t5NS4^kIfDAo
zeDw0iSHVQv_Q+7>4|x7T29MC4mzD9&pI!bS;4f+zDvIauU!N78jyq_bj#n!gjFA)t
znxj&~C9&c8u?)j7ej}gjdnUMYE_Diz<MMNj`a;)}P|XXA>o*D>b1uH+HxtBmfW8CA
zy8c=_=aZk|g|P;U1-^>MDX~mvR53=v)lSx?@uhYWcKyOchRo0u=zBwqzwM(w*#?~p
zP7qq@1S%eG4NR_<z6>Ftv{7)@F2aFj;XT2?kS?j2b}_rL!8a;L?%+$OSNf#(3!1Oe
zvB0B?Cvq=E9*@D#{{+d(r+Wo>U;e}nnLLN6$II61TqpLl&}Vek9(t)yu4SDlF_Q{9
zL?ToBr)&frt$etc>|m@^8zAfh);n(fn{Z}KP~>RA#8+=7=R?75J<s2mk9g{4`!DZi
z(NU4DYVG3b@#Etv(*Jf3wq2J=K!>Vw`gh#tR(oEiLQLBl&CO-za_3$)^08-_u7BZn
zOnF0*4WQ5g5Rw*@Gy_hktB(K+$kf9;A><cCmmEQBtus}G&5kR@9j%{kXcU*MQK`m5
zGuv5v$Dx9}X25~4yI2^w4%Vjdzm$$01^Jn`Y}@Z<%>F(m-KIrEO>`fH5I_H>dT7qh
z5MQ@lS4^loWUi_rK^!L_B;Zd3_%Abq>SEB|L&Lvt*>B@PbP(?^!2}PiwWz#<EEid1
zsMuIZQ5~_wk&OY)QcA@+1z7cU)-z*<`oykl$y{dJ#Dq!I2|J%Fs+yHIRUT^UT_f^4
zl`ORM9bBmNj<Q+p2_xu7Coo|>N;+>V`#iM+`=d`$QxDsY4>>KuWd&qpSZZO};MO-0
zPbI(6H2+n2?75s#fXmCd(WlXBYd$YT?9`|(tL(aqo5g)>`6}-e;$Y$3OyEWFH$dD#
zEx=%-X6MGzhk7@gG*bYf^D@J0JCG$g4r*w_+4rK%u$kyC4Xp-IE#=z|T*r|ZaV#L0
zOOrnD*K?W_RY2`))ix(w_8`85xip+JEJy*cGbCR+5K;r{1fR4I{pZ|Bi-c$dT&WJ>
z%*rOM^J%}~t)oQ02%0$UhrBZRJnWGHkhm0}S2am|R$qGw9w%`nz+aM-H23pbxTt6X
zl<9`5jy!ht+DWs&mA-u@C?If!^KON(;pKtmP>*-tq)1BxEyc7GW)^aTly#KagfVn(
zuFQq#rB^d>TlCZug!Zn1e12l{t*-kY+w0@LG-*-^;xmDjbr+M3&LVum!++noygPrj
z)b1OWnbTkqQqv+r$m%3fyj~ZvAkxK(#hign(AyNY3`<{F!kJnOl5n$(0KqsvR#F7!
z%N+kEfeI3SVL2Ex#RKr>+WgISqsu>EtDQboI$Ci)=j>QrV<@fNsf`pFnw#kw?ZlK-
z`UV7)UP2~`-Xcp;JeUxvHkpknz}ug!1zJhTo|*Yp*9L-tIYk3p;8B7WXpOUx<H<cQ
za=U<9wfT7TEv(wer7=#)Wb}jECO*rPcr;D6>ITmygpaGGFjUoZ<Dvu8HW8GT$K~C=
z89kj4Ma!n&2L8ya5)<8@F{mj#bjCa56eSb}A8M0iK|kPBWV(B5)r?SXgtQSUuZ|Xb
zLk(jgqFdp}4j^O$5@Xe}ZD+`>0U;|z(yCnAa4|x*r4VHnh1E0;`(j`(UhgzJ^NbY$
z$sTN;K><0*@eb<op1}dl5v%R->_QDx(gU<(FlAqnV&n!v1dQxUsa_&h(w@8x$(s`G
zX*LQsX1`c^@zJKxlJKW0Kf~gw9_hDNUzAytlG=fE<;+oA@foc@gzzA4^Mzl6wiMIQ
zzWMN!TgmCdbX^E6^<zT{Dr5>z1{a|m#hSh|4Duwd48UvkO}MI9!EbjQmoTBWE0sJh
zTG7BYi^5BYsC{?95Pl#DayJv~95Apox>z>C<I;Cs`jhQ3ib>d^_W2;@V>eY8AXK9{
zK7NR2nw)#{86mi6E<b}irW2R7+u&08eBAdvY2{%;e#{7HD~NbKeB8eRS)v5kt-3)T
zXn;ttkgueGk8IZWa@6yaORSghQjaZCy)HC6EBP4H<OE6$m=Gagt{dq^4M&|VlgZo&
zYVS9R9(#Hn7G9;kN2&YaisF-1Xd3?I7c;E&KD+`D2prV)=!?z{3`B4lCigKRhy%K9
zyL$|$<RL~z)TBMn>2?_C*SzuFhgWTqM#i=rVijsR6p$BN7X+;5iao)urQYd6W7d#)
zvT3rPUu3C)J$iI!cHWBgjeUJ4Qzt5MPV*HagNM(u#ZK8Ml~be!C~q2fTq@&MQpi?E
zOfRZ66!e<-_m_J5D?lI#vSuqJAVE9_%uxWO<XOJQve4rL0>^_m(pKfr!3#my01zyk
zCf+-o-CLc}hB{Tv@Uyp)f?x5WomtuQ2@3w|T$8giPfmt|39pro?P|-)6Q#C%|ApHA
zESEOD`6rz1m8fqpitMbEsGd6+8rGwZIhE?Gv|(Jhw3U6YzFpaB{T?`|>t}iA0%^8|
zg%TW2SG(3=?V{nPufbL6h5?MNuGVI=&4d+hK*zaBQO;}7J)0F(WP!=34>w2Fmg{)U
zS5fTm%hXw5#5oZ^8ZRn4wVjD={`k;Wx-?_<hpE`Hh!6*(A=42@1f{3HkQ3Xl%G>kY
zrTQ*H`h5J2(esf+#d`#aNj>iyIe+DlPLSx^YVJ1N68SdmH_!R51XXeJhlF57G_GnB
zMz{kIB}IwPnNzX<Ha0fSvO2xwJXSkM2v~Zbh$-kkV0%?@Ul;CQH5Ldc{w{8CE~ADP
z9{Np@(AK1;;}W;MTuE1~ZFNYzuo3-!cK$W<uWQkU1G(lwE{-lL;BU$kv=Hh6MKPKb
zlJjeYEF}vS1z>BVmiSWu8kZNJV+t(HWEG3+WY|g|LnpMYz?Djl1j4EJm`2&{vT>;J
z0nxk1rXrN~k-B~_^yQ&)u4539?KjT-^&D+sHRLOm^qrm;H3w=)A!}{r0jo7r6`G9<
zaLpRqfN>x+kijceXt|}ir~bI5ndre?H0V6Ga{4E*IOdaOEGAKTl%!-n!DCYPT<v?$
zf8D2lV>^@j9^O2(w5##SfIqJOtIac$xP)EFW)yhFdO(9jH=ciAwhuAd3fw*>hJIi1
zn40xDfPC4ciSi9IT`P43E@J1Hg|C;3sOj!O@iYi_yFf9NY01%W;2&jV+I(ZFevkbK
zAVBwDuM_FH{|;X0&(A;Bkj@HQ7zMjfUS6OWpn@HF3H|c4u7VgH$39sG(MK{2DQCp?
zQYn>LZ$a5u_!=u_+{fj3FD-cauBIuCug!b;%}(Dhx!<=-sC+;1;>y6$U(ARHYJFK9
zN)U*2Zc^M2;?c1qM5UzpT;T(xPE_eTLu(xptwW8H<suHM?^gfxGO&E{@VnEnYSX`m
zLW;|L=#i_~2q&AjM8_p>n!k3>$JtZUSIUQeiLTT9;%$qZ&la~G_LO?)Areo_Qs$gQ
zA<^&&7-Obs(n36GV+-YE;Y|~jsU-`0Vv!k0rhIZE=&egHGsiP+X6P43R>!Mc6f}6)
zaw*C=#kp{F*g(;P=v$C9$H_A?s_4`Ja>IM)*rcycLZ)r!u0QHKwl#5R?CHQAt7a9v
zMEc5kyx6r~GCAfrwxz4&Ut@2R?nTy~TIqiga*aucAc0Owg8>2uV)C%2Pi}D}5lKN4
zs{zddLEqa{-E>*8#V?EwR=U={c|D2ZG_H2Suo{xGZOfV)s+5&$qXKZxq>%otdp925
z6}T(LpSbPh+g?{3ZwTjq#pIHbrN`BCQSpFN;fiIG6Tt6Kz{j9Mh95C~ezdo<P7T*B
z_HPeJ%wz`M`Ryq}R0x@A1Uu-0fY(7#i%Y&DYp2!_JqQk}W;Qt|DT7LK6nU;X9=|ba
zr4k?1I0+=$Y$%q@$MC4alJQyY#G>V@2Lm9PS2%qUh`1=h6=X{1hwbKrG0iI}Le&mh
z>nXRX2J@A56l)9kC_37eVlB_dl#ezey7HEcZEByA`B9%{A5E3HCY?b3n)-Sm4iipP
z4jXGeE|j@NX;Q(eAPQLe(F8{vIY0NGX2_4c`Q|_s)vpEu4J!XUr;JyqEPf$6!<Po%
zSIT8fre&|1dh+3Vt{7Dx8Z0eq>=f>hg+p3Xf?|J`-~Yg}A}gA%Fl|PmQ<;L-Fye|4
zEFbcznfZ2vrv|Qs<h#kHLcjXvEPV2W7qX!A8Rfil#h*E9uHPK-`jUk;#hEA30lXn9
z_ggyJEGO!sIw}$$o_Lrn68%JTT259jp-ATUpsx&mfAZZV<;&xM^oykVE=bX61L3po
zqUE_IeWkNn+y8+q`F_53R1RD=-F$L2uKjYkObeWE36p65g(ozo`_j9UdxHntsI3$h
zMyY+3a~J$vFy!QpdaSvRsRmJ#O)Ea-ArDDiEJp@i8AS)%P@>mX8Vh!x>iQJwzJt?4
zRmykP7S@Vebew#CH!NrHJMN^S<X<B*00{1foN%&eM$UC8Na8=xr+URF-JsCLO1kI}
z8Frka<T?pNfgHaQ`c84K5yMpw#<7dU4m8v9*)4uL*O5tzms|AS9+~YE?gWJopgzY8
zuvE`*?#ReU+a8yN9oB8lr)1|y-JPV3LI8NF6L*p8NDKvZ{wCb&adL!=V)bk&fUFU|
ziTp5rga-}WCt<VLAcIam=<~|!bo<hVKkBA?*kxku=;9sEO`i*YYZUFpcq>m(m%uoh
zWpQpo&MKMcb$YjhMY!&pzZ0B{@KzdZSB<mu<7Fa`Wk|VRC8XNx(=Hf0LpAU=4Hu}q
zcZ?$d?Tk&EgmBO_ZJHmC9GU|Tqv4GU>x~(<Zzi2^37*?_;+xsIy${|QM=F-&WTb65
zpjcd0Io=%w=q0Rb5G}JmG9>=Jx9}U_OUbU(CHjzPn0Wq{eB?(ekCn-yEu!Q}c$|7z
zkiGwFYWM3Kz7;Chm1rKir>}($Dt7<&f81^8{NjN|vsSqwwT7s+|EKV6>6rxTQ0u}1
zvr}|uX-?&8jrzM;-;S#V<P(z}YSou*{*7W--g&)^b{}hW{M?x(5zPn8z@lWxe2yoN
z%XUfkgT%l}QNE`3E8LgSex>o_uZV+fO)Yw%)m@K^QOqTdjPCh0cV+_b!?SvC=v}!m
zz?W{8XB%-cv2)kJ#lOlXhXXWsoD^$bxGg7C82rDV_paQrC_(G9f-2}-@3FUeXaOD3
ztlBpE%ERav%TXJx2e5?ZJB1RmT#uTq{7(^|okNXN6>6Q!Vn=GbA7rd-iT$1MHb&_S
z-_`lYrIV{SqTxen8oKJjw&9g^8fop%3ZvmR?6|hiId@~6>kr!-TV$43B)m#wnHorC
zuI6_`oyx1Uwd$Nj4cEQz3FVI{D<^NCU9!lmch_#Z8cWy-d2&i?)Xp6y>}T^*>&yl+
zjpV7!*B&tfC&z7B^X3aWF_nJMLB#nxJs|&?)xTrSzn%3KA|py4&2B%x7uR8JoWe1|
zrh@>0h!z_pjr+cjFu6Up^j)!=^QcoQ-tUrTa<cB)qsk9SV(13yk@4R$`z?xbA7Ac5
z7HJ+<hhZXWl`AQuSnE{oUc5&GfmaiY1|vaV^b_hQ7B(F~r_=|RW?SE!L30Z_kw{aJ
z(WhP>=N@xp-ou}7qj>gcqM&++bG`na8-M!i=%%*w1|^m)gsT|M`@2tZO>qO;Nf8&5
z(M_+Qqd~^jdrcN<`{v#rRil<f`(FOtks|C#0E=N1QACb;#yA%yU2(}q|3lxT_QwZy
zXmi@|rf%l0-xfBMww6=@z~oF;@<nkaFe)F30b!l&!H9p-&KHEgVBw*{#c~BC-?8yR
zu)B1vQE0jewnOAL%LK4Mz_d2;lH6gO+4PYjw{>P|I(Cd%{xo_mUOL^OhM`!0BQ{%g
zNlc~6!X-(-7c|EA=Kzu3Vl<Tm;~l6Sc!d0t+GUXw*ZRlYvv5B6v^XE>Ab1X?7**8Z
zxznb3+b110x(Is3w?`>>FCYbEKSob{jW;{jyzUa<{Bn{3VPh*lWbyXi?g#qNYhzn%
zV?SY2mX`)YsQ8tFcKTw7Ex~s+ln!Wi-G#$WxEBxBItoDJfX1hSx)ffsdVSTEo2DK&
zw(V0}P<y2I8!%S|jLGbZsK#*eM-gEm3^qix7%&ryL8n9kT>M}R8gv3@d%VYL>Bf6b
zsKZ?&$<?<s=e4(re569^ddqp?x5h->7fFgSZJ-A)iSu3lm9HBV3dVIu!PUimpzCZ;
z^0!o!QI1?NcdU~PBiNi`C*xsKztKkroCw!<C?F3a5a3}ch7`sC_6h4lmJ&?ggvkbk
zgU;0ll2)((B(yy6y87^P-+S)|jq@z;TTYf{K8>w&j)9d^3x#^m1pfmGQWj={N7<AK
z4?BA*W$8+;|23H0bk0YFD6?jE-NF;2ihPb9$+Ar*oI1CXM(LX@=50aw!Q}2-1H?Q*
z(+|5*`R{B9)5;>cZo0l%QK51+Z0_5=m35JqIv5M%a+WV#q)znB70DW)-TE5)KAQ+e
z!Z5ODsq0Mz(b~s8T4qcvdCMvgrgd%w{MPkV>tFBR&<h6=WRL06b;q}G{55`DfH-MB
z%3E#MYR<(=#QR(L(#=l?jQ&zbE4L@7s~bA=aI-IynY6(A<E0i|ij3JilC2U({X>o8
z^?XQITKcOBz4(of=WTHODaX0)SDd~zm8@<95U8*cciP@svdaBOP(c1C(YfLL*l$S<
zufHw)9N<g^cidD_$@CN+=k&($sDh{%v)D8A%h%|x(dh<J)WX^Bs3s&-*ZeX_9_Vc@
z<V^P#f#X+-pcz`STexFi67i*<B2eW9Fxe9ko7%Q1Qk1jy5=15G=xD?P!S1N_wm!hi
z)_OErrFWbenO#y7qIHk-5%D{XqEE4Ywl@3FHHN<$<tV-IzPkSL{52Z%tB<6=6eT<&
zI34=_UAn-6%%SYIOZCvBMmzU#;%sTJZHkMZhOPJvswvb?j-LP8($2r~-`>1`+sq!I
zhs|Q_6@IQ!=Ojw<KgxHh9QuTTle`l)Bb&`(olE>IWlfW2H9PBVHt1xV7M(AdM?Xgn
zWV6RfT_hXFCG5ZLI4c~=)dvXscWVbYXoTv83H%BfidiR1789+Z*K{2bGTEAXB;DB4
z*JGC^!b-9=8<^KTGZ!yusQh(Ndf!8%(qsOisF%R21IDv6=wJ<#m0}X$1>s|A7!pws
zNROqHC7#>>n^nuWSt-c<tfCTg?FXGNg_;C>3&jQ%1{M|inifsxnY`I8O0-T>Z9^V3
zbDe^Y#TNaEL|t*ctrq;Al5ufjW8j)<wIlJLyaLp)V&o~zuA9K%)?(eAR2pVMwO>9|
zrQ!UC?Es@%`J!K!I5fUJBTk2zz#tpfedeS}<!NO0bl^9^KiU_9@`x9bR!Yf=@NH{J
zc!MZmD$~oC91X*!M?y6$uv7CuIfHocw#jy^RC$LF-9e^V$i+Jm>@)DKaD5v$O<D#{
zPQ-LF_%G%~<C<tv*{QBXY{{88+LWkI%DI6SYIoEw$!??9Wug35?$4t;zvrHBKRFTc
z&A9W6S-UB8)B9sdc{<_7*ozrsfv9@HkHwJ6Uw+pv$I*HoHg~r6%V++`Y;23spDCBc
zl&nsie|-#*Y$6Pbp{!__hWdNHq13M?VXcpzGOvUO1fWH=O|z~zGlOLS5{T1HEiTn@
zwq}n3zgU~{_E<e6Wl~!xD^-{hAuDVce02Gz_n-TunK1eZ_!g!~n8=1ezgh#jR;g#N
z2&5Eh42oOrO5OiJrhx_9SiAj0#z-1>xr-O#$}@)1RE*Ca@8!R@f^HL?IyU51@*YSn
zQN|B>2TX2gt(C6bo=j-?mO81gou+I$Dx-N=f2nFIi2@RUa+dA_3o|)W(^t@P1r~^}
zSBnn1O9%>`Pst#y=Kv2$PFP5BFE1P{i<B6oJ5&TF3fJ5R$WDW@FHXk<jDJGHJxw6e
zDmJC#2YVPVJzoPB9?!WVZMYEet~*MS&kiSENqM##k!{TgWqD4^B9~@5{N(T=#}r27
zY%R45g|br}aB7ATK#@@n3i>_c;!_(a4&$r<NlNKKr+c_G-nu-Xusbs~nWyYk*U#0~
zq=aaS(m4g5o*DFl0k}a;N4q;Alv5~=a9wDQMR`D&4geRh(Q$WGT1c)-<Y~NgLCs~T
zs2SMqGz{x%?^eW8t0V9zxSh2eFCDr>NIO^0%uq_DHWVgD`i7&1wdWO?1`jL7?256a
zR!?RbpDC0_CxlguLEdHU@w;)%3jp;|7%l=r=1f6>*9LH#k^gtfgWQ_wodWk3U4{sI
z<>5*V`0Jn<XYcoXlTS*X=Cm#4_W5iFHsY!#@#sSxLzC}0KN!U~>bCu_=_sE#l{8lv
zGe00~_QRcU57V~u(sx2Q_uJ`rHnX#W0ZN9><d<e7!5r^ihiU2^gAJ<YQdif7Cl{!U
z?`<vqX{Wx9)B!SOGa*6vMhfm3o(Q)@xr9xLguk#Wagi&rmhO?e@d|5J*r|{MsNn`4
zAz+(0+*k<KnWe!Gp#0~G=QBFs@2u4?h8Q|v<|Jym!|j*)Rw2yTo~C<t*$?g3!+tb&
zQ1+;!!Vlyt4;g#jC@%`g6P^V0I?(ye1O|5^^Ha2ewSr=i=-9InL;x>OG<(^pjwhnV
zO=%nB|G}M?tl$9B4gW;zj*q|?%3-HSKDHK<!;?WB{o_F%YWv8-LjPa0K|i*4&fW6$
zwV2=CmHDSpqbw!SVrE>Gi!9Yxr?3h7l$^Y|k`M4S&9Gp#4<?P^z2y|XbKL`gspOL<
z9`pr2;mFqpLbauED*2KAd@%=1;K0F$!}LQr(qmNyZ0r9(2;KwZ2-m*o=T<$?)Kto+
zY_?(U(4_*OhcLPFHYdIsohT^A`E_+h&F7%&GB4QZ$^uph`$PXl$3&C$x)V1$&4g6x
zEvMckG>EW6;^O=Ef!51Ub!ir7gyKu}0GbQ{je`^afYzQ8+i_PwB;=0@#?Awru5MtI
zEc=3=vp0x<N@^jAG+zlNq+EfpCMUm=mVuE+Z-_s!M3}aTDEGg3q4CC}J`yeP!_mAT
zz%TP*`DVR8#a4|QCgB%$3$gG*7;#{{@xsycx~cB<=T>+_`$uYjM2O$7pEh%wqC}5(
ze@Q;`YwjJIz%UJ9o5W6RZZpfakEu@lwM?H}`m@I*PUgsDVE7Le5L$>N#+=`rh$;mS
z``#v;N!&uk`~(8H+eDKePAT)RLYKg7ebM6FS7^ZLl~0=CVh$&mXYy7FWgqYTu-I6L
z0Yh7!(fMeuYvH$fhL<cG|K8)=<{2p+XIr|viGa9-Kt_AMd=n))5e{32^&P)3vYS29
zgD+1Ww5wv-H|+M8DPw!1#-HpyIH@irMnPUPm9LgI+VmYjr``a$ylZ(RyXtolU_;+K
zUjZfLs}I)vW(Uk)p89-!fWDRs@`_WuF8~GWoZ2%Og0J_s#`(nx1l|1@FzIU@@#5n5
zu;*I5EizSk^wyR3cK%Z;mo$+%gJO4>r!dN))ou}64H43k3yiC8@2_;zE#}X<3CXgI
zb3G-DPc?zkIN?>dR^5~vg8Rzrj8V3fNYRxx&?SWH9;Mp_R(qEU&jPgZj0F`37-psO
zA{aZ*_)yBsN%tuI`I&HoBQq|~MTaDLchp{!f#yWR!FOZ-H1n*@OOYxZA_6lkK&Teb
zY%%v#Gma)+bCYoE((`<?`Pu2pQS0)xIqI$MFj-pq%&~pt-=j?m&GHG{k$e8`j$iB-
z7%E-Hl0{HZSsfG<QY;Ak`B~F}*f^l*q?@;aWS}AW!}T=R3zR|`*G?XVEl9lIX_efN
zEbxN2@S`=I$lm5|sJoW=ct`w7O=U17sC0IDr4iI_sa?-((e1Xk<}M_Cm)^v3%wYR|
z;f|TY5AGN)zcWX@j28H8*@F4GtD+w*hWpUGr&=pyXQ}MmuA+Aoy`-g(Nm}#@C?5C3
zUc5Q<@L2K72L+<|O`}%?7J_DYVaG%IZ|4A5SIByiF(w3s!{q}ciz&#dbsE+<*AWk9
zf*Rk6=amWjK`Z(nn%t9exGKGWBqN`J;;U)3oBewBAaKpkr1^$xcHb*xnZ+;P`U_Ks
zTh<C%;KA=qE;6(-j<Lq!GBj;!E4tCFl_a9;-p;Y?5F_>0aXtRgzonbf)GZv1tR*YZ
zjHr)2c{T2}@pp-I$Ui)G(0;`U4|^mzd}#dE5!SpRlK9P>AG(UXwl+Tk@n21~&6$Fb
z4;dO{DDme^MLiz@bd_-K2uTKwZ@Ir>>Z?}Wx3+B7`v)tIO<5ryufzZvNzqC8i)#3S
zeeJCnwmHuz;@B2O)TvzyCG*}8zhDAb2FDgil<e8yIW>)OrOVS|SG4kDSXV`IG`y?a
zjeKNn6j!oqup}<wwK!(z=)gtY+`9e~zD?f2?TBMnMP5Xhz-BSYyGKCL8MrO*!>K2r
z_l!Yv#(LJKPTsCVKzqPHu5-LeUKoSHnP>HB=4}=}HKKfr)gnRG`Pi0PKw@MCBqT>8
zzYwM?U?L`f7n*&$8gH|HTT-jx7TiE6A1{L7zY?(>(2bGjR44<c;AY$u$Gonn|BQiU
zl1~NP*=C0&t@^v`;mKvB^dm~oWdlcHv+IgI>uwQI1cci7n6Ox=|F5T+@1Nv+AZ~rK
zD#-z`S2>T%#}bdzL>9z~L~r_yzWZ+6V_^^}o?QQ0D7J3H>z|wdfpFUu?lbnWn=3D!
zc>J4WlvZgMq%WC^{e{*N&)jy~qj0NhLI*vV+@x!C2ybZ_|0kJP?E(?ueg!cc!$tQB
zH<y{5z#!~eeSX=oRciOUQ5g1qY7tv@$9+(C%kguP*M+>?CsA{yS3r#s488Zqd+gm4
zUl9tb^IR+b1STWW<*Ki)_^=j>_WT2m`tP~l9ilq*eVa@?!h<cF2hTNokgn+ej}G(r
z8e4Ndg*<7Z5OMHHOI-Q~bM#yUcmPhi_aEq7N|xm_=C5mXJ}3DzAW0ADdl%|lOy!GL
zYBBc}M4nXeG($A_dIFBus=wt9PNHpeRewkVEcS?F1rH3G_rt#MBEni*)%+6ju(VsQ
z8G~#)q@C=2?SHF}c$~c0_PucV-g}jTbLoA&H7d8|yIezk^M!RCZJO_A(9prk43@rR
z4%V8hv<)IV#kt&UU#Qh&;Cx@45-UtA9m%#0xg%XQ0bwCwr|0uUio**}A4Mm5Le!!v
z1#d*Au<Zb#n}g8>XJo|`Ue?h03A*R1m^`ijNBAG1;23lQo%@8r_og}bMJXxqf0~_J
zH=h;SZuGZj$)cW+wdY*6@9M@orK-k3*FL99$762^-+1*x(YXbPH~YW#$PJls8tQ2P
zj{$Z}G1h`SdRidpXHPB)4updX+qpqBn8m1$`P?)a`2_&Ki=ahJbae||i3Jc^F=vgd
z-n?)VxzfY)N`lLsH`a~FI$LpWgt%lmv6CesySFrlZ~H6uy59{(of{*p<o{809{yB+
ze;mKa%yzHrtZXta*?Vv9<sxKruaS#nq^yi<&$wh>+qJh8*>bZ-LPGXPR$AZR=l4H+
z9_O6T`~7-7pW9yh)3?L#nJc+1AoE%c1g}FNJL3mIyzHXdN#;mKXNhg{w69aBMXv1J
za@XY>q*Z_c;W0wH%-!r^^Mo`SCiJIHsIF!Ef*;(zH*Cw%XfBYvXc?BB{?*8YG65+m
z+`-So4X5Kiwl#bJHI6o*rECa)xJc7x3@*HYV4?%WCnk2;{J1uQb2SYF`{Xmx4#<aE
zDZMEwDg=Q*V~xVULfgP}zlubiMkS=YkDH;puE*ZFg0-gj{fEj%{wGq@LN&bzqP-O8
zipw3Y?2~MeOs%jdH_?*pM^w+Cz8$-Z6}j~#9;ua|Z(S^AY|pF!^}RTBsG{hWAFa?e
zRfN$`OGnhI%k-C!x~}A_({WJ$n&kqP7{AOT0`*Bh3ZHEh7R^sX96I$QWb|OEZR1Qe
z0BZ;&Pb>5b)^xiFZE6uSeh5!bjDKuLXU`T)tl+j>{@AW)IX6jsvCtfMAxKi8*w*}d
zBx4l9)CjzzMn3x|tLz^svCXox8pm0YqLOIp1FfWm-^P#;ffv7>+2|i6)T;WS)Frnd
zNe`lsA(i?#r|WvtqS%ZdB9r91zcA?4u7VZh-vx&$$P{ABqt01~ti-hysG1swK+mjW
zg=BdbO8CivCQF%sk3aZeffZ=DpfB(Y1$VLbXK<jRWTC-Zvzg5%#-p*5eQwDU8)Z;z
z7-vU)X5qkLXK+y?p;Vrc_h*P?9oObtKe&XQH9S+|zJz8PR}a)=e$|6~`$&Co-){A#
zkKY(xeIv4eccq)Zehm8-hY0p}9GaPojFu`y@I9?u^Fr|2P<lW69(D{=53RGSt{g{9
z|0*(lK+@Lud2}jy-VKfLCY>J{wx`$Fe>iXvL47cKEKx2~IP1CICdp<7W&<IW!)5>?
zUf}ZV)}i|CP7Lv{Y2uCG+H1;gQ;{+O9TI<6mynt`MeE%8C`F5yd8%N6&8Pd9HPL={
zNUCj9fX6<0u7feo<cX#;Ebt$bEzl|Pm;z>Cf&Ju|0_ux?;g*27H|MS&|5!>XMxEMU
zk;5Ew?niMnqC-~XcaqT`MKvg-G`NcY)slfYwyAfD_*~2;UJxw;t%>^4laU3U=C04h
z_qwji_7J?@52OAAi4}aoJpNQka*zE-c|K?Xqw8DBu4dTQW3s+B!Z&tmFwV_`UGy^L
zPl!51K-Z)ia39bt2VOhYti4>XXnLYSyDM!QyrKc{DBI6n(yi{PJN(6V8mvA=^SRNk
zu_i^?XAJ*Np`UN@4P=bfGJU2W1yb;nH+&)s;`MQ?gRVuF)YE$<e#vxX?fB2%Y7(`-
z))?S#NzTGR02382ML7)aH>8nd$IzJ5@Zv0TN)tlBNAOh_2E#XO#$I`rS>{r)XU0%a
zazHzY+#D8U^EQ}Zvv1pBM~t>zYW=QWT`=9nt7H`TuSH|2q;BwYg-4Qr&Xz#ApD8J~
zAVy4FUzbe8d`u#B`QaB#G?=QaXa9cCNO*Er55u3tbA@9o*MwYY{eay*PfOkGt@#2l
z+t@r3S3*`CW=M9|&hGy`LSStsAq<%7ftsC#LAHQ2$}Noq8z=2mzP@2y%rNR;V{^)v
zMSrD5-bYrr_DO0v@z*?}%S$Txs($l0(8zH(+)MNQ5U^t`+qt_&_gqx>3y<j5y}X3u
z@e9Z+fm306S~q}taN^DI%e`(T(HcSf`@6ovzFkIRSdY{WR-n9vrz=amEx0l@$LjU=
zd#V}!WONoxsIHXLPe)f={E3_k&t`gnH>HY?S8i-{E}peXmR(iRN@!I=`xDAz6+Wd|
zhA7VIKehd_7ksS-{hZbwU}#^zl)HPk9+^~mKnK>7%IFT<x83Dlg5M*6KN^TC5$%Gl
zG6fAyUr>Qm6slrAFGg~Nof%I-BUSrEJ#~@<(WkZx!HIZC=xZu!7!U{JLwOw<Yc1xZ
z(VTB|4QxQd6;GY~Rtn!%)Ea~g!*i0FE^CS6Azvtf;d_L^`NU=xK(EY!W)vH?N+lzK
zrlsO5B}5gp|FMEHHXwRF63tg-hXkEtK{6k8gyav`(<zG#hV}iMm=58Ql?oM8;BhdO
zKdZ8NI->}OXT>e6Wdze6N_+chWd`3lMsxYp*d0lGI?~4N&@2%lJQ(2FEct83jurE~
zot-*$emb98kFklMZ;IOVl;fkuREca~;Mt}p?pwBD^u8Jg-^7XLdb+XfP2uGueht5-
z7jFcs0!{?Gp@Th!yIJd%310`Gi{G?A&lS_^xz4sF)_XtmsB<C+M0oS0lD#hcKQSd{
z-;^BFOW`leTZBp<&#i#WFe%(NvMZMp;+U%i9%`h*b<(acg?6HCJi&G9i9$8>Bj8y8
zcSK8P7=5uhYgS=^UfpJG`55j|%<h|h*KXnHNu_ZnCyr!#ZYZVD+D~Y~Ad7zzBoc~u
zP_r1Iys!0TwhIkzXh?agJ1eOivskp3Ib>-?UDe#E<|+`dl3E~_5_nZ->N~z22J<`S
zUD&w8{F5na?B8|gVozvC-Z6olW4n5$8NtH6k)~zAtWsCX;cH_tiyg(drH~hTLc4b&
zX)n5gaUn^@Um1NfSL0ja#vUxgmQ12*AGn==E<^?paf)imjkBU>vuPEr_;to?0x|h4
zM5%FDR`3v0;k-IyC}3|?+#oXbAu1AKgKE){WhwBp<wjm|z*zjf2@jOdhbOcr^wdYX
zTS7+-FZTZk?S!XZ{IaAAYo?;l8Vy^X3u3QZv@eD8b=4oh%@MAO;g2aFYB1Z!y+B)b
zyaMx`RR0mpw+dMGwMh{3@~5}F$=Ka-6fF~f2$PG4bUI(FSKmv$rzS*J*zH)&7o?rY
z_V$czI~=y6d3=6RBgT(e(f__BP{5&qKv7eLQ9i1!PsuQayWrT2-Rj8DnW*>b>Kjk0
zyTvo{H76p@sBkJ=VwXxYm4PFvd3Wb6j_HXcsS_8_f_j(oqQ~CS+|Gor+TgISh5)JY
zQ=z6t0KaBH-_BkWnP2`yip+~FlKB<5UrJv`lAanihhd=we^){20&hN^(v8jN1yUMu
zV{v1x+ncon0UM!Hty_*!a=Ic~wvS*&5WjoxH>|u$*Z;Y_R-&^T?rnwFC*@sMjr10*
zL<*$8w5P(zuFpp?pS@(M9{=!HSW^2=G6W$=BAYM1rEjP)*Nf10DDnIddi$1oK&WcF
z{7$pq(VpR$J<NogZGKLj8i=fWm1yXYFc%7g+S>Z<k6Ej&52S~mQe%)0nhJ~poHSU$
zB_5@@iS!`h-!HXT!zNT8a9u*<1GkzQ4vcd12QALwcS<U;gxrKFfx1=VVT-UVTjoSV
z(geK|H8w%kAiKjlF002{EK0rUj^`|9Fk8F)fN0StlX*92`#?ZV>|;My?se>{tKj3F
z*5~L!$tpV+XZ<F*wlBjXfY5ba0~#l8%B`Pa5&~|^Uq?7Rd*P3Tq(V#Pi(21%!Tb#m
z6squng2AYT^k2G99)<14awNMQ$%y?WBb69fZ!55vVd8w^<S!ziuy21f!@x4&(@9+3
z)f-1KE+*e#P}k~ql%}+J-|ifqsVPYg3b*2o5bdCv(cjAI^+6s8P2nwg3<w~S20iGX
z-Ty$1>Qr=HmDi{0RBzJXTs9OKB>8AIHT<0`r|mn$#;jx)B6yZ{O^EXnBu;r-liT8p
z1{b@%5BL4PiSzIP+sOS=y0$<U!7B?`QJW_VWIljMMWGxG=nJYYO3|~C9E$&e6g2~6
z)G1U%%q68wg(LHKGSexS@T|w{k)XiY*Fm2xV2d>rwen=Z2$yGYy(&e1ybi<Un)hqo
zXLYAK1HAnkUh>snNS#j&uYI}_H$X}Mg&R^yH=$Cye`KO7A3l3gCryt<o|;hR%pZ5D
z8Rq?yMa!he{+?%~!?sC3Ob4Kp0nzp3hxV5USZ&fC6-xVe|B{RnPqz;jIj5I%;cmOj
zlUpjI6-HxfhE$1FE*tul9U27BN1NTX)MtyBF>>J=DL~DaVMd$<TzS!33*9y8v+6c>
z+16*gnsYCGl9$V}t@z}Gpc|-}Eb7gou1X`?wy(p~85j9AM7TMYD<GJwvswC1zBT=q
zIB+0t-S!Z8JW_EK*RE*6Tl~fML>HxzLC})bT4-J`!uKEBj<Z$zAj2XpdGouDB1F+F
z^is2Lu+!ht4H%{QXR_0YaZWMzufMfnRlbgxw8f{P;J+E}c*t1IEna;&X5L`UcKZ+1
zp*GU}YNCT;M6@isq2Aq5?kSrYz@<>Yr~2Qbk^v$%2hvE&!(mC9KbChc%LdRDttMmg
zKn~a&gy0nBe%`!W;g#U_&l4D}-{-Jm9Jw<EjWZ?=G^Y5J#1SkD)gn$JcI(pz&G1Ya
z#-%Xv`Hh0t7Q|_HYhe#7LOnJ5LptO6Q&a3a4KHf`2z?8Wi#V`J?*#;D`*xCwr*t9D
zl*Bqo0=a^JA$*#|>-i{%_8qOlQem>mJ`I*#-Jh25=GYAbE7SV{>>`H!blGs_qOzVD
z(4BU`yRZRc<&Mlq=9m^|DY~HAPD1IS@|{3I!_*RFTw(C$PclKQB@(u2X8r*mW~+?c
z;a`e=mJKN|XSnEs&WE+230<F~=Ah#uguow)jbVh0q|2Ht*=QLFFth4{LHW3ReS=>j
z-e#10H)a4|jbqoiz;%k2y7OPw<eWd`)$=Q>)|@rxCP*AO0s2wshy+q;7Bmd(z%VZ~
z^j)afDzf94bWcQdb*C&^7kF)NwYv5(8||8OT|k(Ebp}wVE#lJr_sLWPy?pS7iBjs6
z?zek&RBznod6(}`23or~ltoEparujw7l@Ca=s*SJ@2(}!8~tknR9R4W)NE+yZMNbZ
zay7eXXg>>{YY-z|-4FgQa+?}$f&w{LkC)s!1TZWA4@I$jHPR;qAAo$`=swTE@QF)Z
zqgGzQL|)T$A$*)*iS?{K)bqIqZrS(!E=8^7jv9O#Pc{j<Z{N0mQQ^%P;C$$sSShvH
zlm*qI?8l|58pk)H>Xk!l@AFvYl!;RJL?HQhHs~}QAlEWg&33<Esf+UdaOK$l#<b)2
z`GP#{&>o^H@I>pYocJqHxTS-fzJ)`&ecQtmr^{+tGUmdv?yv%;M&fJ$!1ohR>oLMS
zH5&<7O9UsqS(g2)XoX=(SuK>kJgmJ`R;I@e$F=C`olcp%7vR{iC6y0Ni*6dZwJr8n
ziGin@F}d})BkK$xeIqBr=TaXiVAoy6UliA@N7ty$w<u_?nEt}><EVT?o;NtaY}5I3
zJI`euzx-a9Ib8)viDeT2A&^A~>wrvL?-Yk~xOO-6hz56<<L4H;>)ys&bL8DMWLZ*}
z*GA!w{{*8mP3<qhTXzKAv~gt(A6<41j1s9c<qY;zL}=UhbZl+80)2a%RO9i`cJtW7
z^=Kf=?HB?*9mRsQuF(L)c$;Gf=a*MPW*Kx{a%-$I3y4SZ$G@sR`7He-%nF~#TV9I0
zmC&1KRo*}BO<Kz6Mm_&N!UUwC0KIw~21+aje0l;E&1Hl#x0AZ93SSd1_XpMGSUNMi
z=}C=bb1w#mpRdvTFrf3_-1oB@$0A!M<%+hFfGJ1WE+7)y?;s(gRS<s9qMHV}F#6Sw
zs;C%4GjiD&#q{O0Nt6?Mm{hynXWLb!E|{E-o^Q6YWkv?BuL2lvdAa(+;0J)4yqhtY
zox8tw#8?)|FVvf<`)8pGa0bq|**QkzAaC^3&kRM~+=eS4UUBl4n|M~qVWCPR&h<jY
zS+`UIPYwNhw25jv^$Z0KV*VjB(~BO|?bKbqlqSh5@Rj^aGk+bQpwfrjr8TV|RdkG?
z31l00O-0l{)vrK}{gp5AuLXd9&zq{hv)-?mZjb!M=pH;!{+Ipvv)8GwBP&^eylUK=
z`J>78)Ll^|f5(Hzo(lc20a=876E)k(nps`)^Y*zr$Vc6<G^kkM&~7s@?{E+V#jhuy
zt4BUio+<u;j6|oQG$U>*iCuyIxNGtpKP8JmUGpb_qfI>HxXl@`Qy@W)*J`?u@1%Ev
z$C*_hVtSHZlHo-$TUvl8$dKjy#9N6?06J%|duTa*5>^dPmT%-gkT+wIkXbrK@Q!Bf
z13$bkfyH?y^h-)v51vHZiDAM6u}$)aZ+;f9Y7@PB;^Vv3W{wG1v+Xl6LyN@Tq<T*v
zTm5{+0(+86m{Qwwb}~l1Td-}$`C`7k>s8w0J003BA^2;0;>)@WyGZp~o>s%p#z>0y
zn3fe)I^zYK6#a&vDG`p;_Cl9Ir&%xDC*cG!?u>;lypeVyk2Sa3tfo(l`5TS*3`IFb
z=l`~8v*j!i+`kP@P^?wr{mN9i6BazQ5Bu~)NwuZ~Dq6KMjKpDZO=dx^_5zP_X5RL3
zo>zO`*6e~_F*O_Wfw+?i3!m7`lo(4~(d7)Y$GFvj`hr{Bf&Cnaz*<QhV@3c8K@YU3
zXJ?5%m<j+?bd2V|OEDZ2>#9mh;U<0$k?Ai(kT#TAb?=DD1c)_NgWn!Dgk{*wd!%ds
zkhiMMlU7q9w|j3^+J>-nxBsq4b<anFl)`055q>aFiK%5X#zqVbg<Rtqul2)|b0T8{
zF_XRXkbveN3v#{UU_R1<zyO1fsCjQKaVE?Y{r6ZvWefL62UV1SX3dGP_OmP{b1y*(
z@T_5@ZyOf=eEDeB`e6!%)1Hos`H4m?JE?54bj+^D9uiNZcsIM76Sc30)yK@X-42wq
z;ocL1ELa=_Xn65JM`Wi2P*I5Rz-f5ZO0Q&7TWD_{B4;gO>eu#NnDL?i?icXEAFa6I
z@OPJk%ht+Sp?CH`*606BU#lEOSB-qEekQQ(Bo-<Qzbmb|UfGNK-L{z;;w>;v2mP)v
z95C#F?JO14AV?Lc86MW&xK6l|K1Gwu@+Kp=b_RPanXnf^TA9j|J2euXZYqfGD(k>v
znQjAHiiadcq?+ef8d--3(_*<s7k5#Hg#3?g&|^#=sYuJ)B<Ep<cI{Ef?o?o<Ao->{
zpKXBNy1BQ|lDk?IN8r!ASw%}OQB4xnwxoS^I+_vWD99z9M(_u?TxE~MNW)=iqCaBg
z$YPMDURo5}v1`H^r%d|X>xL`GA@Ovvdqok}onL)m;8OwH$+|&2fKDmB3NTP@U(#eo
zD2RSM%}#sQN6y+1Q>cjq31XdY%{~QK`Ic$Vy;#;MD=gHRL|(F1kB@Ib`>hTMJO@Z3
zR(WVPMYEr(r3pMI#&D#uq$g!h7{R#@=(M_6N9_>G)*oCSbUUO-V&g2t^uFyvKd~1H
zIq919QAq=7T(5`KRy*}n`<NC5+~plS*MeK*K9<hdoD}&pN2T`h*Sm(yo6gCB#?PO(
zmr12h!uZW6bYKXLJ09llJPBno@v^%HHkZP^pK@Pev);A4d~B0^)&eitOy4w2CZT?c
zPf_*L%rZrt$wJM*B=&p8VGA!CYIOryC4$dvVFn$KDy&QrW5m>NhP{TTtwjUjwp!Ty
zFod>2`k5)DR@!AV7Z3ET%kZpHVY5-jaq>^9sEpH9=K&MSps{RWk|x^xQSv84wNJ8Q
zvr2_Yf0~*OO`nUu6ItHV!tKyU&?1lY2jQ*SByoI;h3sh+MmnVjs}ab9A$mlzbpS$G
zrgZj_K`{X*pZ>hPA*I;x>_FqaH6_b)fwoEHQ?@^Q&wZZOF2WM|$WO^xL_+KAhN_R8
zyo2w$6%b;dfBO}F*Kk4^0q0!!$kzNVOUveb-=t24N(Sq4N07tPZ8yu9oM?pfZi<u?
zE_Hz^QiBD!gsPuwQdTnWaxLg<l4uplpBiTl7rF!F68rQUuje_Rse;S40}QK9v-Np?
z<1!Xs7s=QME8P<8(;eEWC`+>?PRIP9ZH+(6Oi4TJLa?oNYRsjfPRqX+&aQs`zEk9P
zqT&2Q_|M@q{!d-dk4<-p5D!QkPiGkgwj;OttmC0Lyq)C@PWeJs8kY^u5J3bhJlvQ5
zvr`>y<k6NoEWOrBz1hY);1mG>J9-&F4N7d&QSc^yD_+({-=OM=DiYqR+~eCIIu+D-
z_v?sfxM!n+oGORR$SCI+m;hY&<|judS9V0x%aN@pSip^8i&=M^iJth#bBo!EmU;a_
zX0?xTPULo(^;FBO`&Mh$4+JW5(_>hxj&J%e6{<75rgQbJzj%#oge|zu7s(1yatGCy
zHMB#sI)((cCY93c!i)${H1b;9v(|C8q^LxGI;{DECI?%#EHM~(a8d^!!D1{~HUNdR
z7H1*M$dYAtfvL|(m|6y;xPTPAkd?3&*2eSicGp^dG!2ziUgE4b!1en*hIg)+nmn;x
zX;0Cs!7;~57y&pVA&}4?Z_!+}U~&khf*&e9v9uo*;)IV`kLBSy5Hj8I2SGH?AI75Z
zPDa$ZH_OdWyt2S<q}#l_z$N=Bn-dr%bi0OalYu+IeK`wyph@KoBfcYh7sy_{yhluo
z?8XscE!4*ujDjGF!+#G<qNE!XVa-@A-tI?hY&*7}r*z2ctmivCWEDC2xo>V?R6*+5
z=hqmdxKP>(p|s!^cArzUCDWW8Y`hKw3z)<i`5XgAYpUk5dU?djq;OUe%aD}r3Rj~~
z_Dx`s4VcnW9jUt;8pxAL2BOn+D)lDoU#!T+UdyKb|GeGZw`tOBtfl!RczRz&#T(;&
zox*AU{A*{xXYuc9NtigEmxL!S5A@&X4vyMokSWN5tvlV}%a(!T`xeZr{(U0ehDZL^
z+7rDx5@;*kVj1u^H_2wf+v7DB!I#^M3$<U?bp2Cumj6A^9!aO`9j`Dfeo$g+&DxGr
za2?jKu9b(h8~n}7fGk8+uQa)r30CWmHz*doW?0@*e#Et=`{2^pn$Fn){`OXY)z7@S
z5wF9=b{Te~Xz~h>JX+t>1Vj`i&xZ|Bo+WuYNTs~bjt~0hs0;hiU|7tgSo4RQC6x{m
zgFpB_n&<o7?PRcBNH2f6#wf>r$04c1CKFf=Mu@g^7$a4)W9Nv#o{e@{EL?5<&dO)%
zn~~IR=Znz*MTe>J(XJ);MP+1Ae7lsc8N{0Yxa-h3fE<jyr5$?UrF^D;JJL6D@u~ez
zgLcVB5&fk=A_MBqYaJaPe9?&4%M|GrcP6s;E@iG-EbmeJy7rb2W1xIrQO<La#Z+q^
zd2snk-b=2F*l-we-_*L+DNOiA9h$pwMPlq%bm=vrrDuv{vC1#Q9k%V1COiFJqOn5C
z%nkP90Q^Q)Y9T^KDgC`djUFFXE8)BRW}}_a7+<?r4#A{)Svr|xz96(AD0UniHvho^
zI+LSukHea_G&_%Bbq3+WYvL)Ok58B)Zcc17k}9@)O%tST$qznz$R{8UvMOU~`wCB9
z?jnfG$N{Nq0V8^hJ&`_{tTa##q9xS{c<}IDm#kQO_sH7+S4o3AF=>WtuZ8??@ZS|@
zcY_ozYTQe0{vAJ@p53LU9AU$?$v=DZlvDU%)pN0;EJMuLB1U+CBgR3sI9S4&`uX>-
zFa2G<KAE%>ThkS658|tCFeM~266h=olMzrkXXb}Nn3J_FKiB8^9BKc$8VrAtcXWr~
z>>+=>!^}!kc@+h$`3kRaAH%S#TU3f1!O(tt+L1I(WMnsS+DNo67CGR!ps#}-4AkE(
z0*+kT(0$#fY-J`!q*Sq3%nsnIf$SL>E9w=~plPo7Eq-jK$6&UzeV*zf*z_Q*hqmL)
z`3LVc^xYtvud6t>=xKV3E{BCa5JeH5TKqIc`UQRQ`ZqDh7blZg!RopvYNX%IV{z?Q
z|E+!4*R*KjCb@2ktZIP+s&HBJc(qyBJ22(l<8^ve?Z%V3hs<WTId<B((}B~j%r@)+
zQKwbci8{wsrN=ZdpK#jt$d^5rnkeLO$x5G>unbMyZmxbd*b9UfP^Hd|Jc6n7BLPY?
z0htWXpJ(R*G_rwWr;*U=B#v{eF)r~zbUO)M0W@!VZTV2Ipp$hU=AE5@>-rd4yB<n2
zURg$y2{fxmP)~9jgmO(==96SmiT7-qMBK9(7ORbe|5)B+Q-pO@#g~x#l|LZ?>r0t2
z5?M~qalT;~hGH}0a%<;Q1T2Ny;up)1WTMO{>r6k@cH7S?fa&a^q66cX@itn$$v+Fb
zsGY1{Im|S@lV^98)aOfR`j8+dMUvKLPLu5#)%^)B@HVhJ@jYfm(#YD#Lw0!Rl{BNl
zvLY2+Gu%o#LmADr1qq{Q>Jt^#pW$_)Gki^HRnv8(EEhHxMSu=Xw~bcvl2-y0R(&s`
z`uY44a$cDWh&vAnUSK3cl4VaAfXH_b%Lfsp80`N8)yPhid;F66x8QB$;Ry?Rf__nF
zS5<q!u>U>r?yb9DrvzG(Ln~Z$Z>&Z`{Dd2lPza;;icQ&n2aDT;O!}aPU*5UYTLn9L
zR4l-6zc@pxsYsAR5+W|;i)5YQ#NZj$S$P8yreMqXVs!AX^Iw!`V5V6e)2~IQpBZhz
zJbnLx?7y5Wdc30h<u=>2?L@WRdHU8<^F{Cd^$thqALN8&dk5d9OvU#E7Pb(jw>gDD
zB3k@ef~fiC%9-3lm>-hPO;p|@SH3z18h^XLA<^v9XpWx;;1c;j5k)HO^v(_gU)@&t
zkV7p-7*>G&UR{D$mUX+hl}Fg#gCvY#T`gU;#VDcU6dF;BT|0iMu;CdG7Jiky<R!6?
z2CNzp;o&A*;1am-9J@S5ITdJhd8{4-o!YzvQ(Jx-W#KJk|8}U(c|VSC(sZjvcKXdY
z@tvIL>H`xre9`Bi3TNtRb4FAgB=<t(;{GatVdpN)G?1M$<*0s-yUANy&yL;xy;$k;
zWQ)U+4&}nHks2Y)+PacAlInE`FbCw5sU9yz62VG%fzMEC#0D0<g}2n~s!GkdUAS$T
zK69sgr?5oEwcSul%wCRr6mM}lTN|Fm8C}(BUx?*xCvTQ15+RdU!1HVJxE7;yRNpM>
zr12kvV08A=#I6GMpBqA}IEStzN{-FQdS)gE8yB-(!9({s_vr-jAVkkUu0E98@3N}u
z_4g0JvjJ}w@RPVUHX318JavM*!~2@ruPFhW%$-W#QxAs5{G|YQRagE#5AOWbl4Ld(
zu9s#uQLh<%acVcz;k|6~hkr2vQdoXUwY7jI7%EyfSxXg=b687MmK39551rK;-8STu
zFrTAZ(<+=vyk7CiI@3JSw!HWM^H!Kjc)3p3$Bq>0F=_$4Zn=a%;f#pAj`g+J3NVSx
zaZiDm>$R~;Ee*r)m*@v{VQ{rB{xIuZO=gn3zWwQdmoKPty<VESx~U+Uc9O_zXWe;j
zCdfH!2D<zAP`di~41`aE-HOx3pQ=-;k$Q!)8L$0J^BF&mXc>F_I?qCSvU40w=(KvL
zn4I~JRQ&OY4s6j^^|O;iDQhOr)NiBzK#R3Wt;QmHSaFW#kN!xSC*@Pb>c<9OphL>k
z*r%p#6iC&JS!AXlA4_GYp%=xYII_Iea}18>ZB>Oxkv|}C>k(iWjznIK8LaCG^g7)f
z^>V=eu6<iAm}jG5itFT5tR1JRE_9WR-71}<L_IwWp*+-E8Sp_QPCPl!TNl2m$ElWe
zi%T%bZa)5<@PNy^GB@3=Z*+aqvnNw`w@BrsuUD$$kX#rCth#T8eKE_|8FAt!en&7;
zl$j6v=J9&7YC9}A6wjd3MNwCcj;4{4c9tMjCKP4g&Q|`h8e9XGNflw2kYYQ_IZ&t`
zKQ@rYve`VX9=Fdlk!+%b)8%ICg8)9R5(Crf+*7AnA(Q1(gDy)}a9T$x5J#M4;trKm
z3AEaNl?);UP!PKJMj?K}EhP&!1zGp{PlpX#W7ccmck>cXEpa~wYSnFxl2eiYyh^kN
zf6pfgC{9a#J)Z{X&ktN)z38KM&9#406&iUB<ebJ<?C@eS;Cu^rR<~hz?lW2wcURqs
zOfwjAYzcv6VIAKoY(wxUwRV^@N|I}sa9}qT#}#Rjsk9m@d-R{|d7A}6a7pcJy+Aj;
z<*<d(u6)kTm#T1A@;=wqIwQf{cmB`YTf(RY%i2d27jPX{_22kx%7P-Ta*}QnJU%&h
za#jN*-EHV}-Y;*fboGnLExi7a_`b=SelSxeDx_rmgE?u1IXQ6=ss8_!5qZG-J-9^s
zjwiirz-=MSnVS=!oNT`rUM|wG=+?Q;+a2<!*PVgRv9h&oPhg>)qC8{mTF$0<iu14m
zY9-$}|K5t@n_qm@0NVg@88s;jc)bd05HP^&T3^5k=)O3x+r_RSnL`DkDkl0N3das@
z<saBN&Y!wmRJZ2JzA?_5`PcMU<o2T;QPw~DD{8m7`_%d#U!y9sx^q6InjRsfAH9?N
z5A@1yNqIeLudStHUe@ZZfJit`Gj*hiSj&d5VbUjvck|WR)-F108HURf><zQhTC<-t
zTMuSQnC$@8KbKZHR56&m;g1>uWwGKW+e<#!DlWM5-O#nn+9*>9A-q-hWywMr8I!RB
z%Gk^MD*}+71v1&^&}_Q?EW0mb;4z^3knCUaVpsgAis&<?Ytyj*K=fb2G{bf){9L%l
z<%gA^6pw%49LqVYNsP>91?B@>Z{SZ{El_rDzk;INSgko8;N6cAYCMgt&)$p_Wz|01
zzaoEbrx{{l+Y?uVezrZf4A70K_{`P=oP~ea@TnAV3Vj)M^me8Hk})h1evWaH2<)#a
zBbh=uiFPQmCI9eZf!Bgqp^-=h+?4AuNX)xCvVRnBa{*~%jtdA&<|{lmN)<UW8F|Mk
zLeABmouusvfs*WSSf*};)AcG7S;=#XRCJX$6)gntw3EsM)$@$nR=Gg!y!7{hhlW)U
zQrayna;wS^q{Av3%n&J3+AY1@rbzV^b)Y9h?*xkzj(n$OFn>`(A@OgIueqx9M#s$J
zf((%1T5*o$4k7{2alL$PYS(j8XNT}k#qVkBO?Al`mYz}e)R*Rwe?DDTr6Rz*n}&K5
z6vvj0EEjken3xui?yV<~PNZZ$(doXa#q4&agCj}@VxcPyX&#f!sMdU#?ydNR;%j$4
zi?E<<yU&e<eN$KgpRUZXX<RS2FcV(cdT%%t{1qq*ca^|+Qi-Fzcp|IJ^<VD?T9W8j
z&g<i#mzCWU8Ri6rkpuNaey%uR&H0Y%OX+lM<aapNMN^^iW|lauF{QohdXB0sD|%oD
zhoew`nmqkZI%jIK3ZugqcZ;lQWOZZK3t6BTF+<-b{?ppA>9?PY-^Q7vug(Vngk`9!
zt4P<2wGLWxk0X@VBex-ctJcwTh*G&}yiC1qlF|+pUG5j@2pUhOqJ=r~CxL|OZnx<|
zE*((enYD*XXw{|TJVj!fy(P!-7MMb74&!|nFUjkdU%eXL7iPt4%Y__l?6bU=^}ZlM
ze5H|6vw<|&{FR8$w?<lTo-^kZW0$cFggru9{+@&_eU8e0ri5r_n@{R5#<x{ZAVo^>
zvJcpsCBFG>X=O7mF?+5L_OH`U#00aS_;8OkVC{zTy(W?aGuC@(C|;Cs{RfhDFiF&H
z^F%^Q2tCxh6;9+T&%eCkenPM|%_(DbVMMKOqK}QqPqv^12`WhGdc(#k*}AF^^a+Vd
zT;5bQ?WH)VyZ7PCP?joVm)1|?yU?w&ruBNS=OH+k2vjVTE%*SW>r*6L(Y3ujE!BG<
z_zY-)GVys%g(bh$lQm!pVD>Ak(;uytMbdv_^|>52;K&2yS<U}El;&~+<frV)3^KP=
z*eZ>Zx1juTN5-#MPOX`rr3$xErrD11VP6Xp3=A^s*R$U~?ju6K4Gy29z5o@y=)1on
zU&FH0bzOTbFNhbVk;a`r*{L>@zpOU0#7*EOww4+zo;-)N6ASh@kKb6Ab!3R7wkVRR
zRF7HMds?uMjaKM?q-qEm7083+(9ExA#M~?<xd`=vAYNI#-_itsZtow72p#uxir|jm
z7Sbb1RQbFrF9F(z$Y^Ri4+lIsF*k1zJWUPVBgl3Y4GvT)^w)AnrRV~B8^c=rGj-;&
z*!zNbQ_wIE@ExniQ`wvIgjun#Suc*c*h;VEgvy%76tsZ<znyu0>QHp{r)%^TgIZ#3
zz&ZC7<kQiD3ax=zF--C40I=@}=AxLp`piCl=hfHeIejLXGO5R};qisE(O6N%&&xqq
zY@_W8cYY=i0)_+i*Lt-JsqyzMVjRYHT!L7t;Xk}>d^~x;e<o4v&mqrqvj6e>a^~Cp
z*|h!-6kIw>86Gf1bk4RL{r9HzAN|d{t-tS9A&}1z3zyJ?lVLMA`L}n76JX(fDx{v?
zy?0u=`7<+v1s~Z3nZLX#oDqELe9z4g-+S`C4Yy=<tNf0ftHs)n1O@1KP`#jNFILCK
zzlzEq0S4a5*!4x3F5k;anP84Z9~>e(S02K^(qPzIFz8ZL$C72-WhwK6?L7YXny9hr
z<j6bqsoi~(UrPp)${bXFI+Da8!-{|jqW15mzpxbE^kB*GO?zjYP(M|9pxlY5;K-e`
zHM3hHIH=Ki;7CN>6z=BWoUf&-_cq6ol#$XV%Lmt$q$JzY{!HoaJ@0q<j-!I2?GxYp
zg>hEtti$YSVYiGJfm`K3Qyn~_YZ^97EHeyJAv3cJ|79+%S~yFRyA#DU^#7S{`jW9#
zMsiX43p{7foQw8jiwA}yy47ii*1qvzE&tp>zp*iLytSI)y{_E%C8RIy_ex-qI7Chh
zdf&V_EZ@(P9dPj)YTc1%!O0)$Yi+Xnh1QM9Ln>6piiY~<nnx>w3coI9)4tEUJ#SHn
z9`Pc+U7QDs@VZ3`p$o_zWNkqnpi}2JE(IzF^hE>nC~AMC{>y<_Z4$np%xUYdXaBiA
zcJ$V3W*FK!G)pQ>z2Nssg!^gAVEu264&doO3nsWf<y!Dqw2m{5ex6f{(Df}?Y^|@$
zkwr#S8;%5!)goLmA3NVGhUYUi%w-%R#6HXKbh%tUkS$e!Aj1*QC^Mvsgtzd1$+p}T
z5M5(C3*(IU=@?4t*=ZhO9Uh9%N^BxHj*AW&U*Sfu7&|EU9V$t^r0JeCjHr9k+`~xA
z8Q&-GX74(0KA~Bh`z>(QvU#XBoJ$xSFaKGj({-3!SpLEpqeEwY>t#5Xm-T6CJUGra
zS`(ih23Kr-EHo<5(x9xZI%DLj;c?sB5B>UZX&RG+rVdV}JinuccQ;jJ!zWzksV+IW
zlk6t#a2?t=h}wnQY!1~KZZf5=2@_on+fM5D8B_?SW&?*Zl##MTlg*Xm%vf9d++u9~
z^+9?*tMX=Hj(+C*x8x8g>LK2j4Cpt8ky0+!-$YU)s>m{xv5DRE3OB9hO@Qd=gZ=TK
zPgv%Aa?~S<w@%lzfx|}sFk64K?BRp%v6O$0I#j%C<3?mw%R*cidrWS{xTY3M;ZaA1
zvS3Cu;>&HMeg8*~<5tmdI1nNJxmaB`56}eb<LqDeAtO&mvUyZ5#mclb(JPg-a0c6!
znOkYt1R(TG^d@F8`MQ$G5U?_D$#cMLt68a2*6Wl`P;Ch=ZgdqAx(c=W1sQ6<krmI%
zS7CWCJ-5c$VJ*duzsY6Pu1-=UK*qE`Z>X63RuSrmZYQV~BVeZSLqTiba!~c;VujH?
z?ZMp?5wo~c+cvL!`8-#Js>GIV0#Cy1ysj43g~g-)(BuK@sXiHdsv_3a7zDL>Dq@13
zf)QYuPG%5cn&ytLm9P1Xb6ukx^WFVAIxIBby?^u9yO{Ax_ve#d0TDCa$aiuy0i671
zKQhCqs&-Yj{~6osz6-5=g_dzylr;v)LmYqEcMQz+dEbpKszg0_Eu4O_Co(Ibqg_<l
z448J|b;BC2DV$zJau|%xs3E9f4QbJ=$<AR9Hx4vu0b1O1Q-)f1Iqo$V<sj+rYZz;_
zkg^Xm-%i>gGPMQd5JGZBSk2JGL(Nq(A7R#7`0FDhZ2QPqrWtqA<O*Gd`Md9z_0;u*
z*INqX0M)OX`_s6xW+43cfhVEj^H9Oob3F|>qm1lyGKXiM+xF-AU|`W3Yl}<2tHhJ0
z09jogh2YDyxIj7}stQ$)24UR~nwom}TN`K#YlPo0EG0cDdHbSf?pFA`C4lEPm9Ij;
zjcu0HBv251X>(y*&F`J^n3J>$_xp)#_<We9eERP1wl(gWc2Y7jv-5hMKf%8DztR=_
zQ)BrL)TD;D?0&Y&eO-YF24887uMX&i<z=d;stqlAb?DbNQx~;i2ufk%l#jHz+=gDV
zfLAlU!_M1FlJ`gcopVs`2_IC|G<dJhxtNj)94$SO4;VSx&2AWVNpE@-sI{cdqV)h4
zwXZ>hz$6Ci@RG4qKg6RKS64_R?nLxf)n+)l9Cm2Fja_iQ_OSXzt(L;Q+o<87C;3Bn
zT?wu|M>G*0gnvQzNe^8uYRW#I?B^n?m^$M=hsc!uwuHYL<CD%Bvsgw*9xXBDx)@3A
zv{aa79%C1ZWiu7qBR|;?dDFu>#lo@Z=-YiF-x`+rleoV9T9V27nY*(u-oxR{vT=@0
z>EaZf)hil#c~t?55KhPk-6oI%Ee<IH%RmWb|4KWn0t)2KD8FwC4F2aZy+D*DXA956
z+w(E}B@YKMd@)7|9%A0>R(mMarBmON_hVd+R;R9N5D_LL751aBxg(B!Jq6wrunr!m
z{#69XG-JmILO98i<f02As3@kZkjQRs^<y{_Q7VmuXvr?v;?S7%&pz&75JjU>&@f-e
z9Si_$)<2Rm4ZWxV{|nmpGx3eLNCwm<<XDW(SjD}d50%IK*0<w$agId`J8j|qE`{6S
z^UV<l`U{N1gHFHZmf}Uc;R2>Htj3pBQ3O<ids67iomJdlQNP7w0n;wOOH%=#CNqfT
z-m~_pM_)yXcI>wb=Z@LQ&l*U(d*yry_ArNw4qolZ`8o%o;WY49xW_Puxyh|maAmaa
zbp?T#WDeuFz|B($Z!X2na^+;n>D{j{`3)}&T&%IXyS8y2gIT{zC9CnemE+9tNO?hJ
zKl^3Z@3SV({W-b}v(g%&pf@8FyWq?~7Uz}dml=sx&Y$sW+>#e4EeNH3>Skpo&l9>=
z#Rcu|)(;Z4y1jH~R`i(C@<i#jQG)%0Bq_M(+)Lv3N_&>EYmo^fYYJM$=lk|K+m1Su
z8G%2VNmLHXarp#{oq&1>;ExWp<KT<hs<`#9jl!Ah^}b}-W;%(#h#WUelUJ}$`%|5E
zv()jhimHVPg})ESIf5JAkLMoGln|8P!2y4!7*D2(dDoyyh7lSay&!K>IhBV{wwv-L
zg)dA3zk@j^GVBFP)H7gr(drF7?5t*iYTTVw6_357kE50v33gUL4Rn7oQe4*|$p6yu
zFG<S0>QM-Y*w2VK!Y;QQxw%%!DhiE;88axO?<iZxGTYi`KF>)p_<_}Za94~fv5VUJ
z8U8#1<#d8U4GH39TdjG7Wk!S(6oz?R)oDCOWf=m(cWQX(TfYaIE2Xp}8As|P3)!+u
zl`*&kpQuhEOq@kV7M!=&iQ{Pv-c}2QO3%mb{~#85^F5!(MG><wG+_K;gni#mj!#x<
zbvB&&zWFVh>G@#Fx5m~DipwK1*9!U2Vo^C2)IFDvpk3LR)s_1i&*zoblBTU*7GN@m
zrFXHeM(a`O$^hrf9-WMJ(p}=F?Nd2mmH6k_1_0OFhI`mN&WnVaQqu;mULfE7)=JA{
zC?4r83mp#q86Wz39#Ue$b^>uo{gBKw?CvY!9bx>t+YU)B$&ssEY}o7R*Ku~s;u5M7
zQxn1IYTq8%(@|C<FwFOi@RDnxP1c%c@5ID=$lH41Y_=tLXD<X5hRuxM#mK0WAu}&Z
zsFFhg$_NuQc9sZ8^dxdAK!;4pgj>3;2ym;lgshfs)9{N#|0AN>80)lPFx%csK1wU?
zz=K`q&&l&xtHX~htxTAs!8!YhT>*h|_-D<jFcRr=>{!R!{Dx(*Sf^hL1ijbG{@LA-
z54#pUasK-PHdf(lmOs9oLt{5wQ}giIeIxbm?o)*MhF6SIBb$1Q^~kgQcF8``xU%{E
zr0&&0Uk)#F7$($;lT=|qhe|rY6`eIKG3))huAs_Aho71u!?3Y4d0}K*iK*t6n<aKH
zF^MnxPi2{7&YNJF+%5AzsSyM{pMa!`ddIy_J`-V$UmiWF+*@Yua*RqluSc8_?tV}7
zQ`_1ozS42;^?RdHR5)+_6I0df2fm+sxg{$Vlo(Q@6~3JK)O77?p-J02YB50-G0BLS
zcH+ORq8je)#qt{g6aCP~hKGZ*@7{b2Yu$?_5hwyd<`jZ-EgZBIHY}3iKIUw$0Vp=1
zmj_HA5chlurA=um0F2{GP2yjW!VNH^R}6WPn7{N1!j1u-J9r97{hHHI0nD<vtA?Uw
zql%xwp6Q^uY9XaXNd0VDpxwicz@+zXKn>U{n9=cz#PDkKA3|0X;(?p7)u)B2Cmb>b
z!ETSUO82@?ZR{FqE15ew2gu`mp63v#%!(%Wa&$ff1)f|?zinR$pwQuVoF6he3aNYo
z1r@(EU_4QPPSd)mA6hhmvW&F?*XuQVb<M(hP)CV<lym~=zcU=jh|#HSv44f7Ggu)c
zTp{)RK2hr82w1}3gA(Hj!UrY`0{{-nAM1*>rNeWQ)Yuka0SzhrRY`#SW>o4FIHpL5
zcv6dGy+Kn_X1%@A9p+;C7y10PXdq}{C#txA4x}7N0CuR|l}6mfoty`(>?((pC%!Z<
zljq+zWW4r`w*>xF?m?3=WSA4lPdLZQH@`5mUROdBXU5@-QnN+zJAXuFDws%q&!xt~
zGdUFdP_g6s6MlEYa4DS);Xs#NEoup`!cjwN0JCF!6k|F1;lZw@k$=m5YSG-D77peL
z1sgUW@p#Ct_{TV(hxBl#g}o`rp2hM<?GKhQ4nhN&RQ|OxE_0E$VE5vLR4-!hk>H6r
z4Y28q>A>EPCk`7@Z`-1<=I`?C7vBAIp^QZ6;lsI~79}mhUozIQv<|uf8awrFt*PK^
zg4m}Y2{l4i<1x`;mL+R`dhMh<u$>rkyo5n4*Jk>tXO8ojAB_pa%JPM*xEJ(M&cY&J
zR{G!{k-EQ=L?35yzhmsPH)osshHd>P&6Gv01rwNj_s}!*cSkE=c6m#SWEQ_>in6Da
z{r!ZLQ@!Zt44<EvfMk0(&MXwBfUcVJ;SzMw{_Gg8zt#tELoqIQTc1M9MMA3ly*g>e
zxw`6R4EY@umvn@^Yf^0XOE$Y%tMjaLtdjRW<lA3z2O5SU2wdHXjcKd9%R6sQuJheg
zFPGw%C&Ke@ZVp|weR^p_c{^&pw=nO*SjLGf=GonAQ}Y|N5io4tw^q$$dc7Bm9YU`u
z$LcaXIk3iFy}s8299!+kd$Zho8VxAS(IV|wU1$(L3(!0eT&xutWAA2|3Nx1mP-^v(
zUbXad7Og~@>e^Gs7OmIAcG|1S89(8EA~Rx91;|?-KwvRW-}DbL>0^zC$06T_YUIOv
zQ@k=w!RK&4-*wfJjk!X_As&@b_wHihkqBf)w=Jnhify+yH<qTn8r&q$<onFIt0DHv
zQ|#4Ha>1VQ12H9!%BF0dZMZmV>vfrJZ$ZFT*1qLFyY6n_G>3J@)7PJ3w@@a$qTR$K
zsg<P|Gm*8O$0hY#yV}7*dNO`fct^}uGFz6$&U3|i%a7z?pgR5y5_4ADRH71!aY5DW
z^|pR8q~@R1r#QYDWl`jn0}o3wbQFz$<<K=1_TxZ_&^lX*2dlk!#eM4xQywUVbx6NF
zi4d{SI>zTR9E{e~B`!va9x&aYIk>_RMBxETib=+F6Y-P|Azo{(MZN7A^ZTzfch*?A
z)CVACmEDzifz)cHVQF<_|2!KVjiCvMy{7&Xn8+G{9RL4L5F*yL4?t-QOVJXfeW0F2
z-iD7DavsI*-rw?c(GY08%`#3dyP7|AP5<`mV#gcF4{y~6l@~AG@^NVP2|D*Akeg>f
z^f%Z>8G6?BrDA%r|H)aYMctWYc2TI|TFXZddlkj1jPZ4PO%Y}r_ul<rQhnbo@OP_{
z+$B{<nE%Ye(qs?Ln~O{6f^geaYg@W(mOEn?mw2z(=ERJE`TI7bgpPb7UP9V+M7aE9
z@z7PUYv@0axmmS9Y3mMns>117_s|2rNXtrB4JA+)^9&sAv;Y(Ek>VeF*uBlIyGM)a
zEPbk5n2F9I2bH89hXnfIN+D@AS@G-)Uo_i)AQ7*7sU})sG!nR&R1r7hfcbq`V9ra0
zfYPC9g+_zlwPV|cCr2Uj`?@lV<u0SNLO34EO150-JIZi{1Mkd^ooPN-?8X}nE`o=@
zTEa<%)~3L__wHO5wC@a$6mI#wu@u_P8m%C{zi<8i;FnNgdh!>k5r2{EM2V;C5t(b_
zov|P#kLKKV#-qoa!mA@buc$=ZW;f#XW(bqr0`5vZPT`idS84vZPESGQDP@9K%7?oX
zrLKo|IFGflUtFOTmHMunUL&e)L)3w@=~$=Rg^?ot?AA4u$xbx7|K=qb?gSE9#}jwp
zdh+4U{{jsbJEW-|I|A?F(o{)BOv7qXPI5~6&v-sr->%dKb|&5ie-p`MECFCe5|`(+
zgjupOadbjSPW7G46e|94wNcxn4GW9V!1Rq(nW~G7Of<|}iz?Z^@G0RwAVw}rep*Rl
zb4JkC5X4*ZWZ3)b8AkoBE2`YpQHd9{zyPalLjl=g0_$cEEm~L)+8<{%CQI$oRo57D
z@x9^uy{rsVmKD0MHk{RYX*uki2MGJ&)V@QT(oC(1bA7W{5~QdOo-D~##R%atP09Ey
zYyF*xv~O0pjI12Nee6WbG<sF%t&^%Nw$uGV;n2>Teccso)%04sim0yQL3vZ9X*)0W
zP)`ZiLBc%-!Q~iA<}4@$4LH(5M}G(kZ^>ZilCPBWCPO8gJW=&pju55ZQFuUv*{k@D
zW$OIcRjqma?3Z;e%l^~^?pjvt%@E2^bM#-;_lgd+UOr9&2_tQT%Kl7poS~Lx$)wE*
zjWWGq(Mb9S!<+b+Ijq)XA>9o!ZPJ>?W=0xnzHA;UggAqkJ=9!8n~c6|H2V+aU?fo0
z|M7*IA?VLOSIE(GI8h{8JqGOByt^LBGRVhTJ;fPX8T)`^8$&V0xnD}52YzV+<(*_V
z1#ihdKFWV&_exYohrN`vxY6zznk9|37RD6i$p>Njfp)MgmR2aGwGIn81WxdeMHz{k
zG9BYnzXjgxeE;{u_&-p{lHXld;S&pv?Zvk-x!s_vW4rmEwiJiQar>ki_zAUzn0}T9
z{gX$qB0rmR%Mo=yJxJ4dl;bGGLjWkZ8L<g;WRLvex^&+W9~Wr#hG}b08dMgAnG<y`
zrf-m_Gcd}S@LQGAd=z1&wZqzFj&XKx%mtW1DiXV9O;3<?D#(Ori<NwiIKHJA6UK4X
zY51w7w<ddAPm){S6h3I?_3NU^Z$mIY<(W0s>#LvN!mQC7f0e8QtUacmVm#tr1@UE(
z;Z6Gux<%%$ecZNASL>?(ooS_()#!fH-^p6-Y78Fwb`27_tfTlUe`+FaH(l|B2N?<B
z+&WI-C0||iS)k{haun%@r7Btq8Q3QiS$$xd3RBk&+r>%^g9OO&q+TGafw~^wto5}~
zFv>toW>mw_6tF#?j8=Vq&kI+VtOkKQEk&HNX5ZBKg&DadXMPC!X?ka|UW;m|i==2L
zQlRDzVDNPci9Y<$krVa1ZgjCLAus39F^)C7iu36yTbTQPTiye>;4Os1t$k_olqj<1
zVJ%13XV8;wBhcJHgI2vc@~Js7GfsRMgh-zcEdjA*kO!(K5>*hSXhO>qjr70jb=PLB
zjRtIMKNUK)n==a(N%7I}ujL<Lh1}{87Wz2&u*PP1WH077-_te7W8=vES>~08vTdUL
zC+?Q<${gWY_B!InQ&@VT+ZdlPK4`(G8)&Vyp-EYd`M2mamtT8lTd$xmJapMfTW$V&
z=wf2^WpA^TbxQQ9*Z+mE9MCRrX4X6jesEifufZH`rgPI>v)g-DJVSmIujGH^T`SH4
z9TFplnH%|I6lBZ11AIQ^2cma6=1;t>XAIxVz)bCDeIb_0&H6#{;`*j{_8nAcpO?d5
zl)valDLE%(18duQlV8%Q9r6i(Q615{&#EH=ZH){^l78y+PW`;$I<$?937+~p6Lnc%
z`^Ts28ULr<)rBqu>ij1#2<MXuzN&!-L)f#86FajOi|qQ?%;8%*fhOmYBr3@Fal_92
zF4>k^%;|IdTE<YRGD;D0dB~{E3JDPbT_l?PKbFqJpY8wq+p$M&(HgN830f<Nz4weA
zx`ZM+kW!<=9-$#dOKcJZQL|Qc+G+(cimE-@s-n6-zxVfk-2Xxz$;o-0^E}sefz*ON
zP`E<2kj!*{!NgwdoT^Yp?{QRPFXshv0DJe;n0($(#7x!KLrqG{XP$hI%-sQ24{qI-
z<JbhNYNeEL40#UQt@d}hFr(bNeT$Uosy1-fsM@fdvNrl*<vw0sOkE7b>!?&wKl2jJ
z{F$kU8=nWES4)8o?A`;=oC8|!?qUH8GeA$@0AW{39UtTe+1sXfe+}}JPYzYbI6eJ%
ziO<7QFeZKi?J&6QVGIcr-$k7E+d=tlK4CaVENh9<<I$9*HQm}z_8l;$i5N!?;Pfmq
zj8>UTCue5p(=wf%T~kf=jafBl<V>vbxoeK<m|MD{y9q6b>M#A{!B!{$%X1$<B~J^A
zhD{)97kW6`t}k3(UFD5Q$_XLX^hLP-aM3hdB<n{~DJncMPglUsN@%IdIMICqE(Ol%
z7}bc`cN_mC*A~Q}%x#NET@u|kU>>#_V;DpeOVzp%2Pku4V{5;5xr`ikp$3nH*sQFr
zD`Z*jMs9Cuh4J#leq^T{k7bBx5075H@J5*Dac;VGqsWb*4&4yj>C9tl?F~VBS0pu3
zwlIgPFIuIgq6l%eMmmEmRIFb+2JPC!mF&Q|NE#5DwrxMLO48S5Lss%~2%|~9Vf&QX
zN0p2-0-XjHN+pdPJ0UKadG(<@(UkK#nMXf>*^Qr3r^s>T%pdBfAmJUS0}{9yxQx6;
zutw(|bXH(D*E=<d;-jjNE{Qh5Bke!Tzfb!8>Evn+J%3*w^HjK#+xfQkVM(R`K#%N9
zi%z{0O92MiAe;^KRl{z2*P2-vy!sq#%pI{-;Hk-urYL|n(@KMv1iI9H9!ufc#Zxwi
z7+bN;b;%<gXP=%qU~G8rE70RL{oq^YlkQJGo@~AFwD-1pdEaHu$BLr0SVJn6Usugo
z=*;0F3NAw>4^@l~7=Fzat>gse)~pykiv0LbP~?lp`U|X&1cae6{R;yXb5A+HYi^MO
zp&Qm~1q@Oey;SV>u5Vl~EiQ|h@!#(d$GERTMD0lkALbZh(^JAQkj2if3!kr7>$9%S
zZFcb#m(z0%74jZnQS8Cn_S^8j3y$ypMc)-sy%>*6eoq0@&bOSEgNxOri|yksR~B$Z
z9WB-ksl68Rkjsz@jX}z`spRXsng*N{JDla%YOQf-2y<H=#x+A6+T*E0dHQI25;s$m
z8efb@u#uV^aG+hEZE<DAgQsm$)7`AwVY}TkKjYnHa-M_mcIio^;?X4V2QJj7zv}Mc
zh!BT@fV|6c4d6&KRkPGsEExWLlep!d&UPVGX;QO@U9;9wP+jtQh-mjth>Z`j?^ojA
z`I!xxoD-B=ZK6GFT))Qm(x`1+6{nh`5Wy7ZUFLsh5^@`NU1rKn!vEo<r*5k3<Q&-~
zHkCW@AZK8nYM_mbMQAmv@c)JFnU!h@A->sed)^Ea?=!WYS0|Q5L_2uTp9nf!V%-DI
zOOh$%T1Eu`&KOZ{wkC#;b^BY($nj<Mnd!83{R#Y0^qL{(qWr@}%nsRN!n1k8qa`H+
z%d_<@Hf33p{I|!0p78sL$oIzXW|x#yO}V9ZxB}>%h9@{n83cIm*fX%Kt4NC8i)+?x
z84Pi|{SIObf84n$`f`Whf^6gA7+XM~_}Cy-|JPI?bL8qe2*6m_t!(z^nEbIMd6P~m
zq|m1?sg_5&cd06VcV$9ym-l34w9mJ?ZQmEpL(NMu+9jWvhCi=jXStcWP_U*e17%-!
zh&~t0MC+iSR8a+Qib-Fja!&)o-h;Ehs)m%p5k###<w7TvNQc?Z?31dtgr7yB-v*fl
z0gm`yWp%zQ`$I0;%Qb-Xlu5hM6L4~a?*2e4m^bO=Ep0XQVnff>t9NSyq#oJm2n86s
z9H6G~&)=oLNZ25Tu5~|uml|*9-4P0=eXowsu<@`?DG0inCRaxzZZzCv)Iao{$Pk1=
z5^HrDg6{>@y<|dMH*BNK{8j4w$d_3P9npx3{XNMRQT;}x;U>Bx)b8WZZB)(L`<F1*
z4zGMs%b#&4PYsob70W78P9OK9KDAYZs!u?F4nkD;ch3@E7j6Ch%Ene>kc@EX?~+!O
zZx~m<qdIa}_B<)kx)}EbRbPD5d;adNAbn0hLCrYeyij=Jqiq#%(|p%9^!JOCpvSUb
z5!H&yZ%UH9g<ti{hlf-^K7mO3f%Jlu6U`UPyK0r^4*JdJa6kob<r{zo+_1zv?&}>f
z>HdxUvepbgcNaXMj}#HvH`WC8Qf9bV9lSeyo7RdwDsP*91#xkrfZz$wR*8M~IFwZE
zM-f=yN9bSA6}qmibV*yLP&|BeO;7r(%sapQtTEhfO4`=r7VsT%y_EeVkD_qbm}lO#
z0bLLHlkPj$F4bI3`)1*EPOTF@lI0*^k%Ba+v%?PXV(12}{W><SxPou(z$u)0XrCaR
z6w0nZtId%Lwf=-@`}<9)w6WhRyy^mMtC(|~CCm=(SYsEm*e<K@(Z(1nNTDEm2xBtW
zoWSo2I?3fGHpc7@7w%&$8S{wt%wW+Jl(2>Lu2-vDAm;FP%WW55n!wu+y|>N$3MGGa
zX^g`H#U6JsN|81WllKoC^R^80QVOLcyO|bq-Y3hhWDgo^gdQJewfzI9Fv+a0calwx
ztXY^OnLg&t<b*aXt=~SW&6yz$4UY=w+5Cl7S1(LA@g4$oI`DjX&DYj5jcgcSu3z*(
zzAnVi;Drg4u_9A1;=WOdl+Y7{F{{ZV97yP{wr%rfzAP?XeJ%b9db<O-(s}baU?avO
zTr=G-Xq~8$jSxx=V@SZZe#x~^m~=mfbVuWSHx7NOw4_<9Zv3LBrNlG|*LdP1gfA<v
z+fDq{bon($ymX{ds!ROz_K&>A($maLAV82jt!}%LC`7)JjMFQrvw(<Voi0K`m9l!n
zkIt@(z>_%5b%ul|H0xpmh0y2C=H2S%S5M1>qNE}yE~&Smnn*cI*aB54kgE$0z%sMX
z=VwE=i#_~5F|5+6Q`K{eb6pY`jvMpvk?<m!0-l)ZrWC;*oH02D=Lcfgbvomg)VKS?
zO<+Jw&RxJYpPda+i=9PfGa?>!Z@4wyZ49%wxT*zkPR_*X=sPo$(<M>W>)f;gZsgss
zu+gq>Z-}7G?55%fxM<>!naooD1J93!`N||8KIQre6gli>hmJ_Y&vsb9RTdlB+>;I0
zHY|HBw_kXzK<x6_Ov7e&*_nU&J1Xp7luD~wb)ir1#2HuG6KMG_z5|-rqONJ7Yp-00
z@6fdRl^1}#DZVeN(qS&NPb1m>Uj)9vc=hNqRp^Us!qb6alcCkEJi&Q-(mCaJj!zq`
zS|g{fE8Q95msdVpz2pB)Y1DTR`@OUOcDAE?^yBNDpbx%QK}EY~kc?0<97Uzi`eC%x
zm4P~FZ~9G@q3d%Ep9khot?#D_<zt^Nj}iF!FLt2z$tn99IgPR;dy0HC<VpDFJqHcB
zlg1fp^T6^c-!t7KpBEC?(x`_0d-82BWc^lirDPh6ed7tfH}=!-urO`)%RA4AN{d)X
z`SNhfBf)Bt{kHx9dE%N-#&K=-En@sRhCw0C;ODcx1pg~?f0kk7z7_OMi<);F5fYla
z%i8J?(JCy{;t&(0{to<HmrXTldfCHa428UMLA9L}qN60E05~cAr5DX|oy0@VzP}r=
zpu!KEPHRk4;gTOYZT?B$<el-nsGqaAq{Q-3js*|fj;&VwbVs{Ef;s!n>YO%3`^kO5
zKn6Sv6C#!M=GN<P1JJ!Q?G_oz7eKFSu6I}X1CfpR8h}UBCG8dIpd+~QH~}o^DC8o@
z2^;2{xoO61`j$w&i27e$R;Ej)3qB~59<a?0DJ|>ED0?%mI-S!iIKg=T$7heWAH<NI
zsz=&6Z?jN^dNNh@x?PP@y3f@<hxXj%dyf4lRJ6t%LX8n#zbMnh%tY@jr=EW>=-+NR
z`)<mFQ8U)r-*PmLajh!(^Ey8GnoU?{^Mlk~o0@=dLZ7eJF6l;uUvOI3rQM)V?b^Rf
zz}V-~GG37{ungl^LWwKC{seMbXFPaOz8IvBYcrG0D_$c3^XyuxP447FgI!&$Pu1Cr
zyVgsb%v$)~D8TAR8Z?H<iSKK7RnC_<kp%@e;f4u&fw_(zCk+l(8xhr|!CcqIqst@8
zK@!0L8XLiYrad+7F=lN`*_iOV8vG!?TlMOT<HJO^G|@49b>+ANN<9~dpu6+Ab{pxg
zJvJ=%RQfSKw9D+3>E{3M$DHZ>HtLzZqJoe=e{;Vgg3{#yfg$ELC)RB-hXIq3wP_#!
z19^K09zcH0i5mDD3U(L{kj$PTt9nEDbF9_PV%=)Pt5({5a=wdAb>E?=5nk8vGf5V)
zwVSDaZ;UBACs6)-d2cNDA*uE=ag9(h{XPQZk!RT$=j0bb$)1ZrknFjE_F7&Cd3Dmd
z3os?2MXjqeQ}n=C<bdK@_KbS=TO!b4SN5*mj-NRd8ps^fs1V=B@ZNEqzgHE`c}H0!
zZ1<9k=zjZ;wH$w#Rx2Z5BzSyIUV-C>>(!6Clg?)U+F#`T2l{zK4RC)RmslTHn4IJN
zq6jiecHO~5BeMQ6T9^}TvJrnj4v$}{`REqmVHA-3Fe&HZ#J0E7HMa-*>JGwNEeMt?
zn9y>jmsH!{N7g({5mqwmI5Wu*3GbNM?@1zz0H5B~n#%c2UO~wQl6C9Px0J{07vz45
zxaMmP8xx{Q`33L0mD0XvVbJ$#dT*b@Y9;FI+v^yFp=34aY5m&NXf~U~wWCKWYAgQ#
zBDOlLPJt~`<m3<gnyyRHhFy(s-#$KESSYHe^A8Nh#k60#lJW+>(Ounr(dh2kv*&jV
zLPS19v*E_oHa(Ps9~1zADCF<00s*CPSm>XaNwwEU`(p-IAhagmd|y5_WW%Nt8)MDm
z%_A?Y@8QgBaDYk<rs|9!l<(}m{q5dTiy%TlYH7(X8xw0u9GXia7>HDXwu@;8J1lDK
zSE!c0XsV0cE%zrxip-HUyHf=@{cB`tUZ3)4V(CnupRO}YUZ^KA#XhjMoE7>rwcUDK
zxk?i>ET$4a>HH#&CZc)M!!MVO_!xQNb=c%}a$da(@U)@gJUQ?dZUNggV7RC>*sd9G
zZ`qM5IZBthk$HA-`*NO|@9YcN%Ex`-mcU@qc<h?&9cj%(;=9U#k67>122ta(M>=1)
zK1M9)k6pg;kmqde{SbOJl{tEvZtuPtNYz(0EUcH)D21945_ov2per5p3m<)>d*VJ}
zXg_k?jWkVO<WDGTULG`8Q^#j!|H$t#HPU^0G3|>_l7;WD8D*Khk*JO-A-UPx=f`)`
zBogWhMSfnF`x{aTUgsuE=;l0Lw_(igU~GsJHbOb`VhP5txJuP<@4xOWBYf!*FIYpY
z)~!6~_r<_QLj~_~tC@{=b><sICa_MvIJB`p3|bmYxt`8`z2D^gXe8M$BeF%j_>{z(
z&LO@dl%6z5p~yb}wiq9tReCTkTsiCcZWK*VQSv@_D<WvsNfCMxKb!Sz?Taz(jyS8X
z8`%#nfxH$h9}4?))(g~rb}I(BA@uPw23M>3z^?OL>%b{RnGPWH2ej0D57LBk1kkGp
zhE%HW$1lDGqbtprl`DGISq7%--x_YZm`z}u{+xj3G)op*=9e;*%Xz9BxD*!^&A_5D
z37D=$g}nSwJ-4{&jLw2CGL@fkIysnMjayL&=t12a9Mbu<ZR^j6duYcqz>Df?V-fNH
zb{oCE1gqfB)jSIx#Xrd3l?W~E-4A>-Dbg<RF`*6PZAIx1bYJA65l;yZM5>GWzTNO=
zaXab19WC!u5gB<VURe4?NosHC3~eRVHv(~6a9PiR>sqAt%C!Sxy;eI`rEKAA9s6rp
zN@K07qvCdnCFxz24X|HK9I6QfJlhcbaXmdw@|8Hn5SVpDL0nVM5ZluIuSD;K8q{0M
zB3Prx6k=w9yJ*hfaYep&N0tXMkCDRj)vKH-D2cXQtQgnicOEj=BriO@_RwxJmL@!X
z;II?KVAZRqQ+|f;0>WT#@vFhU7r)xN_>Kat-BLLe@*6o@rDeG-x0cO2-e#$iv64M4
zptuBi$A)>V<~zC`=<|bL_t2T=I%a7*3pt1hK5!d0zY$G#%;(q(p3HWU8j%RGs*}G!
zv#-<1b}xSPXHLK%Dra3_y;-Ej=37fAmEU}D>k2*1iCJ{^mh~y<8OE``*p>gX(@STk
z;!v#b$(eNL6FIl<nTC@uQR@WC^kT7J+uHXU<MYh{4z@LtsACzdbzzKavP&+68!De*
zeqf@rZ^ocT3>8wYupvzdaRr&q1j3cuz%(WKpcHHM_Q5@y&C3YT$=D^{^t|t)B;CM9
zjONXp?GCd`{<+pn7lkC2ha5rghqsB!=gwN{89=8VHS?0Cm+T%s?>lsYEIj~&b&B6r
z=35`VH_QVO{w~GNe0zpwGI&Y-UlD#5Ejt6Wrq?mw9+Kojrofyd)qQRccFR5h)-@~D
z^29}e^C4V6+;7dm)^5jW(XZvw)kz5uH13q5lk*<j$lUN}@S2GceJyhIA82rtf8V_M
zb{3!r3L>j%4M6VH+s>Z~nP*Ay=huDX`E$Kd#$T@`Y^A8oc7Et&K=a*Pb_C=!vq<!Y
z(aCbi_yMbYHkX9M|B7&6mwX|kezo_*S&JjdhZy3wZ+12EmaCz)yqrpxN%>VQMTuh$
zVn<_kP=OS-9E|b&L78n%I;-vsp>wx1m_Z)9lBX?fK#A))bmY-<wf5v*MpU<kxnJYf
zeCG=KMI&Uo@+JAl&fHgINJ0w66U~He^uL4OyrpOk`v?5ks#4Hji{~fwo4<Xj2|Bqs
zVxoY%wpNht($raPWIxF$^|QCmQi^fm&yEP+26iRi__H;~IsA#LpOL6Nnug@j==V)9
z{`<=#+(L?A`K6tSVKGNcN_M)}(k*kE|E>))hHrKCfVI~m)GZeG=GnG|N`lF^StOz{
zTCDF;H?6i|L=K(Y@Jw&(x&fun{^<AI(Zw3{@|bzdKcP<+-m>Gd8AL}mTR+ItLCvpW
z{UBK$-d<HeZZ=1BjO7HW?e||?!y2=4?0%p>3EGP;XG7DoropdtJ)Wj+EZk!_@OJ+p
zLRAQ^nymO)$+lQ<$W)B`=Hwz&%E!xCxIr&|(=_*twNi_8r>Y?6(Z_3x_0J#7d~yCk
zf@#GgFDmG0rZ`p^P(;*YT=_CYCPXOf9fY(!#T)kRLU9V%OyaX_7#9dS!2JacD@k_;
zeMYi0rkM!a+WWd%3$?_TsV=!G&yR($6QKP5xht(FQO`#a43PCKOkMqwKQIkb+FeyC
zeHE_O?;pbDA`|vU9d{H{@)TXsA?C+?B13{qGUHyf!{U+i!(i8DRNq*J9pE>uEW^=d
z(5PZv_rdVXZ&8()nr6jE-R|BYInB>GNwa#*-RC6A=VX><KZUA;8b)`*95cR@I>acH
zD#!uf`v^nf1hoK-mb~n;{J&v~Ee(ICQA#T+uhAiBr9H0AM5Z47^<B?pYG8*^pgOA#
zR9f7Kb3OOw(w^<2Yz95x2yS!c+2>B*zp!+~!KJT1T>6dDGjI9ebPQjjT9`De4xFW<
zK08UgAquY-pODX#wg)}d_b>KK=b7mbO7k;+zD1HG2ZPNEN`iUsxfeEluYP9yH;7;4
z`@T)7D#qK|Nc6T&tqKook?g(H8e56C7Cgu8>E)Y<8l><edeJl$q~l5me-o!Npb8p?
zg$>3m%2|gEK-IB>wo@YN<|3;uvNwgP{2BhX8e{sLZGNu~U79Nrpw>nYam~U)NdU*m
z;dgC8nEQ00*DvDyfPu|#ti1yBw11^cG{u$RZLxRrd=MTo-PwODQ^dXV3EBJM^Q5~%
z!#BP6r|()nw*M#z;emdtsM_v;2Tk5H7m~DfWAKMpp3gCWL8zWCd}FVN#xWw+^VXmL
zR~7BEIcgTmJ}A&TH!u-eHSwtIQWlsV-sdcfZI%99l_1f)e$X~s0`|@eK(;Vv%pm!#
z`SLd5%H@d4ibw2<1{m)kMR*aE|E29#d*^?khh-xrl9?UNMGQ=vHTGSv4=$1L$sHHC
zrG#^D1NW2m$sjfndD1P}Uz`qJvU0PN*}QqhXa&t?$jIP(uu$0B!k3)xTOPyZ4AxA0
zvr7DJyr{=WSz)=e1BUtW5zxn5R{LF9n_g}jwa3!y3PhKLSPx6-rRn%TP|)W3wC2~r
zr~5XUF=c>0N-h$>SPLdC;6tv@1-u}>n_rrp@N~QFgcX@q7erqC^&I`RPL85)(#>OK
zix-I$&vHvwNqMza^}rJA&ad8Vkm^L15?Om@8Q5?R{ATTWB5C`c|N1A74=3{vVXFr}
zdI~O{{{OxdD&c^;82R)42U;EL`E+=<9C7Y6H#F)1$IJ7!lmk%xB$f?SKkAKrkbho~
zD_i%-Df3=mt~Ezu<bfkV0qT?gL?AjN`jLDGZ=M{42B(=CKs#%$`5KduYSmKWoz<P&
zIM%cR2p*ahPbXDsVUmc<L*MR4ljMF)^U;FEw!rLZ2{pI}ckhSUn_}&XCz-NO)^a?U
zS%B5XI$q6qT|T=X2?VVGsXTP6;ni;+nTkhena)0lr436UM7Btk%@E?X%TeXr;;aV0
zDg(=z=c;`EOiKR;^7k~jk_PN$@lT{-6__`cLsv!MFs>Pry58gWA`O!d?|gQYn$Ukp
z47}}Xa{NhY&Wd<JJ?6OlXj=1?+0to+<H`b<J@USzg|yC&B53$Kc$H(72m%eFLI<4{
z8G`kk8HSIm07aWY_v{$_Pfy>CBUgA-%kCfRW4kliQ}Xv=Km7yNo<}Qb%pAdNtJ+Py
zID;yxO)U4&O>=~>1Xx3o#0)q=$TV6=mc`>0xX(7(!nPLj6`C`={_LdQPS}LjmU*#W
zt+<~{54qKSjY&jx*~z0Z?bSdyxZUy#OHgINJJad4s*9boiX^jK|Dgik(stBmCi7SL
zS%F#g^(xDNx@eoG$cEoTbenZcNOau4_RZEwE!k15Yu&V;920m`e<rhpO@vH+JOIF`
z%iVnm##^L8p$2w;dj$D2$|}qtCD(;Gm6Kv*OHm2;u$+V;1~3gv6^*hUp?8Bk_MBr8
zB$K&mp&crotyAW4sKI_mN9JNsT2|sKy<Ua}HD2>aJ1Ra(ADW0GHmWkl1x)&IpXMW(
zX_d8Hu6}K%9d4`v9$yXr55Q{KTn5a3U&6tjM#{IrkHiv09NFQsmX(iFRZ#c33HYE>
zcvl)fj*U&TPjP}yLInvQe$XtxOXXLNg81Xm-{z?T(pew|yTla${refvox=S=)g8Uk
z{Lk(O7YcFFWU_2chAGNtAjew;+le-hz=tVSaDpGzNHc|bn-~lRZ4TG-KNVplGTsrd
zv-ssfkj6#VC{B<=1fc?P;5)UD$G6(+5F3LG@pc~m@JAXVnuRW~lad$h*VVk<Vq<!w
z2fq*#_-E+7_EoJ4`fke6WNm+Uc%&wgprmNvP~A&2hQ8@`85x=8U7?&(<{#DruXfGH
z4aYlruLXzy14(R)KQdJ7QjH#}^C?PFs1<*N7!Ve1Zz@#KgVshFp(-sM=4S-(hz8g>
zfkI()ym_=Y8@+R(dDi=xdk9i?*VQU))Q^UBy$~`LKdSB+){k%!hF9duxe}wmg<QHw
zBr~%!iyc($c$!Xr#Aa*nratMH=saS6L&K|^_NOlF1DNt3h+p)Ms)+3i&vr+{fIm8o
z-`J~0p7F*I2BMG4nFE$3{@l?oL&-hN#fgGZ^HD$1j5z|Kj<HjIjfO_g9oQfIcDdRa
z=~#lX90vh=v=L&`ihJ{-<@TzFBz(*j3(IohHPD;XD4yZF-~ibcr!Yt<^y+h+P_7z8
zeRj1BeQSUb2`aIPW+PwfEGT%x^hzi*f1xG7p8;=EyYyM7T3+L&&X|R!+@n#qbTO_j
z=cn}2f3v|DnVvv>8c(kV@bI3_XqUHnDz%IFAd+)a<>mg<<3=MF`zto3$*q^#{%4Vv
zuXuY?A^A<#D?f@7<Sv!oC+fkU_L*GRO^A2T;@v0W$<nXZ+ieSm<g-U%PE0fs%1oXK
z1aY=dW7iSsHxl0?$kY{Y!=*37Xy==?rrBk2kwULqc%u7NCtuxOCIJ#%!7>-;p}v`x
zG&@L`i`rJ4*fd!h4W#R6(fuOpy=v^y8`IIQ{c0eDc+gfHSfKoSHORXqh+(^;pD}xo
zX-rW^iiQR`Vu_{l<A&|ZK|Lb7`yLweGiTiip)rAQu?u?6tS?il#x^pomK~M4nr_l5
zpZmjS9{ikgh$VgFs%SVWrX%5LtE#Z5)KPP7R+kqKB5~Dk&t@HK%5fcy-&H%UqPzkW
zL@oNOOdXooL8>ZOam<u);;yaL(-=@Vz7KhXpD$+t-#sv9lTV!1(F9cyR`43f<t%cd
zISUEnsRla%R@nBY5_6cH$8%5qJOHRbfK&GSW6Ll3bk>#yA0X9P_Z=ZB>S@9`GY1z#
zN$RRS!qZPK{D2UH?CCAjTsTp)A0EW_7M!6hk*-zbh1VXvA7G*tkX1KFH-RoTp=nOf
zrQxv}KT0;64*4j|N0FH%5sp6|cy94e+AYVYP;b=SM7~D&pS1fGSFX@FV!9_T(n4AD
z#pFhgH?KWs&ux)^FpQ^Z9&+j|Ko@l%B5{c1SZxk}3X%EwZ0Um0bEPq2TSKJ9vRKmv
zSs_&&3i@oGTV}K5GMSCIeB=5lAP;dBgg}?;gG8Dex~kXO)45Dx>Pn_AVzb^>tX&~0
z+^a5+>#f66I=S&mu9v!#R<_~H_qKO0Cp`d%jz-$uVCc?NU2S6R-v2$+HZOeyaT0?t
zOTJ-2dm}U-)wY8dGAMiX6kjXxZg7^=)ziiVp`y_<TT<zkj6Ci6*562l4d>szm+$aT
z_<QRr(%my3yVALgS-ph0D+sGgK_`nVzEzlbDDC-R69<CaA?+UvErzAr*_+~n{{lx-
zQh;9JWpYp&x?4rds4yKc!e(<tkVFnV()>ze?pB}d5-ZIoKAOma<N31@wOOb!qWS-t
zIgK$JKS6rs^%Dx(0y~Oy1M9e^u3KLP;{Cs!nbS1Jyd0qwf)}7BgX<_6udf=JxI#-k
zk14f+94&@g&s<a$(Y%0ECxwR4?!&x;YG3XPA8%2uXsy*Z-Mk(7$}qSOU$)xzuD=lQ
z+;tWlv+xME;MSiR3jKdYu}y+o#}lPf2;MJ^ab0K?h}mkYZ4azSd9bF;U*OAYd;jH<
z>ksD+-EL-I+l3mtNuVAY46`jmT+c~2JLJe!WxDaOet6ST>t>s2=ocK9Jv%T)paqZp
zCMopZH0yUz8ThUuO`-R)Ls})cVS)Pe6#GJqaj=FK6yoXt0++irS2O$N3UQJ841zVB
zM@@0OZ~!hyzupx0UTqMtt&s?oh>l%?xn(aSa~@wfbUlBloH;W9m+e#z#*@@3`;Lpg
z$>+yL5aZ{-I|K7Sf6*?imWAf-+m$p4E=`^T-`|u-ASu8FRKO8#HY!z7MOgW}bVYqn
zxl^^rYl1d(5+U3GdY{6Tue(LkN6Y!;GVEv^*Fg1tY*plCghEw1OY<D$uTTSTI>hid
zkVO4+{I@bNN{=p{^c(oUdc9SWWTAg$Ac;2c&yR1}q~5$mi;Y&8xY&?5s}KC5n+TUp
zX0VsFGgV5OZ<XFYZZzGzyQf4b`5jhP-neMQk=LQLr&737+B0!fM_7Gk@B^ra5!%&R
zqb=s3Z4r3h+pS&Ph;lP_eg?N}5FD?o?Q|Q}3M9NoC*R143}8g6=KDm0oURo`)x=M_
zSBk(Y278UJ$DB}+>fxfE#9(24cdrYU*l{cH8)M)TR{f?^ox4adSD7Owe+BM%(!6+Y
zOglbk1HH?7LkxAaP~#{=IRf*@sJ&obP4-JQhIwl#@!T*cRM0w^2@F9d{-g<id_7D?
z@7v^Q-dXQ;IEyXSb!)O1O4oP$U4WgT2UKL*YzqmN4a8~vx$x59EyFzLr#tKa9%aRB
zQVvTr{|QWoR#Q|#x!yxlj7J`9qvCFSB>xr36;c1HoAr|49k?;#ygF=h8xep!+OsxG
zN3CDM3^*n_0erK6{NlKd#`HeuJ2O57^U~tZQHIFtkV(cs`V+*4Gw8mW&HN<8`^I4B
zxVQVpQ?&W9B4^mUaP?#K<s0;d@6ZsiH-7imt(x|?_>&^bXOfV2pU1q%S?mPObcqFg
zXX648az7{F@J!10Z0<|&0~v*{tu^VdOxmWD5_&`CUGxoa+Aqj0_OaF-X!l;C>E_(X
z7-#CWJkq=W${`R&xq+%&*|1a^juEdc`ujwwxTpC~2p${0v>ZO@7u6Q0d*mt8Wd4ue
zkvKC>I-V@{sXGRrIJeK>eX{l)!unM$5N=u~jSrpeFEe~m%HkK(949*d5+ju5xgPm`
znn5{7?^>S;`@0ydPu~slKRu^eQ6KdxWw1UGnMfpy3A{;w5tPxKX}DQF^AptN{4w{S
z+l-m(n{fNG?77lYZJodV3>2GV$&tFjOOrob-d;?9d`!Mr`?f7`Yb-db60pBSB*@@w
zl)bj)E3fof68?#)ACo`Sn(8s-UJ^R_+gl8k<Fl4izqXLRy#s^wkJtz~0zy3>7rIKB
z)3?ud(bUrxC8D3r9@f*6M+pfVv)R$>=oPQWO1E(xS10L69Df}C?3V72)0-Z=++Js{
zW;3*w3<^~;HxoZ7%`2SPc1yd4S6m0qDPty=;(yIFnK_vMb%e=<@aGjaER+kn<UrN3
zJ5Ww<NT3MdQQXFmivtyUR9`OW*Oa+x49>`LbO=iK-=Bah0&0WOfM}<&c-urKV%s(8
zzG<6Rb<+H~xmK8WvodvDUB6~)fyLZto#4Yd&C!Qls0itMvKwpVejIBp4P&(`%o6#q
zI|tro*x(!bZi(bb^N-oXmY%pw{>_VxA-X*!QJxji(HRQA+p@KX`)!xLKse3`BPi!3
z&m$Rz|24Sk8o2UVbDLUP1gP~<q@-q|8WCc$v3B<lYj<(WaoaezQ^+$Fs%`rRrpNwq
z)32#q-mqfV*X1b}u>H(Pp+4}}U{$6u>DmB!tvHxSB_{|-g>hr5YUO`>v)%K!ZTZMA
ze8hyr+>l7?l~wzqt)eBEnPuxIe(HE|xlKsIFPojMI`};h!+cUeAJsv{{;4xwBit2w
z;l>HImW0uSH#dufFw-M8=1-WJcgF$@Qe&lNmDjx_LVSg;4V^TaQ#ix*=9D|sZ(K7-
zBTwuuW^)i|#WkgQaRf&MS*0@^%0)C36sf_bAeDc{qx(&-0Ew8h^KXK!PuFTa#popj
zqOua+bA7QqthwVNBhmCm=+#%+uDGm<FR*ljtgFYQMWvDE$?yk8SAIP4AuL|m^$q*k
zeg=)N2&v0@W41K$0$@=62Re{w7biaH%8Y5E{^Fn6z4#<EmWSn<UG}ytIe<Gul-D+N
zNT1lG@Tlibzi%>yi{2y~JQ0KYB>uDie5XMgXd6d3r!saE9H3AUuJ~h@o{l~iZk%;m
z?%Es$=U+DEg3we?8X6+yJ)hv>iuriq#gCkmeuX*QMx`L76*AZAhr)sOH?0)AlD)kc
zCyM=m3)(-fmhSqDGoYlIW!3b8*_$EeN0~DNf)fze^DQFi_dng3(XzrMq%d#@_m(V^
zT~DwK4_1rQcWd6(ze~A_%0P(JPYrfi!HZP53r)4y;O;xdw}1HhAtpK-yI=V&#S%9t
zhplyGZ=9fuDYdP)Fk_F?D}9NN&62ih&#K#GoLnXLx3Xi27Y3A<9<~JAtwC=@w@AF0
z`LhwKMAQFhSe++1P{4N8L0DmC5~Q(fx8%&YJ?WA=1`;{hia+iLU3`c^un}2Bwh|t~
zm?3`Iu_xu%w}=a=tWRZ%B|;LO_D!I<8<>~eyJwS%q&)0*ny<5pW6-+zcJ4EI;X^eQ
zl{TT(`RPIB-X;HBh^1$O`x1;*El`pE)^qr^kIomS&EV7c&c^>hR2wS);7N?lGU2C&
z;E`Glwea@ykPE>Ym;Ik9_g_uS(eUtCIM-C?d>hN*)_I^}g|P&=hvyVP!Gi8!C<sw4
z(LPCdw^H$(!9JQ5mdCv9SN{yva-w)<D5rKp7M>z=F1}YJb^!O4Mfdsl2>g{8!lf7n
z>oXju!3~C(J$#CR70Z~k8EU}1ITeO+%RZ%2v3faxMUq(LWglFvT5q1MQ`bU=d!#f4
zD8+vxjxP@=jmMU-L8+2I3XN{DxH)bdUTXT`GI%kacv%wpE9a(;lz_%@{beXVGDlFJ
z4l}%2j2nBoQf2-!a^!_%+A|X8dU%Db0qmi`lKp2!Cs+^25pYd>+LKm+79?$O8pqNH
zWLvFn6;=b{;5dyHBHx-$KkN_>qe<s9LYIJ1(2dIudSC0!RTc0l@dx$<4Ldy6-b@%k
zoS98Lz<5m#e6M{!=@G7#_9iq~gW$b18N5rpwoaB3Uj9MSjDrSqL5H4CGVjyLmtz7o
zCsVUboGeB?n$6W;fwz7lyZOiD-Nv`!u;X2KK`29QN$m|6Glng)z5<+mlFGNOQ-3Bi
z%v?*{s&?T&&@6=J^Y^vq_w?T~BT;qz@$lq8N-uY^>yg_?RKwa$>9>hU&QB?WLKd|R
zhOW-qzn*Q^VBB4v3GIzG6%QCKlR9>y4!j#v%Tr&Tlll%<`uo-`g%47r`RHG3W<<@T
z8R<&iUs&f3MqVes#+QGkN}s7eY!n;NTt%|iRwoQK;+LC$%)GVKZjRu#ukP6`Zk6CA
zTkBBmg%8K{Llvd7T*vf}tO;$MYuRtI0f#A{i}M&?B3XzfExK-w31&qN)UbwRaCh(|
zp;>dTkYikCMOrLwj_}50d>PI>sS_Fo9C?|7!q;;5V(&w97Q4}Yo1+plUqp=!5G40O
z316PKM2??@y{s(q>ON5!-Q|0j`4(5r$EVz<U1nYrmsk`g$x2S-T<}7a@YnlFXFpd3
zK2!!8rLdlrKCn*$3)8MW%Q`I&a;Mzyi%~L6fZc$?aLv*9I|aNB*urq8ts6+5V6VI2
zG82ah4C5M*W3!?aG{UoVY81H&`qhL!sA;!fYIiFMbKofL04&y?d3TDRWc{Rllx=%C
zIVUFaAIPRfXU}+I{gUPjLG7-hDE7i{4>M;L)uV|?A^4wEplrCtx_m4W@1ZjfUT_HH
z&JfAP8z@Rry?7tJd~PJ#!xXBsmMzL>Kb6UuW267h|9+71O5(Z2n)C-Z=wDF<ZjqMB
zwWG`fcS8**9WZouRfN#a)3~48;6untfaNb&uj^ffpJkp84`8~+L!c@Sm8xH(YYSi6
z0!|hGTcI$v?e{C>$J%7o;?iIcFP#{*Yj5VYLGvdVCo)Hn#e>tPS_)0`LX|utIu?dx
zw+)m~j@tJI<)LoDv4&gcM;YgZyi+WlP{Sob0T(o;sWRC$@x?T3DHIF}B-@u@yfV}R
zU%H~4%btkuL(is>^wPiAJuM9~z9Vr?a_30$dvXL_^6-;kbr<H|jlI-2smqefvzm}x
z@7}Sq{k5K%(-(XITsuXDKWU*TR6`x-LnyEYKw^iY7n6?EIfr(~&G_v~@84INvMRl3
zR+@MA82rlFX8owjphXb`Kz7(PLC<lZlX4aXK%2^=2&&RqJu(JO&@xv6+h8nFcXd+3
z)h_o;BI>Xc7Lw9(n&rd#m&Q>fzLPvGRA$ynd2sHH>!oiPPV1h40R28b9cn?Fv*&3l
zoCHN+!aVyF^`zGwQ{-PUR`W?gIA@VoM(yFU{+uz$1lWwI;9cQR=+2uq=kr4)zOEe}
zI947r?>+f2r)=;58)9|31L$OOp*$;~+RL_dH3V5keS;?0AA;6f%A~PyziyCkB<fav
zGP()!pUc%*m_Nwq1|OPc6^JHejGhL!)BFHXS-EIu&VY@CrG-fwZZ~mqETNf|Y&C&;
zmewU7%ueBFPN=hUaSsZ13eQ~b_iw$|7vou|GX5AVIy_$a<1?<dM5#rR6LJ1f!{_Zt
z!j?y9B=ZKoecJeH?B=s5k<TMcM-uI$s6T@il1cA6T5)r3GfV?heN~)Y8%(rj_#2*b
zeg;k)XaYqqdfsSZ5j0<JC4<-Kzf-)bJ3Zm}eA-l2-*x@+i48}p>zLv;6=klaFKNG3
z0yru~v=ndYXa?RyN;k0~wlZf0bR<z4gPL5y`?$b+GcDNz3iN_Oyb@HjFSE1+%F!^!
z8mO7{o5sK~VX)g42fhEg{3^^J9GA^gJcEm>PO`3_$(*TEijqYL;BI_QUY)ZtE6lc3
z569d}D2`Nrym2Ee7xz?l<H%LZmv3h;@IR18n9U8>rIhpQ!~%CKzw0NmT}IuNAHGcg
zHu=s{{1H<PY={5JR`GttyAHx(rMh$%E%a+ja^&><8nv_6!t?gp<(WMd&bKe4I&Qqh
zWJdNYv#~r`Y5f%A;-7<lhs0PS*o3UFafqoM*eG1!76%dj9Z)@;83r+h+(KX2;0)h4
z);hP^webCq)V}*-U%{t+Udj^3ZfTeToG|XYwE4|KnNpa=DRn&38v2SRodrI$HWjs}
zF$Ak2M@$8AO2Ei}*9M6M7jq_{My_WgRBZ0`oEoH8OBRMq)qJf&+|s(BBcHmOxQh}n
zh=$M_QCgtD+XhkKDfLJ*x1<bW%vW+&jD-{%CU<F))%r3If)XGzavNt|=oa7UPtRZC
zh|&NTJT^@7)GYauPMqm~@e+IH{UwD1E6}gsbV*mjdomWYLdxI#NMWO@;jE9#pQ(+3
z6byb+WSN`BxX!R^43d=~y!*&0>sxEQ;MKponrb5p5-IUNT&|we8LHApAC-HG%~2@)
zafAEz<`bf${0DA+1Zv=*MY;9Kx}x~5{?h4Fv4YQ=h?kIizZ7@->V5sKjQk&d8BM&D
zccl{zr?|QqceK=s%iZ`FVU&9AL2cJ34l4O8Ux!u%*X@{X`QO`km~=CmK_3_a)H95z
z&fb!jy3Z}oGGDfdD)_2X!GX_Y^0+2)6>w_IY&Cv9J7&$CZ44P;zh!|eOutb)SD}){
zn$ixJQaSpK)Zn@!G#;z?R0BIG<eudl<~53q`Xv4F{$6YI0FRPZ5blc&VobGo=B&1U
zM0M^2aG>H?os9=X@5yYLNEXRt#wMZT3S;r$<kXaHsWsA(4=9j&PAF|&U_tIawB{)W
z6#PfSwpU|?U~T!Q7M!Bb#cLaLfA29%r<Hvq71ZPG5w0S%A(!jEw0dO9RPlB2Ppsp^
z#l{XN#4|}0@E&iimgJPYqao7o=IA3-v@~52W_h;$<=FYiX)Db_Ta%)AbKnh`^iMYE
zk3!Lhfon6>j%9`z<+&<xm+juc*gz@-9S4i!_`HZ;nq>Gj383yj{V8{u>)eXtlE~wy
z#qku=(8ZICQ-2H;K?e6J&%c**b0f0s>0?h}A@~>D^$GI{tE&Wu<oXM#;Y%&8SMSzX
zUGNITd1&exoX<m1gm+P_mzAcH3})_aMx&|}9n=JlY&&3VGE`T1?^@bMiS@p{H;XoC
ztBIa1D%1}b<J+)4e_ZqPExpYJvD7X)9Y(JRW<xvWm0o)hsW+*>M5&PDe#RR9;PaWT
zEO^jxZXHq`xuuksV$wmCanta(Mv%|ojt=?g+W{iM*LrtH5JE$@d>uWAF>_94{BARK
zm`Xdjefy}RA?n(v>m%<9suN78^lE#;6*_r5Sd)WbL#rK|q(>ME+Q#O|!3l`>dOZ1C
zqu82?cNl;8@6vY0V*|BIVLx1kRTjEF_Gfdfh?Bp9xM|(7p7|cZU-e(T)Di9voBL5X
zXZKN~603jOS}v(kS4_W$x=;Po>JaMnC7j}5Gk7%&xim1-P&%;D_Xl5hA5+eRXdE&O
zz~XJkUr&D3sp;St-l7Uv4cZoZo;g67@XJe+4*G)#T)t~hfR#-1E847#Fw?H_zJ2yq
zUTKi1Xqa^j@IgJ{Pa>uR8bq{g5TW$>w9Q}48NR7wlA2M8JU2*zh64er-!frvTzJGT
zQgMsIkz03oOgV?!Nl+`iXiG&PUhVp56nhOo2Qz0?GFLVPUg)|pDTj#YwFW6EJ(lrx
z-B^>Qwb_|oq=??xB8fx$kkZeMq(t_DzlJ6xA7*`GdG!<P)_N4?`y(rD=j|Wa6f0B-
z8?8CLM^1dtqH)m&5^*Ythn*ExfH&k%l_&)BKD(V4>o&w*Kv$rgwNa49_y%Ad&dNn?
z#7_S0teNECU9JUf(6VLDl=Upu&51jVRW&tfE#!S=4I#JQf5kuD?(b61>)3u(;$+zl
zo#UUF0}B8uroNUA8O)~_p#zsbQ>q9~(GfL64y27fdLKKT7el`(TES87rv4faQxIv`
z3KlGwtLmq(>f&ldeqt{ti(k#GIb>t#TFd__L;=-o_8Wwop{t{-<$ph7!d@+ow~fA8
zRTGnwtWp}wQ+lGw9<2lPRDsPQSeI>q-BnY@XRzkvKEJ(@Ydw<cxvv|8Evh6!3Pkhw
zeTO)V8+71K@n4jR>z8oX2xa4y>&=Yu3E4pi1%d;6ICe0mkbw!`_UW901w(k4y<Ft3
zO+~ea)ZH;f1-uSLF}ygImN-`(4C?8%&@;?q57t3QlI6e5FldGo2>t4t?gN$@#<J7n
zK_WsQCE~Nzh8B)RoKCIb{XS4tY<!CaeA(o8s)18lqK`9NFWb^6%4y$2UzuU1_?I;G
z!>j8Z7xx@e;PKF(@yExMlz#5z@nGWvToa^*u<Idvi!H|3C-W$N(rs_aIcc~`&&mMy
zPFbCO!giP3jR=!ug44f>S>POPzj7n<pRk?DJiMIfVKp&*3!83Ja7)VV_Owo;M{Ki_
zVYm$&F%a$5ddEt-Zak(GBZrb!&>RQcJqZI6c_0UnqO0W+lEIkqiMUQV^ZN-vH2RDU
z!wem;KCm`fixV3=vlE;asWA>^*i*CiJ0USakRno`Of-?eCX*|A;XRk?@~)>{U>sn*
zlqyEo__}KkL(n+B!l|e$8Hk4~|MKa<>u1;DV@ByEsy|%hwl|GzpWf=aK^%vDQ~Bhb
z{SniR=V<()!sC6*Wl8k#NhfHUEFv^!l(vH(WLfjNSK{+Phk*U7q-L$3+kY2%7pHTb
zDsN<D%Vv}uWZjFkG_U|MF%dP0Am><@e^Df+7kMSR&T8zxXlwBtW#gHL$JGorL{L{v
zlR$Qir(LipSWdzTwcht7#!I2X^dxeaR+Z+Vgr4-tRjfhxvQ|?|o9I)Y1LSyZn^qL#
z4gYQa!W~py)@s#naAYGox?Cl0whxaC8A-s*?(<$z<sIm7wNk3l*-`yq^7m^nPZhi5
zJ>CKQ9f^_K6-;}RTz`I&GCZuVg0(lD_b((jD3@+EIEt$F(F*hKY*F1k5Stb&eFCGh
zlG(1H!fs+|u=TwroSc_<cdn~SlX!?%<Cee^zdK1PwiH~IratF`6=1Rjz4VCWSI}+~
zbZ&+a+~@BJ4?41_s?Kc{hQ69Bh&(6zyE6Nvu{6cNSVcd~lV5Fh`t=wlbAHZv)_a=u
zS)w{YQif#JA2d7ExrGpNT)iqfQhDki^><y}b+{q+=$Z5Vkaw{qx6ygsx$j;)kzA=r
znacD;&2PjOEs9!wXVcJ=(TMMgp{L~QGyeJ_&jXT`=p0bWsl3@*$1`R2sq(i<@yB_=
ztsLIkYM7jJ+QRp0vrR{@nP*67)W9|6$DJTWM?+_hUvrv<pCcwc5)dKxOsNL{R;h%v
z8IfzH`Dy^|nWdt*w8{tH20(xK=Hm`+b*~epaXbe1Yk`+D$?Tlb-xz0|F|{+LSe&U1
z<l(O!QBfO`zMF<Mp_ysl{#FR<jkF^N`SNCs30q*y4GN<d8FNO7Llqgl24E)diN*#o
zZiFmUZF=rgs%=7dIX)&6%g3-**D)sVfO&VpwhABI5UM-96F5VWy`=-*$!S8hl6M70
zIBl9YiOZj^cF5Sfz2XY9l_gyM^sTZwBN=S9hJMH>aJy&BSz*h-y78V>-<@m!_(vyb
z4GJK-;tt@z$e%ObXtd=O63f&Isi_z{cm3>NUe__F(*TK+RI}McNvU`D08I53y!sKd
z(+RLOI6aZfR~2$ZR$e<gv`LUH3@<)%E6Tjy1*1BY4`rct+IF=qOPam%ZW=GRc(~2o
zRKcT?D9FKS>K(|ZsEz5~R_pj6)_5N7T{~hnqy}e-UCD)lT#8T03bfYiX*b2W%`}bw
z5L}wq0M~;m{CJe<v$xt&|J})dqUw}G<`R>wVn<JW6g5q>UH%gn7k1y6o4M{QLh}<}
zkI67#9QZLEX~1&H*CIfcBKFmI41n8ww+u+BfjAH4lw$h>dOuZKeIrF6V?D)+9&c%R
zC>0{M<r8(-08I<`Z{uUOeV5$Qyc7i>qm6VDvbHY~d{1l^y(`n6c_Ba-Czm5$^<K!U
zL;nt307dLibKqRU&vKtd_{e<=4rkj$GA0h9e7DI`jITsZ-}W-{87eN9&Tvb-xc0q9
zS(f(PU9e@7J7#Wd>eeJznR3}E@kmd=Yn(cF%W5^G4o?Ugxb3nc>(Je{!PBYpD-JV}
zlzaL_op9<YNxktOD8+W^lEX}o2h=d{V^g1a*kYcfi<{a=IL1vMn@cqIw(PfmS_V*M
znlgE*PG^L447hLGzp3oLAeRbMYX78lx#Pw``Kir;JhXZTL#s{C)|lVIj+|%qqlJD;
z_Hhe<V#zf|uNN#0(F#f{INBxNHq1ZIk=YRw%Jh}Q^bnUf$;cDdZ)-Jb&8oK8MRI?0
zOFm9kzZCjotsrGpRNPnu4608-#^Nkb#Dj#VhKI)xq0uC!xX%+p`evmuGtHGt_bwV>
z-uQD7Q&3WcNA3l`PCuIp>KdKNzA4WH?6@!8(t1<1*j00=Q7OqO0=$AhKRM&#w07YN
zdu!>e$WOngrh1U4SXPdXkeuo$P7oDwj*SSkx#E2DoA-QfQT<xix2y*4KMJ>f2-|<#
z^J=4^_nU?6OYmWwnonV>L?@7^?<(g*kQmq53?3<51B|Mgy*mutIx&-sLI*hZx+d0v
z6B6uxZBu%~x7N%#C%-7B#V$2FH0@RI5S-=hq5Jx-?<jf-zft_XW)PCYcQ!-x@=!O?
z;aydQNTm~(o#Lp`?j=<cz2^K4cX+wQN0UGH3zU~1CA<!E`tFAWh(WUWo|Yd^p_zJh
zcBTLnZBg>u_sxepQ@kJR)piLr79@&@6W^%P&3N3E*7(I|kL0k9t$xDar%jJgUgI0X
z$U_{-{}k*RU(WYwjd*ku+;z(nj?)<4vN<N(K5IzN=#d9-=vSv^&y^4QV(rJG^Q%U+
z@bk8Z7tJ4-nCzqOs6(DlAA0fw>OUQ-3)h2Sd>2*b`pB1dN!taH=G8-&cNKZGoGJW8
zH}n#hfRdQIaMDY1^u;y3vNvJtDjM1p$~>M`sbO9BCbyjEod@=Y*L`!dRe=-AU2Wvk
zv;ZImmrZzl;7in&tqBmUobl&djLxY`7*t+ZUqTD@8<L?jq954@qAt(Zk<U<VD0(+8
zRNROENwCScWCbym*WG86@2-+?N0%adIyWKy*Ndco8dH^xce>zegOck$W@r%uOHp75
zP<cU$tZ-Crdh4?z8yU&B8Z^<X#u=&sge}9NjAF!KS$Zp>!BWtr_3EKltH0{>!+aZ)
zKegq|RW&FOx6HR7#t7k!>C{>hGY;my>&dSkAhPPVdeyIc6_3iqd#sa96BUT9VmTdW
z`aKJqp1T+%J6y_^8m?FDY*>9%*WwHl%l#IAfEq;$|1lG&0aqe!xpyN<JQ}QC{jbbE
zd6QId=>7S$zKs7+T=-#9jA%;@UWxe*$86o3aHYEV(ewoY&Kt#wjc*CJ)B(N*y(atP
zIy}+9_C`@1Ur^G&Pgq(K=AsSO@s_8MF@M<;aHeY10Z?G`0%A_;koc2PZ_!tAUmb4j
zh6@xJuwWnGe*esaTG2S-WL8?C?nm}`)tH-~#cfv_B#q-{V$4qEba(BMJG5p))AHO}
zvQvl-X7IGiVqgV=7POadA&~6vhcfcNsBA*-U;6RzW5LXiRBgw$@;~j(Ol~JN&768i
z#_Xg<9>ixSXPZG}3@t!rgMTyOUqpVL9_q5Xm9&!7xPcP!>D#ceeYEt@0lhL1*ncZ)
z{2xW<;ZODZzwu*4WapH<*Eu5V$g#KM*vCFbabz5lnNdVWoH+J6j(v{f;20q*B{}xW
zPWT8BqGVJW-{1ND4e$5kzF+tCx~}UPdPdOxv6+Zbwm7Z41O0=J&g*k`B8-0acUdX<
z@HfY)Y0JfdE@O#gU7?^p4W@Xa$K)n+Ml=rWl%dVKW@K){58mL>1J~y>RV~p(<|Ymk
zbAw*tfHy5#@w^F!YqTIBU~Nfntk^woHd|y{p8#+!j>vJZ%w0Fu;fMW|AGv4n)ZzQG
zh)dNZjkJ<h5=d=5oF4}$y_1YD2z;CV$^Qf%t@!g#ivK@G`07+&yvxaA8_N{BT3Juo
z`T|qQdutb0U=5|gK6izcow$4PbNV({e|uBJC+{-|B=7(91PP~m21Y4xI9_^~Ekf&x
zg_Qb!PMI<>%Bi4PJtW$nDw?5v_25t5339wTO^M!vRcc544Y(CSpsPUM*A9oVAasT`
zv>o<q0xl5UPGJ0~oFl*5LCW`k1}h~}a3~XTgM%HQdZwndr_a5M@`)gIe!U(RuLi&n
zwngSSbOPukun!lKg=xKSDp_{F9Nx5~R%JAckIEADhly%35%Ju2#???Xc(Q-xJ>a-r
z=Kl}0jQ_=wlD$9(&tpfCWQ6dvRcSIc3QqzB%0Ni>ogoXFOtztmE$AI`$py9o_^IuU
z*vH-r=aRMl^Qc@mn9{pe`>$^A=8RZ-SG`|=w#v1fcD&2m+5AMu64=g0h@w#!wshtA
z+1;#I04L?<Y&I($GfuZ`FX8LQf9>7FI{Ud?ek3Hk;c_8Ey2Fnq?98F#WSIA2+NqND
z5`UdY0YZfn+5~Cc8xFhcg^V?w3sC{@(emNVKZ5qf0dg8ZHL&l!dX{FZxpw!gvh$+Y
zB8tMtld%cqVI!;8ZJFDD-vCfKD{Z6(a3s_+yUY9XwvasrvoN&6GIo(x$r{BgA*JDs
z+jit~4##E2RK~G5Q}jaUDUu)^;}a(T6J^ISGo{6%^~yrjG{l@6#7)wsUG}pjM3p<B
zK&D6;5MiM8jnERrpVZy%vp6_9uvyd>-ex>Zfq^&KkjltvK0BI!eb^n-^p>e3JwMft
zDHRI5$5T=oU1bhQYzX(e`MYCl6zNxWZVl&7S+miRuiMuAD<k!9Jb63>C2qmUp!K)D
z44Wj2zv&nB=)=qQ19$g7l(I*iu%#sh+?FM`5tQ}tUo&~SNj?#bD#=OZx0bKXdnd{I
z%xyVFDzY{)@6fG=$vlHKWqNX@-Tzsx(ad}Wd{K+vCPX)3P6*YEtY&1ypm^8;bmTX6
zmI6YT)sCuZ8e)JZ0tE}uj<pvgp{GpW83%=0Lh>L+R<dv-HXH{CfD%G*>tBu|?`zF4
zCe>cjerU^x(e#-g&!f6?tK1Ny$7}9F-UF+sx5$iyYa58ldbjLz?hnFfd{%c?<GyZc
zeh~szv`7+#hag#N=A$<puDjDYDx^IfR8c*!&kFWWw+By8P9m+Jq#X_GrEsqbu~z;M
z@DQ^z_{hvu^rJ|biHkkeJ9AieeTQVU=iw4wboIH^#jMf@wCQbzH}}rH+wgMBAfr83
z*lArKx7@ZUQR;C-213*978}bM>3;y03wuLvv!raD06~mMKP1(Fzdnqk3pnj6px<P+
z>#Q~!D#sQgpMz7a?Fh<P*w$$3k<~FYfrte;dyNb;k&<whi66+}R56iMf)FMZ5CGy$
zo<>`GSK*AKsggWaMhff_HQG*Lko3H!$HP6hej9dw3l#^hrT%0?A3?Lm;hf{MovN3F
zW+U=s60=19qFtV9W72~H*e6T1I90?v^;3FS2aXjBAhn7XOqN@e5NpaQ%di`MRBa}@
zF`Xm7<YI;)aVOvB?9dAEX&>w}&L!^>>?d2*c30Ew?W99b_lZ016zu?Kl`XO^jdaJP
zU}I_v&xYdUr~Sy>ss(>Q|L-rbMY|d@O(R3kHwjF;*Z`KXZy-P-Rx=!C!<%q^qM$U1
zJ=^Lo!xQf51uYe!TmM!JUfmXvUe&u=ap30l>A^^SV@?$HjpgEO)>bF7aP(0hJ@@-W
z^!q=TCxcNQcT1jN|AH?c`h;e4$eXlqTU=WlHi9wp58K360xH9;|0&ARZD2c8UML&V
zM+Dtj==vbhk)(=~J^_fF_9g;}6WyjfpxQhEi7-tE7cKL1T5V%>L)iSC&B}Jw`+a<j
zApyL~sR1<_Lji0?rzBHiaTKBdWll=}5U9x5w3r!}(rszM7n16G@YPXKFoJ|3cHg{X
zGHhr4C<K(S)(r)M^<6>&)UWX?^wQxSeIBI$1(8B^Xdyda!TIq2fYMXts5JOX^68-~
z)OEQ-u)1E+`f>wnGD@^eeJ~xcaZZqB5FN`Nk-860oEAEENcXeYh)~cb;X{8d1xEwN
z?YI0K^-_Q%Ytk-yHzFN?RLAU_3YI<hT3b(n%M%<1DxD91A2kiEbq<iAo86OK*q2Kz
zTUzu*0kQQD#=p0URCf^0B&uh2489^jv_`FBj-h~1ONKo&8uc$p8<F(IpKg61k3o&i
z{X!sQJ5pGVtL$VF=+DeHyQ@>5{nRv6O!sW={g70;bzt80=bGP|Cb6%dhgFgw#Vv5k
zlgmxK(@yv%%NIvxq4w`#=ed7Y=Q0|P0yIuchwc3ukb;lPZ7FLuvAM&JFezVvh7NSk
z^QE6a&L^AjU9gwDeLlYMi3i$Bt+Xwp0vYfsuFKi$6w>a+CjtEAGl;e1Xg|4C)krw&
z?ny!wTF$+EV}4bxMI`}zJzt;$b?ch3OL;#1dj92T-&2Bw(F(<73}ypHG6ridtns&l
zjMgi?PV6nyC8C$;Ypvy>g8kWcuBs8#+Hz<kx}gU4e8SCxMV2@D&XT-oD=GyBsvgG3
z`}hu4Q+Nio=T#H}geSJ0z08p%1vj4_J&Sd|RoC+SMv2});8+vI;V7w3HP2eIUwcO9
z0EIugfiWTuP*_`)Ghh#W=C5DBcj@c<yBcFjBJ$H6+f&<LGOwiTP6q_%Bno9vAMJeR
zBoXiI_waV!YEbAUq%0|p&q3;ON!Dv@cxaxp(${;wZKdN~h8bGi^A6k`;ZyRjN)hHY
zxz(ZfRGj#k>PM6UT_t&k2iEHCro)|uGd^|oDO<*a(}AYf*Ne9@<~}PD^m8UP1g3Th
zg(!#ydT^d})qk1df5<Jc`^^x+$VNgSk4)_;uvE?P>fUT__FJKMFvdfD{z?XZ@^w-N
zKpWMk1DjT(Q{gBtP@xs#wsYg(gEJ7V14~Gx5F)nVLm}0C#@_gO`?PK4=w$flS{av9
z`dt7OChuQt5xRi<T91+ahOHP>QLMo|?*kcm&99Pj=kmTfBk;`jK1R@8qFv~TGvsmF
zUez}mR;<>3NS?vlSTcla-Ldc#+A1gkw;xzSYhaFBew%77kT%6JV`eo*-Yz3Uab!v8
z4V^t<BLed$(pDBlO-3r3%|HH<Fu>9e*C27$h^AagbR41uhm>6V&zf#^ljo3~49pr7
zxh$b2ldG)B;~DmClTnfwV+>dyq+(d4mJGFKhhHY6gz>XN4w-<;6U4Y7&4#;Pg{wlV
z@6aRxm9sbh(sZzacu5}TV8IQQhAxsu<!MurlJ-sEfH9reSrCuhC`AW6cX%sjDE_O{
zg|oWMqx=t={zG3@7Zvb9xtIUhz4`FrTO;Rni~j+5(&j6QJ7Rdj;cqxz-jfMs)6ulo
z6yb!-gr_L7Xz;7`KhEFp9eeSgI|88k7n}<`(&j7`2G3tvU7TO6b@WWe{_SyKd%O6e
z?o6q+UBD$m4IiE~d1oqLZFYxPnA9~&WG)B08q$Yj?6a!KeE)^O2*XgI7)Y{DX{$rL
zd^8VcLsMqc=4QL}u2VnlKQVYWfAy!c>Gdt;<`&!uOJh5kZlm8cKG`YB<kRQH$`Ebh
z-$3G*oGFa@MNHW-_HU^5g6GP>&9(ufF*A)r_>WxL#|(4XCD-DUunS%6j5`E``q`fT
zm0rC4LTi65<Ng-KMsSnN^G*&$+a$)HnP<;5uEo^)L9B99weDTmn}KPewS?wL94eAl
zEn{?9vQ~$zu`#1eB0&UcEDxZ46a%?oF&o8HZAs`*&<rxPK~DnK+%I{Wq9vdLx5i$4
zYc!u{!;Km%&Tn-LxGlHPc!R$AZ>U<B{?<73!b_=dV4qU~Wa0YYOXD2Oc+Z?mSPg`$
z&Y=>~S#T(G#1zw1JFvh0@PRWRRagh)g!JGj{XEg|EcAYOrx;mVplEzqOK3qh!ooO6
zxxfBPNI(_W>#91TUH-r5oEt~lK&B7sMMAAj1yk3$_8(aukiJ!Gq`E<*NEuE)BddU(
zHIl;nlwX6Kr<4!M(9t}0+~<hhwcyuq>XHAmU=?{ShgOS2-M}b@8@l$NAKG2NliQ>H
z!es)XqV$#_3CCMrG23f52Ghc(=5ug>vr`UXCQ4k~c@{Ir`d7-ej`eq6)0_=L9Ql4k
zB_Tm9jbEC4(|{_BPKMe{X*AUjZoZ2%3ETX~8}V^{r{IOGyMRomqt2xexTgD8B4W0c
zE<v={4$`K)hntj%tjD}%Z#cJyw_1tH_cW=5Mwxi3ILWQ8@3Y7(SNZ;Ea0>Sr<vXFX
z-}6vx?{*>yfs85~g5L)e@l^WdXPi^fM#OryLIn?VEun%|rQQr%N^ngA=oNg2%y(<q
zv=f=(R;5vInL7N;e;94V(`;b9g9^C>4RuJ*CM4Rm0%4Qcy$S64HQ~M;Zu}Nuk$k*o
zK(-ahy+CzzF=D<QiIT|kNkJcgiy4l7^~q=SW_ku}5`%+Yh=2kejCk@O1oQ038sFFj
z&q3t2%8+u7V-{1#vmOcfL~~egwb!A0Eg>{`Yo$xs;zxR!T1D5??J9z&3#Yw<Pvpvc
z^Szon9<B#sE1cH<1I*^Wb)S<-k?PI54x%lQDu|)KZ^^9*KRSQl^G`m8<JpzCiLQHH
z`^rmd0i6~Pz1VWdt~m+X-$Jbk_pP)r9CQ97$9LyDV=Gl(Y+|mpR&VyUjLn$Ii2t5&
z`mtx*vn>?v!7enKW4avVm?DwZYK{h}ExiFci*p*PpBQm~7YB95C~pYU6z{1tEITGw
z{gf)#XJ}y<bo|wz!H1yy8lIj7#AkXN#}Fji<N~S_pNh2G(k*7p(v7K36V3Z8nf+vm
zIy6rSj(z}B5b8E(3KU6B2(`*S_JD0Y%&8prO&1jY{8{jM-isa5P{(wyrn>RI1{tu0
zSrhtVH(T#fv3_&5mNejaGYy6w`E13tTwj|1amW6w;aNY)=1xO5<8XNg!a6Xwb1>5G
zr-==`ShVmtL@Fz_;c{RDN4U4TanefucN$;V#NGSgTSM{=z@YKd{HdG{3rbXukMQhk
z3s4p*EGt169%ENy^A;M#I8Q0tR3@5@Q=KDsi8LIH9oP~qt>pp<HAkZ(a1ZkY@Qdwc
z5-*6|nO4tOZ0shZYEp73a6t^s;fABx5c=m6aF(bxs)GIuE7G!E(mqf<fJ=j_Hm_>z
z8~Et`!z18iuJ)Wq@$H1MrjO+&H$&trC+Jo~v?57KhcJQ_&q>~Fj|Z!CCL3j;OB8;h
z%D+<3TlVuVxu;l+Ec|eZd+Q4H+O;dW_2H+@qU^e^E-#Jm3*v3>W&Q;%I}3#b&YRJB
zf3Paw{^Sh!L20nMCil&Ed+C8F`{Y;HNoe#d=TRP=P)W=h0z%*N7(?QD9UbuGac|ep
zpTX$<Y`M+q($)^g2<6qNI@gn-fYDVuC%U{0-y)GSws*|Uef+01FGQ8<uq=*6q~k`_
z!$_iRuK6B9xo^6Pz(nxov02R1KG7E$3Fz1;&*mr?j~-jq>9J&%3;ikZAIG7pz<(^6
z(T^stT6_Ir`~29*Lx}upU`%;#4DlGeS7cPiFHl%>e&<f-(Ek9Nk{#jptM|R*Pp*za
z8YSngvcFCvAL~rI%p^zds(l*bP}S#{rf8^}byu=T>SbE-pVrD)Hcz4TqbCv4&w7d?
zRf3eZ&vsV|(pV-L#HXbkf@0o4q{X{M3ix&OE4?yyXK~}@%2YQ0Rz|4Tf!q0_^c@bf
z+3o}QQ2W9}rgzBm8089zMfHa}YG2<yE)&#u8)>Ms*tLZ1S#~}nksIJ6sYjE64{cN{
zYuyJWw}cj*ds`cvoaOHgdv;TVm3haXTmtHH8-)jNn*bYx*bW<QfqAP-5S`2&Z?vs!
zNtdI=!_>|Y&NV&a12Y%=w5dm~$F|Vi?41zNa*zwX`K<^(AVWlLLdr4?H*qbh#S>aD
zbvn)Y<~&XXGK`)nO=;p6&T+bIdR?Xzw9`|!cU7PiYnR;k*W`7C8JVJJkf+IcbG==S
z^mkE59uypK)wzD2hXZUot#|yO<eYZfw^#n1-_?Ggb5Oy6IZW#59*6j9M+(hXX3$~f
zjrIFD)sz6oct=2K^fcTup7%1RupBc6pgJuKimpvyW*?zy2sw&#9O?LPpVo7I8v5wr
zi|aiEK&I22pWmuJ3`dOd;7_byG7C9NRry2=IeWhslRsfwNS8E0L<8+G+g(uIK;9Lv
z-bAWd_(LI<LLq1Arl|sm@WlC|!ZQ@V9=wyc>V*~?sKH?+Hhm$>sJ8uO)~LHc$NBP;
z$YyWl9&OF^6QIbUYdY_AyU)x<a9Ot67Vk5$X(ZHqN8~Obvd%M7__LjCaboQ#ku!0z
z9hR=l*7en{eVB+$&&%O{RE{2$61eZ}w7pD3PL>Xa5bJ!?KqeL-UPYPjUwI!uYOW>;
zhncym>=K_2#!xN6J{{_{U%596DGXc~{f_Df#*omTR7tK&(se=31({BvxIN6wfDM90
zKKq#YXx8rU7B$S<`DNywUL<}h+&U#{B1uCq#Yj3hq8u!KPtrt|G2;Ty(J|$ANp*-X
zo0uBcp!t}t@t-8+7+vi@bJ=Y&D}?Sm2dnAinMi&U$dBD90Bm#HjeLSHimKSFMqbJ4
z@Elc|;!g+iiONCNOk7>I>yM7ah1)E>)?X-zg{DsG*o?`mBt^>jcZKB+|1h!0IHzp2
zCU?qDloz;FHP&|Asy$_2FOX1|S36U?A#dM??I4b{3_&mE6;>k~w<o#YT0t(coYoKl
z!hh>DDjE8OX?h#Y=2|`vs3DVU+}Y;RN;FRpa<mRz9CTmB;KD(rJk$-}rym0Dd%J-s
zq#Ae{b3d_ib!@E6lo{10P6l~X>z&PhyoOIT4+NNP3Ny<z>T$Q8lh~h-1|2#@A;{05
zw|B_hp7!|7`qz}ug`e1-+m}CSJ5;x+NoTN~I9=;$*;SYHHn8xh+q@g2(d{1V`1F4O
zJFMnn*nX7Pxk=g(uf5m*04@u%Xi+h`*tsEfR(~e3U%%(pi<9lXXUtpc20f_Do$+s0
zq3=4t5XFXtmV`~n_3G_pCMy4^C^2{-x`EVPX7$WJ++HDyf@R?(jH>Zy(YRo-2s#H$
zwm_ls`!k{tE+-d7pFN+SERmgsOby^BAe@5U@yAYIOWw*H=x%DVL4Fzd!xR~1_+#>B
zb7fPK+KoF~O=uH-vTu$_Z}bn^qPV)f03mIw!2sOyRp^zRsj3PMR??6ZMN-Jso=DA#
zm&{BrlN*ojOGl!nZZ<BiZ;<UeR~p)Fu|`~hE2;Inc?!ar!kQO>mWmt47g3bo0fn1-
z%CS++&g@PH1?B4M&k(!HFZJW^A_u3m;sQ&#|CmN@ERe-jG5Y3TU22G!FdLywJa3MJ
z_%c;&Cr^NGW(S+}jdq;N34>krVQM8(4jEb}R9VUYAd*akRluq4DczT(99ZB$>u2}t
z6tABp2Cdmj8;i8EIPiZgV`Sf%iAP76koXctXLGXoxrYpkR+^*P5^zqzBZGTj-JNF`
zU$~F1{xqevgm5yx#v}Z!`vQT|kyBunBmxACM05g%Nqmz4ooJwhsAv8loI=&s4alA8
z?K-gNDNbRGB%t#2xm)RLWIMz&gXj{VwclfEmB*MaGf*sZgnY9FVm+Eml-7QjL5K~v
z&F--n4}H1w&{^2wlbvKW$4vLASm3iK!+6{>&fpU6jVR;aEw_$v`Q?ebZ-nB&lRbYn
za6$QKY8Z{}?HDZ0V~$^hYp$*!W(O_bCQ{AW{@DjJ^R5Qdz|v8ok&h<2Zc-7shiz}U
z?f6_ewX)4+i_L@4`K?JKdq}FwWHQg;FHycjgw~xW6zi3gVU{bAvg<{m)%_-tv|#Vt
zom<zxk#6Ji-YFs57Ve?m2R}PBUP}p8ZRL?0aEyWmIlHSvKXuE86gN@~o;$|vRjE|D
zH-SB%FUFh~t4)mG+gpZC8kjn#-Oy`SBctp}xbEB3HLDJBut(a(x(w7WZWVoCDu*6;
zTpkQLu%_Cww4c`q{o@6s#IdVkl`ReQM`-b+t5gl2(>;e|;a|jkP{JoME*s0GD*v37
z4xI97lcjwJ79($lLPBqxbS9vQf8b2plFS`<<<&Gm<cCx@gu$M5kDq)3yG8uR9?xUW
zqVK;KAGC7(Mi^{6tHn#Y-i+88^ouv6_{Vgr)SJ3}5-FIULd7m1wUkeUHfvk1BPxou
zZZ^rotf;}Vf<@9!)|K-z_d0B@)pUV{25-hT%+D__(cgJcUU1tj{U+PEsIt&b^&P95
zO){cRt-SYN;vZ-bay=*+xX>Z_l_4AUP{xw47WgF}*@-4>%<Q(eH!(9!qC<J#&2+Ux
ztg)og%nsrI<jFEW<feDd(<qtWM9JR^6edH=N5atu$l#v$pw!bh4TkuC5}*Gtq-c_P
z`|ZhTncECR;|Z&KmzT$^w_uYd5{v02&l}6+bB_`6+e*U6?msi%b?BwwyBY*y3ty)g
z7LH?e0%cpw!u_aICWD~`0Bk?%TjK{|-IM9m+imItZ*%44+N;P7R?|k0san>Bq%mdP
zKfqz;^=@8~j)+%4GEi6CW}Q6$2=DBnT3A`(32I~tG~xkWmsIPI)?N&1SHtPs#6B6_
zgl>|Wigh^9&k(oVF%*g44i0LyhS62l?A}2xW~XGjvwBM>t}UP-En`1rqBP5Ar+-Nr
zAPxQ=K&K><-^G+5(vbItT!byKyDDvWfT(NglFV%oS&gXUoH!#-Ju&JCbs$KWVlWwt
zx%(D8n<Qa5>_BNRF9P$N?^TE4Mjo_r|4V?;rgTE+A`!pIL=sJwKF)UIP_@saY6M;V
ziJ5&k&*|={E7?>raAn4u=<ZuNQ?9#A9!BoJ#V=>3U(#?!Yq+u{?J^N%f4CeR*sFGO
zjC0}ZcV~0;&W0f+t4~KqNOD?qKp4v5(fp5#88LUd;a$tfU%iMxePdNc6WbCPtuwd7
zskT)iAnS!y_kJL;BW@g59Bh29r~t3+-D-WKGkUE86O{87ql%#4Ebj|Y#xmhG=QDEL
z)FR1pf+XIlJ~V$QgQYC66k;0RHklU}SZo~?-|;@>r{(@piv5pQ?Tb=kLn=10pH-z9
z80c#=(NdF@9XFBU%z6dx=ByErZSS}};hYC1#l&g&^Tx}~Dxqzf5=U6S!IfaaL+(4l
z`|?H6C9pNmLIL*~vocTZHk7_(;*dOHhO6{}x7HFd_2;xa|F=gIhjG&@hQFP34H^$<
z^#{0*j%4w*r$hk5%p&ab9hgWL=_z1=i>}QgTMl&D8q`fS0@0mm?>vd7TF5c;;^Tpm
z0ROL3U{hJ%;wdFn{pq{(nB$-atAIv5-sV3RcO`#!5Fuar7vlFu_JS293RpfBNyX;4
ziKteLrcSMf7&7Y>0q5?=@hM>CK!rw~dF3=rpY#)*uRmIGE7mLY@-{L%Vj?<VhM!3e
z7{h-Ocuu~}#Am~?G^&<~n6w>s)xM}%qtZC>Rr~iX#>Cm;6lO_`29)=*$MT+eIs0{b
zmGJ+*)$Sx_?whh}%hgIO^mU`~4pa1O9Qi9V&b_ivP&~qAfr|fp3*&m9+aQMvJa>v_
z2N#u+`61fxry<qIt8ws&<PK=qGx(bj4x3n&aPvybTSDPht2RQjFrTZptkHngnW2<5
z?CEUuPWA}D%}@JxC*SpS^GB~svuo(ZhwBgJs`k8i1Y$0&2kCG*Ov9d6_HD<)GXTu!
z%pn(NX>92;Nv{wv-;zR}hw?kVzTt`J#%DD>L9uo-j8`m^W1{#)I#h(Lia*F(i0#?h
zPK1Pex=}vF6*+Z3Z3#1*Ii-BoaV_w`1P%rC=VXL?!1`m)h@a8fNzgr<{d9*C8!JN7
zw>Vj{JPvgE>o3R4{vv;7oB74&xIT)lorJ<&V=;=?#AD*U+C$hQHD#(PocY$5H@lTZ
zzH9nxsk*@S`%_06F#o(0*hBg_MtZYMe4q`XEq8zQ*Afoz1Oy`r-h^p622`4-^5J8h
zq5^PrwP{7L&b3_p`~hho*(=n6B;lcP3i)#H@C{wUee^=Wt4Yz(;0Lj%yrwm;NDw62
zda1Q2GECa^m%dCJ>kbCzvVwaK@_3t+4N`lr-J40OaUfMW&A*TG1E4TiZ%J(K42^ew
zL!Y*e!eO_{(VJ(`i_Elhh(@#975SlI83`S;L~Ht94vA<icwl~HR+5aq!b*-&9F%CK
zD7vWMGx3B?rf836+n`rcqMIx=XQD}8U2<%usn(f}dTFy+{g&(sQsZ8Dhv31*8B+vG
zxxC%`g(}w^pO0;_ed&|HvSUWwd&$c}b!#q5&i;!u_eB{KC$hU*{GQV)(@IoQ1$@X&
z3Hy2ZrLxUtm+haCWFMukZiP|b&FiL0x$QW|lxMrarhBo%N)Jn3DX7LJK6=khWAA|M
zG9Efh<cy5CfWVq~XA@cfVuywFic+Rpqk^Hprjl{x5YXkF4i$m7Ix%&Z-)s?m$Bo<)
zEvF$R{)MVZPKm$5R~JvqQ-My^+&Q)*N=i_*9nHHjmrMhhf^`HL_7JJEIA?&up7{r+
z9K#0?e%=I-9AJ{S-|_<YC?cU~{r{iFJawlK?K@7{t0g>jKncwL6OH|r@We!0QaF;2
zl16$APq`jC>8IfOG9ul}6m$j~O;L;kCu>AB{)#E-4R>`~UVJ(@&c(*z|8i3xaO3Jf
zvGK>flki=7d^}gtpkuh5`eWA3^4%bH#=S*?=VPKdZ8%=9B?-qs7W8)pgJ7V6Ub}&j
zPePbzTF4_V0Ji`3xeL_ft2O^ggCs4+_CsfHe2kT#4EQzNr;}=?HVDz1=6%kQU;bu3
z?fa{jj^;DxdoGSh0h1i67}i8~3AcA+`<H?60bekW8VX&;9`zRWq;|Ujd-|2AuK#hd
z2!PAU`X}JHLqMhQCKRY*Y(o)a^m4<rc@yTv`_@&}$M1mQFH2^)MZfM&&$riCR2zD@
zLQ2$x05z{?dtk=(Dj~WB+Ndm)t9cpkyI9e)N0@Apvrh{@OS<YlxC^{XPpUO&5_q?N
z&-_$S+dTY=k=$74p1V5P`#g%Be=Tti(!}wg2!vG>0P!lH(5?4S<QkD7-uk05eJ@ph
zWHAdIGJ5Tk4zK%=Cw-suqCb(PcBXo`(YamqOtF}j?3#=uW8fO5^pmr!mDa@9#dI}t
zQNbJF(wQsr^XA2>L*v+UH#i;$xi);^tX;jh>wpF+0Fr*%>fNMyASxNOaRA*8ws!HU
zPy@aTht@RMaJ(G<6xvkIGhj<72j<tjla2cx6tT+S<eawjpzP-F+NLY(##X;c6?>|D
zk$h8+pE1376!!5<U$2`La}fBSsRyeke~W9OMe4RJ%AEXq!T|-;n(!Z$WeWi8F0FTf
zm~^yofYDlIgTuwL59`^Sj2V%#yk9N^7`3lvT-diwFD7zKZ-9*G^y$!6%X>IQ`m6Xk
z&Eu5>k!Mba_r_aj=g+HTYxnw{=3KGPy(HHo{>cC*?H)d_{cw_4m&Bq_=emx(`U2bY
zuR7+90&T6cWebzUtS`r@@q`dij(EX$lc1Vr(Db3slEi#rx5qNj@o?l2=|Gs+^OjUl
z)E9?t2!tkV<+y~YNBo90)mX}wgV;27nTjMuwc)pn0%KVzc@?p=P>I`>at%?6xYPpz
zy{M60%V{lr(n~IDc7E{|eYJiTooV>KHdkrW>TlL5(tv}mWZ$sb(C>wbPEC=)Fo6i8
zjKwv!iyZ6a#<mknn?BtqG9Rdhd#Oy-CX%LW&np*FKcVf6eY+kDy2d!?cyjiF5AWCg
z4{+er?m_Ug^ORNnj|s1J;p{a5eTc$DxrfuF>&Ixzkbne~?tIr)hcc_)l>aciQm|9|
zv=+Le!Y(uD>>WGna}FuJ%PI)D5vCR-p#J7;(qi8KKpP^xcCu;ko3OvIV1_34+j+*>
zU`LK>|9-&YnB`LrZGYL<eKm9vRgts0uefOrtu@-EU9{$v)fW>}j~8;x<>9NBsE%n!
zm8o28rvWioKKA?h`#a$8im*MJH5`Rdw0DA%Z$B*SrYbq9MEUtRX1}*p!A*pqin*8P
z(o};%%y}=PfTrgNaTLv8xO}(mDr>lP*QIDletwE%_w+nQHMId8?wJbQM|pVgD8OiF
zUHqs$rsXN71W4QNT}xaB=eb@FyN+^q>GscgpNjpGJ9IU3JCb6MoCTfab*#h&88_Qy
zr2q3qkL<KH91BOh;?WPaVaL@Ohk57V0+8=juZi={g6E_&)Mg(#<~nu@`L-S{O{Ct{
znZXP-8zCPGT~)lo7S-@i#POal;GoM-qfz=>3wEm`FkSMP@p%gC=dHy5fbW*a10RFk
z2R}+*N#}}$q`*FxG>T9-Ua8Zn6uGo9f(4P1x@S^K$Z$WlyAqMIJ6qox(dvb%IZNT`
z^Ik4D3F-0stqf}OlK1yK%VZ<NpVYTqU(Y+HWH1h^EecbO6s0U*74Xy3{I-rdTALK{
zvBmuI+DNA)tPWURC>Ga!mC-r$Gg?BPDM(C%#tz}5HO}eIr`{r=@57m-rRf>W?!)hh
z(dKHZn5Zd~##Ay`sWq5e??~j4_I!=ZXA=o*(h_pKC6RqrtGrTRnQV_LB%_3UpSzV>
z7+bC9v$p%ZJ%EFgwL1Ki*W^WGhhB`Vq<-n-gk!(``x1wK^+x_@(!GC&x<=n;UU9!p
zOZbkK<(rbnx|Ovqn?T9gthvv)Pru?ytgRw#p`naV@HCD4Mo<Efz`U)YL$*NGaGXhZ
z+|ReauyGZ^NCn$w*goCnuQy`&F=WPjP9yjzH;jvD%@ID+5_os`ZEc7K;8bsUXhU_L
z>Y5FX$qczRCo@(TyfD^&R~4;3Ha`;h;-*n*AmT1Mn+N^A>?zy)vjiv^b$t&GHiZTZ
zhQgRh2DljAg&S3ZE7?25{8WK^w4_bR#mN5v2Iw8<zqk7yc*S2cvFgvtemE;~*Ylxo
z5C7$sqzqIp*nf#{7(NuV|Kz43Q2)>jdE(by`6X$Vh%sa`814w4^Z;J>UNvqJsjUlK
zjE`*kPLjWHR)B`hn=VLBGGW2648&n@%~^He-g%nix1>aD$KgXW;L(}3Gh8*WwCcv4
z!zq<eYvsnIN~V_XId6W2G$@ekCclnV;I+FNMfW1<1lfO41BV|oG_NDi{JkX>O18X1
zzW2RjpJ?gQ#*Bp!j%V8jyZWBJ1u2ZU>N7I~MH#AqyKM_;B}baR*|^EVWW5~1F0BnA
zf$@xVZ(19{SCjRxL*~Iw_O#~|t#&9sSO(g77#igQyZld}?#S}rwHRXS`$tdpsBYyd
zQu|)NwdKq{d1P&x-u)loioU&BLE6xX3ob#%bhL5n_<6>Vm?+X;R_28OFP<i!Ouv%j
ztAUdIe$CJy5kJRFit}(VS!qda*fKn~KqXnw3Jg<*I09&e<t)=%FlIG6HMnbQ7BSd$
z@amtySow8msP&A$8~MizmT4fbm*n3sFE8;^e7|b7fw@Boxym<#t=8Vi9wjBnTUd6m
zzVdl_37D+&Ypu-whnJ7^!VJ#aV5<^HA{ytK^F7`ab2_~ZvB0}@b=iJ0E!cP=USsk>
z>XA=W<C1$Wr-pD~lZDY3QQn;3@}kWZunO2AzHH&X^oBVSVSxfU)O{#BSN+rp0vRlC
zR!HakhUSOWG<OK45z}z~1M^$&Q-vN-eFeTBm=)d=wR&2iBP}+ixs#peD#*@eegLzv
z4oDbKo+XTgeozA>Uz|6+?XSG{;71+#TM_%zLMf0R3T=wWbdx|$TI#y&zn9i}I?!5g
z%Qi<pbGxv3G@GXOa&`o<i7dp#-`%WY&y;9+bURpfW;-bSD)avUdSwoxFH>%&l*iHy
zMN_rP7e3cygzT1|o|LZzq-p#KfYs8|Wl==>KnX81&MJQkcyo1X=zy>BZxUNG<Y$EJ
zS<ma}IECfS%;u!$n-<mV+n~Ouz^g?!f*m9$zQj6a;+>qv*^KDScAdO5O20Kj1Le!6
z1Fv=>0Zg6BlHV8b+pHm}dPkTWS{MnNBaErBFnQS-d|L_2(D@|RyX}rWzTBqI%VE86
z@YG4BdJ6*I*9`+zsZgud08XF5O@#l%3|8md(!Z$BSJ3h?t&zOd!2It|vxsPHRJWrb
z1g$>O)Wb0=QCmmLUIi|=i3Amtfaed)pNN2lF{wy7Ppdy-3ohcSuWaBa;P~&vEK|{5
z(t|)L16}|qkB4-h^1vWMyZ6L;%Nf&lW%m;VzghCxKK3e)31Ab4WRP;EYAfmZ`<FgT
zQJXE#2dT-Vxn8cry?EB0=Oos}pRs6M5k_<LjI|PVusk7*vvON(fi?qUk6+L+(>bR9
zX(+$VV)rv+&FU>$wLmNNBxxVKn9zI1y_Dep_;S&pe{`-}O8c|PNU2+wG&b{$druCx
zGIzkoy0K@s;Zt@$t{{G7&^D2I_60S1iWf~a+Mym11=Uz|Zwpii;fI;;4>`z##~dKW
zjrilh*W4d8zOUxffRd|%hw3>d5YEQCL`Xk`P=dGfSgHeZ3pZQ~QaXl{aM>ap0pa@n
z6@{gX3EX{hm5j1DLw`Ek@H|8o8W5$8Xr=%xyUDpPYW-CFhkuZN4I>Y*!@TPRj<DrA
zKd;&R^2AGIj~Iv-SRz1qi}#~bMsXa-#5P5n=9~Q=K!Ii0-evLgeIs(PBvs|xpL)>W
zCx{<|oxl71v;RfJ<-U4%Jw6~qKC{dp{o7%02*m4>MS=gnk2*~X8TmjmB`E`w$z=Oo
zYp!tU^p@zoToOsR-hMEt@9xa0HypN;qr0PXNhRg}OXfc9zw5Negi=_1YQeulF(B-H
zBGqs;?+|We8u8x&C%S`V&Er8hYr5I)rfCtQQcWQ@b?I=pDwsRWkD(`4L~&6vPiW6R
zDvT;*-!=rI3B>tFXE!Qq_{k_kVC_9feonh<8HMwd`C>^mg9Kbk{pnI4EG{ZX1u=F3
zoQ=e<<!6+us3L}9{TRKwcOnO$uWq0!Yx+W9-n+t3l6a$!gO$88o^{qGsS2TT`AxGr
zjpH_q%3FA(7S(3B{%>J;l(Q(pt_6-4IM2WR9B~wVNObI<G5!_8%v%b_pVygYAk2qi
z?FDGiw~dT?<qBVpP*{+!$?9t*;B4KCCM>^``TLw}Uh_Dw2h9A9+wVhR0YS%m<}*$=
zXsxQyO5du&gaj*D666M|^*@P(|87DgxxR;EL4!Vtv}s*sGiA@ytE<Dzng4m--#Ke7
zjGKs)+tpkR@tzB(otOUwbj&rGML%jg1Ab6=mopzQb6Hy}7%+4;85Y|-F3Is~Aw@5(
zMx2J|VEh+3qoiG@+vjgrPFp{tXmz7ssq{4S@EHm9pISmR30s{YZAS&25&){b6=T?^
z<<LsT?Yv>$J{mWyoQHu;U>3VA<pQKY!-^fBswa!wSu3o^5zo)KWdR`p-2q;8fnitp
z67Nk#1H(2CwwIqWtW;?UK*BDE_R4frmfpGa&}p7Hb`-t|IvFm=yYU-O1Ma|mheD|l
z^+je^-V#o43_o~8v|nCm=pDJctiQJed3XJx>KK{%x|HVoL#cA75arH9yVB(M3XQ1k
zn`OgII^f15qkKfq>YHOO1b1F7K%?p56B~Eu(q~R}af<F=pQf#z6SjA{r)DG*Y2>G*
zVO~J8&xsX3;&Lrx!jXGo7V?WW`<uLj**5aG0`{pltGg?>Kwh*&Mu(<f{|cy;SF!EW
zMij9#Err+pf$S9nPq527CUx_(hHO2)IujK$pwoA{@PHf4|BxVL!24Az{}_iHHGA#*
zZx9BeaFdU%2cEea#otw}0T+&dFX0x4mIj^c*~j`!x}J|&y=1esBC9u)GAf9CXWF<_
zWdaHptO~nw4f&{Z$(+q1H}<zPVNcTsdAFL%C-^$O&nj$F`ayMsU1q^}`?JCBrpWGY
zAAEwpuzWc8RWb<!KxSsC4c^_q*<BEj({!zSKR9A;86n3|E`QZvxsGTxkjR$8N(RDz
z?9dn|h^xyH3|H(S6HnGC)2q6D4gT`@1>q9o+^0JAgFH#A&W<`#cPAs<oXB<!bG1iU
z@H0dNx(rtDBmO&qAL6;Q#Q<1>#C#w8Tq6>|JA11uSQWd|PWkwKi5O-}Lz2w_?X4Zm
zK`Y($mAPTo=1rf6jj)0=X86XUy;<C?Dl`jYg?z|Ef#Si3PH?B07pXw8(;6wK273@Q
z-bBN??`Q4axh(#{r{Ll`g1#A0xd#<n-G;4P1g?9Z<eyb{z73sSAHEt+UF|xc>23-l
zihf1>^8CY#-FLqgh^Fl}VSkhPIIf83Kq-10cRXKw{C)_%aMVe6%lnz(_7^AfBH;3>
z0f!#Ni{6<eQ?JEpWV1evED3ZefgxhI`ox41G#yJ=HTY{95O*sjY9g7c3e(`51)P)X
zx%6ZB4O{dFX^HC$v(=$b0@Ymw+3aI!3ng|fHGn8lm<kZzDb;6IE3RbW3dO}k`ztRE
zQ=n!#{`5ZdwFbC#oO0b47rW1L-T;Z7cY|XSF;WYz#aHbl;emEsA|0BiWE0D1(nH4@
z83heT376c*4flzDa&H>tgRvMvm3a>)u-1PstI2;DhqrjHexHSfn*Wh*PD?}F5<bYG
z9npV}(EdMv5L6~mI_RJ!M}|Buod8;f_lwz8dR-(|mk<}wm+9YU(lN41(m@B%trs7-
zM(D`pqCRPyR&mnomL?1p6^mtfY#b<I;glDuWk#Xr{W~!$6mz>NKuHcxD+F-dyqBA*
z?cL?uIBA_buoZhe+aVEzS<KJa<rxS^keiaHV#GJ*pBXTmSm5+8k)78QWbAC}?lP^|
zA(RD*1EIM7s(P6Nkm|B;_Z_{+^`4N|SA?yfh&b3~Skp#$<Yym=flvBVx`#G=qYQms
zs`b(nk6uRB+^2-VoY`U#DOx>FDr)I9=>AJXm(WA+?S?;kMmmxg*pqvoO(txm0ybZE
z+PCMjTNPa7;Z8%&=Bmkk?{=4J&@Z>24fAi}|LQb2SO^t+QLS4zH~$>Wy*1ES{T^B%
zP?lcv!g4zVJp{bU)cX-AoTU;TMmxLF?l}&^%6R8e57+Wq;Wh38!f;<$TL%-3alx{6
zU^{-D*2w$vcn>ke|3yw%Z-{_O*de8>tbvqk7Lldpfs^XI5%oQ3Yv`?cfj}DVZFKjN
zPrOnTx10KG9Ld<9bG`H4M&!n~z~^o<eluTA&rkk9g5Lf#u>Z9r+{osx(Fod;$Z$UR
z|Mq|$CPlr!4jh^|CT5y7cSiiZP8O0*3Y!l&p5{NaziEKehigEE1EP;Yf6{tbKASMl
z+tP0_5S7!lX}Oj9PvHkEgp=NIoOer>;_{S@XKG#3o>1#&tr>}x(GI2dcPB5bL-j`;
zHbNI`G2WVvByVK51|FMAye^nEi5&9N-utXEo%)M&-u}-Y3c6lswkGb4BEPkPM8$5f
zx54&eD}`7xY0rr7&MB&&f~PQRFWTOxgf14T9mW1c#!uy@t@j0JewUZ5G}S&a5VGfb
z+6<i(ANHIF%!I_rw|z5;|85x}Q%w_8ZfU8!)u@Gd`$L)D<=J+x4$kr;QAPAkhBf|m
zIk0e9-Fe{VvqUGyuX`3(y>m<&KLr$5n&G2ylJ_W6l_9$!e*!@+zI}3fkwW24dAy_3
z!5}B=5O!CIue-RgF6|ld0sYw7Bj!GUa>Wacx@l&$Qofhm2K>92Z1!&Lxp%i(O!&Q{
z>9c3MKgSEaD;$_Gm&nqzw&@7!*>DzISlvhZ8_gD&P5*R|CdaXuG4Ou<NU=<|qiQB_
z;V`ECRDeSSM@;RnyJqk)LTss_2cwTxIL2RSk=<oMD~moa7;OqkopXfnsgGc5Qu);8
zu1d&`CzkT}Px`s;UkQWjjcpMl-a!K?EETdCZxe|Zwo7m`ft+BFOUTATK>_T*BYzhK
z%9E;$BAX1*da3R@@(-y5j|jL*eKa6icnm9sy9X;aHl1ts5duud!~Q7O--JKPmB{G)
z<iUlp9v0f(F0nSZ%q&l8Apm=gYAKb+*1+VP4@`2>F5HXlUPX_*UTc2BN8Wn%<X74%
zkK7802$|#&unaWjj7|Rtc?sr+GqQKHgI6D*7pUg$qvq7*aA0VTi{|%VFex2D%s~AQ
ze#)({k}18Chm@~wJy*0ft^U-1bPm?!=opudevcd8gg#|_@n;+N{##G$E0n;Kk2HTz
zXX}-)`@H6ojfNqZU)!<<?t2=w$y8yz0{F`;_^eDAgSB9<<LU$xyG#sW%#dcAGNrwp
zxM3<_Rhnl~I`cV{9KK$5$}Jd&R-xOQG?Ghy+N?Gp2VfSOdozs$xWA*Z9q6-X7y<yJ
zl`P-%sSHb1mEK7!<vMlTD%b^0);-#AJ~e{3zn$n|CT?xIk0xtnqGTcrc%&`c#bME;
z-wQP|a>P9cbKy6G3XZ&fz?9o*do|ylk8u>RfBg^H{Yc7ei%0kAY)$^}#ju}NlE-Yk
z6l}r76hS$5e(wWss%sV#QS{-SM+AqZ{;>^k5i0}Qvn5^3f=g%@#nZni*0SULX$oyr
zu7*LG%iFX62WWu)sI_ej+bsL=t(F0${SNfIW-;OV@0$5{vT%_B@JP*lh`|sGv`KT^
z-obyb=#swa-b#;P)aoTGb#{LbBYTBNnQ(KOpw<V@SpIS`V-32jHJ^l?FCM)WYGXeM
z<E-ILRr@`s>^iY{>0$nfMC2K8q$i)PZz#}RexZ7&T_AoxXGtx$qC(_ls7Fw;b&Nu}
zY65t|OaUJSnbyG;7bZ$alQNcg(}kFukvaN7Y>NoR8#wVS_`4U6Lfs*D3syzWIZ0E2
z*kh&)R7sQMgZ|wBsA$R!)gNE1BBSwpe8JHd3l%psH(1Y=PXE^Ue?JDs<s$BS3&?J^
z-}5!M4@5yeBo*DjONF35PG3Su`l0BgIf1TU`BN`9;R=L=vEYbb3uE8SdvQ$@s`w?K
zU-+x!Amin!*g4H5c`dG_a(3}yYe3yy=B}Cwq!FF95Y9-bAy?jeam|VS3xLYvZnb=h
zveXG2*SrM0^zDBDllw&ahaz?{iBU2SmA8&b?^oNeGV!mN@`-FLJ#;n=M7{ntpV;nN
z!wkk=z7W`XWT?J=t1tZR7Ko4Sz4jA-VP@p2_wFlIII}i}Nr;;&+W}CFqpQ&^Q`LV;
z9;MZDe4P}N+MCTTk3|5ELi}~tKGS|p5j}wbhLcJrWy#8&oUnjI-tX5mgh9s%4bu#>
zi5$#&@Q{5QH9$;~_FS%|pZ2>DX&&t9tB|c<b6q$QDa2d6!ydRE8x{x~8M71eWxTV6
zGd##x2zO6m%IF=vVt(k@r4zeo6W*kCITpeB2*Fnd0;O52PBQ*rpf66!*jVz|czl){
zNs}sQ)9(1I0OvgSoq^GT4F`=%R?<#+yM(A*;|H}H;3v>pl`V-}ZluAaQHB7MOZxM!
z%42;Yy)1Ew@9Va@-?`me+%czee=cv;_3dbjO|k2Z^KC2=IKG?GI;2Krh2{WCN}mI2
zLCnX#1g2kM9I-o3o^#A6*x*CkXDm2BswFt0=E+T@H!H3^J-;oV{qC9;*{QB+w*paY
zjF#0m9KsY$ycaUB%ByU=;~N*i)|N5PWv$W59Eap8ja)Vxx+f*F7kH1&^OGaTaxl%E
z%(#L2`4vG9ss`)eOsc03E9fHGwS&0_XH$u3)t55lG@^!UPL(L+mxn!SD6j6Uks_i5
zT6?+*1<E&gwna1{!GqlMDqIGq&8C-6z@{s?LA-p!#GssK*{2z?j>Xk{8*~oUB_i(%
zf%;^I`ZGuXcbuutGBf&nT>Z7MVX4KL%npD(g26cRj<$UYN-BbyV&nhYN4{L31*ds+
zrqJIAkL9H3BVg0(_XnD8TBxeH?8K)el9Q9qc=25BL5UvGRJ#JDP~>bJ8(NWIX!vxc
z0gX{?^s%O7VCu7XEQ_rtMM7hj`^3Hl#*Pr;ulI&sxqq)uL*#5%CuhFXPuBoNgIHhL
zg?p4Z-<OKyT_2<j9qjNNry)(<BjY<#O|2Xr@OHA%z=S#`?v~5P>)3>2C)i7S+BPbW
z@g_1jfH$>M43pA79h;LT7;s8nXwzr76H)7(2V%R!Zfpu-5QNtjKc&eAfSW<HahDhb
zOxeL4;ri*E!JwG^OANr!_3~HqGbP&b3o5J!)mZSCe431G+CO9!JR#X_>N{!JF|kf^
zcpzhkXabxy7{jtd;g^sqC@RoDSH4LcTK^6+KpKod{Ja9w*mG=}{(8yb9}nn{A!uu{
zCLBl0R{P(K_6>!@sNroyHqQ<(PD=#8FSw|RmKl^0izZ<X(g&$^SjOS?#(JA|o|``}
z?h{?|q(H)16rWgIDba|n*%Fq43UcL4_}??W0~<-OO$R_()Aqe$cP$|1?qAUDouEjn
zGPkwHFuY3TUK9NDtW?Md!q!G;$;L~`{|z&=QR2(^XY^ngT?30pD3p+Xjewk?W_>lM
z32=sWYTv>UDkggf1sR6KNj=~V6em#)GiBN$d3Q4Gf*^CHgzj?;(1o@)ANAl>{oR^l
z@o6Vg7D|1f!Xs@uB0}7sH)H&<y=cF@u65mTi;tf~Hw_$|pxWW0`3_u0W~DHLUS<+a
zAG(IMxtE<4Ts}t_?B7m*<9W$>A$!yp1~}Y4F4rv>7nTZ?j)?htAYS%D&N_zoMV{Op
zzNOwL6;cb+f`T^XrLf#%;5TQX#TdWzu8k#6P?*DxnGE~MyX_oMmx>I_^kTKQ@}Bv^
zuCsS0huPiMD*|3nS{<cUsi8Vb^L#V!HPL%=CiKu!R)nUxjK1OH^;;nTE;8)%cd|!;
zEoShSoo3@7@SzrWI;O4S*_C^iii4qO`ROw@8t3_3yT;Cg2Fdt>=G{KsaISoz;wJq1
ziN%YGSkloUV|--6CU<E<?2_g`Ns!R5e%TC8ILnC8hSOV)hKx<T=4qu@|K4KhJmwDL
z9h|N5FxpVcrj9&woU1`#OhC1q*LF%K(m0Qy<@w!!8M=z8><m#PAait;j;e0TO%k>O
z=)c6{!p;9kX4ummcd<UiiFoRIf-dtuL7UvZy{kW*^5+sG-3Q)e**V^|!4NsR-_x-H
zB}<9r0kzk}!px<xR%+N&-qe#`VRF~>BYQr&p@)d@N#)A{B%C+UM&r=q@e?Hx_t<&I
zpYYUOUuFmxa4B3hLd9UxL<Y!Gdr9u~(p6+g#^!UiyV;;76_7ZP0KCE)#MO!yh|!*5
zR^5tz8}JThCigeKx+^3su;*!inB&j%>kt{`l}=S5O+O3^-w?DCfXA2T2M<FDbiG+f
z28PA#LFU#!AeW2>8?$uZjD1V&Of7_1&lCixK|zvf01oKIqWz+|z+2c0|3s_1lRQXz
z`H?QQk*8DPE|HMW+Rc*~C!s{7lo1z(lRlME5q<3dMgU_$w9pYUvk1QkQHLZX`Uin`
zA?cNFhMq=}@j#t{;KX_CLTPM9Gi!=1xUknh_zV1|-DCiBfQ)6q0#;jlaa>A)Cwnyv
zJXNq``1(c$=MNkBncF*;jSS%0MQNyeE+90?>6$U5Q8v7s!t_Gqxg4o9#cr_^_URna
z^(^9c%7)zubQ{;@@#WSf_Tl;zA^VgdUkwKg+Ld5N(R}V_Y(jK@iqKMOlxf!l^;t1k
z25oFnB-NbPu_12*AKJgr%%}+#SqS(clwFSE;ERtKTK;RmJTLIhbfrcyI0Nez_jWs|
z!pK2LMcJ8=8#&-SGiKM;z1y=L??h4h;wI54bt3*P`Ocxo;fVG%#<Rcj<AMK2(z(Yo
z{lEWz&WD_vay}<B%A9g6XR{c#IW$zvX-JNRQqG4p=kqK%?BFb+Pzsqt%wdG2l4Fh?
zkj|gq>;3!w!R@t|{oywD*!8@w$8~?8n{+BJmo<GB*xNs?Rz()lvE+39D*xd7z;CNN
z9=N?-hVRvAhD%eGCcR#Y8ySz^oD+u^Uw962k9394$^>LX>X-^L@pTj+ZPEGRdhVbW
zQ~j}lYVHWzFjLlARaY1IK(Oi06@5Tey+x=Jng~bmY?uj298g!9IF2`|Ejh7&-@XWh
zKVn*Q21&<Hn=_Ef;&1H&V-G88>iLk7*o!!Wy?|GCq3yPxJ>Mb6Q)&oBKw55|f-+UD
zNwh!)ViPo_k+kz~2e8m6x{t1$OaC$o#Waum)6pMQO9m7!Xj%=CU604fieM9{r=d*X
z?frTiO;s$$o$olOH)pllY=~d(UJi%#SLF(y`_dqLAtUgApq%Dggx0gaTBju&KcHEO
z>b`25ta5qj&ON*o;im?@2OeW#T9+twRgKB>nIMvRA+vls5ncl-n0d{c+2`N8>k?YR
zAq%Z|#9Ct%mYK;Nwo=C>!2x-m@Yk|jPrtfOwRweUk!JE7>?y9p{Z<4WBtyxHsm_vA
zlj~5RaLk~8&9o93diAP!ww~>?*ft&O_0#X2f6h9UXcQQUb+?ZF4>YQ{jcPXOz5Axc
zx@;gqM&s`1<E~03EwaIeJHeCPbYQcmS=B$46Z0SrfNxk2^*D-XnDzd0JwCs1YIyZ+
zAE(^kx_>^oF0ILZ;sdjEhjHlTM5U);CnXIfHV+!T){y2c$buv14HwsfQx}OXOsiDi
z_3kwnrpmZG%jsHm*Q+O&+bMsZ7Zylc#qQdV3knimM%*b3=voFQlK*sNNcqWhEZYWE
zG5vNTf<Czq=Si~LC4}J~#1-qz?y<Sm4k<kw>Utzcmkok8`&7ye6$ibPqP!%g)hGCU
zy__^ZB3JYg>U|>Vhs2q#X14*JziP<<?$i9ep9IoyrAIchw`tJ0c4P?LwUlT!2tZaL
z(({q_oF-o{4jN4=x|s%zg|GuTM2MajO{Y7TCfQR@CnNa)*W$d-?qE#q;!?m*=c|f0
zR;7~M-oM#VvrvAQxO;K7R7ggE5Q;Qf&J|+-z^d8>02D1ee#<9uHkYqGz`%sX#r4P*
z3}RItY&t(-B`B|Ct6kVdOG<to0+-41E|-F2P{*_3Of4z9Pu99*F+pv!NW#u-0d3#&
zhvIj{HurAy@;fskrL%CPric$`xP-5I6jcF!`d}<us`x!!)#}m=KSX1H8oQ97!KBV{
z##H*`>Y*#oVaxd|tfvnF&cKMrvWjdX(?G_}Jhy8lhm~Nqg9qKMXb-~+@U3tVwAqd*
zY<b&eD*Kye((gsvek1D}U#{hl4ibAou6CENk$RLeK@=r&O2#SdRfkT!boAxquO&5-
zI4nbCK&hr6(JHq<A<}t8wTh(pgCQ=^uCR?%YRe16UFDznZM$&Rr1YoD@4?o<*V0<e
z$HC3olv}Q6!wYkl0pJCj%`&gdiI~3g9hbSDJiaz4Ugr276&C#B7>b4T67sZ?sh)}Q
z4FrvIgDco}QM@z1-Rz|b?`IUv*v2XVCe1&a*4;Npn&;78NV}O}A72;eUU3Dz$)aqC
zach4GkPfD&0J!^sWEq)R`}ayyWPw@qoAT)&c{L<U=WD#rDcTl~WRg3cN~L<$T+4as
zqPgR7Gh^e%FG7%=w2XY=FR9suK=1fy4c?jkJeRY}uEe6XlEcQ2r^h_28)KhjxEbXt
zcQy&$^fi(tE7`>Ia*AldJmVd1m8!*Xcb^qhfQEPOBDDeHH_)f}`%opK41C7iaOFYV
z*!d#IaLvaL-yf{W3i8_L3hbNn^juO_F<9wQis;z~W*r-VzvdeRw3ySR*R>7D3F`DH
z=5r5$|5c)JW_Th*fJmH)7j#y11QKe^nI;iNlUlt)K{{_)2?D=Z--i&K29YePIjJ_Y
z7)8Czb<+|zUVF%GAnqEOh`er&#umD-P6L>z+8y77-{?8{`+-CX+_3Hh5NdlRF08>y
zyJy2%j#e(xcx5p3MVk8+Y*pmg&^D6RPyH%rs%}D_Gs&*yXVY!Eq=V6lUT?H!DrnlM
zBVk^c?6{iTnWU#mVeqi!s>0hZ2@mOC1tT-_>SJS5aI<akjtFV7)9=7}UmUorTRHG1
zKKIE)p_oda(*k$ecVB1$Z4HhJh@-__^=d08_KHJS7lHV6@gkRKpO0M@AHx<g#bUhv
z^4R@_nVBt(HGz)GbYNZemZW2Z_cp1>+@L8h<(|40WGD}wSo)ocwX|wLwIIw4(1|PJ
z8a|>7MQkRt<fJ$6GU;VlUvsRilVKCi0R}4;{dPQ`Cy0sO|I2mLscr&PcujHGW@dA8
z<_o!Vqmpi#^8E~bmLd=j%_>vu-7jDuo&kkxo{SNI<GkNFfp`UfbG$Kucn+(x$%uHR
zf6jh%u1f6fo*!r2k@(;+^-6{LXD^I=cf=g_UbZ$jp0gYYt&-V^-1U2xZEZ=G3mTDf
ze4;z{JJKv=LW#)2nUlZ8O~!)L6quZ=>$$!92EZzpuP7Wh$ipP$o5Zi{X$W6^W-s&F
zWND6cq<0T26|y*p2Ty3t_Bvq0fI&FIBsS5}u=%i+Fn{PJf9-i*sIU#u?c{RQ+uQTM
z=1!gdeB!E1(UI(?(SZab8dQmU9M5GS<D_Nq`A=Qv=b304=p+#C>15jd%ZWYSX!`Xy
z{D8t)^%uVc%(W9_*Ks_Dbm;_Rf!7KOxAM&y7G9+h;bsE<L*kM@yCf?uRyG|uYIPF~
z7T~6M*toE6LMwj~x9ujnb_W7_&}<6nUB!Vx&~Sn|@$1_r1(DW`ibKBM&t`)UW$zYg
z?avtiF$I(H5D7G!PBvYMG&J1<gHisp%XL_YxP>p3B0So@;}`*MAJ0^5c*v_#gVxU}
zL^9pVo5beB7YT})4X$UB9^N7yN4bb83pBYzGe9;xy0qM}7O_F!$vxO>o8WOw0C6r<
z@;6=6@3WBeobIyOrkAQxL3W|gvx7<w)1m)?>Jzu$=MLb;9d&D3J~9C2Y@z&BYtW5<
z9K={OJZ?jL{w(&D^b7Q*olLXkB*<`~)yiqG_NuG?3~Lz&BAFR9Ew$?uNI94*G9iis
zsdXZ*0<VRY@`1VKS(E8f0gU!kQSIX6kJs78vSxnqmpHb2gz=!_NDPv)mn~Ro!2ix2
zf4M_&TCE@c-!ATN!#`D@4;NmNK2ci>N7I|&Y2BN~IuK~J8U9TRs-WBCSFIC%X?6oD
zf6*+ra=H&-li0J?+X7$1A|1*eTRHfl;canOq;&HhH5^cN^Cy`-Hlz^3qe{WPop|*S
ztS?7|Yz|<ptECY|y<h<*2lI`7`Dci5sz?4`hrp#LI$1X1mrvjGZx!KCUlO?>b(b(o
zaSi|Jq4cKkXLxp*_&QS4p{C2uV6mV0g=fFH{H0K$KW=0t`!zJ&_^svXOW~@O^K7@S
z$|PdovPMpQA?N1Na|NF*j;lCtBb9rwq5{vWNhwQ9U+pb&A<v>HFrx!n2Yiqmpju-h
z1>)hRpnKPIhX6XN5R71{tboftls*Hg5~KpX@yg{2p;!QmFCQc%+@dR>gD~0$x<VX1
zj=X<@K6V|Sf0*2JG1=?u{64)w>iwa`T4!%<{ef<E^Q)KH4)6<z4O_Gv1w$}wQu0%B
z5m#t(PCg0l@D=fjwvS`kXhBbZEkB!hJe%wqvN-kutG)xic{QfClNkx$&RC;{a$EZ|
zxrO6eMK*L39+S~OK0#^aTD9qCv4w6SHw;d7)=;f4Sf4-iNguF2{eehmACTac$1)F1
z)k2JPiqKZELXktdQi?2MB2wbQeGaMfN8tSmJ6a;4qCGp7gH;Q7wtr((ZuhABW9)wy
zjjl!=)YI2)C&kT`>(#Zg@R8i9VR5cwj;zO&v30CvH$A)CIVH0XfXuOyVZ3{RSZlE+
zgh3tKOm%@oC-bZXu%f_geNM7brNijzfnvVamP?v>^G)aJh4+>kI9T3K)XSOG`w!5u
z&>F*A!sh6{t-JiY6@y~)Pvk>`to2@v<tgl<&H}U#t7H{)8dNnTAsYN8n2$8zH{)~5
z&B9Rmu=&KUlqB(51r#A%Rje2?-1jq_JQJFFCuqyief~qRcN^2le74rg_d5-@Hd1LJ
zVz5DSiaF3tFp6F;AK5e;5{S%$10)QI=QH)(I<pBXE+g9vfrxo*vChI-aNea<<kr3~
zo^1<q%F`z3VnErC%o{<hl`f|L;N94K1Ln=4k1$(0_VNEff(xBnC*_+W&MTGq-VYxc
z6pUb9{uwWz&F$#<3vIq3(VQy>IEARn*)}@w6Ek~vD<C`y*$1F{wrw~dN^}GWKVhIm
z)|p3ho<+h&=pWk%IQE=G;y+AmL_Yt(#+RygnU;r#?VTaJPww1(L(fwGn|5Ns@_!(<
z+GwTj!n%<I8C^|>iBbINHI%SW5Jlk9aQHTQw`%dupw+#POm~id1YpEy`EmDB>~G~e
zm|KCu^TWO*)yik(cIE(MvFe)sPs86P$ui&6zUS?o^;6Oj?2pczlbcyPQ?qXA5@nfE
z{K+LWLT94GIC+M^2Zp>&oa<27P6jskn|e1$nL|&&0s)u*)^f-KL&@3AZTk@#M`Q)Q
z>l363rX4u(Z5N{;3uWEb<h_T;p)0jE9@rE<(Vy8zjx^sYpHH)4Xge2jXPWYe*U4BO
zCCnDaL%9!ieftTI*sF>n;i6Y0f~1?EL}~*Of9~|ZGSa?q+;0EQc+l@EyNMORLjDC>
z+iMhtkGn}>4<N3xR&yiCPOz@S$|o{TY~4lf$4Q{iCcM21_7@hK0pcy@=nk#zlf98s
z4ET*&j^sMw6(g(QA$XeAUEN4jIaI<WI!h7Ka+OU;VG_;$AxT7jCP^_&J?q^FdbMvK
z{(%ZmtvcF7Z!`CeBivuFdWyO~nz|ZQ)u33`9IlfSryoG7Dr|vU8Jf7AC@JqKID2hg
z9<f>7u2sJ?HM5~0`o746yIbK<rBEtNZwnPCqnEpcUz)QUoLs{9`sX&P(fxry3vROD
zMIa0Ne!EFV1zq(~ZTmUO&fXOnJx%U&Qh4we>x(vdI0>Ggd>gH`99azcH1V(<`zhCt
z@Npv-lii>#Y)$5-vW`c4aNc#@7-4X7!KdJp3?kjh<<iC67h0xSvGW;J=e9`-@Ruk0
zofe;oaVfQip+s<>R|$+z^gQE*Pc+}7h?^0SWKvU-2l+SZ7SWQSemEG&jkpgPgN+-z
zD8YR<bbCQE(<1HZKPLxIBk)Ic5-9;>|I*%lI8GQFqjIB?r*3I={XRfXm*CGzJgQ<7
z1-po;EtjY!sBq%AlLFZwHn%1@GtL+J!?G7$A7aaxpX{G~Me*<2-q##nHF1sF&}8_z
zluud#aKGC1hekZWv9^$ZRyFbz2>q|)k>PxMhm{EOZcKI2$qFOayMc3o)uBmkSN=xI
z+@!jBPrf7K2!0X6I~i*$9fD5olh&6q?qG6Hn6G5Ff2}BMS{kFJQ{m9((sEHpGU|pC
zqptRMlpqn}jy>+4V1+o>>ycAweSANy4Fbjt|ETF(e%J$!{P{WzEh&S7FZAK?>8>gc
z8G4R4Q_oDLV74~zN<MhweZG0>FMiZ~V4eEy4)W-})KX(3Gp~j+=JHCuzdO()v7#Nd
zShnMl*?L&FQjka3fv**4a1<assK%0uHzJYFXrDd%>I$xl$FgNeH(;p26z=;A)WQ@>
z<EgGG!8FnFAVVAwUO24BREf4GuhXwiw&tIDCHMTbgS3QP(K~+f!JfB^d&luG9dzC2
zOKo5{T~BS?pw_S2rTUS7Ezf0jB$Th*N?55YKIk21OG4Gt8>;u^Gf1z#YL$L`Xrqo!
zmNTBaH0%=nvu^8c;i`+64*8Jw;P#}#FV?L|wh_QrX4*x8&CAxE@A)JF1OsSC@POHj
z6t5S>{89Hyu%T4Yd(L|_+$;Bxn=oRhJ)(N*!!CO=tF_Bvf8Y5BP8fW(_#bHCWAA}@
z!O64PI{!uRC4RaNYx`2vmffj0;5F`hBLJ9+V2XDr7?VUZ<-CQ*k_rk!hL!tfd>?*I
zP2QS>VbZ)RG2g!fjH<^HWzUle@)0}X_X-~AH<`%RI_}$_(7IK?4DtA(B${m62e>0Q
zDwKaZtRSEff6j|dD{TSY5I5Xp)bQ7Pm<w3OOu;RZN-?8oI;3Rpt<{AkJ%x_MJ3FwP
zAG~#oXNEE+qpamq&rE%SxN|Y213_=?9^X&UXG!$!S|{L>RDVy^{UT(xZ`=N!{rcbs
zFYP4Mu5vS9rh^iVbv(9tZ8)W6s;+J%TD;($IOR)G+3)<}7fmiDef4)=zmsPKy}mh9
zf;eUPO!fxP-bPl=b8DxP{y1{HvYXsslbe@a)Gh78ZXvp`>g#b+?x2SG%Ze{O;6Mz#
z?<|3GN$!=E1Lr8YM!0_F43Wn|FpzXhWZEN3Et09#>?m1p)yn^b%H`Pd5u=Fn9Ily}
zqvPvV=WQA`Jxl4;Mq1mhms2+C=ys}8ea)v-vuVvJlB<b-qs&kC2Rg?D6ZQ+-Rze}3
zQm0+PUE<aG{@1wlHe1Dqf7bPE@k6)NrIDSkpKy9>lE0#x7tf5X4}_+o?zC5BZjiPO
z851Val*PD7wuu$Hjy4h~ENg+Rxp-KKE1t^XYDGMb*@&&wxbdFTA#coZFt`QH3=sTK
z*Z3>$IKlSdu-ZfH3CMuLKv;GOMGQ=PE_d%lE~aKX=yi&xTE+-C0}cC8<(M>x(QL?+
z_(su-*Dpsa=&bx>-Y4wVa?snC9%D7Qn36vm^tHoTX^jN!=Ipm^p9y&q;`JQI2F`J5
zmSO%d84>3#>x8|cqB5Comt<Xt&##d~;#Rn6z-pmdkt0x5u&NIX)vJ~(1aD|&OKRHk
z$5iH7t5I~%;CopRAs%tf2NT9G{3IIbQ_`03$6h=n&&eqQ&W@IxE_XmWqY<3Ls;&?X
zE0clu@X}9SL&OzH-+9+E;{pkjSne(!5_qC`)1IrKMg&^;)!1;>_ma1$5r~3ok&*g1
zaHZ^i;T<rMYH-+G@RE}^ly7lfra8W$EHvahtoaT0o2!$>o&o2ExTwJhxMD?QAI`^}
za4IV;$dZ9l$r|#!QzZFnn(Z7afMljf`6Yvrz_sY6<qns|q5KnMXf(lHIG5?)x|Hbu
z0--gv{;H?Q9MaF~4%kh%1X+?mmmSo&6I8PzT~AW4IY`XJi8Wmyy7`zKbqVzFH69J_
z6{UKO*Bc8A^%zJEFP7oA7djFPM_liFDBsVfio?@fI}JLoP!SccNk7Uc&%tmc_yJ-x
z)GNy&G<9-IFH0-B*)_Uqd(%?pi~2FCXRm7?s<DURw|~q|`bW&i9oBIH>tN#*D}A*a
zzdNc#7t_%HavluQQ`X`m)a5p+1e4_0Hpss)S^pm5Iw>LP=lJpHBJdz==yRuO-8frb
z<Xe&mY-qY`8|h4_Hn$xc!>fEcxlj2n1s&7s)Vr@#G-1dLcf*o~yhGkz>zDaT+FB~$
z=Z<u?+k)&|wi&LQB)UZ*A;r1As62wNfKUXNyUlumH4T}rcgm8@+PLYWlmO~`t$$|f
zsQ{H=ZdC{oN(#252_(6{OyV7Gl>U93<0SDmy~9EB+pBl|_0JGfq(?--Xq}?tbTq2T
zQu8=R?|-`MXpf;64Z_^*-z&%6Y;+Xq^(7)|)@4w1N-oCjo}sJCc|mp9gsI>SvPQP~
z1URqj^9NQU8wXXFC<Mu{oam6;hI8`taMUWZ5(!4flV?llF95fJY~6NZuZk1r7cJNI
zd-t*y1$-JvQ?2?(&A09m39q`nS^()?=>iCz{{zj;s<z-JZe#SfQu0QyLkS$GSG>Ob
zC%y~hOJ7I|G7PB~KmTk`$RrJ!2CEy3d((NI3#u9|KbBbH<9IMz2$+)srm5%ENCA9O
zCw*?AdeVyCWxurgk{93h;ZMUG)8OXE9gjQAiodF}eS#9G$Q#jvCiMe9tJ{E0^J4^0
zG>9Ql$;FVMdvmKXj6psnz_ud_1>CIeRb6msZ#2#%*rOhDL-;vnILt&%8-<Umh<`B{
zF}%_32Y1Gg{|^KLLz|M;0>11O{tcN7m?kv;6JyE$a~J=mhpjs`<8Zng`XZ_Qu{@;g
z?ve}EksVus>&}hy<#26aifu9upgHVDh~|=@u*jqi)+yrDKwUzZu@s?0*5aeYOYiqq
zug-7&Q9o(NZRh%@;g}SS(8yO%wcN6z^$%P+TrIKfx?+OLJ<h?9)J>7jY2pIr1u%Ew
zW59Ul=?V#MyBB|A-STfg^wl@VxX=B!HC6{g)43-Q;Twj_-Q(d~T|?lHqTgj}*<J2&
ztnVThdbD&tn=uyIB&Z|b!2)cu=G%5J0CLLvAKL7sTL_zT70@c#ApD_!L0m!Ia$#hF
zU4Bv-^x9=3QkYvVWhdj}HE|AohbX%@lGa|CB>d`}rk$*)MdC%zoj1HeCw+#3q7NHa
zu`+p_60$*Nrcl*UGaGfCqk4Xgvk1QBD&ois`Ay01w7~&QH}g_bemreEz)2mV;ZpVH
z{o{L&Y=q8vQ=v+B$AvJjUwZKtnhGKWZ#w(yIV}ZQH~(l$3XGvo>wr-#3I9SVOUr<g
zR~fC_Z{N4L(@TQ7eh(G6KZ$+Y+TOEpZf_JEpckMwaGDev)_?Eh2{AcAm?yjG7V6AO
zC^~NK&UdKL8euq+aoyo}H~BU0GZb8vM|<v2De`RYY~#vrY*_wDrx+g?BX;`cQ)5&|
zp0mxJqUj$2uuJ3ruI8Lh8w_`PqK??cJ?K2uqN?$6B;sb|gektTV#Ol70s=&bM3AKl
z0p~EUg^u$D?$WdXa9JfswxF7BTEcG>hKS?G2{^HHB<G=^xE_#+T#PQA-lUjEt|*cg
z0yg;u|3)H_J#`tm<X<Q!_UX9-00TI}K$7|2lV&#R4^CmxUI9EPzMNM<_xRWAy53&=
zD1mot@(r@*+eC4An7yh}pbM-&wl$W4^w){JDv{rw+jMmbQ&;|~V|QE4Zd~y?)jwO?
zI=~|;*wP=?1K~E<_8i-?x_5M^J!^gjtQYsXGZlnTx7MTFRI&t!l#^*g<)%X%c4|L^
zQ76_s_p%C)>Y8$nk+SJg`YQL<0)9=CSDTchF4RvxxRI3Nb9+%}nOy46`OgqI&7b#w
zmkk*UPpwyRGLuK*YbqRP)3hNrwimF}u(YmVW%#4opVHaTs_9Ux+H=wDT}wM?H&=bp
zCjv8{kkDrBnYD0@fj^jmuh>#OpQmK~AWT7ayKq(L>7%+KL|SUNIsJRBpJHI09}SEF
zD0dFJg!S6y=xH2cy{cGY%O#!rVc4zm)A0h*qc%@+bA?!OGvQ>?*1hhoA+>K$PvoVk
zyBMyg@#VMJlcN2ZPL=)-a<Cg^&$y-P;!l=`Y=4k(4Zq>n<IM0<v&)eDQ!h}E1nrW$
zwJ0PZL%L@W<Rj-UIvx*ceD~LZUApr)0!S1A`gI4-a0F1;YPh4pXR`r`Bl=mak7)Vi
zbR@|>;5m*^71OakGe2uw#+=Q(`&w$?4Vj1NCRG^FitQujDTzyL=)H6NXd~<BWMkLp
zZPRCWt^Z%q#?2eMGOQ{IwSlO{z@DiPnz`$)qy1HmZ7+72(M<;qm0KwG*QqH-bH~8y
z9wxW=4@`w$e*fi*7_V9%Qpqlwz}`4Q5}cnr3n{Skq`gel1gU{LX^Li421=x}20S1^
zWoQYYoY<^E;L=KV0tf*!<0J4^`uGt4lS3GySE2^&&H?thC*X-%PUME@=O8pVuwYY+
zd_U>hin6rzX|SHERmh4g8~E5&qvNQ}MJlXcADrq)^<SPs*k$#B<qFfRW*C(M7bNYE
zTl${nwgI_S)D!nF7aw0T2$z@Cx@@>I{hoqxij_QQu3?*B=`dCXmh5mcj^LD)0s^=6
zQvp=ykQow)Ewi?}cHBYRhWnyE<eXnP^(O9h`%~{=IzvR3%Uu~<^`&a$y?vMoCgG}2
z0jH)*h56I-_tiHwL#=J81jfD{$wl^6(nD0D=We+o{(MJ)u9a`GvD;VOL#G<{HovNm
zcJvil=m{sl*Lz6O#%x;3sY_Zzs`gbTsFo28oB{*;4(F4#HFKY~r;}e7=jb&Z>cFiM
zaakW^3Y_$=J1k~GdFkTXq}}SVdQF~WNA4a8out4#2NOB<F>CJPTrOh+imtxIzt`s-
ze*slA8jA=sZIHsgw<)EMjtcc}sFvu}jfPX!Ne(B}HS8S4IaeE{hrWp5LxMVdpI;5U
z$<LTQ8RhT(>wh4Tf9;pPJr!G6chiodYX?aybCqFoJI8@R2J`LDC({NF30=nFN3olp
zXTBRntxF_xh)>7FaL|xg?!x24lK+adJ_M)ylii8TNmO1EIVJ~z!7B#^Rdtvkg>8m?
z*0+@&UBx%MOsmoN35wpyeYE}e32gHLU+DZG^se0`rchC?eGts+Z_LV}g5dPGdJ28u
zG`&=tm!^<P<)QQK2Qu+NnX(Cj@(7T#<CV+plbN}zQFVoG4mzZ^$<nB*?r^WuCsB<X
zYv=zRI(OrGA|@icD+Z!gw=a?MTM%VD_OtC{L1JtRWN1_*>?J9z-s%z1Akdrq<wB9y
zv^wWE@zK)%e;~{U#HfsG^pVYceCeQ}ZalQ;K0MD*qUwhXrclq3N$8LZ2fApocd<pW
z_1UXEnOr7OEr^Pta&&lq+HYEi_7(=&-18UzoxTRTQ5z5?sZB(@`J-<8QRfq0R#e82
zgO%|s?^B>gungRQPKIj9@>TKWa7Y};OxtQWayRBNdslkYWYyTufhvL|(NJ=9<A!t<
z4OOhOh*=03gup}7so@51%2lJj=X5Jh=tZ0dqE^>k#P5aIDp9_b3YU;JyN2BgMZ&B#
za3%zvYkbwWr8;wpwk?%r|BE7Td}But)KmmS?a}f8r4ehSG(c-@EPIA{BHBoBNyRw`
z=dR+Coe|h}^(od;dV+&Has7!uZ<!+wY0&tl9vl{gSg+Wy(pSbgUr@-w4o|Lh#1Kz%
z@o!Ad-edtJnvGE#)i5N!8HO>t&&{@84!A}vju_`mYe7c_^gBT5zH>;C#V4%EGA~sA
znpA}4N9Z6<g8U04KVOolFM&MsdG=-}h;SFDa#-8^@mnRX;&oP8)5Q6&cbVbGvhSch
z=apkrpVr;x9of6t?eJ^yQ{K|YC1W7#^;#%wLlSM6yVk&a=F<H0G=ZA+aY&{3glTMZ
zp3tzeY){bo1HV(NBfvJxAoVG<Nc<*_6?IKjR{Na`Mw0g@sXaLH>C^jocTXpE8#Qnx
zT4V_lZVyx*m=rJ?RGlC&WzR{eWG~L;vj&~g1c9B{eTYtMHflhTN6}(P&=kXg>l!e~
znE-;39Ds<n<JT0i8YiV(NjPyKzCw!4R2;@e%*ZvAt4=a*FzZBpM%&X`WVwU)#iilF
z+YQA$k3V{_8kZUaj1Zw~cDEt7<YCW{raG9lsc4Drw{2{H-v`qqGI|cQisdf))Vv)}
z7qe9QYi*k(WONna+Npfk4i66B$fkdxs6^L`kZ&;0z@|GC<T)KRs&%DH$slO^WIMPW
zYT7FZ`D6-xAc`)EI67Bz>^ZL4Je9qWIJ=D#stF(dHB&};rO?yq5gf76VtxsR+_x(>
zah`b9i1CiU6kh$)+;W1dC54LPwB`4l6`104P-*RLd=nVvRl`oP-XM)zR~<|<e4SW9
z5F?bt!KS5DvG!B=@o>HL_=r&2w`l;AhFv_>dS`rIMLBHsb<3lw%cXS@Gk8g(_7DCU
zAmbUf<UyGzrKgY(Kh43TI`EvIj$sH~&_wFs>bPXp(oDiET}KXf5e~wj4gE!JkKP_M
z){hpe&fsFV@a7^(;>0_;e_o@rsw3SV=)MKv?HaT@Q>G&e%$;#wkLcS75j&2C2;RC|
z($+;yFKwhw{t^zq6v0db!i~d%<AOF+Rd3hnC?j~!9&$@R>l~cK1c;C6oVek?njaB7
zT=dwsw&m%`hs`bKb{AUTy)5!1i%Sk2)^$5Nw=Ybtt*VTzlJf@(41ow?9z)9MZ8Ok!
zdQ~4oOJzj{E#se{5bnJ3J^?HWWI}^MOSdMG;n_u}c+GYhy2OLkNe=#L+y5f4nDGB_
zo*jB64=gPtn9o>5pVEYt+8Qd1AF@3&ZQ;8KgzG_@^9>rbogG4<HzJ!}6+AYIbdg!^
zI^Fd_FfAu8PAFJ=1V~G4AT5+-nJ<%pA?Vo6kH~tJdYe#E6F@bUz4*ZV6u><BfEb>@
zj*52czYH0ad&bJ{cr{Le_t<bL&lK{W4^@A4r#&GzEXZZoO?F_!cnQsj!+bY+^+4pK
zrEqbFwmQvZS(sJ)q3%1*X?(%=^S`!v!Y!FWYh{{pjly?|V5Sh{MEdTi?>@nr=fXMN
z$<l9Tmw`3ELG=;w=il+S?iBV0?9+yUHt1}ETan!gt6`~Xkt-5uRC40_<`Z`Jl66T2
zo>Cs8M>d@`+`5-L;-*|M+g;FY=qC+(q!#I1zgsESYilU3;b>NNRc5APwm?Uo=#0{$
z)&<tKsx_`=Ef47SbNNh7pRibNV2b3y4VaPFpyeH7jed`GBShZS2bNQ#_p*&I&6m?4
zD*#Dc$qIBpms$fv?S{mtlYw4=u8mL+C<JaB8ce$TV%o&*+PS@&wECm@_jc6Ngvnul
zVQx~~w2D!7+|(gM;UwMMe3QVfrc4N{<f+_7304ZJv2}5%2=IH~d1*j#-c9?-^A%(3
z<mfq9FcwrgU3=Es<$9>pvRDS76OX2Pf82LBg8GVjZglCrxwvFy$}~N(?B0^Ei+hYt
z5{`ROe?s8Z8%`OY%(F3p1JN5~!u2WzQdPdSXtA}Ew%s6f>W^McQ)>OF(!ay>HFvk7
zqAa{-Xy|l8GrAIXk-+(1ImW(B=F(LRx3f35*5XH81v|0X*|V4ECLq)RUE>G7<#KT+
z@WafGS%V}1p##`y;K@`95JX70stVzyUhPc9rv_mm+WLshnz~keUg;*)v>vOet)J=W
z!n38`-W^YVap%1bFsG?=81jB&`bsC}CK4cwTK*e#G5n(`eYfs@$(D}CkhtD+8tNru
zCxH2}d}ducUl^COacLi4@?|6b>0dKu)Jsvm>s1r#Vji+*Qw}CGLj4`ft5>6%X|OV`
z2uz77I;PoOik)yU4px+;Oz0_VMZc|_vl9vf2gn^fj&T`^^qh{n?|gZ)MpV`0=B{(W
z)4Zc}YzK63FjU{iclx#6>g=`RT)@gFtB%RwA4fqH5dgqT7+{uEMUrexmMa0D!Y`D>
z>Kb0Iz*6h5?468F%Afw~>D@<tr7qV*wnA-6C4+pQ@RB*)7gn@8xI9Wdl=X_PZj8dt
zVf5YttNbV6eRN3ku&8FHIi2HjmVVfbWgk7+LuM>jj>{-FW4+KSHR~eX<I#Go?f5&>
zhKlQ?>Kt7C`L&a0*8corg&s5)=4iG?TC;_zmXQS}QE^$g->rN2of~gneb^zDNt+jX
zlSib|HYVqcmyf*!NIio;g-$NC7px}xi{-S}iLbj({&Nu8i|F&(&!0C=Gu<W$&WW#X
z#bG^fgh88G^Qx)O^+x&@`F-UFi=Rx!nT|VTXt&+^Y7@I{&y#GI<QQb8Naaj2tmNw(
zme8G*OG{7+Cncr{gF%2t19%SCL=|25zZj8RNa+!UUF)begO&pNHItc-PxunIp0+`<
zl(JTo;r9{Qw{<4?$Z)|ENI?2rvMLR`OKUe!@3AvRV>^lmo!tB&&%dj;$;`<-F<x#L
zHJ`v1d2py!R&Blc6jNtA+d=bliX-}v4Vw!iT4%Zd>AJA;v7WU~`%k*gXe)-qxdK6o
zS>74%q`e;lLl~ZpNS)duVeY)HIyLuSuT%aTXBVIqHeCSa;=$&+Q5|uk4PRl_g8g4T
zB*l#!ez0D??EKtab-#V&Z80(&b$3*Mqr+}mB$*t6uUyA5m8Nv`=QH;llmf1IZMvo~
zLQeGVBCmAEu24}5<};-5krh@cKj`y3Cf+fKQ`@Ef(J4QJ6!&fEmNtj|f_Bp8N}g(l
zNeHh}S5qDJ;PjKw<mo0a+oq~ka;DRIA*q_16cN3gbje7HEEj31BS+WWA>?*Co3T-9
z17E`};nsdCQ5IGq9mme<jT+o{%DcQQs`SOM{+2$8Rn8iW#x;$gD6)`8&KkiClWZe5
z-6JF1!!eAurtheYKV2e*4@kerdv?#a>~hL-ckPR))#je%F4c1Ks_T&P$;YM;7|Dlw
z+u>~@d9sh+aK(TsL$Cy2y{Q`!5lH~WrR|~`k?%GIkFDw3MB3&Xp6=i<m2Lj$Jeo6J
zea+@w?E4q@V^0J)Ec3W_IFHu;34%nwwYqXR>X`3XNzNPjk@gSz#%<I@h}LSrK}YP;
z<=iQjaRu(fKF>1K19;wd%c8!W`yf-VHKcIFO$)~8UhVz{OLiDCP*`bl0vJX{@U92D
zDEjb=o2Ja0gieWFPd&fb`%bPIUm)R5Fc^}ZB(c=qS>Jk205uH)3bFkEDl+!`Qh@WQ
z2?iDZ|05<!0yW8It;i(_)M>zv&A0=Bi_#=Q6_+X`!H4L4^*Fe+x>t*U)4XU+vl(?R
zs%^z*LG2-^DSXx&8`EHYf`zE_BbL<ZCr-x`U9S!LHc(o0an7GcXVQ9Z#14Cg+j|qN
zm3c7Jrye*Rw^@S7ZjGI=&&y)wn?loSW_IpPc8KL_1r>cRzaFMS6w8{4ld9ErZ}kEw
zy~PgQO^mL=4T%FVy;3O@(%-5Ud{FBME`y8wV%?-@(H>G@Q^Li?*30cq&)^Ylu)WIt
z=4819mr_x*1bgdsON^tMzs1QW-$mQAD(i)u!&Qrx4e`OEl;7uTd&$snFU*ln9+gZ0
z?|X$Q?jUGn=<|Zf)!KC6q}c5|q5`KCEh-iJfWVZU=6?0PVFK&K(+A^WR>4Qs!MB8x
zy1r~3m2csR`2;5mvB2h<F##5VI|b)F9J63N4aX<0TRSh%yOp5&JT$Ff!@ayINB^;m
zJ2sF{G0o=~`8q+J6a3GdGg+J;56fsDo1SU3j`A57+>W?<)%Tl@=YzTym`BwI=m+N?
z#<$#D0-K-93s9gTDq_O>^y|i43!9Ne!z0uQUEYW%O~%AYt?eJIbIx!+rI0W9AC<g|
zH}p6HbOngt@U}c1FrkS`x=<VnXF=fovt^@~-U1%H3>*v%=@14?j3XQWvWG+JB`cSF
zKvOhi0-5S^JWUrRfFs)Q{rGynZLgW{Ccko=*GWUkE~cq=m#rnA`}ePs+}nTpf9j17
zCOD!ei5{A&nZ~%IOmvR0{ZJW(zZ1Q_W&hx;h3}W`#0SO~QUxo=4Q&LyM60fqF^zgA
zrz!6hekc4DwEFJGcT{!vRr_k-P?>2)T=dE|JdpjA^E&K&hY=Z-u~wV2KWOcHfis!(
zQgrfz;A;VfL3FcCbpF{6y<aSePk?YyV1FIO)m^OYOC*547>oL)!tZ)!HYK#EHPhw!
zrLNC*^2cA))du)!UoslJg;}~ymn%Arv0+R)#S}j1&%>>51pV4~lo6hEsdi7edsw%J
zd{@(fQgM;1j3e==@cXcbbsbW!JUyhT>ZJuqiue>3O`V%*P>I@x{tWrRaZ1?PGt$-n
zNmozs?fgGmakCWkDu1mhUz-M{;!rI#*uwn_zCXoMu0!oQwni2B|B}|z7iUc9F5v|P
zPbn1uivY6`HeYb?+Bd<Zz^#;wvTzq#x=9Q6)!8D_NTiauF}x0QbsyWNf~?^n#G1J2
zW~a7_Zrg{!NF&9BbFd^Td|0VV7O4oPg!&t<N`0z4Xq~uhYX1k<p_>Ft^x|E^b(wOh
z8sGSApBu`JuOWee^161SNh$Ylx{MIs<gvtOxQ4toDee{K*W{qVSR?@-5f={U`VBZ;
zl>c#)1SZ~P$8X<P$qG3$B8-|pJgv)WA@0?iDQ0EgMC;3}7#`9TEL@J{25<|H&r&|{
zw+_Z^I%`hBU!AduZCLPXyW*u{7W!@%fqb>b@R~mqG@8=bMM?BZTYIuU19oVRu@dch
z*pTc^Vw##Tu2URh1P}z0fO68RE={Q}<pL)Fs1&ATrMERmlFC%z*D*UrQWGics^rl1
z|4*j4)N}{HXtOhcG(|il1JVvkK_gfXrzKoVU8&X65~KFe{{#x^0KGJFqBI2)`V4Tm
zX}<W?>bjfG#5(<$O1++10-^P|#(wf-;dBX#Vd4+p2_x*;SY1K3=onK@4(@cP(IlNq
zO7neEYCkb3Lh+u{$BSg3RZO1o#SlHOqI^n0U{Qt&M~w`c@;hDcjm^1C;(MBXx)9!Z
zJMAKx8e;zlm7jH#8sMFKF_ij5QL;JGYkEXJFe5+~kRI7FnzeV-DIzp}ZRz#mTnMjR
z`<LLy1%^S2z0--!l!y_2R{a?}_6DmfPF<ojD_nkjaEUlN@C@xo+>N6*&+{6cQ_K?}
z4pGxN8Yq*k{q+KGlKpI&U>d-hQB2vlA9t_lFP+kSbiE*a*Lhp}S_i*TYvbh(F{2u)
z{@C2@Yj=J}da5>^TR3!sPB7ci3}dv7JLa7YvnT4AlthZXzF0l3ci7<1c9H9bY{U3m
z-KPsmetrKsW^JWf(zOXlkqvco`3&l}&bhjwjRLp_ct9<D5dX22l`F>0>0~Rp%92RP
zwRRUAl5yh2MKUaU188cJq9!Z4i_;=iU3DCh1pC048a-upNGXlezCB^NG<|~<wy}^P
z@J~0bln<dMa`+1~3|nY6&eP1C!Hj<Pngl#^ojWt3VMbHvkLu|I&M2&fZBy{2L&?Jy
z)3zCcex>tJTsOg(uuu-p)2`H0uRUlkQa;IPVnwg59H7K6zOK`eL|c1gomp_+w|Y_J
zyZ9-5@9(@H2kY@Y^x{ebA8M96YQcU=3@u=)Uj9-gU?VHdwBAcUm4E6%ZQmS;*|}_U
z{%|zojcjTB8zEDOZBzJt+}vZeyFr$+5=?hwds@1{@rk%1TS>Zz@4W>o_F(h&=@O4+
zT&*`F#0OJ)=OEm{^7n?yFV_fTNQlE}_uj+0+<!X{g1nHmnWFuK@U?CyC;76O>`~N{
zW7!?>VqR_A4fD00pgiTaZx7;l*6cY$k;bTO<EZ*mS%=?g`aD6rzS^f{2(_QMW154e
z&Il+J)r7Ezd)zqrrXOG(-+wZ>FsENtVxWij(Cfabov-3rC*D<W<50jc2pH&W>|=D-
z49?GF%90Q0O{%((Y6?XWT%;*&od{+@B?msGq=_;tMm_g!F_M?z*%1(eU5LyVJU!~7
zs~`1ud&=^|yhl<q{ZM@rbnF?2=Xe*-XAYtCh4#uroyMN&WArZFr-D(7%MY&^e2U)$
ztmexPi7}f@g+W6aN9#B~GEKHlpiG91JoCnl!+?U4c*RSTFqb=MKA{X*MBz9jm?uTm
zia^q~vlNev{@wExMX6af<)!_PnhooBVgADml-ZzeR?Y^1_yv!!JKi07-*GB_Qc?b$
z3}%?(yl`wqvU$H>%I(e4Z~L0{>rX5CAGScPX7itRz1B3=8;s(?hE;*S;NLVIKb3~Z
zClordJO9_G^0wAGuIr!31<WA<eF3MP0vc0hKwZSe{O<y&i~dh~NStQ^fqUW>CkQ?#
z2%NyP3eys#tYN82xhzO0)hGAAlb_1HdlJ?*E6MI{7^4q+xc4m{XXECCdJep=eDot(
zUM@Ef0gxI_BQ^WV(4Uxav8k+gw%m_uJ9LvL(X%D;nm)y}tWQ>sV71099<<#$HH#My
zW2nF&g1~PWpxemVn$4vjnc8BLG$qp^N3(x%d-}w)kw5l39>-hWRjva`I;vGAlf>lN
z?uQc}v>RHPTdwJjohP;a2dcyE3-8T*S=1o#ng(}Y%y-pJUbjxUeN<=OOtK%WmhBXx
zRg%6sc_STyT_4fqb<|MnafO#Jh5GV$g?txhZMrrDhiIR_u{x(JcZQ1e$7hh}juvkX
zcI`DU*W0E;g4BZS_wDc9Bk7p!X_1VCo4Az5<kkNNG8&#~HrdXq*=u2cRPz}Vu@mhh
z-$s=<f^3P!>LdToY6quJ&oO6wTn)G$*)@K>&y^eppZ}^crOV%HH9TnHo|iTMDVWSd
zS~iZiTQ<D<ykmZxCVM*MOMB>Xwxql~LZ)7H9pwaZ#mm4zvi3Ldp{yE9<pG1%Zc7!O
z>rr60hG*pwEG<Vq#bF|daqGFSnr7B>fB&Cd7jO@~LgL?yEbYT_2^6JV0q-9(RyXTK
z*yGUqoAzdEN0ppF`83V8J+h6iJQNT6&yI%dP@0Q!G_t5D0ym{xKoBP0>Ah?LqITam
z1)r(zl-x3V`^1F*Sy~6JAAVomkG)&8z^VbsUMJZO6;eO#G&<aHPdmTsend)GWkt{l
zqV84kzUJW`VL9rB*s^`B+0R#?cx$Hz$RxGWl}Ndbj))b7uF&YULzT5xV;3{}WqcZh
zf=4~%o-Rg6p?^4WW*F9$Y?3T3yAJf!rNX)GjR-L<$ddlp=;x{isEVRiV6oyb8O%@_
ze<=$!lx=^kdIzhee#h$Ujyke7E8W;^Vg<<)S@E}4Ekckb5XJKo{h@fdNo*)N;X@HD
zBErEbUEi#i3QD?QRK<(*ph7r>2g@`ZxA2X8A=8S)+;FDZNFE~qb<W~m8?<oY{c8;&
z8(5$CNN763f42y*wKoM>p1C!}<8MFZjl4CJ3<NWMJB7~HK{dp=NIZs1k@-yV{Hw|A
z6mDu^Ld7zX1%dHV>-<ztabk3@pU@>_6?d&?*GDKuGpwJiGkx3jgH}g-X$2J><(hni
zU}3~ap}G31E0Qi6BLp(2r|c?X+>#CJs`-CP4pA-hh2YKQWTg`mLDxCaBB>EnTde<L
zL|~_T9$I<Rae=S*FWwU`=sWZSjSe$~S1%zYR#bCOn5s!EMh!+@B$&DK$Y<$tR0KI+
zbtB6`+6N%pn|ZUja*nqr72$g^dewIN!?q?-b$Rfv%1!3oS~cZWbiw5k0f0b7&;l{S
zmQTgHZ&KRjNuIOZHK}TqzW0#gfoW$TK?C4IP$8O45U6^H(`-7xpbsrl=%1JgylwbM
zp1>%b2pq{kk~O-#o>L@{J@BPt6A0oi(FOSfn?sUR?$T7?0Z{*lIGgd=!F>ShCkS}q
z2pRwZo3}2IGAL^Ao#(kLMYFx$r~DphX=`3f`#}tsK8}()xYVDvOvq0a7b7A@f3Ql-
zhia*Y^lNN!GioY(=%Qgx-&Q6Z&EA$}daQ7`yy;Zr#mK-N)jQq@yy7Drham2Ax@+qx
z0*{i{YMW!kw@|JG1K-ceuoBgR0KL_Vk&{-nWb|WKYn~z!Ay7wh$Qg@zpi!WT_(au(
zhI7~|40{vSButB@ErzO==31_}58BF_N?TL947Snnbx)-jF8O+RZNoE{(y^LNYflUK
zrn_n*BI2?~WCiaRy4D+v#=U0Fi*J3zM-1+KgS~G4>1Lrl^KQQRtsG89iY%2Nz*^6P
zn@>zSclyRvBvnVO-SeNpF?G}4Eyg+(S{9oItN1>4Y{IA0N9<opUa5Fou53I|!fp&X
zZlIYep8WCCf`c>vpvuCrtArKCi+JTW?iV*cl%301D_Tu6^pXx*6_?FiZfzZed4}vF
zURD|VU1_>ea`xPJNj`&o&1Vg;S6?rUMprfJ2PiC7K&U!c5i4^Vc|DOEiSKKL^KpbD
zUHi7pM*4=i)1?#kf)MK6Gp7w{3fyFgU(+IMxW#pfD`hY}IgfExDR;1<a8R<6_xC>X
zk9sD*VI|iQ72KN#%Ol<AiJPrh*ON=xWKn<sCff&^1E<BYo8N~LjAmUvbVqrYJTnh9
za=l--L7aP_nS<nJwl2A*xKLFwd3be9Gi3~`3SO(XxpvIKn}OBw!z**p#j(+Y!PNq*
z4TAh)XAKYQPj2wq1e@TFct=0t+wS+6kX_COeXD8Rz8G%4O$eJn_^4h6B*!#c-6nN)
zg;b<N3SX4egB!WkIpSXZ*EKTk2qIoWh41GyS)?EB#BUq!-Ef}2Kk?Z~D5lQO`0Du$
zaoO?Cob<$sdD?)XPh5cR=&2O%$X(=RJHt)aOrHl&Eozr=J<bY$o?xAOFBCbYIGL*u
zTvxzT8lG4AeTYvy(Tc>9A}~uE_WnMxToLM&+czRUiqbd2zgio;G;Yqb-gM8S@?}A?
zdynLG9Kujr-7jcXV=Ygq1JQ)7Ie}=}MnTr|&YcoOGVR=ei3d}u0L*_=0!Xvr5*L`&
zK;Xs6dC&ZOPF_|Q(+Vy|jP2a(mN9a&1(prWfQ4Y!5Z{nzuZ>HSAuBD|(x<S6t)a6w
z<x7-(qYfK4j5AlnXR=SEr6}B>9{UVLHuj5V8~$SLY&<}mIU)N*f&pS1t^_47vQjI(
z4@W8!H%TD<2bEOkWH|rTY;D`wJdbr#h=8^c>v$TCJ>XL$T-emG((*+~k5^7<k=Y-0
zUzG9xKmzz>7Q-uj{10hy96CmBQ5~OefO6yeD_i@?Y8EQsk60ylzyh7B3Y0d5y%u;K
zdmoe&2><_2P;7$0UBFv_fCvDR!lrW(kQO@<C%EfBqX8f^+`wUr{RDAZq>{B>nylMJ
zQ$2&nXIKIXRR~&8lSs>*gi%mFxq}9^%t&U?HQ9`sWfatv?VD@om5ia|U-G9N;slm0
zoU9cSXlHfKv1nfWInJx7aLho8GvXydv+@e+SuzQ)w*eqdqy;!M9d+~Uy=i)?ya92=
zFs1?YVKJjpOTB68r$gL5IN5l((o!gmTxOaYD`tP-tWLZ^4_J64;rkstT`)7rrSI^l
zNES+-Z5k*rW=n$36}H%DtRh}Z<_VNuYns7)Y@3|Z(X&Z=KsTKd_$dk$A)l(!jHCDo
zT+!Zjk!}2k7$sxV32d-f@B*QA35i7R_(NjrwPgg);Mq^=2AJ?$E4C<HDY9{_q@8>-
zvo`#vE@IzU8uHsH)a#O~+sI!BEfDPIoTDYPXEj1j`KgXqEF$L0hU1qlwcqyR9znKA
z4Ko`ZF^6k$SrzM2ChaKopf$lwcp`5l+XDpkusO0e3XZi}(2n#n+0^I`D3d{I@rFb$
zRC^sG7*awu7Vc+5Ex5IS7kc=6B{x8|V}R6jzW^hq3jJj&H<0^W6u!`))%V$sL)pYh
z&?%i8{N>LiXZ`;`MaG8Wr+7^_dB1O)30yViH;buPF`jyW02!hCXq@#t;pj^7o>O%b
zE}sw`3Z@NBG8(~A_2+Vm`L)8`edN+z#e@2VKuyw-Y2tl4ti%S6D9I)XA+tKVrUPm6
zsXXFg($nhkSS1%a8&+9T5if|<@zoYmZwEG`7P!U5U+J7N5YMY63nL~@{O#>)u-|CW
zie;jfpMBc7c`<w|mfjLFIbdI-r}X6(hIj)oKxP)W$o3W71b_W}nQEG&N*@?vRTNd_
zq3%9^K3JI&dM17wvUg+Zhg~}JX3icF(<UbpN1nN9v}||eqw*KRBU*7mzRB5l>%z@n
zPBl4k?eLa4@CyV5UU5)EU%@Dt4>|~s-k@ES-OfQ=-Mi(YtgmlALDfrGxb<3DQZ$i7
zY_?NvYmlz4LyX-Ex6|&pvk<?SE!9cwntiNW*}clGR|S6a8L6#RCw)S6%P;7Yqvic-
zm!L9ivbJ?luHHom+jlB=?&Mzbw)u5)6xJu(TJxeNP^4Shj1ZNo{hH(Qj3MyAm1H<F
z(sSpN#T{_j`XfKyH>#DnzJQYjzTPqKb!E(nzk)6cm9CQU??!hUtoN)-U>&+b2-P8+
z0Uc_+8_yvifMG7+`w!d;+qL}~7E{SVZ<1Idqu7j+hpB@)74#+$YNZmKAopy!ly7%2
znlf?O01T@1m9oG3l>~A+-_@LK)_(40Uh||&n1i3of@TR<tlT}R6tj9Q5Z$fxMy1U3
z3`$Y~yriC63P8mI(=<;xB>^pBMH;NVjK*O{VS-e*bj)4?Z5Mt>2G-K7pI`=vK~ErX
zA80mkaZZak!`TaT1=N^YCw2WxQ^(cb*;}lW1$8g}_b~bY+BJXyKotTJK-fFMhz7M;
z-ss4{5KFSBaNKN)Hy1wjM_KXdk5-+uJg*Hg0hFF%hPujy7Z(tjq4%=KNCh&lb5q6z
zxl*B@jDyVn%-D3Qhep5je${@CJKF=Ax;OX)k=Z>Pp~+iAY7;xs>kz+TZCj;>>M~Su
ztNwQ)d}!G^tg8-n-cY<x&2t;|@Nau+Fa*^abW_UOQX>(({%G}S>rFLDyNID4Jr-a3
zSTbpc*z+EJ;{UPq?eR=M`u{_23%QIC%H4>JE%$Oa%w-q1xqNa78wrzGmk+tGxnE*f
zE*pj^p-4)(E4R4|$u&ebx_8d+-T8h0+ruC8$F|q|^?bgbm$r_=8<h;kP{iw#2ME4$
z#+b_YcHg?R{9R}G;!>_&9;MSb$eN;iJAoIG&Ab-Fs?$>L@5Pa>1_Z*pv`Cqmd-kYY
z8<`5xlEuhYX8Zifb2a&z8r1S~5zlXRK{+7eTM6*t<qn^VWg13{@pTB3wzh{wmQjg`
zc&B-vEk;LqCm{O`%ITCVGD1wm1sFYjY2y7PYS2@7j2iGnZE638roEQ*&r+$W+EKx!
zQO#$5!WI|1^z>EV6JfJmF|%Lq1}2TzJKQqDqP0hqoBQ`)QSEpTdY_G_OVYa)2#p%t
zdQe7mC>JjTba3fuRiiTLkkOwJr<GMxkqcyFUyMb?)_rnMR@HzD6rP|I!1~nS8m8zn
zij_3^3x`I`;MzB2au@TR2a~g=6!n65IJFjiF&KQ=G(9=PrKvRvmKCP@`g-rzk>~)0
zy1f@{nFz5-N#-T`bdN#hyYBJZW?GhupS29loJOQOB>ip3u(_>VP0Ce76)Y!+BRe7e
z;B5T^;U3D2nm%;Yct(UVXSnCJ`jq>AJx4cU>?MaaOV46O7)4=foeSo2_14a_O=AuF
zEUfy@8Q1MY!%@bQGhN8=csGN<?J!)BTYERg*-%or{&)3q^WEO7uHWs<uR6Rk+N?Pb
zUm<lj&J5sH@1#%PaDvLX`OS}Q;Luu;09h(@NwP{ck^c%mb(JNU&JSOkq*Zw~jAN0T
zC8y0od8pG$(*2vx`ZIYhkCBq0FRTQ*Aqm`ib`V;J;9RPusFb0ktB;uC#WmWOFcTAn
z<g*C2j{8Nf$J5_~g3O{du{+j>aK%z(TH6T^YlSrUTELYEP3wsMR$WzhmIXXjBYZOP
z8v|}Ai^dFFJ+COk(G>M<9N{QJO|Ivtx)W6F!DJL|J<lZIKw@qZ$^9TV)C%QK8!5QB
z6Zzx5?iMq;snVOL%>Cod?J>7&Jr-*ql4Dl%zYZ;@z%G}Gu3dT>Tv6N0nSTT7LbqSD
z;WaM*1^FfLnTMYG<*EoKZ9BGXBZW5Qw%{c@7MpcP28e#P{7lf1AXlSv#ef_``l0!6
zt?}#Xf$Lrw#nIgX9nCru&25tGQ7W4DWt3qE&JC6R2*_Q<jSJi=VN-?HDXO}n9ISL|
zF(H9NU`$FcO+_$ZR+!DiqcnZg|7TW^zzzbizRU`ctOggbukfnsir;bqDdjtl2uQb9
z)#cRkSW3v#&H-=se~io#1{_pe4G{{=(v+CFU)Odvwb%MzhWgAy<&zrh$j+A2&VyGJ
zB^RPn5G=FA2z6oxpW1INqlir97IlW<7^0F#nrDWF%@VfC9Cq%#CX=a=9zyGf`;^Z&
z)~i-2;J94vXUZ>L^v7NeJhoJq8RbY~0tp<yUqxz>)voYo93wOJH?fq1V75dNy+vMp
zeVXcF)FQjCF;qXMuTbV@ODkI4yub8emzL_bR5opD{0_`QuPQU$OQ>P(c`%;`Nw*Og
zC&5ePyz3o9@^<9MEkW0l0~b1V7l@=;pW$v#dMl4#qe5CC-Yku;qRnd<Ud9P7_H_DQ
zZS-s4<6%Xt3OLf>`t9xI;x8J*{jI98sI9?9kE>hxuARra-9OotV=T>M8yuF%;EgAC
zH#Ud0b5o~{!pXDeG<{m%l)O&J+_|TXvKTs$oSajABJ2C$)q7c6ol4T|@{`zWwI*66
zIx3{8<`IpPX9+@_2@m4}zv=PsB@efIuYt2_ZU<6c|Cp>+uRsI`rQKyJu8i}Lf#wd7
zKV-C|g7?V52r@hAmJKLlx?>Qbs5rnwyG8k|&bEA+G{b`F>1jZEfqox+f{OclC<z&8
zznRtX0cA1Z-~e*fbX%oKs9EG=1MefIk?V~^Qsutt%BMOm<+NPw3%1pp?n}S<>7D<b
z_E2-iN<i!0Br(ZHQsy1jW-v)2!H_M0`-DEuYReR`SY=nNc1d>$skEtZm+^*42y*=d
zj_I~SDL+ZB3aM=wR0<HgTRxSw)!6)xpy%cE7f<}3Xdpa)d4P?X?dyh;dRb|0LP*+q
z_W3Rg%dErGmQ5$}wKJ|w-QM#nJ8R0?2gsE&_qtcL@x8$%$Hm02f&2dzTe4b_4+eqm
z;`5~5D4%CF&RVg>?O)Y*C5#}?{Y;{RfR1(REHc+}8K3!2_w*upe8B+%i3)v37%=S9
z^E~6W)~5c+O}u*xA_U^aC{mR-4;eRcA|!WeQ0n~?B$T1sqLc7ztVvklxmDMJtdhD)
z*UzBdL_gDw5#@=j^t&Ul`+{X6%WVyTS2u-DcDYGbQr|oYWfZTw4F->d@%uaZN)_(7
zUAFQrvmQLu?1G7F1=CNI;R=f1{=jBPGcb~#vY2_qStu;3Pt_5Ao5!A#p&V)gW8lKu
zhA?MMSk7y*r9K>E(Q-l)U)WhRK4y{O0j)C+D0%8bT4<{|XwGz-Vz&lToXPH~N_qT&
z7DfPwIOzZz0iC&bj|l;F@r5oxK@~#L8X$UAf}XPxkR-x)Or26?J!-Uhr$&Sqq-gl9
zy2;p;cS&z?ak)>Hn%6Do!NT_{K0?kWR7>_yidgbE_dDu5@Q+$Ki;=Zl=Hg&9PVq$%
za7B_tS|AD3OOS3tjqC+jrfq+nz&Fl5GazNjS=jvO<8s;&fKUOoVnEChK#BNQUESKN
z3K0SXJ=JX>HH`{OP7<)UBoNXg;edjIs~X-7AS6-OP*7E3IzfX=V70<O$;-{xiSR&b
z_2Rih_K6^HTuAUxuNK1*8P>iE4*SaZ%G<t>@$W2n*t^>E{m?tpD{RzY**Avv0`ozh
z91!eOD)!hdWY_2Lt3h@9i<|h&hQ$3Vip)$$(WXYz%gUNfgL(fH>4hOgzK*f4PLoxl
zC-4tLy=!0QyO?^IoPElSAKmAwi(P#T9k4R8SA-#L!(yo3Q<+69-(GnEgkDv>mayBp
z?*87zlqYi;gZr8zDc_6t?yTpx99wOQvzffBg(xVfjv?@$|A%Q3ci2ZpljZcL6LJ&|
zo(BA!$tZT<mJy9H`iZ$%T9sqAy;#T_);#>=bJRg?X5Bfzr!_QNQr<h;&k|?}pck29
zSnsedGH;3%P)T9u_AkvldQOQC&w?_SARf~DzAy8KOgEpr3g4e{A!*l!8hw*^s#Q;5
zq}eAbrv%*|I{nHv7T5$xoaUEF`2*w}9Cd2^`!g$V1@-WGE_zMne&|z){8JyH)td7S
z`YOWTIp4lk964VOk22>{C?}MX%!LQGY(U67Q5d!%C<yl;hy4YdGLw7+RlcJKL*>F8
zL`){Rp0j?nB#wY&Ht&1_9w=t-xa@qUSbH3nV3EWEU8}xiz$4mC4j@R!qYw2ztp95M
zW=ac!#oRM)y?2&epd6}<nNhGI^oT_DMdHZx6h!4sjqU--Xh30`x;Yl`-m%ZiBGHek
zkY<|m8%=qaF``>@UWSbx9aJ%r^nER_6<6$3^61P=$IJd#RFY@FR~vud*TE`#k1a{L
z^lAAhc@#RUs@AWdSjk3RE`R})dpRf={~_YG2ieel>#qvO5K6O5;cT62*~~qBwNm-q
zu-FIJXD`9`zGZ2FOX;a6->PSSi-hg-L1*eEZR~JYtUXI#mCW5(E;JT@<-B-dZP!h_
z_^Vtt+cescEvEc0rTp%ld|Yy`oJc`qJTPi10{ItcwVMd*`)2%{I~<m1FJMGMKSSfF
z>#06+K(~bHD`$T0jwm--pSs7oogF?ArXc%0k4rH>P6{ihj~}XaXmGB`TxrlmKf`!k
zdLd*ji2iNYW!_nJ<7KY*u-rW>8ba<HB78rIYP7Z7ia{wr5y4?HzFtLDcMt@H*4w%?
zRS#C~&MW(ZpSXG(HJm*^ylnp~OCufx=7U1c;5EGnDUzqBVFER|v=>z!)6VYFFPEz#
zv{H!!F}~y7@k~T+5atZu{kDdv&x8DC0*4JXhx!d4i;Fk&P02F@h9wHEZ_ShY3@`A^
z6btY{&3Dbg(C_;MP&@M|{WnJ}-S!%l<_!B?-)r#n26U;8OK`w13ea<*T!A;aSo5o>
zI5DMZ+a|s}ExzSJ-Zy&)c9+M3@X`qcAS7gwwlPaUqh7QiwYZltqANP-<SN!?zF90P
z^_(f}-1C=-%o*UT<zw))IGO|aC#SC}5ey9lG5df!_utV&0R+&%3oP&#ZZNN|9F&+V
zBn;`+sIbU)G#9ju1B5z>=_K@d{Af*o&LZ1D`dncpm#T$i`I+?kXM&fU-HW;;o41;w
zCz%HKwFgIR4NJlVBQmR}no&_XT7fEd>WlDn{h_*@e=IF*wF1SBL^PWbL%!v>98O#f
zY_*-f<uG^Kbh>zqyYu7AWuvf8>#Uzv>wEG}xa^I-m}&j$=1R!jQzPPM?IbBzwclA$
zCn&qM%_G<-@T3D}Jdqh!Z{T?Ah8F%>399ZjEGfzBkreNn>)_`tbi-OanBW~9I`YD$
zGyu@zYVvkkQqNOk&f{BE7L0G1g;HXtA{_R6rl-C_-aZ<4Plk180dh54tb`iP32(%J
zWtzXYAxLC3C~H5;pK0*(vkT#IRNT7XP*C7DpJ`|`WOK#v^=`+R5Iv-`$uWl}Bgwb8
z(z3FWL-<Oiq~!SZ3o@W@O|(OQ<7-xg%PIuOKTmUua|Z6c^(rE14Y4c^STd~2FW1|u
zR|Bjy1wZszEw(j}9PZAf`ncu5<z)Xotl*{{10E4+weW6#>6!yLs9~@V$)<*sVFIn6
zKY&W(sQ&UTXuT~{tmq49P4TU=TD4+vu06Y^P%aUZzj_7{BJP9>J`_biPmbI5jrnR=
zzwgj0Hek5<Fk*0v(y3t@?425T*TUn{>HzVxro%TgT2x!6qvtWXC_K8CPb<s${r;`-
zHb`7U&cJdj^F;GdQLe70xz=Mfae0fT^s*MwGH}KiaY0QyNIYh>9yKTVM8u(KaQN{}
z#?_|SpXNLYZDY3=C;3}oV=6SQ^j(P171nqZO?v1!(~QW6d_pfYj_lHMxR!60vLXw}
z{CHl{*ZTOGL@MPGe|CSvvkz)E|M~ExiQGv~l@E0>`r*d^38;f^dLAgyK2&$DJZi4_
zlETDm?xHt^fun+mC~x5h?khab?{&khk;}iY)aHKvgiAKaGlCf4gPNE4-=eIF5Z2A!
zOpTh1K#%+p@ngk-hxOKaCJC~pA+!Oy9=mK-3Kl-1ad?IhJ7nzerawuk$-8DtXh4~H
zIrY~1ZC__0)Em?MclbP2O<QC4@Q1%#$TO{HSSd0GS}XlO<Qwpu6~pJIl#QtI_n4NB
zE1p}3V54JX(GAu`g<6jW-OxgXTDuw*XZ8YJ)2m7-XoADI;0$xLFDP!swcvW;g{8Ki
z$O3kf653JIne$6FzZlkHoP@A~cVel;lmjKC*9{x7MT|$rutTR@!<A8J!(>)i#hQe)
zbf5IlX2Nmll?$9TYL*6w=Tj}n?&NTSeAcak*b2)D3aDr7CiH1XgDB&wVV_Q%ku*?S
z678h1F4FM<8B!5vyJ%G#(SOv#;ntJ-rY|8eQc2q{S~MbS!J+)%D;Uw%Hl*bQP7zSU
z024-lP{R_)tXGHhWAmv~md=sHgu;K|AE!?<Oc92NQwp-cQK*8$;j+;NX=`%>CutsJ
zD5%1f8@OOALV!Dy5FSv?R275&&KP|_%B&#@NCB%7c%1rpfC`L!b_f=TF!S^OJpGcV
zK&Fv!J-F~D_s7+uE|vKAQ0XYA_o^Mo$=Uz$2A*Ua;|r>^%mK+xJu7MMngSuT<W%$x
zevIt+t6Nyt=GFEvGp|`Ev3wLMaU@!MG%MNu%d<8gH04nDCF!>4f#$2%`)9LmWhPA4
z|9Ly<p8nOuoV#2hYQ)AY$mmj3H_3`?xig?K%YOKMs97`3vXQKfRu;OP<$coO9^wJr
z*=43RvCy7S$(4mz14FZvPWB!2#jr(=$Lm4ydWJ`6V>z%(<J_5xvFB4zv}-3Y)WUti
zXxv^2>he$7M{*r<u5Cax3X)BWC8m>a)2wh*mGp&*oWvovQnVW$(zKtqtgpgOFd@$>
z*8Ssn4Hj(Z8uyRGM*sAKb-Q41I}darVQzaC$d|d0r&x0Q5rW0?8x9P0w`xuh(w|jm
zO&A8n*778kO;KloSaO54ZI36e-9Um5=9_jnsamEsxGphm+kp$%L81j118>a*@3N2l
z(cof4IrZsER93pcQysDw7(y<1D?Xk_aE>k%)bC2h**M3qvIMNClX=ByA=!Zrx=tWP
z(=U!F{(F9!ZvO?D0<yP~XkMgIQ+s&i^kPAzMnst#E;a)m7%!BnIP?L3c3f&pat#-2
z_Bg78Eu@Hxaz~prlxVdy1bJn`+}H5CjK<)3jatf11~r9&n2;#4I#DZmB5>wm#+Xtg
z8>e$uz`iM6IrT|vFm4oy7$v%VDI5a2c%rVAu9UC3R%C^x@5!A97r_>ChN5QS4Fe8V
zY8Hnh&d>0VujQJh$czAjyi`EgH-O3SBL=$_Q`ZmVXl$;<S6dxT#CywTLVDlq8shVg
z|3~50T>(A&P8ps=FSgfAtIdBw9WtN;X$Ei<@~%8~J@6y5LB-~6Dq;Qivo$BdbP=*^
z4-mf=X@bS38r7>N{o;HyuW^*E&Ug5e1<f05@;PnyV+Qs91Z}mi3j?ouoITkjdm>@j
z#V_naA|$gp(K1?q9hvt{dv3MgcJf(_snK?xIYU`_b`~)g74A{QwOj$7c)GFY>2Nm0
zTJ=gGW;Dt;@Ov8Z;S=(@yZc!e%Q(h+;&7V|qZO5nRdy>W605&(K!d`bvq;#0CK(1_
zF4C1i^r{aCE~+fGwsJxgtX_7uc5YKx^tfv{Uk(f$BGh98ksETR!CblQ0bRCbmnDaB
zkCmW9%?}^~k^FT`IR%n=3XeP#IY0zqZu+$b9aQk{zNWd7KbccagT7T~wl{DY{{;KF
zKtRRcSc<-1c;z^Oc69{NA7{mMVEtiEVT1XdCb_4We2hxM3mVspGR>TnYJ{Z_XIj9@
zo_0Z%j|EDl5h{h?{WA63joUe4Iw7z@eOzjW@*RY6mm^d>wKfUKpX9uR<mCgF8L+US
zEXb8GC7@Bm1&Hvj00szrfXH9j3?P#MVX$0NZc5Uvz%iP30TP@b1vpkUq|=y|1GM3$
z{(^5xr3Hv{^ysPwwV$eXt_yJ>9*p=Glio}qDm0gpIq>25t|xNAq)5|<&;8Lg8SVZU
zJepH~0(|W@%@*Mj;J!gPhBzwwrM~5%z=ie6pcCzliXAO;(HiJJR4p)Ee##*ZZLb-5
zdH$po>A9#(ps(nd90QuXo^Q$Tyc_T?nJy|);hDJ;deFqkI{OO@j6UhDEX{(?61gMR
zJD)={p%YuvCHqZeZKZ)T_qJW@9e&XsM?LJlwP4TZd|o!y@oMG+k55f|qetgktxWwt
zIwUPL2&hntl5@_tlW!^%GO1_AuX_-Cv#7^3$&}VSK1nAD8Q-}aSq;8aqwdC+zuvJ6
zN{*Q&?1YkBZ|B2e%GRs0tX!WKeRdjokwr0aGyY6;8g)i}_2(6+>qR^BFoV5<r%9Sy
zJY^i8hueBLLZy3N{BqSapZ)lwZ|$T7<~zmbF__Mbf!)a+a6s?Q)a8WJ3&x#8x$Kn^
z;?`ZcqrL+M^^-)a{DlPB8-#NO-L6ENz+HQ4eyj~th9x~CEhmP)Uz3n|=YB0$BBjbM
z%DjRb2wot6)Wq#*R(I&ll}j<uoKB@EC*ORoJ`Eqbrk2A5qNiv1bPr!tAJN+?<(_PH
z#~#*QN69T2kG6Tx8-}fGqJ@qC<xtX1#nLVp;KfbJ(L+>HpF3&MecWhAuaS`uc6NIR
z2-|kGD@o726FGN8<Z#rtJug<-G*T-~>d-EN_ZpMzqYy^@z=c(ZB;<ckfH}L8kU^}e
zfZCH-58*+2&o2sj7H}oJVzqw+OhZ$H)#QN)uJtk;zm~M$7G<O1lS0wS6_tyhCvXZ*
zK8cn(E<ld5bLl&W#i}MUooFq0X~!l!`AF2*ngeX1&hlT;k#jWB+tgg-ALB{uEO}_f
z%Cx}A+AfWE0D9!dtEjjtr5v=on#eNm4|kuz`dx`9p_aiDXSEOZ&K}`G(`{nkVIR#@
za9dySeQE2$M?by*b@(M6@WUu@@nP-8(D_Ww{IXxOb=lC$S00x~`pMwtuqvPB>wQA#
zf8!tDeNj-@%N%YB_uh70s05Nll<z22(MR%68?mCCJSlD}wXc>_Tx03PxHnm0d|3zk
zC^lAkoVuRDaVS@*8A27DCHy#oYq9QuyvJNrG_PV76{B58s(K()nCnENA0r%^>>=+d
zf3Ov8;DZa-_sR4s7kn2)oY1m#)c84TJk0;>t*_X3XNTL%qH>tS8uPc$0bbj;9WHP+
zB6N%mlH1BEz=G6ywvn}uh(`jmOg<s`NI2NBW_a}nx!@ddwEao;KjZA*e;ibK48v(9
zoi>G%tnH5#I5+tePdSgg(JA8s%wOskm^*h6NDCTjvyQhKKAr?GL~`fQ7>a@ngrzJY
zwODje-^;ixNbS=Ciwn;`oJCwP0P=ZJnV2MG;BzKlf}F7SDY?IsClX#+C%O_k2JZ!g
zcNR1auF#rOfFutC;HQcs0JPEo;%;D$|1)agK~|2aBXCz(V_cn6AdXW4c@{aHslPvi
z`wyIe{T}}RhYv@0urfnzD*Dgi&lx?y`=37*U1sYP{#D;QH@kDhi0o$dR&ZZ_yD6?#
zy`rj3vw5>RqHuGhyxGb|14|pFb?UfSqN2?2x#u<aU;||QhXy4TUOa8ZE!Ic8E9Wz3
zp5i}p*mubuuO|6Vm;Y5V=6Lu)E%Yk{s#9H%KZPCa9+4Z1jPiJJD%v{{9q+a4bE(6m
zoS@RE@yKsAi95+#^<wmirwIYwgEeBgsZEcs80%BmI&}?*3e*0m_hdJc)Z#6XuqM5Z
z%~74rD~C1z1<_3zx5uaBB2=k+CGm%nLE$ea7W;clQjD5b(dF|$G0hDGnd3iSr3<7=
z{TCG9P+qb@w^CtuWV0g!f`4>4FPeIsF^Dc0Mvw7J%14(g-&rVY|A0JN^kczWp4UR-
zT!?c)s>X9a@?;cJkR8Ej%W1tJcgWv$?(3&}(o#<C-*#=%qTA1F8_EW7FPC9SK&@4X
zGR?I+3d!>h+69~|<yl4*JTh`v<Qg25uzJU6_jyHN(uFlZj6p9*h~;;ldI1Qo`ab~I
zGP_-Kg@bOG`AN2}icbuydaKNBAo^l6fsc%%w5xL-&kSgJ1eF_;wz1E})(}G(^db}8
zj}L3gTjgeMshN0`cA9CtM>SxT+;h#H8!&ZE#mYvo70l{mr7S`-^hBb!GTmBdUeYTY
z?QJ2~CgV<`&bakNht-=X_+E|o9G+77{5JT*LdR+}$uAcAYsL%({<G9`ZocA}l~3)3
zBsS8J!Bd!ixLM;DX$MI~0!=QXK&Geo+*NXJ{Qo)G4-muvO#6v)BIL*B;cK7i<4?%A
zbKGtxwzdrM_w7%!wqO36qEUXNn?#8;jNlDVMv}1}l{KeKqi+zpKXHJG1QZ3ZCn=Tu
zmmOSIYaAh4hZS;J&~%gOYplQhv^t&-4YfLDunwN<Q$I<XKKxg-A5}<h_AT3fo;Z4a
zM775OLM_`fs~;(=@}vsgES$Als+-)2ob{QS^gXH@u`r?gBfqx2p~%>G>3wM+R+i`G
z{if-MwMS^Eckkh0-g@BLySh7j<L<TjxnI7#=a-~1g7b@3;coai3yiSgOH9a#1{o~=
zp5e;0Z>G#<P4(1nkIgScv~?h+A)r!rNW5XHLDehCUCoQD0sI>i%2cvZDigMX1Axe;
z5^7@wcQIGXp%oERX($gf>noU|Tatr)=G&C@28CK*Bgp>ha<bTS=_cD%eYwWs^JgG$
zKbA}{hMq(io*BUf7*fuTb0sa`g5{g@vjA>ZaBz#MX2#2{2U@G#%%{2u39M0c0e<I1
zs_{3Vn5L++Ng)^w6hn8m$}kb}^26A^_{R5EPYbGY1}LtC^ed7`-YS|X8kgXpk2}Q}
z!HA~j0Lu|di6sySZXij=ocdP7Ck~|q%ItXrnrWRS-WJ60*b8U<mE8e43<BsqC&^r7
zL?@K1o3meGQ3j}GQuruG06PK~KpKZopb4!7APglG<@y(RIY1Ktp`Ih^fA_7xlT1q|
ze-9EqNdIKP^dGacFOIRDB!4}h`L|PN!Y{E~8zMl>rd5~z32={|;{d}6v-Z-6tpIew
z{3}AhGI2M>9Ha1+IiJwzc}ZTD7Z8TFJ61aWo+8HQ<iL->p%K!nP?%q%SkKurGdXn#
zcK+f(2|E!dQdq)XG^@AKj151@JhqgTV)4bvr!6G<j_c{%75d=?=m)QT!pGYbTa~5-
z)%$ktx;eK$IAQZgNCxbz<i~juiRd+T+Vlf)h4zXn`ZOVDZ~+L?W9eOA(rPOBP}zge
z>zX*~oxzKxcV|U;LyLzqr-z>fZ!wYDshU61W701^P$(xR|3pR0CAR*gY_TP^N4EX6
ztWM%M%vAATS2Yinn(h4Jnq+MH=F7v-OUBrOXnwYZ?XArF>z~|$_?rkq0koU^cynFq
zNwQ_MA>bHiE2<Tz$Q?4q$RNKPz!iz7OBHkmQnH$u{udyC45%DeMZ)tj?gUl$Bbe;V
zcpIP$$AGP8Gj?t&3dLji_k1lfU?>@^_<8#8l!;K@!)SpFc@UjsnLeVutdu|fu=euh
z$y2kuUcTiA)j@<lDbQ-85|^h+j5y<Fn_iQaXN3wH!{sYxiw&6Y?%l{2o%H;KR~8ht
zcMA#!zsndx7wD+`Ab;`VM3a2A0*A})>GR|7=$$tIXoM)tcK946awxO^Ht?nWQR?{d
zNkrnf`lp!ML-|nAF6gIolS#;<VPmJJ(nS_&?GdQ7RlrfedLQ3rpa7Ky!gmUqoF^Wa
zihh2pV;4Rn@jcr%yW2jT+tTZ4n|9S5lh+M$9Ln6g)9NZ~Q{&xyYrA;Rf2M8^Y@sZl
zEsc@gK&!>crU<A}z33SStC<qzPTA1a;`EIhPd<6w5#Nsxjt;1;g|ej&wG!+P*;){(
zo>x&{9xW*pW(q8Os7svleBEJsU#5S<=PsTl{Dh^0cIMhz=u~xDvX>LB@9^<zs~@_M
z@BME`=kd>J<##`P>x+5J`_k9D?Mdxsc3IZS%$VoHpK>h}XEjb82ng|NZwjNSDE>oa
zlU~s30PY)Bo>7prP*o6IE2iTm?<xHSqx2e>mvzbemkm85!?Dv1LDBY}h0+ouo_5dc
zgP}W~Kw?|L&zbP$(~*5THuhDbGzS#pMb4CdXtM@_Ix*>V>x)$dVoWtA#%8{tsY=n=
z*yBvS?UBAT{IKSNBzg(rcI|6z!ugZ<V5Kam3#1ze92{VS)Q+v$E_ZDwEVYUlf2*)P
z#}_pksys=n%*ZO<wb)4OI;Bvp9ubFl=O@+r>l{6;6TH#qFo9g4Oudt$YbEbU-GQg-
z5MBVp%hdR-B9;JD*p*imBO08?5^%62<|L26X_Z2V6I4nC3;~A?KM+YvvewQK4%8kA
zY$nJu_>#{W%3;5tES&jr`uOO=fcWe05=5A<FbDK$34l5;kynS1u5Z4LvIIPvfJV;&
z?IaKh+{piD^8}x3?ZyiOTr-T#0nS#Q)3yckLOtsZ<m;W)0PRzhuUY@(=n7}HA`<Yg
zE6PpBceGP+PN6;qr?IM_eiH7hYJ{!bP~Yk6;P&52MXQe-8lWTozVE{9iFTBX$FoAC
zb=elne-$78)ANPE&nuwIuuNPO&~3=*%DKr0A3HaDxl4RLtF$$@q^5VqfJel^z-QLB
zR+%z<!t-5yTYKwdw^7YkC6mIrBBGluqCY=;_max_m+CC07-c6@XH&ef=y@wx*Nix=
zuZ{z4v5!gnr-CLuyaijoX(3wf!ODKg20O<-mYpmvgy_ER_ckF2KloQfvtzEqwZH8M
z$+fNGpswI$Cfs(W;#MOUe;s0IcUd9Sz{jfz+n`>{xnJ5*tDO}YNkn?YP+(p#sU~SY
z0pq}#;>C|3mPRsFmVCD;QNXU&!5<GIY+LLI!}$FUl9J4i<}Q>$K?=sMp#c7tbtR$v
zXYv6I4p6&Rv)nleRJ1u*@0cHS=ey~+uDUur5cqJ8-PG7NKbl2Y5veK`aU1cn8FQYF
zlCUZ6f1BXK^rD>R>eI}BM{tmLocXv;1yA3S3smaxtHp_3a<4^htArvOFud%ljC=)N
z9jz*jtx~BZPMe)+@gybiqvk{brNv+bw|O1BeakS+VOGEP#)zBx6^dGJHdx=gOKj6_
z`3`H?XK>PCG=6A|Q|+eWaa;W=X*-|;WA!k1IbT|mSHKSbcCU|{XC_}}+g!r?W!uB%
z)mG7@Z}EJ8kKHK|CnS|-tN#bPJ|U-@?_iDh9-&y@Cz(cW*;D^198+&FlXYCOd=)JP
zd&9Zs?H@&*kRMI-AJpC_x!knlutB#6zcS?vfba^E5asy(&(#togXLNUJ9%P8AXbj*
zI@N1TGsFkz?=#n_(-JdJ>K|;D8m~6qe&|}?AEDAeZOAh<YjZMfVa?8a?jj-RRO{_W
z$tUqk_8eFH%N_k|vMr_guLETTf7i~vw^V4a|8_NHGd3euapjnWFG3S9Wdy}IGne~3
zYc7_{XT@)JUgFz6HxH?g%)7al_L-S{2d*l0IwsyH=%Mi^N6IbHG$y9Q5P+Mn>figf
z-29Yjy+=PB%Ub*}{9h0dY&J%=v?Fd*145n@Pqn5~|Mr(D0C`fwjg3e)rg~+jRu*5R
zeElk`H?4_?N1(_LiRKn{p`6pM1Q&;*(BcA|zUO^LV@(z%91Z01?O1tP3CTX6iK&SU
zP6&~=37&9q3%on$mL35pH$l5}BB~pk3Y!5OhV?+iM1s*F<2BHTX=%Jl!xT%uac*5s
zrV|oGT;LT!Kd+7JZI(pL8Z`GCt$Wnst(6{G%U$pgf$ro^*bt(}Yr*RVQb5v1Ett;`
zGpz_>)>{fq0P#GjECIrZOugj=LR+XM*1lepL&$XEviy&t%4sc~rC(%sH>sMOx%Ny9
zKZ6pOgrg7=-cKenxHk)AL9qmj3Vyy?9`m$dN&rL_*rouE85|2tBOBOJn(;Un4ecbD
z1!=R+u%`roQR7^WfMyT`<*MP9rC4~4_#WqH9l0Abc{)f-S?AiUX-r-)XGZ&k#yt03
zKZ@yBVEqAt4&G=SNb0}yF}O$zqc&+~Sb6#rK&z~9_4Ygk%?1zctE?{73P2nCiMCc$
z>!4~uD;5Ip_t>#sdZm*&vIk-ViZxWXoF2KUt4BLvC`Z!SI9(CV<-6zJJ%s(ZljPa2
z?J$5_i0mLKv-x?6G???~8ieH155;5_G2P0SlN<wRqrBm{qwDZ8nyW_s0|BjO54OEm
zwU?mN1%Y*N6`wz$vHlv#sUTfiY!c4#{?%66@%a9LkevnU^CND#I>}O0>qHvsdqn^H
z7VpP>ur1}vpOlTQ$xY6FbIh|dAKQP`v=HWsa$R#(*4IUiF}Xu5F`$SVxAN<ntKyGu
z71`kw5S0?4!KMElY&<C%QwxBqfSkL-iRnl{j;R5Ezok$Bl9fL8_bY)#R#?jV{fRO^
z2B;f44%S^wRud%cUKs~4c!W_wb;bt`k^SC06<@2q%6eR!^B<62y=TF<s%5rZYwgyG
z(B6BoMrqHWq-N$dYt|z4D0P)*O~H|pt?1GK&diY9XMbdvbwGbJXbNNS5^p0tQm7I>
zZIoful-?<azy^~-;zus-^4W`89iOi2b)Ncz(#3T&kfq-Q7AUyi{ThNUFPkkGBlp^z
zP!U6V%rFE>ld^OPxUD2}K7NQCkYput4L37cJjZ<#&v-u+oYSMBEO|<DdvflJ=5?!1
zjDf?{kj%za-{yk~`7e27Rnkk+WbK2K;5)JZ4mb7A56Xl3Lo*(nEU!2Exp}AVw`NXn
zHr;U7S<?V~!BFTI=S&pNQEU|U!TFy~T2OkK1P$9S4C68-xjJd!&b8^^IqfYM7<<uZ
zaX_I><?M(?h^gkcdwOMZ8-e6$Ny8<Y=zCpTSfRNEdhX|R`{fU(*^E@qMz4OP>(zVi
zFltry3A$^k{(>AbI@%83aQcQ_my_qUn${4ONA<sl4IwriJwL)XNZ(w#NN+E`M9Z$O
zjo8z(YZm|1kEp0foz%Kv<Fki%!aK>uEUgw(4I6jwwNC?q1ou7`-z(p9wI9H5_5Zeu
z_4ut~>$4@jTKc3#7yT`+$vA52AVgb$sc&s?X4m6VWjjrKTg;zY7JJV@WbjLvRzsGI
z9MHCP8PC~LwyvAjeAdOIW;b?|=th(kcz`43E42hq1Rhf>s-#s^wd5Hj-3MO@SdIBx
z#jxwu8eUS!+7=tF@x<}>NAlx0gEFdauybreRQ(-&Z39pKr#{k$Vx2vo-G2uHntWdU
zv4ZD-(FeM46CyU*<$5%0B~nACBE;?OQN770BcsP8U)S9;5dEP4fYxfyt9>C9QZWv*
zF;0*ImLa@m&MgW<QKQ~cl;c(@Xi;w|>ja+$_-GC>M{-jJqa{3WC(_QDsHuHB@gkn#
z-k2NMgPmq2d|Yk?0<B^RS6z9MFaZ8qa*!=f+Xrk?u7vJM(YFBd>R^t*c(q}I0OG<>
zM2mM^i9=F(i2?*5k_BzvSCt@62^6RGDCL_|9(;0JGtsd%nbmpq4qqL>KTNqVeAQ-2
zP9J~Uyj}%gW7IX)hWdHjg%wSE6I+)I|FgiqRPW8RNU?I2U2XicRzn-7ezDamWi^|I
zCAyetO;w+;ep|mk_-n!!xfKvVsp6*nZqm#t@tTaLjb$t>eBb|duTPj*!ceWwsN!rm
zLo*_6LMuo{58O)Od+VHyle#6<GVr0rNT8juE6{*7Z69h<F=FVNa;iqZ{?97^43Hc;
zb}gb-At3*-?jKB>-vjXEynQvo?|9p={_cH|<4L%UnV`!ZN{M@J(%65*5b9pPWhGBy
zi_1zDDzBAh*ee7E?t0GTEvQZ5o=z$rH7<Xt95N-FXRoj`gVecs!GqOg(o#R5TVdB8
zNKY*X`hTny{WPVHC;(A(te;aApD0#hchssbDix;>_MHN_xajy=J_sm8oo0Crpq$AO
z02CfS!AX$sH07+j>h_b6@!RGJS#f;^T_Ag!^BX{CO2r%k%26|8+(flP5?om6u<a4g
zRs2M%cqFJU-+&taPwG@zkULfC`8C|~=7ddawxk{tP6?00$4N$6pEU&gxKG&X2dq(0
zGj-3K8xK#|yHiV*b(~;Xgf9{=qt!rDF49EETseGDH8Cp32!U?22&dPyG#IE*nD!PA
zQ~oo?U-1W;U6ZvkYc_hTm4>G$A^e<F#Ktk<WJn0-nBfuGs-M?$AIZQ4dz(HX%W*4M
z&>>_n;AX+(`%tlH&YC}XAbSJv5Utnt&Yu}K_vMpr>%GM*$}vs(B@mfg$1h`@d>BS%
z_zq?VFu~}G8&>&qAvF5wKX>0ilw*)vn%swdMk01oZbHa@6{BsJn+4;=hcY#1F#0&f
z5l(_0pD_Bg&FFjC6NZ3*fA&J1ed7>g2r)z{8l0i*oL!dC-;?LZZsUm*oNbBNJs0v5
zc{8KC*tC<_WwbAETrW|@((NbLpB*X^oh50twrLi1(6rtddGA`3F0YN|)?Ivb<^;p9
z=K7$!bHV*{^A_LkX)d-*efVB=b@@c6mQv+;l^*{2=DsV+<Fm1f*Jk(QaekSxl4+8*
z+j<Q5NOGZ<uXMEVgTR@FVG%7;U@^a5?@=@1>#V}_lrRIfU0FiaBF0iZ%Cp|1I{ah8
zZ_t$(dP%wN@Z+C@|8XvyRIRo%X+bVMonT+i8RWmCcJA|kLHNz4)}g)P^$7GL4%2WY
z*x%6W9q!0IJo%|j%wL}treRVXV3Q14Je!r($iX;&Drf7w!$@h326Gj-t$}I-T5=mb
zsnGUD{M0*YtAdsRj>FkI+7HUP*-Cp1)i0o(O?A&Qx8hkYJ|X9>f%07mq_;i|HklM%
z`muUz9!Nn|d;1-;1?Nh(GvuSniP0)RQT<S~-f%U`uumxTAp6+DA9^2`TyDO%mLF<q
zeSS$kjiT?00i%(Xs|DEqJJcsb3PUaI|F=>JrUQl;l}qsH3R&XWdUMOe?Cb4iX9+mJ
zZS<n&Yb^}8dpzfVS9@V2e6h%1T_RZ6%<0r|ct-_4kMmzwXPGbzEUW{}vFg8U_Xx00
z82~nAVCxFCV%eaTC{`#-SF$VYFWmwEYg-SHwL*d?MTm{n1mQ&w_)N_wI6G)i@``g<
zeJPM-(<X0@U={B)R{v3%8mn~O@no23CL=l^i*E_bgW9a4WmCaVess1pX%1f0q_%}f
z#!?~%%fMK_UiY-S>MVu$sy15`-XSQ(wl|2<>L_v9N0#~T<;UXP*Iy3STjJpZuf<Ml
zE*>yKztRh<f-#xpiG0GSClnbCaua!~{@EGHvXVy8i6tkyy9H})7e?zUuV$9fm{%HJ
zzm_lS{0ZiWg^ym+7W#2mnhNBH)+c>Ip!d~^|MO|?SH540F`Ck`wyHIXU4LhB$}FR9
zQ7-6<bItQKzfWE{EhJ}8i#;E^I=dKmY5BX-JYA%(KiRz^_KG?5)Q(*<G}|omrMBgX
z!i7ejtNd)I8=Cm}*0d&FEuNccr&YDbu1_Vlyxzr_cht>=s3m5u%S}#qJZBuK*rJ6A
ztp_A1UL0DZWv1xaeq%0CF6a<C;2;Tm8VJu-!Hac#oOrtv5>dfp3P$~ORoopsgtYhT
ze|O~?s#PjlafRb4+{k{HxA@qZ)Bh&VM3-m!AEm{9<?jmZ2|nHyyg_2xsQ6}aVe13i
zaIQm^led7eKG$B$<(?}Zz&T*&HIG1jET<(9c)fh(;JW!t;aewYv6{C1x;3k*a99dg
zG*I)}0#TNSeh_#KB+XB8`T|AGSgE}5UdylDS%vP7l14v(4LBC+Qj_{lOLNQNbsl6A
z<=#<ZR+jsj&&Ev?a6n)3M3H7%Pe&a4`RKRsWy|X=wurYQe4*HXD(|5_&usb(#8etn
z*QaVXb2Qe^;MXs^lkek|JWSTbvK!peOvKF9^GA-KwNBA(wfSa#fCxNZEtE~n`S`RA
ziBcYY{suB*;{B-xxi<b2>t394DrpuX@%f<QG`78T<2k}Xg>Z4<Nxt6=k3F?K1NUU_
zU}|VHD^*G0KWg$jy~GM0MMd+?E{7<zy>fShf!fA&*5#t}XMOgS*Z;9_JGMN;&Zse@
z$Z>bn2JB-VQ|Z$vd3IlXaSDN+ay(96Dl>XS)JDZv?9oq;%6T95bR+5Bi8ST$Z<5pK
zFBodJ@nRuUe1Lc_@`=0HBA|#cnVMT)V`FoI+*aLIN{qr<nc^kVk=K$i_;czx65sy0
zy5$phc~zn4%y*~K`}OVobv1#5CAo3W2b+e<T=OFayf6IcO8vt)e;O;c)3g4weG@;U
z(YF=Fu6vi8>~5867Pp2-j3THAe>o#L>F|II!!f7DSjuh4nui^CwZ}iMUuB`UOV~1c
zbmPginaa{$AE2UQ!Ql4ET*t>P!Oa07K!&bbW<H3}LlL{Ge<c%kS?P>)Z^Ptr`!)ty
zp4%_5CRtW}$suBA-6maT0CM?URYh*csl!_S!>VqL>)5HS%04NXOzI&66f@w!XULPE
zOY2~CM<4_WVn&I<ds4WoX2uIJIfv@|4ct$XI*!tykOXWBj{?*Nu1FyMUC>Tdu>~A7
z$8XU9zDkWooKV8Z0-yqxiRV6U>h4!{&UHC8<6cZyw7}gBmRmy!P04%;=z-Z;Bu_rC
z;ZCiM_}@^=ImSOz1RV2#ydFbYpGnb<3fTbpc30Wrlo3POBM{jD5W%8`Cvmj^fl=#e
zu?0{h@Ke*bH|LA5hNu2@nDG%>yD4si^?6iQ6ZtmQQ}5dG;bX)}>P)L&Dx!#yFyURC
z#aHPU0%wj#?KoQeYVAcl#)U^hofcGWh?az!CqMMr0zKU2veOeDd1}?I_RT1gU!yah
zUO8S#8iQ3a!D)T5SGy+X)|;wHmBSPc0=C^g#|!@DZ_V@z8)|fyLZda&oyym4?mCP4
zM;ecBRcg{12_Sy9*7b^9uf+*1$*mmQo%xk%FP)nK*h)5(vpHSj-m9UdJ$(F*s**}R
zeL5$7wC{CZSS9m{1g$heC;a~Xt$fE8nWo!i4#J(yHF{H%5xo~?+|(Mcj59t&di%x0
z?PaYQWgdvptrN-xGar)2*8Ai5Plu@lA_maQwmI%rNxY|SS;+5rH8IO2RdY)`DiFLm
zRFQ+jXkB9tw{?CqLS2XtDrB3TrKuzn&)~`Jk0eiZJ|d5U>=X(mbnI)C^vZz3*FvLh
z6b~O&Vj)q)vka&zgC>6B+)j9-{Re98Oy;41&_{uA$FhHgd1RxFCq4k51j#(jZkh3X
zcgpq1!>=tqPv>%S>1`bsE0N}UpD6uoH+c~XuB<J1<Y|dq+z;YsZ;f|axo79+hx62s
z342&g@PbAvisf?D7iN9d1k%nH6M{JcpmNk*iq_w1v~TN7sHuAuD4UV#sG|hFp;SC;
zx>`)owGuWuej2Jb!8#K5XmgF{OWpqc+XI%s=3QyscA|brB9gfvd`fzPC0+m7J82<b
zR>Nm|r+dX+b-!e)?s@I;3&c(Mc&p17p@8ZmtRrrVqy1ftFFXh8SSp5=HyUP=J1R=n
zo#k4h^ww7N={2>tH>jN=<XNL_>pU*%psoEb_8i`DXh0qAZH0SjW=9UXZp7Eh#StOR
z@*_@x?m@;=E1w%Oazb)(P6JtGSnM5T2*|!{%0oPbx>^<eAe_+@9MrkOlSg)YxAqxV
zH&EP@zwI*@y6n>*w@{RJCOR}<`beU?GCCXfBhzK@{F56u3%C0ePS5)hP!B@N{xuGC
zPtA3==Q^y+hV*;1ue<R94RBwg{*if;_T4eo;GFgFe1BZie?dKuE(`~#ZxmntC-?(?
z;`)@&HMzL<&ok7{a?#D|l4ya(qP4uodY1zSx4rUo1V_R=@WF9Tnh(mV=>!U>SoiYN
z_Vf6asy(#+RId&Z<7fM{6l#c_WIEMRO{4#f3;cY~p}xhm(X?sxQt0uEjU|HWfntA%
z*D5uit5A~zg`TBEj%t;cYn%K$zL6WD&2taMcQW(B;$szRldbf(rc2Dg8skEp(>>}X
zKot|X3esfJv_RUljjj!;=m1U)`9>Gz?a^aw$7eh=rz4(AzbBJ$tzlA*CrFgM2YCj!
zMKLp|c^GWud_I!%b~4H5Lk4{B<=0}5+;0gVtc#LdDu-LvcI`XOw762&wr3LN<i@%8
z`wY*zxUhBgxmIu8Fs2h7AO=P&wC&%@VNXdgIg!Ft#f4Hh^OO9Gtf`-0M8-!T5P%{B
zDp^x75`Ph@MuR0-7!QPU00KNDalCi|FcqgkCfySNw>mKXzBMjc6+0PSKlKROAIHMY
zDkH2ZQ!7)cMtHhxViq4jlEuMSLb+BZTg?A&r!$nj0N77}osO?KjmLR}+y60OvhQF5
zN(P-BfB$aIW?d1WZZHR2Z{0K$2UI~1<s#Xe3lZ!8IENXh^j>7WhARAEhn5;vWgyVH
z@Q$Y2#b9CUyStv8JLcIKAC5#HhO05ssP1+W+HFDgIZ>Bp;Ulzd<+)Lhlk>2Zzt->e
z@FXoWldQ28Xrql+Oo0r6Q%^oqpK0|L{ifo-=xS7P+T&>^MNs+I?jTEB$GVvtXo<~_
z4YACI3qD%s?DI2KwURPZ^I&6q^<%N-FNA*0v{t?vB_G}>i&?DcE8ECGoG1hDISdr4
zEjPX_E&+52LAZ}S^r-H})&2e0_61FqO<i7F*O-P)MXU!zTL-l3>9n|etJ(SUf0n07
z%85P?8ybhM6t4_d?%W8YGu}65ig)CQ(x2ozi~gF0RXj!3z49E@;7S%l3St4|Pydck
zLVDbD0NrolHiEB8rr~olb#IYS;Ni$8<J`Y{Ljb#D%de&Fbm_+U6<M3yW3I4+3jRH-
zZHFq@R5VoRa`AaC)R*`*%4tRGB*6TL!g85TE|jVAr^>*22lT2%%n=gaI)=YP+Xolk
z$P3$Tw3(#4xeFP7@c!wXraCX~Qh0l^qj6_yJ#z$VMRzyu^p@7nQ9-A0vjqm2@w1nf
z=y~>9;ztw;HmGFxX6xov80HA~<5SeRmDUOmYjl3VD%Cwhrh`{I3_==7tSg>Ymg?Vv
zD->Gi=A!SjF7{s1P-8Sh864RY%QcTy&7)`~;g~bW2M49=XN8=(8l^T0Gj(arf;88u
zR2=|Gxmqx|;r=OZ^m$JOO=)Rz7QL~SyMy&Mbde>mS<$C>NmX}H|4kB}bk<+JS#tX=
zGtTg<>3ftmLT5wp_$G3|xrSNrd!kp*geE)@1X6zfjxcm<evb-fWa>C4ug&D%zei)F
zMGfRyhl4ijJorq!A&hm|fcEj<SGFxKH42|<AKC(!9X8HyeGxhDu-@zPp4QV5y>L<q
zrYzKXV*;a8J`fx2O!9>&zJz2ONS-4YjF;oiIzCji`Xct*ZX6q-HuRb*Zv1Zy*(1O(
z?@7AgI)+}2u&ufRZ|zPsceqQgG<?oe6E@m+-R|bew_+k5%a3n>KV?2HJl~a=bnH41
z-MDi1>DrQXuviyD=hcCRs@GflKqGe<$ksY;VYHvHY@p#&W6>CYXA{vdBh+71h<11z
z<t9|>dUmnUzc2nZ^6RB50q@0)?VD{$-X7@>2wJDZPx$Zt^k{Hfe)aL)wWhI(H^&xv
zALmJh3m#YA(TWW#Do3BJ0he7JiN>XMkse@LY5L(;pwoH1d!MJ9%SZ=xhtsgS>jCX*
z`}<e=ysEU%h2*^^nDWdnlt%N{BHGkhPpUe=$zRNs+phGLky6Gth`G*TR=EwZo;ucf
zFsePT-&}6$R&(wX&(?sJ(0LfaMoX%A{ut9D@@cq8+V@l_0}~sf92Jo=?+43?qSdM<
z_6LUllI?r3<+DPsW1YTy!hebqYIyz+@wVp{@x^B+jdT6Q6#lH%jt`4CDNUWP++7kq
zMODjvBDzSxqy6S_ZkHr`q^%m!edQDrLIIj;60;yX3`Z$S>nY%707gw$3eq&klNktP
zuCQW*r3Cy1aa<T7!xj~NMjm};n9eDBCACR5g#4Kv3A8pf=+gOF3Ncft5w_110q&QI
zSpc-<{%egX76A%`xD{jz<fat?>n{!%fblho7R<j`Sesk8|IQT{UtBd%xSWJ-aS|U0
z2F7G?J{(j6gg}Bqfk?;`b<(nIjm@fSA|^S|*T#48lgS6H+V-9Ip1^;IQ7-oRgi8CK
z-KG9mc0OAzrwrVx1oeEcR)M0uXgOPxmWzC&(R8lng~k{{^P|>-<V&m+k0=tlxoL_d
z0G%e}jiYfHf%biN{BQDi@!C}S`a$u4L*9F!wK>Eo{H5CW44*?u(<goX*ICJ=!6ynt
z+s6SFr=GJbZL~*;bEWTXU+ukw!Cyr*rCe<8Q(He2Yp=hfvbSJN!^Ea<ThXeNZ_SlO
zc^vV!2`<kX5Tin=p!LacCCa1YQRd9@1j#X?f@1J6KKr{^qb6<xLFq`;8qz;r@>1bP
zP0D8ba!pS{zmwrjzXMQ+DHfpGln2Hq!3PyqNTnQc4X*C3xyj^4VGyz<gvB~SKWH>b
zC=kNv(~)p43Hhb?y0JEHxd&o>Ca(r70`&8sovd`92|AD}MHl=AoO1@?{{`u0TOk9A
zcs?<X#Eue#-o;!x_Daia_$@ibY_~2;w}Mxir$#Ha^#?O>ywat}K;5l2rH+CKkHnMj
zAsW%8L5YMF4;;FnIDoRGto9<W6*+Vt62lftPHnTEdNp^es5lj2J;lJ;US-z7MQN85
zZ&UJ=+}hWtUIyfb8pu&7D_i_W+_j>I8wpfmvc4K{KnIepa>N@jInua9+x=4(s~x_?
zIh`O-aQeI60CK*^+q<tG&5@IQ+X;C0{pi_jy=VT)TustHvD2iOS1vr1cSvh0Nt62V
z_<4LGpUN)=xD_wgc%5@>hQ2sJUt_xZKz{wgTJ$C>l<c9b?(PR&fz)5S3E?_%cE+&&
z+i_1GX%%-jVsvkv%|moiaP01$`+N3%(;tc5zp7{2D>^n}eeGV#AMHMFQkvu}*hS%K
z!(Dv+tEx*D`FXcfBm`4=XCuQ|*#)~ptTYO+Z_1Y?zWCHWnFI7m`6fwYqHNFMFHFeI
z`2@{edX>1o))qhah_#gF_ih}y#SA4JS6g@wGle~G40SB3x)`81=72WiH-8Ks3-#8t
zD-rv;uSL4I=IgdbXc_5wNuz1KDi2fIIbXciCN?cHLkqsXU5NO9M4e|K+yD3WV~?Un
z(wePZyGEjRZ4o;mw%U~#C6r1Rijvx^#HdXWYOiWhyC@NRv^8tCHR{*-y+8dwxgX_0
zo+jsg&ULQyx^%C<Z*N?GwR?pBw(yd-R<?l&(q4tQ*5r><O2gylYyD@;zk2`hH|1+P
z7Vz(m6exah^*v)((-SXyRhe(_H5F9$bfHgRH_DnrdA6+9Ngx5jGOPw09xXl5{1Tpr
zdY8f*8^R`4qjE;}xS&#rJVzu|Rqxx}>DxqqGrrt$b_b7d)0|7#B5~ha*2dZaCbNW$
z>Oq|*VE+)OfaM1Q;%$%<oTC@9iIbN|A}fQ`L)s@A-uOB6I^Lsa*lBX^_FZ*1;X#K$
zj~AXYEp2{pWsfs|SZ@Spp`?3WrQaAYPZ`QpAHvUBIP735ZQg53G5KvUtu;&aRP+Yj
z@p9{c?S`lS2(C%#?@+?cidDw!IZQx?!ivGAF;7{6&;i!q8hy`cbRsJ--x69x*d>tS
z(lS%I@k9$GaL}iPZlsCN#SR#rrPq#eEA;I7yqwNz`e|GJWBa9#q~^F-&ov^J&;Clr
zY{K??d=665kpV71FuI7YiZ5nZwFGi0fSL!0cP_Ic924@^|3FGK#*zQFroJ(T6M%68
z;JVTCRn^*wDD}=Xo9p|g0hl++b{Zo<RRKCJ5It2H_*>*`ncj1M$nLk?X3IY*kd5-3
zf`0T5n+zHFU73HG)$e}H0@Cjq0Vr^<x0FR(&e~P0UY^(}t@szo9L%O7?na;%wu?Ry
zJ;*aEFIk;hM*O5EoJ7(Q3xW~l?Kf5a_}^;sZkJ<~xxnfr1+7<ZN%P*cedHiy6gFgU
zA_R9#UUJsSDq*he%t$`t9DY%u-}Y;B#h<l#ZrJTgVWv+-G7w-U?ZDAeM#4$I4^EF6
z@1=1$Dq9C5k;^-G+^AJ{Cc3eANbg-%t#E^to)5cOqZF(lBg^<^e~u-<(V?LJ5lM5=
zGK-@$I#~=R@O?s<^K?Bw`UT#T^CuU4Olgo-Ai2kn-u;O#6Vp;*s&O&XLwK3+U2SPl
z{yZi5FBKI11^~^RRw{lxD{d$$)_;`c8+EZlOEAR(bba5fuUzSK-2nt==}(%5a$Q+G
zig&KQh3J3v!^`V|whqQ+7oy8c2azF%h~wqbSnolA^5xZkf@ByBLnUVA9^n6(Mw7}1
z27xcKa<j5)jOCvqjJ0OVy%+j1@tMMUmX&n;C0O(7bjl8;cq))KlXuTn%^lF@UQD>$
zg|RgSZM1r)+XDtQOcd~NiNr~j)6JDT=^6wxE^WhJYUNA_nX1R=0SU*`07ZVU<Xf4{
z3l`YRmC03{g*MHJV~@kUOSrMCNh<ME<1%es!`kM5ptG8EKXE}49r#<4`+1E}yws?e
zAmxo5JMGI)F^Ugmg!*hx5XiPN#Fk^lJ9j%{fv)*N<b(0Oq=3YO!v{MnJ6sLobI(5n
z(JZ~9*>y%HmxX4uG{|0DGNZrs>apHF{Gjf&t;x9zZsO0zzUygr`ZhY^Mac!#jiRVv
zovbO%x!$x*m+hir_m4KAMqEl+I+YpvBu+v6b{61dA4}G`?R|UoeMi(qY+yx6cGC{s
z+N;i+{?@19NHI_6caNp-bDh@IEDYU5<S95h*mr3pU$pj27<+u5W<`5A<cIxZ`dX;8
zs9<dp;Y-br622m>%Mzx#jv!cJ$W@oQ5Z`3%ZA9M3Iomy?dCE`+*6+O&f2ny4&4IDM
z2_n;fqltJ{LYDaa;Ltw6u*Vo1LXU~Mdbj;z$$y~zc>|vtFTB8cqu!uD{x84{7Agpf
zYo1chqdg|Tflzm?OI*L${Q`R^tXW$FU&~7|U)wH|fA!!kW1ATY<7()Ux>Uj?t0Z4Q
zHx)s4e9siFefY9sdF#Ed4BN*O)Hv!X)YK~zg1f|S3ExUd8BFp`LYpsw0F)A6xye@s
zrF@9wa}foWO89-d1s)NZtGJf{uzR^<?!r@K2k0wc`k2={6`?Om?m$Lta_#QIKMOJ%
z<9`tdW)7R1Y*(=sE_p^OMStapD(^R3hD<CzX<E#)OAE<%*u<^B8aHfE0WOn`Cgel&
zFIa-voc^z-)#`gv2rcj|I}rv6r+)=4Y6hDQ$aXDs7K+4a&|3>_nE}H;vN`idOiSFt
z5BDTb`CbTkf6KHlG35<TvhW<{_G#?SJn9E9Vh+!YUU_6ho1Ve)Gvr0@90<HrU%JVk
zVxe=y;ijy6q+Sl4X2uE7rxOar&q}h)$zK4ch|=<IUo}ndOkW8L;FN{=rT3ARYhwSI
zVu4Ux+M{)#mph>F=!yCQO&$k5@`(jYDq~^MKjR2UqzoD*HSsT&AiTPwLO#A-W=t6n
zl8#FoQ?@hyXEp}VdZ#DVQ%5b5ez*L5sz*)2+cA5ult2Vc`0DY?`KSfsfFday8?bHe
zS4^m@^+I1fCU%aL=|NNFTQ8R<qN1lp7WEOmtPXwo@SI#Uq_pyqUBO?)j6p&q$rChM
zgVWjGs0c%u$lsoNfZ0c8gOFJ<1yJ6%TF5w7S>k#2<T7q(fM>E=U9|I$l}r2wVb$$Z
z2$D3?BBKAZIlG4%be;E$=jA3sMbcMU#v|*FeT2yloI!X!BZ}iiGCiJO7y37A?rJg}
zZ?HxZg==L>MDhM<`4A=n!w04?@9MG~A`RG@)!^xcsl^lQ3&t6K5#!1um-n!GpY~%S
zZ&vsma%p`|yNC0deZK{8dSX2d{sYN3o`HL6IKtHKxuNDle%Sm6Vy=60^7T#jl?C;w
zr~U2PRjJ3&C;jtL-?VO#ugo%OfdA)q62z8%b%1KpKmUh#_17M;=ILat4lk9-O7`zh
zAk!`Oj?LBHr+*o&mIl<PW{UR^YH^Rhl$i6YmUQ1qi#(sWLXj{!@k@DbJXO|A>lnZL
zM$b9Qp275FY^N3>nBFK;Fa6Ov-6BJVT9Z9(dD$14E^K4w8j_qOAza!DA8nLoA#*DS
zs90Phwj|{P8~=2BFV%qwk?J%N=}>(yKjZ3L8<BxniY@;IxP?e!7r@JxE0=}|@JA;}
z=1#!9Zun-wKx(R(j}$%`TC8<P6|bscg=gzxf5E&+z152>cZ$EYdH<(m=T;v(@y8tg
zrKtnAwG#s;7mg#VmQ%AbKM?m)M&4Jq^}Yf)?PL&rosykUvYa+G`y(5{aQ6dD&x3hc
z_!N5c0^*ZL+}9p#K?W|eKEUwQm|EtGeq?C{{&mls5-7yvw1qetY7V!DKC(&3jJq{g
zK?jneZ~O=xu(~&tYK^6g9fUpg%oTg986uXlLn%p%OMU0|gmea5X_*$}vgm%@mF$Mj
zQM3Xdg&B)cuY>|YDR?BHf(fQTEUjQKpyL~m(L~aBn|b}ho^AP_+Ob9V@VOQnQIjYo
zg5v2x{UF+v3vqJ(M%*jJt+n8@7miu@C;w2*dHn^??d;&dqpTg!SO4i(nmB%Gd506O
zl{ZJk-O3-;u!R%<dZ*oU<F{-v*n-`38}7V(<&VU7J&4V^o{gD`ucf@p=l&lA`MZ6W
zZx#3LyA=tY*qB5oJ&PClk#ojb*ElvzGgWr=(d^Bq+p9j}ayYMm)ROAD21BC%hfZ8j
zC&Vp+HSg@R>{#qlxb^`IChm1>#X^9d<8gms)rSj71HUy+&+9u(Pq6sh0&%yi#0$25
zm->HEzkeHO{E}hBhQHnFk3^e1eVS6|J<>7Oc48}OenrNM)bc_4RwTM_cOU&RUufU(
zC*aRm=>bB>rQESjG71~>;B{RfU{CBeZRvF)w{&B7)aLhHa}4;z&fS9Rxj>xvzSw#7
zn6lD@?_|yOHDioSq!{KWGHNx&q(^UoK6;eFiY#@FDUC`qan&oy`sOSDtllpVuHV^O
zlr17pgo#x|LLR9`=sAfP`>PlN;vgOPZWV!(1N;e{=wal<X@mZ^{7oyLa1u8FHozj4
z#1=Z!N**U+UJQS=QDm+tw?8rP`r-VY5V;E5BBoXkBwE^dd+0@#lYTFA`&E>)O@HmV
zo$gP2!R3HiSsvye>*b3?;rh``#;Xd<)x7EHOI?(yQLuu1V7)VU{rthcW4lwp_YWG6
zj&IsHwm!6}$!_rpkyN|JSnkMRS|S`U^VYU8mEDRRu-~zpb&<KHM1gtxAO8jb+YsFO
z0JP`jyVB569w?7)6iiF8GIo>>xm~RX6lWShbfnR%D5g!CqTMddthn}a+Hsh%&^ws<
z6qbrI`AU?T6^1n_dcuRVHa{7`I>IhPZ;0Dv7~S9^#;9aZgwC!k6}|1PLK0X2Ny`6T
zv#J6nDYB*`ir;0f(R-8;Q#V^@?rTucszZ089Vnkv+<I-|U-@Oku#_;9LKG}#D~&s0
z)Rx4-YZJ;-AKj5WF0I$54l6jyF2rrlJzOiU4rdfd_L?jW!>tTcDaU?78<`-skM{gq
zc_{~(dsv>p)l6+w_OdyGujfW}c`wtxIUV0F5p^UDnt2agYOo~zoN*8YNK+vbu;}+-
zW&;Z$C3BD$xWHQE`}g-D8iTwNx$E7lDi(x(AldM;*mlraw#n4wostN@Q*fND03YMH
zH9GBWpnxA}P?8;_AXV&P3WUn?Rh2)S@KF!b{Myp+McKOL%xOYw%DV>mTvF=H_pWIb
zhAdjDD&@1vmWVOFh5C}kq?M#Afx)=3szL;Y;cHVW;Ao)$?lHc$8xdMF&5q_BSTGuE
z+(t(!^RV3tky?l)OUeYfK7j}bBKi2Rc%r%AqdgDoc&Q{lp#Y#evK(5n-e(16sHNgx
zrr)<SzTrKFHi}Eh?zJNydWVL(=5}R|*arD0ye@dB!as(X9BRb7)G2G+Eeqw?&IlK{
zmhzGLds=8pt2aU`2Ap)d-c_b7XN6T27Qav3e6Ul0A3eR$q-y~UrVhJjebW14qxlId
zSO+DfO79ZuhjnIm#_TnaFE@>|`>%`7PQ5mNgxCmw<OQT5o6Gj3E6Y@m{$?V>52hXz
z|NS#O{<uS1r*+xX=~!#*?}_!7A8f)#SKqyT`5)-Dk#RYvtoLCtDtRVN0i3>TiDq7j
z_#4)lQbA+(ac3@1u7v3kRsuOfQS3C`#0YGW$Ok|G)y+6uze0KP<_}Mdaj6dTMk~6j
zv+sC))N6OC`?i8Z0m(HvhC%jZN3xhIKh~hG6aTXCCT`2+eez4a&4SbV`M2}kKCK;|
zCOW7;7ZI-{Yiy3ylxoJ54<<j(Mm~jb@u(;;virK$TNB@MwrYMo`YNUWA4sL-v{p&o
zEqgI0qRMM*0#YOCH&Pj>VDwkw$DqllAF3Z9*Ah7!Htg&_`8jx3SmeuTJh3*rvt557
zP&@hFl-pVAoeMffp;!L)@AxBvKjP7@gpSv`8ul)pGt}?1_v}g(+!8*IOfopzwcmYf
ze3Q$`E<%e{0~_D-%KLZO*P3&#q>!b0V&gjkNzx`=+sJ-pdVGx0NaI<yWWw8bOR9NT
z=~P}AMiNTR)H_&lytbq$O}_iZ`T%L3T(4rT5{<&u8T-D+TS4*D<ASApk31&yB7RNA
z+M2Ekpt37)wff9`)l1a7ZnBz%gc{Ca;SCVM%ebcxb!fRwTG>{KVwCZR|6b0r!fyqQ
zj0WM+`Ij^O2%4;2m9H?GZIh?+QsquyRGdf(7{GuG(iX@}%K>W?HVP1~<D2w<@?p9h
zBG3;glbN=_9)*7zM8PEa(WxLi9cJjy4(;S`z!$Y;Gs>tGspU574$1n&SrbIxo@?AM
z^hJ<u<4Ucdpq+uSVOfdNTMn6u0o;p@&iX2Os8Y($>_`1y#v7h|mZ{6M@)4nvD~d16
zt_by!kY@tFq{a(9AU!mJ<UOo#bM*Q2=hR?^tSMO^TUeNf6zCmhiqUw2rnt{$m3i5Y
z8lav~46a<sQmi!umN-}O^lJRUSWlrY;OL8Qknf)x#~e^2lcq%a3yI#12A)OyM9+V`
z;HX#t(GHCCKpYJ~MGdhCO*cJ&>Qdhs4<yN6sVroNXv^5a>dYNdLI>d&A!Z@kaT6k&
zgc7#w$KjC^M*N=`ykOy&%f%am&{h^YwrOIEZf9j<jXC2JOU9OGJZ#hef4^pFkutK*
z#n)#oMe){#i^aszX2aO)ga)l;d$Jnh;F^G8%?r&?drk%j*{N-s*QSI#cze)DZm&#y
zo_<egI`K)ZzbP#zc>?zzXo-6iClRiVx7Aav?;D#O**(0oUbGTgq+Wju-OlNz`k`xB
zwAFm_KG+!ly|j8nBSq>!<@aua=06#>^GI4L^RqOY`G26Fq|V_Uk!0Z=VU!gXWLanh
zojD}ZOZ67xEHTlu_ikCzJX)06{F+z)9P<7~c<97xytI<F@Svm=NHhsaqXgt8mSNO|
zgQeLezss6N@?T=UkiDNWtHBwFH1^nrKe9U|Wz~fv-`;Y(6xXR={d6MjY)J!EZg7F&
z!R@8Sp7;7JM!1r(Y?%rS^NNJ15h6`Nw!R=tDeEbXQP?2wn4T^fshG)XfKJc%<oD37
zp{N94phE$6<-UtNim<DBDQ6g71e!+pGuAF(4hBiyMQl{p``#6vfKEO2C`TctLl*Q~
z1tetS;v>Vll}sBKSc(49?-9Bx@2-)h;-kbJutxDDQD<PUP09w5DUI*dT~N`s4U$Wh
z)v-9I)wNmSD;A;V%P$Ch-OVMnjvRDa#PBI27Vd>e_aX);#4Q=k+~Xh8sec#t5Wn$;
zd{=K39DPwvo1G(Aj}b^$%y<JNyBvGpt;bTR6fW6DiAJ*|!WPV@<N0W++MiFmhgQkK
z7nH8F&M-`EkYx89S2sn+?9c0NhF7)!GA=V>Cd(xk?c5yj?^JMB7$)|7-}zvOzQJfB
zH!Obky}Iwt4bT}7nh;p{wX(GC8T}<|lQnRSo?uD~pjXw`v%EK$nlb9WsbD%`_AIk8
z!t|0ytK9*ERX-m6HHM&+RX;YfqS7{ELM6i8mq#~=Xk7NdpIOiKYT2YvoADU<SUROG
zq^t3=VwIJQTP}oLI_}b<KJ=ygOpvhkR>!GLanSfQZ<pV0>eWnX*24pj*QHfy-l&YN
zOcO1k^}NF1SaiUCnr@vxSt0z>l^kWkfZXuWTIS$;WJTrX5cOub8>?&beX%T>mwLP~
zebrwQ>o0!nT-$X;I<??D+Wbd$%jw2PK1Ws=1l@Ko>Hk%fy-t?7lB|(KkXbQ;3TNDq
z`YSRur<C#jkRj7OyF;!1f0ObqpJOY(j8bO_QcIk@TisK&@$1c2`PP}6Q|SvbXPdLK
zi7J$~7&TdcTv;QG#+AZ-f2QWBj@PflwNXD%%eKgu$!SnX!PQZ~{63crGzoanc@bXb
zPv<kul{abyS=G%8&1jwjDAgG2`y1L)TNeWwJocr&45}fjm>NkTs>!xeccCA_g!sBZ
zz7^LPq9;>zwQACmT3Y1J8$G8Yp{R*!>0!{kgjHN&@c+_3rz=ot%Nv9WN1SZ*^a=OM
zmBy5;Xi40|tWMM1BIUrY29Pu}$lU#Da|!9v<+J5LOR7{0H!uZC3cr}VUC<Gd_np$9
zw5DuU`Z(<E2iMKpep;+6t#RL@p8c2ypDOSZgPq{34zm?a?_K`h&DrX7xYq>!=FJ5B
zmpHuQny;^V-sCqBEGy$d*Cz%lD@(fl2dXd-gB$bFvCgR+DcJh18q&ydlD>mSV7J>v
zBs4SEGQ%T`I!<Jj?m+kD9ogFI8{f|p`m$UxCz1AWld^E=Y93dh-@9O!+LVr3nID2^
zbYo3~Acu4xDd`?I7)!4P(}WqOtRetKwP1=}QK(3&5^e+6k3Xxr4Ly>KWqhtyY%oh&
ztTB8>WC5fGX;4ciuqc9&luE^{jl);|_)B7w#e*2kt6(9@X+S<{MU7!2ru_GMwf^WB
z)j0|x&(76MNElN=%n7}nxYAOpXU@b_{1PfGKn%uqG%8Mr06*a+u}S_y9uHzmL5n08
zHUFUqzHfZiFmbcDZg=~GB$J-uwQ9r-w483RjOYLAs?Z13`Xlj<{dAdx{#h8GYF4zF
zQr#m-zduyZ$G$EuH}Zmi7tLP#u6A|cvCN`XsnRAuni5dyQ>$EAhJBwyZ=$T6Gv;!n
z-dDK{zF6JF6Kd=+OK-2tHbp37O4L9bJt7~^07Rww_wwJ5Wpw2w?*lHK`$$1Af&rse
zSS6tE(nj=Eo*n`)s3V!}Jn#AWMAj!UxV8SwaNWsYs^6_;RBB|BaM+87{-A{$tT*Xg
zwprN^oCyxwuG1P_KS@;-$vGadXBrEExtJb^>Q=t+ESEg5tjI*EY|V+RV~Z<Gkyf+$
zZDE7*YQDf8nLAUFuc_VeM2>uJ0`oZxk#ABHOUZtgeWtnLdEbwBk6Y?P5=XXt0ZqMx
zvf;Q*$^{&Q2L7G(i~a-X&7~?G!@w49zFhZ55aoIf_He2CI}u*8-A>M_v^ibHN#*?+
zZSa01tNC0apQ;FS?%^vyJ6INYoja+rL=sbn6HIv(dnc62*&@kErL3Qjt$bgc5h`bY
zYHliSl7BX^gtLb=vbL@HN3QyM=eJ8vLiaQ~PE-uT%3;0lEm(-=Q`Sh|v;pXnIUP8c
z#Os;d0%G8+iUSybYBW|LbfZezM~oS|x-T8^Ka|7y$tO#Tiyp(Fgjq@DNvoi2OGS;j
zAk}dBvAwPAr?Tk<s4<x+=i}O~#^>6n!s);0qU}-@OsjpXGZvc0EA;l4+DkW2oTnE$
z2p`0sbU;TfZ3&&)d^JwlM2ojnJ72ANwyn^xI?(L~ry^m9StqX$gA^H_Kb3B#85C4W
zHfj_-*#eb*t^ITI7qYB>BjR>GA=9QM0?$RdpZIF{rV8$1oSxR-ux>9Ps=DCfWBg+K
z*oROxqtI${I=^$eq?Lg^TEa)rJ*5C0VuGhy7i<?<^Yre|k>ptg#Wuj7sSJm}6w4<!
z%DH0o$?nFx;1zDt?>!&c*S0M3i@s9&*J-Oz$A{V`t$`3PY0U(Vi%v*gaez-XK@B{o
zwcnE|%zABJGmzH1-BER9*P(V|ygK>o?ekip582)4Hw<w1Z-lrSq{u$dZ~3xv;&&7G
zzC`~convN-g%|op^lV&Vl^2suaPmC3Et7?j$%8n|dT6H-5D-~CGTWNj7I|n+P2)32
zp^NgL$)J7<K0Z&0@69`1@ZS6wcx}b58g8fF!2K%FrYP=u{1D@t+DJlRpsAReuh5@3
zlS<4tJ<bSg%8OS6=$DldLl4rL_WTkdDw7Pl^SV5QH7+HRXv=fwu2w^D@!7Q@mL%Vg
z+a><7ZpzZQPF6ec4bflLqgvEJoO1>vV0tIs%O}%g5<%xem<A<nwW=}iy*8C`?bOyA
z;q01xtBKv$$dY?<v_GCwPjqn_{Bm{+uf6ayv6D<|&`#@FQ-zAyE};{&SGE;?r%*N3
zIQg#F)6ttQ^Nycn1EkY_ej^??Dokk;5XTyyEF0ttN2HBJpAzr{cP&)ZznTSb_Tqs9
z;(z8O1YpO8=8?I@0m4)%P3#OyEoc8M&j<s1Rytz+%o8=?Z7MWWrJ(rk>PbCja=UG6
z@g>D8P8T=~Ra}Z<;;qMMHofS2hKM$1&|)q3T=4woc^Xt=Hc~1@QBg0id0uX1s<7#t
z@qGTrTlW{WWx6KA*M+E+)}D7;MJ$}LK*4*-eO7)T;ob@(HY<8)%dtfw*k`74tSEfD
z+08j$TGdtL)}z@P(^8bYL3rxaY*g=Q9kH=ALXf|(JEbLF+o^ybc(E8OR&7NhknxrO
zLJx~`_$rF^0s0FABhX=oTH4o`=RA%7KWr6nc}rB;2mFV1_sWM7a;Y5Q<MB<t2~)|@
z5Dot7v_ZjEILjAwg69+SAP3LfIPs6$B(0TnVzq3vn^}^b7cR7~CWD_p=o?aE=_eJ+
z5Zi%Td;VT1*;H>Q>8wib$sCbO3hOdQN}J|U<<F7adfl=GOb$0ApmPscw{b9dRfG;z
zYwlrlp7%yNLV1dxO0~R{VXCiS;kmk4YV3X7QpLAktd#UshSJEU$iO>VtM6?e5}8({
ze(3nU=`a&>)>mUbapcdY5o3hzd>Q|M`-x6#X*gqj66rLOljoAaa*uCIK#~E8X{|LW
zQ+?|)*DDTfzai-+!{JBNDHXUMihvR0Emy)Z^X=Qb)?_u{y{BGrO@}&gO?<tpN1exL
z+U>i|uU(zbH!BQ(@hA+5T7=#Y>3lb`xg&(7{tWzT@gL~+16$MTZvG*8GW&H_J&jSh
z)K0If3NtJ7R)j=tOQ#o{o|zDF8gOp8;*hkGE-yYg{SDRrow!VrdhPe0_o5*-AGUTg
zDW$m2_>Oxpfu<$xapfPKYauDv)tq;U{Ojxbu8*Ip(Ux-Vgd5L1sLgGxvJ&iy9nWvu
z5NI59Zb`r3lCV88IsvmFac@<P#H_To5oc5v;aeuyz|O80=a~$ApF#C|p_W8~QvoHp
za?kYh&2cNmA%{Cc=mLwo`l>1g_4N{x{<cX1JU6p%Tx+a_#}94iPN35Z@>sMZE4#l9
zJGG=lA_jvGm?m-9SIx0O+igiflNzn+Ug0QW;0G1{(Vp2$ib%1}H;B&n8P~M=gX)gP
zI9=iA7q->a>uW6^l+M4fX}cemff%CRmes5GA~C4OM9r8MdxaYB<tW{-=}KeidVUwA
zuosNQa1WCJ>++H$@=_XcnopTLkc)CTY79vjVJgZx43|8}o9ueZq`lYSnm1^>K^_NU
zSw$zu6xj`rEKRj#rA>R-#lVFlc}*D4#PR?J5@LJg4Td?zQ=pp6b=+|dnPKPh^FI8L
zPG}I@DOqo-mG!$ruWJedX25h}lb^NXs2W{oRj4dmm=FVoK6GNwIiB-lb77;&yY^EE
zIHQ%i+}H;DWE&3o0C{&ILTl*9UOVMxp7@Suq-@lY^3D|AR&jaVJnX$F5WBtbO6T=C
z?o;X(FA0KE);vLZb%L%cF=@>tfTxril#O+~2sP*22t<GRMHB-0InG0@ZFFTP^!{u*
zFWs&sSEF)-IBJresWl*JA?aSlq%DdIyb<QT-dK+c$bU}Q7&vl)OgEISMy4l39NZ0@
z&cG|ypIZXz-Be4g`1Ci?E5&pFfd(r0!nL}dEzW0rJ`*J#YcKHWoF$U5iB}BWFP)v%
z$yhjN?Z5TddKA$(?&@MDdnt6d)vR*mX6|JzI>JunA$0O*<12m*-m4gT+vmzplce{m
zp18f5M{QwLE4;kkMu3(l*ph<Fl~Z@@?mk3~hsVsmwoKZ1LwU7}>hW);RJ11ja@bgF
zGL2zEO!*J!MLQjr*`r8k=l&IrGsm6Lts1Ji4BCLc_lCBVf8E11zwZ&sb_XpY8R3tZ
z=qM$UWaE(`WN3s3QN=*qoZ?&VZXt;Q0dq2&6p#?f2Q;DT-aa3B4D9&>u_<`iiY{H|
za1Wj_nkmz}Q!xFOCgIUxD!I5-mEpS6NlJ^n8d3^JNXHa=4XW9d^0s9A>0)H205qdt
zI<UEyNWfIKCd3mxfO22ASiMM!F~aQF3KGY<UQ4u@WIU`bD*q|wH{k55>KxB)ZmBJc
zUz+f;hUh{ju1)nk8|8JImBfgt9w}h-Y77?ut)kRug0E6i9-_~kmNX@LEzY?Lg7$aU
zV=E?D@Eg0b-t<yZzaK8OR+udQ+GT%hZ9sh^SP&K41)e=K*8UqUZZJ9XK{&Ln`D14`
zi)j|gtrcf)NOF9R^ueY(+=gT)qdL9<hS(_%cFR?CA{YkHp#=Gh*y&RkbJC)a6y7}G
zXwyl>f_v~H;a<ROB@m#;@BU9bp`T$XA5uz2cc_7W@=rYH{?kocN^U^-<!I=bl4R25
zq`RXYVG=@o4jm#m43AXSS}G;&a`cU57@?bEhK5C_`Ke@Iy4X4m9?M?_=TOjv944}r
z&?lzMK`X^~UnhBku_FNlid>G7n8&VY(8alCl1^N|c-$U~_j+eReL#BpL#6QJtYhq9
zfxg-hekN!nyHhJu=DJgfTm|4TiP>XIXsn)m=&x_scUF}mUBYB-<h4o%>C;xmCiun`
ze9H1ZtBvG6w0}s$Y;0stq)7L16cZ6>#)i)9(6a2&xaI?YWrreFnLXQIbJH)B-mBr#
zy&9lbPY9*fdF5Y5eq-45@l!Y7L*o@co>0=RU=pZPa_ncv@qEb~PIdn)iu9&3bTW88
zqcOQd_FdhSs7rPgoec=cUQVp|vcmc%<U(PodPL6t6G~_PtNQR-V~KsgOU(KgN#`&L
zo2}8+6(wV7W+h`OUJ3*Bn`t*cvT{e3c8ErtqX!$7(zsgGZz^hYSx818U)Q?n>-}GI
zHThyMkONN1<et*{3-6{3pZuWCaU<lnA`fS8Y{oh5k$7>&N-J9gVpfHG7d-|<ag@By
z9HbFSeAsW<Hf`jiTyHY|ztO?lh&PS+Jv%42d6U+^CmOCwBD98i<BOx7p~N@5&qvus
z6|a=vt*_Izzwza+N__?afOAwT4MvCuw~3@(QL(r_h%{8^YRT$9!|)*tmEqvqFekJ0
z8Cabcpj+E<*R|8{FcHc^@1-S;to+2!bs4!9_!H~ulAPb|-uApKQ<QDCzt*v%fN?p)
zpsbVAJ|7ESMeLn+>KJV|5vM*^PW+DpT@IU^ESC@|2b@1Z#W$_YgnRlXQD{_*4{Y2)
zD>H9>NuqqBMbbaWo<1U>7G8v@@yS<@*l}H1-P^K747N8^NMEzCEKK7Tw%<hs(lVoo
z$tnIs6LZr~?F!way%78?m?q`VK743GRjqqfxcwcs70$vj`{n+~Frxe*dA(JXsF&nf
z8(GvEXHR+sNt-_J{DLrE2l2rYOjG20^_ClYoSWN~Uk`jtaZELW%;*bJ;GYI&zX#(4
zA~i+~FTLfg`m6QO!YqhBByu2F&1TR4wQAOv=__{&FvmR~_Eg1h2Yz|SZCcCP;iHo!
zM9oeYq2qmzu;?>UBi+^)E!vZMs&FBkF4JysoU-E?$JAM!)!HO2^6pN@QMerT-i8np
zc!NvTw3$*+*=iiX{KZ%q5O$fw7}L@^`KoXM_8At%Pp0M`jKRfPgFPj}!}L{+?7|7U
zBXH4nkvZ|KA*5(?@zAH=25U*$w=WEL#ywE3m1>_kzHS<+Rr_b<jJ7t7*W)RyG{*S6
zEzU78VH7ot_hz0l?ZFpaq%?nj&?JyO-9X>>TzF<PEprerzs5BAOUK2)Jh(8MRSd$W
z7_ycf3Ak-$C*xr3ENWkDMnyN0wzJTccO^5#Ff6~N<7&+uwgN{mKn@jmM%r6w9*Nfm
zP7h>H^ME^w>A+mT^4&<06PN^5uA3*lyZ0<9X}-+KSS%WJ+K@NcV{>-O<&7k)YovX$
zq%HT#DF;vGewusnf9pSS1foT_9O%UjfWW=18D}GA4`5k6Wf<_01Did#V66*FHa#u_
z0FU>@=L1`nX}&n2HIPQ}9yw>Qkxi_>33v4_ik1m6!956oI3)Wi+a#7}fMW~@@3Z5U
z=c^O}qRpAh#h|_MaJ|hCJMX8>tnqdJd@fJk-Fq`zq%Bi1TiEmACzWgT&nI*)j5^ja
zBhh}x?x4T+i(h`glW!zHzrtHuPpsJCjh{#nb|M$LDFys_;?bmDpxo<XdMW>IOrsD|
zx_sr5)oBw_0VwkvEmb20Y})^2i~e6Bj_0vEx;aaXcamFp{aNp~%+DfeVXqmr5fTl|
zFJ7Zm?7F%B25sXVDobw4S=dh?5fK`n&r2zuF?XlDOjrz|lzJMO(L+`5&-B&h`L*7W
zl^*g(@+r#{_nYZHZ>$`;x87Xa8N84wG=b{LHQ|F1R;CYp7aQ``(>wXd>JnUJ)a*ib
zqj|IXJN`SA7Q<xqav+fRAqlwOJ*&~k(korjAb9$iNeBNj{fTZm9jP%a=}T>xdoKJ3
zqns9~UR}|aa(5u{MSu1IyUS?j)C%_;V592tSFfuFr+EaVVwYe%EB&^_h17#~y1MH>
zQ0B_f;{i8W>%A517m`Yt(g7kNnTm3bCc1~O@g$2ivN6d;!dgRo0z;ob5u0J+p)RdL
zxyZ^bVO7o%J)diW99W-rho>XOwu7$xc1c+QlImsPNlrjNp@Cymkrv$|idK4-4z-d;
z_$+ChyeV34Ha3BEDOt1oxw^ByZfTx04SP7wk@v(kmjXc3?+vT(?mKh4tZB$N>5GX!
zpxjjZ$@=ahvJko@Kj@ZUz1~oT3`^Wj^ys+jsZnsZ*4(o-fl)mmjJAUsS=7xxhDIm4
ziKlKCmcD;v;falCf8%#@$!g_EKi-~f+$!A)nL63g$8=ich!vP|hO9jwKWi|^`n1jd
z_FMJhw>6^SXu4J}owaE&V<cIOC#S-efU~rM(CILm0x+oI7wYQG*KB50X>)$JpVzW%
zW#=|GdfuOhV4sJZkmHI=1bP-o*L4>G`9Nlt=zEM7zwl%*XqWxC8fxkXUaPVQ_#kiT
zpLY~);!&AKk!Oce{CN8@z{4*B%`2z<w`Qm%W&_x~G*lni(@P}O2C)F~=3Po!Fs^qq
zzAoZ5D%n!mm1lQfRHwjJnWM$Z)2q{UPdQi?sJY?S@38nL0FA%T@)ema`JM|E7hG(E
zL4DAQ(fzq*oW^<!FN;n~sp>Vgy0wZFr6QK~w6@ui$&H`ZN!z3HpKfcS<Tk%5p@X(A
zef^Vh>D|?{dd(MB6%kXZa)PO{JPLTiRaFBFO@>7e6Iv&RtP+``$|m|}`p^|TTk7=)
zTlD0-&VQ8e=hQKK9-GQ8k#Ct1nQ85pn_Ocl)0uI<auUu&>8P?9Jua)1>b&EZ@T%Uz
zk&9-g;tua;0d5)JEE146mf}`0^_D{B?!~EB?l3X+Xl8l{Y;D>iN6PQd0W}TtoVo4D
zF8rE!0afntNP~{$j!|sN%J@BANYx#`#)dIjEa&TTGyRA4ulK)L{($#E=F+ng#?trD
znPa14r<EIFzFVF~7KJjpyk}(37Oeu8=Pg|;_j;@Dh(Ql<q20+C-t$VHq817Fz@C8#
zFc3Fxn}m?jMIa1VrP(YWvP+qqmQQ#gkc%A03o|)4kyvNdluK4Pqhu9xrO@G%@;u~Q
zKgFSiv&u9R8+73$2AO4ywDC0lU(ky7++Ri6F1;w(tq_G1ogu=xskBk35OZ)9L060Q
zUS;fXv>7nLBWFvL0C~fcAJGE$&1g*&<*`vBm&Pg*362mN$%4Y{q3p$c?RR6#V)WFU
zpCLH@TVt<&OdcKVA;yNvRd;!NL3UA~y4b<e+&FN{R9r=^;mZjP5@&x-wLZ}DI<!~=
z<k$fb@SN^jdK-^Q`wYRbnlGF|tpfa2VY*`@ipKU+vyKNJ3*pP}E1am#yw5LD7L~;0
zuO3&V*xel?N(fxa6UN@drl$ZN-`uB+gYX8TcLoskF(|^pZbsA<0m43*Er8OkF`a3y
z57b`U1g~bZt~z~RqE!!@p9c&)yorrR3RSkDfCMV=?^fo3LSr!M=^U9`ZPx4BWQeV}
z8vG&?tgzbH$vhPiwwkM=MPlNu;L5bS=5AO7|6;N@B%}T|EAgXtoSsCtS4riPC%i%i
zoKaNS=u;X7d$ju3$#f63Kt%sp_&Lo2!eO!6i1@eVSu_qyNrU-*Uu+=NTa6OcT>9+l
zxt*i6X3iPVB593TP_V`fkX*^Aw`&Zl6j?m2-W1pnJ~1~I&x_wkQ76B<_`zr{53tNl
zN1Mh88JT*ubM(L1VQ>TcR%F>@ymhkrkAKpgfukBEHm5Ul4OqSL<v}45uWz!pa;a*k
zytw%xmTsTWpqhSLQSy@1V?>3ts1HQ!$_b)qHF0R`44lIs_B`~ERsS5;Ox)@uRQ{+7
z?|NA3sM|=Il-0BJCm_2!mjTU7)E2ERa~G9jbjL_Vqf9(iFxoC@j2zDk-7^|23j1`s
zxCnn0aSO)og98gJT%bEQlBxHX)-1|qm0|sYvqR*GO}MJ9<|ELCCAt;$KZBdpeMD4j
zCLf`kk5~Cfj=3?&=T8>qR$J!n>qXmAi+`kgxY$<TH~!M3O19B+X$x4Q`!#wq^21Wi
z%HPAk-ye>WT_A{>EL9T?(x~Zd(Pk+8oBjnAP7|ZLKl+l3$7z<z_C0uN@acL6{G04X
z?TxUpaFwjb^3APX5B*KFYfX(O2aT&Vw#6_Zz5CW?!8{F4A<5&|*)4k+3)PKiN{R0o
ztO;%MLOm3*bw9tFJQG&ORD9%5dydN3%rdXB*T%P<+blmR8UL}>XqbVYo|MfxqhBm)
zp+)WL-YM~wrVAl1QkIz@_e+JzV%05ovh3bcI(<6lx#ItS!HuPO3A{k4z&(s;1xqX3
zvi)+j_=ABpwKzTX^2NRy?}G0nMb@`k#C~lHfA%Eo8+^Ylck~m(?~(%9Ys3U^*w|!8
zX(Y28ILu9Yl&2G9dlY7LiGML&&npH1UbDaaRB)Rxt^WmO=`ZR)o>EeypV?aEj`x}h
zL4VA+3>)7`Q(M8yE9v1J{z78Wm8kUcFMkRHF8+QscJoX0Usxf31`fa7!y*e~m^voC
z_NP94ET*q7U(DI?bN<JR1-vcm0T3pO{$jyg^?P*yqjGX%!*;9Sd4K2&>=s#@>oX;#
zE@y4J@a>4o1Xf0h7esR)BR#YX(2eQ7gprhrOz?{=$?8nNH9p4{psWg{7|}GN;bHsm
z0WP9NT!K)-2H75`x!zGT_61eR)g+4R>>bp(5Gt}0SC_FJE&$D7u<ISv^l*#sVIErM
zZ4c>7`1BMEVV`iaSI%!M4Q3*%>)r8iHWqq+gDase{R)`XEj#dvSKyHdG?wy`#nd!Q
z(vRfDib!Kc0R5M?G+0fnRr4L)7artl%97d*ZAW3&d-TYIQ|v@h3#Pn|*W-)v#;S#A
z7H4s7co`^#oi2pZYyCZ72BEZ>G-IOmnIx*hd#zHs$mR@!DS!k|XkRFTJPnW-+4HZ4
z9NQIczpt$Lno>pf3{XS${<EV0TNm2=JAO!uEp0JO{CDA+<=aLO_JRS_{zZ9E3?|lp
zw1zk5@N4IVeV2~JQw?e3CE}$v4CkbJZFC5wW4Y6-(M&h40?aH8G>o)B{#t()pjjYi
znVqCF<bmK3AP9Ef2m~d>!<$F;O>46+``Y^)7};lBS+CIKKe8>1kaoN*+WGGDl}VJi
zqf$|EIC<$+kU_W*d;f?U%>9NS`l5m1a8Ok|08cD9ud4OTlS}`1Tg$;t2*w*+4)>c7
zVFt981QjzP{UBrVEIdaF8@|^9k0T{fX$6Sh1tf@Exr+G%_&#w_-%^I_4({$X=r`bj
zY5alBvg9}ZIYXa#+{(5mK!o!dP2H|gyk248BM!e*z#yB_fAGDy3f%eBOY#p>t~Rzh
zDS9n`B@0OGJfM$8F2XZRMh4Z@LAIkS<*~Vk2E;+H?Nu*ILhQ3wr@I71sSzFYO?WO>
zGskwRC+jRr5{E6dRQK&#o@h;@mhHNni1@6Li$!i%th>&CAjAu^e19)!Y&%RmWG>gO
zayurdLP5ziYGR={^Q>Ii;ZQukumhs;dAi7j^|@w%3+4mvE+sH(i>B&f%#w~i18FDV
zwIQ;LvnKoFFFV6~hw)sxl5O@O%6Jvzw388iN-K%&$}DFpw9aoe9pgn;T(RtyB@)Pi
zfqT%Y!4N+OT&|6Tw|m(m1*aP)x|J=n?~w;S{xNcnAA`?ti@$I>@wq&;8wV|XCruYY
zqKIlxWrDVk;Dy$n1t7>G>qB!15ew+NxlL^D%6i%JY*>p;rZ3wLe@rTVKkV5rGkK|;
zn?3Yf_cNDXmN7|)cN5pIq^l$!JdfZwENJ|Qx9d%BF&Mv9?_Zk44G4BmA=;$xm+uA9
zezQB0L)rsZNr#)i<_*2q-|JMsY#esZzzzzgsYjXVs2>|+7mG@}_&bxlOY6`%k1erG
zxM9hbqyQY1G=88We!5+aji_7#+0*2%XPdlZ5}B&3f?Wy?wGeL#N`J<)Z$Ss!a&0Gl
z5ihNa1JYK$cEI#6`}IxG`{u@H32)tlm6f|ZI`{hexbSkL&1~Vr%q=6oC)tps_8Wls
zW+7&%NW?(dJgO&z*?v-xyqy)rOsw~`Ytp%|u&be~hx9XT6+VDYYzI<q+u$hCw$4KP
z_YQDo>t0q{c0Ov?U=USkRVRpH8G_-V0|^qh8yC5zjXJ-jl4`@TY$=bs?RfW+(Svzl
zPL>@)x>A&~di^u3l@i<e1sbi*x%;Ey1J5TlStqMJ)VUfNqHKEq_#<YymuMoClfn`K
zbO7eldT0Gl!Cca7OS|r=67kd!r3`}%OYC6{Gj(z!R?X(mHhK3KW&5_aRk&8|REqxO
z;6ld04s^&}#a`M`)m6>4oKW8Vn!&ZTl{umv<r4C$aB)}Mm6gd-dJh+D%pNAvxP_DM
zl)?w!w-YxNe#<t)$~CfU8>{9KC&Igl;tZm<4O$((g&o69bqv%GobSNcQj0X~1@orT
zC3-s3&lT5O=-t`a?2VWmUB&{e%fqTjV%7Coz6<2&`5s9Y;3TI-<e`-cPvX`q23xx7
zs=48e)m|b(H+Di2qm*ch90F&{c^JU~(|$&zxWYDA&S5jPxJqIy`D@mny5DJE9q+SR
ze@ueZIer;p3E)K?8h<cAx<1*tMw>jY0WdIq#I|{hW3>n%>w>R-Z&MA-b7EIatTD>v
zVSSZJo5_K!R~1H?TK%JMiJ{hSlNgFWSwixPzgf|yDSbrkrOP?F*p=%4Ggf4@=(7g;
zC3;v~12~D|=Ga%11S+x9EZe2Jwkbw-)ikrPl|(PgA65ElDWh|?Nsh+MF~LbX!O6)_
ze?80Y-+W){090QftmzVP5dn<h1M<K*K9F}SQb~^#zUfr=E`{ANz0Q%XF<wYk`FIAO
zrZ%BAN@koMbL)d8EyonbekYOd7!qjIplhX;lY^U?Fv$bzB`IvWTeQ+A(kD`e8vi|L
zl*{^aXcUlP37X)Wpk)K(1@2ycm)S})PNF}*y9zY+D@Lib=h@+_Ve*}RTz|81QsXc-
z80HIjd}Dv1wR?IMaQ9~=isuHYc^6R=JRPhTu~9NS-I7V@z&$5HMy$7$Q=(EioC}i?
zRbwGOG|qQxyyK2#wy=2;DwAr;!I`pN#Mp;87fQ%GULZ|>mGUz3-%UU=*tRCryLJ&P
z8n5HtzudQU_(gP={upATvO`vatQ34Re@_D5<6hsQVtMPQksZ61JYS9d%=u*|Uwd-x
z+loF`ur~36TZQ@vjHg0#wWF(jQXEz`Sg|%3l`toZa^y4bX$(b4Acf&p%$U><G0Env
zU;M_GR$9O`QA%)28FpshLNPZBpVeEp<6yz|o;mU1-*c>-rdnpkC$H%B9y)hA@>Yc3
zab9Ks(G0rSvZ|54Qh%m9g`+Q}&ErjY(k~Pcvl_;v8(orf(n97sZwshYa(4-WpdY2B
ziZkg3Z>W>G8}mIZB+1-E#DFKV&SIC`T{f|1yTnFW$Y))(dI#0cy1S;ga<+P|;hJ)O
z<ermrSR;c%T&K|lEy8<Zc^+DGq$1xlGcAPB^icy6L<5|#$qj0*?7*n-C8$9Go#WWf
zJ31|UZx|uq#D7;-^wqr5;Q7i$Y&1-b?&O3usGO(SgH3C@J|++iV;j5FxxAdGu(h_z
zr023wL#lTuT+q5Hw(YHF3jNa|<ezvY>Z4}Zl@|ByYW=84-in_DkiN<vtiN93Y*a~#
zX5rl@g{7_3Eq`?Z`2lErd1HL;zd5~%2vElm2EY$0O0cq=k;DZoZ~MqL&$tcZV3`jk
z!NdUgc4gWqz|5Ddk82z{=lru&rt||~B^(5r;hf>^g6)Xrpzcb!hyH5hlyZ#~F0ubW
zJ;;7l%low0(l%&7$1=PZhA{n>Em$!3#p}|DH%PP@cR$lEZ&d4wg$F9ws8-e|oL$n;
z54W?F@pI=8mGMH(|G4xg>-7DW!<DV)Z=C*!{b-cmxGrq14)gtVY=_cud2sPD%IPiK
zB_2q2)f11lr-QggAQtvrX15|8uf#DsNoK!dZA@XjPyGbf!pS<2+1tc<wQdvm?tQN{
zY~)J#)P<V72!rP+#?&#3Q=@=NM`2b+P0*a2ot89Z@D86;3j#cf37dR9$eQ*2zrdjg
zTa(Z0&xa01p`)T}3DXW=)z-5vGx3eopVo!mz*(zpdgd1em$t{Mr^s1~Ey~}x0zSJ2
z?HCMm8ctO2)O=Z*F3KM$nwG_H?uoQ*jN>DD0dsiNBbe{dMGcqqw3hYy(fk(Y>ubxr
z9>61Cm@^@h(PdJRMZqd(g5f6g$StelIx8Q>DLgX3V+0xx>NqbO*Jkc5W>~4JF7wj@
z=&Nt{O(tKT4)osil(w7$b59b+l)$X7W5}#O_z9HTWlUjGRnd+_k$o#L0EC)_$7=vr
zWk|acC3Z+DaO&V>#NA&O3J9Jt#!D=9+pD1p(=$Cj3EA1ny9$+AoOM_b#GUnH&~}{*
zate-|qNNbZ&4dU_|4Lqq8LP~+h>;Ql7+|4P(oPe-<M@i{0CAQdE6t-t;O7tb!Kn`#
ziPAad^Ua52lp#L!{9AX~Wt>gfa(>9}+U6-6$ltjc_dgFOX<?7P3URe2U0}~d5VvP6
z5?G5d`A417j|Z|hK>`F93Q+m}Wi#x|kyzsWm|&(Eg8;@V7E1JQ15ZJb?GM`t4=uvl
zUXdi5j5DqY+CHfO?qglsmK`sa61VFlc#P)L{NV=|y!1t148dWZUcXA75GM@`4UQ-q
zuxCCc-x|o~9J11yO&Q=e2K&xCtBt2%NfnNZL}Tq{d)t(Laz-LcZ?yy-?&JO+=v6JP
z%XLJh_}BqkY9^)PLUgpbeiBE^eAacmJ38`YiYhOyjZK$N>?k^|DJS~nSWG>XQN>n@
zE4E6mW58yD0jrGIiWMEykr-dG1wWqG(Vm-YxF>)<R?r=-6107K=xKKxy#W#Nqkc8=
zp3rk;*<*l?M;-S^@dkNa9g-=WD^HK1sp4(x8x<4vUzB`gIw)#<qR7fZKJ=1ZnQxrw
z_b3Q;Rx|{C^ehm2FY+c_q5_0!b|b4TUU}%~nk>#rFxB049=f6KBg-kCzqyYd7OnJ0
z!4K$dz0oKq{=TP{Szj0HsAgHQUV~7(yh@Ss%J}bWD3=Go&OoleGkJ>&HoT!4S+%Md
z`}apkO7T2Del)W1&H-_bE448%f<dSUQ@rY`B7;D34TX$0pf9z4FzBm&<eP3ks;VOQ
z)hFsx)=B!l)1KkiCsT?wBy#!f3wwDJ`y2!px&bdf*qG}cMM;{u_pGKc$jGM?`gP@a
z`#xX&G5m+8X*td7?19zQ#m^FfoE4!_)Wfn$83bHkeMre{cmBB8bAxJ41MX#Vx)ClK
z%?w0Onz9VkraI9C-R1Z9F$yw%<P;y!4xX_P9pVQzv+Ro0Bm|&@ot8#j0n+WcMAnPg
zu}O~L!#LP`Z1bTp?b9y3>1HktXUT#`XJKWDN2;jU8cnw3-i_AVC|CCC=b}FZO?Kxh
zp0UVY?-5VQZS0LLd_JFUmBpU_)Ec|qF5vmF4CiqV#;2>e@!}S?dnH~^>jfeOjPao{
zb|_--xQ<8b-^sb*v%ljYKiG0x?55J+Z#6dc{)o1nrq3QmI%Yj<6=$@}CQM;3N~t`E
zOFHZ>TkixIm`}`d$$XKP7ZEe@^maq)XZ3jhd^8>{cb$&0yl*f}oXrEtf#=6{1gS9s
zKtknnaGuPnimj=za4K1`ULDL1ayJqS{hzIARytiH>IA2aWsyX7{PE8yCzrgkUJyAc
zAN^I@QfBca$!_ZDl4N-o&-H6n(1Yk1a)ka3H6=)yDOQ0u#X?OALM_c1jzhQ9XpT4;
zn2#U~;3X{O(0J%Ha*R6@Cg<))at9+p<Uh0^4>=Qb#eGT1hLSt3vH5l$zvVMseuNYV
z&P8;YbVi|{obklyv@kFcOoCtEDyjUvv4R3|n5;Ru?osP5U&gjxanlgAris6DnxC*I
zoe&HVj%+zYe)7{#CJWG3T)cH1UAQ_q7TVM(N=&mC)3E;buh2250XcT1Ilw?{ica*j
z_BS-+Cd`Q^tgyyS_;VURJcxZz)?;y-6W+IH5j>y0IoTm=*il*_TfhKdoPc!m(tj5Z
zPEIXmmjJX6=RhfmTV4XFb08aYhSJ;^pwqL_WyY|B^kg1$R$CYtH{v-*ij!smtPnMZ
zK{UnLVJ4d;Lv_gnW<2)ooc<BE-SCF5f7vCzs-3xJsNh9B{*79tnK(Pan#z0+YfX)^
zrgoE<Rm=*IG%5%IXs`Lq6Fksc7H_8&$r{3!QbH4!!P@cdD%s!1pRRQ++x&`l6f@<y
z?v!R%%IEfbefn?qsP53(*SKHgr50!A1=}@aE2k#=0P)*ub(K!M0tCF!Y-}Ee`xM++
zN8n1f?^cU`oh9Ep+NNb!ao~@JpQ}UMP|=yDbkavUcOdRMpJHTUpMox>y_SyZ->w}Q
zeuMWjk7`x9R$(JHY)*aIzH1RAX^~R-{{lJ##r&Xey;mY)D#g03#cY}Rxum0V%Swu)
z<rPO#K_p~VG)>IQN!}ZU+IsM7pc0mhH$zh5w;UwagH}c){KtxHmkzvS(a4zG+AEv)
zL(??_-ODQyPCk^?G%sr=S7{$OE=@fkX&dj*;t7|au|>w?n%T6qL<0a)n&nnChNa{|
zij__UZ-;`t#a3Dl3~5W^y%@yu=PgJkn1NIzxMppmiry;7!D}MTs`n<ZDdMo6=R|fL
zQ9oLXT%_zbG-ni&2O!Q(I+TY76quxnVNGs^v_>q*$%|0j4_b@FvTxH;RpE)jHO(4!
zll3|!JJ`LUNN%*`Q@|rVD^es>jH-+b;=LQ_c=BHR8p|AH61gmKQb@a3jAx?zP}?O(
zp~HBpa|0Vy$>-*8TIE!8S=E)~byJTl5*!B~LrMY0c&v*g3O@HVok9B27yu6xfqCbe
z05AzOyL0svz)#}9r7MnUh+reqoV+;20FIMPZU@r0;6E*9%~119G!!tJVW49P#Wd0=
zDJn)FG`(m6z|+C1MsgkvE^Aa|%GWJ`Yn0YyA@R>zSJXbqP6pjllh(ST^f;uoAdr?J
zeQD~di;y}}>RNWP7*EU}xv5)?vJSlsLORg-PGb*`m{P{u*O6D$1~LSj9<=En_h*64
zJ+3BDW4*icS~jRHtxWjKx5_=LD11&`jIkdIk9T5}kG?~k@m4J&5=rxP197RR?@v+H
ze2{Gkb2(yL9FbK3GPnfsnu?O<k+m2o#UoPQ1&j=1@vDFhfR2W@jBTO10z1=pmHAI1
ztW1!6!#%02xXvok3`CHFwDV7338o=~<vet!t7Lmrl!m$k{VFoYp*-}>WjzfdLo1wO
zsj}{r4l;Y!6-Rb&9gTZv#ApRsM9Hs4lzr5WI+L@xB8=y-rP>8{MO#DW>sFOXU4<rV
zLm@LsK*Xe^F)0luQWBHerHnJ0PW3EeQQoDDDDP4sYohS%pSjzQ;xsjQu)e6-yoqIj
zm4^gZB{V@KW%MSphr)XJDf`X@TShV|FhStcJrTjk9N^Li$)*Fi^ag@}(gC}DJJ4N%
b;2+MI2(ldVDpkXFpjQJJBAR2zGy(tF-;ugW

literal 0
HcmV?d00001

diff --git a/assets/sample4.jpg b/assets/sample4.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..ac65ce3ba02bb6bf2d0fc2efc024b98f9b340468
GIT binary patch
literal 210824
zcmbTd2Ut^C7$AC6XrV(uK+w<=5Tt|-8bB~~2p!R(Luk?!u+WuIq=NyG5}FhRK|#eq
zk*4$}A~-050(M2mGUGcpsN?Q_yYIb?=H5Ri=brq%ox}IJ?@JJ$xv7~cgg_vmH{c)i
z{W0pD$x*5w1X)=@3J?UbLmUWR2niq-@Qy?9F<}k>Cm_U_V+a7#5Ziyjc?dj&0%Hhx
zCPI}tpAUSdBh;8M7Qh!yQ6V&Q4g}piE&CHjiGeZ5-p1M$g3K(9O_a2()Zh^qGqB!|
zdzdFRG)g(tHws5oR#Vo1)KpZpbX8P!)iiOcsxY8Jgdpi8`cI6I0ZHe6!C|MReqaZ1
zJT1Qkqn-m;<v*-XtNp_0c@6}zfnP?esHv;upng(-Jeq_41!v|VAj}qq$cjAVuivWA
zL;V`Rl!yKWcjmGDn$wfV_TTU3vHyU<VtE`tV8FXP%&+-_6^I|>fFD&nzsAR_cz?lj
zReV3@02-^e@JGNRsYWnq0m6nL#?N(N82R7tGEl$4=wE9tF<5?#Yk~Q!zs>m#XU0#Q
z;WaV8%~4~n10;@o&LIBO2oyV*2ESl*^ub#IyfMF|sew~d)m2x~RaN~@!oQ6{Zj$#P
zULz7Dy7UFUfB(L56Fd%m|Ni|Wybf5lZEI<y9Q+w&Qt=KbE%fsa#$c@k@2&6uyKmtE
zzGLEJYJ$_y2noTNsw!*a+zi1m1ZzKxGa3oo1yj!<q%{BmWw4b1R--b(75*0K43BT|
z5guoN8Su048-iGW@)vajNap9WcOVJovz%c50&^Y&$$~rljNsP&*(3xBH3K;I$GG;^
zeF3N|x(>kb3!V?J3E$;^+;Q+7@K_r*6jk^s==&2c7c*nyKWr&Bre>BVz!^Xg7uk{;
zdKd|~6A~H~LD_GFb8&UUv5f+}H~`kF5aj6{8D?c~>Hx1`W@3zs1e0L+zt_jvEh?v!
z$hJ7#f8+n3Ab#(#h$w)nEtswD;~VJ>;1mD{ABhTs$KL=L>lFmUNZ6BLBLIQ`7J*^^
zEqE7m&K699VV{stA20{TIm{=-2Zrwe_;gIPFMv^&08Wdc`W^xBG=OD-qeG|w{u{v9
z5MR$oz*{bOJj&NQ0Kh5$=8ABj7y}sSFFTk2Z?M;IaFp*+fF}eQ9}bI+p!x?y;qcxB
zoT`qFHjeCjB-l49O3BvKJIFJ_2WNaZB+N4uNOg;5_$f#L<}D5|SyNR<Q&UMz8L<D?
z>;L%WuUY?R0G936jQ4g-nIVLS{~P=7+5g54S3!{WCg4rZe`8)}A?Ru<1c}W5Hzrd7
zLA)stbY<)}dW2zp35bdc)7`!M$dMz;R9|mpn4!O3|F08%&HUd3zx|#v{CmH6hui1t
z=NTOwg@c*ueK`1VbObIk%+uQ!r}RH7@&EOMzoGRT9=j;Me!daDq2Q;SfR<51{eiec
zeW+2?!=X58=>JZ_|Cf*bh64<4=`{cZ8UI4qKa?T9F)W07ISio%IU$sLIrs+gOKz4N
zP7rK)&O4uN={<nK_sr}6a6z5{ACZw%e;hn)OmV<@M@PiKFt7>u!U8-oFNB3ep&gJk
zgol(MHAoB6gGi7GWDeOt4v;H!2=aykpb#hmIszSo5}`CG1ImF4Av$y(Vn7X03)BW(
zg>FE%pnm8cG!D%`51}P!1=;|4!AIyTNM6_w+z3I0C}Jl<7NLkxLue!RAWRUJ2nxa#
z;eqf+gd(C5#}LVg3`8D+j;KU5AX*V!h+Bvu#0262;xS?k@ec6?i9&KAu}BG|9C9~O
z8)=BNKsq1~BK?u!$T(yQG8<WftU@*+uOM$B?;&T9OUMo6CJ2_CC}ET&N(rTnGDg{;
z+)#d~2-FD_4ON7yLbaf-p$1S>s3p`3)Mqp+S`dvxE1~t!WV9pN8y${5fzCwJ(e>yn
z=sV~s^keiJ^uH|JEaEJREV?XaEG{hmEJs+<Sc+I`SvpwySY}vOSU$3{vI?`xvud%D
zSzTEJS>stVS<kUvWbI*{U|nY2WMg9!Wm9A`V6$QKWQ$@;V>`>%#P%oKB-;wx-|U?1
zJK5FPP1)VpL)nwqi`Xx)-(;U;f5!fmgO5X&Lzlyb!-pf5;|vFb;~K{p#|p<+3_nI5
zV}NnM1Y!~~C72dWA7%mbo|BVPic^Qvj+4rnz*)k1iF1(iG3RG40WJkDV=i~DXs&Fo
zI<6kB2VC#Dxw+-IN!)JSQQX<w4cxulkGTKl5#&+kG2`*!ImuJXbA@M;=M^spuPm=2
z?;+ke-Xh-1ykopC_&E6F_>B2H`A+ba@^$gu=iB5L<X7dl;Sb@@<iEf_#Q$7?O+Ze-
zM8HQNMW9-sS72EXB`7UuEa)wmEXWY-6I{WvV&$=B*g$Lswi!E$eIq0wL=<uoIx18u
z)Gf3mj1ra;HWLmOJ|o;FJS+TVo5VK5ZNA&mx3z4W-1bRiy9i0dSA-^VQDjEsizrUi
zL^Mb=SM;jrq8M6CLCj9<h*-H;pV*7-*zJ1Ty|>f0w{4#nM~W+mQ^e!NtHp=KHzjsR
z?3W0WI4jXB@nVPY4$_Xm9R)jX?pWI?xO2}=>du0lf9_nz3E>QJ!MGCK9o!p9aY?dd
zv}C2^sN`2Ec_~M!B&my1i_#crZD~L00_oe*Z)A4LSjildX_T3l<&f2u4UjFC9gzJb
zCoktLcS`Q6+;e#mc{BMq`9}E#JU4zXJ{(_#pC+&pv<X3kbA)jPq=JS*fI_Llm?Bb9
zQ<188PH|!v%PyT=p}VSf-B;pLGE|CDYE*ixEUav)oTPkJ`PFWj-EO;ccMt6TS4Bf5
zScReTP*p(HTs2Acn(8|>1vPKAQnhJyZgo@jlj@!7Z;6UTUt&3NPD4<`O5>EqZH<34
zwKO9%TQt|SWVAfA%CzolW3_FyGqneGSaghZPU_sy`K+s{8>QQ(`&v&)FG%l#-gA9<
zeP4Zs{<49Tfu}*G!O|Yw9*;d0dzSV}?)BVTwf6~0n&d;OB|S4F7zP?P8NM=7F^Vwi
zF#2SyYkb_eXCG?c{(ZE4qb33-jwW=IN2XGy{-%wlZ}w~KkKf-zW+7XUbIJG3B+R_c
z8qD69YndN6zhl8=VQ)dVcw(tw8DV+N3Tb6-m2b6REo&WWebolCF|*0HS+te64Y$2+
z$7*M5S8BIPQKcNC4A=|W|6zZ@{-cAD!x@K%j(Ept$J<W4PVP<(PM@6jIp;Yqx$Jg1
z?sCsn%$4fe>Bi>f<W}ppdBEgA;ei!*4fizn*@N;2j~*O4Bz7qHQ1>6ae|Y|J*@MNy
z+2ex8H%}YSD$h+XGp{nQH{QnH#oik}B%cDGHD3eYJl|)227Y;d&;1Sj^ZnNX_68IM
zyr3FW>D0G@<iPWRpMq?H>Vm!pI|pA3VGr>L=?WDH4GO(;c*o(RhbO`m!_vZ*!ga$7
z!{0<$MASwiA`eD(MqxoRH5`qPPK|yXvnQrB=I<j;M=l@bKN@;;I94GxJ@#3gX<SV_
zI^H|}_A$w0iN_uvH#}Z>0&&9Y#O;&PCsR(YB<xS9PsAh!C5|MiB;_T2NOn%XmLido
zkg}Y*KeaK9CoLlF{wcjv=TD<gQ%{ektEZn$|Ci=NyPKhsQIzp5(>wESmTFc>*7t0`
z?9nrtXUcO}b3$`wbNA-f=ke#o<t^u1<zFel70?Pk6#h{-R75N~UyLb^E?z3JEa^Ne
zb2jJfH+lelrqrnPVwreZdfBIQKIbOON#!l)CC+D@|58D%n5*1hd8JCOs;HW^I;MJ+
z;lvoI(XMHz6|2pt{iiOpZmFJ9f2TpKq2Yq~g)@!F#^}biCikZC=6%hbElMqw7lkio
zT!Jn|Us`YVXq|1dXuEw``*KUWY<pRUU<d6A;!5n5H&+9$E_J$gPIQ@f-M*%Gt>e1#
z^|~99H%hyOy3gFi+)TOo{m-~RH+#Z*UflA(wS4=J+YftPdZ+K$-5Kq(=o{=e?e85h
z9Qbq4VDQF}?oiiV?Ymd+Y2LdstTEg%qA}7jsyTXPOlz!jTxa~+g#N_M$-R@eruI$s
zPn%8Oo3Wmmn01_;yYGH~Y0hWv`GepGZ|7s?zdlS@U|FCo@+}rU5`R?n7vZmqOPWjF
zkM}(ueq#UR;j;Jg#?#2B-&RsqxmSyx?R?hoT<!UdHPf~6^#kiG8(|w?U!=X{e_8el
z|Em2p>Gkj%mp3bKBi??0m-$}weccD*hufRBn~y$*eEj-}_P6NY^`Esr4}5X{^6cx8
zZ*1R6{vrHx?O%(37ruvpO6(zI3~_T{I5F(poET0X9&TPioRFY^fS|nij%_$)f~tx#
zL1~w|uK6ByO;fF1N+f4PQ%h@#Jw<Jgo2Q$thq)ca23`cg!^0!UFDNG@Bxj?cq+#>_
z`}MsC;%8^6XG=gKBq1a}0>zK`eh1nKavBzpK7bdTtH3V=5`|`AWn%|Og9#m=-GM-%
zkZ2SO3mW7zh~waU2+hwTAgOA|Dro1)CKZlVOU$igmo{p>B}AEcBctvWk;H)!-X<a{
zCMze8Cnyj#G_|yKbdC3!nC>T=ncF)!Iyr+x%iG7-&p&_~7#S5EbL41jTyjck+Nsm&
zw7mR+!lL4mv-GNJMon#9eZz&e%k3Rku6A}^yWM-IuYX{0=<ejy^vvx2xd-!)pDaIJ
zS$+0=ZT;=L_a8Pte){|K7nm+k+i#Kei?aWTE`C545{*Wo*<iX5$RjY}{Ad<QRaOB*
zJ2ua7K`FIFcC1luW$P^tX?4mQA+LxDjIa#xvFuxzG$v*L-v~?kzoP6HVgJ!J4{@PD
zvCWU-hxS6|>WAu6s}h815WWJUOrn)l$pIpi)kF>p5(^R@CQu=-0$T(Pj|k|4u$>hs
zY(+dpY#Y`qBjLK0^1L(j(4wvIj`IA$^69Pw+k)#<ch4&YO4|^mnx2~Qw0TvN9(OyM
z(Kf7I7zW?TqO=XF%3+ZJi&Yq!K)cPYKX7p?<N54JELg~n#AyOyZ4tj;l4z`3BLz^q
zEjvM-)JycSMXqL!BP=YDML8@=@B;u<EEgVL5NqdwAtY?AX6XqpgsVzSg_lmso`wqS
zp{fLoIfSoej-vP`EDcSN7(7%}!u<KIi)1apKuMN4p71|@X^Di{gruRd@Ja#zH+w$Y
z00cY)7DfJ`f(6@Sj)$s=5d2dBD^(@}U~yY+$Y)$pHB||<Z3D&)IYCBdZn3t4rB4KY
zkKndZD=?7eRMKQ8MmVdz1({NVvDyYbu((pJ!r4}!q?mcI{kn-9%v5KQ&T1?>07yz$
zh%&AMXoE3G`N5H&{KVBT(W~%avK-qE#($$0<_+`Xoe?G|icHK@HlS*qadSrK-&$87
zC}VlJ_tkmj3F{^FL{icfs$R*RicOxT#FlL~@@)oBB{Gl`B&HmYG{8wF#}a-CfpnYZ
z0Sn>BFdl)2xr;4;?6!C*La_(JPlDM8kN*_D3dZ^aBnI??2Dw=S(f^tc7)>A~wEvu^
z3Q#YwRjDdT%1hYRA<JAqg~@hg-j+-`wpLXUC`thT1&QtK=6Dgi$ppVc_&z_?0Ax;I
zH5T&9jI#^l?yTv9l=1y>lulOOz=L&@83`f8>j9k@EY00vf#m_s_Qqsq-$T>8&WKw9
z4_}%u0iAD80PKfg5r{CkObCaW38RM-rg>xVh;S1*1W++IrrJP+0H`YQ#{?AShuTxd
zf2aH>4`4J4{7TrZRBJe=rx&Pv-SfMhF0=)P6Z&7!M<dA{0+-j%eyz@)5e&2p+8LlH
zrQRU8|KSlXzq&Gxm+}kUagNjHSDGu8kheugk2W~IQ>S<6rB^u=HSJ2EP3YwfBwNtY
zO0f0(uTtZ}R|4@E{wEGIx`z^zw`PF49$4EE=_Y{n>@eu@OK1?&Dlq0~B{@LnZ>9sw
zz#c(`jh<;JusQ(KfF&*qln&Me0#kM1L5K=KAiD%0XIM42_;j0wID4H+Y6?|(XmKSY
zQ<HwM-JJ!1wU0f(wOGYPb24>gn@A-!X0nr$rb>54kgBn;a7aMmJ7HwtA1?yF03gjY
zmPA-*DPYvoWE)h&bRO+nwg3wRSU<u9;I?HOKScIZHVH7@K%(fX1Vb~@9l7dQ=(X99
zR8@;3{iSo}d)NHG)CJo=wI~!Gh!cw_kjdL0>N0)6dMQZpit3WM{j6HgY6E-nU1{yw
z3s-F9N`yl`RcDHaJ5!E{`!$&Nn9HEc#EzCC&ub*uMIoL}H%31Xa&bIp7QX9N)b?a&
zho^p5DK=ig`gWE^!0tpVShnnTOD;?``&}Ypkoix4Na8S2f!$ex7cAWW#qDhg0*&=b
z-~uZB^Qi)$%EX`g2`hm;;20WK;+`#$!NOvNMV0`(DKaksV>$Ih_6b}QBvvq4E^~^{
z4>)YP<v5Gg&FRXZqLqv*#UYBBs^)CXiV0rDSa&aQNf7~t_$P3jup#~D(cI8LzhDOs
z%L0}d^Lxk*7%_<jUv(Q$R8!%M!HfK7MtE4<f-tFxsY*;be|VoQP5|iv!5N#?>L>1(
ztBSBQVc5|(4Yf7S8?4E!G-gMKzQ`&FwvSKs{*?ZWDwMe4(t2i|qyDaEy^+uDJqj)b
zh{<~6gAQUTFGY3w^b$G8I^v$vGG59bVWWHPGL9fW1x}Ll2ocKae4}nec5bLzR=Tx^
zeYNAIJa>PPTjXWVgWC{w&XmGo`lWV9*n*0FvlN)jpFsweCE(|8;)4y0Sb+2V&-O$a
zYa~uO@Rh*jfLo6nEC}FOC_rpK&2dX7L>WL+M7Q+zhtVQ2mdt?2e1?!f@q;DCA`5J?
z$0%S#MIdt=>6bMPg<?xn5n+!&OrSx=IL3RnW>pvcTN_sI_cx=)ZM0*EX(!`2Lb%lv
zzd1~W8BpDwyKvZ!!9XQ!8Z*}12GJ76MhOcT?`c6wf{hTO{22Wij@p076<^^sK}t&8
z$^;~V$-)W<TmT@8nW9+24BN6|rU^3r2}~;mL}3(hGN3yx_fULlR;)(Wk$Qh^edp5C
zjmLQkIYqZ^d+N_x-~Bk|RiGzNRl(`mmE<FhtqY-tZp;v^hAKL}9v*0IY~#bK7zB!4
z>J!NSP{W&DNYa=oOH$aLf<HSN(CNY;FOJ1g5_a0gIkLARXP$eGxmjfie8xGb-72TL
z#|UVjUbDabS`ak?W!-cAHZ3Zx;gh6TRq(2bP-$4xa+wcE=C{zrbS5fXlT1ndylfdL
z@I8P`ng?rM0#nvN^;3RJQ7l<t8UH6}_#p;UvOiLg_9VcpfFJw;xysK31!4AR>p=mU
zhh(k#lLTqCDa(o@7dTJ*Vaty)6f0|&Fw{<L(sOF(2jW!ZYwPF6+~fbg`L@=4v?5}<
zMX&OFi19b$_q7^dc(N=?ie*galfFz_)z`l^8<E}<vzR7fEZss#`}V8>8^4c3b5mqK
z2PHGbQ{}Scpd(zET~*YkiE%Rq$kqU#3)>}Z>BfezYOu2?!9uR7F#2i6Op}K{;J7A1
zYLm!J4Vkfu>9{ehAm@P-Mz9X>1SPO@*oqWep5licnG+yeE0ma(LuV&D4|nAKhPzEc
z*18L~x~~s;4F?IYem2z6|KeO;vC^nn(B!Y*Ojs%pBumD4-_n?9Q(X@*slCp9=Y&w8
zgr@CL@#Bi}c47*_XQS5aTlc2A9B{=1`|g=8%jYpo_q?31-FNVm_0Us?u0j$1Ka8$C
zXMgFErIU~QSIG8=Gj`_e!Rd0nslo%|@7=06V|uyN55!Eck^*J4vybxdsN0s1HIxdy
z$W>#3m#&naGI#Hj;j2tMEh;|C*d=S|Z0+PT<nCm2ina7mro_y!*fb$<2xw9j4N?aI
zl$iko&u$KJGkxUG%a$7fo(}M}ssx1K+cYnXC2+$a?fB_pK%{_^x1W#DAj_?|UjRE>
z*`LY?VcUCPeFVW7$JDb!Tdp7Y8+Nd~IYi)wq#XpKYII!P{T%z|@636hX*N21|Lg5M
z`jy(?tWA~|59+O}3{JY856E2b860KU7+d{29yze{+}?n6+w)GU7gu)tc~0fgU|iV=
z;^%vIK7JqF&)B#bM@#dFOg{5o5t$sC`wne3RQ`2DUPJP(^6G%QeT>BCQJWJ~wNrdO
zg88%gGU9O*6G)VSBMj~RCoaHBVg^|;!D%81;=SDu|ABEMVQq=bX!64ceXKAfG3om0
zgmXAeP(UA*nVGT@_61!0z?ubYmC2``L$Eu-!O>+aqXWeW?uRjYnlZV`Kv)q=+X#wn
z-NWm&*d}hH&$i-M-QFt<jk-yMySM6;KKz)Q%`3zu{LVdx)OccDHDstXM%VCRk!W*J
z2%dFxd@p9T(DcC#0=?B$A=9gAf56S8OE<4-t?ZcXy;PrCzeul<yECCis_rmIFg$Nw
zJEXJd=B-#@-yhp#Rjc*X>CN2j_+se<{xZL&nwzg?ocR^zy(^kI?GlsCt|x{#h`+Tu
z8c~O=Ea(u%tLO*?n`Am@x{=<nQb?<nXQFJZR06uXd@Vl-NyHIlkIt>qr_B`^!)`xf
z5C!A~ZXR&*{-4gkM4JQ4C4xL0_QT);>mq@08de!E&#j@YScLuUId}zT?hiB*1RM^=
z2qaof#Boi4TvIuK>D++YA{flXjJYV&$$@}q?jd>MeU!$zKT>vA4y_O0Q|I4sePSxz
z67Q@wm%^r$U6r8oI?;+t^Qvt)3C(k$+f5M5mFbscO%|U?s7cpZW}D67@5e*>*;bIP
zs*V0KE$oA|s03A&!8!Ph*!4N`QO=#k^8J69Ti@W%w?Zt~xfc9zHGsHn7~OGDwRxJL
z;Aq4Ev@)X-&>0ZO{Fp;Oa+UEePypJIKxvHuENfT65J1)m7^H;d0F-ybX(w0@_oFu3
z@`p@KVlWFKF=Men9R_=M5N2{Ey+(VnL@evg!8;uhM~V;YpPJI>wfe(8?o5?z+vL;E
zuwi!AYo!uO-X@562dDa?L~GtiFHT3p1ciAcmdK#U1}TpOo`N^{OMOzMm&Ewea_1Wt
z^IEkgo6;f$3GYR&gww0T>cW?4k9IxtnT_Os*Adi{jFMREzT=wZJUkz6Z^oaMyxZJr
zE+9@T?i{q&JRO@}hyWd2g{M$F2z}beyBn^m9=m=jP*6}|KY?Z)C+K9&D|KCE>T+w@
zctPRw>JkDq;g{rpcoL@B0iF3T0bvF-brLI;scf=9WB4fcAP<C7Bskdy@es)2XLx{_
zhqR-6$VHkU1&O7pdQfSq+<FYOX!hr$1i_Qtk~y}>*zynM0TpbD$rKbFiV0X_DlJ^E
zNdC7ZP-cL`h%7{;r+F4OtslchxXt~u_GN}e+{alzMlx6%itP_4iN8~^vNL)qI@D9E
zwiG!+<Z}0c2%XP#8fC(tCVs0Q7%B5htb7=uwqLZG`nn9mD)J;hM@NUqbtsvR7<`J_
zpdgoc%}L?Y-bBR1Ra#zB0R`4@rVq&k6&L|@FF+_1hLMAt0>D!<*#Ran5Oj4*6cOZR
zlapc?JS(!OBnk9A;MNS_M*^rQLE7B@OObwLlfdbNsF<|HH@IT91B`7?)0k6uO7EXO
zY}6P@Tk5koAmQLxII3rRJt$pMIqt1f)QD=G$8eocCZnOpERy_USFn=|N~lJh<ScV;
zakL~~;fq58MTrNygh!AWeGvYPTq$w;c&1~O=&5)=(q1L+#U95aj(zL<s{AW^pGG%x
zZeKcZ?}Ug`)8I_*CtR-XSEtk)jYPcz&HGLXT+iovc%X94zS_e_EbW7mRCLZu?ghs}
z+&ev5kskkLk5>{GqQW+kN%QKf87v83rk#rQOXg;$SNtPnGC~H3XD>RHb>-fZ^-~h6
zl!UE{>63m_E;vNN9RNSzF<@S9W3q6|kudWiFM%H^5-6i?X971~g27jS%nEoQNZpJM
zaWEE1tm{@EdJdl@`|N0k$~|g7S0!AUQmH+g@Y<4fd!m&VGK02kH7K!G`!sGcg0DBK
z`p*qZt&9!VYe}u!|G>Yt{2Eu(|JY)_GSDsQS-8kkvGLPSovqrtEWF&f$$c+@3fn`~
z9Y}#6+6^l-sL>!<))H<FY&W-wO^Ux#Q@z@U*CT!)phoWZChC~7N!Q@VtUFB7lZGR$
zbQA|fHfvLQHbs{vUrHM&r1J^rzO~k=U`P&bR_6fqQ;U2aKATG(!?IH%d)DZr{cJR2
ztl-pU1!9^a`fzemgfa<OBMkaj6ux%A8)l-N)3=pt(n+E<wUQvk*iy3*Ng_1aV-NC0
z>_MJhu65l@-`4&Nw0VtjAJlQ-Z2$L71I0J=Nk*-nO93C1TFIhJJq?O@H&A!iA}q+T
zUc%}Qn9WQ#{?|?))}AB_?AsP$C0A3B`O;|FYe!J_%;Ahy+Y=otSI5+w*Z3ydr=oGW
z1NY{yTj)s&B(Qk9{$<oqZOErYRuuHSXM^PmtHDB#N3@lxm1<E!l9~J?fw}p<78#Dm
zW+gU@5Wpyp)$}cjPk&qy|0l5f&QZku=HPP^V>0ej(WpD2k)w(cTJPpO-UR6gA^UD;
zl==<)Wo=^YubpSm+>06N=H^uRTSly$5DyweFLcg~ya-6Y+o6~!Gbdx<z1WcSY}n?S
zb&%U(irem8Nn~RbMRM?>J+I>PNC!8bs%w^TAOK<Vf9P_7zz2z$1DgkN;kS$Z&um^2
zv`9<?pb_*q4wP=_5oKyZY5jsh077eGmDdfjA;yjSVxjT?#{2rt7m_7BwK=2Joi$-1
zq5GSO3m2Q^CAqbXXX*Dz4qcV5xgAfkpHv0SY&LwXKbGlvQ-WSHpOoE9*^qpvre2>t
zVy9VYyTK+c9b$NTaLumcecZHelR?Q;%rgmU+T$@(4gD8^?1#&q&6akbiTpfyU188$
z_(l9=gI>aDZ0%Ln5LYg)FSC7rE?#s!#)@w>A@#j91Iam@L1GJ-9vt>;VsQHpC^=B?
zWP+XjV=<j9Uvj>FM9?m-Jo1yS<yN}5$k$|x<oo)LVz`YykumE4R-DLc4*oeuUBoW@
zt>ufeX9_#VmPzAfCmi_V@(2Yu+|zP(^0r53ckMlOlWHijSDK8<eN^D{VedH$F}*ld
ziY+r>;Nd24N?#xBC2rF|$^s%X)8zut3&R3}IZ5vEBl;){FR69AHJiQNS=VcRmkTn;
zRGv6`P1Kz*v9^5WexDIZ8O5jz>NerqWbzb3B|UEDj;fCwIM6V4z7$JPS?CUGywnq!
z>o)tT0l#rz>fFp(Zdc)6O0ruH+-L)BZ*XmgBN&__uro__VH)z*&-C<H7o0^F?zn)0
z-OE8sfPK8|zCrA}p80~%(Lda7+zwl~uPUIiIA1X&bD!Hl=t0xs;u)if7s8JgO?8kR
zh2idb3y%~H0~odypfhW`NqiH)w<+4-;8^8~YY=L5@VRp$ux42Dehq;~UZ>vmWkgSO
z)0LFaepYV&oM(c%?-u5*+v<#h0v|0|vEQmQ@Q#qKCMx;O&&gGE8FPKj&zxmt+&=ic
zh4A)X8XFN2cD{H|!R^0^357f3_}3h34T+X@^^F;q1r;6)dMgP!&XwxNiG5HRvoEh2
zKTmY7h)h+c&Y_4P_`$9j_BgQKgRY_+H6b-s?kx%O+%VnWtg_{-{a~k^119=qPC+A}
z(F+H`P;8GD#;SL=BxM2r7m53?{U!KL8OfZuNl~`uZ4zl`Nw3NNJKDQaJ!`|)34119
z?^hE_D{VSbTUxncagL{#dueB%l~%*Lf`m_6Re$MI2^S3Y&qE8dp;x^MOf}y0Qyb#k
zAIEC&G%6Nk2Tf)hicc%mYk347bP})kf9U?!#>^-93Xg*LL!q1p{-H)sr}rG|y0x=L
z+0QYDarx6~m24u8E+~<3mhF&&zTj)aUC7A68cbL$1Gy~GfLSeJcb-hb^NCMn@=y~}
zqE~ZGUBrUIG@DM{RXp@pP29$kstH{IB1AG>q$RorR@Y==0f=g7A_xwapqLX2B<xwM
zZ+Md>{;18+cXPxi`M!3ObnCfZ;OunBYdzSB0Br+qd5QY%JWKDwO~doYh=Lwmx=|Hn
z?^y8&^#QiTK1mPVa7&TJ-m*y3$4ML&#;&*4Oy4Ib)10BjJRKt2=9;0|xfLPVt%4iG
z#^Mf8@Gy61Ktl&~{X}57;XF+c9t;_bhuk-Z!zw~NPGaetmNtuw&!r80{xKuFa+d0E
zelWO=tH@*APdz2>$l>F+dO2T7NTzhM!o!D6LyLc3wV~#5#S^!qid34kt9DcNU27v*
z^z{ecpbQCm%Xpn@>dy)?ut4vs6MgMAG2DU}iLx&o%xl#Us+3w~_6OT-k(1-VKZ1s&
z(G$=?4uJc8TX_KB+ppXQ&I35eRMpegY#L`87DMbuV&{ey>(51j)MvMCC$;S9X0r{(
zMbuyI(ppbbO^qIh;^1rVGl2<&xfZ9WN_Gar*8YO2wpbW(T!7GL+3a-MCt<2DAahR4
zdkXFM$Vh~r_Ux7P14)slsp*Lq{)fo2rcVFvD73fedMKH74dwPvKHJ;vkFT`lI=!+V
zx%x;q(Y@&bwz+iG$ADL=Mn1%iKZfJjVDVmK<1()4O}uZ@cc>e0<5pz6GRo#B6ck~u
zHBs_Pz~NPFqvlxGkOX_nu-=TiSccIOaedR;mKjujw8lV)6#)Np%4|`hJwk3l73caa
zYpc7=+!<f!s85OY)=h{fZC~z~KjnQHEn(K+Ib)Ztbs{?KqVNlOR_nt#-yyzl>qlRq
zS`39cCN#zUQQbOzY~88?p{xZe48*nkqZPa4O9-IxM{)syXEL^WHt|#`*%#PajtjN)
z5gk0B>>(1$28W*dnqfsQEsG5+t%M`kvT>RI?Gx*StmD~*JXJfTrL`V?=slVLsB)Hk
zEHmhjX}`bA(<*0=V^&1&H^on}ES#zn+B0z`7txvXsnXYbdNLvyOWFBeleE{In*Ojx
zG2!o*5~uc_b2}h-OS|f$SXO^`qTiBZqXn<hEk(0?hVN1z6&<`+G2$#{zvG5O7lV?n
zA7{HPb&*@;cnMxS-=@r3=E}TG(0jF4dp5nS<Zyqk-+$NeY_^v?T5RSReBsFj_q9wJ
zJ!jg~SwWw>&2{E_4!nUIx9ne=+?hDIj9bcX|LdX~2bVxVzfOfn+0J8=Tv4@>;r*&+
zX98Iz#ciw6j*{wL_cEQj|1{rRep4-QY;rmQPO_LAR79{-mB_laufc4yn&T0*CEuZ#
zs(5bk(XMw{y)8k~F*_lTxLMV?kx=NJ4T>0QORi{V8IabW3>Bw<Zpb%6t^H?PIbs)X
zb2W!d+U%1OiM`|fD1Sr;_n|JVOSD)f`$Ri2Q-$qBJLXCM!g8+pXzWv4j-=KXq%XE+
zV{RDB54w0F#*Ueh;k_}3rW4GWJ$ATP(&J_hax||AvIs?x1_kX=*jBt_Q`t#rUwV!A
zWNGH+YHR4q3mp=cQ9eW7r^c}tbL!q`q*>x26T~GGLv@9Qrjp^edHr=X6DLe~@}oR{
zp>L#%CG8K9@jD8l`7beUxNly_Qz*K1!Q=E`z?n~ZEpCU1XrEW7|1GpFR%C0qemJZR
zk4vg;N~S-zUNxj6YE`KLe4y#z24{{e6}E6c819zCt$%pa&K6+`tUD3FU_cgR8a)KY
zWN<)ZtmfdD_JG?{jrb?89OIm`<i}huUwxdF8D(1JYs=vkG*0B)**dTMs&lB$Zp|ZU
zcIDG`N9_^we8tRU7jq@CTEfXV6eelBBH5au!deXMzF~ySZMQ4ejlu0EC5G)#=}?t%
z>RgF*xt8fFnd-B6w$7N|eOJeGvW?<(bzgbfHR;w7gV7?6tFAIBAD&zZUa&g+uqI+;
z`<=>NemV$o&xRd_o!OgYPABf1<tcjG#x-eLap&-rNb&Wa?~t&fdE~0wu7r0TbzY_x
zU%wjk=CKBLCx5#W=k(#+S;4s(`y9f|3#Dv$(-W>7m|cCq6}WWPvPHSBtIZT#V9>~8
zH?OI-jE4mg7Hgj71!Xf9vIJTMC4~eN*o4?Clk;r?1`iI<yrdVx%jo5b0~wGYZBiio
zh){Z|{fC{+dkkz=g|4;O5Nf!;T+24Gyh{4iyl@)3khxjg-NV-Gx6U){*}0>g<7!mX
zit7i$KBoxQz`c6q3I<K5e2Allt0nKR1|FJ8O%pAht)Px|+q8Sz8C#Cf(rZuJTg;PV
zL|1Q8#Y{JAh6g-rN+0E)w>MuF6pskNeW637f{+Ts%X>EE3(E`GZFW5?n{Iu0`NN1_
zYDSJrt3GBT*;-n(A+pt6jz(R-H(7hoz-U>t>6|IWxM-lfQ1=0s!~3@O9aoa|CuH;9
zCR#t~PCI_a8*^4l9A*8)Hz6U6&D`f+52Kt>7~@}G|A@sP@D<wX@Xi|h@}UFT*r`jR
z_v1o6z1W6%44Ygf1IBP?+-CIhW*PqF)2GU-=|0@I`XssurwzJBM4mOH48M%uD9eA;
z5X#z|JF4gfBKX_C`IF?a9XM=z0({0`OI4GSnLBUh@Sa+MEh<pwDE4KTss@oq*XNng
z$Uz+%MRnkqQ#9e7i5mYOBTCa=#o;<^8!p_u3vpWJ#^oHu44xbLFU|6(99fdrKQa{J
zA`Q7H-Sxg>aRI0RgUCLSjawoWyYTjTyylqfYC4~bZ8G}(8riT<E%$j8WtY1hiP>le
zH69|I#B8TgMBolEe6YfT%vw}`$lYORuD(h*Vv&GUUCHu~r@uIbu0z*&g>DbGu!2H9
z*83rxwHZC5g=uC@8x0&w(M>*TGT7`>9Wqi@{bUS1dm%O=O1r|ej7Nl@PDRCvTzHQ3
zh~PHec5hVIRxLih=!(Z+#Hsq`dqIVTD8kaeg&mp}COMJTtXFW+pg(H?@AiNSz8wR$
z_4u|9GQcG}){ePd0S-nmHyq*G(9F_H&@wVruB}no<;p>gn8+`SkM7|3M<TNOjm-3&
z^Q|7eSa0ag+Rf!!nzc`l>q6wg_ZJ34?5yKcEHfr<;bvS89}-5K5z@Sw+<mrAX!33s
zYmETPrNi&=wg|_#0u5u4_QzJB(Tu2-(-#X1c3iOHyh(S<NZLm!tGr-eAC3D$5glH)
z<>Qr5EIW%aE1PY9FlHb5hk5DP)Jn#+a>aO)oC1Y-r)cS*Vu@3W(9RtlM#s`mVhZe)
z*G7u=qz1kdZ@gF|-#J`b%b~+o=Qf_2eEaRQq^@RN0b@@&PNT(m2fH&;kKaG|^Bk3X
z$S>>0Svx27L&FkJvUQY<iGEEvpbA8VG5Zo)?@6p*tUkEM2o7UKO%xNRRD{Hf6=-G}
zh2}-Z<l6ijQ(@Ngoum9=HX64dRGD)3wtqtNkQ$_urB@s9A(QAO%t(H?1{rbSuv>0P
zdpz~D$LRuXaeEPOnuoUA^?g1A$O=ElVE^-h4B?&PFL+qo)1KxJHRq2zDK$NS=rJGt
zo>Vy_ZDi-ZI+j$=Ykz2Zh*$`DmQPPi6u^|GrU?z{Pl-g521_%SyZ0xOdd?I~yNp2U
z*drAv(drNsR!I+<LKtOFubhz=sn|C$zx3XIAV7VL*rF#aJ$2;jF~`5#7poJ$IOxoE
zRz1~B)TLdjZ=7!}Xwegt95ZwhtTW%vF9~tclMWPQ@#@rs*F?0(6`b%N)~?(BfqVK%
zw%0ke!yPi6eRwWw6FcEcm5gUZ%}*L+<7$#fvV{!k17oPmr4pKn#n1$&TT){b4B2XI
zuU$4n;GFkKq?=02<fdQcDI~uMA6s$q`Ls=RZND|>W^df82(XO#hYHD`iDLkZ?*|tD
z>i)%@i6TnrYk?#JT*~}1dFBRs0j#eeXBVh0@Nrb?8M=AsSxw}%0_l1YRP!OpLuXaj
zed|#<JPNi5d2!kjdawVz@u=5U^QvRs6qhERh`r{lT2Vt4mJm9BL}ctrfKW1&|JiQ9
za<v4*hUFd2>A7eSAHx{9Lak&~3167F3fZH^@z@N|kOrk?4b~l+^xJCXB*2Yo&}<U1
zBctVowv(X^<8i8c%X{I+)W+?qLrvSes>Vz3=A68N@x0-v#76~pE24!2@<=)55j&pL
z0{1yK|A*JKsSVNY4Ou8Z`e^g*_r~RkL2Aik<!Ki#`32KQpSJCz*hnaNvc_sS;~bMG
z%aqpKS=jAsR2h<MwOrm-$z^XTGIztvH_c8vS=D75u32C045F&3w{HsB#rSS7`oyV_
zW}g!{uW1t1w<zkKV@I#JgtBH$<N>KIb1xWd=`$PcNRa0PQqDaGNk4$inc*Kq{<Op_
zZ-<L5=z!Q3cZcTf+b7)0oSGwii_tSAZHzAcF|FN>aqhu`zx~D!^meCl2-O|gqn>ZI
zP8lE03;T?o(JMszTKT07R?x`@l5^MGE-l1n<OKS4bcOp4GREB6vgrNf83B|H7oJ4s
z6}L~UMJ3jZAo)7vs0-dAGp^d90jS-!nLEf!_m%1g(zIW?8P4?wWEhs2oLp(W-8rB!
zO&z#flHGsKw}JRk=`ZCN4gW=PGH<i5ps7ubrt3dhe>mNF8b`AUGj(dxylG-nXHoK4
zX4Bp2ypK(I>5azkkaZ+w9Tl8pIFS}aer2)KS4F`ze)Zz6YKfqkihp{Gs;2Gno3)>N
zXRmaXq<L<TL!s9ux4|avA8wvqa4j4~T&#a>1v-;v!^$d;E;egiES{|#J|Z2oDIs=F
zV(3I+eknm}`@<`zQ`uIDgXqgkUD_>F^B3(w`wy?+gQG88J>*$dAfId2mNV$_;C@2`
zm%oM<zB(#%6!DzJ;posnAVX4M*A;~q!m%i+d~ackkczjqx4b6F7Q6-V_7r5_<81Cd
zwkI*G3Dli?Hp9Jq=PGbMn{}O>*)tcs1jn@d;~cN+yo&*mbUh_1{6G@TrbfE2{q1B|
z3!1Cs!jwdFK4~x|BRf#4r@Xz#7E%8ooHRHlx>|b14GNLNi=59Aw?+QLjR!qi{7~a1
zQ)sAyQz0UzVyH1@{JMCqy}V~lq!X6?#tb8BPho!CHb3jr&3%e)`GfVB4)yMnC48PI
za5aZ3YTfO+Fyi-0Ez+Wa%204k6~ebUi`!o+JN@sFj6AN)@jCURxb_7%!QHW9su##=
zs+1<_$&Sp>$h%hD@nNxjjyd}NxrH^do_z~Jrftgm<0>OVNdtjnAy&>_l^q`1^)%f#
zCRC1E9$D&R9(JiqyF=#X^}`Pm7W(bh9R@<FO`7x=*|f*y?blDWoD@TFtML>lCLPrX
z;65N-7Uwhq$_ei`(&c)BPl|GG7ZZOLxXgVcggSG}4Wt)opxGVt&wNd%sB`a%Msfpc
zWhuN|(e1C>vZ3A;PtO>Co~~CN)Of}%`IJBNOH6I<fVna6#mU6cljgfr9+ppsd{~~$
zX(IBFq-(Z^56al;@vLPN$jjLQta5mr=Pc|-e8v%b-Q2^%tSJ0{Q6_gW`{fwC*tX`a
zvKw|ZpzUV{4rxx#y(aPN7akn)p1oX3q{(FKbJ6pqY68><JH=5WP7(i*G)5?)A2lqD
zq+V}0hWZqmeeP^}9p-D(ZeAyu{w>g9S{b>dgLQHC2?>`jyX;r(FW;@IsgASn#RL=$
za6k>Iw-fsVWWqYxx~;@}J4H7cJv(dzg?pp>i0BZJmv-$w$LA{r`sBB}Wp7VXpzU%S
z_6*)p*oD4cz!&c{eOACEXCd5zzL4uoKU!e~+6WeKlEeaflt22@;6Op&kK!5}Xakv&
z841DAw^bv!D+s#JUOe(nlRI|qfgU59uB%wq-g3yDK0T?lXF@wdjjhAmKuUXk^zKx!
zUGSk9miOG<4N_l9-s_<0>B@Q<<ROShO=0BH{9Jvm-R?VoYd4193=$0vPVdc(uIQFg
zeLEOFV%?l;r6R+cL|rg8V~pU$tW5BCpW21t4<)WRD>YnXqs_8L&fU(U9c10nF*|3i
z4mQ9$np%9VN(HmtB)?MBeeQFdJ96L5nTZ2hYLy=v+r-`Jhr>*i?2C469vb>+%x6$S
zn+Qz#Bs*PtH%~OaocqA_J-$hoo`ot@j~DsfB4k_Nobi8|a-`)8rA_BTW0uyVP0^b=
z>p`KB+qr|5=ccXV2D%!eSv>d5{tfm!cF)m^HN}%pj6Z9ZEF3epw#qW$R@>`M_VK>8
zL9U9AKT@1u=G0fyGMSpDAajo<d9!10&?2pT4uP{ym<?_ZK?@wn)^kMhsV5MSGI`^#
zZCP2>ZG!Bt;wY`o3RXeoRrNNZ&K=hE-XBWUiG8#W+Cnia$I45VX<kwt_WGGkx?D5G
z4oMH@mRD92bmA|ny(4w$u?1eG)^SBWow@2GEN>C*Ws9jk4yBX&`189{CeD@55g+d^
z>bY+u^Kw@x`pV#gr_F-qI?yb4AUoq-MC^rpvin>6@Ip6IUB2yEQx&m{L>YUbP#zN(
zrA~h#GW+~lkc3+8Ekt2d=}?$>xG{uB9*ABzDirPQ{VvfgyYH*wu85i1`qqnq&q}Wo
z4+w>Znm0UfT<VU|e-os6kG$B{`Ka0;$W(_duKu#lo!}RpPa07-o}B93?-lTJx@?)=
zF10<*DK~4Pj4^nKYe$)*FDv0voLE}Zy^O9XsUZmm`a^{|5l^du`f8=%{<&dOLj&%h
zMr6fo!P#?;%>!IwS<7e-+N}BWAw9K*Gxxl+)6PnzW3xBMyes$An7EsDqzXJtQe!dS
zRTF-Ay1wbS(bQ;a!q?uKD5HHEktGG6eX9L+)*PSavtG-)rW(iJ-nrbYQD>k#)v6_}
zp8JSitaO7nZr_!tFX0ObUuG9jD;FtyF=};h@@aTH>=3rf@!N1|Y6QxRtqpsS1ko1>
z&Ha9lK4_#K>#{Eo3jDKIx*=^3b^D$U-47owYBLBj%RA4B^b->JUyeW#5d<##9&y`k
zI(rMrcO0JS^91y4<dR<O=!vw*JW}B#kicupt*WZRYq^>YHanzmVkvmI0hoYM1r2`4
zjRmwUMeOKkECqT^w#JJJ4rMC{e6EjBl1yzl3q8{)^=0sCbsL3pEQ$3~yvcF8sN%qj
z(I>@x?dbyzM_S81wZ+u$cw(jteDWam6LIpNRHWbuRx;zS{z#fvaC)ui(tFd7`tL1E
zY2t4|{j7y$zto9~s;dj0ZlxWd8}+9^J7){{0&{qJ3<!|UY7^h`?U|RlYld?JqHB)Z
z{L~L}tj*W7FnkQC(|44R3BpAQ@TS1-IA#|Ul#N^GOFTiU18)j~-2$+kNt`0x6-&7;
zC^Q%tdDyGEBlc)|q(kcrOMoc9qLqq?sLcK7TGxyq-{mp~$C&Xj-1V+_SJzXEqtxQ7
zyiW&|_v{zN%2v3Kd^PV}Ni}`7H)Xn=as>SSP1@7h8)!{#f2k!)9&XDvuU5bM{LyF4
zdZV8zx$<-jmo);lDxa(Q2A+r{r4V~bcScl4O>p=_;{8GN>Cb4VOX}qVy5B6y!s!c_
zaymLutu1Gpd5*q{)_l3T_Vqdqw4^lp+x_<&DTypQ`Wj4shs?Xf!Xv1Eo-Qfal>XX&
z*tE~i?}5O}e?JD1lNU;zy=M9E@-BvdyRcGlK48k)+U!tZ(VfexH=TBTn~q2>TfJ80
zO;qujDErvVWgH?>VQ31DSR0BEp_g6mv{v5$|C~6U_R=;nXDzQQF2yzEvIspVrG*(q
zjI_FD_=w#(F`Z(GER<I8!zh$NpL3RaOz-#<e^qYsbX3E)8UooQ_T?#er|88?{LipW
z6rXFyjPO?-<Lhtqc77NU=FB<pgfnZP^1}VT+C8>=ciS6@qE*l%)C2p3IFCrBsin5=
z4Ge!C*wP_G9zK<)Q+doNUVLK|6~19(doVS6-|%x{V^U6=e?>#q6ILBh*O5TA_N(cG
ztX1!gxjr`01FAUgs+H{sOt(*_SsG$9vxb^EW3})j4y2MPn+`jpXb^oM7QGNif?IF<
z87$&Nzu0sVpNOEHRBrS+FSn-mqP!iEL-al_{#T!{Qz0E~9i>F&hoNW>v5BlwA&Ur~
zU8Se%3qNXFX?4n020mIb6V-R3@x_VL=1aToQ8c|DsCAgW+g+f&-r*=^r4y(-(N^(w
zda`chyj#IV{(ELv`d{}-3ZDlqSccQ2s}<s^!lqxWc3AdRiJAtMJ)nr3&?uO#z2yDJ
zDq`(c+G1@*+x5gh>D5OLpDE!{sOpc1Tg-l-KP|#~N!IB@%inr^BGXPjmXSFMnj5)&
z_G;3h>Xwf88bmk9MbX!+kL8G6t@zrpX8c&Ku2lnjkNHOpaEEUz$zTV04zm*g?}<>f
z7He5UmY!UC&Bi9t7<j%{zJcbzwp2EK@9r}R#iTyTo6kgRxXZiompIE=P}v04=3WuD
zovCPg8WE54;2J;iS}G}F`BI<Oo7H+t)q%k6xdUs4yYw<C0+E9i`{PZdQbFhkXT*vU
zPs8pdhef2q(~N|%lB2R4g*&d4N)XhlTAX4@PwCCF^B0X>%;u(>KJ`{;Y<pO$k>NSK
z_a)0UI#*dtgJ;Bz$tw#e$1eTW)g1Pdy3}J)Yos(ym-2xAiM#-$sQiZula?dnGYkvw
zYWaqStBJGue6eM3wChgc-jaPDdE#QE3aooCnv?Y1DH)k%f=0%jueA&a#_ugT9N(QN
z)CrsGvqj%Mc7t}I#oNOKr5n$g+S%c>^JtjK0sGv8xksGUy>Hcw*ru`;h@tg}z)iZr
zp618lBSYpM%KzYO(5xkkc+u3RVdD@9s|e>;vCqox=WNzLxZiHIlXI=#)$*0LIA&vk
z<zRJ1)4JYZr=ql?`x(uBiKBtqzG=%QG)?K@j7-M?FW;`ggQ~PAChnP9LK@9Odllaw
zFbnYA=65K0M?hbwtH02YbS+DpDyb>UL}PmwGKDMUkWgsgKi%t>HXNRq=T>5GU7slk
zp6>8!(Ql!qc;)oli)=E~_k4Jw?!NLJ($Q#QFYiaMYtbhr&#yc@sP?zDZ`aHkcd4`S
zgNDnzC10I)Iz>e(QW@OT@CAP7mi=EW;xUJx$jMsmL-Buyj*8todx=#~Y*JEcs-e`p
z(B@5oiSj8)cJG&t<NKUB+bGo)#Nkl4wX}Y9sexv+Q~fQPL46;RhoDY}@aqA{qvMy5
zSEB}vNM`l<<CFNdr#V)-vY%s{?#X6g!Z&Q)?{J{MjH7dh`D0MyK2tBstioRD5^BN%
zCR~eS-JCm3JQ!0#r*7P}Z}wZzDQ2&DOOo<An3FQ*mNhQ3zw0IG?gk>1yjmH5QYl^f
zVl#PA|6<A4nNv9?h_M=!&PL;3iYI!L;%D7ke1k6wgR)AtFw^6Ts#C@3C9f+PvfC<8
zDrgy~yB23j9=|*AhSBeAto*N;Ml;6)cBM5t97(fv$^MSy8L7%7!a|St$-wCLIc)<%
zC0)XqxG=+IG%d_FnE>4`==@@fM0IsG7QJ0f(0J)JpHxfTXZ#3k<VlqkbG^LZZNn~m
zsykw6Jvigd9de&L?Y?xbeaQZJ!J7I*NwqGmPWkEuuN37>Y_ENGJ5_0WgwXyX6Kc*<
zes*+zfPR7UsYzkthE-UEf{v|PZg$7q%#$#AKSS&R6$SsD*FQB?ZkIMQf032o!`98B
znG-LjG)?`0{9C)OI?~Eu-)tgoSp$7RRE<Cm^TnTE#RurPYL=g>@d}yFw#@X)Ik(YK
zZz)<FAJ~$asr}MD<e#hg>I8HoZKGN_#bHKtY^QaZgY|i9-TM|g4+PH`lDh`ApHulp
z>&$i1gW`CvDNdZa?Q7!|sE`HP4DePx*oW?69vd?!oCd)doYDi^)*9FBN&br;bei-Y
zVP~iQ6*aK5zBu)^E<1uv=r#rG-61lFh3X7gaCbL}B^c^al+Lmc7ef;L5B}+LITb5T
zv*~&7WWndLbfkg@&*zb!m}#qK%1cB%nxLqH{s!Fn*gBB|{=AJqP{9OgyP@sr%Mx!L
z>bfu-^2oNbEU6}+g>_U;oksbm>;dSl`^ZA55r@n4>Cre_nIR`5-KjeRHO3+}?`Ny!
z5;J$5#85>mjEXr%!yQ(;cG+my83nhvbdc-gX9Tq0sOZ!ws6JtM&kD;6`QoHS7(FM8
zvv1hDs_0H4ueWl~(pwB`Q}&MZGzf==TUJFC)$QM&_?k2kV0BGjtP!_Ps#Q1MUoxOP
z{Kn*Pe@}B1r6`5CHyS0^VQJ-(gL+iXg3bV^C4U?P0_Uulf1&}}IG*4<E$q=j@9YxA
zR)yHVW`2M>G-Cex>f^*0R*{qa{#_1-Ueey}eytf}kp0R0@&OAC$eSh~3<CRv)VV1R
zDSWKZY3~3ZJzcTR?p^zGO;^W;BbK6{&UpIuyrnN%7%LyNn6<lFhP}U^%8$`*i7h1+
z7No5fVD3ENj-)jz#)G`-#o3xntU;eE%X9hYzJT7N+T3Q1jof--D#L9&+y=!`mjnB%
zwu#icOp026@Dp_3e6k|eo+voi(<LHTJ9U6<&*t>Mf@edc=@#EcZ&rxWmpmVJ$-b4~
z^vW*i{Tg;$-ObtfAEg1Wc8e!i^CRBJv?WM=zBca$#+uz(q+ZsI9hQn-NUNNTC_WGu
z>2xj6B2HEBgodrA>Vc`bQIFnLV&OHNxf2EFp7V9%`(rE6uPOVkmmnVHixFoZ-VNiX
z8ikceY{-q|$sSW!JzaBffR({&qg8Nkl;>!z`$A@@_;L+O{W106kj{|C369h@{&KfG
ziS!fm+jfyfUb{A25Phvx&KG8|R=M5Q@{rP<H*t=oj$EAlk4j1|iw{?x=#WFK6CgC_
zkPzjq8msR3Ek5%~NbYh6>%u8ET3)7B^6=Aag>9#ILyom_&YWjsmNuKoRS!iZ6keH2
z-C)^<>UN^3OM45EIriLp0Se!~Yv=Ta*+e$XPLHOUx_`oiJ1-Ty99pyOPgnMKU6P4z
zO6RXUaW25v_qp6Z6X^HZge+}Ti$ZFIaMwVjrIt>kozTGN7i?^-MLJ@}Ll;)a%G8I(
zsrQ^Z%0zET=+3IgnC3d~R%P!Lck-PX+_9N*_JYEIYAmbGj-+ebwU0pI^R4-<0r#(_
z@Km)_A?91kqSQ-eFoc-Zj@Ars%*^(=3}q8vnINlmFL#ML`gw~k#jtuCx6`3fRtF_~
zlnO;RD9)7<XOs_QMSr#?R3^&>?nvIJGw*s*EKKi7eOE{{&bypCvF9j!6rh<CS_zcw
z6*ACs${Z2BO7=P*_fO#de_lx5uWve*ot>Uj^w-Wm<7k>Pg$Dc0$vb+L1>-_rDW02s
zkuveQevjwu)v8U|vf;+W^P`R0`a4$+)+tIx8^j{3)?-J|mm4O#Qjdop7(YI7D0Od2
zfBBscj$$mz%#*W9@ToFpwgiq>gG9-M9nB#25VqH-sJ~`&>%AfB?lk$Y=<Z0q?@+UT
znL{#KJh?BdfGv0IF==oBS;9F`!*<cwEMaF2gc^Hhd)F{Y&+fj=Bwp{!r2oS+b`BPT
z2?N>{9#^}OT<?1rA_H<I5-)||)6j%)kfCvdBY&`$Bmo@mhg55rQ@huFUSy=_Mjxvm
zD5L*BWW8rN8+`l6ZLiuy?JXo|jifbOE5s&<S+z%uTGbkDsZmlZw%8*GqV}v7wf6{B
zMb+LNsP5~D>%RZ@b3DiMCa=74IGj0u-|;y|y1K(6)~#g<$2!du@nT60TbAiIDJ*tf
zY2e|4tvOz>LHT`A&~y)cxN*q)8D^4_6?EG`&#C+cgsvR&?f?Zq3`4!1-cO|jCT*>o
zDY!is=a{RbBZgbvpu%(WM<W6?m}HIAE;yqIGM?(x8kV~$S#WI|xe|F=tgB>-X?Roe
z%p9-eC<EOm3s}6vO~{zG`Ahr-Ws<d_(~cX}1J2-}EI=?!_H+@)>e}kfyC)#`H5CW!
zS4GZN=jBJ`o8q{kWnL*rQd4c<(y#iTTEy_L8cq7e(`GJ}_DnKf*AOWCyJY{M@L7uW
zV!Nsm`$tt4YkKsPVOLj<j~-mRF+^KZcNpcPsFfLYYE>j0KCjanbX!a^+LSB*b$rc7
zm+hk5nm+WYy}6yU9w0w!++ZaBt3W}tl^pn*Epq!alczpNu(ggWRzK-<i8t#S`Jr*y
zHHYLkt~6ErXLy++P?4y5E%FLRlnAd-pMqIo8c+;ihgYtm4fQ{7mJ)=G+T{FM(;rr}
zs*00Y8=on66(Tc-r}#f3U)=fm6TO`?;NPZjHmCct_+)W#Z?NxHS)#v@f%x{=PxLsE
zW<hMNo-g(}*2151AN|5V^zPOJ<r@1&u6xfukzdqpz||IsYq3TTw=AOSV}sRu(f*)x
zfogs;vfiRB|3w);|6bzgyQ_duQpHT1ZHflGLMz4OGO2|e5v{JPOE1>emtTsUm}?R;
z2nlF;h9Q;6mm0ScHkeWgo@<IfqP|t++SE$stHkk8_NYiB8O!-SvGqO-!`lAxyjA|^
zx0lr#Tjh<ALX<mZC*cdRs(Go-bI}kQjUdIjQuEj1GfQ*eshelVwx>hnBAt_lAWli+
zaxyELq)(Nx!{TJ_jUzh*U0g$8Z%qffJtlaw0oQEalXz?=s_zzJhYlNCsuAkigybr!
zjKQ!znXW0646UuRn&w&fG*PdmsNQhPt89r;!{+*oFXFXZi0qw|UIf<;bjtKvIpf57
zFq$|q15=fsLT$eyib2%l+i&>3ofx78)K};URRM;r4qYqhaQ6cXP6ypbj>VjijB`UN
z8CA_UE_NlM3-($7n8^_V6UNDqiTt9c_6+T?z!R@%ZV;UrV<8cu)U8-ZECHp{($sn%
zeDE;}p=nG>I*D*Na*kY&=>xdlzdxjxCs7@{;PqkGz05bzPA>Cz?_YpG^Y4kGtgLfL
zdzUn$vS2Ik@AOCD@XR*ZnAD*>Nh;p?QtZmp{hPxWqLsaCygX^ca{DW0^IJTKDV@zs
zSz5HAV%)CGnk?xXRO>+4Dib}&g0LMs2(=gYGz-u>v}M=P`m9`kc9IfSt3aC+r%bhZ
zR#O$TCnfJ`T>hl?w{&Uyud*KQNU5>^j4*-ysFwex0xz`>;gWgoOj2CGlq{2!?YJCa
zCx7?r4Bvd^M}Gaw2ei*EXig6_u^1T)NOIsfHZc8eCdaeMU9L9gO7`*;X2@OsV_uI|
zsX~s5ocPx=MtVqPULyLZ`H(4<1=-%i@(=v;4O3sUn@O_KrMW~h$u768k~y+txCu3x
z*~<>ADH}3rif@04rMWcD>+Lf3P#Ek{n4~yS$N!V9u$yN**<vP}yW+8tc5mO|Tpg`d
z(r!Iiqx!>5bv*sDQiXi%$tM%Xcl#Plm@k%<nXE8K>^sZk;l-*1KYpw9I1lmll+4DI
z;JLK%z^5AP=sb%gqMYCar;{b{u7L({<B|7z${7<~*mSF~*~79;EeILDx8|#so=N5j
z6QgFzg1f1S6uTr-)}SF@fV?s1Q<I>DIj<zO5>xF#W8>%XJ6oQ>I)>ewS2_JED`v2o
z+c9t#n$~1*QGO0yXTi*QTHHN;z2KX0@7QSLuhX~?-5L)QCi)h%yq)7#tJA+Sp37yT
z?6QhYnr#2A)6=1p?Oix*kvTGa*RU@!BZqncOea2u{h@t+SdW91w;7L^2RnPNivesm
zSEN<3;?@yC5th2d*@;``<31Ua4o_#o5&r8f5a)KQT0Sy6PThA}-&tgSAq@*GQ%j43
zHi6ahjK+QfNm~O~^tanJn9Qjf;UgN#1irAoA!BV=*&g{~;gyykr!y#Hi&yQ5U9IB%
z>P8sfvynmfl^`Yun3(96E7V*my~})Db}GjpQYF}~-8J<MVElH%s;`Gll8=1jiL|Gg
z&@0;IME}IZEXAXa=O2{o{@hJ)y|u+#?GIv^D|YW{$4N)#9MN7(pVl4>+<C?Q{_o6f
zocr|Gv6WwonMF_2zIRTl-db1?Nb#ocRzvt#YWVY1h<ilw7c4&QfX_xO21BEgek|)5
z5Sx_ag5Ls?@@jN&3MOFMw$UjN2;edi+FQL*--n{eKwEbl=|OCNpfcWtr6vXHd#O6)
zF4YM4mb{NDM?V0>4J}+$UU&3Pw@Y~7Iw;dFOo6w2cY_#xWm9;eMZBI`q_N0xLAh=`
zI*+B55~h50!i5%<{E;ts#Vj{LF)0<*C@9u8st#I7&n&ewc?5p-HIK3f6|0x+YT;1-
z2jUczS2-{tq<GYxe;PF&U!Sk<b+X=zFM7)f>v=YsvwfmmGrKwQt=eQ8vsQhodDmbv
zgZc*npFZMY9u;9u6;Xh!xVoj3CUv5Wmz|jmx|&YS;$&Z_Uf^iAMa;Nm`IcJ$PGO)O
zkHvU8%(%)f&y7)Y(CwS7|NYSH4GDg$z3(_yi~{x{w97-5g``V7P;B`8u%sSlE}?qp
zEG~O!W9Guse*tVrZ%G+5#nS5eKM8GheE_Jl9fej`>{Nexf9sLAaSC>&$kAow*i0*Q
zRVX~C)BL-wi*pih-0&n8!AYOh%>AvZU5d%8DE2)<LfH|ORO6bv*)H1bfmyjfvVz{I
z8B<W)M~rKisBQ`qAJ7*FnOn0v?g;$!Pww3!1hbDfx$x*pcjeFYS8imbXc#mo3xH!A
zoC?s*Tz-2W4u--v0kighW*w7n$MIMIaR69!$^ZIHT$jBQQq$7qvb!658Q!f>2bD<O
zNzGgoSRcdKzy8=H(fM4GWe_Z|3g@mD9sY7S38c|b^T;3@uJqbp&fsR&q{On^+eP$|
zb2>4{p2AC27+umbmaDU+Zk44mp3X&aQbe3rsCnd0eM3+Dw-o)~8CKpZ%wFTTQYV)V
z$LBYJ<vS<tthUVY7ce9`i+gkwRmOP4&}S@{CTbpkpKr)x9BsRcT3}{hVg6o{-ZYwh
zK6veSQ56~!rF_@Gd+R|7O_3^M7nSZ!Zs)CB<9Ev2VhFg(Pkf#hwvl_Ktup1dorb*;
z`8+mwG;ZCJF}83M9X?<zEVOk|_2XfQ3-@OgGa4mG?#zyuKect_Dcnth*6tzqAi_nG
zL4zBQ3H)R!m&H80+<F%FMR1A~SF{H(O`+*;N<sbXp<^MbOa{DZ&nOKobD!prB7bQF
zD6s7c5K^L!kC(s0`eJ3ua)^7sQgQ!|d0k#+$xIxnmxL;uSZ>H{8Ibr-_dl9i;u64u
z;3WeWp5-Q?HwVX88+!nRsk`Pt)2#5frWJ!*-d|osS{Kgj+h<>Kn@u7%>nM@0<^DqH
z%1Cj4vEEFn>6ZaJS%~{Gnw_x|tAe`AoNdxQfhFyYd0``{F<TzD<zV!hDz{<Kd|?9r
zS#4KX%47ObNdM1_m3L8{=9Xzc{08;Ch^_fx@aPuOw?&_`BC;~wwJ1y7Kyh0csukQ<
ztJ9NIiz&*|iELunOPrLQJXZcLSk}c?jFo#+NgZsuaU(H17WTw{xZ9=moFm@i0Av`D
zhzSV>XW0H6zES=j+u?5%zdu1;N>iEtS4(<IfxPBmpbc`*TAJ<F>j@-m|IeO!LQN$*
z`@+x>|NOhp@X7nbuGtZomfTOqq&~l`n4J@|cjUUpu!V-WA?0?>>in_%W+@=E9fFqa
z6Z~!&&P!}hs&MJtz83!_6)~YLzJs&rsWE0lNe$XwX=M^;On{LBsLqrH794!wm?1EY
z84XAd_Q25;r~TW(3TUYD027T3*EEu`v1I6|8mO>hA30GhV8DxqkvfiLBVw3Tf+i%I
z{KQmKL-F>6aJsSd)I@&ppt(5XFnQum9``9dv`ieU^lPIS{G^=o!~_&0cU@RwWEhJ7
zxXT7-Eo#Bl!F7kK$tB`dJetZg0xqi2GmR;I6;cr?;3Tx_1Hw8`*E+<BW3HI{{W*b7
zwod0X#m7*meP!ZEi62qiuN0|!=!~eD6#n`y_|#SO<$hs8|0&OjjuU^zoyrXFCV+Ub
zfsY}T@qGt|aI`s6V;G&rJIgvuim2duTGV(+HhnzfO-A0G``k%a3z!wL=vw?Mo`cy|
zj}M)8?aP@)RUI|K*wUM8w+80&XFw;m$?baItwL2jX!k9GvX8Iw`=^SQ98}gAB3v83
zr&3wW?2u{`fuvB&Q<!y^QPe+ValjU^DrVs3`=+8=Go>Fkq~7J?7S{wNHGandUz<?0
z1Kr@1>F@OOc$aOJ6%GxxSQ?7pNhj;>dKzviKk<hAD-7jyZ(oT)e&#D_<Jw1`fg<Uq
z1Pr;$>PjI0n^>gX=ki;hQJaL95Fu?`By7Ot#9DncZ<FJ??R%f__l>B1Zt#<OQg53w
zr6Amz4gOI0tC!l@v3`;QGkwYFusNKYU88@_>DQ@#lEnd&FgISCBWaT}ZyHr*^ZE_i
zi1kTYiOaX22*#g*MxGHGb-n+gcw7>TF5yKWa;AYYadGSpl&@K_fE^aRn4=Asy=K{>
zDbp*MRVTvJ%MC!wUJ3Rm_(jQ?h7T>dL<$m@Dm242!FqP=S|1*M!@E(C8stX>E@>>Z
z4C-L@fyNTouksm#nqr^!=VGI?*2;#VZt2WdMT=s3$c#DNJJmrae{FhqO57ZuK-LF;
z&Ga@i-ghG_M`IGPh^4I=3P?$krF)I7wyvqcT~IE<Rba~%oh5uLuYwEs^-+Vjk;514
ztek)-MNAJDBFuvMQB4=Ue+UwB@?0?Mk@!Xju3TkSIK4nHb_mW}z*{&f&_(*51H_SH
ze9l-Yau8jONq<A!BRr1!9>X!4l!J5XT&^YhC=Mn@uURVRMKJ15GZUT^*dzPTAIU%i
znld}-e)m<7^wf}{WeQeWn}PVJ?Dz0nny!yub!t=@wmrUxC<L3ZM-Xp##VL43!$rfm
zJ<6*m`MfBitv8-S16LANMdp}VO&lu|W9dx6FxNY<Hf^&!d_Kgr#~e#$5@<J=JIWB6
zZ>@WX9&YW<Ufv{IY_RUkT*Z3)w(B~6C-?h+L$sTn1;E6s3vZ?b40;Xy02Cy<sPXg9
zjVu{FMYCf|mH9K4v2Hy{qbrQq5*ylS!hKEtx$t~mw-{l;^Lfgyk?`o>wG6cuHb?Fg
zeH95QH9=1(?XM0;1`6SH+BG?Vho30-ZLu+*lC2vz&uZg0Fo2ZEJozQ7@f6SYj9+7|
z_kzsv_cOeCE&$0-MjgZDg%%6dhebR5lCipJ*(8I|SzJ$%Lr@P&^X=;>fh}K*o$Gwk
zCJw%Z=Ek77mV>b!d3n7@(maY@<>6C|!W%D+`ritC!V|M&F|szWLDi&Y)>^*Dc5ZT(
zgWQ8;Fi}h9yb?A?c;ZHu-w+Lqj-HF$1Wb)^5e-J$6mU#hje1GE$5#I?l5#@lNYO$w
z?iao*rO(X&<6I@b{BnWaYV=7~u(jXEcRZ7J?AMny7>X44O<%Q1p{>HP*BA$S5|L*t
zpRS>pZa-fXP!_k?)3dPCr^>F9Ok;8@^=%M5mzQ$(n_vkv3d~eKU_mVl4}rf78gkue
zaGh%rCR#DhRtrQ=Z`x<G)&=rEs5|&A@g^`QD4}l`=d`jam#LDJ=95?`aAg^WT|Ian
zP<+t8^~k3yUO0i%vNdoe$M2eyR!2+8f~a1M0TaUQKZ9&ayB&F}Q)nvnlR;I#_p>?o
zTE$;DXyl{iT#Q9vH7WTM$#sul)W&~PBD!NxPieWAcqSVO+xzP#xKXmVg-n0PqZdlw
zS5?If6IZ4$&Yh=Y;R97t$tAytnT^v_Ileax$hYH_#KT_cH@}BI{5{5GNdceiNdh<@
z+uUTM|LCTU>;@gV%Np_JYViCO=>uem2xp1y+bhz9U^OU~1%tFs?=ZC+kOmmg{J^HH
z&iUVt;iZ~<Y5tQwa--(V$l+cZmt_DZGDH_Y?_ftSu<KSI`^uYKa-gZhoYb`DFlzJK
zo91bh<Lga?y!=wqPShh6UM=USFj@#R3lNuNQmn0e6u!yHNMj025<H{R?*F1kc7Awl
zMD~+X<MtGmXAQCF#7T-#|DP#IhTw?CYB-~E>_=g@r4DVG^X@DR-VF7<)$5{Z-#9>V
z0@3keI2jQ`j}iTPf8@j%$H+@4KdLOCp%p;!<W_{7<5>~BH0pTyK{7p=f{&<cfqy`y
z3RiEP&Gyx>enAW=RZu@&a|o+x8qLsWgb$e*+r)V}J#SoRZc(&kys~b|VGv=t;$~pe
z7H$=boHm+Km@wlKLwE91mI7q6F<{a{euyI5H_WX_SYxkJs=gaA4e(ycMq3!^Ss3dP
z%xV7SfY^e3^xf)}UxXE!EoC_t2{y~5sVk{d2nwwg@j>Rv-{br6E1Fn`c?}7kf5BZq
zYHuEhVe(KG=W5nFlhzBlLRDY6B!snr2NV4qsa5%}7%5W1aQcec_3nsH(9PB`rJV3*
zcO<;>*7)w;L|~?20HV>-q`T~-bE9!?O6HmPqHaRFfAI}t2fbt!HaL$=FkT&Y_o@4G
z5Nfj=Xt1ZxP4XG0yo~~+8~bXkXXE+e^enWm>73R0XX&Sn2+ExOu2nE(7Yg9tCHT<W
z3<Y-be}UWY`~goE_|r`rw4DK^EgU8Gtwq<7>?-C`m16;|c`zOm%}Gy_6$_xQ@tfZR
z9(b;CPv|Oj>8NC=xA)g$=dP|p$q(cf4(AwxdXm?NKR4j{)vC{S?@&FvQElpMOWd<G
z>Rxz*vB4%T6^8S6MP$hTy!#38dKrL1Qb<4f_ORv&YxWRs-ed4_Yx2XOH;34eVbSi-
zFPXv0ITxJ*gCPS$n#toaXQi%;3>O4*#<-HNOA9Z_J;YCxAQj>X$GJ_r9{}A|{Yv82
zQ0vlG#gZs88nBCIy=w&~fZCXQQciZD10c6=X~-E;A>hTD&J03lMtMD#<gApVrWAXU
zUEA5Wfv2nAh4iHbG06!|cLbZub6=EQ0MHLuI#^YaAthPj3!;3AZ*O@|%8Z$ieX`U?
z&6_$u3p9Nx3Y9AKPMB?AaMF4Njws<>*8W2(#_zT&J%?CSuh%Qus!rVW3j3G<HFQzQ
z5z7xsGYOt6(S096d(o@~W36I&+avS(kXhO$YWPu6rUqj(e-V?H_R5DgGCA4?LT-x4
zq`TQ1P~i7=?x}4;YM_#ifKR&7htf4KE`1-=%ixutiyNDqX}Rm9e9RlpXJ{%3C0=dU
zXgudarKiYwac%vc<xe(~2<DujxaNCj?$DB|?cZP*aCg#__-}COuH*`FBvAZWqWm#z
zc_&Bn!y2$9vV_xgAg^`}7y<VQ%rLoo9`$q1KKY30Z&TUoGtoq5G4iqalPb>f))5-q
z5n^S!OX%9i#X7AAS{3F-_#@4L&Z&q+&e5jrVUPN;d!-pmL&UOoz&U0X)9`)_#Z>q@
z?_Ty0yIp>`@(9maeadoo^>3hbsBKxxzf}veOHKU0AOA9R{!K-3=9VzzXoG#24{NQp
zzVmR%?d(_((I_%S>;{O1X9%Qnla>IyaZPDc8N!?9@__q~lvuO|`E;P>)?7n>eBVE0
z3{GUORgl~v6Y>(WGwwrpa9V&asI>4{)}!}dHS6Cu=;d6j4ZB7Jvq+*7yLGCin~S@)
zrw{cN$WihR9(8D<ABO(7PY9eL3N!$}LWS@F|8;K2u<Xly60Su&pGHkY(ModuUPth9
z@azIOQ)oGb#fFt9X}Qm9#N+TaN;^^FjwMzJiy7SSQXbZ9xyRAbV;=Sn+Qo(Qa*Ga;
zOhJ!k5Tk5S=4zfYXS_{3W5L1*;estXTEA%y;?%~_F$}Des5dp}Q`C8mv11IuznV@V
zo_LkPBwp!>HrM-o+bIyNm;wrH`1u!tIj7{L@xH<8lM$fMA^GJN#O7k>POFO3nWXJA
zf>9IcTW^v!5wu~z%EE$>p5YP{EZ~jezg)Zqt{pfm053@znpjYjO9tmWYn<kx5orHT
z$YF>FdsSc@{8%4-WU{M@WJi}?R>=f2DHKnHl!s9R3k8I7SNz6HRLVIsk-&bJUbBr}
z;>6fzYSr5?818GW*?Dlrik>W+$q5Q$+m2Xtvy1Zn#&tNzmfY9$xq!q$4siSijE8s)
zD+MkVX^7$)TG7_7E<3gH?3G{x&^S5IphdP2$}1Z^5sRR*SS%c+q_~N~I9Ulu?Kq{2
zTMio(K_pw5G7}fog01{nl&Yn0@;{8-S6`9WpgTF120HvEP~ux}!<RNAatJ2e;X7Tc
zz&qasm+Hgc^UW)~D9|KGyW;zP^6pu++r%8(OuLPx>^v?XT0Ey{fVw7p|73?)$xEGm
zzinXwA0FfR%R^U_Ga%7eKiO@p6}~9@Q6w|Qq-Zm(J91!B4_v^43(*<Hhzy%gswJn>
zYW5KrM;$;s)&4d(y!i~)h<1P0^PXet+_IFJbb2juzoMnH;zeB5vpNi3O{4X09E{A?
z?!LL$>-e-ljab(LGa7HQxh2Tzk4ef#M~|9%{Ym;Hn8SI+{bi-9K%4p#nDw<FFP$c3
zW@OFnacqlD(x?gd=HiQ`WypXk3(gieG1Ay~Rv)@!s<0om2P)-~t+$mOroCc^+^%dc
z4m^l`oY{-tCJa4~xk&FU)YNLT=J|m2ak1q`>|)+R#h)<}3fPRi!b~%ThH`()DG@}e
zt;gYkYlQb3{9FP)j<LM6{@$X2Gw9{C<5nSq{`#Rb<%mwX<BYWz4}-`~)uv|xq|=rT
zxDNdRCS-&wJ|*L$l`~w6nh2T-E&(ikr3eTnPWjV(hukWH%_r5CmSC|^@F@R9c+lg$
z{oxPVGAVvUWP#^405WLl8YubE>T0&wTBUvL4M|GhrEgUe3Lrw<BXdBvU!>(*fn@q$
zr4pZ*{4FEUfLGhl4wff9Sc39<ptq5*Jd48hO_Q9RcBXxcuIqTM`xC(Ifs^*1ydf9-
z-vk0)dNf|Yx)Um(XK++mKheb#UUSc;g5O#de?3Q;@-=kkn|uV;PGl^lb?$v={jv~!
zD$=&B^*{Ce|981DHzJ!e6DAQ~E>Ut2a>Q<Fo$9NNF0s!FKqtemqs10z&h_wcYd#Q*
zdCXe^Ux4Bd%f_hPf?z#>DDxkDr+6^hm2+M23B%Q=@U3DyiVe?BHAE_LMqUn?>VFm=
z`%N^nf#2iOM<i*@@g!sIUknO~{3}*iiyhBPO3zqALLK8X-y&j1JKqb`C^xvgci3*R
zO8VHmZ|zsFaj#$|Ps;xJwi<=J?Nw_E&A!PSeMd;1Bd`PY^INS{ObPx!x*48WOo}97
z+K6N)kU7?Uat1|X*M$9qsGuP4;6bum*G%mi(`r*%az`Z8@iicFi0fQk_VTMt`O@xj
zwiwf)a7%?})RIvu3)N1yve-kx;O(X3kzqiopWlJ>HIK<m!`Z%c8t_@@x+@w9;s}fI
z#(6^!t~|k3p3j1cz-31_>Vvy*on|H~NCz?Tm^)?<^kfz)?AwF*mS4~-P5G$A^`+Vx
z8)dlmIE$n56sA;(<5KDeelCgGnh1IjBgSn~O`S|hT`HX7&71icvpDAmX1vu}C%8!n
z35C4CK3Gw!#pr=`rRf3@181fJIypovjO&+1^&NK7n;F1~?D--CZgUYNZ*)|78v2ZI
z^Mr54`uVq%wfO00n_tgzI}Q{rRs@Y1JC3y3ejNOKHHep)?CDTsCxyF)p6#p_uvxZ9
z<dhW$C9it#IFYK6@4R#B>f6~)k;LnrFrfO_a?ImB^h8SzQwvtx{wg}2r&^HgM`;31
zwviSG`}PENO`3(Ps#Q%x&(4}RQPh67)dsGG))vc*qiOc^Z7gB9$r1;B9t=|LAmGcv
zNfVEQ58(ySP}vj&(UYmzU@s)BVSn1LzjBB;_DYmSb=!H>C&DCWYM%TA<_bL#UEDU-
zIB7|`n367Bc)YPNyMm2rX<x1g54Nu4j!yjpp*^j^UL>XrmGI1G9NN5%9D#9O@V5!D
zy<^9w>rl}+ZnW$ARs{?8^9krdKz|ll{C!+#E9sP=Oe2B0OTmcV`T33aycza4b~<YO
zL*|#-l<{Q_1(sr3XVTHgnRyCJR3`tt#B(eY1s`#qL{~`4s7DD@@M>$jrAnKJh2(Ps
zdaK1D8$hcxPn@-{lyZ%ZLd85=FVtis5=4n7uw_W=ymC!CtIwXalFdvtC{?^}OVs+i
z&jQ8pxi^^t@xIjKTD%J71=!VQ?uOi7E*X?So>(&z{r))RddHu^k`q0&0C+a?sn#%d
z_8&4^*1;QWl1W@i8NbymgZ})s8gPg4xsM(rEg?<ELUv0rp_6oQ;yEDn&+qQY2y3W?
z4gcZ5{EDb<_j7)0gY4pW-vT8=4;`k15w{`Vpe~C?<xXc1x~2Ht0RDv$93$8qm$=Yu
zz(fKFlOv=m6I`HL>oddbiXx8ZdX~laUxDrS`~ZaUP@w}DO7DVNiU0foHz{aWj3Uq~
zc;v5`FBSYlR!37TAuxJmeH&!)2`R3e#LG2k7%~JJW8nC~%bF<8Z>2iDuhUNzD72P(
zAMc`CdjAY6?~$&<r>RT3Si2NisdEN3TS1V7e$7sei@Ha!8?<5_gP|5HqdXQb^4+_I
z4yKrbySgIB9$Eje)?S*@`N<WS^w!EhtbR<0{u$D)f}rs)d6*RJytEiEKQ4hgnxr__
zCAXV`Fm3Rr`y;l?`EwI6xLxRB;LLf|b+yf6yN~X&&ti)05WOC_ByA!GT(7hu_Sz$F
z`Dw?NAz<*7ykXJSJS)E|LHKr6t7r@*e+Z7U0IRJPxSY28+q*>dKw9oU^vg?xBB^^z
zb@h6zF!Hfn#apOes2@`T!*5)&c~+w>=a?z5AH`J|9}oziOlyft75@vFu}e!VCu8<*
zy74K?iw)xH`>;|q`dCAOHXu-ou>kEP6Kjbm&UA@pqx?GG?a5u`1Z?O0rXDzG$a80E
zr?)Cai8HE5hFQNFmz&cIVa#S5GKB`wJr6y!9qFjW-dJ?|MxgVPQ?XXVg@$Nx$qxI{
zqPp*~3V8F{4YnjP2U?JPN0I0)OF#V-F?jX6U@cHy`?oUqG+@k_2OT7j-mCb@e~zs9
zjF-kBX##cnDPr~w_!nhn0@8jGN96^KrI{2S?q1Yhxmj9`@yY{EhiAD>n&4PiKAEt5
zg45DgJ*yPy!#}X13q81LL9RU)Px|r?J-D;roj%i9!aK`8*bC*vB)P*v5cwuUipjHV
z74`E4VAYreY}u15{10>;2ywv9NtE#{n7RG~E{tta&u8gSCTDEE)chSaqATBrA6tYg
z^HXK7Kib6>9e+zMIn^`Idyz!EzL?9j#UEpeoYdy!l6(lH@}y}e_<xPTX4?Tfn8e+F
zAvIJbej<3U4+AZOEu;hCRNxoGt$N$_D>GD2<y6oN<jc{Cy_ITFmeD3V=IsQJu|M8R
zx_&E#&9;cQFh%+kjoFLyYWVWPOyLBtOm+VwA8W&mUBQKdZ;<%y7aY{`@5fVmP>a!0
zH(-i$4Nu;{UL@Ez=H@-=T9`VhdLWn;@&H)*CS-oGwZLOadr$Lrp~D4dD1<<4%cKqH
ztdow(vlncIT)1Wr#D3%yLZEb<0Gi5!A)?&B_G0YItNHn%o`0q@7cooG)dnH^NC~$h
z`07K|fH#~KG`KE54fUGeYStVNRLY*?k*-Sn2(+xug0u0h5;y@-YbNr(S@(d9bFSD<
zL%vBD6ayji-=rxGMrSJkAA?(xg00yUy4$ENvl5~DJfoQ>%;1i=;%q>fqU*-9djbbX
z7ms1peY>0kL8nvm4$Xo)l}U<nR#dGBwpKtsj9r<Ecd<>?_nqk%%l49~$5z+<LeL7Z
z;J4<i^@XaZmU#*w?X>qWpZ9KqM@$x+SLigJx(hsn-Z@QP2$0f>S{{}9`x)SAno!HW
zmi0o5=@2rGKcd#FRo=GF*mr2nYq=5*O#hNubhoFdD65<l6@j2$_evM~i0PHdy#O2+
z_;b^~M!~n4%W1KKN%~l>`36e$TD#@QXvDjwrJ2N23w$Y?sHQ1Ri>@!}AKT7X{{WdK
zEoM#|^a5Y!xNF3Na|*t?Y1des0|-z)NR)hnL$;qh+Kb2#NYwmjZ1Cq*rl>7a^K0PJ
zLdFCI7Hlu7^mkt{z95FDB6gH7tH%1to3ol+?I2dUFU9@1_6={Y=9HbLrPROnT|XHY
z9zM2zdj<_yF?I2Nyy?<>vZ1I82u}~blNKru1065&A)j~uty=87<3Jhkg=O9RckTTe
z#Wb(5J<`^nAAVf=Irlx%PC$yFb2--jM{m7U)}#KtQlNFYSjkB{Ff%zj26lCM|CUMQ
zwl+NdEN^_b>oy?}V#lxI#NbG`QyRmuO==h6NiR$-n!{uL52=f}w7zrf@agkP;VC^e
z^ABp9ix|zA0*U|-_m>rO=Vt<YBLd2t<R8p!lk6^GhW}z`sW@ZT&4v&ik=R;k=fFdU
z^zcG@FKxxgpC}3L4zJaM&DfWw^S+|k^()+jd2XGHQkb^F*%h4wjglY8mFS$jw?@|w
zWDG-Q6ch{J4H_BLtlzI;qr7FWj=G^rVt_I2c@O-$UvkC#8YizWAtj4a)G@3o#TZCu
zTkH}iaG*qN$vf1T<aQbTM)0u-XGAJudLTZ1bY|dXsXCsy76~hg%2WzGeMWqhe<j=l
zZRF}*fzTGkC`jim`k|+|nOl3*Ck~{`?D=F$rtOOCxhSLU$DX&&KFuo3GyW(QpbJ*$
zs7GRNm}~~dX!u%YKC8tb>INQ^CM&oXT;MmcxH62&WYO6dS>t#$eN%RGCqeOxs!x?<
zf`zXabL;r_`}WASEn<GnYq_=~;t4LJUIF1(kode7ja6^yS$qC;q_f%TlNu=W7L8IQ
zeE8)f%w}i(<L@Ug;~ptPH}!&Bn;RTu&7L$#1HbSjil$55xuMrCQ&k>P%VC_ob!zQg
z*6Dj*sj!O!s?to%X9b!Ksa-R6Y}B%0Fskr#qI!Sq@Mm{4=*?Q1$gE-XB0*-wrKC1p
z>~yh0JBxEBF-0Dm_<@u0g?Ps&D_lM$fr}z@?qm~c;@Ml}Vw<n(4>ezGi%yC^23#4s
z=?SRu@ME?}9jdogPM(GsI@x2xPbB6<??qZfD*ZzyeZAHBAF^vlci&&TQ{)hi-8~oi
zHN?4h!m)&Y^K_wy_g-tBp0dNjbdYmnu(4|0qek?=yZoHH+6H`0Mg7KK$7<j9s1=ag
zm*22`uJB$(aAm<KSDE=$F4Q|6nd=wZQDqAXBdf{j@ALN`0uvn=)>SYbib!)jV?%P1
zRRzk4G|Xs9kepCeNR;Y)U)j2{_j}H7mB1r09Wak5O{GSj?Z%+h!zS;_8`fEJ9YFjT
zE!1larp&jaT%El|v)ln}X{3(t7xvtQralUC)XLjzh3MzA%8K-g?{p=z_sXBT-Njk;
z@m&p!rdao6LcdT9Gp~%mYAejJ+_<M#A8hV>=Qa)@kr%b$;-fX^@VjNH%rTZGYrYY{
z85Jj`Ct5F&VWZ*Jg5guTM)$ksN9s7n>`4f3H>n9<KU)xpew0=G{(4f<EtndM-EQ4I
z%33rPyltbC5OidFmnA2t9S{p~e;uaKLu|OaUop_zbed|gTh{UZdbSB)tHnaz&^{m6
zoEzgg!Jggr`@XIERIIPg(xFAz7)TiP@mtnDHcNtOjiA7<pM@XTsU9tp<f-$6yI9q2
z`N3aTrLM?&^#<itQG#P|AC0YJ;+~-J{yL!o#jE%WeDMjsB08;U8;*bWv$!8i&pJ#b
zj_c@&=&jb|6yJZ`V(1jVWg<^;^Oxo4(CKr4)T<c2=h?T)YMHNP{+v(wIMeV3tVhuK
zD0=79%;VN~oKDVbQ;z4__r;^s$p9mFJg*(6+Y}T>zP(eIfboGF-89(u8@zA6G>-QT
z!4dQ?d71xTj3Vh-Di0%ZG@T?k$sIROuy1<!vHQ421-X-6W~D=O#&6JJ*)i?t&9xNx
z)=sn28}9woaI)K5bFB9v&m`-;lM`0sK%@L2gH%fz>U)PrQrw;?po*=n&oD~YXr7&%
z7bfb}%+JCMZ8J%A!m{YZ%e9dX4ElhY%V+}mdm4O@|5j7~tAN}g$Ybw?TJA>EiFhAW
z3gPY)Hyvq_+$jK~aWlH*Ygs$2W*If7|2PYrT$`bhQ1KXEIef}nLCMo8Z1KR8AkRoL
z7F673D6Y!{fxi5ni=rmob0oL&y!TOHQe%|LUX<#*!W?bn7Q{3`4)9pKjjc4*N{13Y
z671cFq_IZ(Pa7Yx02BnXWR@}@=O{0ongwA9!<%Sa1-ULLYOXb}R)Yu7I75nNR*a+&
zsfWOOVx=z-s{HILXdka_0eRbb+blI}Xt?zP?|iXV`GDnDuxLj;ur34$NkaJJxj+A&
zU$m2fjB@#g;1TO+quLDF-rp#zTwd)?;p`%b3d(`ZQqoA9kC}<R_A?;7L{PAZw_d#Q
zea8mY9{Li_YVH^ts<hj|`*d%E!@_#qnI)0?>Txzjqx+qU-k;VP9Ib^pzq_k`mDNHt
zihL9XeyRMmqMidNQ$oF?Ey7k~Oj#y@cM>ttcqOQqU}9&3wA~yzuiH7^>6Yyc(j!1G
z?+x=t8movfx*-)!W0Go7k;@scG#PEC4~0-$XF%=rRgz|?t-}e@o&<B5nlamya7)As
zeyf4evUY;nHC-PqI(K1y%f-o4hN?#CU~RLswsh8SzG$^u-!ZSUo;z+zA78!4&%><l
zBV_g?4~urqa~#IkqeuRJI;gNBu9z+58*G{et}skqd7k51)QOKj^2NymzjUoz#S4iR
z4Rx{Q`Hu@X?lN{j-u@_ewgg&=INnUU;I}vuqtK#OjudXc!Rk^c5hwu9G66q)&l%Mq
zSe3kgMiD?f57&S*sRe5=bAFa${R~>)$djQbJtr%avI;+=T9Q*>27I8g2Uhy=oL}s9
zM)MwsT7P^90JKoMQ7xnyS3CG?)XqLB^w9LZpw4AJ`^n*|HE#(1nDBG)jJd5L?eS+d
z23O!I)$`Hma9s6R-6;2gawScqaDJ4j!Ol6AbfD6SxF}X4?|9g8sIn!}fJB(peX5_R
zphAhP-=u!)8HlaUwvVRbLhnZD|3<mpg?y6y9*7h(RyY8JW6*2!$FZ!GDL(17d-sgJ
zgWin2!3C@QW`T5n+hRUXKJwvo)#K3pl_6TYvb<xw?%adQAjUuEO3_*`IIgeNYj^b(
zzrRI<cl2Dj%IvytWi){@)Tq~|g@)3P40%^FUENHRJG7v*1G>xzW@X^3rmE&D-7Qwy
zT>^c<f7(}YOcMv^LdyMrp%jeLzp$61i}MTbAJl_?$dFPRL!nhKewL9OHoEV?Do10f
z6z$CWH8};tXU<l#aRMXTLVK?7n&D3tMa}^<g%ti0w9B<O*+CbT?aXm?F3q0`nq9g}
zT-xOVUwz>|XM{Pg=!nY-Y4QI<rm~+ulgJzXsJ>T;@G!&AC%$l&-HszAvz6#Z3Wfg<
zyZ?Vq0IAwbe>t~Yg^Ni;tD6Vd<?XAJq}mt1^<{-fN3(M35VX?R8)d4lf;j)%%E&!G
zT@UKrd<xcQ72M?HFAAIjMa7<#7nqSL{+JVt2q~c~I6u@Sztu7%{EN5|;C;D{Ws+7w
zeG;#nG_vGTS?@nK+7d$(%~qzex`+(Qz)mKN!T#rAU^PGiE<Z}@73i+bP*0HqD~g0g
zr!9<-@&zn%r-NervYKLLMqgnvbIyTSvdR<Nn#qnIq#^*d85-BFJSgCu;C+i2`$dDU
zilWXoH6J1;+t3R&@9Z!`(b)L{_+8NnUwSJPpxdudJw6vd294UT0SB&pMfY3<j7@c9
zn=QopXXbg4RFv#!c2ajFN)Xf^eMO^$dI-hW@uq*w#~=bDCc65zGK}A<aHGd&TswK#
zoU!*o#VO_e(h-yl+6yxbFo4TF1dJl3L+;U5;yjbkQAvV6d1V6-PtJG}!KCWhAW&W>
zcaJo$e$MeIOsS*Xv(rYo>>j3PaDP7Q-X|+F2uR1%Jq*V<wj|_!KI@#kXBx!)(ZFXW
zWtj7Vy>?A#rPbK;M1KA|X6jDce)w69{P3@1k_{3@ccikE5Mh(>cfC?+P+waUz_^3u
z1*=F$@FooCiD7xlc(>C$3IRjma}ASx%4Q?<qHS+LNq*3R>@vlZt@pO+xBemPe@?c5
zx12#p&=m2K!2RgB4-mTOwOaVr8H7?e2nFP+OmLIwy$?lrPV*2%brOORF<*rYW}a3&
zQ{a8neViAz@uIXC@#pkOsoya+A5I~@5`i6k5isj-lX()nZ=2(Q|2g-Ro|SY$q3*W9
zhlao{_6QK^4dODDcWU?g6)L%pVW;di2k6NB;a|P<D#{ShTlAi;Z?{{XuD};_I8H3`
zqDD&*FrE7dyjoXkFTPQ4&t6nnrzrixuBo!au92801r^1@*lR7(-*Ewr3dZO>_h|z!
z`gDVaXAt`I5<FR@p%5Chq|IBBMkHY{k%v}1&4~Dse1df`qdazjN1>5UWuwMmot@g8
z@tI2(hPZG@XC*4uW0P0PS2h&xhBW>&wcksd(jp>y|0R7iHsQv;VFtEDh)ys(HGHQv
zh=veQ^aAUZ?QWD*|BlnX0BQ?NXUTXCeY~2~v8LHu#XfE=8Gx4fs5$Aai869+y>S%h
zgI&H~;(b%_X?qr#gSGhTRtU^i^yYJ@oxP@%a1zm`Rhs82DKcF?0?m`Hd!7|?J4<@0
zJJ|Qm+}S<lg;lxE--q?_?_3k4xXCOGbEnF;XGVz|%e~(w5fsfH_*F(p&RRL{o!G0_
zG=t>(Chx|t-=t}I)y1PQ(+9|}WduhrByhsbz9$~PUfBEqD7`mVS+qr5GJNsz=e*u3
z)HFz}VPyyUlX(2M4xuSh6>XY&g2i<d4P5!#K-^!>)Dzi^huUoi^F2asH_ES|*Ig#@
zuTLj_S@MRkKY0~jd;5csoys$<`9~AzsjiPr5!R86Wuz;-cN>R!ruQAb1pdJbPg51|
z+E0H}tPVl;8BkC=P@*52Pk-b>e^TxzIg|f7_y6q>U77$SRqWp)k(~7Og%8(JkWVG4
zlF$74rzg(JB93QD-XUl;`*s7Qyh`m4Oq&Lz#!M#IvoD{4fn(%Y7IUI!%!XnG&jj@Z
z@MIjr;k2ZXAZyg_YdeHo_)Z?&h*CePNDmVAX|HaW6grRNj54@Pa!_9QC~P<B*Gb=s
zPrGCEgAZ%`-c|gqa*cw2L3431(zj$&+{_T7Z|PvA+DY>U?aP>W>l`+IgIe>3Omx$&
zKgu;t{9l8O>9d-yc+~6tE!oJEQ2NTfMJx?AqLJGv_opV4{Jha3Yk_i`bJdx?&_$K-
zbSXiF?fTxEjn2P-#K%`ES__jB`BynSh$ouI)?)SA_2LJp5H*BUA<C-64Gp&1H`2!U
z&WLzTl2P|3VW+!@Vn+2oc5eE)6rTF4&sHTxD}ULZ*Jr7+aeoM8+Qm;oK7{j|?fBwF
zikAh?p8i9o?sHTA7nd>r%#>0u<edm@t2S|RaPVDLkwVK@6t6J}^rnPkO39LONt9o#
z)c>=`qhgH3cnG%M;1SLZpChrJ3Zv9Mga_oTVKv~*5yV<x#6l^D7_mCoes%#PYE{<5
z889ADbMvb+U0t9MB?R|{|0vHx+`{b-(Y}lKYaTHr#vskOa$q$;{)c7yXAmdSLzw9Q
zqg1|(W3k}X8L!A7AfB#Tyj1{^HKg0UsqUlo)(m{3aS}x5w=mba*@Hw*3X7YYa^}Gw
zD&bZ&n1E%$pEOuiJxao<G)S6g5vYs~7@do)Upy~!g9yjawu-NdKNL9B89|Md>4n?;
zL$=5;GclVPmR3EijV$RO;zY9e@G=DVMAsb_1?d?Ct{9=m6Kh>QNK?IfgAQI1Ys8f@
z&5Wp1drao!{`3T}e$O*?&XI0=m+`AV?@B=J@f|h^EZ0D}N7-v7U*=dV*-?f5uPB?b
zM(KTP*+iWvwa8K^>dHp)vwCmoeAv>$)B9OLbok6g;5fn^PYoZYG~n}7O(;@EZ7F#6
zs742JTW;bY9)l%kHEe9v5<L56<1kP(z^aBpBm;^xGYz&QT}akXq)chm)%rE$qkTg?
ze=LnXaFF6Cr@yEbp3Sk=;#5|I>gfo(8%Tw{r=fi!faw)S7m81@psq1dO#e;iQ?jbN
zW1jUXsYbD50hTs)I>6J(%{^&p-ewexOoY1Lnc1nc36?T7w_=}!JFP0n@7s$#<-8B6
zpj&Rz!2DX+c6J~fgxG7Y$G}PVEFxQ-M`}JfrF;lMWNIlkhOaHSm=E4US~(!mYSn|I
zW4hOB)(4BjN4*22^A4s)_Sp;GYN`ZV{#{Gi4xYnnVnCuqJta4PX_K|B>NmXAs(08Y
zRU?X&p@82{^+kv4z2%2>ZCiJQ9~s|mJJ9Qq@`o8e-TicZ5%VM=!b*L_o{{q6NwUrD
zgX#phr-Q}jM`Joum)F1BYOgFOFl+xqhAu5|w~4SCcX91xA^}cy?%{Nq5n{(?UTy4~
zaTcGfwz(t!c)vHZ=hB>TsdfZf9k%i162>^{#XlTLXLM?HC1M^Qy1g~}%L&uIYt%JD
zH;nU*->7A-8(7^?3n7WYte1+q(|@G={}c5|H)yCxD;L6og6AjiVE3<qx5l!u6NY`+
zw3Vo8hx2-6|1s@hBbT&aMPTxHU6A51Pf50tQ)qc7|D3DV6E<3$$j4wfRqw}6+8(OF
z15{cRqZE!m#Y<M0XX~$JVPFciu;(L~$G@pnvdhD1Gg3}BEZP*$>b&nhpn41Sn{+w@
zE0Yyig>8?mBdMF7$x2;7D^tXhu0@!|LdJ{j@Twffjhxex*hseZY_;gQ3T=VJX`EH&
zx(A*<>olFoF=~fHTLPf*qsCp$f}&d}AFX>1>9yofX|1C8U6du`6&bV5Q&UY|ux=TB
zf8Lkuv!WHZ2L?_ilx!+UPEl!pmBWAJ@h4zCF6h2Wc_WxO3NP!o9M@n+W4y?TExn4y
zo(u}c6beuQJuj(e*vY3iJ9A{yr62Akxz|Fj$ymwqMD<ZLJdaoaIdr+XVO`|~Kmu@o
z6R!&(F8djUo-k<#SW`*TKV{G=;KT&8O}!T3IU{wGtXoyR0$csYj^vDWQ5r^b@fAa{
ztGY*s1{QYAT7`Ym{Yv@%$cw7PuTtb2+J~mr5OU#gjN5RytE4{aOj6hyDS+}c-?O1W
zpE0v>uOop~*To+LrV}53#K}&~(30LLD@@&-XD^`s_Z>%8J`d!d_4F(xdmFFeP9SGe
zBg)meyiaE)FeWs4UTM(Vkb3!~=q}=QW|csL4RBXS<W6PmE`b7unJWTusX{WNTzM1m
z(HvG`CW=^k_dwy*RJ(3(e~ApqG;q2|15xxB1P7;R+D@ZPS}47%+1TL!2;ADYw#6*1
z<4!X3oT9vE>i5Y6Y7O^1UG3-$@x#v8Y{nnC(5ZJdJR$?rBqvYlhbc6WL#&y0nwwC>
zHD~dG-y-{?!1FbJ<x|bS`!yt4DNX1T@6>IM>!pn2J&0P+Kx{bgnz*7~0HWiQ*(&R7
ziZTU(-BUkq%fiJ)=vM1J7sH0JIyvIEM4Lc|X<bxYvuC`S;#cgl?kN9a@X8`}^$$GE
z!=@GCla?H#WzE4S@O77hp!Wn_Z+^q^nc0+5ZDKebPi^B3pnWS8JoT(cM8qa$Qfmm8
ziJoVSvCIvE>3b#TKpkNeH8SyN>nkgvHNNBXN$6P~g|<@BFC%KmyV$|+;>bqD?5&SQ
z`r0>Td>Up#&U_wCl(fR`)cI)F%6{0JH5CF)yQo?#2m3BYXLT{hmocqP%+*;qj7+AR
zscu4xZy1dhCEM$4VO&1a4JrTdQF&+^@J4N6z^KP#c+BY$cr6q6sh#=xXe}&6g=;x|
zJwSsN@OVK{2>YfA%DyDl8GZ$`Ot_MDUhgs05_61Fox{Bgj@!w4llGd<lM)&#)%JCv
z1u$jT)Ry4eM7?i2Kbc{HxWiD+T#2kSwCB12JeS4@`V{(yKFR=o8BNMkTZBK*r(`w0
z$CRRf;`46hqc-KBucn+6JsveUxH1zxE(n&)D!d3ad(z+R&WDuz!(-Y>xN`610_hPI
zAy@qAMXTsnyZn``1|oN=;JNdYh0EZ+)JDNekKDXFuisilp~N~7OAog^$}$5U4Whn@
zs-{ido^^QAY^FWIR%|eOf@%{Axg83A`QhvI%EEo>Muk5M0OXmAg5Hxa^-KOq;dvJ}
zH*l}Jn1X`jL`Ic&(fc|53Aw+369oZ2rR^>np9b$~db!!tqNPjHE>9Bw|K}hXNM0kh
zoU=FHALhCm)A1H2qR9M9q^4But@s7VSGD?#)AcK!M!=XqlytB_acQ=IAsfOydS0;_
zJ#q-AOj*k;RPn%4U5hBnbjo;3k>^FXOB#s<>j*y8Wogn8*eEUUD`u28XJ3ZT$emIP
zI-QruzyhzIj0hYGZhzLVH<(6jScWQ%{sDKVX->M?G8&$G)Nx?Ryw7P$b2l>Bth&&J
zpE1k>Gt{0_$~8eZOy}^dBfqT1gMSrRxQI%(jzau<OXw2E%^tzz1cZRCPQP;Tu395X
zliL2>l~7@a&buo$Xk6dAWm8&Bxg@}2O?FC!sK4jU>2F^R@2zE!&OOkn0lQTwgpi(C
zR8jM95ajFg%|<lcepHKs<<)GNCar<QJYNr8#uuH5JFNs>C2L8ZS`V*A=$!tQdOhK{
zK(}%J@CDJ29sc5o7%X!Ir)x?kT006y$%)pUeaEav6>haupRL(u!a{&d)XeV?ZABPy
zIXFbNdP@pYV<_)+D_Klv#O&)fZyT4ZeI>Wm=E(aJ_mQGh(HIRb49R$lBb|wYQEgy(
zWUd)TFSio@?$uAoCtH_##M~L?;j9BN1z)7@7=>*@N73On%jr@M@T*F7&*Nnt`J_lK
z<gc;k`e8ZV#uL8rKnvQhsRDM$tb)zFV&Cg7Tg#GhSA|qi;+cd+F`5F()te5y&MtwZ
zrA=$bo1D+m&q(LmQr$0nU~d&5l!Pt!F>TV&bG661|Ay;_9?08`-5`!_2?{R=hZ@(@
zFDs_n?Yh6d_SoERzUqE3$g(ZIFwNQP&agR_&>d11+s@}IDI_<j@0%O>bAC^vbIyAv
zF7f#TyGoi~)5Xq=mw9-L&^x4>Phdax9s9@;$rPH8{8RX4EzzIt`ggxZu@{e8spq**
z`1m~m*zctGXRJcQ2{==yCAQdZFHxrFr5~jl+Kuq_FTg89#Yul7dnR%ide22A{<sOV
zkKD`=?c7RsF=)D`2@SfEL}Ffoq_Hq;TRs2%S}RPoD-^LYKjAxX(vRA@>vm-^%j`N&
zP0whfnlH7T-n!FFo<2uKE<WoIlPY=uulN88eicq!(Nfv6(khs1)W>vf;2i?I>=QF3
zBk_QYph25ry9MqM+$lV4bkZheM$p7Z=g8A1$M;Ax_ojojRn=5m<0io!VOMDB95UYj
zgMWRaG1Amm$W(FoJ;5NmGQDan<!YEK=`u%ilAgb*<CDS=!nAUB`F#u}^p?+)o2H(c
z=_SRBMQ^}bHwKJ-_rA~_Z)KjN8a!g*j@#S?CuGQl8Lpo`uJuZFJC--)k9Q;Q*WUH=
z-|6I2-E7!f-N`c(dg)Wy+yXD>Ikch<0*{T%Bv6ZHhy9Wbxg%Ae_HqscAK0nQ{pwR>
z*Ich6($4kDueMTI485}(9p&b1wvgw&swila;6lJUMZ^0@0CX{~cv*bHHZv{G))uSy
zj4uFmVlcDV)NStjRlb*Jhwwfwo%P8}A7CrQWTz+fXF)iJ9oOryn$pgKZ!RnRa%I=_
zFDn0vJ{}=%DV;&po&@X^z(nvfb17#7W1s#u@D_wrr)(R#sWT(;g(p$pMEtd{|7=pP
z>1miC{doS$LZo<(7$_B0dwrrt&joZ&-FO>k{ZgoFw0`7rI{IE&asVAJb|RuQ4pyKc
zO&$up8z1#ZY;XI|un^~Oat98aturJJ^X1X@|62}8Vi?IEr6eUjL@pVHbd`HPf~O>6
z4Ts#Psaw8Qw|>7B|NH;2^_F2xfdAk2=#Y?<7DkVd*hoRTbHIQBO3Ow_2?_=slWs=C
z7#lF8Lqt&;895pZKpIp+RMg)+fB*ZrpXWY~=N-J{7{_<lb$vgd^OS%J=2&yOqq97Y
z`iF!D`54@FF_Mzgq~>dN`lgA));lJitZ(CI(bRMY=Eu?n38Ej+tgcTr^vO97MH<>%
zo&~k~lksZ={VXn_Z^rtzC7&w#%!^9W$jIHj8XIUZvw0Z`o|CGIJ^bb%K}Pq<)<z(+
zD>RSbO@x<U5rFfLv35856jKH$>nB3jX59^*6;h+Z>*R(gWyYVlqA=@L$QgEi(mXCh
zv@)wuuwE%dXZXH)t;c}{UEOUNBdQnx-pinsDFadPk%7g~1UNBnwlj-Q+}g>7zB8CH
zFeUIi>5NeJ>?=P2{<7Z(v9%~Mk?Pxv*a>Fc7{%o}+980NDg;=HqSXx#;r->PdNUd$
zR>}&;DM_Tf{7UGM$}wHJSsRd6A`MyF9TF&ifb-bV-p1!!?px8-@(k&&BnnC!(ull8
z>Fc3#q|(yoP!xU4{tv#(YdZM<(DTknN(iwS&ihCb|B2OE|53aCp-&`(x@FH!j49ea
zt%s{KYQCHB$51BUvp-o-`u!bFz_3|jKP`d2#s8W*!S5Pn@7I~u9wOY}Y*>wp6jppO
zmzb7z-e?>+Vfk>W@Sz}hJ{!8Pa&kd|p2!o*&*7Kd?YiYn;%s@_Mcb?JDal_0HTijH
z7gko^t=>qq1<n=Ow+^$Bm2YJ2n~n&kk`86#Cl!HUz)*XWpxU&496r1|00BN?@kOYm
zKed-+Xu?zm7XNapoCkFjcN|L0=9VMnn)AFvO^+cb-<m&0dq48pYBZQMZ89Y7Wl{9?
zG<<K~){)9>rOt?Xo_{Z+6y~t{5cavxznmYfm3H<tue=yj$jYEK6qM2<*sYm}==n3i
ziWKkOaQ(WU-{ZBfXj(5JrgmWSO6djX$T*pr%#IyfqI$F4l`5NE!h<ROdla(LZlF~4
z&a;Ak6<yby4c&HY%QY)h+xO>gayT^q-V7R4RTrjOsI9{DzwpmJfR8UXA%en<Jq`Et
zRcJ)=f@-PyZ2aS*__}i?Kx?XA=oY~#3$!KUXIxY47dzR%XNTZpi$a0(2E{3~tkPO2
z?Z-KB3jvggVI-H{Zm{3JqdxG{3sjtDi24r1ZzzXM>=`my*alOxoIBF6ucH%P=NlL7
z*7oaK1QMPL&sV?Qv_wm8DGAt1;@dk?2X*~>D72O`$E#mkX0v{`j37*jWHJ2iJQ>^X
z;UZv6BsvNtc0libIe^8+UCiFrygKAt?)2CiVk9rQ0#@dZPL=Ndpane+b?f}Gx6)Q3
zkDljQ*!fIkUrT6$Mbvo?XDe%+1@s@;<+|lXGWNTpx(^KvADbqu-N$mT!<JF$NeV>a
zjj94uUZxNSh@hH@YUOq^f0dDID{1?SY_aM1-+X;U%BR9ic%Eyat)6$%WQgh0>lJ+2
zG5t^OBlh*-Ty>}|znKyCmb<x?Xa9BpD|1`Y0=C+r^_733XEebU^7obm-u^o@hMn>i
z9dlfX&jFd3>{&iCxRu};Jj)t|RzAtLWYjliMg@PdWXQ50b9(G<tm9qsz6-K_M2Q_a
zPG{6UR2Q>{)c_68AcB^#!Ccmy`pSvx`j$cBEmB9e>+-t|O|)(nxFTM(@=V^MOjNJD
zAgcppUc@4a`1Fa%@!J<_-A`-7r<dLwyWP&~=#KjbbD;nKbQ1pEpUK1?GK33*rG}nC
zvLC2#zmR4qirit0`Wo%PHf1GKP&YWjeUmAp{pDsxN6}I#+Z1Qf*na>A)8?G#VjQL!
zVXK@)6yHDEP*9<>V@Jq5{7TlU+Tm+>ER@n3u3X_dJsjCxv(g3_5|^M3Etiq_u18i*
zH;Q+^AAv;TT1OpgVSi_zMkNjLq^Zunrb2c2Y+DPjS#vR?#lllmT<R|!Oh0r%$lc@R
z4<XulE}={34QvXT<qzvNIYrXQd;8E4&KRU>&{?A&8!X`CF9-dx2BQGFoA$ffR(@9%
zM>*$o202wb6a*@wnsf#tDQjZ`PC(>%!opt&ji~Oh-L8y=;^q6YeKW%!xwUR7o;};)
zI<5H{SLq;9q%4z~zUv6X%i*%rZPqRWy7!y|_`zE#w&U{de@U4w_>V8xL#2=dR-bF1
zM9mjv_-mqp{F^DMN3Mk5J0Ui~psF4&73Bp_AI3h*`+WDS-vwPcbx|(U%nxcG@{i4+
z@ov{PI85-OEjv@icCegvb7r*5OHuXF2>+UmH}Rg<o9*iGe-p&%&S-RB%<7HTZGV;T
zHXl)b+W{l_O2}Tluwg`}FQ|XsQ`5)`x@>w)q~#0hH5WPt#1A%U&7wJilAlYct_4Xq
z(ol+of@s{n9J%PaEhQZyR)^9YKJy#Nkm6cf$W=AzA;)qD2Qf7E(9F@7hBW*zxKG(S
zX(Bt<mF4EF#tl!IrMS2Dti#Rn02Uc_8BGe9(bBS@Zc~jFi8>3|GvIZK$F+l4p=;^A
zRk(Ebpkx_O%!l$0CX7njEYBLbIYdGBd9azIX=&<}&Ybu~AL`WGO^t2ijkZklFahPr
zm+2=9W4y{Lk<{}rb;zP%hI0$Z-(g*Y6$&aJ90_~boB4-PIEsaJ({=0n_Dvj*@FJIL
z)gt)nQ2gk2d%zsWa{@;cqtxPyfafHbcEeEpAAg=8*G}bRdS*k>eE6P`;epHJuWdIR
zc>f>=nBM+<hc1BLc<}Ibl>Pad^slV-NT~~8)g!|*<&qTdsgNN9D<7EPBG*e3ud(&n
znnGdE4ukxWT1%$r`3%3@X%p1OVz73HDWNFXUnafEfH^J7K3g;iK(&zTX|f;Qp-BP3
zbr9tyafc>WAcorho3qUm{;Q=zRx*mum7aW;@vC7N4Ug=;iEb}+zo;8G%LcADJ%D&>
zwpzoBjfM;hEAu*Tr@nwmjJ^&U+-TYiJ++TXGU;oO|E8F&^NS}Jy92~_{h2*$P;T*5
z*Y5(BNz*W15Eb-AeKN@-ytCV9v=I|N;K?7KIHVMv*|yA4Zqko8w=>l=+LmQxBGZ3O
zAtm(!v|LMFHw!$+q-8PZ0<hzhOM~)$zWGCoNbzn%i_L|mJ^jZMflG40<$33H)SVAp
zWzWa*9q39=o!)tV`o{aM)X-rRA;Cy%+Q`ydW>r{6FUPwSYJY0G>e~918GNlL;CA-`
zOr)1tr__8V>n+BIr=c^6dEdeLW<zCN)gD|*KO@R6*~7s_>B=*)d~2>}k(l^WIg3w$
z=5^3t%0m9UqF(K64qNGSIGn}I;Hykdf9g<?Tk?n&?4B*j>qe>leGa3em_f1V`OBxT
zkQ+WCvuWr@`sgOit%OXqvP8@4QT`$39z8$z4ki)tz|tp&bJgUaae2+9=*H!sT}7>5
zr^vCx)AOX_z`<h~d2Anp53jz1J8$AhPrb{t%B|13bqW){FFYs8cWx;^)QB1T;<WNx
z`*v+NYyQusH|Cz@=;9ao^yF3Yd@scqG2BKWx-joE!XKk+`v1Cn|4$*6fxLVrmr<pz
zu$f=@OMLx3KVwqhaVK&J3frk9hLV01iRiqU4Er&gR^htfb0wA}fI{?FTDX=!mZsl~
z9Wn>(k;ha`TKQRw^$5!DXd#9Jt5(Y^iSiMa%qSu+f3ImCx7$-P%^{RhFndlb$@;Ey
zYrLa(o<5iC)9r%L00&hjer$BgRbM4*vQ5!q{yU06+WBR{A0TFHyn0qwC}s<$w{>B_
zZN1!;>W4-MPNO^1^Uges)Jvlka$lgG3Wyy0lH*c7ocfV{io3?CU6p3YN-RT$I;PU2
zG@=p(2>loa6Mi(^|0+53<DBUAS5tkyw$-${Lh~i*ZGR!+qI&W%XALP24qQ((+;0&X
zl}vCFBrY{Ni9Cl9X)f(m*@Po-Pq{UGWkIR?6LnpVrJ74cl$E|g%c?4Tx=ZM`vNLgQ
z7BsZ9#TBvK8npm}2tG<Y8s!>!zxuk=o(YW$GC0={ILxf*&d8tJ`)JHYhU5P#mRj<8
zvo0kw1*vFs&WgK#9WL+BJP|^ax_MT4ptqf;iu7SvsOyTwhKz6~tY>!w2C(Tg$ig6F
zn09r{@x(z%`e53^qs;O%S7LU+6f#uDrGq)nu;_@c7LHnCnJ2@;;U+rxxL8vrDR93R
zn@#TqUM6S!_%dT=4C1?@py}wwb2%Iq#b*m>K!=&oMV+z8jnA!vsE9u6mSO`8r7(#2
zV`dS4<mh4GJ8dV}xbT(!S}PStfi9_a1Nwkpi*iqe-c{Gs!ME|>_`;<vO6U6yvqf)E
zGbWj-25#NB$1jAjFKKf<5>$+0$UZRIFF;wiy+13dH{-Z3@k@vQ&+Z2uid;biHduP;
zhNnvR#YR<hY({szNiwi^jhm=5Vd>1At?bojkG7#f&_w7*C}RSNw&(+6*$z%7bRX9E
z)nD-uUVO#d@w%SX52boB7ma&zzxPxfnIbe+fG2UCicqK1hf~PM)OUK+-8I7-|2PUk
zYYkvE(Y)TCx6GbPsu5;#RlsA}PDX<sbAe_vAC<o2eX18yN+j1F?EIShqRW?8|1^<k
zM!x3zx?OD+5DG&zY484WG+<iVFxk%j@aV}#yPo$$QG!$ZPK30zd&cM2vTSjnYhAn_
z9U(+GkKBI%%iy!lcPvufJ(jCZZ;b#A{s+!&_AJ35On>P#)5xn1rSfFdgug0m%eUj%
z>@bJk<KkANLuc@VDF`ClGrTpwNPv&j*}g)%KUpKMcmqhI#uc>!JfxN3OwVp$(zN%s
z|L9pX__Wvg+5q)5*vkZvi?+F<+5skOmjl!(i)NX3(2>#xNxq{EeBnd+dF1!_6|p(N
zDKZ=nyB@+R{FMccLY{ElMf!widvz~aDVa3(Sv>ic?BymCHtBmaHE(^-@vv4Z`5u1~
zGxEDQ-`((R`~iA)hHlchj`Ia~IQzIT_qYE59-jEC!v6!f6TgHWzrzsx1BDC`l+W)t
z_nUvg*T3Jhv=1e`La$oe#=7>soFLa0cq;EWcQMWWykvESg293JzP>C)^?d&<vp`T*
zy|QofvX>@1?}Xr@lAD}zHEQk|37vgm8*&F&^eNc1A!zqATq$p1bVp54Mk4xV)8YN5
zi0nlMH?EzOhl8;f52_)nT?f!R`nRV)(O)!VT}L-baO|7?o~z+{Z1Ib2?1Y^10}A==
zvwDbqYO_k)uCUdL)6Iu4_xx1cgI@6cFwOc@Z*+-m(!6>BvK%DqyhB3npL3Bf_W@2G
z`M?GP1)ivR!q7C3$0B;J<s=(EIB5)N6wGZqTj{}_%Y;QEsNZEp#JhFPaEQt;-^2&q
z-4%4ax6n>_LjJdL;Gn%@-D}m9O(RYw$)@~Y$J(s9fAv>0cg{Qtfj$}F#TUNaY6*`I
zOwz==mD(LdyA4nM(|GgI9f>xcs!(;z#i2BybTiwSB8t=u2>Fm{rU<2Ie-d6iM8!E1
zI~|FQu?jOn#D18u8Rncb%`=u<hh#s}SOgD$SHE)p=i3Oun8{+0k(ig=z)qjXu}~JY
z?UGR$>&zbxS^E#5X1lFSwJFp}N?F>K`E|M>)nQgNy7C5FRwz=HiZcloxk3htdm`wa
zw9j33u{cBC-ucjAj28halFibgR4OlZweoV>km{<}`?8goYd?_YVMJDUaMfp4@TY-z
z=e%bjDo;U`KRD1i*->j-jLoxw@VCYWx0CL#f4Zy>MRYEFQ6J-~o?+`%Ich(5_?ZYl
zsZI++B^|pp>dJntv*2(@pFOS}mnrbfQ}{?G1cBgSowjVp!(%05h%APfRqfOuvC+@#
zjJl}Spt#z(^Z=eG6E0xa-Du9K9)NBMp(F;(H1A7wAK{1B3_QxRzC+j~+BFv5`et?Y
zDHagi5l)l}GNB73U>aj#PrfPlQcB_bwxBw(EGY#k1=$A}6X#4bg96cxDKFS&XTN}b
zCmbd#3$Xi6-i&$9SWruH4hPuKMCR1LQ}rPFQ|k)?$bU^0TR$-Zs)%pC=pNFU?JZ@T
zwd+BnD1y9&yCBLg)q@XqxNMWqxzb2OPl^<UW8@wW9JiZrA-xv%qK9ax9D?3#=;_jz
zH@t<sLb6}vp>IGZnRfX-ow12x<ZW?kyWjjlf$TZ$#`#DH-IpxYLIyZ0B!wJVo+cBx
zSv;`r8l`rY!<uu=rAd%QJke;8ch9<Du!rqu+WH=Y1X|3(&beg)$2ey~uGyb>FB*z@
z*yfKBxq4!@4qW&c?~yhI6Ytu6@@?<ZAebKyzB<eGRIBkwLx0K`aZ#Rx8{M-)EC@^z
z?Z*rk2U<qHmitS>o}MN0ZeaN{T)4zHZqD10V51Xl$R|&fhXUXBZicRhKrf^@mK8W?
z1d(pG^b2Ah(=i`vtyg8LGH>4XXQ_!m_A;@#u_?R1Ey;yf@(uE}c`K6<A8c#026Yb~
z9=S%UHYaK1EKrHnP(Djcx0b5p<*c+%<MFKwCjtz<QL5Fq>L{1m8MG@O?pgzDGuCwl
zrPs{FI}EAP?E0xXEZ>PIk(-TNJ%}JST}&5qws(v)W+V3M=x><Mf%o(~BZC2E1IttL
zqOFQ}*PL7-CIB}iCkA*LV2VX2BYjLW?AF5RaJ`UNSCU=(#o^A)Y}5j3y~xw2TE9cw
zyXXB&fn(hBymtYuxP4`KP1>-hfseI5(VEu4vqNmDLzAEs4a-da@R(E?J_^i8<kKjv
zyM&4hvgtpOD1`HOj%#k-%D)RaM(b}FkvVe6(wuf#2|fKAz@FKj*}$jBF`Uz2n)a7F
zQOQ-9af5GKPM<~~;w+gLj~Rw!ETZhLWGi|tLfSBE+gg~jDlJ1uc)*B+p?V3h{Z+DP
z21e<6e&3Lu4x7OD#^pwfv0Q%@B-^6)In!cQt?wFN55>>Dp`y{UOtWxfNd4N_q0&w&
z=iF>Y)wJYwIThMTseSo`tq_OJPfO9W1)rFU{%&AzqHv6c{;~bq`E=$9w-ODyz!^iW
zYB;VCkp^)lD<!SHKU1?EsZ2ZfcZ(XbZpBsj-rBIf9{O7DeqFe^;`3=Euh{6)inmRx
zO`MKy%oE-nHze2V`1Zui%*erMVe&luKS$GlO?PrW@L$FUbvr*lN*&V2rp$gGDM~fl
zMjI7h?q6OvJm$jgj=J4<ik_%Eh)1*PSBz)TYo+{xUJWa?Pg7#8!i?;g(2Vp+Q_W`4
zTUq%HKO~b&Qsc8tXH2o8(^5De8gU7*v837PD*?Wp^3A)_p}kG?0;hIw;~-<h{tU_{
zX$8<VDo5EGX%0mJbR5yT&yU7lI7W9Pj`4I@0A0Cjdzz82<K)bsV+~I-eavh!YC1Lr
z_o8|c%dXdb6IB5owC}#%VcSjaKpJNt?Jw)lo&dzUApii~>;uqYX|y>|JbWr6QUIb!
z)OMRR``{Af_MP*dS`x-J=6Bw^WUmu~`-O@N+v+m!^{|EYRx$G!8fnF075*?B#?`kq
z2%?hT%*qv#3H1T`OTo7?!{C;6G_D`Xn%PGC!C+8GL8e3%TBJw~PuP+|@|x^|FSGh7
z{dG5)lq~~TLmRWw&TzjgrJcNSMRnPU7`hwJb=bGmu0Me3$FXuXMrhPDPuB4^zzWi3
zl0;DrOMy!tIy&fUQ0*pZVK$*->v=&Evcb#_C(A@i@r=cl#@go{mc!W^{oEJ^RFWb+
z7VSriMv%{-DxjA+(Q|q}>OKdSUHY;my^aQ=kLd#zlGeDk>|ftKWR5T@%RpGrG!xC5
zLEpJTV%hV@C}S<ih!pbI4xAl2$HO12*}&#_=NO&eVjY_IA$);(cVeTv{y3mhd6{nI
zLg;P7lzb3Z7-<3FT9*DJ#mW9|hl)OWtb~A?e9UY#K}J{ycx1(*y)1OMW;|V*=WgNu
zAYMH@vagm&-45Efycw#9j%c0};8>`28i6NivYh;(0Y86^G?6vhX|6x%;t>oXcgpgL
zM<7P}1(11*ZJe>0jJ}xq14!?usxp)7y3P(ioIB(uv40rCe<S&a22G4>gKO~lfp3e&
zV0+CZxRJN)z%1ch5|7FL`}J=}YJJ*m3bTK+a_c_;2Zxj`7;d7)pR~EqCFew6q7|$N
z9oG*ZR=UGIMa>3>*Ii>9jRJet9C1w#vC#Wy>EgsPyPJx#hBpO~s_=RG6xI@ShiAy*
zLcVo^?b_nAZ0JU}RWFl|768&!3-{3V&=;F4pSdj>OZ3*r_oT<fSU|Zdy~PpfLVjE+
z7EYDYM2G%jU3^VS3kIAXE#vkw+wrvB-9|csXUgVg9;J+_B)lg0Cqz6C>xu`~bf$G$
zD!M>F@pC{7fy&mo2PPQgJMWf3jGTVERDV_V#<9nPxC3u<$_ljp)NGraiFu;>zDyKy
zM(x1tRDS<I07$*s*&AGdqJk5DS$&9or=pU4)hh70N9o`2*Y{;^C!o2NVuS22h{zw0
z0)s!i4{!*V=2LZ7sTKW0w`bj}d00Z?D}E$y9PX_+cpvuB7Z~E&5QgiklDL!<9GE|4
z6Ex<m?cV4&inhnm<D~LJrB)cQ)F(#V>uOtQpT;C<wHh41g4<RYOnX)lR-Bv4zeY7%
zgBV`(t<B4ZJF<j1w8e!TIJZ7qcpgWGYxuq+XAr5B?Rh({jNuI-s_2Y+>rro?bc?`7
z$(p&RF@F+>K53aAtl+}^3q=rFRlBIwN@N}(HC!P2#f-_f^!i(0yQ=U7p9efYex}iR
zP<UW3W}ucoWTNqxtI^c=b}Oe#T83%k5w{nleWu(@)$AY%SDxgy`4HV|!eSp+^b3|I
z8ScJFo|70j`4kBu`TkQbS?K;tO#kZ!|JxKXN>NBv0%re_@<elMeQsxT^of{zv7CLf
z?(lg|Br)zZd~3Cv&h6f*y>WX$P)2DW?WxA?_oNub@;sKPQgi_&MeLlww9eGcjVwjw
z3OI}MC`NdSYBnzF7vt3dHKC-e&G$tae+z5PWzEbAl+@|oIY%~}<T_c@l%2#Dm!e{q
z+7bm8HNh^7St`F@d}N#v09N9q16=g<a2IuJZ%*iJ?Nfcj(m2r1LC2N)k6MidLAOLQ
zZJ~+;miOz3v|4_&o<#Z+y(L`!$?5}aY&Nmf;aY=%iTeSxYNIGAH-1Rn6b$J5QG29A
zA8IU7c1_R_aJw^$ste*X2^i9sW--EvYu0j!UvvK)0uA7Qg?#H$BA2@9UXzhev;HGh
zO%wuB^$Hptd3f7b_6TjDHwkDeapq~m&)7)o7$a9Ju3YCz#aN^gCX$6c<Js)r;Cx3t
zTp9i7Afobc3)^JK%(8{J4lA5{Gn&Jsc)R-K4c3fPUK6VJ6QNk7Eo{-B7R|Tg6gJl2
zB540z7QVvk#I8m-@OQ7Ll>*kt<%r&is*~XcD^nY_1RcRdySU_*FsTa)ns(nwO!U}k
zDocgwnDlgmW9iu2rex~{y6FnkMc7c;Ja)FT@#rfZ*Q$|fr238wpQ+G{^{QK^dP#RT
zRfkBWzNK=FZ)|aAMNc1^@_}3r@^?jOP1U{7yrT|gPcv0T<fn$7Rw=7TtrNx_US>N}
zOt%td<CIHU8fbC7IzjYx+SSQF67{h}E(cGES9(vrV)=*m;b`o@O10!Gr~;_>y-<w9
zaABx8^pfolZuSrE*L)~Yed$VL)Lg-z!de!zmjj=eo5{Cq=zjo6Ry~j)HMG$Lox}Dn
z40@&g)*whw;+caTYrD+ZkZALWZLYB#YWoGqLPy}Q1G!)yje1|4sM}u2H%&yEUEPdD
zuYK*Nf>N!tHhfif3cBC5Nh=$P3-SVTO?`5eI`9=-8)i_Y%9xKFnE8$Q{x90aOCZu`
zNMv0<{^rD2r|NwgJ0vA%Rxqfmil>@`*u)978*s}dE&TR3=^WzKLMT=T3V1&Hegm%2
zE%Bb!)%`D*%!W)=a%knF&F38smpXLYO)15Nda~0#A!tI{{f1}!GOwA5P93)(4%P`p
z#{oJI8UT*v5x{7vo6E7Hwc!!FR{_%1&LNy^&hXmKVRosH(`He{XD>ao4Fc4$p0TN&
zS#lH}->M{I;Smu%>c9%y!*m_WCXi%QKq8*1&#}O)01E2eZWtqqq|0PmXDBob$)uSo
z4A7R{yK0vuMWtt?^wv?3#(Q!@VJHJ|V1EwMkO)~Uq&@zFnY0h?U>}X#QQLlz(|Yr#
zW^c(hSBX{N+L4_zU%;!|SaSgFthBK&Q_@z0)lR3SL-AJ~u4?guz)2t(p#4gaf6zT6
z(wFdC)Nsrs((A*06}!vED3bXnkY;gcFyI2`109P~U=A(Qty?eAtW483hLB<fvg7L1
zu`{xWgiT>+!1qQXXA?id{!zb;l7Bgu<Fv!=q$Dwy)W5dzb_|v>=_%ge#jKeSk!O+<
zv<S==THmM0d$6Bn4OZLc@y*yy`kE_y`x4R5-CSXvS(o9+@>uH7`N*Mk!<xVZgzM+w
z<Q=mNr4OYZ>5yH}xyLgUqsF|W^+DQDEoloJH&9*I)^Y%UQ8&%K1q}(lwZaAZGXvKt
z=0&em2xl+FJG~Mz%J6UfaDQp@q-n)kPiRP${J~P?2Yx8KlMv_tKd9-+dzLy|;bs)@
z|8vx&C5Mv$(*LsMO#f!+eMleBrzJ5;fxo=3LE7ZT+;#ZduHhlURI9LI>1BZ@`Kg|F
z$%YE!+m-;qDBINq44FL!&iw)90E8-<v3i+YeLZlDz<jh(*dqd-0E;qps%s14&`76Y
z)6vFvZ}K4t<BAgoq6%E)g22MqLM~!YwOIOAxqf-vbA#6^Cc9G7+&Y(-Ukh<G-k`^z
zY$c^2&o`Qe_H1E9Zt)y3b%<(SrCp1mY2L#sV`SX7;2y5rHSU2b-p<jp#*afjWE)~i
zu}tZ6WR=u_z<f0?&yb071}MRZ^n)lYPQ{rVtHUgsA-$)=KD?Ck{BEqxxIT$w`M0eW
z`0V>@v*v?r*ORpsO*Tal6Rzyg#;M7WFAfye9$r?%gjEC8a034{Q8Z#SPZ%}xe)j%$
z^F6!jD|&;$TcIG{PanV8l-K5YTL{=2`ndxIRN;%wWmzbFqu*8%nZTlb`+4^E=|IQ}
zXh03eZQi!28W>k2?cGVWdKbZ~5vhHM08SRrr?>LbMW*ZdWzYDKSSw4*L5ma`Ga6m}
z79;5t{UHT3&-5^a10pOEQ9o;HpH*yzh?#=e$tfQREX){%`Vtdy4g#=2#UuC`UcW$6
z>~m-U6ow@`(HJbW5EMq=ffD8klb;XH@Dr$SJ6eXm>*?I16TNx#Q;?z)a^^h_+p6lq
z>o^y;s@qs7)}nnL3^U@0^a`8wo4UD>K(F0CM^3rki-H5xwfwFS0TNO}06B5MjV{g}
z(W*)s=dy>iSo50SuB7fYq$N{9;6~*Lc(&PeP0@-LXVTml5e9!Hp!Qm}BmYh;dSFA-
zdGplEZrxJl{E36@nj@peq$7;_VRAEpKyzz)Q}nU}^NK_57|w9{T7mVY4CU^<o-CG1
zlHD4fDZO)&qJGY}^>f}CZHGAf+glG_!>4<#L@o}X!_T<8pS=^#d?jONmwCTl?eDc)
z1bLi>!1?ar-9WD{nW?D#SV41`svOUoExtZb!|-kGFe|gk&v*3If4aNGflPRUDC?gx
zprcA6j{4s;+~4V&oNo!U&i4w-g-@@=ts@Ajey=$zFKT-o&}(?CVdLcetR!Gb9LSs!
z(NpP#(AkbHBmkMy7TVo`b9#jy<Y#`^L#^EacHxJh-cF>`XexLOsw1IJHblk6hzOL<
zuY%t%ffqKT`Is66#L8)9lD-N4CFQ84jHyAY)3?$-%jlN=Dd#PTtd;e;726b__MxNn
z3bgS7u-P|EaITf(9XrU=jxBURYyV}P+>eEnf;%x!4*J61^XqjU$wUM&&iocmOiwLt
z`?F~tyyjU=W`+um>s}e>jD%>{_;3Am_YQGQHX{*xy55tEu;xDXmI0I|ky4uwmuj>I
zznd9@*XVsn0MB|P%XwL&K8(+E*iw(54E<rW+I1FZS(YdYMlgjl$D=CL@LpFt6wz61
zK?A-RNjhMR2WNL_*!wQw0F4T`Zzwo#%l=M;1uR)(t(C<K2jt0`*G_1{tk>60jrY&`
z7>5qE9qSTh2V^V{t!72jw(Lid!sj6Ulx=+XKy#f!%n$2#sKC*m6ldUqq5Hs!zs?<I
zWin2=a;O)L9!D6SNmvVk(_lA>Dd#EQQEf}fQlsd90I8+*#_zg*a+vdT=l3ZO-Jey?
zj?&KC-|xv~|JS1V-!GBmsxtXml05eQXAdM*Nqfs|{N)a$PjIYYSa5f<Nk)vd<b5;Y
z8%mRd(u%(3xlNc;9-WA;z^`d*wUMARfOqVqvT5nbEye<cF|A1{v1qg&)k4-s^6nx=
zJMTVKh93p%Y@s?f=F1+^*-skeY!IZ@xcVwTVcf_F^Vcmt@ngBN>z}4aYvt#YoxIa~
z*+w&_z3~ebU7YD`Qa5k3G>Cj7LTOXP_)~FA4<Hd7df~g$SEA57sI)K<VUi=2ls{ef
zNdN2ZAs^ScHm~=kkrMjx%-!{KtVIhcNd~3obrIXK%o`O~c`6wYc>qjW7}{aagvKtF
zf<W_lTpjPF|7N5ETsl^MhLT_X<`~2Ouu-yShB=87MuZhCKHRR=WNwIa_QE2M`mmY~
z>eTLjhAfIhx07zdEYZLXabu0omLkP=&xsD0jiMBXSID`}ElMdqRoUwZz(PU_<932Y
z7fA69Zn+yXZsd!<UP5mT;}WdJ*{IbUepr|P`YhH)(|hio9&KVBg-w5!-4MMO4&0A`
z@bg*tC8Db%FfjztdHz%*`UVt-lrlIy@eBOL<8pvlm|LF5#Dv6QiE7{mBaKtjSxBci
z=OAYulUe#7JSCrwC@pR+Q`~B09J>gKO@64!c4Y5U`qR6MYv-7#qWC!?!WS2{1DQ?Z
zNwABuyZTsK3kTR+APdWtG;MXLt|P6;QQg8iBtYtb2mtsN3QPlxYk4+n0U|tA6tap$
zD)B6`n|o$jd^I}L*48$<vHQ1i)tk%FSivE&Z=bV!DNv`A1Nrm0(d&Bj%%z#whlF%m
zmA^!&Ml5>v4kJ)AH}l8oo5MjE74wUdof;-mKc3C1B*^p(7esq^%JmeUbxX*}4mI8D
z_1(|pz%(sfG_oN2_|5MIWaKG$d&B|swAOs>F%xf|M(Kjq#BG=I^Et~bC6<mV>S~oY
zmvuOT4=(A0QOkzzcn2)K|Iy3~;4kz-Py}j*`DFX$&!7n1T|=(n<rFPu#DMh2?YM&4
z-6kFiqV@OAVP(U?8MkI)If%EJ#+;IA#)76QY8rZDRZSQ6^4oz9S|kUaSOI2APn9S_
z8t`mnrP)!`5R8MDK;&a)m3;_|u*G(Qz%tbk4uDr8Yz9JhWs>vDgFjE23oIH!NZXCS
zm=h!J-gIbiiYeB;Y>b{?djp)(^oFt?PhHRVwF(o#qQnwB3}LpfLbWX69N?>7@op+<
zOPv{tzr)ym%K0t65H$Hjy{6g%US7R{8pjTegy|d4{T@Co4c4&h@L}24l1Xo?=P$PA
z)<EI4x^Ko>$<7lPU+L<Zi1D>BuMe7Co6roKg9Ng9-L^z<udx$)1*O~en`?28eotuF
zhvxe8o+2fKc6%f%8(@V*5c4K&uAa4>9)r>bfg9(OPE<@62Kv6iR|apESH~)~%Ns5*
zd~xHvk#D!B%QPQ*LL;`x`l4eu>|7$|Gw7P8#CM1rb(i-#tl>7(+sAm5khN#lCEF{O
zekMxo)yIt-u<YB~3|vY~d^Qa1Y@_vaOy-V?`XMckPoz(ohLPMuedDdC0jyPcpgLe7
zo3oSsTWHN1jCp#+>Q*w>vdP$?_(qDumY}osNqt0m9+T!`pu}BKBkjCRHzNx(@{O20
z*eL%?T&DSNSCI0*=R4+qyKeup#@<N^0`8csnfjt`=Pc&&jhuP=3-r7xbnDmIam{l`
ztY^k!C8<rJJV4Aj=F14YfMRZKrd-KN8sN=m5i%vZlGT6ROv%`cTH^7~i<;*IDBCXB
zUfBi%ktI*hiI==~M@|+PPqzxC6?-j1FBf@%5ZB!r&%vuhFYcoHLW^vSql{8gqdn`4
z$o|~hX9>O?dOTz3ID-wWxsD7)k|;zVKI35{Am;-Oc{67aSdvLDoD~N+rhN`EXvjvl
zGVW`lLw?y^9UeydJ;?)Hki9{49RTUem(#CEF|H>dVsF%}mto^qEs(;~D3F;v9`v~W
zL<QD@CcS6ks&}wmXv?maOs9+L5O_Fb_)gxC8?+)WrsS>wQon|(ws@B~z&BCW8Z0dg
z6!bTm|4O3*M!LL<ld%la)d2;mH0jFR&9=R^5t}ty?m<}*^DGt)biUGw$Sa{h^xLOB
zsAB}U&Y5DV7;x)&3hK;Q2TV>vOzq93U4L!F6%{&2yDf8J`=#EXGK*z6_cYB=6sM?n
zm_wo7(ni4xmJsrGEGOpboX%LKm)J(@xwMXr4CCYsB(CQf@QcuVadcL%{UUWFxM29v
zvqUOEM2U`|Ocjq}_s=LhdLmg1fTgmO@Vgpj#!3K4ZD){}Y&2)mrYZ9|y=Y??RW2$i
z^H`$*u-nRZh#7{Ui1H%Iv#PWPnSxN$FOJ~k<%k}9&U8pr+26uY#+WF^1#f1C)v4NG
zh;O85;)3(1`Sr=*rKVnkv@4-CFMR8tGKvT3g1DTGYhb)?7yOrTh8A=p@<7?4n8~Xg
z4&rw{fc^t$L(g2*`1tq8{ytabxX`S?X{>2FYZp(k?k+u~fqpI5@b5d;LHn_NNK3qm
zqo@|M@4C;T=$;#R%*Xuc>v-4|_wq28tTDK*+%u6|N(!QT4$ASR)7iy%goMpgvl%9f
z`)?p$l$%)f@M5Np!reaoS|H<VzJB7<%VzRc2&JJ*T`g(Vf&&v8GSHuf867nZHmjqo
z=>9cr<4xvef{QBkTZW8CfSD0U1#Z+KFJo<5QJ8eKi9g=qUPm|KjZLJs8fCpg;8L-r
zcvyM)zIl&DPDSNejR1b*1+=wgczyk?Zo(PSoT=oA<8NjWj+?8Va^>a#WX4(_lvkx+
zbQ`b3x}u%$=^G;%;a@%7lXE?g_zU*@lLzT>vwHE=MGB@PW$ku-ghVs$44k^Ox|&kS
z_$oiK-T)dDMht1I=Fc;u1S!)-Lscx&3GpnB=-5WuP#S1f(VS<L?H@D`^K#vUz5%|@
zsIT6Zvht&S52fKdzi+2}Wn>`&f-j3}3p*rY^zpp1K=kRz?A0hV8NN`5I3p_Yw)p+j
zMVALbo8)(g?_53bI++V9OhY|yHD2MSUCYar-(7<4LTZ#Y@5%2yrXVy&N*J;UI*{Ud
za<I1XX25Kb%MSR?aS~V9Qc;jWGCS3g-D??J!Dj`ZefX>HgIB}y2|%WTEnk6Oj_Nq7
zLeVOKqYsBj;g?sFl3JwrpAP=EDT{(Ktp(em;F`<!O*dciwEE|3{lBIaCm@f!_#mgk
z0bXZ~t8W$|9dbi*Wc0#Ghj)1or0i)*%Av5R0P2CbrMt~X#W_wkqDbnn#$V28#sWn;
ztaYEr?vwTyz;+=RaR2%^TC<!KLobM6eNC=Do`FEll{Vyy=q-KcF0U_-Wq9*4BJK`;
z5qc0+>(9WEwFGd~<OBCJ2+_Ep?kLF-;#VOO36B}@q7Yk3-38P<y&mXFd}6I!Ol=mO
zM2#nNSgc$ZU?B+Gljm6)$+3jz8G(nTeS`TLrg__0yV_lStKOjawH>g`g1BlUGQu-N
znblu0SwRJe+DHQ5rskq8{trM-!=BRx1_`1SXE<xb1hN|*5S^^@^Ex4hI3wh59mdgi
zXFbq;hzSD)_LZUq)4L{Yt6L~3i-nEJ7n|N3HT~wAaBdysiK<)-h+LssP;M*7R^ERD
z{(g1sV>Ah>+DG;)W#^hg`-AIH`T|l!fCzX2>25@{850M<a;&3j!AnyRQ-;=r)@@8n
zm(u7J78yUI<Ke*F@Xm-!osH)JI<vEl#tdoA7QcG2{|+d=5l$8wDg(Rgd5g-psbVwy
z7uTZX>G$aAl(MX#pz2od$A{#v*-=zB^}dR{`gT=Th>KjBx84Kov$-!YlQR6P_WnRY
z9AE_8T}dMVB|}*ObX}Xq<k~kmz?xV3i(FTtILA!-QRqk>nE6qBdLxhxU4ES{|80sI
z>LsoE{uaD^<~M%6*Ur<W43-&PDsc8O6S+^2*Zpbw@vMpWIMOcC=vu05Bu%lAS$Ri8
zwH3`$N<|;f2w6$JfX{ttOzOqyx2sb%F^7c={Vi4#DoM0d$ofiGu-ZEeYug66)xC&?
z=NLu~#IJJI1sD{FMG8tQ=w%kZlZ>BUC@5WiCsg-*(@qX$q_x2&_cnf-><o$z2+V!o
zh4mdbdlG7Bj@mP52E+`I>;ph4KC3FA%Ij}Cs<&@CGW#F|Io~I`l-7}K?<w{4-$YGS
znTHwjf2J(2%!LwRgZmxAuCpA1&9hu!z@#od7fqX*7h**j`wXS1Ln|FIEj&zP^ftB%
zL#|#{Ut;>f)8)etsih@R!`<%QLH<!LrSlP@FhN7OevdAJH%Ck~1TfzcB$lSCM`JRc
zfqDD!yNBhoFO<kEcGu3OVGzAd==%QP*!M-lYQg8cBRc+$%_3h*;azxS->0+n2Zv9S
zdiMGH53Xf9!+#@1U|-PN{`yY09=xTkwJq<U9V~~JlNM4UDjoTo8KGq3m4}7XMOTNg
zTntx;g(RAm5>2AGZ5Qhhe~R|*lx44ck@>`gc{-BtK>pD|535Bv;K=N|$Z>$P>9E#;
zPAS1Pn%6wU{Zk8+Oiy~u^nQzbgsJ}*X+p^a)5*U!QX0^G&&x$Y#q&<B)UktQ)*?>;
zsa}bqPH;4u)#HSu%sj)_!@=HVXb024HE#F=$=SaJ4U$cIqj4XCP3fW0E`8HIhdli&
z9P=fVFLn=ojqp2r^wwW<f7f@ke7sQ`F|#&qlxO^z^VH2ZfJ{@!Bd-!-|L55BU#s*#
zR~QBJzaMFDtv4P_CUw?!*R~`_PA{n2IBosq;`%G9ma)ZFuCr%GC#Zhj6q-yGPtKM)
z>&8x>)$&4QJt?IXNj4PaXB7E+`g}}Y<#}bM{To^I>Nh>}&anwmG+eUHm852#`HC=7
zCj6~WO*Uz{>V>0Vx?8$KN-!-CjD0C&4edjDBM<7LI$<BdMVpWoVx9I4jf)iyL0f)A
z&nqx*)US0KTD*_-%vB^x!Uru1*p(e1G&F6NL6R*x_jd>~*M}QN)05S)<?a1`*C9@T
zKyq&2F=8yeo!&LFG906*ypS@<4Dd4wt<L`etSU&&`rUzC+t5hm8!WSqDp%5J6hl6<
zRiT4-2qZ7GfejOK8b!&hmJYGZL`jIwU5jO+?fn7Y1Yya~{hWG@PTMhtqgnCrf{@c(
zJJPGf?L7zIFP1r|$saiLFnf_+ye>xZZMhW1GDd<-Q!paxTpc}!n`-S<g3NEgYh=q#
z$q^%qB(T!hksJA4xSa}hP<5rZ*ghh)jqix3ZV@s1-|deG1`J3>TalP*ajx{9H&0R9
zY!(;1w_F!#fsXkZMRJtC^t2szWgv7!^c!l_@leKW^qlt|d~hg>E!R?JN}WK@1B+w8
zDk>mUHgS~zCTMx74Mo~qOFo9`Nm;!?IwnY}`(5@@CxzH{qkPFr4x+rft!S<(=mC~l
z9UsP;_w@Ti9ZN@ls1>cB-4!lOY;H@P-W5@njrtAYVVC!1nN5Dweuy^kHAbi3eFx2*
z-EV8ChC&zMWngFQki5HMmwMxOqxQZyt8tY@jl-as88ezeiV=~59(Oah3C-u6d$}UM
zmy<P>_hxZMiCF$56HghJ9Ho6FiYa85$6L?bCAOR6_CZIEvf4%mBiAQu4{hzBnLeMi
z%AV$XuZ^T+L@kiY6HMOe!1S1{6o>t>5&^;3Y@)LmZq+FazR;wysnt;MVn5I{`?kDp
zW6^qb4{V2byCwBrSak<55F^wPq2@9%qz^}jaqtvJ+kMaH9TM(7>gL5^iXGEXv_@i%
zd#Fd2xhS1t1jsc+SI;xehQeH@Q3SUVo4@lI_ps5FA;%64v_KV3h^EE5o~4<i_sI!P
ze=Vap!#iD*<2?bs&lE+rvS-@~Rb+{EKf1u#(BqUU^ER?UI(^a*de!<Abc6?5xC@<4
z7keA;xxf{KJm^}{3wP*6CMieS$%&F8*;%R&TtpRr4y@6X+1MgR%qTy8Y2`Xwnld_@
z*%U7YW=GKLg;`=Ky(G$CLMcqi|1e2KBYu@WrWFK7xoNat()c&i)c8m`R39Q3_<BI-
zIy&%bI4tZsm}R(AG@urIXpx0m$|2kWF~g&HwZ?xks9oTTe!nFvyPw-#6E06`T2|Tp
zSrug8kt!Px{_*Mv`XlZAV9`&ck`l(D-3(;%S{~%&qL_&sdt@}_seD2_$E|sW!4jNA
z75O$_>ZSK-{3ei_FA>VJgBw~v%kImi-$S-tmp?C>R$Pir<}wp0F^~n{!RlBt7?e?&
z90d6<rHWjY10epJ68RUM17u$KYaBB1trTqdGB2)sIA0$mzjpioHVP3F&FJ$J5V&-N
zJY(WmNDCRoK$*OuTNG2^B8GnOT`IH=q!)U*gSPM}muBshW_~tXGWx}mJ&KD<wnd73
zmxJ7O3n6*{XT}a5tXD)8S`|)o0CLQ}IlUuiK7T089BH>VTr1c@jgs_-!%#Z?;n*0x
zv$-36nrSx6*iohSCK^dKFCzr|h`|R%>}`1%8C912#l{k@%%6d1ef-ISnyS{coCcuU
zp&o+LqcSZgF_}@k!O|Rf<!{$|BsMPl<&aO$9K#?n1p90-$Uf*iExp>}`TII!VD(V-
zIK(Owm6NT}r*ikFC6#m!E`>JE=y1rKeUp><THxmUei%l&-6l&#9gTk{;To$0tWlLH
zvO>gA5pUX14Qa$M&6!rxFSpogezc=(Y^87zFVvE*#mnMF&$k<LyBBZ;cbem6)yNnj
zGVMU&K%cRfd$Cq7D2)6Gkfq%weiZcBp0{Y)uqeWthm7$fQ3U#>+Fs9>FPcv8x;kW6
z1poB3j*4{xAjvd4+90k6V`_Mxt_MqTO8)lVqfB?PiWZ99N534fIs^QC9#A#)x{#Cs
zNl|cs*FGYqnp?8gM9%eu3^htj<i1A+cD=sqgWtX`EgKOqRdEMS>TEUnYO;;B)?Mnj
zY`;liCJ%(?H7P!S$o7#f)|N0qMl>gwb#WiP5$1c4B*_b97Hgzdau&Ut@mNO5#g=)j
z|ED#VX(Gx$;LYfl1FMX+@{rl*MY3YGHZeUCo$V#Fj^kn3t1GQ|A#Yc4-*A;<dkT_J
zzSe?x0{(IUz1>;od{^(q4Ti3~-@6Zf!F=nkXXDL&^grU`PP6u;IvqY*W2~QGhmQ12
z?Z9-!cEx!GwxG*j_g~xU^U1kpe}9~e17E6&#XJUSFtyZ=Z0lsy<Zn02yG$g9q817U
zO~%D&-|T~|Xzq(DP(yPRNwAaxRW&-=rsXvj+J{{Ee(%NY@{5t4JYnzKRs?caXM4Dl
zzW2)Q<@UA5-pOxDdcQ3Z97&Eay4K?F>XaEO)}@VHBnq2kxZsBHn#o+_Ypzq^k$0oK
z8>PBvzNh8&kKS4$S5M}ZK;m0ma=-!3?n)|h7l|3df!0_4;<HKa-8uA6Mfnx>6yk4D
z!>$N)+@D<{#}dJ&-5D~T!_1q2`*lE1=YkvJA(MQKdfOrtT7WlXBTf93EMviR6~Ba*
zV?ks=KP*Fu*wx)|g;&fwyp9cQ$yYYPikz8KYVFxOS;+jv4lfYn{b+aSdm`^x>f3_}
z6r8MH`KXkJez(B?xSy($>sb{4kaHdY&lLi75hPOEr;w2WG$P;_TU{^(|0^0;0g}5q
zpnj^S2O>a-Sy!tU!&j6logP@9wrH5_RMi_O<63iB`Gty_lDKJTK|-J85`0&MArDk`
zlRX@1y~JR??~*R?r=@(1-q#Mwux0KKfSK*rz1xk#3?7HrcPU46WaMA_#y4wV5O;sb
zn1cMucI_KK!IS-KKPl{Kqn&+B%kL{^P0@D}Lf>~cy>Kb!oq0vnK60rZe?4B)`ZIZY
zvC2A3nh!B;pHKRKM%n*F+{pe_@(oY=-!E5Z{{845-gxkmMv|*TZt%N8Q!>-z243sQ
zweSt>>p}12V=XnnT+Djf_ofKH8=@J^BI-n7OpY2S)5RMMCEKvnQH-()fc1;*8+CQe
zaN@4ArKkiZl0{le4V$XPh{}1JMQTY4%C8`w+LWXP%G*_`&g)maYn_@N{~Ks7B?~oA
zZoO?>Ynt$yZ2|WfmOyybrnC@gX)RN553d8AcWi&wF7PI_qGuS|@E%@l(m7V2Um!+H
zJFLf&(${}e5`{Q8Z!q#gTvy*HqG!>mWA~#l`H66>hyozCE@~wgV%SSQ{G6sau(Fwt
zpKXgq%O#JjG#V)^<FZ3!sj{Kcjn4HFcIf7!BXR|JAL#%Q2l_m+4}ImwOM$eQy;?b!
z9|LBZpBq1v4T0T#iQ-cCKzB3~BJOtly?8r|my|_ZlUj;Rmq4&f)e!(UZPBzMNDm~n
z!5376Z2L4S$F%$$?hKkPcK8r!mP3EC+@NQ6xYD9i&msfkJHs9Nq`)F%QE<{G`Nfd@
z9SOBj1eN_>1qKKqLFNUjAjZS(yN|{Wa;=II;{AU6j|Uy@DQ*{r{mtjSm$&6m&C_Z7
zu<IMefy+g~s0$#>PThG4r?-oh%|o{dp%Whx<wMKBDO<CPl7WtoWjSW?_Sf1tE@i)S
zp3SLm*&Z1lzLUmQZsG8jFQxiu^h*Mo?8$T!A*=bFl}hf9r?GIJR)>}NJ-ZR+`8R`z
z*nMP4D~Z#!-Eb+~*rQm?^Kqi{-~N;geVBaPZa7E=f91zR4ZMfL=jnCrFThgdhY+(&
zFO@Im`RaPQ7xjqm-HN5>guJE<Uo_+pi*QpFvb22Yr%Fd9e2(*};R^+Jj|ni=g@v0+
z^Ym-&ZzPP53v>de7TN<}hdXq0LD@BATpEX8&Y$NJb^iR~+lVepU1#7`T*IUu#KRy(
zmP6h7tRQv#rD?_rZ)aVog4Qr7|0Fv`BQ_qMCH`{Rfv%;2`Ldx&D^$w;S)t@ay2&p-
zpY~Wft@<b?lHT*S$aT@{iD6wKkIV0-&Cf<H6$f}*U;CmfxbN2Uy8+M3*4+QaHaD9X
zkm7llb9d`x={i|z*8Ch_pCOjU29=HOd9GnL;_5b1@(5*^xOkick`GGdu{ww&TCz>B
zzW35OH0f*|)%Fb8gNd}*ALZ(MgW*m(M|Lhv%qGAMRF(}`UJ9!K9x$O_u5Uka7l86s
z0KWvFQMQTLRKJlv%hk-5#;ZzOLLH0dUK#bInf91w$`|}5++0+B9PFa6-DmY0mX)4r
zIE#wJj69JRuN*8>H6GmsCFJlFu@FZabWG}%#73ej{Cw8h0rj;HwGNDk!8=rRT$a0F
z3u7+Z3h#kMNW2_BccGY^?BM8Srg~H4(ms42EUp;{7zh;Mq(@-H7<IsC-G!_nIaK=A
z5Ty@7T%tiU%+k-Cue6JE?Apz!IZAFDl%<2%%b18Mh{>d}^<9NW0J6_Gyfd?C-=|wE
zdEx;rVeqo*x0HbrAguqqL5uQvt)jvU*Fk1Op8<0HwE7N1M7H)HM>{1C@<b3K9Gb0V
z*p?#!5#2%^X<aHA1GU$sApFd4kQ*_D>byrzw?%&jV(uLi!e`eeCcNp)dc6LSynfk7
z#0Qyvxt*vrfk8;>LL-`LB*e*}zW;;ni~BDK=3laxEOI65EvU(ZDMH&0KD%b^MSHjz
z;m?mDeAzR6nli+qpSx0iO?!TlXv3nU<hMxx4$)oTG%XLCv=m{89RUY3sAEKwOb>Lm
z^JP7T=;A+c#@w1j&<UMU?AqINbLJHdcbQfF@0r)6u@G1jP{iS^C^wmF8hEVR0FMtD
z@*vFw*$g99_YIX&nj=eXaON=xre-5)KFi9}HyEYT2@0aaLoH^gvX|&1RR~i5<eRQI
zuLk-VmtSn%N?fRRUcZj!bR7y1@)@zO&<>az+lc?7L;s|`XiB~EN`!H_7V=%FzjTma
zr%(-6!vWW8Z_=Inob%H~gQ&e30p{glnwL6Y_Xg*aG8<77VMtHEJftvA7bQ!5Pfbv|
z!cPlNW>*r<a0-nfX5u+sla{opqI$yu-Ey!ddfiMo(@u}0Iv6S(%=t-&g(#v{m9crn
z-s(rCA;M<^KPGPp)$?W*cgW5NapbfdN@ut^|HF$a*FKP(DgzUq9Qh5b^nv&Vy8jnT
z=NZU$`@Vg96}4g&wMkMcBvz|cGe(TqLGRkcsI7F|c1cmON{ku_38MDas@AM6W>K}b
z4!Yal^X>hAUd7A2kn=jnaeR(vDUI6bxhfzq(h+!#h@~tIz4G-fCeJ-4Ot)XZ2yd5>
zV)-5=iu>ULraN78U|4=vmQQ~!d$mHT9gmUbQesj1`m*7xVZbtv*LBQD<@RY_oM=!`
zZn!d!t>DnOwy16_jwxxWJpYri@UWy6*8(YotSD<dF-<8T<!ogz2MRlfv0MD)j0PDi
z$QLtdXauIbJa>TV;;R@vOz#pG26J5=T|Mhy&MFn7cbO)?)qXEw=f192(Snqb$mOhu
zfsd?)1JVr-nXjblxLkaKNRO@xWJ(&M?oE{CV*IEayao-=jfJb1O2`L4b6hwhsCiY&
z1I|yi*cIe!#@H`7i@DcHdp0!tvi#vvbuJz^@J|wu3x<bxrPcLCA34u)W!cUn?7Smv
zzL|VEl#LPX@_zr-;8#49#n-C6RlRBR{y=qSqW<sOAMhA?eWP%hfP#%c*QZx10=|IE
zkxwIaMJhp>_d@#FD(qkWdSiYDP52E}!-y0JhieJ!{W5st!Lj}i<Q6x~V)ent6Ib8Y
zS&DgghHQqan~Xd(gJ2FK{<m_%JJ_YuitLl+YDAEMNzZeiv7NZsZTN}9U8h2?@{S>S
z9AMMMse(LM&$#={(jBRZke(!FLwRIi)j&I(UZo1>p`XN(KL8^ae4}{DI%HXz)-Ae+
zAz47Z)t?^c%osFjuj>pgoYsYxIw5t!Tx}z@uvC<@MmocGUC&NIcv2`2=f#*-Z{BUc
z)WI9Lu2*!H)<~khjHkvl1$D82qb;H`ap?WO`DHrMpqTS##tPdk`O=|<vHea>*Y_me
za-Da_Ywcna-yImwHF%W|7P$rAb$|9sTay)$gKw6bECN-9;Idw?6OaOI>G`&vpV16_
z;G~bbL#sDu&pp38DX%w1aATU0EfMZBay-+4h#+?adTS(ywyWdIt<Hmx<Ay(>kCi`N
zJ}%*k?7Zvgfl$K$%3NAeJj7XD&Xi!1H9&a_Wjv!yHPTnJpj1|b-dh2W(OcWEasfBt
ztB-BHQy*eHXDb$R+vn)0_L**RwUz$#)1kQ+9lurao`TLFe{@=+-SkvGug4s26O;KB
z7WlG#`nIlE<A>|KxMcpkUh$7NlaxlivBuLS{<XvD<@tYy)xSR@X28s#J;MV)N&zE7
za^hd9t|^_qt@58es$UQ|;MVHpk%0zTk?e%C3f_;xuQNG0g4a71s|6yy5*12lI3dkb
z;#C}d`p$uI5K3BKOIU||9X|maBj;D+*$<|PWoiQ|cSyo-btMwamP#AIU|2e~EK(DZ
zmYm`2;IiFNKK{|MzG2q)<9m9Y5SxEE*BkB<r1OQx{T}9xF>?ql6*q&&7^Smx>8nF0
z{yI%@kL3?PTW4WYJnSA5Sg~zVlHSQC7z?b-Q9p*4SI2Ki%~r~V2P_n)%YB5qBP<YW
zQ80+nAESe(O?%*UF-^zVx3}g(m)j3@VnXcQ#+eP$rJckT_vRYC%M<LSZ)5Zga{_3w
zMYgI_BIElKsA`K4J&|IiHO^=<<`7^cM>4_9M`N~Y2Wn{rwPx0Di^A;JZ8LJ;7^OcM
z4KukHBK;U^6uMjs?u$3Zsp%%br+pa|VyN)^%ES@$5SAqfdx#oxZAb7hwP1KdS7Akj
z)^1VDu*R%rv$@$aA?zvno&q;?U5PuVrQtQOKKqN;+0*55Q6EY@tuy@-Ix$bFyXv6t
zsKhD>>vQ>l@?05m?fVlklvnxKU6p$TCoA{8#-Q+r*fm>KWK2o!?I3`&G;+6?=bM?z
z;5*Qj5*S*QTMZSRYYb2HUfK7T&=Vs}sEujKmT-mZu|u2K$B+BAWVRZtMM@$UIZ9B@
z$Hw@kD&K%_{uyCQo&K&mGFA++%%QZWYTA3Ra)maMY=a?IEsqegc2!4ZY<kVp%?uvu
zG&N^p0^ab@I`w`fNFF(OeY|7EbYg6iVt*6nDzH^HW?u{-KUw0<RZi@Kn7ru4xdLuk
z6=>Am#bwCK)nn)SE;;8VZ)ojJcWTPQMyQK59OT{dF`-*;2smiMg_=lKLf<?hXrQaw
zrT1C^C_DJ6jKRAk#ROFuy)1#C!2}1#<F*^WOrxo6`i$Y%a{Uv|A(;7hCt8iuO4$1C
zXy@}u=WRW3!QWEz1~sZFBPu6mfeHn@HZ%vcinu^JXITzl+;?T}YT+yW{)Aj5=S_tF
zo=bbK`?Z1^bW4kSkugT}UCWhBiRiij)5XHcn1#-Y42T^|dj`=WSG7Sv;Ce_YCoT2^
z_<e|;@V=E_x>S^nnMV&aE!jl(#BMd<lxQ_I21uhXknFQdU?{d4?R&``9S=|5W}M9+
zw`(zNdfymH*nVI8ccXxKr9c)H6t5!uFAv2owAL6~LJC(S$%)lr?}ncbdi6UClB)Iv
zRBTyRGu=A%DQq=gdTPjuejl=u<#G8NI(w3cT-i^n_!{5nl{M6rU4>0Gle&N@%SqDs
zWVA4Vy?g%#@us1^;Q%Jja2PI<N`cqU$f*pWq*DHY_x(9#OKbHbPRdKBWfLCsR={6_
znK)_OuwjAM)}5JjAxlkDkIb?OQz9n&&fcC22Sv{i%f6qEZ38|Sf&&zroW{??ot}Fl
z2rNQksKJ;l#@t_i)K_H@FKB;27L1QFb*f2|DTiZ>G#lYxv*wA8`gjw?^_HLV_l+(?
ztu%)KvUf6m+1!gul{d3@if^@#JYeWR;}7T5aRBwGX^DL)$jvxb`)Y?()aAmTJ=4qE
z4!ivNfwg7bE1-dU4fJqaKq+gc^TyF}3I8=~{T0nB>I(eV!y$|ggZ(jgQxDy40ueqh
z_-vkaq(#ZHE)~;&OK9u{PJ3FI4!Jos439O(BQ0B`V63KU6NpQC8C3-RqO?t%emkGy
z5AV2=BphaR(r-Pp#Y|LmyfoC%rb8<E#+ZjZ+V~2&_tX7ljtZ`<GUm=7_8J|g@~b5a
zw_+o=NuBf$b!s%)N0eft$%PDfJrO%>cbnsGn669pXEC3-nkSch`7hwmL=N9PgzasR
zyS=QHH0$-tEM*FWW-}R&&*4@5G&?q@A91K70^Obd>lZHrUo#A8Or=xPB#M=UA<x;r
z0^R-==F~GL_&P;%yVkIGz1;A1eHD}t2^GX9kW7=~%EN=sG;bN|E1gUGY`5e3ih&?f
zpi?1sG^YP0@+OzvOMZP`&;$DBa0<TJ+7lTPhknI>C5jZ<svhX}X`|of#d^povEWfm
zZbFzPOa6J5C-B#Bf=8PGJ;t~?%*;Zz)ix5r4`XOC;8puMQxS4tcgws(AVCJ?<&5Aw
zuKYc{MWHI*bs*`#*M$AU8yJJ08w8#ndZ4*3JuT%Bt;s(U>2kQ({wADW;_Ow*xAd}x
z$kRW2Hmy)loS(^nf3e_TlGdhe>!gSrTj?iXP)633e<9086;u=P7hz7#;Yz&f>K`EL
zE}u}65ZCc~BS*Z;@BUxetqV<QXH>A>k0y@Brbd~J%j;qDbA^cM0irUrOy*qfn$t6H
z4lR~0B_U(S95rQ30~P8UsHYL7?Rd*^o@BusQe4rmV@PLOW6_T`sE3+MCnx4lgNe3+
zut}AdpvUCBc*9E-1acRwCBi#YoAgN8LZz>1olE7@IQ@}pa8&9`gpYf6SKn)u-59j&
zUOQ~P6`AF_zvzYN2<Sf&Xc%T`sA*^sAF~a06FDWT3iCiq21a|_f=5c6B_86kwy&#g
zgVdWOc8Br<vzcyvsp?hb_SZu?9)(u*m}gzQL@D*0^CYIds=H3Lr^&%wwZE(o$oGuG
zA`PLN4oN113b+amcT^9wQvQ<ccTDYG*1sfTzypUXmKzvEIO+^Mq@Jb21-DO0W@$O4
zJgR#&49bnTBV@AVO>-x?I21*d+kvF1jI!q}?TC!Mq*8amet4@k!BC1SDZbzRMB+5r
zLD^+yHdw2ab;fa$$>KglzB17vCy2>jiFWZ?2yQi-)>qwmY44{#C<t24`)nZ?77sNM
zE>8dpGq#ypsz|Hopk!^J-8a3F2e`n*E?Jlc$n#8rR@+)WA-s0hua{!RdD_pu!d0W{
z{qJhTS_mfc;7>+zqneFW>EH;qkh8~oM2_*R9b*KDv%I)>D_x%O9j&t$F1k#+S!7eA
z7QzLeWE82snsrBjEoQ(HlOx&BAP~1;9k-gFLc9ejn&#tH)bnDuE_#6Ao}`{&3Xq*C
z0D0pbXv&v=2Yd1T`on4|9gDmN9xOFW!;fn1!FgYlR?Zh$^X{_Vu$61LTl-Iv&Er4U
z3~hGJW1jrP`G3Hs7<`lfAFDzg4?Vxi+)OT)*_H-=g;;Jj^2}D|V+?ZA=GA@d71F4D
zYh0n$f!$l#-G5Uwww%83vq$+Pa~j6LThs~>9=!rT4prT)%>lX|uyy-;RRAW@|KAk+
zok;1yzVuZjbl0xGsdH++g9p_ec|AJY?w>G4s$)f+Ch2BL%eZfXKvmk><`1+7x3PS@
zQOqfY9+uIp5fFmZ<Nba{yjlWg6k5(eDLgo(3)b1YBkn5-Y~h9`_0+7cqWMDPS*I(J
z<YwHAQxF;x%{pEEyJ(>Fd+j^SRt^HO(^2?9kF&)8n!LA`0!|<%Yq925<nz1kPh|@W
zdi>Q}MC_6=)#E?!kQIrao&fuMp#xoxsiN@8{MJb*ot2{Jpt!H>BjEsCSjd5`H2*hR
z;2ec-DF5CH*7x~`YM#aCUok?fv0~aYKhgB8=AuSFteA{(p+Bb?W?$S`s!izk9^;kP
zc#!L^hPRfUQdPe~!x_3`AD#)k?`ITyW+MZHBxtMK&aiLe^+253wx}TC&-!4gy&z`u
zdR*ah^Py?OW7Fxzg!I=o{NAxv%$``Ez}_ELDPYDu3ohKPt2)Xjrne9!W%yq4=u-14
zc&{D!qCz%CR$toZE17t&1d1$oV(dY9ZZ|gwDd~f2@JN$>ePdTGU#wFJ)Lx>zYxgkB
z{ekq-DT!(U0Iq=t2uQ_Du1z}K>v@70S}OCWF7);XuRoCS4RGyndG^9|0!B~R>Il%Q
z2xe$)_P^Bn1#@9C1XGCOS(zar>NP<uujp8s&Y-(TGIEb~l59Oj56wxQp-s!EyHcyj
zn|;I#c&h5`bZ#zwXk}$&UL$|VA{Vo&(2-qYXg`$QJlR|(nhYJTD;IF2(YqkzqQy@u
zJk3RmxID4QN|NcVVZ&<_kt7wI@MIsjg}JPfU}VUsia=?3>y32Jvj-<ne<)+(2|J_i
z_DoIO)OK-=oKb8G+2GWw<<6aD4G+x&gxn5<<(X}V*wU+W5%$T=#RW~bM-}#*rbp&J
zXBvaz)$tFe4TITk6Q0#3IhQ0?8XBf`w>V{oM%2Fc5GcIR!k!>oqI1x3AENk{m1~a7
z{wihm@x2i9IqBt(BNZn~w%4TPo7Ft+v!<S0eI`wI9*yFQmYM6cPLRz!37D2re4n$H
z;qR%x9PltifwoQHLQ9@UC^e69#>H(cr9x}CRhv4d9bJh-S~ag&x;vB|Jf;%s>QLeq
z6$m*#W`Xx{NDni5>*j>H<lgjc#va|Tl2MdPaaG=XdhyY4ddQm1%Tgx!^2kyV3BOuj
z*~cQ|*y3dowC;9G<>3&a6EL}%=x9;z-hKGC={oqxRaL9p$!>^zTqdWs3x%I4Q8~a+
z<2T`AGj%ds@;AOF-=U^0HhK3Qn0emr&b1xj;iD6?H#7Dyu7)f=D-PXwhUjW?r{6Ou
zY2}l?82%eV5hJ7uJXHM7rT40o>?NK1j=s7s?^pe`=J4~)=`HXdL%W<dK%$JgcM(qv
zkU43{ieo|)g3i6@fvI+7-2nz1Bl>Z=N)z<<oefr5PJ6?#Z<jbc*W0RCWlCzN5vK)-
zU-CUZGx*;|{oB-;71F+Dwc$NkejKrO(#lG+W0t#Y`(s`WV2ogaLoWf#d7f&8k5P8}
z1<73BFO=6o@zT&p6*arbjEeVKnEc;Mb5}&(gyfU|c}OX+zA)8b-Nu)IexzS`!f2v6
z>fIWrs-m7t%m3om_MOn*pf7$nau@a|V$T{zZdI2#vLO%EJx})Zp@Q`O%ORsH2U5!Z
z4@)me{9FG18(HZukkB1sqSGbo-k`Pd{r;Toz~?y!$3Ef<v1^}Ki%YKj*sh1)Ah`Gd
zo>3qxLuZc;7tC!UP_UiN!kU1yBq8&Aetb06@r%tD;QcnKHo03{P>L4KQw6uDR})`t
zfzoiPan<4dR&AOGCILdU`KHZ}r~*?hnX+0^hXDX*G~LM4sx9z!H(s8pXEoL5O-7P8
z#U|yYpTV`$Zbg(9g2w58%sBqQl_O2Q_fYDXF|vtgl(pd?NFr?9WrL+iQ(PNOue<$p
zT#^1t8cu&TR(imarAYguNn9xJplR@RbuN7+*!y~{FMxjIQ<h7@<~oDM$E+b5Aj$GA
zDdL~-t$S9gUfElu6=cKiq8M#R%_QgHns*7SK(b?NV#6>M8EZ&q#hF>(?}lysA007J
zB4<iYW%cyvl&)*{-5}<c(I=0qvnD>`^p&~80VLTDEwUN^Frr%0L;VhUg0ZBfTo#Co
zoXZu+#)Qi+<-NCvu9;GXg4L=Ch)dt7j{cp=gVz*7mrkqf2Z2@Ge})WtfnC{T104}$
zodsKWhl@w9=HDR|tG~nKu}U!;h^|w!H}UeF8Sf5Ro_F@AOO~YTR~o^wJJJtgYAsME
zo<dP3x6i>e`$L8JhVO@Z25fvY3^a_;Ixs8UXk;EfaQf<S=_rZ1p2o(fp8Mg-yiNJu
zKF8MmeQcNXTHW~bRhZbF0WMT$`^~Bt2`ZWqZ7oeGK+z#ZpG2qzT%c&6+a2}%I*3+@
z)1vyqx-x{#ZimV>rBRQv?dHlOX^1gFNu<7d)z4CdE_WQ7mCJ#&J!9$ed;X9!)k8BW
zma3!h+xG{W(qOf`D;13Y?4;2|zTX(RCc2I5(kN`?TiOs~oGGK;b)J$J)MNDcT)D1@
z?9KuW24m0FMC1$JY<XHh;n=d)=hl5m-ZU4RWyAjZrGaS+O*-5NB5BW9j24K5#ukZ=
zW$}x@`XON$@p>ooAr;i9=IKL#;3e<-fl~<Qp2TjUuuItfP2_@D14Dg{S)mQqPBc=3
zsqg|5t$fe;qXVP)vAIo5?EUxOX}8G=R48l(v`Mr27o@wC9AiCl&8_5KmIO+NUO}Fe
z&R<Q7kAKK$)+G>|UQ(d|XKzpxBQTk9$}Lr8O@#7hV6)GOoy}5kYw@r#$5^0$b?EGP
z<FEQl7o>S)a!2pgv7vXH(xGQNC-qc14{TR~EF{-uB#WbT<i1$AAFj=I<+oPc=koj`
zW7lA5^2q-<h&WeU%Vq`Ec^RN*!D#1jg3E(*Jh<scyf@29wC=vx73#0r-@1%H&|9z!
z13)ZDs0=Z<JdRroJt--lKM*S&$w=w9C2VVc6PuL_l|S^Epcr42#f^5(5+BG2gQhIf
z&wStq;YI;j!%(k$!fhF9ge-n^Wo&|b$b*k@vS_Z$r8axOW49K{ehef^T<LORbXFX{
zLzXi)zc@M1zr1JoZK6<EI(J-d>55%AZzyB_x9T8R<hABS*zzQ+aJhvcdLg6y^<CSI
zT&h$9|FSvF2E4DBN0zCO*{2P&YLStXAr}=Z&zSDo*UJ@Xgai~^5dR+?$;*WGNd)xY
z`|Wi<F574YIH!5hQ*4Un{{;#LuI&GIZ)5-0(gA7_rvH>|e-&-9Jg3zFaKz6tM2-98
zIh4;O8DaNT&AFlb>|$!wGjq{xuhP;CFsIhZbi5`-xY)=t#v*|vFBOt;$~z>Tb&f40
z&HM4Fjd*gLS&;;d?Nl4)q^3)R$KrP61^exu{C2!Ho7kqTaHYD)yU9)KD%#u?l@4xu
zJ~L(+sXfORlZO*wZYOV9tzbobeE=iDuWBPH@x55jnf7x{Vl*_juUt8!^@|oK=E(Py
zi4j~Q(wT~4(N$Q=PgeZ?7^J)kro(A{t;$kiri0%HQwOP%GA^I)(oz&}xjZ_yvAm2P
z(QA0caVLA5;!7O1zX|btGFKmhFfaDpp_)EBu^Qv4UHNRzmEo%o<h)V@%_m1XE(O}J
z2LMMuX>0UrRYPW>^*%tT)K6!u)N7}Jh>N&hcm<}jSH_d7v-kG*E<I=a0m|>C=9sFp
zWS+~{>hZz3=2Xs9Lcwqi&tVjqWV2@*S1wtAfQNjxU9^e|g7-V61N6z|{j>nzr3fmr
z%b`Nw;}Phi;5^=fn(=>oga52%J9SoaGaDUaA?{y{PH{6_qdL_UWm&o@YS}Knv_1G6
zDS=z~G1uI_48=>tqap$+;!Y-5F1DZDbpaizb?~O(w%kF=%U}Df5pj10{L5jD4M3AN
zdhjS~hmOW^PJ=>4%P!0W#$L#Ga3l5!O1^gQIOyTplQFxm7%(rRJtg@!87zxJ$yra7
z`bl2VrN^RZ$@2ReRt(<RaJt0VK)q#Gr(ymm48!1g(o^|kGe6C1XOF0FqlMCAbrq(&
zC_NVJ&?^FNTOa(LO)msp=cq1Z(okMmO>rn`F|5ouxrCe`c?d6Qz;o&j^&7G~DDOKV
zD&K$19;aYLTjvi)$})`nt(|4U1K4V1Yr_qfy{$hlOq(ZQe*?Bc>ZDJ5&7|htP^vyZ
z3?YBb;`qSoCB#)%`2OjEtM#JTr{!c@`bLPl@K{hmw#-Hg$dfI)vS^0h2-)v4BeU`^
z-Su}M)x`{`ZMaR_3G!>TkU>HwvW_JR6xDIHp;Sl*(1Kks`8iOPYG?))+_1Li#|_Q$
zCE(PMr7I}!QkETfd>kwUZNw_THiJ`XU}31OoWD7X6v?JKZbp-}?L|$J0aLlntm6)$
zsiUE?jG+3eh~_Z-b3p%>H0k=Cst@_37k7pateL|-Z^!HzU$-f%e;Dvc>R_o^nU-IF
zCBfxrrBJTSbU9#&?gZB&b*2A&uuL|k`e;lZ4*!U==%&GPo<%49$xUYr==>)PC#}#i
z`FhdXhS5o!4Pym_#FXbJHw<EGbVde<*?JD1!wavT&oN~2Y|dSs|BSR8i*wDPINOKV
zR<EYiIpu#csUj->7Qj2V;<J=rk<1i_nK+u^_J(g53_bS`Jbfu$E-kG(8W#AR^Y9Fv
zYA$6Xg2wPx3@xaTt2pDcK@q^ac@SWqG^mf;c=|eg(0#lljN`|Qg1g+>8C)kW$am@K
z3hvo;A;_#NoAZ;f-m-nXwBgt4>EBD!1t6B4K<*ol89&xMPkf7JfQ`v;;>JXAMN}1T
ze<a`NZ{vsmP`y@E_{f9Uql%8AR@sz$v9_tR&vSQ`((*?j!$7;n0_bls|38$mlp)QA
zCkH@)05=5{j;c`Yxty>3D#rs+_5zEaErEh&<R3yWd2hvUqY5!PqTBtMfW9*bl9Qr4
ztH``FCY2Jl{K?nt+w?8E%qt<f@L2YKsU4~ckHjhC1ibgu6w0VeBr?j1=;;_*E2yji
z<W<#Wa?zh^?K$BmZ^X{X?VW`fnPaIQ?AA}WkmWhm7=m=Bur=}(z+f|YwOuMk%aP&a
zxK(OUt5OW9Eh#U!L@jEC=yU^?rBDCX9-w(Y&6XpP6E|C+_ws&4z+cPiUnSrvbjs`u
zPbC@C++&3`1%|<NfQ`j58O6yP5Zq3F`MuhdJ;wIRcyt+Er90axMKGir9ioJ+&;E%k
zRuDdLG?gFDirx@6l*)-<smiHISGzltx4}xiW-X6NUXwVB5jn@3;V)ks$hQxbn7!)T
zhxf{HIIa2~bk~Yew%JdUd;!V&%`YV$C7TgD3qJIEgfCG%Cm?%?7!R`+?6)<~pVQi|
zaEBU5CE84>;mRv6G;GqNU_0=-@KB{hUR!dGI^rCE(ie(gmU92wf4ZUe%CozZ3v2yB
zE*3fa%kTdqvoMuTG6g!$CulJDKp%k`wZR2#e7BA)OsL{AmfyP<Srz8mr>Zf9<}Za9
zTWs)c63qoh5x@6uh-itN)`j|&Y@?<n)Ox9uLCOTiAySeMoN%t+=M7l)Ghs2v-2kta
zNw?ihJxrm`cc;`FB|Vu?ecf|+y^GEG2_IRU6)~K$8$Kn$g=`s(kT#^`gb+AnF+j%B
z_X>Ed8cII)Lg$kRMt-<=EKiHbdaR~v;I&tEwIeFE(O;|oNIe{j^SXI9k6M%up1t`~
zi`<@b-ygI1>xps<$5ISlT%<Tk&3s3pjO-C<G=wnmbJL){p#%ecNM_v3W6X$Yf29e3
zv}Ma<mS(9A-wd_P7<)AQk^JYVPzX(p^}<C^0d+|BiffYPxun6XQDt$3waSx#OAKo1
z-V|{7%fVbCWv65-PfKh@yeR`(E{~R$&Ib98`h6h+|EVSL3Hj!5@|$-?w)d(R<(mp8
zr_b5gNB;hNA6wh2f!WX=S=W9r%7HIeOQTKMHsZ0`#TLw-0ZA(+otei{Zm4lzp7JM@
zQW0rfE>Cu>nz>O6wRSvKg4xBHAt7y>pqG=Gg*i^e4`S9Mob8)C6<J4OM3=JX2JAqy
zV~;t-cdfpRi4yNBS9<(11z%4S>I+L^?*>BXLg&qf_)NPqbg5X&T#-Hvuh){Um!R<>
z**2E3#FALi*OOz-w^hv(@LjkLg?BY~GgC^+bo^BD7YG8sS~$I~ygR#m;_`b~@$Ewo
zLhy1y;anKI($r^~9VuqGM(s{l+54u|5-#baWmYtps%TBKl0WqR2OH8?wY^hc7y8vY
zD^gO7W53MuNG+?yB8i7om`3QjU1b9PSp}NXBG)vPz_z`6N)oxxQ3IX#V8-!=A++kn
zd5bK=caRI=!-JqW3U+_yyWp=4iU37HA!POmq~Qt0d{s|I!cx9+n#Y(|_7mV4{==X5
z=OfKe$W(!YO;*C6_-}#c4x-=MsaPdSzt&T+A3OsDd9q|L_=lcVZ27f$Y}=ezmQqgy
zGq(({o&wZ!&wDyf`VQs=>9>1s{H!Erse~#9<vhB;J8+|{ouyjZ>%!@B>>OfYT$uAn
z6R~2rZ_p~&w&Pp1!|6Yn`sjj*J@Ri)_t%i%3(T)TRRq9Cz?%DS+h<C5#m?+$ClJV*
z08{Sz4P{t+@aqF?&n{)xi>AwJ-LG>$**`Zgf#oqts`P+1{Wp76iy<@PG@lMoYJ41k
z=w9*YQLr9E3+Hc4a?opjseBzH2Z}32qk*~Dd-)<H6!z)GzcB+|uV2q;y2QC`FgV|i
z8FEDxi<qj$Fk&2_0OQOaz1(aE?Ke&+@l(wl5SwA62+!$I1Ux^Qv%mX&6J}ed!aq49
zmwRJyj&NtsHf`h+V;2EE7{uk693&0~PT0MJ7$zNhyHbhhQg$h;v3YowPn63fOs(JG
zU9l8f^cPC00Id9#n5{=03OO)zf4?h>9X`nm&so{7b&xd~IzdzCsRpI!0?Do;(|N}|
z>!E)%*&j)8p+w)vvv7f-AfYh(kv{ud7XD?;O&k>wp=0~uFGbn;6!*}I*3(UJd8;cN
zxrT6QT~?AJ$ly`EXtKX&$WWya451{cDaUf23`h@eig0MY6quo*U6EX>%?9Uf$INTu
z^nL7vBe<FaC1@9LLeoV;0vxZBu3Aa2lx??^?cGe`dVC$!5fWo_Q2L%9{`d6C;+MEh
zO6S7YCpSc}tLu(y&r8T1-ZoD#Yw%CKy2r{&3O5y$UFvqYXyb{S9;Ok7Xb9Oj(}X-7
zW*3i9z9--bGwA5n*6qF;l#2CSYwGJ&l0xX_*)}P8VaVux|IOZXe`k0~5cHJT-Of`f
zbRmVP`Yt)JplD38u0jhjjB&uvn_y)6ku9m6vY3JxD{E<5e5@c&S0Fhq9RNogGVl=0
zrx|lTrb!OS4ML0&Xv5E^5juVIcGr`{a-VdEnm>2hj1g^-P09)TG;gM27)_^u;9Z;1
z1+aJyN5^Q*p$B10<=8DrweRUOpsS%@#{IR7_1538gT)~<3)5rqboZ%hcN2|?tUHKH
zd<-kT^~Z8c(e=-PRO-QGHL;J`i%Ef~0+e8#P~hmJj5I9e=W7RZb6KfP8<cN~2TF_o
zt_{N}54@H)#^Ht+wslu?b6z|;T-(o>f&n5|f6TBA)utOARlZ=z7KLVKzofXkp+^i!
z*ax4TL?3l#N6={*#mvN(`UbO<P_lIekB$iT05w184a)ANj6A_$_!Hq{FPRl2|8fsQ
zUf2rllFm}`XS_K*x6KEPyF<k0D0o6=lS;U}`V7x9ND~p|dWE_;>6t0;ApG6y>x~Z6
zrypEpGd*X2+y8-Tc9ZKrYtV7XogmFt4DU;`3IUjqNX)~3IzBBp4@7$;oK|Nv@<*ol
zH3*~&oSn13_}h(iImLP*^#@J%*FZ8_`hlPEIrM-{*Yz&!4ubI$tjaPJ5`d~$jT6cn
z6sAUWS*ao^6qd@7D(lcv|1I%-jahfg9?3DqncMEgwzwI8{Klp%VdAM%?n@5Xw|N7_
zD3x6QL#t@df=B8o$JUw6f@EV*6IO9pK_+t0*oq#CTZ$inGsd|fWH-;W_xgs~$rYvK
z<2sMlwC+SY2wzu2MWnG@wLf|#FJn3P@%$WR!8u#`%bsh)W7&Y<ahK^1@i)<5IR)-j
z#5qv&Wh+YP`ZXHZ_c1LUD*GI-Vl5G?PNQ;3sL*-)hAES5l~0=v;#??6=BSeRdc-9s
zbq9pf+Gw=E55BK*sV{&FyZlLBb1D~BI`dj(t%B+F1w8Bm^<p2kK$atBrU`FVP$A4M
zTczX2i%fnWrT^5l|I^put3KZ{-n)hur|`)^fJl=nldltVJ&I}~L8`dP*z$bf90*iR
zj3V8SzbB7g!0ctM%jR1N^po^^cFNkKoqucRk2xkFI>SVE{Tc7%8}ewD5ya}w1}U^L
zvpc+BcmiW6r6s=0C^iY}F#_Mk<<Y?=1l)=49%7ICa6|@)yK7eW$?iIQxYv`4@L0F%
zD12>-zP%bmLF;Z8mq5YL?tOnSHN^|CN#5w7R@IHYz|4GlnV3E1rSry6%xkE!`$2rR
z2;*g~epq&oRA!s(%*IVaHg?aSEl-0(+t&+}a^u9o|B#%ReTFN=HV9>07w>o=ymE>>
zt)$K~;(TS`^jHaggvd**zUUdh&s=`;+AP^<N_X-Lyw^x)-^geY!I_Ydt#^)x*JP<S
z54dSxxz>EjoWWv&bwVyqpKj*}2K$${)~5@2a;(>A+~@AJI`%cfRcr}UW<NSMl}7%L
zE+$WXO!5-Rym=CctF*6uSbCI`mU9TjYcr_$vpUWsrZ}-mC-n@pXgQ?gac$P76bzF`
zq)9BRDOkM0o+DU5;EI5EetgOEW@$6~huc7byvlK^@>5J0=$@vZE_HGPFu%VSTb67T
z#Y?({R_1YaE-EYNIr<uAt?o=Cj>_cEbVIk_`0C3T!nXRNFd>5|!_2cavFFHV^5oL%
zth<C&7?3`McJOlY8q}(O-JIT!QT2Ls*L`rVg-Y$q$bpsn7w`#pT%Cik^mNgMs#&^G
zEhbQHf?xJwV>dCMw>sW}yxb1Zire^dIcqezwETvp09|u)vYZ?v&-CBPgWUvaq>w88
zSxHuScoF~2Uhs+J6ARBjd(xfYdkTnuQ^jU~1a#bviQ(*Hds+AVp{n!)b}|aiI4`gk
zb}Ep6Y_4jk<x&9+2UZWuHZ^SxyV#@dGU0?tT%YgUB`4kMD!YE}YYf^rWcle`&y<W!
zJAGdh`$H~U@qA((?rW70C$@$=?7%g1VJmYS#X8+=0Bf(wn^emLnaHQY@zxGL`d;37
zA6eXn4@~DpI?`EDy$;0+qsNhQp4Kn!xp=HT6<G$SJXoIf8)t<VAZtZSo{il==6d!!
z>J#!F8N}MmVohZSuGqwuvR2R5_($f($LUT@E;zB#_@L&@H!z02VYayo$zErm3oAxF
zjt?U4;zv|QoU0IAL*DwIvTeLEeqzKA!k4{nk@UrIZ4z%E1*RVubmigZ+Xhzd6&kEu
zHva<Rep)&7`FHu=BdxP4!HNrK{0?3n&nj9w7TDhVvmF1^d?4E}wBAAzY*D2sC$HeS
zv=~c;tIS@1tKF)A$lj*eMgH8zds+n3KF)j~GDBaSd3U_o8uc#L8KN}g)mk^Qvs4M8
znb+34#B7!~EnbO6pSQm{bB=!uo^>t-;uyG@TVI}hScj-luZ5?r)lc+<3%2?Ri#c$a
zW>^j#RGkaDX}}l@(S@qHHI6h;6d2ln$NaAI`c!c{Gos($NF#UDe`Knm&|l&^19zyu
zyW_Pn0H$CGnvubaq^P$#XItu*Ab(6sqk2C<n<j(4>NaE-m)F-l<HuUsl)7!-z|ZBW
z-h6i(056#+6>e-HBCR_i{_bAwnHN_$3cq33^_L%OVp}eRIIvRGFdVeyj~G>6o;3j5
zV=X==Ja|z>n&_4b`U}J`dZl=9<dg&p>H1Jd=272bK;i9{x5i>@BXHiT7YqY*lb^tQ
z&8Pl~jIDJ!{yTTb{mm6(`m4~RV<prCYt~e?t8sr$Tdxlq<_kHPe#YAf+lPfluhELN
z2B1K%_)LayKIjyu3s^I?Gq&Sm*LJ9^C9|B2RCzy6mncSof|r%~ueAokbO&dq3wd{&
z?d4L9G3G1;YRGgW9=X#3``U9L;*xM$dqe2Z$3ZS^h<ZnMMPeCH$u&WuIin<Kn@{f}
zr@&T#X61s(0n$s4T??e0#(QGUoOv&CE?`lbAeRYY2#uLi?5&opT^Tca7fHqooha*I
zf3@3-h2YG+yxmckJi@G`@4eTo<UA|&5ThAv&hUJjQ6MxXh=aKp&)`!-L>0T??Sf0U
z2rPhhQQ*D*5qigsoccxiG<hhaXzGY@F@8}{lWuOHn3R_6MJ`5Nasls$^B6QJa%kJG
zZ32rjykrnKDg;^f+y{Xb91|2(Hp2Gd^%!gs?pN(o#tnKojI8ZPrx>aBxr_{KWN*9=
znCkRhEgIJ?(efli6Xd%a5BOUa&lSzYo|U>!;g}R%POB2(Q<qu-ULdf2Fx<(j!(jKD
zq@f=+2W3T4hIDd(xyJ{C^E2=@(d;)bZc&0Z@wi-glg#nJkm|45C#5vkJ(6*}W6tBU
z%tqUsN071UGC4`lsiqu<c3z&I!-^uZKpmvW_IHZzHH;d8HgzD&g0o5~)A2kNarOm?
zesAG*YUivw{5*L9{m=4b_=z%1h8qRRYDW1D+LSUff*|u3fb}3q8kaQ%!R2u1OMv@g
z1@q#Pq)I(dQl(75E1F4Q^4>q_6pNvNhIb#P=A4e*HuDa1*B1UUlGHAILI{977(XAB
zJNw0Yq)jR_4sXS4o2=t!%(NXVm$qs1&5P`n=vfiB)SPpR;YRJY3;Tv#GB`4SShzO*
zoQlt9suokQA62%<<LN+ByG{nyTiHy*#(c^2gSNxHa4k<m6MJvxh8DFEI#wV`5F-n@
zAP&vuwQ&{~@F4ZA>xHtKQn<O6J+9_~#)GV1{CheuXil4Z-0e<b>oy+o*FNZxhucq(
zOT9}ych&4SSjDEj%KLDc99e_(Xvie<8)y66WcvQwi<nEMnw5CfTs+ofT2?j_X2<u-
zWymA;ScW2AW6+uZmVEGrJ&kz-pRnFB9C$gmZMdUX>i{!|8VcGmbGz>cA+IueWZprL
zuTq4>Y>>mt75|d@wZ+JY@i!iboP`64JBXs$^gXjeJ${Zy6))%{{0Mo(JpAFKCo-SN
zX)Xk#XDddm1v{?N*?C$DpUNH6g0g><A{KI+wM1GfRNT8i+0Ld8{&0}Uoj{z3zH%4*
z`K|isPeV0|%{lMDhTHhPzKGsI4k95`b%De+L{RnC%eVm&IRz~jPYrw((&<8zCj?$U
zYRyBbah}#?ys<E_&IP@}+~=J&5}GAU8B~|j_dDA4hl#5<NY|mvq}euxPi!kfog3oF
zZ<m*03!^o=P;j%B<yFFdM#^SMts}f6&}-E;OWSt3w(<Vv^<16Nr3NGCdlH3y_A1g#
z4g_Hy)kbdUZKE!$A~Dr7GU0iWu~B5q{k}ImlAEH%7K3gJ^8&fW8L#EK9{fDO6f7s+
zwdOSok$fLw6Iq?gx15^H<(T#Xve8be{|e8oP&4qncPjNgU2nNAK7xgK6J2)&=geCa
z{cv(F;GNtg>v0AuqBs6ro~otQi&ezl^t9rDb9#A;B#;RM07w3tpbZ7+qW)_M0i3W@
zc|M?07O!p@)OX-|nIDyq?ecr=_j}g4XZ($T=UOG~2;Cr4$u<tnB-Q6%-4L!4TaO>e
z{#4T3etI!oD*Y-SSa;(C#httO$|P{Q<zZY+NSmY|gs+~*y6Xr}If(+Hlk^Ws_D@M(
z()a|**E;XRMlD83d%bhuMFrtHLLcg9jn)%B(hqlZ$Q=1LtWC^T&<`^z7n@~tqf)C^
zlgOk%W~p-fT;~_-a{tw7ZHJZsdzux)XDehVkbWz|A7<SqGKQ8#22WSb)va87Dt20S
z%<K;F9T39NIqaDfDM?CslxSPk;N0bPu@Pc-t<W-IV;Y(QvG+Wb%(Dx$r$yz`b%ha-
z4MZWrpp-$|`sI<;8^c1!&>K|W4>&r|qvdJYky%N2f!;CEr<bUT2*o_pTLH%_8=bjH
zWGNPlRvPuArOG3+o^~NfqlZRkuKoJT@{y#Vrrr@_#3-VPMMfNI>Yu-)+IYS(+j#N3
zn=|CnGi>I2@70`-p4ePI*KkdzndgH?_T<ci$ATrakA6qCz@OR1x0_FEtm~jLAq`{p
z_xf{$tL)Qu0MUz{UuEV%0|znNWuS|a^l#HZw2nW=@zC@Y!EjA=XTb+l2cPbYRlZLu
z!l0ciZ5|Qj9dGtUbK0_assf^a?LX4APO8hPtqn7E@QPyIM;<lJQBu0N>UUZH+TLwB
zMp-u%*PBj``@jr|@RV?W!LAOz>$?7J$x>o&x_gduQIXmdC1vmdTFxD~pVO)#D0W+$
zeyJ>=y`^imVgoAzU^x_>oE)7Lv65NmMwsXy#Gv1Gxvf5l*K^G=MQ2}(?7RAVe-Hv&
zxe62->RvQv7<OPb@AYORmkxrEtmm;&Hu}P+vV7p(Wg4vPb(FZj_lm(%<k`;=s1`vt
zgLPo;E0wN#1Cb3kXow|-lW5ChSn`Vq3(FqfZ<#u2@(P+)X|3Oexh8SP#!x8#XLnNk
z3MrGBrS6M^9JEXPTCum%#^$}j9!x=f%eL(kcwMBCdU{${YJ9?&p?f1@t@tHJEW!Vl
zqKo*nngnq+UE`WJ7p#yw|DV1lA6J0ZkpI}lN?P)iqDavEXj4vWoN^N+o<m!GXo;MD
z7tJ)ua@GtLyH&(dL&kn|_Tt|T>a}lh8Q{xp@rIa0Q<ekW^?q>2clJOn&F=JCpHY`u
zQ~JjiW+HVH)nQK6^`>>z`~QmO83d(4yR2m8&OewHi<*zjU>I_`e;!eN!Fx8ZORU`~
zdEo0FhOZw)a}w49aVeCUswP#2iH;@WD+0$uUe?+F9`)lJ&iXb?E^ld=by|fWDV|#`
z!f$DFNk6@($5<WG_`lEFOcgmO?eclMR#hK^YBYa;@h4Qh_UFQ7gegP#uKA4JsD@@Z
z(l(B#JrtJ#N$<+NMsA2l2S0dq;2%G`9Z>OE-H7Fi7WVQT)$5`iK0(v<>(%Hj2YJZh
z{-r=Btmbx$$Bn(4cXJna4V?r5^De!fp}MAu%^<ugoZ2z*-c=>NMGIo_gM(LX27Z>#
zKCm<^BQ?M_v~`1gb431-S@1?^`gT?$c!|aTX;%}0Rz!iKHL3-tnL3SJKI5o5wR0Ax
z3C&;0*}E(!4-pKkIY-A(65q+5OG{oAN+JAGV5|9Dtz~ZQ`kO4dlvtDrp?P#kco1<3
zpsk`8ayu$ou-+OM3UO&-S2(gVrS>5PyM~&HDZ{VgoR+$Mc#r-li2Tj)9f^CH*;ZBw
z)X;t10exSp^<U?}|35FICjl1wf6=|b=ijX)_4<F+v&|#h<?P$&tN$yWFr#4Xja;z@
zb|bz?;_amw+z?a#Dhpp$aYcGnlO2-C5BF@Ih~6=UdWx`z$oAtZ3>_;YL7SUdMMp5r
z-Z0frQWV?#8w%zU)_zJy*gTedlNoB{ud=Vc_^@Dx6#gS>MD@;?kLvoh?>7SGY*R~&
zO-CCwy<x~3Lo9WvjQF8a#y^<#TJ5xiJzaLq@7(^X<{o>x1)PcW$JizxAkKV1jMLWg
zD;6(v#Nus(>xg))n!U`XzPU{lQ-<kEv6GhpeV|+DB1OQ95_i>q&&C6b7IFIXu`;_~
zMa~G)!BF~h97$9)^j7FW3TG75E|$wX3>}~i^KL;z{0PBrx(rlk7+|uS<p)=>9HFfa
zX#vYKy1c*lyLi|0rKd$3M97zlMXU7fQ$2wKi><-mlCKCs58$RbKRVgGrI*q?aMd(E
zakPW$8!%4i=;}bHwzP@{*>ouPo?`EnZwrllA@QC!y+{0fZtJ}1KafYPD&^O)p;T4p
zi_>Y7$O7URBLJBHMqGL5p=SHS`Ht{T$?Sl^_GcT;!K<6|Ag3_ZrPhZgQ4jJ|q`BL7
zp-(%z@;>Q}Ke+j-z7{-4-ce4Jixn2=C3VgXcz%3UWY=CsToBCF!<~!lmjc!8EVPe#
zmLq0+t;bQ&Gi3!I5=N<6CXO&U0RryO<3lx4J^KbgWXl^g)uc>yW-fC|vdW5KJ<W;@
zW}^EVK3B?O<rh0+q6>QVeEq(EdJ8k+5&q8<)%SS(p3%_A{KOBdv+audD;rs`i8q9U
zp9-SOfqjfv;WL(Mk2D*A{{Jyfm+tD!cjCZR`@C%RVNuoKI+)Xh`}_8}Yc6Rz2YT#`
z^e0YmKSA!%G?hCPgTY}hxKtCw;sxeLZ{2Le$Y6hv>q9ZZZu0@X=&{?F$IZy}M^)lM
zzG_UEk)H0sk?w34KbO4#XRd2z$)9CefH;i2JOjxKk3po)RIx?$iuPPaK|mCXblk#p
z(XCr{hOyb%CiTdW6z+I}+n7lt$n7O1>rx~Uj^ry5%uTku%{aPXDBnswdN58OgmQ5=
ze|qC9H;DVoV2k&S%5%eCyLhf}Oslr06wKV~p78LVw={F@VquA<$`x1zidT<E)-^+1
z*@Nf}<ZRERYhDqhL(h4hX_a1RO-v^a&a+;}Bmo(OOjD6b;Uj8$jr!Tm^<zrKhplz2
zSMDY_tt_AJk!MLGLE3V^DUuPOzB`Z1VXeO;xYg{NvwSWg5-HD<nF#?jH`brawE0F*
zcbwxda_srhd5gmhum91V&%C@PQTuQ1ZL`sY-%~EL4`eWR1uk6oMeuGtEjeQ$$3?sA
z><<El1+YtoqdjP3D@7+=vByv<FjJ>5d6Q$PD|Rq)msQ_YfJ8|*ER*Xg!*@F9S1m#m
zWp7PW0TpGMc9Xa8w{>c{d;2tP&mh)=3Lj$crS1(Jj@Q=4+Qv2XhgW1;^@gDb?u-HJ
zdA~JAbEdb141eEkwD^rN`8qYe&-$>oLhmhMu{HbqQ~CFvK-3NnoJKB_r_lVH<|5i9
z(5e}4_pbabsA%=>T*-F6*FM)jtvP{pz7=KrPN{7OOZ&7%0CG?wV*#|0{E@^z{LUim
zBINuBpGb%3WubViZDrd9=_e0as<C5@lP14sT!U9cB2+8Jr_gV{j>Xi&@FGMi?|*3&
z|23C%r2mr@!hl*?toV1q$n5a6tR{JF)Oj2L{+|3;n&E9!b56UNb0+-DKXZ!ZPOnzQ
zvyCO92XaF?<L)%it&(EJwKD;_hax-h9-&oIZ2koNp`=92q>9AGwbQWw(d}2VR#-a;
zN|k4FAuur``YI9soCOYIH|6xZKRP`k*GqAChx%8geGgH#xKq37$Hxec(i&598`mU*
zdB#`{i|mY|;*!H~CXC8A6ti^09Dc0Fl__)2ZZ~d8ZfEgBO=C7`tO(y^#5XZpZY=&%
z$%F*<i&l+;WC6fbc*)7gC{UR@U}OPN|4u2R6Be^=xE!X>Frth+u>mc2bLc}RBv$X*
z1#I;|V)HeItv+!{Wr1HFSWfilD{-g4&~9P>b)Y5q=tHZvT@JRNQ0|2d#%86e^~2KX
zw6C>bwS}hwA$_)a=9MIUoG-vb(8U(WG$Cp9=WtY2AHrMiBRI<NPUew7Am~6Q(Zw3)
z&S{OKgZ;-4l+Ppp-64CaubVt&b(yU~76^27#n%eiq|cr#`3&AReOC2g_xm7u7mBB{
z-M0Qb^x4$)t?ZwiUXwKro<gw(z7Ttl<l^2Yu^U(`x;;P>gEQvnkSE2fDtoJ4V>Mlv
zyriL$JIF{fSSb%=+YjeM3M65kM7?ntH09hT+USG%l`MLy&&go`KmAqF2s$N^N;VbC
z8y1fFUn(=zuN+QC3d{t8|L_|_g|tBdUP18%N<~u{_D^M6URr20*fmx$q509)%<o{A
zfA+P%9B2zWL$ru5PNXb~X#bB+Blf3Vpw7t9!ZFPCU(XHaJbBh3@DdsJot$c(c|N;G
z!IaJ-S8>ZvMIa`UvL2YmM#h_kMrWK9{+xB6*0Tx&W}_(`vo`h)Mvlab*;?>&8}N|o
zgIuL%)MZK0gC6|o*vhrn4CgpE2s4pv?R6ZD+xQt~Pp{RLqYFZLz&`#-CA?0)=s|dS
z@|i{-(Iiorp+2I3bDHU9d`sq({nBU?gQ346-|Hfd_(Zq7%1mGq(syJ;y+xM21pG`a
z@9k^d_(wr;0C74we6AIl!l1}d*8Q(yV7i9JCn}~<g>+&uI*_|OWKP^!Z3%zb#`y}D
ze;nm5chSc)puxeMey-wEPH2d;glQz#M@cO=>(+{(BB7f)yY)^D*ITGv^;Pl}0fQ&c
zc5IdH%`qazBR8tsw8PP9M<Qn+)$}p2U9P&Ndx`m7F7N%hk1NsmkoLL|6Dn_TRG|-Z
z)a@;oip(1}6Yp&i=FuNjj$hE`4b0`8rsUsNrv+aqaxg#2V|@!s_c{@^i4o8bB3K4N
z_!fLF9@(4E@(%uKo=uo5H<Tvzj&C?d$?)OP@dkNF!#@7irxZ+$g}Ckx)$XjFT87Bs
z;1TbhtA&M=Zv$a%%H_9gN18K-b^>eSXfr$^uHo~Yp?0~OrYy`#zMibOnCqrI{1x6r
zmi(kWq_Y4gYL<C6N5eYvV1g=N$F_ZPn!4ux@m_Ca%}tRGr^~3^jg-$Lf*T)dU2d0!
zMAB9|pyF@Xh8BxBZ9gE5)LgRdwcU@?p*~D&5jCdKRql=%*RYO-1b-vtq_mGFm;xe#
zTNvl@`^S?8vyk(g5UXHAI45rH^v5j9?%tTv$G3!o$}ei(E?gaUmR1m-0sq<51Zj9~
zwd7ya@c)nF>G?ZZbE5xVq=Bo--%iot=`o=_h_k=vtoGco9P4WHGob3>r7k9EMXx!?
z!1Xo+JeWK(AVy5T!QIpH>!&}^jBlyTbWGa`6KxwNDVB3NVW&j;IUL<71`6m6iHw$}
z@oxP-Qu*_fPgpacT+Q$D*ChPmsr%QI)cjc_=Mr@LAzpmzLvxcm07a>%+7`vn^L8g5
zk~>x!c#wg`)PGgu>FMtNLM@CoAaj`sm_C`|SAMxvqaKot6tUMIoGVvgpzsV8r|tpN
zhp9^?L*8qQ4l55rk>tD$15|J2Z{G_tnO_7(-3w%<Olml~(H;iF_%z#nc80jL0)2}3
zfx6wof)2Qh%c1|eR9-Ce`qO!mi7$Wln9IXw;!rIizC%5#+^sQd6-|;2+FIGKxi?F@
zg=YY)iuYr-XpzgNnl?*%Rs;2JSWJ-!<#Mw^Ldp*Z&J06U{Gf{oogF*FSOL2%0V_a5
z2ps&uiu7#6J>4nH{(CpQ*Qa%6HZS}(B{fF09h98>Jy3YukXN?YY_n@M-JYYmZkO6y
zglsV9DUmrRF~+^pmg8DZUtiv!X?jVluycG^zt5_L<wzC0!1{(qfH&SqE!#DXmr5lI
z4}#Ji8nylENY?Uu)t`P(IT9Wb@F-5-0!SaQrZBA}+2{^3S{i!qvX`=Ft-Nh<G`oIc
zHy|HMFnGgo85=1Of}1rOut24+v}MrFZecSC;xBG}N0Q|LU7COfDIV6<ksCe+{U`x@
zH&>(Lwa-@P9iha_JoS5G!{yG7Q|I8<8qAd@F{7cO`F}Sy@7kGNKI;(PbA9w)v-8PQ
z31`nQwXHt~;RaqQ1897#IG8gn&Gyk*S^9bTB5Dal_>DE)0WP8|(EH{hKzO++&u4Hg
z*Qr-z&@|=P8Je<W!m!@f{8N(hIh<N;It9aZA#9ml6#IR!4*m=h(Ywr{$K%O|DW~aL
z$FiSy#;-USF_b4+$3lJUh&-b^w?QR=U)9$kPQJOel4Vcf1qK0rEi!uUPvU6Piu9Kv
z{~uNV9nW?jw*TWH_Nv+1dnIVCs9k%{Ac#?`B%x|6T~tv)?3LPr7*W)$YR%X+qo}Hy
zMQIgvbza|huls&{e)sP`f10QBbsop@Jjjf?l<I(L^D9$odZ&L6iRpB+73x|_&I+nq
zj~+IpAyRmMnqbI^T|}990U2tb+Q)dgc=~bq*G<=~mbS0Z-+8MXuBP&Hw4;003QS=^
zem{QTL=8;q>a&3Z!2{u$J5-2Dn<30)Nl7h@y1Ca532iegdPgIvd>LQn`NzG<gIM=b
zcY-&NE;cwoklMRH+<eyb!Kv$T7T@AI7$b7zPJi0S_t_8oeW+Mn`vP*@((#so#Z{|F
za^{sre)~Vo#czZfk%dwzF6;&;WhAr>A}`IasR@g?u<c3SeddhB2||SUOCpI5OSKoI
z{bjPrmA761q>{{(p)zFiFQ1Bh?6ZGCJihK}uP~1|WjZ~ok!9CjhuL5<Q=^o&1L3MG
zP^oT$`GP-*)NN+@ikGh@^KDE1^sbd#xe(Nk_rRkCVOk~=9XfO}tj<IuR#U1`Lv#G{
zc3o%aup*uBfq5Ac6db-*5Gn*=KtKmsRPd88=e)~yQ@2A$Rf|HSBw6L7Qg<E>t4~=5
z-j=e{!o8%7Ibm+mq2Jp6oRM>73w7Oh&-{+Ng!Plb3vVf;-mB{-x#Ib4cpXGOGv&ht
zH!Etd>O0anU)xXlMwAr3k(%)>u%>i0$@?u=7uOzwai_QNQ+O}A6KT<A`TzPP{#RZ;
z_zN$l0pNUa2T=XFNCagtMCn#6ez2GPYq9+gxBpA=*6>xP*t6;C(AJrCWz9L^*!xcM
zKKC6XHownNF?<irnF9H4bA!T5uL;8Mp?p9p!?HXQT=U%JR15QZu?jp{<#uS$O8Og^
zg8L0uZ&thCxdG}`x$;+*ldG>}9hwgjt}OosRgDS_+sd*}fE+UA)Yn(Lw6_o{#+8Zh
z%Cr4fl;WN2db>meu%q%OanJID+uj!Ng5X|`!RX{fm=TAt{;-n10=Ck!BXq-yg7HpL
zCk7D=fP*GEPD1pI8{3m^FO{fo3*fNE1=CBR<KQ?nm?KGpj7cQ(vYW!pjFlBbf6P^J
zSQ%B+Ef1TcA_BB2OY}l=5WG<@!)OD<Ma9wFq9$`>Cl19Y9~rDmkteuXl%OwKwL@p(
zvaHRZDI&m8mbhxC>YN!(Ja{uWfoJ$oPZF){RA%}s+S7^w62uI(BYc#DnQJa*3AG(L
z@zo7e9MJ9qy8iTodw6uTLQnPx2lG71s%4iaN3uj5SOUoONMG(us9JWM#1*p(b>&8v
zIM;9T7(rrmFq-&bRK#<sawoQdryf&fd1k}my{_{-ehwzTKk^V>ML4!p3{+bfi~GSe
zi#G^u7VS`KTk7}%u@1@Y{ZspTAoC|X9wsedtR>#wonU3dDov}%4gCQ>v|owqVNX0I
zj@s&*FJU$rY>zU=tMb*i;Ji#pS(V-K^#)!B^zU@!9(RUCTGoDY!A3-Td#uS$ys-)9
z{y}lOa`pmO`TWg5=$V3t8W$M*dVj&_R@kIz8eDN$EKPS|>#Q6Vvy-U3kk%RekH)XM
zef=Duyl=MF|Gsr)-n<5nT-Xn;<-Uv7lgfb@j&|L)9V?SH_YpKzv;-`tX2r7J<6dRx
z<Noi)7Ptbz(hkjPC#>RQMOxKj{Ord@247I@!sYb{qN)+guX#oSINStdVQ9Iv9)?-v
zj~trlO0QbWyy|=1X8u<rb%M@R@F1t?7NW&ebvx`q{*xDMdHy~z?KVdrjR-k7h>nHX
z*3-}UBlxxL{0Evi-s-|QESZFQbOuw1NWROb{sS@mvVdRqdl<&8>(qGk)9f9`jd!LP
zPN~t5L>*c>8lAeYsYO}smBJL8U308nSY?iN5IieSEl-v8`9)Eju|G!1Rz6ieE`{qQ
zkbU`C@G$}p8wzs^QlGB2JC;r=$p18|>!r*<IPa=Fd!~K1ep}}2mb;>Q)5CXPnxBRv
zDbeOQ=n8wbUDmuLKJav!6#mReMdXz|;>OPj;QMK_zOKZRb4{Tv_K}+F^?W6ZxXxWi
zy5V9scBjdx#<iUk{@9yYnaQCe82{cd38(hQ>7UPg69cV13Kz1n<Dl~i-<qDK-48Lr
z*eQ2&CnW(>5m3@D8e||it>z6RU!!DHhfMl*;)dW<rRa~w!T}Wodmlk_VctmMpZ@Jt
zKCZaI>ToBMh0&FRv6md3;57b^H~rr8YE1brCZ)AGM?`Yp<uB;UA8yy+P76gwsO&2n
z_SJN%t746+Xi=~AdHk|gSq3o7;m?;6R{7TiI4A8N6?pBl+VawVy$vSaQToRwCfH;|
zvJ7c4%MQQ{{;DJY`-%k$2?}Ysfs6KEM!TK$48<@+=_x&O_p=S}IjP~7>~x(68(rue
z!Qm|rdBQ5{4j1+o_$6bVFPl=%uo%#&nD|;#q*^shqfeDZLx@(yGgnzJbY|`tt|}_B
z1y2)9n>W<q2dpv{0AN{+J4ux!W?Ta{12DDHs_IwTX?!wA{mFB3Wrf@?WZJzQiJ#Qj
z->}hXI$pilUm1=Ib;%3odonS4T0f^0{2R@=G?>{|9&9x)nv$p~i~Ed;=D1J05^i}#
zKBjR(LV|}+Xg?G1po_;Z7A9}WP8m)2^nw|-22b#_kG8M|lD9h{;C8>&IyO@a0<AxH
zUraswaetk)CBF!R1PP@_NLy4<i@z_{H*6zEhmoyiGUUGFkRP980$qsw-`NGs1Uv|s
zT{R^`*Hj(Y+sNEWk3GVmfG=jY-(s(-fW5sd*Ew|jKX$P+3ee%>_#_f}QtRIn>_r@A
z5Rp=Tjy669^A;W{l0Oz^2h^v+lyaa{WXMnFl}sl0E3xP&b4fG`GOAPG`3R!#EPJ-q
zH!^3<lA?Uy@JJPFZ3ADQz|4w_10x4=Jp_}L*)rA?UXBr2!%_^tVy)QAPvBw%xQwAg
z1ySMR;d+cMXrpZ-QCzsDerW7eiWTdzcIGTr(O(zsVD*IRE|YgLQ<FDw)wYfJ*+ITP
z5hi8<7Pix(035WoVr-h(6xmv=ICf#RhwsNka*NLeo51*1+LeH5gmhE>r}0+<Y47|3
zd@sPEy^kl$o#Fr3coGVdNpSmVi_j%Lf8i>udD49G!Wq8=5RWoDt09|d!m(L7{R@th
zT_J%o-KZYFe?h(CO&wsrxqWAYDQuD7v@uV$WCQn&!gVU6_9#w`vlCdaDQ5T6aoOwV
zKT(`3?2k3sC}<Z;4`YAggfE^Cm1K^2rU7H+PTI+A-bX8?p+T=9SfpXRMgKkP-*T3M
z+6nT36<7*S%&?<<+BR{M7rd=bbglMGRvklTrPk215csdRZJkwq@NRYYjf-i#7qn`=
zJ<KaJwSSM6abJsV{xn2s2NW30ESUk4<*|O3({hu8t4TKS@#YbvG8$@K-_LhV{hCGN
zj~RL>0ZT{@-h9yE_A~jUntw3bcCk#hO&Gtjy9<YIAsaa#u>V3ehrRb%LRAZLI2{E^
z6xoimS{lGxQWes^DaR}>iwfqE0_6zZkLdPqi+-zv#Wqys9N*6s!)q13c?e`Pw7L9`
z)PIA-pWa@o`R3P|98|TBe)OxsRXH<tay6=M`I#utevPUM8)U`4&dE82N!&={W5(s_
zR0ikl=}BP3;^a&QW+p!rhMG&5d}<0dn_Wg0sPpnQX~rNoKI{ZHKCsxUY1fcGm55RA
zQ>A1^ldKr?^+Fv+r0F0^712x%33HjF9ya^2e*6USaq8Ilh1hxaPnRLu&1ld(-5vji
z2`Z6j5{q94i&2uy=edrHI>$H-nnj+AHO&DUjpdi1n;G~H)v=VcE9bWxJ;|fXDd&I!
z6E!xam%rVGL*zY2l(78BYWa><$(s_-{P-F@sz|OE_c2r9<FT3fb@Vc+Gim6Vw0G+d
zTVI=ZWVw~<FI9NA6l0X$e}IF()|bEek%D#*KyU#JFsgQ<gqk=g-^;yv+`6yldDH#h
zA><Cymr8{?57SP2o7blI^T=fN41K&B@W0MPj5~yl;TM|4ip3x6?X5wN%Z(t6u_-Fj
zf=On>WLo-OFfhYLXa3GeRTc|FJ%)FX<cAO?oDd{i4Z`v?eQUS+#(UG0+q|D$9G8f>
zo9fTK?qIvc<3(&JSi10nh7OaBddVBl;an00LiCsjQza*Hqvh`Bj>qX$r4U7Fhn!(K
zVGNUaR?A$LbJ!LZmFcuoJVH%mC^{Zi(ai-y$b<M-0w46tnZGnc89zH+@tR0T+p>$B
z%2y_AMH4N~>)GeRpkC_Sg|ph0*%wG?x2NOkO|eNcmyO|z@(cDh+${FMDwAW1U_S*c
zcLVq$0jzO$6UJ#Llapk|swm8=pu}v3qE_Sa)KBnmF-JxzH9Zwbo8o)w_yPQ%81xs3
zJjrb*!DaV~s)&78D6r&t;S|It+&wQK3IkKr&_Vw}2?7Bt1HN7eDHK-&82@+Y@X}Kt
zt;uNzOFZZ*TT)3*M~+~$Oxv?Nas5=<{`tICR*byE-jE1QZ{g~BjTIAg(Jwg_yT?0`
zB)^XkMZ-ihHVAi$uf}>f@*BeV%h!nNn-;xU#0p<`*_s^d%tup~_YE)I*M#2!{(dv0
zoTFso@f8G>qp#@7Q%Ij`lSh-LWpNYciOuBg>QJ(HO#13P%Wlm+vspuZ(eRV>MiVWj
ztxx?)?>f^2Pu%SH!!|CGqiU`W^8E`!QSNY@bOL^^*3h$y$MvmXKqtLHG!;k$?r1s|
zMj`P%^u6kr8Eg7tH}O&LEgq)$tYf>MMcnC?(+EZi1w{e)C>}Z!J59Wq{|gRzjys$9
zfyG;;8=(QgVN?Ah7(&xnPOGL4m#w9MG4J=sTZ;)D(4+zb`UkO3CEJROSiaW)#&Knc
zl;b|8t$;=UtwrBjqU<fnJ5LpoXd^Xso@+e}L1z|Gh8j52B5T`vKX!vziaw$VbIi=m
zSMBvXOm=x|y}F+e4UUytJAif(rRpJ+8Esto&4Q{#=XS@1r`cl#4@~M?ARbGD-HY2%
zx^^o%*7jGG$N!08oF<_fr4O4a;XQQKY<hgPVdh})T1gKJy^kz=nu+Z6^=gng?bb|L
z*0fNA=L8ZVCjH%(#LCm%_>}vA%oeD*GfP^XVgLvT#B2sF6xn(kgBVZ)eQV;}e{HD{
zA_rk$qm`<tTp8jYs?p#msN$HiDJGZxn#&(H-j{T@`hHeNHTmcVq_L<{cJ=D${dQZR
z5>G(p+CKS?({CM9hqdGu%W%v%mt5LuQ}&`%l9G|J{hsfIp{s9K-5}=}@+gyaq`s@*
zv)HA^uNj_gv!c?ClX=%`SZ%U_wkV^D5no%xCe~17A>3LSDjZ8;XyF;);oT^n2N&Ck
zw{@}<uWwZTp`_dn$3{`3+TFtFBn5v`vSMhu1QvDym`Ht3g4oA!w`N)B$j>QPN0Ti7
z%cOvv7Mt~5)d`2K=M`k@@G(?yQFqWi^f~jW__L2f=HkJ$I`xY~z2vMcE6<G_8clGv
zX|Kg?OxGt!utap@K@jP*;j^1wzTM03x8A?U`<y-v#x<L2pI^kzO#Yvfn**py^*??K
zC<$OuLCTqug7fTG`7&ZP7io_t!dHN?CIRbqy<K~SXLOt!oQax_&8I5P!04H1d^Yp#
zp0>C(Uk&DZ>OQv>umtyEIe86X!|1vHYD?DXD!*V-*{b^Y%mn$11a4HX>BP{9sq?rl
z^+ayeY>VqDwwDki9b6rkLX``%!pu$^?xSwsJD?%^tqu1mI97n@UZXk3;dN{G+y04P
z#`<yA9}1uD+#1N?ofVt17*TChR-@-@V9$7GCvCFIA76<3MG<3f(b$LTkN=#f>gH{*
zS99hDU+Mdq8@^Uu5*?)e4Zx8evnXv4m3InxT<#8IR)}B04qho@*P;)QEg_&LOwx?J
z*a(kkV-fMgN-+dfS*i)=Q+6CjE6-~37qOcA?PL5@S^hpRA8Es9#89d%-ozW8a!Ag0
zqfz41i~7cQO!2^aW`Y7t$aK%*G#5Rqhz$Kk8sR@2Z4fE;rdLC+N(N-dz#N<Dx(eTN
zWc&-NFPtjfiR1lb%EP6?rI)moFlAMC)xnH@{aaNO)mb@%@U15x$`VyJY*R+Hx)Q2Y
z;-4JU611v$Mr;q1N|;>Hyh^{)Rot$HAS+ypLezme-opldBF4WF$jNY=wRn2E@`QS}
z{%}wePl#++VG$6xHb7ssQ>xypCQ=c0;;~-J`aRf0`POA_mv*^J7srsw$z2DoU(uJI
z;6`|>@|qJ}YCkjKSqFhoQpqc(QXOg<jcA|yZ41;ciQnvR_w%Ni^eM;V0&|abGXz7P
zJPD0sb75+)Qt|w{9l(75K`5VYJ8PU)8TQ~~Xe3|EAeH^wrp?oeyJ8Dn+%&`bilcpX
zR+TVbiN&(_rH~dj`!_Ig792(Z!U^4MzaAHC{j&YZW!Ypj&hV@$!R8UeqKV+fBQ@j3
zed?8W9i_I^_5UFIrsO<r7$uMK6gi`*{nvfPY*0S_tV-yZ6auY`q9Z?xQ^D7U-fpRz
zpn@}ulsU?8nNIfxqcc-{Ij?C{E3l9Aw^x})JPo>_6N-$5opxMT9d%2)2f40i7xeM)
zY^i!=60mSp@2me@a5E4@u9jX!NYHE4^&J{;K%bJ-o85PTqb6HtWQzgh`xD^zm!IW}
zR<~r%TAcz_FYB}ZUvg!~dcP-ceq5lrs!^jocchhzqZo05*K(Cq`|uL@td!AmI?kBW
zTDr4xT6&Rm#a#`c(8Z<WW$7|`z~Tr0$J2KPS~I@E0z#~;7b5g&eCGbR^eC;imu#de
zMtR7tbKu9G$i&1Js}b)mInztI0HsFHRdpX0|6$J<GE>?Sw6o!v?f{+}SgvAf=nG%T
zIX)$AL``v#lW-e;hFF3q3FgojCtY_Z^*+;3@2kg(_1Y+Hqj%6{4g6TNL7Fohp&`(1
zVj>i7Ch{<Rj+kwKna3t}A#189?ZoV|drHb?CkdIs#C8OAnaPj^ZPZy!4r<R+WsZI{
z(-`)|!9#5}<ZI>!)8_r=NjsiEw{6k<Klg0z)(qPf7{7dUgheZAIk`uR+zcnKba8Gy
zwF!LDT-@Gzv@iHTelZ*>+oEdm=2q#W;3hfem!%VSxRCCQ2ij4MaO)+C4qMH@VX%a$
zVJ4-<<g^h<scceKNyxCs#fuLONf(-`E{Tb$XE&OkD?{d&WV-T<F2DVD&PQ4|REfNN
zCFP8|uvhMF#TSXvJE_Jf|Gy8azr<1C+4}zo%m2|s`k;{iR=ml;wVLjq?|aWk^}iUV
z>s&$?%idOpDJ2AG)Ka|x`gtNMGY3DS)j!dK*|$!x?k8pjED6Jx!?WT`&m_3^@D?bm
z(^xq!J%F~9ok=SN=BQVzNauL<PLgu_b<8;t!lg8D;u<|i*LT{uAEch^WT~OMFesXi
zEtM<w*8Cjz6FchL83hq{wI;jdIE#1c+TX%dnvHGpfX#8T9L$zUJW<z*sHBsUk~Bs(
z9fqZsLLdxSQzF>!tX@wd1CuTC97UBzh}x)W*U>-Jo3I;u218fViJJPFF|GOUXs8oC
z6<UsfV&+=y+c_}J@LL%cR_y=uAaVt9fs9Y|+FtyIYy#V9<{kJPZ#Pf@CY+2f(NHZy
zHRmZ80}5*`42t)aa)5*NQn=^iE8^Oc3Rz(Sr?r;?Lhf+$d$sbwWW8dYJ8j%&yLTNk
z<P$3!CjtLtjn*dsU&Z3~+meQ03M_NYyM<rF%#ZE`;v0`9km6NBBhp0)XHB7W7k16q
z=c(>i#?Eftfm6LSZI`uf>kCw)b<FmBF>l5Yz4gYhT(rdTUr;ox<fgdCmB$C-->Em#
z9DlSjHV(a7^Soc#&AA^{oYTZRA8uvz&9W#Q`4M<>*tVn49VX5vf_E#k%+~BU^gJIH
zs1_#DJw#DUV^Rpi7n{iiQ&i8aXjf^)GVPQcw<uj^Rm37cUqSm9iv`N<L+B9b#p2kx
zt_3s6GM=IaD*@{+`se%?cUlnWQeUE%8(C#as$hZ{7Ip2>hX-7uJzkTLMX$R<r&Z<o
z+es314V@Zeb+V7k3d7;DG!9<L<x`d-a5*)&%v&60xW{XwA~(fPUyLSNo%G=9Se)ON
z%6ewKpq-|gn+Cs3*us{;b3G-x?{1Pf6_w(V<{X#qS;{Zf&%F}DSTlv0J4uz;`B79Y
zm7l`&x*5U(rMB}ww0=S(R|;4?Qp6KPz5Jh(F0O>tGpaXkP4TxZAopaQ!-<kV?RBv@
zb!4i0NW)E*)Dj+Yh^CWbL4NwzXPYx8Hp)l0;2gxJ)cof|vmCp!qOpjnj?P<`>AqZe
z#e<f}j=(R?Cw<kR2illD-${lPNV3U(Rce`7wFs8HJW~9|_(Lr<g&YgRoEnjWgPtL-
zq|hB|MOQH;sRxH8%&S&4DIxTgS)TS<aBWYMK<1pjS@587mZ+ULfzE^@73=;HRS?^4
zRgvd9brp;U&Jk#u*kYqoZ3n2<6J+<-CefnU9Bkx*WsJ9EKyey&_y`#*9TWMZu<1rh
z5(6xF)MpZ=S;hnv;ZPQQ7;5&*;NDdDGsl2cVOB%FK8<p}`Y_svScUGac!_v@vwe0N
z<jya{+c$FEDn|Dy5tQZJcUie?iqO1;giyXc4;hH<<m>Q%;EbVfj?1&wf<N8Vfl^vI
zqa&vjRr8Hg^!M*UA(#1Z`qkd6PGRVORw1As<(FrY4-`PI2G&+MQ*N?_2sY~1Sq)fN
z{cSzVVjGB#alhh~MM@>0#C&&{DC-fX-*%^b9aeZQzs@>{ol>yAQhq3v-oY(c6Ij=o
zslL&QUuduAZW7ULm%%&&QqO!y6a&@ZA6U?@*$r%vwCK-DqPyC}8Sv6ToR0e6CtjP2
z|3=KIz7ugR2^VvuiCy^r-~RshR>wa39}xj}tH(*Gzwk}3S=0T05^~<cdauV#*Exa@
z@VD0^HoBQE39`(lT^Oj23z&;jnIm|`d_*hhQ2{)v`O@4%rkFFrNYGB^p!ykMSX4(C
zJKFpgFsCO4+$xbp347-0EIEZ!aIBRIWS_qx|KZWP=z*)UncwJ?ul<2f_6*0vuYtOU
z6G{>_fs)c5aVesplVN`k6R_Isu%vkYbYV2fhuigdfeo!}g(i^0?obeS0Wu#s5>_E_
zm=aT|_xN4wD*mo1uG}$hS!0ywd+kMkO<d84h=rK$@)XE5`^vi`5XEJQP_!>=cq1<}
zxUU+0-vC18dgFUHO!ed;2R%U3!@-4KG2PWGY{6QgZhMw=lb9WbeO1gZj*!eW8m&!9
z3S#9#kj-kaa59gwFpzm`PO5Q$qbzE%dDcqVkUGa1aDQ!rm0>?&N|X~nC?2QM&U_ul
zEW@VB;d&8YgmYjeJH?Oa9K2$7a1-o$ych1Hk*~)9lZurQ_cjJeDB9<^Xx5ZR7y4-b
z3*u<px6GF^4Hq{VC_@sKqoyr31-QW6A5IBk{ox64Xe<^KTMot>l-q$tgrM{_wB~OU
z5*#i4emIp2{Ik>fLM5Wb{gaO(g~w@Qyyj4%O%Ml-)V-NDQ(7L3T-U<E?UP_s>M(;F
zhcJ~KWW73poEn8lilLI&=6S_0kK!=_5!3*J_LL}hGmn6|d0KpFrGJegtIZlM$Y^fC
z|9Y$!bm}ugq9oaZ-shJ}wnpt(s36AN8{xm%loavfvCLL|M%3Yf-EuwF=w`^fDc?YR
zC8=tli5XZ}>lr&BW5$BPl@BPtAz8Xhw(3o5$)%6o^SvG3eyJ7uc!yfbCu%_wR73DH
z0);^xYx?l8O)SG8*tK=S_$3mf%W8}@KHwC#v9-8Qifk3a#=r<1$~SCT-HOh@a%>lS
zGW7%<au!6}#IyF(hx^=53=D4>3ZFQbCEMfso-YuJ9y<)9vU02IJOo0&cQ<4<ne~n7
z)cfBuE-}fdeDTE7c;tgeQ&|lGs*CL3N5hSKQVU5AmqU#%%~*9K=u#h6U1f`~-X4K$
z@_24qo6>J{!@|7n`-`q>ig%dt!iuDG#FcoMJ{~5TO^4A0J~6)?)^%uq{eaWLC}@y*
zr8BX#_M)_QBeApmeB{1d?U&y7Dq<S18AwtsQ}oX@^`&iL4oK@C?tM`^E#0BB#onVJ
z97WQP3t7|}A;_N#&LD2YbIRQW@4;V!`pR#l1|kZdPK<;jE}u?i9cAcfa6f@aC|+KK
z4$>8F&H;ZOrLs6f!ST)d;(2|c*9ETLYB@p!VcbcC`J=o_J{@&{9ap=o{dst7>mZ|z
zwUMfh^ycQnpK0@TU&J?$*#u3`2)Ta_^_u)qX?w5TNczPN%@+@6x`F${;c4nEDD5=T
z>PCsdhnu95rBj|^RkCZT$^HneU21{cFgKtTiXVE>bxSd;R4yXC_36rma9f<+(wO$l
zOsZPF8<v7_@%k4O*j&z3xw#`+T)fcoa(h(=<>i|GMWfDskEO&-LcCP>8;aU7h(DK~
zPF>L9+>R#h3nNgLCGgnnkL>ZUJu_l~lz=?1$Og#yTHRaHIMLx@6?Q7v;}qhWQsv5N
z6P1?W!Il>$l^7DJnF`-Hh4Bjd#|4T#Mvt!g#7W2{H!$4h{!qn8eF`)9&Cs3l``B$T
zjODk(?^qFgyIPBHU;YIRn(A1O*dvP)|6cz7=MMN^`Nm(%%Kx~Q*%!jTwrzItUF|(6
z>An7>6Zx_3mx@fcqu}PXY7N8S-S)8DYq|rO0V@@0Aih5lzAKqC5roZGduHIea;%k)
zJq~ys^Mg*YT0r^@$Zl~b0Toh=3-d1PcI3ya^>tI;(`w7+o@Z&x1P%mR@(DMwM?xwY
zxf8t2Q&S+KU?{%vK6D=?mQp(z%`Nj08c1warbk+_Yzp)GH)^MeOK8J+($E50FWXsz
z0FrP6_QCAQo_Shol(mikJ$E7J=Jsoh^z}jxGsb2<dXgGPa{p#2B#jbpSb^9TCt6=5
zs|z~T5|^)+EEA&ov?-h>uWs{)3+qp-Qpwo7%=71@xw`-an>ktC<JJK{32u8>+p((J
zC*2fmsW)6p600~^BsgG=#hoxE*3w8BLwJH&r7=uY74&7{3yP|5{+ZsR$~cM0Cf^}7
zm*FAb-=7c*B?M!aHR*dv74j@pQUx<_4ABn+H<NbfdXWw<(@yYi1^V<A)q8F(CY}wq
z#nS4_N}fAwWSjerNp}ZY91R=cOh)^NyzzZI%>sNDyx^70l&VOby`O#MH19{a2W-Rj
zSj8*uuJ2Z>(pr|C-Dz=ToI--?Zf>IjoE_5my>LHG7dVe-Y0~?`o!1<0og|Po&Yx;x
z=33reiVQE#Op4ZN#n5fnCBWKF-?Ebe-TX`Xz}O^r`(7G37H>o~Auf<Vtpk#VzPn(x
zkfeBGt<Fs)U_l?7QgdF;fH#PBypcgt1dDm*K|QVp*{NDx5oBZFp?Slye0hYs$pZ_=
zFuV{~Df6}$m>6Rd?}X}DUlM`X9KChAJL-XrwjcI+eW4&}D`zSKj#&{beBAacZA$OY
zKUd0$PJi5;Z4GnjZ7@TkUOqYelD>y_-6OmQ@XRGLFHF*lYxruj_*{slf08gI$Nw?h
zX4Z;cA%2R1IUgPrcCWtuQxd`lvtcw<#zeH+eJNLw9a2St1iMG6T#4k};FXtCGZ+3~
zZn#L2<C2->yNbNgFVzh*CrMdtv%=9ut~DG(-v^!|4}M*(=;2U3nD*YYxBqyPmF>s)
zn2&xI;&QM==i?T2>K3;P8+kay(OJ`uyzRIq6IOY+&mb&NtIW_wDNR0--!H6^deZ!y
zB)*DhS^5&#wy(v`C-JT=%PO1qeU0pD-V|5qG0)RH%sg55v$k{Dm{t#0X-z%8ec;9S
zY;UfYrpWylG8+7k(=%9<kiBW>qiCfjQ`LgH=VYgULH;df?cIZl`)y)mIp3?@nrZ^2
z8?`KoI&}<oeUt-9SNGttVX&$3O{PR%-UeI#RxiNV8e7DlR7yIL9Yk=|!5z;kHBlN`
z`$qSmSJnF)j7l!2ZJe0q)9(fwhqYSX3Am%s+Cg2ae(35L72v@e-jVi-Ltip$FkH-M
zdBwc0%H4(Bdu#!J!M@Ap`uKgs1q<KKiPJ{j$-P&L;d2$_9gq{l>g>(CZy#+Y?)0E?
zO?ibM=7<;;9t-W|+XtujZ8z{1b9ksyD#IP4{T$36k2Nx<T<zY)&hjb0lJqI2TaB!d
zaI=XtTi*Hzeeovv#zg1ZwLCM%oj<?SY?P2{M|4!${8r3nA))XJs>F+Y-U4*hh>bYm
z3)NQXUuTql*7bLnGdlG*gk(#0+%oMaOr%sbzMP0mzk?sj<fMbL|6mlp6m@)_OERt+
z{dV3e5z4!ZcW{QX?9b_o&(*)V3XdP*PVMK`A$JIgpI6)DF0Ox*66AjxTb}5<^u{P!
zMd{4b?{S7jyr%QBq(A(Mg`e!*`2Np8&HbMo2%xb5nkGQ(605PCD~{&Ky0g}OaFPt0
zb@<2sR#<ka`W7~p3qW!D*y!(Mep%Y6juSRjjq?jN^ZAV?0AgZX6^W+ILQh)dHEJ5T
zNnR@vqu<jL&P4H~c%c9tI&&a9PQd<^yB#^e!|FP;=2^9S)K2$yU86ZkfhDzTry4(H
zP+Xd)jc0wCWb_fd+3VDMvzcn{*}kDz%Xb3r0n$M*v%7y7LJfKTYgeMzSh-Aov)w>I
zSrSe-wvRkm>{RBC+-Vw830j@=a&$=Quce!?_mY<g=2fjNh3ZAzjf9D1u;vNcDUXje
zE0~6aF>)u_PxEo()a#AXsGoEl)zExs)pBg<Kvd_ZT*`)Sx%;#f5iB2Lyj}U(0GK^X
zqz1+$8P~Rc4<#Rl2Y6G}d7OShj^ij<;2v44nq=0tj-pzhk1wa4lA@+oua6F4thoKQ
ziivPW-!kNGbo0nlWr6>0m~yk9&xQ6_r2>a3{4*5g8O4=Xau@fc!Q&{?tq4pFY9czq
zO@elUrBo*8ln&;XoJ`f5%(Z&(;unM1+8Kk2OLOs51RMns!)Glimi4aLd&(WVImx}k
zfsUchZ8EK><L$X!{q-E2WyKUdo7gECjvd7iKT*Sdffm}Ny!+n@n3gxS-=M>z3NA8z
z)L)g$e-UZ5SFIrL$Cb9=9c}gkP?SEly$!!i0b}<L4;L8*;-@TCC5NwA7>bA6sraf=
zap^*OQV3)iq3^(Ht8j*f<<q;pa(>Wl023Bm%fqW)qwGT2Qnw2<5t6FDz9tAZPM-2}
zPd4tzAi26(y8IA@Axwe^rGqz626-!8uw7m8Bz2gySrT{gMw18!mnQ&I+!tIu#^t#P
zI{PebO4;LEn0A4N3hopSV?9%tY*63@drOhEdKx#YVy1!s;dUTHck>Skr-Y5leYkNb
zacwxuU`hBx4i`r$P)fG-y_wRercxAcvoKt-Rf<_%dC>`a%)Pnz<Z<q4OG>2tob}Iv
za1XXj0A^UgTNx{67#>5~jND>s`VnDs8??yS7BuT?J0o$TP1duH%vZAni10aG9r9DM
z=@)e5aEoVGyK;xQ?jOdMA&CtsSB0#GyD~#PgcDJI_VRICr`4KhOyoS-dblQ1Nz|#2
z*yQh0<ug(~B_NRx8AoU^KUwzcQf_EXJDzgwpfp7TG{hX&+Pcrvtn@b16;Fu+!KyLw
z`*v?im<3OfYxL6#OIh-d`-U;7tU{Oz!x7~kW72N9sHh-(T!lH|owc9pnYxBEXXhL4
z0bPefWRAj^=AkY{yco+U{W$i=A|;)OSbKD0VKZpbDDA$EgYI`uBrImrVGGy>A9nM^
z6AdObvP?%~>UxVjSiWia>z?uLViywb3>>!CWE>MHpPYYMOM#OG7bBfsV;#23MoCZi
z2CR9$(Amc~-KwPh)`;&&YH}an3J%RVbb9wH%36<m%i-Z}wc7GkD;?;l*7XZ3G-(=w
zrI>FGxAXHGw{XRd!JixgF7?(#<s0Vv=Gb@d;y-B!^gun$ALDf77~e{6a?i@^W8n|@
zePn)`EUBgK>R?)=ma+l3RMlOtfZproFP-`YBdfjn+0DWT=M7LpVa+){b@egr`2>6K
zY`JiFvP3mVE2jE{6>M~j19Kty?@O^f*e$vM;Pn(R+21>ezx(?ePq%^cH7VAtS$35#
zP}w5M6NJh{tw^E}75|FyST;K&4NlHqDa-xaLdPlZCO$(Qro0;Hb2r$+=bXpGD1{!%
zY8assyhdsBBozM0{O$vD4O||4z$$pMdXJ~O@c#<fK%Wx$udfiW{r+|1*iQgM7A*Zw
zB<wZjp0ei7bLaz>v0uh8jXKjwQ=U;&@+SA%t3UOb7m%88ki<(g5O_0@4qw43y}|2o
zM*xuM-V^<?op|9K0B`@-sR#$;YtY&m%WhE6+QXbBf6w)nqkU^N=75qE9XSE_SuF_z
zP~9aVXzygauMeiQFYTh%JN&bDbHzAk#li_b>dPCU@DUZu&=<RJ_bl{M!+yVyO{3aY
zY#diU^3lyRFF?cEJZ5wBURm^9n!u2saLVQn3OYoG4=Ztzf7b^^^C+-@*#{aWfKYXA
z;Y`z7Zaj|d_D1G-fY{?Hnq!37J8~6GemP?@k7NKgA1-hD2H!2h3rkBEN2VP+ko-{E
z{2}E&^ZCH;#9fnRn7dJv@)M3C*d*0Dn6(<ipmTiU_}k_5Od;j8K!HoUKok1$_Ag76
zQ~n2fA^sNjtkb3y_zN5U@i`e9VW^Y;a<kIP9RzVG5VDJ{Qi86|$<}}6vRVV35rza>
zH5aZjcQc+=P}xaqF0_noH=ouFq4Y1?(vcd#GYc&DRj~F}OwUt>>l(#%SRJFNLg#f*
zn=k|Rt#JJO{F!vRk&U_M8igx;HYBg8Wk)`1%i)uRiu)lLHblu<Y!3*nS$Qlb7-o!O
zO2xOJS*gXsod~K}QQy)w@?8gN>7X+hA7m|)!Q&1SItVCH4+rF`K^m`F)>lQ+j2nd#
zv9H${GQC(9vR4WBhIJn;cR1$Q38g#V*|>-5#z?t-R;%z=9}d+ld9!k>=;ZVVqRr{u
zWDDDDOJYQ{4MtIp6coxOzCr3M28#MJ?mu8^GO;!gpa%yVt0|rQr5JOZIK?R{%Sjj+
z?X%cg&+(4b8JGsppVhm8)goE7`3O)tW=<uS=~~@CMI_&r;&!=J7gfftgtWz%c`Oh7
z$2M{3v5!`W>X;xzV4vd_)5nOPN-SQ@J1xcU8qo{~VN;V6oPL?m^!%FaNoNVrCQJ$G
zUh_k!w0P8)ULp%A)#s^Bu+$(uuvb?!c(1g7PY&$wD?$ns;!u`HFItC&(6P<LG_8=S
zSbV8-mo_UEVxENKcO!qU+h(SZ{iIy%ZfBI8GB3Q*T1D3$;bsUM%}gWml$1>SH;G@)
zLG54m74gk_unZg250uDxq&`j0^CCi9Xf{?s?EUBBdvYf8=4J$ECQHa`RM;m*mydQ{
z;bu#sJ6SS3Fv4`C^pIpQOk2}vNN9xEu+u77Raa`Vny@sJOtLR(x#hhxB;@i{&T|Y&
z%e0=UwV)AknM}}k@q>3ed6%Bf3K^pY6>jV+(k13b_7i69qcw^uvgI$rscHgv?4!95
zFzsu)Q|5N3A;v)mR&`8f?+4zYN{2^W#UH(yad#O_+stdyZ|MP-H$@#q{JctA=lS6a
zfeC9PVGpNq2FcT_9*T3p#&OFHZd?ne%q~M0t2%kc56<5QKL6^gVK#jY8IiYTZR&U9
z*}9jbRbSJmjpsH#i9cCF0!8PInt7|wg;4B?DF>!|dQ2B!J~9%ogmbBCS&Ho6=;j4W
zOC}_k-DnPZ0qidHe(|;@3X5@)29~}#4wczU{e*Gj0vb@<dUV4g$YkijJ&Q%FwnPzi
zyqd;tq!D`@_c;ubZ&>&AM?kLbKGpYm(NyngD~e2oj`<<@t>;Dh^P4}Mt%E=0vfpm3
z?w$B(4n4*<$V|9ShRCF}c9l2r*D260xcu1@PRiCVEWCpNx}Am=*t|2AD2`79SbwQ7
zJ;|ti&hkHXoBzG|0fN&kPzv;4WF_}6AG^l<+*<vJ&iJ2=?sxj(rT;Wx%LbJna99aO
z44bKcQaW%?8te%l7M&8Ke+i-Itz+-n<Rg3otk}36oXW811@c2WS&(q@Vork91+&WL
z|EHFN8HmBjKeYY@JxA*#`RdigWGlHT)sw6B8q3|WE!lRvEyIor0vw5+jSx%k%CwLN
zl|4U%14|d)MpOg}xWM^@qY@9D6&4<{TE$=7zFE#?zbhg71uy}1={e{eT)wfjn5B}%
zFKx@??{0c4VZPcyvfNxoHHm1Ld4lbFwSR)oS4y35dfflMm>udeE1%WOH$qo51Ls57
z2;+yG0QISVSY9t*{j(J#Lw}w4`hs%G=1)qP%3Gel^Aua_fSt-WgDh>tC+cVd?uduE
za|$4sxF8&Ik>k1JM~Zyg-_XZe=9uG7nI0QmF)K(Njs@B4^VNvss)8;nlq#!a4TpTK
zz#>d|p@B6Cb0cbELBsSyP$$es4fDKs9@?p{Orc(Md`39e(1^w`g46GZo!Eyi!TG;>
zs;W<edHQNZIsG8SbLZ(>_cX6L6mv^+a579?@EERKQ}gQM$YS$2p7@DH(Nwkc%zzz-
z+KMRm&r2U{kT=S{fH``hReMu70zh~~QaUt0VwxfuK-~L%s|HKNcORMAwJ5HBVC}gZ
zJZtU?R;LWUN;Q$?w9HfL3Lpi#I4+ocP~R7Hyze1fl|Z_|NuA`gKW(XU)Aw$ICPrxk
zrk@(r-_29Noe{j2bL(Xbqo22(yCSrlXAnUdxbPegy9{W{2>kBRHpJ$->SJBZ%)^jS
zXCDz+>1SD?bbC{Jm!f;kt%5<dCQ72yg7X}HAi;P<-0#Ie<yo%du2^hspe%OJqYFeC
zf=6j85lt0GMduePcB$KpJa<f5B00$p4Z@U*MUr1H&nPBWtF=l0Ii7QI4YGT!(eRZ$
z#XTq5|G+4GrDSmOkE7pauiESP<L;$r&zHuYAr^n!iJ}tRoLMY-<Nq~^_It+NV{`I)
zQQ=mL{D>6Jy!xSZ-T+IswZp|v`=<@GNlNi&=cFLojTX-v&#pPXbaQdd@Equ-p3+V-
zXPgK}>cOS3U<q3L*MU#x6lh<F`XqN-SeuY@8(ykiw|DeyaXN&L*^IX3?|1Fle0QEJ
zNxYL56UPI7S4r)f%<aN?QNU3n=tAL6W20<aHr;C#`_zero%B1ec>EO<imA=`PDui-
zbWKxFc+)r-mUemSnB<u9h&qh7)BIat>3ZbRma;-c>AEQH=<e#;HH9J3-O`PkY|y^>
zPKsF2WUs*ztcS=u+T>Iik4sk7yX6qZER=F9hAn(h<iPot2{YO|r?jYACL)mj^U(G>
zuL<9w>nFvjL-$!7dBkX~%;fr=`2dMY@El3-#+^6gadOcE-|jq_l)zDJeGof4-FM|b
zy}h+2!6cVT^ZRR<tt4d*$xFFMQ;_u8&g!RxOfiBi?XmDA{=D9G<B>FIeZKp%m7)F=
zUEHP8glZi|n%H@;!J2IW$ur50VakvX?cOZvIKB*mThRxP4(sX)I#D(FvV9youhlTS
zAypU(A2*VVYIJ?RQ5x(APkZ!`mp?z%s#WrBj7mV#^<8J5>q>bU?+kNX&D!j`N?3=|
zR-Y-UZqez#C}wJFe7w)8!J%BYD~PDch~6=vx~90k^Q;uRll8_BRdP%*Cp%Xgwd=F;
z>Me;kUbREv2J=7g3432hI`h)RE<R4~POf8Ze)^MQ4ZQjj|6#!d>ta7D|Nj(r=Kn%}
z|7RbyDma?Qt^2LvSi{dZMbn&({skogWevKDdVAPh7N2g`WKOf4H1K=8o>&#}kCQn2
z1u~AFBnnP=D7!9fY{q~(C!;_kf_TG`xkBMAV=vU-1FadrnxYvtm0f)#VqxoD6dM*3
z&B?V8BB%G2H9#o!P97sik^aUr*GWRl52hj>PFGNdZKg26Stcy8c61k(o027|blNam
zYvrg^QScs04b}2rqu!`3f1YhSbo0>Flx=mB$DTWdQ_1e#)rq#K1|#LC!MS+`f>n|}
zO`=+~mi)p*OE3;PF6}~Pw`_jtS2g=$#4CsdO!g+X;C4Ff<sn9&F&Jakg$ZIznMQ_P
zcmjJOU&6qO@D0rZek86bILyqJqjHL~P~Go3ZYdpWup~Esc4pDAj*Yc=mL|Sb1)st%
zKCpmA@j>yK{GgCpJ8(RZ9B<qAkqFIl&v;B?aTX41;%Pu$o>OTM=r029Dss*1k}S9D
zoR0Na@Z3dF%?E4EPa8`wN@+T9K^oT2>zq#XF1*XKv#HXCQ}a9d0Ge1Q>sy)Fh0fme
zK6<`76biph^Mwb1rlcwmvN|oe=nQ=BrVkX_$JbL4I2OV!D&sUCu%M-PB1?_6Of61%
z*-a`Oy_h~U$c=uk6?Vy=%WF^`KRN?z5a8ne(&f4QygX&1^NJ{Tp8{@B7obKml+)C4
z7BWh^csYs$Xl&#tsL(sIyfNXLLEK!8cQok|u{cN?U9bPW<1i`#V`)@MTO<xO*~HG(
z_1E)DPEUeJy#InI5sf=&D#43_ju=H&bF0Ve&2?#hvcV`H!`za-GICR}uqUl0gy%@I
zudiOPMU?F<MBF4VvBfh+@&_gLV4*D;Q}Q~{hH$DZnq>JleL5#+DssiYlpT!2K*v|u
z@GomlqlTTcTw@7I+T>Ta|KvS#T5c>&f*bxat*iSMqE4Em5ltv|NccHJldbL^?m?G+
z<7<kGU+1+%t8Y*lYhH#z;%=dMG5(SL(Ej<({wLzKp7ds&efDwCTV!>Ga<u<q;-x2I
zVY+)wO;SwdOhv&Lyg$8l#5bKoyC<m3Q^Xse6Bu;SCimM*qDiv<f)IT5xvq#-6&XsC
zn-=foKwa~lsMR?Y5fa?!BuewAhmBYOS5KS;%how@_sz|jyH8*IP%ti>WKOB+Sx6rk
z=r;#=!Hbs@xrJFZg4EEJVu$cbnk_hzec2G0+$g>j0FUiKK&NV#m^WmEeo9q~&Ik-6
za5ZNXI=kM|oMOKry5-<=2LLxP<6;qNwY4T<QH&>4rBIj^v2`b2+_qZ0(^1OWHnu_X
zLeSFdRjj0d*@rdnTUrAndrtnrDhDpIOQ#qo2P_ma`6L4qpQ60m4%b*l-kWom$1qC1
zxkKpdh+=X}V?$edOW-M=U%tKj*2dR{zSllHaX7-*Rd7B!i|;N4z4aaF6zHaN8uVLt
zyk_W_a)BhIGNs7+^7OuApVmfcBUL#g$7pa((;!4!Y0I&Vc1fCV3kRJWHKaSVqrtud
z1YQ9ec4+2URo4>%gu$g5exXtFX7L<^z7TX?=arfNy-umpDX-&otXG^|d0Ko8p=A4e
zAcS9n&A#Zd5WXfGPHX}9ZBa5|?l|)elIy5ww=?}VrilEJsAE&PW_^r{owIp$g(t({
zN1nu8MV>nvY4Erm<6?KQvyGaZ&j?~ccWV7*XW?{5cyHR3K5vK6Q)lx`dp+j=tVw%-
zPg>!>khCHIuln!(9tX*esObbl&nuel|16cOd;qMSoGLRR%As)LrUZ-}e5Iud{N|WI
z%~d@67HgEOW6NDA+!m)YETj(;l@Wr3fQCU*Lig$t0HMBb1~|PE2tW>7N{Ot`%`$OW
zth;|R(lafov2w!SrW*1SF$6San+A#${u!v)A75s?zi}-GHvK$9F=rlDe9k3l%)9P%
zaNA~zT-oQeV0zSd3S-Uw@XS1wSn#w}P3S;ovVWaol*0M_MI=%1NE(nOeX%zcrQ(^d
zS-fl!&Qrc&%=Kt<u1uKwH<ZsPBSNYD9yCX7#|LH6_ij8`aRl%C`*om^KGjc(w!{u^
zt*FsIywzZyBEN>SJqOb_z;c9Y=j)$Brl%13prsh^(K=a+uu>GCsLtL3)_~us?gnQ?
z14k*wwNf53Wl@u5RFB@oQ{j-nd)|RaWvoEW(Qw7J?WZ;HUpLp!X?S^+|9Q&@$O)S@
zS5jv2S$m|s{J-I<Ju5ER&`t!GEy%AhGRuC);5&N(n#kJeq9_)W!KYff01CG;5&dEf
zy59cM&x>~p@xou@jkcqzs#u%h5{KQGPvM@;rt%qsQE<5<-YM170}9PnnSxl%cH3pB
zX6szFdS+>P!`NLAfn{3Pq1GRkXure(UI?fbPvWW3^s}QiFW}(?h9sfCoc66{M?Bq%
z+6?v0+WL*cVdflj33-gUsGUmg!jXNGU)KGHEgt9Huw~kc-G%5JJ~1Xu%2l8o(9vIX
zN=Nf6q@~hm@S4TE?O6cnhJ+cbbVV@LF8*N>^j9Cg>j_e9Er}rself51_malJFeB|J
z=NRV51Bd0omvCE)wnqQu`y1N|Vtv<?8l7Bk1#1;kNHC5!AD^pH;LiAioANXmMxH6S
z`{W9A)Y{_qr*ioi$q=2uth^Oht877_KK2dVe{jztO(_A%hjsVzuaKmZwClsaRtb+v
zp~T{kg4m52g!78u8=8@r{q1q>uMbu-(~Y3_qy9KnG>BLhZ{GBi0NTzZ4R`G1d?l*C
z^kgXeor3-aY3@0-*egslPvl?o&!I1==|TtZ<eOZO6E-8G`>vqUXoAOfO0w^4G0hgm
zTg`npY=Y|frMBBJ4!h5C_$k+IomG5NjhI17hy&b-*8IS@P}i01lZ(C?H`F$6V3o~5
zObuK;>W3!BM@PK7TY^+hhCB$KKrg=}8~tk*l$my&YzDtzfungZ9D#iz%khe(W^3X!
zZ{wEikv9^ymj6lZR^pS;5qmg>A3r9^HLx0IcmZ>#%CxF`0&`(!Kj?Qoq->k|SvD+n
z=4mnYop}&ULjGfv{{j5O?7Q#h&u%ZpUS90*SyM?4&8Sw`l3Tf|5%Iw^-0WYF(r`ra
zqst*h_1T>hi}5PThzSQ{m&9RN(Afw#sAxn8tune3$Gy3yr`i`HUW>KzrNQgDnk#F-
zf*~32WcJ{+gv`x$E1ZiTp2>)4@DTMrE5VE|Y40<VgY0|z{6epJ9>><@s0Vj)oP*n`
z_}rYxqoW$<jfY{MeJ%>ru4)Qn=<bf2H5j#wFZ0&*f7i`j;f%T(VfjiI#C)S50h#^v
zMRZdPIy|^j<?UAf)1h!)V!VTZ=(cS|-S!`@10>Kmm{EUa)a=h$fQ-dA#WJe2ky#IE
zeakr@HTc(-&RY%!_8H)Vzp!nf6P`L7530FlXdO)F@%bHW`n`z7UfGw)>SGBFTS<Qc
z%^Lxjm8ZERURed3q+~9$XMU(x=-a}blu5OBXlzDzn_eVQ?6s2KU~wd2@4pAP{9nx}
z6L5P<!H9fW4YJ1O^mdrQz%>2*b?gY5KSKF!#GJW|Oeys6d849WY3u^&=xR&q1C9JJ
zs!dpd%GdktInh$BX9Rz%SVRT1Uc&xNb#4ardch$L`|vGOcjMZI9LBWbj3?8^2D)z4
zc_A2W>?*#ggYGfM=%H2&M|JkUj^F|zz1HGwVp#*lPBBtC$}u&WPb6G*G52Gplvwh$
zy?dNuV);=_@AhxAY1C(Cd9j?<J{1a-MkRk+pU(xXf~jWr?Gq3_KI%LY@{nKqjv71?
zhEpo)I%t+6Yc0GLK?pJdpD;>w7z)F<qFURbrRa;-V9-)oIvw;WXdgIeR+k&_V~s~e
zPw-x$vBRCqx7-Bjww(0mu}29n?E9UNK~%#!ZpRHOf~?y7dTOYAzm<XRi35|4+~cn~
z)>Q6-@Ok#aL-S{NkOp_GPn@9o9^$acQ({DaF{w$#hq*d0?7|1Q%0sB$SdW9wjPXv7
zE$g26B9&z7iEtp(Tx$!(K=VY1KOW8|SNQRg>Y{4l%;1!Ad@^@(HOR4x55^`78kR+Z
zqJU{f)lq=10f*5e11+%1)?nG|m{~Uy>+nwgwsU#c*9)I2J8Er!Cgt{+dkX=U5*EwS
zg>)KCj~YW;N%MY`&@s}*A)LJpz2J<!s(Tkn-NZ4V7npvG;vc$*yngtHg=tP+l4XU!
zF?dw+*qTJUfII%O8eRWk%w3ogisU?(PKic53hHRi_no3vv3>F&1VpoS>`|If{y0~q
z$>ppXjz9=HooNYJ!ct63lx%yta4XXOq9D0a=uUcZ;@PI>q-(~w?#KEYVS2zP)#&$b
ziNGELX-p=nG6afg@DazmgjDo|=u&-r^<FDZjWxxYU7~J&valj&(EDy~AaSfu@D1a=
z+i#+6#blmRHx;SGvVWgsb#L$Dd=dC(QO-f-&Y&|C=ISfNE68+^uaE16%5>N2<sPN{
z3wj6J%{FTYmp%10y2|yBX-s+APF?^g0sl<dxZ4wLJ)=5v!DXuu%q!{iL`$o%6>d%V
z;n=N!HGAJ$dV3aO`vgo1d^^z<4o{kuQ@aGqApU<$opo5#?fdsf3lajGG6bYyl*CX{
zM-0ZOv5f|$V}!tH3_`+@ql6I>qZ!hv-k^*gF-j4=krc26z0Ldi`2GHV&-47};U63x
zxZ~P&o#*@gdcE)0KO=?2kR6mkTuO=;*c5gk#=+S~^Tec;mRtTJC7MDf;PP74%(l#i
zB88WB-&QiE$;<-b_1ZWgxKB8<fB#I-;CCS<JIDEnfy?`Uy;ntQ@r@u&v~xyp1a4}s
zgCb(7c8h_<ANRk={mmAzKK^nbbEy8&gM7c&0#k;KbuGc9wy9CmA4m1Vsk5=P*@&)Q
z;#>SXepr=Xf`%%ZGmdU236NhBFf8*Qk?L%7=o54hE7k1eN}iC+)}<mZSZ5q{xDQ-g
zE}<0bBh>*Mc)n=`B5r%Y-e!AiEnno-tj2Lm@`pKiP+vfH?APk*8c1oS59rqxiDC+q
zQTCTEn2&*nIKff+jU&xn)ud>NZ`8iCPYD-v#R;+Z6rUXs?x^ew->Ce_ded=pkRCFn
zTv@a9?7Ct@r23PWG*`fOl~Mdf{QL`rH7f2A`!)1Ju;0uJOAd)nTu#s7t4qSrd^>kD
z04l5c|7DIqio1WX+y6yE)KAVYx+dL?*O4Chv5;sTIsK0x_60OEZ&`BkopLrs>R(W*
zykRpGg*&1e>iOlRe6wQ~un6E*@+8!CsjOe#TJP1dfq%?vSwUYPIdG%`&SnfNRv5)f
zSz^`dvFG!bJc2^Xh%nbk4GA}%y{2029D12zc&KGm^I+=uUBCoYsdQ!7TYP4VEuT<0
z+}c)|PDys|@OU!m7H_|eVbgXrUMRc%?I)qz93a0I<CmTafexM1Q|N(wSv&pflnC_+
zhEwKQ%?CV{CxbteipR+Dl3;I$R=?Ba9Gd<vAZxcE+BUxVZShzTKEu1F>24$SME<L!
z-YuHl^|zZ8)((h@Kg0^vNt<Yj|Gd`eml4A({D{FSBrf*@KTY%TQIk+zwond1SEtG+
z?@>68$BL>Cts7H<2Cy`38Y~rOW|+Y?m?~r&f*nGDvyu;F0E{2Q>fR!5*Hb(^{>fRL
z>HUn4co<ZAF4EDSE4I8O^}QR6NDn6F@+H}@i-Pjbxg>vj7AhqJHEd)nqiBKaY>*o9
z%Pe||*IY`BsX2k5EkD!*<ceiXR(T3?I1?D!ECbU@zZX;GDRkE#HfLlvn9!D242`x>
z>zUe}IcVEUZJz&qCBLhuf$!k)MI0G{e9~zYZw$NDSgod45?SBBz@9pksxxcN;-wK=
znbw9dZ6Tvyd4)str|Ie__vBJF2#_i8b(;ZiSN0SoG`6B{hTectpLYn~RT`LGUw^oi
zT9U^_@Vj+eZ1uW*QLf?CyH{)EvVeN&+t~w6VEZH4u&;IJT#&-gv)XAT!ZWEF`_3~f
ztOn*{xrElk?v)r=p_*m``1L&kX=T;a&3F2p<f1So&{^nZeRgvQ)Uf$%@>g7jG{+sd
zq`m-u?2V8mgUnKatAw|zGzU2`M}vxIH4FUx6<)>`Ajfkp1l2ex)5}A+Z_7qe;0ki3
zkr*kAy-2%7OYUxwh#1pfi4P^<!E@9idZX42CD!!xvI{mEiFB{fGwD1vOteEiw<ea!
zlz&vK?z}81adnY7U?+2Cd(|{rqjEdVN>rMhciJRbX@N7XpiwhOgOS+Cj4))mZ|vM2
zPa>63#_>)<$%e7_wU<W81V;_x?{;R|E7c6sn9GmPg72uLNF~vD$A{&L5uD+!4qsHI
zv!6ADnK82O7nZp$t<*iR*mt%#o%wO}fd~GeI*#?WB~<@sxdfp~qNdw4*D9p$WN|%f
zXGpN9cMW+S!{&|2zjmdj+vI(wbOGCWOj=-JBHg@)*mq%$O?*8$eyL%8;f?1yue<QC
zwR%RW2&p(wJl*`gtBZBz)}GG4ph)~f9@8sQO9*<F`HPP~TObit8xDEd<W%!$cpv))
z?j)yMJki0=p(x`={~^BZoJ>v+3Tl;r&Bc*wkcSMBDh}#*4C`;Fx`Ws%cEi9pHVQ&5
zOn0*~{ZVYy$x{<r-IRxCUuB-R^2(dEyHn`@&^c?XPMH;ak&q8^!R4YQo`_OwTkhNI
zLPFj>*j<0Z8_x^~ysJ{j^0@M1IQD>-J@E%VO1awd=G7#uvPYuQq>AygN@=~8re8s{
zzzh0{2kJqhLuDqOYMe)04^;%ll%)rSD=`ZJF~(EzDK4}OztEOpfcM1L09tKH{l}{L
zBr>W5IFY*lUnvJ@=KRmm_h;Yo*PqLu?|TzBAmaYL>*b*DPc7KHtRCH4gpwQK|2PV(
zavfS?UBDl>&_>KWE@sXi+h?kj-k06nP#EZTJcggkg@@Jf8&y^mXyuTUSs(uqptHdE
zKpLs=6o7>FMRWf0kDhd&`Z%oCLQrMR*GyZJD^Z_nIf6<aluSl)nEv3|d8_u+;x3?F
zrlHR(V2Tp(J#}NY^^eT-SzKHbQ&Lt^J5ba8-^wT{k^+Ruck+I2mrHgY9t=4&PG-?{
zg~K-+%~BsEw9{N%O~bcDw98Z>)PjHsZ3oxnDMYm-;60zz7S07Q+WF&a^=xd~DgjB)
zRgk7H6pbZiDDnQ<9##SmolNNdvM#bNCyrXD#d7SMe_HX7ta;n1coM<vP9YPeg{mft
z(^4GnQmnXgU<7E*DSSp40ga@_7rCSu+XIyy1d2Zp2!O$khVAR%x25Vuy_H{#%J)TH
z9#u{&<;58UtK&EOs*XDt87}f>x^?n{Hb%sp?&E7{P^`9eIo89E@Fn3c!)T(_Y0MSi
zVrXtF<2XjZ>Z*@pN17xc;q&;cTEkMYBsw-BnzKpcq{LD1mNXEVlz%k9!?MP2ZMEY|
zL@7_o%(tpKtZ^Vg;f`R(djsnzP8a}mMOj;zSk#Mpm@Qv1J|jyB2$+2EK#eYQS=HXI
zvPVYF&gBB}#A+Z_%SetFb_G*Yl2PnFm$evLc9I0zj(Wapc~41eR+;<7cF3%S_A~T_
z>{-#pP=`}d4E;Vghk8jJX!BQ1j}Y)Y>0uz!Deu*s3uu>0b-5ifW*Fo*GF&KSAW>Av
z@CYzK8?knsm-P9Tj(&U|=d8-RU&|5WRkc5+9#kVQBt`BF-^aR``W8THpB6GE-uM=>
zXL<D=sd9wue3m_1e$-md5Yo4_jLJ5Lz!#Oy+c{O2SN1hSpN^coUi)6~CGb_R?9-aH
zeJl%n15q*H4Il#G^l2RJndm{LOHsau7Uqvi*p^f-4+VLU8-XF<tYh#(K<4>tpEf(S
zWj{H}&`k3-=X6)z@r~7tUsn2{ti$O08!}&<rWv34C3)0m3lcFCH-qbnZ19wcv0*;;
zBz<<5M%#nD^Evs5voZdr-ahDS_Gb#1oxVM!ci$}&S=Nz)Ap}0@H__pZP>i*Ytw-h!
zMY}v3^B(;92wAe9zIZjdQKF}o&@Y9eLB7g(qZQe_vw)adc!&sV(U@~?BFy(j3)xuE
z+VXVp(g@32hdaX4Bb(IJo572$7E1t@Vro7ybe*Sz@Gofi7LSO~L}Jrs$FXu$zE(HU
zxf|;**qEypXlTpYn7WT;tc;~q1cb^$&z&c_#`J0I7n@n}ZF`<sTIz{*nA#KnJn>$o
zL5)#>_0L+_4~Wpz)aO51_ttF|bO{zx*AZJjSqZQK=SkEo9=k026K<H;eZ)yR`&JJK
z`Yw=&;NnU$I!QuE>Abb!VkU~wQpgUk2Cpjk>vzpD^MvUNGd8g;8kT8nZ2_#aJY+}s
z7@s~bZ^*SB+~wQd4ta^=TqH=+d;D#5zGkE3CuOVl{vIKqlfJTpEYe6>rI0Uu01poY
zUzojpw65GRaGh|)>Gy_wXJZiGoe>9i{$(+fP@*$9Q>ISK)@xN5M)NRvTHW$-=n#+J
z6+EOIhfTi#er#P^?Wwx|#LfR791#bUidg}dDXdC}@-HZwvd;T_7Qe#x`kQs+U$44g
z@4Oo&0cIg>t-rKJEN+h!HZtC0WVC)n{n-SHq3TejVDVpm!gx*Q7v_@?^8-gzgD42<
zKleZ2rTwS-t$`So#0sKLsD<xP=uF%%P1`R)(EG7H9Ssq;B7<_uLmtVztnt$0Qw+I%
z?!Cj+JKoPCaxQ0JgI=6fe5wl-XAt}aH@o^5jB%euo5IG~VU4{t<jN;K#!q%><dPN%
zYv(jIBEUEPE0n85lM}XzOMKd>QVTETg77pQ0I0-UMu2^0a9-XiiTjGRp4szZ21+pW
zy|scoh4qZsQQegughhFDlQ*Fl>zi?aP?ir!>w|c`w4KeHPF!jvhzKj==HQtoz&d(!
z8|z`Ew}5py1Fof6stUuJ?3+&CS?+JjC!4m|f~ACbZm6Hm!bsLrIMqM9rDu42x^Ky!
z%LqdL7_(pAm}$S!__ggxB5pAS%?(bB4OK#hDzf}AsWw@6)Q<`cc{lZa5iwZ{BHazu
zpoj>YX*l32dsKin2}A~hMd2%vk^!i<w{rv?WxqRmR`1^l;{WuQMr)9vO5}VlE)a>T
z86Y!tu)<a*)^JLG+Abw9Nw`;820y4|(0y6sm!hcFsraSWFy2xmY5>ej(S;@~Wy@TS
zXV=q81G`UtDyU3O)KN9QrT5*O%doj$<cJ<K?h{uKQ@~;)qHUV-wkMWuS9B+|2W1sd
zSFd4?CgXD+!Zd2|b7u)V4_cup5nll1D;&LcRhtdh=J8ot=g=+JEK)U#6Ke@OP|8{t
zJxrRD9^mXb3Rvnz0gV*O`n$*=h(`Jt@#WwN)gtw9w_M`Q)Z(xfx0Qh=$c)CtnfM>}
zNcTHKG8|_N17}P2T}p~}^2Rikx(H{915I_|x$Ys5yQ-CpaYEPW4nfb)J`yk|wdi=m
zB`*%ICtP(Q;}PlAkh?g?)2Kx?<&&;OK3R0=*|`RSYV=8y+sly+YwpC#<5``xf5wyp
zhDs(0@^LNS@k457+D_)~S-(+y7xSt}Mr62|c;ywTT6R^|ezy0G<NEurcW8Wu=q8Dt
zCQeGN`f#1A_~&afP>6+7@ie#Z)11P4emu|iZMHQZ|5bd3tq$GCwuDb?ihw4kZ7<hc
z#37zcvjPsrcT)LzjTUd6I}Y;FeAILRbS*kc$NA>>-%*Ct>F^Uiy5M=Zl{Z74nRTLO
zQz7~rA?t8NTQcr<-@hQK)3zp*7G7(WB;lsUgo=~i%^kg#uE8B3#c9zU?Vztq*PhM)
z-q$gDi>5qiQ)*l2aQhH`X!N6Bdte42Y|u5$<!n$RRXoHCOKjR`lgT|AR*-ux+!8Nk
z9!yaL&GNB-p9k>=nu~KtiW$<+ZDUxydjrGbF{~f-j_PdmL=m=P<q<@Ot?L}2->Ob}
z-PkL)ACfjfP2Eadn#S94Pv<$-!$>GkD<paN9J<f@{Ld~!K6$Ko*pSaelc$;boy%*I
z1#3lPuFO7nLbw?}MvSt&+{PMo4`v{*<T3=$?37)6_j3}hvAvQRt|+mbRa6>;w?6Aj
zwC<DEe^L8SY9K4r(pju;rAubp@z0&E-okQiFw6Kq5S{;9o&h@7|6OGOk#EW`rQ!|0
zmyy~j2@5TC|AP2REs5f`j8%x@*R~4*sh8_?Jfh~%X>bLRst6dbAs~TX>R`FFl(@|V
zZ8HmuP&Nrzdr5z)of8<0W#I!40|Jf0!bFlMc@90!OjFxY_$Td_%Tz*PuiDECGMA1g
z-IM!DTX#8sms|`yn4W}iWq1kg?<?ilr(XP!lpP{*PWz@wuijoT)qh5l4_o+!aWWj$
zxK253K$)MZkOtw0ImC{df~khGJVZ+JMt&?cDetz*6PoqTMs#z3hR-X4;$i|$Ff6en
z%RXAaWH*lwcyV(HFix;Xjc-}m^cc$pdlg?CXRRyF)F9*7$j(}(+7AQm+5>rqgm>%B
zcB_*-`K%t0kkuy|Ov3_LhYk1tX@qg3^_Z$cK*JEQWoU|)0$a67{8Dp(p!6bLT?`g*
z>NZvo!7JTiV18}a<<1!>j)51a!dmFH#rqcSdlV?Z%}++6x)Y<rnazZ}xZamuzm9fq
zbGiw8Y-xvy51$I`kjB3ybpl5j)yj7I5Ywki)e16N>=RTb?JtC>bD_7&2P$R{!guaS
zf;R{!&%oE`*{r3g8bIA!x?-tD@mVfmPu8#4>+MUjX-1DG36#e}CL)9s8<;GI_+#e>
zWx8h{=``nT7LW5pICZ>doY&ob$Tls;$>kC@EEbNK`rHp~PN#HRvH+9c1?emv?5|X*
z0-y{X(OSUCmPG`9MM3)oEn4Q4dL$-pt|x-wV)h{(dZxXkEpuy!K<~d=p7j-|#l&bd
zB9?Di7Tuo8eA(w`NnAWaJ`1SI33q83-W`zGrSoAwxFTk3CBxmWXE6uJb2IVF!$2Ij
z8<%THn2?DrqT~A)FGkede#5ECC?x*{T2ztoAHqv+h{l}34fOeuUYx1qaLUExU)mkG
z*0ef*-oUJVq`RH>7P=|0*(B`mgnRL8%9bUQhu%gjx?8XL!f#3~KAO`g*7!u}f9J1i
zb%hY*2Urj^9rnELs;W{GEOt09zWe;L9X)P6xf!wr>1mMk1?5<uu=S2uTW6tkOJmF-
z)tRzuLkae0H@v?UV#ad({slQxV4{#LRy|?ozt_&>0IErTfvOe@rwh?~H-wl-BiYuO
z@yejckIb&%U>1h5c(ZLCD?nN?9f#&<eHPvQdVx6DrqA9jYH?WLI>_-g|6h>*&2+MD
zm1?A3l*h0jvvA|MQPAZeJaoVRCyj3ABgDy8PU4#XOta2BDA?<AsLT$`LOm(4lc|xt
zBD<7!X(ag8VfDGG_ER3wE%mKgN<$N|-%>A_q|lv9g@rK#RXkn2y3L`_oAc5<$Vuok
zY?NxA%RHAwL1^~|C>&HP-uN?~$zmj3P|W}-ZZ9kmmoO|qV+s_0t<LGrgpI9qp7gMH
zveBXI1(RuJU1_?4;BRfsr&I^CljGjQ^+$yQI7B8#yO|z~ZR!WW{ES$rmf;4Uv$KjO
z*EpTpvGj*<R0yPpm}){#>>&NFx}`6x$t%dn^4P@`z*<b*)x2oC5^;>kk5h6P=m#Is
z%`00C9kQxLmzGj{Usj9L^<4Sh8MX3qeucJ98=PhtT5U-Chu6PD_TJ%HaV=}eQWg>r
zKLHDhKe6in?FRzHLpfkjHw<e7h&OB(J_8Hde?jjkhgaD&O7r^<;u@0Z;aaBYIrtwj
zRbn|kWr}v>lyZs<Or3M&xVC8n-!6QqJ_`$)vvtusu1{l+$DZ@F`tw$0L)7H`i5^;@
zye+5<kWs|b><+xQt?Cn&=aebe!t^UT0sU=WKfSK6;GiAVmcMhnn|eTeWuv#0V!j%Q
zkQ;HNE~T9tE#)=YBg2=`b5WC9l%NQwImNv?9W!&n5ZMc%VYDqu&NE1_YC3fdYB0~X
ztp4n+?cLN_Sdy8s$RZhV8Kd$ib6`yjR>2IH)H^S%C1tOb24+6-Dmf|jMh{a?v?$C}
z<X`{NWcP8U)35Q?mB5Dk5INz;{~S7ye<`^L`tl>4a9HfQB^)sYF|8P(gf51Ic?wK*
zqMA+@z=Wlm@|Y?`)&SQPINL~pcXGi8>{3)Z2u^BA5GV_hmN+!GD;I0V`k9O^qQ*%_
z!t<YHTi1k&llU2i)*zi5fli#p>HTyN5m%l3D({0)i=g#;{HSZ1(O_FOgqMouM~ENn
zR(P#yWu<hB0(Tada3IU%`nbbP5NQK~jK-WaQ#2Na-`-7Q^~+<5q?7JEXDyLAi5vJV
z`>F7Q-m)BYA5|0S##6`xiUrhmz!4V)1j2_u8jd(Q$X$+IUUbZtuI+}-L>=0`cOR!H
z*6XefRPz;XU#}c!JTyhr6h`eP<;+^My~jd*S*(D6yo<vt*?i4)WpU~XPWz6#+=UoZ
z;jgZMk#3M7I#`|tg{0aqG0R@dz3sclGW@N0AaFiS7LaK`BF`B)AXW1BAOo+4uIe3W
zh%48-ikK4@TWbX$D^2$c7PvWH)K&F6e0%05=CV@M_hk;rue9QfSzEgq#MT<w%?BqN
zH$11U*ZF4T5HZz77c1Go587mPu3Jb#m6?SVd7EtvT}h3KRwbEJbfV#GLEh>e%f~ER
z(B@O_7AHfYZT&&Q#ja(=q95&g=Jp)4&o<r<$&h8sIikdzEPvy$Yw?(W_x{U$fq-dg
zpIV+!hv*Yy2fJgG=Wp;aSNIId`?L$zmnv7^hr0|$t=(vu=Gqn8V&>Goj^vJQr1A$c
zuHR@L$k>s6q3l}9BKzHe7<B4Wfqx8>vTEe?c*$TWvsTtS<v?jR*Q;QL8+y{kP|(n(
zQ)d(f#*MTlV#1v%ZGZ$X-*zTmz$GAHN3L|Y;Cl9Ft((SW<MH!JCqQ`mgTG))8%e(c
zt7-gQs0U{*Y$@!DBd-z|*)?q~pn>+N-%=CbS1(t;M2Ct(I}vm86y-Nh_wLIGw9J?r
zf1{l}JYIyqFDNbd<T`F7yji78?WqhD5GNu=w+mj@*}0V09PKSqvVnr^$?~H^%*5aF
z0-TA8Gyj5=lH>85G$A%-)~C2Cj_=HvdX|3{hmZDApyMx8F#$A)i)$qb0SqzVpc{JF
zTwCBTR_Yvh;>)qD^cKG7C9ja7go}dox;0nh;lA(V<4wTHkMFj_Tz&y0sWojB1vTTb
zjcYcB0itTpD;74&mm3AWF9jWm17q*k8@D?5b7jKZ{P*3cw1JX>X|W~5z#H2jNxMGn
z`9MdL2@Q>H(LvHjc>$a?C-cVZ^3;HjOB7gez*ngieoO+*2RRPee|1y1GC>$vO&9~P
zi2t3?q5pI2d58e3*%Ux+YsE?)#Bhc@-E2RfIMUYhK?_>SquY?>T~2fhUb0(UtLWRu
zj7iqiB}R)eBM%~!5CSwdGDGsz*EfMkHvaC4Y)t7s*nu+%DanKO0QR<HSRg5m-PmJ$
zS#$98_)l9s$n$OhKBO%MYYbfLgJ3BN3x3?;-Q{d9lq9xF2DVpJcQ;r<z@{f&;j?O=
zcb%Vp>0R6M9KwkY0*sfpEmhcYYb8(FlZ<`HqS?>0_+d#WI4jQvj-$Qg%o&=2`~n7;
z<n0qNqD2{YB%h*ofP^)=;jzdHJ!gQ^B;?((?<}E0IrBz|#u7_>XR4un`wY0~<h$TI
zEXOrf)%y;D9V1o964^OO9RFzmAwLH$B*~?Aj5N{Z;cALas-R2V#{?^?6^&GCZBeaB
zDY6E%CQU~5fRbi-o|R&PNLvb-fb(u1KQ+iUnpHnXc24}h8genVHo3F=>3dJJ*xl9%
zJGzd<5@;vXEN~-bB~jzJF?pu*{73#ELy@LtSC&v1Hyh<I-aU{+4{yLyM`~MzagP3~
zd&8RQ0}mEP#L7LBw%fsNSL_=$J1(-q(f}e+ng)zkxs>QF##5JY`7XE2eS$Y?Shg!a
z9>{=`OyupbEC>NssvY7EsDTABuQA-Fb){eP?;<X#AQz=i8l#sZXZGM98PQ_W<t%gP
z7J~;KIFhiksL~`j+!8S_;Ks~(7|7c<a7E<9P0ul<C4oWBibI%;DgIux&6{$Eoa|WK
zSDSGzG8+9+%$*vU_h^HA%fLbYp;yRujQ!*OiA%?+3V~;rvNws1mpqT#-ar15A>(NG
zH9HC)Z4MWFCdCrHe8Vi9qpQP^BciUU>B+Qi9B=7%IYp+BbNfS3{g_pFS+s=#vt@^f
z^Au_=y^?@?@T89fLY0LDe`%|ILjJw2As*LMNgrN#)alH*BOb&b>!SKhT={g!=`hL=
z^cq#++<8d(6B&`V{?|Eyq2}FRbhM=z(m57%J2~2=SIBMOhvDukp8b7-kNA<q<w1Ys
zStr^xKS{N}9>?SP(`O{{Fz`aeiui+Nowoevqf#PS=zg)U)QXoWJ>>asWpVQN{51v#
zo;FCuFOQZ6L@wNRGA@s7Q@9${GvuWMWdw@PT&vaG8q}I41G;zjv`m3#TiAHYK(t`9
zkL)(dRx1!bV7z@M#_Pm%0ki*?@uubEbY(?)9XJ1X$c`M!u-X1ck!fbGi!(ny0h9Q!
z*8Ob4jzh3axVs10*@%l~N;ct5(YG~md(k!#rC)>kW&zI^eY#uU4+xjd3_|4_=k#Ap
zG10`=h8*zZ2<BrY??uW@b)9CdQ%Bk-#iHIK1h27g96=m5-E?_$4y`3&C?nYvhN+W8
zktxZLp(F7^kwqVXdyfCZOgpM)>!>{PR^ZHIsqNLnu;vujQb7D{c3GbhoSH<-@*nfV
zl`=z#gn9i8*&{0Z>l+D63p@AvAS&FrH2AUZMg-9aHQ-ls;k=CJjYIUBOqAxiZ+^P0
zos|^Ih`?$T;@lg|#w)}%!o;}?nnW9=iPDD_%V9D0KVYUrwnLXlANoO*;9zGmKhbW&
z(%n?wV`<^yfq6^0tp8Kev`x?ct4G4zZ>)6d+6qq9B%c1iFV#QZr9aO}fDI-DjNZ)l
zt#}VRKCK$w)I_yS{N#OHou?@VT$$pa+M^82n+>}>eo&*xIU^F+l^zw2S(v(3rIOR=
z@LR<GAqsRaUW>vhaa1@uhpOp-^C<oq=K;4q-VhCja#riq?9ylcrMYm7KH2K}7;C3?
z02g@<69mhe=)0ym0=chJpAa>!(>j(F^D1|xLX2)#V1Q?_@rv;)88FyPwy>WN6Xl+q
zi$BHYnJ!LvP!X76FNVj8;Y>3`$@e<yrR)j?&_Ng8KTZ^ktw)6^<v}JVsm-zxq6NAD
zmWMB%P(a~yCtY1H5~-3v`mEG1_>jzCPwGs7rg4dXC=w-P{7-0*vBE@?$DJ-p3?V<s
zMwBKsQJwga0g=zpLKw`$(b>4BM!$?AKt)HFs<CeZFG3~ZudGZO|0q{N01<&=AD@sF
zmr|HDQFn8&QGax6R2jv>fI?|G;SmXu4StDW2_hX``w)a*Id#@;t4KTLC51J%Q-tSP
zdw2e>{RyoDf`XcT#48DZ?_TE7aqk@)U79^)_{EDkBV}m}%o+q=L{+iF*Epr>DJG_x
z7nIrT$|o-@8>p%yeIa^0oah=VYaZ|*O#)7`Iy#%_nTQkBehO7>{I6yfA5=t)t|B=G
z7<b}j&44S&GBe^7Lv10EP^;lx;0U(#!li0R1!^Ja;QFMT?B4QSnvnuy5D@kR-R3(*
zpb08VK3~GClKky_!!t0K-ByY=$8R-Ep7oA}y6oH=Qm(KI92@6`^NH=87~@x-e*muz
zUU=BjTsx>+)f&mOv2x|nebv6Z&~3_I?r6Xl-4y?@+QNx1jS{+;@@jfylME+?m;Cgz
zZFWZ<k-^KQELKoSz&rzy5PGL2^Ac7AtJ|hcFD2I}oMCfK`Df*3Z7}WRDP{RK7yr@&
zhpf;YorAG1Q7&Ava{YM|XsG$lg;?;Iu>kxX>aqGW_MGE*ahrp`&${L5i;Cey3tUB`
zf6M8NP8c0Sni{SeB}ko!d#Ar@u>Yy0Hg;)ue)Y0axSLs5q*+@V_mm*BK6yx~wo{+`
z+Wa%wq=7JX!$X{DmWx99zo1dAIa(mA{5!q(?x@0dg}jbZo7R3OyMFYohvz6GYh`?F
zUs?JU`%-KYw;gYsnR>tza$svb_K_}(s?j6@Si9*1N?x&9strC$yOO8ixGI>asKi2@
zIGMC>e2THX*Td=JA$k2TF3SFqYi&3`(`IKg)14)LqbRG-%2B!#KDfMY0`v+ulul>$
zbTCj1*-?oxDzFR48{dMb>uiB<OZ`;h=1jn_Yi<Q>l`o&o60WT5K_RcwF0$3bGVB&9
zxVI~bEdSV5@&_h9j;p!=gonX<4`F*?Q23!`5`z10_y9anN2gANtss{MIyakg2$0*a
zitSPQgwr^ju&hk)N!)u|2Swm%>tREz7~ZUaKJj6~&R{PLM&U)=R=!Tx?=xi1m8fxe
z+x;pCh|vx53%cV&G0ESh<rcEIl)9Bgb8?_^j2Lj4TbAXMr<rk(T9Fr;1y4mS&4Mda
zLp*^6z@ulaPhG=+YuVj?UroAdq{#nlO+o)9@c+-n43NkIJqYZ@7w~zo+V8i-)$-jT
znGae#^+r(dm3!xOf7_1|Qt4`3p|+nIMM{rqzu~itb-pt{_5qtm&<0k3V8Y5FFW`m&
z*|1P&&B5H21xTMi%?eRP7VdftDu!uG@MJ%xvXS){pI+}~(;62lGHh;Z`b;0M^OUNs
zl<_7y9)0CI`NIK4^_vZs8mF0Fg|Ny<<!+gxb-{_=KWTP%4H<S);Pg20CSQ@cI*^9X
zvyn#XU2O{?t!OTnKwXIO*#b(jx;04eiU7oYvi`QYg_SXhY!i+OaNLe;{LEmJ%l<%k
zoaYLBo3&b;0WH#f7+gH<Rc|)ylFkb0@$_ATt!ZIT4Bj6iV?&#Qs~EN?{YfS&Jc=-G
z3jhXwq^+k7ED7qY(jjuv_5hg%z`GiVdI2Lf7C!j=iMi!hpIc=?#H`hm!(IZR$~9@@
z7vzFR=jznr+EViLcgq!7%pY1+1cF`|nd~z^PC(my{Cn_-6=wIdYqBEe&eLKBZA!(d
z-Z??V58i#VSvOD_$W8?wk9%e_w~SEM>L%P&c$!rGZlY*l79cCZEU6hZTFhxDbNk7D
ztrAq?DRoH7=P&xtW0Q=Ol;9AdYMmQ(l&~*TG^kY0y>jNzJ@S%prqHkLRJMn_H?cv)
z@~XGR@U?|g!?;FvIYnrdcPFTpCY`RhNzG)|Ws5p&qv{00>8i9!U)j&3nIX4+*D24O
z<nEB2^{SMkRsu~e$odE>e@|@j^th@EMfNZMr0Rm?`cy-fv6*K%QHFdL(klI`Yj;a{
ze+Wq=;&Fk^TZxM8kaHGS4wSm~*$jB?n+hgO*Nhyt;EQ$bxMWl_P9ZPDiE&P$E?Up&
zdGNqHn8gcL<oagqgS?Z(%8e!7SMnLwcg2kwUFg0D1XRp{k+(v<^(<qU*P~MTj9riV
z<nRdC=s><(ag*GF<=iqvUV6D{>3o~q!Hy*y3|Cal&Y@SNAEkTebtX-;nLF=c$>~-f
zotU{Ql#5Mj0rl?b7{8onZa+K3C~sFj&vxYIalTx}&SxAp%>_vr)P7Dje@OfieJ$n^
zI7|Ru^QLVIvzLG8_32Zzaz(I*SFiUFGCSgFwY9j=<LRo;3DE6}terg%-Bd2&RqtN2
zdEBNskd4-CXwhZ7iLj`1E)jgAl*MiRM4hws`OTvZX6Qt}IR!2I`S~07$4wA;t#Yj!
z5`vCKBxc<sJpJ&CIUafHg9J@PrnOi*U)m5E{+{dE1B|PP<DJ}n%>3q&t^-CcLO(z7
zKnss(012XW-y=>C`}!c@bi*zL<f%8!ZZLaf!$&j#X{-)wc-d&2l|<s+0)qzX0j&$$
zC8}3m&TitV3yU}@{)Yi9qu!4KWFr|MaV{=N=2m4O=r+&gbP7mo4qgS6FBvwp9MsPy
z7GJS9APFr8$RiI1Li7SAGkg3;eZ%P~TEWkQvmJq=Yd!&&!k=!~3kbga)=BFQAy<Yw
zsrnNg-m+3rG8isGIq#PC07;Bf30cpXQ+BaHQ6<*E5h<LepL$`Cf_wp`%%sOxA2ICv
zUIj6q{NyC6FiGnZuz)}8)jyXkpdJR?7lD{tTFD<n4{+r&VNzIA|2A>Dm?HT@^e}gL
zh9HtIH5<5*R|{=B9gxbl&df2>!xus&y3^VBfVe}<TMJ<RB$7a}rAhK`(LfZC$!lwI
zJ&b%j5|AF*A%F}Wd@)R)DUy}I!zzsO?npA)M`S%O=y;N_SzU2`RA-MghN619)7#6J
z6D+3AV8h8yhL6AJp}&Pz1PXE+e8y-_WJn}eD|gK5c@>S>m$r|!2!95?SdgPdgzz7K
z#e2@rdYE775%$*(AUn#JbJ<bY)kz(pPY)5?8ebu9$$^Pw))z`z!Bs%~!UlOEZ(M4m
zi9A5J%9gh6-^Qf^hZoHyv#51jLl#W#;7h`CO#Dr%xLR**jwQ=)#5RmR4aPyE1i)K}
z`NG0X;}lX6pETJZ1P5#Uv*~21@=07AsRtm0Y=}fsV3i<vO5y@;p0sfuH#wMZ;LRIE
z5ODC$xU%p1&p_QIWGRdUdvZAybUwZ)c03-D+JE+~&($y1K7lR?FWr=qKJ68l6xBQn
z`%_LhW4XB1or+hSWA;FT<{lm!s`kp+VZBOK0Fyd!c-7|(AJxyZIV}kR-<5ux)tAdI
z%*R4l)dE<f-0&nAi)0h1USk8FiQ-}kBW6uT`y_mUU*nl_2jI@oo8K*PY3RYzjT?al
zVrFqPL9#JGskF0UBO#?=M8q&Jg={Wj08p*YNEAu)6vJK9-)A}OG@iIkUhDL&shGV;
z$>)v+8Nw{wA>s^_EZ=NA6SZySY$MR)LvE2LZ31!cl|n$+z029QCLA#ubGUwV=pgjb
z7PO=;>%5$?O~fX*J?w-zc{pfR<-txX=9qDuFtMgup@tbweF*ZKqkSvWsRUtrZF88X
z)DbqYO6Pgph}Z2S-S4DQH|3%^&fbJl;arMXPSs~;390mm*2aN}-NG2Ij#<GuR*E@c
zX(m+NkZ|ED+9Tj~NAcjq#CW)4&kdEI%qrPRYfo4s0NQzAWiNMzJ(4C&L#yV?ib#6X
z2+NZ;qk?f?^MB4*N5xy`KYz^6Y3nl8(%JFa_rq>k!RsI&auzcwX;M4suqbp-P;EZW
zutlX-t2ng2kJ>%3?Y+m4gUrX%CqyP2zB#>hsl1USrScp>Q+B<j*43mO5VH|b|LvLY
zlHS|aNtsXEo@@2xJsoX&XGr9;@@7a)s0;VP-9}x=pnplUgC5CJgk#_}(OE)9-$%H5
z+bpi>QT<}g{=4uQ;>4k4xfrRoB-*Hs-T$bHJ(SDv;35<Vff`gD5VIB~p3k^N)hA93
z^#$CkEV9dmFG#0+BiRHd`u9`6&$Idiv}DAIC-P(q93AB3I6>FXrAe}KhzEc!(-dq9
zL84&*F<?Mk38|n1Pi#+r+2Ftn#&Gnz*Rrwa5LdcGr1LlOwQ_odqL+G_qFw1-ou6uD
zW(<ERL$p-N<`iDmpR_ExINyH1fZ=p?kk6TIkasa9riy#v*RR9s#dqw*K6|?^J<Uq`
zMcRIUtElNgq-ztyl>ZX=KnlW6)U~DjPXwciwfKMYX9y0?5^kgaV5(BAc+R-<dA#I*
zu<CY+(;G7yVcL*$w6nkOCVwYK79%hfOSeE=dO*4?NJ6OH=(3pKrFtG$MH<Yr3ex2R
zLERM}$p|f`VrwSw5(GSq53QweDDUJX1~}0KWea90i<!ACQ4vUPKUuKW>8;Uoi680J
zb$vZbdroz@y`7wFeuWU3TdBl@OmkBpG%1b%s7|}G^Rkqbe_Bi~Sc>i>wdCKQfUy|c
z&srlK*~kvJM@6<c3%@U(S^*tVxvs@~iMdLb{exj|f%;B>pFof5!aAT>n1+=D7+55Z
za@J>2uu3ZBVp}wn4y53M$x_yGV+#zNZm|k}y2?rAUTthDYUVK|y?{L*B2}r4NcxJj
zh&Os0dr=5zqd%e`c|a7@1iS~W(X{^(|Axch9|GiZ4p9N@W-pX8K7#!W?J~sF8A0Zo
zT5S)XoafBSiEHbc6EDZ60W9?Sm?r$!ad6UUkG_N|JN(bbgW_k+EOkbd8AM=))g{2E
zTiSh^YR4c}u$TO)F9gCtr9hxVPMw0{)M$Ny-kQqT_!L{OC1_@IgAfU3(IgcirE>nv
zu>`=!Ra|&SHJQ|sFJRt{^Zr`vlTv%~Rd4v>1sxm=2K0N!w@As`KYLJjI~+*!N#^nu
zYh)B!QRE1!t$3h?oigdoUixyD;51zvI3)6Rr8o4{xRCchiQ2r~elL2T;fWJ9TEn{q
z?C#wKd&BbgLsT>YAyESYV|JrCc6a!Ry_A5g#$3E@Nu(E_x@=CMeY?{pZ9Gc4_hJ-v
zeiHe7i>prO#pF2|`J%+g_#{8Bydb3S%c>*8KsamZumL6Q3$mDLYB&y7eLQ|4IJ#nb
zgbrW3@aD72`dncT7-_^F7QdvI+MaNu!ra9Yf!e(?`E$%I6291QL6@^-O?bjAa$leg
zMyvnpKK$GK0JN#vS!7$6_N<z>4Ath6^GQ$YT;z`SkT84o{B@eM``N0t=FpE;kHfnz
zL>6KWhbP;bPoG6Vvfbyr)kxdH?P+E38tvUKefcH(XLq$q>-ppPlkdv3w>7pXpEGp0
z{wYFCz{G|tT&Mm;26|^Gww>rO4!K?N?A@#VA9{hDam{x#$$7eUXZ#@hpGRu)Edf4R
zM{-hXlmy%Pz&BPt{$xy0fqEZOVXghINAaJ&e)xJL$6tZszC5tSBVbaFdqP}Egyh`#
zv1r2qg7<O;ihlIoVgqFJap7Vg^F{n2`8EkSW6yHCRf-q?x?Rq%eNcMkQaO?2A$}u3
z=eZS&SXv9ph|03sNWE-@Wig@}lpj^Gf&W5-qHSKn7VfzqYz3bP#d}e+@|F^rS*$Tf
zBJ``t?<ca0Fa2Uz6;0XDl?SYH=i;4g=nQ!BcH{7FYf91NW+tfl?;-<rWrjK=w&i=&
zNRIWua3~)>Ml{;@dG)T;_y-A?)NAEOITZeC4?A~G04D;>(y<gEg$l%6fj<`h3BCf!
z2q2ZJXsC$54vL8mKRj4o6-X*sQm@;@jdK=R%z~u*vju^38u}KC@xGib#`HLs>tgjQ
zkk{~s-8U>~gJdNL>=jrR3oH)lZ2r^*=Rd&+MLI|!UQq;#W#QhejUKMzJuDx36{pSZ
zfQfOi3Yc7SyUg10csKZ-%U(jpv%ypck+K2E=?kfJtKz!qw_gblT2j3TJeqZ{9yasS
zR(iC27NQ_g?cNNjLO+pxOlADCG|J7pn(Fuluzkl&nR8O65qx!TDIQlkq@G{<+|ud%
zj{R+`&s*uuHj>bD$Q<PAq*p8?JdYysdUDCD)S;xLa7PwH7oQvv86%h~Xct^Cozh?K
z&xK?$PJOs2Mv7kNnc5QLFa&Cy@^|6GmCwHPDt)JT7%ZZI_@=yWLlYG%i^J98yG+q(
z^)5&uHe1f5T%0XuiN?l9OGL)nz*X?5!9@8F(%@B5>E;K}SuO9Xs)W^$Ik>=;!iSEP
zOUX~>KLUKEqi<>@J`kJV$67#YzD`GXYTVJg!8|x+2ekZ#=T*Vbz-u`l9aYrnMDKPD
zYXz_dp8^HVDF>ft)u<On`DnIdEZU+!5#giJ;9)PG_Dnhm#ew3G%4G#eOB=Z1Do<w#
z$$?M7C2Yy-JdG`r!WRR2!*ApmRe5KL<s843a39|@<Zu{J$XXB|ojUqZ>PG671&Pha
z!Fi2=|2|X&(x!GmY)Nze=$yLzoeWj1zR**&cUeLCM|ZyIJ1Sd}5lYUe=_N-Yv&UuU
zKoqNh4xeYiKUR*=)w<)-9pu(%nJCZ(KA7i|Qh}iKW$^J$fga2qdlQg({H?-qu6meT
z3;)^TqHV8-dUjzQyuvqs5x-iD(7N;351I9zV2%`f%Yi9*CEQYdx8kqHq`AwfhH8-S
zD|fZp{{^YDk#a2XWB4fWJIzeV^X0fAVQ~&ikyN1)I~YqSQyU<apg22=cQd_}ow8-K
zBec599bw_z&b^IuJ&{-+`|Qm~$|cI7=unoJ!_U)ovD$@oGjmr3An1J4IT0hg8W&~b
z)`{kZvt9R~4)c9sPIgE&dRxk*j1O0xHAf#*x+Zku&0{RK;&0=5%%$#w;p4=`=yziC
zoc2E@aCdD71EtF~ISgBx`9B(nnZ_x>uXKjwZ*?>p7$t>fB~a-5TD@hD#J`KZ8lJTY
z`0i-5o$Wmield!XcKVyh`+<)?j2{0P0cr5wsi^PV^)nE%hO_J2+PY|r8#dooWCPzO
z(?JqrIW}RsRqU*H*=QgIBUEfXtA}XuMad+KByDydA|1q~ATkEiV&(y<e`Z#~JnMJA
zN+2tmSi!HCqjX&gf!{4)eRHkk%XiFGHrBi`Eg`db`&3=GXt4UZQ{Qs6FTKIB*msQS
zG=Vph@)aj08^)e`1XhcFxb#7jShRO&lN}cBesra7+AH2=|Gzd1V2llfIywI<`zZu6
z3Lt=91iXMy@tDMhsQDGXJ;3bY9`+Bk$b@CqZ%hfH0~wn#${DD8%cNF)7oMZB-4|Cf
zQ$w<UKC}RU?*+P{fxLPWv9(P=@`{UgvEBv=94*WhB!`?N4<87AT$7Yxt==Ok<6mFo
zPaVU6j1oHj%B-%%=nLld_>B>BetSz0@__itKPy7(1zw*po=N7ubosgU?PG9V3yk--
zC@K6$euBddOO0*a-5Rd&eHUo)Dc&sme9hznsG0X?Q7NnYHr?M($n_{ZUDlNenB~Je
ztMc(8Q4KN>D-$uIbrtv9@^MAa=aiQ}PytC-EFjs)CMr~P*!NooL;sC8v%1SP&DN{4
zUVlZEJ~aHpDcsS<{X~;QV>;Mj%rt}Gqb;P3yuE+aVi#d}$W1YAuVpjbfBG|s^$GaZ
zxM2Xx<8m&V28sWyb~W2hT)n*0TMH78u_%Xh-j24OXt9gW1R+YdNih#MQdK>TZK~oT
zMX`zmvC&UKRbtvA(zWfP^F8HayTh-4-J;}4C=XORr4G=QLcJhn7FB$2&m-eXuh@;(
zMHTlpg(fOOCLONNyszVTDh@}V?u1L!+_N%q+AOh_JjL`&$_S*+v$>R5OO+0m|Bzl2
zknccz>p}>R_2__k$qtnqAOjp-U7~vf#%mm=obyq2wJcH^_c-Tf4-FD0vKpPD{`vhq
zLPSbs(bKpmb9QTGM8PQb==#K(Df|YriS8tgAKAR?_pUHa>Q4X0XPjJ<47{wJDOeE1
zvX4#V`S=|GK5rQAx7IbA=?V6Fy0-U~26gqnM=aRlrl_&nP#w$-q*Yp<iZe`p@N{3G
zs%E)56q5C#wACR^Y%QypS@I5?71Q#?(a!xm=l#2u1vjiYdf9!jXT7s{+~^UR;A?_=
zWJejv26QPn)hnG@qts9*J2u5QVR6xcr1Mnwt?4m)>-iF{_~h+$bW#6QGQZ*}BHexF
zjgE1`{JhT*+$~xrh`5iEyi`aODCGOAji8X2Q%K~3`K?3H3U{ZBt^J3TeLP><Fn#t&
z_v>7KH6Lg4pYDAIF@KqFJFxn=2~6N3l%NxMruJIo)@fm%hCCu0uixR`5PhARZ*bk}
z+g^jQPj589046{6;n!P?=txG^En*@Z>Rdqy*!+P7<373ZTmI+2TAQuKRB7|`?mN|7
zY**6MvODQ7-s}?aQhSN)a<EW&f!(jYuBTW*A@E#kkj0Ts_~ZrZowq*vBL=4eucHpX
zAlZ*=EMFo$|9&SM(W1LiutF7T)7$Os*vWEyB3KDC$zGpl9r9omfHqy+88O^X;!Vs$
z`dfd^rNP;+<8<lj!q5iqzrZInR!LTyi7X6>Zct$fLtS7j@uny6QTZk^2mk#d9?PQd
zzj9YBd@b*U;x-Y%&qrobZ3gI)D(16ByHw?zv^*hdYFPL^m%MzGq=zUdoSi&wf|Jq@
zfJ#e*|J{Si_g_mqH!h;Bp=+002fCctJy4(c(EI!>i_Q^E?RYia`?UdfH4wm9{7+us
zzXC-dD+Szk;Dy6~nl}RIQxIS<?eXHP_NwhGjYbdj<+#eDjms`ua1_QcwjuI&VJSkl
zDlG1&k}1f;KZ_~xP;W`28ZknBkMa_%+)bi`&fC-%WW~=9ACX)p+W&*G&`vXlK-cv#
z8l_(Gq6WS4)avO|VwAXH@T6u`>Nf}bSB1yb)j@|^v)osD%9e}&9xNSuai*t&JjPQl
znxm`vs-@zcQJk#igzl^k4a&mP@{I{Qr8_c?&LRHJA&4$@fj>1IT4h{cUp~@OZ9dg=
z`DJ<mGjHbG;PP8}CsIkBTa)gr9`3b#fOUxI&xvP{f&>9K7PLx3N8tf}o?jU!b@Nz9
zO_AkeZi%;ekb?mSw6)&NIup++MMo`!J6jT+$R!V(PiDd^PIzJ~Dj2F!-uE7LCO8`S
z-XIi38M{sG!HX)=`IYS9>d4#S4z&eINlw<H^x}m+o|qnDmr*!FpY^uJ;|eR#Q4@{&
z1QcNGd#nro9vol{>!(ZSy;`X=0Se($M4S|0)-JUn7$L?<v2fmaN*?AmeA)H2Y47Z9
zf60%Cg}u8cd%YS~YFURCb}b0wPfpJKzzf-0@$4UE3xrD8PqrS+PK_PGUVPT7Bgx$f
ztEpoh;IAIW4cAM>V}%jC3h;HN$ZcTYbHZ8*4q_5uB8sx}TK687&6ICdZB`qhPM%Qm
zP!#z9+rY4;8t`}+@I*-!z4tx!ysPr@gYO48?U{Uu?ae_&)o0VBcv({}iKACJdaD+r
zI{u{oj$BEN`ozg0k&?~o7UKiUbLU*byb4QEee4U{oSQUVuCs`dBfweOqO`rukK{HH
zWrfYFeSQ9YDe*%|I60Y*fQbGV<Z^2CuM7)<7;llbX($f|jPH!fSux7Nn-?ySkvYg5
zju-&Zc2`oK=;{uIOiqVH)z1xF1Sz{j40cI}_cpF9M*RjFgRtYaw3oZEyGp~qwY8Ek
z6_a0@_DSw3J-$S2#gb-H$+y<H_iAq^r-V50QCdX{u0P_Eq?VRiHaYB)MI0W@ZYLP=
zKhkGIMZG!3eOvru{Ds<n_)+ll&;DcD32i624RwZ9Q#anPG$gdfa|!?0REFlW&>V2J
zZ~<VXgknM@ioi+4i7FO64=MJv9RvPGKI?fIZ_{uSd`Q0U>|YOGc0ZiQUC*~#2@vl{
zaoFh5N7Pq`OjXKt&UbCSTRQX|e^Ip?Oi>oxcfhWnx3d=^!`M7Dh8spNT^;K)bCf&z
zX%D<&w{J5No<(x3GxNY%P^YuaGJJZm0uZsmcmosVwd%3wrw?>PeO?5D#3!n)i7#Ph
zmZ&E;cC^n7nOk!{EZ+=&{`8~G3wMP|wSPfnaNwqxlovlQmFBQdiJvvotr^yt;sNaL
z7B&_~RX#WLm|~G^;88g?*?AD+m@1!iA{@=fCd?-O5gn@ow6})o7r~&$r<<*81A(Bb
z4Wi{*9^ZS*w?Kbal7}MqVjuWSmHP(n^~IC(m3({_KiF{9X$a+dViO{(ynNCmYAX3T
z)Md@I0blADDm&?Ux`9_6u+{?oQ_^z3gVqm7u?YuHLfJP-@60^^kat)D{<LsleE;JI
zLxcF_)B&R*8djy`4#d^t#)6vfJ@&<n6yE%>nflv)!sTh~MmUpvXOmhTD?`zbYVuG(
z$HW84^(~9PQ&1k}1mnx-Qz^O3yn=EJR~?VC^kcw}1HAu4IWN|M6OF@Hs*0}yLY1NC
z+K<%%z7euvdb<U&SJe0{-BRrRom%sLN5n;mo{Yr%GNirY4TiNGdlx9az|}GQn-4E9
zB``32eR@FO$Ka^ZFYm_PLuc*Ag~Dgb7NB1n3_H1186EKUIjJM6X4VuZ!CL>02J@a#
zlped)IR2g2x;lq%g6Pxn!QLmp2bDW#x!X6=Th7lA!*8weT#Sc`GMZkYW@r4fS(8vC
zv~uUxz=J!shS}F36RmzU87GZYq{#5Tb6HK#lHg3LE0^>@(4omSjbZ%+hi><Ztmg41
zSMWwq6<w%$mqyR$@YO;U+x0HJCc1UPO%rhgO^SBR;u4m9YrZs23LE;x6Uh;VGZ(2|
zR*Kr>E7<^}x+s`674B96+{gwK%1LobJOva_DHJ`W5ty%rUlbe(`to=c<H5xn!Qs%|
zLD~P(T=9?*1*<#Rm4jbzhUfI57q<E^GXs};dP$&Rkv8Xx3eiz=Q?2Qng*+x=96I4v
z2y1l;C=mWQKxCcim&Z+Uo7cz+x`^u+9PgH}meep;(1r&wKu&g8#VVg<VCgrEPsFht
zBBe?s<c6Mw?7#pc&ruuV_;=4LGNFNA?xlkm;D&P*+uE{zI~HVAsG@(Psahn)U$X#X
zpQI9b(G6di!gYupHz0ekk@t~pnsyp_<;U=!;g~6t4l-Ta!U1rYC3)KXPFQ|z@th%5
zEGtEGztN(dxLE)bs(#u(tK|p}!H+0BYvb3Lr`(nBfxnrp)Ls6@PPB7(K}OtGhia1+
zC=czX2^1}3dMy-Uw(@C{@yH>VZGP4C#*TNz%kw7Gj#aNh4r0$QzF=fnzM@pBw@yUa
z<G`9QqsE;clHCM^Guv(A!F=uYB0INV?+OeDbf7LT%a7H|fi6ANhG#Qtvu|KljmRt)
zY1|@tSc<<Gy{<wdJ>={#@VTZ75hD4Hg`C-3U92q^9lF-?M8W)A1#znELN@-2YyVRp
zoqB^QTbH!#P{q;&iox-5llWLErrfV}n<GmA%~Zy^YSTQ|NiIFoy)3bip*KXgQcsBb
zop1|jS6O>|DBZC?&xfN0Wu2!XvYQ$l*&<uO_!h>;@+p3vHHCsPr-3xu2~H*I&n#%1
zX>&YeX1}d?BS~9li!`Zfy^mf!^1rA1e3JS%nI0)Nb8-)y^n9S6wS{+kv5%DxB)Pvr
z&Ow4Mt`r#lAF|%UtLZ;{|KAub%}}~)l*B;V(F4hi9N?RljgU_D%>;=tkQOH3fDxkv
z=~7Tynt=f*(rtmQpWn;(=leb9d(Q7KfJ5B7pZ9fLkLx3X)z$(!JDTK8+a=h(dXGqk
zUIU;yU&-YOi(rjufKLw6u+28ENm61{ov8X2H%<jX2XyTr1ySP-YR#MmFtwr=LNWM-
z8f3w<#qu*H2B#zJIy=B91i5~S8?oXlM`|J&UUR^-%4wi?ivm<<o4V(gNsyEuv~)Y+
zFJu;@p^f&_@9KRT(;-H$y#nv{UAda4@mC|8`>T=Jga2Nl|K5xM&CJsJZ+AknF<teG
zHBEftn?BeL*=Q}*+ROdgXBCX*md#q2m$Bk`@=V!45cq=dM$R>9!6J#9Njo`_Zmksw
z4!|;%$Z8j9ZO!RRu%b_}uLPP~|0||}u|rs~gNI0FnTR-O?$gMgeYQkxrxC9po_G~G
zV$#|{z){-y$gjNZ@9*j8P(zL|w~TlBrV~@QMwhPya1e|z#c#jAGc93j_arN>Y1y}I
zEeTkvZFV3*9lqyJ)CkSVJigUhlHt!C#{p<O(qHe+p-o$`ZBPB4O05ke@Z`%=`qCT|
zO!(x^#02!wctP`+-r1_RR11}S<$e_A!sn0`H9V9l@XR^)`N%iNDM^$?U;sxyS5W)n
z3Bz<Hl2ba;jfU}9;H*&FsmMdrwCZx7GhC0W*I0>vpDOH=a&*Gu;;eQtM09C<?D?US
ziHnEl(u{NCku@cp&~+|JXXKh@MYuAXXu+wpiw5vI)D*8%vfjvt53+0Cy|HL4*Y<$D
zg|uisUhj7%nt(znSut25a|LU&sOlTGH8svuxL7H`=t+_>K`X2J?h94Z>AdIKe-c7k
zEAUHBK5bGIG1QAI2!YULhbSp3uJ&J8C}jZYLBWVLd%(~{@)a}zlF<3qzF2v#)O+xx
zR5vZEPIJWH^U^MXQH{aXXha-vXaX}S34I4-V5ZN?nAW6a(qT!y#@_51|8;s6aMnx#
z`%kP9M*g$H<Y(V|H3^IN$_oShSG2!I5_)uxiMiQ@SG<LnhNjXE9h|^QCUSIzf6nYs
zIonsw*_Cp+$E(9L#l|;Foz_akzi9FZa#5L}PZg|WI<<2`_erm>go%lyQ}f$kTy}*r
z42}vBk>;BcSeoJD%Y`-bN9MA&%!e)_EJf8>mtEt+@*X2LH3t~$iQrW=2VGy{kMmhm
zXc+r##S8X}o4$_~>}@zO#pq$5TW3Ctj@_NXis%0^ao47X`aMw8Qo3P!DWo`LHen^x
zfq-C+-s!<TE8L4RuAcW+$$gS8XqgAgNrjN^tae8J$j)ghx_jCPJC@GWp?~3|<0-tD
zRbqc43~(%L&ZxzV6BAWaWRn|<mOFepD5D?qUUHfC<tlk%qs_7pRo?XfBxIniyJ>;g
z<xoAJoS;ig&GT1Y4&PeJ8*8w3GZH!MZ2L!3(szx)nq0Y*NnEjh)8^44atAlIKs_-M
zhyG&aN)MHh*CN#0nqg$p&BI{p;nv37*##Hh04IS^fWfeZvXHPesPk3PS?y`QFqTp5
z@)F|KG>tve-?`fJTS4@R)#~mAgYPx@B#`>eeG_f~_r^|SRc8mOU~3of>9#OEzd4Br
zGg-t}GQ<Ge|6>k(T3O6|tG$3(9tu{ndtE#0yO7VH6lp7PjAW{1Vr;JTgIE$v*`gbm
zpIv{B8Vt<e*|*DJYm7M)38dSyPZ{!dRv|iA-(glva^<WRF(+Ir-6)>EqyO1aV)Mjr
zQB1BTkj_3VSySPQHibUzHoN<y97=tn)6lErOOD${BJ!FCIZp{b;46P0dH=oiP-rd(
z#8W`LxAFpZ8{lJ_FEl}LXh2xl#Z#GhaO7njt^qOTycLvt;t?aN>y=Nsy6X5S#c<F*
zA%J2;^dUytk0u`?4yTcWi(#A&h+<R}bolQ?3ixXaX9vkhkiG4vzOAFcQnu=}S2d1I
zviRcO)cV{o{(zR6-ozrA9==%}k3{y{uc~tt2>ZcDS^Yt>I`6HxACoO31x4+|2|Llu
zPO`F6W~Q30PJYrM@D#thO+HfgCcPjiK+83B8#m}(QWdQoun}8>Cp8(kWOa@=?cs9D
zcb#f;rNxP&3h(Y31eKixmhg`3J8h*I;B3xUy9E~^A{0_yn@1An9Ro|y*0r+tBt^j^
z$P5KG)ph4=9CEC%B5y^5c&EvRBPOX>_(p9tOs}%CL(~7BFvN-DyI6K%Mz6|H$erCL
zoekD!guJQWC}^H~E#Gq+UejUG!g;g3$Yso`=?!70^p}D?l9JEEEs3n9T5-)-3$)*a
zZ$jxMzD#HT8VPs_;9Rn-do@s>!(F1KDjwC;XPJ`Qz*^}R)WIhs4$uyujvqU&z|^CI
zX_Gh`2k}i(Lg^fRCYo9Skr9FA39|<0^}bt?Z3J3ZeD|QkDnMwS*+%vNSddod)M@}e
zoC6{SK8gc9Us8#Y9gNo`_LCUx_#2qjkSqm`w~Hqjv{^w3gg9ubHtW<2m64{Z&aebl
ztS1Ggr&ucea0ltun@>Xg+8BUL2HQU^w2gy;yL?4og5h1I?nS3f%H-^;=zEtj2<-`A
zF^>zT-4##<;8T-Ef~8AEV!d0R0jRMVi{wM`-*cMk!WQcOILX=pOVgm?XG{YjH><WG
z1`O7Pqv_uQW^m^kCFCB%B#CY_(sHGX72RH1<z@i3pYdo|PT9`Q?`~C&Jl(1;@#Fu1
z&YQUkJ`<)>RS=GbA#Wd?=p-@3iyRaOKQMC4@YyxkKUAdhM_QlD!L^w!wPA{SUq6Nj
zCoZc4vO~bM{CmmBtR>+I-f|{P`ks5N`t?S>g5ptj?`a|PVdswl>Lv^lTc`En*L8OC
zR~JG1R~88%lfi0d(Sl)XAOd;%5IH{)Co%Sw;hMpGqbPAdxIighGg?^K`KCEiu#^Lr
zD;+V`eQm%|{2?FG+RVqDk#bru?O`tuj)Tr_z(K@YXa`M*^3&~#o$j~7TX!7R{%taL
z$)#}W6!Y<W&n7g3jxkjrNg|+p@lyoI4DeVWeuB0ndyk<l%W2G5ogZ_$r!{n6Hwu0b
zKq2x_P+HTa0bsq50M?JjD#jlRz9ud5qO_!V@kkf(8!G%~l)*=7@sJKX)_3VgX2?<1
z;nYUA@mA!@4?!+o{pubjKEYq>`PTHWUseUoPpe8J_<a`&n!_Ry5IRRfA&Z`wxE!RL
z>Z|STSWXUkHSrjH(C>J5VD<)<GhEq-_<zhm|J9TL<Wrgd^e{a%!l~!ItIHZ^M>Oi?
zO!L@fRYu=j<J+8k&y}BNkKcqGx;;ucd|gzXu5oB#i{lp`f_r^LRv2g??F<6Bz~!f+
zu=KwdqB(gA*dhUBVg?ah@(i1DJ%uI_c8(>C-IC~fnoQi()m&|F9_q}DQ-~{x@egIB
zx3agE(_Ybw({*I^W{*C;7f{9VOnSMo+jS<vd43+ErXm|NDJ~t*Htb$AT)NYm-#vLj
zHGS)QLO>Ph9$lN?eP>GijW3NNNQS;(4u$fk2c~I+R?!$zZjLRpNKbvp*E=Zg>Y9HB
z8cTnTCnS4%wzql=OK0GjP8(}^mY<a<oX^xOaqtTf!*~->aBKYzeP2n^a(I_SIJyol
z{>VD&H!O9;aXU`aveh*%O<1sqKgmGeUviGW{;OSU<UT6ZU(YgP3p`>HUQ=W;rrpH5
zEz#`Dy0NyNb6fk^)Q_+~ur$TfrX&@<+-Ux%rZo?44P=XLv$849D6<-U?iqIYM^cLY
zQ|z7YV{2yF8y7KcRd2kfY!gnK9B(d2&)|h5AGkqp6U@lf=6|~yQgAJ)<awTClrjSk
zY^TP=m@azXsd)u=6FZQSUa9E~BnGFo1%2qSEM&$a${K=&1Y^V2<Rcj4Y5)aoqhVUy
zRA~$fp7EHvR@@nPj3i*w+qkGS#{#Pk@ft`H6>1Nmp$92&wwUIi^pBeF%c8d+pC0~`
zuuSMD2K6DHtr2gtgNV_Zuj8MF?lxHpY@s35o%|iPdX6s7esljRta`~|e@1l-(EmwG
zN>1HqmKBu1Pn%{M(@T~5P`q=<w=|U~Zh=fisycLqAnr{O1vh6}LIa>df2cZ&mHw5h
zRu_SMY^nC;ht0}V<2;w%N6ZDmsQGnDu*EF}e>kdIk?&WA<F9#cr{+z$_kLau=bPou
zcWKBWEUub4jxIMBA4N_8>$jBTgEoW93NC`CnY^L}JDmpM_PYT=f*}pLa6hy`uhU40
zr5j$_CqKplI>8I4OB7@TqNekr{De-;bG7`8)f7~)LwL;amSDE)_PxCJ!c<!MwwpvQ
z=ahC`oAb4xPzgYq@Zsn`CI3W&S!9#Km+?Gu(i}_sqGH#--7_6)R*;)p5Ynh<D~{Ey
zBG!!rOp2z^$fsoMu)1>9_<GYqyGc9Y`R;Wdg<<*Nh22tniC??~V>JZ3mzwoTJ|1zN
z?&vPxSlslJc&_(^^k3#J+KS%>BQ^)Fwd{1_M(KQCG>IhIgS4m50Um1F{I5MQ;N_#I
zXanLp8l31bPNuK+CMFU<p*j@6xgig#9R*VElPyvU1sA1hL?-(34t<*+3v{PuH<~%Z
zoxO*)qA8;s=IaRr<4G`J2TD1?v{*f4jVIF!2rm@z?<1HWwtoaZ!0>o$re?QRqT_tC
z4m>Gt-|i}tXa|zrz|cd-I3e4xqxx#rg*Zvk4F>|vc>e#N*G>IfjznKcqU)Fi+=kC2
zlA7!F1e{*(?urdapOO6%30p<m&;0Z`oA~Se@%;NnVt;yz#6h7u_~q)PD8thyS;*0H
zTQhn<=aU%2Fz+$}D+^+$=)YIgf38CdIneUWv)AtKMFAsMVgz+`wtHwNs(mT93ge@0
z=CBX4PBf4%935A+cPb5{jLvyX`bBT*qq-1159xwIQFWmul<Ufu1?P{Dxvv|q<biRc
zJrdX5vpY(cay({xcWlF*Qt2CJ(XIxAFP~B*@Dq|P(ZT{5Bj|H{#}n#zG%FP1aV#QK
zuz6_wX&vgax;+!dM^pJIRc~qhyu0jYN&Q%nEPt5Ck}`WkQ1MCCYAu^C_1M;%FnxQ@
z8?<NzK~lu6(VYwOp50ZUf#5!-gfI_>@a$u^uHnA)7X!?G%X8_p@M)`$sXzh!nLuX!
ztLw%GhTa}Oy#PGzVyuP!Qi-JMnlsl%1`7{tY@wPXNGnrR8kD#IC$r_FC-rhB!k7{o
zls^bCYhcXFzDlB7QI=~zNaAG`g`RdMo~#PXsynfI2D*D5D-FrhR~fr&8fUtXV9-Lg
zc8M>PX~~Q#T$_+%JC-Lb5~VCT0i}%t#zTFlbGyF}n$Rx`yK_4gCbe+pNOqM2p9?X8
z)3T6A8(DlNNdv)*%|b9@Q|8#RNUdi<r56io^Ox}g1ckV<4mbmaPl1q^0iptbx9g5c
z_K?A9ih5B)RA*=7)+XUyCGo9ij@{HN&3U81SI;vNCEeOHs`fe_DA{pK6cAmk{Yplc
z($zVwNGer_&Qf*BnNdF+T~=QkxWD9ZREb=!`R7dU<^lh+=aP-?jDLi;Xa<CgaHVUd
ze@@gR_c29=LOni$K1>U(g^r!N5|#sQJxpm;gB$`+1fOsu)7^aOc#AJTKj6z1pA9VX
zGT!1Hy16vDh1#OSqfFp<n%atf>EIp7^|&}|X5Oi0zA`M5=fTrd!j6W@4YiPqj#B$>
zk>7M!CiOUdodolY3$^VcQr?{JzI&4Ml&V!28~PTmgm_PQ5)w^)(9+f`?a2R3_HA8F
z_qpYY&?tDGqSYlf@Ob@>Xi!|A6oC?YD(LMyBA18w;q+Njhd<sl242a#0*{oa56K$?
z$&sF5SrwC>5oS=AYOP+kc{Xk^^Lx=auXAnU-G=Tv-8K_3r~h>6p@ZxFFgx5{=G;bU
zG_S8H?H#A^PT2^*7P|lhZ=XxRu9ggqg?A5*>o43^NEanSi7#%xo0#go5qCDKAlKSt
zSZl-P@5p_K`Enw_m_SqG*+Qj)0`h#o^3|WA<u-bF+W`BmfP0Evc?u_ZvsOn|M&0?a
zZ6C$E81vw-l^fw+Mk~wJw++<;ePnJGKKnM2R$ir#w-V!lr=bh-)ZqTe!?|VnJpU^7
z&{M<Hnz;`ZzR(7?MQq&vHV4wM1Z>nJfW+}23knjICo|}yBEKyHXc95oQ{Xlu0R+90
z0=e}$U)KyBl1w1Uw$4MF&{VfnYvF5z-t2RHGpaB;Xw)e7xu~n?X{sCYkvjIq08y+f
z9Kj>ZJ)+ud2L)~4Ra_`*=a2^<+=-_;+C#o@MXo?UXiwD8fBD2M*<zp=C0u7%rMZ?G
zo0XB{;3@t8v}*$AYgiI6f`DTY1rB7{fB|Rq^vR~?3)8<lROHtJG!s8R07X%LY2srh
zaA>>7)D$pnlDwP~rN7_<(%J3c9}{=XICnPzp;qcxAU>n6w;(o}1Thj$M%=Z8I{)`5
zlLx3PU}(Sj*qn=0!h7mJ-&r207<TX;q0G0li*v*$Kcl89wwb#J56zzqO!BJrlp2Yx
z{xg6Hd)itDavTmWh}C4Tn=O&pI;l@gce~FTRNnLZco7t4cFmsqL1eDPEGCJq`>Z*J
zVI}^S=iP5kLLd76T<x%CwfCFGFSkEji?QhHC4k6`+deoO190~ly!@%Idr&kiclm*|
zeuN?;uEgwJoZn{&X>N?xJ>zJFO^(YL>kuAMc?w*cjks`3Q8S+J%o!e$l^3KlL_0c}
ztf?7vll@z|p+32$7r3t*t9jh2V(IGjVq(g5pI58zEi`d#xTsHhRVC?P%u!RYD$gfe
zwZs2aoGp3UL}pPdd<28%@B^ADZM51oLN`!zQtl=Lp8~jBzuQ?I%;W9NBAuOMb*Xxl
z4&xL}e7zu+g@O{U1xh`s0l7+wjurJbOb}F+eNq{w?RxR=I!;_=!d>%q$8lOx$<+tF
z_;kU0(c)a4SfIPj3?u<x0IdA9;6O&_-cl28pP<qrbCS>SCmS_RhiSL0Kzdgrz}#k<
zY4=}*7mxHEVT+;4;!Ce(yeHnHqYu51@0Q?K?s*jQop4+$gw|G^{H*7vX4NyeQ~%Gt
znuo+|#vtuIzccM)<IyWRlaD^wG3G7BBCFo?F7_}PH}fItREqh5n&oD9aFRsCTZQ~%
z!b1B(XftR<Kn(K~?Vwd_iwYQcn6h<4;O!^P0snXR<7{JiZq5XeoP@LM&_n~9I(YXc
zWE<o(+?B<2&cUZNgAPB3UcyqY06D)q^{vX81U0kU>lfh1?XE6dqntQzyd6$ZFqikW
z#F0WjDWUV0gM%+}YPM&tqTt5--KVYB^;MXste?I7ByUq?CjC`iB`PTM-vI%y%Y9C4
zplb?7UmMUpxKdO%Ka;n0Q1}bzB5#W6`Cnn}@!E%GBYr6??YoS*rP*cJ<O|1L;FuF_
zvd4HlkgUYIPIP`{%IWvF)L*(Tr92L$HOg>!KWDhIRR>ghZAFc;NYrq}jqy1L-P$Pi
z_tmA*DjnlqS@PHYa50Y6pD?Srh&lcxBmNUKF)CPDHMdOkD9fd>n@fW&;M9iVuOdCC
zjq$Tq+MKbd^B>vzI={zFOC3hCfuPG&+2;7qx072I8&}Wxf7Qz_hmJ|4ZQG9DAI^#n
z$`IT2GEj3y{RuPt69ntiSbS<+`Ys4(rD-|1mHZ8z+5q$La{SzOEjj6w)`{=Ixpwfs
zdK^r^1!xSNnT_;ieBc{un4jzeF}B<8-pAzE9F1u}7P>DAs{Zu)$Exmzc~C&)jl`XX
zt42F2>pEx6oy|nHM_Qenm}~T;VK3XbV<VtfdnOK`5|?sT%iWj*TL2TUdwgp#5R5ph
zlG9%V{Rp@GHV1xDEkg&$X7XeNhptois`_V|y_)5;93axchQ{@m>L}kcL~O)WMqfF+
zcXqgiLtu~@doTb@lS2d}M(+pQZN^rFmqHMn%;6;ceglv-ul7c%%Zq9dvPv87Rj!8y
zOO4McMofG~fc5d7uGbn&0UMNJd(Qu|S^n=`>aX44Z_mXDAm;tI<+-f=K5Wh?^UO~8
z@aGdrmyLDv!SEP+29rVEXd-K}5pC7CNmdqmt(6KokT=|+1lrPksqYcbgujl+U2SKn
zS^yFy-v8a<EfM3_wL9`fVo1CFm$(8*&SF{zGym|wf@iYQne93PCjWR=P3pD|eyx=-
z4K3rCb3K!jZO2NqhJV_ij)-S$r+=KmwsK8teR2vCzW5^&JQD;fKt0|FI40a)VrvW8
z)y%}C7uoydB#-Jkxy`KGXfe3Qa=$U3UX1pn4f5Wo+6IiG&UbSONl345w%Tes-D7}x
zmCj|3+{Y;Md%J1|I5@o%7^?7g@_JTB4Y(&gqVgJ{4T$aJDrhP_8cC+WKPWUpzRujV
zkyW3s(k53FH)df5ZTku$6slC5Vde*8m%^_p$<gyDw2q(gy}A&YXIXJc41<;aTHl)Q
zZE0<E1;KI!!Af*WhTK+cUbx6G(tG=s_i}5KP{Vyx!pkM}6^~lvYb*r?1!h<Yw#XN_
zJ1sG$$HC9wY$bwqZ<4<xy{W9rao$HYpy&|aYOj-hQfosg;J)QjjzG{Xyob!lYa<5*
zExaih(-weB@C%mB7uhM>O)5NM^|BKrwPzv&i`jq(0kRTsKLMlyf{BQuW>7~*d#vvq
zn$Xv>@~CGQy$WA!zq^!;yW&Q-su=UoeA!vQ=Hw)L`75(SOc(A6`)FL~*-Wa-(&^UY
zP8q}-9lj~dihcx_W0C06fNUp|&;Yx9Qz&Jh*LNRt{)h0Q@ASJwLbRfEyXGCHf8$<1
z`f%2-t;jV?-RXF@I1g|VMBMvyzc)5!TGCxKLLH46*+>epw^v|ay0AGC6mpgCu9(B}
zB%mQ=A6|m}-~yP|T`z6iejkSs^nre<fh(+;T`_~XQZ)(Pdy0~u&9f#X5Qbz&P2V`0
z+6g>EO2;ABF{>4w%Ku|&%s~e(-^EA@VGOq`co1lF-7h=L+(y?E#<I?8^XesZGH>5M
z9@8B4Riw?ac*W$_@aDI0Ypp4mmWQ9}@Jpao?iDK1Mv%%Z2ET(Bo|7H-1W!BVsRX%d
z<WWKq6WN6W<h6b;Q^Wmwpep;>V@&!!cQ(%J72o7d!Iuxi1*9U*>MPCd`hzO(&wW&*
zmCD^Bl5{eat)8ear_u4Fw?*ow`%AuK+UnYSR#K=Fmr>&$^N(5#vb|Q!L1Y!D?r!;)
z#CP>Uo>${qceLJl!e8`JH+>bTGizBZ?9O`mUE0^Jwto4%>o7K`Cl739zt#W=-%UV8
zs3y3$;qzm;2`J{nX_s|pb~mD(^+zK(=rz+ef@MZtKZ;C0y9xZqv&>=)3qS+xvMk34
zCip4|d<QxV%qs`P?b59FIrYgF$z-HYcKcRH(=qbgytUSca7<^McODJ{u~JO;^&p9%
zlfu#7No4^7Kn#WZ3_j?~7kmevMECe~@D`IOZAaLi?ryLA7*jHdQJlZw3-7IO)ID>>
z1dc`1fh+~ZdGY_XU;Y<r`@ha^;DimtNGuV2C%VoeA4d-kt6Ix!vTwZyKT_cP9)Ii+
z`K@^Kk*sh`E#qk&oOeK$Gr`(g2SP&OTv>+p<bTkJ)(iP#boh$?@<_rz4>92KY*>)w
zZ2~Y}L4et2Je1Kp{6G<c{I_lO3imoMX~uSn(ws%5{13=H!a#v|Slr|z!YP|pXLg|i
zk%>p)Z6Ft_{ob5lhjY(_1yL`<SG$5(k0u0WF=JkgErU@breC=8>$tz9{g|(&yXL#6
zC`uO0PtBiO(tJjx3_JA^lbY3%aI{6GFLUd^CUt7mnSUfH2o?O^T}}baP&+{`7Fn$$
z(Y*nc_vS=zn;j^~ws*_A9p*0Qhbr=#LcvBRx98%hg6E6ycQk>DrAt3+8cbD3bCIpf
z_n^?EY=ukKL_vfjX<nKM@8*Ir?hJcUh-LYdtGzm%Zy7y{!JmG=KyEIkS?@%RR^M;)
zBDl2X+O5IY%)%qoHE^3<R#_Ff3DkO&^}X||6(VM`><(INPRHt7uUDzk$8w|+#Rfih
zftM@e(nQtw=$nceWq~F(lKC(O9pK7JaZ9HS(nw8d$$*zUuffqp+g}vG4KrwK!9eS$
zKz20@7l%V=Ya(ls*cJkk1pj&n5V-8>=&g|syD4n?Q+u~Z@v@DQQ2-MrUo%3RMVQr$
zCte9gmy-azB>@tF#5($pMEVp+w)-vgYDx7^-Aj;s%)CoU85YB~D}akr-o9t(I>ER4
zYS*<x_^Vb^8XAu&A{Bd24DsD3taA7Hs`W^q0p?uieAUlwMmy<c)J=|FOPLr?cjgz#
zgf_n3i)#Xn3z`wK%-A6{Cm$8P5EWP+IB*3Cxenjv2UrHP&=RPc#3j~Cb84KDB}A03
z$rAM)wF0gO??;Qnx+x&%s=%QG1&p$KduRPxgW>KKOT$kPw}X;KKjK7t#FK*wKJy^O
z^v{LA>S;d-j_>PX@AEFqiZ7)+8HFrq-n3p?zSCZ$;{p&2e3@G6l-WNO{6I^0nY={#
zbNwzfP-t2%uT^R|tGsAGK-_It%&Dvha?{=X)WkHCK21q$tvccnU9!x3&3W`9bQN8+
z*5s<XNoo$-GdnuD{>s{)sFgtK<xpg*GQ3^dLoN{?;zRAcn4+)SDHSQ1CRQKA_8f((
zzT5GsKgSae^hMIXlNSWSjG%Bwf1k7Vp?f+Fdids;#1A*-QO4)#8a!4NU{2dv+*z;6
z63S=T3v>b|st2cmEK00g+co!g`S5A~IQ(lHr{`KX1JlNQqs3OO%)_*20CjJ<H+R@5
z`a+qg9$`E@d3>X-uA_VIL6*osJ;<<NaT+8e3><}offv}=J1Y(rD4WSNq=}BVCU8cF
zY<!KR@%I$<H~f(V<k-U!94Q~CB~s&ds|8NUrprg?*dC1bDq3Xl?)dYI;QZj1S@ioH
z@M4&O>y+XC$2w<J-;2tn=kVR&Gb($5T!mQK1v=SBW_uR9POo{0$So((G82&vm4Aq2
z$q+JUGBWxxr<SejLrE&kQ9X8H3P=fou~noQYh1ZXEiOPKl>d)__kXA-f43z7y>5x%
z@`%z*yO|wSpTVc4bJ*;fgp-DVM1%7F<rR)Gty*bP(W7Y~s_0`W6F!kE159<Bhp0iI
zl5gEHi*bQnU>8bgDPyv-ePZQfAno7=1-=SkSRhzCgl9=f@l7At8#K+5l5a=4P4n@r
zHO-Z;6_g{Xt=;aeM{20l=Lh}h22)AcO7JxioOTFp>@$ZQ#`)YU7GD00tX_`PR$%}(
z@y1&CqXXK@+;g2_ZD5~dQa|%uj`XaK`)#M6=2f~|`zHwtiTJ9U0U1HE;a_8wz1^S4
zL^?8z7i?|agx$k-q*~yR*0)gt7idlcvU@{O6fOBg*4=;D@E=~T6ADumOWVgYy{UVz
zs(mKZtfhcd1qNfr0KhM^wKYB1E<uYYEUdzy1d~=HD1qM|ZH@RsX4LEzB|EBPB*bFj
z--}Bfj1v)Q3Th6)=+Ou>r;~uC;#BFOb+jPBJFt0uPO`s9;nrS7)VnTn)6E9t-2k=I
zIbBx#QA}xe`=)z$nZ`d?O3B<hKb)VI(xR2YH9)101X<8bGH`{@Ur4foBxz(N1U;{B
zpk*;mW<F;xi04=%M0ALYrfG_kxefqhr^h@A6eQ?RCq*(6_RDvTRl1hbI@<v<&q#Kb
zG!CT4V%*Q`ND&rYLBM>1n5r!_Oq|z4bDv*DHHn*yC_^;P+&eltW99hyhHP}o&13aj
zn@e>MB7Ck=xi%l*mgZ%%70HioF7xxvD?3be>0(d6h4P-@T`%-rSwRXKmNfdjx-sCe
z{LP1E65P1CW7x6KUg6T@x3)Gj<1U@PU|rhU;9%8V%atedMxII2Jmvzt5`XRrB4tGR
zJeZ}fY+tRi`@xRX!o8WHLv+LuvJGPU8BA6J-_h5C5uT>jrq*0BKB;Cb(GpqChYm|%
zuvSF`>x@b?4fydZM)BP@CzuJG)@}V9XAiI=SBS|=X0~G~#NCCH<NIn{&lc|Saz3UA
z#`fqpfg<1BlP1s$z(b#daHeK(q)@du-hd%6-6vk{D;<Sv1n36|+z?Izy6nYGMBnJ_
zlp0DHq@Ui;U@i10k86JV+=>)f_?Eb2C0O1p##LdYb=3xXw|}(OQy_7EB0?_v$RfM~
z;d1i%(Dj9&S1M}2EV%HlEP5wa;wLA@@G0!2i-<^1R=2*>@Lj@aW$S@g&{Y+shPb0-
zW}U_zrWn_J7h5e0E%iRCiqj*$TOVpgYudFlNx0B}oa{8hZnPUciI`?^fc2zZ3*LNw
z$IMBz;N!|)86?-uCHE+@{I`OFGMLY$L+@i_s#f;zIxwtz8YCRElV3UqCYA^8pOG16
z1t_DD1bR96H3AolIjjO8AF%_`4tcN<m2Y4V{A$Y~FVUXi*l^jMSUhIiQAcu|&6-Y9
z@W}I*{M}qW7ttd}TdEWjgLnTA2=Dl2Q?$rNesa?87c1rbCB>)6v5%D--B0}ct{%}i
zH6cn>*kfyYHaiU61#J;5JF7c(=(PGkg9I`jjIEXO+Hx{AZ^}PO6WDiFq$T&GHGEJ@
z|1V^-2H=kXZD61UpZ~_w>di%0m(`u#I&L)JlooQ&K4T5{X#2ZG$)I1~GdR==6hAh{
zs)0oIgiq^yxOU&%Gb~i1h?!V^{Z~*$Ra`9L?_@#&Nfux-p+J@R3_??UaNDpEkZ83X
z(>5*A$R)Baw0HA7BYvvYB*erhljd{AnQMJI#8b^yR}y_vg8gsG(+S;~+=}L#&iHo&
zHwF+nD;&)9T5i<nN7tXvE9};O@0xLsjZ-Sr|Ikir5^2C(>(%nx=zl=-QrDlhNNlV_
z=H{eA-@KguEGH3P|3chSF_z5s>ahnNneCWdo?0U3TV*TPiPAZ1vashBPd@foz!aW$
zhVOl&;9Q~AM$qDBjod3J*Y-ukXf7%19;wK_va~%lZDA7*m3^&-tf2T*;f(!VgsLGO
zYDdp3dn`n0?|m@ci@tz?-Z#0tJ@Q;ZLp;PnTRv=nxq-WO62M)jT8cQw{ieD*z>V6g
zPDC0o%h%OaBQd-_FC1(DH#dxhDjy*%*v1=fdGEZ{!x6n$P1)PmYuW<vY-_+>eMGjr
zWJFvBO(+ZmpRzLQX+ulMjDp6-#C!y?Z=!!pfs@Zs8DxMiSNL7l8lk;`r-VFXHUqKd
zLm^zvI00S)-L{2?U?4vQu8-0SK?Mp7e=Gks9i}}0*y-oXMg)_!OSJrJ=FbgQpG6rN
zbT|aU%N0$e9xa|&7p1#?&c?ZMlC*MjJ(|j*HY43_7`5>R)7U!~-KuI3_KQ{Hn#iK(
zR;!NGiF)(kd}#Lw9MMw^9cC=H0j0=kvXw8f1&(2@-e_046MZU2$Mf_aB$R|QHPHnY
zUK8z5cwT;HNf@pestqMOLpAjc9JGW*%h};3#`6Tfbm#k>)s^|YA!=1jb<{x0IBChf
zrB1rl`42aG5-U*Xx3GDAoeyIpE?NT3?Ddabpoj=vzJquTpb#0c720yGwm;sy<27`v
zCwvCquH~Zy&!;Tfp#*9Qb(8HgzX%IHb~GC`_&zI8w|UzWv!fR^abj1kJtE6~ALCpi
z(lL%ELf-U!)31ByFlMzR>G4*9z}THGTABN>X;DnI*4My|jg-|V9Ljg6Zp~t6iz_l8
zAw5a=N2xS7e;>`Vlm?YmhpEtgsg*hP-Bx0_;-XS!!4#$aJtA_%M5edA5Se;%;u05@
zfIcr<_=P+%&FNho>ubfwW!5zp=@Ig7EfFU2lJ@Sg?Ccf6p1=_{2*Kk8c8imd+@CTb
zFem`E^aN)c(PtNQ%&2|=N}0vaDAOQ%R4kePSKZhE2vRKqY)UfB#$PG*z<(h4zZi!+
zS=b&@m{D63c^Bh#9ed%+vMR57-dGD&?xO~mS--^1>1)qp9Gos{Uio!gaOYGQbu$_D
z2GYiTge%N>9&YmQ^(j=BNY>RfS6ax&UD-3y0(kMX;&mR=Fj-W|8Q+aJ7!P>~z*bz8
z3RhE|%ypXhw4ykzhzCo5N+j9ex@x1gYIHj+ZUn&S<z)Vdy!D|ufiVBjJhPvhGhDtY
zMH%V=Le2e0i2L7c9oP~9;Cbm#_ROibQd{1Ku{77y({v8!+(y-g>IO$ImqpCR$RFte
z#bbnN5@_hyXf#a0^$3z|nY)h?p9XO(DFu3L;-_V8B<{$@|5#!^%{BA0|J#ED%wYb4
z6nn;7RRL|INMXV=qGgcwtZgB2sqA@XPZu_f5xj@{;FaLq;N!adbMp;plPQeM$hS5E
z{DP&HRzybp9XMofzG4)(aCg8lnD<g<`A(~lZi%>4GHd^DMU?}ybnk3-ce=9e!R}5|
zhpdil^H=z<b=)toL0kLN5(m!+Bj+;w=x+k%1r3!)(%N<4Gjb6w@WsZ=wn_eW&$EBC
zVtUx>*SC@5QUli<c*-c|hcjjhL}&B;JmAKJ7pU0)+Jliaf(eY7PpP8)9g?EWZ8yzu
z4vB@Dk5)z!jD16LkViKaURPF|cG2qx8<-hLWxV9Xq=lFKC~TfUj)6Yf*Czq|e~Pra
zX`P5d4#g!8;PJ2ULW_@Z%_f>sHwfYNTm_4$Nd%{+QV5|4<yW!4LcI{CY@-Yq45DS#
z*3mSe-RL|83i`D$2mG3taamhibeYRZ^2gKq7op&Q*E}cM&9|3VEdX;1`eiY}7?vla
zmIejeh7(d#Q6g@=1M9#oNv<^9*d>NVPqF+ayaszE|1<A2JYYc4&Gpm$z8=Xx9o4n_
zpFC9Mv>SH;P~HABlCy52yAjme!m3ADyanW<d@^gU=Ad=l6mdZck2!aq0c9SME0m<5
zd5Fkxt~DQFwvM|&x`i9}!CzS@kpdcUtOI=w75ei*VK$}&<ap%c{gpRW4{A42IDIxS
zLEXX%q2uq3ss?hJX6emlZf-++fxM=-tvpXBTGmg`^6J0HR2s?X8^Ff{Z$&N;owy`0
z#_f%I83b#`?04-6C@53Md6Cg#$1gQPQK!{D1+N&E9-WJo=77G?*&1DIu(Em~Th-gc
zH<G9OK_O>>?|$KQP*8DA=)R~(d)~|b9GK3DqMmhorgiy`^@nRRzI-s8`>jJ5I5uo#
zLxv-A;o<$VtI~8{%u=a*A+LC$2JEzqf#RKtIF;)B_hK_u#SQ2GZB@C=bb&pe=H%A?
zyq!rW{LpqI#DI+&;txc$gm<f5+e=el|FSQqz_C!v3xLn3<Yw(-{qY%fTSa5*t%2T-
z{NSW@=dzj_K$L(Fc6#W3h$O8ow6C|Y@JFmJ_$zBF9u)m?L?QJcSO2{@t)a;u9WXQH
zWSii?+Xa0x**mKrw%f#1!H5^Q6n<227n3X=?y6Zgd?G&!1<K;F!gH$nn<Nv!!mxN~
zX$>h4q|n7w!G~r4-gExDWkZu-&xKkhsoJOHbIX{Jk!NA|U1-b4X=#^)pBMdBkdbXS
zllHV>SAHJKggsuq_mk(1T`>c7Cuic_;_~XPV@KMb2>BIhGrPLlZcfu|TVF4`iW*BU
z0nQ6h=H_Ls{AMP`hR1mye$BOx=n-+xqe0Jj%v`fKr)>LwpuDtOrS`3rp23+R!atl+
z!knpK+=9)chQ#He=3g3Lk$?h72c&;7fA8JE>k{~u|GKC^jCsY(fn;X6pEa!)%_9#z
z9a+jp8P-{M4HQ8AtUKjLQY|M113#S2)AdldglVcEn1x-nEFJjDH^<u$f)%4UV={54
z_%0H9Fix(_`OoY*T$#j0s7UmC1QSQ)^$N<Hvi%D?V82R;NaI9{Fg|BnZ??e2#L6w#
zISckUc<fYMx<_q&-`PfObx_-xS@vCB47T`zNqnKL(PYiQ(7@!W#@%u2EJ!(N*o|4p
zM$2r1K1Q=RyUUI)QI$!YZCPoL=NRBzcD3+-eR7!Pa=ehs79=dVHg~q%F8StA86x^*
zt)ihq#3MA%TvUsrwnAk872g#00-5z+E9(kuIFcU7Zs!Ju!qi4gl(6*n@`4xi7w<^U
zeZV$i*|KIBoJiKn=eyN{J>Gp>U0Qn1ni0zEne{q9ax|7>2_mH5q|Wmj_xP#qa_%Cc
zg>Tw?ey2`QBCeh(lS$l(YhJm}3zV=#j&uZODXdY=gkDBF90=K!>|1f*uF%C($i8@h
zaV{uLSN<0UP}&aqArFG~jDTFa@A-UOZp(JdfP*S<rq&;H7wAK<%YBX-tcwun+zJ=L
zSY8?n9VPq2evoeGA}vDdEt@Y6-+L4F$ExexL{r`9C-sMlh~4pOCTXx|<L9ihi?$lw
z*HED5n&l-Qdk~0qt3h#ctit$w)^iUIP?L@Bc-MPj=fKyt<F8CvRT&h^Dm!#y{aLD?
zucO}?<Jd`$lp5pewnA!3SW+VVmI{NHB#;H%1mh)hzu=d3R}l*CFiA?=l8&^}u1JH{
zG)mIwwq?D>>c#Tb-8}4h3=dTVJJg0~gh8Kj;y@5CoFnjv1qbLoK%-gk?-86li^J_K
zl}>pSH(1RT(@YubhottYYVPlqbx9)<&?IXI0#n!sf3=t5+wf}_f7bNy-4N_eO!>p}
zk4Cj^ichDxoI_0#d>;S0C#XX-m|!84QBAuZpgY9fJ-D^*f(*_*=Cev3fhmLagpUuE
zJDwRcL8@g6(iiD|u`S;yvk}^hqC<LsD5&r9hq;7#Wj4=w)S+#*>4bfxD<(CGd5y!V
zu7iN;nYEGmQeyZ@3J{*mf~rq1Fq5^#<;Bo6pk%tDgXX_f5Rp7Mqp0{(;h%Q;6doG4
zZ=%gYK*oz1nTC2|3Q>NKLHLD`v-;T=mKy7^?uT$4<=jKe!hy%z?i_pFPeGB~gM_Pz
zk=AB5#v=0>O!UsKGeH6;s_%_-4(-Kfut11g&GKzN0yrL(OpJg;*nvk$OpQ1YmxaZV
zjU50|h&{_53=BWNF!jw$b9oPAPmtvee*$=+p6Q-T<QLxqH<$$SL--iLL+eCBZ6t0y
zc4#c3y_nDG*HltW_W`@XKUvLHhN&r#B;$X3u~#kZdM`b!sFA#AxHR*Msc3AO^2y(k
zkwyKoE<fK@3+cvI&boS(wmChUeX7AVXNUK;=sqHWUc2HLHL@3J1J=LB6BgIBrB{VM
zQ0!uFGeu$)8v#K0IWrJH5J>rZl%~sB{O_#>KqFyzw%_bNSI;z;5G6;EUthgZ<-{Zj
zvk`#dbH)4F!-0#m<=}a+ST&PQoiAP7h`|4V&KyCtxp*4`-38E}f&{=fYx#FGs<`4>
z1O;Ad05y&u0Hz?#&DlW#Z-zR$>Zo}#Bv2ylWplFm-nY;~X`2bgB;N<w9g{%<x+PKz
zZPXUi3f0xL*A=No!u5X&@#fQCC_JjUElTEb$F33Mem+2Zbj=C<E$i$NgdUZcXHi&)
zglAl8&&=uUbLxRGHmr`{)mV?ft<16MJvpNYJ}i%;nDD=Xm()nvA(uoa?OA-e^@%U8
z(Npz}k;UqyQA%_VB}Rs^Tv5yZq@aWfMpQuwY$%_66E3`EyH9SQ)*K`3{2k~EgMB#M
ziDX~ZbZy$u4igTPO5bNlmaXA!`B`<^xYeN?toG2&PMnszrj(idrnVrKuK+-WM?>d^
z2gyzLY<_$~7UfH(?TGmT9X<uY_J9cndVA?h!V9!VbCsnERz`0Kj5_gD=i9k5`at9j
z_B4mH8gS0pm?_<NPJ-#EL}2>@!9%$1QoU8baKwBMKTtWBmm;BIM#n`{FSQ>fs6QZ8
zPBTa;Vt^X*Mx*a{>AEfbu>M%#?cz^C4>^StOXEsBDQ2alg$1QQ^V$Jo3d<cLRgr>L
zpO8n&A<LQ|kivw7bN5{v#S2k~XpIcv8y$1Qct?<S13|6s2M-UZ{z8`|9Qe_j^1XWC
zEz#fYqvO={a`QD4pPGwGLtCp9=zMFm&;u1S`*`0Goj@DS7`IOPhnm4W@<tLX`{y?9
z`%zp-Vf_L1`qjOHsnb$C{f`)gJ4_#O*e1}5a-DzL6P9^K9{bL@D1LdB(xkRs{KqLQ
zYwEyi*4X`Kqy~LHVJ{?f!AZy+UT@TNyYzcz{orCNDaCc6eU!h3j?S6lX4eGLw?))8
z%?^O9x(v^BJs?&?K2ZgtXZjI~))|t~`S1^CmAU!GfNohEfnl3R0D4RVHQW~%V%-B{
z0u8m@cIzg|pVV=yJdFCJB>|C;WX_{^cdfQVSK}9M+G3R__eQaweti?|X@0%bg?YGe
zU#{Z$A&#0bW#Qno&a!<rBHgZFhjNej6EJ6|*(Gc%iDIez!Yj@5Yse#tDyn%a$tl!)
z1a`W&Ef+z@*%A4pHnqy-b5mN;;j9R|$vnVbbe-u4x>HuwEq19R>W5Z2|MzBfY-@0k
zK?{1Yu7r8oYEHEF(#HN5w|J{~gP>&bd{1~M^b69Yl+`}Shhc2fmTcnZ`LR&4wc0)_
zyRdX_eAFOt^Uy8bA-g7Lvl;^IRK;Wv5fzR|ZW5{EM~Pz7t*-t}_;cAa+JI~r`etRv
zUr4_&sfqh{D!q@`iF30Q`jTL3-@E}Dxhmfm!=Zri(+Ih)f=+%(TPd--3S$czM~v@O
zTbew43<F#Jc&nXvxp1u7nea~b_taa(hXOTvtVce20UoshBGixL&}(`*e+qcaZu)fD
zpU2}3muBz({H+qw<5ShTj}VNxCDGL_?oq>8*HpaOIzSATk2J|NQG8wr(@9@<3Rpam
z2#tr*`2ssTYeLtw4Rayq`^%WM_K@X+NYCPsiCks*=BH-W{~gEw;&lIp@5=U*Q1&3e
zf+Jr;uMySgK7Tu?9<KrPMEv<6r%AWVubJ|qf4KPF5P$X8#*#p$o~pWCyxXJXibEtA
z2u|pnZQRoqfq@qDuy%CC?Bobst{jjCW(-t@CjDL682|P>^CGydAr*}GajDvL<sAkV
zbBSc{X%~O_S7a7ETmNT|(b@~YtXF;O11f9k>Pm>W%o*!K*9MP0*2j(myx%T28{ynq
z8YU{(CC{2N&FV%UJ$x>{dFXQH&hKWg&e1)i7&m5P`M5^8taNIjnE5+DQLrIOmJmFC
z$8O)P!;h^>{a#M_w|aj!*N?^;+&y1v&|pQh<x)*+0b3EV+&MKLQI<XV`am&#I9JPv
z&oHRSYHCR>(w33pi?1eoVPVh|3cLiVVU9hYGHnDehTh)sq&1Z@xO8Rg<X4Pho4AGA
zs#_PYkl6rKL|jz6)6d&&A)%f}>YgeZzv$}hYLxtr_$f;{1!o$o?QYJLjIV5q=HsM-
zD6+9nPLm#9NEWADZO7*ziL}R_nR}VZs#zA?fqqc1;YvZF_|Nzh=1^jG4uIQ9ABE6H
zyDoAI0Z*xvR*G5@x^qr#vZFe%;)1~*j|x+58W|E-{YD-P6h!wwBr!VXY|+%mrHFZ0
zwNkCFf4|%xwf>bJ{%zSjBH`oJEbUJtQd4#BB|y(X20_j9A^qJhQuz~gji;?OB8j8R
zF|^JcP1@Td!j1u?vWsY5;TWdyQoA8z_7autGsD%6se>2lHj+c!dZil$CDk&tJ$R->
z5U$~uMH5bb#KR%W(lgg~#gVl8MQF{{x8Uw~+SEecqu^Bn(3qi!cPvT`IdCrQq;Xyv
zF!3J|xEDKbCh=JsBHD2;_Th+NF0=Zj4%?%pSNxHO7us~BL#{rpy6OC(`BI2zWxKtF
z9#s;oXC0|nWZ1p|7Iwl6U)}M@__>%}h+Fs)<fA&=6nJJT7Z4RA*lPs}5NN7XId7C>
z)a<6HQLnM1#3pFmLm}`xS1BTahR+%he|EuX$sKS6E%s_nWA*qMIkCv6dp6<Gqr^p1
zxVEJor|lM8+d$=jG7;!W>!YqZ!w$0kB+Jdc5CmdEsT<S}6(XIR(9@f>$(SnMM+>Wu
ziwz4W|0(N2J-oT37}3l9%8Gs<sCV%}bmarwb)N)&QHlQn-R%&qY$evfz;aRxr7^-?
z@Ftm+pW6Z=LFCpxLfj~`D2;S3&;~A%ir4|9UFeFm_Q<*L(5nP9AEqY50~!B4SVx32
zaj?79D13*TKx-+K0yFW~Dfd-=_D;pM7u#DN$@^HaY+#3n-d<KYuI$lR?!^Yf(l8#Z
zFYy-F6T-;!auCAg6VFGffZ?=-@9sm4ZAT7kp#B{A7gY^0_~Vo^W-QP3FUHzlbwtj}
z2V%}rBcVo+rwdAqSCIz#d85Xj)hW#v-twi<x_t8B@HG)-sn4t*bbl`QLhp<M^&c+X
z9A%TYT@Gzj05O`P-cPnN5RR&O7Gd;h>-bGRozF_zWQ2mjwvhM&-E~z%jeR##iqOor
zKOf%PFidv#Qi*^N&OIlXcdRw)-+pr(XrL<#YHH_I^>TzvhnNR<1X{J&@3vOF4Hy1X
zU15jYUZ=0yC(NdP*?H9>kb4IcYtl2f|3<pdkwr-R?7b&nO~dt=V(p!Z1~A`o{gYWy
zMHLiUV6#J2Ot@H__q1tEv&%htFX$nAkGk^03vbE)Ha;Njhx+oDjtOLx3jlhj707zt
znz`m@N7fA;+1@~2gX~*B>yDi)3^@5pU-VBH-}SZv9TG;0?f|xi_wWD4cV??E;_#~m
z0FdUC3YyVFiA7GsB{=O8Y*%;cEUH<fy{74aYOE1{@?Yas0VtJZ1G<EjBgFwB9$0h5
z9f-I?YiIdyMLG0c-MrFhtBhbrYrTTmHIXd8ir*i901r*n=<4K4NWQ0LA>nCK&L!ve
z&9LiN*q0??bROg#S^D3w7r|{t=dz8&3G-*-Q;S2{U-mD*o;GWc>}M}uy8kkG*EH#}
z_2QPRhku;Gp@F5wj{V&`b-0`qczqZ)o!0TBXwkPyWMCxPSun*VC)HSEHNn|Q>ZT2^
z3X^F22eZfmZSKP>B-RV?fSkF4ZpM5VvvWQ)(*#-Hmd!5lNVCJO(y$}e>Jcbg*YSA}
ziwG3#%TL!tAro@^?Bu3!vd_-=?0l4|g>@KW9ur^wc5$^oJd1o{#R+YnJWnv@BKV5K
z7kOzJv}`lhAk&ioAoGMI``STIb@YMWKnXi5ZxfX3_C__vM3h0_UhM!zV>S+VXtSrc
z$-`<b@P|~7fOMEN!~F(F#7!Gyv@NZDz89o=*q@hDUlg>SqnGJh=1h;~Y+ETrf9|4d
zE{&&{Sz<0}cj_0(<zY(`Uy$Rg%XGyIKVA@y)?CAwJyJVu#=UnJiSw`Wtqx{gTbTFl
z_y_G;MLoir$4&EGX^(X7RL31n4fKLp<Umd$tgCCD23yJ7GrXDxHk}zu<~ZY}N*#<w
zw0kQ<AJFh(QRQ9?bgLT18~QrXUST0_Y6J_-T9XIVl4uoS!Wb=D`&deU`=_$cn_5Tb
zG?hOgtJ6xg;l89U_R1Ckx_p0WSHY-Y9X?@_ac&`WQ!BlISED(qMCyBvlTV(k6Q1u;
zEC8cCZ#?Js0d|Os@!M$L;-gq(JW6A1KW##phU5hlR8nrMZ(p+sO2`9UVp}>3TVs#J
zjA3Gh&D4sLq!euw5$uvAGZ0-V|HA}&8}^B-HSIp#Y`Sym(M9Et=I;0)PCuMk=FJ_S
z0weK0^ekW`^S5I5TVg<a%{7;I#=Lp#NFl1`BUL?R{^VMafnw3}{eTK(t8bZmQ!DWV
zYiMNtbOJ}-_g33PC*&313&S@&suG<U7Q?U>-pGZ;E#+&sBkZ8=Z<(uCgd#ZMujyy%
z7j|!4c|s&EYqjb8Opw0DFQBOyktf&E_xA0{N6d3rc)oLhg_8Fj2md(ADoj1M>bl{U
zK;!hSpGRLtT;4yStF7DU(9Q}U@5s1m8^r!d&&5X5&$Npd-XDp{6M?M-0_Lg?=YhLX
z;+r`*{}zF3$;lR&VF*8f1pWzhk}l^yQ2?2cE|-A>pk=zOAa;5`ClG@ie>Hc{kr^nX
zqMa1*RORL&X3PjH4DbKRQE<QID5@wyKa6+`;fr?%$r|q0`AWN)Km_(I_&h;3=hTLT
zf$E;^8Si_^b7+=8=`#}~leJpfRiioqlkENlFcp@bWNT*S@8SU59K?zpS`V#eMD+<i
z0nQa6BEyOI()dc)X<tGo@?7RlYa|P2M4o4BOe=S6<!CF|xQZB-HIJ#MpCG*Ard=P;
zhg{!KM_3q)nDw~MTzzlXy|^WDPw@bz{9aV*8Mjk6U{3Sz2B?rPm%hK&sahQf6tLk>
z&Q{lT{zG`xF88Ll<t?uSiLR)RUfbdCQy2==9sVD_`Tx#ED;OU7Hm3D`uds^^oqy}>
z+w4MnwkS>B)f5H1?SZytrWtwb>qhcf<0~)2o~nw*K^<rb(C*BPWuAtMZ!YH>)@mXW
zTq|1wZZ8lJP!#{)0%ur~Pd*gn`?msB1|lH2<!tk9dLto)TK>#j!E1c@%c>&V#UGZC
z?4S~SPvunW>QVk(>le-xF6Mprg_qJw<BtaCFPZaeW`@8%83KJ}x6Y5fbCQ5Ayc1u(
z=5jevY?68qanG4GLc^_#%pdnXqi*fD2I;odCpEczlD`@Qbl8;*qx`r~xoo&-wuj~a
zW9hu3*?#}`A7a-Qd(@t#wwR5*_YPWO&)O=gwN>m*iUx^I&Dx5JRjY)mqH2%QDysYQ
zd-eVO{qG#-aPH^pzOUzXJuWu+ifY9rm(kzb+20K;9>;ub8bnL&H3kMLn{Nu&stoY8
zIzeJ&*4tQcOj%qv$BRt$6xi2E5nE?gxh`ELL=z=<RBV>FME<z`J`CSTEbC_eZT@DY
zer$3BuP@Qo<QZc8yjfF<&PbPT=}S<N+``J>z$0y0plGjXD_ymIaso=0+@?e*n;7?1
z+%B9zl<$C(ubR^Lh+Xw@ww*nd3bP1avC-Vt&i|;W3dI7<;7aX^<q!`;sw9qBOMV8U
zu3MdsT2u?LdU@>-;2!DGAU13PPK6|G$FZ3K#B*v(lr~5#1O|d^5kOxqtF&Qfw@ZAL
z_5l-^un+N#G#$$#5nM2`>s|kV*VGyme)wv_scf~>7eR5FUdLbkfL*!eHJ6ft+8w%|
z+REH*F5w;|+dU@_^(#}qQ!N>m7LtcKDVn{GcYKQe@+=?S7^^WH>^qo-BJAI9so4G~
zSuG;9s<5&;{NmdYM}{jEgEf-XVmnQ6H2=uXEzm6J+x-{EGVuDzjozVzKDH;eR8YK5
zltRT66WR7tfh|T;-!MoQbkN>Dvgg6+FD{m*T+ZoVWkn(@?J74~N3xb9MThG}U;2AX
zj{I3WKla-`vcAUqV03EP`ekk!6r3kFeElAcyD5vaO4kV+IkV0m&7-w=zOXEIX<P?M
z<k4?fc4?jpP8woyTVHD}_OHKJSA+MM{y5w)l&V&>)WLG>-Q_82uI}$-unf1o!OKf9
zQp%i8YL?)fl6JH5;27cNXD|_vACpJ90q}s792$OX7NU%UNOp1k7ok2}-M7H3nD@GV
zB}8}Zjrb&u0T0@v(|Y;+AJU1gkY6chU0qT&3PTX;RW6QDj~#`!xkwet|ADl|W9${S
zWik!h)@`0Y>F$tOIPI}hCylzx{cNU<T)9>~Een5-$<3CnjQXvlM3J5Q2b^etGD~Ol
zoMWgtGs%xG!pFy3k(@)NjY8b(yy`vS>OC!?U#7TE*k5MH`2#*?CkA}Pu7R1EoKsQM
z5r2jfd||fQRFGD=?f~|VKc7UAnK$0CZ?C(IJ=>O@nJ#oYq=Gj<p5gk&tUFgl;ai%F
zy`FM0T^|GQ!`S*q1)$O@peq4FF;O18S~t<<dd(I~Dtho7#|yryW&?@@uCUNk0PJ}K
zHnyX-cy7c0x77Pv{V?d1+aDLpD3Y3F5u0r6DTG+gdI}+B9%kQRuH6%Kpdwe`Rruf^
zV{7OsvqAmmEgKg6Ogt1SfYT2hL)aXP!bHVl$Syy=ZRR&}d{~dnx*X0>PSIxiT3cBa
zP%E);i6wQ!Ue*nao0oBb%B2{cl|n54VD$E!YJ8uhY@0kh)|1e16`m47lX#nzY*+mj
zj|=rQ(;Q)LtUX?WmyrWo60fiK;C{I!dqf`UPfi@r6*+;=!rjF;@TOzl+MJCc@GtAK
zwIXV^yxa#@6FgUohj-3*KYZr+^eRGF4})u|qy<`M09dln|9^m6|8uAP#}Fh|9y)d3
zyvB#W5q9GGiy;Rs^L;>-26;ijmzBN1<NGdX!bmRBq5ai!BMViI*$k6#+|evA6QF}t
zbW)fN)=dEDIiv-y$q{4q-z@#V1`6N-4p4PqV+zH{(xFUqJuPvMPB06!#85+XT4ol(
zke(NNQl3$kNBePvyesGk-YPCo*oR!!vUB%J`q_;2|JxCA$vk`S-r10xQ$p`np7?=8
zm6ce{(Q~-tGr?<x%HWl4H3{&EMGC@Tn!62!-AahYYO=#^GxAMz5_m32*!SbO<DigF
zv;+`CZoY}f@GMyJs4ZWl%W<$Xv__X(1?c3RwTlQK<6TOh+;GX1pw~J!cU}T{s&V*e
zuk8Cg75yah*=~C!Xz6>q7*yEVYXx6;PO`(9Y@!BxoCfdsrs8w2OA0`^C|o*1r#`z7
zX7cnpL~x<@>FN5WN)K`-Wb;>Q^x1I>sl9<-Y`X+jC0~xRT&r9?-<}KXtr%OU6vo%N
zm(AFw3^9r209Qy)(_HI+TOo{<YzmwF0<)t@U>4y@nc2&|US=L5Q>8OmPpYl(q^F+2
z7h*srHX28dc9WN;k}&}PpOOm)!NoG<bU{KS+c|UoV{6%kH*->{m#6W?`Sl9#wG=bw
zsCAwTAebrHNJ9Nolbzj}7sCXM4O9JgX-xy#_(GhSxp(Ky@;Vvrt_WyUsV1?vWKq1x
zDiki3yf5CB+_dpSX1B~Mn>Q^JD>qct%}CvF;ZneQxcVcRllHCd;TdyrqQ!?f{;o(7
zOv6x$k;eJlgHT)axBb1W=rxIkwB`OqIIFs$G0-S^qV&?M15tYr1ghCg()9~j7<D*P
z%k%qf;Yu~L-kuu|?+YrrmSQI){f!8IQdVG{uaX{P@X>^n825E)v9C~_PRf=uihjHz
zAzSQDpV%*XsYo6xBvsmJZmN-4$+BL?`wiO7UG6tx8a+*%&*{{Z=`Rjx+(kQD6l|8U
zVJa;ZC*rAAwRg%L+B+HQ+(+v3uMI3a(?6mv<vI5<GAEcbOm8FB?Qigsa1naRak69b
zi=sqWOOzG@k<Yb2gu1JbBvbbA{A?{b!hI+vI};_)+i1BC-Yk)~U#=m>_ntZ<H9ny?
zv#rrXlhF@RXc2CjfG}?FRHtj2S;Y5J{r9$m4fa4ynb(a!$0xtVzjwYrvQ<xeG9>1*
z^6pdD#6T&}u+4<qYfPWcYM-krnlPAQ%(L?zXQ`w@N4N1~rAkrlIkU)<Pb;Qs?HPSv
zWgLOgS(ZU&__fK*5FZHl4a0gl?DA%%uP$E*3K*^U76j(dFF)t<7?;!b=yE&8*s3mY
z2=6p&bEti|tJ)>Z{;P+%^cnn9$|U+3eQI8CG=|~(aw$r*2Bfn1Z2}IO88B_XY^XhP
zFb;_mE>`o(A^IC03WVlB3lISAMWk`Lmszq!=SOzm;ZW?B>MHlLyb8rdtF^#p;e<P!
zMqLsl4&rZKGaOPR@32{SKO>*!M%LFF{q^ht5QhBh4u#C{P@#-Sx5YTY1oKX(ve>;X
z6iQ^HAc(L~YBIT^<n*>=^{a&Lau9+MpYh9LbWd4OB=y&5<u7fnYHE}=L!3msE&5)u
zFN9B4?skzTWWU_Qxx!2=9&Flc6oz{8u)3L#+3QD1MBR(!xRUuDy+v|z0P!O`5SU{Y
z_QS3WS&&U$EWoIKL_{GjOTL2k03Vrx>VB1Qz4gZMxu|&$M!F&MW@XHn34FdjL_Qdx
zP>1?ILXYKtw6MH5df<7!|EnuE#sy+nG;ybnnm+cpWb0xxe|-{8x!sXl#dHD_yjW|?
zH3$}|xb)s1mk_q_Mw_?~@-nFd@huQ-nC7gsx~SF4tu|dq%D;0QPpJzAEQ4Y+!1J7j
zMnP{cEQazMApdPWdF*e_N!^(7R;EsS@18;KUh8sF&g3@(bKaKNdK1rR5p<kaO_}6|
zL5Zlq%Cou&_dR+K=lBT)e>azs^eCU)oI_(^=%g>ZbwN;?q0iA_v(SfxNiKl{biS)4
zo|hxGK9xU?XsX^dJ8ZAdn&W10L;aC0p*S!hzQfxnzV&%`<w$ci2Xu8>T3;UWLOJ~;
zg!_c_&`<KIIky=5;CSCT-Y$r3NhcQ#^v*`qfVUq@E~i?Axs0bh#};i<-pid(n7_+4
zePBw7J41T~D{JHVU0UxTfx7thm`x3hKUNiA%69Yos_SJ}n*h+95Ore6vAD-z=F2Cx
z)3@em$4+Fskq=yBRsuO|LxvNL4czpRuA)_Ooc!cgG&ah5csRd7NvPPcH3L8D(b_C}
z4B4zkZ6zAP2gQ)(n_`JCakF45)y04R!y-Uk$p9)T>w<9fqcdhuEH|ARp~ZA#H;-vN
znAEUlWbm3{__>kzl)%i>{(u#aom!*k{P$fY^xGJe<X0z(poijw!UN;5<<!x%@HmpQ
zjoJI-vSO7odtuHiUH88YYVuout=mti?p@;Mxp9Zq;58dHlk$YZtd>zOs$Or3fnD+W
zzQ2>vFs&aQ{oMaR?-)Ow4A+U>4`bE#%6hxE6ZTfq_g2Fe6CS+2n%R9@VfuJUefEU!
zjyNNzN-Jf@Ej`|zw>#S5mY*RG;&b13O`1gQmB?lDVWmScl|)0C8MW?RZ_m~X-JsC_
z90oV=I_~xXxj;~(kkwtl87Qt=ZMEzPSC~wIxG(nljZb(8%|a$UN^T>kUPgr_GnX-!
zU?+hF))!F0!sdQcOF<CJ$aO=RmTVi2YgF^=#bs^TL5{B1{a`kEL_DQ|E-!E(h8Gbz
zcoI~upyy|^9bUrO4S4G%^J~EaHj%Qoy4<C&U5?Ymc=>Vqb0JQ0WhZK~2d_QVk;^wB
z*_T*>#ma|z8N<PH=%a@+!e{Eodu=kezb7mCe*Ghn_=vt@nSM2NHDTH%|D#I&R0~@6
z?)wTu=2!zspRbZlsO0vs{XU5mRu)hF5XUJ$pDg%_V9G8zPyrIj;?D`Ij1^<%Hw<Lr
zi#yWN(8oaDvG%-@-pE)s?!Ljvv_M5xR9ouD7~Q@2a?pCGLgIxg`|C(yeWb}xTY6}P
z7$sgBVp;J3Llt#WUOQ+Zz?{*|hDcCXv59jB^OtFy@(QBZg`sJN^5_0{N;_UF|FjHQ
zU20$0&5bWOShc(45Dd}3hT&og%F`|A#Bbf!g*M8xh7^v-m&6f!@TK?_nC^X;qJ!!A
z61m;GWtnY>*Z2E9*w3T6Wle;8`M^D>7N>J;u`t{Wxbr=Ss5N%Wb)EXtgQWP_t=|H(
zF5|-hyLv+*?y@L<E6<{$5Efv<JD58AByPJMC#$n}uZx#YNneD$E(5pvdwqeQEazHY
zaJY=;a@anHX11dIm{L&i3QOYIv6zA6a0PD`5W$E@W4sDN7UP8Mpn_bM`6JxE7?Cb(
zxrQw4U)v83I{Y1(>Wwd{2Uo^kLq0m?NlJpG{;gSn<@Emn6bpdev=M*=k`^dFLr)F1
zp!j3B{#K&rwamg!P(1tUhAMv{sX#2A^mHxpkj9NaQO)Wgcg|l}o*jD7B?8xGB0n+!
zm_p8%rv{<O{Wx_&>qcRRPAfJ`1K^kZUwtglj&0z>wlvkyeXCyD>ywzYEcUik9X2S_
zDPvi&`$nVDpshogqq<SAaD!x_bkB#i-gkRBFF?FD#v!V!$pU@3UwZeS>*AaZdLPlP
zB(+G;#gh3<%5H%_+nn7JNKtYZ5x`v4ET;xi<M(Sc-;?E*stB8jebw&#Nd;=zA+M{O
z6D6`|uir^~LS2f%O3pBPDuELho17fo+DXA)EhV5lhg1ZcU0Kej9}_F}q;vPI13nOo
z^NX8IZammpRw64&VL<z86b!4^VMuqH#rp+<?06S|P<B`UlLCqIVdOI5oW$ed^t6A4
z4HJ@cygTIPUF~blJMikTCUe=sd-+FZ@{Cc4LZ?sOrGg@=S-vL`Du3PzoHu{+jCAf^
zK67Vt?}=F{92yv)Bw{Y87iq)#Y1CwJ3^2Mgh8@{2p5Cmc+AHAj*6|Ltl&!e@GL7U&
zG)~VohK)WIbOYG@WGyf--z%GhPB|f?0-FwlEXc90v!$s!$|Oo)&Q!_|Cijm{mjda<
zkj}E#5CQ(=u~|uwfHb%{Y)Ftmti4<PaMs>TZBZ}6)2A%QWp)C&(Uzf4RZ;p}PxM0g
zQaHB<itS9Ahk`LDIva)4HP0nD7sdrRbrrp+a3xz*7uVj6?E*{tuZpZk6qew<J|prG
zBH_jY$y%|bp)$u`%FVQ9WI11UIlBk%Tdp7NofEGgwZlS+oT=^9#Aa=NG3KG#<=vbm
z&AyTS9Cu*-e22JQS0vB`Drx%kPx*N-TcQ&^(FmqmOL&y2BB8f@DBx`4H52S3;ce@Z
zP><(2_Vz_f#gaK{Lk~i=R%Rl@9BnTy6-TbA8(uvhV5DuL2?$UhVKgFYRKaWzB5SKs
zfv&(Y`%;$;B7wkzd<PqQZx?m34>NOlN2}L0X~q$Ctu6BUf^#eig3tWkEj?5f%xE_=
z;7Q1jBx#9w0Y~Uy0vna=MBe~hN$)@!NSCz7SeUFgPT9<ly{-p~2MLSG7$6WRZ6FoQ
zd|%J?J}%_Z14AyRxeyn5m9nnuTe@Mn_L2v7u5ybnj>FbZ)a1n0$Nt*Vp#*utBkuUT
zt&AVw`0(Y`>C-^KiTSke?QHEdd4$5jQrSn=^ST=40FS2B)2EuFXZwBSDz2|zdvi<Z
zX?fUR(|L=}4a&_%AT=tv_ww3X)ocuX;IY))Ub`cY2Opim6b5Vjc(t0$+3NSiuNMya
z%RQ6mk;~em7gS(vA@*+jJ6?UTiPSACKP8awSHp^QJhW<YnI}--(fL!tl1lzhdq3%-
zPco+a3A(3qVsa-Yq+FI|9%xsa-r(7Io9PQj;}BGC$Cn0E1E8IFRo6RMB;CA1Z;gq)
z`N8iyzi?8NxR6GUHLpU5I|(UsM3XB`Svb~`zhvhlONB!_8dOUs{c|r<6LNY#*mmeh
zo?>}=vPxH>v{68BD&ppsG6lZM<86l(%eTxlOKkhbN#uVTT+M?UZ}|hq?25H?pV})P
za<;v}M0DM6eHpWG{xl}*9oIs9c2APFgvghQ!t7{{@!d!AOM)qLKbm)cq`kNI8c5k^
zw8<PziJOk55b2hGB}}__VnW6ucvZ8zbpdd1S?>?>9n>1ZZU+?%oa3d5)qP=D5MhrC
zoU28&g9`Uq)S%}<Zr9Of>ns`dcrz17$3dre3pryH5FTBukmCga<o55j@^8k1flbYS
zH%|IR>wlmSu9uPR_x-D$l*ZR(=~`I-K8z`ixEV(({YBl>n?IRb)>mIGt8kuwA>!xa
z74uUL+Rk#y$rCH-Q@n6&9cX?L<%)#WQca<60LEysf5b^PV2Z~l3xU{ba$%I&@Ty~x
z0p*+Ge{1rIx))A;b}g?>J?>4{Klt=A`rL8GhtRgP^;IV;exp~Yx9Y+)!DqV3kCV&Q
zde4+}?ykY?;8aEiy}!A+r26Rn17G$!W6MkVD@3}FF`$#4I@4T@6y!jetwky=DVdQJ
z=|t$6gm4omxWeyKQHy}k@CX+N46#v((o)%P3{r213wB@Wmd83;zYfZP%JWwl%2*59
zon9AQH7;m1_tKGfS!;Vb5q~)$6N3sO`mEx{_3J8mq0MHBS(o5-g7PZ_W;Vkw1xAU{
z2^ioRt&*6B3<pAnHO<IZ;s6g(95L`WmYIO7ED?pl5Mng6B#9#mx1y{ku08IQ8ylX<
zpSFccBKB0umm|3!Yz0|6%w>L4u)S$@3E$OchyM24MPeo2;9ZXlIrXYMlLxftK3X+T
zGpy9i^rIF9qH0oz+H6WnWeHU&Ry<-`R-*oWbte|=n(BTI3-@_C`|$?OY9pumB}1L0
zuO3`sFIZR7Nng$IbXFixw^&){?pYcHL82g^SlNP#+pxc^klAzT<kBC5pKU_2BT>q3
zhN&{eb^bDASiSe?UgP?TXt*Jz;9FZZC+ob<vrm#uF<8*bS{#OGt)Dm4^jKCq-ag#U
zV!kH>je1Gha&t!ki<Y;Vf#Kag+#Q&G!E5AA@|H^=*6e3qds_!~?JN!62W9ebwTX~F
zv{*UA<1$*zH10~wxRhgMJrd0PkrMT<pTr1g>kpRP)8*_HHKX<KP1YQN*sh1WY~R&P
zZv3WX>15O_Hr!N{vC%j_PP&t(K<R>%^J;>Pxuc65tJ$J?Ju*B?Tx|!Hv<%{W$@V{^
zd-Oh`3S-fV+sXWF7UCvvoo%xftpe9g-q;uvMEpAlaIFUzfIJu`Z7aCQKJopj?GFxx
zAamlL#s1Y5Syhi-MAc2tZ#C)K{c(O$x>@<UOSv|Tn6}R@f(-nkhcb?z>)r`C<8Ox8
zoJIHuSoad>!t`_Upnr_vFO=Rr$s3$?>Ke?lC6VS`EO1HBxN3`b=7`m-D?sE>0uEBP
z5-e7>>)>bcPkV4UF31^~?*aj4@u)Ht?EvA;I{nrG^Beb)6E&sSVsT}WLGx|BKf8_!
z`H#*UzMC+KTCa5fRuz7qo4?ZhNwi!F8|Kb?WqtQC-}w0loUWpBfSZBQT*!$NQg@Dm
zv^~ZyV~RtiJuP68tIT7?EVbU)U1M_fRE_ym)4oG7LSBhQDD_r!o&2Er>kg0n&$Vp0
zUgs5W{g1P*Q}igWAN{f7%lMy89~zX$&(Wol0hUj<{o~w{tJP9T5G`x)AzXD~C`_PB
zs7gNTK&;*;|6Q-Dacv#9gk(M!qR-dHTF=&P`b<wxR781#xjtvzyO|4VO>e!zE@<6m
z$kTau#7$tarpUY`#Ln}}uK?<}6<NPMK|cqd^H)K*VfJs0iN;B-ywr9kp7Gu!9$d6y
z7e<YgK<T(Ckrrk7Yu7((T}=0nLw$%ArSOxrK{J9`7z^3*(v2W>Z4Sq6y8O+Y6!^@!
z-r+mm@Oqu}sQZFC7fVMb)EkOh*w4O-+P6)X`%9jzm)rJPkOnFqyv}()`9K7$puXI%
z^%oIZ6DMJ?+v3mf$iewc;js~YX&3cTEQoX{7~0goS}yb|0mcT@f%APA(da~dDjks@
zj*(XTxdc7k%puc<%lxn)-f~B=9>H#OLwe*vJy&>9a(4MRTdQCi$E-#DJK{{=t+m^5
zzSU~j`y_VDEm^r+IpiNr+o~=39SEe%{4a<6Tax`_nyl;zIbp^wDTx0T`v2#3`v23L
z@jtzQ%s{p){{gi5@JuYg)laQnp_E~GgiXO^=I(Ckyf&U!eb$p7|NH2%%0aFV<>huU
z&TE`3CstgKD{?zuy$&>P%B>Da2E4!p*+=xBUihEiiHKAP4DeyZM2WJ3YZ_rM2Y-v-
z-7*fX@bL71GMJnEk=iV5u?q8M)zSEGM&h4bvkKeneS~<?sJ>DxUD@-{2dZf|jhacn
zCz8qK>1w|z47nX7+^gl`HFD};t)nT3Dw>L6xaA!T-R^RS@zG>0D+#tNGUux|hMiJM
z#)<2L(IjC?%|oGnG&p^@t3!8}Z$L#zep{=quWx|m_Ppt69agH&=y+>l3dC_JkUG<z
z^<ZK~<=2w-p#|Zte9wzM&5M+^@6FHEsd*5s;?ZPD<Pr@A?VFX{X$|c;PN+mh*e4l_
zwgb#|Jax|RE@pY+@Tw9rxy3|4!Uy2ZMvpl_CSJwBM{GjA2LtB1!cC})LBzAFiffHP
zW}TSUGhfhNh0UE?8Fx5Z(aDRKIp|`@w6#xegg^f8EyDRp6euyyPrU%idg{2~iylxC
zXP&gbRA|Yl&GBb%lzU)wSmM4?+1H=7DLXTa$g`L1DDFKck}XFW{UqFZzv%sB*8h`b
zVw38x=#l?{>>jHas==mvS0Aew#l;|)whiy}_`cqj+kiK&Mlv?t`OC<wzahW0eOt%x
z?(m~kxAPkZXK1%`l8LInUT946=@Ya5aC1NR6D89(ag8>Jv<N3YmGY?*A+U81H&5fc
zv1dLLG<Ki<2fBG=aNTY7#RFRZEkr9UUx1{-A>Z_-OoQOUizh8?s0H$@E}$<;7MVMh
z_n@Y>lrt!!Y$>l#oECrLr26`RSj$$u;q=eoV1@*7oM%wt&y^rbU3C4lmBGRyS27jS
zUYm`SG&n+CBUNli4~kMy<1uK_d0*z$SN~9iIqEZsal41xvVp<}$7vo&K$fnlEun!f
z+(*FEuL66<JDs1q*!yk;#S+Rs<aJVJhc%89lc^5VUG@CYhQGhdQ_|94*M?6sFEKJ&
z-Dk{LqW1zw*Oo9MXw9D;7$dF3O=p7LCK*v1a7XB(2s0I!mpU`~+R-Mmq~PFO#bb(?
zt(Ee1wk1?K66cTmBShPWuVy{f>2=pvI!d%kFB{fspRf?zo|0sdnnWxFMkqg>XPy;#
zO%Gm2kM|ISOVbVcTw_ma=nY1#do5=0eT0>tNh`aMhrev0HBCl=+_dG-^+j*+t!<1n
zA4=s8zlvyKrPO^_9)1w+Kx2&7i#LXy3CkJivJelB?oqTbS-W)4kv9iwhgA&Wjhvss
zkCn~RN1vz9Fs*eCILy>@(k{>6-sX6ov~LyaB;0)oh%mkv5zRtvGV7midyp*;6%Lzx
z7zmCv)s;PH%f2lqGi&ZGJVj>M?3gyRL@8->&=rQ-w4M9pjW!54kw3EYSL-E-Uu^yS
z(W6I*Z2hNA^SHxOkWhlf@4SHvC!cPP+Gml&Eg$-AK6QDH&G+|4?`5&v49l*6uKVUA
zy>Bib*ACih_02;i?l7Nxu}N};nd3(sSyp%X+Ws^O{&0Cy<!3Z=Ak}HyWB7r`CW=|}
zREV6@zlGT*L)>bQEzt9<rn6Z-S5R3%q}iz`#Yi2Wt|!#4$s5`15KxsO%|`vxX4-x=
zvdu@A!t}kKE`v!nJ2@A}B*KWyz5=LZEC+zO5HPQQv48+t+A(yzPA4+@1JFPjg<hCH
zg|WV(tlU||*og)sIqZp>6B9hM`?`%vpB9nDv3&6N;JLEW3ix~@tnHAvq9s{4>l@EA
zJZh=6C+p{1lHbGDuI_$cjD1yRN{2tg*8(r2RuYs1B4xCRn|JAMaD{diKa@_o#>N*)
zo^fo#rpKt4fAtg4j7^&AnU=ybds#pl7r6i)=B^{oAyck(SH^&Y0QxDF9112zmRXUA
z2Hm?Y7_;2jo*SncEXrZ85Z){?_oEz%T=}i?&@zMm4%IF7bVCl24YpdKX&^*zbEg?3
zp8uiQh^`q45x<a7pV?#Z^~dD=Gu!`j+#7*J@&5stja>PF9OvVHN7g|Jq$<jZO(oUD
z&WcOP#Os-z$MsFL^S3MJGL+U&L_w@Xg>K~o2)&G>Y);wrk>)onv3o^~<-!6`s*2Mn
zts%!!MR?4TR^?EM{iksPMTHb%qHF*%Ll>lwq9<OagPk3RVTVhmvGzjSP?D$wpFSm#
zRf5ebf0vZg2}AUa3JSJ{x5#RvVO|BbpOBna-0N+5!^soKnp~Y8uZI=WVt1s4LUA&2
zq@K>@Ur9A5kVV2yJueIbiZs`~iq@lZTDc<$SVOaqFbkxOVd!}!?2+|aNvWOwYbJA;
zAa6tiLt`y{QMZLsFh#v2B+zTib-%IApx7ki!*QM}HEOfltG!oo$^Er^q5Q%p+m%xi
zbFQ?%O+AxMC+|~UWay>t20bb$373JpEP1Po<!Qh%Q|)Tsu!6t`AuQhnGHh-5-GT@Y
zvS#A|kvD_bZ~`lFOtso(r!dNl#oQAZ6=+;_K^?lEg9qcJ<Y2EL3<h9umgvX<;P$8x
z9jAkUcB8Zlp&DJa{!N=&Ajuv&c$%(p=#CQOyK~51Nb94B@~Fq^9~foc9~o-ljc}K`
z7I(xo3cV0YJLm*=n<Q;xlN`vW1&a_EhD2@Af=?G@K~h(ds30t3HO{w#>Tn}}DtZa&
zOZ(*S3`0=q8zH!dS0={_`Mk$T1CM?}AIi0anYJ~ah0{Cf-)9iVAbP6iBkE=)%5ui+
zB5w8trkYOgpH&dYJH5#hn2r8pZu==@lmF@bpyx3?qTR!KuZ3LUt!YecIsEP~KA9X@
za)pkn4-LF2cH=76^_*j(vtrzWj0poA;nUm(EdmRA&c=IpJKbL@Jm=t|atjTA_UT<{
zNWAshNp0`Cx~{*m3q?b<<ZODNnONM9IKawgw@dF|sj4cM627iG%ibspEGK*bO}&Ct
zSr?=o9sJ@;kqGkv{99d+uz2%VRGO1@OTu1a?wolDYCcad>@71TyP{5v$6Nc#<B)l(
z5dnnG&WWX+Qku<1=?fYgLO|6(s(Th}>#cZXlnO^5)c=>Eq!e6)o9-UGZWbSf-*DdX
zn}NN<PCP2qVm72X^fy|v>gk4(hoXuh*IXVe6Ct?Cw?*}$Bth)WM814hu>Br1;gi)u
zoFXN8U9C-#Bp-&t>(D=SB|q>Y*U9pGhQudt%gdJPf_GmCPwx+_pG4vPpVDup^d9`y
z@xPh0b7HPn(0u8=u=3nSCrY)xWU%{%M8aTh-mxuDO&hLXzG6IN3Sp`t#i`JnhJ$=U
z01xe_(lbffaQfMD+1&{?81kwyjNdT9*U2guLhogfMkuq7_V+3Rh>{p$JyTtMTzcBH
zLaWNpIxeO%`^xtjusEc}yGFYwzpWtD%FKc}#@<{gO)z@5_C=OzvEE(VHZybwrRtb=
zAd{4|IMeNSE%c9XggCJk>u(5#*y0{v7Ffl2EPw+YhQ1=19NX79qp9SgFhHI)9mKYy
zE^e5QB!V#y9q%mGCSU7lo}gT{eL_NCpO_7rm!Uzoq|8o}t{bCn<QCS0g~cA0+ZL3O
zZMSTSnMfb{#@i|X?j@QhYx<zqt=M1u`8IzlCgZu>Oj`5$d4^X<pwwP#@5midXal&$
z(RDchwabDJ-JPwyTX+DJD#vYh5vm4~1e1cAGA(lU47V(D*ps{m`2??LWk5WH22rB-
z`X*MV<32=@OE3Nh`gueh%ywdSG%iXXV$jknFbFc_6pNSwqfY`63C0nY8QUeoPOav*
z3drN_of0*DOqTi<h>QVan2pEwS$ty_1hYTL5dnFVZ^wdK(EQgs1_Y*vQUA4Pq{LAL
zN^|@V6ux5nS=-5SooC|Cqep(PMaZ76Piv==$D2(EcjOw7ivY02IJ_Hwj(6Erweo(u
zC!YbJm#B+Tn_Y$<hQZ>vtPR2WCjTt79<#btIm!R+>f*5&H^YAl4X_vlFV;D~%-j13
zwi=#!X36=)!r_;lSCvS`dj7a_)9#s-?5R!Bd;5=?OhLRXS9Jze{*6Tss6Gc&sI2%y
zGhy@QMgc{)EN4T%!^V*4Jb$L3V>ZvVd<LYyn{WwYVx}53v-FIp;8}?k6aOGL?8q@o
zW*B3V)(^J(q}BL2D4|Ygahmb`MOh<&PW`aw)XO~qfn+|XexDRKjqFzFh@q&sbs4eC
z(bdY1%p5;eqkQxeO4fSY(Mu>5%q{LDtOu{N*5*Mj^5vf4uH`9wudow15F7UWXBbh<
z63Okwtj-i@IED%&qYL5|<vt|x)l(N`>%>^G^<3eO7UMwbKI3wUFbWtRL<gw&R2Gko
zVB!_ZTo%e|7Q{d)IBa+;;;$_{z#D}X{x>YE;!!>Tt$+<f3g-5+xd$gRf?_q6d6j&l
zK=$TU@8aTer`WNV1oO6f_0}R;OsE9pc`+Y(0*$j!D3@uu;;O1d?zluv5~;0)4BqFh
zmD8bSw17UzY<?U~D2LypzX2Ed=;@a6@3C_!6w_7Dkc6I6w=T(3XGOy0=7gkFx!Qu^
zgW<1Z+Z|;f4A1hRo}zxCo#AUO7Kt(&TE(wW4FY{>HO`-moy49zYu4$&s+-}yY=%9z
zgX19*;<u?%65`Kb+CoNt{Xm>1!Z?~DHQE(=Vkz3`Q&UkZ)8aoA!zSr)dbZj1G%zLv
z?CAl$RA>_L2%CEdm>b@bkw49e)eUEXPYugVa>e?4eC`cF`^~(6vBK~@KM`&@&y^@X
znzXh)j+5|96Z4xP#k!+Fy4*)IBar;X__EZrD@6v`@z}=r3N1>>sV1l1w@N;B1s)H-
z`jH3xvK1iHaf0MdznSDw<V(e=GKIA+bSd>EO-TRQrUkM)ia9&XudEdF^_@M#UB4F*
z3H6l3?w(w>m_LWMG)SFGH*^u<MQ_3hlgNzax@u||9;{#az=W)o^3M|L7?BpagWAix
z&iq_C?{9GHERk68&P5%6xkJQ$N66y-n_OvE9}!gEr00>dT|fTwN>_DnW^BeH6nnV5
z7|H<`d-wQeZ%g%9`b#6POLpc=^GuG3kI=F+OZ)I6#k+c8o1HQA-Kv*APE!|q)IF(@
zNoB#&O=3Z_-fh~Dz2+H1<|kX>b*n*c8A~vDHzG^9SyR!*E56GAM#;y4Ed3GCTVLkI
z*F>SOe7Mll^-T=(j?iAhe&4<Renc#36fk0rnU1Wwy_W873sysB5_$uXw}Cdr2VBw(
zSv!Z*RNos^(zz>BgZym}E}bRD#AOFre!?W1EC50kLPQ<<L{JpT1njl`2QnD47$n+{
zXY})|%;b(ibly5(dS5?hCQ9UOrbJ}>=F0<lP#bYK=b&Ho4|e%gc56AOsUGA?SM&5d
zn@Bz?mmN8v$6gd10+*h?9saqPGY-$no*lwx*|{qXRj70ieNHy6w!>_0BOvwthdudb
zM^JXjulN&F>WJi>wbz9r+{b5)^7a-@>93-2ZtRx5+F#2MIJ8$GS+x2g1DzQGIB8U@
zgr$Tn3ve86BFuoCnpkvPl+6RYR_zP>xj;gvL8CMqu*=cU2sP{pU9NPipY<DN5mG;F
zeY1RoF0lIjj%($`mD@=DEuvZ~P_TR_C}}%K9V8m;JJN78i$4@@f!-W1p!a{1L|C3d
zw;oHt{)5sI=ao>R{_`yGVQawm&d}2XpPj78+vfZvGgh2jhxD5@YODn<=fV=r1?@t)
zU{>doO@eR;$KBOaN7*s8GL-6pe4O@Y!z>zNy6JqL!J|r^A~D^6a~4YqvEZ$W2j(o0
z=^_K(iAEqjsPhW<Wli*p+eW3*yr>0{na}ZN9*HERlFyLowEIkzsL0hMkl%Z!b(8o4
zkh&ig-{1MtuZu(UQIOnsIv1ptFCh-JMWdro0qRMj2@^YwI_=|g@twUBTOkkAEBQd;
zfzP?`QF~oaBbT}mQ`c+&;!FY4mp5=XqP`@&$(|F9F%gN+C+BW>fN`S?@IaL0iRY(G
ze)6;L@g<(f2s82a#b%4oDxboA-j6hDd&_Mb61OMciK%Pyj%|BpCC*OWf}|~EqQ1~1
z4&2UC<p9YX0_9|g3=8RD?0YmwkxG#PHpaW35R$P#e8#q)@vl*4LLP)!M3G6w9$Ft6
zxDVFt$FX)vk@<keeS=MTA+)R5_p>0?`*qcjxJZ@7KyIWp+c?+7VxvjuvO$rquFMTx
zFa%`;g2O?<*Y&Kmz)yvkB0>x(`K;eEFCTV*=}lbYe=HV_5D2$e>zNDHdx|ajl*!3a
zVrR<SLQTZb9HO;O0qf2fd(%$k-Q~#$wrd9yax00nl*b(#+#8hp9c4lHjUovj60H)a
zz`>1?p*Mb)?JGckzEu~${nWZ;(4+p)?^KgW3uF>1kw}P)ZP$IW2j;T(c%jzQUae<X
zQv$fiz1@v-X7BetIC1<+uJ?-R2f>g-cU|*3p8S4g<4vEEmL&-Wz{_0~NHGwGE_7PS
zc_XVxYi`uWHB;6|A6sYQv)HvFoN=0eV7b}-DZ+G`Lz&`R<Gy*A!`$Rr*UIE@jt6iE
z2QTI53hS1s!g~h~ixR<0Xp<nn#`XQK)GUeHjzjcgmH>0o^c{y~w)~vEE%;MN@xkr+
zJ!j9OKWBskb$^PdHtW^Wo)ct^ismLObt8Fr?^~I}+nvqfc_F#hVykuI`}m5A!1tl&
z_LD8js{K##j?4;d58o@@{ocYoLXvHC`0CS4L*^`$N6v*zpn9*->>b4fgoZmi>3~u~
zYl0}tu3sLy-&-@d<Jp$HJ$dcGPW~k)%(={-QGX#%90R<1--L1--+ai?&%F&ju~vyS
z6|fB$tBA0fQW!t}+Vr;cZUV~EH{Z^;sz~(xWph^OS#fSOkIGrggv;=o)n{jR*57lR
z6rnepO@j<7WbX_BTfg<o7`eV#jY9nQe6Xb}bi1|ooMYZy<T*StWTjh%OIlaix)(zp
zpNKB>^u%EK`&9vdkixRTHN*GEh+Wl>kE!Oo3f?G|QY1igX!YS|f&T*$L>6mQUfj1f
zsJtvy(9=ZY&-uYYc}NAmiLmkwRLW`8`|gF6eG+6qe{d?!$&1W65ic$7GVXi^3v)bF
zEZVs4Fwprq`gv(J>y^T@s*Qp*&Lb1r3^zhfD2?X^6C|mE)Zhkn!z%A`>D*yYWxU=#
z_!c`j*QUAUcbXOt=?zO(1;e};mXTd*RCU!c*`^!kpCvpC@qRz~)VDf~X4Guwz%F<Z
zy-YUhXBFh>sfMU$$~O%FQ?LtiSU7P4lu4qnGOPG3QTl#8m2=o#V3>b$$BA-X7=WSM
z1sotHqPNXj7YMXAZ*=?=GefBPUGYW1;E@wEwo@3D%H@&S6od#tiA_t?Gtfxr%apLc
zh_i>f88#MF6e(Gm{zg4Oh#T=VTY+H|uN0WXP7UK`TIH*>pxC%vk9ynq6DRAs?K9!j
z#Ls4AgKVazF6^BU+b93$un+<P(Xb^$6S9{ZO^tI`UA!?|E4U&z#{rgkoJNhTFgGdJ
zx4K+|yM*cr2}+h^5N$!vywOw<&&H98_ZpT&?akC@Skkzv<yuL$5u__$0b*$Ys09f?
zPSyedFpUDS2TOpa``&{FHPx9heYR7t?kCj4f7r3E5#*=Skr_i3$zX2Tk>)O2#|}nM
zWkp3Ye5JPwwUZ2h_vcIF2DCqiOdxx`Yj4S%P0X&Km(vC**M}D6oCz~`UP#wwSn>&&
z3Oh&<xxZJ)ZZWv61U)R1)J8f)%5L=)89(k28(n@tKNZY0<Yy=!A^Q8JkU=9ULx3Kg
zq>0a@|KqYmyvAGeJYPz)^jUcp6&mqt-(L#DYcU*adNMrGWX%DJq3G>?vze{04}RMD
zGfwz0Q%_~)27~arA0Uil2oPRw5G1;tRqjLISpWjBWdH<E3)W!Jafp-5k+zRutiMYB
zo|O3Rj-i&ak#;rTUnU2YmiNVzcKNyo;uLuA4acsR&UT&07e2;`@IoU7pfPSWw|BxQ
zG|z`<=g8QA+^e|f5`&0Ghp>SdV4!>ONcAIWCI}{EC5K8^^zJ1nE=*5Wlvrub4_l^S
zqcd)4!{>uRY@<MT2^io5om^W11qkrU@@)dfz&2J|p{koAj=^4XSMiKJ?8v5YCVEO}
z0IfIrg~tDgrRUGil}1-Hw6SzLX#9Z4p4)FM5brakWtggW&(c}_PjAXsU4iNltCf_T
z+`CY;A#sbDX(U?r<-1P_m%>jlPanSETe+g+zr!z=8Z@*#-8v6IjKZ~uj8VCno|6r%
zIYC;(q2eAbsKgGr)XHPA;R^>jOo0+tsMyv|D|Y@NXuWMVBqcTFO#JV&=v~v4I|32a
zn7JH2f5f?>^n<#<=ev%P8aCf}moiR!JDv3QY8u74TYfOt-*42iC%Le#IV_zhKUCSV
z&9pe(?tR^#()Tn(%^PGFAryL~o$EYHXnC`snOmT&Fe9P&6V=GWwq41f*y3~70WS!(
zyr3;xHjce5$f&ybt*p@!y781{MT~8VxGA2?J;N?2NH5KiL4DmX+6~p(<CeQX$lCFM
z9lk!6NFwlm4cgOE;n+-EF11~)Uo7B3zBP=O7m##f!A858TUw20y6`_1{;3g6IkDq%
zY}~8)t*If%p9a;}{^RGv^o(75ih+aGr5YwK`uxRaO(r_{3#}WgtFJ~{ypE_tbJk;h
zd2?2GWTmXjtc_%^AkO*PI>|T!;>Nqi>T{b>NooBGC)9M%SA&&1lw(^J0Fc4}rw>%|
z&HBAVt8<!>tnytk&-o*ri<L6@xk9oEXIZ_1mX|Yp<zE9(VDcl|O-jgD>?7vBJTECp
znzdJT3j9h4ITP&>!7Z`<ruVD!GTX8molgyX6~mEwDRrZUdcYo$Bgu$Q=M+6^)vr?D
z=t#eoF_*a6w>l?ad1_015|Emxso+JF>)5wOzVz%p+NAC*G%kVY8xGN=_{(meda}#w
zmvOCYu=rRMr_{q2zeO1y67lb>w0oI$Q&+cPZrFokt3`eZ_o^CZ*t3xjl5{jL)iUy0
z`4BL&FrzFOHpNXdq^;EncQ@+?)v4gZ6{Do)tx!FfdL!*u$&MEfz-4c^$0s%3lqfPi
zLpGK-=ODuy80yI?hgM(71?x$O6y#@(CnXsyvU2@pW}oxT2xdE)Rpqh*k%LiD_r~LB
zdT?0Y%UM9=Q+UEfFxS|;TqN!{GyLJlKXwJw89$+{4M8Kgpg5j_xCa5j$fbgvMA#et
zZ}o^M6l~HQLbT+o(wajOmLP%2981r&dSjRc+4Iw-BLf<#GccERZ0<N-dM@kba~LDG
zP!s^dUHW}(lUD;)O3!sw8%qJ8r_sNrmSH1p7=^D`T=uoU@O7NOpH|U?Ts<*o@S%i)
z=0;go%+j$tf5<R)ZgWt*!2`~!d^CvzbN^_GW=VY{51wGozePb3vlvkilBXS5;1%Ge
z6;rvwm;g)`#|}$f2*AYyR(e1@R|rB4INz`sDwTSETGcWe6@SrgPz1?-<!{Y}K0)H*
zSxX!K@|_=zd(X8;nvR;1EQ_oir<}1?Z9O-AFZOn6A$BjnwdlWQm)AWMw(EQO?Dg&4
zQ=Y;!uCRHD^%kOQoTOgPJE0<bVf~xQ=Kfx9ALcx4`$>TBAlair9(!-f#woKyW(COz
zp&KLoxkAm4Y<)2dxx)uMUGC<ckrMm)lZSqJJ^VGb3<+B5-&^u^+Zf;y(mzVP(C`A?
zCg0vC^PpGH6{HFyzV*}JEM-l!?O?FJmucD;PR)rPK{2#Wjv*?~y!p${Tw1FK72<SH
z3=&uXF~oXw*Q<mwYqU6!m^w@oh$(ju^6?Z9#Ta&9+;^Xcg}F2-R`_fodSpU7Ox7y!
zE;_L&Ske`j%{Q6Oqn^tq!T*6Gu|8=uJsSxUPtQRBzKIUq^-6V8<V%T3te^1yAXAyA
zImg3Z{v2A$vL%Ws3^`G}{z@G8C64O~EAzLf67aLSo}+^?V4gwI&m*&-+Y2EE<`C)>
zlV!t4S2l+3q7C$&0~RtqSP)S?XlD>~2gfcC5;PasvKws1q^OU%F0PlV@`sv<o4sPL
zlD_2#?`#KW7xfvizScx--q2=O4{)$MGN;yTU^`tdZ&<yN$E^?D&(}`xknrvEntBDD
zlL%SY=H!XjW#KBH(>VUv5+qpIF|~~$?uaQ8WyifYAS>9fEClnB0M(^qrPQI)SmwEF
z73)5(T4w&-HLTBLvl#2db1RE4n*-Lbme>cYjvWDPeZe0I6FutPoiZEsS^l#pLVcmb
zb!)G(#QkI5v9{2ypv7bzgLVsp4T!5>#EuO;`h3U8j?+tMljhf4?QY0Q?+<6(l&3l!
zStn{?)ZA(p+1PnV2deO|H|^KFXEfSmC7;*}|9UR@_s4k53ioD@GQ-7ZUD2>0rt*y5
z+Xn@C&hai;xzMNPdUFPB3K6b(3+T|mV<&g#?zHAn=UXRl#_3k7@;lS(zmx#&a4Pb*
zg30`w_T>Yc=}0nwQTUv#F2JWFph-k;<XB?ilwXoKR$M$Flh>h$ZoM$*3U7`8_|Gy6
z3Z@!(W49yXTIhbBox@wfcsP6M#c^;|`6p;gl+}=N@t?IF?xVW2wx9CJIMI-7<VxQQ
zyQ4T$Z)-zVGo>_*?7epei!UV?tRuW*wOl98o@+YBHln}!x}hJHG(%;>xmJ>wJ6#z;
z&??gxLYxAT(>6{%3k<{cYtY?K5=!BgkRs8IWV=`~&O5xgm;e^>sjA8y*O6~^Z?%$m
z1a21I>qP9Cw9%$>RM6XH4rYWzk`Q_ot?Z=gs0O}}dTy*ow|{xSEvh@>r)tjGlCROV
z3R0xQ`I|_nr|eopH+u0aF<V<$d<(tZWxU#RmPJ|PiFmoOUBO-U5VPN%HO~0G)at`t
zc^cO~hOqid9C{`yxVUti#S?r-`owasvL)@h2K<gdgZ<VbGxxei9ppz}+U)zlU4lQ~
zg&VgXWqHMZSv4Hiw~@Z)!fx^C^v6#-<T|~mmPo=?b-)3$(A99c5-qKA)mrk*?=nRG
zpbQl)`C?(5+BL>Kn%j2=knuiG7*WiBgi#OyP<|?vTqAQg12WcQRl;rQiOrgvCvay1
z$?nxG+>ZX4k@FJSZObF0f$N)@#E_V!3j?<gs0Zq8Fh)M;R+!auz=BL|Sx~9`ttDN5
zIi8<Ai?fSVpvsO~R!p;wtrxGxL8cmXoVJi?K;i7VT{4COu~?`O(iK$~i60!0T+Eua
z&}qB$AYXds3aZK%uEG8<sFAY(*ppoa{;UDQ1blJb+#OBmv1`JU$N%oPy^_gQx;F^h
zuPjJ>DgRR2!JtzN_eVk)4sqx9rMr72z?6Vo2)UF`;=5DG_FELZsH@W9m4FhXfd8}3
zjiNw6Rv1``h+=e&FE1qLxw8B<-pb6zUpwVmT#E(VeUNQ2W*cQ_zOi4Y!#Mq=UdWth
zSu3VH=ce4Aw&PSLrF(=&gGJ#jbyIWUYIl|}{zRPsTFKIO7xM<wJs1_f0^VYTc=1HH
z2Sl!M*-f~KC@;_L>!JSPkoNneI?Wv=iNX*qKEdSTzbDw{TLhM!vNcSz3q8?!6g#l9
zS^1=hjl)z+Fvv_(VHwNknFS+TR{JB4VQrf~VN5UgMQDBn;w0{V7_GQq2%?@QJBLh*
zV$Q#`Ys0(6knW3e6-<9y%-H#dQH-7XSV@N7OQTs}Vz&S#ZsrPb!yC{1JctRsM8Q5p
zK*P)cB97>XB~Vp~T+-!xTEU&vIc`4MFX6~2#C~$uOKCGw?~PrWK0M=|F@^cZMoY+x
zI6X|Tw14@BcjKg3*w<&JjX5j0jZju~<(BnZ0r)#fGa`#sL8eh`RB62pbJFBSAnUlH
zeT=#f(JT2a^$-{p#i{Y{Bbpl-*2RFJq5XNK3_vEUExJQMwcPco84Dz)S$^1O$Ogf-
z0$U2WB@Dlx0o=A5mPV8@AT;@j0gb;Xi^7E<lo=W(5e2e#DxG<7&!{&rz=JmzAIMg+
zIlj>(!4XGp)KRhAXD3(Ggj6t6zCW4*PTPtV11^muB@;%qO&YJ+DbB>0x#tX#=-^|&
z8>~<qi4~C{S#=JGjvXr}^^-*exD%*-@uxfsH2=s$S6NHvm#?L(ykCA*&%qi=CN})j
zLgsLf!gJCU`PRJV!;O=N-b}m)n<<^*Qg}zHav@J~sBqJJ=jv7~a`kz?AP(ocb{zh-
zTE%E>UKnXA`F?Tst3i9I(ST+mM_!L0b4zFZJS*evK}tWbZ~WBSv`Bk@w>wXLIcc}s
zS~^Yo%&qT?`N5Z+A5-r4BBFTw>VEUrpWnz|>w8SS&IBIY$kKJueSms47FaavadKW7
z7@2&a%sfx`WG#lx!nbMnKHIE6{KyS9dg1iqU4mXlB#f3^{=!FHOpryWz3ZlbW~^z{
z3x;@lvVxzi+-YaTN&G$KdN%$C;`-!d1yP)<bA3Yw_R1i8rPzkoVxHPx51M@;+FmwU
zbD<N$)gyD_lU_r`_uC#W;RDAj?yENzh4!oZF72g>rP=yb4{>G?s<0_PmfUwIvBKNt
zL!b0<mKE5A*CBKpVcFV1pHOG|-&QNXowoY>_0{o3jcx0Vv=mL)xB}U3`5x{Dn6p(L
z!~^Z)6?l3SFu`iJ-@xAy(V&QGtF96GX%*{VBbXxt*G7mpTP$!6LVKO5I{2~y`+rvQ
zj;A`_m}6mEB-!d%$8FLtS23%k*}Yw+{#;Hfl%K#xn`ufw$PH5*i|3ZI6pRwzp#8m^
z=6z54nY9VkyrL5rJ@OJ(%ouoRD(xP)D~e^Zmwe7c39f|B^T%^@DM|a;ZJLx!F~2^x
zOPSuS{56;<NNTH5er$Tcc4`9C$APHG*l?Fu7*I=$)fNh-%C2UW5eWggS~!7Gu6kcV
zDg-)>1!-R_KTByg(oK06T}dPO21Zm>LOU|<3j+eM0DMf=hiF5i)B7co1j-nK>SA$l
z4T14%4HXQ}WKdOTVjY)#;FwYptD7YuPGd+LrMe}mVEGg%$0N<H1u%~qfsO&#tCzBe
z`jzFYL}8N{KAdjt+Wv~!3wZcS!Sc=ejq^E<S`v=SI>LcosMZ^&JV2)?D4^B`s$hc(
zOOetv{>Gu6ZL`M`>vmRK8szY$`h20?Y$H`0*(9EVPv;IMSSiCD$W0vLwPqq(E`HXS
z4S^=aImcY+Uts7Aam-KUxkRzNkk0?R694l-!^Y$-_?mjz6pH2bC}n;N?!`8E$UDb(
zW9H!{qL(%CxdVO<9v-&7r+jYH)K)%Ho*IQ7B7b&WE7fDg?H0oEfgkA((1v^$^7?#P
z0=w`O$#*v4KbN1H@)2FM=$G;xa3{0$JiR3^8s|!MdFqI41`=93LgpgLJe@d<qzOg9
znvZaAaaBTS3_3EC?x}K-W+4~VSfte9m$a^!Qo8l*co#^!pkiphyN@+|SYPP=4Y~9*
za#;=i1D7&twZelgd21dAs=X^f92AiKh8I&O@!W7O-H$GDt=KHI)Z5=&E<E#?j$*nW
zSdFQRYY;3%QaSH*0frJ|7+H)}oG*yst^C3b=^KOvxt2vfmTz0njC#Kfgxn2yD_^?!
zf+6Tj7^=`yGUK`~xb6SZ^zM&L_kaBVX3jZg3L&R3B8SnOGK87)oQHDGhH^f2)ispE
zLe4RVhRu0$K9w%1IffaQgyfiG2UO~+_xE*uKA-PDAV28wdOi;K+YJ>s+bGFrwN6}p
za^X<Dys?q5J1FM7X{+HODnZiCBg3dKyN0?G%mx!n8-`BS5TwOo0UT}O%upgzj9fO8
zstnq-CW5ZB3iy~}!Zu29hS2CImw;cuP+5!U2l$2_yrCih(2$^X!NFS!#6Fbt7aEJ?
zN(8-oZu@|(_WUIsHOhrlP-F7!*ex^Db5~5{I>-c-OyahqUg*1vYY<{AqJdNIw)4j;
zzDJdr5t8SrMi$QJ@i__*xduGzQ&hn8kk(LstopgKj&GM;7e~d1Y39j-GQ?3UqpUzK
zsh8t%#wq}Hr>~V=*lvB^>$LVl0Zc7>wwU+Z&KkR`)bD@t!;#zWqY^KURMm_{OX^m_
z@g`GfGn46Fs-y&`vX0-_2JL2^$xOe7Zj1&^ZZa){6QAx+Vx0q>J?XEj2r4YdA8h98
zbh#g%>~wC+&{dIKs=jXY02r@SeRQ6Z>s#XQc7M{^6mluL?{<EO@Chk8h1v6Zg<07y
z<vlD{WTe{kd9wUsYY8->EU5XGsc-L_x7#Yf<rAL&Z}j;ahfr`wl-HNU((Y)#mthhg
zS|_mP4hY&bVgF@6-*_g2kZh34KZr`?Y-TaHZTtEv_^%$#_U!3?-4g?zL<zKS)orU@
zTe8`AJDg@CnTq^Nd-<c|LeEPYsLwll35+PZS7*nKut{!F?MEs)R3J*xr_M(HsIvd8
z)I@YBQOi~OU}`YMN~$twd`1($@p11>Hiy+Wog=cVY|)arbFPamT<%%B=&zchQ@aIv
zt_q1x!mYUa-t+poSjX$CF**YBt(O-bb($DHxja6NQ_hmRkj|R}(s#g@{B3Fb-Q%b^
z9gKbHukv0fi*C9|3LWJZDVt3L`6abye&J2mk$XuPY^zu&uN~23Tg$JfQ=^nhwGcX?
zd5q`=E$y1WFDXb$G|Y5G>XUMW7IU2g4d(MJ@Htx2W{JdHdqjh9(DJ5r>{x}$W^Lr9
zi-o1<-Vl+meYpRHtsJIIXbM6<e8ll-=F&lNWG-dt4oUz94E3j-Yh5lDhLe3_>H><(
zlIR%WQ+h_W+T!19WaJm}J2iN{96(&qJLVdaAozzQBnKh_#qo}rDd7kHIs3UwAei0j
zLn^Ya1i3}BX|wkDGZo&{yUThy&O@K5LKro;Py6tCIcMrYg9y?R6&kYx+3pYs9YY$?
zjjvAxKZ@}iFRyB=z5T#-Q47>ys*d|P;NNBOqB(hC55X=uH+WNEwSRo<L;c*21seTv
zf?QQ+igD27zn~%fe7k9U?JJ2@w0c`|nT#-Gp@vJQK46VriyK$Mi|J-?fg0jgE()-i
zU0VVmA;@3^-CFEMh4Ma%^EW@@3vzlJTGbxdI!MgrQ2RgN2MFK)k7sapx~kwMY?X9P
z!!J~N4u_Jvr^oB;Bz`AYxIDg&P_*LLzQ5kv_xA<QvGka~SB=qWJ=5(WAsl_+(@~2R
zHBI?+_e`4Ww%V#f#oPp)?`Ppk0>b13>|wh!A6)8Eq+=3@8-Z-eQHk&O5|vA<eexYN
z7sTEMzj|2V)H9c5XLeEZosz4uju$d)qU(LzOM-UN?+_Q+T={c$_!n*GdL_iHj6w?f
z3u_A7A|`I=Dmwc<RV6)%k;&E|Cg^a#u4%B>k@0d7LN%ba(Q8B`2&1b_*UZQ+b8g^5
zn+Z><43+SZ_AL;SvlDLx49bIahoFhhis}hZt|%cth;cSuefA5J4-?m48k%2(?5URA
zO57*YL?lXc*DCZee0iG8E-iJLc|wFqSe-jl+*5D;7*3Y(L4+#<$q`90r(rlja~=a`
zC1pEhI=zgTkR@K}5a0*NO2*V^$4bs1O=OQ7CIKkr#W3hAHD$8jZ`?RAUs!!qn~1zj
zWX#Ad57~tG2Ma&p3p$C^e<E5P7&-!?huGr%ovz{)4BQl)2G(>KnOuGa4~t6VbN;HT
z;er+qi&%@VK?zZ4;e=YOH~j)5qc|XQwF>vOZTnyI-A2!ZO67M6&5>K?VJot$J@rwF
zi6Xd0ZP#)8?&Nm{CNcqUF2}P!vRm;H@jk2P;ZlX5YgoLbu|MR!&AQfj(}ORiuD*y)
ze_K5MF$xoKS2VZfS<HkUHb-#RipUF%Cf{ao{x1DpZK<3P>?yzP{NM&P1gt5b{em*(
z>xyC~G+>+G2=~Pd@hfuP!fqStDf&bkwYty!um+BLu^-~zwLM0!{#A2W>|a?|-L6t!
z?|<WIS%6#I2eHAcUv#e=JAYGDa(*h3>v2uKOtEa(P?DD}A?bguDq8?=ye~YX^TbT|
z>O#(xVpYD=yYF6e3UbvuSG$#Traxlemk7Qs`4jl7_oyB?cB?b8<t%>?In;Yycwx^!
zM(D)z(?hdeN3Qy)`AzF6D^{NkPncWu4*xC4y=MHc@*hSeasfx8cNg9nl3s0-E?oo`
zCk<pBxX-xC4wr(nlebDXKYKwyIVJQG+NUeeET7~j_3gYP#xj5luMU5p`Ho}XwfY{f
zoA{-%s%mYS17&c{#%*D%EeNWoG?DEf7QcKd+ZJaJ)Inirvv!irNR26$bI%!L^~owr
zzW9s8Nw#DRb}R7wQ=RCFdv4gOvVhc=u5ER?Z>0+>lO80#vvrIeWP<O1?Gz6^uDPcT
z{R~^=n6x#H^5@D$@&*4S`8v%y$c~R3=_r4Lf&{6`!>m+I%jHQ9`74W?IYJ$avx1rm
zE35S~@{3mFIc`%Uoc1;{<q6j0#W%pQeN2&b0uKZnQ{@Oy0oW3U0in7Eml(sAb86<v
z2g}b4*=Ipzv=q9g+&boqy>Ff-YC6<(YfCDEwELH4#HMFQ!moZVs#1&E#KtB7NS5RT
zw0N9_WKB+jg4+}})`T(;kgJ-FZ2CphyI{(jVhYKd>nI8Hnh=Qn6<dQ_=*U5Mh~;80
zM58TTeCsF!zU}+i))7fQiQSiWQ-%2oR_3unoOwy<z?e15qdR3oho<)Go8G%GtX9jP
zg9!t)z}_T(X67$fn>jA|UDKn@mRyuq)Gl&T-_^$VJ?(<ESRAlpF*Q1&O2PQ0b6|t*
zfe)-Pz$w84Kt7->_J3<2u=^r{h^UuNPVy<J5I?MiPlMGyl6czkTk4XMd`sPd-vWDX
zGw(55zL}DhceU%a)W&Q0^MRQiDTw`DGr!Iwy}d<zoP~#-kTm#XXX2_mnP3t&{@vFj
zq3Dh2Ia{@KEe(WHJ|;vfVdxb1-<g@ecuPG_#d=#i3cSA<UM=cjoKUarFAKW<r2*OZ
z$o=Ba0+nZb7+ahk@3RVy;0(Qqo4+EyGQPNneb-84X~p@B{c4F!bPYWSC^#H8G1XSM
zIyh5!dq$yvPXI=VLUAA<G|Nm$fX%8`yDm_`mywO_2G}PT4}djo`*(4N#hv{X4^6t2
zPL+TuA>>v04RLos;fNTOs_$K(#L=VaOVDg5Y-!BUKw^}~h4yO{(cF8>)!N(k_wW8S
zC-lH4w(jOR;D%WaLdpr(+eLE616wbGP<EuIbsk_<AFO}@NH4u_6wQQPJ1!!~@=6*p
z{dfy_GmhQC_iicHpB|dFF`Y?+@?Gha3&-)r4`WTEvSFa&wf<0?d|>&w@Lc>)#;cHc
zd;SQ8u*IN0n2gY(f2K~DZShpB&F+1=DW__{Yj=v@qXuGdk=~VklmIXUNPvG#!-4{K
zu)U?*TdgdNkHCwh9}eGis9&To?;Br0pZz=F@ZBR6OxkdsIO#m2+GM8HJH%-oxw&ch
zxHzUULOYawWQm`>QFuZY(&DmO=ly(Rn5GE#LGhw2`w<Z0i&AneubH$_{Vd;2<t-?<
zx_U7os<@cY#$Qu@@vc{9xkd!`MMkcGHc`u7v{HskvF)IwTgYk%a7v9H(H;woxqe_g
zex?d*q(2f&-%c*_^4A%AcFlXS;qUvsP4!n&3x)5pVlw^Z7rOt|tE*Zre%t!=gq1fZ
zA~jiIy>c#v9u$JnnJkCn#yLhH{ZX_DTIE)|cVDZkn0$)M!pAjQG{t9x^r85tlXW-E
zQ9iC>PN%7@Y3xgvn1q<rMBbwJt9uq^($5P>6Axa5b@`nO-}WlE%T9ZYdKwsbgrGFH
zmsQO1<oI>viW7E+v0Jp>rF(X-yV)d<Xd$88qdLhq0^anYW;nP=4KwRsY_@_Om>O$x
zIiBt3Nu&!!E1lQTJvqE;5xjn4`?fNdur>if(bsAk?d$0hzjtt6wk4M(Uge#ZpE$$x
zwoiWBfRk0Kr;2v;oMbfefL%E&N#^rA9SXwlZW`Holhy^mG?5M);QRp@$dwl?0vW=2
zU6jr7r#V%u!w{5trq)4Ph1li9O?4jE;QxX8BHYXmZg>Y1`DHxXv0YIRuGjIBtZ#ZY
zwid_$gQ(0e&e`ZgWa`QhMTWr!ELzrxAkC9R%oZm>vexKt2@GO2hr2<00wx;|AGrDD
zG2TpFxkb3+UNmPFPORegObg$XlT|m#k4@uFJ$l^%g3c$}I0+T+TPTBuchP5|ATnmn
z7z=Ws9M^(htQPa?uVB^=$sAjff1-V)gN2rlijTIDBCo<hI^jZ2*|S=?*GzhTMtMFL
z$v=DZ-e+_5q}g3_gmV$9mZ<C_1z4TI9V_!>u7jF`O@ds-(gh4RyTrQA1W%QkPTQA7
z-nysY2RH7hQBv|>th@PA@ejhQH`nKu7xE@P1_<cS7H&H|hTjB~h!___Dj+Bc(f=GC
zg7wVU^SQ3Rf^0H+8Yr*R)UAoiiF%BV_}=8HV>0jfhT*COJN0kfr;(=OS6pAI@?Yyg
zSANv2arKfRE#ImQzo2eVh~Z6d{g|_a|M-h+5G&*<D`47*1OjbVHf+V?%LE|@q>l-Z
z)^)iZ=k%)=apY^`3(HJhxLc4G?PNXI*>Ov!wKaV>%A@gCM8jgjckyiScax*Kn?FTE
z@Sf%V!Y8^rqLxlh+dtjKiX4Z1Zcm4_dQ)&{j976Q2U|<%K56xjbsNGd>1vKrWqo<|
zLi_&N(u}7!fn$k_2kf>NY2PyjemX`aC^VREn7f&1pEl|57GXAyT`>(gEnXU-KC7aV
z_Z6Rep-|ibFpQeWGLy|^&Zf|1|J}~CS<UH$2pzhGc0%x_$kiH0jH~~H_DePWDqq{Q
zX8-w|gMM|hBwJP7xBhu**?NjMS89&aJGXn4zgH$zjT_(j6->Wgw%_t<&rWU3uP9^{
zrP@bO4&7waPZDYr(C*5U_)-l&I~)Y60j>N9n#7GI3@L*?(P5vEAj6~jf<PRfO_be`
zxh(N4tGwF?2R>CV#jlxh)L^PQ5{WmQljW}}zE2e+2Zz&9wa`=paf*^;ECo_!x(B>T
zkze;&7j~9#6h*HYv3Y#1$UqcvxA{+i>j6K$Ff-s&oi?Y#oq9r0r^Q8pe4ONLEy)Ev
zHYBneI#na^oT^_{jWKPWbq^(1a~)%IbxXt^p<Lv*?ybztuVR4M0B2roz*!;Ai(IZ^
z&2nEAi68PScxsl5gyzQ^1<@^SLi!S^sc9KjlA9$m90{Uo{NipADPwakld9rswredn
zNb$z^txPu`BYCX$Cm^C;)i=4Q!SKOIg#$}tY!6S?0q#m_r@}QdEAZ!jVhI>Iw=Z{H
z<wWdg+fl7}t#=6<Hp*2gI`Jb}1Xm}1_5iJQrGK}gV)Jl*<T9Uexh==9mntK*0<}8r
z7H$TfY9_bBSOG?s@{GCUS8Iwr%UkF;z3ZoOf4q`+WWQeKGvfD2xH+8U9la4>cIX;7
z0{`}R|6{6kOf57m@!P(oZsmBTscM3P4~RxnwRivItIb#EON#XKtf1mTBwp2MfapFy
zS{ddW51(;Fc{<&$xPIy9@K4ei!))^yuL;x5qSDhV?J>0CIZdxlog0m5#7toYbB`bU
zPF5OOyex}0H&q^Kvte|9U|8>SyzuGlRHA2{lydMOsUxyOMgIxbN`1|K6?)pShb8;a
z7%>sJ7*g7kc^Y4Eb;GZ?udnAO1+qYUxBb<2Jf9dgUp~RzF+bBld<s>m<<)dpnmFAW
zCcjE`uRoLfBt)6`hO^Ns+*-u8pS_5dD?a~|(cVIhn0GRexc{zEvvNm`c=jygMZSF9
zPtxX%=c%_xm$h>eT|4}y@5fOyzx(V(EznHD*#iYd8IdL5)usJywxIU_69M~>ts%&0
z0;Im+UEmG@#Z(r9Gns6<ImGl<1qljU6bTby!(P)ex{G3`D{9)9kU&!`0xwR?a0axR
zmjn_g^N~RDN7P|F$SbZ|e0xFz8naoJAWs4b{)6`7i-B_AV8oD*zmd6M{{sMF=CX#6
zA5IcH&MDXm@ihnO)g0DVFo(KEh}KIg;xY}S9#2su2uXMW#b=qoHzRe%T$Uii;F5px
zG0=VS-KAfHy}xePhh5;X64Rl?Ty5`|K6dHL{OdhpNA)fY@<endtsWOiMCIi@MV$sJ
zB>w*V3ajJ0h{nxf>|p-yJ0V7|bB^ikKI_p<3V4F%gtAe5Qve1!{eu+>L2O%Mk+S>N
zK=&r*Ur*WJM@r|x2HSwgBtWJCsv#Q*IIKkfR~7+A2w<400hBFBoxUiKH{D+b40{Sj
zVNx85LxpcGCdVg?@z3=}$A;&fnFg=l$%{EMc=IZno`wXR6;#)rcNv@qT+PAOS}{Be
zH-zX@z6bc<O{Z?%Y|UP}vm?DS_Gaw#<%n(Pj@J`m8`*nt64nt%>ejWx$uEgx{Y9$b
z{+VpU|4{XUt%D4k?C`27ZG4Z>Ba@rZX76mABYvPPz^U)6iWQ%T3FwY8SZuZQDSSxw
zO2J}2@+DPC;XiQoa3_HD8R9Z@fVhKTsRV`YhksQJcJLSghg5vR*pK@UgN^Xy3D7Wy
z5gnlhGQd$6EX9GaZ4CJ~33S)oT3YurPV*s18usz|Y%$$8$(90fzd=R*4-~iUe9v_!
zVHf>b<KZ2QFo3yLLpW%#QMg3qrEJ%`K;%=c%E=nTpQH=_xe>DUx@}DcWbP$8TDN~L
zlO<YRqy+l6bJkMhv2ChDRBr;!=E|7D^`~}hH1YJc92nW<^sw5cFTCu=RAn-_hi)K#
zeUPBKO<TvGe%HNN9B{|nOI>Z0e?~{`HcR);%kWr%e$|pM{3V|GS|FF~NF%((Q*p;S
z9X#u8v!XD0%8=>fJ{u}A=Hj(sZ|3p!OAa2({`zFt(Rr`}*x=lfx<|is3!O*js)3mR
z{nHdxmh5+f)wH1NQZP3hA#&aX^|!=q4i{Vf`#P0#{Yow3skL2M4>%o1bOe2AlWX>o
ztdZfAY`j{~+@*_St>+#UUMU<YFP|)G&I}1VU-N+!^n_X(Ulczu@F9&F+<?C^TFIGF
z*IB+|Y20_0`>6T#l?`@DR!ku^vi@n+xqDV0f+^yzVomNHpUZ?0h`P)sJynt`%b<mA
zF3Zv;c<*IW?!VXO+)TzRGj8E?Tm4kl3?u3kGKupOQWp6%&KYhl^}>>H+t=C5vZAJn
zQ~*?cPPeN+#&>r8j!Mt9+pO;1CizPI+wyn_)yu6s6+{ob5F1*<Y`yt5z+BI;;%GH8
zVV#AsIHE>-c9`EC?A)L8$Hb3P9@PDQcN#wOVAeJ4fZUoK<=Tm$s*1TuJljamvNH<N
z_k7x$@gMfm8Kb!w>aM0((yPFtKF{1@-z=aYUd?sRxEcoVv6^&WxPe@47w6>6xH?a9
zL!l3Z$q-V6in3U$gM~I7p1X@uhxcV4B|A%f6bZbYWy<aR2sle_zUBNy^Q5{1WB)xX
zgnOyPxjrEB`(^D#-IIVIC04o5kDP~&-mU<oE3CUdqUTCYX+9}ev*)CJIa!|}Gdw;4
z=sU@x(ENe=SJ%;<QtSvBDO?+0l<lxd;rwQ?z4UQ0-;f0k1RZd=(oxY%X8%5f)1E@J
z%MPP8;@Jr2GAj!11cNP?KB3G3gb#O(%ofSPNLlDJqFNjZs%9{cc7Z?zkz3M_a25jf
zmrLTGv^uWIPqN_<n-aY_cBaUoaf(?gj{mk)IS_8r_~)$co6>#`+u?Qt9cW&-xrra(
z8U_?+^-&~{-4XDwD*^}NK>)sAli*_pGHJGam8kjPvhH|+VUB^5jm@i;-aJ<L$9O4=
zkY~(KMXMJAjj@zPZ{U14McrJ#KiHadSq6<J3(cK-p4w5{Q*u9UxrKWE8DeF#tG+jH
zkyCSq2HQ?qXvuBhV}spIZOsk7PLrYsfk30%CUhA#<&Vl>?(gK|ZQ!AZA&wU46OrOn
z%6~NwD1<WqH|J{Lz=qMNf#CASa_iLxFyc1P-gsFSJp48M53{OBEbLLJzxv%nX{-94
zcU>uG^dv!cTvV^F`~H%b;IdvVp2X_gF|Ft)3UeCh556Jo-)HU|G7??7bDaZ76DisU
z;s-8;7H4OtDYZq-_XPNr52Scv&UGd)mi;?SP%S7KAF*1WzFc2c!Xu8T(}{b_TX&$s
zz2lcOsS-a^wpuE#jE&V~0@8uOgiV6bpXfy=v7R;RIdcf;8D<3oLL60RI%rm<Ao+(o
zb<*JASngL5$(GCY;4LMp49sph;8k`3yQ2+<JcFSR1cz74yR(CV@*_rJOR|GVDf>%{
ztq0ZES#su!8OZ?cu3YTVYGo81%O*UP$m}pDEmph8ig(9Ou-_B@`c3B=@Z95q@(V&&
z;wc&vwm9jN0%1}SN4e?M!vfir2zso9N=i^#FIxvC(Cvfb>|W98X$@^+5J9|Yqx<iE
z;k9i~@mc^OfD(gHttIkpCen8gXT#{&3q+X7N&UQ8I|-nrfIaJK)j-buoJzzG`a@Aa
zT;9<=<UHJRUcxIr&V`&-b^0n=&)*c)Xel!SulR6>FOA_6WN`iz)>T(?!6zwy*CIKx
zheUs}5c(bvPs{Zx*)Ao{q9W#NBc$~ju6f9`Na?>ZoJ<cjjd(+{`I4-{&jH=J2-4d{
z;Rq6WXO(AJD_ALQi$vC7OIci9bCO0Guk=9BOP(Uz{*dannK_{|qEi@GS9RguJGI#h
z#?LfV<qokN=?~OAOwR`86=ta!V~Ex=^Wb4Vm73c6SHAJ1lOHm93(rpYM7rvI^vRtU
zq8xfxJeJkP^?XHaHh0yL{HZ<8QuqhgU^xNbw>>O4dZoI`*)a>%WHH|y{M;+e&h~0T
z{0nIDK>~egs!+b}k4UC#Xug586+PN3JPos)`xontiO0jDe<gOb1kq#udF@)qjrK-o
zkb2iT%LN`!qAIltI@GBonHAo<^Nbki%aWX_7zx-3p=mTmgNJ9TpZBbaj#sQ2?q}{F
z855fEzEa>IOp!`ggt_K3G8Fqrrjj6|hJn^Yy?m0LSJQ++j(xy`Ay7j`K#jel)$8qA
zi=EBQ#{rd*XQ{aVW|B`LsnFsNLj268DuVLWmRJXsr)Y3x>$R6QdzND=XS7;>%{JR>
z%xnpy!G|t$Rn{nueLXc8AIf{bXuaMoXt-0$domHFL|i}Fd)>Q(JM_4i?qR9Z7dt!o
z#=?c(o0T089k){y7wtUAl+Jv<E;6BLl;})gI!kw+)8Mz|4zb~od%kx6lXK347P5<Y
zeS{|!h~u2A5JcO8F#!vA8aP*5@B>O%G^hAargWJD_=jIk5|RCa(p?}w6t|5wRsjk2
zD{_Ejq0(H?SC?r<CZAAOiJE0h=?+l;oox_ZGh6Z}=7<bBv`QKP64c=^DlXl5>q_BF
zeB?B05jjeEi+aO?&yl7FGd0AWJ~{uTebfJ<dz4c|XZ_4gx~xH#+&#roBrkwO=CxU2
zfJrtCHW3>|g2<JQVV4+{0Z}7!HpG+=jz3VgLb$kiD|?RfVS^53>|6n!#^@_wqK)N9
z**T`-f5;Hb<9q|-zveKULh4LiRG+7I*7k^9F6Z~P{?pdg;MXmj7(-pzGTJR6Rhqom
zqA5$y<*S{t{Ofx?SbN*TNagCK6L<hGC)X2r@vtJnNE$#4)BjstQ`m_AS6%>YuK_4$
z0U+E924~+p^1D(f__mtAw{7cpHS$2U>xxUwsfh3I^MKFDc;*Ukf~-Zbu*K@<{Efj=
zdbRJc99IJ7Z|oVV+=}cn&(AJkWlHzMtayZ1qW7H-0?&j_>I*vwYjf^w9|SCJWxr%L
zt~+-}DIUP@>{K`Eo-hOmQ7BYZZgLc{<!mS9EL74jN%%WO{tXo*a~^F*^hX+PzZ9d(
zW-uJfqRFAY@EeLroWY7o_m#kPY%j{R?n{DAa#X$$K~J)VX=)mhoVS^uVW16#BWPPV
z2Sn~-)C@7)xKKt?0B5qDkL2Z%E|k;c%m<$+NzZ2Ez+fdrjmGXmJy;-oF3vtbU|uQ3
z74Tp*%Xm1adJ%_u+XDO?B3$H^$15G`osu=JFu{hH6qJ8jzk5cPfFL{-PVh9kvs?Rl
z|Bw7vuZQ?L11-ho>;jEkEtUQxp<Ubaxa!{_u~Dz4BQm%pdGoY1?a8U#PA=~|ol$;7
z2Q6ENG-WaPu^5z~U&aKEHZ8RdGLjB70U<(Pd%VK0X-VhVzgl}cFWli-ZPWaRos@_2
z=JwHdMK|Qb3sq&WX^lUXoC~*0*vBgR%w_ZhZuLj@^>@E*7Q27qzK#hoUG-m72OKb1
z3u#LWNl~3C`QUzD^-}j_eJ5^u5FS*9N%`Rf{#$|{1Se@PVvyIbANw%<`h-&CS$Qst
z_arW}4Lu4vyXEWj{yMeK=lp=%#W(P#;PTululLq^#=2)3l0%%E(w_#FMra56jVx_r
zUHg2E;U$V~WqE-YqA&*TRW>;wQNA}{OnEEotVWBLh+5vynr=f-wf+IDdiV0l#+MST
zdF<TOYduF`k@$t8l&4yes2$K0DRKNT|MFw869wbTtxb1;aE;fs)ZF=sQ<o!o#5VhG
zyIw)^&qkncizq6zsV|<^bf!K{l}V7qpg)$De3ux|6VCjtYIM*XXsjNB8n!iEy+ApX
z?6UPvLo10x(Kz1C^5?~8#YL+{%kX+^Em1wE#T>(twWxP;h`PWR-t$v|3HC-R!D^o^
zMU`sG{F<^J#0bCQKnFnT8$Bgw;^O3Rvj7Y2ZlVNVS?U?Jxq773N5!?C#xpG^Gb#_U
zb&dH^;oG^;U&#&6bErKd4I8q~a*t4JDs~~dtAZE0MmX^bBsKJ=`D<Y-Sk}NvpUau<
z;Pd{``)+3=9j|Q;pzpW3UxJtz%Zq)C@SJ&MrjpxLO)0$Hz2RUq$^~L&hzpk#lOWo2
z3bH8|e~9bU<8HN@X>2wNKV_TLWIfXR$*K~>+C5|X%az9|Ni~!T3?DWHH4RzL)WZAk
zBe^3;`nMfGNGW}iY5X>lr_2#~uey>9`nmaNH`4*k_$6WE0Ex(+llw+Cn}lg5%>qxl
zKS$4B&47mVF^9u6kYq(o&KCeUNwiBb29AKV5Xa3kFMCjLl*m-sYI2V^W3$eEs8xH#
zU$JVT;F{z#$=_7O5)A@xmXkoHIKh69cr_f9T5Jn?rC3U0w=}ehf|x6FxVv#Q%#YSC
z70(9>E?t@Z&;t+q5#=P1ykH^o#O94>$R)6CwBw@-rq}2xR@ZXbfcke#H5z#F?2pQe
zvhr7u_hqckJ~7%W{@fJn(W%ZTU3_ulDBZyLe$cqd8|d?A&IIz^=(OxY8TKt;eg!d%
zhTmJOBMVwmfrIxL(ThOd2&jD>|4%yVh$JxoPZ^{(XjuUvaf(l50=z8}LFVSGKkGDM
z`X^?Jr{KC1avFYSdhm*4(ZMjywkyw-^FWcISJREgfRk=l6lOHpNSl@eAKiS73w~Kd
z^S_u{3csF_UYGp4n+)9FE2uQXfYhsItjH`zl+67u$#ybv2ArB4ELK8Sw>LrgNfOP<
zq5$PkRtKnuGMvb%JVGXdK$~SS!g6W>fI{cu#Ul02aY0Y|{y_QCpwEJGcm?4fa`ea?
zi?GNm1)z<avu^>D8^Q0d6U0Zont!NpHg2iv8<3={Tzs`s{EkYd)qklhS5<4K{u=BE
zIu%K!lMo6ZRjm-yG-8VT)4~ze7v$T#r=+!=WR#LdDI7g}^J_-7020LC!qsHR+)H*g
z!x{B4O!%I^(T-||?swPk#*5*CYuI+~e=TM|LQ>`rQ3OtSK0}OUNNW1Go7s_3D%lSk
zU#>H$5P<E(Q^3pR3xu;b-AJ@RRV@C4<zi`y$J|3W_A+^2zA}#Y<;D)|Y2i~h4$|zt
zt;Pby`7|~quk)FTPVc?##T4I6A7XXG!ta{t7Oj`L-e>zX(9G81McEfz$%qD);c`RN
zVSwvtU31U**KYRWD=9ANBTZRXp}<x~**FWNlxWNgVx*N`J!Vg_)p-^xayr}AA?uQn
z%#0aiL`-LVadHOz+FpFM)f%(@Q^@}@_MbqXujQ{b(e{O2)=qa>bUCjwKF5T=`P)7N
zZz1~<(q5)2%@voy8{X-1PDw2<q*C44G%k?#7D#Ip6Zp+;*vMG%Jz?Kfk<`$ei23G*
zOSsuqzf~=0=3&F6gl4GAS*K(5^fj8ITbx5L<BQu>^H&jWWiB3_h*e1;dn(55LdheE
zd+mToB1=uNY-3cfaT**cr8FDV;X4EMSra{7A9mkw!Na1zAv(*AESg#VUF8)SY=PtR
zZ>bwy?ieaB=3pypaz3gF*%v!BOsj=1O%8ARn@CO&wT!k0m<nE`aDSvA{}lQy!+!}D
zsqflBSIbUsIjYGsQZ)k^*o!`Dc+2Cz%w><Z4fgUbdAZvcvmV;^?(_?Ry|{WcC0q2A
zf}it;RFzPNi}79(!z>#3-uSj@Y+j(%H6@Mp_1PO0o%4~r95zJL$BDthH^<(U?=w4q
zJkc}J9DYNR&B~0BWA-e;e+@T@&;J_XJ}V(l7cG9doL>jb)qXK=u#&*gd%uH0Q#HYp
zi4Gb&3Dm7lHW8?Z4Jc;9$OQ5n#vMO=w_y2SfB><}D&y+H5&W6J&$Ig%BV9HF7DR{q
ztq+R!(h!{sp1F@S08GlB<%VjT<`1zA&|yvKV~#T}bxQD|dq7nJ{L5a`Wei9#ARl1{
z3k@dJ<eQ1z0RGbE6)S_MbJ$Y~t{fGuBH~|nx%CXa$?9EwI-&d7>m7w~NSu3Kr7IV{
zV}v)mRt}3Y*Ve45U&>aks#x(^7VP&r9{IgZ3)#!O%b454n=#zvgE9o@mmD3&q2Wd!
zCs3!D`P(SAN5H|YrpY!+(1I#j0&F*0fdAn@%L;V$9S$1*0h)vWj5UlOB4H7H3II-;
zAyX5q%J#*NsYA+Oz*Wi*&zv0zxO|$sGDxX2N@N6Ux|c{AnY@Lej7SE!I8cn?LTdk3
zse1h_65;IQrPNCn)~XK;YEnRm0~q=qNUlrqF9y3!!Y~#G1r>`QQ^<Z2`n;U%>li)^
zXsuJ~P92z=3I&yHK_xh)a1kUQrMm?^WR5@Qsh50rQ5>gOkII6QGTsB|5h^Ev?T)#+
zWvBzFDT0+8#Uh-?k5q(5>Ur^6$7dc^fmx(QfvP*oW;HP0E55`H!*<EnX|0X9Fptuo
zx4Alzc&oEq#5)3UU$o?UC*0z@C52SPBUBc#vXZ<tVV~on)3q93LWUyT_SC=yr#WGD
ziy2d<>+i1aK=~k^3x93P<>;&3%)Wj0LqzLspV`D*xAutHq}(44h=c3NzE?_bX{;B3
ze&YrXNd-<~i6p}RKuvNT7NL<}(EJW5lFkp9IlPI?q^}IfBdXIZw8<}U^5eV?$BPBj
za#s)WU_#01g{am^=rawSff>t4Ke6Aiv~bE<;R5}Y*)Y(^K*PQLO6?j-w)XXQ-R_G6
zQcY3r!mX*>7j{!8aB5bl-c@er;+y%D2NQl3wmw~OiQw-wY{1bri*TBtxlB;^)b$Aj
zIEx1VV3!Se(j#H|6)LZEW7Tes*!pXg&y{sU+^lDwF%=?yxPA~?Q1l|MH!8kY(IK|=
z7S`>adlaDbrO7gPBkZ-64h<1WGlzM(n9lhjTJ)gn6FAeoawyn(V6)BObPMzCifeOL
zBx~kFd+X2N_yCpCb57S(vTXwwcgAXYd31yy=w+U$-&7YI<K`?<yJD&LR~hS3O9?fs
z#$O!EOrCIXjeLwmwTZe>)q)&?^5fv3``gqBkY=0tMP+-MhI!H{3CEkanLWd_9;xkB
z4o-s9`V*>h1~A?ItMiEZ^$n9C*_s?d+>X8-A!(2fWQ(ubehl7|E|?M9$Ig$f*<33a
zl~WVHj&(g%{B`hNF%@Fe?ln|><M*Sx-9nj`u$ROKXHv)Cs4~8)p1R)ePW!h|FK!1`
zw^X!L6*-)1YouDYA=#Lxa@^Vb*9|I;BbJjWW4;FB6-3W-9!)4F98qMWNCpfx#7M(w
z5Kj|cwn{z9I@8rm=#Z$#Nf1Fm(`JYu<ehDM=qp8Y@m(T7N$!XPKwYrfCs-c?ak^m?
z$~6Ck>5~{l5U5>v1P2A&b^z{~MHJ#FO1+30wy1fgKZ497n?9;iw@X?|`F^n~Uqb(m
ze%Kdc-bde*?EW47qKh{+pnp0B(1KVH8u(R+f>3T>1pY4Y!M{@Q63Jg9l7f7ZrH$BP
zCP{uG!at8)^1GIJ#kkz%^TtW)^$QRA?u3{(U5#zAO4i$T6Yb{R9i05Lt;UNle`Dr-
z^j%@_*&lhY{rAhX*7LlYzpe&zSU<$C>X@|>9V>8J_lKq0t%*4xm>oBiVB=s(g9sQz
zQ)IUED6bL~_Z*+M>$KdezJqdiIQl%8bN|0?8U%=b`LIBj2>3rh&@>V!RD%7&h5r<t
z7{_oiTa7M$o8s4Zf_4vH)jm;d3)G_ry$(2*8Re2>j!T@}PII(6?eA^9wRm5$VRl;f
z6Y^~&`+9_Z4v}NFUOv&8t0GujSdu9S^y38u{5+Ol$&~_{=4=85kDX8)n|VouUF%E+
zu<eoRsR17WrRA(Rl<WX$r$hQ>(~(BX|IPKp6oH3p7r??p{-YBbA1m~el)62EIDls_
z?~Bs@Z9~u#{?v<EpSW0}FV(?qYwS4XCvb=<ovJy>_G_Q#va56TDrsO*#T^wkt(U|C
zk3bO>K!^Y{U(1|(HD;P{XSC5Ziref)V}x6zEw>`>EX2^ID2tN5#jN9)Rs^nSm6|XS
zuKcWJZRWjBm0OP62w&1EwkX%!d>=8Vo_!nB6iod&96xw7>m`6*BSzVhG!-tURYLi1
zT2|WWS_N!vX7otUk#pO+v19m&^2--4%zmCtcy;FkX*?IMqg$fVc10thrYzY=m6H@S
z9jC3cYyNa*G3oG$+_P^RXgw}g!|CVut%6H#Ir{y)>isRry%F%lx(nIE7w<wrkYT6@
zvzTWR7f<G)O*uLRxPG3uvH51myun3m<#|yj*M8WXoSjC>2!-D?%$<LCNC+wKe-4nM
z$~&jHYfehSj$kXD){+XxjW_A0lCL{*IiP>^d569miP6p7ZD)u4f`#Mky0mrK@}joA
zrMlBL${^i7l-DZOo)Ql}j1&oZiAyGuhMrFsDvWqR9Ny;SY=10oSQvc3aMttU;p9R)
z;4LJ7IuRxy{HYJs>cE3+iOh6OAoB>E;kz?bqbVE&4j7d2lI;-@q-7M$?!dP0n2Quy
zs4dY-+V%GC<Nk1p*_}G%q%a{`38XuN*|IC~fe-)kxg`boHK`)*&(x`bT$S~3Pq95_
zmfoabiP%T;2saFz*$#~M3s_Ah{k*T0hiRdJ8t}|rJz9ExfRA5ive2Km6vZaL=v{L*
zzlb;e;$-A-X26`@kk~`)Sk23;N=dBzWW6S|$d%a=IH<focvz3jUQN>wc}L{>iz)O0
zXm%W;W~l4}OE5w;8*s8I^w<fFlp}x;bR{4HwCK1@5Rp1Wnbd$8i4I_;ieQMpb({ZT
zOCvV(E()W+O6|AEQj@;(F2XZP{o_MszH+0;+QY$G&K$<Q>53y6Py)FEM+i!eC!2ZF
z5YozE?Fl#oVlOHZL-rFv*|NzOVo()&oGwG$%@R9GGsmY>lGEgx1<Bd(Y@T|SI>&^5
zNx`;^M7UXa<@e}BGGrggHF#yK&e!&q+>h6PqZjw*8EgJwlDKiYpFnZFn|QNga2mKm
z0B3>sFv*Oe)(esO9A~zXgPD9W6p-d}&M%Bl>`=t6?PUn79Uf_B%aGc>!ZO00Do$KX
zVa$G+_=Jk<4Lk?`(*eRTz~-_K{u{%~{vSUFnjiFE6-gyX;CSkN%;Z<=ry@F>7oB1)
zjiX>iK~X=PJ=>4!H6-iv9q^bF-Km{;D3{Zj32VKmH_IKCXx_l|dAaY3CGr6Q47v7v
zW_74P(M(Pu4nBkA`N8x{s#G>oRz~XsAvncEj8IT^uvicE1sTGbc^5ihs%ib$4%mmt
zJ*tF)J{vF+oiok$fise2BFKwL1UYa~x$wtiHqA#%SZwsr!q>Cc*Fq(5^TB(a;;eK2
z@{;_$%~HajnwK7a$-|30ug(?E*njc}0@3Me64xdz@NHGBk33eiHB(w{2#_lu`w-<)
zjj^>>?jdlnJuc&=+>oXIR+;GFmtUWoFA6c0zi+n}MEYKuJ=>^2t3nNZ-Mskuy{@W7
z^up%-Uds0~7f$eA5v~D4q;ktKSb8{J55Q=?{BI#}{yaLd0;efgz%BFUw9SXW6MI%?
zt+ZEKMW@J*><3F;oSR-L4jtM37I?q=iaJW-yU8lEaa`5FZt=#c%E8zRxo6zdon|Et
zv;3s;wQ>(R<$9gN<Xh2S&IG^hSf!kQO84Cdn!qCVH7FA}E`pZ#)_`DoM`WI`pR>ce
zo@Tz92Gg!GQ=)NJ%IbGR1F}^DTc%5B-tg|5DF`Rxg!#oelB?O$BiSxP4e|P}dmmEy
z{Wf9~q1EHUry^L@L6j^{#;uFtsv%a48_#^c^0yMtwU$u>;`A-jIQj_bs^six`D;i~
zgR*9G>)oD<jP>ND`;l+l!VK?cgOtck>wDZOdh0XEt=kwa^Nfes0wfMBlUi_<1Jd)9
zLw117SptozTiD==FmKl4xm9h~#9<Kw-8rY@l@wvoc=kK}(zhnFMyGF~>Rc{q%^!#W
zq;>5-AL+W(NOq@?q5hxf$L}KId{1W9PZ8e8{`qaBQ&+r#s2w@0FU7g47on>0t(owi
zK9l$-CSqDx*{VNHZSTg9v@QFyA-i_#h}OXkt6j{+ftj8rgK%Hs&|JrCo^y$%1rVR(
z9-Rxlhe;r6w5bcSp3Pvp&U9xtXi(N&Fo}i6_a+pa3WB=zCeCq(UzVPW?aCgw0RK}k
zAZTPRJB0=D!?p*=rZGC$kADqT-R2kVJF2b~CA>1e58tZWMsjK|XTvh>j{I`&4I46#
zp`f5g+cv(eM}XLyP1HhvX7q>x;_8OqK#)Px!;+f>$T2*X06}uW2O_UBFGJ)ynvTBi
zof0T$PLij#c5+t}pFHstx%s+36@QC1?0?zx@h0z5RxEeiFiC7I6u8_H)i73*t<;`S
zhJ&&rY?qLySOwCsbR0Y=DHZ4s*kxC%LK*BgnqfROve{AYSL8cSo3=k$wi9HgAFy8$
zfA&@tIlN&!+4tkRD8@DJ)JIP%-N*h`kyo(C;cc5lIhMvHF2$BhB3d^@-=+T$ezc$p
zlUWm$G14Zj1f?WNnH$S*V?csZz2x(F)3f2wB*GEJAa(C>8)ZcYhqEU#1N^A1)q$Om
z`B_T`y<LyDc^b=rsQ`cno{@gE$GzFn^y%{V%4~zuGhYurxj`AhU1x}#Y5-~m1<(!t
z<9ZQJ15zI=B;fS{^J~nus7^3g=7*x3mBLQ`7qT=eBAB`97oU1I<Fh*zH!=l&<Gey5
z`&)RcXqnx11w-}{D^3a&4&1p46<s|jU*;hR2aMVnKyHG2i)h!u=K#~N1;MT<1BVST
z8-W5M4?NYHm+K}j8yN|(3l|v?HYR1vBqZm58fFvvC6{f?&;>C-Yboi-B=AQ5oB^Ax
zR{sy=0qiZ@&tC#z7J5T|Jn8}h6^xKKx1!`0=;e)B!#`b6LP)Tr^sz9b#x@SrlQv;j
z)$q<XZGKVM^FsY3E4A*gA-%eD`L3_uV!o3_v05M?;C7t91%ZDZHk6=T&G5B(ZjWbJ
z-SgkrbwyVAqvt0)Ql%QFu++K|sj54Vy=|g%q*br&`QKt@*W}H4g|a(WbXIBkTPn;B
z7N=pJh*KbM03tPSEp+8B9@O|3C2M<5z&u2iHbK$OUI}WiXDmomAzi{@d2;#3b)E%d
zQ*32`h$8rNVv^e(?j843FIm+$%$`rD>7~OlVtP|sUdQsY+f>z$_L3nHwe@17*o|BA
znv43~qWL-jP{7_RyN!yl0Rbv=LY{o;X3eLHxei?sK8F5Wx~X2RDpV$$yqQ$+3Ht2*
zm+!rluI`H#A-vf`1#7Xl+&yl-kSnFlr^ni)q`bTnQfOqi92V*)!Xh;=C6y;abtjfY
zP+F5TnUTp5MpILs7!t%mzo8ff5T%K>k*vwJvGuk3quT2~k{3PGvRy7F^36rIsm*Q$
z_1#WWA6A$V7XRgD*ex{<1D#K-b!N*E5R(elsv`&3#y)s2dd=J7!p>^Fd23-Hi|qKP
z#YMdH$ME7PO;iJBO`B5s*&)~1xwKc?^V3C{qH*SX-oujO*-?LEQ(J@k8GpAa`j!uK
zrc~Sfolzgfr94w@y;Qnw`R!uy&TQ}S*!lTvd%b##wA@Ks$+eBE8I>--1O*x=D^nAw
z6A{#)*fV{S;aAz3xvBY<{8R8UkG@xU(*-lwhs8VmnjPbmO@j<@we$7PlqG&4yt0Pk
zq>;JG(vI%_qlbyVoG=I0wdmWy?hm|?F9Kc7-ph5J2_%6`##&aJx2yAL;!cT$I?AjN
z$PP!sqxAg;tv>paGik2^)%>FPMcN}(mik*mL0qi9+$2itJ%^NTtkVn?Pq=9W)0_aw
zN+0<g?ju0}Ce&YgcEE&1?px1vw)Zjjy3N$pcZCQ&WH5nek3LDzNUuf{h*rz;EI*<G
zpi>EdzL`{@lZs4Xl|RIS%q$T91M!-I^WRX}qmcA}aXrUm4#~N68Qi&pG70vHaRX4p
z73`bjH#ZSYdk;L_8CQ9GL|uOxYiZpJZ~>3J{paoR+4M}%qFb21;tz|&8UKJdWEyAg
z^(2h!+iEzUX(WuykR42g5BLHlj|f|%vOyd=L04x7ap^ci=g6w@a~5kVFF6T_7thBr
zs!Qh|Cmw{>c=)<X(Y?BQ<Si6O7I5BeJw}aBIE)hqi-vSQuCvs}{v75m^b`9A=m(DZ
z$>tiGPxL(t45Jt`89E?wI46*dII1?<1_^H1ZchgL$b`mFls-=c%2tiD@oVIhVuecJ
zdqFuo`4uYrWppnercZRNI!c`8J~fVFojVO60siv~0*k&uB0Cgt?81STi+~2)Xtr|s
zNQ+(4AS=AN!Qjv_1SoRKvs$_^?bCcKa|VZ-UJVe2bG8964FkDQy`;@G^P)xBYnl0M
z?UMO6@_KX`p~`KdY1;olG#lS`w)|kSZ$}*5Cu|DJhl7_4n;?0IiUom>0VOcX#vuw7
zkp9j}$E&Or4(eTfhzEhhYooVqAc_SMqZCNH-PE`hp@c0To+=59<Fu{IctKp~(DS<w
zX&M(5A{VXNE+~*#EsPBh_B_d!ZMxzfk=Tn5KEtV}=g(?D&2P4w3$*vl|2lursdd1I
zk~a2YeTlOW6)m|a;?FzOgaV|+l@DM`S@SxmF1zUkYL;MT;1GTQBtbE|u=Ey>`j8-?
z*Pt-l*`Zt=zS+#(c(utIhCko?yzTLHEB1SR-xaPem>t|VbgNHLjUR}{S3J8}BPkII
zqzBMQ)Y5&KU8JBR38M_Ye_0+fi4PM#<YR@+30N|#YU&n>UYE(=8a3q2eG&XspO{?2
zJ&LH;{p;CGfatu~yGYhqV!^5CzU$wy1#{C>C7d`Z@aw6FGf`8boUJ$8YJu0^^~AN?
z<#+flt!`!pFi>ep^a~mgV4n=hr`hR8VFzqr5m;HUl=3h$DcxU??ma`hJUWAD3JJYs
z-ODf+%Lvz>3$tZsovzV?$o1+MRdnlde@F;ec;49>640u?sz>vbT#wP0x$5vLTxGJo
z`2zLn#T$7VS2}sTOO6nSv0nq{^XwHXFO;OT{&cOeEf+12LpfIz-f?seSIs3=E~&tj
zoTYGlQ$4dT+C}t$i6(<ql$ppbcJNdz+;!sk#d;wZvjYA!ZKM3*8rEAb!++ZS_W^z^
zxIr^V#f#XAgVzmzo=ZRKkZQcFp++St|LS-WnvsmG-||x;LRAOE>J9JHRWt3|Ysn4F
zy^NN5C}aA3sbC$|!|xcnXP|X8Pge%U*~c$858l*h%wEgGU)GJ2K}>Mub?P+@pP<J{
z7Wcv`s}f4i7u0rY?&tOqUDeQZ>ntrjE@i<Fn73EuG6&+{&1|DtL2Y$g=fvL}_CL!*
zzEpXs0wn$J8ECcqT7p@)77`1&MqhX7jfQ?hIcJ-rDxJgFo>qlhd8DI;Ig5wonoOAa
zhJayfSdL@|%LGf+Elp^0THM(=^)mg1V%2Rn6)@U4Mu#v|0qfv1*|nak#j-0J1^<Z#
zE!Py>Fl2xSsJqL>^q(XIl)KQ9#+8C%AF_w~C4tv!5ERMZ%3}RS%1Xv)5Q*ESnG-?V
zJ!R;#MuqV-mbP3K1$O5eDz;plbX~{ZC^N`|xi#H|t`*ImE#gjehBZ=qLi6lRXLwM(
z#Vf`@F(wzqNwmAiVB5^xEEX=>fl&dNVs#KOUx8+3tDM<I3L9fH9s-{B$7q}(gUe{u
zAXI-b>BWhy_vekbdGE)AHtaYqH<@l(iG1dF_Yn)Ys}sF3HX~kmmukoFcP(7|x;YGn
z>77!0T^rg3t8AReIWUyuyFoL&>{WqB>K#v<>-kR6SgJTG86}%)3;z$qB9}oX6d_i>
zgQp}}7zrKmC>M&^tlQNO)KQy<B<!n7b6c>q-<r-vfrs|~NuG|9Nhszyk)TY5!+%}n
z|2yh}5daP%P|nb1D0V<qr`dzC$8YLbJ$pMYAR1qvD>7fB;rv^rBaM{m2L_vN5%lN8
z(#o~bmP|_UQ;C&q6)QUy9U|5XA3aMZ2Pp8x)UW{)uYRiuoh%vgXKS1!dqaspa5Z8z
zDf1=_Oty(I6wZbb_!wFa1AI<)1{kr@Y<^3CG7@1C42}dw8xhQ6uKr7Kf|o@2D6|+%
z*}?H~b*dWx=U6;4$F=@|$ERlCly^h+Z(P#bA8~eWuDNLzv<Kg;gY5>u`%Y?Ea*^M)
z&iD<}Fu4t!mXFb`P)ePLfF%vb<)?J6k~yRyZfnJU{}ESbXw;ZO`|z(S`&v<=mO>~q
zpei`hz0<?RZ+2P>uC~u<v9{1r>Fx(X&@kFcA?dRbM7}j6OSw85YJJP?K%WFN(z{Tr
zP}=VeSE>R|@E)ZOa#u%Hu&Ber21U1CPL+*XC=<{`3NE1p;h^JCOnybM4?^Mz<W8v@
z(e8oL$efzDHu`Vj`tVIXV$$qP&Pq;*ys0k83iH_#`c^q>R+2dQMbcwNN^>Z_n!NT!
zzHIvFP{Gf2N;Gk2Df8l4W)J1q0Hc?1HZ;?SUv@phCby>4hU17T-w%>2lPCv>5ZG5X
zvBAK&PDlhPmqZj`arQ}_^&fn#nd{*)zaVJgS=a?3(R#eDen1)e^<H#&y?fi75SV7P
zQ&d^6%&BZ=58QbMbVc`BpWpjYYMhYJ^ia46S7lNr5Q6yU?u$$JD4C~?i=$j_Uef(C
z`|Vrv{rVEt$F_tbU@UZ9Ey*i#y4~5Or^kJ^j3sqIv3e`4@EF}=SvU3>YaRKntjx9G
zjPP(~lIsS8;+>6n)B-(>;1zIpE=u=w=*HajCHc<zo`0o0+sSFB*nPSTru3O6b=ko)
zce%!~tp}|Jc%{fa#!q$fw59xIgGY=S<}~m7hYZjfwaSE-Ns9pe?th@*#~rkjxl)(j
z!q!dAt$uv;OqzBpW1Dr8az=sm)yr05&r-&jCeS;RRp6c#6;XDQn)80Ke)4BrfC~J*
zS=CGsj#gM&FB<w)HQ>k)Pn-#U^>p^`c;s|jSq#*&>FSczdF)jA;=a43GEClXa@C+X
zPL@klpyuwBP`TyOF4m|y(t_W4JxHK3@{z^mYh)kRm0EG3gp9d_CjG^4@>UIAnVX+H
zHw6slLXS54#rEP|T&O-uM%$giBc^w@OP|J&ZPVSekM7qKZdFGBu#kV`Nui6idOBM3
z6j!6*Ez$H>fRq9fVFKqin;~ffw+70J1hDlxs%COz;!Nu<GOd!U7E1Za1RLCd67na5
zh5A2Eb;_nWsGzE}bU~}d=el&0kZYp7VFCk)eTSu?2WC{n|HspN#<ThV@87XU?ItZT
zY7?tMqIQEKHnCT0lNc4WiqfEz6tzdHY9#iot<~DAwPzJovnZ`vt&Z<?_VfEcuAAas
za^{`a`*<D4^DsBhJSYKEz57whrQU88qKx$Io#!p4IVycP7NeE>qk8DCJ->;(OdWR_
zoPmha&&ON@1`S~E<rgGL^>Bj}fHwhEE>uix5yW=P-#DPO|DI6sO$fD<C0265&B{^J
zo-#VM3YyJhJJ;wJztD6^klLg(L^@hh)m^q@ZQTTutiFs$f7IH((Z&8OJYrs=0&XSp
z#_;4@vs1b9>v-g4OBoM$Anm$jW1K0s>%34QdG{zuAhYNN4CL_#7KBfS8`HF^V=yj=
z!w+Emjy;++9s(w!>-5M7#mCZ|v8d5NXbOKN`pq1BjW~mj%RNXtyJ0rg%nc_g=>YU!
z{6D8Fkc0w|)czm<Obx|KQmbXO%-Q@8<T1}*Zm3%VUtKHqvSz*;%p<g|QYOD_pT~4>
zON}G8I+^Il8HSHWQyL^i?uZSGudd>wQ9^{&S5`6`Bxb0d)r=L0;4-nzZz3o%D&%)W
za_4~e^L_(1h+Jld#%U-viIgTuGR-%3LpW-n{-=q&IUJCl={6>R0Mzf3VuFAwX1x|G
z4>g6+39@AtM-#wb*qT6<8kMIRzFRjBMB;dgz98<BXh5E31C_p)dAB9msf+!Q?l+QD
zo|H<JY%7!F#Z<y7wPky%_0+7hV)<%9`klGQXfxQ4gA1CIChreqCRrdX;%LOnmMW>h
zlM4K&E&S}w@Y?4yvpfr0c)0pSo1%+~)ONG|i3^Rw*WNF{_Z-gM^AMjEg6{5LkGsgP
z5g#X3@ukh7Y1mBy{AHF)`A!+l(G=Bo5rP<u8k`;!ju(1+R(av`nYMk@_wTUF3|J8q
zTFy_nogdv@tgRip((qkrS4SvFfll*M&@1BWhw#4iu<HG=3)mNFALvupJLH?Npw_fV
zUI=!HTI;Km;?lnU)hcbF4&^SEAh<PXjl|KrLc$g?Ga&EmKUljdK6vOdO`Q{d&o9Cz
zjpJqcM`FW;H%H`;rt6pQeRXi)-_E!Tc{1|b%ETG`?D68C)t1MxM*=Yw_X0kTd><6?
zd@?#*>(x0`_myi8zNC=(NKk-mtL(RJvuUD?NHxISRaJjKwa!ra4J)_$%Q1<~mf{{^
zxEYZ2vPRXM@#aZ}(zC$iY2+)fYOl%J&pbvPHQf-egB|yzd;){RX8NnQj(p2o<TEbr
z)#0-W8JRFv3DJ4IA#zB|+)#pY_5RJzi{;W|zKpx0BJISpqc>{bq%J*sT9x4-8j!BT
zQm`@k3Sc`x_0U*XX90pH3Mb`fVv}$But4fUe4jV+U8_ppbSOlE)?<#;w?6?b&2<WW
zzj0ih9ChgzAK}TTGDw9%{wU?)i?l(=)B<)G8iP8kaWbixHPkDvETf@L?y`FSENV`z
zg*W=Eu^*?tv=mZg>2rwfGOn58Fz7y5^II>Nbndp}IAdc&=XR!%gs3y%ea-(M`^H&~
ztlo8{0qg&Pyqa$@ZOxzIFCVEu_$xe6hcD9d=Oh{;O-;|23%!%uX;OXF)Yp+84V@;{
zjYGn4_4{bM=(#3ky-V9fuk{SmHR#Q!ZD)nF<M*cy5Z%mWK>glTvmFcJNqUv7CvPFS
z0~zBr61R6y3^--sXhuC4Ac^ybPk`8`a#Vm8T8nY&0{v|iHD4g?$9goH-g_l!V}hna
zK_M8$r586&-TcKr+yg@sKF3;8FdK5WLYD0_7FZ!lkSz95JW)ak24(1oXwvYvr8b5=
zqOYrBL&Cu=3;VMRF9M$Vpi~^C4WW-gb7YNjDyH(Yy9f{s+608)gXZy4=W;z-UP-H<
zx-erZ$;pZ6F~hmQKae(>x&MwwjVNy;E!@1Fi<g-8t$w++h$vS?BxMbJdUbctX+GhL
z;abQQ5rN3$kWafoBg@024wWO$VQ_)bUGWb4tT;}uXB&TTD2of;1e_KX%}R9QMk|I@
zGA?n8(DsSycvK=|I+nx@id9e3YUS_ONr1{oDb-;rG7Zz&bEi>XZ@Mon1b!u@YeG}`
z=?mUlIM!duv{LxgAWtc5&;=9z?I?i@E&)i;8=x4mk|4w)NXG!h>72A^czprcyF2<f
zUi(ZKIQp&$Hae#==jnL>TMGrA@_gqHgMc>|v-*uy6iy{JU!XXuxzn9ySg=sYc_IHQ
zec^J37W%?El5>a$!GaY%E6peA@Dkz1v_QxzSQkVafk}a}++6lq2Ku-@3FiVy{!zYA
zAcli5r%p0A19E1EX2$yn0#>sruaIv4q>v5E63A0l{M-=K3TEIDLDo^UMOw0*f+wto
zns7UEHU@kl7wmH|m-GHM^P6z$-TWg(W5Ot9XNiaST&kWTsxwbEv??eMt#vXEO!76|
zfabfwM4wE|fyr05?M!fZje)S&nTPH1I14WR#<`dhLF?38M5Gn%OmSfD0G)#(k8HPR
zjV!hQT<&q|-zGD6w4dEkOv5c5qNVt5TbgVaQXN{X4;xi6wVR+x{P_SH5<&=Dwd`kM
zDBp@{rxiGkmm_|Z?avT+6!b?5jnT(Ken>CA$o{e$kt2LV{!<6hI$8VBGRQ&bUcEtO
z_``_mEsU}xE<;S#RLvw)U#8-HGB%+%rMdTd!WTX9X3nulX+Ag70{!#Ol$kp_Q#Dj>
zQ_H2fl6;}HL`U-u;{*mlFG@Xn6MrNC6URJht(EpL;&x1$x~*I(>b#W^3J|mdUQG?c
zY-#y8zv}RK+ve4#!`MB$E5jqF9h|EPhxyPkqr&OkP^IbKI~=a@^m)O)O#E~~eq+hA
z?{*AcJeG<w-T)X|rV9ibmWk=`;@e)g4BWGV<go&hhTrMDD|ygw`D8E(3K7=6d)V5V
zr>=voHBO-~xuo7NgrT@z6`YhnAG**vhRQ<mii|-^wkm5pGg+vxQDex4+6PNUqs4rW
ztqv7~)1ocRZIvC}*u{p>i%%TByt?jt^}R)2IFa~Rd**|Us^k}P6uPUt4smApm@3~^
zA*H0fvFyFxT*JOOBj4<7<Aud$^ym<DCfp{f!bLQ>ZQryj?-5<Gem)hk9%Rvv^P=@M
zQ$H<l9~aMcvUs+jQobQWc9~#mESfMoS3V~#RPcu9-V}4v=S>+UZYPlia8`16Ec*@f
zF-rH<uAUp!)z_vCYzUr#_dVUEPBh3rQxvSf*YdNQ4`j+CqxZ{3!INn<i;ZhH2H>$y
z)rI70cJA6rR)@i-d141c*!v-+mq$YcEIHnA?^|Sx0_ucHA+_oRft`C)axAhUDLf^Z
zD7MpYB+GGDcw9YznD(J!95i%t1J1m#(xfYZST|f!<^Xbvl-$Z?L*dN;x~;gTpH+2P
zjSvgsV^4Gn<%Ucb6$;vNx4m1qwYrEm5CzVinwda>`awqkQ5FmE@dbe%bO=)N_gnkR
zzeaR}Iym9ZP2Th1xgPIezhHiw&YQtIM&%=iZ}?^&o?A2OseqT?uMwz&=ZXVZ1up@9
zRQYD3i?PN`Giy2OM^mYsZEM9iP>fmTk;R4^iIYcVxSBQm>PLOa19@CYTb_IGyq4(q
zMC9C!4I*@gGl+|^`!kL=_tK4fqC5-~$(L$L+Z)rwp^*^<cAOc|iHSQYp`9HI)6n^)
z5ee(*x(m5Ee8HlpFC6pqDh@}wApZ+Q`Nf1ZWXxgGER-<|zg7Rl@O}bBCG^Uv=t0wS
zx0`SE8+GvMFyswihnPf%0CtL=syh<E#<=#7O<B>rB<u+BW9xNUt&me0%#@6UF6iGb
z<Da4b-&tr(i^v2){Xs(HLdL7S{`qBpGMtnOuZNV(R*JiuTbY#8scMQtSl-oCE8P|p
zNvb<3EB4eypR-+vO&b-BRWl8Ty6B4O5_N^hzbwtBq7w(FNM?NIG-ifoQ>pz3LTD)n
z3rD*WF&cpTbdVePAn8GyCQ1Y<fL_dea|T3!;tT1(C&2kf-a)%fMHl!``;^Eo);WII
zh<BaJfXdc0E7x!{WVw`wHw;&JUap?erx;1xRj$@k@e-$8A_gm&IV90ou>{qc+-!Zq
zz1Biu!?KBe&XEDz>bSFKafq_amsr@_S>KG;eA&OS{#DNRS-WN1HYLbPdGnA1zhkL)
zudDpVP?TAX;IFRPybPu>IGQJ~jnBFmDkk-TH)Xc!)5ELas?V!aPhYfCkU@Qukj^ap
z<oFTAa}oEc0dkG6_!;R118j?YGZWO_4Uyu^pXkDO^I7(b)21ag8`{z!WyszpDlcY)
z+W&IheKXbiSMSf-10Xt=rR#5+tJbf;#i+u+D+IR{JguJQwH3FYeT3xu1zj{dGqE1t
z4Vwv)a%#|gWs?+ogKEr|g!{6iYhv<y=({xIp#dnifIir9(d*fjoGrtmME-uY1uRr}
zR_JFMs_dQqd(rVT6P-|=fs9ADjR9~K4X=(Mp8>Z@?ZY8Y+4eoO)?l+ewaWx}f;tXC
zH#|<Eky{)PPcBwItsXA+2&#q?fh?CAhBW*`w2l%o7wTIP)Xp0|uYo>6JnUedC-#@A
zu4i=B;>P9-_ykbJ1!+K*-N^fC$Yu&djm4X}mh?0(6T?S}?M9Xg7us1DO++p>ZFMw6
za9S_waa)$ZytJkj8I<angHcR@V|)nCI$PTkNLei2X}-;Di9?O~=#QaS_Tka<%vH;h
z_WYGN7nL2Nxn}bOs<P+_PZcNXR<lrv;kb#+x`+yLhvc6w+vFm-YGcjncHCZ`y~-Ao
zKQ}UDmD0>t^iHfg%E+Ie&S?l~A?}!fs%(9EKYM6jhc@UBQ@_B88;24!cTfhKaI<$V
z_YZy##k<I^)Vu5JMfMvwy!<=-L04R`DZQepNCXsm(e3P8?aT6FZEyZs5&d!%P|}fA
zN!t!-bl~%TNkU-w7ZcU$(^_^4KIe4%?w6yfDz;{ydFafJI5En|n{=zOVOP5L-C^z<
zZ_UieytZ}l7onboGujD%J8z;)7hRQin(3v71>o7B24I%uX79j}N30WogQmZKDYj@-
zXfJ>5g^a1%lmL*@)Tr>GYKnw0VoQN_A2(%hE|;r6W6)u$xnbpLE+*r}1<2Q$n_S$W
z%X4*(4ov>N>nZ}7b(DcL(}T&>Z&C%*1=un?8`C(X$Q@_@q9^07j^3Zn*Pwpq<h#(c
zj0t)l?oL=Zutu3FJVLQN3mxcPY<p|D>C09PKe7Z;>h<>r6ir_lSM&=foiHd@Fc=nH
zSeUl`iU9?iw{H<xf0m}hQ5=mPCuIYH!6Z;q4aSQA9Vo)1I1~f8ku(B5B#xf<ki423
z(CNVIAzP#+J)5W3(n5IEp$9d5>bR~Gh_TPBM@cj3tRw7rb3)vN!SJ53(#q`KC(m%7
zMA;j^t~ForKCOM4;Tko*ub+mpqZOEWt=~-L@?E1MYFu{!_$TGk+H-lnUc9mTtd+`|
zO+k^v_sv7zqTZCI<2?-h!P5b{kV}8)v_js3n;u^xf?z)XX2buw_kfr^;FpB~mRT`s
z*lt<CbD-y7;AVY9g5^i#Ma4XklXHyz+g~$wLsgYeju#ky2^5^Df-a?=v-+JD1{M4)
zWc2*hKzuCF1dcE0JTm!ekYrYaI3d$S9FzNLI7o`du15J7(D@{Q^X2$NWJj<t{=~si
z7xWRJD-=G!T;c-^+zW$-A>IX8(3I*Ciaa1mJDXc9J4cY~<=`Hw!h!f<3aJwV23ZVs
z$nw>T^;bH8u1<TBmO$3#6NdrUgvLG0V;Fqg0>uMyG2!x<dng_4+=ku@N-Hbv-ZEt3
zWB#xjsAj8cTamQ<KBordVU-qM6LjVq`_*Y$WvjW}ysp`ZL%DiVCgXOqppKOnfRIOZ
zeYGuaTBkUOWpIP_aJN|6Qr0%vO<7^Pk4r}{40C>cHnAaq6BnHA{dB|XNABdH;k_Eh
z%ca=epe9gg1y?lttyfYgEVXEI<)C$iSHjqA;gQTo8qr|;Q)qvG9;Y$SIv>rBrL0A@
zK>Tv&pUxp(QJ!O@q8@LW9n#$>={r$^zPR1X(ZF$G<|8KEc(qoRocx$`0DFo%SURru
zI8o9~)?Q8rt!~(AZqTeGyo{Bx>)%4NyE@C3xKdA?(mD&OUpFiY|0S~pAoeLN8&4#}
zTW$*-g*-NR(7`;noS^V7D(2~Cs@hz_*jmC^IMnQF0W7Zn`>25Z7g*e|f3GMZDFx^3
z@GkQ*=aa0CnIUYhUmLRV)8kmkw}j*LM$#+8l&zI2?q^xNX{5?=y8JLV#eJvY&a@gp
zSU6#laN|gAHkS)SzL;2nizh04LXG)`*mpQq=PEno$)_bp-kWMSzjssGv8Oee|1D~8
zBPuGs5`V^a_p3-vl)L8ItFP8((wN*3-WSqRSIrCh&EMm)^j)3BJ*ymTQez#nv$X>I
ztUg+hagN%TZVVkrcTCs2-J;w`zSRFE+9d58%t=OvN7D-zh!u2;wfg#j<NDO&WWJ9Y
z`WB`??K~r9(}$kc>Rzjxm=K!G|D8PFP};UFp36&O=U88wn+1w<MEWi~RsTiYvU9Oa
zSQ%=jD5sn%m@4+)Ma2|9y{w(JU;I|ct#Io_ZOFZkVz$^5NI`kiyZ8PQt#s(6=JBNu
zTZbhgb;2!$Ir5w0>#n*9$7PXm4?3<(9fz@jE*bq<3|+{WdM;>Z)=<8$zc4?Qf~tSx
zVX>GmYx}GHH|#q#WdHJrYg0%qW>0vtOSwPT)~2DFdAkn$C}ZELKS_BTd~q-wWV!}4
z<52At98XDdECrA9AE7SHOr}ze%3sA!grzX?2xLt!YQ^zcnYk86)M$9e<@jt^%Ze#g
zr=m4Il*ujJH$Sj=Fb|T(3+R;NYBhNk(*}IK#^!za%=_K7EOEc4JBXR)g@+gsW*68N
z!H4RPPa){Lbn=XOZ(YWpFPtV!&%6pA+*LcY+1m7>kxhBecv9v)MYw!NP~D5bJ!;V~
z`cKbFS|AI{ZIP4PDK4^NhOF}11^F`=7G`5lm2N%0btIF@P^C5xDP~`1ainaT$dTg0
z#XC*ku?*cLfq%Y$k2=ysDUA>dl(zYkQ3GuY8`kArAq;QAzjnXX=d$Dexnr<N+&6w2
z->uhex+>ninB!7PRd3N`PQ72^55nKJm}^v-+r;jQAvev|yguKIAt(LrN#$L<w|b}M
z7d{R`*HsV0UIPE02M@5f0KG~~y7FA<1g_KxYK*yX{ELMbgH=>c{v&ID;Y%ztM!_lv
z=-;j2ACE@AntNE!bpLYFEP7C5_$tbZy)ZsC<Tn83{pq(=tT_`OpGL&)%*Dh%uaRH1
zB0R(!q)2kFNS0DhunIlWoGFbaD$t3WfRz9AS_r5DzFnhZVCnpNfIu}}!&3-~z{CJ>
zI1$8g(sK%pOmBdha@AyS%B&38xGLIk`vOVlidQVFB3owZ_BJgPRG;yuh;f%rOZmek
zm>t<W#Ss8%N=>}Iw3l6VSZ8?r4SyDZ@%sCWJqK`T;hHV4g(*JU9Y9CV#BTwYCE*;5
zI6T(Q%Gr2(qNtUWEwpc6qNKi!lriK^n&o+Mj<6U|oJYLDEd{pqI|v4{z5hyTBs=6D
zY+nE9z+>T&%xRx%{u@@6sK9B0E-Szd^UoDLljJD$nC=H@m?1V1)wIJi)^R(Q!fUMD
zcNW8jD)mkD#z&-BFE(60MYFcw)13Ko(=i{m@;bigWj19g4=L0$%43OnD{A?nLz*Vd
ze7z#HAy%Vw^FGU3#VB^ePFE^6{|LCu`(F+bH*VQ~<*Do9Z2!_!oicQQb1(Au@y9OS
zOW|x|AvL9D<sU{9&G47j`8e%6I<lH_1Ttj@g2HNvU6EBfoO}m1HB&Ez(jo~dGm&vs
zCGYd+T~FyLml&o<rBI>_>(vD_%5^WA>_LV!2hI=VY!jMj3rs;=_im7CTw?SGnoS1q
zK8ydX>-oH5IJ9Q;Pr&+@LM0|qywj@}Z}H+gOoJ3}I?7nCZVOc)jB%^{Gj8!x)A!<C
zjcpKV|96&BsmNpwBun3y+bX=^LAhfmWx8iQOCwSj`y*T&`!IJM`HFJD_s4WLWg#e1
zI(Dvp@p1icUuLx|4Wi`JCx%mv4KiA;$F<pBDKT@s0Ze>zF+ChXQV5xF_K*50U!?0B
zhMugHddenlq((nCj4l&howJ1r_bE%VnmjQii%TMPtd(0HP~`Ph6}!v4`K5+=W~Mms
zsE43~vg^{L@=8sdztT=>DzlZmd}EbNrE>=?Nh0+ty`02)kr+1JJJ~r^qKw_~=&k&R
zEi#F-FJmqJY@Je0)S?Zy=ztcN=W`*==B?-w$*^;1<?U>=Cnx{ato@rK3q9SNlbgRR
zLEjq}pGYqf*d;ACE!&uc{S%)w2evwm4X(7qo#|Vo0UE}=<E^$4XrL;F3&_c^H9Z8~
z1B0&g0Kv#>bs{wqz>Rn<Gl}RDumgA167EpD)j?J6u9u`dV1~@Q!x3w-QbMhDKBdbf
z&)Ut8Yj0&yWtqft_4?5(NbL&JP{ic}GAB{=VP!S)e5_Zg3E+H-mz~rs8@5;1#6Fa3
zUEMaJxz^OW%r%$sJ(DUgP-z&5XSwEMs7#O__+il(`2cT$@HHfl9}yIQJ><+lU7DsC
zH?iP%ER050p(xvS-u>=_`0`+q4{Eyz!ZHZYKQ||AiE^i|&$Tqc#&a)56K68&;$j6s
zGjF>gX&$xeed}LkRp*SArn_oB8)f@sI|@I&SwUq$bm9Jo=8$KGQT1u)`4)StO1}#>
zucZr9{E<A@OBgk~&FRHm;S!?6XvO@2;EFSfmZy;IF#rEdM+NlI6;A&f=m7l`F^S?K
zZJ$3BsBy{~aI_qEi7AfA9mku7TGc;)ilq09G>?q{7S7)1(>WBH&q8Mi0^pP!&Pkd<
z#K+ff<@#`cq#&N)bl8@rlq9Jh5j8VeEOhCFqB5CLa^X{W6qSnQVjkcl8s!;u05KI4
zE`KZGo1mG2X1R$T8YJaADrxQ82)7K1XAL-~h%swa8ehd_BZT-4jWYw*XC@Rrlw~mw
zsJfFXW#2+MF8j;2z*Ne=iu8pQpLvRU(fzD^hfQ;sN?s=F=2~+1J4YL#OZfNZdV8~3
z8Fa|H@1f;iBx`Mg6PBfO=WQ>Gz3H@{Ss2sPgMEhE6?1=kwJ{SL{nbhb?AbQz91jA?
zNKMks!V4i$3xuK7z^+N5zKT0??@5IUK)C^bV(MmiD$;^xW@Dy2tQ2Gl8@V?MlJv9e
z2Kr<5J|dkGHi9h_(dj8omCDtl40WAC=HY<5I(-gCH9{QenC5Xp$k^^+Fdlw&w>jA!
zrfgL<C5m3)9E2pJcd_)M@`Vj)X2JGrAGQSVK!^^FQEnsE2^tlb{A-SwWu%uH!R7iE
z7C26(nq_=`%>{hEQH8#wzHrBNHkR%u4xb<OC3sD?3S>U_KCMHIb)VvQ^xyZ>Y+nAd
zK9w^hkO`#l#x!6(0C{dbu;Fp7!vN`ilKwkhm0DlZ)WPz;AD??(ZPkSMwCY>b0sM!z
z;VezwB0Eln+MVHZ=HYe+#X_Uk1e(1vL=+&mGv_^varC>BHfAS8`QVk}>JVjb>BejQ
zaOtoh!ZmOIo8R_phb%=3N-1xa)A&c<w!CLLx1|a0h}e-=Ji?nk!M~e$F>R`aAFfRw
z8?{CW+Dvij5t68;NUNNf)S>lT#|i%ftw81%pDb^cdJ1{2Zu6X!8;4cD)qVP)L#nIt
zMg!iyqtT68!vw=6|Hw%op3J!kbjf?D1x95lO%i4plQXciAsUkSK`>QA(IK10sNYOO
z9_f95#$U+0f*)m^Z(Vc08fEKbx@pFjr`y%lim0oIOw8XZ8}fXSJPvp%Lp!AQE(_1-
zOS5O@L+jKVmh22QY&&&3QWm*6_$Om6{p)a7Z@8VYHu;yLb4!5SAEm31dHz78`R*}&
zn+ko9S*>rz#SRz*XuMXg#l%0y-3H_wIR?{A43*ju8Z`=M)Yr_+-CiM3r8I5)e5UU=
zO+%@7Zm8gwc4M1mDgLcjwcC4kL2}I91uK-=KufKPDdJg+zF1y10S6AHUl^)p;b0W|
zj-kA+Jhh^Uy18KOgT8g@%Y%BQ(Ii#1Ey%geDsRpn3M{w2E>dI3z)hyfdJ%|C0>C&T
z;<=+lH{q*uc^w4s6~mx!L6X!Pa*S)$H9}8Z)6CoU(af8{?Han?^^$#1A04AWT+^5P
zE$7x$<8R~!SE_f8xaHD*?2Xz>`1=apSz^DLh37@4g^KHSmtM*G+idW`;<9Vx;rB3p
zoUv^qQ?RZAcV?CwGaLo9QgRpP@&O*Xf@jiUUxI~sgk_V=v#SRwVGvF6fV=>;OUi3a
zvR`5f^5;T%YW(S!@V4vRU}&)x(2)xUE}#zo3fYOE|FRGA%z*XI0L6+hr?+YQD5^6<
z+GJz>^;{Ri1ut0txR~(4Un`Bi^ud(u2YxN^s;1M#$Y<Kglw*Q{Ej-rxJYtG_de0c>
zIii1#!yY48bW9K|Fh&F`gK!RgOsWg<E)$vNp~uuufPw*1*L8Q_F{eUN0V5z8Z+MBo
z545q9Kn2_{dNriACC%BIWjtf5S~vTd3uT2;s5510wBryB;=gipW-D0QgRY0K^Rv<E
zF?>RE;^yVhUyUAcqnR|YvTq%^)1qGyTM~)K@)KJ#o(c2BM!#-xOa%0#-Jdf=RkQGl
z1N^2vVQ$B1tW9TmpKl-X%p_^UBQ4gHfxFs*B}!LlI6!L1BkkEMQsZexE3wIIXyxu8
zee3c3M=aCvQ0Yh_&>d+gmAh4|#wj~YXIerm)-S{0wB|@uCMu31U`C6xIQ40&!;`9s
zKa=+F+A`Rh4!f$~m$NlJtza=<;+tBP-^$?Oo0J|H(YJ7Vb^6QskshCwfm8WQ7VXbU
zC-fAQVn)BV?ouX5AHu86T>w*Fy4Y~hF^=b^w1SDXe=Ui}jHc5}_U&W&ucrnf<D5h4
zbK`Mh^m>n82T2?HS}ENruq$~B`OGC?S26+yg@uH<5rFgol`d8A)mzO9J6|_bd~+sc
zUcG;;@Mn57a4wnrHgV<)z3j6F;`-HM`DInkBtBai)5WWphP9%!KhigdHPXVO&QsB(
zou;0aZ#fctn2Lt&tgCWyCu=v#B-wmxr(HM)7^@|U8$wRpkvTA2fB0JaGpP?DrP>Mx
zHL&}_1=|+4{VMaimTj!)_^gx6f9_(dv4nb-CBZ8zwXbKQc8hg|we!f|fApI?!XMU3
zIpIx&;EX>b_bxkR_~8^+o}25L3!YV&|1KjHqm5jDY~}Aq@1s}&tiocvALs!yXmL_c
z_=>}Pg{R^*N1n#;Eyk{`d~HWYmC{R<qB_p;3BQ^)PS+mQpL5<zxvFRsqzcq0URra8
zdpxNy#TmX%*g8InOIMubi?#KgEBy*~mF>wu76}=k=)`?!t~*AO?myDdlYX3Wd{!%?
zb4GOJfS-W7pBQd)=XH^HBVjCnb*yl^26RmWt322VXE2RPeU1ZZ<LvZ2O(P~4%>3Y7
z#3{lCyF<3cUGY7Pag2|H;%^;ryjOR2wASt`X!xB%-Atqejvii77EMCQPgV4T&}Gtk
z_eup>^c|ee3K{NJn{k!y@ugJG2sNCGL~Tl!JU(C+Si5mqM9LS|CTdOtd2z4k=~Og}
z1SxSG(e<$4AGMsBRf7%P4|05H-KUP>xyoJsyZ|dNZS=ifxWc(E2QKoXeaD2cCV$p3
z<)J`ZB@OEU_mogC8*Wm<y2RKRb!kkw=JSUno1i7K@v}gjp2=tYOh^$Z`}@A@^MSe)
zJ>RJw!*8u_9uE;&GiVEAt$*<Z#W<W@P6U-RL`kKj3Q^QBGz`J130{wX6{gl$fojSw
zWXn?MV!%DV?A9a5UwJcg;2!|Z@W1US5IFsRA{Y}8Z^M{hg8CG)Bu9wqro69pV!73J
zeHoIU<lXj@p#~lnfV0mXHKvvw`m%s1+|*-bsg2dWPDHWk=Ol$xCIO^u?7y@s3nG)n
z(2yQM3g-x)pzV=grV~o45uO0Sw23Gq?k$QGR-TpcVF<m|*s@;3NH>7JODw$14+~o+
z(yI$0h5I6R9jUF82aRWjw}`gXeY*8QQo0E6BN{Wye$b=Xej~SQH)H>FIiwD6h|oO=
zHW^7vLkas{W0I68#c8<t4c80MnPKJzRP&W<7>8WbZl*BpPJFS~>Q@D7q!&JmP!hJl
z*S`=QzTezp50Z|TQm%D07hOg>*7`!9a<_i9EB+WxP=OGs&6`5|9fXqQKx74Xii4uN
zgE4Vsa^t$E)9HQlK`wu^n<*|Pf2%2_&nO5LRwZ?>l4C;|B91MRAC;QsIDBjQm}Q#U
zdL43CAS?fI;<fO5Jk{^(+e))W1wna~MP2a$80{S7x0syMdQ1>ZA-mtDecz~cv5{)G
z4crEf*9u%yw*OE+8a8_}TyS8ZrB1@jsd4ryGJp!Ci@t5d-I2@-_ZF|cuma;hP$_{d
znDh{-KM~<5ddYr}(hW8WdUa0yuJD+FAq#^_HJr1fJuv)<9K<(s?wP=G)&PDqY%jPa
zC1uOFE`<(dAW`F&mW<1Pmrj*Pr6ILEx^D1Lrx`T}CcRWyz!D-m6u;g~#9a?5428tG
z8+vMBxx0p^Di$1NaEuVEzf^C7ZRGS|rlIbDlRl8B^PBYv1<@}TrdN1Zw1ud3hsRUU
zyXMYZ36rJJ2W|H~qX8JGYU-J>_BFma`Zmg`R7x45BQuGTVjiY<B@PWa&1g!pUV6$h
zfG&HU7ajqnl@AQtG1@H!1S-is;pPr2g}gkV<!CUoptCA(Y!%ig(2HeUwn9&-+FE~F
zf65mQ<qnbC#lG&b-s`uR6z9UOSl+?dP_j-<ql`jZW)7&hZGXlY+@vtv6XE%-rR#^z
z{HJ|z{}s7!zyGC=d1K+uIPO#5eO*e?g;-;e@744OfH#rK_|ZY*2=(y%j2sZ|4(`$c
z@C+y_o(kCkaf`1ge(N`!<yGITGM%1$wdy6Iyq=`*`iuv+MIknS&B)6Lt&zbj`rtVp
z2b8`skS7;U+OSNfa)krD#T~+`?=>gq^ZRYPbc{S^T0yQ3yTX2j{ny%3GvstVomh;B
z&Zx@EzEGxqIE(vB%8mcVoZv+Qgw4c^V*?6m<=L^B!0?FaE$>h{MCG5)`o?_}c<|DM
zv23cb&XhL>M;Ofyo+?u2uDbEm0WxII;{66Wd!wt40ciAA7_s_1Qw)D0XuH(OYS)?}
zU4H_Zjms(v*cd8E?-Zua0QYhK7xw<-KxuTIW?rf58M_w{a(c)v<6<4nQOAD}+RzOu
zbHQ-@v*jDY04N3Z<i7=@1&R^S%cd-W%NI4X4r+iTNh9IB!+B2Jy7E`sRR4_5>z#Mo
z_RhHBLjjz9YEJK+)PUWtX<-$ViBK_0WzSGgZ313xCS3W(L5KGPxR(Sns}U-6H`vOk
z0SWQ!=m$vhATS(ZRldyw5i>7R2Ojgo;uyE0Q*0K)DsB)CNyc=^jeL(&q$-|>r}qC+
zQ748w3anLra^3SMk<=^-bfuH>WC;~>bl|5{dtBSihQelsI_xsu!X<Yu%;mC_F;lTg
zP}eNDx~7}%_X{<MDTAzpzTr`q+?Bg4NwEjH87mJllHRPOH#oC8hrON$oREpvCColg
z%Io~=RgE6}IDPVGG5u9VKyc@fIB>Bp2cFfX+>>hYvCu;>AR90k4i#G%3w|9suDSt_
z3pRdlleDm8bDGQ8C{Z|sfDZ<Z6?YRv3ZxNA(rCXMX9CHwHr9WnhJl_Q0YG3>7J#Lz
zD$eC=p2*8Q;HXMdl2-A<a#^dW&s5?^y++NSj^WLg(QXHnWrL(^Y-?4{Jugb|qv&_u
z4~w>*K5>Z0=p+Kb<J>6GP7eDQG)LDUJ%17F4Mwg*?E%sa-zXWkkL&`&ZBlJLOJ0;f
zW>Eh`LEy)Q%EGj8pzqvjIHDSGpoVAt)R%m;>-)4lH%BH+{MEvsWyRYxNS1?s7ms+s
zw<bU9%nE_#qF4^EOaZ)BUQtP5#4LU1WurtgQ`&?r(s>k{GPj>I%%*?e?S(t9DV>!~
zoKeshy8#?p<>ixkq?^8ROZf=5Ed4NgjLyq>;3-6;uqw6a#-fk5Uk;xoTB9<NPL%l`
zIlhZngKSGXB0_10MsttJN<^y|RiaLN+#QBAX`%J05}T1<K>SRs2t|8j(bnrlxYqRG
zgwfZ0re2}B_p!~{j???DjnFdw_){Gdss<}`lFCqRM(-n~cN)-mepbh{jyc^EhYi7(
z5o!Y$&g{O!#PBuAH8u<C7gSSxp`9#>uVlRCi#yPOUP^_Qq@n2>f0!7&(q192Mp~^q
zywoBTU>a<q%w4UAXq8w=jZR8wA~3BzO}bM^w+>$NUd)Yu?li?74suGGeX9F0urPdr
z!H8pkq4`#;iZHo&Ed}RErj!1B>Z~iFd^`JT5|V`}ooqP<W3fa_QY*mGR-7o7<X0`-
zJ2r}GC|1!r;nWsYODe6(4$|#L;Am33qIEkp>8!Sv{->^zs9~@0i-YN{p*^%;l?3-$
zx!r_NY^=r0U@M`A(-(m5>HGIRpm~zi?<A06D&Tkni`OtJ2lsyM62XcCSi)Gwh8(*&
z-u~LvUt{6ch0@Eg#cEY(^{PM?`o`(i&1INGv@<0(K7k`h&B|)Q*Ig^DqNGAcOp0F~
z2k|nX;;*uMn{AMoGubMK^_@4}-E!46kI4Xo+>*>0nE&AVADwTW&&@_%3f2#<ZW+wM
zWK+L-)V_k~M05&Xd`CW_0||IariM>Z>-tTXgZ?{>fR<^X_iYmOkc=3JNu8iE0W*%w
zKRQ>v627SrwGuhkn7w5w7WHt3XCLFCsmqQPJRQ}#{mw0QxS~4P*rjh4ODD)PgMsSs
z0=FZ7i75gIwq!0Z(rLgKhlLl<1jX-R{8%v<jvMR__a2(dqkB|9KrcuMB3>g<1Gdhv
zW6Z!}**$0ox)+Z5jmJ>m7S5Ld;<DozIxVa06dK-so<bk}>#%-jhLO07AH?Hn1AaZ%
zsJ#4Z3B##wdXK*F&Hq4ltO_OV#w@MPJ|C_c?)E(3cpZ0;Hl;iK+jk-3EFI<-QqQ7w
z<yZ0Zn|`g{B1<*b!Z#00cb0z5Ba$v{2EkC}lBrZFSVrm^t^p-b`Que+lh?pTK+RM7
z(JINm`z+}6yE0h@<4CfScm(kg($#Sj;sGF2qf6mB!TXr(RZFpSHn3INcOMr!-JeH0
zHkSqQ_Cr&m31;EX)ouJHS;aGv!jI@rD5FA;(St-|A4(u`WU#GzaWwlOVk)o<ER?^w
z^;D_szA>TNcdE+dlecB=#`T0ze`u(zo#RzDg%OrVf&6iIRqqtWzzOuyK3X>f=H#f1
zh%dB(PaUwH=z{1AU-@UhXbFc=WQp;gkDd-2xEXBB1#=C~z4}qZDrNXx^(y;L59-B>
zJ3AIXvX}i2IQ4CKkffY9t{O?|!E;%tab!ha>!%OFVv{b)kex-(*F*7it{S7*FII=N
zk2i~HhLQ2A(<!|khuQq9)-So=E|*3GTb<J0nK<MqV~PEO<LKuP#P&R-j@42|^l^iq
z0b7V@vIu}K;)gM=PB2`rWf{rcl~!hnLaf&>B%yiub&(&{Pb)d6u+o219xXOWfrjVb
ztj{0z2YXYLV?PV2fP1Y38AzXBDSdpI`k+~7!D}P`-}c<-P4><oQu@-RLHoea9jONk
zka4${^)D4w5_RvSiqR8*ZHvkSuGf3H6L)#4RaEwFM&3@&OmHkitibIR;(;GBm#qvc
zdHiDZ4vO`5=e|KYz@HJ~0YBvMp_g2T;JV;#0;L3f6YjTeH9d^Lsqx!8u;LL8r60ka
zzHnamPO8x5bM5s6ujpbcm92s>GWE2<AQ*6WS`tA#I_#T<c~_*(h5A-4gU6fUar$=^
zebyHVD*LPrqK`lORyB*3m-abUUL(rm(Jb#3Y>Bam?!>90024STyt`jqi%eTJMR5SN
zG8ncz0>>NWPqQPaXsgDgu69IXsq`(J$;zL5NnKBKL{@4T?~H7XGVWWkTHR3PV^RpW
zwJ&P-$qFZ0F8F**4wBO9M-9pl@CJemX+LJ27<A=&hu3Phk-M(jqwG2=0brYrIN-sV
zril&Cu(Z!N20%#v=8%JsiFdTx@y}OY-3{ez9#EU{4`B$`?%<w}oy$Az-qTX=;qBhT
zh%IRBT*z&PbOoH?d4*%9+s-Aq|MNioH=<}`pzbIv3^Z+kO8Vm;?8RPUD<b)!T%yDK
zq#QUzmeNu1eiR<#p~CI&c~UMS2)m_&8rIs%*8^l=S}V-qB0;G1xiK_7{*NM%31=*w
zfCumB10*9@xBfbsES27@u9%iTfv{s?T%9(mUrW;-=xZ-6Xde{>N6ZjWWZ->NMgj`!
zbax?$WwLdO05PA5Jy?rQpo6KKGum0H6E$}29py||JY0X4sHl*Ib=9}>?rB=%+#NH4
z#zfp=QX%qv%wM58?YG}DKig)V&r`1bn#$=3NdwA{wG8h;x@F(@3s;jleDm74(_S}^
zi-1oA5iy1C-DISXF;NG2-)p5NGdKjcS`fyslFIUlw_Rn2KR)cY%&4(a0-s${&4j*S
zsB(ec!;z`CD12+VkEO>Z0q+zHUc3ZLvlWRIZuDx9FmffPI<naKlSGA@$-m$<L|dsY
z|70OTN=$rDFUwVuuLvjz#@27I$JuD#wr39vY!xnqj-!dW%=J;>bQ!TG?3kh%T<9;S
zvw<Cbg1G?L62P_sl340Ah+OzDvGDegWsOzNG!Qt+6o1%FZex_pIz{n{d<49cF8|*|
z=mp95s2%$i+2*S(p;C&io5+FZuYFFHkGl6wLG$*BAATqe!ha_J)<=F06im^~8oyTY
z$~x`c-ssb=*D)1Rcw7D$$8<sJ8b?FntXoHAt|AoTCv}Qi2QhRJ{67#xOJ_aLX%(7-
zV8jMs!-*gjerz!C1*&f^mFamL_lQZq=K4LvfyndO@pE&5S>1R#gc9O;cCqqcGn`*D
zov{TKDUF(mh2J^tqgk$chFDVxftrZxS_q{mI{t<J0^qq3EFW*TMP>Euy3`!MaWD#W
z<W~wVD*7MD#YQaD^{!Vi)Tj0ZEdU=#{l-+trsM2?o{|5plFmbA{QmqiixuM+*Gf54
zn*Z6x!?`1+P-+k25YOjwnSf#_y+~|=G3R0meu}`s7KuB}8=V-$af_5Zc?1a;%7<nv
zY&cns1f$&(mdFApIeukVdH&KPST9|U#Q+=^)hXpaK2*=@v2YYK8_Wy=dN^fHH8W*+
zJw&UN0bcs|;I9XeM4Gqv8uTgcb7n}vb@ptUpm~fn-B_(tA#Q4xsDaoSXkv<*HDg?g
z4}EH42F18II!E^Jfm9EMM^?VThpJWq_p(-nOl!;e^E9C3E6C1PYF%zQ1(*6*=4CbB
z-WXMtKwU8wVQW@6R;>Pj>$KPirS~P22kFY4zfIAx;r^DVQehoDv-XfKjEXhZ*8dhz
zXH@)BKE0{hxJ9D4lLjE<T}mgY|8CZVF)Jf$?5h+Rb(ba`e@pU|Y8HL=9ro7!Aaw0r
z#EjnIn@-EXO?TF}0>?3{=Op({j<NTv)9dZG9cU-Zyzitd_<F54MUT?N%2kZY(nkXm
zu_cH+3PcP!GUKBQ$%XVJNlr6sr-kb+toYXFvP+_&v{B9ZtqOo-n}u|KLJ@&OUgMH#
z-*>DE_;6~fPy+JJX{2E7V3t~T#1)EAqKO1i3(tD(c~5j9MPP;{Q1kfJl%&(?C}JBR
zMP$PQ!&hh11l<%=r#ywNj9Vcz`Dj~y7^rYM8pwU0NW4Lu(<?{9R<{UOOj#I!)H~jk
zRv?pw8gL}ipgF}ztLl$hsT0Aekvq}cPxg$Oh@kMPC72IGGrTyxZnWRmQBD~0KKd`L
zPFr6++Q$F#{SoOGm_K{cU!F1Y(<xX+I|X~wo=O^S7x+@IZt0!Rae0{wBA!VxLrh>h
z4nhh%s$PWw`g1iraQt@`o7eH50tl!5?5@HQPqOkOX2q*gx;6R@Gmu{Hjl{cOSf;fh
zPSC7d`IgjoG8pO%pLq>h8oWrCNckqGfSbah>2wf@Ua}bVAY90dyM~F^WTlwnkMTpR
zil8KA;S3!yE|F01H}jO4cq)!_OcQf&Az6XapK!;|R_zWiYSS)K%J2qftVl?WWDbCB
zYANnWx&vurA{zgjI)J(p!+*v1)a_R<Eg{k=uUt|~B=+oW?j}r*JMJ7A{(RqWaF=WH
zZpF`J{K}V?J|V9mFYS-lFCXfCEUIBO71Df;@Sms&8g}sfs5ABq<)$Gun)PzYw4rcV
zyyIQ^%ZA-3f$7(8{h?_SHph8(d(U1r2B*)AKaKiA&c|DL<Gv|ZIdYvr?W4!tO>M{M
z;9QCM>?cVKr9Pn<0XBN;D|xnFEkyoQH7=m3dF3y3^guyey!2s=JXW?l0|Fi}GpxE8
z1bfQtn@{>~)s`yqyKpL{x|V?C>Dq-^TDDuxSTD%w;{&IMM58*TBmxFv&}aAC_tBaG
z8a_0Y(>Z?9U!*qC+&6{&9P1iwXr5;XSBm6lYrJ=wU>D94&h8J<<o#6}aZz8ZB<%Ih
zxh-ea^W$N9K+f;xr+)q;P2yK`O8ETU%U(D7`o-?#CF+92M83rnl#>lj&Kj@SAPbf@
zj&6DsRKHN#rU;#=&(uS;`X!>Upv>5b2^s?+p{3E#FVC_~)GO5b>HyW=T1UX2f$uA^
zJ9J*5rX*h~M#}$zfz^YpJa0+;52Qi9HAes!U_V8J34vQ_ZEFPHexX8I3WJ6ZNCWl?
z=(^?N8vxFrtL(#-Z=Za%we)yzVXD3BQ=_l2$ttL@d5cu#B)PX|MHfX?PCf`m5&s8r
zxLGEU$ygu~J!%J$@hn${xM?=AWRrZTN{jBh5k!((Ql+2G7TR5P0#UcVYnp(ces&nD
z*bTQ)lVg)wL&p+dy26sXElpCk9m}Fls5Wef`&#L`I#6NKIpa|d4aA^%;o835aB0yJ
zPteMKz`Wp<TB$WpMrmj6dzvAAr*dqW%{x37{@tS)o_?h)nRwecKum=)c70?brQ`Kl
zsAncmDlX7tb-Wkzg!2EudiRWPoKArQN?WVF$9alm4c<THRMK}K#iIT%$pS?IxGWee
zJPtuZ0ru@1*B-wf`<F{<i&zcq`?!ot<ark;TBCg;E~?hjs@e3DkGt1B4aJ^#CRxE-
zW~eQseho1QP;OZvV_FEWc(_Kg5PzlJKpHyoGnB;wHRuhl($Xdcz*uy;M~!%c6j9Ji
zFF?IWmi4m%auEag0FGv;QID9Q>6fJuPys{433N$^mcnk)PX<`w<MgSQ+?}&Wu6K^o
zQW?`XaF;?|RP+&NP6N6$x=5XlW_Ru+=KZcJ)2&GdR`;B>rlzLlK&E{MZQCGEt10n$
z!+1-(ard!;f%}6sZf0kT4QZK-tV)ITH=Dj3+9Z0IjNvW<YKahCR91w+eW$v)a&%|p
z_oX3?Q7fdyf>>LnJb)3dQgkt+@JJ9rX?n1cKx{&0kb+c~^IL=Lc*NBtqSw5`7#eWH
z3-Jjw3+6`86*lyIdUOa4Us27cN7N3O2vC4Wkh5am7C$q!CE3c>q;L)IAdkbh{pxvI
zKj|;WIcV;y=-6s6>!7>)Y9{Gja{?`^jc+CMP?2OMFs{({bN{!?Gjsvx6;h>upBru;
z6cNLq`rV=@wSGDi<!MbmEe96*BZ~d_&|n^KFW_Z6@Jf7LTmXq5rDyh5u)B8Tu*XqF
zGb`EpYf=B_uu1Lj4g6!iwciQ~YJzIl3yM3$>a}p;CS=0|SilfEC7jAmtf+Wt!*M(B
z!^{MxI(xvMTX2f;^g9EXkzaWv25WWT6jG{4o)Z2LP_ozz8ierc!Ka1O6lx56+g<o3
zOf1BF8xw|4YmGOCHqH?G_|CWs1pQ4D!Pqn_bX>V?8z45)F{Fk=?czNw4kS@`vfVkK
zQ0&x+X^jPky!XpHk4QZqA5!Pv<%^fxr4#6Fs1#+=*du=%`qQo16vERs2>K(PFqg}F
zHW>bIv|+=SNS-(h)qc6!^KPq;{NH55O#IZ+GdO{XpjK8ZTUNYQF=`uxMs8`m6_S|C
z5NIM$9rsJnPEj$;r4m7|=M!d#u3LV=Xnh6}t)xJ6MKQrs{89aA|F4yk@QZ1ffyWkW
zrTi*n3>X^!^V*GC)uK_5&I0F3CKRQt#3~oc>>?&arZzVL399E$Hvn2WsVhr3Hw6pO
zSpPsfq{Y>lOGO+Mn70yDp=o)o(lfJaPx2IZOsMR>sg#p<tb{pWpMud0g@X{)b?Veh
zpMEvFDw7oqwG~mU8d4HY$AyfiNA~BNtxtW#%|rFV1}Sie46#~@9ryNil2~tWHxjMu
zqM(_D;9)&R8nX`Fv2IGD#t$MfU!6%#)%pUNqH0-1G3P(O8C|KUSyt1r`{wmC*H%|b
zuH-T@C1qtsicXiAf#QIUxwG}MgpSTbCsYh<&G_H``1_9_J=~fcUC?Y+amc5NqnFH9
z0>@KXL+@v$*n}Y(4JLvU!OF?%7FN1(;3x!h?`)^E!@$j$$3!EY@b^d2wwZ@k7hHpN
zh2pV_#&l!^0dUZe5&t!WrKdi`cPQO5*9|-;rM?KS_{U91R56dNR8cOM7mPBI;;XPk
zDHxe@S4Bb8wn^H8A|zf4JnJLvhNaqsJYr>rz<Lc#)aJwd0{$x5(bA3yUQ9WT202a7
z@*-H$l{^B`Yo&L$HFgvpj$ylI0Qb!ynjr<Yt7z;IlOjnq6js52U}4q!T`y3Y3RQFs
zEUB-2zcGzciZ*6o!|tIb8gx?iYm7%*klW37C-0}-9@-)D2p9!Bk_;oaeoIx{mmz&*
zBIB46`n?s$D;l(&g6$gr9N7(x(b8HnmP_gY{jW_%A_1j^QO(7Q?w64PeQ80tCzWoF
zq{T<SBj2pz(f}F22p`<9Vq(GKfC`-HGcL^?$i3KikGi(+UcWd-oP`~{4lP*58S02O
z1NC+^SEm}&5yoE1lndtO-(Eq@+bg)^U?V(HJ7kQwQTwOQeY2Dqa+}VD2bUohvkPQ1
zt}jW<k#}n1#lT&__C^d0iU*8LFhYEiIc+G<FSY2ISSJDU^aphLSl2bbd1O?aY)blw
z;3Do!K`iic1Gn5p;sA34WY}c)ZTduWVa#IwjR#D%jeVh-x@k_Utq=&3B<6?zrgTf@
zy41m|Yf`m6oo&c!V*ECqze@@ZJgxIN9(@9i(RQV{w6Q^W!1HALB_(g92aU^j*f-8>
zo6p!Hb`R!AoOn_gp-!)mt+Gd>p;nE#g+YBLvwslhb2IWYr}(0dF>}s4TK4CLTUlD`
zos!94glN5yOnzLf79#kz_0DUecXm}LsN;7?q_^9ztj`KL=RPFqn~CtvByHWl@uMv{
zGTB|A84#j_zP&$LEAi^J6xH*qo9d`zdoWn8S%N6&)oUtaBCR${EZ_g}_e891ZqP_=
ze_GgLiFtJYW(p)CN1<R1V`~7!Jn1ZFMLF#afhf(5zn2(4ktHg=&J_mthf$I=J7MGk
zRVc{VyAG%7mI~sFm_`FpToGDKX!lkwBfAQs4a}$(?9jpUXD^g|N^|G31Tu&Nj{%HH
zdzR)HFmr_5=aKUybNUdE%*F7E?lR!1*6$RPr&CrTk0auF$OzENPx}12LEwnO>-?G0
z1qGemlB<+1ZdB=p7-!gAeQ6qn)3bUO#Oj^)QVLrL8<Ov{GFGnSm}vT@Or~aQ7QAf=
zqY4s9v*9>Idl05#QGKpECf&!SLUD>FT`rULKkZy5M0oZ|D_#l5{Osk`fAr+J1H&p&
zR(6M&8&4f2o($wFxOyT#xau*~;1=#h+Ol7L7rtt%%WQhvIk7V(ey$5>&!%$VM3u|a
zg~KGPgsufSx#fVl0EyoSMaNypt`M(L=v`wBeYM;8`cZpy(MWr4fC^=K-)Y;gC{WfY
z%|tPU!iwM>#2GV_$FSx;a6R~OS@1Q3B2?AgY8@7+N$TpC(GLb~AktKd7I9Khd0;5x
zUdcPKlMLc5D)pjnPapApnYj-0KVzQ!|5=R%YxQj(olO@Wu-<49kKnNx&D=(8L+Dv*
zPXw?`8PdUN^AJ_|?Qm_*`TS}0Pdv$8C`)aIA4V|O6-p!;>Y(&%B&K3gpY24JbAK`f
zXOj_z6|tlC33WPv1-XNI34?x#9tOt?o`QF?X&+;#`ax<}o++Ed-<o~}*7WGo=%T()
z*p~Itmrf~Eqk>lrV(csH1&>TX520Epc8BmgrSXs9go{X)K?%T9Zn4O?7cDA18Un@b
zw~klTRMai!*)B9KOXY99DA;~xUZmz?^Z(Iw-tlbif7p+$_J|#7ORO4+tql=k6FaoW
zZlX3-bR2u65s6hoHArGBt=U$O8l{xlRn;D~RkcUa-Sc~H=Xsugy!zswiu?Y4KjXUI
zSHK(RFy%mNLMfLZyrS7U7-7BMk0$}awZOA9$FF%hmF2K9KVK=X^=Ar~n+6~*-Rxz|
zW_n|QK&=;W-cn%;tO8+$Uo^2fE*2QHcv;NvO#y99v#WkIgy__1{Uz;7(`Ww@4bozi
zWQCMfrh}lB)#B}C%Cx9USHOy{Wkk(^asf{Ne5TmR6a^R@02??iH9a5_m-p{0w0GD1
zl;y@2Tvbe4v|5S&ax~T2C{2#v3!s;#Yn3{;{!NBBoO8c$4PFUPt+6nV7~F^4o$&q@
zq+~2=_qwb(by7S^z#LO;_uY>D;TZvx61DIj5?|v76CIa9_~pniEiqzF;~n)DT|t7A
z7oKVx?ng|M-S#ryb{g7@$C~cCN>AZI7e_r@d!i1Ne=Z2hht;3*PHsR28?reG8M`Sv
zuhD$nV+}9+8NEK&AM}xfQ&y#Y<E)b?)vE#}rkO8o_LE%lDkX5}FC3_lhrNxU_2ApZ
zE)Ki%!hD8r%TU<}V~(PiOO{KZ&s<^6aUV6yp6CqgxEqS4dWcjWJUMEcvzF7ITf$Oi
zYLuQwY85TOKLltuoy!ydK3^Y>=%`72l><ML!1g<tb&8UzYq7E`Kz_Jb$2B`8e?9u|
zt&xjf%K`*QF=Sigf1tZ=rl#x6lo3JIsEtGCJtY5f5rs>S6r0Rq4=*l8rg;>?L0p4d
zU7hIf0Q%c@<GkjE9TPJ_n5`D%Qd+C-7X-Oct<UT2$k8r9MZ$?76tFhnR0sB(9CQ7g
z;c09GjPOZN0*?Nw@##=+>x9O5qeAI<&M4#!ckC)KxH;uRoS;M&zbDYcE_Ds(#-PTU
zHo_aML)c4!*kH)qd;*F+==17AR%GCOhV~g1_=f~5R0$SHujD10j(F>I74VjME&BhK
zG@Ws=dNL9+6q9i6`>D~!RwXHoP>$D5ed5wz693g*s0*v(Qrg^j>1<&b<L(#|{dSmb
zFirE9_pbErWU3k3PvmuE1sd8I2pCJTKw<;1`eoZ1H>c@eKzzI&igg$Zz)~lb+LdW`
ztx}45$hxZQu%fkc`Wl6%8=@Hus7!|ydKs^TcxQTg`O}>7qe-AWpHq-I*36mLJgrI(
z6}qk%Y=p@ElR8)XK!9MR;k@)0|8_sl!%O^>7MGd<JFs`NcajEXx0#5kOB46@V*gd5
zRYW>{F&!hnk~a&SuV1{U7b>7BxjNfh-t?_;_12}pBlxhdi6wy%oSw-7Y=Y>Q1VFMI
z4ZG^559<_U1J7X?@_2|Ww`D4^9n6Nod&iMsoD~eW<I<3To(l^yAu1owm4YyJ+zgN`
zN<q(j{w9WJ-Euc>7da4on~LOK)%3jl!;RgLdsX&+5WjYm?ht9+gfZ^0Q7qqd{k?=7
z8qThOrz3^?9qO69VP#Pr9D81VPa<|2zSw!ngmcV3a6I$v*p-=x`WEh6Kvm6UZ*)rV
z1>B%>sU;eSO5?YPsxM$;#fT^?R2<d$nFQVm<#Z!rrdXN>5iF1|*;3kfw+Z^09`}32
z`_kosS~&?NZo(U>#fJ3NIJ?w9Qz0SP?;X)#h0-9`#+s7+k^}GHMp^X>we^$gI0YWi
z97%H>h~(3B3>S*?-0TTORH9ji1aMNLw%Ni>wcj7FbDc+g5|r9$>j1{_C4hzaWYb>)
z?D(PVInAkM-jgcTnT4r`m3z1PdI?pVg}VXAOH-MHSNz0Y>u)zLb~lIj#630IEd9&j
zaFa3VF<tg_M(6v$WWr-j^H0S*fL&_Q`l|R&89wUsI&I&=G@@B1kmq?yz0&ZGjV8BN
z*@0$Esi>{PA-0V9k<{7EbtQyW;l?%^QLAkVDb>8H0CTL)NRyrwOWF5H+<cSBDxbQ?
zkvCu=TYQ5DAteJTdKa%LQC`*<Nf2)qGa!uC!v}_5hG`Nr8xiDt>zkIEb?pqNqF=q1
z3C=g9*D6#Xqzv=R^i^UYJdW~4_^?$_$uQP`Ik$ar+vTrBEarxFgqgrPBI$pi{H!bm
z!Hf4psBc>%HF_t%ZPEOzJBO9uvApD`0AE(lM66{~DBL#%^gg0>yt#1~+oM%eVIJ0Q
z9_RM5!x;yIe2Qh!I)DROVAaQPEhRzO*78(~)-R6y$-ZrlF6+@M@c=d6l_LfR`O!4n
z_Zh(Yq-Xe+h6W+>reKE1ESY)|5#%vP7L3GaZbp|d;892eeeb%}HAYzHF0C{L49@k)
z26_AyU|{kgLK~2?2|4THMOF?b)syMmC;3YkUa`vP++}nTJjd<D8~$U5{<44m==$N6
zDB2gDJmQK8jZc^iU(uMPi5T#gAM%Zv2?=7Q)f<pQn+$N80gpqG8K}KHTGK{88>2j4
zB?S8}zXO$NH=Dou*=$-_MI*XYR3>Buu8yJP+m)?DBaZtsSiB&<BfU7>y#Sg<(Tq$w
z2dz}mtL=6Ev;I@a5nUKBj!RJzfKoz5+?@bf?BTU(;nw85AH%wYt6u>Us5+@?T8nLW
zjKR?c%k$z<0Rw?Ep%{JSd&^y!j#Wwoy5c42z}}qirPRKF`4(lRr?F`(Xe=jX>WzKl
zzJ`_}=Ye(LYD>(rBLM1(qMjGMSn4h&t4RWS9bh>rKwz+xv{e4MnF#$o3a%b38&x{<
z%tV89faRdRZ8v=-H!qCLdjUSWHZ7?$q%o0$P>P(Bem80{FE&#40v_O$c9*F~CS-^e
z1MQBi1D{YFQlT@)B94Upo^^l;G+0ThH|HJg=TPQi7HiO(F?uF@n9vhUsJcF3VD*^6
znLdi+CmkRMo;d?!_fvl<S7x|!XK}s8f2CMDUokQgwX4^JT$Dk)$Sd|Xs_z(m^LQd(
zsy)H!P47j-gQeWwwBxN?rYwwSdIL(mcPzv3lNTkY=z(Ul{TWoP&{h`z>;n_2NT_9V
zGhW34B^L7_g2d`$_HDbylNM|*uxhVz8f(!A4N$x^cxbUzY&5B1xmDyTxl7r^hsRPg
z9}&IrcL>|p5%=i6dUl2>w-1~V-jM#gF_$#0<nbv1D<nin4?ZjC45T;cMVAuf(y>@=
zC5Xq@1WavcT)XGkx`Z~vp%($3*4idgYqP#A?PylQWF!P7%V%cWeA~cX1jjX)<W3tk
zTkS-00t^|!AX#9LGHPy%=O{S1{4Av|?{o>fb37nRl+q0kxVdbx#s979iV5P4#dw39
z3T?Zw&wZ&`?MH)&#iF%uK*WRe5?v7mR^DzyOwH_Td;#Q|{$ygRle+W(+%%xG045~a
zyYIbz{pVK>qDpU~Qo$2xd04KdD^p)6;G!~PfTdI*0-(%Mu17h7fc)|U{Ib#wD`Joa
z<&r_3=dYixC=YGqXPByGzE>>t26#?%Dq=RJc>mm>mLAA{G9U@u0~X`8ypF&4g1bV+
zuiYS3L6GeLKvKkh$1P2>spTJ!XtdAF6gmLI4_o`A2AQD?V+t2zJUnmz!&B2LVjx*;
z5nk~Zdxcs7ROYhulMN2qry4mW_kDNc;Ia`x>wnGTZbgRI5Wj~P#*3iA>(8g(=H*w$
z0UVl__ex92;c9AnB7gu3^3FLfb>;Oj&*?YPE&9T5I@}H9EauN&*UAbaC)cGJVn9UD
zF~Fhz8$Ndj7pSz}rFTrXCb~~trt|!h;4iDUoKuohP%ezXUv89%KkMn)Y=vqZaO0H9
z-|~O(Lx9LUwUP!9Ybb6mP?1@*;bo8_oC7H&fQN%-4IO|`TAp)v3;(vU3vU%Pa2~7p
zp!tCJZBMk>LVE&AR(68wdGUgvqSo;LK-yGeMY5LVtxLh>CmMsZouJ$ALuJ5hP=$)d
zb(5zAJ)5>~zhF55Zu%UCo=q&u(8Wn|=W&mAN-M!5AysT5_BDBky)6X9w%qS)W#uRG
zW(eHdEW+l9;Q++#oxr2S&IhCOK{+M~=;VtFH3K27Jj4)@YW$lQ-~nth6KB)$#(W0*
zd`T9nMzj><reuOko@Tv(%r*z*Dp%Oc;S#@t7FWbBO0)<Yi^>^{nS#3-2Z!k=Hc0^j
z82CRU9K8y!-~2%)=v$HvvCh=!kEe<0E<6{yay_bSzaYfZr-7r;nLC539nVJWgj`%s
zQ}f};L|uofQ68jvhU+B|Yeeq_mnT^Hh|pYM;B0pCZGwKD5CI7%f*nL%Oc5e0`^Cta
zXy_n0b-s>>lJDi?KeSJ){I*tN3>HX^hS@qR-R!OsH=%Pa!MtLV>HJ?vQUuF_#EZBb
zbFrobT|H3CB^2#xdy?H56`4@&t5tIk1iXKWyczazWhmJ*aU##QRA2r10PWMwbEPXr
zeN**SzajRuTv{6`QR3B^beUjd<?>%$cVK$~g$XZ;=j&ms3$WDLvM2K6`!t8=)0$Ee
zacX2hQ~W++rc#M#3l8}4@{GK$>_)g=Y$}}1yQMUCBOrBd#7`g={8XB5&E|Rm&;I<f
z`g^p}&L|>CE*&|6|0Up=4lJu3W#l!z_Ay~4ChXN1%?LkC0mT?smk^bPVJti)awS!H
zllZ$8H5+xDMb#vxfeqjP?{dV5ff9GFCge-1o-_2m95%r{PTn?d^Lw;qA4lPl7<*uF
zEZrs@X<CWnD(~}-`6%iU;d6iF@)w(Dib90w(W@$Ut{vg$XN!hkUS#?`?GOOzX_=ar
z6oh$RSbzr#Wwt73^_pbeaadRFEW1<zDSVXRBeG)|xrsFpbcj;@Xu?ufzh44rs`9*f
zN$&!l&G}LJ0Aw+ToaON3B`2VLSE(JHAchDW`jIj)|3(r!q~0)k<d%x-aglThlO{1!
zdGtYpDNna1Jvz>_`&2aNyFONj40zZ)hf^WTLr5P8J+*b6q@=VJTf44k32nof?JBW9
z=C7zcPw#js-5{&?KhVA4%g?N6cXji{LnB9LZ{BMmeNU7SNV+eunE1nH=6+xi*B@pY
zC0+F8u^xX$fau8O*~d#(FCwPm=NsyB4iH~{W+QLx2mJHH0C#))6XfLu1!H`^q3^5@
zBT$LKVh2*j$nUy`?e`~RjSC*9sA+r>CvFY*ixMP+VpB$J40{g)0->C<R1!e$6N2f}
zP#kluaO$oT%0-_R^H(YeW_wPS)6~Dpdy9N7;f}jKX)#Z5Aue^EbpJHOlRk}AfE|HG
zWjMVf6?003J%Qvo?h+cP;J*8|0FX%jVZ(zX@bKh&vY=BypClkGB48ly4Zo#;+a|VU
zm|=Jalw0pQptvDkezz29pcjgazJ0HcbD=}>!ph}&ad(*TyR9J5wI%MBrl`?=5O6Tf
zZpS!K^9<%s%vcg2XW2Bpl*R=^qK~eApGDHI#654U#D)G#i-V(O6}P<zzPFeXedYhj
zMtc8Kpy7bp<fNyzc$taRV4fT)S7aS(Gl+rUV}_%=%ZQ_4@N5V?PqHx8TI6_(a_EGQ
zPG*VEHcE;8u6l?$YW-;+&*gcZY~d8s$dn89EbZIu9|pLxC(#0gyi*d8G1ct3p94hD
zyDmm3X5I4IprEHywnQeb>N%m|?=}K&ezt8gouQ6t6pOq}56j!+>G!u4fGQCusBTnp
z$#h<xdT}^FSjwi5=FF1lj8WaIvs79cRz%knK5M^=e44E~Khtob9}g3YF`YC|0j84$
zGW*DEZXm{UG?SJpix%V9bCH3k#RkmZR3MP<(KXXCHv*;&1m1PHP7(5~78%?WK1<kp
z{n<3_kjoEsy!<3Z%McG|)D!2Hs-Nyn-Xd4DHNlLPf2=+5yz8=+TpCnNt21`kws;op
z0h}Y=4z59?@eePv*6wR-Y5)_G%qOgF*HpmiQ+hVU_4OfNt@EY0hmO?i+-PjvXC2Y^
z9uh)a4v!WpItH*N-|->?CiQDno3Yr=nnpn(?=7ZZEL0}k)S$0%=3XD;{4;9%LLdM{
zuJApWbZC7FGsx8?001V0iUG@MrKu42fSX$@B0LY;$_*45|F?;80}){97rtEX?n?1p
zg+6ObaJ=32vHFTdb@gChhr#T~h*WgNAJ;6RRftdKk4h!KTV>MLZ+67{Xzsjptbf!9
zQ}Vrd(5ORvFFWMa50Z~qUff7D>jmpBxX(V|k2)AMA{F~Bb^4JWLJ;Q*FdT5A$^(yf
zAd6PTD==x!TBlQ=S`B-QO;m2MzJ8l`3hGNu283AL>2eg)djG+iZ|?=w4gOItrcp^M
zTR-NBMkHBOaIyQi%MdZxwE1QN{Nr<B^;#nhi^P5JY$cbf;ZV<ut}&?~uFAWdo-v`r
z$0~5%KiW@woaeqNLtIhXw=a;b7@tS~Gn~{Af|8&ZLCp<;BHy;)1;mT6PwfMj&v1HA
zKq}KD6qxbt{tr9eSs~0HN~U~JIu>-i8k+G6v5Xs5&bBr8Y0WMuo$|XE;c`5W%qc1t
zWPCz`FIPNjmhrv(Iz7N57ka5s^gv3|CuX)HU;w&0n+BD>eHMsK!*Kj%IjM4aijekn
ztGF9460&#;3<@FcBzBEF=3xp4o*BLo@z-fZPcY7k<XIvp1TehVT6RZorDFcor-6W`
za}*&i%V~TCUNNb!KX%pQtDC;nq>iHAFROoBF18U8Yn_yv=?`ur>Q^MLn_;Eb3^Zks
zbUrCia7_j0;=_kbeQ@`|Vn!c>y;fj;6vBbzV^o{VRw#4z0h2=&FAIUWZo@=oe9?|?
z#zG;k55o8On05=^LT*5no-e|_?B~KWVm+@E^J$LpooLB~fTsu;or3>c!h~47IADT&
z@(LTRXbxfgF7CA8NTa!ojdujG4HMW$SP}q(%vFV9&E2?^{jI;$fm<*VtUD2WPC?w2
znNtxHeX1$ArF(Z+&hjp)%XQ(Wa?AkXhvi;kN`iOrCvbHjD||Y8v{i}XreyKNev2<q
ziGoeZNQb=0U!54zBxU^iSNWc*lP*`7T4UBih41m6a|jh{B`!;<CV+G9q@wn%Fm59$
z+^zlrtB2O&K5#1m17QidD`PCu48M>GE0NA2Gesa}l+Vbl6Jw?23Y5nxZvKXHfQj-z
z%o=56ijV+Ix~L9<N*T0d6(UMOPJZ|O-=*pXUI_JMhTCRp@-4t5t~u`N?iwrv<wTTt
zBmf5DJr^+OJ=JWQ>!EwE$R(j)7+t<@Bo8yj_-F5Jc(*p8u<HSFPh0J+mt7FfqjcAI
zLgJrVLLsisw83H#TXi^sbU_QmDaP$Pk0$uhcnKs^z$4o!3ao*rj`{#Bf4gZ{>7TC<
zSQk3hzvPQJ6Vrvy!NBC02DZdlxB*q;pP5Yi0zNiD*KHz7tzgV!(iJ=^P3Mp6PMLIR
zrSW6h>zNL%0I-K<1YXa?8lLt1AINp=LzhmBV1f4&wA&zH%=~|P<5Tud0)L1()!UbI
z?E6(KxNl~pto3R>?2gDo>~-H;R0`YixHHui47(cL)~noc3Hd($!LWJss}P7dS<4mE
zY1DL=Tu1Ba(BAEAaBA(_t0!C7a+`TZ;zZOTUIVsV>iWf`#u1;-_nkqN@^%djrC#Oo
zY_5{eCh&XgwrCDnSD8;9){J+#I46Eb*JVM>*!kF@W#$Dmh%3Q;a&S7WZc|IV%EJk8
zR};ic-9T2mRt2s>NS>nHF8FJh$G|#JLj=1o-^J1S2N+NDU940V8_yeZe|L>2BF1PF
z-&>8rzueA!4uea{es!uVBGmq)U}#@U`HR5AOn9u_8^kvxV@g>kmh^8J9ck?tDPXyS
zp6&TB?7Mqsn5s4Np>TO904m4&N*8bycaVG$)6BOKu>Rwv;Gd+KNObk%lD|i31E6qH
z;s{7L?sdP?7X};677iNvF6%<b<5|3QMCUNE++)t!>Hu;fb*{0|$&vGg=~(x0jj_g2
z<@B`rOeA+iS#9qr=;_2mq<$_PK1aceEJ|>=aoSeNHVrC3ckpai0uLNipD+XasBOQ#
zZ1(fA@cZQv5BMAB{d&gFmL?<YQJHaWWARWK+f@ZLI^r9Zaunn&=2d#FPv*LyB0R_B
z+EWIv@3f?P0k6rjGmb17;tuoeMc<oG>U*y_IM6)rt?Oe}!<nzY154$5%`X&@FX6td
zlTXxoYtan^1ObW&zz@s^rE>v8_mdJ~kQi(+Z96YwhdXGwZWCTT=x!C<@T_sh4ouqC
z64T^x_1#@suU1l8Gvo8<c%kjQ6eYu4f^Mp<Nmu)hKwakJ3-_eNFT$&F<dc7DqeLc^
z@)SDbY1)_!ZO;Sw5?_xN+L<SD<m0I5>+feqvG(MjZpW-j>YjJlbEUo`_}t}B=#l5f
z`yq>zH$P|M5>9<-xx)DQN559CzWfApSlxIecxAMw$D!hiGOnWNpZP(PJ-1Nn!3qri
zQk*kfGLyz_wU%Y@rga8_z=vY!nNE|<29nYibHcb&a$^kjG?_FZXT;}UNnpt={&ci*
zCS(RVtWgcS`g2x#6&B9}7g`NqbN(A6vU;p674kXLp40`2My?WX2L_uqLQ`KWuu6+L
zz)#Ke(@g6xZZSaH#6smXh%o%u9NVi#Y?>}ko>9Eqd@1P_zb5(7;{6*NF2op=q!fN-
zoqXx;G-Tow<(QpSymK+{8~>q#X4a?<t~!($UmYQmK(!~UNT<%3G$3TT5ZQ2bpsOX`
zSGP@_S#A}>6JnnI@rhm2l;;!g5u-<o{#2f@H|SN%qk`=KEbomVIiN0TBJq_#Ylpz2
zhoRW-i8SfRbud(g2<nidj^OOp)pKq-RfINrqZ|et$O$7ZmHCYwXabv5;X^W)SvUD<
zb6{;l5bZy)*~#(2rrG*?e0kYk>+}PYnh!kue>OTNh@>5bJZs%&W4Gou^QpAFN#4@9
zSqbFV*t81sjbB*wefG2S&fr;wlzU1~IY+hy?Ukk^+>67cUWbC&f#>P-s>X8>m;*b>
z4Oh#$!OwGf6d`l}PFH0NQl82!4S@o@vhaVGDra$~qZNEVp$AaX8=}^uWOMgeQM$Ln
zv8wyNH2suO2-B3nP}je;#oU1jIL#-t?Mwm3v<>Iw`cjP^iD2;{MUQJNpf2Q{Kb~4q
z=nM#vwWyl^=UOxN%>bop0d_0#jsr?Dxji+`-Qi&}Gp?jfZm~J%88bfDCjOYaE<#tM
zAa81Ukrbn<{^6#ct+EqeQUB$RaWy5NpAgiOC9qD!@6Reioc`gC3plFUdh}!JFswE2
z)C(XV20<C&0QW;d3P-t%6gG{1jm0$Cf!~B)mJ$rl<{6r1hudxG?F5%gDed`)c$VB0
zC3ylw&|Rc@n!MYa`U(!=2vGg0Gt!e5cYzdD>84NC>>5uEtw{^K2@qkN@zrVeXyqzi
zmA7dI$MM&4<ePK^HZkxY{Ec%s2*f+%&}{<M;j9T`yKwnMh-Ve=hTw9eUi<5%Auj5l
z#HM%mNwU^DXnqL9YbQ!rM#7-TmXaB^w0|A#=UaN!$NY-fJXU_0Mks-Tb3ggiq!m7T
z^&&4RuhXX~7N}j^kwXA$X$JY9989q`JsZq>IAFr+xKO%fQk|{&Q6Q$YY>hH(^`xcJ
z5<mYy9un%6g)6QU^&X>oS4%d0t~c+9b0pT8=Pc_Z+obEjqp>5>EGhArRyjPE;%k5w
zjR{ztdo<^iGxYo{t59!R$>QlT<L%o{P-TH!VFHr(l>DK~GZs5({tB<GnQM@t!j>gt
ztyzHw!8?|niz7{sTgws#7+5jNZ{QbLU3lnUR8QEo(!mF)8pEINSy6_N0A0Z3*%$gW
zPL7HZRL<G%lG#encYFv2&8ew>PQ^-V$Lp`k_HdC3=dqEDn+F7X$Yp=>qUzViNJf5o
zfn7w1tL#^>K=vn*#1ZS4d|PRddu$|+iEHy|m&DM@WNv4WMVt$0jvWoqLD@o{f8Kxd
z*Q>_i#w)>(Na@6cR26(=@Oj)lv4Obr#yQihD#y#}w=JI^2(Rc)dWP=#D<nY2Z-ltl
zYa?q)MdIA4e)pcviaK+ssI;p2RZ9QD3ZZd|J@p#_fT3NN`*oSZ(PgcgQ%Cg|H`yC<
zp!JtoO31N#%^oC_(@Emn>ik&YD0hX;1Q0bohrOA?`;&E3P}tv|ch^PIKw;iD<d$n}
z!lX<08S+tf{BV-+uD96OHWm8{-y>fx-!zD9-o;K}>Kccec9oAUWb6G=A(#(fFM<CH
zdjJgAEYPL$gRsQ~^2+J{GQ@2kU_jXtDp!6uhgjLoKV8C?FcRz3P}SYT@n+9}Vgm7I
zW#SiS9IK7_6owqJ;pjsbp46W33AK4VW-GUSs(XRcL4UC+VzC%wvFD@o@NR2(1%6BQ
z!2^3f%yZ#I+8On8PyaNV9gHdzzkgVKr}aEwg;fkXjQ#_K&Z_8$X@iZH8~s=;k<G#$
z#R7)W+Wg@sKM%25+nFA5Hy?5{yhq?l@6ws;5<a4wh22)P^Q-eKNdiRFi`g{sKzI!h
zaB3t3#9dD-Dp^h1e&bP2r&y$4o$lc?nN%Mfhz2!~#wO>nPR+t&2=AScF5Nm?r;V@s
z?n61xz?5eyoJJ7=p&Ku2vT{8%3$0f896sp*V;u*S6F2P=1ze1T*mmh({rcoxa?8DB
z(-(-X`76b6C6j&oCtn)aUhP0fWopvuUenyT6y-=sjB~m00o&atti(xnamX=qA1OqN
zrU<9*w&uLP3>+AWTWxo(uviVN5hpICo5afiSy4J#n#Co}{Y+js#G{0plJ~|ylQC|m
zOk<alh7b#Hm@V)Mr1!D|6gWG}IES=MBoWv9Tp*&XkU}`(V0A^dD|e|_cv|vf;w)yf
zbo~03W~@2}1&$lev~VuAH4d#S{#<ma>!18x)J58J^uwLjbxYo*^-EUO6`3|&<}I*v
zq%PNfX&lVn(Ng~yU7E}0=qx~BEr9yD*!xKWQT$cDx0<ZbNmWmPDJJfuXFcgQfOF)(
zFA9k-zIke|9JldH#9vC`w<j+o*<k7Gw84nyAL{)+llXUCy&9C6jzyXt-Boj@+5G#q
z%OLixTVdvTMuG0#w+Ea~z8JX^k?Jcm-D%=6@F*6$#8z<(IKX<X7(Ad4SY6^AxIof1
zPu+;p6}RI16a4OeEwHLoj`X|;78HapYh%R^7#c)AS!cF(E4p?-&x0pVn(4U#M1hUG
zrQqT&PC<AHI0wQPVscEcDNs-4vp&jEUqj2Nrx&B6r&7JMjG&;u?e9^;yx;0F++O};
zH}Iqp+N#_9`>kf*{K^sh184a0%CUOdXJaJ3E6Y`)^WN!7ra$?~Xsf+V&J0D!Ss=+s
zP(Y-jo)4u4aTA6QlmPoKwV87vsZ2uRKkGc?wVX?W1^7+kmW9IaOIKY<cEL5N;$E%^
z5ZfpEO$M~3wJ@39XAQL`<rjPDB3~r{lYT9qwFzB@!;K41TFOqqNJ-d~MNX-EeDe4M
zu<||wv`CjvTlcXJy{)uEG-$U&p0Tq0iEmsqKNvMWjq`R48|f@0t+l9iBEK}kZ3d-t
znHVv}RK7B|gH2%v>wg;eZR^0!r#?arhs}xJffnZndE7BC|0!Gpw_rnKnvr#%h>hy~
zVjHj*vp0n{;@)<>wdaKH(ym8I)HE$Yn}Ck<6wMIB4$PVWe)E=vRp?7&|50>6czfHv
z7?m2sFvJ)bo_DE?h^jF~48Iu#dtI^-H>Vgpcg8T?M1)}d1?(mXMOPPjmRX%IGwzEt
zCBI;bUYWM5-R+^C_irJttW$mFw!XH|I&8h~GnmhBM52VWii^~VSh5~E0IhRb++xGC
z>Fu#iQD#YWY}|;MROtD!VS`!OupI>KiHxaomky$wVIKNDAa%YIcLkx+M@GL(Z!CQx
z!(AV!>LblGNOxK5aM^P2J$((URqXbb$)q2bWM;aEiN3v=Uvq8QV)=?hP9=r=PqWZl
zsA$ibxTRKIM&85ngXx7+T}uOln;l-^BOTW3J}ErUJ`?Fbn2W-n%)FpA0g1+6=guy@
zHr?}2a+`|VGIECb3F&Ti#<fG{E_#hxUQ);E-z8>SHg(hLCrr0#DlIv<*Tl<LJ?kbo
z$$k$z1SJ}kfzK90NSkw8=^pd_!N*Nz-<^2A|1v?gWm{fr;%b=<UU3!#DNK#!Ne5Vu
zkOMowqWTwDZ=D7Tf`nI8>pNLPI2_ClXrLm2??a0ZvO!N|E|}2q)k0m@xxH&h?KeMf
zc-AqM$czy^aJ(M=DU`irXySp7r5usc3%H9Sj~bW(he(ME`$_L=1HfjK?q=V4IN~bz
zk)@1Yoo}*)va_1{dH}2`Lr8A<4${mKtWo{6tCmFF_w_6-*tNPh_>>C+KB+*8Fag*#
z%G11gg^~|(CnrUx7l$;y31pcq@hND1!dYdi<&$1*j{zXV!k_4s|33-$L|~qGC{_G7
zF);Ye;f3UYt|%En+O(I>Vn^je#LatZzD;P7_Z9Igr`!3Ah#)8I{v^8;{^U$CnvJTD
z1)x#2t8xURnb`J9{A={!U=(8nIA~^oBgVPp<VSI~%erhmv;b^x3sAd!lr;QQI|T8z
znRvm$<c>!t?atO}%u<vpSQlyh*5So)uo@UdUu(s!NCZkx!?bF@pC3{@a6ekh1S6_o
zPY~y~K?lxYs<^{*Las)u@+hA^29<VP4|7M(m^{+dYbJ(VaC$SzrL3MJj3s29yEZFb
zYmhqMF=#&`#r{F^(gIoYL)^gal!oX{<)8%Fbv&4V+QpP51&Q_~JA0QQKX1r7iEtkQ
zy!oQ&wJN}|60nir09brw;{TkvaPOAp%2x(K@TQxCdye20$ray`kF}kwp!`=NlW!Xx
zaaED^#tv!n_ZsP<7MT+A;gcdIZaF?NkMJCT7Hp}r6hoc4-6k0D*v(5--eG2eT@%QQ
zthEPcH+o-_4OTUd)Wm#R3F-3DYmNV7-;@W1LHOBupeID$-xq%Z|9jv5kg`P%>@})4
z{nyaFyT<TOnP2ZyGS+XriF-B1kh;aWI)>j0aiT`+3jgYYrxY2#r53I`d|;?Qjb_ju
zUJRVo(-!3DyQd@qqO*;hG%U~F@Zwp(G6jGkPpabqdo_pG#R6Y3bi&$Mu{CQ-6SdRx
zv!3rkK;%Pkw67}qcQc<La^*buLfRp^lRM_M$V?88^1b6dVSTRokgdWtOBni95sDdc
z52weWdPSz~-h@9n3}hr+Q?C|3hfsNNXfn#(7=a7Dg7LGJTP*VCuWFqa^M4G1WV+{j
zUT#*6tzs8SW{O9Z-v5d#O)J&-`ULzv9BW0tqMGq9CfR%a?cL?QZ?X3_u6|*x$Cni3
z9Yd?a<c=E`znEIdY~AZX`(6A#UJN9SkMnWRCU<--_Q;Rs|A?|0(yKY2pBJ+R;gR{V
zK<KiPQl-tHixVu5a}NtAK3|V0xphm25~k=)Eyyo<DDn2n>x|(!ixnY3cVm^5xJk3O
zr*uOHj~Hq16Z<Wdk@}O702-^M@*!MaHBCtCw8y+kcYd|)4=h(rHu!#o7>}vvQEPIY
zs4Zx%YiRJ8J$^dC{lcKB?}=hY5QsQ07et`^iYGcKnN!OZzqCaMFD}+3y}#;HjyF{4
zKkWe0vl^?9Pn~5$Y7|>Rf+@UqH6t)c=%Ew0sI#d&EHCJ6N%Qw<&i2ghhLOaB1HS9f
z!2DHdM2*NAv&5IQ8qB~^YnE!ux#O%%=`tQ}PJ_~yJ!ruh<GgfKyVZu<+fFvWgQ^7r
z6zS3HdP}+a>E_ir(Im+?<dkM<Kw@&YfwEZ7LD4d{4|KeHZ9ymP!_l|&0@p}yVgm>9
zUFAbdUM;jR`z;AIej*O^pC6y^R1?ym#5@1|lJ(^+;t%O&)j|W8&ymn3RiNHIQF8zf
zl_Y8U#E0d^hSpbbn%ySe56wQf5E2~z+-+4zb6i6rGfWxT`7q95lh2B^I?a}8JBQ>4
zIyxaj$|onzT`LxQFiGL4ja}BziF1l@ip~LKte&dE0_5L)q65gx4G7Vb*#nBHa*0nO
zaQ}|6mK6PJLY--8C^FS6$${6YppKQ2*g}Xi!h&mudMjZ@GeIl*4ko~MDa6s$jQh>f
z{48#r1=WwiRxNJOTA~iD9IR7R#YZ%5VXg?4L{4(_u4dedh6%E{+df&OOoO1Ka4>X@
z;loZ5HmI(Er(}^kbJ%|GURz0ogHnd)sU2{Edvscfz?DYnci1zyKzkPsOQ@`vRk^S1
z3w)0;54b?7eX~%g?sv`kS;6ju^jBzt)AsO@*7MW_8Rv+2gDMsfUaFiP@L)7Cs2^?T
zci)CH&K?u0eKy1%rz{;z0me_@Dz9hcJ@DBQ{o48fj}GEJ&N_<&WNz;O!=1(r@X<po
zci5174z+2JK^jjNNPc(6l)+nS^+F<CE<9mugo{?T$dQ4Lu;P1a0$Bjv3K6JZrcV;&
z3)b7+X|<W~@H7BN_g}waYa(&MIyNDoJh$~`hH?h-bD%=0htDJbFd2)E0f}+E!M9s;
zf6v$a>-B6$=4wIxL6g6@&SZh1hv=lv!>XVuEf@b@l8~y2wGcjTzHBN&`AU(QLTM8B
z_18I`#Phb~T$cq5QuXH$>yrOZXX$G|NXZO3%I?Egk!aQEwxIRz<+2QQ!yu1+m!LkO
zx3y_<MV=<-+^s0<oYU+n!*}KFz(WZ+-}~eR6?<E+YW^Z_oQZh-&r-!YbtBmJqbZ$G
z6!G;FLB+^$O0!Bqv6vEa?soWUHLGe-?(F>p4bflpuL%ErDQ!%O^v|s-tc|IWZLvjs
z&fAgPR{yeZ<2AlQrc)8mlnlt7Ppim=L{Q<BfZ6W|#gD^WTWM!-evjJtIfy$gQ8r<*
zb>K>v9PciVD$%yz759wgSK4?<^!2AF&qmq%lB&<!ZW~ndC?+*!MYgF=kq0HnqQuS>
z+WSG<{d<~%wf*2=;JPcx0Y39UV~PAVALZc&i;n-NI!$>ej7u*_86inqHV!SO^PQwY
z$-59hkH|+Kir5%>K~8}@*8miT5=7KfshkOcOmkJLy@`E%Jj*p4@i?QH9iP1C6A~F*
z46RZ)RzAdX5)+Q=sZJr7Bi1xWxT}GY`-+!;$j^3R_Yr+@3NK4<jUut(>d^M=(2Xla
z92jP?c8uplEmWw>cfY3uxEb`XMQ6WU?)#~(e}Vcb)0u4t=Qc^Z&UJykH>_{RF4Tn2
z9TPnBA3LT5gq7`7>IAZ$pv{xw8lg*kT5l`{YRkEFKx#Vw9R}>c9tArGs?7p)8WxvQ
z-uRpOjil)82)rre<f|kO<C90siG0?pvgbPmj52M<K^7f$Wl**x_jdPK@j$vAlIPaG
z{pL)p$OsK3Xr3lbc<d!5BJJ9$QS9<A@H649mu_Ht3cT7`ik~-_C7oX%+n!=u2>s#y
zTvFu$DNnYYiEFs8nH?@6C$JzBkVlR%9N}uUfzKhQZ^H~Hm9rRh{F~X>?p*mvz^mXE
z{8j6$vz3~jO{5-U{<=9<+d7%?W*37$1+Rb;t+x*VOh{(+d~gn22SP&u0+ir1Jct(P
z3{lIQ^v4X6f{R1cz-yYe773G<sdMZO@zeV1(mj!0I!mkAu%Tebga=!Q=nq<5H$?5|
z5=_H8V#jEEvSvvxqH%1zh*a#kvF9v?w?X+6gQO9~MtL5<WFyY$hgYq1aB&R6I?zty
zip<>OT(}<E<frU)KXzGgScAxypS`RGdPweDg_07`Kc?_`89L9pI?0Vad(toC4J~cy
zubA*t8^L!+`R-cz1ijtt{UNYWx{rOD7n2HWT3{HNskFO$y8XX?6QBXvy6k2MrG+k>
zcL|CKQ~5e7u?@822#2xyx|D+GX!UhVt_0l3*Vl-^rW-bqd|BzYC7!3fPJgKt(4F^A
zm+{5IEd%ZN`ye9bevSsBP9R|gBty1@6<r|D+ky~bAxP7p5qp~oQamrXI`S&{k2|gA
zB+3K5&A)KbOFu=?!-jlvoJ0$zhT^Jwtkj3qBV`IM!?bqQW-;4;(?T};gX34aw&MLr
zP%61}Avr*!<ja2A5EUh>bf88_zIl<;c-vQRSp@1YSSI-rfU@RhZ8Mb}KKnmj1*FNy
zZ)3n5z|NpK(S-5)UC@&eb_uVc>$7aF{L+u(s8je5<NZ3(D$&u12jRSDONfS_j+KdV
zMd-jc5ymmSii++*0WYW@A=ftUwBkC-tp^^bCbCZ57OF^$_AK2I(z1Qne&8Ub9Q_WR
zNI7}Wf%n3V-A89F|5On)Go;zu>oy&paQLf<2iNc>WiLRjV2ktvE3&yw@<{tY&{O3f
z?cyO)3tqG;vD4Vfl_>{KC|6gUf!Q4Uo17cd`pXXtf1Pu1I;GMadq_k}4jy2kqc{|5
zvHEG6a`=d1-1Id2NBO4eck)B9ofTrR`db)(s4OV2>bT*JT?9q*`PoQf%S=(i+lEtt
zq>b>qO(K9zf_Uo>n9_s9w%CGO+beU(RQ0mIcJhxTu1bvPys$P`LNZ4%fX)iAJf6sb
z**%;CY?K&BKVF7QTxc?E`u$W#;zM=pe-h4<?joJ*WM%<u)1WM%21tH<$<?(gLu)J;
zzmgOEy_r4YCVP^CTK%a6hoLQndspX%OLF4|T?p64d1C)|0CLlGO)PcagN)rDr02Xh
zb`J#08Ul+*G)TUI&OtOeiMLdT4zSn*WqC)Gs8f}zar*(o10S2y3#_?0q(8YEOVp&x
z3CxEkSYd*|b5<CFLT46M2T1&o>0&`}!i~*B$*drqF&zCy3A?iDT3rO11TP_4&n#0$
z>ZAcDp@2J)J6j&MG>I9_{^5}!zMOt@GY+dAa<-GrAdSl+5m1QN^MPX5svb>qW-=|@
zIBuitGTpA<;W4ebC3$xF@87yAch)2pCq^X_ry^1Fb*5{L#ki`LRt*m?)%S!N$q)XC
zk&jEY&fv!VD^WqWcOh6AIiW5YfZ5imxaT<ADmK_U_`3Ct<pVbWnjPn&+)@T7O0}{>
z=4X(6{rFMkw9|W-DmV5FdxWx52vRDjd+Jp4^wZ`Md|8IhXd2o9#;*3qLl31Vj0u}7
zkD9z97b#0SIZx2;K|nPIz}o=SCc!Zfb-qkZ^%(IyqQ~I;`<uac5(B$Z?k0Gke0v=Q
zp|$^5@Bh&n>##lL@waI`SYYq`A^XO$Rlq_?hSv5<EWOytsl74=VX$6$-0G*R(lbbx
z8By|WR{KzNv5F6v{pD<1zPE+Lv#3R|mN1TU`o}De(Q=UaeJa)AzKhT@N&)AzQB3>8
z3@QBFihk<fImJ`Tkj$^yE<L0=_1kGF4^>0iT*#lUNXyg?IBxUoXzrKX708DNh@xcS
zXQIVE{|(8Ra`q6d2jB-Jp%O4p7rXQf)Dp&0vqVeDi2Kzgq$4{NGSb@R5J%u|?>@g)
zbmJ;`K0#WYvrh^Tr=e8vC;^D4Gd{PSrEC~lPnWoG{+y)=EA3&OVs}AFwPSd<)!1F@
z%tr#7y`A2}dH54&Bq%4-iMu%M5_$2?^9AWY6F@#^Ro=m*UpiOlnR{7ZBGJDBR>u(~
zA$b?g<Y4Vo&yL!e9+UTGp3eM2`b1O*hwj<VS;jFP1;DYht{!@3DQ%OF?quuf5+1$L
zy;NsmWZ9^v<`OZFssR&qV2vdi`WSpbs@o`gn%0D(OoPQ+yr}ZTp4Z2x4w!yhw4Rn+
z*#)_VM&;g%yCcp6_i8F>6dMekyT2Xb_3rgzIltJig+vi4@IZZoU5QYg>FOf7`?hM&
zlkwYW^L@UIwyC$10<oI3dtC8`{X*<zlmKAo4g88FZu)}og<)!EeM8$Xv3|#LpBt<0
z7di%kU1x1M@GcuV0`y~`Knwyv?LdK_uAt=|^~a^um*jDy*3Ivx=d3(hI(<j_?KvNR
zYvin3EJl`Z7Tw+sE(PB0nAATj(Yeo|mGA>KNm)v*OX}RO7pZ7<g|IUfpU}Cnhdiyi
z{{y9>dLM@YfE@mt4{lvn6g-Ni2R$fQE6rH@L`Ucet`!`CDu3_1g@K~EE>-mfV20kd
zaeMJa9XC{W>ZwE&WOJZ|!OQ%>Mnj>Kh;n7BL!qzda&*v6^Uu}##65+sy7A`Gg5Wo8
z!h!lC6g%H>^ViNN4C4&W9fOYrA46tk6h|zuvJjJcfc*i1O(361E#D@Bl3-{MM%c+S
zBMbC*;+L-tZQ-A5t$Uv}Ijr|n2*r;ckf2LX-<?nVA1HYN?Yr6Xaa_ELqS;K6osl+R
z)B|_81ue9gaQ9|LeT_uzAOI_yUdo$QNS!Z%6SIFT43`;E$$-S6oxA0g;B3@Bk;NY;
zlho(2H0SGKjgJByI#DdhC2<&z8YYV)*TE97$C@@%8hDs+t`SwfBBQMqbV+0vruERx
zU|hkpv6zY-N5$BxpgiBd>zEPDAl@4UnlM_xwBn=;gI(O+*}$4vmBxj(!*`4(=)`0X
z7sjukPM@mYUToZrlZm<bIPPoZ=H0U+6$)zohUa;$#_Kiqd;_AnTiNa#v8aP-BA5Vg
z@jimgUytzO@WlHaj$&4g8|r?xTBeo`-PvQ>uYN2nec(22VH&J+KPww?G1<EFMdAsM
zG(xwg{ytl94RT;DvtG$`fVpq?A<HbHdOp`dzVWB$nNa9lyZ$n@bDgYN8Y)yQ9VsnT
zG84YK;nhS9HH!>Q-~S#r+KTrJsMEl)KIU)YyC=z|qQ%@rVev$@W9FftIdRCb2*#ms
z#2XuLAjs`a=dL(CTYMEL4%jyK?fVa1k)V&Mv!5)X($&#|XDX|YF&UPlrWf@6C0{(&
zyVNZ7)#jPFETQclFdG9LhQLk1Zdw@``cld~<(8pSYBb_Sei8;hQu<Zj(zM~nZ!JoP
z{&$aCTakR$Iw&@aV~NH~_+*pDFy+@s_M@%l32wmq_0c^UvoWttPh#0~9IrEN2f`ar
z24+~c7XfqZUy8C;WOt-De$ECui|qO+9We|B*=CuL1Pnv>y<HFy#59TA@Vb$D=!7&D
z2WHIPQpr=ccJ2<VXf@U^$mK?*&e!v9=8}urIwm@jBCpCdsa`yj)kgw+*-iH+Vc$cz
z)64dCHj;BI7Q3P%7q}hL1CGxhnZazAZy`Crt6GLhy0AEmHE=xx2K#6zLtjvDu|t&y
zXHGk6=l7n9bo}b(et|j+cGCmO_7kBK5On}_-z7@f-~mH4NS6Q=UF%Y{lUOey3~rFS
zqW)mIXZ5HG@3{zLsQ(55e~wdhUFfw#T5PfW4nJ8eyjtxmGq_l4QlYap;7aY#7TDQE
zJarb&7@8b8Q*KG{Z6=Flx{6($3G980AH`YB>^VG%6`R>A#PK{=xO}FPrY$9GRA`pA
zC2UkYS^MO<LiE*{t)KI`dbN?>&Y#x;z98b1V^EgqFS=}OuP)|rMrs;BWu;ROoJ~Jr
zW&{}{%9I?+9CUdu&nRTi!_ttakKfZk0WTx^=`C>aD?R!#L*zp_16iYuVO$)Gl+h$$
zvw+IBv|B8eK9<mMKv@Iy3qWyJ51@)=DQ@fyOd#zg!f7i_`B8k?y1eg#xDo~Q_n=9u
za;LIuO$<t|3-jmOwx<ctn(QpCNp)M#nu-)6sKu~}`2|x~gtg==;(4Wgf2`1P)ud7f
zzYwgvL0iyfy&tPao#|0-p`4<1g&RGGWvpBu4-P-lAZF)qA=;U}E0-Cg;Aea>sevX(
zjE_*Z-wj|kO}-u}1cu-7D`8@>!Dc7bbNogGsJ(u(SyX{b9V>Yh&UN+}g>YNi&3Qu8
z{NQ`SY;~k@$}j*^4dVHwmxXbu-IO~e5MxVMW{BUinvLWCKoqNsqN|rj{Hxcr__ad?
zo4?c6p=;5qgwTs2MJ~?BNoNkxknoL#^45E34J~$`Gj>ALT)1wM4DxG>KLWFyk6f!~
zZx0_!B_>N~ck03-d`J0`w<x1*4&YQbJ*aR*Oo=5ImS5C4vZ%&eX3`_q_45_BG4)Q;
zhZdWf>yA{^=XZ_&T36TB-!7yI{WSaG{w2<JS^p~hbQpJf1nj4`!lV$83J2}_Qn#%4
z1s~hV({<(*Xy2|g=)?41VP9BXJ?+e?(GN3nxOI91O~6aDMIP*3^rn2Ps;vI8^OafF
zR$JVpUzylza%nv87dzsIEU;6h8oU&sk30Q$!xxoHAY`sRBnbaRjQnjv{2Tv%VInV;
z;ps%(#Q-h6gRG!m_sy$($jI0J5DrotF4nFKjgnEJE7Ow?F?{z1Qr4Orh1kaSocP2>
zdKQ}>tcJ{%=MlFT3nPNzNA5q@ss1+a8uhf1SK_>Dd9E-yG@e1`dSp@JDU5hK_4FF4
ztvweJ*gjH>*|e^F*Ge|~DQt=R8DE>N$$OY6xGfIDjREbt_i)fa&-*KGa#?vt1%-%&
zLuR8#)RJ6u9BI{-XZ17e5LxdnsWW_+8VQP?yV$jRMz|A2WhPL$pB!%1Q0ZksMOi{l
zS6;a=&T6(llJBd($-$XWnL5NuJn({_-0mQ7AKi*(h)y%KC6Ekl@t{dHkS=~wQgp3z
z(*&_za-pY~*xfo?K`N^o`qy)ReKEtA`499EtuG|T;VN7W9Ktc<x3~CsPHuI%Z**PY
zX7_OX>oJ}`=U>9LWuCaVQP5jNZ=s-T9|;zK^mR)Z{uN#ulX~)<cYB!duZ**ShYx4A
zzOX2LS%yDnwol`dgn}#%9k;h>0X5GxpOx)7J^Y?F>TWInk@QSRZr5F*qRCKtLDaHA
zR+_+~)`~Ng7+6_?z7=ME&RDuU=|aSuY@Jk}xNQC;N5`qvIaw+uN*6CHRw9ZWx4w>_
zR3e)0A@?vM2iUs~PBp*gN`QQ06AwU<5{G4bn@;AS-N3W$InFUb7TCKuP-~NsO5XKL
zmcEM}OXL2u2Otfvn|tT4{dRUIjT?Mz0^Hnn>YkPTo<UBiD^Y=$;*AfZW`3h>CKdH_
zC1lp<9jx--npEw)pC0LJEmWUYNAX}QCv<c6cysE4&wpH%#22&V1sh&MW+R>Fm@x_m
z9k_b!A|*xjy_NsCg+SgE{mfg>HtCk{8L#O@=jV;r_YU5H8Bldz7+qXi43B91kT?`L
zDNzIb9d*={0FSP#=R?V)m^0S`g*^f=)mHlDReh%0i+z{+BN4ryx*gOGUHDTM{2x}d
z`5Wk~%=T5(c-sEa>MUPlQC+gq0&fG<Xu=M+L%bwNw<pEgKdyLoXMXKE6I_t|K5S!A
zyVC`TlJwk*9_>=G$-Q3D4||NdD9sGY-)J{czTKWasRg13m$<=obT|aC`}9D`5*?5U
zUo4#qcpkaftGM(O0M9zB3uL>%S)!cFKsKl|MDH@MbE)A~Q@p}b^sUop%Q_8D%<Om;
zJ60A1hz<o>34bMtpG$rz(`dgn6`Kr+3qZWtz%4P7{|DOhaJeK(b}uF_&IL{hY5J<f
zDZeI`1o9P8qK_FWo4eNR>Kup*o&ti79A;|@T4%kUjHWV%$M;7RzwUe5H`fY`U-BXg
z!Su9orz?S=sk^4&^64*QGAu39>Z}a3)|yY28!c-7e%jW87ziyNj$J2qNQ{e826G0V
zs0{=WGhlg1FC40a=WUX7SHi2^9;BJgZcRr>O{3=)D&zKi#6Wctr26YZd2^Uj51_$p
znLPOnK)Nuz2o`WRBTr%Q(!fLzcs)Dw$_oa6<L}1sw%S8XazAlN%qL|3-<CS?WTFlX
z(SQOx8w}jiJfx_zB~gfVux}b9-3KfNHKhEazJ&Mv52Rkf9?hzN%4beaTG)R=53gA`
z#JPxh*yummMY%bTMl%IzeAYm83!vuUz<vVun*6gVRc>rsarTcX+&_#Dki4Pb>Yzyj
z-|857{NeQ-iQz-b-D@vW!_@5|<$?CQ9;R<5?eWd<6xDPuRckZ8;80(L3>{O@*jZ64
z-YUazD`}{?ZS@f^GyU4SeW<68LFI`Ek8E0<*Kxw<{hW^^2nxJ_hZjSg=cBhM)8S-i
zr9#h;Wv1_Mv{EhFDZ!N=s~<;n-$RZP1FEL=*IO;z^w*cGpln0L*sU}Lm`^LneW7w2
z{ee58NOsX&#xWseC^M<`^!EA0-_BV5>P;Ct98(ReA7*bxYxxFd*BTo_bg}HWa3McC
zS)H0^us#h1p%F^pK|{~{&`>NN1M_H`m`c8bV<`WsGa7$eirJVEeU5x&!cZs;Er2g`
znk3fsgG|{F!D?IOH+jznYvR92H?jMELuis{ftbuK2@NYGzj$!#zzUx%#J)`^%76KO
zf|qLut9o#Tb|3j)6?L+5<sE$auIPiNlDW^Fr+vGT;TQVaG(XPj!l-cF#vaIzImwl6
z+ID1Te$k3hEmL&ue)LwpZ!*s407HLGq&>g!Bez{P-+RB#K(ED*;f#-(7m64k)E{7~
zxc}>W#P}>_%3!zPOJT1q6CiVeypl2d?tmf{#TUx}sJz@zz|m3#OMB<SUM|E0UPCuF
zrwHk>s`+y%17Dk1cU99LDv#}=uk$T^R(OZj=TZYHouDY^*jPl+)-T6|&5$J@)#v`j
zw8QtmBJUYk?g|-u3UpjLl?3sz4+9mjeQBV~YR>{}9|0Z-y-i+L7INayGqQZr5v4hy
zkV>5B%lbGey5dy33$XT=TZ%*MNOvdqaZV5a4@=h`&-DKP=NfWJjgZ^WY^00{q2@jd
z+2*<=xn)yw>B73row?05=F+guT?vKKk=$!xM!HFe=qi^=r*nR9=llBud-SNsW6kIF
ze!pJN*YlDI7fV0?SY(NWpVPJiHW*&nR7TqkR4&q7m4GN$egr{@eDw5^>64bc(uOZb
z9BofgR)+c-M1EO$ni%Px*|GDz(rjgv!|enE9qDT;^^(F_VY{_2sL!XXWKC$*sMYO%
z;-rN^PO(wlZeoT|sfhpDTJXF0FMg+Y%hEA8OkUz&$mJy`=8q5k0>0()v(;1fl%TC*
z6WN6hDH0Fm8A^{aZFYk-3Ha&1cdI83p;^xTHox8(dMN5&$R*~=nW07X!aXss2#w{t
zIVjV}{*m(h%n*r%^xQn8i$<X#r8MO`t2H$v!HWEGFRT@J>nmX!k6;`cah)1-WgMld
zbdkdogI&7JPyWF=_18Gl)J-^C?ZcaF4#f02`u>L<<8Z`VjIWKL<R{A}oXr#Re~Tj(
z@r7VY^^@}tf;Qdrlbsk;ce#Y;(EdR+M@n42o0it0aiq~`GSfwCw3uU?EP4?w{VJZk
zH_=S}HNp69guU_Z4*45al!IsfAdvFSYc+!laTRQ0gbSP?1EipABlm18d!H;szi9w>
z&hBTLqY`&#2ixt<W2Q45;!DS4ft;gJM28f^-mSWg8o@=)f&=YPX;-L-DgM`&YY7Tc
z)zv0<NJ?G2Y?&sL&}SAZ%iLvh@v*4<HV0P~CMeEh@Y-QVYJlFL>+<-RYI2U`va#J=
zRdsHOz%Hcupo$|pktmc%jPy)}$|9dU;?hP_&)s2M5;4I_MM_HS-t$j^2KA1EI<lQ*
zI|&SwYRu~8SLmU3Pkz)cP1bBj$&m`M;zc`QN<62mZC!>C6u+9b%9EB4IQerUZ^@^m
zcKlH7r~Tw;l8QK)3?R$x_AvPeIQMyvJ?TN(G<{yL#nTV*qrCF5*x6c3CCf-}-KYlB
zCw+S^6A&(4H3J-owycHwr0tJnp!&n1^61Z7(kby_@G#L(R_!>SJxYAk2Tg?5kC4O<
zVy&0l!!-vq-M&cFILMf}=NlM63=AZM3&<>Wq`jfqQ4HXcLj-4SmA5m`*cve3%Ip!?
zd0b2Y-lpTj;JgP24Zx5ID*k9bs)s40dMs_x$gk-40n+1Vj?`7|p9Synbxx+~J>^>E
z``lMi)e22slo`)+krf>S=;8&&7l<IPotua>rzlB-Esr+Otv#KN33;L};}tXjuW_{I
z^ov>@%F_Eqo72ZiY7@3ukx6q;EX$F?2FFi2V8xPrm}aMaOQUU-pDp{eozcCfmqBe<
zQQcj4T<nu?x_7R1YdHn$ROr&X-cT2M&gmqPAlh$kLN!r%`YbW+29qHrr7mU$U{mNj
ztXMTsC~}M62Jz*qK#(#bdz`SsKvF~rKI?MQ+Q@QIcQWJon7xlqY}BQ*$>eC)-n6ZD
zl;k|F)KP_o6jx~#Np67}l{D$ujNjvskuJbuc!uCXBFm$waZZceXul)+=6@O-V;^@N
zDDV1Jm49tQa<O)eYuSAZ+Tl>At+jD2rr~)MS*1G4ay;k08{X)lo_#J;%Py|@n2hsc
zXpFxY;<{g`QbiZS03-fMO2kd*KHX)C6%fvTgGt<CfPUC<Zjk+kDRn#C3VFX{mUcTF
ztq9-+@kRrydp4k1&?z8z@C<{UM6!O>451}5n5WoJY(v?E9dyVD6nb9e%;hs)Rzv7t
z_+vQvyw|llVUw&#I?GPC5zVxRac+}M3Bj;eHLvv?%$!C@mXFI@8&Fpse?0o&2P@RH
zbUaRH76jcbcGT_$Y!x}mrOCK#qAO2R9ofCeL1jZ%Kq8SjdIrhTJEr*SD3xiSq8e~7
zt;FGOg5n!j<5%LHd$PDG9Z4?3yVTnpta%cB1;Jn4b+>8ri}$66xVn(WZwAj^E>%%~
zoV74t#Z1ib)WsCB`-GKxN`Q45AUo?Z((gtEp|6(EQ#0=rtfA-Ht>XH(U#2%XKBWw4
zU(26Z!lXf}-Do+UTztG<BA`QBA>$~<{2M=);<N;BdyusEwE$`FPYW_{yRqaw>qc4n
z^o)62cIV~sgvsQpgphm_O^|89hJlhivfX3vS3eK0*n{W$e4ru9@S3M$L3BGf)823w
zG?5_0_IKZSEC+yLU<6XoCMTtj3=fk@Cle1S>`kX1aub@20|`%Z3aaV%(UQt}1IZ>k
zYJ&j}R8ujZ2=WEC5k`gS5nRB~mY|rTlj&b5*kr}~w^0E63^>7ZvZpSVf*SxEy5;rA
zMt9?aQY2iY<rh0!ABI;LpU#}B6W#PG(Y@$B{tQz8@<K<h+c9eb`uyFMR6!)49Ux~S
zUAb$hI&zYc$kMQN-X`MuEL4bny3w;(=p#mXye}i;LS44J*&Moi@(x31d-+7=jpm#L
ztgXtIBY2;C&uzEKjZ7wrUwZXs&-=hW@ATfC{f5>QZqshP>5n^oGZg+%QVWXae^mCI
zxfeR{?8J+v!ck9!ET(v?HoZ+Py3B4_4}%cuDc@=^y9q*q)^5;ef)frj(`2MHz_}!y
z+-c;R7q{8ZUDZvQc1|x28f)cgJ*2IIi!EKkcMKo~4xviLET%t<B0_|`aab1JQ}YHR
z{?4hLcw2naX3bmFNRD)lX+@qoy?v7YeMHBI9CbBR$cTKa8(lP3d}+i4%j_#JJ-9QU
zfQUhVTN`EScvdHK14d`2*Y0PEPDM~ptcfK~%k1BTxXfn$vz82_hX1_+h-ACjZ~8#o
z;|2Jf^FE;uZ?XH}%iG-)Az_Y3kLXl^SxCp^#0VuINY;8zK}0AI(`568oSah$PLbQX
zZ9>hMXIX-?Eg6@I!8+=#+Txy!Qg<1C(bz8c(=U@rDh}lPkpO5z^ucJ*t|jRRp1C_y
zJWatu5NAQpO%Vg|k9zChk$D(dQVeUG)c?K$1a!SwIQ%I;v2iw$XP|t2eEdTgkRum|
zN4vkSG6eE?gu}k-gHe?FTbkqY+iyA(wXfb0`9Mxx8cSqoXU$ff@;2s3Yok>}QJv$I
zF)a-WE&$$Umajo&eztiNbe_Ui6CXzX1O>$NJgW4>WjkHe(!lZ-4pLx$7zhzl&k*OX
zVY#$>d#+TLSHp{T>WT{Ls6=7=T6cb+-JR<L4S#)ME?jw+$1}4~gWgq#D`@dZzLRsk
zk0;CbLPwj-a8?v+&Gm|WOA|{~W}1b5CQQeKK%9J#7JlA<PBbCr&(CsXcFZ!?xhRwq
z9ivoxO2b52<~VFMq^+XC4?x!U)Nn|s89oc-9Y=y8r7rTAHkwv-v23{rYLwJxpa})f
z)ltlLAbt^47Hzi%{ZtxHI<)EUyLy945HAjmbT=wsI&Hf>)p2Y81l#g?-<ew^LFxKb
zexm*!3n<(t7fYV?aRRfdP2~LA7Vf0jZ9h`el<so(WnAh<!qix$s~9?{)csD1(F}DQ
zAQ*B=+%F47DdGb->b)+AS*_ASWkINLLydCmM-S&1iju6Z6eVNzP(r-Dgw#39jB6{x
zRr`A9bFa18WmvQi@!;kxjF+x*E%YKQ=-b@^;)k$?=3iCYUf#Ynx>O;UOp8%CtJv|Z
z(gV5z6ranGh-!imIaDZ=s|nygXuj}H245+E9Mq8v0i>SQM<Pz%=(;phMNJ*9Z&j8n
zb6TYT*cba8>TE^s@g2V{`}XL9lqX~rY5C}*J<W31Ru@W3;`V9YWVR+AxanAKMd@(R
zEw>`ebgx(Cd+m7ro4+T{J}Htr5B{ek?bZJ6k1tx0lfO=vr7f7(=?_?r0-_f=o6i_4
zxfoNY+vxwpO7FT9s4=t}PLgH&NR#+Dq_{*?W`8LB5g_ofF3S81$%KC*w0oZrlytCm
z$_R>C^`M<adHeFiS^Jcs#Zal9FQA(|Wz8d#oh1vd)@)Z~Z`9pU2{IC2<Wk_Qd^<r8
zQl$bs8ul`Mni__zDt6-ROVnrQ=;=iNVZN$Bni40!3%`(OAUR35XMN>oS@4zG*Q(U>
z>9)_6`>;hUdA-{rO4a&uoTCUMIUF9~tc{k@7&$*+1W*Ykx}I``(!UT&4mD_EAQBag
z)34YW8v1tSpNHL(jUz<7!82+}KIg0ZdsVj#ADs12-5sJ%!=iFc+E-`Ubf>Eu7<C*I
zh=|f(e>g7Pw^vi#&lrEH&L%(w@{5wDCU4Ph5;Y&J75$dZ`e5W2IYOajQF64Pvw377
zD1H}PtuV_Jwm+HApM&;_#d54yiJ0Swk;FjpA~`=UM-{;prALw1Xy>2WmsbDD^Sx9&
zw2*k7ICb*Sjc3|j#@4elg7j$jqMeO>&NoCR>t=9yAx+(nZkQCrlEhnk9O{TBvB#p%
z6Y?iK(RAfW1~529*+Y=kJ%=d}LEFq^tCd&)ud4a19-g0LORJX0B%QqPlocxUi2RBi
zM_RgO>(lGc4~8Kr4h9AXh2!is)m&L(9DvgSF(uxd6{Hr7_Y9ltJ`mZGqR^>G|6*?Q
zpAFzvuSgW)-m$bdE#M9cJv!Cu?*7wzw$04^@DBy~p^tu%!GGYdpTIB6O?IBQR*iv{
zsO%{nAN3bBXZKTfYno^Voo14ixUgY8djLp=|BA3j_?A=M#C!rDnfN>hK^WIcJueLP
zCK53oGk2-{z%C<4v1im#ZQw_*TQFklDxU+SzyvFoiz+rZ?qtlBdMa!<x@ts+?Cn-K
zL|<^kYgW@)sz?ECtFY{$kjlcMghnW!4hc!Rpp#+X-RF-_)`}?g-@dSY;x}>swa?Vz
zYk3_*={F3F+Pj>uzH>1hJVNhEFF)SEn14u|u0C4&qn=csr4T-J1*{Pdk6tAkho6!e
z=z?{KbR7-7fjTU(qIl20^}%`CrY}fRJ(R;mPv?0!T<LR(T6Q?_+%nJl;W5$&qm2Yt
zRzjEIvfh#O^~w)7j1CHJ{C?YrYBTvnQ623Ve8v^Cdzqt~Avknb?UmN$Uvk!)0*4Jk
zB>jd*dH_y;#Bm4c&Y_unhPKhC3w0(5uh*9ThS9~et)q{rnX{bnI0P(9JdP=<`h%xB
zDtwyHXy(DD05y^yHwCy3MW6OBi~KiS3MF4f8e1RSZAl=ud%H!F4jnN*(#aQ3zjJkt
z+3+W}&L~JOZY!Hbv?9})9h#`c;3FkpCTbY+U}d6qza-ac=1m;-p4RTN6ZEH&id^8N
zJxpKlSKc-nXeFRK0M@d;jGS4)jD8_^gA-MV9<ziE>;pW8q{B4T%6vCbTBIu}P-Aqw
z>@KS=c~;9<j^t59eQpC^;L?7tYg95{w41{#kC(2SBC)Y`QKJFFL|IB+xeGzwCr*c-
zH1DPp_90hDyZ>?qeT)(9b2K{Ulf6Cm(ZkO+F?BYl1cXMsW`Yc}3-R{!fX6~jXp;Cm
zZ_hSrtF7=~NbcE1W`IU(B+iTOZ+3MYYtDr&KWv?OFfRO5JAshLkZUs(T_*qfV1-Q0
zaluvV5N`M0KjXvu=X0ct?$ZdWTNeGe_@)T6UPVg1?7`S6DIF^w7Oq{vHNt*zjc_@)
zw0Q7-Hg&Q03OxUZBGEmBvRcfiM=@!9y`J-xG?maLOgtwMOgNmgLkhJ*4L%JQ_d61A
z;iPl`Txk<Whp@KA(UGb2;Ma*NPl_*gC>SB7Ie--lZax6Enz;${>HY@dG;K<<ZJ&sC
zYKYC%I_gFQMUZ`D=bx3(rk2wG$d#b;5(A=0FVyTWlr+{NH%xEjb+BV}cJxmsX1gII
zpk3;a75wcJw=nN4J?}9&TbvBs@bsL58f^OjRTfDTT5^SKsz7pfYcu`wAurtJJ%X|q
z*rFloY|s5{;o%7CxU&~h!+E^pv%BWvt?$*s(WRq+|2NIpRW8IfdSBLp`#T|P;`!3;
zU!KeCK!qfqleX2eZ5@^Meb}}*J{CC(fFWSTKU?}=z#o5OOM?k!PvITo<gJ8ZP??^I
zwp(eHL;Yg06TDYLm-Z$7v$|7X@aRKR<m^9C!`DZC)QNXQkJUe1EJIXcznq}9`KAyD
zWOBTez4IHJk8H38WSsw*+!S;s!8g<Yu#ed?$pY^+vxYJUFN5BrcP}ZKFWmd|n$cm&
z@8Ti)><br;Ae+v47Ub=9Av;a+>wY-rL@bu_e>J&AIsIttC^54$fUUFT?_c1#Hjs8o
zXjsWu9&ffnxmWgM?<Qu&)_OkOByM;a_{OxX3$u^!{Yam{BjLmO9k-CGTZSOoHw(c{
zda2vQ6}_q|^wHaD#G#~hKgG7{L47&-U&vI{MTL@_n|IFzA;vV?(f>j|onGUqKJ}Zl
z{*$T8F+cD5MaJUx;+e?N`L%A2ivI7hr+#<kK81+nIA2x!EoI(hk&~%)jOxf|j|dl0
zh@E#s4^0?_?%)3b46}&;wb;=Z@z9Y)OwtJYG>FmuOd^^83&Cpt$-{7(sn*FRo_=i(
zMw;Q=3UgLI`Zr%iX2(lhPA1wwD=XgkH!!q$QxjD0u&?0_<seIb7f}w3!ik&sfkv%F
zD3<}w!S2fhq?~Z6I1+jouu%zt8X4oXJ{aAs`ZOPGfUZK!h)J62=G6yCfjX+p`m6Bg
z9%@Oxrtsqm4<6z6Ds(-~yeVyd*6v?OXISOU1iZmhE}rpse;qakt!z{}K(AcXS+BFZ
zWA%W)3q-O?KB#XWIWzOz?(bEfk?7Dfw$mBzQ2nbf<9^oJ_*c>KlQj$tYOBGw6{Gb*
zbbqSmZ#HpIxYe#1^@wZON6?G_koas7gyRA2j+!ySio2k8_L`f(-q}Os|I85r(cR*d
z3v7v_qsNcke10~<`C6c>uE&u?x7cdg0&Fvtklcq7Uv?9CZ*weUK*E~REboVoUL?)+
zS>?3pSfgfZo`xlUuy>L(QFJ9kq46~6xC;d0KDu~8T4s-5TeVCOPZNamC9!=ptEEIO
zpzT}$FDSjyKlI_s`aZT-AEFj<!N5?8gh6n)!u(NT09*vR8@Ztm%0D{1Zg9Chn1Wpm
zyU$Cv4-eSF3?W;W7}3R0K5Mth<DZo?Pp)Zidv8tezu$_Z8#*cTe3}!UmHK<hJ9?Z`
z9PGdBUOmYQay9IxJ2P!<&W@lHFS3+`Twmm2Qj#7$s;;B^AK{36mBKjuRIYiYw1n9)
z7$GYby%~4F)Gg9Rr(qYDEH$$^0eKHdw3Fp`znlNE7LMaYeSTP?FD>5}Mq1L&E^IFF
zxMNz4`^_9BFDa=?b*YGTtJRm?FT>N*4vh;4N-M{NX^Nocm+=@Wz<AkDMy7(g=Q4#P
z$XrJ6e>XD8&cQMDwBgd7vqy}tRIN10SDyX#=+^P=dmI*J^+;^P-ucy=!$kp_(dBKE
z5rZCS*_y-$DPLW3VM!gcNtw-~lV?H;V6a%u_YHxuyWZXB3N68rt+G*itS0m)p7MTM
z2bA~omEy<~hMa(9i%RxI-|u=oFS2nT+;fI2ad#6&-QxvRPkOO$VtUwT!uE5ND|e9?
zhV?%@OY6_}Ih>UA>y(&#8`TkAS*Ot<5A@5uLvEmbMNIZ+<d33H149HZE39CoG5SgL
z-3>NXy&xnTVy02xcj(Hs_=@_%$_5&FY?GK=dh-p``$}b|xfRl(OoM}}skq)U2f`-h
zH>F%;AFK2n`9jZZ2vkc_RH17h#BaJAku<l_*`bwLVJG4gji!{FJW|lg>;AHc5s+4x
zcMB)FjSm0_1Cfsy@z^SJF6|`0^?YFZt)G%?-DVC<v~<FWlHqfrQ($0?I18Yk3id+Y
z3fnC?m+>tAz`Eo=A8OcZ^{C*}x|Z$+7_Eco$JnuL6r8P4-vQu(E!RZu$-z>`?}NED
zO9qRSrt(KonDuJe2W}!aZjl(#(==6J35$e$i(y@uEgH%joeX~WCaS+_PmV#V#j2}t
zn{miu(qHrN>#t4tBTmK66$jp$cv?FT73@{6%V>u)<s~<2MwD50_No4kfhSJ1#$U$H
zs{uP6bOL~73yN9nAouXr(W2Btbl3HuxQ{uWXHNd0QO8xFw{95Mu5*^{eVq#Docg4r
zz^6;ZoqW#~9vD$Te03JjNmkIi$t+46Pp)|8fDpzeS;ngr$qKWrnYJ3|ZM?U<+#_~M
zH)~Ip$}&$%OedH1`2@B&VG=o9t?A<3B{At=c^8Z{a5H^Jr~~$D^xX0`hmeoHr>*gs
z#$6)2(FBnDBi46x_cT=K+N)%RdO6$>6LU|gmXTu}1n_Cmtp6|_Avs*tdGp#FY+b0r
zURkgVz&nd=twex_U^2MF<857}&)Z5N*T3>T<1IUzzMmoK+x|~91;8NmxU<xcJE>>=
z96nT;ESg56`K=kT@Qz5iAcEd{cT>7;+snAA3rzuIZQV+JS~CgEoE$7hy}jn~X%kKJ
zV^Dbv%99OYitce@)qX$-P3cHQ_>PL4K;Z}5(R|@;HqmiDJD0|Xklo#N8GC|EPpy2o
zqt0IStQv##J1!pAv+8^LM3SwS-KpojFUJ+dzuoJY3m162sDMi7y+nw_`O1_Hc{8M9
zpmg;rDBuqFnSticY-@%Eu$<5qeQyVWpfrvR6_NqQk>f68A>(Z)iTei!0`t7z25#KC
zE+ZGI9<J>papq5p(}w(?u@kB_3!x`(&;1LTa8<NzuPWG{@6f!h4edR5H%zi##X+Pm
zP5aBZ`e@RiLORm!S!edSHkTkL{A4SwlW802ONhPP`J4kKdg)+OIL=yg9EzeUY@@mX
zTC!)ax1P0F$UOF)jG2vIEG>dwt_3Ykj_~uZdb(~h=eycrzbd94yjvEtirl_J_kHY!
ze7%Z);q=~BX>;;ebgh2RKx`#NZ!E6VbNNlJf~f@VLre!LqT?iK`@-(e+LPI>w;B~?
zgEYt|N#qXFKGj~A^AFp0ojA}aT{C!V?`+n(!@?h%QYM*+^j<J2S=sO5dY$86?jjlv
zm!r6nGfwu<A%;{YNVf!hDcMI=&~Dt=n6>H2IJR6M!Ow_NTHqkv^f-^6Rx2sZ9VP6{
z&9PD0t;0ur*m3P0>d%<lAmypD2hZ-cijvmB#Wq8Fo}DIG+Zt#wSj0Z9fw(vvQaa)z
zpAAhu3^>&*uF9Y3AlpmN8oO;jxhF0+fQE9G+~z!?956l4tO>WftBnd|RZIMyyk~sy
zic!f0jh{kI@pE(sV+8J60HrmBSq1ls^f>*loh#-!$Fch-Um|0F7;8-D4JGiO&(7&_
zXH(FxqkUA;u|3RPO5#Wz)fG<|Du>AwoMotp`aD!BJyKI+rB`vAt&QCFja93MI=(}y
zamiI41pUOxVFxt@ajxJ-%U?MrvW@BQZSAVhr|Gz&TrI(NHrmzU5liChRN0~E4+VYd
z#vUTvn?1%|i27zA3`zWY6qA`PH&Fay@R`_|;jpK@f1g>d{+7=*!4a8@2km5>{|1^6
zI%Y2@+gh!UnBDKJWXRny4Q#UTr_=OevLX&SGrr^9QA`5bO<3Oj%Ybri4j#!rUyF#S
zWI{N=a0(%050)wDQ4y8;ArFO`q22dxS9^{2kr3a5>jTD|fwTMn7ARo5ZukIyUTgfR
z|4c_;#IqM>s#0@JU$s7;nq2u>j8V-yaiLDBv1d}6eI%N=<91+qfwo$U{$v%30a9S9
zecdapXGS&U{Woge1L^)w@8=2Vvs1+Pfl#}JQq4e;Lw>QT+J=KwTCL*9L|7%%puNe`
zCu^;8!jbaQ5bL>n5dSh;f6C(pq}<cOLt3FT=UAYx+50b5R}Q~#T_PfG!OGZl<cyn0
zcXb>_c56*LgS6KKGCE-(yCWfK`2dM_(cDT=P&sS3gBhqYS^l9g=*_@_vy1Kg-0fRz
zZX-0d&4dwqSnv?LRv+YTnX`n`R4>>Um?*-(@oA3VQHP47NhNy&4;o|tdxq<X=9k4m
zMZ)(Hj>T06`30E@xB^i89{P%Jh6)em5IDDWyqE2>TF%(f^Jgks&TKo&d5M(vl`><=
znSOuE&qnfM7L_XDUyr^wUuU~LJbDQlYw#5tuqTS5*m7oGT43zmxV_N8s^e=m?nMF~
zusQ`(B=6SFI4hpz$-9Tw@g*rCh`$cE{&j8*?Q4EAQnum?#rXR31+?##he2m+s3u-l
zt+U_o44%-NJ&$)Bv#}=@I<ZGcQup;t*jcgc-;(OBP})@U=c5}Ue_l%M`hBFFd*jnK
zevS2pdPJmk^9CFC*QpK8?>$sv`?jfUj_%Z}ak~G3sS<u8d4s3WeI>#ODMD4`uGK|p
z;Hnh8BI&8$$P%YMx>&jqcA<;=v=S{xtHt9=R&KtAU%&5ml;@MKe}t65_fH)mo!WF~
z!YuICPErzmkf7gSS2IIQ(n{IjkZe_v-J?d<N3miY;u6yqEJo)?MUrEyYgj>593?L8
zVnlll4z$cz5VO&}<QYQwdF6WVpvHzwxNpWds(j)6&6o)v7kr-nAcZ|(z%$gOq-Frq
zylRm+uU7Jd`?-?ul99kZDEW1QcvW8zITFJYiv(slUgAZs%PEQ7=lc%BU%udq-T#ZJ
zG|j@7dSD29{kfc^9Rs{G8O56NbYIrjmu21LAdw8>r6w}1XRdAktP8|@8^`LI1nX2@
zpQQ>UV!Z+jj}!axcl?(lubZ?jCorH^Gtq6qNg(Vh&y1DF9^G}e3QKLS&t-lOv*4AT
z{|8Dans72V$*F^#az-b^$<}{=5TdTL+=R!+lUdtG*AzJ6F%89R<%`DjGFP&qbDJg?
zZidIi1Epx=#ls}$&<eWVb}1P&1ldDV%$I_Y07#4;7d}|Cr%+_N4bT$g0!XAEIrGWn
z|1n=^QZkUe=rxw?uHGLGXRWJ7o@hWWPs%@wisi+wtIa<#Svz{n=<MfGsU><{7IwDf
za(*((^=HishV;Cf$c(u6W5g_1@#*7#A&p)t;Urq}VEMY%3&v&U#Hm-WUG3^auiqy$
z7LUB?qugGx=`ebkGg)qsLY=UN`9|Ecep#w*J?c@hJ>xvP4+7Nhqrzva&O+c~A->Wl
z|9U7BId?tzT*}Y1agMWozeNh2;U-eZRs-kL;UZZm0c0mI#1QCiP|bjliL=Fu!gFam
zm}IW7d(=VbY-hw+mtAT}tB*6+m3&ZY6FOpYKdsdbN^DOMd07hnxrIOHrIkK2P$s_U
zgA{X~!8g2cUi@&`au@STWtSy<aPh{<_GL(|yVRO;C0wbxN&leRH+!s-6ptk{H+kc!
zNS8jOJe0lKrzpgY{OGdUXCaifp`cdtd&fVYudENfxA~KL{k+?mlpmq@HMDm49N<`d
z3Y)@5%-dYexYVZ+c=8Ls{gWB^V>6#@4!DvJCbVe;|EQe*E^#(fBLv+ytP!l+C_c%E
z$~<^2|7~+LKRr7iA7DX0*?fz$8)<jA-wG+iTC1Gf^@(<|>QlaNz}4(2{KgT=ei(NI
zyb0Yl3wqBV1{J^I2CS=Y-wamIzk1_OIl_4#qD}1}UFKn*E_8gC!_@&&HLTt+_@aq2
zT6l}_hdoD?(Msg7sxBT`m$?2yLHfajlF|Ly3W}1JF<5=qplw}^dB}x$MxbCkNvvYF
zjBXFR9O(*8wmi=hDQ|_W@blHlVDNFuV5r1wg5!6D#E8TOc6@dPECXR;tI$U`#d<6+
z?Y<h)x=wTQ+Y0Yr$ls!q?+4INPj{3Zn6Ar<m~gQ7k?{lutipz-vSc;P5=KjLj(I%x
zqHBYkEmQo0e?xQqEeGYA!3Lj71@ZvXR>e_4nZ$;h#02uRZWhs52TortkzF=+FL*ZH
zI)ju7@i&q|?N@U!=C^eys_$3&!C5W(fgV?&8Q)XTtJ>wBI7HAUB=#BF)$^<dh7n5h
z^Mn@vQB9fB(Ub+Q@U$wq<zL8sUwpMpR^4@pm!vy=N?)-q5Ogj=bq<|);UzVebR|#_
zlI9Y#dMVFBs}(sM?QycvDKxFNY)RU64RpY)ODwT!T{vxjZ)AqY&pp7axBbRswd({t
z@gqSh)cJxLachy|+4-L0J%(AYWZg#R10)|ws|55917Vto3%WbAPGhdA0eudQTv8lD
z+zR?`4>aLh=rkar+TsT<_79A@NvxLH{koUiq1uwG$I1wz(pSshmniOTercO^RK=%_
z6$@D<o_fvRWT!NLB$8K;#hz~##(W9$cKFg0Wl?W|oxN)>(&jC%`T4};Q9IY!U!jMD
za1b>J%VukN$=u?|r2#r?0By$#z9oPul#Hu59~~p{rR_=PKk{vI8D)pu>-}KU0Q91%
z>RE$T&pXU^CJ6%qPTXWf6zLUWP)PCC=$pSK)K#G<Px*_SC^Dhe4BzUpZYQc>ES^8e
zwql^WFL7gMF4Z-g?)oe;$<8|Sg9GFG_#m%5@*P)-jW6QMR5RytRSef1rQ20k-Co{g
zcyW@c+PnX@`Jf}4AMdR0LY1uCP>27hmTaEp++MJK@RfS($aF<>G)x*SMQ5mj*S?Qk
z_*vX4)3lG6XnB+GS?hb25f-y~OuG)1Vv=*?m6g#wN$_`A(bo?JB}u~4jeBTIbcm@O
z#T2k`rxDYwlFg>sq!T?wz_Ts0ht?kM_md;tO55HD^AA-zM{?<*NjHkp{o(If3Ty}t
z+-GD+I5ZsE?b5@zGXbe6EI|_YTt$dKwaDVu0$E$d@bFijjX-!0&$Ll%^d<>xliiFW
zu~ktTKR28=-J?9J#J#RjJuX#bN?Q`6^pr!LXY<`i3e<5Li)|(qvnc>(X){Q`LNSX3
zAe8!ARU<5GlW?m`Q^)!%0Rcc(SHQrvx?1~js6PI_u*I#1R(f}x8iN|178Ng)a@l!d
zj{{9Ju0khY|Ht!NINwEKKWM|}5RQ==<_w%6(vi+oNIWh>*2VD9S@=n#KI`QN9?1$B
zId$^(fOeB~xIh^?nwl-+G&-4BXz-fhHCb1#oYH0XMfgszI8t%nL@o}tL}V*qOL7~m
zuOho?Q*}0h`TX)o4D2I8<G!>10|fh8D-5-}5nGkxyo2XlFb+RFF#zZ-XD$$bIuPi0
z63p0v00D-x@9y~`Pn{Pyy+AcS8{eblvF4`Nn*1Q4SwY9d=QqgrHRcx+XgDx<R=WvD
z69zU%#cG!4Ehj;-&4!DSvYB}{=WFfJ@I!?~NuY&oj4(nGhSQvqkz95KQbto1*;6Ei
z6hJCLLAjv&9TqA@m|uDh#L9cTEKx^>O<1zPFbkNapag(o^TZ{XUvEHMmC6Z4_#i@}
zju0weo}iFX!#&I~e_&Of^?1E*<lT$w$;bG2&g<M&I*=0lWxcT~bmZLr+32rX1ss*L
zxKD*yWe&U5^V>$j4F;4UppySb7|(AKA=N*PWcqd!W1FWd$d*Iyn~}a#wSHxph%3Q^
zo9>klE>aOxK|BLU0$5-k1+yrC%%Rdi9|}}uAp4uUMSumHVR4mbU8Vb2z)A1q1H-1z
z!Y#gXq+13(ag$x_5%a+j<{$TRipJ-?NBRiM-b22bZJKhVwzD<${vjMe``17li@pI;
zw4WSlXgTbe-ShUZc&l#%@Y|8P#l@E{?RmDX(YN`%{RIMJZzFHF66s0j&Eecm7~yGM
zx{cgDp=rjl_V1q7jnv14g95=Nu9`EjL;Jj=MEg)i3di9&zkIFZMTwX4Ms3kI7jd5v
z-^sIg36SrGk6z1Y#s@xgu`ijrS3CBry1D8Z<@=elkyj?yt9XZ7WoMOL)XD2CuN_ay
z*J6`t?Xmf>DIF{?5nYaMQK<Oir@%_$2g_?04;`EDy&sXkIT{T7Dq^`ecxdlSJDxa}
ze8oMA1l%8|h2KCTqnzESTY8l|wO_Zyk$H@IaeMj#KX!V}+e^+=0Aw-z+h1TersLq$
z@!hDK7alXKU_6ph@Wq-DoPRPAPyq8!Gm}v=R@Q@?qxgGDeVteJkTN7g{f>_wuiemw
zt1dOctl79BR;$^CYAO4Qp3=Wu$<i?#VaCAuDn3nvXC@_XMp$QyJm=t&X|{a^uxwl|
zYRIHbVwx=Z$@ao%j{6VKwgT?c66u9fnIP?RkB_B4SFZ_+nypnDi9Q_O=JI<&serBO
zzKvt3bi6#j-<&6+d4nsaB}WQ0L7T5cjj3ny?-qGR>uG~%7=2?qaRxoK&}MFdEB`Fa
zPAlx1j{4e-v^!@uS6#w|E8mM*VH7H1bV|>kM=lb`sy_k0CnR&@m=S&m;73ZPz5g8_
zZzLlu@5OPhf*wTEnUCc;&esYCjdu;!B@fsgmM{Imy7%7GJE=B9Ks!bGtIBa<Sw7pA
z`K+q-xSmDX8Q{J&`=IJ}{a(!aUM)D+4PuY@!3u~2_GtT!kKyAH!h&lB;X5$Akm0n8
z#$H-x(BjZC8Ru5bz#1;BdkiCHuQv~5p!w+uZp_?X_D8H3blKmMSi9A`!nnI!q%$)O
zy{Z!YRtjM+i{9B&b&Cv?Sz>_R*182BE3sLpuox?Qh;Cu^ZT9!P->qME%OQ_+BHv+@
zM4-N6fO7#R+A6372?6PY17~U{8s6uwh)S*Ru4j4<!LHt@eu{+`AaR~L{w^5*Gw0Us
zHwuQgkmb4Rtt(*J2k;=Eb=)E4JBkYB3W1A~hd@h!3`#&S+;YVV70T>}cer@|G+52}
z_Q30!<V!E}6%8Tlbxo_~g8oWg&-3cf%8Nz`PoKrYt6q1nFw2Jgy*Eq$kXN1#M#-QQ
ztNw~G)0w^br*huy3D!1N`vUv-QT2I&QG--p(P@*8>f1dN<6<4xE(sa?F4Xni<hOn>
z|8)AyMf8e@iniiYrN2%Gj**{`VU}^E8lyALcZ1f&%kTP4RBP|n);$jE_v9SM9Fn>=
z$TtkCKeO$dApSMgGbbL6*IY$SXCHhRGyHB*#d3MyVgH@{U-=CdX%9B4o}#@$!F3%d
zc6}7FDg$4&kDtB4o_0}GczktF%XC@%yH45w2bSKxIo1w}EF^PavLG2j37@l%pYa~c
zbM}%R1CE!9eSnX22_i^b?1K3kyApX8LZlB@`fHqAb`F2Ww;9?Z(CN0JHGv|oIH`kG
zN7_$v>=s%Lm%Xl06w!1opc{WiQ4X~pAw_^&@LFpXxH?yGfGHir0i7rwaC|L^w6z7s
zV~*Lid`QP!79|KR9aEi6#pso{p#!&{XjhtU{4H-9WF2M1tzkCc0OkoN4@dfThsOk#
zY#jNd^0vLs!T9yFl1BG%N(SzGc41u(@DIY@n`Q;1{v6_(4dBH0J}7tTF*AECd$qpl
zUB2PGJhS+dyLS)Lyy3~t8_G~fzaPa&Z%lj6O#3BUV4kEg&3qjefIgQ`$V(t;AKV{f
zqRIHAfVHXp)j1ybD{pE-GEXYP(#^;k>xhQDKVo1eE#iz3b6UbyCO>$DzA<u9VW4pC
zWovT(x1SD$+`_tG=>zzd1h+@h)6tXn9vGH16_0$Q_qY5|W*^qBVoZ=_Vh4PDmc>o>
zY8?dkfxU?8HYhN&5wjG+C~DP*x}v5$qxAOHbkgrD&*lN?qhOc}mySR$R{$oTFe8KR
zEY-5tBUp=5m@Mx^K-n7@s)J*o)lI~Ss}dUI;^bE@8UE}F@yAXm-Y>4(0I**8d?mud
z7BmEyq<Fe0NM%R|cV+9A-6Dczy>)G(!O*o=Ax|hn??Elo9vYHhRd+i*h&E~B)j4o%
z_Q2y|da2ZRpC#BDsH!BeLI*e7j*!AQI$+-jaWghYg<Qp%kjkROEuSqmc$1@|gu8;k
z&%O%F#iw(ASrp~?B`btR&0gD?7T3bB3=8?B-=RIS<9MDx<z$QLDKCFH@ZII$Giris
z&+W<=|3ai9?w!Lqv(0QA@vesDa|+TG5JVV(Pk;t;!kkW|5axz`D2d;GW#*l*q;IGu
z8&tR9?)KcxM5n&*FnX1Zi=k0HYVs?wn(F!M$z^mMd7+o3unm@^sGs1S#9HOM=xyla
zhi4YH=v)q=w;wM#kZEo*!hr8CuDo7_mH>P}A?we86d01hhwzYAHENG9z(a2X>!2-u
zJY-Mp_7GZo@c_<ubhq@lU=^day%J(lri&ge7I8m2Sp-YY%?WUeW+tQ6Y8^rmK>9Ty
zcg6h}zZG@4V7k(Z3{4$cD5fV9pm=L+Otmz@Kvo=R(q|Sq7ND&TydQ0^samyktY1t-
zIJ!@7RAd0oCV}w2@zo`)m>)~(j+w&p`>~(wV7K_0V3*T|zcNp!pL`Hhak#arEh=s_
z=6FdlQ1;!AiqOkR{9%0B#G|%YfqUctsp0ApOX+H@LhKOBf3eXkDpXk{SLQ;e-9H!N
zg`?d4`Gt>m4t;bn$*CzCtU_zbce=;c+2s@tC|E2UH?pADXYRmUBQfG5Nm}Z&Yqf4c
za=2FPI=+);rpqr%bDt-6vw{*h29WK9Ik5lO<tlAO7Al*Cp609840?+G(a+LV3-i7L
z|90&|kDy2J!tTt(uEY9pw>Q5`p2zqc8Dr<Eg_S?(dW;L3KB!~KY=kxKs%e@K`N77^
z^Fa3m0^agx5K1eqYGffxk2CV+dP-vyy?e^+C;O1KP(B97EpstcEG&vIWXN|e)E`xV
z)14ch&_hrJHx>baj9a@*6$qKzhBkt8#d*}9d%wJ%@3Ufj8at7q!~$#1F+gqDV)ubb
z=)Z6O>n67<qya2Z>_w!I5WG*T2$qK`EJ(InN${9Yc(3Em^V?VRCE2=E>Q38i%b0EL
zj36Hf>cFnK76R&ROG2>&+&LVcA$VDo_+P3IvbDSZ%B!=bzrFP5bKmGsK5|&;7@+%~
zZF$Lf`P++C8QF>^`F}?o?D(YDVLyD}#W4=7b%c^4KXrI;+qbcelN4TlrS6N%fzZj;
z*_Q3l{aSiafS8XdZyU3}+`Eo0h^ZPLA$`p_R5*~@XXfa2N=|;f$8EQIvmpLZVb5Ie
z`4xqu0qWZ7`R_E^j`~MkC+ROf1vFUTTO=Zd6IJFJ7)q~V$&es0SNi_rzI8*ZA^y{2
zWe+fcpW==X#wteA5ggp-Spbg%iAOS!)cq+YH#HZEjKn`WTaK$Jh4yLn&UG1JLGn3Y
zxj8hNV$OYNUhi-qs%720#x)cveafX43A$rxzJaXorD|DVNwTG{c+N4424~A5hLjw1
zCJLbh2s)q|pwic=5tae8y@|qaaV7USmt!PI1;2{~R0eOD$pb6;j+Oei9f-4tjV4&x
z?j*0p;;!c7UTVu*kN(1vht87oV7;#SMvdewQ%K=5Q)xCX$JD$=qY*NQed75xh4V_g
z_ie9P<#EwVrrt!0$$b;ssOO?>loCU#`UpCKw|Jh7hVHf8?NN6!vxX@r9t*c-^>+2Q
zMCRC7k#zD=gS`E&8lw{u<%io0N#YjJ8KYJR%|C>#agnA1%_HdO<X{G-dvCs2)eg)z
zSy(mwszg=kyu8QuoqkI|)#>_R`PKsED?b{7hUJehC^fA<ZZ2Bmj1Y&KDE~qv8xN#*
z!Ro&Ho+juw*Z48|uaafJQwXA_z9EFVi8aEZb-h<9N@G#Ro|S0Ok-6Zwj3J@wS}tss
zE(#PE`}Itn>B8^H+I63b*rY%d_=p35EiecJ1@6Nttws;N-j1HG@X>$S<M`vWB5c}I
z$BT#EW5ORB-VY`g1iF=mB>(?R)%xEAB>unq0<hP%%-<oEjCeo}CE)$r`N*W`qc<?F
zeTen|xj_2uwwVzJYR*(dwN!ix2&;g+1B`tsgitg%Q;l&lU>NFE-XeAy13Cx~I0iCz
zAt|aoj<pT9beTUQaDtq^v}@lUdZ>F`RDB1mWi9J2LB3mmJUTCxhDyoH5+v_C-{gZ0
zboQ%cJT6%BdF|k`c07uFcYC|a8q%OR??zXkQ{Zgr&ihYPca|zLc8b<29=ZX`l~~3U
zlmw{OAYW0!MdJB!PI3JgIJYFa&DdNaI!RnSm$v(5cB7F0N1g_N6{7pL9{FuJR38K?
zh53JjfL^7o`o)OTAwtz6JV<I&SJ)xfjyiF#K<Cex(U;EreSYzI^kg*fcF(cKPK}Rj
zYRuUsG?KuXAEdAyOdaE7qtU-IMNIa>>@6|~hzJf~4##wVblt`WXZ{w`f{w8y3PR$t
zMPyt_;+7XUYN5gc{iKmb%jaXK0-xY?Ez4d&6W*H?ACa|>oGZMyQ|k>5mE#xdNPLI?
zU7ShrjrQ0j`Ngf|%Y-90$SWtt#9MkCW42aPG-?F=LG5Pa<mVRVn9RvQvH3Gr4xxdL
z{u05`78&E%2BwLPESVkcBn_Mev%3KZpux~P>iIaw`Dm0y-;1a{Ez*3c_B?~`j7CUA
zn4QxpL%fSwR-4V<)M#SuuNoC~2l9sp0daIi4t5oY5OQkF(RsdoJ3jImsvBMP`E92(
zJm*Y$AC%4mgPs@qc1i4y_(JZ~D7Xh+G!$cVg*x4=T&NDE<$gDJ{`NAsnrm9szG!lS
z7geSsgExU?XyrkZp>C*U<Sau@H~}<wLxHmm$l3t=qF3~mJyZhPKHtO$tRRRWR8l&}
z#Q-IR?1|g%VC-xcUR<M-m5lZ}-{(k=AFV8mOERg2Ja#c*I|MH8KC*ON=$*Yo%R<*n
z6?<)7{ei26+WG41hn$YL=zPK5IwnI9skPSx+MO+s9!mQ!9IFRJ`uul2v{(KAi+ndt
zHrgA|ym#;eXnNTD;d|aPsm~L9NN3iT?)S;5A?*V~&4_U<posv{AS8&f(|>9Mp%;}!
z@p(XKL4qKbFyiEf81B7Y$P;C~QcXJ_rpq?EXpb~}J-C1Ppyz^h%gfAo8l<&dA-^9c
zoEQ4lw?*h;NP;@W?Nbm1FWl&Hexg5Yy!wtg`!6cx0FLz0p(hT9vK98|!?Hck1U`a(
zo~`VWGsG`Q?Guizn#?w%j`^FT`e3(2AN@}U4}|Z}K*>Xa?c`A=1hKhw=tHFd;1Lj#
z&?hX5jCK~~K_R7lJ{ZFw0dyhhjw1)&`SATcMD*Lx-;gpgX6SrHaRzW#&)m*~sfNU7
zp0NBEQhBrS5Uv#%%Q;O}P{dP5Ecq(Vtx#47yMk=5l|lfl%y*;26$0vKB||C{3R$m(
zj-q;3c~&lv6lo@PzqeGR*G1(>2MAPUGiIO=srocf>dF0gyofokPKo-?a@otzi}255
ze6O>HryEJT)R9$^;rP6Yy4=g-MzGv^h8S8zY0hN(`bTGp5fYgtKKHHar(lzS)UM&{
zu72`rB4$xuDC{H^CN5bk*C38j(l&`U+Pdhe`B$3tI?YteUI(>0v9GQ$*pt^p+Ao(g
zsXEyfklMTRr(YK9q#7!6J%(oEa}XpdLq<d;*kc5j`88gyFkR}s`GM+dJLEl{5}*DH
zvDoEQIqR{qw<f_cv-#QPY!HQ0dGJGOx=Q$0T;af}`<aNL+7A21!$!|@GOq8qh0VN(
zN&nEsJ5hqyqBbou7Y<2~Dk=6d;Gl!D+(c-#P=Z=Y9Cnr#IJe}vlapHP;+<A^)V}5w
zy^Iu}1{Kglg=WSdabaMijsofg8a=ZxS)x(T-y=j-&|)V{x7{)B#740NQD#{OqerDD
zZ@&>5I;wB^bpFpAGiuoWa<ivnp22Zl`rLHmz~Rf2H((Zuilj0f9^S4_@QrK;B?xG(
z05K<Q4I$vM2m5^`z&`o^IsntOz3^=isE0!cA(3&e@4CydRu0JFu*ZG+?x(Blf2T53
zBANnEb@N9>;;I4rmmk8|I{%fcEZu|`fVgc(NXC}H4L}3nVLMF^%1naS9oe}HVcd|(
zh{mB#I)Q+y*tWy)1JFboC3VMbZcE&BS+ZS-F!Qu}Tj^+%LpaTfY~OA5k%`>}_ep6y
z!c(fj))!L~lHbtl15kb$8l4$?|2}%dNvmYRf4ymN$oObFK$;dNLU*gT5PPTBWqaEH
z3RUC5N~r>|UeoO#vEg2m4Bx;LHOhlEq(9dmAS?IkE4P+U%<NSA@E;%uz;iPoKm)c}
zoUIuOP7YNfxWS_7u4|B1U}6qHhSPT=F;WON0Q{s5xgj%xyN9zWBl-SJ(c!08Yg4jX
zfzf>NLLM3zD*8KebkWHa^4pa!kEd_$B^X^UU@w}8TwV50W}bz6SAx)4C?U&?jRI4%
z+pukt`PF3ytT_-K(5z}cV&tk1xFVD_e{7kzxNkh^SUsT>y6QoPA6IliKOZZ7yoVF~
zR`p<5g7P;g-MV)8#Vtft`?tRXtPp`6^}E)9N8GTHdYh*!B_q{$#bat6t*ISO(Ym@s
z8+RT+3-9^+{NF(#dHdm6aglf4ztc@oZi=f@zKlxFGtpKnr<jY2LA<`=%`957^ajyd
zCymBZBJ?`p(H=X3n)^*vVmy?=Ms}+H9{n~}`=hN;Nw+Y2S0D$1&24cz+B`F^)SuWq
zWTJH(`O&^R$DoQ>8p58cA0@OuJ)(j-WEu0%pV(K;JT=+WzY(Xnf;$6ZaJKi7P;I5{
z!%bAyucnosBI*TCPR^QgnosaNsf(ssCVMUbJw8Dxpi~T&rJ%LOMh(mCJkL_9<u~j_
z?!<DdZ3(1(u-b_lL;;EI$qHhP$5lvXL*F^u*4i7{bK4F|B;#}3Z#LTQ*TQ2fxXe(c
zai31XKOW`X5hv*%Shvu?Q|fN?`Hm$&zU5t^R5OKr(FIoI+_5NqS`55V=k<D{))&>j
z_WAYNf|}4)tV+I0&UUaQ2fY(0oXR-BpH|}hXFc-YDVq;&6G?m2Rz#w6&h5laDIwLM
z&UaURu(LXQsp~(QvZmkZxLMCWxPrUCU=ez>%0NGXAmXd*(I;Li1A4R$Ak9vy5i$iU
z))NA8E)D2jmQrp0S&J=S6YB-0B<lX_tZgYgpg<AxxStoa?-<AI%QiJ~V}AR2#`{ZZ
z)hmr3Ztb*}ZjcT8aC?=QnXOmVno<(?T5uL{hwhJNop*R1_7U#{^6QCD6W>qO8*Eri
z_?W0(Fv*{{>F{3OX0EXBP@(JgK$DV7$Z)^o?>-W?{|m7tDrMdqt&fiU%dGaywQo%|
zPiz883SM0FQ*7D4U-aBiUl^|90)9u=$*+W~9oM@^oi#&-zhef7jvF-H1fR}9GO8MH
z)#CEc=sD^Q_ltolzXoS3P4jlq^sZKIV`W<M*0>>_;6r|eDz3PR)9*7m<p9Kf{5UaT
z0B<N&GDczcs-r?6SWL=dR2V-QCx(QAG<Qgq`BuqFve`m(1Yn0L4Ly8+`pF~oqXK48
zsH($FT<AxT7O$9R#+Cr)K~fHd%;;V8T-%Is2g2)mT@vj#Q542qY}1O*kEG`^s-W#`
zf-1@tu+FTOfaQR?4B{gg&HJFzQLC<Mz(eI26!LEJ*75yQkJfM59>`pN+pM6L1LHC`
zUeG30Jw4fm9Rs-8zWVOaL8Db*7yarzK5Knmam0Rn?tAOc=daL1BhY6T@p-jCnp9=9
z&8qHxxeUaxTu+g!+3$$6dGH;9lzEX{Z7Sv12KROJzIWr|9T9%#RO!*3?O*hiqvUHV
z**46uRm`A{5_+89xZ`-sncOztq4!U;HdQ?wBQ8sa#mlLYMW0U8g`A0DWO%RxFkBR7
zWYKlNjGK5&^0E6ckj10L2B?g~2%(v^iu_Sy?Y^Ak(bW&X{>a2ONuF!mpW=KNdLW-X
zq@bv5H1PSD{6FXW&;F#S{dkr7yRx%ZKmK0tdBRjZtnN#gNC3QsCng3AMp|dA!1$t|
zMYq+#JDK&kNe6{AaE|WXY2cclZa4hey+)93#}d>Dx<O@%7p2f+&=L_GE;fY0JR>n(
zdxYrrB=sfEUV@y7>1QsObo~9XhxT{#%84nn=}V|y3(Sc44Q+*G^@#8_?iViX30&{>
z?rmo9I-TQ$zlo<+C#K$NwSA!6g?rWt)r2a&J4)z1RLBqmR=uq>FyQ`&MQ;uITN@-`
zxs61o+FAp3U;M`-!6j-JHQwWnqpEnpuGsa3MxW(7uTH6c(|?kC%z-Q!b-(0d=chA=
z1L<Sg9%(6W6$Z>+lrd_G4=tl|7_*$?^OoTX;&j`j+0E={H$I0g>p4csyxipP>0Jkc
zx0_)nN%)J^T^3R^;rSQPS2<U&llJXa3BU3C@hbawz~=bvCfEZ!zGOme#r09-U*RXe
zN*%jb9fh3o+V{$@pO@Lws?dVp8<3~Ui}Pl#KhF!)Ki%eT!l-UuygxLwUVryc;^>au
z3u=8-x4iUD%O#y3CN~NDw&k-I*J?<yd+O(2cAjw5HW5vI8EgDFr0BWq<pu7ieV&C5
z27)<-U8vba>2TLWv!#ji9t>xAuoATLk`>5tPTT3J6P<J1a~mg2Ci$Mhvl*aTz=x2<
z^=1?{^h5()y^A*LLnI0`1HpL^WzHGq?RZDIs%PJ7)lVrnxLe&o?i3)MY}C7{H8~gh
zUleC6{``CbIt!<l)!ZuIo3vLn9W<m`aa%~tC9p`(rAb-18E)b+MgIe5Zc_Zg@5vE|
zfrqlhSZ+cYcufD3e;OtvF(V{@kMOl&TU)2k*wA1}CoHQnuF4j7WFKuJ4dtSIG>Tmy
z$-)9uk#8V|nB!<AZB(BIBJ^n7p&x9w6<39Qfk|NRbASR|LqGw~ppKDt$Yx#|6eGta
z33pw&a|q{P;x}l^JJ@wF1Xj0u^acUKiwf|V=?^d|$$IV4co?3AAE2)A_pQang(<6z
zs&G);Xrt0D7k6rq!t3vpT|?;Z!1FbOAc)+ybk!PbG48ESHpZrf%ld9gz4`X{KNl(p
zvADwHQ{$9iYs4YGx(vkD6qgY5ir$L-9UF<@zIh*h^m<GPQy8{t<)kjuM@iLD;vKi`
zt`mCH%r7fnC@S7}ni-W#ip$cc$9%RGwzs6;Fo!0)tq3cl%_Aow!k&3*Ia$P1oH=jn
zo&J7>pGhMk+#Z(M8^Fp;lTP&<MXLAQ<$U$KdNQpHeWTMo+6~p<lKQd=Fx_V9wzf1L
z0-SeZSftF(8m{k(5=*@o@4;pd+0_#bv~sir_1IRN47OFs&_s(Fstj~*p@oNTpnFAu
zU_$=?0t*TB_OYsxb)?FO%0p66(@ZoBuX!-Hw~|&rb#q*E$X%I_RynUv1sxkXF;8SS
zG-kWbFr>{e0%%tuy3iu^yxfY^h4KY&9Z*EJjgeC+q{<!*I96;!g(#!~($E1*NCK9C
z3Mc@jpaO~jF-brL6abV^0#QsuO(CELnu=_BFP?=+r1ly-_Tg|y7~_h7#gUK&d2%R2
z7pSEHhemJCplFe2lpY;`^vyf%mnwLZikAe-=4e&aZbws}#-&^Poa7o_&LQn2Jm1`6
zlkN8&rkbY`y`+hwzLcJGR$h4u<xeznp!STa_w%b@b4>oxoO9NnDi3QQI*iz+{kn6`
z1stee(m-!654AMh&x~$9l=)Dc+Ztw{)|uwd-HH5Yu{_L@D~p9-L!J+<c6K+STR5GL
z7^u(Iv#PW@C`pzc;c>wg$@r%4bRJM{t%FfXd&ZSrqH_y@#aOhAv*e2HIGLcT002_#
zVgcmV@gQqPPpw8`Zl<cjxf`K57^u9a&q@%O2%EU4tVcW##;ltU@`*h;rx+W!rDBlX
zg?d%{ImpL7D_921yDOJkr1rtv=B}8>TLkqSwu(Dcg4um5WlTy0iv~IhkzmhizEH6u
zzz1r7mCYQbsE-kQQ^aG^j!<sHgnrcJinLk~^)UNUUovDAl*BJuJ*WWB)`LI?UX*vB
z0MVKTAB6^h3Mc@gfiR#@0_U1a06J1;m<yUI0B|Uv1S8Upyifs3I#2*8plC-QDXjT^
z`JiQaI`dIC6apG~6xta=ccsQ@4F)kv252c78fXBx^%X3K9eUG7Q{1g5H36n1H#DBr
z473A^0m-1GJDk#k%_5ND)}A|3Wg);~o&`%7;3?R}0}d#r1J;}cCIX6oYG?}<i<Zx&
zU9WpDji}$bg<Zrn(_NAw2}o0S70p#$6owmBZq%u2s=$R6=SnFW9ac8k#1BxSsZZ9q
z&Pf#I?^Rxz1z8U@AFVKB2*oqCF$yVIW|2r?<Bs&#`=>OSaR@$GWhc_F_-0i`m006)
z)Uw;q=ub+EH6(_4q%;hT`KrZ!YNR0(<&!lv>h6ss9k`~0p$u!@oP*{ysxux&PZcs`
z0Za);dX#h(NJG@%)QQ%BBs8nY%>x_`Dn)G212q*7rcOHiwZz)nEZh#YoSlxk9|;}J
zp^3Jzs>T|m?u%$Vj8Wuu=7veu=un7s)i~w3t1_Oo=TBC3Lt>n!peD057}VND1qYfy
zhp4Awkwer}swp-!NTQ>=AG;q~NrTveI<%drWTJGblvI?77Niswi58?&KoTjbRL}%~
z`_t6bB3SWDnqm-|LTCav=~ZOwNtT9E8j+1O+7U(%i0*KSZl{*zs{<A6;iGD3^YJe8
zu)kVST_-vWQW|3r{VFnXK#A>~?!|0dHv^j5p2Xi%LL<darm9G>>S^M#VjL;Lqy^0_
z04`{x0ZTvyEdUfza0)2^Qd0n;fC@?g4J85)(`kUyS=!ad+3X~2@<s(r-J;mh#y2B&
zbDEDBTJqZ{KMpA}G18!D3a?*kZqRyD81;`Up0z3<#Q-=c&uVtTK*S#@^{08naX=9{
z1mM$1$;LRQAp++aIHzqLr~!Du#V!G$2Vfs+LKIK}06!Y&tgPLvV`#}7a4R~S8&TY^
zbk|S=o|VpDvrj+Fas@h8xK(dr2Z#@w6@N*MUD+6QuBdAybF%%#ivyIXpi(6A++f#F
zD~lksZAmo4Y3*AmqFOWWBJyWYMgvu~v<#tu>MJP6-e_%TFWCpBD6|9kYM7D~TI07g
z8)5;aa?&bK2^loYLGzqeuv(J$HBmUH2>Iz<@Q+fM@fA2n?OFiw7^jHn0m3?Xij!iv
z;TIh#;wqNJJVotK5mC$|mZbVu&odyS(vX9vN-2P66k>oDl7J4hv<xUCiU24y08&sP
zXvd`h7@~j*Fby$=D5L>LwE_~3ou~k#z9<;9bf92HLC>X8lyx)#Rr=IT%>afPPNIT@
zrTS7D5|kPWF4#ULBh*y#{o-*=Y-3Y$v(yHdklIR<L@%`ij+B{5aORu^B9P(qrzp)P
zQY5DVQpPw_he`$jQQm+aD9<zi?2k%OObgb-YG=8t88!6SX!6pEIj6XU(`!iy2u&)r
za@5$QN*b0vWl`%~6EC6C&M_K~Kr1$Xy<Ep6ijvhyX_8ft9n{W-n1(eo8Kwwu=}1ql
zB7`HFWAvoUh>CyFC2wkq)<%t$l#pslXI*1ApJb2U9Mg|lkqJcrA*dK&=9&TnpEKID
zB2O~em)fWjE2zSeNWE)BLmmzZsSxv2Nb~oNNWXTe0U#afMC(8j_N5&tC{Fb<aX^Vu
zFSJ7>XQK>PD%LHh>OwP}r;%CDW2S@F#5%wM<6-Yvp@AIHPvJI};9H6#`J@1MII3?#
zPRaxkTE>8MT83j@Y>v2;+|)*((F`=)RI!CFC>kE3o(4rpL(o*JDKVl@DjUus8TF)s
zTNUH0Db|}I7E-2KlF;?31tO6m)2WP+YDF{%l}$(uG$uTBqNtH5pa|Nl$<~uC5z-ox
zhMJAsMvUxm6lB#!UD#(CdpWtgl3!ED(xAr`?!j6q9#u!pCdC~&r4?i#?@cC{#xqk7
z6bO<IwXbYVZ%bl&nlU#OI<-`aBvVgTteD4!r)H!=gG-t~E@^W>25ECh0+%!ZW{Ln@
z(og|LXaLP602!p90+N9ZG}BB3HO%;~ZX<{uoSMqBv>4HW5y}Dh(8JS;^BG~cw=~8W
z8O2Bq$;Uj?f%&R{c*5s6qj?<kpkd~Eno%VHIarg5f0o;GK*)*aket&HW{l^G09tS~
zFaqAxbq6#6fRNY*yPYL*ro;a2Wm{u<8?u(T%CIoN&2bL-ke_-m{Y77KB%2^16+MiG
zInl>sU2)NvGp<AuAT=D8io_7OBfWY}T^W)_xfnbVN?{(AFdH#EK_14pBe^2AEuqc4
zXZ?{`McPbElgI<LCzFn%i7-NWRE8nmp)r})r~;a+WSD+yvI|s8+|@-GsRGxaik9Mm
nP76##ij^Rb+M%!{Sn5$1wPxfY;wo^9+NR(;%cq2;a6kXquXMNn

literal 0
HcmV?d00001

diff --git a/assets/sample5.jpg b/assets/sample5.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b2d1c7318bc4bc42f257644e0b304917457fb18c
GIT binary patch
literal 166400
zcmbSy2|!X=*!Bek6_;GV)YNbXToOgo(gqh4QCtvB$);QrN^@zonn6=jTyjsz3K3CU
zDoaaSX3W&ovJ{tWowBrSoA#!fX8wa#GvD|B-~WIAhr)gD;oN)AdCzm+_j%uQ_<QQ_
zdyod*+t(X{!C=r!@CW^UAJI+S!HI_;1_L5O5Tpbt!xlksAR)m!9i}laD*~AZ)1NPc
z0XY}8<hxu9!$SyA29wu>k>=}bfb%Mt{k)6;a^pS@q%dCxg3jfdf0N;lK^YVh7!UzL
zzW$z6o8|IzdAq>fz8;gKI7wS<lj64E$hP*jj*z__iM+zjc?HQ1N3vgGw|s?z0|enR
z1>f2O_sBf-U8d)n{L{8v%LUmU+&b6pU*-1S+G7tDLy)|$3U>AmcEyNq?T0gq6~4>y
zr7#G+&_Z~6Dg67hGD{KP%lV}W-{sO$<o7y)QpK;&1==lD`YMB)l`4Ogf#yol-|LHn
zu&?DnA9V|DgUOX$2b*sv7)A&Zf2$*x;s4xIjQE$V@co`mV&vCypg1w=s|;=`{#Sd%
z-`XR;C;DIav7f&W7)$sGF?n7mSRl(lHTg^493$|aAM0-;W#@vkcUa*-T4Cq%ef0ia
z2GM`ZhZcEIzFqhC-><~by!t@NyzmV=u)h}0AvwO}@0Sbb@N-|w@&7)0Q*riWo8@+i
zi8yaEj_r;M3<&|>D}%SZoeFR{%I8rF!2^H@dLzd?&;)`FKIPxSSIf&6bSE!Q1vTW?
z%HI%#`leF^AB@JgYfB-+`D;}`{igYP5M&OH@@upgj*o!8u&qGe`L*16;dr|kJgfo8
z@|V26{GRfo<<}!legt}TUSUUCL6Tnu{XMN3?(6BfE{GZE?dwkkQyYR*Y5ts~&2XTh
z#H1~$Odk(icmx}#conou8MK-NK~XVj$&3(hmi!LBR8L$Qs3e#FdA*-p=;fFVEeMDE
zr~JPqX~ra{ZUIdV0<|4t<I-Y)oCV}f+qWdk%jM&SiQXue;qt-7q=E(lSywJ^Sdgvf
z>nz9=xg49A6btIe+ngMm7%P{1ft<T-YaEaf{y@&z#);bw<Z&RIZQ7d10rDR}#w5l?
zr2*Zl%FDOJ#UudP4#=vhET$)rU4h|K-S96t`d{*vxE-LK5ahWzd1oqTL&6puK8AoJ
zxwtsvXmQ&&#ckPQ6BHG*F)B3{=eapCIVuT^>O!C8S3z2G-Qs|fok%WDPB!+oK>goe
z|JN_S*ZRK;^08g0@g{hF%wRfK{we$C+JDM6iy+AP9nekjKV{L?5Og{lf^={GQ)VWF
zphZ~_bn4o_`lBPqOTv~d$t#vE+rE9fEhjF<R<6+Rum5L-?=}Bx;a|_Ql|S#h?r>gl
z@ljhhZNbTv8nb!R=B=r?wB)FmIGoLY2J!#+!hbdEU+u7F#>K~_#wCGAg#natk~V<h
zPKxDh;cQOAagzS45&j=u_OCX`#RXggL0kM6gjr_`X<WlVi04-z1#J}w5qTJV1N%O1
z{>ovHeCDlQ`e*_7KnCB>fBuIL_<nE-PvdOB$%{RiEL_ai)NOJZ%nA7i5}4sd5C+l%
zR^1fBLpG2-v>aLqQ6MTrhXNrM6aoDR#Xt#AB9sbkhjv35P!5z2+-L<PfQ~?7ND3W=
zPC%!jbI=8-AG!>UKoig{=pHl!Jq6bLJ@gq+5k;69OdF;LTM9FWS;6dK&ahQ5D$F0o
zghjyC!#2Q@U|V6kVS8cuuu_-+RtuBD+F)I<3$Q`hDC`F8K5P#58ukf}fUCkWa6`BS
zd>Pys?hf~Zv*2st8{jGMba)oL2wnvj!CT;`;1}SR;S=zC@Tc&1z(J}YbP&b}8-z2$
z6A_4DBjORMh@TLAL?uFmIEpxfxP%x(+(SG=d{jUwXe;0pY!p^1&=f)yViZyoeo`n<
z5GXV%oKomj7*n{f@KWI~q#DuyX@y*Y^hJgvHz2nobC8ut39<v(hnzspAm5`<C>@j~
zYB`FAia>EuyHEwFL#Sh@Zqz910qUKiqN1Lnm7<$spkkEb7R4OJYQ+}CUlnDFGm3vG
zsVFT~vRCp}Vk;#nWhzxFH7Wg~BvX2%^jTR$*<5*ra-edo@=oRb%3|d+%GZ=<ls}_2
z(Uxd8Gz-l|XP~RlN6~%g+vqncDk>%_E-Jw)92K5QmCA9I0hRkIA62zfiK?Eek*Zr&
zi&Ps_yH#(fzEM+Cvrwa`vDLPy6{$(pE~?#C`$Jt@-B#UKJy!i^^@HlC)Me@~7AY?>
zU*x{%$3^LjDi@txbZyZy4P^}r4Nr|Ijh{3QYIJE_*LbI?tx3`h)J)VY&}`Bi)O@U^
zsAZu=)r!^1(yG(CsP#Y_p>3+|sU4%eS6i&zr#*u~VJtDe7%nCs(~7x@d5P7+lCfdf
z9oU1|^VoYj2ptO@U!6@l`*lv}OzM1EY`EBcaopm(#YY#*7JtxPqD#?@)8*?P)1A=$
zq=(a^>TT33(L1eoM_)mos2{AqUH`CtpZ>EYm?bNh#4O=2IkDuX0o;ISz%<xpP-k$(
z;GLn7p^ss*VYT5!!)Hc1Mie8iQMu7CMsrKGm#$jMSz5mI*QHNzSe!d<6Rrx^i+gEo
zU`#XKYFul4)%dfCrAeqsrpYmrJEmw;XVZAoa?>8umu5@N7-qZ8n$2#SE1Nr;Czu~F
zzhwTw!qQ^3#XgJE7LP4;EqyK1Et@TG<JIs#;8XA-{5Szca3*Xd93qSm;Y3Gb0`Va6
znibs2$%<oj$ZFIYY3*X2WG%A3Zlh}BZnMp%+2+2jj;+6KrtN9l7t73+v6q!ByR_^t
zJ4d@sc4E6*BrOu1lu0^6dTmd%kFh^!KklIB;O+3UL#M+lvK2Xwe3(4tsO`vb+~?Ti
z_{C|tQ>xQZr@7^3%hxZjS$^Fa;~eB%;5^`hbn$Tc+2x$e#}!U1wyZd@;`fy{D>tre
zTKU-3(lyRi?E1jX#4XCL*6rRZ+^Y4fgsblTVEjYW527EYDW;TIiiGmWo#4)OZ*hO&
zVds(R(c$sIbA{&~&u%Y-myZ|U>nc@?8cG#V?|PefZ}4vRe(B@rv&*NOhNSt?N@&-8
z4Si#LrM@re%jtXQy?&~GA$|hCX@8=Bs{a`VoIz)lF>VK#2P6fY4uk@I1Iq&M1X%{9
z1f2~=1qTHm44!3@n7f&mLbOBHg*1h{XL+#pvu=grL$`+Zge?k-43ma^SnajCboIUP
zW#N0mFGuJ{a3VU{itI49g#B&}bxp;ZnMlXToXE+wmTPyc9sE)M$4x(;U$<yo)Vh=F
zk?U8lZ(9FnRA7`S>Rq&NbWQZj7|)mkF;8PDvE{LIac*&?agXBN;!EQnZ*bdCwqY*e
zhlI+6XB<zCfb)t=;~wFD*ciC6VdLMMR&P3%sFb)qu`5X{X=76FW~0qJHjgG-CFdmH
zOIeXpk@7OtFIAETOIw@Pxdj8r)RnFHt=U`eZ(FtP;I==uhiyN(Lvu&cjw?HfJM(rv
zO7~8$-=(lCW>?Q{<J}p%@9%NnQ~ML_r|6%0em4C%>*pDs53ex;ov|@vIMXh(H1qA=
z)qBrm8D{aa9%TDuH|MD5q~=`Tw{qW+T!mat?$tbpyz0Ea__6$<e7pS0{67m~3Wf?v
zg;j-r7sVG{-S4#ja51Vlsd%#Fhmyup&C>MJ2W5=1Q{}jFe)-#qbrpk^<jNxl&<D02
zxL4(0)md#;U0nUAAVDy3(Bt5-8iSg=nh%F!56KQw4j(;YcqISGCm}~TRqIoGN@O9b
ztV7jptD6;vi7(YV*Gnb(l6=XRhNOmjjm*Yg>2j&G$)IU}GrW0g^IS`0%SfwNYv)m$
zqqWC$j^!VRj&D8wq-}lM<O#nMJtv({9&I;ouj$b4;Gcq>+Ii~b>4ekwIwLwqyXaj#
zXI7r+IBR>h;hgcggXgj5_y2<aCF_^Je@*}OU3YT#vkMz8Jm^{1bL(RG#qr+Y-m86n
zeFOd8{TDB}U;1^xZQ$JCiovcS=b_V=oi3lc;&`QF*m1bys?*g|*Op)F9B~;rGwM3}
zi|hy4g)y(O{&C;&%M$?;qm!YNQ`aM}-<yh^dVFKkjaN6f-TZuucN=+|e@Ej^<z0ij
zqTdL=9lPgr@BDqQ`&Xtzrf)rndGPdM+QUC*vS!t0D<3U=Bz<iE_}rZL+{lwPPiCGb
zKmGhH=eg$dniu#N?Y~ogzw$Et<;<(pSAV}Qc%%2G;Vt=X&%2;^ci$(z|NMdfhu$BJ
zADus5`n3AfqtDy_RQ$8*3*pO|zx@8X{deNufB$_8@`Tis(JE*qH5Igqy1LpTZ5&ox
zOH13*z-TefmO!$zCD>Ryte~%QaPnSmZ9`e@?(HAI3}M=@Vn?xq*3*NTf%2Qc)Ya9s
zHMK3USc^bM8^^%^@7LemkfsuHn<5VZGlt-rFoY)T?_Ovr;50};AHYk_Rpc)i9HD?j
zDJp?Rg9;rWc>zNp;0g#NQUUN8*dFjbq@an^GA6mBw1cA*O;RxS86~w!rXFn<u*}hy
zW)9J*naXIL#kzX><`$NC0+H<KwA|Tcg{K$Q+lS^$4`GFdtp-FZCN?g9Ljs4Jwq@(K
z?K^g+@6F22*_WHgFD)yts60?rEfCd->m?11(xww9+dEF3?(90#bFsIt|I)zVkZf#x
zV)FXbjhpwUA3U6yee`(l$*b3I-oAVP;g64><oyEi{6e?B_v}CQOB3`9uAqQWP?Yxz
z2H!64xTXTqn1s@D4_1sy(KfNqP{MeW)V5tvHg#aW#73u%qIJy3_sw6)yEfmm|6a#3
z|3^Lh-m!oBbrVuWfM8n_p$Yu}3AVf-4)42zEs-{^2&O5x5gjOM<emh$J*R+^f!W8u
z^sB4>#$@}NaqBd;hF8BRtq%$-twJTh=poyfov5K;qXc;D>1*t=XQJZ%j^l`WR*BOo
zs1{+7X;bI+qz}TBL&3DDfZ#I9z7BV*=f4j_A;mAqL$&B%ckJ1i67UdSrmsD;j$Rb1
zA`{2D{UM#E!?d%8msPqER9)NVyf4i8D~k%$Q~Vqfbuqo6OG&fhP(87|O1JB&mR0X)
zm9)%;yi`u^ukcsH;b@{>!nV=SM>sTv<=*w=?Z#}0lKE1b&?1AuPJRNl^kCS@7e^W~
zy+dYi1gY<YUiYH7gn$HCW(^)r>q=Rp^Eb4GHPBee=@Hi^h~Xo6Sb{K$3SruzV2Uz0
z2ElwONDRarfq@gyD+<zB2<Jm|JVc&9!^rFTQZkE2;QEP>xw-s`f-$`WxCx;h3Sn1i
zoX3%SFR|jQavV+GNld7YI5LkNDkgoAW{ubl69ow#Wx>mBC|W`()p$Sq@CuZF2VRWm
zV;I_5*9T;Z<8rTC9PRFuARnI%9pH|t{&FOlwYNS(HT`Y9&|{Q>ls^wndd_%GQMIUP
zFexSE)79*KKr3zJt!)p2Ge^L`yzTQH0i{6;9r@Ow_VR_knPZ@B5`9#bMBu4AK1NZd
zCBO)HsE!Qov2ZmVk0>cgW8SnS@voeuY!MYJrL*FBx|~AFz%Z#0qFUZ@%x^SOZpw8q
zFZAkf$#e*I5mL834REJuY5MXZe{?ELKv`3OK7h3rx%glMhd!RxWWaDn<((2^NhuFa
zfMOhp_rrLaS9PSc8{`oU0Y^f6_{55<{hAqP;*s3dl;&G0CK)d&;!Xq8SIU%u*5Qnn
z!kY(X1kJ*IxCWUX`@@$td#o_F4AzfX=LbU>?80q)g*90d8O+O?MabaJ-8E>+iW?4-
zM`0Nhwc-u@@CVZZjmtSblWrFu3To}o`|4gsD+Js|w>+}o!RTs`pW>$rYn;9!go>Bq
z2^^?Bo~Oe{0X6!4#fMzC@L-C{HyzI-f(ikJ>eC^-5cwSR9xqf7Wxyn0EWn5aArgd0
zF&XNQ9QP+tREHJoNt!><^IgYKu0{mi5KTQXylNjmT|ik;!3mF<%i*bxs6nb+yHrj_
z^or8rMf7=;5M{crCL4;XbhKK`jVn!`t~TLPBC{qQ?+WeBS-1ZMtFAsFH|9XS4IGV!
z>V@!9o_@M)8yU8QX|;a5+>aV;dSQ40BLF6HqyU^RbOH2Z{`_ly{#`g<_5T195f9A<
zeFXCYFH{e4Y1VmwEr~lrr(l-^k+Q}UDXNZ3Lg1E<H!clgvhIa=#IICk*@XF-#ss@a
zOwJl{!#@Re@6vatsP_@SJn^*gk%^_cm*#@Bhsp5xVnMXlJUSj4&h=*$``>9g8{*ZL
zBjE5X?_y6cRq-v<yk;?)5$e5Gs@jEKAIc+o+hCwBNn0OTfnly$P$Njpv_N1wAop$d
z1#Zpern9ZeWlhYpiJ0E-4i{`{eL2-(y}s}#4j%s2DqH%Brf8DHJszV}-A6Vx`SL!~
z%6QO*Wr4%G`eg_d%vvvDmGp#)5BP~tpd1?-Td?*-^Ab7{4HUq}>x$$Wm{-I9{sBye
zh^I076*YYSAc&9vQv;?Y2u4fLD9_=6e(F<=g^)fQ@*rAbV9Ev~gxn(cpwpNU5=$nC
zHjNrKC#f5RjE^ABGFx#*=^+l3UJG6Vwl{{hR{)(T|KuN2+ES0ywsB>`snQ1$WOlPM
zEp~9SrJn=acBk1JAv<a|FuMN*zqH+ZQ!20^?o`Oe9KaV5$i~Ahp9as0Md7zhjxdz@
z=Q-oO^QHnQvpfhs-|GLjXbu@l8|on^0S7~pN_ZMFeTXif0}Ay*B*2JV0WNmL9S3W|
z;tnjXM~M|AjwF$TKbPXobrPbMn>60;>og(=Nl8%`oSW=b!*VM<xK0w^h)?6ATZD=o
zI3aSFctlWqIc2=U6j%K^nP@QPfhO?EnNU4&HG`xgLTK7B6nf<Mk|4*Rsv(Zht8$cl
zBAY!@B5<18A@0vzdt#?F@quM2fph@sb7d8P5m4^)G!22{lpW(1!*dXi#R|&}_Og0+
zG<hu6#fELTygo#8+qD-HSMIS+lrU#L$5+E=PSR8w?FesSNq5{hNmJHadeaWbUKD>0
zwe9eTDUWZ7;f|QJ@3W{RQM4*Pa#8n26iz*|?dVXvgSDr69VbxTJsXA5^L_?kge_Ok
zzp+BWga9p}NfZ<?zhe2k1!}7k$ZM6&TNXK1$|3Nd#nhxI%XkSeP75Bcs4GE38@Dw6
z*yta(qS}<<RHe?|Aw<@riq$Qu7!X+ylmL+%2^n!dm=05;L);-IzM3=5Gc)Awcnt3e
z%@W354%4oO*t|a{q)BP~#xhg-vQ8p>WVndURsi-*Ciep5cOo9{r}m-JY|p(Rdj$#p
z>>Uq7G`@Kf;B+txc0_)hHzD(`^WUQUF2FE`1k#~IyLaNCNr?dR1g=KHhd_5lJe^A3
zyth%bVp2ae+h5L;mCIm+7H2SYg`~z#O+jHJEJsuv8-$9x?WqTEJGxkrFS8CKc6fau
z=2<Og3w}6D+2^>t9<t>O{1oL|&}C>{gxxcz+7!UBApv)8j^PD#n5YKX`^Vm638OeL
zj|X3(nx3<=`t|A2j&FC=`43tso1XXiX(-A55P|d-7Zj15;MK{lB-02SRl=?yg@6ea
zgAv3uyGA}KGzaXtUAm4Q#?_}<gDh?Q{beWT!o-_w{P3bjX%mf$XQc3M%v-%Ef0ag$
zWsZj$1G8NxjCHlz;wAjpGVk+quMKP<*5-3u+L;lBKL(m!^P#P$KXqFrEjmMtW%<Et
zh-lzyY~}$%NRY!qt|RnU$b2j5v&BOuf&e_s9!xZF0#oY4K)@XN17bpjP^COAz5*Fc
zU|;#ann?ylg+_(ErbVWxDm89)A6Ud=i)lFrDeCNGayL6!3>QQRQL+&j(}$wXCwluD
zMiga~;iWpoms$f2p>X=~a@987!RFv&7#LAiUv|nBolwc`8$y?MIK)&|W-JVTFh$KR
z;W<U)S9Za8r3E+PmWe>ytR7WuDov21T5cxiEdmV8z;7K9@w6)CT?F5QBOTb&g{p)l
z^^o0|nn})SI!A~S(4m#UWPW93U=WZP*f81VFk#fI`%Oe88v!qWD@HKLZ^ivy;}>a4
z!**kEr^cJb*dz_j@a-kfE%}|t9i~`;vUcN&HLVOoxY76!L>T-aB?;{Mch?!;Q$PGJ
z%=`9u&T?|K1Sb|E$mP(C;S&l$Rp9u7^Qa!@ko3c(enwMq)kPX1_ax?OJ}+T|b!s3v
zhuyK?RnUndhL5zj=U|GaFtE^r78NMYacf+6gA5W|6svM`0+3A2BPkX;u#sDvH0w8X
z5BpiY%3${AYNk%4+d<;~A53amkY)(+L6yl9n<aPpxg*tw#>aEp2Q7m$f``Uou%}F~
zl_LtbjuRKp^0a=-qb9BSG8$qciokRrRVcz}J{!1|?@*b?3w}Yvi-A4o;~{CO>xjaU
zxRNvIYro{(dm6f)=q)4t9yZZ7E<281>yyr^@F9mQ9o}>pzvwBICvY{YFpgxxs;*%)
z`A{No7CB&O6y!X2-UwAmMgS2~jS8~6AVH{+a;PMG?Mc3vW>h`|rH=JM7%-;=2`~U*
z@%~mQ+z{VqTxuRsRN{3x5X0LceIQg(4%5o0jDAA=)2|P5ZH+jUG(KuQ#hYDG6(-pf
z+7N{Mih4PF5h{tF2J4sh4zF^$AXL6k--O%o5b)c1?qxntX1>v|c`Ngk_kQPQ^M$}l
zIX1#t{oM!{@)$6TvLLwrH=HFb(0)C0q{WDpNe^f@LCwP6vT+z^O70h;^a~r|$|!35
zoxZL>EKN0nB{gaGm9;7Z82K%wi<H%zcv>w}x7@CLUj3pd{1ciIalQ)iu=l6XX{i7O
z)CEvWM=>CpNF1q6ixu)!RGElX531ED+~TSpyvi_1U5aD;PKz&vcp>L5FozZoV|F$_
zr6I0jF{K;om6Y~bNEGlw#XYqtzV9pEiMp327`4yr`IFCqT9S$Fzh_ViZY)V-;x5Bj
zrXI^GIBQQbNwt};IX_J*rdzM%pQfC02n+M)L$SwR3x+y(cQDNh3SuY9=?bPb=!~KS
zSf{i&3b;dHk!hXfVRza-QvLaP3CwW}FhJ3K5EqGIR68^wQa*Vl^Hv&s=mO6U%#AR<
zAeslcNOcnTk-Lk3)4Y2ySu5cJSJ*`@NKsV{^(k+=H)0I4lu8f02vS(nU9W}S2JdAQ
zaR-vAexCmBlR0`)SQg;MV06qOV==H}fOeEaRpLIl26(iX#axCN(I}x&NgAD!P?K<`
zel%?Iuo_OMG9%j@GPlKR#8}i9Cg{28O7SJgaA2@nUC|UbD#H+AZ5~dCIu-@9?TQF_
zr4`ea+8uMX`tMsS`XaC*hiLvDIY9;>Q-d7HOg+^@TnAdMgtt?uf=}&gn9#ap0~j<L
zqA&c*qeIYty8H?qxe=T9&kKd~4s8Bj>f;wdI6XN>t8J(!k02Pvfc<|p+vPzpfrFfA
z$Cda-=*SUhZf-#)^?7k%OVCk5rJEuyl-JZgSJKBs=)bp)ojx%$4G^}Z)ol&dOMMyR
zHwyY-eMw9~(l%#k`_UX0gV?)9irPNw{a3k~p5;;BL9&bkMFY}6qCg7c{3<c&$~Jz>
zb>l!3Py6`>Sw?F3biUF?gE#(;9{LGz-eM`dy;XR_XqrG0EveRZ(rMN+7u)?tsB=iy
zw>Ewpdkr`KMT(*22S+OYu&o&Tr}R9-vTS*4XkXN21@Y^R9=bzE)$DXU`Pa_ue#NX5
z4sCT4qJG>Js&wWd@>+V~su-rdC+fccXI;0(&S&^jAY#ma{fZK9rpLZqWB7K|`taRV
zF4Cz)_Ig%D)2eo*AQ$N(IM2jVz=pK*kcqeU++-5X+hmm|@dPRYsGCiI+M~ZZK`;VU
zz`fx`#Yh{w#h7TGQ8WE1A1XYU7<TBYJ4y=EFwZrZsBXBW!H%|<ROB0sh3;oVC8>;v
zt5dJ;@{COczr5{r9%8L$y`uG{`3lW*cAwHewf@9nQv9lRfqoq4Ia|lLFEp;tf>h3g
z;Z5Z1T$LHUa0r5HGWy_Jf}SeuRJ0&HL~}_IQD5w*BjDuHlgZGtS9CSw{CRUz*BV~H
zgwxqn;|KgULxd%}m<j8V6!9$ShwyqyMjZ``bcMlJC5-+hjvmWy{E7?XxlV}8Y+OUk
z=Dj{%h_ajIdzuupyh}V8776f#gNi5#mR4!A>Z_w&i0{7<EnxD1IByCU#CawLW|a*D
z45YbZrx10xyk8tNCOZfa3M}~il#}yH07-n^c_>amA7}%HOrL{?wKz8hqQo)!bs%bF
zB~mmxydIFd>(X@YrCO;!F%eHlEH(_-aXKAiQDY2Z0%Dy{L4@>N#ZRd}Y(ISW?d80?
zJN>!_$vITuut8rkXv#4^&WigG`h#a`orLd2_m<sQEZcvg@i8IN+TG;287scAghaKo
zY|r0a?S|x*I;{tcJF-W45F<jq6|Y1aucNQSt54v(p7Pi0Ia0!@(?^aQjq^ad$fwc`
z4=E{f!*FP^Nz3oqEjQMrg?Pk<nt3(HhG;sc-gz={CVl)w$tFDJ*&=#L@erkv^>7Rx
zwyyHz%zlRukGR}1?)Y|9NRkyCe>vzpUeknca4$%~rGXEnrcR$q{N=H<B-o~?yaUN6
zq{vmNTMta(LMFu)-v{Z|0b?a+wl+3)5TQFrp|Q&p>#{Fdg53H1r9D^ac~&8xqqMVQ
zDj6ACvMO3FsNm!J7{bT&vAR>cclYqD&33Ke8~Pa`A&1LkU64sJ^D*rR&!*SIt$X+W
z9^G-jOXpQ);jQ|WaMQz&1C-r1x)m`z@ks;{Es9U4EC*iQmrpU*t%b^?tB_>C_4rY>
zFcp681TIFu){R@DN>`Ip-xWL#^ookm;zOZN!(#3+VZCnxP-XZX&6+a3D6Tu@5{y|7
z1yroihfXhIkL~Rs*~|FoF|*P;r#d*&0aUBD2|ZO13s4OnNQI2&7-Z<W9(HZhk25wo
zhud$WTml_A0Tl8`)sxh>6510v=~(z*AR`N;Zr;+;zcQ40TvP&jDkp=I({w0m&Y#TJ
zr!Jt6p!QE5q8$pp%X3gOZpqgG7Sjm?LOHcj!?b?8#wn(rHY?HJixKQ-hpR}nBEGiX
zD>t32kJ6@!*l^}A9gg*#m5eK^s81<6M<w<KuxtZZXbJ@d#?9Q;fzo%2>3wJpW(=>O
z;*5~Cc-@$r#YF_zk(3X^6z-_W19&k!07=q!8f;T+jX>3wXYxHUOqXsU`snSkx*y-d
zqI@t<Abc&9BT8eL)c0=_Tm>44J}j!rtBpL~ieIe~H00GCbO~NOly)irX{~aYVP~0=
z-SpRwYsgAzs_1KnJ~ny81xV`I3Uri?sCWbc;N_?Bv2W9@J#tpZHY89Qt+#?%%mhvb
z*{3T+%&Y&O4}dugx+3NdMMt#Xu&GCgaE!JZ&P?W%`0O*=E-^kaq#+nyHcL;T=v4kz
zINV3<ZXGKKLbw(j#yib;r}UAOAI7Em`^87^5G!c)EASlsu^#$RZ0Zr~#zy4Dyi*(a
z>%$}5<|1|uG{)f!WYfjHea0A`Ka&adS>PAmx)OMFCLZntl07ENXGNxh1f{#es9~mg
zp1Z^tFEQp&aZ49)_jw|21vAF85HHm;naFL_i=)y-#eIElcz6$MkaD^B9kxM{EB3Q#
z{Xq}wm094J(0yaSQTeEK74sNdw2;&^2U)c!43x#vQq}htdsCf?ySFxLp7O?f#0tkU
zW*bY!<*}+fDlmsYz5-Z2d4gp=N|2xYhr;=o7E>q#NlZzJr`o)yo}+-wr#!X$jXaK0
z^xDTFbj*!wd1Jm5;K2fbt8|dnaXn2OL5Fo31w0e3G`SiO!i|mFPH|6>*e2mf+xx4}
zubsyCEPFt~PRZgk6fv>{T_&(6uerf}IK#=qNSq<nINNxRqIZ4iQ#J2qK|9S(nizBY
zR-vj2)7d?QFtnH1zoML}%d$TYd8~?WCLKjWzEm7Km;n{agij_`Ya&N_1j8V#f(SiT
z`IUpRcFcN~{m2u37{r7(+~sy?0UK7};D3uBZFSoTGa1=n!^#@ymd5Cjm1^o@GJ>ii
z50@V$sruImqVA1pT1R%aHm7vUEghPj*#~GHi2A4b6o53p`PuRna(4B=;MK!3w20$F
z)1*>jv5sk%k8Ypuud>EcWz-t7>@(61zt;DHOVgR%d5=y#xLdr~_Gr#M%qw5Rfp-!e
zRE?r8`Q%z`Q3f{pJgSAdRe)lhE7x<!T>D{0jHufrb!0udI&5cye;8Wq&aSj$ZzHJs
zGc}or6mPt46)?2l-qFQ!OIMf;c|5>sVribb@wF+eycmEtKu+ld=+k`sxbA$|>Us;Z
zt1N>_dO`||o-HHj1f?)DC;)X-D4@eW2-!&+XkD4c$}34B#$jn0+a~hkFI@PYp<LOA
z_1gb6_&{@>Qk4BmQo+!-Nyf{uW{fxJ1Qk<$FpKg@P6+|M1N4Oc-w-(<*5E>M#^RuU
zi9W;uOjj&~=7IPXt*}Fk7QTi|(^GehmH;Cz2T?57NvfyWc(Y2=(Q8~OuSpL@wR3n;
zds{7TMa8DDTxw&nyV2~@l|NOetP0@dM$MM;wB?>%#IrWJzh|zS3420#n}cETm`3H@
zwf3`gkF!*^S-k8D^2b6Cb5(TN;F(ASTl32WX8*qE%HrOb>?WoYLrHnP=VU94ud4zu
zM=U0l>T~(J{0~2GnT&eM%IZ0y^@6M<%ScADv8u=2h&8`Y1(Dip8Ef4$JT;!NHg`)h
zs;>v$JUR7n!X&)OA*?nsx=bNL0loHeaCC`5l}Y23PM0Q+n0qcMYqmX@!HAtJvu%7m
zWYd6Q$)I`=T;R}n-RJ`tD7pZ|tRNQ#tTkAX>1@CYm17W&%PV!aG}9_oP-8W$sp&>f
zr<ymx{&slHk{x%bLHWrCW`<4jwUYkiH1#nc`_&ed8*|L|vvqwqVvbujL%U*@XIeeD
zY23qO5m~7NQ^od=JLw<C<Az`5!0c7Ws-nd)G%-2H+=8G=0;zlOww1S`MORWn?I-W`
zk6Ts#!b-C}_I+akV=6c$JRv8LUie(>hM7n+PCfC%3BNBlliSN=N9?-7e~l@BU-!&@
z^vd3f{4YmroH?)4WwV|SWBT&Pat>cQZ=3R3<x9<;gwIHW`vb(O^|SRl<!Ot`x8LAg
z-b9FW@kMqTHc@lYwkx%orp54zOHoq|_=e=;ub5WR8QPcd*q`pMPZ=g#6gscjM!Xr8
zJ6!sFI_qO(h}k*ygeJ3}xp){r8$hrJ9@d`2qwZXcs&iCrBe3p`AnRR%IMXc|m;ERz
zUlS>ELQ^j1feXt3|A@fievPW-(HXE<s$w!<cYFx)1Pq{jKB@)GXTj9MYB*LVS>&9I
zLK6@?O9WpCq+&q~x}?OeQSc4E_Jj=8l82$oPg5;cBM=Se+3SNV8P9|DMPBmBAcpG;
z?%F&`d5njr+Hoku#{1z1Fg&~%MoSNYO9N}#8pRJMj7cwju09CpEW1pPc9_U3vz$Ct
z_5#h+H}qrLD<3QfO+G8&*?M<k&5g4`uF)@As@1=h{L&5&Z+&#fvHGq+TQxL)>kCfi
zvi@||gL+Tv*nMy*>GoS#Jf?fLWt-4rT-y#Bqr~1`6+W@vo||V;xKsKG4bwb%giU!g
zS*&e5Fk5>lRn55M){z?YZ>MIc1a<Qgzo76=u*_1)Qw`=rkqdrD24WQ8f53??0eYv`
z0As+S@qKHH9B*A>k81efZl-u>><ZDSCwBDym{X?jP`}So6<>4%x`#ekv3_IXrFzHK
zJyHOOyk~p2VluBG5ah0XLzYut8lAGX)m6FB#2wBpA8qa|ag$25Mw+(0<Whg)_R=hL
z`vMgXN?~=2O)4sD%X2EkR`i*0M*CR~GV7N2&}zf%*N<O6?hJQ5v(B2?`E2Fo`q$aE
zQDVzdx5n2#9kx+Zt{HPa#lK`8unTw1ep~-~wJbk7%Whfy@<F2tUw+R=Q(3tezB7D#
z<uBzsqbsJKSyr~+*L^F|(GZHK9Ap`&3p7>;bQ2GSYT<#s7S*WPVf2&9$iDhorBK(2
z%kdxgInw-ZcRXinPUa3C(6n#A<gXuF)}Jam!m?P!2gznR6v2`NW?wx-x}N%ZA|4|k
zsENQ*=$ZnSNUGJ>Fs3oD-*%8={s*T3z4C(jtLNSKw?GKgRjh}+Icg@w{Z9dAZ*t^D
zS00h$Vbrg<J<O4uz|-lv{Q`N0(UaNlaa+Gos+!kiAyg8GxX~uuC<#0zGZ5O^vhmhl
za{cNoY_Lb%+V-r_+(dI62o15AlhDe^tdwUI5(Tj=JwZ(B2Bxe}S4_h9b8?R*Q=GBz
zqmR0qJZze>*41xV>wA6xGg7_QXLwEC2Sr(#z9H>-2GhP~e_3Eg+AzWv4<Za73~Y%L
z(f$Vch4COWHaMBMwwT2q?qh0WLo8Y|+Ni3UWScP83(R0u8B6&(nb%%~Nek54E2~n!
z)5kiqn;`Kme`K8y+N2px%B~$YF43mf2^9R#g_wbrbbOB4Quf`BrFgl&XYxy1Y;3H+
z8lJ74s7i$jM@KR@Z38%7t%Aaf?ne<)s67|>_|9jT&e{Czyr9_e3#7s3Jo}Mqr6l5>
zg2h&Bz1*d)_HUR?s>yiRQyqr$t!4vF-%P8u;YaN=n_8>iwZvI1p=z7|f?WBmw%O9u
zEBEe5I?MZ(p)?JdrLl|o5g{w9tc^OhX%ScIScQ8Ps;AsryXx>AR;ZrML*|a7xdTJN
zZKzrAi=nqRSYo{d#fOQz<4=~P=u<7MtOUw@QfM-o=<TfUKtXZRz<WDWw+pOX3uW3=
zo`ZJ+BnC8Bjw$#k2sQfbRGFF#C4XjgV@Zl1u<sP1Vss5mg5UvGvq+^Z%viy$$s>N|
zh&wwNJ*r3hWiJ{ZH#FOLFmJIQ*4#e)l}sM5%xAD6ejH@s(rE5)Ad53G^`*r;aCVH<
zPpDi`qt3}?gN)I<`Tf2!Du>+H6Yvc%7JxFyE6L5Y@U0MK)DB#Jl%xZ(gz;Y?)DzGG
z0O2tjhgAG4JBC$|-2{??jG#85SEB#ZZV6H?;8FBWrJZgY@djbJ{!EUfDoae}(fzjx
zRVI%oE~!f8UUe2vVV7BG9P4U&6^R&^#dzIKFPh$)+@523Q>W~?)lHtp?N<M#nw#C`
zawX3@%R+8m%IN&iq_KAJg#F6Ns6iJ=R&U@hYKc0xboMl^&qemH(G&UFcOW$yZ%;Mv
z)$}Rz)PtST6MtTd(|#Um5_l#bn!ZtwC>T*#MheOBcW9fVXbe63!pYzc#*SLLh~h8w
z8S}Fq@bwNCE0?B^*9P3QwO<~wGX4k@^`X$JsITBvkbm~$SuN_~tepDL*;ae@@i#o8
zlZ((lN@}7<t{P1=9$xwEW6SMA40TOZr$Oi+Iit*nj#`^!IrUdost6pYm!h^a_xiR?
zMUDyppgGEts&K~m%Y;o@h64Zb+s(>?Bu=EYUj6W56;RtF=6HM!L1LxdzZ1K1>4A*3
znxx|6>Zu#ub+O&e4vS1vhQo8TOZuR7v^p=nAO_XGsEMH3fRE`X9Q3w2T-)N)vPL>>
z5N180x}&R6K$;6qDQS=CSbiCp$9htoXxb!UpMU7imWdonYm^n5K5PCZpD=(`RSJ4|
zoxa^$jmy0_5?YfIblH<(jMX8K+G}JnU{OSe?}?5Kf{?aq2sB#*sY?!Dy;xCJPIwE<
z>difBOLX6p!)=Y${Jcf>G4RV+Ku_m=6}i0V%NR9tI)D4y`p*u3L)-F^7oT05^L-!T
zhWF)9f2j<l2eu9D3T@0ASl8z3Gj%v0{UAGkEPHSI81~Z78w9}(hd&B!X=ir4l_Gg>
zMHf<jTaV~)4qrsyh8>JK5j@4Z%mvY!#pPt5Q^%+`TlG%7{&FGkd6=KZGS`sqcVKx(
zo}rpg&m+n2Kt|W`fGI`|8<EWEFh>)6eXQO}vIgDj6)}T`U~aZAWKh5nWXZqr3G?|m
zOHuZdM{(r~egWpgg<?V&#<e+u1l@m`@`d??7pi`u57bIZbI1$XeUKg)_pUF<don*w
z9N{K0!IAzp!Z%BPUwVdEtZO^#ZGFj1keHB%_6JPUlcwP4*0S<9KtB)^TW5?5T6}oI
zwC&nm+MBBZ&W1#sK?rtVhGlI-V)^Knm1=}#=Z#pBd|{hgeM>2Q`JU0i=G6hq6Ah7&
zr`FC=*v(_>O+qveB!7NrAtd)Aen(%AyQ74$qyB8V*+S4S7VQc@`9{CZdRGq7>{(>!
z(Cp|Q!m5D|gBj<2H&`KPq-$tj>Bt{{#e~6A7<ncRkfmAh`;?-JBc$E+JEJc&dc0vv
z;zymAggDYyUmnv;y&dPxd5wpQ)|BU#q)sFTMVD7iXOs9bsw~k|{xIECA=tu<Uv?=Y
zv}d|~??~PYALBlsFiL1M>ye|wHqWA2J|2E#Kljn5JK)L8&OS^#FPKi?iE4;z`)$Yq
z4P^nbIIE#mEz@uqC7Dt7rQgE<a${qWqt}4wi`29}nUqx^Xt%SfH$7Aod<`3~$&Wpt
zUGKlYXdGtW<;L+Hwj&PNPr)l!rtakS(<aTP?b;{1Tg+P6#@n!nv~YoOR!E9})TzR1
zp~p_jn2{DiViD4n9kCNzMqZ+ShT}jmp=k)twP_I__U_APa;Rm3jN8d$Q?f&jg*t}4
ze_dJC{_BmLLwl>XzmKZdkF!##`uTlA{a;s&%2W4Ot$Q}}R>*leN6o&nY-;gm(Sc_w
zkMT2$?e0edmh;T-G2!(-JJH}Cq;tfvdPmJPb)AiJ3ct=hu-s*z3iT~UG_eP{DmYE$
zedt@vukl8iMM1NRpIwygKe@==2M3C`{`I)-*@KY~aLYY4>1bEpO0<h-)jW-IBfQS9
za;>O;eL$9vp3Xfm!pxn_SD()9!@JK>dCBf{jr1URb_2}ci%;xjjZ|;l864Pbz8}HI
zFK`?{UO?VfA)cp4+miq%tTr_fnv~ffS6N(S!UyijCN!K4WFzy0Nj_^pdIQXv`8fpU
z%)B_i2)s~QKL3R~VDRL5M<YNv9h_el2J;UrP0c$iJYrSp34EeC0<e3qazU2G=)O~y
z#+at(HNthU?D$*$Ug&$7&p)g#3%`EdCXF=s(Ffj6(_4cPhHBa3RVWD65U?uaeB}g^
zJ;~S9hX{6_uR?9}HkY$}Z!E^JUN4rFd$<u@svGe4l^S+>j<CAm?lX(%!5VKRRfC79
z?FKIy@m<zwPuJAxtnJtr9sADF@a@^v2~T~P{$q?Lv&H*Rfa>%8#r8j+?F&`zYNN%M
z`*t8*CmrICd#q<_Sef*@K%p+`j6{VECOb^K2c>^B;Bw2)H=8Ad9TpZYcT6-ah_xBp
zM8BeFQ^q!ls8xa5X&<Etv8U!<Kb$DGauIlW-wWNjq=wbIX}48h_+x@!ib(t}A_uuq
zt>94?Go(`7s(J}l#&#-s6}-GQnSCI0l>aDx0BV5IkoBT_fMd)aCU8k9H21d>p3@Oe
zznRk?7$mgG&o2bYd4EG^5)y1(z5a&okaybuyk-1#+a1Ll58WC+lMk2U741Zmo;MHM
zn7wUSKWq5jlKkuX^xdM=MQx6#TbG}*y$SZ_CUN)Hqmy2wtkJ8y4^4D-3#s+pnD~rE
zg?Y{f6YUufpYwIK^4;nzTS;XbxYZ`2p%=FW8B^D-Z=4H8INAvYMnD{G3v$@|njVf_
zS<mm#N($-Rck|?iEF)3-Uh}-o8h3bB0ZWn`*e~3pfU);)Tjb6vS#G9@-d^6}t9fNv
zxf>n;`3OEuu^h~t%WUh#i){7~Knk9kdH;AibCXedxUI$LocE}_pglb}gNA??pkpw>
zKsgc%8%*!Q79u87ml>iF?8}SiL&>6yB8(^_KLDeYKK^RVBX0b&=t!8K(TZS7HsDM2
z1+E3Q$#4L{AG?MUmLzlROAIkN5rro@aFV>}z5J~t;2stt*M9}G|7lNNyL>C8d~w9E
z5hkhPsDU{^>7QSI1M7{$3jr)x1HlBL_L#?pgH)fI#X*WHm+l|4q~2@7)P$QmqOgfr
z@c4Bnt>3qa63=zW`Wy1jnpyq((vXPGR<LFeIR@JywxQ%`Ggj?w#gU)2{@H%};;iFl
ztaU={aNC|IB@buY${hJx2e9URq%AQVCT*EkpSks-!7=zY<8jEZ2d`9RjIf+Wt7DBa
z&je=(|0r^ubp3sn<}1k>U>G-<35M+6#GOI^YF4!g?VphMa9~z)GWYbgB4Pxs^EPv}
z>s`q!t>STu5L|U<W6auXYc;3(*BtP8&Zg&8Xo%gcobS#!g;t+rUeM9DIHNT(SbcNv
zpP#9r1#!IQH5wn5A&ohWDxH|B(K>yQ(|+C{Yb$A?%?urLWWisr4>p9IcpL0=7uK-C
zzKE*6Twy3_ISH%U_9_f)qA2XLbP<yhoBi|EPv;I=AUCfGZwuz<9WYcexiI%2(D6`y
z=lzDI>~+sx6H+BB_f4&87hRkh9m>Aq%y}R?@Yyvz<oHpSE$`k+b^3sr4VpDN<@cCK
zZCmH=k(;Kag8l1pL%#82!qjR5>+2)Td|-}K_u1X&$}RWMyq)#j^h;*hv&(aU_K0q$
zhh(8wGH2AIda!Tn28}WgW!`_dpMLEZ*dojb_QDWrP3lCnjaLf)MAnGOj?0zrn^XGE
zw!+Cai6d$JGbj|)N5;7k%RdfE5BGVA-zik|IY`XIEQ<B7v`yC#aw9?J(4J*5&I=oz
zy6TlGp>Bw@aOh)>csb3znQZq|VfSA3x>#>(<)O$=o(DU~1(3YK_8hQck<B4dl=*gC
zZ*y}}whc&dY3me>si^n~Y;{ye1E#f#WxC}-I$A{_w05o^0NfQ#qASTMY$x*trYo~2
z2lI-tGOzY&m}#@ui6?tT1Av<_#LKtv$pf3Oy9)h<@T#X>dmigs9lrn)mA=|hP2JN1
z^ZWosa<2_=Bw!)oTbTS`ZTFm?LctWJ2|Y2{Zj1qfk@+-FdHF(0TyK6$@WLcvknQxr
zOhV~%q%rzj3$eXJu$@rpmY|_=t%?#_qAFAU`3#{o+~7uHZ`q)wfP<Bq1XHTA#?mNC
z4EFKUoY=b;Gbs9hJizQj9H9@Fhg6(ybCXuJ9PB!w)_#_|O*gSF(L5RBef0vxOwhJF
zOg81!FzdE4+{EmPXX&0{F57da*ZI-;P0{=_1ydh&PBNI<8e^Q?$~*4|9L5}zX2Y*K
z-8&LYOUSVsqhYo<b}6ihK_k8Ro`q-2C>`9s-U-rAL*s+%@m*y6qN?HYnWkZ}%Ok^$
zp>2ns{bHknUey;g>YM+b6neB)^B4xdH$HV)wG;7KFID+k6x&2Sb*JlKen@M|BY1jf
zaRTju0;^Kv_e~GZ`w^BLzxih!|G+a|8h^Mlo~xwEZsr_c?BX_;FCC$Bor;JZ&!R9g
z&v=S|o8ymjnYP>(#!zO|)8KQwM_6C<?BRUW2Zf@J$y?p<BBIfZO}>XYpdiL`^q$V&
zt<!^|fz{QG#_WudRiQ5U9S?%Fr!%vS%Arp9;ZC7fK7T8jlpyqw>2FdX-=x^Ui~B@8
zMEK3-`^O2b-6FGYti1-8cf9j?7*2JVqOWN&LL84A@vY}p>$wpx=*_W8YzlAPp4gc6
zgd5M#$Df(l?tk6TYfE2+xTx+{xB!iG;bys_g9tE_@yt^>{i&FO36=JKhEp#;c-0N7
zFrC|?Q+ZnEm`@{Dk{i&rRw~rRhS}nQccT022j110PUdItHJvO=mY?5t<1BBrG@kP4
zDee?MT&dj2Ov2P$ENAqT?Fn@L=y|2?N?pzF+F9_ku}A8`u0!jbijrrvhwFwLt{T+N
z%63-R#>_0TFCO%}sff>^2Rht6uyB!h)=}dPVfWCHgjGRW!L#}<7t2#=e$lP91Fg~%
z7I|SU)X6iN`L>qsflcN`AvYKHt9@SuN16s9z&zC)s-6|PvMRHhNvFc(8`u67r7X;~
z1xGeN?fy4$KArC<1sOfCu#`~<Hq_AR3;SS^Uspll+2l!T-u(2Dlz}`c!-6EKOQFSj
zSM^kB+@{$i-rW8P|ITsn_SFyC<f62))ffw^FAFzpOjeR-T_49d&`lA?Le?DbZS%%b
z{q&de^-ZS5wV7i#Yitzj8T+oRe70A1=zZ`U{o?GR6Z`4GTsQpj$YRQI@AxE=tm{_S
zl{aNI8EmgS^^<$A97r)hU#FdDz}(WZ+lsxF%B^Rc=W_P(qqUudP7G)hCuqZ?sI|y*
zdU`ElwskIYS*aV5j`8@V;y{x#6&J!ki;D4<_(H8`@hjpFYD(H5!tS;pQVSpEn9)Rs
zfs|L=F5ORUjaxbIFLk5~>N^kj8xd)LIMyPw&2tUY)~Z);E=ZAysC6le5t!0LZGtN&
zYbMNYU+_s={kG9gIN=y;wke~-YH4s;qxdvx!vmEF{etR5QLEMvWvp(A7z#0a+MZz7
zl(H7%2Uh;~C=&n=UIrG31z`UL;KRV;1tBt|h#<sCKYx!A&T(KORKYg>V2b8t9M$rC
z+UXCdh+NFHYGb6dx$@!hPZLV#RV~kR*7Rb~IYU(8;fmCo@SP^*YcJGMQ4@VDPJ3Gy
z`C?bS>I)IPu(wn5x{<uv%0c>=YZ=9Vc(7{>6PBvxkhs23q0bo|Z2<DVQ=a22Uf6LH
z&uq^$orjLwEM_tJCz7~-Y88u`-fOS6)JCss-z9sbpZhE4L)~-F$ph``l+UZrTi!fB
zx6A!C?92x-I9tXw$asGweO){A{oaR*R+T-+oEN4lzvnB=mNhzR?tb7mD>$6;dS4a8
z_EY3D7q7dV<@%8iW)@o%2wXCIbjSw_9dXA|^><i@MaPBl2e8rltg+08%#)otCH%sN
zDn3>!C3~a}1!GwQyZ8smEB9-`olVLejroM>GTkRiVz8Rn-1hb;-#b*fvc+d@!XxY6
z8`JfMVAtBr(}3@L$iXt9e0kLzBJ=4`z<I9nYE&7)o>&SZ6}>19H`hR&lduq^6BZ(k
zg>^HKgauno=9kcb2#)n%Q?sr9<eVI^6*Yz<k#D*I|LBG5z|THl-hz!_o#7^mJmy==
zyU1{~jt*H`k1!!4b*&$vT`YJmg_}b@$KEns5*+}woSV!GFTqRT`g8?j@}cw~)GtRi
znKF7)xi5%%v{L8G;O9$g!g+?oCA!b|$_g`iYMF&1o|>qp?coS&)!FTrO7!N!Q>5`w
zDkGuY>-=yt&9$?SLp7SpdoPXgv1$JK()}jKCgWU=82rfRc*89Z*%i!oVZxIZA-ahd
z*30Rjw5%bU!OPJU`!LOZ-=iCXs|E>%O(KgG{RW37&L`?jpz`3>)5ZLZiepP-<GXep
zc#~x32KIe!h4@Q*j&D`oedG8F^u`o>A(qG=$($+pDfZxg=c6ZjY}yLO{blAaNuisM
zaP<c<X{!Dork-_-ab6`BgbvV7Cf|xVnm+eom>-OCcR(^X!tZ|=VkG3oZcSrkxqjZy
z)irG&6QT3j`Q0rq!k>U`-sghg4}(jdQoEzNe`gH5t52C?T%0VNuITCc-Kc%DQG1#q
zerNtux)}znNhFttWL@@g?($Kt-0zZ*wiG|++&J9*UiwHq=riqo-E-5)19l&&K$LWJ
zUt+uL8*;ZkTRv%c5k^|ovE}Z=!SMXwjViWHX?SmBGV)_>_@jK&nn`=Y;@Z8I7RUMz
zhP7wW4lbo8#+bC;jv71Wc<o2ZfUl3q@y2JH8G6ihk!!R&wd&6Mp^lcmTqAf?d5J!9
z+Tf`rZv4W@aEvV#FUd0?P?;L<8u*j=;UdhV$W~6GLiCS)pRbIkJVfWYp=qW2`@3qY
zMIJGC3wDjaAx)8f_ja30Q|}2DL}MJC9TA3S3dbQCr+ZA}9_LmQC5Yn=etZBxG|Nx$
z(DY?M)eSZqv8ihe2c)LVM+p90(8%CGN-p7}KvWtU0M(U8_C;*=cuA#P&%IoL-d&q2
zKD(xPXVVKa1d(AIeA}z5Vnx|%<8zkM#w{b`)!UX-Bz15M13klsoXGw^QEjR5c1)Gf
zsZXmJro^nvA<~}fOgppezcz)oF|;lrzC}FW!X9u8SrG)Khv+6n29d)|Q5k){)ZAOC
zUx;*3xN!OqC3Aj1HON{o?5masRSUnz5Q85g&Hw)sYHgG!*9!f~s)@#68@T-EO5f58
zm>{G!Seyow4*?vT+;1hsEMagd)XZ?$_+<Rqbkp3My%&=|Kl&8HAKzddz~+w>4_hQ?
zYk(bK{-MLGZVm6b;*2UoY=BhJ$yns#_4&~c-#OY6ZSCY+*ys0?`0QG7^X5tIL4y!>
z&Lv~ABqHKg&BS@9-*&#y-D4Hz;dF0*+OE)^U(cGHjH}x}>1{i_VRB$NmqW<qv?iMT
zw)C9ali0WVK8(1<X{*~J^ihNa#LAEw;(Gl@1NOe}95+Tsw8bo<U#H&eP%GcZG;$Hf
zI~_~OHg(MB9-O(&t?WH<XDTl1fvoG)!&2_3H8fiv%#S_DW-Z;$pQ~jwks^HW2&h^l
zD0b4RIT`MxX5QsLG`8A4DM@@ia5J$9ou*X0E>ar8?^$-hBKMg^=(z)$#VaP=4JLDQ
zbX(n)L9=<AG*=v4!)ruwuisK^Y+2%Xe9!T8zS_Na(pkftFYW20??<0fFAXhCG-?;U
zahr2lR35Tb!0D8xF2(=W&fFs@T0gU>J^0ZqU`C>kZx8RR2VCfpVAV*%qnH`P_c?Xf
zxGq&6y&OmqJ^cfAsfBNrOqN}Vxhp76Ez(+^<~O^PtsNC*<art!JZ+n5)g(A!cTnj*
zacA?X(V1))``|8)F4`(U_Y<o%{BHIi;}9wEXl|V{(#I@cf2PpsnR4XNnR6S=aF%ls
zC#u)2EWhsPy>%=Me@*9}Cwj26ZR(b2`eawFxwy10+s!I|`4GS7XY7zyhV(+Cw#U8(
zWyN$(o`_oEMqPQ;{qE_!gIG(C9VRWA=vQow_#clA53}HH(l#-&{^%M(Hik#Q!_=0O
zN5FcHKVs_=>6F49x^DW2@%?Ok3JrPJjkEj0Q-9<eB)&Xd#^${%*)<|N#;w;ecD*87
z+#HPUUu-xxX#c)dl)I^VgI+&%ZDG!Bp4FBs)$HvGZ~9gXoVH%sK<<r;+^3c^-RuA-
zQTla02<dm*d7LHxi>~*8W_y3*$AcPGY9&UD)=Fd6h}LWoBSw%Q2<^2=jM^>QUKAx}
zY%yvicC<!}_UctdQCp;DtF&5FdplIeZSU_x@Av!t{r>0t&mlu{IH%3?d7kIB-dlef
zUgHMyRVt+oUuphpYL1dV{L;J_>yh1w=xMzyVw2r{Z3>_JiQcFpfzGF++}ZD9(%idz
zhX}_zXH@Z5UMr;?mf4}G4Ip?9L~!E3Oeuf|CHXxVX=v1%{<0k8T!@f!?Ogjs5*<)N
z2VL**>yTGEAcp=m?fsG-A9hSAp)Pm&;r~xzvL6tbz-*X$*b(Q@u*57Fg}}yD9f4TZ
zHu#->UGR7O#;vIDv^)p#^0obD(zI>M3RF2k6BQ^TX=5xv(pE{quXH6NWek`jfhD|c
zA)=cralEOR2+|>=t!YE@gl{N(f{gFD)Wwb|eKmK&)sQRrEZj^5hP&6&<c_ennp$LP
zzQO1Im~+#bJnQxkC}*r103k*~baIHE`aI(uoDin#0K}Oy;OulmZG-~v4Tz$N>71`;
z8X6QCm%M4kGX;TA<8SFUD?IDW;8k8VF~NA~q9#BQcMG2S-ivhHY76ypZWju|<OK77
zYBj$(PUtTURc8LBz6EOXwzR6tgs#}%bM2cDy14bZvDMw7wDLm2ylhUF0jT*_ke`%O
zd!F({#1z^2@tU)ooAqT?VlRG1;?A*r(w{*e*6Z9Mkj9$)GrEX>Kze=XNUJicF7S!i
zeX-xz^F}iLEg|9>fvCYpG$-2nJe=ESW}=xX4IqAde&a{LWQ78#c|bAs9K~5HQFTPc
zs*57%-Qu?JI2JbYNP2?68Dm>Pu0IR-s&AcDbDMShgFA?K%Wg^P`!?xA!$qPyvWUHW
zTK@WI1ZNN+Db`@Z?5bv;m`)EB*7v$Tq5F5<4*L<tD4;cV%2&~9(qrYC?4wWUw$!zT
zuvbR|syT3Osub=t^1>KhbZ(YvF=U%4BEL!FCXIWG4k1+sIkFjqqXmywQP(dj&-ZoG
zf3_np2odCAS>i2BzNE;#=J0Ae<$a6b=7u%(ptILipp;sars2$v)hphIWQqq;caN%_
zgTFqJQG(R{O#N=z$IGXeT9$~`{yyL2y(zK_PJMCw(~pu!A)Phpt<)cu?;T&<iR3%G
zlhcRNf$Y`hdJOynBEWwtpYMF~of)tBkzc?4_*rb0gG_3Uzk1luxg#6V_R`OZG%H!X
zW1olBg|yJFZokRKDzHdAmEJ4Ckn52XNSy_fM3OZ$WegoPz5)#^ZF-Z8%D-E=XBb!e
z=)smLSC)9F@Our-yG^4CMYDd9ne(~0dw1OyE;;ZAjIrM;rpe}tQdf;{a($}5ImML#
zo?G`$O)T~<(`~kXa8swuyv{zf%pOvIbYR{rU8IplzN6ygd+yQZ)$gqOI6&83sSm)F
zhf=~r2=SL_05k??%RZ~F+$N4FMRD4T_<ASv6%|N1DgbiLL)abw)ekiYivP2w{1<>Z
z{B3|q%8Nt*V3tpHl?8l$zxr(qup|DfB?IoK5QTv78!@d}M<D=U3$RWiUjNO&QDT<C
zKC1Q~s*az<&;N$`L;d{xZC)eJ0MUuSWCVp957IVg_p;D7hEjqhf4?$t`xUIlTn_al
zJ_J%l5;BIAiEnG74u%0uiVvK)pR}}6bF1VDeB8Cj2pt5e=BM2CyFd6*JR<1glF@F@
zCx|LREF5)x>zfle{O<J7^OkQ<4J1UX(s9D?n4g-oq<V)mbP&zIZ+Nk~>E~Ib$>5$^
z&!`%z_q?=q7}*mlx)zyv`OnrMyJ0D^GfW}W;vTjFtI<NottkXh1f|x-^Ioui79R&|
zf>DBZ?+mJW_z0BDACxw!msp~QSO#A{i`e}7-o0_(+%M8JyJk(P6*E;D_2_A1usW@C
z3)v-F^NkXvB0AdeB2>l9e{8~bm^iAC72B-zpfP{C;VnM%J(j~`_6#zy{>FrPsv*>4
zS~JqRF85I=P`KW2!J%Nzhhg<xU=1npc<eiwn!*H*9DG=A9N>pi_zfRMz7}}r*ye3I
zyYpVXXU{uL!ezRO23Qcq(HCm2zlaCbmODniVA>fj2@bW~JNLwh80+g+T<$#AJT6jK
z+g4_1>|Yjo^LkL@z1mXwsyi1sgDo^bH5BluCdO{rN0(kXY!k`#69=lC&rqk!-5;4Q
zj~<HLx@O0?GpfVLmh-iks#B3=e0iA<5)_~dVAPHkJux+3sV7aRt_dX(l`WB|b{B`o
zl$QR+)|iQ--kU8=dAg14dSVP@=jEmh!8PM8NBPQmt8G}uw9Sr-&qLB{dt{}Bf0wDb
zt3R^6uoW@FM@IvbLbgMXgd{6><Z0)dq&Miz{Mecib=k-rv}&M9_%UdZULB`f$;S!D
z%X&Me>{^eQv!Ul0IE4$Z5VU(3PZ2|G^EBtZz`uvzCE{HNQtSZf{B&DK|C(QpqsvhH
z!y>N<4ReFu&wnu}Z>C0HV3mZCBB%xS&f4~Y%)N&EcBp)jF!jBY9}f4zp<;Ntt2-wo
ztmWLhxYwuh44OT6GRe47e(^b&c;mKmIkllqt{1@$7rfUt?O3a~!%Beh18b;nn)UpM
z?-m8kFIH0wsRk$9h8$gho^q)h`5V`*-4<yl#XEQ4_nCWS?fq+!{8yhL3M`g*z&Z<<
z^Dj5RJkDGrXx1plN$JQI;ffl_8??jX1yIvpC4F#zAejSxdPp$*`|y_?F>AJjnVKq7
zFgKf+Nj(%AaWEABed8}`LAIsT=YUwVs-qYK1T}|{!#&Q6R;`x?^-BxNblrujzveVH
zk+?iPj~5i0rJHl`%}_wj5;_sAg^@My1f4sAoEMW4g0wO|lY`OYdOF_K37vov8<?J}
z$WkHlVVoSgKdfJ@tZGg=$tbI!Z08bl3Zh%H$fbSID04`?@DXE9@5R7q?za@A?rPxg
zb<Xho_{6WQ9Um?C-k7?fraP9K<KXLvg{&|AspqZlc`j4A+Vk$$r7dxj@VP>h^9oN4
z8mI2tayZMZ`x$uke!Re+?CL>LNEG{yLHYQFq*=AP<9eRKX3%UG$Bf=`89Z?&2%?5;
zRsw6A{{b}-H_oU12ttw79yN>(U9hzjuX_O1cYGW7nz?TH9fx)+$~5xcAa?P4xo37=
zBi=pM7$u{uRN~!XgZ&2Su#KUU%VZ6wIUUj+Nf*B8sCfid>Y3Z#rVL|rYzuVMzhywe
zuL3Ax;mvTq&ZMiCYjgDyR^pi<(b=Dv93_{Q<Nb?wYgxBC#Iyvua`5Mt-ystsL4EtQ
zvrj+ua+F_M{_*<kmv?Amy}_T;kC4X9$S0($iIb;4-|W*nJ*rzjq8>4RS+Qxr8ZDKy
zskZR<e&EryNc`n^nqRuX6BtaNnzeO~I!kC93YNIPXDRt}yHSy!?eQ%K%qDN1c!Jjj
ztCd&8;nloPbe$c1|8Q`ALxbOiZS%kp_fi+!y>^}^%+_?>dXrW>OL%%K1O9XgCUJr7
zg!QI60`+s|E=;+e;U}@<W6>IfUwQ;K0qtV$BDDP?JEvbxQ?tZJ;o7!d6OrwY2F?*@
z%p~C=D(DAI+LQ89l!oe>;^ikXY>?1ncj8zi_5w2PlGdMBamlBXRLrZBGp7X4IXhi$
z@%X!Q^kisVKBC_=Z~_D)Xd54;*v_$O+xnn*J1EpgfY4%m+McG#6{s+Ma+MNgLd})q
z%eUoUkQW>j`*M$!QrH%86nyfmtch?!Mn1+a@4SK|ht?H?&D!i|hE4@p1~wWR#&p?B
zADC8wSU<Pjk)Kkm=%F^lM7MhH#R`O~q(Ox8Ze~g?JU)u<JyT|u4|`u8p`1Ju%AHN9
zgJpi@Vx4@ChL~k<mUpsgQA%UMpjx`2u%2`S{N9o30-3!SiEy{JVl4V0$wF^!9%f^~
zNcjGz^$g^X7>!)V!*c&Gz62<(fH50bb^$>*5Am#I{)z)nYoIg%Y%dwp7$^dUMr-~O
zzygi>kS;u^^#cg;U<W>|eE?Pbx1IA7f!3`F<Q)LpFhA(bzYx`d)fI56!8vp(NH@{c
zd^8O3QXMSIWvs-~yR7;jx44UsUH@kCm3L$Haht#1seBjd<YC_#iZ|b^gT@3L%%-61
z*3aIxyZ*sv^!sqWXUZt7Wy#bxKm2=5hadEUZY|`%bO|(XkXE>x%*nnFx$eWyG<RG$
zV+4H4KzX1Sl%@6r6P}T;Irv$3;Vc`}pPBP_hl&*W_w8pBbXNm@o>%I6pY~jR6&;#6
zCepPJ>qIP^Y5lw%7|*_v_cJ8D6FN7Wv--`+h;-J{I@)-QlDw{)HW4YJcNx}daI`qs
zYsJa?E?mT$%^GReiDgf%KVK}d*{kRASX*<iE|IciC=oJz7qW)+vAB5qC%rjm`6IOF
zjaj!s)^+05c27qR8*>Fl6n3=hIHSB<DL8`CGHjK^@hC2vvnifkHKwVfk|5j8_s!92
z!i^)ajy*pSQ@ok$5ohF<p5B<JtlBOB7I9m+mwq>wA$lxJ6^-ap&MlL>e2&3kgf;8Z
zz6Sf&ED}DpSpN}e5}pk7sLoW__E`sZ|8jniIuSL6NX=65$yMHdkE(fjgDrwFjUT!j
z@*=7=cKD+$>-daUqleFnrm*r><%{jT1iWI2&nxy8p0PW0at3q0U8hPP#~eS^(^=Sf
zL`<1?v*Axj3Ywo#x}RX=k2rBID620tntqcfp<X;uYBoYT`TS{!hzx;`!Zs0D{S|ID
zLJ!l`nIxu6#@t{rQr9X|kK9%b<(VLE+3P&m>ZgB`SS<rhTuORKdeYV-oOlB}I2O0;
zZw@J>0I-F~oKl*$6-AxhKOX@qpN-ry<~sU@Ky#m-?SX8jTAWa(G_X4I2mL8|CZ69R
zGg}h%{QA3hACc^+$inOk>hJCfOqL_!YngmMh}W8qTj&NB2i#><R@Ca;l~6FCo~Oyo
zv)@oN_mVAEo9){5<lzcJjT$A-XUwErL|aBHa}Fe+T{BDP7kk&7myfA^QNW(CDa&14
zJFe^#45}ZqJVT3Bn+kV&+2vSH8pfQ-%-q`Lvg33V6sL|rD&7!+@|%)kn>rq)Dn#kj
z_#*sQ2ec`TvkVJc2Lcncr(+m^;LEG52sa)zmM7z?uQCx@FYIyL3Q(J{cu7&CW-}LC
zJ3^>Z{qcf9V4^<gqyM=%U|zC}`KbuDi_S}g0+LKUSBug}6%b3IS9cPqkprRTfi)xV
z@arIA0=N(WKLzdq%=y=~;6Qj;S@5qOMgs^e2|*mr_y>La@b7R%EBcNdj^O+ViY*s~
z9v@W{GocdAk+_2jI2IYH$lO(}SIoX%C=-M=&Hi*?S<p27#!S>*Pe4TN1mDtp#C}_k
zZ^g-K8zCoqnb&~TiAmewRaO`_05eVkS{;lrUr%XB6+!%Cwo2yRidJ=~h0zY)n|$sT
zcj>8;Rc`Br^s^)2s2bxeGY1lgrc-KydKNX!E#FvQ+OT^f3mNg{LQS{rcs=-HaCN+Y
z7f8r_E72-Td>{KYFTiN0X0~Vf6PR{KZ89MWERqP2D;&6<l_8Usktgx2^gye=I+Njb
z6_2QfOPDIEPp2@!BS43~@?w{p?q{5`5I?_c8drfU17*t&U2IU2z;}P^YPCP1`He+h
zDDL@7L))1a+@+MZIAcGqiGR|OlG3~Dun!dV_;65C+zVE6m{uL3O;&0RN>kcQK4W&E
zVz8aDm5|CxbV$1FV58`WtI|>JDRv>Fh{uFW2z*|MVZi?->*K({%4Q-8V-zCJ@Y~;$
z%SR%irC1GMF)HWh07qhPgZXCLzt`Lj0H?ZkeZAm5{lw9~|6va<Vy0L`|1Z%Vg-T^j
ztw)(S7b{8X(%Ik*^JL9oRQ-7ECHBtOk$sOv>jNRz6u4UXCIj<tSOBmPX<wCnL#Ir8
zua!8fuOydDr6jbXBcx>kJ)o}uCV+Ac)W*~N{>PG>ZQ7;;SNmJpBOg>yuNB6!0(0g~
z{#G99`?D~M#68=I<;tj8eaj_D!!tTQ#5wlnD?MB8{P5h;dMe)gCf{*_-hyz?RL6~_
z0(SxHn@;yJ*T;;M>K5Xh;vHU!yaBWj!1pc2am)bcr+U?6uaz(0unF&EuW_Evy<d24
z{-&HjN8_%w+LKutGx%yPYxNqZ^_|-^n-bS6>6Tu0LZ@kfbU!TXnzBPicA-_53U5;_
z!%%};XVE;+M@urJUe~$ac;uG$hoGwQsBp>BZlBC2$bjHB3ogAnvQ~`wM30msPjg$v
z2<?PAG5H3pt+a(lMuS*B&AUk)e{I|1LTK=pXtFhz*jVAwDc1684lEt<mMWKFz?9Ra
zIIZi!{pC>;da+cDR(U4$s&93pMChr_*5xh~SFU#Gc!Y{^6n&#@N^PK0-&y!S75s}a
zI+S*rNK|nFl|a6%h8yOWtsbCdPHW+vcWlYbg9<rNZ+8L;&|fh$zX;bumi7O7sB;s{
z0{Zb?M*(Yq7D9s*|G#w{UdK`dI9CAFr_QluJ^<g7VjFF>#54JsFP-n3^s|>cjMF<y
zngxM;u43?WlCebdnYL0XkZu(}*7RgnJl1lwWelPEe93TFs<JS&`YoF=^}tsM-!tOm
z(3^<(2Q*ncn)!FL8&91T`)Nn_AvcMR2%bR<Cw{IY!#h+y9|Oay&N}faNxp@GHs3kp
zE=RK36+_;jgVWU9ke!!|&|-5T=&IRmF3c;+Hsa6c-yCy`Hov@9mK-&6s0<5I2#n|k
zYFdO+G$}XErO(Rqm2c=wvlMUK(HdyJI3}O=(OBx*N@`G@QOd<RwS8;dHRsV!iqZ(8
z5U3vKSE}E^pXJeZLX$4vCJ?R=tmfoX)NED;^-Yq^i+3vCH`q8S!%!nSkKOf^@m(YL
zw<gk)hQpNInS?-}3$4R)9V?rSM>DB)Cu>U5R<&(~$i@`uOsBtOh;EqE_16402q@YD
zjp7R>o|(XU`|zAgt@XQvbdJEw$xfM{FO+e5RCD&FeMy<dhg1K6Dnq_Hr>&J_J4|n;
z){H_~-n!P87G9ly=&)@L8GCf>=nwe!BTozZsFouSzn}At1;2aq^;)BT+SSa6tq(cY
zg^2wrSpAah2k4p4_dr8#?pOC}foJ(G=3=C!ImK1cxdkU{7js2#gHI0>(E)OA$;hY)
z|1y1_zbe35N)F)p8)Ne(<+C`&<TdA%VBCxrSqq)rxq0iQE?q?XrIOu?c?n6KDW}(U
zE|C}!kKyMO9}1}T@4u9AESzN95CKwE)`kdlTAN>JK$U7vmO^lWCc91wP8xz&047L4
zjuv|6w;Ln|6S`y|p+ydtO0Gm>?|R7xKr&xV&%$r#5IH>;y=S2)AAX?2R86y|0vH50
zhaHg91YQedV?MX#R;tgAbAA}%CVQc_vU9*m-KupvUK~|#I8r{R2DNDn;~xEc$2&M$
zKe}${NDhsd(!b>b_B-3)1AZjEohFz6c3y19b%X_&0#Ix%uuH-%W1w|_nQOGI>0CKU
zoa#XjlJWiNU{YaRsd=EkE}tWFjy=%5t<jQ;8c5b>*5tz@&agf=zduW}kzmxL^Vkeo
z?d7DW0?5iB3?UL{{AT8Vo-sf}bT}Uaav{sO5RmwU3t59%GCgDOXgtcCy^Z(wY4{~j
zIv|G+>l`5Gf%}762Ab>t^-w;l4vItySbJ?54>J%zPp0r6(D%R?Uixn%=UYO$79kG(
zId+}>ul9B==`Gwp)>X>oDV?1lxmrZO!_m5KaDp*mJCJRRR>JnQi!LX4@>Hv`FM@lM
z*g-g&XaCs}U-U-Xfsc<DKh6PMQ8325%0<mlYIGF$D=%kA4Bv|j8Ja%K`5*evI3}*w
z(}<%M^v;XHwbQuPGnu?Y?E`taoU-4ODV`COKYB}=C{roNMz~a*+qlSb@SlbG;j($A
zXx$qpb>_A#3%_AxEhDR>)Vy|FE;vCa4T{EHftI^j%H<v|t<<Fun#nl?;p7TBhQGT+
z<rrP3h-g$}rzl&99@8oXjM@WeGAT&c`4obrx9h#32ij|Kmhxw~3y$scFIAhbtnpH<
zuF6}KP7<|-CSck=mRSzI`}EQaXQEr&nKNp9{HRP}0iA-_9ej;=YF<fFd4s(P(i0`S
zN7E3nE|x2>DDxz|UtVMC6;$D!`s;%mU#S!rRuRn-?J?L*Y&yKlHVS;L@cdnsbAUzh
zv(wMle>YwqyR}N&<;cYlE0ZeKT&A(n;$IzV7J>G($f{owDELjm0f=bOYZaA}!~dp4
z0I1Yh<ZSMs(npOum98{iPO)PGSWfE}c#m-C1isC?)sl`~kcdD*aS~ft8$ZFD1`+eP
zIg`l6+$Gh9FUsNgzz8K|Si|TXDHr-0;F)U?gBCxO3s!@msdtq7G$^mm3Q|Z&V<^<F
zQvLiDAfgf1*+r87%5A+{FA(;!{Kt+&<b%{LbfzZ@?MAuyVvuQn!>Y<QJr@e$M7_1X
z0r9%N8zh|;U;0E|Sj5?#^0KkFGe+E8(SHmO;<44`QP|)Unw%0<oLrUATmi0fe%Wl`
z#(*b=7?}O6gfvUJ2XjA)==<mdy9LU-()zkqbgYSgJmGjy^K8@}Nk<`)G$Z)=thX_@
zbrD0u$DQI6RmD5oE#8EUZh?gyG2X29@%Z5Xnywx0HaCjRcx^mUAm7$BHTIn~qwS|x
z+;`^86rMR=R1j`!)>baX+#r*&2vi{)T#9U?v~he!q!RE}?GBL_j-2i_*`q6!I+@Xu
zGARxbS}Kl$_1OoDDo}2}kg9`w;E50=0LWd)Q-4_{#2t5qiczjq+K#P0nq5<pbTB6W
zVut_SUH?tsI*4%imB0mz(TFi%2rG$m(4?dPTb#fZjQ1d=t0qBPNk#a;JLhB7q|f-U
zh%Z0!^gOQ;m0MDQwvf+c#y6XAZ4)+gLArH18j|*oD1H@j?r)Ezm6tUy(U4FXk{z(N
zKIzUh7nJ_aL{V)h+77aj7KDUS0GkET<^}J`fL#<DR~ZuzrT(s^5soN9ywWDLs-!N{
zsSZsL;sUuouo>*bWDc9Swm%$~=`4Fh7iBYiu;Uyp*~Z`Y>IjK&A?<}<?qgY>r>$;7
zi9LHd&5C^Y%QymcD0YFQX$9HV4BL=<*9%an1<EU5y(!}6X?4TJlFtxLJa|-f;}qpt
z$5%n8wwfmumgUz7u9xL1*Qmamz0F@U=X~xokwg?aG8Ck6Zxm|YWB1vSvQ7Ki1;`0R
z(kp|@N=28vY=f13=hG%j_w{N-mtn*i#rhJV6k7L%2}=4sx!UG4YB@y=L@_s^)vx(%
zSE~67S{cM0G9#UNaRaUBkRnao{Ua7LWVT9naDi#OpRHBNW{kfG@d~%*RLKlO04-#*
z^7akX(IV?YcO8DPE)<_DhUlegjMgG}Fu*ldmlvI>bY>=uy>ph~!qzhcUvD%>R0x2z
z)V_Hbhr45jTI#G&%oMaBT4g{6vWd664IenwWId!%G>72B{i~yMH2p4Z-eV4Or>-k!
zzH?>jB&ld`mQh~mb7`+(@dNwaGEP47z0Jv?qxtQ<^X|(Ty6kyMw(tJ|#ohY}@yT#A
zD7d7M^CjF_FjWsSaq69<c_8Y`F-psClu~});{;}^9`VYsIpLrbyR(l_{TqvA7!w#`
zh;-_^-|(h8J8p}`;e6>kXsy$9b(|Q~?>D?x=`~`3YpZNU;K!m{@{os9czdI3=2DrL
z1v=rh@7AyAQb|2ETbAK=-|^q(K3_g|bLkOX_u^^#Ipbah7deZ5ooj9}(R9MVpjGLU
zZt*qeW4%W+;fDDaP@`oRW!j%2ds@eM`W-+`*_)AZqH(+gB6GxUUg{<ZRIlyqV(4Nx
znI@=X`ik`GjMC$2R`A~(W|vi}k6bEf82x&m(XLGfXOxp#baP3a{&srciubn2x!D%8
z6$pY~*z!T+BU@(sO;SWfy<D!CcW{GTcjwm*nWJ0I{1VzOW*;AaScA5SePevXjr3pB
zjjEG9+*aWT5TKI>h~&d670v~A%%yXkY#go7Tr9khFpvy0gyMf;+P^{w{=3@(o%2vN
z^@~|Pc*w@H4rt%`btI~M@MVD057>7JK<wVHcTE5*FV_Xx7PKHRB2UXH(QK|j3CAJk
z;Y%sw#ob&kljoD;czY5nSD*koc$5l2g@LCKpoGhfZ=QdwTML78cs_TM`vitJ_=$zc
zy&AtX&>E^!#&#LgGvmoWoF87(FPSBQV5?|(8jFeLx$eX2{aH?783^wqTD(4N08)A4
zbmG}F8%qVL8L@~`9S#v#J}f~nj|&}nmoZc1C~kP15sptU@Iopi2DzuzeB+IXIpSjv
z`^s<blpai=|2SWf{C-F&n5V6Ag)@Uld1!pm*6ERVX}6y!N}TSObIieEx8NSUr^TNb
z$upf0#BXwWwAuYwV|V?7epqLhclBEbFeJZmM93bo(NmfwxH3qw=^FTpIGy<>(vAb-
zo6T_wQX3=#YH14g;6JinYTy-A5gPhgiyAUXU@$Yq=d5ha$b5I`?!E)rD!eVMQr{{a
zK61Q^mVcK@;gBL?5X?X)Y~zjLx{botWqhT+^{ErE1drKGsuz$5#tSXbIVpy?*C5_Y
z<4Ta?LEHql90^H8kI?0X&e(`ytS+k$fn03m0$}Q8QSYVNbXzOO%*wg=ltzOhE>+7^
z>t)^PYYMVd6F)`w8DF)5T#D+cCE;IdLD3ov*f1GS02+<W6n?X;_YWb_Z^za<bWRo_
zvz6c8xKgw$>#a>;mG@p4DUU5o-!qEht5{=m3lDm%G=Spwpj|VETElan{&6$kMXheH
zWv}N&r@Gkj3Sx$y8&8uzL0F#IznRPzH=pv=M}d{$b0*?j%DZpKjkY1|St|2G^yw19
zA96K8>+lyQ;~LLS#1Zaain~6pH!L}9w3YgPyMS~2)Y8-3)xL8kkvyT#SoynJaWf-Q
z1J~;8-vNRb+r8Sws!a~Vn_t>&UiEAb&N-l5z;=r}_FBE-I6XwUkjMDKTt2a3^i&@a
zHRJ-pwVl-prH)rQ$?mUON;C_tZaJi_J8-G^=?>}pJ0HcD1UYInl<ZT7N!n-7E|DH{
z0cNNi9Z9-(Adbw%A220`x7x<TgW-cQ4dox*%~#TShMUK3oq3LzKcjq>7+xpqwBNXK
zUI&DpX1-p8C<@d!QaMW`8G%byof`->afEwp#F!D9IP{Z#z3>maSUD3rY31}^bdzZy
z8#g{$;UGirU*-vjG<4wb@O3iADhYt!o3icL*yaMz^x><&Q0jl*cqokgRm#BYa5hOA
z08~NaoC4?^Q05YdPz5-(fT{YwZC@g4g#}9zK@Vx@UHjcp1pPTM=k4_0PdD2ty{||)
zQ|06cN2|E<0V9})we=C@q;b$YcU*3ZpujsF*l;?iad%w{0G!@Aj3MUP6Sz%mx;ofm
z0Oz!(g;GKpA%8?|jo{vG9e5nK>0jw1AAweDxfuLMP9dP<^OKOpb$VyYvAK;FdM(t3
zoD>0a*_5^&c}||I>m^9W7a47D+$oJ1le+o>^ZS|kh0fg=X}%6~H;!7Q0;#r_9<$00
zKjH&;P_Uc0>vgyMWbD?JT_;nARn@~?n$NiQ)>$t*P>dtvGV83)+`udi6^Ya1gtJMr
zMr2YsWu!*pmbY+F8ME%`ca!F{Qt<D}n@t{FXou3@z|O)}yf2%vY$=dJiy-!X9sIVf
zQp-96kw0ApCoamf6*#l`FGdV7s>Sh}EfbA-?|0DN{GM4TX>-$lXV;=L34{e?B5guc
zJS7Fu12|zHYe;YVmB-D#JeKp_^xMqmls(#*7)O|q`Kcuf;HT0L0r_K2rvSdy1aJo?
z?$~_J0Y38{4}iOKvqpsPCtpdaUweJ_+$UwPNsTGWRa@5=EOMf{dBPWorrF+@%}!L2
zzMCjq8$+Os&zzfH<feqcQsSWz&;D5Oxgbp*=e-tmA|N2r(xB#BrP5J4U#hlqtZ#qI
z=3^4v{_I8KjmP<}+BI<Ful1ocugp+l?eN%8t+%#QDpD0K_xIoPC)!GGMQ}KWt;~0g
za_A7<i&M{&BM2NTZ9d@&<d<I%Tax>wh;-t39t`4zr>vyQ4PKpTuv2@pn*IoK8!6U1
zXr=zy-)NF3Oq@>DcJiHLD+Pvg7+|bEK&SZ*IiziEV=jP8(uGn~J8rjC9WS#o*)OYt
zct3xj04PW@2YWpgVP&+R-yQg3AFSUpkxBPFXZ{UG`BZN^7N2+8E>PikVOyyrg^ozz
z1B`<ZNDP+T+mPPsW>BGtJUV2qkbn!JC^*Nmy>osMs%#CPj38#-itDNikQFO@M3d*E
z6Ehdt|C-m3fS7WMuTy21Y+iORrs9g^hioL^h&2LVrAU(Ur&D}Dy61Ytj=RHq=noZQ
z@NkWyG-04q3#iws_r^!>QUZ_2>LS)wut-28&^ZX~i-%k55&u#kl?J2nfI6Ha<@j1s
zw47hdd<Q_N{)<Zkj}G_UU((6WUjsM*E4kAVBqY+_I&<$=`#lKR<NtpO*J){R`Vzx;
zXNj`tUQ&C$FZ?SY{FS<4)0-hvSAy^4f)SK>0B77$pkqMwqkH5h!uy*asxKOG+9(OL
zRdN#W19CaH{)l5*y++uZ8xLeotE2|jX*&6?-(lQ)pm5*xL}A>#3Gq~0o>(|2+7L$-
zO!GeWtj`^6^Eo`=70|G8!PRi8Ec_J5sj50|9r}f@bFsuLkXwA$eG_nvZo-5Y=3+G8
zDOvJp^>H&8sfK%C$PZh81lkuauk{L%N~%?V9>?k{H*6TPP0d;;CbD7u+g0(jhH+sn
zZFDFMkb#WYH`3M4sf`PG;f<0*;Qj2GEORj9XjiqwuE$1qsge#Am)!P-$F9&6S=3}a
z%!~KUU(P9U{uGRJ)VXnIvzWcn!>VTkQ7bfNW;`9_s0r|8{#;AMrK#+_f{jR&!LiSP
zUUI+XN2$ldtKYox5fR)PWv`z(kEx`7wAyrhc`S<%?o2J8hYP>-KHqtk_*>_nbQf8}
z^&d6%_;WL-=;~(YVq|;rS!J*O+;4bcGHIQ!w@viRt*py-k{WD!#HzSknDji~hJpjF
z5El39I0|0%d8By-inpKR;<TiA;_iiw-eEYU7<LR3$*O<k-yipZ-3yW4Ul_KhnWbh)
zuV6)~JU-nJ>*qU1DT7m`+58Td4SE~e?#j?9SBS`-UaCA|I$4=t&l9(;lzBCrGgNu4
z+mb_$7HFg3Dylhz)W2LGVPO`ItJSOKH6Mw?iKSn<j^VBB3kodNF6SSDZMUNqn8AdE
z*+?r&Gmq`8e*)jEKR_$N0*xhrfsMTwaeQNr!U)M)&z+fD(h%ee$E=KgSaaHwnA{eD
zcyl}A&p~svU-ju-m2<C5@zvlARBEF77(MdO?cvCGmsq&1GjRJoAM0v2sFE>-yLOxu
zc2etAA7?XV^RsN8?d7Jt$M3L<R*=l9XdS+8Wlr7Ztk;ft(O#@(>o}ncLO3n&R8XtI
zY<{X-L#pWpw~j(cC4e4pNRga26HX;>$Wa{`ZCI`Fk$PUFB68wFdON2L^z?H1dbiv^
zAl$h^hBh?$2AzXW^wkFjVt!C+0L5TdJj=A4#HC-}ISNd_hYPGyJspU;d;~6$O`?Lq
zLo{w9LY}(jY$lnNAcpvVixdEs54_R;8t4x2sKY9q^*Ml|eif;t39#=1tMI?(gPH^_
zrK2rq9`(1_$;Hjw#Oxc)C*0Q}_AlFO=UM55v-hbvv?zjktbxN1bUFcbWa2G6Xki^8
zblZeQwJYQlQ<Gi!KC%|oJkM7h*m#i6q?E{2dwx;%aRA>Bl>&eDp^|dS>@ia};;?Er
zB`$`31gY9oTOX&u!*u4Ywj|=mtCiq_&T7^#OhPSU5v;X{$)S?AZQ{%Fc$e7iAb-^I
z;+=cu0r?a`&Cb+PxyGePZ0F0Ks$-&7`8}+?WVf$v7&$V{T1D<b_Phd%t5GjtiQmI)
zOrzBvGEMkvHxn8WcT_AK?ZdFrDYf9DpNpvN3oyP{E`Xm8wyIWO^WIvz<M$89)v`}F
zJ>7W6Nsh_h!aGrvidukRvhft02#`PREnMKB6}WB7Y^$oFQV&^kAaw(X_qV2>+@uQ*
zdly5@Ry;6W3F6FY0v{RLA`fiVeg{`&kJ(u^lX?#0(P|HdM?YprGxwf9#f}ZZRv50N
zO>r*BF2tQ|3zGF;&PK<4J9g_%gOMdfdm(&feigmycE)v>SQ{4UI=W2|7C0T&6gd#t
zPh}v?#NlCqt9C=j1puk$aXw8}eR7iiG)MnK=aNadY+g~r32u_<$tCA)ro4iq@#9F;
z$*usSl(6o6#?WKbs<yk?P_BQzjyH6m_T+S8*2!;J&u`8iCY|4~!IzNc0|Q?}UI%`-
zHJg!e-|W^Z;##HMV<2zj`owA43BQ8RwDCk)b1L=okZqURUOn(+eas`A)J6y#c?R%W
zTGCmb%V0K)J>bxRacdeB_js>r4@F%UmVAqPqs07TYEbdxJ~$HRT=Jc_AuyBHE443!
zFOP1?DEn%Jw%zoQjXdMNIXbGtb6?>DI`Xk&tETEh-O#<D7<1oGT%<Y82`6%vhPg_r
zJPn&4d6`~Yt(s)kzfBms4ji>s+Nu7#bwPQ;!}oPaVUr#w<roG=k^GTYBy|X6RW5|B
z9E_%FMFo?t07;F|?xWP<K>V!|+^18TBvxsb=~Yih%cpePn*gxHl_2kR!)7UWm_jT+
z3-Guk-~VOiy_jZqbp0z52m&f)uk2J^!>SKvXtfPli_3g^i@wPP-uC9zt2#1D8~Mf5
z9@4c3?Cigw6X5wiTza{>@gWC}1fxK4R^2p5AP|2897O;yF9*mLP{6(-Xa&ADT@K3N
z^akUH*;9`a@}E&RIi9wMZ{gLF#i@0Gh2dl^01O}1W$#k075(d-{HxG{k{S35=1i>t
zCuX$45Bk2rjMu=iiAxQf-~ojE@?Hdsg@S?eR>p}>vSPzxzltR&Q!ISRJ{Nxk`L<*(
zHj(6wtD?l|lDaXl&nGk+^3_DcEmSnq%{BRbQVvAY+fU9#y);>Ke!u}Ay!Z0havwpi
zXrQ54Vd_susU96C;OGS4{1I-NToQ_-w~9YrxA)iu=|AG-K{%{mjhBrBhI6IT!@sw5
z<*|~_oFR_9P~_U4+<9Z?0$5oH)L1h8akhE&{>VQd-o%cgzLLIS1Kzf}q5BTu{q)K~
zic53*>49Ft@2s<|G!p8PwX@Vb&ppAf<|v{YDhb#0<F%<z3d9)Pj2*VSLAoZFTi}#w
zd-$(eAc6+X%~pkzs7U!0r3mFTu#nxz5Qkhig_JuAoH7IBsfcsjwiZo9aXU`?Cl*y-
z+*5&8$u)$!Qq4%9F-)G@UCNaqjAEbe^)Hk+ooB}t1n`q}Lza=YT}recJ6scj!`GEt
z=o7wA>oUB2&a0WN@#ULJm_`;gofY{iuM2DFbb%(3D+}Ir<*zs^uEZ#~r^@R!`rPyN
zns83dJFPgQ3u}E4LC<9Bcq?R_l8n%6qMw=~C0oH()?1cuPdEd19F?Fv@}-xj9X*~c
z_KOb@;<Z#y<pO6)0sps~JrS9Kmb9~DGKDUX8)GsBvGNDGW*=_SP8R`|>##J(B2CBe
zg)5nk$)?2`Z*<L!l-3miTgFYEn>0fUB|S~5p~d|InrXI@lE$6W3QqWp*qEEFt_<1o
z<`-?J=+TvcJ!1C@u45ssV#VVYvB)asHdrl<_~!Aw^0mE|jC8A3w=<J>Q4Ozjgg6kA
z!UHdW52LK<%w`8;Q8-4nf_JZ}UELg^fT@UW%EQfwoD8RUC+V}&CQX@X5e(Es_AV*a
z{K14nx}3Mti;A^rIY+>FWKfhVViAtLfpyq3oNvJAL)DkNq+F1}%eCUo)IIGcH?G`h
z`Z)I?{>uU{A;lP`e?z&_1z$Ke*0{d>k3ixP{Ybbme}g|66ZX5ZZQvxuQ_p|Bar7GW
zFM7V)XqmZC77$o*g|{%8{d^kh>Ub-biO>57Bm(W@e-q(zT-ya&F;ZY4T4Wr*S<!=Q
zTp1g9g-CW3EGIcc{_jAc!{iZQL_VB_4;n5o3m@#TKv4Q{{sEwD4MP5>s1K)RpJ<(o
zb$;Yn^0MhO&F_Oa%*ymjf#(!$JP(rORV7{kJ)(FT!4{A1)VAuq|GGKv{?~ICZzMV_
zqOSV%(rw@|>Yt^$?TMxQqVcm#K0F2mTvzm2{*2g{k6u{B8WC$4A`AWdI&s4S>e?zf
z72o1{=kqd(+9c{vtm0Y2x{v~r3CWtAZVqyoZ#D001N}YRi?z}78HF*Dri=E}-7VXV
zUPJ`_W_Lq*R^uB6#xuyea&O3?DSJ>;+jz492oY)z&2{L1Hs?8BmP1_P;2^z1m{`Ll
z$W1FgFS*(^Kl088ru$Ep9_LzDaf)hr*^5HD;aipk69V0D=*{IV4x=w^IfNM7<7fJg
znk}V$<&TJnl)pwWR)8suDUYfzXDu|o%;2$8cC(0jM9Fj%1VTvp&uJQu0x6m(C2gz*
zFjfGenf&4iPzu^yA=r|=Q46nZ7!goSJ$kCT>8^YsN!Z>xZurNLy03w7tg1O;sf{)c
zaIx|XFIK(#Rj1y<F>AT{*l1;!6Y%>N*!3?TJ1)d*do*!0CBwQ0{HR&5KenN@H*-#?
z;7TyIYP3%7?X8b`I;<;KUV8k_z&Jjax~vnhL37sn$m(@#7oq|un^;XC5en%&o23Sx
zP&SFT5kv`ZH!gAKdT+9Y`;ni}zz&sXI=6LZaIkE-TCU1Azf=NOLzmc5TWv%D@6m4#
z2^1r1b1V&-cPzE&cjBa#EFySNi_<;LC*KYxtQjqFb#dg4aysO6MCv-_F`7+;*9%fD
zH@kuGP(2$WUZtwb8%E_w(RIoU1HZ=(>q5gebA{EkkftkMVo!%$SXXXdVqD@Me-SD0
z;jIi~>58mouxs{aXJebt%9~mPz~|B}<<O<5W)Z9W@U)ZqdSc{9-pL>|fAy(IxeL{4
zj+C2_(wPVwr)OcD(GK_%@<zJN^^zs5!pxnKv2MAB(b=}V4)xQf$v?I;>me;Bg?I=4
zlZEz3Jk<xL{uj}8E6*^R8a+}cOA4Q`<&j4W>e++8jN$;2$Bv%bXpS1?L?7du^-&)V
zGr#0|?)4)br+rx6&^MK;O(|&BKms@t2&6;WzC(F4BaXw5C|ELVOu67dghZWeq&etF
zBKj#D^c1c}0X0Tm$$b$m&+xx+Q7IH!XWLT?TYpI#Ig+^L?DpQR!^uLxkl7GEGU59e
zIhgE}9^te{^D27Iq}t`+9fEkn6i=kleGEba;J<l)Iy-H%FBLF7&j^cWHP-pS7h(vS
zPN<sy%-@Gl^??ZTAOQSuBoDf_ec*H015TLJ06?#U9u9;l2mTUc0aLKv@-nrfn;@Q9
z367&|yj|TSmu6e(!D?VBXT*bLI=um7EpX6Ck|o%44kXWB0Y+;luvPK5<Ca#CC!Wf0
z>RgCp`J7%b-yE?zq0FM&c0+kF4w~kNBJB=Uhd(C6z~OI$$UTMi6%{#-c;nLJT^W-l
z0?#j^FXomf^nHQY;&ej1K6rul3@Ma$zZ4+SkzAr&WW))ISfWoYsq=8BG=-fiRp++b
zyE`LDkh?^7z`enwJzp|X(F13@xSiOFZTX0&XD$@<acW59p#vAqocQ)&1zk}6c=#VZ
zH|h8me!-eDZ6G2Gw0T<bW2it?%~PYdoWWKz_($z0mE(rqVkPB6OoJ%%r&hjjHNkD0
z31Ve%$lTJfu%?~+5$kU^S9HckdLs9ll3?zc@9c(^((ZX*52yU8a;nJOW`)+49{#g)
zol?_)Y=pBp`PZ2MTbM^KuLthx5LJ>8gWZ%)S+p1fJmLHW!QjZ7@M@(PY~*o_se=Qw
zSa!i)(Vl2n_=z?yRTFCi#OZSr^=|sAKiC4osW4%8){-i^GE!28#DOTtaGS1*fq|4p
z!AKWTSo)V^&d>USGfPIYT{mm*c~)&wgvuYDk}HvXHKYl}!NDRs7c6gH-?<exS@#gA
zryjYE`msFoh?PabucZ(P1qv@LbB#Cnl^hubh>UILl=u)U&M>TR8|V4Bx(`nK^$~B_
z*(!URF%M+U?K+PZ-FUiq=2O$OdrC>6c+1ne5K0>#n|)7#IS16n%zvEw+PFP%Ff9^_
zmNj%fDeHv+BDJxbB?q2#I3gho@f<sRu3U*PT90U4dIfSMHSO)$JH~6)+Rlt($)o2A
zXE+<nyYzaGn_s@q??KprILDY-HqMC-utHvbdr%om1?ab|ahojs%Z1%w-if<RYEH32
z{*JTP8_P=#5)tm~E%dP@4w*h>MG=(sCM$!DvT;R%qc6JYo)3E~MYDCUEvF%K1nv9y
zX>Ze23+7MZ<cE4K_n!DUcC~VlZG!Yen`C@5WF=bD0J7o5sQf#udtVSo*8NWvNosjQ
zz<^CQyYkTS$qkaXj6O5NSYT4GnFmOt&O2W4!T?Y8PYzJ#587&vykRFviifVfaIn*V
zVl~G5AQe}XAk_y9)OB}&L`g3)ibv0~8T7&-b)cM6^BCI47#zwFuU2$=_q&N(9<f-}
z3fp9)?5<{MieS}e{Ta(hxp31X@#Z@=#~oH)ad=Oue%4>HZqwy5h2>8;8@0$1*Y%mU
zKR?|JzxQM7v8MDkUP;U3{B6YU+J3{!HW&WEsCv~wXpE4zLFQ$;<OQ7T7q!AqMCuz@
zYw$8rXd<(se)hJ?CL=Rnqaaups=4KaI-FGgbE9^k{`fV!d}kgCG!HKx?2dtKfOZCW
z?GTLvJ8)NS-%ac(NtSS4!`Y2WcVrt8F;yL$M`Y@yn0lLuM@fZUCG!xIOL&lRS-hir
zsIJr;yF6+BPO_yq<zmHshN>WH(L2)2dREtLn`m*#g!ch&)h_cJv#`{3mSha1=75`M
zX&ikZpPlYSxh44zw||QpR3Im4*;_9uj}#G?&D3_^=ahUvyWY7mQ9N<!ZlALI5yFgH
z(ZR`E4a6sMk`YxOh_yYHk5nTf>>XX`av#=Lu7AYQa|v6ydBeEBRcvn+u&|f?YV%ls
zZoKB5(rJm&@*bv@krcmPX##a;j-LDk>Gn^sPRK44_OUd*73E#2B5&LT?L5$B>o&I;
zWk+ZXO}^b^d|J3HJ?LzUTIS#Y<NOLwzg!f%IRALeHm_|TP<^S+%vL=PmFw<jGX?^;
z3!;TSKwrK*Vc^4R1t>W2<_UgQNrs?w6HlXG`7Qu$V&x-!XN;3!c_qTb+Kvq^{uACQ
zLk5tn$6($tFkcLO2*yS_GYvSml@hA3>$wLB-7xsjNj_|%EhiA!d~i?;m=|NpMi_OL
z9vsykLY@UZvDWZ)I8nm9ydj^xXR1UEP@s6_@RN<j)=oX1h`KLsP#y7mpWFNw>+H=x
zh8xGuva68G<j;6}Um`jc=omyRiDD9KF1nENNLqYQ|8OVG7D8DA3S>p9IJI9NY37>#
zf%wK_dQn8dr?P-)7im3gwzlTW3fY?~S!P>LQ2JDtX9OH?B-1(T!lD>kxH|cz;6)^R
zFn>6^mZiS8cyIk6-0*>&Ft4hZ*aRM<%<ReDNFFFx=LFpPzjgA4^|04Hh_LTW(q~54
z<g;2exrP$kzc%3q%Z811T;JMuX{^Kyt4#kP^XGdt+vA<vZ_;uc>om0MIB^$_uRgo_
zvShx&qcT2N4G{D<7Nl|Pc>Jubb@zDrTcYDB{|g4AMBj?zx}#3s<~A}p@`Vm@(Ci?J
zqEE+6OA5W=$kj*mz)H?boW7Ch)so_gVx8jCIzuk&21;+7$bdjoB|xs+bxEa!{S)NE
zuDnlPISW|tEm-923~WTTB~2^X&dv!VF~!wDb|dtCG*0IR(m1(4H+ty;r|P+ZEBIxL
z)~;*f&uHxyV5G&3vJ+2aW~Wy5-Flpy$b5o~lVUN8#y3~b6?*1aQ(?-p(<iGkr#awN
zJT&R887Y#vFdufaRNlr%0r*g2nEhP3lH1Huj0J%H=}X<hH!1>tICRvVsm>b@>N=}V
zb8nd^o6fdvAk^*SjFmhMHcRk@sld7O2*85Ozh&<h{(x$buaGd8cElHo>{)%oKt!GZ
z@{ZIVMmg=#<JD|iA=xfZY3FeK9?MD5g&ZpW_576AcCQR@Vs!9-0s2Gu{XfVa$8U5X
z#Lz5&e&C@5;to+jF8Tk`L;#!iNA`{La`1h`u<9z>s#;PBG0;<m)ORdQ0N@J=Wmxbu
zXW6mHX#JYoc6FVtG%1T);ar?H5{}Wqgfq(ylN&{V)ufbPYy!do#?G<<oQK+(578a%
z1lPac4e$nq>1h||c1Qu#si()+Ox=sO@bkHLR^+7Pbo{i1TpAoAtRzrVRSi)5voVc6
z(=KXFs}gT{Z}c}Os@~?k6sGAibF9Z45}Y3a{;=*0{yE&%79dam{n5gwmXGd_d|(&4
zjPI06{%K89xs*$G;@lDMYTW=l%VPUA9i^yM%m>nymdEVCFJ9{H!~L!m(z~J8l)CQg
z(E7%^yTyA^e7(syqaZ+v6$5`b|4R2HbTs{Px6uS4ghyZPoFF+{d8=j2T-x~hn<tv^
z!Q^;W8=(1~1|n}BZPvkD)BaA%LsCGVV3y%DhqiTJpr>o5)3yr5e?|<r_>eet9r~le
zd_V}YDR7)|ikikjlHbrtVUnks8gOb?3Gh0bMYxOfkXnk%gkHa2Yk}v8vvH`>Wc6K=
zw;&J^S+LcX1v;wX8vgX|6rZ{($F23v_NGtD1{6mf7g(=EJN%;HLr9TdAQiCbyNhPz
zdL-%3SHu0N(<9~nHSj>Tdq=4nuNY$Rb5j&d@M=Sk2Sxk>`nQ~u-ZbKnoUcS<Oo~UP
za%pMt@7D!`9o+<?C#{;2-I-|ek2`U*l5IQqi2QMl)f{}#1#e1PZ`#DX#tNks<ADq#
z{27RdNO`!+mD8zn&Y`C=76@HXRk>ODwDW#^-1Pc6&yQuS8`?3o1?uq9KEjT6+-$f>
zgcCU@u2j5oPrjEZ&ma~qoTday6o*wya#!nY$q3y1sTeL5PGVv1nq7RNPrb;gga;&i
zk}6YdxnbdM9|(!rHqbFlZ_6of?fRO`kWE@AV-YS(1nVFor<oiOskcJ>!JjOI=1W{o
zsS^usP(HqDHmGjnpc{`;M3iznO=|)l(eU1Fl9rf1Vj))t7tF=M$zC6n0jE{vr-!Uk
z`QjV)(uA}tG-Q{F@#CgIuOZZ^+A?J(v1<gr&9|JiA(TL@__Vw%vCg=kiqlce`KCWV
zIZutzDq1s`u{ALV$@7MQ<XdsonY+PI+rTp2rhPH8@wu2+vl&_Py#IilV$+=!PSd=n
zv_+mk#1Pg)ZVLMiP~=hx|I<kO)uqJam>oMc5>j3bJsAj`)OgD%3#94mYe*@C@iRY~
zC**2M-59dYm-&=x^@onsZF!n(9<}t{4M$5&=`F-jifH$g2z#q}=;C&>y68m#I^3q%
zLdoJa(@l$wTdwkWP}NY?VG!|uri%ZJ6jE~p&Fa*WKp;=M>d3zbGh=~GMC6bmh)s2J
z(31dX$dh{-#HZOCbG|GH?AQheynNxnAxR-N#_;M>K(FP@=05#Z`{U}ysd<-wK)0Q&
zMjIu($Ch``ve~anF)sZD!t_Igp~56G38;a_K(_Z){$I1f5%vtvptA$puYmwbTsgd%
zB(gARJRv8S*&)Tt%oiFKig#`kje33lWpj3)@(<rd6mXbWx|K0FI==c+u=DkgE~*jZ
zu6sHyxMZ!}&UpEiILiE-tU{^yHF&N(YG#4uCzZ_@dIB*4-%F9+iSm8yf_^zDFP-89
zXjM7+_@mh5A;j^Ye>rK*^!yp+IOLCUcPKGG`bx@0`GbU?+IA4?jSZqjPIWEL#RR%*
ztn90s-S8@;Egy?{OcajIU&~L#x<0&bLZ0nB`T99JopjCL4T<p-e4_CxD)~EL3%S{o
zC&hcaoKh|7ql>rDLx7T*zSA6p$E;KY5Z}Mh8e8j{LDD=pEP*|jSLsSkzdDe1BkC2d
zr9|tmc@QQ{hZhFk@!)&Bi`FLTxq8?eVy{)j&5RAKOD67ER=l$>yr~7YvHPf8a5O?}
ze82YhKB>fU8Nbt|&LircAC)f8!_!9yK9vmBh|fAT+KOh!uQb{!L^_9wgz`HunLHoh
zLWP?t;)YME1KYmF1^fIUyW$q1$cVw4I(Q#OPziIiw=91mzy^kInT4C@-bUi&@5=Yd
zE_6%xdnFOkJJthyq&PCA2w_`|MpeaOjYntL!0%=FozsFVoKC|!bks~7_roiR>#ZaH
z;BohS=6nF1d_=ux3j-WxEvt>PY7$HZe3rH|lp`>pSAhLXxWa8HCgJ_F`NsW#9z;S5
z$^dP^Xa)B7ZaBa61m~M_`Px1W8Mxgz6!+}rb3sBIOf!X@k<TEPEiBs=Y|n`E>s+*C
zud!{&jhMjII@FTnoNiJk`|`_qozF-cjgN8>JLqBdkzESHyvfX(R3VC&iF4SU;q2<D
zA*^?s=u(hckf@(dL&iJnftEx8h7k~4f2I0^#z2DF;BqY&ZClww<J>U!(@yWJ34xx0
z=BFQHtRo`P!ZVEe@@2K(E*8#6XUB%J7afR5EvQ2vMdyO&?MhlT*t^VG9eBOkuB+pU
znD?wEtRr|{uDlrZ!oSffjQSzac<db<_*-R7>EN?E2;?bPVvZUq{b}+M)y>jPp{EIA
z;}b`2m?xyCSmwdmU0*(ObmZpKrE;)Lm7Vi4KyUUT%3YRtq)a<x2Up!ApG(9;{fDv}
z6<o2G6)!D&P)DFz0t}B6rU5dvghd*i&-A~@#4kN7KrL7t=*^+O>i+<Kv_=9^$SC$3
zhh7s0?!rp6;mFlHJ)mn!qQ-lYQoNwFJ=!Rcp36F!tM9z4zHnwIUDr(FRUd5l|B&_G
z(QNm9{C`lhwotpoOpF?dRoW0CHVHwf%dW(z(Q4_kON_=QRE-Ehv{a0?da2sP3{|B=
z(bCdIcdz^R>2=?~alXHEdYt~zo}N<gzTU6*^Z9r<sb?<VIR_s&tauEE<vS?t?d}n}
z@_jSkr)@+-kLQnz=%u51tx)pt8&>xxwUhU+_P^7qnlVSTc@^sZ9V90d<A3hui5gZo
z!gyNz?0eLqNNwPu&l(--1H5ivFhV~2V6I8hW97>aUuU*Z?7~X4x1aSm*@$SI`lK7_
zN3nxs=$%%r+fPKOnJb<7=R_tdv|=jnMCn~)wZ7mz7pE|>k5pO7Q!cDn*k1J$?aM6z
zkNj>3#n;zZMM9>p`t_|oBh;|UL6Zq9U4D|&xlY$MS0i2NekJMB_SkCmP%Lsf@<b@@
zV5|FVZ|^D^r@s8gjHacy6jnj)`iAofW)%!sjX18gD1OFcK59Zy;VEF2F&uz`HxZl<
zV`5=0i&frmXCCZ(#-`GMmBRL?%tx0M-fSMu`eC|7wTUw27#|PWY??TtBW3=YNYgXz
zKVhpM)VQgXU|;c!euHu<JM;`8Z*1s=6q?dL+HhPkbTNb19+(I<{W<leX|FmQ0MM`#
z(%-|$SzsfiTIl*kJwqw$_?H7rpZ56PIA=Z=*z3)<w@*--(Cp#3O3>F^jth#2q7XW7
z!sF%nLu}8zcx&#yn5ewUdM$*Vk7DRjV}0M4<wr-^`D7Uvg~DfTZS>N?p<I(>&bWrE
z6s<H!osEI}<n*<wq>jG;aJt7mIBDsM_Eey+k)x%)bQ^EylAY|S4;#|ti6!0dC_O`>
zL10AqkSct^+1h~`C|94#2NbM_8{jE~#2t`7hit~?Rh{^Th?l&$6pym~zTWC6h;k4q
zg)iIwXu~A&zFle?^I5bHBAC7xU&<=F%>&*Y?6DIrK#fYf+CAJ@iq}{FE5g<sktk=I
z(~He$_k2}bG}YXs*{QE3dLSL@rA_I9wD!kic5(<f@~wMROdV06Hdwm(hPJbFW>JMV
zD^gN}Atj7WFC;Xm3?p2ZhK(HT9b-4+kdl6;8iK8A1==AJk5$Ml(oijbGmUq7g=RUM
zT4g~5=*QBuejWO8#rajxFnAnD%?tzdOxYKB8D=xhmuxaw(*$502L_UE+)2&*ir^7_
zH;Uk+iFEkICNs>vV$7Qqa!Mz890dV&t}o;gJ4M82RZaH2<BRmELkV7MozF@}>|Sp+
zJN=jiTK^W*3zud=1i)l&P;;TF;Z;A#_14|$qkO5w>NN){@RSQNoP*$gsBi~(D4yOG
zeT-ltoM+9rf)4nhq?iB6zxK7*s@Fl(Qpb5aNqq3!{}v7ZRR{lDV7Ue)>OxR1cK?n8
z{~ia9NT7ZK24u&7-u*3aAryK@ht+g1E*P^_xY)-PXAjM`PO9Ws^zgN>j&Ejm=VrJz
zaX5J?ic1BBXZok!o$&hkHvzj0EbfUq0PdGZ$OFTVE%ymdy1Cc<JA$Y5Ze@FnyuFo+
zk+#Utx=O4?$G-I6HX^1$DJ=#KDX`yp%KjH;o{@5l3Q0wP{Zt#&BF`G@>4u4EQdgZf
zZ#)PddW}e&<F*MsnbaCAVr8T>9U3%BT5N0V!#Ikeu*Stp-sjev-+TPMw=Se}uA|bA
zc$#PF*7ddFJoka=8(&YdBDb&o{%gK?IWnQa?Jl(E*yf;1sPEXb@z?&vHM~xzmA`^M
z7qdy(7F7{Laa=ed_8x)H3>$f4@wNIcKch)J{E|-Jz@dY7>V!(+^+sU~>krFBvkp8R
zf-S#PmEd_N)sGfF=D)E(EE=q~a5sRL7zh;sWo+L7z@qtwer$~NCuqyohF3F+olz4l
zU)TmeMWNY2DxQ~<YI%e%sNkF(ey;?o^x(}bFGB~}VoYu=c^|W|OY-@<HO6VyKRkj|
zCwh0?1TK06tqMSj;aR}bQXfqVJP^ol4sWZ<<mR=JmG3MHOmMcvjUmTcjK&D5rIqnd
z+7}*q2ZYtvSppf|osaqoRc>ENt$B})?8aKvgBerx3@{mss)_TeG>8aKrzGICfoJf%
zno^aVQAeAQS%Vkpc*xotqB`#ziJWcEbH_&jmFGGu?Ve$=(c!XSpU+O!OY}fjUV~52
z?9(VziZ?X^Z3>bi=PjxVq74Oz(3FxA0bT&t(FiO^D%m{BTnyjaB^4vJ3aN->lW9X7
zq%iPo^XajS9ug+_!a1JExYyk-$*1}r37uKjco)$Sio4@C{3dXYcj%#=j^@tWGs#=b
z;|lYc6GFQQv5XqQt-CVR!%nfZU(0Q89Ot5`_}sf>BmHCNH_|<2qWa9+!c5lYx}Z;_
z+Fr?OmUE*X*4-uzC(Wi<w|#Kz7LkzGz&E*%WDX;;h$@iK=1_lpP9fxooE%+L1X%L^
z!l?(WH<ww7qV;2+aRmIo$MY;x`#Id&AvW@5PrApi6OaLRjK@?=)bhgB#Vd$r3vcH{
z^RQlzf4|!h`Jj-UAJ>YqwB6o|;kNx>VjTeX0RvG|L!Uz@ZQ9BPxG_DbFSCK}GI%0d
zRm+^-v1$Rx<0iPaDstV06P(m8t<D8P;*%M&KR9+az(pa!qlb+Y?|8UX>R*tRDkW&V
zlm`gWDb?;gMUFT8AgR~Ahn<VCvcH$N8u6=81F$)Y?OL?^aMvMU9$&73QEjbVI1v9<
z@K^uxkNLc|ccU6-BXvS?tqIltK3M#}A5im^4*CBk=3NYCx#fXLfk^v<d-?N<33S+!
zjw=e~jq&%<d|Kg`g|5JFs%}^RoIS+2Eow*7jOsa>smc3||4H=yM?=>E3|78%QS!WO
zv#CY%UL-w7MjMY%<nmBKF-Uem25DA=N4d5GP)yy3*_YG{ISKpyKEd^{hq3IKL_Fz$
z7JfMR{ZP@T%h!SkolYi&mGq-)0WMKzWu^0SZ9Fi+9!NeU?{#Z``mOjF{W1PcLa!90
z%}`Dq2p<Hf+-q}sPVRP?wH7B!yC3%~E!IU@zaer>?Le#gU)Q!6b-ToLtDNf$-DBwc
zMV%d(*3#_mdNlQyBX$|H?H-0c9qDZifr*;6HS{$sZ>eFYlbHRG)g#BqbCkjdd?4`7
z*lgOKOQHN``(IPlHH#h^W$-sWGc|*E;ip=l81Yx0=L8c~l{savP9HOVU&6B4>b8q6
zlyY#kDd^fO0cFe~2OF~!up>T-*ex1xUHfxD#4|#Wk}aIKvQIr#+q^u#wzNN~<vW_}
zdVH2!hNsciH|MQ0S7?}dE7X7#tmQDXwu*K<a`n=ziUB#+{kS`i64$o7xb?7Vz9sOf
zkt!gav^?_ESrGiX(LWR1?!qlGm_A=`2|)hTQ>m6;+MZBa-QmcQ?q7&=n`?)((tSAr
z+a9a$#)a>nZ&nQMljw6y9BA9tn5sexE_eM^uqdoh#gHzJPYqS^h^t2sDrJhp%2f${
z@?wrzLl|(fXLI^|wL{n#g%>dy>H9NCxwOmFu&_onS-v=tmxQePjFCMpM^912IjmQZ
zJ3lN+075j^%TV)u_;BFEcYx04KOjMXkp1cTPx^Izw>^Lm4@RlJp;zl|o_u6)bXZho
z;cDzcChhrJBOn$e>(Z?{++0os>?hQHD!NSe7#Kdbtf5u}HLqI+3bSIbe#3_FZRt!@
zp`+SFG<{V++KTPzGj_=&F*34v`Ei3<V7hHh{To8TC831XkHdtrEdRTeJ?G8%sbkdO
zQD&T1>1IfGUeq-om(JH(HBiLGK7*%XR*X~2b7=F`hdrEjHA>rRTJjRjRN#@yLW#Fc
zrOiKIUWRh7#$kzf^B<b^Vy#3>c_-_R+%pk=5FE~!$mmAfqJs@wNnz&F&iu1T!2IrM
zL~bsc*xhz7%wo<JqaHSnTxhYC$ZMsL>^5-X8GF}Py;{$&9kD_SN0enh6+<2=1pFe|
zTePSr3t|&!(&K**n#!%RW*|Q=&-S%wZ!RNyh=!_tVib46VDGj95)<`DfvrP$Nnt@+
zhux-@`3Y8lU5k`XS4tv1hQHARB4u+xhpd=L=A1ak=Furrh$V3ovx)He2rf~@;94!>
zu&Ssx1Le-s6X#+D&Ax7iu*4?ThUIOQ`bq1Cj(ul%7LL(ITumk_{(<HDjKr=%g=@@(
zqXaDRjr}9B)*GIfuCvtWZ#QhrzmL8pkQ4ld*JNFMbI~4u5EDSI82It$Fj2!n|F@KI
zhJ#A7QY5KB779=z0z@9**Y(wkX_8W!cZf2kjh1HtfX!cd;ceU9BkQk?6z%TcIqHaX
zOE`AZJZMl9oGg?<b+az=yZP+8KY9BOHESOPW~l)IN#H^Dbf2G&edV>Ofeht;SY!9|
z)$`uPJ1|u;A%5DnU(K%`#C^Iv6XYEl{C*UBfj$)JJE)q$^YyxB6F@Px0k$X_hG16Y
zbqJ^V=a4tdr(^uzjlnU$+AQwZtzN$&+@IvC|7;?H(|t=c`0Hr=rY+s$0taot^$7nq
z_;3Wb2ClC>LHsz5yFZ$EZ7c7h`xi+I|BP#K?{vx*wLYJIy6tOy1fp<uyM}%@+Ty6b
z#C}WJAbQ2aw=9gbz>DQ-6FrCwR^ED6EGlmJo*n5qMpIMdj^y{b2}*sI&9t3Dd6+Xj
z0E@CAs0m<sfQ8-7TP4V4a3{2bwG)*OEKCIzhU}4Cf&V5-QeBMzbGzE#x0|bZ0avbW
zoJTpG>u%ny<X|aCi9D!K_BHj6g2X&UDv5TEA!)4&`iRV|OjI_PmTG?SW7bISNDZ`K
zTTb|0;>%=C9DS~MDMk5Kx;HaS*dRiD8*|X|Gv7cD9s3|)!Sk5c`52`)l6DtGNf(b8
zy0$k_NMRjutMB1&3$5;co`w;xNe?aF$e*h`|J79oVSeLE7qdlA{_}~aan~J3i?`y5
zU04aUi~I(sGk5lwBWu>vQ0N*W9ML4>oEs-BG`V>8p!$6P>;9=cs-a84qcn4RmBHmm
zr{bU!W4zoP(X`xa+VAhYW&uB4)DN0ay@TIr_|)(Y`g`s>?)B~X0vY1(Ij1k^eQC>=
z6Gy{5mOZ|BywTng^*AwQ1x5EZRGXuJn*AYCI8|S2$WWWFhWzDN1E3~yE<<gOHM$uq
zQSR|KngQV`pz{09V}(*%XCebz6?1G=3%4rmQb)nIpY$xLTIG7zk32!E&9^(-R&94W
zq&iF@op7){te<p%K+P7iFx_W3!I=K5!y@{mL)J5^d68-2R}1uwvG23#r>?LGqf$%_
zod(GJM`etu=<kcFb(2LQ52Z6oE9c}NdxyF;b)6Qip4V&nDqS3&vs!JRigEW8-O#Tv
zr+GNb0`VA87(bv<0c^^7K&__|Zs=D-{LU2M8JkJ=GRL_{LTWIK=;u+5A||{Im<uZ8
z!m_#;*oGa2&D84~CS{qmXGbNt#j^EZ;gLJQ@KMlkp{(rPyQER+N2~r_pZ_5G0XA@z
zb%gl^7xp(HrrP@$m=h%Ez1ODf4w4Qj>E%zS<{_}^ST}bPrb@FSSNDRnoZiz$kar@?
zm2}boCqBQ=Lj^fgF4Uz2QIR&*I9EI!g5<oL<?+FJQX=F?P_A{_53Fq6L=9{xKD;qs
z3i@!2;UjniC={qgDbd@8HAfVIEXf=r+T0G0OVeT8zI#MG>mSvn8|IP0BzvcFYqnlY
z;J*{gpY!Jbch~?nHP9bX0rbzY^WQs9eg(%FIBMXVu?_TyKp2mDi}>l9(njbfRhjKk
zx%lHhvD~-exsaT7)UPOB)w02(-7f-zGS4dQgbJL~Lk<EKRDjQ4rNp0rlqIm<uV`u0
z-nq(jZbXT<7)`~3nDt%ii$0lche75+lQ(gA_&4FA@64~>jm|f+3R@j~W&Yf~1ce%}
zTRYq9wVYzH=Rm(&y=t?^Rd||`p9UW;3U`oE|AGJAW+TfD5GQF)4(<l@dr|onU+DZ=
z&CX$WPoy>UI9D$XiQVXCx`z}I>kOM3qlkZBG2V80C$Pe4x5eL2_C7Y*7rs&K3)BP~
zg2bv|d9gZ;`8KuOCC{ygE~_auyI9+sGT3+h6I)`Ns$k6rOyTKzplu}9=;l6YYbw@l
z?Ej!n^Lxj(E2PlnuBRS>GBStE1a}QUp~CUP)UF0{8CD66nFU6WR|r@$@D-MK<{{^u
zWNcMPHaSxZky_MX$@Leo-2fe{4fD51iE{P%@QQSJ2z&_S*{m%GyM+kXcu3>%+3c%x
z;D@A29nm!)CQyu5Dn<+sEu4xsqaETiigz*08Q~}Jp;?&$EN37Tr;~S%J}zre)KAQ=
zE3<oRTXaU&s+qf0kyG{`&=oqSdG9P$<8g_N;ak6=6E6$>6XxwcYOe~Xd(~ygr7Lsh
zB04TSnjC0*mjHz6+|3#5(z*>c62q8Pa_V8(R*#P^l9NK;!0&jJ-5O)MjV_jG=D3j>
zVY=b;A5=S2c?th87sCcfL5RIeiYqj6HgvHfx^dVD6%C9%E?v>Jhn`~axhVi5$$<Mt
zH{~!HMB`pB2Mhaq6aaGG!GcJAyW{z1T*Su?W$LR18?#zxnSPf$4fb29PqHa2v|5~R
zH`X{5hZuSIZE3GkIQE6M+#ZQh+p?A)dvU%D<Q=wew#&E~|FbH6Si-`NR>pn1*c;KE
zdAQPR`9-?tSi5h`*<oM4m5_FrBP3*b!Ehxtmkb3x@-S;b+6>mm5g>A>9^e_)t9{ap
zVqP;dxL9fSJ5q+-ZbvGEYqC`+_!tjG8<1aU<aBrk!YdRkLV~T|`>8xeTNNZ%WsyuO
zbW`~Woq`T@%uDe*${@B1X8qjLSLb6r9F|!(Vyd2MHHNb1ELT6ydg0Y%!C7PKQC>PZ
zCoLkpp?835LCjWL>UXxA&=2=RMTY$3(^WHYF5&y#XymKB1@Qsmz<SEjYVsKrQY1<~
zFE&E_!=j3LE$=ije;8z8@~}1HJ2(s4q_*Bzl8gB+4Ku{OBiWgAF2A%rQzOEABzjXN
z8gZ-X8SDPU^TA|t?vr}v*>i>%ITUovlC#Juz^DB`;@$sq3;|At|K=@HfTr+&PllY1
z0+<ZVr=&iMFTY42*!cGL;7@9no&Hb_ZHpqT3jz|*HxG?O<0UQlLGBsTc`wi0Jo<d}
zt=~CGs1TII2PlLF2(7LDRgpoGN1tKLzkrus&c0rht4J=!pZ&)1M@*$}=Iy#AG99!o
z&ts&M!V!0bgJnzkNfT%4Fjpejf6f+I)~$Gz?iH~qAj-*?pkCx)<#MX~FWBlmVVP|L
z<<xxi*Rk-Jl#hOhk{&2<?-tfM?%gLo;Hm6_=&fX@`Q});5YSJKA{ZJ@IY!HDyk<p3
zR=1wQM9xKdS=Fya@WS{J#$Z*xYSoITf_L0|0;{B^v5v&rzo!b<dsuzHEx=}UseRjh
zkJ9muG;VKmUmMJ-48PK#&pGM1RZ&i%&glt+u7hxz%@g<h$uv!4aK$*?0oEi(H+WbZ
z?9NBzf)n53QioAtG<|3y=d*OsL+GWu-b&YCSo2}{koS45RJSxmkttG!qrZ@S1=j!_
za;GDt*u#L%BhbKvt2}i*jJ%eL73EdEj+!QUm55&qPEpnOsm6}@nQM_j)doDpC`V8&
zpJ4o=1rQMwsV@z34o{JKH<1M2KLxN^!)Df+pN!?MRDi={*LaZ;;#rLum!8y>DzVLE
z+hr4(KcT;t2dpB=7jI}&ZP~U`(nnxJraUU-Tbr|3E+sd~1nh)<?Lboqf?M4h#~6kK
zB(LJsXSC98`pTk}@Q+e;#Eqm03Ev4jijmA#1a06Dc`+>r_G|6(PFeHryoYXm79lo6
z-a_#XA@@t1b%v1tNTR|9*-dS}Ajs!T5m8ug8Rz;kC=;_|=nNNd1|(XJ0I3C#wuG0+
z^Xc)W=&XvBP{+#O<3A*ek12-VyCLaj?2`9&ptWXDAA!<ZL*!suGHjJ*a0F^(dA5VM
z%acr(+?mUY5}RY}kbubHip-NK%++=mc~RG`gtwycd}w^MW8yTSv#9U6ydg^R3-XCB
z>7D7=0AF+5EG|utMqU%Kd?j5x`#KrCC;E)oavl`%1oT5hH4joTNR{9rXa2x<XiyxD
zptnyb@Hdk3D%D%l3z5fXlYy_jeXQVhwdO{<HSvcJ>5vUqM{=<yEy63x1hy|WUo6)V
znk4(yFCsprhShIXkhG^`0`zh`y_LhZ&0i>bnK;_rMqFP%{4$z^w5;J@${Lks44pAD
zSFa&$oYYNv6KO)q-aOHC-LX|~5c1Mf3Y}U~v!TqwJD}ug*-^QW{6*WM+HHLCCIKxR
z{+gQR{2G!Hcp}JS%y6k9`}jg?EkGMG7-NHs0UKThUkyD)oiEKuo2z2jNornR_Z%T2
zN**mWtYy~P{P6KMcH|n*O2v2l7|p{Hdz$aoBMsX^%$u&rsno+a8)E>$(u=)3nBCMl
z>F^-nw({Si6}KrVh&GZX50t6?U7!5dz5VA-^WQ!d&**viGX!+%dQ(?6Ta({)D~q=^
z2W4RUQYzJbUhZbITk0nC?NaOZtA8}buReek&0gzU`1@Ce!asMf5>~aS6X-g$9%isC
zh5IeQzx}-^*M4OyBu>ZiwfcT*+K%R!YVg-MyZamecr_kZ8MHOXHaI!L$nI!DRZl(I
zCh&EZwb3-B)qqB|9Y}YLH-6>D;}!jxldvTU;4HfEYaF@ztljbaSjVvhGTq@`q<9vu
z$2NA!`8Hq=*OhiMC2bx!{Cb<lui}M@GrAA$`=?ZQmvw<DXuP_y0LuxXWCprfE{;@x
z*<PUx5hvSW!!0o@R=|ulzfy}y`NDXsNjPHG7EQR)B4U<qI50jOvrXB}lY~`j&(SeY
z@wIn&aJ|#smelF|Bqn~StOm*OZ6B_tLI4PeTTTMoLnFn26%f#sWkk!5CmNtjWt+&a
z<4rJ$P_zST*rLKva2Q|aIInJ)Gd^CoXOPH@z_i4hBNVC?g$|k4>@U$(4XOTINKvmD
zAaI~^mM#$Uo^T=X0yw4kIG!ptNd|ezg8LvT3dL{j1G+%9A&oxvgR?w~U(YrHVw3~T
zn|bbsOcifQ`_Alre~X-m*N0gGqYIJoB7kTFY>6a4dbe$SqHTPLq}sh%Ib_5Q6I77S
zsq?B*pXKGT?yf0S)2(Bacp7LB2oYsYXC@_+g&c0`M%s7W#NJcfZ_ZObO-Q%3(sjQo
z(DK+_z+YOHo)4XTrk=rhMP8^ohi^#7O8LqPI$^Bcu)Oer1iPYRjts;dEuFEJK8<jd
z*bDIaN!Ka9q}-Y=IXpnNPMWl@e!u~88`Fdnvky4md1HazHret+#HMJlLQd9{&?MzG
zbxC5@TQz+6+%uU9hj83adW)aH6H=D-aPUOV^{cArK-h2@Q_)2AJZQiRO-zV!4aHu3
zeUGvmj6QK(h7uFCM4xt~c52DTdW}mTeF8@4C0G2!Puh&j8&hDnZy)+an+<R2G8oo9
zmTCTrxV}F+RPE`g{U8it9`mZub-^{YlU0S62z4j%X+ran+3GF3OROB-%<CTRuSEPB
zG=1HksVeQLYp6r0Qc5agYD=}<Aw}0%hB7ZT9Hc`P$<7f`*CKiF(VH9Cz-XWG&q8k?
zf$s843bHftnFo^~q*>cr%^_MbXR9iCS+DM%VrIEkt@QA>`mZw&#Ouaf|0bwQ*m;J;
z7?Ud|vIVC7SPo%2_h`7x;HisTp1wcIEXlN-sP<2xoX|x{)3~N4cxEsMMJ}y6Y;7y1
zg^-Jy6*diNJbwJK(sp+Hz%9P2YIoHa9CExNQaC6b5TlF0ScZ~3REjc3IAmTY8lKXf
zCkHbzqZxke(<$9y#S12zhAaW2i~6p)64dXta{&jq*IZ_j?<u_EYs^2g`&#>9>cjR2
zBbzZQ-ng0Mvo~(!w6wRM;)jsmm;=+m|F*S<VmOa55vP;?kHZG=pa0*7O&n%5IBi7I
zXP>D7S(;V(56D;L^)fcld(FAVLoqC0^^Ux~{^s@M4!ZCT{?+de{Atd%s9pWQ!Dj$@
zx!o2TkqPc&$cC4MR=e|svl79F&sv@My42wm7+Ry_!xRkrZE{p|t-bonCy(sW`XB!w
zEIle|LMP>2Qg63N*rcXF7juHc%i#dhjLxeR{Qlk2Gs5pL?Wa4m7)(f)M!6ntBBO6-
z4fdsyztXVu8h1&BS}^eXW9;B8WWgLr<L1ZeP<5STEQ<;D!M-Hvl}alTlT#Tb)Yxl9
z&MIn7wPyG}^mpfZq3~3?rwMfzc~NLOqPHp~xU5q<>Q<+bOM-Xj_u}fr)-)-F=%ba|
z?FDNlcD#~@oV@m)lH`jjC)c{a&#){s23jAaSsP3&$|x@6g0m1lxtT5MS4e_xY^1Ps
zI4~FkM15ZnP~x15${YbwpJpCZaU*j-(#E2+roFD4bQdNxqwM8{x{&==^(0)LkV{1^
zN_L$J?x3swj5TWrH;1cmAC7a55&}giTxU6FgJVu+3iI8OiSlfz6#@wYb-ikAsL&ym
zm<tTSF|e@L@g&B2#d+GH32RHw{@&V9=X<>f`~^Wcmqd2-YidxsA`@gcN4;}KVuuvE
z57WM%n~IBmg_vf~F~aoUe4%OuxP;x&O$WaKHfzWxUfPI@EMwNwAhhIhl(1;myhd>X
zSmk{4m2k^NfR-rcdBE`ERng$#?mQQ+CR=~J+-+U0JJ>Uv$wScdSnwk0N3>OOHUptV
z0U+7tj%jLro$id%=qiX?ppN|Aaj%<q_J~^yYmT?BI`1ayL9CrX!)`R;Q{j~9?y(h$
z|KAD!>Y~fxH5Y64zuUafyh@+|j*+(G+4EAI1;O>ydZ-2dYeu`ts*?neaMGxqX0&K+
z`xw=S{zZ3loW5?`+kbVnD4nxZQu?H|9EOOxWQDl-oMQTtu+3%rH1(9A?DVBaVRldR
zgy=!B@fa{w(Q5>L&yK5kK3oY#F!@682W;mEf*;CIqX2M@yR_LvWrQ52b_nhI%TMD2
zeR0-pBhm_3i9@9xPell}&iONO9u6a7q&IfDm)x?zKQlKB=qmVU4IHn;6aFvn@;IYQ
z&r~6T6sp}vPPhwg6={w-BH2}f=F+n$i|=MyeC>P*c2>WhTnev>u=_wXZT3SsuJ1;_
z+ejGUjXkDSd3@saRtS|os4e%wf`fVb>3H%i#i(mQN72hyl$4_iUtO=bcFd~fY(QFZ
zi0yA^0Vm65Lp~4S&D!jz*#JS6azM?noe{X6!OcaEuZajWza~cq7KFwg>FLh4%_J|&
zd09s>4-YA%6!L|lC(t{W1INqK$2Z;hn03qFX80R5;iv4@l#egGGFzv0A->J9oML`H
zenvSGbY-!x{?vEeyO2*k(x!LNR$c}5#?;H_)F}TRH2-a-_QU4^zhhQv9xyry0FF9&
zA`~P~{L^Up_q4H0>ZNb2r%jp^LZ>=oTW6L^*J$y(&zo2niDL3pf7PWb@CR)&2OCCW
z(XWm<K0o?gx{lxe#!Bj$AH?60XAj!cNttXt%p>^IoXq~#3Cy9`hG!Z427B|E=}~p6
z-V0N2Cw{bI-@bH*W_4#w46xbM-j}{ID5ZGuX)8V<Sw!ZuU8GSu8~^~!^}@<ya@lZx
zo#xf5DtgSoNB5`pubq$d;QYk@I#XjUzyC&uw4y4kiVoA8u<dI=s7X>NTv{wpf(3J&
z`l8uVsFnLnu~KGCH)ON2Di|M{-<J>9G@?+0#f4&P4==^lJYdcHpgRrR2ouh8M<NtC
z!i@}er;)M$M~y1Lq^Oh=POlVo-Fd<T@-(90W138cPCAMk*T|h7KWN1mz|zzJfEDc(
zRjn2iZnOwsODl7zS*~56Y$MVJ>W@rBN{H~f?{;M)+k|_oumCzFM|e?WSOb&Dlw|^b
z4Ru`a3(tlESdKE;MXk`;)-M4S#H*PSSS+pb{ktX=q|5KfRO8*C23n|VZN-$Tne6eC
zkdPGU%QswyQ#~rhtesKkrQCLrLj5GWC?O<rZ{x*T29rmbkXrsYm3ko=<oub6MG4FE
z+0E`@`Q@6+`26?G8{_KQdILdP2$DeW2Jq=V0`|xBHhn*r9^PCe+z^MaD3cb8pB?DB
z7l)}Gw!%{+n;B&^=W`Z^p6k1zTgaR&-O+>>iM4uS?#jGY5ulJ%U-^pFtjxaH)wJPp
zf91tX)#sRyg1h8*?g{f{cc`3=A`SuOz~u@a@ORg?37xA}hHZrTN5@1)7@Mmd9#U2o
z%}<EvjT`rP)~9U!@Q`u^V!pB%@U8^|D7f0-t!gF~tMu$f-aJ&o%6G2)1iRIC=!NN|
zqaW9^z17+Umjp=Rsh5(b5LR1ha2_uRzrRG|B>RL-Ru{oCdW0=lQJaBjeFVQ3{Q=Dm
z$$J;Hq($wcVMNjhG~CL?H(vx*nNVoa+@y$)FP%=bn{*z5nW~@D(=;^p=%^L?L~S^}
z-j!0N^T=qtbUF^JY%M)o2;5A_50+Y6&JTGr=n(J;zn6D$XF#E2yDkD}#qJ^yL4LBd
zfsSgX&9%r7mb$Us#9tNo(VN|h3$8<5;V$}LHJiFFnbBlaK8ifkTN1zWajUQVl9yIo
zl<j;n=l0PODyz0KU^C?nsW56mJqZ5h)haPH2*Sg3P0>Szsf(Jady3d_Yp&Oac&cWl
zTqYKoyNaS+WsN=~MIaO`*NY9D50<N`(eBkemCKzoc5(L5wL*7`7qf=o7hB9a;jWYd
z>rXccMNm^aV-7Y1J@%K1S1Cf5b^=w(_DY?Km#6`DF+}z1Gz<BD;S6!>t$IJ1f_qN)
zj+hJ-f4;@nWz9bO@R=5Tv7*aM{QrAR0%QXJJx)Z}@L6swmJ2{!{EJWE1#`U`DHR`k
zX`g`UTzmw}Of}5hx$I&yLXp0u@o9bF=V)WtS2kAD8oyt6!&Nx((xT*9=zl;*N%p_?
z|6$#u4%#()I8h79CqhzvEc)llUIm_nec<&xF9jQ{#FtJ>So~c4*6>rWLY+%*bn014
ze`dsx!|9_pO@FE}qeS`m9Y<U@RqAVl0CH6wbM?E-pvFn#b1iB2_&eB=SLOeXsBz|<
zd3JS^U?g$TC3lK4X3x!Pg&GaPv0N~EL##^5=kXT5oa#rCpJc3dR<+fxn{6!7N^hyZ
z;%+W>5zM*i2+by(jqLlN%@aPNdzbtsM8REX#4h`sLU$U;E@-Ft%;nlefvl}z(m#^P
zH+VBH21_R(3L_$`p@<pMLd@m*7D$+Orq6ngs5+K{_DpUjqr91xHL$#)fKr8nS}u#l
zv8UX3sC_hF%^xm<9h>3hBAl3(&BM75Dt5Qgnej5?Z#|@n$%~M~?xU%$DpKS+$u#LW
z_rkJTWNd4(*G4pFR${2}xwIvdmB<uEIURCKF3_Hu4}0*Y_ON8nhjx(k5l56b%YbDl
zxReS!0h!g-45ldGKS#e8x=sNoIc>wK1GS+~8K$q8JY3rF15Jw5$VJ?mH32O#D)F36
z@OlNp8;2L_p_Dg=PfmPk3(Xfl=y?sH`}DUs_I~8GiZ(sR7Vy^clOBbXJZXNTeZ*j-
z=ixfllj)>1QoXN4T0d6ICjiJEnwbikwq`S$)iSh@?)iQ2kl~DU<t0<qA2`2qkqNw3
zkq%~`sd8{UY*u`sg{0?7%Jmp5>ok@u!HK_g<$)J-y~T~h`5~v#Z+tLS3jJ|4u=#%`
zPp$;}tKF<B$hsRGJbhsy!Bgd}qX<aP3JSiqaAz1&sPNg{7<?{h5d%rRu~(khgO|8l
zv${C-7~T~|?k921q!O^9n9!!!w$`N;QMBh1cSDUc)t`rTqhO+?TEVN%uz7iQr=L1s
zIYK=jKCcZxS3z)UJ{#m!+G?@mDyl;Y#Z3AVpy^^Gta^%*?78hz!Xq14hDfhhw)(V)
zibkoCsVK?#v!<c5fZlRz1=|mYlyTooqN!DL@`~2Jj1vM9Lm-W*-sVMZPfZlv`Alxb
zZLQs&5gZN~*C(`Vml<DpkC$=jpw6b{eTs>JDHF5f+esXq%xS1$y*qTk)Bhe9U;R0H
zg-;4OzZikZvb6>;c%p1t6B*$3)FPf<tOOaU4R50I!3&_*8U^HJ%@9UCV1ZH>wY+&b
z30Z`$OMy1m?%QP!$+4%S_HeA#DHM<%CTJ=$T8i?EVi<DqN5)1&=G&3Ka~Kgc(zbVW
zWV~!~*9Y{u4`WKGV=7A}0cFN5cZ)szWdFro03Y&K0D_|g%K-Q@7qqzmB_Owaw2JuL
z{{krvRM5v+2Q%qJ4GGlN&HEQTeiLPoZsznmnOGOqO5?FGO2;plMOXMBlSs?A6TW&0
zCvi_o>QZ|5k2jHmd%(yctp?li{wus^@M_x?{@ZP!Tb2{)8r3hv2=lNp=f9)5XU=UN
z{q~jho%S_muxve?ioc2*s7q}P6DtPszgKqgWe0!NHnxiZLUR%~LTq+ty^mdtkBN~R
zf1#{{5iER)dhN^~LvME+-jrx^Em<z?A^zE_Jnna*P66a8u4>_$MfMJ0Wd6c4(3(=|
z3SLuD$_zjonOWo4a-gKb@2>m_z6TC(oB#!Rwuo=n^&6a;NA}iq)UWWaQR$L=@g4H^
zuU;3-G%4xQcn+hjVNGS7by4!Hp|&Ro_f~S^>lxt9hCTVcSPHm9`9KF6E-Pe?7>Cy-
z(=-hfJ*pQaon-56!6~`22$(YPZ+t6{v)q~j&{Vmog2Oz00jGDJxaAa+y&z6`HhDs}
zbkGnMs^r40(p#<z_Pj9T7_2@qRDG^FBcqxHI;rq84jZ%53$75>vRn|P^Bl%Dvq4tn
zia0a~3h&6Pp5^WsR`&_tO3_sm<+T!pf-Q>+U{b|P6rr2ROf4&@e^`lsyvy+m+n9Qv
zEYjeu#vl`-<XR`->kC#+_I&mxwDpJ*bX}gJ?vuhOMAoUAn9a*K1&OG^NAl!zMb}58
zT7QlzS$G;f*m~9I*@fqA?OkO2spcf(B9MkZ@Rygz;B)S))7{@st1YoW-$}Ix#{<UD
zZ{3DZd4_lbmG9~Spujy?)1@j#xPd0hgpU_?ccV31qk2{@jyy6CIKf&tQdhkC*+G0%
zt!}635Kae+f{S*6c9N?Bo}sMDd^&bQ{t@{WWHUnU)=R+ubZ0DjBSvuj?q<~Iw?3YP
zRo3kTOU}-Pdf~D|8(uoSJ&cxOfj7Y};V4N_5+XaD?J}J)DWMOW|L8t6&TB!l!2+yO
z`c|rNc(Z!4Hzy2A8$eXWe`lY?LhR1cwwT-He(IlAKO}pjyUgD+nCeLZ=6rS59k)r2
zdP9$7;Zs2c7V?4klBDxopND;MDDKcGMbA<-spCHKy|+r%8`IQ;&o5@iw+tnMD-_uq
z-wJ22E`azP(I!Ai{X9-93LoK@4i%Oom4;~MmJ~vU99p$AJiD>Btx6}uPRGAfbqsPe
zPa85quX^%gTCp1S&?5+**ht$csvI@B8*StPL12<-?j<iV+zmllKW0;MMnp-NG5A2K
z8+@xVs1O0EmOlhcA%@Ldvkycf2A)V8L1om;4<+L?k#FB)J(Suizq@&O&)o*>WczM?
z@jA_Od#UK0s%|uRjV9%t)6Fb&I;Exkzm|)@8sl2OYQ0G_n+E`ZQb9y`2`H<G0)$9_
z+N=k1uqef{UYm2<rB-6Hr<Oi*tWrx67zi9}g%`1VxcsyX;U))sxO4*M)Cvs*fP7Q@
zstm4<{<!(&Z9O|g%~P(4n_oA!f$5T!uu}TSA@(OQaPqS>sn1V`B)`nPUM8{gTPGl+
zyz#G2q~7%(BgeN+v|YRcE|W-F*gJXd<PGPnD4t=tHuAr?F5$KCSE0eumNl$2I~|`^
zh;K6QPpi6wuQ$JY-Ov{^azk9>&U&a|PHUS0^hmtXHgO_-uZ#U3b)46|xKv1xoRoRn
zY4nWM*!bw4QO}hU+7E7{gXafteb2=%XK6UhkunUp0lFo@9hKJN2E=h`XUm!;HAWlV
ze5k7E@SQs^g~)M)^yQ&ZHsP+n@A2!&-TN9q*57$aqmoK{>P{ZqNg!y{>HaO<5%CG_
zRc5TPUXe|$BSkppRpaI|-LJ_Zw=o)P!zwnwMF9-_g`9(&90$OUj~)XH=Z+g4rMB@&
z=pSJkidx?*Mx8JN4}`xbgp7gckOmV1-Ib46q{91jg^sZ{C!uUJMDCxJ_O0Hd9jdy(
zMytMqrVI!ov=t)6W!OfkWImEjq8`X*rWj{y&{r#HuO$uhEXgS}Jx6CNA5p<7WxrJ1
zSZp4F3RhVMQ;$%cBYMQ7!z`AJ;5m#)*{8+$0lqrJ+iT?bZlaI3I$v3CipNrJ3a-K3
zPpi?>Yq?U1^B>SaKMZsq+kBvi9&I!eVwdQd6v-(pDil&k#W01+oKOA<gywjb$G3v#
zx6@^tQVR(jZ#$ef60NCI@%o5aEkcHCg7|`50zo4z_L&gd9DYKF+*DX(V^r|WiI3o;
zrx!H_2Ch;&*-_mZnM&GRYYaQ&I%TxZY^&xzr;{Q<PJN@@J~h{S_GKzNQC_pTJoZom
zFf7;8Vy+!wbmo(*5g{@r3N7XeWcou$HSMC8Iro?5z7~O<Acs5(jv*jOH>~B8^0>q0
zka~FBOSAl0yS5N^yn0d20WrrL0yS#=%trOLZk%Jz&8VGm?H7IEN9}>18hwpjtR$|9
z<Yymu0dUP(sQ9W9c{Ov7U8E%AH6jpLtA@DvrfdW^=8GA=yyJ4?3-a|!-EH&8tjWu+
zq&tf8D++RxapD_A#BC{awlDA<n?dfp{Q`hfHW$%+7~J^i63U0J-p2SgID=kD)z=XY
zW!ruXX9(8wXWkQP)wN9^FX>KT+9%1@51GQF=>=$KfDC=uiHZU%k|X1xT;Lnf4n-4V
zXF)(+S%?&l=~_UF<@XcWGmfC(YWe<XO*8hNCB{P$tn<AjC6RF7nyH}PRSWs~{B%tq
zEY}mQxim_8e7i!u#bIEEko%-`c4wFV?7;!)?$W4kX+`C*o{`O;^(kM^ADNlmNCSz_
z&IJw}7yv9#L-pUBE3jCo1!^(MUZ#720w;dmG(PG^YT-Mg#YE|5`Eo-MHM@N|2oH^p
z5-)Mt)@1u%nZQ;xIqPsjv<Y+M<&zFSL#JN65kiBn_V1gMDJ>t8sZMbf`j;M$pEPX!
zk(?QF{@9PI05$$^wEuwK;`4KapM4(l`7{lPM}dWeeZYa_vV<9&RS9IbK*p%v0+Yly
z+H&ku?5(M37u1=fH_iTfxf2;L_&Nc;uKvbMx*vF>kp_N~ye3=Y5B-u)6qxs%ZY%Si
z9=TV|MPO#@R+sy5&N1JHN(=%qO$;r8`t`?4UoTpsacH6D*J0h~I>F8S1pRI6eWvWw
z->N5AM)DTg*Y}1`A_{Bso_ZGh|NeDv^NX~HO_=AQ9>uM@3bJ2pp_4%nV3WF%u@qLl
z*Qozh$4dqKS(_OhdOD#YO9jHC)ROtme5+161ZJJ?j)=$!OA@i~U6j;!V2FQ4LgwS4
z2fVfQltBPB02X$BmFl8cT|8_kH2*Pmo=T|5e&#IHKJByK?5&*Jm+2?ov73s#_B|3<
zs2YuOQwGlqRPn#^dU<NMc-Eaa$LCN>zYDiUKJt6MsyTpGqt%8Cj3N}T!9g5<md_hQ
z<ea)%k&E;Fk65^WzR75@%b6;2Zkb9h_y9amiX3q<6Dk|YQQLuWPDRS>MHDu~j7CU@
zB~UYaX>J>33Yn009S3w#rG|EMhs_)n31cL<m1YoA35^lO_l$EiTEZ0ZdaLt~I?xzi
z`0Sx*FMWQn(R%2#q0!Cch14E9SVP|mD^(TDn30}6$f@pDw1XP@=WWTOHMqfwxYSxf
z_r>fPuqG6OOu*R;A<kM?Iw+3j>G{@}dpKB$>QsKx0Ms!q-|-(wBhe)7p6dUACQTY;
z?p&QQc6}L+(bGIRK)f<~@+&Ec^3FV3bFZG}=W94NtRGyYpevld)$*=H(~JCRbc7^u
zx@`J|lf;kfghqy#T_P3Vs<eUHdURg8?K*X<XfqMv`X=s@7cp!YGq3FuHw_arg|j7<
zZ{cVq6-Ms+yNeKUmSGJ=0^SS2XRDNUinzOqPuI0fZ?*y$^30a5{sCsNEqP|xJBHP1
zzWipbA$nkGE&s>FJ;l!`n_RuFr0->=BG%1cxyo9#4aW;cQZzN``JKJA0c<SE$M2A5
ze^zdrwL9s?PwDOx2UKYF@FTA0)E4%r^a5Z6*u1BZSaSN%5|tpm1JlqsatNc!!?LVZ
z;thXNTSc_t<Xx~P3jkR#@H`NN#aB$a7-61}1Vh+F87H({!!SOGyk;z?$~Ekmk1W*Q
zA|qk5Ni-u9|8@~no`kdQSOQiE1i+;~86!n-=|Ck2Dbd5%l=*>Z*x)F$rsUcdpm(jg
z?MX!A+53F^j5hi!;<n4Z)Fju(%B9JI#=~LiM45B)=-J4id+)XQxBVQSxZ9&n!Pn*h
zo<FpaFgx|HDng#K;E*qYK&cM%=vcmR@63X?9~YY@LbIVj_6#V6N67x0PD=_~5<@st
zk&EVR)?H>&XFd5Cave};Hf*c#KOj}Jlh0~pwisqN{(i<hoOBlt>>ZcFd?tIWeUBtm
z`qk{OXCbqPOVtjItFAOWYbQzS2As-M2d7*bx+--ZzAU*-**k-{#?we#ikE%q+ANFR
zO5tLAEbUn~+X^IEQscY>r}*BGR)SY1uYMd~*!GijYw#;iFfO>E6VpkV4KJbvxj%DM
z?FpaJWW?|vQ<KvO*J8Rv_2N{2>ENx;J6)TqY>crdJpnLc_J-~M0gVo<ceeVxL#_5i
z+zTNpd}vuPJk55`P0a1Kr1qRi2fjbb;Aai5wt_$U%1htc`m_$|dkxTR`U%Trex11q
z>EvffX;7nT-IJD{`RIbi7~`tauVXI$3-xQ7?|^9c1z~OJfNd^+>UfOb^6R}Ub+eXR
zfL<@m+j4=Fq0NwLu1nLigE>xhXx9cgk&*nV9(R7YbDz$mk{vDz(YT>(9c6Pmgr5}r
zi8Bt@<8EGErQT>%asZwqWXjKTm(nEkFR)c$e8f#5i}S5@HlS8#TPaD`<DO@^AfJc9
z6+^@3N3^IQb+h&Qoe1s-4j0veiqZ;z4d31T>`;;`Y5HNI?s^;`1=xK@O&m~6UUZsF
zWKmkr?&2=n2lr6DrGrg{&FXdpYEH0-{D@pUf>6%<v6KcvV=9VWXEl-Pp3vrzJs!p6
zrQ4<Zs-qpXS#n2U>s2G~c&ITtwWU7_wCzb=ITex?p4vOduRr4s^m4cfZ_>|>vyIbC
zVhk?1Q2DfnRR!=BXt<F{%fi{ZQYqNRr6;XrH)C;rH*SAB3_iwpH>kz9Y-lJ{h$uaY
z2v6w9z{O!zp_9&n*e2AldQNBxr#XrF;xwe2DPE?0QE}<mlgG2Tn1Rw^@nenJsh43%
z6COv#|8fK%O462(Za;ARA#|pUYun!VF!()(R~j>1qSYVCr|DhjJ*h%w7b4kbq|7BA
zT6WLB&wla4+q*H6XptGu^Odg@lb}PEywlNkTG$N2@OL9`cFcj@lMc&EGpLd}o^SRm
za`28<D={c#f6rrPJWN3@tY58#e0%t5rldv+&RVwBw$nroXzrYA%`t*H8j<)EfvSb1
zf-4>t53Qso5-PEht~IRYPW5zHGN&t21_PA!d2|Y&Qj0S4zmAAltbKS&WCiMu8ENPA
zwT%SiZ5nzWoF3A!4{|SSte$#BO@kXd7v92~wU>6JufE-lDiiRsZ?vL#t2ZW|ZGIvQ
z<97@Js_rIp<nsV?m<Yl-(7C86wxL9!dvA+P7!ikX^xS9GPtmevaV3kD#db-xH##L&
zD58xt!~CU+^@^;zhm5i;o~lA;5an_+mjXGS3R1+{U7QCDS-;nOQDZ^A@V$u4GJ_rT
zX<O%nkY@~?REE?ZwttwA<{bFL)2NU|U;T~~HIfa!n`>?gb#p}%9rhtDdg&{v51G?(
zoIj})07F&`96>DX-QtL0O`~-Ah<Iip);C^jm5Y(!nIl=0$_wU}B=h~dn@zIZ==|uZ
zAG*@3<N~=B>H#mc6-0CM;P-ES7epP@iimvg3;Ji#-#JV6t#7}GL`<R%pX{rf(|5kS
z*SO2&e7H=ur>ESbIDIOXXZ0&-CR2WWLtDMfaz4butPmjOe*wPbI|{(>IC;RMk~#$~
zg}ErS)lMgFoi^X<N|TaYI%W~uv{&G_jdgWd97F%C{phKdykaN}OD`~IL~6QwU_c2z
zD~r>Z95=^aoepCt_)ezM&T4}~NWd?zPM21*`xT%sj8VdPzms|Xh4r9=HWu%)wjvdf
zll4zNK6*DpWxpk^E{Xd~i$tQoU5ZfA*k%c*v!F<=qI}zpuQROTZrjj}hQ790!}o-<
zDr*arD3M+79gRMWv9Cq-eF>^sud$-&71q5bbhn|mRxRXv%j4_a7HpB~{oT7Zq^>6N
zqFiSy)=R(xXJ-@<f@Ue=T8##aXu3dSd<1i07)^oJ8JQow-F!`t9AW!2k%`>()U<OY
z;+TiZ8ZPPCeUdrB5tjH;&A~aIVr2J_YWy!=2ddySqqbXR^jn%Kn*7{(^Rk<SI90SN
z<#>20uc2hju46w@xY8WbK4L;A`Nqj;uXr;tfg=+3TFfx5*{=MjvKbiu<Xg-qInlp{
z70FJ$#^umyfhJAQ4;NY`!o-sMcuvc;Z4o7ZSq9&-i`9|n*h_uPcP9q^t0UqYJm8_^
z-{t4{*Fvr{PaNM4ik+sdSBLJIqv25xTkYfcKwFItpJ+S;S32EMMy2%odcEaWpf-WI
z0Xpo;#w=rJRfbia$$`i>Xs?Pm=jO_CC)|^pMc2C$7rx1zVW~(jI+u3@8<x@_eLGB7
zVvGTqNEk!KIr2eMA>`5sTTF?3vi!VmS?1kSA0&K+61?3ZV7x{qI0Xod$o(iBytPz3
zsf>YiyV*ET)_MRIIYEo1kD-|M^@_H!hm;rOC$e{37YKzXd^_!#5zMjb7@Uu1Cd-21
z_(N22Au}+_4EHD)sNVDRBbr8&51}+32eFa!#s}~@-9SL3fT<OtsSZdDfL|VEV%0-a
zxQ6k?)r&)AUWHaLJ~wMZgJ$+r^tQw8AM&(NS+x`qzplPQg-9N9i1T9FhS`8QqoP$`
z$0}bSh}Qw<`1~HlGz4{?`_r#60s8gQw<aF(QZq}<JM9t<F-Z(3Y|ZFN=nUj$1O0b6
zC^?k6_9n0F^<SEvFO*@L?Z%A4VrTfcBxRNdP_hQ90*{<RC}*I(J4lsMDI~Qq2*Z?#
zg5Ktb!x6#b9*j@>!a&n>6K#g&YBz$JoYVsu(epNm9b(gHEeHNOS$In1E$|{7w4UJ1
zAL-6Jd3u-HuMp><6+Cw$zP6yA5n!Lb#WgILclkaoI_T-G`7W*vqkTA4{+O?$HIHRM
z=_0XQVIlEK=-WLgE|KDLN&yzn1>R$E;>n`KKYLJFV*Y%wyej1d{?z0ih0|Gh>d&g1
zy7nQKv<09P2FZWS&jL^p9w3JnQ7fh$*KIU)#lxrAVGH9%5;x2$s)PV*9l!aj8yaFA
zwHD=p4(j*g8&u=n0N0Gqv~ZJb0DoU|7-dRsxq43N9$PnYuEzh@=<Ip8a5yu*b4oSk
zpLMPd{)1KmqI<~rZ~Px#(Q(xV9K3cZL{Y>{o=-nQah^Nt)(JzQn&V3-3U8#6cxA{x
zR=dV?8%HUv@?PCnFGK{}v7`&F(IERcgDk#FFCHJ$r4xN;n2{nmp<i0_#_U~r2_4tj
zefsv?*KX(6q=NzM|JJ$p2dxJ!cjI}Wl>LCx`-#REGhe+gNnI&hI=N2gy09ADv@q=j
zwMyAm1vn_$RBJ`srOQs{la-cFa`xA&B+pQUI$g<gCbqC-pnWW-#%mK<(_emRdB!nL
zN-dLTqVJ7PZ92)$04;gQq5m>EYzOy?2Dq3M41bg4B5hGR%3Y<l*FC6rggm=FJtDNg
z_aZeXL0$KFdBktQANY(A{RJAiTgJY<@ER&ZaX8I&Nr+%MwXNE4x>HCpr@2d@a2d8<
zr)R@uimN^K=`u9+Y)=s8_2Gv!qdW3Ns%&m`cq@j65#XU_Zj2AB8`c9|!q_G~)v~%%
zLtgBA|9H5wC9bR-)wAQDio~vn|82MvDR{>=cdi?DeyFh2N}$e4a3{WeX<g?U4?V%l
z!bHb4#rYb!Ln=CmTnSlcYlpt@8hc*Anx4?|UZ?bP%f6G1hjmfZ4!CPO7y`L_Um2xi
zbv?bazrVi)p3f?j4PE&vg*{xspcl|M*FlM51U}*pRoCneY)?k1mr;8yG;qO*@kH=v
zw|f0u=$?s-#8zXVj>TJC%|#%>icXN!)P9N-^}uSk*c!efCsTWi#;<FmgPlz#Cf|D_
zI399TV_*3{+ydZY$1CC$Cwt;RUFX3nKIdos4)*T#cnUhiu~$7AKO1VqFDES*062B2
zYC7I$pO76o?qX}2&#uwVSZApy?{Cl5TbZ?=Xd?1gtcj>59@O#`jrMD5ONlR_J#>@F
zrC^+m21p3ynG;5I1Bd~nIo-V5iir$KaPi(NL`0T}6Y)q&#&r)j@_6+pQCK@VHz%+z
z2^kS7`{hNmv_5JR@>4v&$L`yg|BI#b4rhD)<9<-1R*gjM+QevuL@8P;geFEp5VS^+
z#Hdl)>S=3_7!|98su3X}j$L%2R#nspMb&ZCXmu;qd4A8g&-45t{&HR7iu-%tpZooO
zy;x2z{?ElkUe&6w?%Le9z>RN}%U%5b*Z38y6s#=wEL2WCy1eCgWNJCMZ1bY=1NKyA
zEC>1Re<1z&cavq?zUo_blyxVZI>n{VarzH-Trn6Rg+V1?u&#8Rv@8c-0(@$)g`VOD
zp25>>N`hs*J~?^(o0ejUqK!I4THc(QLc8Pgm&1oKj@huwoy%13_t=4R4=5UD-5@>q
zbCuWpK1zoaA1Rk3OpqG0$F56g<(BF{rlwL!;&EVWiE};0aqq2>W^f;>y>jw%d*(|)
zn$LC3h1%C+{ikxBO&U%e9CPWAG~ifE`WnETgl?m4Q7`rQ(&v<bNDIC%rU^qdB^G0!
zIuK_)^OBT}n`f%>aCRY0Y%Smf@c)IKsuoRCiSL9{BTl7@LnTZc>XHS#;jy=|QUGsX
z80Sb<Tz&QA?)O|}Gt2o$T^36MwKqv`cCJw{QhAGl5A=OD%z93k-e~QUKmQd6ppTsM
z!D?C0O4NQu)YwSd{S6Wq%>jKP5Aeie;B|!~TnUX4CDh9-FkhNg#><L+dco?MQ-MeR
z)R$3jo?^@e8f^^X7SzQ>AAI)Et(iGaj&}zT#wH0l-*!ic`f_TLTL5OHV=viBK0j&U
zcHU`EO>2xp`8!`OIC%^4>gW6cLOF0mM1>0o9f?YpPA;_8rBf>vO}DQ-OY!ddH~l;Z
zrlNc%MzoPN0fEuE(dX?Yu7}M2y=Y%yYb}V=T5%PEI=AFP&pqui^)Wm<?y8^elODO%
zQpu?Y`!mvi`f5^Z<DoQGov*4ndw0IhR{%%J(anhG`rD@XPR{C;C?0N!_7QO99t~7`
z>sHH${kQEhJ31}`IPghdZU}n?A<|&X;0nVyjw{FmEut0KZI&#0Pe$a`vv_+xCo5iD
zkD!{GuntF^j}&4TuEX%aPS_IzEqNACY6X<L)tCaEZZ;Uy%=JS9kcKEp3oD>%MrV-y
z9vAp={5^b79bW59e67sn?J(*8+KzC7P-jjppRlza8@;U1e0Uk4F6Cs8_~&qgG#y-9
zW6$>21V`5(1dbIV;Z*dYQRsv#ImP%{(q*QtMjBS%y|6-!5YSzhW0Z}rOJGdT>0{)r
zk*3kI{Mq%FO$0{xyEF4%UzJfGY!;xo_D^q@&w0?-+XdoXBwz*LeP&KbhBXGel8Js`
zpz%F%C48Z~^1vaRPqj-<wL2%|>`j6|+6t~sXn!Akg=#t5so><9V7%YBJdrAOxY>Dk
z(VLg~sda@q)cbR^2k23^VBmQRub<>;+?I_jR4rK@a-ICA>#q!r9ti$&%;PF=@jc7v
zu(L<YHY7`>!RIz5tgvC_=?z|N?IOMr<pLyJ{jZe-W><6G7E5Ma6=^NuKvb;&j!6Z6
zWTLXTSF@i&NYp3*&#)w=1l<qrj^fnrxZ~q5)XW|iL%BJC^Baf$88Wc;`GBu<%uiW1
zI2ENbXX*Syv8gBh-GuCm`!=b}X7N(}h+rWycM~u_+yoMw{ufzSOS0@%7OwwY7E~~A
z?KC}RrAi4<{~9563`bph7k2m5cYRZT)Q);|*DU3U64F3CL@~-8I8TV`$;uSY@#T9l
z&*`t*>y2BmKNxt1t0%AV#(&MJi>{a<_28K(Nu=Yes#&ziodPo7<8XwXj<5u6^$t%F
z9oU$s7@4LC2Mv(jG|DqsgUz(lNFBIe#x5i2Vl(Iifm`NP5wo<{P=INbBq>}~zlFH!
z)~lYnT?*VfCn|=HUru>u@T=KhIgh$p*>AC9<fP~xPf*AO2(c0fa6WdzG2vZh@ab*H
z?J8amK@p?CfY~+|K8(d2F8+UbhyO=a<6FO^tK}(_uA{BaR0C}7x0Sl0RZVQceADE%
zrlq*;c<E~L+{*WZ{DJRII}~5iMauE$*_GSvHUAhS@p+tc+n$^<w~47q1Xm#6pE`2<
z1nl7#uNmJqV3z`ryBkGq%;QJGcT9z1WvY*IJRyrqw+!UWA)(&yzSdh=Row|9z+~Ti
z{mGCqUc1O4OxW8Z)kK|@oy08OWB9mk>U4a_Teog>U;Io&K(^ta0)0bvI-alw0zoSE
z^n?Me*bnxFer%Eeon^3{Lg3`cfS+#9GWxEusDa_t7_!fuQo)mLv>At3DV?@%XztUl
zW1Mi+%?^2Qot{X|<qGB^?V`JzLd%S!S>wr1X$wsbCh^|ENo6o}ES>uR<34NLEsW<#
zS1SK9^1O~AXcx9I1avCQjE0Q$6C&|(i0pf;OP)Yr@7*k)eQ@SqMsPuB0Y#QD7ILz8
zc9B`e)(YAx5e-)DYF%+ZbZ7%We$(Z5U);3C$#Fq?=F0Qr^RW#6VM|{64|li)#{8k6
zdZC;ld9g-V^fCiBVM2M?FY^b>+0E%Q*TT0*WvCdG#oaSQb=8CSn?kf88q?|H7mBYn
zQ#X}=ZC*1VMV^J2;%m1m7fC&(!KT!~W<`x5dV<YkaB#9JbU@Z-WQoo%MXvR~eH5WT
zX6In$($1@mB_O31-+54HhZK>p40I-S*2wPhW|Lp#<GAKo0sP%l{W7AZm?)2{Oj$_3
zXIQT<U7jRVcm$&H#ZDHC8*KI~LTAxOJ|DUhuJ^k$&2P!%M2Hkc;SF8!K;Xfdu&hbr
zbKA>lNlq?c_rA>ET|Z})77iLGl!tMy0WrM!K7HtJ#&FSQRp7irCBnllGP#<BsO857
zrzrLNoQQ6O+!gY=(t0Y(Uo|6QM5&X?cr4VJQ>Up@No_Y~3Bq7QB>-$>0mQp*SM_sV
z!&hm*G~+E{IXcT~LO!2*mA263U>bDsyk7Wq66sL5plwreV3FcPO}vWKR8#D;SE$gD
z-WNAr+QAdNVjj9%0ivFx94AOAovhUN_i<dya!1B57vxQE`S1}xSnDk|uWak;C^2G0
zw&_djD0@*YxLtZnx@so-{O)2ylw*kVW7=8Le*wQvyavJx3JI3JLP?&Ey&ek=^RkLv
zIwup9*vWs;1X2}@mo{R63M^e0ybuGem4>G_Kh+!L$f%cWMACDb9m2+M61))TqL9gp
zJZrWix~oXk#_py5nK$@!bfdmC8;LMH^qTw<UpU^*h1e-xMpzjQ3MC)6n<tBK!$_Gc
zd96+a)?VYZ*w>C|o5L)xNvyDnC9nPV>C*U#yPv8h63jZH^R9YfuZc9}ig7_}3tIxL
zEDi`>CHl9fXGdND9&-<zw8b9%4yg>wP9$+b$0<JNBgE%yo(^n;m9-=T=F$J3-T!P^
zP*)3Ahv_lZ0oV~s<3;a<H>=5T$-0^3LOsz_Sc1RA+*3`7Vd6EiNin>X@XU4bpU<BY
zvWl0k9(n7&JRDXoaNwlV8Cq#D07`76W(s7OW92s+r}@vSWy-}ruW5lc=xXgFVm^qU
zMcthcJ5~}$arWn)P~KR>6$^xk{OV0TU>TV0%LDTi{PpJ8?!Ov-QEIwnVX@W1g#PAD
zo1>4CW*z8e#_w`tcPR73C0l13`TjBpH?;l9pKeZ&5|$nC@KuW4$|*=Y0k7hDb^dIS
zX5tlx#0OWsPrW;b4_R4=U1SFIplQ?>=}|UN6sf5G_E$%;detVcnm8;ld;1taC+7$k
zPM>%lc3TMbkCDUe2+y$L(tBR<o;1V!EO68PIq6fMu}$MG8~(_V<N6B0DX#p5B4ZHQ
z;)1Se6Co>4!&pB}Gvg`%xdiTGs)6%(pcz*WI;c%j7Cg-ZS$=lkr+`Y7MdM}_kHQx$
z7(m{on#4E6AfQQQh$oy}G3;Wnv4O=sO;kNaZzzus@97O~`T=+o7MOWqbRk1lznN-X
zZ}a`Jq}Q$%oIrc!tOnoYU-OZq{wl8$NP50BkYFfL<mM6l-C=_iFM^>#kv^}^YF^8p
z*nF(pg+*}GGP5$0=U7Byh6?E*dkjLHVXwB%eqgt<T;V%{T)l(iM7nhAKR@f8lJ8)?
z*R<SAhsl#ov9Jk>7enV%iC58C>F*iDZ=CorrQht=nWJ$Qs8Q5c2JHYrv00g+3n_dA
zE)Pz++_Z8RaC6}p{I$qT9|ayy6%E4Rf#vBfdD1h=YfVtHK>@q-6#2p7hE$U~#x0fE
z#$!>$yKpqyy@0pkRU+RY0DcnE#KmgoEM&&yK4Cj_o`3w~!m__V`q7D?w*h&vg6`lz
z(OLa1Q)1i+LM_}^YqrWxJeQ<-$Wz9TWib*MY3O`6jZ4UnoxS5>GDP$$U+veJ)XGT_
zw`e-2{tXvls~b}BSu%IaG5oH)W>X8t`X;UYM}b86^o*EB;SvP9T~Vz2K)~eqTokXm
zVx8TmKae(V!6_Sci@%>S+%tnZlVi(7#4JyZVHZ=T^o1xhB0=!Vc29me=2o&)x6TeI
zU=2qIJ-6pmHh%7nHN_NqqhEiv-H6&=_SS9J8$9eyN9?f+dfDjAY(PlpRpinr0Iqn9
z<tI|E#z7PKg;dy@kAs`+3wdl3-)I$-Pbr^+gbOI%7K^tzsu7JO*aL<LhgOqr%tIVK
zMCDMW_E?B_n$X%FJTo7o5v)8gmKG3wE55My-tES>f5gU%OE@ZC<!yWwfLN4nZ_;Mz
z4a`@bR7(<N$*!$;?2vqfZovmkU2$m`z9hK`YFSI>{~f&k-<Qi(zbUeIvNF`*KaS#<
z8sV_6ixFP*c&o<;Qh#y(n|TL?ah`h1!T|(B++>8~)Cql6r^yxi6S|zBxo7gfu3+)F
zxiNTh{SC9N#Uo<AEtjlGU3P}XQjfc?ZN0_5Y5QLEy~X@oNkiLAGcpWD`qS(4eNh9M
z3YZMNUD*?Xx+E*dJ^7?}kI<33?JxM=B@*@AquZ{4*R6fnxhjq%J+;#rT}k>ysS1hz
zvu7}zScea^Y)AHx89FDs$v(02LCcjgMgc2FbIEM!{AXB?ze81hl=0>3omb0_Q_AX%
zQR91Hqk%!&vX^i1`WssC?vn{FNHY^hW`Q6<Eq@AQT%b?W6lv5|o*r<2g$R`tbjZzq
z|MtD(zrL1q$wSywsE@r5^>65izr&K(YW-9Mk*IyrYEG)FE}5UqOcAYPcmR~KG5&fZ
z8wp*wDCjcFiNA0ZHUd|)KbpXwtOd+v8?r%R#Zb-jZf%BxRk42Uj=-mTtLei9NbM;q
z%|Hjnt1BE!!wm!68;2*KMCt2Rw$fR+XP@pOx-Mzfibvg(Z*G^|$J!%#o3_ocy%sHk
z4u6$_nLF^e&qUQVc;K^%6F{kmEWa{^uE18y^|%b19bYEbrpnMMKBKZIOxT1*;b*UN
z4bf5PmfsW|=G6GjJl;-3-SJPbrmQ)6Z^N=ii$hholb&EyW`jMBr~3E}CFhvJ2L1x|
z$gi_g*N{<ewN(YEzgS~QdCzUjOX`w=i)B5%BQ8_`K+my?<n(@$z{+Yi1e^Vw(+X7>
zfgSA%{~yTBYwDyoS>@r1S)4&x|H9VZYU)J9sYC_y7U}yakmTYRBL1h*vdXKpf>Pwk
z5X3rZF;kui;LpGbY=`#N7lI}eM<3mn&U^O^m`Wr;^282774rh|MagpADbGxx$;#9Q
zy|o$b8i%HrWM&_uWh#9b6vl9|C%&eHBJQJ!E*!V12s@#EAuNNaxRro0eK%WCcT6(@
zBBO6O6-)Cy(3Ct9i}k&AV|Z*85XkiSDV&XO`}Aw*nub-iducN^mcCJRur#L<OfBm6
z4r8P2q8U@<JbR4R9A8Ge$NkzQI$367H=uf`x#FCC3Z5oWq;(k}l6rfou(D$+t|2YC
z!Vgg6j6h0DDodo5x7-PfcUb|r?zCRWsVXX3T1k=u6;2|_PsdPfNHSWuYUp?z><voj
zDMqOLFg;l1?|r6V(=umtUz|a}m5y<|Xh}C+in8PLRM%Tazk8rUo5^`N!!-7t+5Sp{
zLVt*wu);A9*ApIRLs+S{S}u=!S+?&3M7D!4aDI&t_eX?+3?Q*vneggotmosP(FpCn
z+E`bc2B9XqG#6-%(Tlqif<8hlsgb&JQ9S^R)CKZD9g(^7(Gxyhj>Sn~-TA2mkQY8t
zv|g)<&##CtDzc2^=)_f5SuWPsODM4e=G_zB&AujMqA>_HJwPxxCUDl^draAf%ish7
zef$o_Y}t7BLeOpjhNP~N&m9#`TD+TQFXiE(UgtkUPss#!F4+0--P+wT+(O%E;yx!B
zXzPUxo88}!?vBUgifx_5%`$CrzcZTj>?mSS9NdemyHV&ZX4I}6OCzAU7l^xmAg(O3
z_fch##MuDqQE<7&*&NTm!30+HHNeXDvJ=C?N0|AfugUrls(x-PS^|<*9j`pxQoAc5
zbvmzy^xjT3<nUHuH%#1W4#k#LBuxqOYph(FFB&$IkT4N25PFMsf?87m0mlF(U|#O{
ztFu?<eK(&Bu&LS|;#+AWBzwgxa6alGz-FNiUuCAqNS18AwHZ!)!c+vg#DMwqCHK~2
zu@mYc#Wvu97zGw+L}#D~u@Xz@i5cxR^2N{x6+f*w+?Ukfh}XP`c%~DXGYGEX0VAFz
z27j8HjNlGAdH=rcXvC3Wnpk!^tpMMX|H0n){gVOVG(y=gdf(xrYNN4u_u#$_N={j|
zfk|o!5VbCqqznglIrR7<n?*|RYXs@I$lZd}3Sf+ZJlGrt*v-|Z&mC?G)lg8Pwv16r
zqA*f0uH#VVRBw)w)XIE@)pMkZS2jE67F{{k$1T3JWXl}EcRfN5tK-O;VY{37&){vf
zt|7xNS5{g-WZq?lYZk>X*SDsrH4ucdRLP7lHY1wUd?z@cmc2<6gJ-kSyT?p!pyuDC
zFwO!J8MsK<tegIh#hL!02O3N1WlfK(sM~)}FH~yMt*BFFOBKDX@yz2Rm6mGx#ZNew
zpKrcczTN8Za*~T3^xL-O2Ae?EsG7&~ns&0sbqY7|O=aF?7Fy;iZ|gpB%k7AopBtwq
zcih*uTI#nST#(2kY7I+!GV1(9wS9&3O(Lqp9UK~=Tb+6p1SbB^g=voG+#m;MdK97j
z*>DCNR7gsyt-tlK+ibT~vta(_#XApP`u31QcS2;K`;e+%3)7bzu`4T|e!VLz(lh7l
zBw!L!xFP1Nj%`Vhuzr893tg;_7$>_{YohQ2mqvt{o<COFM%sKGfU~u;a|{D--lS(4
zD9#s7%=+J1Yoa?@6(RjSs>)|HzlONVjvrZtuXu=JWH@uGT@S-{Dt#UXF)M&4bd|+=
z3J923@WkT__GQJXd^F<`Yb67ltlGv@uZLYAl)mBGG#@<zXLD9*ui4`f2c4N4dlzi+
zMEFaSVe8xOhh5+Nl(GEFCu98xkq!Uqo!+})@>@Rv6@I>c<vZ=YZV!^;jaJQ+U4be!
zc<7~Y?bScJe}zWJ9Yj@DxHT&_8bpPTafsbvnd#vh;9v71{?ii1IvH(?%@da)fKi(o
z{D)prD5_U;L>gE-JpbR?5d!CfX4fdZMO(QGRntTYbRU)w)nCY4zj*$sLHl-RRi|QB
zU!55zSZe6#qZ=Hw0gMVAN>*`!iYG|`$x_dcI^mVUzj^Cbm-tI6F1>j1;Phs?+@SDk
zG{z_AVpC^@1Z5#kQI1N>&|8zU7KZaZJ3kdx^<|V|`3EoX&H~Y{_3p2?8Tm<JJ@xLi
z3_z@tnT7RCF3oRViEqdFJ4_BM1nQIhf`hxIZC~h=lZym6{&~q={IVY(obN)n$v1^K
zE{p4dcg8!ESY{5$_+saWAq1x&6K>blDSf}n$m5>z$btHsm0->Q^N_<FK&HWAjDvB?
zYwj9q*D2)>y*b9}x;l5Hw~7z>N~EA>gP9{M85_C}>|7}=m?8R=8XG|)W3c32X0kj9
zEK+!pp{t@NzYKUEyjWmtV2%lhDN9F+2$~<)li$Z)=5Kwm2(P*ektwAD+HpVfOhNaB
zcjV>vcry=;vS*|TdWxb?rGR-;A08eR?X!kBkLQLPiMFf3ckI;ZH=q<&nQsAj&1=NZ
z-N<jCp7ms(^|w*!dnJ`vw>wxu`;$Dw7HbL8ay2r_C>z=Ro%rg3zpG7)7AdDB)^*gt
zB!f|NI0O-`P8gm-j^9Frdh_?w(&VVY^5i^VjSYeiMJSleC?OoVzH-?_3P%&?rvj!4
z?bu@z)Nf1p<lHriKrTD*2>u;e@U3G;m;i<01CtL7(WH5=Lb(=<jN$O`d=nQ{afYHl
zQ><Ds-f~0sY&Q7n!rpk<n4$BBc;^5iyJ<Z>-s+yX|7VYb^_1cl;uA3U^Rinw8!x6Y
zux%)rTS$>~za;K`<==bl)_2NWea%~%4eH)}pqSvS+WmV@qUqo*4qu8l*j}UbQmO>;
zybtlHhtqK;tu(Jw`VBRIE}YPArWJ9I7J#Q6EseBPR2}X_t3oFhIl+^}j4P(@7F^sS
zDr3ZE#T(9DH-&Fn@#I{(P3XO*78+g%HAF`k8hD=7YLqob>n2yl85G|4N7=S8ZgWH0
zi`lIqb`r$yD7O&khp5XWt(;id1dufTILgi&A^GdWR^d1Fn}RDR+?$Vh;0<<sa@tp0
zhyLwt*~pjG9XaIbUz^v*-qprxm;OU-ws;1q)3LJ$>1ui9MBTMZltfxqbcJ<v^oTSb
zGAb}QTjAqg^+_M;Qbw@<%8Q1fBBQ`hz_ZPlYKUtez>i0?4!Ht%Y2@tkgvNa^=vSz|
z+*>=oE*!`{je1+LSRqkvf|qr(fjASQwRyc%)a+RGo{yNvBVeHK<1ZhISGjq5J^9pa
zowYGzY?74I^mDu<GD<){b@t>-1W`j6aVYEW+1hv5m5r^)@$KLBo*$05JTbNVeJ@OV
zyUKH&F4a{t$M?C){Ko=yzRqU{sHXi{GX#&51ct+Z&O>0v45k{Z2Ho>&xlq=4n@&ob
zD%G_80vi#4GcMY+)&Az$i(0#J&bXybmP|a}1bi=<SOk<JstiSf%&}CE_l@Hwwi}-e
zPCJSohOdhgMI{6J`HH<%9sBMG6uE&HmA-z?4)Rp7VcE94&?7)WUf1j{=>HS^dy83C
zGii5!K4atotMK>M8$yvY!_LpN?d@!Lr*C@=S6fZ9;#Y6Q&4&1sCJNCCH(lo$Qsu85
zEk^7w{$|f;<DAhnCqf}Yz|gNC-cRwbzdke-x1(n3I`p`_?L|g7ebly0Ya!P`CxSfl
zKaliC?^aFm`=0xXu&vyX4N;Ezdx8ckUmGR}|6-GF<$0W(BHzH}mtowb%FjS!1RW*;
z@R=e|%O=BY1kJ^#`CnP#{HUNPkh}WQIzW$t0*eDg4O~8b&$%(3)rX34bD?QGBVbBs
z(eg7_y*5h036S(MW>fh&wGYs@N$IcPRW2Br?-T2VHK+H@+P#k{W>|SKl5)U4j$G#|
z(G*DdGGM(#u~xgB7aelq((2>!*Lh4Kh4(H~ufmN|zUIJwf6)Kk{#f*NS<{o~fau}G
zX4-CQ8?8>qCAMV(hz+RE+~s<&0nMPy4fJ$+aARv{IeJ#sMqH;p1UunC<VDT{;|y7Z
z!!utxQmQky*h{Nemnu2oq11m1o=;F5D$`kBeD8fUOs6$>M#_eD{<rVxc`sLm==e*W
zAA0qV*kiJ^fRU-CA}4o--mT(<^(7uFqt3{;f8KAsG;-vN#`c2<w>aJQXJL8IoeTc}
zA!=s6&4t)HS};tw^yD#w;F@5;DR&bZAw;xJL7GVqhByoLyB&7Le=n?-4`W`1NK7pt
zAyRgjIG=?LfV%$y_1?2fYGid=v>VnbSsq_CKN8^YiZ@O$u%&FIjRad<Q7$spYbV`n
zeB6(eAAcKw>3?qnH6|`KaIs;DQr(WD?0<t#zVZ5p6Jm;@&6RZ;R$yG1Y^^6U{N%NT
z8IzEd1}=e^f1@JSpV=vVD~55AGIjNqteCso9^QOI@v$a^W0cM6i$L}NH2M^BARk5P
zyK(ZPtE3Z!+1>lfz~NckhnTK~j8_;gMrM2ge=d`=wEDh8m<!dYH!!WSQ8gaQ@1BZ=
zbL=%XaCXqibmF}378;+;EpmK;A^3U%RsT${{65jcO~3xO9kDE$e&}k6{eGFTID0hI
zyCMeYkkYV<9S(0@0D?H(tkHOjLBw8cGEMV5tAnG_ORte(H_yA|KNp5TQy|AF(|r4l
z_1HkayV3K+U44ZRC7tHGlJP({?ZM7uRH_KPvZHtAQAna&sroxJMB<OX)+2~Pdx!=T
zeYrQS5qKi?y}Kn}(KUV55u+*O3&XLMd1r2LzF4Bs&->x@UJ3WCu(Bp$(dnGKdhqEA
zZ6gnLz%UO^2KF3(&MUy4Ba<2o0n98OMU>qdq^PiqP{x=tucvuq=s@CN<9o<g74BJi
zH4~In0-{?ySZ`jKk~!f}0-#%~^!Sh#j$8c625vgrYO2Egtcgp;kcdFBzC*^@S>h}7
z#G67ya(Rq7xOyORhCW_Psu5Ieyo;b+h0%kriuShFg<iyIvC|$APAEO=MQOg)uO2ds
zdL?bwBxDR^Lu(i{;_vY>?eB!zhRS{bs;5j}bJW|@c<p@P>|9r&Ij~+L@W$q88+23{
z=eNM>Q1_i|DhAs4^E#b|gFNCFt<fY;<Gi(`g`VlOIIfwa{@^y1TSh&a4YFtnKLg`v
z*lX^P*iwp7{8Btu#vETIidn5Eg)3Q4NwH{nZ*v&s1_mtyZ_)Zn4d_>0y{=BO%x%Cv
zS7{@*T$EDe*d2Hr$Z|PQt8nQFm7$B5(YUkk>*IP9qeTyz=$kJdU~?qx9Cs}S5bdE2
zEB?qWgS8)C=FKvd#Pnyxe~%yC)NN{>xr<hni09!?rd@L;zmAZj_@&F}I~R;JwJCVt
zVhQ8Qtl3%1;kVlr6I$sp(-YX&X3nAM>%`D!_RF^=zG#K|6j$%Ns!!UB$(2;?BN~nz
z@*J1Ge59<!Pr21V)4><#YOc{i3J#->CTpzNHh+!VR$ibb>bSgK$k^474T!zuWJ<_(
zg*LFVgD5%lP8unnUKa)sWwL$ynsP7c)T~PDb590R9`0anf66bcnH#K8l<b<@-Gy?V
z9QW(h;J|)Ay=LINQ=reDwQ#<b@Q$(f5;qL<ao!=T<i>h6ah<TlT>Z;+U8un#_0fp(
z=YjOC?G0tog-UPWcX;~WWZ&sz&pS)SQm61Mub<JVaMq*y(8tcV?)_JKyv6;y{Kt2<
z##VnrpE9a;M{izG5?0%qGuzW?#zbuPNRk~Jjo%0R`NT<@dTrm-D%fWS;z==je%NjB
z2Lo^`ZlatSPtB5t5MeeS2}$5puU}49veXJowSsMixC!)YqXu=efXS#4HKRT+sVsI#
zjw9<7lXRMi+RYYkcjx-f3>m23GKpe2X7&_I3YeY3^Uf#M`K9QP4>Qx@bbB;9Ki_7*
zimT>Ck6Q>WqS7$t#TrPKU}RnfSAipyOdot+tZ{PBys4N++WJ^h3BBfocy<Oym|djT
z+YWMlbMks|Pu>$0G`KA{9xV@~;m+BG=A{g>9x{8TCtxM!L0olvo@B{~u(C$yT2y!s
zG1^!ECo;{yV-~$r(3D@yUmIF$Y`|5;H+6w}VhK!p)t*(}zmb4kOj@n`70pfV+k2N+
zZJDxf(4?CUN5#a68Z{LrU30QX>)vQ`i-9Ncq&d8?0VbvtxZV;DFP~me5NbXtlzvu4
zVQ|E?ahD0VO02W3ra0OmPjXCr5*yXG%4e*i(!AX_k@ip#6n^kE#p0F<R6Mv!l`>?8
z)NSI5S+wsK5B`_%;#=I8i@|bd%*@%D=}$-(wy91B9JEKkk9kYxi$jewn6tp?2-#u5
z?mSirsbHP1{s-p2PzYBSl`scN-}v`_c|aAZ%(j881t>y+T@ILUJ^uF&l`d6+la8A*
z$=#x*J{J2AVSrUUWDkETwU!X_THlu;`9F}}&<Rm%mPz4H6Nz2E*&d4gbX^Hn9{A!T
zmHz7O{mI4f*;W@BThJcv>{60zHd~|HOwKC4=G6MAd3sEHKFODvT?hdj5EA#w{gz>x
zTi5Tn426$RMfeyUCbk=AzTA%fC5q9|G~}B8m<!CmKjRD5)|8_9e65j8{LkKStXj}X
zCVr1Y6No-9mg4C~sPieQA+D#$#t5SFl)43wFXRsZ_<Rp95M%%?%L=QxS4_Ojby*2N
zqxroEU=YlgTxAH)b`dK{W=-I_PQzzB#t?l(rwg!m>#?Ou-1+(FiIddhpb(+jGQ&7M
zw4rUDDw+IeCm2EvCjf0JVR7t-ff^xG+0{y*@L!i^lL-ifaCUjqCd0uk8Vhk@Z4Nsn
zpO1ADm+I`wzp<YQgwcbv?53;to^=+_cB7vRdu9?Nn_BPMmW)Yinb%>bV+tB>JFDL|
zHjejA+^jEta{WesM73Lm|7+BJ&w{QE<pp%;H_*tb_eXBg7Cntdr<6P{UENZ<Q?e#H
zJG1>#zxLeih<+ozs;|72dG!K6v+s_v{x#mV-JK(t54lC1hJLhOMPB+ReJzpxtW6eY
zN(qu{*aIE@>|1sczF>uu7O27wTIBancd4{SElnwlE6(88`&kbp0Yt!IrYxb;#Eg>H
z9jfWz@X_58Z@3KW6EDh8rZ*r)FXM1mRe0omwG}tqS$c8n<CI1LJJZ6?_Hp`otN)5Y
zBJtUgFYeBEb-xO!?%#>W=HE3r@|BHxE8Kh0aBP+{Oc5y?$!HXvmxqmI=ro~ftBSD)
zDw*%jy9CtF76&SM&!*Nff>NB5Epax+1^X4OWZ6p%5tyB`D#46_N&UQ<ir4^HQir1E
zEl`Tf-<vGTsT7*}7v=>H6#c7rtIIJmhu65;K*cGucaoInl4@1!gY@X5CxtCM@B(@3
zebNtdo;R3_(^oAOOrmo=KXhydAH2n+njoCzM82Nz$0KR~s6W$1i1b`cj7`6*3kj1g
zZF#|~BcSIT@|o48HjbmUA@qr`Ld`~KlOBzNtUX-Bu~J-8fGyRv*gLmy1H=aRwwU%#
zrWZ46OF`R^$v=yd21~`lr<uR8YGHDObaCNR{fm){OM}W)7g?cqyNoMNzmz|-vK$v-
zdHXOalnV|mIDNrYU;D1^w^-!h@bhU6XxzY_Mo+XRUvF#D%~g282>B>~bLkUCN*?KH
zEH#6ZSVi+JZ_xoYrT_rV0bJ@<q0mH#wn54{FV#}t=V~eV^vUlj5+2==dA<sb34}Ta
zVS2|MW5#GQHPtrFPF7z3q6HxsopgtAWs?lHTTz}nLroEO0X&^D2A7&?X8?i^O>?dF
z>fEIZu1QpTbt^@h5?%hN(Lbrki>D;s6zo_PbG5`jN_W(3H`O+WC`nnn^OQnMwtv&z
zg6fN>oDlesFAHJW35-e|NNYX~BcLi_?yf$>OlezB7r%c5++E8*I}2oHMfC)Rw}A5&
zT#aS)ujvH%;*MJlQ`5P<z+45i2a)~PHe{J*U;lf)Pxm+R{+*yV7#p88QG17;W1oNB
z$&d0JClLs=TmWXhulhik&|r0MrTQhiPS7cMjQ{pu|DnY|XJDKEb=#al_B@48nk$Pg
zwKVij*HM8VNsH%u@c<OMbL;fA(UCLy^HNN2pXbfW_@}`>{ki#{m&azD5S&m@>Pzp3
zQ0Lbook*g4DgM78W%DTstWaQ8nNFivw=j!?g1$`ceU=1N!U8F%;s^YY9iNi)C$)Jm
z=7N?~U-7<pzg37PrFJ5qZ6f=Oy8?z+@vW7|Cgcw|+HJH3Oh|@Kr`|A6>*=cn)&fgv
zbOsH8_eJY<8u*LlYjD0mFVxlNQ<7%_ibj5%GFvVUds6eK*PsIgG}9ZKYJ`Dd-mpfV
zyISL@MNMcrBpBw4T0Qg@r`MuTtL9D~lDg^<n;&8Hl6|JV)Edgl$dP%m%o#f@z?c2Q
zeZxSe#b;-jm#VELW6>1J$Qd68Q;j3cJfa2gD_3aJ&#5Ki1AX1dlA>hd51i)BkWeSY
z%u%8nG~DFWdtJ4Ku;$1h(f=M$O8O7iG5n1==1K0uw8Z17jW@F(#d=*5P|A31Mjyk7
z9-^Me-fX*EUyGWV(_Q|i(=W2IhTp5m@M>ycn^BfJ!P!gg?)PgF5P6I!`ITnviM%hK
zTlbF*Md(cZfJb{?srW$;qqxp18AuA1DDo93;|6QEMbJhs`Ej!RR;ypn>+!D~vNd|g
zV0h@B|2gx627CXHabnW*<NczUIb|cI;C~66f<zgW_Ib@G2*V(%K44(sqoT3BK4fKl
zBBSpnq~T-}UAv3V=!WQ2!tXbT83{l2QmvTh8fT`8e)rg?tTs1Qn3V=T3O7XqHFdiS
zMSG>kbYdMsufPzsh}W4YKKYl_CxJ=9-8nMWCms*pLS>_j6R8~^a}4MNQ!m!_)1ICa
z#s)s`!qsT0>IknzCs~*INYuu@-y@TxJ1*T}QSVn9UgtX5N-Zf6{Ni_MuOoz>I_%Ao
zdJ9WtDaEpy@gp|#RrwmG4=gCY#8d`B+K@)EC5%+ScwX6#Z#*x2&lar!7eucG|64pS
z5f?9Rc_e9}Y{OHXp$q9tc|n#Ip8j{mJjfjrXgW9&*BCa{)#YF}ZDj6XDS9uZmN|@#
zq%F6uh=1`pdo>?I1R9?q!O~PTVDU|dt>ea<?sq^}gj<Nbrx2Y11sjOKJ|mf<$Nrn3
z;o>b;P3hZ=AiX0VXj+pX#1vL-BsS26qJU2(>fBKZuFy*Uf@O^}u$r+v&x_OPTqBSL
z#!dDIuf1+rYN+PIOP~|UsFha)1LEEA$Gevi%|U7f|2m%;*0V`M=7ud8yP`avan7ng
z{;4>lU_D;lu^;w=IeD!qBxmBoI2*9L*^-DCvV>QA)+3wX0$SiTa&`<tL)jRfJY!!9
z-su2gr-A<!BVezC`~9PUE-O0Moi%ZHJL|ezTa8qs1+#TYd(mzDj_J>z?y!#WfTu}3
z`qhuU343;A@oP7K-8eD5z(r?fG_d5m>I$V-o_1JMuS9p-zToE=fzC!s%szVQw(TX~
zJ^4+!Qx%#r^I}*%CFI-?qs7;5b}L!bc|(eD%i1pm6D=grX6U>_xoRw9aon$MXn`U1
zW~Q?8vFSyjyU8P=76$o(V5+Q@tPj0p)LArbmMwzj$b1OG|LbPrerla_PsVFOHbhN1
z-nZ0au}py8cG~An^0di3*-`Xn5}E|Od3EeoduTqbr&g!!MCb_^;QzFx4Of<5U3kxq
z$_tszF?RUg6is}gz_Je$=RUN<7*m%N@`21=@?plIrc#~dvWd2+1`*8hAO>Ph+=4wv
zC_H?QgcFu|q{)Hhlk$KyNdTTctqyF4_^^64f*OTXAsv7QslLI>2G}`6o^O0n%UTnZ
z&=y~|t}U+NtX`+QP{VG@dykm5YZVxas0VLW3|)HG5*ojU0FU~@F@Mj%cogxovf@2i
zQIvwpe8;#cwI-_Kh%&WnxIznMyMasak&DoIRj7wy0`4dAm<UG!KU27W-r9Uk5+2=^
z!~5)l2^)39=<JB1GI8#JZxieAwIMw@IW_={<&)RWa+y6o>J>a3!8kVJwfPA78fk%w
zsn@~X!ajS6!1|3{SZ{e@E$Vaeu0hx2OUy}x_*;**nz;;#oC0voG-{1JL{~2opkHR2
z6#gROYNI&RrhAqZ)?RrpZJn0mEU7sZoo=;Gh<|QF*f=dT2{%IOF%CJdbl5ANUcoO5
zm&L;L8Un><=!X1qEl<~d=(BVlCRhAU-wXQRF=lxOt@H@2)@1KgbU>5-b8ly<jc^?n
zCt5RXGo!m~W=x*FAb0wrA4BCzOJ7SXZoYz0B@q!LTXoXqpHCk%F{4i_2W)oa?z+9~
zEA6(t(J=;g$p(SEZgdzp_kM9-M;M8rDv_r-2I^*>{lzAN&mhIk=YBmTs0u2jS}e|d
zhz3WkDbwE?s_*`*9Pi<AOJ+s^q9{>x%-+ZPsWw*~z6gj*-M~5;x(G2vtiIN@H#-zD
zATl`T&|yPfn;$703%iA=6>w(Xsz31EW6?Y;em*>&V!lE1&mVtbAn+W~t)XAaQfi0}
zOj0eC<Vr4~wgcfyU~lC+_ov3+c@-vMoy25|?SeuaR|F$xEVrrN3ch_zY9+{gy%vGu
z^8^24F$N_CkwPx^?XqSRGiDA^Mtt7a(ZE52@mQa7Cm2{nLp4mD=`&I%JlyQlY;vtV
z74uo*DOvE9IjvGxB&tz(aaR8CQ3<;h1W3Ie4q{V0t+t94jqF~sqP^pIC60LtG1ouU
zuO=(!;~nxdEZ*3t*4ALQ-{3!1{Yw4Y<qD+Eq-+Cy6_4+VZNPlHaY}wYuI7_|*Y@^X
zcY-E<13%vxrd=yw2y_(7vjp65R2$N8T<RUq?BE%+;Qx=Y0P2N$X1ZIb@a@w+f8{*+
zA4qiWdY5?0Kf%TSbSvRS&_n^vxUnDZf@G!OV(jbflP2vz**z>7c~APk@+0jWAi$d?
z3I)|w_wFP~VnEoETZtj7HGzH^&HR=<yEB0*eHY^K7jpv-SB%Xv-E0u#KkUc88bOb~
zc3=8~(eHNZ-|?;dR}~t37d1A@%$t+aL;?#IPCver)N!^y@-(JnBkl?!y1l1Ch*TqT
zd~>%}*XsG{N#9)6eMX|JZgLdXEoo-%)vGJYbX;KG`<Tt=Z+#tGEoBb0du~n^SEn-W
zY(|RK`lzFS%?Ejw$*b)6>5r})+ZkvyjcTbt`t^N2Mi|+wDqiJ8>h^`HXF{$-8}~L9
z8>tVyvffc*?PdW`>u|j%TyYqo(0Go-=Y-Jbpl@f%*uY#2<hU)sjXf@=tq-g$jWIU)
zkSwF@75^~+wzyrk?%}C^m{DU(jRA@;7A#@F_G3a;0`T_cpY-_HCKMu=>QGND#e{15
z6l=SdrkNqrS+fl#z6C7}XZb_Hzp#LN2`GO-_|{Wy+T%l_OZcwXT-+)T!moF=-yT0_
z^nfjv{xbUQd+cb%c-sq@YG$j$H%Kq-)THfOoY=#P7DmDly~t2|_o6k$t-H2m`NU#&
zwEOI}T5CNXL|yUXSB7b0w-CGVTAO4E^Y{2b2y)^p^n+tGKrWDOnw-2ky%D8ImN3L9
zu;3_yJBe>1exMmfc&7U@{~@z7f0FH)chX4b7VPJ+=&hgHj~-HZt`;GXJ*hY7QwT)Y
zckf2Hj~vu`oPHSh5n788s+#bVfi^;zVF&u>hJ7zxf;gYTUx+(5(!|~fVEa}*h0azR
zI0PQ3-!qqI_pxLo9xoKf?AJX$8na&?v)>T361~6AFgd|F3=_Wsmh85}<8|q?hjVf$
z-myn!E3SL0k6e&=5@r+Fw~x4lcJHHQY3;K~!mCr-fwz&QC6}BFPhRV5s~95s2=X|O
zg?e0S`)<U&bi-J7)2JXJu|vatsngX4R*Vk=s|pG@BKHOZ3X<^RKO6ok9%N_M>=}}$
z^H(GzuATN0^YKbroHelh&1x~_22aSs(`V&NhIj)x^tuq|j=Ib+2F`b-!zLb}xXJ>I
z>EyXS0bo2$sxY*pJeZzzz*Y~+-sYXgldT6K6RT<4?faZJ)`y$h(Nw!r_-%_fiJFR$
z0<*ARYzvKNw*GP6k?;JZ%I)sZi&JJ9q*6(_`uJj-#aVOar5c+g&m$uuB41o2jE}-A
z7{)+2ljLy0OXs6!y65$M=S=y9Hj~29^?J&@0_^E{BNZ*?Jb8Uz346@$sPdc$Te~1{
zjrWoTt9|;#$YCg1X&DFyRWpkfg5?IA?YI0D%RDfH1ulhgri;^jt1S*EYr%ZW-{ii*
zvA&jn&GdL4vkxF<6<O=AhZ~LieDOW?@mH$OuQRvJq%DjuL(YYiHHFsZz9oE>D0d_a
z92anf*eL@A2>$(hp*E}?njw^&{b!_;Y~>pAg?9}lP(-d(it)?68tV4S>h1z9lcYZ^
zmR0iT6Y>W=XR%X;B#jSSiKTZll+T`IK6bM4yVdgO)w^p3uAQM&K)cZ8I%J-D<o*g>
zZVi{nN6HnsX;~oKc+V>ZDc!2Gul2?6Q%!-k?Q$KObh)8PzRR88<v+|lR8qBIkjvTj
zFpfqJ%{o2M)c?%Oyx4uFUFOa~tOaPRQn~}=0ki=-^z>@PZ{z)Y1}NzfeaW1%;j+>=
zJJE%lL^b}pP|YI*&6pnxz6eI5`Yp~;qV_YKhx3<ikdH^lsf$KMM;>6h70uKv@;r>3
zuz8hBmR^@LG2S+YdW{R{8A*>-_R48L2t!&%NuFuZtQM-9bjU(KP+!O2Nz}uKo^h&B
z3^hh6Mx!UIHn}!Q&pi{3%ht$U!`l2QqJmbUpg8I|X3EVeLF1}vP`wWAnR$&M;kF=Z
zdMAlK<v62jkE?C*?f>(wNv~6(S#Mt4VTZg2RPf>px{_1L_l)r2!IUETZ+GZ@(u@Py
z6Lu?sbLTKcuet;FOm|!qh%Tv~-+iuP6BW&$ZoWwR_KKdQDi|>>ZEOKiQdZrUvD9m~
zzo<5smwwQ)w497u?RprJG<$sL<7wutD-269dJMb*h(}4NDN86Y`L{SrzRnjP3jiF*
z=1%=?i22HEql@E#pi;pHl+}ULuR-2+so^=;=E{4rU@vo8EVI(L3mY8d_feGUr?;u#
z&R6TW=@>1!sbI7!`$bYu1&Rpy`?hm{v}>QeZ(vaNUdzVmjO-m?B~g$C{?Z(*#ZpC!
z+lKTV!jmZ5YY{#}KMEKBx%-<e@t@vC@Rri=8aEDnN0>L?bV)Mk^vbn2kD`+rz{&&B
zD|tApS1U6Usp~CS7T^~wF()ppBE76oe7-xbUUUU_NF()GFK-ktdOcFqQ0P{M$jROI
zw<!U`PC5;LUv>-n{bubmK|xed@-&uoC~i6ZpGTow_mGhYK~nndrRGL}K|I<e{W#8`
z;OfdtS%DFCM6vuVnK3O8j}AeL06HrQkt|)qqf!h#H0YNRByN#t5?ww^;4Hn#L|)cw
z8|a}@7<-3;&gTuSrHrrlS#AzjaP&Sv<TB(KXD-oqEN(i_Z3tuTUho;)Z9HosusP6b
zwjYiqKvW&$E0hqbpR2GhOI))ynijvHf^XjQcmx*7zXLhv#4MD~F?BtHPE=OS=e4Mt
zVvMMue0uW)rH~qZ1wjBn9|KbkMjqXiJRhe8=BVBwpX9H#V+d{OQyZS}CnVbube2Z4
zJ@5U65SLd%459<>lwz$ULs+O#mNhEU?9t;$HYt4CFp<f$Tl$VszZsN<H+*qWd+l<K
zgtf-Ucd8mW|L8N`UD!NVv|Xzr(wrlssS7D>z4NW)?EbN-s`R@S+7`-QSKKBe%t)DO
z)VaBT^pPS8e<@yOYsJ7n+sM6_ATV=)t&H6a5C&1N);;r2^QnMG70}YGiQCrP)2Ta=
z=1SKv*uLQQcjq7e;zKj-b)+5iE}v`U&0p~ov(?5uLtJ$}WA#7K2X~Raz#Fq!(+ixi
zr*>3=&1qv3-#-)Ls*VbUsss46R1?`JLt*UL%?6!%J)5$o>puT|R(bKj>`=e|d1&E}
z&xGL4!s)I)hH-;39aMV_dTH^<L^pCU7$CfHa)<|!Ml~q#m-&a?A(%9x-!nYpEB|{)
z`B0@nWL`TnGyY3o5q5a4Mg<emF8Ee!<>&aC*N-RnO7c`l|BfwmN?k|*B_LW7)jyD)
zWaQ~J(M;Cn6tuV*jIPKpA5judZ|nybT?f;jvK+x5`vK{W&(VU?xI=DK1G#*uuSy{;
z2bNWIfCe^x1KCu*F)w4v0gDvDu+%BX_M`Cb{9IAs`5y1?@0>A!h$R%QX2@;TBc_8#
zivb5JE_GU9pkV}Lq@U*wawUUmgjMD0M=>CCRU6hAe?v(|BPWa-j4ll_kc-YK;f>FW
z*9}MELnvj3UeD_xG*tD)Du(g|>{di~{$<<L%)>H`DZ{P+7fi6LSOt3Aac}h$HHsEg
zh#4bv44d#cDfI>OLMP6vcFUZcFldO>?z4&%8PVjS-ZJ*(8^P}u&>ym~TQrh85bZM-
zs*+Hz9RM^j!D+5S$%cN&Efn9I0LSVa<v_iWC=BJw<u@;PZ%y#f2JErQ*sjcmd=ZMN
zZ^!g(tliTzZ29|I*pf-$!e5`{E6vwA%{yKezKegG|6W2AoLM-=2?FTCJ_Vf72os`j
z-h4Vc=b5I!@QVKLh&#HF=SR{<{c`&A=j~+SVJV0EOB6%20hh;u6&<N|gc{iq8VlUa
zk10&;lp|MD>TG(MXAdOcN9^qfzem}HOc@JaYCA%&b2}*0d{->(B57a`8>+isVoRC4
zX3i`fSl^7-ehk}Z+DMG74_6FT(U*R<Yg4B3OOPTS&%tG!(q9oN_iiI=XA3|3wke^=
z;z96H;KSM8<p0I_mP$Gf3uT1@c^VBmDdz~`GF;|3d5N7XD~kY+>5+T1`_h`)l&?NN
z@z>!(>S{JSGW;Y#K9>5%qF~#h<VoE?{DI>?6D=0479Xa1-+9y6MvE5k!l!_m&C350
zykpdx12R^Otf$&|Ya+y(2br0kn;3IzSizCA4^xa!V2tdfhUUdNz!b(ap6gxAi2^t>
z2*>#_!=oIxdE44lt*lHIQRR&ZO<}{+@DTzA%Sv;_Q4e8wv98BVzMzN>iOjYddna&X
zHHjZJ<*0x8g;24PoX8hV1J6Fk18|*qmC9S1Yku26K^I@rtBVD?nwl}&s}GJj_Z_8M
z-(T7el-O7Ha;T6K!=HOJbyJsj;}p}y;?Nl*Tj=*@VD%6nC`SCX;hxS#!asYe56q3I
z87|+cZk<;*ZeXgDY5-V=z(^>*DAYkAYz_a;m~&rD{+-2XX1RvV$%K~m@3gIVr#*iF
z3&9fKnyi;r6xY@;^xW+Ay+2FG;8DXHuihOAA(YBM#a#3qwddaF2c3IBU_H|0XGI>G
zz%B@A6qg>W{2yq)*#tq}0_}%zLD9pha4!LQ7BXzq$Vi>4cjb_Wsp<nkxX{_%;p)Ag
zq%XSV|0JG{uoM0Da#ivbqKgf&QDGCYa@3TJ@SXVg0m=ISs~tU|Om9E1@!-vm7h{*L
zY2rNK&S-Gqd(5?+25M;T$sj+Yyw%MdNz5r&0V%{Fz`2Tgjkb1;Bv6y~+Xo_4t5;z9
z-e=#Smiox*4&i<cpO;#8*9n~t#|MY{qMAK#6?3EW{FJx(>)^xM0vr%Jl`#^ki4xLl
z3eC`K2;%!O@BBS!R<9z`J6hk=Prf~~sQgb+zeYG1C`8zi_)J~TVbmZwZ%}%QhZHDx
zfRs(Q#-CM3+XJV*-%`=9gTp~vmK??@7b!NbH*p4@#>%r%INy`S#*y39rc@c|x3_q?
z4M}FDO2$e38`L|>zdn(D3<(912P|UmuvQpz0YY8aUC6soQXeLv+{(T&=Rc2(j14dd
z(e5)bllPpA(<huFcbxKHK!3~yE1F_S8QkL}J$JG#0|u-de_oy!kMLhrG-K#-u3CJ3
zMPHR1%MbWs#wBftLnpg=)mci@Ihhj1y!KjVeaJ|t#*D0DUUxXJeWnk$(DFg1Rk0hL
zRd1Rgzbv2go;qY-rqj;qd*SLl+^Isi*-aeLm{9gA|Ndw75E~Iqgzh1@GERhdnWHi*
z=-Ftt&f2-k{#?P4moADa#SX43v#=2ztijXcD}1XAuT?<}niY8jeU$1rV`n_D<`waL
z_x-pvh71nzx_vD5U4>EJ2V<f|>qF_)&btp6|0dmJOksR0G^m5mTMteaWu*{s-oas@
z*BH`kDT!V5V|xE10g=BPGv@Z$KQ7iXLZ2mlsUE{R-l5o0l%)P_L||Fe{CxN!jMvLP
zR#-N8gU2Sril8jdIfrrJR$Ecz53OGSA-=r<9}ZewYkS6tUv3h!<eOGp%`sh^2>ODP
zyz1>cnNTfTdA<Dev9%l9gg7&Uf{4{{|I=VES{9*YH*lKT-b_hc)}4bd5nu9VNg{XS
zU8Z1Mklo}Fmz?pMP^X`J&#^+}<Ut+F+3zKXbGO1OAEu|P|K%z3kW!@8igDO)LD3W-
zKmp(barg~BSRorQ;yq2Cx7dV=|2@4<%S7^P#8#J3dBdpuWk9#VfVCQv^KGQ*9yS^0
zOFMCCZ){Q3I!u4&>7Z3cWPuzjODNfFZ{#P@^aW9%%lP#A-Cs%x>AjsH6Tb$Mywp*z
z-JbgP{}1#lW%Kl_@AD=e<Lc|pPY?3LCNxzjU#l0(uYN{FKF84uy=2UR5&z*D^Fn+w
zpvxwBpl$zevSv5`^vYrMJL|I6_9sj@(z4V&_u!utPoi3{!_Ve{gBo<|n;wb2w_X4_
z?n7D3-2Xs*_J{oDp8fFop}5epgo5taL23ObzA#LR#{(+LcS$x&;|dm_Ym{NzYsdZV
z8vhST=NXpt;<s_!Bez(JT5bhLieOq=C=T3!8*RA)9I2$GE%(5YxKcE8AOec1scAOM
zm3ttPm37QjX}h%3X`JVG{?8j<d0oQgk1yYG-=7<H<F}_bb0}@Ief=9dIx@AHIu*vM
z6I1c}aD2M!oC<DP&6l3|{CDk=v;532dWRAz?tQ_;P{mt<N=<e;6|K<jrzg-{t1>X`
zCJ9*-GlLE5*zW^o`eq}T<H#J4kW+Y21~*|vZnH`$IWFyGPp~52BYfdSPm{Uj`jN`T
zYILfR;#}~*pcYYqf3w-^=k(kUIKtD5i^YE<JmU9Vc2yiV?2e?xlz+Hc;|p5x9~;ko
zjCdBRl!j5%fs`hq0}vjT&RZ2ZantIo_7DfM)ZI<CPLN*_iP>ctCJP?80Lbz{nLI4u
z9oqrk6bx9K$Qu;%AbN#ymfqkhOdQ)6uqNlhs25tRIQl;LdR9a!@VN5R;Pe}=BL<+?
zIxX%BiCWZn;nLu{zlTPg=O$cUA~EN+E;OA;ZmD2;)@f|!kuAMm8rpUr-SC3#UoXg~
z_FT&y-(?lg#O{25VZ+H>dE)DzecaWLlW0T0JM)(^=VU`?9n{516wJIhX@D28hmIx5
z{&(dOw}w>7O3t)$mG}vvx9kmd>5If<IR8oa^_RC=>$vAe-Xx3%hSl&L=e4y97L}2w
zbpvS@tZ|l!FS(ujmSA71=eBiSQFS@8um8Do*G<(UQY*awQlzAUmZBrMk3*jH-OHCG
zPaSW;)O@kHhd(xtPJwJ#jwGSOq+$Uv3P&$SxKgg}A|7H`@aH4AG_H?<bIl`xR2oGu
zNmuydBJhkouIle{-U%f{!Ht4_7WaZhN1N#afs5Xa^%TFXhUWwZ*&>bFlK~RMqob*-
zXo%8b-cU*19UoUaH1IQd@4MFv)U~!IQiO@ONfeWAN!xRS6&2;A9-WXvuR(uzYiilU
zTv?R0nBC0_`Aq%#bg3f8!4p9*5eIlk10>ec9Kzu>we1_N?jWJui+IY7$q+Bmuiqj)
zWdaLq3MvCh*!a|2r$~#RJxO@Iqpx#bqWc`)pkZ$E%8p_iBjNVh!P2t36-pNp$9@@K
zIsd4Q-171;``Ecglgf{NH|{`69b?pl9qLnJIM=La35QgNtY8McBf{;oT}7MpGZOcO
zPS<^~Vapp&gZwMtHY#$fa<XtIdFOvb5%(QOxl#p;(<21S@Mvom0XS;bmWM9?c}I=L
zl&VCC`!^ME&4(c58<QU$6dtRt`vSlkWrqx(`mwsuc73veoh-<Ffs%pB*-=5b=wu_@
za%6qXfvyG9*=!W=qdg9`P=BfRFQ~Y}atPY;52f(fh`OA^MEnn+8+emj3z+C7sMAlJ
z&jsCn8!%TEpG&^?@Pfdj%N)`{-iSmm(226^h|eCh)mkHAM{6zWJi5=G;}FIMG#^lN
z08{-HFI?C%7Y98o>}~H6VOXrh+s_^R4eN-x_m7W$c|^j8K>oR56^i@M|H=zS@vPV8
z$4^{x-rg(?{vrG)9o)di{b~K)iKzthBm7Z&83V5^t6TgXq=RJx8u7{npUmZBso--e
zGGQYC=pYfCs=4bN_(z5Q&efocI;XmQPRS<nC+iTTNmqUH{r$WV3w0W+P`Jn(x>%};
zpM!UQs|~UK5T#M#8Ks#QR`UaQkv_k$`V<tYLC1>*GgxXgw$o<F!MgRu^VX}~!Ag60
zb+uj=9%@*6H}S$RVtC`Zqmx$$I}8k5i^w=fR@2^sxltp)YPi(*(ZN2UpztnW)pg7R
zurUVlC?!K%e!0H3(X_D))TmwD$;o0`(t|_FZcXdrK+|AUsc&`Sr9cJ&G#(`_W(rbk
z5-})+=xa#hrd=^9{4HY}$Z)ZAoC>XTf42_30tcg)dukb?{r5GwgTg#VN%gWu^5nCn
zUMog-B)Z+f+L+biedXr&VZiDst+)tkX7J_x^O|^1<1X1Zly}dUIzHm8+Zw@=%PWhl
zXQjvt6?R{%-ZkgyIY8xJ8g53iulwH^gRKMlkOANb*lqnGEcb6$+7=&}N^<Da!8VpJ
zy96k)7;(z9{IVpobR?ulTcK~n(UUSJ(2FL+$!a7J;@c&kd_Kgmus_4~qSEs7Vi(=}
zB*S|8{PKV_Zlbnsk#$W4AkTO%w`X&mp0ydkoloYCMyt}-C_v@#1yaLbjOYhx1BnRD
zA{b&B#P!;^fEI2=+glnx9p&5$cqIzikKHS@vl!7v)4^7Q`s3Sf)j$NIb7Yg%@Y2KN
zUg5h3q=_#bLzA74_aWykY3C1fAMh(z>%tFyjdOC#pYvr{h5nts&469$p0$_^0K%%K
zHhQ048;iO!oSY<S*P`J*C!q4vH|{PTH?_PnWoMk%c&q*aiOHBEF(KjzevF4?2wxU6
zFL|Hrw^}Of#1>*?K4v$I___m&r8*jA#HZIr3zyJvW;xDsD@r~9<Pm3_;vKAvtiN`6
z<u_U6IC1S_$<t-^;NH!XuWWxCw^iWuLY-_xBRe@gk>5_Dj|Pk*6*f|*!()l{$-0tq
z7!5zeqyX5mq66gg!<Y4C6{<5vijq0Cu1MRsf9PTf0{}cIWKR3q80QO<s3_qk#T15+
z-uh;au3lGD0TMWdsA>I+jaG)_3Xu$;dQ=_V@mBlcvVq5|C)7l8M2gC^;aHJdnq;dJ
zUBl)+Ytax;$B4t4c(J0Qrw&`v&zc<;ZT_vkP?=s~z=q1)^-vhP54ZY261r74qsv#=
zwK;Gj=0eq#S!C0@%g6rt=IF7-!^KC)O#ULv`U?8yH~$5l6Wow;QILa4U7HSq&#T(<
z=Svt61@+GdXx=P#7EU+I76Q6}ydv^Nn&?3<^OFP?s5sRAs3+9yx3_cCMw8E;_K2P>
z&p+w{iW9Rp{!9%o=J_c2O&`fPdhGQLEhnz|=DUAEccV_!SM&%!Qs{fnV;?fqTc;}w
zqozw5OX4j3KI)4Nldm|gTwG8*>ybmQ3T6ZCMzv}~4OAx4XxzF^@@z<~V$l$tu^4qK
z^u@!cCJrJwrKR6k-*Rm%&9`{~p^Y3O2$SKl#y3;pw>JJVo4cUm7hq!S+p1l6x#7J}
zXAONB*NxI-V-&g|X;t4ZDMi`!0jm)`jV!lD<M2f(+Tx`CGU|JM9)vQU+iq-4(NYoH
z*MR!vb0*`g78Wk576*&U2Ev84pz;I<a-Oss$kKC*Bn0~erC?6eJ$R~-1#`lY^<@hV
z;o_C#Oi*N#(@q__a>X}%P8jDQT#*XK^MnrbIp#(Xg>JjU_y`@}^~Fpe-s6;;n3=Ib
zp%2(+fE}U4bt}sbENZ>wRw1IX!h<rkS8J`-N?i+5?uD^cj4alYUQsm%hj^O@`srLt
z;nxz>3LOoBKbvxY*e%K}D!BAexUUG+IfS~LTipJLt$T$SNn5U|p~gl@?awl->uX{k
zvIklZ8A~s50AXbfagr0iahGGXIcz7T$hq*Bc<ZakCPsgX-=p|x-=G>A*(Tctv0DsT
zs7CKUj;XrsVbDoyQ1!IB9o=E+T1(P0$m{oQqSBw#%aX(Y8r>T+kjNaq(7IVzI@guW
zbtL1IEH8Qf3(|UhbaD6$^11sfn=`$!yN$?tA;5G1yaow6+Ht}X=QVOAB#|-cw5{5q
zx~6i)Xiq!Vrj2yuHs`SJ72XtjFJ^tw)D6|FX+fBCr*9%pWXURf$9Vf#-rY2s*|+?p
z&R!u;m^Gks?~Zw}T=8n9_awd8kqoEcEhX2^*ku{jo!NIaF7hsIe9hL+(Pr~*Lu<j(
zxZGPsJI{0R#XS>tE!UK0=*AA^Lp#kxU_G!nzLj{n^wbWM?lDXPa`0pjqs(m7&K`(&
zf?!a(7$2tRgBNj?DgyRMaHnZAQeN6IC<lLNiU={9am~M#GA?k|4CYJoZVqRe7DRPF
zI@#ohTrRSS)qWcc@!WlLWm0?n^1^Qz*|>q-kEum}g+!lutum9yBtQE$?V1>+G7-KJ
zavmH#93C<%nC*UWHXfT?o#mru60*}Gl7`0s^b0hux6V<fj_<<7EBJW35k2g^?MtZ)
zdzlPTYU?5~Tjxt0^z+S|B5lyTy@LwL|M1yEiU5DBSuJ1Z_EHQ$B=PI9B8NT*T?bT>
zW>7JZN8z)YHDin<hBa-j;KSdn)HxS@>%N^FEC^Hn=fY<{AfFaRBq@nK$T)w%x@|J#
z&Y{DK)|12ik6$_3SNu8{zgwtep7OYyy@LvN@7+%&k8Q<q$Z8LE>>W|SEo&Vwk~HBd
zc<Z0WQTPlf;y;mhS|ThwMGTvmm@dHejM`7e_?bi<^*`eJ$L5&5M|9eZ?mvE3Hi)rM
zU>@>+XYHQxXGFxQQ14$5^U%-k+c+J;Fy+_m?PIWiL5K6o<Zckry0l^3Ud@z)K>MNk
zEP3#sTB70us-kdF`i-ChoHsbz6ndE&%u6*I8>uXzINePy1zp+ta;3acvHGI!i7oYg
z%sAr&$zdE!Zb${55~yfK1nAjbFa7$?o+w-ERgE2~EVVzPeQ=vl62v>TGWCMc)nXHi
z0B7>lZd_RDQ1LDhjFogxenD|^0Ijp%_TER2DQM9Xed=M;_}nw5cuTi&c^`j+YYE=~
z?HTS%k`1{f?nks=JsgJiaDcdlN`on`>3WQ5R|lYk2+@s%WKo-y@Vy8jdg%T}P0~oa
zkuaR%`rl}1ZE7tIk_YoBD0#ofh}Af#tjsJ@%UZY;Vo<PH*2;_#P6WU2rg6a17gY4+
zad;L(^bPFNXV*04b`_Cj*mn+w6hgck3h!emO&dx|aXMk+l6zMkyd+(hr^uW!Ohe~V
zbndr)xtA(#D;U!=nk9@s#?n-V2i228<3rvQPg+ox1^jZ_?4h}?94bW)Hk_;B174XU
z=PmX&0U;_q`E!<lVwPd)CaawdOgUi=n`Qr1n9}n|oK3t8a7}<AaihW6#~ZnUL49>4
zd`e9sJG1Yj*E-_JRMTJdiqYaTh)h2C5TXmjiW_V$+pt$*B<eUF>9~6=-cOfJi(l%@
z<~oqE4BY2~ztY$L1ts<RA)mU3RWMh1%7kJ-(;~d1;ZvL<g2$T$X_sjfm78}fA8$Q%
zkq@80FJ8&23qv+Wg?IEMJvnmNCf}#MNXIkbiT=mfOvTq`SGAA(>V`j^#Vx-_6dpaI
z?pWwz%{A|L!c`lOC6EqtwW*E5tnGmPPBhRL(uj!lTkl&wm}cR-GPz-sClAIS7drdG
zXtzzn`So#?dN;tW^jj@n{eVx%Y*_unb3A5kS3$M~vlJyuE__>%w$-me_vmilKmZha
zK;M^v@rXkJk}fQX;z8@QQeqzoS4Pd_eq;(xaQ3ubP33BJ_cnE0J3mQhM)e%gNLVJl
zbtq|7a}_<u(Oc~9`{{7{p+N4zj(JS0Wyn|mGa>rtq6r$K7XFK>pX%u8)y6NMcICQ?
z94#+2z1<uNP5t}HS;F;%lsV9GiFi-=Enqo2)7Z;;mbe3!%$UVrK#0_ma%B+<4xw@|
z$DfOs66mPL;-PK-f_16yS*}FO<gVRmD?L)Ln7t<1{A%(wTu|q{h1xZhYV$`&`Sy7Q
z8q^ewPg9UyTm6%S#RRLHWN|xBH#sMFp`va@j)Sf(9_Vl!JMJLOk@`B}v9Lp`VVKat
zjmUyq+!W7z?rsI^DivP1Gh(53MYd+tEV??oaQsjC++WweNj>_|{KKcWHTlyv*7G~Z
z*%9W*5Fe!vj0W(!_R{~WEdv)?{F`fhgjayl(g3h6DX9I&;|6XpesSQ7m<Mm~97;Jg
zX109R=mxAC9{9iF;-%aB{p2==c+*Ag-qr6>@oHGh_8D=|cUHy5aE?gqMZSqw+~L}?
zxru%FoXaiW1JVM{jak)~=a??MMJryy3!+08Mnp7mAxBa6&$4<|JWVHmmwM|Rc<KI&
zGwd^e>Mvcd7ZSofTl&3P4UACMq>yj7Yqn$5oSzd}W(L=r=GiyZ$&ifSCMd<A(yje2
zLw8Dzpd~$gy#-Xb$MYI`RuP!aPp{#Y;vnH!o_pSD=a4}irVDPuwdE#JgDiFgMDVu9
z0(co+{(abv(!dW1n{9d>44z*Jkg=G_;~GdNkt{K%0R0#m2nehbHKeA^BIe*0<IO_U
zP)#}s0eWT$e0vC|YK|c8gDmxc#2?TKk&IRV2x|yWO@iA*<h-F<Nk{Y|sF{ie$SOt*
zc!VmrHLgpjJTXPd^WMUoZe3`CC$S>qexYRO$%D=0?lVekLvU4fbbH6Gs(zL0-1Z2I
zW|Ir`W_4BKH={O;^e6%Hgq*&ZO8T^5Z3Fs`;L;h@k#i##n{8{G#^_92AoM$H#4&|u
zg`Q{sXk{efWq6x)Z+d~GR558;Gm=HUJuq`HKy$E$q7ZwNL^%D78mtHAmjY_PM!gaK
zz0NV+jdl`gB885(M`~FC>{2!M+NHJyKM8K;n1LD|R&WW@7j0dEU97x@q)RD8;gWY_
z4zSvS&K2wrzSeYZ;RbLLe^CT6|MTM)d%Mvv$)LI=)X&Q8u49;>7Qu4C(Y-p{Dt8u6
z9~gu=ExdZ$9@?}BMJ4x8#eRUL5f?<ia}5CPwatyTy&VfCZlKj#Annp&^F+6Gkw)Y-
z*|*EpY!~3!Id0A~z%Oq#hAh-9SZ)FAJR28qzGled+z;oyXmaVCV@lv2^N6#R_|mMl
z4D)s${h6pF`>c`V?!WGx#}mLnFoQ|x(nW!Mm}~Tjw;o~kRceT5r^Fksl*!zJuC~0U
z13Ce5hIpya7HNeiw8G2qpi$iEqE2y3Zp_Cpq6aZYt*G;3AJ+TRSHe(2c*F_T%C%KB
zj(Gp+j&ZwDdJImaF{ubO5$6=$vL|=1@Aj{ggqen|C^y2=`~nlpIHt;OUQGJc@yVF(
z9jx1ATy_#)_C@v2xlw+2I4pj)n=}`rA~WSpKtp$$PpQ2IZq}u93vR2ub3yk#?LtZF
z8q^p)cm2KFsym1<712UOHeLJ>QqQ)R<0uuMN?`T1IAVy_NaW35S2Esv=0VP+VG|RN
zmWwsSk#9_t$BOA0_-jDoS`^k3Zo1Lt;2%zWOU-Cj(Lu^cQF?vh1`Bj$VYsEoC^R>P
zUa9dFHIIm_z`bPDcf5)$`1z#nt(ob%*Zfq*fsaDd2!qJCYwv0g_Sw(R#P5Kf_+9=N
zv_IiBbI7wKdC8WZz4B{OS9=<hor*+k3XScku$Xj(Ph@+Pyb3}>)??2$ucAta2Pki~
zY&qzg<wf%TfCR2dSIkDmAtLffp5c9c06|+RM(8}y*L!q8ex?+HI5(M?=26lsx0bcr
z!g~$cegDz*13@u+=}&BP6Rxj1wB+d5xT9sMEisDpQ#Ws%HIlA(v3K)O=<3!5`($Ys
z8InuLjT$ASI>mjIm2V%vwZ3qmBqnwoeRIUuPpaCgIM=jX;`Fp<(91Pxs0{v*?isrl
zX_8DSVCw10Gr4pXcHNG(;4{-p%3X<r-p`E2N8UDqHLX{ws^F9@J4&(3z@y7ADb=9I
zh-P8BgiN4)kGoIyc=CS3!Y7<L1t92RRfMLSXNQ8Zeu_U$J4+v%PDLT_E|i643-cgo
zg$xWJ*p~&k355eJ)%CelKXBKuRS5vZ3kBonfPL%=9gv?COQSXRrW>i2HBvpoX|-e?
zWQNEyxd8*jEA@l50E2=*E+`%s<fy5#(MGco>ZLRR$Me*drA(&cbL73i_X!vFzXwAf
zjhxK>f(e^sF;?aD&;E@$G&ZlJ`QoszU=A_8IXRAC^x>c<@oPr*BlvqaPGwcr_{RmD
zHVmi0F%e~bAKl8IMiCQEyz5pQV9(#iPAgbXEGED_8y;O?TaKL3?I{*<CVPG`BTj2G
zfsG3T+t(ys{K1Um&(`S8_9%yctJ~0k`PE5PF#PM#L%yuP83Q*?)XJhk(`IC6(S^on
zNc_vy(D7-kZuIa)L4o37dWYg86-m`QHAD32HLl+nApAQ6t`y($m`eqYO3kOV??yvg
z9vna^bQ8NA?@vGIdJ@C&B<MY&TV7?G9LLEK?oI|~#&e7YgnQs-tO2GSTzWl%{Q|nz
zqFQ?DqVy_K?cVlV^|{3b!g7_`haY9qlsg*qr^Wenw~x`+2m<=OyMHPE<sDk^+ZaC_
za|df|4BzMouZ6++Q=>Z#m7}FN8>W~mq26a&*TJbd11}<|l?B;Cu%wdz-o4KKM>8On
zde&wdZEd5q`tujMK)qzD|6a65;zwUdzjnu@-6$j`sM4=R{nqV1*@;60d;mKCHsbqL
zrCM2s&tpG~XN>n~lFHYUPSo+QMD6GERO_pGH#WxB(!5XI)-M`(Xtva{=j_>ie#<Yv
zh4^cPC}v^BK&wsd;z%`ycBT*Dm>k_~>fVe4s9~`WK6BuZt}{BeU;Sgep3<(RM3?t9
z)O=v>Z+=!cuVjdnom)esvYbhvM;yI)6+M?&`?^r~{m%aN65?q}>}JJYtf(HvBR<0e
z)eWmPg6N?$eA-`VV#e;mFNXF<MUoKZd%mw7x&DcnR<2hjuk?uW%;`k!ePb?P9dfDg
z&!hU1<mU$}`96TGP==qIT(@U8rpFK2)iqG;!=KY$ZOZ*MSP7B50Nx6fCOGV+o=OH6
zRv6kCa(&A?<9##7-Y_wEtPtU_wHz<|4agv^(QVfIXH*xXEZ>??wA2|gR<toMCxqN*
zLJwHpqx>W-NZEDM&HMB~od<70dnwz0>&4OX6hJ-EP`t;RfU;UCAH0$nNN0!`UVpb+
z@YnD!`6eIUDY#eo>Eeeo)$>CAbYq=#DyW$=xlvc>4P-^g&;>`7#@ikQR_Fued{j~P
z?yVGjim>6diijf6YE1NAo*}Oopy!l*BcnUGlHsD8R8gbNCLC&SeFzlU6tRCFvL+vM
z$4eueUlvAII$Ij!_m<G0T2?j;$k7jXyWas$wJ{9XWvGnR0rU^tM#mpzrvclZKX03p
zO|E9n$1i33>84v?GanmNrWQl{rp={egk{b&nJp|qgok>RpVLp8)OLxbI*^^H*@MIB
zDkDY((gCo+6(4nxZjd9XuTFy9VnNT-Sq<`EZSo1U47Y_8<5fSAeHrj`&9zSMYA2jW
zkRi$hM^@=*{oOFcZqHZomo^+H1^dRwGS%5qhFoH_VyXn$VEl^H#XPH>_aUR)3z21`
z%^RA;hHi7tH>keby#Tbi<5(Oy;*u|dm?siJ-lvK*pef)tJ83WeLyb5OZWi#M(z0p{
z{60h-PxT#-i%1Aqz9H*XHWO+O{9D|*KBMHCxmpHjA5*jtKavru?AA3W7v^_H_0(d3
zlBU`!*X7xDY}kjK53&&pT7af;M77?0AGpOs(ky#pwFi;A8!X*niMHn4>3o+-pG$jY
zTJHq!{ZYAj2N!4T8=-^aHB@zs;3RXgUeLYx2i23^#Xke&9>b3)^95b=O2+HSne*Bc
zkol6r<I2Lei$891lM|2X&gBf^%U)BLdRA5b@pTWa9nM^^bX8DeOq)YS>x(~o5CJw*
zWi#Cg7)V51Q+E)bF%p14rb?TRy-ZX*;YDb;+V9R94b!5I@#STVF|qEJW}}K%%4=Ul
z^{8tsET3jBR*4za$?lQji^$|;%>Cnkqa``vz4dGA1m+F%+rdY>E|;(4gsaANGjJh6
zZ~KnAwZHCT<2&y51vx6*nV`Ddw+bj&XMz5a_CD_X!QbpMNvJN{sA*bT2r&eBQ487i
z7TtE0Q_V|$I+zdFrkt>0bfAY)d)bs*`#Y7XM{X9TjYLi>wAWCBvtB6fGQ@-cHg~<8
zaM>*zhH&-{Ys3C~MJgnl0dO<*POR43&KV%$;z(;$QH*Smmw$Tk^uv3a+ud_x7h@Xl
z9OG45>f&7P2u`0!YW)0dIw_#=hW53e;3`AVUzZjP)Ihx6iP72`2<Xn9QyCwF=W7~n
zl25tG0*uZPOb%U?=rlpS@W16oWWj>SV+K&m{kDA#z-8kAtf|GS=yQqB-@S;+`9A%x
zOQfxi)bmAB@3;M4#mwDet07wb#@Zruj@=S@>&I>U+UzS*#4J|lSfxtXyJNAyzQkwl
z{f}#8P`b0(aP5w&dZl!o<*kjcZB~XaPKZS<4QloHeMsR?o^qnX#KC{JOC`N&M9fXO
zk6+p@q12pX4%YQ}+fS2GJzKE{4;_MacgzXlwCfmbwG&&OZ0dAaDdBNwkmgW~hyTSI
zdPCr|!*e>0C!i~of%<}bd4pWCLw}e`si~4r<YgLNVM^`5r2>*H5e?u+w3_PnR8%~F
z5Nb*9+r_;BxU;6(PVL{oA8%i?cdxuuzZ?Gs>eRd2u(q*p3@0cauJ_!xNA*d#nis4C
zNvtqWCkj{Qh)o~0R$Br|?F#2-6&fKrC-y*9NQki<9DOv=Bt})K@aTaL3lDGC*<t<V
zx@gThkY~Pzbt3_#m_KVY$EXJl=jLxxqT-|B0SW(5+w^A%+XZFiQ+lmKQw&2pa;7tX
z9>@oMHR7$l+Ei8*0b%+BvE(!upqUht_wlJ8(JnC#z1dGKR{ga|0%y?tvn%s0wOeq1
z$*(pi2pA7zeFoR*H;z}GM|$K%?(ewTrMh;^qU~ylt8M9VaASA^zkF%K8hKblP>_j%
zKDv#cM9+UBqqFz#$nXy0;TBGjGg}wMiPO5_Kv$uYHEqP00Pg*m^I#tV-Nx9?72oV8
zm*B@1)6yf*h(aqqsUuqZo5h)E>_92Pw*YF1lACsdZ&-S*z%56D^0J)}2A=U#;+^OH
z^w7XbgCV?GQM?3bJ{dYv)g}`6&n}rgkN3~qHL)1r$S^nGVlOYQ3-XF%pN}e{M4Ez6
z7)NvzR9eP_Ec%Cf5tU>Xj<*{bjR-cMM=rHt@2GSis#i8H1s0PfG59EFHw2=(M(Z{8
zq28lwgs-rJ7d~bW>vs-q2giL@qE0w<1TIj3GSY|`MRKIP!H;5nDANdttOWbNKBA1@
zf<Xg$oKE2TEMak5zBlhWRzZgTnEyofS_)4de_}N8u{rc(QIu&*r#i4@fRMWDi+!>{
zP>pzk|9NLeq1l;xBHK7UlIK%;0@OS&Y<t;nFMxIm>y&#_(0?BB;&C6-U_=73!@ztp
zVrW}nhzg?s=2ltN(>wfX%9V0ViTco2D_EX2S0{Wx4v1hBO7L!y!SL@0HH!WVQkd2T
zf@3lq_0EV=X%C3>rhb{(o!w?_nQq7Y0-uf%5r~sk^!^mAh^Ezz@$a5C!3fDx&T5M_
zERIxfai6rZ?h^|3t1JrpMFQu&!tzpGcz374oO&nKvnB;UH|gR&pS7M|B>xL~sWSiK
z*e}eNi;rF3>Gus;57EOPZf)tnJDQ;!#Yp8*<AnL|!CX)f__K!um6kG>c{2A`LldA+
z<t_XtcaKb(1RU~yjyZ_`C3e7wH4WK5>9Qi|*S#gtXKm%uxzF0864kR4>~};2#q;2g
z_s8EI_Rqc{aeZIq&?dP8alFnK3_uc1O`6GBJL?O=w{N<IcP3j+{eGV2Ax$_Z9>tv=
zBfNNvGmtcz#tb$-C>`s5cJ!;C)|)UEu(|>6HW{^Cf)ZgN^r_q$e{o{EECW%sLdVE#
zH!rXIf+(fWdK}ARgH6a)+a48G<<4G4iq<Q>#b$4hmQ~JQ(NIf%ACSorH?|^XDWzZF
zT>r9aNsUvs^dk{kX=M-)Xd2VvTB!3)DHDhP&Fm=aMqs=F5PE*8+eYBYF38b2Vs9Wi
z9y@(5sTev!9gQjpHfTIe&{2JsK&{^w6RlTMP{@h8@@z((IN&GNPoGFYW~P*^@ybjS
zrmOXUg{f)y=#hy1$Bk2;q-mVaI!#W2ZU)CbDDRf)?T5ph-wYn1QLVLr6bH)j*XAD@
zO2Hs$nitGDU$NVR8#E3X0i+@JtyK;CWdYrRL<U}b8U(drs)*B7hZGR$%lo?qH*&SH
zL)4_EgdV};C*V6J4rkHV=>8QB?~)M{@7x~io)4u~SdExF4$-9^u}r09F}LvZ4dAby
ztxuo3pQbpu&Y6O<)taV7z>p-81y$A>LR@d@gG1WOR7<!RSfG+AFo<b2`=-iuAI66s
zaL1ZYgJm4z?KeC@s2UBw+*xFJ=hVGNLDr8Dg5mZihVVM>W(w$#JmHXx56VIOXSv$P
z<N9|Lh5QYaAMg$D0}wCr$FujA7n|5Jiv_-ecjXeKZq`WWBl0n%c&w+P3Oo$+8rhPj
zlgk&V*KFQ}9y~X5v79mX)^8)!cavMZ_sI#DiK}i8=mp^ki<AC0c3eDR74xOfPq>)8
z;XK{En`muMeMOH?I^&JpRj642<Qa&_O3BC67E2#dUjW33fB=3T5sU<z4d#XJhw{fY
zJt<iXQBi6b4I`dC<R9BSRpU3bs()Xm>cN`yY<$}BVjHF-sheF1fRfQ2U+5~i=!|qZ
zAiOim^`5d}+3QWs%Ui+y@Pv7RlK&YG&3XIsvR7%1;Qamhcn+}o!1|XDTl4}^H@f6P
zPI2WRIatACc(FlEb6l*JRZE_)*621;_4&peE<6(byVmm03^ZlkKYCt9LY#T6<|j3W
zHMOhyVux`J;C>Ig#W`QZXB6Fsg4*td`$hl|LR8Yt*l(C>RN~LBVd(^;L?!u@P~$aC
zqW@56WgZZ?6lZ~91a+UfZM;AIE?4zz*q=Eq2+`wg-hj@Fj>5WsL2iqErBk0)AGSK=
z6E(doW&#&9A`j;WMo<J<whKUI`078BajD5$kb2M*15N$x?K3TWa3G-@i7)^L>;K%o
zU;YKPl23CSv%mvxMhamD41i<^=~n&n{$J3us~Z?%;yq_l`tDW~5d;#cYh6biraJ9$
zGZGqqVpzvwgid)PM6FGhRNS#Qr;42)CIVsJ^9yxfV$y&Igx(!A8K3W@rIND#O*85s
zw)gAnwGUKQhF0PEcFlg?c-03#348x<jsTPi40LIcSYL`2A0b$Yw3&>g{RzzDx*B<q
z`<U)Zi~<TOufzWomKCfC#AAq7+JHaWY+DiY-cOp3bDk8rhjL`9WbCt;4+)*(X8`6%
zf*koYPNxAQXEYs3P6rfHM+^XnujVy-tYuP5Zsp(w7QJ;UR%$>*PcV9;Svk+(K$cPx
zn%ZxrBK)aQ_-4pg+FggNW>#C|Bzkb618V^QBj&V|sA$!X1V@j@0z|~!f^miCT5Yb4
zK#If|!`${sgxX6hbGG`CQn9fnD3w2zcMZWIsvSmpaB+GeL8l?Wq;cC9Ba2}&!XB}~
zKwmsh1tl=ae$?!DN0b^u`yMKU?CVt&8E9#^WlY08GjLK7Bv7FfaYVJ#@?b_Jqw7$8
zb7WIRP79tsgL%go0sCqiL5Zq(o)MG}w7B9x0}$)I4;f)yUs=L+>jG$W>BUN>$5eD>
zeA&EGMJ>R8R2G?vdI4o67^&^~)#>}0&eT~ruzaA2QhE|LF<EQR)oMFq7<%l^HG=n@
zf<&&3)F$NYEcxj2#m%NiIpz)DktROZMdum@*WTTr{}SQdUAkcJ&OUpZd!QPYG3Q@n
z1QPB^%dCo!0Jl94K(<&0m)-b6@i46|2O<t`*)3zCaPzjS*eB=>?Uj^;%;Y7;4_r?C
z#Vq}6h62uIuD5gBG%afsIba)EPZIhFT$@%t$-3N`<9AeaUHu{LVEOXCNCS0c_*Sy!
zxg?0ERXb8<-a{M!_(19#`fTT*dx1>@QDFxO=bf_kNn>E%@%JT2@+0X1r|>^fb|c6(
z32tznUgWwQ#V?kjq%_E}=zvFFTFda&)A0XfQ6&W)67;}@ya}S1`;vBBx<h!?H^&=`
z4{MmW-SWN>`~LfRO{WMy5{2v?EZf&p$lpaeK7ViBT6#!Bcf@QNt1S;6MGQN8S`-TF
z0IWR!yfCR2!d3hc8Vv~SiYij;*!F|XSL+qXTwM5Cm43}l2%=b{MesX&Q~Y*ht|lhT
zJ1ci(7bzosJwMO+u(GEE=rs^=i+(JzwD9={RygBMhUe*JtO#@q^&_|(AtaM~I#$m2
z8=~$z6EnWAm*pL@C`pH_H@VL7ph}}&3)FY_5U<`p?Wi$2)`T#yUv>DoV*mY0*e}lR
z*MC8hk9?ks{5Y_Ux?9s)gVe-QOQ?m9>ODb$nO_MNo_X0<b3@Z0QFv33t#7`E)AK9k
zb_fGgQwu<|bm88^ZS`%B{|d#hJ{Yv2`4(`8rYj@r_Ggq_<^%w_ys?WWkrg-P#_rsC
z8B2bFhVF64gwCVnua<*8q%0;eQ>iXr>d$<*pyD>8Yd1EFN+6z!Es_BHhXMJ12E|UX
z!#m{BW0uN<ksuL$4;STQ6g+8uy~_Ja8AI0r5T5(>#q&5i5ejfVu&C=O0)(inVO^Aq
z$_+ZPLUr0JLmCK!Y}*v(KAflr_)rSeI1N5#exE{e+@h3JvfO2ILtgRMLB6a%8pzft
z7sD9a>{A@Pu0mr0^vMe&&`~4d+wE%80YHW{&CxZFeo9t1vml(_Df{AxdtjMu<H3)F
z)KjOVM-%$)VSAgi+DMj;*1(0d)aU}Z4lgv|lgh!MQ=5f+GlzAb_01}hPCt2XAp_re
zoWd*P>KL>YUC}PzY>N0{+%?eZgQ$+@rD|}g<@=k;@+Egp6FN(fzzUGMD5linv&a@Y
zZ3F(@t*appLoMUORBlE`Uv4RQSUP>5cK>tmTyUizMdi&%qQ)u};0XjiYo2j(H<bZ=
zm;@D1OfypHFi9Tle}*7UZ~{z8^-e|%3nF-aJYdz3$~6_&o5Y)H>487vfnC!GW(0`0
z(Ls^O&iW$YY~9Px3e~|pq0XJ@qq_7^)19;j<)beOj!*vi%zeXED|FD9r#~CDr#lea
z<s<~$;pbsvO%sU86!+xoH9Nw0|9<eMm2ew~2=5T4lnUBLjV~ChPFH{iICt=2X}-G#
z2A7^lDOJ!?Rq`*lC1NDjd{q138)iP@o0~y-aLko*&pYNnqS`Dr{B54c(A53UJ{WEI
zxmTW$r0_6oe^>U1_h*QSg)F`3+a5oWQlHb;5&4Obvhe-oPdjcMQ~s>mRx}(wPaojB
zg)9i$sH?a|Zn9++Hfv;c9!J*53h!q!(IW<;>DLi?VE+SoW)M4SjvALZ`bDR)(_FB5
zTJgw?>qy|8<RI5D`c|B?bWDlnh<Yt$#15SM11YB;LVb0`LlE6aj|2DUFs+{r@9*x&
zSMX)^0TSW&w(~mQbeptv-x5-g<Js?hZ_@roS_U{dJ*_T%Xu#=@>?=7u_T9&$9A$nV
zs1B+?i9qNWTC--6m~5@}#eR{W-*x}?XO+~v-VEC!i)qvVeT7cq^x>zjmw&UA%#2Tc
z#UGN9|IZfIF$M(50KDX9v4z3Ctv^&-I8=r~Al^ls;WGltI!*iNDbZ0UI-(CcDARPH
z`7HUG`UN{^E*=i+$_IYY#56~(MRlC<NXcQzp#Ojrh8wyfUllB#c|>k;tRk?;8h7XO
z^kH<N(#q8?i8s(r+_d-5YE{3ar>j$dEq!mQiDR4-@p+F``#Yk@aW66sl>=-i#GX_9
zp$&o3git^W`Z@0*;9Y{VmGsq^TFp8Y(bAx@QPf*cTf7?CV&l=Dg)djiwl7?In||}I
z4Y0Rpw+6}ayJhN~OtrI)TJ(OM^AWl=i&U(W-q*P?C^ESC;_v}8+2Mrcxq&{qCD5?K
z3Tm)ocFxDA6jP*GEkEddv{^8v75a_qeok?aV+RYDETN|5K9=@8<MuQx0Dv;c2=jYL
z4?<sm;)H4oP|uFu^S4vfFLCUrH|wgPJjs2bFUxg%Jkh2yDhT%q;oB*6(>ycAbHx#S
zeg%+5bwGPf1rRnHX%dC~n*XznXu1dOH2|c-a^gAGKAu`I^B33}xpI!>sobyDh2pMl
z86FEfg*`HWsle-zSQ6<_2#)%?^stf!V3&OzAun7jhI%aIB@84lV1W&EY7+;<I%97V
zU0_M05?%JipWVG!rSQpe1djVc{oswhQ5;&F<|}!cAOt+Ntk=j$8)F+?OedDVlQP{f
z59=YL7Om856GB@h#u#&Z?p{wcmT6%hmy8KEEHN%O1n?<Q%!23wSbuxN9SdowtaP&q
z9BF{pNCrPEbp*0!ZlCc(SmXJc7$XL@3%H~}Niwqf_@y=O=e^o*V9)Sy0|JdCq{B~d
zO8rDyWr%@y>ZHVuAWcH?c-8&B)HeAkikFtn48CM95P>xRRZ(S(>=N-V^*eIjCGxZ{
zOV%ukqA+bh6xtRvFZ;z~hyW-2Y>}Pu?Hu}OFx@3|JuyL>cc*9+Zm7G5zL^+<(@Fqu
zu)p*X(@k}&`X7`enS4Nf<H+jXk+k@DTp<T$S=J_O^5rp73*=YgqI+t!YKI|0p*k!v
zMlEW5W1zoafs=Hpzu5Apk4M!;%?uyAwB7iraM~bh&sMZrlL1TmOE-@8fLuRp?<G!i
z0Ez+#!)$W%a7<mejiCi_6JBc7++XqPRe7L!@recZQ`Lm)wg39Z>3*xJLn=@7PAkPp
zt(HZ0A{;Q)C7mC{pNv?4P+H2SzqS=*qs6=>^@<PAX;muEKv_@KJwXWe3E87czUha7
z6ZCWHx;N<9@L_(8KIOjJlYQr09u#-f)fv!T>u8M69)^ZgMqW7Y?VY#8gZl;cu`dp2
z_@Vux6wi%*Q=xdhGO(LiWjRS~%-X%pp)%0>o(ozzKZs&7_78c@&VQ+Cj-a9P5*Abv
z+oVSGGcBl7#w&35t*6q6gNtDwI|n=vAC0Twfe<er5<nzt#J`|>_k4sJ8mf*gI1p2R
zF<XbvyU?RC=fbnw8V8kDgI?Ljs+axbsnxUUmySJ)mgkzTs(wpO{`n4Y*6UHupWK#G
z3Ar;tc1FDb#KTf+ExW%8u&IKyJI>9|T1|!b!<?96ljIKI&;S2Wv8DtJ6=(me2Ez@N
z>D*wZ_{BMR*chus>#NM~`Edq3AiCpdcYVaZ>-*OhIf3($6^v6rH`}ePj+H+_6h4GJ
zk!;&LdHy|0ZTE9^eyY63y5Nl}R)UOI#q(3~6;IWVy+WzQy4xVcC7bRW;}iq6>FZ-D
z2i=~qiUE;CL{y&@Q-Z*NCvOwDnIq*r%x8V1{bM5@l{(qCzq<%Jwi-KRx?_9QnvvMO
zVf#aDcaTa}tK=X|N&&rFc+KE;F#VhHiKBt0@T1~UXai)CC%iIsu91$)qj-z@Cm{0o
zC2Et)@_S<&6Zfi&d<AV|C#^@t@D9<fGsHXsNmy$f*)TV=`8$g~kXpQFKeJnFJ7h=i
zoPD7G&(X9%SGtl>b6Sv<#R?$SRQ>EuJ!J-s_po;+XKDjm>Y}s5itl=ZK-)pF&#%8d
z-t}U&K)(_85&UW38&|csmkH;WX{|On14hdT@&$5szFv?ud#_tXwp*w44Sj}a^;?g6
z=fKYn0g4%%m&sq=`Fx?ysyq=3r?|c-YGBbwKrw)>mYT}Ry5sT7|J_k%Tw*L%aM*KN
zKpbfu9-;vy8licmHY3Q6$HA*6L;gG(4bBHT4JEGae_4an;(bV+InneAMktOhBu>7U
zp%m}ve1t=%8Vo61CKnf~%3+@va@ml)?<(CQuqK$>8*^z3l>Ev2_zAuc1!eQ_nA`ih
zR{hVA$D!+yZ9~&%H47MLqUfCjhp@uzFBT#Ci<7QUy$@l*kJNfl3bi2%c_K|@R25hd
zPX%RF8_<KnsibJ%=pB>r2o4~1w)7d&s@^?oe>^Pl)-lD;H_lyA>|;!yqOtmzfVJ7b
zic@8SjL^1kM95%C6GTNAq>3DvCj?vs_S8TPdbhtToy2%^6Hg0T2Oj|WD!;<GUMyU0
ztZ~zAFeXrFMCdCZWQL9oQOEiZ+-o?Jv2+VZpPSC}e|Pl>3n8Vn+&ndKPMi&2;W=5n
zdhFaBqgiOB{bp8vN?xMd<F@!ef`q(on_vel#XTtSg(RtUY1>n(gy5p_C8m7$R(I@!
z>ygTPJpx%Xfr%WiQFz3QLjgba#3y!jK3QwAzqNV$q-i~%#e7{aEi>e?vih)#&$S$t
z+TT*8Eh`g9RX_k29P<P&PcEPFG7`EkPMSM5a~bFed0)DwzG5f!@;=3p_mmB-V+|aK
z*4ZI3PDd9VX^+ibO}oAa_Q2q5i&1A%%<_Tz-w*>7Gu2i4kgj+tY^)dYM*ZvrKn``i
z3wiQsdDG6x-j;H07_TOHQ5Tuz@^ev+`-LQZ?8R@_=DFiu{_DRsIo+7CSpp8F+;sBJ
zA7vCIDcY_48#0EJ3v`%TX2E`RhC3L)pb%M+YJ0-_!IiEFhPyO0_mX>GAMjOEBaL<d
zGTrO~CNo98-z<G+q`H0K-qZA54Z7ehufC(+3D8}d5mo<wV)_m`MH#OPmy3e$9-nPL
z2Od7odnF}RTtv9#Yc~qUg8)ZK=cNXVXEpwHcoo^TcLjzNg4FeQI$yZgi@Xh51!v=d
zIQZ%Q%REKTYb;<Jfp>2H7@PXe7#jy-7tfcs$~3P9iIn=u=0@;U?cEwFX&w?G<K0PA
z>WfZbQ?y)m3>ojWbQ@pM^p``%FP`#8Z^J#*2FYoCbETSlSV?RGY0-oM?zV!x!c2iF
z9%lX}crP)>lbYlb?T*HCZHhpLaLv0TI56$>bvsXw7c}3+r-|fYL-D@wnKGLk#0D@(
z(41_grwcBCUtuJ8tCf44EX1?{=PP19&f=ca{%zB>zdb1cPiDLBqq1~$dGJR_iN$p6
z)cjGS=YaHpS0*;%If+iwxk>k^lcmO8*L)jjX6^aWYoF>z7o!h(N1t(QkY2BGq@%vJ
zeuLta5~C8=@olH*Q$|g7_U@>;hFURqtMuM5a6@r0-Vv$wpJr+)QN<8dhP5D(WI;gs
zCPcANYYW;~a%VW0%GEar_vnW4JCsaq@v5)9RJ0D)c2z1-31}dVOsTOV(}J_8rdo|e
z5rYGctV&KCZwX5`CIG&@rAczEG3zzZE<&e<sVLHg^M8(A(+KSkj8Z#zL2yipY&ip^
z7cnZV9*dr}GrpiU_Ui1%roKwrxZzE5maoxk?m0&2QWm?3lWBP#F@69H8szg9O~RuW
z)sb4dkt5EJZTSVyKixW{6o7wC|Jrdt?WxNvHM)nuxnZ-BLM2L5y6x!=#a)qU*`zI^
zBHVWZs0L!-OqaOlrprGAG<9QX&-qUx4}{Tg^X+R#xz8u2`oq_gko|Cvi{W@6c7`~w
z+zg8{`#Utu?MP;iv5!1qZpy8~BwG})VNZMjWa>A3t1QOY9!x4%P}Amw6S=os012Uu
zH4OFDH$?`su!SQvGW)Z+QF}Xs|2nMaB+*>Y-f&`fX133&s`FP0GO#(PhkUKeXA>}$
zBE|89z29#;!f!3SOkIBP>~4rdOFTsB;mES!`10o|E0)bpy^y0ql!K1a`0u}&1Cx|-
zCX1t85f)-bIsgRWN=-tlK*HIQ9sN=xE;an64&Sdn)_)zabJEiC8)t?bb)J7l*Dr7U
zNN?B38h^{<4EDi_hjhw(JJUE1+&m=Tq|Bvu%56*|ilAxAES)+HBs|C{dhxdz>N=4!
zdzLG5Wa>w2W|AA5+k;ptu5}keY1F*Gjue5Q)#%v&D(%BqKwAZxzil{mdZ}P>gE@?W
zIo0h++9A{R0d`I<Kz>&(JsNhIRF0_t`zVsFceONU4VWSzo>6Gq574p1WW?(8N;;^c
z`M_VtcwJHOVnF^gj~XwFIZ461ri%OYm4+(Qr_vy+D+fc9nYzg!Kk38LiIJmVtjL0h
zNs9FBjqhWwExB;pwVPgK2I|GhuVWJQhM-<?fG0#9t7E;6VUf54GiR&vnvxWUDFmfs
zUIkfZDfEXp?JC0K<pYGttn=0_K)&|V)uyz#J@k6;hXX*4Ol+L2mdSAcx^U}E5)&HO
zz$qfB0_hJ7rnZ>%C#Z~XGY<mH-ZDwgV67#}o;txgrXG6qF4z8sr5Nv{_Tz8q{Ar3`
z`O^i@AUQ(0ob0J-*Rfdy$Jk&r&mZ^@0v{%DYFsl`c-~Xefy4wd#YhODn=SRFUS;xb
zYl@xSPqY6H$~r#$ycFL6kOIb)Zz0~_6GRKNdy2WPO&Kxj#9l>@BvU-xDn-y<l&zxA
zl7igWg!HR@wNT63BGrYaEhmCO0EuWGB`xcAk6NYfbIM6yXE{;Nty0bys(G$Q8s-V?
z+fxAu^&YyIPMdhSoF^i|@5<*SSbMq2R2A8Y9jXWZL;|h+C(hn4YeNz;0+^2wG)k7~
zI#w@}7fcUI$?)Sl^+%2x9~+L_n-z|mTjc#+u=S3TLv?#+lb(j!Qwz1EgI<1`LiHcD
zIPMYIUj4nW@l!q}siL?d^iWdU5g&f-W`fW-a?#DsBXVNi$Zve>;v4~xNk3(0u@tIi
z-3n(Bsi(=O$rTm_{6T)DVh6yWqE){H{|gG%Ki8Uk>*&MLB>dmbn*{<Rz5^K=EX);h
zMwfbmC-xz#>clh_1gU`i-_$^Rg}2F&Cfs{)yfo~7LgOvMhdQou$!$K)=XpAXa9b}>
zR8C&!I!M@e{VJ>)qW$?m*>Bj5^@~s9wqQvSvG~U-YATV@Zna{QKr&p4MJkq`Q1=#F
zm1Kme4pvda+W>6!4(Py~t~=(~TZ<1f18df8)d8-e@gX&bZ!4B0#G@t;$J7v*`R$|3
zz{-N}BjIF-GP*!l`};b?9}?-<Q<S5MRLB~CbAi>-5c>Z4qiGowkjLyS9;|Z3Pb$)K
zk)2jKWQbUIz%V`Yma~Lh4C-TvAg5WwCTrSXX(-ts^xpl;J~~Mf0MlZy7RcADUT%O^
zht~LcVg6Yhu9BUR6SMR3&3M^1Eb4{Zt<7!|0n{%}b=$a!I)L~ql1QM$qZBPzA*X-@
z;1%*n?u$4vP?Ep#p|XB=aSh5pewPVJR&gz7j^r^BlrKG=s8L3!Peqzi(}`|d+BOQZ
z_kB*mAC)?0WuB8QR;)kFKMMhPHGq?Yux;r78iCG0U%~v3lJMV)>vNL*U1tp0Oq(?-
z8eB+=IhNdql$tRT7F1G<e;d2imt{3EguI=WU-D(j=-JL9T!f|0pP2Vu=QPQJ>r%e(
zS}x$;x=Tky-UcXJqUV*OJsVi1rKP~Kp1NhP^6;m(4+$HcGi67p>DdSH$OTT)Fz!M`
zhacF$UqsJ$pWeL|>GQUNbmVCuYihR{GF}z|oLniNIZ4}@<6BXfcj)D^e|$?1z(7Z6
zzT#0D^a_aR@!KaWJDU1zSMaJMA!eSAqp3<IY0~$Ju_T$40_jBUd;kq8!$wqDtB61t
z=rDQYMYql<_9NYO3A*099F?R*C((E@O_*Yc=khyD3YF%BHfq1ewU!<)xWE>}$bnz+
zQZ?OCdjJj&g!G@&4<UK$oMr6FKLOH1;cbmf&!ktokw_;t|56n<h<-rRy#VsxmZGOL
z1|6Ae#WNq6mV|WNJomF&B?Q2C*%*8n2EX#45$lL?U`1+x*&vQC+1}TXmdY;X(YSWw
z79Ms{mP`x{hQm&?-vNRi-P2Ny2uZ+-A^eJN>Gmj|;+@a6q%$|hwg!DI;eKrYrZk;}
zEG-S_08e#=1en;25L+nmVxH;(u5XpD20DiHTqJcUKd%X8-p>pP>Rht{uk7?G2e^Pg
zce;%Yq{|}9e+ozSX*9%KMSFdz*{zefT<dTcYZ{Ia;t}(9FNV(gKM$a4e6b8^`*MRe
zSKe@neD`*U#WPB6s-HFXTGfo}EjQ_T8@&(XdqNu{ffR6{8i*q9q=X&f8o=Q|SD<Cy
zZBF+P-z)-TIe$q5<7n{{%wzi3?O!et2Qxi*)L}jec_MO^Erh!}2cs$gM<Xm5kYnPT
z2^HE;8xg5vPd25L>OVw@_P#ToYq&og?|0?Xg05UT!&O!LJun3*wI~a2l&G#YRph;&
zNr2h7=zFJRI8-xX{W3_z7!d^<)<9ZGJ}juSd$XkKA5r!}$9gs;9VuVAlzK6e3)jmH
zFKi5LHFwhDSR&84DxTGLy|QNg;#=E8Y`6I%J<*D5N6LOe)G${wUKj1z?Ku?>|K!bT
z8~iZ1r+kl&u-(N4m3ImMWarBq60_0z#&XH&f%i&(z7tYZZYm3px+NW)9Q?z}@-ecI
zMG_<>4YX2EF2tA?FFv>NOL!uhHhQ7u$uAS);Vf2Ulj>Y8W`Lb-0p!O(b3`cTnK5eP
zzgWUH8gk`Q(K{SWg(CYPK1-VN67Va7XamhAl`BPLNFRoBi(b3qS~i;8tTI<6Giy-U
zmg3J02YjX}yX?jUb9FA-RGI3cqF>zRk9G1kG^UH9e#5Lb$TSmjEtwcj441plPFsGv
zFl1HRIq<_w`0#;Ob{u9GSKYbD+^WWR`Han>6EK*<QrQj=7D|54{*eVbnV)^&T~CxN
zDb0v2CTtqK(-I2gQvL6Wxu$gWfBOc7!7i=*$xP_0YX+(4rI?U~P<6)m&zAAJuyC7D
zZy($(i?Pd+4ERj>pHA;45`b1AIcZ6kDR6|gh^np5mcF+l6i2<1nb}jBK8Qyt708FG
z&j9r_T1gMMpZOGr|CH+za!7ii{f+gxR+gBFFH_HNDol|xPC!bgx!C(?Yx8}<5&*H^
zEn)HhNILg$ru+Yo&zXfBHaV7aHiz5ToJuW>VRP8#v{cU7NDd)7e39cq&M}h1jA3#<
zRJuiuSurD(N|Hi{JJI>>`+N8OU046Pu3X0Z{ds*}&*$^;1bk3Mt>CaG^y=p@+EMmd
z_C5NnUr89frQhId^M+|ZSEobzjO)!K)oFEiijBJyn#D;UNq6rDcsO&^!>qRGT}2PS
z@eTaWdLY$jw9Ii3>tvEp8OzMWfMwdmzR)r%33Yf=Oxzsx6P;B0Lh&rqzsa+hJ;j}m
zd0+_T^T!0iiLul5hF+ii`)r&Hz;W#;-tTe|sXWd&%Mn^B`*XL50UZZDC&Xrk3nMav
z1K;~&R=y_|A$HoT%;)P>Td*2tGS+I4OUE4XMfz`;-ARxI5+=vwIfCWwn2qvS0#fbD
z4y-zpd~KR&`iN@ve1)8Wl4YT#lbOY7NWa_h4v{D#dZH|iw3dd0ZCjx$q7y=Z-(@#-
zdLuHKQr}>2nR(Zl>6qq;ZG(9TuQ4D<{GA<%gI`U?$=oBNK&fz5K{BIR!i;E^79GbD
zZ2`@>2=-cPo_;-D4xTeYc^lh9kgCHZi%-4BAfS%=oN%4~7x}83?>hWjI>xfzJwkGg
zl0)0g{jFAxh`VBj8JW+YUydM|`!DsGymM7#oO#K;U-j=(#l?bg<J|d$;40|mQT12K
zf{-}?rO9X#N?$2IiZK#bBpTzz!<cS#tieOBR^@Z0z%{&Koo&QJ?x$-dcjOUqeyVxF
zqh1x9yGzh<uCbrppWf9%B7~617H`P*=U?J!sekCxELxWBZ{=Gr`!u*NJHg$vh1XcP
zsriDeyqan{aYF(l-Hkk?W{8_zwLq$bM`WCr|7E|^99K)8s8Nwq7tHLiCmorNqY|*4
zjMDazOMaG|IyKuam@ag>KcnCSWPY?*7hic=TXIn$KWl+w28(TM;z~b0s5>?_v?UwS
z-#)J+h<ue@HBnEK;Oe72D;nlb7g2sIDEri|_kE%((GEsht6#BSCz??WH{RZ$Zy~;y
zP?8CCN23O|#mlpI^6V!rh2M>p9B)Ya(h(ct^X3LT_uAuGXFvb@o$lp{qrO#9#ryk&
zGc6eJT;<h+b?5uJ!_t4#rL>uZV#+fRCapD7+H}e3pZ70{S6+9+f4Wa84b-jLU5eEF
zzB(J|UyV3yvmaNO_}2*+rQA?67RFJ6N7WCT)ef<QTOOyoIDD$gr#<xW6M8-B_96}@
zB(mM~;}Rt&*oRAKVJfe*9iMAk?d+G$GW_Mbd6fEWDbe%jK_{KShduU7oFnzu3qBJt
z_%cn7QmTTnf3FoCGUK+>pfYudAO%wCqSU^Y;0BZ|4D2fZx2D+1+hNaZOTAn9thWTp
z>cA}EUyPj@Uk&-v5UrE@SKMb+dqbC>l<Mup-9HWve835t&`xaX{_rk!MTi+|s$bI2
zA|=&?D7Z}7EX1#}9q(fBQri<}&92vQwaA+GZvi5GT_dRH@bh4-N@Fph^j?fP_>!o(
z%jFo_(YJ;a3domJazj4AQO}!ptU<tBQS<NECY|H#w#9qeeV+L-JcO{(tcO|N)`!R+
zUUn#|Q;>=g)=572c~_L$xnhdG9i5jwEs$IceYV@*8`NQ3K)`gzjXhO^pb(Zw_E4>u
zR*@$@kim_OjUkF4nXu(~xihN9`7d7J4+YOX=}TGZZ=QPhu2D^^&5Yd=%$%6{GoNVc
zd6J-+sFf-t_H09q80_7Q&=IyUnFwYR8B^`6ADFLt`tg<1{^0(w`Pvrh6O``0C}4eN
zZiL@OO<WA5ij=@hcmaO9wiq<(Br{jL>*I@k2F>gn+Q<Xi8G7+mX@7*+hEWkk<;MJy
zikJWz%Y)LR4A#xl(Tz|7UzZ_BY_d@tKyEiKL~eka<A{$!?u;f+rN#sp2O$HTfY%&z
zsi@=mZ0d9om`{M`q49s(2BeNf#)6vI&*Z>V5XBxtHI>g{eS$x$R_heHJrOe7OBWdD
zlI#MT?B5(MlF)7+cp!6TBv|T$v9YHx!<-Qicat^!RZRIG8)(B(z1qr<nL}eMH*(C(
zsi4xAyvs{6DvlV;AV^kl8u~c_=77p8#Nvk)h=~{QZtUvaAKzBK(+1sqaR6~%8{oim
zIBLumL<`m)XO_<5L9Bq|UpoY&i7w9yWAb{Ow;LkfO0S*P^%;mIdOz=3A9skJ+z|I%
zB)^J%EMB1_o2Xg=KZc#;xgr$gA4t%PVpV0U76|i^Uf!BP*Y#B~`<dZ5M@SXdS@_pV
zR(sjnFHPmE0<X+mvvZFQBr_Ltr8yE!&3#HQAEkQj;TTOYpS>)OO+yObP1QX=K6pky
zY@x1c<vNUibYibQa-^TFbKExEJ*xd#&3TK&{GNvTjTeu<s+AG$y%8mebTNNRZ4XXr
zUz<PScr#NC(GYn)E0sxntIn3?B=h$a*;!g1d24uOu;ih|{9%KCBb1*_t$oCo8Do!3
zyfveB_D#5-pI5MXV#=EOFC@cgPH8#XK$w;54p(iXa>8BJJt?a@90mQrB8wY?GAUnB
zW9F3Al3RO+vp4inPUwHwoNWJPL5{wRCVcKcNaSv#54EVpge6tayQ9(yM*D@U9C1e+
ze7|>*ED1BB3K!-Wc}di6S26YZ4CPV89wHr&q!cub(<jY4J>4vHit4zYqk3Tw0fScL
z!fwUJXwR4vZM-u8zPx7FE|RfmSxrk&c^1?Y3z86@at<$8E>v!}XbRH)0);Y}QPh%Z
z0Ygps#eUvbkEqlCu<@_UJ2t|FS-i6GK)1fb0|n<8qR07neqDFrb3!#6U?~?9*r@vy
z!m0>Q1WEWW1i?F*K?{Z26jM@`aztaA|2jf~mWh1E;YU4-y*`|XE3kpaEP)Yrf%CkB
z=HUFmszs04+mT%hs%tRzpbb-qlsMT`ud=$WzR*;}6p3j%GsTT;8X5(43@nTY;<+i@
zy(~a1yjS<=*w!nkA?6Ckp`$14A#@ek%z@bdWH+6oWLK2T#0N`Ff{iBB6E7D0r0NpG
zejiRN6Irq}z)qh~ZKHYr9o{dNHPuagEDB$Z?=ZkYD0R2LM<vu8oKNt_80N-;O9H6u
z)&CnZSR$NoA)GM;JlRQPAsMY22RC#7*UR2_O#3J3Z&?!+If)XF+a4b{sri%garcK2
z*&}WDpQ%sMQIs^~FDEVBN!vWFfw&qL=0PN?U#vxTgZNqXIb#8k8`9CGGjUnT212hI
ztUOG$OFYoVT5EtPF<f~>IvV}b`KtWT+*PitMgA(VR7P0Dj|Tz-zI`NNmKoF!!tsWn
zwOlCX2+v)DN|-u&wTMlfaA^7CiVz1)g2UJFVsLG~y!PCv7K4_~Z=YBcxSeeV*9Fg2
zgdE@}!K}7m=^wo!U2eOnuqC1~5)d8XdqA1zpB32fl<I$39HAWO%-gd|eQ@14Vf70}
zXcO?*CqF<|xTE?$aD3s<CjD-?%68=b^MZCAV<ruwsfFg#ASE6k|A#qm^YnDbhGA=z
zO!Uejoz5%6nl%fOi#a^+BCgP$WS3w?u)7GlL_*X!Oo2!yh~B1saW|Q-wB?Mi*p-dz
z`gSQKO|lHDd)9SQ^F74yd6)R*yV8<4T%R@TWm%%Ds$O>U-gErP;X=j({4CroF{tGF
zRSk}&=JFYg5it{Ho7rdi4uAG{X|!~Vf9KMyytB$a(@mmhp_`F%kzea<m28CY#^Sxz
zpqtv$702AHI15YDfA6Pu;v3s<_8l>F`0e+0Ir=}U=`M?7v^2EQj%zmgzHGveu*Tny
z`u*$LHFuimueAqxn-h|MJ>u8)T{-ttmVPS4>%QSu^OcX5hh6ZnNu$cF`%HWNxc<+>
z`DryJzt0E>k%;(gxzH=HNp;qqf1gPIs1{g1aZWG-rz_X@GrsBYWk$E_UY&7!W}LMC
zIlk~6XLQCtF>b>~*f(&iCnME8(%q<6uNXnQnU`~`q)CF@E6VgZ)gY~lnRQq@RhH}C
zOAT&=G-oECTJ$|6mvP#5Q>RzEnr=bu@2c*$$!+A)V}W~jCDIg~Nq3t1-IF}?`{Uls
z6l1O}QMBy+I6U%94Z+j(4+40v`TiG&sg>?o5Bf!~lv-m#;xYS0xL#pHYlYtGBloic
zfAa2oIUl8e!oqVnz*{MB-hnM%zIozB@j4Z=JGP*^X_tkVrv7FT{q`+d?)0r<<5Uu@
z;B_wVX)5uZj!Owe`T{KTzyWQe<S)%pv2h>>R?%DZtmI-3b?UIi$E3H`%_0G$g<xo8
zUeE49w3w20<lM>Psb1ryh4jk}BiYkgEnYt|(I&LkXmgUe#moD)dN+TKUZKy+QSKC*
zT|ic#?+rI*rYbD=!Oc^e#6+}MH6n0rV@*3zlW%a-zN`F((!dudi}}^YtQs+;d?zZe
z0*W8{(kO4%wyWQ3wV1Kd{nYbxRLhIOlYT$hC1rhRB^G?_L5Z=G_E#3FjzuL%i5;Y%
z@`1b&ydEwGHjrz@Xb6~;0lY$a{(eO8l%R0*H16ku_Bpu-sD6+0)sKHs8)z}vQ<mB{
zU;R_=dM;l={Kiq~0wUoHQthe~t1T{Q@YVO|>m)<Tsw(D2L&rm}92`AJV;x*Ohf*{2
zAO`5}>dZm`sft2CIin_hf&5>A$cJkSy3U$m1z}w_WNCg_N-U@zky-=W&JX(GyaQ4b
zvCJ+BYK46424&MXw-mTngqG*-)<`h?`9e7S?HO&aMnSF&Di7i?RvX7;ncY%XEmImH
z3(s&o1RYAX=jA==@|hlj5B81;2sX2z)s&g{go8rIC0_ru^=+ql{;5X3UzAGYxV}}3
zosq~mRjRo^vv|~w$J10Mf&_-5GhUbh0Rj2n*>~f0A6_)<+>aT}`BwPI{-vBpgh&tq
z*<Bhg+2za>*Fu*_dy<TWL+-l)F_T%_dcXJSX)evhvPiqWn;<P6<7hZzW&c`(XtD!?
zRMXxeadEHreaX|R(d@Nuu@^TrFJ_co#`!20Rf)&i(xHZhX*la-24so9x>>zbU-cV)
zAH!*0EY`}x>@t}v{3aNW4*6$Cyly@cUb_cc^EtDvyu!TfR%%UU=@afmW1SUoLl*f|
z0n`0Lo*25X@L<r<i=WyjF!YVhk>pmZpMO0zIeh!Jh%>F-cWj`$J>&1h*u9)nUOyYr
zW4OL^Le$=NF>-Y}=sy&lf|rs%a|+M*Z#i1-MM9N_0+<{v6}>!3Nmcg`f?2x<x30g2
z+v({VJh{F5jfV>b8jec{t}-?tCC@7yl=g53hMhe_>z+Gk$v(-IM#Ist$8~Cd!*-xa
zoX}mlB&dmqU$h{TqcBlfjc9}fH5G~kwc7I#YO_o2w@DK<TZW2VK{|LxT4PX1TXuN>
zpBs7Yfs~<{7sgCA$inzy_~-ZD?;eV$m>@VxX|4cHoY2Y;9R~)eJ!lV!T<72ISCSuh
zW=-@#ldFj*>^wFvl{U{BB$>U4u|_jgboPXWZ4k*I#fhzuZ?X6<<kN@$LOcohfrfv-
zA1bY9<~L|^f+v23tE#_)G+UaTB7ej_E>Cp%_8}rjGTozDc7&;-+g-;B=1@XX@j=Hd
zAf7W`cGykOx^SylsylJhCsKQBSu~$$G1TMgfx=EUe<fZ5Zfj&}kp%Zh^7+Xk_)d%N
zT!ye?^8x6Eu&E=SZcj<+ViK!!lUvLC7rJ#POZ!j{sreNUF?W?wh`ZzxFqfa_6*I(f
z&e&b#{m)ot&#JIhg$p+%M^EaL_Ugc1Ds%eZ*#^TOMxeeYihjX(1cj&)4xL&KvZ8KK
zg1XvvFsyDp>e;ris4SA{1PAPWs|*70h?UyhvyDJD1a9;HRZi)t)QHbAA%ShD?{Ruw
z6|1}v@2mY(Z*yJBQAM<QrVo;Dh!KqHAYO^z@(Hdc%yg#sLq@(OFEl=$^*D8qF;)LI
zf*9^-rY6pU7uFUx^U5S;!X6#Ngje-wEP8r}{kaf^s|=LW!M8AXQ7y)eE+%Z`@N_R1
ztGU6Z$D0}%Wv4}Las`XeSHg<p4HXtE2j9N%nsTy{**;xg5*2RAu!I`9+vw$9;&YGO
zZCxdW3OEbxe>%mm<BVj!+)Gz16iGo#mb$|>3bTo(Oi0E<j0m_Il+fSYF<5_zRDi`U
zxGdi*$!Snq8I@9)?sYGp^P^)k1OgLoca_(@vN%h>pZ0GSYI%v{!G+)+Qdg~2tPdei
zHp`72q!4J3DStkMrw1`ssS#$_8?AHuJhFd2I#t!A!yiDY)wA=5XClhqY8AEM=`-5E
zr6~^Fxqu=FsYoWxXtth;a<B04?!Kslyz$KJ?n-_{|0~6)@TUX`A#iaNv7zc!s<DTx
zIH6iy!5^Wx)m%#FL@4P7K?2nF^;YhGARa;U9%zmV&w<mPB9&eRK|K3pOK-^BJmzbb
zIkvNY%yV(F+_LO&<eYAHjp9+ohg{06vF6Rb(1P}Wn2Zf1w<hbv+s-GycZ7<xUZ@K^
z*_B2oFAON`ws%?3_+WH1zTyhc*EePyQ;JrTn6l!mAsk;Bzq&iAf!vOLZ>O5oG_!hn
zIy7o=ja6AB5lAva$dZ~mKgNoowS5CTE=IXG7<C4ATpupa&UuB|6-Sgj=|()jMG37l
zXonGOZDD5&vN0Aw3i^zOktbh*YGcMXy?)LxSlI*i5LRXquBa~Joj#F?pJbcId1nhf
zbm`i2KldxAqGt~$)QvXuhB3lf!oG=E=7*L0+Vd=Hxd9JsX1jo&G<UvShdMeoBXg|h
zC0NiT(Q!3Jum675Vn`sw8vse23E46<IJnt4eTZWHE5diXSwXA(r7T_P-zA~LId{Tp
zLTL{cKU3E0d`rWGe&@gWw^`JG?(l;1HoVs3zI{o^3QIi>JBFVxO*C24+-raxr+tbl
zm~h&~?_U2io=XS{Us3=7uQZcU)Vqce%@XEFauoipTb8Za8xe~2fUxfF!f6$04tG_j
zb*)|CtM=#Q{cL)0R=8ymQTCF3Fn`_-6W%%nZ9MbcN}Q=^ShF-lKFu0_4-X3SvqqOp
zz$lkTAxHbvJgV2q$Bmid=vlm&$bR@C)DUYd?*L^O+G%Lv3uJ#C^`5L1vqQjuZXe?s
zbWGvJ{Y&R$k1`*@GklWN?uOgakN7fW>k4Fam>veemp`;pTVh4#Kwi2aBn`mdGK2s=
zS1B9>z2KCs{n_ZN>^yhRM@8HG2tB)-)i~pgEL_F8{Fviu)_#XF^i809)4eyx-$tlD
ze2!`85V|5>X`++c(Yo-utE#_vx*`vE?6O15<TKz-HwHdS(0d3Fsj5H_t4pmNVg!45
z-D^aQ3cKA)9sR3Z43{V1+o>Qx2dTj|&s%Sey@K~!=o9scW>F*YV4(mTKnn#uZ8(+%
zdf)3RP4wJ)9`Gh{phgBnv+efbAi@nUqJ-{^!aV)*v2;;c`mG+SiqJtpVYMS7^~}B!
zatW9P<eR5Y#x8>{u-sRX=@h~*3R=`?X-@ey9&TrS^u@OyF(17)XY$8&`=H1ib>;k$
zf4{S}q<D^FDT-*)|B619!S8wasfkX^<BY^;jTxPDJ{DS$4;UhEZ4C?FY9@2_!ZWj*
zSST4`P@hvGWW%p6kt94i!0G+j_0Z*cnqath)NT92{9<foI9r&{fOxveFv>`<Un8)>
zD$_4bp=^uDohEg;c^zZF`=B*w^vsM)p7PqjMm@eLryvpb$;^`A?dkR^(laf>@wugk
zG{LDSerU`=LjGB?!LaMs-#rxpv+H;Ci_|YkUpJf2(b#zU^r4ipblbq)w$olSrv@Va
zQ{?|Ce_Pm)*!YxIW`_AxFb&8*=Lgzc{4Dcr7D)4NPZ1x>;8w2pooj{$93NuXUiP>Z
z%sqKa*kY|->T{se{_SQtmlgPv+kJ7AEx%5BL6%}F;}lBLyBOwwq<v0ur9558$%<!d
zHqKO2#g&hVS$|jeCNf#Nv<Jk444+*x2T5V+kR@o_cVqADfEUR5xclDOnM8$<dG6oF
zC-qOcJ0gqS#XAqYYF0yj@XxQSB^MIxvtWDab-~`2cCyG2OzmW_%VZwy>3<<@hh2oO
zdb6lfe;xgIsvVgFP}hG({!^zUGmZOKtC;6hYqSD($-Fpoeykb)v#P^3T$p?B)`aIQ
z9U9U>56{;;%>_HclV;$6Hvo#>t9@LfT=4#?%_yEs-@nl1<~KN5BXjQTJ&)$EA4!yy
zx)KkO44b$fVGpy~O(rL9KWL1RyiUlH4AJ49Xgckg0fVfsK0e(Ze{kWVSsc|V(9H1B
zF3CLLb6OmEjjdO&4v*Z0_6}Bk^FKRuEPX}a2DNo{yY(l>(wdzUN;RiAYeUcaS3x1Y
zBSAtQd;=E)Vf+}i!CUP+IN-`3;2yGw7xWYKiSCS_9y_qQcgYN5{_wx;C!1PNk?iA>
z)FU2ZJdzo%_Q}st>d5+b&QKdI>I^RYAhZCQ>9HY$zgJhV8}^a}eu#omkSAONuia93
zB18+FJ_i9CJ8>qb$!U#h1$3xO-31LD6G9mwwt2dzw$P_$wr@DAM<y^GESud$tlG*y
zPqZc9e@<8^yGLAT+TbT8tjccJ-ZVL?zN43*9`v;?MZ*~@1*#mxtv(ZKjxb%sfp!T2
zK~;UCa*c_e2_hSm-F8ETU-6$L?VeQJH{3ej@XYG9-M>WFBFDn?3Y)iAC0dNF-uh_R
zi+rQ}q@3souIGOU`Yji}wVKoZ7U#gKU<C(FVSMd<&4WDufEju;!D>TBSPpPfZ6c+{
zUHk>@a)L;jre++Y`Mr0lTro-Odj{%i$*;6h85zlIs$4rktLxH@v!>shPmx5!GQg)E
zWS!>)J#pyV{J}@H8ikX_BWN!LJ(C!01`iE5pFo!ZRh*`E1rV#d8ZCWu*QY!3x}FSt
zQd65rE*sTS_t6sAYc?0nbJMfe4Zv?sNU&WamE~fN&74@Yi(pn=y|!4rGkb*oAbFKi
zcQZdQoGk<%N0y{ZOhWkiNRixp#IfiX)J&@Q_u}e#DmKPrDtMV|m^86mMRA5@xon-x
z#L!1827>4$@d&3Uy^*p$|8Pm>!5_HipL%P>;^LC$dyVUE<_6*Y<CXwtb}{o??{(>%
z=!+kA_c+n{Jkvm(_=a=REi09^TJqoXe@Yz4Pik{cTAIK1Apa;%scgTmec@KMP>a&1
zjPDa~dJGvOiy?2?n|*d)T`7`VeK1?y*yF6fI4)64z#$yh)|e3W`C92G_xRb1@ilhu
z-V`$ml`tE_mn9j_IuBsQ9v%ASGQgQ$th&@Wh;rNIXHLQ$M<kGhEvUJi_X|u#sj<74
z7Fr<5OulYMTO2`Tpv5t=`f0KE&`M`vx8z~~|B~OelINH1PBf`U#^q-+5t+1by?4ru
zW6=>IG0uKWCHKs6Zp4w_UH{Y;-3NOj(W~4$@BLttN@e94AXoia1g?0krg+9n$lFH?
z^@}+9JFOZ2lrkgToXjU{B18TQxe|70d#I1~XT)b)b3*Hm-MyNcV)a8{#w;u`VTry>
zCi#x8x%O{UPX}NOnjj4@A+(Nh@eIYhcTK+>{yLy=_3KKLI!{w%fGs}lqF`Ec2KLgg
z_Oe@EBF0qnX5@fIGkkR+8~4}F*P#UJf@g+sN$M`Vn>bzEtJ*%-%g*`LAXVOo`}*j6
z+ZaCu^&XO5ZjJPuZn-qH>{+7oc1;Mssuzj0mzyyVK05|u(yTE?EE9Jw_!q*Hd+AQb
z*E)}5WO6SiuugQ&xjX9j*uXisxGst<?l9b_Hk=axtNG&#h86%p5gc{Ry3`zS(9uj9
zJ=`v;26F&1z%x5CAo-<Ma@C~x{>{R3CoyjJJzd9H2~N6wd%2miHu(|k)O)=fLtk23
zxbkQE<bN<G9;u&i@g$y`7<d;P#@Mt5lY*x!RthH~J&R4ROHn9Gy^IqE@LWJWY1Ad*
zqZvf~cTyI|`}of$%&MULv@yw!bkQFRiku7i<s{eGCt?~9PT`T(N=qYcXmLPOWI!-%
zEJH1ApcVz+KL<kl-!SB<v-$fPygr!W4pLh1Qr;|B$^|VL<VlfDkP(!cqB>Rw;%hgw
zn%kD<YYR{6S7)uUwSPGwuvT<c1$}$(IxCE2d->B>=}XDpN2doiowZPR=pZnupcO5|
zp79^~q<+ApG%Ln;{;9;-`m+DXh2_o{fO`tHi%9_FFR&vO!ihl5@sRDdu92dTbHi^+
zTO(SAo?QN#mx563&##kmhr8MHe6>pEj@Wq^2t97(3hS%SV&(krO!$5ojbpvpzi{(y
zWbC6Q7sEKAaJJ^CGFsY{h6NoXi98o^M>^G5=B~y$LF|qD9#h)6?*o16G15(Cu#@XP
z#%IyR;xrwQ_vBi=7NeSnl@BkA6f&m&W~j89oC+P%RkEjiQLr{t@e49NDI{~sm$%CP
z?G7(aR&9D#p3J-ZvnGwVz`=>Yyj8So6K#-D5w0}MV5{d6f^-pxyrRni(RA`aAs~vz
zFf*c^Y{f|OnqofZK2@i2Zs+M{i8JLDt=Eaqyz!^MHlAkhk?W=J55A?*BsaPbHEJ~{
z_nh52X4O;}v!7aG9Kw^={PV{aVS>Zus?TegS??|%&oAAT3!duOk^Qi|x5$b8JsR{y
zz(Wphf8@`|jqx}$7s8-*n&6}=<waa%%p#4g;<?&!UU`?jd0C>~uRHEvX)u@CQ^ul(
zTs`#z+Ks5gHLx{##*n00dM5eSYg>uwc`Ui`GHU-IrDAJQ!vH%G&c;&i1Hv2c_YP_g
z`ud>O7dY!E{SAOvho{_;fKwB-E)E)~9LcA4zt`S+sj`uTF3)+(0t1-1uf+YVrFp>W
zm77ZJ6xv$Lv#{lxS6^c69Y%vy|8TYPSjJ$uaAofr&&e=*In37GWyT(TZ-OSd5n)$%
zGz}VY;fcKg;UXQc7`F=E6Lj}%z3$CV>jI@(+P)><#)!gtE#CKg%`SwiDv#^VA*2cX
z)MBh+^&Z%?cWhXpU8=iq*Fc@!N>%?y@&#-@4WbN>aJ~|5lH5c7b|fl->mnBL=L6pG
zF`yb+m^Q%5cWWT%j(?C3uymLQN{U@fkbK<Xi7s!(J2--eSRu6%>#`e`#l*&`PNfg-
z2a~u^C1#!c&2!v#9Qv5w@Jx2(*FnmHPxktt>nUI6>~yw!1J0PYzoU$1$a=j8@L53|
zU+gK-kO8p=5PcY+_=*{uf-=`y=XD+t-deP`99GrpAM`juP4vXzWXsIGCR3Qt!NxeY
zDZafP)>9aCiRYKa6!-HGJB$&5srq{e<s%VVKFZoD<_+n?&jqWE(fZU9D1<5lF&}g(
zk-SDD;^jeE)g0ElWZvoEd~NOnq=R*H`{jQw%sF19i{^L^w4i$I#ZBsV`6<{M7MYpu
zXWqm=h#F4fZIT~9@b)iT;SR4h#@@Mc_{q`4MwdzYlWNA8m!$$@X4-%)1g<25{|)Yt
zx`H9^ubFxyxBWKGwr)HvT)O!m`ME+yu#fNom(P~U%_q6OGbIZ(!GLiKF~(<>=Qz2L
zXhLL<Xk%)7OIzl+-l==_@R=ZFfepW?MAGi9HE%!jPF;%P1C4<OX`+a_tc{>)Bot_X
zqGnowixio;KQZErOdGFCKQmK0btZgJh3vz<$w<$<UyF;N)nq)Q>R8alB`X&{`~SX?
ztV(Q3f;p#qrikC?K4W%Q$H1$dR~Hmddg&#`2@}6s`se1eyN8X;hCMGKG(SBlxz?@z
z?D2o70^Dk9DOL;HTAq_+-JsE9?lPkHebs8GG19-eIOF)y=UfrPhcRoe-4|h_c9;20
zWe$@?5;v$fkfp&4maF`c-zH+x|7Lh(KvEZ&$*VJ3a=*ScT2isdT*0S&WQf1BAdcMB
z>c`PXARr3|DbkOv;CMoy)w-u8VEZ_J^ZMG9rQf!y#JwxtnqOK{Q74R)08A~x^EytW
z$C*N_yR2qi03>fVnyR~Lj7{R+GQUM@u?`6b=nJ`{eFx=zgjSfk(cpB0hxW7bfa>9v
z2csUlS1!8G^?1N_Pc_jw*0l=vSZq9HWMA0nV}hZ-*<+rT@_Dbc+vL|=Aa8+?VBv2t
zh9fZvtG|WQGO6Qh4<rI7xfe^y0lyIT?NOW%cHI9kaQppzRsQr;x;Ulv*yiyCEf}=s
zLU=$<6Gyq(EE>@#lks@nOAeXJ<X-$9pF>YS(dTXg_drs4xk&3ux~iPYgU4&^tksl(
zkKQ^a2}S_VBu3TBAv;>}g!hW$TXKbQsm>*gwW1WcxmPN5g2XM72xIC`m0pdY>;O6}
zJtgf#HO^DOeB6DkX1B|JmYo_?66pnhz?1u7PV(rcbX<0W8;OA>yjUH0h-!)HhbyPg
zDFfcT*&$?ZS7^Ho)!x3~6ItinoT5J26`bwl@6T*r-AJB>zn%N$IeHjW#wQZ5M*eiS
z$KlS%earLzOu-nOUv!hu9$0xdd%JWtBh8A{Hc|i{&9IkXL8j~>8j0xs^R{0Gp4%zv
z+0)VW7`mZ;@aJe`sG+Lk9uwSY^$UBLRS+AW1H~s@ymi4++uoqS6Caey^ScbXV|Rfy
z!`|*?1ECm417Vey3JCG8;qs>uWUN~caz>c1h!<_tLn(3*7)^(<nmCG>AGNt-vIVPm
z$R_Bor%dmc>mA19<KDrTN)`0Th5Fs}mFurdQ{P2U9WkOjgVV`^5~ux!loyO9M$MF=
zy<A-i)Bp=3%l&T(_rwc#|2pLUrD&y&{wZ`Mu8W@W5I&T@QpNL7WY*Rqa>GH3CJtgk
zcVLS?<FtnBRs3jKf!Gp5nD6TUsp`3-WqyfBgK!D>`ydT_tSwybj_cm;my#MuY?0p7
z8B#{j*_x2c@((dKHL&tov#J2uQrJJsug?jb_b7RXeqpy!07j&9RUzsN*VRWRb&0|q
zyT8)2JuC?&l0u}o&s#O?_7Y>?Yb6@F(T}RaOTM4`G(7J*z{GvOG9@A0dG1<Ccispw
zNlPF9KxOcPVc`2y?^{KSrd3~lm%qfXJNh<cdMs+KOyuXoVUDEz-Cy)UL8VW)t($X;
zLuWUtI9B7%Td?!nKX-7x2<!gfp5Rm(ttRAjep>yqYMqKsbd5l_7%N%7#=|GE?(X&^
zJ0f+%FAKAQwL^jI*Uu88g^CPfbsxI0(rpUnwdBp#(sRpJu8jr>Xp9nQ=6yK<$poie
z_ijQk_nEMMR3@rqDqy&>itK>z0i`|*&n|stbjME5QA=qRN~a$3e0&<rd_Un(yKs;s
z4&<K~i7gk9Rc#@$<AY=slJ`<H5($jh{vMhw2}&!k(<7hySnvO*Cj9pIoFO|&b*Qgj
z{_k(Ag3eh{zLFK3Y7eXvkeQClAxE!xw@}I<z*|hdCp_jqW48Z59_jBAWct25Q^Utj
zZ~Ue6_b?YZ0f3b_$$fO7vXjwp;6pJyMjXA2uuW<9W<>oZomJ57yy(~i7lF1UECpjc
zmk2F+<06BH-uF6o7^76Wa=qfb6ntw!7~P8DY^_fwZ-9~5l1?nij>CsTI6>-)c)6@%
zH{s60W_6{jr|d&(V~#A7vr)aXA1agH^Lql+8D0`1cAzoyE>kNUpq0v4VvW!;6wX*v
zsVt@46YGnRstI;(y2hzB%M;u>`w-)0ZOxTa@=&dCGzij_@myLml2_8t9<Nv=K2{%`
zzuR9j&?J+L^7gh`{V+r+JTwU(Se^@VT#eEYfjW!Fee{27&WQ{>F|2)6r{0TvOg_=c
zsH~1OuVByslP;yHrs&X52?K(H0b8VW*IuN+`&K*Ap&{s%+STBUV0CEOz|(;_cmU6e
z(wPIM=u9he3QsFvxG|S69FP9Q!%r%)Jh8hNb|}1fA+sTAuk1fI@$5dK4)s(rG*XPs
z#EXAz<f0UfjWxBB(_F%#MNBTA3#s79eJ0=*WhEOQj*#E#^*7O{!)kN8Z{NHB-aqsQ
z>`q}!DwBVOteOt1>(!EczH{7x-6tg%#ORVYV5d+KDW2H$hClm81D4Ry<H}d!Kb=0}
z>Gd-g)GNg2CyjcQcKFLX1shB~Mla37(gDB&I~LU_La^^5?WaOY&aiug!;ZKA^c1q)
z)SbLeec$j;Rs6o&>dffeM6UFBbKMt?;06#LGtiYj=rLt#bi6vm0P?K5WB{EK{Q#8v
z*c<FEHDWm#=`_KQ)nP6zJ&we<IzB`0+<tslLK<P}*sYl5p_chOo_6|R#wP#z5nGjt
zxY$REdxEdK<YQ>MPzh@0I6T2-D}G=9m72|ul#pXf-r?_=uf5gPf>x?qd7rKID=w?)
zr1Bfu53t0psIUrk*lNd8-qAD0d87XLkZhFl4!dLjAXWFI8=~lZ{nDJG&X_o@)tpdv
zSo5A!gIx?kJWYrb^E|~Nhtv?pleZCY_?=<t70S(XhdmIQh<^QsG$Uzd@3uU`pX|aH
zs-!!37(_~?tdteZWY!0cY84b_Havto5fY$*%X4a~L`NdYym+Q7<T2((<UK3xrnSN@
zMRAV{^mz1i&hlZ*32J!Rp%O9>fhEe9EGEAE7oxaz@mh09UJudZ2i%=Xm;<{Of*qqL
z!ZhC!+ZgvRALJT;X5#*R`@}2mVKV)?$<uxJmOxFC*HcK}GNSiWMcA&VgOG?6$WS9R
zHN$`n#II4p8>mu>*3X)bw}z@K$F|J(QG>!VE9CPP-dRV{cU)Bg+_-dgi~Hp(#^F4V
z+5b*D^NTxDL4xc)SN9ka8N0+UxLiWXb+l4x{eInuUQph<Sz(!QoPN7A>8gaS*r7Dd
z0i9FiMtOXh*fY(RNK+I=*XFO%U8=2HO3B{J@BX-1Bj33Y!s)%Vr!5wFiW+x{r#Y|p
znI@I0&jkd;H+xEGzwdl0WJ^4j^$|LDnvCblz}873zMV5+MW9v1fD>Q?zJ0v0CQufI
z9K(gASo$f-L?0%xW}YP;9Of(qkOv+;4e2h*(1ah}!q#oC(#)nKgtxX_8+D8Jw6`(j
zeiDXTZ^AQ-<JDifNWg^Z+0!n{ProI5JzDiK<P+8}-!zZM-*-?>XiYsj60BNz8Ru8t
z-v|}QBfuF4J~^<LSKi-1gh?;;=)bdSxf-1P7yTNxE$P5Dl=nWIbD$eg^eF-$M`qx`
z?Pxcbwo;(*7u@W#y$Vv0#u7V7Qiaf(!aQaRE*M}bS#0DGoB+usOR-X1lIP^jW`O?1
z^C_4Yatmf~tLZZ+x&xqF`loLpBd*+(kuE}G==wZYgIhum_s&m86v75|&jk9oI);*i
zEv~{ln5y$?WR+tJ^(~G@24JL17oH^uGep6AG)O96esc3L)aiF8A+dGi?-(ZOTi&{x
zR5SUGxqtX3Qfaa-EwhW_w#XK>BlNg|^&y{S4}Oq2-x>Yj&BN&IJEhg)hMiZ3V<_J;
zcq*7$VFelx672g_kBtwS$~@ox#Dn0)tUOQ~v<-obqDX(7r*stZMMj)^+}U?J_gp}g
zbCbX_xc9P95aOE67IFd@B2llNH@rO{4m%U@AJwbVa`uJy`_?^e^THc=2^28lRsEzm
zyjl{d5tq)yn5*{?4WyJ|CrC+W2J8sM1$15qs*X9U-07!$#~;|NiVKNyV;_(xypOWx
zYkIorr{YbITT0D}u<fSul?o%B^Sw(uu(_&2Y-Ug+xfGU}qsVm`q^qHU#fdW-!V^2l
z$eY7xI%<o4CZ#eGcS$kqy8VMrHHy_OgJ#tlQ4m0vvwDM0`9wJl0eus*k-_Q|qxl#|
z#n8(lKSz#wkIR}9Mw$Ppt0w@ZYohwE+sGeehmMsh#Gg2yVa5G{;OCs_?^YVk1<nPt
zgp|7S6y~i)Gu@#(HtJ!;s`~{y@(FvC_0CRwK!lB#nZ7D81pzu%<2NVztwvn0xt2yU
zDd2K9MVcyxD!O-I=8b*|lCJPQ;iu4D`c#@ZdZU_MUNM8r@c4g@X|owE6+Txy>X7pO
zD<+{m_j~!n1RV`?#v>dw)rnpqtbxN%G)5YG6jzr<1e!UX)}HCuNNSc;4!x2bq+MRT
zP}A&C_TFa~vm&3UZ`yy_L=#!m>&v$bxz8^vmpxL(x;F6CIZrqmvCUBIf$%j(_cJne
z&HSZvQ*rOoo-p5<0-3z+B<Cht=X60()$!W8Hp^+1#wbpB;+}Ye7}CscS;Wv<L;kDO
z29_z<c?&bJ!0RO#%6A+Mdq%vvC_{s3o5r}k@P>p@UguE;1he8gs|y8<2`8-|tC0N<
zj+@Yp9#Or$twTPH#2e5<@+HvS(q&QvlHu;-5uv`O6Rk*NB4-f#XtOd5{_v8Idk`|p
z*f=~)rN+WQd3k#e4{X~rQxvM3d}k!P)Uokvt<68Y6$)o17Ym?gTY2E}HRV|owLr`2
z5%5@sg|#|%mn(`jj1tF`IiyU{DbRD7LItTl{rZcUDF~<E9fNsP4Fe(7`amLNhVI3_
z=YmlF<dKrf^DdaLK|EOU3P|EAC<c%B*b-5Dv}}=rkuxf=@<PAem@pmk1!(Bc1PEjN
zlloahO^prCFUPpWkxhKE&ygY>dSFeRg$jHZv^!L*+e;1{V%I-CH84=?vrr)Pwe9Z_
z&m?b^-(N9#i61IGzg?c|qdutEit*3hbg*vVq4MZC-)kfx-$(u3a2`;ct(fp$GlaV_
zHuo6i!!3~+FWa8S?k}Cxe`_VoGG<btvsdV%R_6q%LLUhd>D04cUtmN2e{l^?*kebZ
z;H;qHu6ty7iu%f?>%K(QAxdO4V$|y<BZdlfC;inWKNa_lpA?(uJ^_;bEZ<K9XSXUh
zXfI6FKlyfrWOgYI!HrB6>=jA(8OEa4Uv`?WwgN;{gTh8Q<2;&CKS;ReG&DJSD!8<0
z?3J?Z+tNe+jF8R_m4+ZMKkRz~?);}0pV&<g;wU)*ZesQ^BVroo<{UbjW3;c^yPv-8
z&P7R!Ft6DJvAS*PUYj0ZSHn^23@u7MLM_ZzO6yJ`cK5SU&^hN(EwH$urOXk~V|M={
zZNe9X`KHb25VL=f;uw}Zk8Jngx`4Tm(e6m=GRa($DNtMBa_TvidiGR#^;Ezo_<x$u
z7ENszWfCw3It%~>r^cRm_^<&?+GRlK9`5P4v#`uehfF`a{5Q<UUVqvJJf`gUd<d*A
zwd~L;mZDSkroj~TaqaU-tAny)e*D@Y^SzC!_KU7@Zj7P~P7p*ZtX3G|{CP5!$UFig
zTo6azqs;02Wk~oBbe{7+e#e2U-#fhK&rb^HuXQ_Qy!~uapPHmW^gS0UzYHZvS9YVS
zRNF2YGdle)h|ddOQ+}rTg=IGQuHt;MSI8y-SJJprusatb)%-5e6)d>x%2+ilV67rQ
zHNZ`7gqMvvR82X#9Q>lOX+nCc&v>8Z-6FhgdUw=a%f&8fxw<Ftq_W6CevE3tN6UpP
z-AKWf?8@b)EB_g8aQ3>h8PqU5LMU<0zb$CqcxI^2=pU$K1^#d09f3iodkvImE9rOs
zn(hdG-pf!7-n;I@fgeAHg(OHbpTj<!w8??728vGHj<!XrKXP$a=byUFG(!}rlU7oM
zMy*t_2b#DCr21+|NpR4z-eAemdxYxa4@_@e>?SWFqYN6oGXrrV3RhvkFe=K~OP|uz
zXx8Ov=9C0g8!AQRh#5N{`BO1@|IY&3rRXw>(W4AE3a?Zro#H5$bL7a84$DA&2M^{@
z3?~(Lw&G4ts5@*CO3O8O7E)Fhv=Flk_99y89&*;reAR9-(BtQM$vYjmT;}BI$lK!I
zBx^5oaBk|Cp0qk7HMxv-Zp)-0L1UKLbh`$!=DiLgGWI7?>ucL3U5Dg38MMb7zdit2
zSe2bUs%IuvSDxPRXGaE;=D*kBLi7<RtgJ|!n8pLcTkTgS+8x$M<AU~<*%%n(O-`t^
zevRyjyJjF}ETTHENES*7-B_|`1RQKY=91<1TEjB>c}j|=A9Z&}KcAOuTWrx`xV=k2
zOabC<h6bKxrn$}t(xC}0lOzsxLp}n?u4Nw=VzRb_zEvrh6WVz1wp&j`Ki}CB;slTU
zHQZUjk{lYLTz4rqV;KY?@2M1islLXeY#$Lz3_KZqRBGl-b6ae*EHUO<@BU2nE&5&P
zr7DVrzgOXLzpf#S+X?S=A4Th1g^o4O{iF%yHyvEr1gD<K@g2_W<2|CQWVyUfx2)jg
zy*r|*^O;F|2f&HSXdi)DNz~p2#K0qD9xBtfgYBy9WX+L_QlGgC@mDSNE=KHDS1rVa
z@jNv-l!Zu^%OE0#G-iU3WT8-CEMp;MAO^}cGa^J(hN<?&QmGY=>XDSjcsBir)k<>*
zw<dPMcv1Z#hFxMiUAJv<N;dF#ve(W*4_964pwJW}yv!F+dJMoKEG=<zsEG&`OgJ+;
z21J6IV4$4pYAVbnrsMMdlXYXEA-9Wt4{O$UQS8JY<2uWfqwU`Vu^}Z>_PDQ7yG|Al
zvBl-4^I@^RkB*228F)zq+Rxz27-<<I7`^ciE4$pA;IMd%J*ga`tpUVoP|kO|C1y-&
zt(78;Xp3geh@Yz}qJi<>Ajw^M*{cW_ApO<M-;%KGUAVW$C@i5xSw3ns^eKXdeAV$+
z>=%^#WoYjBI##92VqJ=T<2l<o&>j7|TRQ1L&}8!`>0Y8%F-edoZgAHYfhy7!IZ(As
zsqmZ;bj}Rl`*+s$b)UC(!Kf5yv+>Yc{)B&Y0KnyT<$USfOi5~f@Xpz2E#K2Wut4m9
zGaPN+56RNbl4qYt`_yiBq@}~XI*U2%i6Qi<-|fekTDK(`nmdeupa<5Jne$YbtBH!&
zIN~)W8{-#y{D7=g+tisKRdd%Ju*a^DXi{80D$+#;C3CmoZjBg}_NoD50(vj&8v2T-
z%GLil%bK6{T$+zWhvdf^UW>n%x5H`d@Dd$`hXt*HV;)qNnx_M6G7AsJv6MmOw3#JC
z52$~=;Rn#&nJTOr;i_)mkQTCLQ!uW4LlAHD=dKH602^loEnKc7GewO^mxIB6Z_hWZ
zQ-;U{bh;&ozL(4LFgzDwiGyB#dEx0E1+O_mEpneoO)KS><>!A!Iojfc4W=#p_azsZ
zlcwGt%ed)vxktgTP!N<c@K-R?VtH>VY*viGSQ?G;pwpTWx#&5f+{7Z8o_p`lcC}m#
z2D(IBG&ve#5LCfloa2NOJ;1#Y_vVDdU>EblzSUfzMW4MoXW~5<c?0LyIEk%!vp_UV
zg{$3Y5LRxv1oob6v6Lh4@nA<IYRLnbt2j-(>ltNsS<M&gwF?j2IrZ_0IvH8zG|TJB
zLBjX=0ZdT?vI`6(x!}Nplct50-OBi@43X>Qfy7EQKy2w}zr(9|+FHu!?eABgce+jY
z9?-}jD6`HviSD4AUG3J5qa57MY9Kz{+iler7T*vXlpuTC%TMp5m3R>%r#+QNxpkH1
z1#k6CUmO`uvza5X5?f4zlE&3jXa^T`@)6m=)*9i?6E6qg-ukPxueR5iA0)?J6od~p
zvEOUtdFr_Qh;uQM(U&YZC&;-E%?!_Su~?p~vxZfKmsPXD0~`RFm3QGm&;lV@BMR1!
zE&+O`bHbMJV?X8{nxcfSVjTaKjI#042RSect4&+i-m0MPSLnx837BnA;!J!Of!jTW
zkSap7^o}z^%5cW0U9Q-@=yCm`Rv=1f{>d1T_$s!YSD21(2M=`zOlpc05%tZFqd{Ob
z&>b!Yz@G3RtsYBN@yrWvS_hAx+R7IE;`~}~hz?2iS8Fs$+Qsjsfjk;X#mKPi3+#U1
z(l>UWY`XE*{=?Mq+zTxp4+H3xdX*^wL1lHX#q6WM5ukFgD+kFR@oCeVFPW`H=L~M&
zir4r=UWg1eCWMLjEnMy~JGW366moCvn)TNEPtVJ9yBOysLrM%j5{i9kZNWx^&Frz=
z-T%=1Tf}Cc)Y3Xklrlxj>)}ShX=_ypH9Mg4&7y>P#%1e_c=1dfe<XF&Efk6*_fNeW
zStzjeG%iLg*~}w;JAVIL1eoOQQ)lMBX0o+#L>12kHa@&dVeHjwt3&f9j}=P~wCq`<
z=e+9I$(qn^8;RS9Me3=QRRrMG8blW?I((FQen0YN(uCtC*Qs_kIkya}MmiFWo~;|s
zI4LGV{mF8qW$?lY^w=1X(+4qPGC!4RVV2Zhy1#HV^thI5+1?)c*{jB$%JuLNal~*r
zR6LooDe18E4V_u&_nlsZ83p}a?P(33zFlYY8$~=o6$k*+g!44vbT29)WU5Cym(o!e
z!t=`&bQFOa4VHu~ZNos=(<NL80VFN-xRrQ{!RY<sk3M@E$Pn}#-tG@!5E=}I@(tEF
zWWE-r{!1V-8t$x3*~BW2`)FEe&dH1Ycwb`r(eqvsC%o>Lep}Ns`4e=9E+yYjI=hn_
z&!x(bo}L+v3SpiyD+f0Py-=D48%7pNyC91Gh}0+uCpgM9T$sRDEF^?AX4O7R2+{fQ
z$Ugd)kIc!Bz8iIeb7l#X7)X;Pz;Z(<kx2lGf^UQgA$ttv3548&CFsL>y{G%GZu-bf
zuf5AH95n{9zf@+z9Kk+KcC88~yZvVv7rM$%5k6wAPM-e^NL2>tF;E}`N?SOfvN&gt
z{5z`&Hs`>Wp^7TYpHv&`gxU2sr#0k3wy^|m@5hR9aJEmJ3VV=UcWWT7JC=wq$bJ;J
z!?HG%n?$}W--Z_twwZKFsI2s*S+PG%uIeaDG&np%=1xpnK@uDD#fd1n4x?oLnHC4h
zqp-S@B9OWEHbilv)u?4m-Q8$~b90wSs=8z*^N>s!FNlcz&?Mh`NX8l0ZfSzsH5xYM
zFh$2T1%uR7HVy_UXN`&|6qTp5nE{T&Et3n)dQLGZ{I#a1wPKSz%EzNl!TD{kH>opM
zdZg3o>krF9zi{#S6q2GowU3@Wr&l&=jDmu<YBh9eZ8=zAHX>JNL>Pq&#VS^?%%dfe
z_4w3P>Ar=AxmQbL_A}AQqpD~FxE?tCqgcNbI66H=sazhNX7I#*(OaaW$Zos^P{W^e
zWKE{XzA5$wdVvz7@T&1pzVQ3!oP7YZW*=(oc&eax_x@DR@P9Jw5gX5t75WWu06gGK
zKi20+<@OS{a<|n5<l^{?O@}t(Ya9#IC{U!hmq0vFGqG8+-@_1}Q{-T-`H;>%r~l@O
z4Kyg!ocvh*nASLK(xx)(-$iLNLyLd3syg%Oo?&NR94xoxEC+bpg^F4I700jbEEq32
zqYlkrJ>12Uk}k9Qy0zrh9Z)aAcn}%>pK{1f^VR{>=BS)~WPa*Gk7hAg;>W6KRx#w@
z35DB@fTK_pqxq##$<Ayy#weiZ)IrrAu}RU8g}9uU(AOd5aVg>cW6$RcYw0S38)F{~
ztsPhs4F$-eWa!$C+Lc4(OG8|m6A5|15SFLh)>V!I=q&>nq=pRKIDcN-VIF4ZoJGkY
z+h`^;@8x(5>FZzbN^mw!Dd2K72(qDY5uu|Zpe#74MY11w6&&iS4fP|lJY+yahY1bH
z@C-<DEPcw@RaM`MBP=k>5+XNf>IzpQAUkj*0}CLWC{V{r%MNT>wX)x4r?fyeKhUS<
z0uXHQO7%lI-4>}|D<s0Kk4^?QAG4$ect{bK(0z|k8l{-)A$0Lpoy<Dr)j>gLVX`<^
zGDk-^?f6>gllSEd$LkjTWp|oU8=;D$7ENp|HZugp21nY+-kIXM2UhY&lq+?UK`FF&
zfY!^m=dJfXO^92X-^|+dhwr)b;0b-!@jUl)Pz8x@OB38Pc635Rp#Cc<>S_%Cwes@j
z0}Cc~ze6YYJ>$}%f_;M8MFwg>PI28j{h4e3oLmThD_qEthD8rW!_L%5w1faQ07J$k
zTxduz;XLZli#$+9iDAT!N$3x-9+cN4sz4}(xL?GFHLXW&YULjcLDS3PQb}iY{W7(b
zCq)kRkha#3EH)N<i6sPkQ&ql_Lq#JV4dNneV<X}_hMz5d(cEVDpk#1MBQu@|l_9{v
zoVWtxp9P3Kvp8pF5KNs^TwQVxy%%5vrW`ol!78n*5yJ7SpFNjwzwEjOYRa|>w|d-m
zgOrD!OKF4!YK37BbrA>smzkB=bWn?kYHnTfx~KF?Omx1|{|1WZvoE<fe_X?fll+u1
zbGboZH{91$QTi)4`^tShp@Kfo`QGHvx!lk!R|oRsr*87SDl$~qTFA3Qqax~#v@f$V
zYmPsg%<<wVvqQ<n)ETGO4owH*n+owo40|Qu6iGRCpBdasEuW%R-3^wz0g?a~(K-A5
zx7Fv~y1jdPT+F5`beBs_7v#VkmU`>oCq}{i=H~Y~RsV`%+8<|}AB!kd{^8BLZD-&G
zd23K;0RB@I#FJ8DXHv6WQ07)mqU{e#a!AOBikls^*|V%B6|sqlwI+GStkjj<Q@I_@
zITBAaDsn{7_x$=gvZvr#$Jct=&ip)gF5qnvr=PE)u@n*Mm^2#JoqGSUM=}|EL^_P*
zOIZ<zRf?+F#g4TQ53LQYd-R09vo9Klqfuh7L407?VY&F7E#hg)rr(p-(ljadhZdLN
zpi{b^+`XP+WuF}F7}ZL*{AbRtgL*EuKGmySo{|%^n`@<r%@AE{aD2lg!Y=o2!9I@C
z?beaYuMYL3m*y3V=Z?i9@eL2fYd1Vkko|}QTR#%p&fl1FewW03`!3ic3Sk(m2}aTy
z50NGyKNtMZVKTJ9t1_i~D>(cYVpRBlJiU87)BPX+zs*_5VZ>18xXmGlIY!N43>(|9
z7|OYA4uw+ba7`iR&>Umo8fFYrPDQ6n<y=aPBsxecQdb9cRo8WWzAxAJ_x=6BO*#;s
z@7L@3d_ErcM~A!ULR#b88v<%5kenI_nUOE|$+lqRNFBh+<TXWh;5Lop<lwci(OIJ*
zSH%4pStb+ypV=Hl`HeN`*r(WpjjdvdZB!W%DpO1>z^ThL$7&k5;R$F6@P!QX5}0|h
zh)Zd|VoiB4)K*x#6e%wiJC8#(X9xy7$^AmM7}>f_xr{zYY-fPFZl>3gU}gcm2}Yi<
z6KurXY@Cmgy?jr2*5%`3d0i)du6v7>wtf2i#kd@CZ0E%VzBgLF4Js_HwupvjhZu$~
zP&jeL?q&WK$Ck;47%-F-BFB%5N{*7{0tJvt*Gy`_V&a*==LZh&hEo#m1r4JSD)3iC
zj|A={1O*)Z>GRcw-N{Ay@6Br}?%n@zg7>V{y4|5hCMCUqa|k|g2zCL+RzV42XZ>-?
z54QFawAFAU2_Syukf?FufMaZ^dDpm}3Nvzapd-xgN{w&**~ZInE{pVD0odtbyqaq0
zi{MY*oyEW_G4?KIS9J!`K0EHYk@Mf?EXH#C`@G1mK}p`AqGY+RN#sr7OIv33q^=Ro
z;FHxF_SS%ltzBLX5_!)6bh>8fjvMF~nc#91(st;4W!FW=ESC;%=AgRuOnyB*H1JtK
zLtcVh?hDS%$pYVr0CQ9F%K(858l?^FhMSH<rqIx#^-CIN$-aN9U!pxDqApt{Q)2Xy
z))rfO;(~#jxrfn!;LfgXmQRy7r`FOmK2K>_SO%v2b_}I<x;zAILcyu*#x(b|7Vte^
zMNYiOoQ%oJV;$dfRB>eSR)WALsd3Ta)uTl3`Xb=MDDaadz?EwM#c!cpqK@!&Osll7
zQjd4r{eGu|`6;=uU$0lmKkct(kggNjO&sg{3Ww>Y4K16Bo=v<Sxx8I&PZ51i=R&hr
z!ZlmRKfi<rS8>5reBaSGnh`ZMDjST_T}v6p=8{i)#jifU3;~xPb|qI*PU~X{{|9-{
z<8>K~wIR29BduUciW!<CTkbmmADm=g=(layP^+MYiO*+JX@@+ZQ@QegBx)1R!SCf)
z9MdeXX6j4_1nRWN-Dj0{$$lVmWQb^Mj}=}3(>MqO_4<hb5ZC7y>TeQ&P2dLl^b`+N
z9baO@wn&8ta9a(5xlSHdafp@e*D2eqt#f=8ql@t5b{)~tOj97-mIJ9AQ>N`HCb+Hi
zq9sgvs91;pc3H{6ck&bOzT`b8_psiPS2vm4#z#x9*)r`S&72Y{tIm{f0_y-;V^d+3
zl@rKJn|<$B>ZpVuMZ)6EOMXqwtxW{7vi=NYol>NC`M1kJz0oDLZMfI{(j&Z>Nnh7f
zn0Wh|q8OK2UM?f25!-G(VF}X3-B;KaclR0rPnQ{mifYlpa4PVoHZPf}1Ua#<FPoac
zm1jA$;=D=0Hx6FA)r+(6<kTGId#n^vA@Eg*w`=fL5|kdqPIL&Wv=>k;I9W7bTSPSj
z7H6Me=Lx&}1^Scbr{}CGV<;iftsQmv#8ay)N{`Vej|P4fBRenBD@gXOgAQey9;$^?
zF+5}uix^~bzA&9;T-bwWzr)`d(wOBmc!2aoQOTb+M3syF2fxZZY;s7idHlP{$L?FY
zhC*9b=#p6(Q8;$8J;<*Frv!F|fa{|(WTSa0{`30SM!VlTL60-*`sH^4VxkWj&3?g{
zfLJm3(jNn4s}zsJXOf%YO67XfImiaaWH$seK>mY%CuvpC9vVEy{^ME78Da5`SJ4p@
zwIwv;gb}~<GUqjwX4KomaT8WfYaS-62UbZ{9C4iEc%SB@z}Paier^4?i}adB-$v?t
zi0F4{hS(Q>QbGj)G@}Wmy_b!$iOm)ud(gxHEqXQPz2|FKXZ0i4xC(|SHQQwo9VuQ9
z>9EZDF&W(wFfh7Qec|lwwnwAd)KY|rY%Q4_!+|K9KtLF*>NJ7}Lb^=e?Nw{IjwqaM
zIT&MRul~>Kj9&M`)vc^Q)I`zh^V_S$_h|QHB~)EYKoYup*4;GEH7JLI1nsd?<`gYA
ziiv^^CSI`yi8yX~Pp5e}7n$c~o7l-Qz3Y{3Z*>v_-rWBi^Jhyuk^%G~Plc~QZ7TVK
zh^m`kZ-odSeXpys!=Ub$LRS+L?C0xGp^^yPpjL0o$UNXtD49J|5?cEpb&_V)WU~k5
zTG#8<HCw&yLULbmIIo|2TSjNVc+$4Ks1nQlVj76oX*15<^L&5hs(P=BVkK5b*tsN#
zM;DVkwLu=dm?FLHVdea;5F;Yd52R>XUO0S49D8`Y*MiZTYR$236t(xqEj%$5&M(}`
zma||5lhJt<%)wV8k@lUhZT;7RV0SRB%NqjAhkCrybC*O-u|o!iPs?WO+BkeAiFzTh
zD76&^*>M^aVoTU<<ooGogFs>p!@IO&FZ7REZ`I)usqGEjSc}_M!+M|S7_Wj7ib=Ac
zm#tZ9fRDP_1%nB<$wE@V;M37ax$(r^*9<-sA2+vdYBL5*;QuNX$^e&WS<pUW&J`Z$
z^^q2uE^%kGyM0%pI0ddpgt?2ud$2QW?k3sxnAxOxqpT)Pr5=!P2kJ8&D9ZN*prIKu
z@Ak&-Ge;u~<LM<dfkFu;gkEWj>$%Q^0AekWy8{uM<RA&UMK450_kQ0uO<bt^kSiHE
z004E{;0T;ekted%10CT2Z3i=?D|uDplwe;>a4Q3LxJ_*;Dj*<wkKd@7`qf$C&|vKj
zp}EJCp9xvM2o2R_g>N0U8H;wmXC5A9WKY}Bxh6;<5@lFz!Guo}OFpTj)2)SneS0NF
z+X&(x>J-z0C^O=`XwX0SZhNL9WWPJSB6~l!$>PSRh@ii!pDJ_bUM(lIM?r^tJ$;9)
zFJqw^)R?RSe-6aPmlIb#-yIG6$-CEa?Sx*%X#lV3wqd3dWt#X>*nU2Y%>W^xYVUCJ
z?e1%|`p6l}3o;p@J9hfj%{V6P{Z;UEw*O7%Q&M8gd&<GzTqqC!6cd)qBHrw!x&pzI
zi??drhQQ&c{BXm8AsA}Eo;Y?iZLWiJz{-G`q=*yL%r+x47F5c+4%;jB&{#ReqzFQ+
zvPqbd2LmooM?qx>J-MOhc{f$xg%aQ94so5DY4|L7)dZ3|k+P#MxAFW*A8B<r(X*e>
z1V`_;tSFY>HZ0KPHLwv(6ja%!MNKzYy$-V^lYqS&jvURI@ctvS*T>>sYAWZZ+LF*S
zpJi7_n<-O2J1ME}`pm1?>c=Q^2Scq5uL4l3j(A3brZIYj0DbF>h1xjBFtw+2Z2nnn
z*)e2h_AcZhjBWYurt#02PvmiTe*z?g@7i~2LdH-=_gUAa725^w?$tjrB^DO~NrU|W
zX~%bwT=pDt;f~y?i)(EVKkvhbKEBLv>I^vWIXNH`sI49zhJQ8F^y)qWH<ZyAvAp49
zzq4XzYseg;Z-ck$Q@ii1`mNIqPgP#c^tV-5L{r8Op)Cg*(aY2Ix|EvP?CNQUH)axv
zagb_N6Zp*NMf_fWjhqDTznUh73a>3oNg-yYS)~mx(y35_+yOmFd>2C-H3pQ9@G4|j
z_f$;*b=<3Zsk%^N*zRO$>pes_;p!Z3mR+Rz!aFORHKDiH0HAAoj!!1f0ag}^?~8*J
zB5qta4LJSh$54N7R^Ghjxsssp@!+L0Nql?p%_2|3sAXz+QLfi$%mum0S^bTR?&yJ6
zy@{kJa}QXrT`A#JTfCX<9x5nOnN97ZHU+1+RlD^=Ukpkc<^%7MPo%2!SbM#jY62Q#
zATyQnt*ioP2w?~$$SDA0K8whh&a&Ivm{(U(U~!qd)D+rDz4Pi$Of9Dpgk}v0Y!s(r
zs&a%pPXx1wsQ9I}HL&SG_?4A}@2&K#BZ2Ur;B13aqb!jEc2TR2GF|QBu@UW7(<}KF
zU|x3aEVZ7mF!_mQgk{(KzQ`L%_~+Fr<~Fsre~EPuydxNg-~9bvm+QBV;KI^VEPldp
z35EAa-Jz`4)~gXa76Wd1$WO(I^yywACx377x%1Yi?JtDL9rl{3qR?>EH~J{4`bQD=
z_4-5M@nap%e-qcwey14jgs*6~pt2^s0?>s_=O$M&thl@f(vuzJk+8e{N!3=r`p83b
zOQKIQmSG?+vuX<uU~>KbyOg+6A))uSTg!wMUFldj4FUy(`7er^xu|ywvo|tc2B&CC
z@jF<B>jJUJ2$8TwPjPfnqgDQb*~ehw;FS7fswpdnSB=+ww|k$t#VdVh@jJcHqT%iR
z_D)qU>maeabDWfNJH5;w`4D7H(#o>Ph@OODo@cPX>eQz01vU*r8)LRN^u%aBB)`oa
zT&nEG;(&$-n4v4P@?Y1lLYw7rzXTjBo7-D6Hd_?%SN_9_Mhgv1TffvO_Cdq3*|WC|
z70{yDo^WuC_HJWc*YwAj^i6=-RE7DwDM8W{5M!23$QP4OVf^AaD4|?2g|3qmmY;#I
z`}lVYl%!^M(R$xv;1qWx;McP8;kOt=5H)s?^mgN9unoUSAhoo^A<#Xbj{59wuMKx>
zX9?@GN^@3dcIh(#4~(X+p1+sBw@(wbMjdvga&Y^mU;Ub@JHxLoyt@6y<X(?lvD0Gr
zp2IP%%#xInCL^bJA(NFX5E}s^b3zh@_g(;hP6sq$$miIzxG~tj_8!>S$QDR5Es5Vc
zeGAydFX|8fo_6wj)XfQoZ_ROG3DNz^g=7PZk}d<cR>vd*!MJ(nSQF(TTnQwH8X#79
zC>UPkJ}_j8b;J9aJ8R`&*;Kss0eR<E^#_r8`mIZOTnRwjk$FNt3+%33`BuB96~<Y<
zgcK8+G(vOqBQ4Lwcf9VnW_{uz1-;qy;+!F#!4J`6-g#YmXMcTynRGJt-$rJWBu?dA
zuS&y`{`sQ2^=&3qd&yX;D&9TP=BT$iEgDZT0IQ;YP?mHEXkI{NDkF9HaVbDg?MZN%
zl&=WD;8<Y|`aCjg3xJz&?_&D}Hkd_09G7WZay3y7)?-u%4K4S*QVR|kf6hcX3jUr7
zG34PPCP;o<Le(%&?P9BcmwsbgYRI}bJe6+3-&XlMLC(^z*Fpw6`_31tfw+WT63gGa
z?rcxH;?;=<{p6B>jti)W-fGA;De2sT7~a~S?>1prW*oHF(su!+tJ@Y3re%<5Ta$sz
zV>z^W#3gVVv}4NP>VvL9HoO?q?2`+#;=VhQ(NyC~Bu3STey5*(8!4TNcNFJf-SJ6r
zaI)+y`}vy2)D_rcll^(`oBu3@W)aiFD)**S5V7i99wbe|CPN9CK_1Re-|Rg6??N!_
zx>4s{%b})LW1xHJ0brTJN!ZE-tP)fNCEJ2LOa<2rQA|<Op_0PR7QniFP(1a>G8*!E
zL<~|Yhs-j77FOqSr|3+3;xw=H;#*In<uByxOJ_<Aj$`D?IE<vmw5P3&kG%ligxe`n
zj4>Ga5^hw&IbRsk{+gwe8qW-5WzQJKIh3|!p@9X&6Q=^UavXe1=4|2qjQ&-9Rl0^E
zeY6v}v8;!{P`D|G5ZjG6kitX2KrRXiQY8&pVz&rf2U8r5BxL;%>{Di>`=?3V@cW9J
zD%9^EsgIhgQq1ZGp-UmHUfaY^mv4(i?1sv!^LlVvZtTlI+YrigG2OM(tuu=dm&9U>
zRInocdFQml57O}lR<-pWZd=d8>t>RW)m*ApITeQgL2cY)2PMYqQ{8WPZB-!8`K(01
zGIwHs%Fn_Z_^~-|-T^WYrVEzFJ}jLe^3G_ypNSF%Ij>a{yKT%w;&3Oh&e8rZ9QFDd
zT&T|x8MgGB8P6E(kXOw0e(%&`@%Tai@j%y4L}iS*P2LEsWuZ6NYcxt%xcyAyjZ>f3
z-($wi)qIIK8GrQ;Tt#Bs)T{c|G{+{w=)gnzv2Q}@GiOo(g^V8r3(;N+l;c%miDx-w
z{FeB7yRrc75aX}#TuNo~`sZoGoT}yTZFHTRVHvs+uM95@r9j?M(A(htueZ9BN%Bdb
zsMRE$t6FmpWGIK{6OY|g60{E=Ui0e+=@M3zbT%&@)V6O})j7I#xL?pd5p>O{a6JYy
zLgC?&+QGzK#k~ZgX^vI>?Hw*(?Xz4hm2wDT=B?Zyuc)9K-K%?CB%_KYz_A42W<kK~
za|eVOoFHSp|FNU2uw$0W0Iv0pKo&$b0^FZPk0luIO3HJ02C>ce2cF9s@*2;-t_W#p
zmg0?eO2`AK_HjM5gpC$%?rLSNB>^mMp_|8k+xI?oTFqBq;#uEd^Sndt10`T`IY5u4
z;=EQ`?;jHOcvah7iI9YJ8Kcv2K<~3Q%eL?d8X>?w<0{ufSJP0q(89xGZI344i6eAb
za(}Av^_v(BL)x5ZK7<{LzL-k7-HOcjW|m^VfMKu&uHUfdg|=o5C*NtQg~i<#XyfBo
z(f7Ku$QShAiV>5t&F7x}gC1$=C`d6c?*3jpzF!`ZvxW_eTy?3>Ma?sytLU$g^05G;
z*maWpN?g+T;CF1r{1qpYzMdeE`~mJ53@B(kEp>nrjsS6;iuEizS!zM9PjIt|w7S?_
z;Pn08S@ha^SV%?5jE9@Q)f0h~X}^lZ_m1KjUt6ZKc|}thtkqRe*Mn$qNPpIl(CS6F
zgr4xcYI>!h^ByH4b{O&i{kF(h`lz^M%p}2ZVDHCN&Y`jzDr?EpWQ~kSRZzw%Hb}wA
zNEBaADv*vh95h>L`*cOl?@i0$k(eV%-<r>Y4AKEAo2jfaq{~!tc*G$}52G2-O-bbZ
zW|cC$0z(qk_Np3Shs9ihhh!&x@=bhI=FI47oKt0^eVS5Ro60c1F~zPzR!DEyLYsJi
znc3XLf-24JwX+Pj8uhQcH1IRHRt}qMqdH*=;xtX}<(6&vPEMxFa=IL`I~V!*b-NE^
z8*MX60!vIG59fT5gY!~#yZh}|&v5kp*H3DTeVxT_yLVVYM%vOzrB^K#DP;0XTGEp3
z0~slQ>crpY?ti}hLWy=-gu2=tdi(Vcds^Re^nCNhLv2r;?XOQl-}!RS6fSXx1D>18
z1RJ@A7^nlq^V}t|)>gek#f%fs1M3QNkG_>_A__fiuPuk5FIob&+74MQQy@uO7`jQK
zY9R2|oESYB(zbgS1Xw&adnV=yHrMHJa|>T?I#g1a?CkA~9XN)4y)5Z$7(d)(rYUnK
zE!}H$%5=!-Azk&oi>`r7;Y{Cp#R+N}B0m}_)f*tn3cEt}DA&~R@ra`F^olF0(+?Id
z^%S;`m0q)ZW~=OdO#G<80uU;QAcaG^8rId)Yo&C2A1k!jdpb;Or}u54b%Vk1nUJeH
zDdhdQt7?^c_KZ3~Z++g~#N-cXyGZ%l9PQ$-jVteFvlMXPiVNfk5a2B$lOWc-!bvbJ
zQB-y#8FD}QQ}20jC{&wtcdSz7%+axl!&-@KS#zD6Btw?ZWN-ujKdZB1Rxx~4R!69T
zQy^iL3MCqP=-o3EwTB1p_aFT(mcK^pk`4_TawFl?TST;awy-f%)plE}V_>$uz}I$R
z`3rF^+z{WympdD{nU@plNl@eo`tqR9T0Bli^G?Z}BJC~SYHYjI=4<TNN0~r2^f@;}
zaYIAMK$ex3cC4_xt>03~oIv9t>vI^2zODTZR=A3K#LGly+Y(cBm+~VnOUK9AMf**B
z%cn^H*y;}^@IeN{2f^OnaIG;S<X^jv;=6P{_Y<{9vGBM6P(dbbm;*m2p|frKeP2Jd
zY1Yi}pL=s2yO-s(?<F2<&@^EUIfZuT%iPAkD9luZ9LqURKbC!<3^%3)e_+Q;3%h4=
z#;LrN6_&a;!^Le^KtbSnvz6*=SDUBW?X+AU>-$F56*)Z#O+k&qy6g=Bq7reYWD7Fo
z>vJ=>mJR<~OPyYZuu7cPQxcrGL(Hfy%d>@9d$VlmsHX>le$p<NOwfL(qT(`?b=A`_
zuG~Y3p4@ajPJRo{_JXEYjtOvEBDCs++<TRyzGA0WFH$3>4|rvmguq`kPIX(HMyL7~
z+IXXtYzTM*4R{P20ft`Kd59;`o&ZYOOpNT)1kj&^tKMmGUXwRj_rq9Ly=gA_n(BiI
z9YYT=iVl_+Z4@u}5^_VcUuL!4smkG#jYiS~ylkD^B|ORfzJUW$4S%9Lz-hJq$5ExS
zqkAy_gV3`c%4Tu5GsjQUrT0&71z=P9P4aiNBKmID248w0L8_NFW41usp5;vtW>sIV
z=N;J}6wZ#u>5B+pQ3RCqiOjc#SD965KJaSFLWf+nbHck&mx+<u!y%cdAv*jVGgULl
z=;%8-wu&2^_C}A2!v%`fiY90>${rryb$6j!+>wWrO(AI~et*F8>Hqt!EMu@PvhRD#
z{!3yT^?i;NDV@c_R%jMUUS7CVr6R=%(@q``$hR2<K?F8;rv!Rrsr-SKxs_6AFtuzx
z>-@inN}D0DwuEPY6IqlZhr;vSSh{E}O}t-2cYho|H2}sn7&5Yym*=^(VV&hT0@+B3
z_86>t@laxe%G*1PnA&#blqc7EqQsgG^4nYrr=-sE#)tE&bC(y1zwSO0$dB?fx97IL
zh+9=n;&#no-WFW5F6~x~m{BFz7=HRbvPSv(Usq}e#QhMoGP0!C8UTOcN?=zL#3Kq^
z5g9BxdW#=`i%!5fJLh~RXJ`J<ZjbM8n0r+CrwvtN3l#jKSVam|YRD3qPr^c`2@Gga
zu)U7!^#h=87j$wGN8j{|oj~2Ho545ir7X-0>oB2Edkvnds&CkZsZ`Dss*-}O=!9{=
zwpK*TDl;Kyb0GI5@z1UI`xV<2v)1ctH;k<YW1ZktEtWQ`c0mmZP38HpN!pqBHf2+d
z@vUB1Zz_)_(&==VYK^HXNL8{T$$mVj>ChJueG7ZHMa{=#=svmcH~V{;5-0KOumu{O
zceCNQQTN1l!+vw7QOEf^Ha|m-D|=uuj=<`QbO1To%5v84AMw=tKIjY`Sbg|!p^vCF
zJ96~m<j~rhAM~sJY^oys%?Dmd^6T_F(*y;rvHwBx*&>;mQ_oFnvk8@lwl0l{c8{sO
zWzNNt8n<5O#NM_oE)P|_|KX2g*1{ixqoP>j-Y+Spq4hG=MMu8T+78Goe<AqXpUDXe
z3thcC{Ik?91zR!tK!W8(I}7OApqX+f5TT?|0Xb)&g@@-WelAZrs~OqX=4I<T9WLAB
zowgT=QY&a#Y16;mb?0h9|6Qt9?A5s7Wa3y+Lkm~HcgIV4OqaC?#=U!afA_cDizr=q
zSA2KKDY$$WyrB%5>I;^!p8t)L@Fw8a3}la3U?{W??$Z=F!bI3RiS;^!=qA5LtuT&w
zgktz8#^w6K33u<N>K}yfZEdtX3(tuN&kDY$h@tkXH)Y%vU2*G7FH(IJ;<T?qO{NO-
zIS|Gj>)_OflK<{3JQbMPWS-9PJ!{mgQa-lx-R~KJ;{9MEiPCBAmC|y>bYI2Ug<hcQ
z7$5lelQ8?9OYr_os#^|2(#ePSuuI{#gw|cWnOe{+?R0l1h+OONZKKDWmiiqtN`qD>
zNE$|&&-Pf!?02V5L4>|jJXkj`&6=Z?;j|>mDpokI^UxW)%2hn6aEpxQTM<(S)=4=>
z$L%K5iC){z-+FT}n%HWRM>4f#Xn(fG=<Mi4a<?;JTow8;zO*S<5t1qp*lgH-1=)bO
zIlU?GqlnUi63Z}m&N%Ur;Smw8C0-pRf(T&iLjY|tOJ*r(V5il*mn_bPZL@i0FLs=)
z7Aq#%&=C=hRS#2#k}V5qEA6C;l^W;*trKg)e(hJ+Q+z$JH;|f(AO?AuM+%o8*l~s{
zGOp^IXC{p^5}1iWNOyzfpyx_b#m!=fHTXaNcisZQanX=1*uN^F*Km`c{CSl?>C`X4
z>NN8g?v$YsSI#mtn-$&xz|MQip)fha?pSwU>ae9#FdOVp2TgelsE|7;h3K*$d76Cn
z*Km)pS7Yw}RqY|h-Kmaf{rn*}Jno*_R}o8FDVCF8gjKp^9)%dW29n<B1G(;a76EX@
z(<E(#xbalaTUu;8tG$Q&h0Db<k@tTVFaDTni(3K}JAKJMH8_f9n~)cIgQ`Y?W6%3B
zy&pK#{DYa+@HlR}Jr8&Yo|qlHt`W3i$4Fy-w>W-Y6E)Sb6^DA`(QO)OyKHK07xEw{
zW8!<|d8X=e(w>*w`}Yi^x1O^K{3PB!gFNA2{TJ(FPd-O3*J|rC?rE5amTy&{=W35y
zO()E6X$~E0D+=v8Rp#NDWNEw*gL&3_{NUCOR+Eg)Z~#Ijv3)~R!~a@o19y?;9YAul
z_bWA<zx)8T81nv+>wDL6Ph{g0{u|YMl+lo3ZK(|*p9{jrUJw0q+jrUHx9kwEb9B-1
z5cqlB-#3zEVCVoVjgLy_S>iX29t44O$>>O9NY!%Gaja5TS@8KU&9YVSC*Q6-CNql|
z-CX2@SChfErTvfIxv-d%@t3FfJlkD2lUh|4p=pwldHCf$T?G)YmXNP(TE(zb7M2m}
zg}^>J`n$c96bY&!MY(KG$n{QBwzF@ri_st%Z0}`OjYSld^|gf3@)us0C)u8*;mXMd
zAZkX&ZTFwx5ir9aj1gO=`KLFjD;@%bnYrc|?p=5cSDuMh^7BALa>0*-GIm;ZngkAV
zQixR>A;mNpI~^%YESxSAQ!Y>?Axw-;su^%S9Ut=b7Wds%qs{W{EZo1ZF8#>>$rn}U
z^+T*~?-hx0S(@a-l=EVShf!svBzfnq-ynDX2f2wGf4@-5>UYK}bfOnq)K2*S>N=H(
z3$~s7s{gJy?GQ>w^Y0wXS^F1P_cz7q-QcUdf$=mPk%f45{(FpHr8Rg8TyHkmx^fjS
zYD)h653`|icz<3B%$ti`pNL~#{&CfyE`P$lb8GtHObQhwd89}tH37mS{6}7)_f|R|
z!@sks(XU{Q9k)6`mRjCbHT^1?d*j?YB3^pelbY|?)dsI7IoT}E`;3{r&R9013AF51
zOi1+xVBiO}#2QQS%IuW{D#t{Y!*cdJV@ggVX<D5LEfp*Z)fO+(62v^h$<7ze^3<sD
zFJ?m|VH3=4Exc>Cuj!UGLoU`EE7vmhK6<B^=WwI`T*Sm=Tt4}=og~*5gS*S&xtG)p
zcLS{j_c;IOv<G6&Ez#<;a{D1CSI+E3NxoPSPFsO-(i~jAfVl?vYyfK;%0of0WGOoM
z71-bgl}~B^2~xw*TX;|bnJ>Al{E#{x&b#ueiQ$!4d$Dqf)g^4eoqI$kH`D<A1m0S8
z3W0OjEd9$H-+shO^^4IT%sRbQ&f3?jEp>R}D@4J;P|T;m37k?axxdvTe%CogVVB;A
zG<+f~bl%(p50oFXP4HZT3I@I7@)}Mv-Z$d!&Q7yQ>WFZ^f1%gFiJ3*?0|`$~>&w2r
z^#j&1^74iAao_fM%3&<o!ZaWD+1K7_x?$f;q2Q^dxymmk&xVx??^BQZv52ub0?)r#
z?$WUhilol-YW#oIeo$Yiu&JuC^B3R)Hh7p;Ap!z!pLr3~bfWAHUxT}Rdyh9h#l#!N
zjWlk{(rCF=JV|MGB7g6Bdcuq2(tFcNS#Mk`vL|%f3_L)i48&O}b|iyoXtQCJuK;?3
zLtbHTyZ?}0^zuyLjXtcrS5N-wb*F({WQeNRO<79ksMc2{Im2G;t{sLudIEN5{NjJp
z%wdM<Yae^&ebIp_><f33-UN+nWt&!!WRMC%4Fa%?gHRm#32X-g+9W6T9tiR<jOx{B
zS*IHs=T$S$aUN-pDckE>(i1eiykdBXUg^EjC?SCw!`=pe{o*V^-C<qcMzN>&p5=qy
zPJ?F7W@qH7L?M6#a%4WZX7%KP?T;`^89UtSVZ>@npX$k7ink-iMxw^O4u2KKO1d@a
ze)$N)J0bT{w1dl&nxzq4z20$u>bXUJ1-NCD+G{PF*M;Z)I<|v;wUCzGfIgif<FcIh
z%Kyl}+rt-lkS63hg>2BYjw|@Nf1zxR4g(MI!11b<$@r1xOuWw3FF#7J?_88)aS|?W
zt9DL*#QbU|y5vtKEAH!I&*q;ju$Mb?yk)wPun14cri`!dl|3O$T0Wrpn{Pq<mCw$x
z!QYHT86e~~tL9e-Qzdl9z$n=G!+UegyX*@-`XZo$5kO>*wY|L+0m=MAk}oW*p&WY_
zVQ3?n*1+)R`{1iFAPlT1eaIsKz};V<w84U<nRf&NZ@UM@X?$<eXN56RmpOIz11%+t
ztD2$o+aDPDmB23VMi4;-Hcuwt@^(ZP<dVU}fdyeYr^g6$_;%~aYvadW#JHl+@QGg9
zzs4&agPtQ33E$C>G0);QT6FTKb^D8WL$vpNl$$6|zV(aMvOzIE*z11&tNB4mh@|*v
zA9bZFXW~E3&gLs+R#D4;0}e$`V|_Fbs_^Ux{GTChA-7kA;hOprgk2$X6Y1UjZk-iz
z{Zi8z6F4ni5q*cP55Bv01d@R|cTzej@%P6MKN{+7h?$@?>Dy=O<y*Oi<BGNTgyG#B
zuy}Ph0-KZg_h*rEnT_G@s&_b@fZ*V!S@jZXHh(Gn-FMrPig`@<<wuPwg>tsHSdz@-
zF1`0--`i^ghFZQ`$j9uzj+p*SAJTGjSbG9R$7weLQ$!)<kXGq`a<0w3AOa*M0wp5{
zWY9<1hL*--Ug*v0f2;gtbIk6%-?Q@f@4a3fK9I7Us~tUm;n2rO(ZcJPWdzenh-_-v
zO7zHW8pX0f74Wf89S^iX4M-^Z2l(!Pz+YdO%{s4+#@0{YghX7<j<{DgzLWkqLEp`s
zbED(^(U-3n3(?r@mHy(MY>+*u=;<pk1;IpBB*WY?qAA-bq2<YWR+Lxhv^Gy*YF$RO
z49+{MO|m{qE3C?<ss(#`CH35?U{(bfDN>W$`U?sm!WylL;RY+zka<NS;)-Qe&OZ8^
z%9Xd}ikTd?@fIA&!9~K^$=U*#nlEm6nq+`r3WEOMA((P{ovD1g9Kb`bQL^)XN%Be6
zM=VBYWldg1h5@y#1-TVHNqLIF519aAOD;+0uHr2hGV)Gk*7kUH`9{S(2<kDhbar<1
z{i5%&I)9%C3DZs2a`T;l>`gBt@!3u5#4Vu2eiJi5G8$F=zzD3N@*qZ1^8INRl~Q7U
z^0p7_ac9B!)lJmZ6IADBiFF>>QGKXUN2&~G<nb9dl}s7XZbh{Jk#PCN-w!?TzBN65
z3DxA^)H+Jw#;iuZCKb8GpK7D`a(8b?zL0Ht&~T=zj?f<3HQ4!)r~nohxZTp0DD5Tj
z`0<e@ED{+xbv}h9zNeG&=?N=)ND7|4xyk~obY3x65iL;y(s+YrAcNWQTp-{16sve<
zazlqzx4T!}%0wV~+!QmRh6x6jL3R*ZAT?x#f_hBZPD-ov2E=_8W*0uq`#F0u|7`f}
z5-ffCd18|5%BAqjO2OZ-1pk<*jELktM}~HrqL#N0cgLF6dP}L~wsPGCxu${pmMXsW
zjlP&c>U{5vJ_;SloQQw=wWDfRqe8tiwxK0!AmI#}4Hk(&5&?$*WCBjPfvq};Ryt)V
z<<6sf!jH6Fx!dIR%}ST2w!I?FDB}HFOCK6na@mVN#kpTE;ZLisOF$aMFb@?4&*iGQ
zrO&^dxU<!7wr;sgKG9n(#IPh;#?RbAN~9o=4q)|sWV1>&Y*tQv)}XwxzcvSqAdt-2
zisCZM2RffcegWgJ{8lz@{VsE@5M|na=%%cB^;ox8yABA21`A&@iXY8GsRCTKEFj{;
z03P20$|Fu+mIZQ*&0tVqQeZ1fQd%|dVpI51M`W_JvCs=5W&z4}=<Y9?Zi=_L!EMb&
z@1v6|B3+hah70w>L`obI#C9n!;!ZQ5|Hh-ZWU0Gn-+4p%^7FfowN#T}Mk_!ca@P38
zQ31B&@GQ9?5g+iTY(u0lalHFznOms$7!s|}V(;lITsgzdp7!N~j77ne_7uwjs}zff
z7D_4G=K!ps5q7bmg(Ulk5Ukk7oxCICjSm@RFyT`@_vcl`rsGHLUMu$rhS3WoWo-Fa
z=WnEgH-x?Fm+M}O^Miv&F|vDZT4B6R(eixRCGMn+GIk1FGdEk!Nn~yni<^rO9X`}6
zaMVC`3K8!*v#Z7{&|O>9X3AR<^6Ff!sx8wFre5s)T<}#?DOMc|67MhkyP+x{%}`q@
z8qw6%OXyJzIN7JZXX#+|xzDX>+ff=wESW(4CX~&+@NY+;m~lu{M!n~?t82#ZL*(ZZ
zsHIX__e8bIm+tRscu(Qh9#fLSeKYmV^#pp*@P~4hUBy!Oxr%o_=?|}Esk)lj_xZ<k
z6lwk7c{uHpNv0T*DTKJRE62XJ5bi<Hd^D@8tji$%p}k&evU>?BFXOhKDpki7#n%$<
zG3M28(XqMnc{?V1kDVQ$J?FOoIBRST6$VQK+KqeugqR>E*fwR!IJ->6T67Lb&1Zpr
z;1Naz6Pk0t$cyd{m|X>4t(XvjZ9L$gb7U-G#!`kQm2AT!tzAu2b#$Q0i$sTG1iAU1
zv+#=XPmUK*yEjf`Jdu)`?J40|5giA*R2pgqZT9L9#Zh7$p5<(PNxt^&BO%W6bIsJ@
z<-@{k^_&udeHqqLFUMV}x7Ts<L4!e*kJ!;{>WVE0a}^@fy_t}N+Sw`vUmR3ef>SF2
zn>9Y!M$*GZ0rJfL8|EHrg~5Tbq1$5GQuo^Qr`XX{r=p0;fG~EbX$Yd-h|M*Dv;91A
zYQ5>1)4w&fY_AYoj>RTluy0a-5kg?}+{Cztbrz7Y?EHuH`!~y^GzAn#yQ(SLAlUm$
z?e(4hdsu_`hkK;48oN&q8s<-(J>x7)=wj64Edi^iN*0umB7i&sUoL=Mvz4Y&gs_yx
zBM{w|MckHaMUuhIH$ojL*J_8N7QJ8M!E4`Z>}{tGuxS2=vYVCH_bQ}a=lH#SQ-uwn
z1}K}pTHAN#sE-kC+bL-fQkP0HWosP4URNn+ALO0jm6#db+@@x`AT?X3Pj7*is||V4
z*xEFZ-1BepWX29m`d<>#yd`gHFV9i?n~?XVdKGC$GA$u^fO101lfW^C0IzV?(|{<X
zAlys_WaY<#_Xv<pokUAw?y!b$u%`w**|gd)8+#qqC&4v&-y_-h8G&(P%TU7!)2B|~
zm#&m{iTY-TvNchfWx#{G#2p0V26nWN2bQ-};GYEDJP!{;u<SgbMIlz_MiUt+M2>7#
zV#}Kt#S#=!E4FP*7yxusBp7i@i{7fl1g!^&DHo0D*A;%?ZNMrO@@21+Aa{?$A>9wC
zg*S7w?1~N(+wlh{c&%@+3Plao$N<F$I7%r)Dqwa`_g2eMrG<^QfPs-oiR~kWU6kK1
zYgX@4%Ubx%thfD8tz~4pT|9M(Q;D^-L>A+%RvY_=k_Zhb^*!$H6fcum?d=KgojM#o
z2if24TrLYykNc@hKe?OKS_~{SOt;%xd{{V-XB*nc;Li16+4w(x(i+RY2e0%BvIsB8
zgPcnWL6qN{jBW3MO^4o0jBRSLfkfK}Ht14^>0ts)`JPRG0l%!tz_3XQt8|p*heHYg
zl|T`tTDuwyh+x=SjdxVcSIvu9ulqKIL6bD%BP|e5nj2;arWx@$I`BRbG*c1BQ2leR
z`D_VMI$pbVY8|csH%*-kQMetI7}^k4Gafus`~14+vtENoy$2n%0$GO6O6OW<NpO;G
zdf)}6>*w}uD8(7arIY`eoy*+rE@U$&_VVn1yGh4C0XzMYxWX5uKC$Fd-;V6B)~vgm
z4ZtMSPuW_FI|-E3Qi#$`$q;o4sE6W8PQBdQ(B6s|=+_J{?tgL(9ALC2kf1`qcmw|k
z8$3*56oM3XiUBioTe~<Du<iw&qe;f%FB1NZm>JS;@Wf~91+8k|;ZDJ}wTI8%pZh9~
z%8QTKEVSA7IrldakkfD1eBaymPe0|8;fFO-FS6D(i>m9P&Ok62!y&V9bfCkjl4S}e
zVH%`GoIIV|Ypr{$><9#z={U(%xOL$a#Pnz6zh4I&rw$5W9~)0Wbl*M)wfE>&0q4%a
z#v|5t8(X}p{Z$Z`Ycw~5Su-e=mCmb!(Mi;EuQdDF2|N<Sl(`iP8HO(`aTT`<!sMG%
z)de%AFIXK=&fIj_y{n6hMX{0dW&t`U9$Tz%c`pw%<dzY@jS#$zS2n!}IQk&XJJ%Y#
zg3L38N}eQ;!&S&E1Vy$Ld(l^jJASxf5B;cS&J%AN9LV6c<iu#x8xw^2W!eEG)9H7#
zm<`R{KD4j)Hx(A3?X6#>Ba04UaOSG|`2RYtTpqJ9*#uk+m4O$K1k+|_hd^~h!Y^$k
z^yDs&@>zC~KCaoChGai*HPiyY$Xd+c5#V<`=kWGa>-e;H5;vqzmlI<##fdszc2)0N
zGvVY#<Uc<ioaVUDva3I^Flw8qowf{Z^@DrvP8gV7A$;)fnbM5;d;6TI>#3hzuHtDr
zzF}A8Tx;#EPwlUjE#^K9XW0Q6owPorIXFT9kX}~Il8=4QGA-$~_e)jXnBwtpoogE-
zWsJ+0WcuwNh7BILhF`TSqg^o!uT|C7Zs13If=oCamypCKC$Cf8SXqUSUOrXhZlqKQ
zm=4DFTs3l)lvBM@ar&32Z!Pt@P211%IyJ4rCvJ?yi%_b$rEKkgFx`4QoI?#-2|&`M
zO|xdJa<F8`P1Z5syfoS2;MRKVf@N6O<j0#1^Op<4Y#-T5+9<JJe(1MMYYX%%PUMUc
z{By5-l3tL9@!}A6L0X?FFr`&~t@#o{f0epM<L1g*`i0<D#0u3{aVbl@W^<bR(SR~J
zZINJm<nW2@=FaeZ=s%dGFAu#nrM#rq(~36aW_fG<sXqYIAb0I{i;yadr<zn0$ms^s
zMaEHjg`eiH=9ib~LP<kmRZ)+C#K2ms6y!<+!7~FW4P;<|*(hb29pC!(NQEaozVBy9
z%S5_EN<nH!1-5EThR|yfR}dh8A)=iGbKv9wKd*vmm{~{W>n+cBWzsHZXke4e)HpRj
zchLr>uuu|Kp0h8v|JT{-g<rEhqa|952^OJRyl$w<J?BOj26{9kc^pa)Ji3sGcH4rG
z6T2$jk6B`4QUIVu8Y~Ave-DsUjz_E!bRM0pdyRY3CXWv4&2-iA#H7J<?^OUDKgf^1
zb^#oBIi^3Qb>E#Fw}9R%J7uG6$$Rieueu0~pvqq-5MLdRy%M~h(z7@-U;DtJeb*5r
zcF>fWPLSnA^QFO*v$$M7Q=LRKQFO+F0EKE_$ad0UN)fl=r{bHAuRaNyDMO|_Qd(`x
z|9_aFnBX~s(A}&yM=<taPcan$!UWl>I%Vg}X`pK)TiM{+;n~yB98r^~bQsiUoQ*Q(
zPUv(WeYuX_Y{tuZYQ|wMP<@r``<rV&i9~h733>3Bp^zRbL<o{<{@-B+W>X~VY3R8o
zFi8Ke)Vx8dbvui4ua#tPmRdpqI_MHBmz6J>$$Qv@=Qy;7JKS~4Z!4G*>2$f%3|uD-
zUo^o7zF-3#@B!?N-Dm1=Tzt@HvHR^yAHHV2iH|dOTtN|?hZ)4vcUs2A?)=_8jJ95l
zxY1R5N4vnK%rIB`ds9J;dZr2!9^7AE)A_H_$dq8(a1qHZt_6U?>82JG9VNBzk!ykF
z+>g48@+MceFH9Dx><{a6+*VX}=TA&BLeJdK$1H2_<cO%*pG*#lSe`LJ##a8M#sB&-
zEikOA7E8xCRrugAekFXYRF55kq4XnQ#cY;+vh67Md#mw-tM^jUXNp3Fu!J%!RPP+D
zt=B-kmJscWQ*MF}N`uUu;I*JjfP+@~yn@wx@S%Nqk)ch1S&WOMeoir$zP*cAksw0b
z;(0<}KpU_yF2CL{x;)Yrw4;otn)l>SOyTFx-mJ;$uqqxN1Rae%cKX3=Z)auS|8?^o
z#bbFxtTlWvZKkcfD*d)tw|7xH`ms3Pa@>>rT1u)$@~L(-cw?P9Y)k2P|JS7D>`^R|
zv)5JZ{r6TDgD5hTd%e|OV`p{Al>5)wM+s@>YNbAzh0C;-F*eI(5`YZApaLlXgG&HW
z0t+Y<WWrL)>`cc<=>6GhmP@yOdUw6nl6^eMz=U4H!j%WLXcw~VHn&upa}nxx<#=gj
zyB$F^_LZ|R0fof=dn!TTUNgyn=>(1^)>H%kPv(5Jy3vFs3)gA&W9C$%h%JMs49mnj
zFlEw~4{(lYmRTHw_qD`ol`G<O8AA|9cvY)v`7jo$izq50D_c@3`LX^%N69u@rnd2=
zLA6QiiX6$%|EjIU5-0nUB!PgNruDW{rmND9N9>DzumcbNxU%5L2ffN;&T5l+V`Y2N
z@Eh4$-rS%a4humr;gGxFkw6|i+-SVPkE<-W_1R=2Ix@^jb0YM^o%h3)g2jk8F{%LV
zj^Dg2dV+?mjIh~@P1>LpHj`j&<eADg3P9Zjem7?GDj8RVtsr<nNQBGXM;ewW{ivM1
zWuyaJ{L_8n^7QW^@@ASqpp!@9LGNASTjw<G9qjC7oc1~mfv<WHY4h=tmB;LXfO8Ep
zO~MXDZ>n&TFukMyOoxo5?Hi5-Nz$u6uIykDu{_kT`UGf8TxRknd=rLi(xKZojx)fK
z7Cc$wX0t78v`Hv;@so1F{nLLp$DB}E>8^T_(1YaiP{uMSC+vXdY{wi)Po|EjQ3zcl
z$Syo{{cN@5cL{>1Z3@XV{AcAO?gP2v>L2=Yuq&2^pZOslC6ZAFuaaNS@;`RJyEUus
zIA1xjSrN7~I^<ydFRQwV8g(aO!2yK7ubC(|Qokn#r70I3;7u`U5}`b&s)j{S>C)qj
ziLh_-2X5FpI@NdDu{fSmlm|~^=h1xBRWn$1tK$Wj*H+VZ;u)K4SgpGu0-#r^h`cEX
z{RpxMTMju2a5{kh(x_t)yjTq2h=WqB1Ck~2@b&aC!c3CK+3!W1a<M-G@!ZlY&bcX?
zQ>My#ePgkRSi-!Z`PZ8NEq?u~WL+Z9@f@O5q=BKO(dUPm$8ea*b00(W8DBJ{wKSRP
zPrTPBFDWSp6|!d<=EiRNN?vA(t42#D2kOkx7u1{DqHg&76{yJ!z-)i55}o@nmOR?G
zOlIfD%`dX{SB&WY3erVI7<v4f-s2Z-j#8<`v#!|1rFHxdQg!I=M(g?q^WeDA(T0{1
z_f6S31PiJRoOJ&c8o)>)?K7+@$|snWvGxfrZxY`2>WBGxngj)z_)Ght?2Qn<!f911
zRXfO+FK_DPk+@r<(rKd&N2gFZ+EB1i_GUtro*(|loX1`9*}iep&bdgove$q&ny|jL
zzN3WR)pEt5xswI+ZNr!wR=tnm<jSMn!PvC{Y?*{rO4wncbTY-Btk&y20*C6bm@HXr
zujBvi*<@EYYph5Dk^BFexU)x0SEDwYIN4Q_WYE4CAHVpt<^z4Irv3h^_gevD;>f~s
zl(H2wFJh7b%?=_#K)2c5oZd3D(NR*APFwhD{fGR*Bq{2F0L$}Wg>BXpZW0VR1U5Y&
z?Nh^-=%GOpnwm0|cR|Xhn0cU8n>P0Ii-DuvL~AOU(|uDW!6EA;ro+^NIBy@7rFC*f
z5Q|nT^JW!>E74|ie>?PQQWE@^JEEb+9HMoWAUlNsa<)x5KAQbsVG*$dOdvuvVqH44
zC*!$aUy5nGy?5qgx^q)s+ia7&XPn_-(QG}>YxA~f(YTki$DA_>0wv0WJO8{>RWN?6
znQU}#Lm@=?`PKg*d$s5Pcon=bD@=-q%LA?OuIc|la<hs=<x7U|kH1Ajerex$lI136
zdIe)lRiu~Zh?OlJa;tu(dSZ5!_I)wg5i#qSZ}vJmsOCs(cy?Vb?c&ypX$o&|%>LOB
z&dq5OX%|s!-eS0LdtcZR_T#K@>gDGb3z=83Hb-^><(5kb^LeUME6zvptZAV81m#-U
z!<)MchQh#8w2BV{8TUQsWUC3WmEdatQe)IPF>tV#KIx+zM2*Fl>!I3W4K03w{clSy
zrmrLyu9SMkI5niH$E=AkF5ZG%`CZ4{6a2Z0#l4t&fW79jn?;nI1@}uWorb~AVvnz_
zWeZ9B<V|}y{k9pXv9$31mdL7(_cePxXwI)qNB{IO<{gg9h0m2F+P5`3F?7-ay)P+I
zU2b>C+mwl+@Rs=VK_xxYfVg+b^<)a2qeIY6nIUZ1dIf(s#F0HbRHofppy6c&WDex<
z>$l>?G4};|nbWD9E*Y>mB|hRs!4VKY6tss5MB?C*Sa#=lTjy+$y)dj;>=SBtIIOt6
zUIeh5z{Qbe&}7CbDFRQ}X}p0dFBZX2yn678mc8F}#X~VeW=4dHQRm=vzz?As;dIc~
zj+E#tN8j8`CO(L@^9Qo0&ghA$y&koRi1XpPQ*-@Em={vL9-gt|Z0D8<B6o6<c?YZ`
zEjQ0i`DplEVjL0U03XPfBopJjt2P^du&8Xr4meVHQM#fn{}qxvn0;jBg4)@9qw4>b
zS0h)J&djgAYQ3I35aJn0u6Q&?b%@3@;&hq-aRHm1P^CO%Vqo!ZZrdk~wMbpWin#x6
z^l7#1>GV*20uj^>6+!3nJas|3h7HL6g>1g~`7u06CLx2^DKZjhY{&x1nLTvb5(6??
z|IYV%rH;?vCOoE2%m@xRD;e{MA==hM+VDZmwt7d~OOuj^FQwvwhSJ$F92}6&%kn&d
zwL2L|QFxHe+R@&8Xs|9>kMmkSn8rCTPj9uKEA~ynY=|n^UiX5UJh5J5H1(#!>FMw{
zQK!sFOs`51KfM-gxSTxX&x6+ajg(P%l0iSeC`(g@D!@)?Pbx&o8B4lrVr>34l3ODv
z>+jxawP3TlFmr99@=PUn1X>DNP29glxvOF0qdUb&N!_gC8Z;tjc|A4l!{S7QZ*^nS
z-Pd>fHdwbX@qOH$2JUdL*X~a#;~4g6)x`q78!37Cx+>~DD#o)#gi6#y+vA^_iq(gn
zzYgD3Cu|Ata0eU+g|aC?a1{)vujAm$YXr2?5ilBpgMk~E3*HribNR0kd(@h5c|q!D
zT${R{Hy-43MC_ig-D!0G;~s3<g02~&#;EZ3ae@Dw;O(8VJPCJ2el5f}Pvs21YkO{;
z*tbZl!FZ-gkCuU=_O~n^LN|nYum9araP+OYbMWPRN9=*2)LB#DA>g==4kbm6IO(!%
zxB!IJ^E&-m2eXvBtTEfgyN1no`x)<FyfEFEYK$59w$k-B^lQq7ne>bA?+9@|2HfA&
zLh7tl@Kn{55jw?6<?-ti2mx~Y?(4@jNs?UG^uJe^JiU};(<!_e9UCEV(o_qjx`DLY
z1U#m5fn{+s7wh!K{^G8dWMa2Iz4-D04+|ffmOuem^@$32X-T1<2e_SpjhFyfQl<8g
z4=z+}2<aWFiqrgpvrSekoO=LY)_KV82o_mdgS3A@Ot<^Kaz%K+ro5ropf?KA@A)xO
z`bZ9G;$2cCd#SxGomw^bJ(Z#{LA(uFsWW%jREHZV1rb2|Vv{3f$jVves#&)b*_8Bp
zZGKE&zHHg%Xc0UDr2iW)CS-vUqC{^N)SwJFl_0Fm^XiPm*iZ*<eY;iT4q(bpR#ev#
z7FDTx-G@O4jWrxWW<WEi)Nm?SOspDz^VXayi`Xbx%*hN*U<w{P&u#9v@JR5n1U3p_
z$+ZA>f7X3evcZHScEFD~ty*3#Pwp3K8MuJMXgdU+!xq@eUAV&1;_`Y5UmMz$&9m!z
z69Or7urm0WI|#qHe9ogs0akjMVsw&e(a?%O^TD7cob3tZc~fX*sm-Pm?6o#W7&c1S
zX`-yXr@;>r)_-oWEMNY+khAV%w2&qdc_K{dwSYLE555suJePE+gIik&Nxsq#gLj>g
zzp76}$sX55B(q}R<paua`D1z%+!zvKN60jI_==cM>~1agb)AYzK`HW8Brn6XEa6Js
zvD<%iW$v_3IWUHO<R4Sqy{9D3NKR>EVo2wB$SD%_k(!xk(z`wwIxlF^?&%*8N++?3
zBtvTD+k0`Nhv%-|k2rswzF|Q?VL3+Ek~?$sPs~V_4ybITniNUGlYCWIS*XV>XfX0C
zz_R%ryPA&Ed7qXZ0Cqpq%9~k65sz>xdonI^pz2^$S#pM6hThirLNUB@p8az+lBBqe
z=uspTmHMvHM`E2*_FYkPWwrO8{G8|!=>)%jKvoz?8PYWBubd@y#WgX#@4236bk?kS
zl}w5mjW3Ypb#jkS#7%5D)FI#YNl2?Nq}I*o<$s>6D*RQ2pg99HPUyUv<g2*RTy<v9
zA@B5eZ0@YpoMv-RT$&-8U$}ETJw(SQxOR+Dx$`=w^vM+>upZBpC3^2Vd%Km^9L>17
zX3`m7@j=-w;Dt{rA;wuww{Lf^#?E~<zzvXhN%~0ZYtHYi9)G>kitn1{($&X|cF%;%
zH(B6Gd!spzc2)Z-HMDcF4|3Uh%D_@l4k`cd7=uGCd{#71@2;rzWW{gN?H)YH4lQoC
zBxry*Mhi7FmY=86lm<ao<3IXl(Nhk?JAO^8Bp6W#%@MmmW3kl&B;JLqI?8edYQczG
zh6kI{uV1y<z0qsuT&y?E7Ry`E^@!)5UDN(?yB(8qka$%qHOM?xoDi;GGkwT+E$EO}
zULX8tuFaiV@I0_lj;8clgZdK{=I3}xhP`Jr$kGX$XBw9m<%mrHb4NSe<_pkpfL?f0
z`tpDKFNhV>jwofX+wqd5&`{W^>8z=+xfd)>?N;Z`=m07VtS#uaEL)|fcxH0&*GumW
zo%Y0}jOcxDfk-2wISrm>N9m@bl`IJ)kbd=FrEeZ*Q6?rZRh}miqBK3bhYsOdp|?PB
zqji)uvTIIjDRT6HX*trg{lViM-BLVey#70tS?zE$4!SJWnlWUXq>p)4I(Jjb&0X+K
z!$n3Ulir2jy!t9btfvsMn+2##_^;&}#7=|vNvKC#!}F`kKJEI4<+2{GcQADo6I}m)
zczPFjru+AQd^5-8Y=p>pwmIAmBZuTT&1}pz5-R6xBxj|%vmA2HIYt_0%&CY<r@M$y
z#Ec}mODGg~rQFrseg9wU^ZorFkK5xe4|PAd-`D%PuIKeU6e<}IE2Jltn!K{v5igW-
z2V>B!*E73)c`8hmiGSEaqXan}tly9>WkLr{i%i$V!3-{L`(RR!SO3xE&tDu`#joGz
zRSpEQunpG$N}<*mbuY;NHcS3w*%@?s;`iyNbHBlPqhJ@};`*MPZ`f)*rM58^;x$G<
z`49xqU1{(~cSHHWafpMAqED)JJkYCL=@1<8VY8F#p_JF#wp`cUM^Zz^qS6~Jzipqt
z8BiQg1YA?m=AcWJuTRzWKx^rb@xg*QCy6{Ub18xh#D-6nyx2gSC^`$T3^3h4m;dMM
zfXBQDDfeaFwqv#(J~BIv%k}G*hGG2PtjELViot2U#Hz)d@T>Gd)BhmFO}#zV{FOcS
z-eiV>iE-3L%2N6s8*U6BPYsz(|E};!&{d>cHK5xBg$=SgKc4afURjegl2jWJm+E54
zYF!V&mFX)K0MfcOy|mCDf5J&~rp@NL-A#wInFy5~VWQM)*PRv$JylmP>KRQ;N0ez9
z6<k;*!3VA{MR+Z@mi}gPy+d(0BV{kOxIE^{?+xvxkuP*>&ukK!Fv)r!5sMEe{r{pz
zyPBiHg3gM=N(K<bRY6P<MVLHhaA6aESBo)T-(ew0W(e1<s}&&_cdrTy!Bs65+B^rG
zhoF62u+mRU7cxoElD*$*UmZm(-zrWL>Bw8^Uvr(M<;T7Kn=V;$)OML<%av$G_?uE#
z^wiTmg1+Fm#j10!2i$(I$eJ8?h)z4DpzG{43}bcsSaA|aW++Lgv}ZUuC!@NyYgXLH
zylwj8HLM1-nVCnF$O{m%iXdQIb8lf)$@jG!<kgp$x6D-y&o%0TAc`<DOva2=FDa}r
zuoMduo`pv?C2Sv8L<u-ZgL~T~OH_@!QZ;0k-I_4uyx7Bi!WddYmfo=Sr+|<b7UTru
z`ILcN^bK27{6XKtt5ey(f613-=uVHOp-#jvUG6wwLKxJRz$!TnV9DB?F&KE|fGc=4
zKs(NrIVrX>UH74-vzA*9j;#8Xm=vtkwgj@R++y*1JGObdeVvEIbnYrZQ;Yv$Pi5V%
z3utqTyrSzYf0&!?8*spMM$Z_XIjw-ivMzfcyD_On&D%Y6l+cUn^wSYgpn&5D1q)gM
zoeM546finR$q%ji)b-qV^>GZ~AVU|VgZ9icC<BpI-J3x|If7VWAp}5ZBji!sZE7y<
z#|)q9*6l-qhnftG0{_1Yy6l-D7qA|9E5?^O!3_6O`kBknd+(O-l|x3PSfQu7I&#7k
zfDIMAT1JrqNC))E&Tk^$C$ghPAB4f}X-Xqsqt%e}P(2YAy0$v>iN{h|U<xqiUbC+w
zQD?_u4}!=7R9utcGB-WjU22!TMz9|y@)YyW_||vG*cM@%eLoYXn7wErm_P*8(GT5e
zv;IAECWknFoqFyq&&okWny<xx_&-Zzk`L{B`&EN1<}=cq=cGqPL%_CxkR?I(P5{m8
z$tB+^MtFe9(7w%?>bO*ogUMJUWzng{>(UFjLxZ7nb63<(!@E?Abh7kkj+K|7LKh?u
z_3>o?_Y|OI7`APTx5oI7>|!1Bx~~J7QnMtPD_Fp@N_&VndRQxkYdIL?&(hhEK`hg4
zKld|#<AT}`?xnbzlJm{F^&41ynOnw6rRZ1x+_W-|s<a+`^_fhp0>(~~&l_W9jLIV)
zpKd?(6@*p!l^UvPM^^2UQ2WQo%^9WrNmfSDblKQ$n@wzZ^O@_HGnNHhB!F0SC$?NV
z0DON={dwa19PePZd~3(hm5XgY&+P3KLKsOrH9Oim<$khwF>qlAmtL$AQC!J4R*Rv;
zijt05&i%rhfO9=8n11ro<n&^=nzJTKfi(2&DdPa{#J5G3$es9}ACq$0ejd4RiW>_Q
z0-c|$4IkbgMxA?+d)Xpfj?h+m%uP2}gbQg?*wYl^K;ktEZWay_o2|qul@m~d_=z4`
zK|ma~cn6-5Y1L}Y)xDDq(rwdNQ5>hEyL;;5zBCNT56fU{0YkG2WEx4fjFB@r!mEQ6
zNMa>UhUEMz%-SiCcR2CCo(Otm6g!lv=?ugo#9s8s?y38W4OVv!je86|L1PMll+xU<
zXEk$=jbrlPRu;4T@(%3S`5)PB*~6W#Gp^<T6ks&?s04K>wuJVWo(9-Yu&`fU7qDuU
zclZ301$gac%1RE{b4%f%7RM%h75Y%Q2P$p3cSi5XqD=^VRQhe(#k%6=m&F@afod<R
z*GPS|)2-;N5i1n`t9;j{Pgkp(M+wWE7CKSG@H|_&XuTEU<i9EqH1t*LweX<eeG_ap
zMK2JJt5Y6$Q4o|NjU6Dm(7eYeO6FH2zPr*ijX;8tXZ&IlK~=E>^vE8sUOSq_W3$Uu
zXQE>*KLPt?Vz2I${<Y)tzFAvrIXzBC0r2?2n68L)Y_UUA-!{p*1Xd2IQ@_C>fs)(f
z{#yY8;8291<dx+6#Z0XjpQ5_Dl`4)g52n4p?W-fR!6<$?<jeW}SbwvRVCFB?1~(lX
zkWSc891#-;D_e=uu8fg&GCgjig5$80NjyYiV8js@CEz>xt=d#W6Q=JM!*43ujE6(I
zADqO{V9x-9*IA6^wA!u?h)d6!lU1PsL@2+Zkzs{9X42ixwfAH6gFL(9241iLnuood
z6_MyEeJWv|$J6tGz(f}>jpDCqP|V)ub!phFkGZPZ1W|Tl+`$%6%U9RncI+SK42?J~
z7JnpzM`8uo_yiPI@}P%tt%(LvnbhCa8$tRxpwPs48N66O@cqhE2G!(&a-)5C1om^S
zpit2bACrQ;2Ejo_n<1YEEu4q&9SX)5PP)IiT>gFPAgloRgw+s=`s)2Nh?1RQbFNnw
ztb(Q+ZKXVD_;RvNBIpWz>=2wzmMUAq@TGFvzUkUA`bJQGZ$QDXqQgY+)b=KaF#Kf!
zwjCzL1tzabP71c|A6$s;?rGCHh@IC-Ufdc5Oz*+M7AJyj)A8FxoQLh0+2sXlfMn#w
zK|y2h{-nI{u%bj?K^;fLe3h!bgI^P&Pml>p{|lqmU|PS}?w$}&7UQfE7NSDoM%ENK
z*uS=@(CkL+Ls#9IV~fR^c?$TU=mPVVtXmy@NwT5N@pWHxpKek|?dRh!>UCg8HVOBU
zABo7S%rPWAy_y3kCIm1QI!UoCiK0kQYd%ir^2AL6#3z(qJkxFSP*bz%(0;;j199=T
z{yt7p1Pfb)KYeyC@2rLXai5GYEdwFLogPM(&oXWiPdJ<zbYXbzZqAE6$(G7Ec41!N
zw!%zrNSB_FQGqlwi+UiIP=3QvtUhdylDB_;O6%;UkxYDTueHayA~STXJ;(24^|+Ys
zqq;!W-IAJEw+BvN+1*9cVjUoGZ6zP(qChmF<wm*O<g602k@P+C_rI%$U)pQZhxndX
ztG=S0j7X;o%Zfn=6V9P?AcF$@7JLw0Np3$e#npvrj{Jc7&e`|zkK-RxnxvyD7~w4u
z4N1IWV1j9<zL5Vj;4;HA9$%$iVK=SlpRws%mkpe#aj~HYvhLum*PfTz&4Apt!^~&8
z2xD|!28mccz;xP~bT2G%^FZ64KD-YBE$uUMzeDswxKrvQEFzn`k1h+^(DLSn^fvKr
z{a)XCd|wkTi;iFV%gCiE7qDWBA5G!^)V)HSatd9uIDR`=ZpUEM?Ed8!Xa{fX%ZDA0
zvDG}3?~nfaOWgqdhs)I%_8z^h);aPXlR0vWqArS^bdtN&?}AXM^*g08Ap)m>@7s9{
zq6vttpykV8WVf>M_j(!7F}M8OW0vR6w8~p7{g^@K^yJ9+yCG^Y4_`YWJ~T8eOEOf}
zUBFfI_MmiBnl&j1SW}8+u9D*OiW>FCP*yHZ)4SkNZom0P24(oAkC%rOx-W=D<wX1F
z9V^Z5kpFX!!&<ohFh{jqJ!pj!#@u^~)}7F!9pSfkv9b6~{UBZOppwO&mxzUTvM<l7
zPs$H<=~cXaft`5u35wwBSDPD0$i1e$v}hJh;#P5T#(;s443KGB0z)!(=t==!wU5J=
zXg0=d?Qc42>94Zb7XPc{B$O=#1(z`4dudlSNZ8MK$uR@+4uDKLx#}!ly|S^#yC*>m
z+{D7)TdD3zt4Q4R>NoKFYK6JyK=%)SY%1sDZ5a`z0IOYLJ=%6*DX7%NpZMnsVs+T`
z&wzT}yQcK=tUvSqh`G-OxyFax7u#_&-7yVINh@6ZT)<d)23EEQ@(~AaP1#inj#h+K
zOEvBd80+SRXbssffyL4F@`KVihX~zEObcg+y`HS>JPKkTJ(Y5G0I(p1`<RfeZ-9W5
zM^=HnB=C!01W!<4q=@0>E{g-05j6>In(v)MlaFnl=i+>-%6p)(qh4KgKQRMo&m3^P
z)z=HJ&KrFsF`m4GoM^<RpW1Cs9D}J`#7Yv#DAm_Y($OupxZ){t*R|wT+NDgQBKF54
zR@k@%938t$L8atpi^5SJ^3sPcm1Ap!E(m(j?9^fjU83^lGOtNj(|<C4_V~sR5&A3H
z0fdM+B)O%kNHA%k2dF65kcg_4Oj1$aSVo}R+2?Ie&-UKXjsvPJ7&t(T3DFmr{1hFM
zExd388l+HhlikSRAj6I%CXk#Dm85B05Z0cVK|nP|EeibK1ERC=l^Z7Gp+}l@@K<|v
zrxg!^V^W^eNln0=A&dJ%1rrwUb|HQPsTOQlJdW=8%IHgLIkd6qEbr`g>rcRAhXW3J
zsgtGl*EY)Vi#QpJk^|&g*Jg{O0D{q~zG_!S%PpC3R?R~=&(bR7;_IE8O(1w14S~|g
z@__!mjJ&8w4avxxu+BGQ$C<rOnk{8W9h?qx^wI5^Hv32(6tTG+*tJ)Q)^Ty!IE_`W
zq)sI3$T%C<+h|6KJD1M0+!vXUbytciKg#E2I&kNJb?hZWPBvc=DO2Bl7qfh`UyFKh
z_+}-vpOuq<i;T+mxsK^~@HDQnravs~)l((8t$?uZ)3kk&-U$DQs$EuTdU!R49xExo
zVhq<V7O5dnwAcB{)loSus0+5jr|2d6Hoa1>SNqH?W!zldYRk1yolY^F&;KyEr%SRl
zq%Xm)h|U%u)Oc@>O-_i>vv=h&$K7Q>0ZP64o>%zj%4?A5iruq<TBxg455G`E>9iaq
znMqr&Q?}JnB^h~yNi{HIYBx4P5M+=f52SZ#8dird^tz2{AIj_4Ki5~y2h;=;{75M`
zpi+cI=fGqP%;Z53s1xmW`TiZ|4r0N9S1#$KY()EiNKoU6q9h32t&LZk;!9h9WAm(a
z)*8x!e|x?da<1XpagzXBn9K71CWW-z28mMtfB%wp!0kJjsJ_@zY9#(2gcOrT8Tq$w
z*kO$+jXWoiCQjczFbttR4xMaNYm1xNc^h?Gi;$?^?bdKDLJY-82M=K@lnaoRLSJrB
zUJLUn!{>bs^FjNHT4&CC*Im~yA~tk@h$u`$y;i+=YG<-3ke%>XiXRGv4FDHWV+)rY
zP?18raxf2)!QK=!P1k!c3>sjwqOLi~iuYs?5z2(O`!K1)9g<6m+*8}_t$J5v?Cl}1
zrgF!EQ)R@<cSor%(t+m=UaU!+nd3z_cY>RtQ_a7!UGYx?MAFl!(C4rYo$3R3Rc;<0
zAPxP#iLF_R&7;4#+4lodbac(!>xZMGrBBlJ9eZwm=xSNwW8dA8rg(H+FWE6JmzKp<
zZph7C)43PdIIet}dwAjqmU6PMvD~44f=-b#zBqlN?fx_4ZOun8uMZRvKyC*JGhO<a
z4qKE1ekq(hiP(oaEq^vq6fT@T%a`M~Jqu5<HMseaVj?U^8CMe}6JmvNLsCc0zGD-7
zM1E+UjE_Opt-Nj&Z1^1*H`@ca{(b-ME{9rvX&JfC7N5Tg4b>WWw(k3s0GZ_fA^2nF
z%FOYE;)co+w~1nq7x+M0Fch|KdHix{9m4mRotDKb^W_7q&dg+M9s{nvAs=3c!B^R3
zYs#4E$EtLHIcVMRQ0_3v>gMQm3(0WPGnkL}BuaerH{*5oy?7tx+5H|N>%r-?So)C8
zdxz;#3pS#8cvH#}+daK;b8IRU&#gl#96o7XAlnN@cdS%t64wd_1*5?D*1zY#<qiau
zl%_j-4<dbZ;Wt$8E7#`d+Wu%>^^r9{SV=g~x|&^PT&>O-!w;1Wy1W%h$Ncdm9<QUv
zf>c^w?{weF4#+0buAm+%n?R+oBM0JEoe=us)^D0Ho-EG3%-h#|JcnELpVr(xAW?Qe
zF<g{({lgHZZYVlo^b^L+yyP7NE^<mx`RxHBTgmi*mCqneDqK2_u72H}m{W9SwcUP;
z>nUdYYPyqngrGVZ;C<I8lBWpF(nf$W(10+~3J+j#=@PrV+iI&mmi{|Kz_Mp)rx1Zj
z<^>ZPKrxz=%)l$xx=NDdHk2Oe)$!X#|509R<>*f3?!SrOS>|=X$<#op2`}cUnSLG9
zJ2I+4E-`L6sUffw*;!N(g~Q#7GTo#7R+X3eG6cNnCL5GAFQ-LY=?=xVrI_weAQ0X~
zCkE_0*NF@I>YgRoOD#O87u8;k2Y%;(p~4ahdm!!%V7jZOekZWurCr{Do<rI*x5Js{
z=6U8c5G>CQ$@3Z1){sGAvr(X3K!E|;0Ki^+oPsQY$qVpH8B=QxmYx)rAgDvDi`Xfp
z1KOpIg!BTxuA|sRB<h1-xWPVK_#W}H1Ez}0sis4V&W{4l5>`sDIp7Ivi$3_<*>s%t
z%^tG(S#@(DCo^?oi&{WmR<9JzPwOKm>Z|XqvEuUl*0UPjO#EL**By4DyO|X=A}?CL
z+jscQHT5kK)=}S8rpKX*2GygHubr_q5wUlCkKMcyr=S@3P>_F2bveE1$w1|gjQ-4E
zD7hwLi_00Vzz+Saa6tXANb^m+<`|kDXrLA8HI!xjTX>;^zU1UY+DcDWgu|f0F^LRM
z);Tw1jT68|0Y6O!VqpFMpDTb|&R0iu`_UfpU-cLvtVeUj4#KY$(1O-oVli_!LrGjN
z3{Sk(^JSgp!7xpqPLG0(pgIa%JC3+Lm)HExf=Z@MR><7$mU*;3myYfN`G8zj8d<!k
z;$Pl>ke6*=YUY7E%%^x~x6PicPZL*E_rAqGmn&1UeUE02&D)+-EgP|F<KnviJgIR5
z``jpD*~Hmx4=jPX?x#a4f!I@BHBQ>J?^r#(etk;=9!1Kz0}m3U*SEf`(W;KW3Bqnu
z)MEYSmjuDDRJK|bYE)u>^Gs1@)8L_QNAt3eN^)4DqsE@sNe+HYxrly&%nUP~+r<|3
z8M6Uu2?PWL@@<JgXE=zw-5DjD2b7qgV_q2dV;NaYd|9W~a{`cl&DlKFf_dMQM<6VD
z<=4j^A=@i=#XBij2CrP!c)8VXA)p?;i;rjuRyXt!Ocjit5Kq^uNHrm!9-75YAn~Oi
z3nrdEXmuJ*Ol^Jj&os~!v;2#mefPEYI2MQ~98g+6*i|UeOOgv9rb=x3hHLn`Rr-d_
z)Uf79a^nBy%hDVC`z5XRN}wJ)O^RN^x2%@-7XWYqEHphgIZzL$!G?5=-UQQW#aaHY
zi=iL<8drpU(nkO+>4+65Wn&09k}32Wn5jQfhF!Ot!tZxDg=TG~p9m<`J7-jXC290~
zWhOhd+2m1EMChaZ^G}tvSp+BX*FQS2<ErwbIEY)ye-J5dceVs}q9C7xHr=m#DOSq%
zt@0nPrcEVyk~4cgvi)GZgJG9vBEz&IVRtXHKHiQ}1~yke3R;Q<)<gjraGL>Rqh*dI
z($O#BA{QU@WU^`{!lCe(i2rDiUuJF|usrc_f}^8?Q9^4fC$R2<ELCR@fH47&Z;-18
zI%EjQtd0VmY|Q23N~>|MQL#laoqK)1kstU<3B7J>+G6LRS#jcn_P0u{nco#d3;cFz
z^5%)~!)OG!z{8{$3k6~i7gmYdH+rquH8<NNH^&Hi&LfE0Dt-GWF!rEMssq$9%Lsw;
zsA+h!?)jCzYtsgCNN4s$kMnje_O)pABoy0MmB4=B^JxfO5aFO$Jaob4FDJyu+Zt07
zE(#=Frp%x!7ipl%#?>HS>~o0rVT4Ce<NGhM%5<c-LK-yzBBR3*q5hu~oX8dz9HV81
zAO39n9k{`%JTkrkB2>7#zP!?UmTow&w5i;yx4GHc=X^aUOZU4VpXSWax=n2+Kt2T+
zg((Jx|7rHrd8BRJhGSv;LF_ID9T)@u<Kf%*?<s8fRx(#_zvq7tiqOpE;N4M|yfoP@
z-Se>ym`<MQtXuFWdWLzUtMO>nI4MaITn8uM`UI6qPQCR;$Opd!Wp^<%%);a4Ko7%L
zMweU;cBf5=`)i^}+VWc+(PE#DNjaY6vBg_{(;DgB`zehoB4cZ=7ABa!j(wOD9wR-I
z)ERM>Av%bI&@x~=wu^n?P31$kCd_eWpPk_1;BNuFFzqkYn6_&<_5Ko^*buI@?~F~G
zUVZ!X1_X6a@FIuf5YX%jghlnd7p?B*e=1W_hxdZ8ck^Lt6A`&s<I)G4m)>sjxW>P1
znqD1?CJ~-IG3%`{kyj1ZST6LY@2C{}3^rqbcw4-D$fk>ecg5a`fDWmOWe3LYFAX|H
z+_x)i+@S4)rt)C%r)iZI+P^1KAI9|;%l&TY%Tc3}Vu4)S%Wewh_im(!!_5u7F*dyJ
zm6L_7n$#GOWhmOVx?OFkI#-#lA`ICuYhsHj$L{`L^);;eEp|pGHSlR86wy^mUZoia
zo9!+bpy?Danx-<ds=g}wDA|@L@UMMmb$7=>I(Zr&!!wZWZg>1-SPe<aHX2DB^Gqs!
zu{cLVfnRFj<xXeou*|hvD0;i2-L{Ye-DqyrpmL~ysVlg9UURIfb>26(&9weo6gCJC
zlynq082tbsV*p@y8F#?iG}=MSG=A{tqhCpiK{=`;S19Vi!Qf~}4;iT3g}fVtH0u9m
z&qyGP5Qa`~02^cA2skF9mo(hEQPrXQ<fx(@=1GiNlOHAl6_$U^c)-ynjCx$NO=>^&
zUWF%nfkjRvACJeG=*$B+XW*SB#W;4M%V|>g!6OrIdG?Qafvs}L-?=9t6g4M~IX7h8
zW97Sg3gq+0_5GK>{L^w3QZvi{_UHSuzkPz0;(mYP2tRQc1?5m>u+qnn^x}cNlt`v*
z5vd!!-BGcZHayDyi=Ie{Gyr4=@Q-l1<bYw5?J7JyaH|9p1*e|={Ws&>?Xa%VZDj_I
zD_PC({e`nsHWgGy@Aq=F-Q7@Z>A`TOIY-LfV)T<GN+$zVFPe}!Avw!`RCqly%yZQ(
zR(!m3yYG02ul}(O{OSZ%X%_*d&<Nb=ThDeJYJVZ<K{aPEJ#;P*V@Fo~pOscOht)yv
zw>i!Jv@jm!E8KtOcZqmI6amtwqFISsgJu@VysL_kGh;_x?5B6|td#Y%;3rElMJv6R
zTp*2XZu*5?RnUXC8da)8sr(jVFFYm(o(%E5dHOHR4G7s&mH#=biR_I`2XVPi1H?_y
zKqST%Va}w3$es%#oWr(ToFS6@eOTX0^WI>;m`R9jOJb!BhrGnNS_rD>!8P|c6pKAp
z2S<=B1tk$Xs#I)6G)1+>{|k?zS35DwH^J?I9SC)P*&R_SB1br(szK|j@5Jon>b6g2
z47V1<0VeYfR?e3W1?jjSJyS#IoNEf-7_OgnKt6a>ZV^JArB{maS2Aneu&IX<UL}&l
z^4=8<86#N97C241H%MQ-R(tOZ&V`<v3!<3Rcp|{$@lfIF{z^gi?m$WoD_suFJ6*{!
z@7Kln1Nhuj>|J3{Q6EdNhU0*DAk&W4*yJgS09LUIEPAWR9;=FW&+f^)f47|zRL2Al
zc{93HJF7K5;>wjl_sJ|Q(f!f}1M($H5-(Jzv_TY1m!c*lW642sj_P8~?UyaCOkT&+
z&GJ+HFyf2+yxTh&Z>>+MJZn;`&q0m3s=^2|FmT||1{X7;D?K}1r`Ru+r}DRy*1%Yh
z+Y{nS!ORlCD*<K++FGJ}6a+-%hIrzIPT|675<o`(3qwPYK{}k!3LNChxsI7%#7>$7
zWxd&I$r+5Q&Mcl2^Qr0^4D8PK%ntse--2<pp|5vW-$5nSuORbD`RS>;taiikwD|pc
zZPXoIslRpLOKvAoJt`2-Lv-L4Ptb#t`$f9bYV@(gW81XL{WaaM`(E^%nwn~nRq58c
ztBv?_{ledm)xFM~8@Q8mpyBO0L3uw80f0x1G<Z;+xZ&=e%&>aI!#&80HV8;RJ)31v
zfPKlx3G<@&S>Y=OQfxbhwN9g6HZ(+y2P&>yv=SA5!@@f|5}1*3`5?s$)LV9yBx6sj
zi}@ADH_NT+?*lci#P@Rof^+?v2-y@Qp_*|JU(NWE5gLM5YvpnR8ZDkkw6rNqNR&|<
z4}B8T9E%B$x(b1=;$^+}!d;CX8;)*xd0uK<^`++ofI@4!yx{m_I_+9CWugrzfbGZ@
z_Pjeai5VlW{A^=teRV7k^Bpl_nop6<$}l+K$h=Y2;45X60mq@EmM8aIKC$@%T$F>U
z;q?MC(97*!ReOqSD@q(QQ{6|S%Q+uZP;&VCX3X4@SU^)ENWeNElv3}2PLD8W*fjbl
zN|jG<?9j;v>?lxW<zyHyehi7lH(V|#0K+hL|8HB(D&KMeE@vxCtNXik?Sw6dyliEs
zLF=9}^lOe}H|%Ng#!jRo>23!<cr9LC&bNw0WqF3-F8l^-vN4s>DZN5S_Jp>dXz`!I
z$1sP7FqwC3+wujKF4GDf3N;$F9nwJM|JNRyHcZ-ufJ*HCf=&$n!YXCR|M*QAr=L+G
zm-w$%*vfZO%qtg<Z4yVu$hrH04m2B|m{-J-&t{I9tLR?n<DL6KyN)$Qr%af=AgC1Y
zV7iOJq^W6Eo~8yZ-h|uAA$s*O0fb8!ajy#0Dn(>UE&wk%S4AS{gj;2F+#~G&d#2}x
z?r~S^eZ7%(bME-=^Ldw4;SU;V_$!oejhZt0<~ur&%U>FvDIhXatR!Q6wZl(xkS&#N
zTcaEuXJxA!d!$^|=U?V^x}0p6c}`FBVP#8Nx-;BOP*~7;M8FYyE}Cqi()!Hk7IOD=
zyuuUnxrb>Q6qU{lqg+5!ey@}2!!q6p0FsLHaUe<!5_u7ut=1%*01*_3kOnKR1E}Nx
z+vi8-O9r5>_5%)_dZ;?i{T-&n;(%^AcQ<A-v-r~5bm7C|VYlGY`Oxx;$)ro%LgbCA
zLcVk)D_FNLHtqUNrhMGV=5Ic!A{DdW$V-xr(?gC>@42_U`<%ykQP24@Z+QMB&NECC
zcuYh|amy%Rj5THyOPkWO-9ONZvn$W0k@#8xoZ_msd$mV2>0(1_4y*TqH|QMK(39s1
z7h4}*2NEK`d`TB*ho{7oU5VQ#njCWK25b4T?Il&QP2yD<Saorske2{)#Yf2x0MJsB
zVYo_?9KN&iN&wE}wh;gmMNEnK0%OMV+)^%~l7bAl#+a@Cy(fFD81%4klGK3j&2f@W
zo%ewlJ<)47Cj3_PJ7#JO72>yRnD_|$Y_YFbaH=h<IYE3Sr$SA`pFx)c{vbbgE!Q}T
zI0m}OQ3h392l4ip?=@GG{-S|0?|KK)13r`N-?OA<!xhW8KS&WEKOGBtTltDCQ~OB<
z*!EcyR5D!Dx}v4B1+eKptQC3mdqJHqcjBZKl|*VibRNlccuX0|vbuz={lX?*DIF{G
z(xB?c*))v-6+lLZP7&ZW3u~(u1r@L<5x@xcTmK3ts`5ov=&6ZE$_9#t3YVMwG-AK=
zRa+SSAVn#cHua#zxX9HfgCtoY$wR<A3W}v^lilFxMY%fZSgp2!zI{<Z@hEz*!koJ}
zj!3vr9!=mjMk4e*jqv|`^o>|rX+>4KP48e$$`9te%}a!bD(YqURhFrqKS!VZ`5&a8
z6Or-n6=kIqp2*$B(sX96udizXGw@LuI#%O56%Q*lr+Msh#J(H%;A+QzV&zq3uJfC-
zO@WOIB&^A<#3j!hyy(*d0>rZh-GBcAiY&*QaCax8f4`e2+E1@#mtjCT?W#&uf4VNC
zpq8p2o!!?#IAZbPu7q;00q@fO`a^cpRSz6&JHjIr1#$TKH_EX_$2ZjnYxMn9cx&|E
zN2`7eaZ>c6gOV8M{4XI&#wa57CAaIjI~QB>3uV?lS;6MF1ZUQYGi=&ag2qv&WMYrw
z4(N8K#5~4CA?pi>e10$pBjlJz)dO9SFxpSt?Tcte9!KxuA3mB>%s*Kjw{BGq$;E--
zx12+Q+buhUoofRuO*->r5NR<V#9t9MGkgdu2r$G5z}8z>gk{)=!luUpBtVuZ0#r4K
z>Z7Z*RG(~fPjK0<^cXXgVf;v$|J;8dY`2%Xi)|S7;B)^MJG?fli*ELjoA=JdT-#W^
z-x+?vea3a?DYLoekIEsww21LHZcGMVwGI*Q1s`J9KajCMhP-bk3=sw&uy{NUd*9_;
zVJoInN6J}{7xq+ri(3{NJD7U&>5lKTKu*|p;HM56WqxG*hxmy3Hpyk2_A0}7m!<7U
z{k`LNcOSib1bwkbZqLhdPnNCgoSe7SxPt#rv1t0$apph6v#|ApG{OvK8@F&_l^9-}
z6!1VtK>@Y5zPV9raoEbq1H1@e;D!PRqGv8k)S5+StAnf^7Yo3pdh@y!{{Q-5s6Lg_
zA;dGi0G=7aTeJy2gNF885+h)841ld){sw4-=^l_|s}4oU2H(wqYMlZLhT0M7z;4-n
zJuq~#Z9}y-Xt;Fxku}bLkhCjBj2*6zHNtvbEz5IjCdig5evdzYv8rzx>{X}i+ZL%Y
zh>ap+;3{0^Nke|sBiq4{b<5b2P>HOiDVK=A6C;Oa7n^5M&=h#(OU=)iuTQa1o?gn<
zV&jFC@}umfDYcSTeB>Pk-eM<i=S6wM$eQn+Vnel7HRA+&?s<^iaz42UkZ8F@J=wq^
zdoSx(=vD)>s5#FSLA!+WR5u911XK0*%5`|S{gnTSxzYiHFisRAu|#wt`5=PZC_{pL
zM56oOApGSwdz4;Yu%Ee9D6e(~eflemN$i<eu5*2Xy%$qg-Y$zTY>pUVq#wR`tVL!Y
zRm~&l@`~OH*FhiY@<xQ6#&Ok@qk6~oNQVA3P4`CN<vV67PQ+s1#zoF87XpgaZ07Iy
z2H)&xc(RmbquJLon=#iRKRn7?L$Bq5Tt}Fym6G$jDCmg%;a)S<RZ@;6PJ1Ude8LPg
z7?yCBQ;lS$@gv;olnyjF7)fcG&t9(xwHpIFxUQf}7GSwLX@wV*-72^-I##2^7b_`W
ztnTWmXS^1hhX$%##T3;}we(BgY8%`u?T2j466<o3F|`*^G<e)hvSBn;|A8i^MBlhs
zgX;QyTGxLD+Ye&xl&l=e&NiXAGwC<%R9lyPpNyZ21MF-NzL<fRfCG}PI=DgtP!(_+
z^%%@=tuAK0{s+eAPvwn8-CsRqmZmz!*dtz`-Ywc_!xmd0$Uub^?4Js33c=;FT-}uz
zG>K=AnP`KPpSTHFmcT5EGXokDaCZm9U4V6}6GRPkl-+s`O-|!pn6K^3`y<}yF><)P
ztDSLscw@lGEp)Ye<OyF4Xq5XjIr3MA(=TuCCWS9g?z$1r6+8V2Co-<QR0?_#y2wfk
z&D||Wa8<hO>e+jNQF6Xqz|TzP+^1_(cfcX^I6GFs6FXTF#W~*T(!Wcjzxh8)SHFcK
zd=DGN2e~$PEjKw<U=O<7^)K_DOLAGd`C3EL2C=gs$%pWv(7pfUsFz4y1zA~3+cC?2
zmg33V<*(oIZ=+U|*6iql`=3J{w5XNZU_47hd->9Z-k^L^Z+=CN(&iYk^?Jl;Nx1HB
z3*>$0F6NED4~_Sj0}fv@_UY$$m-mZ7uz6@mIpFSQS7ob%i@lZ7w2r0j`p;7|-`VT7
zhZj5WXs~Iwkw<|b!o(8=F7|6^=nbhSgfp!XtvaQT1itTvRMPj}8+G`!;h=J*ml2*b
zpG6sJ8J`#16ii5GQdr?kZZ9Ni)o|O?ajtEByVb5T*Ub8J?X>+jNWuQPJFKus8*vYa
zrvEcnxa4b^W|pFX9Gix)fnUwo;ewV<fmFukk0tK3P}?~8$DMg!-GjHqZZ+LPN)U=X
zW1HyELyZx8^R5M_?@S~~R~MRHE!tW9oeEQ*ByML7ze;31OOaAAWFwxV1{2@n$A8~s
zguH38aOY<@!PRR?Ah>HM=*3~gP2rQ!P$vdMPzOC#U!g}6PH@bZUQxaE;ib9ruVsAK
z(Dhl~%1Epi+LCo{c&YXMEH`s#sc!3)<(Pk&zOU-ibbeA{Rd~Tqx^e>jPPvvP>K}$x
zUys+pN6RGLz;Gpo`FU}K$(O~pdrdlVzD|!AofM6XnC(v+L(^YZ1FUCwlH*V-%an<h
z{cT#hMbifdBlE*8YEAsC8r)-&&ZcpGM=7ZO2l2;@suQD5rPQYh5CL8|_XXf$Vd=XB
zQJ>Kt&SF)?2k0TKYh}!3g#u|yf;vzXkOh#wMy&Lq*~Swb=n7DMyfuI!2<Pvi=XamS
zvefl2<oy9%_4m*;o4r+G=Z7mZA)v9^V^%z<5<It%qGb6+>h$f@ebZyfd!LaU3NW#O
zUBupF{dZ`B`I{zEiFx#DT)2N-N*YB4nJ!;Ovc*cC!{@G&E$r&{D@Wlndq3;VH(w^S
znOswJQVk(l1Q?kmy&jTrCq?CkCHfi59#^cHRe8H#ZPo9M)#&)S;1htLPtAhZ6$lE1
z{Z<TcWaix2;Y2VG(lSHW)x-jy<vm_MWHvi=Di_ZfR>$ceCiPTbP3*pu75{*j=uxzo
z;Wb9eBOVe&=F6BTda`gqKJIjBVFL&0%=QqVbOsx0W<w^_Luf!oR^6h|`_*bVUT5J4
zAfF;Pcl}E{Y4IQA>7IXh^TrQ9%_qBSWtshzrk<)??Z7<V+j3-a@s^lxZrojOs=r`<
zo&9rCy6^=nY~HR`a-_XINk8w?Skj{%YOuM?>=k0{t5d94yP&&@CI&q3JBgH&3_F2B
zyzB(J5Od`;1u*fp8FwXH)ZeBqtYoR|R3)~bAWoI?SG?~$khn|f79ZgcDL<-Oas39b
zRc%nk-{0L|Ru(=V?IoPl{zVK$UtCpd3i*sv_{sY4)CM-sFWg7n=D=3{+H;djXA@S#
z{^lULBs+y|rXzK~d0G2&ix+N}lOa42=q&9<q9Jk{<j7_sgp^yLvajKR?Utm2y&oHe
zOjjovXaf&CF(H;zLv-0)u`3BXQC*n$Z{NLM%@mWOqc>lB2S7>~z95iwgo<v70BG$+
zmaEl=UCFn8#QRkOfjrI8tZkj2zIS8BT>0ixza@TMG!^R_7oP<rgx9KXm>b>|y$T>Y
zM1chgwx94nh<4I>%p;28u{9X>=jaY|jkioEbL`F2Biq1qGlpNR69r~s?Z>pG*C)6Z
z+0A<@qGo?Qz26=!d1t+8_*DXZ%U~l(Hv=C=fX$`X?}Q|C)5Eq}S+#YCs)(Pa@x*-M
z`EY}$ejLHFk!nFQL-iKIHPq}TfUpzjrAQ!)8ntbvs7?`jvPJpAyh7kc2Hrn$ykHw@
z8LY0q#3gog1!%z4oz9QO@0@9yAk(@)9d_TDk`reG&#exueV-<y@*HE?LzE||rATQM
zgwI?jsWGa)9CaA8F*<rz&&^0@@07Kr2B!2(i_;`F;Pj_tDDKF4LT=m<lP~qdH+|L~
z=j5t?z(J^8gQ45l(k|&*-~CuUYu}3QX(_N1YihmZaaLKOx*M1zz+o+~L-+C281oy{
zFZJMs^>R8OR%$G04-yv|NAK;*htSi=u;R~o(O<Yap!^{)PcWhK-(n$`16&4+1frK2
zs*}3(ZOSM43mtfD;Ly9BpIg3`xaZYYU1&VgtZVkCE7{}tNr~^n;X$c`RyKEY<4RTR
zyUp*Y3wo==Ve}TB5+MQG_`y%H?|Gjc?HnhVUI=#Q`UFe~(zMr9M`|rv<reQs(%Y6V
zSYvVZ-sy8((ot$6KE~y9?A|195h?U}cn0F*Q!hGKbUj*!^dW2ammS=x2jZov2KI%I
zda>PCUIz47yKIYCfON!tlPelm2Ls%$mhVrf5|z+2vi6@t&J8HX>$-(c=WG4_{V=BP
zf#!IrxnE<aUy_n+IZYa{x<EJ<@G+f&ruqQfMmReF{gs+Jzf4_W{h(pqYW9%jJb}<H
z!XjZkq|akVIg#544Qj>&_K|oq%%ePO%tpPmY$uj2v^)*stA~K0H&q(A?xTRkC*UWT
z!CczxkKy#c!y!|>6C20@uhUIin*lR(4z+q!{|u~6XRf_)qg+T>98^3rD>x%&4@prj
zD?@o%PWUk2Ibv$M3Q^6i7e=d&y>d)5aLn2lDH=My5SZ4_yzA7g>dakvxojA+w@Fh>
zT*tGjGetxjodVr&W_xv32HP*QQ#Qryh0glr@v0g5y4JIHrjB%-M$&bS_iyI~4KoI+
zGzfWVwr{pg<O&XP&wq<;J9{tSyU+5KCY?m{2m6~$C7ML%<@jX?Z6#n}fkw##7NU7B
z5M1g=aoFN3fP)BhIcD<U6Cz3;_|V8a(LGquPB>GEL>KkKr{X-Rs?Ws{$2;~PiUv6a
zHq=28eUL6?QGyXXCO<|7RMVxWx*Yqo)264b++#}8X3m|NbSe`2iO#6(z@rX5zFGG3
z&39iPVx<W}(g*K;0s(D{>T(u6Sn>Psfj`R*Tyoi_ZZumNG@zFh?mtZ43x8031HZo+
zF<}|zl|Oo&y3l9E_!Z7t=59Y#^)WBZ>`U&A@h(`u+)UHtn||#1KSX%%_fx&{&Sk-n
ztOXK91gYD#xoo*$96i-Ut_HitqAyHUx#aiMh0C}_rb7;zTdf<(9D_JSVe=F(V$)r5
z5EBH50LIt9X#xQ-_bK*gC0FsG%umc!s^Jd2eA(p+oEyML5wbEc6F0$EWwH960T*bM
zL<{GWu3-tSO25Vcuk%Kv-GJj3j$$w16wj3ITSKoNSJ$d=YIxI)B(?q?;%UtO>kf@6
zzciJ*#fwacIC*N?!wlV^MjH2@0_yrfYPPcF7F+W3n70;zD8FqJz^KhD&Hb=Ty1zM|
zJe2*l^#(lNSAu7OE*nEevmXNMHZdE<fgeNZY9P@JdyXlKcqR3Uj19bg^<>eJtd9Is
zs&8O$qdLxuiPg-QApi4CUR}(fC)0`;=3Je!x375VP35=z7*prZ+GOuIvuMnlf!~x)
zR$6{?%=%;KdZAxWi6XbIB?fuIAD67;{%>>rKa4eX$|v-`rC~;dm2oI5gd=Ktvc<s4
z<*0qT$Gr%0nlHq*^X6$o?7w<0Pu?g;di;pIR(czoKM^`A)#fAeQ|zdIK&r9ixi3qC
z9FZ+$?deBg(w2U-y8^oI;#83(n#AjPr8IIwTQ2Rl;8W7KyEEe?!wy~xz_WXTm~*}P
z6Hgk#MeFLbtld*pBggIha!O&3tirq;S984r&5GWwh5>4Zo}I^e0>*z3{)o*}vEJgv
z7~9}By*}YgLuQYlqdpyYdC_tQlzC>A3!9yy?2rtNofnXHG5I{KuCC;ojLT1YQ^)<$
z>5)$oUG0>_@#xlFs!!whfcUv=RwA*|JZnD<ghh&G(;u{4ozlFGeMMJ6VP8!fa|XdC
z)lL?UT$b)@t~-;FzaVne$fj(m<`~?rJ+D1(L))U+w?jIQME(1&qbAz^R7g%3?9aPl
z=s!dGT@K1+s+m4{CD~{ha3T~(RB=QUJo8LN<k99t2pk|R1`t&a7aY8Qku;##d625`
zi%5*Dl0kyuCK6a(sd0Ywa5E90X%hqg{a+etOD7p1p~B7sGGN2+*`S6${Re43-eGni
zz_#wqYCls^7#+qT0O633<IZxT)8A%m6&?v_IH)`QH^_}ch1#erO(E@dY}V%Cf+wRD
zHyp1nve3n#R^Tkn41=Ng39`kLc;oz=#eG(>v!c1=%V{{>9joS-aatXbFNPv7Q3ao!
ziqfSa5)8oJG(o9hr7yXQS-8&?EdAApozS@5#Y&hRiR{dt1E{w1EVTs1$}2q|3?e5N
zFvZFn`u?v|4R0t{gs4?+o_u#_T52PDX{BpJ_Gsc7@PkvNbI-X`D~m_Ou;8|*lZAk6
z2+pNzh}o66rZ>|H$Z=9!p$sK~5y1{K-BpZH-aBFbiqBWqFTBb)^IX}Ovj>}9syan-
z42Xgr$|7kF{1!B2U2sdKa!FwfU6sbi3=sc=6xw=iFaniQ45-d~Y1I}z*{xplp{@GE
zt!Y!<%3Z}dsWxq<`>3v>&VFG4sy^>%MAAy(gmt`-W8L#|XXyKA?zh|UvZv-$I}h0^
zD_BijH~Zx-LHz8ko4)H81IIOszWUYvX|<tOe?vmIRr9WE3{-zQqz&Ht2${qosMLu$
zyWVfkVoQYP1?!<gvLP4Jm>+Z<la77&IR$xoP5E7-`_SQrOZN#NCR@s;wR;bhBvm{1
zX7g4e8UDf3wdAg+&%Noz;$f1J?9%mt#KeHV<l6?rsEJWA4wqTpmde#Fb&>Q!ZEkVS
zlX|=r>J~IXO4(0IaLzN^P^ZxQIPaULkN?&Zby*i@tDC*?_5?I_cf7~R=A5vFU2ipU
z0g*9JY^+T3Lq9ooP-i`pk`+YanXEp9>B%9Z!rPh=SvjfLzgq3!V_h6nGrXi8;A=vs
zE44-}{58u3Xrf`nsK(e<Oj)pB-_)^KYIxrLQ|jXv2dv*aCfi?m#+p+O@YD4nHXCM}
z`d-uccy)WR&9i<@VJp}FlvPYCH}chJ>GK8Lzui(2|A@1L4{VC;>uo5FOdv^zO12ap
zlqR6`QEVdGl?+;Z(}G9Cl!lCbvd4j)$-7#*oI!gxMXp_o4ogls8^6Qs%RBU*2%bmK
z<Po{U3ZJpje+#<)PX7GkdHQ$w)47Iko?pM4a14#5&%eCVU>x0xKy4`(aArS*yoxLc
z<%D?fv!OWrLu|Pa(zbNQW+S7<%z#nKFSrmw0f{YR5Fq(xiiv^_OByp%61G=MmK12d
zAR~_`q$bmWDUWbh5q@OyL1y0ns+jo80z~K@Nj!UtQDQt)b`d<Wo!2fEAwGHtmvj(8
z+6J6fP}LxFOzfx4zaP)HbQrW6l)`VvCaSW_G9cH2eFMLl7t1ZY@@vUHDoe7fhT<T~
zDbG*8NW2|8XrrIZ8Opt)`(tnXx8t|0m%Td~xMN)x9e9YNhlcUKZw<@*l3MO<HNQih
zl8U7ZQ(AbsLA4Xv1TSd&2fr{)?t=k5z2m|ANcvyjZg>Lwk9S+mCW8tB+HBZ-aXmu?
z1Xo!C3y_9Kf;u|DDT6H(wr~0F;6=L67I=sF7rm&uoEAkqX4Rs9%t1U#4Qaz=mE?kR
z2OxUC;isde<C=lr^^PGrXXzD1wIJcdsuNpEPs<8#=`?KUcxSOuJYAp4N{Ld1<y9PR
zDnEY|o5@$LZeX$pQ3x5i9goE85Ov&=RDIt^jB1=~s6Jv3!Ntgf-|{4ZL>JL_q8A2K
z%VwVb)~3Itu<ojE^oG^7bAcD;dcg>9M=Q{M-F4uNeKBn~%3aIKkb=_}le#GSwRP!$
z8#qF}i82If-yFkb;VlsbrSA`bOkTbA^1;&TtlxYTM71-Z^tN5dF5P5h%#5)B<WxG1
zh}n|Nw7~Zknl2B3^PzITHH`I*rwcaROPP~iD^fm*Jtb>KrmJ6v9wrY)_3FG&BzlB;
zcVG|Dhw!^Z%93FwbG3=UI-RU?m34bhbJs5KO`n=#d;T&u;p$ZCs>j<h!8EFPMW;5E
zV~Iw*cP&w?Ejm}c@P|t)qusOk?+tQT9Oew~saeeH5%G%i@n3H_EWcU7zEJxY`O0tM
z;`xy4@%o>tHBvO-<4u84`I&at4(Z{C6r7qff-X?mXeSOVy$Kk1y8<eOHn{%6u<Ah8
zY^GRRm4UnnIId3v!iyhhKP+xM<HtG{gY2Z0;is0@V;WZnDK!Ju$Dbs4Xsrig`0feZ
zYQhotsBQM^UC*;I<uA1ap=t^eTJL+0X}IlP@hW@gQ^{XZ$_+00y_KOpiT!Fx@9579
z2g++~(%kWE(pYWkPNfP(>8zk|xwN_?@E0XHogC`F6QT#Ot<Psy3rq`sx(=ij=H-eQ
zf81TD_H=coJHS@?dbh>Z_xr6nrjQROpDz&NHSZMu2Z358I7pD=D+|eId!{XF$$~(u
znA48Z`VLbk+Z{8LIx3(D42o<A4f=f1(uy#An51e2lj&hophC+=77B`!U;=M%j}~YX
z4+wdF{YYwBAPe9)zTm!QEL6~ekEviaQ2K~!@bk(9zc0hyCxUwvNa2`*O8uYsPWVS%
zlz=Y;_yi1agwR6sL!rA;C~#IVmv+F|%F0jITk72V_?-`5vz=|ILz{u|8(f+JF2{>d
zG9mP&_XWq>Y%8~*8~i^WF!3J7S$hn?y40SA9NZ~;@an1L4&rmwCN3Tv>b+zB^_Z7E
zxs*16^+-+X?#?N1MGuQ}EH<4Q|3DTaG^!)^0bM|&f=k<<8vM!ux1L_H9aMKSbV0oB
z;??AncOI8EHpF}BmV|j+h(Wqe7e>Lo<6L>e7X8Kju6?;JH60X~`0A7GOxf!D9W4_$
zxHizt<lQQ`9SpZ)jqLqQq`*T~&N;y=+{JX5fRiaC5Wpb;ClJ|Lh%bn2quu>gf1<WI
z=9b<ch;AjaoT&#Z#iQM|UEJRgAgS&jzG`GGw+vN5lA)OQymlIdVgKTuE8e0~MGbpB
zBjjifYvh|P^6#nY=|+zyW==PY_cCjT)w+n|(XJNwB3Z`iEskXFEx})x%lQ#H{wqb(
z#-WBX2SQ?iLD(s|Dr(^J>F;??2l_5x4!8WFS7da;61{=*^mA(9IL47h!}(-b*5R&R
z3$s*>l$M%qQ~3M2T#<A!|HDQl*FgSB8mlxHG4+54QZp9QcMHuC$Y5v_k_hAuypOi0
zTs<g1BL83m^JZh1a$vs{rlO+aruJg>p|_F^1rws{NCb4QHg=aPFRIr_p{JdGJ9jPW
z>Sp}?scRlN0WJdRSUNg~(Ke!ue-LR1fv+W9QnW#Mrk*eBeA(Z?Y0n3PL(Cpkm<Thd
z;GJuild`qytDg5h!Em&2&*MV@-B-xj000<hHlc776vY)$N6Gs*K^n2!@^?>!l=!Yp
zwRc*sym0sKG(Vv7=4vf<e*E%)H6{GvVZHYWMiysxEY$CF<7sd0Rex!9k&33gKN>{e
z^Y!^>lgqGM1sC|dTHb)PR?~5nx*=1O*pxYq5$kH%pmS|K|0*{Z{PuDz{&7mmZ&^|+
z$2NW5o{A{fO28zmAs5fItb62`gYnHoTj)|73#^PJ=}ld@A;aoCSCEu4Wq|x{4Du3V
zn}d~7vA21USk3}29heX0Lu!CwiWIn-1B0SlLXUGcEROTP&p`Ns8eYy0iV8b4mSY0G
zsJIsbTpyVK&-DTPn+Udhr*mc4bYcN;V*xKR7!immaeKaB|L$9FssDx3*;wV{ji?&&
zYXUy{`T>C0;r5tff>$|Kgmm2uSfAl3j)(>HI^d+sGVd1a>qs20nsL>9WOd6j7##(M
z4w1_jDwzd?M6bsQ-hQtr(7DrxBZj<jE7vjMAlQ{wjwulQRH;%JK@LBE`d7g-i{+;B
zZDV>vc#_J;9JZTg76ENhyjDWa$$xtf4c!%oX_~(mB)jmDNn=)5O067<b&#;?d=gL7
zk6_5%a=KSRsWx^cv0Vk!`yj-d%OYv30IEPU970AfmZ>A)De-vrUTIdoeGv&_Bh4y9
zK)x(my1zHkHnLqQZE<Fuz-DKOm#gCxji{elC540dM%^-0i3(Q;_qFz-T2#SqEQ%ds
z!>4Lt2l<z)X0BRPA1oqiiZLSbeUQlJM<QjaLt0jzJJTvD>VYdf2|}g0#LG5vEhWrt
zY)qy|1}~$ho%iHYimiArcepx;pC=d&2>J+Qf^f(UI<$hRx7=~F7?UnR=x@81y|w5c
zQ0yoDTn_P?V&UwK$lj}df&&vwwLqiTBBkJr(R1G-YM!36^baX)Y236?%joB3nE#pR
zTQzrL6KH<L<-YRvK&$No0`DeX3->_~-3thFGXFtjXS4o;5Xuuy1ibV0=wuYVe|N6O
zqten#>R^xmZFCBga9<5V^`wNUA1?024{iM4qgE{bYiu21i-El>2RLhiOQ5JQD+{=$
ziA<o=dtT9VK&W$#k!Skp^%H-tHMCkphhVEcVzjUMS@$`eGy;pf?KmoSW7(kIZg~j*
zXoj=HBb`HG;zfYRwuKmS%}IwrPlJ#IJ4LD#ddjPj71{1ff|J<yzr5>&&pEIOQg=cE
zH#=ZQE_6a(Ac}$}1Npxvp98&BmTV>?sA-X}0j`oFDx3u(3^?7n%J%{{N#P$8egSYX
z2Wsy%q5Hy71SFOX?64jHdoL2G9MLgxmi!bsNW<;X91nQP@eg>6+W=4=4|c<@20+4S
zDkt<}1YKoracgJRBV@rGTEQ6X8ko}mT=U$sYkQ?NZ+0Xf<SC<|v~a<t7S)A5ZCQ`S
z{BF?(t5wxOZEVu)xxh__!=3Q)WgU7Hs?^*Yd%wPQZ25(#=HY8S!<iQYb6%fp?Jcfr
zf6=52)3D_Q71!I#(wQ*U<LN(lg=tZNn`qKakr&_2xA}rFM3aKR>vC~J^&mLj2QQ_N
zNDu%oOb!5XAA%S->4dx!Jk`5sC(un@ONiiEZw8OO!}v~Rg}r1^laVpRLj>9U{#2Gu
zk$9bqCt4Qq1!Uafa>nptt|p2mzyTZ898Q5|PHA36iBvqP62Eqs^l;{{0M)S^F;k1#
zHtm#z=YZXxNePqx?Bykfw6J!x^19PvWnAyH!Yz0Oeqf2Cl%u+w(CNV8?UrPXqF$er
zaF0W34}A`a>%zR)qE^NVYC*Dsnr`(2kphKGve?0Z<ubcc1FE+;AiF??6ED?Eys4mA
zJr0|Gq8Xg6K(;L)?DGL2APQVwl{ZZ)*XHm+N8<+LQswXVY)WTHH!Dv$Z`rP!HljM7
zfToH9&^D;qP@pKr3}#(Be8`w!v-T%}^gM`u=u*cW5u0g5Ff{Lao>+pW{$<oU$<*=@
z_I|VntDqOKF1HpyUsT~%HihK=dTe+vz-zY&JcRUAvi6v1)UR*}F;lRgc>cm9{y&=D
zJ)Y_R|Nq~~nVdE`C3D`Qj7=0ZhcPqEHp<fLg-Nzb4jrWQ+H#0B%rRDJ*v3dv6dhhK
z$w;L%<JCdEB&p~euU;KqulMh<KHuLTT`s*Y{qfp(J|55e;dZ;;sDR+eORcs}Tu-nq
zJ4kcr7In8&5*5jO2g_fg6_P8jqM9eG0SA#q#CM`>0Bd%N1JJ7;pbfJA6=ij3+;SP`
zI}gb#oRz1K!aGx|G9*Fokm*Rs2>_jgfzKg9cz|a;QhnkT<gd_g5P$#&$bB0n;DfIn
zg8>vcP7hFuAzF&HV^DC@dz$aHQ9$a!s{kCD|Np`Nzw-dm?1&)s;K_Gepu&V>XnFgS
zl~X+l`^~+(!nf-9HZ^F1<l|(ezy{Leu6gdIDedaG!7w-D9=dateuH_57R>){=+x{u
zDLoO4L1Ku;flp77*-hvOO0NC$g-7j8^{EE`J}27DH`3f*X=5hplLbJh*ZhnVK38Y^
zU;Ek<d;0B`^Ed6OySU|eK*GfB<hU2}G4a)YoY^mOg!_TN&)Br=IHuCc<}(XMdqMp)
zv<^!Cf!T1R?vD2QlcyQvG<!<!*)&)}B6Am+Z3NL&7zR<`hSbkM)+wc~J?LUE1D7Ls
zj7gs5XjLRz2}6&1crFJHx<WTRf0_14BL~qmr(b=(r1?;6A-DEeh+x7#5vvWZ6tZ@0
zYihNdKk)6m5uT_Q3W@$q!dOkk-6tbjZqs?3F0H7XkS{2A31~b^P1ZgoaM9IZjI!FV
z<s8aA?flVIk+!x>;qjF&T(fogW8)x5XmFdeW7lh1mEhGDOG4|BQw^A6q=hf4WBW4Y
zY}Sr-<-wF4^iP7Ev+PD=Z7CyMn1f%!7Xd$+5xN}tkq&YZ+@J%sJb1VO0<z?I<Df+L
z&fmPjWqfk;?|)i;9Xs?C5Mtiqx5xbAx{8v!iF4B)XwqRPwcxhe7;Y#UL?Xn-KfoGx
z`NFOkEswH@`+CUg*V>J$u8y8k&RMy{SMrC9-PNj^t8$Tb6{0+ZJD4dTZ}}al7>4GA
z;Qtk&DJvRZnM|uSIVut$0m{H@Hmf|aTZDix@ljCXF@p5Hsagx+!?}>4r$-7AVr_~x
zhA6l9UBT6J8iv*M7u0ey7-gJJSZ1UL9oHv0sh^BaovW-H+^!5#-A{DzN)?6E?{9)T
zyVgfK1mBT;#t8jyt^IyKN$>a4Mc!H(3+(3pk7f$4bemRacKBtH_B>GH{;YiH>2iBb
z1cf=?LV|b#J}`)^05LBv^APV)&I65P2L!<I^~5aj-=F~W@C1?w2!}v_`+Gv}#Eq<!
zMT|9Nsvi_OPzUvG`cO%n6jVFTV76~qvx7A?hx|q>xDAsDpYi1*{&#CHX+|0nL(O;8
z(iS292XUgDx_9R{S)@#PIaE{z>6c!&uI|wiV8UFHh9@&=QxPvh-W^<qk@#J;a)AT5
zE;9VZQ0scb*ob6@4b@Bwe?xtcY{6^a>sqtJ@yn}KYbfQTSEc~fEhzQ5;WF;E5xt#n
z^4zR=y01xII&PzsW|-)DDKOj_t)W16X?FJ*9GKXz1FI+;anSz*=6WV^860WOr|ODc
zl*=4#tEj;s0Sl=iwpaq!zqFC$YaMK7L^(@xIad79dHtQ%u6@f#WuM(T>ygKyVXVf?
z_2K>gaiS98>?X+gQP$X<!<d00*nY+1calZ$jGI@{*&w;Bc%`_@cnVoh-k+N>s}^=y
z&10esBgrEU>yM*smZh1p<^x_=sN>ilCa$iyUp>Yf=eOHmNWCJ{QT0@Iwl0Cz-Sc%6
zs-*Q}G=rQX;LLMym6(<ebUDgi0i8Js9PU4_)dD&Lstf2zhA#D9CFnir1<A)sP9jHM
z+<E=)q(lAI*H3Keamo67ar&6RM&Uk=o)jJd0K2I#kM7*xcUhF-3#OOr=1soEp;kP+
zimGsS8Q9^nxoO?jItAOPTZOeEv-OKXDi5Zp4ut1q{=d7DrylertltZ$omX0+`|0^)
zEsY9QqM8Hld7^=sg|Gtn-walwV<*ahj1e;Z=30UUMBmO8T%VK6w3fph(RluT8sbDy
z8P_LwEzIGebOpCyy>II&1=I>ZR9R<oWv+=$WrHy}fX!?6(CHE_BT!F)nu@quL;I+c
zAV6lACVGjCklu9zZ3`&406<MMPSe1g0Iy9`g@IR(=9Tla!ya3%$vFD|{}=+52beH=
zS`W2OU7S-U=++KuS6;;F=ScJ{fL{}@DPzu;SjbpJN=*wlTelJJ;X#(S=8tlAuuZ_1
z6YG9S8N&yN6mn?aM08ZkYqGBFH?O$e4CPMNkOO~{?GnMu+C`>t$A(zR5yG4^Pgs$i
z*kxW>U?N?6+_ds}=BtyF6K2T*O_5J|Lj3Pl>(n?gWrul|D1W>vZZR!r%Ymcq;x3Qd
zvvB58DC|D(8{Ni}H&bmm7LFCtRcvdZg)MNw0oxb`j2OtxASr+-^Vcu9<>=~iimDY@
z6+Pqz_0nM#o^TPHxIq|3I9<#D0B`k5aX$A2+7KY2V)KE5QDD2QdwQk_ul%fKm$(E}
z1G?K5<yK`FDOF0m3tmRt`zOpAnql>h3KF`=i_1Cu;Tzq}QIYbBtYC*A4o#%uib3px
zEy-X~n$mT;K~^O^D5xk1(>~?)o>+Wd_xh~$?aN1RXY-XiA}jDj42hePw)yPx4NH=q
z(Y1<ktwwu1ND=}Rvw4nTMI>1+8U~soLlIEgKo@|LAV)Jwg1E`ykvIk4fVr*IILS8k
zY$E}~TN-7YeRasy_KW`B7<&u{QmBxE>OBGx=|5t&#1{BiPn_zIt}?`!8ERR!-#>>M
zaoxAKN=pV>EN?*&ck`gUDn7irh!8qZX?cU{P-o~MR`(}KU5a3!KuOa9HV4@9y>47k
z2~4XYZ&dHV)8uQdQF0PBv|1}b!6$oabIP6JtFG2Sl0We<5|1*iHTB>zI5qN$gUX8}
zNR}xnQn%T${?Zp#({s%WK_<z^WsuOJW;Ru~zD>su`2FxVh<XB!90oL^n$a#)rp*IV
zY7kxzRBBo`WiujeHznMtNl1ew3u$N^c)x+*tDh}-jD`pUAf3RaL;+m}9I(sKnubFI
zu2`KT0iP6{MF>Z22%3+w;6j%RQ}3L_lZePtX@?<?p@fioGzI$r8oPO!mal$o3|4&%
zWJ-6RDSvlwvcRzS8fJkRIJX!#geE~CDKqB=8%^}J`Ns*rgg!VE7WefxeWKUhHD1rc
zl14gyU(dWLMKY^BqU|7Z*zyE0{lQ#)Tr?cmFBeIFRn2^Qd#T~vXoB>*Rs=+wxJ5EM
zhRZ&AnpHe~3FtlQY?p4>92;^xHOk+LH2s=^{idH>Jd<<gU+SFU!7-n0;2D5{g{3ud
zt(9!Z4d#P?lnfSK`aM|HHZbkarWJW#+lcJ1NOKXHu4g({F$*^0RhUN2gWd=-iYfwu
zrWY3qhOTZ9U~mLYhjlYItq4#sV(hP~kAU>6Fl3S6;SVi+cM5X#Ny{h)t!wF)HwZ}^
zC)P~LN}&=>9`rJ7{68@uZh#PNg>qY8D>AvH-CU;=#hVBVia+5_vMDPPLr3?xHde*q
z-$G3dCwlxh`F~mG*a;$0)fPwL(c01!0XG(3XQpV$C#M{v!|6i-oCUIuV6QGnHw0t6
z;FwN1f{I;^e014#c!%pc=^r3VEcS_Up>6g*bur>561c`7Qi)bax$&0tinF@YhnnQ+
zH!pKPtt&oC*o70!_Nh1eXGTf_$=ofPxTi^XDF8g`vW+|^bOBsHQA2xlh5?dgHu0!N
z>){<`MAb5i(&hm|tQs<{?vuj6Vjq1AqvTmhQldRVRO2u`XUJZ~E|a%jcK^CTQ%`gH
zTML!Tu{Z9Q_$eljY<{i`8LSF2!S2Tcil0yoH6_>pta=02Cu$M!p;Z{+Zc41<zDQ4(
za5nXn;BZs(GT8+_@eHe&_}2m|b5RjlQ+=|X{fbj^9)qv?AH+RX=fOn0``P6Ag7TPA
zBUp-mnWahIYmT9J1UU(O-~`n>(6}D=1Jk~U3ThxQ-=a0A02k>XhISVLLJLkaI|EBk
z@FFw<sAX{7YohI-oHWQ?Og33d`xH^%4G(3#_~#L}v7@=Gx@e3$0*yxMR$%vlW!$lg
z3nw_ntWq$zIU;3=oLGM(9{vC_zy}(zCH^yl2ZI<l*p5B!Ux#65_ep6ZQI3lQO5E+l
zJxJ*O2sr!Fp_y*2VC&_0x?2o=QLc6Hxaq)XLneQ-y_KfGXO?UySr?Qt_rQ@o&h`r0
zTjBp8-I+xF9F{C^Z58+JlV0G-zsz*m(m9)PRs59ARIVs~SW5q6&J%;&ZcPi|=;rhz
zkBSj)tgr<l&|NT0z)r@O*@le|z7PWjvC7!@(NcJ!BhAYSa|&DG9+c)UXw+{yEuf}l
zsd=fg+JaT&IMLu^5LydNFh*9c;v6L8Lmv_a;{(=kBtF<o%q>pt*uc|2>{%^XVR>6W
z^To)L!g8_2u3KP|&1fAkRCh_zyC)XN?BnkRU72_Mr2Qq!Oqi_?2<}PPFDTIPn6mcT
ze~-<RdII?WqW2C)kex+~Q9Sw4`ejDZM!-pNy(J$!^N|#}XoRV5<3Jt672M4uS^->Q
zP78yG1i(h3F4sqnh}5M0ZnrsO$gTVj6^(kbqRj6Hl^R2>09~z{65@ss=`gkyorYZO
zArI61zD)xD)8Z!jCm-zI+!yQg;0%6C)=mt#)<>#~Qbpn-oOh%PzXF8C40?)_0DD9e
zQC3-}c~lC2T4qjK*8!QQHNexZPBXcp2$bl+>DZ4|WZj0u9)pBbg~OSU@z?{aea7Ht
zW-8|+-&UCXyCXF8(Und7;5S-fHl=u1{F(K&$*l9eF;&4%f-v_S8EF`J5XBUPXHV!T
zRks8jJ|uwq3)SgCOLDI()b$L*`j(^wg9>7<`>INJDdXz%1>!r{T;ToUADo5Ogf$<l
zmLwskA2}B}mD<c>T^qDt-d?qc*X7F-e)K8r`(j9E*O$kSeQ7z$n4+hduTUZdz^U2?
zsu6EcbOGCet^~0PyvvLns*r>D`%E675YoZBLnJ^rBL|&5ATY3isRuV9pie`ZigroF
z`(V{A*(vJKo>1y8XRpngLs5gx8RFoY=_9}02UvVYqvHzf=CLJ(7J0<t>_b~}p3*aw
z*CJ>#_En;~iUId1_laSqVcFI3?<%)*8Pgf-V?l{itWq6)_t(Ov1wh@x0Es0>Sx#Yl
z_k{bjQqIYgzfbqQI(@et_-Mpf3Uoxln>D6D%7Q1)@8#qp3v<`FDD-N(Qy%WTSN=Qq
z*_X!Botwmaf6;MirM^Xv5Y}4hP*@0rN-~uOiyHv>#F4teMn0{9;bLuavy1E%cJmFq
z3S1QrXZU2Du`r}z>7Rna-nNkQUI%14f-bWbN)G5nVX`Q&7CC?%z*5YS$#q5IBsmN^
zN@)@q%grO<j$Gk{qYns;l&^vb@(hFcQA#lV`Nhc14|JVP`Q%1sZ25EyeW>UU7I(jd
zy+1ina0Pc~WlT8#7spRbp1RX6iWF9sMowN=Wa-d;z<IOqV_e;tsAfwARI#nGf9@n0
zCK(%nA&#(uC5w}Kr4X<uRX+y`!V1oeR3|@LZZDPdKuM5@P^W-6UapRuH@h_ZSYiI}
zw9)oWdp005)&JDdV@))2Gssret{>lD+`QFtX9viQipza4-_&i_!9bre3ZG~rruD)J
zo;|^qQz~;ZViXMmaW$gO0giMWiuuXp{<)g_Ss6>lVEL5)I}p%piV@CypzrGmnfrNR
zGQV1u5xMC=(75%gaNia42W^&L-tDvVOmNFJd`IgELfx4(Q*B7w-LC|<lr4%^-b4#z
zp%uRHi*=B#gi+Gp$QxCw)cvKUp+j{==HX(53+GHG6JbU%$&Pgr-OOvNG`VA)3lyd+
z4j^9ylObo_9&jK#@eT2d5`<eAH>dy*mgGK*lAvg=^i4$Y<k%&4%oHoa<f(L~U@*+K
z%9z10IJzRPN`N7Y63V8>2VWhWX;LQSBpwtbsN7iKU~qn})*X1}$RXsQk4RHdLIqTs
z3lh9O!htuA28*G>YI#854Ov4Gq~28G4bP~DI!x=Tt4ETez@gx(*<TKPQM%x{kBmB@
zOeunTV@!1rY@G%CN)J|#Z%y7SZH^$L-Tt$H1>P7RTI%gtknw(h{xGF1rqi6<ye`g3
zy=O4(wbR$EJg$a|Wf15q#X;Xkhe5nO@9N~TJ>P@8!|;w3K5MWK%qRNV4w_wBO;=ZM
z*m$k#(W%qNr^XPwMqWx55`MfKy}Ua+Qt!mAkAdl3XMIuhEk)_e`QhhP_Si_m@fDHw
zR-KmuBRbbS^&dAGz=i*@8^$+sk6a5(?=G0)AdVK?rI;}?Clh*$DZ%;~U;~Aj&ZK*p
z({5t#yBAf*7T1TN>`{i&)tHf!GqTzeg_JpUMNdGD$s#|h1D*`qV+?38v;%Y>UM})j
zM<3)kHK@bGP%kED6a$DFkW6vxTm`v?ge-E;<7W5&3ittQH~L^qmLv#&AMWcpbsh#+
z<2Y^lo}M4hxA&7up^+F-<|DCX{{iKCJhD)`!CF>oAkN{|3hb<<p>tJMNyaEIvB-r%
zS18y{tXwtUzC}OTLdYQEHL^o<Dw}w7G-Pqi_Fi%5>alMfDSz(TSZG_b!NnPsnhZ9X
zYq(GwvEi1)1JQ9{q$HSad+&H&+)Z1wRo{-cpcKipWjvJzg$z1HLruUK%xh6KL0k*c
z%6_I=P&g9p`#=`1$^3+@22=6id`3PP7q$pIL;wDv1z+KGH8WOPD<;NPQe@c~rqP7V
zO*Z>nRQSL9L<v)~VLw(aY<V|-tS1AaIC15?Gb-hfdYQT?Bw>3;Lg8T=fIY;$!3d_t
zyaov53=YopcZ(*#p7kME;Ro@ch%u>eb*@_juVOm%#T}0HOTzbyNN?%VAfC=K8TJP8
z?D~;=T>o-zU?%C09{(*2l=8b9K2MZOv%OI$S(I5ctN01xRhc44dt~A{)3}uboZ>s{
zvJ^uYBi^DE*U^G`OZ&b4br5X%_{lvk%|3tXjfq!ZKC&}%e9cqJhVe4}ld2cLD>26A
zjqe3H8M!8$&C(lt-N5adP4YAW)RdwMG6~`%NYuPt0BAraN9#(J0F6K#e5tCoMyw|g
zu3i0NXkl6Z`w{Ub?KeBaoBlmbsXg%cOaGy^^_zT$3HzMaSM+!;0lS2llG53tkD%le
z8o4K)Utf|AaYJFPsL<KuXV<<;hwuy~c|<E3zN_MJH=Mv6SH|d^q)!f^H@UO>w+H-#
zY`8&cI-uNiIZshGSKw7LS4YBGuiNykZ$r-^W<-{#w>kZ}kiR6w^?qqREB!QmeO^6g
zrCYRx_s3Jhm|HAOhluI2pZen|l0W`^K-xsIKd0lT@Pd6NqE<T(Qrixv&a?({_ow@d
zBOkOc_iym?!xy4Rq+93;^?q2BdjKfBI9xUX=jAo1;vJmOrMDmlGzBV~T2a(c4PAFq
zKb_-sbXj%10vdOks08!-Dr+m(j*uG;3cZ?}*Bf&uOw~Qsv#F;q%EV4Kk-mnW1f$xL
z{+T*bIN+w>^(rpm1wG`Dw(jRdLr1&{n52SIw%PQ`YX~8D{6}h=Q%sbt*wSI!_g<;O
z5;rh&GM`VDhM#1FOTEZLWaZ|=@nJm?R*06D(x^qq>70pY|4H)PU~QkgFrW~#Xb8(C
zx^U*#nG%ObcAgiOSAP?<&ENNo*LfakWbjx3y)LC)QCwFTEi<CGGx2oy(4{1njPE6`
ziF84<aY+P6aD$AYXKKYzDRU==cS8yEYVzhP8695#QEybu{#Sg<;R^c0NpK`xFxZ<(
zB$#0gaYg9q<u%*QzNbv|dnYh*Pu+iL3`c&Xd68PwohCqMpIR##60>QbD7-LJ2S%Lg
zelU{<i&-})K-!sU3Yvazuwu}>r{K0%E8?t)U?+y!R@pY<fnSeu?bSENbN(*R%lw^q
zNCudi$4FT=>&$KmP@$05v<p?68!NmgINOGE*=8fu&)=8zw9YzA)B0QAj_OXSU>OgP
zzi%@{q=Mx2I@2mDSROI99uZbiN!zRY2F65{S>PJ<a3?Yb_vk+#jCB|$@@J+QwQINt
zvO|k*A;hh$!4V7{aZx5!V>AcSMc1=h{#%yK9TQe0F<)WShQg-fRw(<@unz*FHsM`U
zT_WfSKt46$%yx+~^lxk!g3&9um-GhPEzY(5@rO)zbzC0qp#GU}J1{r#3DTjwTGG3;
zPuo^tv<6d7h3=bcHgA|GZTRKJoUQXx%K_dO>DW7)qa9YrBD4}*`c}yLdlU`?Ud4}k
z@U&W)>0|w?`xyGr<rrPO*063|^<eL7;#p5O9R$7RR|@3tNq)}><~A(@VL9ua)2fxX
z&9LV>l{;)UmmrrMQpUHhd?S8F?w1Y~15$gPkyPomSwBba!0ZU{KDAgkY8t!d_!mK*
z!ZTzQ-GexSWLA8*u|VQY2GjMMqJ3o@!oa{}kZ{BDz1nT?nw*wl<-d65@$n~UiIowd
zjKlwpvhhmQ!lwF*3W9M~Ma0|e?MF6ZJB4qWGw1k%_{*uSPj7C19^fnI*EHYUS=Q9z
z<u^b4xv`kq^^TKqhxig-wBzyWLE4i-^Mv%T^eT*l(@|HC5k*FO{Mvo-wpDFHSC=;L
zD-AhvQQZK<&tIJ6G*RqRMc&lFDfSWIsGvmmlxVMw!g^CHaM4lKkpI-s@X}@Y>LS0~
z@sH%>JuR4F7Q|8=uhi8&nJrFSh$~tB_~h`SVj}y|g(^bvhkSokx1X)`vEzv`pH!xl
z=fG`0BR01ZWTEjRuA4{KRC5yAA%gPN#|n$iCpTt5T-sL8h9hVFxVyM1C?~c@Ff6^g
zuGKB+MccMHeq$t|oJ)*I*K<VS|M=}3rBd+bVLE4IlBZjKC$DWv+R+6+`v`Kx(-2bD
z-K?sz>}r_svto%HUu7;tGZu)qYu^sp)C8pUUJ4Jr0TfL5i`yLC^UdoXGz4(I-uk;e
z(eol^tqLSEGUvHY_|1YdvecM$YDGBTtwm5#sSmd3N?+@4H0KPaxI{@*7va_EIda}j
z_2T1W!)8H}X?90Sr2MatjLExs{iAyaxRZYKflpto-4uM3-ZI6=X4L_eEu(3xs_9!N
zvTpJvE8o&CAGJ>KY&+K$vj7H7r^0#FK-5z!8gvwtwh}{T+9E|pG}K@(!M^qsFDHX@
zU7-tRZJOl-uey3rBaL$Tf0EGnqg?d%`7qSh7cV=!DiD=$=2@3vct-;5MAwQl+B46X
zi}wiko6g{{p%6%n4`nO)$>$jF8A*y+gg_o~&r^V|{EgDucw*m}hZGw4wVXk%`NZ9y
zyvk4B|6S_9ZyIq$K2d|fDh}7As8(J(qi~2f7%f{~74vqxAAuZlKQOM+onfBRzQR8&
z&~XBDJq>#2v@^D274Jj$G^kVIg8<|`X$&k-Nip<Yi^v2mrKQx<K^%tHD-jAZ0A>%o
z7r;acw-AXn<|0BrDBi{pF3H_s7MGS-!hRhtWOuC&w`fPmZ*^{?5aLddLHclw%(Ktx
z0xLL8hhl9lx&eGUCw6yOU}p>t(eWV}*XFC7E5meS=t$`*Atpc64e~fb$S^ntl!K<1
z_sh#T(Cjekg2l_Rs&AV%;cmxuIfR%CAKnR|{SShDW|@`74hdmc^Gh8n>)pZF%*>hd
zdb{ECM^(Z2!ob`=t}xyQ!_JQ@i6hUf3<Fw$AH(zxH?J*rf$4Y`48qNn!h<hIfR?lB
zgRc}Hw(7vX7njZB!y3|{m<}rG34Uhnw~676XvM*5K}@NN|LANgzjPnOTo||WuGK%t
zvAVX5b^4iHNaT<#KU1`lQj7m}>TYHOX1?(Np(?dC^f+7qxbA;9LdFYc)w|pRuxDJn
zJA5IaiX9M>@ndy7)w`KG&2MQG2J6$71aKO)0s`D%37O&Hr3PgfjNzW&4vIx8%SXhi
z8J_JALGdDRer4gPuYfE;{3z%=yQ>4W5PxS9afLIT8K0mo!_Nm88P+#Vhcb`@eewb&
z`7^_5v-E)z#q891X`{Ysmn4L4sNcLEA_sc0Tg=o;UA)-3O)sXno5!Aq(`jve3g#hs
z5l+bI=Jjwr2gNIEcXx!7_F3ZpxohNUd-~wV_Wg)8c50sBL2`gxoWmv~bd2L@FE}`1
zMV5Ea$r;gYPao{<LnFg6;ibrP&ptbT-qd>LpV=wzzZy}TCvZEi*8{xPh$CT}HW#qk
zV`8h%2wcGq4y=U~7FE!h1}vDU2=1GCpy%pLd)wYC(M|s9-Ksis*6AmwJ2vZ})~YSb
zwXQxl4=M{#Zr_^29hx>b@$CBV^-NE}6!iu2j?MdJzdU-;3`<%0WZ`Sv`c<n?iGeFh
z_lAZyH^g)@CwE+ahyHJ@#iMY=U4<XP6E(1}tff_oI8rNr4$@I!Gz>&u_r#4DG}C^o
zPg|jTr<YzJF4?}+LBDLCGD4u9|E@|);lF-%fd+dA0`4^H8fQS2GzYz>;h6C#aqRWB
z=diX5`|<a8B%k@rE^Lj7@QK`U-n&Av&9Zd=L^C1ffrHoA)|BHYH~p-eTaT=9kAifZ
zu-Vpq)Kxolp8K>|hde7jI~opU9SBm)NmVnety5p+Aom#5&aP;4z>(SEa-7^@ZC+a>
zaqC|XVx1yVx|9oXbZpAS%^^yPBG8G?xz2%1cfSaV^mN7@lExt8X^~`ecS^uKvAG9#
zOrWK!1z4XnH;59Aggo-j2gZLXV6H$jF*xFpN=<WF%HlbsY1z@s)IJD$SXmgQ_$)_a
z^FokRcC0w5td1>*TS4URd82To-0d+nm@=+E!!wKl@d)79Iui5}Z0WGM8n5fW*_MyD
zj1$xmBR4x7A{-rY>l&v2G_b$X`=zmfY{^U&V?bKyeQ;juO3VeS@!uwKe@nfod*oY<
zCSN`lQ}wx+feZ6q6XN&Bcel!)?=U%N%Yakhb;lntXX@%-GHd!u+Sw*$wx_R7sau?X
zuP$Q#QPs5U3HKDl2L#q7jQ{sMk2}7y8erFvy0F@Jxy+ZAmOs+0etFi}bNWle=|HbU
zHzm59{3W)x`G;0WeN4!7p)Dh&YWFX|07zUu`vG%j{+KeB*ZS^mo&wTlpCeJhe3YJD
zZ9#@7>6#Ne2Qr22UvOREz*CCZ)B$s`75ISgy*veASiGKW0DV|0*NcqT71;{d`iY(@
z)7XraW}h<u>_{7WEV5fVIcgt;>7V{`CNfkpC=B3k?UUk4SK|M|9^);QNexK-IV^n@
zRGJLFOYEsmy)%8Mv(2}R<63V<!aKMr;nwNC?VAU9*myF9nkto&OYY&nlF4Q3#KGua
zPjny-sc{z5mI)2jBQ|HTw40VMK};T15BA(@e(2O**Y{lw{WRRdr^Obs<i~(ixA-N&
z%MODpbHyvaa-i6?{;@YV+jZBIcjl*ueGuQMK9GI+`^PAolJdA)a}`CAX%1v1t2i!;
zP?Eat{HYxl;H4{3SuTOo3nSryl0oSo+eDK3$Lg|U4`{(d@FPZAW~_=Ar#qtQU!cLs
zvbca!dJ1la`tLLGJvF{PVtW5@(VH9jo2-1RocCmj%Sy6S2G6~BTteB^!}B`gi+nzB
zUAC#DvQWAwMo_oOO69AcJJz2AGB>NOi9UQF)G>kp>PrXRl%8_~gNVb=Cs#eXC?63`
zR0~eIZ0Kn&!G^r{;l>_++Ld;tQV(-|o~Xf}Ks8B^8YFY7cvPSrqKZ9`>e1g$Zt>8a
z(DU<G-<NNOK0T3=&Uve6@+*bcz_qqE3-`npeC8PwVp#dxM(w6fKB27p_YPP*s}8u;
zo*md?a)D{KL5CIW{o%|02*huL2R0nJOb9uGD%FMw84V?92W^5mfqWCJ&5i>c6~17o
z&||Wt4BKRt<fyohsT_&r5_{}@J=OyaSJ0Dn%y~W-{AEZ3h2_CSvl#(|<pu>&%a$^b
zU!uu)<XG7)^0H0}SegtXBWH*Ukm&&gE<$cfwjA(|vHUtyX9VU>92SHFQY?6m535GC
zIK@dU&MzF^6oWDjy^@!vwATy{^#ts*pxfbPZw)=M@sT^T>UebZ_%U<$*1M#PKZAFv
z+h4u>I^kIo-{u`;IUX)(+N>Wl$v5lxbv(+|5kzMZ3e8r(syj$e+%CvywaV)MP1>;>
zbMeLSHAq5_UqNh*+~vT;2HhVcH9NA+w?pH^%O>18hPP^N-yD4pP|^5t7H{?D-`aO~
zu>1v;74B(m{M?U`uhOhSzxMS%@7YvxKt4!aPH@jGr>f73KFuG6Bh<y=bB(n}EtL4M
z+NmGniu~W7*2LCbh(+TtHv}o=aQx&9zoLQTiK!%rPf%gi=yWZZbAPNd<bWLQqbiKK
zZ4Ig43K=LLRQb8cgy{gK{m-}Y5sB`|LZk~%%U9-<%(GG=`zz6(V|zFJ{isAt>}^Iq
zOL(J38oey*u}2zi?cOQT(Q3AoXem7)gu*~{zr5hi<B`L3E!x?-RgMk2W87j!<F#&@
z8}zEGs^V*pG#q;>8mTxqabSykn&m4V9$Jg%^^|tS?gJ7%B4gTe=cPcZUhJhP%S9Bk
z_Q2Y`qxP%QZFn2%6^t4OnJ~%H7r#auSic-I$<Sp+@udo{3da>U1JAV2-q88+dN?D6
zayVKRkg+3_64L6|ZYsDH+T||WOu5;#SD^(7;F7?^3Av<z1xBILYS-h7T0w51Go%=c
z!63tT4;&`=;v1&j6w8tK-9-?N{zO702I(cb;n4N+{FVbh2A^%|q3?^c-sBurV}Ish
zkf4GCHp!_qMOI+w1vCsrJO^%RTS3r5{FjU(hBD!orql>UsQs7?YWy;8o_vns)Uaem
zQdXAa&2OtBoFG+d%Z%u-cd}6ss+7tz5to9z$j(+lHnU(D8n=sP<^6zm-D2C$iR}c(
zRkx<B+jkg0dLOiQ%2qtNj!P{o=@g)HAc^{g;`44Lf5BR}Milx_gvBiH-i*N699`*u
zDxmP!1Kt!?XNWu3<uy5Vm-6W(7A$Iam2g&ku@&DB3-$xL5yy7jYV)iVR!v+LjEL!{
zL;F)@<SSxGonZ!8RYNDsDp%%0L>j)j9IDwh{xl;NRq|}UY_aS9pc%uOg9WZ8R6%p>
zR*~7{)=>!jX+^V+%n7mDXQ{RL9;;FBBAC!7v67HuMHP{nb;b$J1{{cylf@)M-J91t
zBDJ%Qd+7Q5wwQYF3bp*xs%`a?prrrqw<nq9GmPMtX>HCWU}#&xealx4jCm(yM-K(s
zb=O(F@#(Uw*5}uriCRrR!!<a@%OgR;Q?sqUO>_k%`_FabA6%%{{TtO+viIAJ<Z9LJ
z{1uP*7reoYxIl`BX7&H~_P?H8+L7VpDjQB&KWvC!@BRC0<DPJ{kHMzvRs&*d<xV3-
zzXD@enw6D1A)98NAA1(LUrE08?x7sYa<BwjIpRi}O|F8o@API03<F0owhWWeKKwJg
zb>c)f;ibTT#&4ft>w6s(BSP+pg_l6{j<<#|ciM4-vfN<)Vn7{X{+K|ocx7e7(3A4g
z=V{qgEuiOsKle?F8@_eDFr?HWAr8Sf${L)|-#BS3iy*UkGsgjU3rEz_t~2dw+rJv(
zKSAQU+^Ij2&5Tt6T0@OLZuTV-=d?NXQ6o{#7gP?ny(9R-dt7i^UI@QWxW9C$gfT5y
zhj!FjN&9Qr1+exiv578EEK1B|LQke|m_SB6<vF<<XuW3)Jpc!;rk`N^TW&=q<}Ky<
zDza|HwS%#}_ik4NTRYt)(XXz==*ZmO`7nmGF(1VfB^FHPh?4CXvbnrtGvyADQ8K@c
zA&yAFKz0fLfdd7~5U!}MPbP;Oh0N4BT*=_BovR3Ldd8)xJvA`_1rg4hloVUj(HN(6
zfDQflY0V_RNvlO&?WTar5;d@LXM25bpq(DWNtuz%f{=+{VsEw)?_B8x<QB;d?aPgA
zo?;|{+!0ygMV!E1OTFKg((ddZka|D(2E9npc{Clws2YL=42+JF!=q7IQ#+5kUh_2l
zi(o*2W44C#`I9nL;MPxvQS$qa`_imR7RbLkoc9n#x@gfLQ%{^Crao_i#s5CcE&PR-
z$UHc&A9%U#Y4nT1O!Z>ps*-yC{_MCj+K+{~f(Lhp20Eziv=Qf+Xqc3(^o8Zp98$LS
z*h?A*Ky5f`^k58LxK5<y0c3U19H~YTL!l5I4a+mha^A{tVjW!*R0+z`)qt!%ik5&w
z^?-&3Are~X(!}<WS+&PxUELIeEYa2+C!lm8(ZIn<0<K>N;4q`o5IeCEYjImc|MeUY
zYwyk5;&?6d+oQMY7eD4lLbouCw2Y)}&V_E05CL{nF?cKcsc7_i6bsV4ZYEWVg1cVy
z?TzmEMIAQ9h3qPMR+xZPnZ9J#!Y427dT!beS?mysPAlUtsdk^_-%r!tH0o1#1f+Jq
zEuY$*y{XyOj7tgX8jS;U+w!~o3%J14_Oz=9Y1j2fD}*0omI#in!eqI2V{M@;tT8(X
zW-)u8NnWzF!oQdmK+5*7{vH+>lU!#UyP^7{`Nk+WQ-{a@>hwwXmaa1!UX&f=8~82F
zG`qA0vqWtC;OaCNu;<IYWQjM#JeS!AE$&qF%31mjHT|wAJwHhx_(nN0okvo(Iy(+}
zD_JHExsaZ9Qa#g|TAT&!&?AcqvJD4`U2LI4Gb4a%gUsO@9z|6hSe?+?f$1QhIFA9p
zc%$@0KjOV>TSyP;fkVaG=-Mcn@kp974_njMLT8(W-ZG6H$X24uoCEcZ33#>`LCSdO
z&rJDS&M%LuZ_K_TPV_CW`S%~`cZGneTbFdBoYL9R2QPb)#&&O<>M<8?;6g|{3DjjI
zw8-mic3DwNVL1g-Z5w;=dfjV}Td(M#s(5|>!;aAKT_A}%o;+dU*se#r<s3uruy<nR
zfcF@1ssOh{B&55q%_%r009PYLZM};@masPhU@y>zfc4*s%MEt*9tOX+|B`L_4TS%$
z>R2aIBEj9=UCjefY7GYgD9+@~T!|yMaEL2Hxlr@tH{o_V2_98>i&t&eZjfJ0CFgK+
zxJ7jK4JLm+o>OiNgaEkfipLiKA`j@5Mj`rZJ=>#*ftO^StGt<B8?^9-jp)yIJv*O2
z-o4abJLoZ-Ew<`h<rupC2_<aB{4T2v`|JLmj=K_av>kt^UcPtkyn9%NAe?aAzlh@%
z4VN)v+)cS@_l`UK(b(oR6<CA93PQ{PqXl})JI*wdIIIr@^1w`zh_qu0w6V5=z4jIW
z|B)S<W>OEf=WY<G)Kk95h3reiL#mVO&a8F=CenJT3(%I>!}*iV?9-Pqkbys*`pTCZ
zd`X)M*>)kEiU9El1&l~fOA=m?=qU)o6tfNKG&mV>GCJjUO?BRF7xVtUzp+g)cPQk<
zUibe&4v^yAhBITt`X;Javq2dQ*y^KAZ@^!1pjVd>@z032jt85)R5Ozsg$>JGqC3*&
zA>I0UMv%bd!~5)-&I;1P^pcj>XIA?DvLoT7q5gwW)0<BdsQ=XnNBc?wUYP@bB{)~6
zS;OxxO?E^27s%cCkMVz=yHd^|Ti#rLjWYa6iVxj_|EDOFgR5$75fr6~gj_3QZdz?v
z6dXV3E2ib(p>}`WJ!NhC(w2tkciV_-v10C#BcdJGcD7ozk|Tw#4poGZQa_jK)$7E`
z@KF~xlVkw|bG?iB4s#g%WVo>B2k?6(>@K1dm~jk%NEmfcU7T^?HSJ-O`<A7oxIpDL
zk5QAw+40APz6!Uj1}%R&FzO)Wmnez1_Gf*~6yJIjUjjnRjf8T;6PfR+JD2IWbbF_K
z2=3mUc`qs2J=!B8vnZR=Y%2!hmVe)fs{jdVhj3O<8mQyKvu!g-m8F`>(@bN$9<o){
z#&;%kbC9jZ8`OrV!;F&+Tt6EfZJvp(y36Q~;P4}^!W!Qd>lreuph)7)Fd~Uk`iNi9
zEF5S1k7*TcbaQCxopX$6y#=qRSfCG2b{ugBk$=bVq+%rqDNeaf2O?Z;Fm1^d=SXw}
z8FC(tqXx+>Q<+48Dwd<=HP(7KfpLR|zODd9h`%C!KL&Z`9Jv0SJq>rH7LW&g!S|NJ
zQY(vK_vu3#E^98sy3VvGXr@)yHg<jS4S2;S!0+&qMV6A~mvb$Rf$}gom=B7-`(zQp
zpAbPx$l+#iYma6=lxa&f{xk!tuS)V;F&mxDd}q4|5b7y0=Nv9FUKAU1v(bj^nDv>o
z`1<wfqRETL?H&h3XR~fUL#I<ijeM<@7eZ;*s@8Pdt)JLyJ9#YG%1a;;3Wz%%`Fu~=
z`S<m919Fi+L%m%Fb0I0oM(8r2P_(GlP>^(65Q4*LmO$msKSsg<@fFZ8Vt`vi>gf2I
zcs+{qK%&(TrnYGODUEmyF{P86K6CihsJ&^?hVR=S)us3jd~k!gI4#M?O~$YP4OE^Y
zkdJ}}I!_<o&l80gaFBy&xhaqaxxA+?J07tMIOA@hj?H8qt2^=iHq$AI0NC#~;Z!Py
zgac-r?xJ+!<Hq25Y|<pp{_4uTZN05*mx7tVrH&o4zqk+Hc$0Xt)J&g%wtwV~mE+Go
zO;6=sBCtDlrqb0ZgF7ri9&a8D<Rpls`kOxzt#M8n#4_9C8xL>?v)qr>1Yj%*YA!Z@
zi;Fn*>0}<psJ5A=07`LUk>zK)!&^(mHuUojU-5}roVbxaN_Q@9c98;<;A={5A`RB|
zvctDGh7ZCp8W#UiUji>7>h(|^)|v3Opq;UjkJO4ljll#M#K5W={1gR5XE#Z*fTfLH
z(zvNgp3j|J;8(<Z$98^?yCXdqJC%H)_C@0@&zzRE{>eaE=!uyT0q=u}n(m{e8@l!O
zd-mJt?)<EJ%e+<Fn$R8Ovy7E{=p_BKR^Pz9SGtJq2s|Ay@9>a74$7xvo`fG^|NZg$
zsQ$xNw}ezt*#OVhGe7ff?2D>9yIE;e;$N#L6v;uuWgOnz*L5GG;3<cI`6P1nME{C;
zy{H#c>&DQn6X}k<^u!XHZN*<blp9&L`%zk`VL0-c5<<-wq5V5Py2`NHR*pGWSwQnD
zo}T`qXtpF!K?-UL2s3b(QG4J&O$0eD9rkuT4l3y4vq+%ctTTw|fvviUy)ce3O;@g-
zPh9<N>muaJtG|f(2jt${6owhVz@ZK7%@&m+J(3|DuZb2fkP|!|xg|g*EI4OBqtL7W
zGxI|vJtlu^%>GcbsT};9F*IPj(x!o+_Y8ixs71SV8~T~zdRwS3!AioOhTQZHN8hkz
zSlL`cMFsb@HBn3K6MEaf_bM^4;PAH38<47COxwHPI(GGkWMGXc3NJdU4?nUp**qp$
zIK2LWn2Bl$3KszwJNdSF$Fl~9JY||71ovDl#_<E4j+`PV=L5~9EwE>Gr}9k8EUhyn
zCf!Zo@W4Q&s;S<ZwM7PkU$o$kzH!eRTi%UAv`+y4k&`|oNwjQn>PnWj4#<?KY*)Vn
zi3tI|NM35VRE{=6OOj11BJGO!2IMr$YL%Nwkr?FOY2Hcrab585nv}PY#r{`&*#Co=
zJ$o~B;VScwEBt?N$_uiGngK0I3bvbc0V06Qlo%a!eWLq!dosUbtMJC@;8;=wEW|fy
zsY7nXE9VA#ca*uHnlQyLVTCjuZeYI?sO;_>yY$(`wI`5RSIuz3l5=*7AvLHIUfIH#
zvx7_*9V;oz{%RF>I@b3nXxbl%0540B@tS^%XoOG;OGiUPxuT<deiYEB+{5?5tB9uc
zR*%ORS2C$YMSd0CdpczgGF>Zd*QhwDJ!C5DW1tddCIn0Y0Hy*9hDd<x1imUQD?A^^
znUAKbwz+IBcRcx#p*bxnCXfq9)(bA%2&*(<HNm|p%ws||k__69*JP6>flj7$;=n!R
zVCt^b&%!<uxLAbH?UmM+lqNmo-ZroB-WpqIEG&h1udI9YP9*TI`r3+f4$)ybj{~^D
zmj3y)Tftk(AT4hFBT_HfpztO4gLjN{XZBs}#F=1Mf%f&0g}4fzHP<4xIk_IM{p|E=
zoqkcgml0lz{^Jv7`iQUG0kfQQ(>4}lSOh;X-E{z>&WB7-ZbQ}WVhetw;fQAFX~B-N
zG|R3|e>=>|M&)`a`J;2B<@;c3SJtUzt9;T*RB{%-%0r`;7gPYtVk8*@s8nJIfM6u+
zvuO@`SZOs<aP}W|`0rxN#@E+h|D+Z}*D{Wai4^d|BeuYSqn@Y%YqNlQTpP&W;6NIk
z7_tNxr0E|wsEAMZoZS?;wo#}=_M~zI7;tX7!Ju+FGunuL3kYEN3J}zG94`QjoQLwh
z5s5SmJus2z#cc|9e=M{<DXzr{AAnSn+q+OPLCKHKz8*%&^Ar&o#nXA*Q?q?vNp6kj
zFJk7gB`zq3MTO<}{h!XFuIwS4HS%(bwp5T@(1G9s>gXqct&I`zm2`u4szW~u(6F0;
ze}e{%5D+b!VZ6>h4eY`YjP#8B1w=4gIXE8C_N67mYHs$2M68>B_A$xcdMopXz_?F-
zp>{s)v&kb$B9MB~>6S$zVHt+Uo69GH)?O3C#sm3lV;~`eSVXr@^i<RL$hGH+vlE}*
zqvh0OCzrmO2;UsGIv1EIelE%=fcaYnMCkD7sNFTS=k(JMb*UELVx1LiJ01J)lQStp
z8>;QS*<aIf?^zB!zd7I5)h|AEl;NtBJ5$-_4c*lvK*4}=P&^*aB;LmPnp%!ntuXGO
z6llHdaS(W4{A#og{ZwR@Pn-pFyB~Bh2@OP>=nCBJOp2CQ7+-9~LqhF1Hm)+9z62`R
zM+7@|Z?Pcm9!wmR;slCIk2Nd=u$_U*R`c5p6i^^6D$P^2K-9+Edayv|)FxFJBW7CS
zEQh9dU{=`-g_k}lnYgBn{UGeL^0&T4YC44Y1;Y?Bkz~vJr@2JC#P<~uy><oGkl>p$
zZFdhch?{y0=7j%ppRc6I9hQ%1E20fd$O}=c@6@kqst<19K!>AxE_)>l)q<95u#j7x
znE6>dkX~n|2VDTKh#x(UXj8K(Aya`&oE7hf4l#d$$o*&cSV^@Gw1^a~GgJLlqgzW%
z{}d<PYQXUERA>4bKY3HNNI?Ac*t-h)5~`lGe__-wV-qmEUS$;Dl%%ZO`#Jh9a~!6y
zN*<6$RpS6sT6JOPeRF53Tmq^!54aQZBUz`z-<s%8|02=Rr*Vq+aXN^CPqFcCB!_~j
zaOx8#E*QwVU`4Vg<e(x2o#%p_Cjvzv(AR>P_Mdig0NKDJQtQ<Hh036Tg8p;%H_zOz
z8LU4mNM$&Vq7<U;CeS7|iH1SMjMUdU1K51~L4cw1iNJtHPf6d;AS%7$@nI;JY%)9H
z1Gnna+)0vyQZiWocYB}AW9X>G*U0TRBmD3omTn55iC$_E)Z%PE4znJduNviVUK+-m
zI?RmoB^3kJ){w*CMF!`~z^jA?6Qm(Py?#UxY^w`Gp%B0<pb@}(f?E<723JVo@#_7p
zKc-IFyNI-HKh3&HJ**eC%CAAz-c9y|z#pBc%oXng0&G1J+)zb_g2Pq>&?8CHvoT@>
zfRP};>6oHhORPC^z9IfmBG3W&%6SjR0L}aQAAd<dLhjZ^H|w`oW^I?@6v8B_+y&XK
zo270iF7N((@Iio*BOvS8G0VC4UXdGxx#GtKbI#wGd#!>3FD{lJnwaLaWZM1@!iX1F
z_vPhFI_^9St*l3rU}k4IpLJhg&`?{pYY#kxffCRwdiU!1q5LbM7P+aSTX~UDf4KEG
zmg*Y?ghmEqjFo69ZG5t#1TG*0NApbGQ9Fyq>?$Hy5YM#+SqxLs=OKqP{93o32V(3c
z3B%+WaErM~v_L+B=4;ZRV1y4E5t%0H2QH=!?kD(Qyh0j6{B$<5uGuNdoP>u_W`%#R
z_P#j#@z=*U|MGsjh=}d3#TdVP68(JhEh8I@JmLsodDD)HIe|pe&b=?9pT%yg;D7xS
zOfXsHix)@UDia)$5751GFX5g>-FA1I^Q1T&cF8_&EH%6H$*pyUs`?XSakbvP<Bp~y
z(q$xz<(|*{P}peN486jOYNN1H=mcDdzh-1k+}F^y(Uh8U^Es@be|xP?9I~D?9jjaV
zX{w6G&h%#lIguXSxM)OuzUCxtC>naR9Zoo56~!LHuG)`P9*o^W^l-7+(9~eerhTz8
zJT}srpUJYfQLEckpwx_Oq1+{bqZ21dxy2<(QjSF<HLh}S5$8y>WTU`9XaotiZ#%CK
zwYn5<mdEwv3YLSQ_CjE_V~>5psusa2_!Kz^C>%g{L)T<Kfhjj|pLkk=oEQgg;cNcZ
zp&c9EBo*$CtiRJvG1{5z;Q<d4&^SiRw5yq2WilxpXz@9Q0LqO}x&~AA9A$Tn9L7eB
zQCq_Z<a&#aKP$OAr`29f{Kxv4Pur8l=Af0L4f@knmgbwfOAh3Wd_kH!G8syQftbdD
zNWf{)$&UiOMwL(nU7K7{ri+KXRrg6C)(K;tO?(}Y^KvemIOx^-KGM5P4f%_3;++C7
zbMo)f4FFNOXB^+-*l(#`yS@WAsrJm1NFY0_<#oo8`azom=n|S4JfL%p!GM~t?ucj%
zBw6wvX+o`4m|_kP#cG>IrdN;ISbGjN<V~p-BL0Y|&NZB<%+jt)1FIlvAPHuNSs+>@
z=2c?sdODX8L8OugO|+}#=rzOSv#V0!-=*J}4@~3UB)?jF7W-hPKxd=tyRoj<Wj*=c
z?<-CO?8x>BBYA;@`L)+J+R|{<|0=6f=0wJUcexajlbeDei|{ZI2h(xV@$5)=vW3zw
zD-wqK2L<($7a$82jiwjqjb9663&=0U#bfAGuQQ2-$m)KUo96BS-JO=Ew9*G_u_iu%
zIsH5+<RoIMQH`%e@A<Xf@^`lm7oAys&iDKEEgxF<%-mw6<xFn58vI%De4^3ui(u1I
z>z*st*i$1LyF0f0bMprlv%2H-Ki5%qcgbh>+&uK4X5x8fMV!Sa*JmfRk80cAJkVg`
zR}=P}7M*8i=z2FzXp=Zw5%}(xpm1NiBh615md2VY&wFO|UTGc00TnK68O9iC$Z_p*
zkcMym&2-eI%6gaCnNNRLeU^D|%odl?zZM-REDbpDmJBxXZ~Oc0x^=I157U7U;aULa
zJujX=ZAkOlvb#@w+LxZ!rX4HF$eZunRnFw~>AnA|ZnkM-SyHd2#DZAK*btpH_zysq
z+q7p+&bk#^T#t=aSOKpb5VwtM;xaXF6aZfU#TPgRh%v;P*llv{ys@wRR;)mum29XI
z1Up6baMEBzkPZ<T0lY}ySO758Pq>%$gcTTVdyXM?)2`tWM)6<rUySaG&T{RdRBsCM
zL|U_`S~6r2z~NEZiV1hgs4uV>fg%qqEU_mUC6aEh;7>EnvC8GnwT-?=>5mEL!5tSP
zJ@xgH1z;+n&jK!~?|%L#x_3_786hqlvK{BkZ>|u7)eL}HKvr}5$Kd>C{E?YdktmMO
z3;0Y)SSIILZx4K0_12)lPriKlX!aN8f6U(cE6^n=47qU_D+ofbJTS6Btv|4ww<qW~
z>kVW1?*_d|Md;p&%bAOEi`4IffEct6Xyq`EGNZth8r%T?N4El`IywWT$?~n2`j5f0
zBLJR4g1{H?KoUGplX?qo3B2KBRT2JU`FrcUPwzYHR=yM8KOg132(malRY0guLk~z9
zM04kz#Qh;Hk3{DBiYy$#kWJdT#kT)6g;>1ix5?1o^gFwux4GHK0{eg+c9*<>(o~;<
zfM`G?R#vv-M=8Nhfy#A&{stA-87{+8g7!;YL_;{axe{l1kFf4Tx9*OMT=xkya1OBr
z$-qY|J@%eqG6yzPKA=TUliLI5CxEgU(!dP|{#Fg??$-Eg1a6k}F(?SR0+#|kkp>m4
z8}ob}UPUY%k=^UDSaHK|fw-Ys7n4cU0vv!|Icx@OvQBVxmFUtDTY=chwCqpcFVdl7
z?K^Q5Ma{`Av4omYlft7P#JgSi<>cEdce4)KztTaTMHIu%;;0q>C74f2*-l@pS4LrV
zVx7z<J9Hb&s|3}y{{iE<=-*p1uf%uAnKw9#hIJ}7U~qQ;oT>|@hO6(y5`j9Q0b`h@
zd2})TV`w!neB39R2E<j0VKw(C9n2zUD>+OhAOh1tTrIy+-#fgR1|D3kCXoRNkb*RO
z^bRV%pUwCw8xSNCmqOnhelyuQRq|&0%Sqv%;XJ?+frE5~G}_NBXK+4hiwb2(L6AxX
zI&>`{sXxUbN^}%TjFSpGLsyy1vmkpt(0r76<|y7}k409=u9(kNW!!$e@IiOI1mc4^
z)W80IBHXJ?E}Cjxk=ALnoK${G`ma{R9Oa4WA6RS4RQ3;}uQdPn@2rXIM=JW<A>YQ5
z1HZhurxNL%*z-zOe$Dj}>hLTVTWL}>7d$p2PT4m<bj;qSGPjE4xq@U50>2<EZ*t-d
zN5@u)rvqp&sv&)Vk*X0T{okK@AQ}L;<5V#6zlTo+b`kJ3<ZvLI05(#bAz&RD?Xsrc
zaN6EAx}|3Sb%$??_AO?aRG-R#*GD^WLVWe$U=x2CD|ZZWn|0BnF7q|nAgKLGd3~f|
zio->Z1D9CDu$(^9?qK@4K%#A2^P2jM2mZ^jvh);9WD^a|WVBk2PYQyaS6~LfgDVSf
z0FNyr62v8zmc}@^&B@t?5@0k-6{D1+A;T`M6_{6P&`~fwF{}N_9n#$0Kk<^F75Lff
z!KyXWJFr%A)9j!<oIUj+Kle#2gqvm<0rB|RZ$W{sAN?zr6Zjf1OM1n|y;Hf?*Npzj
zEHT%<a59~hQbgV&YQ{@iBxjb~4UheZ9PX4M*4o?B`pY;M4LxE{)kz#%!KM-JawPKp
z+kpA?h3F{EoBq5D+X4gH68|}r3pi>gSB%8Y`dLbgLb^W&Bgi#L>n5}CAs`d-NQs41
z!{QNuc2-bP21q0@TTC|5Koem+)IP%GwfKIGWXc6*JYe~cLPCMf<kq6pFwbL^CI$=A
zR9gi>cGnqz%U0=ZZ>_$e0^ouhR9$gnCRNv#iG_;NA>|YmqU%O<QF8q6zfzf-tXH|+
z@oF~?R;J6?m=MX|RirPX|7_JE$!%X8CiN4>U~|Xuz{CcKJ(^GBLhEjlCaY%nB|x;L
zgwQR)EdXkNePA-0r=?lSo0Evv*3{n8v8q4q<BCQ4sXIgKmMvMG*tp;2R6#)8ep@vB
zFTwtY_kQ%$87GDFn$~Zd-};BO&ck8y=cQ(v-D-WI1Ah^jclK9CM77rPtXy$$w&Py>
z=GALQ57$da0>BLqW<`K6BGazndH<|jwSK}5@E@KIiIm{4@C}Qahxh*=2pVW00yF~@
zH^?adLtZNcT*5__ATs6uJ;2<g*Tm)cdY>mXZ*H>|reedAqr6`GY^D4z<ieS_ZeZ>K
zy)IA<A=+8%4rbl{NA>>|b>?wN-~R*u0N$Z_pedDssRNHfg|fqj5=ADapsvUcwG8OC
zWp=nPXi_323YlUS8lH*Eg>AWYSzeV(Mg5kuzIJD>wpwnrwr{`JoA&tr{?H<n;PZLC
zU-#?z!oL}qd%&8Br?`JD{#W0SA?3?3``@bzUlmp}!aWw0YL=}=&A-|znQwfg-sX?$
z=@McVvkKhHrT7yo+^y|(o5`+TQ%~%SC!3c0s&*`^PJhWhoNuV1u)Lm@9;^kg1dC`0
zL;r0KQaD(3JL2Pk!9?GJVe>nTbBmEOfx0#zRJj&<svH~^6C?)MZAW1VuA~jge04ij
zwY!t$^?SJ9+UeT6Bg@^IA2-*F4c*t^Ms}@!GIsKqqep4?7ft&H9L+j$3@_PG%G&bR
zi)^kxGm`LdjUlymo1c{+jjMgWGckR^%7@iNm3LM>?rmjd)#5YutG?+=k#e6JzTwte
zQ_utiL5`9ZgjgH(WattUxUpTD>(<see50yw2oiI=T8|8}k=3>8)tOB4!YcVBE>%e%
zkXg?+K%D7OTnx)OrD`H`NDr|5o$VNV!a;TGvS`SvS@Yx2n`4wmb@kl7FRjSEw^Gz6
zC#L#lWIrL{rmpJv&(6bj-am35r(8NFU2KYqKmH4yRMy}QGp<t3H1bKZ21*AofGZ#g
z^Qa0MYFomw+uPyvLlILH4_V1*M{EJJpXMU%%RF_w{>(UeqIlgE;RRw=min*eoNWH=
zVcrM*WPMeer<r$erFoCr{R$bbmH-o{XzCOkZSOLu3LvhAC6Ngy73oVjGBQ}x1DVYg
zFbvUquTTW%C|%W$`9pjOt)I{uqJy*qOlkwX1DL4Mmcb4JAKU_S4D2fHCA^$*^M@ms
zhno&nHbl*vzQ#c+oP1!=ErDtphN*e0d6IE<cT&@x^Zr0$lsG7#r+?ld3#1!Y``Zxr
zk>wg**wo3=Et?+Ch}MTY1nxq}jtqJP4fi*_QH(2t)l5@ZJ{d3rs$qH`^TTRF^me6>
z&C-L*PRydKc16zQQyOsOa`wUMwI&>O3_KLr4Mfvp$j)29#3Zl;h`u7Diy_Yf|9Hjn
z6b{^)DmZ&4_+g9Yg}5^Hsn!|AmnfUC!w|yjYPiJ!pfM2Si!O@sySL%$lQ=qsGN$4q
zmQ~&vKhzNN%%*rMIh(sGhb>OyL~s4>wlR8jI4j5?I{f-w;_||ax?|sU4VXJc=jCqD
zTBItf3NrFqtc!!8B`W8Hqs6v^Fgmaom8fwr>?xm0&KuJQc4ak=Ie97#A;9!!Oks`=
z;8I7Pv9`_^tna27D6qkb8!lKM>`n=`&>uN5()A1=2ZS_yR%7Lw;PsP_=FX7XpEK+6
z^1Q2_ot#AVmQgp+XO$I=pYMzaTV4XAnzP#EfiCxD&duBGYCH+Ti#gY~Z+oQgSsc0c
zw7X=4Zv(<leKGLsV2}ZPW|kl9(I~wJ5ObUo&MF*V2X=5UMysAK_M&LSXBa%+WTiyo
z8{=rf5QAB|vV#ekm>$X$rV|a7ZqXN9dju*%<O@)RSPuZb<{Bwdc#!=ZgH*^_8@TQW
z2Y{d%I4;GxkB@1gcd8#q0nA>i0%pR?t`Wx81k8Cpcxb1t<!o8cySzF7v6=3**>{j?
zCsD}l#5L4w#s~?+PnZ;3zGb|Js%tCTb<38-2$&rk@ON&?jc0XrURiIM*GCg^SK>kW
z`C;xER&BrCmqq$)(l^|0CpZi1vGm3zi`H~hYjbk{kVX?n6JjNmF8y`gHI!YqNw>_-
z0%fY_0IXT6D$Td`fj4D>^{B8u8M1fK>w&%jL`-IR02&d}JgLw^i@AYP)Iyw=2i8FF
z)S~kAc$pZsiNnriJal;Sei#Tf5-2lCxkild($yZ_5+o@}&HnCotU~`?;f50#lQ{w<
z<u)tG%aNZG)NPJ<{$Cee&P_(PIk9%0L9t}Mxzna$G+)rzxr*242F054pnZT&_ol-J
z4HQ?$qWpjWXJ6wMV|Ql&K9akERS|Oi(0GD!rvH(_On?k>`X`mVbV|^*9#WP-qqrdj
zbT|+^l}Qi-QJ|Zprc?scmkOEZpqmiyWu25#a^;LXNTv#LU+8is`++Fb-rH|&u2LoE
z1u>fT$A3FqARAiIc5Dd^A4@NswtXP3LeZDBddXU0QDPTmbf|tr(MSugJtJJ7PY_;(
z0?!q5!;;`&s)BB(CwSH3c-B7UY)Bdap-Z1?Hs0x+{UMmGo86zAls<c(h)qhDk;s`K
zM*tK>z61(};Y$AufQ&w9DAEC}_xNrK{3R0@sRrEl3k92ON$96=pbOD_DDcYv{~_$2
zkAaxLSqDfmJ`)y=kg>F+mj&CtQ9aYQC#)<hd${-`miXC&mHtLMi9sJlj!vCMzqt%p
z-F>G@$ua3S)Lh?>*b9;TnDq)``PALq7D8g|W2-#?aEtTK=59;%BVo@wp5ec4c)-PE
zulwt|c2SuM#5P-7)tEg9GwVcKUD2f{K8H(R$7?ivyXd$bG(T0EZYyJ&HGNd*;x*5p
zkK{7^W4R%%y##S>iZvb_eej+zt;OOBabl(n6tmc(yJ(&lEUiaK3{lbyojD7D%A~~}
za0w_pi09dEnodtXU?MhhLuE6(o+-~(X~=$SwTsq-j3j7<J97m~-gVH{ueJ*9Kbmtc
zsKUuZuU`KfyQ_V^Y;pZi<5HooxFAqKZ@l5_PBDZGjEZq>U`N0YWQPmukf0%>dcq87
z?p5DTTlR%Z5iru%9P}68<K|^d<lovf{)MI*W~*V(3%F$h92c269#9Wf;<f#_Lq-PL
zfGx%;Ac-q~LHz1{$hc^6*_O!p+7r(%N9$~^KCM@mg&nvT5OK+bIn#d9v8nhnQ!cwL
zw;jU<GNPL1c{0ssnF+<Vkwsl~=ahFBlwGtvlOS&pk-f<<)E|@uJ7K^blvxK;3x><t
z_MoI9!H*Sj*i(X^+%Jvb@)J~dE>xA~RjG5Cm363k4w_j1XFo{XCsb7#3`QhFW(mxx
z`g#vUW9Xo7A{o27UP5aZFs*F<5jKFEAl(OL9GGRt3xMwzV*`~CAQ`|A8|!Ps`K=Dm
zj!Ek(LxcCum3+9|8rdWsglgb$DhyXrxjcY}3Fwa#RF}7ggsp$e%2+oM{8qn_3gB9D
z=8;yHm+di+JZogX4F~G4L)dU~Fm(Bhwx%7nOa*5_ic>%Z2A0t3{CI18XLPafH(2f4
z=Q%Ezae=piDA8}yiA(%!t3!yeK}PLZRoWpZLZOl_#`Ah+dV^|%P}t(gAr}sV_8Z)C
z7`=xwE;fBkjDg~U5A+f}^+^U%v;uK6m<Et`6mR!309Py8`$R%)YJ(xfNi*)B;q4d<
z5*O<_2|k7@uX$eX<cS)e$NV>Pp|!8_Mr?cK%d0=d$KL&Few>04uS@e&4Qj{ZdVE%j
zj%^Oj_ke&n3tn+I4KOO<0B(k(?-s9#!l|TFSUm1t$4w)D|5^7cw0U8F(!qSV!c*S<
zGV$4>7X2~Jh*&EaP>CS=@5f0bpok$zCf|e2&k%30iL9wY{FFNOp0cEn;}O=$%Zg(6
zR=mo*k>`~~8R+xgoYc6XAgz|v?<93_Be}F=Of_A#vX_CyiE-G|V|o4rsl7hw$r8up
zH3$0=wUt*%1d;1P77yIHm@MdvT*XT^pA)7O=tLY35Hv1Efa(SkG$+N1S&=6X8Q0lH
zyeM+@IL>qhon%Mr!Ekjjbpn}B98mCK)*Y0b7eu!IQ(H)@VPW74QXgEbw=7>k0YCg6
z;I0Uk=g_afUD1MT3-Bp*FV%9COs4er8@)#dx%nFD&C7hqasHd0L<QK2vsP;d^++tM
zK1tpEAk|k8u{k#FL&>56g+P?WqKNOzNjz@WU7$SRGi@vSfOr}|Bx9!Xi@xqZ(RoV5
z30QM}8ts>!DDd{f*f%pzJHQUmn+Xv&I0*@=uae_J&t<<@n^P9FX`g+A^YA{Ww|4eu
z->BM*qmgjF2hP){Lb1G$or9v?7;DIL#Ix-_sWCd-C}5I;q!B=gK_fCsbs{$b+%td{
z#^J!Sna~qnwO~g{*o%y!`|7T0&+Sr`(?c0?{nOg%Xm#hsc{OLR7(%B#GBS>4=_h6O
z6?4}mz|M<Y-Hp{H%CG(Jb-pXF=AII{dIF-zwLM8t1)b$ku{}roL^)H!UAPM`kBn$0
zNRKM3?v9zggul2$=VGNe_v-eofQNGuVS`T;1|4&KXJ=gaOZsP0=txU5t-2mGa*A>X
z1n?9OIR_`9V`jd8DKRaEi46e=(1kVs$|ga<ok5>f`xDHrKW|SCI$b%@I63%%x|2dJ
zF%`5nJFXgO@p`1w7%5~)#lBuVYmPjBDcszPVytfDq0WXqI#(oDAFT*XNT-C?+EnIs
zDu7)gHg!UHQ*Bv*RRp;Le?;X0YXR4mD{9dMnkGnnp+S32GrKGTgE|{@99U8<!v?;u
zmtE)~!H#Q-+##22i5P~?ErN#4hOQTjJBN!zfo&CZx)K0s1Iyw;8AQ`%T7}GED;@#w
zthRn(O8Wb4!+zD^;+V@>-~aq%-S7MiZ2gyKd^zyx6~%Igd7yB!>OIumac=@+)5??o
zgeA}ME8Ho`&0Cfi#)|}n_bNfLHAFuTaVWXB=SGAr6IYqzaeDh#bmHl-_ZpAC>a8&w
z7*N_rJwM`**E$e-l^gh0zhV0GZQAZME_LCcU36v^Kn|kB&cl)#Tr|!3-McOQb43U=
zX#&lJg7YhoIq!kk04|Xj;(*dg%Xk>rFEJAxpEww46|5(ryv#qA6$BD8gK_X$II)fV
zs0|pe@UA+TFPk(}6Zr<Rc(TXB<kw0Lb;f?1Q!V*z6P5DHZ6^f>I!`1N`usfXw0Vw*
z9@5<dixy@{_Vsown&0Bm_;~a50tYD<EUBmnk@bSHa6vBHiS%Gd?Nr@EAp?evZ+94G
z2;#bYS*P|6Bre^tQ6LEumC33}xaze3*1kT|y)vT5g+>|;V52bHz_4Ti9XMcT0Es~c
zJJSpuH`IOQyw2YnSB>3%Q<FEdvh|SU{;4mAJX&W?^1AYy{oet7r#N$94d%2#>|nyC
zqgn;kO1b_QjS`EcLp4T^2v^t~th=Fb7C$*YG}G&QW|=YORfudnh8Q8nJ*a<B&lIIy
z6Uys=n*+%FG+5mVYnm-aLXSU~QA~)5>9w8dQ&Lun4za}ngQSQ=W+Gc*Ab9^%TxbtP
zI&hL;<`_U8+kc6}9D|W+gD-gIf`MUi+{}=%wxIZF&RT`qk=apWwKd7K>-?{UiL7|i
zkwfoK?|k97^T&{xxI2Tc3^Shrkm5Yj1NLOg5a#UY-t>DLYSdlUJUWi)cs`M3ddDEd
zUfFxqV7wL8H0<xTXBXS;TX(CMhgZ~(@iqp4#v|!S8|$prd_}^=?XF{bR~4_d#@O?i
ztk9MB2Zk#raq&)Ca}VAD1Pv+-Br>*5V)4(Af<n#U$iIO$cLrqlXa-0`gYS=1g3d(k
zWXLs$w^&+1?-dY5%R#lbXtxk9bdyKjye~go)3`VT;HcMfBMqT8*j(0EgL`IWdA@>*
zhy7E(8;2Gw9giY&ru(S%e-1>)zQFcABBg7*e8??RqJQKK`97B>Hs16TWkIblwuj;Y
z`^qJ4>b#x>&{7fK!th~DC8q^;ull3bvqAB2OHHiUy+ze-bDW8*P5Fp<k590&15d8j
z8S}X`7G-{k+8XMwh|O(H)!S%d(4HzrZwUBm*Q1^(HT;iQ;lv^tFo(Yk+?99jP2kDF
z_@<H8Y3^fhjS^$b{f>}|(F_fhIA3{IcQE|wNZ?Y!qW!MHPcHm7?44Ql+Pk<ZVtlTL
zpg7g*___MOq+vdq;-$})1_nG|5n-G1zMu8*ZJUzBy>+<fq0?LQfs67ULCuQ!)cqxz
z`iJG0jDfEAt^qCj@P@*U01A(_%=NBMw+`S<Rc3F;g)GjJd}s%CT+a1}1kZqW0mu{Z
zE08$>p!~{~<DXMs_8Wy^VwzrECC@|l26@>)vp@n1GU)%$pCEiFrxNXyrMLRg$q~H0
zYPi1$K+Wg@X+<Im4Y#n400kAO)QECLjbrOxe|Sf(d*#1%-@j*O%o(|H`W(5biUNWE
zVH8!s(77#49)%QFo>9JU34AUgt_OEIBGzeh2I11JJ>fO?WqbQTK)D282XkJOzEjHa
z5G)Bea`V7dV1$ruMVRBfMZc0t`!@D45=wYgWpkrBRDx16o^;sRyx;6>ilaJ0vm@W4
z3ZapJ4euHmQ(UA)d(*&8bpZ4|vZ&zZ0bN%HjV<=jKMiW01n%4?{|?D&G_0osO%Lxk
z(|}qa*|ybl<f@HVyXvU`H<j=VjD&SRaF_aBwUw$SKjYabMnaded``rL-c?0h;wSzD
zK+_Wks>-2r#&4BsfH&lKa7-AyLCmAS5aYR!u<8`%QAkji4j1%NE!|Ei=KSGU#s1-j
zsmj%)v9|MJcz`dnk8EA|5v!_R$4P$kuI!sTT45fq6Uu2M6Om2_74K=yYK<cxW0bBF
z13NU0?R;^0%}1>B?xKRE{EZ(D1(eQcZ)Z<=(rJ{q60<XZvVG~0O)3<1*tGh*I=c5v
z$gU-i#uND0c21n$TXpTb@<OARe}UUA5CYsz)Put8+@?QIC#uf#Nk8)(_BE5r;{B6q
zb64ps8gQe3I5|_g+)^lq+6;1(<JtH*Lsp3s^Wr2$)PXo2WKKb1EcQiWjrV{60wOfH
z&|qdjq#$a5w4B3;svlB#kj%0K5}#Z>2!BDR&_Qf_@nm4HY>^Hf4a>^b0;0Qrf^|b0
xSTOdl`$qc!Gfa{I<t`Z9<@R|JqBf!H`Yd_~<evn1u5rKrqNIqr=e8g3{vXEPC%pgw

literal 0
HcmV?d00001

diff --git a/assets/sample6.jpg b/assets/sample6.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b9c8156079549cc84bbb62cc4a73e36f34a89981
GIT binary patch
literal 302445
zcmbTdc|26_`#*kW#=b8jGqUfa$S`DY>_gVECY7C$WKAhGvcwo;O-y1i#u7ycm6Gh)
z4N9o2A$loNMBmZt^?toS-_PUs$M5$$Gv~gq`#RTk?)y6D+}FA9`+1$;Z+?FO_$<vV
z%m5Gw1Yj6H;P*#JzUgJ09{^Ze14;k@umKz(UI6@u0)Rvi-yfQtQ6_;z0A@xGV(ht~
z!~fD0kRky2!)IuLkbmU)80B)1+8-Lqpqs#g|Hv@3Z{Yx@Kk@+3mn8Qe8r;s{0pXq(
z!XtzI0S9Ykb8QU`rvu(WJO9-BClB$&1xKg^`$kBjRMb>705w&lmX4|h<1dNS)<GTD
zL1_bk*iG_(+6Pv-dGB9ZGfDctZB0`6ho%5PIsmXTYGG1UQ&UZc{HNVugLI~UX>11Q
zZy6960%U>!j=x(4_R9qSd;Z`|$iMtcnN0uE*E5;_l}XHG{ij~W)y!o3hi05Cll>o>
zag8!L{*@;&n)i3xkgBQ_(0^qp;Qw4ZPB9APKQwm^h50`;IHy?nA9=>TDfzqoj9U)?
zRPaCi0~+!_Xy$)t5eA>-fAA&$Tn88cfGZg)2iw7E4z9sL+2?=V1vMR{>i@noe|P}E
z^%DRfpBOrM2YSDM|1LPl8T-L@aK1l$#_1~<v=5^e@PDm85CH%)q?)~hrJ<^-gSF%z
z`lOL6(!uE9(7|(h&@LwM!4rI-%V@9-BVkYn&o1LRg=jL;KYW)1KGVN^8wPctd$0ol
z%O5_2hWIh+|Br4kAofR>lfh5?BL@I-jQl`H@J|lPKL-E`2habNfB0H|a@H-=zxj%V
z4BZ3!|85}wApVilJ;?vPj+%@T@cSKL3~;k^aB{G5b8>L<@Nn}AN(u=I2nZ^OiVI7s
zAdspm2<4;dI+h0Nnr6q3Dx+Nt&8%#&4p=n<H%~V^4@-Nj?ZHVvJUl#t{DSgALh`m6
z${M!+pSRzw06!ZGh!qb3NdaJf5QHD}yB&~Vw3wOUOAg%0KO3WSFfp^RvN0OX5NKey
z4-f<bW`Zy?Gch_nDCVH^F!3`BNFfbb1noUpr9+`=u^B~dGDfv8g|Nea$*OyW-DKxD
zBn*R#$jK`xB9u@Xn#Z-Yb&O3+&CD?tmJW_i&MvN}+`N5!{rm%Pf#DI6m!d9TiH?g;
zNW7JlOvud2zMXR?H;-IgLM^40l~+_gs;h5!+}PCor0rFEM`u@ePw&VmeQbQ<&D+V3
zvvczci%XxDS2n(W+uZuT{p07(fnAJV_P1I8TK2!};%C?eW@3Ucu^!k30!J}+2tO0E
z6p}^2(4N&ZR8U$imJMo@QB?bqT}B=ImylQ3FvlTT)JM6E1JnLk_W#bXoBzKo``57l
z+BFGqK^R?vAHolu05+!_PK8_LB<guQv<xhR_Dz}98@Cw!l_<&)kW=N91J>$r&J?-t
z?sCK`_iRJn@%KyQ`F1gA5<DbCQAYaN_ekyz6|Jtb2rhr)lQZeBwC9B2Fqz48&Ypo7
zTqx-$6`VGnkyhqr<t2!14JT0LPkCmSV{?-af1?Q&6c?qkk2a$s%N5->emJn6-2^^6
z%RiZM>ZqFMO6<6Ba#j!py@@ZlT&Ccbr);K|R;`TTW-V(_aOlL<Ku5DT`X%|qLjpd9
zS^-<Fv+4d?H1I=~d}Mc}$s^Nc2ZeK`=fSc;%Nf2$0u?vT#%pRVH0xAranw?kp-@Q}
zt^FxTe3M<k-pbPyfZBmf0XY~IbP~-BZB?Xz3BZ37OSiuZmPib&I1|soDAGU{7>0}+
z0&mYy0RALcEPzsM?Eru<+iGP1>Hrku1qf&<A+t5mvs<Lf_q}9k%tL?iAjjRpCA&Vn
z&!{?Lz$BpWDo5KT6;AQ6=m{Rgy$HKbQI3l;Eg|<Xrh4K5tNno3VU|p*jZOqNFVV*@
z<6A*-I-lu#Vz{%QZk;tTQ7<WWb>b&IrB!<s!>lJXSk1al0C=ZwxQ`|o1kR1bOU3gl
z@2Oj-fAki+QfVa7xj1NLSHizoZdIUhnAt7vmfrYg0ZmO;mp?`8)~<Aw3+1sYhZ5cr
z!LW#L>Nijk=v%bc4X^xUjty&#qN?Zz)M)w)e3+_(j2B&Dy({Qn;+zauYnxWg2n$hq
z|E_58Jv85ps_@QBA(4eV8Lxl*dUms6-~G_&OuehTphto4-+h$mu+6qK6#EsX&aBU3
zZN66q*Kh3?#V=x?(oM4DhrgEiuUcHGNcId!7dMw3D~C7VsYj+7oQSq|59Uv+_SWzJ
z_}A^KtnGozAGwVp<r6iCl3K9Jo6YykJ*>}s+h#V8s_4<X?g}zob^8jxYNi^f&Rw1H
zHI?EOSLx3G<~o$l&*O9Ph91GGkZ|tt0C4z6<$~F@K7+T&d40|(#Q_QEPrG%uhNip)
z&U_+vk7jo|3x8fSeN^R?dp)%RtExkno4iVw!Z^y;)s`8%N4RCy`3GI%I3IzMs?CU~
zKqPZzt<z=kJh&3={I!q$ZwE`ym%lA3j1GT8zM6MNCTsarZkv@Y1l+VD7Ao0rW;;Vs
z%+)cvQFuh6?NQ^CfJVjR(Zq52lITX!k^C4%Z_zTq;SDb>(9E1yx2e<mTsCu~{U||J
z)+CO&m>pRTS8nVNVKKKH`ZC#=*eMA8uK%f;Yseoeu44y^D_k`TwFU%tMj}_yaX;;c
zwkbd)WfsqKSr5yzzX=lU3f!t3uZ%(t!tWdTwx{uoNZ2UxsdmBU=SP`Nz)|sd9#w0I
zVAfiP3QtS`WGFym2$MUDN1Aqc$e=n6t<`*z&V+Y3<>>L4y7^fRv>bLE7wuYlU}Z6k
z@wYdcZGw+Ttq!``hIU9NrmGW2B08i}zF1#3mE?<9%;$wB#x;tL+8{^L*$XYNDHHc2
zF;-7_9Q{d3Px6ss%lL1jt5&QH#xSs9>2g21W35A)WM&N<Yw^BJ(tE?Iz4tB-e<bjn
zfZ&^kz?UD42bGkZR>!j&;Zl^Z`by})Z<YvEVNp@6f)TeWmjKSzzB2&CJ40R9bpRBY
z9J=Hy1oXnx9n3Dzk;e~bD3k75rfn;ocM>LSde`>QV}wlqbZAafRSru45%8xM|DAIv
zG7A4q1ac1gH@uYLEIj0(_yf^gW;sc&VKVq5Ssm@JZWaLa!zmDyAHK<tNMJA(ITN6O
zA8`o8IDAlvF;LkE#8g7||JqZuvJP;mlXUaQcvZ6($eHNA1zq>cp=p*KdtDP%RkV#+
zd^TnLjMQt{eHb?Hz$VKZy`*V!y@|?N)GtN)@EA7&IAUR&+TfDwGxhjH1J{gIXXz6L
z`ncK*#l7B*=@<9(LSGJOQNdeF+KXAcB{O(vx=`#o&fE@{WZV~|->-?b!janjVtKS?
zFq<V>Y5Mm~<WHsDcB=*Ys`QVzozAmbQ)W^2lK4y@x4=G!r&u_GOeHxK!P*YdLR&Ob
zAg#sz5KFwefW>0&_->km&3wJfRsJ{RqA2zhQNckWT>7w8&Wbq-|NMxtBy;Q>Dj^Mx
zz|V8f<oJ1#a9o+ASE4#Y>*-v5;rT<!KJcIovSC5^ppeO%<;~C(w&&S_#!^W4x~bJ$
zZ7y~0sIEOi+moK$wKW&3-Usby(J-89m+1JmYApgQB^qXpGJAJLZzz|{?-lfooZi{a
zHQE%Gve0NrJ{;5k%^T|Otf?$z@`=?lUpktjqva$)j;N>yGUG$)Icf&w#5cpi)QaON
zT%)oc(D@OTHZ@Hyhvl0->X`d@SB{5AYvZ*ee=Sq}6)IwNS)09dE>%~|n~n-aab<L9
z$D`C-%NxUN&mK9!+M7k%>erf@Q9DdCTEZq1ewCa^aG>9&K1(m}T`cM{HrYbJ&J{UW
zBQd;6m#ajBg7SFHwkgEMkvd<gm8+32JG2VakXU5SoKjhV^||JY8Rm^BWCr4=`3oo|
zsvLaPKwe?w;rHd!ue{k)W^NuaRxN=7;#y(|q&2{!GfZY;h5CP^Fn!+(*s%kVflUfq
zFo};$7`l)UKGEVPsDf-a*{-~4S9$7b2x~P`>UB%ACq5cv(?)vHdow{6Gzg82HK5}+
zm5!9BQ>h?K02mIBo_E60Qq4HI#^k-mX{Ni0J!jAJclCw4=O$jXa5Vy!El69P5p1R?
zQg&|ps&RM+rZozBD`OCqC<$LSw_&wJAAhV%uzBGI;JkDq%i7;-+Lof2Km9x#qja1}
zYnvuzexzqHOLW{@BpOS?bXXW|>S4XVQjGS^DgrmIOu8@AZ+BNCo?1YZa02ikX@v*c
zzgD(oGF&wLUIv}~rT5K((C2O58+x)USP>Y6#wB)0N{p7FQb(n8*}8}Yh%&8yRp?TK
zo_O}KY^Am|HJ-;5$6`WCv296}v9k{BiaeYSiG1+mq;bJyR-9>2E;E!(vH$OZ0b_J<
z63u=R-QEvvSC#{y*aJVn=)!P{%E4<sodu_E<sveol&=ml+i29DjdeFk*KXA=TOzQ;
z^KkQgw5$DukV=gCG^LNrR5aLZrbPYiByb)&Wbho<M%pa&%EoWtvYoB!%~|QtChXzd
zxC>M2=;rwI*=K3_b)C~NJJGhE_g5_+-49{whggM5a%I`yO{%fkh_z{*PBgy)%?xtr
zx0H3r4<~@s)Wcy~3gj>wQb?5|Q&*ml4Lh8AB?H-v@0V;AJR^C}dvJznwC|G+*=vz+
z%8Vt7BifRrMinu}@(0WKDZzXk@r#LA&kVH3E6I7Rx3plz?R5Uut*Vo&JzT_h&*wzF
zN^8|mtYs=rvcFs`IgS+?j9$urYfn%bSd;M?yKYpV?w<KHN=Js0nU)@k;Q?({O)B$|
zHUSGLRB<KTCfG6Bkr*8z>)=ZKhGFvpf_s)iMIATN@?ug|rEl%I4!SpfE887`m5`>F
zf!%vQDNMn)(q_vLeU<8+h$OQUQcEqoy~pSdM&3z#tGt0`a5QT}KnB-S*ef8MP#dT&
zd%H!WBA?XDBWb&BYH?=Oo3A*Q8QLz7F2;+TF~>41<mPpLv}vEh6(GkF)Yp*T`VT4O
zdQZ#ikO!KdC0f`Es)U&fS-^XUK5~~620z-v<P?&`>zk^Mz4g;$bH!HYf4EVxG&Qow
zW8c}kkjUi0TP-!Zez%fG|I?xb0_8lxV3*@-Is;{BQ%&>AqHUpv_il&o_bYri?bN9j
zZJXB!@+PyH1cSP)99>k;*djf4T^@}Gjh$@|eKW1pBoO<|W5Z4}MMl#qPQ&6&o?*>?
zyZ1Qy1P9@Qa>=>$I~UPP%}MiY_099Svek8gOW>+hz8izaCktNLiR}?yd>gKK|E6c}
zdrAKFzL4R${cqPJHmL2aGZzQ$XkC>ysd4=(lz$!d!pkna`Gt=AtMA6+Z+M6|bZl!M
zUb;ANyX$k09w|JlL-+N?FCs1o6TQ0WROH$#B2rd_wO9JxXBgAms#Olc)k}Ud##%S!
z_U-=UD--=0g@IqK^_GVVD8pgeW;gsyC!ph_bfQmwMDbihQbC8g5Pv=K64dQupP8M>
zaYcYo>5&$E>FX8hECK6#1q(|&4LCbZuY_d)w$)){k8}`c>QH;L(IcVxgn6B0!3}Em
zN1kCaT89qib9y(}?v(`>iGcDnh#BStUCBtk|K9CWVwc{lpTmSlkcVz8h98E&5^i?a
z>)WL$yLJGeZ@XkTN1;Gqdf}#k6ddIb9TIxzo<thX|6Vy>WmKGLQEBA+^zDcSTa~!+
zFX@t-(l@RsykA?hP8G=1{a$`{o`ua|;}f&g8z-jxh9h`TPb-DjtoTV1hb3CgY*O7(
z13se)XYY_d>E}3sRg0PHu*><lL0{6yFZId}K0e@o_KF5*cWOe}!hKxn6$ZX|m(V_B
zq&wox@NIk~T$L`J2x2Xj66=skDkIgmd;*pI1}3caZE(OXmV}S7H2e#8<|{>=+(<Tq
zFVN8pyUzGuICCRYO#8lMOogqRvDez1LaC|hd(LB(8BW6Gut~uiCC~h04_^?XO`mrw
z<jZ+QpLclbR|o4{EkMM40_G&#j8qLN%z<`hiXfdhOtjCakzf?OLYT?P)g<?&1DSl{
z8LmCYm6$O!_QOZX>Ct%_S1)F&F3%^8IWchHgD&B7a+bGM`H3V)zgfKxO%kd0${Ds{
z0+;cgHG3<k7M(8Vs=h0{TNvood~DKd|3}H1>f9HC@ggXke;<hnE)sKhs0jz2#{ULj
zrMp(=-a}or(-T*uet9K6O%Baj9b}Z_h0`VG_LKz6+hvsuySz;k`dscfJ9RoPow#T9
zgc<~zq`SJeJ*UQoRlZM^q~D4_eHLxIXC2ypf$TxQdt#-j_|>AQhP3U@BgC8?_KF!e
zg;n*@IY|jg+0nyxKf}ty87EZVwb<>jO|fH@8Y0?zEb^}L-rYOZRn2O*Nt{iAq^Fi=
z5Y58B*xxe?T6zsDZd*LOSEQ=_HvOs>c_-oJz*xs2VdUBK$@8uHxZ&xkqt<ZvCdySs
z^YGmy8ZVI)@6<@YmmQzRN+0&D5MrJl$%?aTrZ2gEKfbYXGFv-AxbcF>*XZjJH)lWn
z2KXxlGo;<8!R+z{jV`AcZX3npd3_>k1!p?F6ux68+Gni)PU0C(<Rr~qAV4rEt|vb~
zlm6WFrJlNdSR~)%Ky*&&uHKZuY3Y<S`_7p#ZTVQH5A#T<`_U7<g@w5zU1Du~FH?oN
zE#$7288v^{aq>K`%H&&%4rITers^xpKT~SjOfuR{(+>b2y6F{WQR+%_o~kkw&t@qp
z6&85@-McW5<hmGi$SZUufZxLEtCgB9E1_YbFS;%ULHfAxy74dx(d~~ms$EQX4-ly6
z_9oxB8^Qu<Trx%=xwf^sa^LI>3r*8yIF`gWZ-t3{4OVU<j><B{WsMgVO~rSSFW&w^
zw-iQTaKvZ_tCP#8c4gi2od04*vsN3izIJzV${RgoO3X5f`3yQmW@~3Di`y;1Fk5;T
z1ezhWytn!-^R;?ttwH|VGNh>E-^vhkVmL0)Ja08*y@<l304&jPKlhe`v)RsPERaJX
zZ2Buhx$B%+@1csjFWkmuF^_Ml@Gdx48@|3eVRDG3=I<%|nuFI==<MuYQk()`E-sv@
zH4{i5tam>SR@Gl_oyxe0K4Ix|rgo#M*T8=&(8^+O#q`K+n!vAC?S*WPhWc%9E1}L@
zq37Eu&dXhY4Tsb`H@Xpl{fW_0EdJPQuKc9CNd_{Rn-kI@4V!b~;88_P6<|AumrvG5
zW1F41I8^l!%yHm5;Mg~sX_l&roGh%RF!5~Wn~bH14u<EVa<E#0+mzBY(i%>H6WKXC
zpoAes3W^7zZe^uz3Mj)$Lu$MTo}NfD673owp^d6%Gcs>L&B19WB{S%el1_+HTQ_Ro
zB*I#HO>f+Kro8r723N++qY7OaSEEJLgn*SrMsmMPCvSC#rHg-%j0Y1r_<E3!7v+(O
z!Ui>c|8v>`hP~P`*qx&xB5X_zlrkzhdX--EXfYpoxbeQIQ{S$O3C|T)d-Lj3U{W^Y
zEhEKA54Ct#tqOpsp8`NJifd?gO!y3*Cp&r+?eJ`1O2tsVa>inzwYDowyUG3g_Wh(p
zH&s^eaD@%caPsgK0a@Lo*+fVCJcA<Hx?JU3{q#Mh{+4E=)3SAu-k7o??+~=QeWJQ{
zcN>W=5bY!v$)_wQDKdA2F!&-ZLPo5Ka;G_I#VEs(>h3QzsP4$7M%furn?R))yGE5K
zy__w|#uF~u=+d*}_paC9dK~E5euOxiJM)$bmg*WWB(O+zUjAY&gTe>j5DDDsV2o33
zTr_09*0)O&M`50oKFI3T5&Tp%oy&c9B>d+H?BljI@&{@NCgS=v@?`g=?f&zRZQW*Q
z^F`htGHh83-MWh&=#C)du27h%GV~!x+VmHrIcTe$sy;5B*Db)xYF;E${flF~)r$zn
z6>U}x%+1T=)#3Bw^l@64w|9<-wzG8EuJx-B_lY2zE2@3fW65P~Vo%A%MgjLkRBf5)
zbtXN0^RWfEq>#rKO0p~;cWEu+MU`!?ZdhL6t)NMn7G4|epnM?%=N|D!8+1S4<Y`MW
zT(%6FXrdhT4|Kl#GdpIOJWLqo3Jj%lfZGdwtRkQ>88@CNiX|NC)#Wf4D66507Vq7Q
zVLtYDVz(Gy^TN+P0$S)VeiOgITlbJcIsGsa%BmQ1kD;=wQ$T-}ZO~WJAo-ZJ0n{tu
zC!QUX7o*5y&s{TUXz|@?m0;K_#W<w!*1QzyYS72(Hu$7Y&9dtv2Q6McC1<?%LW$N2
zg<d=sJsV5a{oEP>{TQszEXn9iJJ_F`r0m`W5bzmcBHl#NHvMWUZb#8J?q#hO#~c(B
z75O<J*?L9h*^K)1GEOdVsE8KB%mZy@qbS-Ys-(79pvw?lrI5}VNV<I7%qhAcn=`v|
z-0SdDUsm0^!rtn=+-SQ)f}da2(C*qVqr;Lem<M_`Y(hC+aTkeIYCe|9w_~ynj!|Ow
zo-~UIU@{T>5`*Kw)*5r{jEZ{rW!D?K)oCs?HtMrlru{hITe&{)Dadq~go9q|tV8iX
z-w(C4eccs}RtmG0iG7mZsiK~-+K<U^ntKSw3mAAY?bPhyUd47ua{YQ`uqk@MjRii)
zI-b20x?^V{`U?55RtWJy4hu4TEfHw_{+32P{H{XkUwh}x=#5f_UePXRln1Un6%#eM
zmR{L?Y2#gsP0mUSi~jkl&IvjG@--U(i;xPS_ylSH!fmp-S+c;X{op1+i>BbNYj#IS
z{qg}~uR^v;E^n22hI6YauC%51#WfYTEn$pxn0q{!&A-3VwJh5x$-ZOHt&`d$v3e*^
zv*IG#>qhZ0cjDPGVVRImum~9q(Fp`6bI?3pM4<@QVwN!f62!x8e<4gmOYuGaJ~7x|
zC!d;ep1(Up=oa1Kfrd#!ZGCO$<+5(ywQjGmeg9*JfoQMNG$#S=GelK&^7TIbrp|GL
znEQ~zYuHq$!Cy6@KkU23r*p6ydFu6Xp8{oKRBWWQpv@+$k6hP%d}V0}*`{o7FaM3w
z-dk&~vZFP+>Aan#1Q+jbT-bmS!HZ$7r4`2wtuDPIFm-Wo17`gVm55SbVq+;PDX2T~
z>lJ^XcA=KgTxm62DV^`*Gnx}C9ImwP?+W6I5*cW2zI}oIWqv}BhJz7D@6DgOHxP0{
zsXI3f+0yUpG%Ye}q98fqozHW7Ac9*)jCF*pGXhtd6I*`R-H8DCy}ACE!-K<1dg0hs
zN15P&?j45b+`_k?I2-uBu_3}_G2JVz`jnIzHyeb7=$9M+ox1eG(&N|y*(TZ!(Ah9w
z#V}C^SS(uCthm;U&ZIP5dJueN-KVf0&Powm#gN*+(_9~PejnUU-gmla|Kvgs*A3K2
z%-DWC3MWLb^@TlVHk#kc=)zua2(o?lQ9XA2c2brQ%(^!=c66pnkxiAwT3%)tailyJ
z*SwexT6Xd|GgN~TRPH+^GfyZ|`!Jz5Og+Wr&gGEn%q<*B1?$kckmGd$_3|rbA2u~0
zu*;~HtB#NGTRj?wx2TX%(*EYcMK=}%@Qv&pMTLZCSd5mUbPug{zm3H%Ntb8vEhY*b
z_3^vAS=DOKc#DY|v>;%<(GF-2i`_y&@yoZ>cbqG<s#hZ9ILeg!^?0MZi$p^)E&Ao4
zSB24L6jpB~L}e)1OI+uhRqJ#=3WA>B9i98|VNCiID(2^X<cD(U3W3+iG2;Fw(A0z}
z%2%rAP+)t=;#28*?Yr~_u@(!hL*Lh=;2o-F%Kgh(5UrQp6(#iXfM<6kj(cOc-*~e?
z4JZ!3iX5F1_+uyUyk@+D<6c9bGD8-)Z>BO_-_rnTe647DRmPK=734{QwPbTj2gvhB
zDHVw{L*VP}4sjqD?`#jont=tH(-e?bq(Hz#yF+~C8WP#6Q;|@Cp&I!&_O_l_uUIOq
z1j^GkO8~z#L%1Jh6xv=Akd_}kN*LoeMv0LWrzFHd?L%tp88QT;X4)Z{9Mut%uF9-m
zl~#5~SQI-z9`7s|b@EYYumtde$PjUi^NmrxxtKeLPlMN^j+}I!enM(VS0H<{ptAE5
zj|>TiQ)8FR(x-ZY%~5gdp=Yh`y?rm1l!n!&ii%|73vfT=c%xa6?5u?Y$s}0x2q-Cx
zcf2&szyqbq_DE9v^ZNm!$38TdgVpNli1_=nim>q3?Xt%ACoI|F3V$utmY%r6T111~
z(S27C06D5%E!`84DJ1RHRT|`WMk!3VG;$<5e~f<&4Kb-EQADzY-uR{4qubI)1!bIB
z^)`bXv)Q=i0M&{?x8TvNel?ho=S(rOw#j@DZK`I`BN=L6tI)!B7-qO0Fu-hoH{)<;
zRTz$@@nOe7J;*uL$GS?Z+@OmM#%Ux+Ih0OZV0IWQv6Ty<2(&s$xz0TP4PXl`ORrI<
z(n>wHbxZytn)YYox34*x&4_-*lSf!p7*jlw#wQB)k(71_d$rAJqDk6S{>Ewy81V^2
zi4s=S)+*(>)zonW?EQ+T1;I4@?HkA4LwVAc3CF=mOaD|GIKI@|vCCXXt(J*TXXK?)
zx7kdf8;Y2Liyp1yDXnqD#w_d%#rH6?#Ac|@=Z~F#u_a#t>+X64Z#G;RyW!&AUC@@@
zW!3ZCNGR;Q&adn1Qn8~NuVrYmWG}C_1(Nolbj`eE^Jc%lgVzUhf+2le{*|rGWhz8#
zi>iiJ)xq+y!{^9ow?>!;x?g~pPggaRKxCXG@s^cxhWMqHw{vdZhiD8778q^!P&=<w
z$^s=2J-=LhVJE2`xYZ!hix-6YlXQkag$ovB4``RxD!R|wJrPUYZ@4GUKJ}Tx-=k5i
zqBI{BrgUuugBgVDWvvooZVmKD9l7>p^o>K!{Aah+eR^eOFTble?|Y5lsDSUMTGjQY
z3covN+&y->y~d<AZpdM}jmTq-5l=BQjpUQkS}gvC_9n9jMUh+j9h#r;_0(4=y(%wm
zf9z-VGfRnA1M4$Apd=@dCCQim?pAcDQS`VCa|QpC+50cT-AA2KDKg@l#PfoNEz8k7
z4M!=G;@P*>hnT!e6`_fWOJd7ato9hTv<6+s1r*cwUp|iQeBq9gXUoR#DYzwbB^BgH
z&|(L{6y`ayPBjFu!M&Y>a=jb}ZaF<Fgfp!vl;X1T1v@StOHi}Jf5TR6R&{d6jb~ie
zq9WhW8pYf2qPK{bs4si;E0T0aZK6g&X2-E4kqUvH60N{m&Pbt9sT^<Z`=JjoEKhKG
zn5SHNJnAmqcu@C)hT495P3>%66kpD`$(Dm5PV^)g7B^YwQ@TE;n-y!<C`1|=+}HOS
zJX2hDUzr)o#Q+(zz0w(YfgKC;->~A}7=#;u_yb6|0hWJqB<3jKO*n8t49~+5$Buv)
zIORXsCc_;uC@hFqtCYfA9Gf5q=0Kv^2^H7!T+^uctp#Nbp{H`@Z#3>HA!a*qdC+-x
zXZ_vvCN^Gh^9E1%OeVe3oV8S4G}<M1TbExvqzg5+@VuVD_t%s|*A$A&D_;{g@?jP)
zW1R7Wo@C&O<`h0OXHflXIlSW}$jnB+&U%%A2L4h#kpf-(c=r<`$tVBzRzJpkpos=9
zf3535aNKPbsXuwtOl`6Hw2JV?1M=zIez9zc6+ww<A-O0Sew-2>!PXD<z3h49D9Lul
zBlL$8fhqB%WNH{LSv4BRx8Uzd3w<hT=L1>6y}2cgP;ufB$?)eg+!AtD$h|hoR-zIJ
zdtma6ARv#GZfe2!@PDY2u%XXsPqoH<9dqq@q^)A<4MF8>^B%p6ft7g$yUTEPs4I+=
z3Q00;3*EghxSMY7B816`=1*!BI=c|A89rH)vvln`|J_TI>A*te)8K*gmj3dxB<<tP
z=PU=uWJ4o_*kMAI+Qkn{`dZ_fHY?(uzC3ayW{BC|MlSRz<Y=JD&%1Hfw-66irTf#0
z$5nU8@lGk<%lmQpg<H3U)~}De(nmxpy;^fQ<dDB5>71AYkZ7kx3<Gf-Ztb59$*E}r
z_3UZTW{S_zuy(Lg>bKahac{C*1mRHB@RMcs`{cVU`9L&>gA&*47fXS^fx1%@luRMB
zkcL6UyA`}i&vZmvRp80sS}U8&9M=9?pWIzXxpH@Dxx-r6J}LuR%B+3lW1Y@Ao}oK0
zu5`-h{zz^YSBbu@C5AI+Zc%otL*~O)MY^OzJosK#{T>n<OHSvmc@XVIm9S|T6(iw-
zm4%E1TaMXmw#y@st7x7N4i=(2c3=@*JJ7NrV-|udr=Czospn0UA=kG0{o7}J%8Qy-
zViE!~LVZs!|60y-kOlpuKQ@T${ZQ*VbgloBQk7NSB{9Pb+Ku)yBw?ABVBDn#gb5ex
z=d-~U`nO6E0&ba3DLHzDPZTk*Uta}I{gfG<tn=2Ff!;k|p2!?NUd;b7X`7OmnETgE
zeH!?My;X`(!LV(RBX6KC2Xh!}qY$SlK}5v&G#={2`herL11)z5fmuJ2cb8kHKzH==
zfDWB#HCL)AxiJ!(5j#P!pTYtkB50eUS?JE=Zs<@W@Kx`e9pcy%s#KLh5NY!SI+UCY
zmZoi%!y*?86q!(=<D_3BirpJ*9aT1F*P2-ZCpCYJb!x+__uCZ8KVNl7WHvP3=#X6P
zkQCM5u`>$*36c`C_g1_F;fsl4+^WJ;eq^oS%^vwqHVxD@@gtbb*Y=G<MSgm3S#({T
zWZ-v0+ksA{kOsO}`bSKw$KuTa>3Qa+&@ZN0Nn0@3-t=rq1?zdf$)+{i`-s0C%AbIR
z?E1e07ycW<@c%?H!WxPh0OtP+cF6vNfw(b-IbeGX`2O@>`)}Yu%CApWR)nLRDVdnv
zS3ezHRGxiBObvAU_7Q691l@mCxn0=b>PWg{Oe54xzwM?Sr=3r}r<is^Lp}Y~D*9!!
zXTQbM?Q0@vv*LS_v#~_tUsFpgrDxB?48o}(Sf9z#kKF6^CD2^XsZeXkVWWag07L}s
zt%OQkB`|5!HhEE(`pyMXm|@djYz^<6ZNTKXCSUi{2nKaTj7;^KpE+MxXI{8Y$ay;6
z$#8{?;0>gOm8tGmBi?GwH#{XH)yh0qPF$|wzo->A*cbRsWl5a=mSZty_xcaTRYJj(
zr>Df{!X;ZumBA0eVDuNKTkC?TSpidKW4XVySI%SB%wyVp#x#As>SCo%s6tj9-Z43U
zOyEv6DOmU9u2YrKuRG=S{l*u_565;|{O*O9`!YW`l6sDg6y24DEIY~xmfrLUh=O$3
zlYhVu2icgTLtQm0#tY))nxpC`)0@0vl&Tx^n{m=%);KXPr3YTFR+VqR9?o*SI7W{7
zsxtKfwCbFf?dkdYyk74mlN%!Ujs|0=i}`tT<HkwerY!2aDxa1X?6Ib_Gl;?07X7Ss
zzHh4^_~1J^9$eOQavD+C-`RI;&N$njBha?s#h;7@mP(PE&D5~|;|h~^RQK-6TB{kd
zrwwQ+C=d!}MldW)c{>Cf7ldJP8x<e(d9+qxJHfEnilcd7%3hCWghfDg4#pCTI^i9=
zT?s$*u_?8UG*8@j+!V@lJ0h%f(sGLIJZc>2g`0A(GNdU<+)koBFU+(E%9=pk<xUds
z2+vfv*C^((B)FF>N|9xKU0>YN?lAhk5j%tpHAO%loSor|7W{NXeC0!bYg_CnxM@{m
zXoXgYy&s+y7`8;5&^-<f;g$zwSgasGmkflp2BBS_m0X8Dhn~FY09nR1LLY0CzyQdQ
zL(hZU&*4Z!Xos7u;;(XG2g{EJ4O(GABs##sHVC6IX)!R>S0g3qAFouRL-(aZMtt(g
zT^6vIsioD^CO$c}d9s|O7F)`F;i_TYx{4FpyCqGPf~bg-%;bRS38mcNmQ`AmT|%<>
zR(>Os%VT0cl5E%(Lir8oJf&~-8^b@L8tCZk+$9v=OqAg6Aph!Apn@eXL)Ilz%1JH-
zUMU54(YI}|^Ibg~6EpC{OZnJYw%O)WwTB}*R2Qz_q<WHd?U1@7g5R*9(EL%1-VlUs
zlQtW#=4$hnwe+m$m=u@ity0h1YYtNU$BQIV%`Wda`|^m_e;h;ssS{K7aUD|G8co-x
z)qKSg!<X%st<G7;r<)5s`N2%DzdDOoCk3)tOp>2zNRjj_U>oRohh>{!dMmG?F;|)1
zD4!r1xn>QKRfzp!>u?Zn)u*7q@;D;mb=H{(D<zpobLh^hx{-tYg<9QcaDcZ~^1FzE
z7;Ci6ICr-sV=ZzGbw4iL@4@@4MfV$gB}*xuj9`mDW0?OdVT^gW5Q=<%oRYs-02HXt
z1{i)wu^)L5Tfv!#WB-D7_s=b;XXnd@80xjv6LiQ<4A62&m*92>#WSl8s9mPIa5Xz*
zpxrOwj$eC2QBeu#2Zb<#GE}6rD0~5&M6$g=Fk`mY$m_<Wi?Xw0RA7%`Ix1WNe$;R>
z?x|NzD|~J|Jug#*%K22`=#}z<MeFJF?ENI9DbD8#h-t$~*64~*zwCVf><G-UT%q=h
zeHOXXva@2!|2i`&_USm)<NB6v<tA0t@LTk-!VDEjFgq<VQ{njQ^_wA}B*+9(nD^qI
ziq@J1{>%0H(P12*_$g+&k7vB=ngd^_gxX#pT3u10rV_1f&N@ZUNGn`+ll6OzR8jxp
z{NdcQBQ!F-*J@P~mcY6zU3jLv+lpYc!Yb<^)T&buZ{HL>B=tNndb>mABij98Xmcmn
zf7L44$Ih%(sV_egmTtuplD9fi`gKYFjQK~?j$X$n4Ow#A!rsH^;*rVE-84TAq3fR(
z?`;YnX%0Mn0_8b1_h$4)os>-G(mwOa+y^HKPkf)`uW9d$4ipC4Sb}lkX>TP(VrqT^
zIVWuog^)5HE|{zC<<chAU!u)N(jImE2J|MrmgMT^#^fAZKell4`{i3E&~Vjx<M=>U
zOM8(W4i{yUGbzhv9h;@0VX2C`9R6=-r=OK8ub4XZ%o~@Wg=0G~z-C1dHN6R%3ne__
zK?HDvGjreA8%oa-SfJw{LdRoOA*2B5ShhG0pmY{57&k<}82b_>Fzkkd`pY`OX@*nR
zBQ3Ie{m-1~0=~KbYFcw`bfac&Nc27*OLfc9EQ9DL>*dJjnpcNdOy9$CYpp1Xey8>t
zcT>-8-5Z1^o=y_W`^8wPz%Sc%Ci4Q=2gi@mv8M@U#Ub@C+*VGD4u)ROj_j~Y*kR1<
zfx&1+kT#Tm5--?TEe9w-TEsVj(KS1pklXeU!gy}O0J6icvw6C(G&tUs<o7r!sfTOI
zU%yf-x1P8jwge#r@w_{k;4$j5+aB!w!@;$oFNwgK!(*}4;nRz<H=^k{oJBw5TCebk
zmJ_0*8E;?$6SNb`p8j0vUXNWF`F`GM2bSeWW3F==OXJb^Lf&|OtT;Sg;}?HnVkWD+
z9KjgYSaVK50ctdd=k^zj*8^3g@m!P};2k<bzOrfr`>`{&HNc`fCdIbL&ALvuGLa>P
zTgem#_7%DOB1hRYNU}Z;U?px-e1y+GA9vJCG(IH)iSu#<G;3i(&#tqm&?*+QxgIe4
zA)|>L{$`>>YQ~f4X*sc6#^1b4G1EGupc9FAengJ~`XFfhhBe!58*k|5!2*t+`Ql<j
z#Oll75Yvm6&T3@M56=kB96a2ox*v2u&3ofc5BoUhJ7{NWubhvnJqh=Tl&#n-j}Oy^
z;7YxCN{r++kjDYMY0zT%L5SzUlm`@2oXc<)rjkuQ_6#qeZiY*EB~2U><{D;{tY3|j
zWt~5@4joBpiY6FKA9IFmFu*=3^17`1I_oD~?b-G&W;}yqGIQ<R`*Q<%a#KfsdF&}l
zsEJ(s0Q*@{V%Mmg`aboLu$uJK7s+Nqb#cxfPZM3^Zb2sG@r%a(Ed&9O_FyR^#_=MF
zmcld7K$>N<DwsA1>4E7{{koia+!x&Vha`G)BYpIk^oBjPx%TNx>mpI#%OQKG{-BPP
z@29ki8jgG7Z;_Akn6P|x$d@xwd7#&qKqGF+3Ondtd+ogT&=SE5FUro7z4IHGB9h}*
z1Pd}oYTerf-KTm(%*E+Hg4bnIJlqRQk?y(YY#VcbmYenurN4OSZ5o$*?;@?MUmRC`
zHN?F4CIct!QCZLTP+>tNhhDl*98&Q+GBy2(dHN==y?xc4^O>UG(BW#;?c3UH?u-d>
zGfdpuPFS9BwP}#Mu-Z)|zDeQ<d}pNA(&(O<?a%oWI<{+!0L*&V2cBivLWz~};uy2}
z3P5QeS|!?XN{(#yx>mj<rd^<&%pAsIwMm1mh&qUPTY2RUliEsOLa8pzFe@j~$cH*5
z+oYx-^&jOcV|gBPOmo$U&Q^M&tiNoyN#XTaq7mnDt6(9ey^mWIR;~n@vE-5o($-N{
zp0<n^@mELCO2oO<lW+ELNjSLg{8rgbru=9p6WD^}!&1zx#p+^)Gc8csYT9~GiD|Q1
zj!&w6C{t?m<!E*MPnU_7NUGPH`+{>h7YC?(q@O5`SMip*4v*6;S-tpS)A={}H8c@7
zjb;hPHK7>Q0HBo;NkIt{+zBjXL78!4n%hxf!=~tA#@{nsiWAgB%Y#Nr$LHo}iUdoG
zoiU$7HzklD6~6+TPw9Qic5ri@(9H)cos=f8D3<W_X%-Z0i3<NUaCa_<I7?3LTd+6H
zOi9q^b}n?hsj29?zS)6IHh9ZVs}1J!zhr(^$!tv8)#TZSe24ErXW!sg)6+hBV_#Lu
z_JQwn?F@hJSnCPCm@}jNnZ4g(f6C!EaP;oCBD9!SK%~?WMGmu%<oHzy<WC;f76Zep
z&L?*5pIdca_aqP9G?t#H_iCyLK9w)7(_y0d<Vfcoy-e$V=k2!%r|9sRN(BrRQ2i=b
zjvo_MUdX>xiRr0*;`mVN!e<+17xF2Qg-2b7pE!1xfA*g`DfM*nFme3M{7Bwm9(0wj
zdO-Ub!%NFKKL;%5$BiRfe;wf<PI4NSk<B`FzPXBI6x4sd9-(R+yH%u>ktR-0bTi(4
z!ea>2;NsiAS;!`D`eCb7tV20f?t59p+8NA}=T-@UD!Dd6W(IRz9Zs_ZPUsk?Bk(ul
zJ|lpcO%cFTfl@hf>TR5p*vou_GKD-{j{;tM$B8m{M{qK`XzhH2kk<g@XfFsdQ9xJA
z)LE959zPrq``SBn9S{l|k%V4=#L{^h?jka~KQyD+j*RDiP|6I=TE(_U8-$xiq~DA~
zGs}Tt=N>lr_g{NSn9bt9tFd{DI9pl$u35=q^;GXAkz{D0cDbv^gy{6!CngtP_CpHJ
zYhqz@EV=qCW-xfqvPC0HdY4y8=+^=SS6~66Ykdg9{ux(797@DI<Be{H%WG(0tirj7
zLpt7N(A*?fMr@?`m7dLVUyhbX2H7neSa9<Iq9o0dh1r<Pt9eZj>3#-nYpx)EyWZeX
z0)3KIrolioj8`MK70*snoy9{r9J+!7KzZ=N=&LbPqk!#G8;Hy#o@KcM#nY^TF{#~~
ziYhCgj|t2;de30X)o|cEChyUk^meB%w-)j0uOD-seDGE9(5e00An`qGNvk>|%l8UR
zP;)j;LnL<Z<xeM97ll(Z2)>8*4tG~?z7=JJSje3*3w?>k*%WDiqBzw2OgOztS~;O~
zXKlb`?;Ofi8-KR<#=>ouujhU)Fv2r59z0NQ;}Ws0pYktTMu+pG?%a#=s<?ai@e|iy
zovf?q=9Ii=dcr3Xl&S|*-G_v|@lQmr-1=7fF0yc|>>FdL_QJ%k%@g-@_hgB1MQ)R?
z>nE1#?7Z@S1F!vleF8BFWZYaTj1HXC$&4x-h*i}bc;AVmbr(Ks-$yaW@3QGzOL&MS
z7thwb@EXM83_4yG&(*}ldu4rwGzbf#gUrWcy?`!Dl|#axjmzxqrravcLKNZLw<$+g
zu5yjNs6nimwrjZ_D^+}SHn;er=WGvhB>!=MjxwaCxIcUw_ul8+IsuY>?$NPOOJ&C!
zYz&d>ea~;E)$tmlf12*9?J&SrWcME|^(x5xZ5QPL)nsiDM?)x#AOdmBCiN?`FpvGU
zSd0flh=IUn!&r3eRIvA^Dg1i5pOi-4rIwKB3_DaeDet(Jpv;MTx&Earbh|^a@n?>H
z6Qpd2^m*@PJ6pJ;LMd4K-8b(^GB(2)9;j(oq9BURdL$a9M#~DXU7NgL5Z-@9@CWHr
zkgCMpQWb;Hg4-<k*5KJY5>H=J*amr?-@K<1#p}kBD9_PY9vtYnf<*J~gSMz*U$4^f
z6o{eLJ2c0MTVjgdUkj_jeEDib^0}Yqe>h3<pH|c^i%b0Qz1pkqEeea|jHQpaP?hC{
z?gA4R1PlC)F9z09Ao)%S=p(d691jB^;y@&MZSRbh^Fv3eGAf}~iOcd+4h^~j3l;9?
z7BA%L<#U+PonS?*(ht)+=to{x33>6W6ACyxGcUP&TwrntscCso@+Ek38YE`z_Zv9x
zF4%LfH{UbN*Phh)qA4eYlbd$eph%~;Hw{mFnBhJ+sV33rE|{ghA6en6-ramjKD)##
zd%As0&ZnU_lLtLpTY%$bPFu1e4~6GYx%yWE&x<zDwH3aE_B-apo32DAZ=3qrsLRIh
zOvp^Fe=@GTdDIg^QW8Vl7#cXy;@0I&deMG1MXSnU|9ms=uGs_m&?D(#)@QP3J0*29
z1LJ#5zE^W=SVYR-s#$j(=iSls-m82qcDYn6@oi~9ZiK>a{Fpc~rXM@SIq>4B#JY0}
zrQLhkb<jKT2EPw>+}t%W_r>KQc|8aD-+=RbOVDSd`$?TKM~*w{dIS0Ep3X$LCaKBA
zV9cqxIru8wYFrooH9P{drExoQJNdz>!|e^#wAbe&#4e5V!!>SO_e-J(#-o#u{7Ts1
zWn3GidR3#{kyiDrFBspfwr?2QKGtx%m%M8Z@6u)-E}0yMrUIcSM=Kss!)yBCG|BHB
z$E0J~4YwJ06YH}x3UnD;Y*E1xRJ~Eqt;veB_iAblvgexA6P|=ks0oCz&lYzIPd1ul
zY(GdbdUQj%^u}#;kN2{r)i}%av$NjyaX)taN9xUF(z05`?jzOM(EK@ib)xg+)WO_W
z`W0-OMI$*+l8lfPaYa53>-U-n&|!%g@7k)XbZIM+cBzzd-*Q*K)Ai3Yoy~sznX;ap
zrpUx-XRcy|Fu`N7D4R>|o|1TdZpSYiETemAoa(g%y)~KM-G$m1g|>Rfk`OI7IE+QB
z-w>W&>~p2Xcg+z%GK8Q^c7u)b0rgXA?Mn^1vXN9a<M$6)Vj1xrFA|$K?Rk^@>@KGV
zcU-VNP4R2`Kz6!NT`&fb)!Rcn97!MIxtYC5F|k5od7oM_rLm&^Qngm)DQQvV;s`LQ
z6<2avc24$<hl8sq)n5M6QE*dQe1GJqnV_s|-6Yjjo$lP-ou^z!cdWQDXvw6u!ks2~
zX+UZ69We}%aEmiUmm{Q63<HIx{{|!@I~3#n(m#v4G$7*bM2i+U&MDy+AbgvZ>TW?z
zgtOY$9C)xlh@l;dhRc{syhpv#j*(S#IcuWrYoa++4uSE*3^L>Z{45zsxuqB+p3V5j
zWoz<UY7(M&>?Rf8n!cnc=_^j^D{`{gF_i2MamS4mXY&095KxX+GI%+9JP!w<w#JS_
zdJTGk0eG&JK$+v9UMZxP1a}6u>6OJmI#m=qggIFdNYl2qSt5g(djNkbGj=%`<*1q?
zuOQ-+skV8qqEJEyg8}<qBF=DftyD#Ba6oD?YlFj!cup6sPl)s}LV?Sd(EV1_L>iPW
zA+zF~LHA$CtW!>xlB+c?`S`Pnt!QpNM65^2dFp<{gt&%diq4fU-WxN>#wQh+YQIy+
zKzAOS_F&=pvWzWy2}gls1nhnT&v1CZQhf8~X6gQVj3P+*W^w+sejPeK<ap7SlfMCF
z_V0@o5M~KU@y$*6UD%qNNuo7}r{gXofYP(zJS~gf8Sq(pQErDL@=}Q;^~RF`5Z|x*
z6)!#xMaVJdYs})bl2O~&nkSyRm9FQn?Y29=*JxE3r?fDgG0@>?Xih;iD(sAx683*(
z?G3vqCmT1M+X@JudBj$Hn8Ox3eRKAssdJ&c|8F3gJe3jm1$6onXM&LWo6sribNl*t
z^e1bR7YSE|^t<1Ylm_nhe>R?}Jk6nd4;wS==y&cONoV4)?v7AHQA$ATW_kU`6f2&^
zhP1ZO@gFI*<NQA#9wr?ZX-dxkEp*>^@xD+c87)$aas)S^#<V3Yv2O$<?h~uL8~iq*
zB{h>8GHvBDm0je4e$*k<z3Zn#-(LuPN0h9hZXTyKC_tO5NcWR)CGZn&?A~?Omq0cF
zESu8aO&h1&9&jqnN8UF(T1OK)8RcbeS<dmZDojcKyl_!|h;JA#`n(Rt=SsGcV2(tB
z$6Q5a#`$)F8^#`_Yp#(v{x!JOfzVrZre8`bk^OiCstcQXQ`OpQEY+bI{!-5Yf8zE@
z1jprhg7WSz*;v^d@GqjNzD`@Rdnpj!ajL<7$>$Puu8De9bhLK$jN+0f2A3(PqCSp?
zep_RJ_akIFAj0^VphFHoHvKxU?x=fVYO}SHK!>+APfMYYA9%)|$amc33`O-GE*pOm
z@RO5CyVK0J!6fflgjtm$1+N1-88Nqz#A9tNncCA<oX@lQX-II;5*=+SH0}eV%|U@E
zN^JfRQ~0qy>Lm@MeS<(B%c-7bH|neHl%Fm~Q-@Ww$_KAGTTxAZ5t0L8LuKo`^eamm
zTvVxF2)152k^hluG{FJma|RY`3vKm3%dODpiSa1WW_jS2kB}%u$y-km0MVUMpL9!u
zz~ltK*E}t@f$pK?bR&r;(TnMVPw-;z2rtJ%4>kKEx_MOJd^v4D1+St;!n*=t-$+{!
z3HZ;@Wn+J_WgrfwRm!^16%F9Zj{WRcWrAIE*b(tK-YmWu*EByS#hbE<t&q{<pq@P1
zewn3OA#hwiOaY=Le%nv0jyQ2ao$Q-^vr6toc{Lj*&kNF^CkcgSFn_+%*u7H4yXuT)
zveV@xxZ641WJEa}d{qQ=Xk_zuFZZj~tRR4TJ3FicA9m<(B$I(`Dg%tKtrRIop<ei!
zDbG3^2F59Q=%0WKX2y~KV4jeJCB^i3eS|5J8Jc(yMoBSr`XU@zRWfFcv=HLIF)BO*
z;N;W9D)OLISk;lA$6v<|xoAp#x@a1R-IuXa5L!}iU}J<+GB94QQ@6kwEOu$DX#P1?
z=r8|96c<r|eHCwN`8<!xl(*Ld>lDc_gi~7`_7H5k&H5D(Xv`LT30mEEfk+B4;v+s9
zeUX}W%?To8utJLI*9WO|P1c({YCn%Ju5MG~j=)Z}`Fl$2G%$Qblrpg`V}F@~ZednN
zGD}goQqN_^i7h|z_{r$wQNPY#d=ws2tZSc=AXaDn{AVy@IZ@wgp|;?>SKqbw_{zI-
z(#q~@{AupvuF3`GP{U6GZaJHOS-+7K1v!;6_5KoFM~%``*wVZEREg#zjX|@%=W@M@
z!<6{xY0k;>1XaRm?kvLVLs?s2Q~7RZIFO3>ZeHq?k*;!ewH8{@B`OX^S6H~(lW04C
zJsUf9R=NGo+2;+fy&o}Lh+@KI64uFClpIq{xjF5v#)vC(PJZj!7P{#3TBLk;EcbiP
z(4CAjZ3P1E(G?R?+i-|gRc9G_a;%Mo$B2Y<>U(vYYgevz?sfXu)62zahoxm!5SpFE
z!pNgkKd}#vCw$;-@8pi8rPs)0;o2>=I!u)12CkLI?%4Mk7+qAlMyw1`YGVgFnT>rr
zHN}tb+JF1rA!EpvPNFy(!SC$5WueJl8Iq!+<olt_^j@KbwoENI_9&sC)BDlKm#i)*
z!xoYi=wt5gMSJ<s3tVUocnze8y2YJD?h$M-@pRM%?0XHo*fV^>zVs|R`C=S$TDz)E
zMlj`%-`<J50-Kc@toH3)co!~OSRFzS_)H^D1%`F$ub{PNANeJ*?Sx{{z`J+a)YjY&
z7hsIJTiDFlU4($R!R_E~fzKf!%#O#UeLtQYe_}r%ueEBXZe_GdUBvTNAADKNzEvWP
z7Zj`D6}3~-gjJmh?QXcnSj2#LAY~VDAH$bVXQrmAY4B9CoJrIEd?cC<ZZH374w<R*
z`(lCv%Mi0BE2aE1Xz3q(K6p16OL6dA_fqPU;mn$e6~AQ~na6m*nDhfs9P{C#X11k{
zPja-SxTqo`@;iCz9}_{9!3lS!MB408SWy<2SS#PEQZPN~UZCKsh*kHmrbVz&dylxS
za^IBq1RHvys_60Ydn4)Htv{caq5jfgz!9$mtp{|rso=Q+b>|DGm%bQ*Ux!#MTQkKu
zzDZgX^{+OH7D?s`(Az4TBQ-#bKp56iK<ldWv81ob7mscGRO7CpzoTS=iW|%Ok8CkO
zz-?Hq<E@N-d9b(G|Do!=!`W>A|Nlhn+M<NmN@7)ry?1P4RE^fIsJg9c(W+fdiQN*L
zh#93tscLD(sx4H}R?Vu_8dW;(-_`s5`Q!Kf!$=N}BMzL`dA**`$KwoK&SExWf}VdX
zT>$zt(5+Ky3PNs86_s6*%}TKqtTrB6r|=A|=qtEqbnvyZtzK%k*~M(Sg&Ba>yS!UQ
zC8h@OS&Fi|ENAct#aEb$tV}o1au)IIgM^g-gTImNV*kUu|DXEA{}n5M-w6Rn1E8WF
zfD8vZ7^A?i=`RhA0Gbs50SfvL5d$Njeb9jW4Aj4A13-+`HetB*fNMYFl{0A%ZWgx#
z?vv9gyJ*;FoTij%k&0rIlla1Hi?pWCRl@V#F`{K38>gFZm<L4|{1M$KYsNYJ2s$|T
z>rFZkPes#_1{d(Yf_j^~Dt4KVX0v#+|J)sq#*jG}tiUmlLy}A}p`aWbww~}2y7y_|
z5-!b1Y0L8YU|e2rj$(M#!c>_+Z-~}xTIw4Vp|BzJ{6Q1Z$`K|U#3pu%eH*00YPJe@
z<l^|tzoD6HZUjDMs|oWERG{!IYOv?L58A6v@4Vh|uGOLa6Af#1jjhotef|Dp4+o~X
z?Uj=PHcm&!g(3R%KbOCLn<@&nMC6!ceCt&Ar8&i6Be^+;lY9GOq(62EK#K&*IuIS<
zXCB)Qw;pk5v)WYVxxVAM@U!u3rP{RUN$?$6XhLuEPx#$Niluy-mvsge{@PwThv(b2
zVra~Axu^2tb^6~nHgDFX+?n~F(!-AGZY2*`-g*^kY%P}gpy5t{@f#(@flgfnRwk=M
z+<<9VU-rg>7kN!@><zeYBct#(`NJtaRdn%>IS@VNXf{>^IFJfy&a)RIHxEHs?qe>R
z$AGMp>@DPd^_7<I=S*5&Ef^QCzrR=!%m~wE(DOUTt71>n8yCOFc)8nqTqn19un)i1
zph7NYzW8Le;6oDM@VkkuCCx2cn5Z$<{_TlV_9k`#Udj6wiaN`2W<P>z(xDjZ5=_#4
z_BG-{>sPg|7-C1r72W%#w6=+K$UNQ)gcWIJMwEjNtB4>=PGAXukv8`X20s$Wo&SQk
z22I;17iUbaATDgIdI@Ip5e<OC?EKOL@`n;wX6)(%xlDh)E+4)L<%ER2*(U+Hb1k7C
z5pLy=_nrkFqnvt{e9J7e&sE@<B<U!MOi->HC+G;z0K-w#QxSlcV5B1&(`(~$1`|>A
ziDtW%^+IqSrz5fj@4GBn#<qdkhlmL70=1)Bybsx7qFSQiNEMQ4c~H|i&0VvbYb};r
zJZn8+`|Gg$Iu$77r%y-kVMyXj>%|#K2F%L^{1RELpgJ)ri>YLW@84?)5d>PFYVSgl
zy-1KBs!DkCWVzf28@XjbWnz5N_TzXHdyFvTtJc5AvaLmZ`p)S6Fl$MtTliuiZzct!
zbyNpdN$nNJ;XvF8R8_qrG+nzHyS^?0e4AbUrDdRj-m$6@d^YWYE{E<HnoT}mBXf3J
zwe;%yJ-5%#+1}oY(j?My@cw_d6m;=t@gO8J9y{^3(C~l%%q)L*BLWWAhQLRl>wx{Q
zQ1y3Z_+R5n#UL5D&*c{!YoO8Y;L_m#KwV3(a#}d_JK>{D&oYQd5Du@PU2h0Ihv4G^
z1$}rE3^bt(J;iK9@ZcbjMgEsx8iXq&M(G4c%;_rH3Q14RbnV>y+I0iux0wuW12mT7
zgBum$!xtp!$QBmCyv8PVys1nT_0|7^mc97z2x-Mmp*#JWA-VYRnR5cBE~8ek6r^tE
zHz(VX*te=y`DgKCjfS;E=>f}L>Jxt{6YmIuEYiM)9lPS^;o0w#y4>QAxAhfRuS%mE
zn9u4TRO*4)x~$-Rd3Ca}H6Ne}4_f{t4Q-<SI5Qc=wHCzhYaEh%ORH*0YTVxt@Za;H
zpJT3*#bY{pH|-2=pwJTqUl~m?@i&erm1)@ls0)<0*}GTkiL`!mTMg)tG-hDrj_emR
zo4KGxxk&aChjrA?2J2F9<CoWK{JW6}f~_g>W15<EN|-GJcGN~4I@+rA(8E@pf7nf@
zWz+O^-c|ShRet+0lTea>d%SXB2O?)V9V^>QRe<o~fkjx_lx{+uRT;rI(drk2XEc}k
z)~6Q*`9|WcfHC2t#Az78WmPP-naQm|x+bJ6bZPYfA0+?K`ONRStHset`;|5>^Lnt8
znnCK018nx%U7LZgb&|B@?8K(<+jT$H&WEI}%~TO=l6h{|aWp2!*AAd#mA?0!mOD#{
zYt*Jv#KoBJ>K&di?A&*FKe&@v&`b3&lUELsYQ(WD-sY38<V~<<9~@P7hju{Uu`q-g
zsF&wxI%zaoz4KUH3G1J9=G9a&+<)e^x{7}jZrasu<1gcDlD;9cJ=V8DZ17@xSR2bm
z>>XO1F*j<jjfj=L9h9p5$2rAj^i%Wn_zPsxUIaD3btxPdeJDUe<*fGiKz3XzLEh%x
z#$5ZM8#1bbqJ=2|!^Yp-&WfG*m994l6^=Xw<|W)|C4u!^vDm2ftfZ&Fej3;@qqzZ2
z$41|h(QIYDSv4)?HS^GbZcrtf;O5W07`*7tSH<2;K*^OAcG9&5cPQ8hl(SV_{94nb
z{njv~UP1tT0nX;6RJs;h7bW-d*qqPN);cw-)bptm|38t>Nk|#`b`6IQjRU!FK)Utm
zg`c4p9I~)J1&#Dr-*le@b|+z^C7VNdu*u{f)iZsZK_p`rIV<4eP!(Eairm)Ny`nE&
z>3zxHW~?JSR?u>Y$R=5e)5{kVHe}3A$NIB-QzcejqI<%6xA0dW2^jEAO&zCu#|b2c
z>6}90I!IV?WW#6ib1)d6)5f=2U!me$e0VV}-H}kDVy1aCSWf>hNTTvZ=E#%$+rAq0
zgbZ+gMdxizuuh#stHlc6b^hnQpjF!a*>a!uaFi7d3uZvyUW##~GoF}=pgD4FuX5m;
zl)@1POsUJ{L*1^KoB>4@JW(=zy-EI_@Y!s6AFk?n)kZ?2hS=L%=OVqmX)bBNI%7S3
ziz<I1bdYT2`^4<_=QC-p9OR7Gv+?Wvp}OqiUfL7<{CbBUQ4#4A?08pju0e2ml7WX`
z1gUU72JX}eu;2)wJprsze6NUO3`)U<(zeKV<rg5cct%)dJXbL*gbu;i7xik@$==e<
z%tdUaTASewjJzw-dN-_TNjaR+Mq5%ikdI0qGwtT&@P?J{bJ~03-`{Mm78v~tYJ_fE
zWqoWjGP^Ztqi&!3=ka~k6}M&HUta4Cn(d2rIzaLI0jYV>piD8$QE1MLId}rSZLDw@
zy(AbOWHC`kQWdA%D%5ba^|Xnt0H{p9wsQl_@2Wjb!xA%y^6;M1_<`Zn3a#YwbaU!i
z4xh_A@JDZ8<J|^tkJ*71!s~U07ag$SM3lqt`t8F%PDAT{H`Je6h}cF<;}bO%)j*AN
z<rmHwhiWe3#*XF9T$fAZRs5=oOL1Wc{hPqw^KsAZ1Dp9Ty-$Tdw4N<!I>5w51KEc!
zjabC7D&P4FN0xnXH!KxpgJ7^jdc1~RRo+RmH25q>eD+wKv}=SJ2tj(RUf|W@O4|Kq
zozB&_p4FF|YYS|<Bw@f$OC6Vi7M#6!cTV8CUi2Sl&i#rsRKgEOJAq?9F|l*inM^-y
zgKU%VUXZ=%gdQn?Rlj#qH~hoF-7xEo{h)5pFCJeZLxs|rcbJXZNX)*K;IA$g`1CA@
z!%q5{H6?6W3E2ZSnBs@#u?7C5eW4fk(kjOO<8cMOEhB{rk@Ss!MSL4dzJaL{IXt<j
zH_5G59aO{UTr&`rbLAgj{H^q|ptKrVWkUOl2>bRY{C>VRj-4FO&MG+gOVLY0bZG32
z%|BcIv|So03qf!3l+s_lpwRRoFo}y1r{P)_{O+>-%Lk06>Ayb_N%Q&rdjEnpF9^s%
zio|2LGHR6JGo#Yce$Tx2Q9`lAbI`06>37e5nav$GPFySm`)0iAJT0-VXe>Ld<dmpM
zXkQo@cW|c$hQ5vRtscpSG6{9mU-puvyZdt#UnMX^@YIo`b61soC#kl4$I?FnMebKJ
zbJrv#GdvTDb>yR#@W!psW(s`L^TlqfV2AIEd`m3~>1_=U!c?WhlOZBA>Ixxt`$;Ym
zYo=EuD<7;@;*-i#FVZEd3~f4eaf^I<3mB|MTP7Jb7PTFV_YoXP9*N1Pti5IH&osZ+
zo2a;~W_u+*ky(D%J?g6xSokE}jr&WGu7vK}<kC04uX2#P+1^_`i51@(b?$8p3c1Ff
zHZHZ&CKPt@fpA=FI3tqHd`&5~%<TzhQ|v_Fd8{q2C374Y&0ub7QiQ~R#s_`sEPf*#
zJNcFplFsle%p$=_Y^C^mWfn{N^(u-IKhe@a<o)MzACVtBHk(uNFq!unTOLO%**D80
z$I+8vW-)WEr`ewI2!?{vS2goF9|8=2cHAA-4JuN_E6mHi_h8GBWXdMAb;`UEg%<KG
z_RUBSy2igz@jGw{&Py`~iB?Ojk*tO*GzFxqFBO*emL^zurh5O^0N0+HpyPH>p0>Ux
z!2*J9UgX~yW01`>@@i8q>P9`gP^3Fnn|+1o+&IcL3;m#l1#3?d!Y~!2D-_k(%otD>
zfk7#=2pw;oWGfs<cX%1?HEqY|NiXNs6yfX{UFQq?RoAmpjp?mjU0Tn2t=FnC%*7P!
z=$^K5X#Ci{AZ%Fuh!SdeL<#z3kPKI|7+7(X(ZP@SbdcZ)rLFM}r_%ypYZjILiK;cS
z_?^4+*`2G7MqJ<Pa)rX2#NQm>4Ju@ku2a4k@a~-}m>TrXdQ*8nZ0Z?uaI`8=cv$x{
zjPEV48WJ79UR#uLeNaO~m6-70BuQy0E^?Ab@$3rFv05zKbCg(CiO6_;P|nru5f{l@
zTmOYpvT{(dYQ1rj<mNXy*qFq|Zg9{iw3j}iHqiR5;K}naYr+@q-e7w;6|?6<z?n;(
z9h7J!R|9bpu*ds2&;4^s*bFskQAv$p`CZv5iX@j<O{rM+j-CIW`Hw=~omP%JGsGx<
zjf`6LuLJgN+kId2Ty2fx-fC#hBX^=T9a2CO{PfQ|*dJKq%XUUHkL2~p_vy0P9_Hrm
zlaYEe*FS64c!djwDB@QLL6YCI2g4C(@S{=oby?XC^|qAGiwWfyq;VndL;b{V%R<-&
zN_I{roQDYqx)VmLwa0rGbn1Q%ek`HuSb8WFb&`=uk$LB*19Tb^nC1wmT)VR}!k_u>
zD#Y&*@64G6c5I)YJ6rPo97z!<9BfEof8WzZh$21(bgluxrQt6$XP(|*chuDHeDkjp
z@QUxwht)ro{{=<;g+MtCEfoNqRMX_A*6EF^?blQH2|WsTk8f6=T`5I$bXrciMLz}k
zu3c)~a<^&zbh@iQmf%_hd|B24(5M&|^%u!23%G0v(3(OHG*<66_+*%C2IY1=Ok9>7
z<b-@B8iP$u$bL)<ZUwvTB!{-_5#)pkW)ZW`e_6L_BNQ03Fx7FA%bl<enx>`6rvoA*
zk6)V{{Cc^cZy^hxPBeXlMS{}$%=mBgz^1jsfUz;x%Q8z`|JtY}SW=*ObC%T80|ZnK
zKVKJzVXDHLnSRt&@Ngesfg3MMV4-}e)X9_iwre5uSJoi-?aHy}syf4c2%)mX)5I0|
zShykE+S;Kw!U_-i=Nl_K9m3OKn)|L?rm<b>f*)H-!LD$GOGN8}c;cvca8oqd!dqx?
z#NtiFY(v3NU*=COu88psM^k~!DDC1IkCB>K0UTE;X}S2?GOTCQrCmAiu8D*bJxE!_
zkef|7DVLDPJEACQKL@GSTdeOZ7oin8_$xU6jtToEI$I~w7B)A>{NBwC@h=Dy@wt*5
z%g;XgeU15hJ!@e$Q+N1vZP>$x%)>|hr4VL`4NByI@A#vR$HU(DtRhA)sQW*>VXbEB
zQ0yD})BOoq+tKaYG56-C;jOAT+UiD`WzNWh;wMp1r8Ss5VeDpunlk1s0_HASjnqAp
zZlQRt!~O1DO~_4wI*J=k*sAW#VXcC=HzBI?#;ugmATy=PN;BG|G)|U%uvBb7mx^#m
zuw6xKSDS3efF_drhrcwdVc;%8d{=V^mu4tw0r}f4zlo|Go!?bU!iT+B1$c?Wm>K2t
z;Y(uWJP}x2C>o2|>o?$gD1NJ7mD>1tTr#0o$LWqwUg=d$$KY8NiJZww?AWi{vMv3g
z>=72O^NOFwgRUDIbDTjOZ@Nr)l~LFXX}ow7kL!F4ViAc~-R684PF?aw$X1q&S0jDs
zZ;&>J>cI>ObZjDr_9%^`Dxv{nA_A;JG~U86S8F(WP*)4O#a0Pzm{}rivUuRy6{&_Y
zwOd{>VAroXYemJVp<{ALe#UQEuIv!q7^mZFlMz(5*U<tzaesc7t%Egsb>)I4!wY2+
zFrC}h<D0YObhy_aw1DvLw=eoy>YQ%dEg1s=cvBYOst@bO<%U^6(-*=_?}oR{%*7|9
zuh%XHolkgLQ5*7Vas0s_oA-9AY&nIDHoltg`2_M?hM-f*aye#$DX9pNBL;CH7}*}h
ze%?6SB2RT3-Fy5v$OIF&qqLlmAO@f8m4t$$=GHW;v%J1HxHD^2yM(LQ_nQU6>U`n2
zn*&o>edzlK*DFv0QIc=I@zE60wQ^pLwYjHKz8rdU3xJ2}>_g?m2_eP!K+ST9XukO7
zjsa?Dum&k9S&d?4jG5xjCnxvm%3{WcP3`ZQ2Nh}oB|jh4lLU&K|44moKzl-w{6|j<
zw&9l2B;wtlW~+y1Je%iUS(`oUyt6zol{?n?=VbIEzX*Bt(dlc4!N=$?@O^VA&yY3G
zotw*f?s|IP$Z}`CoB#Pn)!t5Ckq;C_y!0E7FnBCN&O&6mqX^{O?W?7KVRzF*4RUv7
z+CtJ)e2ZoO^zoF`{RN3f)%f#YCs=rH=ilkFRLHu+Tpq>0UH8uP{5x8t!{0WB*1w=g
zhwovFf06QpO{>-O)nKoErt+#mAbd<35~$uc1s)sjm}hmTK-M^=99Ds5hfek;##;oS
z%5}1|9WQ|{klDdL_nt7*Cx`oAexq8onlK;VTq(Y3wZ>`5<bA8fsyPNLXtWK4s!)#J
z))<FYAU@@XOaZ_sn}dNX0(eD0P@3{LEd|~Ig%Yo7ASY#K1&_y=!tLcar~*Jc#piEe
z>O;JM_Hf}U*R4nTFt`(FLxFD)?A=}9nZe(<DbuHI%B;I>yc8=z)Dg>#K?ges#a=GE
z=aJ%*=Y-R%2JrKBB(Gb~lU#}ENdg0w9%V|bAaji%A(>JdTZZ1rxPyxX{P;VGAv+;p
z<z-RsiIb<wCQyHp2`0SC@_wbX<P$G-`8lIDuyTV;^tJ7iR)g=v)G=qVwWH?C1i#8o
zDm=)Jbxz|yPWmF(bT6IIs|PTf4J|O{AWd5{WCiBtrtJ<U@E=&kGeN&6ddq7Pw|H&w
zn}YAw=y%+v=HbuhvwCA>a)VMgiI4P1JZ8=wOUreVnra5B(&rPah@BzZKc8+SIs58b
zUYN?h{-hbkcX>RDo>6C}fA(DeGF39qYvF}+&oa@PYN*{^_nt03-a}JNzTk)&G;g#e
zxa(LJbI$@=_n6)2!^IHB!69(WjIV#_0G*0aPD>52l5TIMhvwh0Y|UYc$^t(lW1pcD
zX*2qfGuN##a>bke?%mouj!eqeAQMiNy=vw1ac(n~gBvA^0aB;IDrVe6J-Qhev^?um
z(orx?O<hyqHTz{N<JtqmLE~Na!P|yIPUEwLfD01v_SxH2cEe=Tgt#w=Q&yXn{fgw*
zWk(nP1;tM%iT!L59ez;t7}~kRjb_UIb93ced6Y=?!7$j4;a|`**^3yDm*phAD{Tzy
z*YoeP@oze$RWnKZ;67V*ysIkwVJ=Z_rH&IC$)5vOkNY_j>&;lDk-powB-`Xj_;2K*
z4Y(z9`k?m1j8kN}qLGiGgC5ii+q~;sSe5SIET_2Tm=P}KIk4!$CzVanSbURnC2ws=
zufKpvWahPcM7wOSRnI!9;^qEi?7;<Jc*J}zSN~7aL=7R<37UFDsZEstOso)OCKC~R
z)z3%TO!xC4S?pI}?|K2F2B~s|mm-V<Ih?G=bNxoD;0+8z=NvaZ$W_$Y3uIhXjC;ld
zoL`Eu=tB*MxtaSnkApMJH=M4?_(g_RQdw(voF1Kf)pezu6!F?6z&%l1F<MzQxTpGB
z&!xNF3^Xaz*M8|Ozhy&p1C;{C=lz9UXSS{;&YgRFbt-TPOPjv;R7_R8@#3^_nn$Ur
zi3%~`T37BJ=25wHio~SRZ*{~!LBn~%WtL<9Ce`Z0odaRE!g3GH(tJ5?lDROdxe8nz
z%einT<z#Pb$<n;k_<upU23uJBcTDJ40FC>nVcadSUtiD|GHz*#?s*yn5b}hZ4ep>c
z7OX5wzhFfIw+<E*VUMC+(IlZbHBRk|MUWK&>V-*t?fV>%BTA~iT1>#xx_X0DUSf{Q
zMYilt4`N0*J2O)pwEIOKgkEPVJ(Vo}dsNkyWw74J3;FE^IT@qH7F9QK*Z}YtcW$!M
z+C?L0Csceq<G!zrySaXn@?zZvR9W|?KF-nJglK3qNcS}`YCG|xxY6ecz;M>}=~`Vi
zcY@^_BDL45*&%En&&3L;XcYDA7k11hH-T!#4*A(y7+w)x<-&JSozOwGz=OrZEbu^G
z1P@4OP-cL%20;~{`@eA3=Wm<~g8C%5qA)=5I@LnJZ=iy)6v5R|ne76Mn*N$I8P7qP
z-GH37fW`EGd2I@*oGt;tSJ{BH7Wn2_m_WG=EWPVm`S&u>y1JT6&z%KYuM9cu`t2_V
zjG9_p<hSuG{@J2@R`>U2%`S1jrt`CDpx9~{X0+g|wiDEtlLJGQgKKVRmeXtMTgDWN
zLD|yFKEQodWv?><w6sH!1^8c3Y02F-v)f->nKcEzSAab6;3azrQ&A<-7%JPEvEujE
zYCwtTdyN(8f_r)4c5PaEtc13}R5!a9rWc>W?)>0tdWIP`cY7Im8#yzV-a}IsSI7o{
zS_~sFR!-_xO*Jfs6<x1+eyn}`DGZ+-tTwMBG(P&_QNhOdkY#t=R>(pUPuPco$O2g5
zz2KIi^)ij_tko%sL(;V&5i&)JTQQVDMq1T|?W%i}cA-;Qn5Ed$OKC}8s|qu7BomFa
zz$e(at)FGp{N8W4*s4eY#R9%5By<i}%bzu+hXX9uMxyF)*(r%dCN|6LO*~jy>*hwj
z5h)S%Q8<X9C6CK`Z=l;ED!5cGAl-Mw_efPd4e$_7cNAl!Q_X}N()8Ns-9p_{qDj$0
zOcXtzXscK6!DC1hmfw?(BX04ZcsTILsPpBOYvK3*Uc0cuKs$#D_F4wU=f4nO7IM<!
zbc;gjCmXDCTq{e9=M1B3g8iA=i_P$WlH~qH4|DGMm>Jlf{q0>uM<e#Zk<-kLXYwg`
zy=<&W*n4+%9?6{z@cl7^t$7-=Qgp2()IyweeN@b!t|iaLk-te78CE>&&z-3ARb|i-
z-_7!B_0-Xa`^6r5@?_SZE`1PuL}~k3wdppED(Kd_UVchk90X?8fm{x=1uSEs2of02
z0O1vBz493e%A03&etxY5)#S4IrbFtu-s~`0D#TPkEIh8Qx>`Kv_Qq~48RzI5Hbxuu
z2}-w2wxwW(b7%%#QMD_OwqVmg)pkB$D1+g*^!A|Zo4JPfCWX-8_%AU|sQv|fjq}c>
z{Bbm`<=VtSF+sDzvq7eFhN%MnoQ<YWHB??K(R9P=IbvE}O4LoKtNn(P<=9-pO-(9f
z*D!lFX=|nt7J}k%r(lgnqU<Us4NG9+E0Vt&F<Y?cl)++9ktcO`ZM||Fos7}kv>BJq
z>X}TBT+ZTMC@YAWsS!P|S_#h15g^R)^(=|ZaE8hqpJMBJR%tDF-E7Gu@6s|ihu(7Q
zzHgv+VK_ZTz8?LOAdJ3Z+vNEamJoGF6LRwF%|^WR5<cx0g~-MEMX#0PX1Tz5jxf7F
z;z`!VEu-jkjjd5Mn1Gx7y67xn0Zd2h<LqT8*`9(<s6h+x>#E(lgMJUb?ZI?MFJTOt
z2Dc;H>XZzSy|}zApF&y2%N3)L=C8;=K~|vU0eCdyd6F>nGHicq4+`Qy5{vx5r-7vc
zaF7CutH0qaJyjTS=5Nj8Z`k@@J_D-$+W_Igd%>#EiJt(HOa-GT9XP7Z3Rrke6F(8{
z1Dib%BqczDm2u{&A;!A@!OpD|4?j}TzaTfi`O;5w5KG4vou^2PI61ZNtD^}Mtm_Rf
zU;1gEeC~_jG-dxXsu|#UjzFeIZl>&mE@Q~dF9^^^5HLGtWmL&k56Ra`hhx1iUOrOP
zUl@OjPbWl=yJbN>ZW_R-bRq@vOl^u_S>4U}8PG6o_0r2rhkcez>^oR9Qns5Z>@f{@
zu-@c0P-x!8x=l_mq_N7?!CF*oaNKbR;3M?5ozh1wz`pdYYRsCqBg_^+Q6Tp5&GsiU
z^EvJ_4*Q7!?5>>r?46mU*A~8uq3?0650p+>vu60p*xc7KyAL(tT5aGcqi^FX-skqG
zY2h|DSD7AiVb?l!a*h3|ywO*IedwcOQ=3suQ)VVZtEcn%1H(e5LZ<v@#Qw`hdGSgJ
zc81D$0Y3nFhOG#@IzBl!Dy}yv8o(zy?{g5fk)>jnpYw&Jzr(U+K7W?r5bGz_!I%_^
zSmSRux?Nc;JsxzcVn<QZCV?1yL+Ca4Ji*~0?I}Z<K+TX^zr&Vg4+%YfzyKi@<nT&o
z;}&cEaZ)`}*hRVrzj%vhV2i1XU24%i$93L5qmm*}IIDMf+tJt>>x;HRXqO+GE#fmX
z;itW!jJ!VYMfI4+hf<;yjqjom77dAJ%V1sPP~94{#;Q;CgS+R)RJI&eN)7wv_<2t7
z(<AkBQViokjUicIwVZ|-@uq2HGg;jrkGe7D<(5!YYYeSzQbXHlMfHY2COoQf6>ps}
zKFrjuX&{+!5vtI^j`mI8P;CZpZn`_73R)v6=_yr!UlegV1v+(@jpv%o3}Tc<>RTEx
z5<@khY}K`TbFQM6tGmjE-V=hPNUTx+%P#sO)voH-t#sxZ75R?7MoYgcZeJC2+_)Sc
zBU4R3WxOo@zU+rfo0oQl2mC8SqCzV(D681sFLHd{SZwJrF$eSJ*0)SzH}CkGsPA)?
z==+J$4V6!>n5ao+TWnhIXRa6!r3_PgYp_V}4k35dhvN3Geoh!%VGd^j4TW?_`r%@G
zu8wt!crQVC3)WAS2|v4qkt-pJvyAfz1&~P6VFvv+5~(jxqb6obfppGGk)(l9s(1oP
z1|1Jvb2Qc;apyG{`H3yt#!hMsW4#=!1!{fy^;)lF8IVxAaji9YdjFdl>YLQADmvM1
z+jCD+e&+E#`N8#IIRT@UbTQ9X%PIVPXdu{rm#zd4zBgI7(LzOW>=<*T5_!4GiirHF
z<YY@_x6+qUKt*cA4nLU!+=toxV;7+*q&9gi#r&3Y7Robl_|>Gfw;|mAKR0*5tDLOA
z>jpGcH8pI<%G%Hh$Z-F&a^vFv$GT&|rcgFZ6@#Y#Oc<}xg=D!@T;Q1$7)d+LR#pY=
zwG#R_o*Tw1+lZB_OaRo57;nEVc=UY1i$jLjBf?KCAC>keEXLtj%5Mn5b2CWTy1_97
zAGLl$cPNTiHf5+^7r|(KhF>h5)vpxyPgc55K|%IClPe=)?LkH7hB|Y7zcTL`vOj(;
zzeJuD7#}5OCy5*~%*~p?PU;k%ij5FB(RBWqsIZdPd<@jZKwt1(g;7bXTnmZD*!v((
z%bHka-Ll~oO+NV>DAMZ7XUlbiq(^)in0BX&kSaA@U``g%!#A=t>Xgm$G&IIZ?Dw{d
zN=AmyuDYZEnPcgcc2@N+TKQKel%3_5q;Q=I5v0FHvF$c@6F3`cB9EGRS&dE_+_R_i
ziYkvf#d57KWmY3(!dGSs*fZ&_ND^<|`O+UtBuQ8=sagpgQm;YO-0pQYd~f2ODM)AJ
zTd)wYD2ljbEEpos5##e%u`O|cd4Z|zdcTa$y8n0+e90O!rc!#u7SkABd{*iwu>qzw
zlydr7de)0W8N6rf>7vfWrm3NGNEXa1a&~R{{y<=WFHCjt(%5yWvzNQJ@J3HmnnErN
z=eT&uIAp!}koWnyc}IouQ=eP3UDdNZH;cWX0pqh(|ALaM?=xOoFx()zapuwPQ^T0^
zQhxygy~3u2QC6l~>J;fF?)<0d{)%h!&ecC7nMjE>*~C~s@fXQ=1Zo$sX*cSf-YGuj
z{b>=zTff%KoAjX$D>L7c6BN>9;DEo^W^vHNXPM2>AWw~ueC1gTG0P59c%_YbcQPny
zdCzPxG&i_Ae55(2hOs>1J?W$%3p#6Vr4l#&WG`djL)QC-^Kzct=6p6(t7^zD{>r>v
z@nu53i#XGQXgh-cT8F5mM^%gXbE!irs%h@HDh~&G7B`?8J{6GHzWa)dX?2z2(;yKP
z()z-PiH&OKv>(gjvTrnG4?+uZffq6|=Pg1kJ~IR=F+`Kz5U4(zT@a|PVQS$yhWsiY
zm3Ul^XI^?!*jP}q--jSpJ<r8c*8UD!+Sn|lJ;+@_QsT?9g+Scb%*!zqQg=0cPh5gq
z(|Q-)T$ycjkABhuNsnA7RoVD^q<qZMB1(7CGZ-;QOQfd>6c-bdUv(bWDS+tj@LPr!
zlMU+v+azwDSNcV`Y7vsZ0ye)T_$mx-9Sg4=r&yFde5pDzSs%mMy8h{QgS%=Q+Si`R
zgVke}>;u#*GA&U6mF{pv0%!VKOeNf>N~>TPk4B{tu#$Rs&+G08#pFb$2qyyGWJn~y
z<Z#QAuQJ#e=$D8jdT`Z5@b--Qm@A&eb(MlgnImUYYIC_B!Aq~&PVA=KHYgwSuU?_Z
z!CBdCS3uvZzBWyl_KpL8+*65JJhzd`w~;Iv3{SnB!qF*T0SCF2SmC)_JTort=K840
zp|qyizU#Fk^bUh}N6d!siT+e?H!VTKe0Pdk2W%p>Wx0x-m=m8)F{>fw+*PwA-mfLm
zHw=!cSl$sd7BsIGUykQ<O)8#EHb=-#q!oL29+9$&xh?MTMT~cgg#@uerqMhG<qTtX
zvw0H?x-RF{Z971B7=v_$)1R5>>}ybiUVyJFaBor3%TEx6ERsZXXne3o0EsM4im)4a
zGpYpF3P_qpL%5%@>ZbwQuh=6%fXat48FD*&M~!ym?LHk-h}oohn<_+ADFMYaVE06#
zn-T;8>^$EqJX0u1iq%U}!}$%I)%3b|zE{Map}H_|Fu|^3Adn~n`!Aq;^tT5CIM4%1
zQbKDNUwnK>p@Wn1Vb!CM4qlrt8XDySMRvz^Ml0QPbWC*sQFvhB;gRqmV^|vxx{CJm
zdc3k_pVH1$|0i-NO8ABF%jmEpa`bFf6@_kseY74ZG5xAz-|5&pwdXgw{Y>PY<fg6r
zoL%iN1c#UUH;$1ztiD0adS6h#5X~pa{QrUuhiZRDkDpfI<rTfze|}E}RMhIz2Wfzd
zS4HUiLh)~EMboaG!P%0O5nv?zLGSCpRK2<Co8b|G#Fz=W+<S^Gf0`%PJ^lra8yhh9
zf9g=|uX-nQoB^&ax=dcN34C*XJz)_Guf^IjbxXfa=fzI!n!K4U@I9&vB=r{zR4diA
zDLP$1U0O{}E4|z1&)1{+I&JYwP4k52GBM7j$=AD49ZsOFi9XZeXk`m98)5a@%pOC4
zx#E5<@cA1q>LR=@(w_u#pt5kq8>*o5rv0(@Acf2zdfD_Bm>sKPz8!;9>056#_UtmX
znPv-E&73wUmc?MV-mC%nwaa_~$SjO{2Q>z2Ynd!qT3`FYgy}qR1Oja?lv8ysVt(x$
z%|+QNyl?gAO!N`whqE?|1a^L;$s@%o`7V>Y8k?gs3&oJse?ePL?H3RO&u?hT+iE+W
z{WkNOC&F@N&6rH9&VIZd6GM~Oe3?8Mx(1h#$sMZsU7iphc!sz76NJr96|KH+_E9eX
z)LGgpWAwf-4xgO;BK7IDe;zoWc<{PWiw{!n`*^IhR39i`wJJ>W)9lf4oLGCiI04oM
z%ds7*iWV@%cwwH1o)0ROyI?HZq4DQiV|_5rF2Zc%x7~_;k7mT5(FldH_K3GRd<7K4
z4KA%PPt@F6-gTkxsRPgxDRHi(A?WTo?(9OT(5fUy3ux^pGrI}wx+r2T@zcwcm$Et9
zLNzev*IO2Yx2ikP?*=~b)i6AI2oq+h&AV?e(wu(@xvcV(iDK-JNPGnz<uSROF$ZnB
zfM|JD8?JRaJ?G~&fi=-H+R)r|ZaO(JzUDw%uY<Mvez;`HETgN%vECeVBomjBFkj@A
zovS;0^F!EB<H;4x!NvtNrfEtg9Md19_K$61b)I>tshhc&muMXclF@6JF!0H~h`eKw
z!LELdfO`|xvD(6zpK9@3X$4x&K_&&Nanul0jI^uRDW^_t;kgRbTV@dWaw`xknX8Ut
z8C$Y%*n-LLi<exgxlC-n2voL;X<-2dtMA--OOSPDf70;M6lfE3dn{0Z)e*zEqRWn+
zPyyolQg*M$J|0RMqv%LS1vlx<GpC1?f`LI~u0!6D3{r^KbxYutVyKmU2JV-1ntY7N
z;(9B3@(;fJ`*LC4n|Pms4P7mJAHpeqO9i+ua{!FOU2(zFo-u;Vfoki-LDW>C0+){D
zH1Uy?szsL510B|p4t-SP)nC%B!{nm3l<3AUMj|n^n^X(KQd#ldQOuAQ&Ez^&JTv`Q
zXLaxqutDs{F_S=AGh5-TtgX^gFXbYph`*LjC;FU2jU&mTv(<STp@@*yc>WgQy4$s2
zX`Aocx=#3!uUnjtNBM8AN*+)ozbeHN<vVJ}`5>aa|AJWRIOQ<Ry{`|jH^^e%*+yj#
z;nU!)X~sMb1env(h?B97$5InC{Z?~;g5r^h=?v;qlC0fy>n(f|H+(R9BDCj}vxtre
z?l)5*?j}7$6*ZrGt^SpkfxP^Q!qz$WQMX&gme+5eoKUBRsgcNfp>6NkqjIGNn)&ia
zAUBL>h3I!x&pm9VO<7c_0lx6qtr5PQSQS?ZCXlmN-j88O%9<4)L%yx<%Tl@tRLAx9
zu&Dx~Vm;8+n!A+8j=Cwl8II&w4iTx$jmcaVEjfE8%|HFR_sJA&S+xHp@bKREf{08d
zY+)7T+~1OnN;LByS%(|ku*EyFQT$3D+o=m-zmR4BczhEg0t=(w$A3ZA#H^3)`ZMs`
zfz1)zMVpzJFKS&~GV!QGvQ^f(wW+TL`VJ05stfHmd^a(-`W(k^2yVm%l(0R##?^Nh
zN)7wCd8JG>MR6e)0Glo!lwa^_wvHM67gVCZz5hM?r55PMyG4TMOD%`MfghPb&Ju)p
z%v2{`ET0#J_J|VxuDXjfB|?q?w@=lJ!Ls+&XX`SbAw8m??ub8vdLx}bD--tbe-G79
z#ui=QMsV2wFHHd~!u&(uN&8II{Qo^wz0arxpl96WY|K6N6AR3{?{7+a=~MXmkyi&r
z5_5q!0AK?hl7AP+08`3)-@P|&)GF`e`>tKxaZv4qWE<f9^(;869G)Z?48D_9a}!B2
zv#nCxbmlLwzmjmt`AW1fZ8iy-b@QmwRAlB}ZNGqrWA1N@@<6Z8x9ep=u4B@_WP7}r
zPbYdFbMP2`lqk^Gis!GfWYXB0N?@&7*SFN~nonR6p?@@n`@-T-wg-1KK#G}J07du?
z7DDM!x5k$%(pchlZgzr`f46LO#1FwkWCMI!?nU{%+##u>dX*2+8Cj;0YJS-l$r}{(
zo313s@NJXPB<lHs_sR%?5Dk*%AYHtG$Ac}ZP%}W)a_-n2C?P-xoD>a<M9p$}JNVDb
zrTG=t>d-f3`?zp*uJ=qnKAy$;w*3gfP+Y4I@}TQa(+N1UzE>;@yZwE{(lgps%Dh5e
zf*~d(3&o=F1C?tU(%F&sa{y^e-eaQ2N_xT(#Zzlxdoox?MpODpaZ`r4uoX7FNmgST
z+}$sCuCq&4Et@_EO|Vd~_??M+&vHs^Qwvzapu1GzCar6a+uxTd++1=Pd7W421*8{Z
zzmT8oW3boqa`+LtNS!|5aD(^Yd{dFYZ5H5S?xFMgonXe!JDanNuOomPyST^zJ3|A{
z485qElKbSmznHe+4^{}atta2tXBvJ%|IV!zx_S$JH<Gzy$#a^Xk5$&~tJ3bv+Xh3b
z@{<y+Icmxq^0#FrPE;AVQRnWKS7hEI8kn)Lacb6_Pz^!*F+`9rYL|wjLb!l(n9)i$
z#tUB8z&*nG;nx>eo>dX^4U3K|n&^T)?c3TuC48^@?da>a>{r5bwS+~E&ItD<#WP&1
zN?nD|-fCwY(rXQ<H3-}@070uZ4U!>5B_yci^FTL*lOnT(0|8D{28i;b%<*1bZD*OJ
zTf^NdHbOlQz=a(<Jrpq|MqI9nsC!mo+Vl0=yAEP(OCV&?>re9Lp=|YO>F@rrj-psu
zoLihEc^hah4xHNis+g*S)U)iGsi=hjTfID(@DU%)584XC+JiB;d-3HH^dSrvas%me
zZB(T3>WUQ{99MZF4t$y)DX=3eJE5g*J!5(2lzuwpsdq{f%{XK&p}x?!UvQwlliT&y
ztC1w7r76^MJdD^CxXA64JQSkoaq1-XE3t@2EH476af6A8$u2b<F;gHNHdb&iiN1;f
z)l2ok$qdarDwl+5pWZ2Rn=N}aQP>;}zg|oF9mFd5+*eft#uzUdybG&IBGRWE9#_P!
z$3x+Qp+iAl!WlBo0G5F%i{O(5%AP#68`&;maQu{%V3hfr0u?SM%eRQkR1LkO1?%CU
zC2l4I<eW%;3yZj(!=^>2t|^^g+%2TI(TWU_eG&4<jY8Lvy1#hmjX_*q>UWO5vR5UE
zD{=4ny*}p4ETo1?2K`+O=Qk2qMn$g{{fUliWm5VG=md6IFYA3K+y^`+XF4D8@C^5O
zMe#A*>wKX6%4}<VwopB|ZOucp8&z(M{5ggCg=4Hrd6Jv8b29y6q;DVS)#w_yCA`mE
zvHI?OTps0f-&AL5>b|q+YXjHIj~`Vd+b>%}0P*?D_tjj8&#eqegO^@qip8+fy*hzR
z0iOWu?JB3`Kko*r!QQ4Fhd_#)$B-t~b|3}&jGT4gd~#lR{Gi-LX5YLTkzyXix(1hU
zFn<6T*ysMTwMd2(;9Z!(w!mNV^?&p%Fs5Sw2w8!@sO;Z>_<v|@?RQS?8@B<myZ-+#
zIR-Kqz#c5n3uGz(i0+<efj3O2U+!`_qAO%;+Qc`6zkynBJhj9V25aS>l;%AUI~qm*
z-n}}UOWVUg7GM<8SkAGnnksJ*U~>>*<q8y(bVxJ{QtJtcnV_qc_I7b%OT;zH$gBll
zCyBJtK~JX=GKaHJOo|bkw*9Qf<;G5N--ZsW7x~QHIx3V~A1^B_M>qbwUmWfl=9yRN
zEA2lXcZMrIVejJpSWK>Gn3;V1P}h$Pap^{5Lz&_U)Img|mOUM09K9|H`wd7-yVeIk
z->&O@%*crb!(hOeRiGTiNkI%rCL%gm4TTwP#H2(EL;^cF6q0617Xib9&&?WHGZxBk
zQTMAC8MhyowuJ?RXN+xFzP+|F&vL2cYFG$moX(UyQdJ>NRGF6fjAVP=8hnQ}r~hO>
zDxvP16;dB<!bv$R2-EC}NNc{o=g`Psgi!&P8Cn^&$<voXlsSi!;N7eG`KsRN;e+1D
zAFkwgU+mEB?Rsv)`&~KV7HBm~>t+S}aPTNb{IcZ*1%t~9ce;SdYk75er|G<IZBMzT
zy_W+O1P2^NhrAM1MCW@JzWBfgW4>;)AO`!da`(Gwm&~;-q>IkMB2YSA>y@~~ujNjH
z;zdI{mVmcKj60J_vf5;HObc(!BvZOvp$TIuMSu;ZPN)0LOhMZ8Y*UF&zhmsd=|@|b
zVKZi3S=Rxn6iyG?kXR@x3FCXxyFx+31+D$kc6k5EAng8{-7Tr#V39UFt_@`W5pa=T
z4F8aVi9awyfSA(boa|AE0SX;k@nKcB_+3k5q5*7UmbkGdodEPvb0tNXpb{mcVxKFC
zW|MOY(1+*7<75}C9>|uLR1y8U$71cX$<I#+aY+JXL%P<&y)El6msw(_&V78&6-^E7
zu^cpGWbU^p-zdBtfhX9rfF}HIyd!;Ja`cQ+S_ds1N<6%NmO*S(QbQQ9w<!~7AJIM~
zY2n;{=Pc+7&%Kt{PTQy!vmSl$$A$h1pOJf=E4-y~P3~+}I73MR8!=<d)+WL!f7GYZ
z#8fF=B!6tw40z>XW3sOOacFg4tbNkSL|;G_ck?=l32!V^d@u#{MTCGur9P?y`jt+x
z5kF_|W6UAB&s2O|w+=wjj2A1FID!nPrE@`H{WvDEfGi$GkZL-R$bVA{OOU_i()#Qk
zvGb7Ju|hF4Hn(=Fy(x@qW`aQ#nb)(j_FfSiZ!E`4>EdpqegH14UY)7)#xD0nlD#r5
zOE6-B70TDYu)C&tZM5N_0d|^V2vIl5(oc+M%^+Kzk+Z4Feu=S3z#gr%STG8(u2D{A
z2ZvXZvFa}Wj8(e_AB}HV8=Js70H~kX+>YBcaJ|`X*gqe$@TeLvk#Kt3?Hg<(I!+8s
zn4mA5A<ZxBtM{pp<{__(l86{V!Wbt~SQoy0A5Sf*f?gBJYjRGy24MG!2;guMkH0Q*
zp!o71hfkt{y~F+&`vT8M0|o#nYw*=Jj0icwH5W*J=TN3%%?{Q+lYfyKO8}(-lHLE!
z72JTs74`3<0lb%=^}llw0sCU1k;!lB=Od|qK`WJ|j9zxf=KJ+3{>TPir-$;Amy&W_
zFFZ)idq8&QP{uXt1s%ZV)5brLQJVBq2&}nV(2EbcHV|WFlbS8-u!DVEIh1=H;#jw#
zGttwmYz)5vEl-+=uowHMye3jKMGXA<%6#58;1Eh9LFP9tgY{2Fn2hJ2)(AcJ8!76V
zdJfr{ZQ1Bld^beBF`Yq(RxZx@wJ4(fljg^}^<$QVs*==PFRc+1sf>sI{&4t8Wp0|Q
zwqws$Qid*`dmzYgVxtYhC1OwwKitBw@UbjqQZyZL!ru(o#a0h&dWCDjntQe$5Ce|E
zQR~d7@Cj*;#C=y=pK$i5QlZ&d+!So^{oLv7y@D~B8xT@me!r|&(tLG3w4ei8&T2Y`
zG8DUBt|qj8zLumcQ(NN`J(22n1RdO!6w5Mq(=^@T+G~}e$jn2Dw<^C6>P@^joQ|v{
zU_JfRcYNX8b}FCLeF*l>iWk&J%buaI7GdqG5g*DBCuvKa(IIq)9&p;&NFzqGf?`}`
zRPRx0#oMJTCwX}W%jJBzIKni!P2=jDfl}iks{d$cEryS6MxtTZy(w(Zq6>mI<*Cda
z8bt=vUC72Tml`J}eJJ2?#~aJng9<~AT#QtxAT)0l%GPY48ic9Qtv68lF+s<7Iz<QG
zKo>BA%aQO$0(CHXYYzrp2Avuh{5B4S2SZ9xb`U1XT2g<|=CeZc{Z)NRWR|Uzy<y|r
zy=RWkCg^UALfNzg0ppV!5+pj7MKlnTApWX=PoN{_?_1_R0bPvcbTh71f5%D^MSa24
ztC`Gljjo*>yzFq8%K3uxNq(HDLYqjT^>d7@jmx~{lj6}jyCMU37C!ImCXr1QJk+&u
z%~{r%SB_qsTf%Ct2ktKU*0h<8SfxM?%15?%aDvt-iK5V{^Wp@#7X!pqh1QZADCs7%
z6AjX&eb!O18oShMxoqs!+sr*0B#(&zT}FXv8Srx0j<Kw?Tm@SYDT%mVq$OdLR12vn
zKWd5=6i=5`<I9!zCKFT9N(3Cc_&q|Hc^r<-45nETy+SmGnB6R=uM^kEiXz8Pej(Q*
zUT4XE=u6MO!<xL(7BDQREX0^6f&{&EbLbDfNn&=O#m<^F>rgC*y}}N5^fPc)p#ilz
zbV$mgV0Jo;2`H8TI0*;9%^YqSWS!co+e5nWVFQc>V4XySW2$Okmhl0xxdi1TZF{$(
z0kz(CuSh1W^WmIR9k=QNL&dmqU(_$><fj6gZrncp$xew%NIoAG0e<|)ZEnXDD7)Ca
ziWOeb`R~3YitWE641kshlF1PRM2i2{|M;Ik7>IHKVR;@Fj6=rT1E-Wu{bwr=xX=0t
zF!)uW`~Oenmvsif|6yb{Yrx*uME`Tz8qdsd7tQ*q8|ZVRtK+`}p2;}K{rM`(`fkN2
zDuR=})Kn^`UOY;N)wpBLYgix>{WeQh<xXTkb)L^q+>OEbFw~!X7TRSm$9_hQy1Ww$
zhfZ7f_1?yJLa}~FszcZia}!A)n3j`&3w|x|o$}dWRVm6;wlSy12GM(k$f&V$<JR;!
zM?&Fp0rT_62QH5L6$$<v5_eu@>jnv==*E1r)%}w%CS#owi{*U?5&>;FbMyHZDDK+$
zUAnYGPZ!}ytMpg%wLu229@qn?KYNe$ac_{3*3td8$oOd2e}Y2OWxp-VCNMj8>RibT
zy8okd(;9q1Bwr#OPBf5Iyd&}_@2!x*nIAtVAV#?~ZpYy7<xI!lYf=VQ448d)!-%Pd
zsbcM@O#`+S__Vq81S(qiGt*1Wau(;Y2x8kmp@Thp{*u#e*mDgf?RS@J*mzuq`dRd;
zp0~7~jl>gFT=K_o>cS}<G*4e`PQA9MR~H_<7c@TL9z~iBMy@aLd^b0<2Lu8Y%9!9)
zp8@d}`;0=9#i3%qDfGVRvcyv><H8Dxif=C7XZ&empJQ{Q=+%uq+U!HtX^?9%-iG0x
z>9R$rubqceuHp>1nT2##oE23}>o-y`JPNiSzUNSo+oLfW*lK1K_*P>%SAnf9t$GOj
zBw<vSd`4~h2hr2GBD>)W_x#;RebnZT$JvTo$Gl35b}s_6!_2IY1w3n`b-$GDu}JOT
z-ja0K64|BLy3*jR29^fGF^+zc@EeCDdI43R4~s416o}MApL8CJdP2m6Jw&S*YzqQs
zZj6F-%{UBeQpKjgz56Jur|BzSNhr14`BMJ$fGlRl*E`sTjXgnQ_?v!nH`5J)@ns}v
zSxH%IwmrW#&l^rUg<w^x6CEVdot_`_NpQ5@0PZ~rhusVynQl0cabN-sWZe8}STKQo
z-Cpi8Sx7|=J}xlTVd6nbp7=Z8?W%Rj*@mFK_K(842$Lg}A)WGvHr-NN_sY648;5Cy
zyF)QWUzK;ZF39&Cvo^kYE&5OwR8e3dq{L;$GRE}DnRub=;hVH&P#liwTP&+tl=hG$
zHpbqbub1yb;TqcLTP+iW-f`!}_XuO=UH3@(D;*Ut0>5w&l6=>qNxe*XX2V1QK`DV-
zQIb#;$=_q(z%OQo<ssWu;uRCi9v>s!=xT3|?MYydROO8ytxPS=>#y|}FFYzQ29DAq
zd5_&B(TN7zbB<zsL@n)t3}C-D9aHIN#Lz364u`eyYY(?vzDxPN49%Wzxc=)4%*gj?
zkP(PaOhZJYJW}W9lvfJY7|B<(tkK(g@nK^)#LY2oUudo?c%j_5c+FII0G@z71{Xr{
zp^G=lS$-^7U-|)M;e-}vQb0O~Ox=n%p*_YlpNCmNlVZvOzTLLPf+UHXnS3F<gQGxZ
z>E3R%FQPr0XP+?p*8-q$(cc%6?vxPbq{<zJNAh`-ZuLyp3UmOcBS8E9Q;htLKmKMQ
z|Iab+0R3+&#x`ldAdO4u$=<S2tg3^$7G)1zr@s?t%7+EyGupYfeFqYaf6leU-`ZfB
zAg14ZYAdt%SAut$ce63{9PE_#`ttO+#T`p|_-aF=oBsRoJFMCNEJcAWG%I_@?jt^>
z;!a3dAKn2dUF?K*n#B&DU$U;;a8v=pHtp(ngl6fw*BnE{c8Z<ai}|xDnFTQ=ic9d)
zL^R3L7Hj<i>ac2QWj&vlB_Q*kDR4f~45iWFm;IHAk<j-&4aZcc&gTTHXl{$4%HHt6
z^Pk)u9-QR)@}2qEAX$6)ISM;jnf*{kx&PyP&03hsk6Fi*0e&B~<kB|or#rc>Kc%mZ
zp(tj6J#jbBSd6Zv!7krpil@_~$sIie)|D%sb7l%e_XF^DKl27fuDB`Ik&a=YKy%4C
zUxb{>&<g5bjG)Sd<g*!W5yRkr_{5ZLC9}To*n3(^E{Vjuw9fRsyMO(9dW!Cr#0AB^
z+|sL4HA0fa;D8S*mUW%sH>p7ZJ@Hyt<yAj>@iTm%&4QFKcGtzfveW%7tMk;`fJf3V
z;}vDFYP+Yj{D{Qha;HeQE*(}BlsFR0>MLVl2@R=!0{jm%Jqn^%1VM;5y>vf=34QTe
zE7dAOnIi~rwdTqJNi0&b8>HGo&?NgL;A`-CnyiQpUMkXsjfHRK$DIqmmj~p&xls#f
zguAZWg7Snrl#tD;vs-1B5p%<3eOh1rA5>n~_m!M!IM?|+{Q!1YjY9o7`>P@H5Y>YH
z0((==tqI$pt5kb3xQ$hlL`N_o)}})SmnAPyB0HzqmLNcW(b&TAz_)C4QDxJSlKVVm
z{Z{FmuxxZ{`H<ic>ylp>K5+W&pzHk>)`-&ehbWJvl=R#CVUE#q>Dd2|q;rpF`v2em
zHfK3zYM8Sysf;Oeob$+Wj!C^SIVEAvQghZ?b1vr4usIVVRC+s~a-6d$IVL2hq|@j3
zdVhca^iR3<&&%`qcs{P{x?jPPigE}EbFz^A(KYGjCSQ0gNr*iaITLE9T`jQW4QV~j
zE%;Wz$&gu2*=tOnWD>*T=V|!tNRcYRL@O&a%`>=e$KhD8<0lS+wRyp(HJwnOMP^^w
z#6acTXg}Ah>Sfb>z_e%o^fDaTiE_(X_%^AtcY~}P(H-n*GDaHk;|fT;8TocON1JK0
zFJ09EEsmJ&kp~xcF%W}G&)L(4cNA<EAn23iw!0;}HpbZGkdngEG|*O$^iLe@-y{H}
z(ZhiPC3saWyp*|Xqo|~ycmAw`LJ3Ywz)&cELX$bhY&1wegP*Cv-@c-?F<Yklc5?jb
zJmNH~p~4{}O*qj$^MDX*AJVXxiAlTsO?B$IXFp!#oGr&D&U5@`^b3MioHC>0ZVu&2
zv}2``DY3+l0-_C+yqYEC+gbQqgJyi!H6Hso{~W8UbHns#Jv7A4*qVHNDl9)Oqx5cq
zp%iCj38`Byfo4E3x>;<ba6w{}qr<%0hA=t3`kL{+9KaOpl4)JIYC$I3$-%wqGcE{h
z$m9@6Lq#o+v=#Mpu}9iW4<P;*WoJ)5q$7tdKLZ`FQ15io?LZXGzn52a>E6;boqXR(
zV#=NpSv_aKG9RDb=a&ZFJ+a|r#jz^>ZE|uVUuBFS$AguM_N41ySri5oaDBnNKXK??
zi=KGa-64ptQ82`EnV#=a%ETn!M?LNVp{)n3I4DsQAlXz496YpAE=<42T$NKDV9dP~
z+j5f8_GUIEtp{I5buUV+d}xeYeqB;Odb@tfgLEtLKl;omKoRcD09Yp=TdGHkS0*ZM
zd7T7V_5{ebe21OO@l*uwfB&QlmGytmI)IS?(w($pH@R|mb<esVPKBl4jGp!QHKauH
zOMk;m3Op+UmGK<B;<DEE#(zhu9ZvYr`s*2VH`q|<$Ap?{t+t`Ua1vUNVc#<2of+HS
zqm`O9O5HbGy||H~HR*vvE5K+~+~?3EE2C{17VUw3=p&l4nfUg%e0r^(=y=oKg|%01
zb?$|#YQ{bI=i|TecV(2rIV9F!$xT(~^S<Eob@t8p^NQz|^D~CR`%KB8|ACa0;hkb1
zzjpWLeV^mU@^1k2QOVmmhbi|%PE{QItp*P5vpYi1mTWE5Zb;-vCW420pT1Vu{hO7%
zV>xZ0SXbMd&;mc9JP@US@Xv&7TWh#@w7Oj%Sj?g74lRBL-yifXo&9MetveD8Y-Y-7
z>?h7USUi2uf?VqS%MKD`R#d8<3GHmG*7djo)>oiaN4_8OAh<@9hK$Q-z~NqLhqCJ3
zAt_~RXuvD)y!UfVg+TTu8TR*poPNCk8N1xft#os-Zi(5$!0!v^$=xh4KJNzSYa~_?
zbXgRSdT!lZIBjJ;d?Q50Xf9P+6}dr-l&2{ufP!v(UdK0p%G`KXMLom{Mz;|#J_06X
zPDQN{!G~EP+kuhU8#PAO_YZkw@74XBj$i6T)7c4ewDlCa(#g%n_9vG+HsbsY=4h<S
zeVHT&4Vr!n59N>+KN<zPOm$!2?EhU3ejMMvRoK|so@#D^4>om)^gvp08B+Ryu?sJj
z#K8ezs4RWLKZv$npG{EJ8`AV{3wM0jAn-PuUND~EKv#ZO=xgNG3#A;TmoJgDxWlFn
zg`G}GbCZM}OxP7$u7Q;_$J=y_!66bJ4<c}pk1(#R#6$MD4LG8>BPelITj_pdz*ET(
z7UPF^vM(yM7%N|MFui#`Zcn%60R7Olu(PnecCP<w+xzFOauXI2EO368Cq7rW+cYOw
zJm_pZ*38lY*=G#tSXWAG`Ago(q<`vDPMfvQYaN%}e7kJr!-5_-Di!>KAdckY`2PKb
zNq)%rCGx03pCio|EOXgfU8$EJm89i+)0+L62n!{Pm@DjxZ&Sr$2bP!+xWKG98z4yT
zVew|5y}l;xE3i?HCg~K4N0w%wer=xFTOp|pD&}_l>vVHJBbGKvbHmfISV-o#Qt&`!
zbB-W<EQu1Zf*dGBA6CfN$yE~eYt(B0Zqm!5RA472x2dnRYCYX2P~(CN&XZao>*?%|
zL;i=*k+|EAiI*E&rQm=3!dj_6E#LG(3I({0`ZK=Tp6Y1)O`LA4I@#G!P2*o=j~6LS
zHxA9e%%`|a<RKYgXsio}Q?V8nzG?V0W0Q$|!Mc@t+LJ-mcWV1KWCt^&)Ts#iN+9k$
zO1C(zZ|7Re8M?{d?>42^d`Qw-g_vm1LNa`U|7Mhvmb&Pi@R{HsCap++W%?=5n)-N#
z8UUmp%m&wU^T$AaZHj#^0%vQ))_Tl>RsS`gDUuTzOl4pekh|CFi`8Zr7qC5aZNg_?
z$k~eJ44<yf!G$CoqNdk7A`*a`#8?Ci1i3t(5DwLqGLqwA2H{xK50{uLzB`!YFxtK%
z7=K6$oNNzwkcF~juTuqRd$>4mQqgJf*i;GyEKHXu_1{G8#4Tdn)1CB(qXeU1vo1=2
z&6xcKn}`A=ccYK|we)Is!Y_m9XPx<SJirl?I0UiCXG5jRDt_VuzB<|Tg<B=>e@x~A
zh6@&>D>6&v``qy#Lod$-#Og*%i`OACD-<1hf+N&#E`RXmKJ6u`GRjOtS;E=uS!AzF
z)4ASL{lkO?Z3rJy5n2FM{2P)<Vvi$_fzwS)3C3~DA_AfA*jv`SsUrY|fw$YR`jOf9
za=i^2&^&!*r|!Xn!%}er!3jVNo^|aK!(nutXD1)bd;g7iHuC6HPE};h30KRG{(JS&
zFNG+R6Dhm)qb);B?$FJMq_1lYaN6iMYVl7-hr;pm%w5d3W5&Lj?m^}c;K<b)JI}T5
zIbIk^^YEUgUXA)20?Ny4Raf*Ufui#Fm|6J}jopf2cB2;SweP<O8GT9sE}ij-O<kD}
z!rp!EuU`Gq%2~Jl5@lm*hcV&3l#Qjw4~lQztN)KQic0~uhuZZRAne9&rJ}@jA=~J>
zec%$+-J|abLc89$vtx|}{B>}Ohj>LPfF%hg-wxWeZE9kjrPm87mq>;cKMt<>djwL?
zD7n=G^7bP{1|^8QG(22+Zf$;m<lFvAuTELTH;QstMb;ABK}!=hJ82~L#LQ<5PFg&r
zFsmK#mCndNU~z;7s~HOV#B^<9Hh?|4AxNQC$2UVGl^{g*<s5wVOl-E<&Upgsg`c?-
zg~EH{ZPTzp?MvJ#Ebmi$n^o|!01$~aABZ%uu!wJR^{>QYc-*4wxc8oD+d<^hz$Eh9
zuN){I{o}UtCIa_IdH^^e80RI<i7s#8uy%3N7VleIo%sGTYC}Ax93<k%;rshDPc->>
zohY6bp<}EepcW6@<rLqr064Qn5KH7MU~}}3o%u$?8-{KC*8r9hcHFP-p~^Wx<JGzN
z_o`N6s1E1mhc-Lxj7)2T$^2D3heT8H6zrX8@J2!OC4{KdafR*O`!~MkhqiBF!*8Xp
zhs#hucxjH)3I<;oM_znzPu3cPR`=2XJ+^-V=c7-w$Z7Rkn+HBL)i(!X9`!M77daSG
zb#H8YSjH?!6f;|np?0sF0*YOa%bOs>Rf9PlF5&g#a5jd<TFzLfEac9O-IH+jQ!kn|
zR+yQJ?y%CO<7K*MN|R<>)FBO2u}R@I6&0eaG+~F&->Vb=4+!&iN*Y*#Y_fj02%pi|
zvS1O}8kX-EhCnZV(A+44(w6vo(}arpg;n8rI8(&!;%=0QwN((7Ls~ZfkTwX&ln}|B
z)zA`_396yt^$93~gZvkogy>bH7IrRm1fsPpr8IwCGb~4MIT~Y6y{B5QVGY8I(>gP9
zxMnC*rZzE^priE!HGUg6i)huktypHYQc7Sh&*Co!=Q;B)-L*(A8&&Lj>s%ddZ^SqI
zbWh8Bmdl#obl6N0_B)yS|LRYti?Co#AvFdDM%bk@!&3DwUp@ADB+k6zRe2^IJ%+cH
zct$#3y$tYeA+a35T8ni1qA|qBvP2wFnKj_UgPByt@M{!t6*oTVqSnZ1RLZX*dZ&79
z@H4_a{}4(GyN!NNc6HRE#dz`Qq=(7(o0RF&8zct+P>EBUE@6Gh&4evQbnKYTr*Rsy
z%o8tM%Yelxt(Js%I#zj@`=~0RHT;rrvd2B3^%vl-bZjKYd@M=F&sfA&BFHj;{UXIk
zlX29#`8x2Y*UO+y7l#FGViLcLFS;Cdh+5y;#UzeG`HVp_`FhiZ5x_SN?%E|-a+IUG
z(Gk!et?X1mMFmKyMSOdek?I<KM74F3`4V+fkef}3=&yYhtaj_$n@nn8Z9+e{evgV^
zy0aU}=P{^v+~P^%YzYmz4aQt!lfDH|i$N30_^Q|+6P81=e_bwZU}_cSi~jzgUkUsl
zDEhF>Z5?*kL;Q+lXVEiz%DG=;=R>HoYbWDZ9qaYx4eMX7Wps_-K6R}*F~tNX(ljqP
zitTrK9RX>$Ai*#l!#$us7`3<$b-vi;zwy_Fs$F4yjTiD7%<>@SUrp;5`EK<xv0pCD
zorRNh2cIp{lrPO?z8brb`KysF@&D4mII++^bJlTan&KA?@!N$n>aSyTFN2aY8%Q1+
zI)+XyZ_*kmrp<mskvWBLa2I6GWG)0*9PlYBV*b1^&55=FuSYSf=YU1s!$35eaiG7g
z2=18N>kYD6Z112M8YSRmse#XAYZbvCe0tE=18E{CJ6N7<{cxzJNgH`rBcQ8E)kUjK
z0Y}<LCY`jMx4_|3T1p`4jCcr{8`2T`pAhq3O=4Zk{}GnmY_!<q6@}xWiNGhu;KA=)
z(S9KC*THdGyGL4<o?xfu<Nt^<<>IRJM28KbCEepkI8I$3aen6R@NZ)5)=dY~Nbzqj
zf-#>@mMHvfx+>-#4JgsR)y-i;GCF`ZhC&~_Ow@pfm$KeXmyzJLLW>+1ZOhDWypvD!
z1aI1~IBy$R*LW|Ewo>>3IOt-gLeb+O=?nw#2?xVIdy?mp)zRl3AA(srPkqOxIrcqK
zsh<sW@Hxm@Z!q@MerlWToBX{K8&>7jLiDiVc>$G(Qq_9=fcBVJ7_x`Qvz+lmxQu%l
z!Ubq|q|?ub2op`FsZl<7eWm*@!CC;Wpav-<a*EH?&p41BIjkbyf)rMg(&_tzH6j%Z
zTf|f1p+u7X=)X~wzeAxsi9`-c)<hw87^DaTO^(0ReCB6-MzDTt?}l|^sGWG{`_^_s
zU#Z{n(0Zy(L+-5<_~XSs#<h!;s@-4<+u4+buKdVHCzE%g{hviS6c1TplPqzQ5aDA<
zcg$3cwTbMvvcZeN|1=w!*st2%|B(CHcRyoYZwD=W^UF_c$)IBYQIV1BcAU7mW9o%w
z^$WvWJ2&z@y-_EIIrQVqGa_k6I}Yl%S0IfPt50J#V;8JW;ofAdhdG?&ReY};GlWKF
zxb}B{7tZq<a0Z&m)>&r34mXWCKm;5PX0p;wpUWq)o5AlCX5W(SgHJBFjj!TNMHY9>
z>QEjwRT+GKi3>@l+Dr?c6VQdP8s$Gjs|0ocHd5qhlmSFF&=;9GO_Z}9sWHkU0;wvl
z3CiK~nKBoQuD0O7ZzYi|p&k|^+g9F8NBfE|1RSzvV3Xu^In9^_(%0<bCI_F49rt;9
zE7M&^=XQ*h6X^-$kFgeIGC@9)j!I%rn*g^SRTsEFVukN7BD#G>#qLcS&P*`VUIv84
z_xjDz&3?DvVFbq{x}iT_;&^V(D|1z@NLH{fqiwX!X~2WfBk%uD@^M(DN{p3H_P=3g
zn`FjuhBq#!$NQJ)ROC$09*LTN(SVTut0^1<mUx*kRQCn-M2M&In<-$9rTXTgJpRc?
zLdv~KE44?OzUk@M1=5OFw^uI7Si1VnPaMBL6Slo)+x&5qfgNB|6A!|0n&@#mEEDAu
z>y{hYxu+j_T(kLIJq69t2BaDz-gDT#=jN(v!_q`?<a{9?xCg~9+MK=YVZE5m8?d46
z*c8RZ<GyRFvfg0f_9jNjZWsw-z{gl5^|H0|dwEqP*n`Yc^R_UF-x&$N`_z4mvxQan
zNG$*^>Y`kYme)Q(db?@dQc%Wo%3XGYr>vz%>lco1aN}Doz1Cy($}dZPEH&+m^8IbS
z+{*fk29M^h(=qnGp@FL_;!$`8poN^iA#nx#!}igW-diZ{ZS7n!G5Xz+)CQKRyaSp#
z{K8rB&CYR-Z)0cR7cyz)fz7tRWgbzw5~RX}ao<NrSYrn(WVehvd-5Xw2MWyn`?=y?
zJ!aB)77><eZdk|4(gZdIROJ?lEa1dpXM@pLpn&~E9uZ{z$sLJZL_91Z?|DDW7<OTf
zH(&j2!Eb(0m6u#Fl+h|j^^~&<Ux5@4NX7QBk$1<&2bPHf<Y?+$S%o&1v$9Rn=v}M<
zxzHFMI}FmevWeqZ^`3#Yr@KltGT-L^@d1fQChADS08!(NYhi(3XmA+?sjvKT>?rAF
z78i^dbP(lKvZ+~pWgG-QD_O&1Hb>1ZrcWY#K=e{D6~Wd?f&dXgi~s1X#Y828&+;wr
zzz6>W={ww6GorW+fmQ4KnzUw<YV}c@v!ubv0R5C=32T?m8e@;m?g2tRzRWy@e0KLH
zK}!bHdXfwq*5QBGPq!JLJnljM{WjvulTK7rl72i1n*JZiIhUixqO8qM@Q&%%0UclQ
zI8rS1;_6~c^z^OCG9Xv6RdQN>@U(_~d*$<k=9_9aD&RwGinu6}heQc?48auKx*jXo
zf3oCee=cZXRQxmeahkP0DRyKjn1Jd39zTHkL-Q8@gQ(9mcpv4M#aAO@vW4pq{wf6u
zU3CmvsIdg)XHGwrQ-OA$N3K9i6hvS>ARR7%dczlfnokiDv35e-v3t6tE-i?f^AGwl
zdHUWE=Mp!2inI6v{#&U^#}=mD8&+1a8IKV%nbYWTiSvqc2+@#9*p^SljH`Eq!yh0W
zd#6QqZ(OFWy7VNW=ctWCb7W5s-Mu1DG?;SR({tcH&g^DO`AcElD$FAZcMSKRaYuZb
z`J2#E`A35Fwn^(xeBSh@d}jnX7pFLp9jbEP2PfJQOPbCiWJ-P~--`zW=G|bIuf@ks
zJmA%Q5U=-TUnpU>Ln`C!4@x9zIxW3j>AbvOiiE|{a`Do>CQcMyPUy9Cp`11eGHZCm
z1=fHtBCsstmDgb(FOk#Z=d-1yz9~5~FBx;)0<Bkkynd&@9kxvMs85&K!KYKnWNL%+
z+RasFb{|k>T(=TyMp-6`I&o-J?7a774gZV)7m2R&Kd_qI6gF592PEtYEi6hq@7FTX
z*q4ig^ki`#qXEa)NIz+Z{Z>*CS=}HR%5rFos2LK5r7mo~h$Qgcu5C(dI1=VL#X@z#
zK(mMly#f_cvmV0Fp`4b7`k*Nzx@UER`ij6XF-o-QRDidajwZqJ9X4a>!rL5Oe>DlP
zc0RQ~;SO7>#x8c6Gc_Y>Vg^hMFxxT96G{%hb#qFtTPY0zB?Xwed=3#Yl)Xq8g$`9o
zJg2JNFf9mDGx@?6xTx=@Id;))fg{Q{-82f^Z>?c-Zqz*NoYe|%`?a{sCQ<gdf0W>E
z%v<ufmdQEiL|anvd|qRsS)XuTLGrsW`<Be1jW#D);Q`ylydGyfQzGAyr{S_>wo!zR
zlDXmaeE8q2;<TRfoMm%Exe=iE@H?G_(5pz}l2M4Lkwe=$zC4#*dUCVqp75|d%8(gr
z)2l;Z#&P%=f`1#z*(Ga>3cy@w`=Qs(Ui5B@E{-zp=y>ER8cvWNw^aA4g!Wf-tVlHS
zK9-Cs0JdweAS)}pOb(y*Yx70$GF1Tn_2;-q*Oc@br+B-*LBL?Bx^TmOYAgOxXrJ8;
z6}5B+{HR~9)OVn!#O%rGS&8DZtAU6(c>zjiNu08!VlSH2Mi<wf5?<yH^k=*_WdQFT
zX~4$<c4Gi4HV%N!gL^J2vc^Mkz;7|25B-1b8mNs_AN}#-`VFOyY|aBZ>>od$TnXIp
zxo+}4(%V1Ne*Mk{3sKAAbV}I42Qt1n%9x87wZc#V8k-a(vbr$5tG4M_!}usW;;0MA
z0MUx3aa4JKt8m0bktMR)!^K}S!SQ`4&=S92I-K%mjIEpwtBH}cd(yima;*;C>zgrZ
z!3?dsV={Nc(;;9*y`0a)_^Kf}9_5wp1hw&j*$1mBWw84`q_F<9J-|0XTA2UTLDv-8
zK#9Z1*~<LT9?%?b!gKYvC;Vs`2G0j~4q64?gl8?GAy{AG=ag@5LQP!MG$EktDvqAm
zEMWHgOEvMJFni$q3!m9XzEW^|3+|VZl8GKGc>8`+zC)J2$;l^50`&D<DG~X!u{E@L
z3jZ^41Wn}904kEb)NaedxZ&gZ7^OEm2^`u?(xk;`?ui$^_x9yT79@K8^`@yI*9)Bi
z{0zzbbA2Y%%`?1=vN%99D!d6ARcEdFjHVeV#m2qMEX-P_)obAUKsKrf{4Lh{zN5<E
zd#ziAlVFJm_O#Qm0}Icw{jPu2>wP7_as=R#R^7R-iiA~HU^Lb%z!EB^s1~WiU5oTU
zBsG}4YQr)Wi0Fw}L4jy#8b?)vf5ybhTnuCi0_TJ|^70{UEN7LXaDp>rz%a|qk_cpD
zk>KO=Yj<1GLGVnDU*fHqugAxC96q-^u$*dO6OmC-TD~QNnz)$rpo(P+$1Cx<FzkNl
zbL3Nms;`hz2J@mdh;Par71Ycxt*3uA$4~aDNwS_AS|Ycigt@cgU-@k0sm94EnGY(u
zpUXJ2!i2BfQk2@y{8gNtg=d_VMD#}#{xA|nNrlz&eOYn$V88L}C%0UuVkyZvDN{>}
zZ#_>rnbudTB;u1|gxxrCNFaK;N%)y#U$Ye{P~SX>jpFmWF5;mjV;bf7LMmJ@(eRct
zd48oF@I&u(pk=aq(|z%?&y+}vNIw0q)=>gfmD&sa3KXRxN~00dV2RLQ1`6t1voL&_
z+MDU4Qw}S<iQck470*n2ghZa2`BrusgKj?rE)D3~{nkH9{AwBeQ6P~9jbUZ$B-*Xj
zG~wf-JXr>|pD>IN0jvyyNW0dE+85?|_sqE}(nLF-qJ^atw#*~xkkAzdYR`hb8FX+m
zqg9TL+k6yi+hcDjL(cQKxi-MUy7WiM%*#28?p-k|Y*KH0jqa=BnwiQ>e`|EA*IGW_
zJ}GwCl6!3~1%45snsx}f!cb1wzH4zL3(#F~hLrz-YWyOiEL;-@(4)|~Y{Qv%nIDRu
z7*xGhAZ?qHSDk)c(+kP8KG<)SBfYiL^-zEF&ZFkZ0&%j*y@C^bzDLDV)F2!Z=-WT5
z%?(Cg#4N%KIx1&wH;uwYbH(}ymsj!Z<t^`31;<uJ*V11Z=6f@Y^oI#AjMQWMgwKAm
z88nc{*|n|vt;B7p7Tdu9mb?P1qIhxxuq1(PM={fZd6a!+3mf4FiqMd0#B1|76kp?Q
z-%#h%Wa5o3m>rhm<?M!eu^v1D9Xh`HC&?(z-aVQ7>j+iV=hIEnX=bc<|E#e$KY*OY
z8AKeGN-S~r<2j<T!f5_n9a~oGvEWT3({*$a=;$lLTG1~-K7u{7h!r?79p1Wck_v_X
zJ2O{JYO&$|Hql<MXt#783B2EPEJ@&-ViM+8JGDb9Air!}^hp??n%YM(>^}r85EumH
zBH%<s_!6`J8>s#RX8)g!2M0mW1$#%Lz|{TcXFlfbxLE$K#`{znU8xu`ySz}Y(cdS0
zy{MFnSI$l!n#4-7Qky68mPj;qmcURvzIHz*u)Fg_J}ruGu&I?Im{8c}qZ^k-yr|b(
zuzx5|gu`Rz(+OJ^91y1#6ns^%2ZVi0Ir6&#<(y=ypMCMlk#&27Q)g@9@Q%s(ceiHS
z>m&;<8;n!0F^0_;T{4-0M^Luj#_|&7FM6VJx3@wJEJ!iQQDH)E&C}ap{+WD__y?e`
z)?>Cnhhv#-azpEH#N!GuQ^uzb&2z`>f1spGLz5R`)V$YVC;;hr;p=)VH<2D%sy?*V
z96%pK2<WZ{_{lJGGPi7&Z}bmc+p<LREtjPW521QF{6=3(4GBkq<6VJ~<+1PUEU`af
zIzCoq8ntS4)l-1q+KI%jDWF|X&Ofm550E8+NJ|BKVFMLoq2EipEv&e--FT+1KV)BM
z2IRu&Vy_!R1;HjdY&KcH{+xd+2H#imJX+7n2O0Z&a;pX^bt6xscg%<lb*|-=b6PJj
zLhr9Py>qt$bo0X<-d9=4e2cS(P-1UOW{{4RUmov}rHVi=W&qgp;i8sv-l?co@jO>=
z_N;#Qq_3`ecATw?JvaRu2mYE<5Nvj7%1FdsB3t!vz1;Ba&4f@Ie-3_+5m>hk-=jw0
zOqsF#_S+`xem4?Ac6uPEErbCWwF%~IGnALtHE38HxjP-3Y&AL7c0y+vQ205qqIa2#
zcB@eY?o{$a5_dgk8ezRF3qya^a_Am04g9axcg$CobYoC!qOLaKK<HQ3L22$-IKT?l
z-#~okU%R?4F=%<_tV-g?2FHMsXq4Y4j17l;!bAUq0*`N_f!m||7cJd;b*GT-83li{
zmB;<hb!FUZ`5#DV)3(w*?g&5m)4pNci!(ep`g;xMoqgaE6*Vbndk#u>Dh4e1!wpOY
zK{6gbLB35f#%!y+PZ}|0J{@p?gE8CEsR!lq&1_hjD024#zeaVWhZ43=z0%9yI8}E&
zUk0s%F2%leY7*0u-~|W$wI1X=q&r*JDheufwlBq7<iy5POhtAvP|5tCRN@|{d$F+1
z*FFs`tD6PZ#@(sXYUTl?b@Gv9wnj7)EG9*{9d4NS0}ZQi3n>p~)A;5G;@UA|rg412
zTKUF8lR2$jRzkVL7tQr|N#%KV#v%Tu`+L-sXVmh=^L)nVl3{)F>F+}lsD`4mJB}sG
zGQlc72nbKT0@dun)SX?baq(-OqpDXzD>&!lC_p0h5*FDe23bVx^M;l5zDzXquSW(w
zV;IU6y}g-TtX6l}q`>=cjZcr{*LIys`C&wNmeeWI!}gomY$I|CXy}?b4^v!eH@T3r
zywq~9Vyg$?e_MR$SIeG6%=wEBCxlJc2~%?kQ(++<(G4HE$w|4T*Pwyu!(?^kMw8xr
zO?$FI^aRZPZ)1{n$a>bF50juyUrzA~&?k!K`g6=jsbPDgzL8d8r!o%ItPJ~BEY>92
z5f2A1XK#W&(C^#t0``T_{%)#ge5uOE>?_!(!xP1F-+S)<4|K6-vaK)(+kLQmjmL{i
zyU=$FGrQi4Fm01QHJ|h53Uo|}4_n5PZehrwWXj-3L(mkgh?C>ba!Av_JSpE;Ju{r+
zU03_cMAf9hL<w`AHf-dgs0eHto)Eba|1l@d0Z{H!iO`-vdjAoP<^L*SLAC(o3dmGq
zdP*}fdG+bF3ueN(+>!n>88@*vy<{``<kqVP0&F#GxPD3~amfvI9?}}l+~X#{_Dbsl
z%%7MI9Ulm01G+P}N(I@>^|>;IY3z6U#5Yx<4{2<D?pCgR)-E7^-4?pka}*IUfJRmb
zw;q?WCES=*fNp?7MkN|gt2}-~%0TnOXs2xT3`cI`%_d8ma$u|N!>ZTbJAY^z(tqM5
zJ3F0aCd+g3{AA2>K>MkVC;kUBCZ}006VtW*2nfjD;d;k*V&nBF@sZ$s{ORV(c&lHE
zKwV}`R*cWDb7xO+-tjG+96KX=r~iwR8d$s>a>n{Y;`dS|@9^og-Ud7F<=E`6aAOg|
z4@ZRmklC&)tl~DlmIQckx%Ai5?$KjEwDQHWU-7O$)mqK7C!|O!;~DC2k3h!3dXXIW
z{W_QAt<??K&<4Y#(L#f$>B(^+FVCF;?LGU&){_Ra%^PVh@O^=onz0t#HyR>pIOSsr
z@?#ob{kl{fr2R0}zYFG8wyfh=P|0@}ZJz+YT{=CT-P*4M9guM`evh;Q&<gTIDZ~@0
za1^Xg)fCl!Tj6Fz`8UA2eEnt06@{9B`A#E4mh@}=<FX1iCa0h5UTYW}?Fv>ZzA8u2
zcvGDYOsQetg4t|{CJ+9E+>!D`^njH9d^PpUQWXcX5~pMHC%!L5hYy3C(ss50Lv<|4
z&6q5~rihk_&17j2uns`9ey1a?74(>ia`_>&Q;wrgq^5f_@Ba+_X`T>U7+LvH)wI&L
zS?y-~xr1+S0@-f(tja=6<g2?J4V15sIAwU<j|zd*ateeMHDsA`mPmH$DTPgmCt41!
z>G!afZ()2=R<j|0mh?nd?^##S0~<_&okJ_-kG%^&yvBB((@+lXo6nv#<TjNM9tAK{
z>M){-3g7kVvgVh{b3KN)WUW3-!f%;LG%@b_BQuaTbH&Ep)~?(+ml*nn#J-j?woz1)
z_N7g;7gP{8kG(~XxFUqntu^^``rKM&lExN5rBh9L#23*~;lPg;IjWTDvaqoJXC%9%
z?_ECCGlLDH8?{~_=$i2%KY9$T{28Q>rn9BHjOkk2ohBNe@t262a1P{Uk!698@q%>3
zI5=utO!~hdu^yQ}a76=j$RYdl&C|lKM1a`@uleJ;B66OwWh1p?HW`G4q*u_+-^j--
z5{0`h-WQ}+=v0*WsJ!22Ay0ZgsZ?*hJXZJh%W#j8aH_GwXW)3=x}y~{`%%)o<z+K3
zQ5uyen)c)HSNgw*j-xM#K8w~MparYueCqtEKgkz$Bb_)JNYp;l%gtBys6$<}J}5!N
z%1P&zY<mX-8GoZ$1BzXJd5xV*D%zgE)L;>yIcoYDz4FbxP+86*7wZW1$P8CP*0q#E
z?dH?(aC+FyXGuZ(*96L?$o59=mw)2i@0d4ZMOnu7oHL%YI;uV_4inYgLH0&FUOZQd
znc~`-p>|>$rN3`l08jnlCn?r9a~Y$;=fJ}#;6K7BYEL8^VtUi3(+=0^tin+<Sw9`R
zEvUAwe)TX7)NH0yX1wKMkC#o=dz26sHWu@1ovz~>d-Dg^g+=6X7ODT+VarS8v7`E%
z(0HTfibgBF&b-&Px8)yU(1$<mlF`74)MrzgmJARMInf-@F@BCB20o5c+&4x4IE5j5
z-k$j6QVY1klvUED(a^ZudYJTnJQMl4aQ)Ti7YI?B2We9ZI4|RZriuUADMe_k3iD4D
z>Po?x6kqzw2%3dT(7<{kYbfyE1BDAw!xX_PinDLCKyFbuB8NT@B;or+1>&Tqz#r0H
zQL6PcmCuuJr#!7D0h64i!PHrKoBLVLmif)3qkiW26Xi$>x1vtl*6@0u*6L9u)EtNO
zu=1151jy#0NAwQh6~jaVqxS|B)T&YD4FxR)>lI-6Iod!#kjnFz#tx1zocyNxUzRF}
zg5dsjzmd~-X+3V?8!yNVo@hmI9D4{FQNys5uO5}|a4HJQ-DzB}?vaM=U~-wCMGJcc
z>jC=0OiWq-i*`<_<#GyRtc3aWZj(ME-siCN75n3-lcF4M4-L9Y#=utp6=|~1a{?`w
zv~rzt;iMA}>NtyqC6tj6(+c|gN&TkA{sI5jCRNsu9?X{6W7yC9yoZmit(tPI<b>Cz
zvsc3>91Q>qXN$1dpTBL!Zhk0FftXDu&%u07r(nuR<a$n*FE2*W^6~h=*u=xnYCSH3
z;*)uTm@l2>{R<LP6PB>y+=-Vf%?IUT@zgJrUf%RzYaOl0_t=$EzXVIs9ar%1dBH}C
z&M;OW_RZ2^)dk)DA=@RsAG;V~_;D#7?GdO~0FvdAa;Ln(5hjml>{ndDw&hwWk_9}1
zMZRK4xM2M3^tSSctK};c%Ukk(<QqnxX8Gu=j@p%m1gFz30$M!bJ?0tkjizho5<>N&
zD=K<>T~AwHg{AU~xTJK?0VPPMd$|Ilq8*Z1L}Z`)c{b3iUZX`SaTO*qvBj#IIB#YS
zKbeU9*rY)|7?!4StRF(@z&9p-YCCYM^rL0~JP4Ef{>Rm8&N~nw*ZyUMsbd^^x8{8>
zEfHUs-QZ-ZcCR|I=~MHN^T^cFfvsLj4*wIuAJkfd8z#xuzZ_rDBC$IBGo2S+xzKBW
z9$eUX!u6e8J73U+u$3NG+qoj05>Du=aZ);~CNL88GyqhqckQ*B>*a}T*6-EpcY(T;
ziuAdZ&ZD0{apX(8&iDEfzgLSioNG>wWIfz+88azqR?p8ZM!Ba-{$t)Mnt_}zUMmjd
zn4FjjMpr0CJ@<=n0lDJvC{Pgr`&p6QDMOn;D$#knouPOx*Xc};7{kE!<+nG#rb@5M
zc)#yam$yL=RH}L6Zh$VYSLyVMiL8DvWhIznR}VHF-sOkV*o6vxx%#(ykPpW;+9Woh
zM6h;YwwG$ym@3B9K+Rx}N$1ks_q**hr_$$Q;f?ovUxbTMTJ)0o4)hq13<1C?XMilf
zh}{^e=MMK`&WixKqm7!g7nQR^9(kt7lybCI2w61g@3;T!a?uGX)tDfdGLz~T5z8J(
zi$S=elOr*qeQQ@=e15=fnb_r}$hCuYf?fhh$h%m6gW@(8R{6-{oOqP?JW=4KmTg9~
z%~^rWP1~y1Jc+a?vwe|Q@4dkvRWeNjyVHJrhxVU;BXsbjayVcm^J><;ZW9?qqS7s1
zZrphbkh;UG88+jJaY4@hE!&r-<~>wF2}^|uDCunZmRYLH6PDynb0jmZS7-}MAc6ey
zkE@>en@LR>Sv0o$_g6LuwKyPUL^$M2dO093ZgKOeAk0quOi<x{3_YNn(?J><im<SS
z==i?!JhSg=*4UucYo4Jg8MRd+R)ToVbvKZC#t}UqmcemySjF<(mO9cHc`Yy;&W9bw
zmef_Bg5GZg%6mwx+T4;FE9Eo}#NdGyJl2Bs^JGa8OASiKx%H->#W$VGbmh8a9=c<}
z<9FTM&&@xV4$&XyT_caLQv9N}A4w@z3gA@wJAhpzvBehSs%8dwH4|rmB+OuZ)$<KZ
z9xOmw+||xyAN>0P1M=$p%c=mm*}VMY_D$TWRrmdpiMrP0nCBE(yxI3Md}7Mq5NC8!
z-lO<qysc?oh)H0{x8lxq-ea9v{?4>N^ux5&PU7pc{(aS@d8_CxS;=1Fr0yWUBQPb3
z83P7Dyvrrb`HR2lVypgaWXX|Sh5Rp1ErYcBV<M_sgPmr_T$JFU!XoIuS&xKJ$*F!A
zcp?9W$@qsHF7lWu6m$uFxMRh7EcqCu<Gjj(yNCpM*<wJ8snjamizuD}D&xswX`vb2
z=_O3!9_Tg_|LHAz_?>)AJLS1%qPd*2LPK?{N3cds*6Z@|LN0VDJo&P9oE$XWCfRkx
zxTZKg@1d^Uwr>$CWI4-TREl;;^8m+Q1+Eg1tp$XHaY%nKv7+<x%3-;T^~W1L4`6x>
zhasv=ZlK@e?l{G{hkXo-HBSvwyTbHGXs=xxa}MFvfyIJ7Seq~1y|qGFED1^u=GQ?;
zCp-uT{e^0*{0jxUgmuXyr}Gy~_;C?3O%IhTLOKAXOXWAEn!Z84`%N>JFl<quHsM3n
zYtL`Di$sw9$ilzbN*6&2|N6lm#=<~xSA(XhTO<t96=G^`*srky1xu)!ZEL`GZPmW^
zZ~K7UiRe)j)6Cy(WjZ$Xd~C;hHuQ*fiHXOK)#JK(f?VoM!m?%GZ#fT8jiI|WftRzy
z`W~2fv^0l$=xH*QamF|3mk|pTbPZ3R9Jx%`3ytmve%LtMA206LnHba<yJdqcRTJso
z+_H*!o=)|BSlqa$8mv*-03!6smm7PrS(-%XAc6IZDFKJr&}~d&mFUenTpa^xY`R>=
zRS}bjsF3M1Gvj^ITG#5P-Y+P<3#g9gF5nHZn_aDGh5Ote3=GP&g^eJa9)1@tdg#|b
zl&cR-h*5ffZNV~WKX0$XA9_t@Q|-?u$>FcQALpJLH?$LWO*Jp*=V~(Zw89qddjyd_
zJ(rgygsW%Gjav#B5Q+Sl$m25fB`L`=hUc(tnXkmzGWDw4XW`R0l<^P1&$ARg`NX`c
zqXGoN!Ac6VmbAtoD{W$mAls{CDyzw6rcEP;l%}|V;acS$X-W1)vsWNR&MKOIvd``B
z=TP@aO1WwA$4}lPvQo}H;6b)@l$*#18&4!4Tpnt#?9oT9{Xl^$Qim0Sdqhp9h$cDc
z2XmD|%*3&_ufnMz-u^_U$n+-eZfs<Y^kH>3h(OhC1c@4A*#)VOlOAxh6E88u^yx?>
z_N>AO;#rBq9ANE_pN`{|CW@3m!5?jJtV@#sUNVszNWznRT<y*kasQrU%qt2f0hB7t
z9!sH8`f1~9Qr{~xO`t-CvsVd9gs00@dgAAYt6Jw+<%7|sO>RC0QHQ0+bQK*S9`?Ng
zw(M&HY~EB-U3^Cr!BxkPuQj=^laEUi-gTvQR+@D=k`3jKj>_vv16ktyE}X+sqMFP3
znr>rkyk2^$9eag{kSpxhx%f>M1=Sam!%fuLa}r-Ng7`2v@E^bX*5+i)CUR@cuA_$_
zb{eQtkPuBfHJ!F1amX1Q3mBKm!P}}}x_D%z!MWsJAVN^$x*GLgM*uQgVk#J%RI}tY
zeARcQ*>D{W@tgl}y&>+LYs~Ei`8$uVyzBxX5Gr2;_{i2{j!&fft#@G29D0ee|I|*r
zyu32~==3V`(k)NkwK(gxOY^m2bU!`OFg~|~0BhQB#o>37Z$aOo+}r<L#7~>_)tCOs
zO?ATm7!ysNGp%5psVZcHypXuy^s90AMY+hl8fM$(j74*Lzv<v;{E1^)0!urL1~rYu
z;7&Z^swJGEui04Y*q}wGfG@7)dkHY~IV1G}N<#|c#5Qltt3~qcKX+1G&fgn7xYC3&
zbjddwxf0qh(R5?>r7>Iai~R59Kd<<)Uny+-5WSw(^T)TH<Z%|X&;Q9Rj>8ZGK8Au9
zm>>}yB}~{#`e7+#1AR0s-4Ddy!CUcH`}NF0rfjc5N`bqJZS}AWu~7q2Hkt(mTi5zc
zv(Ubm%82M1sA`UndD^gk)7_50@lV3t_Ov;)^!0UXuhNI<pP#%C5n%t-RpZRgczOnk
zQ&cmC2uayEb9?v)e2WqVnkd+U>a8+fD6M8|pf<8?h|$a_Xl`C;6mDDf`68<W`VyUF
zDg_jUb%{Q-lxiBx_&9J>dBureT14AP`^ydIX+ebEhw=q)7(CO}Yfm;Mbc3PZg2y?}
zaq&`3G{l3p3}E&6GE_qbJD~@Ha8+TuKFJ*FzB78<vi?fy;jPHGCvu1C&jMHDz`y$7
z)Gu)=nMzu*&lpC;e7r`@P_8Cl3B$akLMq6Nm~!f^X@IY%MgWjGGcx&V%mF9=8e_e4
z01tssF<(`H?@ahg{&yWEeaYvLv@v3U2WZ%}u`yDGz7)PvW0C7C`s8~N4A1bP;{$?H
zYX|^$-BCmvQ{~B1nEfc6?_-|#%_jOaRg()t?*D+`lg^N46Gx%D@DK**x%Pyh*Og7j
z;!GZ$O4xX5tbD>oQ6=F$bS<KK;)R${R6pcjUZPU=0uQ}mnw~ua)2eJF3D=>a?Exw#
z=${kEAyA&EXYR~=^fD0^wOn((fG!ARaz#6g);ukwR)Q2Xo@2lLd2{8AVgv7*J^QdP
zho`%PpZ|r#SN%v>ySUPzOXr)Enf=OO53-=tV-90TL8HCvxUp1wk9r-uwI8_f?(*wB
zor&V$$_JV&U->pcBbcothIG$q>GzT_4&igv(R4ZTA7jG{VqtYeHs8(`IKY3rW=rIV
zO6Sl~x~c$FMUXHW7qq$AVI?9^(48JdxQ%qR#tOqIMvDniAR{-2aV;+$S@{~1@HggH
z;bhU?7IJMwk58_u@nzSb9;4+NfpIZOeX4gr#P?Ec2_?q2<Ha-)q6Hu)e(`02#;js0
zjdlK20@aJCU(h!i&m>cHd}mL3q%<vgj}6qeu3zo{UfqrOzD(u@S^3Z*qP`+Kn65k0
zHxG$OvqoUE#?mHDyU6NE9t*aHCc)^vS~eV!#^Jea(EsyyT*Yvi_<(suX!_KLgBzJv
ztM5FjUl_HD_1@O};-F-Z=4<XEBwX?M(ud^|_w<|hK9t#;c+009OxL4~_;md&!j>~(
zp*jdq2OZgt15!JZ(>W1v92g{LUk}HsMzB*^48!2wEtNAYtm!VQQR&UdieXdga2CgB
zgdfHiMvGY)^#hT91T?0Br*5CeKD`x}Ve;9F{k6(#*d*#GqC0B&GWc{Y#{nI*GZHn?
zeF2g#L*iU-<?XBZUZ&EAB06Y5!<LEeBHO;E;ywaj%?Vb*Si`j63x`#RuGFZ;s~3zk
zW0^Hw$4_2Kc@5O|hVlDPa%a!56#C~%E6{{huX{K7c{x*NqwGIm-jhS=T7q*uzw!mg
z(tl2+3($PIWmz_@W)aZhi{mC^nzli+NDP$66#y|f8_-r}*9!p90||*$h84#GNdNy9
zo93XT&CcuC?LbK8O?1@<^^qd=4yngIPP)13*yPq!+55t<O-z5vN)WnyRa|on`I<``
zcj9?O!<&PmNEU+3SkR#?q!2mELIoMH^EXM4Ab;XO;t_fUTIf|^_KsRQqybqauv4v?
z9gk2Ch$ly43fwQB2Ga3`3Pg{wcD#rkRlhHCMM@&yM*1B2vQ?BlH=$z#+O4fJ!gjN%
z5KA}1Yus|D_^>b=uU1@AUU}^(ojSPWWRkoFMH*A+^7Jv|CESESXmKo$bTsvGu5kNB
zjune1PA?+54}NbAtWaZmx~1C}_;e81peT`We+iFwV^i1e-}(3Ja(z~S%PsFz9i_9)
zUVS=Sl3qTgSGIG%_%#iAoNhsnZaPnv4L+7W-;jUzpzcruMX_4Coj+@`iFK0liY2l0
zyU0t)o0YPikK=r#%{v9nh)h6ADrJ&3<J2;x3Jn%gYrL`dJ%;XQ&#=17G!FN>QeqI}
zWHM#(yrF7H|Ihdy!~g~~fXX>-yk8g<QobsdbJQDz1-VD``h>86g0ZX1v4Yv+XUhoI
z6c%|i4ckfPE$My5ldv>Ud5Usio07S4RrKA0U=5wyw)l3i2|K!aby((6nn_TsllEH<
zzX2+5Hd6B56xW^b((51copVS19IG?(A*v6QMv<3;Ekk;82DnS|MYHWCTKj6Q@uc-N
z{w`BjHrvmqT3}mmUmPou<zV=n82E7K2i&~>b}At|*XJD-<Sg8;V=C-0TWUz2uJ|nP
z5HTY*DY;iJ9@#N$lP-c2gO%ogrQtg^KpsR(khY{t#^RnlpcPLyrll;Za7G<kvn}*}
zcM~CAM2?nMKFlAVj41}o@~a1Hl9_qs9M>ExyAOD-)KP!|z|M020XG_;ixJXw4*%`o
zU%##8FnQ5&#3Ef%ykfo+_gZ`Tug8A9T;WM|aHrXiQQ^FWDWQr>?Yf;!+ZTR@SKN~I
z;(STT4p#4L^L%GC`0rzEak^)@k}(wY4G}~FA*nV<EC<^F<2{D=)`=zN9Y!#n^!TYP
zHG|}&62L*~{I~H7rZh*5Z&|g5RFm30T|Q8h7_(=4Mbd?iNk;drnRBt)f!|Z+$%Pwo
zR17L6$eN?<pp>meQsc14D#F%g(5QYXj&L&3GO&wdUr+cb&GBn-tJr18VAYliLKmm;
z_DBNZ^>4n4u$*0HJccs{*`p8zLS~vEcr0v>)`?no(Hwi+hCu=v7xgp`Bd-mQ#E+5>
zQIVTK-OH&0-s|*xiReBCR*t6?>g?xq8Xf-xCiUEP;yBesSu4#fEakn~dd`eC@QM`~
zq^UDk#J?Z{tmk+hPKx>pjRrLg%U<u3xqOAYNV}oYR5@7bTv21FT3LFhw*u+Z8Qa3_
zVYHfackIszF)1hQ$AU1rI!~j~%Cj1IB@+dyKZ-JGT}?6tiwMy7HNwfHXAlthB(pON
z$#w>}=ENkfQ}Vqs0y%5gCKF=1Yzmw5NzkRAw$vvq)~FIVJQ@dXcs#Bm9PdYxdqb9R
zS_ct+I^7h`wO4Oy@4U?w&;EES>IiN(c7KO38RX*h!;H}?CGy#N#Zqd65hS+oVAI^h
z)CID{4}FtPV(;MNDzg))Iu7(*$^PPFa8^TVxs3PNI%@p%yC6wm!Gi|2Un~frmIHZ>
z&mdtBt$W{VTeb=Bf|(f5rk%2z1D8>RjTUanHkc_Uy3>DLHirAJ7yAE(@&EofL;C-M
z_qhuoz|pnw`=9cfCt2eae^x&#3a>WvSyXe>*ovEPYwkysC4hn^$B@q@oMwdiI{r<c
zs9}+n&4}{KI}qA2?q$E+XKt-Xp)p$$JUdGF5yVd<5#V_NGtne*NIsU5GaS`@+jt~?
z&;w#U7O(hn-~S+=fDyU~3ok|C3tRtYaS|!C!mK;T%dx3;&1rrSd6MBItmB0hSz%Xq
zIkX7xriB48966*siyi%BwG{f{^wi&h4a|mg?;6Jp^Y!@~&an0nkC=+|Xw{C%2d{-B
zdvoPWWTfw&KVrBa2zEUPw9lxF{9UoWzk0~2$E^(Svg{fE+#bV*q_^tay4#skYvgDU
z8TNaV$7tVd7$UYN@n(GOC(|uSGq^a%?xJbaRpV0WR7zGS(blb{*9QE(I=bvYZn0O3
zd#$@Mn?gOO^Ex0;R{V43F5lnm&X9PrqEtpSGj{&~h!)3elbKilK3|8G5=Y(y55`(?
zx-)aw*WWjSPn>R`ZpqL$MCP%TsuH^S2RCd@wR5ip_kC3NQqkxOD6Y=1>kCv|9MGL4
zrUwei?LRQTt->P0L0(Vek{p?n3OQm(5&+ySD<V$+W?_7rwTv)1>3tvY9pa?$uh%fv
zHQ0K8e|XR|qag!TBm1#YurG?~41M;LftUAB_k-0fYE)^N1Zvq|MpoyR-#i`s#wLgS
zI`7=7fZ0~&@U2DP)r<C_n=ZXZjo!n)GMAV5e@x(0HCRRG)GjCI*BCH`*|1v+VU<$Y
zGR1c9SmrAo%}vh~6?9;5aI~X?z}T-@sKyiIezSUBBsT~uI0|e}^gr|Th50b%BVjrf
zoUERQ>a?_(RgYp>%49EU&sx8bFikwGv7Gqn1xEDp{vGbyC1YP<t_e<3g6URrqxd%y
zJL7W$18~V2A5IyTHONbJo@B~I?YN~{r^W)1Yzsy_m%<6aGdEpFNq3J;WF7~${rP;`
zpr}gYYFWF-1Mbm;yB_G({o6wHv1S`4n~h~in2LwG2H+aY03sm9%MG`RF-Y>nG`$$(
zyJg`Smk-p#LCdwMpfm|h+v!-rx{e-X5`rgT!kKSMnQrnfeHhpJl8=u~PL2D2AnT=q
zGMy?&go_$kGK#4@f~__2I}6+D@gvW6?8cjo*51$R){~+{+gSt8AKx#?QRwzJ<me~N
znVS2bHyxeh{6w%B5`<lx3R+#t&nNXDE$D_pOovUk-lhqglmG>|N0@Hn#_|XrEu7#c
zh69xh7;1s^Iek41G_2siOjIxSy1)8YJ@`CYZtY+0qt)%mdMo3E#*9m^+EWc}Er$4p
z+*TOIgF$t_u7*2);8UA#gyPF9(reUdc>o2plAxCzNHXgR!V+dZKN;n^sYXu*bosI{
z<Fmgy#k@<>CQ<c1TpxUvQp9`8_i5e64=W|Evx=R?MRV48Rfj>9Qqv)iR=I#Ajq>*j
z%myF2Q3T54#jWBiN9BD{Y2t-C2wr)c@ShU`_xgbKSLP4S@?q)6r-t>)K3Q_U&53pk
z8XK2lGcooZ%V6tRcsq3}+R?wBpM}y`HbWvpd|Q{E8DjYG({-QN)aRfN-Wj=*zH|lL
zQvF+^WL5--Q=?eQ!hLnQJNcQ{XMSMIT9pd3yo=O3L-<&znyKa(l;=<ThnU&Kz!ULJ
z;KoS4WNs7Dp@SfPt;(P5zi;DLd{w#ohnuybllwg(>SjDLy~wi=m_Lt7ACh(`y=Pua
zuw7u2yKs($+XbWul(vANHV{B57x}N;<^OwyY#<YIfhpH0Adyn}Nb5(TUjI&c&OZnj
z=_GA{RM8-zFMcV8Z%2mTi*Igmu@V7W*Z)8-^5Mal@|z#m2M*Uu*PLT%3O@*=epH}4
zNW?XqKFhpRnwP!GQ<z(R5jAEq<(j@}H>7KKIog!G!d{i#8?2(`P%59a(}&{YudyLg
zew3y!Fa;`2t$e>|(D2{bSmK8v=g2GZIKA`f7RDZ$@n-MWWzxD`zjo~G;EI%t<=8$`
zaJeeKH!}?MWycEnQT1<UWEEs$sb`LPh3F-f0>)7i1u~1*mEV2j(@Dgsr9Z#Rcl6VK
zePYP{^;Z&W+I0-1zhwz)Y`qE|<91T&SmoC@P19jJdy4CO=pp^`N?q}%LDYM0{Pidg
zfW~&s<2=fZueJQmzaLe;(^x%wjNa{7aEm6yX}sirGnEuAakRdCZr6FIxH<`NwU|q6
znIXW*H3Tuz!gHwiS`P?m<vHqiJ-)?_|9O+WBh7PBe8|CCDlW+Iv@l(p694~rI?s5v
z+yCu{*n75#y=qs57_k~FVw2d_*4|abj8c1~L}HcL#2!UaN?WRCYpdN>&7!){q3w13
zzt`{ey6-1>oCk6`&(G&L-fy*Ki%DbVcdW!6oE>#~;{pC4b9yAvkx{t+_v@5xsLuIo
zCJ=j}`%HlTi8`a10khQb`}hxa7i9YO-zM-q8|SS$Yjmeio7KEpYc6oPzqRkBXi*qF
z_C&yAr09LNss9D{3{S9Vg-cK2?;_8`Ky}~-`u(eWf-t?`D9wzpf}E;BI0oHluvbaX
zI;CzjBT4X(w}wn}#B4Q9(ToYv*k7PGZH<!Ew!UfB@?Fy3>t=nx9NJ1cZseit^xM*<
zphL&xtjJcotPQhYbYaoD&E}}b+#B@4S)o%1g=(v^yts!W<bvOIol<wo>O}^YY#4lQ
zj6883oL10v4W{GJ;=U%%^K~&yj!GMSzChD5z~VLzjNY~v`1&Aw373(oT#_BF*(yHH
zHjTp!N5@7{d_B4st={c|ybgx;x;*SLhT~J^t8K}|YZ($5`|`)Yq>)9nlcUSk0bt6#
zRCN8Tkp7kJ_|ihU6#3k4;zOXN)2`AE9q*z&+I#R%AaCWyy!(k1qwrBvl5UgcpC+d{
zO;@}fK|$I@3JkA|gx`~*rQQIp0#=Y8b>MXPW>?Bpzq&`jiXEup)IEA4PFDUTuexE^
zNgB2c5|KAkFlP#hZqYTdx?JcQCpVHOg>$sg!HB?)?5RQmZxtu1c{ug{`g7hi>!cUr
zpcgd#UdbRNHz*^lin7Koo-cS@ae3fQX^(DM^~)7M3B^1G<1i|p@`r+ZKPO%fGTC9M
z$~8wDi*ZioQ7q^VMgRocS7R~WRyrH7B?F21&w=(=B)u>_C_K_36{)CTi@I2g$3^)H
zYpxKGJ74X6i5`<ftZwtjg+9D;WLB0yyu2l57RdqFevki*I((94Mt$&Fdh5d}WOvh3
z>Kv;azEa_?*z0p(&R;i%_<)i%v~j;3?5-m7rQBF1Ur=n(Vo3n<aI6Q#?*dD3oLz>K
zAM@ys7{#=B5sXyWG|X;G`E+_$)sro*;;C-KV2QjQe-)Cbi{EY|jo4?~ypKSLz_(To
z0j)GXQY|XBRIGt{bYJ7K$@VgDU!7DQ(X&Vcyxwr$5gFS9;%hM9+$kSO1gWc{<#6fq
zrwCb}8;Bvbx%Bw2rQfd<RrSw)@}wUa_fo2d%yg@-UH?)w#2CyG*!1A}tI9J)@R;ku
z$ymUquNj}n>HG0UH)S_F0z$Vz@S@NjsiFwWwTY+&H&lUYG#jfC*kij)uZcluv+bjA
zZG_ucsPKgf$c3Zwv#|cc(a8UWj2Qg>r)c=^!$9?I7M9B+p_0vpb`JTRV8CKAOY@%q
zYg@!`;j>oGXs6l}3{#NT%I}g&+C~Yf83|I?d?{~XaQ#`Uwl>Y9U7>PSOED#3#Qj`)
zWf%8MR$FNo<YnDwEJ=Y@HMga3&Kv#A!VR8-IuWh1w~{zYDaB}N=A90}>C6x95mG*K
zJ`r%L!TCT-Scmwtqw3LpYt*H7G7l=83=dKnh+WY7PTONKRdtbQDQ-)YmC_&e|N0F`
zGcoB9{OKa-Z0I^R&zjMdCqvuc>Li#+;PTfr6O4Qb`enn*Pp*>mfz{-r;6HBIJy&(p
zI*F$tcLf4rm&r%Smmzvqu`I`T_;JtFKdmR7&bS4Td9<TFzpXrQi*82ZG45RUpb8u%
zQcmFi=?Q6oPLe~SQ*%hgfwCy>PZLAdT|ecsdskak1SHAw{qOZ<y4gQ)JNjzJ$Wh)G
ziD=nt2*s=I*a3U>A!qJaAs@CMm=O&N`598O?nZ+Cu^TQ2Y!^O%)EODoq=vojd?h;y
z4^9S=u2nKp(oi(&Cu8`JiCM3X^7066e-rp;vzWK(P>mZ4MPG^eFXTD%DxQ;9q@tlm
z6I2P>Hxr)+YmIgKMF(wQ;{whMA&7n9W@8s^ueZpV`o>02<BL~9RkekZf7$07uT^dn
z3)s4@r%9MJv@%Jq^=?uoQFF%Q@3+N30UKsq;YZG$nTy{3hn(A3ch*NlxybVcZr;kv
zxik&sieI#B;0AM%y1Md`46~v0_p%z*isA=Yrk*V1@4iOql_{Xd@AF>xG(O5IHc6=w
zEZSioUyEtW=&!jz95(Mesm@Nk)X{Zf6F`ay+vLpu#?<mS<K>2i`%kQMluZ}!K0??y
zC`C(}BmIGutc=iJQOfbOe`=VDR$^ZHDVs*H&|ZF=EbZ|Fc}F6Cuk9Ml!IgLD(Esw#
zNBrG%%4&=4Uf$l|RuBCY<@EU2<((mijaCZyQ`OZ})LM~f<dvnw;`8!G+3sKhXKl=<
z8o8i{FhwhztxT0IN_b9?1Ni4q3l$%WziVb(mXlunOT+DKDjew5H+mWn<&F1%d#UlU
zH7IgQV$AcTmTxfnz#0sHp~(^n1OzVXU~y;$7~cla)(Q4c1<r`)Wrr%T5ktrlqTy1O
zve-QG)n%&*D`$PRS8t4LSKOAq_+X3;U`sXMTe*ws+ZmFq-ku>*qhu>509FhWD4Ybz
zNY@S_#<rb1=X4l%YeP!~$b~_Cm!J0_WcZiY+x>?46?1P_)&66bzzI*l(CBrA5|9j8
z7@DiVa%Yr0!&i|l2A&cIEC#x=BcLc6<OjLeAD~jwvLm$ZT{bX(IX2l57&}#pql?Ex
zvbx<PprbF<3*k<%WL8!A(rorBeOC=#CsaP#8FCpNSS3*vzV0-A0e(-6gk_q-soKc|
zpDtivoNv?w7W1_sjNLHnZ8Wp0r4E3PoYmUO=g1c?Y^nHB^wHSK{YPp5(AO#2ri)bL
z8_*Cwgv@bY&Ay^TNxZi)5B~E3^bsEtD0)rLG|6-B8f7i{M@ISR(+YT))c5t60>hj^
zdv?8>D$E*U^fw;`EeCy9ChiV|F6D=)Xg<&_Gk4b?qosnMpqX<hWy`f-k4LTITSe&{
zi|>E*R)H#Ugj}{eul)+Gv;@kmU@rPA)2Qz5j~R(3YOQqTLE;Z*@W&}xE!v)U@%L}{
zwVLv=VP=)b5dXtV0v~c227?h26u_tuA)0-GmE66-x@6zrh<)TEZqB8>RhF`&`EFzd
zm>X$4dy|>1z*=P8J&rg0MbQrpdBbh2&03^$l&iTWN8KpzjL&B+GRPCLX#Hu28Bt<|
zii*0>t;@l$Ds5Mk)CCrP6&_0C>D?UyuQgZcTE$vEI`KVt+s|qaN1xjSq+4?V@7pF1
z;b!KCPF1N$%@Nd>G)dZyp!w%x!8B9O+zFr&b8lnTUBXTI%8t5A!p6v@$yK`WvnC5m
z=>0V<bc>CbqiU4*rT((Puh+TBS!CBXJk~_Wq0W<KdnrNrZFPb)_FJq+!7Tr{%t<A(
zkKac!Ihg{Abvt!^jk;Ns1gbPd>#ABo6cXgv3NIia+rj}GD+(iUiHD#7X~vEbkY(^n
zQ0%h#`&cqZ>q8;U!Se;9hl_(7O(y#of<@>>q4{}5Ga%K92E%Qbn!^qs4)t#DU7c(;
z59N8s%<6ro^Qw!Ip>fOu3B`~=mGk2p7mRcs(JiUTD+pi{$rbr_7Fy@alpk7Kj(wwp
z$OGGPfY`&wr+rn%B{;0+secx2O59lEUQK_b_k@l>Z)Wj)shoZ(%LDCF@cQ|(w`zIN
zG`Vv-g1haP<E*28rt($wkG^oRw_&lRY_s4u={NPK96igQ)rv`mIiA!vhj!V}k9&Tr
zRuy4Dd~X`c`siBD_=e6-W&ZN82p{Kk%Zo(h<!Mupow|$h2TyT6p0GfaYyxdZp1Lvh
zhUQkiy-c1BU6Zx|epdW_oWI#DIFe(lM&XyuAi%20&gTeVF;W`vdhMIj`@z3XRc~*F
zPyS-s*|F<hmG5q!yjDaTAq+;5nNwoJ9Lqc~d$%ab<8nc#^Um`&vsbV7mE!$O4JXT9
zvyZNqT>te8*n(EKNj+4T?D!Jh92=GOx`UY*YP-KTv^ftcQh0Zuf3+3l^3ZbtZ~vja
zKC3(cGW~2Ad*9w*RcIMa?|S_eMTjqN{=UBqzX$AW^u;ApkVkK}al)pZ%3NObi)?Bv
z#(DJmkZ1u|J6iz8-#d+WC^2K8QDft(w%0`PYbl?#tig>MaMpYm0L(}z;Hs26l&ch>
zm%Hgg2l62iESQRLmYP(X1++gOpdfbtK4l1Hy&^@)9uw&nv+KQwUq&=Je={S@vlUBL
z6qt3~G_?8pbJ+mqi7T~v*J>-N<%Owyl>jdHa|_geaR)maG1OA*2h4;wLkYRYjgTV;
z346V?$G=4YPNYTQwVG#YVln{3w+OM7g+sFVYt2K-8daA_DSA&DWg!Aq>>x<(2LzmE
zWA6H%JMdae*UJZkALp@KxFoV4p*UMXs7dp0{+;6Mi?fE}f!-0He=S;i<xg~g*Jx%1
z5?ed0p7*huB+oZ)2)7bOJAYXEcT$rj;nQyvrVIDYcpvd!yo=8zd2Q}d7_{s<h#(<?
zM<UuI*)Ru>1QX?GDw$iYq_O2^F=JDd*Fn+_p+6GO7dginYO}e2Y!h5@OAkFc{>d+r
ztf~ZSo;yT7;>xr3(Fqb;>Ut{S-Tp6#bDZrvkA`?bji)(ZJgO#!(U=WtjOVn&MJRbb
z)xvn{22RD6*{)c)E#512Tg*ajCdq9J`w4QuOhi**9HYps(pcWpDiG(BDRa)Z60JNh
znGh27*<saPnI%f=#H#|!9d45A<@yKLLsY#7=DqMs-=+M?7e6bDX2Un0Gg#dFwa{K-
zgt>{lSN7{16Zcfni{WbO1(~AT>q9$6fCXG{UHo_ZKR!i$DZh;;J`m(sH=a3;&0_dh
zN@OI9oopoXv)};X%so=ndj^C4QO{~D1jq)8_Pa3kEcEsZ$32!RDyKuD*%w&_XhGU6
zQMuYSdN$lFA^OInr2dQZ*-502f+%@J^$D-9=Td?n!3lf}*>o)V`J$pn5gljx=Vuct
zaxo9tL2tV*(>MP4I>P3jq0%MhUut$=2ix#jACjO%YxpLj#kn>;fu=qY15f?ycg|wS
zhF`eOE+*JrB=ptp%IVMDlmx^-5me!fT!i*--k=RnGD-V2NEtU3VO#_kr@0TzOW{eJ
zy-gmMGCK{*%bIRg#$vcd^xQEcFFjJ|bIYz4I2HpDP7C8O4q#loSugT7KTTK9k+GxM
z4+oX0g-g&P5YRgE7~Iv+Q~R1)ZJ-Xxl5af3E}7y7Jz7(nS?lWc7GCP&+F?PaLMLh~
z+p80}78pL_(1qjnrEfPo)*scHhL$BWq%ln`swnXyepf%-{}+T6(b^$Q-ddqzNWwo4
z>`R4s^rd@u%BE^Ocph|lTkqOfKpkV{L|rCEYV5{+rYAk`%(VqrGMqv9NTRA$2UeR>
zImOK1k8Q22(4Pj#j_eaYcY5WmLhU|X<8E-};go5O9QG{tme$FYHxjhl7cZ#|_PV-`
z80*#z5e(VS0w5hg3IFVaJop6IB~OVlhnPxsY}GwX>A&+&pWb58oZ`oY_(K(t(1~SY
zXn9&y=bQ(GVj|y~=t5e5Z`ZLC4s4M;?uCpMt8Unz-n=igWHDRJ06Po_ca1ti8!p-r
zdUXhNfK@iB2vW(=c~)02P9030P1`CO)-^Vi<LdnSAwb&p0PP*aCA1UjtWkKBM_DTA
zP;n7D?dy6<7xE;qIdOjL6>ZdQ>&eGb?w+*hfj$7anbv4hK&rn!_)q@Ugg10nDly&A
z!6PV<sDIkZEVMFnqriFb)`lu6bPCMH6N>@z6Y$GDUA)?0{35KT^2_4y;QLvd%Q1Ya
z<NdkHG&??RtcHjW7CrZkzD-_b5w*^jy8=5{>r(BT{F$LBiwd#ExvE$G3gMJ^Sh&}*
zQgzjRsSDl^W!4AfOPeka?CE}FmuMP>&^&N{8nKhTr=tEDUkEpQgd0w~v-e;%pz5dR
zl)m5Uqh3ehfcYL17nVMi(<3u_kPb`qRtk8i7S3dbq7G(1@lV0X1|zBxsBjb>t9@cK
z2^pqj-Y2BdDbZv461}kY7o@}1YCB9cAuTa>SF8#aE_jUVADcxF{tYx>5YhC+thN2p
z%U4zh(TXwFYF)Y%1BJ9)yUdxg+kw#o0i6N~F*Ngj`Riq;9uL=ZjE7PV{$?$jUdF`&
zEK*0H1MnyLY2xtw6mi5>-ZMTnEXf~0qjDg>S0`vUDGa;<O3u_uk2wH*OUv~Dj}pT2
za&hKkju{f2vga#v#Vj#CxA=l><ns=k?z0zj;-B}=&gY7|j~Ksupav*tq+u5BZ>#dt
zi2hHsG2R@E3=|V-Dky^Q0Wb}a30jCfx1l4KYU-tRW&SdB=-^ctXE^z6zg|C|pUA!G
zlNFEZl;s4zV9GWDgq^S*ljOhg9d@aOfQomuUx*~Dth}1{X7(eZMU47d*{!*crtM12
ziowD_YGUuuC2N<#Sei+Of8Wx@;($1a-%5!2jIpV&cwSB~MhmJ1>tJanV?)jGEL<#B
zFhZ_srhb!A=VaWcfLE?b3VzVeHxF*QC^-+DbYz@6URRZa(dB9U>AE|8Kkyk4cU;=L
zoGV3lzIYHw@w%thBNnF3w)kaGKuyY563+4BIf#+AR9yH&@4WCXA#w=))CTak8047I
z7^%+@7~Ml{KhoYBZiG}a|7RkXBM$wyNup^ToOmGpRqC1la-D6dYN`v>aO2=~g$|ZF
zrcZc;c)sA7Sq2;W(V!hVp*E#jO?zviF^LW5^iV&<Shf*TUkkDs46*T3pdjg0;tsMQ
zM+E=OB7ApIE(-ENgJri(TzL|uy}4`w%_59g>#yZOc22HA2@C+O@o;rm>7yv#Kr%Er
zkr-168c4t`C0@wm7&!Z0wY>XJ{$RbVLK1$VBZX1epN&HRzz8)f3(qB%gZ2{HZWsKM
z|EfF+cYsXOM9e^rK0DOhzJDlJ^>E=1V|eF>a%cC)>LcMD6N0Mp?UUsuwJ>puxwN(G
zhIk*+#`7mW`akE}f2EbY1X!Gx^S-DF2B_Ek<Kp`P*yM8eGzM!01Mj{FKcH!;{^vu-
z<$`}fT`QJ*bmOBi%M#mMjB+oyo1NA>rmft9cQMpP2O#+@J@liFU&|Y>`O>hXP;^^2
zLAoJ;Cs<I+qq8M9vsEV2_RcS}P}?wfozpf+DjzSP?ylrcqwmDestEVxSGW+~r_D21
z>j1qX71@`m2KW=Af*!6=%+ZU(pKG))XuG~uH&Z)vTUfypX2Vmwf6@p#;Q(#2AYnvJ
z$@;r4smNdxTO^&EdWedu<Loc>)C}#?FRGYH<8Tb$BSVMEl!EK4+l3MjeKG|uz}!>|
zMy=K2us?GGrb;eV=`d&UI&#H+b<Qf|i#)<MomLuUbc?%kpU+C8tCKV<4I{I{>qO*W
zWL+baJQSV1EL|D#NqaHtf!G*B!*s%r=6k#=gRjLjr&aX-%J1c9$KCVW8NF+Ecnt!J
zkX7#^$sPQpj21_3G}C02%e~0WCqxaH9QHQ0D$kyCyPS&QlIpV-g!Q4vZxewfO(=2-
zWGZTVK$&|KH0p@htIZZTH>wt-dafjo;A_iXUG=pt3g14kxAv4Lz9N6FG|b97ZCb|X
zW{ewO#k#D#EZV$d5<b82zB5E2zaIg#pM5V+z3*`TA_YdY-PiBo(0Q)eAiJR~{jq&o
z%kCZzo~00@ubUsmn%&Ev8R0fNDQpNTorko)d#uN-Dcl_V(#Fw#VrLc}V4_9FD)^Go
zUI`~l<?JBYTImE{sktP7Wyl+UDh9G(Aw?;KQ9UT}-I4~}Lk%TQ%%ObAgjO*KTuixJ
zbC@=ku-7IzEX1Ku%{FSRzIkFPBSTqv<$qNJwAHpE7U|(AL$E;4OYV-z@XNEAtP{`f
zsTn*6-LcC9wS*=kxee|{_YUqoexkU?ua{qxeYF<aNW)aQl?Rx=kqnyDAMgaLQcNs<
zo~Ek{l$1yS8P4bIBZZgz9w1-+r(GR}`y~7a(RET1Om7u;(*`Wl;%xVhEKf*3{u6nW
zhvAY-v62I6POEc|=ruK~2dM_<Sw_N5rcEjVDLeF_^F^FfBp@kRkS|@j)Axq5*Q`Jp
za`NA+c29PF^_qPIc5bIaO6J?}$%BB5#os@nM~_VVUn`)35yD#YxvI!PH)XY2=o==r
z7G^JOXtIqdG424(WUcFO>;+S$SlH0!4aszyNbhBj1w}R!w_+jqJD7h#wXz@o1=R%x
z-Y~eo6m}{_(;M}`Ok3!4n7yD~&K>OI?|TD>6}(!K?ehq#&?4UzF`<P8@(4(diao*7
z{AI3X45M%{;W-0HTuf@{LTBuh3*k{>iFbcIhp{aEMvvCAH{4zK9`~Siz(e6);bu?4
z3q0Hh1QLp#zJc&H4RzC&aO-hFBLCw=g0Yl2K~{dq7<8Ergu#Gng!U4E^?U9AHc9yj
zX$L*m(5pi%fs;G5cX>0G83Xs)q;#i8`C00hinT+{jE$x%?oTwN-*k0UoX+5B7*mrD
zR2dod`#vHK^8-_mnw-<+&1got398o!T-x`oaUnbLMSR4};di>a-J2!C^tUXFCxni&
zb5*wA7AKZ28@z^etc;7DHD5R`@H!ySAr=<qiyBEGh)Tn}U3a2x(-JFDF44-Ko+HjL
z(_9QfrR8S~fo{*&c1zNUnt`N*afFPJmatGHnC-NFbmfm+rN)42gkn^fzF=n>bn{GY
z1?2Ggeh|7UN7tV;Cgn9bl8HE|Fe{k^Mm4Ov+ub5JyTv%KtmRyRgMz+OSCQj8H%g<Z
z-oq?~FqQ?DIk!H(qHI>!ro*M-Lf_z+Hbrx-hU%+Nm}Y8?^!5};2bb>c-taHE=Nr01
z@u%TS58y5JU2E!-0zFZPmXNY9@;2M|6Z#gY?!fkX8B8(6CQnfB-?xX3;b)Vu-;zJd
zy(E<(WpXWnKQk$>A9dC4Y{iG6IcA`cn<V)N8GfJh>kEb~;l|J(4dhT^fkF^BDR%2Y
z${G?Z#!-;x(_uW$`y+6@M^}DHb*$BAwMntw!lMP%cZg`@_GX~Bs3Cv~;`7g$`Aq__
zNcK^!#aqeqc~C(>zcbOYGCK<9XJw;BQZWj^HSDv!wCfpY0r_hEC)i*E=X|a}bLoXi
z((}jFe;1&aElXaovxp4fT>pIi^FrfGPeA3kPf(vTnB#3%d(SHD=o*)~xHf@uPctY=
z$AiQ)9E_5etT41-tkTKB_XlrA0C*2rjhSScgOFor%|`L9Jg9bbQ=PZHKvv5}Q{GQI
z4cJ824j1fg4-EZ20->?dr+dV5_pDx~7WWcYt<6>aD>p0a+!0rclz3>d{b^pdeOmRv
z(m6juqN*@+X_r52ET>hW3o9MMafoKq?zdU}mcDI|F%>6)G~)adANa?W;wbCg_rnN<
zwxL!5$k<sJ4hS@sMo>>;SwJ~cm=|85OX<bh0!Sk-_Svtd12nk-`s^wU+wiKw+I&gw
zBipZ-K>(;=x2Ge7GaH6IXh$)(#UCToo~S~h^K1ese-$cU0b;lAuJQJP6EoyZ?>D93
zpV&b`^eDTBUk15-HJIK~26#^ZPLT}Hnc~46{IX;1-9{VqG-AXYjQv{s$DhdpGw;cU
z;XObylhOYFIb2jtkqoN(Qy^jde0U|tgWqPf(bAOj7Fq2I_Zt=BqeoJx#{)qjd^xQy
z#P?;mM|`EG_HASRJsS^Pnu7N@_}WaXhx1pCGI!9>bVlKTva*dSbtw=1I+HGsmmvoc
zyPtvt%42$}d#okyu@)DId`V-)8tjsiEybl3FZTq&4q>)V?n-U$;IthUOFvhU8ex{c
zCKWbBn}1oT3s>c44&LjY>pUFr?+Bd%0|^6pV6pBAVFxes75ic2TI4`Jr3k>LTQ1kP
z4Np^2Fj)MQBNXa_EL3-6A!IH~l%_b2xzKEN$Z8&;(O^Wqo8*4!8T?D%Om7pKP4iJ*
z)7P3?<1!-MQCQ~^la78BRnQB)SfS+_&QziKjc!)1MA!YN0fTKU@3;rlxqdH%>B5{Z
z$PUx8EBZ1r6mde};qDgnI%=}NU)X?PN$lE9U+pqw3{x0djW!xv-CnKU5)6(zJfmtS
zNcT4exvLsWfB+Q7R`(_Gm1?p(l5f!E|EH0FgOLjCg^+w$K&nGyj-gw7rfEZ^e0#>x
z(fbvQ7-D;|*UI%nJ8?9CnP@U!XZ)gW3y?JLlIOSwLot;Nvhf=g0gzanX%<rM@abtn
z1j3k!kytvfftXS;Tb3(28Kr4?`Qj3C8=QRFMBMVd)VThOr~Tl-PW4~VQd`3lHZNTU
z<cylc9U_P3lnN_R>+7hh4G}^%;F{{^c>lKb`A|pJ>9l`FTJ+!y2s~xTy+f2^i@CR_
zxv2rIyH;xcsLhY|W!<hv2&jpG5F^lOvXw?2SQ=<>pH=!0bFLIL8SP~8CNTCsFy+4T
ziRz)=uzm)V`Ht)*q%7wC`5U6l)M#WkaS`ARnan3}%9BQpk*?NP_~y$O*|{cr!AF*4
z#v*x?xJz~_n~y`Sx)bWj<E5Zd7{kEi^v7drmn4|^gylA9lVcS~O!B(#wkI!D=je1a
zAszdCd|Au2SbR3br^1n$e4m_BBxfV}DZ^KQR%;~=ZH>8_mT~{i`ntBAJ;_XY5y5(t
zHA~0CGe!s#z5g~rB%KF*FeL%Lw2c_0H@cVE42aWjw%#Bro`K5)8E*4z_<{Jflu&@Y
z$61Teaj_eHk0p*CUHsXA))>AS=bvHYr>OgZi<fMnb4ne}$kCN!+NuIlqOS$79NAzL
zCSi-OO&IB->Jl2YE})=7gyAl%Nc-f~jClSz%L7k@U=)4f+#pn8;cfe;<y%8GRuF)y
zfr6(it<qun{{=-;Bid4YoI;!*0yaNEQ&ATr=Y7nZ!wWiO3r4xUma0=VMZcv-4$+Qc
zd0E_q<3D1Oekn>SL*D|beO0^BWF$+a{dE`F?37k<CVt%Gwk`D6-;MljqA#WDwv%EB
zQ~7s}DH9Zl>nhx8hgX7jmIJffQ-xEdRkDZ4rRusdEiX%+U2eYyh1LcZbQoNB@pp<b
z8!-{Ge;He5nXyWj?j}hd(X&<8)BYuqWg;i|@I=VIfufy+gba-UVFTj;-tyPYzP<Fq
zP0lDvc%v$EgpufP7*)wscRwIEP$_W9el%q2+=kv~@xrC&<*T=`cIz0*?Z;+qe(ec2
ztygk!7L;@Mi7=ottOUE#_RTtYg<mBgED4<a%MPx6|M>Qj?Znt}_YDU&)zU?{id<1-
z6IO!`BRerbx388QF{C!Y=H)xiU=`E!d1y1db&e$MyvXLOHoIEsgbMJ2qhFG6#5i`{
zue)V6<p?!$m1C4p)(NH`z!u77WLr-Si2>&-j|W}pRvj{s(+s0Ig=W+Q@<@!^CO|D3
zE?O(bLz<_Nl>jK5z&ZQpfM%ZnT7HyZnTPqCse#M8(LHmi&c)oxEDNRiy^+v9I^$9}
zCCOtDqJkITmk5p?Z&WjT4-fcSAl<R&e3|NCugG^o;8xA?*!^>jq$PLx*rp|H!i@&o
zXFAx_r&Ajk>zHdzQHLCnUAPDBdl~K}WwL!W-KS-w%fddLQgkWB(d6q02r4OHVv9nj
z6{L67$5FaRxmC+N4UqgCTT0*w<+<@jMwXrRe5bc~g-amL#zjJsof2U#z=@Y*bBCVQ
zKj2GnKa8Q;l7WsBf7n#7M3Z1<PJRR<A!8}mXrtv{(49of>o49L`%X^w2LI~$@uQ)x
z*x<MC6MqQHj_9rHs~F7#hg>f&HtmR3VQmS!ZrJUP3*BrV2GcZyqgzdA@3L=<m3xp#
zjJ4sz?^O!7T%P`v2-B);Hp98{FAv;itQ-K$1X?(!LizHpyBC2qpw5XE<J%$@JlJaR
z4f$<~7-XW#WHOZEHc>4qEg!wU%g~)!dN1>&lkH4xPM^!p_W1f~c^a>y*U#!My32fn
zW^$k|?0=;!Go+H=6qs$^Mw$-`z(0Trf<dnrM;&Ee>~Jt1KJ<|H=Hh0KWq<ZfMM+(S
znLARG=PfOjbWlltboTTs4NWz^bYg&1AWseLSdI&9t&bXRgnxiglW&t(gbmV|W>FNC
zS${1g&C^UO2OzbvXbXHBy;_48*`kD${I;Ped2}!#b}O}4Xz2-q75ZRtSqA_w7wNK5
z(`Comv*Osz`0~e_RTYiEsAmNlLRq4Xx(xs_KFjxs;^o&d+qEEbi3_gD$2Q0W%M%Vx
zKBjW?Hw~HudVBqz5LIBC)U&b{`nHKY0rKPny{0KnI4AlS4gaY0zGq4oWrNsMNWM|t
z;?<Kn^xi4-=WA)BY(vq{$A2LLJra-|>?QF#Rr)#Q`EA-vo{$5}Z|Suc`P1O`yki6<
zmG0-+HHUJ!G<#RzZ5TtL24Q3S-ZDvpD7J0FB+m3Ar1a!-40LAOcHAmVS79ozU)!T4
zO>TVqtgcf0AyL`?0_%)Sx2h(dN~glU%$fkREX&uB0SgC4LtnijcS0V;7Z}gXB;XzG
ztYU%{<eW?;HS4b>fZm*`d2w#A=4vvomE_`jy=ai!G<{F{XEk?#x(foY>K49L;ilSK
zt{0FgU@Tbp9{qv+MrJhhPjlJB=t6tcbb(3(xGo@A#TbyNE|D&9Mb>8l6Z)@|2p9l<
zVSk%H@2?K)Jw^whVCl6+M>cdSL?C-31rg~QY(oKsjFy<1Z)bjj4;iKcEf0;duMVV$
z=(m(~jcny*Ke=kQB&AxB&mAFAUh=HgVB1rgd0c1NOx1gQY~~{hsm7a}S6)m7=h0i?
zw-aqT0sEZ=$frK#&qjG#9=L9<;HFXxa7<hmO>0|H7|s`b`AGBZz>-_?G4VZA;{uo%
z7I^<_WP5XfTWZ?1oK~AWosaJC@_L%ku)O8Bo!*pq!AEaGzHtvi_!kT4!sMONGfxsr
z!~+Em$IJWl2Eyx#))f6iueBtaCC`|xKZ_tP2kARTRn~_&=iKsGUU0i*Z`)U#!=mf=
zn5cmI7la~H$1u@T`m<a4tlxQyxnjyJ1D+!#`HG7p#RsRX=*1KHmV#{P&6BPnS|%hE
zS%tpI_VTU0-q>%byXSmu*PbTsV8gh#>qlIrPpfgYQONb|i?mw@J>$JjgAFnZRSG}w
zi|VTKdsD4c6**0j0*~pP9{z0fOP(GRevFkByOeE3G>DP1|A0fqr1cgU3Vjv2R4t=-
z_E=S_+x3+~)g!2tR=Ao9owgtQ49>bFV;r>U;GNGYvsTWmFjUFV_;(MXvusbj+kCz_
z&K^1&RUfrS1+QRyNDLe?HLW0Fc5y1$QsxU#Vwr)*cY#Vd>Hs37-YLzpz7JW6-g2i6
z?^ydPtjUiKs0td+7fjlKdCIF3d2-(KwdvX3={~LfwiwChiZBEydQ6P^#>4sQIcAnM
zi29ISfe%avW{zU9wxLd5FM(Huz*lAc7O#62Gx;7yK*^CX@}g^8pc*&ge#v8Yy37)m
zu2$@c9<Ss;G|1727H9xc4XyA>u9!-O(sW53o6-<J8eHN;<1O#&J1q0u{n8fYG4?cy
z!zV4j)D$BAfsIKy{JF2%XWuOt^^)Nipn8;RctMV79D<A=QP$v@IJu{-7t6*sPo{jW
z<MN;H_8z&&N6Hk{e6P-BI~7hvuayLiJu?$B*p2zL`n)uZ{rbZtI%c>P$EqXa%7X2z
z#kI@U6PflzJBN{{Rkx)XKJtF1?6sPG+G{7{LKJ$1a%|DpL|wK+J9;}RihK%s?*<@&
z`S=+e)JNwj+6tYP<*NUTX#>C&JkL>RHAT-P`7NK9A!yrz@!#Fu7ph1qHDB5$F{(Z(
z!NP_dX%*LH$+C6g{@U}9uKXAOv3(Pn0Zp5V*x_b8{P~V$Oz;5o{#BPWONZAC<`t8G
z9K+*qwsQw{n(lC8_R*F<pwGD4?WHSj`*XpnHUuxh0+_^~1wXzzTJxb>aO1j$vs0XA
zy@zXtDY;~ioU^!iTf{od&!BJSD750?b*=pb#+&%jFFtw;A)#|&SGAy_D<6$j3@Ov7
z4u$sprP7O<qlmT`VWxaG8Y`%76G}-Ac2-G00>aShTe<jI#7Ww`Ky6Dj6%5-Y-fvn@
zcNm{G4*h|i5I@c_`;e?krWHYpwmKzvam7se`HYPXlRx0BNr|b`D!8*cbb>cAbe{4~
zP9!biQuZTGcMEXPD&sT07Z&d`*R#O0=k{?m)54?vG*Wkx^uIT}QGS;Sf-y`8N+!|p
z^SwxW1MFQ#vPdRaBHpoK_3_jD&pPbg?#lZWW<QjN-YU@?)&IWL8uRroK5dk=5CQpY
zp^>?2S{u5O^%j2_^`mu#1{>;>WIn-(Z@silY&9LP<&g}T)}pldkaNSFc3nNcwd|_o
zn_YoZ<F>3@9eppMOx32;SG(*prZvchAllQx?=1pqMuG-AvHP_kZ~qG@P9&!^<zJ9O
z96g4?sD^ImtBosLJ2>mCtq6*)(K_Lar|#=MIV{vLLkl0($!Ko`ZDiMIPFX?(!{k%o
zTzsc8@S(DLp}j=d#bn}MX=qn&==u?oW5`J##J2E}k10_!+gH+w@|{}hP$~)HQee2n
zv&JEB?;6<%m2Z<f1c(s4SFG_Q@LD1VaAD!B1zm=JkpmS2TSQCo$}xBviA}N__Sh~f
z$*fVfk_j4k#X(T%GIh*hXuI3rI!wxy#aUTJ!y)e5pO+P`EUUqWh?0|@-ErNa5#uYb
zs)itK1sx@dgX>jU`(&#Ey<ls2BDaEjdo*7g!^RK3!Axs>K^R0NKkc@^N`8U8_1n#*
zy)s_oFogSTPMam%SF(e+@wfPTamJX8+Hf>+M`z?>+0(QpYXkAV)Bs#xq7F-7154;M
ze#v6xgN`=C%S)aWC6o;t*oaYDB8mIRPU-m0P~!7*{Jf!9tT>a1OIjXuyW8Qh|McRm
zkyQKVFJMnz`0o}<O45Os+76b!zPwVRP8YbS@m0Xs%(9usqGLo(cKCLX{fd+(^gETI
zw&RV%oGBG19BPAJp$;~Ug*L3;)@`62xX)cBuZ%uBw&>|Iy{3p9<agkX*KWXz^Sdi?
z_hQkEzY-Rl_Nww7dhNZ$h}{I+nj-Q-x7g*z@%>iR(U_X7Pu#)tQb+A*q|eOKqs07I
z6LH1{T1y|sANr_X3NIW~1TUW0!m0+{(@?mjlB*zqP{XRa;T8`^#bc4QZC;GJ3iWd<
zA8(-0jU++y@%qIKWxcL+f+2^lA7YFT6z8uBKsG4`u@O|7oXM<n8``K!p9Hpexad-X
zHU)NlFA~V9o5_M|B$pC+;lUzZOd2eKn!*Q}9sNJ2*$sjr;Eahnb8qM1wiuQoehd~~
zaR(FH+ees!elUhtjlugEsKyfP-InPt+bu0qTjP(h%oA;%c7P^4am*zR!=E1D4%s5T
z04k=Hbn(ieQzmJ;9P(R<olizKlamI|dEa*KEhQn8Ltsmf4;El~s%;U@J_tpeT~^lE
zrH=Bm`D&k)YLu0s@4Kaf;CFPlwXJJyCCtO8#3D3B789qR>u(Bx+)17?t#G}$88bzi
zy+&cvT3*^g77*F%1`N}Z5Vw#NC#M!#@9|3GhjM94N{*q6&5?WiLs`0iL0r;(O_-;n
z9mQ4jA64Aa`J<w}*vYQB$*4(35o+iRyl<=0qY5Zq+<9hZ!1s-YHlIxmE`k0Bc>K3i
z{yN+;SCshPOyVqj?B+97qFTfte`nb>Qaly@m`G*u+Hmhp#gEr~ZK$w$WYy_=VD-Fr
zxN$vF@$M$D{sBNqY(f7W!W>QSh8gtgYP1T5J~Ulr14iZC-Ny@eUs}Y08YtJ&j#Mi~
zfsVg8#~bIIZPQ(Rl0RW9<KbHI<$kk%gS>J-)?EDy9<H1LMM!H%r6V5_mxUeaJ3<C%
zUZb#goK`Yi?vpCT>f}Ou4=8T9a<S`=jdrax+7WHBL-n^7H}~sQ;=78ek=mTCCYO9A
z{YWJL@n>fH<j$3Ry9CbzG*wC0Uls&sL2AlPU8WULC20MX_26#mZ`j!)@weID&`~}n
zY@Z2w+()X1mMfSA6-!F!Puxg}YP|d|@VCRBVp;oxUXm^j;~6Fz?4PR|dZLVi8Ep_5
z#X^y+L2r<U(arQUaW5Cz{0u-<>&v(K*@gv^zUdF?ErL2|X2XS5jT-bZFLF_#NG`X`
z`v{%wFLkaQKR<*bM@=@sYfl&r^QYebQ?r}JqQvlN%gFc9&H1`FBhpO%ck%vrqNv(g
zz_(9*__nsPPml+1$I1K5N_JPFC_Ge!!xTOv6Yz3bMcrSM5yt%RM!Gb~X_Qf-W|jXK
z{E@VODW=YvkK5prO^9Y=0i3k0zM7SFePwPX$85_NBbnv>aIFM(spQB?JDZGHOQ_gV
zbJ>slxsaGw(HcHD=`FZ*XR6ZEZ%AgTS$@~_kn!lc?aI(6W71Sss6)CvQWN~dCd0X>
z@khhA$fkL&l#`(movZ{P+G8$}2nxn8emDL7v@B3gz-Td%38(|OsBhQvC#xAR3K7ao
zy2XJz=fS289RY>`qZxDNv$hHbHL#{sS`q)~BjgKseu5{Kvl~6`bt=)UJW<J19HQBh
zeS4SqCdc0xIIf|7xBP4>KC`{Hw4jxc)K2Cc?KUbMZAiE=uD#dn%k`1=v1dn~OgG(F
zSEQPcdx7R@Rgzrq`XY9aQDiGY0<gXPDUH%RKnfIcy!8;;dVI4Bo%hKYdRmv6tA4lJ
zCwHRe&9?{N+OA+o14PxgX%hso9!LhTHoX{f?~ldBla?ut(Fr;m`4IIik2bNE9XR^p
z8s!VaK808I-lX6|bkAV0o62a}D61F2{MH0~w3E$L2=@!PHfL)gw=WO=W3MPXUKtKg
zSna~Ldo3lscSs~z{G3rUa5h)JK;kl&AAF{YJ8+_;)yv(Lf2%TWr3^eBS>_L-rZ~&u
zLU;Gc0%vMB6r5?+a`|V>stj~qN%cPN^5BT$z?Cj#mp7)`T`sI1d|mvIeR_8qhs`u?
zp#ZsvH<m`@_3dhGUP1jPwAFdqF;z6=_@7d=z_za9G3c~)(_2+{?50^lmcYh%pW8Ey
ze?d1MYZe%6dL_t9VWbY$yA3rgu2|Ty)rkFl8;7@(Lyk3SAJxvIY4ne<Vj34Kbq}pd
zS;)-uIlE~0cWcBn-~<*8K&t1pim`c+W@!LakO#?zF^o#N5jd*~^f1|Mb6h}{m2+KK
zR)$&GHQ8P#m?e-aH`jke$dZu~b@A$wyQjuG$O-vRuGF(V-}I76xM@yyX5bf!-d2aO
z`64)cK~kq<1kSSKGai_ezreVxc9}1|@z{_+CYM`Yc|QWD_y1Ir%`D|x>vbcb7R4)h
zaNVwPV>%4B_c(@A38fWlk70gqBGUZ^vI<exzWuVWp{yWUvUFBLQfzRFWxan^C7R{e
z%K<MA`B!s69-6@IfnN%;Jzo|1ksTzot!r94%0Ul8M2~AqQ?cYRTeKHpg)S0)>RygR
zJ4!jTY*r1mYG1d${b>hE=YUO}l2j1*zT7e1Z>7&4<0Mirr0MiVn}|KZSqod^UwP#8
zqfo|hx`g+M#}4`1jII7WQZ6_8#$5XSv)X|RR3~E}1(A`R0h%v4s{0pI_`=c7A&tjV
zz`x?N+1OVqsGRywgkK4fnc(2+N$?PooH2{eY~Xs&d|IEue5@~-+My<~9&4k`abb`n
z|9!A5E5~n5ekt4IZ9a|{7i`&c<^F*(HQ%rS-oFyR;uov6l=N(kJd*NV?^MSCrs1)S
zjq_xyoi3b+)T4hvndah~^i!cvxx^z7cDyIzqw!rTm62xA)cAt@>}nLF4u$N{9$O-z
ze_9VdLJmV~AU`VU!SE(P6aNEJJ5<TY8Q|&*t$xmZpOA!%wM)EVeEhC0kk|PU#|4Yc
zAKR1$U1Q1oGK5TdH;)h?w*?Z4-fyan$Dz8a1gB1jmnSjqFaxql8w+eO{sm#QDsV{l
zoFqu(k%Rpz!7VF{sS31taatzwBDCS|Zt>N(0d64Wh>!ANFrq(GV|Yy%|9*vw_vF`6
zt*^-O!hTK;t9?_7ctR63^w(@+nN~4pVObCHE=zm+vq&|gjk~0{gZLR_o2aiyK5OU!
z(v=PJL5>aHauITnqT84&E*Sk8F#Yk>sbJebp<W(4Yf*=z^6SMS+5$e!sa3bVdQIn(
zB~<?FfzBsGnst)uL_@92t6Oh=P1FkyWtE0#B@Dy`bdB9<e&^rT0JW$MWA?cOJabsP
z)pO&!7kC+9SlscUDn^4X1T96V;-1K5I;Py#b9*yQFHQ%*z5?0y%4>e46LmkhX+^O5
zQN4Ipq+b8kaYyQv8GA|^`AIYL<K=k7RFEUSo<>$z+-^L0OEv!~U|BT;h@`j7h?+qs
zYg7!U*k0Xm@8_jb$oEE1A;DU?0CKieMD5WL8id(lVSr;UK(|Qj(nZ!Hz{b`9W^;ii
zM+e(pqA(@Tx%EfkievE&NK)26Cv=``ZA^uh3?hr$QfHdf$A5R?-JxQkM=lw=XAv)I
zpA8uMj;kMQA|z|WO*}8Sbp7&_(Hx(0(f?gaj1_=PV%bW`=z3kvN`|u4@dGO>Cttq7
z+0K7KE)1a`?g|`zZkc+#j&M&0AL;WzLnYQ%AG3XCvn$_lyn>V7pLrt9EzIk|!Rd+h
zCub_n_9)j!==<6-kJFtslgkU(s7^>6R^Ep+_<~b&tH>QbXwagCQF=cG;t?K}ZY+4C
zL=RTGb)Shbpllsrmek8hZB>O#>e=SvIbBGw&yZgavezFby_>~$y-s7}y?~g$w$#d;
z8?el+DY(K$l+VA|cpJ==ukeueX75dXL$&I8JW0yuyG^BNiHz>o&-Ornx+`Q(c}*LS
zMPZ_%E*~uPC)AT&CU+dd<ap|J53z#>HlwKLO|POgsfk5DD$DED(!(l+BHJ<`<5lvd
zzLd5}+}}VCc436d@W#?dol=faTk)Kt&3MhGdC~i7VEJ~zI1oQe7B6x{2o1y}=AC%b
z>X5fbO>HsY5i|{)%O7O>U#@|4f~K1im-+HvcT<|3&W09`(+6D9!n3t)C-6)EQwN&@
zH6%K|w4#)8zGBeXk^bc>L$+=BX!wm5dXRnLA-$%ee9EIxm?j+OWD1S4e$os4YDPls
zqkN6f>i}-Ey!+J@L&aqATnA>-AW8M@eZ*&sin=DW$uX)btl{qY>srk>nf;P1hewe}
zM9Y?QiPUl<)2h(CTLY``#&1k6%|Vtum%4Q^iXqls1IQr@h>ylKi~}>9ykT*u5GW8s
z|J$sIM}u=ILC&m5k;96U-#&lmtA!zJsqK^ADX@@~hx%S=E=iM0cTXuGRH9%I3}?EH
z*qk2|$DTwSn+>}5^pTZqlE)-^!v9*Rq>|o0B<we{);OV;Oq01x@TDSw1hdhDjSwAf
zBmzQBi2mKzEE|suM3W_UEJ^Zj`fqu}HP<l&J$R&O`q_&4dg#ht;ZCA%lW~r2Q)W<s
zn{@6LJ2`mF5P8o$SUnk|C;`Z5wuB<<nHof8-$VGS7;7&o<X=y^?{-SnW^yEW_{KAB
zLn{?k-b7Am0Wm4g0|yv#{vsbhsDf0JoaBgo7+DKTNW564%aYAsojT7Ox84`7w!~@y
zM11=RLdUw8RG6F;e`$8L8ghID%~<R9Lk&O^SJDg31E=~{;)v{aD@~}Sw5oB6W_VSZ
zV`II$3UhEZ_5`h0;QG$sTH3U(99$v2OpwU*=eND!ms~@MTu#2+MM2G9b}Ai0xcNrz
zByEq!+<8iC+I%aZow;0SKWU{&%sEeo_-E0ptRK0TAZeJKw*aKsr#EuJ`xH>zf+tOg
z>R7j_;x#*2O~?%23zUfvd1#wb90C&dr{SMWDwY94N19^X8#pAF@#61g>-Y7S+2`?f
zjQf^g?cS|lc8Rl1dtLtkUoFr5IroA24)K&TUx#c4(1RFdmlE(;3-ElRQ)|?P705YD
z(w1(uDzQbDON;Ha;kqBx;ZUDWaeslZe{)Z;B-bi1nfG!~#4Pc?B$xv?K7vBLIos;v
zAU9bfCiaTuKizs)p=L{ylhr#W<Vi`W{AD$7$bH>LrMM!Jxz~dMsC?*)J<koS(-!(j
zyppd9kh9DBi7{J6Eyj7X<@p~*uK|HeFV?OwRlxe&pY)LJPYuPwuOqzDr84LX?yUJc
z^V7HLbV;BVs$VAcU8xe|Qbzc%`l|yak_&1~)kAYdZbmyaM{h|h*RSm+HICIw7m1l4
z%x-&DF_z1}a_My#)-yfMvRm;UkBD|Lpq?B#EPVp>>MVt6nHP|FZbIZqX=KZ)etCeg
zURN9KJ}{FF-%HH6`7}U1MC$}?q~bdI#L0KQ*|Y&wL$nTBu<b1_(`zvm$Mdc=>YmPf
zf$p-rHYy=EJ|?9yz00=iCmzIv&Na;&_)OiSe{^GTC%myI{;SQ|I%YKN{G0IV{=&DG
zRjcoGS#BVOYEDCGYZhLCwwM`yYGy_$7sw0(U6w~h?^O=%l%EbpTbb%MHm3&TZ5XDB
z_b%<WMWaZ;=Z=95ovu=f(cBL#BBwAb(<a&t8!;svgA<x9P%GEJ0nX_ByN!EYaHm0^
zFrZb6{vZxGd0mlKTK}=FatNLIZ27mSnspND2lG{)Q(%P&u~A0+<*kiXKyyt2HIJyE
zM%Y+>P3%##X=EgF1)*vKzNI+(8!H?JNwHu%Q`^(`lArDt|IS-$n5ip_>nN)fHDgFM
z2SCLgxw<2g{-3X?bReqVdzi<n(k3v_Uo(1Xr*0dpwuv1F*(r<XfpB)-ZqZd>l?P8u
zM?7BD%RITw3*}p7<xQMvRUId1dJIIhDk)r4TySS3D%^M#tn=fO9E`IlDWtL-kk{dM
zHzPz=8==?x__-mT1cn4+U@TfJRpd;OgK(uE<4Yic7l>Uod__!t$4F*Vw1EI@L_CAh
zbvCYn24=pNzzc4bxgiVgn}R+-PQLHOT(n)Bb4Q3*+v!3upGCSpTB`{6>&mMGb*ftT
zKE6+C#z~erg4p2htC#o-RC~Sb!-}=2uz-1Uw)B@Y%cZ`AwusoI&{WY>JxpBfc>U8p
zDRi41d0beo`p`d*T^z_qRlhj7C{>6)lNIV)`4=Q7+!$;ald><VrlOwQXAqkHpGAsC
zq*rZ9I+!$ZvDJ{+*A^onPGk)5QMllgVt30lKV{|#ba1b3(DrHHx7h@gwaflzC4Cu2
zk?!^MbSz@L+VoCt^k<}N599mm_1a!(Dp`gRBKZ>;GeTb{sPw*KxCP-u2JfA$leI+h
zPSkWPHg+DW7@#k-lv<jiRq%UwK0)z~(Ey0STCGH4GuN1j3Bu9)N3N(2oJ=#sC&b}L
z15gFg%#3MN+Dt1PrBH)?nYhT5M<2te_gcaXu$Ro|8CooMFJ2Gr&85G6?X6;h+qVz}
zVz+MaE$Bs>Ur)Q5@XJc0HOY5M7r*piDP0rZy@Fy9&BpC|-LEZ14d>*&ES9~jVsuUC
zX@S)UgJxx}H=!7J;mDaWuFKV*XDBU|=~Fdz9o>~Wa+%%<W{&o-MKTRb5|9Hq0d%v8
zNODO(_JaW1?*)qtt<Id$g2Aje23ND<`9`0lgiFm{g#4IfoaEfLS9(w@M*k~5U)*(U
zsXo`Pl;`fENAE2`NvEDQ6Dguzfq_YQO#m%Qo{{9+YLP9$f@dK)unS5U*V9aal;lE@
z&ncF)ysuo#@%AxiOyUzUd^DEgsuKGz(2$H6$Tp=`x=*oGAuR6jqq#!1euYl1=}zoo
z$4JCg2@Tdi2TlfR-IFxC#Tw;(I(C>IDq(%Oa$EO9Cm2?ao@Ep?7ujxg$OmlzsJM~D
zgrULn*P}N{numC-5$qZR_qY_By(jT>FiXTF{<T=xuQBWak|&KvCuV#3Z7brSJpP7U
z6_*?oA|w#@Xdg`<7MaAU#@q01k-KmVPGd>KBeXB^Naw23%hm+qJM0vk({phvp|ydr
zQM~C(Rf=&nSs$nqc_ehWoZX%C{udO!c(F+*qhxNWY}_wskT!st!nT+8k=s!@Lc2nJ
z2bI2~KhxL5C>WnMFgZU!JPa6jAA1Fx(g1h61?XoHw;8+D*7w$lxV1JhSdrr3rjUq9
zE6lO%P~Tn#y!IG)6syEvCp5122gd~r45qJ_|D0(Jmn{2)TICMY{$fmDU~k8p3OQI@
zpzV9#TVF@}r*Gp$MJ?!s4DN~;Zp88-6H;t99dE%M9HRH?jWR<_^SK|xsOo1YQ;NiD
zPOK;RkNles=HcTvlzit1ZyQdeG5R5D2rULiW`wf!I<lDR6=1F@_zR_ifm?~Cvb-K8
z>UP#|mVoss!7t3;tMc!)4<rkbj~zY2q7f}lb8e!wW*jI`%gSJn)}r_Sadeh(O}_uz
z-$o-K4WnU`Mj4D4CCxx`bf`#|$N-U094U+uqlD2gI;2C?FH)l$L_|bVQYB3AcVGYe
z@t*E=o#**EKF9n0O^VQXt#9N29~&`N4HYNpdIjt_@=IP-a#$XRF(2V9OqXW$TMzl3
zyz;xEyjdu)xE1@6inacv_>x(CRIpP?gJ!zRudMFquX4z2YxvKqM5~tI5uQNW)6x2R
z@eiB)DymG9Zt~~9Vlajwy)owvEKY{I__g9ebZ<a6TUSN;pk1@ekc_)Xhxl8leeLt#
zPx<QGw?c4Fq_a36=AeTEb2twb9W73|z2=P%iYl})>S>|9QOHw)+`Z;l-TNbg|8o4@
zQF5lx)wa$kyO3BF_eL*=i_aicfspgYA}Qs!CDz8Ne6xJLtg0JrUI|H*$mY_&^APQ-
zqV`Kka2mqK+r5wxlx*2X<)r)tGzl`d5R_)1xy9SpePmKiH&+Sz@w-l$+AfyWy;$lE
zMo1gb81s3Pl%;M;FEs02#H3}@+eX=BYr0#RWz*C`5(XeuHC+JQK-W_y?_$8lV}2`u
zi;k<#!`W)am!^<s-TLc&F0tj51cCV;m#Sb9yN!0aWQqoO(pkZ+dsSOdS|+Z;f~peD
z`F0O~?{kZ(8!?f#^Q^oS5b1?tFK0>5U(lB7VhuHGKtN6@x;4l4#TS&A){ZmxL4sU=
zk|~uS2|ol0bP(tn`o#~X9x!DV=_|k`K5OFKb8FieQvBTf;FPM+_N4IG)bG_$YkA|4
zlzZG#mL4Hrg=GI41z%H;W?m>zJomVcs~P=cEksvhMo74p?ao`*!~F<a{lTO)<9oBc
zivZqezWS7f0|uBJ4e&1?H^zIut!~N@gRIA5=W>Qiok&CEF2aKW36Z%;23}w_HPyAj
z`y++GN8Y3sTAWlYLkM7PKg2mGjC>*jV8ziKHbs2u?`cjvA3dc0qf3(8=v()>B$<}g
z-3Ngw+~XmuOF_&q+5nvTyJ~Y~Bjj4S#~Uj|Y=n}lSQRj1S!x$?@21YQP`bTs$2-hG
zW?(f5j?h>PRtPV3%LX`-d)gG*-D1Cw^@s=A8<jD2<2xTNC>he(V|Z*DGyQ$&NqXL9
z;ex9zeV`;Hks2?rCSV;$yGM!C<t+8weGy=3PG<tjMS%qtO26_>qe?;Ok;siE=Q|n?
zy5)!TD?%~eM(#D_v8#GB_!;J6d(wI>zhd}E@QG?|nP$Cu_BiXCIfL7&UD;fV2K(*`
zXa05@&$jn&kA4a6sP%QH_(fLH@JQ*0I~iKq#H|s-_!%Rx(oaL($d%aANl{m|49RXL
z71p%Y8@s0dBXr;64ElWE#D}g%$7b9W{nGer4*cDPL7-_BJEZ%ZeOyMQKc<k+t;fW8
zS4HwxA<l^Ne;@@n6!AYb9c(;eyYOG=eZdds<!^%~++QPKMLk10*s}rS<QLeW(y|pa
z^M^T-Zck3_YUilQ;Ih=H(ui+h=eyaFOR;PJ@s;*ehHRjq&9pO883&$u5k=BY*Tsw8
zR9|U&6y>fIsa~`EWPc3(Ucn>Bxe8%jZ^Nk}Bw6al@zwU&12H%h;i^hzqP*%Ra$dhs
z2QZx1xs4UzUYak@mCCPWJW2D;@{`)zvtOSfXC*uP>|>XT4$V@>HLovyH01OKxPgWh
zAQowU!DnvQ&EKBYh@%Sy$F6=h`RtzePLAjWVo|rF>AU-y5Q0NE4du5hf#}i}AM&%}
zqDbYmC@}ixMtKekSin;3CGmAnUo9-)I(VA5)FW?24?E;!_^B{h-I}$ZjMgm~OB6(j
zcms-QR><ZuUovG(3Tz7TSYN7oMg-)71FF-&^FuQkjeuy<pe6aZ@dh-UV5a#7Jz0iI
z-`Js7On)tBnyh<z9(CR80ulJ8qiQ|u4Zq;&8@6Rw-bs!}sk`qC*<|j!_sW_DPt_|C
z*bL1hzFKu%aXDEmAR|~DTg;}6f7HscNkZMf&gDxD(+hZ$YL2Pk0^EY=jRMMhhf@k2
zRpBldK=bi*v@CBj`iN4>TOSQc5R?fGNiWu**{rmY%Luk`DrMt|kZ8W^WHi<U-W8ak
zCem#Eu(C*1Y4D2}*cGeNb0#Sgx4J-K^)DR&Ew1+97Ga44X)uGZy-^(=0Y9ouHjq+!
zX3vxszOU(5;GZc#y=o(X)AaYelDT}obj)tHtno(sfbgS!iQO-Y;o2=nK`No*GgUZ4
zCi88KLRmVgnpf+;&rVW7meT0H9#EkT5;>mI0%(h>2`o7&vc=+b(0UL~Mm#Zz9Pp^F
zZJZ)tnP`i#xI31Uul@V2#NJXn+O+-=c1!$o2UlCHzu<Xpu0ETTti0-ZQ-UDVd;WtP
zjK57;E{D>ED_IY`y@Z4KyH3<fA_E}-Z9=s$w0SW7w4l`W7aAYD?KH77r&i<ls%}*w
zRt@I{zJ{LH$JiesMzGIsSAQc98#FrqCMtqQ?=S$d4!#7DzSK@W6EirVdINb@6JWR4
zy-J|FdBX$__wr*adJn=+gSJ=}=4s=xBg1NQ>2K((VCA#QN*_su_nhT^FRWzO|EOh)
za4B?kD>@Tj$=}^!QuA)KEr>e=aE^hW@{P>>%R7F%+X`T<#WBdnMD5W@U&GflI%R5&
z4g{>Je%TWrR~nFA_M-TJ3KDeYQsURkc+pcz*I_uIq`=#cK!fElxEvnNlOC>e*5M?f
z>`$=O8AuXmu%Vf{t02uSYHF~VjO7^XH`TgR#IRo=?fn5r<VO9&uQu^-@2yt;WszRR
z8n%G`2g-5CICDsINTa#8-%7hm2kxc<L<FW&c-f+lbthGeQy|_td3TiTBY`K<9h<C(
zy$?$<IelltQFCeloDdna@9%B?Kaitv)v8TP1v%-Wyh;#N!^%P{i3Amy0ay#kn~frT
zJ_4a+Fxuj?HoeaeGUQQ(R(s#!^?Nq}RYPNMD6BEkxX`*fOy%&(l(xux`TU$xU-g7Z
zIl40@k}s%JJ7n!S!VfR_BlN6V7X9Bt0~wMe{6_AZ*lbC`S{FaDqaPKVz|5KJcT5E*
zO=8T+MtQbic^13k3yv)StyV33W3DArj$u)!M&MC{Cx`w>0dcKNVh4Cb`dBRGU7G1h
zCK_|G_6mUiEODKW22WVbuXL6HG^ayCm{+mc>t_dMZhgO92`pxj=-sHy1Y2+rE!{oz
zA(jBlx~>KP*7zpfI~Fo5#R+h^BFM0MCMHWNhS9zQ<FuU&is(vklx=}M6tJP_QF*QC
zh<uo0HVT2%t{Lq)u2v|xf_@wA+1?e&2ybLf6mA6lKn|Mly3>*w)q$cumyxOx%#a>L
zAn~#x_1TReCH!n}56}$8bk)c;y$lp^?G0SHA1|*8Yi_)}Y4tGyzI)(cn=ng;Ht=%f
zz-X(IK2wzXW_uK}c5L&!5b|)lg^&4#Dc@I5uDaXHDd{uX>142gx!yxp`&V+GcF0}w
z-!x&V75-Ks+-N(j^4C&`IB#Ad`nQHgJ%;hl6!cBN*3To8z+?Vj0LA%Uru*WDQ}XH|
z${l@OoHhda%;$@Nf;0aPcr8X5weP!ZFU<7;9uu2|XW!A!9!WwS0nL0TN3vo96_dH2
zr*rPUPji#1^H^j{b&86n!~B)>zYL|0kNen$CBZur;J!&6r54^zJoOk~WizV@{Xe0_
zzH7ajpX1Z>ieqC2b9aqj3N$<x%CGY_dr8MqjW?ZTAEN#qrmEz}pAQpZ2Y^=PNVx~j
zK}`OBYeAc}00F_E(l1L%MN(H))30*m+SLx3n0%z@&e~h_y1P*JPzMxvZkTV#QrIO_
zAysSOX=wR6#Pj^3_CRfy<f^e+dQ77-VJ9He4$f&Z%3IP_P<y8s^sM+b)=7U%pWb+1
zbAMz7MqT*VHop}`PX*|OHcvkg3TJ3gEc7raXLI43<92X>;K2wN+{pFb=*_XbWI^3#
z_@)5KFz^+mXWcC?As$yM1ROZRQdbs=t1<w|Qacz(aeUTn7cg-I`L|Vx)A!=NX3Z_6
znH}%M=4G@4#$QGgaBJna%Qa&vQV7>y=}aR(p+UK^E*j>Z0jAwbZxGeZGF=&{;m&Bb
zlel<=zZblCxaXrp@WM!Hj4I9Foy6wb{~?P`jpw9Hr)*bF!1GGn2;y8%{ws7&ve!<5
zWdHZ`vOvrC4AS`mLcTp>OY*PsJp!VE)281;rGOw^t;wLZSQ7B;hzZ*#K>C6Z>kVTj
z#-yp%_0r$%CxNYVU{q`6#4gX?z##-n%CU-$U1Q>sKIptQV>Zm=BbCtAankF`q9yDf
zw`OAbjXLiaCXR<oKHqHE9EWmepFJ#>X15X8CEZr&UxQAsSn_B{Y1UUM%%rg0b$Wuo
z=v<Wf`CL<aLQAYC@rE=F*!bh=jgJBD$l|i*uUCD=|JfI>Bf6ldn}dt@UDM4Mgk%!r
z_@BI!D6Mr(HCUk_im!?l_ASZriOyuI?=*?>hd8z?ylTuUi7Zp64W`pN%jk2x9w}^6
zAA;Rx0>5$pX%qHmX-2<Q&f22T_ZBupe2-z<$bie)bU_yR&f7tQvaH)8P<v?TU*Vy<
zO6d(jPQp&mTg5x=>~9UK;JW;QMqDa%^$JU<R@<w0iWkE^Srla>DZW(mjqE<gJojCU
zuFufRP<B-7t(k^6p0dj~<a_>JHVy;}OF}OZF31u0Tu;H^r_wLIg5|#Gsi<ECUA(`P
zxwHIRg~mKWwOU@EhZyKc8)+wr33<!AZpGFrZe3U$`<$0iUeQjcqpHPN1r%H%mb;^S
z;$4^-f~N1y2$?HqL#DBTy}rmx>bkngL~!0G<O;H2Z!<CUT_S)6-F)Yre9F>`A^gNl
zYu==Td5$iewj!*G#yBkmUptpE&f4U+6@H{mKTHDJ_DKWCnr<?#nL~_Z&s7eO#6{vm
zBq2Ca4QoG#nfQ4Nt^z237nh;0+UXM6E1;=ic~n9xn-I2cTduE&1FCADIAe{Ank~3o
ziSVp=LXF1eQgf5%>2>+5dbX^@ds%)gl%<?BRe1V=S-wMhkVb6CSKI4pXicAP=3$;W
z5|~Qh(g)n6pgkF|Oo+H0P99E6T-`R5rufCsInn=^1aTHc09t_RlS(A+K0q2X5Qp%&
z2%UlQC?+~);E!$u`%;e?H$0tdd4|vbhf_kylaOw1szyWaq=!VKO;fXs@hIviF04<a
z$t#5c@}(t_vbQ{;4{kBJ)gy7cc$W$cuL}#pN~1&ySY*b|Z5!;{N^~o+ZM|bkOTuk7
zBGR`HEUkz&7TsLaf(NwwcTvqninqJNzj#^^+agro4-nHUQI=x0*wf<brI!;1Tb`r4
zzZWp<yvN)B<%r0bZ1Ni*?0MCrt$<TEm53ePH#gIrFaw@-3>URbrlht&<E=ub^HOGY
zWz@O=)sWAs!5h|~s~w(l<IYeXs=#Q=tC8hjjN#{vVmka%Z&5WPdp3Qy2eBAhRbEXg
zuzX-}c&+X89#qw!YB8SH9!1U-v5wrmkfQg?1bsBOqv<4guf*Vc1mZ|toN>Xg?h|ZA
zZ!@d&{{j4=y$%-xjsAx3HA7sK+`S&~l=%p-%~e^gREdluP6fpbITC;b8fjW;gNs?}
z)OWGmSHt|qMJ_tngGE=-g1(PN!Xb72-Q9Wf@f<?N<^A2CeRi`jSJ<EU;x~tq4lKC+
z)3UE%bbKm0G@NVxu>o?9;#HY^^q2LkQytWL9iWV_vQe0%KJx8^WrX2u;S4vA&soii
zh(!m)XJWB*Le|Jm`li$ArQRjo-C#P#_f@*9s}WY^Os@zYwo+aKTMLTe<=mcGzjXfC
z+<<gOo1-_#y{vhJ@DC+5^<h~Gs-yo03x_@e>D2q^a23c}_idDBd!jXLlxZYfgtBKB
z&dTIDnJJtUo9&VuLC{xwPNqunj{8?2VMnkSn}R{Bu5Yd-R!+ghU<RcjfdtiQr((@O
zkvbF|3RRJ9kGy_dT4Hit@R2*Pg{cB-i1W*^u*Q?cb-DtoA4OWKO0&>l6nJDZy0`ES
z>BlVgs#P>n!7f?mU9bo`fm8IAy68uipuUma4o#cHt+TgP@mhCmS}3N35cE;rX?moa
zAm4Fn(x?a|V$Y%vSQ1AweNQL}0>FFAL<eGOAAmC1?~H7}Q4On+J6X5Q>rmYy$S0rf
z;ltS7->hDi-b^N|v<YRzoFf526n}li<?JtOrY@s@PHevz!?@n_lUtws^!hXZxN}%x
z`K;8QZ?waBr`(cHp}%slK6!gvrz-nLMfM|@XWgTd2>GzpRmQQ#>|53K2+oV6wAXvQ
z+n4nd-oc_We+SaACg#?k<`R1JuPbtcP=eBfWi~6%M&81oU~h*`h}-nqodi?jPU37E
ziSDsRFpXW(+K#x(`m0`NLLo_;iWbh%1=P_YY^yBt3I_V!5^4EW)H4F#cw+sINYC;*
z&212u(jLgW4;cbxce5C@cB<#nK6p&ub2(X_<x74CaNq^*?~#TS21(bjg&{WZPD4%F
z%?pP;XM-#^hC`+?+c&N(+@-gnK&MFiVA46mVfj|FTf70J-0(H6jPi+;2QfcS)pM9x
zFJrFxWdKjuJD=9VA&CdaOY<O`O_CWQz&{u0ujB;fWbsM6cvFYi+8X6g!Vz5o-}{`X
zKS}X|mY$Fm4Me@uZxmiol-hao-#qtK>baHIoJuj-y@Bi=jSv$Nk9HCA4|G@eQ|s@a
zdO!H{S`~JzR)EUlCFxdfpILJ8lRAFB+asmzciTm;<j}Tznx_+jfV<_GUP~r}qp|z)
z)j``nszJkG#P9hp)cC<+o0I5{$%7#J<5zvE84s(?-e-xDj6Yxr!3OtDVgeO2)l`(M
zIrX2@bOk@EzUeebx_RBuWiEx^z}vKyfL$uKFOGg?_grkaGVeV<O0zcYJq(GnSLaGY
zX)4;chwF#6-8Bj6+SWP$YQ{d%eKV@=8LW7@iNv+HhtKC?zi_c#P9Jw9akI5p({9w#
zGXZM!b|8D8HaoR_d{vQ?sfQJ8V9o-h`I;(WEX-a%d3)$cnTRZ1$T0mKU;dSLcpx(j
z*rV2C*dY1AF|36)*U$6Ojk2$>j;CW$e5_-8;~qc0qJM`;qoyO<bu^S;yPWuS46qNu
zHt9(9%$!X-x5C2`e;2$^Dx!6=Ji?Rp^B<PTRtVNM4W`|6Vqhhvb(mZKeOuC>huHGY
zpixBa`>-zEEsE}F@lP=D`p0ML^F=PrGQ=**CP^LhL`t5;=cW@VQ1ZeIO&>YueV|RZ
zlwK(gp7NRA)Mi+UJ2GY26(hVW^~5UbxT=vLn|wa+{twVu;XlID5cdt(EzcNuO6NxT
zhMbT7ub<%4T$reIQ8K`@<D2+n{TmzIuKD08K}E%=BH<4sVKnGdq}_$ovbnyg=dJ?k
z!cVR;9W|-WxCuqC&og2TaK0w@oMaIRSleTkS$uy}(%cmc=qaCjP}IA^sdoOBNxR-`
zBevZvV>!c+=b5Z&I#W8rPUU2p>#N{9IoEs&Pa}zx_BJyKivx4(Zf}%Q8P!CA`8qA4
zuP*SzX_r8;FQ~tJD(YeNjgPC8(o#>~5(&Ip?CK1H;RhK?h7h)Fh%IhlB-6Y0tZmyX
zGwUonov@(qTi<`}v=Y(dFBxySh8@eO1s2iR7<<qa)945FeDNDZ0!lb1B4Bf4@mi5b
zomZwN1^T$j)`PmKIrcD3kZ-{je!BeI^$(A%jwA<OAi-Kws4CYA*GHvS!cKmhH}D>x
zH)T$JL6^n>x7K>YKcs%cSNC&j7tWeOfgR+hk1*258uqqGPh}Gr@;l2D|1uc&DCUzU
zjd^a7Y86QC$H(|Xd!_sL4Z!yf0xLR<x6)L}$b=KzkvXHLB#^kr&85CsV0nwDFXa>Q
ziI2&x7Nmjb&Z6sSIQt}#xcacN3G~zC|AGFT?-nbhj4$ys%G+aq)+t8i{|PMX<nGKa
z8Vi2)<E-u{&sfu~F6d{STAw-D#7~Z1_gnhDPN_mCou7nifA;G9ut&;GF)bE!TKkeE
z77;-ulvl4{_=<N~Ibx$M+o8R{Kfry`zrXJVof5l2E@)GUdLWK8N~l>c?tlsvTCqZ9
zlEl0v1YGSnpYxD9KD`eZ9-`WLpmUp5f8^urg^V}1(Bnx7mzGVf{+5p0;+zX9X3{lN
zw%Ql<Ph`CXJq()u2l6I3zweLNuTbR3oJ=e3Z#inrnKR1y6WY+^ujDqFnxD_F-XHgq
zj&nQ4O1?Bu40wb%zAN0Cs{=O9KCRS3TO@>0eHwhB)VvOCK+m^UY&&UpDR(5W1bXoG
z2z`ZB**OSCwC>l=e^huQ(noI#Ci@CR_rSz^E&ba?4Qn%E8C$5Nv7=KQQANWsxTyA!
zy7(-nIJH9Rxe*S0q4AHa!=3w!j*<trK5Xv!;YQBbw%c#2m2?rSpzW(7)=nUxI1@}X
zesEG!i9zcwKexi^AP0xoHjr9%(#lTMx&u)7I%D6PylKQrw5=-FC_X1At&Z`qeAW01
z{gAjKX<eOGVLRQ=Bkl^t%8UswK7j*X*mm1?3d5=a%36i?W{Qjn_szmqGHHWWZGBmX
zs>-6_ffjRJ2n0x>cl4dyBwC)>JzZ%zmPz19Q(Pi---()MQ+h+i6gDM&S;y&RMacf}
z8~Bsd!7(?wcT7(nw<`X!K3lq+V+u+0haUQ*h68iyWs0!L5I1UpcwPfVEQHwQ#n=-t
z0sB4ydW-^_)bo#HBanHcw)FkGtBfTNASt+C+2>q7FUEmBtdxu7R`f$_UtTe?JH938
zi4=jw;<Zm+TqbBe%3A!{<0x#gUr#(U2;}OT?w7i`^^Mloh2dVjd!?G^;Nu?+9~Qij
z<Iko6t*!g#?k%C5YMrYov~C0iDE3TA&#C=1fQ-!?31(K+G$tljXFMmi{NY>c1-<M<
zX)I-li;XnQem%_NN{=LKJuNbQB$vTELO=|oD{TQ0cI%bgU8zIg-k^(OMbnIIWCDzU
zo9bg)ymZVRo0Xl(6=9|b?h|FFXK{bEqdOM$)$9P<NKMSNSR*ClCga3RwtS_6TMaFb
z*B%Av()4J|YwO6q4dzMn&0wDMn1}2GBeOtA7B3|-*?c2n$ZJ?te!9akA;Cu9qVTMO
zuoR#6VF=Mjc@0WJpVusJ0N<e00(`7t!&r;C*Cw>#w?l4Uo%X1e@$K#%attUsU&({r
z^e>Mfv-_$GT(s~HiA)O{d&hD7pwDLS?V)O&C}8z$JJOT?^p@hMEgDD;Qn!(t$mbj!
z4QI&u_1Bsw_s&%$J^XVYMXAvzpTKD9b^c<N%Uf_?pEJZ57tOgQ3EF()t9Aw~DgPGF
zQvHgiS${~T06z))p)P8`YF2Rvc-M!t{%!L<7t-9hU#A%0@!nOmGIKNAJ02gdJ$Ma^
z4lOo}K1KW?d3?mnJ=c_aobd06hxLug|3G2S#{?62!1{UzIs~gA*1@ZO7I5os2Y0<M
zinf^w;+}8IH=A9Y<<rBR;=Cs~zgd`Er^ckF*(TB}-Eqv+dcE%{!*?yX_xAOp(rBeS
zUaJ(b9_gu_nbf4;WcMk5Vga)juuC_#;4u}k&iK;bGKyCP>J9RyDzwq|sPl%r=4zRi
zd;46b-7P;g?m!0y^1k-$<DKd?Qs2Wv3!w$jz^S}ot%SNEOL!?gEG3HDO*gImy^FV>
zaJHp_=)8E=f025e!sZ?D@Q*WFGm3_CBi+`gLcjN%JB|NlRH$R$Ew$#&V*bEhC0INB
zq&I>G85*d@$z*__B2i&A%~1dbjy2NjdBnjdBv?8p@YQ!rHpX!6(l^PJb(tC@Z8iu}
zEu|Sfz<4NZzLBC2l?~=<jW{^KaIK;RNT5Kb04kQq#RKQ($ur(@dn;8YUuxQ_+e;ol
z&OkTvevol7b1ZOayUG~xvwrFOwwl+=nmijh%7fxXU%7-p529Kh4VsnFK$18r{72th
z-NR-bBPTM%Tcu75V7>ru4K0ozZz+`|<;x@`g~O1v^+af9h=l_9&l*@L&v%3XE70zE
zKVW{uJWVPxI;qXF$o%YHl(xPl+&6<aJvfrR*9x{%T91A><%f`6IeX)Kg~z4%Y`Y~W
zGS~ZwmK@iQhWig?TDQ|oM80V6k)93g1%JKqJd2O1dCH&GZ_-h|f${vTF$xf9Z7raE
zW?k4U$50J$Cz%{_;$qP#<-t@P#Or2KK|ss<!GQ@Zuqko2>03qFZb-=*7dTm*<-Vol
zws^!DqD+6NZm+p<E5*6<Zd+ik-sS9EW2>^4XH(9@;FA3x2#vtAuS0&0zoN!1k0&UL
zbwUZ6X6Uo;Ozw1P0NqJLOy=30_m|SvSD@yVw{eHQpjEO^u#CRP2pRscoE8eJR50uw
z;6CS~rhd-LI|Op1uGKZ_FfQ!sKyaY!Y~bFI{K9%ypp28rsih4Md0E-t;Qw*j&(*u@
z&l0t}cqJY5pJQpj8Bik;Ka!*SX`MdYd1!sV^$DtK1i#Q`D8&Z{_TdgW-C~$M*B2q(
zbIXxZ|GY>Z#w;F#esr$V4RNGL*Pu?7l;^CeS0{FbbsU-$3-PBd3a`>uq(_<tLbF#t
zCpCqX-rG*vkD7PzL{B1e)_1f+GQ#VBd_uNzWN<ZC2>#*w5)(hx^+{o{<_eug$b4~!
zo>SJ&2}kpl80QL@CXd@@hhm_^=~68s$6NT7hyV^tG?n1pT&c-imFdE2ayI(-ZvHIj
zfobl;G1qU(iPOq6TuPB5Kp+I=wc8J*;m7~XrO(7#WLpdIe5?i-2yJ{^$(}LcQK=5$
ze;|fRwBt(x2BODns{}nrTVN{Y;_`UG845`q8E3N%`!VE&5l=NlRffj}gX)vv;<8U>
zCsp+$V@Pg1BNAPoO*-Rhkp;Xv+O;_8hW9X{<2jW02zTYFNpMCL1ltZWhW)l1(a-Xb
z&&Ez_l;#=pdn9wDk0o^HkLp%k=|leyWOpeOx&3%#$*P!EZK=)6$2m-vz_5Q)t?UW!
z$xK}Cv`IowwJx2uz3O^cl^Uz!vR%?-RzH@DQO~63t7+y^4yb$4*7;_y`UfGL?foz>
z(nYK`$eRnD%V!MvY}EMB>T(sI%*7sVXu)+K`-v;R)T0Z;Xjy<QHp74)xmtU;!8Wr@
z8OeoFi}gYB!0QGYsoHhjS1{}+xv%Naw?dUGZ{U%Vn*_y)OcNh1ZQ$3gC*YKKw37Ii
zRng}9T1R%iPr?V#siiu2b%7I8^gobVLml+B^h|#AH6g>#!$wX463j6+$ufS|Su-uT
z#_zg7?*<;?o~pG=u~Bd#N7hXiRgv?G0;$t<CzW2u;ba8E2$}a|teQ0k-2DWA-0fEq
zaO1_{rFw!-!+R3~oS4ctUpSe1O_u%Eu74<YbytVnXEBZ}k^*4_W+0)Zcx_Rw1tCeW
zg3ke(N+?@MlB&&yB3@CXvjWh>&(fS(Y)7gCIO!zsQzGR-rLrKVIkvcYBn{4cW<C~J
zN-MuU1_3X{oTM~IT0f3wf3TuRz;qvK)W9k|ub{4#Ok^4`Rwo=wRf*#xo49vH2m*Gt
zv#WJaWEcr^k~N$DQi2S%{6v>lWpoNJGW}<WV`~!pNlVu1S%aLeOD)~&`$pJX?n)(L
zimU4wkWJuTas0c8spYlTQo^tp!k;;Kn!#AS^?*Qg&HJEeEA3I;`&JaWx%!dHEh?xb
z&`qBH+b-EX9a)=hr9IQF@tuGH3XdXNr;}!ETLK_OwC7)a1+BXvmj6svx}N4qKUo~M
zI0MQtx1`oCn*n{}M@9ljO23Mb-#F&+5PD?$i(;PGX&B&jQw>CM0%3)fw4@LN^R7;e
zWRDOMJXQB`k2n!#nX^d<tF>{bdq<)*<&dNo;WG$u<rqQIhtvcHU!CtzWiQ=C9hv$p
z&E0?~R1{z3nspkJABk+lE;y~ui4*98h80PlI^ur%6P>M=r^+6}AeA7-<?MtZ&J<R{
zMiHQr?gLap<r!CuO=+s)&@>fDkQZTwhiS%4zIz~cV8LE=bMUS>&p+99%OvmOPm;`M
zyc+Lxs#uZq@?g<OqM5w@fD_(c$aPtJMW@cGx%H*?p5%2pRZe9WyDAO7CkuxN>H@<c
zbkj2n;#rD#>QTW&#@1h8Vx4nTN!mD{_+sV4uGP^6R61(c7e>gwwEbzmX~7Xn({kf_
zv{$X|MT20!wWlL%sN^5YaE3($-)dtwzbRcwQ|O<Y5%|=4rNh)<S?*{UOkj^ta+6Vi
zDEpZMyN7?=@41|*I;%@4^tkk$xbht!k<x$nM9ejx^FmewVP9`>W(XXuac>p6D(S2K
zVRME_v8`UGgAS|PhX1pFtLc{{UPAj(UJZL**PM8&yc?$r*=Nf+HqtQxQ*WJ`>TXfR
zu5J+4o3*mRN^KWpmWbLMhG@0F^K8YR4w88LkZ0%9yk%6&=l?Meip{xFpNu-~gIpK`
z#MJPa{dkbSyj7-;9h1UU&LM1@Y0I#~Y1(3L;Al+QmAAB&y;RcUwWDnyU~Kl>^7$UA
z|6zcOtM@p}<U`?BmgP-G>{2)R_&9Wkb9uu`Pf~#ms7LrDHQWBGW{39y>C4Fdg%lhl
zD^W=^<-tplhk1n|GxqvSURs~QMKJdknC?@?-D4M*yNEkH8g?TiiEdnDc*EO79?BgW
z41{RC12PlyTx9|{FldABZ0TaPqQ&Z}CQEjD=#V~Mk5Et?{a%)C=DXEOW<1McffRoy
z*)_z;im;9!I`(tYM%QXr#nUL7`8=zHfN&Thn*J^+-FpI|J<j@~tll}OJPqf4``d+W
zP!DaJr<lzjn>tOYGDMj79IirBDk+-IPfrg5)ygpp>#K3jCde6li<g4#Wvx`L54%Z(
z7-BQ9b6w7?@3kI}yy~KK`pPXwpcN;)?@f1zAUb5Qf~U<?)W=*QkNzmuK##R@9`z_M
z+`^P)vkfWi8IMrmqHJ<z;tg^rCe=Z~x@kG(qM1^IdeBjdU#M&D$c1#0*PECt6#uAj
zT;}9<J>WT?zKzQK5y86^cwmz(K6m$0mV?sqa3*!Er<Q1Sy{|{fdwkNvM;xRoD8<iP
zr>%(Wo))lW@*qBf6kWN$K11{C7&1&N)oR?^D6Sw7h}@7bebmsGsx3!)I#&z&^{LCG
zIeiG@(efy29@;X=oGPa5mz*9sW&pP%zFd6drnpwgX_FCGxJ#;j81f_hCC*sTU!U*+
zDc7Xe4-z9*v2y?WzO@MFkypBZD=OO%`qg2m>!t*L7KT#rfWegHHHSn+M}nRfT@QZZ
z%Y^-+Xwto^V$Hz=Cj?2&vO+^*%@E85S`U<7co+3uHSb_zF1UBH*ij48TEzfCJ!jFL
zJUbp_`UpM;1l<dNF(i;hf5yWd9E+kc1t`U32)-4tW{P;BKRnR4k>%Y}D$7rEeRfaV
z;dVK}3VV`BLj%y_awu-f-N|a*H7h^31dU7Sovn5HhZmjhV=y<v>NF=ox+_JaF0@o%
zqzg~c6D~cgQF8$Xj=V2F=aFjR3A9V_wBequ@uYXq6c{1I+=Y|2g;)-YW=5oO;^>s}
zCjO2>ZaDS5&MEr7U~fxL6t`#O({9`wf;Se^i2=a_HAT4P$f}LGXWvX*G(K1}YJphl
z{s&N17Z3t8=?={9aNrbtV@%vsyisR0ofl*RhF43MIFB4=ZWs8m6NAH;Yx;#}cXB=?
zE9xq6#tW(!o;S*=iz~ud6e+L0#fZMNClmD7BGXOqQAVmF*rhh}Ax??*FjFDre_T08
zYGw7Gjpd_0xUy*|`QF)(Bcx`S<tdX*fzA4hduB0W2=m`uA}D0Mx(~%E=<h2_G5x?*
zuOhMSN3Nl}D=}9IF*8vZmFQrssgHh`5zTA=+WDr(TXxieRCL9QskUY1-bPjF3LbPz
zFGpr1H}H2E6r3`xoN(>geR_ph0P6$j$$+^InTJu5t3OLDk9J*+XzIIh_qEv7+S2QW
z59BBp?R0ySDn~OVF9rSw>NDf!QQ&>I`nb8L#`Dd~6=mHDttyg!Pxf~AYn}kAqifnQ
zL79T@7KIzM#kyM_gdf#z*Mn6`H`AT=+fyDs5nGVw{ZJe=r(~f1T2%a{H?{-3QJJ{3
zw7TtXO-w7I(a&$NdJ+hF<l_!irP2gdjX0SqUJSPdw9;AD&zDr*X|=e%rZ_!0d25;M
z<i2bZA#va`x7sdQ%QjXk)V?t|;*}Cfs^224$4vU=C5z#NKJNEx)5y!hwcRRhKy$1|
z|CtOg!Zx(JKgH>;vQ7FU=r@ftBZ05BG}u<tKRkaVVDBY@a9($JDD6_3kfM-&=62P}
zac`pFrRxl?yZTs2vCy{U)h->uD=+?j?*i56Mm{WEDdvxGfF<>hJo(sg{+6_z28e_F
zoyK=7F`U<Uvk9SJEeNNRNae`m#FFn4t*AZTo0UF#MX!JtZMoGgj88lnCYG1alybs$
z>%n07Sy@IGtVqv5R5Q_E(p7j0a4{M%a7ZI;S0prP{YjSh3#8cied+ho8x4P8;=KYu
zsG88T9zR2O_X+6vf!@|=jLXi5`;&j!@k}27#A-p!3bLQ%;WQFwRm+An6TxQ;#Wxw(
zXhUTE!Yw-fgI#IA+8FXy+Z&tJ>|VUnX?QJ|YK-x;ee>X!7MJp_>Ooq59n>_|++6Fy
zQb8bL@b6+MTeVtzYHZBOMNj?G-D}-bw2KYz(M?eO2NECm-mMka=>UDsyxB&t-;Loc
zPY$?Kt%4Cl=A~}&;_A?f`v`hNY3{F>zT~87Coi3muiE3>6jJNtN!D74wwLFylU|l(
z;vuRbA59GM>j$_T`guax_p3zBr4=WcJlI3EI!Z$S+NSxzx~Gh6UwRc`w1R`tN{=kA
zmEFr_LS#heEWCT^`NWJ}_s9}aoOfhuw&s3{`!v-i8)u!|B)eQEIHp)~*DsOwBWO6J
z?LH=weJCDBc?T7k-`+0#lSX5F%MCti_($OM1Aa2tG~Qd}QKOjtv%o%q1;Uz*{w`gI
zV}Y9fLM~;1b!A0-6dvE(eXcZYCd#%Rx*xJu%WhoN>|8qd=;(Gw%PKC$TA|$c!#C+=
zOS@L2h|HetcJ>WAJekVs?XHrv{1;|R3S#rM%DumU2I+IsMj=&~k}Mr=Oj3FRRol#1
z$LY!<`Db2RDGkL1e?W4(_Ho#ku$G-CrVxax@d(%j3VQ9f-n#@Fb`=*4+7os63XzE4
z?YQzn`F=&|rGGdKs$izJZ{M8;LNftzsMF#xl((*BYu+c#ol)0#_ofq=YW`at2b>V=
z_xL+zSd)eOxk%wTS;4&YYJd3@V14m1%{PWYk8TNnGFLR!-V7Z~f5q(Y<k4w}#i`lI
zE0<28>S2}WD@z$OkS9`F?Rm6na~KdOAISVEqnai7tn}{;<l%u4#uvP5X-*SKt1lJ8
z&yE@?Yhn!@MY`C-RB%G<q<7E$mgn@{d*wb8W7RQH^r&PiC2f;=9$@s%JGq=AaS5LY
zgav~sRZcK}udtujh6NHdKr+I(Z+3{&xMxI6+IA;@paCq)_rzvu$GDJ&@1F_21*$`d
zs9|a4fF*Z@2Or=}OM>G&hEy5MVU|wNi3D@jo_&jPv-E4<um_|)g=^Wdn}LiAfu$K&
zsWeC#T<;#hs=I5IH<Um0?(ov)B_(gh82@xrPeYY@8eu$EBi)KyXhpTx33E^;{G;!4
z<*p(BkW556x(+kyaJTm;K`CZFzK8!6X~F4yfa~hikp4<spqd@VbBM#y2-rUeR{18q
zFY<82Tusj7eYt_L$l$Y{pLHs^Fn8Af5MR?eutWtz>gGc?eiWBW*;(Vff=}R_LS&R%
zVJRm74S4oKGty>xz(YuiUQ>-|VNR>gq0MMK1>k54sH?6zc@SIxn>s@MWw8_!?=yj7
ztbKd_50f6)s<27EOgZs`?|kap=Tv*QX^JM^_{9#!t+d8<#zhvuGMME8<AEll7tS}K
zKdah_8aWbI5B@dJw@-5*_K(xSUjP0R`n{evvE}}|vP?lOLTA9@TJ&O&(Ktxh+NdDx
z5_01;WXIh}z3KoZtYcG#cWm->y!IGKVHaL@N~`7Gw8j7n(?#VO!K)e0B45gjn`;DV
zc!VY2Se{iGgYE%iJFGO4|3%UF82}ydbeQ|IoZ)PyT68Ffd)%9mbWQB-s!ZUU*}nPD
zk<l>e$zX(_bVd@#j2*`7y98bYabVFWl*62BpC7qyN5B~{xp!L`AJP*Llt}QA%gJiH
zs`ngOTT&a%N*pr$?$x2oW*LW(+{sH0p>Q6~hz28Ulz4wSjiBMROp;XjOSVo_X{G-X
zTg8BiG+kL^xi{kQiUjXQ+edWRK5EjDh;Y$f{(4}et$x|2okG0)lovBn-7@%Vw^APN
z3D(0FcW~e)CTQtUKT8>8SKFhE0nJYj<De@DI|-3O-wZd_m=hTQ152~d&EB&R1EIJp
ztp9A?WRk+XCrvpHOwrh$c&2=w(<=TibVk!NR&Gs*4tfe0{-Cr<Q15R>*QP<jg#OPg
zwG1|UCBEsP3NY?MQv44wH*f2~!BSaJ7y>#%o|zXzC#T+eI3#LnbCNTm;tid|{Fmi%
zO=r7adhO6ms<(~n+2o{Ht=!i;zAH&3qDhmou0mCL%(|QtN7G}V3p3WoSqg8Az#r?w
zp(AY#8MJF}e635&&tqgV$cKIB60bvHoH}fbsp1Me>OpFx<g{(5MsfFfsfWVjGSHV-
z!o%cd)7ah8t$$p%LG1$wF{*c@*HjU9>0&r%6o?R)HVj<ii)(}D-0WZ>BaJaL;6AP}
zQQ)|B=hTeh_;54#ZwP*+Ga$xMQQyZfymqmh9Fs|9GMtxG^m?D%=KdNz_75vw#P2af
zGh(i8bG&T++NYcIx*}mwJb-U9Tf>47S|2n{BGXiAwI6@<P4Z{6Y^im{pLCV&w46H*
z2DJr33b`vHVJWAC=dxSgbM2wErqOQ744NUo-`V6oQ;RHU+MgxkfYug&hQ;_f?~wz)
zs8#DwFnz%B&uOPB>+x#}A(@Itx}X&B=sLk&2|gJ!^T%>H_7k;QiJqhC@72FcL@_sg
zThG;1;7<-e@lvw;J9$yBC+@p{lAWk(1a`Bfe0EUepa@`p*@_vhWWae3Bj0)76Y-e`
z0xoIsS8^lJji%b){xF@yRwd<i_c)tGV$p-}_rvnwM*6_F`+g}c(NK2t^M^KUlMo->
z*ur|_OjuCVoQF+I>c?N0U>~r&nMyWtP{WvL+$(3C$~JOg`)Hh{gQ=d_L$%H2r1$n;
zy8)HL!&^j7I7CMyNCvE@$Z>FRn{9bF3wb6vzOPNpz?=KNB!0veU+%TC6<jkS#v0Ij
zrg-!p2e_UMnMCvf0O4pwN~KaOKXtUNh)K7@>9B!5FD;ypER&gL7=kfee%QCmnHO(C
z+N`#Awku84@D{LXSzg8g$B@GwM%0M&8PLkG@H5ed#0-hB&d%XjcVhlnf{l;%w80g~
zJVBk;SEj`qD*IpEOAPXV&AVEL@;OeM#hJ-)!;lD8^-e!UQ3)tAQ4)SGuz*i^>4oDM
z3HZi*<2@i}{h$)5HD*A?`(v$rx-HV0111k>32yy}P(I~nIvY6vgwN!b!;A`oiYXt#
z=rpupn|%R4H*7maqsON(vgyxOtlHqhh@oOiy`cM6lh3F5Av?B0tSy^QN|E;bjo0Y<
z63pA5U5C>65VM@m6p~t}2F@a(KLrO;hYb6NAigyVbg3Qzx_7;3Brd->B(11UQF3*B
z{sC{P9yVvU%C}VjceRp=Igfv2q(XIz*vm_%iLepqa!R?UL83)#GbX9C_=Q*eZkYn@
zu2p8IK2Z>7Yjgj~*Qxh-i)C5ZY1nj0a;6n$<X3v4J>(pJ4}u?U(}fQ2C4-0urbI9p
zJv~Qqoz1HCJ!pOj&RS~m?37k8lnF<Vp_@2`KKeG~0a8n17D!=PQ`7GbuvgI@kN#q_
zeqOa6Lur(Ue>lMnG)k-0uDln|cqUKbt-84;0KlTCvj5<1uJOtr8K(+{J&WkEbvRLb
zBJ-jR6{?ieA;*6U%Pl`A39At&z3Utc^13_KiB23Tu)>APe$p|%BtJVI61|Xau#hha
zD0;I+CA3q4kLuBJ9izq0S8_|_r+)~(^n&dTNp^<)J?XjdVRk4VyLr6zcOk`}_X`jm
zxyWj0(`nP(9OJ}qa~ewNf>&aeUBELo42+et84o?*mdZMm$0Tq9F*QQFLB*8`u-N=0
z*sFz$czjNL%IK31n>g-UHb;T*$D!CFv5Au3TmOw^NyFQ&>e&9+I@@W^f2IFB`}<O1
z?a3)XRj%#PGxf-zYjp0l$`SY`^<O;qmi{^zuXDyV_7f>+VI4C^g3P&6Oo45NSp`F4
zmNu6*nnqO);t}J=FryJrT}EvQd%ob#>=(WtmyUHgYWe){ze;~9RMxY%9}X5sWvcwm
zV2$%eCSy*HKw}uyQf+MAu>~~~9@Na4$e(8Zd47(Z{XY;nnzOG>Iwp|3?%iQ$)#H3K
z6~z54puo6KQ+n&(mRt;x<YQpm6C9H1^Gsa3`!59&&vee&b<@q^eO_mu+Sx|R6;I`K
zVrDg$SHL`eG3;`N1}#T1#qfS@OE=M)lI;Gi|5sBqZY2djz{KUG`co$Bazhe~@6j>y
zD5ZCyp(()8`WL^<8<*({T%;sNI_#I?pSZC2NL<Cq;0|Ee+zW(R`nLNYTM=$J6XC9i
zVsv;-W$**^;edc$vzS4najvt!n;sV(^=^|)ac}md^3b2F+UoCgmj?r;sQq5KGD-jV
zv*;eXEAm+_isUYmv5lsKpVfEn+l|zv=PyX^mLb>r#U%m}g*39-f|}Bl2{|hs37Os(
z7;`_U0kc(0kk}?YkkwV(6A8CZi!E5b75w<#32IlV{@(TQhWv6R_rS=-c-30o!@{}T
zQhLt^&(k^zg0>(QetKd_xwk2`KkA7D_p|ao=X42DruE5rS&Vsf)S?t;91s`CjIeCO
zs{D01`Y6bm6k?a4pSE8-azR0v&M6mZUG$?VSSI&flOj|bur52@!uV-2`TLw0-CI~7
z-pmO@)9q|3&HRuzT)CCkH>G=4s}S}`wJ9uAv^SkeRpxD#$$HT;V08!KX}_t2x}2^&
zL*3~qOJxZ7z=Q)n9?H75X=;muQ|=5Oge;r_C*NU3C9tq>jaFH5ej(7~yL8gkKsXw|
z&mly%Wxv)l5POqv51IUn@yS!Qe9?-snA5m}l@}F{U)jmdvh{AXDnT5Gb}7keckhXj
z)9gmz#nP#n$osqZ(w>^k#{NoN`5&l=``Tb<7h7>v74RVe1f402j#ndnJ&%kN61~N<
z`KT#U)5v!#A;$FSbRE=e`;zWrD7mG)G43`fmB6BUBJX&JJ*Si=i!-L%_n0yLrJ5BZ
zP9e%(cX{`Lwf#U+*xD*n5JXDJ{7dvdT?$+P!!C)AcG;}msP{}BS+~SrAL<9sJn{2<
zyE|8z9h2tEbFG_~Doqj!3)SkcXR)9n`NW3839eox&;jq|IoDf(^6P)SfNMcuWC&(B
zppV|^dDc5FG48BtYi)*KR%q%2fRmJ0aY_dHDiSO~SYu!nDK0%=&NZ|?aPmU6-b!~I
zO500JO|+ohXoy@$h8*n{&O;<xi~@joH}buqx+T@WiM%f~k_k3$IG%9}t6SUEksxKt
zk(pP5k?PF^#A?#CP_%;!F!}Jc!%&3uuGW+fuHv9n&B;8VpOu-$7v+N5z>LExcg1le
z?CoZwe!#MIvpC{tWM5j5r{lUlCSyrg4eO{D-)+a0>gnN27n1*4Jty6Sk`dDN?!2zL
zc}wrQZt;@@v5V#{JMz_4W%uG7csP|rrO369Cx5V334pohCz3vB;8xeo8yMe)(vvr@
z7-^qrnW`(M1^co07G|)hVZee6{TmsE%9~3uW%ayM;o?mL!l15q2i9U6_noEiBHGX|
z6{`zJe8=eL68FH#=S>m(m1X1_t*^*)bI^~Sw(jA)H|}YEnmxgYky85VRF7n0f|psq
z3w}t?bcc-tNkO~~(X##*Yio-o#{GI{Bb3OltKbf2Vx&cbe$<H+WW^3<r91gk6&Es6
zi+RMNzl8s+%u!J&F#&~|mAr|ZIm%g&4h0LtsN>=+gdY`$X!EH(95<5O<<-=zpJ^|6
zp}eLnv2*9mIkTVbxccMw(3e{~)wiM}?P`WvC&C80>!kuUC8?A39cqr;mpw76rwp82
zZo{*Cx9|qbh+EnYl4Nxn-YRZJQ%i2{+>HMVD5OAqBi}jw<&uoP7z>wGB@?$vv8gBb
z(7Ymyhe(M}6)VO4@{;%1C*pP{6}`op^Jm$@{$fHK#<@jb+6tX(41`u|)_v-~65Jb$
zs>y#XlXL8_zKlFtKO^H>I7=isFLv5>ndWob*`(9w3TYtjEtX@FZm5rDptWjEW8u``
zjEgR&XZ0*Jz7TPA^<LpdX^&YGU_EIxMK2~o3ODLTf+&TtPt{ntRpLap3W>jLGGC?;
z(kx8e<TmAFE%ZifST8aKT*H2r(%=8Up72&<sl3)<WP}c?aaI-Y*$eQc?at0VMX^$$
zf)EFj^w2|Ay)=i1yj(2#JUr4y7wuxT#QSE~Tv(*A7@rKc+>AqB@(|AzdK=qYKEtDQ
zm&gaZHu=n~2ExY{ZOxZ{DpW`pV;9O|AY_AcKf;fKUHJ2^ASvhl2V+N8RuzuDNpU?y
zfIX^*L%4w2nv?K%r<N&<LFQEOqn{MoQST9qYNaEF>DIH8_V<=8*Iy)@K^tO~>rwEm
z+gIGVJETUA{=<Il<7^6ZE<W_d4M2ZuQ-^XzYtS!vS0k52J%p8K7n>1wrX{aE@8A)Z
zitIag6~6yuF>W-|(uUD3+bWew&QZy#*%la`Yy%MSETrLt1rgs)W{iDCyFnynL6lK-
zk6JQtv9#urdfeCg1VXH1VwvS-4R1O<CT1w=LC7h&bxnnZO9o=El1vK2l(6Q@+36;(
zS|}X=@ORQPiJ;Q7%eiEcf+&0q45Fv&a@tZttvI&{9UiWe_Q-^~mLDm9JEsNf#F<Z3
zr??yQVRp0W*_v2VYkH<yLLjp{Jk;nO%nj}l%*ks1j+7j6zq-SOcSqID+}szevGqO8
z$6a6YugCpYp7eDXs+}~&11J3{?Qqbahsb5FtzJ{_kY><QqCrYt=h&FKfGt1aRbnlD
zfPOI@SA{2bYpS;m&kJj4e~thg<q~klq4Jsc-FG9$i-K2oNXX>pAyDSNgCp}Ua|+>S
zy^6uFGC0;8(X26ppo;vn5F7laJzrU%>ppQ*sQte{k#q4%e7|z({_FpNq%+s{cOHot
zLehSdZy7e2ywwhDKk6-H)Mt^xH%e&`X8O{*xcIXy6<$4qYSJp6H<&r?dU(hL8_J+z
zs5TPw8=V8Y<)ddp)m$yfJry@{KA(Ju$kcr*JxN4K{2k7iEeWD`y}C_G<NPRc8SDcM
zy&l1NK7DttEIaN2jQ2unwu!}1if~QT4JbYCQK*o!@Ud_J@n@F1H!OwVc9W!sQUvS0
zh74ySy($Us2#@1BRBe*3HcCiTg6I}z&&d#B^K4<j<rFnLDZ<2O3S7A@c1?V7iRrs&
z7Rr$&%d?w0=)mW{V8k~mcM=GJhCk)Hw5)NOz6aOdT6JE#a4Jd&8hZbg?;o_1&84_4
z<_uULjh+@2Mjk(v)*7k)+S|1-!wvSP-@0*tiICbu%Sxsi8F2O*l>$pHvxYQ~7T$FB
ztbM85R8rK|*9P&D-wG_OOx&nLs|@fbed>(Wa8(WKZqD8QGNmcMs{>J_orcN1ijvp&
z>(+@U1k%W(Q8;$UMOh#zM`I5o?h1x?Mlps?0bVdlhYz~%LkX12RoRdS28p2;#!5et
z8p8b!i6M}*m6gHHmrXhFD}|Y;BRz1N{1n}$_7^oUnvRL)&%C3d8iuzXdZpOnq)`Uw
z<>1*aKTbT6hEIz5HtFMG!iX=@tE}!+sI5$Y2LAGS?C8N<r9pp;Ba<C37vpuqh%)+s
zmMyF;Jl$9Glw#sbsK*Affaz#kUdWE2q<-7$!@X5=1|0?FO~X-*O|EZsPxw)9HNvW~
zV%-0vJqVypDh*7cdXOPPn++X=A6$<Gf}}4twGyQV+Ekt2)B6hw<@-uKzATHr<U4qU
zdYbJpI&fpmsOHc|*W1^Z`?C{b5HVMqe=crA)uGC0n0y8A!|SensD{_$&dj5c0|D(6
zV{IBa5k?bN0C%laP+#Nu|7be*cqae<|8M4;2*Vt6W+>-!tT}IToH?tU*&H&*(7}k9
z8FPvy$2lKzK2(az`OHX#<d}pUD;+++>;1jme*f&R{kgra*Y$ip9{2l$%a(8NWlyQ;
zJ+~vB*`l9Q2?>aL-7GL2pY^YP_<x|H@gyzkH*M+Z$dyLQ3&T>ct?A-sSCi5FtFP7V
zzB^H?17*56%Ji8q-ZEs9uxnu$!H;PJn!@G+)C{GMNVZ;0FHL_7TXfY*n{k%HOwVzJ
zEOT+j;fL}Rn9gvHkH<hgz9Dky$J$C;g+;0=*9+B2x`6EtW8qyzJAOMuRMq>ikEBV7
zHFsjpqaOouME4u$`dfq_VMjIP&j=G5Z_QXn@RYybd|RE0lnUb?$fA8jV`u-Kjl#9R
zq5SUs^92zrB+F#t|5&;pW~gX7Pud0ACQ~90sCQ``Tt<k5ka6{K#Y?(Vg5L8eHL-sc
zIp5Ag*Sh^vOG#;OXW?yv$j^u!r}6UC2OL9g&-c7Cajg%Z-W=|f-QA5aCL}L~#A3Wj
z`)FrDiN|yK=hb#Zzs415jpK{VD?Z=z*4_j<Qmq7~po2%`MbN;0$4f=ot$)6VF!IY;
zdb?_Dj@L+2Xef&n+(6Cy2(u9<BMy5bW*RnT)yZD|oCxOWH?pdqzi;jgVrN|)9ZrwQ
zB&C%3bO#>Bn_#h(cUV|NYu202gxLyTaIg&x(FHE8LXqPk>W&%o<`fo<v>9S9#6GO2
z>ZyQAb$?`=218#O5mA1T3)=5Q$Z@jIUpGw#1E*8*PfZs`cGZPb2NOv-u&4C&Ygm65
za5{i5iI9&~?r_+ho4bEVt;AmzqfbLv?7CR%-mZa9-(42l^8Ost?nK>=Z`R(}9Sgdm
zGL`XuqqHkAC&ZK)7n7+oZIj0mr`2;*yBo`({-M-#ipO+_6U3PIDNXb7Uvcg~Mw(3N
zQsG1$E*GtBA~W%0lZ03ePEc!&oP$@-=*N@TTim&Qcz)PAc=IZ1?7;=cEqzB(N>*wy
zv$7Htu>wxZE7(iz27gR!L`_q}r3s-u9p;}BfAVD$ji>Jnt7Ckvtk0)O`xBerbcMCo
zC<O_JUs;+p{qCDs8**7_#?@9<yH`uxPMEFr%W99K!?9c&I#<(*#JWX6WQm95F&<Z;
zY$v@7XE{n|yW4kB>s?#Zt;=a%Hb6(sPjV0@cJy;5(>jJ@K2xY>Gt9F*IUtU|_Ogak
zl>DM)Vt+vP?8a#4gvC*PG0At>hU3cZEo)z;+iR=k&MZ#*zl6@xyxo7frt2%peB@pW
zL;8<qDP7~8obIW}67=R^&w}wWmBe^K*G1g@PA56YZh<l_6S&%}&V7u5LikWEH#lnx
zMev>Hl?itkN}lB2#W@L`c^wDbP{7j}C$t``)@)Gtwm~uDNm;izglbjUEzS{!B3OWa
z+afs^B*}5;m2O^faUrO?4a0w$RuI%Tl?jjyX1J+u+rF489(h)ErQ*qh50U|cW|AgX
zjjRvp4F;8-B2(7C4WioL%5C>u{%uk?KbR6<pFj=rH}}zwV9V#aRqi(ytFGu?Fvs&G
z4THSl7rwU9tjU?DnGogM<e<H1cD6Gl%vt)dF#cId6GuujCqlgNh!`dwsYk#_OohhK
z2YVQ}nDcl{b$N%jY|YK`VtAP0tf3hpta{;n7PS5sKeiFvL7<D;`eE17CI}NQ2zk9Y
z6`X@@9Q-Wr00D+(SsKH+$di>-+*_aMC;CV^J-V=>ttNJUmvrh7FZmP{!mP+YHu=44
z8_ijO0cow0k6}G48DO`+3UVY7QbaMU^X0w-Ivx{z;Rk#|3-Dt)sx<@+i>%bBeBqLs
ze&#tIE%Deurdh``TWz;zg3q7uLeNSL@fH5ohjXfPhTG|Y7B*m>rF(|=Q#}asRPJev
zA=gtVNJ6R}i+yyK<IC4YE{#d`*>Kd{ZA?6y)LeU9-z5u^Qre`8ImiVb_uuLm*dMb2
zY}zR*!)FtfPP8w<?K1nwY(vnT{Exw|LuOS=N$&|2I*sS6fKpsd3a9!f>g0FQN0*I-
zq7^Qp1zyK@drW-W$zm({!J4~onchYTzYD=QQs#G4RnC6FaZT>q`aU^LVGimBrBpmu
z&5VyV`82#*-hVV$*ZyaUVwOF(n$OjLHTjl-Dc(KHb}a@<Slm+yTCX6r=3!c>mS6+>
zUTVRq&~!ep((CndsiHjer}K-&wh0}%nq8^t1mO@JqQ~L}19MEF|2d#8jPEI5xK`Mu
z0giA=HPKoX!5A1=BxF$;4eQ%O%%1X<54OupXfNAHmYvfP3%F)DNDp$lcn?pcu{1t4
zCxh;6wakPJtRuZwaw!H#OOR%x!I(*~D#e0a_H6WCPMY}N&A+X6YtAlfzJ3GsHWZZp
zuLc|?aQb3jRN6>Ly-j!BXcudMfaPOT6aLRlwql_yE^p}`a_@@Db<Qt70<_&alzt!s
zcwVYfyT<CbY+zbXxL8tah3r+Ej6O#1^cOEy>)kD7Pd;>Me~o>I2FTM|k&rEP;s+A5
zI{64hzpCDEh2d1SmPlT{Q^O?7+W4S!^W~K`K6>R?TD0ejO=#|_ml0R%;M-trIX<sz
zn=-LNm5mhIAlhoF&6k(rtTq)R8AI-G=%?omC3w;XH6a{eWRF@#jF3M!*fd=?n$Py_
z=ad_8Qh1a_9lw<VR+&K=kd2l;UL1i-YW94ylr}{7<*l7mH}XdNe7{bf8rT0iysR8J
z4X%Hu3BIQ@$%r0|bqJ#WSr+Onb)DQe794G6&B{TeerTd)4$AHvSo6`_wF}q(Ui^G#
zT3D<9s?aO^xg4je5wV<txDuh9&(8|mlpnYL84VQ{)nU)Y@|(L#Jq2FO^g8DY_;hh+
zZT>EV%}+lRdx9m^-Jj!(nYjpN#Se|4)hVW^OF)`3HF_#`u6=TdYO6Dh5CqH)@dtG1
zLM@42tUjkty1hvx;y<(+`s{G74i{X+UoUUjbvq@u@fewm=Gqdy<n<`T1Y^~*2@e|$
z@hrknWf-#gDm-I<u7U^o>Mcs_XcyNeD4nV6o%M_EBRY0A-N2Y*!Lx!m!zhALb5-sW
zn<748)<LTWhG<`zm0f1h7*ZVkB`1p0T$Ns<CzDM#>(QOr@+;L6>iqv!Jw*pL`oegT
z-r1YoK@hm=HrhA|Wv@D-V*6{xyQOZQH_zQ^03=!bqv7s+BD;D|v47orPg5QGz8nJQ
zG-T~zz09dg=a*e*oG_23arjI!sBjwKXq>Q<T5$?W+OzI$$U}Xsit^TQy~^N(b^}pn
zBy7uusa&-BUN=BlB7DWFoI9K&XM>7d<iT9qW}P(fvBtyn7$tDRG=NUsGuX(2bdK?T
z$xD>|j9}YPuGg%v7u4X{pm0CC(K92oC(5GXG&aXnXsBkYAb4?6IG4{sJ;Led-kXas
zclviXHY=vHLWbo-vPRM-_QWEWVe;Zed*+$AYsQ~q=Q0(%Q9&a9_ij-uEe${)yw>bY
z9TnAgAKylU_2J)*jjYcwr~_&{&r(P4Udbw*&x3t<jLwRowwqVE&sP?i?i$H9`ORD8
z*g1<ze}JPhSSxi&^`rE81EO#Al{Ulk-1D5Oxs~xFl*KCWxD&`IoR4QYRQpQn153_p
zVdL7B!)xtYeIV6Or1Wy>TW!usR-b>N1!XFP44&(n>0!RDuYoX4rEj?fGB^lXt|2wA
z#6z?|k$l-{#cmIR?*PkaR|Cp%opZH9CA3DtiTU-whsW#7)1Z%in%Q1@lZ?8N9aj1Z
z+NB)o{aA7JhMkU*o;tzL{x{D~bcgrmkDaTH!;^^4cA@KLn$0Ty4<m)CnUc;Y%fI)h
zt0-B;j=!uSB)a!Lf`Plhw-83@zg^6<R<M6u?VeG_`wozWbR-<Bi=>?4L2Xfwv<i@f
zhcSV9yH|2)lW=)3K>G*uUZz1cs0zh6ZNcO6V#YyHOV!F0Qx{`U?4*F?@4HQcJdh#-
zvyV~(!=f*V{*7H3Gm)R(K>le&|2X@@%l25~o3%=RKsyUZ?GYbiZ7AP}#ippLp>dZQ
zv_)3pQRIYMQ_W?Ensa|%g?rocgb0-Re5+FJt@QtK)o0c#^7GBLTP4-LQ@Wmax4?C>
z;ReZ8SS};yGD9QTFS9HL<`f`8S)IarO~k)D6xZ@B;T?4|cfC}<Hw|th(0~iHHI4k$
zbFAk;^84Yng?TUlnc&kAn&LqfYF9Et)l7lpGtOyPC;3xwPsFwx-#?}q!k){tqycY{
zs%k%IrLs#anO}Rt{EI2aXz$u4H1!iBu$K~4zACC<bWp(kb5e41u@#~%WA@o*6SqpS
zFyhU36nind^Zo;1QQ`Am>zHc`w{f*=Ig=q|qd-wQ;Ra0R19JLUj(cYH8p8>-Re$`O
z9Hjy}1?p;lhS|3Lfs|Z0f;w3l2i;@)<>_@?m3v?>o!mRot;>y1+70dN6s)0jdEWi%
z0$1ayn|^t>n%%1abHF#Mnmz0Ko}|K?i4p-^ou(IZm_MqD@PrlOBr7BDbG*xok|2eE
z0D%lGFpkAn6X~%<!N7Jq?L|2rhq=rw<Oiz1<=N>9x%z;KFHMN|Q)YBSt$)}f2PIP&
zr&)|)=a8Z;o+9l&CdJ`Jrf;~RS}*V8EeRe!&yY8Y^IO7e{EfVCd+u_Y>Ej*4KerKA
z1vjBfuT4KadZ`j}IfA(~jrdaCSx88wW2E4?7T+Fw2=H1Y!akIzHN#cOXS3##v_!Gu
zQWxkkbgN!#7cb)=SeD5CJpF9IAtzQXDd%8t5M<K~nWPgGhB=vWM#_P!qFbQ%vS5@P
zvBKx8wZ)0XH_6s%-E}j_E6toLw?ISi8i_LiS*D%0eWu0xkedDa_F@6&Y=`QT2k+aO
z!F-js6b2KNz^9ldr>D}s+89c&ploSDPzHCV4qC+N%i6e^!6dcuihuGZ8Rz^=s3%wf
zbUe|0F>)Wt`g4i-eyU_@d{mAoj-sX>oL$1WZ%?Oej|H}?`i>Re8GP<1LZx?;#39mz
z-jixiRX;aW0p$3&rllGt)r~Qrm`Rzac+ee#$|g2l6X+``44S2Z33dv5+aKG!afC+=
zS{2!%)wxO0_qw;(xy;!u7{yPj(ZIfG)LiP!Ak;I9Z%B3E>GTs4(NHv_v@~ji!2MKC
zgZz^wi=hT}v7hx_7`)T`zRyJ_&oS+lQuImPjk;LR_LMAHJC_)~=Wg5lO(5$%gUyfB
zNbZhRtq*W1hkx3fpNdx+4MF8iU3b%DxnZ?kL(816>q?h>TYQtXE+klsP$}`GLcm?2
zU&X7qc$rn__^Np|Y?V62WAS36l7;&e(|B0}Y08*RXaAU5eM$$rvl}5H<<K~bFf-63
zpjOCWJ1T>MqRPG(GY%c~iFGG}<hBj2K9OZHy6}aa^N+~s+=Eb_)k4;l?n`8ORJ+A(
zK7f#aUnl-vj%-VJ640i;G|e?#6mNPd%i{GIC7M_L?}Bo@MSBH7!7}qgmng6Ws$(43
z(Aafs;?dWXx_Xz48;$((MBT`zKPROL)ff{|t=Ytjcz6|v^|vEB<<)do9t?)?xJC#g
zZ4U~jo4GVyMVqYFK8%i2^_zvpIxU-pHmb<U&5zx0oz&emNG6?&);qList(+^wUDBs
zi+4T<-=uDv1}2BzXLpM_)_ALJs`?fFUao*bKyXCraqZa_xO1u_&me*jI>1qI_nXg^
zY*w$XV3t+n+ckv_b7t4LZ^@KCrxxLSz$MGlW&{_ygnb!C_@VRLVH9D)sZL3sYszmu
zqf(9O8Q)m`ZC~|>MfBmev$|nOC!<~bcPy(-p2$i@J(5F;7`#CC7X<h(UT`42tT>u?
zgN5&L7_Pa*QM4ARsuA|XVUO`a5DY>@E(8IFzOB@*328NNi6j=$EpDP0cKaQ5k}%%;
zNbAb%^+U1=11xP^C2X#o<8~iN@w`h2aNd2dV6MmsopfVCTaBJmoNJA3>uNC-#ZPoQ
z8b<o&Y53BnKU38a4KflDqDUy<4{eYD@}WSV`*P!n%*%MdQp|Sp+H8p9;ZcR&4^A2@
zf`vXKGZ+{USp&}CKB|OO-tShwCt{=6#8k2CT=3V4?&uP#7i-JxS}$-$&6SnmL$MDK
z5cxG*0)x7hb4RjYFPqNzWfLTpJ(uQ;>;p~GlR;WFQFF`vJL#V_9VN;}j{RTXc+STj
z#do^WvL>ywnm|P>$a&E|U6?IhOIuB<ef-=(?Q#OcQuf@fN7{91NQ=78p33r(sBvuE
zuVA%@%r$Ud!%^hXW|p&NH8PyrBIH_}0?G$|z?FB(?f`2)Ax%d&AD9_zImWwtwW)ZS
z7t(9Yzhjd#u3)1k=O6g$*T@FXD*3o3{BCDqAub<-Q<L-$UD4_pA43I%pRLe3Z!H!a
zLk=1en0?j77KH3WVfa4KxkjU2y_telN?|mV^_i8VyEWph*+|3|l~JSUqOHS^4Lw2+
z6wG`jhz7v0?Ot5YRl6t<SbTU~@nFI{GaxsyXx!sDS-uTL`*p+litYPkKy&#wZ&RjV
zlTNSo@1D^}YS43uyKQ^z<UWqY!}2AQ-1~kwQ&OIOO|wb9598fwcye;(snJDkMXHdF
zlE$DYg0V0m%SoU`aF<PR{ssHm+mUOcP_L-a61}kL$5P&LuOeR@Ha<9J$Xt~;IqXBj
zmfqLBM&>7JJ1nM~&*iJlH`RPd*|hr~NdJnj)(&w9`QF!qa6tOUH+Q0)DlWn_A&M+D
z`Gq*vq<Be^@?l;b<(kR7NeWtlMZ<YqVTVjo5KWiQ)>L8r>S8bBv~^J$|Ke7G{bSzd
z<LrWc3iXzUR9VOGgz1tlOR}@UoAbksT_yU@_jeqn?%k~>6^K0Tn9Gp{$4t8;rA1qi
z)82TKB;%;2LBVoWU1hzySO6UMZlwQnk$FIg97|=JY?6tRK`$d=hgt>1hn84Y&3iOG
zq<+2<nbvsD>`Z_<37Q79Q!U`HV})+28XnFprLwPu4m<IYI11P7O`AAAF6?SLHWmb5
z{nm3n2Qks6xzn%4B+@%<{Jjb&<kjXVpn$BECtfa^$Mr7x%02CqDsRO8awK&0gF@&F
zF)Q3RwYWDT<fD9=(|fHuGf|wTp(e<I(se@cfF|EZ?n_^*G6%~Il<siiiU}Q|Z}@1=
z*%aRk+maqB>|rjMOTPeHtbDKbmSrZoD=-5`Td3Z#JWJ^VKiasJt4c32pPzne?{|>(
z;2UejTXM}J0m5kiWl*x&H~EpfJ~sY}`~_Ao{VML?XwiX0NSiJ+4*fE_1dmF-`3%dp
zgW!I?T5cMx5oe}rCdW|w;&CTrzu1@uRNS=Xd^xDQo%0E@*0@FOJD9s?%gIYPS$b2m
z)goWz3ui#(cM+yPh;&kNnJ-WS&)XB6km+w?<mn7bd0ug8ZSDd+C6O%EQ=KU!lcWc-
zb}90=@7Z#0o}6_%+Cy`9ke6nq&aEj2V~Yz4>C-Yf%8x6ULlj<UDS)@Bt&+y;-D)^|
z+E-Aumrj~0E2$ACs)&}^RbagKaV}#T)Jxk!LcSNT#PR&cdtC@G+y8GgS_}#^0ZA0I
zTX5$2&;1Tr|M(Sm6jB3%ifM_wqrD%4Ib~*4NqdLWupS6VaT%1om+0~z`sfYRGY{Ax
zf>9QRDubH5>DjS%#u|3d(l_MhT9gF}R@=O!vHU)6KDW>JIh{UIlV1kgv8J^=df~at
zrMFS_>MS`#F&vqYD~<nY*Qj&Zqo)F7>MEyoQhqgjl?saE1!fD;e`@YN-t5s54MUt!
zdXRgiTjBh(#}DoL#03j=(fwzJj<yl}4;Q~i*G{H2en)`#y=)~{0<g)RZJdsq*}DU6
zSkwG)bPNYTd6-f*R(r2En^^bWI1WFcu`bv{V|;$Qf*Zk|Q$ayP1hoh1Rc9J$YI4@u
z37B$2PA(%$u=(y(u!N6I6lczuC2y~yB82#*V$?e&q0AJt{?Km6Ci~a^yu*ehn2cb~
zzY3=mP(2ktXCD%*9X8|Pq>N1n3&~b^m4U!tlI2ZYrVw}k$TXPU3X)bo7-Uvm@U6;?
zNUi(WZuszOqVh?j$ARI%SFDp;k8-g0d1?Pwc2v~T-k|Rx!}5JDD3Bf9;utM%JqW$p
zZS87uHCE~Fn8ep1lN#_3{z3^zF+*y4c1SXN<fJYdE`h&>c%20W5gFFlBS26z%of3{
zaHf0_au;6e-KzJl?;L9z%RxW1>-3%0|3FcNt?yN11OlHH4o5_2oA$WtzOy^~dXzAk
z;5;~|6ph1bCOBMnWno!L>#9K*FRz09G-l0|fcY-6)zxHh^Xe0eCa(<UySP9#Z#PV7
zDyV;X2F09_BtlNgiasg^z3Ns?FR5l|U8ucW+#vQp(0S~I-i;1mq{^zTEeh7I-eYA;
znBkjrADs@HDT={*Waw9el*eRQ4A4y+!#XqEAqCYJY!ebrsf;?|9|qt5ULauxBsp~p
z*qJXaFlTLSisg8D<Q`0)f3_9}I<833Hyu4w8ymsW{^O5UJ*x!wc!-uF#w7KIQzcU$
zs?-3;JADJ3ng}PuHWxq`)%oM&pX$u1>t%pvC;|LGklw5MOD2747HK!|A~(#lx=xD>
zoC29YZxIjiJpWhljlZGVXC$rw7726PM9}4KA5Cai(L%k)wXPiXgz7Bcx}lGwQ7R|o
zmB5-IF)6tox+_edl!eZzF14(*`(fV`ehElBx<r}jIsj-_F_jn_yO1M5tRNr&ky_h2
zu9CB9L6OyN1Be@GcfZxbqoJ&j1$o3JEf+OTHT6OTdvwe0YSH*(FXUqD-C<SI@7Zdp
zrOs`NQ}+nGS{2n%s4$HBVrj#VwjeBJtnq|u(I5_6jtL(Ml|PG6jy9-9QUO1F|1Lo}
zkD-rvP(vC2!kuWE-r?(R{~{z_<$H;XSsR%=^wPQ3GDsPZJ+#%|6a<~(EyaZ{u1hlx
zO&FN;<}s-wFQ5M`TGg1B%&AuV7w40h-yDXPPUU_JUJkEjbcgh6UdWbs;6<_UC#%vi
z3*KBniAU3UWdesyFh0w@J)KX&7x)3RgU2ZjD;GoaX^BG+eG|L+Y^Mw#%#S^U*fA5=
z>qITNBCA=x;fEH$?2#sGB-Zx}rFr%>hmV0|eX#G~YDPYkH!i?<=c+;j!~Py-5Sc>W
z=-ph(;wF1*pOiopu3Jq1DT-i0z$wqASeOVv_dw2%c7yG+MbTrvE?zHSoFM7yCiW@^
z3DVL?HT>n7t_-|t8s=?VN=;|5svea^eMT~CB;~o0!>9-iZxYobLm85V;P@~xnpZ`|
zIDF`ios`U3C?~%K1g%-dQWlNB(q*gJ@_%5TsTYP#cb%N?4hGeidJlmcMtXi+^SV`a
zT_?$%0KFSvjCxqRq<>Pc{|0o%{h}_1y_1)+>UfBPuLf57_6@kX$VQV-$-Y^E_w-d3
z!VIvf<(|-gh6%pchP0S?>F&f%uwv;(bL&M9p48ef{D4TiW15~^&wFdm<S#y;UBxYa
zTs#fdnT(+#B*cEvu9}0~II|y417bAa3^uaP)T9mg(_F$t8%XY;Gz-#I+6_?wMi&1!
zxA^Y^_*!0rAasCo&L@pQcynNWHu@Gr6Fce62dYeS!8`OT<RFl|8}-h(GxBqCMMg6F
z``S~{ZDG5_OtykJ)H2%kyz1a}IPZh4I3~M{vjbu;?BI;<l&pSuW#*{$G=U6(9}1aD
z*klcq16JAD&2*PXk@I<kTs?M1<2C;N+YlCw0rAx=LQgMEO{!pXuHr$C1?3%)*)3l7
zW9~P^p8!+VI|ELS66H4J4(rxjE=~H{d+MC6QCL%)C1`6VTr%lQVpgX<xv!9<x|+nN
z<W(@ufG63BxZ5xqqt7uR`&G0n_-KrQT~)H_QbZ7K(6$2<pzhGjQ=P*yMh-wCQ>2(B
z(Lre+DgwbD<q8~K{B$K@<5NpH1a@W}dmj57lpOYSr36&VU<T#LadRhQxz1V`?|0Up
z5zgWm{`<^cLVZ$Wh(9*m-4r<%@S*QjRfV^|3G<RF`h`Jh#4%dI^@7A?g9IWbkxv%R
zpV4Y#9}lRY|6@MD$}50RXf~(~x>k45u}Z`y@k095eFoHp*<s4?=K~#bX%}2T;TOzD
ztItj?c_FD|LK{Mk!8~)I1zo&T?W)>z{JWyrcS@({cNr&V*mMR`bzs$5;}$=PlkK0q
zwurrxr&6@uDZQLwl{l9B;Ko_2H1jvoAiGXc`Evs&Riepj7!|3*ho%}`6Xq67G=1%A
zZjO1U0?MQ#ThyV&#~9|_fYOu#;#8A^xY*@TT<mN1r&78m;jsL_s(Tg_qb3sZ-k0Yp
z9K<{~<MiP)6wZ{9cm5EWP$Jg`(lK<-NtuAvqTwn}MFIBtHM{k&0l66}azo0F>_upL
z<Mjnw>iqSJAe!HEK+>p7`=7|ZYwtbxyVXs`x6AlsljVyUvZS<%A)5JNAtOq!r4o9^
z7+YjrJIF>7lvFQhTD7^2Xj!GpFzuU`>4x{TM&0#sR80diEURB}0uaGV);VTm5%&6|
z->j{PF&UJ?c-N#)F|I~lAFAAI3)0suKa`_xJ-0t6MV8uNM4Ty%Z|LLTmv7&ZdmQ@a
zN5PR;&J0V*#D&pcK_3M3wH3I4hgX`oK`!VRRxVK-&Mb)u)wzpP>?A$D9((#7TOn(g
z?dj=QaaDmL9K$!TYMI7Uv)TlKRak)BsstHrn4jbill;oDa9{;bN)_)+7BewUqs-Ol
zg7tz9e$N2^OM~@$NBmn`Xlo8D=nlkxryrLkXldIK^hE)+!x+pXp!O2%sdEwc4a>vQ
z-f220z1fDm5G<rLXl$fEZpq{n^?s**=m3;)ytJx`_TgmJi=xS@UiWaTl9|96CvIeo
zW_8ON#K<$OihlKRV;^ct$81W^^oZ67!%9N29QVn%V2BXMA&Jg|VAZ>)A*<G!;YAX$
z_e*7rS9Ph7MNEloir8zH$)g0<W-E*g>+&tgZ>Tms%NwbzF$aioCdIqcI;WG8Ed!rk
zmFr6>Jq)F`8cC^mw2;6%=w0zO&eoyEQB}S02~E){Oh!#1uJoEhL1aZcfNg`c`J=ju
zhwU4~NRHPTA5Vs=DZ&MKj+6`6X3O<3LY$1A*_WZ9*Ywv0G}Y&wRqvmp8Pw*tQ55<#
zn9Wa*1U3gjK?5F`^2Ooewc}z|3s23>++9md_NK@`%w)b*#KJ=nm4<se>_{YREQr}h
z+PF@&XQ(d-I|FM?iIy6>QxI#P<w*^uJn8BO_<v!~&zq-kylC2G1T-$dx(dUG-x6b)
z%ljWl`AQ)@eobEOy;tGU$9TS`88WlZjG?dn%<}j9=JD!(wv8nh%=BiTn=I<Az5?K-
z?_3f=_zw+F8H$Zwwmz`wVLd9%U^Cl9u$h^ItXd<11~5H)l8J=^@KF6v_fdp}sT(O7
zwg6-1_}}-k5dA)I$fD^U!v9Rt#u07ggdwieW5SNe&=H1Uo;XR9%YAvoYAZyhzaxo=
zglj+ye+7`^i~SKrA^#+Z{z(sE24BHd<sIm5cxj-|Qg4&BsAAz<9>`HdVeKz2?sKKr
z#`7cTlO21fa2st+=Hif7^4dKmNvu#%F$V$J!+#|Ku;DaZ@9wVl^<e>Q$mXE9%Pd9z
zx;4kdnS{rgO%-nU${@rRrhSAE(MaOGecg1??Q19_>E@UB3|{XqwNU&&5U+Mqrt6Kn
z^%ePXR%Lt&bP~lZCWi$Ymt!Io!_Js=4#2c2up2pX;4S;ah0w(me})&0yNn{l;{A1J
zAa>XxQ?Og+?UhaGIbZLX(jgMnFsRQV9gp#5JU+FPmJ=rj7_u5jJ-t_^k5h{yYwcMu
zC&cxTW6%*-j=+-bAU++WDKoZ2z2<w{zQX}*Z595H?~98U3yGU<FW5bXfa7g&8L2Z{
zRy39WqVRA3kKrnFL7XQScqoG5neHVM7Isjwt_`y#`m0RuC(hMh@7Pq%M5vuos|Ag{
z5kcW6HN2D;%g1$Nwo^Z&Y!8FR>e3!T)=wSE%E&B&Rg_jUm#bM4JZoU|ShaEf2Q4VY
z8;`5HDZLgUSub10n4tTL-I70JNR#zk2v)!@S%ff`le#W@-9b$4tc51bnq|G(dp-ci
zzlK-%7s2@GWK;aghJGS~m}PO@I+y0vwS(SzRD?C^?a^ZVL7Nm3_oX}wvte?5(?0GC
z`<>LO&ABr0GS<=L0;vK~x#6*A^@Kb-v~!(V#&AJo4dZ>-OG`ot4lSQlHt6*YxXnsl
z<MDnfCwNj>$&O#Jf%Gg4>Ts;l4=zQ;_-61p*EA_@vNKycFbc;bPbwb7SSRjUq(F;z
z&;)?~ux+NL@m`gmm}P*K2cI>5VUEoh0(!p!Fah4*^rpNY)m>>S#x62FV`22PG4XGv
zbL7Q7_1{{#zi9ojo=qnf?=40<trbEE9iWi2W)jP!gP3=ExfNgBy;>V$rXd7Ge~C=;
z-X|*Auip`fdYgrK_9?$F!#eI!68CIf2A(-~X0NAAXwLE$n*8<^F#q~<NaQY#&Hdmb
z7AF2ZW=cF&PzW}1(Q57sw#6iIn}Y5TR#$fPu6t9mQDQo8AhJ(vm;Bo33#|j2@=2LI
z7SKhkozy!G>Tyq(nv{4W%;flrWVy$=n3Ef264PT^LH)SAeyxc>yQo$6fv4u@24!?n
zJ{#0WXXCj*2hrOO`FO9yk^h0X_=J&8nbvakY7M5|oAr*-%(rFozB9=hdC#4!J=q(_
z9pNMUeCJc*-8tAptY#m>p5s5nsYL@uOeQU5J}F+|Bpo|v@E3XRGd-NhAx#H=6&?GM
z9MYX1f6h8k-`$oL>#mvMH4QNRK5TUNjU{`iz|-nlb#Dr!lh|mwq%$e0*8O|t#VCHu
zKGfJxt}-Su<pSZ|T)vWR+H1`reW%%Lq+8n)4GJqpQXdC}t>4WRN_xd<@s-GqKVe0J
zMR;9KqRt94;f77#&c(Z9&Y9}Ye%TycY0(h#rX8y9S*;4))Szu<0xngK%=Bk%oCntR
z;$l?;eB#77=m5EOuBVt)L}2iXf-HLi0m+_~d7IDn(a|^g&U2&rTTVZ|dy9Clv?Zuh
zat&^9g7_-shP<G6L%OcoxhxPU`$7Pq=(yJEy;=4P_v0!rWqJ0zt-{thkIsu`BGces
zAg}5zYIO)mq|jPx0+%5S;-nS0V+86TO+y}g7(za8m!^oDrc<i_+bR~xR*k8D8FlaV
z<^uhk{V&U(&h}0hTZHUwBe9@+t=t2EAvjBW8^LC9>Y2f;JAGdVWYl-@u0LoSH36K@
zEG{}gux97`QYq=Y%aH}5SRX~3Z{;va;Y)$kUu7ViF@XTs2H=81WE6zHkpz!$jmJHc
zyqvonCP@#91&h}*VCNr7D>*hH*O{Z_#}(KjK*C}-nI9I<JX2LZzI+k$)O=kt$b`E1
zIK80YudWS^UpuMSYd!*)V~3~xFk^^M9%AZ|W9ddRZ%R)xXi1Z;Md9g*FzGcCm01b#
z;PhC#s+_TrIgZK7x_H~Bh3B3f(SoI)mC^pA)(Enu_;?#VY_chs-}Izx9k%ECEKtm(
zW>Y-FX}{3gz!^m~rjzxX!szrS4zAbD8nV!1v24z@;TM>ym!+w!IcCsugN=-&oju#e
zZS>^iYt=a-Auafc0+0F}fzr6j=WrXsLE3n5O?@w@m-JoD4Da@$DQ_1EP&3uxJZqQ2
z=IwoD*wE2O^%M%*;1D!lxLIcgoSr$~13X51A2R1LhmGsfzy1^_y5F4H^^RZBo=Wlx
zB)zNh7nmBKd?uop{%)G89;05tqR}Q>1;>K2TYCW1LKkfe95N<Y@gHm9E$w0%(6ybE
z9bDRg{+?Fev|p!}802vNStVG#4(0ug$Qed#enr%DJbGk#OWbJ|jAqf+n3fq_yE<Lu
z@QCM5-=(GjlfJOaDeFtk>3iJA8f)?yxlZeAt9nA^p{i2g1oClCn$1V-($b1{+3&%y
zW6|vw&0XZ(MHws7l{U5cIgUbP!c)HV`(TWWnu-GNr3-H=`-3F2eh+i4#L&Ng>Bm~<
zM!HAz<GyJsa^2-F^VrbtkBzDi5M|aHZtoe5{#C7US@p)q-FcHVqbRMa294YNpE1h`
zKhmzr<=uH^l3DVRokVUkWDBg%>$itMW<=;tj_OiLu_Ie&ctxa(9(`!zhbCs2f^1Z-
z<fsWLCG@FMTN+1qJgt>23(_qddANq(9^B6fx0~<xM*aBs5-S^+Va?Z-e^X2EN7Z5G
z{2mJQ<?mPH-@p27QvHe-t!xlu4ALntBjiNTUD_)dyatJGzwKQ$GB>`P34S<G1)bVF
zh`Ps`HP)VaE_4e1=kWTsKz6QjS6N;1*ndln^J%Ck*gI~yB>;PF+x89e^YDl{m91I*
z-m6~Wjx8JE>XBRw!;|DdEHx@HcVN1DLV`D2F0)I6D&zzI7_Tc7k$Xe-{Guk+(Ptqi
zi_hi2T9RRL!dyYR4F7tk;7?p8+oo*Z!r=|CDBTrVeJu@fM!z<cq2%XuKnizWU3%GX
zrrwR8HmK~%Wb4RY27-Jl_m?e7Rc#IVs@SY(RA?l0zQx!uLK;}~!gkLrphWUmqaBW*
z^Au@HbIwJz^g*22^{JdAzElUF3ZErIdz9W>%%8fuT=o8vqgML(J|_~x3(A<0&;>*L
zn9c9?inlxBLhk_>fztPFh^|ZvGR@)6wt&98k%|Aj;&RNsV_DF!iBsQ+zCN}drnL!8
zGWJ|0rqz$Mo53pb7~W^bWKkb|lQ7fRwI94Fft2*~m?1^Umb7*Zag5~Z{gv-bgQkC|
zblRVxjx~z5=*zz+?AfSxN{6r&E8ag>?XEjAVcVzB0OyEY;4H9<Tm{ZA!c4b`v_}g8
z6IZEm`=3xv&u*=eG1{2Sil<;+zP<D5RR<p`OZpOWi9u`Nvw5apyt>6$YA46CJKe8(
z{H%to=~!U`5tB9o=E?P4#yc3v`1d_)_3tge_2NxOn+l=^5qzJ$^_uu|&^Z=di%+n0
z?*N~-P_;lV1A#;@QDXzZ0b`QZX9n|nw{{#?pTuoBQ+cJAiT4XY+X!$0yD6pMPd6Mi
z#qE_jUp)o-CeMICoKEKg{+IQK&<9@i;*wb;YQ2=7c6o?G=XRgro5o6fn)l!`fNnGV
zsstgcZ`t!(P!jOW(!GO>#15T|5<KvzYCBTnKL^?Mykc7*^oqAS*!5`)!diN}1y{%=
zST$_}p0JyN)-l#OOR?xt?B#+c1^`CywR21$5&RhX#r-Fb&Og!M7=Dzrxozxzyjf)4
zd`;z8P|qQa@^^V$8(?yLRf6RW=-2!dt0WRsLz&<-@zZ@X9LO5j*8<?;V#qr(R@XD`
z-0!LW=wirsV0N-?whr9ieRF}7B``@Zh^Y9L`uFp%NyYi(#h&`s;V!YGAwZd(%q7dV
z)m!N)YMkuq*v<lKI%UVqO5A@=RWfTgA%qBqwz(~uql6KC{Jb@%>>Wz1oIT#N4?|hp
zd<p%QIh1v^R<s{uNPvp-qIwI|2eh9xN9BDk)E;L3w#M6!k-qJrz%ajMk9W$|^(7x@
zw;TX%Bf0>hjwaS$YTBrx8IQ@iQPFkXDb5}fXBHTOlXTAb>U6G?@u-T{8$=AP(W+&T
zS)b^*gK63vS$I?Orw;<9WLVFYPK<SuV-68KDsS)a0N0{$iNRv#myCYz+8kkWOBq9}
z7L5?<Y|~AJ*$webv*EMX;JhwI8_0+HR@T$c{Gn%yW4!0k8peB<R6wUQW!c=lb>Gxp
zzY-T}%-g0iJCistfT|rb_<A3e!34S-qA4hzMV{sG_qx|QoTLg{##muF(&tKYO>7TE
zz!-{}+S7(NARq?fB=kVZbqsRUkmv6`C>8X{1@-QpmCoGYSZRwNtWTu`c=vm&E}lgh
z)9hLfg~t_{+XpYTg7r9U2Ywzkym|eu;=?X1@88b&@=^{mr7*_}zcG*`6;kV2URrv@
z4u=YsX}tbSmFI?L^$C3X&*t#kTa?DKWs-<Ly8@siLoa`}ysTc-#UcjyX*LwxZd^~9
zmSH83)tGV91EFs=3;waoH#W@Q%`1AbdtWG|)oS&s%SC>%3zM8suGb}n??T<d1ygvf
z%mGdTSf5E0r~924UX#9&lF<)_4_)#(uS9HrDsC{k@xY*JDBjAIG>f!VM@f61)V#w*
z&H45js&NH6v=gfCQg6x(&Hq)Or?+C9=V_12s>)MG(-~T>*@X|ab}=v#(P|GBsf!7o
zT?)7L=8|(qco}_$cE+xW#DT(y7MZOFjQL3YFwC)TYl>~heSQ@kY7!Qfa}%=|bRaP#
z!t2cK%AAwO^(6g*%@{N=;G<z<AEEDx2_r_8W1Pu!K$DdtoqJSr`lDQC&)tCZiD5IV
z<5@uRf^SD@yid1LM6=@eY&XwRk!(_;jW}bwLQ7kjH~>u!?z@P((lg1JTJTi#D@y%E
z;mi08o}yfStsQ%%h5EN9D3<#-&HYVJ9^Ta?mqxZPZU)3eG^|<{?&KLyXH@-!?NM7U
zchya5pVw%x{!O22wIh-KVmWex&T&H7@4p`?SBy7$2^gI@0P5ikOPg&by?fy;r!DBu
z7e+s8uke8%IMrR;Sb0fDLV!rq8WjcArh`yhZaaVe-R+gy<$CV+v)b9|195KS&ck^9
zpH)LUH=;kJK;vm0+NCnRLsDBP(5$JP6mzYH$L_<tD~Q{op#iwqEQxyFA-(A3mZ%1z
zseG!CAIPgb!py7CHT&YZ=H>5q0~Gc%7jErdxuJ7DhS7hk&bOnD9I;SW*Ll>tjqZLf
zwH&G|lp=0Xk7xC76cAtAXri?{mA|b)9RGpKm3T3t3UBLcRXz)}VLSEBl5Yy0?;TfB
z_=Z;PJxsTX^DNI!l{EDPdxo82l9)b@@XI|FRS0H^cE4#Zkx|`SV^KP16pFXz-mHXz
z)|0~uEtMb5EX7INgZ!YXFuhfxyXTTcurkB{K#X^_zvAesA)pGT!FM8&)%GFUb@S{3
z43B(J2%h+|8EP(T@F#QC2~DYU#9(;bLyB#Ni3X!u11`I1c^{*8p-enPK^w}vGgoI(
zxFXZp!FW*}<JIQv%kg5S>oMB}UC5XtfMnr6wX?|84tyVX(NH-QXO?rJMD^nu&pkgm
zYBR^d1}dBDIc1_gpcQ!nxK=rD*K>0WzDS!h5ssOA`9!P_bgcjNfm?`+*LWz-=cdcB
zX>yL#W}$+2dI2$h$oOuP)j?117J{8w*R!xSj)JNrolZjHuC(qH$4)Z*g_FaOK2glI
zuRs{W^<I5lbl|E6=yeV{e6g@_7B7?E;<{IOr6`t~v~D+{rP~;j(1#;^kQ}^W30xUA
zyKUlz&5THPD|ZV2h)lU()JuFMSIu`FQ`n6&bxFHtQMB&-$T%N(ocat$Z`SK`1s_{`
z@Rvb++~NudJGPRPX{>kl^1jRV40RM%HU~tL1lr_8tIj>qvAOz0oCD1cGNJ-ghuFws
z*puHtM4yZ|IHs|9$n?(+$756Qx}?1Jbhmf3(8Ff`yiRR>g`4K4$?*}sXmDMKoF9C>
zNpP89DTuKy@d!T_&m{%xV|`}UGF6ztVy(Mg-#$uOidnBjoHuU~%dO$ONfj0<bs@jE
z0cT?W8T!d!-QL?FZ%V4qf=_2o(hZPp&;<XZVj_#Tinzma#72RcD4h7#<lfD48+w^d
z$^KlD`NC#|HyC=X8#=JxAZP0-D}Ka~sWcXfIHfBeR1a_rXc{px-YMG?JC^UENy2s9
z@9Rif4n%3)pmRk3qDg<($6XAK<)njeIB+b{O}ob=N=G44!>?|CS9Eb_Fbm`a_3zzN
z5X1_66R2HfXTJLSY2B|8h2#I$-^b3kJ@1=S?(}KNQ6F<;SQzX%@GP5FM#*M@hpP57
zaBpKW#VTlL&Ig;VhLj(^3D;Ee*BidRlyf#)Hp~+E-NFX=c(p`PP_XI2;Rx(TTRJ{W
z=?x1xy1tb3UJSO}u5(nWCu=g`;R4LeuD@>5@?nJU*(5FG@7}Jr<&mO<K2e#y8OtPZ
zX)7%q>OTR|V|N$a{2!>^0CocH32j$n<EyS*fW5ySCwnEHDvLb&D_|Fooc`=BY|ijz
z`WZMn*!0fYpD+H&&pCn&7p)+>SHB%v>X2>^4@ljE)?M+aIPnBuyZ5r9K>GQ;#{%ES
z_yyFt6e6!`j2%Is@iTIpQMam#HDz3pk4m}vY3Jm>T~)Z~DE}D--6;k5e{4Zh<KOP|
zB>cjx#5H=7Y;@23x9Wsd=&U34`{jAfE{j%K-jjtCh<?fmT~hAo+g(2oU#ZW#FdkMz
ziNl?tbcIIqam>>`{j0&XYfl&<X%Q~`)thE(!}lWo21K)Z^p07k*$J;-TnS4ORY2};
zwr8+G8zCY#E{@Vcv5{V6fg#vPI4#H*`AGP(NF>dL0R0gqU{T&o3m2rpb4@dDiLU|b
z9D69C!&2gF*v@o=0ZXp>QcGa2HHm7vWn*v#Z(GPRm&vPr3u+<e{lL8@EMckZPv+?T
z_YGBnUKje`Ub()QvTZvUb6n3V<#n)c@kL*r!DoC)he%=8>eE1}HC@fsvCHtet6j`y
zeL8#Uw_hoS-^GGDrq9dQaAkvNn;&ZD<u7MEmD_+$rUg7W2u7np;12h{0aF@XQq<c^
z>KUJ1k7~ad>4-DIZEzP|{%=(b0|bFe607xgt=>5m`N%izER(jO&WcsLCU!X)5gul0
zhNu=P%rge%i<gFiX$~S~8!!>aq8oQczm*H~6-0>a&y`(&?vw@Wipw7S>O2zNdOm(g
zT7@QQ-t_A>!@WkiHU6#?aX6adfsx)g5>55wVBZvrb9#RfMAffj{loN!9o406TZVzX
z=n&2^4G-tuER4^I`+9U+^uh(Z^C2QxpNKP_{Gkux5RCLDN!dgrzB2<ZL`xmPIa|qd
z497pJ=3B=jUFA?PQ<P0C{Q~r3ORv;izR?QN41(_$uB;Dmaar@}f=mPR+UM(%Tp8fM
zSbUuNT#KMEp}^~Zpr`h}Xj7XVk7+9G%f>IH*J1dUL&+n>1P8`1Kex?YUqM0r(qcI-
z8=e`@gL@(Ms`)2XoNQlXsq{3&z`A8}D-maJP-ud13$cPPIUlbSgUVGJv^15xA_2L_
z9$NKztR3=5^b3qsQRYEd0HAHg%VC8@=C$&Scud~#C`hB9>7kx|u@_I|gFo+m<S6s@
ziG`Q9$k=?<a>uIWS=5Gq^6PsVlli9C5kvjar<k)6;ksn&Tew*%*}>jk#>k_F_{Ci0
z!0$3m|15>CH+5}~+3)8B7%4wKB#d2R+$X*2b~BkU82kaC&r-5M2IC4Qn_D)vYE;&p
z1OyjEt5N}v;9>`@6tJ{`w9u<`8x^gs>FSZ#AmMLn42voX2yDp6!H=qZlqei5h&C%%
zZPvvj&IYsC+L!mRK+4g)3Z=na(hVWy=~wIh+=4i)?>JPA<(y#{=+<28(O#)DRLoSl
z*XH|aL;BR_tqWIoG7DkA{Ej(FYoc2btmqCOzzueWI!(rnho1S-rxeXnJm$(BDj5kV
zJR+~K>NELHA(v)+^zJGK!GgA(8$CBQc9Sv^X~)Hk`u@d+5#uJx!En%%7~pLvJ;ovo
z!EVJQBhFiW0^m`g;P-Vv120=1!nY;z^%AH!999W%=%q9sBfiL=VFqmO(!dY@p?fKT
z>B4_gE{ibgSyhgnUOvO4U{<!RLU_iwpHoP+g!TYLtnKxu7X3GwDlF$+m`NuxWWa8D
zQkA!xA-|DtJhh6KK=hQ~gBA@F&hbs$!;Y3{NhfR}y=ZO<gv>EpD@I&sbTK>evKI@n
zfJ2#qR(f3E9-Kb_sqA)6`=K35>-uH)icW8)G-gCcZk;=S*^Yc%i}eRxUu3}zP6^q3
z?62KhDwx}E`X=vgPaVyOJalf6Zt_#Wg0IU?4BMoKS&5Uaa88oO4kF&0+Yau9PV<U~
z)Fch8nJk`~eYxBhEm|;xTFDrE($r^9wPA)!w5Wx3xAd8;Kb?Li)$~|cO4<nDD_NYm
zWxmYqB5gT*S;-syJiD+bvaUn+Qol7{jau;9^VFziOekCd3E?wT+=J*r4BQT9h#-II
z$O)WVf(45Re!4sbSd-<EQ;oVl^ju!J3Kwhp@ZRG@ezV*~HFu)`$&dUfa_d6?#;|)I
z{aRu50m7aAwhi2E%J}B%M^GX=oCBCL5)fzp>+q_u2ZYl&dKYTz{1{(GU$q6u8bo*O
zVd8I~qhF~OQ{Znq$ntc{0{)EOV;=9^67JW04v1SWlcM&J-i_vygVKhfoB_?#PaVGU
ziKor4`BMC++^t7naN7)N-|8MZF8)%{pXBhhOrr`@^i16*oVQqncdigjX>TyGNb%|P
zU>h^PTPy1e^33a}8Cxi@IX$&{nSJ5yxRi6=wQ5vC&do>lepzq00v$1zRB~6Nm`I+*
zlyPg9Qq#?D(`fC9&lYD;akXRSopK$}AsGwK7nqET-rE3-o4Ri6ti8;MpDT5|*i$8X
zdrX{Xr@0BTFcZGT81o?X9IIWHWw2Du%VzZf-qJY6*tDRZV(1{E#XavVk<-Xp-@8`J
z=sDj(^Rpn$0%beY2B;@yMs`i$K*#7H>T5t@rWXr$ruNu1-r3x8s+qCwgkqNSp?cGL
z`c0RO>Ptba+2voNC+2FkhZ>x5;l@b@L)-TKAic0XXKD#8^TpaG@tpVSaj~bFz>`k+
zYjQ<#+avWnPhm-`gsvZ&Qx+-dooSw7;=%;&G1WG0DTf7?j<$#kWWTO_y~9OStYCX#
z0ovs~PjC71@z6JYOHibc{Gz`Md(VB&yKp~A)B@??=jx<GAR9X3P$kyBZGa;EqzM(T
z{qX7weAdvPnir}4C>k!YRk$PkbFJ@!N7%OWXsh;w#*rK8N0p#RvJSOUVw334n#Zh<
zoqFEivrH+(O^(EGTpBQ8rutfyBwcA`^;+Zvd@n^|+WD{D6M>+;LFicNGQ>qS-PSU>
z1IEbZl6^7Y5+fn{58Ql{f_}Z?p6-IZag>yl1D!78W8&y%!C{M82UlTL3;WX)W91?v
z8<jy>svso2hkkqOAKrnG(;y)KF?P;dfLRT*MsewL>yHuO**EyrvWFHR8v{FHoaJzs
zOBopqKq(v&GrWp`2+&#Xoq?{}PXX}Y$fsi@`%)##a%bg?nPIMcInNmOj^J9zGi9l^
zvuXUrA{(ue{ajW7aQ+m5r{=eR=;O@AxL}o#K9G6Y5>2ezA68bbB|?@5&yT@YvKX}0
zTM6_zn;C=7L+LQWAMpqpQ)#zWaU5_;GZuK3-r+m~^IrVivu$}`Ao3Smq9(}eX^!as
zOrXAb0SV)FkrYH+5|*oorCWSX3q;gJpUzf|J3<<Z&*YL=ugdo;d<bMm>>*jpgBgzd
zysr&O7q)^;4IX`i53^pO&!>h&FxS`m)CeNq&pggyT<P!>{2HowiK7vI_mFA`1U+Oq
zv24A2h(y{zxj3Ydo()POg1G7h2h0%?v{M_F2vfx=b^gr%`0xxw#D6<U26l@XFjh?-
zd<LY3topE`qEmRciq|Ll)+O$yGcj`!>tk<J75S$~hZxFZKjqqA<5O>!6QqKBzA@oX
z6Yl-$4ztYR$@&g-L-Tnu*9dVe<|bK$Ru0Y=hZav>+AuLlh%jhF>q3TMq{8kkj$2Bj
zp;0lXwqdfE)#rF42G(~ZY74u-0D`SrzC#^U=;l`$Gc#R3qFnHH=bs2&-8i7H$rW#d
z?-_G*B`M3cXdrVgKKop~vRn{`aH9B$4wn0x{+^dCC-z~b67!bXQ$1HJ%`Fhs34Opl
z+ek>s6c>Sur^sT{(fy7&r{5KIJ|Bc=l}dTb<OoO;N4sBfC(Sc|{Ti>9)s!VO+Ty(v
z)iV;rxE%8r-aexLZ9{rc`e+PC$?{39SNJ-~tYl?Y?{~?_qqjoBKKcA*?dm9InAbR*
z;DTA(|Ap}&R>OuW@$4#QPc<OqZg||f2iXKgI$X+2J>jhj&0N)O)cP0k#f4)DC(9*$
z=ha-jAW9TQdK`hLBVluSd^{rOI!kHv|6kli|JS4j3cPl%<>Ik|44VJm>4mkPpW$=0
zZ#tCfQc^n)`%`O|cL=u|p(2NV*AsCykLHoRf>pjl`O^&2T7#mAh~=X#VrHTgtd{EY
z+j-$)*m_g%W;w*{7FIr&51*cA8y2)7It+L#u3bt{_{A%+%%Ff=$v){b@-y`5-1&I^
z@}{GgiB-;H&%msxa`FF1(s>6`{r>;^m<QP%oWrpW;gE8y>^+Xnu~*qUS&<{M<8;EY
z<2d$qj1VeCva^nnGLlUwp;GGoeSLnvzx?lT?)!c|ACKz-FSL6l)J=&OX`x1FsN2tA
zc<!(XqMLQbx-Wyd)mEHiabf_qp{Hx;Vt?9>P>jtrY^Z5g93(=%uTo>KuQM~3c>{;-
zn(Bk{`)8Z*=8sM|bdC6mgCxI6P`y`%3%se+1{BIANGiIeu*yFbCAHM=h1k~Di!LvI
z<uZBePP472uFhilZDq+hkW=d@lDiMQk8<iP&-SgWUoUYeE|(CCM-aj%g#QCMbQ%Pi
z5S}Ag!ryajWyIK^tGnAuMiULkP?n<{LEq`Z$8IQKgaWlzZb3e^>1doN*Zt71&sCt!
zP>viV5jW>SN=nU+QgRi1EQcBX!e*Q8{B^w+Ev;1!hrdQZAgn|~=wN!68y+jr*#zU1
zXOG_30&TQ5v^x($;|}Eh13gn}?mh?~4?_xLma5$>++Ko^6>^~nr@j{-Mr7IJ_Wu;|
zaV&h6mez17Y3w^wFyNx1av#}`HQEGBAE|VHKG7}J%CugRb%#zTIRXVqKR;&8yc-=A
z?;_hj$uy0ua`<!QJd}_<-$m`AcDmmF#Lu!iv)bX(cRN2nvhpyj;o5+jEh5<|bePLe
z&w3nwpWX3=#95ttqNCR*Vc=g@=DJ}JuI%Z~7u#zx?`abH1~EF)&Y||}FdF3wEobdP
z^F~r0H`N+u1Awy{_$yLcrzsVM<A<1`Cop)pX^^K<{neDe4G1~3U;1B3@b6eZ*rl`4
zI`mKs?+g_rDPuk}9A;z$+93zh@TSC`s}YLs<>J|w*E`zsMk}9-v>>YTMgdEbN5Uta
z$}{?!6x-q*7;I89UEy$%Vdx&nIx4t(EOt4fI^?<&b57EseHIA7iOB2j^=<SD!TIKn
zEvyMzqlT}IUHd#WQ$T1k!JIyxpOl;O0^^1P(v<j(@AYn|XaD4f_69+6{$B@$GR5UV
zuWsrq4S$E*BAjJgT=FH8zBt@?sH{BPM&^)yl6|8_@+y1uZlI^4aU+q#E-KHpVVKli
zMVmF+PfHjMprj8IvLs;er}vZKcPG3?qLc-4sm%TZE%*>wu}siMc7OYKV;d;nKGb<v
zdYWn~DNGL}b{1b2rzyJm7s()@-czQ-s{|m>swaea4Ps`r03xxN`mZt?I3^rTd*KI`
zr&_VPnNacu2lVqnkS78iZ(&s{XJy*mz;J3N{}ck^UoFG9zW_3o1*9f32`!0BG(z4Y
z5M1RQZLKDK`3EY3_ib*p)hBO84r1i|w(eBWABAV0xockpP6b0>Hlfu}hh}-yl_kSC
zEI#y{BJY*F78aL8X>6Me3q31ie=oux*4v=yIrp=QCimcR+Rid>hDBk;PR42BC{ujd
zbqlfyOj4$Msuw|sq<m<0wE1jkH(VWoP7t=yXM(s<IieRN4A1Rfb1`-g!!~Vl)q`7s
zpt8cA%~fYmR6{Ef*V5Fm#`+ie(ACX&lWm~@c}>sLH8&e<e&)mCrL5W$<gPEy5zh1Z
z6)RYzRsoAlVr5e()~|R&Jgv{7a?CZa?2;e8sa)?Vu}qAWQX;mg9qZlTlRJZr9X^ix
z^R3M8_I63=oz7a%lje<Ki-3dFA<nDXjoNq0$Ni~qXB){*EH|UNG@Yhq^eFq^ZbQ{}
zsGCyn@{V8SE*sI}3r!5n{tx7;?Ebj7p6*e-XW@X`Msq;kI|JEpoPSDM7*n>&9JBJ5
zz8duMF~B6=$pG@a5tEWOO22)9=xdRH(dk#$%+%}Vr|C<79qBCI<LwA?gxzl?Kl;{~
z$xga?K+202JcFM8-wI!s37H0Ln1PtQ-vkK8Mf;xpZPiO$+1>3Tz0pFBK*h|au8Hpl
zmbG&yW-k}|{Wly+3M!5dogCiml-^AG@d1{~IJ#!*nYFfqOR}+hVfOU$7hBhuANAhl
zPlm&PQItjoX$%cN$Lfug`et2>Pm|W4=c(rB$fv!`_)r<iigGY27DF?xu^D5z)dT3w
zD{)Sy^dz|={8>FU(js8{rN!Mey60=uz*y0SyQ;moFoCJ1R>3~xf0P20r(=2z>b%Vc
zojIwVBG1}gcFD_)ty)7k3p=ZqFGAz*qYgAUL(n4ids=KgR1VY{73JWqkKp?hnE64e
za;o;A#>5jZotO6q#EPyQV7+TtUY9^cU&e%}__P5v)Y?zFqM8zRtnXq<Td?zmB!agL
zX%ePA5&=G~H?4aeJr}2t_c-T&sRv*RW_`L{8TjkB<cm0n13Xn>^I`HML2E165i~Ip
zk7a6vJ!27V(ukb!eHZdJqAnXL+hKUO&&sYy#Y$gMX4v+~MVm4Fed}4Z^-9|3Mwx5S
zb}wSiLL;8IQxiCz^P*>0-;ZWH3%}+`rrk<|gYH2fGKqZhIOQJ=k1mT*TuKb|cOkrU
zFB9P21)}f06qS8@wDykM!G$HgS)ssfbMHr*BW{QrHD3Ca*Ho_n(?o&J(pg@7Xeh{L
z5C&!S)-z|k;WlEwu_`Wp-1z%Na(dsNwi?ixD26tC7O4*t5n`1rHw?en)i011`92Nw
zS3lqdr9M8bbK=rwvGCxEZW_WbJqah+(olz!Bq<V+XEdrQhN6=@<6Sz=zY0%*Z~KL3
z(W=GS*qwSZ2LlMC^xK;K!#}+Zg=|?LX`A?7HgQUeem{I}<dVel-A1N+Tta9D{_u(G
zkYrfpxg|B6!>SeVdZramTRVhK7@-sTxcDNfFv1~nmaUr<*d^wow3Z3G_{Ryu+6gwn
zT&_57&=0bQx18iUf0Lt`z3kcpl@%zrbV3wu#7H9#j2X0<QB#c%6gQnCfRw?|8WC;7
z@nNB#wQs1dF1v*&SIL>r+-r>Y1Tm#LK~C#NMLN+3N^qjhak$oHs!7cfK7(i_MBybZ
zr9Cv*>-S9QP(01vpnL}K#G*=}IyC93kpLlI99MI!7bf{&ZDT5v)sAflSBJaZY$}LR
zYR3*Y7FQNJ^xBoD%Iki2SvzGpq@yhzP9eozm+|XaY=%pO3BLhFVJXswGKM0|_~@58
zTkbz|%q>C!KB&{5)mCTgsfao8X%0L4g!f#`RRN3qRUCs6c26WUI*a+FBlnDvoY$c>
zq!>#v=vhTP`(_vt>{6t4z4fm@3k^tJLY~1`7xzEjz7jBE4|8BWEq>UG71M%}hBAvO
z@jT##ta=eLHK=m#U&IT;Y8vMM4{9N3C0_+w<NS(bmvozBfj(#2mDkVGf2a{N7VAD)
zr9<(kBL=|DoA$b6BTo?2cwI#Oi^?C%+)?OmRm52XI(0JKvL{bt;d47cm>pbcTVCBP
zT}buktmiY>_+>+=zm$XjDMs*;9R(U)gFzu!3>7{~z@HerhJmgzS|`2v`Lh~K#D}0L
z@%+nFb_mmmt)+NLNA#ywygoZd!*gz-O6uhfzZ9dK-=k0tz~I1@DVcZQR`_ZAWLxY;
z-`{|SafSIge`fDR(q+xb4)owtR}%x$BwfDmiu^+;IKiNx!YK9uSI0YqES8i+woOj^
zdMHiq)Mb{F&yu3*O_lp)cQ*X6f*0$e^Y)pn!`(VJ@8sw0AHP?Eus+Y4uF;M)kr$q3
z_f$Eyutp|WOZ5M2YVvb>M3~?d>aWX!MRoHBG|8N2CS~|*1r5*LaM~Xi(VJwyU&Tsi
zajoiSmP+R{bEP{@Umi^lR$kpR-G65T_Tm3U)9cjwta~;|O|rs3@Looq%bPa4?|hFQ
zUJsl6SJGyhIGs&(v!j_?Q8+2cGqa`FCm?ch2fhQz;m7*}SX%vVjwy}K202f^DUpT3
zLbn}C=|z2ldPlJ+vvU7LkmK0Q%|TJ$tUs@M_h&fOE%fI^p0_iz1;_Oc1}rJ=+P2|w
z7%|$@Jo<9r;J4)A1b}3`dbeKo#VJs|e#UG+L7UONU6Q<5dfPzrYFO6=eNrg$PiyJh
z^^%lO<^t)Tdj7v>BVBH+77HFVBX>=L9vw6rv57uD^R%+?<Q;kk@dEPDQOe!IjASLL
z5z7Zovm{ouy5^Hcp^XdOAfa9;ls0l#&Vih${(K?mqbmCfb6tN!GPg)~op;3z7giM7
zz+Su)q&W8Z@OrkFI-92PX&_%-_cJ4|hwoHyrriZL%-`kA8RRIfKD*eOd$kSW7%MR(
zx~eZ7xJ~R$`1I9=UmHC2q;wy;UM!smgm=~dB>Pv0df&X^J{8d9kT$g8$}otx1Aq0&
zg1@qWHLy1spsN7;<<n^QT}`Jprghn!Z;%<$v~b?~!wH8DSWY-IP}g%NzDSIFY3mUg
zsgawC$=wh=t4Ba-&S!O0Z`1D|9j!=L-qpgn5I*rYydv+oI0xr_=SqW-O>fuF+*A47
zBHx*61nmI(EWbB?mcAWU2@1swnM&tISediOFcoLKIoFs=Wed2*&E9`#a};D+b*NN!
zRhPc#+1J;iU1aoLl8%<n8gmY%=g?qdvUE!m@lC4uvXsHlsb_uZHSII8ivDqXVw5o0
z1@600B-fqokGI|Te>5UKT@*^MiiXa^()vD3^^iZbWnkY&Xk5;Jz~LaJnvC}@1kU-(
zXrV5Xdp#bg(l18YS`Y-)kfB(+*qx3QIw1&HoK#pPXFMkBElZ5=w6`J3L5J;vD4sw#
z(@1%z$co2pAXFXfVvVk_b5Q|>;wY~Dps9ed@=&@-ASdVCn5UiifOi6ck3PE8#&n52
zR)tdN#7T}p;#)`3>2C>AfbLJ6fDR|bVO?B}*!FvVQTj(6DAqp#D@t86#GL@SUOkp2
zrE<-wo(g-dQXyDw370SX&bI~IiVsn-Nw^likSoxU;vQ~j1$3D@<-Z&8OB(C>6uv!O
zud32HAFV6K=_7bx@HxX*Dmb*4*(U{>v-Ku=juf?Tfl}70-j4KOPbaNOFJ|3ZM7;^O
zpn7A3+qlzx$J@g#tIBm-vG3JLr_a+mJAwrMb+pwl8-{D~NRKSxi?Y>-5>vfm7{%L$
zM^p8UfJd!7UxcJ;<2XfYNt;Fm`q_v|Sjgg}wt!SF;XvA2WnfmnNmWZ>|5_5ASiB75
zj{LX8N1*cP^#aJ-q1L{p+je&f{Gm=&BA2EH1OQ%KY1D}?k!Y8eig6!O*GCxw4FVD7
z>*jNd8O}8lRGSyWHIi@8=sx)bR*c-}A*E6Id>3}fJLKMW+GOZzDq>lAZ9a<;J}0{{
zlu~?uW0s5JkmUhAZCaFF8&vY`MCeTdyi9op@O!eLG06XCMHaXdOzL1(4K*SjYb8~S
zzb^u@h|ghYdf*Y&EqWus6*6m@CaR^btefAA_C(lrKGkNsK%R9|&PWp8K!1yDe?wMa
zZYFo@N^8oIbG5c;)L}{KWYYE69CFjsk3%`7t5heJMR5yutlHM*+qh!)YsHXu7@SD@
zQ1ds1`=)_`QpLt*>(BNiDC#=m>DSBAeO7<ECm+z*Jdt+TQfL-FkFz|VjHL^u;0dJE
z9EG(0lV_oQXB9ppqWJ44RgPwv#+ETV0;j2l1fr!m_*PkY{I3MbH32G!CpZxa+8o)u
zOAcgE6h!r1HH@3*MDm8-rIKMJ`+=ZMjxQsx73NJN2Yp@laDQ-|ayo7^=}}crH+K*f
zZL*6lj3Ngiad{0s!D55k<$?8!2iNTMf)Wj}+gDz6n~>KH*>Vz9ZC#34J3ET@&k?zv
z(R3v*n7(e!Mf`CN<tPYi*q|wzH06}Jb2l(Bk!LLOaJJKD$))?1b#U_A2|QDhUgW}n
zq|z{~T8MtTJl`Sh8wxX`?KG+7bCXqRiqhzu)k3;gmo{!j8|IBmwCL3P@u&j?=BEp)
zGSQ=-zIKqPU8Pu->B1atQD}#Vk#V3#YK-${#9O`NhP+k5+Qn50Y;AgZ#aKU@GlL(#
z^A=nBVpHWJpgdI_VdyC)uc$(j@Bs6x?`NfeMaWR~fMj*XsQ^k_;kD@)0c3@D+u|?m
z--8{?jh~#vq}{-2X>WTB6`VIr?os>I;2O1=+SR+j{nyjnLoPa)mAUM%a^+!YX*cXu
zB5*|1Z0r6M(o7#zAnpBzXrE4Sv${u`V<#U_8GbN{{TX@T$5r^*9|{+$yDd#CH{Lf8
zug}AZIqFkka$z&Jzt*3yXObBJGPQQqrsbub86Rhzh|~w=d(j>{Uk`xk-c_^fl2Q7^
z^RX1=swE5z{IlG@W0_L@IiU2Gh8pt@H57aG0LKcEyI_%OJ%h^tj{iwq5GilwmotXx
zFyOD|L$MqrKk@eW%xd#m5!@R?JtHvW41(kTg}H3mm!9d=ay6dXXw5E&IcBLWVN?7l
z<7`ywCB+CiX(U>-Ta!jGIQaZ5p%j~HJ){z2H!_HaR!zYmZ^g``14!(Qzb)w=*M4Pl
z@gSRAHsswrgIkMYg=p%;nzXtTmZHYMJ6d!UXcdU0)nFba@Dc;lKIvYI4FF)COziw!
zNvLh)lqv4xLQv%i$3w{rIPkY8Loj?}8l=#E$usCnAAd&R^c>b^q%H#()k#z_n_4@A
z!V~o0&N%sdXpiH{)i<R`;YJe(n_!<?x4`%&PuN$7sYH|u*Ur1ivCuZ!m#xE;a?J;|
zl%9!ceDr!DxSB?wcQ1_!7EINUCNt~I``ovY#`$|A+PG|!^eS&fFnO!~n8`^>0$2A*
z1bg?+zfK`Kt!(W-1C!u@d?VwdO{pO3Zo-7@g6yMDF|l6**e4AC1L;5Gcrb2oUc6|>
zoSh^x61E6u%EyQ?K2ungu@EZ*`(1@l57%}&Eqb*wj}*V^EPt`3=Fv*ps(8Gz{<e0z
zuW3lmWh+A`Ctjr}RJX@#=ypikF;Wnvwy|fLv0@S|ny+rOqu?Pu$fFRZK$Aep8qIp}
zyIxEQqa`?&0OP3*_f(qD=LO1bv7jCvD3=N+n*y_6JD+Vb@HE8d-sn{oX$Iga7D0a~
zJ2dj3)nAom^`5$XMR?-eV-A_g=(=0e6eLI9#S0sAdL(no``T)q57MIzx(E$YUU8#D
zV3?t0xKm&Hf!83Cadh);J^YCOOE|G703|pUKBQ?pK|CnrwAj#DHl6RXJcjSI)vCB>
zOyAp)j}O!s5}E!6eAL@uT8g*G%}MXR?Uo}J$9&Z2Z<~M{>@k*fg;U1G0JlJZjo%-;
z(NjkIiG+;u1R6`g=JVg&TZ#T^>MSO({5O?Z_VNcROve+{Qp<|-+dU;tI_C31OjYzQ
zL5v<?i+~nom?+gqqn>OIO_9guh-zlFqsW%hdNHfoqy?leU%!8(X6UV=swG2~V6l3P
zO)yA}0=uc3zf-L)qXjKb>tY-Ny4aJdz&FyP{3jzKd#++daps-H^nO&hoBo`F#w^sA
z<+0N!hV?9X7@aU!kc|9oi9!Ojh{6(y;ZD#MFj+`d;d#odFNyxVdyuuwnDt&M_^TO~
zz*ZHv8UBSm>a~A14#w3V+|B*m^ALnKql1qI;xRN!MRpy9!yRUT^*ss|5)|88n%#Zk
z*fXWzd#6SINtLI2s8MHlNd+MKU_$8+dtF2RZ<ULuUo)fj17bq{L)Wh5o=Qgk*=`WS
z*eR_zMmf|ezl)yQaY_CUgrdE%3fse5du#b<-7K82%{Lnc?X*_;873O2{I&`5XEx=G
z{)&DJh#V!{EMOnW!_qhi_voQfTh5y#>?S8&U|*k*dZQxdx-(Tb@h%x{j&f-CD<HJW
z`2PZv5xDPf1o5^=`7O_}c`*|`w(6SI%fz#qngZzPQo%v2WKX(ZUc@c|nmkSl|6=sF
z2xhRg(S2ySC~CxA%=>q1<g`)7=}k5ZAt5s$e+)gV-5LGG>g++a6h4m;dRl*jQxZB>
zKqE8{za8AIads(0dTD&d%;`<1?Vx*8V&Uka{+qY@0RF>kWLT*-q>aP8w%XS<p{pjk
zA?dQAYJK_1B@QC9T_Ex7;!A^=EJ6fK%EhDXbV0GhkhKrc33-}_A~lkM(Z-slzpXFd
zXbKk4YD-F9W`KQlReVzS4$o1e83UWRxqbDM<#*JHsL${pTmPW<6Mg&!Vg3&G*Y$(T
z*m6S;?YM;`gIz4w$}sDB<E@x!nW|ysGP*%Xsx*MxugQ`_ST?THPTAvH(EZlpWn^a9
z(7l%(1f<L1YWT1BJc!fb0Moi4ArPFI7;d{I_);UWIv<tFw$O-vVYE?S=7qtts!*KU
z4JEsQ*MvN(vU^6s`*|8z$$if507I>^AZsw~)!&O8`_(j|zw%XAhNjwtnkPhV!uzJH
z#0ZVZ`P`Bfly_i6I_#XThl$N8<#zTb;}g+8-vtZyzPPear$fUNDOlFmDY|mouEWi!
zq8XSh0x-Dsauy~J#ZEY<%gE|8n36c2<a&@-+Wd1p#S=1(J;adDmNiX@OrlrZaaU-+
zhA^C!(chAb#=AUT9K=Mr+5hQCwZA*=#p-_yCl%Jj@2l~Lw_7@_#IuITeZieVx_D6h
zZ!@LDIt*l$S+tA-uST(d+qRH-;B>04vK4&b-fc}l&=|W@&vQL#9S&j&L>tJc%wB1I
zQYbzaYV)-)J4rmj8j)WUJB$>FHM7bM3(T)n!09W;O6GowDjn+ay3HvvMnkYf$A#`$
zdD_VDDD&^GX@YU|62Mm}3Hb;J&!HRASZ5KQO-QNumu}%_<h>n^3j)5adE_8$DokQp
zgb_$f8+&hl=BNVaK^V#d8MD6L|Mo!6X(6=P^l9w;p3XG$U(DBQ){zdG#rzN5ZKQ2n
zOm9$g#d-eYk_sC?ffGn)wteCR`)N)=DynfTE3&{oh6hwMNH*gzmsqtwxuX6H+)8FR
z32GFw+P?gKr}Zd+!oidgb7Ebw9xtZf4~8nZMf|La4^g@)`SOT*mgbu06BZ*V+;g_t
z@^T9ZJGUF-Q~sFyGOb6=E#p*KTX8bFmK0q@lOip|ep)48C|m{T3qW&=UZzTQg}$V-
zE5eVLkBib@y#xkI8J(Asl5iceJWw!^w>~1FEGx#RBb{?`hgC}f(e3@kRU8#J_&Smg
zSJ_ZA8rM<TEoWis@lEVMkf~4e1Q@I=PRXGrwz+HfffEMx^Uc{5t{~>~`i<x>@Cz3h
zdQ0pv#z8e4$cU?8E?uq{0mFH71p`1KqJNQnf7avKJ>M6Fl!0*}HNs!^0I3Nm1oBPQ
zQfN&eX4c`@RR7Y#5Cl%jn+ynu4hps9cp^Dy6R$F7sti-(?eH1#(b1Gl<r1?54R6?>
zpEh!<@uqp3`=>o^a#h}bGNOjzg^(T=1x9>1x0Gds9Muk)1F%pkZ+eEtna3k6x=79?
z#SdDv+2%4V5x+w&LE#Zii*jhH%r_ZeQ;MiYM<|X~yceitfgmvyXvyxOR`Qw^i3KF?
zq~iy~Xtr~BRg!0^GrTlY%vf+ksNSXhKuXWsy6i{jDEdWNNlP3+oS)a0erSbmt%z@B
za-w3?mtp$f;29hc$UH5^MX3Ag*d-yv(WMZ&JXSM|tba@W@>2X1tnciri`gRIcOYM0
zxMrQZ36END+z@oi(Po_5!ac$RL!us<ISk#7cuy7G$U5_jKM}u6jTbs7S0YZXxR%>1
z4RdjNBRXcHuZ9mv&c9+^&H%01tZs=IQ_Jm38C)Jx7Rbv{NWU!`YRx>Ct8@)}Z0x!H
zC`~8+7^jw?-**dX7@+fN{sy(C1gTSAhEDadXg^z+iAT<DREEGC((61EYULauqFqo!
z)Eu+JBx}t4VksitYgsTrF)o)OuvVZG=3=?X1&ATrgR_D@qHMXD-^``1p--E9;NgRj
zx<E{t5IoX<<gYBUedxqKGATDm)$5-i`ASkR2)+ZY5ewRTDv>QSoT?=DoB=^^(f-!t
zK!}^mZ_LoT(v&jPHjy9}dd4*^`XH5#C5J20yb{C|!u902wmuC&x2Ogf3o^jZDs>4O
zrr#oeJFbXk3oC(sPTaaT;M~yVM(3nV+Cgq8S3Vr;a`8?Rjj`tv5)mO773q3VFL`bJ
z2YPRWxJ@wnQ%gn+`VEl4LGzC9FOmM`rrLWcjl$NS@UnNSt4`C-Oi8z&x3vMXcC*^R
zkulm~e+5nhg3rN?nde9L81T&iU~2_mFyj*sW$EJ|)Ro?^7g9tYNGZw^Da+<e!{1xN
zO>xUe4IeWGpKCRxey=iUh$OUnd~KjbT56v%h|N7?zQc4gJm(q(@(sPo+wfr}!AV3+
zB)osUv_Dw@J1eRO3s&YoZydYVe_BrE^G5bPRlzTVa0{#F-`doCm1__~ylcbklhZ4&
zRU8~|Sw42SXuFuBGXH){_VWmcI(j&u;SeCP@)0!`NVUMz;)SPh%1joqell=g{?VsS
z{~B*MntG)p1C)~pEU$A#iA8QgJG+5T<JIeGlQokmIn(U<rR|hof(NABtu~qE{W=B0
zv_h)B!A3kXYcT6Ek!Y*>v=qs#V$*?%l&kiZ>{e%NI7;Z5@Ro4tL~wWMz7MS_b`Jen
zDSf7(@wPckH9z%IBmx3=!!FRqyK6T)jZF^gi$@?`Qsh^o1oG1BZd_X}tb??L!lSPZ
zr|_G(uvY~+y_L2xct`tqj&(j*Co4~c?*)x(pl9BZw&)8QRd^3P=VLon%u+nPp8pB2
zoBz^a_vd?-vSwGd)(sM@li{q`xwcAb$4BG%O_Ec`fhUqNOyeh+5i<DXktP(x9BQcv
zT`!!0iPCf>S*3L;ZcheaBzl>RIjVA~mHs|R`)w0^C{l3W>j!^*8ut3Qg{54Psn-}9
z=)?w-I<gYdFH!1UH12KgoYtgb1ml{7h8VfDO!hDU3nLYn)~E-Bgi^}!dC^}i1X8nD
z=}Qt-DWI=teId@u`yM$kz6{kl(eRN8ZL}wAk6?<y>S?Oh8i7(C`fFY|>xT89&Mp&S
zUp>$vN2DZ+O<}|8Yf-~+AOJFShBx)rq8m0*A2q)Ho_k;aBbDX7JZa6sr$SAiq%}fz
zN2p||w_Emk3Y|jLC6ftV#nW-}IGs@rM=I)$!$-*j0(4?d;>$7j=RJf#RkGgTq`BEv
zV5qJZyV~)2LWTz{U&)UIebeGceo1<d^AJ$3#j}yA1xF5XE`m<w^qR&xRX9+isM{uC
z^t8#$T*$VNRYKQLabNFYQevp>re^=B=b~NOq<+fmog*z7hh-U|@OE6-?sKAE$MgFe
zf(Hxvay^h|vm1fh63V+abJG6Eo$)(kd5N2*WC3k(W`G?ie(|qAB$ZpFB|vvIA<=Lf
zq%t?|x0aL#{gR_K09Dskh@{E=2oS?y=zJO5DY56bCDui!s(t#48e-0>&L$9aUyZbq
zzw}oDi(HS4xm57S;?;N>Vno^9A7$FZWW{(kGNf?ke;eRDfYt&~+yA#K1UA7`AwV)x
zOsj?(28rT=m4xF_7ZR<-Nb>x6bc(wjRocx^$B+1eOs_);oAXrEoE^ac<Ez@RWb$L&
zF#6H`EfY`v?q)elOjlSfuq5V$!?b@X)p;qc*%VbM{;49)Qtim6iiKm!Mrdfab6~Hl
zoJ9+M954F@Y^BCFRzIe#zZ#Rk&<%=GR%#L(GhkV1dgUmKT&>)8Y6*_P#!Qov<PqWt
zPBH@Ytg!)Jft&}f(T!4g8*VhHHo++Ep6O}jcs&(UoO}vB7ULmsxmpUugY)y!H^5>t
zknzd!{v+X-K0CPG-Lcz~hFr{EBCtk|HvT>4r^$1D+X+;4BNJkD{smh<o0bw58m<QQ
z&&hE0Y$!E3%35`AMsY2%WH<)f38mHsPl3NF)HCY<*5=d6=o(Nhm*lZ2i<x>-d|Ao)
z`tw*`>mSv$J3<@Z8`E<qE5>AJQZ3fLWDGBKoqi#N({KEINS_|0Cr~}3?-e0bT|<qT
z{-44G5-(9hJ!@1Y_Acfb*uh1K&to%AA=o6X65sxOqi(xn*M06pI_%e{{VvYL$en=W
zIi{&N`_n!2$1BnFtXjjpv`5yTfREXPSMa*fcP!u>VHpM;uV&$1n0ShOEFA4h8e$R7
zG7cFQ=#-G|kSzVzNux_{-r|wqHehguW9?m;P><4i0qQ92z?}iq#r_<Uc>5yMkvIST
z%a(vdOK!y|w}wi2VVE8A$D^*^tozAXuh#1VIqwvT9VGOTJxuYcug_j&ij!>w?~yti
zP@bXL01y1#J(^C`W#_|Xxw&vVgJ{py1YT|?%&KE1rSjo+?&I#mA`fYrR!&x>i+w(w
zn~p|(wiq)p`umyr9z9<y(I2F8D9JL+t`DJu>PFH4rK63;J#AO2`<qV%mse{RzIBC9
zV6~#dz}nC~;+qDJH_BG0=l36b@#+BX=eKZNB0;*chvAC?bH6#}j{dE%eQN`hxt)8O
zVd9!MC9ej)GYAfLJ{oHLJs13*Rl}cc4W*yn!>2F&XL9Gb?wzL7RUTfCA2o%?#sfXD
zjEcK7%DWO+VY1`H7$uUyrL44jT-vX`@jo>F9x(P@ohe`>R1`rv9Vq!Qp<NDnD*J4X
z_C*ABv%^v``X{>b@R~+&m|a&5etR&{&S*T*C4-*ckkVA=E?$gUk$OP1+wFVzyx`FO
zmQQ;BS7Goe`O>|4^S2jTkUr<eMT9{zis9vpj2%kr(H&mvJgV?%KuYWX<F#KNy6-Wo
z>x(vWy)$S7Dv&!+@BDktOQ0faNOq$(?o85vO8Is-%p-=0USXwLlKy_j<oszZVxyCp
zF~(EvtQuA&tFM`gT-(I3537aR`X~W<Z*MO_b)XzjLz6bBBDbXcdycuCr^|`Rn<}dn
zKHQ6D126^J>m(K-o-c+ASuEbrno{F@S@Ee(y*C6j{tP_ZDU*=XUk|&)dausw$DU29
z-TloUbt6B1kP#*_*B=&y?Upg06qQ*XZN*Oj4YDUkW=vVSR9UJEL}OkJ?2q~^;eI{G
zJL^+UFJO}0w48v~udnb1|KD)akmfKjqRP*b=Ju4!v0aEQy^KHaWxXe-7m{zv`6Z}r
zY409VNK{j`oeZW?=0wcJP~oMdkHq>fT5OjVao83z^%W-M8qJ}pPR6Ubj<;g><de^8
zXgS%zonD`0By0_g4l_Eg)VU<GF|B;!kRFKUAPqNb<V`#JsYDj=6e3LPwvD36m6sVu
zLKY(aR*}Ec!S7eJUstmT!s)kWZ~Sd?V`^vhP`w4wm7?qEq!AaCnD8Gey8CVH&6q{Q
zI>A_9b&Ayq&B(;}o<A^q?(~W9Dn`Ane#qiMwTZ$bnUTeJe9iuXHR0$sgy_$&=Bf`d
zA?p4tY}Ppo9k+3LFts1`o>x=)uK*S!qX#ST=qXdXpr8-d!;s#mNcOkt(ZC8_sIP_u
zJh>0QqOla-%Uw?99Sllkr-Ps?Vt$|@tfhgunQ-Ov3ZTFH|Cnf16V%MVXSyw^gHLH^
z@((0H=e;RVAV{nDr%3U7HHgo!tP%=D<v9!zZe-d`*v(ZKNyA*$s969T!n^&OMuySO
z+gc86A)3GE`vOe$GIp{*p)`6LvdX;COITk8s-Uw&%b?Wn22@{Hi-v2-!niV4VyUhj
zIYiv`qB**1=<yz0{g>a<Dv*5_SCrXH(%6yVUD;q|@a$^qlM4QryO1t=%L}7{YcT?y
zCGF!`8R#j9aQ^z}p8a95`X0K&Pk)xw>CU3<H6|MtU~8?*As<K>M0@sefl&FCef1vj
z-so~5x`Fo6i?f*p*i+PU#JThd6Y3y9g;F4dOU1}7=ed6*Yw3avB&a_&qcA{Q4PdE7
zRSn^0V}3UMDNhl>g)GA+14^%tFL!M@)(CHH3K%*ar4C>HiF`-&RkNmU1I}MCjEtsL
zbK4-*JiPg^Q--5}n!rc$qyC%HhYQMMW@{lw6dAe8Z9R)0i^~g@ee%Ko+RUgZb6@-S
zS7*K!*$#IjQxf|3Y<|dQ>=+;eMXKo_sX!~2?hVdO*iD6C)p2~?k>>=^NxDCO^egJd
zpGx`Et<05_sW}f|zD`!}7rcsoZEd@GgWR7h^dHE}8eJe#JV#y(aAOo5&n!IJ;Zh`T
zec1iU166%?8)Ig#gXmN;&1#S{*J0Y=kFjZwXwp~7O0aNBR2(|kXyFpF@C&QrR>%6(
zbTjP&7MKSm)N=DoBQ~2Gvt)gxJ95MjBy-ip2oOj15h^LMaH6M*bOC;8ZUomAqI((a
zv?+l1ZA2@n8?t+Ey-h(Jj4N+|30z4PNe%B(Kr?zKJ1<^!0rg51_6Y4$O=%O^oG7B>
zp1cSD{+<wQZ`_}6KzX!?TIW)mV?8@joC5>!`R2Y*Q-kaT0pg<<U%xM&)g4xxh(h-t
zNYbjDC3^xow57AUa(eY^Mr`y(iV24p`iNzKu5)AcQbxQr6se>%^}TlJlML*u)@9*-
zL8tlFC;6Y5gKVGskU2!2Q|5!q&PTJ@b&_TtFby4Z7a&Gv^<;Qx-69)hRR!5xr2pkG
zZ@`vPf8!P%Qh}nL%Pq-F2@K%JMX?#X$6JN?rGFmOI$kft9f7O8_6vj|1A?m}+ZuT(
z{@>qzBnH0=^12af**ajKzgml>haxHo&vy9zVA6Qh`RYB^BNa}pU_g=W8-fVRcmMM1
z4c8DS_z5;6lVg<q(r9q#PfJ_AKcjb=<A!T<eztyfLNFeZ-P-UqYtZjU7Pso_niZ~m
zUy+`i-}O5BWNhjWlU<H}$Y(UXqP?{IhawtXC^T-oI5MUAn4z-bS&Cb-1daM`ZUUcg
zC7ju^FP<9JUoAp7M3-|a5}sWbo!#=*e4fe2Q~&R{jMjF|he*%LD4z;<vwp~LCM2L)
zN>=g&!ry$^X7;(j6DvMN<y65Bq_0RdcNZcSo?&LBUl=o=GN=nkbg>)Xs%R1h)h@4Q
zT_R4?2{He%=EDPYBm!!r2FQANBV=Vhz-pd_13mdZfXQbjFWVIFOZ>-m354C@YBKUM
zC>QXRUZ@+qvpv-XJ8MqsWsMfWsA~1Of#}pdJ`f`y#k@Ke?PKSYvikbI65b46u8SA-
zoD;NssW&*0$}C?kodgA~NTGH!4_iKK>*sPw!Z9$3dLC$5KAbm%e=q)kHMh)j7FI(W
zkDW~@BYZwyC<u|CW`bJ1s?#<3ZOu#?ibsTeSJH=hPBYE7QYyOs9yADeE@#mBGdvkH
zNO)5pDdbtNP8aeDxsbiD8|aTj)t{&B23qHbN%;T1vQXAvmez23;+)c)KlSE^Vx7kH
zYrmXV*f^VXyczWLSH(Jmh$&X6;pgm!#hN=u^O_&PIlnMLGkO#Mfs!i?wV&R2l5k`Q
znQidL7T$Z=r*V_2uj)gCAy4CyA<F=ib#tfJJ0X7vOOa=LZlI5@$r7%W43nGibKqZ*
z^pZ0I{5!rD#a@3bU<$>^gNUqWTTnHdL;|xc9en}=5C$()uA-WNfG83dEYxfPDwfo|
z#vo0UQPMu3={nkD6zJTA(SJ62J<aRE+*K{S9Fw5sqFDOXvhG28x;#VV%$|Y7|M<Fw
zWPpUSd)c4af8%Deri!PlUR!(qxjk8xeIHT!<K49-4+9^K#ah=-W(>M;)g<P>$i-6s
za1E2<(clURm}7Sz9JA`YBI9AC3?h;w^U`h?s?5JkoQ^j?3yRflM*NL5xFzaEIJZLV
z6@K^nJ!i^lc&A#sn^UVM2FQjc2<=omK7v{zWeg2(bH2#G+1;l^z65b3${k-QHAlDm
zRhaaOedrHOz>VfHDcv%cJGADuO|RGitIaQkCl%Ew6<Niu21KYGTRM7MVDl{G8J@7R
zIkjrI51LSAD4o`!&gbGQadJiM%-H6M6VKBcswD*kuPI|H%68jyPL}W_tYj!a)t?D)
zS;X!!&!;E2%*9hgBmEzbl=(Fx_br0w4R-igY^aEn4LT~QedlUS@<Ud>92{_B$dAne
zm#`i9dO3XU0c&_Bb_RxDnlNqb37zz#YU(zRl@zLq+~}0K>bwzUp10ss!E)`=bcpif
z;a(Th?Z|~X%97|Y2reP=jSE!>L%hUO4U5b!)b$4kSJGSdZSuxx?FWifTy-xJ3h}ax
z>lk{GVt&Kx5A(Dt{)S%Z%9{gq3&^XWLy8cfKT8Qkg{`>S?@IC&ipeO7=R_?es6Gnx
zy{v`bH{qBxb9DAAsF1jj=%xsWSWR0$%QUv<2a3`)x!_J7@MYeAD-SvmL3g<lHXM@`
z*`3a=or^5mp332VW=IefV<q*3z5zLY&t~>^YJ>1Yn<~CAbGg+V_I%kxUu!&J(8+K)
zgVZZQEk>IMYzr>-JNtqXhdYIkx8E(L|EZ`>Q5yxFebt&C(%i$*vkCe1GWBbxgumxX
z(RIkbg1^1qNY_8;8EyqATpB_E5&A77aofzTTP@fQ+Mj4AHP+t0jRnuon;q)bHCX&P
zZw@K^L6(f+S>Mz}D66vs;2n;M%+?P<vmD-@11!szt??aWmD%RVG|DT<VxdeZ|H-I=
z`cpUu@ArFCoz!w?{!uFo({6}0HtgMPw~j=fK4D#seGB<i{M1WbfffH+S>>?sQYY#G
z=R}J=oools8b9g*0cY@QvU6u{Ro$nRM!gO!ea?;iU2Pim#z0If&$9}JAbMI(Nqmvq
z1;V_vZpEMTa>B`Q&iQ)}N$;LeZd~>NF2j*iVR<;osI|>F@K>G*ohVI4pvrB%lTJ;Q
zbeq<>Yrazt<Rbc-dTziROS4x3ah>L#HN+qTpapPAz#y=2g~fA91W{Jpt8^smDr=kc
z&XMs%P@Lvh4Abi$VGk!kuS0X<rJnLu7R5a23a7h2a)G7lYw+<gZMtf3G6@!ILbrsg
z)XI&psvx?=zgP_w@HDBlZRu$~n#RDc)<{8*xo=`*s$ZAfoIYgs-+4#BF`Ky%ufYE;
zRuT!#XF~TK>hkiLBoV35UfF`}<&S9Ixu3ND6AVb#b_1)iN<hD(-a{zi{Z*ijvkBpB
zs?)L!emN~>H?WIur_M%c0KEYZxYw9%9GFxu<Y6hESBIx<_QNUV@n=83U_gxPS)4RQ
zQ43k1>Mv+IHEX_F<9u2}1-`pePWcWLIaA<G@OVhYr|b?HLtT{wggCFN3Q{_Bd?L%3
zS_xpKVa>@WzO0LX8F#)oi<dugD$pr;m6)h{T`b{=Q%<swfCC`|%LLF=T>9Yu?pD|v
z+zAi<xT7?zyU^?=U?9cl=X5y~Dr)o1C}|3U)8GAo{xk0$+Lhr)tq<*2NA9ZwU|iI|
zVnp>2>SH`aAbME-KTupnVT$#js{C8JAtcT5pbTj4$<Hzx`DtDwYDB&#r5@pR5cI*3
z{QM}`Tiv3TQc@Sd{V{^xxt<yy(z~=P6_MdMvi`O6w#8c>O|N!?jLS1T^*<GQDu3x*
zaxtspkbk}MkK>>GJG-Xcm#sKX8R~>S8eFP<D%EpeTS2neWZK)IIfZ<;nGY99FH0;w
zyBy>n8wp2Y4lGvV%)G^d%20<n!L*@-e}~q^7MWjDKp9H)+fuoc&!6QyGnduq%-<b5
z|7gG(kkhS{`H^3H&uIF%$U_@|SO*Li7cS>*C|qTsxh@nTzjnajt(N5YG^5nK#48?t
zLTUL!5q?BCEBj+9EHSl(riG*?pit+);D6d`HJ<$&_QVK~%!U4%j;&`tQvB;5nQ44#
z8U(GxQNIBSyjG)Xp|L-xAHQ8m6U20yKaI<~6fkr9U4WN*{8!sCpZyBr$N7=Z($7kN
zJbjSI7+Y_a)q41o2Imj!89h)kY&4qx@y}K*(jFtWl#cxm6q!1smJ^dbT+@J;2XK2%
zCDJ1L-b1=~>GLK;jNj<iE@_3sjaeKP)#;>)>O^f{Sdy7b6uSjWA?LY4!*w-!go%48
zuvu(oj-U7^uAQQf3D-t`QMLaP2q3-fdz+i3v|cU(?oj}14T;x%LCabw%g0**G$ZcN
z7qw7Gp-2s-l%sFpSHdV6`uUPbM&5*V_&sjxZP%VXo0i+JPBPw9sODlGCGFV@Z@}&F
z2)}_5=#+xD%0kf7yFco4YU^tdwuq*oFqBdJHb!4-JVR0|z0jlb`Xn8i!vd+lh9~7w
z=c~<Asdh(Eqtg1I8`&{)Z5b+-vJ`_s?A^NBOb{+zxKeRj4SCj(_5@6m6<r4lSjp6h
z{%WV|lW0g!pEyuWqwkM~HGRoH4+$vmmMd6b{9{}XiU;0r8qmlo`WNMG<LaYFLSu=&
ziYG-=$fW5^;~91d9RuYH<Yxw*kbC6}k#9T!O0JI%E2~irY6h-w$6@twmNfUc?^*Pr
z%@bKe=ZgLJY;K$iFgW~=b@OJg(UnK<7~pr-A{v!H&dtpDkep>$2C|h#)B^(#oC^vf
zMMw@n3tHZekYUXHXFereYgbb;e}YqhJrdGAC{^I@H{?eJp@Z@x{b->TS#NNlsX6=0
z+TtPe{r4tAul2MFh?vHSwUl%G$_T!d%VGBTseze@or{N)`bvs*w^e@M+Io)Al--9A
zRG(O_qvEm~)kqcHZl8M>o;UHAclw<InsZA!+Dlt<rR(*#XE(*6henUN`l-JO^x;lp
zl^=vWOd=Ewl{cAS*zy}Rw3}G0oVLvfS+zOS)=d@PKVu9oMGACQz(hAiZ+8P{UD%r&
zgez+s;$qFBawRW4e<W6|&3o?tFDaA>XZ~4_;7bDo=>WujDQZ-Wcx*4ft2Vy2e3-c{
zR+thd@VS&<#SYKx`79c;0bfsm)<60BW+zlS$9!(<<k}~G$XW~)+HX7>eLs>*6!i3k
z00<)dSV_4GE;CxDH$W_E3yyk@k8040d~W&><~h1|+B|I1!>QwOpJ!h(2Z4UGz?`o_
zR3S>n8H{gq?*?{u<@u&u!Ed>EVF)OJ6{Q-Ks-&?;7i!+QdgA$mUTXxQSn}WA>rX&U
zH+P_gYr^v56<CUGu=8uKvGt%Qm%M~UJ-=AC@jZku+;*5}7$z)Kf_Sd9;}EhhABh(^
z@@)f2Y~eYdv8Ck$UpFx2`JNn8T8?EG(|BiA+g?VkLQzCLw2gEq<g!8Y)$Y*Gqmb}%
z(fOH&L3h@J%QLazKm3wR4c5a5tfh~A{BXQ?n+o0=M30Iz3f|Fng}Fh1Yw3^#RpqSG
z*@XVM*A8Hc$-B2*y(&Ft)K7S|cKWkQ_Icug>!joJoK=z%zj75yt~68&z?sY&;qO7$
ztE_NZ=&~-Mi1LS9S{#~P*z||Dj6M1A(;Hh;)bZ>B@w0OPb*YsYB58T=3#JeIjh+tO
zg!P5Q=PIaq0bMz*YyJZO!CnAa3pmU;y5t8Eh(IBR1yKlGlnF{d>KxoG;~TF%U}|1m
zaT4-&?(Q;Bf6M_Y1F_v2duBSCo$cDHGVAmAU9<;4Yni;+mhaX_VzK*k;?v~}?VYGY
zpq5fSj7}S8&;lAt?hd9ioDb>Dk19wU4@Ghu^V2`)3J=uNw+u?MqiksR&%uz3zY=;G
zV0Gj&NirU!=0l>R=F+}EE&{1E6yW~!g@7V<AP-5oeRo!~3<cjEPTS2tE?N@X-dd_B
zd*FQ$`CTo#)h+qSMqw*Fno-N|`DuCnA?7l6e3j1&OY`UgLd{l6=1ZQHmOFkfK^#$T
zta{p9Q*#VH@SG*DM(%ZglXSjO%c*(jyk}Dok*S^)(Gd3T>v%irKak&oxuJ17Bke!=
zm8w$gM7v&+yLx?%=GzC7K++}skkG%ZQZ)IT!mT~bVlGRMcRW2OFMl(hbLLSa@-A#g
zwvujIa3N|`1V=C&vPbU2h^q81MaukDzYmmPhFn8ZuT<5o=aNPqXAoXS_un%KI4!YQ
zqy*^jBOqlJmk$N%Za8DDYwF0J9sXU|NA+a}SxVVOKu6I;_JeWP;YHrlTyuL{L={Om
zA^fR$P%NE}nx3;}_tIYeggr8cyhvYN=K|N}*>n1l8kz1C`cQ<h>#yKZKZaRuOCrt(
zJ`(NLL=CIW6?H}k&E9U6gK9bvu9CC1+_ellRkQ0wO}VuCs0$#?u1Sy%)_#4ORx32a
zb%gBF)B65>7U&pt^m;vP0)~Cb^rL|_gq2m5=B6uI$MbeoSXIWW)W&j-<^FCA+cU|=
zjb=0Lr+lQ1A~3j@22RXUH*<QNI{jn{3;H+{3^ABd8rAKI%bS(Nr$YOion92{dQ6aC
zEH6h}xxs@)XPY}de7?%ZS{62dx2f9;Z6^_!Ovle{)9wlVY+PaC<q)7VPi^<k7tzR5
z5gbdFwMxKLdiFBwz4DPn!fj-vRIxU0C3EqqTSi0v$3qMaFn13Z>jYc1X}Y?jdFzB6
zTkqn?055spl`$?OaV{%S%i`yDo?@NCkCfIf5BECkbaJy_O}_iZ<VA~%l{>rXGS=?)
zsPrf0J$sgZ;zC1|#S5jwOc%TRhQM3RJ|a@Nx~HjH;FI7WrGBN1)%aW0nk_7=Y=AQ^
zFA^HXUrU<nyLFYpJd_w`TN2WJKZ8z2M>1P!hd<E7+@&M#e1bvrF0a~c=6>`f{Wi_4
z0>y-bv;qe?S|}e$=>eTsO*Z`a)%*9K!r440?gn&2s}XAtQ-+PD<O2T(pMZccvmjtP
zYOLiG@OA4t@l&Lz^i9WTIhUK`p~9n);$n?u<;M^&(M)W>w-tZx@r4if_`3OH8=Afl
zhJ!m~O~>Q!zoW>J806z$)Ao380;#)Kj|I4-UIaE<%pGgtpC(`6ROfqNNROyp9<BGT
z)VQ*uthuUYU%j_)x5D*2h3m03I{(UAxQ!SPT=IC#VXnQ6?#0e;vSj8)tO**oJ?^oL
zXSIOmthNf@E($m93qQI1Yk_cu7&~yMWK%pl;ED^e=mQjW5y9ml!y4<K?WFf6VfyC}
zvZ^pV^7}i7BN6I&;e`9=aaIX-cflNv$$C`89MZ2c1Mj5hn?dKJRw)4YTfWl2@4WDP
ztOU%)?JsurWatKyF8lAhfb4SLKmOtjV7$&BsxuNCzDe*JdipR)_8hznch0wGyL2_F
z%LQxw2tsI&XdCxg7iH@SAGz4?O8QDN5o<3^m!29Y?PG0w@pky#6VnGTEFXFD`sHc(
z-kNN1an<o);1lIFqxJ^py2slo2hFh-KTo9d`5ZQPpyzJBS4yUpMBc`sCPOXvs&Lk~
z7vH`pm?A80O!GErkAOR)zb_xeXA2~q*0e}yO}S9?pYog=(G;a3=sxmHNl*3OG27Q$
zNbMY4AC^sG-Tub|u15d(aa3G$pS8B=a0MPChxm!<N(%cZG^I1#)#>frFMY9X=wT2Y
z^gwd75?#XbGLTkVgw8;9_GH*epXL4`{0!jcti?f6l^tzhR|H?&1PQ*4l|(oJV2{Mr
z5^yJtD2Qs&nsIMJh_#|hZ$jvbUVpvLZw&Ozg|(+fv%HMKHM#}@VK#LW+%pR;oK&@@
zmKYWPSZz)Cep)qT{*Zu4`ije^ZMe4qDI*>a3TM~Lf4esopd+9XYLbBjjJkn>0fY*<
zDRpG6`o1Ij4o#6~tiCXzb*Cs*``31|Y)5Tb!_<2`?+nA^Nqf?~1Gf*;>yiNc2Y=9G
zC0f#%&R_g%Kt=qq=(23^v|-<u<vF|1qhj30-!NzEy3|Fe*07Z=o&~a`wch5F&Yg|G
z;!B;1ls-6c5Wal1wDNbS4eR+??~zO$0*p{v8}S}h{ossFv~C~Go{82&Ou}j*YH>a1
zwLF6pi#!PQ)s)~^;1T;{%>zJj=d?A}>>Yp5FYT4O>$%W=q1z%>l*$*zyY;;E2)807
zlHHV{Ge+65f9d1<QDsyl!s_f_UQ77;B{P9VhuxWn!H<!9rC+r36r#c<F9l9u)|Lb0
zUcK7^*OMCq@2%lQ(R98@{?>E}!4e^}0o-7_)%~^%{};+!SF4a#8M(#u%JoPF8!FQX
z+MQk@Ayi^Cptd>XUf^ky4<8UDf@_^9SJOo@7CO%5mnuB1w@_=OdGi}|Xyr9K{JS=8
zdWIeO_o`QxvAkJVKbmZrXZzA9d~oW{X4kHIJ34HPB6XyDM6Lsh%lBjmh_q?-o(!VT
zqOkkQ7nafPH2cx!e9@XXE=&5kExy|+yQ>)gcyLnlV~a)Mx{Zkxzyv|F{|$KKtdqpZ
znfaywRZ&9_vQ%Mm(ys>eW588cqK3*ui|z_g(xO(-VZA}~#SAt=uET#ApOjTn*wGMu
zmKi~E+VS7{q`sg3R(ew7kZDcOw5ZDPzL!nAmwZO9(vTbx14Zs6S@-q3_`a_OS-@K^
z9a+1wmNtpJsoPAwS@Dz@?BY1%-xZpUF~tkW6@8dLs&LSK^F=h9?KvBD`5tF;SP@Bt
zV_b~tPwgsj>!bN%6ok31r#<XcVszH@p+ekxdOkUbnF5pSR8mx(USzsic^v>Y3H$Dp
z11zjwdFqiwJrVeP>;!9=WW4W^iQ?j*tWha|CKoax65%+*Zuo}y9%e=#`Es=I@^l23
zq%-yZadehpO}>909vvc$j+Bx#7z|{<XvRQtbg0NicS;C?^c<rR29hI2DM*SaNH>g7
z5fDj7h_qn*pZoW`+3Ow0eIM6#f3NHFInTH;*dJvQ^^nNb7!c5#{0SpUNNk7hN6_j?
zn8NCI>SFu@Y%cfT#9NpLY{fGO8fc_i=p^pkoBp)v^1VVVb_yHv6|=HN&KHMV9^Blw
z8Sl2Dibi!ALwME#+GB^+n6be<i{T~g{f?!BYxXLkwAgoD3fgG52Nf$1Qt-?p^zbCZ
z@5i&3#rpo#JbrKgv2^_A$OT4SKgAUnj2UVuWX|WC5!5p%L74&1h6#8)z}|tPDLSZ_
ziZeLup{jH$r!2oEKgn)9n=~YHEH=Pk)t}NMQPR2I#b<TJ%V7E!@;Ta3$wY&!Koy}5
z0V}7lZU6sD-Asn*rkY;5Ba^I)YlT=)oCjC#_Ejulo%@*+%yq!i&OPx1=g|-4ffdk8
zL_#RyUAHyQqg!IW*TGvYu7<x0J}PEHbpzH_6+$WYW~Cf}2qS&uO=tFN!u<lX*kAL4
z-wvUsw^Ms);69w<0f(s#0h9HLQ2_u2()Pq+zWz)9i;CAqnbW;;gZEcLYhPF8_@+y5
z!TdZy1TuvAd34gbb!e${cqk4q$ILRZyy&xT?78OX>p78~H4wHd%`LAuXki#$LeK7T
zuwFzbcZ+iBdWxRaEo`OpQ0#bWJR<j&IdPoxPVkl;`cmP1OXpGB&&v3@@koct#4CE|
zZLILrptd__D%Q=<hw@TW>=?YQN6|%hE!q-28cGc~{c;im%M#n%e(w$*5M>wrKF9py
z(*eL`&|WMWL3ia+`rEL&R8(~9cc8Ok&9mWNU~c2;yC`dU^X&C~3)NM{H%Ca7wRwVc
z`kgA?nweVZBkm|!Ti?5%$c<qq;$kCCqi5|Dqjya&vYf8RtWbSj=$s$q|Amc1>Yi5x
z_63{I9Mxtdq$&+i$5QNu+SAqxnT=+-rCnCgMIBWLwcgP^Jzm%2614ae*hf(EO#Fi#
zlhxQ{q#CoPLe{th%^a252>^#)6DZ1#$@W*Vn8gGGTf(1!kK7*=j)B(VE7+b<kML@G
zPqvSkzzk0`3WAHXl+QNO^1hheDa4u4Zikb!;4-$HG=nXMmxTow;EgD%*x}9eL7E*C
z`t+w6fZ4@%nwNLS*ca%`rgrRL{S0+<_=}BZZ@BBoBH`yX<b$DlQI_PDi{R>2$LNsK
zg6n~rA6Vbc>Rpi<OC5D4aR#+bU<yFj4r&Qj?<fqTX(%u2V-err4Cw`kaqidXAVPm`
zigOxsfJGTDauydR*lCGns+2&&15C~2VU;4I*syB})T~bn<vo%g`9*++E2%S<3ob0E
z_+iZQwm%u_y=*@x+l*!9vr+7Oz~ea~f9--hb(%LHU6VR~sY-|<=ihS7+!8;lH-U9_
zUd}Qra!+Ws!*l()G?zxI7rbM5p6gKLAiz3Tm_K7kzpbCpy%NeJ^+3zEHMrtokMtK)
zQKGZFn?Mj&`@Ox9yYXAH_oO$_02DDLm%9+<l`OEhD=BjydFcsUdgF$)bIs&m=P|u(
zrV&T;sr6g#F26@iS7l2*arU>g8SM=QqyGmoEC7lANaO7h*nAVk9GdxwUM(&T)l;Ar
zF%@=3)XXOWw=wKg7bre@+6yvAksdW`c3jpBcZ|9PGK`|h7*15rD`7q2wkdOa=phYi
zwg3@<WEFrry2>n#{boP_8L!P+7?tz#Y9w6Y8Q`_u^7a9yqoTXzs)J{jVq@n?Vlerf
zt%Y2(Hu=Vd4zPght6Sa+hkL#0wwJr(cB~Cy_q<FFe+Zf#EAzyPd)H6$6euNscrS5T
z@-_zpjgM&Z>o`iWVFq*3ILPine#8u`4kyI0`<YT(o1BZawt7p+vdEiM2G9x`wr?)w
zSetM2reb`(2II^9w7ezwjn7nj3csZCI+sDM?A0q?-F<k#G3iC`okI1`^29SXlc8UA
zY}Z4M`j3#OM1fF_u((^0n}Ah?7!Q&mj^I`E-rSPb_46eRtGgUaFnu~lOB1W+CSj@!
z7sK!kHK4b#T3CJw0AOQ`i_WVHR}O)?sV&nTybN*zaTz3A=r4QgyfrRkFrwX8+B@Lm
z4WWZCQvyNzW<LN-LWg&dyi3qWNqk-{?{Xg}6Lc8!*jLz}p+Y2TEc7oB?FL2sU5Rk6
zH!MvL7?9%AdHh!Qs-S0|Y)dx!hCdxJ8w^W1xZf(lJ*G^S{SQ=VW9}!~*AoNIrmxQ(
zgtqZ}=N}m1o@~Wv)cY}~@L4be>0=-dhXZDUbfyd=k22G@+ZueKJOlhg`iPtVzK`o%
zKcJDO#le4-eb`h^<GVi`P~CGWrVP4h17~WuA7~PJWYo@)^C6ZpuK7QZ(u&G)zsQhP
z02k?6ug!H?nGI8c;)x6i_&fr#KystSg8MCwSp6A1r&F71<7QD*lw9Q@fwL-umi*_!
zImbkEH^j53Lz<_huZt@}>DQQIDUK)iE0Zh4H+~IO58P$BHt`yfTm=0pSwpH9qCPpT
zr4hU5{N{1YL3`|EwAa->pVKIqh@||8*5e54DN%iyElDW$lCof`JI>C79h&5RZiBm+
zTX`lBE*!UzFW!0}?cE03qS#QUTpt#o=`Lk-sOko>b2IlJ-8CT#tSu)(#rK*ymmIQ=
zk8AQLC9yi}h33P8^wKe3S8H3Z1l=2JM6kh4OGsI-$f`WAu_Mc>?g)o{ShZUly}ZLm
ziN<%8!#u|y<rY`k1oiw9>3VPZN31p*8FgN|n0<cgQB4V?F85z4j-G3_n}pi^gnzkP
z)kPdU%iZ!*e^Fb#Al<}C9Y-dhg`VaNtm+;x*NevL^`lk(t!wcvAdb3E8$!)k67tqF
z#{He$9-(WnjZ|tJ=0y<4BougXbF*A>YDykvI(oue-5~zO*B>U@oVXgv|3;D2G?P7~
z5q^7YxqP8mAu|l_HRT}S7Z|_P)322^(}Edir-u#Nhrv8^QJ_kL)+!jZ30YSLem_S7
z;7MT@WGb3#_K>Xkyg`|)dE%u7B4$X%GA0<fz5oD~(Lmz0UyN0nB_(FKTwn^qPi5K#
z2@^gYhtfx0&O(+P6k>BO%Sq~tBU$zC@`~Q^%rSwWXnGjh6lJyq&Po_?T$H;D>2WPk
zY%YZDu<}|7$wgF2@UO>Pd>k|pv4jM~0<bkKfPWSfr?GN@BfNB3oPNK*-#iCa6Oz?+
z9p(<mvb(7GpfA)9lRU0PCSvsXkt?y8rV*4I_3=_{2Ye(s<hybj?k;7xtbA;=34@Wv
z8P~^y67b0o$&w|~%s}do$B2Rv0~Pj>L@X;!mlUw(d+6WyC}H!giX=HKyh_WL^u40E
zZETMSA|M7b9?INa@MBbH|E8}iloN5fEz7Upp7+htA8zc&9W-Gtgs%4tytiZ!AU!O}
z{O)NJ+CU=vamC^>b>gm@$(=_kQFn|S9z(HwMj?Zy6`pmEot$4~sg`oOpx`okU_qHn
zf@u-d_`e>Y_DML!5NqWAFipeW!vci@BUCNyOjkTuQU~BQl4e1LQ*FH0ah<o+S_hvC
z_u43~c3ykEm@*rzECF;aGiOV^zm>1Y;yn((rN_=EinMkdQ5O+89PDYb-{dgu#G(3v
z+lS7|i;k1oHjIf>Unu(aw0#n|gK?0EWD9qUp-!=*87KnvM8)Y*vrUQXbs)b9Eb|Zj
zi9>?O5^Y~@ze!DZ(C~5j<!veU0@F}vxMxJ1CNqz|*4-E>a1u>}Ov2*Nekclx1&Is&
zaV98aydZgAndE=&T|&WcHfDSX#~0?VM_W~CbBt|gLEE&=dP70}`WiXtMSEaJuO47!
ztoz(-^w|uD0pit;kC(#X+lKs7Gy>Nq@>nCxXixXsUa8NC;r@JDIWp3TIQc`#RFDNl
zL;yMX7%j1w2<yb8N9(2=9sCYejl_{>s+S(E1QFizV*nShSScs|t!3UCtZsx$;<+*m
zns?cCt!=&pH1unDFe3dX0#0XCf^#KN2{Uk)Sb@Ij-v@Fcm_iG1M05*_2L}@Jb;tY)
zO)V9+fX7*~2#OjTVIV0M9r5RiD#Y&WSe856-|V@vyd_bG_MC0x*~qG{gWum8eX3K`
zQgNMBxSS+YKf6Zo^*4#~YVmeLcCj)kLQy{6C#x}Bp{DZ{!XVIbiB9%fg`=JXk7Ti@
zjs%B-lat$WPS)9q)ZolQ`tJSLof$qtQ*l|jG#ZZx+s9v8J6e*9GLGSsip{)}P#-%w
zKMqG8@7;L%i!8zyZla?S%QQi6(Ie!+Tg^Y7N^fE#h7|Vn;um;Nn>9l2y?bVGn!^9%
zgMKGpsAOK<$2+bY4?hUQyT@&9CB^sKgN{LFvaQPYx2<n|wU#KlSuV&OZRyxxzACIj
zW$c~^S;*<lKKvhubNK!Jx6WbK%o;MoVxsx(4HNyX$A3IKm6jtLp|Q!}hq1~hwMx2g
zF`)t|4#aP%F{WYqY7C$*bcgNqJ0_Z@N1Dlx8D}qjg)a3MkD9(6Bv)jozL3z6Ds*E8
z42DKqgb%KbI_#VX6+%-=Y*Bl)bneXou`{*VA~Y)zBoi|*cU?aFdQ+JuXq)Mzj*Vk-
z>%Py{22V3%MlyZ)Aos7C!j<3F$^8WD+}(fnGJ62@?N#knI_-yYpOqg}$TNxFt5$kC
z*yUj5ZvOdG$cK!JDSGSDKujq9#>!8o9c&etSxZa?9!W5A$Lb&P$91b^nYUgA4_{#E
z?+C1ui#v_n3G+)oBy49o{?xvjxsf@s#LVI!7)coLCNoYWhfsTkjl&Z8H^Q}%e<H1s
zQ)7_5yc0Up2cEO7d1{`Z^?<^8(s!NPh+iwQgcdK@v%oE?0f>E1{<yn7_bZD1?p27o
zNc%Yka9I9(vuCa6!osRmFYeTQ)&C0t>oL-eF&|RieC?X}_{;*lm^i=yuKQe>^Zfe1
zmh27ZaxN9e!@*3r=i4qVHNKUCP8Fv*>p@9LLc$xsL&{f*>j02x4zCMwg}<#c;xHXV
zn-}(SI#}NVrgyWMN<H}7dGoK5-b&xqsmveLS|A|<y-2%3qIGYJWZ?OVSx~LlKTB?z
zzORr#(MSy8JpMchJ5!$ni<<#M=j@4N<kWy+jl7KBgZna|^xHhbyxc6&u&UU8VHOJ{
zpm~MhvJBdAS(<$h(>+0tg7?V;>eQkBZa;H!VG_lX_Ivo5qA&m`6E~fxVxT56#Gxbs
zrZe}C<`||uB!GAmgU*Bdb)-46c8t*Ya;%Q7J`X=8P=P>i?IcrI4jjvfP;Y23dx5T|
zRk$Z(Y~<r{hqB<LH|-+&tC7sp?BlBZBJBo%cpL6zXGXV&8qxl7RxzYroGJ{0d_4e4
z^MFh1Tv3#o&5#yJwNFS7UaOZADk&mbbdfT7kN-+74Mxz53CSZtX13x8Yf<Gvn`k-@
zZZ{b5;e(xm!(8D#0bYj->zDD(I_=vwSi6+JQ+`|a{Q|y^$twSN4Ai972}fVAcxfMg
zD@Um7>Vs%z=db`=4UxHIqC|2<AL4==s_v5en|!@_`ALALhoqxpc8kp<M>MUMCu6=@
zfPq}Zun9CclvlG^XJeFBFwFOl^ZLZN$&5S8#L`X=oHQguWlOm_KgxsUrQ}GUJd%P<
zd!Q9m#iTjx(i`RF&O5dy=A1f2YgH;?i9W2sFqc~O(kwSaDN3F|DQ(@d(RL?&gH9#m
zG6QvHI-~vu0w|3#>g*d7{q2$hQ4AJYP#}0XaTqYXmJ-h@8Nv}>Q}WElz4I|#{^bGx
zC?zzTdsw+a(jYm8oisGcDXonKIZ`d;#&Bl5u*r2|=%V)R_Xy!velqVUte#(k3VS-G
z^eNYX?YAgNFMV5FWyX^5edNmAjOT_t<W0@6QitrnA5fe{lgMya(lcX(m+7tT$LhOJ
zS!J>)O&=N#AubGsGuBs#T%XHjKnj0!BT(>Hc&H4UtnpKr!&?6Oc{qznlqVK(=U}9b
zC+m&_w{OY4F;jSW&1FW(Sd`gdQm~t!pMA~u5r%6OjLy{h!cKG5pSWpS%<@*>#^^It
z=*x4;!vq;xsD<HjfgX1_hLj&7-K@$Vq^w%`3b62f7mm)~aXix3wF@wbSS3<7La+LO
zS1V!$O%3DtO##{IUT`CbQ6op(zK}KgfT2p!{nFj=E9yV1q8h}e`|l+^G+w+%$on|_
z3H6E$_T$P7qb?`><&me7$iX%i!*|{vSQ+>_fN9rf%w_tFnlvs3*92w8pYsZ{&|t;u
zj5129xS{Pl&lD7VD|%F)$ZUi>`>Jl0Vmgvk9CQy_hE_6LIx_18FmLxHbh6V`W@B(G
z!9>~^bV(JRI2XZ)p+6T*u@8lWC+I{RE8g(n^P&m|3y;Jtl(=W#Q(rp^M}L>GkS?q9
z`%Sra^SOE#o!7#HJagobAV?s%zHAJG&+5I1Yb8aBX#dXU`yTy8bqyl}5A<B<c(F_M
zV+nuHGV-SC?%}&jrXm*t=C6KyI>6+QUjDi-Fgr?pTaV^vMF3Gvn7dSS4_V7{7_0gU
z2sAfe7Oo)5RX1F>y}6z<!Nc8H$&mKIaAwI^WPNC3a?L9LRcH8Qg_3-&z#XT^8G*PY
z?;=lx?)dESxLsSietN)D274yq_r2+Vlk48)PulC~L>_wfxiQUj(cCC!e}7x@+&ZD+
z{pV77gE%qF7P&M#^)C3-A@+}cy#uoY_Jh&F_4wp!8gdoKi%wHzBIJAf3IXeK^AP`s
z&&g$*S?xPJ&M8wibGGv6rQq|zLgoc}3yf5OWt=q{M@tvm?~eIg!Gw9%lnKdNDARs@
z4`P)3al;Yc+;>0ON7Zs8Wo5>%L9IP>-RJ&Q>R52G0&Hmgq;g`L8<A9Hv4bRP#lR+o
zj`MGQhD*n|9db949`Wmwfo|$erI``q{g|cbme$O~1ztuR3y1Bv<GczNeDPs4N<K`i
zn!0ARx@O?2h7@ejGlyfb>=tdNDyLRZk(a)a)HH_cBZ2&(<B9*Sr`cygkN!9tL};xD
z(joANqpQ+CO2fzAwhzAz@b?z?sFF$ry+UeWHHd*X{{z)3(v~aX;YW5u{+~*Ade?rp
zAoRq88^UV|{n-o4*1sDYD;49n8s;ttJu|K~?&XizK5u+!hlsUUIhiOcKC}Nc2fyXm
z)|TRP{cXppxZv;k8y7pTeNwE&|FJVDpEZ2f9@@-q7;<^<u_mho&giL6V|6IFr&t+~
zaver{GU(YNs8ewF>b=)i#7Vd(hC>)0a87UxT_eA~5o|WBz>SMm&rw(yy4A;&#fiB0
zMWAu8{%!<rT%>*J<8f8ZW*w%cT!1gbGMc^umt&*xuO`RrjDmt}cj+O^@eyPv7{bKk
zcO(iN9}sIS16oH>l^X0|f!#!!#REKPUn`moqv57P?LOs6(u_Q<9HiI_OQbfU@N%Xb
zr~yM(2Ez=d?r|h1Nj%Ln0|*s^XR_B1rKozVy6Dp{LuZPJLH}K!7e!|e0AiCe2|+mo
zo%VWWfaXwfrA*x#Ky)={-*gndutEqQ_P1g`vyX83wnwF$dG56k^o1q@5o}$mna_r(
zome5Jv_)O|7KDsMDp-XW5U`-h#;jH_Lt_6cg@G9wJoUs6>7b%|aLP)?LN!A}Bq0=T
zD%8wqL^b?}GEK}<`gAqmWX~QU5tNg`S|09JA!3K@c(ccM%&^=M`B!_(%>N8{yE9%g
zYcS7=P1ak!|JEhg;xH+1{>;{N{NMEPOTAR;YUWZ)7LW>aCy#-hz(-uj%O-~cWd~MZ
zIK&=JF5fe%sd12GE<t$*IFePmBn|H>_abQwBYs=wX;B;x-Mo3y+XXKG)B%#{t_<=R
z>DUA&$8(G1iz(_=9MwYl(1P4}ribbpkC?U`z?<(zcioldXj(ZkZ+q-aEPERh8DLy+
zMsQ5|+e?_IMXpvo!|XDya2abyh1K(B$P2{H0NdxN#5JGP&L68p)(Npg=5(@xb98ng
zK~u1UJE@AB5%3`D!=*v-=)x&g*Vq&wthb{X52r|JJ_=Kx?KC9;C@AB&<>&6l75=Z<
z%kqFz?~ndno?`yxjisC$&I%zS--I)|4vgI1M{`QeEEPb9-{y1KeH|Tp9t+R#sM5ol
zo_xH@8TA1Bc>bzbWM*ZWQ}N*s;zg0?Z0IY5T!!f1E6)|AwKx-KO3G=-Iim*j(-f<H
zp6H3+*+SBlM!AVT&C$;L8hz?c5mr&!*JKp_4b^)03*0Ih+Y`p2J%;+twMT;Xe_4Ak
zdCOKiAs89Xt(0QUUjXGu^g`i==31uA`;5-k-Q#|y)a=bR_vez|qf&K~yzspXb$X?e
zs$uHc?IKB)uoB&%>PZ?JLC<nrB$?NF;9GrzRnN-FX6If0oD8i=<+{z(l5zJ3VuU8f
zJ;!lA{IyiIAiKuI#p7z1G=Mswzn-|Q0AOU+U+epf$ax_fc?gRluwiIaA?yR^W$(IC
zz+K3yzcvn4rPe}${(3auOEi!_)DmZoZMehMZcyP|C!;6!Qbv;$Jt3wE7}&G4tYsGk
z>zjiJot{OG=ttu77On~y{IAv8Nja_gGO6L$#d9ZIoZhL4r)%G3iyd1HcJ95h7dU)~
z6PW3lgw=}sLsy<?7#cBY!9nY@#DBXYYOGKAj2JXS=W_|a3aNmV9gklS9l581R?n~d
z(!wktr|kRggb<NEMq!s-Q?sgsJr!p9P(1egqplGqG@2{c?!HBj@5yX0>teC}biVy=
z9)iKsA7t6;p59*e^eJ%4*$gr_qVAaFd+_%;^3d|z57x=vE=D$3)?YM)vQUx{DiN9K
zk!^v_`bfZ}CU(Q6X}GQvL98l;<X1AC6A`N!qKQW(5UxV50a<yyFJdh0R76M|U6)K)
zqS+dlKDO*fUfJD@b_~tTd@Ip-a^sXn0<gPGJ(cIigzj6HX@=L?)!Y!>YBPN92PVpW
z?rbyE!zmfi=ix4g8pnnf^3WE0u801}J4&nY<C4DOQ_h(cdlpt(#=adBaat>nc@C1Y
z{g_!!yG5$f;=L3@AIY_lcO~&l$j#ZceWBikM&AZq^6T41KjBie)}fz(^M?cTjC8yA
zhB~4fP2gzgDvzbpktQRE_6o&5;<38-+fX&Ki@0ANR)v>=S;ofh7Nbn4uJU3SPrmz0
zjof&;&@4)K^*0=Hz@Qww<-Vgg@`_R_i3t!^Xwwr+s&~{q#MZoMzgBN2zUslX&+@Xm
z7BYhGuqgJlmU0=naXOat(u^(QdCvlNVR6($_M!NCq@u?a-L<e6#7Aonnd4{uLPSud
z^n3;SDxtEu>FdYix+7a%rqYhRy!;)Cw7a6XP9asxVX0dDt=$jlKk^#4J;ba%LaT2?
zthNhxRE7*Mh&|R1rOUPt_BpzifAD3Yscy3^$uv_gZV#m-m=|n+d?RqN>T=sci}K!z
zIO*{g9JGJ0uw9Q4(n3!MvUU|;Ee*15_2N`N{&7-pqpPYhm3Q_qY_&2-1RLB&6FJbE
z+bwux_rQTFIhmU(6S;4yH&|&};pZii3G&pG0S{+{lR_~EWFq?MX%!2KYG=L>Se-0P
z+`$Q96$=`CI^Z?Y+EF2Wi5P_jvl@1JM9^{d>Q=FRvwX-l8;{&pg|Gk{FN|cPMnxtv
zrC-dp|M38Q!%J^NCOqT>Q~=0$49C0Xu^<#AeudprVWxOKL)S><0-Mk?$epOYbYxZe
zyNXa9cg68sr@F=m?7Bsm<85-Ev-yEd7hqxbMw%h3s1*i0K)XfA%|eFp$$B|)(}?uQ
z=-_x{`Fv4Hu%Ksp65{*XhVWk*CpJT8Q2S;v-`nCF$Oiy$0KAxjd_xFCBt}xS2i^i7
zNIi3J4Veq>U4LBthBx48bWW4Ca;+t-yuH!Y+4j&tEz9Dw-T0Gm!IpOZ!ow&!h1AHg
z<8hS%etzU$378qCUNxU(H@ix%<Fl3#z)7h$Pb_6@Tz#(KX;U%$$b<i&Aqzp)BRbn@
zDsY@z2e~$vc>Ap40<hf_F7OV5)g8P+Pqb17h$LaW+#uF2DY19bEWWBUhG!%&jt-#@
z+@kA7$tcluY3XvZc{Yr<N@pL-j(Y&!($OU;c_q&>wmg=fs@9d)W8zBSQ8Jq^*zYvr
zgotFCM$J(CTZ=~Ic+3`{7WIRMmVkmq;|L(_D2ghIN&!d%cbbW?byQy4GN$_KOVdtV
zeh0Ks951Em7<8rSL2YkIxJZ_8cRB(c5vj}iGJ8LaRbo?JE(0VGz*k`u0<_xyyHt?_
zLOa+HmUy8=f<z?Gp0ky94Lwzj{p>?07WLm43wh+x@4f&913tl;s%ZTO6)06l!IB@B
z$rXh9c#KmGaqXJo9S?bWGgda?3(Fy*ox>A(`4M%L3ijpAjs%?n)^FreWLeH(%3U9z
zV>(!metV;dJpw;hhDjP@6Qi%d8(}zxYB)r!WP5h3Bygl5B2>F1cjtWmU%qO|(E5SJ
z&1cVMDK~xNV)XB?Hlwrr|9TPhwK3mb4^?e8p#|9L70-!~+I^Ih(;5$wk(jJ<WGN^l
zp{j6FFFKIGSqrq<mn-_z&6T&7n-SFyTT>ekt1rwq5OqX!DfcMYutCeQ5PxDH-q;=@
zSi{eS;n3eS=t%RWm-)4VnU{B!$FP1vjcE$yB-D6nUq25InzAC>xN^BE+;}OQ8)!=a
zmXBS;WxQ<cB>05rNQFDA6xIBK*XQ2g^)<E6HHQlO_4FfG?wI7<V9FI`hXGRZMt8$^
z5tc||@M)9$=ck&y--9e){&RKig3LELWKTbTI*QBG`m;TZ;xO|ahMhL#@gk3}F(c?=
z)=EG&h`3Ak3Q|AR%u(*#+1>V1|HZAOKrht}UQD&m8`=F`db4)0?XEfD!6du+9#lwW
zW&$Q{(?QDBjpFu=x8)*FNc@6Kw2k1cSOVdMTDMeRYqFyx+dO6D-R`5xrIJMq+|*u2
zy3luRz7FA^B=)kN27%8woXl{Tr)<g%Z{SS<5m*`2DuT<J!=K@oHOt%;j<9R>%2;<g
z^SBu4tcPf41Ku^PGfad1)dGin*m}{)B%Ma^qaFOoTO;CwL^a*?0b<rDUnjR-CC6a#
z^QTL%Eis?;*^9oHe6t|I6AU|b6w=z+MhESk+_r6WZ;hSo%HE+xtGQvpLk7bOY(mYd
z0kMV5IvmlKXT<fw)mSb)4s;<G{^^@~m2N$1?HZ$(M;#2u7nV)arT$8fOVx%oM<Z^}
zlg0*g7Xkx*QQ0k$r9@g;bnfsfZ$8Uct1wH+cZfIL_HZQkzqqB?dQ<$hZ_dBpTgO6k
z(Go4~I{dE|xsM$5odt}}F8lUe@-;P@VdDKt6GTCVN+kZ2bAhoaYy9;rWpdwV86FrI
zQg1Qmvw?H=o}C7mK$09+8<OP}uv06!YeqtI{Q6j09nl2ZB2XPa<<B|OgLY+}(d3*Q
zJXY2ryY<g%GgR67E=lLfq=Mrwm+%5>vZP3;!*9g*M)C78*fD)M0OERn(y)ogc!k~n
zl#_e%od2EWg>`fNKVu`Tp0fe8cN{Hdn+%+^??zaAH@R0Lh23)XY-iWA?0g~>PbTvU
z)>>&i@d<y&b;b4_jap*6sS6nlm!1mIGeqPnj#?X)1&|9Fs)>+K+&>u}A{nUg$3Haq
zH1F`1Gyv2(4SV3N;0vd<G-5E+ku?Q7c`3=6C#$!(yCiHeVAJ`)Yz){OgB>JhNK4I4
zQ3_t83y*F-*`iPkVpgPpukWHr3&za~T<W{Ta`iz%t~qZtasA+8O8&|*T&B4<&O~^?
z$Unv^X!xu?qd<@l=4a^4eQk2Nr~?|yS>10~4rTNg_&xg!2m$4bvgSx>_GVZn1fWwz
z&8n%fxzJqKK1lAF8uPRm3^^5NXpz|N^J(vsMayNxD58ySz%eZoGCNPP-6W|BsMprr
z+io!Kb(Ram3DKE-2g+!JhHUCVoh;O_3ATzy_c_V$`rS@R18#+ncBP-|-Fv}>%>CI+
zmSZj#zhFL`<n+2-=gBY%OE8%g4D@IXwcnkYYL4|TTMhnu#0pfC+MZH5lHDxXiD<&Z
zzcP2Pa#QUYp_8}|fM)ec6{{506FzA<zFu;k9&&2N*e@b8InOQ2JY%7RXq^S$CN7Gh
zr{?-luxDk<9sp1w*2>gkxkV0>yYjXKWZL5EG3`t-pq0TIu_h=B&8}%tyUk~{8ESK;
zO~kOQ;$lAvOaVh&HpeeA6`ZkhO>HxlG0FqyPk$Lj(`B0-TIIz0v=c|Z0>q(8=*se#
z^n`(R<Np^A;0MhzqdSB_d&yA<<MT_pyo>Tw{qicdE3PS<9?viw?a@4zYH|y2Zt?ka
zIkZ*Wy~(}mAwiXwX@fhhNa9a%vMr^3*>YE!FE@cx6Wfa^tlraG(Q0ejDlV<Nk85@s
zQ!@Gm6Kc^8Vb+O3!gp*Np~9C$*95Sx({H8rthI!-SyfmUEd;f|KqOY$tW>q}n5V+^
zxA=Vs!PI%W<B||_;WfkQw)XG*^+7|+5;#<%pr`*ymL67ojiCLR_ml);Af!Vo69JMj
zVF>Ir)9zS*{Y{acjf)H$N@hkHlB&S6rT(dsH-%!F*NKydP{>*64lr+ZApIz$zcFyh
zSji$`i`=x7uOMdq_R7>on(~T`2y_yfuz2{yaJNE`HM|RPn|oFN{${JZI7}mwoX1EX
zBW}LaZ%~v6hgXAt4NC%ob$c_zHN1>C)RD^g&AJ<PJ}*`5pdc_gEwpCmJ%1!(BK^E$
z!NerexpCV;VOZu%_ZE*w&@Gmz!ByEZdU@6NmlnBS>4m6vX5G6H$N-9SXm);)F9mI(
zFLw|z5OlXlzYJ2z{BNZPq|+M?U%q?qVF>zW96`84uVKGi#(QHQSxr6*WWwDqe>UGx
zO=P7H<9^*Johc<XnGX88S0r7ZKr7T7KWHy4Pi?{-XnZxxPm^jhRx}s6K9e8c7(yA4
zW~u2;Bt8FY`@@(<5$bR6jsDqy9XG_Z$}hi2s&G;A;Jk=2wsj1yGdBCVcwjnIDbkCt
z;-MX5P7pK&J_xv!mjWNC>}9Gyo3ajh<7}j4s`lIZ0Fu$dmfP@S^vj05*6nPZP5!lb
zp`uEHg4ma)+Y$=RcWy)y@c%Gvsphud5DXoa3gCS4^Dk=uvOhW_7W7y|Y;ZFMmrnLH
z8G(OUhU&{ohja^m=jbY3pb;IF`nzy6yuah{Qz0CqsrKq?`0v%%S&CReX|qmxD-(sR
z;hINxa;JVKMY4`4D@yZM=Z{DEANr<h3IDrfjBpw|MBZ&%b`GyaPU~h(ip4y!Pac&B
zYzR8(H}{pw@nz$6?u(dx24ewpvQ?KjR*M(ke9NM!1hY$)6lQk!l^G0LDQG1u+_hsh
z^2-2DwVGE#gq<6<auoM@Hi|?F<-~*f%?KB2E?S~x7DzEJ`P041H0RWkle!LlBO5x8
zAcclP@8pzqtG_&_4>aIt(b`J~Sr5L!_%r$beLoS3bURWs40V0TqH|-=oo8{vrSXdv
zr`@MsYL3GO#<Q9qUsA?(u_+vCPAD6-{ArHvh%N42)+!!4c2h5w4WpRCU`N1Z<?<v9
z0`xa)q;gHuJpBcpsF0||&9q4A&P^_yH#PhMH2Vj_C??eDtklna6|kD%A07}SovhJ7
z*q&ye8guIN22~XArJF7jowsa-{>+J#QidjEw1z6X{de%5F3?1yq~*%{Y=-GV%2jOS
zEcRXU8eR$nqrqw`j3Wn>Je6nyhy%%sLJXp%4?h-VV?~_ktcu0|>OonsD>3Dttl;Mt
z)<eU1n=RgPLxts<590T`=nQL-?m1e5HI{;m6r}D*3_il*71e5d{Ab~A$0E><g@gXh
z$&0+<Ys+fP^AP#$R4FJCB@}wZ<`~LTmQj<oRL!iC)y*>v{<M{i6LjlK|3d$?Cw>Os
z5WW2*HQ+gyae%6EfJi1ntr7aU*S_M@*#AIT#sjA2$nN={Kl1-p)d&aV37;1)p1#zf
z70P2SKdb=)7egzLAFf{Q%x~Vgqi*QG`@Szq`rWnZ#}y6*Idqd!#Y<-wvnJO1mRG6}
zTv&^qlZK#Dk#^uZeFi-9k-NkU(p#_08dietu8Kr~gv}SSsHE9ikkh>VkQzhLhp}ew
zg~|Y72L@J?Sy*e;XNt~i=gkIg$YxylvG^_x=Bb|NH&)NvtbY`$pEU})J9jjw1R*0L
z&5$o5jw{xd)8p0uk^qGfo&Q>k$Xa!Y^TB-9wH#rtF6iF{XFgA*9jsG?AkVVve}*_9
z`U?E~|LwZXf3mnrJwan3r(FQGd{s1D@_(RQ|FmCR&b3A#*Q<r>sr>`Sp?AG2{{<b1
zr?zv_D_6htG4f|e1TI203ltm*3;zeYk)?37v2WjZiQB953^)F<Dk9M{+1c-1@;lq#
zi$~*znbdpD0vP}v$w2qC$-5wkDuH5uRjNr-<8AND9iWQ?zFMtYe5Zk4rZBmHxv^Uf
zOTJ-Le@QBRV9PV>qX60P?TfC?i}~tAneoFtpbr|d8O!b{6*3{?fVhG9F`JM4(vo#u
z&s}P1mvP4yxlCI_Ml6?Dwf{iCgs;`Rj!JNzRn_GSwOj#jzyyw~=m)u3avY%LMcIt^
z%Bwulnv4|9>3e%xJiIOU<AZcum~UE!QcO9hrl{;%X#}VWovWAH1^bn2o9y~`u0C*o
zD^n#^Xa)2@2*;vzcmvf&z`@B<#AY&w!Gkj5u&M1(Dq+=Z=L&NG{~nR3(=+anFnbV7
zNTKo+cqKz8mD=;WCF|zZ8|$Mn;e(11EA(2E->~C9><DI8d`qYdJS1~&fs4X+9JKO>
z_KxqqnCho_s5L=`dzCeHu#R*2kA$6_HRLpo9A}iIWn`I->nJ{Jcg1I8`7nChfN6s3
z-UsDnrehjte<18OuhxHIstC)_Gn+fiV~u5?fZ<`lH<A(xJUBD;XGWP+6X)W>tbvPW
zH%+egU!5M8Zf@;P4bMnAY`E_c_5pWW6EL4tzDXaPoLG|4kJ`!VCR#}QTz@9OF~$6F
zqkWw$-kvPdc8<sRb^YTpIkB?kNH>LhywX$p-AA)fz_k~VnR5Nv&5Xqv&0PinF&n9^
z%@z1}bBLC(oZ#ORM1EDEh`m|DOF>n4gOyL~Mg^4B2v+{#I}3TNy>jpb>#h*ISnMA)
zf>)qj-`Idn#=$6LTwWaV#&@O2I9QCA@9~;!_qG0o=d|BiPY2&gmqV_xOXf>1s~twL
z&ZVfO1^M6WIOZNm-VlHM?r)8xFGw&5S$dV^3<v{eLEl%8Dy+n%Chgz*c)ng@zY{JM
zP_(s>#b=|U$lhn=A}jDP`ulOEqi0Qk0A&65tYT0lhm<K4|C^YeHD896-^==NcA{v&
zKX5&!L}Cuaa$5Z&Fe<FlW3-e*{10&^5u%x3jyt;h>!%cG!4}7teG5X-v|adlTISuN
zeAE~C2d7WDpUD5%H>gOuHM|AA&ja~A7D3`M)MF-wG(_Npv8V>+am(TuR(-{xiR7jZ
zCr6gq`Jixx#z`x&<9h$Eygcu?vI6aYnC0J?==&E!|GBCYS+C`h_f6?!Jc4CiHEhpY
z4MvppyfL#hI>2ya?|WHI@;>3y;*#C7U|;PmakhvFWHs(`ze)$@L<QX_db>-*7gA)b
zHzGUoNgRu(DlD3-Lt>Uc6w<3q4Ws~|)v)hD^RHbvg?BHjv7o@;ChK7cl_1SaGgT3R
zTL8W)?7v*T+6;1&=*(dx#toWx+<dLbcp$1S$2d4o;1hFnE`;}6fKSu{Pz|a;s-43^
znpYqi8k2oT=(sFBDXl?Tzf1T^`YbNB>#`h>cfjF;{(4H@m*&>)1#Q-tc8@E^$vC!t
z3Ob+Txn2<0AAA!>GzG6#X{Ajh8a(6<Sg5!((<PEAr8YaF@?W?8aQ#?t+siBbN7@f(
z<w;<JAVWH`E6~ngka#(vmGjj7nZtTIE4_0|x@&!lRQDjMcn_n&*|&EeI`g6>HXnYD
zU>oT0(RZ9x<Z88hT>XBF;IF|~j<3I;X!j#*u%vQJvV;V4xQah6_{mBi$yQIr6B=<X
zsZQf{bKZ2)e&zYPV#ezKKpgc%<I2zKZUHIaC!qg<_{wK0n;SE}EpaKo?Y#fCta#I_
zT9=uAnJx%u0?O(9jG{o0vpW6Iiy+Ou{#z<$G_#MYUfR>&7C-7&Mu9HKkBHqm^8|#9
zp^n3HxnH!W?a)H0zo((zql=T8*K1yG4KigCK^ruSn1exk-|h!9@Ex5i9u7J=REWTc
zHJ;Z#taA@cj@d_hGxqg2G~>&Kvg0L!^dA_~N~p5VK2ZTl5g}FZ1Vy|tgrgBU(FIV=
zj}2wtj{TRu0tN)ya<TbzrmiTDI}y|#F__b9FTm)p+>FLm9TKV7h$GE}1vuSJ5N+&o
z3Bi)1kwJNj8qhqB7Oi2%#Be_*zI84aMBUhDTrQyp&B!Y+nIYR0N_tRV@Tvi}3o(~C
zb0+p5d?=Y0r^7;h=~q`mPdCdAUW4J9kKMqM62#Z2Pl<0Bsnh>ZtMaw+Ur8`eVI`(j
zpcRn$kG5$Y3SbtNN>1z}jNhfZy~p0sNem6XlYu_>-yydmjIN8rdM3iF9LAf|7cHoJ
zakyE79RI)o%8qVgYNQ7hC=Mr-U+<pBJqb8$>+ar7q{Q`0KU2fE6u4F!y$P9+-qC=E
zlcX92N)9qkNOP~3(>dvnNtRQzam5sqA)0h|DllNei@zmfJ0KvG!-8WlvW&G3r(VyB
zb^OusSR}qC5U6nH)JtvaE@h0YFu;7i8it0tE!ud(Y?l(8FYk~ml)HPzQS7(xOE<h5
zh@qpzc!uS%f@ui!_EKPfY3e4f$#1Cc*DRfe>9C9WcyElhDUHDIdDKmL9_yn!5kiZ6
zv(Ex>zVuBlB{@`D8Mu(u*St}I3gqTE5kl)7HUUvH!#y>D5cQ|6r61Sy6aA4gpa9A}
zh#PczN~`w#TfDn1SC02-ZAz5?+9kd?%V`;%)1~3FN@k7E_8C4*zYmdLtx3}NGH(*m
z>17qRIT6e@x3hVv=u9KLA&jGRU<LwpU(0w{o4tdqg5M=Jbrbb$;ixp_{7Fp%$xMSC
z%df2c5!E~UMxSMcN0CV0Zh_yZrl*}+lO&>!Sk5bI{Opcx;Krxm61~*Y#g2hFO0oTD
z$MeStooMPr#=0t)7`e7^f5n;T8qXW1r3gkG`BP1Jz@O>cRBCNFRErfGl7RryfEU4N
zL41@#$c{uw`CnVEl6>&cNM~|Mxz5Mc`nT+P6mkLfXL~gzW1>zbW8-3taC1#z$OV14
zg?m(w<G&^B!DhNtI{Ytt?Wx+}H_K?(d<o+}<3S?5kFifwYCA`D{%(SWKE#$uND6IL
zK&7rl66htgc8r~+sQWaltt>L17?KmMvDa?#^t$b@5fHO$fnon#Z;Ra6`v`4+_#p&-
zpoeeQ4eV3MJx_(FRe4^{<kJ_FPHF(Hb<sFYr_%#z&-h3)y@ZvY9@99IdN!!*0Ensx
zz0g-46#iK8{gLzjUA*<fc8U2vQf`++>WA-jrwjIi70t)OU;%a`cOWMHfw^1P+uB+K
zj5S;YE;)VO4?>Y@b0#di-k4v@Y!r2y3a)!z5WK34cY&>h?wQkeRVLhBneBO>DQkTC
zN-kVrGh9FD@ATe1jilyZ!EnWsR8uBt$7+pkxO!zL?}#HA&Bd#xSND*K_Cp$3{33#|
z76OFs9JkLZ%3bebG~Q@^T?8NlPxRhqo^q;@z*WGJ<War*M>w0|dUaiJ&vz3LO%-%z
zQ2YQp)Rfx0CT#h$#8$JuypbiH1)HJYA~BGn*}a*KR^uDbMwSu(qjrP;UsGZ}`OSno
z#=P+NRG<_%Z?Z*X!c)^-sCniqltmlvrD18hSey3E)7_m&Tq}|>x*$|SY;X~lwMb0u
z4E9At5R&OCHZ)3_O0*E`&mDi;(+4@U7!4h`KtuDU#w@*6u{t7td4V#--VHdZ&wH%P
zFP-H?kyf*)#6}z9QvDI<Tvroq3E>hI>UOTwgC-ipl#4q>&OP(G`>aeY`IPkU&WhEC
zI||u#-V5$O`;&nFkw({4G*8808Ri>DWD%$#xmMtHenR!H%8@yo^?Qm!@Q{X70~$5|
z@<dekb&knUA=^OBmY`t*ou<m#j*dX9Laf%%(x(sd2v#}?wue~I{c`O5&v*HMOG&ju
z#9oH6Wk9Tu>VY`C^T%qHj{MEt=jTTh3)SDA9mmbsvX;JLzga4u!9dbKv29P+-82;A
z4i5y7Y^Vz5#7?CPNq1C2<fI|<v5Am<LwZ(sssK~&yg}ysQXpmSY&mULLd1a?3cP&6
zOG1nKRmw=Lgpj0SI}}k{*B!tVD^5FnqmV(5>&{&kXM?BsfbM#Z?9?4E%^Ky}?RGPj
zDOc4JioO<-7Me8EMa``KG(CpRP=^3gUMD+Q=<u1=%N@+;B|=We<ubltX(O)lENhx)
z>}WEzN09xFEi~yLyehx^cN?VSR(owCeW0qkgaHIveIBbxuLs%C+kPSfShn)yzBwFl
zcLYrboKy;1J=aiRiDV8iDqQ~;Oc&?0?ZM*p(S0)(>gov*EYpTrH2+maL7g3-dxz=j
zE%#W8wnRwUdb1U=eHm+M=C*g%#&eQH^bHd|u^(w~zF`mKP&y2+X{^28DSZvbpB{md
zYO5N~;HSx~629P@Z40<@?o0Dx@XLH-ewpy>Ow9_KC$k6ZR>USz$>Kf!&hc1)!?GMa
z!KOHc`?TqeH;ta{4qWm`ZsE>fgOLn?Y#j=%*Xi7ONxECm>;8hK4G`_G!+6mk)N|yY
zMp)mNqn{kfn<BvmC5WX7f?*7b;z2%ZfEe+nzMR-{mCg6C5+<v^GRA8{$8vlQg4?{;
zsVibH8Mg=Y>xq|04}X`ob+tPAuBUunFvYj|lDC3(!~nK;4?3NSa1&)M>XIh2zyxbc
z1=cpMmzZ^|7O<Zq6_Io_g8GJme|i~8{(PrWy8IQ8izrjmt4`6?hKFCWCK@#WaVRv0
zVZiKMyWN}6<2oWr!tv^W>immZj6ZN<7e8HgyCOe>N_n&K<*qockn^e<pFgH4HrKJ*
z`SX3y9<~`{ljO0cp2RbA1uY&@1kN*d$?DN*qug`pYG-DPSn6W&AqCh_zrc<ch;yUB
z=)!4D?vPXjT&aI6@J_JaiQ<Mz!fcxj%0PBMCr$k+LBMKhL;-?B!_XfWN>SV{cq(`j
zRQ?MehRV)H8G5225i}vqo5ForJSJnf77x@C^k+}hv%4!o<bb}_kC&i}dB^%;I7v?D
zDiu>ddpB7fNtEC5NtsvA7r?fmy{~s%#+3<R^SwQ|suK{(`x7$G#5J6af|70xKmPo$
zjM=+cmTbul!<0B<P9+n>^w)YPFNKm2qi-Ch;Cj*IEIILvO8*)dy1O2lP>$1PrdNO8
zYoEvRwMCnL<hIs!l$ZLC<M$&>-4v{oTZiO^<kP4K@*M`rml4dm;mZbaX;!=Bagf*s
z1N^tSfJ+BOliX3|<JBg>8cFC7P)&m~2s|p;tGdJ$FTUars#t&wda_oRQVp6L6sa6&
z>DU4wBmn5I&6DDd$pm#*u5oihvj~rUnu{Z72UocC1?faU&WwuFzkD~w(tHJ0Qr8IO
zS?tX!)bU#Yxx?wWZ1D0xlT{??$}*Q&$S&N|_1xTSF0U=f+Y~ZjK-sr#yJPH%@BSaC
z!U$4)zFBoa*eZE>3tice(I^>z<I0(xb<U$f&~nh?Y5kXztUe)WW3|j1H~x8mzK`F-
zttmVXs!!SC(<|-ccvAOeI_=?c-Ay*b=eDz)^IR>axBdrOvd?%?`JS{8hBi(!o^e9F
z7@%IO>Nrs}8E;rk6shiZ&U~1&bz!ZD4)$QW`Ywk%>5?Obf06he-c8e{_(=Vv#-8)j
zH$n!25vaKuWZqV9A6I4}jN{jPJp6++2h+6f4lA$5aO0&Tu$sITl8q(kCn^vmZDouH
zHO-(sDN5+K&7dQO#+YR&HI_N5)AGH2Q_@Xj2alMfHF$_ePK?wqLDM*en;A6JL{?$L
zgi-9BujjLt(+7k+QkQ{vl~D5l1&88g$*1A1$vUe^=Cu)4ja{_N(ChB^#pOTb5;!5v
ztN-d3;f&$N%#AwmcNTI#bAA%_s}|?0OY~l#useL)LMxmu0sBUo9XjpNDBo>sbc}2n
zq;V9MXP<<vt53P>xH$TE?E`ir<nsM9bMeKIo=DL`s$Uiz{iv7PnLzhzsb3AGun?Ei
znLK77re+Z)$iO`nb8(YD5(Au%VA(rsghwGCVrL0^-wq$ketWSWr-~~M1;TU#7li|9
zmBuyql}IGck@0Nnf4Bd&giOC|W$^sqyL-Hnw8q_DH_kpUbo_xeY}|}$GYh<@IY{-D
z+d1Y+Hp)qv`F>e!!gUEok5i%~nMefYi<!trB($#a46P;GBs~#xLv^!y>CTAuM#?IQ
zS<jp!)j+sWjro1Wr)<=&ylKQV7&&LP0J)0`F3B(!kw|Cy3dNX-(D4rMAvN7$gHX_m
zPL$h19!sFg7Wd4gp%dgaAtia@GE{6j+E448{4kBT{}DTX)cCDwis9<3X4lx%<#G~r
z8YCzt$cVul33BTaaXhJ{j$tsA`_GdBWU89u&1kPOCn{{RG{{BgXwMng)u(JKokzc@
znIAye=ke`EimRIWE)_-ETo1~Z2(;b90XVuuOT&K6mUNCx2k{KcVYi@6`Y_Usma07f
zE4MRyVAFj1_qcYF0Y?~dO)fyCB2Ljm_n*FgO4t0@3G*+c5Ha>bOB2L1IBGYfaLuF;
zV`CI5f(z)WWG>}IL^~nEYN^_Vj`Ob6Ka<{;i<qzBtTv3%6qM!V)0}Vh`J&@L*kVuf
zQt_FRPWJ>Gw1tMf$|z7em3@DlMoelXu%qh6k9c2RZjsEJ3q>$6iEbg{V7cHD<C!~y
zRljvE#bAYei5<RT)om?)R&&+nwZ^4EU(d|z>EK~)lRI}T_U%xKn_MvlE5GNbA^8S}
zO))C5I+y!a)^;4quQ=f4d<H>#iuaVC<VEu!wK)cbIx0Z18CZJ5_v)aXeI4l@u@;b=
ziv!%tgskV%%T7f>x}-@7S@|m%vg1Fzb`hP`$Ao{fulu4tIDDSGE6ZCp{n97kkimOT
zjn$)V;<_~m!<XEJm`;Myh_9AIK&2gnjy6WT*%`IpMvGr>^Ye>{Z!_(G`dHcjH0HPU
z5({cJ6Pu+80Nt&;c80ut7CgQ%OVg;0-k-0WYrw^?UKdBi91B$~H=zkqZFePaly3-~
z=-tGf<!qv#`okphmcmu>3aCD0Mx*9(7koP`i(#`A!o|egzzvG>Ozm5FTz!r{ORfsw
zxy9SMcOM>Lm6tUL9+!}J_gQvjE~n|J-F;=(#{|mQ8&<q7Zrz%AcR1EO(^T>m4|%~M
zoU!^Ubtys2nsIvy7u@gWJj?;AsyXYIWnG&?URY$a5N^n(8)~tP`IiLYE8E9iitY~L
z)!+KARq>Qb=;qGXdXvPCL^kv*)KL?I;?$}|CuI<Y&33EnmRTRgADz~n^Q#qZwa9^c
z{dA7dKt_Q;FTIA9=h}q3Q&nX$?bs_RbvW&+KaKM=?}U;Sm3Qe3V!A*yEewy%Am62{
zly|hMXYi3TruHD6ztZiLeWdm^6AzUDycBzU5JekjCaS64Ts4bgU)*eNhCS{wlpmx~
z{v%yZXgV`FH{;<)c<^fO#9BoPMKxIE+GfE5GCuKrc;6wv@g-i6F7RW{vA&*^-Qw2C
zrGuUy>f8SVS=crq$w<2tSM*1cL`ylpE|%O0ztqd;KOJm$tD)OYTnN4OOaGx&Gw=Aj
zB$DT@+)Sr@7G=tMto!I(+(U39`d$i5k)qD}%J{an*XQS6WAD(#dV+Mw&(FSfTCB2h
znbs^(PicR74xfBzFQhit|31r&npCVhTD-zdXrGb#<fSxYe<kOGVQsF?a&vvX?}?PP
z12QmmI_1N>0D(Z_d6-<4%%#u5_wD?VAQ$f7o^AwP>4#zQ3Dc-g_1wXddEY|DJ@5M1
zdpXc)u}C`IkT-p)h|=KvdV3kq36JIen<st6`M|P9Bhs@twl?>lbh9h9?Xt{-W#kp2
zm%|QU)F8)_<kW!;$|~yz`xAkAlbctFNTN^@;YmaR>*`9xjK4J0tept(7JJI+)B7%_
zk@V@FFiUwk_3WuJ#*e%Vsc$V;9tnDmTonib&V4enM25F2kbt(A2{bvT5ZABvFenSj
z#2sL|ht!%y{U1-~9nIz+_WuyXs?}(X*lM>mqBgNdZ82&Uu~kd$s9jnbv6VDNEB0u$
zc4<`+YR9ZqwMW&a6n*aR^Zd^59LL|cbKLjmb6wZ_dc8OdQTkhkDxYoBUblT1klc^>
z2g>71)yL!?nMY=sQF%Py@A^2Nvoz!w3v@y9OzC+X<zeGn4tG}9C>TW<_axVLSqZ2B
z1%vK^k_=yOP1gB~$LHf~Y0j9|n$*HeVfktja1P-wix!*szIp9kyfNANJa?|*gZn(T
z6GSES`*GiQ2q+Tz+i$#liePg~VQbhT)s<^5;-k6347ciD@1~lVPe6BQi5^MjOR)m6
z(c4LcM|4nXLr<?;DqA(rlVd#3BOy>a{YDq*@??intzE6VX#A_s^vDb0Usk27|3F@C
zqDJ4ji>beJ)ONEXv}ub956czEo9|7hy2ol?XPaHk*sWm01H#mHNbyY}GGmx2Iks&h
z3;>d4kWvq~^kX?Ov|b83rr$KZN|Umsg_Mz3cK-Z#TyNki8^f2tZ*2*N@p?vMe=Yb0
z<t4%=-t)hSH;Y!Z%#oS{?&Hu-KKa5lfDMB`eiN&;(*;|?m^m>PLrN5JAWb9TB#@UZ
zCyN?~lO#SIJ!YsoQKnn;$ZR}bss?2-Tj1=_AhXK3rWDchYHu;i=!<fN$k}R)L+zLQ
zYBHC|A0PN4R`3NzL?}aNtRNTEA{tx|2JCMC=_-JM%+3FPpqU^Hn4EfHLOk(i5It5$
zerWGDz+A-Z;bdgXujxAm34@*xxI3B(!5=@u&WRi59XG<CowD0Ftu=bTAv`)w9I{Ox
zxs(5ND6cbU+YYTIE}(S8dqiw=;~(~pdaMy?Hu_f9%5#sLxYaE#pNaCJOAN-MC+rcd
ze374R*AwX<yclD+qCq7uAXKH57O{uRdCk@ZvaPd!2Ue<dVeW?5cDtv-dHKeq@yyn5
z9>%?{Vt+4u|Do5^xwuq<h}({Zf7<(W^q6+UX9F)xaw=@kET<iS4;FHumX8GHb1ZP6
zB=s1gthM{vF{^%hAJy%-j13z(vB=T5Xi6_gIex_8Nv;W3tx2M~`8+U*$aLz>VS0@w
zn<#2BiU));8-{6!1DzG30VYs2Hkz5{$|XRLliv5op_o(6IN7Eo*SOle$l<6=rdE8=
zfMWw$4`9|wFo#DD;(=%9HGgW_Y}vlk<;*|4&Q$<fD3sr_nkrjT4zBv{cJC)W4a^_O
zCYd2dHLQr6e#P&Zq>yAU@jUEGMz@S(ap8}f$NA~x`b1e$-pJ6{=$kZWZIJ9N@6cO=
zgMT1SA5nk5hC4fH?EOyzhl=5M(NjzA_=qtWt~&4QE&b!y*|Md|Pk*90kVaqAqP@s`
zZNi+PpvuAUQHocu9z@9WnKM()4aW*$VD%nUybod7JBDES%`mvmC)b!zpq@?^edy9~
zszU?v6Km0wJ$%*N9U))t{(P&h?dHwuG0Ohru`{dG`Y$sn#e5}3^fx%Es(umKw)9`m
zUQ<RU6(k!y^Z(YgW!9I)&JL}Tt|E#+nXd21y<_e90-iIuIl4#vkY>jkYM97=z$eeH
z780V4Ty6T>cQxE~oISVm$<%UzZF0`_({Ky&Ybax$Kwu}EMVs3{U21mhF^Ke}D25c1
z?YsTH9d>nWqEmnlYXZZ0xUs#61Cknk!ceY6I$|Sq)|_pk=b`8wpshn5O!&gJLG7iy
zYz_=WH6)>;y^Apn?7c+_P|$q-_|3}G5WUW{IPHG#n+!X~WA4p12q!2$@VFvf|JK?M
zfr6Y}qX5gYLHRCL2aSTS<y8*GhhkY4`&iZbWWkl2MsB}5*Ics*Pw=ZB)u{hCRSFfC
z5r=GCoJTK;9Fs<1Mzi;sz0-YcGtms$GsBH%S}#UtBSh!n`CQ`ajU?Y5%4jz3*jVG+
z#&tdy8uR0dJqwXsTf?nB!NBN|;f1F$hq-{M3MJqBf1rd1)^~^bbU{kDU-6wyzVdll
zUFLlvN3naK?hx8uPRm}bm&&y-sUEZCIN(NWv$6Yi#Ob-mE{ybUQ1grD*~`QrF1J(5
z8>SagXXBHbaxY3ZM&~|Balj%x!QFY&`rP^o9xmNE=y6{95&ZG%!luH!IV!P%qB$T@
zH-C>cuOpgsqWFt+W{N|cb|ZE|WVSiWdNpD;19{}RM0$Yx9gf#1WfKs@btC)CWqgCG
zRo5AXHt2YMDs!|5YA6ENKk4LT=raZ{L|Lu=3;oF?O>%`HZ=gzYgOaEeM>Q2TGKpbg
zYkw-V1>gPyIhGMb6dvZ5G5{U%Ts<(WP?Uioq(*5Emu74XoKna&=<hXXv`RcrA_1(R
z?1Y++GnYy>dv{dm%Hr1s1b^9Kt|MGuQnf1DLlic~aHKj?jWD|(zvfo&d1VE2^MJzC
z_qx%W3&g(W3lpIldy_|Q*K>oUH9kGv_j#>DlvoXTd3jr#GH^<c=X68(kafV}hr@mS
zPpcsTGr}={O#Q|xBd7t9^D_hY{n^{xC2@Fkq1pa&{_<E($3kQs%M?j};}mCH*A8rl
z`a#+xuM6Pu;gi}w^R5pFn6(=6{{vNM&Yle?mv#Tzyv<(yi%rL0;@o4{%9lV#!3Pb`
z*z^&rIw~h>^uWv--rwQ)$_l};zY{aK0U^a~TSeO<lt#(Wq$MsdP@9rj27&=Hk%7~s
zFYu;0B2YWSfg&c`oFla4s%f`QcV+EzoSvH(gD%=t%*`ba^2Aar7pP<36(ZZYWAno*
z6}m?ZDb~@n785a7x~Nj)K;~#uGK^^iaOW%0HSjtbkYqz2t>zd@ijdzqF3VZM93Y>f
z5y`sh#?va|buTEP(V||_|6BF^@2^br|3Y0fCFIvYuBqbUR{4w~((k;k|K3&+eIw{U
zowby5Xp{M&e-=Cr{13hiU`vAiS_U-}I)t3R%n9I~9(q^RN+lE6Zk-o6;|4xxg!AO)
zlr1N1KDZg5pWB=u80n(`#mvP2)&ABU_xN=mkHwUQqePEAUJL*sX^5-^QCS5CDGZnz
zU&mPO$)JQdqc_aTV5xqJmLSp7FNft{PIcQr_M->Km^A@N(oB*{86luMz@~qKy52XQ
z?Lut-{HuHwdxh}x^v~Z28^C#fJywLs3QBm#%1QPQv{~r-pmo#$<8Qp6qjN=Tja849
z#86uMM*c(<qK<sp*Uh|g$U1a?CPU`V{4=cg!V$k-f40mjr3xclhuQ_gJ8ABroF?!I
zx<rv*l0)9=Dr-iej&WTpt%pELRo2Pc6tRbMU3zA~H_X#DR`a*$IzUR9juzGt|FU6y
zv(&T4Rj|MtI9XWEKLndn+WWi~bix9@;ct2bYcB@%B?vyLk-qEG-cXyf0MdZCax_$I
ziBnm^M2LP#((B6>(x;K1a|ns4m!co=5*7cPH4yw2m2E5YnF=$?)j12XtyVc|WzP1x
zj&0vJqM`YayT_AAnV;XwnNn|*S{Ma-D~EH@@55)LM;J-zHbZa~gdjl@@2x3t#t%}4
zM5a!~UtPX%WI8LPOt`VIoSduwvJY<|gxAsXzgU_wg(tZSXI*Xx=6j`Ab`a~Yb3O0n
zYjhoNQyi(eG4(8CW6E@jL%di_lC2jtc7U57eUSP2+iOe6+O~O;qIf8pMG@I8snDfF
z*(sXNS*^!J@<es`>-+#P9#CwOnCQFfTp!0`o<%8L3wM<t`teBnu<BL$2m&khd~E!U
zpxdk=QHgB8%K05A%>6<%T~q02a4E(36JOXU8LR(6zPAmE<r8T6u#nDLr1FL5mn)d&
z!-{Vo2_1ohWGWjkS%W>NyJGT>!t`kxa@R+bQ(Iws6R8WRcZE}i%ZFX;L|X(O1ZDwj
zqJghMnYR%MJP%0bAT<L|c{ojyoiN7yWHpB7%eiRJRI0gjdF{kHR#S@+tUz96(*O2I
zw%6>=<N0pm*Q%Ywex`_}q~(We;gu<KvHJ3sTwAtG4Ft>VY^D43p+>&JKfJ%#n#xJ&
zz6mct8$G(;JpRy5XLns%>B;-SHqcHAupy91A|3WIQg2Ax5B@9`I7!d1ifR!bl1(=7
zy>AxYj3txTzT7-6WXSk=jC1|?(ljkn?~~p|@0Pg_XZ;Yz%tlX$`h&<Pfk*I!z)!+$
zc2}rY)uZn{sXGsSE3)*^B(fw&ApDCJ)rn*BxAWA<-M?p%_0xZdW|u?5LLyP&YQKj=
zR-=V>^YbopfBq40PITi&T^Kn((HLvD32d5qJ9JA;9RU;FnoKfqg74j(Ni(DbkK#64
zDT^-}?_D>)<<h{JxSc<iGi2N=yyY0QcirOPi0yEF<|KG2AjL|S_qi&Mmw0%2{cLYe
zwGKky&qr&5i4P_5tY<7^mciQ%YMb`WN3o`WuvzKazqZoXXi7Mib*Fn?icvNfFaKRC
zJ0M@@!+w=q(Hs~Ky(<PHIU3T60fzk1lJykd%Xp46IXn+#9q2R;7yNOLx!alb-neby
zGNTl;!S5;Wa!UcNIF=M|`vd*xI&-pS0NaJVh^Ww!%m<*AQfwP~)c4oDtG=!f;S)=;
zPm~^k@8uqYZ|yf%+f{z_VZ*%4khnW_wfBs6r=+fV%_=K*z+3fZR*aFfnrPHoQA*Aw
z>)`B=Y=6SNWiQx~M1ZZ#V7dA7J-~Lk49xf>G(KMVZ{SQp01Hn}(4Ixk2~F}4uAwBq
z%9Xf?K}SN{{Zw1c^8rI__0Ive&6oME-y#QS`l1>l#M_V#eH>-9MFq}=J6}7#t@<wi
z@N5?3Ns}kfiQ{iBIZ~m%2?-$Onct(Yo@CV_zk97`(3|!q)qpZ=yztt~YEuTfQ=q+t
zHeZasmJy@pGHS%D!Qo8w&NltJV~7HR&pLB5rfS02dVyLi_tom}#0JT!wJodXxt;~y
zhT)b1WHLC%%ue_+4_e6|k&}B!mBD^OAYW?`k%xmuNiWHxIYJ@uj>nXe8~_4=^Xd<P
zC8xE#h7}Vc%Jcpm=i`2*9nA7ydGi1BV?YN#A%iyv&H~t#$Rjf!qDG<=o&8k<)<sM1
z+{_S&pd@E5e2#xNsN$15Q5qk@c3iOrV~8R%x+*}!??vC3POpD1@Q(ErdL*0nUgB0G
zdzUB_)8Xf+?3QzYod;NOOfJ+Ek(NMMs_vD5v_!Zby}H0~)X`RPlOd=mR-Rbw!Iybk
z&@NO%l&m0)cpu%}?0jq<5+o}nvi5K=Q~>9$7X10A1Dy*yMReoCJg?U}l4^X06f*}P
zh*}R4V?2UcD+YOal{dvg<Yh_POw1yXl;VIf_hCwhbf)mbN-3UxD>=9XFY%{&EMWPi
zrM&f6o{g0^VQtH-aYN6QVkymNUeSW>>2C!tDKkk~J;Ug?14orsBLQh+M-^A$Pu^y*
zrbRrUVFRr`|0u;4(m*f>#`x~sI~C2BRIh20$@g?}UTN^pF43^TbBy(5MGHK#t@-LS
zD6I$_^$Q7Bf&iADA@X5_*X#6HNwRkKExq^9HB#zzfuKMD*@Mio*xS<26juvUMy5F`
z%0*_q(lq?c+R;LvuZSX*=z#Vzuy`wJoe`nC@pG6<Rt2YV>Jqaib2bmb;1au8+14$I
zuvtz(L?8IQ&x||XxoXzLmz`EpcV5;v|F?E89upPl(pWpxe@Cm|ZYAw`_hS=v`<lwC
zLKb}=ZVfjSkgBP^MmM|oTR3~sLa^#?{_<)ibMtdq`f<MCr)LG;XnMQ|Z-@fn+YL+F
z(u@qlX@@~$2s$mpZd&<VgtqO2MNJm;zQRgvzPFtyzg9dsC2cd{Xsr;ji+wSI9yR0;
z6H8|>G5E3ji?{B5<1Vcr^%Pcut3PZI(p=rg=}a${2doCX5pYi`p%gf&U`lf>6Dno{
zR;omf3H-&Gpn!YR@x-iVBtT<|5VbK}{k`0-Wass8t-NY0mrO9bxBXmfg#K+)kaNry
zawWAVpj$0ZiarR&B4xHMaY_VA?P$Ru>kcBG%+{)>As9ezr-kT+CaPQ*!707er`1=c
zR&q|tZ!WZRKI3XzSM<_Z`7<Wc4&9XuD?u|`X<R)j$(Y3f6gyMKHT~$Oo#|Kk4BDWt
zvzM8Hg_?{8#*zVBdK4;qNhxgJ6cnZ=4@~*K&SMs^h2iEh`9#zk)(k4CJiJwVB9g1D
zc&5214fy&f$;9nFEaBLwT8K)Q`P-pXU@*Pw7z1NUwmsFweU+kGf3bzk!q_!iBgbz)
zu_)&f)^)$Dt=9IkWXBxve}4Eds(iEPLpn1wa^K>F@JQWIw;(jmzG?DSV{c}I$&neu
z{Mvl^hOO&aO<{KS=HuB9QGp@w1hi^tkk8LYj1<wJZy))&zPEY5jHzqptIt3-_cpaE
zILskRs60Ba=oy7E<Xv@DKhQs_ar{QI+v?eNVo%p;Kjc0nTmM8)GR{LOV>I>QuO3rL
zLx3{kLw)?^%wBy7?lb&a+>@M*{L(X1h?ig@{Ud?Y+U{mGbw{O&dCm)i@CLzLSlTZo
zQ%e2_H4fuXZQjv}9{7>-GO~4<odR{ok3W4Sy>Kw*B^Euo-XmFF$FZHoQTX?_?i*~U
zy5+ybyNk1Zwuotir=<^;#`4GLk3L`VQky!HFqOPGIN#@qG8q2Yg8eq=Q*O)xmGB~4
zJc}QDIe9SGV|<#^I2;U}QR;go+S+$7DLjZOvnUYhfPHJYr*(BvX+={1drNsuf*b>Y
zs^fmjv63m_%Wj{cP^6aZ>d8CkulE&BmQpPwlwe}2jkK7Qb$f#dh>pqvg1>C#ZdYsk
zOrR^W5)hHu*<CUJ7)&5EI!$j=#(=S!7iJ_3k2W+1rhJAd#Wj`kXOI$a9ZlgjtBkbR
z$a+0j|E++HY+v6;9H*F-){cObj#`g=yWI1cXJPh7LLbY|h2vjjSa@PCM3oM@eKawS
zr|_mBv<8!hA83_8RL00kNhPOKbCnG;+ER!JaFvH{;hb*C4;Xp*+&r2M9wv89U0PQ$
zqWt2JpX=svEzc!;7b8X=xEbNirL$mb3u4@HmxnH+11Ay!^(;X<=D;_l=R~pSUFiho
zY}kyR;H$8y?P1p&J*S8t3%YsjwE}y2`(ib9D0D%JR7P0@TaBjQ8-kmYc@lt*=V**R
z#;b(cRI0RCs%0^CnAl|<ux)N;6{!w*^NMJnCL;mOr;r8D@SeJ?4&NdCS*h|UQ(i=Q
zs4}@H)4bZ*SY#$Bu13ZnY=D=sSPnl-e#fM#Bp>L=X%V@SrS`handu%=%hGg*g6GRw
z(DAGjyg&^E2)zr*V-Z_=RDf0Re#jhE4&6b14)miwpuj`$ssF$MAW}ms@^6Z%qV!2h
zYa0lveLXT`6)!uLGUk-ZxjtGpk~ZbD0sH2g@WKAgz7hXnOMyGOp&cT|a|8y^kcG`v
zRiZn^OXaIci5yx<6t`wrgi!f@YIzQxMW*z*g7oNsUrG}XK7<%FNNTQTea}Y_WxW?z
z9oks{iHr>G0%xubrHf1~;zC>o>4Xt5y;HU>-&;y?Ct&}#R|@;<8HKfIMXR0!gr6S+
zFmP*q!-`}ATEMQZzf)21(0!Q#Z!(TEE$#0>bV&ZUc=-dkYe6P<YQw(Xv1C^ujXFpe
zD{9K!ChfzSZIdF;Di+Nd(=DT)sieA!B_Lz3f<_H8y7;n+&1I5mI@fRnsqex+&dj+(
zOZdp$S^~DLogC_DUh0tYysP`kp7nLX1^D-;d_~U%gR4}G8E=_nI?=AjN#kUw#oNo$
zk{a2T4;*N6;C9_G>7{(HTi1hWccq4EOY|*n7&A~0@UbuZ8$tD~Pm!+6+=rlbH<2k;
zzeTN{ZXiZil%-^c3uhI`h%<-NC<!nad437W`tl%Y&|mh21QnOAZI-YyLrxVAF6@4A
zm0|&#Wvd19PZzg+WWn+LS1bR>tk)ua>_(acg~zX%bazfrkI+X3(T4$~xrS1SuGy}V
zuYGbRGBX?Y*pHDPm`tgcHt02hE>7r$Cn`zm&Ij)K(-Nug`^3rSs@!J&&K=BArK)vy
zlK|)t+{Syg&#74%*^vEK$rW=eOqRt>z)=Jmct1-w^6(V2p98u#^Lwe@Wu^R$wrA1)
z&&QqQWP-TWBZrSg;)Cb405DSTDt}AN2Y!HkD&f_+l*{<YyZHgL?9J-~O@a~Q8@TyK
zdyzN1ef&A?wZIy-3#>9G$&@M}$&jG5p(Ck?%vAUH<Hw|aLG0L%4}AA<hn;~p!$<i(
zFdT0^>S0(qx2ez@+`lqQzk3C)f8+O5b^E1|GUFCfw^~xb#Ub&ibo)lbZClXCM-+aV
zYsMz!+@Hp(#Rofd+68`N0!w6@vUhU@fAB3t)`!wNy)|I|p`*wWv!UO_P&Zg7DlZKE
z`p1-<$IvB9M+WO91zN-G>BrhZgre!m#&BM1+?mu~TWN^n0bY9l0i93<ILMWG(~ID=
zmPF%3GoXBURQR%XGE`1l_ZAvd|8|jXj8vEVkcM32PPdAFrEZ$_lJeHqmO$KKU80DP
ztg0(<y5%#ww=@~Uv(l)VrjB=xeH-djpE62qLmz273y5ty9Q0l+Z8(N6-h3GG@Nd71
z)wC*KN&~Q77+vJ-p>nZkJHr=dzY9UA7@;F`ol!p{_;s_|0jIj-uLw%8LZ0gUTDR@N
zeTQ-*#o3?8L!$Gg?8H_JLGez`B!sN?Jq6bQC`_58uRJ6%(4y<o@6)h*EmO;4)12PT
z&_Bw;h-}PpPMfg7Vy{L+5hZdYo49aSH$1!=uu7lpYc{IQ71Epsm0hH0{e}A`KE(d`
z!-zi6TvxInYdg>Z6_R6{2EdAHR9;|<@sVP4=eV^f!(-OXRu+hN8h7{}CsKNjSfa7y
zD)1zKdkPuFc<3L7p2lqCRR2j)>-=%yO@&qvEU>Uw>%3{>o1!1n{1@=*L8~B?coNyH
zr6j9aUTjRZFbF+a<%wpw=b5H1&mYW*UweI6wZe(L)nVeL(xA_|ba_&&bLFwY9S<$n
z+iif4;ti(FH)_O;xZ49+yH}=Pag$|V%0wfAQs1@0>uWw9Y%aBT-s)a#fqjP45QK1f
zeJ$p_zFT|(997sHg7OieTk-@-(h;@sj2~2}AkiN$m?ogSH6JADs*alm6d3$2tzAH=
zJ^u>NUE|u)OiPkB0^JD!zK#lum^4xjL%WJ6_Y%%i<NP~_gi;;`e`K*^zYuTPeVB3V
zV-YT0aOY8TexKp`Yt9d^A1DKbS&gmT&^FiOd(c(7sJXY3ZHDvDiYe~1b@9p9pkC9o
zWHv$*)P?!qTVJ?;#jtoA=->^=G03H2Tc*9cNrhwsx8w^cy4VP06F`$Z5io1f8lq@a
z$_bYM7S67;UaDEOUcwWFdn?*LWT4IeVSoKoN!`oe2UZw&nE7Jv^St-r97+MC9LOhm
zg5WSOVS5`vx1kSc2{LB^Ec4SP%%~ll`uPFT0EPE2AeGak)(T+j*_VM^=2LY1jiBcu
zf((bCSMltJRH=}DG)UwtcPWnn=*}5_qz-LL4hWTCrep=`x&Y{yf;a#0mCRU%A>u!A
z;2y$8j5GHs(@W*0!lTzT`25#Wt!eu-z}^G)#}bl@K6taXTl0`M+%i`{aZ~yB5Wfhd
zW~a^4`MvI|{tyqBNc-J-d~%w<;kDims;K;=I8WAalTk<WTaQ5M=Tc>B+(#L190``{
zZPB!CkZPAr+c-U%l$WUJqI#FP?^>o>d$uwIe5O<(+*y71XRqb<2$rv#)O%3~nOfcm
zVe!wWl;PAJn4Rc+YWCZ=bAdJx!^9-QPAh`S@5mHgHb*HMk;C+qK;nf#^578LrOM`3
zEvba=U+E@-_|agd?`*3dv{{4&-HQc=hrhuTHbtpV770x5xX8K$orUd+$K~8ddem7B
zwXD+bwlO^2IV|EL+^Z$J3vr0t+qpzwui(DGTU!?~`;w*nKMPjZ3X?XM#VIZR8-)nn
z)nu`xkcPR1RTM@kZY2(mz>TujCX4mHQJaTe*6X;+XmQ>^CRNcfzC0Z^^Sp&R8+U0s
zvmlqj>}^?SO6jZ>^QC{REXdr`OZnasT#nMH9b!L2CW_w%U5MF=UUzPiSG^GH_{J!(
zS4Bv^Dxob`<F%jH6S1NF{b|-n^}^POPbIbf4W2{U$*oSK2Nhk>{^=0YqI0zMgCKb^
z&tLxYLgT<6bGUwKrKTJcj;38&Nh3VgU*6No_0w$J){q14US7U@VK&V?BC_<u6C$a?
z-#$Ly$iF;dr}&)nT{#jOL;J36?3Fqn^1#m2u0lvDJg`b=fd2(;?jVR*FZ#mOPFG?)
z#X?i{NcqJ*5SRS9Pu3XXVA1ScD%uC_s)rKUcZQvJ=8KoU<93L>rRpMoTrJ2fdpI`7
z$DJ2i-O1HLW<--@9O|fIr1zqWo~bxOr0cE-s5@Z32!}{a_Y=K>Uo9`&LBU*E7?lI7
zIA7q;)9`EMLk2~6JhG__BbMe%{(;O(bh=GwrN%~?jh$RGuM7ScLgqGctOHfNwieEq
zuCmo)?EeE+I2QqhsgC-`F9HJ=<+1f<dlB|a<bY`9`wkIpT7@%YjHGBjo;7vfh?i@C
zDi*esWKs<hl9m>k=_2RCkg$yZef^!~)odRD5o+cw{Uij%;o=+BOlsplo1$$2+v@!8
z`K1%F_eh}z$4d9y?oEj?+c=6$xKGqmZ9{z<*v8flPAZ#MCX!2n^SMS*yO#+k*Lfvs
zKDiP{hpnRGe*G{TyH)?UHajh;gVA|QuXAW`u<6#r&+oSjq_jc|bob<nAFJP0<QH{!
z0h!Im<QM()rB%6VB0Ufj#Q!^MXqVAoO_uIe3fop?>31n3Rq%lvYLj?oD$tOmq{4HT
zw%EC@(LO!=ALwYS?_9mAuiBd`gpyD5xrJI_R5D1D;I+RQvS+I~KGYx8(l&rzsjjoi
zuAM<h!3>}l;Fo(<1K}a4!gl>IymHPj@yhxJENMdAvq>)%5RZcUmF>!bX}V9kBCLz?
z_vjC<EruLEkwWM5O}(_@Tmi)0kG`f#IM5L=*Yc8DKQr`WaF4=S9m%SGsL&>OrO$uU
zK{`WDUe)s|M@LjzjWzepI;e)8ymc8loI1?A#VPVR+HI(O2f;XY#1`15>eD*~id*5C
zSBVcDSm|$kX+GN@Ac(OUPC(G8;?0jYx<EmZV3O)gSwxFl|23k33VS8#y$ybhtS9sz
z$nIJ&%CO;yk!1f<o|gJvFGs=V_jFuSG960)Kul$sx4f<eHE`rrpNDU|(TA5Wq;%9!
zL0;E*aB`HGPv~q*YbZ>sU-)oKhx;2&2rgU(F36bP6H7Ij$ZuK*&_n$JD4-kWU?Pko
zHF$x<^d>Xk&r9;^bO!tVV!pJ3Qh4_k0P=j4DpuDfDFtK#$38Uc^;YoMZWq@sEU7(j
zUyGgpvx+CsXJ96taWbv3vz4x?-BBpi49LhJ;R!((qR|O2=_QRMucvfp>6W5F6T8PR
zFG|Y=eJ@LH;|v)|ZNi>dNhk=^4jtTNCapqV$T}0&cj{~5Mcxx?3C^N~?q88%kAImw
z8ADF}ius_YCotcfJDlLrb}K;7)f_KcRT((<WZyMon9-YQz;n~t>ZwvORSvDF5s1AA
z8f{2|W#~CiCK-!LVU%f|!Lq;^_D)+6Pd1UmZsQs#E1+qI6{BH<#OV<badLD-^awa<
zgbg`MQ&i9q$4gMV-;qOOpj%9S^)sBOBl;?K729?NB&0ZM0vt|0E=4G8xP^uQia1f^
zVgfF0GtH%wRN0ilqb>POO!`p8Y!6pk5Du-6^$SIy?e?^`2q3;-hM?2aRVX9}u$veQ
zVEzQQnLe+&NfJrf;c6%=<?4Ag{3RrbjbIg#NTsmR+!k=}YCmbs^VZZ#q8R9K2wkFO
z@axSFZlv4l!x-yQQuWu8ABL~LYuSBHPL!{5^-jLar8qwllIm!@ub(n(nOFm|&)7fi
zauYI33=e)-vX5MpI(e>EtV!VuZ|AE0i>B?(wAj0G=w0bWw=6ubu<4T%GW+GKFu1TL
z$w$LRP2@V)MsY|mpLeHx8Lff<m-ltb9#d)C)DiJA>gy$tVnLcurb=t2)k5EN-AZPj
zJHy;+dk?EAIa^}vWkNi(G})JFsoRx<2+`08@B1AIreydFPl0{H7kVNb1y2xVKPn@n
zgdk*g5YUxp5P^Cv%V+9H5Ze+fe>2w=N!OV~M?@)~!f}aS-?RP*8d-xym4%q-fw6Hd
z!!Ms2_uI2f(>WpYWj(JVgO)iq;}n*4nP+&30Tvl;&kAA8w(PaioN+Tp6~E*L5?~ex
zW2Tsxo=ZzG$eQ8+(Q{8}&y4$gUzFYw$YSKe<k*+$#X+xf{K&RUtb1X==7x)OfR$Mf
zY)qd+$R>H{rxk0W%42brT&;=sFYKC1AY}Geu}71e9PeKtTinSaPt|^vwO<$oS!6%E
zVOhQwVr>+zJSlxk?o(7p`KJ(iSKI?H8~bb1RFPSw-s?%M%{+5}PgdGX_~8EQCCnMJ
zwtJF-;^%eD*d2JXollUKDTw$`p>m`vnx-qDcXpgV@YxDD$uH$3dI5>)op||xB^b0O
z+vCk!zt;<khO9V1d(&J(4U8T!sBZq29pPoxcy1yOvwWA>T&lU!DZhMR3v$oN6|13C
zQ_Z|`MYx&ZZYrlht1>7mo!(p!QghFCMKjlKM4{?#waeYC13_+k4?d<Kq3HD56n~*{
zrIf+*c&<<N8oIt+*03XZ|5R$Zm!u4KIS^{xRotUM*O5e_flM{_lj&8dkgUWo0Qfo3
z&%di;MR7!EqlB53h@V<g&b4h`=9vWNXq}%#Tbxw7U5Te+nXY+|*~g8$IFG_O<-paK
zF-L_cV@h~|IwBg}2%u4Q(5o@8fy7bs!XbfIo~Ufj0n(6ED8V6w7`#eL;WWt<ShTsE
z;bN@KU;>^&o8o<V2(3M^DeBQ-MHf+1m0q*l$T=fWzHNB0CvHL5>sq2?gRe8VxB5w{
zi0k~;J6HKxw6YK;<$FIEK4|eR7pLn5OCP)|rCQSRncRzib9eLVE$6tWJ)2|nqkF3j
zyPYjYxk6$8KoeP;WZQoxKF7n&iZ(<rh4w=S-y&2f$rYlww%k{@{z}j7Q3#(r<!J?~
zlI6)&W%7tJ_Ki!%5w-Vq|3Ioi19R=YL&*CEnOmlxY6CwxUBd<15fpyr=U@sTI%hTY
z;csVa)Nk+)dC%?LA;8-hyfQD3@29Jkmq0QURQ`dk=GV<GM|e_j&u6;FBq`Jo+d1T>
z9~l=(N3Q<(?^(&#3xnL{!Rba})|mRGDsLeXy9<WdpHk<#6RBKk(T9xhXOgXcD;x1%
zd`mmAp4{}QC6>3tGBz~9eXU+^qkQgPWY2vb-X5e<OZ<pi)nBWocK*RCTgNgx1d}X+
zr{5U|nYJHuahjy)lU{#F4zuhlu$V6yd-QWO@bs;)6N?nF`LG<^{dVWZ=QiB$g$%*!
zjEvFg?8~8F*;eu0!cA26=-b}pD>fzB>{_*YBQ5!~llOlDhuEv$Yoxs_4`qQv#_gl3
zHHrafa-NO!Mx!XNSn#sHO)9Kg#otC2^@d?3>kJwDK^6uX7*LK<)Vy)uAQi?fh6r7@
zcm{te3Q^4r4}6U=(0MdS2xc!Z9A#icBXo%QVy#6h@+FPz8Rf2&-9Bm=4s{L6ut$3~
zj{~-hm)@esmdFF}+@5S)CR=(MWk+i9N-%#TYQw+o*mIzG@!!CYzSC9jl?Tt~-nBwo
zC%`}d+ExgO7J9js3aDGg(qdPpk}G5k56dgH_U~WYtgiINGOe-=j)R+P`ixV))y7K7
zBLQb`ccmilJy~|atM<R<=mk%TqFCibssn_e__1JdeC?)pDhI71ve2cDspu_#b`0$m
zzB5yBbV+(-6TzG_Kb&+4k48vMq~^;Lx}~s?dO2AEcK*srar(G1WGG!U1WY2lemf-V
zXdF<`5%W5`z5dH`U?q?|?Xs0DZ^APU#s(PnC^Cquwhr?y)pLapgCde|fVdAm;s!wQ
zgD$D2-k~i-LViKGEN5T^LTBKTl`c9=aGa$Z77XJ^PEj2j5iU1#5V)4)1kAVqec}H`
z@#=ta7hgOSl+gU@6ft&7NWn5u-enfNl~k77zxMVve)-;XZtgp(%V)c9=DROfLQGb4
zLQr&55h4%tIjpAr0E1zFgbBKy4|$4@^%0dGvC^l@HwGP5ecamp<<!87U%>TdC%I|p
zv2xw3?S&L8t{&4VJ?{D{xwcoX_h9Mwt?4jAr127t?-aT4WMIC<MNffCk@UKURyshF
zg=Gy)8+n3u`*dBRBTXZ2Rq5&mZhT~$BIB8JbZg9}kHSa(#hRcr9gX8Ni|2yQKJjfv
z!wX8m#p1+bI2eh__qLKZh1@rA5-~0RQg)Et#TVTj1}$Xlh<Sl$sApGL_afsg2_Ivm
zA<tp{#hnnH*&YfDFm}vUwq%hPx!wFsz9K|CE{%k@$u;&=Ue8(Zh2f94ZAd!K-y}H|
zFRlb?=fcgiCFJ^gj-h~uz7_CbxVjD>rRnHugfBi9pfY)tflRy!{mS#|HJGYWpJ&iS
z;mnl6qWeo&MSl~_R0`8HM1^YwZa~E`V0P$ox#%_2sW)lKbH?<WR@&a!-|^?O%!8Ef
z12kX0EwQtjpKTAk?{tmGnVD=X_A<&vT8r>0F@y)0Rx%4a)))5(9Y|6kdX+7HzVsRl
zBXl}8nC}Cl<@vfqOAV24*wV2H3%4Q-INjitdzQa+*?Oyt=$PN4{*YJ&<j+W8aXZY-
zrvSm3wX^SU-|#uCfie(b4(B<U{xYOZ7G*HkU{dEu3Cc3-tz=OYLB_;bAzjKeV=7*v
zvx;ALj*2|`jWyz5x_a%Tv%kHXV8ip{A@ELh<^SGXX*7~)Y%ms4L>W(GlBu@)jTybe
z-xCu{Kub^Uqr~XNR18lEH~P43E;GB#i-hw<q=W=|>IHdn|1MT2h_x9nZ6TgN8yGt;
z+}kli?PuO^4ebIMwoiAL<*+ahDw*5qxaHZTEUwMR6lu?`yrN(<B*RjwfM)WgkZ35s
zS}D8;m`p%SpQV%);_-n;g%oE&g}a_iER5h{z70etWHe1*Pa`gE2;h%C!#$Mg;sv)K
zl}(xQ5nv-f(h#|v3?^`q!i2lmDDuZA{wU7$yiX4X%Q1GVwTKoygIc~<n{iZ3U&?so
zBH6r=Uoe=v$MO$!)7R<QB~JfB6z{>#qA!uiLH|J#y!Y1jeF7I9!t@W_BkohiN|aE{
zhs@phS6jA6-SwfxMag|4DssU*@j;G4<K1b+t?enQ_5{rz6EWO|D|xx!b93<7Dg%M4
z^rX*qK|$%3LUd7i7emg7!L`n_ccj2DVTv#kc^c4)6ZyTu;hJ5hkNfRQVjhbbJrK%C
zK@P+BN<14T3Y|O1|ABV6>pMetkbfPkBo@_L>e*Dc4H;~fNsm=C>7~7NfrGlezVu#S
zR2XZOKw$bENMl<#_baz+uS&mqP@keqX}E?0^ccd4xSc79s1MaM7eWh9I(eE3mh2~E
zTa_*<!;dTD`3J!n1xt?%5a^t4gk}3WXxrWc`$YfPb1=i{ALx_+-?jTcxiiVuJXY4_
z355`{^U4ur$j<{`#^dV6$h~epi*tdA4ECw*-V@o?dBw?VXS+7GpU+g9PoH1j`3KrE
z2g`Nf*e8Na-^U$$MtYgNEiCFzo8^ksO^0B)S-9oJAl{(09wPT4xQBNEs}61d9eDey
zX4yviM$1P=1(YhX`aG0NX3Nr-MzV;zsRsRggV2X6r7FJfSU(t;6b9o!FR=~qazD#g
zL{_G#n^Q!)=gYIfL7xo_vdN5;VI+1=IptB#e7q@zD?O406N`BLP(~Ojj}*ZADEX<4
zvd-K^ed+!OV#RWW4crR%PqW3lc;o{i5R5rfq5lM4UKvoHFy#&@EM^|9dq<n4%5(Vy
zbl^Nck9A^c5*TyahP<x;o3JW8ziSjZR$DzPlGW}?2}B-0%0!Q<aHUAbgv@CaRq8dL
zZ;kQU>p<#J$myI##WBV&Le04?0>ZMKaxim|a=)$|er)N)Te}h{*ed$HA~FtJ??O*X
zq7a20{ihN_kK4uP6{|CoLFVB(y=VRt7!uo6oHKCA?}p=^rkZe7=)@JS=FEkbvk|u~
zmkXWUrGGsZTiD*P!lGYAZ{>?@XBhSvw<(%oLkmH1+xi(vP866xI2f>7n@5O%gmQ@c
zg<5VYCbo$=+@}M!R|SIEH_W62Pgcva;h3niO=<KJyc1o_O8Sq(`5%>yB?<I@Xh)zT
z>!~gGaH>$uu#Bqn?T-PksQ^yXJ4W!=&K?>JROm_JoA?YP<+r>S8-q0M(Bm7MOdH3$
zpL3R#j@TAx-`%rjn}*ciMb*O6-zp5aJ9tZ~cRcAwi|y-Ieb*~<ZNj@<nyo9DOZath
z4};iOa`wI1J!=N8OJHl;Thn#u!mY(C<|E!b5MWA*G!<K7mQLE&VN@%&wH9luG}lQt
zLCBCw;y(?>zarZsvI=OsTEOF<%R|z=y*R_BipX?}!GcLtxf5_M^rT(!_v%(B#@yP8
z=FJbD0(Udodx~eewI6fPN-B*^0tQs7bZE?(#ptT?QOPwyDjY(Ah^Bql&WtE&RQR%(
z%6td0_PJU?9q2vK8mp9ZC?|np^{~eBB^AdiWmHPOvd_HVX!UI4#~$l`EQb$64dA}~
zSk0I;<U?nx*;dyA+i8h`H!(4<Q3bjFxM~S<^)}GZYrrai<JICj;BDvf)^nvYR^-ll
z<9Z!QACP96^W~-dx{RS#fNr~rHZpKXFtE^7qD{S+d3j${3hSAIU{LYWVD1Yn7iEkC
zwi|(T>+MW#aRuo`iU3l<pw#r6)okHsH_#Jqe}SfLTs=}wCe1FS2Qg?Yl>`Dk;;Hs_
z71}2T+0+^BH0_fuW^l+oxVblj53p<4+cpK(@*dQ^@NNRYr~G+ev28xmFAPA7$#I3U
zdhU)bPd7&C_o-z&I%DtFbLr?;7izt*x_kLjp!rfFU&`TO$^&!W?eS*WH()O)<*Xk7
z!MS(RvQk{gXE%__*wR*>#Wwrp<K)|i`cqE`%FgHs&%RIh5><Maak{Re-x+6y-l^rW
zwQoupmajkSzV4iw5ksmpTzxVi<d5ei`03S9t(|@xD{owtP}RPSD3jq0QznA;e&FJJ
zAJqOOmUyy#-6z98RDl$m&`L<Xc=koDoqB5rk(6OT(*IOFeWi@mq(6Mv$H?L3NH8`^
z(bA!^4#1?uz9;KBJ}Cvq_pZiv3jZkPaWn$)9s*?h<zezz5#?bJ1J!?0TBfyPQDv7H
z8k{kJTfc1&LnS#c#|@Bk%*W>V<4yV40!zqFX=z;JFxIFN`{z%j1Ob}lSDv9XBbZHA
zVLn*{P4V4u;UCSZ_yTo%>*u$xTs8*ze-sEFKcn@aDP0KX3#+MYSyc*}f2?#|{SWlq
zHtm_o3lmI%^v|!7JadPIS9Y#U-Fo%L&MxMe8>8^pBxK3RWczGJmvfc#(OlV)$^a-}
zlg0@eq-<UN$HRcbDh+zWA6}sR*}{naXX~Sih|l*t1r!n{mw6nIc;etcwr>XC@q1Dk
zHLt1-LdSRZQPxe;`_=M2BDFd$2B`FZXueVWn79WYeqK8umokc@x^3&nT?!Drc%tDc
z{NwcsFV|kKJm>D)axi`!el5>gc6p74Wtq`p%A7?xHR403?Y!}7Q%gqaiM3iE>(+D-
zM$z}_o!|wEWIvPgBZ726qHNX(yrCdY>Ad!K`O#Z`+&!VJ;@lU#d^d7V;AG6g^>hfX
z+i|U`XI&?GifJA+eFZxO21kgBLb2$IB3g*EeyqX|_N}kFqlQe($9>Tns6V;^jt;)`
zlh!o!VvhHWLBI70G82bJ{3_S=?;=6pzDhSf9=}ytGPBu`>>C^%t7-+#QhgJxTzY@x
zR-N4k=vK=eHNj}Vj`zmeAXQcO<Tf^Il<!Hnm8@~nja2!C(QYcm@JXpLr8tCg5CHY(
zBifH`>UbkU6-wjuAg)1$`=Nw(=cVb?Jpg3rLqDcl0$gRebIa_P^P4I(rGpGJjUP+v
z%H~TdsQA3kC&ndTieDFi=e8eJ8>{OuvXzph&Bxj_t(8O!+_v8vfb&U85pAQu8HS*D
zMBe;4aZ>Be(u-&MDUY|)P4WP=5A*0VTO6s$@)i+9kJ~qadK;B(vD^uOMTud@21msv
zS6Un8+YhWyCrdj6*nk}chk^2>ak?0wMq-^^;E2HkuqOkFq`4<Eu|#<onQa;dBxTEg
zPdX`BTI?xXkuUf<uqRIyTCz-HV>S{5V@Mq25u$Yo@U%wU3r~tP^KRirwB`_d*Io9j
zM@zD{7BL$dT)hv*C{<UhQaY+(qqEXpDhG!BA_J0S6R8)TWRzTg;Gj4KWztnFB}g*e
z3$x%qjnm-sSo$o>spgEg<hSWoD9M=m2U2{alOGO?P_(>Tx#cNcbL#v(3{FH%iq!ny
zsYEM6_8&VmnpVRfLG7T6PPm&~N(aS8WJfns{z<oUWWvBweVF?`I4Q0T`)lx}$|&BM
zw1#Y%!M~~QqAs^JQF@7Q4R%xI#&gIlJ;o_CPcqbb-k<;SvL_^^sB}$=uoXpJ#WVJp
zt9xgk208QxWDD?Q-4#g+?=+^O0~3{L2`7$DVE~4Nr)x8!!W>LzK7OU|m$3j0B(9*2
z>N)pueR`Lp#835+eQ^0Pz-)q2e|SV=-^<&9eLC(nH6%2(PbCU{CKGS^V#*14yy=gL
zy+?$(1Z;|Wt#-;_>XWrnnAJY`4|<YS-{Do))!U?46(d*EAJxLfnP0AWKYqWIM3DOO
zPBF<xo;Y9VdRt@}Y!=(^NgjA}l6`x?<{MKe5Z9x&SQ&zm+*9dYA=O$)FwTiHtZXiP
zR1mYP3vDIqg?C$LKKc7HZrD=v4QOA=Ry%csBK6S=uZ^UE{E?*xr7B&%*Id6bbnCO*
zDTgD?q^p{^Dv|CZVZz1=JaQYx*7QqpNpJM;QM`x;7bDZjX%qxa;Z~yA?Uji@e4Zzn
z9yKxb(WR&@(FH**cmf>AcO&glF?MF%2Dj7{zrC)5Ja1X>*fMRJ4%_sB_uYt#mq0J6
z8FaO#%Jzkv``EA>K0x8IZ8Am*V%D2Jntay<&#Y=*(S(53QVepT*jVa*k7x!zg=)RX
z9od1K1I*sCzrBOFHh6{2aG|yY8G%lUn!Gy@L2E`7FE%)%2?CjKh@l@D+2&Z?R_EbW
z0w7}}_C8j@VrGoE1r)fjnigXxbzAmD-^5qdo`0a0dVLGw5^wA`zgk}7$jn{xzorjw
z(Ud%L_y>~g;ffaun24b26*@>zpzE_#VC%#9H7}se<q68~pN}Kf%7LV=QQ8CEI!q{(
z$$Cu*NQ38oSss?sazw4cTymAJJygoqEi9ytz9KRxsbrPc;-8^Dsv+l=s)W1$SS&^R
z%JI49posy<xDM}RN{-VASob7@kfJ365L8m5K)fqpI+fPgi+WIDHHH?#C8}u7Mo6ia
zj-ggBLE$u*0OmKaKA7GPV;({ee{4(Vp*Mm)39EtpG=hc3zaF3B`SlM(UC1M1=kg;F
znEOoTq$LQ0ftciiZq)8Rs6oyeQrY`KP5mEevO;cCHtKmALT7!!iE$(hPQ4xUs_Px;
zVNDBE<N7Ar8Y}bi@vq1Ff)}0~a!<HQFFxN-bO?Z|wMKm{M(ccd0^b(fJgUtZT{%01
zc7}d;52O3&hR+tBZ@c~TBIWzxosVu`XKR6~LmTW?z%h66@~aG`8oXbCC&hOE>vT@S
z$w~2-^9_ul)A)s0*AbTnr(U-qmP|FS4HOP%x>4V1swJ6LW@fM3nTJSd7-fDZwc2|g
zO`!OEC2(Bazq}<@e}75%sAh{y{boBXJ=`4JbAGSt&f6-z&45SDqD+S;Rnh^56tMhp
zAEVj=(@#kx&|ST$Ge%oP_?_I+u41pfdoyLUajUAO%|*yD-@`mryP51$Q$MxVa9TtP
zZ0v~U=<ez5n#5acEBAk6+pWVF9x=WU(J2ir*pOzNE5D>9$lUF+{c(H)FGI<&BSIFT
zi*75uKjWS46X3-g_^_y@Koh~di3{W?elcxdCiZpZAwcZtsjn7W()}E67t{SX<$FBH
z^#0NFxj5QwC!UWKS4YdMyz`)UdHqD^$b@mLiqWu3B(UGX%l)Y!-tf~nui6ZJ0gxHD
zhj58T(IVhKQ{eM!;50py2`lH5uhqN)&e$yaT!jszE9{C`8uk1vyrg0hFk||Oib`r{
zG_V)~Jr=jX_<;h>MiXR)iYZ>i>aum+zJFOSMa>{zN`G|-I?SQS8uMww>1j%u@yQwO
z5!AFcS~PDe=ylxU9KyzP-)Uuefp)(tXfC}wt`PbwizlZ3%Z1OVubw9qy9CfQjLOr~
zy0nFhod!m1)#H$*5rqZcim%z!raW*u+c&AcuioXE=o*baDz0Fnl>zvbA#95{v^kkV
z7yDX_Cfn669bJWT4u&iAQc2qM-CkUe<+t>xdx#SWOSlASsuQ6~L!~s*Q_5GST4G9L
z^W&bY*^4ZmU%I_gg`#-S@+w}@xkp6`TS*|pJ^djQTME91@>DrPF+;Xqls%3bXk^$B
zsit$*`O#ZAkhB<1O>jDbjf+?7O#i^VbpGxg-P=0uk!#TP_gy%6z+FLbUtjgGulDdJ
zYxDoxoP|bf{1@Jn!+uwb2XW>cTKpL~$X_rehf3(!?`_Oy@?*aP=<#@PA<)i1@xQCj
zQ_;xfx5sWWsEK(Ru6Sg-sR@2$iwbowvMawY>c1lS4-`%@@xU`G!|AJZI{|o0jNqRN
zK>eot1@`7*U3SSf9<I#r<Wa%BQ%m+MbHua#FDBZ5KYx2cX6_E0?gr{>N{CDM+S`v-
zKKc??All#<Bi;w)qiRbgri>JOR{HMzMc~GIBX-UhC%s=PUZ9sU?8Y($&Is2e$=9DE
z7lNQk8oJzzcF~Y6I}XdltciP>+n1~=TVm<&Z;5R3W5kluBDNAg1OnrZiKuVUoFB)4
z+cHBPy|bT^zy4~p!h(gSK>Eh@Kkp+dsG8k3{#b$z`<xoQI~Y1$&B^Weh|t2lA=^~0
zK~Z!Xf^5hZ#pn&f<cG4AWNgY9Q6~}$;T^LmR!w&PbU;6$@X~KqcXS%(nu_9G`|Mp5
zspXdr0*otl?D(|I-7wQ@{=Lf+DCVYji;amHJe+`5w72qm_mz9}A0GDAzLx}i@|Za6
zoDs;<@?uG9fuZDW=ccKDfkC$Pth0Z#)M15fG(*2Gt$Y`Mn$GmiNf6mi(rlt+<vpZq
zjGGIkoCkLG!<q^HyiFxSTyTIbLq2KvAE>^)E8zECJ^h;JY=$;dzq><aJ^DJ9=~t`z
zU5MFiY}FCB$#Tfm*#@*;=>4#~!p3!AU7eaUo&#uw#n!S{-YkX(S28c)^jtRq8$_2m
zSfDeGLyHh`#V<|AGSu<HCzRhe*#2($gba+{(r$;x)<t({Zp*f)22W#33H^yZ<f37q
zI4-`FX^E*7%(52yjIPY4iEg4dfnpUgiE($#Ot|!i?xVdaEhW^aS`X&adWBS>k3<Ie
z78q0pBx=4r1D~WvtO^$v>eV2{Dj4)unoHjGMHi{YO~j<DB1EW9!=iS}sAv1zJ+LN)
zg%CUK?3*#R)J<LRke?&MY}|D$h?1aW6?KXcU<}1THXv+E#olvBSZf1`EW4y$h+Jhb
zaXc(UlgOlu*_2(P0pd*}mLr<MhO5wHjI=HA$%u~z69k<Al6#nZFKWo9D9cwEx7Ueo
z1g`sZCSSE~WWXW3G@xolSxPwq6c0pF`-qO=*m_-D%H3U#_}z2YC|H$|>mQGkeAO|B
z1Opms;N^%1`Gj7gu0P8nBfb)W?7o%T!GR7LpI!Uud;8Y&l^mVY4fAvDmi@X{;fSz&
z!?a(b&kt(x*0rHdWItbVwJEny4F)VCi+vyPI)a-~r`8~2Mrb?qun-owg`i6f+j409
z8t!lwKkyx<@ydGB6FfY&uzU!^?Y(6JFg9-;?60P~?k{n-FPK8A)~|_|tY<e=0D{&=
zU=?Fta?CjIXYaB_;*E=GfH}Gk#!~_rYk!;*D4+|i_H{s@w?4wy?R4c71$j>b*TIff
zTR#$UlgA0g$MQUPj!Pk2?0-q*w<*^jRctyw*|>c1;Bf@|N7^HplYjd<kg>Sr{Si@W
zcdhKljS8TH*XkpHn7kEu_a|Jz^A)+?U*I#sVa2|AEZ1<BobE6!?>QR@$k!N|*57Se
zT>&o`wsu~8DP-ht{gD4{d%K2AU*G{#_#}$DLxgKAO02tlMC)CFJx54$4XO+TU)~Hi
zXXiaGM=tmU01mdb2|R;|050{qDnMzv<tLMrHuX?7lN%pcNF`N*Qdmu;f2d;gKUBSG
zIMnYS_dR1Uh!IAOea%|3l^JFjgt70t7DgD9WfVpjd&Jm_NsR1dD?}&?BV}o<Ns@i7
zY*|vj`}@C+>pt$|e&U%2bIfvn&(G)me!U!X7HW~H{o~SoU6Ft6VVxVV{?C0!>TLZ)
zPGEQqvKO&21Zby`G_$~;sf?BBO5MWf<GuaskEBD&rRsm3t5OUKd-qe>K>J<oU89C_
zlAZn2@OsL{*P~Untvb^8CY20*8fg=tDoAU+{wFWKwFYI3RkXrD^~f~K4k2x{jQAqN
zA;aO`hu8L&Q3S~}gq<9u+1a|+hkOSKBivWEG%>SMhE7lc@${a09-V|3nd`9SMpHA%
zYrQIsxMJF|O%_2}5<e7PF6xWuFF4K4Z*z@?&+cVs-@LAyKzBDS>T5Dmj~^D^HWDq*
zm-I@#F~Y>xzJ_yx?QKew9hSFq*J^Pr`maD3Pl3r8)yF(dR_w2@J4$SzQeQeGE!5<@
z<7+?71()XgMbZLJoIzdiZUGI!vfj4|UmGp0I_%G6w(qHbXc$&QAB*li?grlQ)fAH%
za=!@SwZ}2zRN{Xi1DR|f+A2?OciIo=w_c(wr7z%E*h&r!l;p`kU+WV<AnH<-bqIV;
z?jhTzl137wvRYHol%7BMGMkHu(^WBFUpO~Dm6hJ+III1RUpNElHn@{LHHr%r&%)hs
zSW48YV4g5ve%8fLhKtRv)3?jj1j9=oKf-s<dF81_{WS5pl1Xn}A*{40%vkx~V8}rl
z1AdiD%ozb#?yd_PYo?OSmbNF=*!X@s9)!UDQ#R)>LE_nC9@+9jcO67Tt@o^5N5plf
z+Z5EoB>U#h#+gVXo}Ge-@0adEi3a4fQtLur+5AXcg?ona-Lr=Y`YY(9#p(t+osaL~
z{-A3L&)5Z!vs3F3i%At_93&??w)K?E2}pz>Q;gh!#S!N7@ddVbkMERAYFy*O28%&}
zHo%Y}`^uWN$tgKVsx8meL(3?lbC%T1#~o1{?IM>6&U}{Rs)W-ZWSy^nGTUC={TBjU
zso#OhZO4pUjd-vb(=XC^B{)}a5nPl<Wi|?LFyX!EPqkUXS%0C(t2_<Ki|<o2e@ws{
zFH#NIy#rpm7ffO^Zs}(By4yd-X7t`~oD`ma{zK|q^6<6s0_PI7JH~J8#`k*hXEBXe
zt)`aqV(xEE)O~5Y$?^GjwRqTD)XH|v1Sd1l3Q88bvQD#ZbQSSkG<^x3Xe(>|V7!Y@
zp{h`F#Lr)pq6W?hHnBoeex4DM@MlxSeg3klNZFVy3(*xEO;&{gYfoXAii=d8TyQxn
zks9^2Jo5(hjk-ulk*;y3WDE8~cD8kTKIij57U(%$&GIhj;QNYhvBz=<PUsTVNNR>U
zZvc{vuCai=>2z2XXwU4X@i6Dl#j4GW7PAB&X}yY+L*UB#?~CPzYp={^lv`Cw2`tC-
zUVSK@nu-@u=mH6Pml-%ig!f88RIl&TsrXuuz+8&i0nNQ&pSC?g=HxItjfBVd3X};4
z`+(UY-=@<aLz+RCI^V1R%*=5I(ZKgPj7aWivOHVh(S5qLJ|4M+<>|PwgRC6b+5n~{
zZ*gm;D48iyi%}}f_dgK(s^<Maj!6adp+jA<M8yFML&rLWboD*Uq+M2qbbQgiO?JA6
zLSd}-@HP_woZ02s{<-TFb2dt%>J~5Lvljpb7*4vD170fxV+#eD?}R&1-yLR3+;VxZ
zli(o>oX03Iy_qx|Td{j`@zxBzjeEtMtAO5<Z|eIn@HZ|rN*cIU)uavU0PLeL5IBuE
zx(1Z;se9j&TK-*&JHN@cTZTM_$w3@@sN~TQ#ry;J&n8Ac%)=`ZHZ6+lr?G}}VjrVv
zH13;HP)LRE?$2b$KlQi#d0LarD41arY!I{pK3gC>%Cwy8d2pP1CbLS$kR$5J#`IRP
zt6$jGJV@FH&SAzGeRloZXHDTWArAu3>jm!E?`ba|{#e%v+xe-8UpE)W6yZjnGD7<O
zzk~e6Q2(}n`1-ZSH#a@}{%;ZI#1NQ2`-tn#opU`&7ZlIP6#uTq285Yjza{wB!BEFD
zZ{A?<px)NO(LQWSRe7M{hXQsP?e~4V+9k5(Yc#`Mc;RrOu^5%6nVtijBE_GsJbzQv
zdGVTJ-nuXpeQdJ`Ya-`rcgml5WrI(=f7;%clG>>PTNuIHN>xgKZx{xmPUJ!FHZsj*
z*Qg-89WOq6HeMaSZnR<@XU7^5Qnz|fHYW9-qA*UXP~Pa}l8bQc-$+9ED%6*m4&zc%
z+b@^+uo{W1xYzAEis+MLX<NI7X#S)0z(4**?o3n#b)+Ll_m)&)v2VH!waHR!OxEx5
z=M{7!QY^s^c=+#23|-0Ix*8p?Wa_MbURR}2Ea9dj-bqAY&CGvNOOwY1-$$~S`xpLx
z_*5@&piJ&m=|buhMtJ#pneq}!dm)?*m^lP8Tq^;(Q`!y2G#Xlxvf{Oyg@VJN@Vp+h
zXj4K1rV1=1Om)|uFZEQ_D4wq&U%P3gNz`NazmDcGqrS+D`Df=%Arx~VpaO#u?o<!K
z;h5MU>qiS!$F`^G_|Hj<yz9bzab{*-o1?B@Do7p@)a6rAUV+d9;`?V0YQvM={qyMH
z$CW1viJ6(e@b>b};BU6p!otLubd#G-?tM^_*-AZA9vkkbIXV!!h85HIustz(lW_5&
zr6c}uq6d_$RTtItiw3mA{(mqBYwYQ(=8Jls)0Vdtd5<p`em6y*Y7T$zUJwPBz^&X?
zy5+G<us7anSgz8ps(410IwSj+>M({=y7+Hc@tw5U$Ecs;FyL6Y*7oIbTd3;KoxY%E
z%YF4tHHC0#FQiWXpq<vMETYM~(;)6QCK=kGE*Y}$^8xLjA1rWRr?lEmR{c-0@BAH#
zdYbz{V-^^$G8qb;itQ<6@pE=%BNNURS(+qo+J$&ES+pV+80EYRok1Un<>v)G)XX&$
zpt5D=DTv7%#FOgFZT=zXKei2=--uH8YGN$Tym+5B{H<J$`2i21o0K}<x}`D_mE-2Q
z;a^e#|9%=eQ5GXvrWH^oewy1wWyS^^WWJ&f0;jMefXAHu`Nv(0L@duflAkAic~FI|
z-zboq^Z?J(XYRr$|E)%HQ&@|_>dT*cp}}AdJl&LhzWA~y=AX6Oet%Exe%*ZmT!Q6C
zP}TVRTF&?IrRUJ^4(^X`J0e9^!HL*sfE#qD81Z69$3_22d_l|!s>ZtVE=L6HZ?&QC
z+TA~O!A82>ms&qvgor2>)?B*3CeEaa@jG+GiT$zR{4V%M@brV*8RM|Tc6{To#NQwr
zLz5Nlrk>>SxH}kyt!$5*s4A~a%?)Gz5(7X)hetsmzz}V~bKq{FLI_4)T=v)=V{4sm
z-b7de)+Rs^A!Y2$vpFRt*^XAf2XT*xqZoO-TcKZ>qKK72ZdV1TsG8X-E>xtSDGz30
z{8Lx&A;`NRJH~z~*!2ebp!DWtjT>l^oh|Mxhn65rp}tr_ShuxGX$I^dp6(k7Ri}7i
zZHJ23xJR3fWWc}wg>d?^oCgW~0Pk0wWb@wp7A=$pk-B-4lXfGEeZRtg`iqF_LM2?#
z))Aj`Y)AejU&1JVmVnaE$p_eZYS@F4ecI`yjTBWCY!`8~R&0WNHrs?Xh|1g-_r6yG
zn9WTmLiD|IS|X@e31k^`2~o<FnZaJZARS>PVZ_y|2q5Y}!nG1s0|p>{=^LV&HsDT?
z$V%nvCas^a+>T$HrU!3A?0}2huFzDE#%Le=+qP<SVTI}3sbbL*%nMOzY*$43f1s<^
zx0$0q7b-T4L)jHsqWj{dyn9=|`TyGM*7e8>x^}JffWBZJv!)Gx6h&SomD2Z@8~#a&
z3--PcO%&U}QgEj&peB3Q6DKA1Hrj0h@fk?bp6LCmw7hRT7HC}rhNbnNSDR`W4QwP5
zR#pzl5uKTAJdENApgHxh>VzQv0n>F<QL-b_R7j%F7|2jboUBJD9^2}}Up@PfuI&Fn
z`^g~0C_uUf9-vsDAdkN$UvvF+NwwQmRDb5sw+dez$CbjI66%dmF_`-3xt{5q5=F#R
z{{9cYk?`R$#*OgM^kztSy--NN*Dp#l(ap2<_cRzCU-Fyfvlr7BQwvwxXxzFO^I)CR
z&DMSuN<}<>x9GF<n_>&L?uA}Klg%1rl3+21D)*(?L%!>?CAjw6*uNk6j&1N-H#cH6
zw^QZ<qC=~`N~Tpm&tSn@%~BTcM_E&^RTS4b+({RapE_73OIv6z7Xif?s66P;Xm|gh
zDcHDI9J8Z_^H|YIQ0XLqCRT*+teMFc6v}~XGp)44H>f_5oaUxFK2{MNRfVcX{9m}z
zi+ZE^DdN@+!Rdgh(k+-cTIZvOA2J7V?%GM%zxKELrT_c!2OsFUMX@2OutVj3-dRtR
z5*mU(ZGNLTpu{vg?QQ=0x#A_ylMX<!!)%nn9p#6A7aC<z_OdAsQCSEiAAay0DEY{y
zENNOh+&!|?Jsi7C=5>Nqbh4Zs*|wNsXZ~HICr!r?rf3lOn*@jeH7+Dgd!*G0rJm@7
zI3Z|`s3ZY<lGsRslMg<<FGK8MHiCf#UR7MtP+4kvC_3bzGI>z)z9;fp)58z1?nWUI
z<@X<tv|7ls&sQ#b@EQ`z{=FXSQ7HvQn|S9Bqd98WymkyOrQ1N^n-cNaQ~%)a%TwTA
z&!-e_`Y?OVY&_h~y0D<ib8=<Q!EL&8@#vIzn&nMIuS%(PXlCBS;Eyxe*2=d#cFl@Q
zRe+u;L6(_w9K^aYky?dZjscw-`$hQBR3J7$)=qlbB5vXWh#UToz2X1;<5zh{q3g7f
z_?W^2-c6K3n84N+&g}b%{f+ZKUP^~ktTQa}ZuT3vbKoEVA04at;Ddgpj{~)X<;tfy
z944qYcru;+d?}=mel*T5z+i%%Ez@s%lo1%&tE+GFLIpqjq~Ib_CMGs}&4wnuLtll$
zmCuWJZ4JMc-{4H|*H3pD9C&BlR^mrkO6$6}GF_`Qr$X`5P)zH|;M+K~S)3y}H@Qj6
z8?*IA^H}0edA#s?pIw&mgL}b2s>yoFlOf=xbrV3J-FYb*o56j0ZuF$X20VNHo(=zU
zqDt#>b|n&V%m82BH3iBakp^lA%o>@YC8$rusyx@_7ZQ0BL8CEq6^7H|(oJqCr!&f4
zqz}KM@WpeQ@*Udu<a57f6w99Z(0=2C$;rxe!#NfR3?!oBdd@@1yT3g&DtGvsJZ}-v
zr7M|V7X<w9ge^=otDSRLi!)so4Z^awQyT+fD`YZY7aHT@;wTD(k3An@2&$PQKK{dM
zc;-N8bI8-v@SyCONs+6AR#`IR#}2R;qW$kNYx+S22H*bzk=f=`N}7(oQ!5YW;`e%8
za2><#@3nKfIkPS&kWu`Yrg+u4#W)j#o-Oq`bmnZcI~u0q#&CJ)1WAptF*t}Aqlum0
zcI_H9>$YY!zSQM=oI`D1$sc%IVEB5rxq~Orz!UM4`UFYmZyRUELCesQKfr^mtL$_U
zf(iu*kd93g9EoGxk7UdAYMMiZm(pE0eL1?0Zcy5)8+w<ovH;}jW%YFJl6Iw&x7V}l
zZ!7dpoi>ydZs*QBfD2^2dnGp&cR&|2sqT)M`O96Qi{*~L#bns^oBu9Sc$pB9cCuS!
z!nEkUlvy_Sq=&d=^*)0gs<a|I9rrY)oyU=>Xi%M`;sTT1E*(^jmwmZ_9k#|oi3Qf{
zK8QF}KYUa4kOm6bKelyyx$S9ldT?frlY<)x2Zd~z7LJjzMzZj#RTK@XftY_>xW{G{
z&Qvg)1G)&E&PL6X5nnBQr3cJPbRtqXtNlX?S;wwu5DCv)H14f(Z@4||D8AI*q@Ph2
zUX4z^ZT!7G?pEEjaRJjG%hzFb2YqL*`;L@)3WQng(|y@J3xzwcx?=SuKCtYRiXPSj
zP43dsedDc}pmv3zX|iTf(xb;-?Jw|YbDB{igl{~>=bqh$d)}y0?+HTD`9}{~_guna
zJiNe4w(iIKaLyOos+d<>FYt+RVpVl$?+r5u6dh6qRqrBsW{xe+cfR_}?9d>-+E^);
zD<ORSLDNxlfPm9ynwFV(<G=OA#mUeB&buQ<0UQaGOB^&SH3(g^eoEaoK;ZBG<N#~<
zso9Qa+oR=jV6AN%$wyjtyiRi-WXz*?$6=%DrV#-Pryd>1l#gGuwHQ^pD2=`npH7^z
zi<uuazOXbpa76G5j^_27+7ki;skj!Jd+hD#s^S=$v$gXKxtd*iGA%6KD%)nFqU}aE
zdyo5@_ARV{Y+;ivf?;X$P~e)tL#Mo&;(>-z=(8}qt9gKhqx7k~hUfKfYKf%EFLkr^
zf7OSV6&3DJP!_U(P;&k~smXm(5(BhKW930}#Ob0jy9Ut`^=gzzwbQAB%q=7JM1d~I
zWUZ8_N?{HeFZzVQjT!_Q&J-<K8FvALHNT*|&_$sFU}sdPGkobQS9)R3n^~1Wj%^bR
zvIdo~P<eu>`i!QGu_u0KJjv{nzi&Sj%<5<RE;Ao(;MetbDSb{p*K*#d+~_SOmwr#L
z^@~p?_q@0*#v%z6CX+{;&67x?h7i#f&ke(EsWy%VJ$oXWYqxLljKyVApF2fe>L1Vl
zbjJQwP(#X{=j7OU=O7DByOc<UcpW%xnGE;J_-Dtd$0aT<>=6?~_eE_mGWw_lfuAY*
zSWDj<QdE7xnLFhw19Q~OjF%P)GVcG(CHqVL<$FziF4Kpjnypo;H<L}l<`-wH>GDF_
zQ8@{XV>=|yy=vT*n=g9R3-<mhJd;FuQBJ!!^TVUVCj6&o<bvS-EA?&=tjqGfS6>&u
z<nSj)f3D_LYrh?Op5CTDfJ^tm7`4)3DEaD~$||#|kdc30d2GP56u0*0Qu#h79y9%W
zjboXIj3caAJqd)aw;aka49!9;9F0InavXX9miGULmb1qGPa0ZJyaPY==yqq4dUu{+
zI{-^@et)XZmFv}IDa7wOh)wg`M*2`EE2^|7o^eS`lCeA#eofuI%@8?wv$iyQpjO7(
z>CD%3#$)%fhhcm-*m<B6)sZ<X$m{-NGnK_@VCb^=9}ZrI(QQ^%jQnN$Wb@?f9fx9-
z6|SGdQ~6N`Rl)LCCS$v$tTQ?yhz<<?YgmEb{PBx^h-`(l)-Bz5<!48U^Ln4q-biUd
zvnNxvd~NZM6wkjMaN_usq6`wh^K8+N{mdC9HJfMHF+sN+LrgU`Zwsflu!b`LArR$7
zWt=|lE3|T+nXyzWWEq{6EHqlNWbQso79{<Ts~liL!1{pmx_gC!eGqTue{VhrEZ4Iq
z#zsO(3~n3GTg^+4Z>KpkgSK@4&TAF<E;h@bncsRM#g5Rq*t&nI`6EM33BOWCJkFZO
zQ`B5^v(<D&$6$VViKB&b`Ot-0z};sl1blV5K8x?yoE|V5T{F(1JbuS{DtoLjcao_z
z$n8#)l$}GNZikE^_<dc3W6qGpxR&l-8MJa5rFfXquV$Z-a8J2^ziz6NQ`%Tbvx@&*
zO-`Vpj)EtpBXZM~+3s@B?V~EyPoI5wYh3y9yO3iCMs^=tFY)8ZbFoz0?@=(eu60Y_
zgi7J)`*-3Zbw1ZlJq|(2{#iLuyudmFs%ur0@{zP-rslv%<4>%wbOmJt3DgRhc9S66
z*hK$ArwnKEdwgmtWKc@DfEP|Wwj$b-c!nj!HxVLGM(OkWhH3Icv-!Z$VsIa9#Jdng
zXUZJ<?UfyLp<A6daRBkE5)?E#V#lffu@MkHVn1Bo7JATZg%nmuU-q5p{(S1RBh=rI
z^5(BEXs8W(w`XX4L1%(ml1haQu!W(3r_}y(%AyFgQ(~`N84XnDa4Z{oyVH5SND)G5
zR<Nd(IP=B9*}Rn?fF8yLDtuciOp25PvxyutMDq`5%)=P@4;wfrpwCFhD4>ZO*k^Ko
zzIIkEg#m=p!@UZ;a0&<y+sn9e^L31&u_`!b4L4B#R?C^IWMeG|DrDg}+wen4bba2q
zR7b55xwl*^dgY~<kF!#7P`4v=zuqGky}^EZN0jfR@?1f{yQ7q^>zAeXZmsJG+s*1U
zeC5>W2YuOTqroh3Zz8p<6SrM|Xna%~Hz&y^UwGv3Ycr8afV_(@k12FPF1!x1b^SOo
z$ulg*EyT)9N&#)%3wd9;O7f~d!;#sXJ{Fx8*t%a1;_j3aK)}0R_&#1qRy!!*p6INa
z)d~aB*E~J>G8rk;H(wa5Q_BW!6$Z6j!q{kkOY>2C=F8AxxgwNT_%=hTzr(nurM_k`
z5OHL3*HGv4vwO>Vy--X{-GgotVeHQ82PKb}THi*ete`d7)bIbKePb>ZGZtitMXHwW
z(^(Kmx1g=B%VIE&JNwpdOGM%L`d<rzO5Utoo6p{2)BSnB&u7lji_!mb9eWB9En`jl
zl&@#BD*1RCL?YAJjB-s+bu0F&=4Poi=Zr<7%2g^T#w9>mfHm<!jd{tsd?<UruqG_E
z`P1y|yGq}zYpx=nK32c`LNUNwO;8p;73JW88PTxIlDE%yOV(_#W4&cx>D;zW7xbF^
zOlAQC(Yj&y9l#E<BGEPxHsA{Q=%x>%Ph4!t$|TB7&_^I9jCdFLVZSHJ?e_bBhs={6
zzssVZ%a}uIQ1qZNaGVLNb`E>;@OKk#6rd|WabN3j_!j?2TikZk-;9uJx_}X2=TMA2
zj6VuXx&>rI^VbTj18if=q=^%G5|7glN7zq%<i7pd`VW-x&D(3?(64+q4pve|H1`c)
zY8r0Kf%N$<o+4v<P1G;I6zuvi>*Q$37#z!>=yF~U-ps=YQ^gsePFtvzplLUa#^8z*
zlIOtY!2Dh%mDmlReX?saaK%QX1Vr&Y6X2_F5T5r4bO-HG%VfJ`FH<eI@~j7?-7)4_
zkhwN_sg~k#`Jwus9h&H-!i)90)}r39z_4qCDF^uv^^%Z9tkZE@*Iz}=1;>uZr7_^m
z0$Y)ki&Dmkx%v11Z4ZL}kl9OB;xn{tm?BYIgDPb7O+#)HxEui1B^-sLN0`9FwkA*k
z;SmY8;pi?KGDk)*@Q_|xNS9({OXG`yw(_^t+WY^DJ7>JjiaBMV{J&Tk7lDTaz&_pN
zR@rq-nC5@v>|wRbANeQMEKG4Qv+}#|F1;?#^+=0amphSXNeE+Hky~fGwDM2B(I?;(
zGf32eq+Fc3;Hj6-0<>q^_G%qCTDFy~q&{b`uSPK7iQTt%W%H)GS5@?#>w=nWyfP&H
zaOuA|8MV*jQpL*jy{ce;&`$J+Fkid0v+>xHDF>TOV?Gx8SAKC3)P~awyJWcVN~+zO
zC@CeeE8{KV2HC*A^#%Q731xvlFf(T%xQa`N#N%@25$eco*TvWef7Xl87YycCoq%G$
zTMUKKpb%ZrrY;kAnQUhDT}C!g!iEDLBminC?DcqpENCF}t&c4965~U66U3@hbZRw(
z0ojG(y02Ons@lvfwa&=$ZOy>ryn>J#GXt%>olXOH-wf~NuHFFFD<2IJ9&F{*RU;(p
zN!38)|GK<xo~gEzJH4Qb7M%*_-;`LHD0`}zZ@zzSc^xVm*5_BB82vTN@=~vTL|<hE
zIzOYxq(YNbpMI^GXXXXBRnrViu_&LTMHRGlHN%Tf9|D=Y<iQF}bx8DF*0~gjQMf2y
z62%(1?rg20PB>|AdVhrL!&v&Rmkg<8U^kqN1^G%J^kKVK6`yRgKHqpf02COG`uFyp
z@{Cok6SpbK3L(q~Yggq~lU2oBa)Wd~@m?<z0H%?0=PihtxJ`ay0pA9;)G%%;Vid=f
zMS?{&%Zh~iTsg{aI7;V_3hN2#c>?DwBI@=`t@6ivxxor-Z`^-BguQNy&0u`j5EXt&
zN*oYp7^^kR1!C6VgfB)7`IqOwF+VK-1Jyn4EaH5tA*RM`yY&sSTdtnE`S871h;Sr;
zKWQ2|KBfDvF2ZJSO_vd0(y0=&_=y6bVZV^P9*Kv{lR=N414GZpBf!S>3}%RMmU$P)
zj-{LEg%7KPc(t3J99w}W3*{Z-x{htY-@*iax`2NHo-qLAHU3Mjhy@o!|9XW<=l9k4
zS06q1u6U`Z`rVT{@rF4FrNVTf>xvW+W2rS4ss4vckL5o9_>NTrzJ7k6U3tE!V0hJv
z1~Sj@{si;R%C@;5n4tax(Q9h9L-Z`UzTT#Ph?Rp?<g$Le!86^-*X7vv(<F;w+xn$N
zXHF)bbFss+{E3EhmPFBoe|Jv&&V}73*+O15T36juZM`4-`&rmt_warlj|D5SAmxUV
znuXf|P<tt4=<<)LYJ7F(PahHdEpBTYGOh1olqB1fHWkStp=Kmq;K;u)e406R?M(2^
zVYTPKMRQ_AqfB-{y|Na2eJGHu>~6j7FQtbefBA-TJp8g-j~ALjA;Gvz!<~M;r(7TX
z`Y!#||6}Xa)&X=|5`II<#4$o$)6rA+Fv679hl@Xf5NF-}JG~MG0l!lP(|eFlA7Uk&
z44t}Hhi6~?mJPzJ^7Mr5>h3<(Y%ax<fH&1jGu>r(KCFMupiiw9-bKLp`RV%Wt*wqf
zF;64hGty^6k9Cn^jwXe)jOw_M#4Q|@eD^QG$kK!i7q%sztResc@XrmKF4gGcv7kwK
zqVkS~d@5jZ<!P9wJ^o?qn!HKe&GNpBVK2YiHQ!#+xSRWAxo22^x)ef*ebx^rab(0z
zrrKU<5R;qx6f<H4Nvc?InLU$XRDT7g9pX;CX?vJ%q`y-asd)b|Nr2~B3DRNEdaXc;
zTTecJj|T3Hej>9e$cmDd`#$zPrY_EM98&2Z8w#+6CpZ}iA0I*Ps$w==F5b?4lJ@DQ
zuYr_0XSTYPIAr1xpo&?5&oB@(yTc^5{^B$)U=6ZB&5HjriL9LBJSbI30`a9%-I2em
zZk}#@UwQS_?sUe-PQ9aA?Xa$kBVNz1yS~27+)n&EJuSCX`9p8|$tU!TX;H1=!3)&U
zlGksUy1px34m!OPRK{hZ9#$e_p@75PopMypbav76pWW?+>=<m^T~=<%0*=BCK~^wX
zG^ZpNgF?~3rF3SPd@wgg1q%a!45&3T=J)$+xQ~`d8i+Uv2WR&w1B-PIlc|q#rC_Sw
z0E`{Z(&Npc>U*|5E&6^`2!qe#RANB8P|@20&aF@OMr}0Xnb1e_e@5|B9O!tRuoq%P
z>?Q_p|EHQo3&O~F`#)(Z(~IUo^!2G%_!WqvK4hR-NLQ#*fHO)zvybSIHC`dl^oxCE
z%7*WfC?>g)bp+XKlL-qGr~(cF`PA*hPfJB_pJzT7Z4}<&!b$y1=`;Jp%l5}+lYU!K
zH5tx%r^=3}tF$XWFj_GB`oB)LeE5$oeJ01ay=oyd|M0ch<_D<d=C+cXA(jL5!twG{
zgs1S>pc8|)O}v65wpVnoGh&2E(2vJBRv^>-C$5-?b0Kr4T4Pz|Hn4q}=I*;zOt(aO
zCB$k$8Y0#5XD4@PY%TkhpN)Kd3<%lcDFd%#fV@Z$<7M1fzVP^%6VZN(^n0(avf!ip
zkR|-E&-}8aTM#AZ2~2m@2eD%)uepyZP7@MT+{%vIuWOOlxCkiK-Etku8ztJLLlDb(
z@$mPHe<pCx%jVsU=^rB|_`l|fzXqmOy7}(~RIl_OipS_NRV*z!tmUh_zRbQ2s6A>B
zvRP_}b`99{C6hX&LNKQ0>OE|KvS6Ksyt*^<^x|Ar8#jgPrs5gzQ0W)KS-wxx#LeOu
z<rba3pGKKN_>~b`gKEVvL|4hvxghMwr=L_qtydJN35x{&8lthQyRs$Y8LltdBzvgt
z2i@7V=|HM}pd*6Y<z?X0^=-+7z}8dridv|e%&Rorov|#O$iG#7<|1hV&z#1EJsOXo
zVf|}k&AANPT@e=NmY46?+2{>Oy1TA^MXM8jx+8BAbW9xxowCbrYwYt;IAzUh0^Ct-
zxJlk+<~~T#v4Wyn)~-mX@AX<`aEcqC<6SJ&d1Mj@Ce2ZqRb~pc?97Nb=592eUs8z$
zgr-1!Qx~+GK&PQ5ntL?nK2l2XE)EK~Y6*QmI_1DYaa)#9jNECDoM07f|CiA+->|D{
z)%{rM`{}#1cc-@?*LyUHUkQBjQX%+2kmKW!e!70D{nU=qtwX)xttg2WTZY=Q#)xA{
zSnQhrp4R)Rg@c=p>jd_Kig&Ia-W?qnem%&K@Y*5~H#W3}r1x_E+w>KzC_WXB_GlMN
zl1MDM+a`^O`Adiioxkw>v6gFHs*t<(csB*+!+iG*>^9F*=cqLOk$ObqvEWkk1J==I
z*9S(3ovAGVghbshLmZE=7RHpluB|rkY2&cC+IhA!f~ASarCX%)6-$arGnQs(nQ~|B
zQ^-O?T&VIF!Q+U@TvI8)MwaS4^TUIbjg^#<l6<tPqse+J%+V;)+W~i9lI?Z4YYRqk
z5^Ha7hp<OL7-ZBN(e-GLk+kP3yLJDpDepSNhtfT!DfPE}$lqsxs%&VdIHt<l!yp;N
zMpj5KvzgeJ+pCqF-x?BdWWzY^mEAyl4`;YV7P}?C4B+HO4H2)^e~$!w2P#*mBC91@
z?Dr37T%q<^8WaQ0ROPN?$DFVly+|P|nM*Pg_`EL-0D=1+LpWv-ayo{xUsc>9lL=UV
z3;S{oG{f4i<!Z)Ja{GoB6{?QCO1nX+W1uA@h$GaDXaj<}*aT4soT`0KhG~N`%3V<E
z?Vy540*EoQ2id^U(_KLKi(G00zc@;ZYS|{vKm532>~;6NrOREv5mzw{+j;jw!e`K^
z8h`lW{1dXuWQu2_*4(beaDWR4y!?!?`}dLc%c4JZYJ26#6Qjy^GOOLJ&(ZcC0eh1M
z>yG=?802QSevMSq?Xs&DY~|VMZdNKKYeow=+OR<J)rA{fbH`rUQ$8De(%GBUJ5@6$
zJDLtBQWWenfl=N&TQ)E)E2|%4-?OoCwai{I_lYC6L4%TU@KLM;3{=i{X!ktfcHWSJ
zb`tRV(vo#1N-`EBY%Tz0m7rc-DHa=ocz}M+xKirmh3o_oWnYS&_rxnn3JPP$I(%EZ
zsd{eFd%s59hB%UPYDGZ(8RV2_VYo~zSV}<nlRF2@0wV_%k!xTtD$|-dpxtPd5?MQX
zJ2%24mUpwY0xyMn+I3NQK0_Ekwnt+b6qvgK{0J&3V+bXQKXf=i2U{EK@@Wd`r0G73
zD~8*UTLUYjsM52rlk*ml{at^38qsev1OEfPt;P)boEiESh5ru(Mm$|;*!DZ4eZ%$d
z0yPtFfUt+G>EjzVt#KZo%$TysXR<PYxt5g=9w`R@%+_qE6<E8=U8itC^D{mv!t(`b
z<i6)~Bl~N@d%b399>79Ugcf0Ik9sB(z-w=*YstRGXz+?_v32$}l(lrT9a#|}fK^F$
zWU@SO@BVVtOn2*2guOxLIU+hvh{<)kR!D2aqHm6(=&V?RXje1$WWH08_{@B?wf%)5
zfz$l+;H4M(f(TdboX2h5rOJasZ;k<E+$vY0j)^{8XSeKY0T5y4RiWw`63Fqt;WNu*
zrfAKyRXs_hNVTdK^@ibsPMSlZ@~6y{*3K(Ke%37#55Qz%3jJm`I_PSKH~L|zC*Q;q
z#L){%v*($Im=APc{|AYK5}AL$n-*B9{wR_%QkEtA)i&lK=$8`g2u#jL-Rl@%wgM@b
zAG7cIY9Sz?RdrtR<;zYt=L-Up4nC_rkb)p*Q=gALobxXqsALJZ3=0T|tS>cabtC(H
zldnwv-gqgW*?aSc528^3P+g70>a7G_lB57Y+i8|~sp$&q%>8moC0_-<towfX?WDl4
zQk?{dp72(A++hSmGC<e#G6y&#(0KoMSthniIMtJZ9!R%YdR3O2dZ6?!LWys-$cs7f
zsA4Hpk9|*UMO{EuR-%+yxELKVUy{t3^`c*Q*oFjN!)6k$_<@6UJ;L}E#a?_)b`8i+
z{qx7lI%P*#=cDD#L5I}a9!@J%uBr>B;*0oNIrU*fHtVhal3m(}TWCpage_tYG;ARs
zRODtQ?g5%FxzG!dl7>uV22E!q$yKC)4ABfxZh%RKkrRf^lOe|(RuM$(sp)LWat8F8
zj!Bp#`RoQ(9mG&e54w6WrZ)AZMpn_Q>Kdxga<5L{+Rvx<&kG7brzca*-gMrhsG2)3
zmNXp_{r-GtXCUo)>UXQ6q_6uoH#gXRc$xQZ4afBcy7nLF^FPr9=ezGUej*}Cj+eH1
zcO(BcXGKmqMnsy6BR7L_$a=lQJ2#i6+`b`~9oUM1Hpxag(>+|CQjvE+QU2nEboXnK
znL`F3|3i!2%|fZBY@}nrh0gA}^m(PnZ{Cl4Tq}~!E0>B#M=(;w0Q960S^d+k^3J^n
z$x5)ZpAAA@9*nHD@XorO=2hwXx4kLHv*=P+rd<_z?(2tpKMx%LlrXA1SJau#qN;Au
z9<F~Ek5%2l1=2!k-gSqeMh^@|&wLH^3ZOdfUn)UwySly)Qe&a0w55roBd$-Jn`l1U
z{zwbTQ*q$%>SvM*ih{aC7-i|}e~zLPu_CYihF?qhdbqYWt$tBtn*~bycWE%`qQ}4o
z?%#vz52e1qJP3JBB)X%s7ew01oza-c{Q5<EwbU&=z}KtH9ae@vSN1lRg4Fq{-9+C0
z>uAwBdWrq~4_iO^Bg=!NT5EKXFRS{tV@To`5~fe#W@F&Y`gio58o4Oz4Yw?<Oz{m1
z<fS}1_i)M`8Kwm>OBb`<jh{9uT|nl37@qg;@J%0eYS;y<ypJAbpC*k!jG@_f0Yax6
z_xP=5QYfFi&JcSqx=oU_gIyCfk4F#1gSo6ZGu%lBB|#?`A88bXkvalGmTq*5dzY!y
z4*h)l>iO5tY@Pw%^e#7G=(Pct*}4T-0q;d<@&~_JKGS85GoI34|87v)M=rUVCHsF(
zI>=t8Jx`gM)w<=1dcVg*9cljc`sJN;h29d|6PpZRTXgrs_dl%MmEXYWa|zM4A(xL~
znPp!(8a@|$y8VpKW8A#(u|Iev9Njp0)j}ES+x-NWA6AiSgV^*akSBZPP(gY|zFwAJ
zSu^$aM9kF-M%Zr4mwH$5pD(Uc0q`0^8#vF4^v)}bq5neieG-m_D@X(5bsdlIv5L3#
zo}S6y41QtSdOUWsv+?R><QJY!n2PbJ_+B|c*<Z5|Sb<WR)$x!*b1kWLtr4alB0(|J
z8KAIbGCvLjFp?mXaL~uaZOb_l6)SgdCm7l(cLOU`t1Q%xKph8}AUN+ek;70>v>^0D
z0S7BO_5+!f0ML6Ak-{U;SEro0kKYdizTs`c8SR@F|82{(w524gKCOI9mLE&Zuo+P<
zEeD%oS<Z<AwMK#*sO*2c^A)u7t>xPG%9XpLbl<^K&4I}BWKUJj{puZMt8mt*ULy)>
zi@TN<Mi|A1{;Mg{_Q*jrZ2ZA_H+=nunfSgs8Bgplwh+q3@+LvDh?WjArI|+y6V$14
zMgBSoSdtGUrKIiC&R0LD_8P>X^UNDU2m<`7iGO#r%q%mI#0{N(bVrw#aQ|ltK|O{`
zCjV7Ta(fFw!wMv}T@t10g9tdF=>W>1_SJ7|y~<3pO{~K%pi`t##A3QHhyYo$zN|*|
z5NZ-$pyKQ7;>_9ej^kNMq%Vr35qGC^F^biXRkP%ugTmJ%>^yJmy5_52y&p@11uA$w
z2xLwtWj=jDkBi%(jE%c#=+vHNy5lfp=kOy_ixYZex#B!*Gk=TM)0T4LRn_5y>VuIc
zQPsk(VciR}dB6B5q!DJdFQP!4%<1lj{@SNEt$7DIny_THAytL4fxkJdqPafZs#2Bj
z)Wd!ls@ZwH<0Vp?eDYm`gkdaB_ZN2^{FoT9i1EBG*0|?v1*G{WMU1$hQ`T!{5KMwf
zH@=C_IBz0q&3Af6J%u)x{)RGkStfsq3El$Ijw_S;n0+RY;pX~wibn|sHF>^s(=9H+
zHVf0v!LG>JRNX4Bys@s}<+hpZ$g1ZmnxZwrOsR&mblW2a?Z*~ps+INdQiZQ`xC9a?
zquHh)$jeUXRk&$}=6nHCeQ1@^y=7#=gS9LJ0ixN3U0Neu9n9B=<`l&9P5m_&7aA@@
zVS&~|;Bd1R?Kh3J8IlpG@)OG%=_5<g(F9rim17Rwz$STR*8yJIHx^AL6l|XlIC3p2
zM<q@6HEHttTw=s61?MrQn~xycW7p_Pvfst(ug)0bgX<-}H2f;z@AZ3VD55=ZFLJdn
zS|KRZDE}P3`ImWp=J$TF!wQe(w#XY#h;&o`(i~@I>>tRWwQMPG#ucYIBaSF6$M&Gh
zQ8Sm#@>bjIN<hOFj-#=pT~hp9!Ca)@=e?S9qkCWc)P)V>bgu1n?FI<0iv?^m$ExMg
zL|uU3^U4thm{q=$GEz0EV7<(#1G)AlT9}lSROW1WrR5KOQlMOl-YULJZLe^CT^z|(
zieJtpF97B*@5V1(4}Td1oMAF7HVFSlxl{jgooRqo$EyX2M5~3;jLsYXG*o4kL+t5&
z1Tkj?f`0goc8GoNW!y^m%2@vG{_UT0x(R44=L~YcUVD4_wUXz*f#=!EebM*200cA^
zAu7s{p4eA^X`pTMgz@sG%l_e0oCMC@s<E%wjsvS(b&X!W_F{Gb>ok@Dj3T!me96E4
zqpbC@jImQAjpOr5r&3^KC8Y9~_kEq`KUr!&L_f8NC~BQ;UYmku1as@_L%4fM9DXFi
zPS;th)V|r*<BKo!ewl3J%0yrSmcE&pnbgAGu8yP(*rwHQC7~zxWS0h=Z8P11wLDDX
zscys>hnw7PcLudmu(As&X7i=^7qb*V9AoB@r822LGv}>N_KJ4FhX9&tDOn%XrTnVb
zUp%z<QW(^+f%fLho}6(p69uZMjjQxw3>`+8gT>pPqbOMfvO?Dk?`~7sxn_j#(DsFV
zPBYYo;WWw<Lh8fm?^{Y|BEs*!`m_K0(#-0tRLg5!;Z&)-K)a$Pv$ZlUA^ptUvh!UL
zR(C3b+P^fE9Zk(;`G{;@`WO^@^Ny9~L*%nsOPiO4e76V+CYG~`{bFfmNnFl`;&70{
z+ti8kn^$5Bl?j}IhzdyDpbD$ecU(At#nB5ZqmNmsuxfh5(OvWr*!^;C?q=L4A>qBs
zC+){J@{+1S)Z{c%xq7KvEa_NOosm<=Sq&qugljAW^E)clA3IWsgGyLohm}h`L#|@E
z%sXMChvgEaxdO(D2nrarqR=rvO+Q?`X6XwGyV^+{op>D6rx^iMo1Ui@e1F{ps)?B`
z5K4b6g{cJd$1JpjSDZO&1t`ypRV#?nA(cWd5K<VQPFMeR)b}Hn%$XIwT)-{6JP3;T
z%U=xi1~0(7{!d6s{{Nbyf$g>cjk7k`hdMbhbBdQ+q^4*VTB{xVHFqgHrA(TZ?PUJs
zw7$LmohktI$v-URiI!t}68ELdQw6qQ9#s(63z5KY62Kmoh=Z~~X$X499LK%@XAARX
zA7m2h)kRCZc4t(d&o^8WqpJ0V{jNRbM~-qB@|`VL`84%POtBX!9M?bpTtl}V!wo##
zXX4D3A7i=HpHzZN7XVry&iDY%X_i4aMFpX|xP7f=BCJ&O7X__m_t+p_8O<PO%89Kp
zGjN9Pp!&;Ohb^A0$J@81s@~w7FduI*1Pv+rsCuThKk(=@RERU0#3+`Z@56})Txb&(
z98VY@tzR<H(?1s{S>0g}@eiJ@&|3)#`vv~`g;Aa!a<MHKw_z)ze#_}A`&rf|GY^Bc
zLA;b6B%rxfOxLs_ZrIjJC7ZJ_h+3wafC7#8W@3^B(S2+`*BF&EZwCcTX3sBqYY_z!
z&D)Oc2yMGSF#prp)3Rv$JeaEp;j!&Zn*j5%;PZTXr`jta-g=8@4!6aK5q|;E{NoW=
z7|^}$IVlB7idpJBJO6={3@wFIVCXANjp~pOWc<U7lc{wU!=cluskctnge47^UiOn<
zO)+sak+UfDKJAplhkVbRV5aRa(tA&xoszw+W1gf$Rx)YaljG{s{Il(>rZuQ2T;MI1
zz|!j`T%~pPM@l}ni?;PNL2*dMQtXLZ6^q>dmLB59ayG_N=uNKj+0agvFTd159}B+{
zkkEQK8Hxtze84X~2UI{rUg*c(dm3xd2GwoR3Ou^8W@G}gAu>{LSb<{+h7mTW<0K2%
z>2R)7L%}{PC^+an_r$BOyxX_bU42ZJ+Wh~Tn|dord5C!YGOn0mxt{-sDedlu!O^)E
z`o{|g+?8}7yj{S3xfuV}F$OKpXzOl<p|(<belfHDbZZ}zUNnSp|1tCJH!@hbs=uiT
zdD<e~7H~dNYw}Sg2oD7Y602xF_upGlk{fw*qWEZ&y(bQvPHT@GQ-hWyq&BAWL{+4k
z>V{>{_hj#@cPU2_$V*pyIQfQHA}}2n{-(pt4_<$Sryp{G!ba=fd%E!?NwKznOQPyJ
zEsc9yxGp9EkyeoJ;RlziQsyd#-_}AQ98i0a!El0X!Hy-MeKi;+g<=vwcW0SL?ya%b
zgbVBdpJ|(Cnnf-@(`BVcpcfXd^F({QKG7Vo+-ox$0`fgI*SG7nT3V>T*Dh>jbUeQd
z+9)D{_P3x$NmoW4rj<EsKi2<UaKTqp)QY1=>7wF-X>$!t(R)rpSvqN*epx8lIu&!!
zY-0LA1_$>`lRZo3uv9n0S66rc9kEIfPxMDOEV}M`^6b~?&_#vLj<l3{<hAPl>4E%A
zzg~+K_0?rw5ogFa7zF53MSJPQ7cntbpXc^!<cvrQnO}LW0)yq}vci^M$N^;u!e_dR
z8K_WhSbJOwo~3!GL>d6&%S6}^ajZoo!UlsZkv;Yw=y*fOj@V`=syQ+C618o4;9tC7
zHl(><_wFrzvcNmW1~Tm8*{I=3g-)Vq595j#YB^mCQcn{S;*s<!x_Z8wFcB-xSwhY)
zn=w6O1*aIWTsZZ6dbW)4_G=bKh}!a6d2Ba<qCwwQ?q&T{<$k?av*(F)iq3&emD|R2
zTzcU@J>r3FiW02G>)xo_%;E?2gh+w35J9-MeU=i;Qo3-T*V|BF7dM?Qylo3>eU^SD
zBV>5V1UU$h7}ZtiKws#mJ}-#~+`{*1g^^Vt+P1lNDFe~xZvA2fh{|$t-R(aflTA%|
z=pvQbx=@Q9d|5RpkuNwt7d!J*EWfqR#Qs7S_lE1Ji*WPJD%QYo_1oPmwZ0x&QphzE
z?s_<=tST6{0zA2@!iZRO2%SMb8#RY2mG(}x0<nuGF&>G8oyawCB#9P~7zG-VWIl1u
zg7gi2Cg3Q5;Oo%~DZ!}#Mg=4;9lB<t7-$NVn20lXD&&9*qnNI>OUD0aEjtG%%at>E
zQ5jEDUsQ@NXVhH@xFI6V_3!5;p4T2Y3RFA(m;97lPXO^w>i5%8cPv658unzjV8C7F
zvQnDA4TLZH4uwZ5cz%dgFqJW`Omy9FV|bfcD_pe5Riy+uy#Vr<>K#0Nz6h99N?Hu6
zis{TXuzHe6H_gLac&w6$O$IHf1IQZomejjjW*i4+{^aBe|3<FaQSkRXW3?t4Lygdr
zHv3ISU~@Oy!o;VzR+IUCj1NESeM{Of{Kmlo1K0r5b;=~*oCJ76*xU$&<3MDs6kh=F
z5>*QL-j_;K{>p(vMgJ1)`{cm-3vb}<t&1bzSE;GTR=7KBIHNKZRT+UU(3t95kw|Vv
zEn8WTmoqykOULDX;Ygx5`0DA*#|8qo2Kea0L8kD>Fx2l_oRZ>?B?>`MoUT)=iwgmj
z>@RfQdZw(7vas~cho>ZPbtC{j=UVTO=ifnEoi*}spX%hZ1c7o)6+}GBJ}<@AaE^Dt
zf)vE4no(31!CsrHq0B$*X{`S-lDNXp{K+4E+ITT#T)3Vn<=k@qwT~MIJ%%0Mx<K*s
z$>qsr$`TE$xg5vyXZhKz(8GoOOTKFK>t)c8%c9)Eqpv~$$31Cr*Kjik%iEpS+uyI7
zy>>-^O9OP~T-0t6y$o99{Pj_3=B@FiX`NTJ36J(frMYlQun+3N`6|J(j<9_XEt{TL
z<9p8yHHAZ%GI+0=0M+9-`;trn8&yPg)<=yy>YzuArwR%?E6ewy70Te=4?S3eeJ;ns
zZ!~;49Q#0q)<W-`qir6KutSeVFQ2U--3?!rLyUI6-!Lu$SlTh~%NL15K49?4w(-_9
zSriOK3?d@{&?0oZRF)zKM*hTUcQC-fg3%MOQYybM$=)*aNH?+S;lt@>&!&#3Uo;Vt
zEgVbipIS-3rug(eOTt-1SFiWwU<Y2@$>x>l`+}ceGnxBNi1xA8%6D>Ku3HhEoPVb~
zuKo+S1Z`a}76!t~YbRiz3n$t)`vk+;PoQFqXK`oeFZFGiOpLqUFPV8aGbyf05(Mo9
zBt6s<QHprms`eWGha;fJ^rx&vVy}hZ_m>(YwveG=8FHkiEyVTRU8vP$?Jc_b_s6TK
z`WyaSvoV%CG;T4*-pa&1y<1X9hujBZ22kq0)}Iez+yo{=ut&n(p12w+Xa{3y-8z1;
zwBwb3M^emZqZ<mv0zZ2}aw3Y}?=+=0Tr%|7UOvcoJ&rU7k$S94Riv#dokjpWJ;&g-
z=u5x`b5t9LE;?`e*9Y?c$IYzeAI_f(jivn;cg#tSu^pJdxrE;zdjA8N<%Q^UHG#)C
zC&d<ucpl6B2YS9)^jFgQ&&2h)>z2Sd@f%S%AND=<9z)q0&}Lckhv{kEk`b!aL>mVn
z_$1Ri*9i?s+tCyP0t0M$HoALFnbU=<w_i`%=K9aYnQ`0vdUSu%`Q;zXd!`+p;Xn8>
zM}z)_y{a4GlavnREicU{snM;(KjUIuk$%Y-wJ|}1D;w&c(H{`k$AgaSm!wifpSF30
z>aLo=V7>;jmCu|#_D<wK>;@A+yIX@AKmnR5TF~rYv()KKT-X|xOIE6~JR8tio<m*>
zR4jUPS3{aN3iuqvjxF|S>Gj?qbnVhzHqZ{n>9sl;D!Ni6;)Y%e<^?uhI(u@vLh&>b
zS<b|F$nd;)EeDh^WTDI=b1LM#58}HhDqtv1LH;OQs-aZn4Z0`{Sog#}^B{xn(Vc$5
zmLdET)8dp%kcP_4Z*l21#~itSY9nF=qvVBJWzJ(O(k_lgK)qCWZ$7WLOP~}J2Iu7N
zRL(#R`C{?He-+eH8S<R6*@ZcPncmq&f4E$sX@~7osed4<^d;ZaLbL13!n^JVMkd32
z-b<3N=ATjI+Z&Fq2F>R5SbjxSOFs~@Kn*GdE@!|*f=t`RuA6|^m|>khz!pGaD9}P4
z3*49BoFVE(Dt#o<2ovQi(KJgXy^jn>V&0g5bjm;yYkhLi)DfmdJ{<cNgDlNe^vV}I
zgJ;SR)BW^+j{zAUz*Y1q6YJ)uSwDM2yPOuW;fYkO(w-S@cr1DAil|hq;eixWiH62=
zaHCZEMiyZzNfIT|_7F21@p6ijqZXRM-|_o{{PA*Lvi2xbWRH5^Yxf>sXj@qzE~X;)
z-!9)?rGT{8`{4zXC28+#I3tnuI9%k0{0ov#I%>ayHf;81g4bK`$^f*?1Pa_7z8i*D
z;0trlLn9Sb)%L3pZ#4_|1rQs+=05F93h9^Dh{VE|G94<&3b|5?*czT7gJ8-*DYa;!
zKF7W42a>+4w9qR@do>f2LOFbU%gw&4eM%o&Dgrkwo(XE`Ex_^WoMu6Bj96AVx!b5U
zoMW9lfG6w|9piKtb8&Y85#S?1!zTFCFOj7)k>*Tr<VIGI1+V9FUcO|DVKBkcm)+ym
z5Zjt*%ab0~DUE9ewr!0PVf@dWWiE^tJEDZ~mkf&kDu%td9Aj6)Nm1y^49Z}QRAW3A
z6iZuIn*b7uXE;%G3tvIG%;fC0D)%g0u#|Nx#r3r!3Z9mxxq7CDRz(-xuUhGw@y>K+
zPs@bYZ9<$wGipL)zwsL<=MFqqn&IkHU>91_xrm~Fcj?VD=9#%7o#ZZd(SPz15uCV?
zT3PGyRx9@h-+8s!vbgwK^OEf|Pu;(*Ip8s?TE&gx;>V+Ha6vMPb~8_b$(aH@95MTv
z>APPN`&In@$&_@FOJuyEy3l==*B+4u&?gt_j3Q)bp{b%!cl)RP+*$>oz}L%hJnMIR
z+<u&=pR#pLO1SSOPHCmFdF7~}vxpP<5atVtMX%J-T@21uxn7id(J0+GM0f`}%{>uU
z&UVi)Cff!QXSNwT;zk)u!uZHpg{_ovPSJ`r_t$hK$)_{IWvHMZa$qCKMEY<o2+6lc
z+dTyC4Wc;=IS>$uV~`>1K5+I2l)sfU4>AcLi6x+ej3;6!$AT~d<jg1ZhyKwz3qM*H
z6hwVnzH-jS)t>d_%s;mGIs^6DZ+bbj=(k4K{YZZLgEeg{q>-VvkdlIatmsj_BLgsa
z=Mw%OlFq}Q>i>P?Z;rjkv5zervJcq~vSkyJ<k(6$GLAA*LdUU3Cypb>-XSDq@71vi
zNr>#s7JYv2@8kD3obkHv>%Ok%6+}aMwf(%kTH@${!~Nm=TOEbyX5Crc-jp#QBYx%t
zO0oM!u>50lw3dyclTuOT#@o$-h%a{!VT;!LdtIqII-ok_;ejfcQXhz(hIIV8=c)a-
z`V1!aJK*7@WCye7x{Kd@-eu$`zCoQKqDfAqul-Hg8!1OO7ZnXx!Vw2-fmu6w11_Nr
z<?KeEOnyzNNQobo4)Q@MBFv|QRJCTBTiiNMgGRLQr>VkbtshQKn_mlRMC@<GU0z!1
z^$xr=JNh_LDbpHrq!aCw=Q^{zZD-*7zoS8%rZwfSut~A))}aU48czg%rPX3sLeWUR
zccOEi9}shGp-kgPJ%yu~`Sc}gxv?W_t=vq*_GARbhhLQ4L8l*7u%%V^x=;7q3xXG0
zr9~6zbeJV+^CBmEH!V7^s13yoD-Cu1eM)XG0Ofs*9X6tj`UT@~jr`i1R?k;`p*k1U
z^I_LKnFFmX@=ZBdB&@b)@_9&!T1f<WHX=x|Yd#~g$RN*fCdK20KMN}Mo_WU0yA;Re
zuYB{@2UntM!?Z)(X1UAn>LgSAnW+fbN$JO@p$Lft^{z?}xCE1L*0q6$RmK|nvIi#m
z<D#i6HkiL)(q)Ms#ij}V*Hbp{1H<bSZcBUbB0SMqHyx~)&>zHfL$&PV#opL{LztHp
z7U1a{f_CBbYq(~H9u7Tz8?!RXumf^VQr85th(M=JFg;?furVVo0M(Em6_WY!XJt4k
zEFSBh_j1O2)E7_d`7e$8Z^NLvZPCD+Hj1vH$+@PC?XNj6Z@8_J-18}k+b}C0jdDu5
zaX=7LV3$$gVNU^GT4;b%1kgS%rB6=pQx$*|fL^3>T#kTmrzAqDZGDErjL;juKzc>{
zAYpMnrF*`f$@VvU0q4&~N9xK#6R*&^->8K7T9u%$Dccr?srw>J1@Q%c9G*$4Bj{qj
z1v&d<WEGT)FSY=RE3o^B62V`8Y?PGI2?Y!`AHf+z!9ZF1^Pq4JAZdMT1p6m~MA1el
zGN*~v1l-miO3UF2d|Ln!@Q;sKOSfPn4$vaZl$Sm917Km5P*A7K7dEt;drWTxK%d0%
zc~UF6)%~2Sh<a{z0-KStb5c*Fzx9O|x@(ciAEe+VeA!HR9)B+<Y)$#5qR_K`s&=Np
zXa|j)pRIqd;jMKSw3DXwp71a-1+E%%TylN9Z2tMy5qO?{#V65fM<F3}%}|h3AM%Yr
z&&Nk}l9$g%eyjZAovhSRE6bWsWee$<4LtZPuSjD=X{QlAtRlKm6(n)me_Jiw-f+I$
z-mc=L#-&jcr?F>OM;nj`>Dn`=)t+hKf!VPSazS`UQ*H~rOlJ~LK=NwaK=Nv}WB5o)
zzjUPgDVbPKP-2ye-ZzrCULpCac>P|8wLE*Hu~FmR(#Aw%r6aRikdiECk<fNX#yap*
zY?*?riG0L#zi8_EW{afy84+d5yIE?C90rm55W2~r05!qCHV{_#?18mXI{|?X+AL%`
z<hvg`;@(Gxok<LDe%rV@N_Qs7JE>bnGur+Wgelsw%&aiN58KB+e_F%1h-TL^v36a)
zY*{1z<+L0MpoCqg9_y-}R|OZz$7))VL}-_5t{NN13I|LWiJPhM{=B3b1gN7@MtU2)
zePjyM^vi}dvnHX=mhnsMy$qFb{rQa3TCt&kRZ4yXGu_TW;p<Ps1x;cnUMF)mDQ|WU
zeC;ZKx#zA?Yg=<g>RG+EBT4sUVu&L;%bc~8rZ`>2elh3SruV2~*BO*|7UP`Ugbp=@
zWu+0T#q_6yc+vt`%u|NS;ep9r+dM?UZT*%)v0)0`pK}RpOcEUvdJ#Mfdi9$Y2%>S6
zhQ|4KO2R&^#NVk~zpT-wZL+T%-8B#y*u*6h_Zh*AWNF)@q5vC;mD92?xyO*HsYo!j
zh*B>BXCDtZHH0nUmwn^Td?x^~L}P5I0a{!qJW4nf5Zx>$Y0wg~0o40%<|;^PNI7i^
zpwTb#$A)tAm<SS{46~NA!w^j11SWx-$$yT%`t0FSCY!F&+Rlxuw?nVpR>76b$h`<$
zHY+;yD>+uh%*(A-HZ6QI*|`b*$2n1X2=o2)Zg$O<9>RP->x5Ra6Lh%$rFj|^%)*EI
z*mkPTP%NaMg-s8Q(m1XDC){*pVZ4TSVvELv2*clDz4f;6uAEf+9qK3aT&tPwc&D7S
zv0dkX2w?JoeO|((s|V)i6A+euu5*`<H&cJ({mh(R-?K>t^VU~Ry;gr;2!Dps9TQGG
zz?-aa?X`RN<Yu<0->8g=_Ea1og_^z=GU#x0$x;VaM&%q%e{_0_%@`n$bc^ffqNW&t
zqe-q`si)N?RMfd6;zwx*fAPLg&UGYPALdctSpyk~w{9bas$sVM97YUOsWh{;AMJzc
zZog{D-2T>M#5*h*r~S-*P7=IDw`Zp(BxUK{54>Dt4$S{pZjpRed8tdFI5yW&i16O^
zbTn$UqZPOMxwl_rB*QE&FE{;`>u_LL{A#*Kn)#71o8OTiK5OB<(rGG)hmuBmqwq?9
z?1AyX$Y_^5qZzxe^h#GoWqcp$4WtWo+zVlzt%*?E7lQMMq?FeA+w5q1<-UsV;7ffE
z&auZjV(>fl3U}&gKJmaF%DPzVR1<vf@)S$&*cY!hs7U}P-~Crh<L;M8)tYu?kh4N4
zvv{YX!PqN71rOZbM%%XUAxExfQYXJFxUMlTr8;kj8XwfL@m(%pF`MLS-&A&|+sH0p
z-P9N3Oh}<P9jEj>G)rfN_KB;CYJ2!1m~610K|al>)q5rIYbS(i=&7S9+m*_@Lk2{H
zqwTO16Vm)wnw`UYH-uN$opwDax@$h|+B&6NRuC)YqUlMuCg<@IfeRzkwgIp#i*IW#
z^{ov+A%^dJLQ+z}`s5F4fOV1uAz$WRN`-8KM+DB+Z#KFVJ9Z=RWh;^Rb@i1H!baQZ
z7Cta@42@sTRID1gLldRLtazA7zjLFYyFe0+XC23G#M79H;1cFYE-`nQiXKc+Jr<|L
zTnUm9EVZKJBYLhCfCx0cei8OY0=4wkH~>3^d1gtZ0JOF6lkP!tL(T|^l_C@p3Iv@<
z>A)#;XlcLz*2;9#&<PQe@krBt2IS@ee_#!^$80hkUlz(;(u<IxffjsjPg<^u|Fjv|
z+)BCq=3RQsK2_Rhod3r>QSIvCsLZWwj7UFX;RY5|?_w7MBDjeXO-T<@SahgG-$O3k
z3=c`%Vu8bp`|0n^{dknm(B`0L$y@7;@w76_Qu|)^1+{1zH}}?0TufQvGt+n5LApKo
ztL+l<r@4}jX@25uRo*0B!}Uk04ftxd^_$L%W6W6xR%$D+*M^B4?b1(>TP~Q@C-FA~
zt6T!q)L)d13Snl<=Gr&g7cC1xFhfIJc+K--Wu}u##{gcACxhfX|4QDDhklE8J!=dz
zK`ocPA_t;BG~Zvbqv(iQB{Wl9qKdYmNON@qpLT}z5XT_7UhHbE@+FRBW`PHtMz<c_
z7h#$mcI($tUJ=2;F>{wcEB;AwV@Z7USOf)~15QW~(-UY<=XYMRqFxT79yJGQTIbu6
zv^pV3_EreFL<Td>$N2NWHdnYuY0KfUX<FjscbOAkh`$CxN!g0kJQNpT=AF%Zx>0U4
zj<66Ts3bHq;d^K9nlYVYBWgYSP_zr1)8xc-kKA+FVy%~a!Y>jh==)9OA61M4kFRL8
z@%B3>hRG;@6ekzI%<yn8xQhh(4akMR&v(@Po~2uL(5RCL;meIU^xR|Y`mXpoOIt<$
z>*aEx95cR3wk9@3vkEDi$TJ4Ah}Sitx$e{AkdMc%C8iuDK-cW!fRZXIkecP0V?8<v
z6>xLgxN5t!_M?30>o*jZrO2E~=#@n40s+<P{X3?GBR#jyg1tccl(<pdCYxmY+{zgX
zv2A`N9S6CL&LW)f=E{BWj5k~<@H%vq$Qna+q;@Kqq8RTil$@0C-R_u6a1FEzRaTnI
z)&6}aWLVoKoMcK%yMSvRw|VQ~ll3m79|DrXm{S15$&>Q1H7Jm;!z>N+j5GL2Jnn};
zaAo*VY2eQXB!d>lb}<Do2${#*&VV#X@M8dv8nhyw=_h)wR8(P<g*dRjj4>tfM<-lK
z1!ZV`ay?Y><||o?d!6V}bV`XV%|KzasNRk|hte;0S4@Pi^c#sEBKjar>v-cpJR`*@
zVK8NHe$V;*NVe=(vR=+hBS1>UQcpxC?)Xt?$jzeOYEk);5KNaR)k68w6wPK@-M{g#
z({$?5dGzzGm}6E?KZ>4uf9`=}$Jed4g+~>7meY_0gE)hF{jPfWVaxBghZ5Uw2ZOXr
zPu6My^ir02HK7DSA!dAJ-Z)VD90x#ZF=>w<sJ+o?d-KL-{#DxFGaYImEDDwxOe;a?
z0aqG=q~_z)a{v99k*6H${{gv=!@+Es!qXFt)5Wh;A|&>LQ7LbsoE<eMr#}x1!gG^f
zNe}YwgF2OUj-yXFLyEH58s*#G0e5lD&Z*bUcU%l2{{!y)nTp#mJ4op$@!zWqczqV<
zWx5v5YIJ}UASnE3{P1Nj7uk{(JX&<>dk%lcawtl!dp8>Tt)?{S{gUo`upH!3jyKa|
z<%S{gwwm|sevN_nXyh6vam@PrLpa4ckB!A$v;=3}C-#P}8Fld=WfYTO04#+jT$g%Y
zcliFr^RWMbw)_p{*e9K`AC$&LVQJ688d|3U8bn7avryBGHCI(9ZFXLc(xImqo@?R)
z3kxNP56ZJn%cECIlT>z$_h-c^O{^ZLSS_FAGjFzt0KtB|bL!vCxEKi<tB+L_93CJ)
zu5l%?5_#+1XbCx~{_VjsxEHxDeMR+VSqT&T*jQ-iu*XHNRD7F{{G{GvJP-v0_?3+5
zW8!6bMpzT431ysM8!jP511vr3I?*V<9258w-i=COeDoGqM^?;`StC-`Ef(&y!MoJg
zoh;9Y&2rMKq&2QfEP`;%2@#0j8ds<If&fl*I;Fp3HsaU_k!5&8F)5ub(h7#m+tY_-
zc=Cb|SS67R%yhQ-1OV*uj;XIw%`&~05?s{gct4QEcr<$6jVoi=cH3Z?c)D|~Yp^U}
z%mh5Lqg(M*qUwU7{@b5;x2(E!1a1$nJA7oKQx~rl&0udn(oIbF!l{##6hosn3gf}#
z4}{1^+I**kq_fgw0AcYEJ}mfSTgOG)0V0IvXX!Ebdg6EzmQq411`z;p#e<v++1T5`
z13x+a0j=_EkiP6+!RpPkgtTs`twS^1%ZXXP^}OQRNu6S9>qmy%!P`Y!oI={Yc9*dG
zQI5j)=4+-6Z+tYOtUnl;U%qRf%l8K#&dl~>?n8mJyACyTaLK+XV%H-7<@MWLlnI?3
zA3WthK4QXwwdmpHAr-9bfDGSlJ6uWqGPLeG+?&S1zR|R;IFPk3_4OyDZu4^|b^iua
z?WqD(DS8rJCBQj2S?Y`p>-QPGEWKK0BjY@r9js!-Xw%2LnL<bEN{tf?P;RgA7E_j9
z0IhTeZCmCgcwUx0BRar7?xc*#jpgKpe^PbG9o`@=Lsl3ep)Sxl{~EXWI|!iQ{=52v
zsc@&|O9u&ylafxK04etq6f{xvRuA47W201R4DYxG;lp9<`9N`;DK$H5+;}wp0v^v;
zDb8E9s|UG8=i$lfF4M<>cIawL1)(QpmZ2C2j!`AY{C>JzQwPPG3uD2lbltL+s|EK5
z>^wWr<5#<2oO7+oa%`DKCZd~l0m*!A>u4(%hAF#CW(_eY`{V>UpARp+eoR0g45qZ+
z70EUBL=M(tIFt!}O*unuN3rIPHe-ChZ&|yN&`a8nkKzZ&QG>2!4$l=egl5J$mG2Em
zP9g)?2U`MC{c$kD`zMx<eGwBCS(L1nHUfm^aRC?!yY|RfIof6rr}bxA<!1LHV9e@n
z<zF>&^nm!p3iwG1y|yiZqm_!ce9Y%X-Ppa6pxieRvwalsHY7+R$OIGwvvh-PKLqqh
zf1XDOLetMZ4#&0S)%Ps-Qdmsfv!5RLJbE$&(b;-F=}IH7w45cbZ9|FsL_9q}0-L3L
zgCK`VgCw<xrx8?yGXn~(u1W{vfY8jl8Gx)b;g1HDMJAyc)N&k)3a*)RZ(I}*mlCW*
zt`!rhzbngJ(Nloo?Idpq8l|=;jz}Cd`_$xg4XgK~AC@HE8H-pf{`m8+(elAL=5<Y+
zkjxw5lWyj4IUHY5@Sp16x^kKNvq6y9wb_<Kv30Wu)E(B(X;J5q|FS6tn0#@|E!(<~
zX6|JHGt=MuutwdI7DdGv@z3<C0)dLLGseyrU`ns5Z2_<)zj^<3ju4T=)E7eKoi<M_
zqb>Pt_4+k>sl6*j4gHJZkgwFgJkxjBpSN+cok%xy5$Nmq;g(hFv;^Ja{>!D_I_EPY
z%a<M@o}0?5@7^QlvAPa=M`N+6_YEcf{F$PqUAqO{=kRw$9g?G-Zc~T8kxWC^e35H;
zd0z5dV)s#;4%%W`uK5{NQ?r}Q#m4^N%<Nc5R-9JWLY(&IfV<QcvhIuW_kUzy7GnPa
zqlQ~2bs9MJ^9m#)yl&^~YeK#11nHL+ZElhKiK13}aZ02RUpgX`Nl|s({nCe9hO<Iq
zq;ApAoZ;f+Oi1+)o+Rz>^JZ^j3yp81mN3`vV91cv!!J)$*M%3t{j`5QJphEG9tKV*
zVhA8Y9rUF^Y9T^3`cxdGP}WY&;NzC~>Me7L+h_TJ3vGQd(YU8C>p2<8pu9V{9XLUH
zQoH{x{>{L*K>CuG$%uzhmED`<_UCiS*<q5|xJSj<Ae}9fO*=q(T!bG@DD-a!^JRv(
zjpD40KQ^s=Y1LQu4gLcn=QM;0=I_d-iF054_(KMpGmx}jpcj2%L`#(&^V)uds717$
zuZ{)+D2!<?Hk$~Rrs#KwWgnM$G6=!bGvRvi%F&0Wv@ud*_5L;oZOh2t7*4YnQ`a-d
z28zH$z<OYFh>Obn4Ne7}%C%C_XD8(pc!2U}zm_`K5+8S7A(d7NT&bVKTL)NrS<<Y4
z0S1@ODr02d5q(7wL4)MNs6_*yg7j2eFfG3yEyZ=I*r#+z1`VwT$wP)r#fw~@38Ybh
zm;dGvvn7Ma_gC{j?kqnS3H4_<oO&pe{un|o;K4NmFS>M>XO)$9aLl4Fjos=iE0Ve=
zw@e{YS&QU2Fv|vK5)g;ks1Qs9zf;uiVnWOY{s*Qx{Aq$Hj1sJ%pzEfslisWqP24SD
zi={B$w~CbktgOWdZ{x{vI&ZY#ixmI{{~WPfL@*rmkU3SFD!eRW<?Qtr{~Yy5DUO<(
zYwAbuFQ<&pgP(cJZ(sSCl<a)8&OqI$V}=-NLgjvhkO?9$*K8!5yLPM>bl`3ut}4PK
ziM3P}rMc3-GkOkp<{nFV=FirejFjSZDsS1_V&R*`h+W?ar2KxJ1e}v7)YEu>IJqm+
zU=mz1<avBmM1QxH8lc^dwL}{|RS56RTZD`ZFF(ipnE|8Wm%T=D<+Y|Si<6i|D?N;j
zF%G`<r512z@sxXp{UV%GcJW|@(lbs4beR_es0^#JIGSktMqG0r6NI7kJ{xSd|M@uI
zW#z$gnzBbLeZlK<so;8`V#n8uYa6#>G2yL|NL}q>vRf<KvU5hwW!=-Afg*5`(EL*(
z-GtI<S<X1b5X7U~v`yM0hn`0s*qFAM(wUzQ==gJ`;ETCe^6>3^6H4P2mvUFNXLBZN
zB`l->4~Hw#d~_huUVYq2l)39ZY!jAJ{f&F;H-6!=Wxe*@w!=7?>adeqdj5~Hg@aqR
z`)l=*aa?BFGx3My((L@N3gZXM^{&>B6u;Q+=&LC7EUqs_sotiYpmjfi?cz-sxsT{;
zzFRGIgmi1@RHc8&idkF`VR4o;`!nkegU$ABl}YcAOYI2>r7j;hkm&Qut{*Ckozxt-
zi%@u4E)4J0$)=~FW2MN<##R-vc@Bq#F5!HZ9QA2uX6AMpVDXtz#H;r0{tSu0X}Lk4
z<Z;)YDMYx_z(gv`zN)etjGPH(r?b4Z{|N>ZRVo<^F8tev2mE*y#oNjVosCEq7@3^_
z^7pW-@xP`huwhW}_eU#^x5a@f5drgIuogmr>%oF-@I55#hctyXVhKGi!t>iU!;+fX
zh?X{MF5Ovw5M#fGe6okNN2QIu!HG`%q%6C^u7m3<xNX9_S{X2}Yn&G%AH^8|>(Fgy
zRb5g){N2s>Pix;VGMRk4!TlegfR)$T%lkbap#Ho!vKqd6<E573ZQEZd{{eqZ9F9${
zrlN%7HXf^}z^*s>*+yv!K6V}ub(>;3R^VD<tmjyM7s_?UO00<~a%BT(?_;j?`l)?%
zzZoX~m3E53!#{HMVD}y0bN${W#iUr*qXJaIVkh86z8mwoa`6U5kutJ;3x8%-bGNkp
zHQqEqwLwh#jL0a#t>?P-AaFJ9m8?HL@Zaw?yzRO8RP+=rYw1?2X6cdZLa{G8^K&)^
zfBKon?DdB`+;g9dF+&$0e-as2&~7}~bGw^G7Sj?vERxv~XHbe>FBGIy@%Xzm@ULYm
z(%MZ^_17`)MXo2e<H=7V>vbm=@$0Dx;!Kr-{k}U3S6%1wE;Z*L1D^$A!)D^w<y&2Y
ziW^(|(LtXkJ277pPm@kN;qo+8TSp`B67%^VhVps|(Vf*R9OlM_y_iM^kv>g?DxK4U
zuS5RoBlR*xL79K4Ro0{c`uh-AJyU0X1eESfP-a`;v#wT0_)?SL7wdr4x8~KtNo9=Z
zNTolkukdOwPh|$@YUI%i?Ox$8@2`81Ijl>)6a^y@ol$$HH2a%4cx?grWiTOU!I;8{
zH0*<z<Ox>sZE4&$E}FFa)e^M<*DqivdVSYC-ENHD6E_JFFcHyV4xe@vQj=Y4XIVb^
zT;Ae`c2gsVvSO$d=JGbmC@Z8|)G3vXL}`}(vRl&=+RH=+H2Nj2gPQ4-a3RJ~oAF56
ze9p{RhVNk&+HFw9(JNyC65+Q|wc|-ME76XmuC9$&zNBOm%IuhICL7~jN_xQ%YbZpB
z<{DK@81Qy2MVjLAF*(1t)`DM5`=MUpRkv_jXv)<)=(*zXK`S`ak>F_hOiAl{mYYvy
z<n+*(wL4+9MRlg9>AMem2ch}-f-*>jn6T5A$7KAP&VLGHFR+PoWH6bpm7FpeC0AU%
z&1#1%e^g#?P6HIXd1fSQpOWY@Zr!8WAbeM(+Z0P@a*yOAVp+aC5rM8rZG4C4i=6iz
zBfEYm5t{E(`RZL@0JVf!<jF?%#S#Qb+t4kEw=&k6qL@T4WXtn>gC@#s*1=8OPKc}z
zTm+voP{ZY;=U%J-egSLbw7aanqi-&e)CMip-!q-grd-W~GRrcO)6=!4)OPhR?bZ0Q
zhHw-a^boTgTAe>lHSO)b#6MHXaY(sW+Ythzif$~Etk!~aa4|3XRefuGRw>o!DX^HB
z@p{ohEL{=fnJiaVXK;|p>92n2Rfy^A&05oGBc?v{^YBHd)YP2j8>ER37aOEc=5*sl
zHksjn<zKK3uZfqhIb%w*&%9InmQi64t``X0hQM05|A)hA{D6srg>7bteX*ddnyqHR
zA^tfXX~449n%PD)BIKGR3B?#<`I*pinQAS?UnIl~>t2=}ZHLt2D?34S*^&!1$|$i6
zS6Cj!GsZ<b1`^Ff`t8Jg3{z{d6u(Z(_AXhe|C!rCWpo;7xT_6mhl9S^ZYihV)f!jl
zI22?Px9oytxl^avrl3OCs1qs5*i<r>)B6^2Ah6m!dJ19OJNMFi{p5dZMEEK(<6$LH
zBqYNor)%Y;%0O6v4YSEra_2WB!z?*33um(CWOG6?l(Nql@PBoDQf--)J?9X$sr@Is
zimqL}Ubr@po6_B=ML%HV_5@IC%al5U_2sL0k!J46#vcjZHl|2IEK!wmAD_ZJ(^W9z
zgE?7FY#d!Dq}yV$$=aW&WP+(BlK-}1gRAFtw$tUtT(T~k9`rj1(w$VBIH<vB`4Hy`
z)A3FLPZMy|rX==m;wyu{*8zLoLl7N1{9+2Fbh~>1!@&2b(sl~ZH_(TrjNGw~qT{9Q
z6cKSm$y*yb4Gr+{+GtKu*)-k}c5POPQ_=ryayxuItaNd*mev5!C_4t#U{Cgfuyg-S
zK@fMet76Iw5j-U=(f0$AeuJfe0y*JP^b|$WAOeJz31z<}ytSbOo#q%DfCY`026oJ*
zm*Zf;EO5F>LsJ7Bt9Yr*;aoc15-umuC@qG__x@YH)RwIrc3X$#al>lUsWbDzr<$xZ
zO&eCQJZGgvUFMF<yn{)-UvT|rr`7s|)&VQJzjGzRh4%}51=nsj^?^KUbJi_qaj<-*
z5k=9~nMl1pvi-g`+II0S{cfYl`4b>Ar|;dvnI@%gpI@9-9I4a9!QMwxcHZh{ixT2#
z)%oUw{tdFfE8~+7mDipzURy%Lg-e5<6Iot#sg<qug^bXSQu0WhaBb7ZSzFrBs;vG8
zK+AI2Up0En*XjCfiwm#0M|k_+ul#bEdxShG>y<UR3i^>IpEk=rWa%>!;t$_&(XlQk
zgUM$W*#TX+gUK`kJjAW~JN$UpyZrWdCs?RP_((cP0qm!+<a$fcSfxM__|B7AZ111<
z2Di>&X?UNVN60^&Ddt^>eY52_-Nh=d=3PtwFSJYE)fuO+7QD`T&-LEw#57avoE;mX
zsUq0a{YZRv%tjSkKeod;J}t@H>35&}FS#~;-TH;a%%$c+tM8M{xRS;y<NV%oa(3{;
z?~uwLw~}x<5oZ?V{{dH@gH$$()7pD%R0OS?W{O6AnasUy|AeQU(1_AQsMxT0T0nC(
zxotf{%uDc(-3m&?3B?|-^T!^I)8t1=!b1iH{U}Vt@hE-(6|(#i!@Y*kmQ_l$=+4NM
z3EzCJIX8RyB(YX#c#UGvcwhD%VKDEe`<=D>v5fA{8>SQl;oIZrU2KhGypwPbW)^nM
z*&nvaTS8k)6({uGHGKJ`OD)7|fSQtdbXclX@;{)Gh%`Jhz*7!cDFdfOTE@`NZfT|J
zNYJ?-TPw*UeD5ACPr7YW<6O491&c~a#Hc$KM*)>nLM5d>5gfSe>V807nkpTTs)MhU
zgQ%epbM+l$Z@IZ)0Sph8q?7U1FtRuM@S@%m<e;FYLn=Gx(8{rm=^8=|7jfgg=v&fv
z4c)}{Pu{FQ{{vtQTaTvJEB1%pSv||Z``Nwk<3r<{*coJ5+En>0h1yr~!0>nY$ga9J
z?xZW+@%OEvnIF<hE2u3?1I-5?AyDc`Sg=2(0QNPlYwq&3FL1EF4`DSy<u@_-w?wjF
z?)XN5wZK517tI=Cfop&XKu`U(&Ac9edRoaZ(CmRqs|OE(+h{xub;xoe5OMBDQHY_K
zuT`&D1Ia)RC*@u%Xz-FQ5q@6T=M$&<1gz{Ur|a5tb?d$w4z2+m^y!)_4jk@_SLrh&
z=Y?vd8wxU|udT|`bua9R{s-I+XglOZAGx;OvNN(v&aMjQ*tw23HEZ?2My`j|d=dM%
z_v*-6<ZJ@|J$cS3KTIG|E1&v*D<9TXOXh1GonTK2EV1(E<^EFFfl$Z%|A6KIvLQ4!
zR8?5t?i0x~y<Q>&+fkJUO2g>dt+U_E!8In{8yVi5bxUpy+}u_s*Cy1#v&NeatyB62
zEitm^(#-4T#96hT_=nbLw^pnGhcu*~OTjaDQ}RpXu{9R6+K@5N0g9@cO+v8@iMGo4
z9BZrmN=loVl~4NtfKB8^dIPCX=>_>kQA~ibpdT%5iXqf%y0g)gy1Spsi_$+Xp;k|@
zG7cLoqk5+WA$EJx(6Y^~c&n@Gss)N;?dc|$kh13LYZHltAXD`LI4^r4DXqAl+0E=@
z)4kA<FIPURmNI=!RF;K{zvPgjMeIMtoxZqYX`ud-#F#Zy$@>H|&2UOAenMMnd>w3y
zQ^LMiC`Il}_)&J0t3u>lDGE|r?)*CYV<T=QRUJ}Pt53BxiE&Z4oE38!+aBV-6H%3Z
zm9Ks1Oto1*wbO{c#bM4RDS_OQ%#`An?1r6{vMT4g8fR}!8UNBUWZa~HL(_L5UBzyA
zv~G$yVC+F#uP-lb3MIy9w6u75EEKZmBzZ04)mjD7bfDd?=BYY!mP=8%9j*38bMib?
zTEf0*%gOI)Li~Xh^TA{bwdn7WDVEvwHN-HX&%5(TMk-GVi}QzJ@w#G{Ug@^LCYXru
zA;EsYLWpk_`!mM0I%15<IV}yu%oy<s2PsE#MOm#d)83VPn$W$G{Wm6CP#h%bXdWXe
zewI9nBvNn9WvTy}Qk<O2Dwv`qugy$RZfcN3Ho;q)V0($s(v+e^f>{yhO(fd*0TUFk
z4RASOVMPeNodQ&Fe%f|yL1!gd^Q9uRTiyKue*UlFm(+hRg1JT=Y(OWo@(TX`i}JH^
zd?Vkb20ez9=mJ$WkwebiL#K4sYaeR`$l`|&U&Ky2^97RDHLQgOuDi2Nhk5)56pggE
zOdOO#z`8u3sBg|<Mw~}4uHN`ia4Gza-0~KeM5da<@PL2(8j)K3eJ7-H(j18V9x`*%
zAMKP1y#7$LBHYXIFS3lb=-jwy?2u!C<*JSJAC7CGy>-%%sTlqnd#^j+h#x2u&t(G#
z@?JSd2DylazS63yW+?$Ju-{kPW)uv{w(6wrRifM*zt++SGW5wqZVstjF8uc>PA%A)
zt3PtEU(c7d%s&$%y6U3;7kpOIg&vHVY7;P<or;;!_P~I;_H=QI*@m-+aprY(N83#B
z5K*y^s{4H*i1o5Z6e(;eI)=+3Y%c{MkXJ1N^Y=yXXj}547CO_e&Wg<;+V<D<)RaJp
z>VBo7!U~%q;D7}ihq8Z-8vYxaWolqOxsi}_V!vbuvG3yl{^-{3i`KqSSeMz~a&jjl
z^doFCHB5NV0&@RO(N_3Q0qg25^P>Sffeqf<Gj_a%k$a{pzwIepk_-YrsD6Bq#<Imd
zq=<5qT1-jay{VTp{*1ORV1Y#^1PsZ2kilu%$3e8`40)~eG*O`GpB8MG*N8&IIjtOC
z6W8&hz^a4Y+O)LlV1u~KcGTUVOjB>ab(<)`ooLHLgtDg`G76;W>F|`#byV&~o^A%z
zY!TZ$<BKySvHw)<ZODkvj<M0aDehzHs>S?>(1oJM2Q`U*(Pv3dJv6uyBv@)2)2msE
z+lT|U4U?DJG$<1oS(GV=<BiKMgK|Zs3YkRjXY9y2Q*ql5H1H_7X~7B!8<n3p5GQ%m
z^>?%ggG87o#F>@xxlB;<Tar?lg~TON#ZgnJmaq^e3S05{UWKscI8>=+9L1JR7!Ew4
zNmCf>Y^|)tBb{x#*2x5LpbD{VoWxkN^<Y>|=CuDhyvkr~y%q47(B(Cg%5EC0$v-wL
z-{1LerAy=M!zyA6v@m06QhdnD>fyLk*rfHun~XuJO7?4AYbkyPw61>Xi$ensw>q37
zs@R_QhE*yuoC&*zQ#4*ZQWJ9upETmD?NDHKZ|86`m7A^;I1uisQo1xOel;?jvgTTE
z6YK4ilKUj<JE3tS?|Sj&9MFNmmjN74`Y5i1Er;OW2Yxly^6F?$z5Zr!<{zTnc;nup
zi|fLQA+M7y7H7TcKmF;`#Mp7yE9l?SrvY{LsybFz++K5IUJUZb3&@`i7T1ukz)}WG
zzdxI@QyM0wt#)mYiuv;R1Wjzs(jx_>DWjv&+VNbEeB1Y&4#K7>Yck9>ieayp*!(|U
zm(H2AVwUFH85UG04J4*3O+(>ZgamT>WU<gH@4xek_fC{mp99Wv%Q-AEjghtmb5&Oq
z?{lbpI~2Ps$rvTP({Hc2U#x~qMxCryUhSW~=h?`gv-Yk0`O}x1!ll+M389;p^Cu$^
zH$`mkg`T*>z-@G!^{-gHQHiSwHt_+5AQP49h;YSCEGwJ_zr27te1s57L-#!C@t$;X
zk_S#(Hc^<29o07%Y|~S&=`Dn^%|ayTIroN_o3_VD%<R*hz-+^`fFGxoU?IU}FueBm
zH0X<YW2<n->oi@G7x|7q_akW#Nb()bpo_CnNUU6ETn)@7?0gIE5}GM(kei4+T23U{
zPS>J`c9NAHAM4fnj9hZkD>N?$`&za?SFXg{bp`9im*t*TQas*V&eJgxCGOx|k<#TL
zU@BqHJZ4Xk6BZ0N_M_A&@`SDbaR@#i&vJM?>PI7;%~4j#ai+MxCYmfI)a3w#WVrUN
z=dR^VG1$S=6(6Z@_x>m!*kk|u>5Mcx`b^=*=OA%y{<oPlz?`zlRWEZzFugu8Fr17f
zWRo?0OGQ(~v0)XTtUE!Y*^Z6!;iNeYa;IS3F~ZuMC?ayYGF@^<Rf#|RAZMd~Aebh=
zg{hGDtb8aqf5$*#^S+4o>CSxCmZQyHI!igt>m)|y)E8+|zn##m1u1h9jh0$%8x@oH
zy4T*jIdT|?>So8QbVGqU5@ZSE7;!5<RGKEx-;YHA>)XCjw`?krezm=8TUyDx!!&%r
zEsiB}5OaU>bC!mPJe)qrbNOXy0VhXa6BN==Gd$11#6XtvqG9hh1$Za7T)AN^j|$2K
z;n`1V_`1f$3ltE1<2gTn=&NrcuF`L6Xk5Ty>zNu0NGU-Xy2L0c$$0L)JQYaMgV~OE
zchA-~x}eNKcFq%1v^Fx4!oSS6^cO8(u^`J)2tm3)aCy>gBYL2Xt#zJ~<5)5J)143)
zsFuzV`2nkQJ6r+CRk$V(E?)CRbJ-EY{SZ!Ye~42W-iYHYRxc5kLI4F*MgZR=`BBsk
zl0i57K1q@mmt6ujT`C>7;=tQzz1LNzio(BhXjj<mbEOx2n<)b=Gq7J(bD@KtLu{;;
zhen7H2cJQ#@IOh-yF*fp^2)KxT$S~$IkkRGEJx7u0a<kZ(647S^Udw20@la>TKcng
zmg8S6bIgd4Tlfxwl&>llrho&Ol4Hw5_XF2<j-@6Kf;eqFScum6;ha8DGFM_I;zebi
z2&6A2XE1O<>wDgNpQub^%$9q}4w6ew{8(NT8|scMIj#>$HK5Wz7h&)KN4gXrm0y_`
z!xNWe^caTi3GG{1NTK=_i(5^<Jb5-%G|)vOul`B8d6EaFU_3tP6DaVW`J^KxP_67&
z)KA{|i};%Z^tvE<Lzw^9i(PiLLh$}#&t$rC(fv<#V?Wd?)NI=tjikFK3F`8V;$~u`
z0GE$n9hfo$$#3<(HWmAGelcUqviBR69te+S#Q?`Hu4b1aa+1NWg~__8vJf%|1LXgH
zE?iqEZc7gbo8AicLN2C1k|DG7)P|ze&R*<W4yUQg%(pCaC+-U?jtK7;xgsN%rmmGQ
zcfNg0ntFfBK$HJ8mN787)<0zo*`7q5^YGzTd{Y5zo)CG^+p}O;^{IT{?cb@x!|5G6
zl@#3>3cQCk>dt)L+}V(NuA}|$c~WF{W#fxX%3)aGFNd*;2gVW>N7&hl0qRiGgBu)7
zr{nN(Pn%(t0r*pq2q^Ges<J$OYY;=#KVTy0uSa2uGNr5@i!7Ey_Ra{Xto(7*AEYgq
z4^z_yb2MvmpT2x{F%??yxpjul(R*?qb1y}%POU%YC7+fa%2n*P%F9Gg`v-R<uqT_U
zPX7TG(=DnOC$}%{_9j@)mP(-c^By16c1Igz4QR3?Hh3kjO3G@x--vBi!kcJUe)M-y
zmq<9_EnQedw;k$dar#T;1{iW2&zDlRJ%NUwZz1iwZB1`?xo6L5eD_x_slIYD^q|9=
zDm1Frd4M)xobZ{;FV}`Y$D~%j;;<j|HqwQrxT1J;XG%&%I77tpt(ckUzVNP=NX->z
zCoR^C%mjS||K^xV@9SmPj74${2Smt)D0vz${g8F_8!3@4>dAk|YVG>nM3%pL9-3K)
zTyzjqegTb;T7J&|wuW2P7xj?|41cJ~y}eQ~`80!@df!Wr+$#4#={7*~mEb#YkpDGS
zzNPAao096+L@h*$^zBq3r0GI_*Uxe`M)X8fwQ2tlrM#6E_I`2Rx1Pgm#c)<f1^&EH
zN|<n|88qZ;gL1S_S-+YNlyrmMPNfSv_b?N_Qb{G`p9~vJSsUpNrlA`V#%!t#KA6}3
zsrj{`_zlgyxr~CtqtksEMo!u{n+TaqV;U`tjo2wk0wU$4tw5~R3eUKEiLjC;7jI-c
zJML;6y40Ve_L^3G2Op4;_RVJ8CUvM^?2$6fHBmw%OGbpOcS-nU6tkavvUc%RvT?C7
zEnn?Yruyua88BTHl&~3r+Uu&$a9-$7O|LU~;P5(tnBgIVi(_y#B0=qVZIN_5(B__Q
zkoYbQkf?+b<M_9EVOiMgMZe-56~Yw&$l#j!4Hm6@(x&dLjMsGpt|Ci{O9u3rDeqd=
z;_@r?Pb)pt*{vpTlS`ArchT3tw2*FMHhjmBpF2BL(;=?;Wju~1uU3h%tE?5;Rf-ms
z0%ntFXhOipO1xB-Ltkh{%zXfibhK|dQt)=U2RavIQG$qBO>*w}b`nhEtONp^g2dci
zqvZ;q|E!MgH#Bba{#D1-ap>fbKyLK={tUW)Ld0oQA~=2gYubojmT&o=WND`uig{I$
z%W3fZGH;HT9BzahIK=14c@3!d2N-<nm-HF+dkEhT&CZaNTZKrP!W_AOdtFx4xsY{^
zw_6Qga3~t6s`h>=xZ_}<(DR*!bsV4Qd_M3sdX~fJvZy_z6DhJCGT%w*rA%CpcRYVJ
z38ql$5^KyG49s>l*J{n~JSk{GZC=3h&G2<sLyS`*_2=iRml|%E-G~@}1Xp(2&^eea
zUQpy+ESglcxScEb`**czVe`nRbxKym#TTW17mvuvTCrB$Q$y%$7XM7eh3Is+3j7t9
z%=HZ&%|kmHkVFiG{Ic;h5q~OEqJCi4M=v5R7)4|cAC9n&Ll^vmwCH}tb!VkJPuaPq
zT6%<Syb$xqg$ky-au`YO`qYHM8i=2iI=^@C^i$JBA%y%WLNd&d{n}|5dLl&V#A7v%
zJqpnV26x|t#G9oB`SG%txY@03`%$n8RtWXPF|Bp-jRU;h;K)t_bTGEq3u!>aK5ixC
zl;|YLBo`9J#Exxh6Rr&-NBV5NQ0fplMWsAwRadfaYUurXgRB^o?G?^7G<xSw%T5@o
zuu0y!Ue#Pp-L39!WVV@F_R;Hz#U+l1iLn=swX1v&K66^Vw|dZS+w`58>ZP8r(S~2}
z$op;)y?Cz8IM^Zbg9t3*eB1PvS7QhBNDaME8-ilt)&pcm27~hg$?f5$hU&nQh3Z(U
zYG0X|mxhm&29a?Y_H^Qq`f=oh5AD%LXVE`b&Ml;nXw{YH)$YRt{)m6~UC=WuGhJun
z{Gv3^P#h=D*MDCxI=|Y^@hjP5lI<#u&Ln~&;%n^>i|SgP7s(E$MYeNqV!3uIi?=p?
z`1fs-@<1I`JJ;Xch2PA%seM#w`<mi9fz{lD7xe0x2!PymA0KrF%`8(^^B+&2=rz6m
zvdzG}$od&10BD0r@>BBSO~WwqTV>j#r4m#lH17C7FcQ6MFK9K+ngPF9O<U1z8AOru
z8I-UYNEjMiA#+ybyp~<n3el2VL_TcR&)hfDd!{TZgYSf|D8&Z4l=)y|ew!S!x~?|g
zbX_x<;n@F@t{yhY@-DmUbWisWv(<p?sLxm^3*jlb49~my<m+=o6A{t)O1||N$qy1A
z?u)b8d8F?rA1;y`lez{GU0;4^Q^s-{jJ2CWhaPsy79U>zooCfNsbYGp_|K8^W5oj_
zzeMEFB)O*QC)cn_CkFtOGmN1G2rcX@Y$iLVo`8;uHSYoP2^DKVAPl__Cn}{%HOLUk
z><}UFvA0vSj(&J8B>`ZOtrHN|b6LKl(+?jkcWn|kyf$|aF-ODWYjxOadEK|{S4yu9
zLuLlFya`00EEN2G!GP6B?&jheZ)q9XdlP@_3v|2&ueg6M>aVuZRQM8gu48n2IWoe&
zY53^z@b>$!rafWUO0JT-uVh6}#GzzO*ybUzyFZ?BXZ*^8z;m~sL?_Rv6tjhLp_A|I
z-i2Ha1sPdeHsk))!-6`msBL6G3$)A7Pm$)!QiAWAf2fa}8FpZ7q$zreu(*$9l)m#R
zDk(_OElce%FvzYli9v;RslI@-W#O(}&uQokh8GuX)IWS<Pqz-4?DW*y2MbCSU12$w
z!c~CRD*YzCYRZby{BzR7ul)8l3vwRu#2V0Y`RMT2qMyd21V1gtI}yjeAyT!3nHP)j
z+fFY84GOFO1LkX^BWw4*{s)*mQtw*1;~uLyd$s4jfCRArLfvLWrK3|ryE!}fhF_Ot
zvr0SfrF9olm-8@$&|Vq^HA`2PZU8d4lBm`M(FSi$FDsh_I8l11N7+#<*_ntQH{95e
zVU;8mUyY}ydW4?h4>X;m^)S3o8DFAx>#>_Z`7!em+X@)47R*@Ubav3J+?8g~WGQbW
ziC{y(LAYjIo>xcaTWxCr%PPuodjNm^Qts6sM#7nK&Gkppf}Pvr>s}*`2eMEGE+m-d
zz%D9FN{oCut>_XtoL+vFQN^jF7_w(T=sPwo`5xfrlo-wcXXyqh@~%i|b{8Ciek4s1
z`aig^zuc}g<I|vP*9k|<GpoMzMJTJjR!`Exn3&oT*YmhFamu#4cE0$%Wh*lTsafz!
zxb49p^i-oibZ9qOvw*^`+BeP6X7rMs-T=vd&^#BCcuh-+>oPUfS;--n7?kOFTuZ$e
ztMZ$C0SDefOdgPVn^ywj+!E%BpEr>H*odVF9<3kAn$emUE8C#0wwdX2y}roVFog66
za^`y0m-+FEFIS8OPnfs~V_4e<9##O8A}sI-&-7_732zOz!|Nh%hk<orOk!$@*pVe~
zaDWHe1ra6*KPzk4vKuj^18s7X=7wUo2hHncI2TjA#cH0-gjF373i$mfxqFWnT51cp
zpvruqDt1{%8yS+`XID6^X`0HtZFe;p=ERhJAC7ldIDC!w40N!N@7qbk5JYZBtOFsI
z25!p1Z7{n#m)l^`HWo|e!5siEY(}wsf_L3El+d4tLd3<=f#64?nB@;(-@o8U?#(7u
zW{At5v3E}R5%nmT0&Oj$tqEFp!yqhaO~>I#N2au7<?iWet!Fq{AcWoN^NfQG-QB_R
zfF|P>Sd`ESYqpJO%|{R`BN3{3tdH%cTL|?hbGSGJvmO!0LJe4@u=h7>T^0Z%JC&I!
z2itqsY&W?i9$l~C>z6nzAShE)D<mb`|1B#jU@`X75^b1B%WyKJq^2$oS<Vv?GC^>Q
z{INX+tHrS?O1QH4<d0-rp)^2~wu7V}+9Jn;QA}xC7)y327`n0BoCnD|HtK@Xz=TS_
z0u)H#11l$h(x+6O+3d@4Y)l2GpbQCw3DLc}qWLNNsKNNg;dj}om<bi9G-2q0(tBTl
z_(Ag{R;ctXHn?G7on~z3wSqyHRRPt%f~y`e|JZ*0<jN~=fB!2M;hPt4y7`(VHb=zh
z!V;{~iG-P@TW7u!?yU{R$K+IOGrpN1yRb!k*K;)z>wCjp%l5nZZ(29n5kIldH8wY0
zr@5hGN%u+xWIfnj0r0?=9}<%Hpo^`xY`%Iqn5G8Pb@aK6HTaD2sYGRsO2!;FUie{y
z!XVlO_K>C8c`r7pm(mxq!*5Xv&sVF@^m@L=iXx%SsYcDUcFU&}x;H}Km2{fb1T+Qw
zd7PJ(mHYSE#q2ITzNi3iyN6`8odV3H0E*>KsqS~Y{9&@O58~^VyxG!DQRn`On$Hs8
zkMvppke~X_y?x*FXXcN5r$j`@+xDV6EB}GaIx=KNUL^5oc5j1X^?7hi;#;&kne^dQ
z<L)gj{js~-%@i+&Ku8GqD&xJ+-#pyC5}e7eHau94IlCb{be~bWZpC9*Z1cMxr6&ay
zd%7{*Zs6#7;MZx=(#4w()}JJ`<IicQ!4&<u;3_xyUd&|hr>26q*iX7G+a8HFBl=9r
zbO)wB;qOa(tNlY4EoqQrcQ2;hwW99HmCoc&m(8a(7%6JL2NsLu|E=*Vb3I|*4j`1o
zs0s`Ci06CQCI4ax@NTsEmFgv^D06>^Pf#|+Bvb<08K|5-FHR<+QfLsc({c#7!Gy11
zPRknCGLY=tgQZS+mp{TB2ScuX_js?YByr@gGgMbmq-VbhMSVe_$Te|)z`Vu3oBqjK
z{0re770xgdc#dA%f56&JowR3n%0(3@;xDYl5+WV0h-NN)i_jNqIJLKXur;-Af5mE@
zvpX=DVd+MRZ>89)%^$B*b<GDpT&Zd#=24AsM#|9q2RO!p`tMufENUf+CVZR0<O03%
zj8D^7=l-47Kbbj(H}0s*VPkUCo(t`it)!OJclz`^`Z<xpHE7<-J3$9#ditJ+Bv|oP
z@|FmleA_{?4vuCaC&1~BpM?&jnFaX&Bz0j4%{`H46>zR~y^&X^vi*r-SAe~))hdQv
zyKoa-{KL;{+(=;3Nmbd;o@Ojh=_MzQCN%KF6CGBy`}Z#!6}!SO*6U?`_i@yI-QMMk
z(idZ<n(J8ntiCwC!=1t)l*k|*9p!qi@Q^((A8?fY8q@Sc(~-SR)&wd&J!VEF&1i{R
z%QLb|&=^NZ!n3`aI0glouk4B66M;>$y9cHY#!Tjb2+MT01Kai+A}8b5MjMral_t=G
zt_I(nGZJtGA3~hNA})nq7t|Q%7~3UM@^}pWtn?O1{iuO@L|W(1+LMzPbrPi8<n5;x
zQ&8t&AI-EL0nPO52X~hjwoP9X=o!DLVm~zV_9f>-67*C5aH*;(<mU>YB@NT#n2KCq
zOb%13UeX62!`X_=bqkn)j|-QD-m;JCrDN`bCMD{22(h}c@D#pWRg<DmwWfO0Ibl!|
zjO)A&XZ@K7CrU5rPhdjqY)^FL{7nD9h6}7siDQ~}i4#@=qNh2%I6g1-v`0Z%Cn`tl
zTnB#|>X8evOJwnXa|Aox$$)jCjhg-lk%b_&0lg#@rhg8*m$Y7)v?r(UE?@jJlB~9>
zy)W}0@UrNy|MUq-&FKDZ#n<8fCCSLYjpe7Rz3x~yb$6ut_|;FPmwc`dTvKO)F>GWk
z3MOCIZBYhmQ{Wsp86)i((6$x>jdR0k3_>GTNz@1Za!i!iKUsJ~9=$?o-4nnraS-D{
z-PUj49-CLEWskh{QmCvMdgx#w;qJc~Klj+;<H-BLF+I(3VR=kZPqWG8VH$cFYQboG
z6{tdsfEg^TOk{DuNOHOTUs&Q(1=<#AcYPSB#1D(-tW9bNZlCiotR>idXqIs0pw#gP
zAIW>%qT6s1YkiBE*PW8u>%l7h5F>vcT=c9N&qAeAI)QG|ZSG0++2odGuAlLu@ij|f
zaj8;ql`A$H>vM>sy&(<egX?T6-!iUCmD+P;p5s5WeTw-UPqkJmyW~-z3P^<Rr+c@b
z)=Fo=4-7A5<Qwh}s-IN|F_$3f>_0wGY6PT5g}+_rR$&8J-WE8*j=Z)C&{$Obf!knG
z&~b%?bc{33dag?AaRd+A%qxP!Dm36Nun)G_ncP$!*sTr%5*%GjzbiQf4R|CFnuTb!
zZ4AZ5(G-M>_VTe8VBrPbfRa8o$u^40gX!3ciXI}gJj0i1+GQE;Rh5yFAsCVHU0Pl;
z7x$?xJN~C#NqaV*#A286peyP|M$Iz<i+(vG$(k368`?EqjE2R!I;W`<Nu!YR7FxpF
zDOb>ziSy&NE#Rk-k}8$A#vmn?rt$U(;4E22xdzfcK@1@9d@V)ujOa=Bu%SFTDUoVB
zm!?ogZkB@Ojnq3S$WE}EGZM-YTdD{REBD|cTBF|npb|w4?}vK~OMW<^d=oEv*;py1
zq_0+X{5j@b6=H{ezw}-XPg;Fx6vAGur1DeqdyLs{VW)8#-?G38-NDPhgxd^4_Hcpu
zKkgPBlz-v5`8t9s>1V_mADG<fENpq5Q*tS{1TA1Pr2TNkDp=NaV3k!wz|Te09m<Z=
z&8F~?)8~soF)p<Rq~5<^{-$Ua9po0QPf}{-XMK8V+g`i3RJ3ZC@aH<~nlQa&>b<j3
zsqh4K?;!Jkh^~lU=hp2jd4icDC5NhuI}wzN`tR}&UvopYjey98Asf1CZnF({eIX5h
z@aLB=w9dm+GR1IM#L*VZzI~Y`_JJQnL4DhQ1f4;Mwx`hFBs6~oF`}O%IeR<Kw&#bB
z68=60sR|2qKQ)FcJrJSa;)E5<zs89A`3f&`C#n4O1!9cj<RW?c47Rz*_uQhs>xHgg
zd^`CwH<=O!ooilA0^TI~NS>Y%S;Yu^avPa^>7<I_k!X8qimt!c9akpVV*ih&vy5u`
zf8Y4V=$7sl7$6d(Yk<J$MoL76qzGfu(vptu?hXM-1qn%|yQPut5W(;N^E>DNd=GYZ
zw$FR-`@XK%>-CF|GN}s4Y#`x~>T&=?&$Q*D&&k-PX+6hRE!17y3ku_31sCJRdq=2b
zJ)6#4g`3k#Mg4vB!ZoN}$xgp`^_!X6G1Cp}O1{0l#PHd`{(^#{Wn3>aiG4FeBO!JI
z3Q*hSt7x-&=B!1%4LA%DPb>43q#nF=(cI4(lZ=VFcjBopksJLcOD?=v&CB$hDR#T=
zo@_L3UFxd3KhiR<<k{}h-UvHJKPl7u&MrNSD0an>4%nev8%rg$?ma6a^!_bsu7;NC
zAoUMdmp%8PT8&2CJ~z_E4I~fRZBZO~YSoiL3@qU{f+XIRva7K)9>%wA1TQBtW*k}g
zx5rUJMyF<5p2)tau=scEkhqTbel3#(DZ?oK?P*;a*1kjwPa8dTy=%GS98zfI*}_q#
z-9&lrc2kV*GFu$-S^ZJisN|w=Y$MHL{dQW(w*_8KJk;P)++?jo#kpugXeV8{owtW?
z_!c~~F|_QmI%B_358B0Y#i?AMq{HM<Ds(VHzo}5dMtkcHWr%;2eP?aRx;%+gR_{~h
z!43kR=sNp)(S4C!TXP}}%sRh6ldV))KRS9r#qNpTu(x$=tgH``x0<mW`rGvYJPal~
z<Px@>=H(XRwil70^I<7g4O80-DiL&2(1^kOp_yt7*5EH0%oXd80_TOYG)G3xOGu3{
zhlr@)dQedbqJj9)7~WI~EzzUTE?Y@)y5`Y7iZb0GFDh|Z<oiQX5J!>YJh`&HoGPE7
zfpXYSoW0b`5kwj?{(A>45$VND|F+y&cbTZGrLPBliCZ0P85|nggSjxkwuw;S_bd>p
zCfpA*J6+fpoy@3Edds01_V?^`8F^Omnp2J(oYUdIw<FAbSj2uc2;<_0hHzWbaK(1)
z(>>|7a~IFw<-yEEZ_k3)9_D3<DM+5T8a}R+A$`u@wMj)l&4sf`*S*MBN{R+9k!(-p
zv^mk2*VQ9r1;1<X2?35L(BHDg6C-YkS1QRh26^nl4tcnupMEkWj@-UroKP|R5A-3#
zkvWx!=Z$ZuNg<MN<%3_2m3~eIR_-CZ@(3yS-OhOjwPIs1efiAF;7NBWuL_Y1XPg<M
z>)vylx&&XkLYBx3g^c1Y^HLoFz3lcxg_+ihqULagHio#K!kr&9TwT`%5zJ6(o7!S{
zebSzlqdS_D)xgQtSZ*-crUY=stC=$u&6emXmLng-K^3%<`FL?ZvgPFX;BF-+2#~)6
zphE{@WkBVEmt<-(h~XSifNEijL|2Fsa4n(+VB;S>H@B%>{BiU$IJS&3HYzBFt0uB9
z=Ay^gLZcheMmRt!ev@u1;2k|7!GPu8$#SV7seGk>fr{EvCvh6MA7tFg-WZ<i_-+J|
zy*HfB&9^pq&8<syn2#Q)F;~AZYjh!?tb%9+q-3e?E7O5<o$#n17ZTt)Rag3tz(lgE
z<`&Bu->T?kRM3!;^1Ww5%0cNHQhxtY9MG8=Uaq6~O!N4&`7>3Q-Vgu8@TCHlO@4=4
zL>o|o)(pAg@6Iv>AV({hBId6qy`&ak9jK7XE$6Zj8P(7R=;$||?*}P4Y(jnSt^zXT
zT#30<x<-$rZTzW`?Q=4jKR1}8MAHXnL<*%*xTdEfCTwWrdF1-%=wc5;9v)}P`N^YY
zfEvw=j~N)ngY_%PRU~n#gfIdCE$MpVk$WS$Mh+~)GSCz`O+0BM{d~NjS<s>H)TCVg
zv8<lCYA$B5)ZhVs_jIS+^+L05Q1B1uWbl&#fyV$U_1y>3ozP0&z|_tZywE<}@)RGc
zcTzB#AdoIr+Cd+Pha1Cmfw1<k&oqqZ1r0Lu^29m@LG}~@rA1P<*ACE{fRMTVNTUKJ
zuR4H%aGso-Be9;NyBn~VasUug5q$qtQ?iIMABj$|RID$(Z1^pnkPu?%bC>J1T+6wv
z8-Sy~ZZxZP;TkEOmjN@kb&c<o?X;R1F^gzypE`d0`sZM6r4$=vdYsEwA<DyoSHP~(
zXL?LJ44;fszA0&QC{JUeT0kJZKao`9tcwba7gU{jBq+w%#$Xru2jwa?K+pZf@9(_8
zA<NF*H5&j|8}Y~qI512PLz2`<nhu~&%q*=z@@<9RWnQq1!2Sc_Sni9;aQ;lggoelv
zo9srM9Nnv}_t&>2WLjrECYxsI58`_--{40x^D3}|sX!se$$O#MTjwPEEuVrp=o~#}
zK7BeY?-g0id9(}pgVrLm8>LGcx1Eb{q9E`fst~;?u>nxol>x_4qm6V@{eh-;0gp~E
z4kQZAl0QmX2<o1(YG{!e8JkWZW0VRpe6c+Z@6~I+7n<R|wNLx-QrGa-0HQOJaK!Gt
z<lcbAIGZP+KSK0f<T$2(7Oyp*jPr~pMj^uE)i%we)L+;|Els|(VPaRahn_!_R+WSf
z9@Y=w^#8<9e_dnn^c&!B3t6S(Hi1B>-S=`?rnS(;AaV*4UK?+&fA7Gn-lBc%;*Ob^
z48vX#meu~RMSUQ>*~Th=dmTT`AQ4q$!2ZmZmr0AceChbN-#ft#p8zrgc<&qKY<QdF
zjt2Zon38ntzs&=E?|vjq{EJp#z`5er1;w3ss&+FD^IMH|Dr>EuOvriUJ%8+J`?IFu
zZ|VB_axsp_?<OF@dnV7Bs}j|m<~`LkR0oHcANc4di0tp3pfZFBF?zm)8I%{)O15D-
zqQ={n7+?&KMITg!v8ft%*JZ-t`e$71C0cR%IadH6%_SYZa5ybHBbuvWNggDo@GiM8
zC$AKdk}Pu>kQ(I$f0seso~a50m2jqWK)?7jpJgLxN5Epmaw%_lWjIf@Dsbtp4pX*d
zT0u7ifl-wXN7??Q=uMjA+L}F%{$Q@^ud1ynefn{=t%=5aVABL%ov~NY<po<`0-C&c
zhFyDbQrM(4AuhY*8lfZXgK%cLj*pl?v8qMqws9${G3n1FeN4F}#S#d|ZztpJ$Hru1
z!`PX<SDUsztr;ux6+Zk`)}eFddq%y(T&rXUFG#Y72w*W1!|uDEc!CXkcTd_FpHG%h
z4KU6yGh$HM5@u3XjYD-&GT9d+PZwD$xc=Iu47_46_`ZBuZWd{oq_J7gD1Ypj`-)|{
zR!D$gKn^{d)xHb-S*#T%Is#=U>4YDvD7;%k_wk2)<k;5O+E%BNDxnpT{n(!9XBSj2
z)crV40N3+!s7mikUvkpzz_2I)pSIm6c%0H@xkE#*I%bZjmEc(&lB*~s&f(&_cW)bH
zLyEoHh}xnQe#1MuY)m0na=uUm_RQ-q#ALYG#>C=}LM<J}aFkhcgN&auTAoG!up6f<
z-ddx%Dk^dhWEh;j<46)K$my3oEA<+z#H*+@I52s%d$Zr`Y2{cvysghsbQ><dfzO3L
z(U&i%93}{v@0X!r3TSfT7O7nAZT>vEZ*pfErO*?>b6G{i^ri&aXd^cVxhjTbE!+BP
zRA+5Aw?uWgna<QB^RAMlu332>plF(k#K+kLgE8oRpNgI!ArPxZ>WI-BLEd8Hu;P5I
z(D89x&${`k>^qN766xD9S|7_=dn=M-!`q7A6)GO+S`|?X6dD5^_RTUekerV6RPrPv
z6RD^QB5Z6|Guwa;5}ZAe_g)W+GOj0mU@8;ziT_B>>8Z?WIkt%3PSHoftvvI5*hZH=
zfIAriObkEcG*FU;vCzcfs{TK*y3_#721M<!Q_j0KHBbtW-fRV>J*f_ZZyd-c4%ASY
z)<O1CM><<)i}BswTh}GaI-q<|ythJpi3$Hyu)ze%lPB6;^39paq?=##NRh)}lDA~3
z<n_*Hhed(l*~-Con^2|Xq8gg9i?iqgQF96*nHkJAeX-q-wD-_avX&C}x&J@{7+t4H
zthUtcI4&kaOnmEcKn{Na`9Sy|MedX+sR6oD?9|Q!Ff#aRr>W2bk6=8&g91MjHOvtT
zX)%}WPk!4qHd|C~G;cVb+t}rX`ZZa>)W`hRE_Nu16_!n$xpQ>XIk}0!u;9Fj;fPW1
zc6=*5buU>-i0ZCX(<1n2TX7Jb%(N+2kt0m&iYG=D5vxG>ioL#HitDOWI}OzmMn9C_
zSMYtiMF3CCYDH10KZJq11tWh_D{Kg}1}T8GibDicUkEAoiQpIt;_nUa1V64oW89(D
zo6L^Pny<pOV!?Y_q=&dv<1(qQhF)gXXf`4-J0asaQl$?n5b%*sAB1tRL~0L?i|a^0
z1uEhJp%m|cwkF<zdR%63gw@LLJvg3|&Ls$XQzALW2@4OEYc1;hAGRDy(A9mQrQGkw
zkU1-sdX69Y_ZSyi$hU{<+uJhh>9%2(Vj{LXPuSf)SMr5;@pVN!3xyf!^xfYnS?F7z
z(VEiz`Vp~}Ll<_>$Hn>&a{N$@oZ;iHus?1--0_^cic?#&P31^WyP@Sn9dGK{_ZlH;
zpBp@sdT7`q>ASiP;6J_*%szp2lC3sv<-4S@dA$jqI<G}r%^r(=ZJ4x^@%Rt)oLo+B
zL?$>X+EavtAh@Q|S%RUFrdhui*%GEytv-5vbyNGR`wP>eabnZrVdi>h!+)T6q{-l#
z^+QdPqmAH?A4}Qpf(I_Ee?__<o-PjLU41SGn&qGi4ylcnd*ZBnW9Pi}<v<P4m7~QQ
ziuz6bfy+<Cp3B3&_q|ha(Td%QVhG)ckQnR<$ENu=O^4@Oqqc#^PyBGgvKc%89z|6%
zj5Xt`7~A(-nQO7*iI=-K6UPm5&q}4rDUG4I@aDPpg#=+ovM(X%pdD|635jTbPxi;U
z7Z@fst+L1%Slc@<YDv$gUpH9Gi!uVa(5fF#b~;Y4KgTo3L;|XVGqO{_+Oy}2kuJ@n
zKKhJRD@Fg?FM9pnYrPvg!nh&sMknxnrTpGiJ}F0#j;|iO;?@1Z<a_!JMSqIDmYRP_
z4F8VC?i*P*;mr0LJNK{i)vC<0T2+@?X)%Fxmlcq)#wX@3v?iL#CS8vo>*}w5eIbJW
z<ecP$3f%^8Ti7DwHk@#+Z*L6MukbH_Gv0_7WwUOS%Zky!uHS=qq!d!@fn)KB_9b0^
zAYrnfN3A+@NIc@Yv_(f7K0(BU>Q@oP&Z~hujuIzK=Y?=*2sj&ijZt-ljf(zi<|9Zh
ze*Jf&mN8@c)@&?z4nn_^jGq0Hur`JQ0oW3+mOZUt;Xj6T3Jlh~kb`JkbTlYXCpjq2
z9Gm-)GPm(qcLk4C1LT;X6cS*gtEexL!H>jq!C~VqKq!d5>zAx8(!p>M76r)|iGIcZ
zl8GYgDYKHK&XVIoKz%Ccwy_PnU?4Xr!6*uWz1d=k)&iCYVs&`=<=MLh2YNrAk@ng%
znRr|c#`ugeSFY9eNxVi=4`gs3@6`5%!HX{rtV$l+`cS`Ep#E(j93*PQ)sW0)`;qIN
z96_bcGi_)2<v%XD0I$`2^!vNAA6DsC#o^M38l%@OZIDi;8Mk`T4H4yc`2_80#*7w{
z`|9PP_3_UthkMtoJz$yiHkxMB>j|G@gWgr{ABzxX$LXu*Yw3LmLYy2RhNiftF)sz)
z-8--w*@N^T8nBP!n@#FlZf*)-xTWsn=9JxW)MA=5Uyu(3V}PWQViNulyl3dTi(*!j
zEOV=u^scq0nREPGJM!?W#4zt!Yoy(%BHyFDo_O+p+o+Jy*a(H*cH{Y9elE!1SZ$fa
zA$72XFN<S~0X7DP#LjN!Wl%N#?$RoG#Tug`>C&~Cdt07R{Y)RT*=g<!$yK9Hs5q*r
z))h2Zoz?Dj>+`N&U^LI^^p}RqZ1?;N`9gtMM<?N5cJm)+I+iM2ju5bBRGt3*O=>ct
z`0uhNZ(MJSFz4rl>6cmt6~#RUu)#-sblM6~m~e?Ie=}o<s7v7R@72GgIy7jjRN9DE
zCWHnxCW8zn!MypzFUc)A%++aSS|uS(K9><YSAAN|e5o769#?w?ymxb=(h1UJzP@J$
zKOW=B-uX_Y=Hs2gN7(Bf7w9U;y-HYv@*U(GKg1~`uz8}RMrEe3iYGy&>SBf?yEW!u
z&D6a<s}}8iCd0aqk-m1->|A;r`k9Q!KqaDBSyZs|BQ?VRkOHXJOtG{clYMjS4USEw
z4;T49nJv`1DeIN5#L~*_J87M(CTj@T&pBDcc(HS##BO@{%kB;XOn06DkUE~q;4D^$
z&$M^hyTJC%S+p{y+Eo2)wag6AN-IZA(BqFX7VH!xxbrcH8}KeaU30=mkcA!?McXyv
z@sFZRB33qvhf<DP4fxZhslwm#BO(XmS^ku(OVvgu_8ADFZzoehZcZ5-{fIZD&KlG4
z(&tWGM%r{k2|cxHpC&&sfw26$&GO6H|2!9PfY|uNw({gxsY!(fa*PE$YJ0?j9yivs
ze|f~o%fz;<JK3aKPvm9)v^7>ow?mEZLsF1Wr9{x#xbI9shzK-Ccs^!yG0xEmODU}D
zvNZC%*h_9N!P?`o_{AS%%Rkn&Zn_!-i2~0&$>H6jPI_C-8&df+y)qZjm0ZVykdf?D
zXw+L%3S6tQXgB^}35pFYT#Do9XxSO2ans+$#LLEjYK0~|SE`_Gw*-e|Z|W*v4=dhJ
zKi3TfCsfn*RchS{qOyzAAAwMTF|o-=fFh8m0o1@G_i`O1f$9!P>vduLI4N4g0aQfQ
z%ZEW{&YaG)eL2Ia++?>v-9wA`j>qjzB3)f}bBz*ueO(fUw*r4U%CJ?kFo&O5e08z9
zO@G6_gO3M+6x<ogDm^r(iEVcr_HX62R4f*w=;ZUlURk}DZ;ugu{Vvhkf0m9rp5Y(S
zMA_t%6Ia)b1-_U{xd`s0zPUJ;b+PF6P|MY3#3df^wHF;}P*rmh(G5suk7c<+>r5Rt
zF{ZS9jw(*RBgJcj_4)>e-#>9KRDGQG<_04bA4@xLcyC<YSM1m@cuX=}X~G2<JwPq%
zw<Mn;4h!Yp*Nt4s&WWB0x?cJb1^mTCoU%Un;=0ja*CuuGS@fPF3yUcK7(Uv&*gTxt
zElXXvuli0$GK>d7_2FapG1&#lF|OD2J}FzV2*Td~8#N$M`jzNoq|1}ljqP5Ps)*@0
zUXpi~Rhs2AaZtlzKZ?T7cb1KlD-)C`MQ<u1{M;~J4QbW1{sJYIBdjTLXT<sD<F5nt
z;HKHOPxQCmPS-D;n)RNt(XY85>>=zq*G9a5-=!_oetc3r(ga8vAii@Ib+M)yOaciA
zDP~0~@YXkh@hp|;bXBs2(m#c+x=!6p(bm`-B_Tf6DcoT{e$xLS_ceTUAkQ#cG}|~6
zA1{bxe61U3KJjYDQUDp>DHEs>B&o~f?Z?ndxxg3fyF{;McU5&)b8q!%!ltQNZ!WQP
zni*)EgDx`Yw>%S7mk-r#{4x(3KL|UXvUka#|I8YTOO&)#v9bXN*k0zgy=fWdV!#P`
z)d!!Tw`!x}Nc4liheCXJvb_(q5^3J-)~J4$^C`t91<Ru1%&XpY^+CG{tLA1KV|R?@
zX<oVP#^T^*?x}@7>4J><8U!g0a%XUmIeCh#w3T9g(jt_U0-WVvneQfpzb2yKL9EBM
zx;JqeuEW24qOOaDpLzFtk2!s&u9C(T!NI##<I>}go}DOSTkcJTl==RCVn`63>gz35
z;4h8tlgFcAM@@-k$^0Y~)g?MWAcL68uZtn(nSeVY$8XZWTD89bofoK~!jXIqk`f!)
zr!7=ZF8G1F(_F;(nc=+gjnW(&_od{(O(lDI^Oej14-?^~N#uojoAthJDse%M+RuJr
zW2=8^N`3GF9e2lN0U{5{W<h&vRe3$9&)Nc$?+c&feFB8+*exxYKJ6(6wr?|zOw!cb
zTb4E6A_wQuDj{2EitelwN`KYf8;^*0`Rr^w<FCv48poaatodc?Q&`*y<xRvZW5McB
zcD_ytt=ME3*Kp46V+|qB3Qiz5xS7i@s-J7O(}T&ql68B_Vv7-8-uS@b5z|ks_t|AY
zQZipf7K5hr2N2w*xE<}A2}Fs~yO?P_ZwT*qIkb9+4Gm`3(+C6dcl`ST)ydKrO*M!S
zLTSi$DXZQ}9Xen+e0s^qt<Pt>>ox*yQm50z{y>_gjRjo>`026(wyKs1Cz=dwF{h?y
z%i|)A4fe$3<f;Um9nTcJt0{SyCAX#rTkB)P#W_@<03J_k;e5*$z1kqI*6}1faCTYa
zU)K?9i6(PSd!Q%PSOvXUl^r1gB@QINPY_BI-}t|{ClL^x0g|5iSeQw$2=6j<{<%~q
z#fwUsXY@AdP8&?tYqMy<_|O(Hj=Y|>h3TkNfOLVQ$(!OsUqW9F-khHLY#;=n=m0@x
z7(l2(!TYm!VHumBU_p-emF+bbwo?%K9pzD}N^SAkQB`ecDsIaLT-}V;8br8oG@WG=
z!*mQszbNM$$Xo|@h}LnS2j!#v|3JJYwkc^hiAOAv3N<h-%u+I;j)+q$yYA2KN+~-@
zk`>>nK{%*e@-qJe@0bd<Vg0@XXp;*dV~CDvU<=}^nTrMkEGhi~ZjA5c5dxH_MOFek
zmt>36M}SZmv?>Z{K<iGXm|ZA3nXvG1*19=rWn5B|F!b0`oVV>%_N!&hb?`G@I5=A*
z#QIeEe>NC;?ep)LA5UK~DUol3lhF1j(>pQae)-&{2?{#miwEeBYE{ID1$iZ07X94%
z4eyxwNOBCLiPYMuH%BArF__KIrZ^gt3R_;v!h;0)5=I~O^dDvOUS5?^a=9lCT6y8S
ziC#!llzWM&{uLcs9H4GcJ$rWT`bE2DoYW}*a4))mBgLDY0_N9<4W;!kxH4#j8!Z)R
zt;pd(XCWye^VKWjKye*iX1?~*_e=HTJ0Y(hopJ9h`}3KRyX-|<r-s?aiHNaSp$2rm
zT~Z{7^p;^0dc~)2jXTA^!LW2xs4hQ@@-l~Ehtq+K!}`b73=xe?dFGCLlOZ2AJ71bx
z>Va?d(6d40C^cV@IiBUXumej+IcSZPw_kc33yFzVk~)IpT@>>y#*Y_8TP=@+9KpNM
z*wBsU_Y$aUv2X2)U#vrTpU*4>Yct)J9aD+b-I(X(MI8O&t7dRlh*thd`69QT%U3O|
z6mc3&{)}ucy3~7{QGv#x`Ba4<r!JDV#Nqo!2dS3Ll~JA*6zq4@Oj|Fa=tFr-C;2M#
zB9Qr)+*~9B2a5R%DD*Vn4qN<&mgf!4`Jg{Fi72DGNmqCGPH?8QP6E1el!kq~NB(}K
zr~K}2F5mIy8`J4_S_00ZPv`?O)@%ez^olGJ+E|+6G&!{Jt8CjccSYA?X2WkfQ#BP{
z5}i{i*8vTYOCVRRV0f6(D{Nod5bOEY$t1Nw2R~T8fR7sT`HK9*{rWWglxAi(Cpiq^
zM!@VIcsGrIB1XGV`l4a0C3zl_=>3V>fm^FOUaKY-O}~-;sF~{i*VEZ}q?}?tNgk0i
zL~XVxQ|TGWx#qu<8mlv}Xhw^4^TlIL$)$p|K>$RE*3RvK=KF?bg4{7I&T6VNf-rwY
z!J{!9W<QZ}ygJ9ZIwuqTmBl)TE&4lFjXHNeRiFPrvF##A`<ZG(;rfR=@c4fq-?VNP
z9+@PK&NN={5H;DjYi*FHbZri91@dvq;<m+ci-&w+dxYtf;Jc4EO<xT2*4~cFsettV
zy~{t-BFeNXL6@wS%oME;V0*s$?L2cYSJ_<!Xlk851amTJd^nE)5;$sdHO{lA`<*Us
z4NHqZ*!0K(s?}=Tzp;I5Jriu0GBqC)aKcihCeu)U*1SqQ&4AlFZL2A1jBboK{fjJd
zTqITsy4GK8sHmTKHjX*d>E@|%7cCBy+=ySBjv)xqZ9+uLfZ-yRK+Tsv3j%@lDoMc}
z_4|$EI8KMGsLfWqV7*ia|6?~-M)P$N3CA(Lg9V{q*oWLO(f-PmIh&%QfaLN*fZ2Wi
zCL5tv5}Y8{4a{>{hO;u94SYO$>~N_~jA6*;0q{gS^f@>%7fK;7m3o7Qle`&1W4Uw<
zah}#JdHhi`5BF`={m{m+`yc3&Azx4{gsxmB0`mY&yksY_WpW>&WKHZC*c1ZyZGH<i
zCkOJxc*iAw63TR4@4p=r-9;hiobE%AR10J&mXyTRn0;P^sMlwKtnVg6vOsLdqs|pf
zn6&K+BKE(_3@!F0mdoW_K90F&{e0ERcEb#BXf}UF?8b+fE6<Z@xJ*R`a|ItllTgjr
zvIZ|CevrFh$L@q@<Vd7J=_T=IKR_1oT-gSo=jpFoJS!FzvpGR{8V<@@Uw!5M>$U%y
zNC(gtgs{bT^S^z~{xu$Dyrff~eC$}PyrL>>_W9XR;nX4L8odQ(Ewh26S+5Mj1kQD2
zqk*%)KSCqD*bLHb7qtu{R_4!i)@)Z-Uv{k%dn=C~=l_G<Ho=60MJhJ&g8Sf;kC&XK
z-)H&k^(kFD1$bv?ljD=DCeagF1fem$(OS|%JZ-<B#L2a8GOV*Hk9uef;LI0V1YQGZ
z{AdsC*qC1zl?G}sQ7Yq}%}y=7eu*D^v5Bvu?>G01^+W0pf%&T@wlk-|_1T|D3$({|
z^@<&S_0CJK=D9;ZKO>kUjo&J7sSKe0*4M&{27+#&WV>`rRS*3k$XR8>zKPrJ`@I26
zrtt(pNBSi1!fCQ)6@jvOV`Pof91l%1@yGoiPSY|OZNF@03Z2G6J@-Ss-*$BH9kzJZ
z4YidncRgzIDelihV`jy8k3ZQWzGY|jCDCuzAVmw~KjvNuakMQ=MuGP-6A+usd}hyr
zF76Eb)d6eP0HLS%t{9u3zXKjK6jX$p6AglruyLaM{p>1#$fqEPe-^^ISW<GZT$7#-
z?V0zcwKcsF0v%n};Gg4gi=bLhN#*tt`Fcx{0^h$f5C~vcw8c%psKrcfRh6l??`_d~
zE_fB)*Oh7l1GS*N5E7|?8yL9^klTyv4;!n<SIk+sE=wIy1lEp}Nx$$`F<1cVy+fC3
z?dul?I5AcT!Kj}h{@!8`C1w{^O!__Ze@T!eEgqg=I0+K7Tju15j=T1U&I7$om!G%q
z$~Pi$mvuxB3?>>fiolS9D)4QevWg}0ysX1^A76Zn+oDVZ#XL;(VK>2;8J&AIL7^Jl
z6wE_cAwkl)Zqy&{GPK!Yim@jBb|v62)wn#r_TX#-+sw_WB;GDR`T-^y8ZdAsGjD&i
zfi0Cr@2YZrH}mWR24O*L8Z$D_YEEx)j0EM(R7ylYajdIaF6Z?0F8`jKkRB_+B3F^w
zRe#<}ObY#ZfcZx?S}Z6!X0DSDI0N?O^`PJBHZl<Uih8Q(J}d|AV0hP&cK<c21Xg-D
z$HQi}ZMMN1(4$QJy5GrRqsP!)S9kpUMs6FtASiL9Xy~n;aJbC^ep!iWaC$!k>E)#S
zEVnC3OTn)#zwYiG6$`-IH8_Kl47|@21=L6Mzg3`nK-I5<w;xaew#u~?F$dOwn4#%B
znV^@JfTP48N`mXpyGsS~C?1p5V|xqYZic~|uO+0DThCqGHy_ISX)>iyZ^zkBslUEB
zcy5mQV^dV|?@a2SamAV7dU%$jUGZ_M_7>@yp=?o9sK)fR9&&1zvxTlgC$6&8%P~zM
zWG8-*g|X`S^7yinntzGYXcEemy9xVCy@*=2Fe<MA*6W%3blRqIs-S*~SX1r<JRO8}
zw^KC!{z<{2>Sw3tb(Uk(6{sfN$sp-LyD1Z4DoS%~W@<#A<kH@S|L+>EdvHEJ>-RgZ
z-KS;diT+fEl;6#5-!QC76>yfNv9p#y>V5g41$3H2IsT8qo#t;cJ=#yc`Dr>@UfJn5
zhK1K=InxjOz9>+DE0mbN6nVUu>05rOMg{V@_;mlh_@3TLN^m9MUy_XdIP_m5SYvCi
z1@geJl<}>3ZT4NNJRSW7jDAkx^3dTCV$X{7Nm~6ET_jUTXsI`o``M4q6IZG(x5<0_
zO)Qm{)firB*5EwJuPS#s4%7H!;q45+(8m10+^0|Y9=h(y`gJZkaLE`J?sV%rr9Gb0
z*BQx?iuZTotDxw<RL$5?4x7icOW?^`F}8uTIVW%z4XffL77`Rv@o|kjf*rg-3>8~(
z)yceFj7Wxor*F~yQ4hIKuDr_}k#uI9nC7#S6YVpWGW$=aV(rpa1zbOe4e)h;1oi|;
zA!J2swK}=e+!R({bdG*fUJE2g;JP3qSx(qWaf(l=7qR9tbCN+U0q-~e7%yoPsZJ@w
za-q=HS06}e6;d6+I%n<Fj?N-ofbfr+oo?%XwaI*`e`@c}NZf=*xJWgKo@tk74l9_h
zVSh39Qy}e<;Oi8x5F^S)vXXaap-W{ouk`i`Eiv)J2361brPXlr?>34Za1gD)9i6ru
zSk0l}Vb97M!sxnST}ARlEc4X|8v%n0{k|sg%RgGWhO52s{{M`nij^F}ezY%F&L!U%
zD!GZ3PBed6yp-4O_AMlsl<qek^Pe(s@KwH2t&9h3b1E7x7H7Ez9<huyV7GkA6W0^O
z)1d!kp=+J?Q-^C;9&fpQeYi=mBP~l4-zV8=W@J#=B~?1x7@k~31fb(|ex#rKEk&T1
zTUZNufN+RbgPc|xwgI7tz8BFtgIi+x{o5_}nLqKnBMG&Ag-z7*q+W!dWm_lhnbf@D
zX?w*Pu%lp7rK+8k3PrjmFezpS{#4z{?BrUCCy=&%G~!^}7p}ZZfod$EyLZa{L{Ag@
zTQqnSPXk*GKk(4PoAd8vr{tlWv(|Uw>w~xVfBfYf*+idCZj3R~HF9FSs9y-q2M^st
z8xpB&MVJ4*^O>g~;g9X<ZO~b=F&LTSY|^~8Xktr0E&A1=snqsJq-><A`y^Ts9kK7<
zUGX~hFI@6ZxVsK?PhLnFi6MN>9KjgVrN&9o`rFr#tEnZU<qhG(k8y}gbxoqN<ISTb
z*DJBN>xJ>fCPR%+U;R2jy!I}kC|GWFLxn);@uaB2h5CNRBS6S#x5#2!avc;ra`^3X
zQc(M5le8n3GsRfQ1bV=rQMfQ6C)YL&1PR+St~Daa!Vb>$F7AIZbpd5T9;y5-ESTBM
zG}BX~f8!x0;EV>&RMTVwup@~OCCrvkfImUg%N2h(u`mv8#C?XF$LM0F74Z<9=_K)Z
zgrn+Uaww(gv)9{&@{hM6N2n&3?vp;_#O1M!j`06LNSy`e2gKH0;MIO#bJi>J#VCQT
zE7`TFokOOU<*y_ysR|Q^>k%to{-r1^Y`b<!u?!09EjY1X)2aWgK>wTIE_a@*b2PV2
z#D@32f>Uve$13Yg6-D=b5;Ph@$(uc&{LF(lEYk`gnN*%RtM3bWRM5uxZYRt}J^wQu
zE!?L`cb0ARak2no^@nLH_=*}x4tEm5C0Tp2efmHWSH)ZJ0PD^GUX4jo5|A3W3Eh4x
zZA?dX^5brHVm|ea#isz@v*-H$B*brWr0q(ryvKAJqqxjt*X!ERMv)l_@@92q2U-Sr
zTE<KbECeTTNjV2L@4<mUoy#`vc%tCtSY8M-5QOmJDWThifxzPzm#}swutqnE6uz01
zDLlvVkyvZ<Zc5YdiV8Fv_#-P<0u6tZO9lj1J0bD@XOrn9t9X_HRO@U^o1n|x8XU_M
z9=2UE$yQ00lqXc_C$i$G$ts<t_(bpG)Dp{{9NAeUi$*AB#xt&#-$;ZO6Q|{CQ0Zq=
zUMyM06r7+@gbnFaM=UgEQX3P_vw2miZqonum#P3_L2pBBHy^7Ywdo$0*!>h98Z$JB
zLX?ke=cR!%ADi?v{PQRjk18orNlo;kGu`AO9AVq~%TT!42D8T)9dzaYj?BN@2%Zdy
zcn4eW1|(#Pi9W?#<HGcVrM9^r0gN*@E2|)su&v)QQnpYbGFaz4a$kJ}hwEvwG=pz%
zQ5h9!S5@fI{oW7HZz{wNX%Oy&G-fDO_(hyKDGrGDAlg6_P9p2i%Mk=tfdHxK;64!h
z1N1piSoCACSB?e|bX|?`_$t$KcJKnKJx-=j1PK;}7C4v0(K}W5@Px?m&gSfKx2T5P
zSG`yP<m7<<T8+e^G?&&Uh4$b6e<>sHRa<#_^`6E2D+H;1rp!An^(h%)(Tr@A>2TG{
z8Fu<=Z3%4Z2WD%$xettIk-5%Uv!)wemb_b;f+Dk>wX3)Oo3zb!FO;F|bZ&us6|S3P
zB+D@^i`HRrmi=@AHtCdOj%zDULL9lg*<{J#Y|*=7V??HV&2ebl+6EW;`T3uEoY`S4
zCPetlx^J&u*Ld(I+s*E&mE#6c3yRTwJ4`Fb)4wp9ILK_2A}Y5mE8PgJfe4gEU$sty
zE84CltKz*kEo0i+Vg~*qpuxURBc&_<168-Qq+L);->bB<`sseivxkA%ZsX9|%;E}v
zkFz@|c*dk{{sx=}wf5h}`Re{G>yfW@_zx5o=QF-H5%|mQBhJ95UT4hxovp2(I#B4U
zQ0qnL-i-h_j9&nEM*|B$t@ZeqN&K2v`V|H%>JMNEyjoOby&lpAyFWXV2;Ke<WM*A;
z)l4mN5J2Gi2MM}c0zhu%BzS?&w`^C@Q~&M*uPS0{Z~0`yT7o-QTE;F5y9=;$c~3F+
zTT{mWQrC>V&iZ5M`kEujN{v<6_|?w{t%V&+EPDM-cmEjs$g??X_o%tBew%c_B9Qg?
z^kv;HOm2o><mTnfiNfZtemdqINn=1VIK5we38y3_u7GPMvav=99r^BW+JB&C`@0D@
z!M=57Pa{ErXBxQb2_gSd$_j49W&Rp<bbN7%%UgLSJhd7h^Pr~HFeNL4W={Fii?9Fn
zHcoEjwM9~s;(Zligf_*G-L_F}_0Y=^v}nE?Tdh=?S07>kJSwb16#J(S1YPqe-AojQ
zlNxhk^x1lL(&^_QPX=Z`pPX9*EgA9}_O_V7_0K{f{=(|9&ef-;<fu8xe8^Qk96mzV
zbvYtTuoT-^R04L}lqK|Z_ud+3|1MXbpy7XK08W?dd@yzwpbNGEQ3(&2I64Y#C<uk7
zN$pNHZ&-zvD9ioO7*`2n0(@6?`O`@()8R?r;wict>L<ynBM0hpnCbeM!SdyG9R#cS
zb9&=n?zSvs2WIDE8H8(R7J4yTNGp@5uqt-P<~{(fYw?7_jDy5&)$|}pMB5k2f=(!)
z$#QfMs!JU^lQ0k2|HV|nIqqx7eFLrIOc~L2b;EqMwIS=ErF^#Y&Hy{`9jOI1TYF@{
zv2KR$OO#QDfG(RF?tza4j?~L*M;BL#7b;BSF!$7@cu9pAd=lqmGpZ`y-SS$eyM7CG
zqF&4ds|Ev;86r(;?ap0TH&!CiI=VO!6&*5~*|R7k^QwjZ)dkB=P14|&b+j!<d!bx$
z3U+K$hGessOadx5H-&hDdh2soutRncDt4pDS*Co9?%fqN4nB3L^}oXEQUj4;QaNpL
z9EagKlJw_94;%dIPiCpXbXaNurNu?!|MaoX8g6Xp>}sC<6nSmoaCMoFBk7tJ+}0su
z)aOg;@aR$4Gk01pmenhgd2+KKN<kVRA#{><3_Zp0syVJCXI-gwl)fn>X?hxyIGnG~
zntV9y8ycPX*C<yeB6=kxziaOZ;mC5P>cF>-FeNRc>RpLWa`MzmWLcN+g(MlXTD1_<
zNNxuf&hR#?NprMKi_uTf&<q#fp!CFZs#cY7R_5cf&vb?;J^Q(Xk&HV#`|4S(FaI^Y
zY^eh_kDvDdaH1p7Exr_I3Y*ak3EWP0{*CMKm;2;u3IR-u@b(Pd|EN>UVzX6iCj6qW
zK&dw*(o(%EhCOxKxJNUpqh8aVBGy>DX4Ks6>r0TtcB|@<0+pVhrs!%#jo@h)yuTS?
zm9eqgucj1!6u%7M2Lf!5S*@U(k{~0Ih*&(>`jMXS+F86*k(AV)0i?l?Q2zvh_jFrB
zm)QX%iymRFhm&`{HzK>b8^>WRNa1AWOQ!!78!C5T8kHqJ0EyFR4h>euvfaC#4n5HW
znrOxrEq`?UJS$bvc1Wfc`qEhW(8Y3~@ra&^cDrJOC_+Bk)2->ZgM(*-n55Wp!6n%5
z1{4uEL-e4@x0{d-kc^z0h(q_lFdWhZddeI@g>eJi<-+M6BSPB^x)6Z)O-eftXlfdO
zK3%`m8fK%2&0_=vMs;DO0VsBIYj^JwK}64GJ~rw1iaxRBHsOS9roVbJ(c>ifbv#Q=
z2CXVqIsVY(41!QkvOtZ%g;p*=)oWKSKdO5={gi-YwRAi63DRX#^15mpj>)|pm<-{v
zbrMN_9piH};9H)!9*G<)afnYK9uO2=zN`{r_cl!kvvQiK))aa{^g`Zq^mSmu^Kp}R
zK7WgM<@P^DNz%Pdk=zB{>Aibq7@M11RMvqLA-R!F>WDE=%qMdKiI1z5;N$emH>M9A
zXhDMDur{{8fQGVzwFW{d-#Do=b$}q)v^He3zR6$P#AM=r;EF5Zu>xiTUoJFfgMgQO
z@B{Qn*u-}eL-{rv|Km!ls?q1;Rg8j46Mp43f&jWSUZyf7)tasbNEOCQMKN^w_hqwF
z>g+&qA}Ao)yL)&QCVup<R+!024v%$H>!RM(f~APqM(mk>PY{le8YZXg?DZMmW1*r`
zDkj9t+53&D@6z(AA4n#mY+1uRa)}yf<QbZA;69z+zhWR#x=}40E(WU;y5uT}a$|r!
zpfs!7ZN<?Oy)d+;b;d6wTOnBh9@`IK`Z8JIcvAd<+bv5+li@pO8>@y5HM)C?Wv0I5
zHo|Bs-A6)Yll6R5H{Fmq_TYQK6K$(Cvl06ScY+`1lC?2KElE)|g|3x^aB=OY+Sviu
zJu49pXllr<mvVyJ-RsV^)_+D|jooer?9Cg`6NqpBPPxMOuav#!F+C4(SWhp{`j^0;
za%TO39W;S8uzPXj)N`QRKlSJz3@7Mb!oS^V`T7BP_g49?5zqevb(g~f${kuCNqk><
z0fu_K891YUagSFUA`~y~n;^&wX6*7k4>6HJA3iR(=ni0KJ>Q7f18Q6`SZrxMU+xwB
zG~E>zteqhF2Tb8pf}Pm;*bFS6#vRiv1E-{oPp_Mg&iB0N{{wjj3G}U#kq=6cNJade
z<o(y=Nd`Jd1lj(Tl{y^D4loM03SIB(?rxjje5udm^Uto4{no1#itGF}L#>PmPRn^J
z5wXcC)|eRB!EZd+E>^I#)>o=H;7JdJW&`*g*yuv=H(3wMsWvQ19l?xH_Fx`thOTk$
z@TKo(50wrZMhX3<n=$>t^wPVLmw&GqT#_1x4~Y=CRLIO4@MbBXsw<6prQ_>S@%)p<
zjy?R8Y){AXdMAE}+$iW#H?Y$w^}Q(}noN%%rzy<Z#sdWyfwLsHHa<6zzJfed&nDjp
z_Xd(A3lD${T!?Px%+^VG$`3Rz0ZwRX(APlH<X+w}rpe8v(|jdfP7o&EQ3subzyoL7
zC6#G_2)a))-)VDQs1Vu{`|LImPbRx^R({6tZ8DLj6fxxTi~FJGvx&h5yjaBreulao
z+y<Oz{m-cAk!02WAh))PRuUdD;t%Aik}z<dAFym;^=o?F&8h8X<gTl6{9R2m+qB}{
z(<jCHh^fJ!JIIig3akAYH|z!Z=;ApG$;sB)wou1mG=<uW>_BNp#DZHJ{qBy~W78tJ
z&rf*?UOEjYazBZ0wf3$APv~B_$Ud5jZQ?4+#svA?jW<#%Df4gP$dYN)37VN1NA(Bt
zIY#$SyYu%HjLE5zncKFg<A2X(Y|`f~6Fbfw*bTNt&T-yNakMwae8y>aP=iQ`<@~I-
zec8u>kzB@;EQ{*?jvDh7q@nee8eF^cs`l65j``R_R|G7j!yTgCUX~KJNpjN38VzbB
z+GJdNVozI^>Llsrv;(8>$4LtQH=Ffx(8yFM*SULri1eCQMaqrpr|^|yJPl<L!L46D
z+~*K}YcdU$RD>DE{k~>4rRyus0=jJs<R-^Aj4l#aT^~?_25m|=guTYjMg8UrFXKr!
z5c;Zjf2e+>jEx@SY&NZDV*(!;bfa1#bdx$I%iBsmiaNc1PBR_}^s0Ec%fkhF>jz}L
z3fGap9?uDIy5xH9T<V*hd&lH7n=gKDd3#kD^|509RhrdT1rB_wo3%h`cSi}XC@~Uc
zY^F69664AF)a<A^?aAJ(MDCOhS}x<?6DcTHhx#~^u8OzYsst21qRkS+v<tL%lOH8N
zzV#msTzoUx=E;qi!KH??^yEdWaM2aUA5t}TK}p!dY;PkprE_BEM;A6^HeG>9t?f^w
z45H0PfTXz9ESda=zhbSkg?3BE71jO|95pnIOh3;l*^|0|bW4^6N26v;BV6xP720)i
zRZdqF3q>1XZC%6G_oXMnieu*9*ZF|uD@k&1Hi#3Q8C>UpReX^0%53la=k08y)=Z0Q
zvhU3NP@ga*ztInLG<F{k)@Rqp5cbcv#+-bZvafBM2byi_`($#Ju_jWwRHV3S96`3B
zWs;DV4qmx(^0qBE6$PHQn|$msDJLsDNG6;y)`Cv<axdC)?RBJ7KDHqKSP6UJOASy|
zF;|f9l)E%wfECw;eNP%QaG*;cYgyOq+MA;S!T`HXjCl0-VGuf_>#e^AAKuuGbMUy7
zB_;Y2HreCKBfp{!;G`KdxS*kGAN7>c<y;>*XbMFyA(eG-((+8a=w4U;qCl=BB<`Y5
zU{<p(8)D!PjV1gQl9@w^!Qr~z1?hz#cCYNvp2Lcp!F$F>(_i1Rf%pAPCPeuy5%S~s
zze)LeezHF-*h#GsHd>kqOb}zMVn5|M&<jz+!x)OIe2|&@#8GN_>L4C@HkWBmf(PDH
zm0B~T1LygAN^Sxd$rJd^#s{oiM<F$HYnEiY;HZQFCxW4kV1hA2&07PyTV3PJJ>6GF
zeZoL!C{q++<NwRod!Pmc18RD~(O?1%1#?`wuHB}moM*si$u!C2m5*)GT@i}anvW~g
zT?(0>WMIPE)#1ax1Y{TvR_+dz677s{xz7j@-ybw+tBLk349=1#)h=ki;&;N~`RDbz
zzvNJoQioEX9=DKG-u9lg>b_g*NaMLPl!}s7Lp{`N{QBHIj1pNYl>_i(-ZrM}vDE8G
zz0UZ2GUi#kB%?Ft@4;V3vTOjO&Q*y%PNQCsP$ZRJ(7HUQ?%xAO#d|(mg~8|4AvEbh
z;Ks?gf>P0Mw<u)O+fToe{?&NjA5_mirPZCNt)0JMVD2>RD(3uobgy1x@qS{T;?cwG
zpD^Fp4>#!F*BAR_Uwmj$J)5DE=@k1d6MMdI<{DZa#nLKxUsJP$m5)_9R`{M0x$SOS
z?sCosX9|u`z!F71=&6eh$3-$A+-s=rne2ftoyIr03Ek6Cg*9r=fxa)=4?PwZvNinK
zIP(s5jDAaBrbq@Usx8#rC9FNEaDF^z;h1!J%RrPhxS2vqZy{^&H1g7JufhVjqrSr#
zGPImbzyA``2Hf{Tst3HVhrZF%sEkUm59T8=XR(@LEM?L9FzrcO1=W0<<+7sC+o|{S
ze+N5r*a>#}02Ha=zdLHi6?2|nU%~=vfpg$)Ck`ZUSAfS7lecX)1prnf<3DRcS(h=H
z0=U|{aqC(SNt!ti^taS8sLcnnbN1$Navi)uQ(%;x#}j$4TTX-6Zj=1mLsuVs5^VY$
zD!sm|4tIw?&DHe<wG^9;E__pAwGV3+9z!3!;NnVQ{@)FgV-ncWr=u}DZ5eHRq$IMW
z-c;claxw-9p?z3XlsZ%(QP0W?rW?BaM@sss`_2*utb9*3W^;Gw^41%o)(otJk6x&S
zgAr1&56D`SHU&F`$spFd4-5B?jyCKs@bf(+P9$~cV!F6B0K2=R)Dxyc;rMGHgk%i~
z!~BqNb_K|-F=Hk?X6s$^UhwlGy6S}kim5sh_DJ0lp}LCK9JG?M(`3A3?+Hg$NPeaE
z$tiM*aXzY}tufBjuGE3n^xCXs@FWWaic*GAzg_H?5fj)IZT%B@WK~1A$dhJnIY|$@
z#jX`SFlN|L63%VXPN3pRUm21-(x^5I<{B5#Iw#d#SMJAPwPFX)mYYCz%lKQ4N69ow
zxG0rcQ;5`Jg%^|P8Ai6TaHS-xY#zTRo(Fbp%UKMy`^Lmb!b-_w9R1ep_Lzr>8}*ZC
z>0HX+<nkK;A%JmdJSnc+0e|Lvu-7gh=`Y_*sj@hxaI17!pUw31rQp?c&H`Xm?cYVs
z29h>-Lq0+MRo;qj{%^33h2J~NN+dOB>jJwvC-(cGSRLPKT|2#P(se*B8N`Gg|Ng5+
zwqYR!#oy^#NvV80>Pr@npYr&2lGt9AF|II|^VG}oSMH`P`{E=#{Lc&8LHzv3EFtMb
z0n_!IgKK2y2Y&ATYxN;Lp|aQx6_Qxq>Z;i3jdZ3Psu*lOAyJ)$Q3XD@-v@DDUvlfu
ziO7cyg$e*n)V(43D$M_nSe3nW<e4+4{b-x#m@T7<oWPn+Lb!a*2!Ez&HZKHZlSUO`
zT~aIzbqvyn14~i^J+L~keLNC@-ZTgU=Ybk?jN4+>@Yu%`A`S`>nW@y?DU*ckZy$_4
zCd{_cuTUuJ9`q^m#@hlSOsC*Dh@7L+Bwf)bSTVpLjdVYio&r<@?DEs$L7c7{QFlxX
zpb7;}o+i6CBIw<0{o_oUT0)HdSq=VDKJ#9%v_!@3o<VrNvRynM!%(i$z}?*pCn4qo
zi_e~pi7aN5S=b2uF;SPEBlkn_2xtJaD?p+^*Ue@?-g@sa20#%(=Z@&hVj$HfJt_Pq
z+a3n8zB$tW><NhmVUlu%RRe;?u}&pnAGVO8(MXX0Ssxg_Wt-a=^b$a}@+v@w@{s|0
zUCrJ{+8I3+HoCqxM4D)>tStrJ2dB$P0Tj(|JSuQjcqbQewZiJvbp;Cy;i($LyEg!%
zrl?;IT@CZ1?-x~c7;kPP>h9ANn>>=V_PHz`v3Z>l^0RgKtxuo2z<L_NWFqW0a=CW%
zg7W^t_%%1<dnASMeV8+)c(et)3$95yeO+O*{GopIKalwhDR0JOwu{N^Zi0XBM#i^Z
zv|mw64q+u!t5ZgMStT>z`hhibr~3t^%7}rZ`J}6at^o9bbUH^J)-Q-C45AFx5Qcvk
z;Dj*&dA^BeCer@mH%O}f@T3tSAxvizgt*g#bnKBS!`V>>i%GZTbdP3%0XH5`4PFUI
z=&whK2`nfp^p_Y_C|)oet67In8ADbAFkkPcjI7V#857=iFY6Z$Xb0l>h#|jys$D?~
zStIW)5hCn{?01>=2~h!m<yeisSdefWZ?_UqA`h&5*83RVCis2Qk?M{nzClipqYKHv
zi()rX9EpmpYF|v`ulDXb20u^M9!CK^40_H3_?W<{OMUy<<671C4X@wKRSb&I?mNJJ
z>@mjyGaLLt#Yd6&1M(>$J|N#2c0xQ2_p6=sG0W4OipJCJ!bjn#9f=KR)(x?flvge`
zWf|#T$zB=Y(H1R;7gU)qu?;3FCj6F}YvlW2{EW_sQ3ZthC!W$oVyHb{VDo-}49G)y
z3c*vGCoFnFh-6-@EJ)_ran1D;0kdbiIqq(#1;4qBq^UA0HfoCwU)Jv{oIhIpO+5aT
z{IMDQE$3kE>5eGJ94l2~>gB@h9Vwc|Qy(Jkt~nHL@N^Flq+}b+brHQvmc+W`3XyAb
zd`fo^E%;X*Z<gifrT3EA@=>yKo9V#eFyTyIeJ8hU>k-wN;vJdN%_<Z3GBbW8kZ48n
zj5*SWIr#@w$N<w^9bM7bvf-^B9{5b?kEm(nd7H`Ts;DKDZ7n;mcO1|EbaIy9$;oT)
zgh)5s^=#CO@B(7~GRr(Zv0g2ADq<9ZO!_wP$;en_dz*d2x@krReFK>!>*yDSNVd|D
z@JIFc+ue$xt3g8F7+;9&&v}ma{+T%GmfAA0k9zVSXv4~W>+oAjKn4tq7r_`1SM@zM
z>oLo`Se5IK=>CF<XJ1(%cheOomcx78y1r?kHH-oI*^y=(z%4eMYj#u>?!#h$v+Y!A
zpP$9`1gFq>9GQK|mDdWqJ23alzWub`5DTP@si(c)(D_YTXQQ<gV^{7%5rD8K7$4U+
z))9Vk0LR6;EI<LWFg(S65#BYH@OUgvZ4RY8pSZo%FIkMQsGOqZjp?WbF5TiMcd{LG
zXQZG009P-psz-bB()G(#(!4?GEg(K|URB!Z3SZ}Q&Aw@v)_)YT;-nk|#SgLCsk{1m
z_9me};wDyYjbB6Koa61p6I12we!*hP!GPbF<!^bA>^Gl`9pa$-KZX^DHj4P<Z^Rz6
z={Ev~p{cI!A-33bs(u97&+`QBE#bRyb^<A`oy=m8H`5d)TWdqK5zZs!ShTUJwkul#
z?XY}Vi3|bN@lzh+|0C(#!<qd5KfW0zXPFu?$DEbJ#2klmKF%>oPB}&n<s9Zv8s;pP
z!(xs(=1fXC%ps-`LUM{EB&Qrdzx#Xr{@r!$pY6Wi@7Lk^d_JhgpPs41H_0uk6ZJ+S
z-@~#BH)k3jd=?7g-j01z#mVVG>v!IKl>y^W(5C4qDT;SJwL|tIkA_dVA=J{?A0>}%
zY-kJjV*7E1syln-m0lt64;p#P$e1kMsw<yX_L<f(H`RK&UV7uJ_4p0Z%oZW~;jWN?
zDqCA=Aq_{}@MClNgtq|jPX;fmT%#=fiPo^45&+;}e2e`IW5LSsDvIlcw5rqg^=W{L
z*X4eh<x0UUE3{N6R)3%psU1f##$kSR0LRX3AWCfJh1Wijtr*=WbNdprg}w{b9DEBT
zbM=3s(-C=t;?y%Y+<QwL?aVC+#h36|&leHl0wZ>~1qxzM7qmAgXWF}a&!s7;(m<4J
z5wB@3?`_qjtnuKlzMa#`f;mryfi!!?uA4<`nMuIBaJnB;?vi@o8&}Xo*E7D+2m9rz
zYYtRZ7d6(XY{_Lr_DduJT+ZelbD8(B6q-y(;~QK!AO=xjPsJ(Wp-7Mis+**dck0P;
z#Y2&Ee*sY2_DGGm&GEhA{7~a^GbtDo3@xwYs^l$5VzvgX^F@x{=H1K$aMsr9Z~%6%
zN_sYTmr{d4iSA!%4pfpW-$I+Bmhy5}oSp>z1g__Nws0`ctbEZ>E*${pZVu7w!SXhy
zso?3gpAxDB5WOlj`G7>j1*xo<ek7Lv2+*ji>%ro|N8@0&av{Q3zJ77|+`{e{b$h!H
zL~jbPwRXn?si1k{#a=o{ZTM$j&btO}f8r){d8CJFPNx?<6bmmI9fh<@(d1*#cF$o9
zjen(R6QjQiV-5dVv?q?L@X#l(Lsb2ixwj1kDGqprbmPx*$&8{<`qyfsq~tx)=W4sp
z;(l*WsIT|Amy&n{--wGTj{x5s2wCPbh`Z?`*IJLncizGKxFHi5Nx|j5-Ai#dQGp7C
zf;;~7aX0l<9+eV5vu;d(-X4)p;B_g;5C<&~SfLYRPeqkUBFk7~s2j$k9OeL7*9}|O
z*ABq%@h)MpV12((Mdh^_Y|8cs6rVpPdnTa}2Lxe5`8+q#0&2$uwl#F`2QtzFV4NON
z<%qNk;|#*3APtEd`;{VVg%V(nZjG#c)IQN=r&Zl6#%&LXpwV4*^;x)+QBHrbqN@>C
z_ZJ99WO4cU-57n&HE`-xDB}^<Xhc3_?U23{P&9w{6U>#`Zgy!-rlz?t0gSl&hW-WL
zd>ry&q+HYN!?4idiNvY(iOJMn)kf&EgpQ)Z#cpyvky-QU!AV`+wcwP7URadodDT{_
zQ~m40I3VO&=E?P$hqYxJ?`#zBL0d~P;GRE~HO@XsxvwmS+bsSOoFd#T5=-z`PbyyB
z3X4v87&W_O5SxaXW0*Rgz*hcQ8ZfGubCfTrFZ*0%>6i$={%H<8SQfJsDb8B6%%W!#
zO;9$_KT}ygvJF(BLI&SDSDd}?RiZO>jX(1GXLDJs8N(4`(2!{{L2OD6l5$F)4pIF5
zM~APpDR_rj#2dz_gSw8`aHoEGG#uQU$vpJrGPUE0I+wI>Hi%Kk9GD|_FQ1Ek)Ev7e
z_cil)iuQ>(J55mf-L)QWw!G*|p^`Om|HhCZc%}Y<FlL52#*WmXw}lG*gK*a!67BKL
zr!w{<<Uga&_Urs*2ykAuK9K(S?7t}Ie^Zd6(RgU_sl4XSwJwVA2>(#W-6#vzxM<OE
z?e9ZRp)7v}**7l7pDlvOOJTJo{j%FsA@`Hlft~N3r2-eI(gzMT4O!1_%?<;ucLfzY
zym2S2o9c>}`an91Oq!ZTlcmseyQ<WQFG-@(II}1po@ci%9^cSYue!g+_d2gYWC+@9
zoXZQ8SlYaAPb@nMy%-*g&J{K>cqwmKmku6}A#KuvQrB)1UK~#!yV5t8w~r*9bDOju
zpNY?6x9{rVMCJaO>n#PL<DKzYF#*Qt@EsG4cUd4B1tTmVnbU}Oqtbv;C=#-EqyX`$
zV)vW>Z)HSm>yq(x!%^b{Pjb%HCU~E5iK1vvr=<a@Kk;t+((4fc#<1!YPuxX?p%cp;
zMZfuPE{<|V_VJd|2yZ22s0brBoob%8MDWiNv2_Q<A78sBAn(Pg%n{6!dzZ?r3yOnn
zAGXNCoIoJ%exDmolmR|A`@K}9ijQF7)E;UI5uwLdW!N^ytVAsR0Ab2M!4Y9J?E{pS
zaLjcz`c*8Wd5ii<j2E+l-;v9MVC01inWo?GAKU)&`ZzM}q6d1s@Tlr$)^^js8_@P_
z5(FDRDwUtB!`y3s6cZwb8};t5cN;Siooh4=U3S42+1EEZyp8Z(bmN=1%)CvoOLyv|
z98YNU2APCzDFqG%0M6`0n+4uv@ONDDpm_fuhHV_$0M_R}6n^kpXeTe$HPsmh$`&&#
zD0uh4HFJ96yALxC7iBT3?!_D%>~a6rNlE58m-a92HQ$5*<6ZP{aL&$6BBbq@I_F4!
zm=(qfUPlX66WUe;oYYSy*ecU*^k0#g5>u{hw^6ulltIquFKFfO(QgvVF*Ecx<`}rr
z<w<jksWmP&NQWpa#-AXd@wB))-Ls6wLZ1c)lBd&>`SBK=CwhzArFvmdsL|(Afv|hl
zvCwMfW=LanW%IPUn{v4eZ(aS*&27Fz2UF1b$`eVY&o$hqgTn*ir`zeb`iACL8pzJR
zy>k6*ClJx{-MjC<e4GXZ(>QJe1{)xtHHGu1yjRpGR<%Kq=t`0*>zz>ZA1>xk0&YhM
zw_db+0ipHxlA9qHD$nSvdky&%dN|w=fs^gS7i^5bTx9j?!GB?fh&77?o(mwbyepJM
zvsWHW7Q>;HpGef86JO@9;Sy38(L;?KaeLJ-%Z`AyOCkO0>jmb9W1DJuNSv*c@tZdR
zgPPelZvSb6Zj4;uP<o|nAnz^ottx#`ELdFlSH$>K;af%*5nfkb%Q!IJlRYfy6kSwG
zuakeBCXMw<<F*+TD)xs6^@G0J(6wScg0_=r;-+fo@ahWskX6(&4Iy*`fW=%;2Fjyv
zg>vL@kk1W0HTPE`d{9{bP3kggd%na6yS-0z<ypY)RlNHm{AA4E3+qnh_WMDv9n3iQ
zDY$-tHCsr+X>P#L{VQmH8&h4QK`YS=^QWT`Wo)>n-)BzH66TEiEzU#tJXKNrZkJWM
z@vFuPaXay)H-7S;Jdw{WX6X~M;<Vnl?HKYp?JM4Ql3#HBLuA;JrYe5KwjEY~g;dUT
zFHbfl$*@e%8tDAPl)Q^_wy~EM7&w2lawUAs^SRZz)q5Qr3p+4#4YwHh<&`DQKbo)N
z3SYuLA4MU9lE-4qq9l5Os4s;gZ}1*@S)Ufji~$^(siQ9tvDn0g)32I_*|Js~=sp36
zK{^)S^%MTAa4+7RChmy326Q_D+f%$4-f(l-5;&3eTzX759AGIuRwe??>nUkr#j2rR
zQcK+n6dz7y_UHl`Bp0@o<v>IN@Sd>3_;WzPM*4Xitz9wRb|~5r{jB8#DfR-cU>q&j
zr_fuuBVYG?+vR1i#Tq@Iys&Wi<=;}+_JS>-Gt0`LHi#kREjApPn{GtGTQIWJ^FQZ`
zL#O(kyN{E6tkrb5#eR80qPuN$r+Ugf;65+rG~%%K2*I`!v1ePNqm2FNV?az#ejgt+
zxS=0ZWJnU&V)LH+lSQ-Ko8#%P-Lm_Sd7S&#_$Ps<x2I1lN>f5fZ<=-uR0FA~2LyC@
z5%%{x@?u+lvigVJE63K`9F9yZG=G<sHEvrU6F`)I(f1GTF_Ws_s}JX0IH>8e|82Lf
zWGjvS%1xeX7ErUEZoC}*wm0L+I(;A=hU0sr()hbODW+Du=UWP?qn$>bSxwD3OO}6o
zOH%~Y?peOeG=Eq4*XAn85Byuzw(e*0X7O}WQDk@fX@!$hq=Cr{Un6#il6=m!8o!JL
z5f@_R?uPVKp1x;S(<3!->;vI#>q(PuZF>X45Q6V?$lK3Pk4W;RB@a3i{sVOi6-{=H
zm|E-eS@FPmJ$fg;mEYDR#YbhY2`qMbqQk{{bosv?4yhawm;clz^~Qxf-Ttt4DEbu2
z@#kxc&t=-wBe(bIigTekkbuS6>8Yx+#eYF(65xS<80%+!7(a`ojQ}Ce->CCuBZp>%
ze}3m~l&$Tr&-eMn+26DMuua9(KXA9}se4b$0{|D})DYQOqqysqTI%iluNeY>=1aPQ
z5;-x%u-zx*r}+@cfz^i~{q~6Ts%}nu6>tDA_n9_MOfft9KM+4;oU2A=)WPG=(6=`V
z1TSc?8PI-x^o8*P(YBsCW~QW7{CN{7S-EV1*L9x=m!p6%hys64_CY>#4*~0xYy2&v
z{ItE#GV$fqOchVdxA@w5^{xh7N7CQJ>b#9omycjVy7Gu1&Ra=Q+OtM@sj*f>y<M!?
z?Q600otqqiVXl)ksgsFukbw#jrc*tfv7XYB5tgecr&=q!;lmo&pdTof4YXg|Avv_B
z=>C`xT*~*Gi8Fomq-e7Cte}G?_jt;8RJ0m7gKu{@^D3Uyj`Avj%#G<UKfJK~iBc<3
zEcI9rZ<iW=O_i06m2In^t<_#fa_cIWm1`m0WhQWLHXP;fsC4(XjxV%M2VJfFZsnw0
zY;3RewuWQg?1qV)T?#O-5)$7Rv9aDv>FRwsuV@3Q+9|5JSul)4$h;T^vtRo#Q<A=%
zLbK9>Qq|k9l&(G8f%F=F-CsMxFf|3lf(G_0xv%~FiasF~y_kP7lu{zuEg`o3xQxz2
z|MQ>!K-ZL++4xgfGo*(Yt^npzT3dpAqbw+;E2vN1EzOA|En>M@C_V5($$g4gVUXD)
zY`Z%EUV|V)Wl9F_3j|XXDHH`0)y@z#s}_t714X!5LV#6kL?|8w7~crtMN}+N#N)@A
zS4l9#Lh581W3`q9FgQ=TsJgjN^!UtOT%GjPlc;?O<d$z8k)l2MRmn&myFa3Op_6;F
zu74%T$ZPG8sZVKivi8~5BJQhMFiO#;WhJOZa$O12veBgtz(ek9SXh>2mVemQ$B}E@
zP)Ma|m}(BkX2}wKH+Ac(n%c;r3!}OnRdSL|(jCx&-GH7zV8t^Ehkw`3N(eevt*lR~
zpktO5;y{46^*+Pf4Q|(sY$BUnSL*$VIQ3l%d|`2Y-niMK$UX1y-3*^g06%rVoH?iu
z-S=Uy-%D$gpaX;8EP&MJx8qakK*rk6hOyY{2KuQSYQwxUWK@DE;h@ClBO0`O@j{Ku
zeL?5&POffq$fKGItI}b$%S_pb@=3y9-(Nvm$tyQMU6~Ti))I*c?Nhj+j{|*JZePl8
zn^7_sH2CmS&X7T58Oaa~h1N#6XF2gZj6vEc_Q7%#Sd0bsI}(w%Oh8`?cWrIH7PO6K
z(A~C|g?`)EMzc5<mJ@I~6njv7M);%sifOl9W7kU;*RGE@YyN)K5ThP$ei<Ph=K;ad
zHQkTBE|bqr2M*FzUTR5Fmwesco+)!);i0h?FE$dqBtonEzQ(gy#3K2$I7Y%u%25AT
z`{{avF53v+lh*D+>MPPT18P&tvoh#~DzvV~BA*t1x}KB#b@T$+J1z{$rT>{VxkJ_o
zJ>q%C`$3Ep^8wPX$#7h{L}0=gwht5f>}dDQTDjuctJS7k#n$|2^~!0op?hL_c+g~4
zgt_OlGx2rYeO91P>t3sYz#fy%vjiATg9sq-M&VxY0TC1}8PLs(5C@Z%vgFjo**Od4
zCP{H;x1r4ton4$CG@Oz7@q$OQD670D7z$e)ZwAGcifJZ5I5$RAAu!O>G{9L%eF3CS
z@afe`Dqgiu%ae5)o}lvg$H4gQ&C70IROG(a9o59C<k|2}uFa_EqG{aLo#I|i@JI-O
zvo>>Kz|Vcz%|zn3(5q0oRay1u-U)Owu;{z@<*WGU4u#t@CNJ;($~magoK)u5(&5<r
zT&%^Y`&PsdAdc^hcrWg=vCj>a1Z0>z-yW3!Z{!3oQA}D}EK;w@N)`OWN~HEAzCH~S
zM}k*nrT}pa>NT2CYb02cb<^lq<6L^L#R6c|dyJ54Ke_(Y&oCdEtM!lYch)m<y%-zx
z;y%#vp(7&MlOjBcS_vsMsol0RJB&{zZNk0z3s;@fH_?1|OFM3V)p`)e1asE=;n3P^
zG|rmIMaV~PX70hdcc0R(;eb~3jfm$34^{LbvNyss(}m#v;-P!onf>YfT31js`@Ux5
z`i`CMbVMAWXncf$jJRoD=aZUg5-wouE_|mQEeZcrmneSa<7mL)eapQ%X?5TLk4TdW
z`mmd*X5qG`sQ}YV?2}cG`~w7ZCz0jCG^9Gnx{Tj*E4gEz(MGV}il}-0uo3&di>HYC
z6e_hNDgBduun^;XJe3E@c?JG<Di=Nfi=*h4=4dCMFEfGf7@Xn!Aj(Qs?PPIw+4AkJ
zlkXlPQS@E%w=Q4$l&qX58_l85riiof@TJ=s$TaNmyefPimf;g*((Ln{ey%#^X@|Ui
zJK~6_cX?qWr+G)FLc~SeMHATZ^Gvz@t*b32UyyR0B@9Je!OS)Dqa{B+1~IOB-1Zin
zwpzwr;8pkHP7!1tTsI|m32?mIV$s%xI!@qjUXvTg^>=Bmv;3B_zmwlaS<LQ8#~Q`q
zeyK&Ube}g_ay`xW*F&bngb6(Hl*)2`OnbVqIqH46Gr^3us1#x~D>Ym}rzv#tiFfC`
z;c}tG%YNK1dF3DNh058_ffL#GaoLPa6UWCcSd@FQo-bU)=VmUS<R{mS+kCV{%iR$O
z2Z5O0UXjTo;h@>2LmNSw3!juBJi1P-gnp~KncnHdQ;IBj6QKK>`-9JmvaO@J2JvU!
ztkc!qw_m2o&2&0d;|;n@HfiP`yeuf{<Yg`_YYCKVw@cIeai=ez%26=AKcFt_^{S<C
zscNad%(#x1kZjPl=>kS%-MGXn@4y^~4<cX;1`b!VKlYR?29ee2R8>262qElyBD`q~
zurjj)psZ};`a0ZZ&HTw!f3KqUDqc~Z5(DL>x{7x#4`<T+M*G`qohEn;o#bw(lSB^H
ztg;kc<aKAu{!XbUsLRAZ2+$;GG)ue44_w)fe3wo4Wu^X2`KNMjK`iWP_Ta0DR~fP|
z1nT(om&#w*w8pXvX*|e1GJw-0;rlJ`eVWDp7)BVtzPC8IB7DG``oZiD11hNUC&f1R
z2)ccc2JUMk!)}jAp!5M05DBdc<=!ymmYS-ofm{|=OYak1_=HBr1PA3~lXvvg^$h7X
zJ)z}dmBez0IvXtJn1y7`?GE|noi)>!bsORD&X?uh(vX><tyS4P%qBW{xBOqOljSv`
zFYXU+saXi_aU&m?%nGL73Ksb`Z*na<o|mzuh4AD}GRYOuv^^~|m+g@XkG=8m(3a-K
z<KexV$r1Sk9eoXo@PDA10?L~|5*t1YsOby)eqIj2S3#_xJ{bt31T(N|%QB~-%F_7*
zR%wJ#%io6s{q$yrLypZUI>nh|S^OUhYtm5Fx8I1sH_Z@bL??}`HVU;g*&)^5AMh)b
zCsQl0dNEk;ReawAM#{<Ps=taDdQAIrix09qPZYpC5tBuSId1o0yH)@{6%Jl<k~HG!
zv?7Jl$Q`-)Z`5ez0FiYO$!{3g&UcH(H9mh~*K9XzV1$M*BTQHf_Nw9I4myEbW^grM
zHI}@NW>i*Vl?d86qJn=-pR^6YI<dULOED}x6~LQds5!IZl1A?zFq8WUY7fE|1$37w
zxa}O7L*xwrtB$#U_&EGOkenrVH(v_M-LcyxhDXecS30`%UaTN3Dw1`Np<KFb)g;mM
z^Tc?DoE(RMDtP6XN^_jJfCe#PM>Q#3R_(d<P+<sh1at$Q#N5K57~skXBCA)@@T@_U
zmyfKj%R=D%2}pY>WwC2>9_2OW?eG!#&jqq%<q=y~ymxvrcYBu?WMRLme#?lrx>{wq
z>(1>Se-kO5n*mN?XObwI{6eiN-9(pIHdLR<Afwi3gf)-gGQqze$RZ*LcJCvJ?pXXG
z5`rCbiA@;b1^01N!r6iC!{=mXm)r|uk4nVK4bi-X5{5hI3$5*??tDKvg<Lkr=<Ep-
zzC&wB8ZmNIQ$Kl8%k`xfQ#}WEa348G^TZQOLQIt7#c9P+3z#(+GXr9RDh>jEX1P8P
zUnrm{A<_y{i|?<d<yQ0}rYl(i4<YhzkS(&l9D{wi)o=XlX9<$~OK=iuU*20K-TlYb
z{d4zD3jen!sd{fdFDQp;$3hSFAp2E%$%m+ioI|f=O0)D1ZR%<N17SmIC{Jesa?T@;
zE=w_PI;s+kSzG%EsRw>>YVBt$aW3bI-T#D5ef93*Z`;bvMsDwK>a$t<8#Wx&USC4S
zoiO&YEobpla}kfe*C;7hq16L`#z$&#vo-V8Ac`MJ<sXR7M033pp8Nf#vDxxz$hBz3
z)wjYAuALgWll*Jmp3}srM{FxiOtDq4Z9aT{qh&dF6D<SmP-C{H?>B5aIxnB}1k{BI
z{E!*<JU%Ro52ERK2wwm0I~70rzO(rXWvW?L{Z=k>jxxxNu(QqSEoL?4k$=&`bmACz
zD8#raLS02YWDUgl^Np43e@XpTJ0G|I6ihW(9#3T=E+-S%<9~UdCoZYG_Z`y}L>I`a
z*YDR|`-jIy8rF<QGM$d-j~WLRIXf<rZk+K9b~)W!5qEM8J<|!baaXswCA?cFv8~N9
zbBcU?PJNrIsc@6wRBG(%-7&-RH1p)}9Ubn~DA4_+J|Nn~4}X)tAp8OJ&``B*D4G%N
zZ{oB1?}g*%BFLuKDtVuG%Cc1Xg;C2~+u8giBHrH<GE}$hE7fbN0>1{iJ`S{b)L+Wy
za~)1vw{Nr^4shikD3-SU`w#T2{k6qHp~{r1N}iwM&mc=MxywKZyMfb%#Uxk$lIYo(
zOk<eaJ-;Q_;R*-R9c|KF0loaedcsNtJNuQ&nx8eMk%n&#SGUk|O&;oAWn0?ejO&07
z)gAfGWJ>v%F=b0sEC*;OFlCAJx7je!D$X7&e#jwn9ah7I49Od&f#MQ^@rb>+>9xr`
z&!2A<T~nq^+uawHe@eY>H4d(J8_kuv%>L-uI$;@J0h}&C=&!K*a;a^4L%vO*1ee|A
zNKQ-b+)do4H_m+$-XsHiFIlPF9oEW~kCS6c+ir_5Pl^F*M4mdUktfA5i(~t4;!671
z&%*&}>WnFzzk9Oeb@_Lb5;el4=ta0;g-dlnp)$!LKV=#SwdCpnxiK#Um`p<+El~N$
zzAW*Ulu~1X9Idpv-++jQM>G|bUSwuy9yIK<gk>7|9RQ5plZne4Xq?iE*h{kAN<(1}
zZyggTa(pMX1#ZDto~f~Y%#>g*bmwq}pMHaSKNp^5@8-qn$Y{Z2;H-D=)EfTqHoBuf
z|Jc@eTeGK|)IE1s5_#p7@tat!fHW!fJ=DYNHxC!NG`~JriD1N*c{+O8u&-Yn9c#^m
z3)ETuvn^_KH2hPZrm~~u`A~#legTdmx|+2jCVm|MF_mBrW^?^eez;sWiN(WbM}Y_H
znyzjrkrpE61wXbtCF+5Ij(yNZBkNBbgil=@tS!#3;KX?Edhpx<sr#VQS5&;cP_Lk2
zbooQ&u-)t_srOwES7V5rg$`h27JRCA*bm*%y|%*Vn)DwCfdsw3|LwfRi+@&`zf?oj
z({ii7<U2P>mD2?Wbt+E#n3D@ALb5ObbK*+ZlUovM0VALQ4VmY?WwsF2HgbT&fMS7;
zy>z-g$#P0v)mBx-d>;kd!Sw5^U>2gRnVrN3DMF<&@u{WCoMRZU0iRb5HAcuUoKPnQ
zDfCXmUG|8J!;x}-nJwPsKvlgN^M`qK31*IqR`M<zJ@q>x^1lo?x9av+Z5Y%{po>Ue
zN_$3Umi3n{a=a(`a4Ck%aNn5_;I9JCL?<wVgz~Q-vOo{l=f+zL)g?e>rqkUA#2bKL
zSz2JdpDjEp05F@%7W!Kig{LHb!=l;s7i|S1+=*V+dVnI*>E<D$#!i_?#t<9e-S2`R
zWc=<x1t1uiCekLe0@`7sr5XQ)33aps?Dsm_6dF=kIXlgTjO>(Z2PUeY%ud$1E3<>g
zA}};>-`qE9@8$zlIO(Xptk*TG0eO<ywSGygRN(<J2*Yay&Qbcg!>YK$a{q;3jiyTp
zFT`?G8?9DBmGHGAOWkw@4{#`iKswD2pG<5q^6^k#iB<I^0Ko3oQ-%ER8nAxXTp6F$
zHG7x5;+>g()#&|4h0$~H7rM_#cjv0@GWXhK%%?k00!Ni-|8}^)?ggwdKu1p~b&Oxl
zBeo5|*9}?4C0I+(5w{?WW{ypucT1^c97u)g%Igxthu=%c*0Vr6Jd_SZ!5)yskQd1y
zRYu_>cV2ub(X&1%-~HueY#dx|475jdL9mDh0T_fhW%U3<76>n}&SvB810h!N?<PRN
zz?mMrLIvxE`_$?oAvd8?RFbygCQ*BhpIUCLQ~E39UT<3Bi(h2^p*UupCdPB-rpp?l
z4VH)I+rmE;FrmX1CmuMDbhBZ<n_mAOpLwT>^ro-0<#t}YOIdU5TfD1qqM#jnQBEW-
z;><)q`FkQEw{4=hKJkY$Zz{&8=1)`xvg@Mk@bj`OE9bFkBYZCF4<04G%u$|+XR8S#
z;~ulM{8W+R9Aq|I?33edIGOK%IbR9dM|}D1$lr1z`s_ZvY-Bz8R~V6%lRPE*kUep?
zS0YAoyC>qFf@OP$@DH492`~`-971ao7;0`%d%b($Ve_V~bySo~Vl)pr1e)A>pO&!m
z#5*XU#?jb}MaqRhlSsR_)Frk!QkPj`@~CIE&=~StB5`8J@6}ns;>L4u&%h0_iK%o}
z-H1aXz0$lI&p|3H-#lXc{YETc6)|R<SLFUv01S6k{wi#1gHT)^rXu(3H#b&mHC3D<
zo>PF_`5guR*~zu?EzQbGZZiHE*Y84D+Tdamy_nc=A2VZ=#($u6!{xId=Pqwww!Z>T
zM*>rqz@Yy?#;nmQd#?t`&CG0~V|KHl2Z|J-^|91ipD&kbZd<XY7LBIyYT{PTM|S4(
z!k;0WtRDvC^RL~#Y1nZ0Zg&Y|EuUT^GZ0J&TF?-j2@CHHO6?5b*R(KcX+Lqe(TKP+
z$oSoQsj;v>>k#QIjNcCwOz-z8D7fF&4>g|X4j2)#YvWzZHxFCc!3cKBN7;iM^FEN|
zYP$P$$q*v(e#mOu$qShvsmH_oYj-~DavcbX?Ubz%)=|=ed?SKzqN^!1+)Eh-zM`ag
z7VesfS24%HqYWRw)x8s+5@;j~%d=43fZsGPzK)jCy-MMevl|YoQoU5dy9*$gG4RDc
zVMKpv3WwZbg@~GsF)8tLSY_4{=~eHJg({G%n_>PoU++pMe+OxbE=a*lN`GzDeXwR7
z5KC9j`G=LoDuquOd{P&YoZ}6e1aRx~gF$|S5_7H#054Q3cOuvpX)Gt7-(NC?OT3L{
zy%)@|O!!<P(r@6({opy<g11{kjAWk30BUV)q6y>L@PH^3&9>v3i8-wXd-xKgXwq+a
zmE$#NGCT3jhPt<z0&T~AMIo;4wBYu3MpWhX$O;?G1x)%S0mQRrB)wZO%K`fL+BNwe
zWwkF-@pAH;A_*%`)Ww|=e}L+x`o*sl!2#HH+DNF2365svIMZ^b@0oV@{X50f`dY6U
zp99wZ`6aPdr{FEMxc%2VeIAnA_*GGhIYd^q^{0m%p|SJSA)hON5H3T4`P;KD|C@+v
zHnr8#7aR-=e%+<pany8|<<;MwQnTD7a56I^Mt-ez5k0mP)csRs?EQ7eJ<y7FQ2M`#
zL^-U!Ti%Rr=m7DOWjg-t7)@j3+_|EFRID8^DU0x|w?VZLZg(X)MQK;x@@2mFYsvbU
zadOq*;%9y5TXG3NbPF)Dc8_UuPpfw<htYkuH_@I{XfDId$7LQ{UOM`r+1{L$YIAf;
zm2)%+GU7bQ@T8r1YxH)KIYSgz1lVHmF|8hp0EM_RrjWNIviayG=G&KmVdeg)*tR9K
zpOnheCcA}c<UUGqbd;peR->*D5F184Dclus(lR$JE4OcrSZbLAnn<+F&?qXE-Cx3T
zwC|ruUI>V@!sfM_6hdp{#eo-Usx~n{`?j2EKBF^)F99P9%4Z>hB(#hCykbtB{R0)o
zaoY4z+omi6m${EEgZ5zzV0DSvm@RS8pt{z<Vd)^?B4pVv`uHLQt*}poURzR+W-JtO
zU!l?rm|vz{%GC#9^%=e<Zp^&OKwqAZQ7A9s)2|wG%UzNUA{3{ns>|7~W^H?qP2=8P
zH;qk~em=K7ojVt{Osup)#>$QclGVoOm|m8k{*IlJcJce}qZ9@vmXca^FL2;(42<vc
z7}$b&Yxn+132~GTWOZ$-k<R$YJ7O|RHr|2PygMsZCw0T(9SW!`tpnQsu+ii$RAr7m
zrjJdx)4%a!@HqT)kVt^u$2hU_`(-J`E+ZYfnsR)D`=r>}PXQU)Mc!0Zobg^Aqm~C(
z6`4lp{gtkuJBP>jyUsFa2lAMe701WcA_`FzGgo9xmnf<^Vp45ibhX)N-grD#k8`G|
zQJG(ZRL4Bccu<%T>|Uj1T=JMNM-O+ODip6^T+dG|C6dI2VI{!j<T*%bhJY@pVHN;S
zo(Llo?`a8;iwaM$OXf_b(xsp@m8e)oH}W?JHbdv^+!%Tj8n&WJ%>-HA0oD35ahiWC
zY~jWp%EWpSH?O&@Fijp_ndm|Qlf6L}+eXRw_X!}`SQaRX=z^sf$^pUwG#k*Ht5<Nt
zo)cXBZ_&Q7#5Z-c%Z2)VhSL9aPfI<p#cB=XqSovF9Dg6Wt7jGPx98Fq&B|IilKjx$
zmfIwkUf_O{{dvk(!eVC2IY647Whal-|CFLWE-;mmGRauQfSD4w1j5!5kVk4-OWfD}
zv}8-dQVZAW>o5m=;t=e=xazRX+B~L3E5A&>v)9RQ+b(|n`jL(ODRc)_rO)WLTi8V8
zpwxLg*^j?cQabp9{gfej-n=Q5IQ8}N5hG}4#PLU+VnpTO$!N6U-t}jb#r$kd)2m?y
zr%-+>mgiis;{<^XQ&x7$<<eU`t4m+ZU2%R|@Riaf_%fWdG-cFLkUE!KM15<jcdd_t
zF+h$nu&vE>Wulu;E+u$8Oy+=#L=<`@E}sj3KWZKG`WaYXdw^JNtB&ymzboPMnU1_F
z_Hr?aMx{C;`kI6+8=PjI>pcR;w&AF4DvlOan5>@8Xw6yE{QD&uFfbloBl3Iinmu<t
z7i(ELjEPpjKoJb|T_GQ2IzI+`sLN~GKVUcwqa<95o;O>S8cY@0!NmMN=ehRyWyv~p
zzwVYJ+~p-?@WJ;_|6=g*zDngsQ&pKsE&TK5(e~)D@4f9YQ_?YN<V)L=opf0R+M|wB
z!EL8{niEt=(Z3|$_j;}EOqsA;O&y1__>~?Zj$J8_-OBVwhbj!F0ru>B8%MUPdX_S}
z%H{h%<uJXfrSJg~4S0kWASPq4cqrSuLPot<xJs9mHywglKABaDc6jF6z-?Q@{Vr_d
z=2qEj*&;C<KRv<mE(El2{+Qge<b86fT-Q@^9UYLmiQm6q%9>9LU;We47j%`AcZhJ_
zC*E_elW#oirCC+RKXd~Esx@44CgCYk_j}E)K_Gov1_WO%WLQn>Z*(oSZc781d4oPq
ziv8VNbM4VbbCKJxHXP&DNlmCK7J;1+afhHnpCE7#L?DnHBh~PE??}d1e)D9I!Gr94
z$5OVYp#s=Eu^z1(?+Qwr<z*;yi%dS%SNp<E3n1<iUV`E%dlmPK6=WR}vo75i7OkSF
z8U|MiFuTR+zuNvpn~Vnv{!V_OC(G*?`BORaL%2*^J8zom7pZT>-%6M}%hRHC2qdAm
ze}|E-(g|4$vdhBM*S_+CYZ3TPJtmWw-d3^INPl&&RC(Qn01W+?{B$I`pVWeqv20v4
z&}#1T<j4{Jbn;y~h|tJv9A%A>NEnU<4L_H)Z6k{|);)rY+C(^A)R_oI5_s>s7Pd7N
z&X`I5xDk*RVRc8hd-#{njg;}WFegOBUa1=y<dHm)OtZUbI-A)9E}V!l$yGZj?YHE+
zC5LwmIHJbkyMSu<>qaYLZ%l`s&2J(m__r*7igcLTPP#rDQM|^uk3^C!jsdidWsOzI
zjUe^)#H!<z3K8i-SJnY}Z-ubOqH0M>0{?+v&wN!ces{K!(bhTnx}5Sv*rDjt`1{7<
zhx?l%<!3F*rxi^10z(&LIsOU>9Kce77+2`$o?IsAykR#Hell?9*BmCk;N2;%<ShLh
zNpt!4!1u_;GZS^8n<o_|KgIcQDsw_6>VxMrEMd(<VL@+pKU4s9SV8WGag?GE&#H$#
z|9Jj6mw+&$nc4LN*_1_9tjh@APT?kvs-9a5!+1;$NWJ7()%|t-X_Anr#@rXLRKyp3
z<Izm%)8NX9YOn7iHajTE<avV+&Y2?;rho?Y*Qc9*hj!a`Opl&H>ZrL8#2IRd(EE5q
zkYz3$a9)Rz?VYYcqq1n&Af*EAnf(d`YuZO}B9vAPJ4G$Wf5Tr;Cec+aLJ-9y3w(hm
zd;bHx0bg4g3iAHJ6LD@xi@#P?2xr2gnOHOAmZvXa$%Ina-0;wYSCZs3*4>z3K{0K%
z@hKkpaR+Y}ZAfc~Yx#M1Ksr3bN5kj&w)@}B5BDh|dob`T50{+pYViu`glpRv#lOoS
z03#+K^z@PdC9oGf#%UZRv+VoaEPH!C1}V_`H}fHvf_w$4Pq3I@DcaalWIZ;hFpK6Y
zUiNv{%NPaxzRl6!b_=|?grDot^aFfX6*qU0;PZ_E*yoN*^Ow>;2q%AOxby*A7=deR
zM9I&m)ZQoxYw@SBX`|OE_Iou-=B-^=M<`uDVe>=_>bTITZb0Nq?GK2+==ae|;mVc3
z2=_*hCo)gpG`ivJ7WtxibbI^Kbw7q<#AG_tSQP9sTp?I9Ys-uy0am#UEm!oW(dY~#
zn(mssC+Gu!uho3`1Kk6ymD8Vwh~Pk}pS1EL$unt9@w@<)9l9x2IU+iyOoN5K=&UQf
zqFpP8qm@nFTJY#n+GeSc`}G7U350Ve!ZC#??lOBCwy3jp;)%K3w=ftLxcLHAKKCo=
zgE%Mi%VR@{5Km#{GN-MLK3Nz6%<=&i6D;3Of~-<X=^!jUSnGUW?+b2uM+~%oBp+U?
zOkBUQoQ)q3S?bU0NeUN^=)qLqW7~I$TSXsfv~zv9{egE%u21?Sm!DF4_wl&eUif5e
zhhIvGi2tWr(A+1~HJ|J+Pb-%;jfR_baia<;Te1@18BY}lo@@Cn^?~2R=Js7T%%&9W
znw}_bU3Gq&njS%*4LuwA^djV<gW=Z<oy^Tm>j7H#O6KIV-vI=F0AO)tp5+;3=PFU1
z^IA06v7lmPW&H(a?TRasVJFal-gqUmqw8hQ<kGCxn?z<8foKg+2>;U652-hCU$3kd
zqUG))(!Hhkl8hcizkaT_jMI@n4G35Xb#W@E*JW}vRo<jlf}?LXDRuOfR&cW}Yidn)
z5$T|xt;~h{nup7}ChM*lXd<6^9jsf~9YA)8^w^x`-8!2;!a0B5a%(=Gj%t!cMjF04
zRjiQs4^$}n)&~Wmq7h%zTOxlpOog65>W?6-4LblEKHHb4Cj2`p&1>g&(Uuq}kJQv#
z_RdH#FY;aTK609AS^t#6=$N_{^F6(ef90SO9URR(9sS_jwJ^`<ns8mzQjA)hzC-9L
zo!|QhnxkEg_kj?X#N&+jYk!GsH~!^-cWAZr2Vy*CT0J^Fo9?kVdX!Yj9VeU9r{Xrx
z#}|jE$yaSdPg#9Nkr$0K4hHCpZ}kL~#4Ph|=t_IwKI&O`8Enp%a&J$PwDF;P#DMV{
z0a(D`x`_)L63f!3yVk%>lFy#JX&U`ianN^wzQZ$rzdsxHGMZ<4jlE(Y`B6ceJweY!
zf)qf8_3QTVHCSkN38mpW2P_6%W29QfedBNJ^viXx7VAz-mQ@?snA;ft6h&keXM#jr
zxvua>PRJF@d%aI_x42>mq2n{rRC8?>OzT{#G@gCVR{<&5k^W(*h3&45JR6gGQ%Jt5
zSMiLX>fS8Rp#-JnO$)j^S`aAuMtArUgw|&7K(@M8et%B;>pbUsu9sIF9@4|or^Y18
zMU6RLUoPAMOWOquaG!}wh+X{(;lKhJBd=LbH27uwUt{}OZnpeM>Dh$*Vj#&I*w5eO
zXp}RLKr~^AI=N`&4ZJ={UQ@%RTikWe;S1_Zd}SbeErkAC|AjTRFMS#P^$3@t%uD_X
zv2-^i_y##q=4QftCO070%IwV3y5pz~t2Xi-4HRl7tzBbh&Wvf8*7Rg`c<I}n9kCp}
zKk3@^x0^fbd5=C{=2Jp={|*wC@cin4SlBVn(Qlj0&<)7Ph5voa%ZcXrX0P}7@(ZO;
z41GZ=Iq5RRO=*=G8YFyinLhQRd|YI^McZ1ba*O{rHp9UZXQNu!!j7rOqN!Y5pD4{~
z+s5RJAC}m$eIor~_vw0%;4@9bTMVZ-&~j!Wpsa_SEsm&0X=KG5)8i&a=siiQ33?)*
zdliJMQZZghH_?h$+?7*Kd(OK>OK;yj)<yqM94T!WLXX!2ILts)!K?Gpn3naBq3GEM
z=UH2-obgXNJ{GfeTe3_DZ)L0jDPvm1utvFFf2Z*FK7`fw;VRQP5JHp!5^qitXjv0J
zAuCE2SA4AC*V1=w_+L8-=KJJwCq!GJUB6CYQt0otX>_8Mk;|(cm=?tpp!5BY+=pO*
z5r0%MiLj*p&mD$ZnW)$LKAA5G@1r$mRe40}?c-B%i6fwO&-ZOy|E6bJUv%sf9Oear
z9I1owRN8cM>7$!VmF*Y_Xs&<ImiIBZ8~(Z>PuyBKJ$+rvJZc?&ZWS+A(r+L{aHYjx
z03fQ&Pv~dVTGoNg4p7*wR4GD8T3_SeWOE30zU+>_<}-pV2htPf2Bzk#Faw`9k{zsW
zwUhl;V}w0iyGNnTH~XwyV}Xd0p-lY%Y+<<jEB#iDyA|BQ;8@VwpGvn5{dZa!Wa2hD
z5jykm3ZtR8M)t!q#K((9;>z2DNU(|UOAB9eAUZr^l2Cmq-9?%ebwHwH8vgseGv3c{
z?NnlNa)X9@D=dhjX$aB`i?}ZQ;BJDK4nqnE<S`Q?+|gn5_$U{$zhNyLwsreCt=NNO
z43Bh-5Fe@-Cbp^>N~Li>(*1lp<S#GR2+|=$ypLTV6CZ9FPxqZswu{oN;4&xZ$}h0+
zm{NMpu72;ZN>|>)lIVHEmf&2l#7xN+p06-+jvzWdMMsU2#a4f+OTD>n1m8z>xqmH|
zk$RM!v^*}_V2x||Q0*6jnGVS{Re1uc{1D772D?JEn~)vcm@H+&CZCKeCjMA{Rqny3
z$-}~aMoc$+T_`TWHamj&#sU%>{yv#|$WWLC=EyM!wO-A->92(FWZy?aL6&DD5O%jz
zZ~8AA=$QIVV=wdo1PS5ZMGfb(X_-0Db@OM_+^px7;L}$W4Lc;+)9Aed=2WiCpX12k
z^kUAN==DAU_LD~kC0oE>`^NwlZw_HthLb2i&}=r#HL85O=pNpn<OnNy<}qd$w+uG>
zdz>a~4%3KEU#@80apeXhqk_u`OV=0Fz0fDK4gY~6r$BK(FyYGt55j|gj>`i(bm{J@
z%vB6x_|Ws_jd^27cWdo4<>HUZMij1J;f~E3`8#BB18pfKD74gkCtqbC{hFdc*<8Ma
zPB)R&NqBQ!D6QhxhtSiX&yABSz^FZ<n{SUSqDlO;5<3;Co~Y3&T=@I~c~z=r=fOL(
zr+%Yh&N$jIjOeNr(<r(xBA)3D!+04x8#M;Sf4Ocp@=l3eGDn*N8yQ%D-=&yB{Un%|
zG_KwSyp-rLnYU9#SgZKQ-&t!Szd8Ek+{>9)gKYV<O<kvZ`aC4il<^hQJ`%jlEtaz}
zAL|J&Da`&A1%{q;WKp~vV-%bJ107v{3=AA8^mTCfw1i;H!Rv5;wo3cTZOOQ!amXAS
zuPx!!SaV0cnSZb6n2@IYC(18-!;v2Kr?f8e5`WDNx*{$t08s8a4-b31j8f(XF2Z{(
zL1&$EYUe=gmYidb#y%+QYLdqiqwLRzX1jG(ZHu2f7W+V9(Y@wo&q>hothUz#(zA1}
zoR@-&85XU(Nbof0!8dsQ%7|~#h|h47?5*D6{K)r{r7u`UGrmK1YQ&!fIQ&(yxG%4|
zg8>(23-Lk`SzRjTAbMR1HdDQ3QHb28abyrKJ#Qv6ELh=vG0%_Afc3jpInNY5D#nw~
zks<w3K5%9fp9&^Wj~WC2itK3%H<!NwV$P(LYOk!Rd&#H2D!!AH)IB|r9P9H2E4N?0
z@@V|l&#WCn!7X=heF?38FIgTFU7s7~J;vF2uI}a+Xo&akDBGPMXgyU|uyMfVW4Il4
zQp2mG96Y)h5acV>l(p~r-*kOc)odT!P%hhNU)VjuJnpVxK*VwX@*z<l--(FeY84P^
zAHJb7nI0@;_+$N{k(Qv|+y%N2R*9kq1P5tjy*Ajag5ifBlvW`FvvIsvEOf@eW8)M<
zV<MGo*2o<DPnmZi)9j9&+@XU<*_=z-#lDBk(ZN>-whWXjLt+@@h5iF|6qB4X25e~-
za}U9x{H0o<)xV4`@YzN7aC2F>c#gVWjK5F&nE6V>UeycE2P<x8Qbvif_eAw>-zIU>
zviH<28i`1b;^Pa3Dc(A5br0Z|xFB~<d+_SFR4?Xn>R!%gE0c&EWOp`zj0K<_3g-`2
zS%uPLPJiu%Z}w|Gfj#&(NI}viwH)s^CC+>2-h;2F+xy?WbGKiuvcwNCh>5cp7ueo1
zP%`o^<z;@eN8e7Z?!?l<G-VH~S5is@b45;xYtv&uY;pB4;(~LGS2%09IZb{*oT+g5
zUp*L(GZ+wyXbPwWpczPS_Gq&8EC(%s=FwZi&ne#hwuMFwZep;L>xx~|sPJ{5@GJAS
zFXH+#vG$``uI_byRMu}Z#RF+%pnk&bp|#{X``3t6&y)@{wY9pYBPliKB<G7WHp)Cb
z^#P-EN0PEOP#)GR=tn^F{Qm<53Y2`AJ;sD7ygwe?`5nh!148*aL#0b8uk$B{Ug^R>
z>J{)N7i<De8k52QlCtUFp7f<1**Ycrlm9MIBWS{NQGl1QHjFPZv$9{pw#MC_S6&6+
zKlH~5Aet7pB;jO<r59=oOKFT(@`H(c#GAf_Yk#X6*<rgQRt7o!(YYM>)$YBjaY|b<
ztNKQn_)=xt^d0%*O#Av*!YJ-4ceh_81X?4nWOd+Ut%_BKgs~g{n8lAm6p`+4hvnO@
zE<M;&$^M8H%a)3tEP`yaad>)(fGrb{INWh!(y8m%l16AD^T^6ql#K?s6_+z7U%A6U
zg;~t;n^3V<SM5$!7x7Pd%F4V^kH9KDvj2e`Hh2n$2T#5AU$G3Oy#57{SqQgnzR~n$
ztURh%c;jVf7W`G+TF+UZN$Ii#jWZf_!1Y2p^X@=!ESSh-uEj@yzGPmTEKDB>6Ze>{
z4+O!9EI{~emOpf_s?njd9{(+bb7D2-BA=qtQDdeQKFYQGn6dtct3?cIQ{U6g+s*fr
zCSz_o&mR!bg1W<9N8IL7?!xds`t*xjyg|6O^a7#~-92yj>dXOXWjEr><s3F0YNfid
zNN^!%+fykCK@V$zt(E=PJU^&rZmz9Kqffe5@N|!bIJ&)mZ8E0_-%sLyL2ZAbP0n7}
zA8~&?KHb3VeObNBt8WUCH5xBhx~Gg>c>!k`RQDo?8pb#BWIx#<u(0nWJGlzKuMrn8
zH%}4Z+p$2dsRf#V6yy!Yg-H0h<<+$*B;9juab8R|IdV*v37a~ZTmk5hjX0^9UTT?W
zMsvWM_Ir=0y^MpxKtB^N%JTPZWAZT)V3C79rBq`}c)n%hw3|yBi*TXJ6A8HlqLTsu
zPl{>+mHZGY0u<>=u%2}Eg$$Jm+*}CJGuon0F{WSr=M@0~$hNt6?qe*){Knc2ai#bA
zqpe8cOKQ-N)FZD|VE#D3%o<=7cC)3{r9i}TSD2n3Tlb);?Xr!{p%m5g3vjpt*qtK8
zM}ZwTODC+P@R`0MHT3l9pW0;#g?sInI7eQYcKWGGTY47X&)=GAwoI@T_94!;WF9BI
zUtdSm%oUpc`IT(G82K4@G`W__@9}gxIvg#|u|^2wR+O^p4>5c{+EvG!Yz`F^Khf)B
z&|1;q&@2sJN~0+Jd#0IJSM9dL`*`TO<1O=4qkqvJE1nPidsf@hin;^LPxU(gHg!fQ
zs(k|R>S#ou_^&V+NPG=1^+k<dZqY8nU6<57zQY7w{HCM1`{8vzYDQsQ8_ueLdtV}Y
z<}w(%`uO}&_rdtYRPL$agMT;<d>Y%TH+m%;0hGVt+p+9o@xOsEB--<)Y0l85<&RS<
zdw)0jce<CG1ynVk#+{4VYVP+OPmxl)0HE%GFaA41kdN*^(D{gCP1t$qz%N-(?o~~W
zm*<kA%9Jfz%^I`XIR2gjgu>@U0FKM>>s0D`(PEeS?0ML4O>Ce4rLnV@y-H~ZHcrrG
zOPQqS?F=7k7+xmxkJfu=%gE?Uf}ZdJq?tYep6F3YrIkm0q6*h<t3Fz#1VT3)zH9qb
z<&b53`#I(?t~AYZ98bg4D?uY5<R6qG6R)|vLbBWA3Nw)%YNiCAZk*4`EdA`RG7I6S
zkOg+a2IJwt6p%SK@sZe~98Tplh?&FijAZEzs!`KafV`xSvcP5`ZD?)Vnnm}XDD}44
zpE`c~09$7nJP+n>5@+(bEO6_7)?;pkF&9oO7;L0|th@-YYW<&hy%zk|<5Ho!l`WVt
z0%%pY^bEhWy0M*n{d4(ENO|CeL#e%|I`|78K9FX<!k-8IVt_H|@Dk;>?~=ySL8!C^
zX1YEasFjs~Xux4$_(Vc|yijrG&!+0jGP5cMG*bQM38;^MOl;&%*FL4!_tF$Jpg)g%
zA7Gil%EB7@p){2T6txi18itmy5)3mJK)dTYErK!iKk^bRuExMbsqM=(y5?ZK_j5N<
z&|H4d%*z9pfL7RVspN>DID9ixFItSdPZVdI%F97E#?X5viU-#Vml}yrHGSDy45sVF
z>FKTJEUlTnb9vMKc_>Wga&(Om?NFRqYvBza)zP4oH%^>ZZml^`x9l5*Zw~=HPG2%6
z0!*<-@hK28$R2nzz#JNm_clZX;KNEClrXFkATnHFtyFw;0UFf=-u7wPqR^&^IPfBF
zS;gIYlhTTG*Vz#Kb_cE)_>=ukt%g2%L88bvK11k#*gfYUz`j>X3~R-(ee15=+B~D9
zx$ynW_XC;a>I+hr5p*=n(WnhN{pIvZ7t{>|58||(Hg7@d|39$g5~PwQw^nzGL525B
zg0`0YBXRkx1Sc((k(@pfh>&|$un_ctbii+!#usk+qQxXg|99d(IXu`@u3wUyxbmUj
z>#TD^J`M%vuMs~`6cFcLFX}It0;Uw<A*Je7H{}vPR0Ta+A8{~nHy(xU#+q11Hq03)
zPVwO-UuJyhVXN`}sBbOYYA@n8EY-F*h&*Ly10ZN=u!a@3F=~um+`usvys)#eY3NT6
zW26P%%g~e08458V5={=T+rpVdCw3bLa&{&wUTNDbs*x3*5090zc)`7qi5?m^CUR!z
z|3)FDFDuA?q~*&?-6JwdI5XxC*<-?17NV>U+wsyO2AR{WS)jdEk58>)ynovIq-(cN
zI`4xrC-AW`y!Cw#Xfx8b^u(yIO(MtNzwo`H8h!waiR5urz2J8zOdb4FGK}ZS!*6{}
zt)}8&nN5WIMYBQP)IYCNgzgYmB*+_R_Ne2<Lye2_;Xr~LEt=s4O&VKzuT!HdU_1QZ
zB___Tse{%1Jn=Vz5q(P(c(w3fvi&<lVLUp4KB4jY-1T?HCMr(8gWLFPkX}N1S-g^a
zWh>0gEWLYGx)&8AH46PI8nOpx3`T;BzXoS_uQ*Lv1z$znS@J#LnMjvkQHHMD-MQK5
z2Z8^TM66k>$GT9u)k6Ao1&+dt`s2#PX$4NkCoh`nAxAEXJ&O%Q@p+r8?h!u=8bID7
zKbk4SV8~a_6xd1<eu6pA5)rp<swNK1C~_b*0-30OK!%w)H^xZ-g__xi7=rXxSW#SX
zF(jz1dkx7^2<ihd^(N&9Sw=NIji-rK_ey?<?2@1Xlwq2s5O%h3<xO<2a6TbaOwA>x
ze(PN}Cu<hn)A2ahRwU>TxjY>16oq6?c{Y-cO*eOZb>r*r^2xZuj>G?B>Ad5qe*gdf
zI*z@LS%;1-yX>+Z$;{p%;mA&oQATo5NT*}Zj3b+@GDFDTvRBB#u}jGI{=PoH+waf%
zqd!iZ=k>g<$K!r~xbclX`fdb17NN^^G!&>*-RvXtwB14!#A^lYxMw0(T2Sq&j=wvM
zNL~e~J_lKaZEYv1u-`RJu|81;Lv8^_`@hVGC9IW<c*{4=1v||3?9(ido?W#BZnkY^
zkV|DYa-0-Wp#KmqLpH%*tI%?{3pH;U_(ZlICWaJk$MI<H5=_n)es76A+5e41q{OFx
z<9BCz9wTPy-lD;itwEq_{>}UdL43xF)f$MODxv)Mh<?Gwway#zzFM3Zv;2+c(dpSA
z48wiKu+vcrQC&lXKc1BHV!Av)_}+=hILF^mYdsoH;nSDW#R$T&X0u=7j8P;ol@qmp
z*<x>aar~I52XlMY>@o0-PvRL9B$G8PG)o(S|83Xvi*^2&4WLOsvaL=cenU0&m*SE)
z#kQUL)N6_-f2LLE|No>aIeo*e6h7n9BMI(us{aA!vO$x*gFY~0<jAcP3{60o4x;vJ
z4$=Df0SEE>3`dookmQb97m3(~^4Ei!yO^W;q}a9tE9+(f82iF9ntMT&LF7rb@$H^=
zu5!C5{QCbUm5|C+trQ4=i^nOgga?@b<f$@1$F3fFF%F(_zCt<rBB<k1dH<1L>Q|r1
z1wH2sO}dA1Q#}saTA(C|F`35zo$)rR9LgR4b!_gQ*X?T(h9zBs)%g}h&)neVMsok-
z8(yeqS}?gdJ-;PLQoyDxq87!At4X5_@{<_{AJXUumFn7B1e0#fPGar-rQ#BH9#jpN
zHBm|5d%&wI&%W2^RdT`2K<2x-LxF+YefSEPa#__cN9L392axR9Ld<(GguU!iE(9>R
zg_Zw0k0;jLr<LE(VqI|k@$hyUYYL;#NP2duuU=GRj<iE&_k-I3g(Wv%p@!WOy7=jD
zuoWsk-H<-qRpb4rO)P?z9F|xMO*2UoDKOyogHs%Zk>F|QThrp)k8iyw&5|d46ABYS
zr_mSJspH+xcSMNZ0CP;CLmVdC7L)zUKxE~C*r|Se1FMmQv|)<Gl8agL6KWcj#TY~P
zd|of0eRt@N9tkBW?#MM#Mqt@njb*uI>n#-iL<#k$*jh!h3))B(p^PMqeRTCM!d-4z
z3rmLS#+e;%t8w44k?$7qG(j;j;nUvsDN~l?<J~1g61FrBjDQ3J;>-lio>Is^H8wZ+
zLr;P*o~`VDD=+g-$QP?Gxr#AHfR_4E!E9zQ{iK*9gEnxWD>)h}5+AD}wtx*@kF>p`
z10SYn0fIGZj&FSvwMrsGCGu>ZnAg@czD(a+qBuc8v`(hx;kR^{zRw0q{#L1Y!Ty;=
zs11O*Ti&A4;CaK+2G-X!0G8Kb6XOg^k~(ejg{eL&<`)|zFwoWYThNR%x^-gW<COAz
z&h5oduB;z!f>1JeeuC=?Tl}O}BR-h*gAq2NO7r0EIp5i|@`_rZl48MI$5;e?(SB|W
zk?l0bd&GJ@h3KgT&Bh;=Kme5Nhg;MVL5?}Npb4qb{20CccoS%tJa}J=o+`lX#CE50
z74S%61BT2p9Slz<J5wANA{xNwp8I9-G26HvR{cKa-#D~&-6{)*+ZW!s;t^6%?!yP?
zoBIOHg`iJN8#sEVFf5$p20#HDccz~VrK?=GO_77n1WF!kb2~4?ivJaQ^(pyo--CS8
zlK0BexIcV7+Yk-Xsno@`6qn5$o@6y?W(CR*ao@J-dPc!bi@V5~%Jxb|YEOI#RUo~$
zt+X0@7g{*MHB}fhdCN>1%M|RJL9Ub9z@7QJhp2P2-=^4%Z3LfB7Er$4e7QPQr$q4b
zxF<w-bEh-xRn`W>l+I037ifVF-6!KGby)DR#0^MTK&?i5Ic)01K(h{B4uM+_Z4E<g
za~{WMcX^^VI;C9+P{;tBjAin5#GzFJpE%`^LZ};=%n8!Ek`ssdlAqM+Mc#BG6UV~y
z8X=C}jcvaTEd)tvg?9DXTcU_uFUlV<vD}#FPc*~w`z<om-SWRR5K8?Ny8R!pRnf#c
zR!5mPe8q45>BUqkA9T>9=ud_6biba&oc89@e*VxVBeYA-M@pFEokQDmmOpgT_f;ef
zltx`K*wO1?`t&+9-oOYPKX=h1wTQ$xO(DpgeIy)D)*UqdwjBh+gt&I=1LAfygZ{mJ
zG62RoJOmUVy!3|_Sna^dd2;D}Ghn=F-NeHJ#3-e=Jn6Xh_L5BGoy1la+UEg<M3%N!
zzS`g8Xt`q~LaBj(?jV!8=ius<t)c#6=)*B!^u*?uOF@9neLJ@Cw!QR=?ZIY4ch3ur
z5C%B831MRa#Adi)2S52fO6_P`*EjLFBd3!mmUo$}r#E_%@S$da{0h88(XA#}(RIH?
zdBM(R^B#|d(e5wKhtd>$_NMs=!&lyY<|PDQjM~(Nl}ji-iz2-&JVaeREV<NweDd}`
zfIArVAD}P9*cafMbE5qzbi26Hkt;OT8dy^jApIsVwb$}}+MXUFST)40WNGzbm%U{*
zeYqc5d~2f5msCCBFPLvZ2KxoPzh0$rmArB`)Ymd~yGUs-iLi-eMQ00xiD~4vUCZ)X
z&znzL=ip|P*3nJz?9H=-pOR<Ov8(7XBm`6!ieJxJQKam-IQ*GYB-8d!{kSEt^5*{P
z$H%O;p`S}z3$zjY8_4CAmM`}V#nr0pR>-c*B&5D^-%%2j8@V|F8*a`F9>wgwVk>@X
zvOn_YJGm<J73;>pR3fz3fzuE-$~g^s6I#5CGe@NdUIrhy|Dpd6Z~;Kb@?kICNCy?c
zE;sG!K0B$!iLa#1{D4pATRzxGF=$DcSiJSlC8nHz<L0}{aal3+!hON#TGmalv1%8+
z=eKhEW8#8Va<QS-iuNV|ScxEH0MrvEew*6#!H@ZeNSwT>WaF`1shmf%hQH)Ryzbz@
zFN5^=efbN)zS@*oZ;PGm%f<!5K7MHsgqK&+&r{s;wydNUaT>zcDim8%%dwTAv5r+P
zCW6L>Jq6sTFWJJ<*fD#V%wQAtNE7xTAJD&fSc2kn#}*6ZyO!UE0gestz1E5*oa9a!
zc3Ep#OR_9;Gi|X!&KWjaA@BV&J@4tLB$4*^ZR-QM-$^7ym^w45fiI3P%aYME;XWRf
zu~JH1C^LR&Q-v!}VlgGApTIxPX>V0xe;wgi9N!|p8f)uc5M9mvMv49fXASQ|Za02v
zluv&T@jk1-tzV>rDB~Q0Qp|G_Hh&tH)B7Ncu}<#bZ2YlpET3B=5KOI4$2OO>-;{C(
z+gpl{NLhPtEY5SPR~cJwZE+LZMhZ>*ol$#(Km%|4S9(u?KA$1j1QEFmu}~*bAkjV|
z$;@{`<(A&#z?~j<@x*4v4*`M@-1TzF3CXrzRXt;S_$aCpsTfo7@6_wUq-u}1o3Y2}
zr=e`9pOFm#^O!W#R~?xyK6^FqWU>?eQ)};C$nVZG3(|P0UZeM)6j3d{`W!Md3^ZgM
z8@AOVk{ce<Mxx!yB|esLQbK`tr$gh{xSrYv2x)yGhVM>-7+{DN=H<k=CDM^!7Tak)
zH=EAmS|Vhv3rK57P~Sstp*3D3r{hGQF2c{mWg-JV@Cs2Kqrepjr;!!)`;pqAu^~g{
z{*%}eJi`TKEI!NVNgAvK@5JT#_AMiMT<l)Sl^@k@r|y{)$*UiI7jH<IcgDWyP19?A
zy!e=%636!h#df5WGZi-T0kLo9pZ4;nirVH7clEky?n5iYQ6a7U6w|_XWGCiQzKFgB
z_>}$;k&g%(2T>nSc-mg{{>$TRC_y0<_6@ti7s7i0OAx9+Tav5A^aFs%b|*B#Py$aD
zRu6LAjjt)#zmTtJ$TYtTFI{x0P!`pteS<J{)+m^h7s=Z;iG=T-Sa_2D)cgg?xSf0x
zb96m);@>yoDF+*ukMrHgVb-Kt>zV(CWb;Q%gs~pFF)(ftpMcU>rC5QP3HxIil54V7
z2hFPl((FkkpOj~}h0^B{*=hNt-yU|m_~3Dv@sFmGBGJYnzQ}8GVcP16`jn3%dexxu
zyV~SK(o9Tz!-~&~Z029KuMRNIWB?6$SXi&w2daFYSl;_Zf&~`j!Jinu*}f`jf!<Oh
zeZKi-+mIHrBSb<)^fKXTtd1=ZyED`@lq7oBz?Gzf0l&K%G~u~;7%7J;SvI*lw4-Q5
zYL8g<aZAvs^#MIUT0UhZN>wWfQcwqUPl2@K1UmeM#D&{fhe*luVKkP2FU&4{%AU<O
z`K?d!)<}PByCBxm{kEeJo0WtT*7BZHUKIFc7MSZTRTVpt@vGw9b)r~M?VYFE_O|;O
z-j8+ZXA(*=J{pP$=DCRbBKwC0ztvC;d*}z5Fn4mYuf947!K>`#x99AT%SO;vPB;9)
zuq&jas9f-6b`quOwk{pOL(D$xKvD|3Lg*<=7!uhJN5H|P)?RoH9Vv&>IvwT7ZAj`d
z(4XRc#|wBdxS7I35b~wn1?MQiq=Q9%cs}&~-5Dbi0>}CI+ct1xqmRsOlB-R;oTDv!
zDR!7ctIhvsGD2L9Q#AG@{^=Ks+OP1-Lq(lqQ!s>E81RXp#i9T7L=$`*yP3$`p7?BC
zQjcx+;$FSfiI)zGQ@cp(uS`=}k1%xJh7>VfFnW}?va^WK;EURYDc71^NRD02C~Z-3
z7gX_al;EY*A!niNm4E1p;6*q{Jlrmlp<I0_o49fV0XuaUoiY2>cQ^$2s!;0bl<Cjz
zJ->Q<^>yoZ&Omjx;a;w40T6&K|K|4ie7NVRoQqDKcoN<%CH2kOu#Vg<Y)fHnWaiLg
z0J~huAliNR8UVrp!0li|DFZMb)(94lOcIPdTqalbE&{(*MFS>cPc+1p+qjA=d1Xdv
zroM7Pc|uui7%xPI&W!k`qQpeIl}v?GJ5%>UiMfvlV$`J>oG^-UqCdq5e?Q$KT<z$p
z>caWbPjDxHwz#k4qc-gx#ios74q=g!FTcLd_f57vw|uVhCpU~RE#BJQHL)7f9J*n?
z{d=ka`}aQ}8ldx$<!a}I5v_CR_Xru?$J$A_f>F+8oLltZAI!P<yZ5bfm#d+vs$+FZ
z&SCyfZ>ul){Ex!d?<#4@Be!YQyZP2jWm0_X3km@%mooy-%GPko09NMnu_{D&-Ry(}
z>?cxGKC1TlxZP%tYBY=_136?&;+%G=U_`2Q$ZBoDna@srU-xb=Yal;?vv$_)`N|Ee
zQFPct?Zph)j~r4Wy>`io3OdN+BHCE+tczZZ(Q{->E@a*{k`Y`M{-nHR)y?V7reb$m
z!j@IMKocxqD7c3epO_{k`ab!RIALkd;_&djVi4}zVMNBlItvz7E}2j2JFHxS=jHzb
z3M98a1S(Mj8LG+XrE>q2ey6vb5<6^~-T^E>n8e__D~>yDPPEVBXoJRI&7;`FZoX5(
zu--+OTZNNPJ~e;-Nm^T_iDJ#XU`off=P7&Gc;JFtK5tFx@wDh*m+9vsk`kv#DE~p9
zKo%z5;8!o9K-foaS5Li-sp40HMH<w<XDW=Ia`g{Y$tYv@9JBhDn|z`P^f0f(oZCD0
zF=MNe=CAC6MkzVY%$ekMNEv=AU`+??9gwQAx&XwJR<Z}J+usw>gW%G&mnrHmRb}4R
zZ}$TD>hz^F1J%f>09mUq!qTVryPY04a23$G$`~pIS1{h*Be%lvy=w&>tN!Pdgv$%0
zmo52rZ$)5U&@)pQB!~#chVn1koT!N|`cV)}kgxtx&(9t?nMff>BZo@ztlPEs^lxz%
zItKsFg1}^}Wds$Xb-_E7KH1fx+YwClXWT=w36fwMt!>U-aZ||;g2V+63qwj@$VE=>
z&t~qj><D!9x2E18YDam$%bA!6?^9jNQ;NFG)BVT#K>#O5u;nd{b<iCwl^O(&@#WF_
z3yLwhQW856GTBc$BUWDvHq{@pP%qo3gXb(;3C?Iq0~;$#3g;EY>LI8AV6qZtB~&O;
zXprr6J3EgBs?_uFbw{x_3@WQ_!Mw0hc*rULuZTXdP4?ZLdxY4~a*<_Dk%#IB@uTd<
zhpe=ys9HsN#J>_iIYgLTAK*h^?b*xaPYuh>v{Smh#>N;vv&wG?!yywc6YF0cy(bpl
zu%vs-9my748_oMBlAg{T2f?Ba^LB;k$8jNBNJBZ!^e&l!UcLfA_(SF6?u_w&T<^JW
z5MMZ8bFAN27&(#W0v$fD(M_tlG>yNEaUo<9P_oX3PT0xBJCwI32O;{!o+}M_CHoXU
z_@PjR7~7x2e@;Hxtq9R3@1h;A4!q&+2dyd-q|7zV0L+l|#O}StiD~D!WfWUJuj}%B
zf`%T&AkYiC1FkD-m0PB)-3@fV+qa=;0@?H&Y~O>soPR8o<DlTOCg-c7Ns)V8pmCt*
zokxZHRs&Js?Q9kzJu|W;QV5R4(G9p%2vezc47;*()Pvx!f?G|1H^nn+;O-kE22%uZ
zSCZHQ)u0I<^q~UGgaxK|Ldc|@4rOl<qaoiCg1Zv;E_&u`rPx4E^0Etjz4mbEX|5=1
z_seE)wQ|Q!X|pVU$~`^N=s0?vZJbARL$XtY^Ipi-P)T$-&etR_VljOx?Xu0dq0Uw2
zevC>@i>nOBSR!YG4DAa$G?KH~G3;F?PG}okV_|v_ooO}}ps^>2LNL1qU>xT+;i??N
z(4JA>lK@#IQ59t321!89dj+wle2Z~BZC7S4q~k);n=z6|9_b-*YYplkZ@0oQx@B+B
z*xC*61qA@O%nYRPVIBhg|FDz?I4x_yG4@0DPmvqrSMMY@fU6q#)XBr86A|;&?^6w1
zZ<4P3-|ox2{OShgIYLG9x?LzW>4T5UGR(^8i#)`iq#J}XBBiH}HZ+Fcg<ey+pw(K-
zSg9!VtNl}6<kL?)1lbp)wkJtwH*iyWWOHf>TqTY2G`nz>%XzoBvdvw{VQVK>^YGYP
zt{3WD`-1hGSD(WW^{*{M-+qe9V?Kr2RS;=-Hs8oVqL!okH>LXnr52jlC-+G|!eEQ?
zs%=`ENKP3A$ilSFGF)kiQFQrgPvBWk(NQ6V;XcWB-=E*f+0Jp)2U3z?jvq;gp+S*g
z5DLt@c0<3w$ENR!MX)pb*8W6x>6fq0m4qcYtTnjG11-qq^6MawKG}&2RYK=J@hb3+
zs%|c4err(k1Dj{kw%=ZLKfm(l6S*fT=0t>ahbkPqMoZk+(vLVbYDa{2PIo7|<}7|V
zv;6QP7qY2J$54?P$V2pNxD>5Su2Tnb3;tC36gbzmpC^SAcP*P1LY1aJtKQOfNq?!X
z=kX&h&{flet0NWWQ90Wm1@y9F_Fycd@1qQ>0xke=vNqKS*j0;Xz%tieeTR0YdcJzy
z6uCVUwGH>dhv3Aw5ZlCc<#!nrv8MG<`vXFV1Ffk@5?OVgHauJ3mWfwudh1-zn)zM!
z9b6AfJ+pj%-_ejUpB3SB!b?+nF)XfW*Y<g7#q2+PzU>ct>BnNPFG{AGZuqy2=yvOE
z7QXu-wDL*l->2DX?)4;E6NZ*XJi-MWUWk=64Rv;QEe^M$B3UsfuK=*A=4&FIy01UX
zB>F_~eAjls$9F`z`cYC!)Gf>6OqU4l0F{rml6*hA<yVpfIbV-M`tj)OVAHS5Aa9m7
zA|i7#|7#N2S{;rO%A3K=fBQx<e?<Eb@AF*??9A?ez+L|FWNSw%E+K-f4zsVO&;%di
zb_j)<Sp{mo|4?R!zgInzl_q<(NbJdnb<cwg@<(qdZUm6z-2BEBBn%w=D{}O*a3ck(
zP0Y?XlewGTE)B#O^^Ci*zD{hcv95kDef}kqJ3fQ)0f}1Q4pn(xby9TVtAF&RbP5bk
z@0&Y~Eb$y7y*FLMVgpSAiEf+Px?+lb_R`0&0upugJk2*+6mmmC_AOe1Em}2SWc~J{
z(wt@+S>$@$-=aFa8;;5o8{b*5bar{`fa)NbdzVO+8Mab}`CuK5L#ggh1{VioHL`)6
z4|GyY@Nbr3m7wi*>chkuYZdZhvuh?ELqV8pOrIf*>rqxNo+}c;(EA>ikgTVAPFq%M
zVTHlQP*ZS3O{AVDKx@LdF#KWJpQB4^%<#_ajc*dKYD-A5TcuXnWXO8G-><{<TNn9A
zKyl)OS;Mo{x1qDPIa8T+j5$XmV5+~6@V5wUz51h_1;yB1cF6L#T~h}L9usxE*8G~I
z^nCsYx&GaehPYQ-3zf}LTm=B9erMEi>J6=YkJyRvlwrp3ep}q0FOr!IiUBw$wg91g
zMY3BG)VfHP*ZWAs^>@p&LHy=ChC#$A!TOdbvL&^*qQVtm-?K%p`SC0}CqWL6i=V)t
zG7U!;bn5U1B@?U}CLgNpi**i@q(`or>-2IHVu};~Nrz|kR9gF1T%Y@j2Zz_KJOP7%
zo?18fm7RpQkh5oz)aNSxgH7{&K9qZ6&Po^fJhy;x@}b-z)2{14fZ+N!++6=;6n*}3
zb!XNYCEpy)DRW`LBtI>0yJ4@sr|FIaPG8mn!H!|n=$ThZ*d2MgSO<+ZhXq4BUp4Zy
z-n957mh)iSy5m01q}Nt)uZ_y$=(v&i!~r8{T||T_uZT~p?HI&BI>BAo#Y`TaPv(EF
zG42rhf<dXia;!p&)JS|UO1W>#6~0dTHvn29%dC@FE7dcHL2xgq>-2Gy+V0wt;Yrz)
z>Pb>@3s*82J?6}u{?sD97zmxCp+QXp|1OXqFNb%iDJGU{@EB3Om{Zo55b5vj`yQvP
z-DPHfJ$~Y_(CwOF77veU$wg_WHec*|?~zP<Jb$(m(X}5J`F5rl6qNHm2)g6-D|=2;
z$VpQ~G|@;8ODRvQHd5~Hp}%DqM-Ca159l36PeoLS8A<v0xGggTAWUW7399328oS8x
zfd!<vhA!xJl?nPxBdL2$=mThj3vOZ@oYAn#APMOqfWvi50W7$Uv~V=8V_4AIOF@_$
zi5Yeilm@;-R2}i#$Q}Yt2=5p}g=ZK~yc7<4u^z8&dk}Lkptlg}yCGKqQprM7gODJd
zjGiuHQ&VSiVhrep5e$C*Jplc(t>qeGxt0T7CXbV&{$9GJS)M$cyiiQblG^)u0;12j
zjEyezxHjfI&J?L#aL00*+Y=_8T~-oX&EwEHwR3~mrpolE{{f$;sT9(|*dA@Cf*2Cq
zi%Rd^e)+(&y{OV>FI0E=E&Q-BjJ8MFjZhAN@T|ST7suC%xLRVp&tiU@L_e$<Fkwkz
zdbv0OiC7LeKFMCdJuj2IC>y`MG*H_Rsy#DG2BYj*J9)Kq`FZ3tqnt3G!`CsyAm?Tv
z1Ra!&m<SQAKS6uGSbRi)AhDqec-ULt_jMM8iU;h!T1o)UELU*J@;)zYDY|#V@0L>N
zIO&zS813WeVT#BfRBP+DmZ^!G`m_f52ut&2*eM%f<l?~HQTah3VS_H<(ROM}gKwXk
z)m1Iu4bQv!OfOuZN0ol>pKGumevg6kMW1B%>`c*I{ZK6k+}O@=h3T9)cW%Cj38L~*
z71_Rh`X!U=bbYD6AZ(43R_#B4%0GKQC%yU<ZP)N~Q!@M6KAZMbc`t{<_Q|wjq0l3X
zL1O;E#bz#t;JcN-N;b5Y(CB$pb!)H`Ui;Z=1$`EiLgRZ&{$CmTmH#^hRP__`Q0gqS
zk$I1EDi)0-S<OHznLwykauaa|-a^P>Db&}8Y@dBruWH$hr<(YxyKZ%In`(C%^U+I(
zO?~vmU4YQ@@|>Ws`8B2GG0%p){69zE1|F@39K>gAING8RQ%nlWHMhGZk4nH6rkKC7
zn<Slgz#Nj8Q$RN!6CZ6j73Wxh!D_j`J)5Y`#UT;SC4=*3Ul2m8W7S$TWwsyQ5npL<
zh{W<gubA^t$2~J{3hm7oo-ZzwzP>sF1rz?_t(+5W=A2O!6JvGP!_G`0ibPpQs>LO-
zH2ED6RpdO$(ajG3thig+y*bYhz;w$Je&o6_CbQ=7=ji^3ylK8lODyzoP1F7oP6cn{
zMjNL^h?0}CdnTC#q}%R5*?yaJ(FcVadI_`oIKA(AoGR<Uf@vhGw3OufGgvoi+l=63
zRKOdpU~iR4FP%NfhdN1hte#%b@C0N>>(Epz?_c4pMMR~2o{WH~l$OgJ2iTk*EQ;Yz
z;ejjOBSdn5b^WH8PGK><F5H?Okwahp`AGB!Y8a3$ykNndGNyrIZ4S$1#2)`P7Fp^O
zevgk`JMV(&_8T|JntrZ+#w7E~hNSFrEbweh_pZZ_EcA)5&HcXkpNh}kUo3rq`#}*C
z@1%NCl6DsE5X;@w#by^Y*>xV;xZj4L!iFp-)*qD1|76J1RB>o3_nnX}^t}<OwgAWB
zVg1=TcdgeI+O3=tr)92e;QWqp!}S{D8d>-ZgC}V&8={Y&uN6K}jI^huXl1pVIQ7+s
zJ)OrK4m~9M1T&~%0{XbA$&tr(y}EDJC9YUbV>k8AH99e;*9MhW6+;a`iiG)U)N$e`
z)R^p5VX~*ge?ZEIHNLALNiaf59g|0mhdCJF9W=P9h2X|Wa<_sBe|zB=!pqTD8~kJI
znwuEq4ajrO9XBKtBTTz~VdJrGeHu|!!NPgc?JnSP-(1N#-vn6Qx82#ikHO7$2?<tO
zM}jeCAG9v8N*yWLp)B071p0WKu=|^O@*~zrilthM4{-+L5oXcBELzM?)PALuTzZ!t
zr8f}fFZgS)F8quZ`A_qThiH30Sycmvh*)<YLcJWGq5<A5*(Iq75Ly6Tj7tKY5^Im?
z5D!Ufzp99b{BW=l^o6{tc)5^~miWk68a11d6Jgs}=Z*nGBC?OC-tk34iEUPz!S}dG
z%I_C#i0qm_z9V4M%8RCr?YOXJOXar==9AZ)iOJ8G-NW<>)L+OxIO+JTr5zfNg{4ZB
zHkqNuKBgdFOFJF6<wJp#SFk>=R_iQdW{4Jm&H$)-x6oQN(+%L<5(^PVM@+?0CC<eq
zW*U(=!M>`phxV6B&EhGYk_Z@`;`<u7(?;be2Qw%~vX9QD8BNV!Pd)n=JHjdN7?I4$
z1EzCqH&500CWr;1siQ9~K4HS*6r3hrP&H#<B&A|uGDL>0s?U<YCSKJVc4h=`o*fO|
z^!_XvuM$~jbDfUg`SuQD#Hm?~1T2WYH>UYTQFy3=^N*Wq@!LZ}EbKg;3<}^$WYlO+
z%iX)1UW<j+u;?J4G(KW_6G>sMQUl;uiRByy8UXh(3#T@L1$aS-fyV%@@+?fs8w_Ck
z>?CX(T+fMPiN$zFmo2vm5@!6Fk*g5+|7tqU;R^XAiIf4x`!){?+QIo{nG<9L(?u!)
zu4hK1Vhh<vq=RK*dIOE-VcerAn_mjzWCm3y;d_WRFTRCVV~eYsu#@B&t}`#K4fj}%
zy}*6`44+ZH870}fIlsyTlwW4Grn2^Ll8HBKeEfVAF0Xb$g{OG=;C$|x=V{JZgNcMN
z{zjh+i7Pa?{h+-5?X8D6=o*Y*7q((^vt`ulmqS8X%d3xJ2=Oqlgr!cSbi=bwA?5c$
z0!<3-AAdF~C5Z@;gj^FB)UJtz^WN;OP+id&KHv7R*JOOah(s6>ODc6vf*JKmbYK@X
zX@m^`By1swox;sa4AE!Nc_fW5kMU5p?FS1+r19?u^EeD7?S46Dn<qEh+l#qNZgxFj
z>wBq}$nKAQcs`bRN_~*htzdeOHOSKP&p)X}ep-r!w{KPBibi1%8lI+nQO)-Z)%{aK
zB{mp1z$BvS8W1$U8+3$KR!i48qxYD5PYY8$$v(|!S7wd<TS3!8343tzltv)dSJW@T
zP8IUV>Ev9d<1zw-CI@o8;TqU~5)bA{x;$oV!e7;jxny5xacxfB8cmzA6uh?F-ycZI
zcvSd$_)E{${W(u!r0bth!~Jp?d(KIUb<UY*Ykxtw6k$WsVnq-)`M>eyE^R8VWZ{Ww
zwe~*lf`Z<KzEvX<=9uz;dHdwQnhfJMZ+FnE7mdRq?ow?T+ggZvW?XlMfwH9DeNS{o
z#U5~6NLeC^2_8chx)I)LK1?&~9TW8RpnPu*fg{k?5|ym4alPd<+LS}uV0oW55iZXH
z7R7oBc-lueM6Bq34+(fnnzVVKjT1*eJaM>YAGd`-)AL)NHtQX#{lZ<3US~(KjQM73
zO{M+DWlnoOms@LyL0}Zvy$yfAoS5aq0P$mSkcn|5X+)c9vVih|9tFbQuK<X1qkOyU
zcvobPqC(^*+xO#Pk=L_h54+ZSYU~mIM`cLTj{3zmWzmm{h|fU@UP{+YW>@1EBki&@
zL)_Ei8KVQUU`l4$mQRp!!}{7i60|-f#PM40kexR0A1~w;?Y*`&`|<Q$SaO%|w}+PY
zYMis5z-tG@H9hi&{3ftRla8WIs}7&{jAKnR`lKycZ1p`%z_;ev?Z%*qvP#>X`CH9;
z(wo>xU#8`#(5LfvUjCHV9{H1jp7iUZWU?7zJEBmXD$&lq=hraCZ?Ey|9wkBb(!DCY
zuL4_b#~axO^@f+42qw9GX|RcqVdjsLKc7`Ht*|&sy4uvC_zZ3XCh>HvOM@nV`#ek4
zdIhIGO8suM^t+t<52y{xnA9Hr-o-MeJ@)c<wuC$o5@gO{v7vTjUh_0T?chvY_Pyp)
zeKzQCCV-+v*%qZNnz_jWo(<S~So1#G+$fv0;QDHqUcJOoLc7zv2Q0&Med=2TI{lB2
zt4c8f`pZOMy-cUr**}HS>1Ovxx;It%rzbuW=q+jc0U7Vu`(WdGqILGiU1_B+T>s`{
z`{Q-4F~JG&_pDk~T{>f7Z&)xw;M@XO-h>}<n=*7rfz=VQkdlIeXJ2ze7x8Lq(jlie
zI3r?$I^WF;PK;hd04&(8%rB558QQlM1n4Rq&!JjIj7~)w>^3XX(DK=a5c%Mx>4TK@
zpZN_9Zq(Qvg!q~*S%2UQT0CRK;>TcilOOL*v!xA<GyN*uI9!9W<z@bah_GoJkF+Q*
z=3)c7#Jwm$m=@!yDQp@f#Per_+hxX1W!y|^BM)f=Tv(Ngb>F{1XCF<{;Y6Ae=8KK2
zKIaq#MXs1R^VP4Zy}gFyTOT(3Wm*aR!;_s_*&h{3cAx+wfa}(HWr0ou0lyMQw=lDz
zFRuQ+vBwliGrX`2nai*ey7_@`Tkrm3nS6`{olVSw02RQDWU8j7<Y;0xzRsb82r_|u
zb15=iOwdV@OyxErGpy0^Ub_{)+IHSiU=yoPS6j)EhEcgMXRJ+AT*mYUClRAa?JGx-
zAc<sb(miFo4Y}eo)Fyd9I4fA9yh)piX^}fKwfa@1dg%wbaZP@M3?iikQnXY#Pgurg
z{&5}xJxTNtVGHACHBoDzK9h=M7m;yKZOfHJ{-(LI=BrxGl@lSzE5`*e++i=vo#I*Z
zAHaJS%2=kugee7pU?au%DQhVEXsMxYGV4g~6s1PUh}yPx%^}l+Y#)3Gg>He*THnw5
z3U38db)K+`ve$`hbLx(DN^a1F`K}629w!J3*RRLc8<F1)7v)xnFZ$)cSV))jH<FUW
z3*qy(_(^8Leqk*GA(?xiqP3fhjK{}~N)Q2ZUABbSE?nU#4%Z$TAP=Yb0~4|VfmL_0
z{XQKf*;6i&P#h&;91=9$Mn+)ABnUE*x~)>!OlRye?Ku3~;X<}AlDiFj{x{CnC*u}K
zLLA^+A(a0LS1`0eh*1NO4R<znxDrRdY?9N2?7a~&h0m!Fk|?Y8P;a>QwF{e1U`m@g
zm#BzXkr#YWrTn^+t9O#Gzqp+5b-Z`Hgpz{GQJ1W5rtta)X=RW0q8E{#t;_01vO@s|
zKg%xyLp*0YIh~cmWH8nH60d&iJ6saE@D;S!BPCO$M279T5!o)E>fBVC++e@-$@J$3
z@C13*uTl)5BBTLp)#n0V;2;#-+#67ttu&o3zZ(B&kSDlCF9bE{n%w>M<h%)TDcb+i
zGVtFLgY7qi2}Lr%38%&NCYT~}SEUR8@Vh3w*cpU4hpjK(PU>6n=!k#gN!`8KEd&YZ
z`ril2rkt_<mlUy=thH^onn#^CN}a+tawRaRzl-IIt4`r<@Ch7t%YDzb`)ImDuD!K2
z74ZplGxW*K;Fc^B`+of9hblTeTeIVoZg-E*;tF$+KKZzm%}Y!C115Nn5+_F64piSt
z?M-v*CDzerr_1&n<wCYvVLcZnV7BtnW3DVe>HTNocrXJIDE>PBaaSGk@Ze$Dv@Hmd
zbb;<Ote<sh8Q<WVYV>e2^`~y7Y9>E(P8Zw0;uE|GU>v8V*fb63-21UJKXqTV;DIn(
zm7ndIc@4rsYy;O74N8`?0*XO)YHkq(=SKVDS&-y5k`n#!>%i`~i+r~w3!$N)y8o|a
zkNcD72)%oXIN_-0@-h%;y>W#3y0_M=Bs$(QM(Zn%h@l3zAkyEad{(y%zVeK-2620k
z`|lV;jDazHP71_&7?MWcVZ$8^>nN^g7>HQWqPDy46El5Hx}y<C?Zj<ct>|YK>j(XG
zNnkT%s;013An&u}Y)ELbtmXRV*w-JueH8E!nHBlA5isO~s`(ofb8jA}MV<j^1JvY(
z1XaGL(O`qmOe4XPQ$DYUE!&-CVx7ptPj_`cwdJ{lQ&`tvS9jbsU-$eaF0o2!>b~*K
zDfhdDd#=Mt{Y$5pH0V`LgNu=&%6i)C*|0uq!w>u%vruJW+BnaJSw61-@;pHbOiQp4
z$+b>Q*u%4|D(q=L`HkPf7N*uWOoP(%;EG74i6{z+LXR9&0gu}6krg+%6OJBMz+Lt9
zic@>{!5Zo}XK1owv>O)VThFD-1nHwY1dP%?=}^OQCYq)@>z@ja4t3E!hUmmZi?mFg
z&|0tkgBnli(U@QlOKyg^dlHJbEG6kF3(%$R(gN&8d7~tvvhnW{Rue+>^|Jbns~|XK
zw{{6bWs0Ei4lZ$d_;E#Q2V+ZKO#+@u3IpnAsV&*Q?UNXc$7f6Gk(TsWr|^AFHK>nd
zvVU)hI9yLz6`c{qRZGY=CWwEZkK;Tk5tbS?{vQ#GjECEw+XzZqOb_Y%$MWIlw=Kfu
z7OAk7SH-LFFlzcdRXbPs9{>T-M1m?|yDb*}XS9p&`U+3SRnOAsm2G$N(7uMmOQ&<S
zUHDaDW?fu3jj)em<H!M?Nf3a5*>n-O{TN4Bl2lx9%t0u_JX|~ncbC!hRul(Y(#L#T
zuAp@ncdp;#bD8w=Ao*E|QvqudaRLh(!unk91&z0qB5hxpb@cUEBPHq!)CJW?R@~#o
zViu_c8%3ESJzdyJR&m<hY!u(LbhpAs31tn@5)^${`ge;ZwqZ?gz&qw4atH#2O`Iwu
zxV5jRQAc;kvw_GrC87+*?eqq}y;-q(g)|a2M`0Um+(5=rLK*omQZjy9A15tdAsqHx
z=(`{2e&|Z_Hk5gd{98<c$SCs-C>fA23fW*6>GAb{u4{eID)i+ak;%G*uKFeICthJ%
zKUrv}jH-I7Swsj$x(?ZrKk6BRgoe0T0zcEgQE34yVrFY$sA&NV=R++MX+0^Yi#(O2
z*D}@ZV%jTlU$GbncoU~|U_fgK1!+-Ib&?5g)Iz^g1i`Yj*W`%2%*ryYAD%KBJ+r@=
zQFdzv%61yPAq4q2xBl|<{hHSF!epA*uLv?`qO#RreaQrS`i9a}OGgQFd6l0VB#HqG
z@(K6YM$}@POzLR-iX39-SrRA+=99Dudycv;*r0Km*rez%aP8i%2&=d)J3^(LgeLo&
zMDlZ3)ZHP(P*uO}Wm`CmH=mU8QaI~yF(Q?lSj-`U)D*gNKTo-2wzo%@&Zs7D)W`&&
z1WWvE;F{%&^)KLT>|p^ok@-FD<fOQtZa-UuAmu}3;07wl3qJZZ3RbS^YL$~P?5ly+
zo;+1?-N2QwNJGMd9>b}y+?L_Me7wDKLjfpWH68mfhGFoQlfmyVVi*p_l4!@^u(3?-
zc03z2rd)hcu$G*e`(`9tPzkB!la+e*sPN?`TWLN%;!^IFzxIL0EdvC0vy_;EGs%gd
zjh!$<Vq>j6JC9T$dIEnY+%AlD`$IY1>*B($xVM=EZ=j4`HJk8P?n}TvZ?^t!yc8Yj
zJ1UHn=2JBW>$76ky{HUfmdh!Mzp}eRD^}{Ugjm%A5+%oNs!;E|-(o`}1iOA9)v&|Q
zRa-R~u=v66{DW`rp48RDN`XfHMFX3}^*f;o<PU#5rQ7@B^0b{v_il&+h|2i)GI|_v
zFb)b4j^2M{jq)x1sjGF1XT2l>M+eCH4!eTd*t}e=_3bIf>`1>W|6Sx0oQTzk*#5=h
zi?{h~EBxkM!;QF)Y1Gmg|I1=H(wu&oOb+D%gDgd(^od>4a3#rCgv7P@)9LqE6yI1r
zeb2AbYg5FKRv%iIEsY=x`TVxOs)!+Cz{gc4H0LB&vb{LVC}-1h6YA-&V56T-16MAx
z1*M>Ov0@eR6f}u)D-E=v*_utEp(C<xsvuIcw4<bpCp(xxgq62NeC!wjm5Ib4dMPW^
zLFxT3wI`8{U#fg3B)QVKei`h?pfqdjR_<S!)4C2|o=`k4w>1d=U-VE+5Ps%gGF}uE
zh9k*V_y-yl5hmG<cvAPSedtP`8EPGEa3fI}7ud4-8vQ8h2D}qj^Sz_hn*81lnVD}&
zAKvwy>j?Uh^Bam}=VqUUjJG#T7v|n8D@@L`lT<0_q~u!z0W5wiaa7rE(p>^j9hI8z
zXp=S+6L>zpF&&c-pIZ8e#>3Kg;*Uyg$H0NMP2+lfKGs9BAb6EHmeGIL<!oAQP;!P|
z6IRRpoa;raHK0CXH-UGFbHlWJ%W^~GC}e=!K_-6<Whl*W8u<8HI81o)_Dj-7RkIp9
z>9FtjT64FNjmWV|C3Lj-YoXE;u)bk2K2@_}VJAjBc$m06QLan2eG!Dt!N&#z;QtDU
z6e69Fr4|6<qt40~!*8hk5RT<+Hv7CXDF1X&TFJA4(pMioABLS4OHQA@?kC{gDq$L9
z(k3CtdXs#w0m>r0i#6i8L=Fdy5xB{pHlO+9p@0L1LwM)Smt*uMl6Dz}o*&E5^6&X3
z{3fgsK0n@~Y5uqly_go=PtpO;7Y@t*4;W4)d8EK2yCd1FIY#(9^!z%bgH>3sQi!6N
z1nSdoC?m_v$e$cji&bO~P9F$@0v2Uj43Yyk=-KOea8)5w1KDMNbH0BXL#gRx`r{q1
z8&8aR3GP)Ak+0{7WvET9q?W(0YJfj@Ka$iKl1>vt+LuZ(utIm$+Us16Z<ZD=@{aIi
z^FFIj#w0h$^~GlyKpucg+<!n#z{OW1MW%e$9JKkX2$n4*iz!fmC#^8-zV;z?$ar2R
zgRHUps7pJOK+>51yfe<p6hoZ)B4&&UT2K1x?GEn|4Ixv&Lo@v@+{(L1;GLQFF)`}f
z`c#P<&F&FgKg(?fUPiWU7<hc<qUtk;pHs?<=2%zy^fg7dUAReS#S7YID~ss=(rJTt
zadWIMbHy=*=!<-Px(d3$$h9pw3v;CzdsMz-u8$<H)6g&lp~uq3`z6@uL&!+DXq~ml
zTF^s}wZmgmF+z~f--c-b)hwgzwfd)~*-N*$d9J0JCJbJgg)XA76LnM9B)@5U)N}Nm
zY5;C!tOjxS8<wH^Zj1&hqdh8_5G0*tJ@F-W`8k_=yz?6$R+V%x@%E-qp2x72WvU_W
zP#p;{%WqJ4(UITn5u+H3qppfS&L2|j=g<e_-sr7H=-;NOCm_4i1C&hJ?VDH8NXC{n
z<Kx!y`@2f{(JP+x5AKp+SUn$3`;Ar0Zzwo^>YznKYJs-HLg=D~X?wfMk==vxA{G*%
zM@y(R(>`Ugn>AuX>pBDVUf<I-7Iq2zsB)PcgCt7HCg_Qld{J1ems}xld4El8+YR)%
z@^;l9l?7Wh0V-}putk;quxRI*hg&vxLv#g``Zv`AXI!%9Uj88h(tcP#9XP~Sz^v7F
zIHcq)BzrGH>ld<%f~DmPau48~Dfo>`0<wa~uo-}2@Q)!V?eS4zeE|SJ76AVq1zk)8
zyG3gAEo>vTJny(*F>y><a)mNH+nRE#%*ak+jGeL&9KHCy0}OOa_sIQ^=1nfCg9A^4
z`-xhH1w8NK<L8pAV<2&y`j%--26W(@Q4SEOtvR1+`0>myis*Aac6%yA&f|XHIN!NO
zb{@IPlllY8t@l@?ldRkr&A`Ih{uiorzo{2~S|$GN{iwl7D0e-1)V##H{rBkf`&A8P
z`M%_zua3;qfmi13cU5&=$!Y%;2vQWSA*mbYdpC!I1)|rVvUlI}u_gmqAL(lfgATI6
z6Wx-MUBRsQp-ku`7!EY~^QSx8@oKfAwLI`uEu?l@+>t<HqD4RsEHcLP?#m{A18&vt
z^R}p3q;(Y7^(uVQk8xD8s(zN1ty6d`zd8zSgvNkMojXb*(129K5!IHf$d<RT5tFw4
zScYLQ|MOYjIA?oRdQlmyuRgHWCs%23;?H0RQ;fg*?D(o-<--isEAD}{0Eak9PS2Ry
zwMTT?oNxYgbNzg5%Q%?~x%BvIw>Tl8WZKDeI(KyVIwC#IWwXl|U;y(pzlyq6H|u_c
z`rPCyFy1#?)QWtTQ;3>B9fQW5JbBh+)Z9QgMw4>AtyF*Z)s6N*9NvF(AAz@z{$XZ*
zixa$W{Fx6Q-h`;C4O!kZ$Xj7{zCn7FG~WEQJJqYGC^PTy)2#MDJ)dKoZ<lp!y=<Za
zU-c71xUA6T;ujSE<Kg^n!U*_-1jWr3baBIpxa8r07y}1Gy^OEHiOSe}fdyh3{+TF_
zFGh9pGPGr)Tm_c=I9H*RKD}wlZY{AkOSI<)K67Y+i*%62NgJQ6gBE9USffZx9IYv2
zf_YONf+xx#p36e3zDw!}6ok>#+fGQyaqNTQuI_-C<K64LWDcMChLt^T42poEz$4aa
zvfKgVpV*se2XWHc$PMdS&3iryrge!}BglNqZOKbv%mUNQc4C)e^^PmUi^C#61hXEL
zvq*I;!@!2GMMl8O?el5eUwPBs&5Y(^@W)3&UiJqa?qRu($sHiY8%!XU@c&-MP2R_F
zJJF-J)ejLUsNh^!JUOK0Dp|;xg2LOhx4pogp&GmvPSh=k%Z!KsjcyN2>f=WTu*ti-
zRs^V${S+(Xr2Ym1g@Xkdndfc2U>>WMy^cWl9v2C5`GE~ci{5T2Rok)Uh|vU9gb`1#
zk7S0&GGM@R2!4W6kmZn&O*uKGa8+H$8}IQrt%LkNKDl<mR%=qekey}n4dL?11|_35
ztEGOCe%&gz%}38a1ZBKQ6IAiBzY`pjaz`uh>C}BWnjxF<-*@pbCM;*``C0`8D9gfA
z`SRW;!OwpF9JwYZhdwzf6YKP2!!s;p9?w6h;%t-wa_=3&S(=R`KkFboQvHqh+_`G=
zGVfLL{IgS)Mv{9iyey3mF(7Eo(j*e~`O|l^@(%G-j^RSBlij5u$P!`t_6H!z0?-u(
zj57fFky@vd@THNOJ~k+CnDp=y{X~Tqrm+R`9@(xtg@^nmil(uK#E0xa5p{S}NRY|D
zbv$$~Bj6tkrGt{04+8~<dt5q79eQpQv?Dt{#9*(@6`ZjX_1$M;p*#D#fR37=(4|)H
zdbOA1%tE!?dZL6A_3~YA;^^~|V_`uNRnAHNca7+<@wrrY@DoBp?_(0Ag0n?nLRH*%
zAQwW6T(j>G>HLD@{@N*FI(=Tj>ln_uRV-fyy51Ag0%EpFe*L={+H89z%R8r;<M*^D
zH~%`zoF-8`LD3~6$)=-?4KU(~Tm;&Z7A7v^!zJGJD1i!4GUD1u6-#~??s^Qlz!5Pl
z*7~uG-!cRVNk{8Tq2kBK;wW99O4#GR;_O{utqahcc9%nBpX}-{i4rJ%K-P>#q>CUw
z*oyxTs17%)<_F2mq#r4hh_1awGNmmF<0RJ_je>|{We)<#eP-~~K6cksg_R~~|4J>p
zWS|6VYv#_=yq>LD17xJoh%pSGGZ#8g=J0L!S|3#PMdLA^f-AL{SSzKMiM?X!ln`2}
zRZ<OjD(gB^$B@b1pnQKQC;F@X*RgA7;F%Sb5Mnl+;pC5)`m~Mj2{+Vz!Hqr6f6nVV
z9Kdo$7sre>InO?%m7e+06c!N+PVg<cf)QPEq2oN2A{HR~I|e<m5*B0V1@@n+)-ez*
zT|m)W4*x#{V1$ylTmJGGU>Zq!SQ0fJ^Msr8@cncKIhP&CLdm5ZDc}L40u+YsstSOK
zvYYY-z{djZVTpVLC$M~a^*9I;=R!BVUu=7zhVA95WTnyGPxv8eqKG>j){PvFb)Cu>
zO8V^7Ud*FM2y14HovLzvQ(lxp0hg3lKLQcE5pO<xqlhbd`?O6->T^#2*AX)8l1C*@
ze@Zn=lI=h643MM8wWD*<<xLD5G1PS>y-mH#ZI4*bkO-6kR-ww$mmbqfBJ<66ixMXU
zRIN!z<o1QQ{=_Yb8!lRwxTcb9Yko;X?5X9H16sM^IUaV~-e|i}g4IPO^8$m)Sy_lH
zlylBGd`mbf>&MsnRgalxmb(fvONUb=p}O_~-4c|?GMPAG%+4>@?OknWVgx}d;A77-
zMXTzIvZ(>j7_F<VM3c5Z+v!U4C7B@76uXoz%QynQ!H`rI_jLtde{Fw$H_|f^e)BF@
z=lX{kTCTU_Z>m4nU0uA1T~<9iBB~b;X>%6**<<92#HRfR*hotpbBGo+Cic~4m|jro
zLq`8JW)6t<1~px~*z%yH8(L+6{*PYlj8X2=oHRcKEM@n}(bppfoepGTY9gI4B=1#i
z(cx%>Mt(iY>fE=nX+CJs9(798hVg#O*|km{Jmpt<jazVunfl|ix3Ao&Oj6Bd%kGqM
zr3hIcBaAFQztk=YSYG));eUAoE;#v5Y%)b~CuMHqvLC?t<3QDeba#?X>q`}cm#{3G
z!0}*Z6nX!`*S4bZ_9YTabf+FmSv5P+`<WZX(oLw)_vedypLY1$SaF%1+&n(atCo$A
zo9lbJ_V}O6*7RZuIbId?qnAh=I;Pu2pa0SOU@*4vkUb&B8&5VYbe!L2KrxlRn-L$a
zQ#>;v!R^ekEpF9v&BBY2sw6&@X+HDdzW3n#%ol$_fjm-rMN4-JEkt(S!`?0#VZrsY
zasD`3@4);KL*p|eNMZxz%*eOtsE{Nq{z2O^ViXz^u39c2_pgwLs)rP<n`=|Dn?X86
z$MbK2^@^$X^QWH>)hxg9ne~zeaS9Pl3TqSZJZRXa1@dx%BtLjxP3f9sk@CJaO2<wc
z%9~k%Z0lJ=Qe^1vgXB62Z#*Z3rwM@K;kqM#=|N+JQS;V;!g@K%6yB-38=6$93bW->
z4E6qF9ULx85aOu<Jx;Sb-fjRBxi8puRMdhN3IUK<Qu`1$z(S6=2^y8K7)?bJR@KEG
z@Hh4K+GlT=N{j1iR!hKk_95_dX4&bk;Y-K#?>=nNJBsU=>sS}19And2JAD@g-Kcen
z`0N_UjEy130*Td^+MdCDKUM3<m6^Sp62Y;y6V;r#@p^H~&UN>tng{us!tF+qRENu*
z*d6alm(qgd-l2QX90VEnlFqn?c;2or^u@(RU;g}ZMvblyRGk{W`la&tss*}WwNp-b
zIsg|7&Nf_ke`y2;@&Fbihz^oQ8>~y;WeQ;N3CcjiFWg*#A=3%x`UlR;z1R7|7BU=N
zGPiPHO^xwNkzf7UHa9w3uPXL431YrYE?2~uX+gUIB6RG6L(jITujhE0zN4L}Vz&Y>
zXnm*RA`=gbKV(aqy&eb>SlYdRH0Mzk#|T$_^IZLxykKJ@np<aqNn)P}b~cWt6!~-?
zz!!w~ELqZ%VobiPqc653n-pI#$0U%gx`38?pH_qy=Sdf!SKjQK5M<iij9nh(p-vtd
zOH!l-pK^_$gPp&gb7r~G^_=sEka6Gmm;XCHhHe2F->{Ck0fLczwY}Q=R?pB#y1P!#
zdw%KNz1?wXLsZE$fv~%M6_aA#)#m(D)rq_v$UjO&&9wgkgEx8SQP@oc=U<;if9Oh_
zAAY9E+ZlBz^IG1M;m#9hPcFM!-ns9J5M<w-@2{8XCYI#2TE{TVBgVK;YTSb=3R-|;
zgQRH-__k18tUoxCg|7q^(x|K+hYXe8*?|mu@r1Irc3E!o?Zvcfe7$y5JweHizQ<Qa
zhLwVU@_7S(@Nw~?_(+KY8NYF4fW?bbu%_dVR7ly5L9vMM`^K==_2;81S&bJ`f^MgZ
z1=YP~vXyfZI<4etsz2vg{WlmoDQ^rSfHWwI+q8GP<Xl>H6TgZj+#zn7?DcJ`mC}a4
zs=2m$mBb=+ysZIcq4mu$MRxU)$7PNJASa%)26%loc63n;2YfdTz(1it-?**=L1~mn
zC5b^x27_w&KaS2jp347!<M(lJB-zS1w(MlqIp#rT_D&MnD&Zg_9E3XA+p&&tkdTmF
z%9hn3WMv;)_V)SR-{1d^$9bH`ea`#+dR^D`ygCe`ML8zI2nsF#3sCjavzr7O$;$3&
zgY<5tSP5+~&&iRy-vG*ggNSWi1ASgBVq0}!3P;{c;#?DCQ*xC~WCCC;&^T`mnOxUN
z_BYCGwdC{nAJzazUDgHj^`qR3w$I7MKMHcqMY&}&S~z~ahS*Ue1TKBsXL=ML7VzBe
zbeenFl{XVSdi?8$5raKKd)9ODqOcZOsC=$4X9FHan-kC<I2e)jGD_UA>e7c-W>C=p
zJ%4`zjO<Fq|7aS7^e_8+$O+`#`V(pmw7l0@{$FmjTE*~3{kJ#TmhJAxXG8HM6r6b8
zM=75mHo^Lw7?K839BxS}K6o?L9+*ec>2$+vgiDHO2+Djo4&u(7*9<*bg2F@0SH`xN
z++HSy`dwEkQirz7{8ch=9c;0Npt?`T>fF3CYO>7uDW2G>Q0r-3*^$;;vq*H=`eC5L
zK3~tN&zsvdNA)W2>iBK7Ut22ivfC^+!Rc+5pJ@rw1v?=B_5~b@oA{Sv3mW8&_8C3S
z!@$fh<=WHN$LU^sRuiN+8*Czuo7R4J|B9o^F3}<Wm8C{VcbZI(Xq35{rv{H=^mY^S
zQc&oiBH}y4jh!2raeT3FRgAG$ZXfrhkM4JOJ%fj~M&6El`^OHm;1d&?djA&NVXq%F
zzWgCe*4?=FZf~H|z1`uPd!J*k$bLqbL!8JrYm~<A>F=vH9{EC<K{AIksB<yv`=;`h
zYIV8}-02gI1`p}A4P-{l-5zp+&xnP~hFgC0<ALyBBgY=szC@WO5V>-H|7uq%W6Loc
zjV$EKHtSW9@B5+{WD;lC0w~C4J10Y^_6fP}pyV@-dwa5+pJJZ*BisrDERHdK{2Tmb
zW%KOqkI|Ymg0G{KPd9gOGLPMuarbu&jNqn&<Zy*(hQ1+EkmB|wuhk=<d^xATc{p;r
z_=j=Ear;t~8@DTqI6o_ZB|r_bc$&rFYm!b)LRKD-5aast#ql-)?jMc9)wmDU3<<*I
zXkAXf{SNoOp^LdrPSviWt9&2kv~&}(L=h7s33u5Z(_tYno<6&&XlLS$LaONb;vHWz
zw)~F@7~tXZh6I!dm?RB(`njlNb9qvAt5_#GcG4=>cJWs!!NP$-kkaC5mV`L}pnL6S
z4O6a)wQ0>hGr&koDNVsC>FF6b1ZnNm;Vy^I#tB1;JtQ@ydSo5Z7*JAGxvS6(i%S_Z
zUrh+^AmexB@h;rc*BtO401GpU2=2F?lsB6a>wO1bSg$)IA^EgoBAAXiOBh!4Q|*FY
z${EaG5~aVBOhOx*apV0>EQA)Bhu5)0ld7EWSy-HKWd$lg;=1r;PE}l}JVk?>IFD(%
zDSu$Yw1wJ-7y#f(sdyCG0bVlS|F3_`<bVBJ*OM|R_@(hbC7~lkGVa>;hYV@pIt^s|
z+0J$Kuy<YV46f>_3)Cf!-B!qRI?+%Td*}mvyq{+x4gP<J#2t;&t6)fC5P9XCPOcA1
z{9F>d(sIRYxe$gWxtK81D(o3RVRD&j3o%6cYueXd^cDyaT<QO|VxRhwShV$;@Tg(;
zhsmzg#mj<CFzoLqWb-61_#C2Q{FM-=e&1Qsq{5TyGru48v^tYmVto}qS4m4EH-wX=
zoXAwY>UZ=fb1k`J9knB=6utHd89g34D%O;uylkXeU5FB;FnvtscAj-iDYuW^A-INf
zT&U#K3^cu0A-}L90?c{hgNFjPbp_Rr(#xTA@&MJ9e&PBrE0)}myjGs|U$}ikbpx8-
z%tnNPnz-_BZaDUFZKqc{+deY|=28A?4}Mx=BMk~3A)Bb=7%z1jjJ$6l)Oz;RF+x9J
zK$>n2MPJhwNUI^sk=A-|BDe#}BEFoEC)bIdAn41>w4o31SCo6$NnzXfmnckYOvywL
zt9eCIyJm}F2YBk%>%k=)dAhYMA795V^6YQxT{XU!NP-H@q-KZ+DdJEX0__(>ZwmFK
z_87;*ejpm(ItccIY>Ou;IA?rMkD)yv$EpHeX1VX3zeR6*n5<YA_9fn*4la}F#91Et
z6;|`SknkFtX=EDuU_e9;wbOWq5OV4h_e+(Sd-#PA{ran_HCFgNUGQ<aE<B-N1j5)o
zyA}*KXTo9!t!==TSz1C0q~00BEQB^_y*e5eJ0%ij&e;J-IZq+rCNG58*zoFgxfmBB
zPUn&z++>q&ZHAemG$m>McC=kHa%G57{6*5Qp{uGOJ463lBoBC_7?J`8^@yJnM=?yq
zh*1n}HdZG4NkRuF0)H{14|E&KT|vt7i!glZJ}R-Y<J#{Nd}Gt3X#}T<i1&JEE3H3<
z3+{w%Z#)Sd<|bx-H+EN@<5QWt`MK+;`#+#d-){zHk~gtH+=wTsCJUeI$1Lcjf0`-Y
zub|2fk!^M3RMCB*bY=t{OCAs#IcE%=&#XIi<}N2g+-BCa-?{bAV>_>n<<BdnXDgrM
zPd!4509RKb=c;nmRnbr98tBIi>Fgb43sJAJz>)*P_cR{t5(^NVrgTw+JkJCBETw5J
zb8BduLCcoc7;`;ap-k<ud*(<bH@6=>=xeZ_cYBq0?eSNyv-c|UQLtEAW=D{hHyBt$
zez@XT9}C+7@m4|aD@IgT-X}wUUh;i<!8i-wh}2Y(U#jZhQEJkP@BQ`60scHJG~wgp
zHzse+CA$NA_)73}VlW|OMe*Fe{kd+8)@gm_H)WBqHJ0T(`lY|~H*H*cUq2~i8$Xuk
z7Hhk%&N_89#JOn4G$4ediDxzydQd3yGsV%N#n+bX@o~m<QowHX9-&7g_9fHtq+WEz
z5ZysZv&UJbIRF0C-$+&0vKK#MpxtF+-#=zH`w7M4S?noADfODqOslk?%=&=JKE-{)
zBvaK2Y{|y<qVM5ZJ8t4_>Z2&BE<Lkn?bs<2ugi~Y_EO${srL;~CAWX=PN_y<5ockq
zR)7;-Dz#P&3jnuB;RgW`$qKwxPLdSb$Wjitd=Rmm=SI1ZL5(C6k^qj2Z~HA1-*6pd
z36ox_mj>tZzTgPiRUQ0Pq2D0)Ti+FJZs)DAoE(w-kQH0ZkAsl7%cZE>eK)nWSp|r!
z;Tq(*HDTRI0@?LJo8=UjJv@q%LeXoi9Dtr0wFr~>|1%@WRTmpBch!6oi~C(GRGMm@
z?`>%CO*djv@zhcsIkZIXl;P<%S@8CyPPPLNEg_*(oo4ua<2_=K$GlYhaW94Pzr~{q
zQ4d+Hy##sTdF<fy09C1@)W`cxqS}k>64wQ&bZCmb6DM&3tZ*`|S7`l>n+<PET1$m+
z*<$gdA0VwKn1#{`oq`HF6nUCyg1?f;90#!EfALht3`yzT=iL%rB$?CDLB;^D;|P^c
zxX$){Eoc`^%zBPfO9q#sR{{_54pBNh3(#_fh?uajR^=K9hlpJOXQ-%08@NEZ7Ur!P
zD(`TIk~!E@1R&NwpVU$QUHN`3)G@g?HKYBnKW%4fd*H(A6V<b`w?)b^<J<|Q6tT_z
zQmX8iEDgzT7Gs!Q<JG7LtI-$Cyz6Yjq12EY6s)D<)*=`XrFRqbXafZZP^n&KMz~RK
ztAd6B;@#7dhim85c0%cj5l2a{4^=PpMJL*6AKcHgMB?v#N3jBs``Y%Jl9D1}i4}iN
z4nHTzq#!F56RCC0{;8gS5UMiE$#~-}V@f7<P+{T6+nNh0rtVKc?14DYz{J?3d;yny
z!fUBKsKAcCGg(P`6<W>kt{GJ5;=!z?*&7-Suix4g<L5%~OMdp%m-k})MZE()m?jw7
zP)9lU#`m1idIl~q4{6&cc>R?l=QH<R$C*IjG+0WZ=`MiXKBHt#8?XNAU^P42^t^Rk
zR+3p_d!9M!+6czAhWtzDkMs?9Dey818y{XrRqKJR5F#}GHLrbbE@(xRe?nBy;N9W%
znJjr;3eBmhelwV;J^*LTZWAp5$f~}pgK;Y}#!ze`{$^DiIE-#rS8aHiTs2$s4yC5l
z<}qH8B4#tw(1ZQ<<EPXfvvkZ5wqzC#?E(dmtN~Kn!K&TG7_jFs(Aud^yK9QtP7lj(
z#0**2lb{oyGJ;y$k<Bj1LHv*;o*`&>#F#>`lE`|U%59+&BPu_-5Pbw`3v6Mew@mUc
zDI~Ej;}|5em&L}N$&<?0iB0h=Zu}O%IO(0OI*%`FhdIL)ay4o{tDd!mE3Nv`?7GPi
zBN{nl=v7%n^OjUUYPm2skdXMHD<ran#huYGOwT4Ur@I468(o9|SLaafrG0!t%L>_@
z42P1E8Jmk4SI9R7E<$z!^e31_lNS|kEqQ0}Z-WC|Ezmu!hsBX*Y_e3bJt53<!EXOB
zG0Z;O+uBdP2a$#R@L>@E$+@Zu)<=43!A(I*r7P>Kq0M|oJ1H4OB@+YocA^GH699k-
z=LPo~mZee7aedL~N{P}}-!F9Hr4uu%2h7>DBgdAMo`!ZZM1J~rp3!qhHB>w_mTLBc
z^YMeK`TI;=r+p!*E4Fk-H_Z5tE4eZwWG)41tX9E*qrO#0Dk-{jP}fdIMH2rXaJmBN
zvCW_LyEa#Hn7x@Ay^hN4K3>&27ud~qkg9k}IQEw+&<<T~z+auvywNey?r@yQUiI~6
zaer~odLjz0mcVAH^mI(l@6X(W@zB9fU{mSt&9j1;_<82mm$uQeknZk6nV}GNjW;ir
zlyuDdpZcG^5c{zUgq&k*c>;811J8zJR?RTK|4o-RjctGg<P@fT`2Xw#cHScyTzBUi
zCKlT3Dy$^wYPjtmj8<h<*uSC4<g`4%Yc+I+BMMYIJFcEua_&g`>Pw8$ZVPUAo3N3f
z%RQ{~O^>6hJbvF6{Ql$K4HZ)q*Sjgp)Z+=`+ixo}Lzi$LE>psC-_0H4I4L{tvhuME
zE}1JnnNz%r&i#nuIw!I%W2I&qD<P$KU&aMewO+WrW$b1Tee-teUW3Zf;D3PahvU-o
zzx3p=H~HLQg83;?4?SVlp9|#P-MzOW&w%?BYck%Gj`IVpoRlKvU8*`^>E+7(^3E%|
zo7}P9T5;cUUT8T`e~gX2w;6X(_140&6UJ9269|sF>N43u7T=<uQ*<0<v@OQCC|`VV
zw6Q35q250zOT1-&YoLBZuVA-Siy8Aczi2I>5TY-H)1zE2@as@_xJUb2?Q-Lx+HOND
zG;Gc7rr)*kxr(i8?VB=5590hwXs;~#R*aX^B3teaC@cn=>MR~^nrRD#xFHw#%K^FX
z$~DcO|4CuR$;iJC271CxAt<)h`%?~rVL+#(w~WDnHuW_?*hEVu7Rprv5rubkB1$(a
z*k`<WcxYpk+-Em>k{a*k`=&|Gg;^A9fACA9)WTQPY{e;M^f+P#eoBc4O343y7o@c2
zT(2m(^bP-Ual?e8#3mRAx0Yn(hbq>MmV8HxPSuq24j`ASpZ((9bN7xNpXB$0;bjkS
z3hPg*Z}%Z5RElJOw7ZX4@O#Hk?>ikHm0mw}VAp4YOLbDMnq}VMPuoSJS=y!QLe)=x
zql;A$ltj``hhI0L++igt9#3yPNjXfv<9PA@`qhhn(#0$8jIH73$J<u2&kAMpQtS>k
z?<Y1z5L2=Bo4sZ<phR`MzpMYmr;toSRiTITvZk1vCZwf=b+~olSN6g7{bcZ}{Rh4!
zJ7jgOi~Pt2%pvlm@)y?^LgAaICJ;}<zR3$JJZFYLwy~Yk5pz_n4Bs>Ut#i?<_Dhck
zxn0^LPS_pOHE!v+3KJk2z)fbL=VqK#y6@xCl1n1Tg_*a`7A0)i6#%~rul@tdVnY9F
z#(yqz=zp2K=6_yN&sb=T|G`&s$h=_$zM8oPBO-$7vGit$S+vYrJM0wEv5);|bm@Z|
zVycPR=gOhb!vm0t2b@)0O8!M_ffjmIhPvu&h!WN=r8D!F$_?DQh1xS;qnSrn`=uWg
z&1Jm{3N~y01GWFI)bAq}z#x9*Pm)(|#$dUhP*L22Ef)h=T3f^*OOBKhrMy`+8$D;&
z5^|wPb2XPPF#H};J@%XR%Z#q2XGYLt<Kc2yyioQKRC^KA;)bW~`k##d;UWXNYS_1C
z2pY~F-%*YI<1k!wdxGlhr#8w%m3&vLKHr0WAGBD(t9fDYU{`73OYtGHMXm)|ij&U<
zu6828uGA_j=H<FJ1%v|Xfd+N*d2_>CC%>Aw{TIFnlMBv5$QRainggW<@%SfBZ_fw`
z1zeEMU||TkI-8JR;FI)TjW!k|4L~GOOP!dp2hx+snhfjhyhWua!AaVP40ssUulK3y
zB=Z|JrNRrR;`f_uP#igeHdtTe(u~I515q*woaP>=v2wf+OL$s0q^QF(^HU+|x}Lq^
zQJJd}MYZC(ZEaH5qoz3RKKn&+OdK9JD30Eo_@EVh!#_N^sWLG0A=qlErKfc;AF+zP
zM*qJy4u3pAb5tCb2#PKUPo!+D^#5XrH!_lU<J>Y*<He;=)KxEqaEV4PSW3m)U<N*t
zGX)d`B?3o6tl-I5a;Zh=612ET&0t{<9P*;y?+kJivkjED=iSJ}CD>FsMXZ8`Z(SkA
z6&F$P3U?{*11YZ|Js?z`Q$%Y?n|zQ{BydPd>tsH00459N@rcU5HCV(iaz)Qqz-K-)
znU_DDPIsl*b^O!N4dS3mZ)S&8+YpCy%agXtAmK`_fA-T3f$er#T`J`=nJLOQV^;l;
z3rFpjsU-a*|9s$wSI2DuALdqbBq$TFe4}&zTVT_6(G$k~2DIxL38F8RAAP+EJNqS4
zkv_M5{;n_B*&V2I?o@emw<<1aQslhs?!2)er($3gGvtTQrfqfE`3y~JLr?AFMUk+@
zpFiS@;10<s+F6?AR2@z)EM?$}HV}Uyt*xop6Tc8A3VN3v?$ySh?YZpB@dQ+Hf}yF|
z-;l9$`7e9EJwp6^tZz@5Is#>P-`sM}$Yr={V(0I(%jS}#xd;j_{m+i&`(}>Qej84|
zIf?!UzTF}IvFcp0{;vX$PH^oOvpY047y8-*2(L-p{<rAsUoGO5dcf!((JmOqJkA+1
zi%<Tg`LuJ}Q9i0#F-LEB6MTj-hpd*pby%r$D%h$`UVMe_r1eAS(Rc&bNYsF^Plq2D
zRL$Y5M_;$h$xwT=>x0IfxKiH<sTyvBzr@eJch59yLxt@5@aGiwB?P|ACP-OC`G0)y
zSx_&=Y=@*c!fenYc}*F^h$CV3Xy_Pv-ub+1`f4~tO=4a6`t*I$5UH$zgT0)0fAb+*
z>P;mb!Ll^PAt`7eM&B=lvu<~P=}EmC?e-6XX3Z#JUpG|rOWE%KGKV*fTl*2*)_`Xb
z7>^i~9fkUBEnxM?9gM0v)Y(P*W@-HoDr6pxT>IT~4-u=RaQ1uno|&Jf@#2L@G1!z)
z=q<@g;)|oym!v->vV3nHlm#f(f<=B}K(lWbq6jtw#;-Bg{*agp1K9yRZ-3dTy<E_!
z$T_t8=ajgm1P19jXcoQ|DejDwsHDo0e|gzCVRQHNeE?;hJz6wV{c<&6aKGPI%^F_I
z?`&N|#UE5^{OIiurd8de_wzF15q7FS{8V%^KZQ$KKc8)|$!!o~Eg5e*nDw2<4?R^M
z>U>y<Bm{KWe6ez<!e4u??3_ED=HD4~_o(GV`S_jx00D;cu+k}Yb}#A3B&nvewo*bF
zpH2g7Y9}CP=lalQO5C58&G6s1J{3{y@?Qhr+e!sL`hUN-qYMuFScrWmmUHu_+OsWm
z{9Fcw(29l;4i=>jPfYPt7}aMdNQzwTbepa7J@?YaPg$x+s1{8(pjxN3?XI*UZ0M|w
zq|W*O;h!|-Mj}p!A+V*%Nzmh?jaUVld1Sq8>Lr6AQ3+({1sqY!tAn!jH#thX&?v40
z4Oi!G=SRI#;}PatS-nkP(g}M@d0ly{W;*`HCf4BojLHg<1peLCM;h6JB;_|U+7xs$
z`>2SEf_|Zd?L28`p2Np!8FbqGhK&WvEo-_UUjYKT2a+&ksKRT>+AtF%;AzI#N5!#h
zK#FhsT#GUzm&8g*OAKVs45{srE1jfNqzNR(Q8@NfK>3ShV_RDX7mUcCC*#V60t*E-
zgH<Z6)lb-K-UeNDWUFK_K1`lz)WoV(N2dFHZ3!!%M{8FuL$4U!*+c+)9VV$Kf?3F=
z2HN?DDn%mvc_eHi<xn6dq?SVh(y6ox%%Xq=&*Fg0!$VnM9%7uvHBLA|ab4q4d){V;
z*6Ks>1n!}zZT3mB`Q4phA{<2?wHV{IdcW3?;050Btxj6|$z=Cg@hAsY&QSTs0$jzF
z#qKEK7+RK6YEMnCZ5;3j?^TcWYt<6Ouo%MthgHpy=^?M<A)rGmzmP_rnv$YbVC4B-
zMI&-e3nhfL{R}$dWC${VBsk?2&*2=HT$BQ~v7Uy;@is%`y?D=N$AjX{0eR<ytJywq
zVie*#UXR6V=n1a3|2u-YRdaO|Z>Oc66gZ61{6m)KuzQ~rOr+}#7SooJ3F_jTIMeyk
z^jd6+$D~hz=4Az4NjH47Zv(#%cn$5D<WGqj)F<s=G~i+Ds;*K7@UtPKf4xevvOQSJ
zd(Hy0Xnk|>5O*Neh}xfZ6-yZjAI5CRsH(SiC=}lsCSt^TZudu)fZg&|iCFT^LdTfw
zFhf?;-53%G+!{BO_;YHKcw)kSFCjtmuc2=80dudKr^4c|eS9Ibn9~)pAPY=6dWuMT
zy8`@iEh5a>CIE^2pomieJW$4%d#-dbUrB6Oh%jQvPMk>04o_IbIzM|mmQ7<(_vY%~
z6`a$#inMjdp_}DdjJ(Ub-u%aySUAJS2cYxU>C$&K_bTj+&qc|4X>eU{$vL#WI##0n
zKVaD5yllWwJb(bzCSmWQQM5FaQjfaYN~La|BWFK+QIRJ%J&t61<S}D-vI_c~zx_Lc
zwZ*~QCNMVzG<L2l{s;VgswSo|ui9q;y&&Lexl7?*wzXwDb8J!a*6~YvE{V6NO@c*K
z2W!c=zU7F5Xa2!A%C^O0+u@5(kN4yvckUQYkE^de_yEjONO-%q?uSNwne>?~OHvut
zX6F8%3B5e;y3Iv;+uitV8Hw-5BcBvs(Ht#$K5aO5dD^(P`!9}#+#zl`BVLvBY3D0L
z-@`$ATRkxII-x?XE(6m`9&L$lyG|HvOEdcP{ZIX+;t=`i;{2F;b4K<PX~<FK-G8YW
zEjtwN5#4Hz_8<jmoe2HClXf9C@am}-X?-K}#Lg%@K?XL_6i-<ap_=PA%f|MNSDjpA
zwpPp~G+5}oHX-{$=29-~MWgns!*ws2FTq~eDC?=AXz0Yp{GN}6{fW~1M{=U@Y%}I}
zVDQ>&8AFeYQ^&YLdXJ+XjbSl)iJ_)eeL>5OrRB8N@g{eaYppGSL~DBUkf0F(y7r?&
zW4WAan`x~Zs6LtfO4Yg;zSp)anxX;&Sl4e{H<B%wnXIznx!z~H(iad!sF*Q&^9g%k
z(&q0>S>G}1s*x;u;;>**eX%nn045MhRiQtrVp}^5ZpD7v(O$=Csn2|oq?_w>6-tt%
z0Av@r-Tk|chj_0*5yisrJY74jJR6hjT(D#ed*y}ds^J=?r!WM#l_aes;wcO$7>*S|
z*}fN&;rW-9O3}IA8nRLrx8?7}qSNe}E${ni6n*qd1Q;lhTTk6xluVu?X}+}c()LQ#
zAxetg7KWS0A3aa5_w8q#y05M*H&wt7414vk7D~0xx{%|d1RmV9>*5(AFLJ7TQk4*J
zuxe3g@9Rv|F4R^wks1S=%2Rc#WwL>w{lfe8_UW&@GG)%S2*DKQT~8KGnqMQum|uH<
z*A=!uTN3Zz7J^e56d3s=+ggj2;_N8V^YQhFN>S(Rv)5+oL~^;ck;X@)li9r73V!4h
z_TIx9rq9=IejZwUG{0r{Z5uvPBPXbGCZr#VyLxyi-ny_)_!0p^$NV^iVk(j6*YvTb
zRW!I;6Z>h<xUA1AIhJ__Em2B?9!rK6F<acW!d1H3<uwkGJn_~cl6dqb^E>8})|0(F
zFffw9QWXXw&YVs@-yw~r(7@4y-<ToWvKFsto=~_<eamwaF}bRC0vm?^rhgQ!zruZN
zMJB?vOa6=v@?xt}FkzVOFtczl`PqMfrGf@R%Nu-lZ(iAmIK@$lTHh-czY1>CO2a{6
zm>On3^8a&<5t+#lnF&izAtWruWF9<&7PHN6`O+gytWJgdC0U*7z0OM1NX^lw7S9h}
zCx4dP>^$D9`5-Y<!+3z$G#T4S-)4Q&#o>*KG575h!x2bMN^hBWaDFn~9|fNA!cFK6
ztug#wtLBafxH+%1_-ve1yfp_b$!pV^k4+-C;aj{L2?iZ)jKA#@dN*(-d}t|aBp+R?
zNgqeY&yvTx_#a@Zt112txyXH4%kfP&krA3apn6_9byp7x(ojF2O174KlHbGw*P3JY
zx~%=vuXQh)urB%X*zE?F9#5|y6t)X%z=lk9BaT3@g0<w0ZFVEDGXf}nNz6teS8;%5
z52vJ;u1u#JP_G!oF4HOtc+x4YqLYtYA(FwJBGh@-lsueXnsfwV2tjdiK{ofHFQ^fw
z8gCm{i$-K;V*wOfqo;@(WUL&_(*$|l|HMdVstscZ3q#F1QUf;j4THpRO@HPqZ?jF&
zLkrBU`qNiRrn~I*f&QX@4~vU*rPz~mDNLB|$od-;IC!}(C_4tQwEtvI8}5O|Ie3c|
z`1sgQXY`?4OPPQtML8?pvx#vjmy%M%@F6(u{`5==Gsj!y8j3+{N`x@{NiG1zLLf(l
z1HmQ2ft7MR0$!-^g=YJLWl4Kn2G=oM4CbB`VgQ~AMW1$S>>u+1S=iM?U9gJ_>sOB~
zZJiIxmx>@_DMh-#w0cf8sx!V^LIaRJD4YT>jmG;$f(dQqchsat!|0Vo`MhQlvoXZ9
z1+vUc2J-m8;P?7Uz2imSu?k-pdOEeLx<)hm&0SqraV-6H*vr1z&(@NOg<TY0;3)Uc
z9@e(fda;`QAFwboAn1I4pN+VI>N;H*TebCRcNu-nTb$e>8@J!;{V$-RZf<>ZbXhIr
z9Ck89)3U<_Z!4bSo^w_U!G}gZIQEk8ne2`?OSz+5LseTspGUvncCo@55>@(2*Kj}R
z`i6kL9xba&AY=-AFl`gsaobbr7O2+v|2qTgIiPI)0D@Me^+Ns#PxqXNKT}S1p*!L=
zKDE5MBmI1$wov_;cW~KV(<^{&ImPGH6gt+-{o?QijkN?aQ7?K&E`5iO^z+x|=J#bc
znPX~Hz=xPGPe+tFCr@m-+=6~esibogvlxEO=-t|o5EO#&JDgfZ>+_~*qJL-9ExXa+
z2AI6iG}Kq55syb(zB4$)@*875P|(n#yT-UIquY7*WHBM;iT~CneGjI8t+JEDruMYi
zW*jASl#rD-f4iI+l)XeO%0@FMNSAPnuuN!W(LrS3Wr_j+0bgSUW#yq4S+ln;*ye?H
z9Kb{VqXTQ$v(1KH)#eMU<)V~2!8eZF1Lc}B9Z&jyxh3!sLb2IkOw)0_6`!XIJoPDY
zLa-ec)<Ln&=>v^v>H{Q!V9};9>_+&|V(!w*?$z_P?85AY%5r1jk9}xuj$GGUoZL~j
z>YAoLpBF6Iij`sIJGC^V=Y<!bMSA$S)Xb6{4S?+kk4vCWJ~l<;(xio+S@`Aqa@IE8
z7t-L)VX(CO^JCgZaj*F^cjkbn`xo!06{6jX$xYC(LG=Axd2_Zmbc)01m|a(OdYT>7
zik6}76RtnA2lI)6xxzpfoEF_LC{GO-DFqrQ(8&TGz}^Y*`uQs0tGI&_QvQR7QT}9E
zT{Y+iYgHNwQ<F_hO55$~l<UimBpgwHdQ=AL2==Dlk+JL6Hety}jZBXeKUS-f-ZDUJ
zGuJmwd89e1^=p`tM`$&qHU9A3H6A)6qWCe`^Z25hN#v&Wj(Ns<_lgBAZR~vIZhm&w
zj0Y{8T_ZHPPsJpmFI7$4{>inSufG{9qF~qIi88#E?6(k!D80wiCtzU#%@+&*+8HHa
zhFd0iDO_#KPK-%5ov?ZPeEi;}=C4E!5zy<?tNnR{=vJ-eY5Y67r%RD~DaK#1PQa&2
zw?Ca(VpX*j!8Nj<>xyYhP-&Fs4ov#^@SS;sCb@1a-CVz$wfl~ChxZ}zU<zq*igkP2
zPRtckXjk=l3y&o_a`Oi<J&=z|IaJW`Hp+%vxVm~daT%&nD!KJaUODj$8cE};JgV=P
z3*#3#Ut!xiDOt)Dj&kx<-Fj-u7lX?dn5QZPzrUaBxid@Nr|r7Yc&W2|&h-v?;+E_!
z#-Z-b=0Qk|@h~IIQ#=C$A;MO_%z~&{H5iRetW1VMivjijk38T`tt(CTY=c6QU*<ZP
z-DavA3p@Sw#qVsOL<#7@tG#nn2}S1}#<?4TT}*(HlsmF8j&(wZiW(`4nSW(kqE&LA
zx|MePtwJLcuESQQA3ZX)T2ytLGKJ<~)k=FRME>*%ZX6yn0=c?N=U&Ro|3x%W{B2e)
z)gTcUz`1_!UMx4=!<tiEHo#ygROH7fm_;7?Eu%ZMF}%sG(r2#ZV&R5qQ|wUJwYrq(
z1~?9^c|~-hhel<lQuvaBr4EX{vb{D<q!s|KnUf+~K;RRU`L$$k;bl!d>i``H=Y08%
z(=9|zGN2Huo61w#1rIWD>aqnxH42U{McNft?JmtQDGZys%!xz{q=pL^EaEx91A9G%
zSO1dltz%KQVc8*O5u6x2d1tUl2*wD~r9R+xm(~l_7UP^87||t&q3@L%<|}YxYy1Z*
z0`6}6@)%cxkxDIFMDg?KP1Mk4Jz_o^1qUrC4A~Be6Uq7ecuPq3DN%ufwYwjXK`#Ci
z;Y?WosO==frx8a*+TnQ_!zeM&mOyiF>jz}<jq<;Fz|0}#5B<KlkQD<J`Of3W;_`~3
zYg5`RE+$}HyhbelnVZI+?Bv@;a{h`3g@YQ5@l6SCnx5Av!+I8QU`wnUjb|5gJ!=Q_
zdNNY>8tr7F0ku>zY6uj`X3Y%mD4PMb6jX1fs=W=ZJzF<X)RYSyG)f?`M-Q!&olC((
zeT6{A)jU$g*^Ok5iatULCY+hFIv@+`bSZBN&dp%hwS}1RMnDjwf=OHsqEM=p_JF1y
zmMhG}n;cimT^9o8y@lK#8-s(8ONcvCI_9&cZsRArU@C%xv8&H7{}MWypI_A_@<oLG
zJ7ErwRKi9oBcG$EGcIcel?v|8g9FA}Ck)7&={m8@{)Ak>xvHaA``M7+SFwOOKWjyL
z>t~i7Ss;`Vp^{1_Z}{d7N865yZ+1v+*1vg^+qd5frNvr>kPlIIe#w93mfwFdYf*Nt
zJ9rpeeBr0kn*nE}>CJ75-m84{`}=>uYwtxam0#b$gK78T*O1r!5f^4)*1xRLyvf$x
zVstKawEKfAbkbZc5`{3L2ylsC-g{-tM_P)Ltc4=u^~~NxzA=tTqA=o`bE5mR@qE_!
zvO@A(k*5md)?IgtFSGw!Y-b0%K|11&E1mvL7mqh9xm;mYfO3~EXEEG(SD{fEb(=NU
z@lnLu)2=f+U*;%}V`V5Bgbl4%y~_@wKA1=5hqaw+q-2V3jvcF_t(74Wy+ehWjXy<H
z(wQSltII%tn|4!PzSV$oeWfsbF~K`PHi=S)oz2l|EPN4A0_!{8?Qtc#6n84SJu-r9
z>V(rV6mQi`^U1V*n!a2cQ&5_17~OJweBe$dC~s$0n{PbT!sH8Wv5x&}Z&mG|e$&1d
z<|*=vHdXw6phvK<9q;OciH|cHuZgrxp2~zb<*FNA?dM>|6Gm*Z=z`brB*b~TXK|rX
ze1d1wBIHip=4)MK;fa6U(%wCFp#c`Irk#$~xW#)}X|II~sYZbOmBLTtCPWAw*Z%8~
zPxKce<*ogw-p^mW#bS22-jChjc8+}$OQ*aH0txsEK1+5#|9i(kq?J?`xT?i(KlZ7E
z+$$GB;OXM$6rm;+vmN9E8KiPH5`qBDh>0Om9>R^tWV08P_p|DJL6$^homVnyU^~>B
z5@~tXPC7OH2N4Xr+Owu&t)4))ilWZ(Z&jYpHj=UPBQewbBHp+GTWXVagj!;zJs>>U
zek{#*Bug8T67zFnA`pv#W2yvJJvPZLOBl#~)23*Q@1rtj<cZ#@R6N6#^Zi3d#OwA$
z7#{6>-U~-|*_d)!mA^NqOIrchI*i`v_sQ5K2#B>V-g<y~vo4^=*|SDR(t}MuD4%ln
z|32uI8-@FnA^rm~-a14!c}3ty8V9;lD-fp{$5AQc`S;va<=sp&`>jHHL+U{n6#<uv
zzdx#PKQeLJ!y;iH;=GK0eySsOe!*+yWz5pK_u4f-<3xZJc!elP3`wYkcR~|ASi~2#
zVRS_$^=|JhJsq?KDsDYJZJn36Tl|4bwd}^kyQ$7M^q|FwRe-eje*R0LRJHtWP>2r+
z!R-@-CMI-BacC<<6OaWQn%<tm(8;g;;t@<!QirNFp|LugInB=-o;;c<PPU%vZ7%}H
zKd~>q&t4K}qGT4=h+T0vjTY92MiW+bUaO^rhhqU@JXul#$T^@%0tkFE($L7C-1fM;
zdVhV*LuZ!5@)gscu?eC#5Zpydys@<AfO5+UWZ#HP=EDJxdBAOZ18J)>qZ?v5T0%?K
z&j%z}lms+c_hqfp282v09KWxzVhf{Vd@2H?&2POO^2+-`xEfc0BRS874Vg$Tu=b<6
zZ?eLdOLesrfAc-C+4Y(njcv7ixoypV$Iai5Im_HwSZGP8(lqf=){ig_SHPh*kluFg
zyX68hR%fSE!!hYD*K$K~J+GGW6fsJPimcEvA~EjrHw`|?D+#n4YjF9uF07hvc~IfW
zdAk>i^-Q(BhD;z$g|O@kk$DSFG#2o$w}osPGHSN__5*O2DCn3_5yWxhgcCfsju2-J
zFYI*%f(t}YulHIknFH{4OnF5-H~PqTskZwS`rz{zCVy6vY_3}un5GoUjN891;Gl@3
z6afpPOABIV(Wooi<G*SW!%(#2H+Yitkq)PHfdD0vbzKK+ad=5DQ}<P*l?im4Ks!ZM
zO8D1jFgLOt{FJ2hua1F{6m~lbW(jSf2@Jr-kG12qA3zWkvY9YaD0+qJbNq$7zr*;A
zO89TQ3wn;6tkZI><Dc6r7s7vl>x%RyTN2$&<=kkUhPrxBo1+UhfTI!1wyI^v4+YQ$
z#*u~Yk)5z2PMHB$*4OdYP=`Egb+Taj;gstL&@PVZVk;6a{b?kN+Et8=Ts4f7EQkZE
zG`Q<HSA|(;CDE#_AlsX@i&KF}F(_2a!}7JEfd2(^tA}=RM`G(bPW|2=RsuuJu&@fX
z)S4><g5Mj%0eTS4Z4h-dR!kNL`e|s44@#DoT7)T8MK(v@cX?^|s$ukDUxQ3}T;cnK
zo^C}yf|C-q)7jNxs)V{QJMLiJmdWt;k3-<49(X*-Kg`HuqC2I{{kH!*e%03RkGPjU
zqfGLuuAveFPXWtbd{Xxc{#{YuInXij)QDz7nFmmm)d8!je%c(hEl<v0=gegkW*C0@
zv#Gt2QvM{?T=`~bxBO82$%fXd9z&Z|WoD2^&V)I4dEnri?sskJ?T69KkcGCbT@img
zEGE#>a8|*v<~RvSbz-MRS|2_gYCTXrs<yp7uMl7BAAV}H7*M1&XB6d(KKZ@LPFJ&b
z=;`}cG-AKsh-)%Fi*432!T$$oGgqL|)4M7m;$J=_-2qbHTo<YN>JLk(^%1jC;s?l)
z8AP4j>n~%4np@gv!M2XzMiBL0V)j3IB5Gi8uB9s2dY%N=;F8%2y)a*ykAh|S&(GRC
z-NHE8p}xiXU2@f)afP{ohumE_i1Z<oIe+fSJM^k@KJ&-IMDsCO(2h1+`COJgx?+-_
z<^v<?-|8clmCK=>)Mh1z>iD2weL<?8F}Z7XN9hecs*4ZL^5*3RA8c9*bYJx*iwk8k
z`C~ja7PbQx&c>rBK0E)C7(6QAZ;UX}8k@ntchCvLQp#0fN$2{4x8~n)H~&%5GmUN+
zd4|4qGNO090#vjmrsnL@85>o5bx7sEW&qt99(_}`C$SN5o4s{YzIngI=TpoHS82j5
z`8Z{4Hr5_8K@2*0v9Kam5{{KBa7PD3=13K<8ucFA>Z4fYFAEOuKjByTnqVIsUGC-d
z_F1kDN$!_EtGPpjCuG2lQsyBmWIh2TMp&FKU_EGMaDZy=eni~gNC0v1OWz~FRv3b3
z8Z5RpX+{$<@G@dt4j3n}+qoZu9uT229iWxz=Rf)}48$yb($Lm6=kY@Q419cdtxuYa
z@vrW%8fT&dt8K56nEx~MAX@O?wWT=Ns<Q|;FIRcsm>S{z7-RU4m^K^hSt)+Ay(x?&
zL<kn?@gRzvL>a!Y?kixtbdIzZ@G$T%Nn2VP?x+^+ib*2nv#364w!A)cx$d_9(*;|9
zM>_rUO$-~`wg$*IEc&HbVLUjaK7?)gESBvvBhToqHDW3dJ`YxC>FTiUy<MFsnH)Gw
z%G}DI*&ZGt6L!!jVSr_Nu`4xxh+=(4siIr*`H@zMf?=&$Od5Z=b3enijvqhoEZGR5
zMERpkBPObyOWf;`tP9_2Ids08?t{F2Y7`QAw0(d4RqLr&2OuH+wRj=}Mg5Yt2II-0
zO;$J{O-t7T$cQ&w9SHcIZ<UyKf>_2Vtzn1yM6YP^*U7f`ruQ-~Pb~U9kKLEUM`KxU
zb(s)_K|g{H4kfi5tGovUi$q!2pVg0F_!9(M1vZaz%mRgr=!JHizRe)<R{+3{#)Qbq
z2rPz~FSUd#Uv)gCwlmE#^_F4`+j7r(G=uEQ_>1XKVDbj)H1@#4l2n8_r?)HptCq0R
zB*W>jp@MMboYnUak->pd*A{zadZv`4J{PpkH+=B$GammOQTd{?*m~IoRSY=42b+Uw
z%0xWe3mcSEwbRsOn)q=>REl}T8qH@zL+Fb#u3tUmPnt6M#9Z>M_wE0|S-$eej?(^#
z6BB1@s$}WE)xr4|y?b7Og(=rk(2_h!oP*5SOj9_NkW!R}Xw?R#2t+!Se`h%xL<OnI
z>IRuP(@#PA0iN1H3PTcB9Rio|Unqz2?@n9iG-`BFCeZF#`LZglhWzSM6DCs?UI{!Z
zhTd4vR{xT}CQq!<HSj;e18Oo|vcrs0vyK<8kXIiU2*Q%F&FqvyX;N`+uikqal3^(0
zX^=e~!Z!b4%)n+g4fUPy=0$MJ0T_=Pc?|+apvKiS)TiA;sslr#$jLaVnb;rTp|*Lx
z`f-*PR8v2h3&|37SV4wejCp_@+ie*N)m>8BK7r3E%s9lp6UkQOmI^%ETV2<O{5jU0
zN7U5Ocz2xAwumkr^mmMyJk$)o=__ir>-C6ABEab)>%nx-a~_^>KT`dNpAG)`xv4@v
zNNt4dLF)oebA@VpQ%{@%>HvZ7>0aZxSJpFtC^dsjgh$<%?)-Z2=)T@;9h4aMzPH4<
zLqdxfj$t=vTRa^i&xZ7CJ`~FPfnn8AT16cc0-@!+2t4_vs8#$7Y#p4p1hZ8MM$X}2
ztW>>&yBy$Un@@(**byO9dvPiCgMpO7${8hv3h}a65>(hq;sZ6knZOUdc{wv<raW#W
zHe7z6;BpJA*O0L2W&10i=bhRoBLw~JT|&UHb#`9+^5CIk=yu+c%@z2LKBFwy&9&Ln
z0U$#s=D#!xb-jHCZlu!kh!~g0JB*JS;l^?I2N0(gDsA<5*T!TP-pqH+>yybpOsZ2x
zeu@k@lfA|t<XrYjGv-bNO?1?QV*=Gev)nU}W_3r5PlC!UvSPa{UF~lhnrB_+HFU%X
zmcbn&%;}GJ=(67Dj!GNH9<Ia{@|+7lQ4s5iY)RS6P3-L2ucX7N3lOypC6>O)m_*$}
zu0IkA`#3>r|0Kz#4e>kNC%{Ul`aW6rE5ON3{>aT8F=kfW%-qX_*O9cZ{Uw{Yn#ZK3
zyJT<5dzre<Iv{Kp5Z``3QMvgOG?CmcMDC~XY8Z+u_22SHM}fX~8)X3Ely3KFxda2d
z<}8l@NNW>wF&ooi6I^~pM|e8w_gW*yff~RwkiOm?q1Y;2)co66w-C2!#e>G(%XtxA
zK8kp8dy)I)68R}4*<g4{((P!XIzg~GnZZM;$?Ws5pHtB|B%gm@)<ddRM_*E2NbSex
z#$UQPo|_wS{8Wu^Km54ZixtUv>-n9^X^<Q~TGTN6a=!5Qjr0`b+*I*L3K2%eazY-$
ztYh4_yhCoL->vGD=X2TnP`tyW6eHx(@}Z`@6HRO!F!}nD#7dCr^wP7V1D6N6sZ*|B
z9y@-%XLtA0&)}!XYHQcgI37n!esM}zt&IlXrv{?7h(2)F|5_cta!LbiBDM7jfg~g+
z?qD@v`p6)mou~qCY|8y;_AU%Iw<RW=6`3Nr!pkBnN^a5WG*zK~(s*gB_8h9@Jc?<b
z4BH%yVGL*`kV&$#A~{$1p+e!V5UEiNf*F|iuD=kK>GBYUv?&Ebbq>J$YLs$De*FO+
zN&rYZMdCqV8qiX}F=7T2NaG7mV*571bBPxF>)c=TI^AfDG^ICG5<Z{~?`(>HT)4_v
z@aX|ErHYN8SdiVFtoey>BR3e3YxV)9@I*9y;yc;bks%ouqGrQ*KO0K;;-0Ry8xH66
z{yR<g#s-O%6t-VZRmHUs^UOr&*63&9#Ro<8^{MR|sc*k4XKv~>@KH$`N7T2H@5Ecb
zymd=&CV>3xcbP)nm(&2|2XoQ2YEE+b)!x|N;qQCbn(E9Zf_H12s%z7uZ5J&FcZA4@
zCpi@a;a@xKq{@~%Gfr`H_*_XMbhqhc;W(dLimgY$Uxu!v8;WvIrfmbZ%%nsSpvHsk
z$8|8+8;W~3!?8B0&*C5@vT8BjmH=%-(c>TQVm#X%kdz6rtuYJ#@UU4fU=#Zk@aNet
zgj(4pNev<f$r6yQQ>y>lZUiMU3GwGnyNcpp;^_k3*YJq44cw3&$~4%P`C|~V%|oM-
zqwnHVeJ?(-o1i;YPwQGLuGU*1-bxqvdb%_8cU1Dl_q<U|J*+<DO;!A1oZ%`8Q+N93
z(bs4w_^YY$mj2ZT3KeFG@C+hf@G==j>}M#W&+><!(yDp6j_WMF+p>G!9hj2yIwk+p
zVfv-Fuo<solbndi9=s)eIGvt}BUr({&)m~sXWeVb8?Zkh@*m)Mh1IRlY%2h-b$cH9
zd@=pbZa@aYxeH3H;#VrmtsptU^4(Q*Gsr@VysJz66sVs{XhHW&>`*(o2Nv3m!)lH<
z?<9<ujH465q09PL(XOv62U}0aOLXsYU%1*C^}gFH7{&8}+ZM?_$B2F-c<#4Y^q`lP
zeHSz=FVI}k{w>eT5@8z}bTVr06QYnDvw)G(cNC(~E`&<=97Tz&deT_~nU4%#;rt++
zhI85|SfBjRT{^?|A+iO<`D9ncOHB-v#)YS9(Aw>dAC$8MLe8IvBlkx_^9vzWP3U-Q
zzcFL5<nKzd5(x2=Zih#pU>T}x14<&~(GiVd(HR%<vQ3%sNa*PhSsJ-lvM&4|dIH1l
z$P)uMvPm2k8pY6WU5Y|r1Or!yDv$4c@=K%;;De{UakFv1fII;`lL6u9L>s3qK&nw!
z8nlE27K)3coZVQdqJj-;rQ)rwG7fkvQQWV)nxJi}r-ywQNsUrk`}$zg55_5im(RPR
z^RD2E5N&Z~xu1$Nj2yQT_d`o%zLM_NMSeX7;>9{R+B`QB7V#Yho5)i5L|oDO6bJdA
z&ndYaq7<psb6G=UBy&*8A~nLK?V18H?)R#(^Tj*)SfngfW52wsMeT2+hPJeb2@7?y
z*KcWqK_K?c?a}w(0@zPR>qQ{A4C=VaLrW*C8G_ytX202`)FG_srH7X-93tb&9!7(?
z?qsaBI_;Fn|I;V4Nmz1=^zA^1F%zg>g@<j&i6lrK49fZvocKPzdn-1vzcg7ial24d
zMb{<IbRH4^_M`Gxu@Ia;Z^V+?K!KWyG3L~l0R6r&^W|K)@Hp)!w1Bja*BonqnlNUO
zRuZxz@w@Wh<<Fj9?6YBw%La6x<}sXe=hOaO51JG|qV9bo{rmX$O(Q1hVK9>RRrRt;
z)rJV>mHhCZEy0l-Z4EA)0M4`DkGa>>NGA6YlOi6<&q4z+NU7w%vS5o>IaI*NaL%@N
zHtX@{V|SVu*~0EnrVc?qL)PiwthT#)v{z#zb}D&0DTl{R)Y-oU;lLBo??2$?!B1Kp
ztM3<XVMG4Cq1FDjn_=HQnL`Adqq%B$6#ho}`JX+X<MhN_5}G>gumfF@9I-r#rZ!Da
zH=Wl=iVE4mqBRVP#6sinN8QE;csh6DMn_!M3)hGXYapantK%ifV941C5umJ52)jM2
zuh2ijb*gc4Ys`Hyl%XQIJ+N^#jmB(w-sYpj@JtjZ9rEV8>ker`0%t^04EA15`HwqU
za}S@gep{GdzY_1aZUpWt_iDf7@CwdMt0mA%|IW^I#kk(td2u~5H8Vhe=KM)1%0B99
zmh1U%iE&Ruy+8{S*|4h*$&HuD08S;*-zixm&=E7jQwIx%2AMqfD%S4=X!m?NbCldj
zB6hBpr1e|y`BXSF7#JuJysr-W`SNs2c|i=&UPU{a2I`&{=&+-AO@(eet8B@SyD4T%
zZmRQJxx*bFQ;qsDY^MK5{b;%(5j<6MMn_j(e4fYDa6b=YLbeGTInmBDV86n!eIn8M
zq18iH6(uv4KTLjFa&26};IP&yHQH!W)!oV^qc=cd^qNKi=czoikBNQT1Bz&3j9QQF
z^(bTZat%Z-3#kE(j=P)#05bsMt*1;D{_^Y-q*T|#baEO5Odze-3ZS_pV#2}V5O2V>
zBv<=z1rrqViW0>O*AMo!$jP&@0VPyf++zQV^<VO_{sbr=*dcdIvL&6ao{vPn!1epH
zlvvA$XB1WN?Vt4SyAKpG<mo@3x421iGJi}q^x%*HcoyUyn)ykd(BE)&YID5UI!HIY
z=QLiH@h1mkzPPw6COz+rvv^_%r%+DGlD%iu=D5Y2We~ikD_iq(2ewsA@!kV}DCaYw
zX?6d(^z%+Dk9CX8ta~>giLEh=)5m3gFRS#*SSFv1U$KNG-phE8HJwuY!2f{br>*M2
za=vZm6KPrvmmpjr-;%*U2<MuUKwcKa3<N+-*>N{&hKf=m;dUVR&NGO(#dejSg|n(3
z5auUnD`QMzC%qfaT_O3O+P%7$ah{fNI`G%-`=gXsT{F2XDKV0h+bFIb4Of);!yl8L
z9}I#iWChJ5h4!+1G>p^9lg;gRZ<&?<?zT|V?ATH;XBhd$vNu)n{P|SnQQi{wg2~BU
z6PLbh9;E{Ak&(2F{hBxQRyX*Ya(+u16##%zNxhY|rjv0Rj|rQFzM8<34{TLaw4xsy
z3p06H%tD-VBE4hkUJ~y(c!#R0-TWCw(uEu@XWkWHXwPnNd4V=$#6Jf6k_b20AH}0k
zM@42Y75DL3$>n8SZ4+Tm3cgM;fuO~5>Z!WZi~j)b8yu5EVB6qQa}3?S!krg~*x3-D
zBNig%(-VY?cFKQ%r{iT!_cP+7`Bhs>*fw%DZp13<i`~<17B)42+7tFfEXSY#_oH+&
zdc2k+$ow^bxLBN)$xWy%Ltgun{FvA=_>(g2?D^MxWu=?nOE|VhOQHm%S-0MYWZAH9
zKbbj<hVsM_npVE~%vIDPVjuEdv`H#`5~c862-mp~W82al&orxJn-KA%*%!pyAr#9c
zb^|ePK;aIijeiD5$HO#=;cU3VRwx-oYT@DcNt6l8HYUmEx{_%$Dmn(4h0i_AXFbBJ
zw?)~KzPR<8fxPdwr!Sc|LS%tFVX(qMCNXSV8@BLiTc<isA4*>l$hDs_Xb9#*f3hK%
z>y0r;Jo%qae`aD0&?GaH>^kcpw2r#E6nE}0ldzDWa0CKH<Fjz~b}&1oCL`$mvr8W^
zq6&IKXv#*igWd1sB}wHKNiqfPC3h;AoS5ohB3!rnI{<yo&-o`ARV<=F1O&yc>%dNo
z8~{AV2U#rU(Xn`p)={7-8S8sDj$1^69rLD<w+LQ#|G~uQp~#h$Kb3lFpedYAuEsVz
zZgz!DH}|#OSn_ePzPxQU-}^-JVsEC?-~8`E;ARXd`;d^FsID+Hq=iNSyC8d`2St|4
zL6$5+|5JG?Y-EHXoFpWO2}~Oj4Kq0a?T1(PH`QW>j38JOze;Q1T_y8u^OFS7NI#?{
z+iE`_uCw9>_=1;06Cp-2VKR>t2l((UVAeUA5(CXP@xfCRgu4MLOVBC-BPkuiT|;X~
zYSDVX)@Wi%1UVWf6?1CJqoy!?7p>QO&s*f_YXDjwX8VVFBV6EeP5442n<a#QQ2A!P
ze8>AxKltVUQFPvcQ2&1%|8O{aMOkNOC*zWJ9J2RFR<gng8QI}b)OF4tiF0KuBs(dx
z31@HFdt^uW{yx9|(?32wpZELqdOjbInvRY_YvmfQTw*{Wwu97H8;oaRl-3l=-!yMD
zwh@zMUwB4XP`brG>f1*&lO?|@g??9jK$v2yHm08zIN0xCH3@J<yd;%13x0m<L$9xo
zhW)H`Vnv|yCAsf)xs17+qy4k(`Cda^p}Yk5BRb2g;h$5?Hq%LKp3@xk`R7>&XDcCv
zD3_Z#-4fx3Qfv(KsNMlMZDXNz)8+Q{%5ocR$e+FQ2i4PU>0|pXY(G2RYcJ@&TD%f`
zRFU!2AueY+llt&vEvk>w0ijc2`{#brmp|JivD|(x-iS~W3gfnqv+04gdLe-ltrHLM
zF0`O~JM?$-r}((od3x0lIhUuMAZ^lkKO!G~N#Vps<$5`F1xFw7lm$eP4r{K~taTAO
zJsO+pxKrCg<BrQAo#SNz#hOLUK_SjJwf#J_p4mM7W^+CIwJiBW<BIdlIP{|j^X7vA
zUjawmpSK^|AbVmCS;Mqy{&|;qJ;j!U_a&(H(c+720~ePbxhc04X=aBuly)$i8Wb^0
zNjBpzx2Ow?>s&r;n&7rL`FCj<9(lw)98P9`*&{-21%3M5$HTrDwP|khnGj@R@`tnL
z!GT<t3pPMOO#GJV_mwScRx}XvV9-c`74^zjLv12Ya5XATvOMsk#6#b+2X}o7!aw)=
zU5NdW`^@=nV=Or|#)hk}-`NeDy{%;UAaNP{z|%fXqE5gc&*0(M%zi2QS=)<X!x=_4
zLr~6@b)Ku^w-=|XDx^{Gp*3$DbB*aN$yX+l921D7&x#AYe(0sk-mn=-@0dX&(>ho}
z!^WB5@#0ktz0%!fC3zfQ1eys4!FMWuK?RQs=jZo_=Q3%cwG;?$g~0^!7Vu=OXW-u_
z#gh9I_v0>)9ssq=dP46G@UlW;@YVDAQHkULiH}j}uueY`oRI&{(&*l59B1VfQN3MJ
z=x*7~`B@Zw%!{LYCf9uHn;w*pg6iD=KC_TEL`EC_$%rkHK711)_3famJ43>U^uW;l
z7xAHBuau0|{w|Nu{#YVnN{6eGvZ-qPPl;!sg!PXc!Q}o96<MsqXYOPBr@!rQr%ofg
z6~Ki?GRVsN8b8tsutC=%_be@kbmiBk-e>Wxzv=hSBF*0Krbz7coY1cCM=e-MS}-at
zW?K2ATN=NehBa*He8a^+utW8Nl=1>eW|DFTGNQD8#RFsv&PE6TT*q^0aIqD4JL15s
zsuY(qRDl3E)I!-#@+B<($ZjC%sdZ;8cx((WjyARooOQaPKKcE&$&?^niRvyoFHvo&
z5EVor7!6e(hRkl1ug6HXAdm07PqO>175viP{Oafa+31tk$ISWdPr06K_9d%Q2VQga
zj9zu+ac>A@|CbkKfS<j!mv6x%F7kK)K1e$=Y=<ab4u^Ww`iNP7n5ek3&3L)iePr*E
zpVh<AK$GkL|Eu1RM;*N{hx5k$i|KA#34lAF7`;@>v*bAxkCA8Tu79eqn0*&tHkSF0
zyLP4Vnt54v9D(hapc7JqyE&`eOn+JO<7w5Kl~?extDd#vJYTuQAE>5`SMPPSKZ2gU
z%&{-sSKl|a2mVx$=DhZ+|GT}Au&X7zW&WbMnMF?L*w`agnDthc)?I}P=INcfq)sKH
zz2<Qa{H(ZQktNrcIDr}~4Li^6do6M5Aw`f0OHYbK-vRmJN8d?SU+`CBJgRW@ozOCy
zn`tvSlUolF)#Kx#&kcQj_+DV0RLUM!bH~A4{KwlZ_XlO}8<U+DgJ&<9AXj~-$4k>J
z$M$CfN5yR0^x-8q<T9R)mmdY}4ZTPC5|_h>5ZX?8GGEMF;m$I^sOlplbwas^abmO-
zw15wS%qOj3uM>y@KY2Z=e*T?Jz?W07tG@hp#Q*nE(d+4)4m697>|Fje+zkOZ`jHab
zrpVzq1iXx2`_)YrM3CQf2V@`}zDMwo81V7vQoK%@AmtnbE(FAuU6N9?vkCs4#{|l$
z0MBBU-|}b%q$Z%{HlUn^{nd`aflw@9Zr0D3P{3#~`-*+no#eN(^RM`5gqUV+JiXD9
z0o4+<Y^NT#CQmz(>Fr=BxAe$q>49B9iy-)DVdKKOih!i8EhF4Dd`P>LL{4QV_T*xd
zu^`j!haHiB_WSde9s7hlyYM`t*Hy_B!+$M_6HLL06vm|W6|b1p{2e*e@Ah);N~pdF
zS}hZ>YBF8>7Z-sL5w73U#bX~5pDhpXXHly!!93Un&pwMvW2vnUG_B&*E8wKu6k2pj
z6&h-#^cT+x&i9e+YuI69(0@#=u$lo2FD41!U=Y9nZjqr>I~b&}4NKxrw2*K`sLG#0
z8<KGr9uGjaPuK_H4ydNx<RxhrBa$9(7zT3skOMva6yO_Y369O=7JV`AI}~k&Ma9YY
znWnz>dhn|e9nkhM`|`<@bV4snKSv|};Xb|km9Xd}pXt#jP*Z6Au1xpJ?4FNKXM!ig
znjt$|X}yh^2;;VcOE$^Fpcjh?I&~R#BRUwFmyYJ6m+Pd4N(MKAI_*bUGyDIQ?Zio3
zy)D~OKcTy&mG|juO~0zIXOWT${_3dBPkP$J{?>@$uY$Z60ZHFC8Psf3o=iQB-R{r*
zP@MJr`9auu$_!B!`>hdwzEboxDL<(x%t8RR4nPVc6_d1;O`rjP_Vk^GHd7>_GjCK&
z%I8*Bm26RxLh?EzmHEy4P0l9a#%5WY1{a+x+%Yv~8(1PmdW`dB-Y46smtC&pKW1T1
zCYP}MrMy3jfBqxUjpsI8dh)yWZ*q891K-*8sw3E)<A*jhCJkmS+2^-I9l~UI@97M`
zREydc|D_!85!~H+{0F$bKBTMbQP`ZTMz4-IelaF(a&A5knt+p~jWgTj>>TYFX!XAj
zfyHOnnSG|=SlchZ-OBOoYb@b>c9ZM65aaAX#e#{<4@^KptsSRQ;4dbg%$U>Z3-R2I
zsm#n3kl3dh<7Ab7XqE^h35}|j<hO8uZ;lg}r?<~%)j`c%BN%JrGmf0g*`F^?17Vbk
zKTC=AY1{ewuwZuJVD;e-jfaTQuli5F7`PljOSTe*QUf&EKXL^o+qVje5vwEo^1u^$
zjNrMG%2AfScQAEO#os}3nI+${TcuO*q=(w#XN}!DC)v*LKaSNIJm9_l$aJV`(;Bv0
z#>Ht+7T?K5JL)##<d~L}zvMGL6f5&77sXi9ne*?<GUYjw$(w=diC>MZfvkLSW{6G8
z6)s$H>rNenr^BQCKOmzZNW(e(iJ9^_-eWAo{plSMf@w~t@Wg$6M{R%NB9>O{h>N=R
zO}k%hK-p_#0hbwDQwbQsiyMXmL%tSIck}9A7bK!fW&4b_-EueYR)jnoR!~0TjxHtX
z&@C_iD3I*EEmMSb%_dNEm&($8+fOo_p30=A5P>Q*U#tWD3<yzr3I=B+75;Y`xV?7)
zTk21w*Ps|P8AxbgCs=F~13FN;f>FaPKz#Ddvpjb0RA3ly1Z{$@v}`JkU}X=`qDbXG
zAnM@-KycXacw%mt+h<iaTU+h2G{aF`sq1b~dT5(w@<Ke7ohvKrr|XxTd8;b^WATj#
z{_q#+`)=tX{I9A+*R}a`{k)U;zLHz;<QVo;=7#nJ0=@AK&bdRScJ@vrv;Y<>yOwg4
zKdH;U7Tcmc-1E5w3Q)(W-4Ng%!Q8ikbcA$+VYKOo8V=}Cmw+NHcf-6BTHW!pZ`VnI
zDVDU+PfUwf(4|-ylI54&oJatjT<%bj0SRd%*E8uD>>m`#HLgElLB~aBU^5lwi)1D>
zvPi;B?&-0G4`Ojgz$dP3A1%?Cm482iO{8i3e=h$Uf}{hvoh-m9^l6QxshS-Vn3v#Q
zSbnH`RXfjngpD*e(idS_$p{0PO*U_>KC_541Jo|B*o~GtL}=KqeWq{b{+r*VP*a;7
zNpD*EN~GXlt71DO%_-tJcl4+3Fp17|Cqy6by^uiV6f+UE-aC8=%z=-F@bBLdRO6MX
zzPQxn`~{W{N>Yr4$rBa~G}EfEw~hPrl5II7zi-+S{Xo6WMF-7E<lp0JSDv}ZRCBi9
z29Y5$Z3qFLv8;OvCrqCFvkj*@bLIY)={HSC2|$>s-(5i({K!6xKUJG-HN%}Ho<*~#
zf>2mZc($7Kpz=)xG+79Sxe92(#ws(#V6xt5*f5y6JS|6lq#{1xN}72+US|OMEhEBW
ztfj7H#m9kOKjz!^-$r${5FhV#Kh~vWSH5BmoHjq$ww#bYLo`0PFKR8DUMi}$P~f(e
zk{pim(_O-7T?dV1H{+B!5~T))zq^wxFY!pq_8W&*z|manz`&9jxurut=vvt@lel*&
zLi!lZ#WXw?<Av7;tLSWRsiqBSw_3o?S&70fz1YtlN)V9?vsSw})T72NZRx0!Yr$mo
z9L*Eu2^5?TeljRBL2h!sD2%$c89Arbv#lcnkhinqYbd6(RzV<;-4}~^+$TB4MiNY*
zNSy6l(_`wM_XTyh2iXbUBg3AT;25J$@DK7hZ~GhA)&S{Xi(|Yh8zt98&`|c1|C=3&
zBpo16X&G;%=!#}L#4dh7RgGoRK0!0|gj@yK_2|m8D_o$ka#h)QgqY64fDl^9gh8`;
zfHeO;wG#cM=&RV3r~tqX0Xi`NPFzeAfGC&<Rz4xEISP+h?P7~%hgBu$jY(*mlXaXk
zuJn>BJu9uta&L&Wy02v&44y1U>AXY2+g#bLNxGn{Y~H@PLk0M3?r0F5P*kIeD26jl
zb6Gf;aR_Rz&E;>5{~=T(jGgtfscDc5&xei)j>8y@?o~APBhb!)5A)Tr#e>ZXS$y@2
zgpNcgd~A({L~)5L%Wm<lyD7*=hy#P2NEYV;UIxPZ5D;9L@05BQyy2P;<HQu8ED&+&
z)hr#DD7*B_AByhEt8d>qDS)ffd&`FgeHqVEe}A;m3RUjC*qX4-DD=HQbD+whzxwyf
zX>MbU_T_<i<U;$0FLN_eT~BADBW2a473iG0t9NV`2dfSm#E<Ui;s$dMP*F9`MH+6T
z9*SM!3|6387-K%IGD<|9BRcAMM{8P_t=eSieyBwaWy-2QndD8C%=J2n*Sddp<F7+P
z-O%Uq?N2+BbbGIF8?G!Y+rD1{EA*G<ZFe73(mV;co(^^oZ?K<dwIihD2~IdAUp*>S
z8z4|&vBLfQ+1J_v)@)vM=bGbW7!mq;s)ZHoN!9jTm+t^rTZ6YA9WGz>R;Mf*ALnxy
z4JC3ts4EazD}#U&+%fj#HIt+e)0K|MoWRhUqtL*&h>;hLEK7(!A?RyESwnbo@90FO
zs?AXXu)*UzJazTBtxokB)2U@n^Z348;aAJo63vU&6kVMS6+99yPPPid!Yb8h;-CbN
z>8RJvHC3hOYnS-rZM2}Ygf6<W*mN7LDTE$~Dr*J{2$pAf|CZ)Cn2b9rzA!I|mur_O
z^GWWJetAWHQUZB#sv58CCvET!qhhx<)F`g+4~^aaWZ`7HMqardqFNwCIW%JJ(1^ax
zG&TJ#*GNh%Ax2W5=}Wrk_rKw@5kHo!``m4=*&aBi+3`xez@GmbyUe}{_-gRiJ^G^?
z@#sz}we+!4Y@ErnE*DQ+$eB%*q)Sj$wm!9lBgjE8u3!A1Z|=VBGOg>m_irPJkCW6>
zxXxarUMI<Fx`C#eu87wqsK|jP?6cNbw2ZmDXYR`1L9w&D*JD;__$z<n6kSt9>EmPg
z-I7~kJ!)bY>!@&gx4DnS{rc@sw2~JspEiAqF%eQfd`O&FPM`M+#cW*gpss57$Jx(J
za*d7**YDpc8P}?jV!QJz0Q?jDoyJFly%gyz0Qg|*;QQ=5G34UOt9|a0lef<^-!bv+
zTxgYNH41rmH9tQQiuSD|8#lyRBK&?yC+G$4>JrqKqDQdwVu`U!0wwvv!!>{y_R?LD
zn*lmrLrIZ@Pjw-|HRJ;<LCQoZn@}de1}hd!>4Px!%d?>W`J1&V7&{|7jWP*NDM=fB
zVwVRY;F>ROB;8I<bj6<+n*9uJ7btOax%cInwZ$6#LjmS*i$Fc}mp7Gg4OiPilJ>JC
zJLPsZCA<-8r$kjr>Q@*ZWv8HbP3NF%K=fvUxhW$9(bR9p@skvk9bOu|#1%lGh*sE3
z$ueuh`XLFBFyF2c2&~OV@eoPRb-TAkpiK0p73@bD->XJr&dRw$tG}vJi)hAZD)Xp|
zk{kvBg$*6DR&RjAmkdj91U8RIx~D?GzF0egEgI+^a0LGoydAx$)<ZQ<Kk$Vr%@-Ig
zJz+*lLrav_3Odk2plqWlcALJZQl9p_-sLwcmt7Y(mnrD@U=gt)yE_^m;txMNym>S7
zt2_(yq}mmcErg79x9sXpgsAeRn2TeaMs`GMw`^^PP{@5Ch_~{i$nrk2^Mfq6*s`1q
zIDZb!a><fL6%)KYa~;hxcXqDa0_=i>(mE??HfztO8ACDyHHD80!2!u<ftS3G9w?}$
zDKW3gZa)?68~v3G7A90o&n+r#H5tz9Q^QURU1z(|Dw#!;mzp=oITM(M+k#6xNACsB
znB6dKC-;=<GFCI}z1Bxo=QCj_Y6r5>cg0zbMOvr!Ao3MT-yUZ86$5mWpEf<cG$-R<
zW!126WPvu%*CSZ-UYCGRneE)sZ@0|IjK?vhOW4H{s?-AZop#g6bkAkX9PL`M{x#+z
z-a~VeL~Yjh{hjk7OZ!CvexRl3CiSl(U&jrD>y=Nt1hdllAHTkaX0VAE1-r^TYdY#L
z<hMzRSb)$W5-ilfHPg|28sk96d`y}rowhNAW;Fr*-AVErQ93S!bFQIuc2Ji=zLssq
ztZjdU4m`}JSWF37_EAX%8t8Iu9=5`O({@0}Z^rkifHtoTfY)zQp!8i1e36t7k)3vf
zs|+UOfPh}p4?So7m|^g|*ea#a&C=@A{6|I!tf6Wh?(Vg#A9~w?g5<HN31s_MW(JKt
zt<K1zGAV{zZy4qGva&>Kb5q?Ax&*2h52@zEhSIvbK$oLNG+LKG=^eEW)SW=RZnWwn
zJ7Hc&%apR$F2(Pu<i?eP9eww#^|omgCck-0S=|3!T#b{@Z<*Dv+HaUY;?PTnWs65-
z!B-!7GD;qna7p}9wJE5BX0dOW|2294MYO37V&);L9BFfZ`7qlAu9Hun8mUKf?=|)r
zC}C7{Hz~}JhtpEY&mZc@B58Kr5#gZSW=OW3k0*zhF>J^Q&9W3@DCmX|A$_(Gz7b6N
z)?J;!d*pE?k}77#gnFkXpXB#cR5O7zC!5w4O}pQx%?Em<d&<VLf<basPMArkufZ==
z^Ena9|MngHn_O(<wTTXf*%aRk@v-i+CJ}j$AHn;EPLHa$?p7Vs$vU4s$PZRz_!1mC
z9K@g?8GiU5FjlH%uy(~iA@!rC@ih&5Mz0uDaLlmj<@Z-_!b+oVPW#hU*m1WJ1>6jV
z@z$|5i(KF0t`IxEDLd`nAvYAUjeE<KuX}7t7Q!dMv7Ced*MV2b#BR5Irly3wa;3_X
za@~oljHI{O$4-8_1ZO}0Jw5qT0y$ehYC>?HNd8V-{LR1W+G=dKc!d7$lM!X5_nIH9
zx6Qa$ja*V-#sz@oe&AR58`Y|2wZ=>b|5NR<kQiw<iEZo5<&>ErfD13WpQltR(yK6)
zH<CpWrt;DO-o!mOikxA&(HVHB^SQ>@y<wpY?@6mGluu}`9@6oz`Ms;j(NrCV$@fYt
zf<EEop;NqoSTQvP0DB39<R;WSs@?qU^6F}J8mHPc*I*>ueX&7#WDKCx=Y|Z;9{sr8
z%N=&R>`C=bnH{3+W7?Njiy?9Qxw{No8tg^lEk%>?Ut~T%ztOi&eQP!So?_CJyz9A}
zxPR5PEy85WeYBrVO(j(KW}pGX)SVWc&^G?wwoPb^hU8zPe4Fmo{?@q}hjhT!AFmo+
z)`lQkmveeU|F8ECt!6)pDj7d26TL-&dt;&E_0Eou3|0`bV;d@1ay@~?=(iHzbEb)t
zPrUBVTZ{Tn-Z*+U++yz<@Eqfu?&4}>EO^HbRp>5BU;R9RS+R%d>74SlVpG{Da)fMo
zNv&{*b%zHr;<Ec)ABT$Tlbf4cdYC`S?wIo@+D^^*^Catwx{>Gz;@+0o(y7d8-wlnv
z#KX`e0z2I-gbKa}E{50S2`8i7!>UWu$_~rdW)eL5?Qrtt>)F4aGghC9(z>nsV~%bq
zlGWAenC}t*r>fGv5&<LWsH}`4o90;@ceM#0DwN14nk;ucDtr9(9kr(*n*Hjy)B9)r
zhku6?W7SFVmcoYx3!#{qW(!q0-@n6gg6Z-?QB>LCqJD}IsX!j@JYKZXKcz#~fc%6V
zKJ$E*svV#W=B}m>1vTjo+{lfoB;!XKNJ=duIuh*)IPk(@9$%fgrobve))gTegX9Vj
zycA(8EebLUZ`Xs8;YCPIdYYw8j3UJ%E!m;MP5g<A#cWaCss+4eRF>*95u}Mk%>4{9
zMr{wNz%@tVgM6~bYFFGSg2?hgM7B9nn}2@4t5B*%01V&uk+OYn#UK57s0IWP$Tin0
zSdQ#u)}Ui(LC0t*dPGTMWfIyS+3`~BU!3@e#k&?RP`sYe&zKT(W_B%GdN~x7=&p7l
zLWklDj219Rog1}FI_y=Qg$y2mDfeKt2s_L_%X=r6k2edjr{bC)@{?DM2Ne8#d?+tb
zfw_xzha=hMkcpd(uA6P6u5YL1`^7bwg<YS18|CzZ2xpP038HB+wx~4r7I|h;h9v{)
z!MHTJj!(O)UuA8-dgpu{zw1-&h2NRVd=u-XNc5uP4!^MqzqS1(`BoVl<ihXyhc{F`
zKb3>P=>6<A&+O~SlEQ-cueFP@T31JLzJ%CG7fA{h0P#|U;!Vnv1S%sYQ@;^iR$yJ!
zN6gUd21_-oWB#ySv_FQ*fiZ}so;355+MShZ)h!{@o$OKb(vY{;YljSXDae8Q4RIFV
z)svM22l1`x<##Qr|J`Z^(#ia+4)tv5Q%*6ybpqpeTxC1&bP5n(JJRXNj^9ZUVK^+T
z8)gOWEh&>^RR|j$5oPiT{?>IgvFyjomVcx&*R$`=+TBM7pO(>Edn=P_g_)7|_>NE4
zkV<@1Wfclhkk98^Xz1|Ov!h>LA`UU$n0MyUGn`LkIh;$RNi>oluvap_>+;W>WYQ#r
zz3<Lss9Op+ZMAcTt%KuFV4*g*MLCGz{2GD-Z#eF2Wqg#nC!|>nPFf;tO@Cx{COhQ&
z03qMse+nx>u<ni_ic)0+XyqHQaa6{o(%xP0F%tcB;oBcJOe$r|T?7W?plmxxiq_)-
zCT3Ys`#IzCh7CYdsf?@`9FA~E@@=sR-x@9&%LhoJQgkSCGY_-RsUzh6Xp%yrWk(?Q
zcU@*$K;+q$IZ}Fs+kW)nH**2AR$zb2fQCua91`0n&56IRK2`d8$vdKE=5yD@6ZdTk
zW5IoXl8Q&gP;S(n*Nv;K&q|Zl&Hriuv=|fVzav=$0gCKASg|M#PE>6*nt0eD0uW8M
zPy5)`Sm*$-!mkH%N+PP`!1xY`lHKX!<!pz7wMWimQBH87rq4-{P_$(Btk4DsFBTT>
z1K?K{^x*j-aIFB2b0<-c8OwQj5oSU%;%lm697@(RL|4|*r=9DeB}1*QB_*1CrxGvv
zptw&~l=C|oR(G)=qLC>+RPMce?U(xOBb?mRp_u5xwb#w@zJ5v3K|M?Rg5&vrF~>)*
z;P1ljx%!Gs#gqWf=cD}WIgy_pKJs`&_emcy|F2X*2U>Fa(enIOS<Ic?_aY5VrVX<L
zJ*kxV=9#T)Z#`Qu3}3zN{N}opEFZlI5ILIYPbq4co0-wit~}|J(phZigl=e2_0vJ<
z{hJG*3P<NAq!lX1VjpG6?X9u_r7Yn1DYZM(`ng{f?X1CFozCKuvIZL^+o1YXbP_fh
zMY3emevwsJbY~{+<}@{e+~j<iZJ|?d^Bc{FcbD2bhYUwgT)UR`5I|sr@VL3wa4#v{
zNS>$e#-6>2@>cKr=%GHP2$vzIZ55uu_k-2-J9#E90x~}WJ>J>=Fubjc-*`nPd1Fh*
z?TxW`pKRkn|D+(ku=+ybgMOI@bteCm4DG-)cdiXG(B|b)#%bl>Fs(CFt;rSpS9!&(
z!xNLOkDtJ5zV3h^hmzxbsKAn;gQ4)r<3<PvSC6(o$lDSwGnFurHCh*8`nb?FTJtR3
zxWPMK^rTa7dog93gKOOH&vuz?<hM|U6*JRT4~g!=NiE%5vhcmXBR0yUTb8v^<05_O
zAhOEf!JVM98>bu&iENMszle1K)ike-@oVZE-}1DwZ!7eST_35wzi$z{xFb-<n?{p@
z4_V{ssWPM58O+kul%S{DG8J%7U97b@NJnn#Gg=#yM|E92v(fvt^DZE5l9Ja)4O&2D
z$0-rQ?W8N12xqRMWqgNIRjh)o=D<44CO-_&8eo3dE96bi*e!Y}w&&mQ(fFXa<HxLz
zZkS9Vx?z@U8p;h<e%k#ys5X;H?j(*SVbr}>1+luYQnTQtZ0bpOJ+0F?Pj=fKGfAoi
z6@1NDbLJ<Z{w$BKuh9keN@I`T=p$n`^h6kAi#_U4D9@oDq=`sP^WnSR63KFVH^J-2
z+hXN<2jZn-T`MV#T*mAhy4|ORI`et^J4Tdp%Q)04DK&!g)EnTF>@Xn@V3a|jx#&SA
z-!!=1;0l0?o)rU~c0q-S(%DxQfcI>7zWnSEnxeazE&pa^w2bW8$mbjW3=0<}R0vH7
zV^xqspF!RVdkc?^hQ=hW*+_>~)Kyq8!GoHF*9mDJh4geA9)?R@6ZA!^t-(KSa=Q03
z`07w}JwQY_?jZeo`$b3<Sd!3z`|axOpY;<B0SK_Shvo55vE1kyfK&QR<CM<LabPv#
z#$WZ=POg7uQoGb3^T&I!NI79ek3PaPn@Cz!=NbCX^z^&uDw<WdTbRrIYf2(%6sRpP
zfe1pdzQ$}&M1luWS+ZGf{<c>~-drP&$SENBj|kDY?yV(j#+1fO5@kRpUZXqjnijDe
zMlTO1+59z^I@nNems{ZBU#pjXmpM00eXa5S%Rw2{Fa_a3Ezucffv<yJSC;>@lL-##
z3VoL1AFMAX!`|LV>36U1av7tQy;1$z$=bq4+c-4+SJP$3^yIRGHo!$Kp?_zgx~7(=
z{-xU0Y3REYwRZ<BlBgCsh>nl3O)ygQ2HTqk?NHgIW<s?j!cYV-6OmOL@Bo};czZva
zZW>x-Skz$CEM@L~pdGFh6xuSW7v3KGZu1`ys(~CXFIwH*{}UOjB0{uHcNo6Kx8g*0
zUME!c^8a0s1{7@)348hNwv$G+$4gJ#{=m<ftCRXfTTV)dcv(At1m(8T3!H9u$;evq
zCu=pGh#Fdo&z;6ACaPyA?;&DF4Xp1JlVT<~<|I0toxS5QIYtcQQ#A^Y8>D#u16W1K
z<ox6b8d36{%I0t8AzAk--@R)R&YU?SO@wB*7AW<}3xESk<s~e`n%>QIun_iI-<dxr
zB?+ul*fINofLOs(I2wsSEfLy*+ww~VoDJ(*qCjKI7G7zvF&S7zb67|H8+qA?1^Wv0
zM!NHX$Awi~*k`98Vpzkiq?47M)>s%0ur_ZPPIMB$PRBhrs6xN!^U8iLe*@?%P?eTS
zjp&2}UNB?G7&hkbJ~$2V-$bA+yt`rb1n5a&ZYB{D$K6_Ri<CcYRBhv^)bs!Ewk$FU
z)9XsVu}7?4&Szl9GE;g|zL^a$jn)b!LB@MQO!SpTgZ?$KT|TLLXA#9@p#AICFibqy
z!@_zc>sOyjNr(W$+x%rrr?>yasLm1JpI4v!$1y3v{P}~V*@woY(Zug(^Fxq-0|n&o
zk5-h2IB3m(P(KFFLeEXW(uJ04yzp3JYaI)z5)l$MBrR}AAln(npulY;bL&BNm_@c5
zOV151m%)TG2!sf>aPvu*j|gr<=o@TnO%Qxx6vjtP7|z`Fm=oD`hh$5qLp}5)zhB9m
zFD~+D)Hue~TyY@wkL$_F8jr7f%yIdfjt4w$f7-X!$VYfD3}vHj@!|8+k81v6CMg${
zyMD)kWZgOtp0H-_Qp5ciq~U9$=>t<v6SfSQJvMBd);XN`7iwto4yAWCBce;#a@#mQ
z4?DPWgjdq}i-1**_zfhk{A!bqQ!Clwy}(Pkj_*aED|o(U>A!g^oSHlY`=z^SD4mGb
zcJ32a=+pyz*^UZ2m5C7aQTfN_DJ2h%;)b-y!5y5gz}{U)BsYP=&{vi5I9F0|XeW2|
zX_g%~C=}!lHuBfShZM?c+#8a-^KqrFuFL3)E{^rdh!DQDm4hF5U4ERhY1xA;B(SV_
z#-nZ^z-Q8Ry(u-6dh-gl{^3dNsP~QKgq1e}CA%n+^>1!---Z_{YEF7rhIf8TtDRO7
zYxpu1>0kPn+lTl*=GFMu?c039eLTK3{wFxnI6-qTF}&2YZe<yJdY*;!3Vx);p^`;L
z;B5GgdiOnC>?xDS$NY^?UH-ned)cn4(T74#+Yet$=nCRj(efvyH*v5kJ?RecnDX>*
zC~Zf{XoV1^@jseg%OCOH80%I?_LiKlVf5s>+OicF;0{Ysjk*Y&8W_n6)$>&AfUXLc
zw5NnD+jsBQ^VR8E3=0|rHY}Y-T&v>vnDU+Dw~dVrEWJ4HH5N_lKiib_Zqu^2QEyOc
z5<?gnZ#O;I_Ukz;Bw9BwyG<3PuV&mFcEOzYSYlQmALyWX1xnYF8HVy6(Rb3V{5n1O
zvXK}+Snmmc%MNhV|7?@v`?GvET`_Djqy({4d2?BU6<RmUV*AiNG<@yE?s`?f%Tv)+
zS`w$=#gbQ_x11NY*d5Y&YHYfLO;S!%0eWrXxO5n;BEHEd@p^Ef-@iCEw!Ob&UK2xN
zE1B>x5tE2@>0LXQkq=W$sW_3&L649=3qwvTEtVz1Vzq#o5=X3i+8Z$~S>y9(C7diw
zTBuikL#t8H&*eATPgiY|yx7T2Rx0na7Uyr{+yt48M74YrsloRBwfD=QxT|0HhyZ0M
z8q21F)fe*3;HyWJ|A?Z$o%V}r$##UAKx~zQ1=iika5^l!Tp{x9!t%;~A|M+)VL>;+
zNQ5L0Va&2P1A>|iF3!Y)^8$C9#H>2-MOpTtAn^7TiAs$yGzv*P38bo~cKVLc&6i$y
zBL!hUC$^|LgDEpA!P}6OmVAL#DA`hs4oUr*>Sd^=yeD@dT2O+`zYEr~`r$E|H*XS>
zi7{k}y<mE$<p`ANkpy>hxu%6WFl8TtqVke0ysR;obUz5gKlA)(0_{ygGO+VLZ|D;>
zyYXH2cMSmFC&D^%qWuWrVJ2jy)Y!E;HQgTx83N#ml6j(q_Y%eOyfGM33p&7@j7`^3
z>&tHJ%l>_Uq_IZIMZ8D=8+kAyB)|D&y?Xt@9_ep$Gu=BmUv`4x5shOV*DEQ%%j~;7
z5yazKJ+7+D=q`Hy^`^cEVyaMmOVCs*6Qg*YS^cF;p|{}<W}mz#3OCaB%G-n{m<^lO
z!?W?=6Rta{X&rf@cyl$MMlC0<6c?M{%o1<L7vwNefV5hvLIwy@=G7=Bp~yAv{9?rR
zih*B3z}(B)F0CfoI-wL)5ADMnz+!UXX5=*k13&lmwmTz3zvgy2R~|Ul#*)1HU~x`u
z7=P~_i&K&%H5GwE7ElC>wMe`Rz1*>#3#Op*U1+udX?#pKca4NL{z-CJ&e4g)dzhct
zeT4XR4|#mEE3RQJD;8p0!T-VYQj}L*PhR0uZvo<oJ6j*?l2eiJj{<AU0Zo{=i=R%S
z+hwBYJk~jzsI!PfR87Tt6U;~ylp)*U#gkd;TmmOT*#b~vV23-&COUjDY6D}+ye7Xk
zhlZ_``%p;hUBkNr&Cmj+J#e`TutMsfV|dj`HUb37CQSOSn(1#)KIsnIKq#uE1yTbD
z$@pXJ=;}eAF5~F{Jaer8;DsVYiz*DLfkzVA<|m|+1c>`u#n@Zy{5&Z_6wP(Ch(jlb
zcM&K~K;yK(J{CU$9Pe~!!on8OERx1^MdTnE8`xEOb5xiNZ2BWHW(}_LJLiMcPg}fm
znl`YtrF6P=33h=54WDM_kPB3T<S3ZuA#;9eM&%v6^H3`~V$ielyS`|LxQ8dbh7IM{
z_Qj6I2fsG<too%oS@RGg&qdrJ{{iK}MM9kT==^F`PLKy(s$GFlAXF>FagS8uR6D?1
z-(M334-NGLaI+(6tVeyLxLqjB-@wV`+>ECqk-`ubLLlph0_M6M@*ulr=-Lb@yy+yT
zhl(4K;j3BD=ARXZ*=F3z;79pjMW1Y`3=rk9vGYZCRGt<ODD?4Lll{_nNc7<MRl~-n
z=xDQV7{uN!)RgqWMpC=VP8=D>jeXaS`_<+!!1uJ-*ZMc}ZtdjMWK6=H-lU^KAH**n
zs?ESqweQ~=99PdAozzC-y@Q?g4$>bnZZ6UL?dJs!gfe2j^Np1=MT4gx!>?@j!N<})
zl%Xt;n|&xlcW@Dn?c@C|9wdQ^>VB?1YS_NHQOPrFi~Y`ES+2Z1Oj6&yg=}GG0rJMO
z^W88&)nnAHon_0!5*fW_+Q8Eamm&DXvm7~bdEiXT>OUa(6fVU-ew_KtP@yPs)sih<
zDaBqjDB#xYNXE5Fl%QvgpbUWcsINI#Ueox4e^d&6bzmlMdIfJj7N=9`Z2hWF_EVE%
zH(6t}FC_O>O-W6YW}!EqMy=og5E-w7u?f%4IQ=|@R0*-=3goWYk-P4T`ubeXXI}70
zLJ7fD8*{ueSbb2LIPOoY)v&SqWB?AM-)p<EtuOPF-t`?wXh4R-mTM!f6Lq8SAY(jr
zXQxNT5%Xm-5B1G59{%Wk>AbRgC3Jit0{NCSBr=E6p+r}49L)M#gxi{-CXzU^wVI-d
zLMBBd_!w4o78iW#g->udLhN2(#EqK1VTBbh0jAskI@9lHSVh+T9bya3rzJV*6iU*f
zFOjbwl`QCPNla+A{Qmk4jN;JD^Kr+>xw&9#C53KmWZddN!^gy;sfqy&E@5F}R$o>N
zr%y5Li{fHV(Br4!bpz1g_05p;a4~O>MW^0^P0Je;Pb|HPRFy;e5tJpnWqUO=Y=gHy
z<lJ(E_Na>T#QEElu)`l_J#=NO9;q3&=I-)&-rN$kbZ*wzeNyH7^n>P;FrmHQ-9o06
zo5<~u_mzVFoX1{jh94RY<0Kw%Q3xj+KD9zsZq(vnF-EPD+6Uz!UF${64T9z5`efRH
zkzz0UKB3gusO3ve7Me7r6IkX^ly=0v4r3cK>dB&*EGNFEdfO+uLe6N6VTjjA8KSpq
zIk_!sh{jwI*Hs?8g9)S(U@z<U>4xT(L31^>)EW1OQvZYqVjM)kCBV*rz%Waj^0WDH
zQK}a1MKse(K3F%wX$e=PYGXPM=1j_(2ObshnW*g{X_e*ZWp%N5HF*M*i#{$Z{%ba=
z7?Q<z#CK0;5?TQZC7Sa4$jJ9WaG@N?IB1trlyCvuIori|DhgAi&{-hr)QlgA<@zq;
ziwWlSBVZih3yD5z1E$*GKa4_>O5bW%c;nz)h$3Nmk$o}4Gn9C@2e>2$Zr}7Jp7u#1
z5$hx}ekAPYgM0{JX9|4{1vnq>sj0-}IwG0#9?Jv^{Ncw_%1cNw7>5F=6*U3ue&3x(
zzsrP3V_~7L@snFETL$FIOVBdGw_P2SCm`YKlhk}n2SRk^^CWD8*2ro(C21YidwBa-
zs0p;L=4D&oRZ^IcQhuaJ$-?IvbaZ0*-X6E{nyuYhr2?0aDE9A`x{y&<h2R8P=ooA!
z%ti3o9ZP<hkrIh8LK4uu@KV=ZLo890f0B5!<sk7@ch{D&wV?M85klV=s}7D&f%$z7
z@M2*hY$<IeWYb;bqsm7&3Jtwab8q_8F5d_F{s(ZUdqQ0^qBNryPL%#(!%O#^{sRK}
z62=N+bgN(2vB>;5%~y)QL010oKFv?GO6sPjmHkL+Y+E0^9=4G@(R!OuxRMa+BG(bs
zZ)iTs_n`LHNSL`$C*Gb9v#AN;xg(=8gB`!AS+YJfQN`Y}T4+i3Cv=Q<B3OZ?e(^ne
zc(_cf88?#1s3R|o>H7-xdU8$l?LwC)V|b;dX?HWVv~6DUy_san=40;ag~Z7gIHOJ}
z1N29kUL`It8iArpm;zQ%frOKMfEAeM8f;eB+Csh*CJQB6*Mo2vaH<^4QWWS1h87eM
zhelk-14}v%^4zHlxHGVPn3AApI4Ssi?)yYON!KRw+}sZUzxJ@mxI-4lN-Qmr)a$6A
znaf0)8h1qSlf$_Kf;eNin`-bOk}`BHW*3#6461<n)_%Ze!0#b&tzSCJ=aJ`H7LcT+
zEzCW_C_Q6B>P@quQ}8a^4^dAb2abh4&gxdJka2-(NXAZf`Kx0)oCJMXL~0ApwEB;X
zZVA{HC1v*Q_fe^#j^C#02j@vNZ1xK4N@D}Dl4EolPd-ocb_%~Q6K%TZ|8<?FVy@gr
z`PXXY>IzQWazKxVn!lT^PoC>&!_569Eja!(EAE0ek@LC=lI$5T(>lE+xCv#GkS6K)
zZ1uaZ?Gqf2^ScJhq+>+{Kvho}j2yW8gXJxldP6C&T<ZW^_zXCW28$Cei}`D6I#Mzs
z0|gMqEDa>dvWZG2=m!Xf7ZurPa0J_X+;#YvO}vM}PCB8KTylTL2*JU+2LfE_smo$K
ze{V*|^m3U?f2`a(zxBCl(CZeu#i}Esu%74o?$|e7eq?jY<Lj`0^s8^1wdb3t5k~vv
zMJ8SZ0m#3;hYyc0n>zAtH72IpgzlxDxNeIe{qd}|USaK39@k2weg<Jd9Os878k29f
zFqOp<qa;2D#6NRLhoraQ*ZTAjtBkZa*M4Wk3Ba3;a^t~Z&inAnJ_K^W6wIo<0ipex
zawlsRUybK<q!|T1oR3?FhmgO_&j^wP^D7~HJGxQJigOhzRnJeaRZeH~2dY`;r3_~|
zc(~i#>Td|WHP7k%==clM&xEqjv9y9|R5}Xf{?0VwFvximC7pdYp=O|Z=0CvI=Ep9G
zWwVZHy<6a$O>fAH;B2<xndDV08*;La<u_BH*|N=AZScyoVRv&FsynZ5SxNY71{sj>
zdiw2Qxvbtc%b`S061d<CfBCXD*EPMi^3?)1GT9x{|4lPe^|i7FcP<#Uu2|_Hx#$bk
z%6-iG5-XQTE2y7P(F8aIFPZ;CB%}7_1T$Cugx>P8^B;?n)~-=D7c?C!M-6_ogCEeg
zd&X+uUmWq2@oGO=sQGt0?rQYzT*|XuLw`k!ce?>FOXGR1mQy&!l}hk^#G24oI+u6G
zv=fs)9BSzo3Wa*I5_yTpK9<{pSQ*k1q3WZsNU@QRv`VnPt(2)RN#E#L9yW=&@Z7c+
zmv!uEw~aAl^C~pE87(LLeTJt$%cWWhRXf?Euegt;vIQR|?cr@BW#W)?m9#@7?XVxw
zF0V^x+>`0kldGP^Dp$jmsCoL;hUML-Wab#eM7v$Ar2J}@M12h4n$JDzbd7D@{T8l9
zLj8JIzVKWWJ=yTLW^yTof$P87jt<#)sf?x;6>;LBuB<P%I&DalzH*D3GtDY&iYRAe
z7+(q;6xcX%N9>Xd(JKuD{ovfp6qg5Yei~}$jHBawUB8wUWOP|fLfNkXcL-QSbNFV-
z#^4&jb{3hBtj4v@J0i{xqW#AUL>~5SVc6QQZWTeeQ_)}!7x@y(KJaKwPh}6b&bC7!
zl{VzvSt@{#J5~QtBdgm0`%;RZFWi@gi$L!*d*qwkpYLXT`*<NK_{`IS1s(8`NPC?&
zlv_6IVIHh%4@oETzF8PD7XNKcSmm-9h|J31X_E#A7J>6xvkjmp*H%1c!~{YGWJ7k1
zL;ZO>7C~kSMX_pB5fhGTYbvqxamET1zAAVijd2oq+BhIwKy#|)-_DmHKfp5BA`H^t
z-t(wkiYkVU#TvwC;?rnu2g;Xf$v#Fsenso%ds1r70C)4UP7M;3*-1hi6cVa>3<mKX
zc_lwIK74s91IzpC(WM?{qVa%hNRxEcnl+-Zq_=G9qgULszV+y|_-mCsxuvs_B~cEK
z9w-mTCJD~W&2|fa`6JR4o7Ovaa=mux)Z|01oxdeOn%1jhBT6eGNM8*^eUUF*PHLBy
zmNGVTpG@f^i&EP#q9!MCEUQtQ0c{2Eg2KHMtFQOCJ$DfEPw2?ZNov{b<^AfGj=)~*
zT#9AkNBbW5e@QvjB`#I4$}mlU>S$s6)E_o|=rR7xS@~lyYr=e=CT$BJT;}8^d#i;G
z>k{rU;z>^;zhhg5)7{mJPh%^B9MW=ekNoP4+lcymKGM_4Yf&kE=4r+Z&TP>v5vu(e
zY~U|hkBn<Sw?CBv=AY{UkSrhPfgW9;)R*XQJOIrdN(8#Mj9}2As5f=^$-Jj_LnPD}
zv$8tc1}3Bo1}PY`N${yqr<mT;h9-v)$f;Fd?e7^1!xX?p@A=6k>aKGl9RN?c;EjMY
z`gbqY7m@-DMw&`dQ)pVed=Uxw%C=i#nTeShVib$%z7X9G%nssCZH|2s#gV<t9Wzx@
z`j*o_J6F%$y^$p>q*I_tKbr(YMZT1*ZM2&COXq~(VuhM;#FR$V2oZ~(s_h^lT)8&S
zJT*uQqM*DnbEW*E9=7@Q$sts?__~Fv1XsQ&QID<V=`XOgqESk;L=U#366bDqLcW5Q
zEq>&{$Q{negwNTSC&HCfQr~tj!F$y4*=F1f6p3v@iCiXt!Ur_tztueDS`pw9rEnoa
z!t{AlVau{LrI~<wbEHfRmM0K8w1+aoLkMb98PjDDmbDtueo?w4rBI7&cS+o3QusDS
z^AEABF+##kT(n`sf>euzUt)F?<FlcO2gR4D%=;6^UiKk`ssk?yZ`aG2$nEpopY@T9
zlyI2vKbgsI)6)aT10z{^FV>GThg-jM*1eG2DhZ)ftm64oF3@&EQz%^J#_Zu<y-&nN
zTpiH_O8Bd)$Ke{o==i$ma_K+drR#Hzk$ZAd8-uxcSNe<p07?RbLS$1XOFX_K6WE9u
z!}(_cp;0*9@|d4<uy-~M)^=>$NwKG@o!cUiQ438nfL5DUKLH6;NbN_y;<WrG)jU!C
zYC-#Spo~}Hd$>oK5=s&bwxeBjIe}23?Bugs{IDB_xG|=~fTVwwrE_9q8Ja_rsjG{o
zwbGa1*6lz^UqCrdvQzMU4CHdcZ|{Wd!+SKdp_0b?_@8K{3dScVQ?TNBc$<@m_Tl-1
zw|2AkTZ*w~R^7^dH4n;VC(IyY%Bp#-d8ni*0ym1Yz<YbtxFf4_=qL~))ON}v27a5-
zQde9X$m*}vJw_@zuP+UQuv^S<T!aPZ>mve#$mrg*>8NmnbE4JD6ISEjEZt5~xDLGh
zMms62b9k=i(~iBq{4ExzCudsWc`#Ah^>6omz6QpoCvmYuQ&ndnS<N{rEerDe9fIG&
zcaHMqoG0&j-!7Zq24!$LTej`VAzAsBg9m&D&NIZeI{w$?Ugb_ttPSo7QBl-57}mG+
zhbSC>vQ^}NqZ+UGrd^jH6SQGuFFlvY9MgU$clBSBQj#t<emW-GwW?q`U2x2EjE`2k
z<fBgJ$!aNYcl4!5hv+MfJwxSl5TqH<37>CJscymCfQe$|3EE7H8)N#*amx38lzf6<
zAKXXaJSt3<>817Z@QR~vF`-WJjg9|IUMO*f|1;5n$r`9H<3~s*n+0hD3Q7ObkL2wh
z6+-+EU9kIl;C3~!t+{`Z<f+XgBSj);zk`)12rfH-1F>s3`IbC34+AG@_Ljn`74$L&
z$<uyZ#GtvvQwHJaR2YDxX(=E{AWp3<CHv4!Qu!MiiU=c;#^l;S4OY=4!vY)HrUPXV
zl686!!J}mi0Ba@-AWhs+aKr}Y1K%pxRYnxi`FMH}l3Y3MM;?C3MzSeLYPBv2hN0Bk
zH*sJ=i#{{Lcf$Ph4aoQp4+rtw3>b=2iNXRm0+2WN!DH&f0#5Tu4y7+e$h@ejW*8Je
zKlJ38RYIhVKP*pvV`DDrb^L54K75@P+8cv^OCT4d|4~wi&tX86ygDcQ4BM2;e8x3a
z!qtDzSRm<O-&9eSURU9xmW)?CqsXS{PzUWj{=;n%7K4U`MA4Z7-!5THp@Hx0u4|WL
zKLW*hgLfoHMc+za`U1@A)7dXa99YzuPx!erHz?cO{62q5dsr9Vi5bL9u~|r%@{_EK
z*}}29()vj{)huSPR=5do{^LH#4YvGtX$sL~LoN*&#flLknb0<#FIE?^A9a?6tV_OJ
z(}~*76Fvm#&6Ei;8L8)uZbv1wx@>1GA}1`nLO6!Dea~E<ZJi5oI{ytT%bAV$m10c8
zS(Q9SX}|iDk*LY-CvwsZ2Zf(^h7uX;7fb9!TM(s-2G~xhFMtWw=&;&<$TYEtU+%>Q
zAUu$SVOg%C!Z3?sBPrh<9fPD0Xi}1XOa}w6Ya%m!@RlBnXtyu|eDz_kSn7GMN&&;B
z64IEFYrjF%DhpUs;RJuk5$8LeYZDASBjW(fA{8}eIFP(M(_<;S;MD^;TddG5rf^P#
z^T7s;MrfVc31kBD-5W;i^9em@Mr;-U5cwE-<S=rl!WG~!&$+@;$GGl03cbRn0M7^2
z#RAKhV#vH8Ju)Q`#pQ4@_Cj#nk7Na8zMou>+eMS+%N~HX%_tEzdo2+`+T5|kSNW;*
zQ1ht|4-b=l)FId`X9;qeW+=u|*|`L4=5-wvR=wQXYBwV-7+W(l)*FbxH(S2i{cyy|
z>CL{U*D6TuhJ4`=j_Q#k2rINhG)9)TjFu18KHOk$zb#L2w==_~vPf!)<{uNB=yODP
z_mL#qD27E)Mou<3T}H|lX(PLb1fWBOabe#_$+AdK1Y5Y6V&-y^Cd?XiaZB!fvXBZr
zXO;@zu6)wO_U86Gm36=(+Ql&HKG$*<ZRU{gUXtM{p*w1$1J$&V+>Q`8|5+@we+Tm)
z5KtGP0UsQ&UJv*#ybtHF>rY&?{nB_pOt|4w8I6wg?mok-T#1Nh+FSTep04{`k9W+@
zw&7<rVJ)@0j$g(E#-t)E{O!hLjxt7%(?Ybif{ovb<JX1{3GT-sk1=yM@J`Gk9l8i~
z1#CP(c(evpzWgtLQF=YEfK)c2;-D}JSKT3hc(Gye3{JWJ?e;=RKttPMEUum5*X;5M
z>&A;IR4uy_Z%tH0wh1)nZ%TXBX4H<CKV6|u_@k{IyOy;AKAh<Uc0BUWo)%lU&4SWS
zoWL=Geq-5cFdwW-$>{Fbt1iYnY@SzFQCtqelE4+UCGqG<j`y|(FwJ}@So9%bze2O=
zYvrHvSFf3)s-57wJDVse<)0@fi&wk)?=o(SXCKdpJ-Gj8%kTzAYQ_++sHfDAJr3ZA
zpp;uuA-sQcISYhQW94=lf++KNei5NX8)Ldf8-J|ZC?--FO7^V2-x>MdSxBSPf3tpm
z=?jtea4@R*+#Jo`eN?D%Bc@Y<<qre<js|Iae3iX@t4EIAMW^bspgUv15hVU)c5@=X
z7)fa1vDiJ9u7KN&YL3@+)EUEdS4;nXx&LInjPW9UFe)Jw*>xe#nDPLX`TVn%u}aPT
zn@{?j7S|Utel4bEE;WR*{#G&c5KF=-9)GePX3^7<jxz6%`8FhCz>VtdT1CT#w+y{z
zUterm4br~*?2$ew7!``ENxAdd_eNB^GAUJ`4T-UVY0}v+qci_IoYD|U5$x+w?C~O+
z2aoVjm^72=Gthp!neyi2*(y)l?_NyFrvF3JiRE`AYF(Y7K3mxP%<Qe9=B>{<!6BL3
zc*>4`M}(sVY1D7v+|1=#r_t32rL+;?lu4%?(;0EJurn3)O5^&*nHjwi9WR$6BZ`2p
z??;@QK#css0lh>-+)xV8vL;+`=hgh}p~OzyHMHWPkzni$Ni%`+Z$1{*Sz(W2(ODYC
zsKtN7%eDiQYuTi}V;O<-N-iK;;Y{OsTatbw@BdGkOEKA8!>7&yH2_y)gx#UewUy?L
z9UXGyBtV;$Bru=SA7&!`Df97V$cE|B*32rJ3#riSJd7b}-U3<Co7b|BJK?xy>}`P}
zJ>=brim?m<u571FN_nzn|G6v(B%L?c1IhWN6>FX#Ccid@CcW$DE+Mys;x&;5BpVZ8
zCE$HY9)WtrDV!*O^gd(*$$c8_PG!p_SHE{#pn&hbT82a3Ty%q;|E8rBbK%3C%h6TB
z-HcOZ>nRx!e3Gzcz`SjiOjAO2mm%hCB<Ol^ZDmh`M8zQoh5hF?e@UQ)_3OtOv&U`0
zYz2L~g)Ld}`P3l<!~bz~-tkob?;C%gb8wKEqYjQOt7RNH#^E4)&q^wxjLM!za!^Re
zF%Htfad7OcB2+#`S(VWtyAm?9S7yKW_xG>AdU)u4zwZ0GuIEJ{W0^JJO~8}(^DrG9
z%#zWaunUukZ(cxr^k<IKerkm&&WtA)|IMD>S~ANbM2GwXZusJk|G0`)XlYz^RkIUn
zvXp-I<mo~v-R5LLT4J#@McPvD*}78P)9?5Km{S8rCH+B9b)T-=R-dI;4rhRJ#Yq??
zMBAjB7|en-SA3nlZp50q5Mwt7^HEc=mNp1{_EafS{RM)5<zC)>e@ovHe5FD}V<DG;
zMnZvv{v!o7OyN1ZJ8x%`WB2HnFfG{9DQ{nMajt6}%d`!=ZN`kMUsOAQF@Y?Z&Ovpz
z+M%d3y-Mbq3Vk>82^?Djty{&=_paG^2Ds;Xokn&TA~>w;(IT1PJHtC*0xa{vbs>N#
z!lI`t-&j;0uJenfcN&0&gFaN^7Q!`Z07`&8nQk%|T?S1q%VH6pFMWMMYnUoNcp8ya
z&0V3u^CoQOZdPM1K!ro2lspFfi2=DL;V5WvRUvRC19uu*_O`%cbe=_95O{}RSFkKr
zk<ZRe^D~1H)W9^u`;Dld>vSgTjsal?hl~z&I|#62SLYf8PK{{{x<!!(_XnU7KXP*n
z5~RohF|Ql8(Jeu3Q_4}4_H5_P*79fSC7*RMz4Ew#K9ho)6Rnez39Ld%b&C0B;f!e*
zg_EZ_QFdAYY#M0(MBMH~AW$Mj<QPU2p`aKK==C{bJs@;0kLjW0*m_3~cukR5H%pbL
zj9|2z8;ao)E_`Ak#YwKE_2ogL$gE~&>^5w@R5{U9V8ifS7#C5S-C4Y0Nxx?>hTX4I
z)KOQ-Lc`vv{8?1YAVX==*YlF}r`HYhidOEFyJnu~9A+j(I8(pn!vQJRNja&Z`q-lD
z#w$UmwAYs)^4P0u@FjCp9W}N&_M!T&?WMX(SS?wiV0TyL-Tk<0?~nW6?xB|gvO@a|
zCI9wVfAT#(^DX!Kw|{^?TND@QZi)c&Z(F%179@=r67?;cbfMo`b0Bu#5)~SML&G!U
z{6T!*Dl$@FYyZlrt0J;InUYOnU|{5A6g9v8@a(Tfm3PsbET)*G-k(2q5G~sn@z}BN
zE{P7Z5;uW)ph{^yF8-Fo`%7yl-}vUHMLyhTz$i;SYK9dsbnW>I5w3QoMuwk<436ik
zT<hu+J`HJHMKC4$$~yZ|vFoQks+(s$Tt04$^YNOg<>ZR+U(-jPg+^A8`|<yia1xs;
zsELKuiI;=dq%9AwR-GLV@(ha5UvZQ1_7en0vIq3BQJ+Mo;HhAc7u$pX2gITa3Os6b
zWR$i0<aV(o(2J+<-5OOlJ!g0%{qUvhz~J!3wM|pcRr=tp-tP}>e1~Ru2CRSVa-yIy
zt3B@8(?4fFtMr)HnOj_s$vr1rlcZy2mj8aB+Om*|1VjAc@CQTmF6x|Iymtz2EHb#W
zOO0Odxo=ix89oW@VHWs%Rv`Vjn4(|sd~NEe*s0;a%$VQhMwWTnGec))W@k6TdbwYS
z-#(IKqfpFp-PVqdM!LedGnzk@U(<E2?*>lIB)s5<8CnvbOKoA#nkQFs7rOMOiK&$<
zlH@pavHD_NnpSA_<n2;d$+W{_k0pDW)deiqfR*C~$bu36n;0)iP8C0Z^u}U6^_&Az
zjP#I`48R=F5WgZdbg4Py_lQZ$q$_NFiZWrgIrkX|=&L}Jly7)Uhm&Sx`%LVxT~nDc
z$@&<+`Z`6f<<46sEYl+q0cOksunJ*WKr(@C8U%Cex;Oh&Xq?lk<zxwuREPf3`2$|S
zxtcsx-T(LBCHm0<QaHk+1;s!P_34=L^8HBWlLH>KW$zZ{j9d0BC$K1|7F4s)tz*dj
zyL81*dDlsvulSl*2=FO$KqA|Hg5md;z_MqOlI@;i0`m`N2Moke>s`f?CzTAC0eLbh
zCS!?|b}7WYnhq;CLo!6XOZPSK03zQdcY<-N$175k)KQu%%e74sHSM>9*ox>yFWiMN
zYa(PQt6^p45xV8ZcUCDP0}`t?OW^Ihj|wT~x<ZYR0|F<v5xgy;IRJ7z$>J|<lEW>9
zwACTQY@F99fb3er#RS_%Te>J3HNuO1J`u2cKUfX!1X-_mrM><3>P;7avCn0J3ep2!
z(X4Xtz*`n|zJE7>yk~2WB31Ere_>|Ze$w_UThXohSG8|$eb+6E@FUy2-hmeT&1XCj
zqOaPEDDBj<{e(QRmU8e(H{4pf!H!{7@>I(Iqf!hjRQ+;yhtB0%0B=NBiCB0Dj`&~D
zuw6Ly*xg>h1=v+_`l+8&vjQ@+s+ITqHDV9CA1z8rej`oqS?`F=9OYj-v(pQokv+j@
z$l>QFD1XFkoS;{2jbmjks?B;-`|*1`#Xu&;`jX`j6{9wr#;D56B0u!E`P*Y^_hHu@
z3@!Lm%V9lcjh%J=%8h(-#dlvjpBLY^BR=gcbV%b6=_qA;aplB8@8E%hcr2G9xY5j%
z+_;rR9%p1OP;o`hDu|dR+xrfv1mKqV^)bo=9Q`cuKs^>t>=#F_$U+`An@uL@2&l~=
z!C(A^PA2}<iJjud-Z9x$u;}X5zcwpUL=LVkg75j=tH-raB`DDf`V`xaeogKG`;iH!
zXJt5y&@VOILz={xt*JV@#mqEj7%FCn$OD-oaHJM6;r$FPpC`3}uOyK;bmGQ#JD||)
zv*8&hQw$}b_~ZB9TKRK<oGIO6tRbV2qyU%PqFsL3dW9rk-rXuw2k8BN|A)ml9WR>>
z2c2J;oFBvnLfp1nz{^?%2!H}Y`1;ucPkk}i>z=S=RgW+S9FdxV?UIiKI=z?-KDx?)
z;LkyS%);Rf{D7WB>s{uj$OYTtPwIXkYrxkMw)yggfzp|2Bz7_1$2dL3fEB$BhpDxc
zYvfD#e(_w$ww2@VtH6lfOcNA?%}%AcN>W&2)Cy=l%q;pc(5M89&~|igFM%ig?sG{b
zjfK2yoE<cq{8&c&m!S!hw34zz?ap}kDXga=mh@@7ia!s@6mad5e~sgifLiF_F~9wm
zh8&lpnDV@AgIujc)`tx9970tQr`_`p$uS>L5nfJzJOzoBV^a6JgH>#Hdd@Ew2BDr{
z@|EAA=hVKecNf5m?&ZN;**YCh*ebig7K}WkXZ%&WyZclI|4JIGGK>R!o99bfmL=nE
zA0xhgw$DLtM*7^yI_sLK#qx23d(R`X0o!zsx?UBpoFi$W;X-t|d-EUg=Npo9jVs>U
zn^f%oBAPE==U5ta_dEOph=-Ta<F2k=f8V7&T>b}W4j+2(&U%k(&QW|swpyGRQ7K(?
zmW-!`pUb@zM?a)A9QGW2{n3=r(R<<qX8F9;Icp+E$S#e`y>smA>l(qJcT=8?$)wLW
zy+B1*U(GHRV5*sn@DNxM;r&WBVA#F=2hjUAjlz%Qe#j2n>JN15f|hjNs^TUw%~!IO
zDnegRdWX$Dbl1$qZ!hziJTn?_c}?VOxib>U7eeEToRTh{xu#$A@sX#+f>!z*2Fx-7
zf4~&YEsy`mHZLF~$Qw=28{5QX%x|C5%Wq)3MTe(<7fL$yx1^nqKQRmV2l&BlZ)Ul=
zb3Vy-$S8j!+W|g8o&OG8F}=rE{Be}()136nak6DPYP-l3r7<w*UI<08z)dn(9j5}L
z#h)iHBcg;K6|>iMJkQs~FliQy*!S^I9A)XJoQ}<o9}Ewn6;NOMRR6V@og_%Tjd-@T
zaIDfRf(W-VRasT!N-7#92|8@vq(6tM+_Fu0N@*D!m>NfxrxIf*ALlQ#s!8^L%lwN3
zp{4o?$=r@C$-ko2`?)N!UM9SQ3aE8^o*3WXmh3dUS1DQnCf8vU7K7{3U}X|Ob6Yp+
zSuleb!L$2sF`{@s{k(`ho4S5un0LIi!@qN3tn7r1+m>4F{S?qPc@VgI!(B*(mDR8U
zgSx0Wz>_VaQz)zkzFr~28-+5GC)yPt3#O@I2!Lpz&hBk|Q@vjtKaH8&Dv;tf>sOF+
z7U9z9X9)#AI=agJn?Isj1g4VN!=aDTXi+JxG^R6l0qu%e!4`@=;t6q{OzW#Ati8(M
zsf-`zd_|Px*~Og!B8n1Svw4S`6kPjfg0yOZ*=ZKahw~fYo#8e$4m~ActG1%i59P-=
zV1NWG%bTI#XUE^Q33G{Yy+o6jg5%6-IU1k>G{5d%b3jfiI-s>kIkj1SPGpP2UdVMq
zJ*MxzZOX>!^V>9*Y83UjHp@~C1ENd(0~SE@-&rc?v~`LVHdiyynf&U^R~eZjD-{g(
zX&ZD(agfdDLzyPQ3fubu^+j?o{)9mZr%7s4F<|v0)&n*%)xZ4u?Js&x(~Yu2hqX3E
zgjd0cLq%pu>HsTC@(3d>3Rp6G#=nA_<l44Igk@rp{6$Y?ulY$;*ISuC7|&Vfn(_6l
z*83D-dG+rs-AgK=fu8D-@6;_(vWua6TsiO!coyoGwjTu^$S&RdygOrLw_{D-8{q6b
zL%bei`qn{sVs5LPJ<kBlMwL39(@IUzW$=+|VMqqo;Kb{|GX73^=NXCv?z<;*Y?y9J
z9tWLo(6%uv>Uo*oJ)gT~Uoewy!#Q6NCI@(LsYgNlK?b$xV1C8Fc1yiyQ*!3h)w)`H
zG<p0;`LA73Y6zk{@U-lGfxHd~bWRU!R`n7`Lf&4vY+>@GB%4cZDaCfXRal<Ll5UUn
zkB~c9H_}TZujp$AEhSX|@&p`9@&yH756C-z1?{++;>XL&xF;iA+7)6}%1bzQ06ASK
z?H}1<0|~}Y!~1RSfA;GIV?V9Hl4&iBm;_SwfA<D&t#}%3W6%Tuwq#`KJqU0KkzOpA
zFHphIo4=)6JN62!3dBUz7e|`hW^P<a8X_<QmsTLc%0Gzvr8Aqj6SZ7vc>&u+;^W){
zOQsF7ld}Zz7k@va_M_~lue@mJLO}3R5CcqUsIU=@^-i+bUeZ6nu`H}8GWBQPVh7>e
z_?V$!V&nNkGfT(UZM4opsH6-LOd~%bEc%D{GjU6=bdJ5~#k*v7=0m!Bw56aBu&?uj
zWQdHWy--(coFuu|U#`H!$5tiOnYH*r$VRAEUEcg<UnfDQs80OdcwdpCv_KHfcyo`A
z;5;!<C@X+QFx^cC&xCsI(uUxvSFv3c4TxyCF<e=apLBZZDlXwz@E}d&&4bql5l9=C
zjLv&r6Km?<{`RSmK|}r>X3Yp?<K))2_MYKG%JOFfHm|CWGrxXbC5b3KG${GqO5++?
zy3v(K^p99*>`}TJ6JvJ1&xB)~bvkU0gzRuMU~jg)8OxjTgPY6qL0uax0CUW|`j<lY
zAH*ku=mGAyK>7f><y5c}N09aRq{d3U5Bun`l@f^?ZTx#hX*;4*9!gjlhr#ojs`;f)
zp^t^c;zP9RexJT#^WDi5-FwrR>ng<=7ci6_@DuS-w!?49;-*^P98M%}(p>V7p!@jU
zTR$f_HV@C<;5?de40=3FPCj*}Z;%1&sMFXBI=km{f$OGT%a%y`mH%V~5nSrS_LC7D
zi>KRm{Eidv`zs%s<3!?0Zs`?|c8B=37$PU})VB9ASH4Mqt7E$`=i1#Cb12D37Eb!m
zgDAUH;r~82gC|0RvSND9gQ3<&J7-LXC*IqS@xnK?e?55l@!~+8d;V($eWrIjEZDzA
z@Yf!W_yxR5U4Bxg*o6q=m$|}=k<MQ^SVt%`%OW3|#KiuF6rQ+ozG}wF)UR#sRl-7$
zP}kNHoA+-FZxw8?A5I37ULgTzK7P6OTMhk9lu(3QIo|r!DM1in@uMs}6m}MwGS|Je
z?Msx?YFbdt-;Euu;{->wt=~q>?~j(Ci(#=$OQq5CBCYlr8QF{Nrrj@}(sUdj@%$7@
zI{yXa8TQ`eLQ=xxf8Bhg;K+9I-pSG-ZtZt<Rs|x_6&*}H0&p>6jJ4!dA&1X(0RP!+
zziZ0=Oj8z3ULz<wVq!x{PMlb_)hE4*naLG_(i0r}b{w(4=s{yd1b(XQ`tNILHLlbm
zTCJ1bl2NDDH=Mzt2sgUl??>{7a6AREPjTVTba_r_EMmU*6kwVYX*SvXZu4j2$7eBZ
zIgIuESudkc<^9FkuIQEb)0Bzd&}K*f>}VEHvEka|mkzZu@Fc`Yk_O+qQ)_`g#JOaB
zm8PWMIJTc-WEl)rmWSn7gu2zJIVS+hHINOe@>X%BrA*WkuA@{KD|Z6qGQmtY`)&??
zU#1{Gc*+Ebw$Zy<TYAI<8&?#21U~Q)_x+4q`(2Y9osx|@MFu&5mR`@!a|DilCbwwy
zV`vH&D4PPCU47zlUROAv$%dMs?}(#u)f5RHWhww!fatcpv}6?KM8Q3g5|N#FXJNF|
z4$8sV<bgVY))gfW$cMjaITQY#S^gT_g)r3PsO+g!UCZzLxp)Al1KVdT&#br1Oj<#S
z+2DzW^vLG}1@M*3L-q*&ZAAbLCx%iQx@pWC9j(;IyG4JxK%;9V=13Mk3&JIV2yT6k
zyW*Ht<75y1jvBEwxeBsbhK|zV3v6Q<*RQLa{p)7Y1yappVF{PCJi&lX@0$nbXCA*c
zT7=BO>Nq^Pz~VZ-xO7weMVXC~<8jUA^<__S(I4dBh=eV~^_%8B?Y*r>BeB}HHtn~4
zL?y7xM{|2gOE|{4C{lDrijKY*M?a_HgyTKVwq(UTpxD`#rw&DKKfT}0AMupNsx@@$
z=<-S9)x?5ry5jeqoQ;=yUs)_?J>yHmif}xiLQKV0#e015gHVl);A!73bn#2w46<Ba
zSD#mo4zdz_;F#qoll-=dFaElw6i1}Wu}aMq;&|Wglx<Kv1YQI!^$+^`k7&*fy_N8{
zKbJy!Z2PrKd)h7_>3j;AxLe4iDn-F*2Se1RavJ#e`^8%(ol&+MT%j#Ah+)rAbc^4_
zqgvLx{{V~QtyA051mJLbW0vqi(|s)YM6;-g)ifNWoj}AoSm*X^>nGilJ|%Nr@T+}r
z$Sl1G+`AOch#82Xu*5x(+BD&7puxnT>bvwlfTTua$(DsTsh(*7G4bheU}l;@Z3Kh3
znjBzYRh`?*1sd9ivQz3Nflvh|*~AO#u6ZmK`BKi1Hw4~7dpPP~wzThIX)@O>&{@X5
zMj!Tn5ZtycQRn(t>NdX}JFc5;?7b-du5owglm`#-uQnLaDiQb`9*|oB<51+6@`8XM
zffx~;UUSJ{D+MqaWK>dwa}^`bp$FtOGG9TfS_-v8sErl2Smb>T$%KHwC9`jUWCe(~
z*YU9KTGeUQ<VxdYSL<6c0ja-Mk9P#<(gd!CB_l&P*y?Sr$tBBz@}u}z$g%q8Ix2<j
z(qK_0jZ7l%|95)JA+Ok}X(kLjt8m)|$Bbuy4a_)MGO))!ZM)r7Dc?GK?Oyh9-@bsz
zOF7%WiPk;!H`5v`XRL;b_{?HI%JdJO`}H<JC+$V!uMsng;STJE+{W_20YfA=J^0Jo
z-&=aAvy)FRO7CUwhH?AQ%GIQ{HGk{R)|We<%9#E254d<-U3HCt{%n52gh!mc<Xs`S
zg7a?ou#$V;TUPw{x)l9E;)iy|nVs#!dx+^u4+w~S@p5vlP~Z=sP7KXGK}Rk>xQo6c
zjU?@1K)O=WO*@zu_(uZZuTcMj-mufB^~|5X=Fsep`Bn;PpF@jST6hSA7!AmL+h$!B
zyY=3`^A%}kxN78DYtcxjCvXB`+Pw|Z(KP2e9RC43g{z|#PY=$)VMni*uBl5um#!$&
zb4UIfKNkTk;Y9j;ID$cjF#8jd3OntJ%5;zGAN8xR;*-=T0~X71BDblAYLx{o6z!eZ
zik(KjOFQppcD>sOSdITIVHIJdIw`93Ix#qTrY*6?Z(f~(!@DUO`apG9Kl+u}LmPgI
zdTG$(K>FqDx##;<(tb;ai{QcbGO2Yb!G9ZU;=&gw-(~@m!`ORonos118nv;WRQj`D
zvt*q1?2$;FU~tT7ih{`tx8ul9jXCGJs-|KaB41b&;}*uEUM0mB9~d<He@KpVIY#^^
z`b_<q1$f^4afzJk$R>OGBFAR)HG0eO-6{{u=c&^7A9qabhm@xHx|;$*I<pxzK#yB*
zDl#_!wO!slKuJk4GVr7HH70v^_wk?4cucyv_$a@^;P>0XnuKYsZL2$XHJt^cQ#^$d
z4856QlD(D{OkgClH>_a2UGW4lz*usbX%5jc_g?SegJpK<huxtfv=Eg}9z)@!{M7#0
zvW{Khh3_>z_g&7SxO}k-SHuF_6}cB5)&4=bHUSDTq<=t;80Vf9Y%>}%w_pyJNofbp
zih5JieiRY-Qaa>x+OoH4x_J|r7CMK4&cdLTg4C0iRr5X<IbeXiwOcsp4>EHnsV(6H
zB!mAcCW2VyzhqOYVhbYn2{%|?8?m6%_l3*C1Nj@95Mv1!-jIc=;)IqMna{}4Ah3-a
zR^s!;=G9>X7`E_z&+ctusOj1y&h6mR96DW%E!neVauZ~M39U^e%QXtXszB>B5XUbg
zAgp1dD(25db1CV>bS;=!zmp=t9f44b?CAtE=zAEBl|=_WlzyG#L&z;Jz-RrW<;}l^
znlaRzL;#p3Zf%#svDy$EYpbz9ucyf>tPqfc*g-v^rF~9TFjRI72`=hr6}Cs%0oxh!
zl#R?i+%A6_e>+Qk7G;()LYGMXqtf<3qYKt#MFBsb^%!Qi$jd^BDt;Ogi7R=UDN<j@
zGlx_^k%!S=QNxic?Vn$Pr(p|b`}oHC2lsr~x#iE?RG+5QSb0Q0NAvskAeh^d`(eh|
zQY+LZE_qQ7p>uwaH)4S$0xq*r6p<emIf%pkVfboasBQPt>hqI5+HMv7)47q<OHb%!
zZM@p*v#JVO1I}3{$Z^C@-n}$e=X>)ELHO0hHR_{u^Y9|S;1`c8Y&-l`ooptyT+IHP
z7$i_9f?tlhcHodprsFAAEyQnO1_8RTOAC)|`yBhO$d95>JC*U-b)2kjKh)rqv=ZT@
zp^{m_t1&+jzd8lb5Q&t1+gAm|sVFYnVUPqJI#(gaQ{zpaqLlumfENQ%6EhJ%sw~et
zOFwh=gK+X6IgXID&JsziP9fV?J0NS8YK+1&qDYNu%SJFy8gZRLU^e4xcoZZJpOl4I
zV%M+>jlye@f`F%dlYwP*QSA2?rn^yzpve88t5-1raC-RuqDd%h!b<$Z1jTb<BGM5^
zoar;U%b3_MgX%{-i-ueFmOB0H6lKcz10~K9&T8~2$cWgGv)w7z@0nGz(z7?rVV53E
z8j&$VWxB>{C~mwi@-Rp`KcpLcwkY;mD5)*2>KX)OTQNn+ePYo#MzpB4cA^zvG6)E-
z40hp~k}8DaaW6n{*-wKMTC&&f$@jIm+Rc|!y=0WV_>J7pIFl0yZbvnmN5~Ml7|o%b
zE)JTnAH(YxjEq|P99fvwTYJ#`(78I9QjmJ+P6-D0cUMF)ET%`AV+mdVox||IvGF)s
zpY$vDbM@zCV2DGM-D?G}(u?>{cP{7ohhI^gl0N!!Z*i;ip+QNTc7*~}KRB;Oxgo$z
z#o)fDGk3+A+KU0c$se`cF@432MMUX7gSCzqDC=hn)z#Nac~(}B=2W`=RyjRzwK@0r
znTN$lXI1mG*{u3M05UInyY86LXu0*YL()cTasODwwv)2jbCh!bkjq_cd#4D_)RRqt
z(3oe4f>s{7qJpn^K&`$7xpcGfm$^V-#2Nm${mU(yaRXyn(AlQa3n_Hdz>%Lsv2B%U
zvdN~&#;5k2ZWqe{Bi7B$;7+`Cs%{i{oK(xPeT&t@)o6g}Rmp*U3&-3+0i*oyxRYSe
z)0(KbG54DwWyy$J2~%bAzy09>c3a@|?sJsPs{~+abVzp05OY~T`hCDhe3sYSkHa!6
z?q1;#oIrKQ#;25y#t;oVuGyv}odG^0#_c+t&6Kx#iy#ErW}Q@XL@rPOr-pVH75rMJ
ztZf>3<dGd3;18S6U4J{>NQYmK3>V3{eSV_zH_;bsdHbe$P|%s>iF#Yq`$V!XpCCk%
zU#yiWVAam(_lzIK7nJ0LZk~E|aJRB@T5<evd3)J$Y+k+U>HV$uWYjIq<~|TQbTPDY
z&C#Pr(04OMFlxlXCzk(Q;Ik%OUB$1KCAW^7U~0xr-I84s&s`#w9WRm2Khy{Z4Sp*U
zw~F_Am|napH4T*zBg)+sna%R?-Fbcj?HY`^vyGsjZWfz+mKdVN^L_1hR<!czsV*NR
zUZtw%uju()8V&Bw-tO@{-`Q>TnJ4Fic5kVr!2QWhD;s%4s+@9m6~5_#3=Excftyl3
z(Vo)aE>W1sNwQ1}(5u<%TDFN56XWub@!Ku}fUQH%NjF<vj!Ab(ev6f$L;<>_3z2M>
zK7ta-i#W49{iXB7t{tAeAe$rgUZs53bk;k_P_|M;Sr>{$ovJ<%i&gT>J3*}fu`om(
z5$5uMN_!i~eh1^AZrgK9xZXb`>J~7kxr1&o2w2{;f<{~T9hNv_BTWDZQVeofs^RbN
z0co^e?O}@c<rGi!d6f3py~+`1R=$xqSh2~G0Ne$7J;ZX0z4TC^hxunsA>acx_~F`r
z0AjK-L+NZTzKS?*`Kz#obtA-|iLevnO;8r%eGX!n@IN3OP&5GCC^+$hhSsU%<?(dw
z9Dse?P$G)t`(48d_VgT!BDlw()d{AcjKj%w{^;8-UCforMsseI(XP;D;(m_m7cVpx
zINYPl<1jM^R?w8*$;kb|0fN7R2a?7)GL`a*bqoV;X5@|ay{YFJ9c%)^??7p~!y=N~
zgd7Z!ipG{e;^tbmW{DhYg8=_I1xu`FaU~oH&K^a54>x4d`=^2{Dc=lBF*Nq7k~WJG
z{#l>Ps7HZc7WpQhkiJ_w9rl=AY2Z{u&dy$Q*Irf?O)e>ZlW_83NxI=pW{rsq+&bI2
zS+l^sC;D!qPYwOd98|^Bo_fwib(f%#cZ8p)evulG$;=EeHL9kMFX7nQh*WuaQ`B{j
z2g~y@Xn=a>OY`CsZ_+hx?Q~H$_kX}Gjpph-!v)++a}=-Paiz*FTB&Tq9|Q>Ag;=3q
z;;rJzHA1Jb7WZI|b)2YYM6kFtQ1RI=Neo&HFDWgM6sc=JQ@W@#Z}yn<qBQE|3)*pG
zafDilE3%UqQX)0)DW>fGDm-xeWt&il#o2tgitd*Sh@&a^BjqdkDK~rrl1yHw$Md*_
zz=o*ttg}(vBO1PPtePm$SY|fefGiY{QA50hwLChq{B7YxV(Ub(hQLb^@2bNzLD&ru
z)Zw%aZ6LUo*(a@9AYCC?4IX8#>1V_=nT;2;n%F#?N4lr9NF-C=gG`01+=Z%@_^GIP
zW8rWHkFQn}i2isaud6s{1$EaNY`}W(6vHYmFZZ%dW%8aZ^Ec`&Hi^AXPZE0Mf`^!~
zSDf3-hJb+3tLKlMZLljEQ&AM_t6YK((~Ottyfd;bTZqR3ownxg9g3th2-lQOpA_T4
z62%#7wcNU7BN9XP1b*1LO|o+f%MrpJyN@T_CWHORN;;B_rIx6WK4|7wa^%N3P4_x`
zCs%0g(|MZ_uxouXLj9YNxX$jfC9S##MGB?SWwp&1{ffodfFI9S*&%GCxjJ#R76gaw
za3T7@io;tWwC5r=Y~O|_vp8mBvL8A#MQ)OYo96wu6tq(rE0RuDD{cfG&}F#Lc2gzf
zo<!N_pG~exDY?G*b*K7xB#cLdP6@16F}OIC`9b=fl>(BVPxR*Cmr=eq5LlJhpZCgn
zJSl8;d%Jo8|A3m);FT=E?oN`>(lZWMgX;T$qTJ072X*O_O{|M=3Dm%UfL^^uX6}@a
z$z7SqDeP_Z4Y?x7!(AaU%cZ|pQk=}C80c4Oo~YlM;Ib;&=Pl>UN`s+n3EDYO4Q+@=
z<h^(5>7VYhT1YZQ%x+g1!enF;kbO$)y|03VjmFyk0rNX^FnBqBST36aSh0jv=C%sT
z4%S%*DcwEc9X5ZRLBJ}_hAyAK)0jy}k#qffg}AuO@Ny&$r`CK)+&H*++lYzrGS_C1
zR!a<!oIK##UJhP6Q&;PI88AwdrMJu;hLlId86QYJcv=~%_4($FpTvW^nZl7DjzpIo
zpLr=)-mX7*vGQ%c_ZL4sg<uc*k`b9Aw?d?TTv)%Fp!6{L?=5|up(b@?Dw4s9V-tL0
zy+z}KcZ_v4y*tHo>A^2;Mfj>pfR{2!-PXt~+0O<NLvbSsvzTjg1c2i#<Poy&hW2Gu
zB_f*8HM%7%U;IQe{N20nj?MvR@0;!0r4gG?|4`|Dd46El$lz<zs$fl3<i~d(-r&7_
zdm(=Nxr6=E{&PeMV{rJ>n}OQn$6_hRMVZ8<r-H_#PZzx;!spOHLZ8s-#80Bujyr;>
zpN{SR0q1h!=f7nc^rz&zTub8|!NwO~{!;O~>!-p}LVFJDo61X5X`S!=Qmm{IZz@pS
ziZ1qHt*iEu_f;+1qVf-w1DOVUVHMJ7+d$b->()ZO6!Nt+ID-T|7=ZXuYY(GZ%5)`D
z2*{Ot$?GD8FF5HB23DsVw<^RcyR%-Yjp@Teq~DUHJ0aa&Jr|MFm?Ck1c7h?c#5$(C
z2-bp`)=m_kA|MU8!sZG!WQfFmF#$EEon<3%oLCbK>kHvkn>6R98XXK$f*@#&fRP^W
z_rq>xrB6;%^q@-q@|S^ROGlh(_C^@G12l;PvE(E$%*@u>t1nLK2Dgn>lL&5$Ee;R3
zj7Sr%l!rG}!-iLaNb;nga^3U=(*n`kJy?IFGqE0L4oDJDvd-*)2RMMpqkiN8!2K{*
zo{C3b`xbB~UQ1$_7|YBSoPZ=0=rqxdZ0e2hKo<P&7}&SY_Gbmd7$l75Ov0_by9Iw6
zrTWF`&LWRt#v~1QG*FPL9MvvlfKM6hS5&_MM+b^t2nCCox?wUYKhmn5ys?I803)zx
z5ztbU5;Y)A5S)ny;Mk*5e_GUPI}pKqZe3UTTtc(0Jy*f(`aP!9T&D-3FDlO6z4$y*
z^=G=+t;+)1FN&peyoDa-75ZG`@@S~=XXMi$c7O)Nf);$T{6<wb#p4Ac`A;&-{Df#n
ziGJn<x1_I$)-2!YS<_Q5F%y~m=jv*>dOPS>=g){wy2J*7#yEH>`q|mKEw|;z=!*H6
z{eggRxnDN?yL(xOa!=Y%$$wP&n4}z88+usw(T{%b4ccCWV1wM31?YD>z4Z8I(aduc
zm(w0a$a%iRj?8)v!9r~G3G-Kq8gQ1JFMH=p=Nz;B+DptNov&00m$CK<TUwgTmegsK
zYUH<t50*ug=VYaxMFm2iQ(_ggwmHgM1(u8<IRVj}?#x^KUt!cdOni0_3h$Cl?+B)D
z=Y8tVu%D*&0RvxOK~M~-N7Ey=j(RB#zUp9L(gyxHYPdpEgiRBl0Qfp9*llAM><$hX
zkV(?S?DJraBUv7AumF(-GZ{!%Y^@Jg6Mk~xPvA1{M8;?~5xnVOdc;}$I-AhZ7P839
zj)2Z-gGY2Lw3u@k$|Vg|DnS4+Gsku`Sico?y5G;D0~|DQ01>jsP)_N^hucC0q~R$&
zb{Ug^@r)Ir?56Q`60;>Uk#WhMZ8{onE0D)YbdF7_M+GiB*z0nCFS%6+g&MA-W-}Rp
zI@yTz+5bYIuBsIZti6`lMSIvVc)&eNtU(3RFCd^so%d~^l9vo6wH<G#&4BKJkLMy3
zjbMtigfsfBnC;>@xOSm}auD;;Jl0?Sxt`<YE!jUD$~dWSLfmjhPB*QZi39ux%gV8;
z&moEvSP+ENa4UsAJkECzcueqQ>j7Fxa70{Gr=u<x!Deoo#%$wn=w%h5FU_(W*SD|1
z5%}nAaZIsF8p(6zdQz2&3?uIaC%#y2sr`=9*IfVa5liIZ@!hXc11x2v+4wH)3WI<f
z@n%k7-H!{4=L;i7f8h4pxA)4Q^wiaV`Eam2dau6eKQM6;sh_&|P-v-=H}c)Q7QtRJ
z!_qc-;LnOmj`@TB=M?wPs>7u!OTigC6s}~+Ut({}L0=iR2d(q_KZ&8Dp*lEKFTV4$
zU}Vs}vtZDgi6A%9)iU)zj1YgM@xbgi71nc8q7jW+F^x3CSEG*YuI~IV*`WqBj*yqR
zr<P3V3-{wA8SX+>_;7~6%$F~tQkxU|3|r{E>tWuNuK%B*@k-dia+@X@4T!%c-XVOh
zF9)Ujkf{shGh>}ax<Apm=v+&aO={z60uF!rLxtp>_Y`{hrtwB+%6MbtL^bd_`G3ki
zjazhGt1KwO<=SBG`A1hft6!FfN;OhtC2SZ*t#;uI*l6GburFBp1iHiek?Z33C0OTJ
z=O}u;xMjxNZ+Ct0JYYVV!*_7e?6h*_g-U~IYFX2hYwI^(AEg)VSt}fvO`kaKTJeDO
zuw0#aV}Yl<i0H97lUjiRp`xD$1j-#Q6>aRFxFfoRa$opCFuXI>*z;fV8{de%5d#|D
z{<FvQ*T6GyKu2U&>!p_o<~z&iwy^)D<P`|9HzKmT=r8rpjU^j`fAzy>M;?4~o<kk0
zw)HXDr<vobmuwbopVooUud^c7k<b4Q@MZ^03<wfnU0`>3ORthn9}8JuKZi4MCZOsC
zYuj}C@$$8t|APF63)tI?7GoN{I8-g_K);FKwG6vqnNx3ly$gJ$!X2)Ri>Lp$BLyQV
z>It6H1B-f`=Y+iumCgofGVXBQwet0<{knXHOxn(p<9+Jy`3p1{0iE8UaPjSODycj;
z!@=L;XDEl*ZWT5|Px7?)OaFWM*=F}lY2T3Z`eLG&*eXbV)57}VOxZ;4woG>AL8|9R
z4lYYam(wVa`Fu|f4&Fv`aDA9U5PK4~M3tjw^NvLkmPxfuYBU&iCdtG{vI=~LFB~IY
zjDrNrhyWx1(-O&@-djtS7GO*OcECH76|Gf7&<oD(&FY#;(Mx~j%c76PN8w;xINNrk
z_s;Hx;t)-|jh2E!7#Qt0U=)l|d&r$<z`S4E)i<nXk_t6$TdD5u(pc)9L@<kTy1ZmC
z66o7THT0Gv0G-0%)?@0%FLfkm+_Lc3N>RrKx8mu~#d@ok080eXvo=Z!pg~(kJFpVO
z4F8nHZsK@Lfw}BluSZ-16IO~CNf}ARuc$uTRooM%oz`}hD{AKl;PrSGF-%v7zo#jI
zg$=xOL70=QunP?`$aWxn#^F?rzXZq6AKsQr4t|bKUmO*2e~|gftX?P5FP?x5qiio5
z^#o*Y+FxWny*N#nmKm$yxiRp=+TY%qqTK7ps9U_|K{C#N_Col|q8E2PoA&gWz?oeb
zY!e;lI~sQJM~bsZysBt#_8SFvCD>VfV-Lio#fk+0Y)g<fr(hL;!Y2)mfs}eKkCUz?
zbcOt=C*T2&XKjN61OqExy6o+JX6na8(heOXntr9Bn)=&vY{mH23JA$8P`gn;T<1@{
zPU!EfmFs{zImnBCswpGA)OahC__W=kxX%A|>B*RRPaU;CXD=yQ9(+T*(4MQ1Vu@%G
zRp?}SHM(QO+9EczrciPtI#LlNCp+eR<MKaMhZ%)^WWkBC`}_6BEE}<<zhW^Gsof|s
z+<hwS0fi@3qwMYrA-mD6Fw8+qKueENRm{&dI#d45Ca@Qk<z^;LcSdAzrnM6htMMB(
zNu98*Za}b_Q9;}-vv7}iREOG}upFDrI*kd7Ikp#i*hxIL<K(gfiRsR}C~)L#ZWZ|T
z0$>jofZ8pbx1`Z7d;i@&ETiv&^zdmj)&kRK1({RM%7Y0tA{jl$O)Z`Gt(3MLcGjY5
zw0s<j1rS^cxCLDNn6p|hP(qm@A?^|+JdOL*isB>)JT=>sD8^m_e&7TU@r4_?Y;+fE
z5Ge?4^SPCl{X4WHlzzb%4#Jt-Qi!v3wod7&B%=)+cz~1xW?V(KSgnei5&VY>QHEAJ
z_1Mj8uTW7544DYdCsNDXJ;<q&dH#Uml4964t>(Lh{>3_GaRcQP3(S53i?j|pkEQM7
z7R{Aym0CN{EnEW7_x!=Js&J?;rbdTOXEQhife?H8wANOp&<~M+fX#<vip$%0^?+;2
zl<<t<k--E5d4c3nPMgf_mu`;Am02-EjpqZj@a_k7f%Rymr|86`8|5xnTlWrZ&Rz8L
zt=45rH5ki}5oqGp@2@=w57oH~WpsJFV?`|H)<DRh<Lf2Cg1)}gVmH(I|C$qR)o#v(
zPv*c6lr0a;x{sgxD}x1Vx350Z7=H5c=V(UHL44Axk#<;<T>LaL6@NFyU%el?Nu+w}
z^c8}-+V<p?qR<nDr|==;v$uX)XqoLeG6mTfferDWTCG`(Rc1sazhzzhR@fk;81qIf
zoZC@PNtxPIYG~pdlFPLfv7uF={;+(bJ$G<8C@6qG=ViU^n~|v-9~rRry71k>I=PkX
z+#hN!Tf$ES?H^Mz@;G0*o|gn7;_$%EGmn#uF^y0HZ%3Yfz%l0Af35T*%B7FPntuOl
z+eokJo|a-?2JL1d_j}4FWV%H^p9%+oZW<4C_s?s6y8b!lG0}Z7oPF}4hUdkEhlZER
zAF|ay*^KGWUaKW+3BAj_zwFkSxlB6K4OVn-+_+`^++k4aZF<s|N7f(&70MB$qE3>3
z>N-psDQ0KF1Wv0AhD!HEeB6DBNt&^&!V4JHiywbmPcTr6ZTFA-@ekM@ICKw|tTENU
zkM0bJd@FBYzcvN-V$Fv*W)q;(zxTPA{k=PM8QSOO{JK`n%tqH1Ut0fIv}7RTZCCCg
zvmIl7nv_p#&u!8!Ym+>=!KxTdTLF^%ee~r{fmSi<RBi3ijjU6<w?dDGw_naA(t5_#
zHcaNuSgO2~zB>J*Goia6K*ru92y84#l=bKyvs~X_x8o(eNi2$QwtMhjK8^lyV6-|!
zk(t{-G&I6%!zpuI<CSuX&gCV4r=*JOy#{{mdioNz$px2i?BP@8PPYu*3VNIo*)U@X
zYFleKf*`8s@%CHA1Kabm=ioq)O09&;3`yDHn{;!-=e3xXszP@@a>v<x`Ob6SM<xxl
zR?*oR*mw3P)eN{TOX#k>a1&=~t@{aSkB~>6@fMTWE>DHvV^OS%2FiV?eyDp%CmNC+
zOuOBsbtJpo2d+^q`v(l??3u{z5(s?<W#jDE4yZIP7^nJC(_*nWY-8!P)pbrV-U)rX
z3}*5C=VVqu(<}=Imtlos+FbTja;w@LVp6DPuDK3|aCZGssB0>sMPTr(mHj=(3hdi?
zmo=_naWp33fMJmPYFC131wAnAtu-5Ke&#M<MI4|UTl_m-cpgIk6*8?kV0yZB5&~9W
zJ^b%4*eVwdOeX-C34gU>!bj(1P-5cjYSBDK!?N9`(Hy}W+(3k<0(52@B+;{uc!y4<
z*s*j1yR!s9HH5xQ1{rUmdj?uf^A0l6itR9t=rV56X$wqe$ubUZy;XRE8cWcM&VNZL
zEP`4aF`*f>Zzn$)S~8Q|#4VlW>$V&m)>%HikaTW2H8c6u|GYE#nrQIQ_uPp!c@N3=
z?-Jc8)1MuVSv$9j?{R8m+^(*VkL6wQSyk*l^6CPDOXb8pXt%+Fz~8SsO+1Ly+LHq!
zHBlBY1Sj}&x0Byng&w#wH=QP1hwUomPPEoVHi>sG=sy;9OLY>;iu2jA{mrj=Y3;-R
zPI9&ha2E@?RNmVDrx~4!Q^=Fo!&un5I;Wj*H|S?Yt|TOs3~+R(H1aE<sId*BV^8g-
z6H*M|O?N+?7Ol{iSu$c*)r2u#Nhh5Pvx*3~-qJ>kmJ5!4$z25JmpCeks`frQYnzf<
z_+OrEw_+vM{F>1vzgZToPV{5#&#jt*c(oSW7L-dA*W?zcGv8&yAen;l2r-zJB4Y6J
zP%_&h_fR&s7#IAeEEI&zbDFS%)*N=-(=q+FaaM+Sz&aH}_p$iCz~Mb9ge(ZbDlg+$
zjZlK#QwBg^F&cXMkz#{r8!^>F2f0IYsl1Nq6s^E0?e+j-7becRQ`7PVECJ@N?jAtp
zW$S&T@wl6q;)?ekqjsW+iYMQfafNU@aYdKcF{ag8dH~&FDeC)>63Qb5mPct2KC_R%
z-$r<z(n}c#X;Mt!`nqaz%f}*(aN8wdT&vdG1tU-d{T08^E}ICxq_*R?TlvLeG|34r
z(pS@XNuFlTZo72olku1qT4nh9;XLGHPBAuuSw^5&F3NHF#O3cg?XKA&%k54UpSX1l
zvDLEHWFkqv_v6KqRTCD-Y63d~-`^A}ngbIF+h@4-TZwwhUR=O&ihiyT_#+E;Fqo@0
zgGDz>-8~qvnhYvpStN_qu!5FOk3Fw4F%xM1_3Jo^d}v`FxVD9S`JxB=S(h#>FOY(W
z<G1nFdFke4_O57&@_W1Wd15S9enC*cunA2)TX(XnyA@lWzt)?5;PCs(k^G;u810tk
zkAq)7TT>n>54zS=$LYSh@5@m<KCu3>nb{+9?CtY0_gh={Tr>Uwp6}T2!ULu%JOdw`
zd*!;U9G${#p+212^?T&?$*b95XBeU1u<)WDu^$@k7c-Q`okamg*LsupJE~AfEj#yz
ziR1iB^9k`a9e|yX9k&TE{Yd<ti2hqoU)7$AJ<=N7CYv9kjuC(>e}HK<*=A7eb^Jqy
zs?2y35U!u|GT0%-u|*-J<bP$B_*Q0giBQz0c~ZX(EK2x*eak2At@h}$knM}eFjd~U
zS7336%f>nD+EkvBm~#<lJKvj^y{Rhi@yq>QIxfl}$SpYVd4+tvHcosWM?U&GF<dh;
zW(c~Tvtcvwu8ZLm;MYUXvhH%P<atU+FMkVWzULVF4tv@x$z(wMRZ?ofg&!3^l2IXv
z+pPhZfr{%2A0tBz*H!!zlm`QgBtAYQ&K8OccDp{&48X;4tVaz;-8g#enHTD&nXOc`
zv+HV{7XM!I{(|fIKLxwbecs718vA%G_A}nDx9RMCl$GP8{5YF(zxLS;b-u)rUM#b}
z4B}(YyV?VIOkN!QxxcsM$mTKs)1c!TSL8=8z73jw01i_G6H56};M(b^cUegR_071k
zYi^i9*6-9a)%#|qYRup5lSTNjKkTIJ`>a;+k<Q%kSWT+xvqp>0rvJtMS(ZSCr+j<j
zS@G8UvR^vWYz<oXDC1;<l>)XbZU3R#L{{37XW9#7@Eb{A4)r-k#k2hGMkBCurRIuw
zc&*eN`5I+fYRs;1ePe@}*J8HvV#@0zK6PIruAs+%REn}-w3!{d!eQKpo!__TxoDA_
zBew*`I%SaptETnHsZ6-=e`&vg3;5Yg0&tzm%*CSQ!Si9an!8DMffz7EI8FXRQF4qT
ziBXH2y*#BryIZzx782Z~UXDsctIsBLbYZpS)g)mQGq-qd>y`xv$?4yiK1aYNFKto;
z>j@<tY4o$|W_An=Q9g*I#Y&JsLMrdt^}7MxYIH!cVa;FOssIwp1-`J91E^tiF=sOF
zt02#Hjixdkq^M$d*Sstrk17}#>M@u}A&-Lw-Wh59<i2g!o<<CgRcPi!hd!DobQQ$f
zxMgofjiq^Os)|`D$YlE&DMo+zYY8e2At-82>9~=oY!86!ngc&8|J%0!U1|V%Nq3o(
zY7FW`gu+f|yzQ;k2f|5c4qqE}zT{A3kNBr-UMg1I3Se7|bhM4Njd}ORgNK88tWh%v
zO><p=fpFvgYlY=vvNR|e-R}Q>_)Y)XMr8z~yKFf^Gv?G#)$vv?$Ql+57Vj$dea_(L
zbNY4tuPgG%>&&aTPcL+TuO}OKA1<bxmV7NLvEJ{xQSn(qGnrZ1&Qz^H<xH&S2^T_x
zTvxG`0?0;5R#UlYI>NKe-PHUx7sM_=@~@XRfBPtL>zewpU9;N-?MUV3YK&+y?ef<T
zC_Uc%s(#yfWbNaACO(tTZNx#2KVh!zHfnP+{uz0v{*LTOGzgpUaW0Ww>^(8p&#M$z
z)J+=Tzr*dKRw*Zvdd8NVZC6lqPg)UAl9%ZdvrDsWK%oRblOd$VAr}4<R_gLxaoKBl
z_L?ROKyqBArlSL?`9pXd*?T&{zm>CjZMkcPJVfRxmq0x^$8C`=DUWgkk)0x!P-TLb
zCo0H9ycKj+XAtlZF%@CCzQiCPKC)>>Et@N%wkJbqOrzop*#l$@xHZW^<hx10RGLwE
z0+aY7)u0_j|IFq{l8GK~2yb#RBm0!D)&c_fdErJe!Q802c^%54woCH}je`jkhnX1m
zP{9`BXc;H&3h_X_K~WRno)ZmgZm{dcjk1NFe!pEp{hdWzU&is&&TMrRDbK732<$GH
z!8G^h`fZ*_fdZ#ZF_R75IU5A}NR1eJYo+~nZHs<3;r#yYbRkxpIdZfI;x~mlustk1
zO8}csL~8P9rZt0$A0Uy3C6LoGsai1@x{Tiz2mnQ=st1^6i};yANS;c-9C=k2D#v!`
z_6XNKLG%0&>+!q_i6ATd(@rn$y{VTgx<q!TxjJTcMMh4NlS?!KCF;cuZ-Q~~nQK1*
z#CHpt5kQWdHdrUN-7c$X11uT;ECevH4h!>(YtutyK6NasOONtmWKN*(7RteF?uDni
zZJ$Z(-Fk8IP-*W6{>RUrl7l}4IWIlUw6FPEZld2r)T6kbWHAq}Ac81>`&`A5HUC|e
z6e^0;uRLbUXVf-szepAvef2|0YjF#<4rWjTLfms^m+S^(1J1X9cvRFBVptug_mu&~
z@m|==8r&_4ch3w>x<bDCjh>SkA2lktfBp_?F2NZg+qI125=}QsxE5W~t;rEW<91vl
z5z&cq58FjOGQ9&9mL2PNQ#^P0ZhY~rk<;8oXNstl+%zq@WgeuaG2>zoAzy(8;F=VW
zk&3r`?jv2Sut(o>=_{@Pg76iG_`5kOGtC=c{8qxR5O3Wu4huQ|rRRCdGeMlw+D^v)
zn=0mNCYlCaE8*-&L8ACmiKn{xKpET%n0Xjk5kdVP*d=<>^b0-MaWgtpyYOt(!ZCaQ
zKj1a`A5dqJ<y~oDfzRLJ*M7F0SFoTj@z804vy@b(K><FmiNxQ2QunOx%K-Ox*a-eu
zrFd%D{ot&HLfeOBTO#E9l|~x3;skhuObyd^q$uuvW6cRg7AJXoNck4W;1X|fjz+w@
zns}8Db?d#)heDswH&+@PUo)JSlv#YS8g?0pIl6(2+7Q2uc(Mz(Eu--wo1@B5rMFb~
zb-(QWnj$6<nRf>kbr-lC#D6C(U9-|1KCRHwC~W|ZzScO!a&n1htZKg;Q!o1AFkyD&
zd9X{Ej5I;06;^>hcof*;^Pjw`R$7Aowd;fH_HJG*x*v(IV%tpk;oEtOy0ZukFsD?&
z#qHR8Pw%R7zHvuiWQCr}Vo?|{tYNm`aF)|yNOQF@(Qky(_kLaY)!Zj(ix04blJ?XY
zqR_I-y*z@6+#mUPsy*_W6Supj!zbwr-oKySaF`NJuDPRC^z6ItHL|;YpG2v?FgM~T
zS58mZ68%CetygRkv_0s632y(x4^<M+LuO?mEq*<+*w?tne)D)nF>Dy!qIO<qcL79$
zepA~A`&ISlC`=k9(-mM=d;!;z{GC%kyt$14#s1wbO6dcQmsf#MtQ;FKnKYE#k<MWN
zAId(RhuODrHa!`uecqB71TqboEqt}*6E2I*fZ{<T96&KnDo1mJY`||a*=rhh3FH5*
zG6kXu4&DLXQk#}G{P9xzpw@N$QADDcK=<5ORJ2?-XVNFS48WsxXR;r5bs8?IRfoK>
z7iqiKFK+VZ(}q^<pXnlVz_2UXZ4UiFEJSMy$ch)^HBb#gcc`XXeCwvEN=>iPm@lw>
z%I!^OLxBx(lfG)JCVqXoaFifUybI<<itrqn;Q~Neuu+__XQ@dx()pepPW3<%PIV(L
z7}{>pbopgKj8C$s^{bD&G9c|P0e5$1YuS|OhZEKL>L?Iz1fx`(*?#Zty|#QbxR(TO
zF{7fnAqq$B)x=1=ZBoDWj@q4nKvMa0pB>jdee9D)dB+dH7i(FW(oXa{wj~{<%c_b|
zA}e&%H|IQe8*a|lL$7$*HjXPNJHjvBbLd`s976wOR@2n!Xf?wS?j%xfyzjUA)h?-E
zogv1J*Au%p62nideW_ufYJ1|x=j0T+-fOk)!zU2mgHxMXPu-r6gVU>Au^nalKNYHt
zpf^u*xf_FvK{ND$t{0{vX+wl+F;W3yK*{%=;2VOjWN<Ib*=OqyD0-OCPm_McO%uob
zg-VBjlJKZ@3lerI)nm)$svgPQ1oZZ|r)dLR_(6Ynn72H;yv2JFxJqJB@<h<GV@pR-
z`<+yMwG$~zS0oy}D=qtn=7c6*dB2G3!pV$f{jQA*BwAL}jh&f@ZobMLp3D>1#)wD;
z?`_c#8FZmCSYWAAfC`ybJ5FoKCD4HOq$wHdY;r+4M<5zFos1G;7R%{iQ6@zvL%;hC
zEZLlIBZB}@rf55nm|?^}Kt(}84hWkri=w85?aieu;W#{B57g6W|5wtv$20xEZ~Q&O
zkQ_3FoXN2i=CBw;Ips`|gisDinB#;Ys>Pfu5|i_pP6(MCM#VIukYsX<az5+#`h0)?
zdDz1aw!QDyec#vhy!z~mqlH)#_Q^W~LV*$BV6#0lRNx(`V)^=vq_?Fn@y6vwDW({L
z=#OVg@1tH(n6U?_^@q?@NH_W1Low~d2-}Xx1b;c6KZ+Q;DV(LTewi%Dw}BHqb^<Kz
zpnx^x8K_m=gt7oyY&|S&8;kDqZjgFoLS~`$QZx$$DF~fcY`1G?<vZfQ?OhW%;x!5t
z?~1Agzt{M5mO=6+3~c)AWZskLY-bGf2b}ft(+2v(pb8^A3w6-uFSjFyewu1(rxNZ5
ziqpxn1wIlRC)@v-PT&&+QV1aTlbFDn|2OuWSuWDHfgwcZ{0~E#YX^3O<z5+?V7$UV
zC|}<K?KHIJ9^>#b+hg{O{M7PLeu{xoK&N$swwU94!&H?L@m8OnA<9wa<!A3-7d86M
zJfPyNdOs|}rT;3v7U-P~IS~E4*YJFx`xoWWVr$ya?WFa9h|;R~I*=3bTGKb^)PM1S
zn$DF2Tj2$1D?aID%JwxzE6)B-bY*1j=b(lwN6c4Eai1gQu;HK4GxT&>ok5B-P@}|p
zF<~)^Rexm?PS+gXe;>z;_$9F0i&dplXr`+)v70v{+KNZtT)UU0{2l1FR|l#D$cIMM
zpUG5|b$SZus@|dY+rPYWg?jV0NHYh&)KxMEG@&EDAUXXrXf$+xOuJie{!8Z+$-)>3
zn>%2pb3FsOBLjS6D;RwSaeVOh^8w!zP0woUfOUrsr>ZH=1s~e>D?x|)?jwr)!esRi
z1_pJ$0qgA>s&?|b@eJ6Uy@s;m=mne>3-Lrxjb<J3wqRIMwGwPs`qU$lOe1iah0TU)
z@Hkv<>1Z&%=8PC1*PIv%mJc@9ztU*3P5Xd3R!ipa?sZHIurraPzLwkb;`uGwSn{8K
zIa6G#2Jvk_&-7*^t7zaA04ciI&;CW*Jff^#xiiypZpTKRH*9+MiEn0ub<6U?NcC(v
zD)~^$*pCGfvwJ%nZkms8gm}6?a!-p~@O&RKdDVJJZ0U;3Qe<?W7jGUe@OrvclWdmD
zM}x)eKA9(Mmf|2uV-O~YX&O}WZ{8@hQTXs3m>WtLbB!3-F4hxe%jVK5;i<|dhydfl
z-qVY_8|GEwJ2Ac1kLFDrUR6@(eb5(Xq_}ejD?9HCPgZv{vZr|m<c85&R16scD8{^z
zvv-WZ#tiEBOg2^~d@g6l3P05kj}iN6Ioj0Fg9HI&1FI!xCPXvX{7lw%vIEx^h^`3f
z2_T5G-+l|epiy?j@HSS$vC<h}mJ=Ohf)E4a@lqIL_&9+nlGU3FQEe{Hv-6qYz&ku1
z^Dsr{6}({uTgd#GQ$yITM9$1pk|0oN->O-1#6?hW$3~#P;z!AAMb2~r_ZkS{!@HLh
z$#pYs!A0T)+;h=$mYn&4mVnzL_=~Wc!N8QCKi>_yEs+8PLVG}Q=Jo@@X@X$W@dT}?
zpo(~_FboaJb`w(Gv0?pf1oFcnW+q&!X`uZt%p6(hTer6RoOzGHgq1Rd@kZ79Man1E
z|9B2i`H~dzB3hfb7cwo$nTQB58m9?N6Og<)u9J^rR;{2g!ObH5$^t<_5hHNC%@ky$
z|1pP4kw8LRA6K-q99+bBeKIC#gb4sI0=-S(!eA6!3Ll?pjS(oXOQrtp(hy4i)be!S
z$%G?!DKJ{XQBYm3Zd|W1*Tk~ezZH}LJ*49O0z1+74`V?cfR?SA=UdRl57O%_CVxH?
zv~!o@XY-{b=LXP*N_(_kRQ))1pHcX5Z@#l~)Wd0$T;L#w%7@BMNnHNSdO0>3N37hr
ziymbQe|nIWm`p2@7{d+^1UXEAor7}8A%U;cq2bB9z-#lw&*9uTf%6=g1o^=P<Xu2?
zb=b%c#Aid1NT|0_J39mrMr))45}V5R&9TEV4Y9H*|MSASHF8!&XdG6c+}q8mpK&^<
zFFKd{2x?0IsOpg&(tmfKL1JdIOH|3u1a^5wHbzeCc~}<~v0lblbWrI*7i*NRmkK&t
zXG*nPH_J+yNv`sSpMA>9205#e-WMqAEb&a?vu8DP6FG6PL5_=8(y32V5Z8$3-dH?}
z7!<sbvU{Vl{T9;4l-xbVhpo1JixRuTw`jqq=<dC+C8RG6*A`_E{glDETglS${o}y6
z9-1JCB~}m+d)$h@sn6{LJs=Caa(LH^;Qj+8h)Zeu=FG_CxOZ_0%#;Cp>up#0a9JM`
z7=s-WI81Z-rDsSp!1mLVa<Vx+lz{p_X+R{r*q3>u81_<jD2E50K1`1pvlhd}%*v0N
zd`PB_#Be;IzBZN;1pJ`I0`h`|6u5RcGAo_M?+&~t?PX&Tg46}$#%?kqgd!3tH}eie
z>Ly}su)Qetz<{=5_yq<$zvysk8x#s|__PGT=PGFM(}2vlUNP&bRd=HMv3c)95PHrW
z+W3i&63x7)+1ZZ2QFXk342Fp7l0yttt{rX5GxhI6cd|g6Mcu%J?Zn74#o}_fSm4Ob
z&cm^CB3(0(kbW2l7ZY<tNEJbH!V~?5cskE%i0BjmPFSK(SmQRY%$Z(ce^&iev|G~2
zosH#Y7pCH9;Kx8mfytL1&jq=B0Cn22%P`dt$h0C_+_n;|#EDA3zwNQyybzd`5Uy7C
zbC(yj7_G9DFuHVl^G9v<Ansy_(bY2N==YePn<=##!$!+J2eA#O6y8T(u_k>{akk%n
z+8?TB1cVq>D05j^_S?QssAV-$z%{CNN<X48xmPM6x4pKQMQLB2%h1vqdnjkU1V%FS
zV(NYN>ao4D9-9n6g>r1%lC?}W8<2Q)-`(w3OD`wKJDD$C$%Q=%^?#<e-9-;xPw|=3
zKP~;f{59<pJ{ya|jUwlvwI58gt*!$H-~C(X_(&Sj97A1z-k4hWG3cub?|o_+nfrKl
zNcN3K29E+wbk^&BVnkj)y9^VX?N3?JxLpMJq!g=i=%Rq=`{q%q(9Gzcs7}9c^@NGk
z&)F%$(|PXd1(70EFOH`6>BJ4S+%p}R95^ZQg5V!FtqU|DP{25P$fKn>S$`mK+g0rL
zvPMy`ju@Q}Y}8+w30|4bY;ovEoM5zg(fm=MWDd%z@a~z<Ir#;eqa-~9r1-B{7AYTr
zG)qpif)F|L%XM!eM!RoWiO|*A*F@$xsF5`m+4lDe9S&Sl3<UO+t+yNp9cJZEe!Y~g
z(x~&S)%3h@V&cxs{eV8%$Mx66_I<{ei)iWIta(SF+gbN^rHYh=L&7IuS7mZ_Uuug4
z1V&x<GlBNQkQ`Ku?o!pny{)?D1B^nQP6L@445XIV$sS7Knu}JZK-G-0OCinhU;Qcy
z%o+w#BATcIz*I9p#t@yjRDk85RUQXAn&^|Z_bdjXrVAqF?6L%Ty5Z-&njSLS{k397
zE*RK3X<*8*9{?m63sEr2Hr#k-Ih#Nhc|jK^&!RIzteVKO)WqD#QW!3simYTO5~MgH
zJZuNPG?Jm9xm?YHB8Yh>v~Z+hCclvBw846ztN_^BAmnLPa)>gBgO$rrb78H}N(F{M
z7ZU3H8f6z^eeMYJ;0DAT1=#e<_bBNDq`|fnm2XsfO2<fhiO#DPV_3nA&LE7})nN)m
zUtj7T?SZeA2^^U6rpRBCbyH2Wh5b~|jmv%a>YGV7M?mgm3MU7}Pv%+EUnIr78ZngJ
zL=gX590UCdZ*!-adS7SJqoE-2gC$({#ctbK`9!E2I=Eb<0K(%THkL3bbtJVvO@m1V
z4DS|V;?m@s0%%|#!7_fq!;{2<HkKjqlLSEAqZO5zD}f;7l%BxMZEAy+5}=aFtp3zo
zRSN-O&zvAo-&`_S);5aZ)7rT#;=Bn6ZF(2ixFI5c4l9N&^cHw8S-J*C&i4}vWsw{c
z!hD^lpl4mRy%$r@d@VN&<`qj%@6_K|y`IUF63+Upo9JPG_RjQIt+#PPYdaSk3SPcW
zG?9X(<}?=JE`-9$jpRqYk?DCC1;nz7MoVd53hG^MXO(dW%oVq%3><sb;(^{*7<pd$
zz7#(4F!P#MuQ=!DA;R&2Dv9f4=Iy8CPP84Vm>^=^gR<uAkp!4h&bwDZV@_@6I#^sG
zYFCu$4w>=b#esITUQ8rME)+RFUK!})3uzW=IS_52p2|f_-$(Znb7Iyi3{TC6FBOTU
zM=Kd>2v6o4bO8j%a}0;2{wGgiVU9YnZT`OAp0jI(T?n6;%xS^i8+}bza`(P;^xecJ
z0^_ckq;1dW_RtJxJ#W=#8uQFZ&MC^Zc&1dTyAFO_ETUF!?7x+V2>fozu^%ijWm~%M
z8^C*ct@XZJ4mVm;qz(`kOuDdZ1zyfs(mOUV7$zLK#6BO0DTCreIn;Ao_T``zD&D5S
z$sJHVZ$T@p;1EGrM?#H|r&zp3Ydt+CzY;sQHK?M4MnZH4`eF}*>B*s5CDptSnMSW6
zIV+R4IiuTFaxlBZoPaqCxb`ConWYD#S=0>?nPD%&g?H$Gkc2}qsYB`hqFG3a<rxx=
zMB(sCuW<r-5Z;eP=8Aytk^?#vkmfD;&Pj{{QmLYQ3em!z;;UxN3u&;DTiw7By<NAq
zXsSn{?9Cdmen<@E&x#OO)!<FcZg^^q&s^@q)2CqbmTI2lc!aFxWCpiI#z+0sM4U0c
z|I93UG6N5`XX$e$dA3-#E4oPt0U-Yl053Nt#?M>zE1l=(FwvS$;5v)5OwuqBa<51{
z2v%YD4GnonEAp!H=(~KVpKkn3kU!%taS8v$#Qzw8R6bXgoi`8M`4oF$vPEAGK2o3G
zph7fU>8?AcDwQL*S^W>f4ty5X4BJ(^|FM4a*HSdcqLGMaS@%}n#j|f}q=`7@Xz6IZ
z!D06a5x(E}@YXD43~dZiUZpCUaJ%EwRp+p^QisLl>Boq&--`ES8;}&0fC~dU?=qm}
zfmBmD*HB&VplB7(FS0I~Iu1Z(H%#71-=f1mG<#P>d(Z!_dvh&LF6XZ$&gsXj7bFt4
za1{}^28^$l%?SIsd~YocX64{RD&-A7t&OG>KkRu{N8j{0A#r=nO9zxuuT2z;*2Y|a
zjJJ)+?~%c4Y>ms`ja1QxMMW=Q)AhYl;AHk;`Ay6I(;FvmJw2wH`~-k;R`o|oma^M)
z_`n^${KG2ki;oW-WcsVd?H!u4-?OYTdhUHoTzt$}y`y!Afc^oSt-SiWssHj=o>b`k
zm;SBC4p+s@yj7n*&~u;y7ozO5k;{UGzG)`1gLH{b8(x%aoGtfI40n&FNRp<omBhjV
zapBIP8_wZ>zTPR<enqRnn|X{&;>yDkXvT?q_YF+;qfMJ87d~GvJmh7}t=o-wPBc)}
z=DksT<>9bc=%)8lM0VHm;Xk8lWrYa261PYeWx?)*%STT1c(YxE4-T75cJW-9+i~)|
z3)Su^8*|;5)?d}Ahx4;<6ieV|kX;y>OVt@kS@;|Ci<vFq<U+gD7d|KBdL@!{tK>C>
z`;}Dz@|x=~BtJFfYLf_tSrd`*61Ky6B~>yAJ<mezqxzHncZR39^)5eNPL-?`7HurN
zYptsqDWBCw#pjcHs4eWzz5;1CW;Z*GG7m-q2&o#@Bq}oy)v`Ylu96eqg!-8hS)~2F
zX(>7Va3pMSeVv@h3EQ@UTIJh%<$hUM$hk@!-m*F-40vnGq>?H`{EPtL4#VqYv+eq@
z%$#(Atf@335W5n$3fE`yn&O=jwL9Rye!|{Lua}cWTwb;e<6Rk=JKioL&%`4K@llcA
zxW>7~*H8&qLyS=on(GSWI+@aAgQ;Xk@8Ru=HX5atHj3c}syU)sc|IX@vLY_`y!hdg
z;&IL5R}m><lPDOkBH?Tf;I6#9g}E<<>`<L{;)y<szQHK=f_+f0H?v7&4M<-!bGy|u
zLloc%Aw__mhlf>fF+1c>hWwrvle)I+X1+RUi~}-gqsWfYYj*m$j%oKpy?vISfcrWa
zP0=JXW4Vkq#RuXJk4@wO+sbKOtU2=<nWZagnaVFK4wazE%(KcrhEmcOIwkV-ZACUV
zsyLrYswmHOo=MO3Nqj1rg&^kGT&Fl*yQn>xYA&O4IX?!cSk8^JJMo2qK3b<eQ;dCo
z5HwWyCiUXXpwb=6r<78g5R%EBPiC{+UiXDs#z~+1jV&cS5?_(fyV&Z%m*<f2ghH(c
zO{gD}(^(p#P*I^(ydHSgBBe}dKgoOh9x>5EcfXs+=+NM24p#3xfAS>!m`d^w%U7OD
zUCKLyb&l=F6bSX3$ua~PI5M%#(4<qM-c3Cxnp^F-R_z31&Mq~h*yh&gL(`7<88s>K
zM@QjX3Fobo)-`FoR~9o&5!P`IK9R~!Umled>3u(Hm8Z9T@$PMlFCY2}P0xm4=Dy1E
zKNWPg!lln$uFoxaZO8toQ~EK2Uc&NC?|T{=Tr!X;Pper&0x_MU!So4V8s6*^-nmC`
zJ70Wv`q_T5-gf4_JMrX8(pSydDYjem>cSj9(2Cb4m3ACwp=@Z|rfr=5=^8nEh);T+
zk(ixcqeqfg<}gH{##X^lECvYk0%20cUIHF`Z3Q`a4#=t?Zzy|Alq>2v;drb`FY0e(
zJp7yz$o}R>tuL;VAud>3r<~vf14#*m3%flQT%RWJq;8rNjv+vH!KF5723Q%W=2DSR
zg1+v<znv1kiKH$9TwXwYeT}S(M%`uAH(Bd6YI)hGij@eoP>nPBLw`b<8JJtdikR&5
zLk<v(R)Yj+B;nlwh$FHrq^@fz4m>QRiTvdZEN;->cdwqE4b-x_tGwMK;*<Gzn?K1>
zt!Nqf7ZWe5cM0!}5Z)+Jisra35%sD?YZcouwcI7w)`cd!1yOb5C4@b!Aa7M5u7%cn
z!AeZef+8i80*NL(9tE?^_&%STJVF~RJ-<55;!lz0N9P>b&b(WHh<?Int*lvDC~Y7?
zrI(w1HQW5J@Dv??ag)@dRZ}dh{YJ5b>&VwHf7dINZpZrhZeL=EIBo7J<}aTZnT--j
zwv1_|D<M8btPT!)w^@JBNltiG@@4)0V%=`hcA-vei|wV>dq#ARw-hF|QK8*6<Yz?E
z*q8~XBi(HD*=T@FfvhBfGlCA%YZRud97bAx4PN@3*qn&ufHsoOe!%EjCeX#zyP9{i
zdzMKx`<aC*hSAXhD>bzs(>s0UqzT5tyCCm9LG=F8d}&kGeJEnMHtgnwt5QZp9<muD
z=$=R~bbV{#iuzIguJ6m*jOm%Bs_;i9XuoH(@cw2N@;1*in|W;$*_}+xy{h1|nQM;!
zEt-C;^ajF~BIRL>k<7J`_}`E1IWKevMz&OhQg(O;U&BIywfFg`dia41kcpc#HV{X~
zuVjt-S}U2<Xgu)K_MQx+rFx6WittBpmvusTEbo;oYrO484zTtaVUHamuh@pS{z}^e
z^iiz9?#up_2-^YX{b<KR8MCZEHgTyv#m0Xl<tO@;C-~x>FOwje<J!K;H&hZYfA{|2
z4I|lM!7_J4*}D=EO%=Uv0gqBY7)IOfEAn*$O0n;kLbh)tvF|lQDXPycw^FxsF(>4~
z*k8>*i?)jD@ZGW*Sj5wh3L3Mnt}mJoE4}+t{hjtj{J4EyJ1DD%7INq@v=(IqB0Gsz
zbVed$V)r5%h{EK`!{^MUat}*9jM0Nad*OW5AT#IQFcNAqvQoj!PObGqntc7jm%3t#
z1!s1h7M`T+kpV6P9Jn4)=0!MD-`GnJpC~=|&5w=U=1JlU-1rk5O^r-F-S3v3FHn5f
zlwHn5m_rjJPHI6D8^tr2PDsq<m#NBGD|kkVDisA8qh*oM=K<L~(}0<pUsyKeCSWB_
zCNSYZ;hO~9R3S}4?NyWQ)FA>qt#B&aHdcidV1f>dh0dA6WJO8^3SB1x=Cbctz<LB}
z!Zw1}T$EI>G@lI4)ts3shy`oikQhT!kc6R81XmkNGh&)r+leNdS_;qh46;1#ZXfoK
z8M__tzV{EMPhhklEB^QVnS@hmT&gsEIVKq^h&pG^(2m)*Wj#l{Oyah5^k;ENr8+XB
z|02YaRAKi@S|Bu=oN5CjSF^}T#qaw788b&9e$(a@wmGV0h{JZwYp#{>&VNHeT~?L;
zdW_QIR(LA|d$>nI-kxLxl!2-JiTm>Hc~2%{K=6;v#7y3Yr@1<>paKkTprl2&njLg`
z{s=G4g^{N~kL?AYJ$tkI{N<+-Y#qC{9Z|*4J`EbhGZkb{9}oUS2X)5yb-vS&<{ZZ>
zCZ~6eez|<23!I*zJ@v0GI}ms*yNcR4?2tMa@}ccJXU+Z`ySpuo%hg|#A%%3xwaHcg
zR7etf`9=F_qZB5|MtbnY9gmAu)9FIj9%0Otej2;2;5lt$@c>~nXfxK%eZ-gy+EHun
zk>bZ<p7n0r%eo)rqiEWB3p3)O4qdv=uQrZOr8>49lNX?sO%0COh`v7xFDq1%<8u#9
zrxyf&^vw1!8N=obG{5jpAscS|YO~Rle8y6rDO8{yl4u=Jj(_4T9p`a=qx8H)%%9nt
z@w?R&;gp*SO(J$$N6Mt$`MeiJv;?ZY<}1S}T-0g4WaeR0t@r10{ew0dK42{C6emq_
zT1qRjyX<k`#7e->TVASFmJ%}ZC#}fgve5^8wCRm=&Af@RXV;2f$V)>yPXJf|#evJP
z-m){lb;$J=Tyzi`a6hVPg`hv(Kux^3)9M6-1S?;RDG)%sXZpgiNA0;+J<|I*&P2er
z=sf)Ex+#U)f#Kw9t^F`lZcS}qZc;Haq_#Th@+EG9eIpfqC^=9;hB-&LAEkP2f+OUU
z@H!dp#}(EmAO4$d&kGi^ng&w2C1_$*XO5>t*@98ZI{8{^JwkNM))4QGC}KKx8;X@z
zPK4xOqwMqY*e*+LD+%>LcHMYUml`w~DzT7P-{D}Dvn7T1!juz(JI^3Xn1ypTiw^ai
z>H#44HoQB0!Q54KAlVV}cYfM{YT*{{7oTZ$DMF==zGh?P(K?VRa}h3Gs8<X4uOhXP
zcNl8VF(?qOR3H;WWY={J5Eebm0tw9ZO3cV&<e)yyP7RE|mjGWbu2XP*5qnr*?m~L*
zbHXuUqH}A!IM07{-pEo&%-LrtCHDsTUd<zoXwxiTGt$sNX2ry_hK)+WzAk07=-|-o
zO?f6oOH!txTLAsVjr(!lG|7h1C(h;;4pw()N(jbP%|HbV*FyWo=Tacpc=N{lg>E3?
z)0g(JIFR;v>Cr^`%a9Hike~Z%;?Ye0maNgiC;GV;N=i!py}z|Mrkf)xBWtXW@~U5}
zU!aZOK6tGx`~aici-h>T(h2NySZ=1K)_d95(8x<c6L;zYhQ|q7W9uiwgtJ$4`RDUS
zMm|*^4^rN{uaj}t&4X(&VhV=sqf#7CkJOL~&Q1$NO<)e$lbx?puY0PEO))3ou%a5}
z^Qu0M8TElVkF~k=K<lDdp#HK#;cec_a|XuIMdBbQUe3<?wVf*$#r|p7g4413R=0ZJ
zQHp;6pv?*Q0zr9w>J@BJVa6i=Z@fj`(Z?kQ*V!m~t74acpD1a0{LQcid%@)+6m{v*
zJ@c>j=0Bf_1l>9zAs+eMGTVJVQVuRg%^v<%)s+SRIVG4uYNt2<-lIwpA+OzUWmaF<
z)2qWewX&Iw*+jS<@eDdaeNWbXV=+*reKX`A)a<9LFt2?be~msgB-2rUyEan6;iIn7
zy`LmjA%W=S72U3eXG}e>Px(j(H1ce}5Gw&W?gaH_T-c#D*Fslr<<^vHBWb)X@7FD7
zwc5YyLq)QzSYw^`qQC)FnViJ%w7E}7Ysds^=5&zD^A*7NQ$+}Cr4J|{g{=s&ZU{9~
z+nmYR`xflo#o)uA$c?}HE`!r~55!sZGp)51(E_l1SE$o8A<kc8T#9ewZRFgEClo4H
zVyrrP4)G37Htt)f$X-j0l@~WUJAv@i8uVq~;plfGWlS{mq9i^9jP3Tb;)S39Mv*vG
z6m-D_P*4#~U=9m89f=SMbWIe33|4Y2k)n9vWh_CsMv^oZnxE)}sk;yx=nau`?u@w?
zX-Eq}h)-NNLw{(BYmjg|Dz8W()^aE&=J?a}UXC*s+@=Od|Iomw%p3CTSmjmEoPOq?
z?KQ$1NMfPfO!_bwn9$iVt9xYy3hk#-fzl|XG*GXbhWgc=k|-+AnGw*<Ko;Zyr%4w#
zM!tsn>1+rM33a)4D~}Wezt71TU_HmhE>h;PdOU`N?q!m?Ar)u1!VXWyLl@mBnYZ11
z%qBe2b14I`yGJObcds@K>VNcGDI)>0nM?JLg=<6%?Hcy}U!*Y6k5QXSnOneiF!NHF
zQUIu>I6??=+G~&n)|LYCWhxGPsJl=8=Zgjk5+dXhvO^5<(*!QovjNyk)|p!IP$YY{
z0r3UfjkooLsEvy*Wz!Xr?=FYEdY!txOIsm^oOsu^5dmBXjL9j(k}XNGp`zILBYchS
z0BJVvo)&#p>F6nlMX(U?55}K|L&7+7Nsmw4SHyE%^ULe0x26I~q}PVt$qD|X#;PD#
z0v6z<Qf2xL4@0n#)DX!>^K1>_tbdkBmcOhcpHI51^YCyRrZ7O4wu_>#2KaYG2eW?O
zUN3w8CV_WFGC67uc1rD;okxxhD_!Dd<^hknA>DMJ>+_<yl%U<ac70r$iP~8>(buX`
z4%wtv=$mp2nfReVnt++(EZ<2K&%BSK&Bp0g>971S-we4@)B5R)39tNyYVfJEs;{SR
zWR0mT;rbaQMdG`M9~&;g|D0a1zUX!H9zInOzOZ#ML+1+!XFJL>lM_@%k*)FRNhD#7
zsXqDsjjU;7qFvq;-})*Ov0g&>4E?Qp9o6~>+dZd1Z=F5^4?Y?YeGsJ&uPob~`czIv
zx^YJhviZN=4xHe$%MFqi-g~!AM))N%e4p%C|MIh51=RxH;};WyCXaryfO9)Hb6_<)
zB@?xqiT(-|!$431`L0WIuqqEo_cQU$Dx04K?&q#7XknZ%lZ$YchdeRSycc3d2dW?`
zSJ#;u0W!%;U~H8FPFUSt=OKCLnJO8W&L9+zEs`H?Im06>oMRhXzEAO&gD8uZ;3x<J
zR9Rt@+4ktRnK>-i*RyK{et3{X@YM1Sou3R<1Ji2DDoLK;9@~{tEeAVZJ+@XnQ6s#%
z)2|$`6e%Zt8cFf?%{|+HrX4^SSdS=wy?AItMy(4!T_$9^MQmQpR;BhjN)3=GY>P(j
z*v@V-M;AXYfk}UPGQM9q3~aGxIM2hGnr-;L_Co<!OewTh669>;h=r&2%4s5@f|2?=
z8i<c5Y6VZ+tip*pKF^E3G4@3co1g|s2m)(lWg*Hw1D8!^fIMdjL73Ul;Ws&}H|Z+!
zKV(0Oy}{Q{fF5aGwd!xSGiuP*g8uV!<rBrTNmC(!LW4WtqH|SL$m~50iwehZzYfVk
z6fTUU%?Y7~rC@|_0-I`#WGQp09>t@f0M_8>idw#qr!NN!H2ITlg}{T5?pr6mkE!<q
zb<NS10?U`lmID*7?q73vibBCM86vj38s~$P>a>AWD%#fB%$SB-FJy{DZ<JM4O}%X-
zj-NPD<Vv<eyqr_#oy+z(8o*_3IH{Gp-gQTR=+LLPPTY5~LQY`&c1AP!Qpo#iha8x5
z>(3-PZz1gfe$hK|LHEW}`1h_=MtjVichj9a2KNDWr<H^g58O<o!PyxvMnGBL_6@#r
zx8r31kM8?-uU}3^UI)cy%0}n2oFdP`E7->{Zexwmw3`~ltJbZ-d6RzAO-~oPf1fOB
z&LN~G?pt2Niek!~dm_V6b$nZucC(1MsO{>SGhht!=jN9~GsJEAWSm0FkgM5lNKst6
z&1qDO)VsfG5AK;oym7*OywkcwtdSR@l5Eas=(#oS8U2Ht1{2e%q4_5#ULvxtM7Z|{
z?R))Idvee6(T?7$Ph*|{a`Y?sDXre;v1)n>Eq;gNxsETSVhKeI@O}+TR~Hq2NuShA
z^OdJgN8378mopaKPUuRDOCG?pjX4aUzusYRqMOhaMdA(r=qk8zg%r<WTlXHCDXVJI
z7|x`BrUi}!D+-WOeXMkv1Ta)g(eE=nf{|y`@1T7Kt{zJG2ic2R;;uTvnxq8Pp7J46
z9O8C1D;M>VUIwLYh#WdLaQO2!jt9Y7E}Y1PbId0;nJ}-H<gz00%;F1eV5!m5dj_ci
z2&J>!0?7Z?rsS!=sx&<06hSZH#7+}f1>JMKL7L+#mcSXjH!Br`?d~(+M@=aAEy+VA
znZ2sapjnff0-BeSWn8X`9HhBG9m-)fvNd*{0N-T!!o?|&Hs;=X*S!UU7EQwPosMi4
zwr#K7U4s_e{e^8kj@PPyoh-zB+%RCh<n#xC<R?$XhApfWDRE2*$IKZcltjXc-wg=J
zYFojNPrFtD1d<w$o3z5W9!9pSAw>eNm8?g^HfrwmG6~rSw2?Xc;VS|4n6v#Fyu$!3
z#sb$4DdP_TG2mnbi&!Y!`;Qq_K$t=Df<0vy5GR_6J_`W(Mmz7Y6soB=bUh$0m(@G*
z{7fwVMk3Q!GCMnST65{zAN>Z!K`TgL3P<&E!FDi42qKC3y7|abe{w9~M_3`65A_s*
zt~kQJ?Lb}!B-ZC$Z+Lp3<7X^3mUv4n&#|8t=-^vL{w3+OlWRHcLq*2ExA8-=nIB}(
zO^Owhu66kwbw$+Bvodmv(2tDVe;HE!*Gg~HHf2K{3P|ROT<<aOW4C<8l?z(`sK5m0
zY<nVaB0hKsioI{nzVhgm)G>S2ytN3o@@SH?u4(kQ6OX||3j$xpw$T>-YgNx=eFE=p
z!(hZ_nK$>fD#@C0SNZ4HrGBAH+??*NvX_1jC54k8XeC{w@{2!|jI}djtFwU-54#_S
zpCy0fL6ntOMRSF8zP+@jmwidTu5x%F@{2KYQ&-jIR4n?i{~yKwco1dj6({ef9^+t2
z#ws)~J;)7TD^4tVu6R`=%P)kKv~AB@rGLrrhWPd^YAve#)FJY^s@+6l(41{oc(tHt
zVaUl2ivM8;$DYKS?#w%1Oon3vvW<Ov8km(ek+Tu*;<%ow$-s-JWWUcxb{eBMMKjH&
z-SD|tBFgkQQGA-wH7xrtCJSsQYV<#+^I`RKwXQQv8JAV6KTDl%#Y2`dzAMz#61L*L
z(A4B^IxNJb!S7CTvHU1E)&y*f;@{_>ts2(nEG}CmmdOqfp%fk?)0-@4s2{HVrVr{y
zJ=28xAucq=i!*`)tMy5c0nv{khL_3iT?i|jo&p}g7{#~W-5sbKMM2HBmSzKx)Hs37
z<vl^T;7}j;Q07-w24b8bWZXSr<L6As5%Ynpj6sT>&FE7iGbL~1rJ4IGquvy99?!@E
zDi|VHpx%THQwLuzMxOA!9aTXpVeW@JT0Z@{YJQUaR1S>(g!BB5oc+G>b!D_v36QDF
zu$l;#hfc6zR_z{^EDKJFS?g(5@y=1NvD26yCr$*D^TYmN7jg$VtvipK!{Kv)@&J}d
zIm>21O3(Gf5$c6bi?5Y<0xKYll+nTsilyD+^xJ)=G3qrQze2y;)a(a>gBp`qE!*zO
z%b<8oK3O0s$|H&N#2aNF&IBsTc-{Y2`^tND)O#+xbc~1NwBUh$eS2$I--}I4-FHm@
z$RpjDequU0Bbf4A@r(AU!X!mwk_wdTl50tQ9a<2R6HQuosh`*HJGS3jkN94ZCVMh`
zX6OmDR~C@y8i(azFR8`{cS0dQxKa71#<CMDG9Lai_~RTml#;Bt{HA2Ts7j^lM`QiA
zJ)-Q7(^43qL6P(Nnr`32>!nSa+XQX9Z@aC@4yA~^D~lF=T4Ek9vHOs+j;ZK6?dQ?d
zls0)u#>AnH%=M5mqCBXk7ks9-rB!;T-e+wl_?)+J{CPol5FmDbGV<#Bt;@dzS04XL
zjIeVSq378TNxtE@Q)^&4{tzqru0oh6-s*^)y(ooaWzlctHec8?*Ec=Mp^pVNmZG0Q
z_*+RiAxEvgy$2*B?kfKS%$L1|_<{WB@8>@~SsH|O)b;!^R2|MHAB$yW)p;$hi5Br<
zHOa{v%2myklg8(Slr2u)e85OhUhQr(Vi+*vpS4p9W_{aXxAG~Qnw5FXBfRcP`DgYQ
z$L`MTXQ$>+A1vir4gazmX><9Yx5d~QzFJ|5ayfLrKwS5cq)VA_9-x&6S>lNFzl-%F
zuSXr0Gi_U1Ib6NAVt+erWqwq5t4o$zNM5eCo>D=F$V#Vqi*vny94bJ4t;{K_Yf8K{
z!i6Stq}BD5|Nh#`a<~rx(f3+#eR9xx!5p?-N_6dK+7&4z#Er$-2UVPF89XA$&Zehr
zlOj^+Z^V34SnuMor5KxdagQN=$#1J3X_4}>_qEN~7CTW0`7)zSP9v$R8=j|f;EobW
zQ$ZlgQI^EzO&~_twpJ-~sP+#$(3~40@Zc7P9;XAe4ijS`+pgR9APCwNS2x1D154w4
zPH^#242~WX=b>(4;pNMLm0*xj4S%vB%WJ*NxU~f=n^85+;`|gRUGOB4V_$(MhNMbQ
zhutZLGJB{#cU-4d`{dk(i&9sN1@^r=@6U#U#1fc8g~v)u`E_)cF<K{%2-$9}T6B+L
z$=66{xKZS|MD8kaPFD|4q8D>|>P|mX{Y76-VvRy7NF@0wRRV+*L0y4oZTMo@=K2Si
z)P?h+)^7D6LO`HMC=XS(v{OlwQN)DgwzY|ATFRA!#gOm}(ZT!&2=Is8rZ-Ec06Oh%
ztyqxXdaqGZZ-eLf6Dkuq5WSGCZ}`WlpUEQ=Er6a(ehbE%sdZnYPzRRP7&(~68EIHs
zk&XI96j-*cVfb&j9f5j^1vMwEWV+H!ispv0G<X9VVl0^m`iqFFGv^+TJd7hs<Q@$D
zxR)8Idp9szAz!9)<=3snQ6NNJHkVNMl<Ga{rjfN#1JmpDo}D}L5+?9{e`)P&YMnXA
z7LtY(4j-?(TvIS7pcGjyrPh)69UFbNF!8!=Z5d)2eKK<14ibhLAWdQi-VhDTC2Bcr
z^b^w{a)m>gQUTMEQnQ-<uU$U6>KoAw4><Dv1pdg>I`Zo)-a$%am}^yI*C_b|U{K$@
zYemM}T$g4U43>QEy~}Y##<q@6RTp0tR(xXBA1Kye;d<T}KYXXY=E_!$b9lKb69i4$
zG2q%n>FS4tzx(WdvD>iz(o{R?In8`TG1hCHy!6o8UEm=2`q@jIwSGSA*fz~VR<)HB
z69#{--<yCgBVYS-Tt0=%^HAhuSC!G7!U=gAy>$A-uB+y@g-)h;qZ8q~H~big)4Amv
zDYxt4%F{@r#~LiU)*iR8glwv&qo!`<1<(m_oQ&=l#)6sm%niyp>(O-6c0Rt=Q*urb
z!37^!UeL87x#VDTM^2v;OET7+*W4hp0@^Nm0Yp6zG{K}UG;ufCK+iv1n{R}GEXCZj
zc=f(vAu+-RlJC;vKkoNEeJCcuczvcyussxT3h%r}_|pH(BMh({0uy#m>jprkIqUjw
zuagBit6&-1B)v+K1EWA_bpQ#HNb<?deHNCM28O?b(14^J2ROCLA}T6gutPw|{p*D}
z?AnD85&6ooV9jDCc%tyCg?nxoFn5RGOa(2Y?xCP*KOBYTlh_USB*~M7=+ot6kHh-@
zyj4i`-y%lT%B!aMe4aCM6=Zfta<!B_l3PMZCIZ1LGT=g9S8QC350@60yO7Yu>8UYT
zHm*o-L%n%9NF;(78i?zYhfdd<TIZ}{k6fY7SR%zeBM4EMv+CBq_vL04(W)Gk?^s#*
z(^dGGo_zbSn@R!Y2->X7Dl}bm*jc+RlcgXprK_%!S#|6NKmVltU(sX$7#lZhjOEG_
zK3cu%GH8F{^UPamMAx9``fOFQ+EUmWiLR=hr8E<OUN942Ap2P@!jT2(11M+$tvkby
zQbrB)QbjK`u`s2GetbSRSL|)&NNzjyXj9F;)3IdZ(u0%tzh+ye={o9bKlpf5i)1pD
z?H9IDtnyzN_aDhOF&DHG5%iWk^B4c$ak)woB74O(Pm3<N_%t8%#kArOKicNMbhmLn
z<}0*5Oh=E90i&UA*3koRn8{!6?k||K^ZMG>>g8RLw7I|PC!Rj0TD^izcXJuS+$^A_
zn_rrx3J%Mf9#)u~W=yC2gMvbkFiS9)+lb0OdCmW&h0Yz(hZmznXAjQ5^r`2aG%=Ly
zo_+TU_IY9F2|eh6U!eATb!zU^aH^%^M=2AD6JvWFxv#_(^YGTG*{(|AC5oK9X6$@<
zj;R*dDI&YH?c7zd*vh2wH?qMWt{m>z&CNKYp&bzDtaVk1Z}(%C3I!ih)#G?d%)l?^
ze^m<j8e?)03032}9cKht_3JYKL0r+fini3p@2)SlWam9n8*d*QfT#DIqQ7lrIhWEN
zUiKl%?rJ%8705a)wJ=Fv2GZXyT`ibJt7UVcQ5{PDS~jYoWL9OGfwk#a;CZ-wcc7)3
zATwwG892J9lLbp=nGo}_^nqO7r|T+{-DA2UZ`!t(Ths^&GNktv9J0RFeNQn|H62Uj
z_{0i(#HpI&&M!$opkRGIZz^9JE4K$#2gewSPYHd7?s_g{o9%w#w5FN*Pq<6Q92zB^
zm)CX`^(GXI6AXWXawL*xZ3=j1s=bh+2Sc_<BN5nZ>hHi7a%@;m^?++J77E_6<uY<g
zUob|<Gf496C+a{AW71llGyUQ0C2(|fA^Xmq)V(=Qd-ojC!MoK_HZ0;N+Qii@B6jK9
z-bo9CTx*=($)U@kYsE9SKMo)Xu-`Z$@81qSqMn<b|Cop8g5~$TDrI8TD_CD-*#!VY
z6-R1?B3AMrOo_gNiG~SC36_3On1tTX00Ccrb{O1yf^(yoIU<w`&u1La<PBLdxoZo-
zw97;q$Of^s!l}IqWz0XD)=-eUlvM855+gwcESXunOi=R4jN(!l*&{j_zh3$=m?DJ%
zu@}z_^`PC)qq)VvcX}Qka(0<y4r#A%M+>PK0;Qr&XR@?g#+gI?aagf))|YN5znd$i
znR?W<zm^)1Hh#BmZd}N3YYm#GkR7X--?$K$g{hs3vj<x{LN6to#Uc-$zglZuFHS)D
zWS0d1S*;Z@xAUe;<JOmaJJ$zVHb$FeLLIedt!;ebD_cDG^@?MDs~Uql3vjgF|It;J
zdc=8fXt%iV4HiG>S%q?!JXd-;nnAbBFTU>)61}{Xw?wQ0TTU_SXi96Qt+0mXp_;9|
zhJVl<dX1hFAI9ejU(J}Bk8Xo5M_V?xZqT4@bVV6l=pR%OsVHPowXO!JbY&>9{)@9s
zkv`u(^d#@RJlI?-if)cdC}ibIc%<|Y`yl<#0dn@>=ME9Nwwv&H>k)x<(ezm~2iCOZ
z-K&=WBxJ{1fCQidL^dUXrL`lJvj20f2%{?`aG3vibX6gWdvwlRw-bavk=7~z`;~h7
zyGzw+cXM{a?(@53gpl3tqCQObSIJL{1Bo><lSS;~S~BZK3HMJj+vsc5`ktJC(5*9k
ztv`+T&ohba=@<Ni02Uw~E*dpk%KZb?|A6oH`QU$$=xXG7M<p>{Y~+XQHW)b&AH>0Z
zvDoTiKfT73HU5Biz36~f4+|oX7Pk7&V4{UA87OFc5D0`3q8ZWLKaJrnnvp-T$56Mo
z54CR%M|&PrMk>k+uWOv#vA+CLqU28$a^T0@W#dA<6Sj}bb}pIGPlWtgFSq{(EiY*7
z<8(h%=Q8r%_j2F+u>#%RnfK9sAhyC9cktG}@QpULae8rd4qJ}Z1pk<PFITOhSIy{u
zkQ7Vaeh`wdf!nOOx0QKNPTQZ?*JLepX)+Kx%gr3!{qgpi+Wbrfm}yPoYxWZSS{Ev!
z&({Cc-4?6_Ghs5ft8P?t6V+u!m3A=S&#IL550bc`yYsuXucb0*4-5@zCdKv=PaRAS
z7kK5ZzyTh<T0cg3rXzYAF%sPzof*EmjuMypv(=B!c9jHT51nk*3u52JOjGd($Ri3H
zJNBV}|3Ui~|Lh@l=&lJ1KUOf_t$#VbF6ZqLl%mI7W^7QnDCU1qN(*g2@Zhyk=RauW
z%+)9)rPDd#?Y`K-Ijx#4u~j7O<9KwpBz1qf{M4r7la$Y!pq+d-2T+*Y-__u2x{j>G
zs!eaPBDj!l5gKUjPiO%l@$bRh45(!R2Y9M0$mpGE9|zS6R4D(G!<h5Bdvb6MTd+XM
z?EW5It?RO2&vWk9o{OpLtj=q9{<GPkfBVIq^Wx#9<&fVGN9+E6=G;6-pUC!gR7_f^
zk3m16CYCG-X;q<^-And3a=;AMu9g=sgmCGwLWxn00aX!JvhP|#Dl_&-+bnGYngPBN
zpEQ@o)4RnnNy&TO>A_-p#%!&3qfiwiDP4K%W$P6sTTvO5i}S6H&eJJJxYym%ED9E|
zJkbb-)j#OSzITk=F8H(}{v0q7gz~`PEq&i>#|rkXOpC~@7$j!uzYF}%c?*{GLXlD!
zcxv%)g;>v>6?*XQ?~W~`8~v9YjDo%%tlB*|C<7iE#Js%LY5EfmodvsAN~`69G55c3
F{|C4|h`;~<

literal 0
HcmV?d00001

diff --git a/assets/screenshot1.jpg b/assets/screenshot1.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..0e958474663bc68257153fb619fc2eee87508fdc
GIT binary patch
literal 369127
zcmbTd1y~%>vMxF}L4p&U;7+i?o#5^;NP;sA5+p!K2$EofdvG5df(%ZA6P)1gZoysN
zWbb|U+4tP<eeb?IeBJb{UR|qd_4L2$ud05Ud0GMBt0*Wd0FaQ70B;e0fTvY}y}UQr
z764FJ2e1PG089WX5(NMaafF1}B#|ipeXM}Q3PAZsJ2K)VF95Lz5Fj=U#P;{|7P0-k
z<{$sxS%95AxSef0UJ7vYar5#cMw$o60kAMIFfq`vFflQ)v9Pe8li@$d#eGgqLP|i!
zK+D8PPfJh7%pvrOnN5I=j-E$`S3pEeQc{wM`?a#1xRQ{Bq}X2}NLbj|&vBnq;p0<@
zvC^}O{r~<vwE+k)QM)nxQIO~W$OK3z1V~R^09u6hXoz|FJ2n6MAR(imqM>78KEuLB
zbb#XnkdaVOkWo?4&`?nky?qhq0jLCMgmk<z=tSBU81$~hd;zgJm<+O2?Ib$Cj~V$b
z-2$Isk-i`!r(j}cVP#_%0165Ti-^j-mRC?zQdZH`(>E|QG6q>$+t}LKgB{#G-g|m^
zfAIP8H7GbFGz=OSpOBc8{4FIlH!r^cR#;SAQe9IEud8opZ0hLj>h9_N)z?2ZJ~25p
zJu^GEy0*TtxwXBsyLWPWc7Abrb$xUD*DfRg%0JEew`KpsE&_yI$f&3&s2G3kLPGXJ
zY$yb%Xmq^jgfiL~7Oq6}d;yrmvavZ;?avtab&g3a-F{<{G779Ro%}WJAC~=}85a0|
zW!b+C`)|AE05~W}h?|E(0FVNlLVjK>`xhC+3z|GLJU@cPquWoskh@|oz_-XGn>Z07
zDMl@*Q7P{g@Q2%f=oZ~C*VH~I$J&C~SF`&&9J^{R5C+6^*gPxSPvPs*qmXk$(jZ9_
zl`j>v+9r-EsH*<yp))Rm9NR<z*uHRSHBPLa2`1KmiNy-Y^~YkBCa!JcRh35LW&_k}
zA*rzW*VY1Zp#V>ars}@VB|(FzjvgsD8#sV!d2h)5ish=Ogt6-QRNiqW!m^fev0fOt
zLSd$y&fuN-epNx=>yW9~G+^Nr#E&1lx~275S=v%?O@^}SnkA2OQ)=-1gMC>FbH8u9
z5`^$83{DpbO9iNav$>fBCL-jtgh2LqUe~Hm0D4Z^izUr%vvT{omny$n%<8Rgr_Mjq
z+f+GEXP3f(LKIjWyg|eOZE3Xs>z$5m{97*KHr8xpDByqF;h~@G_C^5dg(Y1c`#F+b
zzHN`eU1Z|`d~7*L7V**opy*8$5^lZ^<)l5D?m`B4;TNTs>%82)Y>MxaI|9{^BJ{<Q
z=Xg_pAJ|d%TXB9kl-6<IHR3OLNgZ%#P>!*5MS^1o1JnJiJ`_Afm96bu<>A<N#a<SB
zDT*d`h1JMtX|7D!S_CS6K`#hLDb+&<8f;9N4-b+R_)}3pmH7%&l^tG;LMKyf7x<M+
z)#<{S!Ybz85M;^Gh*&$?L<5V_4S3`A-&l+<q}Xi}!?gFIXPe0L%mJsO<F7QVx(#Sv
z%@GM5rF`!gU|!+lH(p}V86v3E;J6AM@TSHH4y*`Fxd6T$ID02tDVXn~V-9PUCsgFp
zsChf3!O-ZowD|>x+hM4?yG;Y%-FVpqHO@NKX|Yc>js~z+n9{$RaF%*$Ip4YGCd-l8
z-E5LQ03XIY0yhP%$vdx_a4M3)PIzx?oc1Buv1c<IM_aFAp<8f5A)IMaa{Koa9P@IG
zom`HxSjP_gduH@z?HXXvp>z)<Snno8P(N;kFRS1k4GSZgV&mic-M0*2qpgyoC>JS_
zwfr7w;6V)|zw54|#Ez`ZuMgcI@|mS?<=WxYkkdAH#N7z4%ud=ep^8^}F8aPqJxKhE
z?yf8O1uFxU^E-YOKWqvKhY_VziHe%H&`GWjn~q>4T9uo==9*SB`aF=5GUG<7e65=Z
zu#;1~;5%!zg?+ngQ^^8&h=u7)UQ`25%hig$ilf~Tn&S8bm@8y83r~0<uT0AnX(Eav
zFKlJfU9Ha`>O^Gqu9t+VB4x8~%|4;_dm?sftZb>jx}|py6DbzbN@^!I?PX|nA?54F
zm5(+QCW1%hAW-c`hW*r?43dj1@D}SmsR(4{8F}s);rdIDid`{O{n_;^zx|&`2Ish}
z`>C6mNzC|Z#i*Yx%;w#f$;m4%(!;IBq*h-KJOSje{8s3m0Bd{Q_3a`~Pk=4S$R~hv
z;S(S{w@bvd>Is15sd@YS)gub+BYVoL_F`+<pYtXBPXG->7id90M*?2k2W6H=LdGXR
z8=|>;EINQ6@8(<l696#s1bAWSruYQth)8=3{@X$BpB<{70HYmQ*O-3SP133u?aCI2
z$#Kbf0@Ujw&QHYMlO<{1M8Evk_p$Fsf>MEFJ$@VFA&-li2${Nm9U1Yor&Mq5(WUP7
zc^V0>UvVEGB(OXpXFYPq+>@q6i{ARJupzA4!H%nJ@6#7VnDpxJS@sgCwwsK4zYggy
zkBfMJWmzVawS9V?{1}Y-U-}{RjsB1C6Q>j(T@cCy(Y<=WdG-YG4?)aLvBfnPr+U`m
zhScRA!U_?j$)=emNg^e5sk-DR0MgGp_BYphV}7GHe#f*=fa<)M{x!c2Sp4G`4upW^
zheV83fv<kQz<$S+|2WHSw|GyOw6T5X`{HpCF~;m(<vlCdZ^u&e9!2Wj%)XXUYE=#a
ze{zI=E98I8La2j_P)8Xd%?_bX;XAx*tue%`c4b|oBlP)Sd+#D`fwq!7%-pV2`9JiG
z*@h2d*L&ibdY1e}IkaqDB>QZDZZSYx4LL&^jSGOB<WFo+kvyT-`ZaQR;#)B%h!q(z
zmy@a6+fM*<hg?-Ae>V=>0k~0Y4bob9sjH0Dm6i$gtqg~&v0>y?0GXj;y!PA^pqv<i
z=MR7JyenV8-y&ZCkTAvu7-M6@f#~5*rJ^i85<UTbg=0K;#otVWLG+lJ)?vY0C_wW|
zx|f8k!yNWz0$mrR({Qum;s&ljXu$st>pEJ<zyHsHUSP`oH!jw5c`ne5jF!Y^i=%#l
zmZCX0YMQ{M+-ODZ>w|-Sk|6o>d|!KP-nKwKk%DX)c>u8%(!X~Y0oTXdJ)mz-0NC;!
zYlV}vcxv@%glsW!ntQa*+W}Wr`HhY7vvFp)30dplqTz#<WN>BMY?#s!8tW0^wL*XI
zPnD6ZX`tIo=#e3=2GtbKuB;sfiNkUbb3wqt67l!IH_vC)?&HbIcg{;qkLF`{%>|0t
zMs!QKf6HZQz@yO&F~*{++Av@0H)(~Ad(vFit$!~YC~|5U<d)de7mnNL-*v%Ou8+{q
zadMP2n1^;~M-G^ft_Kbi4>KS)4t<7ry3qlmuFG?XQ-lme>MU)jgCVB)mS*MisT>EQ
zR#rOCwgnC&G}PTZ*pglpb#wMcgZ4YSdh*!O-}@z;7=!NOH=U^r)mzsr&w(D~s^FrG
z8k5+LGzc_TxhBUU1d}*3(GpwU3x%ZJbMx)YICM(*cxJ;Ou#>7d1HPk`^MXUbN=MT}
zj*$bC(N(XjWF*jsqNzS`gPO~1z_nH}!rwlLY=b(|icHoO!(%W*mTXPIDKEg3rhwTr
zGX&_7n|lz+sA%C?<oSKbp?vJ83>13Hm%~BF*w|L`1OV%NK75-Qz!t@ijCHsHUA@<Q
ztR^{UZa(%S7!h($XV_xSEQDSdW0JXcGqcLt6lqdTHuU>@oui~qG4m~aQp0&!sZ#1T
zi>lumL_K)92X0ljbi~N%Hv1h$k<m!4-(DaU6zS}92rHnY#vrj@>hJoLM{=g$+Bs@M
z#8}Ye&#q6ttFMq7H!(%}vf%?!Hi@=vMiv=ap$;#+KiyEKdqa0b#d^FNChyj3Y}5h5
zt80ykf)|hsM$&2<Po{k>3&Ay3_!#!}o3p$oRdf7Og6+~~_YVk31vjI@jk*QoN=F^J
zG<{{NK?1^n73~4`w%p7f#|rKpGH^NCOb&L0F4mpkV$3K|uYc;O_3$Sbgu9r{;doq$
zYy5feTH$bSqdk?de*#>b!jGs_S0$5jQ$qj09VDbPjHoBThR@W`{Zq|tBjPqTS2;j8
z7bY_FEj4(;pB09r!ioS6w1R9kZP2aJuFfTeV2BF4Di$jNN$+iq>G}3wfACl8b9j}l
zk3Nq=3r2uJViD`-J=ux2K)G&Pv-GV1vScaman2{e6qH;-J@GC#W{N2kjQf&OJsTAP
z8DQ8LS>#x0CS3aC&kmOxg11kAic>$)eeORD@Q50q6K118Ebh!YFw(&!3@@n*;9U{Y
z*_66RDl78bMkW}e)u#c9B|v0`W5w}n!sxz!y+G4kFJnkcTZiYbbNdeu_wP<^6xGEY
z$M3T|t*-HO(>dwh8a?Ct!&FapUZ-!kYx}{L04!fz_Uquo8E4^JgXfO>dSgr>=FO3h
zmzyjRQ{`EqD+!0`ADl8<ORTgDZPj*JN(2d}<Z&X%HD%c*N8SF<98o<t>JoK3Z8KbK
z^1-)bE#VZK-;fOI2ttakp{#P8im?;ZEJYBt(nL$M5f)%Vfv1N8shpEDdG*mljr{PB
zq*38Byh2dIg@9+3OQ$41;hw@6*h}6Y{UgPn$;k=67?~Y;=Q<H<QK<EQqA1b|jkwea
z$RY1B&F>($@u$VDzUgmF8R3;tp-L<Hwi@CP0MGgqYzRUfkq>8e;AkbbVoYs>JQn!6
zsz~7%9ES__w@hNEqhU}(j2+cf(tX@jo*{o1lF9V3EHgPZkxb;oQ~O<cWI4(?C1R{Q
zu!9ShFEvM<t7*80be6M_i&1MqA=-Vu-Usfk%wlKCm-DUp&$4L$7;tM$+Xyo&G8gcM
zyQJQn56Q->X>uyvj%~#gLYX~nLfooGrgG27lReioKSPv`a7kMF%ypWhBx`riHqXqk
z-E>Arv?#PQCkwGY?@N^MQj#F;r>buVic}aS3}{7+7m$(c{5A+1lIXWi^qm>9x(<|n
zEtgYzCkGv!7$trP!gHLKA=n*yH|0|U<B;93$1KMdPQbaU5a%i|)UhoKV{ybUovNI-
ze$7AC3Z?C~CAwByv|I_A%ZF~wfl1D`@Eu&AiJ;TkO1!dt{wyZ_<J4xsbQ-=0Z~!&;
zTqurpuluXoN6Q8YIi|BepiWPb1L#yIUqBS(Gq#}pHAyhL3LBd6g3VBA1-tNpSwc+9
zG=1u3!=m6UNY6X{4=RDFT^J?5{77-7ChF0V`;bw)#BRTErC!1-6%El*RVE$jVTb-7
zRxSk)QagHk;d)THdyK<Um!exl{x}Axj}prr`ZK5EP>1#^i0z&e;y!+;@LU37P8O!!
z{izPP<3nKhx-89J8Pf^AF<hHn%YJDchr5;+MTn=eQvlr#3<s*Yw_%EqWW1;O9W^<r
zyrl4+^lQ>+QE04=JX>gN1avsJ<=a}Byemt}%a@j4ZOs$o)g9%oEBMq8KJNm9D%dFX
zB0d^wph-Da*a-|dkuIibDD2=WziU?Rt3WKQElLzu+R+|DD!e8>uiy4I|6s4u7^RjN
z!EJs5m|bJdA7k7{gQOllmHQ%f)k&=P%pLXy-*b2s&&e{Sild@hBWv*a23dHE+-n#2
zflD>lN+o3b^%sHpTC8lbqdeO$*UR~d+;vW-gFP~oz|(WJ+NRrRJVUj2F+5Tf-6wus
z;Zo1{=H5r>bgkJMHo8SDnJVVt;7~=J4*_H<?)An67Twk0^(i!C?--j>a^rb($aT;~
zhQM<UTBg?*gXVOpN^^bIW<ve%0yM?WBK$yr#2<{;G8V>ZM)HO+J^VypVy7eQXdXCp
z4Ne5ZbBz>zi<WzFTRM5`xqkixSf>IiVb|lGbFI5I-zVZee5Lt-L<UHAZ;0^PO&|AI
zrt)?%ob;<c019O8r_<-XU9X@dlga?N{J%l@{{V1_B7|dC^bbypMqCTfRzmCMX7ksU
z#zw%nB4}ppPH*6?!mnL<84GzHx`ro!0VFf+!QB_(?y>lvfu=$VbP0hu;`#*w#NrgP
z-~gPEFUIv|I|O0xuObBp9_(`tG<B_gf6H4AP%43OaRm|pb1G`+V>(FXW7!V&cy`wi
zrc-Yk+L~<Zv;6I!pwXX@KIOA7SQp38jQBnG;Vpra7(A-6o%$CBqJbLvNqp}Viz*FN
z_+e|e-%1zGzs;ypH0*TJXT9M&AKk}d`wxPzXZV5u)G@BsuC2)IF9I1MGE*C@YkQs5
zp{e`Fn18>%jUC=8LapI5?%ezOWm6*4kIa>g{@#=PH~e7B-gg&e*7Kr-N}(l#LVox_
z)(k6i4F`0(R{fj!tv74L%ZHe=J(L(rIXMe6WANEBBb?-~7pCX8h7c!rk_l{1^+XF3
z-NeZ#&}hxp>j{HG;s3lBxSXX|p|l`5POH+QdwT$3PcPP<W>w+ne>;@_b;P9GlIEzc
zF^z1rWpqfe=)d0TN^#Wa3cT!!$%n+&PvwFbR2&Yvni?$}$%j?#fMJ(O4b74KZ5vZy
z%D%Xfk^o9nN1PQysZs}<L@SIWk&+0dgj&v>7sz8W{JxjK-ABR`*`aq_%m@eWg+!#G
zj#t?HML`Mgf2@J5C<xb{#_!iY!k3zMc}Or&NZAw^X4;Kv$Voj1CXFq&`muY>IwfB#
z=XNp6_->TPe{j%m-X=+*N<je@NQA|l`<t4}QNF{5x>fX8pEHO4ToO?GLCBGUAwrV%
zPG-g6bViHr^a|%H4;3ZWh$1XKwaQsUN+pT^_ql-GsFUzIszJVI#0xRaO>>fu?$qzz
zS|#yE=A#_yVjeekyN|3YjKT+J6T3Eogm3j?dKtd>gY)esJ{T+Z%g|@x=JDGVIN|n8
zeZeh`dSg2N3O{Jq<W=bK9Jv*Zjf2TeME3JUvJm^P*e4)D5fZA=53kb=aoLj-qf%u1
zH$~nPhd?X2ON95$bDbDs^W`64{+XI<970Cl)aYGICl&77N9RUi^j+M`1@^wMxBETl
z5NOF&$~4+h+Eko#D3NY>c?e)SB{*K6TQqjmM}tD92|l*POXG!g%PAR|4m1cRt@Nno
zor54NOIm507^9(Ci#%U9nS-!rqc<jZc??)I4da%skG0$$Z<#~vj|P`I%VK<U+<492
z=VnJ3(r_3=K`99qOFrs{3xwsaqc_;Lj=zT_T!h>I7+_dICZ1044&Ac<k+^45FI=TE
zM7G?YnRt+%XLP1~ze7Y*9>uHMB@_9Wn6DJ;KyVjyt8bDt0-F3AeRAYiD~tSUH3o0t
zCyzG@((@`RdYIDllF%8RX<2nOfsJ<J1J8s*8#5z3<h`tB-L;8h)7XPsN)pn>y6THx
z@+th}qmc0|p0G?5r7DBdG5&}q|5dg>kpKS1XF5-TdsQ%%v8H97blOyJYYR{Re%}z|
zom!ykH>sA|z*~aJAex~c0UoRqTu*=>@PIXrZ-pu1btBZJd3fqQ1+yfKjq0t)(o@f9
z#j%`r=pCP5@VxnH756Y_>v7^F!su=S5^((foXzqPNq6BGQi%wVmX84ZeqH1te!czE
zqv_bIlKv;admra8uC6)9)Cdu5t`=}hA`9yJ501fF4sYjAepX&9i~Ks{2i2b!SY)e^
z&KV!|2Gtr5Tm;JHjK+b{hr&PR#Rx8)1UUjkGrx+tNuikGh1bG<pN~8<3l|F?@D-I8
z!rl17yo8CvitIwEj<qjIdcx>qdI~;?8!M5<2Dthgl+7G9TsOLUQ4ul{?mB2yTajz`
zPWkhAaM250YgX1y0Txm=*g>zdBCM$QRCk;?>5rY^A7LZDg`hdqRDtLZ3I{5WZxR7w
zN||$aYvxt&1X2#mVPn*hb!~Bc87(Uw%<*m2i}T*Fsb2ZOOJ6!wa`}h2Jd^i3>sc<l
z(lqV==B@t)GymeU7=tuoPk`8c&=a7tPb%ZB7D^H9Zx>x8c`iic7s*`#>|d*eJc<io
z-8T_^<B;NrKH|W^*8$C@Yy8Ku@a^&D2V9#eu6@D)p0lgh6=|eN0@fs^4?&RuO3Tx&
z)l@`-Ue{3U7sF-{QGlESM#t!8B?zg}oSc$9I6oUzU@iuKi~<1hDA`ISk{DPOKfyzj
zBY#V#&VsRTR@zOKw<T|J#9`w-8z=XWW^3YFFZ($$WX-tTDoV`$xA@>$tw%+RK%r=<
z4r83cQqtY|j^-WExv}u1I;mY=y5rwLTht1ME}!PbMWz--0**t;4SjhET|+yDB8N(s
z@(*@mm~-+P(=jli?-K0=j4m&1<(bP-{rCWypL8m|SL19|x!R`x3af1E)7595Fv&{&
z`gKEwCU^ZMk?Lhjdb~!LJ4Kz)Oh@y(kJBcr4@R>Kq1K-|St?z823BJ=mxH2a@Z0d+
z@Kiv%!h!!*2<Ji9uzs9=FakTPDP?1tt;l;y>D!_3B}xIuOheC>z!K+=)>tLmx2@f{
zWy5Ws4zt{~nTK?gDjy1cx4ux1M%)PjN9`ZEn(N*s_YQ}{Rzyg!3~U<M80za9gOAWs
zUp$ZC!z_ll$G`nw=nX3pp31<F0pz?^Q>~OCNK>0WJ_kPjJDCaxlgTD#4fkar2}#Ot
zbz}0Y@faw!g&{D=$$LoP)Q<9n!y_<NE}kw`fgoU3LHVjIOjXqb;rbOs%)%mcT_n&z
zwG%L?K5xI)w#cr|6`P|^@7GUSW3zJTd92*qnA^A0wwMeIF%C&Ex5%K}K-quK10K@J
z7GA`K_VO?y&?RpH!mas!>nCcPfjioX<)atEK+S4RA9=eqrRG`IkPxJMXo#c~KV)AR
zB|p5<NmlIoYT1LMIX!kNbUt(A=htAT0!)U~fwYh{cuY3%&E{6R5x&@z!Y^BqQo-IW
zaXV-*|I9rB4&k-s@}QQI6?8wRzFv>EnPkgNR>upGQxOon^P8;P)O#Gl6ON89AS~3Z
zQm&O_%9r>Qd!vUgJW3@O-BGg%-2qu^cg3l^80dCjPI-}>5g3FzlYj79z<cr35ZX8G
zXt2az4`Vw87K_hOmAnS68cs}`SQYSm6`b4dK2vtf0|{yzz&CC_5kUJVs*#Xe?nUA1
z8YR&SB9NqqyXO-#v#0_2lnL)i4%3m?j5qsPwRg1~PD0~IhP?}^N11Yd=44a236-n6
zZSbH>H58W3&eaA24;d|-MPNf5$RZV@Jm#Y;$5BK-6o+)5S#w-Pam>Nq{VH@j906vU
zr0QcJJpt^ErWwZ%-@}!8D1jpb6SK{vui%=Zi-GaKDl|SIrKOIshB<<&I)|Sfjx|AC
zzut4i6EyyoTV-?5Sbe?(2Mw!sksG#^9loGe_%oCPL<F;G`V~A!2F$J2kRD@c@8I(R
z$`a2V<&P??&kM0GG%O0^)mVrzk$~bXSDINWV`8!NdxTumG>M?kionvDVX?$FvD#D2
z>zWmioMW}m&dp@|{_65}cXh%?A8{iajOEtgM_4yx-QHjh{;G)jy2$k*qo^s~d-@Wd
zFx)s`ts0M>K1VD2GyGL)AkbdE9E04!LNck8j_O4J6;~W#i&uq;mrm8_TAALDM7nxz
z364x5T!wX5+DISbq+op^+D8ALbsdGWg~_-uLaql5_SOZg92sYs(%_9>`}Yzq)xV!n
zm%zSXHzH{WR|vmas`F62wiS<NP$5pAwkwc(gf>zauQ3e9F(+*gPyY?Hn`8nIp!!D<
z0ji0W9xHb&V-vdn^02y2m@_Ju8F8~AY_I`xNxKxxSW_yiKs+@qG%2iWiF1zqRzs}`
zTJ6bK0K(ycTBIE`kMsX<aR1Y%L?iwmUOh3Ob=UrH6obt_Hz@{<M4C9)pA-sUvq$*!
zb7Q^8FcV}!*0KRCcK^Wr2ki>aHy1+}C1E(J&nHVEb7i<n2;{~^!^=*!q>ni<o)&nP
z^1@7=DfD5J<$LURQ5}l5s@mXyNP&cJm|!+QaGEr(b8u;@rCq}(_Bv&85}dE@+sgO*
z^{=tV>bIPXVjoK5r#-ezeu^W@h))|l{{+WOZ4#<|bA#S%7!BGkMF4ciZ{vvn&Z(6b
zsf^AH(c@I-(6?mXp2V#&WHohV_18j2=Q0QTn(4Ro&M0B*8HY<`C7Ea_HYANABI4TA
z19WZTI`#>Dk`rCNdzxaNABw-<1R3+ltBiCr2f0pNImS6T2RCt0j<cgKo>eZE*%2wJ
z>i)ieRpQ&!CeQHsk{gpH-UOa+_B&!SzXO{44@QJsy0j!`<v2druCMs{9M|E)R6z;w
zVt#mG`)K?7tsz%Y?U{;Ci2=m@qtWbm*_s*Wa0ui3lCg!?RdIvOtw;q8gC^lYc4Ex!
zSFrizHzq>;ccUkVI-_eBqE7%!?>??0gQb;klf&dDg{$T{#Ki&-dVKw08IpXOB>ob3
z!!7&Zf1@L6K{k5T%9Lcu+T4;d_Ir{7iH_csz3C3{tbUXjGJAFk*9v7qkwlHIu7S&o
zoakb+qv%Z*<eHTQ=nD*%E#NbIIiFawp=v5W?*RR`j7sFm5tEoHTZ}QSY_Te!O&MY2
zcnU#=sS4Z$b5#o~1{e58p|8MBU)z}B*>{Z!7sSby8<j+bxhV){_^-$nB1$ASa?d<h
zOfA0p1jvSVRc~!?{fXXI8_#Ln%=dpK#Wq6XbkJ3Wfpufvku#vp+DtyZUpFV;^IU`^
z`q_<QIRu$exi6EG&8p>{B)rg<KFyv;O7u&Ej}GK26tZF~G4YLxN*~?eZV2=%G=^3q
zVS_f&)**~H-hfMi?A9_@P9k~rp6yj|&WN4Ep$@S=azh^RyyNdy^jb`x(eH{o?2Yss
zRUopHQGG)w)n&0O)JDEK;!)3?n+Nxea&kXaRAMT8R*ldGaLSQ&g0r_HEwUCSfgdOG
z(fpMaR|I;cY;7uSO(5p3ab06Y4>Xy(T8Jd5^ggDlQk3A~Vz}4O)@_TQ^Gls3NOj(&
z=CFMjGS)a?iK*%`uQr)1X*gCQMexRYr`9D|EqeEL-cSP;N8B4C(I5SSZ-b_gx($-O
zQqBy4B!A+Ca$Yh(8}ZQ~lL3A7iAU%V&U)sAUgY4Mce3~#^kW9feSO1Vd3!do8dp26
z7oh&ek#sitf+{~%g^&ImRdD|gemub5(Cf%u2`r(cyZj#E;+5)$nYG4@1MtKbz0iWM
zowOa0KH#70TbaipA8%AQOCbgAHR$>9yS`o(3vvmzP{~A5zfzY1Q!l22H|8&CRZ6*W
zAzTr8joQ7!FrG5?(Qz}3z=G5F-F}zq@@C-byc32=rwc6;tPdoX+q!-nF7D=(zh@5X
zf-)p8x)|Qu7o4ugLsdLSKZWJcoJn~BDHfF%T$P6w3eYSfgGK18fK#(H@~p4C-DWoa
z*u@zI;!G0PCwxaxJhNiPDBGifs;+zFkke9adHU$450}MGq#>A$A~$$6lPWv59*P&G
zu%m8<rm@`9g9N;zDTP88&SW)NqTOz{PUXckb#p;^BpWvyj6P>PS2OvDA8G&*<P1gM
zAk?Yt;j1_M(WAQOi?Z7)6LddnT&cV8eh9KZ#~VE{Ul%6r)%wioJ@Oa`Mx@SuKLJjF
zmeXrl5lm8o`H1Wyf<Ua<H$G=0{kdv$UH=$}NPGF8AaZTA$JO^HPXHNWM2^rKksnO#
z`<``N-C~Al`kU57G+n1kt=f1uKL-9y$ssN|5isRMT=l<pd2?+n`3mt2#rds>``xR*
zz{Zkfyl(!Hn{|yxd#(Khz-H<Gc}}2tj}d*(+F&HLYJH6m_%9t!1jbJ|y4N2q|6>65
zdzBN{C%`x8UwNw<2t{Mp{@h{~KBDwK0lI#_`~LH|THGZ1<{vGO#3R?nG&~Z(zZxUb
z0oSR9*mQ_7ynFwt^zXk)`*r*}iGDyuWN(rFF&7O7Umi&nBm8z4v#xOwnM@)L&CthJ
zBSS_1HHIhY>&`VH=o4V)<v*;@LCj9iiQxk(5kkZ|B71B)Xn=QZw1mieW(WR5O{LEa
z-leps`2RF>%fd?V3ULGI?o#9M?mVNP0AVFr$94B$#MCN10gfr89+}GtY5zw{^tH)n
z0ffYSzkh0_hVn?d94WQKM*E0@Sc1C{wqU#TWStWu<o(kSl!S>_M?iJik9YNWuJ5*#
z$ymv)rUSYSk^Nu%lNYsu{josO#JYe2;x-T~pdcHK8seh5!*27P!5ljHJ=gzb(-S}s
zk{<uydOH<MZxWB7KjrlfRSX(5iD31=Yv|j5a36buc()G%btv2UmxRD(9oz4mPE06-
zsbD~GT+A~id!*>GQaCUF;twLf|H^$rKwM#1QA$>cO7C$b_Ts^dfO&fZ%(_<qA|0CA
z`Q`>}Rh|E)aA=U#rs3C*r)w=PZw%@h?dM;9F~k^M!d`4l#A*Lw3E|uQm2j5nkV0rJ
z0oDY)W~LL~(4A~$BbFLm_uAIwI5MvdbBye}DD|9njvymSd~jf_r7|zHTFrRZUL+^%
zlK8`rWUpxLwV`fI=gW!Cs`sM26Z$FtSP3wdw3LOBTmE#<-5VPE*{S+#KSzGWvaida
z#O3Us1u?2Ns~6eHt^VVD)mJYfaiSCJltw|a*P`jT9w-8gEm6o-=D{?*lmHhLrHyV^
zcm+R9qb#Mk6q%X~oZ)wlj@xVdq}374?Y-f!6_L_D2kuH@O)~}c#QQQoO1ZOG9U2#c
zDb?MjkJunqG5_U4#H0e$#(3KV+VX(Eq>a-62DW)09XnVzT<dfLGB3)3p6)*&R0mh0
z9KRT%l<HmP+~f^g<CQD~Tx&lb&ae%OrR#V$WQp>-9C;Yoo}$)mj)(C_;Y!XJE%BNd
zt);YWuz%_1LRo^(?3_1nbJO6}^jJqYThH;KF@UG4`cKH{m%<y9?SIkKqRnd19?Tct
zhuZ`iztfo1(!c;&<H8nNa?t`6!c-+EX<4b2vn##h`pwv_KUhILl}=jyscf5uNbLj5
zlG4m{n>>p?CN;)Q>}F4rY9@v*h(=7#91%6YFNIHi&7Vl@`x@%QbHD3uoQC}`Qvc~;
zy-w;lZ7=UJ74ftha3$aSR!*hl3QA|vkC2iPr5kifvNWq|G90)b^Cppm!_t~*#Xt<C
zoB63CnQG1!=}6sgJ2oUaaw*8`-7czz9_k<7zuDGV5WEZXTNZe=_UxWnf30rs3yVG1
zT!rjw)QET$SoTB$lEKS1si4=v(dx&ZC70&3A@;urfgby7Y#V^khDttn*LYC|jUAib
zu>4S*UTE;E7evhMH0oIDrEzQHRjwBUVMH6@<QG8Y3+(c)SYwqu$64L#3F~T=>)2D?
z8NK57o>n)85jt+)Z=40i(zYB)+DfjDhhA=~5*=!ACR5`~jyredT|Jk=TB?QGgkms#
zt}RjK9aFhYO4OGbz<GdB8U}+W%A^T>_sQ}L+;e@pza8%i1@vgW_4!_WBv5RtMd_@l
znQm~uw+&1IEi1<faxoyV!WTL`xnjZ2p6KX9H5KC(BSG)H+f1Yp39_G9&w0ykVjaLO
z-p6gxd$-oO`NAuwIc`&7`Z6gq*6B|6^ZN`}U&{l`CYkFA*@XDkh=sg`@H>TepG796
zrlZs^=(~RGM^H#kM44g3lt-V`=Kgt!W)awq<<Rc&rhtj18nYp{wtDC6xeMjvO%dR?
zcFko9-aR^OQVeCo#jRtHLcof0%cY~pOx;DI4dOd5HCITcETJ%07n~WP&C(}6ZRf&H
zwjbAY6$w7jX-*XXZLpg0DQB#)Y=|JOVpysM_5|>3`C=-ytGZv;%1JQr#$A9z<VG(m
zdq`N0gU9SE(H_UxXd4(R6I**Su@;;6DIlsg9-#LrQ&hK&4zt|yb$FS|{3D5m;qE?2
zdB={ByD$}Uc3!3)OPRS?>eYdbSPsN5v9%3XY!p3w6cPdofR*pD4yX)61aG3{UwPDn
zDQ@JTLzD2G+Qyc;a8Ix-)#-QT)6xXLRsOHV^e=h!mwPrm0{ldGjO-Q$=BgRi!=3=!
zfmZwag(vQVJeMb_4Xa}x!zOJEPH3;3a%qDUmYBc!&g9V^Q%qV+yGfKDD16fWFry%~
zB2d_MHM1K3cun%g(7&}7TDev(9Vb*odo;n}Cvdo>H0w~H!ZVWvW^;a{{`r;(tL8h(
zcOY@<#aY(%wX?+TM=}@7#ti|8VgB|Zq9({!r>@592_W_NQ2StgMkJ-p%&+2}oxrTD
zM+~Vn)o(e!FSf8cGLaFxI@tz5$@ZxU@P<1w(4WNW*AOtL4^fF0Gd?Kmk3?2*+6|yF
zDmd*e0I-JY+{ngYBq2H*B8fN2k+m1T<9V;&UQ&0tWVta_YWcdcp7-sfgC!yYCtaFU
zXX}>+7pP!92y+%Vqp}Hc$W~x=Xv^rz4G4$g5w|5_f^>UTK1~ggjW}>+R~DlY$8<p7
zvKC|`Ck5~tmOF#}JoE-E%7-lEZH;OZkj(%5<$jtMz5AfoVa`S&<=}bGQbK{X9rm$I
zR?g4%%-=$-fwJ4J$b8%%#}6=ubm`s$oAn6$8TO4y6{-6h+kn8AYI5o<#KEs^w0y<`
zP+g_5)dl>m&WrFs_G(uyUzbnm+YuSMjmP(M2H}>nao?V$>*ykNAQ<Vt%P@(7&*~!K
z9oNYU(fYddpQ<Zm#U{>-GqlZD+efV4+t4Xn2~)R~-Gv&}ZHmw&!jCNYNKmOLK7h!-
z1u2okJYIg$5sQoZ(UYDZ-8RZBsPN|o>ZV*ZS<KrtS(SBg_EntYG4a40f8z6UUT%%e
z+KsyP4+bzt*HUinKUp}B_?wcti~^F#D$bv*mn0kNqP1Uh254IzYg5WR<V{Me2%eD|
z&t+6u)blA1#46cT;te$^tNQ3WX}B@=c*E-X9e*X(>obcu>IW{kDRUA&W3c_TO)`J|
z(eQ;Vf4a)Imv_?=@)LWo5!2t5Ll?PKcWoA>3U$l6Nh(7*_SrH22+6@x?a`Cw_<Rg}
z4=)$!PnC<(%X9J>d{w-Nx>0eiFg7G0(I=TjoGWm9B?awPFJDMWW(X(u!*l5cYAgB1
zra0D>%rWT!PAlj8ONQfu@%NPT-H<S~4+hW3v5?|4W-od{gHHCK_)bBX$VSi>_+<=d
zP}|MfQti1qm!8?DL#$fH7t{v1x6y`3j&y;ue|qPcyzdhS+deIO>rvPYdR`?@V{;5q
zbVl?|1WM=d)*kfidJ^o4MwETN7$<l}yU^o)<7)hwye@TV*rX`w#S1T7?2vbb0~><;
z*ZR4Xt+UE6MfEdVo$Aq9)Bx^18d54_$3Kb;JB3b7O;vHg?DxMTVwLkCKg`4<|KRTb
zhL5wxW{@4fDXyr|nX;$~GT!LXEGOQIs}U%s>}N*njXcclJ3|o0c&`&~rR%!bzKcV>
zjr<@ctEOX`M)!)HLwAq%v=1n|X6n|`f4M5s+11~A6|c#k``;0TaPk&&>85PN6dpoW
zje19-deL5#t}*NpOPgR=L^sROgaI-3-QeYro=m(tzX_j4tM(V&m4>rPn7FegxZuwM
zZHJ62eZ6y{QIIRNuz}IgroByI6Eu{hlm`@0#0@a6{=H*P@?MfA(#UVXprT?{>$N24
zbnlgKhR0H{b1gps7SBfm?kIQf`9}KkSMU0MemY=yi;oM})oS^u1Tq>WMTvm6J9`LP
zD6f4SYsL3#16!Top0VAu^};>e<Wm-NYpz4IM{bwLzEm#PVWc{=rYkp7%pK*rKgKL)
z82Ey2nihEP!9|MSQ(uY}tc?6wmI4L9#X8)#>$e<k7-qt&47Hlr{XRbdt|k?(l2Pr3
zKRq}&#@ZM(rPuS-Z)?2ZmWtSzoW*Br@YGHgnEV#K21B{uLit0a#>abPFH2#Xxs_`4
zW$7EHQD!47*`fy*z`S)XV9~c@kF*<ac$jcF;xH`F@KHgAG28Hsi}R-+@NG+AV&v|D
zPGdc@YrJ7@-NVe*cL4>BM#r9ht^D|?#P%%gp(69LX~q5k#i2}CQ<1*j+On?5xO?np
zE-}R_u|!qJBXtq^{POKI4=l+Y`lt`}o<qcziNR!!M(K0RxID(-xbNtIZ@XXg3V{<E
z1P5{-bHP=tF=3v(Ju<tdOT|`~<Z2f6Y#X!=N1gf!87Y0BRFxg-Dol*Anwk5_MTR80
zdVweo{@1e08i;_Sk9hKtwsOa1+1@?Qm{{J}dD;g#oE|^+>84-<?XSSqA#+8axJ}Oa
zo&n5X!1n7ov*bJ9QV_wX4ZXNcUPq7pAr<D2_TgwN6TO|OmG^h4yNO2Kv)?$N;zFgg
zH&$}j5M8o|C0701OZ0CW#}x%EZ`s?94e~q167+p7C4%~YF&@9ajc@GT)X>B<r>Xv&
z@X_NIJ275y1)TOrHM*k$bWISB#MtWse_cg5OIO5^x_f*y6#wuk<#t@NGu+u#Y|5eB
zg?_<Mw&f>YPQ-@BhQ8;qS@nBQsYtBa+Yq*m7k?v?TF{HA_unZR3;oLTJ4`6fE%OIK
zTC7N}@&H!i2v%f0S}xR26TwO^pbEtPDy;a~I#7VA&e~tZ)}aaZ7kMysSGDZ7q=DCZ
zvJ^jut*px1>Eb8U0`kf|C=A8{7bQ^}nwGxOWu8NsYmIptE?WL`XfjsC?EaJEBx;{8
zY!8X0XU|LTm^Iz;#~}<_x!<tZz=)s{Rx9MA>Rn*23G%{ZK@lfwHr903UZi{FW9<V(
zDVDyA8ltg`s4UGRC}jmFDW}TX6Vb;^QbwI)_BZZ1sEdD^<eMQV@-fgT+BAd?ISW-~
z*f6B@0!IrMiMJbaQT22lI*pSoOii2mrOU;h;l$!xl*WxFTyUkz87C+_u%oCW9W6im
zxgr1MvAOSlzqXvT3r+AF8XXKJL<K?q{#ER$EKTMUA9rZ{+-E}7y;|y7tS=T1Qv$^r
zJY180{nIB7iUumU5I-eV7`g-|42Jhk(aK=3HxaqC+8+#PEwbyut3>yww^S6_)nM6J
zIgz8_o*$%{HMx#iYb9De%12ba_OW1#-d>T!FYKtC*cJRXc#9ZWS(?V`1VfKd?RU>C
z%Ldu`A5#H`UyU~OVWs@eq1M`mF+(g%2NG=cUSfzm2R+wL;XqyNU20*G68W|GBm=Wu
zjTN+%1(eg}q5I8bi!!+tGN*hgGPXtdDS5SOW(xTH8~<brK?!FB16R!c?X<f*i+75F
z`9()uEgvFY%$kh~Yepj%Fz-~^a@yRsxi972!1QW%zb|FI8*79c_Gu<JMwacB4aRcX
zpCnW947q(~ZH1`IaQ`6|2zXb;vremW7~L@pay5k!-v;#*IwpA@_o9l`JBGQ&|1kmC
z#=6s+C~}j$6wyK#xsbg0x@oztv2Ap6Q<cD5@<%Mq{;ntXu&_GpS2Q;Lp-7qY5-cgg
zkS!iZG_gh%Se~6|O{uZ@OMHs@;$fm(#NH}Gl{Tr}kjl2=kY2=cV{!0ftD#{X1hP(}
zv#6$B!`5Dg{a|}^%Hf?q7v_wq{)K!*2crM+nSxN$E17k%Yo<6&`pLnnGG}T&vRASn
z_Ccb4;Vu|ib#voHF=paFlLK~BokD8lhiD#-;0kflq229sUjT-RYl|>3cMc2jLD~ZM
zD?vYE?Vw7<{BGFVVyBXiyJ@VrKL^MQqCKf{OH+Kjt_LyW%%zI#g`SsEo;QUv23btl
z2A3UKHE{*v$)$p1u@D6pe`sj0%i$JLHFyuGhLU+0FIO&eD9VGl_asm5cY~Jiy2hzP
zIOThcqsQ{K>kI5-BSQZ~InD1X7rI;HwE|DDvuBgCBk&)*bB=WAWA5*L#-iYaT!fV<
zKu*Qy!euFTdDrZ}D?GT<%V5o7AJGW_>hJovZ5=aWvr55kn&f)oQ}V&#!1CXG1VhY|
zW$GD6@3ld=fz|giSiOb)z9mjps_BMF)>x$zsq-NnS1|nf1Z#a219H53`YW|NwQ#LP
z@6T(~*I#_mnDlu;-2@h5!s><bCI&z7AfZ<hJ<Hj-`cB`=UC+Gf2>36P*5?|V7ipAO
zcyqWhKIuk6j#Lfkqczgg9bNj%aOvHCy8dKOF@V}Ix&HS4kni%1=bnA#xXYn&Zu___
zlW75tX=0$C)p_bSAmG<uCoZw-&DhuUH9qq~r((flVPtVjwDMZMwu+Xt_{QM<NKjcg
z9`~%`sDb!+_`>I1{T2>98E3eF`3wK&PUEMg`imDVvSQJ=9Z2)eUye%{Iyw~AfV4rT
zr2-Ythzh@)+;bop(sB4Vy^Ip?^9sYTcs44Dds)g|EgXv~GxV_R){B^APo#>yL=iAd
zc4CK9T-?dTD<$4U1ywn4a$CjOtywT=tSZB8V9R+(lVt41YA!J{HoRmnJT8>)W0uO6
zb0s&oQLZ%}Xe!;OdMTBp`zM)Iy3cRMK;EdrF*x%S(Y8mQm^>fLJgZU9pCTAD#H6D}
z!7BEYa;mYCa8?s{>ft9d05p%yp;yCON&BuhT_}N4TwkyUGG&L1J97mSBXm$<c*{|D
z4Hn>ev+!Z2T^rh8-_{?sU1sb+>eR7T3E!*`p>NRJ%`-P9*=UYno8#jc{Ny5(QL=%9
zqc@(Ka88xGPObh5ZQ%oHHS@y8HgGC(jN5Yxm;oCq?m0ohwi1JIpJ|!c&C{>t;_2k4
z%9ET>FeGAP7AJS3Qhks+D^GA7toUVff~~Z?*H-*?8htgf5aK{ly^z{}*k7(%=n%Yl
zX+=Vql*}f-)KgnMK*f}>GA$()>SF0>)`<*DbMpyOnzgsfB6_1rr{i5p;6S9xe#C+f
z-&6`+IH5$m4HqQKG07X%=CG9R=c^1gp(KVgTaJw)sY-Vf2geg|`gz9wIL@}tjV%Vu
zfl#@)09?fWh=5USlNN?l+53Bkf|elCx-ibD%`1F|R$jz%w+b>KtcA40>WOBOF5MaJ
zn5&g@!suA@wQ4B<jaF<C8~WkJb4tI;iAICr{F)$@zrjFaoyor?bmqDB=RX_R%{xs}
zYxRhUwa_3V7mB;4yH5bnXLvqltSwD$t<#9gsM4-LE1~MXh3Ih6vmVSL2ij+5JvmES
zB_9(64&+MWzn>pRuERFt-EO}MgsrL;%gpk;6<x(6KG1}}=30dWFNOz+`P>&m3x$H1
zU2c3d?>5H1Ka{WztPF&Rn;}BIpC~d%+J&^!)AAAJG`Oz=^rkIujz4Wbn#Vuh02Qv<
z>R(_f&aKlP6B%<2xbs^alS~jHPsgr5bb@+2Y8d5zOoZ5vECSsOnSq>dzqfZQvMd1O
zghZNuUSGSEJA#+@&>*kA1~t0HRdGgXBtC;-G~KfxlARQIw|`j9D#mI}@QLTLvA_at
z-n;)1stn1=e(rM-5OcOvli{w*R6PBWzV_Hz*L1iBRu%f490jiT+_)yTk(o~pC!oEo
zFKvKjKzPVr`n;Jl&D2e7BDNt}vi2L{;f=eN@B-!$8-Qgnr64+yJ|O2(5nU-vd|Gj}
z`3l|=EHlM|as0<vX*b~L-7n@KM%|cxOdyl-Fk(s7G9TnRWK2l6@P;KW`k%ZZUb5VT
zNhr_pE>Bev8Z&d;i(DMaYlfw%zDo@KrJ*5Qc9mdd=RTD~2-)o<#WnVd_|$n(!pHvN
zR}rHKP%bp!J49#j#wxyvne44ph7CWA#duY<kL#5-w(r!+?R3q&8aLON{?0%u)J)#k
z{@b=MEXZ-{a&((kSqX(63V?h!SGQd<4<Z|~(J{l_RXEzxt3g9o6n;$X+ozaKTBVEp
zp4PiQBv4jnyDUD>&D5;1FNU-!?bYa<Zb-Lo;H2&de;Gpfc_;cZp!<C}!B;P#T54ML
zDoN9mmMcmmTPifkz$R9%6{U!;LR<mRjg0iD&@<NMMmnQFbRD@bOR;?>Y^gd*@j-oi
zzV@$G9=tfr6s+MvPDgcog{<Wf(;^bp!JBgUZf>H}YHVV#LEqd-u3tcV<{#q(rv9iQ
zSi%6S0yo23kGE)Xyl@O81&;bepe?&gu1H$5ZJ=Bt-Lo5YqwmSv<P~hCV$^0DUo>J&
z_IOmY62GCCWrg1t)&z@)CJEn{rZsA=DQ0QfuSRX%7^Y8B-^4sX?6DMK-l9Y0=kUY^
zJ~uQE8pj>CYJzj%=&DU|Dwj;@(tLUtS+uOfsdvRH?+^T?i}1i5{qT3!b&Cy2%x~T~
z+=DK@wlZU16xEPK<DLwLvqZv$6mZH}+i8!W?GNng25oO5Xuv7mV*a^uu{t4)Q&qNn
zl%D*K%PLL*v{vgd@C8eDpBL3yp3<zF$#H(a9QkAqZy+8>-gL^%CRc-9?`~g9d*AQ*
z%QFFVr<pETv+&9Y<$f4!UO;Em9enbAKtkV1knpbl{Hl=MbC2q9$3)dT(08x<<h52s
zw8-L_Re*B0k)d7Ytni(wgC=DNa1S5Y)ZyQ>P6JQ45^zifR<nprpA{QBBH#Cje;kis
zgBK^j>`eV^k|;m5M5}ms3HI}RMYtV(OkCI+JIc<79rp`Z!ynG8-8|#Zi@~iyzVcD)
zY8bhX`n*5K{oAzsu-dy9Cgm^5fRrR6r%wPCtNp2GJ!5`Y1(Z-N`--`TEJj&B_S6yM
zD6cMFis__ZUsH;>ow2s<NI<?SMoNpMeiWIu0~@wZ(@>gakuQa<%61h)-pxroNacBQ
zX(lNu>Ua$;jZ`c8uMWdQqh<6(Ri@aVBXYyfNTzK-QR<#2?RVL}9BH%MTP=kz{J<Z8
z2XUcP!6Ls%;Q6KwNlAf>Vg49nt>j}{foj{^%12JKW)8YPsCZ~qrcIR1_y%VyB@14V
z+efam&buW{xqWCQgBE*CQ>w^QA|0@~X;#5S+9d|hV5aKnJ)M#hAqt5^p(au)ld@)!
zjVrf_%gT+^bPnarDep8Shi7OebH6r|H+D9R7c2-dqp#1!-wAGN4md9-QpFu;ii;)D
z)ii3}SUv#~H3h%T71WrR%C@a<I(h)v`NMV0kcM$V+Z?W|1HVX(GSNFSF-ZnhpBaDi
z09*hM*vw4NYRL{7<{EErC+t$KC^UiN-aw^H+cz@CmQuk3g(nQZ1(HPijv%tTk2WN$
zt;Rb)zcxjN`*pPJccTnVQqI+NezFDrATsR4B%J)YR56DBFj`dcaB24H>!;|iz+TAV
zmU^^`-VeHoYc=~-u?7-h4Oi&y4;blaDkZry-mey)DR^BF3r<gU#_L+aAi)bxGcZm}
z5{59$Ui+{?<f3e$lGKnZ-^1`Vw)=Sn-*yV)NO5UiGd--E<MpBlMuFq3Y&jdoaXIrr
zD6GOea{2a8M-It;desarXdF>TIcE8hQZ^}n5Kxehc*_opRRD?8YBOKTCx$R8AyT?0
zKvfx2TUFg(TM0P<3!tK~V~>JjNG=s=Q<kQ{VN-<z5d_!%eY=oEK$n;d=5K8Gs#{J8
zQEQ<+7U$0dRfls7C6@dMtTKnj1|bM16!FG2p!6!pgW~NCnkt3Bt^*bpD>D4gZOwUw
z#}lAT3N(hQR7J3BxFW$JLld@gKxxE3KKKat={5M-+%my+L_ZgsN23UcN>mh&bBc&I
zm~gJ0p&9Jx)=G|$8cNTc7{K-2t`8xU$L{8~HFRpE$^GU=Ppxa7Iz=7Zg+16k9~GX-
z!A$LQw5Bf~#G}a!)2Zer9v3>gz?B@<uZw6lbJ2fm80Fxks11kP2EGy)F2fkidbMqO
znmU<G;r5&4;$vwk$IfAnzD)x6ltf?tL?t&D5c{L>Z)>m?(M2D|jh={vik8Nb<sPB3
zQ>mN-35Z+;@xWxLqN=-!PaW~(%8@aQRlG7GR}oZB_iO~iW`9}M#Mvb>a<F`Bpz<-`
z$0^Rg8eQUo7e~5VgLIZIa%{+>IQ5kmkwHmp9DKpbW7z^oM_JEC?zKk!rawq!cqW$2
zBoQad;*YEIFq$X8U;Y=jRq~+#oDJ#x(V`fv+duVRAQ~sJB<D6n*2jAsOC2jQ!>u?l
z8ot=occwFtE#t$rv>wJY9q%yZ>5;&7tH*RxeD|ST_hsY<Dy&AALh~4wD0(gT3u4}U
zgu{ZQyHrwRsi~qQo&OB$7V2S#(>WE8E4+ES_1yPEIhW+NO%K1n28Re`d*<fKgY+A3
z&xQE29zh|=#0E1(9kxObJEo#%1W_o+;E6u0-V^HHL~Dk}C%~Hmu(G*5#J%g2vPaxr
zhoq7@;1N!-{<R0sC+uErOjD<PA&0MczE#JTvffL=K^tdwe2dl&fqJ6DEdVuNemz@l
zzhq$vAAR3z-*UC0BXNJsUn^Io2D)h<11Qt;40J^4J?bp76O9O^d+_QVE|#&DKU{JU
zY!t246rQ}${usQEZ-So3O%U{+$vq}bgn0b4TnQiD&J9hqL7^o*)!kE0<zTp_6%$W<
zO83`?IJpsvUDMyDV~wqfeb`<itq?=JCf4~)YRz)Dsoz7z5PTWJB&K%Bj#~5cGFj6q
z2}-?&kk@1P4YR2E9u`wA^ZF-zds=<}4^?Lw*7W;^eLxUJ0g-N$PNf?}q`O9UkFL=n
zC5#-+=#IgZW`rVLqZy5KhZ55G?Eib7H_zK`FLrFlaevSII?wZSIr(mLO$^o>*T>sL
z48qGV>qgY`Hjz8NR9uiBw?X_0^?Eq70}Pwc)=Y+)a8VSW4J&#GQ+g?=b@pshm{W-(
zeI-J31(CPV(YGkIM_Nnecw!GI(XeX!RM{<`j6?GW)c_w2RE^f9bZgTPTrB=8I*r=a
z&!^-<3A!5~M9}RXAX7{gJ$pri7<=zlQx2)EO9hrxc1@sP9>GI}g9se>4})|p!u4Gq
zqa<Ei0jm8SX|Jmb*Pt_Rh~za~|C@9I&EIbjBZG%b(X_MO=|#kyxV7Yk7G}yAj+2~O
zpX%k_r1LtMlmaij?DbBk*aT{a$m)c|)+wPxCW_>5xj;x{gpP&8ZZAec$f?BruL*z$
zjXwYKgsP5)#KcRH<32XD1J9q_LbF(R47SUco*2LD|5`<hO9>+GSL4mj!Pd6;&~>s~
zE9u1I5UVHbTV?qqSaD(B3^T)KvQ*BqP|38mdRUWRx%a^d?6hiE&C%zgzTs!ZZ0j~w
z$8~yQ|9yVyyHsOtUhI~>m|yY{rQQ7@eXmE<RE0z;I1YMN@;RPfP@@Q*yaM}(h4qY9
zO*hP#x@>6G3#sNSbSWNSRewMR;0u`;-sJLt?ls=c2i5gER7Sm;dH1RtlsJntR<_Nq
z51biLtk3LCV4iOsf)O>E6(jP-W+sJC^4y<Qe(xg^WzZs<sAig2V2IL9<WOVcPzPC)
z);>(XV61ql>XTi`Xf0TvDas8v{^S}-JInF~G=xDR_zRzCgc;-_I1X*CS!#&TVf8Gm
z(Q^C?&1X3$N@}uYa?LJrwoYVe&3ZyZ<i<FMRhFdJ&_10o3M99V5|<a7tMx8UVoXN}
zCFp(+7bfzj_M-^$*=W-K#3r%^_cEETq65{KII5(JXSZB^>+a^d-%b5IRT%BC?G78E
z>YGgxW}n3)-rlhd(-bTGX}2Gtj-wK_ma&J4bR7LM*q@xhnwgX%(kI8{z<d^df78!i
zRp|X|-7`lyKTs|L;RU%U@9`k*NsHTIu=!iPqZ+G|qW>oUcJ(RV37`LsI`B=+BKzII
zS(O*qqDl#tvCVrB5I980>PWiPmY5V2-!mjGBUlQQWA?OGTgm2j<SWSj_@E}Q$%<Q0
z=A73*N$oSnS@qkE&i}Hej=sls>Uwfd%%F=Brmt`nXZB8KB9N>%Oy;UxAGf9`Qxg(4
zE(oXJwDSSPpnh)@)I+9s;BBNX;l(ZerqYorDcbxdy)iAqrP7Q7ItRpiCTOE}M;%9j
zC#^E7eb!5Wrn>K#gZUZeMl(r&ab3m^Z2e6~9ya$yO#=JeG1qE&V!ZwW)m6&Zg>gw3
zq;)qCFcXk9Gao+Ws{}DN3OH$%Du%-*!P)D+=rJx{q<+1}fQ6vF2Kp#eI%0QrNlZiE
zcfsgsB|7Pu493XdKbtnVpNUo`B#C##z+?MJqCIUvN{Z}W>b7xRk=$o^TmRj-X<)@;
zMz2m5U=b-ml<Yv6P-WC{M5>FDL^c>JHML#A;!`IwVmy?LUX!3Z^-v6cH1&l9!w!sr
zS)qebjGhU1v6W**s3_u?cW@R?ORZHUb0d{GUZ9`RwSOKKZuP}NZf;S@%o&W592cx*
z4obhSb@}%zVLfDBrFX8iwvPThMdBE`162G^TJzCbODiEYRkf_xz%*_oaMby;v4`if
zg>GQsk2zN!5whwX0PfxQw2|+%B|7T)XAf4;`E@GLd9=Kcg$2E{9w1!0MF3Qdl}*4q
zZ-+Se9^GgS6B8|rJfQz`N_Kphjdedz5zq-^YT)PW&R4XI#;`*y=cxA&6R8U`X1oCL
z4Y5r99W~|Sp9xLm$O;#lVdzANDso@dnIVSCp>O(1b@PsRN*XeTmgAhl;I?aPxu~Yx
z*UU5ZZR=E4vi1>1^2ikm^?f7bT;I5|8K%69ja7lqwr<=rFPpqrQeAH|Nz%M2v^0R4
zyo#kyQ$9mR&`;*BD~%}nGsgcpq@|BO3uv>~KnrcBzK51#SB4n*sQmW}|Cjj19J!8O
z<{nDM%pe`r@MXd?G{+8rb|L}&89Bbp$oj$Sn>O7B9yyyK(N5qJ62RmsMzzl4V#s46
zENi4-E&4XDUoY`jf7>a(X>jpSNoMM+n~*dWt93+r%~G7php4spgEjtROhwMB=~?^V
zb!ZAcSBUl&M=NP<E!ybJZ|P-i^~T8by8~?!)&i2kn`Ch!<2u6pX;6)iZ=4~!MAtKN
zCSFF%<v#nLK5ynm<D%O0=ryS_@@NPC&{5B$GU|D4vkCC64!nAJAxbzU+U`z9!DEi1
zUJf2^bTI(&><nRlPZ`YMkaP)NM>-?LL!#G1b?!LwVK+y*Z4t`hqBfaFYn;a#X`W78
z%Z`|>LJR*eytN9=RA-DG3gdz@>iR1%UWe)M)U1U|^94GTsBUtFXOr8iv|bz<O2A@j
zH-nxuB@^K5x5cS9rCaB7`mN0H!5FY3rm~f{as)0~?`Y~iG|*V$y*yWGQ9n<B4Y;z(
zwm8b<-s*}|u`+^IDXb&8dx`pcJ|jLL-S+e~p9i&OORot+>md~#n_fkHRx(Ig{vFGG
zQ!S%dXA|7dpW3r<Xb<ViV$)-T@1Xx_0Kq6FlfAMR?b*)Tu_+o|`jRz=7*D}0Yr&i9
zoEa>hL6;Jc0h8s9DSyY5QFk0)GvY~dl@@a1FjuCne;5s>A`Yg_98>jmV?nB|Z7EYq
zE<NdJB!>Aq>25}w{MW2g=7!vE_f<<X-a~cc?CBJ(x&tFszTm{SuhzfN_|)n=`SjCN
z(RxCkBO<LPF?J?ib9DPpzS5d*(sqW!=N^t-(F3Q+NNt52ZS@_TvC3NZXhC~PilTir
zCu%L1U6JH=kLaMePj6LRw=zH;BQNVMSSHUlI~6!Ad5@JnW)4d)MDvVDA&Fb!0k1hF
zi=t`Y+lF!Z(H>^}kacwCo4}gn^f)KnHC?R`;NAZ%k;MfEJvlUSvwPQ^NcGn)-<LRe
z!yj0;uA-OI>+$w}_y;dMp08tyqkN#`S7`06o*TbQpl8A)l1xNHWp|FGsH(PStn56>
zW{rQW<HN$-$v7HKGogoWIbyfi#i%!~%tQ<1J02oU6q670x#tO$zo>^^*x}Z}G_=K@
z2+`ytDw5h$com-w`;uJBpbCpWAS&dAT2QH!dr~iQhNhW^9>tE1FQ0W|wzYCd2j{)I
zRu-#0Xta3(?VqxJSBY}7Y^qx7l?HLX>xCuIf&P|i?Zyt#8?6{yx%?axsZNfmSzBSJ
z?0Q;bLQF+kbNfY><%4Ozq!XZDSnsF6&cz$TPNq+uP8>gll?Nmb^t<(S9Y>z(5}~Ta
zRP&w-ey-rRKCZ|vaGi!u-Eq%Ur8DWaQ&o6tBI?`XxK0#d0gO8u{qV>A7qYcUCwSq<
zPvre-`rV=TbcU?7YL0sVFbeFk*uHV0AZ`TJ{=i7f$#?7p%_d;6Nj?X-q}9B;s42Y=
z1kYrj*fcfpm0FpvtQAo%^J<Oz70jC>8jWR`N~{PfXw?F0?<%q>R`q}A==U_NnsUZ9
z+r_g|e^K7E82+f_mM-sM=3*#}4q3uzljC)b%*_1@Bs>}rHrVxK=Nx4bPy7muqBA87
zr`3MgGH+>M#;@}9Sq_b<2U0=j9NN_?hxzaIN?pn_&CfV01oM7roO8ohy0@%Zx9<a|
z%_P($+QT8;sfVUYq1gk$KF4l-B)iT3FjQX0)AR9P1$*dv&x4j#5Ckt>j-jH#TjqlM
zKV)B@k?M~s?nzSYG_Y_{L%`jN3ucRX`d`$&{|<Cott|g!N+8DkF>zG|b?tu-N{i=K
zeqkf~=2O=p#E_@?D1Ndx5x2}o$F+BCPV^?5Odc<i<_iAxCwq>64&_bN+#@yVyZD18
z{Uo*c%<Xn=i(l8(JHApz$8@v1(`CViv!WLqp)TewH*ga|)tTl@ee!n*CA)D}8q|Fq
z`0X+DHl)TXt&agu)(d=Cd)na2G(K`@;sYqZ_{-SVdHyO&R4^64quiS1!%mujs<yse
zwfb<FPoEsuSQq%6XUwtb22+DxDrmKdK-gO2BlQr=E3$qWQhk+ZeXEf^ZO0F4W^#9v
z0`7SW`v#s~FNpOOti>V&3|1IXkC}A#UWsXdz{HeEI$WQ}_TQCqfnQ#No&cPs=Vm&i
zxD)E8ayedhXN2Ktsp1q~l=2>4Iet0S#=NUj$(=yEo+g@a9Ll@!Er7luL61fS5RqGG
zW$mXy7u)yh^#lXZCGzUcH$U*8naduGjzfbl{a_&De2LtGS}!lEd7rTnzx$#0&-XYp
zzjHHf=deySm2?hft43QX;eu^_DA5uEZTO;zLjPSpvCwcK^b4wJfCyT;kZO!_7NeC!
zWj=a#?)~5FT;=?(1by9qIKQlAP+^U~fF(mG6oS_?(2-!Yl(SJOTWQW(ED_oZnrc@P
zy?x+_-<V3}?;VeK6}doiL|cDGU-L~C7(IuVjwUXdbDEueRi}&luhO^@0|2Q#eP8(j
zH$sBkF4iDRf$sMe*q2p2+(x*I#-5OLsp5`onX(->s?7A-LWm*PpqV4K)-I*wA+dhr
z`_nU<)sogz&Dr!3ZO>G|#Ez=!k^kD9#`^CbWr#s;v}?s8JZEO7_IGYGjM9SIL*f=?
zs!JD%1hTMme#9;<UQ^(fLA%Zcc%fSgXxACtm#JrIHh$fu>F{b=Rgxn)8*Nw=XkSsx
ziI9?=ry{kDXfBziF_0~y0<JC<sSx$SM91fybu&AeJnDQ7_2dFsg)N;t#;X}o6^zt+
zqbaoQ&j<eKb9<F?_x7>!*3%V39m!7{U5t&4my9A41@U10I6A%?4Ui`B#uF_>0<Gd(
zgp0j~F2n{8LxM`hv$Rn(a6Xb>xCpjNQ81F?bAT6W)bdSsgL$r8ZwB(7`buQCS<P$C
zbnwW($eE*nn4~~xN%BB;xFk^L4GFxT*-f65>2*YdR92PU1iI@&xWd;XVR1(8Ufn{K
z_T7?TS5V+e0QNhUtZ*wScdjBxY$kqE1BlFPwR~{0uNwyGeAD9!z%<c2zO0uGlpcGY
zP;XqgNvb)8pIy2Hk~r_VO;r=DO+a?&$g6N{KV5r%<ap}gYYjt|8#cPa+&@whgQi5r
zo4m^1*kFxnL0qmIv95I5`IgaqdBVQd*U+u<Cch?j?jF~(ZO(!s7@z+9kzicV!Q_om
zd<r66*cijTB&oajPiQ7x$=kT4=;Xod$fRWN^3mCgi2|JlQ`Z}w<w~-YwARgQ5?A`d
zK0A}{Rfaz-IlP*aaqdT$OOXKQP-W}Hg10Vql;~_ZJs_J=skHTMTi0TKmor7OI=}RD
z2XIOr)QX^_>*Qc^(YRopJwZvTA{_%gIJrid5@pW5O2V3c*LG!Civ3m&>i(A5pFQ;>
za%^p=qYAf4j;Uny$!aHC3B=hv;k*$qed`%#M#;7y7?I=aN2DQ`rJ@)#BnBcKB1JM*
zyDfC}p#&wv`ly4|Jm#)U{cCf0E2tUGyP+&MfX982e;65Peva4?YvxFj)EeHULr)4+
z^dN9b^c{t5hj?{jg%yg%bI9TERh<706BV$*Al$RZO=pSrFyX{@VsTP%%lzenxuScw
z0(OX=Yp@-Dta~PQNnO9ppkOUWfJs{dw?PK^@0I0h`L$-Rh&~n1)SQNfFg`Q`rcvj>
zYpBADYV29z`m-U4)K&P(jq1cF<7ZF#0_;1zWn6;8jreBcwTYy8qc|gf9_zy8F)3>f
z4v_o&IWr^;%@i&)kVBKG-qFS6B#*Nm_tFcv@-IBGHYyLH1xqv$D7-C{)RoG{xiq{!
zH$-n_BB?pibG>gYpH<^BYjOUvwg3yexiw{DcdOfi44%uWT#T67SN7G?7qdqlyYlpM
zEZ9ltY>Rw#eH}l$;ejd<<J*Q{rn)fXQVn}B^RBvevSqH;d2sALhX*Zg4@s{+hF7J=
z>nBOa=F~<vJGSnEKgYJPls<V9fpmP$#8sF@5X`|X`XToGP;Vyv+e69dqaLR$KL)mH
zC?1Axed|Ar{C79Han>G-<oDI`Ovy;R$1^^}B`MynkBYT|$2yLq0ODJKq%(hq|Dg%-
zb@nu2Wa$-gN0iMNZx>9A{U3$~Y3h8i`T+}_!A^sOs>~$oKpE}Rak2RIFQ_(VgnrAj
z@9Czpg@+^t^Z5jhXcC<_bo>}smW6iJQYyOn2$Jok4DbCEr)3JsMY-ljDl2rd-7^N}
z#8JDDTpz0roIDD~Dv`O=yk>v#9wV63h*_-z4~>@+%e5YH%WNsLbyU_~+x9?eUmHk{
z!QW){DR@iCYBI*ng*fk`n!I@*V_GlZgOWtKahZ6EH484Kpf%l@6jiNHxEh&vjel(4
ze%y-pFc2+w24|ALURa<ka)HPGtUfaW>TAjRfCl*RhsHI2e>JPgwaDI?1bdwe9)EQc
z;uvQgE@`2J)dM+PMu7t?9nku8rR58D85x5r$&2zg7@?V+O88~nl@LW->TeYGq>YwZ
zrxQ!iJVDpZLOy|NMpl0btD2hzt{l@P&-~poM^a7)(KAuwDwnru1Id0gn?3>=?lM*~
zQcV-ZlVfGi#ml};(O@>k45$a_iAS~4iw)mAfdh|@)QVRY`Rr2^n1=gePG*xpdgm>*
z*4*Gphr};E@8Vd@mgnd<lyI8PX=vy?#dW|}8;MT$a-2@rCt1d$KeLk33O4=<e!N{^
z%oI9$^$){Wr1rH(UK_#v@Qz;pe#q_p0D^Sd&u+c(m`5Yc6*S0_?qX><eGS*`Z1dk@
z^{~<~s;1TZZf)3SNZ!sG*2g2V7m#Yo@`Xx{*Y!CgxW9RHHDzXOR&RX<Dtq=twAPfo
z7t-^X<mpB2QJ|cuk&oBM?m42PzO(saXE8m6@z+#Ac`*E8o=s27ppy|qaF3Gj-fIax
zp9s+-MWy=F1ZI5Dn5u8*49?29RXtP4$Sf$cJK!}n7%LeaCf?U4_qF@9G1ZAg(1{96
z4B&2;J6NLOPpDvgl||^gg`C9MVda8$;9{l@<<S?1Kf&jTmz`y-<E*H1SUXQacTzt5
z2kMdZ*;l`F)F8eAI;?5N&dcvhqm8n&R8h55FT{$NN+vXK*HmI4c<PA%)?^FMZ(mWi
zYXfYBPAZjEEkEQhhq)uW;33khs<>6P)^%Jcr%&+&;(C0KCWZDGlY-tv%c!O=NbS|u
z+8jTCNcUz1OUHhvdK#ZGk!K&A0M~L~>wBi1KG89DK>U=B1C6V_xuZ4?*EiUc6+gJ^
z0dDlinpV1|1$oDk=ly*FMtclwXp7?ia$B&#M(D8A|0E20E)qsl;HR%p9{f}5O^Gsi
zw6Hjf<dX2xe&6KuPeH8-i1|c~5~d}@`2mevcoDRy{COq89E#d5PgUk*Zz3y`#oi1x
zD#>da9i_&C+NI1TxYnDDu!DKuM!)zxj{8Hf%$Bxq<+9yuNrAXhq}S3n_(Lrl4Czm+
zJG-?VoFH&+o#Q?ENpWlz!}<ay<bsuK=aN*AW*uS5uiZJXd!<w7iX-d38c0g}o8xkR
z+4|?GWr*wn=V<Rw*kBufN6Mb!<Ol9IuiIaycGawWMCw7M3KQ`?)j8&b%h=7vUDV|R
zL*H(1N+C#plrm5`L6!@)l?!_1d`@zkHqh~`Itmw&>WH`2YRo`8lIUb26-{U<E+Pnh
zKtbLT-X5b<0izeS8NGERQ&1}k$LOkO8I0*w9CVwMtShAn0jU~Asrp&YOxDc~^YR{p
zZ{y-15iB1Dj0j|HtjoyhIo0=`RA0KPDR2)Lrw(A1H5kCZ2>S}%JZG({<|)eYksIrc
z`x81)iEcO4mx@#e6BxJ-kKI&L)zJB*ZdRYMvFA!P>b~!nI&ln=`ss_HvBsXgAn|2C
zAP)(FafNL?`FyQQ{^Dxpl2Krie->g8G?^Uiq2o^OcBO_*>L#ycf*z`bligsPP^Aie
zMmZ2v`BDwKGnoVf6RXEbk07L!xDnuC!d2nQ&dIk9C5vBK%&)^}@Gc6XEeVSFnc8Ot
zKIh}R{uNo+6Gk@UO8zl(+<@goln>VRGREZcvmF34e=odO$zI;^UXiOghNPe{IpjZ%
z>Yh;Edl+J!oh=VBd7P%tz!d#CG)w?^#|zXVr8%~+7b7Foa`8RbYp?fghS-AT<jNfa
z;3eL!s$6@>*48}@qhAX{jM+2czV<$3c}$~MoNjGF;q$bI@A^{ACHxR|V98!0mS0&L
z^|P+e`_&Ay9LZyN<YrLKo)o9JqnaMq0KNXA%KgBjofd<#J88%{VYWWgpPT8KDGu%h
z5@(;0PAQfawZw$C!xBGp-|vH`3i__}&)8F@W|{O~<9=dDk@T-}p4HJu*lqTD0YlqU
zP^(<c8<XYFr7mZ$h2)if;X@iKDg@*6wwy*r)Jnd$6omNiE6e!H(B#g+)b3hDsT%64
zTcddkzRT$#+VBVkxApTDmgJv$jMjGEf=IKgfghQ9`+o_0?SCRI@#Od(kg=C2|9FUT
zzdU0`H$`teQ@?gYbhgt}LFdRMbWEKOb}3QJ5TMq3`czzx8p5JQ^+CpRu`}{v8cNlW
z{wM55KI&k&dwI;h%l)Y$X}9UJ6H2TvRmIK<VGzkHKk`@?a`NmHdTzF*kZE1{`zRp2
zGjpMdR(!~y)RB>(<8=UdHF+?B?$B?R92?^Yk0N~D9KujV(F!B|#+b%eh(E^w=xFF@
zW(pG~igw)sq*k!WAMM8b@ay?<;KR-?Hgf(FGeza5ys)Q@a@am8jdL{Hp8qkqVXOJR
z=V`;qQiRFW%wqNzP#CNssKnJNp*=&i37e=p=2bzKPe>q2@OU^eC$Owz#}Dx+_~^5A
z2q<uAGVrx<6P1qg=p^PylPq%>Kz!`cX!<6v{oPI@0K9*W&~(-gz(cC&96D$vuo+U`
zI+q@z4hS(5K<9$NWH=#vH$k(HHzc;Hq$bD+R7+(NFE3zsX8Li`fZQu-Hpazj?Y8Bj
zI`6Qj>dz0IB4;OAdn+pd<QJ0{7-BW0Q{5gS*kP^a>ZPt{_A`MA>k~ofnx5I#d?WtW
zN38+b+(}mT)dLbz<Ga!ETnOf^j=*{rnu=$tcKiBaXy%q`+Vo?PZ7Hi%m=-Ai^8~ey
z-v*bBt7N6hQZz;^!Lz|iUn<A+(_Kh0=6o(y5w(}*f`ac_xn=os6s7n8Y3rNpA!a9V
zNn`dTesmt+hCCO>z9yHK;~d_o*e$?SDG>>D?F;st3@@$*PA@A}C^PLRI*A_uDyT+f
zF^w}K<AQF%3n>P53E4?Tsao*^WNNm5!OhV$+DW+qRIA_G<-mDz-k+4Hc!OD%gF2B$
z%~Qp((Sv8fPPm*7z!EPtXMKlXpS5e=zk4x0RoEE&$>S13Tbz#&f_ZgOdZH~$iqdLU
zrFl$OKDEsp|7PPACa!IisN8G5Gz<x^9Oa$YbG-j>6d6NLuN)LVyenO~niIuTWnQaM
z;iPTcMQX+U*;HsEdZ_dd&XRYNAmQJK#oUM7DfjP3=`6e*aseU7Er)6@_Bt7SFEMo!
zodlIRLE35ggRRXww+jqo5?z>5>CAn{ef>G^)|xntVQ(zs%K|Q~a=H9xRw_dJ^28{=
zkI-jR4GBbeEQ@Gq2j>M1F&p%TJ<QVL$iL3D)heTH%gfB<%z*$t7Udby)-It5uS5B7
z?ODTkhYn41TQVxg+*Sfo7ZoSkkYOq`++v0bi%eGHD9@e>lHynJnI%Y(1F@q|GQW#7
zYXES@Q)CZmDJV35uPugE0d~Y6?3#J2mz>{^NKlpoTaeZn%qB#ac@_R%&S|W0*Wiwy
zb@A#RulBcNSHB0*R_P>clX<FFyyG25H;jq6%Z_B=lm<fj+pyb;@NJ01oM+f?zdfgN
z-BZqN`L;TAIk7%cKV$94Oo<WN;JcootBCT9O_VZgimYm3G}s_*<RHXL)L2Y<IM-^4
zO#6p1GADg9t=qG7gHK2D6OU&MSEsr0$lKb5<9(Y%pmsVblL@c-JFAn5!WR$qXkw`u
zC;uanTKx0Y&4)#>H^SN}Q$lX#$kE8*D-bhrFHPG2+1i{+@#2813geb=leKo(X6&4G
zsORUaDE(G%8%DpH73xc>ul9i?+FC9z_@z9KP5#z_-qGp>JCQ4OJFEco80Ht$W<W*n
zK^u)7!YMTb0Vpq(%#Rx@mx4Z@>{=`eQ^DK23F*q|qSfVBGTe|*rRj=S8E508ezkEg
zL@j?I+?8rdPr32pQ)173=5EKk3rR0nu@*2%tp*;Ya2NuqdkTvS2aVV$@O&^x=+Yu3
zf_n^82G!<`RHvSTA0{>SWu6Qu8(HD0Zlq#GSAUcIZgSsqvQf9Grz#{0Q=8{97LI>U
z`Q@9yPr@w9=;vd=uO0?-Q@aV=iGMMo2Uwc5P;4#2OyKhamQwG&J?+*`t~>o5sqhTG
zt>Ba3L!lZzA3h@V|7cY26G)o!cFiJDk<`1)kyy~%vz}*;0aSkRRdxGyE*mEMe{2&r
zjEet~`Fj6(Bp6$F-EES>4+b?`0*+dmnp&!OVqg9j6bZaeLcztKpD0Jy$j@2?ybIq_
zJbIHN8>K(A@hx=y%?~O>s}lkgTd(iY&%b|BjYzY~%|jaSTrAzdxpGF&RLlKrpQKyQ
z5bYbI`rmNz7;X_ojQGN<spP&N?N?B;-A46#+=RjSs>U1|Q&#4#YSgE%f}Pq*tvT!^
z4OOyQtVP;Zr|u(#W%iAfnB*oYz9q=>agJ0`3=5Ia6rZ#a!#@1MP9Hv!cqImg`E3}!
zh~F&y%BN;<@rKgNspTg3duxteq8!aGbY}9$7CGzek90NZm|Xb*ZXr6^edrKEt-b}7
zE6fr|I#`bWdU6bJGKvHy=r(`D9zVr|9yS(?wM)%TG{1=<6sU2)jLuRsVhV1@!dEpQ
zDAiT*jM;+lUR4#ZlUOB2hdSdAZ-U(xqm+t>m+xk{eG)HTMxB!1b;FR}FN1{512Rki
z<!ZeZqLMiB^;n21G5lva4<w=O2Gh;JU$Ri@dKz8wmqL8EV7Yv+8tKz=`W2Vr67YHv
zI>azWIiDg2RA&3+^=8~;7V+IF7sPu}7_D3HMKkvOQli48=WFFgHZ~tU-@?F^e>G^A
zK#s+Uxb$;(ZDYx`qM%THu`f>vPYL6)58UrwK{vC_rZtp5gs;Z)+1HQPZfGEDhB>T=
zf))nn%%0y$nYakd`60!>Z<60aUf{x#MS?X+$rsxB590u7QZhnKvwtuK_@&XLCb4Xm
zVu>wZS&D^Rg_Wh3?Ua!S9WwDM<!(hx<<sHK2~@G#&)YB51og+p@4?c!KJ4xrKhi6a
zQ9w+NWg&^P?WT!Tna$;mqe%BhmGtF68h}}6+vcl5l^PF7ir09+5bVUBmM;2fZl?UL
zNJ)D>o3(fr*B1(!d`|LSp);UZ$DGO=@iBP?AaY900i;uP3bb0DS0loJpNMcG^%e)$
zHnHiPp7-@F<V~iev<Y%`rD9VZchDVWdcNGf3PpVjCGxyBT%2FF+n=ECZhN8F%X?E#
znDsDeT34C=g!KiIXxZdabf<D_SVee+zScjKCUIw0#AIw%@C}x~2c+XMeZ+J*HQyf|
zJ*Eib%-t#S8Kc-_X~%zI(lpt)cYNQ`;5jOeN~@M9I1}5mw2e_`&wDm&4V_mlT#vFH
z2idm9fR;IDc1<$nU*L0i2CY16(vu}U3vLz{wN<MdK5fB0G$Eh=${nO&w^%j@g}rU`
zYRpN42Sn_fyk0)vY@dK`sHrH;X8cvtdv|e_>BOhf@rLy?ct5+e?VOTnFo-CzyHl04
zjI5oOsJvRIZBOLmj>h})**#9cqCm=Qnk&;w$f+^FH>qeyXQfU_$Ju!|UWntMC(dGB
zNrLzS9*o*D*YqoSLmG#**<?TB^8B-rd@9fO2NM8D#aeMF+xJ$<VLMS0$<aJjXdf}v
z>Y_gUg4<5fB6+axw?TZ+%3;o{l%~Wc&eE9K9uv8w(|70abFG45-Zvtvy@bOE=HCrL
z;>|2C$1`M;`YX*%NI&-csj-Afea#<A>kwHD`9baTBGt6E_lFR_xn5Y(NmAmdJyN%f
ze^X*sK9!}hI+3KmF{}LfpagWowLj+PcZ=KHFSJI*9^p;{1-)aHM@Cyqg~#C75)r}b
z`MwjO`3)ibhfQ_2OSdHDAt=!gp@BJ4x(vVMVD{SHzf1+OIEH4{^hR#(cT75=)t1LP
zi|1}c|1d1Hv#R31-xc1?GT&{~+YKIfkV`-oO#K~OPe`Zb1n(o*G2V9E{-73)08vXm
z7CDzTa^To21PdM?_!naCad4Ln_%?{$IiwCzkm?x4<8wF@co;(ILV9-SjrJK-4c-ih
z)Y=TzxyjzFT#w;Pn|^#dt~`F6{#PBix=>a?`J%ntsiJ<ej$&C}5?HO=+b#UA-guF9
zD7*(zohf!Nq?O@|h2Yv6OE2ap1XHaKCd@fpxT+bn%uSBXy5+3qM5tCaA{F1MX1y8M
z)dv6uB=JSs6ADED<I3qN;+4%({PPpD*@Tu=6(8do3olDEHDWV?g+o;hZ}&Ev2C;cY
zS|~&Wos`4*u~JGPrYh@vPTJ=`$*<&eQ=Z*)gR{v=V*1+bPr%t5dd;cZ^xi1$1Lu@F
zER)*&bjkj+`WoDk&F>r;snvsnfLp6S%8|6oYkvigb<<NMfbKgx!RisRKfo#s5~FHV
z=|eQ{tIAVEj=MXuD4w{V>P4glA4(dVL#Ebw^oruV{IU33vXImC?7Jwqf|o4_6C8wr
zOYPJo!&4!tq&8&uS`KvBB#c!Z%psp`3u67<Royw8oyp3DfJnH|*n`ge)~qr_@#qLi
z)t~qOe6FN6*g*4hOpJ<$c`*y_D4RZipxuT*0$Y?Bo%7{9gj=2}#jZdMb(7izxz(3?
zUL;_9%vPU#W&U-~c=beMU~!OIfhNrVfs@L(1PVXeGB=pz;ugsEVMp=*{j!P|{?TKp
zNzJF~fY=xrY#)q|L}aEvkN%)R`LvB?CANa%mkYl9oWwY<W&83uhO&R?LDyDFDBkX6
zu&;^9N*eDHakIq01Zce6iDXG%K7Wh@V3SXQf;YX0gZSKSgI^zjTbYu|3>xQ{5wk4j
z-v;XjoK1|^PVBHl1eMy~+TwtWCf_m5HYu&8EF6<s4JuGD+@9-8NTa#@l_Rtl1?_R7
zm>-6Y*gONUEjmw`oc1Bg=c5fmjM{o3X1is8l*MDz7C-)=lo>9r;8^+9rMM|u^IQ>5
z!BSy!8`Mv2JfpCf;iq)E-qgJ;*CL{Y%>^=uB}`Lw?@-0Ezd|XWrU%HLE6ZC#i$0wj
zy~wE@>41-+c;}k{m|(h?ZS?f3ZoOzFnmOJo$3vG;dI}tv@|>FRJQUyXE5D2LdIsro
z;UAsF1v$){5A@0#2i~IOksO;4=&lbMGO`Uy>~wPCrsU12P^d7jnBUyImNP!npK52H
zRA<kf%0Zmz{U^p~ohGRk@K%6M@2;;&J|DRyhJ#bzpkewXjW3PlX>`LjLcN{X@aH0L
zwe6A*%wNi2)ewY7@NSlIV+ba4a%z6JWf2e=&{4(XK(dL)O3Xk$QHxe%4&gP{D2}b>
zU(Cd&$LHSRVKAq?(Th1FP`av;Zssw(uVVfmmSwH1^<?rrnqr~t+WF_BsoAHyj^ou{
zy+8`VI*FsH#+k_&Lxqgh&IAYbI3L&H(%1U&`uNJOxg80cJT~qvbprZ42}8KGfUk1E
zL`^g*=l@q|X=rp#E86ATy4z?|6MrzMOr9)}laZ0b(46@(TTteO*h05FW75?$C}C8R
zPr8+-9NEBpqrI30nhr+kE62T>OBoLIXx6x!HQNxoIP_LN)yvKM(1Tdv?RmJkzK)YE
z=O-pStj$jE>7<j)@!NZX7N){1^~nBL`Qmy`Dk7T&Uc92xp!O;=tp-t!E4fw-?cSGI
zz<V|7_a^qfUW=`bcY{Hw3#LYMp*8tm-gFvGEXA+IGe1#iP3UKu;jspDBx<V+?g|#a
z^uK@w$P$$zB(Fj_`tvt?gIIZvXTU|EAG`tzp0;&bQ39x=US)R7mrc(PRSDg%^#{V=
zpv{P>P_d(iK5y<A9EnLrj`#))Nu_ce(RY?(WFLMhTbp8uRCk8=nmgJ)a^O!sUdmcy
zw=U0*0~4L{zQ#|YoeR0Y6Vl-FxUpQ@R~yAULylfzciLnr0-pEr;#31|e<XOf{KE=r
zwc{+N>Jof6)}#(Y8keLgx+VpDUprg)Sx?Ni1z@wJzU!+@_=rC-oJi;+2%;&)BV_p&
zKubP>ip$}BJho#pZkz2)1P0bI$2ru%uFf{?STXQYWk)Y5g^`T*;<->LTUQg)EufUd
z8ZFR)<5~C@*0a1aui1+hLcDaAo9*;3Nd_4JgSPv9dP&=9Q+AVLYLY?60tD(*HKydR
zWIap7MbP-x%wyV6n=wPW#oZvaX5n4>srDqSA+F47(-Cs*;nscA#_voX<8}dWF-V@S
z*+aMC_p+WX+n~N}{~UjH&;t-A&JTUR<GQ1F0q`Y5pL$wFDoH{@)=?ILdiouSlIneM
zmPiT9h^)s0^hEzl3+JVPW9xjh0U95~4KK+Qee-$$7EqoSOhZ=Jd&b0+LLG{I-sj!A
zyj>ahckb_3kEm>lcSY6HtOdN(>Ap7GZFoOd1Z-_iMqU0GDfqleB_?Xp^`+dY%dPEN
z_5@2lbpT9Ev9F|;EFK}-KaVL^LleSp>}+*@h*}@(XtV#!YlP>40lmGC;qB(QCT{Ji
zlAl6n^mG0u#s<ASIOf3u))S3A6M+}O@8^4Vn^enZ{MDgaYAkv~>GdI}=he`};^wM4
zj}CM`K~2!sOKdzJ$XPUEYLaO+>ZxY4qV`i1#(ZL7zNqGFE3u#ZNgVJu!gVpBln;v!
z7LNaE^-D=u0n^GIty<~Z7y97};~vL`_*KOwy-8t9T?3n@6R^8R^|xmC1)drtj=vUv
zbCmVIPnH?24RgwFYYz8qn6m0rdP$V-AA93VoNHIC@P*vBPC+Z0_K9-)^x_G1TwIGb
z8DnyoeJ<)BMzmXoDa$YK&!sdXv|;<E*Uy`@2HM2GISw@~AJ{yv%X;s!%L&d=cd23(
z5!q-fJ*Cy)e+69*KXj6}w%Z*YUsrPTrAd)+X=Pgz@)ImrXnzlGuP1ZrgfJvIx|9+p
z$^nE!)Syv|HUt`@{NMzFtnM8fqZ*L!nUcV|aPM-9S0d-QYa){!Q_4WCJHT;l@KCG4
z#=Y4UUElCS?*+NnW0cQ`Qh)B1G~5vS;x(v>Ob>iyVvIXR>t}gtctx&kSmOGW6VGW?
z>dWG)FnPD^BTIwy6`(D<?9klwvpGUxWMk!kFZU}1RoyTP5S^I1qzpJOcZknPeau3h
zs(P7yy0W(c@$~XLS4=u4`qh@yjdKrjM+g0UoK+l&483w9M^rODx#GU4^zJzy<jB|v
zBwiu#(qScf6v$uyt5~*8y{}^=e9IBcmQ^Tp@pz=id^a1nyx{CxK<ExQ5?1G&bWY(V
zUm@L{d1Mr`{GHu)@HJ)G8oVPgA+?MkrKOM@J6Uo|5vdM1k*>GjyatL<(Yf4PADRZe
zl4t`SeN-3M@pUoCDTNO;Qai5uKk4l$P%bcPhT7uQrK)wCEWNIl&GeEQQMnTl7AE>O
zAhSF}mET_Z_Nb}B9aj}hD>`(fYmeQ+zO=gBA!Xak$6vP<pv9f_{l+G$OOlU6GC{Gu
z)TmZ%-;;Rlrc~#zr@qvDpOSr%`QndECbz~=A{uCU8S}T9!u(x1+Y05F=h}p=wM4rN
zU8++F?5uU#Fx(lx6|powaEtX~3Y%R*+xvl9;xg_dw%O=80Y)y}r#eu7K>vq+4qaxK
z_}sPpUBL~fcY;B)vXHseD#yqhS@)kOk{l)8H3KESazliqhDa6LUJGDQ?vfJvP`eV?
zvtRkiw5T%hn#}E;2)#1$+qbs-Nh3DOOcCsWxvzRc=}9bUCDhA%T-?kpjgY^uG`)_1
zk_uxgiVNTTcA-n$6-ZyyZV9`>Le_YYRq_Crv*aHmK=|US!=%;9&KWX4d*5tkzqC;T
zoUq)Ak!V$Wjw6v|RLAs(yCa4ByHKL&sR5gpII%D@oy~CFfd*}(E2{R5r#~*m&#Kz5
zWV_oOTAl1i-Ot-sGDe(fl`6~f@}AiZRi}}ZcuqU!mN&d19B1tGnkwWD7lH*ZuGcgQ
zM9duDi}VS+GPnq`N_y4Xu-AHLWt8`eQ`}lQ!9&JbHLrB+g;t<(+#M|oXV*l(!HD+&
zhsmY1Iu4Ly(E%#gU3=uqZ({i&DY=lYspIc6b2)2K9|so~t#p*qPO^WP&nMi9%?FNP
z&)elCvqs-P%iIg`160k*)^G*a#^PFPQci?P?YgNtw~z|LCePB#wVN-n1(+&7yY+q6
zK2vKZl90JUzTpRjD8>Fqk`fq5oEB^Ng(z!!6G^+HWmmGVRQmMN_|rWa7l3t}&M9~|
zcPc%xax$$XFck0)gHJ}{eP0vSpy`NEpw2%G53Uq?UQNXass7?W0aTvqQdjF&1T9TT
zHWM|QyH8$nM*NY__g*?}9AG`w{6hF}1gZvTZ6l}<2$A^<B+KzTfju7b<UM@zmz-cK
zybs7euYisamy3}8udgrUy~%dsd99dl_0p8rwS;XWDd2%7lpV^mFayoyeP?-YMzlq)
zmpfOcvZlb*A|;1L{25>CVznUwgZLV@|1jvWlCwL0c<WTFhNq!}(TY}J>M{nBe;Cz*
z=sZs;mFko@-)MvrULEBy_~np-vNA@kzjQj0>t!ON<NS%piY`tausRF3z{oyjQjw2<
z9~JhJOD+EIMmb)-{PKEE=gVhIEwNVO{+2U$s6ctV49jt#1EuS5@P1tzxaAvk*4bvr
zy5*%5unzm{6CAVbQypy%Eng1qa=;(_jO)kg*P%baJ8b}6(>2}MtTd};1~j)2m$ph?
z1Bl@swNwnHq9m_|e0a%#_AJrw{D&YyQ{E>Z{+ROQO39#2%a8vvEyMUP8Sk3mzKB}J
zZ?PmZmB9Wah2OmYFg`V-1<emsjq%48{gAg&aQs%MNIDD30i%k#(MQ$?AHfs%7Is(u
zEQX~W`pu0R<6)MlS9$P=q>r5NyiTZDg-fj(Z&?drPHuqqhoTDe%=mMo?n7%R4Dc;g
zv2KfP&mpkH-nk)+wjhlEN}<-;P8HfxWi@pmTBJh~*O`PIS2n)lb|0M|x)2<SOFFqp
z_&xQoo#<tj>iqPpBYJJXYxAyu7#!#OK??GW>!Rf%&Ve)JoQM~`MXs!Ma&5Xk6QfTj
z6`$EwstMq>+@n7aVWI!em>juwDsxuQ7ET?J50!zQ<A>^Z1TF+EWq;s4t8~V>=WM}O
zQ&MUWZO5PX8PZC($#h3|kgn%c2r9qQ$tHz3Wz;w$8BU^tl@tIE!1jZKAN`!1LZ)8~
zors1+HP`Z*3SiffvOe3aqXUJcKyrEChxLl`#bvIyC)XrQwpP-p#^RZg_lq8_7J&d_
zr>rBbE7d%}n508(<8<S@f()(0(_<6Xm=6BReQiPxjiwTzi9tg1GSKRoQ{A>!6;R9b
zpj-OODoc9<`P&sO#j!bO`eDS2&yqE#FZZ-2)*`e^^9M|8yuagpZ;3XJf_^CLT6_-w
zt(^4A#-eBOTV>~Fu{53#_c!OchsEkFtW?qUc81n{la^SXm!KN3zjab~MOD>L#eAE!
z0ICV7G|MPx5b2<lMD(z!8~@R~+iMX@%+Pw{c76af{D(2a89<}M&-tdeVp&&`Nh-S(
zsfA83*oqf=_53k%bMDn}`+V-ne`w-4rQj?WJ60v_aAqSyxzowHo(Dg6)$WQ)d__X9
zd=xRYOEbCWJ5f!T=KEEmqYjaOU`a!;ZLr0*ukO+q&oI?F%562gU19(rKPR!xr1^ZP
zi}o}2rk>@^`g4&r_w+SB${(m?Y+rM)a@ur>GBTi-5KTuvONh<i_QhZN0KZ{LIr5lc
za1r`-EK|7r1isdV>IF@-anz6OAtB&?M|OkV^dqlh$CtH&J<$%D$C_eIfc!<44=%Hr
zk|eoM<-AG9#?kSH8_z*n4|)+d9ToB7`G)sniY7BkMm!IRi$)FNBT<M!%35^)wcqMT
zdfJJ`yIb%XVCVeWu#<K!vzB+9w$FzwajHl?MtnwO6(D9QWu#AC34Gy5HoINdbpf~A
zSucGm@0<e6|I1(AxoST%j6l4|h4yxQ=UQT<j=d;UotW;89_>0jYu(agJ3W;U_svRf
zRA^%PTBPhsu-6j2Y0kHN<FIP|GMJNY>MXERosvny%}koA62^&Frq5sf$KqF6MFV)-
z`4weSYb`TkL}i`#lJx2EH!&K(z?XOUHKXyHxD?oz@ERHf5SAxkJEFzk*3vrox6d?u
zcSIW}V5$N%JEj_Blo6$ZptI8R8xMF(r|U-(fO0)ciC0%t$Pk0Xc1(eClZ?Q=&q@oF
zhe2_^dp^?&td)q?_5n+S+pOn}`pJ1K$OW8AWUo0I9&M_T5Nm9#Knsw)ufC=Hk;w9i
z|Jh#?MvVHPdP3Ply(Sz^wx5TFJsXlDBP-~YW-VBYCl-O=m5go0LvzFMP=rD(_>qw_
zyM_CM?M>CJ=roqsR7Laj$*H+Z2f^;VJ`NzIT`)GOYDvvDl)K}wS=agf#RiO5aARm?
z`Pi6z&HAsV^g_8bBWVXnbQ^&Wxhca%m1mv^Pu1I1ekQQ6)N;%^w<(YO%VqkviX5Ga
zl6DAwR(==HyI5K;y*G4G!Dm7K!(C61(!z<%wORZt=4-OU-(Nr0g|57u%h7xMlPB=_
z8vgA`*DvMm8hvu7BVYipidSQEDiPk-n&x8;k4L0UcGWJB?;l2jZ-7u}0~%6mJbgVG
z<@Om0XmNkh-=KGs8Se3mvSQqy<dsAMWiqEl0-_Jp2HBCgR918AjE_QDAkNF=R!72!
zT|KV#Tt40oPmi@`9nWLnDf}VYR~LLc;$5B>&x~^3GI{EDw}nbEWfB?@AN-7Wf|-S-
zFa?8h6D}wV3D&-we5=Idt>?2`*g8b7+;PS0uhmZT{Hg9O3OM&Jy`DPmvObnMi#E}{
zf)=Q8OiI+p_>!rFBNzz(YRM*C=ZvJ<WD1uD+9%8B#L|FHWr3q6iHD;Ga4XNJ{xIq;
zX`@{q3zJIq+q5RVFX#=e--p9JAO=sDKW_2x&VhHTe>I9#hQ4(E0*3Dj<_O1hm6xHP
zx$>c9c31BTGYSq09$Q<^s@o(xYi?aC3fw2fM*B$63msMGnvbRs>*ce@53a6_$1DZD
zEECj>oCf5R+`X*3Z^F;f6(i-nk*&_0VAp^NKv~6QF`w(PnL~KuGxXISHKiaSs$B%(
zg&c9B0I9Bb`7XA0juwrHy}S91O9*VxUhe%b-l{Uv;e~ObrFdlsuv}R28O{p}EDqaN
zu#$SWh{P)=gP#HlJ)Fg7!-ntu2XwT)v5MZa8>h`WRgD4rWLr8tuj2CH2A6C4IWd5V
z#f)9w^=uqeWQd8^I4aeo;SQ9IUnj!NM)_6L;q$#c{e}ji(SFNs6GOy%F)Ti>j8i@J
zSlEF(B{#QWu~yzdSV(?TyMFo?%F6De{^n`73P)@}s<O3+ON{#Tsrnc7Hh+>KH@2kF
z`J-3AgQc33x%Y$x+Hzbf9NTKs7Z(;(90jYt06=4ZCL;2qbJoZ4gWDdWu12sO6nlbq
z2A*8N{(1(?4=k2K-E_W99Dl3(^mI8{6kaY;yY*TQFK;sb%lIms(B5Cl8kG)6*$&_y
zB+;4iQfQc17|bV|U#M(Cib?9IyAQv>L=Pj?#fo*>-~JXzfycX;q_}LcN#((+w4IX7
z?7Rf+B$I{=<~&XifLae~Mlo8IN%|FVHWeaOz~*4~b9KU1$pOdCg$z#Bl;I!O0_EJf
z5{>^fnf(9LWbH4|$QLK+%R@!yU*>J>vrpeBv!Pq#(cJh`VKh&z1r5oa8^JQ3KR;B}
zj#-nuzTzTC`%;Es4J4=DTI?jRs{T!OH&6kzt_yhjuy>O<4Y08!GJC=<0j3y99k7__
zr;?2CiP+Z*=>sz93NwX}z=go)#xW9Iz4e>hQcdYBJKE+LAX3F>*%u^fo8?ZP$lkbd
zqeT|ydU^I|8<h~1jHnH_Yy&HVaZJ3Y&<C1}nluvQI+x`yo!pw?nU#GQ8u)dfpQI%?
z(8?hSlCZROzYq8dTMY%-m6$08oDpOA_8L#R=NZ8~*oMVfm;z|6`M<FI-s3eqQysl@
zhr>p|LCY|WMDwX@c_{iU-i1zyKZB?#mN!yK68=1EG*b1fSk-6JH3~e~F&;XxSiryd
zoDKFFC}VH`H2ho)n)fQ@L{|xK3wxfLR!#%^oOH3WG^*cYjPnPfzm|fY94QfMPRhv|
zgOplvz^K4CO@6hhW)&haWS*}Wb!C~Py#j177f>Zr+_X^eFp`n$<D?oE!?v_;C000#
zKEPGZ*w|~xaJHGtiAic_Yk-IAtI@2sgQ%b9J9){APEyYoY@3-1o>J+D?2imzCl>?&
zEGAnzl~l(=aO6x`MbfmfD1pSPgs#_U#QIxZ>B~j*g7@S-v`mXRka3@n`tvYtrnkYi
zs!|?g)hR*MVPmbiTapbK-aX(jKc;+x{ruU0x>J7fOqMv&_a7e9piru>JWc4#9nt+E
zC$(wjy=W-cVh+~8gCTrVZo-->0yr1pW#O+J7&xPwUMvQ@RKu5lkYUKh8?I~h3i+5@
z{`^*-i*yUJ1PDqid-qn0EeDXu_lv_e=T>)!bEq<j%RaXNxTk)}x;T(7;yF>`a_WiR
z%=}6@JASB%Jl-3Q3m^-a?L|I2&$6+6s_s+_jSNP4Y?uaCLrhfz-{Cmv@C{qui4iB;
zh&+hRxrJ7lJO0g>Seoy1*ZMf9Vv!Vzo|3X01!{faQ{R;<jh$Bba*WI8JXA@h3h;bO
zU6=Y@&f_~9>!{2TP>ldMygcNtM7h)GGuZo{1M^<cmw0)nXhY{z$x|6g9ia26UVO~{
z*HVNwziX45Yx*bB&buk=ol+zbtoT$IM{~UnatW{IOlm5$C{&{-EHF3Pg#Dpy{L6b8
zhr@cT6%<LIs~`LuzP{CwO%wiF0&l<klo(oHw<o+S0#~<XIxtB4q>_~Tw>D*B{a|!y
zL+z09HIfvDHE`W}XaHfGab_^s;vKCHF&j6YxKl*Yzqon6-2ti2lx9OKcl&*PwHjo_
z$LfJ-hRvihkbGH86?LX%)7gFJ%E$QG*#42{c^V=dICApaoVoS&e8skE67Ds=o)o^-
z({ei|PoLMBVs=UBf6|_0H0VWsk6~kg^efiH4QM%#i`u)p2g5SjwBkDnKgRo8Ehy<?
zcy+;(y2bIgy6A80r~i;{a`#C`W$dJmptdgXQ^ChIU%9uNWS(Pup13R)+Obv1o(Aw=
zQnTB!C>D4~>Y{KK2Majau|a}-3E4Fq?DG9i^CO=BFnGIXE4^EbY1qbVnO?Twx2kvN
z38s0y6pPS><Ywo16dSMIM@TCplT-~(ODwwxcyA_D0;ogSD5iBon@kcq@=b<U(D~<#
z9q;eNLcgv4Jm6Dj69*JzdE)DuHgPw=C!6&0nGLC(BkgGu+@90BTK5MZuKs9_v}<`#
zhdw?C@|JC5WLgjwR-Y>(qX^tJC@MydX4~?u5*F;R7br;cs+VIn83lK(>}I%^_rP^y
z9d%3nQ-huZEdjl)d@I9KIO6&jaEfBMicNNgwCU!}oZco|+~N`j|9Qc--b8kg2QS(9
zXv3&<pxdY<P~;eSWDF{h;#n=OS06o!Z)BMc60Tkq?*P?TN)jyFwQ%1PZIDa5V;=N(
zR#P14TA%Wd_D*|cwgDIJ#b$Hq#r)>xy^KJiFSD8tVGT)gr0FKJDr@rYJwx1vKWx|3
z=#UX&*@}->f(ZQD+Ulrik`IzN{(2t7RexxC!xduu_CotiE8m|oJbwJ<)hBY~9EZ*g
zSCW*vJ5^KDG%SC$t2llQu47n3X$#89IWq=x&<Z@>?8QCRaKCZZi@NL3zLO1nnHKbA
zU;Y08|3Co0vF6=srbGu`ukL~OswzHVShHk$7sGoux_iY39V^jc(xi-Icj;UcrsX!#
zv}!1#Ek{XqwRSDgR$htW<ed!C6P)o$Qg`JPs$Wx~k4HpMGHSN3rXuNA2hj~v!=$b+
zYT4!po2Nnk)q0c`F+-;yRRiP|G$VP2A=)H*dMuX`JHi90z<hNWu4`A+Zr4zdS{D%p
z#?=|gs;MTd){g9o5R1B#y0R+WID=MsEsyVARd6~hcdD0VC602@t--Y*ZEiy`JhA>%
z;Qi@#HEQvXa;?^#6c_E~s8wJaZ})TBuWA<7(puYxg_>V0<zUzYliL)Wmo$!u#kfPc
zCZTN+w6+G`Mvy?baNG_(F<Mrx?QaF6bBI;fKE&p?jk%f6G}IG)3zumgrE3UIR7tcE
z&jX5V%(Fu}tkJlHf{LS%I#tVd%p-MWZNJQ#R%Ubn9Bdrp+NYXDS@x7-c8{ApS3OB-
zE2DaG))syL07404W%=VI<I=Qjt~}V6e_r+H-qJg2E5hWujk?@g6m{G_wUe?Csyw#N
zG1j)MpR_qJ?RzqX<bGK^_o`*FQq_&3^$<!kQx?s%pYGOn)YAGCXE!Fng{I>(Ce;hn
z=Cc0JeQOQ6P8mRKr*I8=l`qSo<<jN5Pjjgd-u;3{n6TUdV^n5YSy>|?;#MR$A9}iQ
zPNJhNHM!oUO~ECxPE?i8m_Wfjs>BjZLZNJd(><%c5|X>I$yzaSNWGFF5%n3#suzLS
z^4YGMbL5H2qMNw`$#LWequMe2TXID(TZnwGC0Lw%<-L8YqMXtp(tO61yD(faKT3Rr
z9(o#0-d9HIDIBi5CfCpBS=QE(-yx0U3C=+^=3>)Lg(dH^wea<l*x3Y0go;??im`L$
zK@(3r>jY6}=V6cu;C`mO`0j|Sq@9jR`%kmD(A91A6~1e0k1|J5w`wuximR$xjZ4KJ
z*~e9DC{iR_Yyv}M4wxgoZ(>xRRoq)$C~4MLw%1orfpoZ$21{Tbqcuo*jzz2`$`{J6
zd-~UGA<B#;E@$Lu+<AI@GHLNLH1ZGMNZmi(6``j`0K2t<+Uh1q#^zD?nX%1kmzmDq
zn;O>YV3A6Y+fPwj7Sl(WBtXhao!sy%o+&*U-zO_xBq=Ez9%@vPcVe@tL#$5vb*`-<
zbvfx);^A<58uKeH4w&v!@ahOI@1w>ScwzYFx(_r8;^s)sYf&Va2#NNr$sf#<G_e!C
zk&Okvsd6Uy1x|z-!7|x6=~x%X{WyJUr_gZR=`DDHew99J(x;>m$t};3?!VT46#%Z%
z<FW%&T7SpF(i=Jj`=XYi7=<ki1)`A0Qa4HjJ$Xumj5Q+BM*dt7xvrzb`U+|~YR=QL
z9mQ}LHA!5`a^=+AZv;GXd5trL11IGiRA<32ECu7ZY#sq159?R#%ze3)7sJDJ%o6Fy
zL-Ty2lljxX;RFSqHMxD!{r>=zaZ8ZDZY(|tjaV_fbB|ONFz`Z6DE|OUldtzh9H@TP
z_9WE&Aqy;Ow@$l9KPaky@Px$JKWvl!_!PMY>psPU;P~1gpK5cD_C;sS;D}uFbt3z1
z6-p$!udxX5BpKrNO#cA5D&41qA&t1ae1Etq(y<Ps>{Zn47Q*kzSqqG3Bx0!%nxd1s
z7~EAm%{^|HaQS5Q^`VTyp59A&3P`v$9YI23rin)+&<4GvAPluhJLBb3k}5!A#~}HN
z@$PDLQ~3HCU@IU4IKlR+(xS*P*g2pKyNgzdvN>9Vy)Y7r09?^P3q=4rFZF`DZ9IPL
zRLE$*ML|JDATu>X`?Fl{b6Uq?4^vVtSRRw$MOU=8Rse%9ApL7+Njin?_5H2u$Wc*_
z)6q_N#Z2}{#x~|}v2#^vU=~rsxJBiR1w0aZ)Nr9`%ErN2Sc8*{SEqxYEe88s=$gC{
z9mK0-K3K~$Hs&C76y&&BlsdL^Tk^(A-M=HxYFymK!EPR3D{;rIXV0horP+FI>P<~4
zCns~MuXJUsONLiw0S94Hn^czJ7tL=jMaXcx)K-FWcI;^e<Iv}Bw97lon@OR+V+`{7
zXL{hP_8&^<G;3oPgK*LkrDbIRj<^-!DA|PSHybo@b4k>m#+ErV7CmbNq^~a-sB5F4
zfKfn*W$RHIV0&-E-}w@u{{XJie>&_z;<}@<GkcItgBynFiu2DAgaODL=DjQy?8>Ic
z4(Ipe)Hh=TcXL?PTA9Spk+Jotx1U<X=tooV_QCZHe2gD05SMPky;kwUe2c=Uz!lG1
zV@fOEIn6=5!g$R^rA7>_JCXWS;qC1ac+2|+kjrApJ;cQIVxap9yQS(;+DR{%zch>T
zx20dhP096UGLJqcFk3+wit%3Ra+~N(v6J$q>{iq_5L!ecw6%r=9o0sEI#elhd(vk3
z>b#Nma7Am)MPrP-l7WGeNv!=g?mbFVGDsZ(#z#X=Z8vmA4wR`;wD)f3UkS~c{{XvL
znZ;(sPj*2(m=($Ri*$ugbHJo(SJQ0DAd!b9zpWl6L6AS$6vfo!zza1c^6rUS(yb84
zCn1p!qSF>n-7@=C5;0G$K9e*>-7;}+e-wlgpx}XBQ?J>GzfKo-%sp1QD(-dD*iA^I
zAUMjn>+9)VC&E&#e|I8+2wWW2^ma1DY<ey8{F%=pxU1PN#5X2LcBJje#dbv|ju<6(
za)rd2gV_1;!M|)woAK3aD@VGW;St9Q??Jb49OD&Kw5Zo~YY4@rjR4IeMuh<K$0QHu
zSD~245l#p`wWZo-a#~oHJA%>{AQDD7r`)z|nRp)Jt(<ewm{44_G88HqPCAY%JGmIo
z=6*x46@2;~RMDS*Hez8PFI@hwKG6a;7mS`OI?}xve2+ppL6lNQURAOyt?;iJgopl_
z1Nw^cZK=&S?y>r$o=&#S{3MLlovi1}Ay|SsVALh4)j|8Et4(%T&dY(lbJTUMVD@^9
zd~JW`DwDp9!mJ!0G)gxbQ^%Hv=U-0sNbqpLgHqL*ahjSWLMg^T1yL=K$U(_9p0S5`
z&U!fhbd1?2i<~Wjd$_JT-v^3v+$nq{8k8rA2hi0>-*RI;siK)>i!H2jM#V}V2(7Ij
z#VNQ!dO{x0*w=mzzq6Mtwq}&)YsKu&sy{C2R!G<odJ$7cZ5n%%UG*f}W==MC8Fwnd
z87BY=;x$c4?yic%kf8nSX1sdRQ=Xe3*%4dmcUR|ga66V<8rT|4F<CT7yq>18Zo)B^
z%1cYw@}!}0(0kWQ;}r(d+h`}}JqNW?dINs^PJ2$g5X?970puR_(oS^Ly|L7^;f#>8
z#Cv<zGQG}(^gT06^4j0aP!7(R^fh=DUN{xbX7?pl+B1$R1fYzZ)m%dwWdf*NUio0)
z<mRQZHhAxh$$fh1+wU`E{z9;U;=1ARqLa|OZ~p)u*`*`_it2{PH1{F3IJuO23b5lK
z{VF7B9gOgPRvy&dKMGj)IjBXjbCxxxkrNW&XNpPc4`MfhQltbmC4>?rN{qbKNs|=G
zby$RU%j=$%l%t-NiLH-O_)o*#@z__V15q8>EQ0%+S019M=qGKdQ`nl8{{VP{p{iP_
z>g1g=<Nd=`QTwbTx#cko&ZijptJKq-K0o&ba^b!vX^HqqwD}~FrdBx$a6c;IbzAGp
zD~p(+kjAXU@!FMH%{3yh@Pw0%j9ZDNf;9QqO#GJ5HD1Q{;ybo5NJnP(sHWx9bFxs2
z(VI5*(%kQrm~eh(W7O4y1e?}2$vDD~YSuV<t#ep@-kN;9i!5#<h(>N3Z{64%rALBr
zcd#H}9P|~hv8tY?aj8YE7UT5icopCh6#$aF`qWW0kx1KEhUnPtDrBUVpp+YSH0`cr
zz#d~1=vO@|;4+rnvNJFDNv;{rcih=hhhoG$WDs_ql_DTV`FI5Os8P5+rRE^VhUr;T
zua^Bd&2&}x&MF@Php1dO{9vAH6piE=+G-~xX${5uB%PLrR*v1ncNLsun>Bj{xM#AO
zIF2TG3dTXl&5kMy3puT{b^Bms3v!GDezoh#J0r-8O6#r8jw^X37b__{Bt!F#gm$ia
z(V}CxKh2H1G2XhcF`u!VUvs`QQiJ8$SbfUA(X|hjoenzt)9hglBJSvGrYR+U?B%a7
zcF4T3WI13!10eLM3<)F-Ncp<~T`g&;$xf$YC?bs`+~{1BfCX5N1sltzLgS3`Dq%a>
zGojsFk*(Aw&<Q1SwN~IF#;R~z88y)r*in;g<@HF+vSV(2MJ2YM2EyA)kUQ5kSSH=b
zjGCLV&{!z9)6-`Jl1*m#(&AR|?~<*BmO;3B<C^m-X!I#-Jx&4*Mr%8Jg^jlP$+3ny
zio2zFrtLhNd7^@6jD?DMQ*%T+Vv2gRN|8*x{i$-2N?4?VXWZM65KrSz8kd)Y0;)<M
ze&C>U>GZErSsZhQnif9Md2t+)OKzX-H<CHey+24(7NUh?belYVC~LXSteuh8MzTW?
zR*xAu$rWqOX%h;eSGZpFjBgif9e(V!xPW8nOPqDB?R|}+?YQN08QdEcp<{mx%43m$
zkaYL2HyKBo4oh@*(w9MWs9A6B*z_b~wmiJ{ub!yiHigaVERd5HG1{pZ3b!JjAw4|~
zdNSYCl@x(fMk^GJ8DJbKu6Fgu_JGe!*0bn1Zgm!}rWGLZij>Hsy?POvP)9ROv*eqv
z^`C`6E4hw40jX_&*R3J5p-;Fzl(i-zidq0EMIaWC^q>c)-KH?=X%viWw%>NE?_QJO
zvA0T#J*Gd>w2tOAeiK&N^^>n&tmKPrt@C4oJAG>rwHvKU(hF!cNFcgG{{Xv@e*XYG
z0C9?KyBS4Cbc!~5{nJ`Dop*k=mr&2RmyaRGalrQdYj#)Friw<lb|&USA;J%qKj9z9
z)^2XzX3?b=Ep`^#OQZlifRWcKD|YJYD56+ng_doqOr5y{HI!<-Zgo!&8>JGbi6<7;
zvPTW1qFfELW90_1wW)6HCA0e^pd_fO!x{Pe{{S&tIX!H0s%cvK+&n>bb7?-GYiB4D
zOuJpN!F=MX_@4HAJKHa^%%%x;qhJH`f;WFUtFjbUpED}^RlT?Im)Ikh7T$PGxE(WH
z70#g;)Md1bO}K-~KkM9lkMyk~%;g=9Z(X@bHBGqzwhk*lHAa!xjFFs^imN`SZ4tq3
z$J&9J7q{Z&j@0AS)jF0?qNj;+MkxWO9l%k~VNH300LOxPCZq-%){*iL1GQa-?;rxX
z6u@&xjCy_)*NIB*UX%e{;>uv0gW9t$EsShR@+bnhqL>9Q(trwD0CZpL2iCfMKmAy#
zkkNXI%%G&8482tT?AJZswT{B-JXDHm!1Rv;M)GN~g!|jhE2GoywK(Crn@}ME${;EM
zZ+eNgjNur$H1{L(H4zkigtK3noq#wU@M;)kj_eD_2xV`W;=3>syj)~-)aBQdzxI^5
ziDlY6&Dd~9Y;jbKamO~~7zMg!x?Cj~Ytusr#dA9w3m}s3VITrWZ)%Qn6j3p5uKZ)t
zrA@8J7;9u!oWl{7!iNnQU^9xNXQ`uVX&V5lv9=dH3X0SBZf@iIw<p-)=GU80y0x^n
zfh}QzLf|3EE$f=}%lWriDH!H0#~!%mytMZ@rmbVmE?@U&-ivn3YxJ&Dut`o>)OFCd
zFkwXkBA2Z~X`nr`;c@=}iYg!a?I-iD!#^!`M`ULAD_piBeEL_Mc$NNQxvy6Qw=%1-
z$FL$(O}ysVn&+#zoJ@&2lToB^JjoPbvpK?n+|~y|J*&c6EP4i-uwWaBS9GoK=~wR@
z2O#=aHDhRLGd62)G#)^#)wh}(#PZ}SNybHU^*bWg%w10wWJXRIxy59Cv_`{u$o^He
zX{#u%h8Sv6Q=Lh5*svTiM}QY&=~a9~e|@G|`KxYZTxFX)FV?P?G@#Y=79w=%){Bz6
zX9It&Uury0G1=Ki0<L&xS-I1-7~)p_%*+*(^!nDksYb$cZQY*JY0Zis_G>NM70%4*
z?#h=Bo+}#q+}z6BMoK3XjTv$)J+|sL;~1=O5T@&ZF~L=!mAT1XMdaMD^fUzXt(^}g
zc@?4)DeT;YW0k7sjw*#=H(>RtROC`+BE`C#k~)q>b+&~rELU>^j=T@^@0!AD?CYd9
z-5TZt2GZ_?s*HYB>0SrWf8id{np=enJ;B_=&{p(fV~BerT3PqT9~lL@Rt@?|9M{q`
z%OWyy5IH}WwROR0aZPg0#$KBp!-SYhL~(<a4U>;dR(6>@EojOaAb4gnFm5*wYR$e{
ziI*RXLheaIB2vufbCKA3Q4rq2bBvr<T-Lrt&Nghk$rZet2xE@tp{XtqEve@u_4cZi
zY_5+ka`&vvi7hSGMGo$zz$D;cRa?lsw|MskBhsczM0QYjM>{+D_bOBn(VkCg!PTZW
zGrMlwGlj_QSmGV1rQM!`cBH(@_N|mE$C1*zUkhN}GY;`!{$jlDrzGFH#%m56Qk8#*
z8nLF&y>L$>txEWgnlD5}bs>&Z&J^^dg7M;D=^<A22Bu2P>6WP8vGE`{OP~AbR<k9o
z{Cm`^dNTE`BJPhmadhQ$XWVJ?L6X0`Pj0nEpO2*$v1&?4hFNliRd`=ys1Nst^Q0y_
z#qFIwKX(<#pz_=3X-jaW@RlEJ$W)G^u_leJrYd@8iV;k^5#6T|=~;6r0bAa!63tx;
zUAwTkjum3?x%;H_uRx0GSyfqM>N@A8Yf`;VRU4F~-i2G;I_CaG5>2uqK4o5}k`E8q
z+W_qtk-^U6T*--|cTAoJDGWa+N|6nQ&^*hL{6e=i%LJ0yGB5!r^sA1{%1_=$A))AS
zU)j9VFAnD)N~?JVt(EY$w3<@G?*%yv`O#yhGP%)sdsNopw~|D>wQ`&oWh6NQyG;vN
zlT_1|;#+eY?kZGQJmQAUeasMh_B6?*J3tuXv$-s#Dc$@>k4m8VRyYLY^))0j%lu`L
zt<E#o2Q`E`16|ShQE}MNvMO!x$GuL;ZI79V`c~=aWY(lRE#h`)$tAm*Yr5<|FsW>!
zfaEakie;Av>zagH5zZX<tr+iCH5g-%C=m0hk=)Y;q;r+Xq&0Foi>A3;8i^OBX3IyW
zd@4Ae=zp>+(}hHOlF06cU}5c4w6)l*J?S;?rX{hXRP{b|>yP&hIs7amx#bZOw*VjS
zSF1zKgkSCo=EM9(n4g4YqIpc;?Ht!Rt5`raIc{zjF`x`oe-Z6hS!l1&j5}$&ZvOz`
zj&D@Cj^X29u%u(m7;q0GJt}5~C>-L#IM6>(dW!DC(_YEcvNw`yN<74o#?wkfNq|mv
zgVMB-)=R9Y`;p^5_0t5u6P~P=-H{VQrZX&TTm#NVDTC=lg}RP4Il2u_C(b)lF9Rb9
z7d1S$c4ep3BD-7+ou;)d?nuYX2=+DQ*TZtl)a`{!qfQwammp+&RGacf-oA#pxjPNA
z%&t`j9Zg3f2Tb5pS6=yr4I^^{P;`P~g)&Dqoi4p^Y%o!{_N<f~i*r$A4dSQ*<VM)~
zRBNrKppF$t<GoxcH10#IeFsdQW|+!mSWlP3vA_ov7<i-8qO}(udpm}bKuH7Ges$?o
z(HvCn-L2Hxw2^IXr6TA`#Y&)3O=4S0B4=#;-2kqfx?c35d+Bq~q??4awacw?r|0t&
zk1uFU^HNw{D!REi0=;^1RXxMV#U~ZcT^AdpcQOnSfIHMldm0p7x!d10q~hTRbDD}(
zC1g4&6%dpoac(}9EK?LLN%@HH-nr?kv$e-Zeb~m7tcxQa{M9?D1IpOvCyMNX>LsL%
z`E7xevvjAy3_uqQ#GLK~arjpyS7{_=N8V0EwiB(zi?Z}t2&lY4Wfk9sCY|%T9J1i|
z&MV56rkAwNMjabQx{l7$LvpvbaCwUAa)9&K+lr}ap+K=%yGlfo-I!T{$iS^>rF1tr
zYjavy>@99uFD7&-_+sBSSK70gQRQG9ta-}kt}D^<Ij+{{L2Y$4&xfv2wx2DPS8o}p
zG?>~et?J~zW<5n(S4TQp*y>Sc`vV7T7e6Sc#Vx|wNEm~&0AN){nrNvv7xNb0R+Vzx
zA4-&*V>NCTW|5ofQo3h$`@vV%qJrWG=27PN<kFo!Xti#KW*Tl&vDR5ikde8$U=Dq2
zRykF=8u*E6QJ#d=$(hF5r8J{FR@j%KIjAkje9TgbvEID>PKyd|4nFlzlHMsL-tWy)
zO}$4GMhlo?j4zr1`qSF}(MDQy{{X(LRb{ykbLHEw^`C`6E8b=?*eY9Z_3KD%=u_?w
zr8~7jh*Ho2N+<zED|Da*PrFT%?ixcfV}!|gTO4M+8^X)~01}Y7#$+Qum0Wi+sqm4x
zmrVO))8B~;21sK`w{`E1)rSU`zWUboOWSDSjENkygK)vlG1ioqK#R7vCDNAPT=6~4
z{D34I_jB9|?X6{NnD<92!oqFGEw|;xV%^!)qiCzGh%O3CcPppeEyZ5Cxo43g45l!t
z&<{c0vgY2#?l<MXLHg>;JWfT#Z3YL(7-qTqjYihi^6Tvk7(upZN2>)H8LiZ<Elx)Z
z(_E|i(5n^otgU0HTuU3Rv@%N~GYpo_K>F5g?y08hI$0LbrJLJaB;(9}TodzS@)dDv
z#x63w`w(jOQLl;DR$J>wfZVA8Wd!3jxg^)O8m+^&qXos8<|xs@Bk&cf(V0tBu8t4I
zvb;L(QMR3enPaIf%7l>Zar)KuEqo=XAcDT4xv7_A&WEw$lPMN-jWM-b&$Fz74&3C5
z=pyh1)UUQXm3H(BO$l6yPnC!KAhUEgX`lB^S%<*$FF$HdKiM>Li`obLAr~x3`&wiD
z(@XyV2o>ea2$Jbpy#ktCqxPtCd_TOkaFR}{ebn<;4}kBWj62Gzo~2DbQ97t!_(P;+
z{{WAq`r?pyKH?*vwINgNXyp%RJorGcC2kWZ*wsn=6DwqWr9UG_D0@PK_!4aA?Q#DA
z-zK7$!4pLjN%pimMt<q1VD^VBMhO&Dnu|+74u}0{`qxom{;X8UXuCybP*NHI$<>eU
z&2!?lj>NskikcXjRsamuF`l2{2+h8Ry_+YIPSV<Gtli|1<U5WxF*vSzj_A^aefA`;
zBW{98AIou#yN)Y5+Uk7@V;rnn(eua|+f|0E?CRa@cep7-LsVZ!BCM9u23F+rQ>5Zm
z4yTIu=O-xWa(g=yt<E_u%|$K2Y?8+V-nmq6%_9>!=KLr>K5#v2k<_(m67^(+?kUt{
z;MWCuoYfRo$ST{Hl^NPtbh=3yao;uQniS72i8DJh5Y2*4dsj1l`O7u#9A(76A?ByF
zIf;Iifw;2&0IW31-E3g@3Mq(!PEAKtfcC$I{{ZqO!~XzXr2ci-XNu~M$j$CkyFa;q
zp|3XAveP*|je6KEnT=DGkpA$~E`ED*Ju8~F=5aD4uFK)~hfLF*V(AB!3H{Jv_nX$a
znb!?ZM7Nl<=jqKl>I~*W+<}}|EBBB}T(u0(Fg9{(IWxFA@XB%xW{$MXt!3o6m2MxZ
zj2u=6+3v68CKyNqx#?NUSshSmK<IS4Pwe|3X2+O$IPF}gjyy(}2)bRX5~G<|bUxMS
z!W@+C-0<f=Z;IwyMB(#JNg6f|rwpeenbv!*fvHD*;rn^i@?zwAR&19*u^Fyu3YP4v
zEy|4Ls=cXMu9$`&cz5kkYSFx^Jidat&3YF5L_mSG_xey~jz?RBrR;=stm}E2IoU^X
zTCIVEj5h~S_qa7e;dVYMgfpn8!r@yD(9&fiM$iE;Jw|a|kA<}LH%89tP^&8Wkz>=+
zs+>9HW^AR(u5`LRmA#g?HI#6McG)LAx@NmQMp9xSP5@f$<VmECc{eFsqit|Jn2U4`
zn#PnzYkNK1mk80wj1@u4F+SAoYoaP&o>uBqv9%Bcy8B#0B|DGJ&(jsnneFBA6q3qT
zI83b|L;PNXxa&PaH*K_8)Acr(Nh@5UI$SpOX2A=_G3!^v-ea?&;GBb9^dzc9T;ANs
zxgFpz?4KQeQ|V0mWKFp;R1wtlCc0c*JsP@a1nBd8iqn{9^3U2r#sJ(o&(g8AoAB=9
z0kw`gdR9_STAdVOWn&WNK?Ft^+xL<mn+G@*O{}*8UVbiV%|&!ND!W=Mp;}8ujj{!G
zJ`rr*PQAo=ADFI0^*N{TmG%6ttt`pZ5xS$YyYl5%{t|F%e$lljqBQR9#IY6}jAT|5
zNa*AqF;_HpC!r*FC)Ttq^(n2~G*T5mUbTwQ&QZSSNn_$D<pFL+#{Tf>SEsVJy-mnM
zll)6tr_7H&QFPhQYBpC^WV0^#^TRD={j2x!Wd8sd{&cKrnri6L@i^aQAG6ma!whOC
zxaoW&w|s#58iwu>aTekCmXaxuLfb8w4N|!ngdcj*62!2LrM!zBL9bHNRGW0qTvXH?
zOWSCXa<!TpA`s@4qX2ZKNU*39uOGxKL&Y=wn@5sW>Z-(kRdLwLUlKXnTcM=*QD%M^
zN5|CHH{pvp^^Hy0i3y$3KlH0wMYCA$k?HpKme%nEju_MdkGonnS1WC`1YuBiTIOo#
z>1ycF`$9@Grl6U^p(NG}#iF7=$)8%QaWGm)&QC+kykUpcq;Pm;7_1(N*P88)$kICq
zuE6ymQM~8xAI_@bVKT?omv?HB*g@WlZX=KV=KhsPBqMV4;=E8w7)DJv4^EXO5-$Rz
zPTeax6puytTKK%VU-m_MaB*41?kBj}<Q`EKUc-$2YOi?`UaH!P{{Y0kfA<YkmW&qX
zk^}yg{{RC1wd&Tt@!=2qg1K=202!ty;Ug1bSqW?^^sEh9{?<!|Ss0k4P#BNlG^$ir
zYF))fld^GL4pUUQOK6e>x-$VNWym}(dUH}w<lP<7UbqGFM$^*0%1L`D$H^5wWba~o
zJE$#TmN_mTm%^|@a<y(7h!Rj@a_Gy0k&3qjtj;cJU8YKANbO1~B(cf%tjV<rcWi-h
zN&Ci|h2N2CO)Yn0?b=}=d0Zb+S|?=DlqauB=c`UpmC?01U0Ip!Ycx(y-OKDMTUEm!
zEsCG+=9OskQ@>NDo9JG()MdAu%Y1=beqzcNMfI;9YRu@;#o5R(jCH1_?X{I{W?@74
zW^S)$E+YqMET_|&=C5s0h8O}xOg0IvOu4NPA;Q&7KjpZP4uBfjJq+5jHE+4mTmHF<
zqiO!3MRZo^Wk+Uf+BN{J`3}ihdK}Wg&yu>oT|V>Bq{pI?4?LdrEx35pt8y#pXw#Hh
zpF_{ZS*}vjD}M<*`qIp$R50z<wyOD&$woY^nitBBYR{Mg=dEK-9Sy2WQaNPE993|B
zZk!6}O3^Zt_gf`fReU_0jBqOx#9FjgaebN=0#@uh3gE41#mJf(*$?pazHW%m{7n4S
z&2nhq@V&~(AX3qMpq{5S=5$3riJYCLwE&98Pm$tOn$OEKC;{SZo~_h>_3EGzohHJ`
zZlm*3^?_jBJ^q!nH$i5ZxiEQbk_oP1dy?B$KtaNPv}+S=am?uK8yOi+{`Kg-moj?n
zYT3yeXcv+13hc{;CzDz_HLP&U9FHtuIL^{}U#(nhb4j#byOQLVbd?wn;ZxhTD%G>P
z*gLyqu>>hR*IZ()adUGk3ml&?^s8=j9!4=)&P$lyFm^Dn#J3q$a-8$@q=FT7juN?U
zpIYgJeb(9Cgre`v&Zh3(L=Y+x2uxH0=Kco28-;#%R?ee62AOv6R=Kv*OwAJ#7bg{o
zABz6aK1kfSJd9S;jqLo6N>11HCH~aZrT+j|Zh7^nr@YnULghdnm6y4D8+**lpIg*G
zEL#GouCIM7?K7~+^{w4Fu7+mXX|qcbyIe?f{?Vyt(E*Each9>T<Mx`JN|HQI=lyTt
zPzv|XDC`v_zxw5*Hgr+_!Stup7=;u7Q-?|bQAph=0m=7i7-V9Q$l`=6ZpC(<8vg*0
zNBy*FsBUFf;V4O=yt-Z9rIHyXIA2xnYmB<zEaeePF*v{#Gey&t?sa-4gqBunC5e&S
zoDo}bhQJ$6ewEE6?s^evMAWfMMnETd1dMP1914(IZJEP16*`*fh8vkC&RUq47S+81
z<{137c*!W<y{jVcN7L?*wbQziZ~_jMvco|$Cy1!y=5m^zjWw0zmWp>N9A>%wgd~x-
z-%PD#Qj@0qhZ(7&lr6nVo=TeNyd3+rS$#aC{KYl#6F-E`@u2?ziPnGE6@VD21RN1t
zJ<Q{=uVJXHn2vLX=o{9nYRD4nBclw8255LU{zG5z16}#nW$^q{>Rw97h^mQ`{6ygX
zn5tVC)K`^^tF1>=Zxx+3DDBcYKX|Nq1jcjuRlAQBX|cy~9lFTQiHPJQ<#zgdQfw42
zYZuO|ZEd8*GshvgmfT6e`9l$!<@Jvn%Vl~ON?_{IfaTDGySSyzsY;#JrmUVZy0p`D
zaeE4znPuM5l{pcU$LCu5hOenwL8{$L<w*+x<f8n=az+ae;o_aypCUP^JY9Nv#~0*k
zXYpO;-Xn|hH9W??%@xmz?}sZKL)!+Jd|!JamGYquJ7Cdp_GqOijBmjs%7;HgTb4Jn
zY8nZ-lqh#r$X=C6+?hc|9%|os{b?#qf|O7MkbkWoTIwu+)ry%77b$Bpgra~k^>gwy
zU;&fGYaNMumhG&hoQ^QP$*p^9c;X=tWxJZnq`D($N2@!}5Amdh_cYxGKy1o4UB^G2
zXD@{nNBEL0Y8G3QdRA|fB!kOvuD*t{wa(QmZc7G-0L+3h8OQLGQ@lw8K^3w`YU-hi
zojo@~ZY3wUl04*bRPHb3kQGz3cQw&0WSX>eCC<vo!@izZK3pWZ?nO~sK^|nB^4$k|
z<3~FkYbciCf+o~O<KDYZ4!8VBCeQbq`BGmQl}~`<CFn<9D(0SZ8htA#v8pGzTrfVO
zpsk~^gWM^S5)|U7MsnQz>=@DuuwQO!!1q6e&7TEANBs7a`PX5dE2=*VGkcXTc}5<h
zy!*rv{OoJg!SykzvBuBDrjkROcKTO6ZOr0DqG<DLnxs~dkCdOCzq&iut@%(`+QnsT
zp<;z_J?oCEv`)%<4%2^mc6|Wmq1Ugxrt(noM>Ws144=JJ$%5=0<nvSgsYoLtq9;w+
z1FFrHI3U!}qDG^j;=M23t#mxc?V9D($Jh0{Ya3?`C=HGP>t0)PF}R*dnB===vgUGk
z=yuVi?BJ8!jZvI{s#y8~TUuX(we4Jwvgz^=dNYly`I=IaiqzRA&q~w09j@4DX$}=8
zSjZt{9c#}nXYm!K+gseo@yMqH6%~_|Su|6-EWOmrY!>2HX-6j&H-_bv^|i?*3LNvs
zdsa;8mdALzkw<Y^o*+egXyhD~arsoHV`Gfg+jR1p!v}51HN18ui*gE*I$-2gNxp0o
zR3bRo7TRf93brU3FIe1&hz<)W;Nz2Bo{j#WYLZ*Yw&D}}u6fU)_BFO2S=iG09X_9Z
z9oL90Zj~8ik|L5R9IG#5U0u`W&D+|$TGcZTw2Z4+cSNK6rnzg`Hr}_OtmP|Q-YW8u
zsUDHO1$gA)*P~Z6tLWFZ_X1^vu-*JT*Pm6kVH%41m^W~;-R>&O&Uy8%Eel#cdA7K3
zWN+zD3s;hMdo!vHY;=q=GmXl7R3bRg<QB;}<mSDviRw#>S{3fDOmX0M;8#C)a?ud6
z{PLcoIRce>$}#5bX%wLwk!kjiB(7xywsLb?M$dvVUppx_?sC0d%2yVwylu#>4+$in
z+LHjk^3VE;aeA50;bLpBIMt?LI6XyWd2JcWsmG-!xvf8i8+vuYngJ0UudP$nAdPa$
zr268uz0uT)Q(8$H#LJRo8O0=QSFUm^Cah@VbXkt%?9FId>hs#RW0Q~Wo|RECoD)Xm
z8uhKxh~~s@eR>+Jd7@ihPaKe+E%$@RL(Oc}&IX+}Uet7BaSxR~<xqHKtW)7jQcv&Y
z>+M;}Hxw!~v>pEd2%Vvi9(=uhs2&rwQZqETZVv*IiYe7(5_mGy9C_1zLZX|%u*n!u
z_XnQKS}2k2S#w4Be;ijPPxOP1nOf}ad^6=BQRaRkrck>DUd5mIM;MRd{KY%|5u$OH
zBk>fyke($E9vC0<)QUslR6S=ul~=MM;(Ucq3PKn$o%lVe+Si6HwM##<%`}Rn^S2z;
zC?sa<w(e8E@NScHYZa{F-X!O4;y;yK()=x>X;67~w&F0yC+8>n)8$sM{k-pVPk*H&
z#2y9+6OuheWtP~u&2!X>O!RQI9OJ0ey)d_4Dbig{a^WK*Vd`s|O6NsL^(Nhg_8luO
z_Tw2D$*j@GJ&ps#tG4n#R*WCkx$rpkuGs8p4}}2aigwTi1pw<t+|p$Wj1R3AqxS<n
z1Gn_1iICDhYI_`xl|n>xqpeFQaH&x9S+bGed^2onR|EZu^w8CU-HDBpm8V&Z6I#YS
zNNl^BgHSoJoBhN5s-^I;Ezcxr3f$pG@pJFZdX=+}?D&uS0=Z~>M%a&p!PIUq9_{Vm
ziLiMm)K*2^g~4Sd56{5&u6fg&n|3;BRDz0E*2WFql{My6VSKjbnBaXWjpoEcC_7Z(
zk?UVmgM*Y;H)LmB-8&Lmosmm%DuRiRl%A%vO2)|?@Tfq;lm@zLQ%)Y~=3S}Xnh1!N
zq2x{FxKUZ`0Gde_Fxv)1ezc(rgq`lwD$jONV{s8fCUT(qdR2>C+>Q<^=S@l>NxPaU
zdYf>@m^_&uaBg$K73Vc%*HhM~9Y>)_n_<CVe;U-&g8_!=(!6QvTG-W{wnnGkpXOl!
z9D`hy<;ycFD)iA5!_9@k>y9g$)FN1so<YE@>B{BF9Q?Bf+!SC{8zZ*jZ}&}gbTDq_
zrPF6aC;pj=qh-u%T@|)QRCZ>?ze=Tkzr1odJw<UvwJW2#i`0hR<+2rrCa$69gNpkI
zxmBj62d@7Bd4*(@z2%{8Bu%5{!TM0`y-3I7T`+{4Z#@~oD=B-s42iaJQ;81T3L=F)
z#AYmb>)N8Ym&ibHIOVFES|ghkK1xO<_l%-Q+8b!w&FPcSR~dJ4ER(#_IcI)JHR9rG
zDLzzn(1TICJ6{SubTOaedcER{ds#GpI{3&ft<)9~^<qu{{Ogw2u@Y@b*y1J8EN<>>
zB-7<&wY*}aKPEDxEmKLQ=~osLTxmva_KY_NIS!=bBht2tA2f|ew6=~J;e*W=5g;Xx
zM)l&WT-#gPJW;saAWX@$^e44?KBqfw<XaiD<&D@J;C8KTA}5@D*<sL{SFs6Qn^D*l
zA%^UB&09-$zy~B(WjVDFlvgkmV1896=~fplZgLJlt~p7%w>u+lOBvFyAQ8`6n@xqJ
zybX>c`GtCw=gUS<$n@#1T6a2met2VYGE~&TQNM&&=QWqU&6dNj;zOr6AY*ND-XDTL
zwomc`HyYBbviT#HlDqk{r%;2UfmQV;0@2Po3g%Ai>1#uayw8#KteA2x&>lf*^yRU|
z6VNTgeUS&XSsRQv#d$K^^j62uS3m2&3WvRYDdopuDo8k)ezXmp6kqDs=})Z8LW%%d
zS^!Z)bf5)Jlu~093hVqb3u!<Orb0iJR5voK@S2ioQ~kC$0RI3)@)|c?{F2!{KN{if
zne=NL%ao68+g&ZXu^h$@N&Zz*PUKT+S{;?Q+nKEz@?GaSLUEpzEZ0{Ha2cjacAfy|
z710__3(VSlt=ViXtdMz|r3ZF+BcQ7ior`hiINS&)-zSREB$3Z6^9t+^nuHPe&MLHV
zE9Ip)je7gjw62jVO3j@1ypMBq(<&VOdkW_7Adhppr_2UP^{A|`2JCN1S(%zvjXm2)
z$eCW`){liFdwphTTw#_n2>eBHLEc8SV%0mD{xs)W$o}B2ez#^Z*a5{!CBsV^qw}|+
zq*1xx&ULT&0j{^knxuM7t%ZwOL2%O{nZ&~&azOr7OJf-;%@JGPTiu;ASCJ#Nj!2%;
zV!y(1h2zqa4H``|O4L^R+S#H=i$ihzOnYF`9qpm8$*N7^>vI!EZ9c@g85!x#U}-)O
znmt}gE@Fz}M2F6a3;d;t8*}>9>vGeIg^?Y`iK<#_*1D#y$W^CYw(Nx;;40J_b>@pM
zvwNw^B!XyJBVyU}Sdc~qGF_U`;^EXCK1l~YwQ@V5BK(Xq=}6V0M4NU)4r&<`7Wobs
zbfJi;BZfaP!S7wRjJWW+{{Xn(%+*ASn<LKLa(=YhMT!j+0WG-w==#@DWB#mENNBf9
zP*9Xm20o<|&PqlyO(m|3AwctWr8)AeBchHTRTGy)S}_H}qXIpu#Bi)ASoJl|*`rB4
z4!6TUzSBlI8;xgMT?;$MMmSQ{v-nYs7ix{2EOMmraKPkx)n-e07_%_VW2x5cSDNxk
zen>{u?#8P;@W3+iZf|;|jN7@Hb*KrOYJQc@>N?XybLFlIj<~I38<|Q&6IVclMQ$>G
ziH%8O#^OejY!Y}L)Hv9eJCZ|h<x4g?V-?lt3y%-W{_}q-h`)G>p8?1`cxL)l?LIG&
zYbiYqQ8nt}m)uoK-a8n}aS5h4&(f@uF7hp;vesB;^L_JI5i{stgy%j4xBmdH(tkSb
z1$9T^A(iJG8sL0DzdP5bgWZ`_*y3(K)|yDs=EO0nXLEoDdcv~Q$~&Ho;k^w!JEh^k
z`&`_)^d6OBc@II!HRslP*zQ_HyI`wp87<U|n&tdKa>n_NdWz0V-LZu&YBq3s&xq#}
z`BrljRO-XMLv7;yN)}!0k|^p5@_!oDQ&!aJW~?4PPx)<Z%l_%jSJb?3EUHq?Su^}h
zDt(Tnb7*sp*P5ls{{UG$tat!q*J<JJh8MT0lj=?OXpcKTh|{Z8Q$vIt+2}e~g!CO0
z28PWZPw&a|8rf^ogKQ~}WfkVSl9Tf~)ijmf!={!)Y*;=H(_VRTr`}C9#ltd6<0q5P
zwRNaH&InhEZCK>*w7p0qj@GPL`Vm^b8q#iU;J2Pw*;%q!73Qg?k#3}3$7v7@l}}fN
zPM<LYkT|T7N1p1WwZo2wio{@6OrP&pO4z~fOMW(p+d0o#r7s!ussvO7`GT)SvgJ0O
z98fWVWP(mc2t8|SO20qaHuAi$5L1?D_%xHWlQc)7`BPe7MXN(9O$Z8R+CK7=g5Od4
z*JG$8V@+9MaPi7Aq#m{CDf`J8+f;M*FY`Dj_{bH@YI0)zfUZiD(VZ06y(6f%m6FmT
z21r52;aRsj>;X@m8!tSO$Q92i99*2&u@9f6YEkVi446F|iq5+5t(=Zzw=EMBkU_{F
zg-w$MDL~P+)9;{wD>#@AhoLmdW12s|3n}_np-&L3xwksuS~K5sK6si?+>!59Bh%Hb
z_p1EB=D9I7y`!?(3Z(C%*p@hXS3!pzs=;<64H?aH8IVf}Y;n}|t<MeEh^{2szcEfv
z<61{^DeON{(^gCQ3k>F?O&;72nlOH~jNYeBB)L&u;?Zrg@TkD*Dv`Xi&e=hYk8{Ya
zAlo{64)WJ%B1qkPU{oT~1Gth7Yogajr-i2(t1>Bc#PSOv#wsS#9whS&z%3;mNyOn4
zU%FPB2@?l72BWvuZ?yS<jrYb6-D;_5c~xmES)F#7;$Q7juq%-%>V#KGr|A&5Z?y7r
zk%NjBr<GBAJqyXD*hV-|eyu~vrQ3iWOLYGLd4*MyIqxY9lT^{;!8dlJ{st<X-ZAj~
zx!T%{AdbKj{V2K*XE)Hbr}(Q#yeuv5!<FlWBl=b2u3HY9o9HSXM2#bdn;LcKK;k#3
zTf_(WVb9dm`&4Q5SJjL<{Kupqel;6wcZa6Y_)+Z}AJyWwB0A;q{xu))8t3)BdsI56
zpKA@{ruQu2c*v*B*!`3jL!Hz-L#$u;uvtN<&ao@)Wevw2Yp#^tNdl_7KdxG)ZRl-I
z5nCgxflPWLD{@35Kb3RR6N;L>0|}=$shf2!+|_t5+8F`e#(5RbYn}4d8xma)Bs!de
zIIOd}RxaI1;<HOr&zZ^i%{JUyH>_XQxu9^PitCTUZx4kALZ4c8b3s5kH6R(JFm9Av
zXY6O%Lqa8t$28vcV20+9H@!#_4r*AdITnuh;h)W`P469gilD3(>?A*6^rxzY=rLUj
zV=+m^SG_s1n18TpCiG)RmqZVq*x)N1mK`~-MzhSCY-D#qT-1IeOhfldGXDU`RL3J>
zJd>POUDeWAy_LKnq_=4S5)<5-)upd{{{WGkFQ;pNnS(s>OeMFQa95LuUdJ_3SXwzY
zXM}Fvwe%F*g`blAj_Iu|%P?E{wz7hAf(aFB7ywAX9ZqXmO*p?p%#ySDny&IkYO1OL
z>6~V%%LkUy$r;0dI#Ehou(6esWjXC$NaL8d&B)Fw0c0UycVv;1&1~nTjVVK0q4IDl
zTYedRGhS@ICu7xtaZwE0x;A=>-nCzmNavdHX8q#27j8)M_P4JUT||mH<X0Xm_=MrD
z?DZv%RtE;LHQ4f8mcp{|J!>j&-YRjjIUS^Ks*vPQDgOXu*P?Nn*I)7j{{XgQ`Bgh1
zh}2fKx*5^go3@s+NCqmtv1;(`Y<1w*2Qucd*-uFsQJszn9cxYoh=s_nrOPW`ox2x^
zr_8h`B0O+2_*GVpc6_VtQQEz(d#fYPRIRLuqt5}l)ZiY9IHa^%lQ^Zg^$Qs%IW>>2
z-56}!9a^)Lv~oO*M7^xkd$Z2wkok=H?@s{Ch)+)S^YUl46SwfIi7Ymbf$%E-0Eq9b
zR>oUBMWkCv%Bg6^N|DZK$8(x#^JsIISDJ2>r|B<a6tdeT#D{deG=l&W`u>$-IW${M
z7gT*Q=Wj9q5123zlfWLevL@2E_1sM#RhIq-nrNd8yH%ZdGB@zk?=4{4b1wImXo&MS
zU<02_S9D$CjHInC2|PX#-sM^Lg8>;MrEO@C#T|sAIlR-nXXPW;)~V>OXBF?*=zNfE
z#CYH??2+Du1V<BOC<KF=sJCM65)8J|(a0kLPNd?U1dBATn~u}p6_mar^+~?Qb?3`+
z$}k6New^EH6;$VNIj>5h=Aiu#a)SQ=XIJcWO5f=Haki$4Me>&<WPzIa3_|6q%_FQu
z)FfZ*$2lK%xql73VdmV!JIy-Z4g8Uf4d1Ie11unY;Z=1BAvRBj?OdyKp{pZ|y=*~Q
zHs?2J$Fl=oqdfF_9ku<iwM@E>J!$^{+O}#j=BMXgXDhzPq0Fp&{d4}e@ThC(k1{(6
zPlx?#{b(CHIKR~|(w|wFg%kjymVg$9=|BW@{7}X_71#J#mOVLLhC<3tKPqtC%CEw6
zxA*h5t3Ak(g9%FxdUH@(Ytdfn_ZIS62DiL%BPls5<G3}YA9|vT>dTBS(X6`dycaOV
zZ4|y#PDwBPMD?wxF0AeVkV4^4E%Vo*Q+p8kuW}!@Tc+bIQhcX^4+qktgI119i6ppS
z(aG|%`g&rNEo&nRk!{_w`rTSeBUspolRqwbH8R{l(RptXJd!ZgQcBk%F_LK+8q7-Z
zmPIPQ=DF8Ga<OFY7_1*twml6EKTj*Ch}E`^r)j|^sMtZ}c$M%v+noEHit^~ceCot6
z+??->$NWxz-~ze&>}M2qK}8{*4}(ACHU9trHG9T@zN>Bt;6wdQK7@U<5+Lk0jNlrc
zF>!TpAXau>;1kV5W~pj)wpw(~+Zmk^1~5q#vumY?ks*jPJ8(I!IZ3x<cE%}2?3+{Y
z@U-&Pqz+wo92|koP=~^@YF7d!bB^S<097t!E>kw*;m%`gJ`Z^H87%Fv?4$)J_=y!m
zQt;H;L{M&(WtLEKsr)?St$GlnCmWouYUs`LB!*3r5x90eXqoo_>%pwrun<VzXx;`G
zbrsy`jl06h{^L`LolTECaq}OoGLd0RKoeUJ-jA(y7GLVcM23q(jMNkq0iR(*c1E;m
zO03i))T=$@Fp{x=&)oV|C%^Mf(%X-qtgPhEUl%x1f~8MnU$lved6EozRV_mFw7yhl
zh5OYpSCQpPR!5-t9uxh7vOiX1inDX!%NKIYuao}OblEg>bGWx+G|^yye(mKR+szB7
z+J?-FD<5~Gk|68Me3n+OHid9KnyWX5ET=hPC?2Khs3cFen=4%C+GJ6fqFFL}w))3!
z2x&4tStTU@0MkQOppg=*E0V?Fy(=T+ln?soYW!Xn(<5b7Nq_IUnME*FBwt?(>B2Me
zN%McVRufHz!pWh45Qis|n#DyYa@|QuS)5F^N6(Q!;}v$&*d$BEb2GINY72~gD>;rU
zcx+)U$~SjjQx0m(FXluIYW5_8;XwyAico#00-Qw8puQ9#9tMD8{#!}>>$X_i(z;W!
zAV;(=4SD|nh~V2KgV5KZft}>hl}(Nk=?|PnIVXzmJRRYcvb9}W`9FBWF(VG1^kOt*
z(pP4!#KuV(iKuPUe5OX|y(`Ij$m*!=5;-q7-k4PIs&n<OTg6t_@{{&GrQ{gNY+zNz
zU710sJx(C_PgY}+vHt+ybbrDns(*4n`|i36_K^?xM73Y;NB;nQ*2a(U*H+XeR-7y_
zcYWFXswX8Ca-%rO*z~;%!Ww>q68)Y9d52kFl51VoW5CDfUR<F$%Ixk{=|x!$TS=Sm
zoDZ!?(fLcj1M#h)OP5x0)x`6@Rk2r4ouksFk#HO#IQ%Qg?L2RAW%8%DMj6Q}NHx_R
zj#$T;vgMzOt?zHr-bfuc9^BV$2BB!dM(kr4+%PdpO%~}(p`{DkIl`1bN-k_Ir-hK9
z<aRZklO5+1hSfYNX!nvZmOxnNwgq9`cs@)IF6l}4TG=fKb#>IGb>WRj+KfhwRQ7JQ
zk#nNzNe)evpbn>P7Br00GK^9wIYt=vrmM1?sUV6C4G3<MRog0qR;{2zC^0x7_BApz
zj^|yZ>UUO>2w_H%WC2-EAcNYw`zU8!970N>V<{1A>^;ZQl`7V1=FS%=Sli7G{{S^l
z50HIplf65&1@$eD*0Y@N5ygAbI=pR-SI}a)Em1<rELqM86sB_0-JHX-MK0h@YgW%v
zy+WcZhUcg})?CdSN-3>QvddYOP<hb?<Ersmu<BNm4Xk$LbGUI@#&_J~ohG&>HWoJ{
zcEvdy6*W<`1XSA$gbzba#WEDy+|QRpxs2pSPy4*pX19=VK?mtjqRo_(8nJ9;P&Y{D
zis<x9e<|2?J9#x-Q4-mbUdR5F#;ZR9u$JddFNGP`hvYTP-XM^&V0Ejfxz*51D}=XG
z8@}KVPfFB^87@E7r~|Gmu^z4kngSGt)r^B641G;x{P!*MpOQY+rL{Y1$DKAo94&iu
zu>tbOB-QcZGjankF{$orrBRHj^E1K5%~N(STia>=9#J>Sg-_mYq}NSrc5YKKugRXj
zg>p4*4+^E_?=y+<+}F06h<K%zIG!LEYY++NHOJjeGpcWu*!RiJa>=`$l#=93gH$_g
z;crt=X|~4h@e~Y-rLai!T@K7!T#yDgFE!Gl=M|IOaM+l}F;VQr2&OL4NLI8)EQyzz
z=C0+9yiD@H-Lr!y+cijo8^Mq@?z44#{Np1&<kzA^m?}N$XSt+yI!V+egDB_|70b=Q
zr&pn#5$Mco(}HUrS&Fq_ci%&AN^^5%mOLTFHu7IB4>&&6pHqgW=QHCRnDpc8Z~aYj
z*}{*dbjM?OY=&pvoF17(tjV}-&N1Ge8l{X5Hj_=aKj_kXN<XbKTLO7C6C70nBWCMN
z&oqFRHutM>0Bbp#MDBhW5H+a=e}cVWZS@t5cVeF7!Hj<pq%uZ7jDB^#mNFV<hrEOH
zs<*c=3rR<q56XMiQfk9TnVGR8k2HnJ%GZCS-Nf2tN*j=aHH8+G%`pkM7}|)ud+8>N
za0BHHkT|SKC%d_kEp1ihDt7?C!hL9JeBP$8boo6Bi1(JE;kN;ppO~;5RPijtc``4|
z4&&=y^OU(`ZmAno<mUMzXp2B{hXnPjR<6%*t(@bGVym0?Z1d~IDI({ZP_l!SJ-bu~
zaB}K$Ppur%Q8R=h*x5JC-#O{naZt#B98{^w=VT)ktco`~f}OoZSg<PYjo-aRcO?gH
ztj>yCm!*uY&QxVjaBFhKmmxQ1zIs=2HD$Zx-4MaS2akHBw<SJSTpGg?z4JUaw8_Zx
zBiyG3vUO|#LP5R0G$VK$hv`^HNh4WmnmL)!e9i4yXfkGRV_t*~HnlabKjZ`d0Bpzd
zs<uN4rn)P1Fr%|NiwWaI86u$8?jn*U83&s47shs`V960IIIA&4Ko#v^C21Q}n~Y0|
z3pO!|m+b9>&u-Q9s;pFYJgH9d7?A)@;hfZr`$c4==CwI7RW%{pT>E#fTg8`KP$qb#
zh<vn}=TZ0T#Nx|uR;leI+N~Q5XOMj>!j{Lo9nRmv0~%{LTsfy}`h~rm9(C1>cagK>
zWA^v0AahSy*&KA*v)*bjPbIuwOc)7n2gn!G6+W%6E#zj>&eAI@Qrns|{v+;dM9bvW
z73{TH?V?*{A!Ao9kU1ybx(T(0xU=%oHCK?2HOE%Pc2S$<v7F$qayx6OEVsOPD=O?@
zeJguFoz^oR1{(zAdUUN?(!|Co$mwF8mAt@tJx6M5%+8^5Fe_+A`m$t{U7}QZ9rGDK
zC^@SZP31OogNnvfXDRboJ)~55TPofVW-LwyX|^(Kx=b8$O?sTaxVMq)QmJZBwZB7B
z_Ib{qA-7%6xB*#;b;kfGRTc19<po*Xx-M=C#obMH1keje-!MU(Ru+$?-rYy%#Xpyy
zYE@+wO*0x1laofe=z5EN`^G;(P~T}+R?rt~mvP)0<BZob(CU;Oo~9M~D+uDK*dUhm
z5cFmOyX5paA-UOq!Yv%QZOxCZT(<DFtbDQ-`kM3owBCsA_Hd5pf!zN9t^6tzUqX4Y
z*h+jq>sRYQ+0jG(seY9uL5NX63Mpv-qJz?a7B2NA0~A+X;iccDg&j<Ue=4Z%XIJ4f
z4~DL8^=s>4JTaIK=4Ut<HG^%T-D!R)g4%pG@XZ^c%CG~dHMI!ZtXu9?Dv8Y_p<Oq1
ziQOv3UPH5S=QVFkvUs6g&KX#7>}%G<Q+8vL<(oz8h*m2XlWMaT3Ii@c6`!hDK<KX$
z5=xA(PJL=9#qyk!_^wSB+eWv6AwnDiK^Z4L^`#VVZ6n4JTb?qxr&h`Pq{ZJxL9E84
zHWUw(5(i^k?X)6Y+eXrYxSO*(X)O@nv)tNDvLwaQOcJ^2QP~xf#O}V9{YTVSl}F<{
zF+ag)H{(Rcw0$-T=M8oqe0~%*iUxFE3?K11{{Vm*?lsFxc`m#-S_ES)#@8i&@Xte3
zwlSZ?L$28Uk4wM%b(ORS&H~8Bp;?FFR&;v{TWxaeqD39ULm!Y>4XWElYdO0YO*t>?
z$h9rh@lSBKP^X^bYZAEIj^d(<LW~(n1a}qZ%1-*8q#}~jV^Z{QhSMMO%lgz1M$0k~
zNco0&>&<siqUA1glBGnW%15nRm%ovhD*5Z_#bWDLiY+Y(3^0w1eQVN$mD)MdZA9aB
zeLbU$m5IFHy}R17VYwsakYraqc-lsiB`f8)$>8*^>qH;%ORhR>Z{}+l?qySB&lu#N
zV@&TsVMPE<Y=3$_wba>vs}&L&Ez%yf1ppa1sP1GUGKt5jsh4Bcz(<l2tJh`oFmGHU
z`Dv2Jsj2j9E$7XYah@|z;fc?SY1ZXk4Glv4MxIPq&N=q1Ygn4*;2_B^d8JM~vAM}7
zc^->uWwEf12;^l7Is;DoRm7cVfBWjQ=4Skn><v0ZyQjWo_TN>c*Yw!34b+T&rm1ti
ziE5hMp>^Wulo4u@GkUXeSx>9#*YA^cBysvHU{<kCD(uq@hx*M`2(3DqI+m?|E;dJZ
zC<-z6jZdas-%97qyqoTy<23o7DV`<}_VP(-a?Q4(c#|FXlALzjYVD=<<kw^UDqXq$
z@vNlTl7xF595UTW5`Uyg{{VebUOmbfSM3jm<owmm&hNRmByF(?jdJbW(^DQ=f$Kvk
z&D0y^9+j0daBHF;3C}&ui`=^d?^<@B>K3(o8Ijzdl-67kZ<x2e2pRN8!ej%$&~eY5
zC-bh-wObM`aXu`y+cbwCcxJnBF4ah$#vg_JE-$Z*{n~${LUIBABVA6TbG^1F3cLU+
zv2*vIkm$z9%6z*7UP+?%;cRVA4>-X3*PgqTJz9y&wypLVB)M=3s*>3?%G~(&;(6Kr
z$sP&g<~=J~D(I?|(_~-po7{iToPYPxr$^&ExugjliaQ>F5^Hv>XY7&R_-?}LXqM|z
zw@At5V+8T_uC*?9fU1B2&<gTuVmYOz$9yn~G1ZbR(mzB1dei*T7|8&8n%bLoyEv;w
zUzRMdHl|mI)E|1KCZhbEfc`W(mmDZ1)Xcxu46wbhMch<_&rH{iI{lvdMeLFl{Ec)+
zSfe<uS+s9>Rs=e<(>5{Xe5cUlfA#CMf)#j*5^#8|>TY#FwGm|bdy2TTvc_=a5<42i
z)uhfVTGcOYHD)o)N|or`ipEbC>I-uw^^R3h(-^L(y$%|XT+F*`<1Ok%#jp#GHXQ!|
zI+i~dY0?t12_hb>GgWgxX8IDvtmyZI#<NH69Buw|zhKdBcPw$*xjl9fng#DfgZ>dM
zD1)^fSoaUou=PC}T`J_M<|F>C@-ld-lF-_e+m)H6;ppzACvD5e>GI~gSd#WTa7LoU
zk=`*}6WsMF>}zRIT$^VX6PZCgWC2)ro*1|{i6TgacZ1K?ttQeil+-07c1=1@m3OX9
zbNZi$rfaull`zLAHMFAaaMP-jFfKeJbkeZ%qro{HYSo8@E@NQLG-vayoEg#WqHe{Z
zOaT4jRKrA@{(2=+gEDndG<rqMXDcvM>`hyW8|d&1sK9#FD6~YU8FVe6Vn2DbA8K*5
zibcy9$oH*MIppoJ3V3oY&pLpzw@!Ei(x$YsYgt&a2aIu1DErqmjZ4PGThax?RT*1@
zT(aEmr|_dL@v^45%b5aC>ywJOZge6mbwRnN+N;=OlUi1@?<N(L$1`@%L0zzCseyK8
ztj)5~A!f<$Yc5D9n6x(PPV>}s0=Ax#JyhcuIZI=w(KJ}J*i^`O$is6J&{VpXq#{T1
zR5K0-psw0ca#lV!6E|M7k>_3`yOvi`Ba+$4_OD3r;aPMWXI9QsWc!NZs;wq-*H(%#
zJ|SAXw$^hGxQ;`9JwKIs{nyPZp!!z~{v%2n(?nXa++lzooYzI+Z6K~Db?efHu;g{H
zN4Dz&wQWSa3e2c%OO7fwT4<&)(p94dgki;b{-=1JUadahrdd3dC%y+Yq)2A3fNeaj
zB+5rJHgoM=f)AZXy=OgHqS)v!4lJMj+U0H=0M%98#|!YY67jMPVO+l^ut#@8M@YPu
z8yNzOg8hFAZ1efel1)!jimTMm@uK3}N^#J@f0?dCWT^J7<FV5Zg$*>^)f78+J*q{1
zMNNQlkZd&_w4AzUtAmfusUle;921&}k7@xClT9Xpimg{5lZwu0(kE5mdx;^{oW@S=
zhZwHabv;q!AS2?q<tsAzF78P8TBh8vDhIt*gHXJg$`@>pdY*S-I%?$KWz^>Tvfz7~
zSJQ9OGSW;41CdLb=oK@ZwvIz^{{Xw+mc|IK$48H3Cpa0$6ru5=s;}L$J!0J4U6{nk
zt+)4a`qnpx<(?f(!=0!}+%ev|WSnN1lp2*67P=UhQ!T^X#U!mC&%pyd>t@^$Yz$z#
zNTET;I34TPhO}YJ+wwK5qfemSjJXUjM^I|@x*|lfA<GZED|pdzr*qH6R^^%|-wBfk
z2jvF<(^b|<T(o;WLO`u0O~yB76e>3K6%2k^XBj+GE({A8lgmFgKpE*+B$nv!yqvUT
zQo(B}@^%y^J;n1;=$7%aOdB~Q`9$}}CX7TIrpu>NQB9iuRA1TtYhX6;0j9+%OM!v-
z*MU|kBdVO6z0IqJSzsCKR8}_<eB7_Ca$*;}RvG^6v7Mmo=BVlx-)W3w5OK|NEzKE)
zePWTs&APCb6kAE$HcvI%PEDhktxXpT{fxW5@0K%GU^-$#r_2M7THh1Rbz>YlR(B*i
zrk4=Mm@19dt)<P!nbNsqPFK0o<rFYY4aOc5lszi_vIYY+--b;{-L)u}))1qLaRFBq
z?!mTsbyAACQdUvkpd`rH72OEg9wmC0WUhi3O?lUfuJ>IueRkJ2E&lAR8vXCc=i)gh
z-lR?|#*yr!xbTdGX%2JzWL19{f5<UC<gH_|%>Mw0atJW2{+%i%=Pg!^gplD_*1nZ`
ztqMlSHvxiAdfJS+WvLQZMc<gV&@;_-IzntvbDWNAx+>;NMr|36Gv5>|``1@3k%sab
z>A2(^9`pq%&UvR!T^OrE<#*iSi(uoPv|GkxlX~{xS4BxBP1yC?aib@HBc!r`EtvB<
zxvRk=9ff{(jgp=sPt5C~bx9&EWGD74x#_a8d@MrPfJbvhZ-IP_=zLG+Z|`4f&ek1%
z#@uzTRk}JJ%;NQR{o4CgT_zmf!r$!*@0Y@I#e2u5+mYq3(x=UN$m^Ep%$Gmwe+q=x
z(jH`X5}z0P)%wr}MGyL=`c#z$K}7&4rKAFihItB#09Yi*ske#%UG>=bRkNqZ{{UT&
z^s0{Lb$${_niH&eg>DN7ytUj@t!Dk-80qjT=GF^0FgYK5739{Jv$`Ul{{Yuw;Yjas
z6OSy66=m&LERzfD2U_|n@7-?w4%|$6WL%Xz+YDpww>6EbMJz2BmzK(M21)Ch;)J<l
zWO>rm($ZPv5iFMI#Sh5GISW@Oo#L6KkGbkbYZ{YjXcXzD^(pG0XLc~oo0xhU!nAxD
z2bK<dRua+K9kg~rT+FcBd98u9eq-ra9vZl5yhAjN)#E4i73NX+jj>PMu}8+OyH)AV
z(Oj{vqmg_lN?cO|qwr_^h5rEH2D;A?>yX*q=+U%H#?AcZ!TZBK1wMvSSDBG};zYem
zh;CzfJlVY4c4LzlsQOm*jozQA>Jm+K^1zI%v6lw~3|Bo?o7Cy08AV<ujm^cS-151S
zPm(aPlhw%MnuZkIYnzvzX)dQ!TZA>}X-RP(lNR%}V_XtzJ+*}wxVD9)jbDR~wd=wP
z$n)EGWcOFCJkK-65Ha8r%{Ep`%UHooYy@*!&OEB+($>d2soCvEWn-Lr_O2H4*U6QX
z3}UmY`JkDk<76YZnQRH_rn|onwtOmp{{V5nnXD1dQ)AEFLa{xwQBo`@pb4$N){m`q
zHbeJfqC-WoNP5&1J3AvS-1Mp!(}-X^@@qZW?^CrnK|ivbpR3r#i5h!=Jx?{-_)^9@
z{Z2b~P0%Wm;<c1+(mW~6+0E{4Yd;RHZVZxGTo_%?3f!;w);62qy+=#BGF(HwbHgV!
zbt7o#V?wf&mocGvD!_%LAwv4sIeV|j#K&#&ulKu}!kjso+fsb1LmpYAnX@C7KIWJP
zYpT{ccPD}rVp>ClU)G;B;)UB7!ivgCwtW^BtT0$!XYtGOGUQF{(t4V)X)l#B`c$*S
za&m34S&LjT9Wz?mY@^SO7x6Wmwi?{(;rUSj{{TExSFB9Y4@9mc^*S2RjoS!Y)}se(
zOZBY@Gp~+g_pF&DQV%uLPh*~Yl%)vSt2Xi|iy5t11anBGNX=8XbxBB~!e`K*3J5+1
zfRFj@C-bh-v|ADtnXBmwsj7VY8BZr2MQci<r!^rpbXK}(tgVHt0|<j0Rur$ClS-_c
z+|oKBZIfylo4-otwVVF{KH%W|i@^7+-o(@Qs$AndOLoEqv`p@k=40*Jx!833n}*vI
zF6YoLYqFGW1YF;|Ygza!S-WK|YVe^ydB!WyJT>6EEf@);9(Be(@f;7OaN_IR+=<tQ
zrGI^$WN1zmMOsKz23wO{l&Rg_j}cpya;?V8bg9DfYNN{XADGufpF@ffwycTzR1yv<
zOS#6|U98&nqMP%Ap?-rk#p`ze0A|(ak2%Y3{Q0ift0Z#Y_l(-`rK8+i&gkW2RvS-j
zVCK6GJ`2FulrGj8$UN2+H#({8YBV@dM<i8gA803`^z^KajvK^sV)Xw2yIj;DJ4}5+
zu88b%PvRnliQGK}MJFWvDuEU8&U1>9UDV|rl=_2;Wf~e?{oGfvZMb|hfO6T-HP`r}
zAlgR#xn*&K=|@93X*5IOf_h`NE4G0?MbB#Bn)i=Ro6x$qj{+4yxc3zlTAa^zh(chW
zN}hI_LzM_PxiSqyT9)5Zwuj`y=e1_*dezOn#CrsPI^^V6OPjMi$kJTM$GO+<#>Z8Q
z2ZQqe0QKtCmb-f7Wn=snvG$JVN3e;n7Oix^WGCfTS5vwE>V8$Cn=|$?AN{3q=%1A*
z_NB+7epGX@I#C!kdD+U1oO@MhHLIK3(y6n_h~S)cHGIt@KV=)8HlKMck&@Wp`c-R-
z`IdJ8F9*`1Q+%jVr5<*Q6H3!LcE(SoRhh6W2JUu6Sy;=vd?~JG_5T2QpL(d8p)O`y
zMGg-6h{}%2DKzHu@6v6pyFEQ?sxMR1!F)$;9-7k-_Xh*}qpewE1-{h*Mmi|#T@Brr
zTCA%+NbH3-##AeubI?^Q@*}p-f97l2q^%TqRU^qZjt5nLI`bWH)#<(;RfogL67hun
zO?frGCUe*LP|p{@vv_sn&-#o%*Zk(Z%^}R9X8WLKxGH=_Vg3S5Cry1*P&X1U$PWU&
z0?N=s7*!YmRI;AssN==Y)K;uZ&{UTaMod&>QYn$Lcc#{fnKVam{#*|K0Nqk+%Qbfr
z={D-4GF;p{55(57F=wLamWySk!3zHX3ISU7pgq)Td$nV`G<rna)Q;N;KMLfoPC=`y
zxsDs*G4A>r=kA;4HHte;4G^nY$+-UjAM~b8gCX0%%}-N`uH`QkhHW=bW^MwIKAEmr
zd0N^Z3Dpe}81&AWt1L|ZRE4M$6h9QU<2MKDDv>OnQ-W#+IW#d5m-VF01Gr|aK-sM1
zv^0s=_z}R@2m6M-8D`peO7m-JBc>Y-ySOcG*j#Z~nq#xh_~!<pJs3-LZpS>xFaV{N
zTZtiIzywsL%1F(#vli-Imn__xzhyW2&Upu=X-iszuB}Ox?q|0e!risN8wbr9>zsD1
z%NwZU*38ndRdytwOxHx(O?sTLjGQ*<V;b@lhWX5g_i{1k@vGK%Ycfikgn7VBp2EFc
zB$TKn*|U0)yi@F2nsDyhTR#4@(>BRr3Q1kRI2G9nt`gkyt5WBRO&8H&Hzy60+rjsz
z;RKmy+zRc_rD-~jeN1Un<Xa3c9CqxxgRwkgwN|*4HN0vMAe@R4vAk~Bw-<=Y(6i<<
z7#Rbhs#5r;PSD=kF&fO=WrtCowa<uYK^i)8_I5WdZ|t=#K4@0xC@;Afj?}htsc|0Q
z`%VK6_2XAc%EtJ2E40zEcDrrC2aMGrc0%K*>0CIa?;GKDfcYnpV#H%CYZ~uRzD>vh
zBpwJg$xSD@q}wp>j-xUb<SscCooyAthDK(>cg=T0mf9SxJDbyOT@PwU=Sk&^0C+rC
zGbEa!8%CU<dskfG<kLB+R8dV0kL@hO?HXL1cSBBVUobPKn=<3o`qpZclew#_292Eq
zP1LTUmT4`_5TN-~_deCL1QW(^Vmwz?8b0$zzS=4qmoq3|^-N7ykuKyv>WCHY(u`xP
zJWN%0Eg~>kTpmnjri#j7bLE=V%~<p8cc|OjY5cMlGhSP+>3WRXq{VHVuI@OlEK|G^
zJ1~}0ZIt2B^&I~I$E2T?P5q6iE`QNqucdj7JD#GPZMoF=YTjF2Fvl_!ZRdkfc)y)7
zKi{nu$35>*<EA@d3Oe;P)4~DV)f*7V*+cGXYj!G%%5X&saAxkGe6u#(VTb{<-n$(%
z1&n*wt43-_#jQ;lbFl5}+NmYf?lyp^;Pa81>2muhbfCAACI<a~fm`WQtXEQuc}#?L
zr3ojbi>VrMw4luCqH`!x#~|`E*0${++Nm0h6IY9pg)_Gl+k|3m?QfMwK~CEKwfWTB
zl$$$Jvauewk)lb@@K%3^BLcyGmC;+^M>P+Ln)?N3>f`+sisp{!>XyeLtYj$qRecba
zzlnd^uIYR%<d4L9G-QNdrAiKKfgOz>G~E9HuKX$!UqpFP*h+jq>sRYQGoy$7QvE6Q
znUGT&Ra#&`_i7JXmI=g^Ff0m+ShvyvV~=3PKWzXm+UvYCmC|xJB0@*$PCJ>@_)JM=
z)hD{tW=3X5IGup%J5uSRQM|pkXbV5tQGh*_Nam_Haz(*KG;Lc+EX1T|7!|8)8zfr~
zP%=lQeHJ0Kn>sP_S4fmf(E?Q98mlwOD11H{hBJ!uRG)K+K^86UWsU`mIKUX-Q%&5l
z0ui|D$jwV~lk+MtiYVMl@`rahW8Sk2@x_57Z9b;5sO=YS$5k|wIqh>&`%8V61BjGl
zpYIw^3y=7mzqCjBir}QI(G}g(DR{Za)^XHq70)nIPSMS7_c4#ch}aN$pmS6+r0{XZ
zvH<?!u9H`mJB@P6%lkUoX(dph<Sn1PeGN6Sl)f`QShUe~sp`?%#MZVXLmWGf8P9*`
zT_%@jX@3QtppdF8k+FCaI4r;Zu~}4#W{{MYrRBGV>1^%<A~3;>p)-sFP{}lC%-{&q
zJbO_>n!hom=IzR6&m{Nba*ek8_Jf|aeIS<B7}hxd0M>FF+Pyk(yEz<!;>u{SBqYfF
z2X1Rf*hvMrETd@SgId>%<yJMNG>rTGHgg+D5H``s2DqF32H~{`?PE>S+~a7^U&gs-
z2`D~ecQoQ)O9_TqE!>QQgM*%Hwa_xY7F9ppVfvcpiOo}EfMcBfYCK@lG!rytfGk>n
ztsh$Itp4Drkw~*Z891nW8q$NQQAKqdGv&DGwa@AknWri`EhsHdTNhC%O|GurBrU)0
zA}`gSfnJl~t0QQUvya}&rJLK(R?t^>K7O;bY-now<>sHRM{gP<TkPEXhd+gA!{Z$(
zRYlW_cYXr7{<J!t{{Sf}gZ6yrZlk7Yej>LF>W+kbuNnUU3gmS^hcYH#PPq;TNaP%!
z#<b;K45L-2X&kNQhp5;{R_@*6P;tDEm-DKP<tYfGa}K1`zQ=t&US4~F1&=){ITM3Q
z!BV8EOPW9<-D)w{Cp8F+=Ofm!mg7W<ha}dPj}dE?Zot-ae(No6b@7}4dw=z+wc-8h
zob+1ZkHmC6h~|)eoO+TejBy-exvfYkUkpS^<BH{FxMY!#_O6;c9Q8LU8>Y=UZ&_=j
zl*;2??V6$!Fgd4TA4dEee>36ryMNPpfAlKu>3~!hQ+ZJno@vEhfy>@bWRT?5Aii{=
zJ9nv@yo-~vAl7e1wYQds48sE*g>zbUlj_=}R*RpVxjuuvJ=joIi)UqV;Li(bxwzDE
zBN4{)$@&UdFDxy<T`Nn91sn-Vjo(^TtI9DpqeazsdkV&Dhu<?7ZrQ65+Mch#^shb<
zb9ZN5YKoL>i5_U}*B_v&a^19t8QQ*;)ZMg4pA_5Xkr-^!gPIX-jc-o1TH;lL86|!d
z6q7**m}nPEFNo!0+&FA|^Huy=0gG4jGsLP7^seKZ?VVSF?N|E*$T??opXpshN~mKW
z#8)4cb5C=wp2dUGsolh^g!ji<$k_9L5wZJQf^tIEC6C@G*w;jMIcM<@hQy_M5^6W)
zo7$)m4|-8dWt$K9{e%9x59eNqtHtMOHwYJT*l;N5WmDos4+^hL_0KiisKScklGOBK
z^dp(DDz&L5`8}&Ud&t?#+-6^hr1JHvX-`DK{b{}=09(N@9EZU^*sXdqhpbGsK`cdm
zc@=X{w>hkigK8Lf995-{iq#_)BPIn`l(sOvYSc{IGw&pOOF*iBeLHOgcB{H|y4ymi
zI^&vC&~eo0w8Q(B`qfL$E??Z{qF)LkKg4BvuRlXs9SF$2Y$2|4;dhqqYT{^0x+T^1
z%uSCj7EWB(LuF@Up~vPH8O98c)YnuMwJsH0rR?HGb>fC*WwsbnF}(MxCi-AU%uKQ#
zThggJ$oi^SNYbUQ?Ap`G^2?r2-2>FqHD}l#pOz&2&3zO!g7iK&68`{t&N2k^@6@s2
ze8#;Zz?gLN89W3Zop`lgr!8HPc6)g5^f_RYC0aAP@aDXaR=$oRmWtR1h9k9dNnN1B
zyD1&lhwNjH);*&jj<vn8D)1{NYKEx!!L1mN0;QJ{Zla@M%_LhwVxwghqB4=0u1+oV
zo0$(}HRrHfi;Z7MwO)xBbNuU7v5VaFrC}Un*j1e|yvxXl{`F+HDqjmXHE;O`N%(&{
z<ZcgIy4#rHz7rz!{HD3vu3XkA?JntU{)pgrZfbU1E=?|{6<x~SERX&qnrn0gTlv=>
zr;65}!gVc>#;E;j%0o{|Z%$|iDD6{Pb2B%g6+|)fFcMTUV}hhpVj?}wH#7|f?2meq
zC)#T{7LK#Q7jZ?eA$AJ84>jnLYno(#dgJmH=T?hzH;j@+Is8MVI+5nF9@S0_dq}e+
z#*W!OwF(j&n%ui@;)rhM5nEZavG?g(##VI$XkT1Z$~Ja3rSEE2xRc9M>rrXVzs)}4
zx1}3O;fQ~`Wedp5-%6-5#k%@&RBixLM{!O_h(C1{t|#;2FvvQ6)~Ku<SU0F_8t=f#
zd(fAurSvqTKQIS@Q;7nw<0IC+c+!uv=6jxXR{5-pF-jCvhGl<aTB->yWMriIxmg$S
zpL&i(>)NKq@!X|!`yqDp99Iu_BTuXdw*W9Y_pdV*B<fjoM@05W@9i!A&1Qgg+mlx`
z!p$iF;OC0-tLTj&bhnbSTMfgwtfo6o;nO+Aabjif6@^~mDrz2K+7mR{L_zWgCb=o@
zYf)Jly0z84rKvIy3vdUmaGI8^(`Fc$h{x+))T7kooQ->L6WGOT6ii!ks0WIbyj7<s
z1)D!g^c2zOx>C5o;;lXVz6Z5l()FuNQt4rje$mTu&`~OiD&ZOuld;vlt7{Ys8ssSU
zs)?)HG=~KKRTz4zQ&u`BhnkBl4Nlzp{#7FCP#<5)z4ZrU#;F#V@9h{p{{Sj->M;BM
zRGb-$BA-%#4*vj^a5|rg^yn|8jmz#<!Q#2At0Z*LgLknEUM|x~AQ$|qX?$0wFL&qK
zxZIV_#Y$EpZx!j=);zvg`cphwf8tje?YvbdbS_sJE1XlenI$PXIqWDlE4erW)~*%u
z;U7w988xGMcsb^~eH&0&pjFQxes#4v%A2$5Tw%;@i87$odDEXTay=`*n)(`3T&$AX
z`b6Y%Jq=!yTA9$OjZm=0Yio1vXDl>%_bOP!zGE|T7r&{lgHN~4Wj}>^`NeezHS#*8
z`^-Y}FM-WTn)uVSr{V*R7EkwTwuGB(6X{(wz7%rd{wW$n)g4LHAP>;jJa<Q1*ygp!
zBYSenc>=P0JTs}s>}#(6>p5Cn$n=E!qv=usTxjWM%{M>me+q=x(KE`9!Ww)(>sRYQ
zGopz8>3)>0AW@T0WL1{+S2#67Tf*ev(?}Uy$i(%l+H8cr38wZDuqo983Q<4{U3Z2-
zZ6^bgWNdzwaoo<Q!c6NOrjx2zuQDomlArd8>G@)_yfdiI1h$}o5<=~OI-GZ^PnJr=
zEt@B`+a<Ugf-!B`&sx#9l0`BrDQ6^;j?3D<g0gO(snM9dV{$<%$ejrGHD2xl*DG$j
zu-kba{cDbJ<VyRTbmZQ{wCh5b62k1lG$#a}YDJAr%(2QF5ve!_8LoLPPKZg`#Ghk=
zQx@IAIRVHgk&49C^zA--XfB!MjxV%%QjF(5gjQ9p7~8pOa!A3w(9=)1x|ZpjIG4(R
z0f5-*eweDB7CF}E9+98sD~<}%CQ{~uWiJ<kb(8+!tOFU(rD*pQeiSu_uQbr7G{$sZ
z4j9%>f4FO^zQwK8)~UZ^TG<~jkK$rLs#_S(;yG&-YmW!&%or9+nB-yHj1m27rqkxS
zi^IBHqC&G7UDS@daG)R3qG`R2VQy`ItqzV0jI43pO3q|b&OpHkImK4gtmC@1jo~CL
zfbe@)Un+{RA82e4TfwHRcLDshmx8ChJJyp+8k=t~B{uL+U@NkWZnQC#(+7@8)h=9y
z!{Z?5in%e4-6h1EnZ`EbrA8)7_Uvg(RFPi!;<lY-^A1`pY!%N>CN!J7#EFgNni!&C
z=-vMS4AwH%EY{z<x#xFC!cfQW9;e>BT`2<iZ86Zs)yWe&n;v-TRBr5QDn^20fTZ*k
z4Hj(BiAG0y>a54^#b+(XixlNM6t!n9k3%m@+jBmNvljOIWRk3BgTNJ<id~1jYZYX2
z)vD%{dtdQ3{5_|gM^p~q>Mi*>924nYjiKGcCZ{#JFfR%M*zKOxtQC>LRo$G=kM)MM
z)+3HK3R*r+D~Y(hn&NN}nVq_tmlT_3(35aT+O~a5OSSu4S293)0iLzFr+i1dgoM7g
zS=WFW4iDo*WpXD?Nb9UTPonDT5KMuO8zmSu2BF}+A5lJJQTY#p^G0|-TGvK%-c;Gj
z>RtoXtkk`;Ey@p08}hDV?$+Yev~$Aoxat^Up{pG*nvork&Ll#4>?z9_1ZVrz%{|VT
z>`NRwl0rLHzL5w}G5jN%#(Nq^b9OPo9@R@xe=2dFz*j7OBWJNn;$l`YpL&o<mKp4|
zqacG)$+pND$gW24f20n2oc?vwK*Fg_{`sa?S%^_UA4hx^{{WJAZ~p*a68`|`Ro_wp
zGgz9X)>x=K3b^TZRCh3CeB{&63ww5v_#|Z3YjSG6EOI)n;htx5c<eo^GRs}K(ytk<
zG3Pq%3`gDf6qDBHRVXJL9R`^V&Gf%%zLDi~&&+dOJ-x2rINB?gcd0X~bt6tC)mY(N
z@IB2{p6rl9a6M~HBCL)Sokz_gi?Jk<b5R$?Bsnlia%#$)in0W1%Gadw^jpGPAG%@s
zP`3toMuOoryLMxNzna?dmA}sg$EYb?r!)AG)%Z%;t@Mk2!Qu^%dfWi)4~{{uN$z$-
zVpGV?SGfC(J*zWIo>Sr(iNBNa6`f{&R<5Y*VJ*m9V(gyP32d`f2xTrP8D_-)0H<63
z0M|wSb?PB^sa$6wgBrV|8$kvOw}plZv98HZP-3|yx!)a%^YUr7AMyC#?iz%>;->w^
zVR)KN+KA)s0ah=c?Ya?@XWvl3dQ~o_5>`-VO`s|Qx8Ym5TA^{c@@ph?J&nbHcBF8z
zj%#?Mnng)v1bS6=eB^W$x<(C^ucGsQ%>9w7aVT2Cu7=djjTUh%FN4vjy&vgK^c9=<
z5i$NND>19KL^Z1n41>=UQ{1Y*3CZ0-HO#FXNLc6cs}?#0Fffwfi67Ijs;3^MRH@3W
zw<&m=#1QGQpn(LkDaHpuT!gx;_wDD*<-ghND``Cr+`1gI=6jk}$}VGFgOgftTbPQC
zyPWqusG{#AkFb>QNa&(b1*%E?-_+JOh@~;ugwM|*ucnri+4=QH%{I&{Xc{>qj1S&Q
z!n-T!OcGhHP7C3&-;OK6tGP2*X&bt9h}N$TGIDo)4R~(4g>~!1+Egse06v+hv4tmT
zG<Ti=)qc}#)>5`)h<v`{yJt)a=bE!RV)QE(!`1jX0<&hZ+_IvP*A*z$q82!)*{f7@
z$me`rHq8!UoNmDVYnAZJEzgPCKyt9hw0|nqd@RSQ*y<O4Tv$`d;<kJ`yRIY8P`Rw5
z@gk~x4BXYuG}(U-=Uj!uU>dsHnBjf{8uh+YST|kk6n2-iSc_Y^?YgW_kT!}soOO3A
zc&0+$AF%~a5RsbV!N=0IrSP2z=!7H&=clbl#^cWwY_v4<_)-pO4TU-eVIu+TYK++=
z+1y9HLROm)fOe+5ZRx?OWh5X&PE;bLmTh=y2`)8cjvu^1uF-rYeE$G>lzuhCTB3~2
zVHYGMz6`jI26mir%~R7nH}>G|B&8KjG6!18Rnqn?Qf;Jlnr4><ofnr1@_MoAYfWP%
zPSKnSx>0j?XH43JY-iqdKBA-1j^i;twbRRKqn@+<+bZgJ4FvP7V+5YRg=912bLq`y
z^>zq+DV5}b%6ZN<r|^$j%z}2czAWE$&&};xQj?`oOr-8^23b&d2CPI?)Mp_2iud6p
zwmIrT@+?IVKYBt){3+%vzXafW*FtW_ZYeR$cAV5~bIoNrs4JR5SjyF;#2ZIeHOAUP
z*Ip!E4i)*Y607%{DeGg=ucrGxpYBElRM9SEZLD}@Y*!swsYKo?%<FYoq-#jRcOx|o
zykX(>W$G(~7X9Mzzr!a>H}bb21e&dIj}q>XZ38Eo%8=@6&o0+Bx74qslXfFhwlEK2
z-m#io=bYEC7};3ile-0e{*;`%RoIeHqi;P0cD@?27J6^m7=H48euBB{zjnt?9ygWC
z@=cu7FgNc_EHkvx(NSJ3&_kM?O?_M?bbRe-YCfb&rg6y;Ou5Z@Z;Wh>wyx3v^2#%x
zdf~*qj_f>b6B^g%a<y(xn*a{==Sy?CI}o-7<bUr9?0jC>(W5`@3ep=GRQQp_qL7=<
zR_RV`EPQ7stf7%7nKFS^XXTbW^y^fXguTgxY-1Zb`c>^d>Ptre<X1grbx}I29Zgsb
z&5&zWIRkV#71f2W%(-Q`tfAEs3w3jm@}FvR+Z=9VoL9R|s~o(~oRZ>C?EN`T0e2x(
zdXrrpf|a#EdgGe;Jm#-8OC6YXtfX5DM%wVstCV~~KjKi}`U+hoHrU2L*%i@WXq?nH
zk#MbJ>)$L(^{ip>qj^7doTjC7x8*%CN#Vs)s-YFvHE6S*F7ijAr~nF(5nM>=XUx|>
z>%R(s*U_G0*eUUUtzWGmX6$YIq4cWPlN5-NQe<NpS<fbgJl2GgioK_CB}3R|(CJH1
zF=(I%S7YHxe4ROD+8b<)aBAVXonM4V_1jiWRLNy-jz>&?tU2#ko*$klZtfaJ2pR_i
zKDeR1Q7SDaPo+$<M7!G?I@Zmk&3FpKDyt*q@^Qs`@vPMt$#q9wEk>m%-4n4>D-=0g
z@l#E7VxBU$8&GY|O>;sMn{u(mE>x68{iYEN@&@w+uZ$_iF-j(n%`*&q@f?QdpGwYF
zy^4%p<aYPy`jlU2DzU1Oj(~Ao$BA`GZdGl@$l#612^hd9*1X#FT%Rvvr`g$uh%Q$0
z8v?Qi5!!&>%~<{ujn1_@5y_1I05MSqd6>ysnOnusNOgi&`+~6inY)bEjp``;A{$8w
z=}2u|Mo6ON8y*njS|j~~TD~&WCxs557+)%QgE_}M6W*<CWYy+!dS<P0q|Wj{5re5m
zNd{Gip%v8G-rs5}vq+I6frU^o5l27;Z9+GZqK>Vai*<g}i+~Wa@r-asrEA#V$#lXg
z(PL=F0;s^P<lEHAFL-{T3tZa>bA=7K;B};o&af;Cjk|XA?ORZ)qX*on#_YLb&2IsK
zHdLJ8)VDHkC5hzqt~tu=(lNHB3z^<|cQ!Zyc&>X?y`J4JEQa4Iav8YiB8j%sX09vQ
z*yQei)gvC`HQMOQyTZxrjr`4Ej#`@>ACO+%xaW5jU9Du}3sPt|?6MBkJT}ZatxTSU
zM{?9~m2t&xS_~kp(=C&~-6VCzQ`7Eex^|4N-MkY{*EFSB(&kI5v4JT607;W4KQXBx
zR$>)KS-2IqLzwPuVUIx2OTGiD?tXf;JK>{Ab$_YdN=a!p<das8X=@hDUx^X4y3}~;
z6pz-jG%4Y+E@Xg8_{b-qtmdwbsqBjyh1JdL8RkDHMN!aI0^3j0lr}o6(R8Gf>aIB$
zjzJ*$*3OsWJy%R+5ZyUIKh?R){#Bx9Gm4V4I?X%cEKP*eZP_qB{Cwp7E1*lQ4_|<;
z=q?oURlvxqz0PS<aCb&M*Teli=QCMK*6eah^<Ss8a@wbbHGMrd7zC&ZC89aN_pE9|
zTcdhameMk<_r|UrN|V`!D{Dw!^k#15tCFV2bWKR43>asw0Ic11KrUHFWEIYrylD0_
z<&zP|o}p@ZUm*VgyHvzJ?B>WGym4HO@aX=HgZNiX0~)4#9I-T&)?gG<1L*I9-||lF
z{{ZXaU;PTZ>X0hk%prv1f@_z%oU@*4-jFUN5Xa`pllNA$_3Pk^_H{g;y;-e{-m|gB
z-&}=OB=oBu8_?#yklo0K41Qep6&sz3e37~6RyGmaMl;DDl}|0wuL;SeWfXGrI+7|p
zjOQ%X8#Ze@9T!Pi5j8~ksz}AMySCmztCsSIL5gfi>Pp7BUmjj9wt&%-^Gv};KRR&u
zkxygHZ4TR*n>fRce;VxkRS%!0!!oxB;Qe!5hcoz*+Gw%u(ydqfv4K|O8Q!O{u1Rio
zM_~cmls|=6xg}y_-xUQOORqNT>5uzHsoLZt5nVCZ&F(T^?|Ri<tZIo96yS3}%N9fZ
zIuH8jzs|i*AN7ktg<a8^G-lgfW^Ol}*FgloD6U7j-5-Sxl;l&i)Xl@cqT-VHhd1ss
zoAD6Nr7z+9t7k;DH#!&ubV8%>p>D~V_ae->tyVvmp&jZ-NlXxV6_Pq6Yg_6L(T-`&
zBRS6^y5NpUkh(8kbM>enARVpGy>4?6i5gpXLOx~Wd(~O)wxDeAYo4;u-jg>xBFy&l
zvwzP{KPta{{{SCH=qm^CAxHSEsOGKd*=S@Q^i%kXtMHX?Zf!K^&TyyDbv4Rg>sJ?2
zZA0b9=sKFw%Gx1Phco784!Oh%AEU75v8{H-#yLLe?^SINPYmrdcG@YeinX)uQ(Jbj
z&m@326<B*Zi27&z7HR#UJh9uU`A}s2Yb#QbZ#4#Q`DG9EucfQ-9}7JsR?}O})?juV
zd7KKh<1JC_TG?S7hTosYyehki{AlVl3E9!}{Ad`T^YyPb@y)Bj;+uzL+OmMkeR!qN
z<cG$}{ujPFezgR-!j{_KLte_?G$DsVD=LdqrUq>7wX<3_kGeHA%_F8I<Q?f;aZ?mr
znNCxRt1PO3vm62ITO*yGbK>t3uAs)o)aErmJNln$&G4oA+Uo&Zk1=tY-<Hua>UFOq
zm)e3vpLI@KHQM-ncHOVC8t0G1i>drJYwH|ovi={=xXY&NS6g!|FTi77Hgj0F2PU#d
zVl*^nA}JIeGgDkJ&IK-~C0)$_025wIHiv0$$K;HVIIc%`%0ca0Q_&koWJb+Q{A$@~
zcWlyz3PC*kQYnoGb9z*^{7mQ3b4-F;fs$K+%~N*g6vTuKL8UHRsXo;Opxg&s(dKp(
zF`XxYoBk?0e|q*1MR}N&@0-%thU!&l%zblJV1Z*8RRDSl!YQ|7whGGS$YLzPMP6o(
z)*PNct$NfUWO>-Bxjs~L*OCb&V~hf6+ENDdcKV9BC2>@5*_7|#FLy3^8UB@;_M}P<
zK2mFmJIviLaSgkPq)vDPrD5GfyRbPmyq1I}OJQ`%0=GS}TM^*kt##pX5lfPN3!Ge)
z(P}kveZ8vE#@|6+q^m(K8P1^ZsT`mJMOTsbj!kh&Zcgmcn7V_RA3c|)VQ6zky7$gU
zE#9~*^(s4~(B^~fP;CmVRCFV)X2TzoakNLVvw}UVg)emImgi4$PuRiqVMVNn2L6Yk
z70HYE!wdW}{8uQk>|RLdYnt)Ks4X-pvc+YTY)l?`?^)3IrjY7;sk6w-78%ZIqP@e<
zBn!qrI*|YZrGaxyv~5FBiUs?*$@*8II_!)pi~ukIt}H|Mk~{Fx_gg8r;M8b#wrSCJ
zX_`%FU50JP6>>JOqr=acT=DTRu`oHQqFmQZq^x<|j^my_E0ghtlDE2hFaSJ!sjoJ?
z&q|CmJlfb5o1VQY+(&eyJu8+)#DZd{QC+<^itPMV8J9_g2TkKOwKaH{Rrt}zIaYix
z-ChMo#AK7ubgFxlShR>FZ99Gx=#aY?J?mE+MH#2e%PBS(ZA_6@SVuHjBMpw#$5&Q%
z!Zsm>YpY@a1(v!UH&);*psDw)HH%V?=R{q!XGv#s6u@DtMI(;^%VZB)_i*@$PM*7*
z_2Q!~4u0!Vi&pU+wX7^!D^|;6p=$4-A!1*pe9mJl)kj7j;bhzBFJ5)c>}L3h{{Z4t
zeewM2O*ye=AN!)ZEA0~uKJqOWtX+BXs6Fc{Um7=>v{}b%2ltg@!vi{!`&VU8g`Bkh
zBzg#Afl6Au^R~y$mp|*j3V_$r9!z!;of-Y$pc$?t#A2#mIcSX}WM-{6b*&jFnE`GE
zYFI>6<koT-s5(-Lfr~`|D6YT5fAR?b0D-k!cQdK*oVDNleog-Q{&f#ZhG@1d*j)Xd
za8KQC4RlK^F(~<0p8mD&<<(Q7;`)B3&hw+o*sgDOsVLw8Pfp^JJ9oHZ&dkf%iRPxI
zX+0LkF{c>o)VC7(BVzs0(wRNV#;2O$uRciMskJ!jWl44B6C*YdcQwo3>VIm!3<(jV
z00*sku#Mf=>Pp%gDa>0(de-o@vaYD-9b-StR*=!0^*#|iOKh#F=O4l=BkYUmRQD8D
zx`q29`c-@AmK$bleJDe*7lZlNKB#L)#u4w-V#oaQ$LecthEL--`>4cnWOQMg>22lx
z#6DKw8A09s>gmJvHjd2dEQ&{N8L{(l52anRi4laK#(USL3M-M^!h77&Odi!7*o^0=
zHRr3^!{%2aM_)5`Fmcw2X2P7B&Q10<ibNPEpT@a8MVd`oUP%D-{3|)#M$Fc4+%d1d
zSmIn`<*w7i&AY<U`-UH_LOAMdauFa#+TC-_XbUxE$Q6?|LrsYEr50dzsTv$DIq6nn
zO`?koj`Cv+UoYM#0;t|ywVWb2$P!3CVmnnLS&TX7*19hV>89rF-AKQ?i|*ugYSkIZ
z*%J7>QS+`A#y`6m%l&Jp_-7I>k#i^?F`H@pg;LzgYShMS8YDOR)5&wlDt3zIj-CCL
zaV!lQ%M^o#HJwAV0Y%oU7<V)EH}c`~RAHN-^c9%Ef~rqFO-wdDEaFMxIaftubQ!5f
z-uCHQ9zdmOQRIn(#_Wtb5OInC)YJTNt!c6al3%q%IV&e5{x#QW9}%>>^08ebTS!MF
zjIMns=I&!UH`XKkaiyfXV84JwzF#l$;IQq@b2?<DEXwrxb6mKEW_015;##$R)1c3M
zn$6WF^KV~oBDmwZ*2<SmKwm&AXdL|Bn`)UQiQ6_q>s+>=P8fFFTl21p231VTE408U
zpbw+I3xCNwxBmdIiGTDe?^xgyNDK{D?<9bDtTtlp>P=}^M3JJ=6U`#oS{6M({LOKf
z(#dl(Z3W$rL0CS<RP}C5lTGl)+a?cr%wXi8J$|*=#2~YXM;d?>aH=V-5i^sUvJs|v
zI-CzdO*Na;);3DR38>k?rD3@dynqh1PJ5L=a%z$@Y|58X`S85e?KbqW+QS*?RO6tc
z^)-CVit}%Xm%etV6{yG%r#^(5RK6o=>~cy-#d-tV{kuZ9RUqL}kKtWUW13r|veS?)
zswYARN|nc#C}a0e2DsbQ>{||W47fcjCjLodf=v9%s(GY7r<Cg~TH%=WB-9pvaMwcE
zhq&9WIP3*gRpOX{`%<XI09dmB0FUTD>!SYvI`oE>lT)-%*jC+<p1zdKYHo>jsJ7h4
zH(kI1Q`x>_89di4w>s&&#TJ<SwQoxwEZ<Q|JC3Ko&Qn-_I`&up0BaS{cusBFb4Ge2
zRK@$nRA0L)-Nl*YLbxHgCaaiA^WKDw8_=H5f>*FLcDiA)%TSH#tgVXF=Vc!}!E$m=
z(^NGHHmr-=s@7Fi(XBns%fb+{X$sHxnB@LdZ%|+Kl6{48%iu$g@mZfCs7?Hq@H}zw
zSpDxxgzc%bk213`Z)9m1nb@&CNvx&>$^aQYm2uqHWJj+?C$>vy@Mer|1J~NHEhbSL
zqer-op!TkbEe$a8{o$#nd9q*$HjMBqekW)U<mHcgic;mNbLmsPCW&pN`DK?sGaL%1
zaLWv41vz*5oY&Gx-kr~kt^7ta!|t<dJPr{~->p*dwBBv}aQBUQ{A<Rkxt0C%tMGl%
zlG@Egn~N^THIMN!)y3tV*YFfE{yEJHW1DJ+85UMdj@*j&T^96TXjb!gC;|K`CXR&2
zZqHiK(+>EDu&zm+aAv5+Dk-K^1hD`P=7fSzLk9l<#9D2rV<p^q9hBz0_ryLipHtm$
zYnWi@PI+qSggUd9Qn}1$IVY_(WN4;Y6yT}P6{#roFAPLvu#V;Vc~l(NaiA#kBt5_t
z$q$UGx>9A`XyZ$l@cwngT(??vz9KNc0grTZn#r4~HIh3Kp`*lF(nkajp8}?M8$7U}
z<E!xu@jawb_+r*oQNKSkR|njjRZ)6BQ>xk`Hl}8}0A1WuNi>i!4bQz!jq^$V=k=k8
zj#+YXP?pKfF(Az?Nn#UDPzaz7qrq^>HH&$I9g2Bs^(L6VO7k%-Ih~kzJ6(byCnl^%
z5hN%XG*N}w(Nb1L^hCCdmOS@0k9~G!bI${yuVS2Rk18*kPG%kT;Ds5O;<kJ;Jl!zQ
z(|OM|$y#3hi=V(H)JXSl&N=3ymLy2bdqMQCJzdP|kHCzp)IL!owO~&fFSK%d*L1c<
zOLCpbYk4DOA5mLY7g+-}mA&I=y$y?bmvvf}RSCs+(pnQO%*kf?i6*K^A#!~yWMs;B
zU<g?IR~;}gY9d8AImoU`uimij?w*yQYF}@)Mkj2FFn#L$?;E=LiVJ70a$3<h-I^2e
zZU%TPDma)fR4z$1=i>hWh+Y@?Wtih-`McMb`2PS_`*x;n<6=@1v#%NTs==$&xj2XJ
z*yc#U3r|G_v)EUxaFP+na1ADKDn@ml4Yg4ew@2?W+C8hJahmgLyVPf?MOvCE%rRA*
z;;IaCMD~ZA)`~TIm|3L*%)~{xXile!vkPEWwJNTLFsIye$PIFSF4rNnv}<_SJkktc
zcCJZJo!H-(k|&&mG7anZm#?^~tyGr>)EZJTiBz#`)jqC2op(MeJ2Y7F!Dbb-HZZC2
zBZ<z@fk*&nP<nT#V$0xprBz4BJ9^Zvy&^&N#dJ;Sa%o3f8v1UA(ty(qqtdz?eJ&`}
z0UCkyub!`6%32=%8YwHX>S=cGjnOImDsSvdMql*{pQT``Nj=ToIK9wCqAExX8-f1l
ztnU%rUD_m8FozRPae?VxwH!kBPR~=zuT80?(5v7LH+B8YlAOF;8uXOmap_)8E?m`3
z=&L|XkxyQ0nHsW(h&j{MeMtUwJ($I^W74{7?H@yz4ZOWc&0=exc?Z2`OXEiMJ{EEs
zsNFfM-V{&t-1}EmUxl2s{v>ybIG~PegzY|Ty8i%O_zHxY`a{Q#!ceCSDFZZBvVxg%
zxrqHKBQrChsS!&Wwn{Hspa+`Kid9pAP|7YV0tGl!fa)m#wbyuP{{Y0J{sz@?+|H-M
za`%V*GH>_K^QbBrqSz~EPLN$e<j6rQ(>2qDn~h6%(9#cdUxQAB6?|llIp(env5Xu6
zU6XvPnp+1g*`IrQkcK;e6^SkHpP3!DLO(XIB6D?Rb7yF)8S>u?&CtfU+Bi;w>sT&_
z&cgCbV*<5N<wqo?%+7)U(W0X%QU^J%i^Bf^@)2+P=6{%|j^dxf#PJvWgumc5l|$Sk
z@S#Agy=qO#<&NB%7UdE6H~vGPL^ZAA!Hf7k41E4(x9DL0GnlxyGQxP}O<mBYS)%e-
z?UE7*Is3!Y714$MCA*z8qs=Z}=Wz*z(`gD@*a1^RmXI$`@vmm2mDEtG!<92lSd_;b
zM>)j|t0H{goq4qNlW$~jN$i(XI$Jp+JCa7nxuuBjBr_}HjyqOPEv-)2+Tl6!V`k3)
zR~K<3EzGiY&ws|bYB$fy&8P4qI_-JdPIj$!ULE;7B^mS${OKc(rpF&NtLH}G@Osv@
zoxqbZpS+{<sAkBuN-72vPz6h-Uoda`v5m&CE@m@K(jMlnB(b*fJ;dn@YnEeyUV*kB
z59ob^D*piAOlwqSJ#3B^X<A7hS0uBXl55pI7}+}beo3SlNhK^lDkHfyz7W^-8!LT4
zsgM~}I9zq~u3F~Z;<S6q)-m92MIfEotm<la<@legRgBv5Iawz+cP%eaP6yJU2Xw^K
zO30i=M^>z--IhnnIj82R=VFGXpk-NxGHU+-gEaG~_?Fr-0Yq;xvN#}fiifz{v|BsB
zh?nbcrrX)85Yj$yJAuV=_V2O)x#|UT;&&?wy~54By93GSIIQig?9H5>waAXPGU#Ry
z-ol-3PtEkE#95sG0HFcuYns$Sc3cnP{#DUH#-W%+C3hGF6an;i!EgB|cK-nN@h|>`
zU9I%O7zxe|IIGDZIUBo=GDFXLEaS@=PeIbIviHX)Xb@^Jz0>V2wk7ZDR8j~cZHNN@
z0C~Hb$IY1C+m#m<LI6lc)%EUbR2by-tnUB^LHNFQT8}CmTnb}KDcE-)dB?3+lII=i
zisdVd@VUXPi<_sJv78E|ghsMM0No=2^{SKjmM9o~oqWPO)`~Y5Q>M~&c<<~*$3h3p
zJuA!gtIxOUw-+0WE4eG5PSu?_+Dz6?=%bNdiQzT-PLTw8$vc7k>u0fyx*eUl3j2Cs
zVyBe@p$52_+O{eQS}&4Lotmm@IBN;{=!TjGc>bnx!!(wB@#|Zn7~9Xj!lFe58kmM2
zPqi*F(wGxjVBbOi09_aP*L$k4it55*+f`;6KGYdc;zp*i^V-2EIp-A>t8vFl=AP$8
z7w#7W`t+?WF>wRG_l*(Us=oorUyR&Jf7?IKuXs-2t<3P}g1N;+ej{3+!)Ar?DSF|*
zi;+?BkT~R4w$zK>D2_4Z$ySSw54CFpo@|Ye9crY^2U^g`&AmUxoMNcTje8_92Ep7v
zD$b_HkUb|vj7MY?3{y2RiGlSM%-6UP{wb9!RSC}qu{zP0bKa;#;yKz@=@`vv4azXD
z?qrtqxP5_oF+J*gai`vfkha^PK43Uk_xjgVo`RwAmaU^$qzmPhJ%wpX<H$QlAa|}M
zD@giEa`#Nt)SBCNPCz-Tw-@8dVjpyp^*(~V<y!Eu@^#m~&5s7eKYcMdWgLDrlj1fL
zYjZlBr~n_R70Dx>KN+dui|ys$83q|a-Ryt;RnYuMv{|%dxy}R<WDi{NLfFDuR6N8!
z-hC_4yaj#dX;Ryh%x#hMG^MezuYA}>HL0Twwh^;+VVdKX=T$eMpW-hU+UTfakN4L-
zkaQnf^Sw{Rx|XQrrnzK{^RZ3<t?9zb?8>C>?2IYy45tgz*Vdps4|?3lPSoGLzz(F;
zjP^|u4<k;3KklFv)#yv}O5Vb`Veyw+_)M{`aivZ8e>&l(J!#hXh{F5^UC%kJ%YlJe
zBdoWlMIpAxB~Y4{Oew%5(bVFrxo5;lyTOn*`RiVI1~*fqv-LYVFG4p0q9B7^0u!wi
z!G<x(sUhdg{{V865na#BX}O>P8h2U-IS{$P=QYx3z8ZU?vQ7w3?$$M`azxUcIiz>`
zW|<wLD2)Pv=qqA4>&jP?yFFC7QDGCD0h+%ILKNKHYtx|v&cx44E~b6;-u?l?ipH8o
zk@l#`Jpius?5<>^WRWBqL~r}+#SVKZ=qpmi(`&;Lghe<2a4W<1j;1304vMf*qEx+&
zzS8_2Phn6P=E3<+D&n6oDK11)mfljWnA@Lf!MKWOt^;{)es4<Xl4fd4xelXYGh0U|
zl-q&lijvmO<sw!Qw@*r&JDNgByBzZ;1{cWor%QQq!vYult94q-VY@!1HaAY(#^04F
zTaKPnepMdcL*<WglfiK5ffJv?u=On(<@~c84qUJuD+e~@TT$gV>}>9=CXHi__Uz{a
zy=pACK60ed5a9F0XLc8gTblRNVgx_!59Lzn%(5vh_!{EId}G4D-HER1i(#W&%s%p_
z7aq0Zb4=G4VVR2)mL&SuP69eP7~)>33PxC;!lXYc;=9gc*gO%ZSX_8M&RdH(WNbEB
z;|KVUdbK}-JUMj5+Ql~>>g0bTT}m-=S3J2@oM!CFto#>YXKo|6l0{L_Qon_z0PLMl
zpdz?xVAFStJvup)inP(0bK#vuK`R=t_F-9a=@-%GeXN7uZFBnyklh{e#nGv=7?8YN
zGB7^kr;M8EjQL_Kb24!CDocUM;8sqnWvXX4;tv$80NU9$>3TDci(EC+$uyg!V$6Dy
zYpRrVyEto3?#PZ3V3Tt$RE!Z**^SR8KI<GURYhYcJxYchmywM1Vf^c~@mf5ZOkohQ
z+&BX@tR2yfPl+5?*=|Yy08uCX^HN7?a9jOMpZCpKcSZQE?tuAVfBWXT9Tz~CiKmIO
zdIBp9RYw?E9hg`}%TuR{(U5<uKb20f$^FLstHkAJvZo&9m|}7K2lA_F6A|}M<wqkW
z&OOT(%F#+6VL0}y_tp?<R^DKR1=1<l>^fGFi%8_HC3PgsjSE5=L^3HEmPX646|Sg9
zV_ecwl&@i_MJ>hh&!MFXM`KHs_6~UWGj*}P)mOh2X4{At>0KC&qv&CV`(C8SrC{qv
zGN7!fd}!XM!p=`qKfX1s;U&$>A4=-0@Uxni#E#T)#V8rB6Ri2F>-~S=DidEwc?)4F
z$r&V423|ph6)Ps?#Rg?2w#%L=+XQ3MfebTa?$Ry8D*{abXvKOs8LcRsI)hNkBowrS
z71wxc{y`t`Hmi>2bv_f9y#D}?$-mz}&Y`Ggi(se2JKMhHrKwSNcQhf^)lObQx#|r^
zb>bNL^3L<y6*HAd{7mUmjkGXtH9WbJ=1l(pb}?CReKp*mZo>L!6{R@aQ<|i@7VWW>
ze=}y^z5ds=Y*^daTL-+lnH|T?v=Dax0QKssx65QIDRW(s$*o!3$jUOLpOuKlYk$J7
zuc$Fr2lr>8_Z1O~i_}$Qq?<zckN!em@EXe$J;{&4h4-vG{nAMw_P{hpa)~?}{{SJf
zKiD;;<2J?Jxb=(rn%|*=_{<XJ^NxgOv!j)*FGPc#t(<z%33ACZXv>yKo$Zy_WsR3V
zI;#wiYSW5e-mVYcApSM&)czuo+l!B0=)V)ltinK6G2HQ*dtEOi{Hmt8rw0mdJz3&W
z_o{PrLQ8cRM!^J*y!5K62!!qj1-jy$NJg9#FR9r|@^>j}*B)F9GupW&ee$s6g1}dg
zd(@Np9Tj>a3(i=#=Us+|z<5PF{H$w8?s3%E^W~7SE1rjn)`C_<hT!BC;l0I6H8lR$
z2qXy?IO&>z>hPm&Wh0)ws(Dykw++lQjK?Ck(A3JT&4ueq#azVIreJ)$@=rLdmN?|q
zt%!S_SA#UY?%IT_!tkG3kHvbY+AjXh8)i6q`<m0Cmh&oT8Z?@<$Y)X@L&^28Y0=UP
zEe;foLZCyy`p}UTUWWtXQ((H2EYcO0Oknj2F<j2KaV(bWZ)-HuA`P;a9DWrKamM;w
z$6Q4h&{HB98;vt^MGYeWKPug_@a5gg2^?%w>UpcYAs%{;$rlJYI~M2gg~BFblY4vd
zSXUPFT1sP?N|EbAl`663r1tJPQ{^h%FYfm*<6Ymv{WE2~&Z6o;bFnfzdWy|_Mim{P
z%+H8-O>^QXpbQ!~Q~{2}^{mumwNpJSlDpLDpsfU4L%4`nsq0fEwh-ic*E!HrzD!#K
z&~&J+$RPbF=q{v_9NS~i*F8CA$?mI)=%8az%{^%=w7@8)2hraJzvP|U{{Yv-zxoww
z>zc9~hJj!1@mt~IobJHKjQP?9UP-MsDQ^}D3KVxVt9(p~k7RS#+QVE-ZLC*&_i{K|
zlGfD3yQE^yc_h~)oy&xIQEf;RG@y5>Lu3ze?^WLG-mHSGmwtw-O?AQMsUcX-p6OSY
zU_Tnlxz-VKBjnbRjmBipbtS`P*pcZ`0a~SCE?H`FSix|L%obS480>2i`C@8qW&#n%
zy?PJBVdh676P>Tq(v-F)TOGuM%aou4BdsZambgyksw&9at2b9=gG{-Trqne^O&(FH
z9I!N&V+OZIZ*n_=T2`PPu#xN5n23R;rUbS_{XP%+=)cape-n@RmLqiABJ)q7jQ%8Q
z+&a&t#NgyE))e?a*ejl<=JY-iPPMJ461r<2LQPC}G^_BOmE&dD`kd1=NlaSAcx(p9
zKJ^Q8Y5WY^)?01FQfKcBRB=A$QciJO+f$iK+(I)SJ(t?7EAjYNkuZ}O&-hhoTC9^Y
z>BoADM^iqqGBB#7)^#>CiRpS`ZL}3zSNqBIu6OVRKZ;{WdeteqRtI7;F8Btr<d8=U
ztTCT@xb9a&jJUayTawOmp4EQDd4=}%z!lX$gqknkIvX^S-Vd1PJv-Kc`Qrl}M>Xe3
z>U{+#?wqEjy)Na91A)y`$p913*R2U&DW5rrU%Zjn_$V{mTuGh_Z6AeTYpvy9P65d-
zPwPWDBk?ib3b|#~<%qBgB4_GFdT)uKj`zYdvt)&Xc>1WSM?GC8j~mEg>0Lj7ufENx
z&2J-MNSso88#A-MaLCv>rtAJ8T^C7?7)bj>i-jJ8*14^D8%9=%=GQkbG>s&zUD)K3
zYMhSJaoV~}2JOJ6HBmzescfS&U&9~woC=l)wOOv%5PB%Cn?q17ox7IBa?<#U#D5HS
zz5f6pnST%GTrB4qr(5C{7vM8zJQG<LPm_wt9Y)QuSnOhl?%-vOMBy{?l6`7=oKxJ=
z*AO>?;s>z)ro7lF+{50kj?C<PFG3kdH38?^x)6W?NySi@(sNQoa9iCfSP(JwrqB#r
z4_<0mp=q2V0?d6VdJP?ZgW<rXWu137r=@n*c54!V7Zv1Jo1BkMg+5ei#+(YWiB(D^
z)Mkae0yW;<f0b3eyUtZl%0F85X~xL%ug_Bbj9ZzHtv<xZ7G1zDdV5sjTX5w@(BAB2
z#oVJiRF%#F#ags<f+l$zZg6YQr%ux}tof<g0M;zuKfVVC2ZK_st7nXvMn09&rxea7
zN~2o?>b6X$Y^qPvs7ZTbhjfyt{3}-|vo%V6hFxm53j&szKt9HuZK+taLlUkr=}Ve7
zDN>(dl4_O;+|0j~M0JfnKkL`>H4bsPqN7u=4O+wgd#~hbMAWPg{I*}pq11_Lg-6hp
zbsH8uNT=|tN#v8gii~u~sZ}D%bSf@|xXrpp<~uHam6fP!c2Hdc^8j1e(d{5Zqf@b3
z_Ws^`T*Dx87;pt!)6e=iC#S7>n5LyGqtU}f#ujHQ@dHq0w~E*t2PKX@xUW5rnOo^z
zh7vZUao2a5WN;}7=DJLeaPagZ+rSa1<NX}B=go5#z96+T1%~_v&NwEzB<zvq!Y)#m
zsn|iOx~l&Gq$_dU=B~x5x#|_ZmCYK=>XsQ>YZES`8T;{(^r=?v-N!Ar>MLbdFLRn$
zd98(LtmYhu*n5hF^j$t)e1!i1fYmrcWX82xPeQDo8MGvoEWeFmYd#XxE~03xuLMX(
zIUJh9>ao4i*t03}+DPJO_;Ier?8>Pi9rM<*E<7!*LRdDRDFdqkT^y|tt3pP6wwFR8
z9@;=xugYqzjJswrK2)eU2BxCCO)0;4ea6(08L&36&*z%#f8iw2%7r(Ga6760019-r
zBD7n23Oqrg^!r?Y_sRbN3YI-{M`j8xqyGTDPxw~1Mp36W_BM3ey*A;uEXc~I`=+;J
zyt4y|7xJze)nm<To%FEv64;PmT7S<){Hg1CYJcUT{#B2(hjVXZSFtkPTOavoKPsIr
zAR~t#m0FZp5~WdeC0U{V@cgQ1=Z`!%tv%ZqI8U(lmOSTh=8#Pv_l0swq!Zk=Mx$E`
z=SPpcDLm-^0K6%CT7ARV)a+pEH;@@pFEDRoz^k!;sb8gY;v}TgM`ju>Dcpa}jsZ1^
zt6SVjFlC8|9)hx@?JKidt#X{ErEPE~Vq$PUm8szrad71Eis`7+i!i0}BeUyD8|5{`
zb)Pm}f3ExmLtjRC>ta~pEP1P`Z?trvOjcZCoLB|T1eG3e398ahyyA-qjm<Q*14641
zL9L4$tl$$vDVP4xx$9Fa0y9uZyM=Y$8adKG?W0vkb2^_1%ia<8$-mz}&Z<ds3}Ba$
zMm_3FTNLHiU{pR{S-Wvn+g)r8u@A<plrBLwt2M%y$}Ss@2Q=xHYb#acFe8TaK9o4E
zhNR55c6QvyCahV*B(jB#3!gwUS9=n!*`;k~HO<OEtFW;kdRCm?9lC)DO|uT*aBCUW
zN!Z#Hn^!XBwBAafD<C|^+#mkEYWOYYhggw!5&NUZ;wxCYm{e9w-PYvuZ$o$5Gg;tP
zQrvL(P^a#nrDA;BOSeMgezYO6-Vx(oe|103wESk7y0>pgvHF_Zp^Ny=el8aWjDISl
ze4A#-=tU%GbUq%v7Z$=6kcnbA9_w3CODahoBm^<zJaQ}ASE)uaPgHsneDsuyjek2d
zYqZGi*NlY~B%WvmiOB~NdQ*g3eDd7$noHe7qH`mnE*?Ilp0%LFY{}z_=EY4dWN%J7
z=yE<FyaDzG_cf1Z<T?}gxno{E74F<CJIHO(GDb)pP6c*aL?t{fiRg{}Yk2N*)Y$X2
zkIT3oo2zp7P?9Y;T0t6wkYu?Wk?l)!QxzpK$-oZYgzzfewC@{cMOg=-6x$Fp{{H~Z
zMG+k21n4WEx4uJS(ryM!VEYP&G;^1C9(>IwLE59TwtYhSI3vj{*cczJQX#X@wF%<T
zG%&H?FkOf}O>p)W?WtddcEYn|nAVoNu{4u=9*1U;Sy_1yAYG#heQU0=wsGP3<CvW6
zIX_cYm64rY5=Wl+lTNj{@e~oH%vw#WAkR#99+e)o3>r<Gu&xPjloQ^tZ&OHYu5q)R
z0KTS$9AmXJL#gnllCx&=b;!;m2fcJUhOKF*0EBsx<L)hGIuga!R3p&rrS8V%TW=M?
zEQ*st(J4QBIjObHG7V0}od7SlD8s#JUkf}9ShnvMp-o*hdHN%~jcivk1Gg3J8U>2!
zejc{B!3H?@Fg~Y~S8GY0U2T#)r&XL@>Qmk00i`CHW_-y>?OYW#XM7crp{6M}c|3qk
zYO?L^>sYfzv6X(itl0WBRJG_o-KppuNgtTHS3N6YW1mA^Gz@AP&G(HZc9;be!1q6e
zz9WxI@Ej2NFEfecB~z%$sWrQkdn-$C7umDPEOJTuR`k~@xYf*<d_wvJCQFw4Opb67
zc)+Wc{v(54PA1#y9&klxR`$6uSC!eNYLGymyHF}txE0E1K=(j*s`6g}-l++dWxg2e
zRA$ssbGg9$YSJXmxYw=IKfDBwLs^q+?;#=~*SW2u4&;<2yP9vl*{HHBTUY}$W{?e#
z{J5%zIjLYM8LvS2TM;*RP1|V4=TR+!*E`F_+d20&NVy*=>MM!vcSBgBw36WCHHYG>
zzbi*%9c68&=}tR@k>^*VAu4M$8%ZO8Gh3N5lm7s$B+{R{YLO&ZrS_&GYd$X_Kkbj_
zUB8LWbTWV2T73*>@gnAtBzJlOb;3oD#8sPj$|&}%DsF8<swZQ`Y}%)nrTtWzh?iGn
zakqSCv(`}s)G`x*9Ay4A5!~pP!gN~WhO~-se(L}~8n0(CJ9)08a3}7T9e}%G{=lhK
zs*)kHH*UGAvprampDSgLdbi=|*{`H??TX5$!p&p3=(ZykiBvU@yq`+KUjV8902IZL
z)j8M!SRLj)<eAN6W9_iqMH$O>$E`T-4CO9NYcawQ!-JZ!r_Ca>Dh^w{b<5!*kH&Ns
zS5jah2aj6XgBuGsA6oOIwLXHD#tyEEE|y-+!K`bw5vLsoLtcd!?jym*D^+K2;JDh~
z!_rLSnHiaXD&lWtYm0}Gj!7pqvp6B~GCdaIZFSpuSdhE$abDkL<=yC>OD;glF!k?L
zj4SVR!{<{S#E3=+VlYQaeYMNmOtQvEWj$##Q?Rq~B2R9jRsp-$KdoxVQ`ID!IgNf)
z`U=i5w60f68O(QMkOd+3HK{fbwMI=bh<U9K55ec_mdoh-zbct*G<zHa5v_EZC?d_h
z+PSHGL5Ths-F)Lolkom^!cKFVD*Qo(_zcUZz#_6GJEc|bPAfC1oAJgxkgPe!CabVe
z$FQh4q-yId6TwL5kOqHKUSK5o%X-z}_I{^iwqAsCG3KE)&>=HT&ose}CZ&k(aytC9
zF%XgX(tA)L$gJ)aa#)-S?fgHWc_<~7a_85za^h-o#>Vs`%_Fk3wFsk%&_u;~an$t|
zh_f2jjkUf$cOOAsjR@T1#MI@8+sm8+ihi|=b2q(uG1F6)5#@=VSiAbxm7bp@m}xSL
z`qw<Y;mT?#)QZ+2@@)l2tv2q`2~%#+GwDf19KO;?+{C!hm1Hdpgl_EBP4wH4LYVgt
z_Ia*((tO$-l<TIo6`@;o!zl;Svo67REPHem-ipP?S93b~!&}6F^)+(UcETww%@Ql!
zkPTAKO=msLVp)bV<Z;DY5#s;~=94-ijXf&Q%$Iu)l|P+qJ-K!c0XusR^(ojSI5BWR
z7!`@F*?H1NR16LY^r@8WD-{wyB#2vso@-Lt_*%f;VBHw#J?japjc9rQ0E{n1=B&<y
z><Rz}*05k+S-tDnpsh(G%B!r|HhpO;QV)5tUA_f~f774-%U&y_!RBiUr=W68ULo2a
zBbx-TH|lqhADXVDvMBjohvQx~+SvL_C%Gcr&b=jX&ZHWR(Ek9Izok;CCDhJTX}!@h
z>eB(pR{GUz?-?$Y7U|e-$I6O%HPKGA+p;-o;c2M0%r^L`a?<2N>VvjHts8%e*5@94
zcs;;UYt)SFc$33lRB77$P0?;72*slv%L>$s#$Fw{ZJ*i@5BHb~wKUFoP?{~o`X-`<
zVryxX_1Z;Llfph7nYS&BJHNcWYOI<wQk)rst9S!Nl6K4tfb=1&3H}x7&ROETojr~S
z{#2Aw(CAo+bu(u8UvUb{G<WFX0U>rAn&x4%n%Z)+#L>7J3d0o>(H;0ns**c>9@ogT
zL5vg0t!3a>i!Ef??#DzZ>r=-`A+}nGWYy?^uDWnJsxr!oka1lcrzDvyV9G**H+H3H
z-9GTec@^N)+d5%2YYZ)pwHclv)<qwUbC)S3wJEe(Y?mTNK5UD~!Rb%A)shw296)2a
zr+2Z)>O1IBZxJ+kIZ;es79@PJR`#oeCD2z9yia90{p3h9S8S%z<@um=jnAhQrz6>#
z#zoaKM;o(&B7wsvHOJ~cFS#*G9lC6hB*#wFp-%5nRH-h9k=(B7vF%QZ-zS>)8Rw^=
zb|^r{6*&VGgjbkyD!t3>kBVSsvzG5iN}>eZX;c$Pq(5O5BreF@W|5-Uw-nCI4O%z5
zIL#thpV|Y~yI%_4q0zZ@AG~g-=}s$C8lMSj=F?7oOfeGtl22NqbX^?vw}k#RqDaG<
zLVsdt-}CDa=SXxt7yf-={OQ>TG>Ib6^lve7YYubkT8*ITCMHI;jHU+Qdm5Kwl%!K%
z3F#OzNo@+C>&;7Trs=S&Ev1}59;9`o?IM(?7ojv(7H#FU7IEe-;;lz5vY*~V5go@$
zmr4>TR;dwgG;JGFPca~i&0l_MoH~7sGCYvO9CLw!M^Y$)OF>Rje>0laZZ>MI7zF&F
z<Wy*6kbJSR{3>L%G=uk13ta^zPXu)pndhmG2N(`d>q01-!ebiqf4Xa5#e;S9Xn*I9
zN9JozhA-ne{{SbrA6ln<5%NJJm86XdmiIE*&a*<>U3ynprt3EPiUfU;-1OSr82eYH
zf~=zboVI_N*N2+;C9^i{E)A=Mm<GrIFzr)3%*BEbLF94Q+PmcWA<AJ=n^GsXfniYn
zJCRJ(FXMs+a=-v9!p6_t<nC`tTi!Suo2I?HE!mIhN*`p50qMnK6=h?cySt3@8Q&|y
zr?0JddOW1~T?^zAtyYXGY<b%e=MJMSSG8GHe3VB20G4BSJt-qYDSlVasjI5kP-tL0
zFf}Z5JQ2P*Rv03YMr5*X`B$%c?7SDH2;pr)Rpq#kKhCXgWj!Khc$ZZE($tJ`wqb{X
zgTX$vO48-6We%M?im1xcxpe*8GqTdP0dk1B8P65z7q=^Kqo?<bV0}2PW~|7?t4TG~
z^3M=zHugI7lkNqAE(U$8oxIg&)vip_9jL!?eJh^Yjckfi5YvD;0;B^hs5_ddtFiSx
zA`QB3jJrU@ac=x^SyozH+Jx}lqM24zK4U*X)vjApl~t`St)sT5xRAG69Pq&mGXPLX
z%<h9bMNi?YhLQ;_S&8z_Q$Cg1j+AR)<oi4RAqjICmO4PW_<*nlEfP1GCOrrL0ACg8
z+WpemXm>If*h&>c?s`|BC#BCewzX0`Ze}|I_hzjPJ^+w2AI`k#9q7{1QA9Djedjr=
z%;BQXe45SfX&sFFkRAya*0U^5?1TI~RP-NGc3+bo_^xXCHcWZQ-HP82xs6hbedA58
z<D4j_2d4NdP!|3bjy&=tf9F>&bwc2b*3sB^IqgPYJ3Z?eXO`A+G%QiU>ME^rPMc2g
zyk1C4tgM5MIvTYni|xw(>^q-fSbHaMbFrN_i*6e`El3`u)fs$EeBbiT7xSuA(kGtf
z8MVzqLVtO;*b1um7c(5oGoQw^Ng>=oqpwp_DaI<mkTovhLjWyC;3`0Dx!In^paPQ+
zqu#wc;f1hUMxVP9D;lpuNOkl(9YsEE-1O~P7a+IHrwv?k-0Y?BqblNZq~|<!u3N;~
zho133Ixi&ChU1>cHF!8$xg360pTE0sD{~VvLHon?sRCv;J?hCIBbrRpC6e18^oaie
z-xb>Uk-y@f2i20FLn-`7e}ku4bk>6(afoB<n$fmwJ1<~xD+;|0X#7mF$BNgH_UJ(L
zJ1Qg8(x*>TlDOxsVCyGsvmbhh?r$%I=_ipcd^oa>LdL8<qN-`rWM|U4(dMm*{?WTr
z9Q9mP)sSxMRD{MmVyaB}rU<N8t!(&i51HQ`O=nZ#G>+$UYsIqGZnM=5VK0O!{{R%k
zmGBK_-ARCJob(ZkbXrjAzK5FUG~Gm4-+;t~WAj$D)$UC6v46yRD4rj;+9vJ8EEoCK
z2Av~ECOZz*s&|O#FWw`ev%ghB!Zat`bgR(aG~fn*!nq{1J=%K}HHf!cWOY-T!jO5<
zeJiSq_fH!fzj$``uGYQ|RvZ+B<MXZ}McElAs~YG$x_wOD<~h<#ebxrO&%$?2b)~<{
z_mO1$>XDWsW6L~QZuZ_Ky$+>!NLhI7YakI!xa@#8rxf7byHXeu4{AP?z(Q%JfE%A|
zS4-eqaW<^dpZR9L&Z2({Egq#83M-}4h)p#c+ck|X$i(~$)UV<bjVZmzt`gw&qOZhO
z7vM5(TN$c$C?(WWJgDT>M?=`$k--IV_=<x7d(<P6Mz*wqPXleo@PAWYOmZ6(@Y_FA
zvRf}f3H)h?n(H8KF{l`ssqGVd;CiY4R6T@3!x;KgJJS)`cw@sm)h_25CzT$R-P_p&
zP%&OcB7D<53J|g_s%_$@MXccyt-TEkJAL0f4hKq};^Sc0y?dWZ^=T_a@~VFFt-2#F
zc`yxUh$gz1YAM^?n&{{gl8&s@w9{d>8=K~DQ$@}C#-(BzKDEy|aw7;u)t02XdpWje
zbCzS8%eaz7JC&FaeQI^(OwuWH$rUa&dE&y1Ns6;=spVDfBPhqyrDm#XRIvJ87n*BZ
zldu$X-#w~@?vZPAluWLAl{KW4WQ<_vP1wi0@YIbec`^o67#p~z+}PaOL^nj=fX!pd
z(>q~SGFt9pP3B7>1xWlTj@6GQv70(zi_s~J0kD%+RSV%lkV)%Ytm<1GHI1^vYBwAJ
zK4NOnyt4_!tLfIaqeq%N?_-{%_2or1l+P(x@N#ih4xbZe%q5TVt~u(>b}_Y$7EndS
zxh=P<`&LUc%uGOGRC7rI%J`PtM!|yfj!j{&%NX~sR)f7Xd6iYB$5C4u9`|Se03tMB
z_2fVF8uDKb+KDv@Wk145qZI7}9KBwn^*Xg&9My$9SCD-W?9rM4$kz4Pw5Uq19(Mby
z>GiG->hf#ZnrWP~6W+V<)a<#L=HjYd@}eyIQfB~qo|V>dvP3F_8Ro59!RF@-2jfM-
zE0PtS1`LumJMC)I@Rp~3d`)QrUP>Z%JJWXugId7Gs#a$I0K|P>`o=pJwz`rwW^RYC
zO6LcTb;~;>BH<ms+g#?hT~2tyE(qy-7vjxJTEA;+-zb2EfknrqT=ABT_K7X8qh3{l
zNsmm{RlWmkHKAQ18)1b2cB^BW@lo`bv3E^^Ij3PF<S59ich-VEk|Kbof(3O#3y|WK
zR;zOK83OZHQFs;8PC5>wCKT?29qU$(OO$nMqi0|)1C$ZCB<7({5Xy+Vh;V&t$;3tT
zT<fNjiFzGQ*6t=&aT^iAu3u5obtvw9*d`OKoiKV<FllZ@ShiiX^0cH2aO6MnvE<ba
zS5>vSBwF07W4A+4qU9^23s~kZuV&s?kyT#ePP>hI5sFA;W_b_s5l@c#;xm<zfgbgC
zObuHf6<^vo{oyJ)f@^VgeG^y#t`o?(JQiN{&pLM2B`IG<c|#1{0qiQ;s9N-y!iTG1
z)7hypm8Wr3t{A%FngXYm-jyOnyRAEuVnJ*wRf$%SNeajh6sriJWB5v=aOBC>geS2U
z<LgdYBpe4|J&jQWl3dLDLrBCZz~-6e$^G{H=mE+0hZ#{L4^fj=qSLiW@#VoW6bZDA
zT`R+yyWOk25~xF-O?44?dsMiNRB07F?oL~{HJp=d(lUxyE8Tc&QkW%}om3O^5XQD`
z{3yD9n9;4#^9ta$Gm}!_?2vn9cI;ZvJU1PsqbuAyw)2yWRJw<QH0!cu9%{+oYvVL1
zsLJAX8;y+kG;3WpVCt+Lf4tQ_%o<y=+^Ry7AHB%yTk@;i=jWBIRcnnt(C#k^ej=u}
z)vPWYLk#h}hoBVQjO@-!#rBa|+{ZPzWiygk6OmlYJF2if`KW7Cq7As4J`X#!vVYk%
zspAq>_S!}2BEkI4S3-W-jIx5c>JMLPisT23jyhtb(8qPaBe<x+AmnjO4NHA){?0WE
zAdv?}kY^RMVdB_#1(wty$smUBU6IAUTTNWqRKAilrq^O<{KDZ!rfZwkb(ho}MHwP6
zdH1h5-lW`_bDB!XtpO3`+wh?Y@|llT<kp8f85F$Zw^LpAirf4&t&C+FsTow*^SN7e
z<PL!KqG=407Cq@25ohPCh;j#|CIj$CIH?1LQJ$IRkkI&aw!79=-XM0!LlfvLvDfdT
z(X?f^gXMN#E<bO2(QIWsWRb$4?s}fJAnQ_SU(&8-x4cD<<_^FBeQVe+gB>0){{VXn
zn$=0QV+zu4$nuXDK%-od@y^BmRf^|iKU#xTvLT4&^H!}+Zz3(zaLrUcBsIC{TF5P;
z9htxY<JzC5nJqjj2;LpC$+UCNO4A2#tMdN<c<QUNAE#X0Pdu?sh$9P`2Xfe|nmoc?
zTIFLvkw&C0KDBt2%ZIAg$qI~bBX7gHa2mDtqROPP^3eAf>-{O7JGt4W#cReSZO!Y)
zO67mUUWnp);tw^QNhp2mOgKmcYu$OS7Iz}$l%KtcI}hnr8c?>Nj^x%%(QH#k-MAk4
zsy4%A<0oP=DnXXyUzFE1tgg$VK;wg3Vg4gJyBTMkno93Ef{I{vJ{^~4ii`f+e>%A&
z)wFgcne(}3HJ>oXO;{9&#KS!au8P(u&9aqI(t;*7XqVEdyg>G(N@aJClT6!+3qUxh
z0%?>k>H|;#zKzrYVs+bX7yDe*T2c^KtNb~@(4!#p<{bWY&s*U(rMo)aL2|0yx>h9L
zF5!$0)yX~1nlIc&2BR;Q#eu;U!)jBRZ{3*ZnyO={K`QUq0LjR!_e-)UWA-gtnIjFk
zl>KVlOO<8DYVJu^MJmnpsHveb*5_>MPxpm(ekVBa(w|b5Ml<-42ZF9SnhbTr=DH}D
z8_Ab}hT^iRv86BKGHE{P`c~EG5$Oeg*#feT%+{vyIXjx?wbsR>eQG1|8_VG~{5>pp
zej8ZGBq9`VuUgO3=G(YeT0GUeC6Y!<h~yoQHK4O+kZOUC-@C0epfm`MKox7I(Cz%I
zMJK5ltg8Gbuv(opzZT0`y4n5GA9~7P2&w-76z1oWvaaRe*E#5-lqF1?T;NCuu2Wvq
zUr)Lu1F-U@wxabjsj;Eq?RZANWVmdh;Agc#so6(qsy~)6j=P3E>fCK6X@~cuh@fqR
zf!4I_+=&4_>lvfmTNiJWb-z6Bu5=(-nl9C)7op?helxhgA|DU!+wz#gkHl9GdG|@|
z1$5l<sqD;&E7m?5U9{FwMdoKHRr%?Tm1xet!gJpe?HVn9JA!^$qvP=Z0QJ`&ieiVt
zh0ZfeL1CIwaX<)Knn|Ezb6vl~=zxxSNaU~;olk_Nx#%c(712Zr=!3V*{c9R~vT++P
zQoo4KG@<ns!%lk9SK=!R@EI44g*9`+`bYMY%_o_#CmlU%QoFIx_Bto<&>twcIR5~4
zng0L?=Fo8{Espg4mzjg6Cb0pB!xBJ9B@>$Q!o;>~JTmybP3mZupk|nBtbw`BHi3wG
zRMu^ah58@mN$es5NCLEU9W3g)lEJ$Q2IEsF%ND^ayB@=(>FaM}8YtmLD?`fxdRL7r
zxoUcJ_8E?9<)y!vcQ@T1t#3*<kXD~0E?k;n6762U%BeQzI8aG7=|^*llI4dp5Ltus
zN^*TE=2)N@7$kNTteug{+p8-riqg#T{o}`=s;s-^3OaYMKBKwcPAV%yTE_+^9kW>4
zoBZZ99PX>BE?6TPD)kvq+WAT5y6z*KdeU4Ch>?N1X*w^Zb3<dYNQz<hL9;oL$0Sq{
z-CRU}cOws}%_k=pVxtJjYF)O~rck0)4hK`*)%4XZ<>A3Xdm6Y_?qgF4)*#llS6RZw
z{-jlTG#h5gjFcy*rCg&KqbO9JCuTx)S)==;Kj~6iSV*!fOBqlPaa>cUD4vWf!fVXj
z5Uh|DW3(FM?;(9pSXeArBrvzmeSJ-J&)uUNobc9e$5(4@9jq%F4nXTuit}ElNdEvU
zaOb6FUFmjzGqCd5{{RrJeT%o5pR`8{<0~|R#Iaz-xI#PgTxS>q&{t*}{FXTD+n7%D
zcLUVdQGxFYe4hb>{o617mb~M`5bd>+Q`OwmV_&$4%N6QK-vXX1%02L`&2L(dOSd?~
zOy6~XtyCk;B#5as2)P^u)y%hZw9_$WW67v<o|WsRWn;x|?&F3(DrouJcWPoJ1JP<l
zE~AdLfIibaIU}E~Z}>9V?tD#caH9bfyL~Z_>s>V6<71w-nwS16H~fRw-M^)9H*&KT
zOz@)>eHooCk<<7N>7du_)qmx)Y;UGN`qk_*TP@wZkhexsK9!WUi$&D;DkhFRBSAC9
z!v$LhHD)H??OsBSne<9mS2XmY`}h9<cC{vt&9`OhdslT8?w)05YS$@5F^aOJS4?AL
zT8`w)%>>ZV&QsLvS5eZgZ5grVp#b_C^UE9Sh`jA4>nn9`aGSq~R}|u<c=pulgJ~uc
zUfjO`lmo31_=@f;a}ku_^f~EV{>?HaCvw%^tu4|)bG&)0^1u6D^^K^>9P<`b*vHbU
zoOxL+fo5U*xtl-Tb4~zYhOJs92GLC^)|keD+La?=xtI8Q(h<)&#nAc+xO!JpIS_#J
zNi*#fhE#ShOhJ|a(*nu2TxOyw5Yo8OB9P0)G@+nmJMh%+KOO0aV@s3Pt4qsfgf_Gz
zr!RV=F%MchP$SX2DGFP7c^vh8xhJ(byi=qnOP$Bi16Q&++H!2Yr}&!2NLkS&ARj?f
zKZ+qp8bpP@UhTzm%6gqJqU@V6Aozk3d1ZSr&7YJ8H5C5<5by3KUAFPWigUL)&0ejE
zK_pj;#=7iIO3I)y&I)y{drut6J`(Ag7WGrbMM$=^HKwASnEhnUsaY9q?wy@RM%L?G
zG|udPb|jBlRAkz-9cU!(h##eNehb6uaud}f$o%VWr!>A6Uh%3E6}-KakbfHErhJMX
z;SYq^@MB|J&-(_od|e<qlqh?{$L4AD7uz|nIL1gHboCyFhPfE>&svj0w$06+{pvT1
zW9J+iU<BnbMMR_q1dNYL0G3j+9nr4E{_b;5fjW~w$v1Ce2izP~k{zfN9Fx|N5h7zJ
z-n;!6{{Y1<zR1;4GO4k`TU)fG;c`eOpeyr}*bLG%BaL?&(SjBV4&G1QAJ%{_CARe>
z;~474`Bf?YQdh8~#*WX!HVYks>P`mcTxZY<<@N1L`%_eBQ<#(GkLl}G(8X&u$|`xO
z1tU`Km;V4EOHcajzs|jHTe>mJYj2(<+)v;t<FU_I;zyczw5isl$^HfYRn6SVxehuC
zM$!0D97pBNS%*8p^yZp&WgeNQ%9^H%y)fyt`&L{U-QI}C#_kg&OTsWb<Jz^sLNlpS
zgWmT?T%*pBXn2oJF*e0<AzTcOO+Q4mHg+#G!w)ab)3sJQ6Qe=Xmhx$R!lf4r-0XDN
zKF?(uakz}J9nEk)H`G>d5Vg1jc@cy;?oU6~x$0>f9&IPhR_8kS!soHg5zV?o>0Au-
zOV-bZEuKIorkYH_9b(NE#b}&{H57+s#BMrn!K8q`Mo|4Lp4OCnt=z9Qtq+KbyBM5c
znr(U~C{aKiMu#TdEl2xCtw|<DZykv%NYzeW^*~XRhRt*uJfcW;dQe36EZj$mqayQ3
znMIH~)J{F9HB4g+b4~9-Vcv=WOwbNUpc!`Y!6+A?QB`S6aUdqW7vZ()n>p<>^P_Gb
zO2V(gZAtsbpk9sE+4Qb{_xFFi?a<dW)~3{7x^nu&per|E4RGa*lCyf6bjzq-wvT<#
zTxXh|=?vGb2LTfmeMm+a8I``Zb|f)K3l11nkW{$=rEg5tb5w-KN5iS~HQ9KVp9&%W
z0CK3td!skO(v{Q#4tJ>s`PWk`71W1dP;>d#bNEY#@e3?VFCFV^SIE<ldynN<t(m<3
zD4ea;b6V?u;4iCGLvv~TCa;3z4QDmJKI>%uH9)zGP@OT1NZfm7y3ywSrd?m38L5IY
z_u`nzk(hO<gWyvTAoHe3!0KzUv_P?{fzVbJp2pOXs;&~uG3+ZRUVK`i{{U<HR7v}y
zg8k>r<z|zES$7g{eYxvg=XsOj8?Uo0NfTg>;*uRo9b-;wTby~wzbQTIc*osM8A@8%
z<zM$Rou}@a*N*k?ZHbeXL&2@2jWBQDu_RHb!#e?1qr3hXV=@uxT+=<a?Gr^3c~MNu
zv;mwN$b}WPL|&k{2lJ;y@v&{(?Yu;y9T{TiS%yCf;58;xK^W?K*H=@{r|_9hyd2kW
z@XG8*c+lr*604K!c&!l?SGsKcRk2eQt?ul3Zet7e=lWNhVws&M?glA38Vv=`X+E?7
znZ-1k3K&g#AAsZ8rk4KuX0oa9jc#_za0PVNE+W!b-Bea|_GDMp$JC|bV~s6;-W9`4
z-Dy<#i^G2m#J`i~BD((o2tqCNe=&v{MkhX_wGiCtdmG+=&(u;op^2P+6}g);_2dUg
zxlz;0{*~ezaKu+8A9;G3Rp==L@uoPgreuiv&;=rggc#<XZN<vt-}zE|0~#&lk@l*R
zYrF6kkYSnUl}6$~A4<lwy~_>@Q_^CLA`wx@eJjpuQ@)!L+9N!kiaMIVEX!{gRqNPS
zT^RDKob~0;H3UT*FDcyT0MxG}Hp$5Y`I_vetgg;_)8>4!Gbg>4XWJjhzP_}mH!g9{
zdYY`#oH<bam>D$^j0TYR>0FK7oeN^sl%Zj7T-7FrHxf%PX5dn+?vgX0tx`l%$O~^_
zl}<qQJ*up-3lujq4fit~B6@pNM?`czU5aHH*vTQ>?bOvhR^Zs!NpP!{QNZ`C6Xuc9
z*Cb560q*W*l4h0K0zZg%_V%uTDa}<&b7(3zaMXIK9@S<mtBATtPqF5+Q*qe2Q;gM_
ze$xHraIYB0aw}E~n51FbJOD*+3d&B*;H`#n)+LLmqj(ucK4FZC%G0!1ZKRS&nDf8}
zx^Xa4vN>?|r47rkZzA^3(wcyS^A2l+*)=UsLN#MJ=txdS6;kE|lpXwv^d;^`P0SXo
z^UoT_#x%5y5*4>+&{q~SS<!ciHU1$gYiPXNsF(ZF0s7YGJ3Y@^PqKI(<}v>OEbR#W
zD}>SF@^tvuxf!f++ax@xYbI$S%}HKq?iu>NtqzxLU5GL_++SMYFD|6Kou`@0Gat&k
zaM073^YL{q`AVmnUtwLR2&^`*xujA!1Em5iM}t;W3{;GcR0JL?s`zX-YO|^8kRRqN
zqK@Z1e}R$wRsR5AVmpl2pUOXYJ@H#TjHv96kHP0}iL~Cs&HjeHm2NzZMipc7%r^Dt
zYo7WuqO@-`yo|3A*@$k|;YJog&Ifv-3F5dWtu}hNX>(Cb*U}f~cJ{?<sO?>pR)-cP
z{_&I<HcZRYo+?GfbtvdtmC-VeIH2$<WQ>#4ifKf4va#oC=D4pGXp&!|YPJIxFpQ8t
zh*z6Y@>NpkchiqO7Lij(c=g*?+JP>U<#_%5Ya;&8+s^jw_oFx_p?kA&Gn%?m2N|@C
zgYzk*oGU6Y``zl1nFNZmC{ii_%#Gza>Gw~yXtgEiO%rE|lRQyj+yY~*TDB^vLkxy$
zmVq2}r~NE_1zabJ>t`AxG6kyF6K*_G7?C*dPmy+=!hsU3-!3W{A918yK_3+<f0mHh
zN=%kGreM_(9mVNN$(79pk768Bj+I7COmk0EXCXkzCjd}M>31vkKMk-&+ENalhZW|`
zy-iayr_|HY@7B%8kdlr1Q}umU&;7&=`@ZCNta&}_ojm^Kih6yNm$QhY;C~Z(R=v%Y
zw0G9FF|K(V+qtDVd)7mQecQ95pGmd5wh=M~Dn3>nMRK}^oj#o@Ne1LUb%#S)QlCa^
zRvKLqT3_5*wqca7w{Oa;$V(E(sX4BN6rIjy+g2fdYpU=iy0hHR-9S(4R*a|cvs1<K
z1ET}nNVUO8m^h^O6+Mkl2aIcn{li-RDAMomwFqUjhs;!qX*mO&ny!LvDVY;^b60Fg
zwx`@~6$^N4SY`+8gnz(=D>Av_TetAetV8D4p~(4%H0SWXsN|`&>(Aa4S%*AGKj9hG
zm54T(<NLy^w}o|l<XZ9m@S@B);tAnxU>uuY{CHHS!+O5L7S!bT+$ggjJV87)tOX19
zQU-s#DsT8kbzXnbPao|HO)%$(*?cprswi7-PbcOT)#<QHZ{g7e!fcgMw+z#YB63Y=
zaXM_x2b$~&MOFZT(>zsaHu(n}lirc2`zYf_Mvw*_cIT~LS8L0LU;?grE5?0?wJ<b9
zU`JUOiHV5ys2V#xI5PNA6cRslYk;?m;h6gIR`xQ6>~eLfGtDDIU&AnqpAcEZMsp^A
zE^E?!L`U|noIlx<`3kkMg-?kdapMKfxgpLzMgCRINJ!(TH7wRUA@PhYTDR*=XG7ub
zXZt#H45axIAN%CjVI{q$qX;r8%7>x<0PEK_Bc7b&h=aF7u9UAHRz(hpVFaFHLN^YC
zoYSJZTis?2ItI=T`3M;b2RJ`km{zTet<O_vj;m0n&b^sqLe0Xc$@Z@m)vrz7w{>t|
ze(4SmsO?=<HayHa_pSlAKy%j>1R=(9GC}6JS?Hri-;hsCijqeC$nT25nj#3>4VqcL
zW^Yj06b+Yw`A4{{O<=CY7|uZzr5}ijn;iKgp%jL_6Oc+Nfzas<jX8hbHEEkQ(H)5D
zK_hjlvT`Yn6($``Xz0^zkPliBGg=$+CgrLVCMmO-JxZ>%19j%4-4VCanrJi|nnGy+
zgwiWykF5Y%a#+(1CLuG;YxrYVI!?13yRj4evJYyBE2|Y3XLH-5adD?BsmNpm8o|`t
zWMpGG?Of7E)Sl-F;wz6T0=dC%0IpgRe9Tc5p`_WZWAa^Y_2RGD!RJePDH#e6@}^2T
zpufC0?$xh-8YP@OfcW556LS_`zs{gz&lPAvoE1Lx+W3Q?3P=9{0ID;d=)myQ?+&L4
zUb}++d{=XNgiRBHz$6n{%VtvkA}v~s+(M_jjMk>Q+bxkjPDNw(S+u{26O_4rwan{n
zorC?Vh;DRL_)Q-M$Wr!5F~o_x^8Wz!)VkaRYjQR)8Ad-U>suaU_{PA@xC}BYO)c6;
z_aXc$hC=MnpWZ1e>MBGjD&{u^f=il2220(?qE}xM;d5NoUZ+hZp=w6Of&Tz@v3x})
zEaU$Gd-_yM;S@eK>T)*|U>fGNoj}CP8XuUPnu==f$7J1+iFtPSG6rHu0|KiF85=6t
zBe<^Atd3bSGR2NhQ&;R?_in#ROxp_o0D6rZTQ9T`x$w2n*+B#wRRA8QxN5r|`i+&u
z)h`w`WsGEzjn$Q*M6>HQ@v-E-<M?w|Lu19pKY5)`jIN_gSaQR8&INhp(K1R4bv)N+
zQ_qLOSAK@JJR5TG>tZn2Sr?zesS2mUdT+!yBfIdUTy+Q)(Egv+ymT6)3O@+Fds5&~
zYBY*zBym6u&m7Vk3L4evehV1Tp+2Pl06NOI!c=!Uv^%gpYoyglO(HL<3dWn!lUw46
zQ^ie<GB0-imBQS<0ivhGJU8&nDYHGo#-k;e3iNwRSm3gb1t$@p?cS;g=#e3T1~OYG
zuM~w<vY8+cTG-8z>g&5mxX1kSKgzr~kh3?jt~@`*KT}$}As#?a6%#Qhn(HzWlo|>`
zf?MTLl51y9@b%LmyOl~H>5<m6o3aBzEn|o8C>ZTsSB0kF_U5^2Y-<In*_Qp^;ZesM
zO#4UTYdFT{&U~un;DR*TgQYQIRbwQLmi0ZWwlO^ZcdvF<xzs}Zv2uNCi<gw;$t2dd
zHm43<UgXIi$|_a)7#_{Ru35XC2qrngFM6$eCzA*r>z+>9n#Es2eHJ~)%>3l$i=bLp
z?4X*4?*u5TRgqpGyG*V3SCLt==I-R&Fv5-ydsC|HtZPYK+(NX+iDX@@p@Hm0a-D+5
z3ZTNMP&)xrBxAh*N#%-e&T&%8M#ZI%muLweJe>UnJcWut%s^G|T@i(<MI4fy)QstM
zHo1;@BMgy7b}7%^KDB#Ho>a9>%o`&$RVp`ILqB5|Pg3M}!bjRv{Y6-~jdpJCE2e~$
z?=Hs`NIFwXxd1#=Gi8PaIckm8<_5W@B=&Q$bw?@po3(i~$>qB!AmEXa-np>PSs~&&
z5n`@sxEhVO{dQ00E3o8y9_!**{ND{-$NlJtt{NEpxD`iXT=@6zMH*!*+>`SrLshjs
zQXMkb%OT4jy1u5kV;3i6dh(KryFC8@P`;Z`xM}8MNjwg~RaNJn)$7J8D#-C=Cnt6|
zax+Q?G|3q|e878CQ2JE}o;KtPw6D$&T706pC{4<1uJ~_nw({~`+XDWzv?F7lw7te>
zd{yUAv>u2}aBs`6t!q6ZXG3IcXmb6ZRkEA=vWoTHc2<a5NSrdcAlDS1xg*iH@1SYj
zD(c!U^~UYHhkOxSvK(>Tb*_54wJzrPH)zTonY*H#$0yWQmAjcY=d)tE=<IXi7r!P!
zc)Mi{>M4rcS8P^<r|zOfKKQ3QRLK*4N_XwH!r=b^uxlGizAYFH&eMV`&cxHbHqTQC
z^sx@7;khEY8rF#0CSGKSbYWcXqZgRQ?RR2R^8z_m?hSF%pE6h7Xcb|jC}WIM7}n<~
z!%@zg(<ARMBB=@NBu)y3{OQJDO3_5l;(5n8HEkGqW1iJAiBV@P>6@M2fK%j{D6QGb
zRC!ApXKFw<aSfWjpS*&Y72hjSZZwS|Sb+=kQy9%5NQtBNsESY#+)Pt}8%Ut36_{bD
z=~RrEpk#`@quaICuWb@_^8<cd(;@774!(BimtXEznjVI#Ms)Tg$sAGK4_cE`p|jMi
zbgNZW{p6%&w*LT)QW}$ezwMiH+qVE~7{RvNH*-lOj-JQFc2^1njpaXZ)p~PhsN0y#
zi0XTT(z%=^26U-LNaC+8{>ybFA%g`x5l|ZLPUzs0vm<KkyblXUXFP}Q$ozg4OLHmg
zX!xo&lSGqm@GxWauREFX-jm!Z>{{^tqYj&>n~8xdw;We*r1+xB>rsLviODUJ2<=pu
zrxx`($@R@H*#k<c8lE!X44SMwQK2^iXb<8l?v8#*7Gl@*miw&{$nA=o`(I8V9jJc+
zMbNoe*|iNnhB1%GR0j7;Zn01nV?NQOvHt*4e>!)ar4x@N{&Wt^Q9hNQ^(XVEiS)c5
z+@H>tfo3vGjVmMfCxPosEu~0|zCq+;tpTZ(4~8_j?C0AK7C-^#-mvwr21YE4>FfKz
zasxI29qMdssypg&K3gTjLfKaYsrJQd*@XonKv?zwWYn9Qo=lrUlq!7>TF2Cv%eVm2
zV>>N7=J4EJ@gM76X40}s+esYNJ&dEfCyHR;=Z2@20B$Ccp}*jU&x-7S-aq;YuH(cP
z418N>Z~QB6&H2?Wj3@E@jvK|tI_!h}h5l8}NV#vNE~cwwDCvr|0Oo)!ahkJle|u&}
z*{&Ht{uVhEah1IaPQ{7!{ZipJOL749L4#d)!@B|^xYQ)vT0;QSKrFwo6Wo9L>D_G=
zWJPJW6QuDK=h!q0%YXqZ82}#n<NDW%smaOXn$DpZa^!8f4{$0$`>2U;ym>Xi&p{UL
zxrrODYDYpKZ1y#a+|!|5SJIMhB*${JAsPyE8htAVTEIl-kjFnv)|7rCDr{j$Fys!D
zmFT0CD5eJPmT{-A`^K#kTdk2DNTedGy)mM+f{RH^Qz`T`f?FMRtb%9(sBR<cS{z$4
zgo}!*<C=zaM`9+Z$)spBoYE6O#^;(8{KM-=3nbG^QUQe0Zo!~OtauAtX*4M9?q54e
zRAPD#!nCzU45Jt$*089oT^Fl7&&3x3^#c;0l?Ojs$+j7oRHdP`o~EQnWYK}um{zrh
z==6Kpx;986orQCjm-hh}^(Qqu%GVi?^aB7@Seuk(>IZs{0;vp*j$hbUZ>}8pRw3#m
zre{6T%xR8dy9_uW<X5B4jV+ep&fTnkD$VT5dP*#{q^FsBgp7gBJ71f0=@{yx`PLs`
z=-#S1%ahi*on-#`2c=Ct&8mDQuM6Cmb(mKlf90-|#4wR{4A}WL<39D)wmixFWi*I7
z<W<XKEI7x^JBrX5XXJj^&tp-wq(dyK0rsFqwuuJSE$VA@ow3&qNADeU_9VG%Y3=^)
zVfc-|&{4e``cr!mPvc&v6M1kFHgnKd7k6^k7cC<mr;2n|w>m1a0&`3?(sMB?)s%Bh
zFpdb>*}4T4GA@&}k=tng4Af@B-b%Kq=uTVmuFp!)G^<=El+(EC?_83zJxStQT(K=)
z=$bskYuX|DRSRDT>8DbT;@w&%Or8(55Ygyy)%T*?Fnn#{Z9?MiH?)vQh@W(v8se_B
zok|A%^{b2ypk}&M;@ysAO{k`4N~JhpLG(2JF5zvpTX~s^s>lgG)v7a=$GF{y^$!O>
z%eV+xagL-4@U7Tl4+I`+naxLLIb2h>QZ@|J4FEKnX<<yE3s<UmVjm+#wqI4<`PV&v
zgwmSS=`B|?#qPqoEq&E2V~@F!arjnr{uXChe)0t`7My9Yf4}`JfWDUsY1H`2!+#8`
zph5PnWf_4fAIx#>(z{b|>c1}}o^pCtkl5C2Xt`p;J4Pw-oQ$_tHL5XevwXX>xrb0=
zHQ^JkE+rm}Yljc<&(!Ij$Vj1*c?5AGcPD@bYf?W8>W_`k81wz(is^Dm*p7nB;cZGW
zpD9T729Wq)P!X%)N$#NXD08rrH?)lc+Qf+_0$FjK5x_OJAewLo@b(^+i}<d=l(h{Z
z(k<#RX?!<3MLCU+AmmmvHm2@(kdrtFwWSnm5^W!e^{uJLsf~I4k~!|<i;V7XL0QvJ
zB*&A1=~^}vp>KJRs%~y-eVhtTM?pH32~f)82b#dKurZJs58+hq0lT^Yf2&Tu_03iE
zH>SFq8gzR>9Wp7G>J??juQV@)KBjB<vh*@#kW7LI{{YpGynShmyGAQIj`BL`#?c}@
zX#lLvho}`G7@!CJD%G8+l2r}Pc{Mh?k{q3)*|ONj909<r>!?1_C>xH@IUcp!&1$zt
z0$$;te7|P3Skx%w=Od0yT^Nel6?&0@isy>F+c$e8Ph^A4T>93H$;1Y`s4iM9o_-<T
zyir+x?==*RNFjeZq^}#AUsIj&9-k(mr%xZ+3@WNiH?4UyuEs07a2#ZJtZ2Pj6?L%F
zR`k=ILf1d@**}=BgPH7m?zS_cPCwYL5x7xY*zb*c8s%TYm#I$w08xicwld0oZ{c3#
z*AIPvCZTZCOi4Y5U@2j#{ibwcs(qAFW0Q{5bi#pMrwHOOF+%b{pa>;w%HLcHd}v6|
zYP7D%Y?dr;`ck}`ol$y|TBJ?QMc`L&@bxhj+v9@33h1?w#ZmjFAH?MIHCPmOi~5@2
zGqy|~qKd2COsp%`#-^K(v|E4IMgDc{J|IABW1By5hwEN{*pcc@+O;v>7QO^Mn;-U<
z`qeFB+w6LS;9(!;TE#AEYsk_b<|a$IdqRED%l`1I+Iwb6zO~gwXo#)6{zUhwx((gb
z)g?9E4UT9(bd^cnigWa$OF>^#KUICYRm*xIt~$=?qR!stuy9g~O!RQlvC!!n<3%8j
zZ@At4D(<tQ#WZbuZx~?1_sBnpdXIYW>r?hMblIKLv((@<8%DC5Zg(Scjp{0kH#OBA
zRnIzWmR1lj6s;nGR*@GgMpQ~ypl(G<!B+$x_0tq{$?7r6ytrDGn5u#8N^vd>!JW1@
zrxi`yv82pU0Zn5-F)IC^F%d%$*;A8Mw)04|Ubfl=SSk%m1~kQ*iQx9BxY1f<*eU={
zYSzM&2VV4>YKdb>!LEnFmf~$-$a=(tilw=nwmLr-NYUtxxA>d`UV2Rra%t=~j;sKr
zXlz*NDs7tX<YF}$3+q}wAF#K&ifLqw5Y4m$+PUXF8%IPSuC7O^=xu44VO{Z`Zbf9P
zvVGE5=xazxNfA;mEJ8~hlao!Y3|Ng_w};uS{59ruAUg_Z&U+Rwt_|0OWt^OshbQx|
zH<_~BcA<`+QP#0}vIWN(=DQyX>FD?GZ~^_wV0vWGw<>=LbH=e2(&E}V5|8CvD1ey@
zbOMLDH2x7{nnaf#V-Loz$8#e@%4K|xe(|JeOw&y*i+=IO7e9E`NVvG!pFU1M+B6Eo
zR_fq$=gI#7eVVZq#lZ(NNKgCbfU+O<xz&`Jr1trZPPcap44Y?@IsWmaJ0iTgo$|lS
zbs*#W)g(H`ssLoYmmvDnxVcfe9=)jjgny;Wy_=A0tJ3aexzO#}X$!F@B-JKoIHZp{
zxI+5H(vSJ<-_EtPbTCbYP~`EBw2fBA%%V|phT}Q+70zosUNFEO!jl?0{Ue_bLO<z|
z{<Y?_N7Vb%dl^G^LIDkqhO81<Fa}K{L$LU3Y_s@>e4Ml~7hkSG{<YKVjIHB{FmhJZ
zgYC(yTO9KE{zn($=Ek`$f4INOxzkc>9gyT1t5Hx@i9z(I2BtAUF2Prl#{#`?Pqtkn
z!?xDj95V%03Dh%nKdm%#)q5io;zi)M(V(*o+vR-yK=rRQUzp;iW|V#tO&Ta-UBm(D
z>s6+e&B~L^XQ8euU7fuRYfuDBk6N=T=H2U=?r63s!^sMJid&S#kvFFmT|=R1TQMp3
ztQ~N4=c6CzT2c7OtFem5Cp4Al9HB)pH#D=075bX6Nv+nzvNz_SYGWXJt_~iMGosP1
z!BC8AVrcguVbZi)991-tm2!nisxRK5?rR2Db4)aikkV<S(iqZd0p^evLHT{@rhpnx
zG}34TukcK)V7N^1tS~95;w$Sb+3aR-F%{0x-A8(vE23cSHgosWH1|@<-B{+5-0YAO
zS=i{gsc7N#2s6kW){oi|bc?t*VV$@FixYE~zhcq=Iw-3#kd`3-01+Hh0-K``nDwY!
z^`<mN2qwF2d=>DdBzi|oiSCYO*Cm!gUXi3`>qI={vST%yE?XvP=n?#nCmjH+omysl
z=?-!M!1`A_z7%f_YO^BY!5OY=RXeVS(v>~UYWyR#=4h_sk$Q$0uIF8i&Yv8LGN{0x
z#=4?;wLT<e+nj(?@~enQ?}J+ikV*aZ{6z$kLaEM8CI?BU!Xt@ATbZzx!6&KP6=5%*
zZk*rlRzAMq=@K`f1b%gW3}d?WJo^6L;`%cb1TK1#Yn8Z#?dE8t#^pFTtEI8MQY_P<
zhFggQaG`fS03B;%19wC4h4!g-7IZf7srlLX{HxG(e-3H789qZ981-^FtfwpJjUcV7
zHrC-n{IP-Vc{Lz+;O8Go#M0xIIi^1<lkZv<R;>#bb{Qk3RAa5o>Qq+cGw;spwg<NY
zpz`Bx8A1BiGTiE?@S=s^hqTMz_14j!d{#}z!~GsZlt)AC#)Vc<*iM{eciKjpsigS|
z*&Rk}#&xd+-`i?eB)WXkGXOo!c6(2CHhFaE+Hy#zAA@z7Q*y+Z{{VbYd>yR={-ICN
z#Y3+Ib*wbnx)gXjTqlSZ@o`BngmrW{Xw`q<LWPK?!QrYUdkOyl2$rHg`9btKq&^h8
z+DLKv8n0^-J)DP+25Kb(uG8#zuH!~64ZK$J4Xlb3ipH%)C84w_xbAe;F_*cB4!ebG
zc$cNG`}9NLW)**pADLUmEw@aA{r>>zUTvt&)~bIHu>SzVHM}Qo653hY8=U;PIO4hi
z{o=?o=rdbIXk4D7Tzt7#Z_1>Eu~iJmlj&BBOY@EP4hM8K;WI%b+N7|?yPc#U0pFV7
z!@quhr(E`l>b?{3y}pm7o4c6Q-7@DaGq$VU-+iiPKn6kWT+ox9S0*W0D-v4ivt1Ad
za7}5qg)ShB%Oc@%lU=DcVCQWI+Gue~`F2!LGn$w+m8Mhx{)EyKZdw^yjndB>Z2%~&
zD+?*4zf%Ncl6qDXcV|^4WOsJW9-9^?VZ4rNIkht=#D6N+Qg)4zf{UEFnOn@~In7G1
zkwMUhuUeOI093Y8B6-4sdsc1D_$;xh%Q?UmmdfSzG}w7CBrEr^L{WrgPsG;@o!Qux
zk*j1d?%l8etX*#HT4uJ5vc<C4^sSox+*!{JYe<3=3m6oV@H1TJaa?tyGz^hB=}U5q
zEnM{DtqX8_1)`O-PI7%~XG}t|G_BN}3c70beGKY9-4`9OF-+OCZa;X}O&Ie{zR2a9
zvPj9)@NO;Dn*o_dG1S!WgC++>7_5@CWNUq5U9-S(p193gxc>l9Nv_nJf`UBykN3AK
zfxDB_(wQuu!h>=B1zT75R>;?FOkHLJHHE;D$;@r?gMre#dDm+d$3tASUF9LyS|Nb*
zS{gKdZnI|?{wbIF*H2TLTQlD^g)O4a2VJ%28kOvJ5Sb=8$3g0An;RD@i6cz1N;G8O
zbA;428FdSTET@m*UrMG=e$~?m^2WqYPEEE0r8+gq+<R4$4bYlM+vhFV4AV<dGN}2O
zdXrY9K3ri%YXzfbawAzb##`2(WIW4_$DpcO$s%Q<Y}U^e=$-{oG<IxqWUi?#4hg*u
zd*Ue0vlsnwAI#SSdluIY2U@AC(hNJbOBO9;DiMbTNHyy^&B;CwhTxo*5wWi_D)&dV
zQ{$%m%-uHZ+g(F&r+`5}O6xpBZZ7nJbsr`%xP2=;I?<xBPm;7IW%yQKHr_`60BBa0
zn#b)<I~>;Zg;QR0Ro8MA^%as=1dl^q)Xq3P$x??rQ;;XAnXKgL7c1(9xTxC%W7qoD
zaataQ4U0)G4RwAT)q(#2MV+(#-1qjcKO0M$o%<ewcUCj}MWl%&lGQ)~vm^%{Jw<R5
zhKbo*f~N#mB?okJVph^4^E(<$?yj_sHpt~WNMYKcX~=A5w~3647J!(ZQ_ys&T=C6o
zI&CH9MLFHObTf6vwQW*GvTfGt0>;D<kOy9CI~><BLJ?7#5C>|%Val3dK4J4!Sw`M!
zMkI3;HEt=hNs8r_kYa-Cip-*g!v;K5DxB7<VKBHGLa3;KBfU{9M;w9N*KOc?aJKgo
z6y&f9-reb~%;)fyJZj-|bx!Rdt_0Bc6+eV3YY8EW>14PBpzH-py|gt}zNW?IuMN}-
z3?YJN=HnG(N7k(D;4sO#TXuLAgQoc`bSjIQ=Cb&MPG$L|$UU)K-lr|Bw}tH#tYmTs
z$23xeqV)>1jFyE;4Rjo=X|ChVTVDsHhk^ERKQW4E&V2>5Pd~w_A6?KN=4;Os10m0^
zG%>@erDTh58%*6muG_+hw)!(%yI@Mi&U<$Llt*x>u_uX<WYQW;=W7NP#LWm(!RywC
zxKr56m~D{aq*mHS4+K&)F561JQ%C$*HBMCW?Z`b%ARvY_BVbjilm;Yr6u^nj<{Wf2
zHtZ{@>ccdKRhQPMv?0h-!KNct^K=l~nA;<v9`)R5i?c(tUItfnQZuQs!s_q~U3TF+
z7Tw3cH6@;+*3X!*tUX3Z{3&%bj>Us_dw)JzrIXN~mld6Qk|fxZuh%q~&eKM3_>e{I
z8LvBTb4~1J4cT4(_^T451q_~)jSk1*V=>pRNgNAuKT%qjQ<;7s-YNe8mPxdJLal6b
z%j15BE#n?_=?D7@{Hp=DA!=Uc(b*E9t1!|ilw?!7(*s|^x_|bUh%Mn|42XQk2hdlg
zO#XhoZEUT9ZiZGKx%oi;wNcpSsqq{K#EbJjxVFK=M*<E#3F};X&9SmN)JWcv_fU;2
zlDEui#4|KV+x}%ejbPF`O}hsVY>c`38oL?Xjy<!A;*!+a8L{x~+Z8p-Zh<7nB-Jqz
zsmv?wio@5i8FAS2S4ZO}ufk%HNnU}<6i^1vnt#yOoN{W(HPIc381)MpV<390;|)k=
zmYUdcgIz_tZ-^;1Yh#M8<xAMHSyC<uCZ&~4m7H_m6%oxNp$uwaG?>IZ(|;O6EJWE~
zaY|?q!%d(LlfiPZ)FzL2%45zw@+u8_Cx^r`I}RL%A52nYFO1K)8FG6XY?+QA!`7hO
zVY_sAnOKG%l``D4mxnA)X^D-y7?qEptCp$^V{ym>ib5*i0Dxn?Q8B+(=|O0Wf2<YV
z>%bohC$gdaDUm(V%UK`Y1KTFO2f+ZojOfjhWMlLg6`S1Be+kl9M7A~##2=KBeeqn~
z$1Ypc))QJJZ9`c}nRi1Ou47PQ*MxcsRQEKiu<LFZ9^$<f<2PO(UUGc3WB7AjQ9SCq
zqX8l+cPYWEX)hR&e;U+8qiGpe{3=U0i`|Uifz4+(WK>nz)LV2ET{aNc0oaFJsFu-D
zx)m9&bHuuYF<5<+0lob@3f=5;#_mZRcKKWp7YDU-dYz5syoPI(Bsu&2ja@UMrl_E6
zTf=uNZK<o00}+KMwP^M*i`4XcN!5b0zb3D($`iE)<LE0EjT_X{wbQ1Y263@{>rOi;
zVHt0jfz;Dc?qfQ=A|$uaL-K=?J1tP3Q<cdGb`NSK`wA5Atc>X=XL5IB(**3jO>@ZX
zQXaQ7@3nfG#zv8m(s*|3&V`e9aaHw8Yk#(rB!n^K0K9%R+Z}1Eo_+0JRxy{$iR6rJ
z;g3pRlhU}Rz2<u~-@1hxXlP@RNMY8Z#W!?U8ftmF)Z(lQNSB?Xp0#F5C>jXyi~&`i
z1r44uDWOQ!(AYKnDtd!g{7WG7?PKq|HPsK{v(2mgb?8*_%k0u%KlIc3SC`zDp5<d+
zNjR&T_ld**01WEwA`n|7N)&Az_pL~aI2kR#?Oh3I5@d^R+cI)V^))A!#Ag5n_ok7X
zEeYdWBM~EZ(ksZc>0`3ju5Tp)jB^lx^d~juQ@=FP+bv@g)W)kF--qN30FW+o>s)QD
zq+Ju#=B5c5>tn3bwFI{c$G1wmf8t55A@bLu7&W_<k&>q5%vmqx0|1elt0c@mW^DRa
zO`S1;URie_=j&3gw3=nR3ve1|9J%N#KBkRnVvKs@+{#*6OE<Z#7&Rs1Cb}u#sGYhO
zlI=(Y9Ez})0FJe(7bZ)U?@KMl0XeMXYZhW&>P2!3U<!vp#rCFVZ<%2k{A-e<X*P7l
zE2MQG{{Rv1P|0|<;pD+$E0QYi#_X@9>i2i|6Gp!*N(L}G(p=lJUbB&sab0y9RFQ=x
z`}gQfj5<_T2a4vthh?nF<$x5iBTU?M^{Ke*XhQ<0xu)uTX|o?iG?wCf8J60Wyc%t^
zD(d$#fT<pGNj28Mo>Vc$daCpFuDWn>Qj<AqO*=&rM!Vl}9115NF!nXshmEPX+~%n7
zG>n_X52)McA1}@MP>eV|+nVNUN9uGh(7#~8<aX?%ty;MGpkjCx(@yR;TeE_v@m2a*
zrU`xr0-Qiqz~Y-}E>v&U?4fvy$iccWz^@m+U6->jq=bI8%}3*rRsIxK<29|}G&6Xs
zPeGCl!n)Sy6rZ}!a`6@1RvsR@xSKl>GC=BU!L@y6T~_4DBYcB|UqMc!FH@Mu2|H48
z^E0_)idM!#ie}(O1xl!Vq3>4(c1)dyOai7VG~(Vy^%$zRyesHOY`di%)LhBUQqm;7
z$eE77D)yY1Qr@*xzA`F&CV=GEr+6|W7l-2{V7n7tQ`q5_$1U*?{{Y15FZ$v?nXWSG
zJ@Fq}(JRxjoL?-{A%NjAUblM}o!~VC_-&QHAzoA#toLg>I6pI-(dSv<2v5wl=o%c$
zccrzooRk|#K8C8CIa@0otu*@?S_PBok8^7I`D8mu?mJb@J;=Ns>_t^1Oj?v=&Cn}4
z<~_5wzctwn&L}-bp#V_-0CuWHbu;xbyH2<t>>A*sYyy4FL^eC<6M}QcO0#V_jzG$A
zROA|o#_Gp(S<z~ji+5*da2snYVTk%y8RA<&vL)PO<&%c`*Nsj*ljRk7-k+g^o@wXh
z8LrzNTVzr;RMi)94QUfLXj#eTLa~k%9GbN2$4<5B`#ySVN1Ym~okt$$KXLxI@Th<!
z;8&1$m#yr#MoI5fE?{U}f;g#C)MQ<%_MCdvfh_%Mw4X4k0yIpTvZ=O!6`yZ<h3-A+
zSfoSdWb;p105rtLO(y3a!kP~qDv4qs-6zt$C&L!)Z=yxJtCGX$YLS&qjGrB`5L%$<
zS+n|A6y0e(#}9-L3sx+wWxRYwIPTt+GIm!qio3CDEj&qnEn#B7edhjkp>^TeVO{Wk
z@DAPUH&V@9*;Bf9GTB-tD(Y3e4Ky0m#g5Gfr38qWGAKRBuG_@kUAKl_5x^>{qwy6;
z&Re1%3*5Z^9CdDyZb9_LaD$1CM{`N+IBv^mUUx`W_?Q9s*Pv>R{gdI3FM!(wc>Z)C
zp6L)?%5FR=u6ZpQllj*u<T={B(83!Tw<<it-kE~9&uT`5_NVvY`t_(Tq}*A5+Moi>
z_DX%lJsgP{dWN6{+k?-oHd!|9Vbo9qB*Um8pKdlBng&|An=!^SkPqcvi>64jXqH5E
z^1rQAOsed0UM9ft47mh?tH9$tQDzn(2muM}ik4h{68^|=31=RrvG1^C``M`}9e##)
zXe;c7y!B2&^%ZYqWgXcRgAP8mpnx=Dk)hpuF|<{Z`s4V9)gMDj3QgmUP1J*o43GCk
zTG;1*8#zxFxE>&sH}@C$RtlZ+7O7^@*%k+Co-QdAPAEK61HSlmU?a7;)MNwZI~0-7
z4z;tOOZLmnOH*V3G=@0X9vN4U=~j+O?$0Li1^)o{hP!uhjlN{3k4m5y8SZNdGqwvt
zdwXLm`*k$u9%Q>$g$EUrW~QuC3wYCM;YL2SrEefBxOc8eBUqFr1!R=_oK(GS4o?cS
z*dhC|3jP&|uBqIo)~<)fO;3c#{tjui-g1Q$0k5b30LP;JO<Jy)>`2Er6$8^56Xu|h
zQW<MclMXAXwLiNCtsIqyTmZ|HSu-ytl!>Owaw+ds*kYvWL}r+26g1LO802gmcBykf
zh7(CZjc*J}$*#bva33mvD%0_7Y`TmR{{RV1w3#1l$dPf3^vy?d0bQfDL9v|g^N(tp
z*|T&FFf3m;?;qR>i7uHJZhDG9B(lCzM>*$;l0*BmiV8(sjDkAXcdZTH6HImT)A>^(
zd!v+r2t`5E*P-|!0*PbpK*ePDHj?)l)axJU_HLknNX2qf9<`NC%`m?W%)7R(Ugg7H
zzE49%GfLYQ?bQx2d9M3FQ8t7l-IIn>&r0Zy=bcw$Br7Ww#&CUU`&qiJY9V~DQHs#h
zBM7DaYZ|XZN^0!ewtR|osMC=ebv47e*S4fOz2fQC21m)-0*rSx=DLmKw=)?Tvhp=<
z`x(+&ITKpK>=ztW%l$bo>`k-2%#cCHO4=&Oa(6iWZ$|#fkv7Zbq@4~rR#LH61RN1o
zVDx9Fcox8Fx^JCvGmwq90M~Su`fi%4SIa_2bCXkLxf7I|vJdU3MnqUR?rT16MoF?-
zOpd@*H*%o_wPj|wH5rxLg;K=yz^G=)kW81z3maJg^~F|8mS)3beQOCbOEtMExC`E)
zdy;w@&8wD<-@_VQ*0D!&^8zMN#c|TLH7jSI{YX=hgPtqC5NZ}Ysa9Uq$mQR|+OfTN
zSfg0mr|&9Lpz3NeV^j2O))J#8?swD1(@M>Z{>#)?{#{)C1uE(H<E^B>_swVQW!UIg
zs!#0>YfHGsUfx9>;-X<9^+q3sCkZ0eXg!$`X8!5?stFj5$MU6S4rsgK_GZ=M$83{U
zJWU}Q-LmJ54nG>`iv8cr^Xfm}{D^)b$>r#_=Q#cE^sWa)fZScQ?7Njad)CcfWYB-Q
zI#g|^hK)X)nzD|6zU}m`gou+Yq`kJ2^))g!M=H4b)3FyMT{Fx>F=Njo736xNc{+rN
zo*s5%_0B8GrteZmeXSICHm+Asm;1h6YX;Ix&R8Dwpd@Q7&TtfyS6c8TmzS^yC+}3$
z+}a5l?QqT9(@d;atx0JWE-s324>dyKQpiqgNW_Uq&4y<4aGh$(>M}tnM<iEDiG|Hu
zZB>+nCj{~<P7PFaBxb5g!5LhVO;E2xP)V))yLVJIa@33sQ&kK|56yO754AApt1~ap
zIowaRabg!N4yqF7nX@FtSV#O%!itYjRY=m*9J6xXwaE_WUt1P*OGS!C00G5Ms6-6s
z9)h%%w7z6VOhfcDbGA%WR}yEXVxEUnT9HKqO4&RuUyI9)a0mcVX2Rs+c$b3XHHG2}
z%|~g4Az1C$WU)N3u9`F(yrr@^s@6$aZpuCm(dLs}`)$NTLmYdfX6VMf135U)TI;JT
z#>nM@YFd_2cCIQ3o3GZosjEW94g1$KCg)X%al#W-8SyUFJb{Yhn$zZW$A6Kprl=5;
z+dV4P!v?oF<W}%}wC-?D@~!hKslXq~hd+t*qA;{>&Z#@Hjm}H++Pr_nafmhh$kbp7
zIQ+9*)by&yTli?qEwm4M?7kd93Tx0@Ov<d%ZHOP@IOi47JLzvDlS@1HXQg=ZCYQmt
z`%W<)Yv;xUHx<!0?;Ln#?h{5zrhA%L#z`l)H78Kv^SB<hX!HZpx@gBn%xX(p6>bk&
zOn;(d)YRMFW4?rNe5&pERcCCqOW6%>Cnt*6vt6-{D@s2RonM5s1$w`Rc<}9~`+~Y^
zY;efpz9F~#Pof|6z<)DbqjIEHw6%$p^k}5H4p*l&>iT`d-F!NE8-w;|U&|uA)U7s;
zcB{QOOU)ckpF7(~vMKM#_OAZ`!}^?JWRYBLf0c%MR_Z!j*1x?Lj*9*Ofo_y^;j13f
zO|yp|-mK{_c;~}A(`QJ%D!uA(t@u|=cR3>c;+W!t(ym6`N|$4Oyrb1kdGd@59zJ7P
zQ|fn8(GON6`(RXvcJL~iI^xT*OC+(ZAd&-vigu|N+AQuZ@N<pIJ;iyIY1*r1>2pD&
zi0Ig-JMaZ-W5r|=winW<$i%f=Oypg%e6UT}6|(FMKx3zR^>Fh_n@sZY6x*p5j%sMx
zrr;bAR_$!Z1%-H0c0EHuP$<Bv*S4!HWaFC2J299tp^eI()h(31TN>!-Pg2CQ2JcXB
z8Kg1j>Ks(<C<1^gO*cGMa@nU~xS6FHrUIOkPA3aNBj8+-?rYScjyqdBYhC1U6;E@H
zYLS&rhcn~3l$%(G-dvOT;<*J}^GWVhJ`s*;&8_72;G|8oH(n}QO=@VHOq$;ITUkVr
zuna~CBDCW1CF1oFy+o<08b+guQ&7}R`>V7_j0f=ds({tSSy+iXyMqz7t0b%sUMY}W
z)06F5j1hKc0=lmq9BFpw5B#<MhML&T{783#R2>#%Jmy2k;arXHl?C+i?fa4u`q1_q
zekN*oU1Ph}VIEH9L%F?$Yk2zPvA4HefB|#QUeqmt>oi}sK`(@reIWiqxfoUaqH&GX
zFo#kJEza@9Q=P}0K<Fw>39QyhCjvDraoVn*4eGML7OjKRkxWJFKL=eAy5WRl&~j<(
zrP;#XTOl%T!({aIG_F!I^+=LC)`~y7S<`*uw8Taj)X5<O8*@Mi^5Aki*QIEImqxW#
z>Jq9Zb#^)b02EnECajV~01TOCJ-bzFYioEDGRo3R*bbE}sQf6f=l6AqJMNXfwI!XO
zhfZ`aplaZ8hCcOgV<_&(X8XX^$Qvjr8Xl4G(l%Q=*CXX;Tvi{4WG~_^RI$nP8~K{P
zi0A7g7senRc2nq!{Hq2Yq|~!$>{wIMtwb^@6i7PKg=QgC5>EgQ^uYBCxfjD;1yHO6
zf*`Ar)tj|N8%Lo0JTe&=7K3mnsAvBG>zcEVf15mpFw)4<Hzz%6%EVc^cdl8zPev<3
z;}wS7s&Yu=)i{ak+Ou00TW82nIssaiEPRoZ(AG29wYd>(@|^bs)S1n@k7|}RxViZf
zvh~09XnhS;FO3SP!eTha1vb3{kSL%HYXkm@zf)F0;<{t8ByY_`)W$|?84nbS9cFOD
z>s?l!`=bEWqn5JJvoOjU%b9tiMpW5KUiDSQW_6+>DXmDPFwzr1#_p+A_Yb84U`;3E
zK#N*rjTWVD%5X!DYpU^#Y}VHE54w$m_|jxQi5QZ`9yG^0hDA~&mnhvuL9#PvIGEJ7
z?Y2O|n9CQ^ZaMX;LH=N915x4z)P7>1-b`RnG*%xo9(&h+tekjIdqls=f+xB;Z8I}E
zvF*~m8^P)l`ZYhqF<8Hd-d_VZd{W_FBpq08HOo$GI-e1y8~A9*ylg9bS0O8zE+lVO
zCY>ZY#IX5*j4`hF!L6G~V`n9|XCBqj6N0YATgz`E`e@Oc3dbDNP)NALwP;U5aV*#X
zwsY-UFcBOIz0Gr0dL0yvOH7pj1Cv^1QqJ9^ax;vRT-5iutyiLo>T^CVls=4HbY@a3
zhKexKZan%{*1HuwB5T07Kyg#-EHUd{1Yc6Vr=YH>W#y7YiOL^(;BU0stvX$oF%<MY
zjUh>OH#{S$Us_J_TgX;79M^Ge7ur90HNj>*d90iFtZv|yH6l@J6}Xd+tv*Irz);(b
zBNJ0f(hcERCt{I$cQw+bt*xv(VTnh+IW?4}bEcdvsYc*}NWiRn`v|TO<#64I>MNb@
zc5<eyaXOvEx=qKIamqK{e#WZ}VizAPirO%398z|6J0A;p;(70Fw3bM`rA%+{#dbRG
zuKGoq+{X}<76SzO*IZ=Obv&xHlBkmEa}V(rNbbZ^Stbb1LmIyik8PXp+c^IKbkr){
z(WljKvn;QUED!v#oAjwu##(!INf+uXPil#HS(X#UT2uY$O8r#Rzu_g({{ZLUf8Pe5
zw27avsEHR{(qE;*ej^nB0JUjTj%Bw$?=Yv$7EdKeiyXRCGH;GsRBur0n&ox-xbA#U
zY_T|SrB^C*-llf7k)<wfYs;{o6A3ykqOVgnKT73v+kdm$DwV??TPB^o<7zMWW~2ij
z{b5$uw+Hbxyh)K^N7AfD!~h3hrBN3)AwTF~f3sd6;zQ-@lPdHQALm|8AL7jPU%?Zx
zie0)BO2_X;1!CEpHb)?um~|RDZ8Y1WG?J)nbgH*6=1NMAGr+7h=4|Lzv01pP(@(RE
z)~Q^U%<G8@f<dH=MR5X;GZJ*cuDWqe*_^dx)~L;5)|bj1anChtQ;9Y#%hWgNTgEm`
zGYLxaO?4)QXylz3bX4Z6NvIg)UBASCN_8wMSGgqGt`qNbSyO6KxjT5QPJ$~$miqcA
za!HavznQOXfF*^ZMnDRP6aicq$E;D?gj=yKV{bIiFW$!k&{sN|_uBkRJPyNZ{HgWt
zT-4&^rEbSll8lkoX;Nl<XSGRvCN59VQGeZh#+Um?$jF`XA*LYSdFfakM3cXlY}*bS
z99JKCsp?^w7_bEJK(72WV%6Efh?K8nTh$uq-9jQ~U;sROiiUe@<|0X2Mo>B3o-5O-
zIHe?c^wgS*JIxd-=@zJ}0rJQ4tyPO?&wArgQ<2qo6f9RNWH6JK12q({!S}Ant3jAQ
zi5a&Kw_AaIVl|$U;_6aHc^DPNS5Lg_hv;wEf|glh%VeKwvZMVB9{X!bYniUrIBGjK
z{L0Jv)0=@?y-m`wLgawe<sLAI$FI)E{#p>j@~#+ZU7A(?5-@SOz^>E7vCC`Xtp?1-
zRaQ;J`mqN;$X7#ao+#sK`#C=&(Y_)|(fBs=X~sYm;M;RsRNUmk`~Z4lr3;9$+|Z7O
zExP>0zbcfP?4+#&lB4&Ew;e#IN%F#j+OxK-()uF1KFCP*s@ctCYKbFV&IN72yJHo#
z6{(G9p)!n-^{-sf;rl+1Z68b^u8Nx+Q#cQZC;lhVdqO{%u3t5^AArhxm8D!vJ?qu{
zB@*2D9ynW&=<0tnTvZjSGt}m94;FABd2UB#Vzp))*{w-<+i4%IZkM`x4D}xmYCqfl
z7FRg)3Y2XB0C;z-%VlVrYw*f>u4-Qq$wS?9kD1g$cSNn~GeoPt1{@MAp>A>B%*^jW
ztC7-**9(~34?|viEM;bvGt@RkWk*>vwi^QQgn9u}40IiGYgUco%NFu^at8D@W=oO_
zD}NV5T(x7Z&WKA(97W1`S5ilxy2MitIIK@1Taio+LFAS{D9Nf$#%)}+X=G(M2W*3}
ztyq!cO_H4Y8rqFz8CZ&xT%&TFuF}rUcAF+Yc}i^kDMYcaOjnyN%@Cq=^`Org?^9qg
z^)y0vt!Ipi=#Ip(;F?34p^29Dr|)7hNCps1CfZ;dX~1(p03$gxJ7<asG_*)O?N-(@
zyOg-xJ*&LFnEwE4SX(M*mXM!720yJ*GM?uv<8F0yAN0t7GhF^Kw4UY1;Y1C~R*)94
zz>&t=rGLD~_-ZA7hLN$mXQf%mW905vW65B7&1B!`^H{NL7LWVDR#2&}nY7@!Zi*z(
z4Q_Lh&e&XcsJ37Op0ok2;mBHTYTPgIw&VC#$B!gFZK`LC6N;{eZ{tLs2As54QT9>|
za=N$tg|5HrtA9FO0enpB`~_yJXSsDb04kox*0VfIaxXQB+!9hVwMV8X=nLa!?tvkj
z!q2<-RmbEjore-b!RbVT8?qaj0sy0}W~IALHqpmCnv+AN@a~+aP*4Hdqvk%f+d6q`
zQp!lFJIRb9f$nP`P_!yd&n(1Y#yXE*!kQa7JBITvS;@}cD?Q0(Ca8fVLbxN4YRm{s
zd7ujSAAdOOUZH6ua9G*~;5w?0sTEN&snF;AV<g{gw^;H4OkjO#wVkY1sIuI@n<*Sb
zIuETUxp4SVH0>Kk?0!RxfI3zK9!aSz+0r~qYhk6mw5aAxouGHkbGfP^D>g+xYL?Zw
z#UhVe@DNzME@tWb#wYNuN5c*z*L5cy421rsud&Rxk&)u&&bc@HtNg1HA;n8Kj>N-?
zw+YCgqD<Dug)9!G;tP1lZIQP2Bep18lP6@(qsLb*d#h=B6fOfxD3R2TIO+XqpA#<{
z*Te8xMYIW=247*w{c4|6o-0_R&h21h2x$(`c{MbWD8z9L4^!0F9JObo9fw>fTYLk~
zIjAE{Mty}!*ld&owNI$6C6mdtE$dm#q(y8)zvohe{*KGs)JUW~cnMTBrOGe4s-^Lv
zRQOC84p`G`&^ZE%0MxPK*nLe{&2&d$4J1@-X@Vkph5ArXkTF>R9qXjiCHVzZPI|i<
zhv!<(xLvuYBbpnOBp;nuaao;cuGGUw(F{CO37}&~T9yt;6bX-EP3b^|Mrh<$Q-Y@=
zyPbVxxbVWb<$y<T%bHAyZjNFrfMbU}xHTQZY`a$MYE6-ubfso$mdX~HxEv1j3}xGc
zHqp>jai->=4O089)7yhpPZ|2qQ4C}E*z9Y*)_L%&4@gJypw4@vo6!_yYtlR<xVsW@
zgdEln;yNe%HfngM4AHlBIL&gio+~<^5qNLmqZd$<Y;Jo1Ym`uPnz~yBa?A$}&2~Nz
zxmCSr&sR=A3f&wvJ`&!x0%~oy_>>b>P}YWPV#Ek#kSdIyYUnKV)@(ey4#K!`5=%qS
z!@<2x>swWYJjNJ3v)Zt=y+_V(B^%@i^Q`_D*lXXY-$Z42sqvtw_d#5~pAmnVx>rnA
zrcr+DHe-(<Rfy1fWY<BB#gcC=NK?tIyX`h@V$I<z^L0{t8kUV>?q}&5G}>X4?Er7=
zLCqHWoblfId&bPyB(KiTQNcP9I%wpESn{%Az0FzEZGP2pxsGE)&%HM6k2(?Nm5#FA
zu%SS}0<1inQm{vR+|QO!TOPGub{h>zq08&mQ0f-X45O&v`Wo`<g__3c;bKD}#dT47
zSmDHI?p)G!c`UUkE&;~Z89v?X-ZgDP-si$Cd}A^wi+ZT<T^71$om=9IJX&2`++IDw
z7~FVK*j4B?W#cBf+zzztXvL^2W9A>7SzEO|=~%0h3xU@(;WZW{49%V?;%ts-SXUbJ
z06NpOnGBlcqUWrlsw+`V`>gA{b#5*!Z8MA{**QMOxm5u%i5Ku0uAZkgAK{kaaoc(Q
zD(bNQ==^I~ibaGZjB)r1nSUCyJDS$On^pQ(hj^T0#5YUnDaYqtMIYkK^xwe~t-jrB
z;rnzwB**fsdv*Zglvaf!P6^1!&jz&Og<janyPC>s$3!(UE+xdI1I1CEO~*CS2$)Hb
z?cJKsx>kXS1moJbOR1EZ1<lF1gckn*Se^Z9L2zZW+T4)1t!~VDQbab_42a`A;{ux>
z?=D7q4k<~LmZiJP2G}K0mBnVfMGa(~%`A2v0ny{Ql5IhhJ3w%%dS}+V#7B1DH!M01
z!o1v5ypKZ+tr6EUc~+7zI5@zsA{UYi=`I5h`^m`q*0?2pVlhtEJs-jP{91&m6R=s;
z4ngnlTRNP$x;*w3jcRV3Tzt)`R8y*-k&kt7h}mZg^);7$;=3y(*#Yz9{{Uq1Sy7`Z
zk3)LZqfKgM>3a3nrK^EE;WnHg=qsnUg(iq$Mn_V@yDGs`c8SYMPNuB80>=pi#tHSM
zx+^?Q8E=)kj8`n1ZtUw&Nf&ITbcqr^02~iWboci^GRK#|=N+p^O~a`Yrlc;FaEHyF
zM#CO3Pc?JR^EwQ1T`Lll6jtoQR2!R<*vr(UF>4lb5rtv1KU$*`gQ-gy94-w-S8>5u
z>Ct?<51X#qu^G*tf7!0eYnC=V*=w(vSxpVnoOU{yn|OMf^G_YA{?m#3GGe&D<@t3#
zgkfp13AHE@xdq04o$J$lI||--zS*{e9jsnrpZ1%K{*`n7EKYr>%-{Hw!F(Ha>A8QM
zcyT$d%A1@RUw}YADv`KPtvDljGE^H70Z83zx+=)xlhmy+%%YihCf$R!D;aFfmqNR~
z<fw&UY>Spjx?G-uwSnfkXsr%<&r#d(>O16DvDyCs$UYDE1#MMmaLL|Bn|S&2uL0<Q
zS2V|Ht?B6mL)?yjQ+K6$Z^Oy_-ww$i{PS`@OxG=C@16AB@cxYAr2hbqnSk!Z{<M;x
zkRMTs*52vrT<<;}PxfmeZpY=9BoBJq)vf%SrI|+(b^6yOFNot#)t{NCWc%$<`(mbF
zzRQk3opc<s-I;R@lQR3#b*C+kxavh9$RpOg{{T>5FHx9xSN$tG9~(O(v0SNKj&Mk+
zqx=5=iLRE5Op#Ad!l#|bmW*pD#MUh2ZyY`B&N?%aDT~fCTrBgR<K&Nw;;lt&U<oGN
zgIP`47X_`$cTvSMB+vk|4!G}C-PrI?B8pD?j;E2&-ds&3$#;>R$QfoiCZO}A{`ET?
z;pad1sPp9YqQEtzXX<lN;+QdMF+d55S_UN)08)0KLzDnHq&&EwM@!)Ac9T^ykGm3c
z_}4(PnY>!@o<uNwzljw~V;OGFYsK=3b)uvA5GkxS`?Q&)c0^Ie(@GHrSe^w<<vDW-
zDQ8}|sk$^sa=eImW*GE1ss7z^VlkPRqYMyEXtH5;h$zi6NF+O-6sNGD4eti2I^a0S
z5w=Y)jjkF^U12>VL663tLqE1!@FS>_>2t{ofmt3Wfne9?Rp>}>%w~>&zBYDR6E2<M
ze?6ZcA(JPuHODO8XqrhrX&jPhV}|XWUxh%MKzOtFnbiAcxkZ=CmPm&Rj7Qg+9>R|7
zn#R?yE*z7EL6)Vlv@Il$fwy-A)S8{2hwS`?EJh1f%&5$BiiS~ASdQRW)hpQqF*9&L
z2l-P2&+a%9H%yaM_{ge6le#d3NW6j6RFI9)k~kiez^!!|47lzq&@UwNblpAA8)b{H
z(wZ`#LO&Iig4W(_0_CxvPAeMP!s2)h%*4kZP!Cq2txAW&U%r|o@bOfDCp&ldV_dWS
z(~h+)vV)Vq=~DyJn1;*B(p$C~fIV}=5xhPCn6SZQk(cR;<2(#Wy1FNlK&$L?t)yr8
ztbfGzZ}(UER#al8n@3_yeAOytO(KcLF`D!rhWd8ME;R;TT<k{m9sdAYsP1J|+9r05
zA@K*qtE<;jBo={|GJk-M2lcG)i8_a!;wbGx^8{cLdUMSW;S?9|66}s)S)|8)cok&B
zw*=Q5j^r$ZbXfH?^^XucdJI&6kbvod*QgbGRLz)PzO@qEsFmz<ziu#Ua4{*5(A3Lv
zqDaF4W2iMpSRd%&`jFI9_|U3+CqJB%?@3;X$P}ohn*x9}EtIT4RMpWRg$eF5Lx6aw
zfyp?gF{}+l^q9zym3rON9F_v97}WSp`2OuwyH84{amRBmUJX`y%<PWD4^a_Fv4)sv
zC~50b+n=6+GIklJm7qdmyQnwa_&?Q}5<eqIiS#&f%!)wlb5CjaqOs{xY=&OO42dBG
zD~@Rr<zS4WbfBV%B(Wx^wSAHT-*ld6B^TWDv~;LtI2oi^*yF}r_O9zu8~iHY^}v5B
zW_345F{8E}Yty_g7CR=Nxo&=fx!=TZQ{XS^?zb9yOa^m?70FyTUTZ3kQ(QW<EOWZt
zraM>fo+}R7mAEGrwJ(H<+n1xZ4T39o!^iB}q%1~3%Km&-$mE{J{{V=i{p1b?2?r*!
zqgNy<@=t1}Wd(F~nns!};tb>GJf16l+6~`350rHGuQ9hempzC3s#Yp%H{lVXL&^Kc
z(zx5pkKU;DU@5nG8Pks}3H(N$Zk4yNWvq)>zzPL-#cFXU?wTs601mY-1Z#rWtw|aL
zS<c*_Q&Oy$2R|bJ06I*h#FiMOfaNjoTsEnvUfNx=#H`U1ka1YbI^6ArXUodU_V*fu
zmoeL{gfxc&y;n%K7M5|uIUsU<O<`9{1s(2gTYcQr%t=x!B#wslD7-m+#;V<AMk;J*
zoy-gJMr(xeRiNINXuWuP-j<BzuCEiEuX*s5(baqkpC?F)0rVBvk0!pz@Y|G@*6vte
zD)G5i9^lZIPQ7`lvCyPX9;C_ZS0mN}<K{n&CXsh6rnsPe`jvb`8vW*^;472I;tO``
zB~SgiZH^`%okt*CicM}0?~l%{_;*x_UlZx-t+=$nRj89Mg`L;L8}VxeiZBFAF&y^b
z*E16{o*3h{(ONSKUjd_RjeuBV>T2|4e|be|5lpg#amh70ar`Iot1`K*1A%CrxUUn}
z)nxHa%Rk&%OCQd>ieJT<>OX=b?sdD1_Fp#wG#gPzVM`3Z#hTH^+7_9i9K>MetS#&9
zDyIUHf@9`rt0NX9)GRV9WfMD)w1sn3C!aY6nkAz-{zlw7cchK>k~6`oBt;}CH0T(s
zb)kYbEpGRzg^wQfR^kvCk>0uXX1BUKkA{SZFC!y@!2XriOt^9ft$EmwiJpZIg&Ce8
zn8Bn=j-d(tE6BGJ#k-p8z^`?j_@}Bjd?P9T+)NxDjanWl)wK(KK}2^9s>$nJ(u8Kc
z<Z#BMp;O&wH7(`U!OXKe{Y^!K-kf5htc!ATcV@?i3a-0u8>2QW(=K%Fm?%hI!o18}
zxoRWQ!X)5TlGbnGYN!O!PIqMFV-?GEdNEoTVrzs@$VeiuTETTBOn~`fPrTjD7QTc}
zl`TuQx?C|F@=v{5bVccbUDR;hS;1Dl%zbHETP9`7lsN=eZKhjQU{^bU=BpDPStLyv
zUM-zXz{RqHeJaF5{SWY}i+u0U<ZJh9)UJbrLHm`xJyEob%ia*0^Pe4k`)2QxTyIQ4
zr2hbdW_V)VrqvmjA1qDD=m{7V>HZWk{{U!2s{7evDqHFQ0PCjp#6;(wm}xS9iY@;D
z6lCMpME-T+3TwK1o_-tPCsCYKn9C^i=C!92S+VA$m9br&4lbk&a&uL$Tkhhy=dsZp
z$PnaG54~LG=ADx5Pjl9;p~<egTO75E8fpo(OLcBk$OU^8e<V@G=(VM7k-)wq%RF-V
z@idF+kL6tKt@C2EPS7qY?GGb#Og*d7ei;w=k=&12zxGj4TcbKDKi$cBoVKM0+4XCP
zv%*SNt4c?fJ&j-3Jy^Tg>vV~Yrkez*!zykcTJDKmPew^ol2kDI*PjP!hb2X4&*nEw
zeA{jArmZ=|sNTF+V_KL`;$=(J+bC>x6<Fuili2B`^g~Y~xQ`G;vsPo!NGm!&#5&W^
z&L1!t>JBN$af9ic*K<InsVerO=2c5=?a>tRS<Pr@rJ>2Q?qkndd}o^TnZ{d*e9SMs
zYRMQDP>!rlPtv(6ZjI^eQ7z<5f4EQOQQWZIkylT0XR*&*{{XH(K~UnX=MxlAFryR+
ziHclM2ux9m1|}%Q02rh+2yn$Tee+C5Yv71Lx4F5`CANNC)3gF^JV}3Yr*wR%Z^oI6
z-IyLE{{WEn&-=sqR!f{MCTg||d0rVer&`t1FRg7PFC=FyI*w~O#_K~^Nhhgd>e|at
zZ#L3pj#JfJjMhxiJQ6f=3^4Si2KHf5Ezrsdz$UaaBgZ-Qq{RSrUIkp--5h@a!Tf70
z#Zw9O*~tD97xJgj&3tUV;Iu2NmF&@zOz}OlOX80xaAha~`~@8Y?VG+QpX_=HTP`yu
z5RYo)5tb`AW*`&Vi4LE_bp8d9y^YGB;c@!cFDEzCXMiMc=R??0_?e?+6ptZ9+@e8)
zir&*S&?Z&E^3$$=h}4@MO{_r&9V*#eih-1+8IuOFHKyBVGh+aZRRO}>50K`mFyK`X
zT(V&@G;QgMk^og5JBmO@2H9@mjzQK%1M{xS#LThjP%m*ZU{x)Q<?$ldh!QblZ9L1l
zQS-O(tTxnSy^sdW<Aqfm3T0E+i(9-cYXzplhCGrFtzr|mpGwV|8S*ygW9n(~zZf+{
zkCAe<3}+5Lv;pllvkPyB5lfCf$Na`CgYY=T&EL>fzK1+^k(=U^opNuW{#Bb7N|tXO
zi8IjDpl1e<$yG}bqq)GZRJq57d<fvESrQB#r})o$v}0GIF#Z=?2=wdSMcC$7!Oyoj
z{cDBR?%MwVUcI?883Rhe82<owJbslj_)4d(MkTs{%k5TVKQJGiam#bD4%!TVjY#YM
zgxK!XkTRrfQNCW)ri`;&9j2yeaVkL}j$jyaD%9i4xZFGQQ6y5>e2tI|R@8|53um5a
zr}3bc$2lNgHuS|Zy%Ufs#h3<OYGtt2=853oQs%lT{0JW8QXKJ2TCf`CqHdI!$Vuk4
z^vQl<PCJ;-W2A|)fYet2GV@w9iEd?1zcp2mesz)69f%u>iK!Z)hNEf<8=7=S5()&%
zMAA?ouGhk{wAv(yPC_wNAB6-@W05f8=0uR=X&}`lJTi6%ik57N-Td2NhmyVOmT{12
z4S*4z<bD+-aF;(QE%;CfwxFV&$C28iU}S%4MFh?j0uRQ!*$96NANu$Hl*p;NISVuP
zd%%tNU!{7rmitWsbAZA;{&mdvxwQHs%r0lq4i_iOAPVGeTN$k9(AJKe$mM)NF%hB;
z=bFWg;{<iCifi36tMME9HJ$b20FbuM?%V-g#+jhO0ZC<)yo-Z@zy`EP<DBiHHm8Et
z_CGDQz6WE-H3hGS6>bEuzCb)He~lCPSu~|-E1S?+zTm7jjB`@oT`i<LR~wH4xMI0e
zZ1n0$RFaC=<aG<5yh+J5%-vjO<}sh+AI_!=Q%SuJgIgc*EFSJFF4Z%RE76KP#PlsZ
z!aO1k%xYL|*}iA%?^*$6VNvdBSR!Px0+7nm-71U$$I`KNjb1n)$6TLEmRfq5-WG&j
z>koFr<&T9G=~mqU_O2?w5!p*)MlJxU`ihi|E=J?6XWa~gLoG9&)JF!ntNU?vX)Hu_
z9DQn{MD;j%BPiuhV_v=Rit*;ru5DexSfAxzOxJ8rCc5`eI`N&8*!Z(hiO40TG7JC_
zLGHeK6Pm-MTy=^Ur5K=~#V#qKhk9B7H&)++TmCBTFn`m+(<V>dXMOQ<Z!Br5I*E5y
zd@_R)6Db){!ir4FpTLVy$b|jYKT}$9OfWE@Rm6$1(nvVwo@CE+`PQh#vqld!tygbN
zyjP9u#n)23eEsPNKb?5=e~Uj~)b;-W<=lM=!ZMbTG3w_vX4Lkro~0vMVqp+|#-9{r
zh^m6QGIk#EVyn2p&1~aiB<@nVY-bglBNEl0Q%-~g)PBOL5XB^n1Jac`(7_gMWW=L2
zt0-5EeQTaun#W_Z_;!IJxIf-%>Lg$EgY~aAwz6lTOGu*=#Rqkb+W!D@6a1^m$LU?z
zHDu0QOVt}b6d-Gef7oih<0T(yw{KN7-_Y~l@eXuS@krFQrpzJKV3VfT73whSkjoH_
zvyonIHaBUWt^r;wK|0B}qj6C`$+kk=n)A<2);ILq2v*2k%G=2q1GN+0!h{6l741;N
zB)Oz{6Rj!9rO>MS!BdE!c@MCmWWKs!&mWebt#m4}=H-cm=(}@Cmgk3ksZVZe8%~YC
zwpugT16;VI+><VgN-Z4?>vI_OHEJ`9Q}i{QF6PXo{3!a7Ap{&#9Wb@mNn3KJ%-iAn
zvtD`QMBQqVcSpr>)mE;Jccjd{7Jsm5_jj9jkX1m=JJ+FnJ-ZX!ubRu|v?Yp;!!&aK
z^*=M7q`%qZyo;X{+a3{d>mvUEI`K8@UDEiD3_rkHjBqMd9Fl9MnhCHxii%!EZ!L~>
zxmdO;rOD1Kl1%GJi8!fYT)kvyX~n=E^=SU?A9G!lQORBCT+vfHt*9NG8uh||(v$wV
zpXMs-bT}e;UyU0+E4Utq`U>YHIg|0CZqP?f1$h<|<`vz1Gnun@Z=aPwY<uSwb#;nk
zrS{Q{ujl)E`~JIDe52)Wtu^qTw01{KU&-+NP1E;lU3Y^oJj*j|$2c38`q!Mlh-Wz7
zjJ-`{kI1@~Fg{j0g-vh)gr22&72Ufy8@o(p%K5h8vB>7Lnx*V*MqLJ;mC5+6Khk1g
zKZd54@lo3sp@*J+qLhw$^{+<6xoSzCIHqcL@=K*jA`TamgIHG5C39%bP@ZZ?k383!
z(Sk2PazP!<T$1N&0K9-PT=I?Cyd9BW%-#yH;M7uJjimCu{c51BOqqpq{<rX`u|bGp
zmlO;!N)mwv$~Mpl3es($V(g~vpaX42<17H6L$->IP8ibz&~!q!9u@mROdX8v?^S#b
zcR_f4WMAZIobP6HTD;zVsc|Q(ky-K@lV!0ufmYcWa`L&(2VquZ(e7h`X4|+h$pfgU
z=1X%$2KFhC%j=p6?oTyj05pz%U#$Q%unSGy%>Z`Z35{05dH(>2zSHy-%4-rVy2aC}
z%CxLO{As<6<MA_Z!Sbr>OP=wND{l78PvYx&3WijOvuC$qO$vVziQ`Mc`h<3%f2vK{
z&s_fi`m3IJnV}9(9;TDrxO^vF;4_PBvHsfs03%r!kS(^Wa<UFXk_8W8N8@ImlM77i
zBMdKmclNHn7?dc$#zkh$OVKVD6(~3zXc=A9WS;fS_?5O;zBvrkz~XLrU#P0PbAwEQ
zV9YZH>CZKOL$$oY(T6kvUfH*5w%Gpwo~!)psquvVT3Sbp<@bJ7T?l^?HnkxI--krE
zZc;L+!vnt+&dKCM5=aqzv-fwL56Y-?_AzhnA$yi(F^u*@z^aVoqZ-YvO$tY?Ie4lh
zolaJv#wG*Q(gWCaXwA376Bk03$UjkBZ-NK!?vMK`R`)sM@UpI@Xt&-do;S|Wp#GJe
zabVW+G6T@oGL@xcsu8?M3~oA%ihO50RVQ&-qkqDBfA*J%E#n{*L%o3<5Ia`)#cfAr
zt*)Gy`HuF%>5eg1(85~ONYk-b&^{f9*sChrmLt@k{;{tpiZb#KbDEl)6-QRtc03`^
zR_#*Y<2A<4(n0rBRI(5BU_IztiIr7;aZ*^fmm{G(HB^>@TZ<sfpY5;eD$@}pDtm!g
z%VN-qcSJo!HsU`v&Q$NWB9wk4Ic#$!3Y=n)t$HUdgxYFbcG&r;mcv_`DBG8IXePQT
z=@ciq8S6=$Re>9IsM?bm6Ap7%Y##^dLfp<<9YvIzqZKW}{I#jT_cCW*O;wOqZ0d%^
zXB82tHAHPupIQnSX;beHtpa7-<Wf)}4RxLhm5SWLe&`{nCVvTstD{4x-fmsLIV16`
zRRra#S*&(L(o7P*wHxwj6kD>57=pkbt!T|6Bzq&;82VCRLo6zh44fYR^-<VFJ<m!8
zRV}#Jp(oGb5`XF3&_wzecDiA_FpiDAdHieAKFrhELllI@pn!cy70$Ix)}*;)nPywN
zT{72Z2Lt6$d9Ety_cIU=C#7dLu92+-?Bx2GmmVL#)h_<%iXcAo{{V$;Y2Fyq^zbB`
z61eKm(0&!y1!$v^wM({<J=cje>k;L`Jk}i+K7z5Xyg_|=kg|Cx=$utHVxiTNX|vxo
z(kxQ2W<7IWnWRr}sFBJ0#yw4AQdY7@O*(2xwrw=a8vN`R>sj}<@Z98*cY1?R(s~Lq
zaMc{Ht)<-}vqVIL+pTf(30F~yIT4kEjO2UN#tk!})KW)bt&G^lIt|sC3Nrw!@~?AU
z$l#8}sLW9OsJot~^ML#80siRgO_YRrUvC9sU+*#Iq5D(=?vpEenvt4XR+n|GI|zcJ
zDCLOaqSgEvaU>;YwGxBS$JUcXdq}BSm|7N^G|QRc2<jEF^9t-ONx(Sbxhi}|p-Jdz
z!g4BMQjG@5(zEVQ27;40`|-G%!<jY{LK8UMUJ(`Z4$aMV-w7^69w46Cn;vvo1L<9n
z*z>Eeb(Q`s*o$uyt-wF=^B5$0V~XL0HcdBpAnQdU#R8X_V+v7B1>2F*wtN~t;;z#_
z>)~h=m%7f^;^tpc(&5en1N`ec7TAip&IcG2nT0=r79)`7ARolmgfoCTfmI#InQ3lU
z``M|MT;jJz4C{2@he?E-WTCGR@gidC=;N)^{{TAiss8{LUcagAFX6|~OZfb)I?Q(w
zil=WT*#&C0EgdWq2eJ{7&U2c-r(1~$<F$2RVw}~^9DPR$Ev?2)AKm~~Oxad7y;Wjp
zHsY>kT}hq=QEKDrcJv}S&q|)gINorww$(lA(VVYg_wmM)6^VlZ+*L&w8Ko`9PVFm_
zzz*^Ct-zz~faB(_Y3^+kw)lRd>`s5$HPOZk&dhsPoh5p|Grk&AFg!=|X}9g)$~ENx
z<X5AHR^~YI4)!$k2|T@G1IWq6Yk1H7eYg9jyE+_C<3}}1nnr@g+hs%OMRn;oTXd06
zHN}cs9==<=k*RHV-dd{i%z9S&&@N|=H=aQnpW*LZFrw!4Jq$%i(n|eHsb+2B5P8qB
zM*td<+Sc24U3VXP^rH%rvXeZyRE$$`(&d>ND96eGrv0HvB%0}`M$F@>RdSW&xFtwl
zl@6b8okPzZaf4Thq`3@fRo>;>wr{l}cExB!pxY38gI;|n@3W$&l%i8~Xj^jfDeEAS
zao)S)uC7d(Lr@!3*PM94{{YdMf9asE8teDbHH-bzDCo9FQPFO#_ia1(kHBa8S6SeA
znLKR35>h)bzS+$brFyn7#4Av2+W4*i03x{m0Iq_(T~~~n?P~HkXg}SSBWwzjemJhU
zEexi`-I{_!#Z>kbHz~XFY6m&4CeFcvtlO?Rs3_5aazX1;OmgU>>TA%Vz0-pb&2J0g
z{6P$Upx3D@XHW@`yC3xws_P?_D<jIhUUTC6!RUXWu4uRKkJgoEXd|l+yYw!?b6uyy
zQ;kaaf73CK&b5}bkt<4)L>?}O$?**1`|7P&>@pwRQ6Gf$sk^}&nzmEIV6R^-YfHkH
z<}V^pdI6f?OZTNkUX+pOj;kzURpjG>Jt{7$SH3H*CyX)@9s?QY83waoD#RYfsr5Ic
z{o#byHR7@ILj8ka)^PqLPWZhU&1B$v8h&{mg1uWAIT9l8bw$Y}@rs*MG23ay)D?0z
zpJV7prE%h3mK>gk887<Np8V58b``=!%h54N!KbX7zO|E%rith{9w-UsMoOQZc^>rA
z6J`z3{rBK1J=KvRx{bnthT%)*0YHT^Q+Ln;d6d3o03Ky0nLr8VRF>BA-AGm?+)rO>
zP1#&&UD(k70EBk;W`0BMn$3GT;ka0&LafXH%}i#bRH?Y5y4M}-d@bdS2H3H;^{6}#
z5luRI4@YtOR^3idu+Bd1ztboG0IrSv>b48G3NlSWx-8qblK6bDsIHnlV!$9@J$iap
z6(ZipML8w8m#a&s%)9N}842eDtz=e^x>}jkwe$f*gQZ!R!YR0*4*SB;h<rL^{{Wtw
znEwDO;H2f)`<kb*oOfoAhaHB$W&tBISERM7tfKI#+*4>i)FNtbd6unr_RVT}l#;CY
z8T6=TotZ>^6?;^gt&WGlvXOf{Bkq*kYQKo43#;C37$A;+3U6W;#zUuTb7`#+U>+tJ
zD_()8Uc-BAlS0{2KPrZek)*BA)VZWGCl$%~sFk-1*uONu<Qy3pQ}U@`wew+@A9;Gv
zNfEaVZqjF(M3;TBjGkAarVMV28J}9VL)XmD<y|+82w$=Mx#hp5S3@`PqoJLnx$vAZ
zykT(?5<t&PS0!{+56d3-hgLzjeny)eJ(-6)#$1(Cu==PK3jnd4bTtK#FyI;!(wK=O
zIBZk4{`n94L;27LuXx_=ZM+y^BmH<-$G3m!TyKID`L{rG@~vCk=ZEmwFpE0f+_?TA
z1vd7`%Em?O>0Ff`ym!)G;wZJUk%n9v&bhW|n{cHUW?Z9mcfSm*%3ED(67jg~T#mz>
z*DJ5wj}Z9WM!&p?rV)*u)E>W;b@WG{{{Sx}&adK4)OJ1`w$sxfyRP4&kN&-R#jq?3
znu@y@hgueboCE1oV=Ic~I>#?zQb>M4C`(Z?4wRi7FzZ&(i5PjmFgSJSeQJ26;0EB<
zQd$(AqnG8}J5;HT{TO1h>RT~xtvsn&amQ+|VHMpMbYRfQBMK@4HuF>>?WxVUKIWdr
zx?`|}G}GzLP!W+$Gyw~Bs}?PZF-dMjj+(*v0<Xw7Eoso;m&9dGzd5Rs3Wje2w1S{&
zlXOJY9w;bjq#4ZsTwl69Xw3wHu7knSvTAP_=;%rODKKl@Wq%hd4Pm(9)p7Y&M98TU
zdVx}0J``31EuYGf)P|s;7Drwbnz?BE96Mm2e$)aMY5d6xH#n+}S;=G8q@+3+T36bl
zy+2E^X*?>CD(+^8fIWGsY*Sk^W5Zq@OKH++z-LjNy@grv1@7C#g4qj57dagUYcGZb
z{{R-Rk~33Axw>N|#G7;1Jery9Y&6*X#@M8P8`MG_S*=R9_hNC{pN(y!xOnW)%-P1{
z!q+!<tY7L3MDw=a?k@tnFQLsCTUIg^tr!(>Dn`z~!x~{=E*W`mx+}f4fo$L|_y-w2
z)z7?}H-7Ck9_CH{q31|(!j6Km<<zIPB*!5h-K!@jdz}gkmz9*}@q~8c7g3-0maW;`
zY5Kt59?A9)PDevsboE-C6`<hs9pW9!YXCvJE!whShIe0`glqlP<kslU8y2Fo50w(g
zp#K1LVy9g!!Sh9i^(U<jny85O(aHY4-B|mIxu<Cm-&`y(+l4*I9R*0_lvUZ<JKAU(
zXjA}TaHFdl&5OnM;Dx$@<0tr+kyLQTo0{e?*xl(m<Z^iiJC%*F*<OaUBC@rPFD+m|
zx#^0Qlhj<(kGpZg%G(@5gXk%zO}9V7N6^%`MVdzuQawl|(j$MnAz#+0-1vE%A1-m9
z_Ju>FZyamtjBBq6Uyw@46MF({7VpBExmMd~SdN4+sC49xh~lZseGV7IULV!to;>N(
zvFGlgKgyrr<-F2-Q)MN^w2-Sl#x8_l8rnM<)RbhMnbv$`vO{NKdIn$2G3VO^*OTq5
zXv&VTLAH|^sWq^ooGG9eZ6+8~0Lc}%;KQF6X}|T|XcWE{c0U*iT|-V~>JX9n))kHG
zOKlr5+SulkF^9m5Cgd+9ll7{z>+wOAPneHN5OQ6VCGj=m6XoG+sqpru*VanYef;2>
z(lU1@r6i-F=o)HF+I%~H?%pfMyhyBX?$L?)E?@cAjY+?4KVQ`JoxPNO2W#WYFh%{u
z99C3(5nWN(M%|paKr5lsr;TEDWyvEMuFM{jJlx7IR2oG~hvmYY3dy)^BO2DT)Vw;h
zRm_QwSkyZS<ku10D32ifQ+Aa+deXQwAq}2-QW4sl5=8dNmld@sU$TMI0<fOuvDoaq
z1I2?Mb6qNoB#t|tD~&7GI`8|V7W^&JUfcQ&r}D2MY{xUkcq+tKb|2%CIcx0}U(t>=
zdvE&aD{IDo>+QeYHQCVOe;PTg=8t+tf_DcM)yAQ;c;p$*d9F-S_fhHPuis?W(ll}-
zE=wr{3h9rEFXk&7>9cJE@2yT2e66b<jy{T-jIU;g9bWD7!q!vikyF^*-pE+Z0gQI%
zt$H5QT@mBdq`7sGJdJSgjO`xvJXXhz?oDqSy;<i}<ys=%$ksEv0AiC$=2s)8D3WbE
zlXR7Zxo`B-p2q^UVHtwFdTP^>Jqp(LQ6UF8R;tp*Vtvcfy$Ua(okFeu0Pih}4xm?;
zc*0CvM;~W7u53hYRwj^lq{+NK40bRsK>JCNdgHOhcHRYhchxTAjE0gRH=J}Q{{Yoo
zwBp@+p<*PK8LuKwisv2}UwK7%?l6E?tp#>@RNbkF4k_WY-nt@ZEk!MXP|Dm?)Or<3
zYEqE%OeVOF;ke`qyAbJ1s2W!5!1kj%q-N|Z)1<xBd6nP1&7T9u=lGOM+9CYw(v|-J
z;(p1K{-ZejMPn`QM6`B1zs0=x#{1Zh<SUtP`1xV)MQ?!}7;lOC7oj<9dRJNCJA)61
z#2|4rXFr8+4QLtE(z7gh+x|v2_l0W>c_4a<sC*}@P5X%7*U-n|Qm3k-kJhPpQgYV8
zbJL38f5kMU?@gYk;rp+&*@)cbljZy>M*=IBPvRoFZD}-k^3qKe;FD$Ck#mE{s`scC
z;GS@Kt9zPLe(_&9Ijm0-pnWxu_eeFAe~BMcy?iX>3I0^@-7D5P?n=`wow>Pe<Zz!_
z!PK?cFJEg6AUOiO>_olm4qKUzKbg)kQ(3_G5&)wnu%{O}SrJld5prdz&y|S#j<pKO
zid7-X>|HtYMKVQ~-hn>Tk;P(Jk$3*P@D&YY0%nQ^8Kk5jqL_k;LlZP+fs(~;(sF)P
ztM-?lVFh5uPf|rhmDP@jMMWtx^jh8KdaDnmEN}g#s#`-M1<ZitdvjT++DPiDPxmfm
z?;OvFZJ`f>Nn*aDt$1QnABEmQ)pwCyJx>?x94wL-k=*rSO+Ly+isfW+kiEU?9;n?t
zOVC+Ms!Z|%5FURTt#M~Hs}+_r$8l1muVUO?xrjv-W@3s)IWz!<l9(O!v~pSaG=sy4
zOAp7I^Y%2^oH~~@Nqn7V)<fDsUa#Uw7f$f7mv+;z;QNus>q6LypAtN5u`F2iCp4vr
zTvVH#KY`;TP?e8(@G8HH;_`KQ5B&52sq{1ZW-A=qqknYbw>&?rww*eMVqtC=dVA6|
zJq`($Sk=*kD-45HQL0M`4;)t~;-UM69<BW-1C+NIy^B19S{Bh5ZykCps2BrGNY9nL
zvn(ya&1V<b{d&_R)9|p};!A%_pX6(6#*VKwv{=bdPv=jeA8j2DmA3D~VTdb;S$H_;
zF<gxH_cqY9%1gvJC!zlU>(KTHXxWvNlpGP)kyWlfP*i%<bu=NyV)`0r;|8cijoF1>
z(;ydmm9!_Mk}CeR3G8}5#IN1>cF{d2QU3srn&$ir#b>%8;DuVZxrBd)Xw4Uuc{i&z
zX#wWAr|~n<r|_YdR$@fngP!$~EG65^+mcQIsWj+%71&P;cpoa;8DaTz{qx$mABR>(
z^Tk4P*!wye2=&h&*1NkS%1=p1pW=1*+BL|nRzS%Fe5mZluk@@NSn+RwJ62Uq=&MB&
zs%~CtQUF@PBcO~h2A=0CwCpCx%?(sY<m6+ar45d~Y8ooz)={5ourSe_R&F+Br*att
zb105Z9CT`<HM7Eb>FZlVv70JPq?`(8lf99BO>|R_G}0nY$+F`{vHYqqJt{W#uIkZI
zsVJ?kL>=OonWhuXWLV!?X`}=%-komSWgOs8M2?QZvyN+D_<GihV}gGXpK!dIqc2`+
zS-m}qvQ>5!2FQrh4Fw^D(hjr~TL<M7v=SX_U&GS0TD7{T8_v`8G=rlp_}LjQZROkY
zg(E)o%(FD?08)7AQdB+^PL>w|cN=*K^$Sr+5fAgJOn-P{glT5CAz%UHrA=>daddGE
zY(VO1Jp_hw7^XSS4NV=Ku72sMZpEc@r_g*gJ<|dgYJU@&^h*mLvDu_33oK)<D<yXB
zO<6C5vkOWK6NLwW1w%acwxr1`DzWd}(EQBOZQ3R-q2f#_TV==lO<Yd1sLyha^32W0
zb6Kf#Gp;n$72f9{PtBitw;S|6wbpW`TZ&v7*wJlYP`*f%e4~~=)e?4B1i6|=t!Vms
zTfnZ$1_#PLDm_yBaWNtJM_+p9dPLoO^*oV<Jh%s%!ql%8G-hMKtd^|l*yZDfeQNz<
zOyHl$SD;v1T4-8$&J<&l-OYE!Ssry~cVt$3`EQ>rtWX@P^u=3+WXJlF2iB93q$%lS
zTtXM`ZeOUWRD;Ukk7_xMc}JuTaVo}(xtQPz>MZSG(QFz?FdG?mAIh#L7A5Tk)a12Y
zPF+Ie$lM^|SI|>*?J`|jZ!HE8efRVf_bH`L2Tamwz8A9c&?;{Gsy_6MShm{r^~KA|
zCM%wbIssYKlUosorl)yG+qT!Fz!IQtJ64QZwX~Sr#R>HsQjJUQD!5j)iDf2^Rmw=D
z9+)*0k<J0~rWrrq8LfBN<k{Lt&YJI5wk4X@InaO4Ho-NMKaQ<z=4h{No&vvnC#6l~
zx*NunCcW7eZhj^BYB>V=^0s@dc@;c<FYu+%SsMD+ap)u*inPq-MsYSSZM6Mb$qmGD
zTwAc*rMi7<5xfiGDe@Cew=8?Boc{niWJYq1#&yTSeFAw4q?Ztn?xPHTwVQYFcG3R;
z<I_=i{{ZWv?xkZ@h?i3~=iu$?AvW5Ka<^d1wz8YySBcOQ<=Zg!o1f)HjaDK(O!<5b
z;#gdv(x!ZW5{!z^y3sW|77KG@G-UJz4F;-ldorAsvo9?RsP!sEYWOus@qU^A0A1FB
z%i(8Z@qhk8I)D9gZ|7VZ>Sj2CAmn$UkxSr2GhfXwSsN7qe3L<_;Bj47g{RC7A9X!x
zkyGJ4heWxIXqKelouAa#n)ss5IrY6w5pjo&$NUJ2@#8e>Nq<|Ne63T9FtwPbxV&-{
zkb{%ySt<x4n(B_iF55~$88y3WZ~_CKYo?63S%qlxM!_rOlb*Ggcai}0HFI|+nz2RW
z<eG@&Rt)JaExC>j#C*-b6<$cTPv!prbaVBr(;GmrJfwwO(m`y~Pvjq#PJkZtIcOFw
z!)qimvO18;aB5qLHpREnvu@>VcRvqu+BEJ*0!#|(v53`uPHPHk^_kG$?vaaNR&84T
zHXJqxuQ&0Ynl#OQ*%XN9`Nte*y>-#Dn=q@aQAbASSGE5Du7bJ`8Vr4%{-V2VaZBSz
zEt+^VC8?m{&sys2VU}$Xe5Kv(T=>=dwd!|e^5v?_dP_B}^io_$h{5M0rFT$knq+EQ
z<dEk85ymPoe3FZe^*XaPRmw7oy0I9#x0#s6qXh5;OXez!?fomzN;lB*XBpexD?78k
zLY!0l$+7araa%<z>T*=Qh-a2o0Z$pNT_HBY{{UL5mgUgGE#2IPRu3^|^{s1f^&_0u
zmpIy`&qh-H+C*6EOp$jf6{Ry-nezjNL(tcj_|yS?7q;r0QupG?r2WYv?}w5n)9jVi
znbp~pDe2EsU3bIF4<l67E>oDM*i@7C&2dnd?^<KVNB0*Q@qSI7AGcrcOI}JCGW%Dr
zO!8y!A}7wx!jd_CYewueu8Ge|t21J=s`NFaip*)X%5@Hna@ni4>y{MKy9cqPpCm@B
z%>G<n=DjuU!1879OzJ)yN1aMz)a_IH*L4GL@f6wZ5g(|mBocR%JkP~PJ}=aLoPQx)
z!yojLYHR!vI&l906EE8wn(BNKLincNJ<-3FbW#2TF|Yh%U&prUnw`AE?@1X4@UCVq
zT|Tu`{u9)t{4{j_DgbB_DDQwPoA9jLWq9-7rEpJ4qgo%sXQOC0pJu*HpOyYoT`r`X
zWrk($cAuqo)K;2(&nB<NXUyJR)N*tOgHF^Fax2n(0uNMY=K~dw;x*2dE)Q`~m+u#;
z=tcXi;=!q5QtSZlUbE2{Lsw&4#F|=K_*}^<5hl>++3)FIZgNfyeEO^HedBtt<dZ~(
zDI?pxO77%jb*(E2n$Zk{sTdX4hrB0##|9$Pr)AWyB-_}Ma!oTe6V=?_mWsi<tu(Rc
zn&y$YQt$nD;3^YY2t!Fg!iogKl7R&+A)zW*A}+#)Tvs%bEXsB!DiAVhEsg7ZY9*-h
zg17uJ3QwsS8FS^KAo?23d`E2Eu=0!SW%$zNX4XVtW640L`PW0GM7G`$KkdZL`HJ0c
zcu(O+pHq-}b*VKiZMwGF*=YB22sHDlU6tDI9eo%In<uGvanP<<u_W>ADJqBz;8SUU
zQh~UDJJ7)Ib?KVh!i(6nXov9?=f;~er?GEGw>Mh7tO{^mQO$a1#MgB2<S{NbnFjBt
zY8J%L<42Q@ywvFmADed`hNRf(d?hEBsGvO`AbwQ`j(_AaAN_J4%+u&c+c01|<9Pjf
z6p1DV4{Al7Plh~4_GVi<w?0w(jDLjsS8*K5R5F}{(2ACdL(G7R=DbihMxb`&&@+@p
zDRUEgHfw76L^jafahUQBdx~N?OPiBC?Jq)2SI$W^$v1o)g|+yb`>IdrSA1aS`&u;X
z`?37#^ds$^SBDZ$_(g<eKqXtNdgi#jOHrK1_H!X9r2MLcQ1%FJ%x-t?Lcxz>DU%$K
zDoq5Bm1U-?<PqFe2znjJMQQkF{zfYW>-+WobafNh?fxrV?6tAC{_~goxc;??;B%R-
zebs87<}v;mwW$rdgOBz{@~WY}YT}>7dJy<gc0N*#KS&qa-MaD@e&!?Fnw7naleAkV
z_^qgHw70O1L6>6EC$Q=I=A!spVq&nny?DU$Tq-|&4z=F5Jg222gx4;PrQ%!5g1Fr&
z<LSrrsUna^3w>)D>UBdzOAM;WRCObYtu5{{r>#O-mf2VBhjm@Ixbq@B^fgTi<iO_?
zHAui5Q<;MgdQA@FD#SDSY3e;HTDa37xRG4tc&M62Wx--<qO>KE6%^7iLk^<0jGS6%
zmaeQW_I#G=O(8YcDY?F9)RehZhGf%EN|B8u(@CIYZOEpf!wN*Rpwrlb3FC~4+l=l?
z)f^RmBDK)#S(1vG8dEB{s_js8A~I?=ps}Bg0DRuG0cdpMfk1`oYWAHWx4VLHx!B`0
zn9=B3pM@V^iVvG?LH85tF<QEJgS727B@x>rFduowO>x(pugZ?>93r92%g<#LYl9l&
z=~*}WW}k8j&21ZJ{hn)%?k_{qNl9vRS6&9Od3@V;Wnb=RgIT(a@@a}}?X1~k=m=gb
zsuf)HXP;VvhOsg&G@FmKi7ughG0y{>*Lk7%Ij#!LaTX)SReERZR{iIp&nbI2>$%u1
zrL~Mom;`}<tH2d3_s~J5yseHDoGodhIIStPW+mH6V%HmX{H1_CmB-I=(mOk`B%JbV
zo};~wLWFGDn|CoFs5OVFpWGGhY13Mo&FEnm%}p3P2eoz_HZ594^GvazEX;5KuUYVa
zhZf4&WR-HqmM6LNtZH;c$3*0G8ja|Q3^AXZzB^YvaXvH6b4}fy6k}-I$i2D>q?W)H
z$zDr7t$en&z_v~}{&iH9*;O5)cA74W7LR6K@LY_$`u!=g#U=f~l0bHj6@4pul{r<8
zS`^&m<jre6H)@Q41z!+|`U>OcZJvszkm7?F#aDY^&GKY1>5A4eGid2FeK+hdeWK`t
zCNrPzip19ShQ5%;6Mo?j7(IRKuKcMZ%Yyc`T>2u8ou*Bx+-#4C$KP|%S5<LwW8uv=
zpin;Qj;?Cf#&z42lo96MBGje5)F9MlbCBD1j({H3p<yoBhdeN>Djjsvin2El@C{Ye
zb&Fjx9rrT@^j_!itm3X*naN-LK%Gu3P^Y7@<W(zwiC!ePasHGfP1$912R_-ZsH+^c
z5}xNtr+izDff|*ADagTu<bFA>mfOXC7uONuYnT{g=Uau^5B>GltmKs`M^r(7;2#ZY
z*_AEUHf8x*2f!Zq99J=@d?T4KblpK%W05ey{4q+!Lc3Z>>+P&Bhk;I=bpT7GwlM_^
z4sb9J73JD@hkRY;6K4g~EI7%1pU#4*$}0CY;_&|f#Aw(u%{d_F=DJh;Yi8YkCbOzY
z_VFP8CJ~?IMxiNDXG3Wpi{giE#JY6v{o)yam1tRiZ280O!pudEe4^R?X_!7vhK(VO
z8>*@LiU=T|PqMi}w-&I-zKnp?NVFXi3DRh^`$(iim1T?q4KkFa-1D!DBppt}YXUr`
z<b7++SYXm;buWPpB+`RO&|TH9;r&5vZ#Kz*jxb5>Oj6kPeHiCPwjhiY{<Y3)$gud0
z2H*Jk5rg<OcrbRaFPZ31_%Zn$r-?N+hG{Hd10}zBM^b$&pULAq8uVcsMCVTA$mcn&
zD{1)aTA@j3TJ!VXp_q-Lt+O<&Qh#cnPt)4&@e>>qS<Tuln@3k-^x`3?I;qT7%h%Pa
zEjr0=T1h1!cjmd(`EF~Ym5Z8vy}DkHI|4lORkfwHGI`dgY^d2%D=&MW$gMdFi>aC0
z%#eJ<VZOCq=1@U3s*Q}>x#`{uwKm=uhmK`CzdGJ>0YB`mV@u<cIw$ZXU|5}+yEy44
zYct{=lrHp-wW|K|*ylg&p1;zyi?u^JE5P&Z5aU+0&;0Zi)A-RdX#;<^{VUT)oiB|X
zeh4(H^{Ys>$w4N)6T&tVOQppyT%gb6Sz@A`5;dVNSvF|>niCVWtCm*46^W}{S=-AL
zGTWb+=OB@a@?0$>qI(MwE?3R5)#!Szp0)_74qbu4$6B=dlstj}uVR~rD%A2SI+C25
z)Jbn+5#%rFRn7heGtG2C!Mj}KsHT?0Nk5x6&(pPUSp1e{zNWRRv^io@k+81E(3-uX
z-2I}(3F<hm3aj3++b@FtWW&7}1C~5iaA#bRM)2<B4h4BXj<PjizN0mzb?(H<58@+&
zS?u*4E;uA*42!%0)DxQTJ{H}XPN#0b<|%WZW1cIHlvg@oij(#&e_aNAVdqD(Us+G`
zuOJS{xHafr%IBL&UN$C(Z-|Pd`=eqi=kTUoJtWa?Id?`X3At*@Zs^_$#d-~7YE2lz
z)%#Vzs-~|ZChZdBoFE>xMsj7GAm}U7m&0SrhsATh_;YTE(#?;U{J({CnsmN*h_8Sg
zL;%mOd9He!iFG{N#e8_HQ*ZicKaj3rkMD`7ukgg|!|u%sO6WW{FO%Xc794JrFXvrU
zHKKFkKa6!AKN-2PJ+`%SaSz@{{`FNag!O3u00|v8iiUKgU+*<>x-0#kR@<Bgt~#;V
znqmF}&}|%irn|)}d#5iwK;&0dUxqC5t9(XP`r<9es<liAxU$Z2cGSI)mqunVtbY*<
z=`#KmKW%!Q@qX#WW8a$5(_#&9rMUptp@^3)HpM|+XzKN}k6F-VwA+)nEA<uSmO?8%
zKrHcjflksGbT#vNZjx$U8PS<W@w|GX&xwlty%k0PL4oO6JKT-Az^`Vr{p_rG_+|Tu
z++DaHw1?|l@;e|VNT!<SX2^!<e*5qh2o;EgD@NKw7i}RF2ux9m08CQifsT$n?geQ<
z62_aGiq3DL-Gio_xt~Hd0$1a`O18*@C#_49O2<?v#-yCt3pw)yaRTng01oE6jT#8_
z4KUnJ6=d34Lf=l6tQS+x##Gml9(}6b9Yax?>lohLeJirKm7(xLGWGIQesxIVn%Nu`
zS+y#Yp>Q_moYcLIR_9T&(*7STvNs)%O2O1L>1>?;02D;<eJcuWr)F<aCa#Jmis(7O
zno4K_r?UXwLJuIwzs`U=FB;6T=<&ghWO0wexG*zLMs)rYcfrOe{6h(tb~}Ex-F!v{
zO$T@fX$z0cP_`m}8a$;FFN46T*#TjO`9(>&(fC|vR=57{Px7gJduT=FwEqC}&=ntt
z{{XI<*oX0>ky2uE2*Kz@1nxN$7B~j$&2$<kiY2hy1<quHrhBax8r<!y@2#)m5?f5J
zJAio~TFll_q<36rDaA=gBQX%L$@$dsD!uNfEz>Ho#ABk<C7JDw#+PX7C=t_m5f$~B
zk5I7Weqxu#pDwtW1P>@<^QX|y?VTrsE`#Y3NAqDD86vptN=0j_HnT9?dWuhCqwzAC
zbPCxV<BlqeoV6rq#f;)#tt^9VnWjhqf_>|t@I9cs@l~p~0SS}x=9(s73pziF;54=a
zzSjO?vOEUap5S@`P3~tO;hhGqzwt6}_eb)oLNmC1YmQr;@c2=uZwT4VbY2#=Q*ku?
zk#_!n{;G<4)InL=lRV$z1^K@5Ovo5ALIi3@`Sq^C#7CigIrA8@7Pl;YLpT2bUc1)p
z^QG{`9#0E=&m$bM??}@*`A15Fs92<!5LA(aQcSHo?<<0O(&_{c8D65KwRTA{*!7?d
zL<K<frxRI}B_=2V!=*Hw14szS=8~8VrkbmWPShIfkHUm7(@iiN->osJ0NhhLH3DX}
zjMxe5T^+oE&{mN+Yi>t$I5n3r#%V}&%5n;>)dx}{YGEDdD2&tN>p&9(E7PSf)KDe>
zd)GtZT|Kos<%Eyjj$G7D**0A@%_NU+wTEqrA}Ajy#Uz&UV}dJ|QF7Gir&ch!WjU;+
zVfTQn>2#KM=3IUilcIDiT3a9*IUn-dKkyoP@Xy+CUB}d|B}ZMwENV+sT(j`ym67uq
zM6!Q*x(}sKb#jp@NZSY4S5#7jY<X2DDzoNh71R=3k`2Qup3B8YaR`S;mK9<`p&2Kd
zi*+|r*$`@OEp%TjHV)9B`qv>WqaAB0?sZW3TPR#kstF`kOwxU%?Os4Ntu#7}XtK!U
zb4d1NFUko$fUbj-?)(+u0~t%Zwf*Z1fPaL4jd#~GrH!~m131Ep=5oE{TaKy+J##rE
zu{kEOuI_>Y#efygcI=K){{T^z`s7;L10ylO3tgXvyfYq`Yv)`5W?`3a>0L1QRn97{
z$?~I8OJ}@sxNM1j>eaYyV~e7S;+))_&rX#kPAL`@K%|qUBsNwrirUe1!1%hHa6aOn
z=Up^cG>$A>{hc@6ka&w$EKQ^j@(;d8e`>+gbc>7ovE&kw=l6v4#c1T^Qpo406-I|M
zI@#`Qv<m>L066(jbz@u>si|CQH!U>npGE8ibvMxEp$6|#Mi#6?X|nA+kC=WHqo&Rz
zwt`S|z~-`%(u?}fh$7M~H}KOi{o2=@-QQelH%8*>Ld?hT6zRvQH6KC=2*}4AR9LNy
zfSgiLGVPYRs%f(MmUq)c*v3O=kH)%fH{zzXYRI>V72HZNc*7sYf+b2aI%{8xnnPQL
zyS1JvNx?(`vG~;A@U}D_(A^*S0RI3QcQE?AikE*EEMp(&wi3#J+ikT-{A0)b!T$jA
zFaH2qi;ZF5R#v(Azj)|Hwx1kL*t2e|@Az8&B=ofZ0KZ54Xf0vm)MURDd`T*i=0P92
z$AkRqIQZ+t<dgleIL~2^@SwFu3!xkM-^8v8zOQgW_aEU|vuc`z@);z&nV9uE%>}4M
zY=~r<S%@;o?~a?BH57}*1)>%hpj3!a%{d3m4@zKruZ107!uAcgwqxpkrm?(DByo77
zOlbyHxcqCyQT_D3XI!QGw2ls2LPgG_FU(BLsh_Q85_6jLBFv&lha^?FqyW_v$<|L=
zN0V@(<k)qktKCc$MoFzH39t^){JRFwed^qA3z@2kPn2b)le{5te92DZPHQWO50<(7
zDXJ~T=^4CC6uDa=k_prjaxqx%D;8{%SxTFZ=;)_XGuAHT40f?$->oQe<y7}IHZ1oI
zCLsokX&B%vYy7G6zT0<lde<dAE_F|#j}&DdIXi<|T8skkPntcUi^f0)*AzGJ0$Lt3
zV2y38#*f2CHXG^<b$&4CO@RLZ(@*JMyiY2B8aVuCtszVb*%Z+WT$Lnk{A;oBR<CnA
zE}o%|)3`G91KznYPVn?M!|6X$yYeK38!>O(Gx*kL*dn}>b<MkOUYu8#X*)C1=8c)T
zrKPQ{gruZ#It(CS2|r55zmLyHal>aL6`bbQf~OUB#~Y>o%|>px+g08F0DPy^R=9b!
zQ8Fte^e#d(is_?Zns`6#*V9i_kDO}Kv5ToovJH#(pNv+fh@MTFq3SEoo4r@$j@p`2
zitfuQ{f}B~ft#;NRJ*gLp;}?`kzQZp=ttV2lzrosHKlKb8AGXKNx6k}1%<+I7V)qw
zGDz4sAB}aN4K&ed{h@TT2|V{4FF}mgpGxYdsPT4m8DHZ0{{V??fc2HXm3a)&o=C4n
z{3oA7Vb-nHX_sl<o$pN1D^z88oUau2ITeFvLLxMsNEC)yQP5V6vh=N}E6B>Tv<WAD
zoc{oHRagdc=czT{EzdGKMV^cB`3Sa^yKX1+t#1z<+TN<iu4Dd2x$15a*`9Uc$v1e9
zQEyp)psr~D0IjU6Z4<i=w=}H=6jt7ZKjM2~?QoyYyKw&i!Z|TN#XDafLJg&~{{Y>S
zZ|7WG0GVSS?FuP;C#OOF?CJbb{{S8lf4}`Jh?_R|vC2Qa&1F(Ih0SO`bna{>4-z(V
z2(Ifym>Wb)@K^P%XRRc8)z`eqbr^FK5Hq=dTBoKqV`$HAm6Fy5arjEt39?pRxU5}#
zXG)vtOYOw&r>ssS514n!t$jUq0(x;>xTpKJYlm9g*R!9?d=rpsmGNGg_6<rq#Y5#t
zcOOjG&roWfBh=}{$Da?ULmK`=s?Kgzc9Xy!)fKE(ra>LhC^OVouTF8PR`S^Kr%rUJ
zB^}WF+Z%Y88+rGs_o_QUCnq%V5R4^gy#<A-Dv`XEn8QF`=cRBnxe(ls-u@L9D<BL}
zibD)hwty3DF4_P#QL=y}iFSrQ)cGeRLi<!&?CXr&DcK7dkYMzuF?kL*8dg`ix^i&k
zT~43Ex*%rzbm@?%_j&GpYQKy%k2X`M!HtXe*dC|Sw2sFeWqOwEasYhcPpx;l`X#T0
z3;^&IR8vPJ_eTJuClzBV<avZDj1DSUss{!p{nM61(1Sxt?txj^hoLkHk(yZ4JAm8+
z)|f?IlD)1E$iam~QJ3#26q7coMZ;}@Mr%tzfo9h2k%mU<XFJ&4@zbxDYaj6jC-kmH
z9CMnf6Ay&DV{!Id*k?=zM+YXoSH-a{*MfZH`NGDoG2W8uO-&^hc)K}fEz>lij}(nf
zts4IT?Dv-YPbI?sX1YHgb8lkf_*uWDHL()-(aj48wKz<Wuvv#u>r@ufc$+dr+fPG|
z4G7Vbi2DfJ@~c+bJ>9_`U9)XJ#76>|Cgz>txveD=%PS_|;f&+;tsDEOXG4hsv}X;^
zBQ-7Dw2n(s)85u%`-`LU8nWe#`Vl~w#Xx@X&OIn)Z0!n0dQc;!@E&8;u6?2YqN#YA
z;!BSZ&I!Pb?F7^4Lio<3z~%nKaKC8FZ|V3~Ev|;Q)n|eB5UfVzq3kI9Y@r}n?Z({Y
zkF8QCCZ!s&#NmOdr1_K@VjJ4JuZIdxiy=7RM)~}yt;wIl%U=?%OZbIt$>KPH1JgB6
z!2bXx{lNb6ihW35!#ZtPXH#$Y7x`6nUz}GI_h)rKg%{DVmDXuP`#;0dT(&c{r26s2
z6c@fmDgN%qiCd%U`uZsOUgmuB+a|pW;#JwT@Z?sA$_(xmm;I1G#8+b3oOAf1k1A1x
zqXMMdwQg4<HFaY`Gg8VcTtYd;RgHGNF&dD*P%Sv(prI)!0H&HiHjs({ZYc<$2I7#{
zWIhx~!%e0GO*GR1r}C)JT0t#~`g?GGwb0r*9Ad5~6=Y8^${s5+V0jp#MAIosFe<cz
z2B7OiE#8>lodt|E=)UzK7C_A*2AG1qC*jtfBh%jLhcdSxO2W6oZ-@BqdOfYc+9Pf-
zIH%k^h`W(PW357Z3Q$URdov<kPGb8~Fvrklgg25L$f%3@)Hd0yslpqw(oN-#6lc`b
z-)|mR!sWfiT6Q^Q4q~%UB#*hJJ8}1WR1@4=C?ZE(`gf>q+OufGo85{rZ?zqPzRMow
z>|3Q`+v-}4&ZLmcKX%RYC>&z4_Lp`?N7d9K<i_~7Q&>_Peg=MQ*BNoDFF!D?d2b>q
z%Fr@yZ{A~oc&1sqo1_8plg>?aC1hhM-JMak$l|&W4fu3VI^9f}=l5!Von=y1MbnMl
z-$T$9VzIW*Ob&g`VO`ubu3gX5+OV6lv85Hv+}VwLaugGq<TX3#@23`afVpFjlyXf(
z71^}ik=l4q!V&3E3#+vaa>FE_;U1OHzwb@}!VZF}w5+*I@UZ5eJLo+k0-jd_xl%nA
z$jQY-?@bWPT27tz+5E`?@=){dSGC<)Y@z!ty8$T}IQ|h`wEfC%UC$=B@39IU5rL;^
zvubnkkr+Qa8tHDXtn>>2ssK3oSGyW?mn>Y_7^ymwRHAwAuYGfWb@rJgJICD}fb^<~
za4s^eHpn`US|K~VPKN~;*%dC~EfcR$!l3Zg*866*WAjAVt(k$1uS+btw1P$?CIW-$
zS-HWjBe9d*apY4mR3Vw8ngBv7HXf7|h7s0;Y*0}0{8JA!0GOqw0@C)N1|GDcfET3)
zkwDPKOpMfcq$Vg8cG4A-sLeD7wD?g1cxJ>7o?rDff#b;**HY5%2;pVsyoj&gKbf=m
z5rL@Nh`cWg;DYkA5$<c1^{%RW0v)>@Dp;S~Rk4>5a=!Gq%}vB#({7`+4%}|X2BNvX
zTbq@PoJAik?rUy;bCl=yf4p)pX}Xz??9SNF@Pp6_*|*eT7d~`~8$D~g4_R8w@-g^F
zCgi&kOE#S(GB)3tff&zEdh_e7EySBnM&pc9#<bL09!(j=&NA{VG54i};#6PuhAYoL
zfcI-wKVXB?%OA?4y~&6~JJ%M+PUB6kBgder;C=Am^%b0**`it=N3YrqKgBbmoR2qn
z{CWPhx#H3emv58*04)dduVS0LQ#`7w)e*^U9YLs_vPc!9c0(guW}Zbz)Rtp_J?qvy
zC*j4=wAC^%`b@(vPj71A#LL~W(}$0}bEZR-A4+%%7o}phHD*QXw6|l7g%}kNmU#qX
zxa#~zWj}=?joc2Fy9eFhpVq3+aHDR}Hzy21>snCNm|`C)xuIonvIQX5QFT0USq%FD
zUrElrnrDMGdbUPX7VmLC+GZG0_d%^qJ-3MuZQNHqI6hY=)Y7#@DWtYSz<lk)IAuSb
zO}a!Ju0>32-zz((bgn@wM<_dfR`fOI-aLqlh>-VwKN@vYP_r_FP3U9V9o{Ka&O~u+
z{F4>vUkXG20ApPea;D+2T=dqaSf~58RrtfrhMoTaU0eBAl@X9dbv=%13*9RW#VGWt
zXB(3>7x!#*+gp?z6>4}n6`5iYNgCz8lmTWj$gNvd<2BQUSCP$$c7a;@PD=WkNJq*m
zt^7w7tpn6N467cFWa;vfe>$7O@(DG4T!WlQgY-4eJ;fcF<~|~u%+_z8J!UnIe)z0w
zbUg|SLr&N33evQdpty=;;gN$1^f2GUI5A(vJp;zXwt;Tl%!+?X;_OEIxJO^!3MqUi
zqe1@c=)6*Y&|&`ouAkDly+I2`7<C-dlD#HrMd)-M8@=<b6-min4SE)fAr^{3$p*8G
zxv7}qV`G1dn=)6E+ZBgoe5=+&(<D^GW{yZ~(LMBdkUwH82VIYTnJ@#O%^g!aD7&ME
zA1MdW*1ea{89uewiug{NIwoF<GrSSdR?ekk7O9|17nU8>30#hbzG|$!omkEEXLMuF
zN>Mz%I9ep~*?wf}?N;Hkf<##)BVn(915PuQ<b3X9R#f1WOx;G_B$W9F3+bBABNaH5
zqc)bq!Z#{uy~gIAid}%}Nv;NWBE84@-@>6Mn#={w8KI0bM(RKtsNFyY>NijVmOGV4
zPW56aMt8~h3Wn@<P>Phj?T|BuZndeQ>3`ZciylZ`tbGMdx*as<9%nAcU9I24qUnMd
zzWXlJ9-j5*lP{RX&m^4HUsDc6C||297#K*n9Fi-t@skzNAz$w}2lA~E=hlOa(q|WJ
zn~7uErM9;h7eXD_-QK0U8}j>zE^VZ;l%s=z`qWIIjmy%6?yh3hn93B8X^60rdWy1%
z<_uz?yB?(*)Q>0|c+cJ&PpGYb4QZh3G0Lxz8kV9m)XugQdpSvNhIrp`)gu|m5r10d
z(}7wu&Vu(*G!eAZENjVQUejy^?|}ZwNZ<BmjQe0<Q5UU@>uj0f(#PdW@wZhtrcyO5
z*f;!4EwldsJy-eHZ{v3V0EtXadD{5=>XycG_>skF_s?Z;9CD|quCq#+SzK^N?mm?!
z$kkGVqL0d)C1m@whhdj`zu9c$k(VS62dS<`)=93`_&Gb7e9y!j&;*wEk9T13Og?oS
zW}9hpv8t@f88gWvr2tJ3g5L5*D55z5EBz}zQRcHqcw%_=6bYN234{G6=5E1u{<V|h
zD46(y>5oGE55QG)F<%=y9|P%)6`S11Fp}`2AFXlzDB1nHbm(%d{BuLtBK`Jc$oYvu
z?NHJ+VxB;$V>^_s0u!c1cRveY5^Fay9<f8m;&DxGMNfs4z9gUcs~nEeB}d@4&Z6VF
zT75}h!#aImjZwecU*%9IT-OZuXMI0~7T{KzREarfU<GgZ>UGon8!Rd|?@<`XYN{G0
z%&|JkF+B6(?T_00NU}6r<ADnk{nOg4{7k+V8ss*cPx?H3pIq}@oeoXv9CJsN;-O&6
zP|d_wbvy%BlJKa(q{b^<0x?jR1kePjakIThZt^KKJCBNdp0#8UMJ6ybn5hA#iU2|>
zq)-HH!KRw*kHU##4J9#*G}BB3H5*bBEZIIBk6P&L$tn$8PD;qU!z)?0OY;*>Mo`?(
zl=iB!R!3SOCYZ*8#%8619OsPC1%c9yrkIT#FG0W7Wf9s%yVZ{XeQVZyJEom8ONP;z
zS)&*wdJ|mrc1J}G8#fhr%};MW`2;WDp{R!C)>n=D50=U(-*Nz;^*E~X6kdZq=Co-e
ziZYb<GwyHg;Qs(wDEbP=x$zQ=h-1&YRx*-DZ8|Pmy^P7Vd#R7jCI_PNNfhMuIW^5I
zv%Z{EG(^S23b&+NEVnHo>Z^~=h%b2scl)+-{xj1W_WjxkNGPBI+upbft9k9DDKd~n
zb;3!z920|3L=*x+$gR#(weW*atIT4Ra*+YZ9eu0TY%JEw-*aJ3IrbIJKabGqy{Y-K
zjqTC6bC0^cD-QDKVJn*Dl&;RWLzXr<JzD-*<`$OOc^u;cyAKHXMReF!^6&<^VV-}2
zwT6o36`k}tIao+`C)^*c3l(!o^GfG+ID0rGIIaa&cPmATxf4A}Woae-$c;hRKJ$7C
zxW}4SGO1BQ4ouwCZ=lhvL~-MOFpK?bnXuFDHJigDk`SD`n(C?VNkJZ62khy}w6|wT
zd3kA|*n~hgW93oQn&L07roWx0nX*qo*il+m_fYU~ZuMomGv(&9t|LErNmd;QtfK68
z$qcu<7evJ=3n2Y#7S?lNuSmJV$ZM@TqmD|<lf(<U<t7WmFJD^4In8Muj3v0`G}bjj
z8P=WK6u?3$q;;UAMEI!seP}2HQ~J;VI24$mW3xz2F#+UKPyxD))CI&Xjwu!=Z{ef_
zg4$0?xoV8AIK={IviNI{4cL$RXZ=lcKNI-2vTyomf0cPr{x{6s{3cDr=I6kWJ7l<>
zvd6u0U{_2Pt_V^wPmAyM;;oFooG8TxroclZd3kK(6%2AH;GRz=n2F+?mic-Lw6|`Z
zdQ(zpnME``^-tZNF_BoZ9Fk8`00lw%zG|gDVN!yH$ju~Y_sRb7p&B0Vrztj2f7dbm
zsgwTzu8e1^SDz=L*%hgz(Pl23;;5k6Asvllb3@JkAVRC+I30Sr0&(@LUMq+;lW&~x
z<M~&!LH_L?Re$1~d5Lku5l!+vBtMl~1FZOaqJL;>Uuc)+irePt-?e(B)I4q#$yHxd
zs+N`*s6eJlcBN-+3hF~@QDzSy$oH%53>O(VR?mL*Ek&y`XU`caB9+dHCW-vE#z??E
z)y3=DW!8_W3z*|~hEbUS85ykT!;+_wRVpg9?DsS@!wtmH@vyNO2D%+W;uxci56s!f
z741_>Ry<nJsTo+N*HTI`jjQQenn}zr^{&ar(6N+Q$WF_n=uQn-W@uZ7Bp$}N?76i)
zeru#ba(vN{Hob6OJ6Q|a$XDiBLZ5zW>Mz}T9PxIjVA$aJqIdK%W1nMQlkl-e{39~A
zOx;QSD~_Jz;-9!@d|&3vPX7R|jr^<1^VYU(%8$T`^9<0BxptnSvu8>nnKR8oryNwG
znI<wR$hZgIm=nS>)u|BLnXb$#)w6>b^hnIOefK!@s}SRDYZrMLO?#;IFA74?Xbcbh
z^UvqaL*S432y2J_o`1%(p5}4co?WWBI)&DM=ez#^GgPzlEpya*o|P43YFhsQs}+7B
zmLFR5u)hK3;{O1MdRL0068KoZ_X04k8%pO>w;BHcmZGMg!p`~+_h#>km<woEzWn}`
z%G^h~A`NL3dS<lUk|Xf-$+py5B6x0<>Yg8ybWStrTvVQo6^UN$&HWm0-c4-BBpi>W
zQq*EL`jKF`Q-Uj1q}k`k;w{{-nQgn!X0h&|6T<Ny;sjQ5(H&TQ&pw(KhUubkL09Qo
zmN>x`*In~M=Ujc%%@~F`9cyb%x$@T!(!O^W8>6!h&IiPsY&H2#)hPV%J!=+Al7Ol<
zV}9P1>tSo!j8{{~#bNoYO8bA2MrkrlKu@JQlIC{`aphJSpIVj&DB*`tYn`KHinl-O
ze+q`N0Vtu2DSJ`?q@V(d0G2W`4|;pJ!R~5$oiTpl67s^k&kfIIXQ92vm;<g%nd`+q
zrg6U`vB_%soZ6lHOxY`(DD)LxBIUkdaqm#hnq1t;7A=)#)a_!=4ImZ0<9A=|`*?qd
z$&fv<R}X~d#yi(mIcr${LOYW|LX&@Wf#B5NQ=!82;$LgHZsHJLuGhvEO3Rwv?PMe!
z$?7^9P`c3OsJ`)-56YjYsSUny$2E7cFoTxGyLbYy-N>ym1Zsn57_27l&!E9mQGr5q
zCC&~2u8+a-7Pyu}(Nk~dR~^yKRizoDBgBQ9Uz{F7fCuGRbE>QSyl1st{o{gAN`hvd
zh=~}I@=v=M<Lh3B;9Iu6(4<#6G8aFos)xjR^}Z9$yisZ`JV9`n0bK6@b_TLrZts^k
z_omBE=$hO#S24am>jo9uc(|E7I~#XqLqgbzZjLoA{{Ye3a^aNqtxYpfh85cRQ|n0C
z8_?W|6f&@3?N(CY4j5FHSJogmx^qVcH-ZTFtj!t$98#H=5@%`8r2^_U;?M^^UryBw
zHp?KA2?hx{27oF`7^${#^QrkwLn2N-Nh2ydk=~e^I-dm<Xs(qP05~6|P?F8He-=v$
z=U{)Bj=)uPFj`5ov(;{sOwvRUovv4E`wH^={{Ws=bIwmRp5vypxlR7)^rlc)<@=_V
zHrn4>U`XSTYrgnYj5Q1Q$IIm?&*MVe&Zof6d&W|{{wmZroa~gS`U-!75ByI=f8Ik!
zQT>Ci)`$E}oBh@PRSFI(hNr}K()dxW;j3eDtH+hf$MX~E=}~-7y4`FoRG*PIFzwGu
z(w>m{nAFy?@<mUD_KO{j<@qERV3@~Y&-Jb^#CNHCuHIb--y2AFA8ysw_Bms$+cSiU
z-6~HcFb1KUp$r0`cts(QvTdg3fC@8A2ctJ(S74R6;-!ri$23$7LMaHK1jRIp0FC%H
z8Rokq@SzMnXf(iKrkVkbG}A#N?an0}*F$3Mf!tL@=c=?dLCSMk_hT8Vdl*A=E?)Ij
z_%)H%h*;ETnhOP7pRHL`!hww9fFB$ZO(00YW+wm+_1^e@z)BG=bwXK)$cS_wTF$Kb
zR)&<J%<lF&eNH%ZyEtuakVLYL&ST(_*jBEbF>4?DqO+3sQKTQd%j1=()%?4JZ$naz
zUa6x-@9v7^wHv9ovX?jqwK}bw*hr_jgK;6|v))6KT-H51iMpnZ$fU-ZVP&mtD*oQy
zDFm4ePY2eB=_`{}IYP+EykR`jNsqiqtX)S>is~}1y!JHHv$))E8ys$-r^#hD*5yw_
zRpN$8rC6g<&c^@>?ov%$^Ct$SW7&KQp+}?IDou|$#^O(}rEh8%Y;zKQ(bW3a6sL6G
zbFws_I!|Pbc`iJ*BL=xWds1lTDHzO*a0OtsXH2w5L*SnZq%nVLUk%g0Dj0OGp5lKn
z`EmWx*WRvjw0%q|N1Cp^2S&IRCjzoa>uibMs><S-Br!!L+~OFQEPqPrLATJf?70Mi
z&#M~hpsjhc%*M^h#*})T^}V&vh%efCGO*{9=sg8(-{0D379u#*Z^y`ksrRiZ=$%Gk
zU*D8y$7FFgchcWnG}AFGe@d=vnccgy(TY(?A{i=-{ppm>e*M}hgotp!@ru>6&<?2a
zUe2u;10Z_V{QRir<mQ!${7GOFObnx#$gVWjlx29BSC1@&6phVnU;%MUOa$#sVL;Il
z8Kjy%0}g~xUqT9cngDP)>DH9f7=o682zaFLLI47dxuylHD=#_uBZ}FJNm9H>KPpT%
zB^pnpl%q>4&ivB34D{^|KPSWX9rG9UHOzcX8?^gIJ)}R%ytw}W8|HNVgDUM2J|9_y
z>R)jmTI5u7(z+?Tr|JlE0sb_AY$CNNwv&PYYIEJ7Oe$p1kXWTF$tmc6DsPs+ebf0-
zpsZT83ai(evyqN!SP)H&RafWfLKzDjaY+IfxQ$0vVOlMT>U-Ug*<~7@3ZNgQLjM3Y
z+8=uJC-9x{TAKi6MNytTverEdJon-hBx)8@di=kwOX5`Bp_cyU4SN`O`=ohTohqD8
z{{VV!{{VW_4^{YM0u2&Lz{5!~Q}|>2YqgaF01b1^d#35xkdzBlu1-z`5{;mk)4YiG
zaf+{ZG?A*s10Oa0JA2m!Ms~qm(}LniY`}BB3(k6)uW_kcm>A220CTu;QA)P9k%eqF
zoY5;oZ)rQF)M~jnU{;q<wUTy6jX~NleQW4&vRAu2YVFxvqEX{Gbt-FLNN?RZ_pbW7
z#O98~QZW}zv}A^>(qWeekz8_)r=X`~igRy^l{q7(d)F=Ell`L&zU9G5hCuY@u5{F?
zHe{-D)a7RrqD#M3o<qqABYuXxOW|SN;a5JlFa0Zm8-^#~Gkz}fX{V1zWBON=XM^>v
z7^wURjrvp>*!7z_5h?kWt2*s!lv5myQgo^jV%2vztrt6CC)&I4+C{U?#p@)DVB4iF
z>sMeIrB7B>RcM~Or7(-bp|RBzewCfzM&EI)NI!~v{{T8qsmiuzmFmCp6uBSt+5Z6a
zBUO*yTIHVStxw@KVB#@Xqc{N9r-b-VGZ?+&(KShp$AFBaa`7SkMR9sZI;0=<)KjLk
zXJ#GyjM(v0{`Lj!w3@+UFn*P_KY-RdBOYcpx`Il)0M+U~6}a<TuLUz+d~}?Vt!-d$
z8-}vfgu{?~dWzmFZ>rjc1RV4|!K<j6w2|gZOGvJeG;tVY>Nxt=Y_R!kRQ0Uos%K3J
zd&f269V5>lYK4DzzjzMfv20%%^`{D))VW_n(W<PjYeLm+7_UzC!C74wMV#M_Ez89X
z$Xx9>HOD&@BR_Sju+Q8^ScUJhyoa8(A8P4r<_@(iK$?bHD_s8ot^6tzSRn~T0~Ucu
z1f>9zfj|?boUXaWM=UQ9`Ie9jiZ+q=1Asf!;t=cx>P?+g>E1fM$<`Q9WaY6@`JA1n
z0QCUWrrErdDXT=D70Avh(r~9Wo80s%t5YJ<RKB}uv}Y27Tb?sQ*A~s`8-Fod{gc6K
znz}IT&@N*quR~H!o5v##y?yH&`^U2gFJodPMU)uaeGOHZOK+4vG<~Y#^|8lZmEq2a
zLh=flwD$oQZaE$4$y~yOxnkIn-(c-Ep%v7gSoEoIM{XAq-0V}ja_#^)&2|0^MoU|g
zdKO$C&aEwuYOTL$&R18yX|-#YBMrQ-<;7+eEOKevQwh{=Dm{%xlgQA@4zggH>wX<w
z?6lj|EQNnMaNOgo@R^=GwE3aCRRiWI$Jf%iGLjhc<ej+dif?hl;U|Olk$l4hbja&n
z$A}|Y{4l)#0INlR$WVe?qmYS&w$fzt*0V(AN#GGk(9+U%K?I&#sl%QyYo%MLL`9@v
zpnB3RJqVXov5~eU-!bWtim5%$`;dVCG>3CS-rLMf)E97h3?v*8K$Fy?c@$T0`LHna
zlblvoqk7WDi^|BnFcct>)cik8THLwr<M~#;i>Hk_ZZ57cB$e5ip2Iu}(B+fWnHrtz
z-p=!MUWz>ha`y|p+da)gHj1*bSlY=n$pk80&p<oXc_VnFXx&P(oDgY)+(bV$D-0R{
zaBeHqz80eY0A$@wj!d}E;Z(OWtMHsZiR4*)O{hBbY1oRj;FZsbq5lAJsdX#top!bV
z03mn(0Cj(rMucLxDsFdD_)*pPLf9lz-5i3)X!jMxc+&6flU_-?a_4IOD_x-!V703z
zbUJD_f#4rA<q|9gLGDi?ynf-!DL$sU<Z{y1>}{+qGj+m_rnaKAVxue-K<aBY&V&-%
zLIwrE&(f*M3nAx_dJ#efSCS@f)XyQkC;=KGqmfbKfGpc8p0%SK3&9mK8YL-;jP<D)
zl$f9dBAY-E%fr-ttFAu^5ZXgj15GrXP$8z8XaGZp6}@7`P(BGB)X{}Sp|1?F%DG0#
z?kc6=z%`@T=8on}$E{S5)-3CZ3pFz|7CAX19ct8RB#tAEOEYi)9GU?1-w60fSR!5C
zC~JlpRma@_0PELc6BWZbb4urQW6K*9wR=-|i7Zi<cs^ENZpN<J58NQe4mwuS)+ojr
zNtbcORg;HP=xR+(dGOF~t}9jI;!pRf)!632?#wB<sxEL%bDoIY)70Xm303M<l7iC)
zZYsRIgGgpZr*4zN$ldxMO40B~hRh)97ZLpA;yjLn*0giAHjKTU>Lzzd83LSQ$i;IV
zEUPuEazN&}T_Z!9{{X~NO>%?;aurXftv!t2bJ1tpZEUedF_D3ari(>Sip1gch5MEP
z08<)D4I3bmP%4u<O5qS>vA`74A-f}{)2yS?Y}(>>8}{YTp!BSP_Ky?a7v?KCj~Va(
z0N1Y98%0w*Xw*(}uKH|g>ROeqfnp+wN12b3G5pPNw^wstTr~5!W^dA_w7$wU!b5tJ
z-HZbi8o=%$lZwu|eX3~`KSj`u^uq4pe|8Vbr@eMD+66!W#wgsVq>-DfT8XUYi~Gt&
zd31@j$n-d_x*Ic#6}n`7=&pm1Ge~Iw6HF(X28K0D8u_-&xXE5c28gD^2Nc}%OiXf`
zlR2PaieIfU4I+Rzrhrk`r8p1;07(VI$0#F_T@|dpWMJ|sFlC@JKJm>i3V#tbBp#il
zMh}Ga=`sHREYI^b$oxoM+B_$+y?LMV_~~B(%xbM99v0H#&&ev5#dGLzPZiS*ffHNX
z$#puz0vD0a6+r7<=8^Eo%Not8T$CrDEeB7nXI8f&({p-s`L%l;Ew_O*8*o3-cb`^c
zSEKN@iyz(7V9&Y(Tu!Xz)am7qVs+3orTe;lxc>mQVwJQFExz`j0sjEDVv=(1R)u+d
zC!$N8+I%BF-CR{Eya}SZ;Vsz@vl_KhYi2`)Ug*iX_-AJ}Om~u}pujbkH^SSjpDR#|
zj@y*hk;FB+A<<n7yPt+y#ANwz6u(1?g2%vm?Y5_HB-Wq0Spi}g9xH#W!N$gZ!QAx1
zXNm)u+c@;7;Vqb%y4RHDD_Nd`is);k-yqFJJ1bz&s$E!z4NpDzlOh9Pp#1yEKT2I~
zjGh{ihjdf=iuUlo!bgva{wc!?cVUv)sRg2c6O&y7NcE3~@p(QS7|zz2h|l3&kr}L7
z(ATL_c>vEfOpVP~UdkyGM(x~3J4XX2wMQHqNs+l@)Yp$uo}3Bun`88@Q(k!{(pug=
z9z(Y^&ku;wUkew($t+M99f%Noq<Yrux~jo+Vkx;taBJxBdY%?49MWjH9mTcM<8q&4
zRw02Q&z88)CcOuBq2_9g@Y&RveF-_Mto%`Hoh{nPuQG`MUDzaKA4=vmNtWJOBW|ZZ
z%BenDaw2KEk!bVn43)dMe2jTAbD#E2df&p>nm-8w!yD&jA5JUGj^?<JfXMi`{8%T{
z*Oo|jwkf^GN8m(6<kWJmMMF9ekv=#yvZh9AdW?k!&CN<VHDs5q7Zqmc;s*LxeHChK
z^K0u>BO!Qt(X?IiqIPBirk3WFz7y8rPbb6e9)9is`~_qDH!N)L<Zt4Te_C&5axITA
z)ld9Qu7~|Lf97hwK6S1+ZhG+eO-N4wRoKrJ>0sW6n~mKbn{RHHeh64wr`d`8hHHe;
zM*U9T9hehW2k{%>-?+_B6^z|M>9GF*I>Z?7^{uJ=DBe3GH&c%wFzsEBf$ul{odaZn
zwX2F=?N)13T4!IW=PLfx{Towjc}F=@_lK=?@~J|{nzlNz<=w&^4ngQ@jn%&OBOfs7
zTr0C3DRQ$WRO+NxEvo6J(tor>xJExP9gTCV#;1|&zkZOm%m@{Ad0|Sb&{UqLcZlFK
z=`yPIUPXDdj7pXCtnlyMu<<_<B6kD>Ox;~4AoUp)JTeG=Y*2yfig!Qje+q>4u5yH>
zrZA+b0drUM`(6J4Ii$eYjx|xj)jd=qzyTO6aY%98$(6Ro6-LobfH}z&J&uKUHDT^t
z3}BjcHu6dHV*qs~k`(UbSyk9EyC73$Y_2y{TT{Q<IKsv63tJgA*w_%jDK(+vIhH&7
zc>e&nhd+&Jy0Pac%U7A0FuZOsyb;JW>7-{Hfcb}Cde4_r){R9fe(79H8G|nbcB@R$
z&lw{dvC5wGNj0f7D#l5~&k-P#gMtr3O}UB3&9{TxRo_#JRMLaxAW&6-HKze*jANRn
zP7O6sW5mW!OxJPYuiWSi=d&^Vs83PWrsEmEGtcgy&AgSzMsj~MP<No^W`+B&_Iye`
zfv%Iljf)$M9(>XKsl>vj#<z&AzS*QtwKqDA$PZ#G&CHU>Id=MHsfwG>k}H{EiK7@Q
z-lDsSC;L}}9z)&=5Kp1R6C_rouSVjkRGNS!5Jyp0XVfLSjyUo2FI;z^E)hDr1(9~J
ztxGiw!zU-TNo!-P)1fYcqSdXLVkZ>3lhop-n89>eNmEf6)_K5U0LDqpF`egw;#jmz
z>(!jnYx<8XVG|6uGGn*3Y0&1i=4Y6Hb@*2~a=RyC?@-Or()f1D80}5tC^EMI`U>ZK
zO$d)ylLH%oY8K*aQy!%wa)VTWXs=fIW{jR3x^g&a2_KbG+{&-Qd8OoyE$s3!$!8?~
zb+_Pdd`%bojr^($dmVPS{{SIYKkTpas_1eBaMazM)V>sVFt_$!h5W$4Fo>`x)by`2
zyGOj97y_3wOaTMGJl4`)^ij=C={X$-#Cx>Uyfp=q0zB6WnCuq3ptkDjmghYmgtcaH
z=sN?=S)ONUNZr7wHzLDltxAqpgUIwWfUvP6uHj4ND5S<kGy#{2&XyvEn26PioQl0~
zchDLdxEInu;*r*~C}kxu5Yl3RE6u~2j=H1pp$JVdnxGm_G>1IUA)N7=N&rE@t(`J#
z1<yUP?LsoCv9}t;+^E6sD!uP|(VWrT#hJOPj@6qwqp=Gd(`gi2v9^-pV38v`9smNp
zN5Z}mT{%|j@k3mmK>p|d0A8`HD=QmPjor@1B?6p!*D^;-Wm}1wHxC<>aokm_Ml1s!
zrmZtOE8Mn6%T;EoV~*yi8?2Z8(Og!o{{SB{{{X#3U6F<p;*pUeWL9;^+g$AQ3hMc(
z0i*+iQ`P7MIK4$9@j<6SmR5h-SMkG)rgHxPO7G#dTWFO)00DpnOVkswJ4c1jKq+EH
zk}}=swxP|Sw)3eBr9nM9RC;}<m31`CxZHU9RV8+eYRb`mV$Cm`F(aVhQfC#DcXv06
z<%&Wym=y$*RU%nVX{&od(u7DaZU@XpP#(k9x1$@#;l<VEr)AV-%TMA9!)+57>|$@7
z$MKQcr`0uEPY>BFF(~sfh~pp3*G27lv(9n5jV1FqON*)RZW?*qv$6cC6jv+M^-@V$
z4?QV_@k+%I%-CAaw}G`CMPMIva^Fftqq(<%Vnszz032ehu)@_OHq6_!-;wKHVXfPJ
zms6fRU~b4B)zL?Dj}fvnG3~`D>0Ku%)5SEmGyu}{nWRwGH9dj-{?5!XNVG?N35rbP
zgH=rk7z$8uX^53toz|E!4?U=$29t_wHfRD{NQ{z<4yL-BSp2wvq$XKn2&g2JuRKtR
zJ!`{ZX7FB~n}+W!_4=CR{v<>p(E>99cLFi>uPy%o8|HM!;IYWda|QnZgl&+Y-rD?_
z^cBy<tn4FElE(m7L^Mq?p10u73dN+_CEVy5>QSCY+!fWT^ImO9b4ur-1<M-}s?+g`
zB7u{nO2`j-Yd}X%YpdB^CXPj3(rQFSSalTBShUg7q?)=DXr$RwVy+tnD2|na;7}f-
zn%42K<wxaDl`y@A8)XB5ih^5WM+UQ%70|fU?s1+h@Qv2Bc@@NSs`+cQAsuOTT?0^G
z4Kyndk@hIUoc9&zVXG*^mGn3<R9l*o>dqE@KK9$?n%+et{hBoE3H2-Fw1g7QdG;07
z>PMHEGC3uFR(n>5rly;t!D^>F2Pf3kySt^dwP_C|a6Yt@z0}z4c{Ggu%*q`902k>?
z+lt1t`L#8t9$g0{M|NAiIlu!I=2Uu}vDk+=gy*GU`Dv)%6^vpix*t<mY}+rPddA%?
zrIkm1{<V=bnY)+$<6iy+=@xl)mFF^eU*8paJNa$KcjZT8UhI-<Q_1DKv1S6tbj&dF
z9G_~LpNg1A-9|NM`^RDli6}@oAa$-U#dko@CL3&t6EMeoWBh6=rR?O*BR9<z+~qVN
zTkUP1yKd+7HSIqS(O1Ei1P{F)Pv>4_4Y5B7$b4Os7y5dC(=vZkUTbi?W{GZTNBC&0
ztxh8%p_`%x8(TFpo$HE_BM1W%PF}TWmT20qN`3QUB0X!;j+Bosde<_e8=kc^_07y6
za(9pCSyA|n>-;CJT#+7#7?1}f)_23$wrG;SXOgM>EaOL#>SsE=-~D!f=4v1_D%TXZ
zJ$P(sLDW^~`U>=LTQkqb?vF?C<feZIOU`zriTYO|;h+wxfc8^bEgXx({4BlVqnoRK
ze}Ml0I>a20sjcaJCroxDo<A{98L9j&a%{DPi#XogRv3tNG>o)p^!x3v0&W}wS+Svr
z<R0}@-@0)q=x=FSgFwS4D0+`d=owPtFv>y5;MYB9+E*}oJDC@@*d@Th9Ytha#3YQX
z41-*?SG#n2_*tndGXy|wt_mNzdK$YB$rTkH;d5GRmUbe(i_C%4^k7AKW%OnXxg$NA
zu)}WAc&(Eux3@IVd9K5P<J9M^Px9b*q%y6~`rpE$u5!en(-=`w0#`v`fLkxUCI*aZ
zF<BbnQqjlga%mL>c^MLAj&Z=NmS`D$DNX2h(2eTqTKhLvA21aZ2`0%GA9|%d2RPn#
zLnT&>9yk?9hj!o-RT425O(Ui7Rjl9IyULXcrzF%~H{)MmdTd&xa>@JOBMK3R;MIWK
z;MBd&{4c;(l>mH>4ND*I2L}v0`qe#*Dd|?kbE9Mv!+O(^n4ss8*!QJv4Px4=(OOpX
zBp|NS+N@aJY;cS+J!wg3MH-3~l#3S^3HGH>a2Kt27KS+dG8}NN31jsllo75jqLOi4
zS?24w$&*oct49R(EnYTZ1GewzD|^7QAGF*@o;CxDaWJXyofU<`+|Q_4zDY?=bN$jQ
zfxfo;CZRkVa1qD@+qGLF8<3UayNx~;o57+b>d`PCog_Kzj#?#Kiy)r7R#ecuG8AEg
zDh*hGFY#nl)|gwk+am#mFgkro!rtN!F;pXgkib@Mn~^TkbAwW9_B-toBQD6*a?R4U
zb<Ypnrk?XMf&%2@wOk;Z)X%BqcbZki_YPxB1q70Kthiy4X_7d>F2?{4^*u^GF3Q{c
z8$%I~nWIAO?rV{@noFz3xRHrxe3Mqrcxu}>o-z}Io;@p;)P^q{bj3?GM^mAp{>X&%
z9RC11=loj`T|zh933TLoRIS0B;TSlnBRK?Bh!0x!4+Oef_!2pr00ohl{wAtBnAZ4D
zB$dNYBX%rL>s<$dKjdbM{<**Msb;TZuhxhBO&|T${#2T52Io+M5&TXohFZ+)gS9)I
zH;pcQjV?QNz!Enj&<f-{I|thQO>Q?}K&DQedw*KnddTLEn0Czn0E!pheya_d$&`gM
zSJ99E09zHoX9l_+=5*Dfi&l&a?PGv@`c>)e-Yhxf9^;{?Z&OsXBNV5DOb5p~j1k(D
zU8y~33=P`))mw>G`cMU0KuVm7Vl6{-Q>m$b-|tgl+-KXe-D=cwi3eIOU`&#e6(cK6
z{;<?$x})%+4L@38K9xWvAwQh}Fw#)KLt0vSCNfm=C`NPG-L@+-1_ul}iiYsxHD@eV
zrX20_A4;mn-l3gQBd5a?%`!zAmUbWjJJgnX=Y+g17L{zX+$d|789aU3>k%<rQkOKY
zb|uRj5`8KOq}nJZ<}LoA9ly;ap>lD_t?L#B2-tTSs#}SdSn*XSCaHooEb`iN`ry|k
zt3`{a`qQcKqZ~K!BxFc14P{=C6q@E|qGiW6YSKPM0v-)0YCr|?P)#ofni)vw{3~na
z!{;~Oetv?t<eN7Lr14OO%U5lSVnsT^f6>NITC^givE8eSfA11%{<(D)odUBurtB_B
z=~lFZXFT<*7@3h=s{lg+522^|)y|RF$ew?Ct0dj?S|np8YG^uF+a>!|!v5(G$`A1x
z*}U^F-|TkmQWgIIT_T)+z(sabjic&#)#`Iq@<nZ1R<!W-g2fVpCNUgi`I_J-xt{vv
zt}YzAxyjFJnzGlG8sTB1o9ar5#Xo^rSnV3kP`slg4%DnNjpe{d&pj&VgQF58_eZO?
z(exDTH1#_O`%~ZU(O_)Plx3@i@vWDaa~04B=NLbQYfp2YuCX~%DGhZTq-~$Vqb7l&
zDbx&O3VaW;Wj#5h%SOHBvdtIaSTZ4v%C~BS%V=#GsxurKlr}n=pcFs8DGTO+8_N&{
zPX)V3*v4oBddE)@fXq3q63UB6hRX6veQLWtbImj(p|J77Y1(uaLK$75TpS)c)^CXX
zSaqu}vkPLJ=O(=MLHkZ?^*WRHOr1Vq9+`C$C_9Ke_O0&$XltqJ!s<oc1RV3+_NQ6$
z#j{FE@<*<!fB>mcE6$2LTOW*48oLOw5&h7pg~1<qX1e*gQaF`0IY@fUjo%+y3~1l|
zYQJ`9&eNpG+5`M8Oeo9LRuiVt*G{x$qugySYa1H|X+f$&VT{rh6?P<oSb^<Mo-Ejr
z>?l#ulR0Z2QyuM;<T8Q}sHiP;>qsv=*sbG7BkwAJYl+@EBfZ(Hog_<$m+4hCJ1@3I
ztGuWUo;%mKCHudT<iY#*AbVLB5V0`Ip8e`rP=y=7&$W5-m%C%JEp9Bwk&1|dYnhs%
zuOIp|{VOukDB*#ljO}CIim>I1ea8u1vGLo>X=Y}T_l?Bo&{sH;ZgSbqPHWi0J1Cxe
z)>AD;_^VdG>62l3jx+OnpGx-XKXpgB=B3Kq`x?2VXTsM)M~{258ICMwbLas4>&~_L
z5^K#$Qry7+kMT9jiT)e)JE(kX(8$n%F<!7dc{o3buXNB>9VbWA?Ak?8$f%kAB1Uj4
z&WW}oWQ;E#>9A^3!yKawBKcT#UQKyUsix0iA&zCp{uUhwr6s9&d3&}oPs-IQ9Eyjr
z(uT^J6&!7jDIt<E^rA77R*7b%vhXRli{=!MINWQ|leH7fr|(xaY&<0m#9@uYjQUpg
zm*BKp=vikZ1JKmbQRU`3E5chNsn_)Tgz)o67~H`drBv`dPX~uKx|Jc@sH$;(&9l$C
zzb{kAH6C?aqyGR+pZS_%b11Jlp44_UVCz@o`=Y%Z4^z*_bUhcrX}`knANQn?es#-e
z3A0~#{{XsNezm9CknsNi3m@WF-F}Px#KmO~<SV9^!j6@&8mTaNWm)1RIvv2(RF5=C
z=CeH&bw`^=jimeJDKz+}4godCN$FxO?w9S~9Sv-0+KIPduag-+m1|NQ(Xq=Xsk0<$
zJYkmyBZKQ$SGJ}mJbb3OEp=mjH2Gv>-P;Bp<E2AvIXN7P<$6V<+Dh_L7Ih4yde@ow
zp4ZCNTaFlHRx}-^$B16#%x^4?w9uI~-E+>dVIU4N7z61}%#n<31bWoAU5Zy9>wgN0
zumLG4j9L#`kPw>O(<KsHZO<!7fugctb68i>3GZCzj)Idlq`l$M!{&UWg&(afpXlp=
zqbI#)^>#g5azmXOnsSD<+}-KZMyV4p;*;3pr7O<ne9j2kX|Nrt*ilstip1`9{uTL+
z!ydZ@b6zcDKC>f#cMJJdr=iCfb5v`{#zfjd=~vb@We1PSvztA9B{Y*vhT<-I3S=>1
z_@<?yE`n9AiKm1#T#RF(%|RR_<ag|7QqZcjKYW_%Q@TSO*&~{I6-M3bR`xg&<*4t`
zjgYJ5l%3hG`uPo~gW581Rv;;@j(N#CipQSB?F^z#pzS%qtqE<Rytg+t;|xCXdQxU?
z<w?|=QvQ#R(zg5zz-m*ESpNX6R7~gan$gZI{AnyK#4XHY0C#GGUC<}G)>c>$?Ui!u
zdIQq0p%M6y`1~A9Uo^OHZk4~MX*U*H8%Ef5K)?~{ifIm}$DCPAA(LaA0(h-S?4rGk
zo*T0Y$?k5oMm@!h_JoNLvUXumtZK(7-1-`1os9BA`a<nsSVlAXR&JYULx(=KCY=vQ
z@RVi?*rk~WkFocz!&<q4G`aUUJCJ}qYel0(y&61{ZWe#{TlrNF2I(Q?Z9--FM-k`O
z)~9(2tF%t;T^>u#I&YXR=*V3CE0wYE#pEhuEfK-s=C!evCald1tv==gUE|6G##Xp(
zMDDwDQ6kaRTTk{64M!juaTzML!dzX;d34gu#h9M`t2pjAxdZUj50!!KRw6~@n)SUh
z2DA8SBKP|<u>5(diH%*6<Xug4o&!JeQ^)@RT;KUrGgq<MYhxOYeN<MTge{D^gUmoH
z68G#)awU3(@^+w9@wMNbs)&H%6x;RBG@b;2k$HORyIF!GcNorprD;!Et(evKaOoa#
z;+;V19wEG(t`C@30QzzLD-`vvfkjyp+kNa-lvj@u7@H3n>ZY@jxuO87=P~o`+*27?
z9z`%?7-daW+CiouAkns%3AT`m05M650S#sEPlk4lfITSzrxe0KIjM0`u|-+VBtJ@o
zpTt*8eiR|OrWUFJ3FeTRE(3E#3<jFFVJTPIJJLi*=waAVWLDbP<a*UB>H3V-m?f!^
zE9CuZ0%}>h61^k9eiRWRYL|!%jm%1agnHIBWoVnxjpTY7AOf5#l^qfv&{kfjIEE%-
zZ~&@%8Le5vMW$bPk5gIMnJnUbf`5x1mFZTCvLsR10YY~wO31g`p_y>Gtr?Qsy{10<
zLb&^<70+v97ftk{e*!b1rD+`X!Np)|vk0Jeu5;2cVYvhjYV<RYS_v&IfYXZ#LnC=O
z>s4iP&J=ZHibFfC5kywuKf0h(H9O&NcW$8Nq;Q~CWk}oAZUkC<W6KK4K6VDH_;Xi-
zRIzy?Y^pI_){ASAwu5JXvFdQ$6SNY4Dy!pUI+l$zh)XC<%a3krds;C$Q|7BCd4`$d
zJN-8SWM)D9400>7()?9#sGwZ=a`!{u=}xUB@gdTz{uXH)y~C|u(RArGTd$ZA0vx+~
z`&Lkm-G^GIBwUWz!WIDxKf6#cHufIXnXBsd+7+0NFbu?ek&otUwoOG`@nIKIT)ode
zxSo6Km!9FvyPiqyN{ftF7W$sjNh>07*jkOAs0dz9daoqGL7*;SYL}lW0M}38sTAny
zM_tD?R?tc6b&<D9m4;T59GO>OENh<eWvIE+T|mo6^DSI-W>VE2a>E4EE2QNb{{Tv6
zF-&6*T3QA)(?J1+B9YMBxR%u$uNbNq_NwD@kC;|h)ZW?|3<{hPj8ayHELt&|Vi1Gg
zoQPP{5q93xvkoh6&eAy49iudXmYw$dRd<(~5d<<zRivL86+=aHaU!wXih<>k$t-Id
zYIJlbvfSn4T#~~8SF`v_N?jjBhCR5JLA}qpt_)nPtaV}4-OcoDkyci?pw5{gO)&JW
zQ6%vn-3qCmYodP&Iq3Xy5VZnUUWAIy_Bt9(S&}Ww4x=?H+@b5mZ5o}<T9|&Z9f;%|
zjWMA9>8g|zwMKBQC_6GE9E;mON@yfj4lO-SxXv<K4+fU02y0Z1xwZT%;I}cWynRgR
zBlfE5n7B`KisGpsc^-r%?z3xd5Pd0*n~<%0EogZu>PO|AAEiXdQ9(UTaaE;dbE*|J
zu^Hz+qM>FW*B3*mbmJCGeQPQxTXFZTu}aW~p`8bJBTjpAG>2B_<UCiL%RA|}uWKhD
z9HI8E$};7qpJSF9&R%7RCb-mO@@>~>{wDOUlF~G=wOAuL^(Uz{>tNig)b@{MvFBzo
zRH<@Dc1*J8ttottYt?GUgBIH<YFB@{NR;pn=C3lpmFKq9Zh^p>;4fadt}I&Rr5p9s
z^f0sMtt)J_Ch(L(;%TjqS+<}#^)>7ovWRZ{KMr|Y#2ES^uQDA@yhic)oR7sVV&h2h
zR-Yt9lxts|J2%d1wVmFd;yWUw1|WROda0)OCMwWUM-k!;5`7-z@LmT$BmV$O%#KQn
z9dUP8EP@Ku4!CTJlv4_p5%UVsZbR7BfPA%Hcindy`g2|N^rU&%U2C(p(Im_&J*%wJ
zl$(^tYQluC1Jth@BXd(gY@K~lsc8@_Izbr7%NprV_e~xpN_T2Ugm{D~)-E=T10!_$
z)i`ok5-ZJn9?r&u1Lv&>`U>=LJ2TG5^gGW0PS0g^3H!0g2>eB0Xy5WP+z&>O{<WmE
zoP!PUvA!TV)m=|a3d{iW>P>Xh_)*rDgRo9>QA)fJYoRNcRM6@pk)Oj}YLA=BKczEB
z)Rt(JG3LD6a!rI^H8yYC&(gGEeDhsRXmGcwt6`{^8_SU9Gx}Do#1Y*^k%N)Jt_rc`
zmCYjT?sL~L7}%o@m7lX~o;uehc-vFn!%a#f9dav=@pX{cv*V%bTu^qI)m2^$<H|AX
zNU-yo>*hTEQT|^b^u;M$_oZ>zp?INg2pus{QZo?IK*Q@rNC;|cYnHZrgTd=S8h_d-
z@f;tRiq2kLnEwC>sAkm0rR%TKpZ2dYhJPw{Hu}SH<!1G(s#<~4n|kBO>rg}<rZY?P
z_03$jWMP5n%~r#5a@_4aBNy&wlhg&STV4URTa5N|Q_yt3ip;;^&MI{v!WmB^sHu`Y
zN)FA8l0eurk;t2eV?tULDX(d9^E)D*a;Bky^C0IG*1=L=y;lqPw*j8MwN@xHGI^mt
zbmp%~O1lKLGu$L_{q}4DUAK#i$Dl@l;Fncl>&0reJj$?2qOrh<GH`0mjmO!phEML_
zh7=h$sQpq8CD^Gu9PSlw!S2hi#GvrX+5D-ZC$Xojtf$7-#v`<aPn*zln)FRZ2*#R{
zL_Y5$9rIcvJ-JzsyN`O`(bsmK>fD0FbM&T%Q)9_|U8F9XsF@*Mu}H@Q`=`>XX}XTW
zBAERRXC!MDZlOsR2Mi5Vy3p<B`J?^YA5N78+T)%EhQT0Z%P$%AtsOa6d@6b~0BOf`
zHPH1dSfNN13<UzRW`cX?m18B4F~vTlvz5NnW{&jRGVLy>@};|(G>Zsew@@93L{L97
zSIcH+Q&ysOVVN{#3)jjOn`)aEo=sCC9ZN$6aa`WN;b<?_=C@H4{@ogu=84$KaPY$%
zGLM-6Ql_}wQrKGR4yrJu<W>p$)-A7g<!2cb(aQZQ$OzkOVAr$iQ%7s?@mTO4%@ln~
zil~^?TBF8#99K!;KlvIW{{XH}{Hjf!$8D~zF5_=iu9L!+;@3xemrOH~+4WJ=`qw;H
z?-^Ae?ov1%ZtzWaJ;qKMi5c{+k56c{PXay6ipX9jcl8Y1e}z^bR@kvxMln2T*`13q
z$tNV5ZEPY~4jQdJOD$|RHe%ps2Rs_N9q!zu=p<g}zvo4OvXENcf2ousAK_|(=S`j!
z`QtxPKv>UVgM&#_Aqxs_;XnrLo+);>_e}tUV}Ev-Vsrc^fscixjk+9GoHA}vnu!7+
zxj(%>N`RW^kHUayrqwVSO(h@>)YCwMog0*JM`{UX#5arfez+p3TuxuTSqT*w=9!9y
zS*zjgDfN9`8+B4sc3k=n_3d_wsEZo#r~u-*t36UWsOXETc&7@<8b~B?t5_Gv+KtcI
z#(bqYazB+Ry#kuMG_7spg%&u_DyYB$v?CiuDrMCtUX;m(Mk{`YDQinOjz#y)VC(!|
zF!rY}fX<ik+00xXO>=tkv)hQ*g69>_)bzR$X?`1N)HBG#K9#E85}8+YQx^7WiBw6N
zMw4ivJ-yA0n<tE~ziMcgJ?W=%8fn>$XR7hesecJ?x*M8g-0rmdl=6uUjP$ITu0GS_
zD+rYCSVKEjaax*7V{gQF;vFh?-O&y!58<@U{grkJI1UYKO?!;WZ{I||Ki8v&Pl_9P
zF+1^$ebL^o+25Ow2*B^u$G7W&TeWzjlA3Xox#cpjGfC;V(xbFY9yTh#<Plq#oxg-G
zF1$f<wn_<+d1uD}dYbk-8#t^jQr;!PsO0*eTFwyVlV&)oyvanRu4|C#5OohbkC8vA
zuRgfAn)czYE*!hD>S`*=?#7r(c3kXW7^jYvf!rg$1yy;#<26K;Ww{HUYK_I;%6X<2
zFfV5=#=4&Z-tM=S*q%|dk7HWfLKF9n;wf=gW0P{!*lmn%<W@GMyTd8yMJ7q+x}CYS
zzmhy~PDj$In(2$s<$5q8qG6h#5UwdH6gH6cplH_9rOmF~^ffGpD_J$E(4K~E%u2+b
zhN=;>=~^Q;HvpPmf~b*FHk9#N*7{6^74d*+2)P}TL_sxdFi-|)A&^Zky;PE4dY~&j
zr|VEOgwQdJKRRwRSsN{mi@`c~+x6(t*C`m@--UZBj(~QrBN;D#$DxK^?aNWGO0c~v
z2pY%<sMyx%luZkM>Ztjgn%;j3Iq3Xy7j|hgSsfvoGe8O$j)aPlZe4rksxoNKbfZ1V
z7V$rb5mJVZQN|5)MoL;7Q>8k3pfD-uaZPBPdx{-C#X`SH5fFQfRh(wu@S+QG)_NvT
zVO4Sdke}~YB~`KCP3X;>a4I+8eiiDyk1cKhWdW#WC6o=k)>UrdWNQzIfsmr)_o~O4
zK(97&<&uj#V=iTKOSWILg$J!e9lEOI_pa*nRMK8YJt?@|mNgj+T5;LQ#w&-^^!sf_
zMc&?3z#l7TtxKmUzE{+mGEOa;Ha5`M+bl71e~G<KPQ-m{>G0I!PMV57ZyPyOs@o7a
z=sMFUmtytj(z;T+vBN2QIP)Cmh&2&urOhtvJN&8dTw6Owp~n-GnWZi{>MPF7W$&o=
zJF=Mm?alRPOYr`|t}aB8jJnK+7##u0uSC|-rSX4<)G^wUJCUD3{{YvoHXTnwyZ4u)
zIG>FPyI-(B^y&Wqvzp+v{YK+Wx>?|TtU8ZTL)^Mr(?@?}saW`m%)=tDnU9(_{*}l0
zhr^O-QW@>TJ+mLFPpvJ*A9hw|L{j!OE09G^wKu+lG3!=smyue-;xn7HO(@0|sKuFe
z2JQjCuDn~dG;v^6>Na+|Fe+4YUG9%;;}VgDBLg+Vg_pX|V-=(g9aXm5AD|Up%<a)6
zZ%x(cFX5gfz7%}_0IojmPGkPr{&f|wt_QK{^)Em^VOQhnUXBN5d6?&_J8y@?ZN6UR
zm-!m0Z*C&-h44{;oPK=OPxz864e<Pm8vg+Ng`fS=S-9z4RQ?opFN9Ob{AsRp_?qd7
z&r@Tj(53TqXN`a&J<Ij3F3(VFd&g7qpkfEHuQw3wQpDl@8C>l&>;3AvY}Vzy&%JF;
z-X|?(%Odhio^x9IW~1dWidOj#L+Mdg9ML8$y?<sdqTe1jjz_(7lSDxvj1gQDT+p%U
z;Hp|yD>oI7uiE{mPL+1#<08Dbq|xp*c(c#UWE>tS^1%1{S53|v4)VnY9OK%C#bYA-
z{ytmjRqBicpwKWBw3rPdkcvP$E+iSJyO0=nH7wc=$c)FedFx3U=4BrBL_HbM{{YJc
zPUodXc1&@IpQ#zF%0^jUYQ?5<X|kes{t%P?C6b=vPw8BasQ!E0Pv6WmTSOG!zS)cB
zkZM@nh{!y56xueUQck6aCodl${HB_dHb*ohO<77av2f;Cv7T|oQ4&cMoc5%Vt$E7v
zk#Vj(_^?h0q_>xDLsI3?-B(Vkkv5n&YnHo5B2a%?-tpzpSgm0l7?wYW{{XJETOMU;
ze(>Tm&T2q+40oW!)VluwO3|a2t445XJS!lJUxANsN9HL+QrPS@NKth-E-~{IAe!tn
z$NP4U5U?$@srIc=llT)dl`u&3HMgRi#Uy@*G|^M|PAkUt2I>ofqb3;(fnIgjgVcAa
zk!xc@$5N8rf=iZ9UrOmMXO8AD(XeXCnjyDirfNuEVMbo6b4vCZ)miHD$pB5ncCN2N
zXpM*jjllC&d$BW|@rI)fh4d2a-hmi7Tx4{vKHFFR*wsz~n9zY3Dl#is%IP9pedNza
zbSa|+J$3+UlxxKXP#jkRqLIi`K}RjBN{Ol@hA_*samnvo-n#-_TeZ?I){OE8aa_&&
z=S{zMbLw&_tHY677!HmvGIWf88v*&6^}iWJZN49+xje=qU&{iT*_l*+^iL*{c+GSk
z0sjDzqA&X7{{YIQ(d>6x+LyeyMHwtG2jg8fqRFLrV8o=U^7b#()(b}_RJBlv<=2RA
zuOPEl+Sd{+dGyJym*Oqo*gP#Anm7$BDiP|SbNwpt>)mXeOT9>=$#X_4QI120t69p_
zvev@^!KE#h%>X@pRN#uywbbt|mk0N;_Y5;bR}pZ3W$K^@A7`7}fK6s<IxW?SEwsFS
zm{Lihy>&9lJ?Q`>)d;t1Z<YnB_Q=Ic8Vp74NCafyVx$Sy+9Enqm;#TMrsFg2xS0Ja
zxi!}hg$QSQLTRuYo@mVi4AM{mNk~kQT;h<nY1~#4m2uvJk5%vnj+S~!y>!I0^UqK#
zs6{BGj^{PWJz3EgWK4~Ll4?OH^{FVKsbBd>5jl4|R1aEptO5wY>c^6O>nXE(XuEE?
zY}LTEboM5$<b#1slaX5-tg)tyNhi{=JVu*t$@N-t_zbDP-<-7LlUX;Ad9ud1&&^!S
z^wHC4b{L6H-T<vnCurnXEVnyjqAA~5Lv<5JA_54m6I8N&Hsqn@58X9%-nKEUw7E^B
z;<9fhA@?cd9xJXW<C8lrI!0|e88>oJ6Hs_$wdaX$?sJumHfQ|(D<rLRM)oK1rO%b6
z6y3|VF;e_KkI2!kWNwl_6|5KT%%uMScynGhkjbnl2P&&u-Wk<xrSRPLaxn8?jfe-m
zbisS5=bWu3a0C-0A67MH#>V4Oyo%1&GbGZs)lpPp-IdRE@OOpc@Wz%R$a|R%>14-$
zLG4#{-xAws*7)47a>M0}9!I(Mrzgs;aK$>cE!^|@ucp48t|ySZj&e;oO>nn!*-0g2
z1%RfR034bIRk*qe4>g@F-ZRc~O%ha+<Eb^DcXz@0t60QWGNzw#I5nr?T|?}8x`u2l
zzbUOPp_TPLQWxByioX+ZIjqZM$Al)a?+)fPnkMHT;;S6Qmr6NPgY>Riwx_X-w->!N
zhAToy$)-?H(lJFMk!r?WnOt?N4>m!lYg0t@DNQ*1YMZfRlT=KYpyr$g<BDXL?9dLP
zw&A+i0v?#ACRq-ywN_)dG?4`fr6#Mp_Np2qJpJQ}k_%Haji7`b&w5OvJCqSl=s$$i
zIp9`ii1e?9Rv~QXykVAh+7GpQ0xfuWmwK%AD62shP;pZVn2m_Xiix^b=tI!r8qBso
zXq2dI5nV1fNalpOT-j^v@{IH}g(P08Y9$7ZVOkOsBzC7(W}(uP8K49rfslCOniD2T
z<X_^?)~qep_lM<O5S!|9RiK`+BoapaWEyq`%nk-Cb!KrsXuC3`kh1p3_No&dlPh$?
z3gfJ~QR>i|p2or*(SuBv`d4#zh0YYc;Q_=T`&4O@l4-|V1KE(>H$hY#zH#qd71uRn
zj><Yy66LcrYmNt6orsZ<Q>JUuYV{I1Z)$fRsW0x8dlOuiv1R8;@?<)Us}EZ4jTqEc
zXP&D%apoa{;#lJga5g_op7m%x;rE6<mEBEAD>Ig(g#F|@a8Z%h)~H=h21^_P$u-wa
zE1n)QP3U>=i1m-{8^0<em?Ah+>V0Xq?7!K1a;Z58bNScL;;MT}jz`kq>3b?Mf!q8l
zRyR&S;1i$9x|Wtnz9_?P#F*f-^AvlN&0!sax5x51UyPsfAF%%buV4L`*PJ?m%@1>0
zKf^@US903SvPR`uPXt$UrfYWkv|nes3MC^2Pe-Xv%&L*Pxq;#@3$><h>{OWZ^PuX!
z<J!5@B#wJi93^x=y<D{wr6unoYWxhn$b^y93YSFF8c9R4oRgDPn9k8i&n<=I+36Y%
zhmf*LxIxf**G(kL3W(Dix_v8~2^OMvVr9=eEs)Jf;>hIlkyT@pZ-(w0Jr&RCUZ?!K
zJZZg;pKqk|Z)JYDV~Ud8;YT&+y^l$5Xj_c)SEA`&mJ8uL%y#JNyahfrYm0u(9`DMo
zAkT}nLOTUNr4@gO%ftLXBNJaB>e6HTqO*YMU3B(GV)#HFl%AE-6B(mxK$rJ2!P`5H
zVqWP{3#+^gH<(C1wZ^F`mF07atw%~K!fAI>TS??danH(4b#`}qfCTeGi;EQH&l6_a
z;2c%;IT;mFGJ`er%}ZerOnzVCW7?+F^x-m@p&vhRJJ$t8%P&)=l(}S?&rccWj2`t$
zWD-QmeKB4|cyi48dNSvNJoi(B?AOxAj9~FqM>VQ?9FIp~;d!Y*0CP-}pT7S9kCXJO
z)n)=3C>T=G0L3(lKq?)lCZvsb$WL5SX4KW%L#2$GOJx1}@!pPsQZJa>l`yflpv_dg
zW|e~;pjKDZ*OT12oMa)kw@S5b$KzJWj>moBxJAB?=4bt3EPs`9QsLpaJ-23?=uItT
z*@+?z+*6^*Y;?^wToYAQiw|zk8B_RHbHedAwM=Am;^@gWBE(L^12ixU-!RCb5iMe_
zGHZ8;Z%mammhZgmJ$qAHqc;ams*5$eFf-yfmDivrpQT#y<iW4);vc|-pT@0hanw)R
z9IYyx8bUR!9UDu!h0Z^_^{Sdf6xHoW{{Wt<KUzz0dmZkut@fLC2Y{`S`Bt`xs;!hx
zj-`VfjC$6W)P%XI*q6=o<!LYm<D3fF(H9Lj&{d&N;V`^OAC|Y9xf>Xt%Djw>u|D+@
zE*lPyd&m(6<|hDRyA3N&i%hb6i>BJ_xg$I+X3FYi%c`}^j1_)Bcs<Ci8#z^!W~G#F
zSey<kcG~rzjO{qhCAg$={w2IOD<CKCsjM#u4ceA{3jY8))w{4;9>sZuvSIWUpBnL2
z%we#sDy7N7nnfIbp*%9&c{p}_<K?RNP+PsTTExV=4s+8!mCGiaj=0~H<Jk;yG;P6C
zgHgD_2DdqjI(XlGZ~p*YSNYelc-Kq6)wC6})8h{1)R0&X1yb0;qqLJ}0sjC92Z*-w
zX219FR+oglE34_gAxZ9S84x23gPKh;{iIpjHk~cbxo5sP7B$Nj+rY;)FO1%2h8QA8
zW#yY{=jm8UH5nHcoa5winhlh<UND2qgqZEryKVpjLHxcL!R=K3Dc<?oy|tJa@*~_&
zuJ8W<RcTFHnKd2j93?0<MTbLEt7U02kh!BY3{`5k1;*YFN@6T*hF(Vp)K$BEM)JgM
zkbS1y@I!M)KrOb1X{_5d)|-JM^6=-8^c9IEk++bfidBbVD@v5y62DQjIY3dxIIe3D
z7*d#2j+B6&2}n3Ka1hlGahW=Y7espss7-al;X)A8PfB78QH;<5w9poSuoN_ypc)Ez
zT+jo8dR479)9qT7yc=>Iz&}b%4{Ntb;D#G$hSk6TdkVF6k+Bo0AXh1{%aUSMQOPF)
zqPTx5Jd=u+i$*4yas8#K%RYBS2g<&MyVpRKjJG;tv3Af^%vP}2%A?dlLvkxqm5#DF
zx&zc!XNlJnA@y2s;4_Eu(aFp?HBvbI%jj3$x!LMO?(A)qAznLFlERp*xlI)HGcJ%P
zO6K+ZQFUbee9OqFS~FQ|6NZ*d0z1}Z<g;VeyW)>JOzYvgNHnEVa8^82{tjOwPPw{o
zy*!=$zx{QCR!r4u(U0Rh(=_qJpSm|!OX0oN-WG%9KQP=+*0zfM+A(kUjNvuQrn%MU
zY~(9oRVc0`jFOC0k%NQUx+Ag8Cw3e2U5~>(PfxM&2#)eG=3HVpWk0}8G-XjY2V>EE
zVdE=b4Ox{|S#KmCyN^9R&$W2$dez<bviDbW86=qe#G0Avh{DF+@z6tbIID|rF`U;C
z>5)x+8V{Mk{A!iPu?nfgfxeYVT$`DbYBEg6WNb&RS7|PePRhGjMLBLRGDC4xq>-5N
zb6YKp=11P08<?b5UaE6cVtU4dsEgYOrq3)n`qt!ARz|nbxQrPNYclBu<LOD0b2{$T
zFD;~ijE7Py&MqNqdzm3|f{aCJ12?$Eq?7fnNg*S(CMdB;LrpXk3g$9RBiu7_(xj`A
z6O%{~nhAQ6taTtlqL_)Gl;jH0EEP8rdU{hKvP*k{z~h{HQqLh{_f0Vr95G|qR5C~x
zfO(*_Pibb-knfy&R=iM%*QHAti>MFu>3+3atx2jqo51$JX7J^<MhmXv_}5OJ)${dT
z=Fe6i3MWc_>ZpSdwK0ut5?Lf;F&e_LFBYRHKX)8gYUpua<1ZaYWXz1Y0QRd1e-m}c
zH9pZ^%s#CvClY0n89S+``4KSSVx6NRIjU1*8rMH{zdDTDNIG*_x=*3BaVh*mOl~@<
zredPa9STY8B{)??nF~@RuATU;Q?Z<)q^!x7<xdU4sIDN2VV?QwDb%#%W^=_#jP8qx
z5ma9MQzcs6t472^YEro$NC(oI4h3B?%%gPqV^x(h;k_%)_^}6a(vZnF3zBj2o|S6N
zaXN$Ay?96Wjh=m3T9QbydKBOhkOpfG@)@RY-2ecsy|*Q~cR8uy)kyo>6mB%BBuu&7
zM`Kx139LS5%O-gTsIN+ll-18S5qeLS#g(~<qTXFX<Eb^o_@`d;16x|R$W9r@dRU6S
z&YQOE&XyfjsGHepa*>R;)~!vF!E=zS#du8j5$Qh&WL@{M<D3JU*U~Q<^Wxq3bp!17
z<|W5)2>$>YaoFXl@ypP`_{7=q4UqJ;{{Y#IdC}&Sw>G2vG)CD6H9QltFe=0_=t1dH
z&5_w?z9LJNTU*it;{ne``qh1R!wTs4RtxiWXvcr0Qq>rI67o4crCSULJ;AHisk*kK
z@uAlE8L0)il&WJS*MH%^4+vpdr(6@rQ&?kt`l>yRu%Ei3-&S;QaIp_AKfU~_70GZ<
zB%0QOG?aN^ce^Ckq}yx&Jxx$Df5jts-jT`ttI%urH#`~3-$?LA8Df1+PXWLK*1X1h
z2-3FcSEJ-trGnX>W;=9r9s@t|Q%lZ2a7XGZqn_Cy@g>tPb2Lcp`H}r7sr*(|7ih0T
zk=7e_sWCs^#Z(8Sc0*&cpl7Gnkb2hW&PA^c-Xyxq+UJK(ze<~2z}4I)cI_DatBQo&
zu??e>85E&n#-HJPjYurgt3w}NoSM<Q(;mT;rb>$JnvAY_kgDS)9)z~n1Y))A?t@@v
zv09^{`?4-CJm$AFyZyjHDdc_9YnHTc6L`C!x~`fW8Q~u{@d4hs$znX?t$CP;^Ty}5
zfUTvO#dxaNCCW{VZWyd3Q|VYo;YRgzn<N64r`D**=kFLb!6Ts+1yP8Uo#~8PDnKbH
z0j8(4AQD@pC$Y6Ji4(oVc^gL+CAmkLao05oI~_BWS$5_^HY%K)bg05R$Oglc(v^(j
z{nI_|WHXRdftu2`vOCvz0Ml%ktr^?d4=2LpgUZI(t~rx!w5orI(B8pCZ|Y=DjS<Ni
z6vTBTVw!hHRF@@7GWt#OfZU3LWKG_n)=P7~y_^#unjoWc9xAfShQX~-=2o;Cw#E3E
z81bBfI#ouAOZ=ZJcBK<+MWq;>7lTkus!Y-hJV-EpwMWL1Zr2$>$W#KfTOL(L-A2xS
zfHG(#wH`!nyyGI4JlJ1BI}OUNACB~O8(JMS{wBB7=Ylw(bR+LEIs8ABY+JTCs=lEJ
z+BP+9n>elg6y|nbn}u34sk0rdXt3YIa2eTJL@UAeuQr$A-Pf8#R~%=Ip`{`rQpmq!
z;JGxB=SWgnG1oZ7V{2C0#l)UiwSOx-1R2lPv$Hi*OwW%`xq)*u5=Z{>;;vY?@50oY
zBd4|@P*tBdC`Cw_%lLqujzve{$ceZTe|k{0V@IxPAMx&k)vB8GtyzRV5gVmYy=+F=
z^rXv3<|BsP8NO5IHzdCXqL(q}_b3$fkvYY3v#A_*rV45jWOJIT!!@$l=X;u76}DTy
z64_h1!nA`whc)Y-DDh-I97sIKO~c8+RbU7nwWM}Asz;hh7!iC_uxA2G8BljSfyHz>
zH;uG?PT)ffjwB<YN*s|s#z^Vy=SkF*IRdK5tZ9MTIWD3Fcsq!wa!YWGB;JKcFZ8>+
zxmM12C6PzWA|oT~UM=G(q?^Sz-eVToAyv;|SJOe#O-<awB9f(@BGia@sbwMgrqD3W
zI5sJRCXRBw>f{&9htwXR&?3qb37*DQieBmtGg?qzTWc}>mrw#5urRW8+<oesa!Ede
zxs`imKAm!5nFu|J9cmEysFHDZ0e4bh8bBDuOBA>`G@7^+H5T1;Nc0t3n&^+hfJ|i5
zgn6a|#4Jw>Qmw0cPy(A)AHzxHpG;B#rXVSe7JE|yGPvzZrhq4j3v?B^;dqGg6`tke
ze?O6_mcr_KghfuEIpQ_qtR&)V<J?9+TIWWrN1RKkBzVf8+)ttHSY9;MKEtNT<+dG$
z)%wwEvX+72dvCJn@Pqf*Zf|mHc@1;RR&_<_XxZ|$bG2&?j3)IPiU_R?7?3v0v_7X5
zOT`12zJ{ClS(JZ_jz)f_p|cycYjN*bGt*B)zK`XM0qIa&J_RO)24!8zn##BWL|}W?
zEm@s6x;*PskI1~4JsXP5EtwSi*P$Jb3GQ?{w9&<*Mndg(92(x!ExfM}!y>o=RB`HS
z8839Gzj)#Gi;;DyNW+x}Xs+5lUMn93TQ)#qfk5{aykqXz=ajwco_Qk~BzX^3VrjUm
zF%pn7BLXv!cp`z5++5FhC!29A%J>6woSKdkiPYzzq{@k{{i55p!ljK{hAl$q0Awhq
z(zzw0I<+lF;jc6OY_7-L)WgEjfy1z)p2Jk`MJ|zfXg5MuWgo)BnuR8gR`R4n6u;eW
z^wg6F7L5HuB~Zr|gx`4AM0Ppq2}Mzf)c8YI`#h%7M&Y?y?;*M9bHz-NY3N^y<0GwC
zyO3b>QX&g8ypnmYN5)nHE490fe6kET)~Y6FIH!?H!>wurEy1SZnMgyxsL7-|jbp_}
zNrM4XM!BI47GqE;&H)s}%B8lxM{3y)Cb3dF;+i3qZ06Z9SNpZ6cPu0+i0=7$%><bj
z61I5amde-7aLzvp3r6%2qGP38K^ga{V>zu%V^Es|jo;Fvvw+;)#R`sHfUM-~u5TWP
zyJ_%?dpKfIfQSYv>Ks?kH+OT_OG03CPOTy)AvDG{(GnRv2QeS5X6(l4G01A{TO-ex
z+cO$}GG}jkm0r19x#`7OBu<iZQ^z|g;fSQ0LX=ZlmJ;$o)Pwb@miMMswv9*B<enNY
zVpY10yW{Im{nUf_O>CTGa=1>L8g1ia>GY&)Uq+|w*Fac_Ey$YQKssmXOeo{3np~|L
z#;jx^s*=j^wnJb8&0$%zlSfT7rK2}txDOhiQBp8h73e0KJbA^_mC;?{K0pi2Kys*Z
zF^U$6n?`D8Y{ZgJsi{#uO5no#)$Vs!e6+x*^vy#7mOWq|JeH52V_xKyqInl?=(l;x
zpFk?2tB`ocYo1BDq>idmN}brd1~71Ho!*?XjP2YHQC*Z=shqWxDw%DdpxuwUdm6#j
z{2L9_4iE%<e588S*Qlo(GdeUmWq6)@C4}}@#@^yKRzuE5Tv0X)?;R`4XJwC5_)8L|
zm47xl^FYtGDz}0!S5Ww&s$8$#c;6qG6-#7sOX8a|zAs6;#MbNm+kAiH#<&t{Ugt$0
zfej-g)|-N+m5oVbmmPZg*HfT)o68GpeA!-$A?@sFS`j$&BSTs6>^h1Yt4u7|&mA7Q
z`#V{qlgVhvkQ@~woYv8NuI6#~*D2^881UDZw(9M2-BnL&>L!>d^5W~<`d2))mCnj>
z=aO9!3~7p*F_~0**KH`-6N#P5-Xxc0x*u~?5zBZo=hdT=`B!B0!=F2Q;!-?F#z#>?
zi~(Lz?wU3-te-0!FzxGFaednK@LMy>$8L>32TilZ(I`0FrBA;ZuIBnP{kYO&g94A#
z(Np-0x_=7vI88+7Q=k628m<Nh<6V&0?C2lkRj=prVL0HPl<Lotn~^b(G?FcND$(0j
z`&lcHjl&#ddsdf;uA;Kl?M!L`mj#&i!LBMTQ)$2I=Po5WvuS@;JrhrO?kr-J2qBbn
zSz7A^bRDa!XD^xM#Vf<r7(3N@&MJF*jzI@Ch8C~1j;EoHrR?@LEv|uHYV>m)hEswo
zX+-9yK8HbJdb7q6$lNNXsiy5F;uXsO01}Gw>PF7Tp@yF%{Z1>z8e832nIT=woRM5u
zU55l<*Be$nX}i0Q*_vxbK?gH){<rX`shEW(Duyi-tO9dEpa$b6q#Xu2RI^BKNKgdL
zKGfpjKC~KD-4~un*DOw09Ad3abYQf#xm+BRm8nQhRz_5pk0h+!3mnp$ss2z{f!dvs
z9PI9%V`<@5%V+l){OgcQvvF=lFiE8~ZAXQha<`dY<Cy)$X1cPrR)}nNV_&?iOC8fF
z+SKQqu%RV#C0<ge$}Cd?cAi0{doBL}G9D?nVM0FEow}kTvP0{NY;p#WXNnP=u9{5i
z{2Kw?b8t^wCO=x{{6sUY-8SrxFx7N0n!QOGbm)d;6d9V4+lI59@%LRv^QxQHW%8sr
zBkr1#iL4_qyo}594wcpE_ab;>karkAT50N1*xR*oNCLVY69(*cCuyybl}(vGsc}8M
z!PwYRI0RP}AB^q=`&&%WBCBAXxU8#5#a#}CD9>`$*0l0~@x|v4XR1%)t0*wD$1*p6
zlCq~=Pe-`wVPy@qMxCCurQJdw?GX(2ezj)ZH0#&Oh@@Qht@LL-r4vv4GR308#H;@4
zs78ruRE2j3+|=@+^T@%~{3g@J+irl*Or=)vr-v@>E*3XrA!Of>dWw9ek*d0#{-OT>
z9_jw~^r}N5t&<Og8>M7w@3!56QsKBDb*PrcvE0D>xIn}!ea+pErfLvNJ7f25ICSiN
z#dDn}HR2e`-m5p>BL=bv;<k>*JhwVuhxWf}@m;z90G=vj{J;Hmt?@<kty?B~L(}oC
zIv9V;)Nro3u9v`v{wHgiiEM0pR896xDdpoTg}+MkOUE)Z$bX2EYo5PEJTKZ~J{rF7
z5Ndq&L66R`Jbh{{JXLXz`Dz58opi5&!v6rhXEmBi4ML<^MMQliA&Ew5g8*?(4rz#<
zD~3amaaN40b_ejtIxz1|Xa>!$qPo4P({&69qZo=q*XU}Q(dI`gWKoY=RcFmEVzrYJ
zT`uA@E*V@7yH84`At_m>aWv42CTaA*t8k^}C)$x>&1|*;SnV@b;<Q2csewAh8uS^d
z2Ep2u#v7B;n&LJX+9(3M#D}1$S|3ViL=6sVA1e*km=i;l@0#emFyVEJe4edI>@KHw
zZpF0;K5U)2q_MgycXz=*FvVbKMPYjFtzHad2-o?V<otDV^Ih9pgTXm8MQS$A&cN-o
zh;+ga<yVpGneKGoLt4v}2Cj~4SnOjVD6I@x67h8(`@*8}G6QCh+%^S2vdpjh*qmI)
z<~3U0PcK)t_iR=NrO@wVlQEB_Rk&=`v}G5mmvVEO%$4z)#iOY+jPX=EE|y`@$3I%;
z$;UO-PvJ);O(YjrliVUlD`ivek4oQ{#IdJ`p|ydJoCekhwNs7El$tq><dRP$)P~|l
zj3H(!It=hC%-b85G-K;QB8B?WZ9R<x0~~|3EJ#_gNtGSV?Jmp`4mwv?V`u;zR~1dp
z$RkN@@H$hxziP>|c%nHW&uYljw8*YwJC_VO3_6NRR%J>zM<J<6EFs{V@2`JRSh&r4
z5ZLo-?kE&uv@snVjTDOLaLRHGcX}SSmzLl$#ELM&rHtQ->jIWZOv3`8MT=k~jMW`(
z&|cU{AADtgv?pRuJ7qDsQ`8(0N;ws<5l9YcqL{|DG>Qru!lB1FqyPm^5!=lJL(I78
zO!Em80N3WUZ05zj(@hY>3x~~G3JPYa5N2H*N!N_k%Spg&_55)`Xull$mpib1>S(5J
zf}w$td`n(RPpxiv9>eXPAdQI1LN_1jT=ku)wsuihl0AVi0H?1t;2xqT6ztZp6A+q=
zoL5W?*<#%O=>2N4f%aEFTJMU{9(4DYsyR63%eVW-`PIng;<!&$Y*<-pWs|)FCRs8n
zReETKJT*8sy=aWtB1*vU#;04C@6BxqG;>s?84mkU?!T2iZX-SMTB)w5ET~CrE+c-M
zO*VZQZX^5K2^30_sH{iUB|)aips;u&3A=VX)0!Dj5(hrjqJv14iKjQHBS^6?n1`o4
zQ(UBoM_L>Abrr2HXlziD%X4K9;AnI);1P<)2~U<yomFdd#V({|Q{s<!d=5uS^`5Bn
z<m|3clJVJz+mW0akV?`4fm*6*E15cv;TIcv)OQ#5)~oyJl~1O5s7cOtM$)x9cae{E
z;#+jxwkL7*>sWVwD3nIhECMuMNhdX}N=hFQl^jE;uMLbnQt$g$PkV`4H&6*zQG&I}
zk{v$^@@2W`v)BF{EH<75H)LS3k)%GmN&c12d@ULAz1)NDv@S>e^eOf@Uejd{ixPju
zUPq)rR}RnbL)_k%z=T)DKXz#viO`CFdIWWT71iz|k!>Ybk`_2q-D}h={4;AV%gG}i
zb}|k>ofl6-DW^^IcPF@Ji1`RHj;5kS#Sx9<dJwcJj%ElsrrX?R?G$>~a<WICRr6ej
zUqLWPo9k9Abj5!TyeIhMZ`4w$QM!Dm5}_3(Q>o`yehOT+!bNlUz8Ji;4YY4j?ObQ5
z<DrhrEwsRuD#~ke(sj6mIP2V3p@NsZ;(3_ZK2;+}!iPREwH}Ork*`eEBgV;b$>Z{>
ztNcKY(eymuP|x<2&;4{Y0q^Zykl5{e0XPKavg8eUcaD6tJl0k3E?6$CS5S<n*&6mc
zZ7*$!`+>p1?N_{aZmnZ)a0kq<fPJe>O*=(c@-Xwf@7(DA9_nWENb()B(fif^01(5<
zziKyrtMoai?zQDF)aIa!OLgVH-OY6RMuBN}Z9TZkOCQXvKb3J&_bGEO=c$U4l(|_(
z;?Cz!lWeGVpSs<RPj7XMS9D~h8>7u>#hO;PW+?-Y#<%|1WwSduW*i#gtsQQ5LEgp-
zt{VrpTID=Vp)6sqt&`>cH66X{igCH^Mps8K_JsqbAe~Q2=V&f-7eDKN3W}MKQW~L#
zXw3i_q@=)l42;vlascXTS<wiR%7lElq8N{ur?{nk4JS>x*hz-z+Xn|dDX%1kFpPdx
z=qN`#m6-`{b53a*K&$D}sDrrIvk!x!VsVr}Yl^p-m2keaOHjkRNVDc{`_)<33~C~G
zV!h;;^|td<+*NrYF`cH1>tLM*si}(`B$1ESr29j{ah%Zg6fa6Q5iW^gk&#4=ZpUz4
zjYXaG8mS4#-#cFiM&H?TzfjMS`BxWw#(SxxW6tAO(Hy!;QYciG{K_eq9VLVwW|4OM
ze6p!Ok*;b(`@>Q@npPA3<Zl7F!Ox{^>9ZnA(68M*=iZK{i&S)05*^!pYopO*+jzL|
z!L1RTmW<HXHkXVa?)9%9@fxkRzd75G-0@0#1uulFYjn~4(N0v3)v0f*Lvp@d%xe9Q
zt#H(*)g6?oTSg@C+A5c~C3E~WM#0FwAQRWtwx-)!nAAy2L#P+G7SJCmXHTJ}u-EUA
ziSvv90KLs<n=x@rb?1riqw5+-Ju!;Z@XoULH^{GOBiugr?3$>{TM_J)u625g60xmR
z#sz4_4}^`#Ja?>3VnV8bV4UY}dWwncYX$EZj#Q}!CI&rESj{l}pn1Q-cJ3X;a~&r+
z;ss$WME1(oK^Yaib~)v_+59iHk+u1*RDSL_O0oAOWBFAdh*P`xzTp|Nv8sjk;<V^<
z{{YL=n(bYm!`(lAs}$|`N&S0PH6_Z8p3R<@;=8ErbU_y*&0x9o=Dhmtw!+7syOCJc
z)<eTe(A3TNb#&4Ax61>G)wB3>N5tj-0E(wQj0yc~t$ZdQ^7P~BDosM3v<(tcQY=bJ
zWU&dOG{8`4>f{~`Feei+D=2TBJJ!w3%$M3?-NPYSBjP|0T8Sa2Vcs9Ql2bZL<-uMv
zSks(L>yv<LP%)xHDpiyoGfYLOEtX~T82eSEw2rk~6iN12!28s`T6)j|g@||kX!4bT
zDx`opK9m7s8xKA&G6NK>e}Nnp{Hm<6h{;!of)9ECi0wTTiiyy6%`_qwPeV|;)d3!M
zR_BG}i8Xs^q+k_aob>5QlFp{zR#V~`QE{}5pwS+nZFS)r{mFsH*0N}{#>H}A@hLli
zNtBPN?Of-I9LcW2$DscJ8X|$SyR~dUpYDpZk6fos&FdfaEA*?CtadYyQC4KgTZz;C
z;Y;EaFQzkpaOSJ*vli`ACkG+RnyH*>R>%Fi=RF>Wc83s;r9m@rYgS_xX50=5tecKA
zSxFr+W(L0(ou|3#Q;}R#R@AmItGN@5(<@`0fTxZP1hPc}oOtx0LrFNKQ=XIoDk(KA
zQEgy-O$?~+boyG13hZ>-UpNd`1zpbidYY!uBObL@RaHDFtQpl37Qq}<E0k6Q9CK1L
zg7IW%vM5$k!14!L!h+k(!6y~z(fASNR$luhpH7|>Spx1x=Odg|L&2>~2KTA0ZltuB
zOA}y^)_^+=FIro;fCmmbir0qkoYrj#no!TVX=H7{)RN8*U4bCfUt<Crl07Rw`Buwe
zh{N8aJ!y<-q@ZIMVxzz&f`G3S=4=cKV;P!q^FV=yc&iXL(LkCkEAv)Sw=M-zGkckr
ze92CUu}Gv^j#g2@siU}M#(4TtU{;-jPRz%g`d6WRHnQ1mJ=q_8?G?d{m8mnk4zDD4
zd8tyoQ`G28DQkET#+r@Ib->A%Eq~QF)~oU?pXh783q*O--d?EUW0PyJ&+!_y9HTvJ
z&wHM<b}XcoVmT=kPLERu;;9hF32H$n6=rP?SE&|S(k5*W-I{4?vQckOfXy!U`-}S3
ztL;bZj1?3y9Yt*y$qSxsX#0Fev*?QUdgt05m7xoR*|S<7ZHpPf{HfEdq^`_r;AIDB
zyN(wyk&@W@(}TkO8BJ@=7LHd&tU3USB6v|vdsv_E3UNtnPqdnPm88<+WAeFQ!!#zI
zVh3w(C)bMEOs7`r%%S~{GRY$_R#!WK27#x`5Xgw&Rr_1$Z5dN~vmW|;kC>z`03Y2o
z&V80io#vP@W<49VchXUXyPQ=gD(UtaEu=vU%E451B-a<>T~-@`_B$R>ADPF0YUagN
z-0j0rP>M%Cn~0%k)yPKs$`4wNCMa-!D)SE0+_d|6wC{#RLj@99v*)KQr2ha~<h~Ma
z{{V!hP~AndbNGsHbaAJ6O_~1y6qP0Ml);|`QAg=qciE#}UH<?omi0O*uW<+=hx_K8
z_GDr5KR)za(2byx4hYZEx^E8nMtxgk(`Go2L5^Cot5cFU>T61$Jgw069T&p(dQ57<
zH7*BJ+O>t}mfH|zJJixmZaG8R$@WCEBvV3Pjt2&_`U><Z#OHHrJgkAv?9(jNo7Fy?
z*78~yC%Eln{d)fZyIY!g38NuD^UwNMJoXg67@Vxq9qfJTCi3K5GLCBoj^htm@Z=Y;
zvm$`l;ELhmNi^L=+sn|LR=8@MwGoCU9Q4_LN4R@m5kV?_?>RvHPAk}TU-<Tm{p0+F
zcU4~b!_p_1-?(i;bDj&Du3UBX726eLb>Bjz<;Va-8Oy&IsVv|TsLyUmtm%98o9uGq
zXKG3-sOFn|!Mg%6T{nr5F1$9<b=nn{x^X`WJo+lqNW=Jls{M~wSY3jnoE+DvYY>TT
ziT18qZt5@OJXCEtFGg`U1IoGwBw*)?>a=^`D*g#K6j`Dnz0GA!UYw)mZ(aWOFS;~!
zJM_4+A)+V9eXKFm9+iP}8o>VmE<ils_O5ypYHlY(qNh&dp`7DxYevrUSd^$dA6o3G
z6w${{>DiC#_|YP(Lde4)j1gQ;I-aa0q2_wNmo2sYsShd+K&-9Irvza78s?N*XH25j
zrZva<-@>BBWMmkm!KN_9CeulP0BJ%Jz;>iGTJrcEg-;S7D#xg-<+17E;JLLHd1EK(
zQAcor_@ONVrC6$$MQgGhyv5_zt9g(8<5v}-=hc4uU|G7=s|Z?o)1Cvae>$Thv+;X*
zdH{z%#GyyhxR}EL02D|lCd(?U2;G`j3L3IcQ`m6#ed0(Q0!M0!^}tX^T3=}1a!U-#
z$j;Qqd>y8=Jf|nic0K~6^F_=<0hW%XD5+9saiYm3mxKi1e8`N#xK(2-vmfF$Oy#1V
zGeve7Vk%2-^@5Q*GojP+bZftPKaj3}U?iGDv1;BoXb6vn2DFiu%>;~hNW6pT?^4Ym
zj+;xJW!ozK)?A9{w0UuN5Bp7;)e$X?y=hkwvmV$L<htFdy3`$$8-XL5RRWK|V$E#P
z97`K7y)`0UHoaomW%x@hWERacPC3*KHL)z{u1+n_(zC3k%NhW?F&j8d&Uy;f@S;im
zs;uO>7|Un$ryq!aVV%aJq`Y5BtjKZJwtJYv;VQ~@p0$apGQ7$g7#;hsy+rmjhs05w
zFF80YFvs7qs!rSZ#ubLw9)xtPS-Lsxc2aH<HvSc?tbR4JS{(0l(f$;p2(<?zf=~nb
z*A?QMro8bb^sStUSwAC>=~vLgKg4EQbJn{5019d|#jDz_s;EU^2_reIrf+Mej<?1-
zoy4<ShPNg#q>!h-YQ<x3HH*A*9H-1f3=)04Ynrdc%!rIyf<`Zc*=>Ar1320rAOY?;
ztzU}~BHma^1{u-AA6#P<*&SkMEh?kWXryW=r!55{LDHIPN)nQ&mK~`DFbva$FeF2Y
z(b291cSZ+4yO<R<FJLZc(Hm`Acw4fX$-4glkEws&H4AaFDk_9)>99>DqO$bKtD?$s
zF`ChAQ19%?k0p&C+02SsM*&AtO#oFlw(w21IKI&R2c~PG(R>Bs9}*Dl8M3uwl4Rh1
z0+^)IJqt$oRitSkG_CfjamzU4@D*oO@ExVvVg?J3;c=1l2enp>9O`mXxx#B62fwj)
z^3x;fl|I1r{Ogywu)KtyG8_!!Jw+>6=uJso6<P@Z$x%=|P5s<c0=uR&#YU2~xg>O?
z#9Ok|_TRwAKYV10*6}sRl_Vu!_ke~|Q1==v+~30-HX4+K$f1EC4!*UCtZE5;syy(Y
z-H*9VAs&Tl{{ZOm`=I_+Za3stA<&MN!-yB@S3%;mj>Jd}6-I1=V~G^{8kblM`aHPg
zm8x4a8T?T^`a|;?rp^}n-1<pe=cISG=X45vsyUaM-JH>SlrC50HH$xb&7-0$%DqkY
z@+YdA;3XSmWj&2mGHQ^wdW_cRGYWFhK`NY5o@s#7PmE-aC<4@Lig@M6VN7Y8+C+hX
z!Oknc(A+Eo-n{zSNc3pzXvlI6XWSn3&d!ujd2nlt@m{$90BlQpc?yWmODXSK(T6n5
zYtNPhceXb2X+C^^yph*&kn%~cd0asYZu{B5728W=imv2oc{&0&ou{=i%FC0{t5X6K
z#U&vVT3ofP>mrlxE3VV^D5q5+hD|0WZrjeGbDk?RSn(uQ=+N7c@*b=@EkcdhmZu=|
zw9_<_F(r>A)QWRj3>!5O<F_=xZ6;_b4JRU@Y=b}xzuJ({Fck1{Ok$3Hl`7y;20St8
znzsvKD$$!LTyrHXW|2slYE?9Z#xUI}1G%qb@U@__&~2l}30=qWuP+|`)Q?9BwYjpr
zYGkh^9+5Q}uBaI?rkKXM;9|+`Pwt!RRqmwn^wxhxuK0W>pFiR?=6SN;d6UVv`^KzA
zbs=noVm_6@Gt`STV7d**Lst@BD-%-0a(mS4CVSSAA&(7DSFH$RxA&(ko|Q<=ldI%Y
zkTkL?D9I+YGi@byZe=VizJv2MgDj0JGcn0ErCkz<!Gm1RRUcBdlWe?U>MLJ%Q2i^O
zo6zs7z2Qq$P}@T{eA2{Ul^~?GWU$X742L;B^&r2QTkmJ<U349yk;z(&CnoGnHAWq3
zNcBaL2^h%ruEiUxo@D4**iot|Uz##cv8JxGZd+>xQ~A|Kn|m45q~c1NjiiSg@%0$0
z^IzzTFpgR6JlE)~MomTBN%m8ZLY}3dXu8S)a}qLoBkc$BtXsc@-W#}#Exb}&9R1{S
zf1O0V?aGa5#LK$)J*TuRE%gc3PvPDH6^kdqT|kmXt!zi8#?415vc-2tUE)6tY8O5P
z8ZCw-xQ!z~cWutnGyJQZ_-)~xXG8H!$5FLt&|8ons*{o4n_80^k@j*&H}NXU_Fo#_
zy4uRoox4~m!TCjUjbnNT&21<j?wZZDbFQ0Qh+|sZFIZfD6snRhRAKm1x-99uE8)4d
zYlWJ1Zy`?uAAY@iP1B;WgnXb7G4vJ3FMg+zJ3b0l>i+;C^GanqSA_m`9ssV&P;XFn
zglsgPF`g<nY>Zc<7}*|mSzLs7r&tg4!~W@_UkR5+MXI)Lhp?^KA7+NF{{T$CrC}|H
zFNP-}G?^sSVS%g}+=~&6sU54%d}0xGRoT}asYQF&CQnNd=_fjcvo}#P{{TAnZBavM
zUT1Y~KPvSqZjUl+RCz3<yw2ylW~koW`D#@UIj!qd<bm5xpEI!CI@Ff)q{(*ap4EC+
z+BPraRf3I`+2-S4HBMAUk@YpzOsjk0pCU83n6aNiD_mUd1!8Gae94>|teQ>qGOJ~j
zh9<qINV{EALW*I@D}$4h_}48zb}l`b@oMu;bw@d@TlsPa$3s)uNw(T48}8f682u{n
z>h5Ug$F~(t`9IL<rMZgU3ysJLkcJr@D+%OF;9XcAam8~`QEc>~8LU4)O$cU%t#C;-
ze#%spGoEXuw9(5}JhQoVN99q?Am}Te+MN)Uk<EC508wrS-UkQXx!GWjfAx#V@7}VK
zJ!)S4J2R1-{`2st>wx70NN5;?tve!s42rRD9!BhaDkSWVj68X&4VljFc&6q@04V%v
zEslpNbF_%s2G1m#5)vEItmH}a4`tiK4h>b7GL4SijU>lbHetAw+m8Kedl3HsZh;p$
zGA{$`R$$LY)$|qB<BmCIF@!sY-g8*b;HlJ)F6KOcv<)dMmxr2y)NGy=)D!>}Bm?`W
z6`g;e>K4EgZ0G&nYZVrm>{P^4l{@t_jlI-rOWVk!)TyfSoQ`^m($;4j>T}Z{ZxL?X
zQqO9&E$VFw9N2zlM3#(3RjHCiGV#<=eM}`rQH9ap$s*f$JO&&z0DdG_42>Ldgxm8I
z)~-f4eHSw+&PSZsrGtl|Y-2BtoAyn!M7jEt{HvZX%^+sBjRV|<0PO(sYjQ0%B`%YG
z@G!^WNtG9%-3s<UI$LGq%VV^Evt53c;V&X%(J*Szo~Yl~WL3A0M>zSo{Hx6Mg0S9+
zH(_0}Xqhy=119mNMpi}zPiEvAtWqaGIvAV#%hIvArL-qRDH-|0k6LD%{{SHa^{OQ3
zFHS8F(y}znmp@udU^QjjKt$n>Q|K#q!EqCQ=lBLNI{q}`z2U!sovxrhc>a}Dx>n4_
z9||a4a0#w<ca}+>SQ%XOXxq5<tm<rSO=?3NYbeGJ4+RfZ_NZq?h@O2j?}OH}NYR{z
zuG{q)InFq$OTp)%t(M0--1TpVA#MCEJllB^F(=dkUSX)tF0~t*fG{PMpC412r?HJc
zjzx8@hr_#(JjEn7{s^opd?!pdI;kd>bqzC@AYf$HHl*SR+=a@x!5`MSRG%@tqPc8h
z=##VfvcZA>0GEVZ`(#&L@m|=xRyQ4!a{hS#01EGl_f83a5gunnBSFA*rCJJGk_tmo
zP}50NOA?A;6uG7XhQ&!7xOHV1DsxB<iq*GjaZW$wtX!XZ$h}>+l8^R{X16Jo$fPwy
zoi>C|_G*Bn{KuNn-bVpzL{Ns|hXhuQgqlvJXxer21afi@LsKi;LrQkOrrw94X}T^X
z`x@PlNfQ7I0qKhLEjL)UxriiksAU){(2CYFQjLyE)aMDgbTn2#{W+-u4yKdP&N_`~
zRr9z5_*OQj;e9s!2oZTsMmHb>^yy78>m5<SYaa~mqgR&I{KOoMy+`5RxP3!H)GS$+
zR6KR#k@cY36Xr6ch>Y&w)DlCH!KBM81bAf<00~o?F0SXgaMQVB1_pZyKt<=sV-c5a
zfaETCr^fv$i1gh$VKtO44+TYNM%W^_-sb03W&WW)m2`BgTM-~Mf~ba9y5g=+hts47
z{&}C}R}%`o4>og#H9~g9#6R`WS2^jPfnsxRze<)>1Xkx7+|0Th8pfG<HJdtOW?T-n
z!RrxuI)kn-4<ej53Y!_s+|pL(D;9x3nJDQ?*QGHE1vz*$#7?IIox2nQG_-4h70O0V
z1Dsc-X(Mc4dshY_d&u;#@7*+<;8hup?25si7dicI@L1YOE^t7=`q!CU%+uU6u_q@r
z*N2UhXBH+`X%a_&a|{Lrl<YmK24-%&*Gg77n`~*KnIayPl!1jy49ywgh$FZ(2)+HI
za$+gxHa@jgn3I!Cs}0RqjkEYtC6b2Xnnr1oM#Bdb(lbn=HXUj<8Kg0k>IQ|Pn8&Hb
zI2_Ue>(;GAe|L=XDv=Eo=07>4R_BTeM9S4Voby1LKAi!&)gXy_vx@e}6eK8%z)%Hv
z*@O3zJxn6dohrf#^P{=k$tg2ka7Ho^nu(2d!5Ff8Lzvs^R5kgM4IWh|8-~$}>~7Gx
z!1b+fGao|I>@DC7(5~aoTa#Py$86D&8mCV|UR)nE^*uSJqoIkY>G87}3}z#rzN7ip
zOvXzm3pn{3?zrR{)==tHWLb*f5Ds{&3pc$)O&TUied<8<s%A`gnZ+u}%~~^L=`U*C
z(k*bGXmQu>*7VinaAN1nN%@sEJJCG*6#R+6_o_wsshhm6YeF29)JmxPFBJOLuzau9
zu!mE5^d-*~Q;<n-TB@q{CY!auNl4H0R<M<bXl#;UX<%$A`q!fwdxspO?}YavD9P_t
z;b5v{bIGdqv4u9sbY(h;bUFQN#a8y0322X&bB1C3BDf8A%uQNG^WH#tMklp#DvoDN
z;*!v$x0fr8w-SOneCD*H@pp(}U6Jb-N(W%|XETIlx->1mC3uq3FrR5zh#1VPxSH9}
zekki#nv@nZT0=X}AXVG);)Nz<JUlIQcU~g#R9+r})>-Y|QE?}h0(A!!ZVwaKx~5W9
z+&RjQD4j(e&M#)#xpgme+s=;@p8JI)+D@Hw^F6#wf87*&lSKQ`DNo^jK3-Pc6c6r<
zRVjQSXygQ4pH&92ofXj9okw+Y^`?Opj9^4mpIl;=;>zK;Pzw43N)$PCJs9GnDf39K
z*0Z4bqaai+tfQ49@UKRkP^wl^*_EV|Xf;&auIv+oU1~n*!8(aaT#y{|Q`j!+lHcst
z6PH7&#d6%Isw*3(#cgGi!>jA%3Wn}X){!|Uib)%R#ZLu`78|N#9%gjPT&m@ZvmKS2
zx$G;*HG8vvsmm^MGl5%9DO$%oB;DH^X>|6}#?E+G2R-`NyH5UhhEhS%(OdMbd9Q1j
z)LNs*uI@GigUx5ehG{oGO82R+K53|)lvgAhH33+nC5AZX*i;KIn8bzpSD^`VMI*_l
zPufk%k9L`c*w>TZy8i$Tz^R~^nC_8_1Nhc;Ii&fX$m*#T7+D;ip<u=dhH~n}itIib
zY94Krwrz}qo=+yUaOIQpM;<3wr{-zv@SU&PvaAoAZz>W`%aA?lXggJs?v89?=Alkq
znbSc5)GTAkEYBWJ4P<GyLf-D-5<uWKaC=u3D@IcFJ(?JV@Xn32FJB0}o<Icltcav|
zCCZ#FbuV@Z^CbPHSzltrn!_O44OEj@C%@}T&Fn;~!a6oXHR4>xQQL!6e#IO8#ve+;
zX{U3)4U0)x9(!^Av+$^^&Nm7(Okz<|0as&?tV10!Nu#2Kxm#p=M30lrPO4RS7!;Y^
zhl_WS3zNE?8Q_|I4mxZ1gmaS1%|W{>wT49`I`?uMw#-KL<Yu)rD@~g1r<gjnK>R9`
zIj0tp)_99on@sWrxd66u4|>QRGrS*u2*~!$HuW@x6)H+asJwF!06naL_rVpT7m76N
zh{H3wA3~(ka!6FL^(rS6PadbH+<dJxc>e%zkLOYwT`u8psciZF=oI;r6T{VtX|0AX
zfu!yy?8*uBV^B}wO*&qT>+B6nifrnmij*PB%)5^Uo!{u!TYujbT3-iWD=s8S7b864
zn~9upEvA{N;yB~C&_%d78C74xxM?E8f=?8r^&TFb={M$6pZ#-UnvJy8BbHwiHSMmy
z;#oW!M<`o&^sIn4Fdnploqm;QZK6oA$Qz&Bnx69RJ1AahInN)BK+(-<S4|G6k}gPQ
z`M(<UZ8~)luH)2-v<qXV*V!7)Sb##dO?ket3hEgNgu&p{)s=@rH;K;N!Fv$cz_N~n
zRbxEy{H@Q{p;?yGiE`Dm0nHj|{{T-vTC+2L@qxBrIpU~kYH&^sC9rfa&$;m$@t&vJ
zx^D!>i-vvxISK)%ABZLJGpW=}hC~aF1yg5X^0x-IW>EM_SFR$3-Ht#N&6T&AjJEPL
zQZFS>^saig*y*QxTysduqFy$)3`g*vT7u>-1g>y1`Cw9xqn3vcGkLcXHglFV)=n#9
znDy@u`Bonfl4in&ftioF;=FwF$Lmf;RJUSd2D&{2kFziP)s0_->ZiHTNplQQe)2_M
zsjNL#?hC6~QXetc4hUW=nlX2|yqdL+drZ3f2B&RwMtqr7@$@z6e-m5G@U6U<<)%~l
zgI<Ez^QZB|@_JEH%|{9sts>DNq^TIwNmNK0N}vi(^axsTIHWVD@cXQ`{{S${Mt)k%
z)s!ZqDD^q4`i!NTN}(M-iFq6vN~?9;-_E6zQU^O$trS-`Q^?#QKS4s-yyq_)ZtQwu
zn9;3aajWU%c^4(ne*}j$xoP6-IVD)N1p7SWqxAfZMY|hDO4es%rfZWVVB6d!%tx~x
ze?IlKX|6?aAX#6?Z|=DMy(>t`J<eLR9Ce1qlieucTX8*yG_dX@w15vvzFQ-gn)0zj
zSn&pzqrzm=E|MSz3!Ibm73BUZ_=Ekp#?ML9!RO_b(=YgXQd%2A4o>D%>2s!wB-*kf
zP1Z6#`TEyA<&0Ko-0GUOh@^xNYMi$C&MJ(`g;`G`lra<t?mRQ72`m;iJf{`XM<8q(
z<(7=yY<1Qxn{@hB<vdlRIc_O5Ddw3aYrd6z0DEGGp%^uBFs-sY*7ZbpGN|aTK>X_}
z>A%t;&-cZ1Gt*lhgQmW1lj&BR@m)O*KBj%x<BG_*UbU2th`kJ(p^VoZ;&?Fe%scRC
zhs0F%iO*482N1cVGyw3`BK4*On`u-G(<GF73OJ@QV{(odbrtHK9lTqsIA0@iUTcpM
zv`p>8J<+vtaFvxTyqf2+(admuHPpPxMWk)Q^50tIYr2$qV2%p1yx0zEMXeD66yqYJ
zp^1ehAPOiEB0n6{^OH=Wr|VKi)Dlmc15+exa(Yuu1tF|5DmDUwhGLA;1R0|=#yfRA
zX#ntQP}l~Tit-%wrix=k$atw$sfnE@f^9z2ucNT}Vg;+!g?Kpk``aG25qpI34Aqnv
z6~_(EsR8RqYqAK&Ls;UVk*m6zF&}F>q%KMCRW%JdO;|)^C1OX&9<^1aXf$%hKflz!
zZKqqnfkJ|P>d3c?vI_H}j_FCVDXb%vNaTJshkc<;s=_uBMBjQh<bEENwb-L1^kxj&
zb=8n93bV#K#(wbndedOMSq4GD9YO6wT3s5&^kuZ4-lr4QL>Vf{nttwhs=1RdX_l>f
zcB#(M%O7gq)o)V9)I%Q*!vh_KbkfoeN0o{F<#^NVbHYrXX-u3}XK_+VF)}x!R=iD+
zwTj%*TN9Q%VzRXjL&$bc;<t)YF*8?N0kGM{SQ^AFkD8}qi)nDO$B=VGbKy@t>seLh
zfiZ>TXI);j*NF=92PUG_FUX9=8(`y;>0ENE<z#dyC8^H~D`sY9j5`yO&B>}_!q)0(
zE$3Xa{npPot)V7kIHqOa=o8xl(SGqgx{AnUDo7xM?_Fw2NaK`TSsAUL1+IK!r#9fK
zlLjS@DRm})13S-%cSbAQNf-$tg$p8e210NvGf9(Mmia!`@t8{PWf<bHtk*G(DtDvJ
z*yuLk@M^?Y_jdfG&G_?JDKzvk<xIb)>v!|sGCV<|0030r=e=(w@_=`+ObngfE2foK
zGg_E#NTPo=Tx4L6QAR--71s+hl1axxH=D~DWpBV{tF`r^2YiaS>K}^ej2qa>PRisG
zJ%s8;VPWh=Wlf~%_Yw~*6*>FL=C$^dg0at~NhjF6x*T3m2`gvbqyEOYh?a_Rn&fAz
z3hLI9Db1wcWV0Y|ebuQnGOmIU#&-z7#wsja>PB3)?sJQkBorY1D#Eq}Ye^SX=yQH8
zzVfVEFxV_HUTee-*#WcbT`oFv9GuSlsA5MY+@omVcdvd(*ZvULx#2C8{z9>OYE;+4
zczo{CNPF{EEzkunbFqCpnp9UkOP%<+@>47{FFPn^<Ewv#Sx8be8NdR)R~>H7Oe#tP
zOu+oGE>HKnS54t4_Q5{p+{J;ZmZ-Yl36-fu<X=x8P#UwmKdFrq#0vX>jhe2klI3jY
zoE+)99ge%=8)>b{xBcTV;}y^Ne^8Ti`J51IDogk4?m4j(n!!8oxzt<hiEE_!Fz4+B
zhzHPBWz^Ev-(-2-d1gJyp<&mwZ@G*y9?G7j88ud7OA^MZ$#*N0^AJ6&bxAE*!#qr+
ztcY4yKg54JVPi?8bL#gSxYN>ijO9rboKr={PA=%;?myQ46%~B(^}<n10*XKtWar9J
z?Ma%$W={eDGg7%54>g<I_UI(2mCELvBl$R{*~dnjb8Lym)&TQSH!YkDQ$}^8t&4Wd
ztsIND7#XcCEt*Yw#2?}R016i8HD&IXd|nyv;9lNde^XrhN3~57dY*<7l5`Vf18D13
zqFkmy#wum8t{N+v$R=a3dWvipjS<=y<LgN)qe_(&Dr!jjjlvz!%<6yMt2Wy1t+;0O
z$Q_6kY1tfdqZ*agnis#}EYu@AcE<SXSXA)%;FnnBw%p7>%A8eQv!1;O({iyVjb}e(
zv<!ION9kOAl4J~?DM@Nh8l|Y%r7;a5Bc(7^BbHjpm!WU;DZkwUswWH&y#UVY(~EB~
z_Zj^vhM^2sN({_=wHyK5)@|6dinf-J!TZg-MtJo#(pgL9Ixyf2aZ?FuYbe=hbvpFl
z@lDS*e}EeEnC~~PLsG4}BMqWtvC6JKY#Og|X9VQ&gX$|3?9PUiie@h)GJ*|PvfcLH
zJ!_?Gay4dkX>aC40UfH{rJQktfGUui(5b0KpOttd^N&jCyb`Uv7F+^b1KNoELznQ`
z(_WwSl05*bQ{<@wiqwewD9Y4se5)s7p@<-0_pD9OIP;>2p<leJdRH}8uVbQ;cDRfY
zzkL@mw-TqNRn{dj=`2+J+^_Va0X>c_1w4)`bC~tLR@F5B0EbU%8zF7L1ozK1<H<yf
zsOp%H-nBGewOlEJ1_%MkYRY!Dr%ZP`D?J(m^G?k>{{XaV?3U52Wfc`hI|}5Kn`ca7
zk~tk$MUn}8*p#$+$;WE+e;Zq+*MZE3pD7X7`B!x*DMa(@$ChaE+)yc+4;?87T15{O
z8bcarszMOb8f1xzTu=c@4>SzFWvESM8~HfPR8mOtxspScW67vuOIwVhf{vd+(-P-Q
zS!8?-$BNcl8z@I&ed^N~k2S1ZC{gsP?Wm{lbNEmLKHDR76&%p5*#>0>knT@y_KO@f
zjl)7m@aG`bn)ur0DEyoFdC&U98K9`n$4}wk5+&P|v%VPvJ6n-gJVUN{pj&x*UA$u>
z232gI!_u`<(^D#xV&%+^1-v_`UryJaA-*w;<+rw2{{V$7o*>ZlNX$A+3i6IT%w(U&
zq1~ISK^*Pot*Gj=Z<&Zp@U*C>9E?zA+j_BC1y^{w@x?MqlEI$UUG0zX&;`vR^{l5V
z#DLY<+3ITQK5FNh*5<L;^bICYGR7tcs@9}^DXoZ>;+ko2q-VV{NG{J6cGb5+PpB1f
z+{UlMdG50jgHfFrbG(}8Z{*uNIR5=}z0X1}@ke*0&P}>=+O-Sr0=MXLy~?)(BC{r5
z4;2zMi!rX(3~Qa&ZT$NQILS|#RISaLyiPQVO6fS5Q9y^SN`uyz6AWY2(gEpIB1NZS
zfeDP)N#T7}pmPT90m!VX#_>0$CYj!?&z}fB)sw4Tt(B~kF~-ry*11%5vC+-c^4t5D
zy1IE-<gv|G>s`GLJKc^^id@qfARJUKIpUBpw3LYoC;>${H3J{Pr5tpo4452>4trA&
zsyODHM-&Yakx@F(F_@$@3`$xAD$}z-z%@#(^7ZDB5x-GDrht$dkxnXso%h45UzXzN
zj-)6x-zPQmxW{_T_HfG7(P|~D0AN-SMzRpoO?J~p5@b>~(OI{bUSG<;Gq*(*+}eaL
zOcyk)UD2#$XxdAol3O8tD^lQu3ZP(DB&T|JFt__fh=HeutfF+u4hEFq*2x&Moz0R<
z$QU)w>N*^6BS?eGl=WfX>qgod#oZW=WVM&dm|?pf#->U0-kq9ex4JCFH$7_HZ@h(G
zN#Ib--iKkQ+ef8o=2@_;M&~E_S1o;f_m@vIbSHu6E2gXEiizc6m$a)otc@nmT9C=b
zW}VU6_bOaT^5;DWt*chtib3sJ?91p_m4b!Iu6tNQZi@~>3feDH4C7|bGq$MQMJxms
z1TJc1)wCK;=8lzg%kmcDuZhA!>|24F=BX>iBWNo#A`652*|YOVcs_!oa$6*y#*|Ro
zBOb?!qk4Rc*k@h?e63p}DH>unz{waQv-KSg(&;yS**#QK=W%yH;H0d~Njy6mw$pjE
zo|}->E4>o-%n4)p)E*u_)wF7+^)QtdmWOTOtuRfaM!JlCY{pM87das3gIy}vtTAO5
z1HEIdNjnOq2O1<{;wJ({CY-SuPaP`>OQJBtVcg|;Y>nGEptiORfDeCK%+;=U+*{?g
zuKxf`*P|M4G_*F3wOf&(k0rEw*;~wRho>J(r2Znhx6}l0g9{D;7&)%I78-5Iy^lK|
zUW(;iS#Wrk)o>>UGu-tQvT7Pt^mu8Zbtj`7X1kp!)Aza_JgUZ0<XfEfwc%TBO$fHQ
zmelqr*&mf)j|_O1N+ql{CY1+3t%g5^X6p05c;1{NPCTgVtS_~AEmT~G-UmCzYRc>L
zr~1TK`f*&(eGf)7p(S)Vd`l2Nyo8VUPHSiTM%}d4ory#$4p>yZWmp_bus^)GB)BBF
zy99T4cNT}BVQ~+EU<sDR-Q7L7g+Oq3mteu&39@-7=bU@*|NXMhPESvDPfwS0b=R*v
z%DBp8`{$Ya{LBTH>-v{&OP-UvrCdoJ)blE$=KJ0DzI?9o)>fX)xT~NXP^}3vPP*%P
zO&{C9e?ok#l~(0fuv3ffoW6r|zOEY>($pF>i&+(GKY4O|x^-^Qn1k#YSk`B=k{l+T
zX{=Yaqr539nPV4%n7v!W5aP@Qihsk-Ihin2bv_@_PRLiDp@{nj^wyMdAdOJH*k`uZ
zTkJ#hW5wv+z0^5NevuQzzyTunv6j6gcnTS5==C3vO~dsV6)PQU9U(i1yZW*zAKL7$
zU<|&^x^^tqHc18i3%n2XcNC1`S=RK2Lo&xj_+_h$`bl`Uq?rB5ui=|^HXasA)Uwtl
z-cSIL5D|w481*HFQL8|ITD8R7)3sL3d1eea^K^;AKtV_9gs|e;gxW*%uecTD5Zka+
zooXamx^}O!2)>#Wfr_rJYVmphzvjtN2pwk#QSJ*cU+At%+~pf>E#nit<&FyBpI{-%
zD&$_3lOi{%soCRQyV%~?y9%m>;U6N?>66}yKkt}Vaat_$$rIVWl%HwT+Ws*4aoX6t
zc|Wh}rwBuBMt8lBHkN7t<T&9dW*KhTW3Bt|^j-+^OoH~C@M{-0rBAIIZNFw~)pT4{
zt6oCpcd~<6$B(*-r3r6G0GimpvJ8iJN@Q@dzFBri#XI^lv8k31X_r>gj`rSk6*kBw
zXaohU5abKEVzdfp$w#}B6_xeNrv@ase$r%c(Du!2u$ui~Kst<I?L!<LER&Sx0-uZ<
z&@iGQ)4n5n^OgxCYbAVhCT47kg^nGp1Gep<we0j6R+*>SigU6W9VU3yb^%>#PG8bN
z{enRDFxywLmO^{%@eqr9uNbCbPO$e=w>}rn^^|FTjcD`M`h7?&?9i-hMzKXQzLh+`
z4$S{v)|0g_LhGQc?M~p#<b`G0sb{Izz`Z7E$vm*0?H=nA`Rn(HL8`!xl;gnKQ4_~j
z-mtmNqpq5HFF?i|+^kviQ;S8*^1Ofs+#bzi#w490@`=08x?UL#ua7u&qFLQ;r*xe!
zOqHjqbhTHXD~DC_$2D?2Z{10;5^4_<+(D14&f=~n^ITO~YYP-^UJhNHcy(7)z8^Yu
zuO-m-aX;*mcjqWi33=}$!ag(eNMa=^{#@!1Pi}3zai-3)O8&617yVavH?XKsi~jT+
zcS&6iZRWY%(Nhm^gkdT@*t^GT%u{GBF=ZX=m;S&RG0woYk*1f!PH)BKLpGCW!~Ch1
z{&0?~=2n|_E8ka}L@i4xsJ4QotzfxKEr;!o>teWu2V%bFyj-Nq^}2LKEhh@5F_!Rp
z^whb}k;GDHysQ>Dau`%u<WZ~<#pP5m8t}K)LXp}kY)qq<m;FUX3YMfZmc*CR6l;AF
z3&tCmOrTF*Vh!^a>)r?wSV3U=D{6WM*d;0PtU8{3to%-48S@`d^N(_}QB|cqjXAop
zxqiIIaPRU1M||mEqFlX}bI?AM<V_*`$rxvH5Vk_*bunsK5dsNYu4^oZEUg}%5y!X-
zvw`hR4a%Bdg>2B(<l6<6FzBe(s<u=qL%MBPE%^MV!b|m%@fcj@WU03y5P%QgHu{^C
z6z8`do&S30=jVrRqx$bTUD4|b4!t=_D{A28eRpNo1a^&ilYS5{E;};ana@Z%Cp8y&
zB&R`^=He#%2H%~Q?DGsX5PD=ps%7{^77<HZ3t>8S2&Y^vNg<h$=VTU?QEtyP`bw#5
ztJvrjp$}%7e~lfQ)2G%qg!UGf&h=7DQCOO*^VreFbzbsZAMmG4kfrt_wQ`4Y(Q1P<
z8bLhOOi9${vLMFXbdaut4s^C()GR&gLoq#*p>&W0Rq$2;t%gSbfTjv{AU+$CLLuCB
z0dndKB}_(;WrD;}9|Ads2>EQ+jWhhpZ&M1W`tN;ZJzeaYj9z;HVZT}W-nR91R~bYS
zQ(CbY&tx{aL?lh_c6-EB(6S3Z8SES2%eY4KUbM7t!nVMNv#z!AUGKXi*$LLhoToN<
zVgTCBN<upXTkg{0v)Y(LWNK^q9d;#DO|W=wGXHqo(quYn6tS<ThUM5ksj0)2>eHr*
zTaAYAwA22Y?bz$xy(&+C{F$-Q;!DlP7T7*bS$Sb=akPW+a~AywIsKfp;w@uF$PTP=
z!uhVqP%nc9O+6lZg|B#~#&t0t&aKF%G}C#Sb6}LNeO4dUxrCp+Qxoj@yz$94@eXm$
zsM(<D5$ZN&toSmNV_m7FGT$*8n@fMU^}|hV4SNV<-Xq7Gban>%(elfQt|H&Wq;Q%T
ztG9T8D7=VQrJD*>nphUJ4j+5ZeRbZCHWRL3n*#u#$5uD8BNPp$Ml^Co;T!Y8KB}X^
z&QwU|f`!hgeAQ+qV#CX#(;*f@$(8LDA>UGs47v;PhT!utn!Owui%7`m<BI)~*FN^N
zw(r)AG*fx+c6_5Fkxsi2SyeUppd_!<s2CYR{zA7fgvKlptd2#XwXMsB0T=VSvdSQh
zd~HDEX31V+uYHf*U|l43#NCsMRqfUKK6CN+DKbHJJ-8r5QoW+Es8)lkmL&$J52UK=
z;U_YE#8P|Tmv?Ve<&0LoUU173`~7sipHEGnbouoq6zc`A_OeQ|`Ieu(`WDNA^|(Sn
zhIp38ycw=Fsp%5a!DpO-s~W*R?im;16OvwHuQFRAS`f~lE3}<2%i4px)=eHs)nZ1a
zXhLPA_5A0DP|Is`FfZyd78Z}=T^vX}i^kc0eEZOylUHp4<0b21p8}a=RoO8_&po{3
zg0MLK^AOT;0i^49owy;Cg7oOX_saN7+y~}CjsWJLE^3UMB?Qb-WJo2wBx8q8Y*b6x
zUVFT&Sr!cEndu2s`oXfpca_%a$^5((NF@4KG50DK3eqq1)*04F+R&o|UMpjxk+tFa
z!26{C<k*RD7?RE|`cm~x)=Nj>_5xW{UPL$af^CVE)(*>6Pc6QwBzM1$nx^^?1!nZt
zi%M31#xr855j8*Eq2L`2LN<#*H)jOaI}(DAnszgX*h4u<DUc`TN{wKRb_s+cIb71^
zB&`O>wCXiI&P8&O2>NZJj&4kJjlq0I0u7rgeZEW~z1>Z9_Kg0Bhn%h&lDX$>i|nu+
zONEStWh9RBdTq5$Omn1;ZY<VSy0Ndn4lVBNkC|`m=~ek_yel&LaSil7{=Bx{gyIjV
z{Lc@66O^aCWKPs>9diK=EaU!f>2wSnkV-@n`46bm?AfOz*koXo;ymbY5#hf@D1kuk
zp4^Cl^%_{mn{xK8)&i6ei%)RGtJzsz0XK&Q-ehDE0k-_&*CPOuoyJ#1;VWXHX7JaG
zjSvuA>Ax*c^9OzQoHe4j0m#;<A4aq~W3PU2c(FkX5dQoRVPybTiy9}U`M>(!x~ohJ
zF8iF{$ys6p8jLMrseeE>2YQ>+?_g3uM6SR5{p_;=d)~=9@3B%2mZf&i$$$Nynt38m
z=HESVCK`&dS${(jV`!$Q_;XXMrU6`wS+HR}>umVz5QF?QV&@%2m5w+@fvBfE@R~GF
z5MLzURoCimOwFA0N^_uJ6@NhMaWGz<0E`XE@%&!^R`45G2j-_!{3AxC6Jz%$!7qP6
zyR=<c3i9`potQMH>A)xf5Ab}iu4=7Wpb+Ph@Fy*R7jL9Mfu+V=6u;>a2%5Mp$a%C7
z?+=k=FZSeQcT>&V+iZ|0nq7~107{EKZ?0mm4sgyAK}olV5yY#=dq$n{JZ7SIqZ>B`
zZo)01s+Xxq@N(XG1sL|3g%Tbr!e#;<FU9*a3U=QS4>Pei@R@~74!S>A9s$P+QT~7k
zH2;9Q%feuvv49c#do2Z26ueTpr5Ar`$?^Lg_~Sdl%fIES5t|O2N;S`wdysJv;2cMv
z=JChw-K>F6gx)G<%dCNs?}jIC$`8&TXhSTB6E-}>H1=*vy{#Tnh!zuI*aA>y_WUl)
z&Nfn5qlGw9@0aG2O7)r)V89B$mfy9SE~rNRC@z~3>$ax-;LC5hQsr)TUW4s1M;fq#
z1LM{1Df6VxY-MM>_bz~0L<gLN1HmK!Rf2x;h75|E-he3ixmWJ$_P^x($z&w|fW9OE
zF*_>q_fz)ghb#vCQHqN|L)hX~wqId3s;g6CI8yEY%j<UlbXtw7843VXKXOI!_ztL3
zaudLTtBza<<f|;<n0V#}7-If=TrmbcSH2m4q*XO@(hJev8!)T714H7Vezm-$hFE%}
z_&ljEw8nDxg{(!^Kv?h6lE+kx74YON=RKrg5H{(hJ^LZx*=KCi>C7ESi>?njuXKDN
zw;;2ZhqH}xRqFIangXtY>tOi?(w9BBpV61e-g0l_ezn=a@(0ApAbtzHp1S=8>WTB)
zR^kZ$_|4njCN4nMtGNYtLg%A&BYTZLv=)y3sFuL?X7aV8H!$p9{bH-Y8&?C3rfmY6
zODm;5D%|RvhYJCg-j(z4_eIN9&zGs)pNs#1?tTSCNp-#16*6U46RCQ|_6_gWbKo|C
zf1U!?VFU;P^RBK=@w+`>sHD3XZX<PV+HjgfY(#lk16JT*yps~N5sF60F847u&n-``
zdN0vqw1pBpS=}Z!Tm`6)G;nKjn-LynwLS`d_3Ub8IR&zF9L>oiC#NvYc1oq)8#(9v
zMi+m+r+CtR^TfO_z7H5_cN+m0y6ULy38M@^QNaHc=>&9FhrvG8{5MVIXH%|F1OmEi
z{+}|*do!$OBs16(7K}OPpZfrtQvJuK|4doHb=nUUaOwf<i~z)96&x!64jF&*#3InS
z&-hQ%QNV@5|6Yifz1hWl3;3bzUp*sHz@B0*^8879Dc=9bqcNp5yO`!!J>FjcuKky@
zbIz+QfENF|@{<hUa1VQ8(f^ksFw_+Pzx%M<&Q!sXCQ}?`3ier(G&tI`%Zq1wGZ^-M
z*Vawv**cJ+`P*{L(%3+4ot_?509twUo78|6aae~|PIn99bI12=GoVVr@&^>A@ijxX
zZ=)F_8#n#Ti770o0EBTA#1)53c5n+W=B^C))PMvY=MTtFYZv>#$tFmW6XAX~!fQv8
z>GMrNs-?;jjAA};CrF@=bVDco(eeb9gzO4@B0f!pYl$MtxCbvwSGb6zdaLC-`zX06
z)`d>^u8VvQF}YiSSaoTcJ5qb^1-^MZ7Y96tIC1AKQW_jvv?!CHzU|8SS@^Fh3wxes
zJVuApH!%lB(p&hOsR7o-uR1dWx{HKAGm7_pktsrs4a@%i;SXpFVzf0nl^I}}2UzQ$
z|A2buux}_(dcW|Zp_8%E+~MIpvw+tJaHVqng4B@3=K+CVZhZV>W;ZIb!%SbOaWb=u
zTG&Ox0f5TxYps8jDTSGvgbDQ9^Ru*P^sGU@G{388%rE~<buQik{U)Mw$a%{P%!C{u
zsi3WN$YRu1vUyBx&GQp5Np%%H`xC_XHv)h-kcT<;_ry_9x1U69#p1+)S4z83Fz4Ad
zZAS~_+ZB;hF1uaXd0oK7O29hBv(KDvU1Q`9Njo6<QOg4+mH_$yA42|C`0u}HbDY?y
z_$HzH??>myyT4Kg>t!x*Fb<eA9|LaoZ+}V-zImc47vBN&$zO`%weWz1`tb+!J>cQ`
zxjD8Yb^TZu06*A+{RTQZW2OkuWmEhE3Y<E53OOE(&IU1zPHbxWr7%3(px&08I8O?)
z>Kh$}=W|r1f?v~@N09BgCMH}LEF0yD$_iPWP;`hVG29x)s%HkArHj_Yx<!!3(o!4&
zGB$8-P(&zZVi^E>3;f%&@6wv=B_X(4C@Zi3nT5umH;>(Oexm`R5;$CYG=2M1`WK(0
zKen_7NNA)w;9!;FIlVz|gSY(;X!Hjq@I?V-)2CauhC^q*ex3`Q;oY|YQ~=yx^&<bo
zC>-$cy66w6OA@H0y-O*5PPi7ZYo7QAv{DL8jIx3a(e{ARo;NV~7#MT3`8BB++{z!&
zWFJrf@O!2hyiv^W)W6UFZA5eQ6{#4)-xBNpO0*X?j1gcqXjso}!1FmA#_yjP7sdDA
zCIB6f2RfkOj|_;r$sY{=(tk_rkGPWaI~A}#{voIp&>hzN2bA|o{Rs(>R-`{%OE!2r
zpI>qXYyg5|3+Q3U3+F!|>6e}N=;HU~f?=)!o6`Sp`&xjW4gZIN*?$xO6Dsc{>}eGD
z|486Ci0_L3*8`VUl8-N%D6Yu=r-zypp9r^MmeT<n9Dm2dJ<|N14)h0<dYyCs5@-_T
zhk*FLC<6=u7>#JxXbQv)iYv0e&xQeSrzD2nEo5K!-=mKb!hU#NN5v4r+lo@2eBS<T
zLhgzAni#BOqN|~SN8TUW$mNQ|x2;p89%a?0MoGI3@heS-LoM^fXy_k6+t200)kK;x
zyr)9mdzveW)t(D~Ksl~}YysG@ti;5kgm7CN(5*lY)h$6E;;YqCPwJ~LhLkP>CUcgL
zA5svGw_$%xO+8ViYWH26!;Yj%(+CsW#DIcNBsrabK=Z{zHT7ct<0wy9fAjwxEi?0R
zXgzPIhMAhjT>sZhd4+db@<A@&XR7xIoizt)uVbiJPXYT7jbvNgn&J)+oQ5&?iWD$E
zIdS0sk**))mVOKP0t)L_hvH|QuttU_S_kh(;L`z+y$^HBgpa7NG#_<>J_5OW?&`sB
z{K=#%nN@G+n-blKEGBLU=H&0v({y@U4D>$+yo|#~p34e9`vnJNfzZ>oFDNPsZw!?R
z(zA82dbTgm6eJsWkkvDxy{--Y36FVvRxZp<8V@LiqbWcnpa)DUcAfzPW?Jv<d&B>*
zpp5cF2{5lDJs>E|o_#bocm|$=jbP6Jzv-|6&I|uww<ob^8Ee?IYc@;)uzNbJHxcWb
z75tYcnmbb>VU220Yo{-8PwcVR@9CUNLd{N)Bm3fl2;Bi!v^O0H;pEu$hyrNt3jn3p
z!m4b7_Ugy}w-<oq{418KJU0YRt>lFw*hepR-aRY`{U|&eh1@JW6}ml%M$4=`7xusa
zOXB8`?%tI+5VLfT;u^3<w7QFCbE?I9tnVG-Uk2C<`~fu~0Lrv{dsD|?;<hK}@hvbN
z8^fA4MYR3UNH#yR=Kf!$GOBxPN7O%{*ofL2j+l#zTb5Q9+y%*bGfbpdPgV2b+Cg%O
z*MC3*<vrfST4E;V3$R6icD;@W5BW}|)pVV#pg(cLlXA7v$ic55;D+ua7Toe8Bbvvk
zp9=6o_E`A5QSocU(mh=-{Q#!?D?h70zWE|D<1-c=v9tZKP(=JxK19mxq_xYkdT@e^
zyq?qD`WU6rLqk2#!qo3>oEoWDr`I^fI5g`;30q-+>AwWx(t$|{<h2r;p|#EfQLRPp
z@%FHewXCFE2z4Ykw~{3nJLcIg<FHS=C=M8YHb^~OcCDAZHP87b=3G<cRNqSHq&@(H
z8(?kX$`MZ4aA(<Qxfge|{MQgGfMMT{%de0W2LN~oH%Gkb&9?z~X61CVt(F^%)qBx2
zrCca)TXBGeh$qE)A=|{$Kp`NJ%mERn=SgZ*MzXk<#ja=Ru6XZEo`jOfE1&3-DgkTs
zdt$yFwu;e;Q@D-+EH^`6<&Mu*Mls6CYwqlMyXDcLta$gkrE3EiC-p1J{P9&!M((O6
znWt0nw&Ua^Gn07Fm1^P#DZs4BcoTYD{g(3&$gYkM1!e$r1M-TtiCH({y`X^@seYJC
z+pp+RIq|6h+MK)czJYEMUxPr{Qe0^Fujfp_jH|3@h_~&|YR!z({Dx2xEbN(<dOfhz
zH31VfkvU)L3|)~?G-i?A`H&&M7dfY-<~0dTni1qYf0#NY=r3bJ2r;7s_&r8<#W{on
zSnnPta3av~^-w$z2+rCvZV8(9&r+aU#PbH6ypxir83e?#IE+Aqs*AX1<K_=Y9=KNO
zZ-C)V-DVDPvBZQ-uJn&RS001@UA};|y-5xGFt!fa)s!7CJFu$ln739!f1rQAV+(I&
z(ai>py6*6o7|$z)Gi|jVyu<vY9RCL-evhml`J%Uv-S>^eWiyj-r&7u>Ns`g1ILCz2
z)CFnh-um_*5NO8$KOpD)59pJsV?JB|_U)&ZKOjrsGiV2NbCtkMW1cVmTHpr9g*8L>
zz{jN8VP;?bt6{r2I<Icm?*@QzZ#ZF1UQvnfZBDHtJ?-bzy7F51@wYB1*A5z!yW=lS
z*bfcw78)498w<rF`Q>?*oZ{zLunxNP&R<XJW|@)&=J>yRfv97fnBx$@R_M*qXbf16
z`5l1czB+u~H6EFB)_heJ+qk3%J9!=PHQhxTyogTC&?q7`n;AQR3fUqa8)X%dVe00+
zE_Ji~x)G=yh+o}htQ|s$uc=^JT1W+v1>qke;%mSuOMHR??nA^e&O%f!HO7uOFgN?n
zqpyXYaK>`I^#wO3Sd1m!JH%IJP9}c9h?|pkd$}QhmY=kiz~ivOjvr5sX<qU&b2h*2
z<p4S(C_f#PpV|$zK<8EkB_?(;VQ_LXmeVp8Qvx^o=s18Hg~f2nDu_kJR7@Be!TG6Z
ziYkco@u^7oT}s?4lCQxnW_Hu`-{0XqL42|_?lr-d^QFdqGuM?Em{?%2avAsPh5mb0
zqXK$_Q)sOnC_Z5kM|4sP*{+s4e$l~ENJ<M23K~g>SA{Gp?=0f2H}cUM)=5cv9jclT
zq*<x5-(C)=U7N5w(wf~unoD1Q9r#s!HQ$7A(zte_OGTB;booluu%uJc@$K|`yQCKG
zl*m^pt`F0ECPT+8QW#{fo<CHFgSA$tg)#9DKYfBC1H&f^qNB=1EbIe;v@{SDN*6V#
z`HH`;GQuZDgi@C^k5E;})bTyeF_SZg(=qJecds)QnXlHO;6P^^9`h5Jh?6_L!8>XF
zOkB}5@_wU8o}9N2f5EYR<2AuHoR71p_aA+xEsY3urf0Al{x<eci@=Y9E&QwU6D<Rr
z!d#@HTnu67`#bd_(XS|?%#llBrH4(nMD7XGE$yAzo2zbUTP+vvYG{~|j`YK$7~CBn
zw1i_vEd@#!bE+g|_b=!7eW7xs)2URh2F;o$bCwCYw!{8{WB7MOR&I%&_@V4>q6ZxZ
zxz>-=xK@@^3iKT)Dzg^t5~b}51c&Afj+jCfMkynVhqAo%QLrQQYc=2KpPfG`vPc#l
zd#;LK5zokdX5H}Tnev1qy3(UCL`Z{c3-nA15YlwTA;M)BYcPmNLuKD!fY+Uv|NI_w
z#0>K=2;-u0l$w+Kxbt*KJ6DyJ=#0RCuW!7sH&=xnYi)`_eRvzZJz5+qNVTogD1JGq
z<7Rhz5b5ObfX>RGiSq~KD4k6cFr3-eD_<_~#8@jfY2ayVsUcO5D+D(k*BJf@d3i~+
zR`&BU>cd<c`xZ&Dk2Z3LQx|s4=M$h17sF|R7PkxYzIdeCFnqk9uFVoZolTWHvwfa@
zceP!@tjN6?M<Fk(WvGT5zvPXKhlZ;x|KlG}lpsmT^tb-^`HZluXYVpTAXSSFtz(g9
zy#9mk09Ym}ivVQ>aB3ap^6p28Xl7MzP$=Dgt!=<WuQC?-MH#moJ4gnHu1{fitnj0b
zA23=+NT2QL_tfGq-%*pL9*h~AMA~Fflar_3yk(X>e=1-3)+d59U9%i_oQ8XEA)912
zZi?vx&YBCpV^>dwq7xiwGhK1zsrGTxJuwe+)rHb57ptS^!KIt}Cz;OovWkx2c)Pt$
z<%s<S{<K_|a}!!LyUS#LtQ1#=)tOU86?NzI)5oLo)FJ;EoIY(IID8e0^;%Ph-NW@4
z>S$qkB%E$rEurgs!7;V|VN`vkMbnb(#zE5|1jPuC<$~5TQ-3BoA-3^V?!;!wiPLCN
zJD7mve&g<lWh=3v%kz*}akZX`7{2So7FKt{asxMR#ZOKn_vlvq`7KK2q?TWNgq&9t
zJ~S=3pY(z$`i=}ai}Ip2X)*Hp16G{(jJ|8s_+;i#Q@TjXO{wjmBQq@)_OdY+a>}y7
z6GB0rjD_%1fPl-5lVw9d_XPNb$~IS0mIqEj6{OSgJ$jt_w98kIs`NfrkxJG{v|^$c
z>|`7#WKP0T_f1RIJ(FCXzVbL+PLAJ4k9e{!GKKBMNtBDMS*Y@t0z^{XbxC4k>!;)^
z3klauJQE!!GHFanQau~;s^@u??V^Ec=m{4(Q=q0Q$R=_U-Uez54+u7@;^5x=(l~B8
zNI17LaE$CN9X=?L*cNe&jRT!I7=PIU9jzE%Ta^b<mAOV+g`ZJ8To=^e5$S2J$yb67
zKrB7&w^U2gH#+6?iwjl5sDk9zDYDo-Y<Am-eN3i*b|;ij))GB{H}_d~FEDScp2qh-
zGFI^W6!81(wdRd@hb}sD!5Mo_ZeLX8pN7X{Z5r0T8|@6?QzGL&bMwVaA^-^@8!Z>M
zsWufpq)T(IC-*_(@IG)#FG+|SIw;IWNtGge&DUE@O0)7hir2p@v)<(rwM8KO{V7TE
zNRQUdaa(eoKx<K<!zoA#TCyelbuhMUD^W7_qIxmH2Wo{tRd0xGGUeT7ZbCf&S8f71
zizI>FDnNkAF~SXpL^CsGnY&cQsw=$(A^c1)<m}pw#bkj(@If=0aT1{%#U8on3opj)
zf_rwywCIhn8x=ZOq)<3Fw2?q2F{kLW4HS%})&*)SXVA4el5l*ekRRg*_fpZL%9y`M
zy(O)51U8lw9oh``)7I3>d=k8DTvC5^9sGYkOpNeo%<Z$uau7y98G>ejgWNG5588!0
zsO8-6@T!b{yLT2Jj>N$c4cJ}PqBYs8?&~;M6j0)%Z^TW9r<5V5ZeLXnvm1WB<W!H<
z6b*c9slzL}4rvJ3yuu8i(FV3n@>t)ApzG#ZME^{Tx+|Yd^BHxbZg!Ljtcl-IIA!TX
z@7Rv6uj7ftEyNJHY}F+aneiyV%u7@xal`yjqqRr4n>)Smnl(Tu7E+1=#2g!cjKoZP
z!Gw}I!{*VFA7NR&oiO%_f=Yam61GwFh%dg1Z~QUnmQlP#oEamM;a+}6wkHnl7WMBn
zWF{B3@l@K<ii5wb^Cl+oE$M7UiG+BdzmBnSddwp}%{|JF8eweA2EK8%%$ubG*nYt=
zVB~mcy8ObbA9$WpaI_HbW^8ezir1H>CSd09nbA`VW+ah)^{SqD*^1%bu5wuj+iS|r
zm?%m#PR=zt^lPIr{o=sv0v}W9v`qtVIv43vf1LNqfho=rRk+*0k+uiJa$cn5a7-05
z=}RnW-B&isd<}RRi#?oXsh02^7?ird!$-)AJC1p35XQ<VynT~oXwN}~Dh{s?p!<|^
zZdmXI21f3u8lVGROMVv4DmOWA0-W{?$Z|#Y>YQ7uQo9j0@5%Bu6|z1iv(ZRJaAKhq
zTywSX6l3A!!CYeg3S;BRmel)IN{8oT4PLZcc4q?hY9z<wsezla1iy|?>_@rW>tuFj
zO=Zsa_j^Qr(m{nRil<|*Gu#j|FO;yx?eV76Wwd5@ysNCO91Ag7Ug)%B8Vx-53^N@m
z`N-{u9|@B>mC@GNRM8O9gGnD;^<yR%l60Hd)Rp3usPmt4edn&Swx5@G;%X~O*x7#H
zRCrgXI=D`_4A?=o8Y!`qh8Id~9Y`$AV~9$~`jf^DU(Dd}op8SW1LDRFP})EK#l=1o
zM2zWI0fr{00~Yfi5GAnYS_h<K<B9a99QBkAAP%W=i||4D$w5eLAT~CtJO~6;Q(5AL
zvIII4Bv4Z!kxjp$S@F(kf;^E^ap_fPe29C?9uWdh-we%LwJp#MuSh@n9F(W!3yo{n
zA&?qJHOh^j9N9fRO>MA9bM+UmpAebUMsJ;xdTnXtchV(p=cFCMy#!NxC4N7H+45nP
zg_y2n?4ZWNr$fj$UQ2u1#r2j*g!G9duIB<dW^c9W_p4SwXzWtm`A((FZ>!D)-ilAP
z=~dLDVk8ZSd#)6Q?c^EwIIk~Pu~O)hcnWBTJ+0=TeoXcYQZJ{{-?BphDXW5*515T0
zZB<URbmll#z{V^phjUegg7~H_0&&w-dEnAjyGv1~RHTTNEYLYskY@~2Nn^r26)S5H
zU9ZjcohwZR#rarLmIsG5cc~qvcG`vmguYbMJM7AdM?ppx6E#-r{dUE(n22tEoM_zf
zC|u_t%5|Pc^#V&8Hj3Z#S9)KJ?&EKSHqY}bw@OEa76kKR%DU-;7M5EuWM66#h(_7N
zTi2qb=E?K=FNsue3l2iWbRORG<{{E2AWp+ejG%}5IF7BlOKC!0j<a0p9?@OKRljx-
zz+a5Bd{pt0{;eIkyG{C5N58iC_qH&Y+Vy7%^nE<_X_4*M*E9S%A7Xqul4%i%SagPD
zzK*$&+(iT}*PkJ%e#D=awt7#&aM~x=NgH5T_o7Yw$uFnK!=f&GH*V3psFc-M6Dsk#
zu#bZjNl=dgAH>sX^Sd2dRcb~XM_ZlSNvWHMPo0!Qn=qY)F_8S*2~_^I+HQrZgS9rY
zi5m)4Sms;qL9Vy8L@g#L<P6raw#9il(y9t56*faW=5#lHg1lH*16({3RZ5&oE;uXq
zw;fUb;&k(CSEJ>Cbwg<(Z#_$z=((A$l>+b$LFLP0by<Vd{zJRSOgya-*$A&;UuYes
ztmYfnziRt`%w2d59}PrY-S=gzku$b%5J|w`{rH?Ec%pF8r1ts>446h-&O9!^EqeFn
zzj~5zIumW>SK@uPl2TRbr&8I?7jZRx*(^%S5tCjJt3zl|4_DyuFdKhOp`^Wtx)7&x
zBjG=YO8Vv1`ED@<E`_8v<+GjU&kRNs72j#cJ}tWBDkBKN(!KxMyU4+w5`z6rF`D?c
z$@$qhxilBffn@va^AC~6t1qAKUjw^0Q@)w=we!<+g#Q}qz(j<C43$8J5K0Y}P{}jp
zJ~)>xD2Rz5hzJERR(Qif36okx<7<b66?^bYtV^W6dSs=lyKBN;qAX!V!UIXK!lH`Y
z`J~?a^8`LGU#iQZVUJMUl{=rkFSW#xAM|ge#?168|H8UFr4ccR^=SWeH07h$9Adip
zL(i4Oo(ZMIr^aLD?Qg6dlkmOl(TZKccO#^4Ey|^?l%1KQ`tB@)3@b>Z6l=>cwnZW@
zh9b>nl`$GeHN+N&+-E$s<GFOK-FTa4O>LhBW2u|eFihH@hgvXW<(W^0`Ko+T2Q#h_
za-KgE-iuK2l?vD2l|^i}hf1dDxTnu77h00CEY>|B#<NO{=E`y2`OHS-4brmhy;&Pr
zE4@wO)wd^(D})BUM+iDl088|SlEx^FNEeS$XZ0oCDU8r|v5*JqlAN%3SdOMpNl+y5
z+f>Tcz8CZZPj|0z0<af@M2GeX?Y2Lqcj#wpgz*brS+}l*x%t#a%=tI^B4zBej3*D+
z@l`H3wmm{?O<eBPh@7bKXU@P1GsewP^qdE|2t#%)POk(W-$AUwgIC5Ung?g5L6#0w
zP+?EOMQ*Aa-f!J^*5arEgUI4{E5GOWYfLH7DGix}8H)<F&@~nEY>O50Ke;L(asdk;
z=C4j+JQ%=|5RtJ&1AclMj|`Z&4Ite${t6g=cCY#-yFgaWv`j2&oiHp($GA-I%bo|_
zB`Q&2WW1_vDADrTuTV$^1Qcd!|EjkwJCE*b-E`LXDS`f7fyK|QqIItNdR}0;N$7q0
zWI-;(Cn`Q6z~zk3-yw6WU1FN9(wAl}ai@0ro3)l(V%mP2&UB1&43`q6F(qXUo6gM$
z$(QS1-w#Aqn&Vi8jZ8exFIOqkuilB5DY8ct8D*Vbrb)!rPl#3+?t)c$$(a=6-`r%E
z&`>vJK)6w9e>EHqg={0d#`L$+egHQv0`75(QZ78f(jxlRFheFr{XOayJUNn1DbYg>
z`^7wVJ4o<cA%0^ml`IFk(p!4lF%CdjgOmXKR!f&!n#U&fGf#WUe5ur#nIR-e-Updy
zHTRU6@O!wYxzfH2u3c=+Ma+2lP*qoFKa<d;6pP~4Yk9dbz6DdGh(IWYjD5eBro?Qs
z(CZa9gd_Fp?A*V$GX@8Vic@0O)<@%lld0_cO15ua)l{_nkRnix-Sn1G5D&t9sOCX7
z;INFd6MxwM0~&88|0dMg`%cG8_Cwf+!4>q!sX>N|&;rpGi;CC{ncZefc0a3Ewte|S
zw%zr(e)*eK$wv7`2OC*vGGpC)Vil|1HERlvF}?PtXhvn`KzD6Fx%p6UEd_)Q#~$1W
z>&VC;H7V(*(mBOtW=3r8<N*RC!jFxx{qlvXIzn{LAT??m_;{<epUph?6G<&MyjuA3
zqwm#9JENpoDOIK|a|=U9Arzx6;)aT9sD^bI7$g}v-EF)CiOsyxJ$(Y`qj|1+4H*j6
zr;%=-^qjnTrrYV&rdp@^HkBEo3^L3ET3&b-GlWzb4~m_o+(qy@0|w2>FJ^Cxv7*bl
z=9gAqoj_$N-M10U<?zEjl&5*x(zlmH;-S~KiB3|kXFBX6Y{jd!CQ*RH$*xS6^0tmx
zgsvq2MAsppLnQ$1;PC@^;Bof1hOYsk<8_8G{o3odYus&Lc#H%y+gWe}mJ7FnbYpuq
zA1O|opKpvidM$*%;P|j$vk(<V9jFc+N`W()>g9TItGNwkY7ATAZ<)xiKQ(gEkum9=
zp-aE1(KPgurmlCGBG!D#Q8gw2yMP!MRE4~k6H98gSPF8V1y6Ctd3fYkj@r+$g!B-r
z6w=k3U+7I<NFfVBtwQiYni7CDpecbS3zDWRp8#AV#drVsMwI{YjVSBb45uu(uE@9k
z;2k<l|MYW5Q)y$Kw`2tQOQbVPYP29tk)C<B@In+#D~4;x7=Q!Go0iY2f03SUemMgq
z5|L>kFw$gr6mt_QPbNt8r`3}Z*4~vRAup0u99Ot0_l<<>KYZ^nBgGK-j>7QJ!$kFX
z#{T6%wtX|^a@>T@+_$-tYDcR;r{}GtaUn2!6xOZdK@MxzQ%j%EkePPKknw?@u2k!#
z)kHC~q<=709DpR)j+KS#d@=osm-sOHbePSIWg+gbe<ETC>#4nYm&t#Q;pyCYRP*hg
zrRwN&-uCLOo3Ux!rCh<G?bpo3(5*+?=jq*o4~@8DTbQ<!#?vIO?@O5pbu?G-(LDLg
zHYh@fQMmEh3j=n_BOkRytDXvWGqG|@^4_H7oiCfpEODQ1w-m+C->PBsqkJupEMmcY
z<lP*2kOVR{H-7^%R2Sl0<J%I(ku0qNeWk#Pn@JB{9=_xOzrxCqvVHX}0@04*YT>#(
zg~@*#i^XyU3YK-}#x-*6O<3Hj^JJWUeG!y=V4Rv-q*`yt<)rv}O?_Kp+tE9$H2Aad
z1leK07B6eAbTH4g3ulln#$a;)?q`NHPx48w*Pagx-qi1evtlNR?_6$wi5#Z5ky@Gr
zV<}lSfX`3;@%5CrAk^R5vMLSMy2xky=Bh&4N6Woxa;xx3I6d>z;4Vl>_<yTLZdWOI
z7V%+t)Mg??YI_(`Svcu5Uf`$jPp$l>L~*dGJ`@UWs#USnOzh`AQ*zmfmzAnWB!xCF
zH}(<^9~QXsN~vT2M2d_mZ~^?r+Nz}M`e`;tdhOB5o^qhJr&X=u67K~Yaqhq_kAkKC
z9^JyF$-Lp1BFbOsyP}x=P9-9>M%MbD^f6f|46`*VMfzWpf1fQE2G$MsG+*XMaGVSx
z2x9~pR_B!?<4je-tBs;U5XP8r>sj47K}+KqUVgSvwn<-|#wD_7CMWzEutNWtu~~Fc
zuf{;OVnxXX+Oc~ZQoZq^?dnOyE`0|f-Q^}p`06^h?|gz!VsfZiu=p<oipP_uBM-fL
ztPr9ICFWEl0vf3)mD~bK&!R~gb&uT554?uHRZ!)@SBqe-97|d9J?=6)J=u(*tRKhR
z%EfleL1E*#&WrvYyyN9cJ3P@Ds&o0@CAgc^Hs)h_SBRKcQX6dqv_s%YnYx2GA-16W
zK=k&KfD!fXgvBpUG#NvmHjBBDS%!0u(yK@{722~srPL{B^yv6I!g)ys5knPem(T<G
zDI|?2LeElqiN{N5Q=<W_X@*?TIm=k@7~hr`kj~MN%(I6@`SKd{@?kEl&6$}tS~{xb
zEyf3SFSpu#U*Xc?Cbqs`L&SJ|ui5$L`r`P?o6t{o+bFjhTSv6Glaqn)Z{tsM@ifB0
zx;djxqAYky7x1PgKbKd3vt%auwM-P-Wyy|Mh$bdhLk1m!?baR9FXT{CVkC_wc(|6)
z2;aI>Aj~I;sXE+I*NC@!ACK$OT{;c_0lmnKsqvXa?hrxf;5w>MFrMCRAB9g2#Lm>U
zk63hsJqAg~M_1yu>Uvi2z5Q-GRKeu?@-y=W)HTt%o<tS5GALAX3|T^W20PpBVMetZ
zYb$@BRc4_!W{V<AW;v1OzK?vG5p0E3WmI!9i)#;nTa3M9mBIc^LM29}edh)VGiY%S
zj%>7M)w4+ApY?8?rHj*%le2q8+&-*9g&=B=W?{WLvy_Y+RjF2(w+#A~FkjE!@T^2H
zeCJP)(&$_|i!0|5H(Y|t4Fs4nrR&l6c<FY~xqj@ZD8PEsk7J8Iq_|POE<^Ht<BkL2
zln9(Aw)jW=Zhtg(btm_YX32gD8X)MWYne}y2YgvL&x|eE!H!<J>FNz0cllf2OUk@4
zn(s|%{(w}?e$PLmUsMF(JY>Z!hMKk6i~0OGwT&DJny=+mb`@deOn{c2q6kqx3dQ$e
zdQ&I`$Fo=<M<#>bSJW1BFmW*SGj|-GhLPWp$ql>6A+~4aqjy;C>MQ%yR>U!S&LpW|
z78eHb^^4k}(;euX_P>qmD9hu<rCp;m(>8fYDG$0O&_az}L7*Mo%_!qfa{770=+M)w
zdg&y}4gM)WO6qr5;v4^kM-!IVGu<Y^BIso7MUEs)fP~DV*R%aB0KlxO2`7k$WZa#b
zOuT<bS2aGXff&(F;;QR>Ll>qs%U0Z)Mktj*gzVZi=CBs6PsfJ407Ju0H{m+@I8tWa
z)DnK;x)678Q^oSZ$+`jDlrBf|vEiM7*l_@-8r!j;tWo+hkxy{SeEa9ojcv-%Bi)lX
zjUiKDd;5R}!nnKom$As4&yeC%odOj6p*~o-h`6I@$JcZN{@)>igXLT%rm7o#an^B8
z+iWZxB6Xi5-(BP%kIDMe(6wb#pwfIhyKRc2WmRXc$`)6ee3xj;DOX?H{zayaKqSes
zKLA0x1FGmxUsY_T*bvw1WtP^SJ~VssQ<%=%fy%>nsW7p0P&i7Z315uuA#+H|S96E+
z#CSpYSeEWvq}Yjef?A7(OXBP_-e=}4;v?7Ja!hhz9)8Tv<uprmx|7vdh)|z>w@}1p
zCDS>qC9dIvXw;#K`P=V$_^nXQT2E|v7OCkCnT>RIzUT?%M0MOaBQHx!uJrvcW3(ow
ziH$ZE9E0TxEJ+hZ3}5GWbcC%?el}=KhyPLpP#GHj*^6QOyJS?9z!CQ;e?C;E=2-@5
z%7W~cfeq<8Kc+osgVpGXA?5AZ1a-@uiqcjqCyqn8BLO|4z*c$lPu`s|GEEwBk4J4R
zMo|`SDL@#z9x{*KFWpZ~3MB<mL~+UD%2V^py{-iB;ImrMkMo_)+*E`KPBhxH^)p-$
z`-_>1GEw2KkW%!wy}-Yc&^<F*^!z}R-ZUU-Usw%iuBRGlm&rKDdrZpA&!A63(arSJ
z@Bv+qP4-2t9}<sT-Ut#X9c5VD_PPKc^GDEu97+*F-Am0NiAtXQxWS?qeQ;Lnj9Y=|
zOdw4abEpI;SL#C53QP)Rr%&M6X^x^!`T{!R?K|QMYp3)uMRK#Gdu)Fd8uQ&cC_#lG
zkb80=(#O^V;cnG)OuZrOt+ycWOAE%w#ghEJtMcey9Z^-d)a-c!PuCRV9k?Ce-gLR>
z+y|4v8s)+A@iPu@6SC))$i{tgUT~dOmc+v&T4f77G2!+*A2)^5(+X_!P8L_XC}Gr<
zO`WELS~T2O=gy9Q>&e^u4U@(l9jWQ!;1muw38?^S_&DE6HL1Z>87-zCizwqmU&rJi
zkb*sFi)hE8E_N6MXCAl7JQAd_zUUD!oRgns45*xAyjp`Hl6A9f{2A0@VjJ7LWq68c
zf+?+nN7^SmJUYBQ5^sgI(MP#MNw2r+gbb|(D{We19O^M!H^BaRCCG?le2{itOJ46o
zxAbAiITEhhaJ#O2k=u0Jw*k8oGo>p|W@bb35KNm!3PO<^@;QhRrkd~qVst@nxMP%j
zqcFj8lYr@_0wmm0FX#Rs!c?N*D{qnLk1Di}bhdoXH%|5?>~u*<0m0h`x@=>_lVj&$
zPl14Z^XtrVie^~}>yAXqFAHqY`9ugnSGg(GMyu++kJs*(s4QU*pMRkoPE8YIm{;2W
zHIMa!SmWo#`iu`m#Sr{=DA44AyZ2nA=4M52Q`4-Kr5Tgydw|2{r8B&A?ZNOa+ojd%
zT=Tt(X4<mq_&3!FRM(plAw{|7%21Gs?wT;L(~PceDU^e{1B4A6uQtn&-{dr=HI>qJ
zAn|mw?uvNct|&8ZUoL^c{f8zfsyPZ%v5CF$PXp17dsD?lTJ9+~&-&K~<08!}%Z=aF
zKF-mr9jYtMgqdfuWYVt<97N9RNAkR@TaeQIAlvNs0&4CpGMHFB|AS~u|I-$oib7G0
z1M$^Nuyy(5QhCB>#Y2q?L``MnJ}orkSo`6*Utbr7Gh?`S@P*wKDUDa-+*S|S*Y9w~
z3wAw-s-ei^A$WU;^zPcyZmuLWga(N%4$u7<IaPK8)NRNpRX$7ZQw6M2OVO*MwpTju
zYczK7PEf_ewGD6Sh*p^d>00zFm!XQCkl)yDbK}NeocPD61SUWWNrv?q&C=d-{aQO|
zp)=WYv~H#{N^+wDLuJ94CR3`$_2ye>3%_hw;f@Ta*2mmjiH53M!6{6n!IwPHz+}NQ
zo#J%KC%M)v1F<4nDAy1b#m6}d;~OB-)1A*y3X!3#%%YZ?^bG)yZ#dB7w))ll&UZn*
zWS{T03nhlWHcQzSs&QzSA<SZRMJBOb#A#id;Dm?LBx^oCCTOJhhMMAyUGIle=ElBH
zi9-^8-c^F>awckK`coUN+B<>On6j=5glCAwg2<@63n<=csyG&v2j5KJ%(nqY6#Bj-
ze#dJLdp3M#mYnV=X&vv&kNuPm30#JhOSqZ732vYZd-?6Nv5s%}(U^HSGy;6AwCbv@
zdU=%SrU&=BXz4`NiG{pk11AQ(56Fpa_Ip8Ujb*6nqEow?gkZfxBd6%?9gmzf5I@}N
z3FUvkU%)d-+Hgr!2thma(e)x|MT#JCr%9m+(hiFhX5Snz4CN6RqPPo@6|RX_5Zr?*
zIFz`3%n%8{n_adHt|y5B^Ez7)ZZCaXiMa(=J{MziwtyWb<x~0B4s@%XX!wo;f5oYC
z>W??aK)A@~@|^|}zIyiY(XzgRXO)ju1<$P&>b{uiK>I9<Pi}0wo8?q_>^i1xl@CjD
zrhOlI%z@!t&4Q+Ayv(%d-6vF#MlcNeWUnU)GNZVIl=%*(4XoOPT!Z5dSv+~Ss{3AA
zP<Y2pg7x(qS-$M0A5gh1X;K|B@6h=g4`*7Bj9Ki=Hr?SGt#to}A}#Hb!-|`zk_BJL
zM1`yP8{)0}Z2W8#$Jh{0+deaG<Q<!JK>W73N?U>a1>;6@TB4KRa7n2^<>5q%4n&P~
zh&}?VXn7T)A#o=~7~$G4v|IKA>6@lu?;ayUVXhCJep{Kr`om<kBF|h~=j>?bSLHm9
zd~;$&R~(Hscy<WSnx7M|Z^j49P!=0-utM1o$o#`oMLq$3(x|m!iDmILY(Ojp-{L^J
zyM+Zh^gX)!aP5ViGJYHjQv{j6g>6#idwSAg=e0xkdxRzk04F}7)^l|tCtsu*%4B?7
zh`JV{5r&%cSR6yw*%rWTtD5&+CM-W^IGfEzl*V}&K-$HVw{{8GNvq(5EWGUbMO}tg
zUo&Z8yRGZ`iKP(j3YlPrEZRU+aR5HKa+F!{l^{M`4)uu}^Dm4<0>)x}d|h0?5%zxz
zb6Ey$-R#0bDM-W1ziQbC>)Pr;dkO7S){rZndBTES?1F*(CeE{v*~yv!tkU<Bf!{k<
zp=Y|Z<;)br`Ac?)c<`4{Na|`Z_~jB-({CH?Gd*h}Ihc+IeEJQ~TbO4Ql-p53la<mH
zKO~^a*f@!xvKjHs1-c7a8KvdsV}F}ErsC^%a{CaFYx+sg5K~3wjqcQPnK4G;+v~hE
z9qVXNjftJxF=Y}eBF<;|)ROxU-7kBRyKhiLAtG~#FK-0!E5)qDd2E7K?S)E*c+l{#
zse??=P2{yuwly1jl?u;DJ$EAz;t||YHTq>i(Sn6WEky6pfHM!Qq=Qk7B<4BmZ}v2Y
z%48c(Jn6H-J+T)7w6ZkRiLWIAb}|wIPMvG{z2%=f8}b<LFJLDOx0%LIHNc7DDT|Ig
zqB(h6EAbF;jDFnW779zc7u~34xf(5Pcc1Ue$>rXl)aT4g4#B+K$j4OvY>&%!CuWi+
zd-C00kV)z}OK24eo57S?RaWHZepk(}M>jQOr#=et3mG9OwH3ITV^n1P6nk3zvLrh>
z1R@T({=6EiDYcz>gHfL3aI_p7dah~ux9}aSLU&j~1@9-dsv4eFXY}(KuSl+VC#{|a
zvv;T9)+k?ZSIId$<VB2545}Ny<F(_mVK3t0IN~#RBM2WCP@<h@n_ih%UU84af!4gE
z_ODmZ!zk_`o12!1jcd6XQ}r#&a6$~|Md4*9OoHk<qtNQ*XtxQgezWy^PaM){BH;$c
z7QQkpSNQd?n%zB6wl+qOkpHRm0PI-O`le0s2jY3IRw8vo)Q6$;*6!Ur++^pJSl@Zn
zQ-`8bm0@#@UMK&#3Hv(5tJstE&JJ0l(tT#14mEiL?b~iF0nPb((2pAWn%+|%4;a3x
zQ)_}~SJ^gFPLAPT^3L@;nQx2_S_^KATNauUp}ekQbanjc;mqQ5&)X-!rlql^m#r-)
zE8(;f$}loqg^%l1Qt`ygC6wh1pY$OEu9E*<YP)p+`|`gClS=^?JLp7XL_kC0NU<G1
zY1rQ);X8N_h@sKz$0qS973DT2urqQTFYP0e@8#!IQmXq(Aa$mbywj|7GvU?$GuZ2n
zBeb7eFJ6C`v={>;HTJ7;GmSf#jh*~ONSvGaH&J_iu{_I~rc7n@uY1&0B}&)YZJd^U
z0@~P{$`VLy@RE8wr*-~-S9)TTUBoWt)ma=L-!4j7IS?t6$M$R3WvW_x-*;AVnJrCn
zZE@p_=P^pn^Sh(wyFdkgiP!NM4rb#&wHn;AhP~CgBiYbIgj2Mq6O;fn=wUBCi8Y?O
zep)wwhx7xSrs1)6;jLA;lRj(DdO@<8NzbY&!b^}(oF1ErbsBy)_XO{r1>==lOF>(d
zsV)#FV~B04xW}DvWcqIF?AlL`DdAJid<)ZeyD;OqvO;Hh#@1<E$#G<s{3eT!9AWdM
zbCc65d%YhSck?pb7%k(c;_jg`3&g0&+_NwBoG?E|C777tbcWN5d+1<Qnea<P&pgxm
zY~-KwKF}b6X|s`8Oi5mZ)w{C9pG32y7p^t(1gi?$3;deV)p2i!l)alYsfrl^;K)4l
zQXC6jM_#u1^@NZ&66QJJ&VdPfXKutk#XIc~&Db76Z0|SYq>0~!Iw!L!TZJ(n%5R%s
zCnOqDEA{v1d1r>uWG96_Bri49-tN|R1STyR?V5(>(WFl{FxQaZbQB&;Fd9Eq%%-Oa
zQ^&XSi_OkEd~r(C_`o|Q>H4xMZ^_?54~zF_@i^u(uF<lr*L(4?0VUU&ZN-lzlA#oA
zhS4e9?vTrb#nVsVPii`T({^KHZ5%&4!>i)Iv6?eL)?U{iTqvD{AOJf_7dFXn)f<@i
zAbSV8Ql>XiHr3JO%JWjL;0BWU93B0Y8Q#MwaA`Dg{GGaNxCQTn&ZiD{;M@#+hKe+#
zxD?a953OzOds4SdAp>;*Ud@}_<n3FJqRSuDEro`&d={})XbAG9_gFAl+s>m30Hk}}
zDe2c7jOK<vwC{a6KSGr`jf~Y=R&K&{YnI)aK#;nrVLAG-%HR*4wUA8a$aq&HE>+i8
zOO7rm0?+m@^+(4?tpqI-^jiy!%{_s=)}^}IKOlWOS7bV6*Fp7fkbdKyL<g-|+I2zA
zJG~$<8#UHjG#IX`?Jbv43hT*dnbGw=>4Wo927)oBMi!cUPBnhqWwx$_J0EoA2!&qR
zz}>Q;GBjEJiB@tI#*P=nS-K{1I_}Z-^0nz~N>C`KV6i%GhC|$#`RFxY9p|2a-}j`v
z!XL55(B*<VmE4YPT|CH!PFMPN>$E}`Ly<rjLm<TS^{-HsJxEm)G7-|ja8A1($rX)I
z%&N_oss4hBiM;|ueYu&;k82f3*DIwvwxcph0_9o!t@!1B*+Z|kW~v?E8ZZPU8K0}#
z8OjZ&^@|SH$c2zpsre+~`4GFPOf!KIw{zhOIppywkjBsald6nattx@Y`N;57)rRwS
zi6TSpadl1IDy<o|+e90(rt}L2HIAHb4Wh4=+=`EH=)TWXF0cK6MEzx3o6*{K4O6s8
zu|kWx2Pp1Nkw9^GibJsAQrs=L1!-}I;@$$qAxLm7?(R@vJz49zxA*-6`Id9ec^vx~
zJJ}kka{Tj9vzz8&4nI!i=|~N|-RPKw#V*UbQJ!i{$q7)<Rh13%n=tnHiTfDU?evIG
z5t^vB#iz&(N`+%J1V<LQ8uU)Bf&Z>r^5<#P1(xeH0)1#C^5^?xH}X-rh5Bk-yQ<6x
z@6m=W4L^7BvJ8QCxN0E6DEnEJJBr39Y7Me4y{aEB@ivIu3VpL~Be|#Ll<Jx9<vz%E
zeljR0#L-<1f;U43-5821X;?|r(L}SZuQiPXFb(rdZa6EpGRlKX{{Ok1cSs71I({k1
zn1~i~qNX=)^G_7)<i9*OV%sK&tmZT7qeQ&PtjV$8K=+$DtY_2odrl1|jAJ*QEpyt)
zv$O6i6YfES71gy~vY+MFvwjcg^L`~d1s@xJ`n#&$f*d7ZG67)H$ic_-_sHI9b4NU;
zo+`H}1`Nksjj(66!%7%%JY?~1YmCs)<(}p*RtNnl)n}Q)8Mp_j9Uv;eeX}p|C3>iS
zdhLjs%<saYaYTYP{N%w%e%f(P?*Bm&M|x&Ev-+`k-e~NEGI0>VM(S8=YK3frodTpR
z2c)OQv4Y3Xg2>E%wm#nl{X2>)Y}cn06I@T8+*a&xt3c&N7C$t>peJs+wal<#UY!YQ
zC=jn$6($+s%Nlq57zSZ!g@08{<GI4%GZrQlIsA#YK}^#S9)$E<VH^m<;;^9ZVO^IA
z^3VWZ(SFPN4^juqmF4R};r5OU34W*m#J<9b%nA)T{ZBIP75(c$`8r$xW0%NS_3cn+
zJtpmt_2jYfeA^fimA)h|bdms|<;V({RotieidVz+^BMN-@wnjb@8yM{-(03-_cvTY
zvPc1b^f`&G?}TS-I+P<eXR$dhG|S*j(mH8z!0MlG7-%YkWdI$Qzhen_8CGMKNH!hw
z`}}Ib1!7le9F*CX2sPK!td_JdM<sFVn~`0gt=2eY_sKkA;6b!XezEqq*kWo8O<<|<
znDuUF>6V*sr6>JV^w*u{_bTvp_$6bCoX|~GtGa@pmv!d5+*9!?*3qjkC*c7@gw;=0
zHlO2FSeP4X#64KXzO94s?kl^Ccp#LHsvTlj|1tzHNOmiI_-@1FX*Mo1V6r(jm&M7h
z=R*!5^G?+SvsA0v^!CP89BC;$e5qGm5CGfjGR}7a_4Te9d;SWj6K~qQGy8UWUq6-B
z-=Ebe*WBVfetM}ftpZUcGS3|;%iH7NQ(f~tpM|-qm@^L#TPdL;(%Rkn{(io+MVvrh
zW79IXZH=}Bqg~fGg`#;xAG;j5{&B{ackx2lCi1AC_`dr2X<>d7Wbu)Fl_EeOFn=dn
zL4TwAAl!1qvRLnwNmcJ2Ii_}lg=!72`@_ar2}W*f4QE;+lctR$JiU{??flbG&dErR
z@#Z)*R@1e2%WaIFYHSfl(-0(7T4|7Z(o$AVTC(|&zUZW-yL_at$L>>M(WX%DJgEB4
zu2*C<{Cyat86=Y`_$=sU%G~NcN+pp#MZX_n8o}&SB+LPf5Yr9EMf_VcLTIWwywUzU
zi?Fc0NI$pYF%hN0_16t-tQxbnTOcRC=)`K3r@BqI9aCwj=v5l-nXwJ^R};V{YSs(r
z=k%56nqA;g<c^-pO7gv_Q^uKPWvLx1iY^;b;&#j*NTX=3Y<JJ~Bm3#h2npg!DDheL
zvxK*Sl~W9<JcZUe6P#pp({)*p2XVKEZ260v{2IC0kn0b`0V#TqnZ$b%3OzKpbz+w(
z?^&7a>&TZnJgX6)$xW@jPrx(mU4x(b_C;;&B;IDU+6%Yje4Zpkp7@m{*QqO>A?kZ4
zE5>Cxv)MYAxOB~dGjSjxy9FeD8BF%XSbGT9qd-YEVdJwK$c5nXPYQZ@NE;AH2T<st
zGyVrjmmJJr@u3^1vAM%MshvM>)Xp`^FZVVG+IY@)n~?U;>YsWq#rJ4V+YG%}$qX^1
z#d8B)d_Ds5kU?I<ne?RmEtyOkz3jDY<P4<QP?EC^TiI)f9Ba+zD5EvJ#^-0A3wy~X
zx(VJ~o#MsE1WF!(?lk=0oS+~1myUf#MdrCE`nq1a0;P2oQ-_8wd1Ostp#Ud#O`(gT
z*qv~4J-$3;QBv;pZb)r%eP5#`@f?tFtVMMjdQfmKSNb*P-l{k{m-G4l6iAWthqbEX
z%aCsTaY%zj_Zw2;^a`Vr-KJ!rRc`$<drE-iaI-&NgUJ{<CzmRlB;Gq{pI$3)UBfoV
zz;M5G;I$?&?b5Utd0MIuY9EtO%r^xI>MwescmafKTz#)W5X4r>;9ft?t#+n3$QGF$
zQPTQ;xzGDoIiLT#U97vCio0w{YWbY#1vaa0ce^)CCsuU1E_bC+upwB-HeZvH^kT6T
zkUnO-4UG<rX$6)v{wLz_w$epLxz)T8Ro#{C4FarK;~x=SJ%U90MEBS5#YITx`$LPf
zW8p?+)V#S^^C=BH<!9se!VLMhewZl5>Vtuny~rI@gV^o`iE3%`1s{Xix?j37E`$G|
zb?7kz+}mf1+a2{=;CmoXHn!%MqP~&$p6)uJ(Vc7m7A{X_g!D-Etc<v(NS}NpFY_$K
zNO_Y7vYP`FsyPvO9I=D}EA9K12zVE2ED9ZV+}~$?n^)&q8t^7Vk9coC-4&KWT9pDH
z1bf-^Or6%>+mVTPASjjUM}lwTeurL1KGM2{^{tfZI&KB-JUtD;9@2X3PTK|$B9E$g
zo$({V(XlIoT@+d!8*^1v9yGm-HQHakN+{u++{dx=8o~vD19wgFxp6i2ROfifHI5nR
zocTCWwg6Jt&ZOZ@#lRq0st%xv?)2x&`qq#A^8l9IQ*dK221>JbW~V#arGBvveXOaQ
zK3?CC;E&1b_d+n=EGT!brb0<nj-DeteQuAfm{_{WqJjVIJfLkY3UvX!(lDpwEawTt
zD&KokY@>cUd~DLH#DeG|sTVA}k6_M543HE=_;xM7AU4ZBS~$s1vmME-(g}!hU(uwQ
zVWJaZcaAxyZqgbf)e=UdrON0`AiSBggREHCIYKehz?@s-^lbFbrY8Gwz)9Xu@0*vf
zl|)8)2+NPRd2U#95*8V$td1tKxnwEXbkqMA89?L)($rDibo7ojHEH=KZg6;&-XSqB
zxov67kXO)mgSa@}_Usck`q9TC4o{YTA7?<;yoDKm3q|fcvCec8nf&JinSIL2RJ@EX
zXg%0VWaA9+^D%A)RoOXFGU6bhswVG|aDgLF7??i59>CO0Dm}|tM0?I)NsT}|YpVyQ
zYCxR95vF>FTFf*7tTYyNH~`h&zR~vEn!__oama#Ns{#Fco-2n3IZ8Btupfo)thRa)
z)y+|1QW_Nt(*N#7N+Oy_6ZSl4+~{&RJvjNH!EKWglINN+dd(j`@=zGX!+AHA&2zno
z&<05_+J`nZuvdr2T7$v4GxeNalz06&D8xwjF!a~I>|%3|3N=h2D!e(fx4`E|93T3#
zct5>Q<vBS~*9=<0fH1<nxeoGa4TL88`w&0oq?yB2q8wu9F@g7Ot`)TtygI9B{=+=%
zP_u_9kgie`u`>*l=c*b1avJ=0XMVk4>r`S;CMPWU-1L=<YI9GwVsw;ayOXkPU#j1g
zSNrUnztf#(hG_W4^M+ULFKxHc;{3kEXeiKAjW2KQ8s2g<iGL`VOqdy~DA~U{weVGL
z>EFP0g+2@70a;oQU^61(CLY1B<XC3-HlU>hZM-tM6dUD9bIk$hoRbZ{cB*i7(6!Wg
zAC@A=w2&GqaJwI?QCDz=W`6<+s#YuDXsNr45vVbE|3x3Sf3~gIo-D<(C_{nJYTL!H
z3hYYDGQLl;S3<X9%U$zO2{4d0;88b60!943jF`Rt)`c?*%R~h@D6>mg75hfDfW*v(
zQ$Co#y{l!N`EnG#%*XDsk(|Q}BAA*lDU51Md^uV8u1V1-GF$iFLvuk>m?SM+)AF;M
zD_*nf0zTi<AHR<S1R{3MumFuvU;KZr66pmc!A?W<m6&y4(tHG0n}izlUjnSx^cg`?
zV9Rky>(n`G&aV}Fp&%Ekj_z_8^d3DGFQo94>foWQS1cJ-F+;Z>>S4YOnF6KN%|bF8
z<r}zjC|tcZ2LSHXQK*_kVBkT;X_8ph5M_kvaG0`NUr}PtGUAH(OXh9$pl-%Ux_1RX
ze{uXr^t>}s{R0oype_5i6DT<RenbM7JUBSoqfHgvxY5!;2eGBFsTq|bNbZ7;m_-ml
zZn;Dlj%v3Yqt~F4(!o9$<XS@fIl^S-Y2-oEpFlF)@$Sbuk!=s4E!U&kjBqh>CW#>g
zu7~0!qRZcXVANiMX$sZ_80ga(JWtjujB@cptumzEU#DJ!Z)!&j<{NASnTpbKTB2v_
zn+RmCEOP^O%(NN$8a$olK_WG?6$N{;{%PSUMvHy(e|bd-hbUjzZk%OR=xUd|;&D7C
z4G}ql8%Cp?S*QVNwWcnHFg^dDhg?iI6<?BahqBpQ7D@1bi*K>pokXlCGdh*=)*dWy
zZV9N5jnyA^h=0A!MR;LPe*B8vEq&L{iDXQ+`n^xUf9*aa38(36J95W?mQil=@#v?t
z2(eF=45pS60~6@`gF#Ak7tP^0F>d$*YbzbllU)Vs=Hv=c!O9=n7+DMMNq*X?%@i58
z2XWt8duijt1bgu)*yfn~$BcqOza$4y7hJ<7C2RHFL`@Z?P}aHR0wQi2b~m+Bd#lD+
zyNL_mW0)Qt^YU<C*TqZPxHsU{Q>O|0jk6Z+o70ZyUK-D0hCr&8wuJ2jW2EqaqXO`z
zUMMk0ho*3|!EbJY%mj=7FMoE7?`G#!djbY*fmnW}uOlgYj41%!;c%eXxU&C*Plf7g
zeyS@P#WF;<sT+OkT9bDbXRwT}4bZ2mI%F*}1*C*EnQEe;kW_&t4a>Fc^#@DrTHVR@
zX2iy_ob?`zZCWjpqNVN%iFQBq)!0oC05WVA*LnxokW*5Z%Tn~1mwRH*czU=fDyl2A
ztp2IV4%iB{b>?W67gs|*4(f7OJ?<(B{~<8`m5o)}{9e8{q|OZIeALZ?;NTK(tqz_K
zSmi;0KnU{i9p8+~+ulnWNn1&!jVY_|4xTMB;tJcbB6<`F_%pzGD88q-WY6y_00P)V
ztUl(i7ZUy#=%%kXl-04>!P@7baEbjiuN-1ay_T3EQQE!*vO7^GBO2Z$vPv~kiO3&@
z6Fkh0gz}?>o2`wM1*C94^T*iWMN{xy>i<~~cGUK3{?R%m;!Ru$@9$PsX|FNESt8Gq
z_1)Dg0Wn+IQclW3&~#8xV)s42n1V2oUuKW>`{(j-xZW1%+z^i*rN;0JQ@kt^!66$H
z;xW+O+*_t@ls_Lr--(Ih8bRSfm|UyYzc{<(=!0-%kt>hy0e3ONTm0ol8Fxa*EUz==
zgJlojnYtbNhu#(2x4E*y7PAVc+M<I<itRV)qlmxr0Me6RO1Yo4v4@8)wL4;nhez}@
zqCFitzXrUYkP@#LtnB!#nzL)kuVV6HN8Sgkh+_6#T{%R&98*eq<cO`i^|ytk%*}LC
z%1Wc)(pIwa_Z*Jo`Hr_T6GUD9ibyEX`}YSB;nG$PRxB)*4TvHoi0AcoGt-T82POU7
zP#Z2FypfYp3@n%R$f47GBPrlFTSdpPITxt0!r#kGG;svZ828|HbXz0N$(?ET`8Fm8
zo}~gL65#5Md(|Yp&?#`YO@7hA+{>H@%6-{3^)#%J#?AfMluVt&ErbKCAl@5dn+wfd
z>Mm^k=+S;K$M*Pr&4F5|j$lLh^}Q<hFNwqAv@j^W4z-s6^e4y4*!rjx)!d3w`g|#G
zd2+U|*Rayt??(Wx_4Ut7SOx*yFuQ|Nneqf>KqD7<a>$czsIA{~SzU>{h%+z$z6!b5
zzWK;@>|#i(e9aMH4l;xjb5Kdc@&o;6WhP*{g1csf<FA_~ygDpd1!dJx36<mjWfe95
ze|HXtuRjpuWwQCocTGV`Ip$UE>FLAk_;afJsxgCJ@pDYpEO)})rUcGyl*ed~=v~#;
z?Gu&Mm?&O%r6?_GHNGduQfq7{938XNvPv|gsbJdDunB+d`t-doNBF`*<fMURvB_>J
z+r(<xkx}J;2sD`~G~X;fMi{rE;)Cq$tT`b2${oJ!X>7<^i8ZEwnW<Txk5^02iCke#
zCs7ze_oYBult}4Y{*oBCDTh^-iO19ne#(JW5b0MTGcfGKM*(QKcl{};m*sIw@jolu
zA3^{2;*hff>fi#B5wso7-0#WQCBW}y^9o!hvhMYS9?hv^3j5o6sGH2t9xbHLVLaqH
zk7+IaBn%#<oHkzyA96D^9<=(s6Xwk^rqrn}e%rQ*+eiiz1*FK%+Z1k&_AIqijQ<BI
zk9kqe*X;GFaC_Gx$d>hJJcumDBf8fD$N=T!lvYh9DF3ctY_0O~BfY%Uot%y5pw~QA
zV>Io`RQFfY>C&Yd^7dl)T9LcGqM$ZOvI{&DZ2R%=2_{GV|7M5HrS}h|lP#oRLqr?3
zC$`K5iT+z<4VJq?7|ZScfAQ7axn0M3&&?DR)*=!)!Ckes<V!J<!t|@NYVItr^kL_Q
z#_SQ0(}i}JS^fNeQ<h=_zVtrSLHB*A-TAZRR`Ki-$BrjNtB2tN)DkDxk+DiG!i70Z
zvmbk?;~`|N1g|eDu`eCn0J0zK;Fyy^&qgVwy+|pgoR&27M^(nM)(M)}wN#LYfj`H1
z^ARYDi3<83lzVnnQ7EN6L1~<;fviivD_TV#!L(HMN_%BO2TrPoo!W4v$U=@!=fD_6
zic#O@Pw>@@>adE*;x(c~=S+XM@w?xoudB4UoE1!yY(a(gNb06cLA31KQ<(a+VpD2$
z7R^TN@>FxOlSh@CTOJD4988m9us&76lf1Z`@~ydltfy4)R!S<W0w=0tKvhoUn#zBW
zvX9(l1WHe`jr4LUfjaW1KRj-0x)~Ff?$S|ycN>(*ozZ{l{SVR=QsT?qt7oU_i3E{r
z1hFhiPpzzJH)w39MeE<sMTuKjbb67cqO<NS-$3k=wT|Q`=0p8BUL^H^sU_ke4(yDb
zh!wm0%2us!B??H|H6zxFnGF7OlUzop*r7u;S`6)uf?@Rc75R4C8zHIs4RF+$vp^jR
zopO&-jrNcM884eyL8kbcl5v&JJ>A-;(R+`#0djHyse(mwAGh04wU1fo+Gi7T6)4CO
zinNwn%M8^QW2BYcn8T<&X62L?>f}4hyC&Gua^?4G57(@o0R<;+lbcL!G-lS33TbYV
zSo_};xIOCeFW;ovQzaNE8}$0tjW4|Y?MohKH-X+4v27XeKRew=X2bAY@DtE*c9qvG
zJE#h8@Nbmf!p3IXxP*@Wl4pso$`IRXER)b1cl%?4v$&23AJNkXO3_Oi=B|EOh(zE4
z(~*>N9!;yl$;MUCfzGAfe`4T~;C^Q|p`j9RK3(!>N$>B5mtYQ|Zan#LU&|uEgZws=
zUGLF5@A<mlf1>{+;V)zpJH%JkULXpKcZb^p`?eVnnXMD=xGh#r*oBb|?r_2Do%Urc
zA{9rU;kBPWJ=t6RCjf+&){y|~DwKcOFs_*Va;?%hXVc)@bgH@ct$H&<5ds`?DX5LS
zY5z{&UF|IZ@&r5U^U{xcw`|WqFb<#Q#>|9`=^LFX$gYn@5ec#b8H$#c((%#FK#f6>
zIitDU+F7Je6!$fZh&}Dirb_vdg0G`aa&k2DE)CVq*kDgA%i?aN(7En;d67*~5HDW8
z3p;p-qboj4lpLc1y?sq)?*#PvS@YK)wuB#-5SGq1dDfC^QDtuMUHi7(G@b4nrER9{
z3J{k8i0jaaoPdW4>G<5lc-}CV6LW-dak6<XdWMs^vBfu=i95N83QY(8P4+_x{|dV*
zr-s5<E2{ehnz3nwzvvqDsioPa)F^ui=!lAiq03)Z<n)2y7aTib_1^-eQjF4&4jT`+
zy&~xcq5aVYlKVU~<+Bj*hTKPC$sqa=CLsXpXm!*|OR*{!BtDWia+K@ztRO{vZHaTB
zdaPQ`^AxLfv$_(6od)_I82luO5oDWzOc4M6yDG5D@Gv;c)YJOC^cL!uOXykO4c5Vp
zUC?YDytwY${(oHzA&480cDRL4cu>(RaZmZ5*7-b>LZyd@BL3kqSum89Ha}+r;k6*@
zmwkmwUe_$}Z_U3lbgOLFo4{-xTH`Fs#$5Z6i?GC1EI}E*{a4v#_9VMmJJ9I){zhem
z=4(b?u1%2uP<4G2mKA-S(qIwmhr1b@WR4AA<BKq+M#D5WsD(g51<;dO*cqTafZZr-
zO_`-zxApl(NKpRj{%!)$bP<siyUL4~l`vPLP{6%<=*)<}(iQa}5BffBAt=~jELdaJ
z!oG&Er7i<$w~^aPXn7A8TG>UyWYFF2lF#xK92bgYwkSWkU4~cYZ-tB-nvM{pxCYuX
zq#7(o%#4$yj!`*SR&g2;oe-7u4U3W;7R@sEJi_F>bQr3?aF)1iOXew4jzJKm>pM<)
zn1tk^L9zL%)P}ItZj2cB^=1zR^popne(hI2IkS)qPZbg1`R4Li_TWcmc<wK1{hEBE
zn1$N8MW4xTPpvYCuF3ZC%33)h=?{4f-i*@f6|?#2(?*}ikXtr~v|^vosQpf^k6_}l
znR<br+k}w*c~%N$YAD5A1^4G}wWJx(O~<oC+f7-~s$3F?GQ?8YuF~ysRul-%01#ed
z5n+DInt>W;t$cSrrEwV{N2DTB;xdm=an>BLHIf@h%;BEqMYT<Ckfp3>os}?A20(YL
zg}fyVi3bg(f7pnPP+M0_{WALvCc{5ecU7CRt=$AawSXR*U&jn?FY3u*IdISJ^^c4K
z>B65*lV%B*NvEO4M2N(n&PX$e<lJe~V7hB~3^(u&&AH%x_NX7WOQq2f{*ah+EH?h)
zVaR$OQ;~B>iLZuMi+$SwE}M&tP@UYklh_v=C0Bh9Vm`7o=W6VV1XXN5GW+5aK~nwD
zyFHpj0MCgk_FW~0dXs+mpcO%Vk8}T=$@?+7-Ky&8m`%3nCrG;6#MSziR0Un*Cv`S!
zbKsF_d#2qsyu~SKdtLHgj=#O|LBl5Y$A^?)+-97rcEul9Q!OLhdYUQZl*(WJhYuoa
zPW?9yI->Kv2B>U-UP^|}Nc6@{^e-Dy0p`kpVeF3Lk9kC=PX7Fro#mIuA*IQ=T4%a=
zUzcD7ex7M#dZt~=jC=!|$*cj0{~RXzJD<7Y)mrZvT{)ZcFV5go<M=fc|G6p&^W*^5
zAy+Fwq_A6Jd`V6=lF|G2fqK(8E{1>Rv$cWd4!=gp7$(QxiTi!8p`nKUwv!%kL1A@t
zG0c9z@BBov)%>-&JqeG;Ml!+QBmGCBh_@TFinur7xZ1WBN1A1M4JQ`LK(*zA979Gj
zkW{7!JuWdR`Kz3|0EATrPe~7uj=L<*((}Vhx=ADCP|Dz~Hi2S!Rj#*i;Z_6X0EN0Z
z3ZJ@Y1W~X`U(O!qi7nKNX>+d>oNw=43(mLu>CR+fdL%HVB5;xdik~|}o828+GerYi
z9GOy-&xZ$`SyJgDeRl}1cl@Mug(rvG%%Gg8G({OiriqT*%cJ7<bl0Lz^3@p}7JJku
z3)SPQJb@^=2zasgUw`4|sbkP<f)G?txdV6e7KwWDIIhHxgy|O-i&2w|`8Y{|nxe6{
z)KRY@5IS~GW_I#3VR7-w?Zt`0Bfo^^mzwbqI^*Z|3D(pjdvuKS`R{<edMI?agY8#W
z`RX9Xd~uwSLGZG?{sva-m&p}Hvr|^D_MQ!}Dr9R*jblP4IW^X}0+%;c_!Id<`QdyG
zh$uTnEikkz@aWE{J!?PIFaP%a)H1em^3D*oxSh6ymf}P2A#`}QI(u!vz?c9MnM$&o
z(3spyc{FPNy!LPYOR>i<ig_re?!3**N&wtn&k9$H9y(AE!Kh4%Ha*Z4o(rCDUoqL!
z=r9h(MZ~k|9YFGyq^%CP6OW|A5_52S?|ds2CV^7!R73JPL%OoC-#c&qM`3t%4}W91
z4oxYy`q=QO3xkw}7<-J;r&D%B!N3+&EdrtU&W4Cd#IL?yeP`=KeLvIu<@i@6Ax+<r
zYQ}Pv6|DfVcjd-|Fr3(4zg%69UpVc(TCTzSQ#0zek}Y4pQK5CcN9C#Ap2D}kdiZb$
zsA>2T(urdOk^kfvC)=0#_Lo_1FhafY@a>~m1G;_ED{=e}5Lx*_(|?fAvu2TxkKxTm
ze-!DK+5&YZ4#(PO$fiwjEp;y9$2_W2*}_f@GYmcZHQ2J#Hgso-cTvA^WAXTrX&54W
zXyUr1$%acrjbW(}07TnZXCg~aE4`Y{IMFao3C|D4TTsy3`y;|({XN)Q-c3)}N(B$B
zb+{2mBQbZXSy`ztesw&v`vsztVPIoz89kfgzT9r_SnE=9J-?|*?u)`_^|Iw2ulfD*
z@Dr~wK-o-Yx7{kMAjkYndMh_YWUWhTS88I@G`-#0*Kr&icnkmQHY-^2YP>VcL(RP;
zh%e`^7aP%TkU@1SVWpYETDV%-&w2_~v_R00Rt4g&AV#Cinu*~Vl@pPTHkFCptb3-m
zI--d_nphv3SDi72M}v8|2Ahwr3)9P+gH^JdJnrjFFJnuKp98*k_e&9KAtZs=0rU5d
z+xPpATDKo(ZjM(fQnn}Cl89oEw{84VAFOmo7;VP8_gNX@<BZ<1Nz#E56bDxu2Y&9;
zT%ZxCpcidfx8=pkb`7KRc2Z*8)D**`Q&KaAe9P8323nRV=zVhr6*LZwxI~%~m?SQT
z>qInFD5t{Wzgx;1cw2(Nu?*ble}C@{ebI$zq<Y)15r!tR-2z~)g7rsO&t2Y?JgWov
z5o-<81&Fs@<v?eUQObeJr{nstfXV@mw5c^BJya3=ZdKnRGBFty^YxE5G~s|AG2MCx
zWi3{D_4LBp-<t`PADkqPp0>6oZcFV+cxjE_Xy3lb*fp<wz0<~up161HDugWE*QSH)
zz#|ea{<1NV``We{D&n;tn6rKo>{CF0ZSB7j{;0>#<gD1sn&5ia%O&3xS&mJf5-q+@
z<>_mT<EXmExt-WfG(~gntt?#gMMZnm-kzZ^an*4cym5hn5EEbWK!+yapj$KRzu?nJ
zO8Yg@1373*xBf^89%=Z6frI{A5Ezojw$h&7u89!z)U_dLk-rHWc$XJ<Qx5I$dMnEJ
z{r1PbWCzA_<0plj_bf+3MFb51>`PvNj>?l?>yN#zt-Vjo{pws4qVG)>O_Ft5M6)FS
zcf0x_{9l7eDQ#(v%G<$91Y7!<f4|%fc3Y;m86#=ud$*UgKu6*+K>IW`rcmlbf=KD|
zQoncv?HfTr12xY;N`l4Oy|z!p4xwlFCzmZ2Q4mq+D5H2h&P<^eL+lS*X5t@^!oig>
zf}XI{`3@qLXd(~f_PtvcAL6jBoZp=5qJ00Za&!GO(0oxAGi4j`rFduY=CikIJ#45?
zvAny&3I^Gkl(U@v;bm9VU##|Nv6x1C#lj1?nJ+09(rxH4GWi9<Ty_o6#SM!JOqAAH
zJFcqdi6!|W3qQOnL^6;(LU~kZb9H{7#kjAOcxGnit3?(eJx?H2WgKLkn7`j$#w)Q^
zS~-*^X5s_xzA+SNWgK8jZrcR#OfD<UYC7Ns&tpMty6)I0EzH&U&QE=N-a%&;s|ZfB
zW@?Os{93vImBz{yJdNO=b22GkV$;dPL}N<5W}6)II~yN$4t0s)CPfk!e7UK;7~mS~
zx7SSna&K~`gN*7!8u?<6ln8l^xM*VWB`!`>{OS>}Ew_ZVj!eQ9JmN(Dwk@~cT#mRx
z15QeH>@`!yt-u&mhf262Hj<CV>P-G^T4O_4{*U|WO*Lpx%z}#h=%9UvJ<%q2<^~qL
zt^3-86=@^%4Ryj-=+!_Up#D%hkxhfmPT>Xoq)WU)-K_lj`0Ba)JAy*vm7ZxjMx@x7
z>OV-M&COL+!d;!B|9q{kA`jwiE5s9bbv~qd(iJ*`GTl-S>ja9teMyu-2ycI=fFmlS
zukc+ZbUnFpS+>-~XgHCb09BNn;k#Lhuqvy#Vysi~!mVZTG<z*0BCJkKC7n{vg$z-n
z+ACdPFSN7ZMv2;XDG{xZ-|f3hPsqaqUU`IJbE1Z)>eMy;LC*v`2cv&&dvPCNsyX+?
zhF*cM0Rgc<kBDYZN9Fg<cjCX(!S@<Xh7{btaX3i`9<T!580GAlg`Monwp{O}2(12-
zsdb;6uYT~2{hLcA02L`rf3)&QRWh_PJmUC0%M7n1Z!o;qu!x-sE)4!q1ePKyA!n=t
z3uu$c@i?%i{GadOf9%Z&BI8?9kD~Ps1tCA^#=*H@CE3ybX_VD~%LOtz0Q(D0g7+lU
zzSxu>*^^60ejQTM)RQZ*aZBl6`P+7?%}lPQbOn&^KMAkZc8|66m;lA(vDk~VP2>mp
zj2*Emfa^J<ke+nDDj9{CsWYDtTGo*~AK79!ys(&I$|mi~J4nxmD%=yh&7OWeG?GT_
zpH@DH7|tsBr5ZjU|M#TbY$<eo$9bv!QTB7JdswcPuFoQ2O(!K-wQo-<l|8OqdI+9R
zEgoIilZ7Z2C1Tn<R`Ioz&Ty3^L-Ww;ny+T3vv=k!(lPVfX~v5cG3gKv&L;7oYJ$^t
z^9dvTrLlhDE8%STUVHU3TOYa*`CaPCkxHiVnVDCnql?kfotPQ1JN^Shg+#tJW458R
zw8S=;Qm;Sx)}Ag>G@itbZ!0R8geWXKVa<EVk~%^Uc0IA-q>j{!41M2LLy6D@XLP-=
zb^h_gbm5MnCwgBq-kIZN|2CCjzuU&NmxxQW0g@h}uaG{Hqm|R4z)xeGd*wIlNMs%1
zgHanEOyDYK=M7`U%qBZ4+L5s|>ozuO%?5lBX0E6+!o<}#-mc?t<d2jdA~sZw;$Mci
zwwO<khSds{$}%l5ys#jH1^L?WPQPGXTjS9a{R2e;i7I+2-+jb3zWWbSWXGysM7B|k
zR6cPp>Md3C@>`BtS7r2}f*SJqJuFrL&qw&pA!BcR{g4pMAaZb?7+LgNfPFicMvPzz
zV#lP}ENeI$nOFw&>P3=<7ngQYOM*Izmwy%T?@Zn=Zs-~cPWF^%?8UrSj5Hu3bW>3&
z-{RMwHAh4KQN%}BlU~Ku^7@Vm?IG1!w4$tumYl`yoy{n*>il;30jTw2uk3C`jZ^J5
z`2xhG(F`;QQPV<ro=A~|7%bYeD2rO4vI}34p(3omJ&VwQ+%7Jr&5iG_Q4?MJ&PY&i
zB}%K0Q@WJbaS;N6!2hWrPW_V}G<MJ!H54aicd5PM?eA<h=<%EP=e$!zbd$IRyArS8
zn_xuXdRq9iQRo8PF`t=@svSC-(+Be?^M3uaMM=jvx6R$EI5O<2e<(c@fY91Ipf=Qt
zyD$1f*Z1jSnZFM}`1POXV!6R~SfUhY7D6yCpU5o$Uv^QfuhWd(TV<D|)Zn+ru~D@D
z_-yq-9l5QHoHe?_m}BUu1SSKiAK<WwxohHfGl!jY`JL%CEN%_WYrfMnyhWTHq@@ji
z?F2y1Cs7nHKYxw!)2p@@sM@|Ic|ReHDlL;hzIvY_OvlvSsZ>@hsdI$w{T0=lnj+uN
z6WS+p9gR%y0uO5m1rm^$DUX%)ySa_}gH~SCVXVTvs|WGKBFXiJ@1*r{eXEk!Ai}i(
z{Tpm-H*p8K_WP<Y88I2I4!cF@Yax@w2VYCh>Enp0*XlDr%cRPb?b$7TzdAG?O!u|~
ztS+n3v;Y2P?gxGWeUnwHHWL1MWLz`kUyw}AS_~A%DAFF<i|G1=9d9dMwGLOmv(b;c
zz9uk|zgT^f`rQ!TL8V;9_-kV%<#{2KMMxmKFb{4y2}9-zCuYNRprnH=fd8G2z5w`9
zcSk{_zl{D?09$}|i94h*srGk>Iw*z&$!11~z_^2J*Ir}4&>$2f58L+kbQe`+T0i4G
zSgO;gcX>!%{p4sBsMFJ&8Yu7S5o>DmF|_xpPN90}QD^Ms19pPM*+uArocCu8^S2m}
z^D!oMvJ$o*PS+?_1+<R7nzzJvRr>36I$^8-?E2ghMM{#0Aq>+#J8aY+Wh;ym8B<oy
z!eUQq$Pgl>1o@3_@u@c`WMJgxr}t+9?vv+tC#k{EH_4_lf5EJcDg@Cy5I=l?&-9*d
z>-b)}_eZZ+GdjYf`VUrXCn}alY&wasUQ*(zpFb4`ENl^-FR%x><S*GYxriu6T$Ckc
zyl{_W5`a{?PuHevK~FRY(>}D1(ar7bjl-XZD5^RiWW#fRyEncvxq6>nMhXTr*H^T1
zW}0*U88IsZm-)Vq#=@sL)`e3x7qcmxE%V)Vu$wI_q)lHY@~3grZ8;Y7axy3V3Y;0B
zk2?yF?s;e6kyGWPstj1B@Rf2lrAglSG+S{L2Sm7P+JVrUHss(<2?dXyjeB)59AG-e
zH|zyb<!))JuGKpk6nIpE6wmYi5o%@W#?nbEESxXCpw6sw&`1jVzE#V&ET6zMV)O#6
zJ`^Gh>Ne7OAV?=l+qFoWt}+a~J_~j?b+1rdFr|$y{A%flSK=pRl8V*ITZa&NYyMVB
z>E4cKDSab{L|aN$_$BX~&G<roD`DF`zfK@3W1B5h*4PeV6dk@P)fDILrSSQG29cK4
zg{c263N7p7wm3fP+n;G~0VTBSgdH;&a#QC^0Ww*!Z)(KZ&_3-G^m#nZ_~RUT47=gL
zW-kDrvG)Aa09<nHt4KVcglpBR8nOK}tt*7?zRlCZ)&SL1o2~m2X7dn_Qj)iZf^5$*
zcqepFPo!l+Rcx+%9|}2Ow4AN4v98Ak^Bbs~HT?WU`<8gFmn0HZdQsPngQB#2NaGSq
zN?7p%0Mf($kj!@ThFd}V?REBwoz5+PJ^Np0BT+svWI^3TP)a0-D<J*30A4>~2k9BM
zPFM{UK`{DOgztnMS<Z}Hhk}de?49-lQ5`|V4Ed!s0<lhCy<k^8gJp3DEaPID;I_*1
zs|67w_>Wu{{Tbdws~BtXnruaeE7GF>>|m486tk*v{85>T#x09n2l_v@U7^G8hEw%|
z?)Hr<;j3FJ<824_?XCRITV-!qj*RM|7Lmf*AJ?vt0V+(^t_k;wb-IL(TQz6_SoiL1
zd6i*Q;ks50*(4N5AAb;)2s2l`mHDI{DtspKl`NDubwIPj#vU6V-B*bCZ)mF9!qe<Q
zUwP#LWPWw@-|z+n7*VX%vNdLJmYyhayRoNZT+jPk1y8f3u^v6aIcDD+oNS=yb@@^#
z+xn2;tHY+8!ahh~3je&UZP$9asSvblnAxiSM`ExM0u%_XRiKO30hg}K#wP8Rr+7_G
zOlAWVxA!blb;k#F67+FsmI0ahhL3q-acXokrcLaP+$h-b{@TmZPBy#X4`V)wRD0c(
z9BulWxrFKeD8W+6!Q3JG_YL2!+}_foH&+E}k!h&v3BSm1ea<yTe?)Z7?tkJ3(%I;?
zMrzw*KOg75H@+!7^UyA?uIBlxYqv$f--H_y$6__XduV7#uoh@UGMXA<w0=tJb~TVg
z2;2Sp;{?r1*A0`S$ofD|NYXcDp)K2vV%jh}4&kk($7Mlo?z5Kl)9mq~OIEfRK42+D
zVAP0_+5b(=a8fFP>&3-Vvwx)eVMK50?IfoypS0Vmc>SX_ROIibe|aO>uSOA#*T@Vq
z$h#uZ=uvujEB-ay9^@a5E$Cz!<#AwtE+)HNI-u>=rfTL-UJ-#qwyhQ{ozH6!XN;0j
zC^?r|qjGW*5nSrq3W-Ta8tHSn8Pbr#z2Z#;Z#)^QA_lR=%J+$;7@ox=57CUMYz8{U
zBY>krOVruO_nw7DnKNdh<u9Q1z7t802{l#LV`2K8R{CQTCW>LyOv(c32l)3Uf*zeG
zttrkynal`DU$(G=4<%P6=d2|^;7I{_od{h+-lwI;xy2WikJJD%M@w;e5Z-`DdSaBv
zU`H16e)LQy*?T4pF@VV>@m`JBl`?IwBpXU!qEf$a?lp`jg^bhKyW+#a{+@=^NXBM;
zy6|09hGBTdOmTg?o>sCHbn5Omvd<ugDOzK@c`KHLy4rf}O6(WcMTEbA7b<6fjz)_u
z*Ec-)>q_&mm5Z92`HFZHbdSkI7r)~vRO_hnxuQvg>wd!Z<vJN&WDX1rxHP3bqBR$8
zyYH!}<3#KF3{lZc*h$K%A2{+>;T!k1+AFV2Q7G~MsT-@1C+H<$&&7qYZIFpOu?RDy
z{m^uSOX2(&-BX$34AJb0M2~12idhMnuQgJP2o@uAl=iKEJs6>^j2`{%mt0hH!BoYV
z7>z({>3=Oy6?Q+E)g?h<3yi^rf3)4_QsYSPmcE3eX@tO(GTngp`qU-%qwxXN(d){&
z1Z}xDbnSRV#eR_-wkh&D{}!E_pg_bN&KB?~@@B9vI?M8<QooD3jF5Z`k#BFXCueOK
zeNa=oCVF%k`lmxxv}yF=Yd1B?$s8*nF<*{glY{qPB7&!$$?WTJ5b#1wmI7vWW%jLa
zF4n*Ay4EG$R|;O-K}opu?jI|C@%@2#5)?BKa>Q2SuMS8VUVXypJxF2YhOI3`mkuO5
za^qc2S|Ej6J%qXd)Ho*BDFIOb+@;~Am9!qz`Q*iA=<L8MP<w~%cEAw&4fsAkC#0z2
zfU#iONAy+h;r|&H&tx|gVu_+cv^9lc@=n)3CuNOqyarz{vwv+tuNltiOHMR@C(w3$
zrIzA0G$jjwX2pB5!ek!e*}kzKc#?QF&C;h}546tQ0Tc%!(M4YQBP8RiO+mx^GrC8t
zd#l(Kgcx7uVJq(W+?tC)an4RIa^YSrBjhAnNdf+W#13%l&|AJv;Cm0<$>s*G8GI4{
zyJrB?EeG>xD&NHz8gvzZ!JhhDEy!^JShC7yBvmWF;H68|kG1JMM|vn{8ms)Y3SX~g
zh#+y2s{HkQtvjzMcFv4(<~Smzvo1Ib8`PayxT5WD{zzsX-|_B8SjK>joE!WQ-<PnT
zSUpPjt4WskXtVGUKY$aK`+5v?-e&7H^HkgNv~CWE<opMz#8N{Z)_-VUF~FM(T*A}R
zEtt*rOEu=8jDf2EjYQe2lDWXWrBq^<^}+L7hFh&0zW{z?UlyejKMOGK6)`Y*d2g@|
zq0|c1DoaPsA8TeDZCx<bzGcy?gSQMCj4-)v&V(BmCSZqQEMwJvmfEnb;bt)SQq`vL
zy>sZ$G?(pKPPy$!x_gLm)yb{r5`y+|A<!;CBg>J*9M4dv^1Adht6P&2Q_Qy7--b;j
zK+`l+WegAJ0Cz~};y&duu~3=2B0N=>*Jf_JA;vN$`?r4v$x-tVh>4URQ(};$&j`?F
zzJShI=&pX^SuS+EI_}Ntk>79A&HTmvB^35@^O_#7F==$m`%LK8YWm+Z{lJC8X+ftM
zVxq*SZ@D<NoiR*<prhtl?JFh)9-&HmB8CXRyWMhK+MsZTYxq5;hW>KSr=RQ1|3O+V
zf^MptQ>X3KYIHf;lP;|zaGK<L#(sGh9{3SZseR%kp|sY(j1Pl@#7))BZyvvN_3Tct
zLUL`%4kpPf;nuh>BsLox<n;p=SSKgV=p2<fzH>g;{D@ku@#c6!vASC^kxNkMYp$P`
z^XOo5Vt3%aq@N|JTf2B=da8lUIo~#??_PU7z_JDU4N)-_>B;NvHtZfMP5ou2OUtjt
zyT!{fvE}AxQu&Ge^z%XK%O6NXw-H5DR8!VwSNA`M`pDrVOJ=n84LvoBWK$wcyNpLS
zY2&Nr>Bt7?V;+~YX7`;4c^r$iR08!V7l29S-SPpFmu#mOh6B7P^;-oYw3XMUdb-+;
z-C=RuUs@cdoiP4koWi}$Rb(osvo(9NQ0c|d$RI_rhm;~%-odR?=a)Isaefo4r~#BO
ztv2rI+wL`$g!*sRNyE0k?w55Rx`dggS3bH~@DfxXMvZSj-<D}@$ZH41&?OA7S#19$
z?q^fLix$hqxY(igJ1=;qopdi&M~?SEeD+Qt%xxOl|GOYUt+M?#U<GwzX+gyj=)zT1
z7c@o*18AV+xbm5sBBVqm(AQ!=wNP9!e#F49H1YY^K^}=v4W-6M=XcB}tx8e)iD$Wr
zmrf0*-$&b`de0!zEL%Pt2V{5zMGDx2BaK{##>_OZnQ+nWkZy^>A#MsQ6{calODaJ#
zD=}G?HT8UT`Jh9IKKr?HLUNhE5>9vILLO1;KR_02r#UUsK2i}n>y*1!_^o`ql`>KV
z<YRGQXn&7)2%URo@MN0$qvmt($eRK4&=Z8c>qL}#R<Bn#7jp6O`#SG*(lg{Y3r5#j
z32fo!z_7P)qdvMVcA$`A8+M4kci#_5=w>?-l~UJB0ts?}-LJi>=S(IrY6t=kcAEL+
zB8S7APG#iFeTawH5*iDn>3kMhC)jS-6Bljja~!Aj^JE}>x!kAvqG{q%HR1#qE11-4
zRi%GL=<0f437QjU`h}3`P-l|~;NZEmc9Sooa!X6Hk^9SL{`P!l38wJsPKl7eDDT(b
z-T_U)l#x9Cv%_AM8N0zW)Q0!8&PKV!@7q<c%&VGiE5Sp1gXg&fQ-1Sc<z>Jvo1VIp
z2d4p$Z28gAmzk;l%)lx=);cX8FkgUyX*Ar`Vmz1oUDonT?4YP*_FA!Xphi#CM{X&;
zhh_U%OlpTsci6Z2PZ=e?ugp%sSuLm9VlodP^)kJIAdn(N|IG}%{nYSkdH&U|YihtK
z_1rW-32*5~i5|3w9_}ebt(hR}K6zE+{X3LW-Y9cfcO73f0J!}$2QQutw3`3^`-<1a
zWs7XOIcm>dpX72X3|`AI5S+9Ux3+-Q&h|0>$)@P2qh+bA)eA_cBE!NEqOd%w^k-M5
z8}ws(8$?bNEx^SC9`JAT!(uhwt`KB6i<5UF2u@1BnN)%$4<#LyrMRhWoA5#sWm!D9
z>c&paWBa-i{h-(TN*wMA?o*w1Wu%W7%^p;gj_Y{$Di*Cg-}_)HaxLq$lo1|KxiiO5
z+dWh);Ih^7&2(4V(E%53n&*GfyM#=d*ii1KWMEpHGCU)j4*uWn2t;%q(R;v9+2CEl
ziq9`^N3v`@g|^e?b4paXkEp-wNX^t?9XW=~by<42PE={uTdzB!OlhkyB$EEYmp;~1
z<7;mg?$+3U{|+u8&Z~b-OUl1%LGE~CUq!S>d4zHg3dvJ+h%b>qiAno6TQg|XW!&OV
zpgnW0nZ*$yWDQu1tteA8oU|8SKKY$YmmBy2EnLWg_3_dEUFs)*=1L1iq9GhllWGW9
z8u*AKTV>IK?IcW(Li65q4KMuq2r%42BPHFll~05_*W4?I9m_%)iU^L{k%Vv9bO~;G
zHF@IUi?Q_&ISa5%lAhTYGA>>}GD`W)`cGfO5Sb949ZlPKi(snoGQY-;lQT8Qd^eqs
z)y|sw#jJD`73keFeZM$b=Wb@Niw8z=gl_K(j^s|xTyhqJ4~7od=*wA?B4TAj4rOHB
zNnBPbYC5S0t44xct?_oYk>YLQ-cMMWVRBk?{z1L1&dEX5v8T8p9@C9QV8UgZld}0q
zfj3W{rBteFo^|-Q1QegKQ&+IqAk7}cuDI_!)D!mNk3>K3>{`G-i0xczIW6I;t&0N|
z#qMYzLO(tUuG8<D9qZ4~i+&hHL$W{kX(dX`@MgZ;l=EMq<zrNk1Mo;sx{3-<sS(He
zXHdG%h~+0A8pzwqjdQA-5VBun79F=gCJ|IuLEX2D7g+jkq0y><O|)zK5kQe-ZjF4p
zS~OaJg#H(sH42G-&kI|2u5JXS#px;Eqls75sSv{j;E!j~x0JZL=En+QQ7U0f3eSp+
zfxgw{JnHaFbrphjVuND)cAo2yM8Bh(#r=Qe!yi8J7mv9dNTjEx9vH=YT7GS~X19$=
zJg1WB30w85?$#Zc`I8B1@7jM8GD(9B;9K8IKs4zYQA)^PY#Y`m*Ouz$88Oq~hDqjW
zLWL6}oD#D(pYO(OQwQwsc#WQ6D#Et_wRZrc5yv5qi1E`2E)l)&D&;0k)qm|$_k=eS
zSOu+w7`+TyXPwE!OSe=dSNN9?30v<NG<iRm7w<r`&U8LkCsdJsI~LlxHOyT=!~pli
z%!DD!LPM9S&>bpy_prQ5Nw#T6FHImI&bhde;{8yG#2%{^7BNYZC>E!uJEvCph4*vZ
z<^KcuR-^h7Nh}qA+?E1G*d4vS3XR@5yOBlou0Dv}z%*Oa3tN{(Qd3P$CUUZ`p6}6e
za@}>iglV1JsZXG^?FnWPIf74O%lk2=OSX2&9Ns%_n50{6Ksui@98Fd|`Q@HF{vvtx
zfOjp;Q0cb>zlC~f=C+GZa~1V-&7Uk{Jsz8_mA;j!6dMeVnrhNh0*{S?YzU&41=ka&
zw$ZB=2Ta%GtCWs*yDQ!fL6z^3z$&~vq26TK(+Rf<DrB{aLtWh|cV9ndjS*7k<Tr;^
zjV2)kUBXMjKKv05ffMsfd1=+VZY9Z<0hp_5<A92Vva8dt2IxDV63d8V9}H^j+;w7w
zkE7)$dFazBnH<6Te6}`uQH;k5Q*$ezY@0)9^6C^sFEK^MGoLJJ3@1LIsui<(!qMy^
zTe;uu(RGwJbO^bvihSHBa$amxTKKuL9y<F-x_J`<H`%pX%hk4Lrnqp|i0Egj1)g|J
z<fL>^Eh0?N8lly`FCe(%fK-k*r9(-7fi%ogv#_*|$zvuYg7_J);_VqV!4tFq9EAE5
z_s7|rh2iqhR5QMKH5(BY&ea=l*AyP9Q)x;ZzBqC4%L*E7h1>~Mo?i2c5$%!G7Mm)K
zB!4fFu32rmFWiMNB%@gQYv#aNDuM1JG3dV~&#lu5rFl*>1So`1DBuYe@&q)^k#A|i
zbP6Qf-ZkDDi2{4>r5WSXO|#l#-A<btdE>-j=yK!%$<Ucb;>9Jrruk##RBvyy)h(cA
zcUDjyNaVRgHWQ_hAw_LFDvFynn*;B=vlm^>b4h>Itd!H}T4RVnT-4`2*ERB&pzgY3
zs9Q^^X^LR&&95JwJ=P~xFVgcz(*BEStl)4=`~v4YX4y>mfus~GXwg1N4$9Ph>LM^)
zCD|rOlK633%`$dxNMcd(AEc(G!tUm^_>TU1AS)mUi<J6b3ZCJja+nhx;7bFVkY*VT
z*3Hkjd_ONN4TpOZ?8~(g(M=nQ<R1R=mCxT!-H%i}Rk3*5#{^ET;%k&JtZ{0I{Ps&M
zh3%3xY}XB1g-;*n`E=_vD@{rYY?i@$5qLe|R1+MhZpx5%XL2G_yFjS4@OuNWN9V+9
zKDQwpOuYpd8gz!VRlWbj0+P_kW|Byce|?*+`N|NRl6V7@6uEu@xgnhNgG$%8h=KGk
zq^j9Nju9@-05+xj>0u%*gs&2RMunS~*MPK%e?Rro6|rsZ_>D}sYI|KIUX>nK9zG3p
zG**+pFhFd4-2TrioJe$3*!O2lzQ55G<HxxvquJJPtrADyo0M}4HN$AAOnVTp^_uBg
z-&(J1zN1@UML794)~I8X-6@eHangk5t5&%~o2g}DhzVLTvJkpFW({NE&tw2weCm`Q
zxhG54+)m~Ly;vZT6ltO-D|B=9UQB1W<myD1W#naly~e44QngjO6@Lj?fhlK;JcRN`
zw5#zvz<|tyae5P9{}V+ke^||`Z$}8}aJ-gLm~ffsj?I~hqy5Rx-W0HE-Jh>Bt0iFO
zY(Q%_V>!C#+6>>RUF=iN-Vp)S>RqK9km@{b0lD<dQ<xU@2*WU@nIYn7u2LmsFPUY*
z5~bh=Xc!j!?o-vzyYipVBCbOt35r?L^I^o>rk02>g||X))#kO1a_#w!1Js_u9o74<
zY`&pc|J;S2JST^kH)~j3{W=Nbwst~ARmnY8=|ysUA@o%_XH3*L#}jUp$mc&_!KjV1
znGAdeb%g1HDiTyI;=5%ml#rJ!?<=N82pJsCWKr@l3=b2S5J(-X0ZY<2*&4I?BW>h*
zfJ0%Z0TvfSzHaBmRkOGsRR(&BcllFe(M*lux~_by0iAQ=B3Q#=IlVc(CXH7fVwDsO
z`wtTPq@@|>_g^bUf3JC~*AzSuErwqbv0-9{rNx<qG1Lz^!l-xry+^zsXE{kBz~q?c
z!?8`7vS$Ssx*)vY&#GUagVH_8p65S^c$hi1+P_-~LNVX`|42H^sHPwP?V~6qA*FPO
zbdC-|X=xZTLZxAJhk%HH;OHLRjIklz-7T@vNOw09-+RCRea;^2?0nAIgMH%K`+dEx
z%RgAekk&n_Xz6`Ced}oEb(35H-pd|C!~iFon8fJ2+&OM@)1uMl!qZnSg7xP97BxR8
zw&E0bF=Wc9JtwfK_Q!lAzUgh+UCWFi*|$>l@UC;{MM^@I2<#+Mer`>=E+o5vQbp|X
zL%Eb~w(K6afs^Q5vPA1dtcUP#SwS8e8=wUak*!+ue&y~v4|!*Ho+&`>>eOSl${kM2
zzVU2S&I^E-*ne1R+C}y|lg?JA*OZw#EfT?m2Y*huP-u6E&7oo|^Mb;K{!57TG@&5o
zdiY?Mti`o5^%9frgb7_t!m9Zn&0s7&$wj$<5S*l%e}fG|1HcC3Zi*i}>B5!t8!oqM
z^t>fEHO1X+48MFA(P=~JQR3(7sdk}0C%EJNK=Kfry?W1fYMQ(ofK*P}MnD=>h7Fr7
zS|aIBXASA}6}n>)Ho66_Poham%dl3~GJC$!*;G6J9abYai{*L#?;63SFvP1lmw#aE
z1~a-S|7BGgWJ-|2*bZ$t(wN=3z)#>wwik&(&0)t81yK2!<I<sXH91sDk%>YGZUH0g
z(_-46?c2HcL_%UedoOSlLn6P-U`<7j1cowpWkqf&TqVxG7Ny>lWtNZ3?;Nq-2aU{|
z{YVv{3DaGi8JBv@V}H1x5$H@Frqd_$v0=sdd?{j*Xvo;o_Vt{e;ZR_$<X#j-j=01t
zaD-ES^V*Nv`N~t?mIZr%*QTfV`zkk*C;QfEbJT?{3Wr(&CR2o)3v(=@{PMN-f?|q5
z+z`|3QaaLFt04Rt3t#dco^py^rnCXuY(h&!&L<M+2L0J$j6cq<>F{GzKZ(Uuoz)F3
zHy{HHGaIF2Wx?$)=$$>#{?x9>t-)QO>1wJ(M}<<Rl1T*bt+QI$hRZWO>gK!!!Ftyx
z!-Pii!Ozh3PI1cGpfQqV8$>|CVX^Y9ncp@16clpTyBQFxUUx81sju_n;lNUs2cR|6
zs}qk{UHuP>d_nvAf1*%(>hhP9h*^yd*=-g7cq*;xI5j*Z+T`y$xTcRTZ?N~BmNM&y
zcp!TN%31xEhnr(rjf$g5(jvQrQ*`;~=2vUva7l~UwYeoL7jfh5Z)Kk7bG~EE)~rH3
zUcrRFEiwsiyBA1*gs0R_?~XUf(SlyW<dErjDW|K>X`U3kF;pJp^q0@=91NQoILu>d
zh{jT41s_otn<ZyzBz1TZ;O36fq2I0|M$F#nHB~hNe+K90V)cd$9Gi~z|E#IDLHoR#
zA%Jsv&SD~B5v=+d9{9R$C~6hc$>QG{;DMvT$fdXf;t2X+?(HHgVWC)QVef4ucUA-i
zdL6OR?X}1ArzYsDOwJLF?jpBiG$Zlw)~?&eMhuS*Mi%}Vovs-UILr2bH8%xCHqj@2
zURxse->^xZ%TH8IkrkU~{t@zer<SQs;hyyj<`kP=&Dh!f$c4$Q|6j3-9(}2HZbkm5
zr-CXZh?^9!bYG-`C{zyNRXT9UiHEDc4%3}2RWV7fusY}Oig_aQM<pc_x~073=9mg+
zcr7^-dsmF4O+(#ijq6rVvsQr-l4e^_Rn+%5;@&a;?AvnMR&Kl&=QA@C!Fu1(nUuwi
zb+#ZcxC$?$<>w^_BX8+B%K+z93ZjDf+T;8zrr2RYUK{oSCg)wVN=@aZ4{WFjI%cGm
zDHbi(OB@{T3!{{-trwo3J58uB0hvtD#_|nxR>TW9I5qjpKs=6Cz`rTbhmP&}Y8XXM
z(^wg~6G%m#9MId%f1FCUUTfv>iqGC#gAAyE33{a}6%W^Uf$e?)=FgH`%my!0PJ&Nr
zP8Rw?*#;kYhCX3S?MVYewASzr4hU_J1(C*wTYg&0%3!&05ufByX(BQ;JbNkNh16Hq
z3I${HP4?Wu0=-oVcTF!4AqyUr=ZaCu_m~(J95M6F7#~yHX@GeB&->CKhc$o_m|oM&
zotl=VxW<K(srZm0CeNk!0H2#-G1fnE^!eLkb5m5TB8fur2TxIazq@Iaaj&s(Ig)iT
z%p6pCNqhhOXHbo~e$Y9^QhfhjB02vPcpG`>SWANqbq_}?V|e`5vBAXeD(e<qrWHZT
zzglhuNGU&?-=Qs!3dW!p(}O1#w<+R$`}`!Ez*-`fzlVuRG6(VV?;3uaCJr#7Yuwto
zNHZXtmNIsgd`r3E>!WpziUS*#C%tM6S-7*4_UxWE?l`YmL(P#4T>6fBi3tmO<-cVV
zC0R2Ej(UWjT&p<ji$)S+;&iDquxxAJCwQpHwLtO^W(UpIPa2(3fUu=O!}*>*Vqz(R
zy*v$*0-Bi9Y7Udot*l0D8G`oO!e8uKZZqg=1VvWNJgsSGGlaIF!CJ+^p-`|%wZK^t
zu(gJQ75vFZ9(hXfgt&gB`<~y>9{GaXL2Qj`E-Uo&o54OXq140NFeLbpk)`N|USQdR
zr!#`7b7&nfrFK4)FUpFI(+a60nXjMhtC@65JY%mHpgH2CmcXeybbe($zv84YJ!8W%
zHgLXmZFS}+Ou|=<*<bc2^jaoY(~4<h(E&Z}ivrRm2Md4dO%0|0vT}ni)sM%C(_t+0
z78It2^(snSfSqA62IF@8VUO$pI$$?pUm(1v?^D*N%IQc8NyWgj48ru*{p{fG$Yg;v
zRpek=RsgAaAmY7=Zn*xGgnUS^&+4ZpsSFYCsRzYr)CJ!9|Ek~K;@KC&@W#;c>l5&?
z5{%CzU9Q$JI6Q3Hepgc#obZ+tWRBvBW{6nD{dg}1>3L4Pk<-^m6rkgzt(Rc>c&Ih~
z&nQ-YZD7xYm3Q(vZwIkjVUMT(R|b#@83e<6fp%9F=o|`uBmJu1t3?&TmmXp|X9KTk
zf%)7AE|w%pb;9%`k{*c{$V4hi5C4QZky>~~mHm-IxA@aW!3u^fwp_Z(4OI6RkvEyp
z`eMutnJ(}x06rVmxFdFY>zeORuzhI+ZZ%^zSB&GymR>|1xTw|5k7S1c8+m*mr}>Xa
zn6TeVg-YHBoikkK_Ya4jAgp1eTrtVKj6va|#-kiyR$cruBPN$YcqI8O#7TBfluaY5
z&L&z_jhiR93j=$Rx?J-Bn&=*1+3h8cx|C{c4h+x;;^87ZuIsOXQO{e($t5>nz51<P
zT<$bM1lh`{`hm^Jan5$i1+BY}b-ss2>Lu(NK6H7b<Wh!YT$TofnK%S$OEGvC$Q!Wl
z_5BFV+;>yy3RdobEyTy$fUHPr><}|5Sd#?K=BccFMf}c}N!@F2fq6Ani5IMXwx#2g
zgZps65sWt@=8t#jKcnF;;G%m}Ti7_!f>b%?;Yc?l9Z-@OAT(|yKh~7lI_=&k-(o$A
zt-SzOoZHRWG`d&&?Ci2`6dzUi`OUbNCMNVoP6K5ghUbZ>^fR}-4XAe-r@llqST`um
zo;Pn;CYpTVuIKniGXz!N0C^n_qS4GEw;Fwmr)E~EG4UJXnCG%Neyf&LJg_p=NA?d7
z8zjg%9l#r8Rri-kUi`~)Bew^~Fs6cOv?;QuSeW=&y~M*_6py`wC)3IsFrPckZ^p%w
zKD8ltw1RGr$dHP-$C|;MO!9jBj-#@g?UP2kMxOSHV>zI%;FtPnT;CrQpg^z8P`a-u
zDv7uX1WOY+*BWN7C$2s=?Ls!^{*lmN*diY}a?_%Qb68Xjx@t;pos3Q({AZI7Aomtc
z7E2yB9@ny07<I6qz|`LS@AW?g$gTdKPgh(Whn*eEaqjo4`1ASV<+l8?K23P|<|QSq
z#%p=OVhu11cyxxA81%*z9&y4KdtFr$ufa6uH8gn2XH7y@J4yyDEyfJr>14S%?ojbI
zzv$7!dpHgmBx$G*s?LIoKvV5X_8$<uqVcu9`mVi=7RcVguK}?T{ZP}9MSr>=FBug=
zA_E`hjDMf?hD`)Df(3@y9lZeutH6lowobotY#N2ccGEIFN44rSLK+f+YS9*nR^QP~
za))o~A;Eb0o9@hHXtwx50^ucx4@?&Yx^qm=1MS`0Y8>b$B;=z=d^j?QYBQZU)8<^;
zavqybckkulYKd1|^6K8~>f+ptL=H;_g^bCfzCK2bXX_mD1l~O91Yf1$-j#_LB_$D;
z&N*M=f(**Lp==X30R}sc8xLHut4qco*M>fL?<$3jl5M6$mJ!0|2A(jWWI+6I!p^Y5
zOokSnNOzSwU;cl7HLAe#jbxlj{Oy>_wriaNEGukLD4h+$Lae=xdZY_31|w_e7Gmi7
zg+Hv%s@s4u>A(fYwH-Pds}dB=7Ekw0y<qBrT2UTqX=Gy5->9B6O*n@fu>v{4Vm$pC
z`V%|giU>ZN<k@DHH^V~C&#gY)NR1v9D6sxc{qcE`=GsnA7sQ~ce&*b5(h2h9rnFt|
zS$AlcA1cb}xWXq51`NhNnHK+mzAx>cuc}$b`yHJ+VlUh+K{S~_f@43aihJxTKVO=l
z)K?%yD^EE0j)C$c3Bm@<Kv>!Md>~Qs00pr~iq294SW<p{T^znKg7H^7RLtGDXoc0i
ziVI(V4!2hlwLD}?t8aVx8&Ba}-HTCNeK4^TYAJRw6ue@f#DiO7-OAbp{YsP73`=#d
z_eLV=$@*k1qz4Ut*T+g&c1o^aBFNsWJ5@J*-+Gx_$m>h@joaRNUSgPAfT4uq*5tV^
zyOr;CGy>^Gud}S?;~ce8WA|~R^kOY%bxC4(te%Y5L!{WKL0@lI(!q(Equk4hDSKX$
z{^?*dHFhcDh282MMHXSsP*w7)3~O$9GW(*5-SmJO)}h{87^k+_Y@-gfw(duP@V?0#
zq{Y$#Z|P9SOYr!;MuD0Ob22%PaqnD4t*5&P{}l9d5Z<vFr{L^gGNWR<_A>#M|FBF)
z0gahgyuMl==7fX2qFpQ+=1S2Zdr(L;OGeZOyX|}0|FBpt{x2FNJ>qD5vMnr_aFoGz
z?S%(_)!W_W3j1eVUXU+_t|SP*>I0YOdB96^t^}<DSqC22&1#i{vVJJ<n5NlDI}JwB
zd!o8MXxj0M*0U^Z+UsiTQ&iW&NjoG4WgS{`B3Z`LCJPfx7s8`}LGStkfUvthZiY3f
zH=LCk_VWWHGM(ROspz)<88%k%BB7jCX%Y|tM>?Z`TEw5u;-!Cy?__w-c!Ml>a!t3P
z^yP;%V)e%KNhdU6R>?o|7pBvlb0gV(*Sw$#G5!RBNvriL%!ePRjqF*g66(sotCGED
z$D(aq%SC%W{1aYux}B)bacJp0`(hAKJ>T7*@@CsnTr7{Eba0ucZJJpF6YQCIY#Te&
z0yq)ul{SnV(I&h;Cm&lxEtoM|&}Mp=R((#@TTz01CNu=YlOst)^J}@{W+T~yWniqe
zo=^p-i#R4+%A&Oizx3nTDWB*N&Vo&zH&CS8tKXa_`jjs-)|1-8B6tbb!=v<ElIKNH
zIm~7@8siqHNk&@hKVqh1$I!FwS91&ah4JL(gRz$`m{ecr&t+4_<`+3PL^iOhhccEp
z^8v}|4vfj@2<@qVSHk%)1yYon9nzcJmJ}5iR#gkvzT245UujlrpHhqLJnNv=KQ==p
z!>VH-i7*cbJ9>dQyfmU4Mm|sdBN~ZkFTWOaB0(C<fF8h5>xc0t*jH+uJJyIL(3|zf
z>EifjKmWL=v&su!Bbr!A>IL@8gf?T;YBi-)KfW-Ecv%+PxF;_&-K0KZAvpT#&d0J|
zzAF$s+s$uhGGfoKt+4x9=1-n0)g?le*^b{*k7SwZx)g?sUqD-&4!ZiI>Bq$Vaa-1|
zrq<K?touMFF`K7~Bt=|VlcSb*VIk7i#!y<M*W8DcL#V&i6>Z--wx?DIc~Ow@-EA7>
zmz!cg^j`Ss#+^}CbQ8n~wxeUFX~yx6Xhg^Pwkr2QL}QJ|C!^=;j*2xfep*uAGtSkn
zI~d<T+guVef|P$pMH?-%3!E}z-yHFHeXcOT?b{b{0myG}_s01`*C_4V!O>dI*sP_u
z2UeO-DsmdgO;{fm)a+2cK;Y3m#GiIQmFI?uC8!V?t7K?KCT#w2do4?KzFY%(`&rMs
zO`HBVx4<x*GBY{Vy=8(;j~ahbXlfs5>E)Y>btv1)!zp%go1x_RsB|dJvC^JetJ=4|
zh=zV0*#Y<W22&P}&HUaW8IINBZu@(YYTiRL^a-OYF-k8Mn>`R7g|c{?n6g=rgB1v_
zCg|5SZL|~Xz>EOEJLrCL0`6~bim~m9L~(+UB6g^4aZ__u*}bx0BC=~5xumqMMmA|&
zbVA;fDbU}Cj4m!`L_~@TV>meb^WwjmME&)k5=@sQq0<-;u^Wdil->uF%ihX65Gk6-
zTL#%|<HjPK2(}WZCrFT%CK-qR&;_m`mCpVOf7eC5S=3}MK7JyWZO8>2q0iTs62Y}5
z9b5BP#i4{*D%M})E|{kqGTUjxnCwua8Zo^Ys9of81DqLcI5+mCLnq#-#%x>i{LJ&C
z{Z;@Nw;OP-#cgnY%ZBE$aTq0eX^Q80g>`ms_=3lUQz5x>7-?sMy|{6;I$jCC&XDV5
z_}jW%YsK<&#qcLNwp^`na=fmQ)tc~hn*!2Dm-dW+>9TmZq+jp}EC5(XVjDQnmr9l`
zqF0>r((!OBwJ!0w8q3f^S<I`t762Yih{Pj}=hkgJfT@vM{AZg<ckEHrmX{n*=r1+0
ztyt(r7qQZQi?1B0_GpOyJ7ke`-icJ7bc5Q>Lz{ST`S7pH5H%1%hRXvTsJ)t_#}JV;
zeg_rRvx}1clJu7Y?A91zkz#9pB-$*{n;k|~s`p{4j%8e%w?7%Qy~&(bxu@y-eb3&G
z$>OVjR9ACnQ>y%5jF`;nh_U+*i%bbY60IlDe+yi$1(Uv-$~=z0VhA7%j(*iD_93e*
zH+yW@dcK>y=W6uZ$Su(rSN<|v%R*(=dy}qEbn|al^|MH~%uM=`QNF|CQ;KLhCsG2p
zF=eppSm)J`K~8Gf=K-R6Zr+l6aA#n1+TbSaS+;^K+)#evVpE5pO={Y(csfJ!2&yZ*
zX(P_ZNeC7@Xqx}v<4m|h7fj?iCRRK#^Da;(ing-V713ZBKUpIqom0O)LG*RwKdhe*
znmiRhC#Dx3gp3}b8Cr>sv*TJ|nv5`k5QY#J;f(ij*<w*uh)g5B*^PiuViBFeZk`}d
z!S}=e3^i5sqq<}ohZ(+qQ$Ol^bLuHCm@IbMi3=F{yRT<1ZGFZgDP3UKH!^B6S}|%8
zq3g+95Es?iZ<0~s>9QNd9Z$t}*cuGlM+h)CL)2)oq03}#kZsrthsYho2gUy@+l*0%
zv#Y6wm9f6<n|Mn;H?JY&?c>G}q(fvJwKO_D^2s4a#N^!0)m_dd-NH9L$&voHe3$n~
z`-xlVz8?4QME{pd#;WKsDvt!z$WVr1#0Q$S9(Z+COnjxXLO2VY6f%>{H#Q4l-J*zj
z;y+Mz@OAE}Pj#)Va&VA}&Gvs~*dNxa+1wM&6w5?dd4FVgx|w{D6W_5lve<B>R#FcD
zk^yZ>g9IH6Wuok;Yl|h%DRDi_dPv!H->eq}P#O7V0w^~$mAuF(Xg*WEaG$*torpI3
zR;qs;&K1q3%k0Nyh}{S<64_Ln-=>7^RsVH3$Fw1MLHpjuy;P73{R@Z<^iCP7+jkhz
z`Fi*`Q$&`^O~JG=;8{YGyz--CB391<HQL&2+D}Mm667Pu=z|SNtyNwgOy}|giy?@!
zo`o%c*hpsL)2&A>2L|6UJ5j0A<vDd@(`$kQ($8ue9b{ETW){>`KURs+n4aJwgf7GG
ztb5^aI}xENlKV9}{JKzvb{x15n>EezC=pg^kB<JEsakpRInYQx(F}pI2C<O(>d>*J
zOJ_D$gw8Z@YRKvwms)mz0zJqUPZfhDdYqdSxKAD!&Uh3Bwm->Pg*DbQt8Z&zZTdek
zMK^UVKMg&WKP!h1r({@`(2lYu#Q%2tE4uI9yTSTUX-?r~TYe?x<z8iPB=}GcYP5cH
zu%N&yV|$7O>oiL_rxpyj5Dlw{NY}Vzb;~}f2%=?>M$MKTaZDAVF81Mwt16rn$xoFs
z7A3#Y;U&Ch+>0ziMwvA^{3<6E#hQk_7(6<EdT!aP^_)VGtlTBx#}y(6U&Y-vKD2sp
zlRgh~<w2cbA9sIPtjwY20!uLW!Mez&gv5eeuqKUkqQPf=zPJ7u^i>^XN@t0Hz=_xq
z$3-bC)!#nYQJttAM{SEXb6wo>BL&#g*Bj(*f4VOixBoin1#1Oxd|1nxKjW-#NW1Yc
zAZt!XocyTQ{SJQJohrUtvgbrGFvYA-YbrSZgjEJQ`5Ht$$Rol%qcd&NdwM$~FO<4P
zDyLuK=w_Y+b|a+ue5Yd%@@T+X{KoUKS_%%P#|a5^2`E;QakhM`?qoB#Ry@5CzM$2R
z$Z!0zo@uRS>ra3D;(4^{pqstjUn_D_eA?Svzbix*{|MfMv1->*Kf+6*h?ZVJDKEb;
z#VDR2|Ekv{dSX4k(}DyvhIZZ}+x_yzHb!$lbFCP9q|IrsBOEC7vB@5xv0am#uB?nJ
zTNZX$<0C{7ukg$$6_}X^+tY_8RBmOnO=M`Nr{m~01sF*L>46=bclsPg9OA68d;oh=
zZ~7zFG0BNGHG7t_;WrAW+Z8Zm;SBBKw0$7}n^o4nmdRt85aXEh|36}9x?n?I{hz?^
zpDNFhJ!GeaoOU(~Q1A3t(^LZt5Bno2+aJ8yP<%ohuah0dQ?+Qt6Gw^cSdxBi^z~1i
z0>6sNuLWMd)5_j53?iHPCHiabvRR&AP-))r*A`v;%q=>wKjM;z)fzRedQ#YI+RGGJ
z@@Kussk#wjRUSVvyF1l9jPoa4@P+f|r;h-49(<sYoP}y7Ne<W+d%hWDk}hs*Vb%1?
zfpM558|(TSjy|xhd#%R;`&&;Sg{~=M$4Spj+~(t6-zE*0t=3smW@>y+?!5U7askVn
zUOKF7ae3W<6c><fGK6<;?+_LR?XAZ;i2TwZdDZ-A+pYp~i;rC$eS?qQGwheKX>xK9
zkb4tx68@Nc4!eU4zr2O_z$lxP3mhgmsgXcuTQe<IoXX_FX@U1#35^?3%ReLwA$#fr
z`+QCbc$7jQAaVThVyC}XMc!sr<+_2hT7t~gE-N_;RyQ&L{i>uC!-r?Dk;Vckhm8@W
znU*|Mr~7p;VM<N)eZg&9hzk4dM{_5P7_KxvQ(hgFCZ)I3=6sYxU_QxHZ$f>)XvQ!Z
zZDDdB6a5nS+rKRDE&hkq(aX9IGWhA?=8|(Arp;nPIm%!3>5%A6&du%r9qe-BzSU)v
z#SDGLuCm0ErQE2PX+2^FHBl&2p-$l4IE^=~4am@`QnrDRQd2GRWI#FY6A!F>Sp+WB
zW?VllX9w~a=}j;M%0@ZFS4LCS;j>48)MW-8?6wCb?LE@|lviB(X647n^F{J-s4|9`
zPrYc{28XNmW07~tCGUT}fK+UZH8~4!MMuLOQ}!oo1pH=q{7FE2z3&XqWwr~`l_Hm#
zWhQ%D2O!c|0XB`92+~e25lzKT;t9yVs3Mu(2Bf|5fmhdqdQ4<{jEcF#(vP>1g=}vY
z&dGPpgM`|=$lFrl1Z!UWhgElzy^c}FhDLv(KGC&}<+X)E$`x%mkkDO7FlaufS&c!-
zC75h1;O#V$(*$hH3N-Pth>dwG+m?)4&?fwx8ak3Fsh=I*3gP(=%Qh7@ng5WecF7*I
z#WKEy#m332P3b_bD^e2<+{5*sn~GL0p=AJQ+vBYCB>2uvJ9Sl@L{(DGKa-wvePr-A
z0vLCy59^mDOE!QDXs^~Lt5x%oVt^H(MZdwj+_G<;<>cBPdHVJUPjOyLuUh4=I7?&P
z=#x}#`4*o3vZ>Pf*&T51O%U+QvCiPx;x@YYQ@66ge^|d*upe7uyBhh-Q3*%ei8TW>
zO(y}VIw@0W3NY}lWo6ej0rw9I{U?}2o42|sWykrJ<ztpB1w;KCyVx3W$zEo_Aj^+J
zj%w08@Hwb-?}@~y3d`!v#!RZwx~)MFg%Vpt7<MA4vg=PTeO1LVZ3@FyWxVR1Q{Jnq
zU8)1gf2c5|_s3yaZ*5_<n4IDI;tb#mK2%$9GaK0Fu%1sYWu;5=oDPdya`WRCKdKd`
z!Sa8}f7Z)U702^HYJL<Z(gr`TDh7)Qh<WEJF<Nl2em6EtiN0}Wer4bdYRVub8lHa`
zk1t6Y7k1OgUHDmky`ED3IGDdUcKR$k@ItIRP70khV!_8A=lWE4mY7WM^{MDBvEp_v
zCF~jXa0p0<<0X)?POPMn8)A@a)2P|YO0il`uf!bL7~H~zjSM=;?~sxDhwhJ{*Sy`n
z0fXGW4h+FfAu<Xqb%hhzJ3L&6&+>e*JJy=ZJyceeV)1frvVZp9R}*S_LR-q10~Z-k
z-JjN_>>aI)_d;>RzvO^wx3G_M)E7u1ujKY;CyUtm{G-JtwN^tde<Q#WW;(1v9F1it
zF9*rC;@G8Fu2{t09bM;Z?vfn;&cH8CmXvNKUa-*g`qhl9bZ6P0NcTIPxKCR0yLpdZ
z)2l}XP~nGlKHs2674#Q|#z({UHBB|W7gRlEbRUdU8|XX?@*Aniolpv174!z57W_^N
z-ihthLGWM8!{B^vTmR&%KQt#W|84bI&0jRphUv6eZm3z44X=?|zA<JQ$qRt)u+>Lf
zerce~^c)=Z7j!D19#(R1@qV#|Jnh&&)N^2Zt|KUMV6VBfxUqdbDZI$NnEj36G5J{?
z+zBk7W(diBPQS;gx352y54ES_&!Ah+1r+6;!fflh&FKygvzjw3UiX-wX`K#i=TmSO
zxa)iYnb$0-9WhlejxX@muAes9jp=IFI%zIZ82~N!xL$LE^;zZq!!nBMB1~)?STV}k
zmzti@7m>+%D-CqI5Q8|u;_8VvqL=IvzEJhqzcNfQTZDVOe~oCS?kxQCx}a(C9d@V*
zRJj?D6!q?FNhyHHizkryS`)=Q%}fXp_hCp5YGOG;dZ-!CV(1;+xBr87R5A3mflXZB
zqz(=$_1<6Q0(NR(>-27`SnR;<gSJIhZ_7D$q@b+O$Jb5vKSab?A~C70Rv90*IkEj@
z%i!0PE92h&g0qmA@X#4Rs3^ZN{hy|PhW35t$WM5DeV*mJ;Q*1$STVBDJ@5yCQ#|-d
z${%bhYe|M$?hA(6EbYit+<dk^RTzfe)^Mu^4*C%i6M5D7oV<J5FO%jr#R}N_q9qj;
zdniHj@Avox1>Ap3PSxbJLa|>Ox+9_Cxx{LCJ<7Oidy^DI@qA-o<NIczq78GvdQ{wd
zg>Gp^SaEvh)Skpb!HM5fm;-PE0XZm%<-INme81MHh1(^TZJ6{A(v#xNt{`Jm>l`Pn
zilMPM_&QHfZM9iOc?bZ^L^;f|?OlB(Tvya%kMtI}OZ_~8{sz!flz%BzL~q_okYHE1
z=Vqqn7&q|k@Qo03^Fw?$8_j@weJWWIOb<taP+WXX#LP3JHhox^<Cy-t-QgiW28%4V
zGA?}ieq!aftTtmZkMXYuC4K~b@ts#eNOGSkKgl{M{Vz<r8_G~rw0BLq2ztU+7rR-o
z621YnX$Jc|8xY`Qt#xl<!NQRzFkvpvjJQ$XEpJh>c}$3%iQWJ7Ua<oblk|_|Fkf*<
zl8vB(Z=Utdx&HItL0|OjmLP%p`f+5;$!-gj!{)z!l`J1-5^puQ5xASeEkRmbi`z)d
z=_jkiUcI)*8uQ_M%j~^M?jzZD6hn_a-_`n?cPuMC!T~e1a(T8se$O;t^jqT1|18<+
zhg}1XmC-U`-Wq;o#-Lqc`+s&x4|(lh&(`Clu()3YT-Hvq)YlPKi5w^8tTCP7)^bU+
z6ui2Y;_*8ysc!&#C5|+?PC8>_x8bgv%q#w~`x}}HDeZNePH>lI#>umB+8wF%`)+N4
zY*;IbQ_QwaD#zx)agZz+YL4tP9j_$3HclWL4vS5U*{Tvarr{9{M?r-fZP>g3a`R&k
z!K>k=JBbUAbUpQG(Z2R+gtJ8^VncllSFal>uih<H5^*TEV9WbD*fO&i^~~kj=j@=`
zztp|eF$P;T!iUc@e2bULU%jtCmorRE*<Eg~@rX}w#d&U5$5Onv74hwrMk#XT+1CWy
zEdYF-BS4nXHXGj~I(I8ebY97REPh*zPjvo!N!mEgMdv!@loH@H-8-b7if$%<$`V@L
zh<KK_&N*f@@7lqT?EGxe8@7)AZI2){(bn^NW=WDbpYg^UQxN4u#Oc9y3;I*u<0<Pv
zEk&~pBeu5j(ofC~=cxajT8Au`?8j+AUcIVsu#$~-)7Rqb*@KleQs&>L+O^%!l-Uhq
zh6i}358WoRV}6eMCj}P&46l$)4I7ReTnL?xFZ-&#FzoCT+xuS=|K}>LE4nM${q(-=
znYsicxi!-AlbT0VJt+&`0wg3f>fE|<;_2|?#nV9Uur*37+o(y6CF*$pf&;yrchJPH
zzhX$lV`uqp8ftd`a?C3hBdpv0!<wx6s>ob3bsY9fw##&x)D<%nDEsVzkt1`N@w*RR
z0&2c>O+~#ed!sI5sK8c#ul$d!Rm};2__UV<nd?mFquzDUW|pH(xQ3gLSa&9|Ha!;N
z{R@uSb;U}~7uk=OBSnMc0UjEETFFK6)I)Bf#-@!Pm{s@o@EMYu+L?u`l6#)zn65(`
zajfcmW0yFiTKY}tOwMFmMAvrCkfH1I3H>y#x#*Zec`wF0)H!^`#^aG!oxLln#6!YR
z22HptIjVJ0c{Y{e{M>1W#;)$0`ueuBroS&0*dvi#-n(z2xYoZv(`FC(d#+i$$gi%x
zrP4Hkgrnu)-NTxc<q4G@hg>bWo5IgOTK3*s-(Fi&QDCi7u`$dcb2^NE;Z(e_?EWTs
z50)A_u|p@;Ow28rM4T!$JU$Hzx#Wz>D_fwh*fjk*i(v{UmahbE)Me`uE8GvE**;)x
z;>F?mt@IIxKwSNMctyVL>-mHW%y1J=n;V?a-EwR3`g8hvkx6wnN7n!W{AEY%C-tk}
z{(a^VO7xu4-pX4u%7!{vD|PEJt<SZloGhAEJ9h$rs84_mKNOpB%EY_@7xnDh!D5rL
z7wInu^S6tvhh18RlC`W|So%#;=?$J5_)n?xsJ&2osn;Jri*}2y17(Rj9Y1^w59U~&
zIOd2pP+`(3i0+m_8$|&mF0xouSMO#b993jrH&)1#-wo9VlA;djScf1kPxt}ij<)O`
zw=>!>lZA&ezfGMf*%mlWeeScw+qJTsF276XGCPYK7X0w<R`tI;D<px;?r(b?K32x4
z2p70uKeH_IpQYHuC3#oICo$^9o^0vkmu$rDy-vR`eQ=GZriRBEqMn~bkF=(h1#!M|
z=OR<{RFgUEH9vdr`hD_CWn9C)5{Ry{+0I~3?26eg?d=ThZ;Ew>+LGLb+s+rNth5c|
z{_FMCrjt^Q_4eldhTqAQ=oJRau44?8-t18sMerbqn2uqALARI^5+hU9;@X^(_;Kx<
z@`^V-tF8MvGh=fbI&%MVHlaIRUzMw8;nx3QF|3tTl*A0I1Vk$7e2@CfTnx1%0Huu<
z1oDgtw3*Dnb57c9T-I3Wxr9vJx$rNGp(Eb>^KAnnM-E(6m1`|c?X*&sRUFmdM8B)t
zRh|VGx$OM$yXJFUA}>KVZd~v#m+)B(3Esc+oYbIC3T$+#C<0A7bw%G$DSUH7BX&vo
zMnygbaWtRtzUL0I*ODhlJJO%6SiLh1I*DzP2j%n(nW(r*rBG_#ZFowuK}2^*mVVka
zk-QkP@%#Ai6r^oo*)ZyF=~33}*z%))+l;RXkI&HmIcW6Ymr3Sf<(+rSq!$7B-KU(v
zsZ1YdE-!^YtDKH;vRuy~Du7nV5ou-vf<5jV4;LJ&FzW-ge-H5oe*HO>{);KTKweD5
zAf0oEg9D++b}lSJIdy{ghx(bD&GJg8;|_&sgG#rJlSJeYGTx-+Tx&`oR5PxPUq`0@
zhgG{^YXu4snBzU=z^8O^d$+3StGsE{gY1t5;89NRols9BU7x!ewbw}PI=XvwvXQx}
zQSSo}i<7=dp*6EongEBKU%wedRl@d)C!Y|Uvc|N1k8(H|D*m};Bi+fF0G8VllyftS
zm!xi7B~NPn7qxXf{_pF5SeJh=kls!(f#giza1fm&dIe+WblGn*AR9@oPtAVs@^hV_
z`4<Bw_?-o66iJq-4-_#{3!gG*sBdbSWyb0=uo})zL+sQn#gav;sTsLSY{%3tgn!>~
ze$YzRzxJtwt=_4o>ZNR<hTrc|9PLqgAyb?x3rPRm&ml{j=|r5W!fGcYjbnsnpy68^
z>ituc?vTs`3Dbg64>8A62(*H8iSv&U-J~J!-DAn0)xeD?{dZ?a{h%BL##2Go2_iqE
zhXce?aui6|$?-p|!iu|^fecS|OfoUSr>Eivg^Q&kyJpjjd-}T|-907eAIaK!Eipr=
zP3P{AJ+N)Qp!&hNnTf6+n;%Dzwi4y&;-X({G8JXi;;vQ7k>QKSr@0Lc&$9c<6VjUI
z@3bip99jywM!X6h+4oTsQR_?|YSFYHU<F^&-q37kh3XXG&E*T%J>{ddB>Jo{T{F{W
zCCASc|A!0daXA_}{G6nkQhQ4y=pgpTBwyy6aEy|%#_UGy$cOj~gr8ZA5l9BjlBgd|
z{TlprC0<vpoP|59rCvcXGNXquP`CTJ0kw1bKbzl}3ive}yNoBQJI}3+8(2ZwIr=z;
zYh`NPMv=1PimZ4Yg<m=VzW}_mj9C$-#isnU0PPh{*pqc=fv5NKyw+S-<%-W*1<Cz;
zP3^G6*>lG-0~qxHK#+=~xc-z{=SEj>`<Z^fM5&U9R_84>652O>_^HJ?3y$t`be%B*
ztkuLY3D&Z?aXVkL%><4VP(-Yy#9LStFa|+IauoJDt3xKM?llDF?&SlR^xQ~yRdkmm
z7S}x&dsp|ZJD>Fw#&+c@$7O}3A$_UW8c#{|^P_;GdIOXkT%Ae$-cbqjL*fMgVSx=i
z78;O2N`b>(^lhZhxmGr@qqeMV;HXo^AG38k{8x)*>CYG%wnO(px|*7Mwt5=L(sZWo
zAI>P!ml`Dr)jHuXM`fE!QbnU)-ZxkXnFW5Jj~Msx=I_o(|8>Z$bq<YD<7|j-xn)q%
zI+!cuVVWA$R9}j~9l?}=4{@XfsM1?Iza1k1H5}N=PB7PH#uyL$gGQHUG{{E_nIib=
zXsHg3n=<%nn;Evh@es5f(2A9mCrBC-uZbuQW|UUWj9?i>YMQHF7M)aGwb}@zpcgZ_
zvNY&CvB*vPBiHJ(&-fI3g<M>l-@f=pa1xvxFl^bUz04BZ9iO~Z>HxfQufkC+louMW
zDAFjlBR78{1f~4WG1pLUe$kqd^0U@Z#e|+cJ1M0S{+%BCV$8#gzNKqzS@Sga#n6f+
z!*l?i!}lV;UH;x-lW->H_j576#Ra&@2J6(N$#{b-tUohN$m~;s#W;Rq6>N6wVpcQ}
z{@JK;eUeecr4*Niq%$IKcxko<R3pNK>uvA|z_2Akl*3pJv`wf!vQQ&YcReFTJHB>M
zm27?RJM*Ha0&|E+-9)$0e$Z}sKfTigEbjs7q__9FHXKdA`6b@SKKQig^sBuSUyGxq
zJsXfTLdFU>-Nu$Ky6#sHJN0$v*qN=d-Y||yOJ$EoYitv=(&urlw$@6;bKr@Js{7$R
ziDMGW2NLg))~4SFYj_8!2#ev-gSmP+(g)cNh9@8tOxtp`=Ic5kEOE+e+(C|LQ`o^f
zmv=(7R-8xYhpMTkL$gdUx!R1-$f;N9l&2$(Qf}sE75%^JFGt-Cq6?u?wAbqMYz)>@
zLXV{88AbO{!oaq7x>t~*lXP>+;2Y?GEW8$}=Ir=vis;>w@VM!aabJ-~rP=>)QgsiU
z`u`$R)-Lw27Puj}zCg*1quE_4Nx3!dk^d}z)$lDq>(OZlZ&ypd#nBA1+ZLc^G@r97
za7(cMkf@%$4KRy#^1A$$<?#1FuqJ;{oLTo8gZy6t`AX6JV|gxg+H6&FxA_a1beH@6
zfP%&!evS_IP-Mj`7$2gyL7ZQ~Zb=fes=S_OBWEHr6=06|GHc{WSM<#GO|NRRtruTn
z0=!P)EHi3KQ_)!cnA>gqX_b2btDzR(QtGSIy5^T9J`ya|8yR22(iP-tYq{^sCXL>Y
zeRtm7V|GF`Doborx{IA1hBB0UXp1HVl_3_deowgNS2PU!EV<`^W~RrOwXfmZmM6lT
z_aD>F)hQUjBs#70#6-4~us?GuDs|#bKjUu=0gEhQI=BkvM`U6p9wR;SzOK*WaM`A;
zqNi7rqV^<@Jc8I8#*-aL^i%#B5tYi_(zj~y3?h$>v+RQohO;>qeF27X=ja*}x;0)$
ze=hkj)!*u~Pl!Es-&s~x+UFiasB@g{ySpcFnQd|DCCir-8`V=mqF3#5!da&}_I)PO
zNtxD)nNVkSRkG^F-MSf`04Ji!`f+?>lG2M3F!qCBjc-q7I8I9sG=5jsaPVG;)C4Fw
z$nT-&pw}svg8q%&M3*`albJCnOd{>4s@KhoNem!PMh?r|MNy^WOe7|7Z>0-Us1;J<
z^<xG4R=5cGILj#~O>L(_0iy$r(Ll8iAF5o&L~<PdXjUb1rV$^3tu?2MXY|LBw=`3V
z8ixuj3#{ZdpIz;*>LWMXQ<{I(xp&t}RcS0LfArUyvCebP=*gmWmZ0Z;p4GDH!~PR#
zuRdQrQKX{JW10me#aUDKr&*KHfFTa9QcgvJ3wlUHF0U$RvQ+fk6ZbqVn5}8)wY%@!
zQ+o?{I>E(gxlXa##=<1|xk8;hUu?P=Gii4QLxTroR*ba5cRQEWgjw}3=YkkyXO~v0
zx^F=vS;uXEx#^^q&L-6ZDdv7F6cg;nthD%Yy-3)Vvg4D&J-n_+laNAoO${nUXdJt>
zFau!kb_$XAYIb*BPKIoby64fF=%z>bn^32SS<M?ZC;o>uQ*z^dCXxT0Hca~dTE_!o
zN!j@e#u$1GFhx)OTRdQmOE)utQ4)3hwR^i)yD?aIDHc{!>5rZE=T6TIsS`i#VKRK9
zx*eXJ4%Ck$`Hc+zqzFwI&>q|r-;SEWdsz27(EE*XqFTc#FG=U<NL=$v<MPvpFuiM(
z0Mf^wf7Q<ej&B|kIcokVP~8`Q5WhJzb$V6z@=lMTHiIXRyYeN3;exmS_$5B&rs2(8
zi#LC%P*+sY`PZIc?p;qs`RW<$1T8Ssx<ErTmb;sLXq(^{bl5GkoQQpCVNsl|o|=BY
ztUh)iM8AO!HH{a`(ePj%T(?h%2~gC1UaGoNGyC-y+p{kjiH2VlLW(M7e-^ee?YKX0
zX}0x1{R58wu)?Pp=?{Abod@-VH9#%0-kAl2Z#a*7HbJ_UyNq_vR)Fryy$DDwQFn5J
z2P%fUBI?L5=B2;8y}(ax%LL05(Z;oUVYYA1aRG)aU`b~%sS4-Q^7m0CTl;!C^?YYm
zgaYT~S849gby<{zREFxnS}(#pM?xiz4LCHo;icRvtN@R0KQo2)B~iXX3@H8LkYnpy
zUxa_O=<Nmyc(ODl!jsm)8VjvRUK})tQlluB5r=);JJmwGnhpwbMlF7_PbGV6OkfoY
z+Ne<1GE&jxK2?_a_k4IZ|M7T)lWpr4Be59|Bn=}ux09)}G{|4j>e!v*FnYxLkI&o-
z1bBH?_YBrM<8hQFbMcLoSul+xu4<~t)C6a#2w*}yP5t6y`<f&{LS7d3JLWVEQ?GPI
zqcq|nMgYInvzbJ(^WmIrEiXwuj!eJed44+@vR?T6&B#6spX48tS|ZzmZE#M*cbAXf
ziQA@KqV8JWXc9mUB$;Qc7rK;XK?k}OHMm!b5hO8o$8c{Bv$xM}LC6Um3>EeQ!Ek*2
zjGRHJgq-1QbDl)cF9N)Ed7rI8zD3li;wt*N>4rB;bBSk-WPYk_+Cl(zVHIHsc))kO
zNQp9OxMB31s51z(6K&tilj%{SIatHqp5a2bh+}n03EH?tXM6VoiE7qczu=CF7Wy}M
zw&o|CacQ+T_4{kS-kg3c1#iRD&u_DQbHXa;uU~_YYC#icVIGt9Dcb;nFnaSyFkRhg
z<JkJKS*D7kL)-TzE(Opwy+Z}Gdw9jBaS14YD2ey?$A=Yc2s$Ae1t~Br@yKZW<FpZ7
zWU+!B1F9wtwerXjdHh$Zqa0I{F7c-v&DSm`zg$-38`K#&j`?E}XjqziRmBoaI4gdC
zF@87A8Q~}tyBz%2pix&2{VuUP<;`8h{~gwF!Nb1chxu{+BPaksrJe~V0*K4{L)qel
z*vA1e5zD-1=&xnh0`o&?sp-#qRQhlQzVA^u!s;+~mXu!a(otrJXws9QDXr)=vU=Wr
zSt-L#2i$j2Ga2t<XOT8NDSeVrsbdBwDwL#?h<Y4uHC$FGMfgrGSi{qe^)u<8MAIWU
z4Q+x2Q>^Ysw3cd<cgQc#F#o6mrojq^GB<DO&qv{cy~_QYNRQ7N`vEE>ydkgY62ScX
zy->HLzVaUI>j?G30);WDv@vg+!0RYxCPbYIrp70jOik3DmS0g&5;^+^;$-<+w_2~i
z=jxa>-5|f(%Cgp3!SDGP4Hsp|1)wuBe_fn4KPeH&*Wa;KL^<GKS|bDTpK;1JEsx`s
zb~W_>H&gT@xkl1~Pg7lSNGgtJQmB#EtfSRN;bxksY@WG4J@Md8CBNEB4Wt6qP(k}-
zp~>V84>k2=nQC9*!2H=JtM^$F=~Q-Y^qTaN!S0r{VjGx4`EBR@Ub*dPpcJJO*@}ro
zmL)k}r)1xgvfQ{yRyM0dl!|tdJ6gKbIAt~VU?M$pH7f4c<+1L3RphncfVrkEftDt#
z+O(SZ89t{y@#;6VQKP1CNoU|{;f6=d<$^lIWy{?VGj8--yn^v<j9{>BYDsMRJ?irJ
zjdmS6etqC#&&Ik%U27||yN@vjYZ=vKn6^e=qnCXkq`+qhNwSLb3fGCEXwQB3`#6=$
zG81xk?`WU)5P#+}ADa9y2}lrt*jpg3w`sb_f<_`j@^JT~Pw~-hNP@^W-wq`2`x5q{
zm?&L{*<7rsjkvY*Qkg{c)ok!zoCf=QZ|TB^si3R9zgkn<?JYpNK30NOHO%ygV9Pp3
zbGk)NU#lJ}Wc?=9vhE0ft96OjcTl!Bkk`x?YLWpr=116Kb)9CGjtPx#gXRqz>Z=+d
zCa)D--(q9<!QxCKzS#_cg$0qS=&&b_fjml~6-Sd|A7>UST5X+Xi{R=TS6fP-%S*n{
zZ5MNkEta^4LQeJ8E`&j1AIy5(MRN6&MPiDpN<Jk`8#LDr7g+~T%~Zl>O-kLOHd<`)
zqb=gFgcqkaTll>d@2J~oO!vIo4+bBI{2k2l6$qRpV>GvlIRJ+9%qRVk@1e=dYe3G1
z#f;cR4C~seN2Kw|)5lwt1&i|XIO;EntuBX?B4g{P8AT0hCg_rDj1E7FvtnH<O2kDz
z?X|ZQw%JK`(WvxC<rlRmOKUPQwa4VVlA)IV(8>swvae<0k{lTV+ER*?MiQUURs6`q
z&&HI7>3=IcqPD-cZJa@B8*3L|yNo~n*(z>~k+i(^zsfhE>61!kU)uco5aw4ODfZAf
z<jxabJeWlUVr>x%ffUBn?g00z6}7DZ-*3{Bt^~hgDo#(uEcbYt>3I@NGVnDW!m+uy
z-)^^1W4yZu!Iu;5S<doHfD4TkMU||ZJP4?TpI8tT<QN!3)@o|gveyjlOh0bT!4@@Z
zLbOeFQYMc!QU7vl_s2LdPD*G7F~(hV+~Nkj5_OZK(cD^7RbSf>v&2T5IdJ#2PZsN+
z<t37XNWdWA_TuI&5vM0F$8P0%6wD3_ljPUWO}XNcf7$QiCrDP57E<CI^5C(%0I0W{
zcz{v^4bl(S=cy%!vr!YupRCgP2=<$O&ne)xnNc$}b-*eSp09iTr%w~2mqs%-)qB=0
z**=q}&^>zosT3{CQ{{{SH)U~aacgECc?nohQ{h;fhepkP#N^B$Y(jx)Az?B`a`p!(
zZr96TxBV_dS%|61yz$)gy7-ZdEz5fBBJ-9S;}DfZJLL|EJ%55zOD%peWf3?)x;=Hn
zJuz8m*_llgPxZb7slMe6)eLR@LN$A&x8+3tI%19V4gG#eG>syqvSvV9@G5KUt44>9
zll9dfe~+GvpTXFn>`vuxC3Zm~BR+|e2<7FzGh^d|=$}+W+2D*|eB;GMmVVYgpQiyc
z8E;@>W0)-ZD(k)zkZiYblJzxJ8Tk#s@E4Vw?xK;%c1*vcwU)__4s>#;gYxAox5bL5
z1;3}*rujWwUQ=&Yb`*_FZpGai6b+MCRG$Pk#)>2ys7RLmiwHTK=QtBa#Gt+-Z(Cwg
z75T*^2NA@n&nA{dq)KnydD0qy2?~vMz0*ckW|tC}SW?ey>S^&%=aNla#xT!>KAU>w
zPpwb-FpC1zwT<@_`q?8L5uK50fJVIGs_XT+a?n9x6gv2+V9C82T}pm@e`!-T-k4m{
z9e#ApqkhiPh=gQ?3AV0cp$VN?iHFVXkq7DJ?jgu+;Ww_O6OlP&XLR(eKq{K&oSYAu
zZDp~&T;3Pr@5<$_nDBLeU^{x7PmFU6Cy)!-`Py%P9#c-HrS4<WN&@^r7_ef6**-}i
zS-@U2tLGBt_!T0=mliz;;=9hHs=sxeKXkC}_XMcM`Rd>l@se*lli~uwbyE5=brq(F
zBGbi%CKaOYep<R@{XU2Oc@XUsv*i>U+P!q0?T&tWo#Mz|LHxcZ`b68+e$3;XJOPYK
zY@1J4=Mo#R$_Qz!!hH%i-Sc=hTP2381&(RsoZ}N03!`K&6EWMgu(c;XaVPdG<-L?F
zK#>-ur{Rp@ANfPiC<>39B#N9!^lGyTF%j!2rZ$MKcU{8=4i-lilP0hEg4dV0jl-2?
zmAqx`3&Q1ltzR*|VMq>L^J4$<C9~=`_DobcM}mr`CgedPI3MZTfl=Ioa|{8CZ{RO=
zKj2L2zcrY~abnyxcPhocx~7Cbz8)F+M8;akd5ToF|6?yoZmi+@0l(+7O<|4Z*=F0`
z8F&DGT3swn0^KQ^W{sXy#o_rq>B>7zqHlUWThtf~NUkug>PZ>!^Emg=r!1XaVwrJ_
zB4Lp;S;T}n=m(vkdz}<FKyw-bt`h%d1d2>DeueQ$dJq0HR9VGsL|<SLCy)DPZ*NAS
z{3{1x;VV~?zqUI+_J1)F9S(K*A{-jZ#8SUwJH7K&#A&l*G8-|Ovt~TBP{@NC2gIC-
z?wCyGzxwA+96=2j-W$CdRs58(X@0TZ+IjGWVwm(;%X8Mvg!*GPTI9ObVZH~{xC!vE
z>zz%sO)833p<U<iuTaq9U$+ZC@({pUAx_r80pv3$^ySiB!-Je>uR=z#9POK|Qm;7d
z3TZq0rXB&#LzCAU4I+U0jI;Gk8AG;fVGpfS>&l8MvE!;BjI9V=mYOXnRB`Y3UE%RR
zEDS0K&y2O)(*M=>;rz|r%l}(h{X?O5A6dmy8pLj_51W1?j8f)l^}7F-rqZlZ={gh3
z&%^itRoa42{=<5TRI`x5L}{OkenG|R6U(m{0{Rej%y;D*wg}Q5`%1E|{YD62x8=&H
zs>gWs+7LDELwNf(xC2)HuJ*u5hGtV!P%859f8Ppc0ULSyJv~taB<-E=TG&rq`AnN}
z36=IQRGcBqfNPUqC|jqbsM@T#8wX_D)hCGp#67qpn_&H(Iw@;Idz&MW)4sA-yoA{p
z=;Gpc&C@q(Z_;mnc1j{gDAq@LFg4aAZN{K+EB;t7#;e@u7kEIv_E`VvCeMdiy^mbu
zfPMJjS*C}kbYfj4#T!<4g=a6H1Es1{7(O*RqRWcjw3%CZJb=nuw`%BIn34vYd0PGc
zq13iw%;LHVnIzdR0m&{)F!9OlydS@|G~+%r;-nW5I>;*l&OPPWyo?@cX)72nYOX<J
zRjH1UcFd;OBPFLRX1hD#d3`J=E-vL{<UB1lZzd&ZC)8Vln{DfO5nlSYwJi)Vu%<x$
znJ1}c#>1APYDi{-&F_n!g*H|3X6d@7aW$(M+Usp0s0+0G{{RU=_P!s#Cf+x9{*{rV
zYqyuWt+l+2ra2@E*d3_5xYT3Ip|N{+beiN=u*kkdm~dEhPW4{m^5FQZ>@bjRK@3hg
z+l-T%tCMB;0$$ACBZ|;I>sf$wIKkRRMQ3UH)ztd>+R289$iRRwNE!bC3N9R~-yqaH
zOK?~$7N*-BL!H^EVDT;cDFu@w{hEg+X&Oq}nigIoypw{<p4?XMi>c3XcFQ3fcI04I
ze9rbXbmF6Px6yBe``{<$Pvu(Gq+QVsSSYFx$vRa@DHOIVy+&yajP;^`H!OeS&OfD3
zRA#_|NI<40LgFLCGW`jv7qHo)63nYrdl<By$0ez<nMjP~K&-bM8u76$I*pywm5|K*
zwU4Xo5!l96aMD&3Df7)EX*(;NZROp?^O+`)<I=LOC1hfwrF)7hD#+oz**%z9<8}T%
z)mB`iklCsM(M}f}ky;kl(p!-vPI3=pQ7EY^F{aZx-510wbBO05Ti4dQSllv#4;ABA
zrO6|*66QtQCd1bBR}#Y#`7O1H>OCu`3biq<q9$3dCi~MlsZz>kd^5KrJ!{q2;BsuT
znofT4%}$zBVvB-{W6g|7)pJtE3Xm`{nrVr{R7zl6Vz7K+4YYb3^G+P3OStT88cIdV
zowY_4uBRR4l3cVTB=IlHyOBam2;H(Gyz!RJPqR~ZX7O$dLRlqQoV%kFz#la+Utnbe
z?+&%nt#cNgt#Hby%ba4Mmg>}_2+mxBIW*GjCv%0qLj~huxZTYRyO(TfjC1v^mF-~a
zQjtJnRBh*<dWUNlSvqHm<EpJDXw`*O+j|Z=7hmUIyP-z+Hj5Ir2zrEPEz!0#mLuk4
zLD9xNXtE!?RJqgqq8A>On|_OJ3pwKgqIOYial3Xnxw2WAy0IkGDZA~lcgu`diA~;T
zuO(;_c*{T87|F+&GBfqA9i(W}YC9-lf&uGZ+dQ~zYUtOOF-BrrhY{p_fva}HNf8NG
zb_I#+U2ymjZYtL{W4pRgpu|FT>snFCxD_%D;dd0zIHx*y*yVCj+<bRNKoMJTJOP@O
z^BRC;2cCkk<dYPo6|l1IEAvVVV>@dm=K3X5^JMx8mrhqhgqK4q^5LakBRfM5YYWCP
zZ_|=Yba9hcDDpwy`qb-&rcZ?8FR4O%E0bQK8~uY%hFIfQd5PRyXOBw89ZV~3k)L6u
zTj{r&o$PA3MM+XN94Tt9lH2OexAH<Goun}w50&xGe_H6n=~XXM=}Gf*e_I;Xk$I3r
z(=OK>umc=bgBnSb(>({JY~;-oxf{e3ZU=$INo@g)iEuWzOjeS9?U~OxYAxD35t2C-
zLR%RYC)yR8srAKaPBw*sOQ}>?O&~5n^G_gVt;YnvyA9F$Q7Gt=+};gy5Z?fb%1A#-
z^L=I$CDUvK^Ilz7WUh2n)gsr2p=g66E)IFkcMuHdU><%$STBg>p2s)gl2=yx*M2_e
zu32tu6vI%!X{0qEbN&ne0Fib-_uKi_#M3NlcW3Yw^f$B+qONehB(__t*yP=X2Q_Zs
zFWDf+8&<sPn(A#8V=im7fDzTJCgyo3bJrD-ZIB=+&7Vru@Ui~@iOK%}bk?xk=Be<V
zABwD`)nwAmyvPl(#E9|*+A=xk{P9^D-ISW{uXT4}G)@u3XLNx{!28_xt(k-!k}G(s
zSmruqin7af1R*ArFd&~y`&RCoG<NqfX|~d@?rmESkT*&|@A`ff4o)uUw67Z)H@5)A
zZw%5a+-&{f2R!!Xty@fBwuOPo5*&|iE2f=JcG~Dug`$yl**?*1@i)qO!R^gfGTp88
zQr#3P7=_wL0M|rhm!S$5-*RQO%+4HvxDuos<o2tVgh8bFY{Pc-<M6LPGpTx!USH;V
zl%MXB^F`Ym_@cJ3L$SAH5_+$_PpaSA>DGwX8|2+4enaV8(URo!IHw7FMQ*qJ*~(u@
z4bAQOwV4=QTW&zi2PBR^O65##XLu6hZsX6&!M~MEqjhw3;pb^3FC(n*?Yx(5xdszF
zwRUKcTF2OWpIY>={{X^B=gcR~S)@k=$+(z+FrZdmyD1UO%hgi0dM0v0Hl2?$xVX7X
zMQCJnXJScHiqi1q-lp@h^NS)5cMS7hCp(JJ@Ah$%)a#R1k_i6*dm8XQX3b>D<7*kF
zc(U<Eon$=?X-a(0Q!4bi7KP0sOD_>I^4TSUPtCY>{Ohi@(yXoHRfZHmKe{nhSC&r1
zQFEv77S_o-98_{yH*6Z?mLE@a?S?B~c){1aI=Y3k10Ek79CfcG)vRZ*X>H^~`=gu@
z&MN3#K2|#JGPI&IK)F@A8#u`3yIX7DF6?uV%YrFaUh!04xNT{1d1*5aj@%!@x%82~
z+*hGl4LN)WihsnBfvfD6_(~2KIZ`T%NK8dLqh(K}cGB4x&6}D<{{Y%HE082!4;@8x
z@g4<xuB&UCx;1~ijmruNr;-x5=J}XH79x9<k3GrgDjn?q05c!LzJC{6+EmkIdQgul
zD_k&OMRPZ{s5MlUP7$~@6d$}}Us#>Dh9Zh9*D+wB3(~82s@z=Oh)3V$*WU)a)BV}y
zTm7KS(<3H*t@nW4^s6(+nT~(ET=f~<6FOPsg6<_(AxXwFTB27g_l;xi>OomEq7aO?
zA$E}PGH4KSmaH3X-I}ylyLCh=yqdy6i~j(bvdz#|kU8qT4^`6ccWKs=`j8E5+HoX<
z*wn2Vli0arf6<4rtB=a5uEqQ`a6jU*ldn1VtbY*4y3#kt-E33Cd}!#F#x;rGZV0K(
zeN;9_3lP%Pk`z+`w3N^luZQ|G{VUD&AVvETTEF4^%%$-n4rvW?$g!kS8qARnD#Uq)
z3?8g<ewE|a_}SSVk(Mqz{b~+=RgeMs*P-c0{SAJmu&J;|a`QMvEXM#0n&phf{{T{H
zSe=S;27fA;6wa2yFDl#<_n(tib6GcbS~>p!6J2g-tkCj)X2<DM=$77Ew5QW(t9nG9
zrmHpwJq2->w;yZNrI_-mz@h_YB=5dfZ+eo{nI%Zab6pQ(E~lYs(FiP~PKO-UE~3Ws
zSy+c5NEOW;?rP1erw~}V>9uQ%8F-|8^3)=X_{Z;ll(QYu3fksIgeU}JrCbW(lho^5
z74Ib^s*e3?G(yLAc{S4wjASw6<c{6V0R}P+Xow?^1&ecmQz-#*dYS<1ye7&t=^1*L
zAl2D5`zw2vx1G#vy-N-&WYR|*la!r}2%g6NAm6)@PfE?d@U@NB=3U57_j(V)lSzbS
zD47SrlSew3CRZQr3|5Vuh24z5yorL5f=5c@#6_Nl9;{yD-Xpx^8k*ibHfzt3=p#iz
zz){rHg-nmn-5n}c#&2>hP%K(|Z5;7XLiveLsH<Buc&pQ7qNG+MrDkADYhm__w3Zgy
z>z+FvE0*z1^~Juc<(#PqGx?hIX+=tr!#Oou<YNB-Y0iI&t=eD7E>X_mis{3p3W(83
zvr9^xL8wQ)K-_Q%_pXxR8Po;9Ij>KdsUr%OcN&&S1DPC#W8W1VvZA@#c|Mi5v7^t=
zV--@Y*Dc5-kIi3CaZR^EgyiP9@%2A<C*1UK>y;8o_f1#Q<R99leNfkhL*qqP{u>#t
zwr)cARmiNU+hxc5!!_G2&RTy0J1sXarN^~`k_gE-9cyCmQEOc)IGROyNgWPrOW9oJ
zX-Zb}Gj*R3taic~Jine6Vi<5wVO(4{(IB4sMo?k|#~-U6pO;#tE^SKW<xL>BneH0a
z`|W5g;enP^Eza(9$G$~o=o>9{J6WY65o864>?)-8CM!m1Nv+$imvJLWo4AWb4tdJ4
z>-i7Xn;xyF!+U>qDHCpnIOA1p6O{qKl~QClGR!)rojt~tbRBKX1jNy^XKLevn$5O?
zI}Z@Ubl29!8b%?J^S2=5G@C+6tD>f&Zk|Mv>rjgEjzB3U;-a(Te=d9;hM}87Z3}kN
zn5W0SMRooje9t2e%bJh0T)qZw>AH{Eb<CE+u>jY8tQkQSRNk!<HKJnCTA2)!^s16n
zl#0($lR(B&PzH>5QbTNtiM!?En3uQGVEH?F=C_JgJnFS(q|{Mse9|t`tTH;*ak11S
zWz38Q(htq~QdU6yq==J4x-y57Y~b<<&S{#JlSKwgZX>;CQTwK}qT=;A`Osh*&wdFN
z<5gOoj11jE^=$2$LV4|69nA7x+%vOvHP?cMjzm_MLcPSOid0rqgq9?l^bT}c(%TnE
z1z6QBiZN_;H5b&Y8JRi8YIH-!2^B&uK{E#KtShU~JVHYYdy;Y3`d2l2eDO4=E>?CH
z(&d2%BNftV&XePF0ggGYOe3`H%dW2-4w4IKqfkpP)}%69B%#czILBJ`9FADRz;z2{
z{yhCEH`SpX6_4pk<?J!_NXlhE2mPZ=Ti0lVF3@qFDWfULXfRz*C<o0WCxMFR{7$pa
zq+KJH!BD3rsxZ|Dw7ap)$8fe5@miTwv<RxpgY@fHp}vLXF{_5hanRRf;F2$&TN=t%
zH$)tQ26OLJZzneQ0yX3o!8JRzb2L|yE8Bx|w+yKUKX!4`(-iw#J5RH^6fFBrU%hOI
zwudjRTKN)UFn0AM^Hjqzg+M=EX*={bQ7T;ID<~ni4t=T>W_Xv_;<;=5MM$36cTW*L
zoCEpSuXsCe?}6lu5m=)q@guCJOCXJMMtP-UXQ7j-UbJn*eFk~<u0vM6xV!;2U;~~j
zixE;*<wmrixw(u9ayKgs9QUXq=JG5b;~&nv`dUpL()by<t=q>hhp)>ju?h;3I#yiP
z@PxL8eLexFn;`PePy!ooudREhjGHpFHM16WEg~+*mR-$`pb?Ck-6B|AF^)aEi9)I{
zMRn6}m00MCPjMr?oX(1|XDl*3vs0zhe`)Py1hF{-s2=s!;*v)!<Edv$3bzVZ<rpMq
zwOeHk3rQIyspmDpI3(;hbnY@)luE^0GZe?SO6Okl$znt!IodPVHP;9w5k@!MY`5NQ
zVm2FG_C2dVTemlwG|Rh`UHJM}J}x}8-^=tp3Nf@BGW;1HZ(|th*!?TettGlh%qwmt
zk9W(sWGSu+Z&q^GTNxJ`#HUcZdpN$*R4h)I$Z#{0P+w0g+S^4ktP+;qAw|GNJP*f%
zT{wAOr`~BsDM^1_4uelqu@+KY+-^BPkgI`HZ{5Rx6?E#*x{%0M%iFXOyQNx;297X3
zQT(e{wT?QE-niFaHAz<7RkMtCqFBaQpgWij^puoMAe+!*MzOM%1h8Hxjka64ARVLS
z_NbJVj)=wB!n}Vm%IB?98hwI>mDDq+B%1QEF49Qtv4;#mO7gD(j{>@DZgf>1k(!u2
zPAaE2;ry<uJ*&T7R98H=HUmv+h9Rb!YCz}w7ykev?0@yqznyK&2Eu)-=zhT{;(SI#
zcd>eAmUD}Eo3k$!#ZuNbi`2(>0g;)xB-Bq9TY=KKZfe$rImXgRtNs)@)wBM|t*P#F
zRrpxGFv-@J8R8Z1^{iFM0|)i1dyx-?5o2_fryH2|;P$NKX&0%ZZFCMY+}5-=te`M*
zaoC#UjY-m#(b$rbvo1lYsX(kb&0mIC<WND$uJ~f9HSg$8v6b{SH1ts3Ykn0^Te=cm
zqQ(i@yyCq4MPF@E$sKWzENoJ=yis!)&+%0LF}otl5T~r4(yEfYS<PGckvMr>*-06#
zZ3DwoT}YQ#c}TO6z1q4dtp!qCx}Je&J-vitB)}kWf~J;tobF@v73op<4k^X+^D^F?
z2G?%+tnEV+3B4E*T0W*x_}S;zcI6^ifaQ4nYoyZbmrc`#;rq!sHS_HCVzs%0;yqLt
zO{6b^PwyCZ6~?o*?aJE0!e<7$sK(LM<*P1f9<AZa81%g|IMG0688y(saQtD9=U#N=
z`I|i&aa@TSao$J($<0aj{=1j*ts`Bn%x8s*_(vtoau5eo+PI$|=@641pa;z$@?3i7
zrF2G0O6cX3B?(!cO{ObDsH{gIte7>>UTT3a5#r~|7-9_-V|f}u+9z}2i|F+YDoh?j
zen%CDc!e(6#~E(boYkU{be-oWW9v{So~j7T{J{FwJb;c57284HCKYsJ5>|!7qYy`>
zb~+xcZz4OmXvZMeqfScPtSHZ3Qv90aQ7PnP9AH&5tIKrzV^XW{TsWG_oNX6&&gd($
z%w0Dz$*EdxtASn_*g`WS0qalH^iBGCm3esiHFVviWi@!SN5R-_<#_;rX}aRh`s@NY
zn2(#Bb|;$j<NP*x5_ZHQJ9`&`=Ha}i=YhpXJQI1D@|*ImStfR=N4blt+s?B<;uqnE
zYSoSzijUnlPfF*MS-m;OT}hS+f9I2aRa*K+wurOF_#T3#LQU*cszocda##Ar?5xRa
zX)aGjC~B6Sr|SOz+A-bF1iMu3V!-sTMiiAe>$&Dst<Ln=^<NCk_ZleO8>=I30E+7E
zi4&;7`IkK_f>CO3L!+5E?p?D5ZOA^Rt*OdaTKEz;x2Yypl`zadDx;}IzSV)~4rrsK
z&WEwi*~TJWfaC&c=Dwzzyivm~#5A?3Aw@77NlgJ>_x?Sb>0WQD{{YAHR<HPfGdXTo
z98w;Y$jU=-C^ec33mW;70oZbCJwoTpg^22S735a_BlS9?@R?e9{{V@E2mYE5=U$hk
zwrv#l$*gMZC=qK{O&zF#Kh{ccG3iw_+kf82AFs^B9^I-Iku>UbSrm-widl{T88wMV
zAo4Ztr6tE;aKF;Jt7ybF$r$PfHKfFf+O^l&bom^8;xH?bbam6@kPc5QYN4?*GC3x-
zbPIDe@|<uDYb}Udoy)(K13y5kSUNhJjYdh&I61C$vo-2l@fF9HW3V1v;<=`9v2FII
zwjK2-?!zBi9Y|$r)$C=|xU+nslucaT%_ekn8=f4qwZ2Jw_!3qQ0NgW*>(|3tN`C9h
zgj{UMO`{)7*JU`aQ_rnZN^Zmu=o&P(?Ii0ALw?+B2Ru}9cz;H3H2HA|BRJj=5=r*0
zIZ>Cin5V;fT1@f%nh_jpytdKOsQ&<EX^o%sSTXbzxlq+fKBqyWX_jzUfgQYRtaHGs
z8m^IJE&E9&q)m`HD@~<hBPY$c#j&!5RHeLYj+hmpZzxa>)g#`gL}=T4m*AEHM(7is
z;;has$xHy5^%a##M%tAK%K8WPZM6RYG?A-&;-iy7wIlaqVXkRXjFrx)SDZ8}9VS={
zI3b2GDlfA<V=RPMB<V^ATq{Oqeyj^xF!ri;r#A7QY9YOkMvky$r8M>HSTmWLT2RH>
zV!6MGRsn9H)DVBHU5)sE`s>ldzulgFU*i0Z6dr1LPToQ0wt5v4R^H)WMQ(Q>$oki3
zZ8(}FlpYvlSEi(=6(mH_CFbUY?dk<&MZH{x9OkLWifcmGCyEFPIIk}-sk9fV-%H^u
z+)K3&HI=7a=Uq|{L~7&1d}QK3hRb^EV*V*t?y^>7qhqU%fc}-+EzWv>0y@hU&xcR@
zjYQBqT4k)wCgAwW`e&sntD`td(rddpKM~$Mx4RTGYyfaGSdz;lsDV+pbt4qvv2uFY
zb*-$b{{S*D<G8Mx(?+noj9bTa{{U-c$OS%J`e!umo3j(VV{~O(Xt2t&+}})MwuIo7
z8Df0^tX)7_LwH4$k~}?(i-T7dnlqL8yp78ZFH5_QD|WwOc`|~aC&&pL0&`gNMp522
z>V&EMDacW6yE3{Gvf`$bvv;XwkKe+-rFI??lgyG&_bGK0z7|bA!A(*@oNYDfUL3yr
zW}KUjT?P$I9jY><uSpuzS3%A)PL)bS)}*9YW6eZp7}_Wqw`a5LQ5MHal+6nN02=;v
zYUU4>GBbN_KacdN+f#{pBe&2~<z%^}R-0Md_aP_iSy#H0_W**SSJJDJ<a6Daw@y|w
zkOgKr<dSR2#l<OE9h6&Y?Be`KdFI?0)cmVhV1-goTJFQgmYOrD?;?!9yd|4}DvO+*
z-Mdz#qFYDXaKBpU^#~SQ2;|%e#tE$Dx+!dPQy^C0^r<0`s08zh(30YXm2=S4k%kTK
zDGKLh;hkH}0trvdI%d0jL(d+S=D2dzIxo!wP8F1Z22bNx64K#{jkw9;y(q^+mCLbh
zG?`D{lOIZ%ba>C+Vrl4wX30OY;8Y`E=}&^)BaKsUPAZWTi7|z8aHERI@inr`rWHlW
zW)-a$@1mluBZHCEwBpw=gt&x-B#*yTNQoyYpDI=4f-zlFO|iQjEx9thQhntrGDona
zxIot{D-0oiwUoJ&L`KaJN@i!2e9osh8Q}U=6?=<zxJe4+N0Ex^gf&6aLmK)?F5zAK
zMl+M1wVKFLaHJZEB;JhQp$m<yNFb4(DhRM6B%ZxRa@Xh**%mCZ_K5qP{{Ys#FF?JP
z+To**8FlaNSR)B;j=+7jhu=}3YK}`=sXp}0k3mL-K4ww5v6!&jO9RxJ%(}No5t)Di
zMrtJ(+T)C!>~VKaW|hhQBbu*m#q<hzJIC{{5)Sn)dhJ>U?}%>Ww9qdV!OW@^NbGBm
zxbZA)ac;g;&ulFO7ZLjLjE|>U_PMn@m_fC2Jqr5Op|`ZzDM(sEP6r^7is`LofWnYR
zxq`7Iw_56iqwLYtpzMneAI=>C80czG-7N2Nv5`BjGm7brM3*ZtsSCS{YlyZUPxpZp
zZf8jC$cH~M$<J!eywskCE-h|UGAEuUZQ1n48LauOB5Q*%Oq}PRTIh8m?5t@{4#Oma
z<Yr(A?Np?X%tu0aWh!fj9X9DU`=il>e|BbgRSX)Tg|HbW{)W3rrJgS`IOSzV>H+Iq
z6yB`itgmB(g6`tVS9@umC*DD2HFC=8NiL&~W4>v#BRh2iwQYryRZ-DS%4wZlnIjzX
zD?zT9j=4Uy-Fv-7QCZxHUwH#0_cd&}RN(QBl~rz~W_hXKQIWUrI6Vbf!EQmp9co_4
z>4GAav5cHm8@qytaCod5ZKJ9SLn&f2zo0!$Wa{^;Y{m$1OMJLJhpli|cd6)YCTFbp
zj4xm@T|S<7Xn_9!uxO{b$5&)?`YPj5<E?k8_XpCjkzi?tvtT75^r`9w2ZF!kT#x>`
zH}kE@22}kk=vTlbdH(>5iwUD;`@rI*D$iqYp2^K{RA#Z<r6S?8QK;Nl`&T)!m2=Cm
z^sC+x{{WD-JMA^mOy#MwCGme6&V6EE>sYL~3-qd%<U`>^QH;z!wQfen@1<tV5lii@
z6L3vlgId7BH4~-GjbT%i-0<DRdXB2GI+hqDE;u#WogLwyclx-Pa4VXvH5gy1y&8?x
zw?j(W&31*{YkN=9?BueNKPh)V4%p_W4QD+{#<r>594@Qj`Si_3H;*k4f6AlrJQ7^n
z+s7J~V<$oMt!b-8q9)X;MUK+EITf&3u0VM)_fJ7m-|G?E0}ZwdjyDd1wQrIpv8U~-
zc_JMmSDJG2F=jnUu9P2?SEGi1yG9tqm0KfD&k1H<9;DVp?l8yeTDN4*3V$0p+bbb=
zsJzZUbrH|h)IK5Bme~w(pWYFH?OqC3c0D;umNz&XnWns(%8Zgp<oec*m8Pw|{j?+I
zmD?@sE3v*<vx+M19+d|G4!~emjlV%&LF~t3>bDfuV@R}%R!DGUKS5ajI8yo(W2!ds
z^%d72cA4kT-mPQDH2^Jjsia@y`PQ=*c@p86u{(}Ub|rLko{QkSlwCSg8U8+gxT=!h
zZh3cd16Ms|WQ{04enlNybpHTnojD2VS(9>bYkV`%Ro0{%OE^(oMw4jPmNHJ<ZvX-9
zTH+<k8xcY`mBf-I%Q+NTxDCyDli2iOxuxf1aZUl4=DE(Bnz~MtHT0k;z#UC>{{Yx<
zt+7MND}$c(agS3Pl{sCGbHnzZ*;<A5q^ASS4Ij9z%c(7+nj|c~YYhDbdQg(&XOUhK
z_R7Ub@9iadS!F^Mu~Ex=XfylNp~@&qoZmw;Rn+Vvh%?O?1J;`liS&aCvAm0?Jc?R1
zpjuISjF-kbEMp6I5kAJNM|}>lCgqt5RjWaBNQo)U^2FAP9V*5XG}1?f86?(S*NC+1
zklQDTV*9I(^r1C3aGg0pTE@SJ^{WXa`!shEl2?EYbuiBvbos`5R#hhzEk_7V$70-W
zzc9~3QlZKmbj?eWX;`>INxO`5^AT09SIo9$_N7hx$3xiWks3Y3jAZ0eSJcsWf;gwR
zhJjlUEe3!aO{zdccmDt$?ewoV)(`wl8;a5Y01M`4J<(gWA;>kzS`*fg)JJiWjHS3g
z-ZhJTFPSdTcopPS{v$X0BQ*a2`3PJ80M|kM>(Ml+SM0$7`=r)2HV70iCi_mT2+qiR
zgVP=BPfK5wr*5uz`czBdG+UZMt3O-15LhyUhLeiQJC)GO(XIE+fQ}dEZ|PkjHFMlZ
z=X^zR=1U#2o(C;e?qS*C*<{BoMn08WBqnnzzau|kT`z_sRFE8e#dDg{e+d!X+1746
zs|HYTUy6gn@ocqe2*G1m{gY0mEp}5q{8s960p_b~7DC%=DBzB2mX;Ga>1Hv>#YqIM
zD}2?fS`fD_bfv7Bq11RoYC}7B-YZ%}NjzScsi{UGCnh5Q05<Kst}A%#jybzcoc)%o
zWpORtrSn`E&C@(`MBj8{6zQ%T!}Hx~w~Zv$@@;isaNfp?l%nrq)~5GYwgDlPZDf~g
z%4U3RPKden;Gf31o6&uzT)nK*ICMBTY=iWrD~_GEJu6GOyiW?)>`+ml1)Pp^lTmo8
zZB`37n%QCdGh2@_o9|?f)uV5e=8BI#rgozf-RKq{XDla3&^|lv=zmI^RgJCm>l;X!
zgQFB?&n$2W{&j<Uol&*Y^){kJX)T0BkJ*ovCm(p@t!rwFnc`Id3<}DkyjbAcoPK2Y
zetc+ls;qb%j<s4lnX`=N>sZasGqJTe!gRy?NIb<EK2`22b-0Dc-UI7Lwv3&bGNCB!
zSG}{hxrvpLPDfE$u;}r5AV%eId-GPQ#meVS9}zevnFY<8+hcYJJuy|o5Npq=9$7SW
zsVOU?Pf2{;EkWapR=v@k5M`qw#dYD@Q2CrySMPpDnl77mf8~o*$)(+c{;w6YoJOL;
z;?{UE)lNHBmxs0g04m^E-2A*@7@nrLp<3E>9O8A7Tq>qX>shUL8Ib(LgI!LoG`X8I
zg*jf<Cyr&tYB}zeMoF&<tw&OBOz%p{#x3plm<p+?+9po3lm7rb+PZM@vqwH5{_T_M
z*UKKFmeU+7FUoxl8+XK1@_ScI)~7uecq6LO5Zq}I!aBGk^{dYW^ZmLPgal9wP1}!8
z!!+cZiH9;%cmB0G9bV?+PKMe&G!c1z#Iu3(aBu)Wg=lzw%TfN*xwyX@F524xoOHn8
zniJ(o4XD0m_bcAZcNNU5JdsNB#u1r$;;!h)V7qoZ3-(eY+&)m<tDQ^V_l}uH)cxP-
z3HED6@X-5IEg@@zl|zC}b2|2cZw<n|t1yN{3<2biYU-s9WOGh!HoZyoeLmjECH=~U
zNWayN@;7hFs@z2s@-f(|3ld1`b6w7yZEGCSoSd#^-q~z*cmZHK)E!glTy-JJ{EoQ0
zp%m~|x}O8ho2iiYPwQFr3H&UVRFlfr<j)G@itKzFsO+E0X5oco-KLqeRq5F7!)NGe
z)z)$m$=<Ur=U9)-YFTK6IB~chDG8ti%_e9Wr3J8iXxlB5&t59W*-(EDe+tz-4mm}(
z#rrVqqL=K-dVnbA5~f2vy{eUI9aIDBSX!If!m@c%f^+hLN>j2eQ`F|Bk>q9!+q;U-
zy_?B_nDNDV6&sZGJ7SVY1#>T(Gr8xUwN_98$9naWvPUeQq!L2gFnjWTm76A7L)dl{
z0v&3(m=3(xUu_c0rifW{j`fV$q)ux7(7_?#mCqbiILjk($@Hya?ulx}=)h#$rv&z^
zP>zQ^I#5e3+e+5*tkIkRd9P3K)yUMA9h{b|D#zYTxW9FtiEU#XN;06Dv$02E#dKWG
zF|g_>78Mq7-8<DIDxC-Ui_Yu=^{m}t;hRpHHed)+!v~R9Pf4aynz}r?J3{v@BhK=&
zuJw$Qf&T#4sjjU}$}FhGOoQeOIj>{c^c>>UtX#5bBakYw3I;KnxhrlyqPVL^lBz_+
zrE$DWtsAfy!5yhZsGd(UP0mQirF1!@!9<Hvu~oMDPyjmtT(ptO5a4s|T<Y!J7cGci
z{a*ZJfs<5_oE~^SwT2<=4`g1`h-qWvk290_*QR)C1(Q(A{m_I~D`N?6j^#%K6prBr
zO-Z(i6m=WLw*%ze#%qw(^(MJSj7POkTIH`*sJNp_a86vuEjQm`(-j@EZ@hqk$ZGQG
z{7Cd=@UwTsT7zlYYPhzFe=*c;m2x^7<@K2V0JQZjH%?o4p^=nHKa#sh1ZN+Wdspar
z5skGUZ9Kwwt<Idbwk|OdG4p_W?(3T0w@vRXVlWj-h{#{-U39gk&DN1xm;AMVkN`sQ
z^`@<-nY&;o9jV2)D<GDUyBNvI7{=b!Jn-V&F~G^-8g4Bu8O=Fssawe^La-c+)Pln-
zQ!BV1T5U?^)a0yGy|*G4A&@HdJGiWWvQCpn5yJ_v(Umz3lgBm7i;MOv&7J(|K_Qm9
zWVT){5lEm(ISep(&OK|TyNheMY-L6ba??#MP8zK@q2}NjZp=zREg?~!nI{!Ik{Ena
zkhaiM3Gy6{1v+x&nvt|!q?y{<u2ZfFsrU9Dje7E0yX<LMYAO>Ho0p$zmPwjM;c#nZ
z%R5SMQ_Gz>DSN9y6L=-D&0AT9%#Kfb!g5XfyB#o#QL`YPPnUC6P2}Jy+;9bRq}%d4
zDXj_-Yj&c=rVakAkVSFQ-ZkCh%-zoe(!A+wCdr(xSVkm$O?Fye{D4>e!&Ognp03Em
z(l9j)b|Sl&{{YpGrC}}00%?Y`0koYdFfe=-{{SN7U-i-d0G)2Q;=X}=2uR_)S+XhC
zMgwknrpI`a=_@3B?Zt3ZHHir+G+t=@Y6OoFvGz5Ip^Gr(+qGWsi2neI*#7`_wt>x4
zW^~%Ohhf!kM9(WB3Ks&Zzk}__{{UXyG19cP&UI0xVUYMP+&=F;zbcWu8*bSK;lC<Q
zB`U0d_y*cWC+!dUP`?hKEC9K0%9ZR_J7-m(cskcdD<Vi*Kypd0vhPnzi(o?>jz%k+
z+4Ef6+c-G847IRDQS+MEv4vtJZs62Wq@yIwF%{hDwPBiunH|yOaC+7Tr)v(M0A|3$
za>lxSnu~1br&ZMCiO1?X#MdOnY#AfZE!@^%x0R3GHO(a{GwCTtG@)*$zL5lx$(96H
zO22eht%prhS3G=UPOgTdFo#dhI<NDtS5v(aG<M2!B*y@H3M)&N4V-ip*`2HF5a~9H
z1VkBFbQQr}T+e?yV=P$mYs;k_F3zgbyRtMaY~!+s%_!sy^IIMlw0nu9)a4(1;ym^g
zRhPKMLN}4vK)J<e!ngvw)76U5vME9rV56y_YRbKp%@qPQRQxL&#D~nWmNGH~WY=v?
zAbGS`HB0)QEvg}Vn^V-ZBR`d6xBAmvaM;A|yb^!nRX_c7{*|7L?JR)()^fF-W$vNS
z{v$N4q5|GuIetV^YX)Fgm+hJ`y-h3ZjV%XCH*z%c3`U==YTw-f6v(`70|0tvqY+b@
zcI;gk+9YxU+IWx-qNqY7-;5Dl(mEn-TcIj5*0pr|2`=Ucqv~r8r$y|JzRFvB21^@p
zzTvv#x7M9;J?)kGH;tQ2by3JQ)d*>+#Zp@Njk(icy`C*X#!%%?%rn}#e-~;3!^)06
zC6hUC$E9r;bTD(glQZ-@wWP|&EUZ7qi=)Z<)^!)W&YFJ-h2lirY!}+NHw_!I5_mPP
z8R*f^h}|=9K+uk(D3U?@vHI7dSXf1C62}^a0<pwR%6zEagcM`U%9{R{d#T0tduIy~
z>+e_>J{!EYvRK+5^?a5c>uPc3jgD%v=atAbshdpHVz@a7rv&;}tXf`2GDS3O2z(rh
z<ixF0HpA@Ey%EV{>sBq><Y7@w>N%|e_-xZzRK$JjqMyWO_Bq?88;SvqhX$)x)ZzaC
zhK?$q5ZWuC%uz)!5}Qo{4cGdH`d6FlBN~-x{{V#jOy{~QdiqixmB`CN8bi%wh9%6V
zJ0N`t#bGhYB-fWy<7T(f64>#72mN#v*I7={!tM?SHI#=Ds)X#1{#s7lWOnXrrG-Rj
zsJ#?ov6-qT2D$BKb8~#Lf6F+A4nJC$Ek^b<w1|AWc9n<CA<yC3ud@XOj{pv7Jy_03
z*xzarN;%xvTGQ7f5oivB_&}?Bq)$VW5g#Tf`^a)@q44u&;z8L)1!+Bqj^p*|%iPBR
z?bfy}ltE~!I0186J&hLT4XFEGw$dI#oYjvJN(3mK%~kb6=y9{0$ZExsWRqIUVk5bC
zVf3VMLLMs{n%L+(Ap4-WQT_3ni&neRG`(8x?mLMly8wvdG2kN+jOMx(xlTD>oup*x
zM@bgTCGG6cxNNd4dENNa?zM<@pAoIq&f#euFu2JebQEb~{pQO_t6E>ldkyr>0?f$K
zl}*?RPal>lrmuMxqp3Wwf+7GaZt6hjO({JG2(=@(YlO7XZ52txFshvT=l=k&N#fVB
zytB1-xRyB;LWK#CNF6hpsXuhW3G=3e(Jx~+k>wfT7V8lNe-P(CooQ-~Z*JB%QU(hQ
ztt0Ij$vrEV+AG-XbYAE*FA&LnbnsmS3nogGOmpvD4y4ddsC>p&#{(3qO*lJ{skl{2
zCVlnmM#fc@rCu=}&+A9ql1Ryoc9Yt<YE_GOM`b!vQAvpoFi&!7P`R#SW>&isMH@73
zN7AY3`kl(72{6Dh>N?i+>Pbk}ofuS$xwSSfbk8uT&fJe`#CVQq#c5FIqO6+i%i2&P
z=_>Z7zG6t8sZ~q(W;6ab$mPPQ>?p$e3g>N|CRO@*w*`F@BDoDfGF%hLWj#f9(0vT5
zEe%aO#B)y?+0IE}!ndtyD{LUK#YI^1$li>(piBLt5MUhDDek4XILQOjvP)9fEMG4?
z3ZFwj+P@u;*JUq>9F_hCQt@r9&#6bTdD~3W=DUGJcPQI>0&AjHIVoKvcRDS~S!m|r
zhIS(r`qdpS{K2RRWM*j2cMSF-rf)<P<ILn{rlqFpT7))N-fA=#2kg#+JQL8@PIWsZ
zy>jMHB*Eom4tipubBb!^#+=%AGA>4&=1b(6O6>VJ590T!?Z30F<c{Vxl*)<-LO9*^
zsiO$aennN4NWe>J;(K?uxGWUD^&n>ly+e65rNoW7TVhAHFg<<i(}kju=gmpUk;`$U
zOB^3)Gj5FnoG$>@OwSM6U&vC*e&HhoS53@dq@JaD)=0)W79Fc5NSY=p!vVTiHX<*U
z{f^ppu@ii(2<cr9g5PeZF#iCif7ZB`<<H?|zYvYrts?#8Ijfo#{{WS76NA*&GdHL3
zqt@)F6Wb(!=Oe9JuC(rsN24K>^{#VLn{~qvPAX(vLdsmBUrM!_7Qn`nMrnbsXL6rq
zSjXR0sHD{0R4`m;+MV<;Z&DMeT+8=C`~^iMmy(a-q!%eoYjnht39S3ONkn0pSPq<u
znI_O*Qz_w+Cfl@}_X4ndLeCAE+%cZj!^rv3(4?9;b0!#bk=GQe6@hO<&3cD38!ykC
z9D*<^jJEp|pS%qwG%cJ2%Df!ssO??WnP>wdZN|mtJ?omQvNxu#KUC3SxsB8)4UP^w
z*9~>2y|kqmV1t29jPIc8N2rjY&rWK}Wlj%j=sDRo>|}$H*1GQjYK?Dy<Zs4KQnQ-6
z98YV4_Aj%cl2zZ1MtP<F(13%qmj3{Nnzefg$siX@{{V6q=f4zhx^9S|dgiLc^Ea>v
z?vLJ>0sieVVSUZH*z?CU)r{twu<>k+eq|@W%~8#18I6X_5-@3|OHS83$yy<*&jPN;
zOsOZP0jmNAZ<vrs9S5a*u1k?f^s6nU+@8uXWjlu)ROtx)>1=Q^D~h$;lRWNgNuorx
zSGON1EZmG5bP*DPe6mh@=A!=qd3POjxQ<Y<hirVHjEd&&;<Yz&TPh;UBzQZ2f&Hk8
z)EjKW)&;yiTrmJl3$=(C+<WG+7w3(=xy50KwuYZ&T)U(aCrJ6q$UmKWkB6r==6Mz}
z<NyGyB4aJl)yH#m>I%f5eQ728Ym5Q_KGazh<k~Z@Vx7O%vE&-dy0=IpX&4Lvn$B>I
zuF8a+>~Xg(=1%T=nu6Fdx`!P@jQuOai}&t&&)_-@YAEiO(l^e*AT9U;f2DDnUaW`4
zKWde-C3Zz8uskh$4cVj2io98?9mb~(j;HqG?FuLEjKlc154)di)Y?7EC8Tpo=PY){
zGteIO)rW1SjZ#i6MClN!#`}~256Vp{%2i}#?1YNb?Ol<piy&6>l1YOBkT(wbsjoh3
zbO8w=M>VBSLyhS7E3sy{k8dMr=e1bgqUp|hWKmH)(bGfWv5kLu4ZXF)+^akwWz|j&
zJ*q1S5bA)jjVvwgL)$Uuf;i87=DG1!<b-}_U3X@em*!iFIBqYrtH|bAnmLSa<0I6L
z4QT2Sy2&wQARX8^t|)AB&HL7W)~A$8>@S~h{PfRXLz=9UxAwYRG0EMUlx%ukjG^HL
zPD2dlqn>?|_2afH-NMQ;Nz1A_DtpnnX4dL9;p*6Uh{zy*H8ZBu_fAK?qf3na?kRFp
zh3IP;D=keDE6nE{oYynrJKrYQ#;4|!fmu#TMZWtRuXZh%YX;37pCI{?VE!Ney3Mv3
zoR3QL+0Ki=V*nm&veMiBBuoC`r%!S#Y>cf9e9OSt=QZ0X!iwjX<%nsfp@4>(X#s`c
zpZONz{{XI!{OfhP8u}IRAtQnD%WYM_&Q2>oM_{@;EB^o)z!k$&V@PgC<~^Yyv*owx
zSEJhYrH4YMxy?)qvy7;ybo=@2H3X8~g1&Lc&1n+}tD9f&lxn-!`H}Vh;HaeWW~@#G
zk^cbS0;`#s`zV&J{7b5`<j9>rxGFFBO7&HcsggX8_60s>KV=Y$#hR~qBmV%t1!`&j
zC)6a67LgD909QRp$s>Anpv~xY`@*B=U;E~_w2fwanA#_0ET_}8V@j;KwKT0l9G67D
zZn}!$qbxC=IIW8b?j*x8!1{Hi3C+&MPMtMRnAVa)Vk|izS0$|Zm(CVkYzH7>y$H(n
z6kg2Y!$BnH)fua%!asymg32NMBD<6#jooQ7M1c_SDo}X?-mJz_yDsY2DQl;eV}-?e
zUB<p;)$VTx%p+Zq4%x08OS<I>KJ0E}%JbjMUD&fR<kkC2HGpnm&2#jy+qoUld+VsI
zV*VrZW3aB)&M57zBZ52yL5?euv)LWEb*0TaW1Lo;Lb%!9ivS9>X9*<9gU152r5nu|
z*Ox6eJP97?osG3eTvk4~`=XqBMn!E=`=2wOinT<%j~UfY%ZnlZ08H^&bD$S42d@>|
z9ga??W8lf6k58Fokr0_WF~@4@Ax$q~`EB~{>N?f5n$+iRN^Zu0T|CA<SPHP`rB;JY
z({5zgu)s!n#we%D6xgQb>}@Trj1jG^ym0P4hAJD)FHb;`2%#?AvB0ct7pbaq*qi%4
z#6qzc;lS<1FlpL$c#0p=lTj@?o@JP<?280Q;w1DOR=Qe2r^TJ9ti$O^LK}~oqjtM3
zSXf)z{{W;$i{?L8sV0WnZQgv!bJP~C?2cC{CPfvDcFpElG9OT<t#O|cG=nSJ-X}%j
zfcG>#4e35+j1LS>7~RHCDhKm5Jj@9k=O(#ot4W=-pSooDfp<VFn|%=6+``Mk$TDj>
zRXH8XsSbS(=SS0$*4+_T01DC#mMS_5nla{^WmbH!v0^ze!!>Ve$e3UVHQtT`4^;5<
zawM;40B0XE9*5~ihHfQDCAXU@U?6R7qPZ~;yiMrJ(mT5e{OI$Z)q33HAoZ?DZjGis
zLzOh#{^<KxOdr0Cn)gn2?C!p}>BCiHUsZ?x8aQfvLQ!1?Vve*CKnYE%Kt*%@p}v*p
z8rS@UYe)Pi>O<WXJ5x_u<7J7OPPGqkDe5vYREzjWCbNxsbv`y|NowmI)K1@ZToa1x
zB4$`sG6sHNpmG8AteH_QY3b(PH2(m@7}~y`^|Xy#)7;e#nqdqLA^s6u<Ve>u9(j?m
z{wADuxY^U9>_t(~3{^c|e<H!~a9`yW3syE|==R^dR>ulA8o956`b>&Ck}BWCPeYhY
z1}=Ic;8vH0U+0m}xvgHrt;LLew^%LJL)PY@H<4uRoN<cITa=D=(s#B7Mf=TAUtGy^
zCS3i|L|&yep@}HWm8WV1agSQMn-K-6U^u2+xL6OZbF19d>}+^2;@ao^@%*XH;ixr6
z)$b;3Pq>RwzZ;kUPaXKK<gSlCQncB}CA`;{OKoivOCxZoqmx>Z=o+1>sGiP9i?%Yu
zzhOjR<*~DpPQ~Hji@2p%mI+Stv?&<ouUSK3r!>uVr?m2i+vT%xqPo7X8xzfSqdIMN
z-rCC2Cz2MD2?`L3`%35DrPl22ZZsIDiWW&3!9ZU;de#-8%NES(-tm59z8}-=lWDk<
zZU_N8wM(r+B(k}T4iJ(*3i5Q(t<OsG=WC(P_;tVHYZ&_f04nP*<z+F9)ztnQC4U(e
zJi<?{J8pSVTsAvPP%AJMekeG}tBh{Ur0f$y3K5FNo)v^0qa4<jvo^!a-m$N2Go}IR
zm6IXCs>Q8S@-V0IOVqWdoZTf|yXIm&Qk5^^{LH8PZ;{4Ls9VD;#?CT_2BA8&xaz{A
z*-4voBBa`-xw#MgJ8)~7)g(zSnUkTeico2hIVOrxA1>h>bq!sk#3j?xLOyk)6Q!A0
zce*p1gG(CstWnV+?<1=13%a!YDW_dsY6fp9gkebCitDAh$zR}SLwTm#eZDhrp^n@G
zQyl`v@i7?MeT_{<-1IZ1tppc(9QL{#kVd9Yk$}Z>T6L}6-MUQ#4<-iI!02mfrqVK}
zD8<RM3foS-9wH2F-xl(Qn~%6X>#(0nx)y(FV5*0pCzDyqoTns_v}(pFl-dTjY)GY8
zjPvPNV=!7n36>?0%kwdA!`i#((@@czmFDNSQVWC@_pGY9RBYqaRn(HsbdlP@Q4t9z
zJ?o|l?^U76QRtdwD{IRac&%E*J9ON8);6asPklQ^n*eW+dsfcercg~!Qm)l_iDo;x
zjw>$lTr$2sT;q3Y=c=zyL%y5TuH2~=x#0tg`A_{b{<Xs$%c=0P_NM#w36#}48Hz|_
z?mAX7TC-|@3O#4S8i2dg)d}TAbfEED(|3u5MQIhTUu<Oeu0~n>*%`g*Nsb7X&Rw!&
zJ!?la<FMtpW{RRsYbe+3%F7WfM^oOWSeb?l3yk+QNf<rH`3>7({{Vo~#Ah6l_zEl*
z7}og*GBM9;tv!?s6geAx%~CU(GB54y?avY*KAEpJ@ub`~IZ@KGsMXR%)4sPj26Y6t
ze@dD+J4e4-)Cm!RJoFsWZ4wi>a%xx?bgM1*A-3cL&35({+Y~-IBN+6qI_&JEv@S~k
zByozs)wFfERy%OrjbRB|MvYw@oVITSpEm?kSw3LHt$In_95Pm83jlUt<Ea&AO}lG5
z*yU12)&`j;y!d$|)hr^s`@z(6T>{%is8lHFRk1OR-N_X~s2J%>ibj?URDAB^tu#dX
zSa=7i>rG(99FfHVXJgAR<Tl<TyF#Emj7cJ|mrpA@esVFzdh%&PHaqIYdpJ1u7SQZF
z`0Gvc3|r1J9AdevKX~%(c``_ETO)7Fj0{z%<Rx1i4CH%Mx7^Y&&?J-n9fk-k`d1R}
zA=SY=uyDuTohw|s7@DNP(q=8RyBrUC&$fK4<mUpg#C%ApEfIT3bN0`2>&N3>z2P`S
zqT_cw3dw9?Eo6?l6a~N(F$ywjO{|hLwP}?iUza<wzol|o&Y;t$loMd@&sycgRB?kv
z(t>lj%TD1*sI1(5p$R=4SCtLVMqdeM#*?%fGP1b@xW*0-Ij#mEe8pgJK^U)ETbvZW
z6I$<7yS=mW=DB87Tp(jsbO{}-);S6N*}&$u!pXVYxzPxu?2f+D;z=V_k+KwEX008>
z<2+;YuBP?WnZ+%Ll0gd)M+T>wb!f&Lj{Mg|>GQ=(C!0IQR%Jl#f~dp_W6!lYJc^mC
zS_$fo%sg9HYjcP3PNeePC89nTJDg^#qseV@q{zlMOm*WG=VK<_MxRYS<D#Se@1};v
zk#fgXwF>M(eD&j@uF82GqamEI$8%iNb}+;x%z2mY{{XbyH#jZFy-~S|;kp@9k6NVl
z8QS<#_+Dw-3uWWu6{kEUjz>{mo)UiLyEAw@#B>OZp;@+{S{Q^lRwo{n-_2)o90~2C
zW+9lIdR4=PjGfp%m5o{RyRwumqA5<p$}OBI>TA#S>(OzlZ3inDuQt4|E>6K&*(jEz
zkL>zQy1CDora(Tmoog3Uj;6U~v4l&BA1Btjtt9^d#EnLOgwv)?ZIK6t;VnBc=noav
zV~Xcx4K&kG0vc(l0}H`F@-4;x09_yX*7FRC`Y8M;NWs=LWz-mN835HtygeLfBURvb
ztYu2*sOZZ2mXjQCVV8y=S3dr2qz!K0be<d9xhh7^#>-tyd#2py2Z2>><7qCF*osN$
zWi89{j4`Mqh`f}LGe*}NwvN?LQC}`?!-9IyfsjVvjQ(}8r3{f-RoL2==R+C56{n=!
z`4{Nf`B?2W#}{XFrb_W>?AF#fWJy?LfycdCYo&|jOcwrC(%OaD$ptF5k3?LsoXU)?
z$rY8VLFLTcanx3jmXM#AMm)7=4(QIJy`u&pr`v$n5rQ%^%>=!Ud*UXn(M1lQ3HhY?
zi5`{0TUedABdN)+HnY5kPvJ%MiaT}fnu>d!s9QZN7qNEKwc)rntlEUUw?xf!HgXn(
zA9~@a?<3NttrK3+v5MHVib=2(aqC!hI#aSXt?VIOZaa-$2&kA`3htv76xrrtX6Gdl
z0^vx_b6y~`3vF>aKJS#0DOOjejWF%|R(OW2!R{=G{_Rkd?!9k8#dR%?expId@x@xU
zawNeg6*Abko7P38vn!mCc&>*^hAle5r*qF-pXFT6{H+_=T{BxqwP?|=Cm5P;fd2pr
z_NQEg5}66h(AAZThPss=OrA|%)3mFL=2i;J5kGpewx+I1D(7WsqF7q2ZGc^S)n61|
zK$c!mUDCe_2U_S+OGAy$P28tt7O|-!t1*HQFaY4zoL(J|X!EBm-N`kj`Xe^Mv?=ZX
z0C{C}Lk@#H8qK+eYpq3K)RSydAyRv8>}Y;v!dBSgn@XEO@dd@RkKMZEThw->C+?2p
z0;aRRrnK}Z>qjw##&T&i*gV^WTyTQ8&V{+yTiXb%CAyGd5Mh1IP>I0E&nK;P;n%x!
zn-eR^B6i}Zg5!E)xUQ6Q4`*O{993fsQ{09NvQjV@cd44X99@~Mqg+V^*tMMe<#Iji
zqlDsS2fcZzS)G#A50Sf4NC%SGJ*!xJMo(j(y->H!gZGpZRgYTw%s24S!BgTAis>^7
zS^!EzObBj2)VI>S=Un3Yw*LT^Oo?uaSsQ5$aWaO{P{n;r{X_j=`c*U1yt<zoH3T?t
zdg8Y<$>f*Jnb?*XRKe(LDQsMKI%_Drh!s2ICa%VMS1UCs>l6O%VN-(Knr@R9?@MFl
zj9`6gwusGo!q(;UuH&``h3iTigV?{O#1h|rKKJsdwaEf$u1~1N3voS;PG{L`j_zvK
zk3ZS%XKv&WD`&A6)YZ@;`_iAzrPt=fQN$14Jc`Th8gwzs?HH<(zYXtBD^bblQ!D$x
zO<aVjOqRu5_9h#XuHRazc_!ZVj5jos(BJS|Znn7p0DONc+SO7!U29H5cV`cf-ei$U
zF4l6~9zK|^<+0|^<7Pdzo9S2D<3bfBMpcY~l8DFe55Uxxmb1qcGbO;97MV9Fe2{bA
zx@bqJ+tTFvlR}s7Z7yUIhb6c02hh^#c1HT+?Kcvz#DmPp>+SWfn_cO#Jk3VuLE=O9
zi*$!+JKaK#_^pX8RB1^O!4SltW7G=5qD~fN(RSr1Shj9rAMFaeaDrd7uc>PADpU4y
z<vm#|NaMUVPxi_w>a4)mTd5D4D!QM;W=r_UuAJ4|c(aZxhAQV=osp{rxBwhf#dg;K
z_CN{0WMRsVS1Ym?HIIFan#x<9SbvDnvi|^(a(}s5GyJ19rYBg%`58m_lJzFi-#2Uc
z*5F9l!!m+tVeI3tbDI*{o!4WOz0>Rs+{D2aOPtkz?Cm)J0L#2p=yIcvo?XVao2P%z
zyj5OZCr#bWv|p7q6b^IlMl`)GPJYb~71UbT**%Juf-<FQD#9&lT{@FbMpgdCvXHlz
zZ5hWWii`U;#ee1AF<B@?-PNhwa%r|wOp&d;Wtiiob{4i4_I8ULF{%JN3feJ=&r+n`
z?8McfN0<US3T>asfGdhxR6Pi5NgV~a65EAtax0QL1a{HQZr49&iLL|0-nq{e+STlE
zMcK%~-pzNfJYcKrKsS4F{VPs?^vhu?x$`4YxpUMG^}Q%5YRw~}Z`1NJT4s4in8k?o
zt9S3`Eumfki<9`)@mJ8!dNk2rR=f?UNgQJ>_kM<fBSEJDHh+HoN1(0a9ofp$X(5no
zRGMSCe5A^-J<lK3u-VaE{KfIq_BBq{1s0^mMp;<z`9VLWX6i(QU+(VWx$1lj>!wvg
z8n-+of753F0CW1+1a~T@!pSumK-J|3+Nwy#E1<)Hn#y~#YJUnlp9E?`E1xZna(ET#
znaIs+OJh2_lyBc`4t)WxDj9PljQ7n<_Z)UDM%ZwD>tP!z4OB%T-KeT0+0(A$XjT#T
zkaJVXa8dRIC-ALKSx<I|jMpy2EO;Kcrk1}Ge5wk1nrR~rYOU-^b8jT^fE=mEI5?^@
zSllwIxdjyU??Py~##@-0y{V5R_3K`7;~4_+Tz38G)gOf6Se7*}n80ztHP}G)lgo$`
z{8YBn=3aRMrGh%W2HzyP8>_LlhbTC~uRgBlZ9O7R*7@YsX`lzSbF)L2*R-O;%!quz
zboH)QOo6y$5O76z(3R24S~}c@Vh7B<$22441Kzq0bJ%_p>qdK7H0cj9^0nzX0;z;_
z8qy+<nSdVVm~Ns2hTZ<}G^{ryGwm!vD-$11)Xy?v_*4L&TBMAnE3`?r;kvkApsLpv
zml6UD?bEjuD-|Up%e7tB-XfSeA(i<odWzAGQ0br6y+`j+i$|cJH5B_3LozPV%aKia
zfju~^<w?qH@%e%TEXDDGjw;k|v#$UO#%(z@9ges~sOz{*E^ytei_|c>te&jg8k%1b
zqgou+u{I_VXQoX>YU2(+*&NnYpT5bK<=rjue7Wjff1P^=hOY&pPn0gi3;|ih!df(R
z(ca4<jkfGEJDNw5@xdDuQDV89MGMI8WGb$6>BVGS+N5aGI2Z#tBBF4MS7VHw>~WWb
zODjJECxMEu2-|ZX9ON4C<E+nL?2~xPHj_(-aN~Oa09xao*q2?WfC;Zori|gG@Qjd=
z0t4&LYhT048aY>-=kui*a@t1m_>Pv&TmkJ`3pmCziuQRFpSo34HJmna&%H+?Yy_)k
z@uT;U(Ic70GTe*p1~toYDvjOQF)?4fdI~vQBhM`@sn<nyIZ5=X=auG{mSdcHQM8Ce
zQhns;eQU#YIo;WuslC~2!_pPjAdR>ous<s9Nyz$El=d*Mx;(E@M;ePrhXB7nonE|n
z=e3NFyrTwz^(63}*|XG|H*6%l53PD1l36^&%V($+(?X(!OFfybC(w_{1+k84=>n)H
z9Q)UzXzoa>ouNf8%o6e*Ju8<HT~d!K+^>Zg`qjfM1y_uY1$oq@o9PEry1e>ry;;js
zbEDT_Hg9cfCsejf1Mixx6gw4Gih_Y50eP;6NMPC;pY~h&)1}nTv17v#V%L9nVsl*q
zu4Y)%O*Je8G=`8F9u<LYbYTL?=6Q0umc}wc=k=|VUqMaY35Bho!YU#FYNj-0>dWU^
zxg8YoT(+qglS>Raaa<L?BUQ1>y6llu^&>SkvMfh&>T4IVpHf#F-)Qeqvxt}8gbj;l
zDO-SjYTl`9=BmoQam`ZEh{r`DLY2{JF@=m8#=geXcQ#V3Tkzo55!jBR9a=}WNR(iK
zf@;hRPbAWu+?jN#s3TI!Z!Tp$j|Q!3u<wXSJ9Vy_ZuFY$an@Y8n7;ZP^Kisdb6)fv
z;+e@MX3&O_E|Lj=C+5Mxrzd4{C9%YKvh`%umw3xYpU$vk)WQv`UR<5oaMCD`?M(c)
z6+NB1YEp<J<eJKE%I1()m5oUE+CVJWZ76xIT{x3$r{1_E@2Tk|bUHn&-OG+bFH>DY
zD#C#y7y_eXN7%D&<elAxRld8Eeh`ost#l^iOVs6zRIB<61?;j~`HZA<io&{*`uZD)
zVTF!Wxn&?4Q*Fj3v|^Ozc>b*fkYB?LidEwINIs&ok$2s>$9nEhLxrn54cnO;>sDZ2
zolQ*kEpBxF9)wG(#cacJ$ipMry#Q#|H#ZU62!V<mGje+NrAjJGQ#$jMT1dTdV*waa
zSLs=k>2~XudE9<ga-^7}iIO9?(V>#{5?Nd+<Yye$Z3I!n0>ubjpkkxhC9s|$z0jXx
zEJ+Sm=~eBtxUMctvk@c2NCt&Ti|Y%qS5VaNqmiJrg;7Dq;%g#XyVZ63M64OT8m&gi
ze$p-33m&I)3xgc&57xFVb-R1@7W#rJvFQHvHFB=Z?-uFJab}Y*%_XNhhw&k<9vOVb
zkoq|vtz#{Ym@=lO``s3`d^KurE_Q**Jl7*SQ`GJ676TI56&oWNKU&ILOFifJU>~h|
zbfUQt<5#IKXv-qP-cg0h{S7rV+h}C}09gF}D)|wgwqo5|JTSD7t4b7SXc;E4(^#`>
z#ENBmd6lvAE-GacYIEi{by!x~>fZNoL<%x<>sye%e7OP6Mh$s$R%fD}(JY6(ICBnt
zMO1zwUdK6mpFIicQlPJ`!v6pUC!bT|3M-_@D5e2PKoL*=rM{KtTJRrkt-s~lCR>yj
z0*}JD*wEYz(NQhJ#%`hhxP2;_uO_F)%{mVew$R?S(dky(WQr5gF}c3Al-`Ds+~`o$
z$Z_9_=Vpdg+{Nc<4nAI<^{rtam1>~y%rbto=djUXpK6vTfHPdmGhALgq;l<rr45)f
zq(u?NsyYFR!0`pf`6IR;Q$#ltIm^j2F456=sV!WHwgdN%O6cfk)a&%9#8yl1Sz3de
zy;sU!i@>Z_?9-t};wFYxR_#>TvsRHTf<u5!L}mTZ0P~u=5qlP+pKKtLRVI|m**yhH
zjh)MK(Cd5(!C{@4ZUR7Y>sY=nx@(^kUQNo9{0#Ksw)-bCU%Y0Ho#HU}Vsmf*04x?M
zl(s~mag+E~h0du7w~9MUVG0Sn!n?9qgI%;*<R(ha#i_MRX`~GLY9p2m*gT(V(bKP#
z+$sx1bdo^8AK|Wq^kyzExvQ<BneAr?7t7Df>sNHy+G{htcWy0__}4U=v9uFVOw1B_
zcOq0=Y;1+3{?M9xw%Pl&Et8t_F{^2Mp3+>&Gb>GmTllr32RS9bF<myGGGqK}mU_}T
zm+_HUaa#Ivd6=$xD)KriJ4DzD6jTzNS8Vn<W4TIcK+R*_%a_G*HhMT{-bSUHV#Pn)
zter|h6|VDJSN<GKBENYrQniG|uyb08BL?bgk`GeqBzexOb0g|^1zmvi4}W^7vq{hI
z8}qMGY;f6JMw&tg?;G=|ujQFcV|x#i)~PbAjMtuF{{Ssp^RGnkb;DkGd_>MmM8p%H
zTF$-P0+p1Gj>6etR>pbj&0y<Ve6Y@CjhG{!UX{xf_c~KocN4<!G@7WGmSBMMn(qYU
zL{_?Ky-X@a8FyLE>W&6E;<@Ji^H}^Q^{j<$Lw5qQ^{ATTO>i<AGrdo}cFDd}aMOzC
zey5#X>et`eQz?yh-9{Et6VttD&#OqbA~<3QT?Xj@?Oui%w?#hpDs<9X70ueuasfq;
z04^4^A=DZOytjp9+y1Pu`?YV~W-dI@wanMJD-6H8-87ApBX%^|UgF{?V+SgLfsVD)
zP3Vz~qie8|?m)8JsZEW7z>iAJkO=oFQMjCgid7_>h>8%&=1MXE;B@U(bz!y_2!34k
ztSW03&Z=f)Es!gq@R-J<KmBw5wdcojr|`12u^~-zU#O*+?o9W`6_riRY5XYD&~Lnp
zxZ5Wk@m|%ZPxhM_rRkLwRJ3BNvMTDw7PEi2YlF6z7iE1#Df~i?#)xXzwrs$s+O8oY
zYFeg2y{E+(w2i`m%sDk{0<9R?2NV(oFcWyDNhUHjlgDakmdq_uIc`vugn&m{@^2j5
zG*U;ouuoA-lUtIf(BeYv{AnG2W%Q~nj%i0h#yF|;<SyI0XC|gw6pupDZFWSTE(zrQ
zYo>sFqP*$uc1L3AjCp?bS>ofRWg?NEduhE76~b$_KV-QRAo(QWhoP+~O46~NRdi#@
zicx{!RUk?d4^GwTBaG;LC$7z;YH`gRg){Q8_BHO>QJVe*ndfIl9<_Z8rF)vOT1K6-
z^{GT^Iw%!4G*KOjj1QiKDej3VG>s1;@#FoU#LXtq6|kh$YYW57E!bsGsIOBF`RKUz
zIw(GBEp{}cV9HxK=B7ne!)^yOPOO?qoPNqGT*$H-FglN3wFI-Ob#5!0mQ7y9s@DQb
zFST+Sq>7jB+33{iSvF~7jJuS{JGnfTsjZ(1eQ}E7)w`IpQ%zjnMo;&Yezogf9XqvE
zjj^z=LU|{(R7_{OJA^>0038J>z!a`)WRXWwoIx7M7jePI<y?NNslDB~jbhpf<c{^v
zh^XTRkyPTG&R*qrGX@1n9qJo&KW&fZ{xB=Zj_0WU6JK7pNwnaO*c-rKt#WN-Lv3I|
z_OD8#)f_aoW^9pSV-%go6{n;xlX<jo2(0C?yiJ)uaW3J)wN1Q`2sjn%;A>u*T~W2U
zq9>T!vA>!_D+MLkanK6&l}7bB-6U3LX^zlxJ!$K2k{N}0S*9c0<?s3(af*yd4WhW|
zS+|;3QU+`1BP~j{g$+K<qSo!D!QuJ*tFzsZeZ_hdo#J`eg=r(r?`_&knU^>liKu`o
zs)Y3KNiy0ftg?NrB~0#C7_VB;uSC}Axct}|s;RLs+}dP;j^<EKD!WAjZftOUO*M(B
z7}+zSGA-HRM+4HLn<|8!y{nnF-Pj!dqO957%OBFXTj*LnOn^T5;<cpADkEk~iFJKb
z&zSBY06z-8w!3mqVO2MFLeQsjLc-0^4h3!KZM(zk{^NgIs%BM9e-FpAPmk$Z$E|ZS
zL8h8{sbD1`qy{&HW%Kk;J8s`Hc^etdGTE)V`U+~$NLRSi#aEb}tbM7yibh7Ea;asv
zIV?I?J1Z$B->&7&a8`ctwX`rrxsvgNa1a`yadU^l_BEV!GI|z-+hkmG#aEHM$6$L<
zi6m7ivQEc}(31Flont*Sieo6UC)z<no}#w2zx8<ZtSfA7OLJX1RftXqt~Pa}TG6-_
zuWtCoWOS{I@k!FQ<09tI9Z489*M@g)J<bejvW&Yklho%HH$YADqaC}CV@j+T0;eF?
zvxT2ECFFD0R-dU}`t2<3npZss1J<y#i`g&lt|XKAK*zN@)73=6UjsSacQ<;xtU4*%
z`Brn-C>bWWUgbw*QM{ALP^xR8@C&2MGd4WLAd1ei{o6L9@uNq?2qV=bU~(7($GvFj
zHt!|8QJ~8(7_UE}=`GIU!qseTmNo1AN7U5v$_4=B)b$xDm&msyo`SNP`AhP8n%=I7
zM8giQ<u2y+q_O>oHvnTi)>YQG3gjsm_Nu7u7>a~>rzhrk?~HFExbZ8^83j*LJw;(#
zo@H!xJXchH6k$J&k^cb8U#(rYTpS-v)X!nW>ijCm`if276ehcrl1p7QP8$y+WSY8q
zQ!%M$StfGY-;1QN&-y_v*b3+DwY_s(zf_9h$Ya8R(uUesHm6RUrE4CuV`5U$CBBtB
z_FgklT}7=}A2QUWvHt*cgIN1nBUs_-y(Ojg7<^>b*6uGY%u)fzp*4HL`s`YibCn-3
zIXu?!sqWn5t%XpZDrSDIs*7z_NK7Q0gPi)**D)lvo?|Z{WFJb=OPL(<ha{0fEM=Kp
z<5EI(s_Y8mJ!%%`L)hqTWs2iThB+HPYCLRDw<G!2n%bj8)1!4+O8!-iO_|Y8a;Bo#
zxE0uVIuUPUHT0YSI3~GHo0ih($!Km_vSuazneALQr>IK0<nhBJ3@{uwP6c{)v{B-x
z%~6_A>(@{qfo=(Gb*s>7*G@rcw;y^}W%5MmbbI-3cPyC7j?~Rc)(I|@$r6yn9Fg9s
zyTq6-S=ke73-*m8mE4Cbx7Qsi=ZEfo(O{_A+<<lKUVT{TdN5|O!9P(>h#xEi+Om(t
zQ`qPA5LV#<;Eal`Ut5L#3{O6<#1vOikWowmn@k2x{;h|#dDg$h`K?#**)rVE9RZ@b
z9^*n%R%~W&q5imiDy(Q^kSOkJ%Bk_QPKKm00K<+Gn%~mn+ioQ8{MS6qB8>&5o?k8!
zbAouRmgSj|A-KGX9-y7?^say$Ql8^QZBB14%#7f-6`!WXzUXnpLvi|=YvB&2!yF8n
z<Zov4uV$R|T!BuW;*pHRyOZi_<%}h+5PF*3>}0n)Fhn-q%C(85ub+PNA5v?bY|Te&
z%BWR`R;(jh$6{r<9Ee<CRaZtr0TpCeUBrMpQybENJ0Am#+gyLXKb1wR_+op{7hGBS
z(V;PU<(r^dyImM_a?Ol`!_rwx07+uFitV<?JZ+WEdR6NU8Y!UrME8pd2Pg{+<kw{+
zxd@!PqGqvVR+Vm5hV8r_xUFph<&#p4c+8m_$mX?AR<29<b~SuM=1qHUs^AcWw-t9l
zSyN7Ib0azEJ5^MLp`lTKyVTEiXob{|pD|mnrDp1~%=%n6*4WHoteEeP_0Nc<&xLLD
zXR!qRnv(DO6g)X}*P8lF{Yb~@U4{AiqpuEnQfSS885G^H4QS|hYFME>isz*5Bch_b
zP3MyWsZTFj?5OlPV6`%?u2lgf;8r_MWRSIRwtIM5e(y2RTOatAqk1J_>rLiq@nH3C
zYof2fju<a~UZgs4#szEV<zww!aNKowIEj2cb2Y@0835kI^q_cVZu1Yyy8g2ia{DF?
zqKj5Xk+-N}xO&%e{twaJ5t<FhZ*$VR@bXm?XC^9Ar(}p*!de<P{CP%xwXvdUT5Y1q
z6}_}b4<sn-T}}~b;r6q>=EB+AV*(_{(xSYwvwL$Kkj7KEBB4`)=s#;yxfY+M*xM;E
z!5U}KRo^Bb?kya#xH^>|QYV7k^+cpdpa<nt{3$rb<L8~XC1}DGW$1fWZOcuugrMy`
zGf`IQXm5wKp1Yh^iz3uk{?%=kRkT&ag$K*{)K+3g7UD2k6;|^HVb6N?@NsUB%+aLj
z=up14dqj>n5E5T|@z1cWIBmYr@vCv>0nestYe$u<nX9D6t)}UoGxG%pp!BB0rpa*(
ztgfL@XK3d&tCzZwgr89@w3N7wrjlU_p}{+9r20Yr(0J}H1aYEyf$@+<Ve=jJYTFg(
z(iNs<hT&LwMno(8c<WR(OF?g=B$BpdpM_D_4)p3IqS>t}UqhX@Y-CqW;Zd9EPyI9g
zwdSSo8qeTncZeGr)U@cFJS&=s?p;5H3vDTiZoLl`?0yK;1=f?WoU(h;hLS3$!d}+j
zWt7MLlU{oSxYSg7j8mTE!((X;YS_MF$DyYk!dn)>rK%w4H2(nj@r$0-I|UyxJw5AH
zv5O*m%V_hyR2cO%)V{WpCOH+c=}yK@q)_F%jg*iN9kEpB)aDYH<Fj`3KD5#$)YC$>
z)}e5gj!a5~8~{hPc(;u2rMthxgdMrX7uLl`Rdbd@xx3<;bCb?+IH1t0BI5(OHM^qP
znG8#}@3%in&U+d~YI-f8jxC9vK&^{Z575^OM^a{hjY4455~vx)XEX?@I2(DcZ^Rb%
z?`-PCZehUpsi*NEp2rC)H<UpiGVp4IgR$bhRk6gzLm`dxpQU@x!u@S;^igi@`^d09
zwS5SCrgwC19<?TfZE+Gd)^HD|Y9_fvU`Qt?0|J^xQ<kWIogqIgM;^JTS?>(2pdd_e
zaYc|hGs=8<Yd4!U#I9Tk!k=%H*EyzNKiU>P42Uu~>MOGgD^=z1p}rPc+4VYql4#L}
zM|#t=(meQ;ywIuFmadwLG^2TL!fID1y0)>k1S1zbbDjleh=j5g3-SZku4)wJ$!T>G
zioMY>J5vL-dB(YU=IU~VV})~El^vIvth7c1t(N+lwQayTu5@C}Z8V>?!Pl8r`PZS@
zY9hnLmXbJT4Y@!byke>*Gu<AJx30*nuo4&p$)TjXeC<^L`qHrZn=|j@x@fksEO2l!
zSgCH&+O(4<!U;GdwM3$xtZ{7Pucq^^ozHVkw@>t#&r%+|E8b_Yw1K;*TdejiydRW~
zmCwe$eax_v^PmMswR)8i=hAjYKCfdUP9;&teD<ZWlq||$gX@~-md8v@=|Wvwff|xP
z>P;oST0Mt`6?kgi%6B%4vmR|(y+ubnD;h7&^EU#$$kmOmrJ<#ZHnw0OIIBNpIO4nv
zRbOjS%b=UM2sZVmykLXHaiNr9whO92$~fK4Z9T6HQ>Yy)(8IrU;>CPO^V{obMZ^4^
zSoEr%LaPLG^SQ<cYEfgB#Fm<R{{U$i**;Vm&#iS9DI`p+go?V4LTT=9M{zdMkx~uQ
zV{;r1!mu_hqTz6(6&1?ic;wWvn0FCxb6nQ7r|y{~QG$({s&7&zv?^K5w(=53C#`5j
zd*-9C{3~{hN_Y8N4o}`Sz2W|RJ0JUow5PF^UWUJir1LBkdRE19Ge(3o(^4U{hJY}<
z5&r-p++X$4{{WqI74$LqP>|Eosb0(E!blk>fl6B!9nNrB&2S`{n}yCZST_!1x`=1y
zE!MpGMc-3t5nV<%JdytZ5H-xWK55TTYH9T{nk-2BCFJ%MGVNyscNEzL^o;$C$K{fG
zf(2GhGWz2N8)+C1Y7%TD(ljhQJE=-RFo@6lXBE+Dei*u6_0pmb_IaqOO|&>xj7`YA
zIU13^a>LLJRmgNpG&#c`sjR(<>RZG~>}W%6XoLY83|54GOsG?WJK~=#R)Nnqsa|;e
zi9TPtdRB&&Kg?q%uOhkW`<1M9*8S!kvpdBaa@|`<nET_8V_Ytit;uHp0A;xOl@8qZ
z72k%RB}pTir@cxN8{$1gTu2f&%xE*8Q&M<ofUufebG-BG#cemuJDjyX1Cf_k66!N$
zak<GAWP>2icr}CEv}9{A7?m42$);!;qiI^4bAJK(TzgeD%?lS7X(o00jO%xFuNmB}
zji$Tb4{5$ds!ZeNEAo#@@}#;Rq;=HQxP5}5wQo97?e|uQE6D{%%LdAN#q${iG4vfx
zWVv(5Cp79i&5DBMrjCLr?yfYk9A^!*44TTh@WMU2{jX4Lk@r`nX;Sw@(30hPbv%1q
zUncm*8+Ou0eR-&@l$hApZ`ns5+G9Y(fcw^l)8+2=tf$bp9S?;q<VG{LLcDn&nIF=+
zUlMB)LE)!1=u0Nm&(wik5RbTWRF$fnyZCB7R^B^@c3E6>HQabN!j}FWvU|&t8S#=&
zVO;W^)sD2ebs}9xZF^l2s+B<BE;|b9^=(OXdnlAH_B;h9vvK8Y^dmKi_H<&cW03Kr
zEvsnOw{S*cW|wImqXwn;b!j!M)5j)AU=KB{ZOCy}i;Y&FO$oeGnAe<-?yvH#ISFe5
zqx+(^hQ|$Wi5vPXA}cjUjB+_Xwa<8h)ot!es_jPpb((sc5m?&3p`=d=xkC(sKu5n?
z<@GBM?90oDNy*+{KU&VM?sd{}ako<j=rq?Vzuq_%+g~5--2&H7i-~SApQn1t5xkNd
z<!G@kh>qPOUH}-ydA750_HPl~$U1^QkgrYbc=Y}vy@m4|fX;cYrqW<QAok|6Xx$w@
zho;QZf4f?`fGr4-M=E*xRa@OMzwZL+mFz8}2Z5Z_dQHmP>I|We&O#CCTys~8JqYe~
z4!c5{dJV6STIP<DEximKXkzVn?rM#FW)Ju=I4b->MRl1AMk#=lhJYiKe(k=M=UyWJ
z0FV4u%lK@WZjA6gQOzE;$svi-nrktcxBmd^htj05hlZGsVO~v7jhZ6HKp{^BjdVJ9
z7EurOYYCbn;?*1J_evuNf&NWIcPyJmk9V7Ndbz^=1#4!rmgOTKR+r7YNOQstf2~>4
zW7}+HJ!CojJ5{|SVtSJ5Gx-*BH*zp5mzpLvGD+#lrwtHD%Z+zG(zbLsL*1)4ZfjXR
z4CA@At54+EzB;!RUqg>=yGVE}YY(z()T6D<=3VUS6jo8KW1=DMNfHy2iY_gTsq&6R
zCnT6zvpJN;NMV91EY^wKgR$^CX4>Qax)0@DCD4!l6U&n*B|o!%Y<0zIwmE0<GK6mg
za%k{aMn!`#c-@X$e(pi0Y12kgtZpK9hFMP51C(s_71d3yLL}YAYeQ;r_rPK^+lt)q
z<(Hp%_O=;Vha=S1Yu%NNWbY1Q-mtR1q8J1B2Li6>=2e6!I8o0YxvrmT<qlR&l7UB8
zCM`m7w}<LIYLt<aWdtlfNl9V{r%Zm8u3-J13)bJ}d$f{xSlg+MphNaw5}3LG&5x!l
zuGCu(AGoiWp0t_GZ{H(6NjlUvH!TjLj5=qU<c{ZUu9?-%e;OuPz~ZrQb;yQRKbY4=
zN}2{23o4SoLmFGTE*mU(^rc3wTb}nKjCoPi*!<dLW7R94)Uqs!e{xFesqn8u!=>@h
z)Q?U%BDL7<iI25$$Eh`UD0HOFLm-WzONpU{x;%haTd9I9ol%OCxCC_->R_+P&odhR
z_@m6Hn^aOz-ArX^3jkaAMQL~%En4RK;`yYI+W=GsJxy0TX`^pv6=!3<wKE8Tk&Z=k
zz9-Zj{JTh)@0Gd7VM4zyqvc%M8WwRiv=K+RoDNQE(@1?ya`QWz@}ix~pYI>epQSN3
z(Q<myPxpv&OW8j|rMt<u80&*s`9ENj%+Ft!HQ!U>`5s*ha(vDn=v~FH79^#xQfDpG
zZ9%1~%{ST?{ZU=tTJ`d_-3dkdo$<4BOUTgJ3*ew4L6S(RZNN$3QG<djY(ie^IA^8O
zEG*Hy(#9ZAO1I1@mg^iA@a~&wEx4{aip9xqdmP-Z*4zwnTiRWus>K72haXDEd!)L!
zxqE~@NVh5#PdLb|)s>PhChp9p)RXP@#@aI+P@SL?oxq-RP;2umU9d6cD85Bp@zScU
zwA(eNtj=5ROI<&NPV1??qyDw$EzKwJvp2+N{wFh3R&TfzJ?ds%J&@o?Z0C{Iy(8hB
z>b1=9=aU#9R4vPv$5G-*Ht91r)A`q&`MYnB`_)TwVX@Ih+zQjvo0b$U!R%-Qigs$u
z&C3{B?P6l&3~&Lfaze7iV{U|2y$*g|$Of5g6KPGu-ldAp-XHaGs{I8!vK1JuNkXXT
zDGGx~(#Xu#8YHujvnU{BX1oW*6Ret~ti3x`d6Ks=lm&>ltuSp0MtajIts0qbJ$b;c
zzd?bXLAL-Xz&`!!3B3)Vj?NuEA1o6$&+AZ3Z>?A?WR>i|*E2ROTWWWNs>|~)e`?fO
zORj60WyGa_b>$)Fp4G_TTtx(iONp|_u{Cs6#WS7MFCuxj#4rz9&AAxEb~WfnW0G5t
z1I|BM>%0rD&!uXyq;Ir2!Q(ZmGnU7>SW9t!q{%591C`>hi$p7j%l`npXnHatoV8@D
z3Mjx)P1*}`IvPlFTN2`oQZ=!_`0~(Are17J5g=^i&{vP$v%w-y8xYEPtGkO$8B=z5
zI;{&+j_wij4o4uKL9MeZApkj1>}%STl;EU}x$5pdFlEL`Z*x|FScV5m<DRa=NUwb2
z-p*OyfDrx_<nzc&ySFTPHReiu^KEnrniN@yarLbT2Vi>DBL-^N4=+%+UsqqkyF0~T
z66!X@kVqJ3^{R=C_eZR0vLxDc5-HpoFhTq(!w{o5sp@F7MqaCU_R~Phps#$_HLYIH
za*ZC`1KCA$VkadVBI!jrU7VfVZ5*F6kVxo7Lk8x$Mh71_&3Q3i@;y!BZfjRE+G(ap
z<MRu!<2dYVobcRp%c|U7Tx7YAJADOu(>%(18}lSLHglO!5~-3am5SOgw??3Fdsj2q
z>4DQEV;fXp0a^29COdl9Ja;+~U}_4As@dFodevR+ai}ikz(K}+>tq)?8(2X>R{VjE
zm6U~Y2_V2VT4cbi8Z<>%25;g20D86>37$51S0#Pxy9;*{j~8Uo#NAlm+`NimDA_n6
zg=C)(>e0@wVcmw#cM8@t<e3ty7PmF6G)+j#*AE#6Lyfr{3h1qLyQGbC7e88ajF>{C
zQF=QYi~wW(YSLKDKjok6S$jDedsufS7IOjje_Ej>h4aki@KF8GDqT4b+QYe8T`uSN
zW6$GRx4I?aji1Zs1Y<NtoRcP1VQtPr<3rUqkS)(&?+UqLqG}@w@DLAPMRoE!82d<$
z)$~0@;RzfZ5B7y>Xx8&uXlV@VwnamcnyAI2DakunvEhjst=M+WbV9hE=R*+E8kR`Y
z)|yRA0~5g?@-4;x09_yX*7>iZkHUnu8f!OHMr4ggR(w*P#m91`?k(+}HQ*6}TrRM$
z9-|y;c?+M;xayj8CSQ^$(Yl(&y|y1}!iwjHj8oX9baOeu>BU>HlWWBxK;c|vPeWS0
zjJG{n-WWxsazcu#%h+*P`i{S0r%2IFA>KNu6{2Y5vu?#7_)M&}gB&uk_X4e3c)Cdd
z3v8?CcUBc@TVtl38}3|PdiBA>DvzZi>Q|0QGr0B!xaB!BwiKX%PVFEr<&U|pvqCtu
zFY0P(>pPm((w&i~BCXB7SCTp!t8F2QWp0=>Rnkg9Lt1VO*}aThT4Z`;SCV*+&+IzG
zm0T>pF#fgGwOjW`9X)6@*_0ht;sTu*^fjw!$4c<wl11fFnETgEPI{Xo&u)1wD)lEd
zdrs7p;yDfvtxH9cYq6JtCSFEQO2S}hj>9`I0ou>1UaV4N!wC#n{{SlXF}5hc=m4)a
zy_DULPKDeR$`_Y81P*D+{{UsPEZuRF^`%F4#fjNRxq$LXG>q=V<mQDZdgHEZ9of-q
zBcRkZ2eZ@FS0ix)ir_V`7doRw3bsija~kWEdFNx2oYgv4?s@gxrfZ$V<oBz#z<r&K
zbx&iO8nK&l;2zx8fXCSh^c9SaVs+jKohQ>7WRK;C<lXq^`qtz&(dyQ7rr+EKC+bCY
zZA{^ly<42_qco=aIG|#Q0m;X6TN8MS;`dGa9Do6gGY;apV;vFKDQa|H71Ea0*{>x4
zJB~BhR!*mO`p&WC#^1XQu}AZvE@tj&PI9Kyqq*2gqg&0U+gnS3u#kSdQW><}9Wogd
z<FyHNB2#j0mEsy^u3G(ypW(}URF+m^RW2~goMN?w?9O_>6ieEL>S;QDSMMLXwH5R*
z>pEDSM&q}E=xJL+Pct7#v$%szzJI#qN)_}Yk^Jk7)+cG@-tW1X08L|6-J`mKkF%G3
z3K~oT_guG&_*9o~psj10ukB9~@We+6!hgVO=&yL^#7fX&UfyrklGrwO!dZvnF<d8#
z(`K_F{`dyFUdNdah{%bGxIXpM>6fa?26?RIq0n|78jv*df4VEDk)e1Q2hEXGwTaIs
zY23}Vxbx-_H}I**`cx54tCt^pJZHJCJ8IGB-)44J)BUeZNju}3o+kwnzY6DH5vM~J
zUKqQN=xaK@rv>~NoHc$SXvKCJ0HvS?(i&hgPxWo|uRQS>#pw4nvi=(r-5H>8DJzym
zF)>M5brG9<3tTFWN#ItGk}cp!`f<&9HGVc~wkb55uH#YA8tN>6<I*psVKU-c-LXq{
z0<xzY_+Py3v{a(a+h~LkBx*1eRIMkLg^xuUthZ$x@`s*T9G{v-+ADrE5yq&!6k@Jq
zqa$3Dvqu7-b5LBx<>*qe>eTumFyN0cGk0TwTUt7VxW{lR=dma?iCRgL9Pkckf@va*
zDx3jW%VdpWjQQmAUF1DUs^gaxw?iV@3<{M^akkq!R>n@IL7$4CO?JcKM=bs(c3uV_
z=(i{RvHYmj?fgJ>i<`)n<6|z-AU#tZK9#PvIke|@D0p;VUS1`%EbOV5Mcsg{j<wKf
zlizt}+R@oC#JMMdT{NG#XH_m@!H(-9&9^%)U6<e1w|p@?vCS$1S1XfIE2&(sb<?qC
z{^Ch5TXPauj%r;hW{%*XV}>0np(!`0+KPsSO=EjsG0FLqQ{fRw0x1}6yjHoRz6zg{
z{SRJ@m#|S^)`kR2<oL)BRnO^Hd_jLLys+&(L9YS-00g=bQpBXppY3@Ys1yxKOyqED
zjQ8P(iX7LoOZS|g#*tW)+?t!VbjC4)(CD=94?K}9*A28V$WdBf*w%`;*mLV$DwUPb
z4!$~4ouRR(Z(^xxiy7#>4Rbn-IZ2d^5KU`VKH@nrwRxv2l(g;*X<tmX_75u_1#!h)
z*wVJ=K6TgyF}y(+Wv!J4&VOi&dL{Svma}gfE;gO#@UG8LoNf^wj)(KFLk|u}GtR|6
zYKY+V`C88ST&j()9tYODKMpFUg=`Q283P=5Cz{R1(>kK>O&XIOkVxK&O>uVYF12AR
zau=VaU)gLUx;kCJps3yJ&gyiOZX~NZJ7)ebtr7kr9M|rjq0?SDOJK@yMtQ7RCz3_^
z+_pMqyKBj**z>TFPB%DD8|pV}8IR6pi15gI8m-}Z8T91d<ctxF``4|MPFH@1br>a8
zsP;Mt<~Z1_Li$t50Aw6CYo@byTAqD%YK;}%W<bmv3`bgl?iydb#FZEv^sZ+8(qk2S
zi!IEJ78+Jy-AAq~I_6!mIxhwIKT1nU50^tB<YWRh$idukRIev#uUIo~$m>qLzjtD(
zqdIiK&2)YYJFg}G0Iq-5xR&8Rg_Jx>F|5l}ZJ%U<rbR<!T95%Zvx@1w4SFoK>v=gl
z+si(j)FVf+=$<Hl_>#_kpU%A5-~A;{dK#(jOg1`rA95-7tt~yVFdm|W4Am65t1~)1
zJb_^k`?Y5^qS(pq88n>w(<DYCyJDh}Xi<K7K+hd%lG-vgd%GpGm6zu`c;hwTo+4#^
zVts?<-N5v#=tMpvE@?o*9x>Lc`KKnOn-(mi-!g>iYr3_yHo9=~zl3D_n&+tXvC&Bt
zf41%xK@2KhW;j(H>IK&IxgA%_Rm{Fv40^RLOQ3Vq+tc+K+UVO|#pTL7de=>CrH0=1
z$x=3EuzDGKq-@s$F~I6;jJ~pw_sIzRLrA4g@%IjTio89NCau`Z+jDjHOJ#(QDFVBZ
zjGD{Utc{eLFm{o}bt!6K<Drq#uW8em?xKC$N}PjRM9u7d8{q9lH4P5hQI9aLAL=Wh
zqFa*<gEbL#QdyV)C_<aq1=MZ#TX)b@ZFbvE$XK4h@kNH^W02M^uP;+lwerz3!3Ul;
z*N@-FwOvo}HY%jFGN$g#>sz@r`;jVV1Cl*OdQO>n7Nuns#9W6R$4cqL&hc<(Z8v=l
zIAde-DEi{9i(;#pI{H^WCa<%LOLKJ{PHV<Do=%)$Q_P2H`d64uE?l03n&P3cl~<_}
z^6QGzwOoAOwbG0krxDwIZhP)r*K-}jD#qS)W)i8!(UZkoH!!Gnehk$lxzJ*bVar0z
z83f}19zROmb9DK@RUb-CS&y08#(t%1bu=T5<a8CAHLsHn`FJY1H4=*5jxFqQT8zef
zlKs0?dv0!G9*!%*=DhBEDOwgfTlf}zG112X#wwH+u-Mu*#oEvFIX!V+ocbPhJ%|od
zN}fVUd5;77Rz;<$`&!7^;kXrxS~H>s)y2YE!0yBjD+z8MS0o|a-7{S9M@Gh_orqL2
z%yRy|l~&z@Ub}nN=vO*RIWh%a3=L$&#z}$=P`DgcGThQ8f0`h<S$5<p&MRY0)s|a%
zTgoF1>+M|_S=rdb7f+fz#}j?x#E#*LS`+gSetK~6S{Ep74pa>M_1WzXKV#-;T6mT@
zLJzZ%&M~^T#;&UvK2hoGQ7WazhnXe5tQ`LUs^jZczqHhU%QN}Z`)?up2i$Pf!T$i3
zXY;0=L=*n6^`z?g2e5s|3&H;YJpTY%!`A#u6{CnuCAiPts*PK{iFAC<I&T>$l#j8u
zU+@0_^{LVL**63BMgSl0*GD;zvhyan@uD!#*-8HZfz@cdT$6^3T9LS8?^1KnYL>g2
zz7&%?m+`KzwQ)VphBlj0$rH6Txugasf<NS2i~hPl^R9r`(8u9ITMacwQadgD!O26K
zQrNieWM0Yki@3l)BCcAwTee%hRP7lg;8z_@O-V1H5zhb}2YTl(Zdos6$2j7#iuZ__
zNr7XY<krQlua#~c>Za_7iaVbUYC`i$SV0+t?IYW@R@d}cb$fhpv!~sE=xa#qa>?3v
zIES>B(qATID(lv+*}m5TxoREo*x$C3AXTW_4T|Py-7;Wu!n*AtN@-m6AevPECOWU(
zUZtHT=d3Jx@+xbXyt#J{jY<2|uh_WzmD=uVY4Ult+YcmhTt|-d{6Z&D%x(G_>A$-P
z9I#fMNE}>kh85gNra`Fa4+Mh&q6#tZU2!<7yCit^^nYZ$p4TLZoUcCBPXrz^J?J$?
zo}H*{K2^uh5%uR4e*VVU?YBUnyJK!D9)&GWO7Lcgx;~Q_WgcvUbDHR;eaVXQrz^>`
z*n+fDBDS^@O_R}iHCtDckpTLd(P>G(<`Va-rP$06nwr`V^q9AF+$$LFj){CMUli*$
zx@Ec~r!LtlqrG_^qjl!VB;6M$0=)=F-)!?GCG65N6}L9yTNY{liI6K=n+l%c9@LvH
z{9IES;hl^L(>0P;W{ue$_k=8@mreUDL+`#=itTMI6G+fnJ`AiIA5&X5v`#MX%-|d_
zk%<{OIj7j$$#<wn2vde{o9S6S5z@71Y2poKbiEhLkA`8jaa^~DFH!YqA9wt<P(k;p
zr|_38>~`KQzmnTajp2}{WWgZz!LBggEXVh<mOiBBt{V{|)swNFHH16GyXH`MHR(2K
z5<skrv;oqsbug*W=C!!5H9Nxb@f>_8J?h7YCc3k^YfF^<+w!N-(py;BkJjFODQ1>A
z0a^Il=zVL<uB9yGXPlbGOGBcYE!cm=yHB{kAPHn1R9?LGr}%Xk$u#UT0NM||T?MVo
zF;7bw3#fU%IGs-)DTsUGvixU(szr{bFY8@>4k)Wh9QQSD(&WT?<C=zScODSF4|L1B
zIa=ysy8i%)oR;}Na4`1urnWJk#L8E92FChNA$pF!$E`&xn519sk&tT(T1Q%UiaLJ`
z-ua0&u|2Z2wQO+{1N<#qvR0YV9b#tdz<KetW>?bSzk-e`zYyU?c9{hf0EU`rfXP1L
zuRifF7n%31qwq+XZj7OvQl7QQ%VSSkLs|6%G7En@Q4Yf^8t3t;br-|uBLG*ESNM$C
zY-7UPuGPKkudzSqk$<~bOmQD|CJ;ssEmiGT&A%8qNJoEKo`su0)onsdTkHpxHJo=F
z*sWs_Ymw;?=jv-yH08J>oSsj#>JHt3z^j_N?9rV^U@D&IT?{x-ENH%iCa&mmZ@Q57
ztCqx}CUG{SBX(XX?%QH^^c9O<@-<sCInT-kHX)Z4v)IXQSsZtz-GF(nTWWN+WL*+m
zlhm5323or#vB>v3e*~x&w;%71<yJgdr%$QsEhe3H41(U+2Zd$%_|AS`#<l2kH)@wC
zcqdZ1hg648LKZ2c-1r=xF^bl*cqE4I)n@^n!z;1c4i8M$)E=6VT0ZsBsW;4qOJTlQ
zkRZV4XdIu%`c>Z!mUorgAdi=tb8hxq5R&BAbX(MXi0Au^U6(tv#Uyh>YkdNgVYvY;
zPaKN6ZC%_=Nyf;NSqe-{zj);RYQ3}CsE5y48J<F*q3OWwT{!4P;BPMI^k_Nqt8Zf(
z@-rWd)UPU|X>Y=!@k?c`&#|wYYxhTf8ei=(W(aDLUbtEG@np#Lp{Er$3eC_OOLA+=
zmXl|o*~*tTtB#;kWH`-wPTZ%-{i4(_h!Y-_iK{1_Z#<~Ebq6EXl=Xo)XUkQyp0Fo1
zb6w{^ocaplrSYBEgU^hZrg_<~M#lxttH5f~Yd&E6sT*hUuDbop-Viz1I#%)cPD+~9
z`51m8v-@U_<|W+Vjkx;suB%74`yQ19E!87Xe=5qN)ia`-_gWjS-V0qOPJU?y(Ok}w
zqAMGVhU!ky^u-g>i)N74X8baFmdN9f!!*enyvyo!8TU_$x1u(62!6Cj<6fmy{uA^%
zjX>hh@HK(Iis{8WO!M$x!)FWQFeF!O^dl8p!xA^yBMrzqg?iabdaon8pYd(vbO|D;
zZaEmi%~*E<um?`H+g|OX%*1~c+=b8;EP4)lR%=P?z|Kh(oZ-t9pHeAHn(YlUFxp4q
zRVIA8w`|r?PUgB4E~Hq}M|J9WKaFHY{%oFaOtIpp6D4Ga1z?$eVlmRTycTh&PCx0N
z^{z<nDxV80c$k00-fzOPBWy6F_Z)Ub=K0{=-78N+U~Yy93<l-LsHinPKV6($_-W=}
zy6CJwF<x^s{`hlFWW?$$+cxNP{35jM-)zf517I3;Sq9~W+O4E<gk~x^z^l<p(hQU;
zJ7%_99Gv%)RwH$}EOs&h+chIZ&X{H(`l+BS-R?weV{8^?Vh(=wIN88+`=vjHJCdm+
zW$RY)rGsGxHk|%d^ZvNLMyEdf<TYOCP*#<U8C;TE)|(bt@CIsGvA3jKnCJP_hV5Pc
zma1*SI0O;SYm%eX>ZFLL(4!Y(>e)g_y$Gr);km!GMe9krMSbcq#(imfO$OnqemySY
z<4}&`>UL&U&J<u*K@*jeAaPhqS1u~Zr>GLQD<&!C+6mKs!Oyi}8CY!<Wz*wakbobh
za$YXB3mU%!k4oN@uOm6F4kgA3#}w$F<^&!^dPwDGs(d~1B-To?06e$g?mZ8!dZo0N
zcQ|t47tji}x}52>>|6UjPFdKWttXQk5~V=~r81<pxiX+o00Nf+q}Ij^u*3G-@>ecG
z@sDcp4N(BUxEuk!xcq9~#xeMriE|MPow&_y_$yjl37XR3u@kq*Pw<+~TOAN~XQo4Q
z8z^-kf;!VLFQSSf%-ySoXLZg4#FIVTvNV|q3T^}3^!`=CNPc`*O(kfY)psPdOvqc(
zw_u1aIK~uywX!f|%_wa~+W-JrP6c*wPLOzmP7jPZW#^|gQ8J$AYvGN$ez$dCpahtC
zs(nb{;=1NLM@-Qz$4&dCTg!%u26a)NdgJw5o2!K@l0obRa#xg;Y>tX)H+FK@u*nYA
zkxK438LIZAE!0H+0C-oCW~F9&N!}-EXd`%>w$@Tc0qI1M+sxijRB!JZ_N<ybYqazs
zxJyYTSQQYebK8oK$g)W{d2#ymrOO)@r7@3hYjStX9A!sEAXJdp*#_O@$Dg37QHn)#
z&$(h~?NTB(C*@S&wX;*WM;|`IpDb(~(==d;+zu5g(xgA#TFu5Kb4*hdKiyh_NFY5?
zKPrb94NiT^l3Lo4%yIrzS&|4<mOZEsVMJ6@&~8&|%<?Tn`5L6?-*m`v`PI8s-Kvqd
zD<TXUW~I3nYe_>dOz<nCw!9Y>h;#E}kyyymI~rE+wMP|XI~-J$O&(2HM9ncWM*t!E
zSDyHs$eM=c9Y+<UR-`?SR#=$M2n6Svg%y`_u<C1|z*lU?j`X%!R~a4YJ&IeM$AvI1
znLPz|7{zlv$04VRZ%UR(hniueBNxFR@-4;x09_yX*FbCNVep|XhMK8)mba6E!kQ(s
zEpBrcj?ig7ckxWv&!=j{Yhz`oK<t0kE^C3BUhH<OFYiNVg%!x%F_LC}<kUyvWYjO|
zmZC-9AmMqbu5U7;73jTaYV4JwH9Qx4<!UbjH_IsCdXfH3Xlr_$+BUMYFlM&6`@_)t
zR*`2smE~gf&5fqJVzWfbSd62O<ZG3h+G`81E@Udoe>%prb-A?(eU@w5xGJ@9oHDTd
zYmuF(x`QIRm}D^M>U-=Iq0}d@D)S+<n4_x$$TiTAF0F$OxX7ucs&O^1?#eS0O_$1;
zsqazvp2%G3@{DJ4ky^uAFUaSpxp67?Jd)W@+oM7{NC6*>dNfQD(`XzP$K_q{*z>FJ
zW68W<Ww%z6;Rnj{5ZLw2RJFB@@UFQArM(T7!!Go>Av@w3Q(Y&6G?{f9<&t$iL>!QN
zp4F9E+6|l0jpmP2i8iRA&3SC~Q9f4A&8Bm-0lAHFcRH-sS8nKVr14u-(kdrDAKsj#
zYuj8x&;~zv4rye%RA|@_QUzyCUTGWFe6-}w3&z(?bEv|s+@c@DHOOAcfFm9H*R5@A
zc(H!zSw~9g^mtNxV-YxE(zC4X6QY*pcA*Pf+!A=&aY>_D&8S{5g?Be^tr2UM;V+8E
zxA<2~T?@le$#ai2P_wc6nxk#1HmN1VgyJ=9Aoi+`vP|Wck$$H-(1w+iFbo`wX1g5+
z!=6M&<%nl><Jz#E);eL_;Cz3mI%<h(N!t)P_cfQ{<(m5W8V4$3V8@zq_hE2)&c~r^
z6UD1(>tp6&%L286d@$*Bs**DxP)|MTX~~y3otB5DM*=BT1J^avw-Ieo)K>7_oRzv2
zQXQ-VV+7Jk5D%3A?K~Q0rG0kAFRVPmPT;(A-n^zeSGv@^gzLK%jI=kUsJhk4z0%sk
ze{*g&j2^Ar)|QkO*F<^Uhz*uL*sZ9mBa;&D&J#zJ%i_qSX53mvB>M1b*Nq4KO(8wJ
zuj^a%GJhI5;Y-t+lS88Lb<vMnjWfe3{#Di7UwIxPjO2l_bB^>ajNY^P7d%3yG>YLk
z{uVyns+qYn+M&)iaBDfOZfi;2HaC1W+ipv<U<9tZ5;DXbh8Y#hQ{y`0@R_>HZfF?)
z020=9eI5;3ia6^0L#VE_WE4{Y4K&jMkM|GKy#D~j1N|f)_N!?85++>?h3!tRCSO6c
znCt0LaiZP6)VD9+x%I4j+4pB1Ppx@Xe~8UG6yV}z`qy7zf6!tl`?ZY65(%z8XOXdh
zIL~^wrbp%525-Fln$BAmX(`StI`T~15`O9qe+roHQ9hWPZFjfcknTCDuI2KrowLq3
zrS^cv)6>kmUOIfnv^7NYESY;6xA7Cu=O>fPiH~!|NutZ;`LF1vv};0oiyApblyj4l
zPL;^E1G^fVCd$5rGIK&7o~e?Fi&7V?>}hi{;N{7tM7>z_tx=OZF9n?2Tz|ell}_3v
z-G_>9H9M9_tzyfUFZY35U{<SRnY5(Lj|oT*if!ECM&f>zHjVz3sYz|UR2L)?ao?IC
z{p4KMjO(M7=UKkQ#5mZh0sat9Yhz9z%W}#=!vNUz$2Gj3hb67*L34L0RD&zDr_4`Y
zwQb_KZz0x2F$Gdad95O(?u6W9Ym)0p7=<yO4&2onJBTeIh)WvfrYq)mP6r~l#Y)&G
z>c2F4^iy;!j%1lnj95s;h-1m`$69xZ-#$q1gI_-5YE2u^mGKc~2zq+Ga5L&+$x-Uv
zwEGxRj^;S0`=hbZFTjW9YSWrfS0MGS#P0*=WgASnZrQev=~+uL7Y?VU<3!z*Old31
zMm*6iwxG(tb-5I}(f<I7cJKYl=BK4Yva7r<dAWSj2CqoqJD-O@neQ7Md1?XtYkttl
zd8=P53<%$7`U<xzMI#<t)RXLyFNb0mm$E>^DQf2+k6x9(JU|_tvqeRq>A!uPmy0g5
zc9zO`jdE(Oy9Lot`$U@05v3H4stWZDn<28o6+M(uA+Idz>StXZYTk$c03jHDwNUui
zsOo<S`W;rJabx%z=FFH<eXFMz@uPzU@IO<I@iK?Cko7qgkKx(KQiZsU{$IwuET!Id
zI`*oLo+0L=1;%nQlTye8JEon!Uc;?>RVn+*Ph-o)xqE3PyC}^(lS_@oV;#L|Sy9(-
zPr|sRE_pkIS0qwJM@^Xw=Odt~S%V1Uz`!{)l1R0wjjc<JMy-`6CaBsT-iI~CUEWO9
z6WxNWjp{37!by!nW*rQF*0PS|Rrpz3#DjLHDIG;&{nn$XN8v^LcLxCUBL=j*I@>jv
zV?7y))H74jix=>GvA-<{F@N7D73b5p??pKhj+;r7eUxWD)vW_z*Xu$9aZgm8Wmj8m
zxUFe{(o&>oaVV}W7F>#Jpjd*_;_mKHq<C=%?ixHeh2ri6cXxN`&b!C?u+J~Z7+G0Q
z<}>ek&r9CDjB4L`T;%V&K0(VF%X)UNuvqC+PgjC2k>1})N2r7Z!!|AqUcB9<_&67v
zkbA^S`_|@~A0)lxGz+hLq*u8cwUr(~C@IUgBYhdnq3Q>jh|R=Mae?=D>yM7WSL^=$
zee$c5I*csQN_z)|<#wF`{u|}6!jA|iCN_KGb!i+L0O#Iba2o0#p%UTJXuFY9WjNF)
zKae=y#QeIACG1?+sNa@TZlfaNc^yyxQ0T>kV^G2g5?L`7JK5%JpMe<IEj!2V^4m&z
z^@fS2+8Xf>yAuod@<Ay(e|V{LAF%s+2I=04-y>EMynUU!Tqv+OyIo%=hCA}>*=bpE
z6c1N(;t+N58ZESj>_;{wT*wWGT-!;SVU<^$mDlG-%K{SZ|EMAH(eC6xXbR5?mXC)<
z-ZeUgXBUsjAAE{&jOSqOYN}w)!dV1976B>)>!F`lOue1<SOSci#^4QzoFZ_j9_K_*
z*_EKNZzx)h$iB#2w7~ki3g<EWhkmwvwy(>Sdz=$Sy(80K5W~G%SbBqujSaMgKC!IN
z607%O%X*Wv@ddW}rYTW^g;7<QdFxlyjg-aF39TxL^mi}QnhI+;m4ZE*XGqd^;i*yh
zj43ZK%Fm0EPPp$F&krI#hIoC9EMbt>hg*we%uSt64b5HuUzi%n8iXH+CV`jjtP;)@
zC!x$>!~G%V1>xaHg`v}NwZXbb5!0tVZaFN~BWcZJ@;8><wyP_m84@O1I4grA#H2C1
zDw3MgB1ZR9e}2l5xH@vETvCqK8O64@TD@5ws-a&}cKW8IgiT)?ZzPAp^WMo?E90kG
zv=z^zWaFq|ny6YhoVy}kv2U?>pfDOw_v|L;=Av^xJl(r9s3C5UeX0axpwFd;$sL<9
zomQPsZW^T-8KXT}9p&mTfZkJUeq=4#ie=eZ!u@*)hl;>LKD<r>S5BRYxco5&L#$-y
zQN@pbfg{}v1zr_3pD5>;8w{;E;HVP-1>2FK+tVlKq3g+rjl;2-Zv?$`BJI@*t#alz
zgYxHh7J8EB>{d`*Vx1CAgm5%A6Kh)1XaH#f6Z7t#8i5x)LTE%4i~IzGw`3s@D-ub}
zLWnqr_y3KXq80dH_~O%5O=+C`vGPE3EKILwBm2{Iv3FL%0SoaHC@KF+r^A%+JfX`q
z4<Tx<CCe&g!4<T*u(j-EO+C@Q<dL3WK$)?iW}>*99#3WUN!S_GEC7A#GTvxj#n<_q
zlk+}Ex8@bq4be_rSkP6R8DRa{6-e(4rptD^{8H3dvGdS7t2YIgO(CS==w}kv2=_fo
zh#irvS&+BcRjoi^&-2Z^UAovt?^=?f(QB}&110LI&Va7(sxe)k6tuc?EU(A+!DRP}
zo>%1-f<SVY%@JUdnFRiZ)Eb55Q|agOU;hQttQgY?R$I_R6XZ;8?1)md+%Ml7EfP$p
z`Z+X=$Ll5hL2=md`(;m3{84<3!chSmYN(||iJXW0Q`I~wKdMM0>Z22$Kg$HSFbDS`
zO5|sv$m(lA&#$B~@;mc`Ela|UUo@Ad6hGs2tHE(H>Z`0eoKFUR+ly@JW|$0YG4eam
zDI^WJTT0agjN4q2ooO<?KO$4z41bShMl9Slk0#_?GJIjGiXYvdWSb%0yL+~{LCF1y
zQJo4~)yhjQL1&1p+*jNcGFsZI6mF-w8pd`OXy3}$>vM40#TeP#<((=;OkY!oXeQSE
zhxE%cH2h>(fJ>){v_ma?S4gBI?EB4#b)ET!<(V0v!lF4ypIX!;c2MZ2zb6>X8-8-7
zCk#=nq3Rg<q479C?hI~nN`c#2d&?EbIY*;TJ}q6B+o2B$L+_~wv_7oe0PNPy6VD4>
z7dW)+z#ga5vwo;;3Fe&Y_58(Q`6o^b^?6=)#eFv4OaG2x>IFBNG>Cl~vbisFXQ7E}
zkVrn~fzdj#wUwtZI(DnIvXv<aRiaKMn{CUL=#YAl3KUGcPL$hpB!BX4tYi;gcRpl^
z=yVVo27NJFf;)l%KWnTzU!`0RaU321?GC3hM7tETF&AIsTp#Lj@ReFlbh^7E4tg=`
zzL=DMNHjVWykAKcly<4>bZ}p|9q0Ns6%eY?th4Baq3Rz}^Xee%NsKl2kWqatffR_n
zPV=Hq?tB0V3FpZ;-v0Ry>GkU>*6sThPJ}$q^}<RzCdc3;)uYhKjG`&^7kmZTyCQ8D
z85Byd!64a5Dt$4tc6KE%^BZci?f$7Qm?i%}#&8xp1l}MFKa@rgS3gVv*-4&C9>7`d
z*_3rLw`K0$xYT+W&a|pKjrp1((J;u8xz|AbIHRaLoUZF6;Q@<fu;fYFV&Wb4!{1{)
zjQ|GuP=+{ILU@3-%C0V5vol-LF3g4j>D4$$jSY|MD6+wxuRJYd(mJ$EagDObNuI{E
z6X_SeGcth$I2ot!?Li-Xf*V@ZGP4lP(DP=%RSAHfM+fJ?;*@I7v0>6AFgh^5D%B`l
z%C?*jy3*@5y)>1LX}Q=8Fg5Uw_wY%ojouax4+#>soM%*Bxe9JP{H7)K$4n%l+@2@A
z2CL;S*}@W{l~**;@GcxgvOJk=$a^&g)0yiPCFP0@f8Fuhr_>8<^814<|Do29QLbQy
zZJIL)o#ARL8mf<!D+x)^RKspycf8INBtd0B|1(?aVtq^1mhn!3!Weh3E(?aQ#-g|-
zZ_^M63uHR-aNletNVfL+ZN(SuR-XEcKPO*1pDm0QA9D4!Vhr9+{^V-q_>`<3mJ+vg
z{h)D5P(Po2`=-^j-F3u5bhZ00wi323CZ)>99O^b8eyDo^K}izZ%h6nIt9b`8w9#&e
zaiFX%c1uoo-#W`BL1xB^!gbz#ccM(bVnT5<=e)nj@1>0-Z+fC|+%``v$#@N<AXn%j
z=N0<88E}c%Qv3ohRK{M(EJfSY8!gA_&Y_v`MlEQ|Ol3^KJpQ7xA3oa<%$DDph_uXN
zTV|K`d$(4#*nIEioncKQkw=<SmYSMwF=eYT5|pAO{PUDd^1-a1#k5(Puw0KeaHycO
zsIlC^c06n`_%2OAy|d%zIC}7?lfI#VKx(qh<wm8sj6k}mcCokBw2NraGgh9u=pNc0
z;yMnjs=ld+H0X_H#Xegt89Fp`sol*13|hE}noi$vEYy(aZ?;<=4HXP=KaCu;0J%yc
zuZcw&s2<=!Q{p#|MDa~t=(ilaYBgNlS~YF0%q?LI-?z9Nlrm3<l8xnwC@L6(`KE$b
zc9CsWtEHCWBtC!SM_wphZvKlW!j;LfXO#qGY?#{%N8nxJh%$kqN*X2!u~P&)S6`0t
z^fBS<{`=a0Fs)Y<i;9w;%|rgviE0LX@Ip1ssZXH1vbRr+am(l05GQU=SXs&6OEduM
zgte!tKnAZy5LsyqyI)6$)tD{zsf=*3aRye<J*+PW+8-ltj@hr{Z0S>~JM+Q0*~>^d
zlof4A6)=qc)gl3J+_XW5Y%)$+#U$|)aEr_5jb*KY(&~Zra@*o|VJsGC0zIa<UukbM
z(sh^n(#C$<P4~mAEyhD-NpZEL2mFe<{uCkGlOyR9?LU6!(541By;3%whR<u}DSlI-
zS%eu2Zem=Xr8(4`zSEi_AMJeTR<Nd3N8jV8aw&2C*<)XHk`dvQ53@zJETj!3-YGcJ
zQk8p6rlC$Rt}|Q6{2@Z^ZyIn|8q;t-9^;7XB3^Sy+CW0A03#H#A@qZ1?u}?$_+)7B
zudGf-Uf0DuIdC49<tOuAygnqCN-YYHqIIjWssrX8H_umx`m=cl$<o)je4Txq+X`em
zFzq~ai2SZ<nEmE%yxn7N46E4=?F)d>@lRCHw;l?6|2QW|vLk|$6TwkZ(MPvwCFW2h
zhhdFVLG)VoYNz#Hh=WGDe-_>As^|w(et4|Ap?SUuuUw#Ci^_`oXPH>}WyZ}%7!PF~
zk2l@A2kYqxaS;ghYX{=G<B-R`RK|$fxUU;uxD=fNlCw5Gd?EEXo=`PGRlkt3|JiYw
z#Bdxo_DJ6lWc{l|gU3?S{j;PA{r+3Btc00;A+|5^_@?V`*_r~-Vzo7g7Kk10l*~`#
zOq<1Vx^U{ITE9Q4E*@AwQGG3FYR)9X+;T~oI;bTD#jcO_qjGiIdXp(r16ci9>;-XN
zq<Bt3tz;oL7P#}NZ`BTUeM@iEaCy1RswxCdKy^ytptkv8y*$Z9B18oq?y()`F#A`b
z7!`d*;rYzybK15#ueRE++c?R5y$;7(9NM%EhATX5(I&1(!Yq@PeY8M0&8v?HvG;C)
zC%UNQs@Syn)sj8)^tXDE{bCc0q)7{rBgIuQaf?|qbZVNz9@WT{^o?c+sM%!DpNR3v
zN}f%)4@{Z(#c&?0)ss-<ZxAhyCKijpx_n$waXE|H_U9-iR~_jO`+xgPMhwTwfkgM?
z?BtfqGFa*@I3)7kBe%>^wiAVz?8XcMy%vD?Q3Bg2CONLQBaySl48&wp1&kQ=6RZo%
zt|ebR3HZxCm^+H!hwE16kinhu)5c3(Aa@Z%Ciecmr!<QY1GbqaRStOdwv|(X66aj3
zBa`FG)i|C`z5?m;GS85^hoRR-tqGhBrj_I)Eghk(_7Y?5f0DKou+Y2)#S3h4%wp!I
zpItmUB+n>aEG9ypY<i4Pb3D4lJX)43XI2+vnv^5bt_+bp+!4EIJl4I2$GRDp(REAV
z#9&5aGLG(*NcXfTgi)I>3NfwC=FK5WF8B)21e0=Nsht!7Zw%Ee-u9fyJViM8&C0yE
zjFUVtdNV5gC@N5IqRr7bFpo9wnq_-G$M-^qfQ>si8&4j0JP!P=K!<Qi8XFALrDCgu
zhuz>@tbrI63+FM6$K<VRJ<`O4(DO~_QKAO?loIFNP4`{c-a)tS!%eW1wHMxYR`d~g
zz1P+sKz4~5Ob1J4Ta6Sc(YUf%FUuef<?Xzt34#$)a+~8IlC~-cA%!VtwBRF1l3LaY
ze^6t#7!YhS-HiLlIx!{U)xRN39PI45i3*Ey8a+vjyCW|1YYj`aVE|NHFrjLBhHuSh
zhR<m3iGAehh|V`HnA#ew15UeE?6t2jEM4rhJ2Bpvu~qDMQsTa^Y8BFCSkjP8^UG{s
zqV-LjRhBXeDQm6Vv;R9Jzp~R*B33`W6{SpCs?xMCf-z|vE4AHuCbA=A;CaG>1lyiT
z=8n;h^!Wa30os8=*eksmnN=XbveHD9e4^9TalmU+Y#FDh1m-g+P(XJs&p$bFk#g*z
zKC<djleu&uPT|EVj(q;FhitF2Y?|MpVp?n0m%4GWWSNnlUPAL{rq!hEMIY0tHpjrO
zPE93>BdQ`iq*}m=>$!6MHW1WZ-2SDBpyjN2{zssemi_h(WLs$>K`D%7L0@_`>a&UF
z;IFyBO|&@>Z%9S^RJtg}P=KC*9vPFU=}4M6$bg`i$k+z^>7b!on6<A!P`pWtG`qs1
za)n$y>tu|nr>BX|4N<MBP2W<~XGYMrapXQ1OJ}9}SwY-lfT5(FZiz?R+O2r4Og5ig
zxK__Mv3ya6Y}W0wP@-O`vUm7yp!JHYf&@e7tc3dP3BB~P9dbr!^-jy-VSDYo9Q_*B
zRQPeY464n<PMqKni{f!ycf9I?{&U<8Q*4z=t*VVv>1<^X#HTLD&CVWtQOOqP)=c9u
zShP&=4u7>7_yZ~Pg{WW9%}H+au!^byhK3T??FE{c+fMF{R*;Dwnz31(=yCSt=L?yB
zgcMX2oae=XmIN-9@S`weF!=ObDea}K?plAH94*<)+WIm38IL@fFP;UjW|+~k_Jr5_
zA%7EY`U!GR#Kye%0yPs2D)s9&Zz<MoNU4|i)eJFs-qd~}ZTp~mFO%pHa9wU#iC5C@
zy*RVF-GFi-2lxhV4}QUAhn~Mz(3Acor}#R>!XN%v*hoq6<w05hx>e*Ijz|de87e^^
zb;kxoiAys9+>&OkW5s=`ucsW9IG!{iv6R~&(Q8#q3eCmo&vB&(zo{|l4wVO9$T=^M
z62ipF^R;oe&%*z-&M44bp0213cJfqKpDGx0fvPI$x%6dar0%6Yys^^h+gi-`TP8>?
z9ZBVJRT<&4`Xn4}X}0<a0Lb@RLZ3fElsvJ0ld{emH=gBpd@<>cnq|}Ka@?-Qlq16e
ze^Lej<AQl=vhGb+6FGIXXnxf)A+HqfmEr$d>~jJx4BNi}YNpQL(Z542==~Piczmqs
zd0fWR;yltfwz{vru+151P-nvRR^w2m_fZ#Qr|wK+IjM^z_P0`e98ng+9#FMQorf)=
zRmZB!Us9;TQp=Gk%SWGsBeQE`@cVYBjR;WP2@^0MAIL`<iQPs-urJ2&4k%ns=M&C~
z*shDh6Zbns+@7#XvBqP0gDFY6X%3&t)Otk^|C^HY7$6HUB3QQ!0wn1qjoKm*R?jxg
zfIptHL-@~KN))}7-QBOmb~2I@w(IWvmdh6}BL>jg^!x(IoJAw33#M$NlwAPS8E{8L
zzCriUi@0N)3nGx3krk?9G==jlMxSObbSx>gi10%6)1AKwjTqFKZrw#OR}|4NGrQ^M
za>~pS)XdaLEN}`1YRRRx$yEd;J__MHqVM!VV<g^9FK~K?3zza*qhH338|(ftg&JUa
zFF+od00n@@Zi2dkFA^m)ep^I+<0P9!);-~?#OHYt$Oum|3oi3O1Agx;n-L?9X%Rd(
z`@v;Fcv63%`)EBY`1cpf`FFMswR`XO4fL1jfa;#LWHPO>s<v;mfs4}U(Sc>fz<g<8
zzUxX6#~o9Z80L2-2CS)k<S|gETQ_X(P=;-CRQrvhT<NpuV4&<K0$pb8$E?Q`hE>$&
z_#Edu;VQLk{W4RqhIkUnX)jHPGwa)uhA*MoIrA<!F;qn>lDE)@sjZV`S(4)^9}bQ1
z%m2C(&4eHGHC#2UjLG^mK$p^Ty+V1oFn&q;G<0jQ<^?d`ZSiXnX59h%PGZbuIwco=
zFv&!pr24l1+tw>E)ueScRXJ7c(xo1=QXgpn0=Tk+bSy>lHJ=sD^YykuOJW3?5r>dH
zOZOj|eJEflxk_B)T^I5pC@F;Zc8t&bJWIQ&NUyCtf>mZm0y0uoMp|U*mP8>?^D+o|
zUp!dr)@!Nv=7y!eb<TI64{4S)ds5&NOKo9&vMCo>YVJL}$@og(?dEU!txPEz+KF_D
zsG<rBneSA;myekH_AJvKJdhn|t4~$f2KV^QR+Lzx&hr-uK2T_t$y1VWmYk&@sof{7
z=zBVWR5v}#sT`uMrkXk@tfc*=in~W5-<(di6@3k7;9+e|pf!3yegWiDglK$hjc$f~
z!&6Po&rPv0&^8j1f49VU#eFic6}IytdX6ViOxKplMe?+m+*Bd*1#cy+vpSM#Eu)R8
zUevVQ^nikOnyQIesQq!LT<5O-!ToneD&^61)9b?Mh)&4$&~vF(V2jwjJ&T4*T3SRv
zgYD7-nCw8rsSw}DNVG@b3GN!lNr0+G7NfleebOo!qWVU2_&m5*L_d)zucEk8JA1lJ
zD)?&u)=L9{fhN)~>&G+wU_;4f;c9X;k@yekT}Z*$17fbz0%F6w4GV9Z48TAuCK>>6
zX2j+*RTx#CO&3mk_>j=|t?Si?dAK}R)6##f(!=aW+Uu&^?a|<MCde~$%1-*bY(L3`
zEpA1r8qUI}wMdK{kXZ-pk*YVtU##xZYjR`li*vM$V8|T8y*3KJe_R0z7vuxI{IK-o
ziW1}eShErcgh3e=>y1$L(ta_9_{ezuu8?1s!Zj_`w!#$n+w*BxC29(FYMpplY6|n&
zH<KnQP}bki{eKC}M(Th0Y$0-?y|fFCRB;dU%9vT%rMv>&9N&+OMwx}nfdub_JK3zW
zT0Zo_%L|KGF0#L@YsuH{*^vnxSj+=yCz2eog-esNQyPW7lY~Q?Evi-J0oszu1x8gS
zd+2Lm7LP$lZ$o#~83MZPYsy7$f^t(HWg;}6+Ys?=L>9+lY4<I4O0ygecj%@!S=_U-
zTvShGCt`UaD%Bt{bu1u^(gg6$VfKsNl!XCWmu%Ipp_{6lOIwLdGU|K_Is-8YzHf44
z6txQH$wH%*tIE{lJ@JmyK}GzVGJhH&Y`nnG;Xu*WR|h@^6{&e6Z|GZ5i)c#7zEl-~
z&L!86`T4o@Y<d~3Yz4p=F>)Ui-u6jo-PT8V{SZ$0`6<L`GY|cnYpg0*1|y;#7urSh
zJx2b-n}x%NyH`h$G^W6cCiTdMmwQSpHHseMCoj(N|1IJpoKbuce7b5<NF=?|rUV;t
zbE)kr(F>r0V#9#Rr2e!ed>xv{#97^77$|)R8L`05MB0b!N*MD-p7U$$S(Xvf(AP)z
zDmo8iSbm&>ES#Cl*UykFpAniIC_89Nj#`O%n^|lfDA?z{=e!~BG1eaFt6*kk!!#Gh
znRa*6jqCk^AL@+ZL6JB+X7Q%rZ4y0d^+naG?(L2e4a6Q2$7u0;h@u+b@pXArypHGn
z@Na~xWkm`ArWIyrrrQ63L%l5A35IAU0-6ez)>~`bg_&<3A_^RCb?#>+DI2qlqRVYm
z6`HBeF}$FJ-vubFVrt_@8Z!0M=G&D-3p_c5JjhgePnF?C6GD70ejnt7O1*=qvYd3U
z8HNZ)AJ}&xGNlVT`8cfRIU3SP1YD3yY@On{-$RW0%LkR=0PGrzURpj6Ulkn*d7JbZ
z;e+b1-bHE8YbX(@N>IV$9kJ0k2Am>#gnDI(SI(5RXb!_q{_t<5Pwm(+<;kSbm`Y+n
zTVl#q5<dWm44gN<rV}HzWG|rch$PwJd9ETHMeQ70qoqv3fNV$I9D}V3ZM%{~_m4}D
zy#v$eKL~s|;VPN=Vm;b_f^X7hNiNa{jN3PkF}JlWa8Z+B%->7a0S00Mzo`7NX|dw%
zxkR>l&At@5MW!0>JMKJtGSOqH?aWJJJwoQkPP_)jp=A{)*<5J&IX)^oGX=W&ha}Tf
zgOak@#V8!5Y*nHACzCs)AR~)dxRl!_-@Gzs+1(TS$cu3^^i?$?!8t4C=Rmi_TJYyv
zDA9vj4ln&^`4l!fdvF6kDT^Ft2Wps<&`v>H1J1Dgd1iu>*9|IS#%FSqX*w+s38o-s
z;t^W3j)h!qyO3dVfif2PnZ<fou6dsLUwIX!QI&oc#nbfV2|!q3U&THYBfo*^kflxY
z>mWzmi5`~zLlQV>wSNbz_dBUB)j>DF<%2mQaDYo1iGm?A^6?T9k~Ymf)#OX)OXrmj
zi=ugc^8YI)&4NyS$4aPcKh$_PUJIAOro-p^Apv5+x_Ad(ASXB>^XC~g_D_k0iQB%~
z;+pWN^cTqi27C9C@M`F_fX-roBYd=o#)c2vgiw;ymE8j2`<&?V-{1VFB=@X-o@;OD
zW@#@_qWsE|c;|_%aH{LbhRRcGLVP=tigc`j`5)T|EOu5uFX0KwT@mh9!d`LRs4<~&
zocuyPPeUFGF!3k{qBl69Vy!NdGu!ac5ajNjiVj;S{79ll5V*GxU@75reUnbJOg9pm
zDy|=CNLli`c3Xq}_XqB!g-k?w$$f1WbBtQNyhsk(3J3sbt?<EnIicEqqH=nozBWl7
zW^rj=UYK>kw0U?gM@Ml9%g48I@2CGEWf~SAfPJ0C0Z2|Gl6v!bAZbPWx~aDl3LHY{
z2hJn7EbiU$n3I_+i5g}@IBLQ3^LLKRxbMDx6Zx@F0~?`uKQ!du|BB%hYo?Dm?puVE
zG}BBRAGvDpref1Y0ZN%q?=^u!&WKyD?nF~O$d#LoehsX*?jteK$a6BMCE#QA@nm$T
zMt6FF2cWn(%CMN=TSShjhE7X5eZ8+YZK-#0EtA2=S79=}TD?pWz_ZJbKDWF$*=o4g
z)RxdRWXBIE@W~axZ5tVTU+?k@yZS|=a$|Q$2hqS0r@tw%k1$4e{uUwYDn4kop_OT*
zx!(84OY<e4<Y{9C)_p25do^oeaK8ER5tdb)b+s>&cdr4z!|9W_t6}CJOuQ-jI$!_W
z?9K|pU|i4HeIBz(nek;I2l0&s)>))3oz_ijI-4M!Z<}52#7hOGC8hsE!s&Pb=h=BO
zUQ(UDt9gD(u@^L~{d~H~F8t@bg<QBy^7-*y<uanw!ULH2R9%V8ulu0htlwO_yr?7M
z_VK*}0UG17vjF=HzJ8n1{Az*pg?C}=uZ$g`v1TKkOalJtu{yK+@P)YN4MLhb&MN^Z
z_(hD<hvipe0wYp7s1bHrDG~7sp-D+WE?x{sflj(?d(W4YY8t_S3Wa=2wxXq^UX{il
zkR$3A%vvm=65su+%A78ZwvSl7?^bw@X#3BH-;BFo<)z#TvLoFRq$R7&4YfDoa=hz=
zD?&#+@Uny!d0}~m<dRPZg2$ADg;OG0q@v4Zdrr@}w;ImZ1O;oScQ08}W|SF{BA;gd
z8;e{0L&CF=El)@NhvZVy)+vh4!rOm$WTmWVt=o%z8_X`_<M0^NJa7*gH@hV0(an67
z56`X#aHDn4jj8A<$`rhno{4?#C=-W%xD|-v4k(g%IlV69zapra`+GvkTfkjK>ET}p
z?5F!vA61{qD|#!VrowhKX*O7$9PszI&4{WbzbmtWNb&d^oW2h)tYoqJUYqYNK=Z&Z
z<LZGU=05|FcAla%O99*YpEBbQ&gDr_#{~mqW>DkIA~}}6CeDlpP(zT5XpKRd1s0)F
z>NNsZ!HhL2EHD>FgQ6ZM-OE+qIzZm-uncYdHqk`hPrx?=m2nhBZ|vQ(&<)%KH=rxM
z<68p>J>*Zm#t|<$m8m@#QRikY53{j$`rG5N8V2~}Rh+1n7!x_OzUi^FS$<WW_8zC#
zbLP<RI_$b)HfPv5%kbhD<ePDKs+Sc23xixNyvFcHGxS4e8r;$VO<x5k?WkPzmR#>-
zZ4b_|F`+#}rX=NU0=0`{Ga_((fb!I0@pY<AauhV1X|Tp~z+rZWMD4{!<38msL#9Q|
z13t`jm0Fe$#V{GT`bl#(ZV7!Pn!ze>q0aS>_AZ7O{OUtW0;T0fQjH)@b?+VA0wq&{
z!g#T73J_7G;~`eyOlw31Ml=`flwJe0UIvO<Q|{CJflh9?W|6y14>Ps2obcreGtC6!
z>c5o|Q;<bqwXEy+;NvAYK;w$HD2Fa_p+0*!ySJRH|4?V3R5(v(iD6`0o^*e1W@;j6
zb2vsKZVzZ)Bg44RMfDTj<Rky?GcBLJ$BC7|c|u}rB7IW%j_vW_l7|**?I?3Wd(y+8
zHm#EVwj=^P6C=NwBYnJdPzIV9KLjY0upY+`p*!>eV#%MEmjbz&gKl59Fpu=iZ5IIL
z^b&Sk+7&{BcSS$_?GZud$~3ml^6XT#Ay)|kiyt|}Pw$j_xNHBdRTrPQ-l(}ISl~Z$
zWMU{0*J&7Jkzc;-VX{?eFXXWqgGbC=as#8^^;pmBlUXd(Qq45hzdzL-S)~f!fu3jZ
z`K_mZq-@dWXDARe`cA0kz&;%5^m75G)L+z8j^)=06|W03xeFyQAxQW4k3ZIy(j5iT
zxf&mM9{nC>y}ISpt+g3U4;EP|Y=B`(Ju$7jSma+;?ylJ$XwEu+WEM&tTNp^Fv&hKR
zN$Zy%e#)|30w%MkBL}j-BOp7l?Gw-6&mR>3@D6DgU&x&dOcuLG4^9$Is;&5FfnqOq
zsHHy%*s~|8^D~2gCKS;~k45AM$+CpV)(jiWSnCDiEJCvTEgq(H!o@kBu>Vh=6AE}j
z=D+P98i-L9KIgEH)Ie;yJ&u#ErbgDcjJfj$n#&62h3dG`M*rv(g3;oUQJ+HyUBAo4
zNdgi-b!{4Bk1mdm<e&uhpih}?FY4=+6I`L5?+)^gmc#fSbpm!DEskBX-zTEnyE_EH
zTVK$aXB=y#o$Fiy<!P*#&JXk@^chq#bU8--SGNRznH-0Y9d@Il2k8P`qxacIViH3O
zRA=Zp^^Qb4bL&cMCD1?aPG^gNpzpN=!1yuR7&8sOb1<-}8#aHqSQEMQ9crt%wJJI=
zFV#|d5I0Ta>pwc_a>N{wIGya}q=~9pO@(%*Zm8caT=E0)<kLW{3;w?@F9DvQtimp%
z*Il@4SVi#GBgvC{m7{fwi@wz4{<knvl5K$@P)QBVr!~ZoL(JU{ChosY!$;8iykJ2#
z-F8qDONO&a;7VJL)ro1jrM3SNcb7;U)yQH}xIHMcy}O#rPvG<~lvJp-=MdkcyCW@h
zxfe*G^GdIXnm&*oZHK4~e!R8R-C>3X8B|<goNpK1SD<>K8*vf&7^Y{KdV&60?3o4a
zoJRV&*Zfcivd_-Ts5EUcKs|!jJ9qEjgO`YCHf-Z^yZ}eL4?H#k;!Ab8N<S|jW`^YE
z!NexO`6{f1ga#}~(b6om`}QB4(D)I54S&Q}GOSR`9cu~^YE_d*TXn@9B1d&^XpkBu
zX<|n2Jegm_*nMOw6+#!KnIzofEN42Y<6r7wu&>AGVqR~QHhq(*?nNwDFoMJvndvzL
z6d#odPP0N&k|7v#EpVE%Os_ggH;_OSpUSuPFSt6g#9k&Nl~fd;jJ;`T<MR|9K70~D
zA@Qz>Z;VfmUPE6449yp{r<8{pRa1F^o^f{nt^iZ(tqE@kuYOa)TkJjt$A86jTS3Ik
zlRC4yqfsK&>rZ2nVsi@e{~`73PrWo(URH$mmL<XLzINTKj3W%ko+=~!FsgoL$G6A&
zp`|#*d*ZTxMHyWPSc;XUQ^ywTvGr#>3F)&Jc3x7cl0M`=_4~<B6CD50+m96qoZM)R
z?9NSj2ekao(M+=Lt@Bd*f~wuM9*ei<ba*_O0A`I4yHj;559hLLpM4?Fgj|$*Tj%T*
z5kqZwR-N~-?0+&GYAG1Twlb})%aupT<5gw|*~e^HH%!7pkdRzUe&fS<hb=z5(8L3$
zzIk65J;ws~KQOe!8>3C1YbLAM!V+W7FlJ;V>5V#DvP2T|%j)UK#{f=qDYgn!^VTKH
zBkIYgMpY#tJ<5)<XvAd%QHav*BcTuYaOWDrcqEtoQ^^lIR|3KxWeF$3c8C*uzg|J-
zHY;=`Pe%58`}gYbyE%O<VF!Li3NP0ZFZhaiPZmUfs3te3x#vIHoqn*DIMw&z=KG3`
z2AB#>Zuz01y<}`{XzdF163h+OBgH79nwgd;nmaw-i9jrJC|2!ehE)5r_k;ON_y7c?
zpg8Jq?Z1$u+e*6lo|dSt@Jmc3`cSn-X(DQIs~Wt7we)B+0(2;iLMNh?9<NdT9}?`T
z%%GPx4BAdY**Ws*$@n|dMuslygRHf%rI*v=glc59C^}E}9`loMDXjy!E5Z(;;P7D)
zR7DhB&5N{8Ry3SV^vT@SFv%`D$l(m*F~00qxM#%8tw)Ze+j8#_Z)ebe_)U?(dHvB8
zM)O9}F|yIrz9y1t_ZrW_C=P)Fue4#0T}hdBStX<H?cJ2$r8MpJ!icXbs5@WQVaAyr
znqgRzT!(reip+)w99#M2ymbo7BM9Id*vm$le-eM@=}0!u6b;Ma{*<x{NEAErL>2)`
zaUK7HoaK7Gk$xTp+TF?SYS=w&`K3Th{BUr>UNG9QS~L*+Bkp$ji!X?RI0@fa$-B-A
zmYRI^()>pXS3=#oG`&G_Em}&!_*l*4p@<NCF*-KPom1TWuOF#lSH~JZe(5YZ`fqDq
z*3G)0m*Ta|^_V-`l<2^k8z5u>Pa?21r{dPb2LIgJZZK%;r$7!?`B>s5B>#SS#6b#6
z;)~-fvT%YQbJFn95brq|Xv`?9#78F)s{C$8SbJR%W%IH)w41>9jo1c-i-#4R|6@^3
z&I!()3Tmf!>It(K6IvRcS9rzG15d2tzqxVh1E64!n~lbyq%~|$qF66!pV~@z%Uh+d
z`4|h;qD@bhb7hjkqgjf}&so@%U7)JIKO%`WbGvShYkb$m!#E8qUstsaEjNSA+wieF
zGh}hA2?pPNGP$l4x{G-G+@bW9ED~`^tO&93niYD=m18bOkTIM~A67E(?VtQVJY#;}
zwjiZ*cB<~}|9T*xuKFo(pH?v@4Q5H#v`>Ons*j?DtpabB-BE3`8YzVdZkz<^Mg5ji
z%bd%hQ|F|Z76)-h)H3zg2ZWgqh=jt#sz#{S?Zd|8o8v*e2<8-8kM```#s#(Ais&B8
zFEo2rC1vfsuHNycZ4I|rA>BxLLHVcNlN5sdei_xz$a~>G<MN=+lvwh9)+e?|;#^we
zo>k>_!VdN__{fkwDlQFx&R#k9VnmoOb1oh{AWL+0Y&T3VrM7wqZ>2i#Z>ogz^;J$x
zj{JU5_~9&u{7e)Q;h$iKOjz1n@rfre2)(+34+)eTOmkKtQe+^3U|yxQc^IJ)5#`H@
zm*xl`){v4VQWWx(c<c_q<mM@n$q$**vOEhvhXU7u06oaTkB)VjYiL^2T3BH18}e9>
zr$b`B7IN<Fx%y^UJ`O<~^$`M3C>lDM@m%~zt-`UKBYc7;Z%!~XW^JB&y6)|K{?hnU
zROU)nf!6ePLw(pym02H>hB_qD)*~8o+I-9Ly)q4g_1BGesNPEOE?yNPjZ)5(ezOF0
zT6YP48tOJbau!$}vG|9COc0_ki{EPb-l)yQKudEn@b--~@!9DR1<;^A26?46wN>F)
z<_ynLbYeWC(PtLXZ~k(4#;*M(2ijaVi`xVR)=S~mdAU=34Fe~ZQ@wNxUdWLB@K#)*
z`xbTU(V4%WI8Jr1C$B3IKwvXp$~*J!=HD!g+|_a|L{P#`FUSXaFO55+EmO!$JcE(p
zl%ipXR!sbdty+>@z$q^MZ1v<tth0w-*kSXBZLm<JDfM8#UqId7!iJvqUI1nlAKR4Z
zvt?@QD!x@z>2SVQr|V*`Cz&51bKx<r{qJUA8ofvdsG#{RG_-2ONX~XlyncW{-4V!w
z5#A{y&LYnTNOc!cTd49%qY0Q%&yZHy6#{FmNBqDn6XCaojXU-)3Qbum22tl4h07`q
zo{1vn$N!N2-vF8Di7~ONPdB=;IP$w=ANRcZ@U)8QZ@i0cF=@MPeV?~+!dlQFW4pzc
zY11YaHfak>(WcKq2<y1QaoLF_exV6!I%u5{y(~`smFoIU{w()V0IgM|E54Is9bvO6
zVdoninC+fZ51u&f-Y#X72GCHF+pN&yOvWc4R-PVJfLJSt*PYCgo8I>SS;sqcRT5Mu
zHe}k_Ye<H^9M%w^3!6!u0q(ho5F|<LN@^y2drJ{gU%?u8T8QFr@8eS@``KJ`CvB_D
ziNqYjz<3~m9=Ckvpc^Ms09-<JlB|J7QypMW2>*{s%{7i2N?~cevLF;Fh@)Fyo0~So
z|0yP*t_<LLyZuwd<FjSNG<O>(CeO@|V0$#RLv49>mR>~K<$*g6B&jOQNoA72cq$Hd
z#m{24PS$MEVH1wFcL>nMryb#vUDn*GD}O#D{sC_C-HBA_(FwD{E8h`vajBUk)evei
zer8sHv%70(fKR(iVu~vj(4ojZeS$MG$YoT~d$0Nk@Htabm6p3EoArBDmRff5kDH|5
z9x%%$eEfTu$@0$JI?Ag66Giz+BBqOA&Yj-e_IRi6_xIdQM5*M+0JnNZ_KC_2?EMIu
zPNJC_AfA0;b$O%Sy5iVwQ;ThzTx{*CycFsY3;TF-QSGhykdc%^U;n&AM3GSz_zx+l
zOa(J(F?KF|&pt9^)Y2(|#t1al>kvn{LFO5&<GL`K;c6d8@57}#!z3?A?tCb<*VkHq
z5?PRpd8$0&Tny~GXB!TMgnPD-dMU{aM-R?+ov*w3%Mw&?$)E=#!`xKDwhc@91#4J)
z#Ym-{z71KS_@AhZsD+z!mgm*-`w^MjYwv;G>@nM$`Fq?`noRl!3V^nHTKYKpMLu*x
z7MTPzEIX_cD37z5OA0&noA74hKIO7cXjX5o0E_MkCujL)vqL#3w=(FmPZqrE0c78_
z%JO6>)$3?pJQ2#Rte0bF679-bb<gZe6j9P|Td`=^vb#&trUK82-26pWPI2_Q+B*9Q
z+sew0Z2uyxS-o1W4$I>ze+1O~C)S~UTiV^0rJxdw^4v-LI;rIsvYa9c_mtwB%<C@q
zhSMyljq-Xd=bqD0MjdR3vkS_7U5;*Qa{l}eX<cL5_PemvP|0~U<Q4dBTd=i7_0dm1
zx1H)(^H!!4S1HI(@xpYW!!TQHu?wh$x`I7tWx;SoS8gPFn+0W}b;K%yfK+0Qswb}4
zUwuOcskCoDXGnd7iTqORfhqP~1I%TiOymJQy7ly*hvT-}J+qxPO3NLA?|(Fu(Avaw
z=3io(FfK!+wLd6(RN+%b>5rEy4u9UE{OtSH>v8;fZpmsTF}hmXb#oHh+r~0jd@?%i
zOhWZMoTXjn({t3WJ%b>PGI0AvQKyG8OL-JRQ|#y?{ZN8Hjo?#mfd%SD-~pxxq4`}!
zDmof)-NxYP*3TriD(>n5p7l%5Mptk5?GkTw!O#dY&$Q)3sy^im<WU0x5N*NP@^uN_
zP-m%F!bPxWo`RiwwBM0Ux|Pl$J}%}iOYF*STStU_$IAMFMD<P!26)^{p$U-3w6ERU
zthxpON`^aQqZ!tvx7)?Eac;`<9mq}N5Q{|*8kj1sD;(E5u(Bs^hlv*!t{qW^$uwKJ
zIH{u{-3h_WK1@#h)Y*w&C0C6LK&5$g<J6D$v!yXS%F<r=Cz1}8hBM0A=dUe54xL4u
z!yP`T+ThY*rhiDNWUFD~e==(`HQE0@%u=%TzU{NMYRE|xjxN|DX$T#CyEJjLivs?X
z9k)W&0P1I63g3mEamiB>w2t0206;f-)&FTV*<s8cXAvyxgOPtox-pjo6AGKs?pf=y
zM*ycNLW2VL%qDgpTM!M^539C+NFTR6+}$3F4&L4^pve!-_-*bzGlRRGMT1JrU6+`8
zEoR?u$vbjl8mNi<TyLJgDA{a;5+}*#$z05peLe~ww1@9X36!vK`p_TD(9*B-i@LP8
zD8dV)PgO1xO1?xbN^@FJdA6h`8}UlXROP&U{rHEkY|r?XEwtNe4A8!AvRpjCQ11QW
z!f_W^*)crgYjBsLg#FbrA~NT?h^><n{lkWAZFC-_^1X*Oczir$Oqw%Hh@N(g8lU({
z%0Z1s_u2fr{^a7VKNf6$F+Ka?8@&zU$(B4>u9ZpKIh9U{Zjqurw!@+DA)}9tsmN{g
zXP#;woc4hOV=LK7;Fd_a_ZUr^)fL#iDrFWE8AZZe5Fd=yZjT0t;qTXcrhFoSl?QS=
z`v|3GC%;qDEb!g&qw>~L%K6_dbqmG}`GysQ>lfNIZH{~Ti|Kfw*}Nc$mh6O}2kcan
z*)!w?^L_UhN)+C>Q^i9JC`UP=0{J|hEp&`@AgK#B{AkG8dVmfY9H<U@nxQk-d?BjT
zN1v$P8CZ_fsUTx<hNdYmH_B`SnV@f?4pBbS`eWqmgX{F9>ey7m`qif0%9-S+m`hug
zH70~4o=eQ*BybpV2B3+L68>`P*F+<(a<ooj4cQ^fk^ZSf75uvA`gZ3G`cXQM%+t?f
z5Vg#ggjh>nGss6nVrAdDo|wVl9a~Pmhi__2-KO;4)s2%XB06vAPFEDYi}}h;kMT4u
zM{kyu#*xatvPWj*Z!{4Wsa1EZ=!`g&4RAyaEXl2<qerM(WQQ7xKd#0!yI0KZMH6$+
zj}4(5xd(m~UFNP|@ENWRxUqOkd4vGP^BR=}zCW#HW$!r8M*5EKxHsD<_}x)i)lywq
zJDHZSyvW4}kQTs+&&v6#kU8+xZJ~Af%*2wn=nUbCP6WTfd+HOFQ$2#UXrtR-4;XO3
zcZ3@T&<+MUHZM(R_t09=zf(&BICT=16q?r>jCOk*?tIUTSUd-4nqTy;nDNjpC!Z`=
zIB&%!e|H@W^bGHB;nNVCP6rEHJxW+&-x^oT#G&d1dz2Sm3(!>C5Ef`4!grV^#m2ul
zW501b<!!sE^b!StZBn*0xFm{d_-fRS!K~-BNDt_TA^RQ_E=rZdgELOMoAnDl&H4zf
z@z72xOd|T4->XCajEW3#*uLS=f&TJJTX0Qua9wTDWjyE$09U}+Xu5u4tXj78Et9u*
zHe0CGmtUT1Osfo7zR<SM%!TF86Z&e#`-{M$OE37Oh)wxkH)~y9rj@WM^-EspNUSe{
z7V#}fZocx<Ue-5IH$<-xL<1H|*^w$N5hLv@cX!9X*ik|Q8eVwXoEg6xrU<@GvL%AG
zXw|eynlk%TS>&ylkIt+<HkI-lR0H7oUx^VB^F)^$P%iw#PlQXFGax(=MbAXcQCKGL
zUKnJ0NBJUEjObv_LtJR~tXBWGLegs*kL;-p%?X$kZD0W3T!0gHdEtx-YLH@LN@vdU
zin8!BdxdXhwNuB2x=spZ>IS7#kWQ=f{(f&&mHnMFkR`>$s&JmwZqKSxLA|9;PMU}l
z?&2yXcJTc`gq$nR@RRIu_!x=s7RRq)<C^p`bca%?DQ@9Zv*14@%mXZKJ}p@fJQPnH
z4R5PF$@M~85~+TXYu;PC`9Oo#+JhK@UogZ@;s>!)9oO2DzfXIWqzB>oN3MT9zQ55b
zIBlUUPzmk!Gn}DYS%fZ#JMF7{93uL?ImSe8(o-i;PzuC*Ejw6ss);LkbBXT&3uc~x
z?=)&HJN^j?&QC>#Ow}9GstW8hK4d~>$c^OEzWGRTQxBMLPCMzhxJ~uncQ<MKnkw<6
zQ%u8sSq99;p|Zrapd;#qLePOHC?Rp{N0@7&;|j~P_kM75<`di`*PeyHWpHPDa|j<H
zA)vjD#3cZk4EOMRJ8yYB{wQ*_>xyI$M}+Kgd#g0sWSb{W9j*|DVpAD~PX)$-bzW8J
zdG9bmb)SSbM|YdeoB3$8K}%QV&+0$MhFLOo(xgb-qA935jLi47Mc!vRpj_}$iVj*#
zfQ=2wvsK#8h2n=CP?z)#PiX)VZ}Qj*#!{_ls89Nd3$qi)?=#|s`CFyq_TG4O7+?Gy
zaqukoli?lWWf4nu89AAtYvK%K7<rRAEV4x>x20|fR!cN|dj+W2ihKUENy8<&t6y-|
zE8$?OFHKvXGEkhZs{QdD&$Y1pM-&WI*|)YHBN{XqGU8=Mu0Hz|spkFXVw~>s*Kf!A
z9rRI6tj%U}gGxT-BgNQWNw})N;JZ+8W&D`>0V5u&ml^kaI1ERD*DQ5rb#g;a0STAh
zwCzW|Q4imOg(o@;Q_Q(dMm}#I-JUh^p|8EbmhY--^Ceemn4TH;h6~PVNJf4JG{k7{
zdowvxwYujIaAGGPYW-Jv9CsH@sI$&U1STG9Gi4kb#<sMkI>k3%;OUdO0?hU<Zh)={
z_P%HC(ekOz=9}Uh`jc2d)scY?6ragn05H+u_|Kty?ar#Rv=Kjm!d9<pIUip{uqTs7
z*^V@?2zJ!(=4ET1Fro|V%ayD*5K-{z*jzTorLV;<a!+NBc_nM=)C1Or7i=mB^mma5
ziuuSMvp{5jUzQ7*Z;`_jixEI=DZyBeXP@Yr)ipd#MGvBuebiX^@G1uPUd8t1U?;cw
zerLC70IG#@NtR@%_U#%#B}eIL)BQ)+Y<fD;i#j8tF8Q_u{j%}$s$zu^gj7w{ncY#&
zfeXvE3kA1uf2U-;DXT^xUmcf@c<WRiOGN4(qu58d26i?b_QMY&`S<hf2aD)HgClul
zewF`_`m7@0Lc|C{Mq?|o!zhp6^<f^iOp>yiKw|_nL~5<`jmRAK6Y@W#N>xhKtoujG
z+rEED`t^yzZf^~_=Zc)x?g=$GR&5xaZbcuiCzdVPSN3ekdnRjZj7a?=;LV><ave2F
zw^>p}NYBbNEd~F;i3AdR$NJy}=e$9_Grn=}v$N>22WGRN;Lm|Kna_g|nc&g1Vi`x<
zpa8Cmqdv>fUSzyQA50?^ZPw$iq0D?Oe<1zAX7v>>K0v#AN8Ex4>?XV1u*ZBr2Ey1D
zSSH))Wv+xGeNHBOYKJ$JJkdI+zHr^A;*l}s*IExk@4(b^r)F$WEx9*k0;Ra`Z7HL4
z&N9mTx?LlU?j>-Yt|~!k%`(YqlB1Uf@b8n7hU@+Z9iFpU29|LLyfJQ{sPHbZ#dK7`
zCG?Wc1^Bjy`2*``tIO!UjTWXAA;U~Z&J?)Qyzk2AuI>~31fb+OwdW9|w{q`X#_q2x
zJXgC+6Y5G`ylS;>PvcPCtA+E`*uwXoi2OFJSFij3A#vzT-c4lE+^GtZ8*hpuwzF#W
zG5z@BUM%<p(`WPS{I)#s4NhK+CzRyvTh+CLe1BjFY!vK`!b%wzDo<xcV6G3V-a4Uz
zmO9h8<3osFJ4I+tr5vJs0FNiLfsz}nl2Rr?t2ymUW~R!aGx$1FhP*5her0e36c;p0
zkbi9I*kj>mA?XCBN`K+pP9s+)Ae)Pf>!Vdw+B9}Zaz)AH>Twal6|C4*+7LaIS&&p-
zGo=j6vTb6&E@oSFL=}qF;82~@SQB_~E*<vugUYjIWTX!;1T7jl=3#O5?=%$}YDdK6
z<P_G^6CLVhgo((TRWOR19tS+nKg>*kewiO|&uUC%552lTBoAvSMLT;Dl{jM#LygyZ
zbq9tYifP{XXS-$_R-wvPaP^G$He2`4VMH}K%~8r*8Sz8agnS>jz6+!~g~Ow{=znWQ
zzkj6*bL9;fu(CgHMgK?!Qm0F@6}B>eqV~%8{FBB^nBA(e{<S0un81d3O5w+`H8#yp
z;76{%9rykfE;P<E4ke01Q7KPWXx|_^7yQVSiD*a~owNA$eDKJ&0g58--Cryp&>Ab1
z1Sy=B)W3C+_gGq-#8E|yh3JQ**P}SYKM4OBu;+kR9Ka9V32m|o$jd@*Qrjx=3hzuj
z15TW8>{6Ak(XK0{Djt`1!mA@Tpq|3WGI5(UBX&vb%ARLHiEm!}^YdYGGPKoO@nUir
zKGE?en2}!`a6MYyJg?LZg*oRXbd$=E1*kjmEE1y#2nv+a)yl6rqmqG@i!9(g#rU*T
znT@Y=L8h@g()kYB(Sf3asUCnj!chgP&Umc?{Gr?Vp+A*H+Bs$!5)BOS8e1V?;TyW5
zfg1woL^AC!?650D4ImrQ<@1zwlILhTHI~*NyBmvTa&Gp^^g33a_U~vItG6SP-cn}c
zX#D#<pN&1=zyAhwToHt_j@0gcLyw;ZQ&JLe@^%S!_GSjSB8b0kPC+Zkl$NKJ*P7Ik
zDVq$DCx9sV(sU(7sn3>0S%!W%hZKsMgcIab@}7)uT6ES?5jfj?X3+a=_K$)1XQJ}(
z1dks^qm`d6-fyr3Pdx$HmcDbnF2g$MZ`EL9o`-Qy?xqky&a<{&4vI(pNK&Vi<bs)C
zgwbx@zSL=o=a?hV-v;z==0kRA-3sU3AJnaN7tkn=$ovGtz5&hcNTB4)bv;+^*K$++
z4fb!%gkSBe=H-Wz@L&}EmU0+Pi{6zqO?hrrBj#iwjKXr=er+J18493lA}%#8Gb(8C
zw(v*1t>C(!&2K0|ntz=4jb9#AE)U$NYKNIN|E54NNu5R1^nH}0GK0Sa5c4!!H@GPp
ziEpKoxkCRTRqr>GQQIiFe3>yMKWt9;TrtWjWr8xyX3*;f_Wx^~W!$j$IxA&XG=8X^
zRY?vo?Y(SSkw-cU*?;<y+u8z?!}qX9v^%2s>|8%FT~UybUE>N^v4~lHME<;0+fX8G
zsYYd}?s}*TT+UKKpxSB>eMaNzw`Mp-)3f4}Z*bwI%+aWT7ueuv9q9LUd0B6VWmidd
zvrk;#<Hef|zOq8Z9+=GmceuPHhEsS6i@oLVa0;}4NU^q}4}1TRMw>})^AN>4L4eUi
zw~t&(+>e}+F}F*8Z*vhap5aFzE}sd9@c>PFZDu1hEy5*<1J8lXDo)r6l^m0lFGI#|
z;bY|Y^ZOa$@24#R$tS;29j^_UmU?zZtG_62G-q(oO>U@r*Rf6exz-e4RjYsj#5iX%
z6653bo0Cnct*(?9hNM9W7zF$qu<IeI_Bt(|pWEw)vVuK%0uUwE%S!zHGF<4iRiwQY
z3EUp(zRF7Wak^9gX!NnGmX+jFVpb^KpNz86Qy1yj<!|3{J)i_E1x~3NwTaQQ9U?R)
z^*r%B+Y<NaCtu2Q+x-Rp2OmM;z7f9}RvqdluXvQ7BU45k8{nMBAj2u?TUITZ<L~oV
zpKD0<)4VQpGu<pvgggf6Q`){%rh1C>V(!lxq-`6NFGrU*pfyakO?mYdrqS!seALqb
zrdJ3$nzg80EuE3U#~2lK^ox;|e|m9==Tgkqm#-=06UWlDpiBW+ooK%j^r-b~#nbe;
zt{CHL;C*OH?!#`%=M$v(b@eIj@74DQea^l8YiC$sF0Ii0<=jZGYLmJ?c7Mc27pcPY
zYVWZN5;(xEeHX#{rTE8^V@&YOD=M`VPP$awncV5$4&aU?7cv)N$;TD92ZD5%@J+fg
z9muXHDM;<KVuI(vI&H$R5d|l>6%D_E(#FB#wvdCCBQ>QNN-XEZsLBa>oF<{*{X)pe
zBis48B$BnxO{L4dzCx$aiXl^0+0#5fG*XG!S?TRRgv1AuVj{Vl3tK2b7g_1WZB}l|
z$2J;*=9f1@9a7re)mqqdo@#*8<lCH6RT4$4E1PyHNgGDwFy_54;g!G^G6KN@Nu1Tq
zu3Mb&rzbvJYI<txdVG6BPNA{L&0Bp+*_3?ao|IBB#8Z-zi`>sTrI2>apd9q6W4(%V
zgpgGGMA_Rc6<9{?6ntD(FAQ8`oUM5ti>GhbMarBiDPliL-jcd!6<25!yk~G^)Y1^Z
ztZDfZT;Vj{=5Y8>%;voUc?A9qcmDvVf8|aj)aZHfrnLpJ6ybsQuS4-eXT$Ix_Luon
z=w|*Tb9NV72vT^igG?-!k<GaLt&@t&sjow<*R6ihr!2~(l^}!cYn_S=Fve-iaw+^J
z=+grywRD<XzHOS0>%p$P8~5x~wl2|LshCkr1g4r`FxL~}@yR~*y*;JqMD{tT`cuzZ
z<B_A-N@>p(h|J6X0QmSnL04dl%C`!7o=tg`J~muPrf?+)Jq>Nx@nskLwS<dBpxaco
zaaof&z0@O&{nR-$k!HgE+J^Cfh%a(;P`3iMoV(_dZo}mjepd+_A8Kz|0y83CHt`p-
z{Kl={4c4x{wB_*&p@Vk9J7PPr#WzU+`Ng=x@miU^#gy;>^rk4Jz?nzNDwX-<4@#A}
zPh+C+rk}DScH@@ebLm@7zrHo+SC=%|-9|QoEx<UdvDX{i*u80aEwZqT0uQLE&Z~3%
z)%>ey)|6F|gs8$!>}k}m8}7yr{06yC5b9d=I)bzod*fk(c4oUU^(QGD^{G>gcPrWI
zo*{V(wU}S3R=|CA*Km#n{{VZ6-fnl$$}yyttjL-COwzFnX^ipBNp8CH{{XI%+CQS^
zx|Qs5K1=9Kms-Z3samkvJWR)XEO^`c(>!0|E9*$@-ZhUbNI6<<w1%S@G;?X=ElB?W
z3_thHL3^v*YSJT45NzqRV!7%z(dl7n#Yr*()ZNw1W7uq6)pt+jiKQi>jaPdckRAnI
zLs&BdtI?181a!@5Mk9Xbir#DYM-4qI7*n<ZsO;aTv93h4JpxT;An6q0du=si!SSiq
z2_x>tNc_cj(0d+sBS^^SJ||y!-XU9MZXt{UJ<VXjbG`-j0~P2)bDBQ_I_c+WydKiG
zQxEm73Q03gOo(+S(eI6(s|>k2cVU3;4R+V|qD?~1@c_y;);-8QD+-#^Wk*V}W!%<n
z+h$7fW4Z03u{G0pC}&cUy(2%&S4}UC8Dd}Vq;=Zzv-o=9Q|}eP&3M~*V^S_o*vTjR
z_35bVbH$}K7}@tFPYmjtvHP$+o}`n*I*{AMeK3A{iNW0IjcC}d7lgGDbf*z*&-d%0
zZwlNP@v=TgKQ&`gGP*7`U|`<(GgF9_RfwT%f?K6t(=2tH2-X<{Z5ps-A^B5VD7K8<
zSlsjo^>>~qE+i@^SjxBuy>NPtjj3Pi@?4ONyyGuXILWf%RuL;NhP4z%eX(wCpEp5N
zo5EV0q_YV6{{ViHgBz^@(s*A|jY7t;7WLoKw7fZ^Ug=j-OLMafm>XL>)y6FuPOMt8
zF#Jmoovh01_|*$_A-<IoX7tvHl4Bd{s2P*-sf#U|cGJlOPa2XIL53c+S&kZ}F6`d5
zK2(}CH1@=*ee1K++jIl#S{(0Vps@b{R%-C3X3nci1N$}(KZp9&y4`GZy~U!Kgvj-z
z-T2TZHJO0yD}nKi@7gA4#JiCP%twD}Xo)k;W`V9{a;i@}5!*G>Tv@H2mkKs`86(u1
z<fv~G&$V|mUr5}1yjD;`7$17tI}-OgpA1@=p^{8+4R!Wh01r`IHJ`lBx_cME;;P(y
zyyCeH(ME5bPTF;zpXS?2NaTvqI|-a-t*5nx!EL;H)^6i2>>fej*QridiRaW-ky)Dr
zbIy6IR<o#+X*fNNYW5`?cG}a#y9*MXK9#d;s4GSkw&ia1!&W^`ntGP>Z9%ku5J7nm
zC0T;79l@_uyVo8IHn?O|P&nt4U9`REqmHZOm5&wicZu%x>-<RMM-d?fWNGlUml1$6
z#fCFZOmW!nHJv_7y%uHMP+(^rYs_!#{EOjktZ~OvQ01+HlJ`$Q@Xm`0%#s8Es*L?{
zU66@G1p|O8R>tJP02~VDJXNTy3FRpqb>^blmfex(x3iaR*(9}RMFE~&pz=>z(8$q_
z)n=VS50N|KwbgbY1GQ2;d>kzsjn6cw2b#a{3J%Ay*x#r+Y{UMUpXFQ;$S2aAjvOz-
z$*mlHswO=*4A*1gglMd7ZK2CuU66>Q)4BRqGNh*UIq`Gkk|VT&$4k?UMB>IL3hX}h
z<h|5?vMW+~Cuk(T#>wYCO{$OYALs>cwmgTcM0)j;-`K^Y!oiw2g0b{sermQ|G`Q~W
z%PEgcvs-xc(7xWI`BP(o<Fm}u&hoh>oRUeY^o>4EPgS<EjC|8b6njz7v(&s~46CI0
zM#mwgE4%~I2e0#^$NI|nddp9+vhu|00c;M6KgXKNQBqbeEus$%_?0!pXv|EL#<_3?
zN7lD;2==c!aB+4!AtfYp+Ny~44mz^r(=GjKUdKdsI#_;PFQ4w!-5Vk%J&EGFA^b@C
z95p|MFH=8J1YJl^@f?cFx#dl8RQTE7OIRZ+UlPlu*_$g^!ph89rSu1;E1RC;TbWF=
z%F;PJmN~CVm%xvi$5d{#qu64BocE_^J!@ontkBVP*)?rS8#u;ypP2d%_33k@HmM$+
zY@uU?MnE5xDvr=+sh7iEVm#m)&W2srE}ipSJx_HTmPv>~J?SqG-CxGK-jF$?tx~ly
z&2mWWMOkYy<#)=f>T4dRoeSfI1JF_)Yh-Y-q*5MeB#KjkRBjuNwNVo)_#gg7xc>m_
zqyGRp=xg1N!hvidtZ4S8vv$pm%!6$y7Ktol+|Mx_vB%|2yWD4JlYr~V^r+}xQ;WWg
z8|Wh^hCKdevabpAcPQK<Z_>D);8&T*b0lXtV8GXDrC&j+*nx~loG`5FEiMV!m)0Pl
zC>#!gr2`ySD>p#P#cSzu3yVgHfXa=vRQ@XyTj9CRYPx#dM=i92BP3S~sCa7b+CpLZ
z(yEP4SIW&Q!<sioO>1R!ZQ$2dQY2Vd*PqF4GIG(d{42f6J04_W^<~Wx?d&xakb)Js
zYV~a^RPq9=w&+{*u3FEQYDZl-^FhScyR%Eg62?H2A1e;@-6T3*gmuACb68Y|zL~um
zm!&Aag{@j*a~L=(0TrWbGYI2idB8Pre)}S<SF?!iO-saCdE;=bg_9$#VTe|l(Sw!g
z*u7wx-79?sbG|Ipw_*s}lf@58X?F<w)!Xc1>GrJTU}V;^LdbB#6@k`Uh+??kh#Had
z;;@YM`JT0)uTdG-m8r5%0p3X*@j)zXs-U=I(2DA{UkSFYs3qKT`EiuSKN|Eoy_{T;
z<Wqv3TQgfr@P?mbhcK(j-H7J5qqRuJ5kik)Tx~Y)&r0blB2|wKQZ~^|EeLMxKzSQ^
zs|0|Ut!e0+Ijd_Zq6fE))P_(&u76kXou%$wV+#JDS3IND9hB;|p-R_7n$J#>7EQnw
z#yv4ynbT|~K1`8$a*hr~Yd7wujH$`0k}JQ5qruq1v#&IZh%m|XFSTXOsoNTysL8m~
zE+^y3IrYVAcx^5;$(r75s>*T&Y~^@ZC^)Q4WOWk6^6hxZ+z(G$>LP5$p|TDIZHATO
zW1Bjs$z4fxF^i;^5~=&j#;e`kc>`=ui4URaSSb7UJ&9s1gj4b%)V0f*yghMkI~V8!
z8l&N>a@JE!SC$|t^%dz+S~KM_3eiuwz-reYYSk_!!2z7`dsPA|M3#eVxwgsh8SZ?R
z{$`pYeGU-PQbjQb-XB`@FA*m1536=fr}V1mXD^8yV#&0E4Rv}Y*X{%5OoZW7)>1~N
zq3O#d_L35Eh4$?uHP6Xz&`Qe6G7dPQ?Q=M2+8nA1=RNC7N0V!`XQnu<@Xpmyq_#El
zMSDg;MKBQ4O)wg3it%{h*3|Zwp)%O!V#rZWHOR~8C$%BzSd@xpA3sw??ZGu}Y$LH8
z1?gU0UyYhYXl4fAbNSarW&Z#kaeup6NNBS;SnZ29rBTxpaguUA@xZC*QK+2YRV(>$
zSm*AJDr30G>|*k*l}}PKX)j+2tLsA6LC}W5a%7j)us(vFs6Xy@j;5d54u?OfH_nld
z;T+Q}$>l-D<64<L$)x+o6)n65W(qo!NsSt5q&fa7%ha^yOZaW`{;-Z~ILQ>z*;&M}
zK#{1xQHqvC`GK!4PVDqztcjxSTKY^YEzl03mj^kl;IA}=QQ8fgo>ja@<yBfnxVB!d
zg1i05o^7WghVImV%S5hwUDoU^XWJyFESUjDdRCOGN1Dh|r8{bj%O4Zz_awS`xL&^{
zOMBw0C@o%4vK$ULtz}VjSR+DQw>g&a!iGC+$sW7uKjAkg?%Cu12VES-9?0r+yRWv|
zTiiPB5`e^dR%eav5NYhq);Ak|hMc|;Q_`i6JyU~LqHGH1mdB|<+C$8g-!xWAEbMhP
z4={SxrkZMMD@2ZdW^O`wtC5=J8P(|Tw@p?(22D$<6_ZX9W6+A}f5Z76a?-8*&R*&8
z4Mk-C0C-m_*z`?S0&6Ej&IUza@P)ehmlKimC<;D@{{YonVcX|O;>9adZ1bNV%(D2h
z>6fPi@)d(`E(u!n;?6~)JIy9lxbWDF@)Zi2^Ve^gzO`~PhU{oP>(YERa=MP4ZV4NQ
z$lk`YYU!e-qfEBosbxG6YcA!m_2R`?_SU6hb6r=02l`yi=#{-6;v*ba@LACKt5k<g
z@a&UKyJ%%of$D3|f8j3E$MDJj0DRZ1nc_kXp~>R=AMVIM;59RNuEeHsCPDt~5@^;m
zq(lDz33Y<XK3NAp-L973U$vSQAzXqzE0(Q9jblPg*tUw^`pC1Z1Jr#gq;}B6NeW#7
z!3?<-ymg7n(^9cs>e>kM(X%Kgr&@|MAwW^Mp1jb?USwC3yes~tpV@h;M^k9W&1rB?
zvQ%8rM*T`sc!n69`IghK)hY>mTVyZ{Q-9wTT*y6>IqU0$yt|g>FhZW07^o8kcVh(C
zB;Kcd6{3d8g#^`I*l~)e>>W)d0LA!J5)ASy7`sUHs>hm{X36m?eQUPUn=z5^T_=|I
zIx7$LRdiD`XIG`S{7Yc#_kXQh=C@lM?{Fac(iI1-F#sPyNwYmkphZ1e=ve8}NIK&K
zKK14Lq=n>YMo#i~5$jdZ&HO|3-4;7^6WlMBi}PW+vG=N)-lh$-w=MESgK7^_Ylcqp
zNZo9C%-NPkY~-~<@#p0Og&lqCr(q+y@THlLN({pWPHTENLO912<<xhZJ7BDlV{aL#
zrH{-)D-%>@va~Kku0>q7x(c{ainJzV_=?zFKz0F~WY-E4%ADk26J7Xsp~+bh%@6NG
zFLTnMnOZm8x#0d)*U;xflJ9q3Jw;!!zGQ8oiC;>_YUtVy%-y%WXN+L{z~om`;;$2h
z(5(EYCV67odEgP$;+`GZG-rs7+Gn5IV(MY$90AGCdg=5F19z%G(WuyslY!ctnKsb$
z3%gL(5{<vWYshcy+HG_Yy%|CO06&F7_==~o>3$!Ci*q65vEsJ^g<P@cxvJ=EZA{%$
zTC}i*)Jm@|I8bZNHB0utx^2XjCnkh~W<AgEWb{xlDqCbYF{1K$W9TZ0O76zA5Qx{)
zH7ivgK~6ldNM>rygkpQ>W9-f;4thnit-ctrpV~J5Rp;;bvrb1A6X9bceUG0*U4^CW
z+STN`w5oAxi)(Ed{_}8HeJLh$W7f>B;Om|;w9+jc3q3!Rj)afA-(1&E;vG)kNYhTL
zK+IspWf|$8?|o{z9!vOx!T`tN(W>2&(@hyGo=NNdaabNUyZ+Ab&9<4iiLV^oPN%5J
z-H*WWR1tfgch<cR;f|9WBIm>xU@TVZRT;VtIW#jvf5O1(TBf1n3-T5a86ey`BKQ2O
z&OB4CHodOTcyq9wv4h!1N|B?`JPBh5hkPq@t=_22%3B0;v~(Yhb9Np&o-5ZHlz+LE
zw<7_Ks(NwhSX75PG=#ZXv8uXFbq&mIxg?BW`qD+lD}qYe9*sA2bk0KQ{+Ur$^vuNq
zXWZ99{vz}^W&AhlQ`G<tAGsBeIpEg~W$&}Thro_QS>pFT)kJ+O(53Jr<#FHd67?QA
z%_zzDt&zlfH-WVH^vxRMTDbvtDcVnE6loWdU)xEjNuTtY3#jy7!lJg3J^U1TVv8*(
z8T6{Q3FchMAn473TyfQ&p2+D@cA#FE6;|l(mmY$;>S+^>j<YKgJ5wYh70AxyRg{oN
zraIPpjrU3HYd)rr;bqfF{HsHOh<K*4;+Trkw-ui|<kiIyB0mFv$h9B+bbsev6?<{`
zQA=SlSlZ3A#>;yMJEP%%?kP`HX&uWNWU<X?84uQp*KBUUeBD8*TMe8BqbrM>cLV0$
zc>HOPXx7O)NI8)XL9QXQI@Z{}n$vd9-^Ql!#qZhme=VB}C(HQKlDtT6$8^~f40NfA
z_Z7y`(H0@;R<wzHyV9iKNIzPb=_FM3gC*6i_S^>phZUIct~mhIR92HVqVEK0YIf!H
zeUYdIkS|K)E_68(7H2@fdsS7Pq>+UfT{q@axzb?>F&NH0D*l%?Ve(k~$;Bkw+||V<
zZt1$4TCk2nOh@#tuEr?tGcsVdIAPwpY5UqUlC|LZms1Z?na%sg!vr0*uCj=tk8#Fu
zwUjiJn>VejAd#S-y|yZOEBvbs;bgio4(Ez73h6Om<;vuX_j4_ci8;mw1#q+Oyt*!V
z$>OrDx<>eAbmiRX^sO*k+lF=dm{ad9z#D+94#Jzd!wX2q0Y*)EKCgatEmC=zvI}8H
zu<2b`Rqr-(V;!QA-*^c=*Qer0@?-fy$7=5i$^Z+(pct)cd_Oav4w9(b>|C~$oB+p~
zwgqOjIwUI{wVkQlm}3qP@luwEQPqsoZ(2ak*}ymzWa<d=?G4k4yYkaBQRb}tiPnF5
zon2g1rc|S0>l%=F1Xex&0Eeva8H+>%6Vz~OsxNU^#+0p6Mnv8dWzUl#$Dyp-uMgiv
z6Oe)MD-Ne)vsxGz65LvlO|En3DiyG}y}yRy=}oLquqSsJ2mb)oT``53&Z;=RWzh6U
zZZr#6&?!g29D$tF_mkQdRz={CO6sPfwmET?s#R&s<~0k8dt1pQDTMAotl1)p<e8zA
zI6U>K<y_60s?u~-K8UHV+evAn{_#q*t_b>7yCL=;4hG$W<(3EB*G3;_>Sv8fC?(kQ
zJNS%uZ|KK0P}H?%WH62^ztIt9@SJg;hBgQDG~#7bW6$Imay<<)glHYbdJeRF&k8$#
z(=|0n#(SfMw^=7(h^3FUb+)Wvju@?U)-shDl1`PJjgi=RTH^jEk{io(AdSnk)Y_X_
zM|jDXf#2rCbTmJQLKc+@4Kk}?NX1&yW+LV?4+P-TqU|)z8PMmNTK9}aG?c&%G}8fw
zxZf8HsjaE)FG6Lp%BKs8cT86tjXDQ<Ls<1GY)NSVStOi}NWe8({#8)i^{+Cg#?2PS
zWCs#t`WoqU@0!vd_iH%{BQ-g*AC#w(aaL{P5kiURD>i6ZM%tyl1!hb8>!{=PIW&OP
z1E3X3{r+|%qKpr%I0GiOkIJ`1?guqe@%K&_vWgbq_Bq>`Jj;h~Q&U>58+V}1YGgd}
zYe!8n$0%nw7|lesDl1cEZ(twCRUgZ}*C2G{Lm4OYu6FF1x>}=lAZlVO&u3{QVz=~2
z@e|4FaZZZ8LmKb*Y>!rN^nf$kvze<pzZ)u&_Ylp(xqlZ&{-R($?nP;bcBqF_;bv!t
z@MO79;UF-m{6zj#QN+K*a%;bO$2I=|baV8%!LGBA^qm@2k3h4L+h`Gtn%&myZSQO%
zGN6$eKw<oiYPLBWYF9ZKyd7fTm|f#z)0~RWx$uqR5A<o0U-!ORhYOul@h&D!#)YcD
z#5J=&x^gOzw7O>~u*g_;#Z9?ORUFxBJr_}R+#oflX#}>!EYY%@cNNW6ace`aj3pPT
zu?ff(dOnrLbsaX8f8tVpjaG|!+9^26^7#BKr=`4)J^W4ioPF!#H4TgW&GfEyvFICC
zZ@6vs6}{m_0t=FQmKCi9`^0f#{{V!~B=LY{@fFE#?s56n7;UE_yXpKW=9cGg;UZ<x
zW?b@IkJ7liq%hsf5$cXd>sq~xWAQR9*H_?M@eTW0yBs+Tquh0=E?)`T>d<*t1%UZv
z85Pc4{{XD7B1+O{tx4V~kY}3f{0hH4%X(t8q5MQ=i~b8H_>s?q<pchDq5W&f1$v0`
zseB`W#YnBY<@KUl8^>}v#{s=7s_^~e&*a+$$<HU+xT?dOL!nERHhNv6<}w$LlZwfM
zTAITCdwaN1qjbvWjGq4h(z<WTju=HvyO}!oi2lbcFvo7Ec-%leDzrW(3Wb{2AMh&4
z)jEE!rsLDZ_D-ASS4?|mqhAwg(y{&Hf_+yNahhf>jFB+aG{+m7<{|#^pj_$G@RCB{
z{VMK^oMeecm!`<SHM}P@lIR+1t`^-F{o2XJGgUdRr!C@L80pTkL_cuY!Ol)=pMY<;
z=B2Jw&X`M@%)SoB+flIE<bGAmdkNkMd<Sfx``2xyHt2!&70{lCmo;Z&rm+72QYz@G
zFlSq(f5<Wq_kXQhu7$DA_X<j47NefwE#qk<I92OF9Ol1%m(WKYw16C)#EzU-HWI^3
zMv_SkvW`nJ$^0s1Ei6J;cQUmNNv;RX<|I9jQY(Vjb#r-gwy{aN2#+`&O<@S$S2cSS
zf#Au5v<we(Q)$*udv_ctGq~i|lXgvx3schVEf(I{B2WNPikyH5K9%HN?CnFyQoV7C
zi*`v=!8F<}43^i-2P2xQE%xWN34_>%;y*SbpbXa|tZD1!vm)Vok?CDjWoC2KS4KR4
zB<iou4;3ZVxOH+D@~=fb4l^(8uF_#wkydS@k~xuBV6oz&Mv>gwiCQ8(DweHq&2zAl
zLa!dxg*U0KMWGdj`K}f<V_vD@D?cx50h}*dbAIwhHuqaP$)nAykM>P@0>vYEig<tx
znfUxa{dG%mQrPZf)+}^OM2(w#p>gG>TIa9)M}Mep3n79N{0qVS>q{Pu3ODzei*YHA
zGBfixdZ5K)L+ah=IVwBJl=csqobWmN`&HdF3&dH6E!5DBFwo|xjSz9#khM|njop#&
zS9E19s=!BP$)l;APl26f_ck;C0M|DDRp;^!O%gdUp9>&ihRt=_4cft{LJNfuq}N&h
z0B8&W`qZNga`zDLhc>25Tm4kv&37<`eN>Z@N9R@iYcp$}EVa;WTlY5PMgsGL+>z~9
z(D9eVb~d6bok|TR+lVYz%XJ(MF`wyPYva90YCb5r+Qi2R@{}IMct5RJM|NS^TS<F)
z6|`Kujud`1>l(GA!D*#<F2^d@3Jt?4=$ZMCxF<B3p-<v2vmTMBLE)<p-GLY~?d-jW
z<6L)xd_O0MwaBgZ{^BwoX7pM>cb+qt>bjMmgmkZ!IP><51OuS_2Q`o5?*;ibQEGZ&
z-wZK`-Y|23JBosg4-4vd7cqT;M%^91%$e#l+v#0OZ~(79zr=bpJ_DnipRgysD(X0f
z)k|@O6?e3o@;REkQu8b7T-&Q-y<pC84|?RS@tt^ofaJB4{+sQJqnw;qeJ_C@F^>NL
zc$p&pwQo<ed)w%3qZxS^E-PelMO~zKUlA|YMew{DO|xxjKtqr2fzRhr_+Ot4S#kdW
zEQMuX;XaOnvmJLKj2`uWNVmXDjQ2IoNosb@+AP`!<|ERm&e@Txrl!X=1@9H`BTxo$
z?@`Ed4RSNN6<ep$si)0&J%Oy&<CFJUY|<Lk;2KRuB!LAx0ZM7|c<EW!R}n)Ql2;0&
zxvgT1o~)a|-|{WT{{USd`PW5WynYmDwhCjQkxH=0G|Dnk+3aqLLOw?13bbE(6X{U2
zRxO@;dnWxs3vj6+{&c<}yN+4WV96>l*za6-+@?ijBRU+3oHEOhE0~f-hT)_}+RE7k
zQ!|m?_;<t+>ad8Vz_>xr`(E|XA;_*uip;p|DH$A6N4hCeMK~&RSTl6=v9+j2=h%rC
zoP5Hu8Gv-pO~RdJ?_79!dyN{Jg~CT^w6+5_2jN)_IC$9=!-K&Ujv;>VtT*t=a$ANN
zs%F`sm`J@l(~CEnEn{-zS5e`6mWI*9gp}RKwRF;Uoz5&my*m?L+edvO%%lblkVkrE
zrrT{)4i4dpb5>20Nw+A)^fa`qku*w4<c+OQdu111Y5U4CRK4gXX75g;M>*mf!#uxd
zjOA0USJA9Q*C4R@SCd>d9pukejCo*=tf<BXQoh2(MM<LE;{0K$iya~2R>YHhqPSfL
zOMOoCE<bjsENixpxk<?5uDzvY6WP2YrzCN*Mw???4A$PEIzm<<jqY#-Eq$5<ldjSg
z$JuSX$2c`++H9aLQnY4|RdPG4wTyM<ux~E>_dA#X4NRTgiiB*g$s^Ptx3k&22JAiR
zmgyAlMm1HmY9?51Q<rkw8k;Z5`K$2iYERy*Zge!6tz~wyah%nrDou4KdT8_9ZCQ=l
zDjUTD(pJYsHJyzUI7nn!qzdW})K<o~1P!TLqs9u9P(^j5VR3aWI^M@ohB%;KEx=+c
zd+fH?aF5-Ak;fHO<7=2=C3_rxyQ1IO-b*q@C$nMmFLkPVABGz0<Lx%?$!-WNhLd(?
zv0mzAzYtI~Ru<8y-qK(XrYj@FGOf0hu&VInDE7s5QTR-&Hsus>I=op$ly>0ORVtql
z#<h1|6qC%*9e=n;{VHR*olTAsU-gakH61iqSEXC&cRIg>Q6o|}#=M`cGb!$kJs(2T
zwA27XrcegrMO(Xt7Beg$V14DqbG^;P)9zLalx*M=&pm3-h=dZxoA#@NkDq~$ty=g@
zrKHi%y}i^24-qNZmywE^(e_BEcX}Z8G-qC?DhTc=;=X}wVg{0!0j8Q@FxMaA-Vv*M
zU3~>Dj!}A2*BsT0j=@bPu<A0*P8QRjdc?huw3yCo%&GiFY4%8A`^V^Oq_KZ4$}e5D
zoPs5FCg)HY`>D-oR5HmN?YH-1RICS^aV*Cq1GH4VhG>J1PeEDoe)@RV`^}T|ryYTp
zoZ(N>v2G;W6ta`gK|}=1P&Ys0SE5j`mi0KTS(*?GuoXX6_N=NqBUmf4RO7!DqoJs2
zWtY_8eJhqvMs%&tiZR7WisO;ciBJmaEMxMmBVPEc!>@U>n-u-jjLlsC0Hyu{vgMO#
ztZHjEZA<uQs4llY0j%#60Qy;D{45P=Nm`;EPlcS`i6@nMGI(K%Z;44W#p$%yrElJx
zSN++FDeAkiT_-u*+!lkuV5!3-N=N1^UrW@@){2s3y1Hb8>O0nm<!bBZ9&x#~GFnKb
zi1|t6X0Q*5Z>=s3wdIcK0qUm!k6KOJQlSoKZl>jytz$mlII=P72;hD-xeMv?k1jJ4
zdVyT-tYx9ls7{`xk)+tVe6cG30J~N0d^c}8l?{M<5t`?fB=tt{#7S;qP2sD{X^4+&
z$J?H0`xV4VmxX}!#bYQ*9dNA)8@f;<N^FdM>Yq!H$)}V(6p$-a^^c*+HKe^xFH>eo
zSx*EWYAXiIpQU7x-_X~wlME}N@Wjj`y9c0~0=2`h-W*uhYG;*rzIC|PC3PIDvY-3t
z)&^^`o`}MGop*vHVwW-JfI-D`y0S8A*FlC4n8)?4dKfQ_l<Ll#c5Jl>ZU#t41b}*q
zlTFd=+JIY~CM%V}`9o*gxar$;q}Q8OT1TOUcY$Ab&0VK}`Ht`Ezt)`(;vmKU00ol#
zLixNeF!Xi)b>rCsslcyL@~M0x8Dq{lH1vLW6f<PW=0alKjtzC*2(#I2_foD49437$
z3u<XMXdU*SAa&k^6)VMZJ}19Pui{9^A}0BW_v`$qsO=Ik@w}{IG^u^78KwE)h=ux!
zis6<*;xUEY^8G7`JG0T6)S){E$|2!;pL))ddOx?VZxaa-yOv&QP?A>+N8?t=-J15S
z{Cw^C*0qiG%HBgCp8HwABRy-GH_UA)rxN#!6uy+0=MoJ5M!1_*%=yOxx+czgD?;SM
z05wM^AT@HM*re|PsMuke?KC#sYAbl`apD)fbE?zZ_heP!S|f>_Zj^uGSqJ;S)~#+|
zTIgFG?{MAhX!*XB#27uvsHV4yNSr$4Fy?_CWvkCTH*XYj6CQXctzus4DvRb21~@WE
zbNW_IU5+-f#p?R&+%rW4uH%qoKhm@9V1K-SEb~;AkTTcLEQ<;*djLP3cAgs2R>mJX
zQTxUpm2+YxXxQG8ySdPr)Gh%PT5I^zt5=?p)QFbuYm>G$+P<c)LlHrq^w~v@d(={)
z9D!4CWhs|w`qpjTxNQFHw;bSB(0Y+8Sm*9MDSb4lHqipP$t-yT=~_d={#}~+GY6aa
zx10h!tJQ+g^CvZZ3<Zg8ZrvpVIp^A?YjwHQp)n3UswVVi@Z8nBv+|%;D~@@tKJImr
z`P(^C*sLgOZChhkLycEaPWz7)>wX)A#}kv#)0f7KWUT{x{#~Repv`#P(=>X;)X+W)
zNZ4W9=~~5InqjH)O2Vp=cQ#p%Qcqf0Wb)S-<kqb9=DA^HQ!%v8K5wA(sBT9f0gx(B
z5xdBR;w+&6{A)_uF$^JSN1=gU^JC8@kkuZjhox$GbwSkzf4FGsWn1`iI(xyo&R{>d
zf0b}7xneWT5;-t?G6@lfUY@n3Vm!?kP})xPgXJEBBQ+?Da{bxW{5n(~9<^Yk4?WoD
zxxmF`cvc~M@hblEPCS_&Onpv2t!Q{_;%3KyE}Co2PI()2hDlC-hP=yGl*_ALN`H=b
zZ_JvmfgRb>d^52Zo-2w&kMw0SW7oI!uJ=H1?LQo9`m^QXvr!BU*m3g@&XX+izY%H;
zs%zJGY|O$*PR#Zh;8&)2LI^x3;3@Tq)CuNOC;LbHqxsTcH^Mfrrs!AxE7UOeR|?Xu
z;m+Uwy<zwl#o_NH(e)PJZ*<M&m*;6czZ!v&+<0lcJK{^PEl=5JP|+U4o<(1bZgn2@
z!C&J&Dj&l-JAi)47#(*gsqYvqsUEneKE8)MHF5O^CT%#L<!hgtq0Mtv*39XlrPrCr
z>n;ABk7}f~-Amv{%;Ue_CQ7|4toVIoC-DNv>A7J-+4OGT*0)DB6?%xy_>ZW4yTsB)
z8!2v~2lcJ*2V9B#KN|l4&oWZ8ti99UqwyB6P(0|PyH?(ufnzQBS3MQ)9W|f0T1LvV
zd+}NBOEERAO_4NKtZM*8Lg(i8u4Z;3w8V2&E%}|>&~aI%sH%J=f@(H!qpfO@!a*g#
zs#h0j8kd$piU(3_NTW8+ZvOzq5vJs|UoigweDtnr?@+kAa`MPnp53cdib(W71AoZ3
zAN_QH=Uo+gaoH4tkkcrImwbVU`@JeTW7<m&hpkH*Jlk7|p_Gx)oc{ni)Y3E&C9m5i
z&H)Ye1Cv~lh2&dfHA@IX-Lzl30abt<In8lbP_7KB`Oi^KO3cbddpo&ou9A3Mt0}Ka
z(e;?MD`4A#%A@C6#<Xs(I5S-&Ty+%UMRBvDEZWmw?G`3BImT4iKQ*k9>T%lVZ_iaF
zbrwV)E0Q{;(`}k4zR)srk($fY^ms%~C93RU`qnkIn>C>(qKruMj5kr%mE#O6GqIJ*
zn9dJM=<JV^qi@CtQfsQ0bB40fFEXZ^e$bGz9K^@|`K$V|R*D^mAk_2Tn?{_RwUb>)
zEbe8t2g^9kYDuUM?DIBq_{ZZ)l<b;iSEQVjmsWE{8Qqz=B;=aY)1+7s_9nSb#Ato6
z0i4w<yP+cW;(?>d{9mdY{Z)|;V~gb#yWsmnZ>9O<VWc(aO>)sl;iUPixgN8j#+I5=
zkaN|E^{o3g8lJfK409d71JiX(z8IpZuijLVp{cUmOAVkudABh36*c3>Yiqk80aM+V
zZ4x+LQ(w8ex@as&z~IT(@~nl2SGBlHhmo!jHd#$XB$Tx@s|h&S8(}T=5go#*`AHx*
z1RB`Wu5DlUa;gG7>$*^k)z2CgCr$F))Hb%vGN`oAv%<X<MQ6(sUuiz%(97k0+*SCl
z(SY2*8hRoxCj83_Iv?T`+nJDCNJ$vlb4@slmS-%tRzblML@NHPdut8dxsq7HD5Pe$
zx-hiM8pYHIV0Lr_pus)r)s5t2c`ima<Qi$K65}h`g|?v$)6H(!U_MZ2u~e{;t-=8t
z@-bM+bvomsvBh|nMASS-ZmXTJ6Z%&%;;F-2LkhM+46ygEx<3LkpYaZRRJL($P~hcv
z0uDj0a~zsgHz)|M*TJJYHOKvQKgywQMNfp|UNrk$i6ayMo^hUOu-VUW;UbNgvEu@n
zd!2TL;LsG$sV@vL7t*~d7~!+f78n3jSiluGy~vuij0t9Jhf{_fX}@h=PVT>WR$Wbu
zd$}%Qn60`rvAM_#)eoqx_Cc)ZKV#GbEP^n6&g#Rb>srL;sP#uL7Nx3qZgB<6{i4*L
zm5(5b(fec<1ukvd3g?ydp>Aw;R*$~gHuX_f1$)b55Hys)DW;eVHOY9ie(kGzJHJ9@
zvB)n<b#qlkMd%@>wriPgQ(+{VZE?pZrDMq3n76HYl|P8hAHs<(6g29raHpxQmzc)^
zRXk$?vXDf6nHkv9p00TMRq@Sc%?R}wL{{zWFe%!6eqEw^u*szMA5z4mExGk$fmJP+
zecX@KAlfMVAWwArAa8oaz8~npNBhQ{M9k}C-sJiUxa2A-cg9U>$U|f0{*^L}^)-b<
zT4TvoQ_XGYk?cUfiLQD42Sj%^`cu}pq;wXn<1*aFgSoDh$+BqT1MdU+R@i+a*~f_1
zNf^4rf2N<HtQ|(<%ZP#0@GBWB)MrgUffX&X04#Z|?-6P!U>1sU&C7dM^kL0W6IND>
zIqP{0QzQBu)}FN*&t@eY?c`T{^qOZRHH$|%Q$jMrw=p}dS$23rMf8b(n5^#w-v*9b
zCt;9xC?~J2UsImfgC>QpG;Mw1IX089Y<g$XwtP=*57^<jjc`=ppIp~HNnJE_(v7DX
zcQ9b-S0W9Zf=AN08$Cp9T3%d6pd_rnjdgl#jXjdc?Uv6qr71hu<-|fNH$=0pQ$58-
zHtpvhjdUDCB-kL61P_$p3eVAGJ7rGX8qQxFP2v1G9G{3)qSWP9?spHNs@i*g_Oj}B
zr+A}M(eWW3hPurLJ&?@b-+b2CZ{i%7-{BrZuCDz`^V)a+0MLzN9xJ9MQs{Kv3Aj|d
zo&)!!T#v%2c&P$w(*@vRPw86pIsVy^Kp7&W(DuCzguWVr1|xFB6#iAm+^a)0h`@rZ
zM>Xfw)h2gg^pv7h50@glj{zUuUHw=3(L>`y#s2^WlYBrY{t>H;dav`Z8UaZK&V8%X
zJi0#%ERQD~3{;W%WA3j$w9Sk5P@B2Z-x=O<itBYb7evxFLO@_PCpi3TDXVG^R-$X^
znu9~(<dsj(A+9cYS&~Uy<>Wo<nzPXzI7r^3DqBq?5y+@;tUiBwSfmeb?Vds6uuD^_
zWigdQ>Q6OVZcZzrCo?O_k1Tsr;2FWHdlMbaTVsGK&8L|zt!$~B2L`!U#&k>JXHViN
zLuk_X#tm_n@dfS8X;tcDhhFIUmnv$mc*SIoMvqbnA8sqL@brYmx3zB*pAgQ!PHqio
ziqRY|bFI>le`Z2`SNhe!0O&DY3uB$`EfmBR9LeTJ$__Fp1IoN;vuOI%YOl81Rs@^_
zfu4D<KJgEmCET%@44Y5Qx1|~qi~$oVJ$b9vmhwt)ip<qAc15VRZ|S;QTM;yi^M5Ma
zj%l9?ooj-tQ`G3Ec)K#(dUe$J-ys0@6`=-!65*H@KBU(%(q3BxjYby~rZLkM3}Q5X
zl%mRcdL{?GIK|26Os1l>BuyeC%*3`kbQLkOfi2z9R0zj788zEZ5rM@y%6cP%@kfe2
z%OCnQ8<;jS{VK18HG8d2JBPV&#aL|sV*<NjE0|87D%8Z+(XF+)LBgZ_>sme&+i|GG
zg>p`DSyj?3-j=pHUldy`8dQuh8AW+c!d#P_6U}3W_)DwM*3%?wZAK8pazPdCx+X<{
zW7P61NiTSbzi?)$d;b6ux3@;jvVsmf3iAllWRAVV8rBQlOzpt0dPJ;o?TV{0$Y&=w
z=~T|#OSB1X;vxdfa@gbXs8%ia0YJ#@nn#97G)WwVN8a}7PF|QaBiO+&a^ac)HAi{^
z*j)9kpA1L7=1IGv;C?hM%rOqs+0eNX#Ncy~MR1NWGYoWNM2=i1R3RB0jMcpkXzi!E
z7XuK>ZzOw<Kz5Qpl}axdzjV%{!jAfXgc!1d)|j6jxvCm+ohRaC(Xj$6?67R{gUx8q
zo{v&*0nPS(LiQFIEi|Y>?V9t=O-y>N+McU3kJQx)KNCBDg|RYCsNN^vjsmOo>-{UH
zx+~#dhxc+jY`niL&4bSO=tt0Ek!P1^)-r0k#kH!bV=Ss09+~9+Rq8%HK67aP8MP`*
z(k|uc!0Gw(NsN=?Jkn|Y6-_$E6Xop!O~1f{r|ZRd%1F^jvLV{aatQ2cA)cDvAzu)9
zcEVD_?GQOhhkW33_|%r8Gso`2xGS%G^eBG}=_NljNsg{5nww%28Rs;r{4B<o#Szvl
zWV0$yBO<WwS7SQ*99K0@hAio&q~wl2R?BJ+^c7h}cT-;KK3^FB0Dmu1Kt*~F!&KOp
z^W*qe2kDyK8R3`jQRe#P!%wVUNXI3nbNZUx@Fu1rnrk<2zBesRRd^oVEg9CA65g&m
z3g{sMD?oj-SW;WFqPD3bL6x=iIIQ3TvmEki)aX?nqgde74!{FgGpQ7A9~r2v+$2Sb
z>S|)H$*%1I9_OwNX5MP{*17X$86Uz9Pvu*rdF;Tv@kERY+Q+p20C~Br8=I?(rY32b
zvU+Ea%B>=ju~fuQLrLT@=WsP(dOv}`<XVsZx<B)-j8~$*6f_2#K|&Esa0nuSqk!=p
zyDAqvuHa9%Caq|C@+FzHl^NI!C{ff`DtxmW*vP!Sk>t4hPGeHLaqEoYxl5e8VL4h&
z+?f%JasVuGR;}ck0$vwoIOK{?Qc<m?-0G7>H<vpxQIV5T%i=9mraZu<c<E5#7B?o1
zj|h0a`87YX+_JjJymrAA(>|<!;U=^uACdVacE?}mL!z3{%~iCu>UDSbu_eT7yag+f
zRPXhBV1hJpmg~)V(^r|<8>(gx%2==?lbUubHg+wBv_eJ3Iqh3lOl|aVxaVL}KDE(D
zW^-3w^h{`XZO3Z68#_j*$110a!RU5I%^}Rd3<^!`W7)gq&T&z(R!nXw%f%vxZa%e_
z;tfD)+Dy~u=EfK6RT$n$3R1LNJn{=sJ*k>iCFD8HcG`4L9gVHJ4y8e_Lld$`1`pzF
z?`|hLe}>({mMb4Y+@32l!*`!#wp(kc+B<@GI6nN3(y^YFHlwdUsWq;p1p58SXTgXB
zU&pOQ;`_Fe2&|%F8bg-B?OC^>(4+4nmV=;4cQT}ELg(gebguG04%kN!Ll9yH2|ep-
zG`W@8h9<uwQHc%Dg*3~E%K0Hl?z~qK;tv5_Sli7loQ~_)uOMc$l{Xck%NnYtlH8?j
zs6Lrwf@w<x9FlphC@-SCj672A2hi6RyPa`sV#S$nwny_E6IOL!GC5;(Vg@l<EBiKO
zD=Nr>4Mu2>%t;k?eN`V$oHKK<)=BDhG@3YQZdKVyITg3!v!Ad?y-Xs#9;ouVGS`S@
zV`HUBoQ73XTDI3A8qu4PA@bYuryqft&-jrkMQeJVA({(T!p3ql+;y(11$iP@oNaaO
zQO%*VTHNKecq6pbq?~YAZa&rKx}E3SbqOXPbe=wy-%Cg%miMUHRGq+rbH*#4nRZO;
z=uIkxp}Ea=z6km3CWs8TnVfzVn%s(?2~_w#Ys>428N8wA94Q^M*1BCkPPVXxVF;z-
z84O2SiA$lX*t2&&Q|0N@R@8GV=*s<9%gG+JLvt#-BbX~R;ab`=DmdYXO0{pIY=1K1
zT>7e-6mB`xw#B=Ps~tKV$nnIZ)beYT)O=umc<f{Q-^Tv{;aXnCF^iMCF|GAkS}85=
zfRZQapW!~#`+2;!Sfb7esgJ~T9;;*u+C-<QK~{x(#zLA*Qvjx(l)z!GSH)xZO<U33
zlRb_!V-%y_v8;j{3TcM8%?h`&kr-V4D?Yr+pBpXgJ_!U<E0xGooK@3sZdCDt&(L<P
z+8Q>YPbwur*j0$RZ(+@3Xo?zx8;iJYheCHZ(zO&|06Gdx5f$0ljGP5r(<~6Jux9(t
zcN%YD^%;EG+aIZ{o8*wh(|-Z^N8wFuCP8MOyU+KHFv-5<{W?{UNPbw)xy3vlDrAd`
z1z$1MduN*7vb~B2!8OfQUqgCs+L}pzj@S$_Q!V6R^6EaOxg|}V3P$FGwrSc<p+^Va
zuZg_PYiQ_*MO+swQu8@#Yd7SK-D&ZD{{Zff<y@Yjc)w^Nm<3ABMY*%4o3s&?cdm%R
zNHER)?@GmzXO+vdWr^m!Dm1wj&N|gOoq=5Ut#p^p(qvQB6pE^gW=%J_&jV6M>?B{U
zXl8fX`3d2fSJe~`%vUYp=ldS14Z|ERaNk2ta<LR+%$hq-5?U;lvE8WVGs*TfV)(~-
z;ln5y#9^1}E1mAm;{Dk9GnltMD`M6xu4{%zq>E859=g3OAMDTp!RHlpz9XLz@EG?u
z*F$=Vn~c>3xtdskX;`1FLzJIWSksEtA|!Y74q3lC)6(QCdK4ZCnsINP6HzWkVR*et
z>WKZFIQ%O*#Nts~NqfwiZtzBwZL;l$y>~i3KeH9ba%%A3#5wW5!aQ?WIn^%x=70SN
z);!CG&2-0dB)!yZ_&kBGE>Bq43X{fA2=!%@9%VpJ;l*sWIbX(L0h*Es%(J9wIc5M<
zy^R*<t7*4l3xu@K4uApfD>uZp<}0BvA1s6QuRD5Ya`$U`m{#gc&0XJs)U0=z^<V2(
zL*qlm{{RJ+yl$Rnhi-?U2E1>|$n>v7@~G^=N-4t}{VHXeCEzKhY;`^nwpk>5##6O^
zR3vxrPVp7!y1rP@<~YXF&ji#|*eUB!+eq`_#0c9@lq-|+o=5WcZpVCACL>0U+!}80
z%v-RK+nCpr-l@kYmH^sFBbvsV(CDqbhRj0lE66{kQgeZxxvq$uk}EDrVe3|47y_KO
z6gM<rt<3@@ZCWx%KuvOeq;yZ<XG5+aIz;~fLHz5A34`rbTjDaqd?O_sRTIMzQ5JX7
z+^Ki+xE0rUc20}i71FjxHE)Tdv(uaABDDotIT_dKKk^HM{om_W@;lc;*yekUX;FHx
zq!R}Fs!!oen}ISPx&ZpnP_V->{5YTvQ^$TEj{g8el2~K?>_c(ykzQSErHiP`G8_T(
zlhm4)1ZPKpNsxq^yJ@E^IoO!bTH~uehkPQ9J9~9JFyfMX7PcaIImc04v6i-Ds~fAE
zI&I2N5F-JHQCdoBTC)hdtB>*|lkTa<y*(@!Ig!5sR~k-E?9Nq`RgpKBZvfBAJ<nQ#
zTiDm`4aT*UqfJ>EMiQv*%=zx8m-x1iYR{5MqE|$bm0zd_t#f-xH*>91QC3GA;%g5w
z)rcI9YcIi0$*4~k@R;+DeAlIimC?zFMgIU8+3toLwE%59Sp9ldcZBYZzO@k^^>vrJ
z%;>!i-&bLMx;Nw%uQk-{<g>b9s!9`FxMyt+imK>z3FefUNeq?imU`57eimDY&zO+s
zZFJh~=a$JB-X6c7QHp8jY_mB(Pin(qiKPDkc>e%8+oC;8BDCZ(Ly&!HBN2I)JY*Bk
z;Yp7hD@Mh}k7|ZsK<YUBscoY21e3{tI|3?CEfjYdWM(+$&{O@Xqus;C)X<=2m9VSN
z6-P+`%eFgLN#S)MPJ3`DTb$UR;WGZ0D@ZPtx!EfLT$I1@@c#hs8X2A)T5|adj;5@|
zr@ilouXd><O&%~m1N`b%;>PJkom4R!e+lY*;0thyPjV}q@w}7ilISw@x{BB{XZyg8
zwWHkm=xl8ZPqAy#T!KKlNT+XNM>WEDrpik{5#FSq^3qc(pJw@o=xVwG_?g>$CP`Uu
zfj}V2mUG`7t1sd&iAJ-leUbzIjcRs+4-DPU^rTl({{X_B9M7rCs>p}Swo<6wK4bp?
z)~)XkI{1g;4xy@$8XF`LL`e5NKUz#?ji$F@uK0sj(XL9q2bO^MTztdx&3O-sG)c4_
zW@|YC*__7M`^(y-WPTl)(^Oc)5RN$>wdi(Z_wo1ng>hH-&qELVHgt?La~IInT~WVy
zgW966!p!LWXk9J9E~NX4&D79Gb07zfwa+i%BceK1XA!7A``He0gNlheb6$lt?xW>#
zPW7egNHJai0K<!bV7|A{KnVPETB0!D#7CKHamjD2-Nh=ED>HCMZ1GiXFC??LS)w>)
zIjVZA9?W!SrFdfH^)QpNmEXrAy9wI}s5+C<xze&{Or7Ns^OCTBm7d(5l&WrB4}^`6
zYOO8>bF;A-SN8%qo!IcBHN;QiyWqu9kyAe|7OIqPV~BbrQ=418nMiwvAod2RHN!h@
zB$Mh*bXepMFtO*FcJ2lM%@zUK{0$tMV0s6EzvNqw{<=T&u8zGG@Ss}-A)uiRM&f~-
z#<g$eCg%goVg@~MO<?MpJ?)#N%tS#89yaw_=A3TI-4ykmB0GIOFQ>VA8cu-L4U`^W
z&N!)R#&Q<=KeNfPvNKcI+BMC*jIDq;G?t+q&X+{7OLUeHK5(0>o46l5nHW2bF-Xsb
zN3^_(;dwE*Z`Qp!2T9|O;c!bXK^1X1Y8=M7;;ZXDUR!=7jyweiw-t+f;#=#?eBM|G
z_)T;B2=ntDQL8RR5=$=?U)vGAgp25FptA9$w16z?V`|E+3+oelRoc;;)*8K?+|h{U
zQb`<Ux(#wg`yvHl^BTel#xCZS8kFN>D%^r9+VHtN3dyZbxIGArfS#g{F?~g3%eZ-}
zJJOqo*Q<}tgokuD?NMVp*B@Hqd||4tp>Y+A4Y^UW#2)p%2;QPF#8gU~3uslOx3}Rz
zz%|jN!F3CJXntn|bnGju6tzgh1pU^|tHqkSXucm=T|*L7ZT0V2T7IH6*}t^ac8NgX
zcE=ymrlR+5JUo+a6SbT+dbxCM-fD(E#<il*<<<2eGXk)xh285|Mm)~PY^I?eW_A{q
zP*}nug#Z9`^{K1Uin6idPHu9$7o||v?qjsk?q!sKHV^@#TLm?FBgAgyxw(*ynL>CK
z*7#Gx-a-qzxrhzU%-~l9l1j&I-R@hvy0f&_RtRQN4hR6%{aJGqXWpeNLPL{Iq*;<z
z6<+s?IY;{@x%!=NW6x~0QjQi*Fl)5euhMHBIh2AttY*D8x;&)!G<8XQ-3Lh{oJa*F
z_2iJ~#z@c43x*X?_*tDt`^bg2isSozo%(Q+$B8rj;azmPtanJS1B^0<;Pn-kE~f5U
zT;sfbbu=StsRD6~<F$F=7_UYRTSJyHX#mOSGJOSBi$T7=xf>y650jEOsi^cV8#aCj
zZeH?Dss}+`zlC($8$@rj#4;3Qn!+yZQ}|h1P<Ks6;iKw|4n1nM%<=ir$cF*`Rv!Jo
ztwY?ku<LtUi@nN`k+9&C(-pO00JpU^axi|Re^FXRbug<Xv6XkN*l3Zo4pmc~v9D2F
z069U%zNV$)8=1F9Xrd%P#CarDH)3H6>dt;^neH2O(zBD@jW8`|b8=YWX&41kPq40=
zK=9?n*HPR~g<?~ld)3ovNj5tkg%}tZ>57zBy?dDpMKKMgo|M30u3N=p_Yd@~=<dmu
z#}|-PVw7~|v92fzX$^52nYZox?Y)4iwdU0L*+Q>1r>8(n$)1QeAFX8~k+dpk#v>&6
ztj$d8AM}kzYm9uW@~vz}yPOJA+z~@ka(+?PN7^dls6+}n0g7x4>xL~evY)y*r`<#b
zmdTEGezad<8BqZgAMUMIvA34$SMt{+a%)lvod_tTVCN)L!sen(q&_kOghkxPy-x9n
zfGl~eUiLOy2=@j!OQFFPRNXx1JGu3$Q74%1{3gw7qy3?YCB9?y6>mb+7V}be$<7a2
z%A|FTiG;4E%DySPTU&{wC+?vBRpyr$lU&Hm?bP(ALJOI)=Pz<7hZ*lmvDabDNCjK6
zlML|okx2Y%WUh*i=3!hj4?|ChGa`=psuA6IpXAYFO#R}c^sZtNv;$dJW|&o>-f5FA
zsiApog4;?Q^MlyZ={c81dv{J^Egn5_Qo1ocF1s_Df+j}83=nx0vufB%;F;;Ii;?ox
z(rFA$voH5eIzJhPBUl(0ZHJfYYNS8vMM1LIRUbBLTX{BLKy!h-W}JQ`O#T*E@q|IV
zn~d@f+*UQGQC)Lh@y|ux2){AebdA$;*L&gdyF$tA%~eDAhFH(x9yhFG`%>#4^VPqZ
zthq*OWw=XX=Z9I7UbV;ZJ`HO4-ZS>Qc{^+)^{tl2Blym47Q<uKw6tq=msz&QXWV}Z
znY%%n!{SFzb#i0BNfiCY&lS|(>5}VOdKeTI;ekCp#c<MG(Q!KKwFen6(tHmL3?;sl
z$lM)?zz5&fx|?fjdr17dc-0rUTI;7tL3y5id_=0dbUfe29y*TZE4?Pt0)BFhBOW+7
zJXZn(H&(6haZ`^f9I>eC)oR`~mDV*}vk(9j%?|T0=vp2y0Sm`$v94KM$_p{e6Ow8w
z>==F9I~iE7g;kfSjB#Ap{o3Jqk*k9c?$PRCVeY?EKK?--*tcca^I7sr;jj%jEeNXb
zA{U+2`s8<}wbuT^w@Br38nP*GLsw!OqB!eQGU@7DfwC2cO6dGHg|!gfl}>Ay?W1}8
z4Cr-u_vw=_`!&Q7igmX$!+b1_`c{0uET*OhZ9j;ic-fT&bv_?`vLN*suBGs!maoRn
z-%mfi6|S{N=4V%?37bxXX$rf2J?hwlIo|Bq?Oh9Fk*9IuBwS%f6zXFNDFDgiS)h_O
z<y}ZM=51t_kvv;K`N<-H>Wsr0w5MwDD#(dIuP&^wBhrGhBLpM+Rrq6)b;6+|=~&sf
zpiQK;77ZRtTY)#3wMpegZ$cnM4aZKVyA?HVW_j+Tzbr=0sKXe-{Kwj}=G7%^g~E?&
z#-(nNH>XFeij8(pEX1FBL2@zHxlXjK9&kPC2~O<O24`MdtnfOX;}yqgmQvYxnlK3n
z<b&=oYr6|AY;#vviPh?sKiV2R@PIe%%IC4KKd=&8UaPNEHQ7+)qIb(D+98g3`Qw_;
z)HMG9wM5xLf-BCCyCc2ojF0T(u?s4xAoW_@@g2#x&_|STA8L{H73fpd92dNjAK;vu
zIkVGn{#BhZ+dQA|ALm+hJ)AGcFObeyj8&wKQa}jgj1%cn&U|dII}>g5U(CnK4m$Lz
z8m-ReSP<|BIQ29fRMq8Tykh__rAtj8MFnV~*$TYXcXGytSoIhNx^E4TG}&9e^Zco;
z&RkdU$LKN{W4nHvr6#$Ff7ieJ#*^5ZI#iPM6dLWkEMflugjtu$4D(D>liSjgc4m10
z0N{_*pxvkG`cd7RyUe!U%DB(;u2aOM_>1BS*skV@B@mE1WA9dvbHR?sL8#ruFNrQ~
zB4S0PNfFPe&;I~gu|6PLEYNAXl8vn*`6Oem-t}DxUm2-A+ju|W<QA*@yW8+d2Zmhr
z{{RohxIHsZxYhM2Z0*5mCTx8@=@xpnwQ~o8{3jNXr;u-MM#KZ}F>VW0jWz`EPk|+#
z<b!hRWPF}k*l;Nf@+&P~>c>`1N;f8DlX)ljzP0LJB-3sDbKyv}I}i=jNx5;?-AyJ3
zlWDBh`m{>Gt8dyzrFwRk^V~!|>yE#~j_MEGXHEbl9^$ff7bfhEzzkMZ{t;omF)Stv
zbfX`5VxiZ+cOY;$E6Mk+b*&?z=~;RndvPb3b0c;wPmh6Kly*K&wzF@{cIwpd?zbk3
zsja~Ikd3asqOBRze6mdJbsa0i9xG9F*9wx5GSRRN_~xlwXg&(jNO^7}xRp=K=Dttl
zDW1ZdZF6HvNZ#e<i2^6^s2CXSS`OZ2(x@Yf$D!E1FGNXr_*JTT&w82sM${L$h$J|w
zGs)N5x!Knfk@0q?_FF$F5E@gD!ni*-KJ~p7?wpn7c?uew3CCPk);Ys*#wo$^j!2{d
z=5vs0Ws{82U_B4O-|{WT{{USd`PWBYg7{HOVR1-k8WNcx${(czIo}ar==lwXiMJo6
zXX+ZvclLq^S9l4MYX;I~By&2Edz(~_NjDWZRrIRAXKP3qSE?ROV%xbgMIA=(`$Fh8
zFi6Hd>tfE`Ko-y+^{#{Q?^O~}&|Xa1V+*aLZ<JMsq0MnuH-2jt0DQ+Z<V?yEwn(n;
z!Z(8FND&qV;X85YJ5dv!rpF`V2@KlZ%MeuV#tmlepwVnD+HhWJGLf(0xY@PaZMh&e
zO>TI{QoDxY2qT%5Lhj(wX*H;>)YXcbFN){0K3&i=A@AH(y<^4_T*k3jgAne=t#dj{
zo5<`{nvJ@cl6Zqsjxt5b{Hn6}n^hn3@=xJfDlqJ3>r4-bwKo3%mzJvUtY64}YnS}$
zl_)NPb!PgGOWW5|y5dd0?^Vq%)=QZlZb*z|)hNYFLn+C%snF^AWcr=TBw1X47o~JI
zwo7HNL1!P#Qw8KbNi^%F%Gp>}`O+_ZN>!f12z#poipTJE@<R!-kcDMGDD@SESifd%
zhniIl$YXEqDKgu$^ABq5bdM6);8{DW$sY#>u%PWDw-Fv_SsTeJ$h(n&2e_r{UFh;!
zvks>f=KlZ^d`7C{Pl>W4<-k2^XR);%+`AkOkFEyPql)8d#M}aXdsji?y;D<zOIa-D
z+Y*d89dTSzR;aJBx_5&-?KMk*dY1-PF02VT_pf2I)tgbUj1?uE22))%Ae<S^88uQP
z-5w{|Aj4xd1=53V<JC=hhpT&|$nIBV%!}y8x;<8Q8V-``e}IGe*P}18d26Y^9rI1$
zojIHMX9M-8YnMJnhqJanLr$N<%<BAcIn6P}&Y3NvHVVE}`d4+~$j!ctlibKo`2r`f
z%@<L3W}L1{$n7;P8q-#_5nM(XZNNU2=UTslp|rg*ETkzJ98#w?rJ=Nv<g@4&`dyW}
z8P+geMhHH&E9w!$D1@L;%aMxKP0Mo~it@<qBP+4mdRDiGUt_V)%E|%!>BP>b!p2-_
z*Ar@Y8-7$^7%{1amuU=MYYpv;5_=I%M_ZXnrj|u}&k|W$LM~B+jTg%-XnRx*sjcmU
zUE7=tHryX-tLtJMT32V8-A^Um%S$gTPc*?EYc|f3Rfu&pA}mHgtiWW_1HRBSTR2e4
z0?J4PHFPSf6b0Z4W{7$XwJ<g5;cH!TGbyg3(Ou?aX{MMAHOzRg{X-vW+Lq#$#|($J
zrAQAIjdEZqrq>y(GcS&A57Mjk;=Jmg8z{1o0$1}juct?djEAeA(y)<eyePhqLukji
zTz)2{7toJOkIS=M^=?gR!Z03B?uv)2DH*Oj?O9hj5Uxr0tFGHnpK7k6Ge9pYCs6H^
z^c8aG0|So4Q1=mwEPLYe!v)-Nnzv&WwwD;3&$;9T6&<SEB(LSQ%#SGHM|!P1xNj~#
zz0E__mWiOmOOw~RrDkc4Yc#dFeF32_9^Bwn4=zZi5va~iIIUttv^&j0;OSl-Sk=n5
zz>0lMFN7u&U6%*v3D3PMi{3_bpYG&S@#9SQCvW<^4d0b<7^h7nnB}tc6b@<PwIEt_
zZewu6v7|IM?hu!SV{OX+05GO`We9BY7z!4}G<M!4=SR0c?T_VL(wqv$x5R0N_)dqy
zm%d%w+LQ9gGCrcZdsvz)*?=Frd^h|Hzw#7A;U>0|W*z0Zj{apM1hF+d5n*eBd$Z7k
zyis-(#cyce>ZSUMDST#i{{V*0e^g-`%smOJ@e$>Zr6y{&L)NO^NKh6B1X5drFfOkr
zx}S8?vys5bt2T$Fb(vW$5utW?fxGms>%&i<4Lf=k{#A57Gn*OwBgVA`-+gf2vN!Xp
zPAgOmXi28ZZM8djy6kXksPVf00K~r8@zwtTrEIo2CcX4=h^+1v6tfe+72Nnk!P=$8
z(%nxHw~PWaAm^uCQj&6uMSDdCn>))t3TgJTd9taVPUZS%*j9<uEaNQ{aw{@@<X+V=
zjA^4O&UKy6M_ch-_?}Bx6YT1z8S2O0w0svVvs>NIGqba>$tJRd+^QWlX-1t3t=Z(d
zvlFb}4xi}~f9OW5s^_V$q-3@p02R=9QpfEV-)UvcaQktZq$xWh--@*lHsvhP{IT-b
zJxylWgXx;ew5=0-EH2U9ObGD8u^jnh0=apc?AFpud2Dg*UVTkuk6R7y)0e5uwZln1
z<~bsik^o{pmC%vE-O7@_NyS1BHL^M7xcG=ITQ*6#1!Gm&8&MuXx09c^Pvu=_gs{`h
z-F`v*>z#Z_n$`Fj&|EspYd9S33=h`0SMNn`gyEKqiTYM;*Aix?p66V5GiGhmDfh0I
zM3Zc>H@$WAM<ku(&hEvrDtp$-(?(3}w4eC}!T#^{tCi5UIbPwKYdWCI5C^HIA_u*J
zF^1iqn?`E0YU}f1a6!k+Y1oG&KJ!|WVe;7H9PJgvU*5E`Fb9%rDvr@LqMW2hZk22%
zLCCKvXQGR?Ru=6LnUFL4)t0`ufNcjPc3R#pFpbVBb5xr}S@jttQX^2K*0IgS#kI%p
zq-Twf1oo``qn?GjY8GYne=r6IAkfL0=Q>ll6UKR}?u@-D7Dwi?^`q?>pc)kX>ycFs
zm*-taB8)g;Thf;-k&M;TH6*rcc?iZq>s&?kl3ME$WDIW3(dcW^skP1;9}`BVn{yqs
z+f;t(`ySQ5abU8*9v55@k}J=b#CDkJW?o3+Wy#(<d)9Z3gWKA}0O4LjPvO$LsK0dJ
zx9<^i!6*F^e@(;rR(06hcPjq?wrh8>?BTvS5?i6%pp0a0IP|8O1IX>Sj)NVlP8>tM
zM#oEXv9hT9sxoUkCRukkQgcCAXx>EQig97ny$804+~<z8iau%vIgb_9=+TIDUH<^z
z1&vZUaZlV{q&^z|0FPUb_x}K;bJLBp$PZ>Tp217_viS_p%U!31EzjB0JSI0EuV3j|
zx9=k?!T#u)E6HZ@h5rDU%C_2MEa3VVQT`Rm{61B>@g>%)c_GcLO*=QPct6gz>Uc5O
zuc2Q(_lfT$W-eze!20vWc7833Oo?<WYxzV`MCNG$9e!Nn@ilz~_Rl==M!NnW)#SM4
zU|*IL{_2jE-gpB;k3!OB@dl#4TNEsthuu$K#*t*67qpkfz9N^zx9UF43vIWI@-jRA
z6;s4|hx{fUFt*TPk-otb;zmyAx&BlE#dxDYm&5wRwrmNHIf=;o%UfR$JW|>&qAado
z{SM)PIp~ep(&z=>h~66-UZ)njWf(9`AoBu_1A6{7qoT1lk!SqzT$Prn>!AI~Bdi?-
zVO_bB?oVTi=fppC7lrtl95NK07<CO&_?VT`OvCsxO=U@5EgIH*vrO|bCS2#%r{Bj~
z^j62p&sM|C=9*vLsEY`)aq^sxQ_xkNKNmWLF=Vz|lj&0SHl&=Uk?R^vepQ%bfu3qh
zVr025^)(%l?=Q5+U7X4RJYuV!D>?3TQeNUJ%Gs>zxk9Pur&`8GOm;lC#5$S1)Sqw6
ziSmJ2VMB7Kp5WJPRib&bmE_n0M+dDaKJ`{4UD)qRB>ZUr2IP@XM<CJz(|is803zIf
z_0j(Topge`*TROj7Ld?V8flU?#AZDJq%)Vgvy%MC&$*TI6-8mDn>=x}K#0nNxE{E!
zc`a^QWrfk9wbNR8vSxAu=e1|+mhr)I0Bj`j<ma&!Em)Rwuvtjbtf}rt;aa{Jn4MxJ
zkYjG+libooqgTcEAWD&`;nyrHjFEZAVNLF3J;=z-cK!>tRMRAwed#11{vZCjaoomx
z89px3*>y#pC1w*YPt;ZuSzN|4(7ru!Qz<0%E*EhiJREeVD+57j={y*PwPt<1EN7oX
z#W%&0wA$0gdPblR%QXJUKWxSc+ME)zxuFfw+Zh!`v|9=8FKTkGO)xBu1_gE=88F+}
zNV9;X^(VbUP3$Jp-0XZYWFUabcrp@4(xNggzl@5q5*bv3^cA&AZc(umYQ{5cx#OgW
z>9ERpncI)0dFe@!#L+(2Q<Wl|p{0!BQ(j^i{7<S|-#n&lxm}!&l*?^K;Vw-6Yk7=v
zTRm%skwxft`&mnKxbWwWF5s16mA3=NNv}oId`oM1e78PagQzC9jEa<fqQ6q^qo_w^
zV)JeVi1};6{8%Nq)$Uq7(hzqwMCQ4Y6e83vV`=&g-TQ}ZqGzpj^V>Pmp4ppz*ox=y
zu4>awq`DVQEhf)1E4D^bIviK9_)(YpKF&ty5p!LyShJ2=Ng6&fxO--_xrZ3RRvxv8
zHI$as*_478n&I|*(Yib1SC>Y4&B^mFWPX91n%mVrUkmBmy5+I-t=~hLbv87d&EXh0
zB(2ipAMeumwBt^UdddAM)7nOLekh#Htc`ARoaVc|2;Vi8q;~+S8Mlo3R&^eSRvlhP
zsUiK^{VSO9dhL68d*H1R+-+#iMk$0+O$(pA+lS*+F0JK~S(GVf^3K33pyy_$pQc=2
zX|YW=?uvnsde+UYoG@GMhdZ&!HECGpolEG9tzX5K`ef0@$j>M{Tn;b~dgZl$65P($
z7RF7npPD_qW`_C>oI10cN#s)d7~#en-n%agDQy9ioczs1cPg*Kax%xVc^A=$rZmeK
z%N$e|RF?d=3DToSB<FF>XqAN7(|B*hQ)%+7GPdYg<DS*%w)Y}f)<qzz41-e9X_K)m
zK2yyxZW$X4D2$wCbW>L?M<`)ZSGPu@sT-IzW+A4U+>$n~TgB&XvsTo%6WHOSIbW?x
zzsj<%iKaNyNX>B@Gcx!7J`d1TaHi+aVP17ljg4$s&eUf0IIWvM`1OzJSV**5gl0jW
zgOgPC7xQI`+#kHzH7K{SejpV>j(}Cei0v|XEtB{Po~0u%O$EfWp0YO|K~r76*s<K2
zeM#s+W;agEN8DU~H9f+-phXxZ+~5J;kF*;jKKDh`$yG~d;AG%ccr-0hrBK$=V?LF&
znU6e6*BWM}9k7WlpC^;WREt5>l<sR+$@LYVD>U|z?o2doNuO`ooDZc~O&0ucEuRD0
zp~a$=NcRJx-;<MW`1(}ZZ-=j^)nI0Ul~C>{9+gmwxiu<Bl+^JxrTm5qbhj><+`}Cz
z$At8I%S*<W$zY{>)0|&9oF^4XvEmIT=HmWO+_OgL0pr@ZTfYrzcJ~Z|-Pv=&$5B&F
zK4x5{9#lp&9ZUXwU-G973sV08pI7{=dow<Gi~WVEf6uG_RYq9aDH=H!45tJh)T}hq
z7|rFnN(M^tP26OV0sKu!yXbblCgVl7{{Vb{D&>{yD!ZCt-?&(_x@&8=rcyyh1mn`Z
zJ5Sqot16L=n}sL_Bcn6*dp2AD05kr24AWwZ1DfEgq?y}E-Uyr~+SbvS0V?<FN|(lo
zulzQ1y7L9Ukw3ik41547%XVoGg%aZevvoN(LdV*aHDhNy8!cDuTcvfG7cb<*5p&Sj
zap6gKct+v91OEUjsy`Xei(k54r;yxbq-H(IHC=^T8#a7D;mP$|m6W3hP~?z0{VUM#
z{3RZ*r$-IT7{V|o9dTU{S{{8`mX@6f>~t*`!|)jrDC63{G_qqj_o~Oonq+rPZE2f{
z06eb?`PEKJF=jO3Dzh?gHSI%C++>r>PMcGKQ0Uj5R-_e)#J?%`J*&>=a-G@jdnm%2
zj7fY;X|`zY6mb6lDL(bP;MLos%HFb8)YtD$YVP=B$n|6Y0290K`uKnS2-R?FtifzO
zNn#XoPfG8!=#xay3(0`NOSX8)>?)BJHnGNup6c!8m9kjpCbx9^a>yNr%bqKX5iJi_
z2NSiuj4p(CbO<Ym)ukU~aa{BGk<ml=a`~LX<)5xG$*UJ%DvWj&)XzP2Gnn&D6{(Bd
zd|W+rHVgZpYo4~pMH17Z$S`@$Zg@fmhFgQxFaR~rP5ag~tMD^+Hr;D^$MLwXG5gV1
zps@W&Wa(K~jq_raJ<gP7$bWn*pwgQK51|#<_BlS&JL?w7jMnO!5t}-lHu@;E$XAwS
zUO*o8ESi~IVHH&OX2BnYbfuxgDQZ@G`6N4h#$%p5wZJr<UCfP@h$%fwwrP;tdx&jh
zBpEIV$MB-kL12b=UnrmUnA7R09Nmbu%U`nS^GYxX9ff(kM9Cp1*14<fcHxuO=C!n3
z3b-JzI!9@vXC5v802;`=gvECL=QQen4H#2@ykY^xHfwP-Wk{?d?9PcRs}(I|Ygv4m
zdS`=LLku!HRn2uXruqoU=xV#6tj*C@Sh)476NXMjG%@yb?j=~(wWr?(o6(uc8v2|#
zwpl<J7~FjYaXu%BB9<Gh3=y=~s-5C_u+nBVwxa?}4?sH_*tXU;NI*DK(zz!cPUW$8
zPMPP_VkLb|V)*#9mryAmDvbU$(@RL4mrkN4@JoRO!sng9<M66_nG34Ur0^cPdLD%t
zT5=&57zK&>n~psyB$ZW|uiZ{_QgYS!*<87jKzZkrRuFf^2W}ENTzSSbO(dERL_j&G
z8U{D5b!r+v3ikv=f1Oe}aX)9&i^B)~7IFU9^sFsKcIp!^`*b~pPx$g7+RcStNjMeg
z+EfA!HQaH5pUSfT0GEj6%w_w#9LBGHmTlr^0G6IOW6!Y%6;s2g`d`D{Us2jOToD=*
z)PunlvE(g`{{V+}>vQ6$V{QRM{KaW}R-NVXJaFS8D1xp#=bFBR{j$G+d?ecDq_<Zf
zt(Cdl+<o)uT~CN^<?(j8ABJ?IQqBTC(;)l2cBEN<#oEocgnT%fRgM<oNc^+|s;@%E
zxDO3!+O%-X;yWy`v5+$<{xSSibOX?QL1As<zX(UDS*ZI=YrZf!Aoo*VGRo1pRvTGP
z2=t_Y^}hz{k@&B|^Jw?PJkX}zdX?;cwL?jUYi(Ft&M_=ZRfoNER$Co39+F;1T;C*W
z-|XhO$z<Ovw@i|2lM~R}3)PU4LC$J_5T854GZ1=3pUR^TyG7#kh0ifp!OczP0zIqI
zK52?y0-7H^vrLN%*R5{&TGY#WWXB3HIPX&}j+`pJRC<0qn-I9|iY!x>Vbi5c_>t=`
z;kFaurDNBcts`TlV<)NCPvJ!yiEt~G@kXciJ2I+%c{mlERpgB%?FM<YgtKS$re_Ad
z1dlDUb;N@o#3?b`r6x4Mif<<yX$=NBr>`ANA)ce)Z}}GE{{XI!{OhC@+`be(g3=la
zLREXf#?nB&cG7zcr3C(6st}nN>?;!a(XUm}g1a_bcLNoiwi+PR?&Q2H3dopGnTiev
zrDN(=G94YY3E$=o@+qvOuEU{UO@r;STf%^-iZi{vaz$<E`Ycju_cIJ{5*>}{gG5jx
zGsH4SJT}uX`Su;d-nlv3=f`u`+LPSEdx^HY?+r)xUk|K})8>=#rya~^vAJhs2CZl=
z8AbwzQhNi|s;!hkCNip!u0nCtSCfpJRcySDtT)4B&9zT0uO=~o7jeK91Tg%WJ?qk2
zB3|cT;CQhe)8m3b2C+41n&VZtcIq6j<;7Fj%0C+`1>{q)ip<!jd-p1cts}7yapR>+
z8toZ9X^hg-Z3M69N%vQv^)=J!mMwQWjA23MuAFYpYRv9!TVkp&8%7092AY?8yf;nv
zq6a?p*(7o%(a-qS;pNu6f`4}uV`%<$&uj!%EZRF9{;da<bH5xgs3L0NsI_NxH*|}U
zPRKzA1Xp8atBI^dz-_ZG5zqCk?D`!sb}vh=Ng}LoAz-V!gIrDap(UoVo!M6a9QOp#
z$$iH;>}qSe%jvd|{h2@#jt`|j!rF|Nwi7JPxz+po*E*)XpD|qTp2szFqfaicG*F=^
zAFX;{hqTG!&|!g3&w>H`Yql=>990#y^EZ4?rlHYdX<L<GMS0e*t=j3RV1-1EM<TMO
zuB>!P^)U2w`+Hm&BLF$d`x@K2jwx2w3yk?z%YdW3Wl{UqMzrbsmuCL}hqVg`wpo};
zqeuwI<E?ah-icwUTPm<$y^&Izj9W9DS3FUdXW>mg(m^J`aqCvZ%RZw7aSTTr0ms&{
z=5l(Sw4#&H?e~GlL0sRAgUe$R#(w0SpTfF&qZ??>KHkUecd|z#oQT5>zSYn(*;zm&
za)@@|l6kJUtuA@BWZUIOFAt8bV41DsMsz1>X5b2~r|K8W;*he=WQEb1JRUJl_hRI<
zTQPiI%_hAm>{Wl2Vx7%`PDrHpG>^iEs`_zVr-WU9vgC}ORnC7JBe^v`6N=QY6)$I0
z>LhRHS#r$Mwn;TfSfQ&^El`VA?UoB&r-gO19IIh)-z$B>dwWuALZWu4*rHgsjmy@u
zG$_e-)@SDj6?V^4hUGw38LpfyUP;;P$db8tIi{NS(TK*bW5vsxR{B=7w<cR0BxB^M
z7o|;gp)k;Pc%&z_a5G~v`~E%;&{LsJv1cE8yxN}|Ee@FN3cYx(n?wGKAN#e8i$$ev
zsKj+VR&}{I8kCZrRxz|xlxsvXLXPzPYF{GMr#Zku^{GSXMQb0EZi(nB8@AmjUc;?5
zxQE#(5N<tO?y0R`{wG$6{{T$|5%`5KgzWFNn{N;4v0h5KY2+#n%-{~xwwk7$KB^>&
zC;9EZ(Xgi#*K^M0ZCuVz7g%3w7S?myv`yxLx60d}tzG`uw!eq>$nmmFKrEQ~KqUVF
zDxPG^blNed@qUvf!!&+iT;YCB)v2dnTWS`PT*oTM6igMGQ@4{@xhCvu?Bx}9BHmkE
z350hsDnj8}Mr<i3rVVO%d83{Si=;b}CfG<HsP?9sGn`$SmwkO5#J9Im?((eu`N#6B
zUJ=${)Zn<dggJ^X31!DL?w+9F;+ie$T9gp8O#8NO3&8J;)lW9s_Q&l4WQ}(fQz6<(
z9cwiu%xTFkOPtJ}DbpgAqb(sQa7avwPxwo;zwb-`0DM;5$j{lx(bM%SeNHBp+_6<S
zJBBNP)&BtGFL(a{U0eCmmc`RnlSLvjFle()_wc8zgZb2(y^hPphi;Py{{W_g`5NRy
zR&^Go;r<dp)#*PEru#*bq@7}HejTeg=*z6H19M%rSlJsKgSYUkrBRyWuB|hp4DU@2
zVZ~h1KjY$0_dl&?OXEe>{u?>1eam=-ld|(m7g1VGS(Oik78~x?7PBTIBi5##<B^D`
z2Cmwln!1^pHm5Elg?p3IyPpX$9u2tXuKxh#S4ZMG@lW@5^E|$FB%iHc(DWD7WfIH^
zz^9Y_?`o)OI<J;DJt7THNzt^Z)=5$k8sl$s^r#{6t+m{u7$RXCa3pW=u8B=WBg};v
zO}5Tc#<q8tiT1Ub+sYVZ&ls*#LO9i8AKv1or^3#RJ+6`2NH#`AXNM^$an(+1%uMyp
zy6IO@(Tso$mEJqx)xQdzx)c8Z*F#;@{{V=ek>ys>gO8cx`lDx7ygswP{)B3{HPVP%
z8y*_5H`<~W8ICWRxZ~+tJ}=aEtad1R{Lh2=3VjIPXvWek-uC1!FpJBdO6yiM-~qK3
zyzE2wtb2GxXk6acr&zRUG5yhs^Nag~A`wb|8Y>RJa?Fn{x1Kr26(p=oARoLs*H2^5
z*|bq^w4EzM7D*T0713C*@>EwfeU6$1)m+z4;g!Xl$Nj(QT=c#o>;C`@x2Cb4&T>vm
zJ69z*+bo08jTJUbRCJV)T27UbdH%ABdz}Vzb1<##IPkUCEsjO;GrQBDbTPNJbSQG5
z`%tzaoAx%wTS(caa52KIXhWyMIT2+40JL*mDQJ1k-c2HLY}ncxV)`h>1}I|rq>yp_
z<5Vt7>n23nm5fcZ%mn)k(`LP!%~iFwX#pR2@G05#1t`CS=loTvPyL%DE#HO;G8gJ=
zgtR$)n&hl^Vef5C+g9mXA(w4&Na`+GkYPVsr>aNutAWAb)lGir=8Lsbp4?N^aa^eD
zD(WyJ+n2{GI5nSRs2g%4l;v~R(zi>RCoJX5GiEkf7jm3~LndpS>WjK7%N9pU&$)!0
z@x=y)Sq2$*3f|Ks^8&R~B1cJeAKElI6FDUfaoWYN_J@bv)UM@S+4RR+^r^eAk>))g
z%=3vDqj#yIyO97OlZs6pDV^tpWHZQTZbr;~D<{U%8E#u6?zacpw2#7aL4NeF;Kw5N
z;d^;`{Y6EqH~dZRvT^g){SQhH_m9xeG6j5+O(1=-st#G3CX+d4?<P+cH?i+l0XV@F
zk6R3_1UET9N<C;gM5Ul$E3DMy^87!Jn~_$j9N2|<L)5L|b{<sA-!+q~3~ICE`_!Jo
zhw<b$PD3WAZa6q_#Cr<#Ty4|XfrT)zKT<0=>rmm$p2+i07u>zRp>=DFi>TpZeKwE&
zy=ZG__CF1Di+%VrOk^X~NG7*D(LNNQJ}hTpl}l$i^f?uye`a-Wh_;A6alE<D_d{l@
zp&w}I{7I;$hhYzfv;<LmbMl2=N(0oNdbMW-kHhZ^y~V~J>Bi{}a?(Gb_M}<k`o^ts
ztm-n}T`^~n{zI^?@8Mp%5Lxsaj0BjMk3Gp4BlHw7H1wu#1$e$)Dqb}0Nw-^`$ljmQ
zxxa~i8X8`tx|O_OTU`=cXyIG`0IyO9H{rc<Z3kD1@+Kic+aiwYde^43@M<0;Ys*p^
z31JSbJD&BOMX{|nYnwM@MF+4-<=ML9u4{u6x<_6U;cShwjGncnXKK2xo@`mL8Cgj+
z7;oYgjMkGp<5tk`G>fK+<|3T@ryLRNYJf(sMESR%^+xYc3xUDU(wPiUj(G}ex$td~
z^5S-0Tjn(qJ1|S$a_o-6@wr4FTAsyrz*zS+E8{&kr6gB=3sp&+;E`PNJ0fLW%oP~1
z01ma~z9iII{{T>CM*Z4wpc=X=?&M-6XxPYUZ~$e`xTAVD3Xj6OQRkBFWNd-WNsllB
zpp(#nNsi-L2;-p?fqT*zhn#>qPyp*v%15a9AO1zizw4v_06OU_wQPsbbri-nn#vdu
zc?N-%A&bd#8*&NC)oo732ZX!AkGGyhMYJ8tmO6E`tLD5Z6kz<Zf%=-j*Cbolk~IGS
zQ5Hhu_?yzRp252xra|^3xll5}L|~6z{{YUlG~m`6LlQEpoSGxB&I?q!XJ>XlC?tVc
z(W^U_JxyYcc|A#Xk@?qj-etFf$@{Lb)uSnNH@p?S{4$8ZmgF4w6`iR|EsdqYmAOoT
zUS(vbB+j}zHjgxsi^@slv8xgvE-<65dIo1_;b72sedHqqmSBGh<Q`u(XONt;F(#b0
zMsWC;s&m$#*wh;oZ%-9SR*PaD<cKmwOn&VqGpo?F#+f1*$X;9V>0PCznp&>aTpH`3
zw7JD!;xEC&R=$?NNh5Xwwvo<bc!xlbQkdUcObm>6vXflirK5`(gh?j$+6#SajZL=h
zcDN-J(UGayi7X`fM&%g>xs>1z#;~ZhHKg<>4Uv&rT1@TdJQ?CuBk`=(=RrG4+<?`K
ztendno3hx!{41tQLK`);*#6a_S<1YX!2Ij1_-&zEeVS=5a9A)PAEj}B%d>B{ahlcL
zhMTP1qN<EPYSPd=LlwoNn9usRz^!Q`3fmC)tHcs&+6ntJV8`=u*NXB_G43ZD*pki3
zG^yyBthH$Cw2NJgk-{;K_0U6Q3QM|T3CDWkz1`1Na8XGi(X4Y{!nV<84l;4?Tbi6V
zQHYr(EAw=%WAHPIYxl7tlJulOXP69*dFxZ??-cr!F~`r%n#C<nxao7YnSl9w_pCn?
zGPa8}YmLBUR?XR*={AozwY{1>L786+s6eiNQPnOcxN|Jd$2jEIPNxl(?ql_kkzEgl
zt{z=h1a>E4FkJi9t;m<c$$VWF+xVXxvUB)W5CC#`r7gt|g%HbfvFC3lx<3U>^2K+P
z#^wICa9WC|!g3mbAKDim_0_+fKn6`v*2B?iWlCT=HuB^L(AJtu<+4tk6(Ht@IG(4c
z&MvHBoBQYuTiREVSel6>Fx;lu6lS@dV@SNykJ~i&EwJU2)KIquX=olHhf{<j9kQ<s
zr?IWh)$U=Z%?)=(5>{gkb6zVMwQr?&QuvJL@SGf6$4?a%_9hEKb4Y82n;F-OzVC0S
z6<{wRal-Y*d6hmkXo|AF>9T8j)BZi<{{VKekxOz*nS7gfbJ%h9s5H3tTW=$mIjMUO
zsk5iY=3Fj*>hfxiuP2|WK?ZY!z3NeKV#6J)k=nCwWh)GlcRUJl5nXjj!m20(u&pgE
z%j;I|-H80@#cE9UH}$PnJ97=4-k!Ixx6F!vg{vRL`n~0rs}tEwV7OalNd!N`?fk2G
zGnSlnS1IZ@cXz%Vvc0_r?Mpi<Ohy^kuk)=fHd(H%;TnYUNj2bLp}L5aWRJ%Jqp58r
z9Sq$~ZEiICd)s?8wI*kXq6`8Q0uR@k+t5v+@U{57SsE`rI75TdvUcV**OSiw0I6$C
zw6I+o@U#+13tS-z8|LV1u+Soo$3b9FB0HRddmfpogp-y0iL9S1US>7Tm)SI{-8}8y
z@HCQR>M{CLJRPQuHp&Y^V^cC0m0C<OEx_aG1#3%t8CHJleaN-&!pWBemAC-o{o1B{
zh4AzucqNdDpKiv+I2~&lX(n_-SSxopO=1rw>w1l<{Nkvo-N$~_v>oCTR%dD8!~Xyh
z>3{2|^sZxAKk+xa{{XJ7{Oeb-gnl+rCyI8ObE?}9`f9(<q|xkkzA<cR@Gt!|{*}oo
zsj0Z=KXi~B@GH=~5ii*!L_HsaT=I5k>T2+3ZLh{2$W6DNE1XWX&xv06VBhY=bH!TH
z8y7F`wT(ZG8`bz3&Uk|#@Q^wG0Ax{b$*rpVOeuT_u_?k=Gpxy(Lex@Pl_NA&%Wk5&
zNFmfNb9)#cz&cl<ctTS@gzg&vWXZt#R<zZb$%%gF%;fZ~1@#%XNtK09%zb^Ue^a}+
z(QJZ5fG`-11r+p+$6Mc$ia1M#n)2k!GnQsKIW=`}BD7G19FIzwUPYv$Zs|JATC%;o
zoVtt<-^42p=SDhbi0$NwPmv?#PxpPRc`MU6;qKP4xpY87&1HknHRr}Hqt<dr>U=kJ
zB({5YJf|RgR^F9vZM55j0PXV}kEScSj9t?_tW1*TM~mLUEV_lsRXobvpRG1KCSn6h
zfc<NsAtSu-{**MG9j+z8XB*h`!0aoNl6f^<WREPf%Gep~D4vL#OPXg}X=xRWxP~Rb
z>c`a4BWDUY=Dc~!npZxOlx-Uun)0vqQ4QY}!rns(2M3zEkKZE-zjVyWqf!W74`Wnh
z+!?c;YoeRFIWaWs`H@)lG?mbBGDvxmN2#u^)+I=kW3_VEI%uNet#tk$k8OjW?=@pj
z;x50yH^cI7hG_BkU-{QFb}@4|`|bSdX8pt}YSkoDlci%{oJmzjb7^c)9Lg(pz~gOo
z&tsB1ovxDEfCJjPXbxgur6sVL(phU3HoA03#Js35QAaf))-UdIEVSnX`_i0OT3Tv(
zvY))vP^oFHPT{A#NSFJuig3_j4d%?xJf!tz;+@w)ayP7rQr6-IA`3fJ2fpK<<xl$!
z+{AhJWmEf+41TnmG*PlJ{BE&YcznYW1q|b*dF83Wu1d#c4VklVCIwwtc&-^8CXSr>
zw`1tkw_r_aymsVP+tas^$F*Denc!qoC4VkHde<UHXtJjJCiTU0TYh32*1Bn&IE~R}
z((X%E$1Ih*Z#;Jil{KAMbu^(<OEK>dHBx9eJXQy4W;Ns($;C;e&$eGInyt`FW3;qy
zGft3o;DKD;nydCpi%9^=IT-cLcTsvo^C$L>7g*7nIphj6h9ip1Hq96yuS&Mfk=ytc
zU;8#<!<KLsv%FXP!dQ+=kym{Y&rkTpF9zga&JSq6sH(ct?ba@{-Rq;F-5n~wLnY?l
zCI>)ysqE!4-G;y`oMNQsdy~ZMlQw!5tAykOLF?iE1SK74Is>IIS_Vl1d9$Kt?=Th9
zYv3l1soXM(rLoOf{_+T;;dVc1nR{)`P}V=>Eph(+C$KN$zd_nKKiUew62M?%-n*Nt
z@fDk7VSLm>pL~j#C(PO9<`TTpJnP4dl50>%y&4b?;KYAQxZ%8CXAQJ-$7=?WW&wFl
z1Cd%BM#sWv5H_nh!Dx^mZ)2bEtxpfwyxuMs<E6E`=kds^=tKC69twil#-rlRLO{|P
zT|sa8W}w#v;{N~>&*D8%<lI)-4%lP#9=}>ejqvY<?sbc6yYD59y|irEAH={M3dqy%
zC)2eEuHs^nLi>g|#Q=L2wQUcLd>bA0p#K0#jeOX)^;aEMo2AWf<F5w2jnM+?A+k*M
zR~-#UKs<X;u%Ar7n$F=$$rBYm{`Kkp8MTfZEi^{iGFk><=t&ha*tne*ra#f79m3+d
zRwsW-^YI@M-GulDu;Qe#xM;k^LO}#DBBKxdL_A;cBs2aJTRV%Xbj@zvOl7`TBO~$j
zs(QD8)<yePksy{f84;7u;A_)9d*hCqoUHnM){+?RASmOr3{!TI+L6gCr)CQq$tp*<
z{LR=`r+9MIMF#A201Aik9-a&Ma%IWee9V37{6I%(^%<<<^=$O#q+*Qu8;n&M--_pv
z*AtQPR;T1!m?LkU`u!__k|6AJ_?qgY9S%H1uPX(<lXrUCZQy739R+k&9taEN%YQf^
z{{RGk<6W?(*~2=jCgXT-O?Z_AA#sjZf@?!g(5y7@N<y2+>fD;wkzD8Yt)q_dg{oWX
z!H|5iChP&g?N|pVh0Su!T1RXh;+dyuWDSI^Bw!LgSM)VRugicr6)k8~UD@eg2mb((
zaxePm{{YUqL0xNPYPJ+nOa!8k#yW$~aZpZ%MC;R>de8<P^sfG8tUoS!cL#+YpIU9c
zl^ho4dB@0u_h^5eWYyVQ3^uQA3@XDCNc)eus#?X=vFYO51~n&~^)(wrPh*4CWc~M)
zJ$b0pftrqoI~)isp!(Nq;!A;V;n5iBAXTbc8OwAY4)KSPGHLA?3d?{`sHXV$QwAlC
zBP-`}NE~ros}7c_^)^$q&Jx)gS-poe`Q-*(zun@x0Pd1kv+!!TU9PZl54~{U^O4q^
zh^N%Ds!l1+q#^aJjf~rpOJf<uRu!t)hq)|$YRqV5nNW8$iaiU$qsh3IHyq0lDnH$>
znTB%0y6Gz`95r3ovE)|HmIsk19ZhQXIn1~rBI4pb>=(=}*N%VvRm*sn?N-{VkmRI`
zk^HMC6m;`8jM?5vAR<6OBaeFJrB5zKQPVZcRU6V{(Bq1Ha+qaT9YL&4gF&1>&(l2C
zKC+}bFzzd*>_O1lFWUScmF7r!$F2=O!&>#$oqDibZjL*FwMpro)x`E_=)Ia^_`gn`
z;_)t^B$;@AdLLTR@Sc@Dt*ntSB)bO5{Ho~GR!GjgU7I%-OtH1_{-Qt)Id%bk54Cw!
zrLZ@n%=tiWMM`$&H>D1KQ?ap@C_#9+AbxeRVQD4UGR{aHg>dyfs*#J2GAu)=>H2lt
zakH-C-~c|gKZ$i)sgfA%qB&;!)zcO3JDx>H``Fmax6vk&RGRKak-FnN3hFeN62{?X
z+6K|ZV&11@W1)HVIZKnC^@ZXe5?WZpy6#bvz(L70)#U7jE?6s`db*fvkF+L2ur-+`
zWOifRlU?VM_ZpFs+zQsvL5EqlCycf!q{)`dyW@FE-CKq3ho8o|R(2x<(`!;;@S+m6
z*LWBYD)as2{<TruRX!7!)nW3rJB1^kFrU_{s^ri$u_#|kcplUMj5({26C?$3_i2JX
zL&Ko|0Abq`*#&Bc^{k%e%V9hEP{%n3kw^|B#9D0HHR#qf@r2LsQGcCvnjWndt6?0-
zc`L~DuKXoSPTkDrjRRc&0Ew42$LU_>ABe(x9)6=4_02eR_pGZ93qgMhL(;g-3c@ni
z<+~^pvYm+Tdz=sJUS&^>nk|g*a@nqq%bR%r0Nt!)5g^p$W?0}IWN*%|M1nZg2S71U
z_AjZaV7p+FL)_qhTE@3;wQ9FlJylM{q=w6~oJMO7{!&>Y{{Vp+rLYiNjnl87=C!oP
z#;I&OyrcB0j^s~cqSm#m-9F+gYe-|2rXMa-EDyDE39R2~y2O`Eq31bS7G=(UqO_BD
zFsBu7QXNxSyVd6s+>jMU5;YiA9>SwtPUu|2a3vw&0Z{q@+Oxi<$|)-vGwK)G?2$sT
z3@&y>65wR@KU%qO9;IOm+FVB@F-+UGu)u6qRWD}EX{f72>9m-&dnMZxG9+$6Y_Ap2
zYS%LxCAUnLm4*QBYb4~L(+Q|bqG-jv{?Uv{HNcTftFem?qtd8pl20a~JntcARZyqU
z3Mk4-c981E4x4cylf+li-&`bf5E=4DGsvf=yMG)%yPE-+vkVTkQ=F8A3Q%#_=k=J{
z?l+aWeD3<!I1UYKN!=4sS)I3o5By8zKlIc3S1+s|_?%t;0M}Ojb<p-Xqw%tlH71xp
z;&E&L09{x4(lmP=H;tcc7R&0y*C?$@dxal_n%wT0N4<91OokwYbq2WVoz(S5r3N~N
zf>}mcqu>u(<y2tc_N=iw^1mZ2CY;gi!HrF(w%gPmxvqG8XzHo(GXq^Y*7NG0)}3xK
zhOO&-OjLh{hg?bWlU$|zzHQSU!-`avqnTV)+hup*6tVOIw`3vB;r&kTVCxV?Jxd%{
zr1&#NYflcm629Wj*}(M`bmp9_j$BMyh5W}&ZEvUBh~wbn<xyPp`b<|-wAT#I$JF$!
zD%IupQMCwNb1@S{uuu75r|VLGW7rduYZ*S`vG$VgbaEm~eKN+{b($+!)zt7nt?6{@
zEn38`z@(0&s+#GcR#&rU9wG@^Na1x|KI==E$1yCK;j1*gdKBDbu4gNv_KSP=ZAw3w
zg@2Ta^lMX%E>`pn{Hs)s8vC9@b0BG@+@ONT`O&0k@!#r^s*(^L%sPxxHhLqg@g2|E
ztX9f1k2GL_dM#z>mLFz>&mjHdAC&qnb7E!71FHcG%I1qUDi&^Q%02j_bHr_pEj`$O
zD&nS8ImxYR{6;uW;S?rB=V_`^4CHmKDY9!tG*giZlTlY<cSn;r+1Qo<R%9opYp=KE
zMRQl^Z$)TaLB(|%bNLz{m<CFzHH|GB9erZGO1>QBZcnQNT<z+WyO<uk&-1C5z=9vd
zkxEL%)IZh_rClG0(t8vT=33mcKX$t3vCAFV+v$Jt?!Mx>TXXxh`c$G>*=Y8b=T3%0
z3KwO`By-lZfCII3t&T>O*icR>h=NPYcz@PPi}=$;>@r4OA1muciRVWZ<E#1Qw2Bvp
z%R<=)u4|3AI6T)aea_4_s^+97tf@7_BdFKYKjYE<rlquhI?Sed0FR|_9g^}m<h5_i
z&E2(#J*uqod6y{|>D1RMM{h#;T)@M&E0=%ut~1cq^v)bW0C7^oG_WuzaaSdDbHY|;
z-JSZ+g?<|qW<?KNlZ}r`ocWQRc?MM0D9jf@hXXYz+o9R$^KXI1dvva5;^}PsK=88#
z5)+btO6{fa#Pj9xNfkB5ju}#CKP#><2XR>QTM6I~8uHbo^kb*2QPlWmGX{_tt~TU~
z<Td-U_p-naNL}BFHK#;jOZcpt!pn_9Xa4+MJ@ZsNQieTMwvJo_T^$bDX;u0h`hs)L
z2|a61Ohpr@2LJ=l;ZkOiHUSOu15-l#mmZWJwJnh)Ee7G%jCz^|Sg=-W=t7L0;<o%!
zaF7nfny!aEPvb;`!v6sA`7isQ)~RaaI<)@)@6zZI{yX$L&xcw~yJ>UYTtubXTp?kf
zlmY#7P<WOlxcgy%fgBP>%m=7B#ah_#F;ANi%uPqZZLfWhrP5mzRxZvn)aUxveaFL1
zS_qt2UfbRO0DTrfr2ZZ05Nmurwv$cp?y+?!D#iS^2eDuO09|b8?qIXkE?27C9%`1v
zKZzb&9=mxCnRTWlghO#A_8I*uUk&JU_=8u9$`OFaF8=`CUgnu+u4<vYCu3mNQK-|P
zkR{3i)E+k<&b;Ht-X9uPsxPfqHaDZpjfX%_QBnrS!p{*UrR17T!{IItV#(k)rE2Ri
zN#LIqN3GhS{?5ERZV`GZJ?dByd`Hk1QHI-Gu#8Af-!QN7spx5ZH7d{IeL@(cF~@2W
z5T~&nYGhLCblQj}DTX^&CnI7?^YI@M-Gul7*!g6drv&c0luMi}faFr3z4REz`-vG|
zCxoV;vUYUi@S{fYE}^5aSfno(c;As;i`@B@HyK?WmA{H~4QekZQiURrkNIdaKkuqn
z@K%Ya4>H!}kz1gR!5^&$O`%bFoTR!;T7QX+ly34QXE|ffO7trf`#zkfr^}kn?segn
zsoTtjL1xLV8Ss|7bM&mEvpou*6j7Ngb*#Nc<!#}49N;L)tRAfCS}UAQ&5~*p{if<-
zq0b7vg;}%H^oWMx2$YVl=Dn(Plwz)UHET*zb}mPC3{mWfKwi8I3Wj}KW&+@2y>wk9
zb6K`AHFP*4PJInn{@2_=2P4v<$fXk_Ux?}utgC>?dCy!{BI(YjcMJzSS38B|b;fp$
zi{%<SV$4tP0LqSnv6ymHj%sSFBvmJN&qwe-{ELx)*GK+!(gRzsg&j*_b44MA27nrQ
z2CB|vKPdTEx%Huq6ydQ{a?AiwcVql%@BWu6{{YVvKljn8=ma-|5~T9rBqunoLs-81
zed|aM$(#XLx1pU!rH)SN4BCvX*f$iVc8Y@{OBlr3jl6Hlw5kv1T~~~XE}INJ%jA6k
zs#_5LBytwh{gUn}lN+O8MRuMffpr}LZX-Lk;nxGLC9zW7GmEw<6lCY-Cz^^l{Mmp5
zEbUTicb+422Zt3$@oiK2R}8yJr!CCstqU<e7-ywASHhoq&#9?a-(y{ckYqO>omaiF
zlG0)bC3_m#EeLO05h1I#t|JWnSl|z8mRaexaYHg#-M~ON2uSoJn%-EycHGxwPd-FT
z*YT~p0k+F=?~-d-7#VCgTm>1-pr`}6>M6IHEO+rI+GAkoI5iTIPR!_vmo3qm;teh#
zrAE-EJf1%a^Q+h}(l8%%jMp`EZ9t^H^y6H|QFaC`2tS1~YcqQ+QI4Cr`d3B$%Z0Tq
zUF%V4P<@KX0)L3nEVmlkz*Z3oL_-(GITgjr%kMK@Yv_$Too+qR#<+dB+-tmo=FWXC
zIV{v}C>xH|5Qij_9N#n<%NeJKUY-`ohE)!8>sZ%TdVQtLlSt#{IRc4ao!Hu&v{j0a
zeKp7KtS!hH`@~m6WV)N$qDbN}I^(@(H5mGuy3^(55p<|z)ot*lm`J3t`qyiy+rbPm
zWC4mJCnzz_nuOr{l-}V83m6=Cr`uXyY7WZ_@}o7HyJfl9uYaU!kw#tQWn6WyKh<<u
z=SDI+ODg9KPIFsQe6lc~yo^PWAcay+3h+fj*I=KJXSH^vbA_6Au@#IL;fSpbH38Kv
zeWVrD$?lT&Ch@BN_T91Aiser=&i55RgkkostHBL}be_0B&V{(Dd?q)EW5&60k5T-q
zHYB9MY=`Zq4wMu%@m+U>bhvM{%fB#u#9$x7hV%uw+G&yfn@)xi(N`5dd92M=!;4~?
zTZ#a!aEUxk6o4`?!3McKDp&B|h?>RN{hD)_iK~OOiE8J3t{npo;;_6){{SAR>0eKW
zcamp3wmkDcG5XWg)-_$n4Tb0MqPdx31q-={R_j!6Uj#3p?cTh)pBpsjR29+NHp?_U
z%GOfYB2<K%ty6L3BOj5iVNl9N+sQ?qknPZ{Wd0_zG>E*r6=V0D?mqOo6!$D#ACwPX
zwS{sTd#N3M=sN&CLBhU8K7c;nCXHForYh!^cWbtb{niz6B6}TW(Mw%W>Q2E$owG0`
ze-Yq*rnz4b*qi?V5yK^^G7UvmL?CxOALUv|=cerLE&Zp5Y!+KHEyTBR#zB#IUo)`B
zrCPGkv|*w{aeVhFb0}h=M;vsjr%g^?MY&BOZ0&9B#4BaJ=aEXbM<DxF^w#p)X*SmP
zIV@bdt8=ttn$^>aj>2+$$l37GmTTFjm6^~+LuaOYRM(Oz)nQWGM0}j~HIF20v<tp>
zZp9nhNrXY9VCd%{hRsK$SfMuuki-g)VIF&$x|gJK6Tvh(t={4CiBI#Z5S_@#dXf6q
zC3rZ?Q>L4|Q3H&Ejo*srFJNeH7-IuGQ=s}0RcLkI3;zI+Y`^u>`d2Tkf5>9*{<^pG
zuE(*(AB~jiZ8U$x-qim9(^dX-jULALj$7xnUf4Cu14>(kFNC#Y%;48yr#Aw)>uz__
z_)kUfrNc>N@iE9o`D-UpfZE>3nBbg>iki}kF{15WNTyFUJ%RrKPLJ{aRm%^J9hE)?
zW9x9A?Fi-Uz>n6FF@>$`{6tXx8Y*ftd2lmlwQ|L`F+R0amV=h&ri-LDr8ZkD67@c{
z*4x@!+9*h41Ju-2lh~*(rj3=GT}kN}>T9NoYkNB)hD3~jcBwsUTsl!)%;LqjHlU79
zdD>|Uxydya`s8}Ey(E>8hC%B<kWpY`ywO&yFU7RqH{5Quq#rA}l}PhNqW-I89-(M{
zP~3iEywg>*Z93y`CN}%0`d6V!T(h@xiF?rWGt9h(Q^>DU(C<Fk;d3(Jv5m3;=y6@C
zoRyY|o?6!tyo`^-0&`tghP3szvuno2WE>OOXhl@mz`nVcRM#*V>@T%_g>MoR1snh{
zE6>I~Qax-0Ef&a~>LS@-J?nt=(YfMh#+M)a<M~$`CpE24i08t82&(80TCXC;I#orN
zG>YvQJ*l5(?~2(-<CQD2GP_t;S7H0eQ2L6ByCXQO8fu($u8uhTzYnsX`Q}Y!Pj+Wt
zW$IJ#+YC;<$p*2tqzS3a0PjS<hJ_#FNXeG5H8Y15R6ZkGnna-&`pVq1<85_MW0nVP
zrJUMEeXFCiY{I_Omco0TuAO;rZKT_orbPLBpL!czw`L1&l1#^@7aysvxW%-2^(rX4
z3b$96P%(}sf%Ad51B!@gcXO8eRpXG~+&gjjRcMUuIr$b~(d}Tt+qjPBy<13QjA0M~
z>?&Jq)-X{xZ;0*(*^qh@Tx2*PR~>E6>K_TA4**rAVO%mgO^r1F0HQzC(=^F`X(Rk=
zWuuA6){`=|YlCfi>Tm;cD!r73Y1%``=qom@CUnK#DVrOy1AQx*er9ZX*0DJ81JqH~
zg;B;xgsI5RYW23D$$=*$*0Y>$jp<L7mYvdc$`ET0QIBce=xdtRXQdfk8`=b$w)?Gd
z-ZyD2HT0MsyI6h|*Gu8~oRa=0$g!=Tw&((Qm>qyVM|!8C+@nAPeApPpZCMdVYaNkq
zvgo0eafta9<+8|Ty}X&S3-G^%Xu26t)=irJ6ObK8Y3`U5-xkIWthnR%Vz%gd5?876
zIkt!mzlp1mkF$L1gO$Ys<lLGRjZimIMNbJO$Sbr^chk`#VlhdIM?j{R9`p>qqJVW+
z0rfx1uXvtNZXt`&b5+phm;1zh2KZn8KPCSFbNbb7ZFXwX{{Xek9R|PNAED_Q4cvNn
zg{^H@=D9H!#xcPs6~tR=7uOywmqv+3+2lr;xX#_2ny!b=RClc+gZQlI`c9!|6Bg7Z
zyp(QZor&A})?bA#wJCKeR>>ZF*?<v9PDE!QmOq6do{N5`PVk<r%mPU@$IXs~j@;w&
zsVx*QgY|hoi|p<*{{R81TM+&vc*+KnRaF@*a0Po7fn{mnF9^x1UgPa^F5)|qdMKfm
zT=0Is2aEJut2^+fa#%>1##Hnovh6hWoi98Ss~nUDO9`0u9kKMJ2btd5UFdqOmhx_j
zcps?a2eB3FejL^<d~e~YY;X3o>YF7~#(fx4djXH6>KA%+H$EQJOo}WY&J&)3F~}c{
zOW+xQv?aXMis9#-nZG*C?uz>?>TdZmeJhd?`I3{+X1H-55<7683l`x3UAU|}$-KQn
z(Ti;YKuPwcL*pkK{02$lhy%e4`Q#m=@vdz}dM|V4UgVr$RM6f{YQdqE(YtMGl2c}{
zhWz{8XvuKQ?!bJ^a%<794%ih3Q^iEKXSak`j7Y81GAJXA)tlx46?@hz;b(OJ0ErVS
zcG?d$%IlY*q*y}$KZ`V|N1CE8yt!#MVrl`ImPlk9aqCnK%$BOkWmWdhb6(rr@Y2w9
zGF?nnV+G3|Flsx5xK}FZRQ+ll#;ifMLxG&1YLYa#jd*&@y$euWk(H52$iep&6mZ-_
zA=<)7y5w~p*r70JmRS{}We50v?kfszWX>=RMMrdNOGbJ(f<NS3i~hPl^R9>**?cJL
zTMnfLkivsNF{h<RvnMUjr2s`N9%CG_3VMOwlp><-PyiHw$MFQ>8zPzcq&WO5gPDvX
zMkjy?70mZBskurB8eg07CZo6tpD3hBK7#?c)~(mkw|-pK$BrB3wHfT+(yO5#X&jC#
zyYR%~{{Td5X&93pQ2PVYly@q799`ABT3yQnDFqmeA4;D_jilABl~+H!1khhYr}3Qb
zkih`^j!7fwTzKM?w+ep<@c5^BftfcP4D(W1arB8{(<0P}IB7WgR<zd9&kE&323pl@
zVN*tqe(y;!K4Ux{)oV}E_%A;|M-)akVJeo6hoRU-Z4R6g7RCT8?OVm;Wj<R1r=j((
zn`&~l+SII^es#GTi1hDmzs|LhiI*TDM2o{l!GZLx*yWC74z37ckPo4*OhjGOk==%p
z<!O~~EE3`ei~s|I2(Kp8Y&@%r+h_S&F}P>iv2*5*yPBwSyRD9Nj2aDOYFxD=$SdD<
zN+)OovUS=(u9&O9TQZK9r>y1gZyRh#dCz)XLs)y0_G>Za!n|?bxmxC)yCX=gEVpD(
zhD7<AyN?3jznE>M3{wQAeLX8VYR>AE)M&C_6g|<uMYD%!b;kU9RSQ24D<3xTjF0YW
zDod5V=QTZIjfib+?Z62bAB9NrF&I^@SF5`=u|18njd2U(A27+SonqOYW6M^t_!*q!
z-AI|sXh-oH>wGC@QrcNb<&Fgum%Lp^XqxwSe`vROokIo{=31VVx?R($Vq<bybNnZ`
zt)&w<OH<1Ar;1HZV-MWj#Xn25^9Yg0%BH;vG1})mw`RK<v8K39Mh86uit1)>V*db*
z6dNdCzW)H}Tv@IuZYqBW&cB6pehPp1m`a)InW1hu{4Ay7VgCRV+mERJRh)}~NrKr5
z1`jl+r2`{wYqjuHt!`E);7gneQrK&AuZ->7o`#*epwKr{cV%c11dUi;C9?ZvftE4*
z`*Xbr_U%n>4B7D3q%ZWyoS&Gk57wLFHU9ubZ>4<>UoK>qW1ee9dAQG-<KCtMl-{C-
z!@#2ziJ;QkdCuH(n#i5Zk`_L-<<$7ur$B{6%Bsfzo@;2!vfCIKMeKW4Y$7I=9wKQ^
zS3G?yTG~H0bWc;wWZtEro+g(u!)YH`U#GCCBS9Q$gQAR6dkyYaynTqx=qea2l-Wm-
z^O3lptsMo_j!68lsKdTZ8ZXVQK>o7Uh?#75J|@(l(sVfd+jwqG)8V6&`=kE=uUS^w
zy`{~rld3dvTdjq`iK37ARe`~)Z&Q{re775GqdMysqVrCM;?7O-rO_GcGPpIP9QV=a
zku9aVN~|AoWX{^`!>(AiQ)#VPp9Y_IGozwnMor)Y*ne8tNauS9;kx-{c}JEU@&|K7
zuc+HsM^C3(Nfa_hh*mIo{{U;=rF}jbm^;F(%sQU+l=XWY>UMWbQ)O!nsaYaW81SUk
zWRBWB-a;zw>P0JRhtDRNhp*dOC6v<b3ob_>_2#*m5JxJyj0OPyHJ&C4RIk+?_*UF?
zAh;hf$6Ct2h&RpO92)a!XwI3dqj$loR>Exc^FO6>y2t#6F8=`QtA9H0=y}oj*;71K
zohzMM*Z#VX=Rt2{Z^u9DEk2E0s5DD)Q}|JLP8pKA%{=t33jY8R=u!Ai)57xz?${Q<
zBD8!(Z_;Jw9e)~2S}ew1?j<vx+$ms}Wz=jvYnC4pJ7=s}7PTJ`H+;2J2LiXP@tIKm
z8Zq_RHrtQ(jw_!BBoA8BpTZK{?=*;((!o@4am{O3*-3Qy#t?Nii(JyNa*~8@ZAmOP
z7T^E|2Lslyt}Y&TGC9fg^s1{bvWzhCZu4bFlNAfkj`fl{$c@LnC)$1KfE%BBKGi@Q
zS~jibAG1o_{{R90l}E(-YDo!jxkmp0$4zv_-YNDtt;to#qde+u_VToE!3MoIz^6VJ
zZ+CCyUX&fx&J04bvFDy4v-=*kBB&uGZOTu4^{%e^cChfrml+CDFrfP96xM>4yr(;&
zGRbnzFm{UCXRkLD?=#-PD^nqNsN-$Zp!I6xJr0KNiW~kVk$=8Fm2wPgTAvZmh5iv<
zKutyDgZHVUNh^(QtJ0Y6PXetOy$EE4ftTQ()zWBRJ))mZD5~!hno+Yg#}(0DIEz9a
z@qt;=>T}jw$JER4>tgOi>{B(Jt`GQ{P4yqjll~e7{AcK6%UGJH-W7B{BTAnMpBMVU
zTXu(~bxh@f+34-M%{{B5u{n&7N|Y&Xb~;{@Yc-TJ$h(=Bka-oY6|69*RE<C#IIgta
zjt)em*)0?nGD+cx6C{eb^)*d3n>l6?LnoL|bsP$5HcOI60pr`)?sY(jbLD^!Ju98J
zIIcR|ow$F5(2a%#SPUzMM@h4#9NE9r)_t_TP2!W&IISLvCpuSiMpuhVt;`OJPUEzn
zN|#OwkuL_WYRsxm@HE`9w_;Awx2JmN^ZUcsuUaFq&4?vZG2WekYK}qUBy^?_b5h1C
z%<-NZ1Ja_pxbvf6de&|?I%!q2vo&-V*kTRxoYo)2OES+5$82q_X)kq-a$eCgw}&MA
zZid#*S8`=Bj@_#>N=JhG?AXai+tRu!>X9_`ayw5I-K~#@pC<-EunO}Cq>#&o!5GeJ
z`il&@o1Pg^y2M+4>Cf`3ekP1IsJS2V%vOtI(@OPMp^Gh&$&I~1rps(N13Y!18%mhy
zpaFW3k9v`I#_CD$>rB=v#g-I+RGSC2FG>nVw}+tfbs+x$^UX_tN?#I?+KPa4h`=>n
z4qM+5@)yGgZS#**ruc+sR}0*tid_Ws#6Lr?(r!M>rrc^bO0p~pu`U4}g<`jl?B+7a
z+8mbi?I9;c$<J=ywQTv_=Gn;)3F<nP{Cj4Brnv@Z%YYF{>QCqSSAC`EI**5R*0<4O
zvwNur&vw%S$LK0bE^0m>)pbt=+Ube^03VS|k2A`4hsJaK#c4;RO>g15>lvkuWVl99
zbHK>OTG)whk2G(GTFh=58LgjhAgi4J04mql^e+~8f-7jHhBSuSFp7dtyhrgLdJ8Cc
zdq?pnhIN>)BeH1_HrZ8J9n0UndLN2(_0+7P)a+(RwF^Jq!}AQERW&SV;=ElAuZeU>
zHOqh9>IgPZ08^GV&-h>acS_XbytA^Ihuw7}0hIL|(a;W$#oCJMQb`@eu<6jt8iLXv
z-jKZQ!0TH701sh`8wnNy%@+k(mjsV-S-sHy!#6jXB%Bk`u~@!Z#LfuL39bs;CU)WU
zx+^jl499_7uCt##JNKf8#AlEG12(mUNpue|kC-Sp^sYz@SEA3Gkg5r(Q39|h9co#f
z&d<Vj=`Hk=C?6u=@m)p7GKC$hIs7NTgIY(>lH3C*2en10N&C<TVa;Ox1Gjpp#MJJ5
zt9c)|Q;PDd-Adx-@<(PooMN}Z$5$gfT$d_|rKagmbr^ezOmTttbIGgcPqVkYP`2uF
zxdo3>dRM4;?dnXg7N0u?k7J&kR97=WdxdE7M?$@6yI2+PawuW4fsR&8pYNJ<>k}aR
zGDnhm4_KORLTv-Y(iHMYJb>M4zDvi9l16wUvn)x|%~DC&K*2cbDuung65Kp8$H5s=
z4@#M8i;jmw;f+=+Jq|eCfFao9KJ}`;B7n9xkIuJG-A7kCM%AU#+{PJ5JXMNCq1mJ+
zfrT{CA%>uyO^8pWOr4M%X199R7h7}ksKBmwR$xRBsmNx@tQpGP*!b|UkfO7vIVbB#
zXwvYcZ2H<b=!$dsQuxW(>mf35A{ZZ?K7wBv&(}fV`?iNvHsqbN4xaRu<;h(!{xP!h
zH5Ir~g_v@ENUGixiJx4Fg5^*WPp1_6i9c|?;|W={7~(&56C?4icA`&G;qakGDZx~@
zADtqOrdy@E@V%XqV{^G5Ko43?IY^Jll>G7W_*K()W_6KP{xRmSxv67yz*QM0wT|a3
z_BX80l@FGn%PS0EcCFnvPP~%R*uG==0ASn^T?<1S=xo8JMud5cT>h16>;<waoMRPN
zyE8Kvm1R7G#apo4ZBao4u2lPT$LUj1Xm3gGi5BKa-(-vO*A?RH5c=aZT$!LHtBS2V
zK4f;$o!2Wo{nr^C#YRP7YE5i$x~3+&;qZx&>XNfXZD|Jyw=4&9nqjeP7}1>w-ZBZU
zT8__@?ZTza%**-)iE*#r2xA0BKKo;!uU*r0ts}$MaNw)Teo@aP)-CF3h@I7p{caO1
zkvbp)p4CY$KpA!dx$69E=A!+?Hd$>TT*)Hw>T5RR#F}&@sxWpPmzqiqn~$;aLN(MV
zusICiboZ@a5%`Pjs}9?F8*x1aZyI)kMl${+On5v?aXhOUSjtNr4lB`g-COKdF-G>%
zgN&b-y-ZSrIz5x9v@R!z?HPi|-dR1HipudOk#DM7z06IYkn}zO0R2_1?u?~MRNuUO
z`FCvb>N3HOLy=on@p%^NxaS<#qf5EQ_Z&&5LP(hp`$D$4EZOmx{{V?>ANAAvR}e=y
zu1Rhxe+gpSOufB7==Sd8I3x}Qb$%M~{nnjy*B1<+vh4~vp&?F6)-e1~g*EBH?~mnK
zfPA!`<wxN|!ZUPIbM02sN7SXkOF2kA4&EsZw2gf?!1|rsh|(gWXPlmEuzw2uifbtr
zV4^dI80$wZ$o7k&(SKz?Bo0}LBiD*cojx-TVF6ixS~*cFtGSFE^G)+RR%b}rN6S>D
zU{tB+G{nq*4ocd7uXCr9@|N=>`Woo9!{%7$zAM<vC2G9A%;v6-IFNZ$F!ePcaf;ff
z#L=4AUE>snTFI4i8>u$HNF*2I>MJaqn)2uIGhT$2<yK`ZaJ}l>v!t;y#0cKp4@!mz
z*td_$g%i+Lo|>P$c@OZe2h`Rku`;e-?S3N1RkvIZt!Q)<-r_T_n|sF(c^mYt3;-}X
zQ(@H1zJK+sKZIaXXnbncqy4d3A|<hR$4pJly62)EKN{tC5SwyvduEAlO+SRv(q@&N
z*UfTpHm>7})AQys;n=S{jo7aGKf^<mbTsT1S!7j=?<C+Vo@+{dF`<LZTcIJ0{J<Vp
z+O%-yMsk-cNRrO_<fu=!`AP=?-x;gZO?3^rO!r=5xx$rd!ca?APM5We%em%}Gv>Mc
zYeU0o!KG%Ca3CBL(zAlN>bpe1)xd2+a<c=6BW~>HnoMrD+IVd3t>q7L#`t+%l;#Hj
z_Z3Fx5<*!<anx6Z9U0d4H#{nUSV~9y^!}B{YX|-(H-Gijznyolg&ueOyk$Jq{RTfX
zTDKmJnrNr7uj8-vmY+thPKj<Re+o1-)AN2c)7T!B!CiaDp-165jS^6o(dWMv*IuBx
zwZ=1*Ak(F+S~>9=R8lzS1sJL74h=C|6G#5=u1Nexd-zMb+<Ar<><wdPwyW_N(0{u|
zCy1N(3I6qCSpfI21JJn5ZB2WDFNpMw7R|44FhVKitD}}kbo+1s1N)=aqAQjzOwP0B
znu0N=nXX*Bo`02Baf;=;Zs%G_D<YKV6+&qUHDw~RPJL;<cfA7-JKm2!-hm3YEgJ?_
z!3QAMS7Ch?t71}27XuEz*0-efvCD{#ic6ux_?t>=X~_zB`=``bZQzvhyeAnvZUw8m
z4)nKUgBhw)?sFb4vlBxN+5lM`DL%)gH^tHz^hlI(y?{TJY|Iw6F#I-fXFkHV+5PLw
zuko|#FrUIiF7#TR*RYRQ70$OiAHvQ502H72mc{=7T?g{6J&kQs<2kUu!XjqHM$Anl
zZy~D-Kd_}{iUEK!-nDB|G)nBJB!$|#PY#^ek=!*pi`_XJyScOm;s<kE`l=MsLLQ(f
ztm*tRIja0o^)mcCKhxj-xu51Lp1!+vDG#qSf6K66#(st@wTG*W6{@508dUg9=s9ZH
zwC!DUIUsgAA<eh^3c5=J)Knq8&p^^0Q%JYmzzx06y*c%JH8@E`$2i1aF&~w6#Ws!^
z$tFW4qjcwOt+45w$Z`60sMg<7x&Hu<i;{}nqFe#_^HpSw?<s6P9TF8gq$Wo0T<*{3
zR?VC-LcwDP10Wu0Ib3Q99%14eLM)c?uw|BJ-+|MfpM_$;E0++x<F^d&0`zAUI5p?U
z>^F3i{)Ybm?;6OtW{y^6;D)Uy@I=BN9EH|mTdNe4IopO9?@?KdD&n=N(3(FBMT^ON
zyGHk{H6t`fW1A7VI+4d(kYEZ9VZ+jz<S6S(!E(F5sj9@C+)z>{x4e=W7Y*E2pNcKN
zv^2GoI1!!4-n4?zgO|LFPYBvGYA+I;B%`mmtvlTkeK*9=+(x_LnTAe2?P+Vk)0rvT
zx+r+^_W~P684Z^FtB$sKmN$|@4p6F$_w81<H^Z-V*6_Li0ExUmLyyX|{7$4_M%dt`
zY0&SF@t2{G$?r`uXC#wCI~uqMRC?34s*VLQq)MlBV~%JYYD+*x9MDoXJTMR2C2rnc
zKU$@(7CNl?=ne<vRnX>-`@{4l@ZrX$2V$nKUt4{z#Wv7`g271qGJmBmhIOAUR{V~S
zS)8tcVP=oz%c)!iEC;_eku}ZCJ`s-b2}`T%{{Su>{{S+olYznQO}$T_hSxdY7T#ZJ
z?fspp*xDPNxk=(18*X?!el*VqTifeb$zdj=e`#>}+R(7ta5LV5&gMT0P4<R@!^Rg+
zaTnf+<phtY6}54sc#W)tcG|Q{D96rMKU$@+65SayXkI)YG+Mk;lh^<%q?&icl6IB1
zkw<f|(kz8^pB||uT{g-6%0_>c4WxMS+xJ>!jAQ-C=wMh+2zbu!Ez@eS**Eti82<p!
zYD4hW>h3T2o9p*3%l*TV`qCLfT?@g!9hxaFF5|h8GrxR+k4`e8x4bDdS5Vwtz%dN+
zZb6K3Q2rz*@G?so8cUmu^YhhN;{E)CrdZdXQ&^+EmXm0PXu?~!(zxAOcU`B``qQED
znB)HdfX&Tg38IztP%DtyO?oW(xefy#oj9Cfv7A&8dQXHcOnO+Elmjo#_*SXqnU8Z>
z)Y%_LhE^SJN0{Eis9wq%W+S&1gx|Py$M~$~bvewp0@XPzIrglpZ7)x^Rv89i)VlZf
zuSW|WT1lRMDsEL;DO^J}w434i)b#?Gx9s>+^LifZU1^+-CXK?|-s!NLhGb<J4hJ<;
zO_s+`nVF<w=Y=jY#S&`Zld)@4nP#^VLmEg2Bx5AjS}6YjS>-%q-y`|f5_+3OXjqBV
zLoCdslBXFpO3PG)m4-$PLF-X+$@NRP;joqC+Di;%pXFTKdXCM`a%9E^DrCxxZi`R`
z7K*@}WY8VGO?D{SD65^vg)SOMFivsBboB=D*jgzR6w^QqYJy-29Q4H|t_Pm#P8xZc
zPaC+bcZsFEP{)K7&i6R_mS5$JsO?$qfRmP~EZ*>HD80BnaG%z*JWVi{S)AkUe7}`f
zLRZFQ6TrqQduxxf-A6k0MhA+MQ?c<R15waL!U2>5O!^M>o8W|RwOmP$m&*e_`K#(=
zz6Mb74DBwta2xm=X!=%Wd90<mv~~g3uW1@|nx3Gxv78scC(@EE-_$Ij(m=I|i)<W|
z>s8^ErEw|(WKo6#xT2K3;Hw+Ny*7z`e5oQwu}I**6|bgvRusXSMv=X^#dO7LdDD6w
zJ+vBroSd|1PhMy`tYQH0uJPtFj@7zGNnYi4)Gj4+mfAf9YsBJRIJf&B%Cu_48K^-X
z^<v0`X_KHheEOcB(v<c!rSTTZ2YyckS1IC^mK)0_g?Bp^L-gXg>U<37rKHb4)$RO^
zLRH&<PC%^l*0@pH$spm#JXMY>s-DEX%UW%$cY1_K-E&&{^_pEl6ls7WZ<zWD<*)Ik
zbyJrtOL`upnqGr>J(7lHB?Eq6<5e_0Nu}2yN#<uc-M2NDMns!uZ{kb%tTfe?y6ym1
z7Sd{3k=)Y#o+H!L)X6Suu*3W#4wt8^f27`-4}7(0X*zC^ak-${<v2B*<n5u^%Vl$O
zPP4W9Kvj)FR~@MFjocb)+n0lQY~uo)8*pYey;!0z4cx_XGF-`OQZ?jp*1L->PSW<^
zTlqv1ILN04mW=JKg`8u|p2&Ua)1jZqoGP;njEbkI#SWdPuz`RRa-Nk=)^;(3YD<<f
z+DDITR<OaUJM8(1=Zc`u&$W6~UEq!i@b+~aNQ}akIhV~(sI9JHbF<Ynk*iwA6Bbnj
zHbACX_)gN<LnWb*=OLu#uxnGFp324ZZm_Qf0siRgMWiora*?wH1oq~c7b@%C7O=Rz
zNhR|OZ0Z{ss8>ePpg<Z3`0o8_G>TJ`M3P<UmOv}X3}d%SNOh|TXISEMDD^>_O{6s$
zGo`oGVv05L;wO&$RUZ&*7b8rPXcbkX!*VmyvgbFA9^w6~b8}&DIFlQI&fY6(-P`*;
zUP4Q4In7elig8IR9O-;!p%US}stwBz%wX1Z&4x4ym^zf7W7X{MS2!ojbM0MS!23>`
z#d#reE3XSv+18B7dZv#zzmxqf$4#dd8q6_+UY%F(vz;;CWi_1E64+&Hi3!4!)1_C<
zd9vM`ErnlNyQjsz;xKv-nAAOrM)0jTTw7`FA@nSN4wZ^FiHD?vA)4vF^sib>aHgBw
zlxM^H{Y!UI68@&Ixxy3ZD0_l2uA6eI{{VfmewAZKAMrS~{{Vbe$cb!jc;4D0rdrRi
zdB14cyW1Tq3)>5Ji&chACO9CrgfurQq;tnp-#Ike)WR+DtwmcKD_v7d5ycuAtat4_
z9s5&dyAfVL+qCozw64H5fO{J5qx>^?d(m2#bRDg5j1Z{<J$lxTocT+oka8o)UiH%z
z?%f$x(Uc@axdQ;Bjt9M4yLU;WwT}vspwB(}n&zmU>v8FIaiSw3<lv9iwlqXBwAceY
zat3OsM(}fHRhL5o_BWRDV9J43LUKO6>YKXxk}P;_xW#l;OOZ3L3kx!tZ=KZo)j4w<
zg@=BkzG6z#XzP~eN8w<yY4Wmvf00}^x5l$~{{UTE`PXEA6nXFXjG#^{XTpFsy%vAE
z-^#6S64{sha@n5QeH7OzL)=vU6kgN5{dD#{1#s8g^k{r1L8r;(Op;?eNHyIP{{RyG
z{)aSCdzfODv1b`GW>;@kCaGLZA}GZ4Vf^coKM~sYH#LkMDIeWyoo3p(u8O}A&q4c!
zABZFd2*+~O2ZwaYwH-<#por1N=Jc(ptH#9Cmn*sI_U&(`S%n+{(03-YrZ-p8`Jy&u
zJuo;Goo%^LsP5+=7Sk`@jfd1!QXw6ylFe>*DJQW)<!}vU&N|g1kw!+vLS)4>F$|Q`
zE-I`c4Ao~()Y`)7WgL!-Jw;S>cP2}jG<7<J5bIiCXow_?p{+eO*{u8`sKaXucF6QN
zHQ!6(x#QP=yHR}^&1kYFv!z8F;FIz&^dhtTY65BS#N05Kq54-xLkQ_R7&-%NX@{s7
ztyX&TE3bU|Oe6i$AbBcnH15>x$E<>|Tb+`A>pBk=oN4j@0M}3HTzfTZQ{p+WzrrKN
z<HblKR2!Jooi-qv=Q&jARf?XhYgnroP4j3lis(Ew#4?Ze$MU5=gy(4Jq946a_l0fv
zh4$I1=kGOT4}wQEO=?8R(4+~d<NM2skHo+W<@ZW|^6Ve+{ze3Xu=SyZQ}0fX#*3@)
zGdd2nuWQ=7CUZdRv?l$-5B961vOhIuf$nyCc9JfvVv<HA5%Pv?kIJ>6)2-lAs+dwb
zcCN(qN1oyDGQ!4+015^v)D$B+Al%FpdXtKY?xFw?ckb!8oY7!y9A}N}(KO#W93W@Q
zMtXzmTr54Sn!e{&8)jXPI2AJ2pCho|(U1Bgf4x~3rxQp&+BKy91as5=B#Eu0^Q?9p
z9l1~{owMvEax$j|wR*uaVWi}UwQHQHh#dOXaydBY4K=aLRo+FFZpNM))3pWy?{3uu
zjK#Y2qy$XZ9MB|RQ9#h#`HmJpyxsjPr@6NmdJushfPk|O<W{iiOt(iZXEW*7@w6u!
z7#=IVp86}>ba9Q#&OK`B?9rt6O!I#oNYZL4wRw#&2=}g51!$C^83(OgcROgt$s2we
zaTcu(eQ{g<DIYbt9Y{5z*y~@$iGsAGU{XhLov6hOW8yglaqmvW1tEIUbfi)`9}B4R
z=Wh7|vG3;>8jR$O0^EKS-p4IJ<H?qMJaMZc`#18glHSmrV&?C>HsfYTa&b*;b7N)7
z+wvgs{L2N6lp}|?xAPYV2XlO&kD(c>mo~yZTf*8bgh?E5nIs?W0m1!gJ<lFH95=*k
zwz~0T@t%+Y7(U&9t!jKMiWs#!+xvz6-MkyO{{VF7pVF(K&Hn%ZW#rbpK{IaN>?3pE
z8UFz5R_BPd3G93dXNihjT_9tgqjw~J8K#JrLznnu+J>R3$#ZdeF}1W@wqJrzKz~~4
zJYlTduY@GkwCl%+n}D&h9AIZ@{XwKz&wM-aW~+bVMYGf|rMQk%Ayv;pIn7*?#JaqH
z6W2%=vVD-;gxF7PP{9%WN%0o1qw6qR*xySGZx-XU_2#wp%T>7X27zfIX`{Q1nr-ZI
za0mF*jNmoxD&8$R;k=|-(^7cD4!Z_0E7ANt8rx}-Dxg4;Fx%-%sWSLTw5JAnhjuv>
zw~33U$Dpr1U&A}&rB~=ty#D}Hdgs!)Em1Rexb&ij#9^2}+#1^WBTI~Wk^HNcDPD^{
za$AN!8nvcawxN9$x`IJJHa#jCRHMx_k5`5jwX;Z*fD|aGWg!-^s?Vgutp-`Yo4;>k
z$*jFjb+?9DS+EoqHH7twG>^w4nS$%huuGxLml-+q8RDw468ZJ7W-8H0@?`HO$wOdc
z9Mwsz5_ZnUNdd?pRiYaxU&(oBV_Sv9MBowYRh$W2oSgfuXD6woZD4|VEuTBGr>;gT
zS}UmJQI-VydRA<rEBWJ!GO(uiZZq4OsMgwiF$TB_lJ&?qsbwT!Y7%+3ERnHjm+t%2
zD22NLP|cFi*t1;NjQW~@aKu-yHEMA5GLG!*d?4Q;71s*ndz}#2T3SVkO=@5;^s2XK
z7P8~}rjVY;nr~@t=MB)wfl(x(otyZ#5PuFUnVe*I5@3)I8LCo0&A_QNM^)g^KFf0{
zA2$4F+ZD^}Gkv2~xRH7@ny0Zb_?bv{r}YPllT>#87nyIgHnx)lrci^=6z>b!wTyR&
z<D{eld*ZDay)KSI?q;6)c^{T%IW=EAn#r9AmKfSr3agTM;=6wcXtz+@t<RE2&H*jP
zF;ef)q|()zQ}}gfoXdA9Vo4<zr{`K0I(?0f3L+s<_m5iAl|50M=H~2!UDe>zt&pmj
z8JBX7g1L#jOMP?*h=>!>dj56L^p%cDLfeqr$YgIm;$~t`b6c8K*pexfgRvZ7eQQ;*
zlHHmsC?w;G-_u!Su-o+pwU)+36&P$($he*+Jqg+Q(w4@Q{vz^QZXw;ux21F5ERCQR
zlM{fF<3C#EoUaCTK{YgR-Y1V25lg|`F~wtA*~@h>Sk#f8PC8c%b~{P6Yhh)2@vlWy
z{#8fq1XolZggqH8t^3<t$1?Rs&1&h|f?n!Zc9)VAOOupr)^(ND%8a!%V_QuU5oauf
z+sfw$)YPk|>Gn5~Y7t9^Ckm&&YlXRV>T6_l+P$&8vJy@=1tF1Zj`3ybx0V<qVuUZu
zqP3FdnnJ_LMmJ|PZkdmAYrXImopp0*EQ*8%Yy(QHJ<A<_-<@)u-jCt=1g|8lLk0w`
zLE{YpF9oYeepF+}ttxQ$PH9xNRbql7+rp_5fUCt`(WPOl!N*#!W|pU0#?8l-mwaD{
z<bq6Oup*aI{r85f=ItIc`C_HP*|S<Wi2m*`qde!u*0?d<F9En?=~b<~EYsVp(p>FE
z8B&8euFN~U5yhImO84$rhr)JfxRs=o`sWpL-$v7I6lMg#In12bLF99`bqjrJH19Sk
z5mi*<9P%rVzVU^U=XJx%bN5tqsBXf98)jA4jUiPZX0~T;`_3vWFB@F8qi_3^g?HeP
zbDY$38+$D^G`wf2-+4C1Iavu}4$eDQA2rS0#6hN-Ndlg%b3~q{Q%jVBFw!tll1Z-P
z!<t&b7?H=#S;*NVqgLXe)M7Jfw-NsUc@05jT@k$B4yJmi^sc4fZkA8|Gp!@k!ly%-
zk+{u4B%5kYi&06v2Rv0vk&MtLU~AyXroIo{A-^i<ygPaJ%`z50#F#Z*1=PgVyfR-<
zl34!ZoPp_8lfzBl`B%3}o2dmOnn|Wf{{RTvQ^$tM>-?)<t6!-;`ORyVpJLRSA4LYU
ztHrGzT8<|~b|zTQ$~|jWN{a#I#xig#32a(4)NUXS^(MPNk*1iJtKkJ_9nE?XSw<nY
zjXkP)m@Gt|!kXMjr==z$c~4hrv2^=kKKxSZ3u7W<n}WTNVy<Z1>o&jdirEn@jh`D`
z!}be%ec!wn8}aRoRU7XRL#ho%!sg+;f~wZ7*+>JA=S{s#BOAprmlsxAuBoG<-lPj}
zY!&2gfr05-_rlupSk<o6eY{6>#Bk@ZuFO8MW|2)Lro(j0lMIp0>^6qv4)vzqOncr5
zywN|)@^%^TU6kc?)R?NLaq9)NS8=xBz%L|pH3Y~S<hJ}4W5zlP&a#gpNr^XmmH-Me
zIpeKsXy`;zHhb4KIdaF91x0mpp49xqb2^CtUFNZn{{W;Bw`n6ETIihIppM!$NP<#Q
zc2?wr%~za-eW#;zHS_fqs>ykqNosHSOEKBY>E?e*<UB_o{7tT0ca+yk*&ct}k&fLf
zYr+TqC((bn-^#RGlRb-GG=18PYv`5D;)l55@S^6Q#Ff-pbQQ&4bH0zlZrYwJyMuI^
z4&&9MC-kYIyE4V-qv~+%`ORbM^ORIo>P>S+YIZZD)g_g*II)l5?d@FU)Ut?x1Ow|`
z6&Jj6(0=KKtUJ0OVam58S8?H$iS&3Sf<1%`o_gZCX?!L)f#&W@$K2<>0Q9bpLMkoF
zHzYSc)yoHY-JKPqYF0+^vy0r)8yJ7PUB1m8PqmR&eJavVyqT&pcymsC{^(M4V$EZT
zsJfJA@NS<ja9MtpN&E$7dR#~ky-(HRYO%9LS$-d9{ax^1s%QGt5AgEf50_9w{{U>y
z_|*Mh%++JpR6ySk?}y#GjlbYO{d$G`1*wm_ByITOm#~p*8xE+{(DWT=Pc01T8$x*m
zKOomfc@wUQELh9=L||6*;^MV9>&_FGx_Re=A-%EHjitmABy8K1mE$zeiLsb(?eW66
zE%?@{`=m<BwudXEPvzb!w?-JPmzNB~+PE<b-aYIKy*m~6<YumEXBwrpf9IgA64dRV
z!p*-IONd}kQ<~+uV_H=BjBvlgLyN3S?^W2zG3$X-6mKrWKmhHJ=5Jq0PbERiw&Pk{
z%GdNFbI-~vYr{>CHBWL4Df}i9-0Ce8eavg>!nHh0#1Q`gynmHt4}qMM{wz#=B>w=!
z(jRhZ{wDtbkJSGFa-aOW8~!r%Ir(cBTOYay-ikjODzCuI$U2JFwLL4gXDkk<LT|h)
zr?NjiLfCTL^o<n%0K}r({om_WCcL+Sg2^hYH$k4Y)g6v0J2JcKGB!jOuGsGX04S!0
zyqNsbs;TMauOsoQwT$_j*nimk!TZJBbI$XSe+r!~kyCF)><1X4<54rbM`D*69nypw
zgk*z`c&{Uif=8uuSJ>;r?#q#$DpfV-v$)>TPx><t_NymQZ}h2$`$n#R0y&}lNfh+^
zpe~*ceqp$N6%F*}*7d@n3V8?D*0+kiG-E09)n;6`4K%sUN{1ZhG|ovovQYm3?O=LT
z$oL&-WReCaKD5aW0jUCv(lOy`18Z?0C(PSScP?8+_9(WNFEpJ`a4V_Pk&cPwXAZ%J
zJu_Dg$eGW5m&)8&@;cS{XKRf#`EqTL0=XZ<=}w;JuWOui+QQymvtCDn2H(V08;f*e
zWQ_270ZltI(Wj!ZYe#Lmg}-yWS4rZYw*LT)ZnigH#fgHn5raYAC#@OAJDNaA5gumW
zYLJh}P*T`v?kPw=g#{y~@a#(+i_87?=kTs#^Zq_1Vg2)u&Z+Ej(s!#iUhwm8)uLh7
z4WG)pYrwXUM&yvi&=noQG~==7<}cp2k<4r5yg}pJ3lKcZQ!vIls)L3H@dVdTrXuPe
z5H$<L`J%FcZH9X{`4tDAk0kL0)81>hvq|@i%s(TF>%1`Uru;nCu9-erh($i2j34J#
zwgaco^!10~?Uk%#qv{T$C;P<b`7K)U?AfquQH`kyIP^HCh@V3y$=Aal3mW4gZg0oP
z*yQK2{4-8?jmN=gbNnh_@^+z=a-R<-SUhKB@4(9s%n@6jAYZV2TeqFK4<bqV<bO&`
zXg)CvYg%y~?a%&&T}H7lmExZZzB8DdemGj5fM9rsO@zLwXBsnok)HA2w(>X?(dkz$
zKZfJHb=vd8<`L*frO}f2MSUD*eZ+HvRy6^#<36Ii_P=&_xBK6bpKu8k%in<7h(XsN
zQA6W7^9^YH%`Xu$rkft-1Lhyfxi`6iLCk`^7I<FHO_Zk9VqOtTSD^S0Ol61dlP(>`
zap~z&&g?Y~OUa#tt@Gosti3{VszUWO$621u7}+A`pBRHp41V?rHNtrFRah(yx)J6|
zdS{$amE2JY=_Y1)Zsu0fluV(ONeSzU(lf@O{^fICyk3WwDC#%QEI#UwPAYVf$9^f?
zmlB4jG!CH517PuypUSYMzLn-H0)$%5TC(G!2`*<y)X8dORyYMmTBCcYK>_l@KbSu7
zBN(jiXqCn%h&Ha#2tJ#QQJ&^YxtdoTp5B!$nj|%NJ?ce`iQ^)nnj==n{ydlJYP%oa
zH`cwnpTei;VIA4l_(Q#RD_l==q8kpSq*#>JpavSHs1ht^Bm0JdGtLlZQgQd6HCo(Z
zAJ=<-I_GHPBe)$Zqw!M9j>l`^D{Z<fNsv6oB>w;^&%5wpk;_}%V6n;GKf<clMru}*
zGj2R3sY3XT7GeG6IR14)T`u+}D)!O>I^lCt(>9FLG`t~q_H9w1bF@sj<aZq_x78wR
zO)2gaV3^s3RCY6;#CgrGmw$aBXsy`p`Lad@Z`pV$+=jZi#z);j{OUPwXy%dCSZKOt
zmY!gV7ry7``PJjE*;^r1Z!z$GRi!49CNY%R%j!NOyqa?yAVdeJB-HveQr*uVn(X;G
z$vLi^CFMKzM`R$Nu3zzeru$Tw4iIgwJ{ZmqH0SXgH8(V@tUSdJBVlWIOGw~?>2?+l
zS81wS9Leg-N<m@;Zs^DN5-<Bz(DpGPaplg-*9NAZH=4!PJ|7<`_o<e}LFzIMN@-9f
z>`fDajpL!L?-s{3)|DDU;qdtcR}9lvJqR?}!fP-}Y@%k%GMq5<t$hN}+uCe&!!XIs
zao<C$sgHXa-S~pyCjS7QM+4j~W*sh4e&PJ9s*-jwsVguVRNwQ={#2T#nLKt;UIieb
z+DQE?NyVk95tYp;?jda+!s6@ZWkN##0C?6j+)t=p$GM2)1YGp47^@zQDBW`&qY=1E
zYg=ad+hFV7x!)9cR_^y#60C_BC_eRQqj@H~96r%2uQ1YnAI#E;C%8hMfEx8*3+a$)
zns=0(fOFEWP`=_-n%LSxH<nt*@iXEGhpSZCDCiNBl-15=LzBl!x1;l`Td?)1It?Ef
zu2S6lxAdvBxZ%{iI<~SOZnVD0(am_5!&1j&$O)1>=iF2DjW8ye5)yc>%q$awF~&|)
z<!sK>tZtHbi_Tz&k%5jYI$bH62NFo^G0}-N=qf7LIinLd!nYxX#iJ9IZOT0ddggV@
zU$N@5!~qMta_6`;<~@yLq}iO*mh2MhCOw=J`P7;-)Mc`ZP_s#x0u+L|_pG~}1~^uD
zwvDgzC-teG<kI*`I$wv%-EVQio-3%0L_jL>iiY(yNU*0kr#>1sUr|8J-wUDp4A$i2
zi3<L;)LjoRO_5LVrYlJGFsZsZS!Dhir!=|aH3p^2N6HFI(TK?%MF3#l;!9an^kp>f
z1Ktw@+mrVg;;n2WwTvUkyN3<@>Tj|jJuBM4)V(WqGnV1@RP?15>FHhfB=a8~s816}
zqo<ai%DL#5e1KQZ*ZwTMO%Z9|YLmxIVzw-oWvpY`xu#k#+9v6N{_i!Pt2f#;SS`W(
zpE0S4Z({mX8l;|GqhqNU`p|=+Ewo}AWI@<dF5874oe|t-Ks&cN>gP3gL?CMx<eXva
zRAnuWm&6u2Q&|51XzEKl`E9XQCJUVY70T&acAKkftsHAIS)^sy7aOn#t|_#xV+l)@
zu@cuu)O9;cO*%V>jFN?srEuJPaa*zJw%3-3+H*b@UoJrBCqB8ZoH9=78%9knN8Bnw
zDT)G0sm4#Qdb4e4#YBn-RX;K9RHW}@TiWcohK@DD=Op5St_UPxHz1z%ZMh>Ew2Q56
zBE}h`Aa*2Hu8NQLSsQ8FyV9c0+8OgF1Dn;6*6LrE1(@TGqOd}=cOyAhM!{3onr~y#
zg}G3gp;2(S2Xih-&lOH`8Z|zgSCxrPSC#vlPWNX);X*#cFFbjb$?F)GRJ@PBnArU5
zU$T!qe-cJ3f(Zw$Y<NeFelPbM`BsRD?8$s${{Znh$E0GgawvO*J``H8{{UNEO^@9b
z#b0yKq41lQhXT8sCM}=^@3R#&J|<<0_*bdJ%sjSb{^=sQty1HC=!kMuRxw(gP3U%)
z54*$r%AO+zxrydij{x&q)LNsKkHXCPcDl^1gYzwPJ`cH6j$2-w?i|(A+;vsp%kxF#
zY7CheW4L49xAa#6c3$iSVzj7TPkNQjooZR(wOL`1rq<^j>k#Vqp?}vJy458a=*%$G
zp*nYHp?~5HN*i}*<OP84E0<r0+RRQ>mKMjV)l;h`jb(+VSzn17qYw3PXZ_RuG~w|J
zR6p|05Buh?+RYR7I!t)@g{XIi6ZllI{6)A2_m+y^LyBs<6TwHLEDwnhchjH#{txh}
zV)%<>$M=?!C;SUFYP#%X@Gqk$9}qO6e&mDjtC4u8P1G!>5zeSU$qsQ=s>PX03pnak
z(=TjpFVH2-Ou!R>)sOKm-aDHJr;Mp+IVafF#itV~#m1_+=25Xp0|SB$ZHc%1-`*To
zEp@X!oCLQ+Lv;Gp%`LZU7T@*IS32DErSLOn#svA+%iC*zO6JS$7+y_PPl(SA{3JJM
zA2f$Q?1~skF92*xMvnah%ktL&lzud9%4C{D<<iA=z&uu`hnp92bH>`ad?zf9!%JXq
z;r{@7t$3DPv{8C|*{q}dHgU;*`5Br?#;Fh1toWfY@js}=3;1j${A6=-Rz9>pa6jIn
zkH(9u@UtoowX18|yJm90>GUS~*n3w~WPWj3t%{cCccnn{Y0v^>E4Pkty?R!XT5TZV
zSB^rD>s^WHk3LJ@WWod(ByJ<QsQ?4ofusbe87BgxS>u1Y+rE?oXl2n4Ybr4c83^Rp
zi(D_0do+3I=CQ23(|QlyCyBF)urb=a=^lbMv{wDg{c9glKj~9;`HeYz5jo-fNf?&%
zn6=0oBq%s+dRC8#<6^N)Ki#gne*wW={`9~I*aY)Zt8F->X317iK9wR%&IKkS>|m+}
zZl<W-+@wP+X#rXVxVlp4yA%7yti>1Ez>-%oAK+h8K+AC<jCs=#WIW*Zt^WWGg78Z3
zo?GP})ll4tY-8%H<=x1c!l~w>(&h^^kQN`j8<bHEBy5%7R=@ardv^uquM~<=e|RGu
zPpxt4&mKoOu8Bs+ULJhLhLiJO+n-k-&bnU}AtK$qGgP)aKjrL90ZZPI-Wa4O@E)dt
zF(YMJR~<>~R*rIL97eYk1Cc?Y)9HJYKy%Q5#8)oaVoP#OHgiy3{QXTY3Bb?dI9CT~
z1e1?y=)7BWv1(QeA1eTPra#>PV1A~ld?%Nfe;F?$58?J!O-IDqhDuq+LbvyrjGyOS
zO^l<$-Vl9G=O1dmm`K^+oSuj1Gf~vzJa+m;n^3rsdOWPd@g}>iL{DSz&rPz7ZhP5&
z=<W|9tVeweAYW)&b~dDNnaK2I01y2L);_$$R@AN{go0f8L+Us*<GB;q!}zCut1XU^
zAIi4z24x*GM<3R<tbT8U1b16$Gyed68W~5PcvZ@Hn!tmP->A>yT}Q;1F?>O@alsKq
z{{W#H7|DJy5BQvEpY`s4^eXE$NYrXR8l8{L8)F{1U;ed6KnE)~iFFHB+vLe^vdBBJ
zZ<imXCAzd4FOb6<RG=P!Qs~Kl5+Tz*NfY0SwWy})6X{$(^6j0^_^8gDe5Z=yFBd*3
zPBILu{VCy>yiO-r7Q9;D=ut=N*{(UpItuT5o_rRA^ou6G)nSVnUD<9sSGZ}AEuDgl
zV*{mT_c~JdY!4>{_cfm<?<F49!BO1wV2k(ji)}EsPm~JrPZD3f=B+VePR825O6h|4
z{zn_q57fWl`DK#n;3`@zx#O{}v8>?9`IvK>^kBWy&YQICH+7^OZ7fe?RhC~eX4<$#
z`@=MbVd^RxQt126PBG95ZkKE%mC=X?9VlooW2Lmev`w(dXj$PHMqd8_UrOOTMQ3z%
zNv@}WMzcOwZzJ=p<r^A8O7=3mj&p;>NT@|*bRt-j2P9PL;15hvWfrWz<H>%ek~=Bq
zl<h;!dsUsN)ATu`@iVpX)q~o{^A2mHt|O{D4ivN$A*LDtV^u9k<4>Rc!$6+LoI-Z8
zG45)-N9BG$jlZ2>aSXU0Dw?MyLuI3{@Px&*nNzfHf&A+vc)tD_SspmuRfx$P)n3Sz
z6)7`D4;tICRhHI8^<YLTP7e@i*Bl8p@y_+YqnQoUXqpDot{fOVhhTCF0i4%Qe{8Lz
zA#x9vppbfd=BXJ<_eU;ueLe%_q>AJY6*^QBYqFLi2*NL7c{P-!*xnTHt0~Jo*G@dy
z8MhD*JXG;nsk|A^LBOmVI?`Jd=e5S6BKlWN;mFlzBe-5G*um5N)TUg%67PuA9U@SD
ztA>q5b5wru=hWQ0YU&>{6SQ;{zio1|>GHH{2Hv%C*yZX+46Metv`0JKf3uq1>|z(0
zOUL)-uGpQ?^GbgSP~5gr3gMIy^r{znQaEWLZKMyo+PS2zj;OmFq}J&^<#fYu-I|kR
zkR8JepjQi~&c#sXM6gZb=;e*USq{=Y4@_5Q{gog01$sC|?y*TGz-dvB;1yd{($Tc5
zfDkbeu>;=~y;$=>qEbnmZS8}P4O&Sk`J!R~_BGDCvPioQ8$2IM@uRCfMJ`$wJ%7X3
zCi>FZn+(yZFON}G=heR5txFQXLc{>cJ!+#aTO-L(=4{qYP2-Ph#)7&%5?O9E-Nyiv
z+KJ8B-ZZ0gQrR7IxaP6EOyhJZ^)#a2E}>+qXkieF=eWgM(sBK%2H#p4qBJ~RotFNY
zHO)hJHK`s<jK0~W9fr=sQ<hdp0;pwp8Jj+!3e3{*+I&rnWM;dtpJ^QUkA(@S+pMtp
ziZWOERt})z8wub0sB6(r-DWi2hn?uRV$)nkW+5acPriD7wR6Unn3fy01mtaF>FHia
zLr0|1$~8uCDoqbx(ZWyR*qu&5j2!(1M|*}ypBXsY{b^5fX>7%}!B2W5>sgx;BMqXL
z?{uaGFAI`28>s&P>A63xbo#OK?I-^Lbk>VzbJX1&RMKN6sTm1wY7I?2IkGX@tX-y5
z(qlQB;!-&mJu)jV!gog7Rx$RFVzgTldQ41NIO;hSrxSUx2K?*4FP^R<khr52>zSTc
z<L~(zCI0}fN&Kr4)i`JT1$Y(z0E;hE0AfL%zD2N?CxE|=Zrc9<$Gm^;*EGwCdIkur
zdS<e0iMiCSE{`A_3QJ<yrmO^kdzz!G&KBYb+m#1#q=rbknma10o=z#Z@?jqxPs%8V
zm@=VFh&_cSk0Fy@wa48OR;zO<Y;`*RxgMTjv$?T~?%qSamHPe_jdwg=71v&S7}i*B
zt-~a854x?`{#4sAjC8kfytuLZHI}`mT8mkun`Ds>ySNPGR^NxOCBL_K7O`8#ScF){
z0~7a~@U5`#_m)Pnjn=Gv&9qRW%^bdR7AkVMAdmk5U2AC3t<IBhk21tO!SgrLwzZvy
zCi!kjbppq4YYcL%o1MO8{3&mdnkeKE?O0^k3k+m>*4DHdM=eQeu4>ND8FrO{fS>>q
z-mGZGL1eAGEO|SIXxEjdbH%F|lDreiB+rBPXJ$D(k=ymfau;%mZ%o5GnX+@!H5wkp
z8<1*j%_1{)TBpmv!_vH*TYlb6opbn^zu`al@tgj-e@f&vc4gM?PJiXG{&lC>N1cD(
zGL6>~6YE;u5M1l1-iY7Isv@bg4;T;lp11vTTEoRRxS{Z()rS&SS7Xpu6@AY7KMA*N
zE4a5IELuNdikcg<JYLA+^@Uh3B~jBku1ek3Tj{1e<+%D(rE8m~bdLTvSMbPUM=^g|
z<FzG6aOyDKGAie~F{QKhF`DOP#F04UkaJoZE$`WNXd5_IJ6N8;)vX#S>lTjZP=|1W
z;2-u)TGM6n?<4;JYl_B7)dC*%6QR~lbcS!Ga;3PV6Y>0s!G8_<l<tQ3*mlKm7tr}v
zk|sJe6j{|G00^h6ky^4*AZlIJHZ3tTNtyscXb#Jq9A=ms9u~>gll{j2RR@fOkE=Ki
z+d?lH;<kgdPD+vHsq#5;F;?vEI!4FnYN{@1iw!<Xg}(K7NIBJQzw4l`MY-(B`>fjW
zhx+?p_gndj=8J{(C)%9Xyha`V=?bBnkTLFQcJqdipJk~w)rXdPtYmw&Bf##Q$&WxN
zx=b2vXjV3I1OBqF=+QhE!!MA-2DFsZjgCi3OR3uEiRNFY`#jX%BtPj<{{XrDYa4#;
zoR9d+(95us%fE@g+O0nnf2xD(X-D{MC;Vufyp@lwf5*800D7GtjT%(=S(Oh_T9&=5
zwr3;`lSDY59-_LtGxOGHscw2!h)4E4xBI`=t~{uUS(pGi*IL;eGJ3I*+^oE!u^-;Y
z6&#NvHax{eJ%Gvl>63enSTllE+>Y*XQzRsJ3Oy8Jfh0f4OmeyU^Ik>byJfWT6~ZXM
zCvdE){3k^nCTGU@#aR+pog>jk-iFzU{<W2<AM}~O-!xC)M?8OvBb=6D46iTo1I2Y$
z3H_y^ESVdM2Pf&)y6G=;^Q*sp=QQv?K}{CY8blLCVIzIxin|@GPy%|NQ%FR26NLxZ
zR$8LmTnCtC2SG??iJNlaQ)FX>1FKU>@?{MF06oJGl?Ozi3E_mwOp`uPa>V}t5&r-R
z+tBVzmkz3>W83oQy*T_xPhy9P7sxIL-jkfxHj8a4>afhjo^kjZ7LqrI@XtV#?bdo+
zw-Dj&C5@wz4^<>!8u6ATW@uS*Rfw+HJ&iDZO$|4T`+fic#szggCQa5+x1&;99*h1Y
zPH)Q!GEFzkU%G&?{VGoI2*l-&;3z7itpZ~L?r=u{bgPF1XPO5Qr<zZt28ULzBJkR{
zAM(V0waKaH2Aer>+sXNzcZ2N&Z}z4iGh_qoDi0INw-<V$jB+%i{{W$iud;bq&w9IL
z9txU0b3l^9NI;KIK5IiC9)G6eRX!kU*ZQ(q1h;6cL_TVObDo1I>C^G0W02ssA3JZU
zuSwPX$fxjygLGAmi1!2vpl6`lm2VPWOv57D)i(QQIsQVqZwZvtHD{AQDY=$G{{Vdx
zP9#rbhVi}U+q_wSBDUmNHX-&E+vr7`;SKOT7T!!!WgG{Elg#lgf`k76Eztg=x{ryc
z{wwQG`sEk?hG=CLzA-QSOmx=o<dOQ<TdAOH-wnHU4hQ}@PzDyGbXUhys5?-Y6p`t;
z*7m02HI+vnF~twzW^dz(PQc<w$@Q%p<r`JI=D3%@&e^}-BQA0B8sN)$o*_oT{{UVG
z>s5wc^CuGh($9;2^b3x}*C#gvrYp6aFxcvT9_ikF)7-M~4mx(POt@{Rv5LZ5Gqwuy
zNT(}tfmKJCZevv(6OwC!Tb<h>z9+fzH0Eaepb~!y@k!+KCU@#tiLT0j6F6n?BVWSz
z(b!(UoaI(Qjy-!<O%rBEkpzqw_iih>OWm=>DErqHgEWAZKz(^ND!^n?Fi1U0aZ`Oj
zz`13#m&@97oCQ5<fJ>CPHxWogooijm*tHJ7d2INS1Yy^K&0qfjgojVPgu{6WK;EMu
zQ?V<1qYqQkzRKQpq@H!Vo(Sg^oUBI-KRV`;vpVA|q2TA5T;OK2W~QE!o@8*ZtAASW
z?R2POMIlE|O7?6201BU>l_Z;yc@bh5T9Hm`jOZLGXaNbPnglS_4OC}MpZ(KA*iU25
ze(Y#!G6$IIIs;tn<0H90yZmZ|oYa(#v%#1)7c1Y){*}yMMfPn{c$j1dIi~gq{7iR{
zF}XPvG5}W_PeLmubP-!x&a$bEit;)fSF7u3Gk7aLVOMC7GY<L2YU1=}H5a`T&*9T!
zQ}^*!OnzPfIjl~Sdtw@7Q`0qRE(o|B@(x7|qJqrNsj)onJl9pA@p{AkiuW*2{hCP9
zUkPKx-|`M4+*cbC{M^@{QTxZ6E!dB|dsfDb#dO=9$Cl|<j%T?0X@=_E(e6Le)kko1
zT|Ep7T9ILR4mc_?R;-)^$Gs_hB~#qLxq1;!!ebT8&a`uP)_-TXY!8)>6_u=AEtZ!&
z-NN7j>sUR_Qat;@w%1pFA&97MAA1w^uVMp=?x3uVSW++yrAXk8YjGSGjPH`kTE;Li
zMsZwBkWC4WX%`#*WzBf*Dl%y5oc+ZU)P5MNTxsq}B$r{$TWeLd)hAvuTa!^G?-)|*
zbQ*Sp9rc))xdfW%^t%&dZRF=59xEp9&W$WyOoDS++OCo{84;O?*f_;s*fC$kXFaJ{
zeS+X2ISe?Z(Z)4MfaZ>dy-VIMKP|u1*D7#$??iSJJ3T^wBTTZm#zsP)a7HSuQ2m?D
zhqAH9(AQ=c;${{80vjnru{3STUOx)wJWiix(=Qb8HgR5y(K8rnD;`?%6q#LkDnSOj
z4NVtS@X?hwF$ukW$ge3HPjt=+syy>jXnNO!SIqFVG3TnV{c2g`Eo^bsPZUp~H1}k7
z{0e-45m_4ykL6;gq3KWEgVKO2_;cm)Eb*S?8tJvsiy7PfgIdR9ny18$0%zdWP;%m=
z*x1v(D&5%m4NDozYoEJ}_TOC9yp3zA$0D2+ers5W^<6#u$Wg~e8Lc~Z*(pQ!TajAf
zCCgC+wk=|eSI`)qQ}F|j6wChrT$A}$TxsRoFJsi#fnWHt^)*`*rDF1&^Y2{^ulk7o
zmCZ7vxh>qDO_YuP&ILiGL*-cldb!+bJ&WpJiALhYX9BtV=}g*;hq8RMx(t_2Ox*?}
zAx3I@k{aSD{mUApCP9B83$=Q+E`rLs#jqTm;;k9T-|;;9WH&bE`qpWf46m0UQH=NE
zu&%YMT}xG#=Kj^*10R<ZXuuf2>4Q#98A2}C8>s5G*BX7jg^jC0ZwV|*wprNsCazlR
zMJ}v<&l!pvb=p%sN&Cm~sZ-^WX3&P$Ivcyl3<5Y~WkO3G$0sJ8X?L?vlH9KC><)9%
zy$R^e9X&NHN2#r>Z~+*9FLQP7YDMt?ka;u4#1H`Q&22`aisfrVioG6vOAV?)ceRQ(
z?v=5nT$!v}DnLdcQC6ij8%1tm3NnteXDf4a_DOEWH;Ksu{hn(fz*l8B4oJWq1z`=p
zuBWRDrc=IlHxes0cGNJO#KOF+P5rd5-07dh&V$0^ZKU7z)B0B<tPJY+fA!VBooc=m
zdEfH!nfE<c>04e5T<Zk-A*zT}*@@!u{{R!l{{XIQSa?2^J;EOfEZPDk9AdhyKcKEU
z`<>K26ME50kg76C;=1R5En`#Fj^EOvk5Vxj*&L6Io5<fN82#RJ>T80tm5PN^k5gJ(
zMmW79dOn{Qo#B|%sWDd}KgbRbdR00TN$W3DI^rdgQ8yU_nutCe5!#yA(Dgk%#-pJ|
z;Ea?6<J5MjVvxr&GVnpH>pfEy4V+H5QBOBVa62e9$P-Hvz2rs-`;XMGFG}P!_(EO@
z2l0xEPgAl+Q8bm%W;--ykw`O1K*KcM)Wmdt4glTG-NFo2KN89$)@A<yWCg80!Opd;
z{SJQO9F5D?svN2GHKZNUl}OrGMcaF#z$0?qt7kzQ^@suN*{mfSNc3sd=A_xJ<0F~v
zK>q80F<41&fw{;PomXiyd@i(+BCKcbqe)UO^&c#1J&j>=u^|{SA2eh0q$mP<QE;&X
z5DhxoP`4?M?_#v*V<vXq9XY-~?iyylCjCIK_eLpy%d^V2@4iNLx&Hu3i~aHZ>tn?O
z_b?xGN<RY~PvXSo<g9&X{XqW!dYvDQ7gOP8R6T1_*R^)c<%!Vf>;=mEj8|D^etOLg
zTb=HgAh^>l;SVD1QV32n`1Y+hA$0=?7u31j-<@_Xk3Fl&vS9~8?pq&#r>ZIpDuMJA
z4#TsGjiyu~c_j3}pak<98;A%8UB<lA;tI$8wH&eYzA=(J)KvIOr2X`Bah^D-*QI&0
zJxH6n3Cc<JtnEYoJ|_PFe9<p~7)SV$IL%3z+_H0w*Ky%y7aACZu0rKcy>`>t9%dur
zWe11Losh0JoMM;dUe%KzMwwT(4OE`*o=ycIDut4Kvp9}1>(tcN@WuvqW1b1>-{2@2
zI`4MGILP40M+@vJ@!6^p&72>+&DH4$x3CpfkwV75bt?Xq#riY+xdZ`1-3{zP;@fS{
z6rGO-scBBXY>8EWh*dv~)-U%*pxRiy_K|fPWUkY(bHO;T54(wy>Rs4bH<dquuDKfE
z#I;6_<~ettm4U9$#0NSuH*(ATsjZJjOZSjxE#)+aE_n2*017axQg?<J_rRqpHk=9o
zW;pk$kmUsfKsu30jIbW`7LKa@JdY2U&ty^hise=&mSsc)mN*!u&RjG18}mD>GPY2~
z5WqWjkA69<=`4-rsi?#K{WZpZgdgct_B^Vp_mPM2yHFOo?AC;Fa}nhGb6OrCx0n7A
zc<ut7u~?Yc`<WE~07^$8=P-ZA%745+oqDFRwoe!MaWTTh$LCZG>>`y>HJhd&Z3Z%a
zi~j)Eu1DdNp=`Aoq#5#+h&19v_B`I<x7^CylFm*~*1c0pPb=W{!}Q;6CVw+Yl#Ua?
z=w$I_iiOGFoL2t;#7ta#K(<BXg^f@B^k`)kzA&~mZ8bmXfj{{5b{e2RY4GlN9$?A;
z0D#m1fv8)?d}(_Nf=fV>DIJQO*4DV{RkoktQ~A*T7H9teFA+}6<8L*sd(0|*4RJ@|
zXKemBnRf?lY{#G(t{Tt@b>(bzmv`%0;g+n=67)p=E61B}{^_nl<nw{l*Hhf&f$F{q
z(|*IGBRcJ888x#b`H~j=tD2+G@1(uv$e+C>xXr@joL3*(M@+s3b6*i}jlP}ahQX7J
zdv~uaW>M3c?WVUlr?Nvl!m=wK$AMmnr`+4irbFeUtfXc_PE`IC(2jW$`HP*afPHaN
z$24~M-6xi<!D2=#jC!2zw;U~~nIpJY7zZOKnq(mx=TdT41RB-|nK48v-9R-2`f@%<
z09O3BMoHMLV4mC&6Ef$znswHl=i033r_E9GfI5#_!gTaEr#(!?kB>vzo)eBwewD_~
z<Y{RV1Xx&{VgCTFcCu^Gq7U8<mF!l3yQk=5P1(q&Cy67^2Q@91h!San?&@oh*PQ61
zaUupC>G`10N@<`&O;XfG?LOdtuxT+%W6k6nQ*ZZ9DiGO}<G)(O;~O{|54BKZnwB)q
zi@}b5({Ubn+72rtT>k)&<=6dmq3jX(itx^1KBLmFtlPlbj2g|OrX!%2_p8tGSEp;u
zj|C6+6Z%t*%*wCD6VJ!Z-xVr?7p`kFp%O)ry@z3nRAO8bFlZQ-$(>2K?_FMkxx9A&
z0B05KVO_BA)&54$;A8wqz}s!~HOeW&R|Ma@c~`j&%MdF=NoF>Zf_fU&7(K^C*kf9{
zWzUy;5%oCy>!lp#{-jrQB%fNhVz$n!x=;_+qEACwYg0&CLipP#KaD);?T?kST<x7n
z85-TBu_Cjt%bvB97M~5`Zka<UIRMd4H<`3a;L`0aW-`MHs~+H0tXFa<X%y2;4=M3G
zoBse2PU>?EYymy7R{UFK8tIyK%p(c2ZU^UHHCOw@YrIbD;T@@64q2Rk%R`D~#v1l%
zAFr6-&XkF#qi02?Yu5IbD-$Wo`U>cvx)WQZ%a9mkX0d9=M&*^=w>^zfxYjN0ZVWQG
zQU+?@!i-|CHeu_Rklsv^LgADpL!kWHvJu!((A2l%2EyBqt#X^PX~$tRuGKPpGirSu
z4Ly$OBtD^km33i$CSzaVG2xKg-TP+-sd(Zd{vVQP#z-SSO7v>zu3@y3JceVEYpw95
z$hp$wwRR<=BxBJ}`8DM&&1!rnAE{gYlKyBgKt>>E(yAH?z0vC)0m;xpll(^&pz#$F
zLVEHlWwI1LCQ*j~RE|1UX2S;*g!4cU=s~~XFq0iNDL<8Udh+K@m)&bv>~qxk(dSd;
zLj&5XLUScaskx_&55})bGNe?D=KMtoYcS*Z9B1CSJJcz6o}CS80W;BjJ$d$>J|KgI
zToYQxc=96?4>PdOG?mn}DAkF##t(k<{8!RN-6nIh%RF@-nd1jexs5-SW86QN7aV$5
zhhO-z^)*`=z~Z_~PHp4=0Cu_LhaJlLoTQko!R6bLS`ny33OWH&%UhE>h;1H2zymeP
z#>KZe>c=L8H@U5*Mnn%M_l^ZexEsl&C*?pCRiI{N)xnxl2THl1{JO=UAM?;tGJ7*u
z#^3StZ~o!_b<0wFg+GN7DCt)%_+eSd+9uAQaMGcS{KYvL?ma5}Ane|}spIKg^zS-J
zn{+2&9rqOg4y#XFd1S<yCmfC|RHWCsa?01#wwC)8Z6|}A;<_ytci1KC_m5wtPb86y
z+GynUADT%3!7O<8tjs7On<dCSeQKb!)b4`S80j-hG^~G!=B!)vdhu#I$+Jt}Q>5_H
z{{W%LeO|cTXXMsy$?4{|^R9^e2;u(#my$E4?Oi8<AuXuR<Z!Y709v?{Q{iNuDveIG
z+<yYrK#hGWWw}uJP^AFD9<{rtACOlqW4eyVOQnCu)IaZv+t#klt3beK1e{iIc8g@y
zT-6ddzZvS!d1H;C25<*@<!lv>HmNnLcZJDO<!UFR_<PQ65afN(K>VwYyK<X@sRWu*
zS4dR!l#KV2aWz!77}Z4<&r9$%#_c!AW6JUewOGAFZ+7eNky%yND0H-_(bUH$o8i*g
z$?{H5;arh4tNousFq(6(G8gD;nDIC(6mkL4xvc4YM_jfU;Y(Xv_PF~@D=yq)cWUe;
z@V=I}QAV~3QPl}JJ;ycaMhT<Kt6EW=lO~VCx+`kZK-RIWF>=J7)I;Ho6l!u?&1o2o
z0c?*>)n3jfKHjHrUkYg}J;ZWZ#zIC}2=7WfKjGNnfX!m0$}(6UoYLfvVf%WoW>vlL
z=96o3(XFgZy;S;xRgQ_I{{YXNRJkH(t5j<DHS`@PPlicaD`?js1B#=lX&Pjg^E2AU
zSh2}9PAbM~r1{&qTl)`6Kl0@fuRYbRqfJ8LBT%Zc5CH8^(pwwBx!&V6n@Oj;%3)PL
z^B&dHXfVYdyBewAfO)JcKi(tKp$GMOHfngLB{vqtE(liX_*N>!2RlZ6>zcH^=67LL
zsGvM_;-Y=$rDC@_h+lf-w4e&E_a~6KU_0Wrin)}fV!K{0c@(iu8bvCkobiEMax`)}
zv1g!oX<W|o;NegM_|<J{OzM*UAIh@-0GDS9{xkF`+prfgpZ0}xULpasu>HdfQ4fKN
zukj*r?rS4i{{T`y-lC7jjcR->%ZIILTOO6$BbFxam~n9#_pY|o;l*Z$Eze5O0bM#0
za21ea>yw((XH@&l27PO-Y>o$`WM66x{;e7L#z5dw7nI|BB!0l>BABiSLBLS*XOPFG
zI3%sqg!KdbXaPmK0vWQ*1~O}l_>pG6WVX0f18WsLRL^4Jq>;f#)2&!T>t1ZnQYP+!
z(oyuR+y4Og_@DjrQ!jxS$M}*tTeJpg81ieR@TA^H?wFCias_tNdLzxR@iBFo)*V5D
z<c1roG4Ha$M(@g^O3cY@M(uLh+Y=E{EwrxQWXu$L9@L60K?ZQlymrr4t6oD(-0kjt
zX$Z82IDm~;Esg-CP!R8PxzA5Q^r-?&5<@@Dj*Hk-*6u~rY}^gPmr<AeK@z>9$a%r6
zJpto7kXbm|aw}H7^69_bv~{2GlMNmI&e`&%QbFzwc{Zqj;%hi=TX84x=DKflqY1dw
zWf)P`yA5~#K#_e{`Bg2CPJa|VPHU6ioH!LHa1PWH+JRVQJsai40ENeTbZfoT4o*b{
zt&WStY|E!ap!IMM<yrn6wliLtSOT#O{ynOm=M^{0hiul8=*h3eZl~pqNk3dw%{TX+
zDw-DiuVUcSXAO1B73gq&A5XIQ`ql_e)%#zTKAFXJo+7#b0E%~AvB3h;Wl{d=IUmZA
z=d;XU{{WAK_1r7fZ{lb1pNEv6m_SL#wNM*6l+SJ9uM$XjK6`Eamau#T{!fH8IkC%2
z<>%@tq9?JyYB2quP@WOdT#EG=Wf1%qU-sa2{MgcEqmB4-0r9nv^xgjepqkazkDu{z
zV~}M<!md5BNsO`aiw&*msyd5v{{W#^TdrK=R`A`B>nm>u{nh^f>(sr5Wol7B{46<G
zbGq0dA6_c{tr<&DPg5H6L--8W@ljsT{G-~or#r0k>IHGfW_Hy0qGVs3T1o!tt|vpo
zYU!S!8rKcf#}YrqvM-$y_f<8`>9(z@YfwO{2rbC=u9<9MNgmFqw$@7;`fU{u7_Lf>
zQ@(=tZZ*bwQ|#3FtlfaFc>GA(qxTu+pAv7HU14pv0+9^;4RJ-Wz!^QOx}L`j#)!#n
z&B?BZPu3dRz$(Nysr%&r07}_zWi5%$v3q+c^4vyx5`v()6Gt)J?BTsndfi7unZ0aD
z9lJ>u?&Galw?*5IhtSciS`8Xyg@Loc<JN=nAKq|1>t1UsnI(5&OhMFb83a_eaXFAg
z$R0NP*!0C)YEij5#z!-BvBh~TYB&sW?N+pVKep<S$BoKZ7SE-5a@_S((5Gh1S76`@
zvfgL+iLUCC_hP(`ic}>ibEHR>2Bm^i9IC*OL9Elz&B309qTPA+i<8>4@+yIBA>>j}
z15ZO$u2apokw3hoP{E#MIK#{jXEj<UK4Oo2RtF0oFv%mfD?E%6DRcyM9t^<0oO;Lo
zYL|*45o>Z_V{(S2&~f;To}#SUJH+9zO4Ku<7U9R9xbEN9y+_0&9t<PV6ZwiGvm&R(
z6VH)yy=uft6?m-9gjk6{gLif(G=r9*44MeS%x3nkm&1$lODC=|UY!?ecJ()!wJVkU
zKyw-Ay>ktaYm#3P<SmG9PZ_J4JC=szbol}OMQt67UgW628nbC2OSsv(D+65!!R~cB
zj@d1dkN2xL?9vaKCnwuA4`Wy`Tf4}c8`((rtFUXDgd`Cr;5}=ZwvMD}KZsWP(e18c
z!??nc>rIUUMyfc#t*CWos8dbFbc{kODTzpAU@}6I1vjwIHSq%pbV+Vui_87~#~taK
zh0&YBx@1!Q=O^+USBUj%)YYE9;RUv{@Vsjl01405uqD|fjz@Pt&WErsje^@}X4Ch$
z#dp?_1+tAj6jaF77El#TS0Ame?NUbS00OP-jG^&J$!8<!TK*Zs>LiZKQq2YNjK98t
zJAiAKW(&r9Q(FT&9coW2!&b}c$xeyG%OD+q{#Dh4)=b8~z+_zvT*?0RGX7M#(*DgR
z4tOB<73kLf8HBb+kZRViW2j351M;3d$7;XfOV1-!5i?<C8$k6vtIFzYEn*%swDPqS
zxKME_@_S@gIo~ur!Y|!Ns(1~L*-DSP5E|LK1umc5=B4a7d~D8@c1%Z3awsj>)@H*A
zqz#epKnbDzzAJxtKdp6o(A}m<?ki>RoYMXw=U0n{KJ`{O!t;)5S+TKT!xen|!l0s!
z^_ZRsA|o5dGhAM+YDDSs;ef%bi6#9lb!UNXEJ0L%y#AHh=w2EYL(jN&mSfeJ=DjQ}
z1uuSE66Ir}R0uEyI~D1fo@wI<v*PLGBZY+r>0Gt@{NVxmituay02W@RtKljhE2Xpl
z0FQM40CursiAEvzon?<DWBm<lsL802Xop*q%drtJSqNNJCexY}Tn)n%52<-^1+tET
zS0=J$VwM}fj%FsU37MS@X=q3OCpMq;&{H#e8s0bm0FRq*_Yd=~V0EPT3V#YFK~`e<
zYbhI|UooRaWMZwF+|#a>crv61{w1!QEbQD(dKY}cjJw+;9!4`(t|yXs*X;B84UCXE
z73)I$jVG^jl4&U`qK<V1=leq-E=d4!U0;S|m}&C)w+3Ul_V%nGt0#HN=$g`|<7mX^
zuE{dHZo?MN0IWSu<i{4!KrVLl?cR#Bv}DoO4H1zE#Kx~%{qifpnMmsVFh9^FK9W}%
zt&nw#tc?Ej{{TAZhro_U{6<n`MqXCBUj&sGcK~twvHa^rxivl#r->%ttxM0{z^uPD
zRC3s*Bvdv)CP(318r{;)P**i&q1i#56|1sgl;Dm*sc)|4y1|vp5z{@ZlI>`2;;(a!
z@kEDYQG?ZvYF#EJiqTz%QOT~2siPI~v!~LvyIV~@NL;q^Nf;H*-jg;;jPu1OB+@1^
zYEAbuWwnr}n4_HZtmznrAn-?8)*6aOd*IuTA)qC&9dK)<*J8s*FwZ|Jtfj12_Nt$A
zqVU?Rx+=!}h6EqZvTn49otVUS9Q5s4&I(lGa#O15)}L#bw{~;e=XxA^j+M=LfmGPJ
zVVqYiQ;e+7Pnk+LL|zI|8t)(M{{ULGuK2G}wDCM}TIx)eXd+fs#?ipt{cF-qsGd~f
zrBU3mapEgGR@JVgxEEsTVHzZIfWdK+ew5j@%X^E6Z=kslq_!@N8%PT)jDerS6x${-
zdxnwXNbWQ%q`bT<c??XE7H__%`ig~1YBoA+HmwXP99Wxlqj2Q*sBZmDlzCfr>Sb$s
z<6Y`b_H#(K>H<fT$pB+DoOK(|`S^d%vTfYzj1r1W>ij3DIkTJ0xh6o_!ESMi=Jg#`
z>Rm>865;l<k^nWMe(EuV7qpg#Md1&&+$`-X$m5I>E1lJJ`@^YRx7nn}nB496sd83V
z9Ao>(VqG5ELX$@5(Tt4of@`JFqJ}L>HAoLb+PS3fQ1xSFhik~U;z*0!MUW5$HGwIH
zJ_{0kD;l?RX4q*jWQycXD{YP=`0-MGiLv{}a(xh1YRKKq3~f1|?ptTw%}XAYDZyq`
z9)MNzX{nT|!E8#}Eu%K+5IL(5+FM&LB860aD07omIONlJvA^NQ2haRJtz+tP`Iqw(
z{p?nc_-N&a@uX4=6~Xj1(d&)7LqqCN{*<ZwG77&OPDSe@SpNW2KDAVSG+jsTvob2t
zw(VV#*yf3|WZPv%^flPpef6x+<a#cba}-*An0CPlfIV_*`r=jvAZF*2Cm8$<b}f<T
zy=2LJnOg<q+m5QEgHKo^QV!ga(4J@lP6xFx%DsVaprCgR%zEK#p7ES%d7#RPx!FEo
zK*%3jO=vXrJhCJCRiv*rq3XwDcR~LEmcKz+xBmd~_c;FmXw*yKMp6DOj0r8v$l*qL
ztqVz3+VTY806-_}T~K->&8)rVYIuU(1aVF=fN@zn)F|UQs-ET%*hXQGo2RuA#pis*
zcA)4zY6>JSsq-@o57M6;vrYhscBl8x^rQuZiDUU=%Q@=CqzWVC^2VfM=Nauy4hsx&
z0qsZ)ZDEp4B)NPls9}!vfuX|EYO5;^tAl`h*6-p%gZEi$#SsPA3*>-nk=E_ZcX8ai
z1NMl+eLH_j)-LUxSbZdlk^S$+yKQ1U`WpWLvK0Q6qCGijD0-a2;B!W5PVst*N(MLZ
zrB0+!BtPp@>A)RA0o!<{H}JD-#S&r|WA$v*`b;aQ=>};&Qp0;>bgJrbKj20^*_orr
zXV$#d>f4X$SD?K%o<6rwGv~%5W9T{0@)VZDV<Ycd$m7kyvGKp!0RI42cv*i2uCHHN
z8%*&m&V_f+2_EAHndF{qG9Qi(Yt<rev)*`5RF^Cs5es{<T4H98k8Z`UgPbSHCEdW>
zz46X}kg9$gj9d5uA_fi2#Gk;^M9*W-wXgXNU4Pd@UWKT$L-1PA4l>FQ(7RHROJkY%
zYmfL%tiRxY`U$Oh)J^f6W0Cf?{{YZvV?%x_O1EArwzpIBM;KG>lbY-Gv5yvbQ$0X&
z{{W7Php@>tD^(sbisXM9s-M8qb?FD$q67Gs6ieVTj<r!r(EOse01sw8mbl^6?W@qi
z*5As~r*YIRavBqV#MwXHHMSpP97#{c6YbQ9&qNh-;kKUB-p_vPl|i*h_3vBA%AUuf
zPTQmkIxPTkTyuA4rA=BWiMY>dlHImNBc2DXWesL3uiR&dc!Ksz-EQty<g|nX>^Q9b
zxW-L)#aSbcTZ_+XQwvTuBVhI(wV@2?bX$KW(E}mgdgiocwMj9Nz^=%zbl{Wshlqq=
z6;XrMvF|T63rnJlG6>Hrn3t#JSys_q8q!Km%yMf}BwI`|xjjxuKc!0^vl$qg+1MVS
z5J&4+s=dvf9>tio3r389%^rFNL&xPzd)Tc>kX!{&zbH;gt>aa4*u~P9G0jOk%{$7#
z<&JBw@N63`=DI7ni3$&(t|rdavCUeHL<s6?etl}CE1zA1rxk{Z?t~=wr30bG9QHD+
zgoD)`28*jLssv`mYrlwL13QIWYAYUe@QpnUEjsmX;|vZ#tCg~MS2!v~##UobdWpN!
zGiRCVYvpQJ5n~u?$!GH9&2v4@DQ;atvc?#7?^Nz$DJLBb1e#tIa=O|m=*n~XQR5H!
z4gUb!CH}N@DnAj82n>F1f~_RTibwF&GomE+ptQ-JGQ;`Ts`!*+!VW*#Pv$AdW@l64
ziRVhedYYCwmA*weW6)MdN;GDL?NK3>b~pf#Y5;I^Njs~Q)tP9>7684;uBSr*G}7G(
zHRw`$KSNrt_Xbw9+p}Ern&+11lO2d8`I4<^d`CLx8Bgn5$72`q4pwZ`$lGL7?^~HP
zc2bp>OY=H_po)gi<`o4*ylkSlQZNDatoxllTV~{CeQOD$rX+AK>xjquRkd4T{uXp0
zDG08B8bwpo?-IhoYnccE6rZg&?5;F$8YIJ8u+(g%#$ETX<NpBar|KF)$*5`8F_O{4
z5#5g5*N1yeqixaZ_8M~7_;n*zIoK<gnl`n&kvDGnvOTj;?1_I7lG@78dklESSXZEZ
zphpN^_^6YysvfeF4&%oa#NXcT)Mp2wZnbZrlsfv9UDT32Yj?s(Jn5tzxTN+pMGqNI
z{{Ry;8Q>tUWn~h~s2i4G1vRie&qcdfX49j$n;a-0A8Mgsb8#=yy6|6VnAiA9H-GW<
zFZ<QY>$<S8)vZyE5RaenuKMoOjMBO>{9j-uwoB;8DZ~4Y)ydk-*7s3Nr)fq9HRe6X
z2Y8)-iQrrP32vjyY{RZS4RFiN65LewJwL%LP{yO9N6mCvf{}Jxu>!OEC9mQcm%l8i
z-`*7gZYoV!!ud3e_@E0KEI{if<2;1hT}G+So_*_8@SM;1hXsEQ+#bN<sNe!>P0RM9
zrZ_dG9aM686$K(0*`Mx?^vHZKZ*<Rbah7Qga8F9@!O~SH?`%w+joU3Y3p5KF0gr0B
zqP?pmtc*uumYM_4JZFX1<T(ETmTFJsTrBVAOw11%;=CIF0L7Q7uR*G%jt3^X80C&T
zB2<zQdE*s}97r^DB_y4?$lM=l)re)qX7@CSM*Y6gt6Rg=C(dg}8RD23`kb3#C_1^?
zD>GF2B#t#dDy3Q`Ms9h<Y3N*QR)hYU3S>Qve;fY*$KAjChxyksI?|TmPvJ!9DnNMx
zvS#R#V}PQivz4Nh6*#HYYK;-SZG4lt4a9_T?rK=0Si!?%cU+#;?NO`G9%Mw|n=Pa<
zHdl8Aaly@Wx*0*HKHPO6{#A39h7~EL8=SrTiyFwxKvm~Gaa?`v(Yn1d^S3`Ju1rkr
z6%(rVh}i!C*HyS56OXPd%a9|p@Y4SP#J+F$>xI@2{7!Da>#Kh{-k-pZS$`3gH*iIC
zehL2okISF>=0BZiw<3QFGqr_zwfm%ASjs>Zr(xl^okLuW<o^H<VO;g<axD&;5awE%
zXTq9ftB)sP{_>t`IQ&0xrpcHj^Wi;@-e2*qMOn#OM_02+-5M}%UKkNX=Kz|_(BgbW
zDSP!y@C8jB;1S0NJ*%<C@JBM0rp-uW3KX&YDbf@#*bi#vNZQP=d27+2n{(qOYoEI;
z&ApdtJ$UO~RL*&wet=5aK*NFnAlG?3e{RzEC?xI3`qnAk$rV#hSsES_ENxvF@Qr~`
zYLmU4=VFbXD@Qh{Mn@{9rHPbA$8inH;4shkTCp|%01d}|XClH3Zo{z1>qJ~?S(kZI
zb~*n54rte!WVd2?&fYM?@vbvm$<;2u>!JR2)oUbVnv|O|!6Kwtan^>eQg$X;YEb-E
zO&t+p@1Qn%0%~k%nSRGkzPn-MM#>LX=qshuygP4ci6;H*zWoJbQl+#;kw;yis9!WF
z0CCc???C%pzwesopYYKT{w#{dHS`sCMnxKepJQ3UzGQ1r`;3!aP}fQb&K9zaN!Mun
zs;WPIg%<sq6Jm~lQNXA)xQuCqfTn~-&rz4E1AoJdh$WNz&5GymoBcL__s{dKrSKz~
zpYdc?EnPmWyYvM1RX?Q^){J$&DEgdxn#tAA$iG^2ek4^-gw2k$ZTnYbj%e;{TAUTG
z!p*W!{VOz7k4Vz(k#wthql^}0$@=ED!5AFZTG<>ANijEi4AWSV%P>()Lkp?kD@a8<
zY^ONdgSQ!=Vu%;`vfSsBO}c?5w3Z^H4upP`0pgRuZLej7j1+P)QYRhj&Zw-}>Qi@$
zHuMAYP4pF=sGsz?zuz^Kz63)4EQl?1{{Xbw&oY8p4{Fbu0^S08j<wYlXrs@rz4mn%
z3NN%>+kRr18SPxCoF-JRNa<HInj_&pdvB3|QPE8-_j0k@Zy3P-Bhs^F1QyB<nL9L&
zIod(nT0%kJBzgfz2YJqZ>TMqM(&Tc?1xOLS%xB!xZ6xo7jW#d$j}#2ui^)1JB$Lb`
z4o}jtG#OswR7qRqKn}rMz0nJ1ORxOb!*@kr&a-u!DQ>(gZKxBIZ83>}?5C&otB1tR
z=*rSZIT0X?S9N%I_;j)7_d>Nttv%!Fa)8sbQhERmDWspmf`l|7!-@onb60d|eBDkV
z{nNYhrnem3pv`?|AGK;$aE^1F{V4Gb$b!i&P~gAv6;G+d4|&sxo8jUA023g66#oD!
ztKvI`v+*o9DoFCx8;`9|K*X!nw~@~3GD|)Dm*4>lTo9+zKR5C<=>904D~||YT}AgN
z@{}K3Q#@w1Je=n>=*e*RJ{<6@r}%HyKgy^~v+*W5X0j9RK3q3UnEHW=*_q&v!&*(X
zx`3WSxg)0HIsIv(C-9zSuOUrpfA*U-+<3z0?0*k5ixufO2lD6o(q%_uH{o5Vo5bdB
z@I+XCJbzld;U@mo_?vS(^+tF6jwv#V{waxXekvp^a*s4g{IgxoyKK_<-^0p>h!R#|
z?edD20ehG6C>Z;UDxSTMve}<enkVoX!}z6pQ}T-4w#!+52Dst7J8JBXXT=lwI%BW*
z&f4a@KB4c+{{VGOZHM?yd`{+1j9i^$t~2s2dIyAb@um2B5X*;Qy}ik;r=bv9Bx=W!
zcOIssV~=X#XR46|t;brat6hDULX_^qIvi0)S(xG-pm?mZ?sg<{Pg7Cgk;$&qaf&&}
zr>SKjl;DcAOFC^HW}ehD9Ff7zY2%>HcSl&G&yuu;1SIV_s(P)Tn{Va>%O_8zMM&N@
zG^Fh#19v=eQZC`zyz|l?SOL<bV~}tKG$x206z2m5yG<7DPMZvt;ASDRL>>L=WJ$<d
zAKeO^u<gxPQ|z#R?(p<LrzaIC85J@#s3;f|jB!$oQPFr__aD1s--_$O$OgLUy$=r^
z4qBs4Mn!6HGtT^150S1D-auNp+mBl2mggM4AuX=Fv$c(1Y)>9VB4k?d>WH=2Pq81R
zSn;F^w{T7TPNW~gub`9nj%Or$3VdI@a9f2P1!Q!iLf%zb;{*G{`PZrVv9|a|U!s38
zPCGL?pA=6tP+O~2%LMZ7R!n3sHHp->G<97->Nn=#uFN1%K9yY#MJUIXDHPq=IsX7v
z8xw#Vn(MTgROvZWk&|A9C+<IyRlk5^>SuOq5r-zRmgkoZitIj9Qfa%5#t%Y=`BtlA
zH}MuD<dN@EI^cs_nLW<B$a%T~u7Bm9HAc$l0y1hYR85;{!S9OEZIi|n^HC#snM-Np
z-;vl1in&`+R*BHI5|GxT8f%^LDH?^X#ey~prb+ke1!IVpEjC;g*}`b?Txrj6fwo*0
z2&#MLy-UK{RlHW_YlZVBZVNE4Dtnqh>n@=$Zjh(SE1tg4W|rlhTbiZt34alev(cu8
z@);it6|1M3Fu-E7k+LR}3vA!+*Pcxj?JJ)xk3DhLt?pw_;rWm>Q*F<d+<R9|;Ru-^
znVb+w?^7*}2kQPGhf}zDT0#jJ;MY2QBPFJx1gjxrR|JDm&tzS$J5<zWx$#6+kh${V
zOv#VsTN-Ko>-{U%!N$~Y+~&kSg?&H;yKjDL%zS5Y=4%X~{{X82f30nZf4W3d*wxT(
zlUC6T6LRRjTMy)Gjnpk#%JN%w9HGT|@1R3TG<Dh(i>dg1t)yZ~gByE)kz72HG#3vO
zDFs^?G(Ck6iKEfHD*jaP4&^|uw^ZTcKGmPuFYTiy{C@2ic5Z%^NzcntY*sP>^{0mf
zP$Nf38*W2-n&~e_>mSCpUkS|r0Eltglm{=4dx2RVSu86va`Y>X7m_O6Hty~+2`9Bn
zb}lOKLt@^*3^aqEO1u;(z!(C(+$2<Jr`*n}wCpW73hFWziU9L(9?9iu^EPmNwFC05
z99M-~__Fmh=!vaDnbbD`^{dxc7`Jpr2Q7ugVp7=CvyaKOjR#>@VH_5$x|$<0&-eP3
z#m8WFiqstm_NMm)W!!%IQi?vv3lUjboSW|04ugu-6C#R?y=rKG@*K4P0Iq_99>%wg
zzvJ%P{lon0oKcfgEyqvcMBFOU$MZ0(&4S_nbsp5J_`#&gOtK{;ZU+XhTk4DvQdQmo
z;P<PeRnmS&o0^c_!7Tp(ESMSoD(f@{jdba-z>UYJ(!B>(n^#Q6vV5lpu3faMRD7xl
z;<@EHm43D7%A=_Hi=K?KpX=&nIFKKD%0Q0q!u;4t?*9PNxGi*T)~^2m>#Kh{)}O+T
zX?$#?r>XX?lfgXR-2VXAG5qUQxfA$Va99DYUED}HG99(E1^~r*mF3MQdNdWFi?JS+
zJ#(D{8Z|vdR=BeBV<I-gdKziiO!IAQNSdsxAi-0NR5r6>aLjlW>ZR@;b^Y^7U04H!
zIjD}J?m4VgI~@|M*oE)p-H}>aewjYJh+Fqip~mt${VJtsE19d!*z7Dc=xiWJqss*t
z%U4gM$-dd5Y~@erYmS_osdjB$e(%uI&=5fP5%k1>f$dd%Q5G2`&j8}OB_H;C98j10
zRr~({k<ahPd{vk=IibKOdiomTgq$7G=~RN6vp2M>yL+{8R0GCA70>HF2+{R9q<<*G
zBN9tSdYbg3K4>P7CzaEv(>VP{;jFNhYioi~bjZzfcYYPsEkSQ1&0qF8Ka~-!t)-#A
zCo;TD)VGjaq%x_DmH_S@E3mlmZMlFot<wh!pPXbhX=zGHMH<kjDcvHaw}bAa0M9zd
z>Hy7jwi-pHj6oxe6Wy3(nv6{<kd68sN{fc9y0NMb00Y=i#0v99YUoVAz30-hZ=?e5
zUVn_$&3^3|!}zi)`JVNAMSSWj^fjCpyp1Y+3|(?uUy=UEtkWR%r#0`erSKsPgVvf{
z4)r!PBWXS9jP*$fji275FH8=XKzwN~!?Q8{E0~&JHdb@hj%#WBF*)gca%5!jU2V$=
zd_K+F$sfv!T|&O`^*Fbzja#A$sy`7`J`#j|YeL_(bx6V{t-osPESvW+pJ7=c$n`xY
zWL*Z^a{>Fl=iJqsdzFzDVpfxA#_rsE*KB?idGh#~KG|QZYy@`dDZ&ICx6FH!-hre^
zfCmDK1|dGU6&p^Ve4xM3P$SAbS!&jvAe6H068zn3IeEo!S6=bxVcp=3tp|pi=qoQ$
zfArbE-ZhNA1Pl1A)$qJ;C8gfcoF1p8W9s&VT(fRc`Ti5db*1hjhO_sXr=hz^V$i-x
zQbGC}#<{jw=5e>~w*ZQA_)*HYWL|7*!HsiI7d<wjasJh0P#I@#hB5fkZn<yyc#TKW
zkPgFb`L2e2g)xi?_vB|j=8yth_JZ7;cf~<FUP<?uMk%42ejbQPVI(X|w)33((=_yg
z%ThjbgheGrdJ5a3WOpIfWm~J_(6D5{<W#y$sV$zNrA#s|q5a<(!w#6LgZImhmN-Ri
zj8yPJ=z7;~(r@sfZ&<&rY>#42>OQA0+nR7}RGx|q(%|(xP*NJsy^l1{Fa{Xsngm(K
zR<!&)t^KH{90WP0w+gTKa`ZRjfsUsw!;S)lT>iDp>o<RD)Ew|t4$)NhIdIFGrTZes
zhk3s~%4;{p<2tGT0Ioy&)bt9<_5MaIs@GapB29^80dhZynTW~!F<q?Xr1(sXK{%82
z%`|xR6{OELB<`;1u1&XrGzq}wJwcd$8LBg8_l~6Vyc;CA#_RMyT&bq%lPdU6N{O3u
zw)}lEX`&^u#`v6LS!@sBTI#+cvag3U2)M|IHa@B;J<5*AZ^DEz-)a|F`BL2E{<S~D
z7zExe)NcDAVgCSq6aEyQ<0$FT9ZTZ7i4JyMGY|L);=9I?rTCt(w<`^SW%`mo!ljID
zsQX8VBy64wb@r_NX}3dYx#*+$Q9p#rm+=uDqUROAWPj22`d1u2Bf77`ah@}oL{Y`v
zpFfpj_;7=BIrKqY6h0H15xJdlY}&7l5*!kXw6Pt4uT7H)6mYK?9Mx5R=~NZuho=~&
zV_a<XN_R2^Q<IA3yjylok2qN|u>z?b`K!Y(d9#}o{nU>%3zMGy^lcr^E3D-Wz;ver
zo@#*J&>S7{o~MypA;2YYbL(By-QbQ2iqm42FOsToy-4PfL6BppBA(=&&D;%bc=46U
z>M5bUtIe9Uk4_HpxCT?kXvRkvsVt62@<_)%qOEBfjGATYx#l)FaqUTom95nIAmfUV
za%${-ZVpT8@wZ~kcEzz%i&8t$YnWEX!Y;y#X6c%1TCwj-k(kFDPUO}K_Lm{^#%r#(
z)sVzDFzH)EMsZ^+LTxlFYSH?NjkSpVMQe0<&PeAxN{N2Sa);(cDn)VA<OA$$8SZk=
z;wiDKRyT$zHaYoE=}<PVjh#KVYSzrX<X`7no;OUJR`lJ^<yX+d?Hq$CE$li{gSN9e
zQ9Y}Fq{aUD{&nj<GHsp_n;o`)F->g9pT!f+?NV76Aq`!R%#DC<GguvHk=Ss1nsP9v
zGStwSw8?go8F>}4;d`SllOmtsIIf6F+IbaO{{XX)U$$hd$6OOcw>-#fS7Cx_)@`=s
zeex?<>}4&<$IB&3PH|f;jNa!{p{q%x70-48vIG)B<abkwn-q;ZD}%spwY6<=oM2{?
zS{lTX;eULuj1sjSZx4j(j=@N3hCy>WTU)%+94ZWxTn*-;)9IRqqaIZG$+z%fT$s0_
zMv!ESW|#V{z1ZXTA-@{*?KNPuv56&Mt&H<tdo+&boHusY5{S}OEC)bo@>?--!KvQR
zC;nanTECbMc&mm#lLxIv?A7RkVf&`_#bte!e(;Z<rChf%r|^mPG;SIoD#fHC7}x={
z=B7;58T0wb%;C-}HMJ>BrAf%mWbDl%W!TvmuI{d+<>q|n+O_NsE^+EB=xE25iboAs
zcF1*3E=l$k<US#v&(*FZ>|gm-SeNb~r?7Z)Q2P$1n~(10`CIx{$BZ=a(WU%jk@CJh
zJ68`tdQCE)gYNvh$t{_<lx|V)RXk5=_HAgggSKooPp^7D$Za#L(HqK|7yj-mwAA^w
zjj_(^&1}kl5;86anQrUtS*)a{alxrIWd|flPV)e&B-+zo7ST!6?rP|6WG$yFV;RM0
z^oh>Dh;w)7+k~BmU!`QEGBo8CN$0g?bz`rU+}yI#8-{)bYQ=9HTq^VHT(XxtvODQg
z<X2>v%`PkJX~&ic9#q}r*kYV2$QFtK^3RCUf5iC4Ke(j+Rmj(eTlkXoH0U`7NvjsI
z`Iiw!dTz<A7}3zLYJJ3l3ny}Mg*?`1)TyM#Qp_>yAuMY}xQ{*0N;-nk85Y@li;Il#
z4%Ib_Io&ItQfpEk!((0vtNIuGhb=$rprzE8v9aTC`24s30C4{RI_HpaN_&oy_fWEQ
z1J<O7U}w^x*oJ1^B_g+K0Tv^EF;;EVF#@xZtW7u+4%OLcBW{q#*EJCMg<s)3`&pGk
z70)ErL|}&owWht&XYm%-hs{YD+yVEgkkQ`wOy^F2_v?%CJX>{(h8-G1R|ldw-^U{y
z8tl9QZRSgHA?d)auVN*tH7{^AJCFCxIwQ?_bKRbuKZO?}YP6>!u{9LbXEe|pUyXF~
zvPXQY!+fuS)6%&?#d?%o;Ey`0wAkgTLBOhFL+zkztC-+ilI@E0?K;6NZ6k<aDhyXO
zcd|Mtt=VMkPkOVa-U#mk_I@jwX(Vkmcr&8Y_`8vS;go0VQFxs<SeY~T8#UV{`<9O;
z5BKqBA|$2XURmuL`Vq;O@Tgx=p5E#}8*=VE8tGGUf@tJ$Q>i4)3vEsdfwl0B=+x=-
z`+5FsZDZV4>Bgi^cg;pdCZ(W8(VsTiW>9hnUTTCfKiwm0j)2!r7P_I$&+<K7Em(5N
zmIQe3OfD4r)Z#I!tv4H3bE(XT-fD%eGIomdY-WvJ228Zv`c^IUpXt*K{{R`SW&AW@
z58}wIWygBZ&}Gh~KixGDUh*`n@G-t1pKhc_`y#Ub&^>9-;$chRLj9n6(@ntjs4Qup
z-k4_hrXhxG(>&jL0PA#cBGAdt-&SwZxeqqBn)gx9Mcz^mHzK;2T(^Zu{{W_Did{!r
zW$JQev(~b8LzY45RZHS3r@~T<b6Ph2tD;9dcQq~hS5c=nZPV#lqNIAJiE9kH9i)z-
zn`S{hxvd*q0`Y~~6#jMCz7v6~NV7;t9Zo5XjLf;hg2x@G{p<Q5Kb-;(B~LGw`qZ08
zr2ugM01<2zpn~mG?gWBBt|lztS0*7HxM!t}%?s~K^c9J!wlxWce~8yV*hMemvq!_X
z5Q|CCq{EZasp<oC%rnkwtL!6&x5Qb{XEEHgE_iXrxu$E8JXgilxN)3gr5|AC`$lLA
z<L6BKgHAEqWBpklg=A2nE%AXE{HPE_c~wzKfGw<K^&f>wutc2g$)p3A;eJt%dV#{9
z^^$%9ftsElxQsl}Mlg6aLsEg4QJ1(UHLLhh3*sXR6m!Ko!tOgQR^IAPQa8&FQA!TX
zwqy9Z)*F3(dF_<_p6B<`5->2_;<lkXydb~z$NJY_Q`3yE2tPBIA2$^sPzTI#Do<4i
zX6+j1-hms+erEgy1tDdLPyxb>R2dDx`@X)k36~=|jC)q^h8K0r-l0Z6I%{%k{{RV_
zdeq{|&QqK)U8A0+xNPL^W0HE)k(LE}jz-Ui05_5!M6B->^Q)2n08H2V(&#Fm?;j$C
zq5<K}YbP8<{{W#^SQtt0b&sLP>L^E(h*qf2F+7^>UDkhxx@=>tSBH1Db5>x^qsD<p
zg5E50C5it4@zYqEb1~DjFnVm4`I=~%Y-DSY8N5pjc;Zcq`qtmYC1@|S%Utmtn(_FN
z{V6@i9g(5n9kyNy*6$=6ccUsd`~;5x@jwp0=Tm$qf>^AzTlQR|C<@2a0#EZ4n9QGt
z!d*+pvT6WurR0CcyR9MC^T%4X>ErBit-1H_{VG_)PGmaFt`7%!rfS+=`C<{tjj9F@
zwK=QA=cg-8GQ2i(LU~+|QCikx2*8YUTv3Zs)bE^>p`qvAH<^F5mJacZ{&kt*$3JMe
zAMA?hqO6W;uI9&uY%{E0Udp&8LKKnMS5v#pl)gC66*ZaBg<7^u4;iLN-A!^!W3=V`
zN2&QM9kgxx+1w!KrD6EF_GwYayXG!3J$SC1JZ%=v98<i~IIJGjo?lF!Yjk8dFQKOc
z>&-ANT3nlV5-Gt`U1gl{-bRyPsu7cc&2-V1y3xyCI;K~=)L2du+-y|H80cy%Ysn{&
z#?7)&4jg|f6qvbHh~|<zxi(Cj`Nz1b^65nGHuDGCGqCl?y=PK4)YhD>p-9Cnf9nvh
z)KQ%D&2t?o++0&iW1m_~p)KVJY@@zwMjLdNG;ldRtD<rDQ#&x&sZ&lR=&i^dDqkkE
zPRAtV<nA6pqsgmGjYgP#sWvbi6HPHT*o9lZX*{z|;vCWC2<wqtmZ2WWoq7z^vniR5
zK&Xpiq`2&C=ohMW3uxsy#H)gSwbyvcFxC-E==l8Ut&AqEGl!N{PW1tsCxKGUk#kOm
z&bo=DkRtCN&b@!ecQD#`R(C7E8+iOhHM1sv6i+aZS-H6RAcQEZaY1y$e6lWmD;AEc
zu|sZddsUdRx}>66Z(7k~Jyf<BdW=^?p+^d7l1T~qWO9C$q&^{6f4L~$pWm;cHBv?S
zjcX&%fm&6=;)`Spbc}j#6{L1EmgLf(-nKnWTDE~@BXjK)yiBBaULLi!NEv+aKsm-}
zk_{#)-^`qD&PdHZr9>9zLX7_atQo5-uW0teDf1s;Q0{9T#cfS>FeFHdD)2Mjo0{GW
z-8!SNnWU;2o3BA7&XGJZedidicF#j-v<9`hAgf8Z{&}xI6#ev#V2LDgWddDWsA%F~
z7<+fF(&paR#2Qk`Zzv8IC_()ziT5>DIsFjLZK^y{lCG+FHMMc#8;cvk86nY^I6bRI
zuM-sQP}<Y{L2Gsg&$0Wdtv9?<gK5dER*uH<S`3owX#u;`GHQ14f?bpkttm_16*&8-
z$-MDxoGL;c=dcxyo-4b#x)9qy8*$^Be7D@Iq;wdQ>?N*9Qb#q<LIje^%5s6eQ|m<-
z^VLYU?HiU46+MsqeJ{}0(mjtpTQUuKyp)n(NXf4iy?-)yfA29h!-;xD7f<0UuM9n_
zztV0KS<s9!;I`e4Kdo{t#ZzmXggf+&c>}Rg9GGMN&2&C4w~lx%Zj=$R!u>(-LfB8@
zOwiK!U_Ah?=I3m>Zt>FqiiOm<?u>#mq#B?ZN|eXlBvcx+Hu567ittAws6#&XD&Ke>
zYLJ=L*+L?)w#ng%Kb>#Q2{sC+)KjZ2SQ$}|HBzft#UyPTE<+3&NGu_@03r@M*O|=E
zZ9*up1}UDk(T0yTNUt$ec8g(3E9=h;S||Z%pbsVZhtG=S_K820a~0v%{v^FiM3Jw{
z^{ovmJ<~}qP@`(sJhm-vYp+_ZtxNunkSXRy3sW7-dlunMk)nDGQ?8`)t=+SYs!b)h
z`V?(q^0d9(3j)Tu;0qeimQ2HHVy$Rj@*B1P0Iq_P+?i}__|yJBE&l-AKhC-EFey)Q
z()dtq+-c0)KpwRy(14t|rNuBFgw=?gsRp5=SeN6wz^>ayoDBtE)k&$RA9z@YXdYpA
zJ6v5m9-Funi`0E<TUl7j_jfHXC6A>(-~C_XSxD9=b>Q>=020|fq^^5jvy$&nxkzJY
zedK_nCZ5R{uZ~51Kf)Jto%c>5UO2`{{{RZ_v<)KDNwqOUyA8=>$tJgwvN_!EqA;}r
zMyYY1@0xrq(!9y;&t9Lxi;)_+9b(N12+c;rif9boUfpi=>7j9)u1B%2F(VC+LtWTs
zqem7YYMAB4Ih+AqEWtCU@NI~B8ump550yWyc3WM#917yC@g0=aq9H;BRxDCLz~cus
z$lj*uXm*-R!cpdQ-hkqz*JAUusaHS4n)O$`O`aR~Ya`E=YLH7_U5|9!&>tpm!l>Ri
z`dq*78q!PmXy=ddWJDURTVLEjaE+1Zd8mw>UD2fCq^!+I^_W`=AlS#(rC65A+TnMw
z!9T<+(Um*FN%c6KWky#rFErbQ{{U6<ueVyrkojb<@eZ}aD7d>bvYkaWbVDSVPkN&@
z#?V0PL1@+Ryv)Ynlj&G`jLbC~j34=EYgsRG7{UJbMR7A4+wju4zAnIk7p+Veyo;*+
z#xIC0?x)mC#S6Oa9H`A@{{UxS{{YXdzxT}(i{^S0PH|9LfYWc!A7z*R`KePx)MV)v
zKBp9wIX;Ctuw6%=4{BfBU;FBsd^@RsaewcswMs^RSb(YF%|M@<VvnwBb?~g@eAh^)
z{oK_mFR2d`6|uPuo5Q7xF(51xZv0jt@TA}K_@DQ|{{R}+RPSRLR#Cp2lEvXwAL}l6
zeFiIFSu7eu+o2^92<NRh)8!QFbJ*b7uuWv&zUi^(D>_TwI_i8Sbo<q#Z`!tzgm*OU
z>!s4q`a0HYa;35Bejjxn9kVPz8$10?NH0}yFod{2F4Z0Dwkt!=pTt}qbM@MT^UXP7
zP;yCQ>S!4J0osZNBODI6rg@|IL1RFS?Q+UZM@^CzWQ|Ik=lpBRV_l?=LtNO6j;uE9
z+tC!}Q`-W$?Q4xrU-zq-d<fJ2Ub92PHji-y(*g2~a(Jp&l49kTIUv_ntrT!ocatAu
zFpBCYILL1Gt**vQDKg-NI6m~3!p?WPFvBT7IKim|VZQ5DMUN4Xd8G@C)UW}^N*j0g
zr~(-P$1nI*a~WpGLF+)xT?RFd_nrw{3Y%MqCE#!Gu1#t73HwG{Ha`llBKe;&7++eH
zQ0;EUd*2VisEo2(zm@b;$TfFQPcy?&r~NYKyOyZ!rvCtUA5)Vj2jvuP-^&wFdnl1t
zpi#w3yeG;zY7J2vCg4ka#*tSYM>GYAB0QDP(z<^O#$)qt0qzAj?q!I7x@A8Q>WzDH
z@azh)$WC+Cu?NzJxupnlQ%JREs7a=|mNiz-1RqMz)Nka}E{yW12N(bj)UHKJYoDK~
zRT#4yW|Tz)>=s5NAPn{VE4FAwzru^#VIYu0F+l6lu)~^Y@v%zNJiwfvO6+uho8k7W
zDdgMF8Bc80oUG~mVp?4W;d8>lZ20<yth+EH(=;RMvVU4=m$1w0R=Dv#l##Ao<sYHV
zN%3CcX7MypuUKFRe=>iqDQ*z>O%NMT_&D*&@;vfoeZgRVm0s}Io2C3GzPVl*!^oeA
z1Nu@ZgW&y}887dDhXOyxt#lp_aVDLr>O^KstA`Eva6g?-P;SUu8$T*mcXAvq20i$y
z`ktjEI)GGVbZ?kv(u$9jGb%|+Niyxd`ZmqVK_fg><&_^h2tKvX8K!hjn~BDFhex;6
zujQ4UmP~B~9xFRV@ZH6<^EynjBVckdT~UpZ%PH@3rM0#;&`Y&QEZ7(oW;NOcb5{6|
zM+&{c<*`+6<`F{7#{dl1EV~_O^HF*Iwc?~XArXmBpr?4wKxVf=)B}-S@K%M7E~L4t
zT;<n0y(w}r$*Tdo!1t$Q`eKj|QgPEY(WS%MJ*L$C(<ltT&bFtsGON(Bd2R&E!|s!j
z#wxNz5lLb)TOeneX_*mC1h0MO9wNr#K<`&pDS$FUtGI!?C!nFd2DK~dSJyG5Mj1je
z&OlL)wNgE1=|<V(ie)E|R|dJ;)aZ<KC7Mk>Yhg4whERErwN}FS<kT}&dunFeKbd(P
z8REMuTV+H5R<o9g5VtLPCmb3CEGsLPFOu|z7WAiZu+s|;*`^qu`?T&1K@*erXfZJU
z8h0kccD8=)b6V62E?gC=OwOi==jMO|P&7iFg&7MLW0@IF2w{U>m*S{A-w!}?K|4wO
zsiG=9Q#|?YLY}#%h}3tfWv6mmWm9zz{d8aFUbExU2Zd!GtiR5MvlRX)o>>*s#=BGH
z9@Q#c%*}-)lUW@mgaNP%9%}S8C84*d9u?0V^{$sd9$tb6tFiu7bp9eS5BElPqU1J3
z47}E}I3UW>ug#iZ$ZXZeVM}sphwt0){#1%jJoc?t#!qu^!*e&EBzZipYObyi_N2p}
zqxn-2QaD-M05e?%nWLE)G~5qL5-u*sLo`VOMv;sFDlFGTHaj8kiJB^kAi9u81F<|P
z$*y}*)Rn{9p~D+ylj+A5<>DRc8^c6%`hK(acm<4zWD-S@^~tSG0`~7sWo-PN)SAM;
zw>kl+MX72K$g%C0kVrj6XZV36`)U*?AwR8J_(*@mG2Uyqj^M44$i;Pj7QNIiEiNaT
zLghLs?NF83a&1MBYns-i*G4$cmZ+nrdcnBWu38{23o!?+E>xQc$?~G2T+4MRM93uI
z8tk-KE%cZ^(=i3{l4(7Pi9=cOfA)-vXKqR(96NhfwXc>fV-bDLynEM93ZL&T6(#PH
z8kW(v-6{Ml*?SzXVxNrV@-&HG)d>9S#qJO92=uQz6S^vw!ZA*3d&5`%0AbV?ISwP6
z<v-H6?`B+;(R0RHcig^>f7yTNYf=?DuZN=<UCe-=Y*vOh_fsFIov+1r7WkJ+P<jNc
z(X9FzE<iZVRA8ca&N<{%8nT_-LGy!HzD6$rUs`BKSEv@Z(XCky7(8S1U@NM)AL!Bh
z)Ytd}4g61;R0={X&OMPEfl=q2SEHA!)``!EcBaCL`h&+EMF132KzTRBKlvKv{{XH@
z{HvK!xq$Vr3cv9s>RK*Ns~kY|AaPncY<^vu`nM*zrLm&iy%*mZsuwISb%>+JL_=|;
zi)NI3xCf<E)?ai>2skpZQA9zLUY(W<gN!aKoQ*JQfuYQ~$?Pi5hyMT}%U}BFC`3zR
zy1&sabxBf7g84y#wAE(tmYu-eFXvS#rtWjbqe{b<!a8z;%)gyP-V)P`WVFBc)g=^D
zt0AMo`fhs5{{VedgTeZCI?MUfaX_lbKj9D4zr8Q#OaA}}mYzR~U;FBpA}iZSU-(5V
zKfN#g^<6%nq1wfxBx=vJA@B97gc4?TYRRLK{{VzPP2R^c_5Sred?BXfb1&yZiUn3g
z>)|~=4a~gz)tg@g+RC6tvg6dADjhV<J-m_9X#N_v(;;LI%n8mao%2~(Nh>zUhV$3G
z4k{?gPE(UP9aqJ1!68d1i<8mYp!_QT0EcyZo2&60@3=aW2Vqe;-qt!{M$whWc!3%{
zKuL@%sKr--^NQiA@iV@TuttO?u0_(YZ&PAwYYr)(Gw(<4oc{o_Ys$}4^{(tUQ;!kQ
zH&Ow;!0Jh^qm=Z`0^MyaA&BEDy!~rnwt82cDP6XDDOn;XYOO7DH>s-F=zK9hn4&)D
zao4SHT_>AuB7b=Z{VUg??@b;iA#&C{x=h6@#(Jv}N+hp0sqG`6@Z6a*eicUUTwcsS
z+BLM7?$LyQiy?vYS5n(H>ncbIVbc|+5{I^`qHNh+?XEZH%4Ps$k<CGGs@=sl)hwVM
zRQIl|bw|#kx&!2>tZ&=vaW>SHY;)hWPd$~b>TVz$^kuGiwNV~?qbgFmIsHdPxQlZ`
z`^EGfMRV5?IynvqBd9fupyJNDbrjLD;Zgy(2im66?ljcCy;<BL^G*wqTgJ&Nob$Vs
ztZ6mImHZcW$JpRhkj-x)A{UK_>$erH<Iv}7X=*a#Q?<2ylF9>mRb%2kIZijR`Yl78
zQE{ayE1^H(B-6jX{{Z*RH29BCJsJN1@0y=$627l82|P_|qveg;^s3F`=)wO0k4OIi
zzNq%drG}SQF!;hh`SgGL>{DCC%6{#s2i>XrUm^Wj_cgC?WVh0u=mp*ra0jJxU+|XS
z{{V6?`{t@sX^~EpWSzwxEWba&7v)IaC%=_S{E81;3WrgfDiAWua7nCrj_Db_O%%(i
zu+$$lYS~A;HyOaqYa<o8dK&0-rp!3>HJaS1Y<lO1E#rR=+815hlaP5GO<JBbhFIJ#
zHn6}R{8wytJo)ZkppZJyP|q|UyYqc0`P-@6K!_%|wztoi`;Vtuqb9h;w!?05{pK|6
zChSzZyPoRSXc>X?7X!8{&SGERI-YRV&xl^>(S?=d&HWjEX+Z7;a~k@v)nxf=m3#=<
z{{R)6UK3sVP4FrX&zmQ;bJEGac^L3>U6c4Cd9(f_20OrBgmfahVuJTY1Dx#R>MJ<)
zL`dfiyohqLXX{GBRL2?hBC;rOoP*BcQmd9cNIryAupCwK_tu-SyOBT#lb)m+cb1?I
zK%ixt0<qpi#j*oc-F;IL$WP32D@gn-hw&o?nMX`iT+N)EAnWf*7Is!~L#TW{xwDEh
zB#)DY>yUCQoY8j;9piC^P0Dj!vDoRP{mUGrgefMHJii?(S?(f6kShq2mhOEi$^!e7
z-jS?FpOr_csQu_D5rC68uD8Rv{?3)XSpNVjsP18j{t?Yi#z~!<sbXp|=z3J0X=o&)
z_jR7f4eLnGp$y!{<Gw4l+HL+2vYxWZpUl-G#m0O{^IrAX=%f#aI^gbRiyx&~!p^V6
zw(NW}cQJ55wzprz1IO!F+QW+%@a>{^w^<MT3YsPCZ)lb+cjGNKRq1T6A^frZYmV{c
z-)ix#`9ruF0<Ykpe>zWahr)FJIhk#|6`|>`l$fHsFVJAs{{Rys=)Vi>tX1;z1yk-1
zG>a~LI<cEd@P*aHnO3-B>ImduVz(`0x6-r-<3-u#lI45k^`@4mJxO0gQnkA?LoLR@
zBN7m#^*9`UxvpPP44S8#pT0oAtSZYxS_;xPEjb8r(Q8KB?PufoYnDF|-9LezZQ`9)
z^21htv&zZ}4myh7@Ft~os@cuwpamof7!JHwr4!7QS1r!ADPt2d6O3`vtQeq>1Au+S
zbGI2Yswz-PkjZ2Z+-!Yoj`6SCq0`JZ$X%p3J9>dw`zOq-bSmqYQ<3nsl#8l6OstO*
zbR*WPc&0%nu=`KlP^ar%i(`j*n7HSyG=N*4)Qtcr10tR=^rQu=3rQ~~5g`LSmd8O`
zI;=7za77$}*kErRI#%@7jKZtc8WAyFFw1)CSSiLk`d2A;dn7U;Y&&&1^z^B2!;Efi
zY5J6A6%ys4b>lscrAp5lMJu(&PB<fuJ-@=6dI=_(kn)*Z{g5y?qzF0P*14J1wjZ!C
z2ce|i4;=`p8Hy$_>PW!j7_PeCQ6(3i)vH%AljSbQsAHVe>wXC9QOdzoCR?z3)5XgA
z(y^kD;X0asn!eNy;pQ*38`RPmh2ySY#3%*XZp<p7@flY5PI7V9syMCAcrr1=uhzX=
zR|0PgZuE&rp$@M@&gbfCv8iQ5me>A1q5lA`i~Q@=z91j?rm^bclG%j5D4t;yN<~|S
zC0;=0u{x1*F_Bhc>qxR$%wjLzas_lcX;GlWP+%}5R*+iU!x1lX#zvyp6+Bpy+Bfb;
zHLUhHsqRBEahk49DZr@~T{u@gS#wc=jMmFzDQ;<K68U<F2l&R(R(wE!Uq`fHvSj{r
z5=p1Q9AQ|DpIYc`ToMPhYtVg57n$1`_Nuw9A+g<0;SA7E6-5}Dz2eS3$rEy12?+G>
zTqVx3udUtOI{<|YYmY&|uO?c}n=Oj^UE6%}1O>p_oO{;XI(6m14AK?Py>rfLEVx0-
zV^2lVuA<bTZ#du+o@+y2((fATIb(B;xa&z(k0rs<m$XRD`#zwAWp@Ex7lkG}Zc&CC
zfWsAVoMiMBSV}R}W%S7{+=Qn?j-HiCqlQizAV<=NC2__LAp{V{z@#`Mp~XuirN-Ar
zAoes}70SwT+gd(DMiie)+tSz@H9#;U7!}=vc9E3!Et+&c@$nzly*IIo*v$BWF^@@A
z?B@sh{{Twy>6>(gz3a@xd_$%1mo(cax{+`=&1l`}jSd+)gIsTOZih=~(d#-PhEtZ?
zvHd~ys@@OEEcOy94FtG3J#+s6>aCH=^i3Dzgu{CzWp?3?E3eY9`gX#3<W_5QS-*7W
zujF)?s`3HH6;?=YE)Yo=A%_(P#h5JQ*bLmB^#qpD{i7&8>E?j3(cIVz>u6#sLWfV#
zgIiZ?wlp4<ooQV(RvI>veeP7nHHC1Vu^aFymFne+^=@)vovDQt^oNuvG_(Mjqcp(s
zZ-`&=HOv11T$A}$4U~Rl+5YIwcvb%Zi7!&nU&-?X{OfAe{{W*${_S%}(QZz!jMWW0
z_qS8r9_IkkTNG->IVZJHnEBegN^!O&EOYu$hQ?oAd5TdwcQu)`sOd`OFn3U!M_%<5
zb4hP9%M%oJIZ``SA{nI*h&4h9H<DDJ+xn(+_Q@$ZF)DHFNV@K*!~Q=cs+0JGRYS$(
zltxbA>Bid*hhNoLb9p6Mi(S<i9NtMpq{@XjzJPVTQ~rM>sE)m=mcYD{rA?RYq1chN
z-BDNmKP1g{x_+&Ed@z!=!xr+`=kYbAHrdZ!5lLAbN^82i$W?CTVov~8?w@VqJH4*=
zPYNFI*Yc;U6Ht!m>FjOgwM@%&^EYzc2hyaw@jO-$eAiiT?x!?2%8ZnG-iX4v@xG;S
zWJ_l;F3t~ASdVu#+cc9((lhc)9`(yoPDgDzQGr%y1K{@+drF_kzk#=H0me_YE9`A;
z8Ps@{LAQo%^lW~0k8LO<R#lh0RvrHA)wgWouf%JW?sOB=8Mrk_MGM22pXh7LrA?AD
z`W#no8>!2Q*%8&<A`d@Og?$e1%bMzwvS&0E<c|K@LJzR55iiMJOquEtG?}PnlC!WH
z9vQmw^>t3t3|DVno)!Ctk&RZUd4{nATWYs4w_WAGK5CF^;*zwP>OtNqoBkMnXZ$LK
z^q=X|bNpttm%xmp{8<QP5X4J;<N8*Gl+0{ZkxPIH$E|l|@{APZ_p!TaLH95J0BPI#
zD;niV``M}Oe#tyX{ak}Rt9f$$U0-%^N)mz<^ovQpWm1RzlT_PIy|l|(+yl?uHRt;~
zJH!=hu33+c`sEl$cW%MIyn`Z}d2MHZ8Op03ftRSQSG=supSbT9SJ8YsEu@zZ9G+q}
zFb`2&j-_QXYBwR2sq+tCtxZVUDAtWdDYjPo6s^-W(0EeHb)MsPaj<_n&Y#4MYd>+7
zHO(sO<4~SRT<%`gl(zTCK4|w*G@BhZ)KG>-P9%w!@TC%==+%>nvXW!B15ITHk|>3f
z29OSF$vWQ<xlIKC{w<*XRmKro)ZFHw@QF%?nuBs%sO5^?S&*v8K2k?o%5lwY1!Flo
zh@X15^r*8FzxyOrGbBj}#w&YCKO+x%&2CfJ_8$$fUk+K4a6G^2X}X=#i2zpIcJKh|
zYqmQa6Y5y8nroX-+kyablgU3?e79C`^DM!L_ss%NLY&&xaxLTI95Yo@d38VEke~KY
zVyPlrfo<83&x8lN9%=qUvyHpt9>W5fMpBU@T1;YXs;B~-h3om(nKAOO=0do`k&5TU
zFLdg`zj>QlQ>xq!a2R71N5s~uw*pbsmzv>^z;(;`zaqDXu2KkZ;by^-Ka>Gj@#Trh
z@7B9py*QqIMQV(n8%9`G?uT@fTFDDON}fH1L^m>?#zewME@X~2-m2K<nkT{eqbhsW
zOrl{dEIR%a^$@q9sbB?T2LhY2G5lYx0~6)$)f-C^gY$Ht21j(hVwD7d!=-QQ2-iAz
zkvPExN%yTG@UkB0=L$y`%`BUU?NK|m!TEkj)N}7i8XfP$S=6P?t0?Fke=3K>T5NiC
z?cbCDNwXjWwns|rgSsx8vE_sobM>Wpb`6L=V@aO+WRM_5=axN9OflAxymk=f+a{5?
zprNb&v1_r?Sz*w;%zq0LRTG~R@FnJQjh16gsXGuzena%E<{X+vbqJh}{X5rn;yhaX
zD$`I8^`(DMYLVk(H%FV@+DT<_VUzCw4<6$+)%a#9rPAWI5tfR3^l1h<#s??!t338}
zTE*NKvpt`lBhzA^?H4jPVg>|j>T8m`)y}ozsIG!+h)Agte|1L#^rjsTRPg+9SVMEH
z&A^5tpq_+vKgd^=TgLkT0E(4k;N4v0XSPizxiHx7JZoWbt$4>y(;|qN?V~{L+XRAZ
zf5k@LNvy02sTSaEC$Z`MC`eP6L}_2MSXx+Vkf<_B!+t-7M<mp;PPZ29%*cov<So*g
zcQKl>CAW$gpp63qpH7FpN2%JvRyjyeIu0r)YXv*=436f~Ng6d}4-NwidsAiA?5%~v
ziBOpS_qRB#oZ_9$lqDr&nAi0^E^9xxIgLuo<>znSt*;E5RnyDFSJaTW=DHM<S22_%
zz0&9?@v|-tbIni;!v#qd#Zm0h>d;y=$gcHMVQlkA`9v8xt`1)u!FhD=JT8NF6t5!{
zbY$Dq%Mnc_p{HkgYkG|&4j2vz1HEus*Uzk79oZgX#dNch3e2J~DFTLXYDU0RQJ!;2
zfvKW4%a*|*nFtuIb@L2m_b%Ze05@KhtR0br)#O2Ye4-YP;|-2T9Sw65{`zKf(Ykw9
za%Rw46Wctn0<o~#dB<vbl0r7KVRM{zsB+T9sOxe>#y4^eN*!2K7Cg5A9&=2MloEQ-
zGoQOVZu95`Y+79Y>1$+jR9Y6r=HB%zccA|O6*CP{IK4fX)5Yrn){7yjk|p~JclL{V
z)VP)H7{A(H^oLaUrOFPdpW25#YM!SjpAyN`(^0b%o074EHB@@0C=uY680%i2A&~ef
zOOKn$8K)AebT|pmCaaPvS#d<R#eUZh`|JGc)V?7%{74~D{8A6)Qoa@v_~LoYV-;aK
z3dHI~xVWn+)KF1VRFJDgzM{JQGyX;EUv#bDv^eXSx{gVy^yU&+<NK!-bJ*smxUvtF
z)pZ>5DMYy!Tmx*4=rL8;Vzx$d+_9xFO+p4^fEu?oxHr5qcj;8OAZ@zg0If?_+l=?E
zK)rZJ+2cO-Pc@`I6TY9q8LId9KV-3*e(1R$T5{1AdCa=az>WO0#E+4?-1V%ObjF$Y
z$Z_8l;>vb6jBaaMTO?9Osy~Xm3n*G_DvX-qsM|X!t!!=CY12n@8{?s?dbA9w86&x=
zOZRLfzkL%oXg#Y>OSg?7c>;jM@^RLc7ik(|7w*b$5<INiM=VV_ZKFTHD?Vp)broV|
zw}=n8DmkEx54b418j32dgi<#4QP+yK97__FQHuI}Bod<+)ba5T_l+Plu^di^-n%*+
z?_(3<WWHXF5YH_5AC-A1Qt~Q~TJmc@cyzuKw$dz?+Sv;an;EN<!m=hx=QYc@xmQk;
z>^g^*c#dv9hqY<j+kKzJQCu$xI_+>e79BsWX$N$3R92CC_6%xl!}!28t)e~zv6F%b
z;+NVf>rA601#hKNINgz&oi3w|S`wIVrDrE&rVE%|vQ7ufw~Cs_Nl3^HcNT1%*F5Uq
zM0G-qwP!&x0Fi((^9ozh!c(5i%9io@l_`x}r=W;rsPkUdRUdh?ix~aXB^B?UXl9o_
z^y~qeOw$9(J|Lg?s-^z`u1Wl>FHegQw4FCsfnWHN^)!h_<-EyRz4)z|W>{?_EuKee
z!bPCExCw6T0m&RyTPyXpwh@TQIV74^(5|MXt>2jek(X~^J!%6Qhk4~cGY*xUosA-j
zb4!!*t_4?p&FxMg#xi`hUHam!8M0J3P$}T$g-aSHR>7+GBRQnE8`zT0KQcdj)#QMu
z<5M@Otu>&@69Cg8gtHThgv)DER<&m9#ddm4n12n!6K887Qb*L*)ScaqTuQW47S~SG
zbhO;D#~-V=Q&n&LS7&XMw+3e%aljR{nnx^SDYF9mU%$Kr8BAzL-A)Bnk^ca6xydKF
zu4y}+(29z(76XFq=(Qqzt_5bz3D<|g=}x(~`#DKh?jF9?D71BV(hJ`WnL^~4!!NaS
zR&tU>71qq0Gtr}N*^UKiMAgS5sJ4e9qnR)=DXoNO+#JTk+PwDv08YDxX&y^?0+uIs
z-Yc^X(z(xFHjzUWvEiu0U$b{r;<{?PL7eoLzK+Hrj8(7UuPvU4+|_yb2BWAD=+}7G
z?N>NHYLC{vLvb^O&l^d{<z9{#;b)hQ_?Mx;_`VYsw(d{*)eJwCV3S<c6{gPGF7)H*
zZ+Km~xcn<N?X#%QlYktWr|=?S{8<@#h5rBqgZNinq{d<Ju#!){@tk{B)MI@+j=$MU
zxrcvp(W=ECGD*UUu)pZGp-T~gmN~CVNjkQBMH)**V7Jz;;Y2Gl6zmOLPZY4voe*#L
zPAkccMwyjL5qlOQ@g<DRBizd8u^bP^w5>!kjnTR{(AP|4nZY<p8=G2OSl?8yH0yh*
zWD>-!mOSFLX*ifgwKma}H-<`**7zgP)x8r>HuFj(Tpgo~Rth|_I^9J$G7H#MZsppe
zlg(Ea@$JUJ`BpOA>5syQWz+5C1Q^HVS+_nOx05AQ1MN!V*v*qbXvx~d3aciUa~piE
zxoFU?W<=I^136G>w)V2M)JRi?a6ejUYBwcgs_`TQ=?JH@isXf`t!jKnJs0i|`#NLp
zil0k?UsbjR0Rlmh>r%HhY|-(ejh5^H0Av6wijZ?#bt9mO!LFA^=FpFNY?KHv<a^e}
zmT@C5wPn<AW7hNsw~NE}Z!6^aXM%mtYN@K~=+YpDEGi$pU9s656Sdi#w-*u5CAv)I
zbJ0y`TWQle?~`n6)RsKe8Qi+I_Tw8>=4^4<PvcXpGPeu1A=_x=el*OSuc9W7zyrC4
z<DBKWsrDjqjG8VA9au47r+}iolUG3wyKtemWyvP6ti8jt4E^%ml)EHH9N^Xe01!kZ
zvscI6JeuH-!p_Q{#EUqIuBEY=jDN;9pQSPx4l`X4mEw*%@%PItvP^&}03NxmUk_XP
z7i9^^k$KGt`^Pl44SLC=z7ivC4!x?1w2Shv&%G&b5!jR}H&q$-rSfA3xcqB2Ffl0~
zc#3I>asII8fr!z<KJ%RYX@uJ$<#Y9*2%h;Dh|mwhx*4Qvy#ST}0G?fa#cKt*GFu$l
zTM}~TC*Gef)lh?u=A{%}(X~sRKKW(0a;i9BJBsbRM*x#fhUIzl7diTh>ZI<h3v_t&
z*-7MH;fSV20d2;a^f4@{`A4lkJbThQkaVUoJ?I$B@;<fNBz{MS3jYApB7Zt@Ik6An
zUS}kFnm|P(v`L8a&1X9kNXDrP9G-etpm?0D{s{2(gB(1vN5{8M>r{^$7WOgxSK-ax
zCs7eE3&fe*&PW?@KPu3=dtVM|DqMjqgKCJ|lk7bQzk1N%IE&p%d*T(2$dvrCuR;DN
z-mK^vb=IY;+resu<lYa^4yJ*gljGUuvhe-BiDS8*Mgm>E#xwlIQ}}n`lVyI|wd*MR
zIuaMK%{v(>S|?YgI!it7xiJPPLlyb~TrP*GT=<{G&3}0&EiIqTW$Z?HAB{_8LOog7
z_>%ptrngg{wO%j)9<j%sgX>1BSlV0I#J@bUF5Z1P#U-;Mrm{u%zlOn~iK8g03pP(e
z-{5Mk#;oh0J5Bq-9IkymD;l!4hLob5u5$^k+()U&aIp0t?axt;m15d-pY2mk12lIz
z+yTbkb6l=n4$ez;IeljC4M)V#>Gtx)_86Of5x(<d^*mQ~Vy52FBLt~`nTCHV>zlPR
zDsjD4b86IL(Ec?@jw_mbBc`9lBY^SlxyaLCW0TC7{#C-M!*g5FR*8*iUPou)5P!30
zKCj2+T%7qg5;5m2YonpfABm9d$)j#MP*{W<^`K`s#Y5Z+uw_VPBrxtO!MeD}V<uP}
zgO2o@Wpbufxe`wyc^I+pnx;Vn3Rf!DB#0f|c&Rh<j{c&M(Dr3M)Tnsi;*b)P!0suk
zdFQ|DKn1pyZ+8fqMgfV%U5`t)M8I*1+RWsX)$MaLRS{aY)NJOnp}RTc(kUd-dOMus
zny|C`)tuWg-haI`&+kCSo1Ay3&Bwh2?l<ii?^ms)Gg~(~!JxRS8P{P$5sVtnehnsE
zHyVsYihT`wwWu?AVO8)zgn~Z`B09Sx&E`2LHD72attM#_&4R=(;xEt6Kb?AS#CeuI
z8qj0!OOMi}Y{f5*C!Hhit5J?cVs#?ab!i0yDryk4(?@~At#&pk<>(fwdq-U{R!1ds
z2U2VUNufE}JZ;qBfU1|mazXpW_F98y%~2Cs7LW(E$8~D2aJyUZt>R}PEl2O;Kkbj_
zU3RG!`&5(rrmb#6DqC=w>sm2@$}3a_i^mCR9#8RqN~W@@?^?lor+qJgcq6qW(In9_
zILeX8sH*J-&0?<6$?BS2x;bP(D>T3trF{E$J5sQ)r|;A2UCx6DShx3T#-X(}snDkk
zwPN-pl>)H(QjfsGf5nN;+}rsR5!m#tJ5aIM2K5;=G?%=}#C$1vnYiYqTFo6KqM*5K
z3I>FXVFfzAzLa@jmyzJqcbj7kH#lHNt!j+UXX1pLtgX{Oc0v4&aQY3G+OI<}`BSZV
z6}~1_FNEvQ%65*_s{Cy=%8iQG&~H-9!5u4mLX|I4DHwdq!+RRi1{Ic<E%=H=va}Z)
zhwj^&&(VBEV|vmE-(>2U$FDT)?+P#XWed$#?p<Z=nXsgWQavl6vP`Hg%VVjoEJJEm
zJxnT237T=OTecs*PpxyU(bX@7m*@WgSsMG&UUXXMp3*pJ{80In$*9S$W4ww&0L@n4
zUqh6~-IF~0W7!M3jn(Yf;1IZ^-aTsU0p<@{Ln@B+><5;3@bB?Nn|aJ=KT4}6E$o2w
zTz)n4*pK4N)YWWbKYG<gFAOW`STd0gqU0*3dYZ10vycTRu~xe^A;d6&)Q*)5@{l<>
zW82cOTC;7D03`By)J-AHS|N;S2{jz6^Jc9O$s<xw98;I?icDyGlBS}UpT?3IKArN)
zM`K$7LB|z^SMMDZ{t+a#-OyCf+XWpfE@N8`k}#n0TN)OmmsW<&G8zsW-!;*P_>N3p
zz;jxiwxs&J(aL3*DfvL>gVUul-$T?cBnq~!Ki(Lvq?=|?lvGRW;XO)8K4nB-UBK3h
z_&tu@_agvu0IBwEgRGXtmhhYjX(y8-G{t^h%m602yPJ6}<}t`H=x}N$9#ZL4=iG51
z;i}_ednQxqSTms;9vzXdGy@AZS$81z9jgl7BMj{sIoQNjmA(XLhxp`;yKlNHMiX2T
zJ4=yMrTeC;I}*&PxX;$DAy!ZsLk~hJ>^HHKtoTyPP>joLg5G|-bZ@O;=z47SH<K*s
zwTB;C>h_nW(>c^FG||#VITbx|$mtv2qL^iBR^VgvX{=gYj)^js{Oi^um>FPho0}Q>
ziuEws!1A%*5;;E}!{twJai6`E$J3hRe86j(vb9Ea;a%w?f5Of4Tz(aYt54<MTn|??
z-@u1W{`r{bt#fju^v?pl4ptFp0CZJ~x-jn55LJHkUzyKZ#wEJ*&&&YA{{UKxRJds*
zj?yFZ$a(e^-CEIAB)YpBRQSuzkwI&>187>pl3z8-C*~`Q5fqdckqPcxv$F#PjGESg
z8Zb7JK8CrOo7Cz&J8^M|H%M7VFbzs|O}8?u#ts{(t-0EYxyS63rtDgeQMZ!=aZ~Op
z%xxT;t10&tagvdop(vH7)1jG*WQIPKN%WZ)I~+H+N{LgxsMZy&qbtd0JZB9VKTc}1
zDF>xu(_5V?N?QzN1ad00u&D$AQZk&D+;S=?Eyvqo<NjI?=Tx4edm3IR1V;)Egp*vu
zMT1(+>Ss&fLw3^I%586=m;u8IlS5C&AU2-^j(>UE>0C5}n$>P4u+<#bNunIW!_d_d
zhq~>ZhrMgruFHT2bOy5OGto4QJBjq0cHv{&_n7h4rn}Q3NEytsIRhtR@%h(uc1N2p
ziIp_Vlp2VR1TpU6q;5&Bo-1X@>c_ETlT>3Xw_<gP6}I>E6*{N`rYRaiDG^C|3}mkD
z&FG^hqu+Hm`nZH+)z37I9-@mK{M<^?pH74FuQ>6woOW8mJW3T^dvvI&tQn;(DkD1D
zL$rRCzc?2<LCa+9$v)z|==Dc!Z?et@S4<51)G#nAD@t$OIdKc#Eu)yzik{B)Sgzu7
zG8KWV>T<MiYpWyMB&wMzc|L-$#`1D_r&Dqxu=L(L)5&i9)=aUvV^Kcr@jww6q<-|D
zXXocRriLhm2SHs<lBq72<q@0$rnINIGuXgz6Lx!55E|lAdD~EiTL45la6ek^E_F22
zyg6<ok7z&@v+IiKqV)y2<ab1i>DR6#U_UxR_iE3er?Mtq+4QH1M@kTyYZ?n9CqG*3
z^(ScjJsvnE5Fg5#IqPdUdYr3qK<AoAVA;OzD>3*d6pSj7{o{`2y${63)*le)7Oudh
zt+O*J>~ISHwMg-CZ{93u_AMpP+1Y@eLFZhzUo8(ibK5n+c&k)hS6iA$Nc%D!kSG`d
zM@}n4h{#cz^}mN&WI7&!KD~UQ42$v>{^<Vz8U`142L9|#ozhPK04jmcwN>#Jtu4Dj
z;cX?!x}AWybUn}T=jlzm8AdlX{8r!Do*T39zL*PMT>ykA=N-wYd?}*Z>Aq${n|mLc
zdGr3VTy**`^{kn(ikgdS{xVCF$IB0x$D)pv)Ly8QT(MUUV~t)$E7*)<`qh1t7yLw5
zzql<7_XnIr&bjyGe>%p~Zf3o=7P4T*qjquB<BH0vG@*O47q?eukhHN}!f+!IlB_**
z(-dmkQ%-1_d1f-;MyN>MPt&F<IU^KrcI0AO_+AL7QzrJGa3q6_S9NHNh&Pf>JJu4L
zk~<+yM^kRuztIolS(n#IX=63R{n*g4r!A43J}Dk0dwBP^lH5vi&e^DwHP;=ENjtN<
z@Zbrci8vYDO>uLd-oN|CwCH6W*uXQ6)Yce0&}cN^azy|Q!`h*Z6M3aT$4YRR_Bp9E
zLA|mvb5Tk#cLUat2#iO6#)c;Y)_@+WS$AYp)bq&{0OpjD?d?KIl^Mk#CAB}eA6lDn
z^)=rggz#&B5?-b*uzz?fH4Y7P%X8JE{4`mLQaaUvIUVS^5{{z%rjXp{y(Yw<Fw9LJ
zY4xXZ@(5-3tvxOC6oWJh#5YGI)^o2kF+u7^0S>tJuT;_rZw@&2U_UBR+;w(GoZNh_
zRP_d{;B}#zK$_HlrNjRKT^IS+rhG_HbekOi0MBpzl+ybwZ{vx;AW@CAYChG0(ppNh
zA2bZG6-gQET`GWdt9*3XU7uwfwe~r^NDbW8T^3wwdwVz{hq1*Ui5rZ(nt{OtipD1L
zCAJ?dihVOpnj-9b*F;R@Bv6sfbysNbqpmq2LmJT;vW1@GA3du|1Hi37t>PP#1&k-w
zPw7}u>iFtvqU?^^PuwoUt05yTT5#MfurmW$;${1$bko*G7NsWUBC^?Q!aGp0FYg!Y
zUA~R}p?;N>Luz}Qag34mtG6gwK^=`aFL9Ak+N4Z<QZKsID}f%Lx2MZfO>R|ZcNOMm
z0a7Te4x&!gQsnXny=et`Bn{aT4wSX_2>J1*m`5U-5zBl-U$k0C<DVsg{-4&dv>RWu
z>FXgMJe=2!SK=y;#+!g6&UvVJHtOd#WZQhi!Rl%HEdKyxzi3;9m-*F2>O|GzY51bh
z-stY?3EZ3v{{RrLKk&M|dgNdnhXgfiEzWpr26XojN&f%|JO^pm(PKW~*IjEK)D`FA
ze~5bcKj9+k3cF~#reAJr82m`-sO<(?o12Kor5BwNx0cd4q47YX-6}RU%hdD?-CFf5
ztr}Ns4C1gq;Ua^09%$yit13xRyY6`S>Z&tHx|+OYLEJw&L#)I)xcuwYm66HLhTUog
zD8$9EwL!c|03Yi`I&zGoaGd!!Ab%A?I!T(XFN%8X4Ao9?Mt;`&9L@ags%xTVQa(|f
z`&3$_pDM?mrzXB{5bst`%<G1+Ov`R-ZqR<^PoS(BaVTBCnR*m-T6(q^BJt{KBwTkb
z7TTHgszmQ{YdD%nSr{S#+|)%&fChQ1L@?N~^4y<FiJx;Cq)QcKJgGlQdLBWcpiEhg
z-^Q4M<RAqn@TFng)3Vbqx8zz&j>loHk{=K15-4F10Q$Bo3XpA@!mMJoW|Te`nUonN
zRX&)_T#LgH0sGllpZ0j9sdpw=j>vqP9-j<He12#9#-h?~^!+yX%#JAHS2$yu>ZKJE
zj<rcoLt9#FC}LZ!@u(az7_1iY9+?`q?^uvKDXN^3WKx8riUY;a@4{QN4D}}!PJb9&
zyre}8VRQUdm$cVIp-}E@XgAjymAl+bMmKE=2LiEm@D8mg$5jMTQ&^6gN~1<%oS_}X
zShkZf5$jy6>qeJ^W(#?BZ_Zfm6q<wr4PIqc+NH8TD$=gB(G;|*%Xa61S`eO<!y~l0
z8RDMYtvKvUGcI^yDoEb|Q)L$-l}N)@ZSnzhtOrq5(1p0z`qT4Sv{;#@KO(278MkS<
zee2L}<S^S>oZ#<I<zAK(W_h^ftW58T5r4#*JE5PFL4)<LIF9FcdRUnI_b&x2(XrHc
zPEE$+=qn3V`L!#LP3E&}+(xkeEQl>rY^A#>&2@Uj?H`5<DZ?0sKPu9W_Z4SnA5$X1
z%yvt=_(vwF>9RHEr!4(K{HlMQcFi|V5bktKWL?(o{F_3c{LRH?Xp!zOyOY+j#XHny
zJr0f(3I$z;Jm8*{#K}4h6mcivRHx$ZPxp;nz650-;xPqCl00WB#C<9ev$Nc>9;I-d
zTYuK4O<Zy2M~_k~RaLFdc~DPc)OYsMjwA#9?kUX_6R#?)D^%kobGXV3uP&8t<dK#3
z6-sR$Tz>6?dvvVjOR=;o{{S(YEtSM1t3<?iAY@b)5UiSvV2%OJWRp^}HgZlaqQ{9;
zH<CMy*EZ=_KZu%N0w86mWL`5!(9qXhP2tN|9PJ<<%DBfI*2qg?tM{v+(7xn@=}d%K
zxLjNxden#ImIM9h>o>VatoU-?CeUo<ak%X6dU9&b^o!*f#sDK5MR&tgk1D&fXE7Qy
z8jOD~0BmNug}I6xG_wFZ4z$SKqCs(SInBMXP=Ah2KPq^$l27>goVf0*!kU|#rE<h@
zMF{y2EjY34DHXENje{S&;8zRr0yKMgB}FZ@uviW$OW|UMlTDnRtcL(sO{KcaWYLk3
z4iCL~vG~tQzrk~j*J3IpTwsEDtczC444!=|>1=aiJDLz}Jw|HEfu0xft7jWJWyR)=
z+?|~LMRTD1*f|)gpGHJUQ>`ib)=aU#T1+YQ#RCw&-jx~TnrH!6oaVRm85N?IHS4=I
zaQKOy#ull??3VFo?{oB`D3&D{?beeHgGn+z?0lfs=8<&Gr%i4^!nk2tQFf5l=bzsK
zFYXBG=C|{ti;<7cwmtfN3Dd7M{{W>2R7RXrTcrh!hneF9b#eUbvGD_i($X>fOfmRV
zTO8Fjs|0e*o+uS3Y9P9kYLGhCdm2YISE+x)&@Io1Y$0_FuW*5fsLwUoYBtbZ_`2g&
zxLwI<0l*x9N8J2zllWGN9xf~69GAyW@Kye~GWmBpk@Djss9uAQu217$Tuy4y%5C^t
z!}II<yjJn45reu{?5kaei@ZXYTAiPVwAKx0YU;9V=)`vXC<CWsZ{e>D%dA|1_JY5;
z2dF2eD~<59k!j0w;_X{3`;U;Ptd4t%nQkb2CrxVi7P>9RiS+@$QosG(sP2KdP`{Wo
zz9fJ6N4y(trXwFd?<Qh;ws~yS6wEIM!JD?6jsXfQth{MpX)hJF{Juz2{9A~}=nZN1
zPId8_I{TH=<&9egNftFDxE+5=k4S(evYn(BIp(?RJx;7L(&o38gfNi<DIAK4ptVTY
z#{wY(j8=TD(sJx{b0twnxk}R0bgd~I6UAv`0=ZVlYcJaFHfQm!H^(=!TWBpBzSAph
zBh+`T{{WQaj-1Dr54}$VHQ0_ucG_DM@RKm$gmYYL9P{s5bTWqQUYVw~BT_arACzXZ
z8WlO~=}*QELB%4_?B_gGc-(sNLl~ou3jQ>qcpOpz-4_|nHy<@71A`n9+K><dpk=#L
z`+-u|e0~$at^7%PnL3nadzj;1tZ*w@KLKC5D~#cWdRHw+W_lFn)ijbOUjb?v=zl7c
z{tzqS*EKxFu4lR{%ix>j1f20)-PA8@a_~P4rxY3_`jKxmZ>lMwcUHEI%yWJe(Gpq>
zx;&q)QhIE{1Y?I&?_Qtb=@LyO#19~x{vwF(I=dVOrUKtiqo9zF)~Qlu$6{+io5+9P
zU*}$%@g#r5qR0Nbf9$50*=4_uCj%Pys}Y`+iPBk)twq*^GK`na=Qa-j@m;N?h4d>{
z^-Ak|BaXAWFf~4VR&BJuX1a~XP{N->jc#=>g4yhOs3c$BRuSCjdlK4h^MbcRF;neb
zzlC%_&RYno2jN{6xH9P3>Pit?88U<mwM|-#9174zUyKV4D%_9x>-{T{S2*ie6S|kc
zJWNG(T0tX54o@sia^p0nbkWvEW#i^kRZV!upgIxz(e<v&NB;m+wVcf<?sO3LtJ06%
zQ>XB=I^P^j>*)N*w)@mKqiv{PaZ?|LMNRwhMF1*AtPZi=s;s=qwY3F$v0h|Uc7q-%
zYwJ%0YgAIWjTz4YMrn-aHAv*vblD7OSkHn0@6A^bc^2y)F|cC1TAP(isSyN<a>h?5
zrDAH!2#N+OwBup(0Qr}ZK$+5La}94rk#Kh$@Ov8J^!)<z)5M!mO006@EIU@U9ipVp
zYB9W=PWtCbSY8>AEV<^a#I@&Ecbh$&EO}`pT0vJVhwq2dvxmfuYWxha-|_mtN=w7{
zMQtyJUZ*5izYju-nu)8OZg(7qj?A*!K?Ge(<%hL!X2SE#1S?n6;9%7hczD-(gx*G?
zpf2adIE$!Ac?%E^de)`ysmloHZ(Q9Q?Ic3Kc#gG~4us4G;!S!KC|)|3RxWK9pz!+t
z0CbwY7KVfVS=UOZ2$LE~ZenUSUuD*&IKT*?epSzE5_y+Rj*Gyrmc;)69$u$f(j`O9
zZbR~009a*5HO|KqH34RMdy~aW0-eOQXCqjWPWW6@DbqD8VxHr+2W+=NPcns!>=*;R
zGD=M$a!qa7_(J;OL&(b#xC{<+TE-?bpDBJH1O<G}ce#BOVAc41Bd5f{i%B5#W5sF8
zVCz1Gwa~PraNlK*a`ZW=nr$-Kwmg>ck=T$ZvPAPrLcYDGTk+?H{{V1ir9LOL3WgME
zLHsM8r9_UJbY#RnCU+mYfGPg~Xy&1p#22zA2E_MKT+XxTbw>{sxzFm_zN2*F*j%{>
zs1=C5aFDag_*31FCbfjyT8dCf=#K8nRo3(kugewMbN%7oxam-%G)?!E;MLUVWk*KQ
z7DPx`{p?^=ai((LtwJ_CPYR8iRGj|+%R~HyQ}Hq?-oWZPbvZSp{sWpX_ofg#Wj|Vy
zGqKrgnVo2}pia>Ei2~p(q^oDst#lqBfQz_h>{o46wWdsTtcWf83cV5OTx{(wA>yU#
zC|eTD%hl$fS^esV(2!hXy+<s*WPth#092=wQpP4}8<A24D}lv!J|Mf>;S+rs!9+#`
zcH~!n724;YjaTLMEO>=KwLA>c#y67!$h>=3i=?GZbXQkqI45Ye=5_uImvOoJEoJIX
z4yAGb09_4pO?#;7gZ=9uM=Fea9xJ8R8$KS0b|D2idokADKBqZxcO+0ct6&k*t$2D7
z9fL~5Wtf`k!nZl4C$Z71Wer;3++P(3Dx3<lq_9hX4;zL*8t1DnX*NSgM+n?NtLSh5
ztQnEDqJPUz!l>OGT}Xf0HE-ZXQT`(kHfeKNn=s8MS^x(GgH97VAC-v5y%z?qO1iDY
zS<CPCH7e`M01`*fat&&#wYh}or?EQfc9Rl3z(3yNq(y>4vM$4pj4Nd0BbmlgFnmIP
z(-k^^wTU9vJiaD&Q}_{G)~m=kG|@%7wcX(j$M|3R*C8P0wTOl<TH4Vc60+w2ilPyt
zc0YWl+O(tVYTdTtvwM_!rh@989kXdAR85`EJDRcPd5w#B9z*@&HjmG}cf;XFky0(q
z+{x4}WfC-!vVy#gmA$81+r@1hffsQcvE-U>VaC_gx@8#3jMIVRB;uyV$=qSML&9ax
zu;P;wzg$K!^)!kX$z=U?SR3E>1&mT?p`}Q%;)E887K#=pP>K}^?ou3ryAxUp6iaY-
zcPZ`^cXxMp*Yusw_jm7o?(@8VC6k#snb|YvoPG9Qd#zBfkqeNcP7nO-bHHC>0eSN!
z4);j&iz2md=JYXIVasmF3kQs5_Ymo<iK3Fv#%_t&wfHxSJnDS5Z;bn|D4}avW-Qi*
z38pN!yn2$*XdT%!4;r=M&rYdS%^VBT1UM?E%&dpMZ8k%90uF-fy%_u(Nj^|ED46K1
zI#|+ajCP*B;FBzQ_q#ib^;z4t@0+&{M3O{hnX(nSg4A2__F7`=blP}Omp$~%qet0i
z+^LRs56bT_l)q{Job}4Vv*YcuFn_7X9DCLMLKpm-v{P@(_E-$}4!Sa&L_$<?NtiND
zM|d<joF7Jy8sF1mK`3K{m!{xPoLpYIej<lZs)f@TmXd@^jt4wV+pEPCcJ|btmgeC-
zEgGs^DXW0Kwf+vj9{==+{^vTRV1{hWyuHbR&Y0#n8C0S|S0%wsnzNRZYVE>sBtKVE
z5jOHwDCoL0_j*WA+ZK1elIupE%GGqo)M0S>*LBI;9>niyQ!v6yl;~+wOE;)1<@ape
zu?NAuI(MGNQ2ug{g-v5lOig|v--!6x^YUBbjtcX6cU0t_D}+ICz1#=17YS`)3~h*o
zv+1c_rbA3!T4-vcx|SJ2zJ`T*I0px&^oSklRBt-CrFC83Q^N84$C~P<JeUwoJ!eBQ
z6fOb_`7<hK+_6c7&}n|9AiD3uy0>>C!kKktq|lXlBy~)dRmaQo2@D|j6fQM)T3rBN
zreLL#=)1WrSyG7K{_}%vjPjyfNQ46S3A5zTv1D#||JXQkc@a*`;kOR~jaO3#_QAP(
z6qqAxJ6+di>U5a>t_mv|Nr{|<wisMR^_T3^Or^;RCD^;@PA&bdJkb3s(M&s?wwr3#
zDeq#}U#kjIe}sX6Drl!3w`5z3WaxzwtG%xIw!q>?s<%BN)_{~6v;WepY9#}@q}CVL
z$;42TzZ9onEj99%fOK;u)K%Bqs9CbtgzU?b&~OR|3pR>F9DXIVAIaO<Mq*g^9^*eo
zZ-!8w<93IKOXqS(Z90n(NrS$H%e+IP-Ki_ZL~N?$CeP`aBpu-m=1Yqu8#dERpUEm|
ze*MEbt@`)58osNU>?gvock2SMYbalnjLox$s$f;ppaQpo=T<awo$&Uib0m4vLOm;%
zu`S=8Xe)RBKa{x{@WtRbB_JgFtPuUvEUQ4qW0A5{ZRE=x4n|Tm!rs_UVtuzM`rN_y
z62kXSmSauqovvte=I}LjqmZf5+#EZS?{|TBQ+P@P@UD<|Np=78niBqBUK45xlR2=C
z*kkFLFD~3T&%nyGMnCtqtgO*ZBm$6Fh7HQRmR3gk5dl}lj1t@TYJu;jwRsbVNim$!
z&hr|~Fo*&M_jW(OdA<*kn=%=#f@6?ySMhMhb`}};M6TkN*zSENM<b>FMXN40+d2Fj
zi)Xqd3k$O%9?5F87mG}_y(|?3x~Wl3T$}%iZm3b;_;>kV)I;`0D-97dfwu`EU{<-W
zd}Z%aIwR~3bY0;8P#TH;&u^mQEqMi=M6?4mN)g)Yhyt~*%j_!76pwz@kXa*GS1=<^
z)hRPUvHhMZc|t4Z{@?DabrbDdt8k^6Nh<2V9AmfGLrrs6{L~y5L_tAPmG+DJAFH~0
z(m#G4{Q6}8x>r}s-;;64S)-P6Z`+;P^!nC*`@s>*;s#I3$b7!!Z=RIfto<=lMqqLS
z`NjKwi%)=1vJ{sa46$9>@1{~6fg{lxxEefc>QBhjdXdlF2&shbIb>~xo1GDjAc9qv
z{BTWv*E;Dn-4#zYcd3SAVV*%BCG!GqvB@os-CjpJLjF7)9WrGR=HS!wFboO%b=Wjh
zEh^hWPEfA%08U)RF0E{D%upyD_fP&%-EXrx!>_T9ND5(dV~91P>pGJG3Fh82X3a^f
z2&dlPTjFL19)_=pO%&Le6Dxhj-dr<kXybl(Mp>js?nD)H@?qe41mhAhKJTPU0Z)5#
z*Z1SuVX@9}>PgyYCp^TK4p&%xh)43w;5s<$ld|3qk`LYJa{BJi)k2AH>xwnjB|&&U
zly=OEv#s$Nakjg#{ZU-Ae}p}+;NvBt>`YRi-cyNSiTDzr`IdWs2etXLWza{i?+dQ+
zh8VUj%;w3}&K49=ZW9W$N5O9d>|osm;2>u+L-7Sx&u3l4Mk?#U9Lr1c+tTz7Jq1gO
zZ%z43KY9;IXHFO%6_qY)_;6R4MUnI@5%sB3MlT2aw5uxapmZ-aGUDe#-`B(yG_0(C
z5up(SVzI?2NS+QD`FWDSsyEai#q()8=Zq!uZ~k4nfhp-otH*9+lRR&u>J{5&twjzg
zucEwZ@XPCb7EQ{x`_4!zD*4bT@UuRnBSp7Z_RY3`C`YT_?AimXd^C^BiQ$H}vcc-s
zb+&CYjPsudBZ|JXZR}7Wxfk3*%;r{CkI@VD(Et{?uS2zhn63YLMu#j}_DTiokp!G4
z_hg*};T}`(N`jsvuME)|6G;Y`yBN7YKF_v^p;k?5c8-hfzGy?on*tVmF)jwRK7nsP
ziG!3N7&=KSsK4s$qa{NlJ+Wrs@|glbZ`i&UOGt|8R&xjQvE9pC4(@y5yEg#d5%fQC
z2N}fuuH~yTXG#@-dbjsiE9;7~u3k}3AdRhYLBQa6(M{IA{S5q@Kjp!f>9zFg`Mj`Z
zMhTm@=5sb?bh-mn*;5Cu=h>2!q6LN(Q`%|aRP{wtQ&^n~qlo*J_5xi%vtaN)A`5*~
z4#XMBlpvb?NVBmOd0-BC7>dL>gH65g6#2-lC#KkY4uV6*1b|;S_pXfeg=PS*(U<Sj
zer9!%YqkAdNhE!RBdVSiA507}h_pnzl<;i6Lw=cpj$nsVJ-vJ-;Q`NvpnKRMIaF11
z=v}jG*m}PamD*|76-t*zLY5t$WB;Mdes<(Q-4>R<Pn`LlZO40E+It8m&2ce>D%v1X
z-8N8TyT!GQ;WiW$))Z;uWf}g}pWY>O<AYh;w;B-bsVsl_B0|!{Cb;SP%)g<+(O=LU
zk6@ri=)Rm=7Y(qAS|~BLygste9)61=tR!sH>d*Bxa`D{LD=X$02is~d@>f$Rp#Qiv
zB286uO1U22>hkCN`AfRG1BqFA0jikH+XMjKV%`MjJU>g*Dlabio9jU4L6fb%;C9!|
zIDctLIo?<G!hUAC>H!^QKKMdl;zj}k(0%x6+t)_;#v_#x%P9G5-e0%f7sDR^u^BoX
z_FfzMigu6(jN+l%<Dl2o>@L?eD2q<R<O<SCAE$+)hNU}o@jZTlcfk@(`&QMCec9;r
zb(90vSm5B!H@~?DejSCw*j1u`1TWr{AD@ctEfu8X8(i<n+FS+<+!OOxkz4^nZ`!gQ
zn&*}0cOr=4=oYkq+B5ZxNzt1Zru&^Q=bP?x?H13Ow*eSm8nOYjqoV6OBE&oj@1Agx
zgskCF1|)eFEite&t)g}RLlFll|1Ei)RN#b-t}DM4ZAsSh1<R38wNNk-es1<6Qrc+9
z1Kj>Z$IvoWwp==E9sbKu;07F{7}0kdq86t!L*HnqA+XL(8COoUizF!r3XJvJ*UfbF
z^|4m3RD~Fw30v@=9wTmz{?53bf7e|q9ZStl^*bdk1_jhuni%A}1?(E&Nl6%t$jix)
z%}&SeU4oxOJhs#>R3g%NkvFz&>S=8Ol8SvF7e|94_V`s$mr)!f+Gxa(W2>qqdfe`b
zc5Gum@HgsC9{m*q#iHxtVeQh_E?b|&veT>pcENdKz64PU%t8%Yo2}9wrJ}>V+#nmX
z*~ZvRcduL5PkGQCm8Te$M#KWsrSSu5-$R6AoY1pUfKAl=a<B0(r~E5ai%0=G)QTtO
zPVR&4)~W3z+R{UY5$Hi~4evJc!?q$tvPwJ+$##R{kNN#lqtj@kjPp^@c->DaZrNdS
zOmfutbC6c7Ia&;T2~&9=I7d_}!RNd8ULlTqj}|8JGpXJofgGt{-+itq0zUGlXEeIs
zA-=i5m{436Px-Q)C?)qqC_npISbqkX0eNh|!&PBnY3aEO^=Ti!es)O>k$-!gRi2Zt
z^=?yJiQH1{l8eAR=vsh%mO#pyMd$@{Z1KCw>Py0v3atl47qXXDN8W|Gd02|`Q6Rl5
zd*dNi<){WQd!elsAh@97wdy&O;ih{!A-R~(C22o3N6{J-u;j9etN2(%jWsDWkXPZ*
z3(FBh0&9Ez%d{Xg<cGKVSAgQI2^UL(!|l4=OWuMF5V!I0&rKteB8=Y&jqAxoO99tR
zuQ;Y=fn*uAjO`7xTQYYI+owMo>y@m;ueFkMkAJ+2lV?~mvJsBSMORO3*OHnSAX>D7
zdL0kZsfYE0V=S>duiujFc%g1Lu)IS3dREalEa$`$s!3S*c3lA-#VGhSUho(ihoE{y
z?gHW-3X5?Nt<dWD&uRHdv#pbf<H|-fznoI>f&ztI&qioJVj)mFUfFv?vBW*l;v>vU
zT_(GQ_egWF&c>;xXNxG`J@B+Jr;@Ym?5i)TCA|`@Eb)A<eg~tarU#M2<iL12|6c@=
z0zlZsy=$_66#=Kem37NgAETmH1OKk7cedtTkofB`$IXf0(jr^jgamRIy_{!OV(W=~
z;Y@`kr_?v+{6&wwonXG6acwgvf4e?nPR^RBJpjcF5Y@>J%XBDavn<5T2>mc>O~*78
zkkB%GK#e3woN5(st;9l?qVQD5VI9vMV?3V<{VlzcmkbY4+1S&2VLf)RcI}f};%7qY
zi5a={d4PVpDxmOVdv0XnyHJ2q5y_<wT1MnRW_@x8Zi_~9<s=?4`)r(&pSFt#O^^nE
zxqo~L{;oy?6VtbRCGkdbWGb!c&_c`Y=du)<Eu|2&w4k($yHX+>->p<Dafsbf7cZe@
zAP*?wKrQ%((k_RIgJU_ZIuVw@5fwymoBt05l|Yfw=Ul%Lj|jhDwo4_-)ThLmbM81T
zh?cjf`n&q@Olx(_L(15#LPdo%+9$;!A!7>^<q5uGjWJVJt!xK&jY{VV5*i<YM7Y`=
zGI}%tFN*brB>R9`5NVCh*moYX>MD_q9gOqsX1X;oZK-TyLU4ao(2(jrf*ID*u^gE`
zt0<1*Ph?0w@SFY57nCEV_0+9yIV*Fqbqvvm*;-hGy&ADjbqplFIPjv>UE$D4qSk?=
zn%p>|d@}$TVw)xnj2B3jA!Sg-JXV8-;4W>M>6%8k^M(=Hk3HSo#+Q?qqV}&AdwEKi
z9ru|lSU%bf%}XcCy8e-{vlExQE@Q(kh|)RtPZ^F8`sm_f5ctcF=eY#rK#jX@d_zjJ
z;jn|T1k~<W21l9^yUfODxCj+G@U(l|>ny?7g}_O>5EvVb|42{*=4vA(od^CA^$+FK
zd($>IB9G7ib5OdcII5{l1ddf^K6hA_-xP`XqgPoYdKrv9`pG0uvQaRS*ekt|hRPuD
zS9$<cyjgdD#E)7#ijDD_JRahVm~Zy^G52q7)hx)rrLye2ra|G<F&VUeF<HU0EI~b!
zi6pcWlkVLYeDj}s0)?!3YKH_S<b-op(4)t(jHXCphGh`6fA);K`Ezwuicmg%YTP||
z<%(iWY0;fnN$0$|3AcEBXL<<s{zSE&CrnhiP$fghexR_IWm_y|cq=go8_)#vM<MY9
zWML&y7ukCm-i8uexq-IVPMvT&_UU45{Ad4ASU?v!hY@MD5ib410zwNUDa+MC?o1&C
zJ_a{o;^OAnx}=WMf1dG|kr#;fJkWZ(70XBDZo<59r{irL_aQUEC`M9)>}~f-W`g4+
zrs35nY}EkO7>OJ^+~T)?Vbjl&XX$Ui0Veu}Fs*fCq6TU+k4D2S(=WW6%Cx&S0S3yu
z&`q?@>}eADMnbL~k&>VQtI$xC^M@xn^H-X60q-S??3hahXaZ!^)S3ev(Rbo^8uHK{
zp*6t;Sm!;DnwvL52l4?h^XUt6m(i;02t{dA6kM&yiTC)i;Knb@J3nN$qyS9-+KoEB
z7w-@%VZ!?tp{7MkfwA-m+vkRfiwP@nVm;N&Y8BJ;zY0hU-3v-Knfg0^;#dxrWGCz^
zx#otYIDo;DW+Vb4MtV+3S&c4eG=jlnUXrw}0p`>aRy%&m#Pbp)WLe>r7N275DSllV
zB&#87@nwu7kLWE%R-sw{6`pmdBFfj}pcfi{el2GSNHB(W<uizj^$AWS#B%k$|3Nyw
zymj53Gke(st<2MuzTPdhnSCtAnXjV6Nz!AAwf}Q%g}D>+nxghw!SPoY!iGHi%Mvi|
zFm0nV8k#kOm$-~k${BP+KmOhysR|FwaiUoOAYwSQZR1h0$6^qDErSE8T+3TsF=%#b
zFXX&rH&6dodYTQkm5O_j^T~HTwOT)8{x7ohk8@2YJ_veINZR4pbDZWhJh$JSSySVA
za9d(T(Yxjov)l2#_z-?P-9G;hML7_t+sUy&8)|fLbEOmoz`1F`&NVa639S5ITWMu>
zvvRE5O|XTBWgd2|4V1>et*LlhPBxhc>Jgke*ZhFC4e+n^u0|{(>3&ee`8s_%YJJmf
zs<WW&bFL&w;RM<lZ){{P5R_b({)xCW&zU1Rs&)e>5VG{8c7r>Y#IIN!CPZrZ&U=UX
zULeZG8(t^9!)}RM<CQ$_4`Tg&=(~yN+|G!GYxj`#?`ySH)$-O<k)u$)I)(-W+U`Kv
zo!TjTxTQ&ZFEt4IEDh#}N<SzsI_XbmHDK2)8b><58&X>A*QOb|Vh*tWrb%=!^(|r{
zadw<r%sqaKW=L^nx~6?3vhZ-o+KtO9NOQvCZ|;KB`|wVmy*;gsU52`0p)UVmozdf<
zY;xFAZovp&*=zg~)znnH<zQm9Mz&po_sN`q7Z}Ujg}RwMqsx7q6`OT4PYZ!Y2wQCE
zUk6^4gwD<1M!#$lRL;fxp~K^Ko!gsN?%&E3@&Q4Ykw~!0gr%(5Vw4yeLF*2vnLcWC
zixxa1!8#iud_Gcw^6nsXqQ@$9RAHQN!<V5)xC+Z7B`a@B&brNh2*q;G6<vTxaUZ&~
zKyqDF(Ys{|f<lESCl6V_KS*a2V;@K|)cXS;GCQ>;bivyMBSy<n!?s!$uj3&?%}{~B
zPd0a4huDkT$B;p2lA}u38?iu)<;FDZsHg0etNz?)%0{odpvpS&5gp$V{S^}>vb9`!
z6`l~?nR48Fa{$E*?u6YvpM%Z(A9Sa{%U%%4pMtxiN%^98gh(KCzW%zjn3ga1i6r+h
zH2oHOY~9U8ZhON3-yyR7)zN&QS=xC1YNOfQArqjyp;K**ToL5Azg7P#!OWDDB)Tik
zWS00!=j(C;kpj9j;V*7$a|c|0ktGd#T_G&Z59crHtz5UW6vMH#X0bq|Eeirbpj;03
zyc8fCpRO3E*9H6I@BSgZN&+G$tBFS^Nql9@WGA#9yxyS}@){E=jyf0yT$batPRE}g
zCS~)FovzE;N>ur3N2>WD4(yadUw;{=<rqPwRxAuXaAG)-Bx{eZyDEyxn13maj0IJ7
z9=n@mtMv+of?*95*t}#nBkwy3*rg1D71%`x78a{m_M5xRiGue(Ufo<4Zx$aLaho}r
z4WWIp>!7K~uTWVUD|A(7UhLN6sLskU*7`jg74sx+ud273TVqla>Q3uIz5PeI#Ldm$
zk}4N(iQ_72-O3|)tb3``S#V16R`!exGC5LXIP*PC%loV|Yfqsy;7LC4(PUSJF1^7z
z_MFEqHxu(*;>4)G%(~2m*0E_Vh+FIdApa{X{`J8cl9F2NVBsp$Ek$@;9g9MG;cFoc
z<*NFe&ElE&Q3#!>iT}NmLE`1!C`1h(!R|px_AZpV=*;39{|BdY2^xi9J`gc3v&!4m
z@9)v>rm?+B>2D{lOqVbiRui3!uyqd2YjIH$rR853O)g|$nZ-iGl4z%B#f+JJyvFF4
zOJ4E_f^6ug#dfH^y`_3;EpdJ2NH~MPfEF8yy%MGc5P}yYK2MGrO&%*sp!T=<IICqZ
zS+SaDv$x0o@FyWei6_kTrC_c-jagr^Kv>Yzv4U6W3I!4-Z{_vX{ye<R;v#9o?inu>
zr&;*C`Iqu<ek6Cuz6a&tY1djZUM(A3I5V}&Mv(4vYB<^*W^8J=#U6~)51*@|Ze%dG
zWTCI#9yw${>yj&z%@C#G@2%Icf|nM#vTq@?YN+28l=P}7A~mLWxV@iS9$Sj26s|HO
zz_-Bc4PEDA*G`-O_}JALE27YX!&TU*7;IutRO6aWb>Y}ri?U;<(Rh_Q=}{JKR%F{=
zVg`a_Q)A3^oUtUA+<T0kzb{IlrJZD^WsW(df?@SuW_Bf9XFln$owO}UPN_V34p);R
zeZrMKNBop{Rt$*8ftOgxlJ=f>l+95aRxk~ZsTsfGc2F<iO)BP_-GZ6y&LlFBK$#SO
z=<f9!suqUUe<()h1sFkPy-3{tym={~cJ$b|b?u#?n4gj4p_mQM4}<4p`Hy=K62`mT
z>p(N2_7i-<&4o$2ClSe=^+r+3bX)ZbKLh?=$I+O#QMJeGq-G~3y=8#fK}8s#x{u=|
zwz+ZX=yADGA^;wQE*`I1^n()WQ4mYcQeI}Vk7{=>o4r<#At|v(rHx(N!&IbGB;~a4
z6Vp}RP#)m$THa5xXUzzH!h)S5=A^6MvR^&QS*<t3_RuVa>(VE@X0`7!Xy3DhHT9B2
zwD$}{GD_Zdb1cp)3U<)`4zzNBeyt6YqTe=Hs^<!pa(qj{H16kr7xwPJfzz&;Mrmh0
zQOyElc%6N>833C;OEuaE`A|I-Yd#x6y(3&5WizL!!y9u#IjinrzPqvDm_;(e`vu#5
zqN#jE>a9?RV6fuupjlO1f+0({+%hEYP)^H4UG$Mf%|7Cn(~f=VpJ_4gENd6?;O}YQ
z@}1vHklm~r2E6mUD%SjJ>$e;unVz04UC}Eb;Zv=qt+3<V`Je*BocMxgV{IL0#s1@D
zIg@AP-M+{W8^4^qE%+-3!$|+jQe)E14HBa48#)2}oXQQy&vMr0uWByg3)r$g?*@!;
zyV*C?O|Q`I)Lvq6k0LBzo=D>t<i%1&4_~s!gA8QptKTKH5c!8~R>ClICvN|t+!tPz
z(|5~Cits%aKvsDHxk}GZVy(XG?3cyhH{g3Nze)8JWhXQ2k@s~kjO{!LwCA70<~&Tg
zi>L=LaG9pu$>RYSlU}q7d*^@*8Bi^X#4CIKH?w<XueM{j$?9UnNQajwqaBEF3=zrR
z0HW=<bW1MXkef!Njyn5#vvu{6hi5q?rqN2|T5<db<xnEx`6Hj_AC%@v_E5NlXwTl`
zfn-|+%*ahxSL^W2H%^3oU+(YJ9)Z=mB*n8AH^Emlze@(BNG@6aqdqBmGPrC;KhcV5
zov2n0aTYR?_j`nq{TvbUV~Mqjl$W<GK8?we5fC1XeFsDX`KydLi<+WvXmkf_)tRx3
zoyGN91H{Rq{A*)3((x*E?k~GuCR_^ENG=XSNcjpQb^2_Ujcrr|zc6W5x$=+T9269`
z_lgj7vvml0A85%a&3<2y^TbOmlZ<JWm*i6{^7&=$Nm*Fl7^6a!zzwK0G4ZQ*etE8G
zWHQX-#C=hnFRccZte42v#W9c6+K+<HQ_OJMeXD!D%dY2DKWv#hEp;)`lth*Ad<Q)z
z5Nr|BI>_%0?H(@cmX}Dg$;twQW9;W>f^8iPPD7zm(Umys8L>L%_f$VK&w*TlM=X~d
zg)x;JBibOz$$a+rLqhx8@we^4C<xcujc*jG$qMXu8;K}O0DOvy(OZaJWQZL1_Da8?
zNN_rTEv~aY*bv7&{mXXfI2qPBf$a0_=xubWNmRZAzl+3Ti{{|pI4?^FI?nGr5^?W#
z;|qT8TPdHFqbdLT(HD|7IK$g)A>*)jXue0|l4k|pOE3F+ivR)?NuTbI0Oq+weUliV
zo++)M3qn#X|8vs2@F17v>`1Rv6UU5a(@b^OAnIYzbH*d2QnFV6YpH5|pY|G?t>_|C
zXq{qkmE-%@uUbY5x>((?{d5ODfwhJ+ayW=}f`6k>+iYobeXU`=sKe{B+}(U?5ll^=
zQJVdm=n<;ls)jtDWN#a++hvOF;6qU9YKs{>$NZa(j}OTs+Tql_`ddP9gA;ER27A8k
z?XCSM3pCzn*}XY~2==+#TSF4C+$V~Ya`z5bX^Q^Pnr5jbN)FaeAu3K#)V8CX?f^6t
zTqJ1L^s<f`$VyyM;_Of4^-&4IA#}!lufp;vnUjy?H#sOw{}3xkB>3`?mKiaJZO-+C
zy8INM-BR0yDp)R<#S4;Yg=6X{piY+epVp}7*-IA%6%~;&oGnTEt>hQ-ye)ave_ow9
zHrxLxW-dlfyTa<-&T$Q}Ttglua>{e#1E>YQAp^EK=1&=2UdRvT3A^9ZnrY=;AH`^o
zP8YSjNfT8R-yP`s<W6V>-@6chFf_Q#ZVj*iR=-Gb-6=g_9LVCmBGPaMFgB(WE+J;Z
zABia$9#B0nAA};-ev~6dRaunf6^#!uM0F^JrldZchR?ROnur~#XZkzpixHb=s^vEm
zvi}%na%TTzuOVH-IrM1Lzf2Qe24Mksxvwrr{wrS^vywlH-@uquu>h)<tle}xRg`(4
zeKydq+R%;ZETg|gUpd$(mD5TTvGyOaQDf`VZUcxfisrc&m2RUHG3S;gRys<w__iAP
zMhH5~4=M{Ud2I*P<e?skJ;p4A!GvUXTv(h^aww#E3h%@%-5ma*9BLJt!12YZ3o?XM
z!$wHvY~ajY`VF6}9fEnhTB>C1m3{W6-NF(-VRiZWoul2Uo?veubkO1Z+?r$@fAKjP
zT>E=oIuDj$f*#fOyIiBpaKU%qHVJc<Ex;E>&bcSug8renqk5V~|3e9A2X~3TgwERo
z_z(Kfb0)7dr5D-(t22Ba|4{a@)$uZ4#9@@!XeD>Hese&DM#!0V3zCuNh!7bUKV+OI
zR?*3<1yNQ<kZiDrd+!OV+H!ehL<#bT`VcL_?AD-Dp%}FnxWgEem}T5aT^FAQTPCxF
z1BP(Qqn<eM0H+p!$y=}2c7@6S5vTc&-TD;rGNf77jwxbip~}(Ys7u8|6v6dv+}Y&Z
zQb9JnpP57^=g|C=0t=qGDz>Ef>R!4hE0w?6R<gax`Hnqlli5aR@hStQ9ZJgNTb-3M
zK0><Nr;{?ez~wg<W=;9q#0SXN_^I^rGc0)ff)A&TSZcrYrAog44|oCCAs7A$d-mAI
z#pzMb{4PowS@IcI{peQR*hP{swHU5#-N7lP;e6oBOy;byCH)2^1Hqliw^iRr7`mH{
z^^<=h(fA9;9DO<wT~N@r9Q~q#-&6xq<)sQmZ`ZKrb$U@`=(D!A{B3=9`((@)EUIwM
z9+wE|>=WG~TI=K=ZA2YfU=dGsI!0#!HV^Tv+_4oWd<hS<#I@@DoyzIFm(T7h?=CdO
zG%{4jmELQg&OW!c&7&*ppill>Wg@bxRgi2I)z5dQ9lSj18gKl|SA-AaKoxs%s{-Le
zHh-@%&w+jzmRBu78@d|JcT>DoFZ2CERhs!tZuLR_8vCshL(zw!&SX@P{Ko`Z!HPms
zE*IUA1LYpm;HKcODRP<OpAoe-B45BvJ9Ro6s8y-GzpuzN(1!FzvqAIE8Ap08#GI68
z`_JN;J;I(&PD>pHXTAu95}$@&aUD&Lh+^^P0;m&y*h{}Hoz#W6`^gp%`c9`S2WXYn
zr>-Ea%^o(;SJ{f#3PG<wvj4;8p`N7dBqub&`9QEw)>r+DphbCNLJ&x5Xe?+i-TKm4
z^V|4?;%`CB(!jv1>QO$s(J9d1gIfwk6k8G#4?|A$si`{3^R29Dn;L_csb%`tB-bs)
z;;(Gq>U6FCLwTXSTT}e}=5=+m{i8%3<fVB@sawGEV(D&7?Ldz>5w%k)XT2@HTg{AU
z#Wu4cO}E>K<DI(XhFT;p?!xX{QdY0Jasl`YPClC-T)jsu#a`?-8~tHlZ_b_i;qLc<
zL)h+CtA(7KNin3>)<zyv^M6;#oRb&!Bia?l|6yGf*WiQt&g!0^t%yCsR8z27_7x{U
zmw$`;Rxga^=E*`(nLc7|g6w#S{tHGPKwRNp!Fe08eMFHBE3$pqoHjWhszLzZvXdz3
zq^~)7dMg%jKi6NT*{%M}L$7IF<Csq9V|$ZVE~YB);Maf|9ws#xT`i<`pd2_5xh%3!
zYxB3b;C;Q-Tew<V50k}Ynqd9U#LYmK;So_a{n!1NF#^V`p)RHD9%u^f0u>!m>Q28q
z8>$O=zD)LeQsRb=+_U^~66kwk4!_EX!u@5d92`3bb7Ny7HK`@h&AZ5-!@%JFn9T>H
z@@8C6wNF0IW)GD(b|)eg^fiCu?;y{wj9I=@o1^$OWSB26^jVsg$h&TNjBfy5L!tvI
zM7$5}g4z$P8+>DoBvhcz^u-VUPMX_?ENF=i9~)V;+b%LoB(<AmA8$w@1a!^EM^+vg
zCG96?^~t+^=ANen8(ozeE3)+GRpe!0&vyX`h}@se80$QbKZ}8gTE#z<Q8~z*QCttf
zh)5A0)l|^9ZmsTd(TNs5ZZraVkCh8*Ob&%MD-Htdj6Hu0-PtmUIx5{z=3T!#^>`$m
zqw4I=aqnXhD^PU@SE_rsqm<-cn1x)2@xMZRQ~+d}IL%pJ=|^*GQ4TTFn<^{f$KP(C
zcn!X=*vnDdev6%!L^NB&=9#1wLP5i{L!ss<ZqM73QsLN3INc#hbcVmg^6|*UIMy-t
zYQ5<JX9tZdby~onQ1(#pg^k*^`^zn((#*rh?xtc$w+9#S5*^5km#R6x@6``~Vmc~l
z4<6Pgnp~`m!(`6*<2}7zi0Cx?eTyg5=ySlexWnx*hbQNGZt{AQWJr-dc*o4ig<)SE
zWcDXkXSFcd+cH2VPI57?#d2YQD0A*_m`Api#{Q4rFX~HXO>v$Lc?x8W$Zz>`Q|B8I
zDviI{XCTQxB1c#L$^7MYbc#!Phlj_cv@=Hmc~Oi_algsNrYlhrT@F>+N70`9?OCBl
zfD$mFal$C1frslMe{ky=iSr^8Jdfn(+3(%OnUQJz#!E9FGPRQZ;1)Z|1C_zDyg2N%
z$of5J{tA%ET@2SAp#FX&H)*42!=~Ys+xiQ1{tsnS9qt^JF=XNvqK^0$Ko@u1=+Et@
z$uJ$_J;V&d2IjIF>(DPE8x|B~d5-HZ@uYE>50ckocl?KP0}RfzNJaX(H0x$E>#EH=
zjLS>KZJggZ3cUKzTU{<K4P-13cwcd!Ctj`TI_%ReI#GGaJWUxX^8B^|{R%=fOIFM6
z{#vLQvg91H&*zH5;!{*vlRGbbm>DsI0RRg(FQ%==l^9g`c9BG5iu!c@T;a!^ZCeA?
zls|G`SRyJu&SppXzR-FXv6JYTys;-6ReKsQ<Ne~8{ipOtZ+i_SNdhD&RsN^Q?W3H&
z5$-l<16ZLcexNjwhv6_M56LXnc`+!8zwgRPM~(Ew)EPzlJH%8WfyGO`seq9_|K}i;
za#o79uWBcm?L<en$>h^UQ1{DdMhw8JdwWjK!rUeZX(``_VfI_aRn6BL!#OPnSh(^G
zo<T6Itt%<58MOq6;=3V@z}(~6Qj1-ezq{I}0HI^h70_|a|Dn+6V*JK^lzk73N3h9B
znOh?SkWSVjs{x_~<K{zhjavn%IURPENt5#Q^wZS=>kP+ykCs6BO}&1Os_6S6Axm^~
zUWK0EaAs*08=t+xFVy3u*s1ks6y3PbYh}r1&%TZ>LU3c=w@hvNrB=_u?A9$Fz7D&$
zU2Z49m#ZU2zts3Hf7$zcL5%2b0Fy}9EH8QM%zOoLkq!wlece`)4Nj2a4uxm)UgpB9
zX+A84;!Ub0<KIOvd%!I+>0|(9dZy5x;^@ZP>}M@umu);_E#c_L7Oo1Y0Xa)jXwL7r
zP{yJ3dn(Txl#g0pKdp*8-Y2@dUX*1ejtji$r!Rrv?Wy#BE~ZG*RJo>53y56Vdj@M_
zB9*LkGciJR#M)LF=h0KTbLUpR^?KIj(q(QZuND)qSjb_r$UrGw>{)4*QByPkw;zM?
zR}E%Iy0XrmH288Mw({hHdJfc8--jIMkCc_rX;lYdyw7x1eHa?~&Ogq;)BpJXP+#pt
z&NvKmLQ;Qubf+L=!6G}PD?Dc+)EbMzVE6JsPY%Bckai9wqQ%oNRBs47Rc=ZcH9t4p
zL>r4ktq@cyKrb$<6^35uR3QnV2P=1^!Mhru%8DC`$@X=e&`2zg63zC}tLEBaG*Ewy
zljhVX_2)<Sp`N3)blPNE0pT_~_{`5~9aFEAk$8K!kJX5CZin23ck;cK<{Uqe4dC7>
z*VC(xwA)(*NK9t#&)sNNhXdly8nkI48JLUl<OPe!JDha{7+PnQaqn!>B65&jNuYy?
zbFStjVt%e->d5(-K0}JlwAgyarC;Z0qkL_T7DJlU@REqfp<xJjUOIMzZ(kdexU8$V
zP?)!}d}YI3p14Evt3ovCi8!S0#$!<lt&_pNoJ6ndEZ5Wl7R~;PcY!i`6$^-bA~i{M
zA<3yw8c9mEnRCK~u|P}xUZqjjqHXOzmhVehyqa9k&t{e@A3xwB7n%)mmuarJ!}NcR
z$2=Wd56`R3|5E2H?R}1q*N162kaROvj3fEwb*@F6$7Fi!UpUpX0V;H%zuTn;HeLd&
z1KwPp7nY<$XK0=0JwecG!z#h~(hD=37akbxL&Qf~Rf*Z@?w>9Hf|tQhRui1;UX~>(
zuAPpu*#Hf>4onn~dItEGe|UjbgRTlth~q+jJMi_`o@Jvoq2<12woD_0qHzpG;7HO7
zK{lS4J*GC`e2=0&4i%f-xQ_Gi0EgR%pml6{UEB_bMY;3+#10!Dck#<d3{aSRb1O@l
z$9rY2qv!!93EYoum{&;}{;?q_u#cZOq#H9-DZ%`C6-i4J1It9(bWt9{84fldlqYQL
zUf8%H)G8~3$=!|TA@V}}K%{}Df{4JH;^Lse0d5LFhAjDaI_DLvNPC_x<afV)1Tapo
zSR?+SfYFe5dARr%;EPl6Ng`{=W`wKk<mHpZrX*oKTI`(%rb}6|MXWj<Pn->Su3%f8
zn1+*8P&}v4l8L2Th6d&HgE|f`g({~MT`eI&)Si~d%cT;njY=h!LyOy?k@<8)&EgDq
zP1It%7mc?)1TIF3#jd33POBMx)NK)&!Fx%Bl`T!B?tikJ%O2`D9U(5Q5}G>oMu&1}
zIzA9P+j&QDw8s@KLiXkCYSf~K5{7Vew*X%OaUKmxHoAY{qX(j(i1I)T?&(hm$X%in
zD!gOhV#5yvVxBAC(BVD@Cy9j+E>5Z6oVrwTY(MvhB@V?u)a_fD->%})O;rp7FJrit
z(jw!brJI}0j8)GQH_?ihB<PH}5*RXj{-KC+gH7T-t$)2Z<t@mfWadmCFW2eS9#!o*
z(N~V`iNp4};VT5Y{k5iZETY|rUVa4%b8Yz12^AuQF`X5PciU`p*GO`^IM}O@0HAp#
z9uka-(vS|vs@TXckXNBT+ywkIr-_|sDoqnaZEYvYcpM*F6GiF$v_5m(nMm!EHDpyp
z@msF1m?#0YY00uG=g`=8NuD$G!KJ-xPj&iXaDN@1v5}S4&Ynb|Gxsr_3l(n7`Qb4g
z&GVFMm(8s(cap$|ZVxBohTysaBR$KA6o%T`H+<IxR(GNb{ho=V#)b{p>_N4{-Gy<T
z5;V6ayS>`>Y2OfnYEipp;$D!!A~%Pa)E>b1uyWxmBU*M>E3o13b?~mxiQUi_^9{?-
z8tiJEEdMw}fQ~^@nkCi92_Q$6(u~5)hh`VJ`I9G`yLpxUMP38K3<*pQI4{#>8*SY{
zo8fDXAg#8ufMk=JpO!vUjD9hcJznZUs^*nM$Gy3gLuQWFcBLCB89ptoG1_M=OOrKU
zWR!Rg7s2gKeWKr3h%h66MMreN+QkY|>qb155S_02ry+gRJ@H}XNbV6##h?|Xs_d9v
zFYYouos|Y*HUe(#PX4abDM}+d6e7K0!C0CFnRy-Cz^j3&mN`4uEFAm!%16h7rxv$&
zfgcp`r;@2-RE=^bZDMo1bD9Qh$ttYaHt2(8k{EwR&UJad39%R4?%&JRF{Z3${F8{c
zO>w48`$eMCwzXVp1dW7i)=7or&QY&~iLv{T!aZeR%mNRwig8*LVez~9jgRVRK{w4G
zsf{L8;)s$4BM>k%Gi7~GaDj>He+otlK4})akh?tOqKztP!%9ceOU)`Q9Yn{%)NEp2
z*051IN^Mth8hh_`?hxX;_OJK+?O$N?`Vn#LRtgdIPFyN7LmMP5;(!k>pEb7**gB?e
zflKT!Q#S(J8M!be{QMaSeZR?0JQHW3rr@8*N6R}kxrwv?kVHj$7wtJ#vIpP_&M0vF
zUJC?(6x<7U&8h~31%Q+$5NxQt!whLFZC_spugxX<aHX!c2AH`81BCjuJI{^W7lmU8
z$J2c!NjII}mnSyz7qmK!o;LZol##<m*Z>LLncvz-Iy~46_&`y4l?nGIM?+3qStrn*
zpDes~K<NeJLWyHy*1F<li9uaz*9WWe`x#}kPMOQXq6&7CbjFV09n+t|65KS*QqY*c
zcXCR2wn{a==5~ub%L;6g3YIJs(r@@m#$DJwBlx-`Aa#UT8YcQRZN8fc7#Aflit-2U
zN~aFcmM+&!!|=C1PwRXPxi>`?be4Qsxwv1PttN_SM258f#9<kM73DCsi8&=K;sY%H
z<^w^BhIl_`()1TY41kU#)MZ}`Yb+7Pf}0@mJ6M)?>ir`u#`wU9tfc2*WxNXB?XWmt
zQ~gmg@sFG(eWwGiq86FsVPojSv}d%Zdf;}Lpnw_Z;{tBKaBdY!3UdE<RE<0L0%b%8
zI0*jsjFt3)#-o-SqF{dgtQ>_YB6?mbM9P#6#Qu$X7ebUyi{1H6=l+_IEqK(WkNn9v
zNiuUnhF1ThBBJD7zfQi!V5yJMkFN+UuSUlSv_#|V&9e8eQYDnPI%%F8IXXWU^;HkJ
zvvAk@h0A=lWExoH>5%r+ep@x?B0~;OAKM_|(eC0^{AB}~uCq8C?mt+T-7yGj4+uLu
zQVya;GIV$bk6z3sgf(=lQx&OI?M1%cO>^%)j?b`>ncd^ozGHbGmes9UF=SlR?3S15
zqne#P@t$4o)l;+6l8X<(hwbY1tQ<*1H5SwBPqs9-vH6(KKePggI?pX++Y}3^aGw0t
zcAY~u^_TY!%p{bB^HqDonUU27%i)NW!m#ox$pn6vR2h1iyR>2d!Gg5Y@kek)Za=Z2
z-_vn5`(kYcPw3S2IG<whg&#mtp7!bS()M9@MpAjAO>=@!j^5RFIJ}W!D;~);B2CMc
z<|u_tr#K2w)YmGD!YN(NfEWItv{j;<Z5bjw`aF;Z^es5;S%(l)E2p7?OX5y7iPuz&
zXL(y0t3Ey}ImvYFEALW`-He_&t}9|5Yxrv;UT}Xs<SleZ^9;MIm%d;5&j)aOR(hF*
zl{?2`UzjT!QuQDY$*>a8WMESohe4BEzUiZx+b4L^;h~ba-D+!^9zKi}I3}qKRMn;-
z@j=9Mw>Bdo+NPMQxM(xyqy?lUrHfBl`V(HC^1h2fGd5L^B|(!&=c0V=kR`(j2}o>z
z*Nd)U-o>RKkM4P+Lyu;0w-eg@Gr#E~KhwQL#J#zuJb1sjsw~c#dD>jiJpCPEqwXp^
zwYy9g-Idp86HGDO2Q36EF?f<bWhIJIj)Y}YeHpPee=fpv>dW_J`KkLR(_5F|xs^`D
z)jyOg#ZgeF!^D8Eyd}=sO|JzR?N{DBTv1U;$2Ce}85zmCgl`ERxX`@y2iaeB3f>2_
zPpf?5Z5V=N5%MsyS&YjYqWU$dUk!C5r;G-21S}y`989*EDCacD>Y?lG8$usEzAgrx
z4B?oMRl;wX<~CLTGL-|Ol+i-OXhP8MZUv-T2D6uRMf|7y%FhVD8*y_K`f_%g<52`f
zrB3P7;y=sfV(JTy2FoTIQE>mNRD9(h4|gCZgtd!<9v^C-u2dWwTwQT*rLz8^q`3fE
zC<U-J8U)gN0ePQi?@7N1{@+#-@;(cR&-)K$XRh_6kQHp)@qb%Oz}Eloh|KkCNFA(Q
z5@?s_=}PjE)*fI8X@sq)=m^Edr=3OsWv0SFvcX#^4NpQ8WNQ1cAA*=mn1H@!#;2h<
z%g1g5cZheVgUm@Esy&1Z6xwBFz5d|p)p5P+r|9oG^6bL1WbZ71p1E5=KH^vs>OOm<
zzsSCN*%3or)odyRR{Xh_ZidHem{F0q>m_%Fx}sn8ljQR59lLa0Y&gXk;+xudhka8N
z-^$T1fZPyBO?fK^)cGMbfJ;^EwH(y0Is59eFFB_Zg&Zw_gcyR)cLDjB*)k~HfJ{YI
zSU1$<b^pf3StvulTvGT;`W?`kdMT26VexWeQxCiY1ZFVT3N+=_E4Uf?SOyEclnKKc
z4@<$TUw}Qo0t9ys|GRM&$9O<YhHU=P3cT^3<tlS;i!E*>AeD{pbcnD+C+l^_6fJBl
zfMhNV<MCGd9|{cPR)yns{S}UQs2>s?+#YFgsBb^Ck>s%8|I2UY#_y4FZ)2ALsYCk@
zrPc3|Sz}5nF3`51>{C~gCd*@1_E0L{h2*U&>{j!7<q|6niA0>{7JV3%!FdSZ2jLZ;
zq=%8q#B<q@LK4C4<z{Cel?IP%iZkFMi@Z5U&du=OlCgrhsc9QUp5&m)r=Jpr4jNM6
z_5^zPI|s<DG3b;XNz|j1l$^xBrYWw~2|G;y5U;`jyx$vfTA<v+#tI%a249cD{0&b}
z>I*w_MiW}7wc``Om7AN+7Z|sSwr8lz6lRd{e<%y0NFuPBclX)12V&)Q??13XR^@*v
z*1)AdKtey)?5pe%un9-uo)!jBb}esCb{2HRc14|)4rb^dYW6(UfYxOHq4<xO{X-D}
zY<F@BJc|Ma2hRxkhM(2Hb8b5<sC%vjJCF-N14n0ulmAc#>1-eF!R?i9ruQkYN#evX
zG|rklaz^70ZGwyZl~$*4Z>5cZ-T<r$T8gG}G?jmb)$kTVQf<UQol-03$168e78c8&
zgS);Y@xZbWr=`D*2z3ulq4b>#&i_#2t@4obfCVJG1PuP6u+_s(fENi9@-|I`LYP?_
z_s$<suYyV5Dz~m#0uZiIrvF_^+;rS`+*|Fpt(TZd|2-xE#^_&R3EWUZ>;F#I*VOfe
z0Q>Wh2FkFNF(hH&dUXNtUn(m%moR^jmaea*mr4!z;&nW1K@mwr+0(^v{F&Ke{0LZz
z2}$0nsFdcBYXnKG*`Iv|3W6x<6am+YIFav(F>u|a-dAGYDi6bN8@LlCBA&(V(k4-g
z86_F)Mu|`dVi$V|gn&t7$m+WTE%npdkKmE;-R<$td2sN?__3C0wL2Q?A_i*aq@ZkE
z?H?VMMry}4mSp9v@%Gkc2RZsKg}s>p`wv<D>7Zy{B4w4CNaVfD@1sHJ6JhdthBDNv
zyJgD3VYb<1z0uN>_NOIrOjH&I+Pz9#DqD4Z4Lgfp5rx-6&#ii9#Vd(vcB&SRN)@yR
zzH<y<oy^3HAl4ds{eG4|lK%v>CUZb_b@;BPh2`h+2QlL#_(PW_gnXe)`xWj}6=+Q_
z)KRMY^VaLPG-pql&J+LlUdp%P9nK&PfSeRyQ6gLp$e};nsFQ5dV>o{%qHVN5q*2B!
zy>R+ISQHP!5b3$ylo+{_6z4F(+&p~4Qg!ZaF}C!(@9tIaRq;GoOyE<q3(@(A@D*Mi
z#w}z=TB3_@(KZ>}C}o0C4G2t=x6UVjK=FeIgkbK77u?ne9I?%bn~xo%;A;zFwgv3;
z$Izl|NeNl}d0$*?ZJA0RqFx0js&*T+PL9J6RKra=5}UMFQzIU_&KjbMCbIuJdnErL
zf;>6+_SYV3<DqJW*F%ArP!2xX;)XnW`#R+M$J4~!_)|8_3F)N4s!=|P?DEDQ?tagM
zCBoZ#SyFUDBIa43rCXikUpuVswsV;)rtV5T^QcT!Nx@K4LL4i-^<x-X8n}*o=-DdJ
zZ$TK!5zuWRbgVQknBCA5M4;46tf*b}m2Q#Kv~puoQ52h$@20@Mk(FahjmrswFP^=`
zHDoo8nq)_US&@I+AaIbuR_*-E{^#msf69homaygp7KG?)+WG!}Lu^4{r~&AF`@gTn
z!o?3MAWhvMj4<FQ0QPV*3-lfPvrgWgS!19#_S`DnR~`cg1fRRLFoL&+dLFUVbhpcE
zInePqHYTUK{X&}i+<r}WqA8&G0no}4tB&>FOp)BNVekUsj`%>n<eRuJ1{=JRkM-cE
zYjArUl0XW!(+3P16-c7EDSZcIRU^hD9Z=6P0)UWLj2;=uyY15M905r%pqHt5cO1x6
z5Zu<#{X>ZZCbO0<^|;Q8UiNEK&?(bfq8r>2)~+ByZNi!ZG%Y@bIxufBsHUNwy|Qq}
zSI+5698)D*ij8$3<u@Q%Lz+K-uC~8(QFv`0r}{@5V4eB1EG|%B7Z%AmT`efOZakgo
zk3W}rJ{uJ~3_Z$7vaTBZnr1M})V}`==8JU3_a8D+kB)Ix1)UEfoz9BO9?xut|6JqN
z*G{V&z-r=<ZotI;Epa|8;9MJVoqvs6=WXJ`0&9nW9`9S9E*W8K^uQFakGv;*3OP1A
ziu(>sQmNp(E7-%XA`tiZ3cL94$&6;rwygc%&dmSqJVxApqX7m+K;f{(pQ`Hy*9x%K
z2Ry)9MS!)w{J-~N3M!KTw)*e*E`~D|F{G|O@`2Ekcv0xj;itzB_OSn*Xo$F#0mh@W
z|C#jvntcsig|Ys{|E6HT@#t;c8T;=cNmh7wEU>nNh`NB8rVN;Af-TNuCIA0hWho2J
ztZmQ?D=ODHYf{yf)Q4Xm?P$A*EQOdBC|9tS^qVOv!KTuNQyVhN9{+$-ghf5SbZVUg
z?S0hvhtgI!Da;6`S9@Sirg&PDnG@c)4)o4>1VlcocCF?8IIApwuA7iPG4<Wv8N;?;
z*dy+bTi3K%@J47Co}C@HQ8|$#9$)I8=I7m>m~PzgKQCMw7{t9Stfv~xvp&n67D)SS
z{eE4w?}O|HsK2c5?09ljGGo{e*;sa~FzdioWu&}O9h_Vf=+INo^BPI?>G4xD80r7;
zzcw@1!vt);0=yB{*+z}xbt)s`DehE{$}=0?DP-#a5E-{1MH;E1Q>rG$(<8$d$p4|+
zvFBvwkvVb3hei`R12zwlNu;Mw8Taei`sNbtF8&RNi0@dv_U!Qu;`a3aGY6Zg-bPp#
zSme$;d{(i-P435~IW5&tu(wq%%@G$MQA2)=q$0!bI8amX2D)Y37uf2TqB)9Tn!vCg
zzAD?lfwxMZ!rn-cmKcA>v2TL;Qw<F21i-V-$foaUMRWU;<Q+^f?$zARip7vRK-UI3
zvvQ0;W03M6v<<13pxsYtH^48D(2Qa6<h1|w`!qv<90w!p4Qv4LO;-$41#RB|e@p+r
zQxyS6=sW~?OPVB+I%&uU&>G+_y{ZCV{C@`xpS~L$B};vfq_thtYvK<OIbg=<&O0Mp
zexzIC|H{1Biv#<X1Cb#(ysA$E&W!f@hmr{f9|b$<alv`8KCV#joCK#CKg9HFPB+1D
zN}%B(q%`x6QPkk^a#+n|+7-*)bj{>m$K*A8<)mUMJ<tIGG0F#L@Qxqg@E$sUG}L|0
zzmA8@0NgzEy0nYGz^5AXK~aUH+?kfeO$c}{#dkl$x3{zgHzZa;BA?4W2#xdtFk!SU
zImHxn!AN39+Qy3{RB**M=z13BZ>HlVR~;Y8)u;ZSb5Y7?K<-+v*!f{}r1EU>Ety)*
z^UoaKrBCp^0YnT$abmnatrly?;PIf@94=W@Zz(@@RO0-SzP{QbBdkNv0N{a|8oljV
zC%a8M9Q$kXq3R7IIhn`;qU_tN3fx~!_Lgf2ta;XaJE<yj<Uv5?w@@ME_3ibF*c5BE
zz`!e8);tiiyRUU$pAi3e`5{>3_J?iG?avdfri#RU_gbq?U+a-Fs`e^YVdl<x>r1tJ
ztBR#%f)>;0HyG&~<7r>qo$EVNWZ!-?qD}MgfcB>t(x2sihWm9;@l}t0YMEk9qun8S
zW{h^P283X{v^@txNY!N%gnJKnZqkqG^Pp7O`Ie5Hl4p+|&R1PqUC~i}91#)JB=oJ{
zhNrwVpYptJSZoF&e0gx98!+f<Idi5ugGE9^AEsz{=>;G=qS6?t;Kuv$kZ^JVdEqLY
z9!=|wI+i0EzeA4CT?=V!Fn_hFCVlGyEqByzfC3>Lw5NFveakoD^uLqZz`)>}uHn{#
zc8j~C#sBGd)_9p*$3m5Ffo?{H{JMQZ`I3I;tAgdEA87AaQF1DSGp?^j?kn#`JF|>I
zHu`tlu>YsMuMUdq`Su(<1b2508r&@e50Vgcu;4bh3~s?KK!5}%cwq3s-6gmT9tL-}
zVA=V;x3ycl?^pZh?yIf$Pv5#dU3cz1-KS5VKA-bxahF-y50Z7T`NDEM;}5UVKQTYq
z;TAV<Fk(2_>03rufDS2S;+2IEobWl^f2Ox-hJNeo(H$W22ka0Z@HG6mw~ZA~nSq#b
z1^!I`1=zEJ5uRDe2!pAPBDwxIJa?V3J+>3N@mSn-UNRY%XAgML@~6T1?9eUm-{%*B
zq2)99?`6@50so-@pk^T@+Zj^|a~$YIYAL?RL;Vr_k{$5B6-CNx@=$_ro1*VC0IL7t
z*Z-~Y|F^dizfl}uQ{Zy-DW%1n5udIkW@!BI(C!Xr?tImqGq4R7$Y!`;T2#?+b4wJp
z2~?iRGZTO+`r_6jg>S^3QmB}n9r_wIG^n%pcBEA!E6KIdKYPxOfqo2b(&6#s(^S~r
za|oKag%IRAkC-k?*CUPI4ImT&zh}ubO7KuWsSpo)YKajU<CX@18t9RWYdC+Z(>m!)
zD4eL<0JgB!QdNm7!>Gx_qQ$8;rj!aYnx)r$9tp{hW-$tru=;3r?^&z@=f;J}%|O*P
zbvsqA*(p5})`pbJ!{UgV5ILAB>Sv0v!tn)b_TsB^^-Ha_^3=Ma3fQ<}5_KzHg{A-l
zyPKu(wdV3+wDxRxJ&qk{m#{Z0M{!D)qUO$U`cQuU8Ym^V6RVNU6nU$E*-MQAW7;?5
z2lL8JcZ@o0;1LSt6u873qw21icQ;u&JH`bPBk`o1ji`-z>rOjEiJ=bUy86Xf&8#@L
z8tz?^x>B8CI}fCtEJukT*B?LABCWlX7Y-M^^RUM()FM^WRvBg?&V7?yMUYIFxLmmO
z2Gn0{L;b^u1c*~{24b>daB(s}Dwtv!$NLKxaFd7O|29ms@jQFQ&`mK4`A;;dFxNM#
z;75mAbrkK%s{q!CHvzQsCpTyr2E0oGLZgig1V*$Nnq_%-60z@E+7J_W<M}ry>sEsC
z);GuID`%o~kt7H{Y~kBne48_%-Moe3pJPHkr~KKUXAkGCUP=&v53~(^xsL#tQ)B#D
zCTs&5$QU&AR1pLV+G9tV6!eqwY-50H=NsyOE7E(gSdk}kqM9)7b={QlY8sQ5ZG?5O
z8SNW-s9V@yCNKZs+{&ZyT5=8PW;)Az>6;i@dHiTa7wK3?#=2e(99Nt?uO4#W<I5c>
zn)ASWxj$vpO%`3$JWFrdlomPSP4iJY$Vdx4qUh{^hSleU7b+a9*4U_mGaTiVtYj25
z<v#pqk9rmn?hGFj(Vc}l++}L%*={!05{56i4zEa+B5`-nN&N+cL`FGQVK<6-30u8>
zXY0~OeW55fzyT=<D`<wEZ4I1%J=DX)!sC6?okwBH@ts7^yd!$s*4{o{?~Fez46!xB
za>z90Qq`6St!A5qahKW=EbOQ0g?)hL0JGPu>OtS>={IpDw9snc(9G_#s%0a=WbWUK
zyJ0r>ZE<B2?TBzM9*<xwgm8^2%;mm-OY_2F_-j(?;W*M50O*Wl>veMvS)FeLgX$lq
zY$>t$m?S``Sx)=O3OZA5GzIovl$Aw1<pDI=@jddiqpznC@j}kl#1cWkVPYgt4tuUz
zqIQ%45D58on_T&um0u>>PXuapaBmt8x%y){?Jwu*?MW5PkoIeUnPRKl^6Iu9Ah+=w
zuOtGRUsBX-gvTi6U3>g6nq?tWat3==RLgz9F!~GFLk|{PWsr9uRqevJk(E2We0e1k
zyChVw!dg?@lN4no?Csi@hvvdxf-*}9$k|k^o$4SoU1;OWR#w89>rF)~;1?NbKE9~Z
z^xgGLd|_+I^WB+y91;ji7gqD6c3nhI<iaKHI1!`$3YfqLzu0@mpJ|hG20Q8Q>y$(d
zCbGHw@tI(vkMaE7y9n|pu&JU-b7*Z&BK)N^DYGgz5!fy<iFyHM>TS?F-&=}|@y|90
z)_dEqitR=lGm#sl8mW24Qmas@Qe{tMfJ3R1pvj2zDqs+hphSbWDN8hbb}<DNJ%V^f
zeEqFY)h^m5f!&9FWTrXdf-64n2Igl!@+hA5)UQ8;XeceEBY5KocXz!eD439g#mY8;
zer1e=%%fDPxRKi~c<+xKrXCq8)i<a{%Z=sVRmb+88FXfmE|Qobfu*elTBWa$BG=;y
z=PD6V6{rpZ!eLc0lEj&K_vdSFDn?aVsWgiMwY2mcQ-okVYS_!zU{RV<6kxabbAGA)
zy;CfxXdzn)0E7zvT%NhhUWrTmrDM!+Ep*Nbu~>ibd&T4s@k(%?h$M;+@G2$FO&og?
zy`u_LgkXoC0zqjceIi67^1gJ9QrB7hETyh=MI6rin|R^30-VN4lVA~nU)|PI_fys)
ze|Fl&z)C!Im!<9uftWSUKqv6WoJsmJjH`wBIZoa)QKK_;aEzCTo{RK$SDFCdG~S48
ztnP%eB>swzP<#{2j|0%3W1ixbCK+`tv_l>?pG_mY5dLHN2Z<%~d`}dC_re6n>WKI{
z#TM?TITdTMq0d>`bxWpMqXuIXv!k#LytkVKw|<CKUAz&>it3>*y+&CUt4vh5^C`AB
zv`!QnfOzA~S5%rSA3YAu*bV+#Zlb-*!>(J8Y@QAuPxri8jyfGn?PG6aVE*jlK9Z%I
zniwi6<h|i_20zDHY{%O^NtiO^?{@3;dhR2*(?T`03G<gKKM<K*JoFxO{^U5L$z%Xn
z)Q;mg*~y=2qbsn#kDp#Vfc+R4fgKYKbEmN!#&MvFJ?^38%pZHSs?8-Tcc$H}43M`~
z|9J-){}`9+Jl+x%erX%or<>-zq`B<u#I{Ub+j+8K+&sQ^oU|4!#;t`mS4AQV&{#)W
z-Vs{9V^qi8l$+;fnieq~EE?l#Op!j31E8A1sjm-Z$vdz!P3>&z&Yqwt;}3=^PK1mj
z?XR)3EOL5<U}J^*-oJ^txK57D$zECQ7{93@-)L(7I8ID$NwnYH5BPu+u#t#)7Y~K-
zvjg9~7pk9Hs$dny38Mp8TFf_{V(!^(Xc8Ih6B-G%7?J&o$&5Kxbox&F6{)@Sj!ZLV
zy5~|8!H*wtaA#1vvh+s-*Tq0l1O8F3`@FGF4zAhIFi}q)IJJE_BetA-C`+{eOfWSb
z8B}h64;`x)&ImIo4P8*y4}Kw8J!mNS7k~v0+?4Nt@j%Om96_gO^$Of55}|paitvPb
zz%gf!A<1)6nFEpi?XZQNN426w9UgN?gdAd48;|GhF{HUps4AN%Uz7V(q8e4fjg-{u
z&Doj|u!p>U2XgZKP@rc$6I9oK_Z>9+b->TZ=kN~p3bJIUo37lVn9qHL{oOi)?p0Fz
zi{WPvq^U=~6nSTYB|}p?#hZ%_ml_qO=L42*=HnCasnI~o&y4H)pW*La)&Zh)e0jLb
z@J(Ai%D^Irr=%Tw%o0uJboYk!*Cj|~jWCG&Gft+-!@HXnwDzoY!f=~pI3o)&?SO{F
zWdb-4cYIfQhWD9mUsz=RJu`wL8^?!oaJye}o$&Q*emojcA=OnV9?6||Uv#CNOV^U4
zxq}_y?3+(6KJSA0`3^Gx@Om{6>%rMz`Pz!%TfHqXJz?_Vo*WZ$%a~?Y22Nr$x1<bi
zvk&)m-_)yBTM`n}4sXvvdq92}7$_g6-#G2)srIMR<P4~4T}$vIZvBfl_2mbI&0(JM
z*ZEWgfj8B3W`9aJ*Di2c>WaI@Xs|^BJz{;LHf-N1`8k1;mI(%V%Yk-U=}rlfLhj~3
z5<TLe)0>0>9_W^>?uQrs@XzU~QA0ZKuJe!)Tl}QF#d77ghUKOU(4OUq9r^8H6p?q8
znso6F_CdJynLeHlR5V^bP#g{vNd!8t27&SGMVFjJ_dNY92x3#HZ=Ss_+h}XAe|X0?
zXluxn^rh$$c2*IDX<NQQM`U1^ET+5@wD$V29y==#Iau84T(Qx;Ox<ez37P1V?XvC*
zkGUrEymf)(!mVc#=bp>oKz;C_R=P$EOglo*GyE9iM8A=qDJ-F#9eGH;YyoVlJ%4q_
zeXRnujCKV?t;i)^@&hnT1Y)+XqY6x;)A_jUnJ9}VgI(jF_+%+*MW-NE=bBL`r1#?!
zM;{mx=Av{n>0QE>nn6q-C^ZUbP8aE0Y`d4K51U0SLY4eI^KeJCThsOL9C(VVfXjnE
zH~q%aA43@I^yoy9WJeMWVP~`<R)q_zI2x!aI<s6X8KRb2>TL{QbG$cwiCtP4Y5k~;
zB3P&d^~poa?};sIq{4mumxMD5{Q2jv)|1rBmS31$m?7~E{gUy*t3EuHrRc=nn%sqU
z)QT&8FO8SpFyPq<=ynl((cUEGJ1c|3(BuFLbRdUpb*L!-b-pPEx~(>zPY3f!nyoQD
zPsYd)8baE8lpd<9mB_86x@A!Ud@3WR)e`Z*D5|p_oa{V2&Xq>S<2Q$VsMK8kVIi_q
zlY0#}K}5$pKz8n>T@?{$5M$}v?_c8pD*<pL=UB^gBU1}n*6L^mybfNTHDVPh#k~PM
znmSpk9J=}9bMCapr<#(5-XGnT+t^;i<e{4DFYktBJ~&KpLHbvp+8Z{0*y)H^hz<|f
zqBPGx)9m|t=0I2)>O7fS)Y!8$?Y0ZH5(pacHKjPp?tD9!dY@U@9*7s7>JMo3;e0a`
z1IH^Ch!qeX=AFiaT-)4Wln`1(uT740S2Nl;x^$pVVZ)hs<Z^F$rO}~y=(4&eGk%n#
zd=1uQQfu5Ks#hTXluUJs;ViVg+)wPpgdAHp@KaBg)t^6sSuh6ifs~;nl^w;3H`07f
zHeoRUtAF8n{wr_v|5_alU4%2Feu8>Gy)iio1pjWRME|9LcFwuSa3HtDUzxIKZlvFu
z?R(oaCb>j4nBs&Tg=5J;XjKm4DT6NrsZ@*R$x$H^TQ{@Gad7V=%ixWXRL7~hX<4+p
z5$1Hp+4@pL$d5FOD=T?(EK6wbo0eGu#tlGu@Dys^$);Tyr1}hf4@2vcyy~hH?zu3Q
z(Q0fDZy0vw7%Ewzp2YAz$LPD~zDdQ&31R;fAlA|GvJmtoSQv3`nuo)=(Vc`sP4#)b
zxCC%XP*3+)^5c=_1Ba3-z$<?9N#@ODn}~UBuS<<!nJ=noTeu1wG4DueqW?rp%1$Ze
zYevAk9pX^6lwAc^rwi^`V>d-HC*`e|fKW~RJwDqtj!1RSMqukuVQt9Q$B`1YJ*l$$
zp|g@WA#%^rM1soi%r$wf&@>Ed>;jBWC>WxrvV<$(3AN6q^S9-EjR^t55^8|;lax=J
zCGu-oIsDU}nt+4Y(szuoq6Rf{uiUF5T{K|H8cJBMDgX|R<rW_Sdd@%@0+VH{m|n#A
zIQf=HxgHt^MY2vE`7{3*?b9_hsq;ADvCer#sl2&I^0I8eH)THfHJ!p#^1(6d`CG!M
z<G3j`Z2s37h$vwBpJ4ij2bhA;O~{6Upt|9MiLrQqG&#JvF`3Oeh--_O-Xhmu3*D)x
ze7_~4(Qjj~sx5Q2Qw3q{l=v0-E=l*qb${QKL~uO<T{o45!c(&iZ4nTkSD$}a8<}N+
zx|p+ef<>`j8G@YF^u%nMb=g*`A&F7B7cvfX*S`@wvTMoeScYf05ywU6LJirj#RY$c
z&WDVwE(hiYTu%-8?_j;I2b62g^=s<fZC=7>q_&0Mh2dLV7^Au4!Az#qwrGzsl?u)`
zz<9BYaX((ZqM*8@UIaG6^19MpfAW!!yrrZ1+(^4@wA&H1qN*a}k!2R(E=!{n?XLqr
zB~Hm-GyQ3(<V>CBJrW;fjTxa!KCGRn8nbK?-9}ch-x3s`KUEA>`EbBk`+UhgxT?J@
zlMS#!SRw<`M7TP-6#pNVS;2?JoVQ59mXhvCLyS;QQ=&U-;BU-{5+g2rpfR+6mIe4k
z<&Mr&!X&mYi;wjiYmb(-^ZcXlg}t{X*R-J_u;&0pT9)S*hU{RGt9WW<PK#&JaYr!^
zFHbCVtU9Xdi~FF|r7R6zZ?>Q6OiA0!_2OIKfa${T9Lj`>-RUJ&sw-!=i5>&b3$623
zwEiNR0mWsgJ8ux2h??-cmx1`_#4dL-AL?n@)rqv5XHhq7^d!=4Csr}a%5!L7kgF#F
zlX_{vh~HW*EeXokILQR;C_6IV(kqiFRsN>%)_A9jo{jCI`q}!chPK)KNIRde$;`|9
zhaazTl-^XI(cC0GK}!zm8v5#lZ={WqO2t^dm9kw5BAE{I?HN8DI*<mk)vh!o<w5z>
z+P~|e+BhNd9Pb<uFmjfOr?ASPD->DkweK8qKMmD6L2tiPB<7)7x3kKS+Y;XG0%NtB
zi1vKLn18}GB!{{<2L#4n<YDWb8#k2X-$74#P`!P#s-bOf>AveEqBNqA)Knf$9%PVY
zw?`&0)HWunE8A3Crr#zY4Z&R}P#^$lJTh>g=1~C0^N~uiW)U~^nWWR>z^$a@{qdpB
zH688Z>R!4#b19FuUW|i-ammvgY+gqXVkERjX`FdUMStaI)(1gtWggaA2X6t+Gu5(z
zwf0M0ZfOR?!a$BWrq_es0g{>yAW0K5&kyJ?=o4GmQJDHEpNe*4t|kcWT2h?9C^1)O
zFV=sGD;;|1_*M+A(wgztIj1#Xo$C|VPJP=-K&n5<hFEJHAr_7!f?C{v!fjRo>9(16
z6h==+0b*5huypzMdUa0a^N37~!nXya8x&!x3)_VZffl(v!Td*U{x|3QOX}l^KUNL)
zFTxY-xVM;bR+|0{O5N+6l4VN_P!_jCmerimaxy7>+~qLGW@@@{ny%L`6MKT0-GA^#
zKm^<A`ZEGElDmo@+xY*~Df@<Y2kz|wD$QZCQ@hCruO#2wpvcqfZ-Fge@o%0Yqbo^8
z<>A)yQrfn7M|-e8?zTgxkCx`uo<0i*Ad9d|uG_@TDKk$A8uJ1h8(}m0P4gdtz1Z(X
zm(*<hPEeALWpQesN6rSyD(@8k0^ZA5wm0f$T#jtK3dR{iJ;Ni(t}wOp)eB4M^>xZ?
z`C7Lc@N;KF&`P1Jr*s7gvB8H;Srqg5x@kIN9Sb&xI*N;{>SHgElda%ncVRofM%Yk4
z-S*z1Xw8^GJg&C?RchT73P%<@ZMH2+<m@qhO$V5}3f>tn+jmxuh5&e>m}JBM0*Duu
z@7#MQd1K9IJ`9^4&x5d<zmX1+FNY0w`U3jT24e+}dHXvU+8Q&DNXfh|scc7*MYhQA
z0PENHx^y9)K#<75jd)s-3u8FVl7T)^T-k1dZw3H%hTd#ZbV#|Ca!^(xG*?#y9p%KY
z1kyj|W%wGUY{&)4^eJbY)6*L7#pr5>Jws}n2br%JgZbNnO6@2Upr}jQN&>N-!15^b
zi)lN*YLhv=uBGtG)J=8%`}<Eg_529nFXl_(q@`dQP(d$*Cg`I>9Zq|cU23S)B}G=a
zNdZOKBhgR&{h;L_P!s-BiCP+YE#}HrB?g@;AirRr(z?ijY4vda-f?f8LzAf5b5QFe
z?NhmTA7y_;QJJl8Kt$CxYq}GIctuRFJFXkxPqp|<Kx@Ta#nPS0t!7H!w8Je7Zq9=e
zwH`&C<%GiJy01@%2jx^}8o2j-y6$Rdj2iCvfQj{O0XOTKF<5-eyWiK}SMZg0Wl0qD
z#WiYE=-cS@qw{jh${j$~z4jO)q*`!Z2c85+iBe=cqT=oR$#iWLT~73$L|6R2sWAOW
zP+C+mzxlp-E^Y%e&p;{dqa{3qCWun7ii$>}Tl`Cc!~37B_HUe7j&){>Cox~@*=}i}
zb-n~=2f@kb;Le2d`S#x4W&9(IR%o)jNW>vmC3b;ak-dmrVwHJrPh-W64GN}YWrO)$
zI==|h-53DqxuobIhtoK4u>ZJ7UFc`apQP45gf5g!A-4g1N&0~@vmV)MIt}USX`In%
zE8N?^_{fue5S=uOr1?)TrEg08TQ9ZopZwEHuLC$Nr$N`to%jdVXT1*ARpL$W{(NpA
z@nNE14Q4$fg;*4>9wY|2RCC>XOerd}xhHjP)Omo|)Mg=84r9JJ!X7RpY|#tio249h
z(z8g-8{8(nT5Xdaqd5BE)9{PTk4gGg(c_BJj@Yi(<@Z&wl%u_R4Il1)s0ZFZz2*2Q
zny@Ch-EoFmv#y1a+tY0m|7{%|HFuzO3;||VAMDST(T_Pd>|qOTTLz-oQfyLe1alo`
zOdC1eq+i}G(P)ftD_E;(CGQZgW?kgDh?Y|Z#n(ricNBoij4%ZaPM={Q_xk4z1^NgS
zitd3I{Q&LhYc=9?*7Bb~a088yb-TG)#_CqW4xxK8K<u;|IaeAAmok@fkR`)Vt!`=~
z^YWH{&rZ#*i>5?RZ5d#ToRyIl37^2_i7c@X)6TfKcy&@qD6GHnQ4%RB*}>cNd?0-T
z(_gMox_|^)F;r#cN30;(gTdE<u21I+n3#e}k!haIHnikMGtO~Y)=PK{HaPRYih3qx
z!U|iV>uIXOJ^TRn>m$vmQ0Y9nu*`pE*;Cq<r7`mApfYx}U($<H%m#7&uz@Cv>3*ab
z*8FOg(3{bcBU}@DIp7-}@#ALZcepTjeWdw|aVZ%}z(&D2YXNT**LA$&fx{xCSOqC{
zSo8xfXGLpRkH{J)VpOL*6I#YBsf5p_fl&Uy>ypFb+x>QeUwg*3OZ36+cbeTx0}|Zn
z{eap87O&aJhxS8aBD<6J5;xd0O>;ym3Uq_k&^eAUMsSft#{iCpb%Wd+4sLhls&lOM
zW|R6Agi&XBFMfUm$WShvo%W6TwY%Ni0%rvM3|Q)u0;i}6osw#L=)RrVqP$!>(8MZ=
zjDx^yMZ9O!w-)MAdkJwx@G_;qBoK?uCuvhmhniJ8aiR9@UWSA<2_NAV>I>kGWJ0?F
zz<mvM&d~#+{XXN~GlX@<CK?kp^f(TeQ$%6<YE@P2c^N^6s{HNZ@oqKs$w$lej~0e8
zIKw-JW%7~Rs*D0Y!@cbnuRP!j&=1GJKq22H`3?uB)ifsMoi|OveKfxV3`A-ZHvq`|
zO6>s3wSwZi(SAb7`;r^aLefK=`djviPG|LnVbrY*s9!+-GT1`n?B~H@V7xzDQ|nu^
zwEfBOp@}McK=X#q)U^(&cu<^Rs}$q!XQ`uUibyBO#pAC4br)a9n&pG}<e@Gl>b+lU
z!LNt&@N~}UVm)GG5#Ff38f}dY_*>0P8rd)20V;Q-IQQvhq76#~sO*CvU%+$9&52p}
zL`wXs>x{SKB$_eV#~SDvT~z&_N@Qc>dfK;^)EzL?2N%wQv(qSsNceT-ce{<fE6%->
ztz`94wqKDYtJ<;z2!Byy+IXwwJhgRcSR!+o0%SZ5cfU<%qWHA39<vS=<7SyXv8;~k
zx1X*Fr}|k5CPp%h0kG#G|0gQ2$o@SnI*EkH!qEDUg*0Zy+J@LH@>G)x-n+1jMu`2)
zVN3sbnOjm}BX4`!>o;L>uE_!H$npQmv9;YZ+#U$=zxAr=XKKU4Bt%xEHp_Xj6D&~r
zzslg%q2wsmo7$%6A+C^wd8T~EU<$<BATm-T6*)%S$f@ydkt~x{3hh|Dye-nK#hC9?
z)c@B9_)hS@J!h`08IIp04_??$TKoPVV!To|-x<Ll^z^p|uCQetJ%&zYvHuV8S=!4n
zMrJqFlcYGtIznPU+}ROCzhU%CPLkSMfk$8IviDii<~?ZtCyxM2B;xHGaxIZISJ>YW
zn$k!rZ1<(3Q=be;1_z}om4Qt)=dRl6KAz3B&6j*=c96Gm6@IEk%8I17)ItX#B3;)Y
z=?yF&>*{2m6SS6ooGrK4?m6TrNi*#-I(I=;dp9CvDT5#`-4dc$zpqI@{sO=#NwR1Y
z!Je5?QqE3RGjOHP8f7d;4onOb<R9GZJWeA!=w~a;+7hV1OwE4*T0BmKtm#jagRH@v
z`d|UztdQE6kXZs&CNZxI4~u>e`-#dX3HL;Z&@$*luZHa>A0eVxJErd@Vp1;a8^Hoq
z^*CypFpX$eCxx>wicA<{Y{ht+x}Ll*4k`7b7$z&tOdH>QWMqoK@YYZOD0#!PKaBRN
z<*mFa_dh8uZzEx8L7a#AFoEen85FfS$piW9J>61auMcv(VrVfBrYXLYF`R>Pwo<8C
z+0<GCqbYg@DKId&z{k2U_;WLBIc+xG!d99-rKd-WVxoB#Q4_Uy-b)xQ&OBo~P?T`(
z-a5Ky*~%paq;sI6-EV1VIOgYvlf{8OGgL#*M#m>gSC)tKWZFe}6Kmn>=ax3D;Mi?*
z!Sg|$KxPfF^<n=n0MB2(bWzUh%W^l(#C>>ElX8aX?Z>PPKvfBt7dQ*0T9u+tz&s-J
z>}pOme<2W;3@$mKMT2?3>L1>|{^ra~(C<opa84?P<l@4P+yd6r11Y_)e$C~!A2s1&
z-1YJOZ|SFY+pyl`-k*j`hbcjpa;|h42+As)Z$tFB&BwA-(p+ab*%pw!46kVBTGj$*
zDopExqvK%L<$c3wcqiuC;cXyceM5`d9iwOBdxyA+6!u|sNi*CRhq4q6zTS~nxv1Uc
zhm?{_;dA|0Y48tcJ;dy^!M9^>-WN*~e#~jQ5C^Bht?;7Dmw4(66py9I#&pVUFi+9?
zWmLVUZ(l;bv5Qh1*sDv(b<R0e9z!e?lgHUIw2ZJh0*-tqQCmbJgp7UXZ+0~s36(79
zzVmURV>FhnpKDCDy(&MvA#2P4HDySUtiE1B2}J2s1(c)5`?(gZ9!&jsN*kLN)2~J)
z-S2l!jF4urH06;HW3l~Qj=Y-xwTuG8+Fh1hUVb)h+HaAo7zi|YD?*6S1KqXAwU#lu
zcW05&l}`2XG8*j-X`Fg#+;7yfm%l3p+MWmF%?H0)+V@3|gva6ljcaqYa<!A%zA-jk
zymc)D)QF0SNpT{mclJre2lD+~xn^v=(%!EP#4KBWQMo0OYR1U52Mo<$1<;=MXo1t;
zLNkVSx0a2ThD9>#H#xei@%D|f0UnLCNb?}UtB&S$%?OcTjp+ZOR}Rv2*!FdIPR{sB
z&e_tKs7?<N?q9?`2bnSeDR0hlsy5hnGI~()<(_MO3JC@}`V&dqcq8w(TE5<3-bkGx
zm|jzk?Y5(ayAZ*LStO0!9>X>=fo|gWCRDc4Qv!>>>KNH1lS^n?bWbBCIrHpksb{Nf
zW|wxgLI>17L!(tIMAx02c1w_hU+wuyr-0yUxrB$hGbRH(XwdeG4}at`DY(W$Wy^0(
zUe?ZDtG`qyDQu{jFR&B-wY>D(=J#$BB1kFI*RUiNU(js@&sK%)V&N8e3xu%KX!PW{
zN#VH7B9#|{ZFFEDUZ?s8S1#RG9p?JD;^zloNo@~aT>BiRW`_X&g^~B(%13X0;OSa)
z<zY3hVa{{0)<i!K^Dgg_p4F4RVoYZvXVIjwYk`FcslSwhxh=Ftqgs>mOsk;KF3XJp
zj5Q;2nECRPsr0RqL?CUA%E?L3aIv2)rlFo7pi1J$2sLt$rB^)d&N1UkZz)w&-V4pJ
z>)a#Gco4mcBdFNcGFc48EG*PTJkd9T4#!@5VPUP%_ETrpa+h~pI!)BEEF~-m$6v(`
zr!SPsA);Xi!AtT%g$TxhYhU?&J-}3FTKi-4Q9jh0uY6^-(BRXof1egg{cmIkN=Q9S
z_|&|**9G(|dG!aU=uu%#(91F(IJeh;-(K;|)_6|-iGhm_@s;B(dayViuECuQaQ5g5
zb%uAzuk@xN&&jn`y;(God<m~zA%CxhyEv@!AY<%nM&eDt2(T~O<rB6FeO7P)8o*r~
zI-c$6vtyUxz93hR$8v%_DBT)GVz`FMQ)%89g~D7(ASb$&^%C%V&L69Muj{E2B=d`q
zOOgDP$p?k6Ya}e^TB(nga!!&jFipG=Su7i+avqzU?hQNgK*{=;w@OA2R=MVHdk&m7
z*J^9Wo31^d^7LF>Nfj#x$x~-%hDP;Ed7K`}n#2#mTUtmKXQ#kfN9k3NbjBnq?ve8*
zt-h3q`^B>m!tXQI)R<Y;`YOj#-6XtbDZmxnAuu4GfPh(K_zRp%^gwQGNo^8;0j-<m
z1n?}!9VFDxBB14Vx3}|NIhMYz{XJ<&vk$!5)4?3ed)~K(Z<i|gA{rt*LlWSvejWfl
zxk9k2lr)as`*Zc8K9_yl@AbM17ek0c&75{o+5z$#+*B$-IQrm}w-G>n9u^0mW2^LE
z7hBY43lyk&0w5G)qd}**nCaJA#-XKbvpJ!`;yxWzhiNx&Zq(wrfLX^W4{x}{Xk^8<
z9$rRj+yx-=y9cNi;Ekc2VGoJ*E6nP=5BUk|0O8&)-<|vbsTc^PeTC3R9c^sn?RHJ>
zSE7_V9gsy&Mq0Ed8f0`yP4dFWfu?;#wYwJmh4vs@J3v79tfTzLp{wofB$u~4$qSUw
z64$QXLoFj)!}x}_dn&n(C##FL(jmvjJVB(%J#pIaR{0Jsi3kew6RR241Og`>Y-S1J
z7tAOIboOVf^R3vBVxtASn3AzjiCnRTVAlwG3pSJzVx%g+!aecMFem6Loo3UFb#1e<
z2?Y=KD0|>HW^4l8lSi>)uu~6vmA_Q!p<V=m(61@gDS1$nCC{6zFRwLhQ6i`eU7;>T
zO&{Dta#AP3x0?0Sk2*q-I4o37;&%kT$XZ1AlW3z-WN)S<g&ZJvF`TiBP~_j1H4&!l
zjwV=j>q!$(?)mt9hYrftd4DNq*H5k`AQ4?{wO!;m7aAbo`708UztA~-JsB2Bc?zHj
zzi665l_2YA+*g-coaK5T*0X#d_C`&G^dCbY;IY{bus<+CnHS7tYiUc@jN^J^%Dy(Z
zlgh+|nU!HI(M9=O@CoOl-`N=z!Q`ht*Yg(PLU{1)1hX~XIaC?TGf$}%r+&6J`aOVc
z@l*+Zbja$gPS@y69?^?XVG`xW4pop*q^1Nc&Prld+W9*CHesoEpyX_mv6JTroBTSi
zWccaKq$1<Tew|cy+p$w4V8SD*OnzpEjp8N`OT-lSdRZUBVqKHO%Ze{MxmpnEzF;<M
zm>41i5q=xKG~yS{iVtaTUvX1wyII0j0GO>BPFQ(zLuS>9&ze;TA~}@a8z-c|jgfb|
zTZ)U9GK~*(=gnPFRdu+BMSJNsq*vKdh~6|!nkSCT3If|doTvca#_W!0bu6H!7-NAY
zR1!Dhb3S)?q!IlbWqx-D4tdj$Ni)EXf@0EscJ=dSNjHo;ZfZpC`)OQv@a9P>`Qt%l
ztZM;KaID;}kunS)l}|zDtH4T1C<0Qqcb$v5Z@L^v8!<ap++XpdzTp<H{kV4Fz%}b;
zsC+3h{87f$Zi5^oIJ(LY!nSg<$!b|~kHeQ;(tIB`iN=dn5`KNGF{6rby)}Y_V7D;3
zr8wkZ{)zZ3+!BgPQ{(6H=an|fcg{31Bf5a&rekUw>8Bt{z*57VD*{e~<=4an{p60Z
z#1MO|%;GWNMkH3z{3!GjxNLWe0{~x>zH)jWW);${z309ZEX@5(>2RyZ4^&{ya$+>E
z!t96>->e$VI@jtxdoXNnZVIKS-$$!g<En*RSqthiV>h_wh-;y0fyoS7I-BZ4DvBNi
z9hhezE7rsq#ipVY=(W%Q*2A~6v;&u;{COCB$~4MMOsFol$f^IzVfxQeN1FtfXsYn1
z*vLefrS^EV4zq9Bkm18n9GksBwpQ4qGklNo1%lbg>7$&3W2VKs9L?riL8R6SyQR-2
zLiNp=J$!bEqhFS(O<5l&(nZJh6MWA&en*>3Z)%+T-uce<mO{FXm6f-b^u67N_XK7w
z&hxOVqR3j9U)#N2vm-JxX-LIOcWVTah;F{B-@PR6;5r+x*~C%p)jPY#8WUI3FNiA)
z#xgHJ)3@2~-U${S$S+afpu_jc-9I+~d&qg2r#NSgNc)S(ynFcgIyJp72QytcjxF*w
zx2|;9cDsA6(hmyYf-ME~Xl~jqO0WL#nFkS25Az(o`JE(_YSIdAj78)!v&>dXPk@T|
zwT`pDbg+2#WZ`OviRNLqmII|e1YsU7Ipuj0IC#>;&vkzu&=32{qQdOjZ(P?7J!ea_
zp61O`D%}?^7_$%v8ddJy{vq>A3;pf99xVeRF)vC`omN3S&UZFt6x7D4NvlkL-MveK
z(fb?Meu*r$sByObNK-mpLdo$s`3e0629qFCdJtken@F=lF-@A9mNsiyd71uu!6kmc
zW}3w^)IE@r{M;lN@9{4{{x2Xu25{X;vaBjNL4tKmD@D1O`zz)~q}QZchNGH!S9+n+
zuYZkf-v3j@(jg@>LY98uc{UXQ>79vA^B)&m6mQ?djMM0&s~%n_B87z293KCDMWpUi
zBS?QL?6Wk@0i{#@A<apFD;3b-Pe=}TTRMHf5};;d)z|KsdY>pNO#)&DYR}djV~^Z6
zdh*RJa&39euGpWj6^f==0(vInT>5;AyFy6`IhLhKj6SuPX;WGsU2IBN8(&hoM;|V~
zSwfpnJSV;E_x96M{5<MSv+$`_YsBm4eQqXs9(FGF=s6lBFIVm|;k#6ClNSC!VdGe*
z6vlihctR-#!5r4IdJ|1eXIAq9hC_v>U=EN72z0#4L-*(U3rG$x4r=>-k6D@7E5057
zT(m*u5QL|%uiWA}CD1iZ=IAByEg|qLl0)x|oiyIrO*n2rMUc&^@h{$z*rPGFY%1iR
z6>?OHc)pB{O=pU^8-KWfwxK(i$G|$pvDC-cuCgT1Zyw8u5#r;`g-O(u+5N4<&TE+R
z{)bl+Oo!Uw#GZ_`f=wJ=-A2vY1vxv%i2$b0$~UEKxrTsUNfqZZaIWs$wK}bmxFBpF
z3rTp+2bY-d-DS$S>skX;1`8b7+aR!)Lp8}$cv-OMQ+E8Pcr#V=q~yRBMf_Hw^V}x%
z3Uz8-L`2*mZT<j_$=|Y?A?j)q;=Hl~Jm_4eC9G|V0rpRWAR7!3i<VI}Rv56ICw7$F
zbj6P8M;wfZl1S^S3VSt1!#cpw&)1k&!<D2L^FXR$2<!4rH$(hQH52{%h4>^!!QDjw
z&5x;y-KpK?S?lRMeKxiBn8G;h;+dR)!-57%+-#}zFdm`m1V@~#8&hlZD#2XgO;#5K
zc?JUO_$XC+kb@H0d#}OLIqh6b)?nWyv7G=z=_Duq3qabVERJ{Ippm{+1!gu*H*&`|
z`2T8r>6I0?S!|1f^<mOq8Fo7c%Kr;6tMF0;#@bq2>PFcxpO^!sb7nj7IyYzf{sL|=
z(*MK&G|@FTFg-?>v1pKl#>7=fibHi+-ap{Wp;GHPcCbLJ&l4j(7RR#d8@y`!*&}Jy
z+mi881#%VuxxRp*WIFu)HS%+gyB>oa&Qwh6bi&h1*46Gf_W>!+2_X>U<=YNny`eHu
z1@oWq>K8-f5*xu%YP}cx(s+~gSOIE!4`_3<#Wx#xb%Hteuh~nLW!RTAjb=q?X_Evy
z@xhuZ-_b5jSytXVzi>K^*+OHh6qs2GW(az@e=h*GzS|c;pn`a+u$`TyU>T|rrkUjA
z!{xq5Zu>c+kvg+&OgIVIxP>&=EV%E&B#F|Psw8n|0cnA#Ao+fg8E}|pu2yYpYH_S?
z#1)rqR#H+}&guY%6vx2MCLH=zmzE;dl;A*wuJua;6U{f!Cs~psXlCw9Fz$!}1i9qq
z*W~JAPCSKnv)EsNq|ULPX6;OU6=uA!t5n>5bH-}xXGi>!6^}=Dz;oA}U{DO%^Q&3e
z>gi8SX}URjSi7$<UKW&Fr)b7(e!O!>%?OHMobNr*ZVFXjh$8C#ql|!QQ%qO*V{w@<
z&oGm2nl3W)4QUU73aX4k_Hzg7+N_Ge#h1qOPE09JFR{RjafC<l$A1A~fZT6^{yV{f
zk>Nt8FZXNu;ZH9&E<dz3q6a9mGvfDtjq?&cn47q>hYjBFnXpAp?;{&Iw#2+H`WzgD
zFJlO3-3(wfcJ9Mhup+6Mh0lCmD)~B6TFc9&vY+KyDe(3$;LA<!Uw|nh(QoffuoUlL
zhF6}yiqym-BOh=X>u&i$$NS`q6OG%#?mp0PXvSY|gY$Zh`K>f%%Y({<0X;jF^V4nO
zcrfR1Rkv{6`myQzq)!jV?E{OzLXGXvTt~~7y=Sm+q2Jhd?bpW+poVOR9TlOOk(c8|
z8pxNIHjKdn{sN5Za6BqC@*YX$5h*11n8y57;G)0y)JY}GvrYWY2csHA5XnaC%sI+A
zkp3^gshd$c>-NpIC!MH|13CTm1Ua>NmC}Z0on6OkSiLuuo=*6ND11=<jy6afPz5ox
zyLc4on7tClL@z6q<LJW0CcL1Kkxo?|Rhv5zlp(m&)=DaZG28MKsIU?WcDr6gOt8_E
zB12DN128@dCcYpayy@S*T%|JaB2RZhKYe%M5Sl~Un4@e;RbURO-?H?Z#Q5~{i!?FV
zqk4!8MnM<vcvshJI2h;baSZI8iXD?WW{j3>29WkEarDaYN=~Aj@^YU9@`3}&od`G7
z1REQ3EXD8`cU>_$jRHw9B96FoH+1VN4se-jrdiT9^f<pzxnd0l`uQQi`eh{PQGyU6
zy_swIpPEGnDkQT5t3S+>mC3DCr0^~)I?3-O`uXoJ9<*v_D$Nu8x^MixETd}x80rQE
zk>uYr+=W<47iLfy)z_^0c>iEEARiJK9(Vtpu{_@<n!|BkoKieAIavA>HZ~Vu8|qy;
z48Rqk#n_akSYxj_r|HP0<^wzJs-8XP(tlyXP7@pWL)A(*-7*^es8R-Aeg_jggh^BJ
zDPD+=MiIX5Y}2C{@V&mt!(*AZg68(=B`*TFbGGAT@!+fdAb$ihCux53xd~yUf6NC9
z@zgIqI<Fs7)Powy?%%09yIPZDzD!B=r0Hitpk~$+UpR1AfEqjIegOUa_`8OPsq3nb
z5cv{gZKs}=69q~nUo-fbTP9O1x`f@3*yxIVs-!bjn-Qyvo?}BPIsbxElb-p^s9!^3
zEB^qay*F@k171J84aAQ*Sar;|o|8rZVB`J*NJaOj5h@6ex9M*QFRsZ`xOox=>L>GZ
z$i)VfhgJw&TxI~q&OQ?w-Z>0-U979nq|J8^4LKXaQ@p>m2Hq&Lqs;ot=dao$&IX?I
z*DR_`FIybQ(YbO9O>YlPJ#k&pew2et<dT8!d&cQUQGFfpT<IwmGX4Azo-tE%^*IW7
zQ=zGOibrsja{NG)Ytyq!VS)ldOSsAUeG^1x+zeY;1nUUG0_5=z-b)m`>-*BoQY`?B
z^T@*!pFKV~7E&wsyD&U?Wnj(q5bfeKd^+KK;7{@wz*rS0b3~haPXmyHQ<+lAgIEwj
zpJaEc;q8%6ODhS1W`6EKX@-bDZ9aB?0WZ4ag}SY2j|@{$#=F<Z8eC_6C($6TLYbY&
z^|U;3G}dZ;>Y}PY**L(Cn|BUbQ}uM@cE0@wd#*#gN6HDVjt57rt$tT|xRv+Z{Lc7b
zy=TH2Wcymu=P7D{(qr8vUzar0{s^1~bjGiWYp9H)cel4(A{)cF4b&AAL>dPBJu~M3
z<~_Un`wuH~QM>_<<4WtCad~*)sr6>zK+;qX(XEE(A)RzxUg$BBiHpjhrv0`*$xldj
zdK4VVIe)4zeV$+Jq1LUnlX@2c`Th&QE=k@UxR?+~PgYZl@M23=&6<zwxhu+*KNobd
zD=!<GKR>Sb-!T1l-i>jchbfpzM+iJvDyj&?{%mNZM7DJGWjoNst2SB9hM{?<*_nE;
zK^z0I13a^B6Kmca|4racpWEkzLM;rbe;q9PR{^B|+jZz)Z^k&2SvX!$ke7ysU@)4$
zI3fEBsFKAMu5ARSi(XIMO$S7GvrRbWl+Xkj0_4rd;%A5w5vBlE1Qf&n<V2yANB86l
zkTdTU{T6q<S#_w&l?6cS1>h=TXwC-{9dU<BZic#n5AUZlMd-s^`e^%tuJ@48QrNq)
zLbBHJp4;in&6e9PTVd1)f;|9Q3a1m%m$kg0>;2!BDPE`b4W}l50pr0ab_`-7T$-ED
zg573vLz$>S2tGMj9)?h1!fP0I?~2X+WP=-f&kBE{tjzrtfIB6h#m0(l(&BzvC{}D#
zL;pVMpXPvt8URH!O5UL6eAvHzM#A@hUL7Y^Smd3CxFzC0)iYC<oXD$+MV~zY+=Vg1
zkY1!cjK2VrN;#>yjwJV@ett;%Y%Ft$&98p<kgTh%b6Mi>XzkQ*VF&-T3Ni0WW~I+*
z`kxP#LL{i0d|hltrHY`x-uy?bm^Y-$jS~`@j;@dU1JiF>)kg{Tf}T78Q|4V}8nWXq
z|2!x6#;$W05!5g}&qIe~7JIGm?^ga(|3&|cH)fPHLO6TKJLy96DP_H-%fBG5m?IQK
zE*LP#M?cnN#iSjsUn=HG9}xb$ym?=!4Rb=4$CDjSiqncUOS|jDWgZ%7C3$s|*<jh=
zKd%g;H}8Sh`sQBPBKH}7OewzfS~(xty`TNC4{_R7=oX`fyFS?>D|<9DqqbPCv^JSO
zw=>_1MT=y{56Y|{2%jdAKz6UbGe5|&vSR0_GtocU-IuQ%Qf3e2`-s}@28_Qk3q+B6
z#Rduf>A2W&)wel|kg`H%+UiZj3WkyU<9;xe0eL74TX|1;|9*0QTC~?l01q+eALe0k
z7Af1Pxq`d+cTcW1XM`xKlgXJpTg<3)zhYa^#p^`B^KX%7|4Q|vN0OT4Fn6FJQp2vc
z)0Trk4KE!bXqCK<&*eobg05+;=5*d@?s0*Ftf`D8qu>!?j4o5{L_`1lUvffWN=$uD
zSb&vt*l&ocV56OR;<0-4N4nOTWe9V`Lac>=0@G8?V0wbs?scg`MQm~-wg8OgFThaN
z&Nkv?VGX!!R1pLU?&bV;E4&gayBoJrE*-hO#@*C>IDVWhW2k<>-wIT9gSc(^?LOLF
z+KrJbKFC>cY&~`&hMkJXymG#Ui@ta=`dZeLoq2Wg`nRS1YT_J4-#|elRCS^Fwq&eO
z06obvOeUVd4#MRYRES?HV&QB$AIud``sQ5_$SnWH!kI?k@MA|`7YPwHZ2@BB5jBls
zsC*Lq5>N^!^Ikn1EVhf(a0N6!s5x;rmdNDTHpONbrTYr~@-2e&WX4I_n@BCYH8H!x
zQ0FWPNL#`2<I2YU4W66Q^+qptW(>vNN&A?UB>p4O50$Bm#!snEi*izLHXDm=n+Phu
z`bTCN;lHnm?}P~^?~f5O1?1%lc@s4LK740sKK2#A4j(TM(U-Hu<Jy2Pt73m@5`qST
zUsN#j4KuFPrf#i!KH2HLbLkErMYml2(uUA~JxY1QaqSn@EpFtSsC4HgD#dLo^~S=}
zu5o4__fxL=?+cHbICpo=R&T|hq2(LVGH`}-rq#`(XG_=0mD7P)lq%Y<MqK7t!tn`<
zBe`09h=HFefipuYU1)PZ^M<9VCYgeg$m57CA@b3lwcZ*axD_(PRD#H<!J;r+U0IbT
z4r#>5ozIx(m!~9YHlwxySkOUqqIYIJe1U>o8|!hqOwLAuLez%N-}-zOy4X$-OkR7g
zxS+v+2o(lvhcpIdaweJiZ`_3#jPQ`_zW~-0T}sO&12xHEiWb+D*-+Q}na;XxF!Pty
z=8FYve4i)VYNNRQ&)vHe7itz%LHAb;OAhMnQAWf#v*8H2O@SQACcSUm`(Gf&JyvQL
z`E$K=+LUor|L`!=4pu^EJSkS1$D!S^XH+Fq6ApJp=v#hPc&Zo$Md=B=xi!sL(V;=W
zNVR{+$TBCuwjmwsfPYt+kKL=8M`e)nIM*-)V)-ee$z1BiKWaSpjm^U}vKJ!MBC`Af
zTcB}t|9%cr@RUbUQ$#|A0&#vRtnl@NV5Zkoe#5<c;;tHL&aeKr+itin?FggbFK!_`
z4-75JbjJFpVWOrdA#t<Sf&^|SgfpS{FQ9Jw@_XR?AFQnP`vCh_9viLtzx1mo5Xd0d
y5BR6tPY8<g+`D7C|9<7eSP0rLd6rQ6_lxX*-Ts#}{%bJ&n@apY(*?S}i~kMFlYv42

literal 0
HcmV?d00001

diff --git a/assets/screenshot2.jpg b/assets/screenshot2.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..1fb63d78f85a4d0604359e6e7ba29d9d022d27ba
GIT binary patch
literal 228796
zcmeFZd010d+b_J5kbw{;*=j(*1P~GjB(T+lK>`#A7$5|vB3cHE2^D2h5sM@sY9k^D
zDn`VuBmu0n4xl(tK?!JWBm|_2LlqE2ii+Z_aF%_ar_XuL_kQP`_q(p|pL3mfC5B|L
zwb$Bv-Rr*Z-|u&4;`YQtU=bP=5(IEK9EgSgz{DdU3Cxrx0uUAkW&!{xzzAmpi0~Z_
zJ{RJq{rx@&#{z`EKF7l^IRL%_OZW_j&o7Ty`25o5um4lxr0X(V*L|D8KwaHkIUev#
zcR>K4lF1Y@iAteROsG^7GrEPDsi~RW<SCYP$LTYi9Hu)koo6kW>&!;kOb0hV4$AXd
zxNzYN*RMhXyafw<7J7Z@1V=S7F*7x@wXm@DVmYw9{>zVvCSXY+Ig@1s924LzaRf`;
zL<^V>M^A*;<;$h{#{-8a7!gTiiZRs$9#C%q@HheiZ$uyxjf~*Y+3<T{WJ#RF<oJ=S
zBIC#osffG0_$bBE|5Wp2amNcMkNC7)W9pQtbZeU#&MqwWOmyD-1zetIz*m7m!Ge&`
zsKrZ`E?fR}bi(Ry6W2(jN$D9IHg3w?oRzmNfBTM|KPdK=l<wPqpzNUXSfxs>sj99y
ze)>#Z{n>L3=Pz8j+M;c3yLSCoLuc2m?%!|U>3P)K*Z+86@X65A(U-5rUcY(!?)`@^
z^TGkbe@yG|nf(v*vV`-(8yOLd$Y17#!*7C5f~66W$stYhizLUTS~<AODTsgZ(NoRF
zjvnF{ljGAms8gKKM>9shOzp3k{qIdI_y5z({+`&s=G6me1RT711WVuxnqOT0lzV*w
zXsX^*du~3ENOaF|XF8-U1?m0);Z4-Bp+)D&dCG5hE>%uBd`+_O+l6(HeX@rXS1e0H
zTV!~>sj0Qm2SZ-*uMhQs`B6%n`A>P8NTcE@>&i6Aj9&HWy;k1hCdKNQ*Sau8^rWOW
zPn~z~AWx1O30gqOiV`dm|GvaFRWJ>4?*xMze7-R+37I2Q-JP@IkJGla-k$`+PTy)h
z+%;ax6pchy9$73jGTDEvDg$k04G#sSNX`1>_cm7^t%<e1)>~eQFCuv?I+-m!R+K_?
zo%z!uvp|9(*1F0jkmSZ15t91#(~83^W~EoK2>22l`}1*Y{}G+Do0Q_8%J1(qVh@k0
zoRr_1=*5LnaCGNQHW5G7q0|P5Gb_>Uc0+AOOPG@aIQeqY*|pDawHi{S$G+Ym@ELlq
z0h}XQ{=8xSd`67TMwUNKyh*#f%rcZH&^M@#PEM6*zAB0=7dwTIHA=C#K(D>@N^Z@!
z%=G4JmCD^yI6V{qoo$|ilusy>dg;X$`nv6Nc~~0D2A^HSrm@=8(KAkZM;5#i6mM3w
zO6Um)Hdsq!+++vD>S-p4r71quP60ZXP#_BDB3qOoN?^vq1(_lgHz^O9#LZAO>j+Z|
z#}XXMjL#}N@&pOAZw@ne_I1net?Lxd5(qhq^dbt-Idk~8@7Z?L=>^GcXeo`eh<}wn
z9Of<J@O8|XD2p^{a7307TF@aEJLTJE9NTuohO72D{>&N8Q$!XE_A{?nt_$QlO(Qy+
z5>>7g<lu25Ym`dAQc`Pl)|oft=VCbMpfDv8ou4NO@kNP6F_FrR68b?jp%Is2YcSKu
zo#!U;v=mpnArt$t7BC>&vRIa?2v`g<)g(C%-4<W$ccJK|v_PE3F1N4in6uNfYX|+f
zlZ4w_&~*yORoyfdXk7}#vG!xMRpd}p&JX1_?xyba)uL(ji)MP7D2x$BgMmm#uE#wn
z?fmN0j#}`cbB0NXl8V@NTtsB+!U#$g;~H7<O;+-59J+3@<5Y*-^uVT9&rK%4HKz&i
z`_gg$v%k~Kf<tp3JS3Ul2v04Km(!X@#Af7Z6aKxZbpCv!?;F%hMC`rHVFgEN)P@U<
zNHX@1#dRWUkQ@AE4rLvq;$NAWTjne@IDhniwu)sn<JN(Uo#eRXzq_5<kQ6U$v^`Wb
zYa{XkS2H_6{<<o`5$%|rW?Q%~Y5w}TWHp*d=l1Rk@u{XazQ-)G^lpRa5HqiVvY40;
zVk+3m@#Cpld2Vc-SZrY%weNy{I}@p_nk>K<bXs7MJw?y*(8^-rGIKT4aa7~X8VXJ_
zwjpP2{IU=if9NdqG>{hL@q2esVg?QkFNwRysTixY8=^%ic&KcC=M5GX5OOcGDqcpj
z;c4*2%Cs|P!E_C2x2Cc#e{X%?J}1%QPK$BYU4DKmJF64tDef&Ri&Pjnr%44(HEjGV
zK_JIX?X4HY?=LeBWm!}ygyIBbvHJ#<yIUBGhe>b)$}j?wQE4mUm`*z-aGe_&)^C%5
zq(3NeL34YHwT#|s|B8cJXDKCG#L)rMYnT~pwi}zHFuF=QP;^PpP%}iccBT=XLog!m
z6T3+SxT^MAeEgs?M#MbdtmW!Cu$d)aTZy!|bni>mF}LHMvbVc!p&BjtH9BaAUd*Pw
zEE7*gWnpzYd}?;O35Ene4humc`#2j|gkZG<i^wBMMl9?|QY={+RzEv#PblA0znzXz
zkK*2_MxG0dY;8C?XD4>0^^o9{xq5yLXsmWbjCUk)hWs2X0`X<!Ri&7izD*n}!WTqs
z>4+n`N@>PpmDL#iXl5lH^N;OS0J3*STUIQ&2igZEJe++M`IBi%SGhIWB|SK7v}{>C
zkF-q6y~(nfclp$xmj{a{z%RR}-|qH(_1gsaBXQraCvy^(Pk?rh36QOf_;bbd_x(A~
z&jj91!^X8ICqPB;_(wD+{x&?q8Mf5f{uWd1dunaYU;XH&-#+X6>e`oncAb38&sbqK
z0cP*{{6;daJq4SFed*HG_TP@*KmFD9%dr2p)&FXH;G<33_~vu4-9;1Tf8Y5(559d~
zloP)^=RZa;{WkpguX7fq9nCp);vYkCc{69vr<dzL{;Bt3jJJFl`Cq5^JvG~Ruq5Y?
z#KTvv{r!H=pNq`j_y6(s&apqg%;q0Qlzg0Yb$s)JFT*hV>aVkW8ENnv`y&&;`w+Yg
zD|gH^|I-Tx=lJEXoVeTn7}3<YYqBRB$>U#tX*+)P?{k0uTg*=1lXVl|uXCwwX2>SM
z{=Y`GEa7ob_(kdZc*e`cKfSzj-9<Bty(0CvaQlvjhF1z>?+I|FZ+wQ!_=R&XC!aN&
z8D`)2gLZ+F9{)z`YRJT5gSbA`c$r<ELTYQpm)Ya0VmCY&`WX~Jd?#8Wj;IIu4#kC2
zR9M04ffIb(zKHMj%()`(TcL}-U7sMtCFh!j*+;fxswG;^^xBzM+J45%C!LtN&tY_N
zV16?~Q;=hQB!CL2CD&BP`fsNoBGh+dR^)q;=ujg&y`t^}=gy}ZoKJOtUC+K1xop|R
z@lyTy0l7KBPI`U}BO>Sf6h=%k;>wnB1{ItD3tXu2JV)BAP>N<dvPKz?r0P%+$4|#n
zH3>WA&jrL`L2x6Wa>CmYuzKJ?#aNSHI7x=<u3hl}by_{T*fIXWkTV!0)efQ*mTeR+
z7`12bMSkg$3C#|(rKTo|@2CK%weJ1}sX`+)0i|A%D|jl=;h8)iDGu5d4nbsvV`VN%
zAa-L7CKEs=2E8Jw4@s0wMe0Cl16{+!jRZ#nPH4nh5k5^C;VtsoRS=)c#p2&+N-oGj
z<@9-YLb;xcfxTgKvp{$t3-f=$LIKs|l^B<<;HL}Y4QohpiquaF=0+~+k<rA1_+tMx
zXtJ1xk?74Mf_S6_ZAlfrtSD+NMUyd7afrg-ita5Q(TPYbAi}@V$cIA;%<XDC#X#%K
z(*P+8XzA_q7`Y`w{3f^&x=~|Riv(qkYNXXJ%V0Ug-&X0Q=QO427T9(IEt<;T5ii#=
z$<gx_0aG#6b0Go!@tO*6sgMrH%?OW&bqkOtj>5Ck#3YbFSe6mFo1b$*%Behh3_E#D
zMrDEj^$VeFw7g>Z>9=z~&oX=9pOUn{HSOL7&EX36!F#VJfJx3@-Y)7>esVIhd-*S2
zhdBbG*3NSSCgD8~otzJhiUs#-DFJ@XZ>sXb_LArL7k0@~k}t)}9(2)GbzN1Ee1fJb
zDmo8p873z-?~CczGp`!O8r-hgchUU9+r9O;f>+@Lhq6DJRfJ-G9#W9aBF$xpW0i4n
z_gM?;DDGa7!%A{Wa~ra0KxV%%@+N0dSE@k5F_h}@)D*4;+%xP^gRAunMyj9Vg;BrO
zdz&<b&FG_UnRzsUTD+?`(2GV<wV{+q{zE;8sC9lOp+By&;X90}P;fk}^NgHNp{T8O
zDE@8bWduB`7_-uW>kbDi9ZB`c8*Cq~s+lCI++l6oIkeN1fd!liQ36xEoNDUor*dm$
z<1hl+7{)Qt(g$P&Eo(?<<XBU^S}VBr%5b2T$fQU4`4EBY1D_|?+fW&6VJXd{l@lT)
z+8T<Le3zsEA>TJTrXcDGzA%ufBKT=}*%H7*j-Kf*YH?9=krr#U7C7?;4i~a;b<ese
zLaR{juz+|Cm2F+zl&Z1R0WaIAU}yQBApsu+8<5f#EwV{PS<2PI(Q#DifxtCWeQj7s
z$dzsp5vctgkMjhivLp{vh7}yo{X=1H(+VYb6|t-s(G7t_isZ`hToqnsQA^hnED9xU
zlN~I?33hB5E>FisK&}jWN47Omh8cM)r<8$cHwk?WUTLgh-iykF2K<B##3g~-XOPN(
zwVHr0!VQ$r2a2f6ya)2MV8Jk7$Du|!)-EVY6~?A38Y^?I{k2kSTSf=YZGQCJc;(Y6
z#upyB9S`ky4quSabKcx|DEiGsQ%(K!J%dW|udlu?C?$B;OgqV)0EyU{hTku(IF|XX
z=;^s1O?>b3O*n6x29B@Hd>#-%i}>m8suO2NKgqVam~DN+NaJB<>CF=Wsoh{|^E7!~
z|D1lDIzvI~scE9OebbIs7{4p5aN6?GZfSpTx+CxGa^}n_jK=hKpDZP4e0V?DUw?_6
zQyqBX{x@F-c%f&)tnfR0@UQ1#e@wdMT)VdRIBP+uH7@9({h@~UIX0lX`}+{plPR;d
z)U0xHyL``Q(PuFHb$5=%C6DS=ckcB3aWJuQ0xbS<+J-Uj)OUK^@lXCQT6b4ZGtmj=
zTseHAI{)ku|KQLkNB6npJMTW{SpDrnOiy@%;T3@OpT3wywE4{5`tt-h-ECUqy`+-8
zP8#HXsY{kB{O#3qDvVTFFjD<uyBtTmr>3{{x{DR1FZ@ty%Ql>TplEj;t>CPg^wG7x
zdi+#tRjlBO9#8DyVF?&AscDu~M5{SU^AB`gX1hUy_lOm<?`6fyoa|<9aW^`cR5UGR
zYr^Z}y=7+9)KGFQ0q+sg`ID5E*@PsPnGRjABJlIdJ6f}!-~`vPQ6YVeb5TJ>9mR8#
z`Z@|ODLg;g#JU>y?#oOnJ{sX-DA=OsY>7G}By5;#kl})O7(W92WsDN&f~q=kULukd
zsNzfG<+~Ixlfh}{A~b*Fs@QnBX_Xk)Xv&QOE_5{+1y*vciH=i<98}@Aw`YkZCfifP
zmp+m1ye!>1YI#w;R-0=3)H(uf2`SeT-IpKB)8T26f;h#>I$)`#H(`JV-K=q+0yn6X
z@LOzB5U}k{b)vPFd2A99l=$&ORJJx?qr_1WV8caKdS)6Yh+W_*GG$nv;Ukn7oS*fO
zkl2iLDgYQL9HW^>n@9WOo`?6Akpf-5*K;I172^pzNMa!tfnfFY(Ac?teuGLQ2Cov?
z?wZcW2x=FXdbd~{+UO}_0A-qjYSm~xP(-xW;770*EZo#u?gB)%AhCpQjS-M!Qkf}-
za2L4>UJn%IT~FFwnM9p-uKQ4X5&&({VivAJ%S1ruNN6Q`0&l^tqlkPYtx{PTHiuv*
zHxw0!m@xho0Vx@kiP>Cy87{_<>`*4NHQ0v<NcU`FL~ag;;KtBf>ZZY6N6SX|lr-*0
zFajd|Entr38y#<vu3gn}x!=L}{=B*I`yc;|B=xM@ee8Prk6(RzI&t}9(Zz;@!o%9k
zfuHB@sQuorD&~`KW5ajX*WgBagH3+_jQP&Y3kYbP6jAkk#FoaOm$T^Kp8LkRx3_)7
zQeMda=BMx0G<z;7eY4QvQvEIFt={uO_u$oWfkkW{b!%lk@OWf8uxDZ3!eXb?vnutS
zzRa}H&#8N^&Wm_-czj>dg$D;xCi#R7?mHT^^I6YSGTAo6J5k1d^3BVjPk|F)<=qM3
zKRIEU;ZLiF+4~GVU!55VdgX6=XZ997KJ(b}+IbPAr=Bgzu6y_Y>NH~!Yi)Y*AE!RC
z&qeRwb^OzEy=w+5_pVi*!<B@u&CL1yAwDwi<NaCwqa=0pldMZqr&`6hHFU1~^Un3;
zHpYC;?>@sl8DBpb4mwk@DCf+jq95s}c9&;hCUqiQAVB`p7qUy=&leM3=X~COGMOA&
z_w5`V>qAs4wC9K4vnK$f1Wq90<>dO;)t*KPg>|$EU~05IBFp&RF_}$Pe7Si-_!av!
zay@FFiGr#M_N8Z9ra75VYcwiM$TE!bsVUN{8Wx=(zPsB%^-oxbv$F7tGVYa|MQPa%
z^4;y|#Y1WWQea5a$Or{uGwIHaDV$I@DJ@*W!~A|9Ui8IRhzC(%FKJ8Xlb(w1>2c+s
zS>*2A9jlM1om2VL2;}Hxs*z8it%$J2pl9;6bPU*v*Ho}@EvwT*E1^xj3XUc*ksoyo
zJ3~=_x-hQ0P)dEIh-}7)4_~XUwINMjA}-RjF^t(;q>a>C`>=2VE$hP)W5qPv#V=NR
z3*97^_0sRp0ONt5KEA=ih}>A3UT`=)IEiXC6mE_A!@v%nEf^6o@TFrKnZ$1Bj?T?$
zvhDrjK|5uLeYr?<Z?N938g~^jq-9srKL`!D1_bK_ItPW&zgRaH`oG@Nt)(q&7=&0U
zNz6)X72spkSdh41zLN}HD+`w*H6BDJ&2e!Yg6VWRpUNW1@ae(OoG2rH8bC%HO$lao
zrQo~@E9PVN5|)?YMIfN7W=g<-T!+WQeFM~;zoy{?!_;?od#F$Z{7@nWsxZy_*f%PU
zX2x3$eK&Ax30#o4B?@<<s-AjDOwUghOjY_5hQgOLX$avSd|PEKX@8NaBa$NB>Cg5M
zxLRv0`DX1%-IYFHswlgL7VW_&M}_Q(j;DfaSbTz=#03TEK_2|PbrMGy#4JE5xF0Y0
ze%iL<qo(Ax^y&H=NP1Ah{&~gIa!<IMKU$KE?smTtFr(tIbnVEkqjJ%a7HXb&#l>k}
z?=Jm#SKyj3-ruofdU4f?=(}HEVb-5hP};uw0bl)z2m;LxRIRx0b8pI&ju_tu2boJY
zekgL6`kv1CqrQIdP4>?{%PkULF5680Sy!OPJMYm1OJ|g4B_CT6n)c!Z)w2CUAF2Gq
zsv|RsPHmZ%ShuQ(zPQ!*Ay*6H-d$K3{%dQGoy*nzJ=)8&=KLtb%Xc@>+kQ?O^!9mF
zo4#x``SQ*Ux`n2$4}SGI+nw6?_|pXuIV&T1&QIgFGwueYm%nc%{s9x$gn4QIv&fiG
zGTMIs)Q8LCH<mBmUSoYDJ(QyrBo*8l{ADuNaOlYUq@1E!%(!|!;Z=$>f$l6PlyB5L
zYo}xUsg2x(hU37g_#Aq4qk>NLulQt0VN;uUYL|swz&t=_TF*n4Cu|xVf>}_*3;x<A
zyViY69}?qZ=05NF(XAOHvrttOJ+GFHws36~F#|;{eT`IkQ2IL&lUibsQ7~68N94SY
zLm{zKwFD%)GsGaXZ9sPC31W%13cy1X5L(Jjb{f&mY7_crqTJ*rJGfiEOS^#nTo_Oj
zAhr<aDS)|yl4KrDD(WS1LawR2aHy*UcfaORpAz6p!nOF5RQN662)M^_S)-xBNr3OM
zXwQBbk{-&1hOh#Cq;+eEFO=YymshT~U8>fo3uc!q;uDo_Mc-~FRb=+&5((&5gciDF
zn4jlu=#`P^Z?Sa#6&ZV{KSdoGfJ<Uu6x%R+^SX13GDZYJQlL*s4R4>vClw?XHBrTE
z&{>oo>MpVPGc~LU!<R-`9CSWVh0wlfEl3sPV@15<xGmz1Se(LB1VYBtz{Ys0To49W
zwqi5r#sw^J)0AzbXOhXtQ_{9}w4pte1S0i%X6R^`HAW;nkax=nC48e)ej9=Bq-WM)
zy>hDLDGR7>vh12{B6UnECq%Dz8|G{99pz&B29+biBpYR@dVJUg-#L7U`hK5JHTdZE
zy1K%rk&Ux|AhDC~oNw47!sUf@7g4UFWzgNTO~DNMH5HhuWXraR`w$uYfZm=J5h?V?
zDK`*I&#Uh8-M}V@-C8vAG=s0OO#}gnhZRSJvpbUZXL;v7`w)dNlIPT&bE(3Y*mR0>
z-YvJco&Z>0&L2PczEk=>ip^Xzo@+V*wn8AXH)-vSH^t`v(?cX@%%v6+AYcCI$Op=T
z&+niY`Jr-r3VD0Z6UHBJdn3l3ce(xFKK@U>9-Qxkd+p|~{OH#C6Rma9-B;6GJqk7u
z^tTp;?-Tss^>jDAonGS!NtsveU-0eO|B*GUdmQ{zQdmV_YIxLBDcII8@7WW_FDu+w
z%iH?gC#k6DL=gL80~XJo*7!7Nh@V$%C|E7xEPUR$U|Nmrg_r2YIrXB+`TVr#6L(4m
zKF{R49`dmHd4x|5(bTL}COGvVYk26rO&IaxeK!8&x-1dl&9kG+x;jmOSGJa1++8FO
zwu9y+`*T!aN|?$+&+7n4RFKFDu8$O0wj(ezI#b-FnpG^^h_J9UiQSGh4n+kJrixzm
znKijd&qc3KRWtex^xhPfZMmKU!Ko5~e_x@r39#rpdJb5xl7JyCNq*E%QxS#=us2ny
zidaVPfm-H{P+YsA%%ogAP_*s&BBc@d&{3i!aW_<Lw{NIQagF}Q+X;VX^3UrtO6HQ4
ztKCz7S7d&cSlCN>B$?Ux7;WINY}HtuEOb?AlShzrr&pJ_Yqv#lAQ5v!!4|z+lgcF-
zSl?p@rSd7<N|+TiqYZcYCL!k(BzkXQX$!+v+*_1aVE8De$A0K2hzg(?;>)P)6ba5u
zVvYF);!V360|`{W!cvNuB&|d#8@&v2zTUvj61nMS7!{Jc^M?f%@wx7v9jV*HSvpQW
zA<xfYQ~)sIhe(*6c!U(b!|>%SoO_7cLB2DEWvatZ?JhxN9EI_K%!+gX)3|X}*y`fl
zZOX#TUvm8T7Kj!turCE=D!klS;m;9}PF5nVOo&kFnF19=HM~R~Jw+sn3iBU*P*O{v
z#p<Oyg|P-35UHG|W>}FS<8j@Ji^b|e7xX|eOU(;XKW~S?bHy&cPa#!7^Z$R#Z~uqH
z*L0=H=hc^Yy#DF9K80;FXE(idnW88|Lq7a&Wx5(7)L}3*zF>T}J-TGwu|4OX3w%9E
zsG*ReV*ZwWCQKt3_*lL{W5$-xpByATNFEYMHqqZycFX8Dr|exY=%S2wx*3Km{L_9P
zjH!}zKYwViWlYQPV9s0N4e#z#uTf!q`Ji$WF_qsO$b=r9B7E2>zm5?d(OenwP`t))
z&~XQ<64MV41)ILsaGK(ktwq`qLq4Gq%-Uij!Hhlfk47K}u~AA!T3k~VF*4TVtLS4C
z91f|%4fFFqK<HH9)d5{P7n4aq+hakHrx*vW>wJA`WT}3`T834l^UNO*oPwhTteY$y
zbBA-Oh?5>1if?XZVM0;k^(q{n#GbF`+*9ngu5fJDx<}8ST~91ZUS}$ug=ZJpQdiZ1
zSeq!Vn^hyBmNBKIT7!GnQOMp)*9J3mi{>~uYMIsuV#iiZk0!Bn?7@1x4be4~-_nj=
z5`%u5B|HpY@QP3IgRF@whu>SYtq-m2*+pa1^&EaXWLVG^SLvpBnH9QWgm01uf!FZT
zsvbF97&lG99D>qlZfGcpg+MZc9qb)taWP#`iInM|UnwaNgNF*?5>Q#}XN^z-hC+|Y
zZ<9%db;LYIMwN(38B@XQ+@{7CwX}wj74fu(;@9d}iwX&{Jbs`MdS-vfNA>!;Roy&~
zP>xGgsr+a2@*&5brQcd=N?BYSrX^TbYnYd{f?!m}2{j{!&IS>o>(T}IQV~9~*ie?Q
z>6Pya8x{boMFpMy7N1#oIc``G2<(0&w@!bFU;p7ro#~$`i>r4#ZVR0KTMBsYCqFs?
zC~wY80RG)0AKyqOz_v|aLl6CJ`{$47Pm2BnvH$i-&K0i{6F~1iIR07bnzwgl-nixc
zHRJLFzHk5Rfc-rC($9@A{ai2$*=YVQc*Btoq!`G(oC*Hqc%l6>!8PakonudUGj97n
zvg^)yVx10Mzi#@A!eQSr+W84^IdK>Ee^2?pys+$hD<Y(9MX!s(@7D6zuEp2qWz2oO
zcYJ#Hg+oz>V%0{pk8WK<^6Io6teE0EI@57nd8B{Vcf4Ng)-%@>ig#yvVJS$vcMfs0
z3&h)coKp(Zj5UN)T9h_$aJJaJ0dpln^-X({hah2bT6%bomxytb^%f)366PiS_HA<w
z{l+ifF3de%a!reO;$cxXQ|;>xL&x^u08CmFx(iEj4i6RI(QLqySWknE5rJDJFD~J?
z4Btf(<%Rd<Po)=A(!vN+jozw96d-vn0w4pAfCf8hQ$dtJu00@(F(8}X>N9jn39?>z
z_<4(fhXU998cWDqQ_s?XEOD&CqDrn<p9NB^E1!Lq*i7YBdh79NVXknm?mPfF6iHhx
zu=6q`Aj^q2r9q~G7N1NAt2N8liu?N%#<s><Yb}wKzNp!@)8ffp>9yKfK7j`wr)xRr
zcYzMS;-bWI1qCudEQo{`bSNN7F0)cYMBKl;%Y#M@bZ{M8gCgi_wGJ>VF)V%qf{86`
zR+WP7axI|q)NNV7t3!u}j74v$nWvuPrO!lZF{28ho++j1+>lY|E{#hRX)oB6_12Ip
zaeeCV!7tkqbrN#E&Ze$hEN)L$_&XKef@nwNm?0uJd@jVY*C49IGFfVOL4^d^b`;nE
z5!)k=-!h0ktD{hmQjno0cNbknt%O2hf5!tEuDqjMggX)%`BY+q(~)~*@ohJ#r*E7W
zV0(9cf5Hzjr1MP)${>uCKq}I6!q8EEQzy)pVxk0aR`kl-QS{iT2s<`~Ed?}DAO}t&
z%V0Ez$q;ckN0KN_8fKQl6flv)(xW8bnN6d6KFOzb{uWJ#KJcu!10nP0a2k#j*Ze{3
zQTXI-^em4P|2gRUZ%*;$dIiJW;;M`kYyD|>*N-b~&%8TWi+z57`Jpdd*M5fF%2QR-
z@fBGjaz*CyJ%5%QdH)pVug5rf;-t40wk29es~#Cb>qC1}4YJtHT84dH%PhN^g3e`J
zjbOCEJX+$k*E}iqAftJ*_4Ro(%m_$TEudE=C}!idD^E>d@JQ^?*zBBWzt4;#Bs%FM
z5E?0UQMU{!d9A*_u1SPXup7Ef$BdRT41-D#b~ucMG~xOcDc)l6R^_c<l8>x$F3YED
z*+beXxQk(~Zma|lW&)YRo8jzmvf=PHLeJs5O5jQ5VMJT;U|}h3)v+QfAWpw2vxV^v
z{<Z4v5I^7aTs<u<`w2Z;nrT-94h|HhWG$KGnjVg{E>NaAtBgQNxDQ8|kgD_YsR5-J
zO$7M12EE{XWgh<-iw;347-G`%7SqlV`<j}zf10(f=Yv<itxUq)-)c&Lj3<hX@=4rY
z1zbH6P_1dkRQy2Z8<krd5=9qVDj~f{WSB`X{$|+K`N~?3X)~#uMXy53#xNtso~)Q6
zHjo+mAcuP?u|(}=*HZ-HTy`pK*YhQ+%9k#-<mb=igf&Ax2Q7PM3bFX1P_C^WZyB=Z
zrhGf(e+u+C1x3Y?a`N~xxxgJ}xzemC@cOR|l7+xjk|dk5S7sOHh-_z(kpz6%wpMmq
zkIaVcspPv#K1mA#y7QVa`nx^PXWqv4)?Rt_x_Z8xzwK_ABLujjr3j6E!W}}!XxX4L
z*1ijdFj5t~xSQ0Ne<b8Rkt3l%7}+hGYON{eYe)F@b(2U1{USGO%><atQCwAgS;?}y
z;}_-_H5QW2e$Si$j){jLP`Tr!>bJfNJ_f~a!~b9+3BIdtb^3P<cB%3Qm-81!TbQI5
zzTUUK&)VpP^N+_?NsD6|3fYfmURyq9l-DXDg#WQ*^QVAK>N44iOfmm`|1+0@t&`P!
zB7!U4)JsoZ)0JQm!s@KOH1xHgH6^^`xm<Z8uWizMxBE%<(|<X2`p`Pp<v$+Z{v_Ot
zKOpn0XO_B`=enCh5F7UJuNPcDB!@2d$K{PTm=DG*iN5{#NQ*9+3<J#<(}!0sDZ{=?
z2nk019ObwByf}GtZ_aQHRQ|Y!g`Dv(=)6lMGHJYaTXEcuRnna)QK8|UouG%)j$#Rg
zxn;}Kj*t0B^Qq;9rP^52uqRT^?P5QzBc3Foe_EYQ{3?A|81Q>l?KdZ{`aTl1ENhtl
z%hn_Z7VtLI5~-Qb{hSJ@Mj&}OG!@c+o(c}%>|x|x6dcND^~&tzdqN{2$q-YEmMf_Z
zz{^l9l;gDAH1@rCNU%K@w{p!`^w&p00+LLAZgZ5nFk1KwrHBOFs*HX;9&cRrS|vKP
zlUucV6GqzJtcHR7rucm!D2<0N_Yekzrz<0M0_LYE0jdXM>NgsEf-(gLszYueSb%IM
zJ6q>aBvY(BkAmG=(0OtPu2{$F<EjW*$`p*SflJf7d2iB`gc%p&)!W5;U>v%3ZFEvV
ze)_xay`Bx%c!=@QdmGeR{FD;tz<T9;@}tQ1w-TFKFC}e=!iui5li;MLZf2!ezX4}D
z$8C~=Z`&dC?^DLa>fPGD?<iM*P!S8q(XobfUv_mR*a$SmN+86A+p#PzhXxpA)AGV&
z^lts)5-omOfnlUq4t%PI`P4v$HA;&zlqf-r!GQBJ#1531DE!I|BrgcYkEt|_V=~$%
zC_2JHCMZT6FGFl#Ml?+k$%pXFDBsv`$S;6@y(*y|c=}$}uE6J7+uqoBd`AsD$zFCN
zv@*eF4c;;gg;J|+;sk*(D(efxzkrS@jqHfd+10r8;UKOKKn9gi3&GfJSt2MB%txsp
zGz#WrQr_geKA6Nzf@++fz9dHfZc0ZFwq<NlPd?cF58i;hV3c_Fo$Keb`>)hAJ>44j
z^;@f_tLD_zbw>F&c8+X272>0NroD31bekx|skrbqW@@r|%5<BjFTHc;En@wU-VwAy
zIq-bmid}-fTj=lvuvj(u;HEI|xa8I1Cd)b*r=1uGHdwBVS^lnd(RaIce7EAA_wC8I
ziwZ7hnuohdFW6Vk-tSq_xXJ6#*WYeD8?=6(LpYA``IzyiKp<Wrb$nr8o1NloSpRT;
zZ^YV~9$%cIa((vk%pF&{d*=Kk-?H}f-8S0;o8l(T6VgsU`eFS#7XCrKQ`pwEab5M-
zSD*c1*`Gbp55G#ib=NN9sBiw#mFvYvnm;7B9Iy&b{j9c~#AB{3UFdeZuee*bz&mVH
zYVwB4*;d~1D{h70&n|E^C-<Csy`skVk;C~ND{j@#nRz#Dm-U=OH}bdh?!9B2-Le1y
z6Ts>_jp?fPpzl83_%31nV3<$&9H-M8!DRCK8t#wwlez{DExB~EaA5ppv%}ZSiVssl
zCjY=J+7C`X@x#nc|DCLwyS!7wy^c2hbyJRK@A2-o+2hXNNqmRl4TO52=%>k$S*hLa
z<P8%p@3&X}cxmkWx@!V3SHwlcq<(#PDy|bPNZYJgWQ9ETW$U~sqN|OL$=s?~<nXc5
zii$drS8@?m4c2mK-V08tzr6<e5HgjWUc|yJ&MjbrUYYI8H%FR%cQ3M<+1{diUA-$^
z;orb^`8jkW7({=yc%?T~Gt#?dlf#33B+o~L<XB6s^`{}QJI}x3HjG=9;r(K|S?MJz
z;G!2(h4}rD{8!sTRMp`CLp0UOsiHI)Vb}>G(54tX8{lQg^yp~=C<U~wHSB+*Wy*N!
zR%zj*Lg~X)fsvML#s&>YqbbMKp)im!S;CthnlQC?ZjupG^;Cijj!wlUSwU6Hy<YiF
zFuX_^Pa7!EGdHX4Oqp;GvMTJNc_<_Kt%bjIjkBz#HZr*0OD6fOJQF=7E6+cr%|Vvf
z<<aJzt5LNl8BZmZo8m1DFcjZVtsdfE^Q6MvbbC>%Ag)3xm90T7SfCBr2pLjLgtrV1
z1(Bh_Sz!H3cqk=YdPQbg4gIPj8VEla^!RLPHVc=jf^lL(y>%PRY2p4u;j_2saaz0#
z?;%L@g!-KUnS>s%031c6Al9=($w#697Ug6f%7-LVw5gj#8q1-`L?78LS1jJBLMUvA
zB{Q=0lvQ=xAWYqaCCVp##c}DbHDkNQTkdnHxv<Z9n}csH2NDIg3V$~i{V-d#rXDbT
zYA#2UP^CF4^JlXxth9{RP+D?KhG%<;@PuGz87;D01hP_tSQa~K8C*#2DME><Ty5;T
zwsC$k^tDrDTWn3&z6%%+(k&RjX#P0B!K?B<>%`4}s(?Jc4^*8!e=MR(b!^H3dYU_l
zGVkc5&8ZG+ycW~+B=2$i6}7?Q3hCwaoTRglk=UO+ZsyqhdgzL$?!jK=$_Mq+Ry0Lj
zHzpaH1+A|tvNmGI-)~y?)+W1P{^_!x`_C^Qxf9Hd+wp0B|E=8#5eadccUuipr@VAm
zojATxC2;riIa44aovn&{^Vw!NV)wRH-<wW4>ps<Svh(4jwHbH63yK|}?00WU;L>#<
z-gn0aujEx9h8Nv(Ja?`uCd4}}pmE=u{9nsQq3n_SY1$r>`31k%4$L0^ei=ynX<FKm
ztrDEmiw9?NHdNhvnpG#RFX-hL2b@}-aM>rRwYKuRtb^z4#=Fk0T-ZOE)qPjB#D~vo
z8{(U6c~q14^}V5~msejrv*X&b;sqyucyi_Ep`|^lg$Yj&blki6aHe)jOGH%`(Elx=
zB3vJ5=Ulr={+u%XX2bFNQ#X8nQ1%#K)lC50c>S&D3NljBx58%8+d$uw#|iD;vr|<^
zK75I`cMqPLckm3YTtrG$#!uHe#-<C2QO$ik^%5<BitRE&i9%vyS6~~p{9&=M_*eHz
z^pU(7J4(5J4NC15d}kk+PmJo9eD+>l@=oNSdT-YD;K%Y3rN0yO({#C?K=OXK7YZKV
zEtL8M0$zBBA5a1}B1vX?8|mmSYg2GlZb*Z(Y^u~&znqGb@#v1oC9~2NqYU+3WdJQ*
z>jIt8>RP+-FwqWzmVx<~R}fv+XVJ|}gQEiwT87%bjwW(X59X=n&tdz{!CSc?7i)H=
z4hcNEW&UrD72S?3=_%WGlYNz}w1DIXN9I!_7}&;Wjus}$Xz}~y)T!O&;cVKvG<L9h
z3X2oohPYW_Rk0#xro%pgmIH<(v~DiBLztuKilxeJ=`z)$YUTlXUw5h57$r)_>XmO#
z;-%*CC*MvNND*frsX0kzcdn5+EF|6yQIO-<5{t?bzac)juIfX#c^<ItEo*6IL&Z4W
zP0}v{rgH<d0HlX{ikK?fv8oI>8DU~675evViG`(~HO6O4Y-mt^48hn&;HeA*b8r$H
zgc_g#(V+pjZUxB|8l<@QphItwf&|3&EQ?aOB4Xm@JHuHp-mp>xC;{27x>x4shk%3v
zaRx&C!NEFpX$^<y=b0sv{S@roNl&aMWx)2a3>ZiV;-eyl1qp)JnlX6Y<4dYBKd3C6
z)}~?m4aiPf>Co305gWtsxh#vGn1>>~jR$|*FdqSy%NKGcK#^6>?}3<o8!r06<bBN1
zpd(HNmQ&*P`w)$SGqzOqz~H+A)^21|7j~F`dK2;S$<j5;vyV-G|L2jn`pOYV0=2J(
zV&8l)U9w|*i_7OR*2k}WU)}M2ew0uCf3_9j-t5d7Je~80|DL*l3E)#Q{s961S#P41
z9>HN>*GJk0y*{vBX_&&;f=}_KUjAomnfc$%|JfQN{A)0Oj|Kd5Fv0&C3;5?mBL9Et
z&HG62+R?CH`la)e-&`Ri^O|+3_1`7z&#{_UH~n=<PE$_phK&sjD(NgLid`%#h$7tl
zH1}rF)xXf8SlZIW_SEL|_9RL((&9yr4!;~qa^BcyXUbbq>()T>fkhdeRtaCrQNyK9
zOUB*8O9o_)fViZmDXNARmrzcPNmN>7)Xwoa)nFbfq>4|u`n}fJ7}-LT<+RDQs11ob
z%TkaDgYG_bHPraG*IPp|L!9%0Us&{q9KJ(QlTFMkE%4d|C0Z3(dJsg!oe~C|sVH67
z7AYWX%o3-Wx|u$xd*Q)1-V;r*-4bPSG~bJW0t2rShU@vBxmUT!Iu+qG`#i=s9}I7=
zcMsIfH53aX3!+lewLm*iN{hlZ43rZINW4<uf?VnC!|%!ZYJ0k$%aD+uOo3R>NJxRQ
zn3@yHDGcQNP$BICm9=b$P)TeJ@;tuDSQTV679=m4wt3A=0ottGpLOuj>QBSb?EQh)
zuv|WCh)*@I51xA;MV7Uru?Ae#19|9Gnlh#n=hv&07T_vjX3`|?lG*Ty92mZ9IAPE+
z60u@~%*r;hLV^p!7v*CLz~_Xr=yl~b5NsJ1`or7;*h#bmT4I4{7S4_>vD-ETg;yIe
zMF5P)xnD9Q;ytzpKVLfod;zre=T#1}lkr7*TsTS*;^CkuDN>e213bAKDz-5jJ3J(N
zxAk*$O+0n&u6-yI+4Dff*g7wk(=Za|t%zNqwv*BlrF8uit6HXt$>#7);4M>Kt#!!>
z#rF;(vi<=XqJ%)!7dVZQK(h%S11f3&zl;!^<xd^_6wf(&?sG<T0(9WuVSkFm-RYxn
z0=(*h#@q4c-~<?~JMy*z+Z6Zr`^pgyXvb3D=au}GJ9^eYi#O!_5wxeS9h#`>OA!EW
z<8<kf@hu^6AFx^2>HFyoG+dN;J6n_UegxXQ<nR9VQ~!_cm2<HCS>o*M5Zw$sG~~+*
zK9N8C9wlvMp(|{5x)Xinkg#~sX7kPGdf#hH@4V0VOS`v(i}Y62^p??#3bQ)Tts7AU
z*z%AE(t@pQcXpl)Z)8#3YmJ}*J$g#VFuk+(4HOYCu!Y)p1gOuYD)gPLjtTduxPvdV
zUuF*(Ba{?t2*eb}^~e=Qt`ZjoiS_YPr+;6$h`D$)G|UZIfQUdCLkty@bd^gZ0m<s<
zmYb0ar42g5$wng7C{QyUN@3z(ITS%gTcGUrFwt#VMS7^c-1zdS`d$^{7LaEiy#S;N
zPHoB(CzkDC`3!9_#M+JqN&KgJM;##Wu%m(2&jox>!zgq+!=am19Dv$v0?<#J`Z>H{
ztNI~&3A0`2+(1Hk=VfYuO-uaVGD>5TA3s+P3uAZ~9jC|~QpDq4tJ~4m?%eHdp^eGk
z>l4Zqltf)^ukez6(ls9DuUP%Z#-d-_x@DQz(Z-=vJ{kR`3ncf|f+*mVCANZ+Fk6v_
zz^}{jnH4at+$IHW!l@Ae3$wap<WLTzy~Hr}R%3W5ho`c%oLzx<z!qQulAA1gW0+CA
ztf4Lg%H1Cnb~OPsRta!w`!P_ek;{<7EGM&6KIDs5Js5-PCc0%%b|4@tg|I>m>YZ`}
zyrZ&Us4r&GOV^-gj}*}8%A5>i2zXZ*eWL5Mzk<JHU&_73&b5^9GBmfqy~Mpmt^L&X
z(ImPSeXSydKs7f%AC}%g1U+IEJC%R68Uri~n@|wCi6LS(MxY5wh*-lkhN4A4efzui
zf5hSP#zOpmfziw+d|v%WAPgUS%u07pX)RCK6S|Q%P-dV0Xl>57<<QInKIN~)6}~3z
zNe$zkiX{RD8RBsypD?&YnsS3Yg8tZ=TTS|D`Rb^)Mh<0KMQKBOpP0EZ`IyW?Nk%~>
zqx<kKg4K~EzUa_%tLMTQFpx62h?lo#k=W!~Cai}r>Bt^BkG6QHv7l~-(&hk}08o_{
zNSIc!xoYq+W@Kx4s3f_vh6HitW2(WQB%r<4bu+9_+TGSTmj~tP5^FLSBd<|P*QGBe
zEV2rz=%RebZJp_EePKS4cx;K#Yz5kXr$S2np|mB-O3c{fZJ+(b>2UI`%LU^4GpvWJ
z#j*Uw{e@+kS79KrbPfR-M2@==0efqf_%$RVVxpFR5T69Li3OwKO=P|Ja|eH$&RPq&
zW2#oTc-k7L2ZX}Pg6B|y%gGCeGF6@$R&R*dJC_XUSm-m<m8aHnwsT?4p!@Q5w1}7G
zHa(}imPYKmX=>yVxWbXqAC<@r<u7rHmn-<>;!qya%CH~nGsw4Rs*t7_u_a8y?a7Fh
z(ugn=1eN@HD3=Eix(y_f<i<mb_<8k4Q9ujD@ny$~OoCVdd!xaH_&M|zvgo*>MGDdk
z1ZaxPBo@y~EwzMcg=1#rdS)P@mN2CRZ87S&hLO!mw;6EeNjwz9Lp9t+b+nPuq}0%<
z_%ceCAw%Uh5~9OTDJqoEMS=L~<_~_HHyf|u2Y8NN_;G<lEhVnsU~7ZXGc_gr*1lST
znz`FsA2tP7RUAn3>uhwfR&j!(&B6eKB!`6w{mMu!!{T*iBi_to(Gfp>i_W^e9&c7h
zi;T4>SmxU|_dRX_^g>2&xKx(=|5o7$I41k=LL}L!pkuAHHEVH2UDK;Q|E!YyqCLln
zByXva&-v8NQHsE!auLU~<4Bo^@M<qBHE%J&)1F)Pvkx349o@iHEzE_${2`Cv`%ouR
z-o3xVW7;@MQ}yk)-(J~cLVJT5#VJK9G4Ae;fRN&fQt+%&Axu0hi%DIiFk*Q39{948
z8rIY^rXsN=LV}d6-1v}-&`i<s4B4tJk*~4%O;tEy5!v|wvVDQtD^(-&yIwiO-~KAh
zL0)Xw{sudm7E#aUs&Ea}4?CN^HYJvo&Hm|f#;vmTnutqjY|81=THjSRl^M6QrI;fZ
zsdOka(=lN2&_|Im7XJ)A?xk<?vJr~5xw2*nh%kP~@pX;_zDb895*E)^axtQhG&tIb
zrCP@_at&g8C}+=~!{Q#hVpp!f=|pBv*zQm$)?C&4I9xKLGW3R6a%~xJa&_m4j>ln4
zOYuiN2cjh1Wjo-i9D^_^+DLSxLFI?>h$0%+80+z7EEi>hc~*?UWT(%NbY~1iEka{9
zLH(^L3?T7EP&6nsf(fP4h>76%W<xM{iiPSC0P#*ntEZu_5wM!EK@{1h(&DGqLp3Xj
zr3FTj%^HTzR>`O9AmdMBQ?!6FSWkwk1xySf*L7S-c{HdT<T5Q2VlMG=^IbcBoD3If
z@y@Fb$KCt#mQDKFiW)aLo*s-Ecq(+p0+Rd(rN01e+L)=y6M$MKzE~(MlHNhDBVwBX
z8(dYk1ouUnN@8yj;YW06fykMqV>J(yHI4nWsr8%x;NrLa*VgEm`9H$={fBFyXvB$t
z{F#B7dw!{MkNCU+y8mA*lEP**xHUt3Dc+uo5Q+t=1&uJZySSZ;aa~AdT9fNMm3I{G
zs?mnZb*41G0&$<u5IA&gWYOXK5ZxjxZgsZCXOR*^uKIqj+#RA~-qBK*m%!yY(tzzH
z0i+4}(0MV6u2a3O!jlp3@RzZKc?{>x%~%2o&Z{BqGi$!IVGyO6`$+6qI0+>LYCUW>
z?t!Z9#qyozS?&=AIj#>Dw3WrYQbuWkt$nE)R<%&hZb0%mhTNk2wLzU-20U*CMeuay
z@|#bBd{}fdGwDN^Og<D0b)Pt)E+UPDQLiK(;31UouEx#B0lb@Ra4~!6k6&eGr80<u
z8%p7Zy4~<f;9rB)1F%VL1XtT4nS=y4{!Gk&P!S#m=7z>3mf?nip^P{@n&gEZhCEal
zt4Xs6geR>&zmGI|=O>Ai((Aq7<1_ENZd8}Q;#=+Z^ZI!>V#T#Xzp!SuTv_cy7e5ze
z=?SN6Z%Tthjhxx{SvH96>q=A>01X&|`~+)tugt0u=<#|Ypbx+rulPJZWE3kTbd^kU
z4KwO3)jK<LLijt@AZ8M~t2K~H=EE%>Fx3c6Xgg{+fMJAcRVxvXSOx-HeK?eO!EzEm
zg_j=RTjDJHB7oVUW}>hdw^f4UX_>ucI5KJ}h%4x5!uUC%9{lSLA8*YI0LO~1{21P_
z_RaBKGm2~yY)6+09+VnbIF+Tq(}tXSQEJNB?(4~c*o@@4AR)B8Q>MWwBNcpCb`_-c
zhW$hm3t4<#2_LK#4>~~2R3KL)6T8o<7%ghu_Q&fEDdG035A|w?>p$LsH9p^Mqn&W4
z+~cCwwe4{Y9Vdc*PBl`xmiA<xy*Gtr)jsDBkgj;uIQObDvC89zxvsSbpV1<<cBhK4
z_G#_w>RCOJ%Vr0^7|$8Gc!T$R(yT=@UO6?)I#OCt=<{7p)7j{ijgOzdS&F^AF{9@G
zFLUw*T!QuUfcS#7zj;jn{Ckf((Q}5f!{49VJZ<6f-)o!(KSVtrpB`+g?%JQVt>i{x
zlkhZOv}MF+=w3{|=3RPKaCP+U8$WrrZ?eoRxi%~>smN)ccbF{SYqH|*{cl7cXVjk1
zFs|FbNi$bZHBY>#{AN;upsv=8drz>-nEEgwrD@>O_%`7<<Jdu%i2dgU5Tsw^<EPgj
z`+}d3822fIb|r)@91&C&N_W1>F`52u-}JvAwTj?*&ZqC;oSr4cfOLWXGds5LKKFc5
z@pg%Y3a~M8!z2<uFL&m4=CrLNu3>nPEzSDOrcIR&g)=r_IDszd_P*uTjn#yi2|t*h
z*Q_aIPinbgCC)2;&cEs%S*}O`EN2-KtG^<HvbEJ>Tm)KaUkWo@BPj1pHx$CcI_^Ne
z{g@k~2)kcH<0#<|B4DY0Pz0cc2JDt)D41RgcnEa^^M^WdY>9$IJKu*~HIv}k$>U<;
zoC{l@Lm5WNbzbGKytX8gPXb=m%DzpgW!ZK*MU!BNH>~x}fSj8hdlUT018@g9o+|Wz
zShA=S{^1cyWG?6A2|V@8FaUYB$euF5t4cs=5+EK!bFcLINcmbf8|hUwljKvgwCpPH
zm~2Wxy4I)=z(tv-b`5h2C@qh?^a1|Tf#B-9J%i%e&p-F^C)M6lEF7KZo-~{6dsA-Z
ztSWn-FF5z2Vr==ANuQ59Kk0hpaL%7)(btwGlHk%|;Bz)`24PBJ9mP#TlAAzPgPo)a
zizU*vfOKH`K8lfp!pM4{^rHxnYk>$qc1#8!bq+RFT0@3`NH)lK2C)dpB;pnrKyw0}
zSr!%&7}lTvG*pF}_2w!4%|ngYY7D$&LwRXoDb>i;L$Fka(>X%fJ3$K|buRXiR{d<p
z!V6Aa&{W${qXQ;Y9^5noT{P47=V)nPy~J;t?HkpA%;F`xN)70c9sCx`-3tr}R!F*l
z?}y^}SCd^iu$*@^tQC%3nq6&CT81{w*SnjsEV8BMk-~tB%e?v^+&0PCcdSF|w#7at
z>UQ|q3;FdevQov8jV8Z?JgtM-_&2thtKZ3nudhGu({uH)YHjtYaqil;)oT|rq}Str
zOf`LdqPPF%1c;wGC8p%+r1Zj^&_gz_j!y3-9dbUO<Fx<4+^TQ<9g9<p6nsa<8oSR)
zrY0@p)-N7+wO3uOale5R8;(zzQ(m~cfVy$2XJgyFcdsIp*}hAroS6VgJ>UPlFVeO;
z(~)=W^-#!(cbyTJ7VW6J)P420AjZBf|GZxS?-T1}`q6ctQq}Ffhg}vgU$<>aq`N>p
zHPy+oO;<Qx7t*(KTIA_2bR_)2+dt3T+kEm8bCdjuA9p-sIj`eTpShK^=e6cA=?Jl9
zmLy+LYVKEH5uUYdj=w|rvqypM{(tje_<w(An*5KPru$AoZ}G9)hDfcq)ffByx2Q0I
zKA`Geq2P^#rW$euU0|^1K<ag5c`SRA2>1*wEovLlx`b~p(+(mc{DvylH7#S>ZY*Aw
zEtS3Yo&R0<@4LQUSc*I3FsVhk=UD@(c=p!K@UBwAxtMvDB;i72k9;RP#J&!gMWBCf
zT^vK?I>-gFInh$6Z?SGm%BmOx1^t5tS!QpUNpN(r3~7McMRtG)pAu#V8J9ij{ALsu
z9l~+}q+A(QkSd7Br-vJfIOYar*%%b~7Kukx`1k|(YNG0za}gy(=gNb?lf5alTLw}I
z40Jd3le(s<3J-`C+*X@H!r=idP4`y_Z4#gl+Qw)?AMKUc;5q#37-lPei}3^D#&P?t
z%DxNKrLim;1MOgegU?V`ON-jJbCO*Vaei)|b=4f;c?d>bp72*J2>Z%%CVLC}-syng
zT$<U5FyY+$*-!9+w>;NHA6(xTcU_COKbLwgsAlPsxVJk7`O7mMyTq^f`E=EDzV@ax
zmA`EWhN5%*f-<pq8@MxALV($<t^t#{1=>|xShh+qqKcu%1#$RPj^9H)PRX)3sAYXo
zVY0CJYzYeTlt9C1m8L5EEnbCO!{W|0LPH-7DSVF+bqGi-dd0tr<jMIaT~G#`;W#kQ
zCx8wiNgg()`C+EuGDjB_G|OM`W<a?Np{=_oPJXRne2FwvwowuFFpnt?pEV@#|1v1Z
z3l_DEE`kh0kMy=f5hZv~U>4>!RjMiKGTzA>4*zfLy=Pog>$>(kk`PL$O6WzJkXVpF
zK&k|oKnN`aY=B5p5(K3Rf=EKu2pFmu5dtI>6%_>ml_@9zaT*B$5fKpqK}4{C4a<BV
z-D~Z&-u>?Lp7P;*IKT5@`imxv48}~xb3fO8UH_YplT5o%^}hZ)JpedRbW*j`d`F*n
zHU#l$2VrHttuV`euH#N?`7b+<>HiH?`2n|N#;Je*OX=XThYo*3sJA`o%de|Gzw`O`
z#|hg1|8xH@UKf{bmun_oB}pGT$}BQnf-I8HqwhsN?7f})H#Fk8vUg$6zIs)z|LBnK
z2hnAG&yl_y`R}Atzf1U$C8W30&i0p~STDOHoIgV1sQSRx+v|!B!<P#u>zi!miD7PE
z8(lE4t%uRU?CMdnjqjaFO4ya0yiYZE(5T^-*}-|qhb7-<Qv6(J@jcch39}5&(;LN9
zV1>p;p~!J|uV9uuh-@;IH{A(`-0I^n1S4N?yAl6{KTn19KXMLgwR0psaqle>Ly8Fo
zz6!OKEpTuv$rF_U5#>{Tn5Gc+7@weZgcFRe1lCO%io+BW@~9j>F;Olb1YDTVEudW8
zP1Kz&&xNc9nZ~*xz=?#b4G+<|IKoHXCM-wsGP)fUAivu5Iy_h;Au!V!sf!)us4Bq=
zy=38<-c34qB%ceAM}eYgJvu~liY34z17*nY85GqDFBwQF^iYKfwvfbBLCLI&=ctFt
zEH65M6C3K>FxLrT)#_w`iPW79nH7Src}!F+oBYOFkznfF&Ps^eu>H^S#B>p3tugo6
z*Y6v)5Bten;r!8<CLL$G0YJ^<|1=|@)Vyu7ZDQVlB5amH&;|ZfYoW4c3=J68btYGU
z_0u>N7<bIv3DQjJ0zulH2^}0ugp9PXuE6yzP~}L#z%3x;BL^68GJ#SG>7B-R$%;{6
zSO<<{w7_@_l3b|;*)o5KD!`JDfkqxBdX{c*Jf1Ra;y~<sX-{%elUDYMB!%jnIbvA7
z@4_GX`bruyLokc?{an5BoX`we-D*jYJZ(_D8AHR$NHnPm;2H(AIE-opTCtFj#fd=c
zD568lVVkgb7&7{zJoR8*rGYQP^q%b73pFA|f9<q@ZJ>u%cnz>#F5Ue~3I9(|4t(Q<
zyjpQ+T4%P5zUEmNWvSaQ(O?pMZcMHI&1?2A*V#WGX?wqS|LLM?zd-+)$lu;nu=~5m
z<9B~UtNp(XoKGDpeAP!qUL}(l#6w%-4$*&n+Pf?*8}X@3IqJ6I9<?d<nE{fcX0rB5
z4U50O4}B&o26N?3i8@EC%b!5S${8qjaVq4jIAZHem%F)$3bMPlxb=>v${m9P&0Kv`
zuU%Cs)CETk_>pQCPYc8~Ae82S6kaTFJIadr!eW`Yb_&#sBc)+LX{Se*8$Q=aPt-T!
z*)-{ZIOzmUuFH-pAzn@w=FkmRML-m##H2OG7ZW61W2kdhb+~xGVLTCfM|)BS7s-Jf
zD`;EIfV!1@OQ;(QkQ@jGub+=lyzack$Ag7nC+Z&eFGlO4TBc}px@f9AtNRUhRL8I2
zPNMb-JYE0`xdZGNxVnsZLhcWm<1&FRLM%9`cAz`#urfZ*CTl_TR}S-vs?Y+e-r`Oi
z15jfmgKC>FGrlkjp|uc^F~tJCBe{4|G8zIVi&?sgI0N{?(31WRAY#zP`%Z-fc2Zie
z6kf8O-%>V@XI2X!(v7^@RwBC~g_#83{-?_VyY%1-ON~a!;s-9aCiRWl9YEJ{w&6ir
zGhN1@nA%PRqR@?yt%d1BL!o7KMJ5Jxo_@uIt<dMdjRGH52tNL>DR3n-f9*rTtIDTN
zMLNn2%M7WukfCv32M`D(VPIBxl$e|9Y!$f|o&VWe#JU$6CFA#109T*h5P3~Ty~WUF
z>TKTj<C6GiEzq;}nqrt_U8X>n>1?Jcgn*EqBckCLL}-OuA0lp5AA^*FvX~ye!C3CL
zQXNB_yB~Z_Vc^?*`x*GJ{pi|(ZB12@>Gz%hOXuT|?~SHmH|(yF*{NyUrVda?+-mc#
z>L-Lskmj}=-T!I6x?lDx@$`&&fB%pBYgP1rigoRX*d6*e#7_@S^K!`!lGZEKmCZD0
z`fl|~LXXq(-g(78yXnsc{7AvaPc9=nH6BFg(YrpbGNFgU%dOwitXKEgdtWablWshd
zB~tkCy@ypkdCdIcPi*_JXO=bcmapfLgU0TG&3;V#qj6rzd?<XudZN1FOFUm$J-=dy
zXxG#1A5WF#JK|a1k&97(L(})=S7&eEryKsKZvzOG|0ZY-JW9NfrgQfb{pr4_gNW?U
zcqmf+S!GqYF_&f9HjX-Sh!@b>9&mv=M{sPv2evOqBKTXrs)9hrCB)aqN;|GwMZ7oL
zd-reXpEj?Bs8v*0fV$5+?ANLCQ<m?pix+wSFbA^IZ8d?zh4CPq10cC<UGUz~{juhb
zdLRC#>U1lyUk*~68A=MDl*7`;m-&{yV%SKjx6b#`@lLU%X+=9L-lTf?^$mI9BTZ)y
zUQPF)tThWT+IFTc4XgrU>~t?Af<T?|%LCja*#$x_taya=&V`m2h$4d(PO28sitvy#
z^8jm&M>;(3rys;YkVZj)SzqT29qp*5*U+uFh~rbmhm@SIaHcA(7Kma-6`(rIvZDdT
zL~VsLv&KUr%#J#0MuaL7szF>Bt0`LHgod{QH-_{GkYx~=UJ`oc{7Dpgm5jLQ>o!%I
zg+Xh#^kOC{Np-q%v?o(6ISF`%1J~xn7f~kRk}>rmb`;qc?<&rc&=Gxn4GK;+!H%jL
z9D#qcObh`Zcd0(o_B!3CG;6>Di7#ZsILQ7Qx?Icce4GTHvj#MqB{W5<XP)uFuaV+Y
zi9H@OjRx640h$iYW?&#5<f}jtDWH7#xNIHHCQKAN7KmLEAXWk06j()+rkLZ!E~Hu<
z`$#lblL)3tD;n0}mW}2qP2WAc4amyt0~&63dhhnDl!a?+bgBuIAx&Qf2NF>r9)SHM
zb01P+W+W@{55N-&RxnQp7{Y-%s?}%@%;?5cNV^4?IbkELUp}#IB~Y*^SSVJ=OmiBi
zVl|f5@pP|d{`K<eI{%w$_nA`bw!d6wPJ!z)4~{j9gKaFsBXlc%oUALCIuuL6SyO;+
zII&`pB7_yW@^n|~e#gBal%xRCLC0X}a-QoY9j8vK0!%>1b9EW&X5jtW!QwT~wO)H6
z|5XG3eD#fesm$@EQwjPynfJ{30m+WvotwI?PTeolm#fP8cA;pp%Dep3{cG0!weL!w
z&1F3GzD-SU*?)R*Z5d41Uop8_fb$soZk4dw#2t9Cs6k>rYI9qE3wP~7Y?p7_OSL&W
zDlF>a(Vg}Fdq%sXX?lSJ`mSc4r`GTKsDJyilE(bNi<66ZTJcjstnBG06BAQ=H{IvV
z#TezyeH9`%2JCU~QAqBvADl?#uzK~5_DXu+y!oeD|E1L@>NeEcL$?}W03hJ@-%CSE
zL94y8j*k|X6s%3&j?t~Y$&0V=9CJ&Id#uoto~V)F{~D0Z5D<`d{~KFoRC4mM4~M?U
zf4p>ksDY1b%E$Lw#{<6uPCNQV`d2aF(|UeyM_Wp!Yx_;&w&4z72DEB!<C%zS4|gT6
z>v=|9mg0NTHP|U~7_5%vN-YYToxPVQV`_=c(UGVpQF57(BM0Hu=Pza#n6iV5*q4`;
zewY#X!00Ky{A;@=89tR0j6_O=IP0iCh-b!V0%O$RP0LSpx=$>tT|n9;w%B2s9k%91
zI2jGdP*BQHI~v3br%o4Ps{>79U8B^U<Ri&3G#Q`%nW=Myd8pv^V58U$&zltL@~%8H
z^O7x|_v*4$kkz-K$<imu4t;32On9mYGb0o$*KPrpCx?j=p{@!@c@U(!@h`eRaYDCB
zjGfEd0_bIm>?E);t4O6p;Jc2|Ow9U0KZA0)(*zzvMc~WjfD|o)O*1{hXB&_Mkyt3K
zQ!j%C)$5S0>VAjb<$w=^2Kko!1kesStT?bwxuZ4~$>J?v_utCmI98Ut$+sdOk#a36
zd8652n~naH<a<@lPn)FZK643kC9c}GM62P0@9&2m7-t=ZWX+wIPPekqk~>8wB{s<G
z%3`vq;AI}}#1LQS%u<PxmYVe|HQo0!`}dOeu9sB?$D*Y8%v|REeni$KZPILz!Ojy-
z{Q9Y*JaBN<U<i1dQr?=)HsFBk<K1VPz-!^_(RJ&=ap*3bq3g@Tr+Q1X@8^og4(5~E
z)^2(4E9YmsXWJ2IIAbxD8FK#way(+J!pP~nV<{_6_vzzrch0pR#$VKI0KM#8fen^5
z;8X!V2)uRRul$r1xuksx{1eJ!F!-gT=+_@U6Te>Kvj~0#q*2PIZGLJsC#6fJJ^A2N
z>||}sH=}DZ@%f)UH|oT=M8E#dOq1Px_4Lubq{DxFs2Pd05<gN`B<JVM^g!LzjjL93
zs4E4ho;8L=&6eig%NnysInsT=yF{sR8s<rN!2|SJ?5Gv_CQ{d7<Q%Jm9F4xI^6R`R
z{f9`*9ddrg-7ekWRHUv@OkvI^O19?+lN<MkD)QqeEP%xfk^Xo)S3d0Fg6~}wg|XzK
z-l-|)0?f<thkbOMLq(G7<@7rdF9uen9lin>f!VvcA4tx3f%BP10__)nLsC___OCWe
z%>D{R4&1F;QO;b(@6N@i=6*LkrujF-X<2vX{OQ%tnTdbwdb@P-(spmS^Fz!vIqK*2
zt?ym4HwEw7d}vYUM2bNe<oEpDnJ@7_Gla+XpEQ|074vTg{MQL=y34k^`%}?gE?wv^
z{{X}(|MUHS9aH{`$9Jp0AqT3xnd`sazvh8e?fAdWeBJ*C-e&pKasP8=jZIE=`FFfk
zb7|pratBJz2WT!Q6*j+^{`Sx9gs6O^w@5~=<((DB_3W3x^d8_B`UMC(kTk^|Trz%b
zJ;lmht5VP+KHhO&T{D(R)fD2M(k7>wP6|`y@_1w}Du!zkQoO=7CU({7ED){m&}634
zg*5;ukqT1xdG%A}7kfU|ozQmcLks1OMS9Q*dNI#OnNIE1uKtHp$++8muZ{QkCh#Mi
z%91YWFlG>rATwY`>xwO+Ngf-_%k%_Jj&rM&OhT@_h;@oYQ2SI(RbR(+7|Ebt$2?<)
zGXPD_1f2kE<zY^J_}qdPIwCx@)C-jMH{w0Sil}g};WHlr%Zr@^XGo-2`-D`xn30|=
z5_;h0zy|MWWh@B;=*rq9UDgTJH2O4ysLq1fz9&h65ZE9aB7<BNOctZc1<5ia5sU7_
zgo1vR>0v-0%)!*lM}{c^F%@*t;@QnnDMJCJT7RKi_gDW?$uN55$jy*=aTA1{XF?#M
z&Eq(@5u{y?aNtECtmX-2g7#=G_)ppsI9<G`U=Ob;nquUp03r|Q?Im&TX!O@!B@n`e
z3UkIxgW%RDErBI*x|$<#@`S|EA^tS%9vYe3H1!$%@kX9-qx|x;J)ap<=<R}rmaw4c
z=)iM?(66nwG2jt_dKc)XLqrbzG#W%K33BMyWfYW?Pz29@^}w<EDyz>>l#K5;cO^NR
zVhDfiL=X9|#Sa3%5f-EBn<7#OEn<Kj42BdmEP2R@M1^6x76?c&GCcH*VXTv3fXvkC
zO!5QrP~hkM+cVJfkB?|#*KL>nZ`fX{<o<>(R{cc$IB==uAyWKg7ztPC-OX`jCkIcv
z{YOl~|H%TgB(M#sHt+ReK*EX(NRME%w1j&mIp9Pb*OrB#3egO}b61cpDof~;SdzRN
zM{KhIfkwS3ki(D(Ky6rZjx<s>#z>#Znrqq`jl=zr0NG1P8()%0eVwOGgKDO^;rAg3
z#QXK}GA4aZ#-=!e?1p%+g+74xMRTT`Ap8~Z|2!D`%T|zMXeRg-bEsu_e#G~xPGzIJ
ze7kF#;IOwY0Iqm5mgLu}d)yAtDTVM#q|R))+*1b$JUz)0C?Vppz(t%5FPY;ZU?k9L
z=-@39WDthu@_b>Qx*X&JsM)RReHHfSiB!+iey((3`VCpAP9S;fN-cno>pUB<+HBNj
z^M_@u2S3y3YvzH|L@V;qpH|fe{gMn92<Fud9iWK{9mBUk`wKuZEPxPAAh$2o?t6_7
zS4aimFBnJ}1%gyA=)#ES2~B{#*zrA6q7xEB0r#?xI{~SQf}}WP9T0#5mT?e|Lj?k2
zgnA%T+%Vo11jncX1n8kTt(joRt+o8A|8;0&%K0tWgE!}PS_TzZ>d_5Us3}7uvu3@M
z7}y;@`yPAi!!PY#DInZIcn||JPoUxf76qR?A%b4RE6d3I?5TS-|MBXT4JT|=O&0we
zE+4`anMC?K>c<L9-tWj-zaBCYoh2Z3L7>V_#6qH#Vr2oe3&%ENifMTQ30k`)0u7Zs
zkwBa7Yt}*(ARU;>9RnJguVjDM!*-`^YBCl6r)}T=F$xtCargGWHgf|cS>;~tQD(6w
zS^vSUEr7XJ3l8~<xprl1AdHf)J+(NxZM#4x>@$$QuY_{N)(FWIbq9Th=Nw>$jGSwO
zm&-nFPos6i!99~1s14eVcHb!P+wlAl?po~Kf|j**ho>i6c5V#rrJMPmsaWASRG^#>
z4rlMTM^9ODxOm_8c|vc+r&>|DyqI=L@ElXzX2A`B*4%)Y-WR#;9q?RrPn=#7>OK*p
zAQp)iPWF+U?JY2C^b_h3!0jKVm)`=Nd+(ly3RPJ7<uF(r<*6+YL&inn0F-JBa*QB}
z8dt&Idj}9pfz*!*^i|cy#uE`}88WOA<I4`VKK)e)4`vY5e%Z$cD?s)Xv^VMdD=fFQ
zK=EwgNQdkiY0zZ4wTH|sgDGFlVLK`$0*^3+)o<O(4$~ILY7`Ts^H5$TfiTbA*QfD=
zKI2vM-o!n_S66~ssZTbtRA<<sg&+Hx{t(OlBo|=WINV?7wu$nTWiVNllJsQm!>viv
zD*sJ=e{F0r@Nu)4%C$*ZcGtRfwJ|Y-e#Btvq=`@QTB6Y(&+cwHUopJCJPB`8<4-G)
zaZTA_@EL!}vv(<K+VEW2&yOlfqvBB?@i$L3a*|}OTZ?k_J7I@4o{pT3hmFv0$_f@-
zrYl>JeDH%?!G6moXL^CKDV<4gzip8sD$~~#Ze;G7ZZe$m_D3Ezp4ECKmqyhZ4PO-%
zO|+oiz_zuV1vb(4OMJ1_!Ol)vU`meLTat#>tM(<0e5iSza7p8A9aU@<IBeUQuz?hb
z*pwtr^8h%MZdhf@&9eZvFP$n<4w=XZP)QBV2+)crT<TI>aBDA^WjrfDc2YN`QKhde
zYiM~|MZ^vc%sKbpG*F3~emlzzn3vFUR*}Byy(lih;S<dz|AvmQyF(u?J){d3gc5hP
zSJTUNDq)$}S9h0ryoRM70i3btlomEpjT?Y+c}3OOf3TXBS=n0tlsRqglC*D8@0dsQ
z<JKsO(&*zrCcU6FHZ(f#A=YY%#(d+zmiLFhkwQP16sm;oFYe&b@@;(vnh-Y?i%AY2
zIZ1No=B<S)5N?81%tju@Sm{+UsUsTFeWWmERNQFZJKD2rl+dl6@s<BsRT?$VB~bn^
z`^zkj+5j`a?`@kf0Y6wXI&_jp^n}|EfHWQIA8EQ>66WHUW4l-ocZ2kPR&U{0ULhYR
zzqd(-3R7&in-2i=ln8r&X?2g#+59KKtu&AlBmrlPA3EyWj$liL9kE>WuUbZJYv&Z3
zRm2mx@PZ>;N$LF=!tcQp#Heti^jPtAswxV{u9k4}mL-afxjm#Fl;}jrRQUsNU+!g|
zB?<~M1NIOI2EJLUU04AY7`-I$xtWUu`BOaEK9U#&%z12a;TxET&oWteGF+r&*MRXT
z1Ob;H4KN2Zo#m-3&5kPqjdhpgIzS*A11=i^NC3oVf&!uq-_D1l)v<Yi2L}L)o(LX<
zu>yfzonDLtd!Me`Tep!6S6z4s0AGKuPf4LbZ=Inn{-#vb7F4K=u;XBhOwcWgc1w<x
ztCB4=GZau@Q(Um&Wu5Jrq=Y8@j$2#xT0a~cK78m|Tc_#6&5tR>l+DVco6cQyk83io
z-rt>ISCCw9@N;{`ACng})tk=srILO+RnwhkuU6Dva(0mTp6vef<HlW{{(lX-%N^z0
z#~(8He5X1+QvD`x-@9tFURpfgtvs{I7z2xa>o&nC@L`%@kCUA-dUOO6B<Q+)m`*a`
zB2agz76C_4s7e!Bk|cr8*=#IpBE!ZQ#7wbZY!rdO`2;vSOpv*eVnUn{@Ebeh>v;$u
zm8FnzkYULsD-LVHizCKYMMIs9I5vbD@{1?zN&}ukB>fW?Sedbrv73KPJ9q0Qwf8-s
zP0fZ33T(1Z>{#`A%q}!n`MI96DWK)%?hn@Lt-G%Z$te^KGxII&RsI62#T|L&-=h2D
zvSN$(D7?brU=5nl_^GbEWr6PJCaJZo50L~XPY%f9+r@f>2Ky3b8M&lDA;JxPivj}1
zARP$X`PTPCWZ^{%8AT^KtGfGTJ)R@X%rz2955GMBN%*~7Wd!w|^C3R?O?yZ$#AQ9u
zOUG!cj2)gE!gra*fD-#_NSVYFythV}%^o8DK7{?yjkh^ET4{GT?k&aC&KWZO-GHw`
zD|(UcD&8aXOL8Yu-V<c%UD3tFX1)e-anLWk3WOg^OPN=68Qo%WAf@;}FOI+0K%><y
zA1~b);?#i_TV5<XRc3|bvFt6a_~Rd;_goR@$&zupwWBPA7kUAYRVDUm07z65Vg-h~
zz$64v9udJ;p!a)Mun;JR;lC3gj~Wg_91owrMlMzvmjmnXg8@~LdjbQ8)P3@13kiMr
zidC`<fjYp0w|C63lch&})zyz!@*didE!6N6V6}9ir3&$yX_$MUZ38zOV&g^_z|@6-
zIqL$$36?@TP6n<JrZa4|1n4X?6{4_cQ-=Wb_gO|}LUl^KTb`r@PnE0d32PoJ2Vg8`
z2tVW7)g^)N%|_R=ofIldY8cWLD?I%XhLn5WBN^)i4oGbHWR9G!C&8epf?=o@4;pi3
z0?^$ngqCP`L1B*d#w!K)rig|N_NE+4A#9caSk!I9axGm!R6Q;yNZ!BsP=6%Xa>={G
zulZq^3xSjKc4za|!}}83O#)R=QoSa$ZKHRT^U1P(2D7DaA_k(xG8Q>97CAB|4!qRb
zP14F;kd?qVdXsVb;T~2{zl_V~YQ~Z`iFoOIJKIb(rmup1^zNrqgw)z=f91om3(QMt
zkpUfROZi$@0JtZj{0W=!c|))rr#s;)Xcz%y$pNtTn0XhF006vS$Ojk|=&@kS!UA@E
zJ_5qn3L!rs60QhbY@a#Md(gG(Oc~4sae11J2-2A?_)@2fV;=^+N%S}wku6KeLMvtx
zB!Xz3kQzskATBGxbk+qTs5m<7a~%-@T+uoUZW$O21u)QMtw9@$Z`w8&(9k_>GNWGR
zube63MpZZT1g#GlyL!rZqDc++PJSh8^|7NhN)_pfc}4p}ZOt6daBMioMD(dKc&lbS
z;4+13V{vSb!Yayvb%dk9qe+M|ateWFjFG^v#UQL2k@y3$8SL-OsJ1H)e8{0D%(+_!
zaW6gr$^L@Fre$`~^Pk&5f8uDmwMzsqxo(cX<Hj}D+R>Z4B};PL&0{O#AC66fSq`$#
z7s^gsSB-71F2G**7{Ynhs~NnfWA%3%K0DsJSLWG#Ou+cpPmd3OmWYZE*!FUuj;IuE
ztC(c`>c#NNlS`#rHT){hSVw5#s(Mcr-&>;1TStxM-u-rRUlHc;#LJm8*J~gvgOSa1
zIR{(rHN72~9r-Z$QhLt+-fo8hCl*Gd;I#Rz0N%3$+GYVZmY?qnU5;3y;CV%<sQaDo
z&OK_%d(7Iq>$xke=gtVF*S<lATyA;&sq*7HTjO^*9f?m13qHN>^^Vt+CY_HE!hfAY
z$br;z8&tZfZ?=2Mx%v|P2R9$Jpz23^xtBkh-sQ4**HUfvN@5c#z+UmsGpMq)_iQtN
zygXQvDBd}o7-ZVoa_!xoCf@n#Z=1VLjLW&h2JHRk^xj23FBCh?&Z!Fij8eBeShwMg
zyc1mx<5Y6+e*K-*hR5QkW)N{l#J2v`_U*~arOS7M)}@(#SPg016@+Xp)C?f)s_u&+
zR5!aA&uy;%61hNWE&8z(9o3}tN$L`J^x{M&G~1dWyLRD^_YaG;x(og&{oJ?7vqLl5
zHq*JMW@X=&FRlj@0~Sed7nMiKu8iMk`D_{;SY{g|yDu^&cF6cvX5yhZRL<IfKWB8l
zO_B|7l)S1V{yXPGV;j((cH0kvG~cK4E>3H6y|&2)0=DQ#RklmQuL?ilE9%K+a)Mn~
z!AjPO8mFvBAFVWwF*r6+^5#<=IV*(%+|i*`Wh`}v7$f|oUt)c5d)E1tIsBP^5X|Qu
zBytGW^o^lG_!q=t{R-{j%_dGc;&HSXS>;WVJh0Yv^$;gT8>X3|nhJ>mvEWz-qO-s<
z!<A>EmI;jcfG1Ue>Au$pbK{v4#n3RoA5tk8u8zYZvz`v1*Dwe-PE1rnb+~*MF!mXm
zh!^DI>Bai!i54ja;seh}uN-XF4Ov}lNU|m>Rv-@-=+i>hjj<8Nq>w58vNIrQ;3>CA
z;1En@E!hy^3D0)BjOVxyTMCtoXQIG*t65OORX2vy0h;&(LyZ@hz%|&C5-22B0BnyK
zmJ5ha+L$^-u3mmSCzxu7m2}t<?dnf$Sn@2OjZR%0Zrr*$<oSAoTJg9)j7z+E%>^_n
zCR^G8p^k(Dc@RPx=>F=UYZy!%sA#fv*gn9}%ty=0Ew~q8(MG?F>@fF<j9qdzz^QK2
zs*9FEZfSxfK<h=C;RZUV(s@h((~@R`$r;8#li04t)zQoIg?ceu8H~;k+Hn(4)jmLB
zBBuc;$O3;03G#|}P^-DC5FYD-9fc}~<x?EtI8h!Mge*?4+YI$G+$R{isUwwVEhfVb
z*76J~TL#JGzCHK+KFEf6ReJDZncV&S%pWQ76K#DCPGaahQ|#t$mI>b|65N%R2^=T4
zx~Wn+uGj@jpJ>UULuu6ge59nCUNKq{3>&f#B1+-~#9#NMES;gCv&I3k{Phm>p}@BV
z_A`4At-fyQkEj?fd;We|<=gbvVOEM!Ucz4V-UIm-r?rOs%USZqX%{Nud!0k>YK}aY
z`$*l7G}z;yG+g-wo$sHNv}u0u>3Fj8{f&K35s|5TPQ1Fc)4P86-rdQCH7C8Uwt(zr
z4w)IW_jKM?e~*S0lT9rP=X|@>^P8&0_upP0wn^el!Ckx8%B`pG{wWu2PkE~Ka#3XJ
z(5}2)>!G!;;WN+XBZm%6-)uMhaQ7K)K{JSUr>yI%+amk^t)C-<EBS=B+)XW``%e8;
z<beGvBRwm8w$S@t#PODZ=WF}wURx6aPWnU^n#f5M-Mti-d_G;F#$*pIRz;4&_q$q=
znL(=wZ;A3o0=}qG(BIHy(i2gl#>Oi7yHgvli!ZyL;B>#9`EmFAE6F!`XHVvwe(|yM
z!E+l6?!&Kd#~)o*e5IayB9Yh{t);y5aLNGosO!TO6Nmn3H1g~YbG_kBXM3uiWW3&D
z|82{gliQYTpL;yAmrOq7J9v4^%V9m`$+3G4yzVW1WQj;;|2Ky#@A|!qKG&IZI{LhK
z#Ksd%FVik|*qW%R3D$QCK7U{Tb6s=imA0!Nq|rGl<9-UswiRS!z9-ha{<N?A9@EX<
z-evE5uGm0Mh{u<ePE7r<?(-PAbrz9@yI`BzcEls)60PX*&BknJGn0QWu`G^S{k)p~
z%htTGXcOyj00>?po@TtC+w3tV`}bJzu>cLID=#``uWd~&j@Zb|yM$4oM{Mh@u;e%{
zS@9(7sOp*oK#ag(4gxPn!G;sT;<8p(QpvjQ^Ip=~hJ{40;ae1Glp@QSHlD!_Jr#tN
zI^$P6R<0XciRnh5?>oX{*?v|J!M=S|lkK%48LQ*##tCC^+R!ql$jvGWnMJfN5v-F2
z2sQ=YB&QE-aYU1?I%G}JrUKHAW&|@7P5D7pg=LZ^TJ>SeGQQ239u9@T#^g~kRZ>|&
zQOF(1Hflm>dSZaj2;4UCU1q>&go5?+?EN>ntN(KNB>6j<(l|1qOz{m&40La$9$0ye
z3`=}Ma$=5fSX2eh{t^%h#F6CD-KZ3bT#1a2I}w4e1;9rH+v7a}1;mpTXf256=ysvl
zqj9~BN<e*;^W5D?N&na~IHYtImwJ-vVk=URNyOib_XsPAC3#^WiB3Q)C|G$3Z~=Jv
zhje8+1{P0h=XglYmOxZ7tSWDoae^6(c+W9jj{%Nk8eE72(KAkHuuz<KJSmD22KFhM
zJ_HB^6O>sZg$U-aEHvvqK+<WiXajMV1uQ42l!R(YYqAx-E2sofOoTFtAa?*P7BB~p
zSo)a|&4sxWdc>C9E3uMZo)|zM(QY+wsl%2wsI}*5ImxZGo(VONcx~~mglMx}-@+$T
zJ1*_O(FC-KgEP@o1^uw}ybg`K=;5d!tLwuQ$Hx-A;tBMG8V08ef0ibL1EpLRu>Gvm
zLDs{q#1Rc7ZrP(T$_!uLXrgqQIr}~KZ%A($baX&5&)guXlKMB4{4?RqwmP~@MVKw+
z?V-1TrUpH*R{Sbj>rDYgoxz+Do>%@c6pS-Lm`tv<S9f%K9Z!G^5-e+tR2K>2UbsrE
zlgT7*T|Q)(D*DII_aqpkVaDStDprM%TCR)!d(bogZ}%Ag!Nys5vx<@irip?xjr8dz
zoOal+F6`FdR=yjTjtCsCrDPUKcGt!p#3*p5Sj~rcaP3<azB$5Sbk2*+6ckHyrKZcA
zAu*P-Z_&?C$2Nur^4;9v{e^PrsG>`5RgttVF7EzD(!x1>8g>1d@L-AzySmzSqAz8K
zDsV-JBgU$R0GQTFywvqFNgNDeiYExe(BJpd6<C&VmW;5+Mw61rL2!WV4+YPj>wUv^
zx`GP=wLevLIxkg(3cia*_SYfkvml!nDl?TATTFpEY2)g_*$6wDECP;m6!ut7ov8$)
zkmifM00dFdk*z9xGqm$<m@OIi1FONNYQPv?e6eh3iX{zZ=)ej^CGiBQJoUgN0i-rt
z5KjzcxHCM0<s(%l7}D+_b%-wR=5`OXnrIE7Ul$&y-6YPai`gtVn`p}aiYKDgA=m}w
zAM;K33tlqZ{UUdi?<bw)IpI>7)*^D2m{5unc5T6l47IF=%8=z-fkNOTqN`|%K5-<1
z$zXqG1=tsA{X8Qc1T+FVusaZV3k1oO&{L7Yx?t=B*z*v8x^4&N2mu|^0}*Wu((_<4
zg2{Ib2a=;navT}VAdD&n1_u6NEYZ+8>MPo)*tX$G6Re3sF##=<bOy^Z&NWI1Ph4~-
zih-=8YB`TX0FFwwM2I9$oNkW)nu&E9-u^m=qA(n%Q)F6~<&l?|)pxk|$N6zBqzn<m
z#a%Mppc(70Z(+jRkcDRg^~D9-0+j>A&Pk@V&_o|SuI{KJ&f07O-8OggRKFV{i|g(>
z|H>-LaLkGCMz1dnd|cpSyX{xE0yewi$NKN{C;w|OCF)E34*<>hMT5((BccZngn8ND
zq@Y4w%f3+CKJ7+SvwlIHpvuxS*Tw^<*^$jgc*$a`T2-KrM^Nj;2ZCB;l5G>K10<So
z1!3%iR^Vz6px*)d$hz{-5*Z4{F)o?Mk<hj2YO?7DfrF2$Lu<~6idaG>uGsS}Xm&Bl
zz<ZipkNMcB4o~JA66acR1rp4_aG{A_-g^f<aRIc>ABq!t<3VZ(z!3mffr{}vRNh!0
zhKA0j0dTZJxmEod5o45Q*BC?O!yzu<ZqoORav!FWA!?sm9c-c)0SrOCSb}ApuEkuF
zA#Q}O)m1}xU`vdY^ASyuP-nJM63&tpNC0anA;ww=i5g$goWjY79E>4@4(zdJpye=f
z&QE6srE(dM&c*|vF>Ib06#0kPGU#9|IOD{DhZ6xZTd))`^&s98G%hy5nA=KScr*Yd
zG>ud{IZlWJuJlV7eLQgO*Rm;gvL;9?-gGBI6S>6;qaauE8X^F%D>PI>hE1FX`d3%2
zDrYC!m=-$#_=Q$zUAaX*<a<Thoxf2=4$C3K0Wa=10xDDm-Ps5w&J)Q(6%|cNH0TT2
zTZ0B2K=gNWkc)2Y!wO5|Lty&LG@Oo6jn*VP_!BT(0E9qMFskE&1si~~DV8a&v*-Yr
z!$KIhXER2Cl%&A~Hfy~P=@yV(rvr#jg<G4P)RkH9BFUyr53C9U51U6NphsCg#0#+5
z>V19Yv1TlhERJ!(z-zdi6Eb$T0HY9E%!7>ZwIDC%AVA1y`s9JIj9yiqugjHD(t%HP
z;A0EIWG?VE;0_-Dk2^TG=SZiLLH7UFPzriC|Je6v1pVIf`~n~Bjp+I<>Qwly%2^Ms
zd0DAp^lj_Uv32wQb5<Ag$A<qXZiw{lM!zv|yM<EC3*TSPR=pTF&9kZIoGmayBOX$(
z@ShiZfq|KKC2ZqK2(z>ajKUP^qc8DHln5i)&EKF|gjr*dJRy-XbbH#%XNVoVFJhut
zuaL&T*Ob#sd6rrj18Q)Sz_L~ML?uQ}LY&o=SU<xcEa}80qt|@#RXATek_n*KvLLkG
z1qPTVa^aLrqQVPOyeArVe5#~<kpS6}H9)N*B~V)=lux-}#I&)agtS+)EU-I(1Uhge
zK$mt&v=Zeb#n2-_@dHUIfc)Uj`+>`=3l!OLY#kY?qnlb}))Mb?@?F(YGJ?OdsbgGX
zdi%@a?eQ5{xn9qatNeHE_D<x!OCGrhFla1=rCj@7^Z4Dk8TA{LdW-_q;2*TSX5~h7
zeU9qtqgVB=nkIGGq2Po2v{_Ux?gk)Z(w`f>_0e=N777mPC9q)u^A15B1W}IpL%bNs
zG?L#YHpL!qSlVSsci!FTe=NM=+sL7Km39sokE%LbtO-1z+!Sy<Y(&!lR{CNHy%pa@
z#@?aBNa9r35ztevq&eeIU|6VR;CsID#0r2J0s9+*w<U$E(@RAIc4)0?BPKB3HAPHp
zKj15JC=CO62+7BSuXcT5u~$$D6lTP<eoy48a_y+%s+?1#@DmxHTGqZWmlV*Ui!eN`
z)~E`KixUFi4H8Oy$I@RgRl|VT>Wl?FP#npcqLejT#>heH_(wT0__R?$>1aBt%ydy#
ztpRq^VZob2B>AII1*C>E_l~n=sf6)Z&pMcEBw;<ifC@b&IfJ2dwBa1FDR6R-gTST<
z?33us19I5rze2DU%fv3!W3%b%Q*v!T68$~oE$W?thN*u-`ep(dFgKBx-{iahJPME}
zf6A_VqJH}O7r;C@s=m`X?yqK<Pdj&TG>#h_-|D%LH;nHY0;qh{f7LnvYe{n8Zu-J+
z+4KGY%gg~}m7Kf49DuQg;``q%{AjuM%fi1Q(xc5~eI$W~UDv6Dw9#?bFc0WmnzP~b
zC?NY3#IhrC2Fnimr#k8*UZUX!*|rbZC60ud_r&Tun?6r*nd^7H_Jmu5DZf$|hx-+x
zM4m4ek|gfmS0{vzC&#@uB^|wUZRT1fV%a5q!(mvnjOBp8XgGa(&<>+^>M;}0JJhx1
z)wnNjfNTKQs8z+l1cT^=EE6~d7%VhN$%q%|7Kp*zZj@(J-1;Idn8t`?fiaE#Qf(G|
zeQL!hVdxyoFrFI7PmAO*3N1;T>ST_l(D3ei?yGo@Kj<<Moraz|Dpu$$AZxyG0bFZI
zbVnV0bs0<x5PR!BQ7$!+IZ`fl6a8epWQQT$Y^l|tE%5!<5fI_Dy3+*trA5ZyZ&s3i
zZDv0aS!QbAa_{b2^{_oQG6yZWG~tP=irM_x42dHZp{I{|Ld0Tgi|yxUtS`P#S0W2A
z=AVKQA1??G;H+MDaCYFUJKSP(1iQeHUMSXELIl3mL%Y@xvc%YjX(!}r82NMf2zIcQ
z7cD8Cg2>sBdQ6IM-T~T*O-lQ-Y1d}e3#!l9@dxm4-iugZ?29i<1^;HMoG*Tk4Fv(m
zhe9nAjB{(}$f$>E!}x~a%H8tGGZ!lWgII)<Eq@Nno@fyQDG^D1y5K35gCi-^0jXs$
zPlUl`x`TlI1mi6Mni&*<Mm`g;sAcxktvC{tH3DTU%9^4sc2|ufqfJ$~Ny9NAq-uX1
z%vlEo>Jbhgk{$-J1L?jdLx2aEW>Wi72pbh>ExVBO0Fm4dm=;WAtP9dztZu6;u{K20
zItRV_K#a#xC?S(uvi7s}h}k&+4#<9_Ch2Tw3Vg!qIHsyN&JFB>TGoJrhgMKwTuKC*
z9@H|4t=$!+I?KLu=1_32Gky!cAjYtxN(*eYZ~Q8EOl(Ch-?t7NkGj23zhb`0=b82e
zyN@G`i(8)#?z?jCPdAGn1ChnMABPnddk;<*q&DVGkR=@!cAtyTc&|NifST}SNz=C4
zYZjqqe&<nBL1fjMr?2IC)D-2yEuWOm4Y!e(3ST~}I8}X7a^)r{VD<>Mm2%I!v_FW|
z)H}9`Dskbev(1lnKBA?vkxwFGT@0#mE@?<B^3BgRj~6eFeTcRozHlCYuscqoQB7b|
ze7|>BazNmt&Dny=Dg#zA{VPgizR2v$j;MF;2W+}+qcxAm+`lWPE;dA(+s>^!7dCmq
zJRpC=#HZM#m#1v(OzdR0J$&x+ejo3F@>m@0M#&nrun)URSaO_9VyA`G;#R_=#I)Hv
z_uaCNnuTN5o#<+k)xUH4v$nxW_dl-{*>Ul&cfl2hHJ88m#Q6p2KejbXt3TlKaHH!+
z+pGXg@SDjr`Q-+4dFciJ?QLU+629sk*Hjt5DDhZq>SX3gMbCE+uU+UmxGBn@OeglX
zuEVyTLiRI}t#nd6_ci6&?i<e>EuPl1Yd2TM4=mSA=V@5koTXh{3U51S;jR&#c+b*4
z+u4Ku@1g+;gZO=e`j0)+o6T|%XD}+#<(U8B4JRT>ljDCn<o>(?7V)gLcgL<-q!!Gn
zad6Ly{dDI_`DUsd4^37(T3CT#ORlRtQ!iIR_TTi>7^Y<JL%`j$oaN>xnQ$fc)f;aL
z(|#0=Qh+HbKII!dC&cdzLDvsPhg<oLr8)cE9t<yf9u;9ZPZ8=@o%Osy*BaEeVbWdh
zONHw%Q2ax+lleM3w35-X$6a3%W`xAtOm{XFDPC1b$D;9rH(1icl&vIj>ATESbGSP2
zllEAuTMd8}jcO!bjM<ywwW6aU0^pNkz^jML(e9KZb#>T~e;gCC5l&<<P`Aj*vn)v%
zzE)?7j;<UAB_OS5(id(~y0dsYcUa%I58oxYF{0!)_NX}V{J8zIIHc~Vplp_Q?0eSG
z)^KuP2$xL@qimGhO^cJqpVS96>)|>>jj#2GdNFIYFJay6s2jS}81=a^adMb;m5^b>
zT$!ud8xQ}rcQ^bZ7vXVM!^i7t1MZ6X2DeI#!sPIq{TaK?z4y8YFyB)QkB|u#I4c>H
zBNlqGWAklhqnu3{5@WufxBJvvPc!qgrm5M+OIS${3dnxg%_rl#V&?hAu^wTi*SrL9
zC1=|hD$K=f!5hJZ4r6cH0M3~Y16cW8OsNoG1I&q;cQp_iikHay*Ep6?)D;S-fP4lo
z0U;4!MiW4W5I&SkAn-I%ROCHw?-7m|NLHGl6XyvMB=Ck=ykGHDG<2p`kB&m(P2ixq
z#G1%L>$0`5P?k>~bDwTI7@?V?D=9F>C=8916y&g^`IrUb{`{=lU7mXy!jy{&iN89M
zV)ve2CXohJJ9~`m8j)mrldEtDmhutQ4G_W1LDh;20|RZrWMeLbv}T*<bQ5W%n*l;Z
zAI9^H-)0N4Qi-=0&WnNv@J;rz>G$~ifYff!b4Mq+Ui@;_FW~l<;@v%MDr?`3hZ}Y7
zqp>_E1q6-Jp0dvn!cuMYfZ4_SUm7>X_lFrosj0<DUbX{(`$yaKZ;vFFiEeGJMh?9(
z{@iusX4~56)d-&N)$46%?=CN#-mxWVZhTQ$IWII@&bl<HME3QY1k)o@7rv0k$C|2^
z_C1ldiq2=WtUa<2XxvD<lpguq4ZkxeIKuR{Afw~sO10Q!;lg)^A7{b6tkM53ze@CV
z?`y5a4f*LZiI?1aRcgtPxm9n9#yXs3io*^QUQ->+NMmXdU-oTqQoQ&4kfvTy3MxJ8
zVuX-C0x%OJBku|Q3ak)boFzmyX!Q`T3u%A1p1~ejP0qWtKS0?hci=r-)@3X&$@*_7
zr#e96=JvMWhR@-L5(d!BO;nuZ(IXerYOp(Y7VHJ7Z)x0sjoY;4dECV4{Mi0RrR&8P
zLz9^g9((Z6b8odJtT+~b9NUudVRpbJdYjeyq3oLDJvyv)U#=DWz)om0UwujKDO<kS
zl=#wkChGLxka`;OGLC=4)M_W?qP%X7B)2a7zif)5lMnrK@0(VCh(v8|zMyS_edcM>
z@p7A%<&gJ}1bb0y{DA+1F(-y9_hE{!;WS;pvG(8+How_Y;vmgimGiL6+mYtlFz?`|
zT&?<cq~4d1JH<QoHmuXo?Aq0JuL@6na`r+}6u6=6NLDriPZs9f(3d~saDHfh$74wc
zve~>(=%w($FMDsT4PI66zcKihV}vWy9^RFNRh<b^BW{~wBZ<8q>g#CFtm>qs^9q^r
z{wTO9_&my|5bQ1SihM~iM5hD~5<7a!*FnEY(j)E){$@v=Bst!bi<`<w^(E*`RrO61
zdciaV8_iG_9S9_NWM3>13H^$d<yi@=%X2>3`Z7SI1*+nK?@vxiIKXw^|Bxbln2rj?
zj*s)#uI*#bZTYyl=X?U*w6NsE%;&m818R+!5F>mz6DFQ2QWWTA2{6Zi_jEm6W~z8w
zOMm+c{@t)rY*$9ga3&~sMwxlaF83Z%5v0Giv&GllvRt{8Gq8L270K?mep!PTG^_vI
zXVtjOH$~XZyMnPDnZ#WpSP}(HPe7S@p|gx}zGIc1zbc|DUR3_>=Ac6#Th%m&%?w*n
zrnI$|+E%qqkoHOfN5a^QM8E^@YkfgY`H51?`Y29VMrvqP8&eGCX*?!o{J`fLS`q@N
zyCKzR4wInaf|Um;wj6(IVOOwkESRo~U`j3mlYuHnJy?t8B9EqfKvWR1lNF~cs)1`m
zuK@s6)q8+!s{=O!{gHA$h6{nY2sv9jK|-g!ivze86gET^2UwTn+6uA>n7e02(EW`c
zGCaZ&>TYY{%;w}6BjH@Bnz%gG9TzE53hHH)b<S=MU>?saj)$X}hhCFfztr`-?Fxpp
zn{hO{Qtd%3{cJ4YP(miMVAw<m(~J*CCE}yKLx9kRD!ZfZK{1hg^dNcV)E8HcpEr1h
zhDXi?hz_m2>zDD`c~i5+luF|i*X_x<&iWot<NLD#<t-hyPYQMyNc~K-o9p=|y|BNn
zeMoCCS7-Wr<dNOCTTk5oQoOfiM5mQG^6W)G3_*7O>j)Abwlyg(_2jpk4WT(xFJ=m0
z*UiINnPa@>5%bNs-BE}3x9$9PXKNRppOl(^<I$G4w=TEt`D<%&)t1zVgDF$>XM6kw
zg68XIDeXmTO?)a?&aekP*B*aNm9#zS=`y9txc5gKl2Uo(;C@<tA-C}1?(!Kn{7pyL
z<;bycLk%s*)a3s09b3Bwnw-<l?|mJmzWU)nMyG9`5gPgIqIKWS;e@U`y`K)mJWqD{
zA~2n!{CVp}z=gAo`_9FlP&=VVzu9`?pm`@{iCW)&yV4Cr-FCd0!>-it`I5QwHoZHk
zVcc`g>w=j~!q@UL|D_A{jv?2EjWSH*TQwOTZ6oFz13HW+JK0t*zkS#^**3!1Q(9jc
zaAxAj49$S`@=Spn{=V4%a#x8t_TeAR5Np)FrhH<DlHmlF*;kUBFzscv?!&KL*Kouf
zv?IbvrGX<SSu^trcAO`+wAl|X7f{)<{E@?YTDh{)HO1iiXUCcOhVkH9&zWpZHHk|!
z*x1<j>w^Vfl*}oL>ha9L?M2whSYp~M2wovhGf}MOyvVg>5_VR3wROCh7Qbd}yG0rv
z8ePZbhL3+aytJhpnjX7m)K;dgZHmE~YFcTRaKgq#m_TJtUIG`A0IupHPGUbd<il(^
zsuN2TXUH5X$x9Ix06vnA2pA#@#0p7BFzsp#P{Aoc?}}y1A%SCy^PT`%HNv!L^r4Xq
z`4lkH1W1+gn`|hlpj#VG_7Y-LdqeWrf<K3A+ax+oy$*i%O)Vl!?l=;^K>3Sk{ju&k
zQt!9`*#MLCRQ=2~nI|D}($0_$IP69A8zJt#OQbVLBEwUrgNrNY`(JOT-kWVGi`xDA
zQUUBx*$s_=(K(;xc#pCG2|ap@6~`%)0%nsKRshIEK&zS_UFC@0#Q@D&_IT|06<Om*
z0rOh(a-`k(GCB0KiXOcS<p}Rn<H@q2SSv2DB3~vsR1Co;)sG&%vO0tawORK}e1Yv0
zeVrjuT7Z>tvco9+xr=W&5V=X`sap#f9?e&+gNNpkVIc9gNhUBf4<rPC4g3*Tt&N5Z
zpZk-M2t`d%>06>>7(SY_6X0xouk?WI9}xPkGYOE-2wf%~K%5YVWzm2{;s(GIf$;)B
z-Vmdx6O4kel7JY6URr_j1%`4|$Trq}>?A}V?B<AB0(LZ@o`JENGGxy^%~k6*Qo@{9
zH%Ox}aI5kLu|&>Iq?jD-M00=pDRzz$QINNJo+RgtCuD`59exD0WOjmx{5TkMJca`!
z$M&<{j{zxXGd@~2gT+*%jMC60dC+oT-zjlhYp}<{N-kS(aQi>g8~!OjSlqIVHTW9>
z+{edb32Jo!{K(c8-$0&iDDF>yyRg$2s(+TA7hlS`_A2VL!SM`k0(TQ)iYvCkn`D+*
z(9&bc+Sby(M9aq|k*(}#`KMn@;XrfFz^A*hsW@TbBWH(jNq1d`H>XjOPcARlN5^eA
zA@iyBP!n85wc0z<19UGkSaPp1=yG|@f=IQf$?}kCF$z|#ZJT<_X+zuE)E#yMImWg9
z$KRrJ&YU*S=ih#Qf^oiQ&2VTbvt4{3#jAoX4rMz074u6p0)&NiO}ZwQmIC86a2sxR
ziC?M1#qzNSv7j;V44}scnzUEIV(b1G50&?k0<~9=nX2;y4pNKePKLw5&?jA{=LQhy
z3_8>BvRto%85}zC`h15E%Q?PnXTD8QAJZ6rX~(7wW+U7y{dHwlTzH0h{w<Me=vkpq
zCa8rIJHn85KzHqF0eCK?4H~LNf1hHU+JXQ6p6GOB!HevLWrXUtHSKA;Q6w%?&kP@(
z-_)v9i5oCa3)KEvM+C81H~bq&)&|UvUnD#Pe!?Rw7c4ryHa6YaNa@8uk2I9(_f8}>
zbWL4L1rkFL6=B5F#l-;jGDCWzM2<#(hrvT>8;Bi+$69(475;`A_wqKnG>tJ7@CLDC
zwNL!AdJBm=ft1EPXh%a$I&Y0uR);Xs9--Rp9H!XTPEdVE&jQ&#l)=|sAasF%sd6!(
zsuq-lfoBW7R?sq*0x<C2V-Q5$!{>OqC)q${I3Pzw_iXW(1PWp1#A)qwG<q%*Ox9CF
zuVE%Z0<A_#f@`FNzmwcou-xi$<Jsc0afRSdoD58R&s&E^m}`?Ipd-PP>U+QWXB0Mx
zWe~du%-8S#3#_pEGJ}1sVmOBZ(g76LU7b*?^oJx3H}3}<D=4AMlMWuK(~T$S48iJ|
zQsU*3)}qhV8cxC04g1_fa@kIOq~Z#b1Lf~0`*OiD3zF04{vV0eduT#mfc^5?M)&JP
z?mA`k?#si9-`|K!wGe%y;?DMGx6kSL+yqes15x}4i*Dg*{K@V$OIR6Ut}}baCCfg+
zua5@@P2;vOfe5zXV01`jwM9Fl;>f&5*%lu+KV7rRl3E^4D$`9UQhk4q>%(+J=b?Zu
zgvd6I+)aegEjg|RYwnx8&-}6@aUl22lEc@-#q%CA2Tn-}v^~}y)Xup&Cwja2+Mjx^
zqKx7V_$cd3@|{}`4tZ3|Z+3S(PUZ}eCPh#rU>`VuCK?}JQlah))7tQO^aNrS?LY0)
zE^R3B7Pa^4n(vP1yYFv#qVes{spg8b7p|DN+|j4shHr;OrM+-3vHByx;^#Y_1!z4f
z)CbRKZYtRJj%yK7_v<Ob?uzL|>{b0*-<K*nRbdBjp3h&2EUai3*vPYHjV}i6mOhfn
zKA&*d=tpp4Gf6K!3}oQSGK#+`-Javr+<htkCZcb(o0zZzJkPtYS^eYqGcNicFDf9v
zRQ#s&{^u1+K1g-VacaSd{OpNIA3?<o2K-Dx1y<j)q~@Rcm{CG)0=#LYX7&AbqXO{v
z;%~oSUw>b-COy>BegV1V)5ed>zn(YxWClRC4qkiI{&7IpOL<`W=Pv9*?Ui5Wbs7^_
zU~RVUD;4XK*9CGx=&w#Cr3pEJYn@TQ#~>TL64PwUyRgS@O6RRo+W|Qj>Ab<<8Dn0x
zJ>SdX&y_dE1m*I#Uf+y5HLLHf3;)dfwiP@A7xwm8W7Fb0Z@g$Ywt)N7yK_tF+nr%S
zlop|)k&ov$>_;UzLoaLkb^XpQPkzkDon(de_{4d-fK<1Ck9ml9(*}SB&;0~=4lqiO
znI5>m&eu$>U-g}be6@8VixvNnU8AJ2oz-K9<;D`S5Y60;V6vQ%h#|n8$fKykPJq$#
zfI~4sgWB(Lr3EGcz1pNlXTq6MXc?c0av1*k6o~#1NDykOWV6m%Fa<Dl&AeMbM<TRX
zQv^(>l5${bA59i1-NgmeaBqJHwitlfJ|*;C;j2!8)-5Fl16kEB5PJ0JIN(kIGD*td
z7=5$9XY-xPoR-yLvp>*&dGaLghppv)+E`ZqVmB(@SDRL3!X(v*_2JvWn(V_27C*}r
zv&=*7#~uenHQb^{l0srFn>wZ;O?-wm<z2~&^=z~P=*_Uc`k&Yl;HPXPoHA|H$p|=8
zrs{k2)Mvk>MugMPHRlfeWQcy2q~9@XN*7x16_pMKJ>0`=WA9n2L_d4lqWkC4oA~Xt
zqM?9nwB&D>^Z(`#*~`EaeX;7*?U5VbylU8+qi9A5X1f0s+bYMMKLwjQZSVfr@VN>f
z%>ds{;z+=WAB}o#g0<|F*D*Ikh9cUzZZ_8KUjy{c3{?P>P5xi=#JdP|?*>L?Btfbd
zTSB|z;P4vTC(gjV#!WQM|8?W*wwASr>A~_m%S;bO3JQ$x0t0RP6^(eR{T%9H6Re*I
z(#zY=Lc!`|z`eoRhX?`solf1=jWVVgcJqNP%p`TYb6sC+pjF1M$Hx@=pNW-!(qo62
z8sj%UjZ071mBOqM-inaRQio^~^A4OasoB%`UaiIykR%hwEC0&R=(YO2a22tQ;j;`s
zas-+)TMpx-QB^y@XkbJ({sZkeAfh1y{<^FyZ~NFj-sVA1GiP7eTQ_1bmMB0dFhQ-U
zt2GZHfc+hRh7$&^Fo<(WQn|1^3RB!t#y1C>X`m^}<~U`Dsx{^ri8~865T6v0UD2@1
zhYa8QG5`f1-GgqC+(4D^U*jI<%ZE2yRXg^zE`nXQ|FiP{VC>Dqp>Y4V?_uoO4cW6Z
zmNE=kYtS&pKK7xqGenjwks>O_kUfSpma(faLQyLF7Gp@FETt$!Xz%k}KHuMc-0Sl^
z_dgEDk$)<4UGM9BpXck$&Ax7XGWPp)v@l;3JnYyR#)T?OcP)75EjgO~Z9{xQjTfI-
z{sRw2;Z^u=2Om$9E+Ax2Li^F^6c<!%`3+c)7X&A$BKn5rlf;E(QCTzBgmWX%NM`Bc
zs^!~P7tu1JVC;KGJVWcT2KY>-pmA+fPn6YMHt-rxiBbYcptMp5nEYjdF%g6?qpO)1
z4uGvZ9|d-5KKx%dP@+{#Z*I>eiAJFzD#+_e{Hvfuxsol3*%%&=69rciir4WfA2s+j
zR)G{xEQS^92(w-b97IG0DWPW{Z=5KGa;#>l<^aBUGQHLmmtyTh$1zDzWvXS)fdxOa
zfrE|2bFb<<*mHH~lz3R2Bv-)gw)HzB6Np&2>++)wcF)i5Al$FKv|GasNKLN4jK$x$
zq2tnV`^}2o?ar<D^z|~?TaG!%1@5gvhp)p1GL7S{Ip6a_6|3+Uw1(*q>*hBYFWCW7
zr2djcbbT=Y$1_JJ%idZ}DJ7&G!hD@m%eCl^+9tIBRVqvP=+)YWmeHl(@8_eypwq8}
z_xSkM@uBZ@Tc3$r_8=~XcAT8ZY76`Dbgqp`%d^>!-pNboSH`WMT=;|@dhdJgl6=G2
zIg1PF+zw-Z8^-fx^FH$7hr3!k?mx_(S=W|qF4ZbyJnf@KJ^GTBQ?q`VyMws=hL2`v
zXaO5diGRQZk!~CMIEQ?ABTa7VG&V<8@WXef=qg7Ki8m*k+r-OCL<#~WeS%InUDrH*
zdzL+`x~1mJnN}7-%;d8T!OdKSVmw;<(@aKcWz1)@n(t>F8KJE6A>8(@p{`{PqP`nt
zZI8+c$3vIKu2?%ehAU)ZxOM*~BGAte0c>?kO;#9M*5;B=!bORoTrud~L{8y6>+6lM
zgRtWNIBx@~{LX)Nc@L+kIB;&L3=^}t%B3Ba_PuLLFMQ?OZPNW9CQ0s{^T-|%EMs0|
zg!sk!C84!AC%HZtIT0I3WH!TCq;@zT2=}~?dP62X?YE2f3hoC`796L%@mVr}q@<9h
z#|+l>w&X+S1IaAob7Ct6W@^w~o#XflGchU<u|Ws<1O`C-Q&1DYMc;!g>Z@AAU?R!*
zLfl~(PBh7<F!>tnI?R!YgXde44n9$LX3=~3>~M0i>lu;s9j`p^UAg9?zYVW&D!Ru1
z;))e3-EZkVc~JtoTl)o`#78I5A$`h@CZ@?_!IU2drv`wpaGu+Ca{G+it(H+r0v6b3
zL^Gb*%4kjQ)ry>b<{*LO#K<Y@i;K=?pD^X!z?>x14QAcjLH%$|A*4^zVb}!<3xGIR
z+ydBNM@B#r7=^;Xkey>m5@$1t1RShfl|ibcqr&C5Y7rKaBlBpK_r*Gv5z+@TDr3e~
z44hscU{ABgG+5G2K(K&VCcP`$G-M-eo+|ifJ=97&;krtnjC@HE5F76XORwVaNkdfT
z969<gpa$vHg?`Vm>z2KZyd-xsI9m3a2Bf#5ScE^xoX7M_tT3UNcFO)9-n*Wk-sKnp
zOh=%Mu!7_@8U8Oc9dM-*;Uu#uSWz|eE~uX7z^gFX60|V@b1F*9hd2<Kmjhv&v?SYI
zX<$WyMOT3ygEiozGW&TzSSibQWU9``K?d}x6dy;R;=%ZWK^6&xmIbd#Wbv3RDDB9i
z_28{Rp2|hTf9IWaeldpeX-v4-+bM9cQ+oNTdw|dkwHlS`7$=OmE{l}T@0CR^0YhUv
zAC|v=a<HDi=moM{g$0=#VXm4((>>fK`_q0vBUPE?80%BjV?GuoZ4AWmP#}(joaz61
zko=XIK8q{*7$3~(e_1h7&%}NI!L1V<UG9EoPYj>qRB#v;AN_=QtTK<jFe7BTH2G47
z_;tOldCYOy@K4~`CzC)L*T_3nf#KcQR{IBnaNZp`|KQ_s@eljKpTtvNO5k~F{CuUb
zM^&UYKZng*Nmbul3Sh`yo$RV}^}D2Ci-KH^?p=Q@{c`0=cUgjAX8q_ELKN5e=}~z(
zOJ3mVCiTN1)n`1u$0Xv9w8{`Y5^mV1Ug30>6kq8&dr@bzTj7@OS>GbL2K4HAd1z-j
z%l_;PIAlhaW*??5W7-t`e?yuQ&m>40T0Yv=DdD4qAk^(^KZ(hIyRl0N@%n{sdy;OX
zp*W>lGn>S{0F_+MyYTZ&Yiu-o%I2k|)$<)8?TGvOBYq?S%oVw;oc(TBTiR5c9p8)S
znjzVcgd#0&*_)nO9Ma#*HO%wM<ErUz-VHZn(-dE*!}C@hS`;6P2Fs?)9-9i<T?)ys
zKJ}0lCiuT-Q~%}m0#EZ<ykxRdhKD$uUiHbEPidm`XlY{Xoi)6~>)OVgs&+||oxQ5{
zVw80bwYAS^hZR9u_<HHtt(YYg7aNbBif`!@GwK7_$@T=^`-G{I^^6^?j4J<O2@@RA
zT#qgMY}7d@d@pqQ^&sY}=5lR@gvcAM_Dh_%2u*Q14HDV9!%W8MltXn2VOotm@Wow<
zSx1#T+3fHqbEboveK{MlLN&-^f`b7OM&wcvJ+S=n9Kx6fiE%&3w&loG)ZcGxxWhd+
zk|#DBoX6(75a+Y5Ui3F&!yWhPMOcJ_HDVnOI?l}rJcs%SmA*%?5y+%vRP8$-f#L+N
ziemePkqzuoe<KoZRcqDPW<DvV{JfuZ{$0t@*w_C1;QMhsYQ$Fre*J7PD<gkXa?~tZ
zX1CGC*Mod42Z5eFUBuTgFK7pwY#2`<x*jXe`a6B+0ldbkbDtV>B0zb+Y2X*~$=uX0
z$5o$Iujr7Q!DbEf3Ajk_(VpOTX$FtW%3Q&0l;+=;9sdB3==w`(#^&?lr0Fou4cyU&
zaCQ^)>}Edz;Cg~$>*+A#y3;qb5B~y4_PUV<Pv~4$z|jDs^;nh&qSy3M?JT{@;!Mk%
z;?#jQ<oYdXMrX0B<ip$jh7Nxk!u6I9eiloneV2j{IG6r(hHG6cnriUm-mNWh>yXa>
zQ51!!^ynDZtH}(0x1Fuh5v8?P*l92b6(D|6*D=%L&au#p(pu}dJWKs$=Ge(=eU_$k
zBq-jMH$cz&&$<VpBr1R$wnDl+^m}y;s~AjQ?Hav@{48D6+Uccc_^4Rlun;Y!njv=l
z=ac@1ybavw`;O%1x!ZG9*<%_IJ@G>MUwuC(*sbq`C&CHbU|=px=0qc{%-qeQYs(}r
ztkSHnrxEw7eTwBVeaL&jHRTJ-#IQ{gy;#8JLQ|xK2$2A|O6Ca_h*EA6%z=dS6+(6{
zCvovWpa<DREgnS#qz?vxlKHS8u!RK<Hz^g#chj<*GoWrF5GGCFJ}4ptZK~iV@XWfs
z1~PwSR?WlIW7c4>s(}T9layXy7mp&$*{oWmFr0~F<Ws5zFAsAOph{`KXY1CAuOX*S
zHS3MB)#k9lFVi5=Rr*M7lRlPd4%_@<im5cM4f95Vm7NS`D<mK@>#*w`$G-oa2&`_e
zue@}kSQJe+X?-F5BpvsU{bBRyd=o}FAV&ACpbMR#c*6>P(e`xfP+iywxG_V6doI{O
z^C=5srzF97cJ;T}GHhnKJ<HG`?7km_Pqfh>XxMi1U9P8{mYfwbtC=L%DhNx6cWl}U
zYdW4tUJp7Rz4K1(&T!G)Uh?|#q6+z<Kj$~IPQ%N^bw_Yd3XdpRy}pqlI!kXpI|4eC
zvV)J$G#71b9hEg3pMCFtBwGji2=jg+1EH@_nuqedGN-h|Eq$TP(y*(0AS^KgGlkt&
zu#u+`Zr8dkMZE-x7LDGiRKYQYWVY!yy&KXgbZVlVeeN;Et-E+5Utesm?;4psMj+Vr
z3;MlTVua+Cc}q9G5goyK+vctiWU|702I6_5MLhFo*4_~)sud&a+gZsSlednxKwbTx
zDa-)MJon|p&!N8=2aF9-=7X0M<wun5p|UpA#S^L8HFv(f*ll8yh>vZ|RO}$WHke8|
zkNk}47?>%O_<|h9E+yY7rH4zrUqkU3YpI;!4H7roF<8y{GD1dk)HGAs!;!tm90qH`
zJh^<%*_*%=vTp7mV1vgz)?eP=Q4TY7p|+Eb%FAFL-RNf4)wpz;pvywEpJ_C>UAlO#
zIBJayX{kE7Aj2RcaLm6Pa3IV%i&qghLCb-;?!bXd1LqnyNU7Gozbck2Xx?}tzc&EL
z^TU@4Y-;A3?{>je{mW&qX=>HbH$z{%knl=yrf5=P`<!gNj?Ob8;X6f+>K$a8M}+Tc
ze{)^K^&3qj)C9`DQx&~R1)KdZd@APGyO}bj44awyAf$fg?Dzs3A+kkZ0W%lc>vfC7
z9sT2)mWO^tx_{h$<ppG?n;!BE9r^B|ufnII?c^7Z&I&uBTYkK${f*|6&VIT6dQ{km
zNX}gFWxull{P)FNT$H;0f%LM6uE_iqB1$zyU}tPZK`=G6u@{=PyTO16JgqL%qd=7k
z0O}=X1qsHHa{V=PZ4{sLhh*E=mW9^@wk>HNs4~pO6!Z8AOS#pY{11B794KDnwZ33;
z=Bt(Z_577JSjSR(sDyDHNp!zjPOBQE9TqVy!u@KdFPz8=1Nu^H!0<bl>)Fp18FDJh
zL_H4Av08MEg|WYq=hQA4tDdCNBOK7@KClI^`!+CBzUlMrDDlj+d6naxygT@N<Qtu(
z?cDUeDohm@9%8TRc4O;s)-KbO^867j-t0ePH=d=pg6u~_tnk=(p+ioMel|yIpVt9Z
zBtF?ZYM;zilL(ct9t2qzfJB-UAsR49DsGBN1w7dDK(-yk*ybR%Xdr)f!016tag?tp
zh$tB2+(5db=s9?ghUr%%XP{*SLC|Xvhx1}1qLlHr70EY(4d|lW7b>C19>e8KP(2dj
z&K-FZJIyYyBY0HSVT5bg#JC>oQ)ObjwkzaS-G;Qzp6ojs*PtaNwHCyr2Mb6=7?92S
z?5}43I;fAYH)rgFadHX>m?~>HFXUIx7dM~0!-F_ZesY;N+ENz$F8=!q`Tz5U?In<_
zX6z0a#hFEmj^MM_-{qKA4;=!ovSeUU1pBOsK*)tLzf8qW&X_>`HOEtxodazS92dQe
zKeFo|C<%Sf1@bW3WkDToVmFtpIwla7W2$9<+!#B01uL9(g3*($sZ=5}5YNev7-%yz
zdS`#M>MDv|o_w5b9X=mv=8-~Tq^qZ2y=$>zonrd-?2`vQLW0ITA^5od=$>LTE_Fi*
zmpAa{qGb7VC-&-uHyWQKMl<ik7bcbRUmQfXL9P<ftT32{(!aBW2KZZp_>zyq!Kq@N
zDe0Xru;sK$->4pUWM??{cEFS-I!-uVj&`_Em-^Jt?v@d`tw9IhoTa;7ahBH0f9a7`
zOg*69JH2$qu<8D|rhC`%<(AsQk6ZZEG<u_#q>68{<7@m_XdS9m@7%7;2V#ltCf4Rr
zPB8P!4M!h$qp`QK=Y?4x+9ar&bmdE>FFNhDDE@XaXTY@<2Drmpo*ud4=tfSp5&nJM
zZTwI!)23?G?Tp&n88KRBy1(LtKJOTOc}`yJGCSw<*x$F%F$K5ZDL4bK<jEuJE;<P!
zl?YtR*<4*y&1F}Mcroewbt00WIn8izOr{<3lk5tYT@_lxJm|g6*c6fUe!6CI^GPt8
zS3uCfu~%7=h+<o~)egG~T^nN#I@>Djd0&R!$o+@)Fw0$)wo5x*j&EjtscM-OiVr?0
z*oT%FaZh#X!k#(#)=^OLIZmnaW*=4ALA!y3fY=#TElmQq0q%y(ltBEmzaAIaW>$1w
z5q^^A&7Wv06{}CV>wU7_>eI3i=yRh<MLD&r8WtMR1-4wyxt9F#*I^%5JKmW$-$`_E
z3AXrmgkTttQakwJC|Ei-4lr$1;}ENSO^~dd6N^7_bTA<Dys26VA2R@NW~LyDHJ0NA
zzG^UdMO-G{$wol9!cWIXi&jEN^F0cOC(&7DV7*c*AAzwL)U8rwud95cloM!zkU%PX
z3I<BX0%Xt_L|MOr3^4-hIU%u%BqlPRMM?!2+27$ntP5N|tfCYjU$P@ev))!=il_TF
z$-pjD=Ky7@4|=#9l)i(k3rN`~b<|%Drnlu?h<hPdf$7MjGeltaYOJy_!G7VVmBK+o
z8ALNccn^{%vM4J6LSgTb;Vf27*jJlz)tqkQYeqx@49Jv|TUO+}Qq#84xbFZ%s|(g~
zD$r4tI{yDL0Q}#5@&DU*Ol?aqo*YDjV|0C!tq&Y_;qXxaZM`Ux);ZNWe(CuiH4D)*
zHm1nJ70v;Z?{w1MmI*Y?Fb<}ohQld-2okG;$z7K$vr$>Hm}wH!d$$t@Q<Gf2ZXecf
z0#HoPktLZd%_e#^ek;i+H;1z7(#JI3=Pja6fmx{&s4oMgsqgZ4Rm9D*1C4{7Itror
zk%%pHe~?F9?$7y~Z$!?Td5D}r&PPNXk*+9lH8r1ZPis6H&XR1`9i-T+1;L?Dnz_#g
zWsBp{oM@;>ZW0%s?LX28;4V>EtDR4Eu(AX+8H(2gMt1!SZ5q%JeNkKA{?H67P@Weu
zFu{Pk4QsWHJ{md9t3aN-J>X*AADrL5wf}K72;+90c{=Lw#$g}M5478F>{<t<vM%pw
zC8V8M5HyUCxaZ2=bl5a9>mg6e!w~o0DiP2Ot}z-K_ti>dFY+=j9@mjiYgT!3m7(q7
zQIJhy3m8|+nsL58XQ3aXKgP4>Czu;N_#Bu4``IY`Z;N(oR1q6wGX*!Xj($}-zCQZh
zgowI=)y+rb1k<X-d0c0VfY*SD)xKB!9yPLw^4Ow{S&~}-J$V0J##IQ_BXsQRF_>r7
zUF*WMBdd}zH5tzAdK3b$?)$XC9tu9`@`fC_Q<mm)ViRpCMR`4e^P};dgY;_!v(e@Z
za2;;mZOW7?j}wt-=RR+ndqx^+?xlfYg!wWTdH1VbM^zhVLf6QwKKT#&EZG{m-1I2w
zrv@#;ShaO9InMWkHwtzOr`U#Gml2X*t^G`mRpBt2xjJlcq1sB37va+6(WKL15t;bx
zR&)@EW&PdB;T!F8G9G=+=q$w%D^2wue^_z;$rv2>xUP-m3&tcF5yQvfkW$9|2(WVy
z`}f8L(ul5ETG20iHmNSs6q|BglV8!d)%s^jt~|DNq5%g@#hhHG+AhUfr_Kwk8#pP3
z$2(UCAc+L&(rFI6XT@1@fYoWToXsjr@g@;Xa&{?Dy){e51YiJPg+u7TGy@eyrY=q$
zfajVZY~a1XhXvX$M+AxjZ32Jz2{ao!pg8Yf9cHq7=)UwK&Z$o(EOh=0TNoSJq%lQJ
z7(F1HO=UbtG^NR)sA%46I^&{sm@$tPT-@tj0C8~o-;Z3;R39>+3DCYQ@P0RfIVs-R
za6}m)EflyyWsn{9=DYh`e6Uwwv;Gt|mr|+?R_h&KRt}y5e2BWxP-<7FRUAqS6HXy#
z<4eq*ivJi_Pao)J@A)eORT@m*qh2`MnscWv=++#{EaC92C!1eqY8_>(Zx{W0hy8yn
z9Kc!i`pe7z49d-|9}fybBX(;}V!h)2W*A!d)1tnTy8`DUpMR+}&z)}-&14AdOr^>M
za`R}Zn-32s!mFA;S&}X<oAZ4z37eMhNcDvYW$j^5oyInX#&K)blp}~5e)diZ%j~QD
z8r3|8y{tRO3C2dvOh$nhvR>1`rwHO1P+BQZp<F+%907RWwB=;tqt*df)gFxXTF(j+
z@Hk1nWG0BqkqbF4a$OxM5`Ny_8Ob$-<&r@N&sXBB2ivLKkTR_rhU5LohR~+jg~Xwx
z6MGl^EW}*cKL+DPX*fv5)vJZc|NbYq9Te2pr`vO>W+WU!2l%i+8BRiOukF4q<1Xhj
z6-O;KT-5UMx!!X3l5w<1%-!t`Ykx%58-Y8=f?jO%t9H$_dJH-+uEKX(O7X7p80kpU
z4VEhxlNw)-as7RNXUMHezD)7VNqn$Irb%~J;OFB)3yEcVBUYvSsuzEsto3-RKvzJ^
zizT+y$$uHMQ}ZQxz|owyHm5r__P0rlWN)uUr#pcw$YKXq>1Ig0(`)lL1?rPBJ5R+V
z#m_3guR>%<!?Aa;96^!Ce(BYyeyet}UtNUa_8c&@8};um)(IHbA!Z9)Nscp`1gL2d
zdpx0i0mJsEM5+$mUR!^v`BGBXa_Rkhp3Xwv(zDXD^*bJA48dVX_XcV6kP``V+Q~U6
zB5%i_j&mLxnSY_<r{j%b<Mq388@GQKj&c~e?3K}47ZFdZ&CL6{=|dAU1+8TfPVIv8
zrG->(bWku__RrE!T+Q3FO=j~$o1RE#aZ<_VWu3EZ)Z&qdcpO~@uZ{7o$6T2?Hj^x+
zmH2d*T4nF3^bZ6FhC|5H2PgG@&Q#2#^Up3~jDwv^qzFiCj*8a;JebEG++!NdJ~*w_
zCT-m`vQcy`(Ivf@+rPS?n%Oa2+XMwn(ora;{VGTvsEw_Z`Uu%Dp)iwxxC@1ib%9_T
z`M@J{59?{qG=USE1LfsjAQ^|%qd>k|5+TO5NC*=a>{o6Fjf0Nar<CX_1g7feq?b1a
z%?xn=1nx78VJ;9et)Xg}Pu;+qnHMxZnv&|Xb2lDy<bnffLNwEQTy=8%n@}|1P!0*{
zRVi?Ju}tv6WcW!AkBl7oO9&B8nQNcL`s5|2J~}-Ro4KlL8!p#*=Sb=I1O3w}c~!V(
z>D~{~#)Ei5)xjd=*784)=lGI!f!2Aq(N_h0itzTi&wphLE-b+AyNZ!7`3CKRRZ8La
zyuf|Y-CtJ_RkKwhVbvW5|7xJdP^alOjU&&=%KJVQKkf_3{xW!cMKh!%>)E&7sQKW-
zw`N@Y-PF2AIFj`iA?&*p#hB0N+BjeP``vAA($14zV(qyL?do^T`rIFVxz<;zjU-o-
z1Cx&roomgL+3!EVyJQrk6Bd7sWMGaecQ-d5HtM<MC`@3lg@_L^F8sQE^rw{}?Ui~U
zOQ8HZJRR&*D49Zkg-h?X#U^&~%zmKMQolHFP#a@o4X)o|IWL-fQVVw|Z+rp)YTL!p
zk@I;D6K!L5r?2)eTM+_v*Ow+(PFY8XvHGu*kL>$#YOT8*uHR_?BZ7Nx_ZUU7Jr?6X
z`M-t+{&g8a&;Pe*Q~yXC0Bkz&jhIOi%cTt~YjrQa*d|X3KnK^{D+4>_k50pCd!V0e
zdz)hcvgzS<4!}p+usK5^CrP_TRJlFXLr(FpqPpR_5v9|-&+c;W*7&2|+4qEd8BQgG
zR@06$K70(1h&by279?GvKDXqN{5^Xw;?XWft3^4gby9>T)NI%w)if%vhhjD&M9;p1
z2SpiFZ1~8S3~T6(s}>W}9nl_-BDt{h2@}IQ8eM0jjtJIzZ+VD3)@aS=L&9`qf}hsT
z%|m>PZ4=iAWp-u$fU<8K7<0ff*&BqsLY=IVa6&}d-7&#O_p3^|K%t~UqoU}ny{(Om
z!ncYcC=t8T`CCOtT<qh9kMdRhTfCI7&UcZ6ci9a1HF<)5l1DB|KE79>^Vls??WkLS
zdd0}e-(8-U>OR9QI!lqUs^25W`ZE&ZNDYQk6&8O5+}Tl&65PX@?%`IA{(c)#KAA;m
ztI&DhWk0JSEL5W-*JCoY=H@wCU>_?i!%QXj>K>y?*w~B+yCQ{Gi+JEzkz_YS9`Q6F
zM3c-VeagBZs_vFTvddL$RTchgxEi*C;ZrH}+c&YXvJ|+@TyPH4eR7T?46@uQEL8E+
zS#Xji`<SP*M9C%Snu&T}XNzn`jE-efNsjGu^Py8tIS$9MJ`%Zcb>0EygF2WJiJ$)a
zbS9&ht~Ep5oMQw*9ODW3keAE#YlM?C3CpS3>8NjwZ&vLhmtw~BCG*DIYg>;j{MBJQ
zSk=rspykk#lk_A>ePMNZtz$Du2B&_dtl8G7!1J!uUZEnnKWu#MN(?#8&P+p`b0nG?
zIQUpFqp_{=3?xAWBdVieNw}r~<-&4$+o<y318}G8q*Vpm>Y&)=JTXc+!bH*@8ex?M
zDs-ZOPll;#;5}eGz>@&JW(xz^3DRRu5{RDw9Q;yxgf$@Za@d1L=I6-N<)r(!324a4
zWE*8<k|#H4d4|TZ-IAGrVAfD;D9%1OsJY!8tOnH{`c-|bo`8k^8K6*8=7^B8>HVRX
z#3@PAOe8XcOBD-VLVQ?aov2l~yc8Vm&J3zsnXJ}ONnpvQA}W+(HB%?yIEytUbQ24o
zX6^bC^GXR~zzVS<Eq<U6Bz@(Y`ttHO$>r?c)K8pm*2lfmA-Im2yRXqxu4<)ra7m_l
zXY|x&*WvJX`+x+vrjqEpgqcA1lx>4`wMp-YLtz%4DW0tO+10vjsm;#2!Q_)Skg_#z
zEW;Gq4IhA<csk3^>nRcSbNJ`=AzGe3XN}lYIzG|1QJb%un-k*5F%3)U(Mngf<O-xJ
zAPVPDlD=6y5kK2gb!wHnx(bDLhw*kaF}9at8z7&c@M<Pl&gP8Tp+Cn=XwwzH42~7i
z`6o5ktNeUA?unf7d=t{5^UmHvOSMv+uhX2pi9@-$^2qNz&MTjOUy-$1Nf&(+v|Xs^
z@=T7s+VaH?jPvuk^(%9qmS+_vS1bdqU<*PyH$1tsLX`U+I#VCyg(fNz*T|$60mX{P
zix%IKHc)eurPpAiKuY-E#xR~IJXiiTdpwfRrjrl4*)QYyz{`u5$i9Ek_#4=^$Q0D3
zH_YRxx9%Y)s&sjDhZLN}J~YAiB%Mg7JEUo$yTA|#x@pP?JhB7T9Mz%q*5V9jv6^sm
zc$!1I6x<RZp1ny+wQqttxVm<0giyUFc7=$X-t*!yD>e}3g(%x!jNW6`0j1a+5)LkF
zy4@!q4o^>`OtpJOfg0$&xd<Dc6wBtDdPXZqZb3k65VBAhyb>$Vod6E4G1J-1mS!k}
z4)uB2_f>`Y0$bZgmxK2SsBZtaiXK8=G)s@@d&lpLXJBeR&{IT;odp+Rne@Ac<bene
zN@aN^HTjb69@Do>cBzMxO}+O<kH&N1#~MUCUS+b2QDSu}*ME_R!hYrNvAD}~&Axld
zxx&eP%aX+xF8-I+<Tuz>G5pZ{(`}J0o3AX9`omE(cY|UwE7o2fi%r!csR|9%yehG9
zD$eeKgaecSl&v$}bh*F+SsWCr+@_~tiK687vG^os%4Pp@WK+_ol#*G*`c$JVTXK{l
z9FJjh^2!&TGM-jBXmQ175oCOc+YZZhVl;i&S8}+pDTu5vhJr%4r!8M*1u6Ak+TJs6
z&~?a7a8#(@-Wzibx*qRn)+}_&sciRYc5s@1nsU(macT@+@h4U{o(X@dpmRdaFW~wb
zUc@!;PvDaEnE>`Tt?NGDj*(%%r_1mExJXJ2<_;*|^w;QvGUq$xY5?*;c%9eD$7h=~
zEwE26B*KHSTz>P<z~~eb9>4<qa0LI-q%WA{cak|z3??YW21gKGAlw$rRCkW`DBNU(
zUvhGdlX$uniY!R<qlu0oD%q*{02}_6BUo=?3Wp1*?(jZXB}*F+4KIKd+hh;el~@C$
zct4K@h65uB@@9{(ak5^s@>h(gOB6cN?IJwg?=(TJi@cV-9xGiCU@8soTuo-utDUoW
zaR@chuqvpHi+Tgl6e|4mNYICp$7Hgco6VU#ri0cz94ZT5RV?)7Yz{QU9vaT7fk+k;
zqdMde&3N-~2Z>f@6;y^OecvuaTjw~;mk~$rS0q6lvQ=|P-ykY-waEyS=M<YL@PS=L
z@$UKpmZ7X9Ju82brYkXXlX|=vH|vQz5}e0m=!#hR2l7MY?B7R<GDZ&@3Ewhn-lAh_
zr7F%P*=l1fU(K&CeZO@jZ1c<P>q<;yA4=?G!3jH}LDeKJ5^p;AoJIP^ZryvG{m16K
zkCx%lz1I}lvjq4SIH!(SyBOoN^HyV-E!QpkR07VV-YlnNs{?ur>nuq`4<KNqZZM_&
z6t|^8{8q&Db}6~>fNmPA8`bgQlZnrQ<w-mQ`+rPkgQ4Z@f90`9FC1hYl@KOI&P*#m
zkNuAOcVB^I$qGRRe>+kJXX6yDwz;-fezEFI(yPHrg~fP#D`7jfYf+ELIavYhbNNiI
zQkK;aHoZ#Z&|yi3shdE?w$-nJa8P{sVe|$^CM~yj^ioM&5bHoZ<aaIqQlFtrBs?kb
zF?@D#m(nqNdWC5E>!H2Uab-yy|G-VW7nVP?AI_|@frF;O0BzuK3@=*ce18#>;UnBR
z_5wX*5*fbfEn8@QBs1uB9P#g4F`wrXQ({e_E5Zuu+g`t1y2SsaewPk%+(IWkk=({=
zOA_Cj+11_mgts{$lW@d}zKT}Fe+lbM#SdEon5}$;Kg<m|%BPLnH6g@YwjccI%^^PQ
zD~jy>V`CQ3EO8TW)5Rdm4uoleh8*6CJoU??L|ukRK@Y$pK8g&Iy57P}paRdoPDiyD
z;OR%}m7Bx0uAQ;0%g)18+14aG%8n^H80^-NCUh~bawYG}Xo3@!R0x2d^|a$jrPG|5
zs=-~uHga4zC6ndaA&$Qr2)8<<Z;WbV&Y~k~_;c0vT!r5No;DcdEU(kK^a`O?y4&X&
zs~8qDv(c^UBI!aEe%HUUTf1`d@ezDh^|8o5Iu<Wn&e<FjAYMKAWItCb)vr_Vj4%gq
zSe)4%#AO<fl-8d~@GR1-9{WpuoIkTTdg#htQo_tbUe<HIvP+RsR>V`8Tjs0~QNYv}
zc3Z8PhrCO6M+O&lU#!USwn7!<JDx(a_&J?F5>HpZo>ojKYT&<44q~Ee+pP6LVEmu~
ziK-OD6v^jklk2uD7Em$vS&8l%fL{28?ekrXpGJF@w(H=^k2*~vxA|8Ig%u{miFVTn
z&)C-HwVgWi_gm^4tCc#FhcAR<b$tW@W1+$<3_weIL2we)w_<d#_>d<X*YbaM7!zUQ
zflM)3RvFPDb3sp$5aE+paC!6>tR___v(0Iup&gV2+;UNGpIz`Tg?%0MkSJ5)AOp~J
z=stC#FCq|@PN<?)l20-#v<WFoP!YYo#<6>_tl)Maf`zFHjE=C6x)uqn!R)^st*@-L
z*t6eqzyJgH|J}CY!}TlwwX7&!1JnZsN|&xVK(FPKnT*&Mda(_z-H@rd7V{SoudfH4
zp>4{VZJf)&^n!{!2+(arp-HL^w8UUkNc4VnI@bqK^uU`ixn-Sk(-u<`Is|ZnH}c0U
zO6~0-Zh>^t;)PW*|L1F;uoz=_n(6bI(yn{-Y&Of9n3@Sl)Rvvta`wd33@O4?w34bH
zVa+bcMET`JG9g<>5jhoPe6ph{K7hC-c?3yKs!r8>h>*(UUJ5rF6^D>{Oyf8Z<F9_s
zw6zn@d-s3sN|<14yO%-9UiJgY<#T4WbK_bqyotfN$Yd|A`@`9n^MU0s6ZiP)V%hHT
zVEl0(SG9shkfm9-Q4FkAGd?G_@+RLwA)5-zrj@kPsmh%qp`Lv+D`ZS@>owt_Gi%2~
zZ*V1hlm|j}JR5^~3Vf%IN<OS$Ly#VbSF}GMub)>{oc}->Gs{3a_^0hQ)N$Gih3h>H
z=Mh>vY#%tvQ5Ew>+i`Fs%2@>SP~&621KScn9FXsRv4mACn>O+yunq)apO6fR49s-3
zY5f9RWZdP!aVBl?QX%{0sb)(zk`e)X6Ppip57w0~F!dF)K|EZoUWU~zF495!*LCu|
zke&DI@Hpf<Pz|UdRM#L-whObZR08_5zK1bk7g97>9bv;iQG%&1<t;dcu;@n1d@@mV
zvCDtwY(Wy*us$VZCLll4V8fh<hB#laf4pbzzg#Cc{9UK1!vFNI15~tHRLlK4;eCWa
zBI8_>^3S<UzO8|o%DR;KiK??t@DhSnqtP6~+`}GONG2SHMgbT#BANgAy_qBn^Mm`E
zzzuKV?teZ<p68bE{x!MR6#dSnpWB|X^vQ%-z?@a~nj(}t$~l^|*w6VR0^3t;OdA*x
zTWxH(-Z-a&ld{6xuY_^H;nh}VSg8GAk+GW7=$q72Q8q?g6{p6Kphda}*^OrA=RzEK
zVn?Q_bIv9ogfIL>`n#Xu9>W)-FIbjNrpGK?Ho=Rmh~AS9n+lwq%3QOUpr=D+$DH#+
zBh@NOI-feb=ZHfFbn22me;@*<9PG9}g%yypvN^F%(G(WEBfW@6a<9e<1e0zN%?`sR
zF{UYGYhMsx|H68_$q+)}GNo}Nwm`2XbBgVut8iHHWswEZni;#OQ^5O&6A$W7NGGC#
z)9!=4Wl^dawI5J`ND-jm77ZoRLhQW0Rg|cD93q<8dnqM06n?*Ya=GN-rNQ)`f&0JS
zWppm(6dyeSmi{d*QgE_HCi*==G=d1(u~G|c6?~035%k{w=i<sw=kih)D_=qwXuY@y
zid?X+9Z241q*$sEpbMHM{F>SxC$Qci`IgzR!1t?@N&ST9V%3;OJ^IXKKYh><Q`Pu<
zRdOP6jTGh<2y5<?x>#MrN2u7Esnn|9Uru6Dg_hB(SKjx4A{F&nb_GYCA!AUi;_~UF
z>U2?Q3(-YffwD{SG6ox24r4oPxLy_1YNd=(KHz}j*7`{=ZYF7h+Ev6|Ny#uzr_k^O
zaY$yFsh%%ZLBThoo(Yy6zgMT(`wt{jcGh{m>w~bW-Sv0dsA9Q!iuX=*Q^z|!TvL|n
zWVpuj#3&Kn;fTGa5S}xZ?dp|^=D~%llk<fmXWJj$8fGf^!xKbZ=(-V95X<|jF2VDL
zhJJda_R~SC{1<A1d)9vI=5eZFc7(H<R!xA0W<2A0fQ|9((Lce_*ZUS8y6QJbxT|ZB
zm#Fsf7rn%5R-KzFKbursOgPViCZ1afU#$;1vUAvRukJJMBam<h6aKz-dN`=jDA=7=
z=MZWbd+62SP0q5Nh1PbTEJHF0=w0q#W3n4aDI2E?G-n<>%5FH_=Ckm&I3WZXDf7eD
zkz6NYo}nJM605^YIv+b*<AP>&W~Xw)N|$e`3mbaWjlVyUFi&vat<FD-eUWQSjJy>=
zq5!6`Tz@xq(b-(!ysy5kWtc%Aj(MZjsOsc%SG;EFJs*O;sbSX6tv3bUo_ji@A5X10
zS7GG_;!96C$4^}s7}j{}RR~8@LkF_h3>hiZh>o_-8FGT4IeUB_IVY{>r%$CkTa$n+
zW=x5ls%^+LVB%nzWE>=LdQiW<1rF<}Q%E3JCd7p8a$NiTTHjdp+9)a#o>uLuOsDOJ
zaTip@D%VXBZvGk%3C(iVxv$mwRTkNMvipo7BU=B>`)w@ysG3>C&OeZbBlt8Y%&^A|
zyeA{g9`58f!JM3edp_E*N8!!6{PAGBi90!eo~qL!s{8!tiMG2b#A(UTo>5j)6iT)X
z+0dTFcFEk1x%_aD2lHLk<?i%CRIg9{y6?Bz#Cm+Qt5W+v5GRBqyw40RyEIq_O|hS>
z@V%xXcQrTZ{sjtYTFLBGD8GD`Dw}5nUMVm{Ky$+e*%JzF(mY&Sy<0U4`Q#=}EAD>%
zQI+y%*9N_*)fmudZs2OBUy*spAqQ_?BmB<!ul$MYSu~%sDdqL;M%a*e>*c0S5hhsa
z%m+QtdCw#>A!8h6(JjLZx0p#fEJpnUxGsL%C{ZAaz&iFQ&^9HoL7iYzxz9E=mqdef
zz=|lWeZc0$vYZXdZ5bV_*;7Y7z`s*nh;nDktV)$;MR95GiTo5|faz|fT#=^<77(S1
zmP_(btcp0Dn)^Q20EYhAYROQY=3(KHOQ8KahXDF4!3<Hz*muc2dwL*5fa#sy=><5m
zYv8Fnyn8Pd<WBX)FWo#sRlB7Ba@q<R0WruO_zjN|Pf+svBz>fNQbarU{oik!8~Tk8
z{Vd&Qb3*Tmvbib9?8@>WCc)5J;5|#S$K)TFC{$R}oLTZFxWwE=8|qx7Oh91!fA(?z
zw~Qo!pF}-B`RPArS?eNK(BwG+FunA{({_r^zjfk^l~Xm;<v4SjUB^pFO+G;W2sVhl
zn}Zd^l%b^N>fuf~5>Ygne$|6t`Uh-F=VX^Q##LzaMZAmG;n)utKcywtZ1DECT}1S0
zxgnpoAu(o-{G27f7~4nVCy?@(jfq8jzn(chB)3;YnXSq~-hf;(uZ7f_nIBmlsreOi
z1<TgE<VK!My8O{6NeQ;hHV5$<f_;A;s7#enGK}#e-u?73NObz9l#Nk;y#<mJYF;C~
zdzaXdts*t-$n=;j9{?N#(#@7-hlwX%DA|ct8}fMs8ZjE66A1h~*_k%rh;fmA(4(o}
zRCSZ<L_x$+mShxc+we71E*jeHK>dt*SZb*@!c%MMmiT$M+T+X*+0o`7*T$nBu{eE|
zlYHqjxh>U;s7IBc_VgPHA0Ja-?Z`N>oUw32w~qYf)4mYXud$jp_xd8ZyQ=D6?IO>j
zKG1u+2?ksxYd#kn8@=MPy!Io_*~P{~N)uh)B*ZP1y$-*bWad*h{J#EHe1R9^#Lxx5
z_KaZ7=d$`^HayE7tv2U5+(qMe?=~V~mCDC*Dnrdb*5?)6_Lo<eP&Qsq$hbqZTipOF
zj|4S&Im>%w3rpdc{FQs4sy^wfnB4P@r}|A7w<vca)j|p%lw3XzWdzh%y%R7B<#o;q
z=kcNX`{c<6qz~?_z;9lfLiAZbP~1EYOLa<N?szwa=<ICL{bpVEcw30nTv7cq^PEZ@
z^M|}X&eJIcH|m~39J`+134WZQ!TJ6XBQ7j{8wYO;gzW}Q<Shs)D^;_&B5VJFz$eAu
zj&eUM3RlxNb+Bw<OvUqzA<Kd-{mx;KGqTqZ!tZ}v>~jf}@xLK#Dm!kAHmdc~My0(m
z;*}tdcU%L_FV<qEYu8t4$mmtOXZ3~|b)7-jVYSG^8lp|9z1f$h#RXPxHrYAyQh(cb
zbu{s$D;{|y%ojg)Aa59}06RRerdzYJI889E9-GOYoXkpD19gX(e<16x*KESrc-ntH
z-D`^b-eentVO3-~1UhzhpRFb`dg>rwouKzm9<@2oTCQ<RhM=XzV_AEAw0O7i9r4sm
zS1-Z^;XYgST+*J$3ax}`@rop@_b|Y7Y25%Pns2WpESV;nFq2-~@~9Vm%CW4?Rl$Z<
zv}Mg=ByJs$&7;pV)4AdF52WS@Rp9%C)snxT(R9ZP?5992WaCAhhQrpbIUe6h-#~)z
zDZ1mjNP{P-v#Cz9pj~Fo^36kD`z;j91k+o#QUoxJFq5J-Fs!f(V9_)tI<%U^(>2*%
z7Q_?|2N|{vu!6!{?+}uFkxVEyW4JCT93tTsae(z5<mJ(R<k7*Y))n*bqG{a6fmkWi
zrvyx6RY=xru>77dl?HQ04=#D^K=m+%>*=WUIMO@q(oXJZoct4wWw{tl_#X1tT<zG;
zwLfsgz=ingnvjc2pT12uW>|&Fj&2K0y(#aQ-n`_lHA^U|c+jIg@<&a(arLj_E9zCk
z{$#E*j6`0npBd1xOWQqdyVa}vMgp)2?*Dy7>$&uzwfF2=Db(}f?Bk}0sOY=vYyJkB
zu7^3SNfCc0$2cM%#U!wPyLPl`t1<NRhpasvrjm=kENpv=yM2}kOC=I2AI!H|8y0%1
zXk3{?QmC-h$;gWqCQAw)9)xIPJkJS_`{JUf#fnlUN-s2Kx>|Qk?unfXi&!;}uUpLO
zb%@Y=C2@j-W>#lYZ`+w-n>HG?@QCZqfB^OT{cYig%f_bamN%Dj;%|=paA4ePan5*_
zgZ4ikQ6IejGhu;e^3=ys?7p%TtN6{7^XhF0##&;Po#ah~T!E<bk+XL3M~|B9#F^^L
zKRkWX{XzG`Hq)Kg6`Fcd;dbo8V~#>DTq+9xCm~nT@;@G$n>`1C_FT)I<@`M2Ig9L%
zPqu*!1wBc>CzhlT-8eQ&{|9oL4QF64qt)(f5~DUmu@3U-{E<c#^!?siM(c@s@Lsew
zD@aH>GbJ~!#&Ani3eF>tD$UrsMZ!s4JIDRgL88uR!b;jTR-WiROlFbz0)N!nhe+$C
zmw#x|lA73Lj}Q0nlgsO<u#T_^(4#bWN8nwmeq6Z*v8GHllOZOyrVDG_Iq9mZBETeK
zb~R^)JZ$+1#S7Kjr+OKZy>NNAPV_guzRNF!;f%tXv%_GiYZH$70ZI2kN_*&Am8`|v
zCO;!Ss;5=bZ${Le`F48kgMbgPz?887Kh#|-p^w3FFN!b}Gm<<-uTGAz_}*V$N|TUW
zetucv(<SEDJ$p?Yzg@e0oL*a-o3}P<JcHfr_PzG(>kRFhvaw5l{^A=-J<mxauFq#a
zpUX<uP}hUZf2(8>A3i;*e)9Vjac7-lRy=anfuFE6>b%wcz2$6rf?WLd`N^sI37RXc
zZNghJ3d25ZBQRfl&Kxa+>rcJQGkayYY&Bc=VvX{oknr@)cgH%b##!Mps>TE5E5dTQ
zdR2R_@Sx9jN8*bv)Tt0ozhOwLuHTKT-IvlC-kLwpmPfrfu3@NM@$Oizji%$N=DS*|
z$22@rtU*dapSg0UCEoGAaFt`nYW;k_n@@7SUs525Rt7QU)*GgW5zJ@3_nKZi{A!#T
z6m7XvR_XBT{ZGc8X2PujtkBjpIoIazz3p1GnO5O4eQ`M(?56EyF&nt&H-Q|)y>@u>
zCRIcBntbBaE`{6B>(cfe{7b0@M^4^ll6?T8L>Qa!&Sh40SqUxGGB26pk~@CLe|}b}
zO(A|v;sUm&!8blCOZ6Q75S)u*ALVLBzkh>m?5LV9(Hd{PLbWna&oh6uLciT<*?+ON
zxgPYUD2i4W3z~^V*;`4o0YmE`uZ_~2mn^JkHAx~Xq-5ISw>6)R+A>-(u0?cU%%yUX
z_&HDxsXnfa57!MDzW3WaVfO5z+-_)+>Y+PN@HqXDFf-yAHOf1FB4jy<zsXF8VIOXX
zEP}mtVObK=LAsJ=6}()+EKF<42?`72!Pc=2wjYQ3pY2hIe(3;OSX>E!;*MjW)5s@#
zf~jm5z(g=r<G{a9Mvze&l79C68R6WDR_<0&{`Va)-e|TcE^9~a`;v68DFp{K)C5Ig
zE+KZ1pL?r}InLIdlZtmBIY}-@A546rUq-IH^rP@;{EtvB?uZ5Hy81I`;;#OoW$;hF
zF7MsRc{x{JV%hnHG?G}oNoFJ|r|ta&JFKza)?HbDkiUvrgc&(Gp2GiGV%4PSWHi|E
z1zFatJEAV7tf~tYe0a-N{pEMdi>0+q+NX?wnKxJ7jKvLmDj6Q8#4;F9X1trHjgCiL
zNfe$^h^+c*ZZbCjHFAa5$Kh7Rj&7Y;2-xqTrOT8GE|w-|fh1AEc=gWC=`)AaHk96)
zb)J>QnIwpd`FOupk9QSb&KN0K2sVFIIOpzQD)K<omfggZvzzDBY3woXIISb95>^dO
z<~DQ-15TLCM~*Q|LL++l3irBrbYFL7{)3K=2Oe>~*xqQ-7iYIA$0lrItwGsr`MGE*
zgB^ysjN5z501hAddu$JM4*fcPr0=1covP@`*4{G&$w*DmQ@YQnG5Yy{9Cp{|#`D5S
zj<<*T`;!;!ylEelx;Dvjv0U)~oi)w&;s&q6zbWTWW<5aZ4k{1R#-;Sx-;}1m7CHMx
z+r`T1f^Gs4_<^$RbmU+^iPk?SlFa`Q8m!q<2Eix^73I;B{0c_ecD!vV(L18|eUyzP
zj%j@~9``<Z<O1bXiHY>I^yIqg_2lbtzLA&2Q#9m*b#^?n{8TPnA~y&r&dD>NbMIU%
zO1vnYxs;yy`~;0?@|72$kL+2VmnOnfLB3H|ziOV<5YkXqaeXyg*w54Q`*iBKalQFT
zJ=kI#<*3Hd1TNnW=BN$JjzHieWB1(YxvbMD#hlW0p&{L#?}Y4c-><cI&uh%EDVZdx
z7AJD2G~ju;lNghBQay&e=4$c+>7*QT>-le)no36EhK@`IQ+CF+0w%AI3#;u4zcOjF
zsjR>GOy&n6DhF>or*YL!y6u=l*&cb?ZQQE%y>h6S48w|&y=Y5Vk20NhD~^@!<`LbR
za0r<ld*rD-0OKvtSV(owo>|Sl3W2RKiZ6(w&U67DpwmJ@4*c{&S`?O&H$_an+rN;#
z9JDyGFuPJA-KP5WR07X8#WSkTtEI7>2_KEJK)fF_SzgqqL#_y4J>z5B8#q%L7Lq#_
z<}S8SMk}#%FlAcmcv=;vnoYc$)aFvmp}%1!JZwipXgCr&G?d^25gKE{&?)j9Ez$=v
zQ`=0J8V{Daa_I^Fv~6tXCHqoi_d12QkrfX1-?R#sW=aik3cRe$Nm6sv^eb^s@V*x2
z27FWv%(LuWpdj}q{oPFr@w^~hmzqTM6fh}KJu~uNxwFwntmr|=DK()!L~7!Cq1ChP
z(yuN>sk9FD8YFqR73rq;BZ$SqpPN`-+;B?QDIhuXYfR!4IoI&^ya3ULqez$Vti``t
zhwA2BT*RxU{L2RC6D#E28n=GBeR?JPni5SAo$YTMY<2%9mNQ6IpM5xZGEcQ&HMh|$
zVUj4yV;oX6ZTH(c6Z<`<G3Iir!(Ibo_ilfL++Mp_lc?oAH34uk78wW0y*q!k64`Ea
ze96_$_YQxW%7D&9#0{mq9F*|+us)i_%zdNP;t|BS&8gctO?nG?ebOPOyukKQ+GZ@w
zbLc*aOcM$pY$}(}rJd6x*&BcI&Bp`Km&5^WNf817_=XOpAb(5d0A5AFkc8`MEp!e$
zwUx}z82+1)lQThU%YdgQ)`uC+3P?DxxsDPuuS(dPJ}WOMtVl;$?*cd)O~GRrFbe;D
z3_I<+M!LPRh~{Tq(HC1oDF<92Y#;U@HQOm#N#-7I(toTFGEQ}`klACkv4hvN1)OMI
zmKY25Icz1m^LX<qaim_(c5?~Ma_v7vZl(#3ld0k=((A?!O3u>3==+s->hy(Od&#d}
zww<J@pRq<zvP8e$t~I!f%n0=`QzH8PYuNi=zDP(*`wQCK*^dBF@1W;dXyEKI)g7vf
za*`jUJPhe0ko8v7-^L)hD!<gB11DZ>vL#(CPs=WYr{-8f*dSlcv{L#s6-d=ujMR5u
z2;A0msC8jF3_3&dql=5%vTCLP0PfA3s_~2)5t6&rvX2JAR)g(n6{2?=Gc39Xa0Rf?
zfp~-HDp}6mKX9hiq5&1Lv0#&Aei)PHIj-Qxmft8r+0v|#s4=L5%r(*K`@`EN^1`+R
z3}j05i#fb$uujay?APej`&&cFRi9M^3Zq{gj}Qet#3Q6j^%Oc&id$D$Bm7F+(KMik
z-F#LwI9?ff<$Yb%VaGrvm)~pL=S$$;zxscFDmRbJxMge{yzFWJB-14R?Z;f`noZ%p
zrumzqrVY_mwi5-#$^It1iR<s5=<ipv%nq%B6D=v_P4<LVO6^x0OoD4AmKV`#u-NHb
zbiGVoQA;IN^7Z|y&q=a1s{B#zwMYvCCC*N5o7ZK_v<4eB!=nD`Cf-9?yNjCfGa^=X
zo_|jz@l;J&vUi6%n<xfEp@$wQS#%#Y5x&t8vc&}%<9)8j+>T2nbaN|;VMB20%<(w>
z%I_i4LHo$}iM{N~Ou_#CoTTWcI;@%tNwweZ)ZjR~14j1_{^`zJU+1yVfh72%Z@0)z
zb64cU3NO7nUuAlGoef$emP_YYq`l~Vc~05~;mpeUS3^0UbtI})Zyh^j7AZy}U&(Jg
z;ov}|RWwb|$sv)P8nc1KOovtydCB~0OE!Tk>|-2-&B!yt+0XVOGS#8;MhY>e@D8%T
zVP0gl;wZ-$;Wd;N?^0Xg&i=w4`fGcS%J`6Qt54rzTy@M6d?;M)lgG^FhZAnt<_99v
zxg6+4yi%$`mw)~V5;u_jW<f8k@;3QZvxX9NdG+;)_$#(H&1@gi;!cpZztuwS%3R$=
zu2*vgE_Log2?Yu)f~s(_XA>uqL$uEy+QN%BoZA5z@t9h(&u*S7P^z738e@XhBjWK&
zSfn9iGb#6QZN)Qjg+lo$xjRafW^?y?|3i;tY=>8c-~XELe0QyW`5{bR^wp;S3%Knk
zpHY?+{DYS~Vv^gQ!pPbeSaLUV46|wCKrT4#(T!Eis(!M7oa8BRbHwxoVW!5|i70KU
zkUU0-K}z6EO}$P{(;p^->DmU`GI}HoPuu8^e_5RVS9W9Z$yKXI-#)7L4YD7>Uf4(U
zhPn2`T^<VbKFhg@WEu}~m8d~{pgO1SsByNo<hDFQ%Bk%(#B4T3&?{%hYL51g2-{eq
zjYqyP&92_-eFBG3KKJ2Y_^?a3O>x|<3P-TB6&A}7Kz%a>f)hZo1p%N3unkiJlBBF=
zlAA^}x#0&-;YRQjhQaJlWjk1Bu)?^Wf~Cg<JeRG+SqPs1mW7V9o#&3iIe@Y!>8iGU
z=r8&PptcO5w1+=Zvj5+7w*RiU{resBezEc``LCqqhar|j(ADY;4~dv#UD8pmNh`>n
z9gd^bk6k)C{I#?rotgh$rAil&brQ1rWF#j>%!VLN=X$ePISRF#j`D$h!QQXbgsJ2K
zd<nNf7OfOY^wKw}NuY~LcT|H?U=7G15XQ+xZuS&X8SU{IfS+fof*@?(<(QbYB<xeU
zFjY1wn>+9>vX?6F^gT8sO2Pz3{vt8RlQ&5H%;Q-q$o!YE`y<TtszkUKCvO(z(UYr0
zH!vp5$pxNE<AnC6683#qYE)oWe#*pz2zT<Ed<3XF3@7wV*zXw|x>_6&xJR#-Ppe`a
zzVS!Ij_P3N%P!XRM{?tV{FC?;!KKT=^OE2NcoWYC&u1%oE=Tt9$bY-_R|nodd1fWw
zQ&H)W^lfu;$6J)DlLB`jaW&KY?l_j*-{VTQZyQkz@VQajgQUa?+q}^g8DB1)Y7ih^
z-q5N)&C1^oe@+!}^$i=ej|}P$uG*DV(|koXP2ILHn6-SxBM)W1YQXhiv$&wmG+F|1
ze+FP>^($Ff&-;uoRu#MoeI=yfmen*iThu4#ogMGkB8rPx&xAchT#MLpFz|af-51%A
zsn$Hi8~^_>c9v05Hf*~dKtPcY7?AFmp`-;QhVD)WL}_LS=@cadb?B55hR&gfMiJ>|
z2x*aSqy^*Ix6kvw`+L8=_Fnt@Gi%L{nIACgy019T<3t6-@UofA?+%Zfsd^@wZZ?73
ziJCL(>*?``Z!p+ag~YPQ`m#+UlF71ti)?5fLS5}<ag=%w!WTOlvKerV3T-rXkY^u)
zDoTHS0|qwf<;_rV#bwW*W@UGm;Z;k@CD<2@V*?wP#TWqWm?M+maj|@6LZDNL82<$!
zpjv$X9q+S5br@=V^NB^+!@K^ZC)pyPAk&^l3H^oxyWP{x(r@CiJ_8mbVsfYLQ%246
z>OtM(;}fo?lcUc=AfF2a4DRU|!-Qh#Eru9spP91loiaG9T3XCPpKlhv7SnE$cyzM`
zT~jsAnA!FD^~-Wdude-ulh<lXgUC!F&D30sp)GY3P2L?Q^W4`sL{2fib(ZQZ<HkJl
z*;-y{GzW6{nYSdYf^0Bzf?l-VSYt$l80ju8-Yb4trEdK3g93ciGHb%zwpY4BB+jK_
zMYZiYH%%F_)~V;?Zn^B+zb4O_^n|GQ%(k|`JNpi=7o(Cpd^q1kk0D)rc;}<Uy%?h1
zTfC=|bsL{%lWhIaI0L)sUf;!G)UexM^1ez{EPk*pY`HdTkY2Eoc=0;O=S8^Uz3Wdp
zqf!{kl$X46kX~pK`#LVU2PByiJP8s9WQl;a8wUDV)Q=yEH~%gIHY9m9?`;Ge7A^L{
zMi2-(gHZ6sEmeFQRT02<rDU5_#nW&>VyL@s(ny|?QKl~CZPnB*@)rpFDfikn<ztTN
z|L#iu9TxI`v>E^PmbTLFl)0r>#PZ^p&@hby(C9JFt6vwKV@$<o<C!NF*2aF{K#nd4
znW<HK>8!?}ZY!-&c5a5T%8rwf`j7+^-p=%AdsH%$YCe-dq9$YHdoV(1@wiP3?`a$C
zcFk6%-y#a6gtmQ43PhQiQ1@tfc<w{+iolayrF39?YDakXCy){SZT=u5w;oXtDS2uI
z;*{;IM<n0Xw?Ld!^~U6JRN;-t6A=^wZKnu)o0x{Jm(2J`>S{We>23N&$zs^AoZow}
zIOlK(fUhRI6?3r}T9Q}ThyE57`;tMK>v4(JcELdG^VCAE*9R@-mj%D5^otCZd+F#f
zBXo{SV;zo99>J8)Dm>1`d*7&c5G2ETdJ&aGWG1Hb%BVKJ#8ZQ%iqGXO#&5ou2a_9y
zWqQamY@SH3^Cd556FLZ*4ma9I<mCn7Ix2fy7N1L@F)FF)f{}92j<Q-6(3o7=5;EuY
z{U^Z52ft`Payo87W3KqjjMLvs)*gje&wOnE*cqxmEVujCM!VUzkR>GzwHDOy<}OD`
zT+&?XG%Xlcv(&kZEc>Y_7ZvEnYPP4vX}9fC>=pB_Pq(?Z`?c0N+^93>)k*34K4Lfm
z@43U1kg_^(f?NNHw^@lG4l4TcW4_&v?&EZ#D#bWX&djj<FY{qEnQW}Ys7yNiip1q?
zf5PUYubmY$AG{g`lJq}(H&K3*^5ai2*!W4H@nMhfVVOVy<!=eA(?7$@>KPcIL+x(@
zKUutKE``0jK<ac{OvvAY9`xVSOXFW4Yl;q6vsVvPg9n~9zd5heWWD#u#Lhd%HOTab
z=Z~~&v--u?^)goL(+Cb*F{}M0hF0$0DdE~3BjhmYM0F{ep6Tg28|Ya1L#}8<e}zWW
zkQ>$n!dx!J&OUT)pe0b$^}vld(dzPz<uCTY0}B&4vx#}TgW8;-F?7U7bs3tjC0pm5
zDF)m=ORvp&?%9v2-}ue-P#OEG2)+_CBYsY*;>r>K?m0H#8@9IJTsG59m_e6!CjF+y
zW(%|aZw0y^Gu|@J&y;$`O<x{7nlM*?$2&MgDamApX~X9q>A4rk(o3*nTmPc)9@aV|
zOjX=?Hd}99F38BID#ZiEh!s^1T~04^b@7gAx4C$GJ%M~4^zoVV*;WznwsViUZMDib
zGDf@sjK<iIb;#q@`pLo%9;kYmt9MAMwV6gsx*-p$>dQz9(YV#p#P?q3{XJr1I_HIW
z7b!jLZHh25mua5lgNhj2bg~}~;>&j5%Rpm|r!%=^GgtVBd$7AE$$A9yVI;vu)j=ex
zrO!7s0sBV5-+Lo`X9zm#k7qDk31th2>2>pJ48(NzrCntbrn2{Sj)Bsla7D`qhzLbk
zOw9xY=6_%Fn>@ilJGqQ+RPFm3K9iU#ak8jd@l%c>@s7BfT>oxUp%S^?9BHbiDfa=M
zWB1tpN4m_`E5~shZCbzvbqMaAf=qw8nwM2o&vt{HbwHYhPsJRkO{!0m7sYs(sk`77
zHvN+~X~JNKkt*M(Z{j4waVP*GOpuS#EVNjSE%cKjDd74|gA8Fb#l9Z#EMEFVw|fE`
zBwke|sE1Dcu`FXHGlHMm5dN&8X>8_F<@tI_`8j5@eEfk01pR+#qzDhhD&M!;onypP
z!ts!f(RO7xk#&529YIp{xE<{*eYf7K<A+(kq(MeO1_o%9Ve&beDDnG549P7!UC4lZ
z3O>Gw_9cQ;za6RiKvg7fp=CsBC1x`rgXL=u;%ezqSA?X5SX}Immnul(*2;L0fjD2p
zz=6if1zJG$lzL2jZ8QKKMbgsiqDR@tuBb_MQ97!*VvIMGAYBsftzp*@DSp^X#$L##
z!_sh=LP{YA`N*nepYnJkJ$Z4m>Z;BL7b?BQDoKn~f`VT%F`=IS#MM-?xss<;B&#;y
zn=*Y?D!4?-mFFWUZ<f0M1zN-D7g;jSc9A7o&MVL!JWQvZevibG<B9j`>hgknfoN}+
zGYNz5)2PU+#K+>8hzRQrpI-&)z7B{6#(F*Dn9m&kiK7=JiF(Mpc9O5i&nkLU(q3zO
z{wVY8gR|5U7l7lSZ9t4o8^;9<;c2uqr}>baAU?=Xm=80Z(KHSkAw){JRy-u*3*;!^
zhYw~asgV>~+R}*Q9ce0_v&wMKH!me-Dt;V;&s5T87n>BHH-8H#v+J*<@~>?R(Z*{d
z0f6P-^3$r;zKc(u`0SVykF&(BO^0n3TKURB=N9d(Dm*GOF&808PkB+FGkOL6p+be;
zE9)%C@+v_7c|xZ7v%T8Nm{wi`ou(-J?07^S|2-?N{J~3q7B#7g1d1ipY@^(mLZn#|
ztp48L%BJ^81j{+EL`mw65Pw{n1Es+k+MZo+BNRE-fD8t}&|HS;wSSBukf9v^8`2_2
zfeRI0G2S8#n&~`%Go6TG&82TDFpLr`NNgLLp`<p?1-df`;*e)BKFR~F;x1L_9ct4;
zoy;vKSlU8f_I;h{5#Z=o?iFnCvOnPs2SqkulMfZ$YSLQBOLizF&eUc}*-K(?boG@;
zh63c$iGI2!_~^{xjk+pwPm9ru2hPtl@(7$bD$6jXZ67ILo0SfXPybNv>##$a7rRCH
zRoKo)plTVPr4FdRvnbu>12~H`&rVWtfjy2ZILA!97zD6`iwW6Fo+iI%e>UPIR*bp~
zr%t7DaA_49QjYS}gD10pq^Wp%sp?^Ssa_WTd_Uab$=C9|8cTb-hm%3KDO;?T*h)um
z*PE?+>}0!VM+Rd-l6Rz5i}p?Qti3bna=ef(ilZ)cKb_Bze}M?E87QYqM}m065f=C1
zlX{f#*g)(H)R<Z3uP*5!3{0T%){$wJ2~3Ko$&zk_GFDr91*Lz$xntsFL!Q6Xqvi|5
z>=6~*uPM5lbl(_;qdqY5D;Rs`<t)?{IIO>1oRNNsRuYf>wPtoUbI&XLSW#ey8s!HA
z-$Y0k8N!APUR0Fv^R_hnIV?h*iMYY^fX12x!t+s81h05n1xV3^6kjkkl89tRynC(c
z^#LIJu3Nmg-cMD@F^4~A0JcP^q2H}9v|LLM<4tFL_F}1yu^5TCE^k?tZy~y-zb9Ic
zi@N45-J9xk=(A*Vk;8icg7wWzqDA!W05!_H_{|Lec-5$l8;{BQqRtfd^&8#pu!`;D
zXs(@Zm`gk@Dq5NzySOOurdf2QDL9!&So}CuVoHABMvY`4&tz8DYphKl%|so(DO8FP
z`Rs9ka=3^_&tbaxs~X;VK(AzDIED8a#jSEonz$a{-kI@$zi{0hL3sP+;po18!RjmZ
z$(pHQ3y1P|y=DhL4nC=fi=j|f{n*>TE$B`W)lN^8XOI77S5)-PaWVt>CtP{?cU%ac
zW%%{W8L@8dhNQb@`qr8~1fT6yUA`_r8cKbcoeW*XTNGL}R!`|e57pENAL<hx9n&;s
zU5w9s_Ml&BHvKUabC$olapPNYRv_7akbBMNyN#Y13_P!pHOwB~BW88q&eJE>rX@lY
zE5E)&b=CHC<TC4t;i6yh-IrA^;mV=@_A4$|VLt8QlI=4)zaV?R^4nj@M2<!iT9<uw
zC^xPY+P=gbEamQ~4tnY)Ab`-0=7~cIW_3eJgqw#=N?>qpyk4O6fz_{7%RXNYxXvVu
zVt`W!_i1N7oS}|AKZE^0wX+E4LDD39C%WI*2}KvCbZ#PNL09*=UQPq>6Vt`7h_JO*
zbJ`c*IL*n)6q^hOvJF%@UV?lTQu#@g=)(yifUik&Gv*kUj&@8=h9%^AjtF`8X^}L4
z`xK_3Zp8xOSyTEk9cc=3!U*FvoE15r;PHwB@I93A0Moem1I|dM&CuC?v>gSY?*Y=Y
zJ(yO^UaCVNct3xGOTw}Z*v`Zc+s6&95px#hQ?)J5VeZ8VeK9$RPdNY>lL;qw=Vp(n
zR+3k5bZr)u_qYMEj7+i~6R_P=gANjVYPcE^e^tuyfKalLT9~V}tCh@Gad#SqBOhQ{
z9iCmAie`kGn*79K^G}gidzr(X%@M@%=k6SuJ19Hu9bJxe@)1e3t3Zsi>R_eOp^KK_
z2$ymwv8*v%^>&P7M^KVs8ULY=S?Xqgdp9%EVqV^B@v|H^(YZ*nABBj*dp=dyv(1|I
zNA!1j6q^|y6_*!Db7fE_wJAqJD)Zl+3+Y-UW7QNfytmYq=|=Av+}5jXc~E!tgZ-y%
z-1LNza7RJei0)V*k>^aNJ%}=M%gZINi7y0T_LNKS88)ELD^ktQ96gKEN^LxX?r*`!
z^$Elw($6@(HG@0-X*&;$^5nkf^_mkxW>YPfi-=LpTF||+?A7^6B@<VZZOtQ9^}GDD
z&C>8szunu7HfraKyugdE3;py=ybrX&3w>j`MLs7&dJfITZf&B23F_dvS3_2Pnx>%+
zeVzs~=gd7GQ)bC{$4SOJ1{!18PWbtq_opVDKLul~6y3QZ$dMzpT(%A3yq>4i(wt--
z?f#Y5*+*XAV#ucjv`=TE3UTpCXsN%SFX~!~i5iK&r^v}bQYPc9%P~RYVtK$|%^pdd
zcWpo<z`FN>VTK70UXbI+y&t6=bvB|@Le!z<vLxluVk=E(X}(0O>SX9_qJ_)G58=zy
z%#d?Qe0Q^EFUlD8yh`kbT?s(PvNuco3zUCHy6i@}F1AkC;RnL2nK@GM%h8i-6OGb(
zjw^gM#`pY0I~I97H936_6tAn@XOZox7XF8`pJ_YJfXB*^(<fB<;Mu(Br}yw$hjrJf
z<3!Xc)OpPOPSq=xn-b5G>rwl%Z(Gz5^zaIanFo_foGvvTmf*=Cw3EO1pvIf(w@l}i
z+I<vWmbI0|K7c5bhEUbwCVwL<iB!q4sH(N9%Y2omRlxgKez$(AW*CT0$s~ho=wwk{
zSwio(WG|oo+I>@1c#yf<GGLRx`WWt_b8s1}9YjM@rl+5=yU`4^NrSB2>b@Zm(KFi;
z_!-3wT+Pn+89?e|@;r5G<o9t-xA~#1$7fXU&TlMnn!Yi-fS!#q@r!uivuXMok#F83
zeBCmMu!OIy=#m=;S{pzReH<b*rnxu~b*skoK<ro0+Dv*}3Mr0q6vAa7q6fwo_;d+9
zPhMPd{iF09=i&YX7G%o@O!rhoS?1IhuHgKvS+7|*AL)%M<5kw5(f0s`Y*Hq*dg@GW
z)Q5H3I$%wgiz|ESuU?L%2w8okdA4o3QWse(@G5qBYGH|d?Nd@!Q2~+jxUI9;<tvP%
zZ+266#m5&cxn7jXLCM%x9RD&u%Y2|hdcKdyS9J=y-GfI}X>?MuNS^BQ=`!(t6nE5+
z8Fmt@mg`ghNVBFjSdt9yMF6hjm?knR@9WMNqem*1eFFZX1@1O&1oGIC-RCUzu~pR^
z$K+W8y!WWMT2fEPY4)9&DzD#3^tK6M+?)X4X6vK0SyP_y?|gN_hqs71nyV?7b%QJV
zPahZkfZX2d%`&ashA1VsF3#Yp#h08uRiJmxUF;;<G6dbZSb614JNSO5(lottmEdVV
zCeQWZ!8+SS&3i)!F8aV^mceNP*qIOCHezTQ-sjqQJHNN8eN2e=%C<1VU>>>g%nmu9
zhDG=!z+0~-K#A;p&kX{`+!5qIkd6V87c-NOrnp<0QFO~V2Qz<=2-g$R%L?LsYtC=j
zNc7Z!z*B+S{M=JH@YbA_NpASeJ!@p9dwkJ%U%{iV?<Mb&P6p~GAHw$HwOahkmj<3+
zrdjh%e66?Z(9XZ=my<jgP>AKIo)F7UHjH{ou++mn_0(6WU^Atk6u_6$e$m?Gi&^hN
z-TKGP^N-O9pwl`tXr9jAJFj|=qgg|d(e|8yf=xOd)8m<N^%r`%-N>(NSjVSRL8K@T
zHs3uBI}7ig`{Fm1_p~E)lqkRB4uFWFsFMe{{Pp7zHZD$-yBKVMO5ieYjW0mffIWk4
z>P=2@*-a1duP;3FbmNh4#91@|is{NkOfu|tgYX{8CMNC*P3*==JYT&4^q4osZ!{k%
z%6h|Y{{$R!XaVySq0>F%UbT-08ilg{;t|qlA*ickreZOFDwkApk^J7~WNs4}Z^Q{d
z*<TA5xunH)1nYdah%5V;hbO8<B=(4;S)@viFW*Y#MuefY#`6JQTxs!>aLp0-?Vs?L
z%gSLxUrRx=ujqW^nS{wZ1cqaSpRGxLFH6b@f4(R!-*~;2s7*!8#Uf^1WZ`Xg8Cy%)
zF#9<I(X}RzWq|#DQl+U_c2baYP0%@>n|`dP8UBsPMOx9=&%VTK(U`TX=!-F5lILAM
zodDO%5&~5Ez~QWTPDF%by1s-j&#s(K0l6yLzvS}+T``r;n$l&r4(QM-*pKs+xzz~&
z;d@uJnzrvT>u(h&3HgLuN=ce7`F5M3&M{@ZR;@LrPho@_Np%Db_NGcDLWQH5Ni~!F
zO@>B1o2-8ITU$Z-Ehc6$wG2SEi(%mUOe$N-UFe-;uV>xm!qYa?lc?U!nP&04#|Wm$
zi?lslqt6vp$UyWKow^jb7H!<WGSyRWWE$O?ZEP<Dj-tN=9~QMBW=qE+5_Qc8W4+9y
z^)>|6K_<tzESd0@Di3J?2CPhAaH9j6wq}{eZ7Nr*LO_#ek7JTh?#nK&-}UYoYAInK
zGK=a46jl{9e{W@SR##&fRN}yB1A;b_8j5AWkMYm60tU0Z0!}q)TbG~q;B}I-0~b}V
zInxS+?QgWm7pHfk&e<yCZix(9kgHn1T%E70sP$M3-*_Dsk8$jJ9<=6Un8*0kfmzpL
z`i`g?08h`i7i@aVoVQt_)>gla)y&?{8P(57%k}iIQ^Hb-M{PbDE`H)7Wa?zZPcKM5
z_<}O0#AW3ILl*}t`_uuK@pM^==A{wWjd&z#ZYGt#e~19mjx6*<tYse+2B(^?%AV7p
z3)We+_H|BpDhHcxCAmmiT=j@6X8ES_lwY;4W@z~BmOy`P$?ZYki8l!<cG9>>R}H?v
zm)hMiOu~a-yx9GCQRVu$E3W-+_fYV2c=fZaO8bi{t>CN~>HBnY2>#0TT_y%QCW0Q7
zSeq6L=$8@eCV1wHZgn-@@1!lPS{W|mvSM1@&kFY7OY90-r$xX)oBzAvnbchGj>V5@
zFJTvZJGP%t&k->Z$;nzB0`ZI^4fDGZsszfc7~$YaYJ5U#T$B(IB4IR<Vi4+?c?;~R
z1tp`kRp6Gd56%({M*&1VB0+B8QVb9cbl?y!W!!&zW{`XGr+x<xIUJw<0)4j6BakgG
z!nP)jZV0inY0?iS@eSzpt{l(27mIFpnvYofaECG?F!8L&{&7R%yK<Me*PV*NY{^4E
zN@#L*q89u~Apc)431bc&aDV+<&FA;7EK@KVC67!-ufBU3u$juUAS&`x4YMVCRvEjz
zLU&ezj{ub<e$IRv^hAgd!n{N6Ne(T{B=qQ!vt)+Q6>b6n1JcUAL^%|fr63SXZ%#A(
z2{AxC6a!#Er!2(a(1?%4Q9PoanFc1lBOvknQYBQNJq<#NX3{NGd+u)U^6I-#ujKLb
zqF%&Je;tEvzhcqKh)tg2=wB9>Wz2o+YaruSph+bV-kw%B9t^zXudUa5W41M33+7E)
zXm9j{QpaZeR8utya>MvzA;DX$kLRzRz9JLt*I2TZzd$O<J0_A(s#JYsfaSP~>!kz{
zH(K8AyduF@8f_mFSa4+^+@mbqIqK4e$8T=!q(?bl&Q2DtZm0f2>+IOYuKMn6P5Jdu
zYBf12iBDW-uB$bbx5)!b8Mv9&PV%2-J)|m_y&K=^Z3eO>@T;7@&$X}fMQCmM7j$=V
z#%SNf5VB1M@clI(B&9cVVj(wX5+brRGe$})@|d2S8XoUlO^E05w4%rcAqFi)^I#L6
z-#m}fx*%i;`QefmBhF<_G@MXPZ>bozCSsK5OFnYsSIq&*;SzaEE;hPtJ{99BoI$dm
zB|%bLwU;T=RC!D3W^g7~BNr!fDpA_Kgr*MVb2WT8b*25lpjs~k0W*92Ho~BIehN1r
zck?23S4@vfcwv0Py|nOJ)M9~u%y1@yA7VwVox0p#RF><tJWVP$ikL|<u?os6_3AU4
zdLb#MGIV0wjw7jqSvX-tmn0tma2x<y$0bYhOc_;|ArK+>-Bev+=hk`YpGq}X#=+I$
z47fqBTVMM8X?#73h{9Dz2mqnQD{`d8A4(3OgeL!^ghv0fJr1?i?`=3#m|`vRQT9&C
zLw$ORR19~SwqyxjOn^$b-&ndu6Z`ZfjxiNW@VXIIz#@ig9!(ke8uGGf!x(uAj^kX4
zv@*!)(crOjH(#z=xOb=>Y_?}dq|8Vuf45A0FSlLWx3buZrRct_r|IxR`P7KngO^y5
z!#jP_EazUcKe4fQk+0l*N7ObFjo7OeSRS3%7CA4Z1e>;=@U$SsgKM5m<@fpGHz-~a
z<67BkH_glvZ%nJ1UUlpa$GE|0b{;OY0HMrz%2P&60}nN}OO9zVS2Ujdb@AUO244T>
z=Y~%;NXNR6KJJYD_Vmm1$c+u8^k?77g862&&$RR{H7?HaDSPinPGT^;;ubdhseWAM
z)SEzL8>mND5Xt$>EQKBsG&RT%6e>%7!;J_yRkmddfKP7!d|W^GgYryjfR+5yay}Qs
z9G!?5O6*7{VH7&oUrfv3yeOi`-b@0drHaIPAR<du8u9ASF=UX<4OK(Z9U^>Gzz!?|
zjW&s;wfzyEYj7!PNeJ-k4e)UW8bDk)1B99cLJdP;J%#?+m|W1JBC@qMp0fj<CqNp0
zJ_mt~@#R;B{ts`@faQai>Tir@Y5^%pp>yNCYF<<z6A4{B-6WBl!ipLu+=bey`J1Yg
zV<^ol@G!#@&}vR$T(A*Om-J?W5D=XTVQC<>GElpT)(Ln5Tp-pi1JHOHy$=kek92N<
zy3RJ^fhhCV%fe2g^qyJC!!fws1&~Gs>XX{jxSOfO6RWnWYZ;wNj7#rZ4Oe9+K{pbD
z>|~8?`E@E;<;tBEPj<Dq3FwM!C0Bfl@jX!pSq4z$7KL8R(gk$`m8S-|?`mqDid!G{
zmdk^X*%1P62HyetlQ)IBqC4@|`5)lK;~75D?~hkP5Nh#0mWKu$4ioNbM8sQ<ZUmXU
zhyU#K1o#jd;>Cxb#1~>&eJi4#0b35}zQc<r-Nd32){TwY&RX{r;KRKu?_MxqpN{Sl
zX>`RT!btRx()}lU6}O54uHVr;UG1By%$^()5?St8R5Em{$MvdzA#}9%)T4KIiT{#h
zvhQKxRkNcpG@{e7jgjP;v3Da6v~D;!tf;v*QPd*4u-AohNMtm*?U-@ex@5gu?mo%y
zbO)8-Wt(A3`~xzK<JI_~%n$-pm#BXWAT_b6BomG~UM~A_G#u?!k*C40BP|z*VdiC~
zC&>@WGpUh%Vw`T5Ju_kCvvf_bDHL+^TMK<rGAKqu$#C6GPO;3J+lHDvKz6G0ptCqR
zCUBAm{@zUI6-QwV0Fi!al9F}nK5eDW(8=WUor^LfKcW6qrVCjE`}w(@Jr#4Yaoq*a
z-(HE?0P<dSM1nS_l>u^kN26XKEbJ<z++p72`p3*UgQp&^E~F6U&nrb^Er#u=h|6!M
zuWS_L|Eflz<r%Ua;_`Bf;iVf#8VL9wBQS%gDSUdyFten&&>c8~xTt~YRk?c$a1>(J
zGjH)GDtZ+K7&tcmsA2vlZoWgndTxfnitrP9m`ng~K&9cX^9>(PG+;}EKq;00(Oz%U
zCqOv)qj=$tTt)e9<G2dkrq4%>{!;L>@6}olm;};EJ<SdlK&Zj09>rbr_f}!M$ea7z
zY*g5OA0C-iR<3-LpcQ!&R;`&Ejf##L(R*;TfZ0nHY*`l6ve7{>E|p)+fvZVtP@|cH
z-SxhSF`Ow0`!aEYn%F?8{sek4(pjZ)V90`LDQ$o=P4?03lm$1+R;iA7EoG;NyvLSC
zgjwrx!}IoS-tAb{oU^S2&xHO6ugdN6w52Ij=9Msd;-!o9E+964Az)^ZK$%IFu)pr9
z<8EbUbiV{(GWJ;NrA{k<cV6$lw_Io7qSgO1HHZJZW4gjBgUDI2!f;$pezUFblr(?Q
zg4hp+Dl%W5Crqma3CBhw?8mq6!ap#qrn0Wyjd{_+V~k4G4I+Ny<2w7I!q8l3bU~UT
z|8vGgZrOqM*Z90{jXADKaOSoesi>!Bk~5b_LE>`y?eju50}WTOE1HN>i}oek+>W`=
zd!di#<?QeHCgTZEHwbvrk{JYgfWbU#aqiDx{x?XC3NV~N?id;H{29!FkN-26^NYPh
zSDCN!{)`6lhk(hPJI7e{A>dd0pBd_pqVD><WmWx+oqO2#H^-)ol>jWdVobSuZ7%s|
zQz%)=99XHF^R>ZKdtgV$nHb#W?*tC-a$O0UOt4LeY6P;?JeBbJ)r%3`Bo1T*f?;}e
z*6riUAd?8a$8E5P_hbq~hhzkI>dD27Abhuq3H-As>zBjgWG5}-gFuLNQo}K%EsZ8?
zPmgn_f@f|JDEYgmPA51@BLhFhjW=1xj3(a<JnY7(u9f&o3T@A1eCG)1QbkC>*N=y6
zy2E0Gf;||b+lH7LPN;7vD%T)jV?3dE_jG|2T;=4_BlyszI<omK9wIp#_C>4Mcy?rO
zgl=WVNrpcns;|Q;tOf!}I6=^@c)~hKN#bnM$FM$M9fsidhPOY-!zZ3Q*l6ASDEL?%
zPGY$qEQR&;Qoy<omJYC_)Q&!EB)_?rx$<khu&{$dIHgvveX=|9ZK6@iP7^*{sUY53
zRbA}WM`p!*qhIqHBNz?3(`MfzvlZ9k;LqUH!nv${3%yi9?TKBz9~4X|3au=!#kS>7
z3~0NiOwaAUiTwK+Z;Gh-;<GTmN)_Jijlw>-*nBmd@W4XYe9>!KP+H$iHYa~uw4wuv
z=)x;-H@<0C_O#wxvCE)0+Nmz%H-dwh7!pmxMz04&{IO@$msp-<bM%NVc<84UL(7lD
zZ|v%+aMdu;Btp@LM7la<Cv$P49sOxWe7Ze-JlTtg(PJUX&A5VmM#Vk}nHmWRR<TVP
zc5I|HpQ?E5P_xHW{(iLHaFGbt=|`H!ehkGfYO&U(ULRB|v@iKbrK4VQd(Xdfp+0gw
zfntWjr)IE?bfbJf{dS2Z9_d9jK#;UQtZJCdlw+(gLM7n0?T<jlYd=;i>mKU&%VY1t
z-vaz@YPurcI0&1e-Vq8W#P@eg6;_bPOpHrQRQyM%m-gv0V$WQH`n5suXNP-lm?WN4
zOJv;|QKH+D*x}CWdKjg&C}9<96KQ8i+e9Cb?(05{Bywbys<JLF+XjM0u9f0e^FJ$0
zsVq+y8f4JEW!b)@PT-rK%7_UYH!F;7fO*IvCuZO6Ni+0pt1Pa^g{LzHURH8B$H+O}
z%i_d3!Del&Lv)$PoXC#|R2fmvU8dj_IWA^?TyGi^Linv|g+i}ClMA|z{$yODR%@1>
zL4|a>@aZdwXd6LgrXL1Kv~+vJe|yMdL+fR7OmL2g4(S{H08+xKa@_|vohiB4%QyJh
zjP<*d5cK)C1k*k0DHGclQGG9e^GR?%O-ohVl=!G~Snl=q_@Fge9&7wm0vBfkEYB5<
z1!LKdNn?I?HQf0I`Q&2I{`lJ)nW<XTSyrpYv56qh$eu*E%URjchqd0vmoEL5^$R)u
zYg|9pgD0hE#^e0D`<JvESdec>No4*6R^g6ep>{LLKxnfFYoxdo_~xU|%Zhnp3M@vH
zegfrDtkJOp8z>YVw!5VVkk}^v_V50IG$a0SniVx+;aTQXUu%p2#>J40bLB7tfmKPj
z{uM1NaE}+)o;@kY7SH_!az@^ne>%6>xHP+Crey9X^W#NS@$N&8S0=H`oUaU|iHdT+
zy3UL@Od>_{Ap=|{r;Ne(#IlcxU%gd9EvvTPy)Cc7C)HzXtY&0Y)!WhZ@_et&r@0BY
znFJ*=jS^PhdU%;;E)z0j`KW7rI5&{7w3uA*j-euOGC?4IvrL~`kY3PIEvhgg?0dxw
zc4gG^3T4zw;T>%nqIo}1BpZG$dcZQw)Xa~h{9!zdNE3aVw$<(6;LDugL1B`|Es?bw
zcGd8z-TwU;;Xx1eJhUIxpL73ztG!PHEPuK6{XsO(gdXcM2*!#NgQeD%OV!`(JU-s}
z3xqG>d=8Y?wNN(y5_=RZa-f80D-I&yatwgr!Y%kV-vPPABvLy%oO_D8sFueVrb`Jk
zu@<)I_W>n=_C6A9?D5AXCBO=RO8L}h_3eYV;y3X+M`TG)pz-PV?by!IZVN}MFU3j6
zZxYXkvXsB-a8^lwZ3HETTNx09b#~IDpRjuC+(J2uM2si{T{OrpEYk>bCmp>7CM}WT
zE+2M#9axCBbzM^&-Q35Od<I4lykK<@|I!rE3u%58(#%3K`F`Fsz(7@lo%X8rfFN_`
zu7C*>d1|5d>*B8q^LZ^9J#DA+r6skg7q9x@^#t|orNVR(TU^uIM0jZ89LNEzkFrNp
z^@{A#eMGUF?!0qJ?%U(yI}TSayZajfxr%18gBtd4*1xV=2T#-IJe6v<{}HdXg5hQL
zW2WG-oXvz5B_>FEqdteT3cD@E@#p4fSuG8)d$1=KxHqp%L1GS-zm?=+zS2llcaN$J
zttX*R^l7-<3x7s*IK5ss9N(PoZ!Fp9l`f;vHkUx6_LLAkd!;8YNf0BFU->MNQe}Ku
zF%sa~b#GjfhYx1c(#W`lG%>aSu~`rR(@Fj!F|lvEG}9l+>!#vIEyNfqLsA-hF9$3S
zVw%j_dR|!V<FJ?{#z*vHNyOaMP6W-^dR}TZs9P1fKhvQdg9`YC$fMV{LM@w2<qbm`
z-@SBciaD6qmD1cSyV9s+=gAc?^hP1~pScdr6d@FuU8JqGhP4K^h2`~&<XD$WeC<>V
zcw@hPfoHiNDt;^qevqQ^is?R4sK-Vyi>hgp;Dc0muy^Jd=O)kVA44h4?<3ykq&nV6
z$@Z(3)GTqdpDFsnf0Iz(?Jh4W<y<0MJW;Czzg*Xo=VufrjqW=ci^o!<k(Nn>_5(eg
z#4|}j*`{~fAIeX2S#`MHn4|zei#PwCe0zA6Q8$^O-XrXBG~!IH0QPzO@4@$9a_Y37
zTJl&IL&k8_VbL*XR8yCuMsYe31tv(_E-bm*>v^i5SQAa8Qo=&SCe8Kq2MJqM>+|7>
zI<e{m*)B!HJ(qNj$H?Y0YW5m=)7rd3@|))vOdj`w^M&P+e5#dn(z;WT4UZdhbgu*i
zc4Vq7VDw~0R6>>i`;8rT>#mmdD~q`EPY;nn>E?NUJpE;#dD=?bCe;Ofk+D?0(ij(7
zLeCwG)zqEE$=m!UO<ucI(iN1F<}9gwnpvx6;ZjGYPE4N-x6&E9qTF4W)ZV9Lcv1*@
zX4@us1XLE)PO5wTYPsUWC8BmdDR_LM?Ku7#39nBkRh3o$HYfFTwZM$Gzs!bnxLT4i
zC)LEebusuJBSFMN$5yjGwv5e!ge|(VoZgh*92y10G=AcpC6U(O?N|>#m$7N(NYHGv
z7*1G1EIxk*k0>_I4u(_=zS%R8Do##5FO6pUtR+4BML1$o{7C%L>bZC~WBr&GL8hHz
zOPi-=+89sR?Qn$T#Q>TJ5T(zH-#F(<k0jw&)-=3Gy`JQLClh15&7}nGG`LyIx>RDO
z?`{82Z~4dQ0qC~z)51i^xUICKy8d0YM?EA4f2shY0h>S85@VH)c<6sR?<{#96ilE0
z64Dj*u0m9C`fi%=;`JCbwYMI3%y;$eUl^?PhqZV=eCNp9!x(@edwHkSM70yY{Wr$d
z+vIz^(sYBHrRlfjD4&lB3Cw?iK3Mz(%Al`$p+Dhfy}L<fg%q<^`S>gv{E+TPnlzXU
z<14spee2}>k?73FEt4seI?Oy_l0e8oXRxXEk9#DRx1*h(`qB&K-4KYR9`t9fk?y%m
zjj~Mp3-pzK;;i3wZN_Pe;qc%_mLR28S4r_|&~KVIt7Oa#a&T=<-gGpPdhdPcx&bdZ
zangClE%@7^#an}>S_QYUEKg``5V<H`Wd$om(H2TjyaHp!-@j?3GvodA_D87!=72M_
zn{{t&Ze;s|a?L;&r!s4GPlr=YuKxFy5Cstf>7To?qmlVb&UfKy(cS^@k**=;@yF>b
zYfsPfzYZ~yT+_ZOw15v)EiPk6aJOi03+Hm=L`IuM<>nz;&ENP`6y>aL4t{%u^(CRZ
zU$Av;KmvK5SAu{y#TV9vTYh&4<i1bV0_!(Y8z#^8uMAQ6bfNI1bs8I7wpxd>1H0TL
z(zINVH6Y{Zc|YDT6npDDmV9+%Ev!qQZ2Zw)&P$p#n&&@0y{E&uV+4<?eeo|13&QT`
zsORi!oS)@?9c%)$A9oPM$}>(iqJgNF;AeJxL1u2E4D~ht2`2~U4wAn>nC5HXMl?U{
z^T$M8X@{Ru96U++_1fSskg~!NLn~F?{ouGA-E|xlOGOSDKO2EKt#A_^nnLT=lXl}b
zV_?wYwP6G-(l8gJaff<D)D_yfR*L=rBVz|GN)zRj<OHG%`LnAee~b4J1A)|U&}?T1
zMi4dwE-d3utFw|PzJNN{xfj8Bf+n;U*@Js}v*N-~5MqzwP{W<f0Bcn?p1vW}>zEK{
zJPKLE1SzyFweEwqagO%bhPARXy5C#h3PqDygPat!qJRu(qHK+QJ{8ajwDTC&qkL+&
z-1wQ17}pRX#1W>`GB)=mZrUy7LZp*zvwA*$39BT`NOx6#WNteqp~RvdL@XX^fT+7a
zh}@|d;Y7kmI9^M00cHZm@6|PKN@j;4a*lcYLZzPP<<#|xB98j06;U7B%4Hv7RMZ{V
z!#6ZnP#>5^P3k;6t-Rdz(|TPU8NB>`&&zDBda!YxkLaA`z8KMQv1vRE1R0}s+po{6
zp6LX_rL(5r8du{=w7#Y_-Q2N-wa}Lq(uq&j*P7wh9TVf!$Pgq?)hA-dp0BI=WrrEo
zM{#nZ4q~RoS1-isgAy9F1?C3u(al)@l$+OPdVHGA9hSEjycl0IGaKBoh;eLid*Ynv
z->7Sa!16e*S4NX@V1(12wyDCS9corH7n2_Nj#}r~;k~y`jWJodO6$=quDq85E6paI
zwie2Izlr?$$-EjKAhlX-;4a%gA?G?!;7Xd=LS4p;<kqt)Hxrb9nU!$9nU|w4KxCjn
z+>C*IY#JOiZ@E#Iarc2|CIABpI91vx;mlLzk)PAn$;Ym}NyY@G7L6FzLmv#(scePw
z4YNY!v8crvB1|*+%V4ud-N6nH<&NEweV3zh74qweUczD{kVCJT@6orL&*5nbsXgwB
zvL)-KNb60nMd=YCK24bO2uLol?8&OB{2A%Cx*~wFP}~N<^3mivEi72%zB^m6xi9Y#
z6U}dGc|`zFLl|^x+k*-KRq#VoUdb>KXq4jQ#Vq7T5(c8I*h<gQDuyJUO9XtXC-bZr
zNy<~F<)&@LvU>xr<x>6S8uVV%Lp(mpHnh9C-$$s31e2#`bjE8*EfZ7X@>`<QyUpX?
zrw*Aie^<x6K|fA0O_0KP)U`9vLW6gb@>i?o<p<P)80O12Bh<u;hQvo;(1~{#Yg557
zwboWH*J;FF!zrntIeN8DqfaYoJiWj4`aVteg$(qW>d^CCctd_u*H3{lx{h@%u*ppK
zEf@HgPmW;b@&d;%2>fCKYEtvJWMf8Uv8y+Fx^fLI?v1(>xq8OnXD%|l*>}-i6@W=s
zl=AD`?;hy=c{7D1lk$}0AehB=aVd0#qDZ}5nBiH@YI}P9Y-t>KTe1#M8|85K5$-K0
z(>^*Esbhejs0N|i=wXf(*BI5MALtP#Fh48EDgwW8GAh^rcR&+v;Hm*dg8C()OkHqR
z-;?}=uK!~1riI!vxLIk%ebkbf3?{cE0cHvdz$!%jFRPHQyYlQw=6{imP3t&Y<;+gL
z7##DI&(v`)?0maFl0%x(n)_-4XRoWROlZx1+9e|e;MkMCC;MC`A|?l@RrX6DOR65e
zE_RtE-99;7Q<X<Px(v4w-AR^RW~QK@#hC4&)HmNMGo7bj)VUA;SbkLbtmka-6!yq!
zyzN*m!zLK5?c#Si;KY=;C`Qm6)j6i{&ISp$j>yVt3kl>RdyB%mO+2u#uZ7nz8ugf(
zrvibHXnM1tk;q}O*%UB@ukHb>x^LZfKeI3?q*Z%GI6v?%;RljCnoI8P2xEkQ4oR|R
z+jM?y_EKQMu$sHD9HE%nBYxPiT;$lr)Ljpop955qhOC3IQT1oz49<tVQCLex{$GyT
zS&^~sY>Bp9tm#RfrVSxVxP?y))0LAqGFQ|$*oikkN~6tHf9y0orGEHOSH3*ccd4C&
zp2}h_HFR4I+Vq@l5`!gUm>I~lloJd?X!M@wb5Up!`?h#K^>KO3EL4-(ZnUsSPahf^
z7R~2D!BT=Xteq0?P*-l5czXD|S9M6A@T1Y7cg{_w@dtTt?`K-aer=28>pFY9Xx%J2
z_}Y|49VWJnK6WfXlp-&Z;ltc<4+19A@rBh5M{YD?Y~D*u3HmBfYI-{fXHqkSc2f4U
zoQLTcO-KG`0Or&0#{W!N{*uah#qinb_)YMIUR0@iBtrccpnz;g?mX4%UOHrqL#!NU
z!blNC9nN$(FQg(aon2a%#M6ifLBNR8Sq&qpj)*aic1+LUSaW}XV=jpCUQT2yiSa8T
zCnI3Z9xsT<Bw?Wkt$XW^2Kf}@J0@&#gozM5!7<#B4wNXMyn;aElnCr<i)<qLlwpxf
zPbMRr_7qhrTdXci>mYZGwKl_|!(bd?G+cjyte#`VKs6`Mc2LTxDOk{qGU+Bj9f9?^
z8z8(Wf?(1?n~iHfD!!m_gx!hpAUssumV(2JqK6@*VdjE*Hg*Ovcma0X$aSU&6AtAA
zHDR&kuv;iKI^`Wy^D_-xZZ|4`X)=iE^VI0Um)bL|uRn{2x!ex3Ki(!t)qDA1!M!9|
zu$zuD0m$(%q?Yt}FeR=1?je1xWJ&_r>=^v5hgz3^KFj8hjJjYScD=QMlJ4m2oT-Y3
zUS&))Tnm1@ASYDDbHf;OQoCDI&P+vg93JG$go+U9&3vZBvLi;#fj%W_-CmdHP6}T5
zn(E65y;4L4e8mb%$dswGw<`AocB9nkdBNXEnAUjmp<R47P@XGTFV~ZeK=3{lu>OYf
zK3}hZFTD!1*uqFl9hlSj53#wyx3fHY$@pS2bx%t4^KLIAmDe3iF3W|k>>keb4x9@<
z*cHDIq#=tL4qv$<+_Wi+&1r18Pde%PC0euA*8y^gAV-W7I2pK#m>E9JU<BJ+dB`vn
zaQr@{@mk)jFnK(=!eT4us!2((XKnwR$8IXrMqhj(FnaMmx^i-AJ6v)<YqQ7ry_FS%
zLrNZ9%hMvIa*X#FuXtV?fXlGvljGw6c|PJI&-(QUMy9+tYKeks7P2v+48#V|#0195
z!{u_@1n562tb~90%m{$Hc$n!x{83E^e`C%T5$9K^mb`b5dc(DOP!LO?s1TpSWB8~}
zq4WSJ{_B$xiD*gK#IxP+M+Ppu>oLah-BE*~6m{|5+E3|(2;CLk&yQ;;v|e#KcxmiJ
zb@Q-i$6*Dcn-G;l0t>mzb~7#S4BS4vz<5tT5dLt-GRM>G6)}yJ`H}e3XmgZ(g|V6^
z>d-02Rq5-h@qvC{&<WK>mU`u{?P6I;Rb|N0BW=6Uq^Yjyz{<FG10!&ZsnY{s){$|p
z4#2wcc$(v|UFVz(*f)HW1rHo(Cd^lyn0XNejuYQ$Y!;{ptgZaz0yEyTS-t);-?G&w
zF~J-sDhy@Bs9Z=EVtGTKb(;3<!N2-DT5KJ5r<jaq%K!jKOR>@T?r2O9O*lr(FeB$<
z#UM(qyhIpLa$Z&FNyrXxL_yLewqbb>ChK9syz*0yrOJ$Kt3YlDZS@Z4-HPRK`pL4j
z`FE(B<Y#DVWn}^0CGhURp?jm4?<18tx~UhG72iH&P<<#Z6gm5^Fm5=I3z}5U-S-6T
z`$s?xoCAqy{&V)ROt_FOE^KN}oAJ%OJIBFPVJB4~_6G^{vcl>w5IW{V0?-C4eC7N6
zVw?Nkx<XfKqBTxc`@rmn5y?Fv2De;)4<=JB`^JT~|ACHynK+WjzU5Cpy*z{le5?!;
zF?dEfsY<e}+JAxK#eN@GpdD-P2ctxG^jOjh@hLUG8-rHgvYbUp-GOSmAHjNapd~!v
z7H0$<3bD<_1fji<g=QR{9$pcFIOv?9h|{;%8LydZv&af>M~CQlgO4#xl&7B)h>Bjr
z;4>8-!-d%w&e6C0f+sclTF}JRY$QRT;;<wtzbQ6fE@O}!Q~@r47jvwk2c0FUPG&2c
z^L{#_jsJG3|G+TlvNA*CK6!ePO$><+9id>q__=YWvPPnsD20&=LVo+&v=he>rz^S@
z?f-F%reL<D7G<%8tDb*RDt~@+i|JL8-0l>UW2H`d41>-%ekg_bfLCb<Kb-HJbJs;3
z9z9j8hgV#-m3bYHbqo;z^!|I}CSD@jeIWBxUdWex*@wkhFy@~hQaqK27bUmk8?H9z
z+2aHH-7S-i%8YhHeU2<uxp{Inje@9~d+~EI(m_EaTiqW*vJL(1k4b;ixQmQ<9+Y*o
zQhgYdhgW&^ahcVIr{3pn$$D3uG#fd7Z4&H>E6+B1B@>jNo%L2QG;BHFmwxr#8sBq&
z{$-;R8t|4!360=#N=qIMOH3A2@;c9)GscK0Z^{i)2c~k%^WOq&PL|*j!@<Rmbe1-Q
zb7c_+^yM+qt4rM*gE)@BCdd5SIS4(g9ibjouSThR(q)u^rO5eA>y74K+RAQpwZv`P
zw6!RMwI(}l$OHXie!V7fz3s3BajR=nqWDeeJ=0ca97SGHQMU!jR2!r17`j@hq`Uuz
z8dKA@MK9*mzk-%|#5Siy$FRxqdxG&(^ULm;$P^CxmZvtR2`>tBF%b7rnT5AwTCwlH
zRpsM9J@>s?ng+DUVp>5{uklJ&L&;OcWwn49l(<kIa6>3W!DpOEfRpM)XZgryYKkI1
z+~P|2Gj4yz!qwO!b*+7U<uFuxW`-h{iDo$WLJy;dDB!j%jjb{durhV?2m`B6O)Iae
zg4eDv>R9=?#`n5DK3w%avYh}ix(!t=(I+^XzRy?7DZ#bS9;z;pp_0bc5Fr=&WOr9}
z;eLmWxufsbr9wsbS?P+3Bm#3edU*Z={)$Qaa-^wLwZX~AZF-r!pL|D#){lILOmi@I
zgEU~lMj8Paj>9-OYPf4!y#!-l(<h;V-sNjA@HX?3asH_?Q@irREIhMaXg?d-`ug0v
zjjjw#%~o|?+wffdo!X&!$VjyF7>)N}e>IJJuqg?Aese;<$oq!5il5XRd0+w4)R-Oe
zK^mqns4lhfZdud==Ol7ywdCM#Zr3Qo3kJp;#hPlm&dW|eq<^Pwsa)`7FA(VVxW`Ef
zk<k|e6Y(8ylThmvw1LQAWLekUGmM??z#}HC2<6L6JGj`QvB-fTx09W-rWT)uk`%Ct
zS5$C?D{__@hC*l2p}l0Fuci2d-=KiizIAr%nbvF$!|bH}sX&2LA2<sBpE3r5f9jk#
z$={p#Qehjb>VGfhC0g#;4LVo+i|77-Mcw~5)4dGEEnxEhr&9I%{9UPX)3W7u-^lU<
zAfkBv@01nMmr=U5K5_$6zy=yt2(l!F7KM#~*y(rF^tC|pDIE|f$<0FuFGr@5Eq;hp
z$tyMkWtNy*AIok@t%k+&Ho*fppX#NAD$QLKa!oelC-Et1UK`-`2IgIp5DEWo01<;h
zXCEc<D7)}<Z+i9v*mAmaDJH`<;zStL7C)}Q)-x`p6Lt&Sjt_mUvoXRMGfA&II!6(d
zqy862jd&(6O!Y)fOc&Vg17%AIZ{28%Gcy6~shymOzm_ty8`3jYWJShy@2kNQbK_`8
z`Erk3j1A9WeuEpuVLX~9`{8T0Ay$m_s~IejMo15@bqW=PhHyCj%B7l^JZOM9o$ooW
zP8HUfB+V-xQ_|cMLgqZTRKLBaoMwH>%(zci;LOWK^0j+P?^%U?!1wo$kyo0x$@0{}
z<CwM$=}M>jMw%+?wz(goWG^cy^P7hG^ON9TymKlV+%-O`fz~g}yhz;eTXyxJ{ij#<
z`vdv<Z&CGk&R1)+<Olb>-4VgsqP8HbpeZlNZB*}JpOQx9@02bP_#jiF*{%-v@X|FY
zN2XahQfRRtI%rYn4g~mXwCvm+_lv6x=pj@^@ZM7N7t_+xkwm#bif$?P6jQ~dBr8jU
zm)5@_!5gfKb0VhF;u~6NIj;-*4pwubElrS~+BNx^Zf`ja%!;GYG_jeg=iKu*Kb9`=
zk<_l$mPfUMa|*eB@0&`i5e1`~i?{474w23(@<=zY$9x0*`}`x-ELyR`NL+1~vNvz`
z!?w(A@wPPA_0OW;<GLGUZbR!<OWwN*%jY-Xr#4>|dhZyhj`ZC5p3(PsRVdL=>OqfB
z#>-Ea0sjp3A|vvJi6Yv7wS>U9Qz4*a`Hy-OfQ>0`I}^|fasO$(D$v?n^KA%s4_kSv
z8drt@&p4N=KEiG*YBOr0`h*E#`h3F#LWdAtO4!P%`S?`O_OD?DIM7`en>DbuY!;qq
zx=U514Yatc{2E-tmDkyugPUaRYMuzFf0nQza+BUOFxWF)V5<tYkZ$$8tSoN0pZ{y!
zFzhv(S)pt(f8+zLtqOPNS+C<B4gSxv!PJ%mb*l2`HI?Mj)e5v2t}*4Czh+)K_KAOf
zVlE-R99|joNXNf2c`|c+OUZ}kVb;BeV|K-H8*NWS%ncmMzf~{W_56g?ltXtu#bf+C
z;s{%?6oZ6tV$`u+%kC7!MRv{{GDlyPo?Y!FzXUNnC3%x&B-Aa;XNvjuPS48kn8=l+
z#Pc3}H@9<C>v4B{nP4K1ssz?Yr1*Z&>1xpQedz)X7rWFbQ7ECPs(67MG_(X|3Z5Oh
zQ%zr?x$RPylc=8h2&$WOb(^i1FLt`1Q&GX^th6$<dTNqnJ@=kcu?U<&qi1mD=An~L
zf%q;d5IjI2i^K0?!U)=5%1}nEYMj_SbK<3`c+w^?9&Cg^@(+3FACEsv%6#ct=I|6_
zVCW|=EEOTykmq>uk46gs^8~Y)sy=)T=p+8~Bmb|rc;f;89u#S<Rx|JAVHmzafALjm
zew8tk9~irgwOXx}VOd$c9AV#qky{x*41vxn0lIc^GQ1VwRWz)nher-d)k7R(TR4|G
za^rOY@p7?H!Q!~V$Y5VFpU$%aKuy}~Q?e2?VQZ1TnNyKQAZv7k6tttaa58R49m4=~
zzO%tK6QeF10ervNYNj%EXaL$CMN)dCZyjtv%ViTuL%*4RlZ??B267wgAPc9k(;*Sc
z%APu@zGh=dC3Vg>S;=&tAE+A-wGW;EvSw6z2vBfa<rM08jC=1b(`(c28Lno!^2-Rx
zQMs}XR50B~u=mvT4?V#*=cOoN8j2C~C$fo_c$^+VxWmrXa22lU;KWX;s&+<XyQ|-d
zrN!%*CVr-@*Zn0+GCG|liq_{QyP7OB>@6lOkJN_|AH4!4wx%O|1p<9PAJHY*=wrCN
zXk^I|{KjQ@i2@qTbxZniEFF80|7t6AyUN3sCA4(dLxCYEoTdG$IQ8eS9i(Y!-Rh;M
znXOeeOKNhvk09LoTU_ksvkpQ9+r*<I_^3Z>V-ea@e5kZ5eNjHlXXDRJi`*~pd<TVO
zdL}<YMH^{GikDVkxDAtarBM0rG6yD8Z2uQ!?-|vE+OAuJbfgF-N|WB35PEOY3B86Q
zT@q>lQGy`7giZh<6ltLcL^_DloAeTpCQ?P3G(i#9m$mjjXTR@0XOD5dKN%x`GvF!r
zb6s=h!k{hTOd9I*E3Zl#?z+d2De<R*g2>$KbZU@@I?oc_5C!a5X_$8q58X~#iuf#W
ztVP-u)nucQLM=@k$RMnCXfKWqT=Jl|>%X>TdogEL!?5V9B%oyiR$w|UiOeRb5w9|p
za#N_$LtTCrn9HzbmiN-7q%LK|%zM9|cH>NGt7on8&-2>#Zns<a@m?eS8B61-t@py`
zyt`He`k=<8sPPq*kk;7_-Hl*R7UpKj1y|i4AX^<dcebzcd}^n}_ad@`f<k-5xG*bk
zCBcUXBMK!a#jC<ZVb&C&3cPfY{Hp*IpL>rD7vE&-Qh{{IyM_Tj8EN5HuDi0|hm@+i
zee4$8&J|CL%}<_4(G%;<cvS%xd^RUY9H+|+RzL3KDuG4c#;nJ&Hv3JN>U(HYk#O40
zT@g^OZBdEprK3}+;C8c+0<lF#RhGHwh&N9nY|!*t-DA?{z)6_N&Z)W*jHkixD(UMV
z%iU&&NW-k%5m*S|hwyG%C{&-&0mC;=JSke;v;?#+%&KtTMiHM1y}sA=v4BDQINOa)
z<WX6gXEsV+Xh`Fs1(8Q>q)<Qt)vIXP>hyxa#=;>@w)ie@Tzv&;;r&hboci#5vpIjE
zX~j)<u1vE-OTvwJMooyf^MGcFj_HU)yLVkR7?#tOI3=hH*V9MiRU)^i(_3BzkpzB`
ze!{A-@yJw8%Fv~!4*7vd&g0=)f1~r2lRfdKynd+AWTKK@UjE0_Jfx3MKtTf9&OMi%
zRA30u?MY{tlnr|opo3}m=NPb2vtQ;KM66~Pfsd*c&?twfdzF1xo|H55Ts0IAnB{sU
ztO5&t2`aT8r4%Df^Nrp4C25*&@fqIB#!oD<@f9t|b^i2#rV~@a7~^U&5{|@wOJXfm
zqO~0{fVIBtzRF55dOYy=^{KGRO^}|azgxq$g|A+t?Tz7WZRZ*xZ|<io7CD%pAlz%3
z!2RU}?=SN?AL;r87*s~fJFgx;ar2p7fz_s0j_|R4F&pzeNdI_+>if|lDLu1n_ii7a
z!5Az1yXF>JJfpFp%X=M*IKm0#c)N%j$Y?<}q=c2Dahey&6L~>DTGN^&w8GqA&&cj&
z4f7Y_>-YLu-+$p}AdjCwlhi=Zc25WKMerhTYoVZp9>n2xzKq|pmQ%qqEyx$Zn_qX)
zguFjM_2<`E&>q&8!^7RMdfp*?@&BL|{F8|wQ~P3c@k#|BHd|T5Tc12#Ac`BWlP=2j
z71<1AoA6sQQ;YnW4RhB}O6H2|l=wF>0i7YqZl}p|s1=5QEGkO3a3gJiOfpvv9~t8@
zlpBysF_fnymVDwTxy@q6kh+L84?0v6hEJLqCJ&v^@DnZ%smfwTh5+I2a9lo1K22t~
zlR|12FML3UMkDGNlFt4q6-FAxQ07ixEWTe*^rNEQM0LdOo{0%v3jb}2nS3Ka?-Dha
z3Ixhd=`7gz;;IU>`s1;Kpf%%$TBMT4gf<W*sZIdZ+r%1~h|92SHa}crOQw31Fa+#~
z+j@*K2eZqfkzd_E{mJrm{*>J6BY07NC*5W!;GXbv;$QQXI=i?QIvLW;dH`i_#DG-?
z8-HCb+sVe$^$%;Dss+?iDNC1mFa+z?MfFs-+OL`V$C%(|tK++h@6>PB!)3z2U@BNq
z_P$|;-9q}LNP>&$W1w4}<GL#@4Hd)3w{iwoJ~JWqFDks?B+CD_!17JO8X34C_1SdP
z{L7D|S#R^#Idk@cVHz2#T5yyuyF2l@q<^LPWwLqx-dDBQAt`3KruOk(gZUroXPlA+
zd#0D$(EDbq$y3RCq%a<4-Tty;Kp%#&c*3sg7IMc6VnA6rj<o?H5ggb<J$Vz$3dy>J
zXpsln{1yB=89@!0x<tBkGl9M*IJ9LD?^eoZfr_`=TJkBGvu`z@GBI;aT~_XR5gFp#
zKbj_15Pp!)(z=heOa4+VBW%S_agKT4cBl6MLFi!6DrTF{b1*I^M?kexF-to-UwEqE
z($MKlfk*Y`swxC5TE-i>C~sft1+SYMzpN+4`UE#irDKdo{xo<xN9P@xGY;2c@BuRb
zlf!RQZo80aRT3jF3B0bF2wzkMASo4?2u2LuRE{k!Wu1<Ms`*G>Mo*?|=ge&}s*^$X
z4bB$N6H^d#l{w=Yj+*>D$>c-CA?9y4+w}}p<lDo(W}isR$h1Ui2nY0?UH}2fS%V16
zX?I_>Z~$}lbFy5<3!nsqsV7N1Ja`N3`qLz3e$vIsE8z*S;Mu+FD%oQV2=KnU%Sx)?
zIu`8nsb0RiQ3+<acOOwH**1D&zo;8h7$+*o`WHcELuA*t)v`|R<waahG=}0f_0cFV
zI&yOfZ7d+Q)${Gyyp@(#wTuhC`-EkoO7XpLuhi^&t(S1%t-m2DGP2QR?gDu8xnBw|
zBB3gfHyc1`=Xlw&^nTFGN+GNV)_@D&UYX>D@b1eDVFONSgo^RYZTMvc{+vFcb>1R6
zsH_Jk0xrZ{+4QAcEfh1?lU5n++2Z~%%YDZy>>uGCKd0O)pqOP=l3sZIAXTq?7qp2^
zN_x`e<Jiyex6w>7KnEJha3CgV>zlJvE<er%y>7}omE;1tkx4B*kBy0>zD|XOwW-Zc
z;@WETE`$-!tsz+xCUF?M8Hpb~mXWpv@>?Q`7L2!-B2?)s@F9kW0Z;#-!Tf9Bb*b&Y
zmf6}d;(yC9*rG)KTZjRKS75%T-*X=Oi@@7@SmcOvJJyfI(xCvZlKgPqmz&O3l~63|
za@_rCt_qMNg+kalLA|7o0Jz)y&odE%x3prfwSz5g*?1n;&8<kpOuzDj3uA#h_bhlh
zbW5m+y+w@7#-!CJL3?e|9}1Ws&Az9657&He#@qW?r2Iv#46W$;n3C0lZ@2Al8_h~4
zG%I=Y=3O2V8$Vi>xf9I&%Zzs#k$0a@`{ga0COU)chn2J-z87zQCC}%);2}KefFf-V
zsNRR&to*bd*~Lc@_*~E*u4R^YubHx1>FMscpjum{mSJIR^HLwEue_VCp4D28zgPdF
z?wWKLb*Lt8)s-Nnq}*rSW!2dU!e?t#?e!@}@FnXGYS$$wnk-(#XLIK8$m{5E15}i4
zzt4^bIPeRUeG>GL2ggRZRjxuW(a046s9P%v)_3sy&r_PzFCXR3JCmI~^_PbgHy?C0
zxA80@G6)Cm-xuUK5smP5!F+<XWM4|BuQVhd)~T%}DR(6Ym`{T&o&FED*8{TXzX)(E
zq4$3+1h^v1G_lDZzufZ;p$Ua_m?V67ushk72<z%~Nwo<mNsPSD%|rvf&(IlvXfVPX
zL52|XVAx)t!Q=Y(!`m1uBIWQ9Pz-$EuzP>Xj4=8MM#+zWVGx=z!l)&`HOJv`o7Rq|
z6pbi^lB>4MmA@X%p{gL%A>;oT05RDmV$F=z(HWF55hKGAaVO=w$N_$DVu<*JDZ_xI
za+gKKN>J5}98zfVOav%1Zuhb#Ai(VyX+YIgMrrIXM&dfHW88K<=emo_LON$q3)Ue_
zxu07RzbJ3z(&7iv>a{Vcrg0(6pYac_UMUHF@^xVwK`9ET#MrlLTc>=9uqmahxvZYM
zJ%Tmi?nWA`Np{0!_blp|g-bfV>vjz9_dNXkF!xp(x06A3S5-y(;V5X=XoQZ&kPPxR
zoOoorhH*gs#mv$Wt2_(Js4a@OAf>mBi+6QWDcd`96P-R4wSeBQeBF?+wSfN7!s4=2
znm^Z&gxm^zts)5UeE4K+sH$hDd_^2N`upu{z_a7N;pg3Z0|I`ew2(?ZO9wY_waF>d
zCYy%E%8U6xbmz_i{rl`A@n7bla<y`=Ws0ZWoFpcj6$6a&>@{A2Nh(MtVK_fML3fip
ze>gydnsFdN51hy2jE}Sz!Y&)QdMdDHE1u4(FXgjn)E<1>tRGI`_&JJKP>cZdHJvz8
z$bZIdT5QV^&x?4C7yY_bDM&(B|GjWk0P7QIhRAEFR;R_dJdX|7&RM8oIxTZ=1sxpR
zb0iYogxoH}ytfvWu)&WM-`?a?=5N&6NptbGAm>c8V_&>}DTyb1?GHfb&UP*&1qGC^
zq}(H9R(5L6zk15~;9VE{`cpI94o&O)1jjR&)(y8`Q)KG$f8#9wuA(y1?M~vijOC<5
zGkX(Rp#ycUkt$*ZEl)QQ8qye(34yDP#Q2C;C<Bc+f+;9HMG3FbV6vZgbJYHgpkYqd
zfv)v{@c8u#5=S-RkrU}eZ3=nAXtHM^l91TQWqI?17Aw8a@nb<LB`?E<O19d+<)K~3
zyi`$rRbq9&rTtB^RVxa(YY8bFe6Yqp9rKZ&wV8i-qq$dUszKjf5M`!$+UAQm$mnPo
zE+`mZY<NSrmSHmRJW&ggG0)JE(@okip;kA1;HL|d>TzSYa3Y<68aeN&Ja3@u0hY39
z*8aVwYin_Y?PcG`C|XqpyJPa)2~EeAodOOWgwHE)j4YXY81L$KOyBJDbU5OMdu`KL
zSmvhqr9d*6hCp7f@;6e;F%G+9XAP5=uQ@t(ftJs8cbe+Um_WVz79daZzX%XAhy~ri
zt66KBk9OkP#byC0pIyn76WdZk8vgD<XuSN1&d@CsNxOB%c*=a(wAcrTxy6oy!0B|W
z|8G1X;ZQNkaB@2-WRB?|&)Y)Da}3p`6_TDR!e7so=Yq0oLbYs%2H5j&qvYY*l4CI#
z2*WfebO|!@pXJqm8D^Hx_s{$%Jm-O?yL11)ZyM?Uv1y!a{A1*xw*7n0uuT-2yaPLE
z9piWt@!ov(Pvcq{P7)Y_GMTnvZ;uWgU?*eE%EiGy(;e}4#2IBj|J|-v@UbwqP46Fq
zj)G9<XX<?><yXGJ)svn)`L_0^i|=1Du?<?LT}965+dlJ_#(h5Mv_0-{A)a<R<EeU(
zybyB^*7=5OU6An~%8vY0^eI*ev^M|XeJkeM5)0bBS+b?BvFXH!C1lY0l)T6aIc0BU
zF@3*h+M~nkOrQH1t5YzS(1*F}P$83x2-?2I<1*9MGc7Ndv<rbkqixk$dzxFV7l#hG
z>qpZf-eoQ#`BbP|Wsdn_p#r%TrOaOn+!u`Q2gxJb-@Z?EO^OnZYozS*tn^TR(9($l
zzbpK6w82I18~4$==bIs7MnKBhm-ZV%I+6Q6eE(Hd%-9bHT_KJkF1evSeV2_q=0xT$
zW&C}KSEdb$-D#d2dHn@1>ZLT**Wh~mt{Kh1(TaO`!CBVWHNrtwKF@u2Y@Io__64s@
z|M!vfe~h3137)M@{_{}Qu<HwCXV^E_P!ux6Gd)$-q<sBqQ{vike-z2$M=^{$DBg(p
zN&ML%jz)eagBC%`FeJIfr6OKLLAiT0<9J@={GgAK0YjYTLD)T4tnqB8x)<O`V<Z{T
zaI8Zzu^~I~P>*<kj8e5I+^Ry7B$k9>ClUZh6ARw5Mb4yhiYZJ<%GM)1qpp|;{PW>R
zMBAjIG00hVqg3S~gjDi;65{a!Nj$*5DH%P*S(-EkwpDx2&H&d<(ty(aq0?)Dr*<?`
z6>Gk8J?#?cWjMPe(dl09muIgaN4<@>GI63Ie;<ykPR$J3s3UI^Uj&SFi?u$e{uD~}
z+cB`bMJxnvQ|4@G(lozFE(cWS=0dBPa~L7&LBD%yT8WhPK9248OlDV~aRd_fnURxr
zjkIj!9(QD9ay?8V*>cA0ML2{)TjnPv?(+!<Mb4$u7zNl&C70LvjeoV_a}JlEr2yTT
zx32OWP309l+j03RlF|$+DB5TTpkImmBmIn7a(T-s`va@X<}ks?>P(9wMOrhREh<w}
zcy{dXYJHQW!BUq&*qpCXPR&w<0NqQ7#-v{SfL>lG7;`i(8c9-7uQ%{5J=A(#VBNCL
zTPO%*g4-N~`cb<tm_tGta4E%ZJ=LjdN_~LsjdYDhBeU1Aww4?;K#wSfuM6@wV9?hz
zzX{6NuQ>PT`|oeHM+q6s?9+Pm{>{l;A<_Qj@;iBw%l0;e6QP2|;ImH!eo;H}3S?&@
zeRPqUPrlttP<XMbR?h^R@fd+}WKYHWsqR_vaBO@u7a1=ZE|uHUSK;@3j*>P-9ftw%
z%Cb7rlnOwrWVHa<gph_><I9>#70L<Q+>h|yjC9Ujq+byU(HbV4O)9ME)<1LSqSA&U
zs>+j{*6Vr1(=q1`&IvrkjDq5)YDEG_+?H!HBvO<HrZMx%1bA91^_Xvr7%@-nW8|0P
zxYqpGj<Gq8ZLLayvq=adaL4u6tnI5w*bM1Y#;2K~`OdjsFJFkSVGmqec|0h`tSPbn
zpg$KG@;*i$Ih3BFTbstHmH9x<?NC#8+?NiV0Xu0O<;mO+v5ZS<AxGI7DrMM5szcMk
z#hB-mC1tm|BqpVV%RbXkY{4ychU=NCo3&C+Y0>0^^DYWx4Ylq=QcVx0$4zXK2c(8Z
zVVF0??Evp6h%sBHrTI%_3VLbQtx2w8p*mPAIpStQiQ4GtUj(f6=k-jjCpO<q|1gP_
z@eRX@e&g^;UD^#P_m#lP>|L<b(JO}EktiQc1cQ}_-B|<jf{7&&smrm-DCum!{)s_n
zsYFAB>t<KE85?xfpkiRfZP<~QKRfq)+$+htjleX*Kvi3=a(mqcS8k-fUN>z1qMPQT
zf|gS7`+m@qRD>2!KNnC@zdgJz_Mk5KyWNZqE(SrXH@vXD>Q}W$dB#zjil~0ukhoS4
zEtP@V$5ExvR<66hPMUIL*va&VQq$M#LB@}KAN<2Q_#d|mM;`wVbSnRV)-rKhA6-VF
z)B4=>!s&9_4oULJ##_5Ta8-X1sLcLF@Yy}}u*>Yv%iX^S-j54$wr{UQZv6UoKnV9c
zjoEA41(JwMFa8s*%~-RD+>nECXL7lgre<{|KgK;>PJS|?6)W^yWWmc=T`W=c7?U*m
zKSL4l!)~J+@5#TrCqJ|?AlzV+y`V;t1^sjTv@wxC7vKzPXP?Xail*MXQAlE)yMFuQ
zyy7~=82qfEvS;8Hk?w4>8N&h6gENg*<Lqe_V0|Z|<nf0&xWRIy2Qy*Jui3N=e?b(t
zAH&uubXn(T2U0bhNnNdsiI#<GN)A86-cK_1CU#{F<>&U7RVS6lWt@`DM5GLAa*#$K
z^JlV1V#FRTc2|}W=bO3^p3cI@IVg=V;;|$kz1pw<%S{wrm6Et_X2Re9&1HA_zs(Dm
z^-ZR#JU2^9SdbECX58fshACbR`az2+eqb1YZyKxA&TEj8;a}t<PzHrhPZ_Q)K))Ct
zLy7QluAfomLK5`ca$YXzc5=lVf%r7PU<?sZod{Q&<saKpEM$>*CRKE3s$52Wsrjgn
zZ0UW>hk(pq3@;Rm$i@WiwiaGn*!Q0`*cN5R3@)LkpRYoC_~K_>gXdSP-g+l3r&j3c
zxCrZ<*TQ;DYrX}C;RHdU3FJd;1v<!}o%VL_t>4K?wJw0qrjIr;b}1I0no?lEwV_84
z!{f@8Cq?60Ck;}~_eyRy1s5{jJz64zB)npu(S!VuznfApS|_JG*S(FvJ*JeGap^m%
zA_KxxCGB3Rb06}sZdFSdnT$KOxOGi))<N}5?oCO-Ys-i%CiyT=)$tb(E$U{;9G2;E
zy=|?IgtanbaWu@x6&y$2_l6H^Kt4#mWr@nmI2UE|;MK-xfQ0JA%?!c15*BT5!U6D^
zT1~ut&_!`34|ed?>vI#hAzah)T`FkHDs#dkYc`e=IxBRyDX=PKCn1T%1(3{p5-wEC
z$vt|ST9`q2{l%#2qsTZDQscVE9475!c2cQ2;C?BQy&eoOWw&FpknHx)H~-GI_BScv
z!=j1K6u%PnEq8MldbZ0Nb(5TZk9>>fL=M>jc1A_Ziu{puxFHu(R5`{f$yc(6ZA@_F
zQQMi?kLsm`dsBZAblnI%Rd8x>VD;{P0L|dVXTLRE<YRivRLPpPJ^vzLyKVU?W2IBz
z<_pl$btNFhlPhS%+e#(I+2N#0rM;SSW0e*(b!~Y^;qh0`S9emaNN<*zEK7@P+++ll
zv--eGB~M<=uTSw;L^!l>XCFNF$}>XD5J_!2XI0ysiyeLEhlw4<Z(;*d%5=2506EoI
z4Hk1BF=~$=;wdpNV&IT_lUctkCC@8Us$ciH^-J;BTc7Ix3cl{e+Wp2IDb)=1zXTRy
zX}^?G4?25UhFBI*VLF8hn|(evyYU9jF1Cez1+>*`)BctW&Utl><F7Fp!e5)r62iV%
zTU__PqiDfj1(J1JK85VONqBcDx1rA@ZyDdOE{CbUP5LFt{H{3Pt;WK^x$Kf<?d+8h
zi>v)oS~mY(QhDfntMd#~r$@WjFSFW1Fe*A#$^+f9nc|6_cl{jpdx1n`(z(?%VQc<S
z=#TQNrB1rtmZ!IeF$1LPI}te4Et8q-M)#Bd)%W|~YofOunEgLG$xnf~;aUHYowEO@
zH)4ylWyQCWC;nSG?tlF1du6;~n_BW00p##d!8q|N?9pdY`MKOZbKNgE@0OKcM^%3=
zrP8Nx#aa|anT(vT;F7Vt)MQE?3`9ngy9T7F!a1)Z^}_)L{Nc-bk^$q+SG1?PI8TN~
zoJ<I0DByt$5)a$TrTNT^2DLF9)^Iz*aTZ7C#F*$$`Db9fsEVH?KhzTpA}FHcU|=lY
zZqRYA8XgcN>c~kXcG6TILg~1W`Ul|kEu!$5yty-jJm71OFqjwZ+Z%p2&RpNo@K5XL
zYyv~Mir~1E08``6OEV2)qNpdpR<_m07$|*_;+z%ZDI4DV;l@(6Q`_&?UON_}&D;iQ
zr2Cm{oGYMsrklZkr??9`9vN3SZ~Ory%0@5nx&z8U`AZk-@->LzoqJ~fiCs?CN0-EV
zUHcKcob|Q@oD4O657(rF89s!k`TR`9f@JvpXnEVK+qJNh%3?_+Q)StBo~*gEIn{!S
zV%E|(?}LpQu$K|jiJ2hb1ni5lb<?kO`4_X`K;D&GexJ$`I*tMS13l+%?fL~zRd*DD
zzpHB#-?-FyB22vf-xuwmcI+wr{a}1{RRaN*sUq5UJ2Kmga#>s>11v2XkvT@$v}096
zX#L?_2BsXE*b<zJueuX1Uk4~|6-0Uvo5hFvY3|R@5@D<L*v-q$Q{gR00$e4<JP3QV
zQzsA8XQ7}G)ZX2B07K1FD2R9V9D=!aOpXl|@`?y*vL1ZfE=lK~bTYHJ#^%W!;gjcI
z$Y$R9_`IB9447KT88hB?>}S1jJ86~Fgehd{jFc^p2XaXdyI-F#(jIGyR!CeWQ}^LY
z(n!<j%zdI?-8|m37{c%HPeg_&jY&9(_Vq}iNrE9Se%4%+Nd`WM0VMuC75->+QTDPK
z^2iN-X!~fZvdQh4aY;HEVp?L%)6}iP)CI}qIxgX8hb|j%@_t3I+hv6VNQ%?`sFW{m
zAbIC9%23uuXcy{feW0m6{S%AN31VSohRRK!kSndC$71OwF9r>mQ2w@FzP?!JwYsN#
z-Ny#8y4=Ch3rc>maypl_@Y}0OEO+O#4<BP->O{AcJVy}<g&p%$gE+340*9GrR*KoR
z2lCQOU%IHhxQ<lUns+xX$M$=>rdc-q(L18C%W;npy@_+U>Zv&H85dmZ6LCGYGjuzY
zkNIz??O}G)r&7j^GSV)W;fK(zW6pX5iMzD0Q*oio+Vk(nmmd9Y{k}GbYgLRN{Tpd>
z<~x0KfcdTRRR?x_>H6sN%9qrvmRwi0CJ?`hUt*WQY&@T$P3rjV+o1%CN+}z0N{riG
z=#fjIycvaZR<)R;p~s#oyA!hZK*9!GU~b|={8PINViMsH!v^OX=>PM>ln>c-c%PCi
zz-|$bd{B7%%d_w%z2pP8s1b`5N6?yE-<yj%J%xg-Pj?uB&-6CFQXXpGl~3;ml*$$j
zs}%6>>qhYJ>Dx#=Klrb1+kb=ur0aOo{&oxC0pZB8emv{tzhVR<*+ZGrmH{#UFChYa
zDuH?JzHG+k*AKW(Tj_OUd_J>*h4N*N_`bRB#=YRvlW+5Z>grqTmf9LC$>G$hWjbsS
zHM9%M;|Gmk5i}_~yJ_sMB!O8ELy>B5c}mlezs}^i>G}RzWF5PAQ)lCOQM%Cs0Scc?
zf^FvbN&IbA4nPosQgzZH>1M<OL(hxK(z%}#yEk!7OdqSF**X<X3i<DB6Cs0_y5y%g
z)x#jk-DSjHI-IYX#qQAfL-hsm38?blqoLylFXuL!c|mDM0%8HWku1+21nAx0*(qyk
zzF#5f!W-@!VGF<MImdF!(hdV3^xX8_2sCO4jj;`k+r_b(z{VLtVqyN;h-M$L{&`L>
zgH@>cPEL#v*f)pH@L^WD-D4qxYU3)mNhzk-AwEo~k>nGj++VN3jJ@W|o{gl`NZ9p)
zv_yt6n5C^wTUV;>_@`&eJWS4#7ypW39aI{uyqgG7oNn;uHdWH(%Tyd+twEe;9!yfV
zo1)IR9IJodHA{Zyf^rCS0gibJjGtyiMHsHBVsG*R_r=jwCgZ0`5;029!SzDuwd_11
zP!LqW^JLBwTnJ{qqyYo(H>MVM@f`AeQ~#I|0A9AE>(bsq`0L#sj&<$cL^|sPA>%qV
zMNoW6yXdCWLH;P*n3=t926%<0nV;cOs%&HLy_u|jfgRq!3lcYOVu@$o-YDg*vrSeg
zrj^_p((?rqr1@V}-TGs$U(BT!?e#+v$2}~*2NB(Ac&6T>g*&fl3D0kSVs+&{ejWSn
z(a%TMR#L(-K6uh|y<F}9Iw(V%^J1qg{d8GvBmRgM-d4(h9rmafyV({u_ho~ml;68d
zbn+k~ZIWh|S@{jSBP8eZL}XxqzbI`O!=S`ZxKWNbf!lq6XlPyFS1W!Z5tU6mBCHD%
zT4S9*yNKoEO=o7YET9#&MG>K&>I-tZNDi^ESnw)@^lD_0Nzqg=W~{>JSZ49ET1F>E
z4+^E~+nFCva7_tWrrpCWWpC{od^OC+vbP#ra9p@N`pBfAIze#n)M==0ZC+2Z?w@9z
z@oXOZhAOGaZ;1&+gaevx&kJ&7po>WSQkVE_i}IAG{iyrJCKWRd)r`n=LwKCk;4%C{
zh9j2)O&7ovQJD(=HxsjbX>K}oAtp)OTfY`a_^KAHoD<3p?S;1lJ|<r{e+qL+)f<6w
z2Z=b2^J{4u$kyAP>Z3z8UiLP*arxkv!-0outSQV8cLl0WjJtt!1CT#gL@#C0%OcUt
zW^51h{gw4rD&|&~o=TrK>hV<);sPt>WlMNy?fgUAkUC^6fy5o=+u{8_D^tI4Ck%hW
zgD|`q^eC4ES@zD+iSAz19tLy!hCzHZuLemBOS-rQgjGB9DL+LZ2!tjE^w)HLh|x8%
z#LT<hu}&|9w5$6Mt(7CE1UuibJgw<?_B%|V4sVp!IXuVwGIsBa!o9kMqGQJ|QOx*)
zQzRbx|6fj8foA+krbkbpD^4WN#3m=+GwUkhf6?>GTRHKEC@aYj>*Hq65SxIvmP>x2
z)nk^m$HMP!7dfnryT1w$p2`RMNSxQ)|BHYo5G*}k=i|$5{y41p{uAxSgyE~wM-4OB
zL*BPgMjt}Cu;z(*F2=Hg^q0odK26IC`c-4@3jLp1aU(+?0B^pwntg4A4ZB<`iqb{+
z=J8FL0drsf{Xj!lRqx?j{L!0vqA|1tn0cpdFvO(fdDy=FwQehV4bY-`s??tcAUaNk
ziCT%a@XY-dd0B3qxXIJJT6b7(YZHyESZk=fi&ydf-(Y@(O4#2xVFPXEUpUN%1St|0
zAi%D<;i;lYrQ?e}6+O!ntrFTbVzV~c#VujBr|%D3cBBHvxgF7CpN@DXM#`h21~{z!
z#(BX+ck-tr42Y26zLN?ka$DYlLirhPKe<7$d}P{%UD8f0@w0If>eX?bobz$w)XB<H
zEP<?}A5_C7WsV9CQ&m+yCuH=2uHMTT>L+wU&p6yMCDbKIML4Kk6sby}5vQ6QcY!})
z(hLBa{^u8XaHog5csU(Dt59on{F3R-e!Nl@0LlDjF!_`6PI(_uUkD{!u1uVQ59y!Q
zMCHehn+YAj2em3wIq{4k4)~$DNrn|4_xz$Ed+l^OQ|PxY9|GRWN3jtRYMe<Md9#uL
z4L>;=!Fw0LbL-#$r_xK-AB<-vkNswcCHYq*RxF=d%qIyw^wSNq`!#1c<lz`xC(x3o
zO$ZahHsDj!K@@$+a{*^SS;?f#UG)=hTBr*j8CFp~qHrEl95g9DQyB06K4i*PD(TQv
zZ%9)%M5=PU+Aw5FXOQ1C=|Qp_4-dS|H3h0#=k=&${<wWU&eRZhrx`eISRu5d?^5lz
zW9g+Xu~egzr<_~6)$H4@@0!4=NZC}L3B<F+ia{g{QpzjmQcd6F6dL*avTSHR5)O}R
ze?Vt3|7Krp<w6nf6mbUvOTdfpP@)zRIgOx=#HZ;F5zW!LtETtT0_A{MUCm{+L5iZ4
ze#fL)j)Leh$}P%jcv7hd7}~RvwQZlArKBXdmMfbfOc!O1tAL68MC(zTc`ZfnD7~ML
zs{FL$KE$$+9kA9P)bLK_IsQl1tx5S3s*5mzBxmhh&&@G9UEt{@wEk=2nl{*2aIbsu
zV@Mo6B0z*UB0$Ma<>C4>;Vk4sN-)pI7uBI(nPk3}8JjHPMNy*S*PDMXU_rnVx$Oqx
zK6EsVr)I;p(T3}2H{6I6B`8c$j;8y2#?YnGq4o`iH%#7}f6+#M1Q!4Dhf}WFV4|`t
z>|okQnkEF@nS&h~%l%m`$)HZKuhtH1?+<@gd%dl)L<L{N-uWmY`qkLE^7G!HJ};`R
zs_J{AmqOD-81LeBD7B}pa&SGTr`cCWJ72U4MjRaEWdFql1;8pG1y?ik+=g6iEv?F@
zlktTd&G=>Oj?9n<U%gX~_OIg}V{r_7Iz{RB&twlQJ#t^4)C4u;pVzYPdfo4^Pdco)
zKM=x~a>?70Sv#uf`NV%~Avf`A`s){EQuh^syZ3;3el%w`ny>iRw7ilNXr-Lze^%x;
zt3-dupFv(gf0Lh7|Ms0rmq*U{UN~AQ3l%R!U26a~(S7Zxv4Y7iH~MlbPPC`lJvI#S
zvyx4nSVK*HVr#}B2!G8SE2mwP_sW|hWG=JRO0dp{C#=EPyr+8(n!ygG%2Bke5c0tV
zADI>M#$KGV9Jb0UU=ER(Z_%ZY*Myx2eqdUdxe;765(ju%)<cgJQZE0$UQ;>#lS(sZ
z=`O7Cui<z4e{YaB@*kx_PTUKK#i=2TG0iKAu`IqnYD1Z96m6;n=G_TJx`ljagD>zE
zQkjho;Q$*>ev8}TB*EMNp%wnq=v}(<+*ke2+?;z8C(#>ivfR##jnWKiY1%kM<ld{d
ztAjrG19WM#)y@Y!g!EIU1DOTZ8*Ni*ic%bjR_xYtRvTBX6p`2p4<-Yq5Ah<$#lh>d
zo2#r0u`JhN=C`vv)P5*2+kKzE*pAJ+dHvLR)UQz@&TJnYRP0-H<pHs=hw>S>XYgbM
zLxRBV2SqO-ByMdV=-z!ee4sKq`Ra${a%Y`_$+{=>q2z}<%a4o|TO!OmNf<}I8R^NB
zEek`edDDwmJZ2qi$3v<|)5}a`;S7^&+;HQ^t+A4!j$aQ;q!weM{a2OxgtCuRS&WtF
z=L#2k(XOllU2k9C@Nh*<G_^vdgZq<Wcx$vvY`^{CY?_%s{!&OD2PH6G@}IA2_GNj&
zQg46yd{Z}f<>M{Im5jbUn48n^K;{;FJb3d~t@#BRfYY^>2Qd8KUVWJQxGid2kHB0y
z*xR}hDH4y85YNh`k6fTp(yzaBS?16gdt4G+u{II5{W9hrl3jXd?$ll;t~RbU0rFr|
z!<-qC5lK$f^5w2&xssif3yH6zOhrSc(q_2HhyXLt@NHwQ08>%%lnZvn(*UsXr;P~E
zV>Pf#VL>Iqfq>4)PDGMZR5(8ygcQuMMc}Dss5&G`aM+HwSEFzd^qWM59u6fDsE&Nf
zgr09;4!W&hrn^YT{$Deo{HGf^NbH}MTD<^YWJb~sn+%{*;SxY_aCE4j_CYVb7$lFR
z_)%1oVup&at?H%4nCr#=nFo&iMW7vUXlau^vS;Pg6FtY!B1I-w;}?K-<}KP}nPR7+
z1rU%B#Mfz$446|t)_7I+)pqRijd&Mbd!I`)`SRl>WcAX|#5-L$;iR`Lc;-If^h1{4
z<l1Y{w_w*|1WmVj!+Qqes^m?D3s81tik0AUPWqv#W&qdV^K}b#|3uFkISfOWzPMeT
z{N_^TY6?`nw&bXVVm2weRCCzp3Ufw&S<4d)OLi%bPT-f~Ju&+MmQTk2&@|Kn9<`9Q
z%^tv~J7QRmzQVoSxz{4O1PX>^dWF1`z>H@ZB75|stn(mg8i9dTN%~n+AM(o2xAJ|~
zqV;g>VfmENRQ{eV5_;56#giaeLrHW4vaVJ_T^I5sY|_wgeZF0`i<KaC0M#7b3CAYZ
zPWhfM^~m#p)&*oI6_oW>l9fnsCqELus3GKnM9kiKA5r_WY{Sh|<>xug=JEhtH5yu@
zGpTFFWu$+fuxb4lV1ncE+uJHkak3l5_-R0&;!L;%`Vk>sW$9cwZsa2_h_{8x45?{y
zMyAXW=Daifd~BJvXX!Y+4ZKT;Yu0{c0Wr{&VwVJQ4lzN;xjo3`cWS<+Zuc?%MF6W}
z8gg=KH>2MuUv0Bs0Dt2p1O##>X0&p9dGF*DQPN>YPqeQOuIKgBl-E98U`=>hlCtcA
zD!k6T3&;=6WQqg-NWx_rf@PB0hNNH<1_VMM(R07_{5!Pr4VZGL()S==Sh0-GxP%na
z!qn!7^SaCGhMq>L>8K)K6Z3km6<QXn*e)F6x>R^c4u1x*0b+mFC3O{+{^|d)TIJG0
zDev6q^;K^;y2w&8Cb%Bz_1u3oLsgvf#?zWrXkze%&T;Px;d*%=p!&wj+l(&HY~gP$
z<t#rY8tK)lycYn=Qkv%#_XZFrS0?L|HXt3tU0|8ovRxZ?`cw3kt2W2-Z8kN#6S?%p
zeVhJ(6{ymlkOqhRjD(}P?kfOD%%;Jr)Q<8q?R<`l+xPx7ekpGF7Jii6DU>g61!em2
z#*uQ0K;yPf2<3@wr*iThGfVQ;pl6&<yAq55{LE}|z3|Ig{ocD(tcUAFF~aP96XP3p
z6SS(1yeUrWqhm?Lzg_6fU>nf?-6N<SBUt!Xc)+6;#qy@#((nytu37q*8kS?59ZDt#
zWjLwqJoEMhuP?r<kGL6p5qWkq-tM1*%kBC|wp5;ZhBF{6pV$oi)-3dl54>#qwCl~$
zE3M@jTVu;Iugvw#9(VXuNl33Lg2$>czAUDj;I$9bt9jYem%<(jW{KT^;xrYs#%LV^
zei@6pg`iKbUiY=4W3sW&yepYIUDdWkC;5av%T0bUpG-I~EUsS(4mg<GTqa)vRov7)
z|1F^zyM>fjZuYJ4fHWs13ZO5HADQ1GB)-^PYZASfr1gX^t-x@Xdm+S5YzNEIdSnI}
z{`~YA&ACPcONWJBA>LB?AoX<BhRie?^79@=bHu$y?{X-5pmtA4>Ec5cohmt5;k)yl
z@EyAakhc$aQwMT~D}mGRTWC*R$+7kKx{69_$mxPf%Hf;SKAOI}ZI^uPOvLC<Sv<I=
z@h8?D?{TK5cfM$TSRBnxebCO;>;c~N4DrN@Ee^mHEjhBvI=RNuZ7Vl1_XKo(%sc1a
zf!1E&c5R@pWUe5c8Ya{c4cu-3<hr2xx8Chv?nwMa1XSowZ&qz3LB=L!`h{%5jr0;F
zuuzn^WC=;dZ}RzaMJBhm;sXkd$7=jxp)g)RqR{XVR&6I_>ehg$ZiTLxF{uZ`XGs^i
zjq>Cnz*Me${&^L!I$F0a5F^W(Mv)SUv7>2<CVBh(EA-_1%9NNp`%W6m)UCqYqPzPB
z>ZrCe@KT5=e+FF~nG}X8O#)4<0~Qld+zID7L;_f*Na99R6`X4t7>lW#B-_TtX_v?=
zM^AJp>o0pQs<t->6w-Cv6rE%6r~xf`V=MLzlmL~*a7Co*5F|Z$g8aY}A#wU}rYQl=
zswH^)-A(Svp1IlL5aTb)n6{fqASrtkwPn#yqq$i+aV~t4*|3{$CZCLDhFG@JY5a*{
z3`>N=dn=VN<0lQ-V1c%-_{lCgZQ_Mck2R2d1KrIsmm?W#(^qAypTB4!X98VjA*>RM
z*}C`P1n@_om>28LH3cEIWsVCSR_*OPlcOqvfilkgpR8}e_hvE$m7nkiOx(`m<n45o
zTOwDY*7rsv^Kv@~a&T8vQ<QHOWq|LXk`9%qbuthttQR7hx{FQADVXrS1#~VEDE7QM
znKd+GZhPf68^nU$67&|4$GgNsZ8!l9Xo@Y#y*s2{&@A1A?3#9#9s$a|c-RG)^;eT|
z750}PLJbIXW4}Q_@Pufe`A>HX1xm5EzX*^Dn5W}njy)`Ulf-2ui%9%Jm|D;%12`fz
zXq^@!k1s^qJ`h{UAk4k(5@C5spc#Czy7k25dLXSYMX-6~BjpGqh*iNHck?ZK1K$nC
z#bj+?@1Wy${@x27`cGkMq&f4B!lY<{v;hJ>>|3+%GZXN#OGzqYxLNnh=lB9_@(I38
zR#Xx$@hUHy^`c0gvXqVuqpAtLJ6_kdpZpc3S?x}@63ht58Q1LqS^;0zwJhT?T8JvZ
zxn>CIXL2n0B&;#FGb6%y__NAP8mGepQp8jRH}ZDdcw1U1EdS)_y~{C#m*fx7FND$+
zeBDlMkQmpW;$qkM;^(rbD9*luv5vCF>?^-Qslyun^!(6xty%UNEH)<w;bKwsrDjgC
zPjArKd3n!FCvQk<(=xeQV!rGcd?!b#>XDw{y}@jz5u{I_&dh*-7Pr%}LA?*#Ek(h^
z&(6htv{>lTC~{*kP(dtIs}M@%+-9awYu4UuwF}fg=wxO$Hm(_vJ+Fs*y|`Y<US@UP
z#5J4ym>t?#zfO2?_4v!%uHk{k8===}jwloA7?I&*cB$f=C+Orn?v1x7d4ZCFw=O74
z%^-Y3FjeNh(0te<q~Kv$uwN*D-61&fHY==hF;dauA>G<ds$sXR`^G!Gr3~h^g*{U2
z_7AW$PYPKmYSJcdWg*r1d$1CU1+fm9Mr{M$?L=|F(^Ubn3H-I=xSoHjTo@CAYZA(W
zC`~l*8k?s)Y7@Qq*2T;JXkB>3HEf$#9x{_^3HB9}5nJTa-?Z%i33JD@$YN7lR$oqz
zeZKrYrR)yU-N_;3u3HaW>7V_-4le$ijh0aOc6+ji?cnE_N3HVgT{EZ4R8Lr@oO%ld
z)k?GQ>e*$ydUkCFF%P)BQ+cS{CUWb}dxW_U;!G8|?eyl%{I=t+Wa6VYD#FmlRzO6$
z@3?JD3Mm+*KmA=$(-9gEuXsp=y~S-=!9wFfpbnClDIBBN#1QatvwE1y7!&pU)V(Z8
zB<Qp{k@T-XQnJK}K+>rOC3HMY1qgT1#6zwqBeO^{>@8lA2xHlCQzix_YQ?}J;&C$B
zG7`R@1^8in-i_)nQ#x(lynH5tbU!U9EdZ%bDzs=8H$DL+s4dzAwj{>C+QTEiRVsIi
z+Ev={ToNM}5x)?NN<%IeV@JV;b!02yFmFMuPXE0hMULmygnj}vU>%OF6APbMa@o3r
zuD=MHxIWzai{J%v7bD3MDzxIije0-SAjgdkA-9WV<LAsjdRs?EY?X~UYI|Q11>*BX
zt(f|D8_O4hojetCm7+7fR>tHGEv$H4`DWru8z$eEnzSw6(C!&(@*fuVX5hwHBkubA
zWG?<1Dmdoxv8NJIlid`=XV7p9OX_N0{k%FlKXgk@Q(sH=%#eJvGnKO%a8q8%qYNgr
z&p8zgRI~&|=Si9B(hh1%(LrV;^P+r<MWMPJ`-6gh0mc|?I{2x})S;w8l5^hLM#vq!
zEKLzPi%s~dr#ir{6vK~Ir57V)&)>;MJosGTD-bYx1I`bea@Kj!eO@8n|2l>~F{yUE
zxZPRet@t;X1yB2Uu-P+|`(2!7v`_w=S$-F{1{aZ(B8xkEMc*F+nR@%~om{#0<q)p;
z$oW!~Qm@fjA>uSligZnob;oqYSVB)rfL;{p!LZ(5g}urOYocb}EGkJ?(<E@mv$OPG
zUe@RYmb>@jZEkipJ@Ltdo08iMSjG`x<!>{Jow84~TEn%6A$fRz<65=4uX-ICPTTc3
z{tOZ*Cj6r^$!=>d$zoITY7n&61knB<VWAM&d-cnrIJ=B``+C+nQ!5%cPH)9xpR=mQ
z)&AyX<`ZC_9+qlM-f`i9b>R^`v}8`=O%G{(ZPNaDfZ7tYVAd*1y{d`?v6}{IeD(3w
zA}%81-8)FLeqlCn?H=`~Qsj2~yi=#Y$1+kB6f?77c5Lk2Rc&<X@qPc{n%J>bl#}*M
zJ3No8f~=DTPZqF}+AWM<NJ3aZxr}K6n;myQWws=p7hyx<fBw+-`p^Z$P)uysv1kuI
z`-10mXa_KQu+Zx5Y^g}o!g9O>=q@C4YFx|2&z4G6^#kjIeS*|%?r+%fS7{;|EOl9_
zw#Ru#%$lJ~b<;C#C;peF3g782^OJ4iY19GA!mEnCir-7x_I_&ze)`Cj!v@3<FNcpm
zD^1Qe-^JL^rv7p^<2ZOK_#AMsg96>h0Ak}(GA-Tzr!%QBo;dPA*Eryfl2GP9!j#Ux
zg(>Ap%Zk;=MzMb_@Bg>E@V|b&6{40c)eNQ4soid$rHx@~SQh<^fH!X*{mHs7xvh8&
z#y9d`;v4w`ZCLV<q*bIR3#-SSm5a(n*QWynW^Iah1ar%74K9P_mSxL~r!=jjfA{7v
z3DO9x!woe0f+`5diBqndTwC}F_m2$^XTn)ao0+wOi^y032(xD{Mb*?pq5&wKPphV;
zY!p3m*hju;#ss#2IMRX``1%&DiIJhsI6JR@5S{UvK`};EDL@}@I8Ym494;f_l#pR<
zc6QcmlMF966i2I<R!S1YDOL5Q7)cUTkfqNgzcYO^V0s_e631V$RL$Q;OQvXUMHx1M
z|LZRKl#_5}Q8{!Je^Dt;M=FaqUI%jK&;yUvPte7;(Hdw2GD^?73wC7S))yY0shRwI
z*D@3KQw3xy?tOM$#gxT7O?=G65flSI6Ifp$&hSHiOTDLq&^~*yxph%4FC5@w(q<vc
zL%g;tyxevj*%)z}J|Hb5`@8ky9k}Tx<HWAiinG-<XTwijl^7c&xh1L9)<$DlQDJSs
zj;DH)Wl;l1+vP?bcQ7FwwTKgxnPxh(^*4_nnsc11zR!|B&F{a*St{yne3G&_z?Gr-
z&|{;CjnmikmfTN6S`3)c?Rv%hv|R|BXlkqUk-sN~gJpW+V53DS-Cj2waIjILAZ@l{
zyfQUdAHt{`?r`3=DfQAro@U}Gv&fODO7wA{pS>C_VAo6J-n*1&t7Ped^)o||&{$0&
zd^`|**2*)@T)7`;zuJoBN6oy`&rd2~NFN6o9GAyC3fjU)YHTxF`O1zhJ=6q$pZ7Ds
z@%j_|@?&D@rL7`#HTTcSNaP{pQKDY(p6USJy#4vc8DFG&5eP69<}af9GJtpfbV0a^
z%2mzm@PJ1fd+OU-Vz47Pf6AeWl$cBD*1QN@-<F+7MG8s>DyRi%Gw?jGc@x#ihwRLe
zv~|eDkpYPwJt`#YvL@OBP{s2dOPJ_0HBTzR>P+0?7^?9CuL?Lx+iIqd8gOLaIKF;W
zwOeY^!UNGJ0JDq^fQYfq_fEPChe_v~_0&6vo*0~R*DkSE5Hro1bpd`Y+!Oc6U5-LD
zDN>nr*UKJ1Xhr&v5SVr`q=gDL01u{Kc4{Z%--)0#x__{qEKj<V-@IhzNp7zZKd+ZO
z*Y+M)^Cw77Z8lfte)PF*tWh&0E>hPE&Kp|onw4Sjqh9{QiIG0dhd|pE?&Q?UI`?hq
z(Z01zigb!HKA?B>z0$Cg8j3rcNTesaL>em!0i#H|i&(6xo{mlu_OzGVwA$lM!XF(B
zd94ak14w8f&qBfc;pV?6w=QfH_bY$}JjcXqJ0z>`X*4ZTzLa<|9ce2_iPSx-CvBD-
z9A84rN_c%+NH+rH=9CsTeK8%(X)<TV{@XJX{=w;QgTR5vt++S(P99QZ63%##9_=5^
zSbVZ|1bpZJkND))T48(8Z=sTgB-G5?>|G0$7-r+1EH)7XT<w0e9XcvtRB|)X3{Nb2
zkq%)Hr$jqV&_LsTHB8!vi!^V?_JzV|&SRr_QM~v}uUYRp6TwV!+rDX4H7UPifFTtb
zf{|!w+>phTuzSlSLt)<9inUH#X|H?tq)h4!_Eo8ma?@bUWV0v;Ox5tMOBPG@qLei=
z!8w|7q$HPX${)JUl4Qy##4=|9H*kKEuJ=-pf!qqOU_O_O9B_DhLv82LpH7u;+#L-O
z4f_EFm+Yc+hz^i6VA&;(U&aI?cPL3Zb|ZA0^B!#?7m%y?Zi%^m1bedDSAkZ#`n2D0
z<lDkQk4QyVxJKHY2%B8GVxl#y+NBXzjBP}+B$v~LWu{7LFT887|HRZ(z!2lMnnJD7
z%_nsM2!5hRJ32JoJBwisBLCjMHQgjc<1dsjo09RtY!p4$RLuxq%Ir>R8$cLlLS(D8
zdf(ZA^V9~U?i6#N*cTM{LKbAp4r<jnJ9f7O-_KpUCtr%<_*Seq%d_-#sYlv8K&lN5
z1Yv<!iVF&_2q<%AU;IcgebpIE_fe!zMUzWP*l&2HT~Hu|&(tbS5Si~y$4$>M8ab*Y
z@KGphGV>+VL?Bak{EXCH+w|s%+1@kLvOm5KB^Yzdz+X)KHfQ(f1_f2s>Ovq|mmPQy
z79CPr{Tq*H01~G}g=ssQqsyv}Z0r<>e=X2+Y}%?}Si19C#y{ijo4IadFc-a=?`V~e
zPwQP-222Wx$d35V+X5@FbO*Zqv?S=oiC*(gKBlrAkn#P(RmV|EseRwPl~(~Z<6Jux
zYuUGf{_s2$CZE&w(cB`BDvw$4y)19>k()@jSrt7m6pw1{>$1T=gj;`r5Hhb;mRXyq
zG7jb8@xO>BZ8bGQ9W7`HZykX#EJdr9ONOD6t++;uiqo0A<cc&QU>_1le{U5|K3tLI
z&{mYo?$ET+%$*LR{&Zf-O=Qlg6FwL+ZyWU|Te<08qBanah-SW{6#??hKB0fz96Ll>
z67pbC7veD<e=4L!mt^&tYqflYB@jl~JyNev^RudSvaTHlvW`F<CE;Q^-tsav4C?ZH
zocm01@3V~h)1E4G$}?Kkur&dU2M-84->IYKFmLDywNL}@IqmlfXcmSd?q=qvZ8bTE
zH{f~jPd?^vyy>^vy|f&>LeQ|DaT~(Mo?!e73^FZ3dWm`KUJWL}{vx2xjWz>^Xvczm
z+E)$X=>x%8pO0<!@f7Fi=-2>eZ=bB!m6{2yDY5xB^pH&3TO}eWLxVtES*416Qvm_i
zullf&#S06hSLGiB2NgntmK13bbFFrS4kEe%;Kr{4LmB8tmlEV6jQF<m@bLAP`-K!1
zSoP7&xv3h9YkVm3eL3bS%Rwv!nh!YhwZ3E92rD7?0C^7R`pWL$G0sc$^`C1R(Z8Nv
z2o80v=r?%q_G&<3<c#GL2`c2aGDrTE+usR73l3H*O_P6G<o^#JB&%X>pTE}?cDn#G
zTjlp&@UKRINIY%&+b+wJjc8{x^Bh#(!KTPQ0JTIYpEFeEOmWl(7`JyTedR8Ok7eEC
z7a)f!m<k{Tb^d%W(!BbNF{G$bl@;JaB9e@C1<xy<0&1lxbt=-)6eQ~@f@+tIsxvG!
zkjz5aa#JQ|Tt4}sfopn(4wi>8$uVgzih!c5C}MP;lO<hsS}32z93&&zt*V8WuS)W&
zmU$hND`dsdTTH2`f@Fd0jART>1FSGqWN%vB&-j<|VS#xI*L`P2#A{JCsS^>xp)&Z-
zh-}4(ork-SDnSx9*}Z)W`i+MLcU1Q8pOsH4a+0mG@9pD9Mjx;3+t&>~6mZE8SNJMg
zx^r=UX+(DtjG_$4*hXK;-ijdoW+7M30j7Fh<dI^r68)^&2c)#W^zmERUftLaKeJ0E
z#yi9TMr!XukWZn1&N6tdqq8ha8-z{8ykBE=+tCKMRDlbOt$-BD65;xtf>niC$Q~M8
zl^!*Rqv>q!$bTE@l@XkyCH;oCw6%lPogo74Ujsv4Ggd?UZR+kA1d}ZTg=GZsN|R6l
zt#=k$G$=n3)UHY=0THf(-j{`TqOQQRmHDoG5N@5BD;;T;6k?fPPg=)1Qm($C7&>i2
ztlAx}8JxwVJNYO`OTvWPu_m(oHM4t3nbYKGVT|VJg6o&IGI#D7tnQ$tdeUq>Y0+yv
zzo>05@bz%Ini6#D#I#_`dc_2l!~|c^P$QE)2kdYlksq2}JdWh<NVuK@x$<e`tzJ7=
zWyQ!mbC%kv)gq{+%JreTYx40=6S7QS18%0a)}%9`fcpQad(UvV+IMew7`=v!-b*lA
z^xk_fqt}Q&qeKZ2i58-aUc+dEQAP{VB@z*x5yl{d6ukt|QpNqQ`~Tn1yZ61{cYk<}
z=hO3zV-{=Hb*y!*^ZcE^GG~jhFFi&9x5_U<gv^7YJyD#2`FCK9(51UN>5@tcTF(_|
z0HZ+nnAH3D?S9S5{;vY*lWgQ)pkw>gT@hX&I}yrJ3sY{CpeUY~MlEnk#JlS&T&|M}
zEH?W~IA!QGGvvZ15xX9r<}(uX$@GGM*4+WOSSu1x*HQI%-wp6MQ%QieJm56;&Gd|~
zi(%RaunU!tx8G9-Q_qf_85BOL2Rt(R#Tp1CxrN#o20so#)VSMj_$M_Ogl+qz3OFXS
z3^+E6n$K96ePFQ47UiyC;<p9gu*o&f^5;9^V<PFbHtICs5VLCAoENbj#BRue#i{C}
zLwPow=YUam6QE=%8OHm`f2n|au6inHbarhy#0~g$Xvgl+6sf-vXtC<|>IjIeW%2ak
zGQ-wR4A#|z>y=Mt1UjYU+t2zA(5MB0r$(kwEzcPfY+f^e`_|Yb>8Rjmptis$_o)XI
z?ogc+>aKibz6(e@&w_bvTz};C?*)dKWo~YxOA_1z!UfcSsM#;2YzZCOo;Tho-~Yt_
z2Rdk5A|0-UCZBUEMgO-sKqRIqEtC=P6S}VcC=b0&476O20hY~F9MEwk-S^R8qE|z#
zvBhhaTMK(_Jf(8-|Ehhjb`-|>GaMTB&Xb57l`>et+PS!0u>sRr#QyHI8o8&u({N5)
zcrL%?`b~Lpey5kP)7jo#<ZH}kCjc7*&EL5@plk8|+k5C?Ps~RJnZC=tlI1)Xe^kpR
zkAL@kt8e=6JqCfZmOC8pT$T^oT$Wt~O}@1&j4!cym6b)~xjJPe@4CSz={w!vOHw0$
ze)5ZIsw^yiIYfW$P78%l<V|OJN3h;Z{1D&$o#PWua>{-YMg7C9ty%K7Zs7!&V4vAq
z3G}}V+Hpa*>rLz3SBCMUqn{?9k?zw)TLGjGRkFuTtMAiO<)(nh0)D;6$c+Olm%XK)
zSij{Kq0-5*l6K1X6(#3&`Xx%UU0!#C3`!$*5_-oryFaOiB9R+rue$n71Jae(ORm1$
zVWat|t>u)jkkBj$Gg)y1WXJc~AGh^htd|I7HtPCIZp|OTRx{F^`B`|J!G*cY@Sc@)
znPn$jn8PtejYRh3)7>l17X<m;&iCm7sbOzshW+ja?1$qMFz;~TI{kQgr=VC|2)_(d
z0!Ft4q;AN0#pHfqupoZ>y4&7&DZOa5Tqd3pG^b;`e{|>WQ|8jU`x66Bo}N^{{)i;%
zF0=++&5@Nx2&!G3@O!>^Fe_<q&HG6E6Y{@6dPMiklmPWR)!rAED!?jY$@H_L#O640
zr$AW)s}fd0&BcP(`ETL5;|JM^Ch6W_&Uxxsp#&L)z;7c4o7c3^c?EB_<(WvU=pn10
zU_?WhTLW6mH9mt)qVKq?QDjDWRBl&=dU;Ucd>EAuH(4~2V<qQ^$^yQuJkbauxu{VO
z)f$TqH&Hq^^KMuTb(U^dHBhP^Zr|LHi}?}`GpddJj#H7H5n&gX-r!zv%`QdxIBI{0
zN6%8QdZLC&(NuqKF)?FRwSD?!6O9vQsDC%_;BITZ=$(1Ec)174ybsv<7q$GvQ<sd7
zm3=Jdqx-El9L}ei$&1p8{SE#C&!iH{7uZpc3BVZMkq8^BZ?QG!mpqC_A;0Fsg{PzK
zWGY69pVOs~^_NTPEK<4TI~5n%ur*nALy(?4sE|575Ki3zLL;jU0j0##+UDGT%M~y`
z6;W2R#$81E0p_kJ66^K6RwO3-F(>;0R>MWU_u!!piTLlOE=!haOg>KUW=y0XmwAeS
zM}@4lk#&~QFM|1RtZd2C=#5~KC|Gm&p_w+fc78n;`>^av$91))UOI)_c}IOuAP$-s
zMDojaGSsHunZJg4!`E5$U2-slkh)LrEVxp>dBcx`-)-=`?)tF}3_fM(`Ki3?nuv$K
z{F6IOvX83{8B83Wh^#)WD^eH<UpbW3@kYQ_7n&x{WSmX)B>i2t`)&UM-J4o1<O+5K
zgM6PfIjEWF<f4+@4Vb%XFoa57$ENnVGNE~bjk2Fob&xQWy@f-9m^i^`6?y9Nyf1V7
z^?UnkUq!$=bF%p>HGQlF3)MH;uVFup$pbHtJp6hy;pxqfa_Qew7HhEis~sp;J5=tp
zhUF;w*a{HMe6Ao==Ykv7<Eh&3Ok*$5zJLBj@>C0r{CRcZPgigaVx^H_bblSCye~m9
zIUnj=cmLEGNiOF)wmr)7p>a!}E?Fv-R1BE1Gb!l3HREk>RK|>q!YdqwN<_C(lSK>P
z{RJ{7OVk;qgN{qVLF-D*nW1LghTz9oKYcF2l1YPih*!by*jm|IENkk&J`o5*5Q09m
z`|joQVGFiv3rwKP7$J3I$WbSo6o1pS3SLx@H%AKEOe?EhlL)J^B1Ny-nGXAb{r8;b
z@-pUqnX0$UF@7fO8Yc5Zf__Y2p!#o6JIK%Sk}k|1GjB(0mr<vpuRh4GRJWM$GmSV@
zkmhiU+He@qu3<BATJec?0nZuN3a8q?k~-_U58LG#CI|y49RNF41(s^Hnz^<>vcAyV
zQIs6b0B6`?Id#FxU&P+JR6N#AdsWgJ?DW3Yo=WT)2PaIxe?tEH;Dg1wz|`d;b!q2*
zAO`ww{p<o;N4J(qcC9ALEP=9~Dwn<*y{oPN&pK6ZrMD1}1K#oUYwN=tT7_p`0yND>
zaqFm1U?KAw42uXwA5YF$esBDo7d$_-w|#jcS@U^seE9()lP)e<>xHdKsGGpBGkx;<
z?P4%4JbOK7^y9c~=|$fYFfQP;Ibz4+MsVhyQ@O^kF+e8atUfaG7wA#f?M9>1_Xa1$
z+`fV-!a0<efR1HU5mj5%;S*3lC8sb!6X)w_Oq8YGf96JVVt8}7mwh+(Lhizcb>d^<
zY&4#Ibv(rde>hXm+0)<kvx3FRqe8rq=Qv`6DSw^*=k*twR~vtUc5R1O{Nvu_)ePTx
z!T8EIpEugGL^JjLH+8^sH+u62@Zf|jr|l;Xf`oI_MfL>`uUB_Dp2yrX!|znWUc8_v
zKHXYoq`CWkl`8AQz4@mP{nT3-Htf84W9Eytb04s=LyLFs^?#_v-jd_?$gD{=Q%mV~
zX=}-}#>XhWVWFluN$ywGnEk!^zCs3`7AWzu>ET}>CR~%Na*uD+wA0VXB;k6CYOH^M
z;nugBb&=p;ZpX|Q!v?rQ))^^P81*wp_R;yz-^y_J@54Fq?I=@ul8&7kZbEy2{nAVR
zucs@<k<%&O`7_UHrrv9GlFlA-UH%11jjmt8<gU=mqha@$G`}Q%Xc}vQhdS+dAK!06
zkgYS*bDsEk-2oN}MZ9<IN50LA&ad4mcY;C%p@Jl@ibhNyFp3fso7S^CoL$`4_M&SC
zUKO=@r(y=*enxq|>{e1P%=0tcT1dUqB2$v{|5_3f1rL6H|Fso4=}zh+G1GsGvg3{-
zar4hHo@n&ysj3il&<|9WX)+_%fS7Lik=e5&R|1*Chma^7Sm<Le&d!ftH*thmAiQ@T
zsGONXrBQHKZa8InXO_NlI0{s#eJSc2*eXIan@1;WUpRo65F$^2(Gft0n=od<*(CM*
zob&Bj=cA9oJefDC07sO}wS0WROI0f{?W{dm{UA(AtvzuDnyRc^%Qj*v82Y<Y@1s0e
zzi1DY@N-9Q5*Wo(?E15ES9Ebm8i+7BJ`x;n=*l503ty5eWthIEYI9#+;5vXy#^3IR
zB`ennj=bgXk9vCqjEb&1(v9do0Mx*FF{0_+DeC$7{zK`=y{QAI;{36RrcfHF+=L)?
zE((eUL6^r&Ktyky4TKW1_My)VxG2WtTiCP82)QlLQ)8K=^dB}>e&Ch|QN`{EQ|zv%
z*D+KQlaS%Z-FeROOvORvd0zdy&pBcRcS5bkccz8?o42eRFiSSVT!s^ZLcF18RBurp
zJlpn{dy65vdIEw@55`B@q@PYJH836-1}B6rR%7P#wkY(MSKGfWmIVp3?pY@5KC(p2
zgzY?$MyNFvbaDK$&61Ku@zMD!{Gx+(JY#3w^g>dX5md6DjMp-SI?x`P0|RVC4Uf;X
z#fxvGv7-8K*vFh?zUd{+Cve(zGw6P=DA)>}khlLZ<dXjUUR&Mu{EDH9$<f)>07i$c
z1Ix!IqfY$<`+}Iq{V3yBYrWE!k1FW)&*(ze#WPTON)rNHP_+CfR|E3srZ8#YMeKEn
zR761RML@XHkt3!_p7+V|JYY}5-Nj{AvDNJRq}B<`IktI;=wo7BoqT{)hui1!<(ncw
zk%)jh7wH#hKkXHyKEUhNW$O>WTDWxxq~n39bj?}qoN45$0KB$0{aInMzA8$@3AAUH
zdml$xLdZ&r!ZL^+g}YcmO$08mVG6cKN+L|u@5M}CT+|f5@&m?Bl9^?5)0Q2Or%aaC
zNgR|>hp^=?r7{k0BcqF^?#=bSx(A(8bL?#`IFIFiVSDZ_6+evynr9(zAfG2NLPpSV
z?R?^4+v|#xa2Cdw>+~j(_o#ew6(PIt61A9_dbdjj{N#u%w4XncwU(w*`eYZb>QbfF
z_<i4)`e?)jjMQbFvU<g2lx)-g*xPjjHRZ3MyxvpAjY!=ZHZ97BP?d`<sKM6yw^Gn8
zN@St#zM>B0&$(ylnCoM=GpI*wmLxBm;-Hoit|tz@g<M%a$wzQvsw3nxrP6EY1M}1U
zT$+#!zk2z57FIoC$F{yRt+J=twHlZlYMdk#^cjy#1PrOP{%deaf;7E}N$(TdB*6J^
zsOU8k{>xUNSVyezUv0-c&~`L2Y;`t{u*#G%DMJ|Ddza+J2CU_kTTR{S6vi10r03(M
zjk$ZU`aq`@&@%Bmc>M$1PAxa_HS2DSh*(Mb?N&sz&*~HVdqI$AdEUIl92)41*n&Tn
z!z$hF?1nK>_|X(mBoayA3L(-8S9NR;TK^M^p9{K8aMVkrVwdx*Sxz0l+^=1Hj{SgO
z*uTEk-k+v}X(Y}ew}`j)*!Fq;mNj7M+cOu}TSrLiuAyP1V>Ba3zA_Z+?&kD<8K>|=
zB7s3@R$=EUulJRlBWEeCs?7_0Fzraj2iwOc%qtv;OMnWjxPw_ZFU+3_*Lw+H7eQw+
z5sC#pNQLCYLA#1x!e@;^PZ5E4()uO{AQsK%w?4#{MI0jSd1P5kPsp&<&FS@PtVNbD
zYnfhOvNFKjepEWWl;UWbhxX)8rxq`5-I+Syvv@sy=m}Yp7hKfrTHIiHo__y13zk*Q
zi;KQi=Dowq4%E7%@VoCOQ_y$yA;ANNJdcT=3*j&IoBZcq*s~;yY$|(;v|@gJPJm3|
z!fgmQs|Wg?`}L-9yTr5jDzkz>B0REPX|0#mE$(Z&5$Y$hgW;U1$fUTBynZLxn-ASu
z_Z6NwuYL?hYNt>$3LRsuC~JY0qaJ{r{`&CY!*<Zgd&zfC{5cFiDPIN{zkshZ_x#rU
zmO$CzYcEKfZT>p^*L+*J-0eJL!Loe9+~4_>{~?VfO_xo+Gt_-3I7kM7>uxub2uSxr
zsZyqrXKF+5)8j@_b}?c$<h>VxR2e;LRu0<Bf;Nu#(0WnZ1zAwgf;-$zlU<AI1&raX
z38;{aTy9_AG3=V?kV2?^s<6!gEaBKp=ei|iSVRzZGy>-4!DlW*XfnEKb=AkyDAoOS
zWH(AMeY#f*k)bz_2)}KJhC`Pi(C-)_C{C2VnB7zSl>k>G*@_Y;-lqyw`xXH*n;mfs
zU5ppq5!XJjyBN}s=dR0Vm_8IZso8_UbP{sSrBOsxLyTnRLaXc+>qb#*{QdqZM@xGS
z*B$dbn{Lsc8*3Dry-ZQZMf4lrte5c3pk~0{nrRvPj!5HrP7e9J7*B?_(Ym1)2F(aI
zmgmKhG6mbvq8f>uh};X0B;m95l4F?ng{<w<58qU*m$713#Y)Tj9_WRs>#xng-?AUB
z(hJ&T*44&LZ&eY$XZBRuuF6}UJrwJ;f0$>zeI4<QqWRQfE<QR@*GYx8Ty`<r_vigw
zwGS5wK?=D=#I4T^+B0Z!7t57K3>+xCSK4NGl=f`yY`57&_StJGt63QetJ!S!ztXPb
z6Xv{&_I^O9J4XMax|V->VmaY1=0MJRmiyp2Jqd%QyD0bgT%b}nDdiarY8PqpMA5po
zWlVGG$LuRbaK4j&4Hk1(Ked|jqBi0!+f#+G?R&DpUwlWjN5msQCZ?Q3_!gH|t#5S#
zIvR!Uhp`^;DpSuEOF)duW*7#LSWQk(Zpd5CX#8vc@3*+{#e|>naCiM}<ZrX;z6whm
z;nzX}4inAq0q1t@W39OXeE}_i$OH(DOzwPdGVB6Oc37>o8Ff61%4cG&s!Lq^N%`2S
zWNvU&DbUO}wt8dtA-1avdLq?Zjw03Tp<t^u$pa^r1#?n;|NEOw-F~-pZHBhATJKjh
zbEW4@R3Px1e*&0;(;Rq_?QG)>KKWv{4d?MsqUw<}D=fkLPu0Br2Ka`X^5|$q0z<XY
zEsvns*J#Rw^%~zgem2qOPMV14;mBEopZXw@6D#v9Xj*kmcWj~4e8j!a0!_KwUbWR}
zl-cll^!D?&QUg>uqyTl`MAQ^zUa4=P)@C%9P4W0S)pmqxcITQj7pdC~?37(W_larN
z&~xT70rs^}Bke{#q>7(>xFs+h$GmNkaa!R06a!rlq+W@S!n~%dMy~}ztp0NZ55(JE
z2jhWg+l@9WAcyL@M2d~^b}i?61k>b09Uv$X{_3F7j<Uk#|6^wNKN?gszW=k~+>MtS
zWXkxig1r>V2&=3$-jN!UCmTQOy%nf5YT&;bj_;D`MjOk8FXn2eXx_DcX2O~z-L{Fp
zHVErYJ_CXuojoL-%cx-<lIFt4w4_vu5^6?Qy^S~W-}iZ)aV~6)navt~MpCb1DP&Pm
zXp%LY5;qj0U_kurXyU%sJ~TdOByHo@l6T5?#7hEb9UQ6!&?(X&-K0htO5nxv_>%s+
z4WJQHK03GzoW?cOOwC5HC4bwS;5<81!O;u2Ow5F9X%vzQi3rQ})hV#aeS!7zg+9{g
z!V5j@0+dd@>G@Xfm483o`o*HMxQJ-S<+`U{K<QT{c+q6N!vtp>3rB^_K-<Xywlyhn
z87Y`u>n)QAD{;=K3i+o6W*f+3OOuNbCigbKVD5YA@k1fewi%P-<hnX}Cf2@z{+ym=
z1|h2UYOTX<JU7=44s<#5dl#T9Y6BFZgkUEl-V?pjO4`NRN!M0BAHZoxSzylsjPD3u
z<lb0N9_NGwDxcSV`eNcWPZa~7HVu<@7A;DAMCTSfdS)WKx)@qedb5x^e<d?xZGUyO
zW>V6&i!--SjgVZ}I_U!X1J#XRsJZvBwif-Rgq}&xxNW;nTBfI$G16Xeb*$dRe*DnY
zz=K8IpF$f#bX#Bo=ls|RLd(DYh*UqM#C2ZpS$3G{gYd;J1Jf{sEUE+c;$z&`I$`mD
zKfP^mbD((p7}j+7HaexRpa&8nZ9?C|uYc(e8NO-K9vBz>%2{6^_Fv_jR@@liE;HaT
zW&%VIvrvgJf7T~)6ge~lTopSZ)?qM$$lnt&;3l+nQ2s4fe6?3+YrPYezoT(qxn97J
zDQRYxrFw%Yh&97rd{^(&cCA^9wtqPt!cIUg#Z%kTVLMkwuj)+q;hMj!m^p>IXg<~e
zJx6DcQ>`VXp7kxM2-h0_xQI0~m!|7cQf`3=1c9V0{pFo1??rp)>~!(EF0~|QVrzx~
zl(*@W{-(URg3o&abW_XUIIj+X^GdWlr>kz&rgq4BfREJcPyAEydcdgF;g~VT{4UIp
z)G$O1=G&h;E$l3sG)zv`3Iv6Y4aN8i?pJH_{<7&n=8=+gv5pAIsdh1u{dbBDBY^vC
zx8Bg-g;3q`D0+?KV-nf9hT4kd?YXou)7<;teOPruMxPG;81Ye`X^f+co8Ca)Qp_%1
zG+W@56aId8zx#rbonG#jKkJd!dU?i~R&uVHT)bJdt#`PYPIh!|G>nLrZH&ptE16X8
z%t#{a!R<r`+Y{R~f1OS4&d)e&k}@iy5s-8%zrh<nc@FoX*gfT6Fg6p@Wbu&EYQr0+
z=28-aispc%z~8}%b#OZ&8>a~3#0%y(;KUur4{&yaD4k{2J!_VOT%#&*NVpLY-uxjq
zbtS5fzn8a1qszf1?i_ppL4i@D@OK;dsv^Em_$De6=p7TQu`U|_T<t&yqsq|OLTgFB
z48lEsjWMAZ(aGKL>BtPKeYrnlmKhbj4F)%gd)He=!!TJ3-r3|UPvk(6+X-59T+nEC
zExC#)CPga|D4DwGz_829+is4^4ff_=?&z4vU-xk<ErnzhtL2FG52$Z#+nEb$D7X!B
z!|_bQS%s)F^(WAZ9`>fbWLgcQuMcd0i_M3bxvpeTX&=JGA0(4QZ0dkK-&(=KVX?Pz
zjm#Zou-LCcE3>&S{-dI5r(vSC07oY%sm51C0AD5*pmf?(6^*nr<S)$%h;B=|CF+$J
zNeHn)<<QvbP);Be&gnW8i5sg;fY=SCri?X{Mvj+aEY{j*GRL0N{tO7cK-A|ZO}L`u
zcF@{tiD8#umNv-V^BT9l2~;L+cA+7~ge=_{Um`V$s_m1JefMw_6YtT5=#h5GkEk6L
z#ZjHbHD5v-q~07@6DE#as+$Q}w*k7IYl!F#;2CCV5mAE)LbVUIQm86E{t5fz59QdX
z{WqkSeO?EQtDnj?!~0S}RFpDDT|Yc;c)9H?$!ovRzPqz!N1X4Qn>kkZJmON@^0U3y
z62Mbsj!k~d*UMbJ7*u)eWA4OK^rbW@KD$M}>6vWa)<RyBSp1|mS+Yk{zIhYF10<b+
zz6390QG8p0l`1I&%$yhBX7}J{l0umBIokj1r@_y3y6w6w4KcRz9AO;?;?|LIl11@d
zeawiNG`^}Jjl})@gaZuomvoHAr)*jYnblRIXo}gB9SIi@NGgp|ATpqd&IsM!I!I${
zU}gno>uo%xnZK4A66(M(x<Y0swvEQg_m)0mzc2Sm>^<_CF^{6qYN>ZaAAeVnCj*c)
z6}>ewIcIb<IC>f;(JZndFJsW4)JoDivyvx!@%G0VkMnlZGm4EOv<QE<K@g44c7Cl4
z>y|Q;@7R@w!+}=TGc}T1wW@4P;yHT;_pu}@$b6rFwqi&Ms((dDYBfF}=?0q%GYNq$
z`TK6Al3zhb0qzUd)_Ch2_W!3${%-`|fB(8UdcTE&g8yH%m!!^=E|Tj`9GtGV?%JFo
z>-+eQP0h%Oqp%>ft{c1<LZh2tyco|;s6mq46s~2xK!d_9Fb+D7X-po<O$K3*hlGot
zi1D^BC+2>y(KJ6XLNItjFGfb@eyt9G9FJ#fsISSoIbRNXgqxAR9;J8Il8M2B1qVk(
zjgqCH@js)Oz!7Lr89SNCwa$4ns?V0iQH}k83xGEDhkPjK2Ps(tqvpwLWhe@?xTJFZ
zO=l957TjI`xVx`RLgiAc+O6GR{SvG+nXL;)-$Cr3AOASHb=xX~?RG4p2Q<WB&u=0E
zBF@t1<tCMs)|$nu5|A<?tY1PcKuoXERI+33KN@6Tgf9-1D5_ZJ??>R=MRg0M%|`+M
z8BTWRV%NwobV{{1XH)?0nkr>{v=ztgDn(|@qa}N<XZLK=Gi>_;e3`#ML9<qEI9KS|
zn%)z!U|!B&^YPoYPOAKQ5DCd?D#kE<a+44>tB;bYWKZ5(b=J#m{<O)5!W&ALdzsWg
zWe%mveX4nDQvtJJm?({HY5^llE$O`oG?)mkjgc9S8Q2F*9v}0kQ-vC>CL%FHRwYt|
zU1h*Qj(KxUq)AR}>gri;?0-yf4I%YRK+zoYb`%YSoWBi$+{I|o6};F^pa;u)zKgwx
z?Sy3RLzy8~WOt9+EoR0t7A3Jw15?EtO4eG<zK^liA!0%^#S-LnfISxwt}gkeJpl#I
zpUri9=N~Q^J64{7o&2d<bQnR|GX93~WinuV--%zBl1_AR=Yg}%mj79F?+%(HbWKJ$
z{%s6E2Dv_t*1=m5Iab{0Jv}Vqy^kQvGV_v0@RfPC=PsG`aI$fr>L@D<>}3@7=CHbN
z<yB4tWabbWhVml(xw^<lqSqOOkwjT;Q^(rj75;*1uMe~z&dMcyLPgGQkR=q>^=$BC
z$En!K^jzf1qvx;9q{#D6jDQ4Q$zgF|ru>jk;ijC7UP@0*>D@&u#j}N3lJl`R5|^&@
z@FRf)XHl`r>tclVxuk3fvF|s`kY#t9t9tExHB$wL_~~0bmRuE-i|?Wr8xo+M6JCbo
z$R6*Hdr-jfUuF0=`nS#5h&Hk3ldo8!p+{Q#p|Cq#Ns{q?A!Cb8<pu<0w@YLs>*tu@
zxa@N*qiz0C+3a8pBpG7GTZ=X`5xJ32cin;E&27rTx)Ml)hv+Xl7uvIps~>f2^S_5M
zJDzOwoKmf~#-54&2#pueKv)S(@1nK;qpj9p73wRz{YbdT;CBPku@Vi*8xdsev@w_)
zucyhvFkC=DKZ7JakjiyEMM_(2nQIhu>e<oA_<xX9{z=Itn#({9kWYR1_fF6aILqT@
zxhWdjv7?6H^_3pM2ZrUi;gf_%!)T=254m9hW&RrF3$z!docCOaXTq_?vDRapqS`FD
z;KTyQKQ_G97>exF<$6w5#)_&D;>eWitq<kL#4`@tDq|(zP}U2XgKl+gZICba?~YBX
zdfQyPhc2_ZR#R)rw#gF>%5yi@ib9hLGMK+U)Edkp6c3>!`AGwf5A7npjZ{BC=^841
z=qzdQLGPJJMbD}t^@S|8wTfwum<ak;Ex^lPRG@KSN8xXv_5rKdm_^s3Ba}T1dXAO$
zKKVq|!@wb$#x=30wVps{jdb-*$hu`BrO`>F1K*Ff0{71kZNk_EP!c~bm?}k!-<`PE
zvDd5L^jkpO9dgLN>0ul~abO<y!rC5Q%QA~=(Kl02t~X`_hz8cdbVd#x47L}wYG312
z9y2S974YeL+bEtLShUDfA|^6Qb(4~QQ$^+XAOx((JhpnlGk{GfT$P<~86$z*ODa_A
zT1ea1!)MsU&XRg;7qw8Zn;~zrjfAZSG`56&NnVyQ&}xzvl6H-^P#4t;q_HA-lWPEE
z5w9Y{&a~!a${3J+MEvpMR3U7Qde6!yKjquj5O^~uJK_WM`%unL1S+5gb_w;!*=Lk-
zi|n2;7ysgVJb}_WHB4kr*bU<JdSr{%=WzUBcs-F89UGH+`=9;nXd;etm>gipQgH&w
zBwthd=z98awDt*Uq?2`msbhDfzQ4#ehMeg}qO=<9DbxPvB?(QNHzQLOO{V}|+f$io
zptU({zkIINFLus@yRd`iKxWH4#6qogtcDUhx>;Zg0YD-n10y*KXAw}V?l24S_8Foq
zX+8g4?~zZlnWcOz#UteFj>a36h+L_<XMt<iYp|>bHq{y1=uWdH({T+#vRWCEcsoJ*
zJ-oSuON7B~op;m5wxWPJ1&g&^k7pALwX1U*N?<Z`U*M2bwILR6`0g1`yj`ExTPHK@
zit6AY%VD-6viEvjA7i%JoK}d?cF|fdnTdMjx=PB`B*=ZS#FZ5ZmFqjU*TO4fMxkk`
zstZkgzt1(!hi9H?SOHBo3AF*jk&dafd2(*(xeFTxQ8KlL$Qw#wWV9S`ob?OH8?~^2
zXO<4hfsc?eHG%yiDhx*cbN`VFZ6<<F=4)7M6J1vd^xFR(rcY;`+Nx`_&{~thAWQdv
zy~EUvjpzOMlI^V%vzW-s(;nLF7O|2rSFZp?wD)SCrt{PLjjn4hKV_z8`erDrH+y%j
znoydT$<1*G+nRqUV6rY1O&jHW@&{n9e)`Vk8T3Z$;%MC4F2nU7L5Ej7Zp-hhle+w?
ze;S5Nq4KL*mecB^=*<Z&RL$(2j6aDMw|(r=-g70(+MV-|!rl4to_$@ty~rrv)1$d}
z*FgIYS?Ukw`#4j#x|(7%xIA6l#3yG3zBrV;7LSPnBM&bv8%#e1n5&x4ZXk*Gw#{V?
zn(?^@x!Pi;od^zU!xQr4i*b*v59?psN*E$jgk<i;#Eo{yYtrsk@GeF)9JWd>+4PVZ
zNc2n!bAd+Vob@}44=vx^V}Dusd5Zp!BgNQ8(6lpk(6#Kgwq}~Vi`|d=i2k7I75|sW
z2Y%G)Ezf;9X<9*J&XU5^1@dw9KfK<w7O`iJTrQ1DeEE7vL4W%ZWed*5ayIw{6ur{;
zHi&DA=#KEXLOW-PdK{HJNp)*R<+UF+y0eqbx%-We^k~B(JA><wy}(bGE?RAz%q7+J
zmkf*GK_Cw2+sVSOZksa*=Hgw}aE*H_!6h{THv=MQ`y6^`PJTXJ0%cLQYUtn1Ez;7(
z@+IukN0Ukz%$eo<bNdA>2sz6Ut9E~zl3m`24^gT=HLNwyq|><B5B;JAsraN9jZZL=
zG$Dt4>I1e~QE4FQIg=L_1OX)ha2PT0f*-4ux-Lki^8+I%ic5e=jQOsksIMol63ytV
zMk?c+ePP$ejYbi|;xX8?YefPBvT{g{3{-`5g4UG%i`B90V+Z%7LokznE|4X7By#QQ
zQAL*QI7TYZM;)59gGG~)cyVE;@bT9^p#r=^vquw{de4+F3*sxMq(DuzTvg!1L+4uY
zOoI38#o+#d)hN%`tkrLKHEm4)0#U319+yRFKk>Bscj#6VuG+(OqGChKp7|Y^$NvTT
zVd$syXMMY8-k{h3{p`FkEtFI4e)h@D9cwrLeD%)szA$|ok+EBJCK%?NSf*#~_nBYI
z=8^hlB=Ze6h1?vE5uN8Peqd!OM`I_4IXgD;>Nqq`Bv=tGVyH(zAogM^GCn5zZShmm
zySE!fBOoQU%3+gV>bJ7AKM{Dnlzukhn^-4BL-*mT?w@-TQE{a?K*eWU>&s4cgZwF^
z__&O4=@KlFB6dU?jxSW^oIMt5HF!O72T^eIv#~xd>l^2Y18dFnrC4;)c4s7~Jqqbf
z-1JuEi|{>p>Ck&&Uu^Ul)bcp#NFB4pzvJ$QH*8M`{~o~Fo>eTA5)VCG@YR44DLO~<
zs?8Id4~CkDSEyKlwYO_C&hU?NWy;vwIp&Ecxe>R4NCxVp4;TlrCP!9pcyg!%*TlYE
z@-IgSlTdeHL_`6RFj+V8mn^Ox=(ya?k4%3L&H9}?V9N2(9}D3l|4MQeS<fZjE?_57
z0<Y8qIRUPf&Qu2fCSjDFn3+B7(@b*E$b2L!c*hj>ju&_X+S42$A(GH{myIS&T0&1)
zTQB8rwY`;PGdx72sgBf^+Af-kdV;9wqMG-0Q8!?ff-f?AC1F0CXEqKDAN~gZ0R)!8
z)b0gyQAT*O0C8`h;ObWKX1=w6qlBR$KMf-y`~ekBaI>>Yb&|{a*rtxTV=7MA8aZjS
z%<?Ir@iO$#J(|E2x-Sb0Buj9;?P`-^qQldJ46x$MOtkKF>4Qv}Wy5WdQR)pMUf&zP
z_!D&|Nh(7oe4L&LAnT0D60$9$%kPgu6Vx7ZF7#&kv)v9yJxi>mn5i2{+OBk&H>AnZ
zbS&1W;hhOVG@=f&M#Uo}!=!DshLkbNbw+oY5EC6CiICxlwD08jaw}aHNTE16U18?B
z!*XyzrdRvx50ma@V0`yOSs_ZcWBcTpCS^!2r#s&TW;oMCl=@#ankkittf##04LIZH
z_Oz$;=dogm3p4Mu)XEmrGrJAgl+|Dl7x??x!$;vI=$SiO*ydJ^i5Suew;?vWZ#!(S
zXGD~<X6Vy|Xrqs<*+ZpWhnqG_Vh?WWmc^O8EggNTnR3**ETL!~8ou*$<Oh#ZV9;2%
z3cmemr~LeSQ2*MW=G{{^)J=);e$_bAQBv!9wHWU7+jrpg@RyEG52upvhxxv>o=1vv
z&yVj!05n@mJJczDp#^5sJ9;aU{$Yq%NYo^d3T@HdwRB#2NZvJ=<x@K4sZLKW=U8|n
zk}>KC{zBgiYwX(;;6ahRs5nZ*dT=m*+H~>*uR2xQIoHWYSBt$&Hs~ec8H^|2(MQ_I
zso4Mlg~MMjL)C&}{J4G7Begg}w8x&xzeQY0cDyvkbj<s8KQCOEZi$>}N~~aEg$Obh
zc5*dFy)j>`zK&qOmiOnjblPGVi+a{-oHPH)<D?a?<WlVJM}&gl{a~RLioNl%T4olN
zm%}~oN}c(p=8_+cZo_vdZx|n$J&spip;KxNcN*v<$g}jex0$k1fOIl*)OU?wc|L2Z
zj`MQQi=Vc1?q#k*PgyFZk4F`~g5ziw<(_rUt3Bg<-tMb`nZGeWHmCL91t>^Rn3p_|
zx8`>PyI}cQd}E7i?_Ow@sHxJAtZT5o_59UpX;g?3##r<&mW&ici<4>r39Tpq(XZV(
zt;%RH(rlI_c!4|I3>0ojjU*47Hcx{5`Ou{&P07T8^2X>y&ugQ-^7HX^tE5um@R5q2
zpo`WqVXzSDTKpq*p!1WaqDQ!%jDYO&HE;f2@ZciEM;btFXTjNWOG?LDO!payjvX9{
z$-8LKfW238yzEL=`K~Ta1r$WAgGf?Rrjy2Z6HU_BQs9X*Z11V*d^Bl&JdiOMx>%~X
z+F(G!kcIdnKxu3ACD+iIOhoR>ch1?_BaT<jA#(8)^@;J&_>1j&%2Pb+D0dedU>Wcr
zJ?K;0fvuo^w)hyl^^4qmdPc>2_K`KGoKk?88eN<0UqbQzkyu^X=MF+-B&!pB14i&0
z`B(YEnpw%Q5f!n|DRQos-C7VDyUBp%-7eXl7oyU^jwgFCu}A8lW)3(zpR?nFVi7r0
zxlo<6TLW~l42~)K;tji%7mD(D66b!gJ!$^DP2BJ_gQ_cdN9`My;dY55{_ZC$(gueu
zwkCwE5vA6Ms@x#VmMZPN8lLF7P;=Eg5s_Rv#IP62OTy4*2TAFD2mdVM?plB_SK~%+
ze_3^fa$RSwvQ453XyF91xa97l$l6+JPS_ZD|5m65o4)#v3m(0gj0yh(wApGEh#P?A
z77Hsz9bv>Z3j^aoMxrCHel7f(g()@wep^~h$3cvzYt?s#dfy+)1h83b-B4vWTP*G1
z&zupRHb-syg^3dZXby`A;sR%yfy8_{N*Iy%m8*m5M-v*MqdST2`Sq%FwKzn0r!nU-
zi>~uxbn;gwT)6$?N*R^iw<TdOlT&#!kYZrKGX$9R7>cc%&VJ{%k22Z0QCcC&HXtWY
zlE*EJ9Ju36FEP<vg3mO4pzj*$%rI2b<t3o!WZfBU^SVLw4Pm{KU}Z|%Sv*3r8VP@j
z&lykHuC*SKNme1~o6N!lyikqP_pB5oz$j1Nf`~?o62RJ1G!J2WBdU&M7eO${nG(4V
zfUEwp6(gy8K0)oO(Nw$j?QaBTj$5@~<`f;n-bkLjZpC;`yv8Yoh^Bg8H<#B-w-=Aw
z_#E5GxV$*_-4I_nDr(G_0rOCE&|I$@c@dW>7;rtYJki{Ro7<yD^#OZT$5Z!$_`|-8
z-?PRoE6C}&zvoigfMO4qsV+*X)WYwoAl=oGD)#nb^e35$Z^Or`R9cQ+EFa(VG=z>6
zfhvxgs%T{;U@EDnCu2hF&jOb@IbRh}r@ZX@3#7FdXL(X{#SjyIm4Lsh+%VaFX&g!a
zL;vFKFLKjgahW_{(+?`o>7Jbb1!9*>%yUcp&qK}g-$V~XUz$=f#aL*LrlU8Xc=CJ+
zGQ5i6Ij{N6Vf`1V((sQ+Kb>);Q1%<==>{%wK_7K`o(&5?&~nT5FA)0ftNI^vcRBOE
z{MU16er8$H+~vBtcO~aWF8G}MlK<DvA8_p-&C%@t{8B916Oyn0hYQf34Kr<v8@kyV
zH>+}M{?h;d#<Pv6uBn`vN0^SYGlv$^uG%5IyF9oMt0V;3>eBC~9u@Sr&2RVqvPE&`
zaf;OY$=?jkIJR&v6wCwT)a+>D5MEECLBUYKJLQqAS@;M@&jo@f#7jT#;<l6GZqHG%
z>7JFI47Hrmmx^RR?)gv+C*ri>^|LW)G7BtW&^o}$r;<G~WxqeN>CNeV=7y{+@N1Fd
zUCSS=wMhih=mqb)pD*#;ggn}0<ZPt3u`y7Y1%CsCdiC2Nq(0^9yf>TcED060kw({D
zI#MtsVuYNyDnr)-%#UJyNhpOQ-#j`Mvy0vRWureBvRx9Q8TQeH`T<+Y9Wz2$nS_ay
zz~7eIh@=Uki5duntM3zZ9k?(|xmF-c)If+<{eGSF-uK7uU}^eVkVCW|Swh-qsNGSE
zy)6(?cU}4Uh(ey1IzmLAHuLN*PRTapo_t1%E8_JygUEMzR)QcGs0HD9M*%MqLb#>Y
zAP?^nj3Og&44$_@HyABTs0*4U>j6S_s7CBJSyxySG}tb6#ZN2)l@>;+z89_l22@|?
z7}8H~*l_BC@87nNJR1)##@ZZ{$q|Jql0-B~pQ_-AZx0?A(F-MHdJ-L2?+7K(J1Kk>
z^~ndkCC!qX4OZvb5Q^A0k#&5mp-%m;iAT*&P8Y^PC1#Yog7hvLIIrs74FGCr6^$nO
zmT3<KgJ@eool=ofkJt-+?1e*+0kY~&aqqv0VKm{WG{8{yjm9||&+ASrf+huBG%<fX
zcygdM_e!51ATmMohVrYNAlLjw#kb|%m`#0U24zyZeBS_$6SUzITqpPRM*=^=+SZ>*
zP!N%oiAXR8Db=2yjkoL)HjAEf{y=Hq*@NXWV7>#kQRa?1_v=Tu=)SBMblFqD#ZZfi
zKZn{|q;zSr0@CWrOIFJlW;<cmZ0|9qdfcbiX9K1Z2M5W$eYpeX_KpuwxN<4|ck!R7
z=DIUJdE1968sSrE?qUgxG;{Q{C@WDDZh9t_a|l~|Xs(g{BS|4841dlX11A9qFCxo{
zAIg~*Gi7nLyZ)wZUIeBMS$kK<4@ET|t->>M^-b9mU!$q1nJn78`Y<w4pYJ$7L)xQc
zO2rNFHLrq-WEFo4_4Bol2u!t>mDUUHpo$g`dIlmSGT54EO0fM%&uhP}b3}8+gF_Gt
z>2a7A8MZ}(u=Ml#?+t<?kYf7Q2P3P_V?Xc)?eETha*~hLB6els>N37Jaj5DkTfr%Z
zPb#DJI0R~$N~)=xiPH2*x*<vHB{Tu-$?k=ra`o-$E#ZiO_$AGSr;DZUK2{|~9heKt
z3({qpN)cBI`WP71K`+I#y47y{1rkty@*4+ih$yxxuDSgJd%wS@zOaS3jgeO$AVewp
zpgVh&r}XJ3SH3Mkr}4jdM*r<6axkC<)={Xu05T#~HW7<s;#WeV7frczN?6~vAEp#h
z?{(=yQ3mzAWeFy>1obfw;xcB8ZhSo`9O<Q3E^YoW-^|PmP&m+pTlXE9d6ZKAYy^T4
zn!#ji@7xRibb_qSDBdTUus4ty{bb?}*A|GcCmR7nj1zE*k?_2!=&;}y**)^hqQoyQ
z>XOEYI=x}^rONBgdD`FMEt9z@P=jL*`kTBz%_l;e?#lvnJYbyirb?(dnjMrRMJIjW
z0PyjmsaWx;zc=wIYl$B)9-D_kB{Z#L+k0Bf+7M>$&QV@r0n{RlI7`(|uOZXPsq7b3
z&>L*RRq%qoVpTGiv)N`os$G&MqU7y#Dj^Zy{N_=YmfF)=5gbS<1S*R%^G?SX<<JO2
zE+&2f2?JMOn?A*E#y}J7CK|qmzl9JvA6RvD!tj%1>a^j%o}?Q>;mP0Sk~4~*3K1l=
zice!iM@C9gy~9M=s`afXH<)-~<npER+85Y$v7QP-d`|xcf5{8qcvhnk10$TAkaCpr
zAx!KLG7_d&xr8bd6c5)*57AJThYPNTKwTrnt45q2rygisZ}J;ZRr5?D2_EErF;8tn
zO{Ijb39*3EGF$BsH&>tNh_tg1H-pRAs~y(^z{BB(-Ovi-#;nbj84tj#@GVoHp=x7z
zROp_yb7(lv@1c$d#TP2iou(~VIqeU-pSO`l&(0M|${*ZKlD=rr{LI(6p$u36umzi-
z7SjI$O<9RXU2r)3?zp*QRLab}F7Ovf0&*qK-_MXU6yJ9EA-TR=bHK6%Z1``oDKI{x
z7?dfzofANBJ8O5Q-b1@;A7*H-%&Lz``(`AiZ<6P|q-tpEBo$%^L7y4TmfUZm=;O4K
ziqfTzZrrs!6?;W%e^6LqA(JAuqZdf0zIToLVppi|;adgh*{v$^VCzJxrH<ztPq_=H
z+%^3Vs;?%hvpBtNn#u@bvbW0g>t(=GC&rn<p=OPh(g$XAjwenEk-DB2&3e~x&a?4o
z`!IREZG8xqPrg)w`5K`?){3S;61t4WWm^=ii|}RKHb?uz*=ntk)H<3b`f65k@&lro
z!)y~={^Bj;pYYOXs<R#@;)_ONpFmUx1D8KB;o*8qBc9Dyut^nBon?|?C2f1*l_}$O
zg}eKj6SsdqV&cyRkB!xcaM8wrOEp`cpU@VMbHC{HV1|ux<8|<O6LXyS!37#u{xB{y
zwqQS*FWg8O10D(gAo=#){0S!0eyqYsGb*;L7SlcTZv2;hIZ^owzwbJE779HB#zF)?
zY3#RpdmkZ)742iPM|v2!aWBgj-J8lS3@z#zL0(ttnIY`>EwTsfY;&`30`JxVA%TKD
z{>VnQepJPe(Zx?st^E1psAZeHhHjn8F2}w2_=~ctY;KJktM3~DIPBR;I9#c@{Q0cK
zde@5Vs^l*a?~hkJ+lYn#VRHfYnh?!ve}Ogy-~gCndDd#s5*YS&-K&NT(Q}Qz_9%cp
zsjV>_W$^7lyK3U^`ALPCVxJCO=T}lb>J0V}p?DTGBzrL}mp=}qcu%N05=RMIP?7^k
z0L3XfG8f)0+D6%|zwqnD3`#rlY~*aON96d*%}I#NwAeE$XA-E$Ta4L*mFk529R*^Z
zNwH)sId|!JtRfNH4Ucr9s#rqDOtkbCe|vw!slRkjMCwAMMxpPHjXb#v@X)Mb3XO-q
z*xRM2^%Ap`&^|o;e###Zldq+&WMN?8<inIHjfL+_>}DQ=zN`Lh!K#lAfCR&u*n3ti
z+}0n17hks|87mM~EQE^kQ+V@?EY?)KmwLIau4)JQCWD6{EtGMl!fX8KopyHGu3(<$
z<}~>OB;K>cJ4h>010}A`UL1&Tz2XHGnNkE(?O-=aS*RB0QOi2@1J1^rx4UAwFHSV7
zLUsba0Mi_{>`?LH&OBM3aO*@ISbt5$?THb<q}l3adriJlg&J<U(a&g-!QQxU`-vc{
zjrQv@;@?Pw4cet{pyCR1wJ0QJWHay_u|Mh*5+??0ujOB=+4L!JwlQHDsD9%)^u{^n
zkzG{X`XIMdyT=$YMmjpUX^ua!c-bc2kF3FK?#zaFlU-LrXOI)sogoE93)$q`LuO5x
zvLNkhRO=BgpYJtHum(2TjfF)QaIOP=x(m%~!e5|`uYW|Ty{_{A-hKA{2z%OCMV|m&
z1M3B@Jc5+qwV@8Os+&4tqCR~YJ>0?$NBT;}ynlfl1Y+-|c|Cy*B>tHAg=*(XeJa2z
z`LUcR?BZ3rB;?_=<ziRh!&_;PiIs$!r=2I|Nmcd+E1`>>pO@tJt`3K?u^8)M)neeq
z=^)xOOL%4SFQ>sGR~NIIxLWO~q(iE!t|_<WRmX9=g2Ibt@2m!sO}L^uVmww%j7y}1
zy+!YGR%RSDFRV34V})5iN-MBkU!RoG^HMuA5?f9ptYfn2Z2y#Qc@lERy#kTlt<BTl
z?#ThUw`?{SOo8rt<TD`wGqDF9ss!7@Irn3SWD->k-&Pygf=_?W@ux^8NqqzMk~;Hk
zOgv~=zSPe|e=`S2iVF?TjYqi18C58=Rysc?ZXfZTlEk=oNpTo53AStFAgN5|X7oTA
zL?O>XW2MxdTmTd2^g-}x8->cd-yV_4weT7L-k8_!>|r0d>hIn;YJ{QcV3;6|mmCgH
zaPZ=!(wed+EMP3yv#lE0m)~P%;+9*Bi@9Btr;~Ov2;lAL1<F#DSe<MWg6l$!5UsgJ
za)Zt{>>>H_{DC&j0TA0`59n&J-fg>3eN&>LzK<+rPD1Lmf`Qq}enFRK4(HoqGMs?8
z$C+(y@O{;-S{0aD$Dw&nfOGeO<J&HCCuwvI<Z%cFM;s_SBDKJ8Q1PhxNgZIW-KCxp
zb1U7(m8^SN`vNnQm3%`g7nbJJyBfqz=pi+KMP7DZ=coC&VE;p51J_Mh@`V>8YnrO@
zXc)!8kJ+$#zwm~?K#$XJ{sG%u0l`pTYp$5ZU!Buh{=SSGsxMet`AL2$?i9DnUitk8
za2(Iqb|0XOLd1RCxPpYGbe+=waD4U`XjWeH_yo|0CLfGFa~<+jiEgzB-_HLdvPB=T
z33RDRK)A~eF76mecXQ7EVe;|n;nLMr+(pCRAK&UeIq&`|u4`fNzT)Bh!8tbWv*#4h
zGy`}3tFg;}{fB>?G;V0*$yEaOFHl_e59>+prE$%F9RJU2DNui!^N&CH3q;L)>la!5
zpDWGJh9VDc0guZ6aP3)E-(G6}<M98w4xCko4-@~(N@f3J8I*PZ_m=ISm+<$h{2yI%
z?&GHiEl*wbWnEGqcD|2)-!T&-Ce$!#akO6b#_pZDsbb2IGYzRCS{VZm&&jwf_>$-}
zFa=@Il1oyPsvSpPBZt%`%c49+AJM9LanADdlSRpmnKzXoG5k{Z>rrN2pFeFk@5tZa
z>%FM!U;riw{L5+!yA15S&7!G3_nXH9pPDqu#ols`d`8SSG2(bisRd79b+!8=4OV4g
z1BjS`E2GsRgZ6FXtaV8vw4>5SDhqi=hHSOj$=ny>lur73V=Fnc(=_<+Xs^AjRdz34
z?K2O%WV42I2&j?h5E*%d&@#RcpjqugmkdQXqOjuvDG)0^h6?n0JArJ~i@U3tlA`w<
z5VN!56rzZm++td)9x`#-6@8Nx=*fAx$5HE;<Eggjr_5M}>|z+A@)9KH&yfd-C<BnD
z)=hW#J$7BX&J!&LTsRH&ODE!l14osyjFiGnXPR#~urga+-aQ$k;dXgC|7*TafFj2{
z__MFs2`&QpewKUKRyGRBT)enHnD11eEyy8_)E<Qs@xBJxrplf}XmV{6$4u^<eXDb%
zFm#1|0o}#$GW@okAkL1a6w3Oj`DVm1>*GqixS%%tw^kkRh{1AQ{B61T9VC?=l0sgF
z8%j5PQigiY0<zI7x`r$<lAG}LOO>~sKi{K^^;BvlH!rU%f;(PLhXcbLm!i#JrG{|l
zg4_G`EGc_-EiQjVHNQUw;^y}6_F*9>&M2880KWNo^|hH7Kp8W7SmJ7p^4fT}F?pe3
z&9R{tx2dz}XhEYCV_v9SN_Pr0gGgbDfxrp2dZ1k6OUhZDQ&qrIcbANxxO3cHlx0RR
z1Ho+M$toV)-#&o+MI`R0ye~TPcco)iM6m;VCgKsGkXa>AL3p7yRU%=@v(_BqorW}`
z5Lj%>Q~4}bB=98kz@E_}N?xe;-4=J}c8I038Hl&w8en1PDb2}k=;an=QuHumeblCZ
zM_1&j(HWay80&lf<`R-|=^sl%qC9Y}D5rqQQ4SwOlSV1Wg|#bX`D!inZo}BxdZT(J
zV0RNW4`r0-6KJty>QN(f=-o<YlumG{hSi{wWrk2uV=YF-Uto{xyn)iPu;E4?GRs7-
zWqi*Nl1bD{R^?2klW(^jC<!k@hAG@{41WaA%n2gSV6UPgb&6llnVPLYTpIFPC46zo
zAXImy?GRSQZVun+FnqFVzdL^V6LO?+zRM@Rk?33{i&#?-8F6!6OvNQe)Rk@8+(<@@
zJpHx#3&?P|tGjut*4S!-3aD#R98}j&O67jB8!ZRkA5Nr0tfiCYhCNuQ>UiD}IGD0h
z7kbL>>sy}ayAad3;5`f}ZD@0qMd>J`b7jpNiCV13W1TziSbFg|1g><2uz%t(T+CNt
z9%U2gQb^)AG6tHzx?ZjJh-0JCIBL;3?eV{-YyWdu`k$XFYh1V{?!jhKAKoatE7v=?
zgVHYxleB^z-hM}bD0_L$h)AQfSjU_x(Pa77-7)a^AHByGA=5Ef5Rw8iU$kBv&qwxz
zyR*dTp&KFu4k{Uu?yE~^kM}rpK>w~mb*74gq>GvletHFp(iz&=E<WlrHgL3mR3>3W
z<4xB5jy5A^7wide8c7j5pvq`Z8l;7+S?;o0v~ByLxU8A~!xZH)lhh0<MpoR5GZU+g
zMEyrY83~KU`$iZUk+B)k*`Ed?WOCeuCgaq5!SI*P`Oi7=#K4VqrbVU9-?em<UF&KZ
zAO4|4^H<R?JLc&xjH1d)x~S>VH~gT)^CpIv*S37Ccpw+)S_u3PT#BpGL>6WZ`ELEb
z)jQ;sjbnc{QXtL5<q~(G#W}K#jfh?@(#_Vkqrc-!CLwo$;ghz2Kpu0-RO-6Lw9^Ks
ztkj~0%V^5xQ<NKgGW=h!S4T;S8W@1cQH6$}taF;^c5+ti;P|Kr>z4b1NCrC@Z|c!b
ztvTm?bRHof!G?d<RM2Oz;oUDA9ggw3{I_zW(kbd9Bm5~L(B1q04|DGw&-NGpkH#Lg
zg`(6RLDh&&%@7(Zv1+tx$Esa)*i`IItRS(q6g9fEwTT(3iYjWBR;^Ne?z`Xb_x|qZ
zcfXIj9`}!X|B?4Q?~`**&Rfp=oY(6)aBdxG{$BHrRFu!%G$setbhtxRKq$dd-(_EP
z+m*gl=G}2X$^w0Dq)f-oFc3w0wF6geyd>Lp@An@q<60lXxyt_Icn=bBH@Hpc>U6VD
zbv=M9FFT{PW&&o4oOZICNcgvEdP>QGfoU)G_{i5f@Z53{H&qTY>BjVRgmcmr#Q-rM
z243fk(Xn?CACjXlDG!-4Mu=Nr|8o(UP+;vxsO$|R<WNO=4Z%0#MsJ6?Hy89v7KcJN
z)&`em5Sz$OUk~$+G3aVe^9sSpZQu@v6jg@Pcu{f}yRq>|wY0m^I>GSvh*&jvV%ah_
zuGqG<?i;O8odwd`wYOss>WI-1SWb#?esv4hXDWSEL8=T!<~Cz1#jhb+lURZQ-B<wq
z2(Sg1RBHoz^|%&cstQ=`VGzb)O4bE!NF8%a{+GnvOdSz%42_~RAVMZdu^2NxIa`<g
z#t1wCQ{~sNG!2wY7*NAz-H>qBPtcs7T1N1|@&$RP2NWg?N_z)JC7r?=1;P*Sb2gpX
zh%G319M==LimBfdQ>+FA`cxX(sBi1nMhi7^N2_Q|`hf>QkLn)WI~+X3<%Iy&@^yFq
z(k6cx1ZXFw<&K5DywJ$sKnsU7uT8P9v3S`#B%(IFx_>}FE_jt0zZm<qoXPN(aKm@t
zp#F1It!E~oR>Al^BgwDYiOarCEWk2O&<k@rXJX=}i@2rMMxz5$S77C~Q3)Tz%PZGk
z5Su!*)v_Hqp2as5kRJ<NKvVF;db(4>f|OC%v5yDaHyt$w>4ta7GmA?T6<`Vl|C=Y%
z|M^j%e~^u{NHhkq$WNyt{U89o(-!sv-jaohlpsr)N*(n1DWzn4DT_u3>lbtejk=qE
z|F#5^f;N*hq`{L6^@F|)3UVPYh?RGa()$#SFAN2|t%ZPm<W&h47TTJkXoMc%$n2$R
zB~0{u;u-Ux|1nv2-9k8k{h-*V?|MacP)pW$&CK~>zpADLrSB~kF|M^Xqj_0KEQ2qY
zsiIJcfBX+>Me+{DNlkBzgPn(^$4|T8%U=TpQsubvg6bwW{R=*?i49FOXqOZf-Am4w
z0yy+SK=>m(d316q%)5#{9WbGY3eiA`-<el>Bsr{b?CM-mrXB!@-Br_ijffvR*Ru7q
zm*y}PR0#&f8hqnz?ylv5%x^1&7Jd;B3FSz9u-xy{*y_ZLGG37rf7{s=IAv_sX6;S{
zDW&B^A&Msf5Tu!%KaomAg?G8Do|+MTRgFH+n{6v)DrULGXYvDPiV<!bSGA^Dv!>wz
zU>pwx(_h=iF;M8hi-p>SV~xQFV2l{+I)^|qc}Bs{Hp#RgM1o38SXIHEW!#ei=;+y$
zWOoGWR)R_`7eh!b8BvbC$6wC6GLRZVTjD^Lx5U&o5Z2PeMQmmK_*62JVWlwo>iO11
z*6jX`m{uS(M&w<c>LzcK?w-;5yE!EQ8wlek0A3d`mN5&es;-hkA)f+^i+?~%x=mbb
zK*(%3fNVJjDH0|WGm96DZ<#6MKB-N;T+ty{8iukCU>K170A+}=Wq25yX<{I-uTA`M
z=_Mmf33zDRG*zFrotn#$fCrQq8B7n>o4flR_d4AN!(9H<?;nsJc1j+WZ#)Fwic9$>
zuCWR@4#ruBDs`eog*nh7D%(am1#6Pq33WfNe%Sf1B}DHOu!KOXf2;`qkU7NLXuE&^
zsp0Lz56RU>6JO$LkGLsgL#%#vKeS;S(=J4bHT2-jsao$wBiWcE^312w?I@Nr=6Ih2
z+W5-Zh(WwW1O>YeBU`0&S@F1QpDdmi31JpiM4IIig(fb^Y5uMYDlV1GbaS!D19vBO
z)}r{(v#icHBBtZhHK~kX1sQ<I{Yv+8w9y74hHyK#!IZuz?dkpA)Etc3MpDu+p<~p<
z*0Q_1(VShPS4+Suqfc2Z8$8J2^}yoKx6|>PhK_V%qNkQQPbgplylU1o$q7DrS&C`>
z1@qZQc|dgYvwXWLqwVNt<A-zCC!ccmc?wo6=86g5{b6&w-e@G)mAZH!NNbmR(5>t`
z$*vtY%4jG!P?fq_7?BB>tK4;vX(RLerRJV+Z~Wbr6%9GcQ>$V7Pj~OU={aRbx(|M|
z$axVEdsy@1hu!jQ<<;fP;ui`MRsXlWBAF0=+?zJeK7dQ|*Jd|+jjIQrVs_Ncgfcz2
zBIw<{c%&V$5L(13a+2$CQ-^olDiPAFHFAX0o^;c$N)&827wq#3L_ZUJj3#$B@b8|E
zB6TY>pm<Z9qf^8h^qoKLv)YN>6usnSSxjNI_A5%=(+8$AgGbX*V0E3P+zAOw??3Tv
zur^jL<aJu4*<&`r7eASeU3mMmjf$NM;s>{p25*&T3fQ#TbZ2`(0^O_QymCt@(CXBk
zssJP<y-k?Nl6&?3%6FKUoVDl4TDqF%f+ty@D<2n1zq{-e1oRI`l_tVMC}cr99-ZN`
z#?i<d>$KhmEH8cz1DEreG!j;W>o)G_$qq$5`f4#K;MTo{>(LiR*RwnouA)f4I$PJ`
z#d&nHFKTZ1S{_mKmi>Lylgj~z4eE_vzqZx!lA2UEA;Ox)%_3Q5uPF%dUg#7;S_~qG
z)oW^{Ad*vPdr=pBbrj2lcw2F8A`ca#D-tQEpERemM-YnyzspO?12|{9Ri9$)UG8P9
zXeteQpx%5S6vHJnN7}%jOB%yE0E9+C4Fv&YFbl@*A@bt6iI6&7eIGdsS<rf$T&9uu
zUBtk)HnXV-A{MJ{8iT*y`ba{wU~aAJnv+E1KELr-KN-fJ1VdF9%am0c4*}~0Oi!W0
zj$MDC(4i98+{Du>=(PDC5JJni(HjUl1pEVf1*modlLvvi6)L&aQ>68qr9q4cXvIB9
z<*Ejm2*LcQWZT39Xv9hr*T4yO!cr#I5|2?B7m>@?Q`j?+_=)3f?JK3cSc!WpnNI-8
zjv1F}=@jF>uYj`y*x+=xh<~xuY&65W>p^fyq#7ts)2<H{-{ElXHTS$@sOj#G_9fWO
zOYxM{l&l$EJW44kC5HN#rOR<X5$oR)yp5EM;~vZ)4G~vkSF$G?8cGj~6!p$VNJX)X
zRax1l8@>=JdGRYu>L{ttaXCO_i_PCZJNa^(Kwx99D$M8>hAkk=vnZk~V5Yw5WdAmF
z`KvhRgVt37pu*lAQcF6uWGIOUseVd_hr(GRCS9mA`X%?VIb@5m5H?>sL<CB~IgL@%
zgi?-<C?27$GHsv!RwqEJsV*xe8!;aJo8wj*Pc5j}y~?ozM_0)^JsZ7bnPZo#vfNPK
z2dNM{5VJs%+4X!au(`ghs-N#ww!rZUkz$v@l>1yXjfS8STqpQsvGv^;v`4+&4ok{b
zdxTCZ3FbAZ)~rJz2k4EpHG&)XInxhB@u`)~Yc?A5S~F$7Q30085BT!Z{xsU~)$8-y
zF%-&_iSd!Fqa*a?^s;ZhiKgdC0pOet99`KuKB%U+Q7=ds<&+xQ$8F2woTEN|W;k50
z^guJUi|f`p0BsqIh!u03T47(H9ath^UuV2<c!n6)e<6?lsdH^)lajSiG&82+8PY_d
zc;l%CP3>Gg-49@)XnXbna5sLC;<3A2zkU01w!y@AMb55U)(5Tv2;UXWAfi;>(uCbd
z!}g8|8O6)f*UIzQ<J*Gv2LY7Dp|zvD9sTCoPtHnTN|d7)k7L(R(YQ%SQ)bFB*uhI#
zHP5lZMqp`0;3rx_<UbZVke@n=Bf|`=8qZbVS3pEAz+?|l;~@ZHb#d_*U(38eNUEu=
z>k`Ee)O?up=_2Gau|3b;jZyQJA0m@tOD;`)2`ghGCAWe5?+9Iv$H0~fYw&Zn^|J81
zrG6$B#+#r5E|=uK&zLh~W}HFBXWR_gA~7=0R=&Ps5HTrX{^Um66|z#ulVH8IH9}bI
z`2>d2z;`x;U~k~%EMjkfDxQ4KQ6;aLVyb37!8()79#GG&tfTiqaHDnnDHJ9Yf_DuC
z(2WV=Gz3UeBBi;YwbCvv;n!$@HctcsA(Q!%!VxJeE>!TLmdG`@jnJD`SR`1T^j-#K
zoXuPMVm<YkOjN_J=E#DtcH%QGzE5(OF2)7H!INY<-KyKhT&L#U5nsDP@EZ|WZOB+H
zH24RCl19XOi$C*mu4l)k?zPHJt&JaJUT&Hxp&G~}q+p)^jj9$o{lKjkP0WSw#>(pg
ze>b`yn+(j#p?z-i+aEt_gcHcVOWiQKZ7?h+zL+4zNaTYV!4XLJXbeO_n|e!v%a4H5
zNtrSOlXI9#-G>!($N_n1UuvH!uVBA`TP@D!0ho+5wIFj!UlxH=l9<LdT0~UJ)m#y?
z*5!iBl>upOfo!xK!q=(*EA;_fJ!{ipwZvhSkiL$oq?`$D^r<Ct9v8Sre_}b?PeLLv
zCAb!xts_!&l#-3<n^9bCOX*9zwu49r$iFP_XzZHVvSY45VVPMZ=z^UdKS68JPpW6N
zD<zU+!y6e*BdwPY&L!KPojvLY0xze9nxmz8ZHkg!9h{FMpI8g30>mn_fTHl~+JjmF
zZ|vQ*XgZ>CB9<C)hrH;mSadSh--n0VE-j#sX;N})Oa6RH@N9$YQAaz~TOPA^c*6~s
z^hVI_nDdUg!xp#Y;bei!BV5NgrfT?Juri$L9q1JEIo7hJ!?$@z!6^FPA|?-c*WA$#
zcAxtenD@9QODR^h=v>KHGr4Ag|5uGrfuhsIWTDE9Ljgrm+_gSt=p_BQ)XPiAvE1UY
zuKNxfK2c4}84OI`Q){M58ozsM9h7}nc_Nx38~vSYZ)*(v_+a_%VT1WAQtfT@2KzH(
zD-~!$QljQ&=sw&P(@`C*Q^cpeG^OBD8-$^~+Z7eQ3Aes#>ebf0sS_aSp-R+~;H$-Q
zxbj-ryc5Hc)z4I;ETMMEgxg782DefJP#Djlav{sy3PbQQ>t|<$tM5BXbnmI{eO~Mm
za<|?#<ERr&<08kjsHdt_wvqV*7DJ6`Vs{!rqg`Zwa!xI0YZcfqpooKp>l`Vw7=ww~
z{VUJ%zP5aX>WJN;Zm0^G6Df>p^me7fyT2zM&9iRXxnT91yLa$o5p;aT<AxgMf~15R
z>h)rjs}BWYv1ojR)Y<N}bl7+PWyR=waImaVb)Odif50o<L(MNg3(Ig8+=WpG&aO)?
z=8Dd%HMZtmdb24)E!r3O2vTruOh2!$zwb!7J%7&O^ftkppC~tA-52ojneoQKfbVEw
z<x8IjHH`(IrGl@ul1E-&^J0q2-3>Aydla;k>SqFPpL_E$k?~ARcIknQPL1PM)9*ys
zi`GP?pg!1#aHmzW@zU9Cgi*`TnLT6cRf#ug;}TDd{V(G*w_ogi*chxv-<iu;gs*)$
zHdvU}UJ4<~bF}dVHI9{^X--tNB#7tv=vP^k;9|@AG4qqsyXBd|pZ`n+vClEi40}>a
zYJJ)DGU03>rv362e%FO=nM>SxV=Joj{v?hCc-DXM&OCDoAmpZhe|{@Gt%zo%XDRCe
zfGsUd`a0_ieMUcg-$VE_@u=z+A>qqt254SJQ_fm|1RXd`{^avrrcv~@zD`75nLOMu
zL{s720ko$fOHQ$W&$KAK?j1d>1}06hbD8|A?n*89((<bdZbEy`O@?%E^27ut^EdGC
zasN%VOQbaV>F?X#u{kWdBS!y#e14&QfaGG&8^OE{4p@Tk>Vf*3A6mIoAu4r{!vC&A
zfpdS_tsC7BP5t^ak7BJN<HgK!rlCZAF8C|A5Z*^=Ikj|J8h?ceLVf|p@RAWeolH0}
zpp1FV1-n_1QqEXi5vCihham@5oV;!pCi1b91+=2KmlHC%eo6@F2&F+TJf+fd7~Fqg
zQ@1oF*m-*Gu@*5wzz$f>!j}^3q6kg75|46Z(q1wb1e7fql%1w7Wj(XY(#QvYCm)it
zki+RNXOtFJk?&#9yp~b#F=hfhW>Upba@y}I?(C+F3)0k_)q;LNY@#5Gx<HKn$u&|8
zq+*^S<^z}xCq=&!jM3tmROX!M(+X3ws=(=qReN>z*>B$JN@#*L#0xJm9encj44m9L
zmiZ#Jb9LIzKFLiLP1i8eSWc(9i&j8TRb-4OHcH@S#j-NKQ|=kf6~ytOfhRc#1ZW?S
zGfe2f5mrTzhqbYOy*FjVQIydWa7~ygvvGo7GGt(lu#jy&q@Ub+E-9Y=R1Hb}MC8d%
z-FG~{7oBJt!fHEQ9s=CRw;XDv2bZ0$je5b>!So-xd#_s{@}A(iYhL_B&wGiu8C88h
z!iyywK)E^OZI`TA?t0Dr(&rZPDf3!g?+nMw^kjKuDfmPe$0Dv_$D?|^H(aRA-WT7e
z>U@Sp%n*t%8i|{yRu9a6pQfLGy>%%6phT0rRSDwtWpDEz5IINcTuUBcBZF3jzHG+o
zr-(yiqCwTQeGt&uq#w>DnmQlaZT`EK<D8L-Y2k<IzQ5c|G`*PHDz9a-(#k%Mljp6G
zAwv=C9<^Be+`ED_AE61Mi+p?F`<{Y@eY-u_O){E{K+QS>Fw#v--Zsl~NNts8<<OoE
zx*fBjaD}JOLakg#MTR3$5pAli+R)(bFHHxs9J0!%F182On_>+EA`GcJMlTh2P+2)B
z!(60{d<)MCczkx%sx8DB)vST^``R@fy>na9>tS~DHnL2^PYZgm*6coz8&+!r7y*)z
zm$Z(%&pKi3XkBUsp=_WaV;E)HZJ18LI8stQm&+NKRak3%8)kiXu!>xwR+vV$3hJ0h
z!4a+kE*Pfuvh-q2)i*8ZReO=tdu^yI=Z#5D>W%}dxbq&$kv1CaT&hy9FFTL^G*B<k
z6)}OptG5|>U%U?5)>deaI!ivNRY1CP;{jf~@%z2IK&lT6tS!GCXRYbVjYdrQUU*!c
zzUiFSqANTktT8j@W#Dtu{qz0b)>XR)tO57UP1I=vv7w%qF1^H<kh=XIdwCyJ-81oY
z-x4SaeM(~|@NA;))>}P>J!~l(J&Ei2QPvQ)_|_)&^|DdsQQbZfs{cau^r6Dddw@AY
zD({o~QjmAlg`;=6_QcvAQ0N5JIo@OsG04b3$KIY!{Ipjy&)gDOdos5ByZ>sA?NfKn
zGrf8Qk9w-^GMwNvS^hAwzwzGhqW-`wHv35D`NdCt8Q*h_=sGSaC!0tMX_|#4$#T*?
ztJ#YUi(dJ2F@5d%fuqd~0+}tX4^U!Xm8u^FcyJVdM819{F4sfMHN({#EvJS;^-b({
ztmWd#QQbJzLJDuYqR)hvcTDf59@kw~VJu$XuLW>MGgO~mT}*cT(J@IdlimHiiKQ9o
zyR{urRXV9ya`uzk^o~nr#~)I2Sdl}5k!!kDmtnzfW?Z6lK$<6$yT0OU^GB8EHzE(p
z4!@|Lug}zsi+vU?-Cf==5HDQ)mGA7QI`<}~y3Cg3*BIY=++suv@^K$nA&`ZlDx&}U
zJXr(8n$5a<#e(&|N;5Gs{R6-;GDiI-I4TAYYYQ#^Ag|1O#_{WFZocQ(h1kky4Wx5$
zw<Gq9gc4|@G$G3z8d%qPlpp>lysAJwv@+*D*q$MziYxSZIYW?sxv<nz0x{{=E3Y@g
zh2NK(i-GMMM(9cGXpe^%qY$`I99MV_T@+4dJ=Sy|CWp0?E__Kb*X599KrKO?$a*qm
z_4ppPmd*GTX@bABU{I8sgKRm2kw$Z)|1~)iDK27?OGbiw6QR4agM3*2Xv;wHMa}#?
z{qXCLPHnz*$~>Tvjy=8)Df;C!jUI!wjzgxKPfmHLb=`BHEC}=Wb=eO7$}gR^kfFT~
z9%OY(sSMWiN|uato%;tAd=VV!1_&YpuMP1-v46xq{;A%_UktQeY@TgAzbwi8z5dQ_
z*=YLrBGj4qe#OAU8m;HPkzJ8$cmDH2HvD0Y+4rLVEb|{l^CkHuS<u#H^I9d>k4$PW
zC{Ct;df5L6O#wAm{t+YbcSNd5^VU%3dY@0707M(txw5ZXC7in70QIH+>JlT+jl_Sq
z1~lRDxAb#}i1%ibeGRQmS);coo=^9pjY(ogqEPzlfmVs}HNlG=Gf5~stTk~m=Tpyv
z$88!lJUl*kwLYAZu2p|1dub>s*5h0C-t5vTsri#Z0hd+@Z;fX=U#m;|wlyVdDJig`
zKRIk~vRFW8F3nPknMItSK7a_W%%w^w3Ok1+%xN4{4@KSk<Z^)b_Uzs?l;pq!HL(7Q
zEOFXG;X705C_1UJ7=_#v!>PuCEX%A%o0T{Jj^nl!*m;sC_~qY0%KjtycZ92lVsn{z
ztezd4t=p-YmiWZ~tllq&Ie1B~<<AOS{Ma@+qSbnNI+G_k&_@a$cJC6Y9o6)Dj?SCt
za6@GTsy57#h`kP-FnJ8myWYR~4p|>ep&#I+fzkDsgkQl*P;wJ)n~141<B1^3iLQc6
zF9|YR)LM@u{5;8SPObP9n?upJGR(wr)nhUbR_Q@mqvt!bU=f$auEy(qmpQ+48c4rc
zOJ>-TKC|}?v~r*lYp=`|)82bwIY~B6^$mYNgKL$~kfS)WLDU8Fwk6-BQ?h$UJoYBX
z=-1<V&NTd>#zh3_CovN_te^I$=}jv!{&DVfeC@D|4QZ_hwdOG3QcMcjET(wZYm2#i
zY}dI5K3NIrQj<MY6@A*Xg3Xm0)YLb|dvHjWv<VDmwwMvAeblC!bwH0`6pfT=7N$}<
z*P^&MZm-1SE$>4V+;Mq@&!yffQjj62fqWAy#!YBmx3OjboM6UPPoIF5tr^U4Le+-}
zC)gKW70h?5(OK*?2sATZVCkp8Qe4fg(3k6;D5iI%0{3z6AOTUer)G`h9(U=^=N?Hq
z@aU6m8d|`C2SD$h`6$+mVAP<b@pLBuaylijbPfmtm+3x~yJKPhx@+9Mue^7t%x=nh
z02lqP_H%7H{S{~@kVsUNIi(i+-X@G?oiKAzI(a)$+QH}z1fiEK3oB<oC`#YrpmyE(
zp2o$?YT8jqaA+RH6%vMqrtcdxIijp%2Iu>pXI&!8lBK)IHI%mrRRc()HhE?H|A3Uw
ze@XzDkZ=Hmr?nsXg}l1p^QXTck^k{V`o9XUF}ucIaD1x<L~{Xf@k4s8uea?Y$IiYS
zcUKl@W-<S(<iBf3b2@_mtK`2cP>ub+-`F64tFh)kInBN}0SD1M*NuEH#e8-StaBe?
znmv|2Kd3!>@iNnZZotA&_JPQqe|O;jH(%jjqws$*Uh5BUmK&Bkd|*InROn0J1j!Md
zr-dO>>WuDluv-|N#mt5qY(sCtp>I4@x$qW6m2-NUN%u#0RzZ^EQxnM^!r%sG|BF$C
zKODDNM|=KND_$17S$`(adk&$-8t$$C#Hq%s-Ew;T@G<xUp)2i*qJ$IlqXzZ*riof`
zWut{Ayi}N5sNC%cKj9gl*rsoTg)s;fk@j^M(1%so+HqhxoU_MbHk7KD63GM7F-0L*
zpW&#yFCIzNEty9*bFBfCyG9lb3QXQd^-)i5jd_q)xicD6q%ew(Fb6TIR-bhM2I{E-
zuB>_<s^B>8AJ~Kk9ZpwDwdLlbWZN7giRTNQ-G>d<zWqT;CPE2*xT|PzhQ}HQspqd$
zO-DZKf@htCE~%)7OovDvO{a}>i3elYvNB?T?A7H)B`oM2fxLjJ5}F_gHHWGxNp{b)
zm5lpPD@so&AeAZU3oI4%;gy)sIT&3EQa;5!hT6B93J4u|M#T1-R)RUTF#b3i3JlT{
zR0*a_xlb=2*6?!2LWq)FL6_Y_D<o3n6KKgwsAsy8P7c!M3gRa8%EjNAQhR+Vppi&#
z_dF$4@8X8pWF1LCxg(?^BjwzQABg>S$2&Lb%_sS(-$=+Po)kbs3vuK*JRpBRbRY!8
z24R%F+(yv|QV&IcPLtp&feih|xfQqzgj6f6Vx3GFd_D*)KP{gFLCKuee~3nL%!S$a
zsAkv?&GVz+9zJ;aT~r2onvqeyw6eOl#=c-aX!axX1GG}1ssXB9gJ;MPA!3m2x;l;T
zFsEj8C1#5o#mcTVBt`M=>vmmTW$G@*_<4V<M3r9F7;0M0jjz7u<8GILxNE;(FPx@B
z3|GNE%<B47T6MKr!*`k)wMbyhbg0%5Qj&D)KE@>^y63Bwb2pX4lx~WW=M#jLu6s2G
zO=kMfj|D`VzZEL%bR=aTRP9uepFG;!Qb6Eur8qYT9^fsU#p*3ycp7RtCdG41@uksp
zj)bMHKiuZJ;OW`6F9&H%u45;-n{`xr)nfE38)X3tr6pCjLn67yEwb~;N0PpzLXZh)
zX_8g4RO(7E|1U;p>Pqvs{GI*cP(?c>)$V$jLdAr8Z=WVC*hR!LWBk4GXl_+wFSXC@
zgXKq%DBRS$fhuz4+Y;ae2@Q<V6^!TH74DDy$IQD@u`?f8qc5h#e2!$(sqqFi@?r=8
zi=bF9`mFd9?>OY4bglhWHbCUDbohJ?zyg|}R4cnQk(iO~|4AF>sqG;YWVIlEp=J1|
zDoXjkC6JkDmif=(e6_ZdE<y&T8`Ex`A<|6ik$v+RZu|{40C|s8VAVl0j2oVkY3-D4
z+C41InR)*piZTSetX&JzD3#dIfyTXi&h!8on@OZ)DJL;SOymmHeNgaA{6WWIia06<
z3<F|IbJ>|RC5;6itCPalG8xc*5M_FhLTSzvmQvhUf-)9Lq2m&Tr;OE3En{B#;xXY}
zEnF3W(#M2xgAL>c72$|b9aXzLhJ-50$+Hi0b;a)4Tua*cJ~L>X1b9*cL|wnIWj&L$
z-*jzLo6AP+>^WPrOz&Ys;d++MusZK4R<D_C`)2VZS7;>^98Z&Z__X5sq4!^f0=^81
zfqDsfMj0v)AA}i9OC=Kk_av)`iOH_TlffvXRVr#Ipl^+}uiKgCFc_1qC6KR`W`R|y
z33;hE)%Utx>?Vw_Q1X%#ENyAT@I)hL<PLQ~LN0z2&a&jC?I{s~4TCah@r0@3bU7(%
zdo8<2OYOG-of-xDV>`4CYpImzHxHlYK}9sTQHof`bfugc27wZ>V+4tMqRaYC{4NFQ
zrcCp2^CX@<EF)MG^%Nfb)yB5f+PL#m(AJp%_>J`ft&-oznVY`?m^Bi;APAVX7sOo&
zrFDFtf1g0Xt_q%{#nCPM&==Z^MPpxJs$@;{H2b)swOQzrRL2xUCx5~5ObvSUpzp2s
zE}L;qXFhLRu=Qg3^8Tt4W#$0B&#tch^r7<4oJ@m&cMWsUWR*uZO^Qs4>nWNJ6wLzq
zIkMG+28Bv;_HAZ-2jLTpx8$cK8fz?9^D|4l9`1orq>pvsaX3=*sQJ<_DM`{f-pAMY
zVOD1Vmn;!pP5#OnJARnmDFm=$DMZLiG9=_Z6|zlfzymJP)h<GrMGQ-p?2IH#!Z&I@
zm|VV4;iOF_T#@Gxr;l@w>Zr22$|M<91Z8+0TV5}BSk<n&_1%_F!X;jOu1jc?)BiN(
zjqi7S={g50Ln;`)gw|ebNKm0iLbC6wBd0){$j?7lwsoYpuXV<owILw#R9bdAaQ)h2
zZ(-m^%;2Ka8mfUk*;jBWR|Jc%5#@=Ef=3I_UWYshgs}A?{5N2C43^yE-1~ReO1f$?
z-{(Ua==zNj<8p?J;O7TbskPrdZa1;K@J{^r<kNwo{OqUiL945z%$3(Vh4oeTNl%y4
z|Ma!LG9z5l<T9r1JLI(>6+V<+@%|U5q0R$(->}Z&R>|FeW5hUyuU19Tzr+L3pbcQ7
zGfw>tAk6+1()=HP$hyz;{sCpAR-XrGjdt?1H4kegWjMT8sNP1T1)25%gsDtoFQcsA
z0Y>kDBAMdi<RZ6<Y@wGnzoUVEr5LMJdS(uiSJ*M5xAow)oTW~Qm{bC7YR93>Zj&8X
zK`mZ!WlYmJy_S&thW0yB8t<%mJoT~;qsLlpurdth7N*#x2kCp-{))-cqth_N(1Du%
z0fi>~iu%(U<@9UJ;^vbvFF+J7jr2o?KZ05S{<&NT`!k-`oH{=-`f`KTr>AA&yi+Wl
z<u{I|OxSw=G{8*APMQ(=SqCcw8$UB}0!=lTBCT~ie$h}5<302we$@$tbEhtU_OMSv
zOYK9RP42NeClESU5ScO&EU(H~i%u{zM)gwB6_ZMXD=+M1`Rzaqz%Fgf;A~UlByzp-
zaj=6)@==o3-=jxWF8u{xO4KpmDfeB@T^<LQKKuk2_<bpneaS19e-WX)u2YJPoKaqf
z<9pcwsI=ylecU{p=Cto@%lOLrkRfoe@n7!~{CWKV5Pxav$z=Gi|7@;GunOIB(fn5J
zewmC)Ch#J*{9l(g*oU;f0qysxj}`rG#c=MwKFSsT>j)^R2d%!aQ(o8Jr4E1lx1XgC
zKm0w1v%Adfx@`)3Y};|F?es7xtFs<95Q01hHhtP=*Ad|bT6W`U-#dodw=F{bl4h*6
zdc4e6(_f(-Q!Zuq0d62S0;bb$wC;ABj4&^1l~B8*j?>qTzvJQUl<99G)dF6_`-$S)
zuvw1PnH^`E;3fV>_xZ3F5;-TPwo@dN0-M`nuezK$K3$%j@o5KqcF_)Hjg|;EA@%^7
zze8lAaI@?%^<*Ylk9xKgKx_m3H7tVw+A*moQ)0^}laYV#Ktj0TS?j&g8k58_<+Fo(
zqv<#9oNRn4A@AV+_8((6^E62VKbY^gu$8fVsCf2C3+R`bDv-1OQgW+tgwHzs#@ypj
z-iT@b8~6Tx&ULZMWb68zXg_cxfV-Rix9()jGW5CZ+z9*_dW{@B6=5z*<i_2_0<wgy
zpjQW#!hq&ZgvDeOTRsj)#cOT?qF?1{xwM1<O|Wsab1k>b=^Cv?V;W3sllEq!WDMnW
z-mP_AXa1e~-QAq{>lm*v@jWE@=4|vW?R&ajzXeiaGmTO?e)v?pj_VhSElucngfc2K
z%_-(gZ*IV+E{z;~2*MT*xFC)ih*O;0&tbPw-FQ3EOD{8L(sDPyZUyy?1&2>>{u<h@
z`L>+bg`oSIQ{Ha@$jUL)62m*{tRD^>$WF_2-V$o2X*5mk^)GRm-ZvA+Fs!Pe^05rF
zhuqNpZrn(^#+1WPPnk<7&5J!e4TdR;F8j4E7~@Gd5hx){odAVn7>$!+C!_PI5XC41
z4kc;9fx5ZkMOaD-vr1!m=Lg}0Jl#q5ZTPEwM>sK^G8FgA>j`gFPVu<FLa#uT+yl#T
z*8$w=l<4=6(A5*EbmL+khNvktL!K~pXb-^cC%Ie?g3nyn8~jn30XNL+FO`*iWxO?C
z$^5=i=&29d>-|@*+NW(yE?l*)sX4q9KWqTaT(P*DVl+Z&IfIO!pi7=0v6MY+CR760
zc=kW1(IED7sl<=qoD+j^{E^beU9ivj%vqf$F8>dnK88^Wf~B*H$*`O$?W|K=%v^0Z
z@1ER3yx|N2ZxFKRC^H5ox(-~OY0tdUbUz;T)dCunQg8oUd|5I4_lb_fSVyFfSJ?sE
zOMw2ca{>LIPG}T%`9JeipHn^sEb{&U!U6w2AshLV`0uLzr+T%6ZhSp&u8SNu{sE9Y
z+J8^{`(&K=|4E~;qd#LOf66W{cqj^wncEj5$L9Y59siulN50kTyzU1+O|p%&11|CZ
zyNUl7?frl9CI9!0W|)Np1iyPKBU+f_#LfFcOAQlqQ^_1Z@|mP0XZI8RViLroi1+1>
zYsJEF{97>o*%z@YQRYb1*A)Dp*0uHlD4g-s!izZO0Oi$muLn0^<Q*Tnj}?WKxrh|H
zPp_m{iVI|Z1V<-a81maa4qr~DUwgE;w63w%LQv6^fXL^p%kv~;bYm^FsnydYPA6IF
z{2A2&>VfR!o{kN}M7H)N?saNH&|EYR5B)zN!fHg2BjlvET#jb=X37h%q^6%ibA%-0
zCMbPRmaK*)1MMb~!bkvRS^TXwK4HOxm-UA`1;@~qUCVr@zQ{ESg_gzaY&dzw<Tb`o
z^y~mGl>G)tb~%cL!VpU9u)hQM<cK5m%EJ9OjW8gx9he~gn&5h)P~5J>JG`8~Cs<Hc
zQw7gE&BAwr3&G*l09|z_W(`ygXdN7Nga?IHtz%8Vg=&bw<FNYB(>9WL;y^>ONu~VR
zW3i5ul&4<buS(JU=!2O-D)IwAo{7q?L=z1brIV}kOgQ=Fg|d4EyTqt0vwMPEzXNQ>
zrW4Gs?T<bx*&MeHuhiZ+sOGnOh~H<?Xjm7?w#d%5%qY4n3|{oglFMc%aE}?c2y|Db
zjLV^`!L)it*a}#qFH4a4R}(*#U2mjyLKPFhQsS@Pw1hX5u9^9RTjK<wUx~65CYqjH
zLB^hB5Kk_8f1I|708XdOSk?qsrUSwcAS_}n3LH|KF%I=)L#WhU`@m~1H{fMHZyB`$
z>9>2xuDpU^e5roo40Fp!v2pwQjPu^zdx0n4>B>`fDl=WRH*{zwD?4ALZ3**YXobr2
zIRheI#9oW*v(0JMKiav)Ub1yrQ;IV!UQlr8oa#>e{vv)_{<}#glI6;_jj117r=sB4
z6^21@dHt-3VsN<ctyzlDdO(azX!Y=^k7eVf`)d9ZuYSyyI0hkqaBbg4X$lSta0p~W
zlU#@9QKWI#LGW^d7sEiT>y)bsIMoTm6u4>*DWYA6b`xs24w0?P3fi=|+M{~<2o$xk
zB#cOnnr-1X?%un$OJaeX&itJJidD@pg07Sv&ZGc9J<~c7HRQSgk_-|trLE4jjDK>m
z_qert``YJYNc3D4<zpaMxBILd=%DQqKVk-PI_|IU13I1TnC#?LONsjCenwPV+i;Lz
zgJ0RhJ7P?znzaH<pr7qxd(hbXT=m|x>TzDoZMKd$Ul6h*p0;z1K<PrB9LwW+7w#_p
zyLLWCF0}mp;c-3YYYvMinO?+X?7J#Z^%{6n4P`(&Qn=?zcZJ#RTZny}6+rh<yzb)B
zTzrp1dkAtS@#mrM8OV}_e~y*;D}OF5Je+sj$6^#FgCG9tMlU%EI6?eRd)fb^Q}W#+
zN&T_6Z{+M{w>WDI*2yy)9<v)<8b)h>7Fv5Mlt598aq_nVjh04!(VA?vb>YTawzHgi
zG#a%=sV+xRaj!%9{m=}`RNO9@>Ko;hjK{H$7zG4P?6~dyMjlMBIt_N3!qlHshrAn$
z80;uCFUG>aJcwp!Q4J3`$P}C)Lq+9zi4It%J4fRxp`cYz=8oB8Q*;*k_rtZF=!2&j
z0a!|9C=F;zmw%0nG__I^PLz|o`t^PQ0=;QVo7&GO`RaI$L$nLUqq@C)+!d%?@+Mb1
z#WF<~O|~4xFzNAHs6h<<uFD*H4T=SE=0qDkk;9<3b-6m0EU?ZDd)hc@iK{uBc=M8d
z9Vmz?j35@p!b4*RvNLv#=LUneWEq9x`k2A5(NXRh7-^~KB@~3gUo1yigY<l=>1Di(
z7x#UH!f)g{-{#A7AVU0k-^k}O1)L0b1x#TGni5-y22YI{yEea?%$%%=4uHdC8TNG6
zG8l;*`b~|#%nfhj6+2cs>mtUb^z>+~0pE(LgE~#`%bi8@5D6BtzY~#(s4PA3d!?60
zW0up<qy#av1WN}Lr|jg(^%b00!TuZ7OL(`*&i}q7%IMm+Clo>)V+c|>c{Z5>HiBK7
zNG84;^#fDWe=-hgB&uk6<xl__kCIo@(W)JD1vFPl8IA*4G&b*uVzk?>_jnxV1XABP
zX?oW*?_o!}5d}1_^{(;pURIimQ$_N@H@vYQAF7xaCBp`GE6OY3Ek!4l&JD6K+xXYp
zVrlWs5#sd%e@OD<%SDN=^(||k>CstlXBOLHLLUxTu!@J+vN<_F@3pq#xmo_=YIe-q
zKv-63BQc;ve`wzF2`MV~4&RpRy$D*@=+_YL4&3MmVNEa9zfcz#+ZQ<kBT>G;)C=B?
zqv0Tmjr>89=`u|1Z6dgZvKOmSDJc0GBk2kW9=%x`n%YIh>1{a;56oieI(Q(F67@^Y
z5&jPIs<ar@jH%F#j@A)kFGYW7Gw&U<%VRIRysce0ZO=ddtf{u$KR}13<afm`>~)BA
zxQ31Fam%i{+uh%j*FYq{#H#`2ZR$-bg>Tgi^2UQB2uG10+t-ni=r8P<syJ2sN8aYz
zS`X1SJv>p{GHCj_?CVdSwT$4A?Mc^P8tGoo*BM;C`E`2VPKk$f9B$)d#npQAez3K?
zRnsGwTmQ3#&z<LA?{1h#rO9?SqnCTS_kSNZ*86&m`pXHdC1(!}@4>TVCA%?MNF^-Z
z+suW+Dkzp1h<$cSP+f<X_AM~PtRnA;wFv0Wk26CD^56GcSO)%kWBb2tasT@f$1YxC
zVkxtD(l9{KuQipkcUFdA?V!x@*;D#BwBZ59X;k^cKcHVs;ul!{O*_q#+RngvSt>!1
z(g<@Ans)GIRs-8&6ofk;^{PAGPqY3i%xV3;9n99rxzQeL`Q1e@K^srU>Sd!ESf_;T
z&Cd!VQn)%oy6@vfY%vTe&E3_*JWD>f$sT(~#aMdR=8oz(u^1Su95fz60cu_0qu}Qa
zTg4Vy7_Ai~zo2JBqGIiaMM^I5P|6%paN|*>tvxEu-~(siL60;D+crAHc_!SqoeJ1c
z`=}ZD=qOS0ZD@3k0^Af$i`d5)Q^(_XdWGT!aUsi5yaq63FawneJ%XYN-2n!@meZ*s
z!^mN%X&SlKGUBDMz7!m0U^Yepc5;l9O2AA`5F(RNY>qm9IUfj00~{E<R00}?1pP1u
z;i#%s6e-wwCnxUCqhLyoy_+|?Yx%mdWKgVVnmEP}9p4LX(|_VQ>d8{VC;=TO!}mjN
zDTSg`s^Hcyl8T29z%wft()2s13%13ci>LkUaTx)rm@V&Z5n%F;h9F4Knhq2HS<6Y6
ze%5ktsL+U+^FHH0AnBrCwU<ph@11zS3cmNJihT#9ulQ3ngVAG)0#>&YrLIB^e@O7h
zYk0=KDm;VyN`we)jHkc){x|{eC^Q=xc5ZUR;#c{0m0kLYM_HqWtzGRhRkf>)q=>zz
zAw1_Xc#nq(qa{iHe&=dT3Yj*z%^vL@&){iK)27Fn@z&#p$LWVtq&L#HHdllBo}eL@
z@L^SLjYt#!x0)mzh!+*Tmcd3{?bjdSsdwLXA1P+ET)>l8`^!%;#IL|fqPE#!G_{fF
zrgqlv7ta06=y_M#(S(}=%ZE(l_(XpF@6ao~^14b!JBuABJL}EG{gRnKZb5@}vOZ`!
zra3KaE2U0d6DEf&NrezNmv^HHqV6iun7dqSF;Rd(5PS{ql4t7W7NHFn=`~I>ivSf~
zO%fRcC&6Jp43?L#{~EWG$o^r>PaT~t5>;%KBY>fXLyX5MxGa+pu63QYD~KsWO#`c$
z+5YMqYNb-K!q>C69fK$O#RE};)ghbq1<zqz)gD+pyl^dJ9m*yqXKDw!-6kT=c73Lx
zaK`@n?&YhwvRpKq@S>|O%RrXR^Fnb%DG+@H#tKbTAuwVYX%hR51B%m>Fc}^F`H@_`
z3e^$LEX^z6r4ck(h25?TQ9Srv@=2D+t4={$qHF&8jaDXQmKQL1zP!G{OdkY1lkg({
z>4M1C%R!E*{%7}29IVRw)Hr^K?oSlhDAunJ-Dza-^>jt9C>CV+fTR6MnZ-vgHT5e0
zva!Sb2dnM>;17NK+jqZ70eisA`}<}<yMRr=(q>LeZ)rsVe)OL1)YHl)*DQrAbhMHm
zD1+;FJmlLKYVGRFhmD}*i%!)LtY>`EDi0asLkW@COj~U3*X(Z&Q4-gn?jjA<f_09=
z@<)vD2~FCE6<3O_X8jmjJ-l6@WX8Ab4BcI_ym;L+rIkDyHIdQR207d`T!iqfs!LO_
zbk1`?Ip$fp;zPU=-MzA}&=uCPONJ0uXCHqx`qU{cyO9Y7UgRmD#s^P`x#c%drjD7W
zE=W8i&+s>qZQ9#Mjo%Glj%JWn4$$#oSc@*`D%nu4ux_{uCkIS3reY2JA&m4)Xl6Tl
z7Oc4nk(^=4VkYW}@fD0tfDuv5Q%`3x=b%b10IQ(Q2zAf6gr%|5hAd~R;wS-oBRN?p
z&Kv}YTd`QAp&1ZCMi@%2wPMLA385&4<)T3x;4utNxNZz#TAsAb4Inv3J21+qR0+y$
z-b_KUq@bx}dBLSoQ?J^;r$o|&tYUx_K%Uu&WwM)pgg)tezq<yy-52<w19Fp-e(67>
ze@_`D=cYUI1Wl8y-=yNzxRv3|leLD(%F98v=;Q^W`X42k?<S6^4Xyw_yz`|<Ojr$d
zk(buDjG6!gU#WCX>+MNFQx@H~zlf2N(wus60qIM3hDxH;K#&xRn{dj(O4Taz$;Z$h
zE&}c@AYBV#fG~n=6MhWGl_N^~a3ZCT(9zbvCj49kQe-#0$!%TRa{z&E3laT{@{X!(
zB%A{4fCYt~D!WHBTMYoT5Lz(9^qqx1vV$7em%l!zm{S#bFWI`<gfG3eg(iIZT(bft
z`)tK$XxvT`mJ&)`NOhgSy9Qv>QUj_6i#bhutilNV#)Fj=@-n?jK-5NDybYGhF2c5I
zZkMSWLsq>@6vW6%THS@y#3CV3hU^gmbAD(PLpOb+bSLz?e05pp&~h=ernenwTJ;t?
zU6A0XgL*na@@8qkSuqHC59b`4ZECd^C4Sp$GPqGFaN3RWUJ-v^M`6BCoDG7>9U9cD
z!6t?4&0>O4U}&9P0SVJ8fnz+7Z(1wkeAs(BE*A2_^@-&Js}xd`eW}Ze5bQJSwnGaW
zjCzn&c6~9MIy00B#!VDxx?fFBMdk#gP;sG`RZeFTio~X!BCr*<!?nt)b*83aI8RR(
zVBioR%Qcfe2j{-+SgTVfH%ReetX3S`reVDf$b=R8xYiz5*VZV2GZFj|nF_ycx0<VU
z^La`d*ck9FLb5cxY-Dw+tn<G=zf*f$Slu5|*|%kVm$%NL{q|p;#`eXf$>1^?&DDUH
zcY43k1PS)7xsCCu@f2Y@t{I<>#mOl08Ta*@H8nToHa>iB?Jjs4`1p2PQq6C1%c<Z_
z#mjk<Va~qKpNK8GXwqu9q@7K5$}k6Br~4i7>SkS$drmK<_nTXwRc0OQHprP!V-VVf
zuuupFW0pou*LXed5y*{1L)G{b=SG_K?d&DQOD^^MO8;^5*SyTZm>7NNy?ryeWM*dA
zS1`>YK<s^7h<J+2Ud7je;nT>dr@m)KZXR{RzDMEtBTQLm&!f*I0|qa0CyC{@d=XnS
zpA9C~F|>5FEx94b{$y*p!{#s6uT=h4HYYhHT7<T~Xv;EZaTsnI9D}Wt8PDw5#9_C4
zo6%jO#pdbH8?s!!!B)Q|RR>z~rG%d`*L}F}K|bg$e_Iv4E11*zAkjIX-^{pl{p(tb
zFXPn74Dxl|ZkELQaP)Rcq=rR6p#1Cfm^Src+GlRq(LV0aK3d*ocQcN<-&Na&U9o-}
zM_jYBH*l3R_{i?NZ_9S$w(2JALuLjutyj#Hc}LMe>>te+MBc2sEh&4mKOcErXWI0?
zLk!VeTk&?IFcGSK_b}Muh)DkJBhvo!CG~X1{)SuKzJ=nlM9yDBVFSV%)2ZTyX^{-W
zH8R;Ul|e`4Y$6Xv+5L<JO#v4Oub^X<Kl%l+1Kj<&*}wh-5e*Sh|K9saE=JO<SX8$O
zhxyo=PEY){b5A>Cd2mn#5!Rj4$8H_LGSrG<6HpnCt?C(pS>d{F-I<T4OL0gw*M4@Z
z6-(SN@S)BLis${Jxn6hT1%-}2n<uBtTg4YQVmWAkPCxmR$)RN@ZKJEI0rBbo7FbnB
zT-x~(7<7Vcx`S-_E&ro#pn7)t7T~JX(`m(cK)Cw?r6yL<N&Tq>!Q*yXLiAZ;_Nu3o
zH_z^QySKakvnokc{6h1KmN7ol2CVU|BQE`Y_QOx_xcb}VuXJ=~`~%{Y7-nk3QUXSe
zJn?CauEPjF2r1z@xPAsmi5PI5VyC$Jq2sopVapl3S3SvAvp(=iTqCjTh#*4StVuEt
z3vL{Xx5*W+#<k>$H_8?x0WbpPMJuLh$YD=gP;El{eTj~MDMalMN2xK1yp)WQ(9<Du
z381)K#{q69#vxj}_wV$t6m_|l0OWZ#G3F#%2#kV48|Tc4?-c|c5qiy;z~+K-M6R_c
zo!HWxrG^LmLz%iyO44b+<Z1v^DT6P$WbT^5e=1a-O!eglN5+C1hFY#dle5n!4izOg
zS|ZsSZ&>p-WCrUk8fSPmALo?4FOxVie3}m0HPJvTRn}$yo_fwN$qU`Km{g1S^!eG@
z^E*Rz1EKoK_vjB&4ZoVrRaT@A70?6t!yZ67Q_haZV{b~&K7XT5df+Xv-^@O!Tae^t
zpsN}e09Qm^CI~`7?m0Y8*T7K{P+C8mKOgohM|94-PsS<Tw^+h64OdK*XNG6oD!vBG
zZiRNQj1X@j`6!s;Ec{>hRzxSx)GvG@kX%_4_!t>XxD<4Msv2?$RWyv8CK`u6@&pmd
zquHtHA$zJas<=xw_g}@|F|JD)4Tmzk^=T+jY_7hencEX2d^Z}$Y(5dP5yF*wD=M3#
z*}U(`w*=L3Jcf(kZvBv-L2B*hZDz+dh8cNTLThKMsB3D1`lJZJa5LDfLOhSaM!fLi
zt`MOn0Iam=QZLwA#-KKN$xCpQ-!D=;fCYoTf$8@Zur?ilyZfO^i!1t17Te^DC%J{*
zNV)~-aZ7gYpsGyB-3ltix$6a7Fmtrab-P0m4UgTjAJ{Ml<Kg{_T@_BcI8r6Z`bh%C
z>{QDbAi|aiCDRdzKpEOm0?0u9$04Uv)FF8u$CnOj$3Wvf=}LQPoQL@{o#NZb;0lGA
z>`bG4|1oeOo_sk!N<zXEK*5MmxHsB0Bg!l+;rZA0J?3!kP=}84PU4lio6XJxK!mh<
zroF{Z-)+{mK<TI<^&Yiif1`3U+w1^r1se_uZWu`jl;A@Pxxb!KYuL;ZQES9Mm$fvS
zC`@<aa&K5Qtm$7qXL5`*PuMzAZpswSQWjY1*wV^%*;<4&k<R#QdOo6$YM)W`vyO(+
zI6RK2>Kirg>-t@-cUAOx=f!05U|LV#*eC^{sJ7my*r=*;<=Lm#bC;5fs>3qYWoR8(
zPb{1Q4LqXH3~3VE&^$XvFFH)J<Wq7F6h+rEai8YLBE6R{g|DQ(8t|QFI$BpZ^KzAf
z24`(u@RJ_uREMP=#>IYb`nq$A>6bwm-^*$BF);e$1;-4@LE7f>{VNr5KYIhUa0KV#
zOs`1Rjh&#cdG^1D=C1!3(|UQLy;el{oeYodQ(IWgpJ=$uN(N3;3EF`rjy$zaM@=X`
zW7t&ss8%s+QVl0ekV=*!oZH4yg5j_1)h~4{&a4dz?H^^;4cW(Ns#bL4+a=nQuHoKJ
ze$BR;a&5z`4!{VVsV*R1_LEcO?C*4VdQYiv=Y!upHgHOrpv04aUQQ-g%jtViEsjkd
zui*y6YN9*65pRvDA-6EU@oSG|UwL2Z(YY(<Hx}1=UqIfp%h2+12aJzP)z7aT232_n
zd0|TyTX+`wK0Cp`3YdbpZcH0~P(GTOy!~lmNHF1ap?n`cVDR0nO&<Ds_?l?38snP;
zMcPaA4ks-moe-8j8qKZL7lrhrS^bT6hTldEmPdeK_1#{wx8{2{Dt+s`rxxsDzbS#7
zWgVhpzNETNiEh2e0G5*ft8R+rn+xW?iey!eYel^e2=R*b<!0Z!AY&*?$mLf&MXf>o
zSA!x29WNPFE@sjBL8-<rog(AkRf$MwOud#vl=m6*Nl<2l;RQpL36k$05vkb189%4L
zFBO!OH{ld*ZfLF@rmAtv12He0)iQD*W|n^8E;w%AOb*EYCUzrd!r)QwIo9LY=)_v|
z8%bVycU<B|&VZyV$_W2Sm$PPQb&U|iBEzVLZ<fRO-^igIZ*nM}$(uO9wF<K^0}?1V
zQG!{1!p}I^+sK7VEx6_%Ns;evR_8RINyb@2Cj*d3_F^PE+}T3g)*qbMNWt>glayA0
z$cKJ{2>J^NfN~KifEc+;qZk_Gid7mXGbK_m04YvEJ><CrOG&M6>8o)$-K+>6{}|M<
z#al`8o@BKDg3!GoJkO&EV3F?n2t8Xb_vMo^;_(JQ%9Q8YAa5>zsycISmQ0d(IH>lr
z?qL}K5$45L!k&a=V8bSM5*_>ZfH!I~gX=BML@uCv)pNjHO|uQlAerxZ`>E-^%~gT1
zY+$tLrL7Wo%+=^`Yf3cz$NeIo(C+OO#OVUS3p@@WvC!SLl<mDUMROg1xCEYazzWr(
z`O#X&^_g3Mx?<71iwVzR$k~=ch@aX}<sKoxk-__q8_?yt#|ee!^<HoL%*HK(#m1)^
z=8ong>5~p?3>S1g*DlL*&28|~DgZo^Dn38GV`qvz!idb>)CM3^Mr-jKb;YJjt4E6r
zHPSmWFj^TL`K_Ga`0nctP4~#r<YuM5q9pr<lt>;IIjx?cL!2(d>4Y}(6Te);j;FN-
z(-h-Y`jZk=9PTkZ4m_Zk-3h-+fuA4BpGJh;rZZM~$HDf$aQEinY`y>5Xoz`Mp`p}R
z5>pLPQxOzF5HZx$)-0ja)N1wHqH2nHCWfZOEG<P#ORFm8pfR?FwkTCaRnY-G%Wv=f
z?%(&f_j_IEyw`c(b6w{Tmt?IxYdv{Zp6UMF_x&*qy?~-B`<ov>Yg>!~wh_cr(nOHy
z<;EHhl26~q1+w$w)+37rnz}vi)cMrMnUfv5#uMe^fyWikcB+`-RWw3OEKOw1WK=Cw
zW&GdSI!6$h>pPnR8h+b1yX2<b1|nI+AtA>B0=Y?T(n?I;JkPTgtF0kN)C*ACKsQCh
z7d%Rv<I#^8bw))+j3J`L69Co0W6kmj0u-+=FLqZ04B4_t&qlg6)MVR{6;WhcUoFNR
z4Nn!)n5?VN=n{779REwr>Nl9Pm#bPINMrAs>7Kq%d0l80q)Y$4N8#tfsNuUg$SBex
z<#Y`hMGv!R?iOAl0N^8NugXKcrs)p|^TMLh_}JO0o`eI<tq0cD{yaJ7Ou1o8SS1#{
z)eblp1EQmvfuRBcDt}nv6una`AgmuojL(L%qpojaaW>vj^s%)I+;~1zBPU<{qn&F9
zZ?o5|z5s$`L&QQDVW{uo`0Ksu&*(!T9YRyef1$B4{`Y`D%`IT>2^hQ+)UkVj{~Ee|
zf~I3Awl8;3i$lNrYO!MM^#^1%^ONJ#<2#X<_&q^SFwq9^t1*5)V0bnJ<ssHr8J*gm
z^PA4_V_eN&sNt)q1{W%2t)a@~(lB5x%6?Y#fU0rI>7iWGL_f!8{W~?Z1vRr@UAJ?n
zo;lB@=e4-+g*gC$-`RVuRa<1Xq?Vc_g%mKxJWTqVTk%-iIg=#vnn|}751dW#l&nU)
zjvYH$k`>X8g|nE-n=r=Y2kP6U^*sFLM`W3Pm^SNMG`5D${Lm^~y|g4iFG#s%v?AxO
zk!OQoJkeh+ULuu|phfPDGzJoCdK6=?0;v`CaIzWESirxnkc{YtQFTGv$kA3BgJ>G$
z(4NbdiAFzT5O>yTR&)Q+m4RgDf)AB*Co{*B$rr>wfVZlA^&L@95i4~JSmpSYdeUu3
z{doY6Gn}`$KKkKM&L7VTm;bc;Ze1iZe9+;L>u3QdzRT&bdTc!T3gh>b+xA})ytT}r
zpJKmNbS;whYo?h+v13qMl!sX)6GVgwhQKhOaL}?8EC^WG+oBl>#$2z7ApBJ^6lCnI
zaXp}uWA}2|n2Y*Ydw5lNJn(#d&PH6g=5NQ*DB=B@)Ow`;C)s>89`AEkTYsOkSw!p1
z-~MumqU^pba<wKZPUzvPNc}fqHHa6Tm`AFpDZ5f-KqcUx76abE#txwmhHsv4E12hB
zKABtWlTo~BP~tn}Iv%vF=X4lexI#h2R>*1Zl9+p7J^1LraxJoI*nEZxg6(2+Aq^x}
zbJLDN2fG}T{U&wFHa&8+b^*`FQuA3;DNsabhjzwpI`8erAWt5$U+Z@ub-1!{?NnaU
z)swtjIwIgOMjVk9F7=*}A~Mu>4cIa>$tzjDV2t4!h7YW?fKQoT9$%D=$yd6>#h}qv
z`FX_icUf_IPl1?n$)ew}yT$Q#7fMRLoF8TPxEzZa6`-HYsn;!Zd7A#HBh(xG5<*l~
zlgd>o6!TBM(9hqw+>q7SF5?{kvj|k|&|I)cF*fbbBCn55BreI4eR7B4!8qtFpTGUj
zIFLnMcC3ZFcpS(?sajeAO^~&tsVKtaDy$1ZVW_g98xjtQ*@)N0bD8d3T-Yglp#-MV
z`)<P>{zA82{b-}`%gSVO*<bIqczuj2I@V~9dTU3&<8)e@co7Byw^sGs4%w+S^0V_g
zthk-)pDnYUE&}`QtkI`JGaOGGaRE4zoY<>^(gFy-7K!6xtN3UhD;oyeQv7{(-D}mn
zjPY*gMk-X&BTc4D2a}tBO`T(}OO%>uW46Afpy0G{YMsXGd8#psZ1-}ZNlnCj)#x6P
zC>zBo?lLcwdTQ$w5k>gq{HR`ycffzxX(Qe0a2Q}f-&2VC{r$xUv)!xd`#)X(fZkOA
z4w}O&4jEB?Kqlvr+!qJGE`2Tkt<!t>7Z91=1n6RM$2leb=S7u;3=c|xVMHH5zV=q5
zy3)U_iU0ie1R#=F9zxs{WB+y~@i&tO+aIs|0o~5uhqXGh{;d)QP4!VL1MdzGPXDkC
zex!1xe%)~wqf`bu8N+P2>w4wqUh?4&;EG9sfl8_4$eC`#kbUsvQ{MlV?)SIr{Bv8B
zeenKUwg0&(OuG91xz>Mch~&8c^(HmrUi>5(i)rEO&Wz=kk%}q3RhL`mDmwOq5Sg&w
zwC~RZB@d=|q1&i7tb=N2Zm<+M<(s}}YiXY0`gr_?7Nxb_SdDl7%mSQy!}_fCJ_txD
z5%!}Nqf$)7)yWVNy7RH%e%3Wi)a8q9+C}LU6?4Jm7bTClDCrbOb<Mqgx%udp`1<r`
zd2(B%;_b=G5AIy7kgglMRl%QBQs~St$*8X(n-)Bgqu5hORhlkNj~n$WNOvrR=qZZf
zH~$!d<k#`4IFkW#8z*pza<GQZvf{y%i{ct8UJYU9qhseD>!wVI8LbVTFkFsCzsekJ
zlvyrE5R3$@WUw%u*C`w<axO>Kzszgc8P}quyB^VUZDn1JWr#C}aAqwWe%^6kzRlvJ
z%mpFl&tLG1sFn@#A>yZT?lI*;+>396J2gTBbCKtoza$JKolP`DO?|8QHX3fSUk$^M
z!Wr`+ov<}zR|>s>S(q%rk{8XPiljm!Dj7LkKrogJ1_T($j^r*3u7Qjxdi|K%CNZgZ
zxBpE3IVQ3@do>5!>X!KV8bIhf%o(e5<+)2#jx9)8!|=ZMvI@i-#H;_HFUu8pOulP2
z-l{C|^`dJ)9*A~dsdFp6%iC41Qvwhw^IG@igI*P;1gynQ+SwK^pDh&ME?ame;+=8t
zdGA(vh@Ls(ShsB2c(){?GSkSNh+c?>I6hC*&i-2Knx;z}cZ^RksN?<q>U*sgKp0V*
z1$;d-2}Te)0mRHS%6QB~VmTVzU(pHa)W4I@7%d<=R^x^<o{Tj*VbY-#7TG4IaG!=3
zOHS+50W#a(cA9#Jn17Ym<I^&f*TiDi8v?9AS_a&DgxN+yR_%^?_9vBpt%E(h;U_lx
z*S`VAC+CgQR80J2(23gw+y0t!Jj3t^yQ4rg2MfAhh<wa9`jqpT(&o+Pz$J|2=xke#
zm-fSRUw!W@+9|53%I{;;J4oY%Wk+BfI{D~4Y=7ITEK{IZvc9d*RllXU{2E@!D9s^3
zZ_N}Q=?!?_ob$GB@};~T%v*khdqzwQ)dFUt<Iwe0o4eUW9|2_(%VNT1mOLsV6we?<
zu}&r$EA|vmR8Mi*r}%fUzP?~Py}&UvxxPOWxVWk00?zYzNi%dTRN|H+h5%CGgo`<V
z^G2SQR|TS2#8dPHVyS0qr&9K+<E77bCv$OO*zR=Qt*ZT7{PV3-(7qa-O#aHfH594$
zqj@nF#ENKL&XB|nGpojj+4ZQv6EC;$3zKP5h%~LrnEA}hjo!R(f%P0O+M%2OERULr
zUx1wIgIoBH#H`U?DDCV`M_-BH>tryuRS>YE)Bk(2h{vJ5KyX^HaPmJldJqh&UKGF+
z0!HHwhfj>IHa~tR9Bl2ck%6Sd{~82(e_QeosLyolu=U=GTdK?S$9vra5kT~JWUHFb
zzGE=0`?IRGFc%x}Rn2@uuKXB(Pvrcp3*?wfEp2|F3J$TPA$mPhi5_#)b_}<Dg!OQ<
z*78sW*_7;Y4C5k!trAt1;~7(VHmnoUN(IZlvFcn!a;OqaNyexVe=jovQUWV)V%GzM
zJ@Jozq38b|iU30z<w)e|kYQ$9sVk@$Uyo>o;`X-fN>6R!g$ySO5f5~IKg3;(N;!fH
z4hMyoBRbX7{He(e?diCF8t5u`xwB5dxfzqh>8DrcDpJ-0q>rKVBbgav;vgU_9n^!@
z!7wgoWyCFHsHN4zKzP2Z?gk<$b4Xsb>U|t%%He2>DB#i;Xg$umQ~1j^Tp3t3C40O{
zAGLFJxqEBeYX{Pn(UvJZ#fp2Ojy{~pZk9I?(T*)hZAlt2VJx`n`h1A>gJnUfuWHMQ
z$~9o)`|3t;*zd{xPi>lBR^7e-*fZ&#thZT2MF_~|6Oj;%Dh$GQm1!IKG8M>q!az``
zfdx$fQol!xnT7i)VJetwY4rGH26fftt(>y~M@5@um@CAyS+R5C#p3sL|MHZ{D`B!N
zCkTO(PAcXVFG(3x*$^ak7~c%qfUVJ^x%kaXo7phetFDr<$1+M`NE)}Grdv}a$DCAi
z8nxGb)rmMJpmVzHwNipo$}6fBK6IBJf7G=xez&aC-Y)1C#NJ;47H3;7fL}V*vAx-{
z`s;O%QMy;w=!$RBhog89irX-;tpdUpN`YSvF*J&=Vpm1-9LvHlDj|hD938n=FI-Hz
zE!mp%l+Q#(4?B6a^@%xTMXEv%e$40dyR8*DdYjPYQTEKf>D1@BiQPfkJ2g0RuiD{e
zaG5r-DMUwN5U-4+UVM^K?5K1?+L6_>oS1*jpF==&{Ob;h)1s|B5PS}M8uDp;k!T{&
zQ&`8=N@bKwr#RVBvD%-LO~PM!<;b5PZ%g{@P(X_t*#m{6TQsc4sulYeS-7bc4(Xd=
zcZ#6(k2Sv@i}Q_5H2U*pn~meEj#NvT?RO$InlgJ2l#Q{&%|=!#0$9xbXLe~YzR`C2
zx;B|{rA0|Cnx{ox0Lhi{evFd~<0AYt`x>{@hI6CAs`)*aPwMCSjC0l{CT}<YOal7Q
zu9T?`5wla6_<p?lO=Ljuy{`TM%U4f=oz@;d(LDW;Dy&<?wD?h)RHv1<0(4;Zv=~6)
z`3Llq8IXPw4L+!j8UXI(K#~Rk$Q7{l4gZbW<NbjS=lywfu<!IAP&WWo{JoIktG8yY
zQC(i7sF6RQL&K;;+4O%Y;d_#<_v%nU415UO@f+tK&=%rvpHEiV&D;?GTJT96P@V!>
ziAw-)x%}^+cTQ+d4+fN;y+~le_y@Fq2l%u4mm+0DdgPl0Q@5q)K2J00oASU?<jU9T
zs3+e?e*OX~A9x4>u99@>MPfL)HR<gd#{+hK*T@5fEB}jkO4ygxVFiNmK(318e?SYk
zf32O6$GY_FzqF{?zm#d>#h?7}uMPSyWlRcg{vUe?p(oWQF1NVO+)(u0c953u#k-0A
zj&WPZYA<g}l>kkCw9*U8JspPoOC~1OQdn!%NI<HoH}l6tru9d?ZathdBHFs7dd>(6
z020i{p%tOpB#;(WlFeuo;Nr0%Y09tBU}yygkClWB76fU#h=W#I!-3dF0!R-Z+5wNI
zBOyF^F5H?8gsYL4i(SJ@n`GR!!;+b1CR@9prg7p9L#mBn{xBRgH~V@r;}sKj2Lh4i
z0ECe^L^PUi2ciTD%_#bLbiTK9O;xgLkT)wevr6fXaS`EPAy{)u@a2N5D+?DcK7s+`
zaAwh9$fWoFY8B?O__uNKisQ12JazsCiT6m=Gft1iG{w8Jvv|BsueSE0`YTH!^dt}P
zVVuAyTys-GP<S%P2Kge_Mx`xY%yBhg9=vM%_<p#}uVrxyWgB&Y^$4X&UxQTDip(_2
z>>qSEdu!=((JDF*4!Juj!ICn19D>b8MI6gVfKK8n8NUPCSy07t29-<#LqLFmVg@j=
zAxhw?B=QI%qu>kOOXD2#X%{cPEpsF%5e7XO;A*^fMA)^D&{z#~5h>mU_C?~otM;As
z^UF3NY~;6aHtRdtQg@un2p}IZ7(*mItr^q|$nwPSb2Xr`H4_lv0s|;&1RElKo6~w4
zr!7`?&d){8&ti7jFJln)!hpJJHjjR^{<<J|wEPmY8IN?3#j{ueb0f{!af7Cu{kqrK
z2?)lZY<t*hvQ)E#twpgCoFeEI3Ny3vt(q8wbc9Q5jERKIw$U!l`ANHEycQJ{aT~g2
z(VcunJ!G1=;n7A|kl1RFAh7w~eh?A>jL`xFceB&Knw)usRuIYO*7#K8c59pD@}}$G
zghbx8HnVeaYzm`!Nmr=*&m%GR?-w2+(2=FuHRWqaq47n?f(f&fn!1exnW;Zo-eC#L
z=unfhWhFWREW1?vRIvP+E-^y^kEy5T%PF4kV`+%)?koPtTLI^gtzvi+RR&1$1amA!
zUriHZsZ+#Xgd1sbT{X}48rk=aqC;0&f;VnOj;r72-yuU%!2}~cJUao-3D8Tq;N5$b
zD*ZLKk6;X0-bBKk?1o@G@0ua5$%95eZ%C}piPZ)&n1?VP{M%BCru<bwT`WGoYNnX>
z!_GC+ATICI{{tS4OF9Yw&mI(qVfJ(+zh?7{?|ABB&N$N%Q_q^V!R{<l55-QR-Sw?N
zljYXwfSRALQa?YqisZ45W+`H4{kpwVc@S@RL>aMZKt*TO-@a3(imE!{bO>=GO@9t{
zcp)DgmwpN*C@rm8T+E2?__Yw@c5NOugN$$3ujWMZC`GHlS<IKCk3*~E(++S9Z&-@v
zqmPT>)d6jCf^N;)8k$)d`JR47YmFwsfQbE(s%9DQAQw_8ea2jPI!{BG<b?vfVVRV#
zlQfvYB1jlCr!x9jERxCpLRT7j$}WKI4zuAZ&E6vE3~tTQi9c6^=4Pcf^w~Rzmx4~p
zWY63h6dxu!r*DZ5lW0jhBn~F!5UQ*?$}`5y2oDk+!$-e|Awcg<Sl`AZ(N5-cV({uT
zMn-m08cZ<SH-`?r8v$5WXkiIn96F;4EHIivbeM^1oK_@;A;ygHh}f&4$b@)VOgHjc
zo$yYY&zEiigTl;<F(6el1gK=~(z*ZWbyO&YJB8ad8DQ4x?_|&RZ=7~9M#h%{wjIc5
zf?tay&q9hFgBB#5Wb_>(L{??F1rId=X;PH&F>PfJW_+*MxoDJ4PwKJ+6h6N-m8IDM
zzYM?ZSRpztoYuo^gAd<*6lB~RVrV}7DR2Y3Ao2Nj8JXt%U}56sAmpiy*-;$S?PcEa
zm;nIi<~$?M6@3gp&dvF~sw{{ktWynUtN@FB;@eNL8&9$I$dkBc%EfE3)~>sNS0X!A
zcs(vj^7FVEntz5{iqBARNK+2M7Lankb`bj6MCyUGGPh&fiG2NB#fmm7*A=OH*G5Ms
zBDI9QXg7_6*X8jELO-z#x#KIlWS`^85>G||D3OiG#~TX|_?Nvz>qcC1!-&|_jAN$-
z<%zMSWv93;U0)Ut!5+j2KW*3dK6_QL|EWqwJoIH{bV|*fIib1FqZGDoFL9RcOXTBe
z-hp}?ZFgL5&>sEbXu(=&@rqP!)+J>_$+QKdu+p3Y?HUBijRU_%H0HY(3BiExY+GEf
zWK3BuZugBze1jMYLy_YbGQ1F6O`b>Kwd>EiB$R?K03u;txSZ^qjE5tluU4z#cnz^L
zi?n?>rH{Dh97Jfk`+pa_F0Tns-#W~75V^krxctbTU&Z!>nrhP|&jkJOFr=qNt<)SE
zwG4QNP;G5O?((Tl=5Wtd3-<i<zRFnKSFgWm;jSx;WEHhLTP*&R{TCKGmFWFf`8U!l
zacxW#hp~uK*~=5GA8j9(T%IlH!C#hAuj+!PMvp~3THnqhV<be}WOl0mfF5~at+&%H
zKxXtD9%#BnYIys~X|$JhE?h4dZkj8n<&I;hvMdf~%)uX*@v#>%=36}<9aVFH7eKd<
z(1MUVh;9XZzOmNbg|gp*J=#eLA50QLD(M~oDnIkBGTAx0KlOz?^sdUInp4(#R7_j8
zIE>6}VZkESghqhC)YBDUBwpRn&=8P!z{t|ZW{<c1)EYkMIDV~5vYmDT)hfEITs`+D
za(a^n=WvCjg6xH35WX>-y^-p%C@_Mj*@@Mch+IQK!Sm7TXjXU3t!ijd^&HWYr<vp#
zZD@*Pk(0{#P6D=Hfh6XNK1f0^4Z>w?KGnepxrGP#@;Nt0Rm03tOy!6#jH08EwC~3Z
z2*yC41MO6*+CUO2P^|QnF@_MT|ECfk-~soF(f*=46hU<!9?@TIsX*S3o6!q*a58uj
zhRX%d7{d)XQR_E@1!^kvb3lcCWRlTJ8VnH)6CDGW(Z>P_zbBsWluf_6%!Oo{Qcw@d
z9cin{RF|r#k+>FSxR9}76>x2rzHu~7a<_Cx@Fy7m0A@)c8>EIw?!&-ipVf4<J_VEz
zkH;PLR3+gd&oun^MA77IM~u52G_C=+%`S%L=78?ifZt=J&b29gTX^_VHhAzQ1gvvQ
z+Bqy>$-<_>`Rq4d7d9OBI@M%c1mv+oco<6N^)e@(yPsmPX!n4V*MeHYuR)RN6uc#~
z@Cf&s_gGk_V<35@^n3gz))XMILu4NIRJthP`q|0R>H7ot-wj<kyNvGsCaX^6!dV<y
zElGnCWlzvqC+^O-KDgwN(4W&FDssPEwczCUA-)@R=gc|q>Oh|dn-@>8X^*skC10)+
zBTkk07KdDZ9#6?*%s!@dS|mhMqZ_48Z_@E-<VO_M2+4-giKZ~njOFN6l9BgV42@Bg
z8C6g@XzaaEzB-wS#(#%|133w&0o649RzQ?sh!V4(|9;mCmn-f!0hbVs!tCQI<9UBT
zYcDtgLgl_bnh^B+uN?FL2^YPm@ppK2(c?vp7v(?n=daBu2+3AZtjfQ}mwzY<aE(v#
zBh;PDXHv@0uJG_bDaf>HHbyh(bDY%mEG(b(a4Usz64)VW<3%B8CXX2Fi_9uU?z_U|
z&o?gOA<S&MvZ*;lJ_vIYO(Xje0VxF+?yI`tJ_>sMRj=ku%E4<Qi&h%E=oXBxmi?20
z=`h!W5E?tT5VbNgg?^CneJ0bov)^$#CiO*ymUH|e!%sWZ`ICy}o6!B5t{w^Pq$v47
z=Am@{gS=j=YpSvfS_|#Z&#R#>LHp}_;)!@q7_h1Xgc_tULkPD7pgQB=5>ZDoOZ2;!
zcQER>LiI;3*QMFW)=wtB=0qa6F193406=92N(R<+KJ=w=XmoERZ;L_-GZUS`yd2TZ
z3Q*Uzd{bhOuV!ttj~MEVg;MmZByg1l0~uA`0qF!mDq|9@z>sKWWgIj<#SR2WY03cV
z>79%*J}wN7GrD38Kqh^hT#Q2T$CPAo&~~gJ1SAB+{J;79;s^*%Tka69(`bvWz4slM
zCmo7r6Iwm9?wmtlK+Mt@$yVy=NQMLpBWoN8(vAU#0oLuTk@`8^;g5|2$zE@(DBOuF
zfsen5n`jJXZ;OZ7NqHl-P0S107V`C5r1=&qo1fiu)`Ty%RwBZ$O|geSuWIksm{W94
z$Vv#;?&ZOO2oH~rhz@(fv98ghHEN|VsV|oY9x(<j!M}<{vEEU)xFknlv4r@V3RZE+
zs?t=2NY~8PTpwrONQ#}jBka@alA<BWZaZAam<`uM_cB~8ru3L;)i0$LJUYlJq-b~$
zs6|RiQra4!s)Qo07SXv>3!N=gb(-lbd(vda^!+uPo?w4+v#>ld_r`;8A2pL2R*Q+t
zp=cKc)FUQojr>~Yq%tkh$ylvbV>5NQ7`K+Hg5%w;J0S$JuOy;SPd+1V=_HUgoTWyJ
zr`$%n#iqg)OhlWpE=KJXPjO501AXm&qtFeq*9at;-x?ml!Njy}B^Aw+^I;9?6yXxd
z!C~|QZGTxEeg9c7j&(Bn++@_&E`S!OItZW=0}PxDS}##F8&FgwB09#okXyF0#h_y!
zUMBZd?W3Sysng*|$G82F#*rrBxsPX5*NPn?Ss)mRvS>xwd$;6S(}q91jyG0B8y?8t
z4CHitmRTNa)9Pn3`^tlf_aaWL5Sb!+vJ9YwfSlsMGQ|u?p@-iG*Okwu{d%)LHx-&7
zW=3%!1v)i)Rn7ZH82B|lxv~5*Fj2d64KsM+l>${NH{7$ns!_~@!IQd?f2*=y>4yJT
z_Nj26-uaW|+qKPL$5R28Id{>!-|EfEb0lpR&AFTRWa|0);i4G?SMgj1k1^*t*J}KF
zWs(i3-N$SVGYZuFKXq$zk9=HnXG4e^?FLeX`OD%Z(ztyYoez#3w{K3WFCyYX)8h=5
zjzz5jllkLMh?|Mb(x-(Xuff68NYA*XF$QWV4jwaNfxHXag+dte$xaI}PO4<87I~!w
zAq#WcAwA&c0`gJvzUU-_33u=@k)z<u;E9f^mrwaR?TnR^Lsoy;-8EPd#O3uWo@|kt
zi5b0})2-EMhA3hgn}HOZ3|I`bTR%Q;B%Y<;(x>{p{tYfrQ+R=VKP1L#4ap<Qp$5;n
zdzR?JA_>sz>YiSBQW6YJsok~;I@LC67fs<dHCO8yI8i6c5%s}hk%Kl_J{Lqe7&D%>
z@#4tgA`A_5Wt#dUEmYCK8j;2bx=}&$JVYcg0n(KYI6GrOXE-Ty70HIb^+|?V8p~Ha
z7N;Wc0@9uhL?jXjARCLTAPF%%jsdp{Oy;Cf0OTy#kN~2`5Deq%F(QjFKwLZ;p+wk7
z9xhMi=JP+nDP6k6A7%C9|G*Lt<a3Q@X&b`jkh`Y<%oNLy<a&uHfq@X=B<4NxCXzEV
z4Y0C~TepNNzP~xDUYi&g$4Y{H9-V%8d*<@eN;NgisPdd;2uzSEnZPY^!jj<bbTkYY
z!++i+d~z>Q_Tov<e?EcF7qX%g$_Dlx<HHL+E#9wZ_qgs2TV%3t_LEWBLe}t06?3pJ
z$4t8kV5PPRo(M^YRKWA_*$kDNz0s@VXNMK;&1WP^oyt_S2FsH${0swm#(*Gi3_HmP
z-`O69h4R$KDT<#=xZP=26~S94q#FH<6seOGX|7PbWL9~YWA1a8c!cujhN{rm^Njt1
zwdV#MV^7Ai3c3d?6byTfj}3wp&JE)wHJ8TLP=yFV2;sy-i&6gm0LL}4LA8`v=gzHl
z#{!irMUcA!bc%4JHdU5KzEA9JDd`wIaW#DN$F5+67P(*RwCKTx&T^}xen-HW@LrGj
z!gsQnp0RJUq8YWyV(#qJu&_aR-Uq`!%~dlDslo5^TS4?Z52fHD8lnq@-04ldN6`Rg
zgxkefUKiRV!?)uAx5F+HTBxULCRkVyOU#=&i1=W|Z}-9H?UM?(mwvTxXJ=on7(;4@
z@6CfIiWS1>E%`5|C{^S8rynX#_Vc_^RDFVb$r}6aGwWo-6~rapA718TOB3}EPH$W>
z-F>5)V%jR)qqDR6Yn<3}>*A{!qpusO<!>&17RsJ~Tyc71)WVBjQh&}`!FsAw2XR~a
zX-n(&spd0f*JCo`(`D}P=)4ZbB(W5Iu;KMzyIPIkjZk-;fu{yu^bpQ79N&<OaVkCY
zb*?n=@`mEXeaoDm>u-_Evc1y5XUAHL&nvxd`r2#%qI7wK9?*7?O(_g=`JH^&wEaBe
z#ObR_H|>7knr%w_ApdMe*VM+sI5`rs{B<{``jgPpj#TpG%tIzM=^@wab5FkREIjyp
zk^gJ&?yL(pIYRZ~$YUGz8$PM5^P;6>UgOE+hL4H<&zb2aaWiHCvc;1_XV806Hrh8!
z+J5?r`E|XuE&^<((sDIsFJc>Bq^$*?AK8+Pyh%{zVL$P7{$5$YNtwp6t0&AWN?oy7
zhn#ApUV_WV@~^|G<+upq^E$}XpxD?9mBakw!IVOl89inI{;cTj@r3&Dhk~n9ze2*F
zJa5jtt9Hghl;@yVBUL$}+!-J7>+s<fr4M<#=hC1{@buz62&;T!R8Q8Y#tWZpYF%fo
zymrsO%{(LiBV;y8=YyhTbH~=Hc38Zwc~GvAcK=(~`<00$ZE2-%zX4Hs$XcG=(%XVc
zL;l~10+PYH;p=V}o>@0BJ;|JyjXS~4Gzvd4uk;kBPrF}FlX8~Zbb1_NaeMD0y=}lV
z?}G^w5E=P*M;yowtM|9vES>NB@`1XJPg|~f=eF*VX&=6pu(h|f8!Fr}uTBYDIdDC+
zKi2c({GlQ1;nOGI1RukcHcs1GK7opvFhz^r#CjYjx`T<Tts+Rx%w89aDh+HLY>rsy
z<ia)@@#(=;3AlJ|YrL?vA>fh9ylIVLu|oA|&WqaJ+3AIVJ{n)pFFO89VyD*F=e2l?
zmZ|Q9a5{gwMMNI63LmWgA>TXYdzF^5NXhFYzm)me7TRE6@2OM=jcho#6p=^X*|;gw
zyc7rL%(NOoaOu7-9%Bd&F-Njb#as|7y0YYwl3v)RVJjM;9Ad%kd_U=C#IiqU%X51z
zN<(P$FG7w*1V>$U?{ivj6%gw5)3D%CP*+(qL9tuX)JzbC<fG_W4u|_`e>4Wuwn#uP
z7qItNP#GaywwcC!{6uwv5n!E#1_i@<tiW8hDI6dqxAGAUn#^3Wj%FfPs^W2cIlj>#
z{7sB19LL9Cs$HoXMnKa+lncC@mH_5-H$vkR73c5n|6=Rc?OZcK0Ke!k0#Z0is+)j?
z2uz5v5F}^kLPqf@<4w9}3@5=5$ulB&s$nh!Dn1Y47PCRZC{_fWXP*5DHG(6AVb@Hh
zu9WdCAj$*}Lvqg7yIknU%f1h-s=lw{BE}_9(d{w1Bu=lO*jAU-%>zvC+i5x?P@?v#
z*FOejtdg&<@@LFA*LCeW{<P&79=w{)7^lE|bzXwp+Ih*|hlr#z^$!XeE5nFWHU+*J
z$0U}MqB)TpezS66_D@BUPK@~^l3%hkOXf>>Hsa>gzF@=a*uqec1{MX?w5z>DFIL8U
zQ$42%7xLPvdEMv=F}TKk_Dn#jCDuEET?NL4LwBKa;hmG2zDhh7bJEmWY9*n>lGe-<
zoY`*U_ax;_+3Jk>Z`wU|Or*syW(#_&^iuRp)U1=ON8U`0PQ04T)Uy`oS!;VKRu`Y)
zz0zT<Bi$&Bco%6}2ByIlq6L7_SX7Jj7{*{YEN>40QW>Mk&rKjaA7>KtG%l*z7q4g-
zdDTb%b^-Q41YB#XMp1oK^(6|gN7HfbO2OqOb-rMK*DE^!;YFjZ_)zFV&6gmPk}sOF
zkb-j`Qi9YdU;Kxq%WXtCbzKX>DMFJkH%w(Rg$oux-EN+4zf8wZ?naj8238;BEW=O9
z2<|_nH&z*BJbtd7skqqOL0^C9dB;YLK~j2fMS?35I|HP(#U!6^AFq5u@M}vn;>%+4
z*I_V;Ey)SL1ewM2&&}vHpT8DyV>|Gp45LkJZRGO`mU0=>8S6VeT2qW#N!lwla@mGw
z-#RsYczDd}LtWu={G6OR^<%~tQ|&dIuRug+`Rl>Qi6O^+)ixutI{y>dWNwQ|ndYtV
z-KUU9i?==fVx7g>URe=-5Mp}D=_`Mk@U(H^14Gyt9Lzm>DMCCEo)^rhd2eIQHFued
z&kN`@XR|1}GM!9_DYhS3%uBa!Z5nGX3Se#vq&Hvgjd;i!W0Ekkr1)H~Ex%jn>C47M
zy=Ga-liPRn!fa=*zr}rMf80LyzN7hRN!7PgGEBagznTc~ggX3)`RnPDPW~IoZ|fga
zs6p%;(TguW*t%Sw#|0{1-uV9GbYN9sCBFrV&3q)N*-872(!2DNzipeIwbGSjgc=c^
zav!ZjhIuO$h<WStKOagv#0`~tCiXCjE_o&kz}=?Wwk**>RfHbcmD7AO+jRo>+<4y9
zX8KR=?(J#@uQuFoe(A##qQJ8py$rwT*0H`cg}wE&xNYCc82z`5Df08C;=VV16D93?
zwQGF0epo+5O;RY)X}B10o8ary1jHPP(_3y8olvJO&H)O)W3#_=Wl157z+XOVRMn!(
zk`Z704GyZ5?B;L_L3;}7=P=7gGpd>mrWKE>lDwB=qv6LuCl^ZICPxo{|HA8%8MAKr
zB*K^m78eJz04BbNddw~?gCsACE{IUBY9L}XR}&o<{ZM~vgDm6WPx-dDT#+gYejkEq
z@y0>kqREWFR+WwcSwTA0aLdt3_{qat1F?H%vQPC37$2NEXA!>})%3P$PnxkvaYZw0
z=GltI*xNpt$EJTkzGX7qd4+00aZ}@taiVpp56+L$eIKbE%O@H^MBE6Vz*CH(FsY2r
zEfZCCdJ2JY@flXc^H_ch81fC?N(1Ru$O73wA=j#W@2YCrqTM5Hdtd@5O7Y{MW>2yQ
z!41QRU8v^E!N^LISrMQPVEh8e(YlDcqpGT=+>Bh4j}s;@brEN!+a8+WF8j={(r4X-
zA=S`LfI`~%z~FgH5jsIgd>WwI-mp5o3$R+}U9_D-8x0!$4u6dzoB?C3oY01k9tX)G
z2OrCS3oZWcx5?2q{w?&^t&ilx?$*O^wgbwhh)?<A^A?xBSf+1MMDGuDA5wu6m~hwh
z%?i=-kGx;-TYGugzmF~cUNeJ*CLdjEr{+tz`Qh|EwM9f|f^-{Ts^dF&bb~XOHkI&R
z7ajcqM7-BIs(qFw4R)1!_p<I&P&pSmarxr!Tnrk6Shr5uz|YnLuMD2&+&ii~08Tf4
zYdE^^FT40;gT1VK3OtT-M@1*|k7_+=25WQn9PZknTfH2w`K)kFK`s}ppe(<TOz?j3
z<`MAiafh`0--8aG^#f0yiwBT@4~T|q5PoqF1vwNN3i8ObvLWzCjR9E0N8Q4QW)Jiq
zH!FxbHU@y14-DFy>-8~*{r8?x0Eh`V*idYiDjT(_geH1_*7hmS8Oh973Y0TDQ5ZmK
zi6N?Jro0}|H+6C~*E3OgtKBP=hFS74ulMc=<;LF9N6xA(+@}(a&M^4MnkhK0ughFo
z5G+deGcvvFm>$ztrx3K*TD{~$wvf2pZNDlIk|lpb*f<4!laR$Wt!Zl%i^e@oG&u_<
z&gAB3)$v;nGka<2#)?7!(UkIy>Lp+7xwyw^Fm!1P*?q|8m~Cgd#-n7lk3Oyi<GX8a
zJ$3F~fomzg#AI;=IH$9o$yt~F&kEL9QE1u3NQl{+;z$RXzF_x2*Ks>)(=Wm`xy+(T
z*dRyE7x`gWLz{5#_2&cT&leebBNvJ+?it+hce7d)V-b_TvZ*9VmzCR8Qg+0<ANCc9
zw>RH=BXHE_A9*l`y~Vso@71G#w;lthchN_`L7Oj0+k1IwwxMYv=MExHnG5A>U1@I)
zO)|+Yk@qwkYHq~av5#+DY^KXoi`LHsmM^;u1uDP4_q?p+&a3t2)v^+|M;QuO;waup
z)%oqyWx-$co8ajl)iH~0=fXykect;dfS>6b_QN@}>QPe>=2_b0oUx^(q@mV`@L{f>
z)c6z-2v@iOgz4=8%4Hqk-PI)s7<G-*M!lN=Vpr!)ervPtOS)|C|EfATxU^M%$VZ7d
zyk&QI`0&T;@RP%afHuGpaQ)H;tihtr?;l0g|M_AUSet-M=Kp*_gAW{2Q7aP$o1g!H
zQjZ&c0dj`N1-#h)d~a_jswe%2QOmoZJ2t<6y;w9Ez4`y)o%u#ST(~%0oi}H?aj)I(
z!2k3QvzF687hdd5mwnPJjy>dGw>-T5@L*Sd;JT)3rr_ioj)uO2|M*I}|J+Fb(m+pn
z{UrXSk^bI5%CT>N<naI0NdMeG$)w#--hXN&vwwOtM-@w1w?VbjPoXxY*@mCxu&}0g
z)!0ea=?b1M^2@v3B8n|#uNa>+U=P*a3n4V5u||4}oD?wFf=M3|fkpchf@#*Qt??Od
zkSuN!Si+8WZxP&Omv^WzEo~C>@Mm@^Gq)bjLCi|h$x&<7Lc4~O4aodPWYP_k7$ADd
z80HB-s6uPCKx>VBXm7|$Y3hziT)#x1Ta!hMlrIbOljcs}UA8*>*B@RL_EtnGQGKUh
z=k*Gn5E**SO;eJ(L*VN!D{upnvy>I29>~Aze==`2v~#=JRYd7dqsGmLYFFSkoKtId
z64|3PTDXYiW)rv-QQoFH`^@r))Ob3wNSS;o^3EeQ4!2Q>3KAIcR#Ea*jOk}b5CBy$
zgJWqR<uEe{n~p8A6T3lGX(bVOt>uwMIMFfCnsueB4&7@q`ZyR;OknI(w<Kg7C#s{t
z2;Tzg2n3jLIxBZbnPw0HWTyyVXNHrFiLQWvaw+$BvcFb7VcQ5GGCNn>q5zi=O$2ES
z+T2Pq1U4A+8KWQ`IJX1`JlcVB)^|n!&RvZ*iiz|6qS`ag$LZDfdVM#O%~3{Uco5(;
zM#3=0+PTgmxHHFED`Rv|@LN10`J5HWYDOHp-}cyf>~(@MO3_;f5#$ihc4S(+Q89;7
zhG|!bVr0+qTegnLV0tM|cRrN68;gJ@Vlmv&UCn}%CO@q}Ca|BV-CEv0s-+UQsQs4a
zkb<-~&UD(P@M5NS?Hv4f4HqDoEaoQ|E7dF_b~I{?x2Y0)^~>N<@9{UCoz<uAq!4pB
z$R3h7iftkH$Xxxi*8RTz4S9m8B_DJ-_rfjz4C3+7U!x(P;aQ`AD9Vs(D|ANjXQD_W
zGA*EcmX{J5IGL?s%mZg2W7<d<RXllA0)W7U$ks67Gq~jC9NtJ!<82;l{n%tQX#9OB
zj6u6r<l(P4U>sxr;>}OQi#qxSru0I77);$3L=8l7DmnJveX-<sG?TsfE?_^5+eo)B
zYtrfZFGNNdiK`@F^VN3)dJ_!vAZ_eWPx22aI7%Kw{2w3vZ`_;9YoZ2E_x^GSN0csS
z{4jS&)C<A(!dV~S1a!nXcz@&oqq&vqPmUSkB@x_}U=V$Mqyr=gtzcSA%juTvs}LO(
zVu2*Y06!GBQN~!Jb&Dd1c#Uln<$;zhC_1iPAr&p-w(vC2VCnItNHTN%9(j(~EWu6i
z$d)aX(_R!qzg@jd`Zjw$cTmz-X${c5LU$@BO(a!K<*oPt0f=V?!5u7>vYf$e<Hu0?
z(e805KD}m4ejQUzs@Mc*Gj2l?F=V|2!6Dsz&)ymhH7VmTSCJDRy^Ys0Ep!P0{kO%p
z%Al@trYovlKF)ZSx&ipb36lYDDJ>E+$v2xbLBIJk=!yl#1nfZ-28Mu}^zdI86)Ln6
z<<+5+!QG3%GK3M_wyn59WT`?!dde@1#4x^-4x+JSu+I}n#}OD6X~esyCQKag5F!YT
z9aRAm<^T+708AhKHTZA4*<ZGfmqCD7#R3MBB7j_AfP#e!NaV^*Jq_u!P^E%c9^#X?
z+rC4j@|JUj>lFLP(}Gk#$yXRlZa_S<`7m%$D;3fWK$rw>u-uc@fQY`;S;uy4eYLQS
zAd1|(S5XUlc=dAA!*ZVZM;AoS$ODEl{a$zzbLl1vZV<I97M+PV{7&M;41Ocfn+wu(
zML>`DYj$dfRoH59%|<%gHg>VjP2PiznuFN7@tma0&lnsm(?(DNun=q#_^Sn`5#X6n
zL;C9P({RibzclITHSrbx=@!0RSn|7Lb$6Q2Y>TPS0KYv}Wy5nua<8^m!#p*!Nz_3%
zA@lYsqEG(k6u7*_lKbQEd*S|*JE+*kG0~RV?D>bDg>sq;4vNPG2ef49zKy~JG!~|6
zMPjDOxiCQ(kUU^QXlPfotZW|7DVBo%htQ{<O+~RN<1m_*C<feXF+9=7S2dYEeX_@}
zoUuuryIc(a8pcb+0sTdelFlC0H{x@+9ra0;Qv7=EWQ~dM)&E3z{xdKqtTXj4;ZszF
zFNWcQSk-={@!|VJdj^gFfdz^3b{qU;dmA++u)mfG$O)Q$6q;dI#TO6cn*+=@@ZxuK
z%|yqd^{NGQqSyfjAM_7s$`?T8x7Vx$mAd7Kp{3UN40Ub97J$FY>{iHCsPGX%!!YPG
zJP#063a47~T}6=SFgvQc1Ylyk<XKXTVGB)<{Pd;hKv>;Ae(+S2)!e2%DsIHW<FBw)
zwAIP5#ariQFWi8=lP@(1m81B{GMQ1J)k+dHS-c1{MB=zq8esNp0kW8VO?hnO>oFK!
z>zXe{leTmb@(MVwTjFllxisnEyQg|H4u{)(ym`eJlJ`N3_DjLoOWHLZfV2cNF?(ux
z3k&(s8XyF1$LE_}0a6A`!sRTkU8hPWS*W5w?_U$#;mbv9D9^>TW(L4c&NfSnEE!cI
z$qMOY6?buA0Fg4$jXJRwtPr*ftrdb2&~geJ6HC4zQkGkZk?Y7r#&9w(ChjzZxqE?N
z%wSn;2O9RLR&2ZY`NuwcfWLonsO6wVbAQ{#X1(pj3UO>%p)mR14j5Bte?Vs^^HtiW
zqXPf&GW{PPc|9chh2_ankxT0}|0p7oE%4j_%FpHIa+d-?-3feP$lx&`+bIn&ocwt7
z&rBt(;c%wt@CF6378W&vC@;|N2Hv0TX-;rGIx6~d=+*C_D;rAUGr(z$@`AsB3oIX%
zba|^^r_lYy%XS!ERb!i1nZH|nZ`Q@2dFgOrVwS#QP#q9@K<PAjzfkuF6lir+-=<Kt
zP|v8Kqu2C&;G#ga+Tw<~b(Y?3)9@^9NA-3a(k|Fs?+2`|EP&gL{A#HJYRvPGQ~&nG
z{(3po%N=wHC@S`k3xx3&{Cj=uT(75Hmtpl>*E11)IZXR>HTJB>B8Op(X@%ULP;*<A
zN`ig%?_`IK2N%N!$`Zu56=Hp5p0;!=+b&>C5CL}LKYd$7heiqs8b>V$u;l}-dN+V3
z4&%Ate`!NW4^YP)Jp;MIw8QEA-^s{HpTmVQwA)bb)f%&}O0G|)D(&`;L(XW(1{hPC
z``9?o5mVat>#$2MnGE9<;+@UqcG<tn9)<b#=dHJBpOD7Y8DrdUf2v;bsq8O*;k7SS
ze`-zB!%SEHW8y~9sQqAQZ$r(BPg%Dn>ovAD$Gf>9MY#J5XbfF|b}YMr$9c;_nSI^q
zrl}MzDa`S_pvC2$@{?uxg0+Q1OVY5H^)r><X2ep@>nbZ1x@5q43rHf=->xeggbLX+
zeM@bBeJv~4<FV6acFE*H*Y2rBaVTjh%gpTo@9$O)kC+yg_J@Kr{^~Q)&8i&sqMlC;
zV&LEodE}qveYlRh{KjfJ@pD%CPv7Qu>*nTvK#nb%N~nn4;ul8`vGQQi#qAZKC4_5+
zh{V`jFzNlObX=}L|DEKgT6ova*1*vZ3jRF%cO4<>?VSE)IPiSCL;dOS4?>AoZ=lPU
zE;a*CwCPq|gBz3Eq;Y_HR0wXFGgW5@(7!k~f2gM&P=Q})!HXF?h5;qxu-j7>f;n6#
z$BS5{o*+oxSQ(u#o5bZyU9sBxij9upGd~f7KORat_$6k6ziTr^xisbKs^14tx>#V(
zGD4|3_y*V+?<<Pg9Nr2v*aT<{B5swX6A6DiV(`KJbJX@4{txIcz=n9><o|z9=Wev)
zqpsO^bhOTl;jV1e{sDc>0o;<I2GWNBe$fIjkJznN_P1ic87!VJ-`cH!1VCF_B#~fw
zpb=p=9d!?dc&dQy;NyFHgMiiZ*M_SGKX;n{fS$YV<395Gzxv;KdgxYwTkDiunrq@h
zO3yQ7iuk24=jNfAXKmI1(cSq_gAvI10Jt4zctHxOY$7^QY``2P1I$s5z#LV?TA|S}
z|6bit)U!|#9cH1eO_sa~gbJWVH!8poE_UsT86>x;i0vIbLLs@=#lHbagQ&$tG=;9(
zGlC+=VpdqPMT|RHV_Y9olI_J-?>`FsE-arUy{&SqR-q^c!Yg~7v7m5?NO{xxoM=sI
zqxuySlNbdch4r*^KrTvV2p2pA$#$JAOaq^r{90ILFE*kv)^i784~~{!aKkLx(qIJH
zYwsBx&x9Qt@XaL$k;l}Xp>a-)*@Nw=9iK$}&^-76JJ0yWlfU)e&1u(jBn9~?y+QyC
z$&58q7+*mJ7{TTSL{-3yz-|Z#24g^QKup3KW9DxxW&#3iQDyI325b<2jX_1&0KO<<
zHVh3)!LGcGKCT6cCZd^_2}m$e#}4}0x|t1MPVf|pH8gx1gN%hV$1AZ!3qV>)Mz#RE
z1muPdCu5B~i+LfPwkTAk>?t60tPGEyq)Cw69n(~i-De#S(^{KY#{q|{T(QCNg!^@f
zw!yhGYlSB!YT6wY`#jdDkh_w==4vCS5YcJ<9k`dM&gkr@s*07e4u-i-p|>Ubf&Gr)
z=qnxTn9(H>AdAoej<Nbt%X-%AeQ97ocQ-$AjtX@rBuUA{a}lotGdJ^D5oV7!T2){|
z1SgN<GT9jr9!gDgyfO_dRdAeN_XY`Iu`-2GiT1+MgkjWM#l|bj^YZk;QUz&ym(*6#
ziPkh`sbt#@v^X*AwY+&eOY<>w451a0utdxVm!f=8Edrvq(t}T|6$&^Tm0!o$Rad6@
z$G)UWW`GG3P$#1|_3Or3TI*IqDSP_fr$ic*J-_)%OjZlCza)F~$iL=F$|fN14N#!w
zV!kP=-uEn?`=<;CP%(&VsHF4@!7duV<K;UqYynevqqfRLNMS8wC{9{KqKBS_Pt4S#
znG=m0KJ--tnIT3AAB{B;rB11@{0I#EvsUd}_61lt7fq)?MK<>qp6$Y$svrcL(fAME
z%_t6B;l5&viF40CE5!c86y(1<h8^eSKh$t=so1!}dpA3zf{DOTVske>JS=vE_^!`!
zkY(i30Fkwoix6RJ%%Dq8<F#NvK2LO=rZZT2wP@>mj6hx+L9{ED3%($>QIVq(z53&o
z8TIm8#-TVUgDDFR`dwTjz_stM23?Gr=ZgOL-S<0y^Nz|+z0I+ACvM4FRWY|hqPLy(
zw^km?mMJ+u8jY#pwu+a<ApkR69JGX6CnjEtJPhM+ESvf2=kZ0swF37T;mQ{2XxbDO
z=VZULG<K82RTxYR5zH*OQzASgl{r1idj3RE5`gd3*@d3y-d2)vyY|4V1)W$hhOAK+
zN};#jT}M``&JtPekP;)1_zV`fsVWlGSq^N#=3`6eU?LC>6~I133~!~v(*ajSAx)V<
zM5E(~yJOIH&!SR%G-<iSJkbcRLuW8iQV~G-M)Q5Afg&m;!%V;&ZgeIH7HW$ILptDW
zfYb<7M)2Yjfr$VjQXRlXFex4m6vMF)`4VKaqhSm@@MciN@>47AYfeBSBzudrkHpJ&
z*`|Lxh=`Ui_4UpZ`ufxXAi+1s4(f-nV=xgunZsr`MKC6^z?jAwksR{0l0}#P%z@G}
zqR2HJHXrkHo&<^U-22rLwND<y!wW^VKH>F@rc9X{o3Rj&J8%Y#!i*05bSGfrDRS*I
zD6|FDJ}PDJW!L=?*C3+eCX|jhDB*geQDmAae$wose0Si{j!C7FRU=Y!Sl+TN=nN3l
z!w#ZzO)f=$WNqP_B|D=_C$t1>j4ePD2)x^7uuq)m__Jt8%Bj|#47%C2hKnz{NYL>k
zgX2hrQ%e4O;=uUAd_)?_q4G>h%Vm-4GLzM`RG$SR<hx**s-G=cSqUWDY$jFR=bfn-
z%SB2Lae8_$L1{B)CQiRAhFXx&sn&G3u^78<ukn!84Ub{}tMjlqwQLywI{Tr&^fB!&
z<wbL;1dWzSn+;cS3mau0Kvrw@6VvIYwc;Tj$9S==hM~)2tVL@;1;=>2@ccuU)D1ON
zG%k*K--{*{9}5I^LF$&-Ylf1=*rhy@@zaPZ*Qf^of3^Q!*I<rA@!vAe{CB<S-;Xhi
zSspsyHlWu^?p5C`>iuwPA^+?s*XOfxfSxP+59k7bGR(5xuQk3&n1et6u4ZOSH1ZU%
z%uK@pUaVghTbLzvh)@w0Q)9okIlL|rGND_g-%G5(JS8<Zq%yklAzY!Q0Y-D3=}ZfE
zxo%_^??6&z{rjX9$vY1$dyYq!&G1rW5ovm@mHfnbd?3q0R)&+Ri6n`Ph*yF*fzM*Q
zE1h#eOGI4WRiUlK`%b-3K?#Z#=8c<aTsl(St);ti;R9K&W}))j5N;u5T$oe=A?O6^
zTkVG2GbA*+d_UvLHd}tW`hknbBgad|ghGBb9Uc(bTbL$U17a*zfk8WN(g6>J96LV1
zYLf>>^8CIE&^eGWgg6Dry0nm2z?q3$MaDjiWQ9cQ+5$68K_x?H#R!B;ooGyDLspgm
zS$ntwI8=$=1ff_em;o@agtU?weKSh0SGkMA0IwW^vC(qIcM^$uN(F*lj^_D(bULRL
z$XboZCu&!=FhWg0cA#FVr$s(9LCNa?(=N+iGw|`G>29r3@k?#ltVgOGaL_D0EoYRI
z3hB`b)+TGMBjX^=5P1SSpqpiKdDL<9{8}!MnT&5J+f<Clur666@z?cQSk*1hd<&=G
zBQ6vFVrUARR2ga*g_AAjrI<^{R56phqQyQ_+il=olKxEouAiO=WF`)zM?NnVz9$|T
zb5%OuxGg6Bb!!CG>|=9~U=D%Sgvt_R&9t^_{%x=BZ38LLs%zBLP$XYW8DCw);F9^b
z9Vx{dGv^{;&+ih8#Vw}q+t}Kf(c4y9)g0Dr7D!!|w)NL55-2X}xQya3hX<jnn=X>l
zO6BgeDdWfa6yRJs6J+iJ%`*$yY#Kr#R9wn*bM*eiBUL|hvgiy4o#vd(2u~S^U3S;6
z%odq-XiJk^4r1b1`pd4qV;Yvg(W3Z9^lY1Agxwe4Z!)r3KE%e0OTUQu!^e1^8ps<#
zhbvI>0548Ehu!HW9olH>+M+nw^mU*R!eBlzzQ0hoiH_hSdMu4K^sKO?_-1q?z-y?y
z(=LF8zxtY;%e{ApkjW*R4?dsehgJTW7Wkh8!T<7#F}{X=gy3oXqgTmr6Tl+M$8}zC
zJXja~Z2`<yT|rCNY7Le8AzAHo)n14y4_JzT4)6%+BW(fn8jg!yG1Cky6k$Y<CKGl=
zo6HVk!iG&28d_$-6NCL3Hc_}>tr{$n1%6-6YAL)z%>$oxJ_{4w_=!1KWa7iq=hsF!
zLd^3r^Jd3_Z;p#dY#NeP;guP&TI83NjM--XAl?f6r^#nz^(Q9p9hAEBD(7J|9L{%5
z^MnNd>GNyf>qWzy`wIs{-ixSWBr1Af!fj>G*F0a$X;ipSQXRX_YgShdPC-A1S)3d6
zq%e|)F6NZr_{yTSYoJvKhBYLcHMcNL<yc-phUXnslsh~U3&o}Z{(4&F!j%#|I6WAM
z3Z+W8Ya!YyCAGy?gYjHOuwD<e(Uu{kGf2D2ivVIMCmIV-l>RT)-aD$Pu3h_0=p`UR
zz<~6cfG8#OVnPj}7ziQ;EOY{j0!md0T|)~+N(2m$Pz{2BfEATclpq8|1Z<!vASwzL
z<ncMvcklh~z2AM#H})Ch`)>^bS(BBlx#qm@>-t>=Fy00{0}`0dRfT!d!v(adfnpGl
z8Zqu=z(m809EH&^KJaW2R07k{5Q7S_cX2C>_fk|4KvOX%$+>4-Z|qmDzmb7(+}y9u
z&?i1oIbQs0ZC}laph~{dYerc2VFtf^15hcmL;|L$M|YEbQmJIfpC+(AM2mFoR`NmN
zQ#Ouh^oLVY4?7sAKx7;@JVMrMlM_X__<GRZWm&uR=&j@8@aQZ!Wh&IZuc;=@#aEDY
zKzHDFYr#14(@eMivH2V*0QE%s@-ow+eRx=E;dJm_Gd$;s+H|d3Jdg)wgX;Nskb}p?
zxaS>H9pOD?^HI@0Wh{GBhc)GhQ^(Ak<m=>@h(h3$8=BNYU+Vq(SNMk9v0zT9y<KpZ
z+ht_4(=H?5Uqq!CL;AS1cks{0tOGYZDt*qw&f3j=FbeE2P4zh+xPV?wNAPtyNn0xE
zM~f_Ug$uH`=j_~dL0T>lBHJIixnezlyqx+zJj79+w%i?PBwlov)N<O&MO=o9y=H7N
z;0};~b6!dr`5R9<Nc;poMF6>xQo{9^9AxeOiH3E-DAdl5+_N8eeC#0~9fU=5mI=D=
z0!=J8*E6PA!H14z@@>b-r9$V^L>cyp+<oNVcR!>5aZhcQP6^$(nf1(9&t09NbKOIC
z)IRCmJ%y6P>I#~Bl=@%z7FNZ`dT%aXF{3Qf^)7!{^%?!0d}4#<fD2f8%GRUD7~7oL
zbMgh^?5W2J0|jmU*yok=IR!9(zaNK{&bDe8y3INqReDgFg5SN{6Ron1(`u0Oi!?J8
zf3^u$le2e%uWMgwK(=p=+hzaj<mdN0JvC1Fe}6~aEw<x8Z^6|c=gz8l^>*i5EOw{1
zzbija{IT9F-f+ZHk6raZ57r`5o(7&>+VnuJi*vLJ-c%@ntU2Ke2F@Sbvzs`Q#!Dpr
z8kP^(B<LBR`*d%0Tjqz21WkJ4$DKlgIaVpVBddOy3|{`0qQ1|{{L43`XPc*4(LK}E
z`lbu3-=?|^9-VkM)F@sfT%4ObRd!*^^pyTda_KdfxB;z`c4r1H7WkLkEKc<r|8TGU
zoV*h2f{wj%e{V-;!gBP+6;bx$^|QqfG~IJoPwS?Tpe+hhcXpB)9m$ub*@u5z?ig5h
ziC&H#lQ~%QZ20)Osd)1&hvJ~p_qR4S2@<|zm+Xl3k5fX<D$T`>$`Km98*R4PI0h56
zulz&K{hu?PaThUl2dKN!fwRygzn?-ZYh<)2oyhq57cCk@tAUlnGss5{;cd^>d~q63
z2OB|GT&l^E^Jx-$t-X}fjb~`e;o|vP;hIp4zgaFZmI^tkIp&0-l(SZ5D3G3Ia>6PG
zvKNF5O1@+f1}hA7jn$%%nh&3zIQJ;H?{t*lNuEWV{>l_>LSqzV=H207hW0|DEVz0i
zB3yndWSe^aIwc9H+=Vrtx&oKrbE%}*1Kazn3(Jw3>#`w=+>OVVW4N>Cvi_&9R1|oI
zGre}eYt!=FW!FGzW$x5Hc2Ild(JMA&(>cArAqj+U+J5rSnA(gHW1Iw^A}SG`HXlY8
z<*HqvqApNa^=dJ|@*>2S%R{_&ouLi9Sp>OtlEPH034~b$kmHpHR?X1Vth^R!qPbt4
zEo$fCXOY^C0(>Q3F%oZxsZEF^UTC8rKRlEIS1|TMYYDSN-0-9t2)YwOh;R}rqVy;L
zS-zvJsuk5Tv|Mziv^>2FJ*MVp(!<ba%n(rLtGQ;+2lmdimdSdeDHkuXO=k(W9}$z`
z-wfKhyY)0w(ODFzE5;v%goZCR298&bLDTV|-LixSy9YFs3h53>`qpYyv#fVcz9<;1
z%b#${<XnL-YFE?Qqus8(3DILysLaBLMzkl=YPEk=SALNiHD^!j2c74GhcZGsm-?FQ
zk5Xw47xug(4(_%BsWjTsLYqG48sjm}EdF*X>RvWJUPg;JWWn+8j!I}OWV*Js&$nuX
zc7zVKWqcO&bz3lj)vh1Ii&a=?nB15SuksLt4rwYptbd>>y*3C}?}&+1GYVOdMOrkw
zJ;cR%>)U<Molx(G8GDXG^E>UrYj$aS>&>*a))?S@*z(u-Rh5>xSH>+KNZG6ek5u$V
z#z*9IkCLu^k9(ha>xudI=H3E5^T*qNoNqV8yew>WUWHBY4QkTGalO0l_+unDj*Ohx
z^_A{)zLooe5#6)tB`@!vJn_`Z03~vC#j9y6;s(?@IQ3rh=o`a#8%@w!AK@(YCxgui
z$n)yXiS9N=;Bn?N9O8aCI%p<UWC!|YIMk3@H2YTy@LxLsO<oIq?^f28l$FK3e~Ij`
zY>9oU{AFg3(vfGw>Kq#`b(crKWa519W5M_I{M4&QB1l)<$z*}uY|UH3!@>og0}<{o
z*wSE|YX9oE&#XtT%ift8tk~Ik)-@5?1N~~9kF#z>RV3WmKRbN$;fY&ynqRC4t~0-#
zO<sNaxwG@-=10Be8-GKJ?qB|N^>?~^vC^H;_4ubdW(KLdb_ajeuc0JH7vBFIOXW_U
zZej1}BcV%M+bb~6fvNc~UWFFcoppcKs@7qd{EQTFz{hE>nG;;t)hm#qj9YA8V((@J
zDCL^YWKr8c()G&YetH=91sd75tWME3ryEYE#Iq9iy8O}~ReT{b<DhcpgvFKw86CGW
z|J<|2TcOz3X)G94{581X%<5mL)cnP-zqjk{`GHGM3E%l+tHA27P_u%;`p8hM%9gf5
zMNYt7g`~5W%^$CYsjwiq$DM!0{+7M}H-wycd1h~<`<<{CqklunOgIT0zmk8s?|lu#
z4h5HY9lz&){BH<l&0T)<o5=|vLvOV<iw(_Wn2vM4#Auuf>QV7{QZcXRSLJ-<=B>`I
zB46!Z!$EbF+U%IhVWbRcM}X(vOcO2BlxI2HeYC`nSKYl4|K{CYVdW<eiQ3rzDpwYS
zL#tX&<pA5KOR~2Ox9)RMXk|t^2272Y6<qiHWdgio!!g?qM-&3fXQ}dfoi$POoW&>^
z`i_pVO6QeTg47^pj5|`VAn7Q&F>1a19usXP&p6G2?B4d;b(U=)740FfQJ9L67{w_y
zO+Sd{QKWJUSFuqMSxQzDu3zsl+}>=t#~ps}SPeg5JE&bOV8#~aV~h*KOJHu}#aMNo
zcdjrxFqLlU*nP=@_SYnPhG(Uq#kj4T#g2Y6dr^4toNjgh?fiuNa3wwM;7oWJoHo|W
zT5`L#Sn_tyfw{mN3Op-WrOYer<Hpi^YdG700j~i~@kXTjkFPWNc#Am@1tf&Yfglk8
zr*a-|h{YL@qj7=XIA&;y=P(289nY1V0%oztXiy@+a*#HSP?2wp(@cH}i^GpY>1L>{
za1cy2vIQfEj0G%$2QSLt@N{?ygb*&pli<t1b-s4vXPH@`vrx4(k)-h#<q-55E&y+_
z;ub*NCXAMr=W3%sL6l6H80MH0`979jx_V%yd#ELUpl}*L_;ly_r&HN&u0iV5p%Ij5
zJ=nJKQH9_=Tmv{&7~SW|5lX#Kjq+ZkD1j&@%9EQrpLWsW?I#JEA4b#UOaIJ{lXdj5
zXVRJ>(}J>*zoad^@ej?uH6=^lXWh7OdhuRErqWqGIZ@F{FTVKD`uTQ%lzn?Hk%@L)
zl&omneJ6p-Z}?c!(c*yh6Oj@qy<K#L>J`#|dz*PnTlK|?W_=ymK-0#MGX2>uUs+}x
zqNZu&hkD63===EXKE<iNkPn_bx>|00UxRaJx_9o3M(Dl9db@|wBsHTuqfaik95l9H
zgmNQtpFO@6RwFAqSHXro^KvaMu>R^DHW|ftQ8Mv+y;Hq^YyU^hq6qU#-ydBjh4|h)
zv(=nm-dK2dd^)}+PwJg|?a5eNAo5;|Pqb)dy*Y<ZwKv0A1G(y{cCUGRh~?=<Wm|Wg
zgdr~jWC@VD+B{>-hX+Dw%(F-ZQ20x7&q@ovqCMwDL25?s>9U*)JOKoeAR4zDe`V<2
zMQWRcfTC2yDE7<ToIhTRo~T7T6(6&2F~X%we71b6_mj>=<EpEA23r>x@xdqk=SaQD
ziElE`-!tqbZhG4Ee>X}jlW{R#*vSkWR3T#gkHxsaax^`LmSf~SM7GF_AK%QKWz6A2
zL0|CF=%>eQi20EhdPDOcPkzb8-M25>+?)u$(Nij$<$B?;=C8(4{M9kSj-dI;6E)ex
z)GzbqcYBK3jyyuwdEJ%zslNHjIrrFCzbG3+I^TzYTQ>|fwHJvBip@LBi65-Ix>jfX
z^&cE^3-wG@qkU8<An|>0xf=CRDVV-YK4iOMopEt9wB+UCovA0I-Wfyx{0{%h6y>Iz
zyVGVcO8@PHCJQxIpz+F7Ej_U{_^VFE+cQD2?Lj}msm>heCOC}S$D*U}zHp#X&M~~S
zp7e(|7C&rc_wM(nY-{|^*gfcf`1x~6w;9n5{zg<&zhIQX!2Tf$|7QVJ=v>P`cA)Up
ze}Er~R>3|_Txnls_7;BlZ%BRiPq7(^0aT4^^f>~`FnthTD=af8L?akJoNo9yVBQpD
zXNDd#Aj3`zQ{xoiDtM8_9GL%F{(4oDC<Vlx<5THJl{#>KQ~|SM45D_yPVQCGWDL$J
zdQ2?H+}$IzdTnlVzOlM%7q-%KsxTanL?StSJG!(~r7LgIKsj@?XaIWRSj*g(2yS@f
zm_%&S#le}~v%+r=3!;uWP%~oxIEsMKB*<=^#R#6ddhP8_8-*K`7${ra-?2HV_QbUl
zQRLg%`uVYJt<XQ1cuOy_K)5H40q+#CB<zFt^OdpHvbFQ0ag4uEmIO8Dx31J6>`nOP
zq#S;}vnXS%`eOdpc)&7<3P(z98e_>9=%}K8t|Hy84T!7y(c>7NGK!qL4a7n7Z7^9!
zVX>}&?Z?_3k)a(!c2wne<mUqy4aX1f1)g(ZkgXDN+*6$@{0<|6S#gh$>K|~o7ULTz
z&F$?8U6;M)=pVP+N&+k<J5IK|M?E@PIuO9*VrjRVBp|8{-tE(d&Uhg(@WZtw(NllK
zUffBTe<)3ooT`Oj1W-w$yL}Ivbe1n?H@22?Z72DihtTufK<=ev$Mby7Ca0YQjUDW5
zZag%DX319mw8J8<gl3J6j3Io2^NFrEQm5^<(mmqmdg1+{Lr~f`u8Hqe_?Yw_?fTqr
za5`$ot5^I7G|Wvdq1~=mPOhyTy!ObzR#Wy&Ea^cYt*2(-&3v;fwg}rNbfv*$L?E5&
zAI26s4hx2nW%(vFBVqh*n3KTPm802;JGc#g?ds+2q-Rp&4<kgreKlI3KB7AZ<*x0I
zHKiU9*7<y8IQV(><nXZu6xZo;f(pJyE~I#5W?w|OxtDNKkbROu&|bi4wWN=akBRzM
z&r?+I5_YT1Ju!eX+{#T1^IhsSbr$d_1+_$YXt}=~$~c!N#rO?xsD5H9+dn`C!x7;t
z)q1el6y<P7Zr_1*Jw%s&=!?;q`xY)R!mH1sKxnFaAPlSkM1%^#_9O~$LFfMsD+Ohy
zJodZ$>&M+K_kG4@-#oYdz@Q6$t^nR#3ZOIrtdx|5#_y!(05i(@18rIi_)x3znL<|i
zI`!%IZ@{;bao$o@aNgSZ6Fi&q2X{5{;F-g~3GrzmjWb&+pSFDlQQsMXgTKClcd*|=
z{`q^-#FkS1bNsMo8AR~0<lg(8dtXC%d*88N{<U@Fw|0K0@<SjK`#%j-am1#?{$B<L
z&cY8^FV)>C-`7%p&@IZ7k=g$;vj6d>u?GK_{@?%h4;1{%6|Bg<c0T3)f4R>8a{Y|A
zg-HF&=>OaM{olJFm3OJ@u7{*Qim*vO2Q_w`84#}g1@+yJ8^DhaoNF{v-%4Q-J*nBf
zS8Sz2<l$HDlu#`T#F{K!w}Y?DzAv)vR4`%U{FM`#iqv392-L%q9@M6s5lBgBg6>ry
zII3MJvMuSIFNe&RD2F3{cS5XqAZrb!`PD5bb^tjD7tQJBrw2p`mEVup+3;nRW%F%J
z7;0>c>#yF}uE-0*IP!tH(jFt^5f0o1oeK71z4d;H-?8U^9-xOu#@2Au*y4H-JNAZk
z9aE|3k>wUwM_R}~iZ;f)nJn5J?FzgILP&Gfr^Kk=Aveq#!CdL0YnAabYGqlM8n){J
z%=xzV%t)W5Cl0*>9t{u}cp>@PR6%*5Qw8l?v3NHB&JDB_8FH2!PvASv8bIX`lmONO
zVoeYNFq>@nYhG%OqZ(x%*tAcfDR>k`1s^5Cuv;v%*d!{*6|mGY<?Eq<)IsBFsR}j_
zpmYK`7uf2A`aij>VAa!km4HpR4*E=tF*0S2r^tgEk>^wCo4S8Pu0%Xh@5&#@3CIZS
zeSYx!5wkXU_5D(g4A+!7bXtvRph#?j;Q_FNZ@x5DeAO8Z7rg03gko46faY1G359bY
zhuYhVmND??xu*x~8>Zn%d8MVp9EaJwi<3r`>HFSCx?W;zG(R=HAM)_8*PdF@702We
zO^Rde{H_b?r7SO9r;Na4BFVimhMej+tAnPx6D|}(R7e$)LS&=DByBCMlI|M}m>hZf
ztqCp2IiF+K`a`ce#ZaD1%!QZ0cnhGdREH3rX&?^DKJ23r+yx9DI-1Oh1s8Vc0cBQI
z3(29>X}CJV0JmV$9BjPPFa4YfND<L-&)5uPOO2NqODq2J@vl?;R<Eai-2+vz*d$2X
zT5mw1IY+H0W6r)_XOybGW05*5=8vU=#_Ee9o}0T}8YNy!cIQUDh}#<?Ae2znYtAB}
z&r!EM$S_gX^IR`2iUn(nTx4d^;FOX`jf~Oyd@W-7bJ<qzw<Dj-ey2q|<0_3Wm-vMF
z{&87CM_hil->JWZQBW}3I&=5e<w+mze_Kxc=fVOUo!kp<*EbqKisHZTYpWCcSj^-3
z5T$+0BL(Zt=Z&dly-Ztaf(>exEO)9}Ztt2d&~{&bqj!X5bQsVwKcf~>y>TC+J<Gn=
z^QUS`#=7S6kqsG2H`lO=lN7@N!QIIOzE!Xl*na7{Azsp(D4>KKbyS=49CbJNk^o?*
z-E4zZc)!!L!AA!z-#$wly1KR(4v*RW($qOAblI^+JQhcW3(V(^0lB1vnz##SK+2Aj
zOMg4IX`bN@H(!X>ALm`EyZj{T%t{(%tBK?JYHDAsPEUrHBnq-j+QsMT7D#{J*=sh<
z`29)NbFT5DKjVu?3)&)?`7Bm_)5Ok0&$ue@F?!if45?tp1Ep;Mp8+ARP)B|k52#o@
zc{D(^;{bdK956U6iBffRQ4T~E?S)vwB{2D9Tq_CC85$9qJmv_2HjbjsAC}Mt?T;}b
z6wKe#O<<S{&(KFOus|AtSS*Rgj4=pgNY#6ep)dZVA~@LwydP1&&JZdFettxQq#(c9
zsSr+w;m3%dvbEFW$IwsroSD6<wRbhJ-T}u3$z*Q*>LsV{F%1pzNJMB^mjPS>e{>HZ
z?cxbY#0SqIAnx3aWeD=qJ0c?{9zqo38IrZU=f?@U`N>b((+^%JIZQnj652StQC}nq
z39%4r-ZiDZj;?4%>mBk-e0VS+;9!_8zU`iD?i90Ab+3teWi$boxz^Ws@J2D_;pn+2
zl(6ts6Ne%o18%pSeCYVYoFI5Ovju3b!;|-)ABZ$P7$Btk_Ga*Qk8Pf%@Ia<)FUCj$
zcaf_<;7T{<y1))UkwNKWVV=^HyI+Mw`WB6G-+Xp=7~NqKUUYG`VvHU8)bs3XlD4a1
zn7uMa$vxzd(_hhj%?@z?`8o;iAxen?A(<%E`E(g2n#W)H<Wk7)tvE(_@PhPq&#P!y
z8AR-SHTk@w2D&EQ3yS|Wj`<(}KD{hyV^_LpU@8~!o#6v9^w@Xz$9{d#Rvm5$x_{)0
z@w1d$FE>S7!gj?wH+vqe3iz|?7B<w`9N45EhlGc}T>J7j<YS7bANoJs%Yy&ZuKxQG
zv~SBEAS?44{)*l^+N;pj{%G%olCJHAe?+X!`mS?RDxwA~4WUA3?CX@Df|obnr_k61
z#)?pmm%3j*ETMtFm>MYc&W7AWbQ5kGBt;J+WHm762$pI$AFY47uBoAl5sk@M!9+!4
zB>OCw{Cuf%Cl0SSz{j$Y10WRn8)FwbpKM@7<$DmUW}DUXJm%Btju7ZWmU>~ygWCpc
zE~OdZJJ_$CI|kVYuX+!S#ncPlww;?hg#2J(!h=#Ef7+DCv~_Q9^j+_Y|CB3sykzuf
zuV(Qux{ehbsD|cu3*_;j$i;FFyGr7#Ons8#gx^IaE=iXwne5|U#(wV<aiKA}+@?=~
zB#?<Np~83|Z_$$rBsx7zCcg_ys63Oe41Cq2dAS6GKJ@lMxQJUN!J-~*m5xQjJSIz|
z!Qx&Oh=NKW<s1aMFTvVNj&an+U0}`4o(zg1N<5%V?_Z>lVI&EN4i<wJ5wt8%D@x6A
zr9#}s$dEwlQNAkRj3}k0+n_`nq9q1dhz$l4IbSuES|wS2=~(egdN+0aSj2V2g4jMr
ztH-;=s@W%bZn=w&L$%Bi+s;nkco}OiD&+Eq1Ima0hUj0f-r(mmANaGFCn9bOno<-E
zaiFrTNR0!O$t4U2vhT#Pi5`A)lZj^s2cy#Y^<8p%qNgr8MaTzsKl<dOT6rJ)TV~R`
zA=E2mBU%;KpgPeK<lS%4{M6@>(Z|CG@8Ltx(wHmQWt6y~Y^sKBg#BX@B+D*6fBoKL
z6P&{^K&V-cp$}*F$|fm>vPyHpBY4}X!=)EW7TK7%Lc2Mp9}x(&4`A&tEKp+<-Zge1
z6D(NI39b?YlxK`ABNsf|N1y|QNZC3<jbF6veQr{#nr}k%o#|AU7nj_?mBxP8>%DW)
zX;I7n%^FD^W|f|;(RDzA>)Yxf=?rhaHOBwT+bt$s(_3Jwx-@+xXcg-ye&2B}XVye9
zl%oAx;X)6{s9Zq9;uxY3rZs_>BuW=$WEteIzg?mWAkeQI)uu&R=eJ&FdKou?ClrC>
z;Z1Xn_KD8_`bXS*6_^bbtbg8p2ei8{ol@uid01e3;wSJreIIA+RzCCcEAWYZs&l?<
zwzZM+>;2dcnUUgOt3SsVUH<yE^<JD(isd<}eAaM%#|6u{BfmG%f)5^F`ki$5ul4Z%
zJPY6Jf4-dfHRbUEAQ;Spp8U_(|9NUk&p%)P=czIN|9tuXJhE;7G48tm7=d{9zYlx-
zpXdKy4=?)4hpi5|%}O3Yy|j+Ecc6u3Kli@jb5TJAf`_{FLLN*rz*O1_Hv7{E6WoO;
zAQlf}p%f>i3}&Kn2EuU2>*WNNd=M_^`sO%~CS=S7RKm83RlH3EjS}q&$W|i4))V={
zcu3UzKD?;vD=+9QPFCK!a!gRkZt3Xt_@E(0vDYEATe<I=(9dwW#keHzT(_@y3nnBS
z1RVDP2!V6~%fPHNCsy`u`GduzF>d&E6F<lm{fLE;a)*Y@l?|4Drhe{?gr$QOseMdR
z&<=AnlWMn<3JeccQ>EvBMIDU&4#LT&7B_&LnYh#6&HBU&C#bq>TdaJPwby(h(xaq@
zD=12}LrHT$a-99mSa)jzc!Xn%@d%y*OvWi30Yrhh`&DIFR<<Qs$a&g=4T@5MuBD~3
zcZ0Akc@Bu0tEg>)#DT{CkqA3jp@)(M)6j|+CPeG2<Skf_@zD+4MNqNSqiX1?CP*(j
zOeq`@M)&6-N*EJTowpmMW=I%;`J!^lE3eSP@RLh;ORCVRsmSX*!{6L}q&hp)?6KW1
zKhWR2d9zsW6pL_%XR+i#q0>ODqhaRKqyEffI{4<cXX$y$Agze>*QdQ5@1EH6ktg8A
zQsg1MI_LI7UgJ-OLxkbITz)xLnuZ<(<ianZe?egZ<*_7P$g0;foWEq}8`+9*1jW!z
z!+K0$>br#K#t}Tvg}*U?OO~GJJa^jz%C}wItCcs4vsKt($3wJygOnms`VK>1V)01|
zsj|}{O@?W@ltSGQ#L=WNL6Bp<4Wnl#Lroh$F5!{JD)Bum=q|Is#^T2EuGL;XBv_=6
zR#f!aFJ`|qWJT|H^U>&X_;Os4=wKxA_QseBd5KgjDxa?XlEg+QkUH%V*%4E5&3BGI
zW_f8<PA&vzC|`8?OzGWdamzu(M@Jo_HtrC*;-l^90q?hBJN1Rpp^}O|S>@+KkFCR@
zPVf1RyJI3`U-7mhgQ!7YA~MY68}|5ng{b{t>f??@xiHc!>O*ys;_*W1XU&RYaSTQ3
z84c-kvaYeH50BvgKeSN#&de0Bf*l40vw+V<P3B9kl^adqHd5HTrU#a=DsS|T?gy%f
z+1tGhZ*lABe$_q}LJ<^*<}+a;kMZ{Q!;8gnmJ~ANV`o^p#{*1dI#vUnWiA!|iYOC=
zpvHlm?YY=}bHZ5MQ9t#%kPSwd-n{tOr#DTnPPZn`YNzJav2hT-;s(4uc9_9K5U>o0
z_qUUY-*wZU(xs*vJes%L<)@W)nX31cVEJ*$l6vKZIX4ZLJt8#2)XIS!50VI;`{Gl%
z1isMxP8h+^r}O)lPlU}%qNDz33C|}hEAAR)M)>!Nk7R^aZ%MbJgn1xMbaFf&6ziF)
ztWl%@VA+qs3>^is$j=z@QeAMEr59ix;%mgGs%BF2?$z0LUI-*+g+r9H%Hkt}_ypA$
zkn!9&27ebEQqF=<`R-E>cV=lDAQgF1QsGKqEYA4wqwc2V*+J;KR4)VVB8r$j$1ukg
zBTlto*nMMi`DB{v;E;@QERGbz-*c7>Tdv(1w0|AGK%!r~I(xh_d!{$5sdO>cUPPT*
zY^ojWN)Ln>MV`3-Uh=B)*==7K&i+RX=fWC^htC=(=J#*X-9$tf1TPrfpNKTf9TRfo
zQ@RdfwR3tXf+v<rk<FUtZa4BDN+?)k!3+;x>!aFc(Z)VnDwCP>w1F(`hVEX2?q<-z
zYj=&s1|K-|Facdh=PfdXEcYirO0kHklh|4&-QY)+G~Ihe^|-yFjw|Qqsty>1_lPOf
z=k={%)F*{Cr!uTly}b1A8~YS1HSa{p!TG(8BbxN(Rxjxl;M?@VKM=Mi_DK+l0r^~X
zx1vH{+|Hf_rKR!vensSZSALP_Q<ZCb5Gy_vFuC2j7J(Zr4o3RaI(kFA+`36Px3X(X
z%S^exeK)`5lUmr#dbR6MG=@!TWru^BKq_$KtOY~p&3em}{x#G#HiTII^)e}<l5VH9
z$}zbXkpY>{&7HiK9rYMKB1IM)M6KcmyF7O}f<!MVG2@xEj90|h$L=8gD%{^cg8O8v
z{yTc-e=J`AWmOAx_5#AiKX=X0kH+LGsUI%(GFTI>oiBC@EH(VTp|V7JwSC3sd~$*C
z*M`iY;by5Tbu9H<@3}Yt+pkt!D8va4Y@Pr`)+~;Ey;c5ULDkU(9Aq)Q5L3=4ws0F)
zTb`*imCB<HyrNS-p6HkzdUtI?{$0FRSN|ss#W#~^$_dipG+H0sjqd30KaLZcFGFW&
zN5(r{e<oj%t<P)L^P2BM-{ehY)!TNQE|o$p4PPndgz1kkH>8VYRgv(j<qUl_YaV39
zDUl&4R3&q)V&~Wx`gC}_6T7;$ggQCdlA=-(qB@ABFJ=qX`hlCjZ@z?X2MO0vr-?cm
zz{YVH33wlA8(sjxpwgh=iHRt=h~p~*@eg=Vc6L{TX9s|1htw5eI)qE`-6mAxg^+Py
znMxIZkv;`@VrwovAg3;OzKjk*G69~{W(fv1AK^mpE=Qc?E6-%E;2@s3pcwK=cf))R
zk_m}*j)0~z>v5s!Jc|5Obtfzi^=;+G+Z``bt1wHhq)YIw(AiMurq<IGS?Rfleuc$O
zC@BPYH&QlJ_G!WqkC_VzNgzN=S0K_gBQ170t4C*F#m!jE*z>e<;&JsWXvy_3bf2&i
zdG0hraM_$8CKS08cTn9~UiMaFqs_AKV0tD4Hdz+%jK6?{F~sfCH|ajLiWyj_6x;u*
zatwXp)lGvoCGQP3`t)g-1rQ?I@#6+AAzMx#yO#Ipd-lq~Oke3M29fgW2W4_$M2o%n
zLb1B)h~SxR=)SI<6sH#vTh6ImSJaLG8{LQWtnflhGg;NBCtB*<n*q(!CRa=muk9<3
zP85xWyZ29|t@(5Kpp}!C)_8QSJfB^2lMQrc=s=n3DP}y((&v2X{tA3+1ni@igt3+4
zD3j@L8KgSpQ2N$SGZ-v&gCh*;Jn-LE4#wnN9keHD+Y6r4-M`o<oa}uZWUf!KXS+WR
zZ7};<bToef;G3O{7JNX%dcU-GEan|N$Nokv&Slf9O-#t5zPQ+}RA7O%ha@Wa7a1{M
zcwpxQy+~!_GNA92d<&uL*_Zt9LBT1H_5#76xBQhC?;LEU-H7ck((+zPRQjgO+W)0@
ztjn{qlp)f<Uq}yxufuzrkTI7@5^8yGPD+@k>1GE+=MtQgAPI{=ZwFBjS)c~D3G?Ok
z6L23pE7zGMc(vY+{&*oP2xB3dV^|qPL`mM4aRRBChAWQzk|m+E5Ae&G&_KV0+zEBb
z#T(~ojtIy1g5K`xmKD2d5f!twxld$cOX;DdKC#0=ZYG6!eWzH$1tcN)`eJYXtZ>jR
zk^vrYRprTBi>XyM#h*I!ZZ}6odPj~q?_3DkkbQ)^8GJNZp<=sQIoz8D?v5-~;A$^t
zjJeCn-Y>IY{)v%`VhuB}bW4H(SBsDa+Of9{F`3U-v5dgb5;37Ycp!InTg?5&H~<UH
zPfHa}vI|Fuom>KP^Qvkj6c9c&sTh6Gx8*{JQaxNG5RFglOOk?Ez;C~YsMD0maC1+L
zIELYdk}wplCdcB0Ru?OUP`QaFww0Hz27O9*_A2CglIZgEiQlhxOit;LvU@A5>+lme
zrag43iYj1<94WdI)MWBrfh6v3s1ci1n=0SENoU_W&mI+N^{%W=8bj|vG!AnGb2aE5
zXOWOWHWDuOAX_$cberYTZFr$mrE9x9GnI75W(g0eJ<?VLHts2-A$rwy2q;Y-#KXhT
z6v!ZY%6jaPJzn;mwfEH1P`P`<M3?Y~z06i6dr2&#66V8!Xx#R>d0o0G`1RkAD%S&;
zkuVrD{T}@Ji;JcG`I1%}4gA$)f5%2Y>Ro<*3K^c$-X;KB^Qp65taWOdYs>R$)fLnB
zxrbOeK>I8a=@!I+nW1D)1Qsf^^D!l;iwMTpNu*YOt_%_4yI*=}T1te8rFV`@KeKy9
zrSwyzb<c-cG}k}fWl6t(lhzzfQkOipH2u(b-NYlHNYB9@A!sA1n_*o!c}y}K+GVY&
zemTiaWG;+;T*OCN(uT|oUUPGS^fkbsSUb6)C0sH=?OcQ%67nkAS)NfCn?ITH)l1a^
z{oeW0UO@2G>ew||`d+J-YWnNSE0zAK1{8aKRd?2LP2jrer7fnTMsn~M;{&@-i6hx(
ztki#A>eBeN)%^<-4N$>x2SURxpnst6O1<!%jrtOBwx#FI0$*!5NtQKm*|xpK<MYyI
z(>|#)U+yv(K2x08o!i7CCoYL^PYjCJc95`69e-TYH{bgVJv;rbTBQU3nEyj@gf#pe
zT!M8&-X`yDBQct6t=4nI$f$QSv};2}{E%;W9O*pub@eM_e3rU3!{+Q>g+R{`2_LIq
zx@_mQ8`npUUYk=av#ra|sws27o3_U?a_9Ml-@p3%&LuE>J{aj)HBFo`oxt6^{?S1*
zY$AWXOs>o{S9<Y<uWxRux~$Gu#;)O=?sst3->&ueMqf7QzZbp1@%bbE{;%!c|ESLr
zCjW-KE&EeR3hgNUGm)GI77)3pgkvfF-7iP>@~#8j_xa>3S#n==F)tOWSsBMV9234%
zLA<n2R#g$)VbIZS@RlMAknyQQB?7H`TmsWVJC&)n;ykjgOQ21g5)-a^F4`$!KHWh>
zQEc(dfs^t<xu*gBx*~{UREVt(hl`*ry%fYeU2+X^9w<@zjLDsNj|bB}gGo>HB{bT6
z4tK?N35X?H)3w@M9fATc4bNXwHKZP%1QD0@YR>R%ZOV>ZF(V`**ei-@%ou0BBo}<7
z-$`Wq4CEY@<-XJ@OH^H@ducmHb^3$`fTbQ{!nJ^O6j=wN&{#|Zt{MnO<Eo1JPqbx%
zV~E$OdbK1FU28o-2CWG4JguLxUe9-m@zu>P*7Y(;K#j$LD30R<`9X`Dt16?&d&h?G
zl>$4N0z`=H!hwK5EJNXQ$WL6un#qe=;*O9#c(=t=ajt5;Dw+VLdQ$Co96!t>npLdB
z^&k3On#}Owbg1~9s5bd_<<7-zvDzEbG^(Ba*u+yh#;*->s7{&k)ZS!P*1TRr;b(Wy
z^@vh~a0RJqR42?%!f7rV0~xLpqG(+ZZ<Fl}ic<GqD8_~Ju!FKdEZr2EgV9`X_-ld~
znEi1<qtN*Ui2f*{h{9h+lzJ(=yB?v=#XK^78mcJtcHa^STk0s7$c>cq3S59+v`4)l
ziDtc3G5;%4zohQyXXl8~@KW?T*J9h-gXi0~5A$^%U#lD&$Z5Z-O7E=nahuWYUyks8
z^OfV`n8_hsxK_U2lC|K_vQOwbG5}%*k82I<={IVohSS;^jzdG@`|^4ir6_ssf!Ovo
z2`}Sdx7j?0XR^V@I}W(5YIKv{MV(irPFYEf_%RRIUaUfg#VY~JSwZClWsxmyu1Hi|
zq(~75lV}S4;Dd32i(JxxPc=h@)xka{ALY`=5YVC9h0FGH_}P3axFlYih`XiCueoGG
zxB3xrFiDIwofdDu-_oSB#`dtU|B)k8AZvZoV>r;sX%J>Ob9EU%_TbeoGV~x(uw(;r
zZNBiTPe8)#Z)f0p0YW$|Nl#{3MMB!j|K{`y<I$X_WJQ$*w0Mir8NHzB<at-wRKM@B
zdN-l>8!a?{@1Mpt-LXEDQol8#N!$so4AM}~+CI(Yo%nS6h;ZcHnd0GSxX4w3Vrl$v
zai7j-=9*-~xidRz&1Q8*-p>;kRVu@3HW=}7rn%c)7xXNYO^SR^ohpJyeS{9m=DoH#
zad7gC_pVP3rHd~XlIHADTCyjT<2Y*1>uq_P?i%+j>z2O2)j{}azN2;XX8L+<^3n-w
z@%o4Zzpm<BpEOO|>bUyd<&)ci_pUAe?(;kK{<3$PYK6HdujE+<bU!-$D_QyCJ;H^{
zCu=MIW2X48vj?yN{bTFfXf9jX4Pruya>WkKMdLbB9ylwI_pSqA82*`na>Rwwc?$4m
z##oNwp|qsQ%xi{~1cI`q0><{(gw*P@WecVboU(s|b+~t<At1j`n61b|oJ5X>YqfEp
zCu1?Rp2uCnR51;@BX0Nv`{WGg>CDIJf*af5)ADZ=iSF(LrX?_959rv~%dpH9?0dWh
z<x3s0lXRierk}4!T?uMyC?wIRBir)CWhjC%--#6;mF<$Mb%RG19qFWz5rTPrp<9p(
zx;Y#=SxGP$Lh}r9f`e==ev@^bMTf1(&`!S{K(?kIYSWFqdka~5PphUX*DIYqGGge+
zCnjN-35RrMQ-dk!ULZpl2j+@sU^j3Z14$lWZ3EE7o&ah53i6#v@GQg#!!zO;b_#+J
zfpL&T;J$~Ys>GvWOsi%nQj57NBq1d-hQlZ31UAMEarJ7Jo@tIov^q9C3&NYHzyQ97
zEJ+ptc3V+82@*q}9M1}l`&JX>HTRUo&V%U(bes}qT=_}>REP^MTo?7@kEfV<$ppbi
z-u;wcQ?*-zeB~N@3A&p5zco9I5lYhOkds1nr{R!?o|s3QXE1l%k304?g{UVyMsKCB
zdilOuI5}Tt@rH+}o1XZbz=6P~m+ZRb)oEi5!d(u7)q=Rb$}f(_>Q!0j>N^*0Sg$kF
zs(qTQBorPlzcNrw#|?Dtii)UOWt5!oP(0UVL=s&Np<!p7HP~i4ylW0Jc9>)IiPV8{
zdq;<{%xi*v9#3*i9n#tOY^4Yxj|Pwj(rs?DT-WnRmpt1Nofn(C-Q}jqW1E50v<0+7
z#Uj&Ty@oob_Tsd|LUL%6{55R^`hEgiMwL6cFe2FQe{oSZKN43o(wq4BwkIvfg6#=w
zb=EF6SKe@F@0;+=pS&Q3A^YwYW~-_y#EBOy+ca6yj|#*yhv6`PX;dOB#_2cw=scyZ
z<Tg_3l_PqCAw)lWck|`_rtSYo(Xo(NIU>IzcfHW>MB1s0Kx1n?HO~+3VRI`7zSC<X
z4#*^yiQFqG?7R0Wh?fI@^GvcP#kL{EWOZDKd@1%KLVnZx#^Ls}b^h`nLJd1$1_Son
z?5xK0ZqAMOu2r{3AJ@8ZOKwT^!KDNrqX%<T-sh?}Q_6BrceAH=75)qk4n7TdbaKpD
z!_#|YmPxGgj*8m+InxD2!wj#pI3=?Bns$vG?bI!f)^!n2O>BknQ-Sq{z{IM7ee)9d
z1w!{-<6}M7^Mdlmg2Zm`t@7_#t8b5nwSP*JF__h)-EaJ0^(LR}Yf(YJ!RC?ou(FPH
zxTqFpy@=nIM3*@C`=v+LeKV|p%+L2?eDQygGoOWi<;<|(Ub)^sd%qaw^2V;wKS^3o
zI@NUgACBt(v4;_}EVDiGXN@rxGxt9GKEAe$rT*;h>yA7?G<*tnD~GTq4((gePcAp%
z(nxvh$xNgI1OkuZKn9WTwA1W~ZQ@(#oC)|n@qDtoi@j6QL^jB)u&13RFEpU}U|%_E
zixo~$9R_KfpZ7GV^N1ps4aRnPi~323qUTh(Fj`CqSn-&Is+9vy*Y~KE=kH*d^epst
zV?n~s^T35i=?{e$Q{MV&gZ;W|v@rZGK8@eJj%^?#U&lw;(S-)yat0F)S@uA|ZZmBk
z_1%4eXw87pc~)R@;D$t!OI&Q$?57*r3vfx3@(>o#)%}?ng6LjIDISUp38hMjJCUic
z>6T+V_?<~Ypbf<!ghK%d62jV?%<O-|Tc94*K@)`(6_8S&26EF~a*==ZY3UrpEJy8F
zM~c5EPLPQf@hIV^2g*^!BqxJF01?VrJ2e|#l9fhxH#EdBc$&cfk3@WY(oxIUZk0rj
zj9KVyz{=KySv1+MJZ-LW$EMGDy5{nrJskHF%J06{YuK4TGl*OBc~yu#Iv4E*yZN%(
zel&W&6;A<9K+2<oG<_B?*VflWqC`tQIeZqwjO4^>9RVY{BV0m3&RKD)@_3P;tG*on
z>MQTmR9UT>hbcMmtv9DuFc0cct~B!%oY1Goq;m;NUTEYQF{htRumFykc|E_!K3+LS
zJGGWu^~i2^A8M<G5a>83=@=<TLfoe~8p^h=3NUVLyYzD^+>V)ru)QcDJ-i*etoy;7
zU`QC~M5|l#S9e9}aD%zUy#W#4SDragqlr%a1`OsQ<<8#L!t_0|$}U$TqSXA<+;wr$
zybgKYc56BaQrQ*HPYt5grTgU0=vr+siG`I<)Q{OGr-%RUX*D``s2g}@+O;aW#nR2M
z+^ox7XlHvF>tzE#F}>DV(wStRid7d&YUHZ<Mz$KelL7({8+x|iV+aVNKSSYgfuP&`
zXx~zKmNqS;53PXFHfOB432<B>{pm~I=YhQzy}vT`ziTc3-yW&|_?e)D+rq=m%S-zS
z|8}Q%Sb9>6xSu3deq011H1~b$>`LlQ>F+zP;-dtuq^(akR4&8;x17)@O4RChQ;n&G
z-wegD*iGPp+(MUyeHaTCep=(C@e5|qpe{;eKGlE)N~m$Dq_1vad**X)wPkuZCoSd#
z()zhtI4C@wSN=g?o)O@v4QW1jKeZ^bp(Fdlx8<reu6Y@JA8wFehXG^_Eh`0`hm6Bg
z`<fR!dW#WB<CS6#Ika%D;RP2AMd9$=81xUh8tTFvVj4wSSl?`%eQ~4xfb=UR-BPyl
zinh(Y<`A3wgi6nY<(d5Ge)+{xirqNo0Xmd+znH}*<_|LB$rn<Wk$kC87Nf=(F9?7`
z`JHgepR%(s*!E`V=A%U3Q8DT0)9`_Q)_jGjm?eS#HgX9Us2o0P2kFzU6odsh#gIXn
z*sah^!7d!yM<)(}STOW!T;O`0c|Ejp3`iU3`?~lbF=Tf>$;r+ueCd9hcJmXq_03jg
zN{{-+lJ+Z9-hTG{)Vsf$bIex{R(?9U>GiFi0x^l=?mcs7vC-{CDSun)kZ_mRc2oI$
z48tj8XjYAnF_6e4RJ>y$IR-j8e7D8#$=OC2Mr9C$w&kBPo%ob@|LV}ja5~jd9Vj7s
zlv)l8!}!IDc)PL%h?6n{WTOwRmtPBA9{NO>US-tLqvaP<B^J;!90S;71?<j6!R<2l
zzSXA|x(S;?8gY*gBN4DT^CtFHm_%1T;fc;s6CniOt#mADvIcC__J3S;co1{)!3J~e
zL)fdjqqz!Q+uxkk6H?#OVfp3VZqW|U@q8z%tLUvDEg7-j9G&H#*R9%+dT9H!@P_R6
zpT1O(H)8j!NhXzdZM$wE)jysg_Eh>}QeMLlhonYp*XLDJG0kd)&vR$nWQPwq_vR<u
zV+V$g_$?GozoL$m{9f%K)tP#G@-EY_ISSF}7*#H=OEX7R&3MI=ZJ;S27F^k7m^XlW
zVAF``(VS@KK+J?LRVOmtGnH#gzBsD3Wv<~oA}+wz_E7*5=O1El0Dc(yuM5!s&pQab
zK$zTg1gL3zW2M5VUzNcRm&u7NLtI_*1N+}vVNb!Hw(EMa&xK?vgBrJuWmSY|phZ&)
z^B`%cEuvlEf>M*q0*jCd@9kT`)rc-oPeMQ|LI-~7J75+b&5Z{wX%XX9ciO7gSybx%
zh_4qu&TT3@jsSW+M_eCI3}Wa?6_cv;;>T*|7oK?MpO|r`$?qA*Kh7loUgZ;guZ7|3
zy4yy!fYs_n_FfnI(1^$ScJB*2f#`q)0>!O8X=KXLMH*EI0<?i*l?Jw1H->heK|YVE
zZPlh}HnvWTT&JaU0CRe2lqg~eW3hS=VZcEG&p8`W4+Xu;dWtR{;2Q;~s3hR~qT>U@
z4P7z7LdysYETanuQ{Bc4;Pd8;KpL<nac~(8(U^hi&Wd_%efU@!*u-`r(OH)10GT<E
zSf@$AHL#RRvV{N88qE}Wg}hX8pdo_aMnaL_86Z<!!VuXE;i!-VAUqy~bBdm{+JRm5
zu%g??JjEZ&-fp-q>y)jp_FO}S=*9^>YA%*Q3T$H>oxe)j^NIMpDRm6`)vUopV6C77
zXRfk`6xhyn|GxQ7ktmF}mpQIWh0qO4AxTf0e<-M5cRau3=J-i&DaeagdHfM_=qjS;
zLQO?o$UP!esA8MN0{cj3@9dZEq|%5y0j(P`y7#3mC$9~6^YfU*o1YuoG0Ze~bJK85
zNg!Q>@Tm5%K>A5w=R9%Pir^h&)0BH*Qqs{lDtUsB+qGaf>-#q4BDp0ST1JcXwGpb&
zA)unY_Ydu6JKQyb8T=K|%%@Ua+eV)hA6FS|b&3&;?Ql4l*W72(7*dPn*o4c6K5xT6
zp+trs<4-4Qj$&(*VPQnC0YmOl{!N~-6le{z4YXU%Pj@q#9gGIHA7CRj6!+KS>MfiQ
ztNwYv&18iuzmn7y40i_va&Dv_EYnCcq&_j$_I^XmmJJq@6f_w{O1+4{@@Ucpy#*2H
zArcUwqH<qjIsuddv`Y*hqqBA=0i3>2v^!VbK7A^$ww$uIulsL^AK)1NNrG%-OJe@l
zmh=YRp9-_%A}9>G*-CqItyM_4&`;SezeoObH<$c$qhB=ci<Tb_$iIjT+yNh!$`aOq
z|Hgqd_Dm&|poo1o{-nIDz0fPKFx5Rtka#R-_eotLgUYcwDkR}mtGjvbm`Id6Ve(UL
zXLk%YOCuFCD(AdMEq2u187EL%j3}u>!A>h75ffabKh1KnLdT_HHYYg*anG9G=@@?9
z%oS_fHM$&cdv+MI)za5gJXQf8M62M_q?D1?RE8AQl<I~ZVrtVMEWUDaEILa|i;ziL
z>Ky#Ki}tPoKK1rN%yTtJnoUHWq|fU(f83qi2<xd(7|Cr+Ef#l{OaTQc5-1xp0J%&}
z5-tNeF3nP2Q4G6utWaty2&xK0mq>_Ng5R=WExn+L{?X)@h@i;lQ7<zMMJw0v(KvDz
zd6KG-fDTp=0Dde6CLBb=YVSsEvnD{(Y#Ob-5Vt|ljnza4Mgm=%dOm@vM5ZiD7+~|m
zIjU_nF^;F@9$kN7_`b_?H?pX<_xN?5%D0Fdi(;-;aIlTxK1HI`)~Wf{XP;N*KGthQ
z?3f$1fi^#tZfh^EW^8-s=4{ICDgNX_$7Bk_*SP#TlrzgVD^JfTUph6CzdlxR=e^QF
z-{9?sT^I6pez3eyivFoWKUOm8J4ko$F%cRMdE{De-&skWaDtgH7S;ymJf&YeqNFR2
zi|MwV3==5CAq>&Yu^c|lRd;Qg#LJ-ux%om|SdxE0xcts_1pL8cFu$LZ*cX;CZaeUH
z+md%xdERNYthW}zw{czVL4!(7*Zg)}Hp&I}lV&R=-&RyO-KS1dvg_d^w~n2YuZsUf
z5|t=UJZXZ_b?ynVH`wG4hV$7wX-^&np6Au`^<qGIe;7UvI{=Am4DM3`#biD6;Re8+
z8dt}Q+85wopUyaSGfW`$fK=$Q;-Z_e-a5yUry4mIU4@o#!t8^Wlp*!QSPd&;>Htcl
z(637{h770WLineO6`x4X!@L&Y6#U_E$a`|^6ot>p(339Er%1fFgoPxc-wSDNA^nQ?
zWR|wrW3#NPVtDd2>!=?g8meog%J%2VK=CiKBRl2Yy%e?I@Ra+4uV4QBd}Jsk)A8Ap
zN6Lf3Ydah!*2%|%+Pi(z-&_~g?xC~VADfUBHm5SsIr_G*{N>Nh?x^SvHBCOfSct~_
zmMixAg%#W_VCyq_>-1;}YsZe=_3rRtv1tX7;`?kiC$?wM?BXsf+r=cscS|#=teDYD
zuTfnpu;(kT7DqaYuj^O55GGbWIj)CQ$SP=xh?zsEw|i1#;>!nrB;}@TEcv$*m9HOP
zPfk3e5#Q1=q?7eJXurU5)y>?nje7h6!<GXN=;)+x4d3K?O+OADl;qByU!5d~9-5BJ
zeEst@JAoW(FDaf_&}{|z#<Eqq^jo@y@$<69rAJ>)zy1DZTheNKNQ~4~k`&hx8v^^C
zgjK_8Kfl@jG`D)gtGxQn6oJjzxFv}@7~-vY`kr~q9gT?qp2@G9;&E|QIjCEPoHhR_
zWP_p48hMD6`e%C?YES$Jy_4z8Pye^4dzK{99Q{Z92(JBw01MDw&ffHC0|up?ll}|T
z1o`*jnsE~ax4Dq&l0G}hM2m)Ra7fP(i(h<@6fSV|IXUL61tYL&f$k%#AtmOYq#z)0
zG|Yne;yD-MtU0R@szcGc@TB#{$E6E<k=9~R8`#!YusAtr=ZUKkVTc9UjU{<OZ#{W#
z^zw19yo&A@@&$d59K@2V&oS@GR@=uBgFh~%`VS9VSQiu3H3tcTxODkC#FAUUR544P
zzm<GTLHH;Kgw+B?@NNK+&b-H7mG~Mp{GQByn7>%1q|5b)^e%XGC&ixYEqlr0#XOMx
zf)oKmZ-Z65Ab>Q(0NE2>p<SWoI#n*ty96W^xd0^%dSYs#RM2i?G5B)56krM<yJiOh
zJ|nOjl|&SWqASqS;D{<_w@L6oB6x|6)3yCML+JWc5s5t{h@0mS6FLJ}JpS{=r`56}
zNP#l&WwACM&_j?=65tscR4d5x{B#HpLn@~=Yn)$hX}UhE)116x>|L~D@CfIyZixJf
zolNkuIa^P}>9Ik6R%6k+u59_csxH~j{BD(hA-6wzM<-68VKJ{JCJ9DQnxiV5Zzp}v
z3_>xH-|Np8oAzE=J>rz^LmAv{JYBdy2t8`L^7_MULD2(Qr+ttkl9^;?o&{ZL$Ae$&
zzMF!ZK3C3(OZBBBmkuVC7Dl+;zdf<oUfzQ0`?UX_c1P&@gs4RLEpJ0AlqGzkOI369
zgA}TsvNM++2_sCa8DUx53c&|C^xREDk>I(L>UcO7WAv8wi8iZJvX>R9mj!7WGU+@t
z26)vR-<)E5eMvfGrPrCewX7*{GQouF89~~mI%r~4>|5g2Kux;tA)!{Zt+e(2EKT7?
za&Bpf&jYnl1eJC`4}<PDvDP%;qFn=Bhvyk3YwPv)YWy}$@0(+*pM=|4Aok)Hx=z-I
z5r^4<t_GpgF3^)_BTdhQUU%}*I|aYgoICw$(l|IU&B03;emfjd(Z_m8x9|B-NldKg
zOLwIUnV~;EQI%)byFzhE6PkkZuXs`y_5YNv{%!e8I;ZeAq>%dP_l*?%IN<53V0cqu
z2jBh;k-NLF`1pO~X!F#d$-eU|XM#pAFbv+DEtB6O>wmlYz{R>Rt_qv7J++dcbg(^H
zv)eMd_<>%-Az!UKT3uhBKK8KCR8CQFwXr52R%k07Sa9j?y-C!(_Ok7f<UOoV;i0Ib
zKc;M~R!fVYxrTq?Ld?|1A2?|$MHi4R-#h;8mr{^@py-Y3Mh+ssP5y?cU$#3pJ~r)N
zf6+GQ^{|25;*(n&dtcC^={3>yAb}zwdEZ!wn&#w&bjaW#>a+^4(23b}BIL?nnXPB`
z%3YNI@+Qq?2Y0|iN5f(((r7gKkjDL)H>K^Ic=&-=7~I_D50}pn%xds1Px}vx-o10H
zyE}jRzW0|!s==jHRax`Rr)ITBl*|rx-V)|RVfO<3BForx!|l_a<IdR)Ym(gvF>P-@
zAEl9U?JJs|DUk(oG?;7s&;Gw56K;Wp4TY5Nd#9D<Iwl$2S$?Y?&WSp7gI`Xd!QL>p
z%W^8UekEqsUSj6RE@{M=?!&ttCBt8|*G$=Oey~pE__>B3|2@7vZo9+?FnK?zneVCj
zyj6DNp`ZB8mj8#cHxGyU|Ns5RzGOFqke#uVF=MC3K6ZvILlVLug)Av$tQmu`#)Jkl
zc12l}5{)H|Wk?bdMG_^ny{pgf>HR(DI_LNMUB7dF=btX6i>oWI<@tEr@3-5AL%5<^
zP#1({;vZ|$=N>nwd*ZT(??3G5W>OwL(){XJuPq%?odWFWZ|rZ=f0<0#o;6_(S<Pcx
z6TY*&gmUCROStKMWwY*{eRzUa%li}9VF~_?(U0K*^`pJ5pviL(G>7V*$o~s?Ue@7;
zkq5?M2^w7~H0NOAP;xMF5Zr4?V=Lcp0!ZOlCAVQ{+Oc@;b(;E@D3stz(Is$}cEi`Q
zim#tn<1Cb@IMLocjI|I9^VNoms3-RmB6axL5V2)xMpu@XD~svIA~EQoU=p}J$%NUP
z1AO*e8HF5_e#z+c<H87@qlm9z)r8#lpQb!Q{li`=gy(5FozgIF34UI=vQ|I^NlCP$
zwrFL0ZHo0eGVQ>tLS>Vv#%DQ!u^1bcEZto}`YvW4UDMNR>k$^P8FlUF(HReapsUwe
za9C!p;d$oq3~h0Ol^pHBELg<*79~-e#mk{BUI;*zR&spc4VniX%7Vi|`Z!df&kidT
z_VKZ80tGJw*;~)4g0RxzSSv^&T>?-%KDT60P|8abP-A&)=Za<{n_u2Q4Ju7%sIdr|
zssZ$-UwivgXfSE8MKzJiVxJ3xrysS0kQwo(haT-u^R~O&1k``r%0`(w3I;3<2{VxU
z&BSz>uz_P?`I=>(>>fEa`NC{f$Wg5Ud(k&+<ZPFTZ|;LLdFv?YNk3fZv4Q#q7OKL2
zrasTGo8NGi3geiZsrpn=jug*IeTp7(VFxuWLc%W<A{}b#ag$3!r;0*k!f(v%Wkm!j
zv10BAPX))X+7FOZUFr7{kmcX0PfeX?crh%{WLNay8ilHT%u$gqj;bwbYCs1#5Dwwa
zE7OuawQ_DhUW^&v=iy%0_}y4T$#0Xy^k&NKHNj=oUo~ne^K)M>yY{8|-JKjz|B%pC
zC<rem<~thbBQqSD9euj4@lL1C@a)=?7lNWMzN$Os7&CTJdXNw;Q;p|!^eiuTu~#I1
z!S(O!hL&hoIz~%1&}VW!dy47>plix!t)>Go9Fb*ZTq+gAcGS_tT-N+dnhqy^>9(1L
zqrUXCXi)MF+8K%0rjR+0;06gO6)@~d@q&bYc9$3IkYgxV(QPmM?fp{4d*+bVY+fc6
z9lL3(<X?Zcz;@$FivWP}hk-U>ulau<l(aJyN59+*J+mj2z1uwAR5S5NIjkhHyz_2|
zkzErhuCjRGMN4kjZQ;j(E5+G%jtI=O+)T@g;SKbANmIO-c*}~-Om$~R#`qEYBf$$L
zQw3%xsFlFt)UGwW^4jO5-f4aFuKN#45Rzr*g8NM`))7)Bo{o@x6C-K07WIQX&y^Qy
zY_{^}v$^#1Z#AB=s6VaQ>NR*ygknWFW+O9d>THUeS7|zy)nrz0iJuoONL$ucWS0th
z<5Qx0TxcQeg#6z*HHG}s-oj4i_{ref9YI25$ClI5k0CAcA}YuA9?N))-Qe0nhkkbR
zXJwzJH-cDSKZw3VkiW#Q;tydm$LS$X(|4MKru};|w$JET2Dk@KvM{p+Z~M&Ux$UU+
zGH=?qO|MLU`u_P+PPXKcPTOXqpstwfUOTa~ZB^bI(F66JbBl5jW!l<0YO-2AvQQ(g
zf?gg#1?*mrj3Hv$e18$AAKGcPKmOGF=Y)*;)S0U|MKO5g4}}8If}M!}W^n$~O8dS0
zQA%+;1!fm8&Pr4MYm=#%**C5`T;F<8^-BSdOO7ZK<9};dCd@8>)ZcXsk)FNzdo1)*
z+&UykI(8+S3odq-ppfTm6uVt4wpDEyim~Nw_x>c|CcmQ@_h7NJ21oOGO}n?{r5HHT
z6G;VM|111Xu)e;|xF;?@J8)W8Mu+vki56-a_6|(!j^h6ni*V6an^-jqF?-wgABYI^
z&A%(oc~e-v))TwiEyj{aZr<fxJ0~ZgCS|6QDTk$}Z}cxSS<2ZyhM|BnUVfSB4}t<G
zBe{@F=N4A>08aQs%W;>`)^X3Lo+-9h(@c(I*Iye_F%qFv7AuV+Nh*>}1gOWctzEm@
zr_jsN<P8)A!o>=vK)fkmqb@{_Cw^-@S3ah<rnHZo@|HYg#dd`3iUO~8E2YValTNtg
zBKKBa1Sp($n6u5^!R6iCHylc+f_531F~q7GZnfAvSoFQl(`+|>u=C)TsTWIID`ImH
zYELPCb{L5PuiGhjcX=EP5_}W1F%*$5@;Eh#44|5WnNj>hS@C$;Js?vD3as{wryEu}
zSf~`gJO>C;fQ=|%bT}~gsjRt0?R*Zq)-kB$Rb)UZPdm5;XJY-@1^V6bC@NSSSSK|k
zhd?RQEa(~n%7^(f_Um{0CEchwrZKOEzW0R^-R}&MxeI~vIr@^#H#q#l-od`+VX48R
zZMhw~4{v!LcXhBXZe!K49kOzJu`Q<giY=%IIyr^!p5#}-NUNiaTvS3g{B;-nf4ywO
zHDZS3r{6bOV~+8W&4<d^Y#G{ZP(H_1X!-`UbU3joo;@C36gwV<MLMj;o|f4P(y`__
zWA|)4sY&tD9*1eeAe2RA+sKXdC5$s|oI^2&!6p<-H4T*t5YFY$P7y>xEr-k9eS_<{
zR?0aY4+~!#|2*oS8=bk=Qd%y=Z77EO;O-r%Hz9VG(~!z=6Y)Cqdu+B!l8dvaws}ki
z18F#184fRiHB|XayL3rjFu^1{UV@+>t0ckd6PklNgw+Q5q<FU)SSZ^Dwa+*?ST$Pb
zse~{QPvG~pegBqc)3>BCb?aE&>>VXUWpAu+y2-3&kJbeCl|&mCO|8jNfT1W0E6_J>
zcm?I>w#Gr$$?TKAF;bk#OujXIZcU#&R~U?UB?Ttf;gWWU0Rt)k1fhVf>^ewQg&cU%
ztNwovUR`G%{~IN${;%bQy?sBGp#R$-ik81A|F^icWmwFzkM2q&!JgTyqVB+0azo`@
z<jwKpp(9*0ZRk4^Zb29jvhd~8S#*{t3*JD!^0nlny?NNkLsBd}gp9B6e6YQRxdxQb
z<#p1i#t0xVrInMdxN$vnIRIs$0^sdj0MV^SPybT(1$8z>!LlYxkL*CN7LI1q;PtE3
zLJ5BDG{AAuxwS-j?i*zdBSL6k^epoMoSz7wL=B6<z<H<?4p5x{86a|j_$wp*DoW_b
zqW5O9MI8+dKCXF5g}(r`X488lD^g3Q(JG3{B(SqHc{#OUE2U`+H7d(2;7lh2mw5px
z9omOWK4^G@)z5+0ydS}U0`N=^_!ruvk_AFlK~0Sbu2w!*2M*4nndH)yLMY^uG-eHt
zr$V{8KihG_a(Ho|>51X>KNf?Yfv`D&=({U{-M}?BRBdXiP+xN?+b1?WV6~Oa>(^l6
z#q2pPL4b;gqOwvljM?LxCIv}u2&Icx+|Dj2b9Jz4wf`Wl-qBF_YF*+`S;Q|Ca8tHO
zJPjOOc!kPZs%;)^3D)p_J9439R(B73AEw|xBa0)+E7SxWazP@QLNF*igTr!Ra^P3K
zy~y))uMpj_O({`%oq?0QjGBb|Y~c;^8Qh<6&z3qk^7tz`Orn-(MeHj@{DZ8dp<Ocf
zC?zyddjsukXC-XMX(lGC6Gj>qfb}8VEdnxh>jD#|NO|9{7i+u%y}!9LCb?EV)^yi<
zDsIdIudb_Ov-b-gsHGY#&(x#u1du-MVl?8Z9Pf-tQlzNUYFtTg9b9wyLtHrC5W>;n
z+S28_mf6Ks+|A3YP!*OY<-+5P+jj3dRy#e0vGv=*<K}O_I2pZwaT&HumEZUwgBz4+
zL_h=bAGj$K>jdT~9EQ}-4qQ@a1Nb3ohR~ihJV5{eX8--!U=_BlgxzqNxgc3Rz~<cf
zUo+K%NzjM?qnYY|AexVlZk%%4^d5_;KFXtt<>xr^S4MJL@XsZXwsSuZaZ}zlEKOf2
zrwD^V9a4S-$q9JK3i;NKa<T#_277K1B)QU5-63_J#8Vk%1DTdsUhJk$Ad_vx7zr6$
zLvoU@QR5do9rZa$PiqSq@>K(bf|L|B7WFG=NBY2!Q3x3)7*p%cjkiI;UNm&vxT7r8
z&(uHb`;ZuvUy*C8D}uOngzU7hlufVzu@z)(X<AJSE1#4FYM_9`V+STNTMis`Y)i>T
z**3dcfsqg0f3)ytINYj^1-V_LtpHPg%Nqp%s1)WZPuGG-)Ic+|R1ofR2OonDl|hYR
zBcz%%>1vgEb`Cm?X8#PnVgMLhtmJ|ud^@Bl7@uwHP;yL-Dj!FH5y2-z2v`&p#LJ<X
zWKD=6Z4-59!y1=!2sGG?GyU4rez~AiD$wjS3WNx(CeqAQ6yv<`^;!M5*!}RpQ>~O3
zFLD`w!qS4PR=ProX@l{MWT675ShUhl4jr$r4_C^Y{<$~U!e0En((JQ)$VpvDzWAyW
zKV4z6hMaF5<p74=qOp5{e;W6_UZvp6`A*g7*{*5Ry!)(lAkU1a4+&J`dHSx9C6hm+
z__Pi<4H<}ls=_r6G{Zs*!R9`dmNq8x`^ChlsO5Wav@ZV<tE}>91Dci#!-kXH?Ewn|
z{Zvr~u*G93LIc_`0pm&6BQdUp1Ww6Cr`H}g&bIHkRk8J{Aj5DkGfXtTcJv&Hgye>1
zPtIyqI60zQ7<}zIUh&}asAA6*#IBR?Yl*-b+}XVG(Tk6K6Wy0*cjf|%W};(PioQW)
z_lDngH>cT)UVS)SmO@}ZRs5A79{mDcfU&zHDE4gg8OI>o!l`P>A*U$$#(UDdi<0+5
zN9!O~7=Yi!z*j2Lkh~YaQ&@62kg~V}6Iocx9)+DXhr-J>=?2zhJikF&#*k*>o*aUV
za!&)ebWkr*?6W<8+xSaWq{Aa|-|FUf{{sa1zy1iZ@0XwCyw`|hfNEmqfCE>{ai=ir
zODB#T-O*I-D!opNs6^GC!r#v6{*4=c-)PyX>euei;%-IfLQ;}JM1AJ`#6c?}nPt=Z
z(?vie|HHu^MGoZeHiQym81i^IRk^i$y5U~y*~*tGe3HnSzoMxweW@HjZB?jL&u~L7
ziTZg?*O98+C+~$2sXyI#XNr!5DnbPYow4&k$x~I9sfs@(1C;aCgvpuV**k_pu<yDg
z4MS50T6Ymf$=4Od^9K58uiwt8DDAiki8!gv3NAiDXi8}&!5jeCqwLX|my9fLeHyMe
zXnIjS_acI`=gQ2hSh?>Ehpc=~gtfXWOXq=NN+J^pas+1J=}|_Ja3us~MB`#m+saxT
z94a_SJFofCC(ukNpTj|vB!L21h0deYXa>MupathWiFIS*bD%Y32-ch*0(=I>9gqM3
ztX`IaxPpXy5S#swnotmC83Un1fv+GqSsaf7t#m|;m4gp_Ka5Ea1Eot|joS~6@4UN^
zh*1bKO~eF>DT!KntQ!21Vz>R6)F|=pSw1R59CC_h4w(sMRrrNzQOK8*bI%%uPL3B2
zw0)j|de%na-HGqmcg;53gH>=wF09(n0UiqHS0kx#5g3FyRO)o_JSbVaA1iahxLETr
zKW3`gd&2j%2^Zhv98a-Z`VPK$Cr$1ZN`UC5J;??NdA>VKe78pl#c`2ylEIczS}H-2
zQ0-SOy-vtK{3y&D-NUkl?bNqA;qgyTgVXcwTf)VP3N&Zz6j{N{bxh!MX>5JL11%v|
zGddcS;SPy^?lk-C$-+3#pJ9}Bn3vzZlLZI>HY<yh1&IC_Q}%m25V~;rscIWB9Z#*5
z{(vVki5~e1HewegI%CRT*t9szr9$5(hvoBKtOA@Duh8Kz4mrNnR2#`!c@!lN%OfNV
z(&&>cU?JREO<K!W;B~`0SfEF1pxlNEG_aGyo+z4qH}Uq|gQ+ij;m7)LN9xaoJmGrG
z*}RtfVQ)@rFCn%J`N!Ulz(4%`c&@~$is*^Kd8_V2b#$%g<u)S3f^bCR$jZa`;F6MA
zcC2!90W8EyYzQ1r6K@ModH;wTdTpQQ_VTk*2jTj@ka-o2l(^e}6iaKeptL;1;0}96
zqkX`o4?*p-N1@2Jm&?w1@fFJ@sqqe17yfXQ)9Z|CU*E9Joq^E2VlgLq^}4=0zehnl
zOwyH-U}|i&t&@9JU!Eq8H$`2owom2u@y{$5eY4^9i;OeBBxP~>q0hAo?Tr}|o%2@6
z0+^VhB-iaQ-qw%QYtT_hb0X|yx_x-Wch7vE<IkAgF6a$#)5E?*-kj3)ZkBaMK1D4Z
zhU{PyFJx@pLEbVR8-1T^zuCseUHbldR?ktah3o84>6g09$d2qB5hid-+yCg|KqyGe
z;VTzjitpS*Pvw_j%;?<Fwh_n_BO+!iA_O<gEaP12rOnTMm{(>z&@TDPc{wW}+Dyh&
zgAe=_7ro|zgLpjWwf~}UBRQ+K4~AhfhO}Yn8R7>Mqq%9lyq`cb7LhvXc0|f6QgRa$
zyXjD(YUk3kUf-x(^${^DL51yN4hhzPr&&BhH*4HVPT8+W`Vs}g)|n1YOZUE9W&|$v
zguI@|&rwGZ6UZv~eARu+jGxB0>#(F(15rg|n4UO=^QIEZ#A{+6r(yB_e1cd{DN8{b
z+Q5Ue=|qK&h~U@i<q_t%n^7$@I!S{{II6CT(uOqMsLl+yrE#Ms#1wMC9m$fyl}*IB
zvJjrrZzNEnaRtU5a!5Wkm1MRJN0UOdr<m0-qiGf%C=&-MsxCcL?)o6#jBNNJRQP_?
zBM<Z+!<NLUh;R^m<v*);di(DenpTPrRIMHn5L*MNCCm~Mn53cT)LPalG&^P2-Cl?n
z%g#!4!LVL&oh9+BlpDFADbpGJ3X}fKfMiyf8l-jw?|Z6Fz!AbpfWXnVwGjRj$!wd}
zF=%!w#eB#5V}KCmODKT*3|6*hNPxv^U;PuR+;5O5%A8Gx-GI$8<kR*0AeRy`=i8jG
zzPJ}|)pl?0q5;>2@24Id_<_SH1fuB1vNJcNQ#qTD+XX1e9NBOgcvwCu&&}cJQih7z
zLXB(;&~gOnrd-JU(*j*jMmXlXuD-nfa9yIsl?a|2@8$MuQE3E}BLsjhz+plPTMCjq
znS`7NOvxj$wX&fwu{-Bi%Ft&SXZ>(nJeoBc-lJ0VvT?-A6j<X^cPRzb0@xJJt&xTd
zIdatGE>Zj$<uj3b1YU(vhAG;_tvhqogb+<|2(3h9NHn#z_Sc`Q4a53gPVhJhYq2zx
zWVm}*E4evM=vom=<gT6DJT<sW0MB#Q(AQkLNj`FJwd~2e7|~*djzBr~LX4o@ny8c$
zzlCN+I$St%^xbqndvFVP8MoMY!ac1o>yN{jDi{ORk6|@LJ*$Y|r?kJ9o*H0ubtIPd
zJ_RHTfpC@;T#%Ez#~mEy%w`%2T10g;wr+ewe2@#s=b#1Oqe?+p3#fMjsL5;p4aL~;
zKd%ouV{9%Zz9GJE(>ui1^k^$@+x*XRxac53bH79C<Kua68`9Tuym!U9quHwZ;t3n6
zesSFkh-s7mnw!1&v8^A++^%7t?;~qQ81co5rB`_`oC(*P{Z@^DUG#h|>1S#~lVSg>
zIHh@4apsKM^s$ck83MK^;o5U?wUO<*`8Mmj@io6sOw-z;pcBUUW0Q7y5lpQ|w7x4(
z<7b6G+>G-({qR>BkM2yC#Pd2l02Nkh=?q=C5kGAUMVW;j6O)YYeU{pt`(qRJIqp>V
zi}R7~w|?m63C)N3tZGGO3A-KnHY}qq6LR_y^D2V!dQLlQJ0iKjk<xZ%d-rvr)4+v>
zD^5K<Q%4<I`(_<`M?}w8bkzIz-(EdY+x;KN;n(!=6^6BhkWx5&0N+2tnFfov;5}Mv
z^JQt_W2>*D;Op06{cpYW1Y=&e*u3SrG_-T=kKR(k7QMO5b^)#`Tb<1OaK9$(JNDs&
z%7(;-D3zO$j1bNzojT5E-$+2;@NdR1sTO>9m<>AYu@&)^Gl=ck*~t_xDC-4h_}%8E
z&?N{{mj2YuR<njNe@Ja1F}OL*D80Y;%lLiK8+VTfzB0K1S<+{jSVwJxWlLU8WUP!2
zol*1ivyxv}v+MF%Tj(FVZN%zn4g1cMALavxSy>kE=^vf=>7ue9+&ZgyX6SpOdm`qw
zbF7B>g|pE??77`0f&=$1g&Vi<YB|08m=zc<_9S^h)7<tn{==i9%t!UJGtJGhukZ3P
zkU7!Eyk~8xL`8?1eA3AY-SJJoW3wW^z76PmX%2nhs3&rq+W0wbUf5axQA<<n{g<Vj
z`gVowwP&N1N1Q*G=O*p=#WZ|8jHUhV^3270nymjwO6PW)Yp_B9zXD?4EdO7wcyQH$
z>8D5p_H?aK?ah}DU>*G5y9khS(5d?_RIiHM&37n)UF?smF}A@5$mS5D{6pClI&pTX
z=(9xL3vSF5TgX5ZnllFF<-3yFzk<(qR2hkn*Ui@mnz@(H!xek7AdFu`wo!1lI598G
zQ}brR(HxtU;mTsI+`-=+hst@QpK4oAhIV^(1qMo|jf}8J3#h+p%ens*5`L=9HS5sE
zRAog{(^0SKx#4$1%c8Rtd(!PTrYdDRAtt~`NwSsvN&0(Ajh-*tE&wZxf}25~@YoYj
zU@61e+E~6i+ws~uXi945t3ZMTg%4w}-l7i$_wZykTa@ob+DL7BjD=O<;q)syvwyI2
z^MYnEUcWwF|M=!hwU2<0#7|S0I$!z(2ylrFBBd<ENucw<fI@;vYsvlsY7E!Iq9<m4
zCr+{oz;N&zR&uK?a!>*X{E!`t|Il=wJW?SIG@G;yscR__$L)M4;V~c=%ikVQDmMg+
zndo3?En`T45-{ABiT8rdLHkirWM4M4hY(T9fXccC3@5d$Yd1|WEHGOtD=!yxNX-z)
ztvF$d;3bxOBjV-x=D_(nHr=*+T*hZ(K4{!lE%Y8KyCHyk7HS>Tr~~^5YN$G{=q%!D
zAA7!oUKKs$q-G4dT|3H&ohh-L&>*yuW?XSMW%bW5_#PvkOAIG5ngK`YT5?+fTu0iD
zgDDQO)IM~gh#=GjWk$pJi23v6gcZ`%5xncD$b;1uzV;x$4f%UJ<x(RFL>qn}R+!bq
zogEJId%>CF@4t2{*^mO!hR$9X$`*L1d3mK+@WkFzrwTaPFM~9#49ctT6lwNqy@y@5
zJh!$xWm=xiw8$`AtBFVQmu_QAB<pF}@z49hl#f*9_!;dtm2CA~-txA%sOpAE;cgj%
zKk`KOCX0=KXBCg=u>6A)QD{Nen4D5k@UL$D?cS?Zw9=7Po_DgxwXgA3ongzT;E~14
zwO>s)<}wVORxQ4$DHB!uRQ`r`r(Q%>WmxmWMFL2_T;#hq>`7`Yhz)yYvM6#n3N<V(
z?4a1eDvDn!7f=%t7gK)x>+RmZhcln1U(lV`;c@pJ-YVLR!A%WX1iZHMY&u?EMaZ@s
ztBGw$s`fb9%f~5nEjGk^M?XC_+YQ=<_l+*#JLhOs_JSe1F1c#zH2!yf$k^vihlH6&
z;B1c(h_&cnK9YEW@VWgT(X&c=FK5d}GPX@WK6WyzvtN0A-?Y86GtT2=aMFj)$u0JO
zzQ$Okmd&+4GI>NCKfkxAx3Hk0Ii%)wcp^GJwmECRDLY5}jcw?P(CeO9e}NU}y9R&M
zS`GD#08g%G=Ir!Ji9r!=hr4=C@Hy9A(}ZsgZ|K;<@UiCM_w<WUd3U$AB0m2Vrx@V8
zO!`ca{W=?}$h<UWaZ<$g0t?@D5l767?yVfV*`X!(9Py#!>N?3OHVt4{|2_tLQu+-A
zzce)9b1RnnCdS+$S@0^zmC|3t!a9<ilFVZ*OX3!uu<)KJqxnal7kSii@fUj{EcdvA
z;lt%Q<J}baONn*nt(uR!AG&I{7Zu~-?Fl3iZEHbu7~$8#X$|j(Q+FS|3VkM>BI9+=
zSC*1z_c&%{vms)wqocKTdUiR$qdl$3U{9%#E-qDQM?X{%u(EF5&RkR!nn&npb45+~
z?h=xg`r_kpbv`;nsZPG;BLdEHyWnDcKAtf=r0wfB+7K2u8|7yv{MApw=PEOhE$Us=
z4^NE?*L6Fo<mAEU1($E1lq+8+dmJfd{4}|o5K{7yAKd+WNzw~+mR%OpuU@<Em)<~6
zZ+t79w>=-8#Qybij?DbwBYg>qWKHQ$adj(;Hg5L1K`9Hb3_~4wJuh_mrHX02Bt{-r
zygAO?n~u^vC;JX{b%6Vc;W*-ltN;4B3qS01#iTA2V8o|gs1C{R@h02rmBr4dW^!Au
ze$NZ%=zaV~2*x}7lD!g__3T640VbX0cH_Trbk{K1wJ$$xZTi?(s5MnzK8%-K-)sP5
z<vxBF)2F`Q>z4Bgq4`T*hd$!q49IaN{wGFCnoHX4<{dQ&J+#R*ivb1y@o5rlon6|6
zJ!C#5RM`5PUly5GUa4V-@T$`P{cfeA_rwd8{-d3+LA=YD>!b^UP$U4KPa#nR5DbUx
zo4iJ3jwHt$)n1cot1`Idpp{Nt;m&&1)@IxYD7sIWxI30T&kKk@<iZG-elx)R%l%1N
zyT?s*SJZ(23)m&Gh)+kHXWJbnii>^Bx}{;9jrMg6(X%C`&8W$HvFu5>G$~GqTjvf4
z4rkWcHhhFz6XMp%dq)p%<iAQd6(}7DeR|n9n_(TwqDB$w26HuIY0UXGvT*Q|hrJ`O
zO?OTk>P8_`MRn~x9lu?9{2<0+Zt7FaaG`DL?EM~6b*h=d6#6W)u#fGqG;`{r6z+<f
zau$IfE^oF%oR&%7071cH3Dj}GK;wXGBg@v3E$-V|WkBa4@nLYZxfFuz&syXnA45Q(
zwCGf-%VRqiIEu%X4$3!n&Om_;LP?>^0i~&_pv5Mh9A?c$Fe3QNfJsFa!3aZ~Wc@}n
zyXpEd6cr2Sv#gnOqlZOUB+$<*eHXi7Xy&S*+RZd<aKA*oq3p*2eTq#%C>(WBBb`?g
zs7{tI4N%qc=EL!sIYTqjH!x4e_@+_u*Z7_66L~)wueC^wMa2gf>FQVza1tll!4m8&
zg);`&w&c2}9aISgQ!)s4CJXf&l8N3xO-uQhHWn`u<JB%|ztEDRmm=mz*@|PA<LkTX
z>qB3_ms96Ut&L`i4AfT%S^Y8}Sj8nxF!jTBkdhOLGj%v5r0vhL`d7UL_g-jr*ix}h
z`ASud<yhxSjuRdmSjan8^_A@$F7U`wA4)6GdZV*n+fP5k)fe!Bsx!ZfVI#~0pkq>9
z3&MG}*|<-%4{c)Xxp&FMg1YDurwWAEH8qfu>BxO&n#;N&Z8*|M)S%Gw^>}In@VrKo
zzgy&m&)4`<EbPiBMfri>%mK7g;0~4%b5ubx<XS@QVM7YX;-(Y36huF;kK;ZnWBXeT
z_#-jhw&!osk^hGiIJL0U6P}>0MFw2khbKhrxcY2uY~D`%<U9({g0zS`a0e-2v&H-c
zX4||Y+}BmbnUlDVeCBYmA6ouR8Y7kthJrujz>CBk!`j?Bq)<O*D*!xlvR>vhM5N^d
zaN@+vbG@Pm3ozZH9L%H<-}eUIZ5LF11XFP;uAH0z28u7Z*#B|M>yYC(NejibWnR$E
zEq2AGAaf+Qe8fF^i@1Ncy=c;BkMd%CwCobf{Xz4CiI4(K$ZkY1eS9h>%%oH<N+(T`
z<Hha}oP()nGpyWiI;a15l}9r!n@irP?_>Fm`}Pao+nuZYSJ7Ri-86jgn%I42OM$cI
z<<h<8$5{@CEl&-5G<BxQo>X+Vbm<2wduth7;xvmH&~xJjEIO?&Se*7f|48}7K`vj0
zNqOX~Gj0VkO1F?9S_{+Z0YQ?~AzU>o>Y(oh0`rDQUJHSuv}xsJ6Hqi45yv}J;SLO^
z&}0~n&w?SpQku?Ct7QSbWyp9aa91Tx0j>5ETCH4l&QmRk71AVSC6xis`D{N+N`+P@
zvm(G-)}i@A;v`#!nj(@Td5_KtdZfwRpiC><-VGg+gtBcpLYBjPV&6moqj{)HMsDdv
z$R4f5k?`Qw`THlRvpU*z3jln{W3A*gq~QNFnAIxmmzQ}&__H_$j~<^Rqda@%=DS(~
znG)}5n|ANzZFEgeCL~T8%cd`X81QNPOxVaY{m9fgn>t^pQh$pjg4gjCz(jT>;*cVf
zi$Hu+c!*!s6qe>*0?*MdcVQDeZf%O==QCk*q_mqwMc%)|z1H*3m|j;=`PB|BDHYGr
zE!$n;%8oKh2#$IZ&XCXFS}RW6qXeZzkx|h&ZDq4m4+TS$Zw+ne-XlQ*Vb4Q|%yU$S
zmO_uMI@s5?XWziI@7~w^mn~?4@PI5XY+&7U$;8|a9Zw5Qn6z7t3)@$!JA@Q2bt^f2
z2$)FA^hHSz-CK!Y!wN-}KWJ3!1kvOiWW&As&>_OcB}$e1wz(Qr2yx5MEi*-1N+rVK
z7*fv(2DZgfG)H}KG$ygw0Z(o$XBi$<D9eefJ&`yG?{&gC+1PQN%~WC%2|O}|Wc>ir
zWMo5&Q^<j2v7*O2kubUXzn*;mkI$-oy#96Uzz1S9V>*80s#0=r(HC`_z^X5wf2aP}
zo(Khrafon<+cg7X?bp!Jz&Q1;4r2+EOO#Nz`J$6ytnV5PwhA3D6oEE!Ju;0cpR#MG
zAB~{1LeOH^?Q`zx3m7B;5g!yJb)#pU%((9g(VCg*CG5;swdHy$5imSqh3>|<Q=7IW
zEBJz;Tb;Q*En^8!wbR5dE`%l+LPc}KF~r_7Df^s3%Cei;h|+$uY;(pj1I|+DFAPL2
zDU|nG`?dyT#El#%b&N-BK_qb|f|7(=Koc-<!UvPXb-J@PYpu^zb*bpGjkL2DjXfoI
z<ek3(!|ATiCBq3BQqb)d>x6p$G}dzGFpKVKThLfW))4$5FCFV1mxV}v&JC*$XPt%R
z>QH4@>hBj9>p6^lz|{eqs6N?<*B@F_XD5+zQ)UAaOirA1>8n>LEv-<4a`Q0MR!WCm
z(VFr#tfchnWbr~)GNigx4m?a%V1m9bz{dodESz}G&|3L9ssf$@6}FH_!+^q>A1KB^
zT$!b%83koZsd%&#j{ng7Efif58LcK1V?PV{57ngtO<-jCFgK;pNZl1HRrYJ@shX){
z;Bl;T3@p+c19#?9$ydBZlk2A(f{VU?WWdmDW>+&_Cqrst^ya=LzEA`MA_T$p0kvwN
zF0~9o5p?+S(BHf8lg0(<*&Cy|(4bS?)=z4|!h9O-yGLb@;Ls+RE6NcZm=XhXwW1I3
z$<Ry*N>v#ba;b7J1G!RN2`#%gh=i79UH1P{;BvLYDqC%x7Cdz?=xQ6oL_pnII|W7|
z;Q${NMycA67Yl}+h!9&B(-aO9&E1bg<YU6Jtn36h>@Bo>e7Ao3?s(3u=VPrFp;^%B
zVTJlC8%`=00pi!;XVG?J%X?<d(k^;|9Alfy5hS;Wv3#Re%l^8lc*nh`*y0i^c&s6E
z0M9}UYgXu^5~!eJK;?H2mRw5#jZm@63|p&Yb6gtPLug-yVQ@-h=-#_&y<cvvnSiY(
zHsOX2wV-q1QA5j8dix-^yo7MnK(j@1IEI5jS8W#*_>m}ldo2R+xPt+;_3Yn`-#`ze
zSg-KkVm$w^F7kiR_x$^bsz&L{2N3GegY@IzTV*b=X0$Lmevp17U0<*ITipDp7F1lG
zChpaxK;OB@D?NDvfHWwcoPl)Nhm-KPD8sr)KA<O*?_VKx$#NRI(65xcC~4<U#x*z1
zE;*YT1ID9Tn?kx0830yY+$P3|Gdh!>?r<)(Yh0blw%D1_RZmwpU~q^xpB2>G9>W$l
z>%jyh=$^ed$MR$t`mN6wUro^DMw@MA1;+~QyMIl}H``J~jHEafrPrzh@hGAD!VK-s
z#K|=CeiJ;`jx_JyXKq}|n1iPx-%1z0*fHnt{>W7u-wzglKEAGco-p(H>aXCc$PT^$
zmoRH(ARL`ymZ9O~rKuoB+ZhY@%3APHXDGW?F~VT)-`H-)oVxEV(bu1Ed)rP^KI(u?
z&5Ho+Z^B_{8EC&E*$6y`q3uI#h4$zG3`hutvLMK=Xdp4I;tXYQa3INGBSkSg;0p@d
zak!+KxRN=6{{l!EOIVy_aC`_fuHdN(f#dA*dWVDrg+dXugSgU_*#!2MVLWm)I(*Lf
z)Rufq{B1MGNMwh8gN;~2RdI{{gw$-Pm>{^M?5b%_YH~_Y@P73!;S^WrW$FXt<lz?W
zwu(!FW2Zh^lj073vIoUX9~7#+o|)vT{c!<Sn_{EV@jaj&9*|nO*H;f2`D{BJW-e;*
z*%8^FS5@0al^5W*9^@o_((bU14n@fuP*qJsVf?SjIChLgT6}GJD&UtHWn!8UUN1Y@
zMMg0Uh3B(J1g4+RM#akN2cj?2<fC4aEffK_aqX7G{>ci5OtD*%apGc;iUjL?RgX=;
zsPTI1skP1fmS==oQHFkyX_AU*Xvr{f>!-^}Q-diz4y3T)c45~iG8<gJg<d?liwv(k
z3gGsRF`}*=FZ$DTXwsw2IyyvE6|?w}vv9W4T8p?y_JcE9L=X+YpOlrekO2KNw~1zF
zsNlF0-nl${!Gouc!&0O{E6_qzj_g!TN@F8qOx0LykTe`b47zzX>+yeOect1KT;$y0
zAm@{Uo*c!uc#i?#yM}2;M*FwK+S|Mqir3a$9tJKohj$tI-IjV1eh&IQCQ1s#;=C(D
z#CA9rx3>0nG-1$Y>H|tYBT6hhkX-jP?l3kaW3S?<<AX=-eU&~VUANW!Dc1rTK9U70
z^oX%70+;=k^bR3KlYZpgxdJl!_u!lrSIrcjEhqPme!Z2xzrnn*_xfkRZq2>!gq?~_
zQQ@B}W#U2BOHq*WLAkVwYR9_BB!tClhBQuzG?JyN_MFQ3OLh-ZXr|lysZup-r{WIH
z>7>-g7LvjFa=?T9sP00%$xYFRUKuWwK$EYg4Fa%O*pE8v`^<Ok<R|a+zLT;}DWBB2
zBK7R=^T%JWer{d(=>^<iu+G&@g8%FT;Ux0)`+sUi3rELI7zf{(nb}ooP~e$imFwL7
z>ECb7U2yuuH*U*m?+Iy}nunIh*9mg1eGkmPeJgxk(~zsvf6bkvj7>vB%>NYr341i`
ze`>b;OG*TXUBGNP4$ifX>dHOSV4TUfM|(}Tde%0Xa52Q`<o*F#WzKbEuSt0uHu?*b
z9EC^0$A<lIUb|*;55LhqT&4tOB4;IJi+EZc^2u@5I-&mB&!5k+5qz&*O0&^m&-oKB
zznBdS(t+q|u6NVZ%+`oS`JZf?7A1k@7??bZf1-O^VEN=TB`zi2XudDDVsa5!t9JvY
z&MM7>5B3(8TIZI=)|`o7)K&I4Exsr2I`db)p#^0HTH8J`aVx#r{zM+KTG_rJG{*&|
zd-V%KFm8OGx_<we#89giktcUMB>E*;@OQm4j;pZeSVP37*~m3YPo&B3ugUSxv;E_-
z$mhE9l_=`c+Wg8KBZX2{gJ!60GLG6=$%lJdTwg+0F1Yy=b?LR)g6;dj?CR_2yMNw#
zmraNjE)4mkOHlHyjscW8GI<v@9Ks5qru+b?vR0I<CIrDbBl*Cyo-6T!9V8Y83SO>g
zb~-CrECufnutwTtlDGlC1$D>@&O?ZWYk#&~N@hcWD!e^}28Hxlsal|tSx^)q0IoPJ
zkx7P5v%FfnDLqr93N16NWxH?BfJo@pRUiL+I`6D|m&n3Aw7Ojv9vjd3LD0;-<o=a1
z{RSftSD#ObkcSc;PO}EQkz=j7EC4}fxOGRP>k|QWf%jr9+)ONT>`&-0R#>aDqN?h&
zk;yEnQ9hdvfC3hYm`Eh|p;j4KZ}(cV)C5DD6v|RJHx2B>vvf<;Pz=bfM-h0X5<a7!
z!yuDUXutsw+N&>%20Dt{SAvxcT~obv6fIvni`}xG3gBR$`~dH#hTh0~24kCK&weSU
zzDC0bi(u|p|Afrwtaw{s?ikJ7cQ=rpX(Ps9TqEaCKh}^{|6&r{k^MkY&5*UyknY86
zfGm`mx{Y-_&0?GT(aPD)_nYX^R`;pnV(1dvZoZd-E4J<o*vBg#ZSWAyZJ!p@SRuAX
znBvb|Cs5^$rn(4J73%~V0~{ckh2=BEd@BRq0O3AdZ;m!A&S;O?4;!GGUt*}q7}gqR
zmxF(OvhPgE&s`g4!$F|Lr{yI7^%43xkSzvB#BngM^w~*YyoEg^s4>x7_y>FU)QOJl
zw|hnctxYZFmW67<4#V<EO^D`+AFWU7?wtclF=4-Wk50qJt2U=sKL0#?)$~6QzxAI?
zoU3hJ!eNi@2b{Wn8q&Oj93v+AHmHBBCJt}ve{02IV=D$<4{H@5D}Bx%FDHs_Y<<R-
zi14fWPPaEPeB8Sq@}`b-@siYr%>`X^_Xv+@&iC)s{RhIMnz`zWj6V~K=elr<Kabgd
z?2}f<kFC9W^Hp>&L4UPTNdH<jetw;6rNZJL=cn;EA@AAK6~CQYzOKHZwSVT*?M!UM
zz|i=Yj(5pgGd=_RhSEf-Al0UG)8V-0%@Z;&?ySId-}-<1^t9GJs$Dj9{@&Jai@PUt
z{*Heah3Z=)!kkA$4ThFV=^T2qLm3r(IzKW`Ps!MPLu_9@BgZXq$^TmWhslNCR$bGd
z<%H+Gt*x3*yC)l*__>K@(PbEEG`)YkBewX<@kkBUv{1caE7u9W*EsJ%6K|@W+Ypy(
z8OpmZVCfGNUJ#vDivKlv`1Cjeld<;3t%vYs?d`d<u>FXyVI8i8y#M!Go#d~(zwZIk
zh5Jofh}OXwC%b$)JFel_y4C-fG_cWKk4HHxkj@FJD&PhA5X_AEL|%3^WFIib$IuOS
zXUjA%EN<8jCqkkrk_NSpWnj<_bw9htQd7rs#zFtM+mfl;!eo-FOt`2!OnS#@+=!``
z_<OP835wE=7No3!dyG<PNI3}IHW8CxKU#*YV&DW38@K-EtB4cM*k@O=3h=`%CL~mY
z&{BobRN+#}v{38sFvRUd%dA3hPF|-_EHOHjiSAZiL8p{d;vTV2ch>lJe@wVD7rXXg
zCD$&`W!y2wPcNG>8T?JGDt`r$b5Hm~n22F>8QL6HN3i0tgMo3Gl|s6NZoFOLxv#&3
z9RbwoLyUbnAkcdB$OhQ#>1Gl=q8G;PU86v}6y;(A<<^0L`KAPgH-N>7H6LZA4+J@w
z<w<S<kkHOVb_3_lO=udsTU08V!;k`G78%Jx7y?k9@gb~34;`{ak;Q>bB#LauE{_Bq
zYqxYci1Ab>IKOmL;r%XFkf-!xa0j%R{ezTLfTV-~MgnfmF(W#mB}wdL+}3CJF6nq(
zNZt7+q_(agdFM~AdN-D_9q?T_w{`+LDV)0NUi4J=78@ZU^{R0n7cwJcBznQ!oHQJa
z-#`uHu~N}_x>pKtS(UeIkBxEDeMV}7=E`WC$x_3dq+gSADmAJqc<z3wNCax+hTIex
z1<$G<xBW(8*P16g!NgSnl4%FUGg4caPlu?Wsj6ho8>KPjYYl!r%}wPo1V{0*BJ1Zw
z%!TGx_ap}6aLG~Vw*D2G_3sAmjw3KS)@_=3|J>8|(*6ebWp{+TNeq*UF{8Swle;b$
z9@W^ZI~=U@E1wV-<&(TMQ?^Qm)b?LabDIh7Sz3`jOfD0z7!8bkJ=&T7hvV{43#@qY
zet!9EzD}@WjB8s<KC$aazKukqWwOay`Q^q#!mU(S2als9F680>DKiB(ScpoCMmR7y
zE(wv@0y5-ZQ76O{M4^;P$Yt6VYS3h+^UdaC-q5n&nmwNF_NC4H+2L8YBBCC7t<HBC
zimy{%{pcDLe^3PCJog>tM~+dqzDb|TaT`7%@uEd)Y4!5o&GUwCKi$16Om?=T{%XDW
zq^l8b&|G`j64G!?AiT8EN5ZjAOCUv1Dy-Fe5O<G>UOwISv-9IS!G`8VPwb3tnTP$x
z^IHj)Y`l-n=CgT5JI?CLU_ZMj$i6S?cxbLP-#w6fd%Qz**l3=6M~M^N?+Qz|Uyjq7
z&1NmMGCm^`=8^dP!9yK|MF-oM`>B-=UF&`ng3W#89j#hd<v9h+Sot~tJ%JTZndN$+
zPvZ6LSS@`g#}mq(1xtFGO-SA!s&}ygj@5Un?uKD2-o?isX15hL=}QfDNMvEQ9~qXK
zW{vmBvfD!3fZ77TLoDs=F2W%E<`|7pr<rVFQW-=&{(ODmm)rV|IPGG@;u@=ue5d58
zi_2Fk_Xf*KwoO8seCZ1MukvZ9?<)8-_Q?FCcJAah;MiR)JA2aOs2Pj)cYpI1XO2Jj
zt}VC>h^P;9>1BBsN1gvC`i?WM<zGdj&|EXEXG$<F`Vs8)M?0en0zn&T*QWu;D$u~7
zNpt}|p;9DGTq#qC-K3k;PJQ3-Vt?m0xq}i=VzVXTh}?{02vjOx--eRBr)rCQjmp;V
zQ2}SxM(yTc1__)mQ>V$#JEA@gRpc-n+U@FsIn@7b!*nKBNin(Me$P%G<6GsMS?-Y1
z5-jD#z<u}_k8fUm&GI%3a#2HkUlhtVBMZ~tO+{NYr1vls(zM7I7}FiBfnIMLh&*oQ
zmLzS(+z#cI8s`}4FdeViR!j7Lj`~zYbKs3*0)|bCQKzG@(#5h>^Y^Y=9KK!StNM>h
zQ&rhp{gaxxrV=mQyzaZpiyet_i^>2yh=;t+@6l~n+)M6Y))19MHMT}BxuaO@++0JW
z%0J@Y)_U&jmv<pbmK~3w8{~W^Lm`{YoOTiUmE<eaEHU=40Vah=Z~)_zxBng=wOLc$
zSyys#NY*I=sTC9Bz)GA1Wke{q1%nMaba4B!Ni=V*E<}#<F{fGTS^zZOUrTM1Neyh^
zsWs;2rqSiI$wt$nd>FzpZn)zxlFzS*Z<6&ptroo})<>)Naae6=1YVi3>WXsS`xdu)
zl3pIWK3|(S8*G`s-{6yud}=B9T)khCf>M|;z3=hJJ;e!Lgk(v!9i?E@ES!ME;fsd7
zqsQ)yAU3EyF{#puYVn7B8~J?hZK|UxGnoYtY7;0HHpuBG$fx?{QsFM}v_i*7b?c1%
zdar_uFo8jH63$izjvCsMU?3-1_2PCMhoO)h2GWI18%E?Z{h+><^odvD^5I6}tVz83
zZ_!lm_Vb($h#wds1SU*Ofn!)#LD{)zy#t%Gn`f6naA?nQ%UCJSaaGW_OGqqnRd{AR
zlj7h`><Nz@pWX1(522msy2OSLxCmq9&$fTidDxWNJ4>!p>7jLJ6&f~7r>I%p%k4XP
zv#w`@emRHnBtB}6`~hDGL-@NH>Q=)8Fd^i2w3cZu2gIuKGEIfdIVd-kkIpxnuuC{u
z!7BV<+L43%GBTq~zJL=KFLev@X@xj-wrZ&ePBkWuhg}K$=j~o?m%{Z!$6T@HXF-Vd
ze=?1-{D1wHGp<=j;ozO17%1ZPGY5_>;o<_SiehD@ftTd$ftTc$M}x<BLL*mq%@UPO
zk-#IA3o{}>;+UK`Gr2%0>tV@F$3afnqTvAZhas~?LPkuPw2<HGEeo#|PSh4kSu3Kw
zbCW_Bz0p&Gr)zJ9j&lS@kh(RThYT5P(sjA*`%0JA3Sr_c?hN$?Ntw;F)?6Jpeu8<!
z6Drvi>M+#siPKoRV3DDIFIP}E#j*gx@)P7@3!)q-TduM^6bB5;ubBumCC6HPUA25#
zU$$sc`2O;hm-fqxx_9pKZJ%&A6l)iIRbKWXjfEIT8y1_%^$ueXZt@#YZZA3Uoa!d`
zB`JFP)32hPRfu78r$jRqC!A606voR76Z3~sr66zu>S8pog~WP%ezbVxg_VPb8^*X;
z5a{|kp6LTwpJf)QC;&8D)^gjQLNKXlc0iRo#0~48mf~|ss)sNnw{eiN8dc<E`hpGC
z)xqBpu)eIrPzW-REBH^%<9YeDAVBb!<plni1(0%FR3s>F-Oa$~oUDTzn<dHH*2uO$
zW95V)P@I?Rcp<(_kQ`2B$VoDc8UT^bidBhCY|`t_YozQ8S9RkO4!(=Tq{cDB>gad-
zN|QXz&us%obrmnwvI+XBCwx_74SpWRBZKP_`TK4MPcVmraq2*kK$xAfD<k|vJtMkh
zCb)s8vMdtPW%hTZZms#LnLV0KrIelyk&AGd*h29g<ItBQK+G?1I)GZS!%rx|;5!v-
z=5Tfcl|kl)xUj*vkr1W;E>muj4XsMX(b-^&$G6J`(mZA9#iLw(_Q+JN!KB~EPD*v=
z-q8^V*NZeJUsR!TWvr6h4z1_eIZ#h$;ptx1N3n&LCRf~M+;uYRCv??(b2}6*vxn_E
zE?n&23Z%QAE*%?rDuf$<pL6MqS^fw@vVK*fS+%<?OnHRioc;>n-JPz`V>IJ9Vn**8
zkan^3GTY*X7>05r(W?$R=z%`oVz9@6SSB*R)w0G?Qqd*qL@81%fY{EGeZY(cPDX&z
z$%%IGf8bhZ4_Zo&91W0#wPq~18V(i<058|>Q3sqg|CYJ`cfTCKGY+Cs)dK4O?$S>|
z1<I&%>byqB`ANuWaF_yh?}&AS#v@{!%Q1FM_}S8CH6>>15R50g7fH*6K4oK{&VxeJ
zwI1ABBC!)8l3QLcr!SE1Bs4@&F;9->sC<(4a9o&IOOX850Nr&74>)J`W713&3Q3M3
z<O4)6us2@yQCQ5y-j(IR$z-D=)ibK})`V00H|?&lS!p{ZH9U#h^$0Bj+BF(iA3Xcz
zUAyoYjC`8EK$xE7XEP<sRYJJ*_q7IY^eVz=h|^a0qu-HfdT#l{v!y|^UlxY0963~I
zzw{v2<4H1ffW=%*Y3I6(*|(%<b1r^HbdEQA-k2}!T(C<~dp{TRCa`TqQmKcHOw}E~
z!9E%^bsrF;&vK6_^w?9sZd1=#xXGjQpuv>n4JWR7JP-3GQeYLh2PcnY+8IyE&n(55
zS@cjhQR#GfI;$v>jp5>p<x(kB^JQWnjAMZNNCX%LV<@XsD5VM)LkBT@NHqb%ugrRu
zfP->h+yKmm;{aQSN|j??g7Te`*_g=yJ+Tpxf1Ryz!c~a~+Fe|pWNRz;v}Zk!=jw2`
zBOGOxMaRvLEM<FmkK;#ayx|BDcZuIH-&A0B=l8m?ExXr$|N9E>Af>UZ+<Ym8L}Xg&
z_bHGnHa<q7MxnbtkB$;=<^TQCXdY5i#R9$J687ASg$&acNW!%+<c@SfGg*_R(5C7p
zu~gQ8QctM3P3t}-2qRHjY8@(63L2>pCJgm6l%=Oum;edCq)>fMal?W4W!Tk{VOkXC
z{b~GuLufaVi(Gb03Q??k%X;*t|6^OuBdH#Rf+JxW-YSsKZkE~z+oJ3!;kg%7MRK9p
z$Ay_OyFn#iw<o$~7G1*5He0AMAzz(}=7cRzv8)!#b~1^Nm^h<OJqa|=StT|8HIfY$
zS7xfDZHS-QM5Veke0@%~oEPQEKU{OM#kDZdo*4-;a8b`<oPb9@s~efTgF3WFaZxY>
zZVlrbC<c<~BCivw2w0l6MFfZoxRm%I@1_VSxlOnE+i(6tXWu!$J@?VwIzOel1T>!u
z!re$WKgFNf-O~T_4>Bot^KTO1P&^LK&V3$4auni3SFmjKple;^{`WvF){NSV|GHw`
zZiC}4<L#Hi+#+`!fLX}<Q4;bmOG}9u<)PnQC7>fk3W}vzWIzlkvT71gLw1)?%eiQ_
zm&mnI=#T>JV?x!9MKa1FdUaGuwm1`}ob}n;maJepY=C^5+mYXH<>3-`Y9A`)OA;+7
z=SN?rvgC!4xqlCXz3_J_>-l5~%N%Iwh_%sfyo*J@$_pEc_8v<lKdVtrX=96oy;>Dx
z&skT5ei4**Vcqt>Y~mKpgA~6}HV$`Kf)$LsDXe@UnxDVvo)tWie(C$enNuS_p)c#J
z&%4x6#HJnI*P{hE<Xgsr(@drvwQWkT+I5_IK=6$=yK8!_qbR4QO&Gb<QQAML20?<L
zH-p_D!xmy*2scWUdhsu;{Z2g4{WV|jW56O{5pJ7w6?MKrnxtvHlCOxGAte-UMmH9V
zf;&-JhCmZ$5LvOoCW<T?ks-J`K!{8Mm=39%`i%;YvSRJ7_yA-mLu!i-XUivJ&}Mdy
z@GLT@k%2m2x{4Z00+oV-4YqKb@7NRKgl<LIeo|9&fje;lEjK4=lI_YAN>I)%Zj}as
z<U5TvDwPOfPn~f)ZJydH7*<~lp<7&w3i`b-(Ri+p2&YRnKOas-hYC?n-RKqIfQuKs
z3!uyt*@RSn!a++i^whS{XHf=77WG)p2iN(JMY?sXg1pE0Zx-t*kKjoGeh>j<l|mGv
zo<_FMZlKB}f4SLGENpA_RY*f^wYit6+A5@SUQT_JM5g{Gihq<E*$!EXnI#o20qaPL
zwi-);+_@*xo+p!PlnM@;A4ZB3Um?pBxcGbDhzWj+m;)DNhXPke_l-Y;gi7%;V_}!6
z>T@4z7jqV*9v@;msz9rrQUJm0CiL&T#0+fENK)d9bY11DKMijiv1*AV+<?iRK?A$7
zgiILYA^qar>5hY^-G^v|#hoJtn^?zCwxU4t`S$~_=i7-FQT3MhpX=TljK89|u0bA&
z@p&RWt)NP9e9=NoChNONyjK9`jYGm+D!0rO8>}EEvNK>6GyzE2^0l|f5Eq=;1QvDN
zP7s=b6oI``!SgKDb^HgChu_n+k3RS1-ahU}M95P;i#tJoGulItK&5H=iG1ST<j0aJ
zT`J?B2Ft)hpTf?^cyy#GSC*aU`(Bu|SkLXc2^JHFcf@&G-cB=DLn_*D&UH)Pk!p#E
zasQFa{3a+E`aq>PDvFZ*>dwl-cI3Of+mw}tk4Dz{^`Sc9SSkAp-{KBMg)45qs$AZu
zLzr_Tva+*M2)XG<Ud;gW*8;4SU=;6l)wp@VC;P^wu`&fk`Pu)AvUiVX`v3p`H-}+4
zn?fU}*mB4+IoF(rVMZ!R&Nd=aPMyrzawciSFms5Ch|+-=Qe!qPA&F8Fspwp<@1xiI
z^ZR|izu)EZ{r-O6|GMOI>EaT5JRi^d{dT)vjZ@JjBO|R1vj)}EDzi4n?uVD}fBk%E
zqxRIM9<oMNWReyww*2x%=$}&f<EZDCIq9=*GcuCpQk{d+GA{27qB^Yi_LU`fe#!gt
z{Y`gbdBw#7NVuxBB}v8i-M){~1ws1jD;`dn2Wsz+oDU0neIQwlbtBr!vSkVvZHRqT
zKJQt)qTE}%5c?D-hhF=B@u$Y4r0KIS7I%;T8JsPBwKjT7Lj!w-r4BR8UAIQlIPcHT
zZeDA-JzHaC>gfk2OViI_;xT4kMF%;IceTG;_Iq?2Q+pmwCIkyE^k0~~c7`>(RNS&(
z(0-^#GNUL4-#o#-{{P}Vk3Bi}`XVq&xUg=uy#5;=3e~yXLo3ct>rX!y_HTSB;(G5F
zMY{kAYbh%|wO>RHkYePgfStu@QjGOmDD;CnJhp0LK-B^B^fl9X_a|G#)ufg$+O!eE
zxTN)GLCMD!4hgxX%KVB%AzDS#raR^oKP_OTF<9|H8asl*jS0~z;a%NY*17W|-LFdY
zsHYvVG*wG%=7}vPRWy$C&JLRT4C$Fh-!E^A(#XJv{35Fjg!W+tN9jRY!Gz%aQf_A1
z@O)igk`VRbmHby-A;});>wTf*tB*dLaZaQ>_4a(Y8x2M`P3O1vGR;`deZO398iXVI
zza5uK3Z{)7s&FXmLi8QXHRc$CE*MFHZYO^c=&-A3FQX5hyQ3H1kw4fr=`#4#OJ?g2
z0FB<>R^m#3JYZ{$_9kjPN8e)9#LgDdfUwVh9g(6&NdyZBp<ujD3QF4-gG+;5q{D4_
zL4}iWLNecW3nB~44oq4GhV6O|L<db)1dwa&pio(1KbELo*Qy@KtK3DYf$`~{-s&G5
z(FmO+U<g!NEqw<$24oeU932qJKG^`5xE01uIL!XDaS{^iYu_!!_Iy$u)#y3HKZW>$
z=?`zUjx^SEyW@plN5B^@xVfs#EvD!?vVEQ)@1TsU$oiaDUdcR2J%lf0A6#hhH1Oq#
ze<1h*06q?n5!*Atgx4P6-ODWUyi37|VZkMqf}H{T3p`7X4xt=ziA2{fT<6_F%$BA^
zlGG@Y6tg)m<IhClEMHqu2)4i_3}6_`5zJ&_Z$^%oqGVAfNCsimV+9N{-U{v@E?D4t
zM;*;dt*xo$j<u<qHz`7(5l6}871D{rO{GP&3KbpK?vFll`B2cumA~?aR%+A~n(x2E
z`*chM49~zcfV1^bXQiZ`N|%PHU9wNpth!^EJ?U{T{Pk&@mL$eN*6Uh5(!rsT7!N`7
z{pCZ<IFI9-S#iF~=Q&Cr@WH)+XNgW3WG>lghVw3hCC$@L%e%sMhw>?4zN1-4#ifJG
zT=T@Q)EzjxAzuCvLfoZ_|2y%AN@T3-%g`VEwWq&weU5DK0l^G3{x+^{jxchQ<ppN@
zJ%7~eIXad+G+tZsAh|ZrN0u}Pxv?>MM$foq^xM^CZGn6Gt_#DOcUJ#%l=ZNCtEIo4
z+V*+%++~9ccW&>bcA}4!KKPMyU@mV^Ipy8tI>Vy)8{hNJf%Ws^;fb}8+F{}Y(YjjI
zBc1OT(wsjUUar6YC2!+h^ETr#Jr&7#&j~pE-Tr*hC(*xLUv7AI&zz}M{3^vEKRgyS
zuXA{Zb<c@Y!&w5W(=QsCCQWbtY@d*Bn&0}U<+ZPE)1SA0cNPKN{HS|K%C)A+he_L}
zwI_27AFCzp=@|C#dA2umqpvlxM)PL=^P?-;Kc#Me5)Qt0t65F5=ZWq|yP=c%pX^8e
zfkYg1R+mJ+Qmf=zb_G-fd^;1P^|N}adMBUWHoxNYLq>k-x?P-h?2x4P4vetwRr021
zTQLe-271=>F4^A<9ZBnpN~=tMs#g>FD8NMAL}|5Z_Wk!`v2vCR>R)uf6L-HKxrgS5
zT#rOM)@)yCeY_{M=h5Z9Gt!Z;OQjv)6ZhXaEI3Md@4kQEnCO2|!oXICw(~AMp{5Gv
z5$Aq)fWE%-LpIU6U&M=QDGJ$fKM4H{ma@ev))|*M_l9=7q-PqTFcGsFq5R_LsT<C8
z`vDLQcttPx*b+i66s$N`N2G<A?@ESO8S||p9Nn1-%Z}q?t(rL+Zllc>C5PHj_+HhF
zNCFdOUT-MpTJ@w3HB@$Fx8pG{H^~F@w{8~?g{wA7KR07I=Tw)=Q?%a{;a2Sv&2&=Y
zD*3Xb9_|wlz=Krkvrv5NUg^nm<Omeo{fW6rhg@Y<Rzt6%;=V*kfOjwb9e$x~{<Cx1
zvUhl(?fA_`CxeCQV|(q7%G{3gH#X{5i!|RNFJw&6QoIyIqQSDl1r@VJwsaTVK|b@@
zDdiRV(vAHJ_>AyK=iA)fdzP9{fu;#IoIc#^Ukcu0MR>0O25Skl5;EVpyFt>->L0R<
zRDn=yzpP3;8M1&){|j49CZYT%P69(w5FoQaltIZK2HJ`W3H1g-2qGor@=_c~E`Wa&
z;w!pY);q0pvj~)u$jTfn$|obG$+CPw<sFiWDjlWN5j&gmH)jb&38fw-Q!uM{j|9@+
zaww{K8rLfYJ{^+>o(i(sJc7|SO*V~~;7@Mq+A0zupj?NZPq9tvITA&>eK9VW%QH~?
zM6uvvCYs6J@`ISe4H>W{&->4s?QgUpx`$a|hK?w(fnl{FApLHGWN5tumIhaH74Bmd
zQy@IbEvArMk=rO4ilH6znIl4Gyc{4RUidEzSkaP;v>8ME#cZt>9d3WrXofcjWI8EY
zp+a@bgF9?6Ldc)Q@kaBeu%;Bk7bbelx0nxccZ>+u@RN)sjn@X35zq}1i5ypl&kC&%
z=E7U8+*$2y`?nv=G$u)u0)0cxS#GlhEmlyMM<C1;vHea2-^w`CJZ9*Pgb!5qyfNTJ
zBEogNE*b5p5PsNNG^T^O?S-M=Fk&n90@~M#JA|u&@vLG|;V6cLEWZ(zH;tW{FVdj<
zg;8Os6&%b0B;is%+GsN6p<^!er+@Z{aru8ys27`83r?!$w;nrb7;itFFV-1xydf+h
z-rs_&e=Y2et23RF?`mrHW8e9IAlXxD^$}m5?moWDw96y;ym8%>{j1ZSt&{I=ekpvW
zPrEsnyVv(kL{aWk^xK)vkJZ^nTZ3<QP0-EiwtAlKFx!^h^60nASMP4qEl+p(IsblR
zy(g{H5^O61?oLE+Y|gp5xT`!^ZZHP+L{&_YS+(mUBF0JRU2%E=^z6RjA@|r42dZ;z
zrOIMMRMW$z{u2)qG`7Enn~(WpXs!mwNc_cax?rWY__fPOuW<LaMuy+xI-hFNof5b6
z?nybFIgfMH#3Yj|H$LEG`>X^vm}~fPSLKo0m9mjOAJ{*BZ`?RE6G6hd8};tE`k@J;
zfJ0-6DwpT@_iv;Ry}uWJ>Gj;14}Kf9AH43|zwLUZ>2Zl;b2A&)t7-ukdVi_&_rA`3
zPox&kvmp{V$b!;R?ujdug`KiavD(Xx*Nch5)dohf{M?`mS&tvDIC<ZDa&^!6x#OQ7
z<qAE|y8d=YE5<bDzj%KCH39-TlJ)Z-L2Bc!_b=CP`~EleX7u9Gto{a-gnv&iV{hH=
zxn7=pFJHDNX9)2U6aZq-7X0m*3tX?Z4A6hku-tEA`{Ue2a$^ybzzU>0n52b$+@g6=
zaXSm2%JfWg&QnP3UUKfzRc7v5t$_IrAIB~E_5kv=FQUMuq0zZhB;ZprzbyNPsj#5x
z&WRQYa?(RYMfM6^;<}L--B#Mjf+*tNZN5woizPs=Z+aWib{5YnW2V6Nd;2?IvtOI7
zyy+7-44vHXV61#uGQ6gVdNreMWbzicYiP9f)3)Ycu*Ljxkg2X4WrpvF>|33vAIQni
z^U^F5#R>@AXml(<136)93kS?=3$yH}xBIM{JgJLzIOe%ee$L{`i=oAX!}p$D3B8*i
zOxM%etV#jVkKFNs2s=bVUF;I@y_c%MlonHr2?1vsfAF*c+MK}{!zM7wlo?%!K^^A?
z>Cyutiv`I60I1~uL{ow0Nw<T$uWL{d@LDlA3WZ{;%a`(T2%>sD0<uP#P*-sft0zkJ
zV!MAe!!i&ZlAf+KSt;!7IvAaLd*LF;SN;@o-s;7f)J`_O-|0Ija1Xb6i#jFsp6KVr
zwkbi7G8x4VeslcWa=7us&$>#yXI_s9ih_U~1kwG$4T=pY8uDGeqf_qrZgV5&MkB}2
z9bF7S?bR+&3A)`wMvV-XHzBW~Syb>E$9)`52e4I4w-QSZ9As>hQZ*1Hx!jZ$W4?k*
z*VU!lQP&x;)x}D1WSc6P*L6Z1d#f;>bHcd6?qL&IpkK}Ysbbr0#>8TByg<Kxc!~6g
z;L$_S)hQwqF)r?j%yCM#&9*$0pLE1gi9E|JJDfh{knKc^;g>Nlm+gUr3*RTY>2cz`
zTT04}NA&NlYa<6FGx9AbNH+OdIydnT+nH_hdV5|Sy@DTfAUUENT0J#+!E~WH3%;Om
zAeQY^FiMqh#Y-qq@tmphRU^84N6r?YN>zbKG^BGN1A(+?l<a$?=P*FvNYUEDoGZh*
zPYnLfO4t7nW~J3v{;eHVb^gb2MI#ymSD?iaY00Ibe?_$u178tHQtE-9L=C)dA?l$j
zN+C$dxk*vRH)uvY^p8p;*a-w{m+FUXD784VJ|F0`&y#~<(|EU@wQu)UU7FtECa)%j
z*nk!JXs<UZZo2NjSmei=kZZw+4%v9^cTo~VF{&Q4hgNVdRDNDMI1#LFKu_+{ydnGP
zL1FPkFR53WLM>?3z|?oMu21z8xJc;Xf95$*IplQCHL0ch5qAzA2=%!ON2*Wfi=MDS
zKG8mOzSCXw1hOBzE@+5PZm>$&MB43?&-zUNX}82ZZOR42OCBQTO@-t1i9SKQiS2xW
zZ*lsHgF4&=2YKol^e2=VKCS4vV{2KmR|vEgXgX`!M)13bLGXH)o)&C(@g>bSh?58!
z)Fo*#MKbIn709amQHle)SR!=xVi7(_!9~<4kOl*UJd&3ZtE5vhkmeUw*a8w>DO+#T
zASadQ%Rz2bNI|EpZ~$~4(hU%hth?YrYZd1%ObbwAY9R)&Br26OZo5%G&kW?1B2Th}
z)+j7sgPZumaJ-jYy{yvt(RU~q|K7}JbK~?i@%N(Wqw<q2Td>`1=ko{^qB+C9j*3rK
zfzbv&MfBvmX=bVbDoHBE2HRUDWKXjU+0Xb!G1?@v$~O-C4Y}}@rJV&6`OG$ay+jua
zln}~$hJehF6d880ZUBn61VZ2l<0vtpVUQYT6;;=3WY4CSAc>L$FE~ICKu)mMfh9p1
z0af;o_1yKcQAw^Sf8zNSoWvmI+Sg2!ifpc$gBH9jQ-|AMHIZt#P(<<#5}8jy?U-L#
zuh*mLOXN)hH3MYMGcemr7t5=hnB@HsAL*yMc?UjZ6>SN|WSW1#e|)9jlfLRxrujC;
zyyg7c3N7pA+z}<F1P=b7f&)P<yThn8tz#|(VUkdQ3c3c4!N<}Aw(F2(J3wgb)p(P2
z(`%)ZUSI-B8bW-8Ie{qbqdvAtF*WbgyU2Y5{Y&mUNq_;jqS!Y0|7(E<HhbX_&;JfO
zJv{;5D)F>-TB~7%n;b7OArD*qgM6+&vT;<RSz#tl2jw}eRj>Y5L(-V$2Qi+#q<%+o
zqUIY*b9cB{6L$0PAW8-0KhhH~%n7YT{t(3N&Ymf7F&-oM=uZ?mulQ*7DT=bG>L;3N
zagG!fnz&+|NBTo_;X1<MaB0LTC>>(07klBP#GUTboj*ATUEDDT#$ut^D;#I78yw4c
z6i$qMm@^uo`E04uOZI8BsFPkB!OQ-_!}&5L|D-bPsB>$O*7%7?(YE$|!i(0C{)RHv
zdMwm@<A$NVlW(<2O<pgOWb6gaB<@f)w9Vy;x}JYN-aaQu`*=;Eso+4hR%V1-jS@42
zk2z=@0{QIPtEgXtwj@EAjKIUN+@CF9{(+p}=P%?hW}sW8+>a1F3_sa?pkF(9a#sME
zEs`PDszF1Efy?lU^EV>o6luO3is(qe*nnR&!~zb2RU?C49t&%rgT|!PA-U=tRY#O8
z)D_2Ncv5ZL$Qp>CRMEP6V4~Dakp;0rs5?UX0o8qpV9NtZY&Lc=(@5wtNw{&IdcKc!
z9Sgz<q%?x#m-Fl!ri4z4;5@))UAh|ovFq@ybu5$cMM=uC)uynUt<1s5pXjR}B@3C$
z?kH>dx)*~Tu@O-5L4e^c!c9FKyX;@`ayWHr^$W{x_ii4DKSa&fX-!dca)0nHOqqc<
z5&em1FyfIcw?><W(_v{btSX=`n*GI{T)}0^8r#zCRAwr9C$W-V1dfp0bu>f_+(9-?
zfqJ@1FCP%(M54`DqF@S&{(Q3j8IHS4PEhW{tig<;G%ltjHlZc*Cyp+b4;>906;*W*
zR!X$uCq8%6@QZ!LJ37n+RsVhjpDHLkoMxEBwjXisUGjDt9$Aa)ZTZ=F;b)wUsj<#Y
zxmWX7z9)hIrzELmH}WtG({tcU9fc9#)QL^)NU<alq0R(bek@(Y3;=+ef(;KzN<PVa
z=tmNg(a%q<?w}H2Q5@576D_-gQWi63zpgc?{(a<}cw1YXwArk%r^`XrEjo7dKajSQ
zd%u6I{__ocF6iC4&kK8huTc&wJ~SLK-E8WUvh{pIwy9gcaw9s21wdW@Umv@FBPk#)
z?C>9a^Y&V1$BDCQz@^Ur>$#bDlC$buc#3~E;Mo`=CFa!{`Q#kYRmI$zGM9$SSfHoA
z%#D3jB_*mQ?ekp2vU%5>kFiO1$}7?}-!$8ONbkCO=SUkIV$D6VDeR)?!_%h(znQiu
z91<C~-|YyAza6P^_}jx)$B^eE+bgU3>wV0|1h*<Cf<kz2u(?AC^*;xx*~xN(I$hO>
z^zp5`=H@00wv*HgT2E9V_@|fLCeDA#ztqaeF5#{ysEWB<w!Tu2Y}X%*i3pfGY2v6K
z?0)4*hIa*V-`j}+&x^$Hy#~0i_L4pGDH<<K9wo)UJy1PwEgGw4%8k}%9bAk#_+!4-
z&Pk1uW!}Omg#K;lNJ<G6YJdQ+&tUquUAot>3&8o-j@Gja7WQ@fD|FP`(|+pG-3K%Y
zA@I~ekcbfU^G0aLT@+-LA__s({tLUS1D=~f0EKZ7h6A`wKjQL7TpFT-D-J~vg#ch*
zqn5e}6mpH7T!Nskz|_XeCo8?QVQD<FQDSbE{phxgd#~KbP_adt*E0g?#&m`(cyT`A
z>$}o^<ZQ%SlV(8wGQan7&l8SYiE*`;mu3i%TCF|N2?7&Q2{9K@k(}cgsL3S(m1@R7
zaxo5OMA2!8<Onld=dPBG()GQ*wFjGQ7V1Uv-=Q;{l5tyO-C_x_lPflU#LWs(M%S=L
zvb5-}*Rff#(k}({M02LlW<<0vgk*!DCN1LzVr<x^VhG5Ct;;x(c&xbHHgk?xy|Ap>
z#LQ=$)F80mLLg%i;%s3hn`PH7vnCa-P>npD2b9fQ@r-<Q#sD&bQRx|&;{vg65_d%M
zz2i8d8hDOTu8O|U0xd67FcuNgZfe_4qpti^iB}vkjUNi^<q$vNkvwRHj9;u9O1!Tb
z8ZYU6v>cm>r8yFNxtL5Xr!o-V{<KtoAY7<iYR4&BjH9)vUiJ}#ObJEJL63-{i;VQ<
zg0hwt^A&>FnG2^Bs$Gt>SDu;qO_meJJu~y&GwQL%OHNe@7#`fQJ-4NFp}S$N-Iu@i
zVvlA!SeRFH1&?2YN!!5>$E@~Vxp4$Ve-Quu$gv;yBT5cDEsJ-vbnvcP4E#DFB%2K2
z_(P|iy66JcJ?K~RE{N5mCruvg`xpH@{mz(jTP=<0a$7ZT;v#XiXwj>(pr$$Di-k|u
z%V(XJ4WGW>zd2WwefuUcZH8zXmHT8Z^A2|_wV|uDO~UniPF}<PBhi;BQNL8@H}!<v
z%5sa(dwP7)UNoyw<js2RtK<0E9}d5gPesT?`#cN97?&yUSIr)EdhH-6YCPyWeb{dr
z8}nT9!9YR9+R8V+VmZgoy@$7YmUBwx!fq}*LAY;;`_%L3)Y%Jjk*dz?rz{`ktorY!
zOjN;Qxj#v0^7D}&e=bBNh1N${xShAtskyMb?RDnM-%~Xw=RU|DoVFgAip*Y6K7KXV
z`&5c^?cDtA;Dy$UQ}@);GZB$FCZhJbkji|RTO5zFe>|f4*|F#cpU;lzX_^06XQ)he
zM!h@}vAQ}hm}N7|Cg+b@tk}IhWcHhKyC8ETPiK0?RNyA*@C9brpu4uW)$;s>saId_
zNq-bwY4@{X-O9Hq)qj_fbv;O;xok?UA>o`s`tIoH8WPWb08O;msC_x|Xv;msQ|wQz
zq>~MIVhl}pZ$#ZEN;@q$6r^u(T*jkYPkc!{vU<aU^?@~S?DBHtL+}oJGLt+LSX$W6
zrA6(s{l6Pwk}C)1{xyWvX4W360JSfXg!(ypF&|v+p}))BlRkqOgeT#sKYAc~8CRZ@
z5sm)YsMs%TqC8S;NyKy66;n4pT+R88SVQ|YUK-NoA53hS-mU1Qh;^lN%5rQ|jBS_^
z=I-ckV|LXMgwP6c$&1R?n{SFG3$|{~+DV|N{y4HjcubeJy;iySS_v=Z*;r>>3g>kq
zLv20#)cGFR<;9n0q;{1kjS<2?m7=d*K8aQ2ldau;GqQr2=XgBnnmvC-F#06W2J~m<
z@{`)<E8#`k=u1x6OXyn%N{b&fajr04CjTVbIDhc>G2=L{%0`DV<n^2rah5RacdReP
zHr*GoQg!~Lt~D8X@sd8edHN02?w^@Kue`52UAoT3<_f|ZDZmwN?C7A8jR)K9`*jc7
z*JVCPdL7D(1_1m{lU*kVzRUgocwi888y>}?dndBpS-JC6Q4~;R*Q=r-)fs9NkRU29
z1ModTh6RJ`zYCIeC><)5p#o-zch#8~{Cq_f8SE@CyAd~W`8V)EUKv8N^yoMcs8C`E
zQ(Xx*E;85=yFn=nK;RF@W@;$crpUHJR3wm&C`9jGIJ$JZNC;1k_w4E;{LoL603TN4
z_Nc;JU#`YA>98E*RBx6<?{HS-n1l2A8ym$a-r8hiTX)x!K_<ijRg-zL*Tq%{dCA59
zt>lTtyW5<$>Tj97q<=?(V-;tkQJKJ8o=w>jk8X+M5+ywzWU@pOnZno<am{k)F+kz<
z^1x<-yBB$#QPgA35mABL46}Fo!c6a4s<Eh(#~!%>u5phI6O9uWTfjR&uy3#}cj9n^
zYG|JeAnEj_4#HCkkUi5>Z==kBy&rS&Qqtn3D0!)^6CoC)U@XnH`ORHSm+;p(+-_uL
zIE|t_$JfZyqeU7n5bz%8*ffqeowLmhheEXow9iT5zStv6aKlsFe4I{g;@0-hIJq-b
z?#uXqV|KVruVZ<IcEHLxtkvL>S7!~WkGY;Cd5Y+DEFx$dGIW}|??R_~CR21e(a8A6
zz6*ICM^!}3hB|(#c&e0qZG_Ic84cXuEv9o}(C7hfZl7f~TqAgSpZxZ<_D;&T$UlyH
zJAW+w*7Nxb1RKjaX_I)a?P&0pE%IwKA=^Wj1fRbd{xW$LbcoUm3)u6rW%P#bJac?(
zdv2RiNGRA|eEHAzf<N#jVZXmBch6Xq{$!W=C+<!2FO3VGCH-e5@44ZU_WNGiM-%><
zn&09SAr=*_n{TL7_*M{RtFhVhy5GlQ{Mgc=oAje9M{-X5M4h<wA!MJU>1R-oDtI{r
zI#5#4`@3>Z9^OB254g<N7s*+5u*Htx&9hqFXN=XI>Spa90fNj#;C)X#`SX3*2r=)c
z$95VQX*V4ESx5b`-&1s>KC3@(x9i>AdK;$2qRy(N=r>gsd}9xvhgQb(5+|?8UM~vl
z3h79EA7qn}*ET)d^lkSe?~ad_V+oRL4`^VIkz@nOMZT)K!zD#MJFibYyKPKgm#o5i
zrPbx<!#h*;vgTvmg51H&3x6tP!1x05dJDavpLt6C)9};LuUC{FA=ighPkBF;_47Tw
zm%qhpzqd+iaq&rsv19d^$ZxWVqeJ^|e<afd==zKYTb;7yYho~|7IBBhN5%42m62Ps
zJ2Dk^{R0`pi7oRApRC56l`3;?`Gsq@G&`jeFY?A<d4I>ZtZai5E4tqNYX+>Y?Cq_u
z)Zb6%Z}O6VqILd$*Y1m-r!R?Qn!_)OeTWk<SRwy?8U3%fJfbgHdG^25RwXNLud82M
zg;jd|#)5linyj%8X8h(kmAJ~-e2TE>B7H_hJSeLnYe5nvhiLyE;TAZv=Z$WIiF{a_
z#+Zxkvz9y(vNG4Z?`A==W09SJ;FTtfuh|4_keJ~OY?tuy!M!^~7Z$DPorkuE_)6xn
z+IO-(himQKv9@D}wb=2i?-laV{j9hHDIk~4))9q9=7ljw9mMsO{X+Q2tBGTGNeY^}
ztvVmhc2bm$8`lP_Uk`n`v}F;z)Y;{qiA=nL{G5RNTr2bEu6GG>Xw$D{@1T)(KdT1h
zIP%EA47?rY_o7Abrpc?X&F3x6-yUv6Z+^+YY8b;ieZ@Ea$7v&K`lDp7&XIimCx?@V
zR#dP{FDPt{x`C^uY|EPW5w}cA42-Zd;dkwn%E1gn(gkGoJi`7Qo^g9SF?eZ$rsoxu
z-%|0u9J~`Q-kSQ$<BwsZ<_k0CX`(S$I9y1Y47gSDB0RuA2Dm;DZ1QL|lS#I-aZ0w^
zg2<$50DxauIeQEd;02{vcwdJ)CBqDKUC6f5lwP9HINeqO0T0@Oy~Kv>1o1Mq5(~5J
zVZFmKy$&Ji=}ZsMI4a4zLKI2^2`RiR8dDt+EJ<Q)2O($3JI$>@o^R3f9VQxgqi!Bo
zSURvtP-{;pREcGV2h@9}38=*(UfM66IU=Qp&#MG3rWQ+j+SoM>n={h9@be1^^~r{n
z-EA%sp$s?CaxZPS^|@z^z#eQyHfys&Qe7-Iwg14JS3*!ZY!x+jS6u;{9s;myvgqkV
zuqUD0qR^7q;Jv=EG+1FJcqNRU^Kg++S}f08tv9GbOGqu@CmAA2nrS%6KzAYqk=Z3y
zul}Y}g9bq)Fhs0WFhzuftp&b;y|oVzaKE4u6&1@bHST0p_LN6Np!?mX;=#sOnZHnI
zT9C??1ef?;2qPH2y5iZfsi&w2H(0K~D;Lmue|4mse~-(R?nK@!hd@j4A-W!Be@gj*
zgRwZdtxkRFv7Qw^$Rhled|3rg=jb6V5xPY1Tvv?o&8tq^wi&T83WF_Bi1%=1$Ql{9
zH_+-NqDQD88D?loM4VvmLPK?hnqO!fI$AQb&*1T!wHv!zm-gobHuc57I~Wu-uoED1
zz!D{1P%U?2S?^M`2Je6?<`Kqx?<lM1TTS?xXs#cH@bP+|*tP|)5zg;s^K<fvsZS4P
zPU%I4Q=>jXf~)pg^s;h8R#z!~^KQ{oim6gQt7e6Po}ubc@95~OJ9Uix9s~09jW%t*
zLe%-Vq_{-StKr4LDk8Qkmzyo1D?U#LEJkb#uQr{~B1Czjdxu<v-ltl-|JW+Ho4*=z
z>$t1SRXgOyV(|;oy^i982CIuXiz4{pS6jrco*{2)?<73!@(3%7vdTNQbu<3_>1Gd-
zsF0Skn(jfZp!^06Y9H*H-|~;UuV9-lyDX9mypx<^)aSKIhK6A|*bklME6h^)tFIgz
zSPOql?zZ|LqEv0|TKj$BeL^%(`D$oJ`=$3#CU#x#Te1YF<;Jsb#Zrc6e)^)rFvt9)
zk6V0Qn7aC{!n^A*Y}&~UACVyBWx^h53w=H*lsrDsT6Zhs*fn5qpPzl<b(^McBp0d<
zZx#LhykiWzW!I2CZQt_)sl6K_e9iZ+r>3vi?3kW$mc0;<mutIqM(U7<Y$RrSQ^>?>
znQV`JkXMQ|2B$M#>8z_;Oa|8P|10ATi@wI&242bk4vs;H9gpsFa}IZXxq9Nm&%XmO
z(P8x`QmY!47~1jB(Rbpv)ccsxiH!X!2c%OA&Exn8sc7Q;irU5Y$<=caBaR=X?yZYL
z1YyI4!B}~UVr<;dT{8?*IH&i@W=zskVeWDAp}egleGZ221+QiOAHC{5WK|^^EHtpH
zOMh^T*zJ`WNA%sVSiaC|dM){ixRH}iic^FSM9XArzK03ESWINS>1dSNyH936!|9#3
zN{imgiLXSJVJ5vvM*;#PYmh67+v(00i)FLVm@xzNj$qBN7m-?VXFFZyzn-qpQdYDV
zqVNuPhkf_yTsf+K(|?P>GVW1ugQ9-yMV4s!OU?7$Uk}|mH%mf8UnZ|~m0a=DC16_W
z4z_bkN5S{Cr5?-sn_M=RR@#;<G<0y&v5ajKUdvt~b|)%Um`6PQYUU<f<IGo4b$2z1
z33XU;8LV`vq#rMusSa$KV)>;!N($VV2HWWiW!Sz!r1O-j^K{Jr7T{<i8pHJ~z|6^f
z$Ou|Yd&ttja4dCFu7{M^%omD#AmAZcyo<`<Kt*($EfJUi8_$Lqn4M$}qLHzLhc%EN
z6znP_sSc5jm;=*%c<dlLWpu-!eev~{*_3xDg&SVFh?0-)dm^nrykU}ETeBK&FqFwg
zn<H`*toWH^1+|oM(;qi*TBfYC9WxtDB7@%u@h$Y)PM<P0UNd7S8#F6+=AI%t8=2KF
z6KqpNwdBBPPdy&V-LFN1WzNxd^<yO|!TAo#9Fa-eiBxL#8=*pl{El>WtMPj&HkB%l
z(I~?Q=z1wIU4WG*ft2{}Wa5Ym*ICI9tiQ=n8BMxJ1jPGi$0`OVHCK!p9S8V~%a4=5
z)2GGgA@eO6Y6DB_TgHjwS&DD4knZ`CD9toPMNA?~<5l&8v|M2UHCMBuZ?DNN9CVr{
zBi1S`WX`%U>L#qxL?;>H84#3e6FhB-8A7Pd$lk>jGP%8C6*3VtBlk6lDU`IS`ZikX
zcE0UrTxPaYpobj5McJj!o$|`w>ePxk-_3&@q7y_PB2thF_@F3rj`xJe51qYJir@$R
z{&BOOSLy#A*zw<A9er}!VPh|^2?c^?<1*=mhUf@ai)}KMl>cmJ-X+d=T(3an7Wf`N
zeFcBMVAQPfQmI0BSYc+F*f+c)qO~I=2IJg=C@MM5e$|MfJ-p$aA?29VXqvrRdIeuI
zluwp&Zp#<>9U1JN_9!bB*=ZIrc>Y0ICu(FOI^ogl+$dkBhxy6@*O=?C_pJYZZ0xLZ
zG<}rfh2)2{Dbd|}Ws>C#^7Mi2e(>^u<70=T-;#H2nlbfH?MR&%VP%dN&1~pgR_(#Z
zBwkIs)V?0yeExGlb@Kj1#^J{}p_Wnn<0PHCrRk$Hg4TR`VcsVQhmiUWazd#|(U}qQ
z(XOg98Vd<3&xy@PU+rNVSrrWZNO)%ArH*v=3w#%e)^fNG5T+g=Q=JSl5uSJmiLcCC
zYELS2evk2C`|ds@-mhNZ=~q8;w7ckx60c~NW_^>f0s`PLK;24$$q|yi<2hS6Bm+#+
zx$f(i-$IbFuGOPdY4&!A2pljvcxBY{xqR`(!cF;L$*BOVG(x9b{u{})`sg6i2AHMr
z2Pq&g0GF~d!OX~D;jpbuy^ATE>60As0CX!ZL#V7gV+=7z$u`CfB@4~P(%Tvgmxq7n
z2_L!)B;>yIOF#PmENz((;QPRjwz5k#1uOtH;4tY^{0S<qd}IC+<~vAJNG>AhEAWBt
zeq7<t?ugzQuG0r2Cz9THkcj;y@Lo1VgD%iR&fBnsgN?@!cW3~iOt4aMfV(HtDE@%G
z<r^t^f)RZ145dY!VJk_uh{dUcGXF4xp<9sb0y9KQFhoZ&HW3mEnsk@N6uIj=cFuDk
zmLNC><THZHY++eRt^<%`25ilDgFtcEdP}a5q_CZIB17cp*SckVAk_)+U|uqMC_>tT
zsFxecQTIEE=p{kxdsk6VChRSFa0;g5NFGWi1O?M=F(e_7JuL5KZGpU^*%V4s;PbId
zE?W?TLROy~(e*|i^$=`Iax7l!L*;l3OeEI!7jVlcR7XUg@=lqHF>kh@a&G+tN!<5$
z$NBiSo>$?2e@y=Oe;Rv|{GUNcXK9y(##*s3p2OD6KaT_XrvF0cPDsWtDAvf&#5`Z7
z6{3Nr-hR{#?Z-OBLuFSDHow5&yO&9i9G!zauvzrX^B$f!wz;=YSX)pPO3VuGm>t`T
z<igzCX!^eFMZ%WPt?W5qvFc3alH(8YE?LbIq;$_d@vt|Rs*csJax+b6jf2l;!Q-Zw
z@|nn3cQ(tY61iBJMGuU9h4k7T4lY>IakOKC{9187r^u^EWm$jy3trJ_1apXQ)XY(8
zwaulsH|LBAb>ZKrgv?GN;aB3wUw1)_J5Dxlzqk@b5`joZaJ?qrAqv__?{<vX5hTVj
zY)cA4Qt2kz25`F%yKkFE%5lQBiyU+LlxeP0>riuPu}~0NW61+73IlZ;2QRx@JvH~D
zo>@y#o*3i>U*B)u+;H*qpI_sDA?ZsyB+EM(vA8r5o^(g1h$oF~!{v%Ig>TadUP=jd
z2uNxr8WPO}XFk<I_-}JcR>dL?m;QVzp8~0)VF-a~t^^^f`-3frev*b0xROKA&saj4
z*!LfC1Irr?PzT^r0aM&qXRv}maw*wHlyCEmwxoILBxZj{`tzi!-oJ1R(_ZlFe#0kp
zS=w~)-m{k1O9$^h4((O|VI+p#L^2<+VMmhKy8=tGQp#_x-No|Ia8VRP3|=IL9XMg3
zom8gt{=ojzttr+SyObDd9({Ujk(^YEaY9Y^;$`|@B2KRrASy1@i_sy$a2#0Z;5k?t
zBwFGNqj4J;fgsXjaRV#P)2$Ho!hB1wf@8@rWe#Y_1j|^(Gg2&6MAX0Mi9`)A6JVdf
zR9y(ls%}-^f@KngP9XWQRTN2oQQhbc$!JSnxDHi1$W2#ls&F8Jm?GUVpzcpG0@ER8
z1WbL|1w5>TT?$+r8G&CNP>c1s1iq6_A#{lrEIivt@VGTqblk%qrJYa)el1@k{Fr;q
z6_u2A&gc3;Fv~bUX|y<jbV1U5zHoS$tute)x|FT#9#O^Y?p&Jj^6+$h8Rq7EOr*=J
z#fbGC`EtJ*C*t@j<=g{<tLL~DM;@Qc+qOnqu3neA_<7rc%iC+=n)Ry}&ag5}%T}cS
zfdHsrMEx#b%CiiF_~F(ev8u|GOf96E)Zg2oqqK2afiL8L`H}qlX#|h7#`>_o?MK(K
zdMvGYw4{Hywc+@=*8oP<JPB*bF>sP854s?2#?er>rg(!KS!jYeJNyOm`cOofZ`dhA
zH`7qNP>o5YI+Kd9Hl!3buK{~r<+vLPT+KKJ+R+y(xIQa4T8@WiDR%5Okgl>%+LcCQ
zKZ)KGD{Cd~=H`|;W!x2ZQT@x+<BmYK+4#IUlK|0bMfn{^LwB@Ik70{3L-j_QdpecP
zYM^Z|D|U7~$L8<4!Z(>bx~cbSqk(k9HBTn>gQw-3%0#cju(_f|kD7&-ec4MtD!mIC
zkP8pe7ZIvg626MgmTauNLH+h5sngXA<{NM}^Fr^~t&eUU0WBy$fx}q#$xW&Dk*}L-
z=Yn3?nY`<o(lsdJb-ON~(|l*I3T+ClZ9+U2giZqTD2Sk!m2<46K3X<>&4}LDnisGR
z?|-T_F#M;s^d|DAF@}v&gb~~+4v_T4d;mCgiPv@!X9|6GABzZ9u;L~YjhH}dibiMq
zuZOonP*xx&5LgM)$PlP2T|}89L|2gH-}VfgF?JsE4U(;P@vklMg`^Wj)e-QMMzXxm
zgkZr%EaW7E2ThkJvyBjEKoj!%OBV~item5i=%#!<F9(m=o>N|j@+~Wr>rKK~I?V}(
zp>KCIE&yuHgF|^i=;jzo_GQOa$~TTMUg)#^9jB@Z>A~O)Y-uKBKDQXuiu;;^7t2*0
z)2r`hIDNg+3E^uevds^`gKRH?;#LzxjuvcMFC>4FhBM0C<`M(=wo!r#f}x?5$e$CL
zy1o>36`0}sbygb(qV8hb0iha^wSd?hovwg`_|XU|5J>MYN?g@MuN~V|8Jlvrwj=;7
zz1Appv3Q&_Q#BUP6peG6auB}$9#AE42PD{1sc3PV0$*E7?xtABcOnD2f?U2Vl%oam
zLGp2IOXUwZuHsm-G-RHk^f-ina{(cwAcoB~c7t4Aqj<}>1R;KP@bLjs$SShz%w}v=
z2LxJGNw=fJ)Yhr(lu6s09=`R>+t#hlb)XK>j%>nQ@#sJD33Kq=Q;#Lz8&#HdPmURy
z$Vk0wJ*jthu-fwUvsXi|P2W-v``GhO*dBbn&8oi^{&?AZXU^`i@XeP+iaV}0A77bx
zD)t}TOG0ymYfa)g$H>E`?nPJCbsXtW{EDrY<u{%G?ZV%8k2d5XCqBXXS$OZj(kJP`
zlw*gs9{>G8<XrE$b(5=+?7g3`5qmeDH`*BF{oVuSM=#$er^j#hnlYKyh&+F?q+qa4
z(6KJ)h|(k7;PYmGd~(dGp~knA|53LjT0hDCep>FzV6kH4L1s>Z6~<k~<{t?3itlLy
z)BY>nHt?~g+Poo%DzI&Nk_T>cKeTLQ@IrI9vO@R4k4q=V`OL4CHJNNYY0lpNK3u_t
zM^&kI8@&ARbx_0}g^6{J**k^rQb&>MiS?37)dcUdKT=P^yJ+E&k+DXPDY0gbf$ydC
z3W{c~iDjA`+_=cIhZ>H%D#rJ!Cy3qkh@1QEGc^)<<4e+J=p_<5%_?{%zIOg~u>NV?
zrXvT3;#H=^>&<d!2wFSWIUzH+FHhxdbW8`?hs3n7eXJr?g#3RG1`$MI?JnmvWC$F;
z-mToz`fu5-cy}%{wUzCllew@uvR5Vb(H}LxrAuOee7AwXpm4%s%cbZ|7tEI3dwqUz
z@%2vr>%Hzt3}(8)b<BqYB`zGaqJBp@9HsYN5VbjbShFa3V%NCNYH9ip9#)nHA_(-`
zc%`^yUz}ikvi|yyzKIIcycI9&%wgGBcP(C9oA$AS!HmJ=@xWs02U3#~TV2SdAcCS@
zkPOY$PUD!WUl+SelA9)PesnBI+Q_$~GDWZ15_#iE;f>&W%*DOF9bq2QLZ2+Wy=L`~
zb%lEBw;DE&OZPa~-qSkyvpp=up{?bx1x$W{c1@=>X*8^t5v9y~_Zg#p!*s#(cT;)j
ztv%^nJ%vNY#`0Rad>Svy_1^71xs@Zya@)sWmf`8nuf>G5;h*ly9zE<Exb3+Z0>cVb
zAedB`5L|XRB1p2g*s7H5F4o?Q$yLAhJNw++oyzrnBv^PJeeEszZ$Sgf_)H9e@D@{q
z%$N`&gd&A8YB+&NPk%>|(#)tAY7}8Zh`s*n>E7AKoSnWwl3}GWD5w;Yq?AdMO)Xt-
zh=rjLs}Lw77Y+fPI2sOCsD`>OjhKsLiypG^r4nobgvCGy1*uJuBz<8BV_y<%z1j$(
zt(OsJm!?(SOm4-_@ah%ASW-Wu@5K=}z=_tkifIy4Z7OzC+hGd~nW^Y5tA4idt{&c+
zB@fIh83_!Ej8aq3&t7;yyqa2ej^>!}dqq4F1W(cRWNp7#-0&QdsF%fACNOL`2I@GO
zRwNN4&3A)C8S%nTX=dw;;Mfv6Wf_+u{k%zBC}j&3x5T67d?U@t;iU9v(Gv41lCf@Z
zORjKihv!(mk?`<hAw&w?F|yPFEX&h2g)G}57VIXBrQsZ<bml3gB5UaSc67D_WMkPb
zgd{tA=S2N<erd|?%EioAL}6g+&@pe8t+{Vu4w#3Va0tm*wy9E6fV%G{IlSz>OEsB$
zN=FnRx!fx@n%9R}0Sh!pfTV|&R~t^E3$g2E938SHKaRGw2q#`Apc6sKdyZb;oa<Zt
zHDeQKfrJ_SKlUg?Cz9{oZ+31PJz;GkQfw-JOIdtJM(5?HsfL2EZRGX}A*W3vDvy^j
zdr&hD@kezt(zLGHD82!vJ3!h?_%V9FiF|J>x9mYkt~1o&_IALke}NOk7bKYcq0UEM
z$&}J>LkA%Y-Oq?2q`FRR3H>=M{tqPOAIRHCzo)cwugDYUmV1a92Kgl(6X$O1QDN_$
zY>ftf^8-h37bF*)4vF*p7Cvy}U6p>uhD2?8l8jV)^Gd#sR<87fOoeymw@4d7$9;ss
z>-AlS`l1guRC#nidQtVA#lglUZuO~>mg@|7SU`{@Zl**WedY`n(3JV5;hhar9mjWm
zWg>TeZm9^l{q(e0tYUun-JcZtBKGI1%;^N?uB4;UC+=rHQmawW_<0<YbG!Lx{u7z=
z>rq|@1-1A2QLlPBs<c1(Yh3WR*ya*MnHrqgFIT;F+O}Q);^x~2^Ao(PD}@KoT&Z5O
z#9o}@9zS4RnQ>#b@JH=p!LfuZ6oE?EjT$%Wsu$~#Z=Jp5t6r^ZE&GKv94+&`%;`A(
z{UIX5$KwMenQ!kJd4bHr{m(6XY4=VnuDUh!Z<+UE@R1K*`7`TmXL8-k7vIIsO#>E;
zb7fkFV#uT8_|iAp{M+DB_Xz=|2L}*+^$y>g{8$yKn!osbw!SRmAzRq2F;ptB3SASS
zT&wE5(S#DUc61On0S<AIADv{;*19&bpFnC-Do*2TV^#0)zF3z-wBe%kG&d~-R6pDV
zrBsu9HnDM9X&EnGv8UtaN9EP|+=K`oYSgKwPl+aS#Kz@O)dugTd3`shyk0yaN7sJ7
zVs~Yi+F+CXi1D{D=VbG}dEUn)4-fW7z!Yum7wqm&ZHX2WR9cxNsrQ?1^L6jP6|_w}
z_`_ZD^tYf9qfJsnvxQe9Yyu-8P)*No`zj%3z@%=xF-?uEn{lm_ZBU&1+Md+4Q|I`B
z#nLSKnq@XtkYIv$K_irF3p3aV$hU1FZ)fuk`&KsYscqIN40>|d(fW(Wdt>iMr;txw
z^t3==y4UX&DI9R+ZEUo90i8GPt_D!u1(k-tWyO5$=s^J(8;%CVtbUXXSuEuW#kr2|
zaMWP3G$=MKb!}39sknbB&z!*ogexireMS_Lf#g#eOiR_JID!(~Vj>$Yg@I`EHX35j
zSCL1_N_@cuH<7l9TRP?NTdlvh@17DmMF;2F){Bk3ieXvSF@0B&KPSTTLCV#%sou&<
z`#1*16shVE+M<ZO_Gvyf_mlaaTg(r5(Ph^vGe%IFPt?;>tyr<pmunBcV18-D9j08I
zx}+cZrV|^&#jlywA>jQIeN2(X6xGXvnMYh~LG_;3fRGHM3>2dCgzlkqv62P44H1x4
ziug%U8-yg;R|(WpHqT$Q;$mXy6tfi$WH&$>ZUdb_sU2?EYgoupExM7Lx@_(mdC~zh
z8T~*8+wfX#T}o1$4lw<zXlPSK62m4%N<;;x!(YNXGi-%RY<>`hQ-n&dCb5J>5ak?U
zA#{xc#OybE$AQesFRXO!&p5b)&P)x2OH6yMw2TYprKsXx7+|D;*35hfmr{j34xi&f
zwqVo43VY$l><D5zPd`!hJ68h^Hk0o9wHwl{l&<P$KXOe^nSOWciB1+|?TSyZ+;lES
zM{sa^mTR7x*TsM&9cJ}>=~JT)W6Jn#qdt*|PZg*3O}??RnO79dcjZMn-0tgzC$fXQ
z97zcY+q6HsWEZ=vHX7BG7vA@9k*5SZ*%v)Q29{!=_wIE(+_GX^z4Pm*v&+AajGpyq
zsK-2ib^MdDCFWt6xlW4e8_uimYLVY@Z^9P3_5etl&F~xzQsw&dS>31CJyrI9mht#s
zP1lFti}v_oV=Pnqb;(~fsWWd)qHG@w^%s$o55Jk}f4De9X!ORshnRTSr#`@IC!KuC
zTeS;3>$v05hCfBI{wvnxx&Po!8Hd^f{twS`OfSbCpGz6KuDp0_^4bUWf=8LFiTs~U
z8p59taF4^{3sq$1E+@UZs+Ja;y?JtD*zLKbJE6c2CRg-<P`qzqCut{or|yB?!9KL`
zC(i!ARnY&rm<ApH{jN*?q_v*%X~(s1u4D-W0@MM*&Sl+S==}$B?_7Pvr<qt<ZD{wm
zKVCGFI7>(#zgU>>#9(xU&E8ycR&A~p5gg@dnh&{<wW}jCiQT^%mz-<4Ghx+TCX$So
z4rX#x9OtfiY=lg!Z(pl`&9FlM*l&$_Zj9V8QBc_0HJd8)@?f!}lW&YU!!8T4q1);l
zk8~c~VO+J&9T-d&RA<$GKDk|K7zw_qINqCMd|_Uxp_9v!6S8aTjj76<F&A?2N&gLO
zuwqF2RA9;qtz0HqL<H($#)Rh5<7anpWe|5Agv_^P@2`E~^r>|2zT!euzshJp;UIIq
z-PG=HDg||2RH-IKTKsA==hveH(-U4fF}f^5&r1+Z56jYdw{O@wPd8$C7%W_26k~!d
z7_i{T53ztJ>i8~Lp{dS{v0w5tTZ|qQm`n35lnl!yT8qNc9MS;^&oa-76qFKRK3|#0
zD1tNu(?qr-My{uafi(f-6mOmi7qv(of_n+5Mnl+g>n4d!G=kSAC>ZABY&MvS<^7dh
znJz5k4mei^(iF#+=341zjQ1%VD9eynZR@(;tiO2hhSs)ZiAJL!VGmyk#77q}wSPq9
zEZOWwDWsn5h-L~?>y#P66Uk3Fd;jd&S3uwk-(LHa6G{<tP`a9tg&2mI)hRJ>XDEB&
z5L<JO2E#6H$+ZNW$_VBkEW(I7N!(8hFcrqGR1z_{Sz?_=dhv@?Y4bshb2klV=#(ri
zgODWEDIuPxla!zz@cp!471T1W1+7*ZtVx}8zf57Mhz3r~fcho0&Wfl$)T+bvKnTgQ
zj=PxxZ91#u9ZELic$^cE9Ei-Rz+XGH3T?ev2vAWPOEh7*K9<nxI@27<1$z(oQy^a-
z%YoHrlabXfF?%LN)8Dyaf9{ihfLbio)#fS9#JM~w5~SLrscg+$Vy{<ja4W!?E4l{5
zD!W%vO8vR~ZbrZzaIypq{{y)LfHxN(Jl^qtuqD`MW_|@!{}leJWPQC{-h*H!Ei9V)
z4@7<8Ke(Gck+%MG5@TMZ91CHB;!wt_o0DY{`Aoz?^3bLHd5<_EU|vX$LkVA5)SHZ&
zgfP0B4k`Fzy<rl^RbS>Y8Zeno?AD562H)E;7#Z3Y9D3A#XzJFWdGJP4c{yXgv@-<%
z{1$eO0f$h$(%gJ@#JB-wmU`i~R!j4m+QON|+dIV{_#WI2(mCA*_@=cRS0|+mXZ4u#
zM=Abj$dC!%PSE0L)~wy$v9WBUHF^dNG0e_(yd(ru*~*JlwL>3w{d&~sbWC}q`LjQ`
z?zP&}Xi9F|Ln5{#Af5z3I``MB;c34J5&P<n%75K|{I`ZOtOFP#Cx7+a+A_;%sao6_
z18~r_<?JOKLl{v@&CmtoDb6n2;&S*_1aq-a+~T7v07(s^&!C{O9FZR+C<-f;ZMJy%
zDMH?|9znBJacsr`>s*I85k0{U0DrTkYN8w~tV%;qmKqp32+QXJ@$7oJP6`C$n8ecP
zuiqJgGdu2d_r#Zl&^y874&H77&h*enQ8x?!f!y%%ntUU%?BqMlPBCjp`@z@tOXg8&
zLj7spq=L$A5fH}0;;~kpDF=vBgCtiZ+n9YQEfvC&T)l(MFAKeY_Yhz}dgFgB+*KY+
zwv}B2b$U^37DAkC_Q4KGv%8+`2}HcRD4U4F$<;0impY(?AmbmTD<oNIiV_imki!Rk
zWiX9WEr`g1SRBOq?anF`6ulqF5rq<&uoMO5`UL<J#s+{HkqAkFEhKAhC5H=;f|wQI
z#SjQFL_bF;kV?Zrtuc7JgU8*J09(4(j8$Z#4wKrNq*FdvG=`{2TrBbtLN@@fH<%t_
zHo57-3fX%{>ATP=^Jv6ppz|0>!}o*7E#3}B5of^WOH*Y!)m2b^Y^;V=`zfvsVo*an
z288d=aX<dy0{H#Su<eR9BmeW;_y3Zu2EMRk<4WNIZN$|p@z?gNsg-8~dl#8?rZ};m
zX8%?*B5FiF`|R%QZV>HeMo);J+U*T*N-{*i%X}SR0m-yL<s}kF_VFAy0wTIyha~JG
ze$<<;vo{53+}Y|DdNKCm*3G^%?Up4spO1A0ihh{7QC_Y%pW!zlZ%JkjE#qys*$__l
zer-mJmoEq^k5rOSmSH`|;Qcq;rJeL6@yh#>x6TE3N*&hK($cA@iYU^~;dm(ZF&xsd
zf7XZ9%o}pSRfc0(Z=~|6Hg~b<Q8Fx^ca;2bZ1YUT<)GIO-%cuysU@|{0%!yRUi_fA
z5K<3<{3<9VA3-`Wt{44?vz_T)jU)*&v#eWmrf0rrHVTf#DM#zX+L&l(`C{BOvV48x
zQDB_NAW2I>40d1A|65?j0ZAMYL$P6MkQqUzND0hAOGG+X(q$jRE>9<s5s+AS69rju
z8^aEuwC<XLybVDN?)PInK=L!Lk0H8bvqT7ji(^4toWWDVups2^#WZ8QUERYATW0o?
z>fmCGD>*G<)O{P9r-S}%91C`88!e(GxI7;0F9q#hsLgIw8ik>x5{cw@7da1&Qg*o5
z&XID7yBxG=_al|!MNR_a9~Vpha=>YtHqO~4Y~6G(3{dpXzURl`HVJ4VMhjTM!zFz1
znj=~{Pgq4tD;Ud7A2Vll>0(oeXPTbBg_>Sq>H{*iaMw#0e`?Jx8Q(pA0bC(u{1PE-
zC;}2uQpwMc!@&}nB43wqn|ct!I5w<AnIUY3WyUk~HIV#Mi&SVZeP6T%5hDRBu_6ln
z<SIcxR?<<*a0Sp0vS9Qi27$29su4q;ghh`zVBfz4YE`cLnZ_Z&p&LY7aMgtoF%BA;
zdo6ZT$C6=MKNDCheT(H;C2XS-yr?eif3Ww~VNrJPzWC4$3Iih2Al(dI14uWBbVvvT
z(%mK9A<BR-NDD(qDJ>}?HKcS%NlQrkdwAb>e{1jkJ^NhO?~ik?>-+;V&ziN?z3ykl
zz3%wjSoCtVY(ub6rwzCpC=LVzCkP-$7+w$ROiWSO{07x=qyV>Qn*WVeAkByOsoa%U
z^iO<}nf$HE9H!?~;a?X4({V}iIpCsUA1cODitM9%iOrrLDmkJNwmvOx^Tc+r<CRo3
zOdxK#jKFgI{RmNC5z7N9L8A;B>AD>ZATCjgO;)*1834U__t8~q3+$=V$%`sP4scpj
zn<)OO(4H4G8g@7QIIOBKRbv!mo-^VH9%aPpUwIjY1|B4(pYG7Z1+sZDGkHXwRH>k>
zenQSNh$rgq4r;fE7P|OPZ`>czlNdw$qOygmC=p!`BkScbpLmqq2WAI@R^*Oy^!+M3
z4_e>~j0LJcsq`?OTXj+EV)WE^1~0lwoJYM+RvpZYf(t6=vlTKZ+(!U$Mi}^`qc*cX
zjhCJ9IoJ=E+l(btD<S290Z+sRZMgx9QO4c4C(LYv5(xls@JCrV79mcYx-O-6VX+!M
zM=YTYkSgW$Z3?iVn~wwTz92Vw%DSi&AOLOzqYrT0B_Eeh1#*Esjgw=KQP=AMMOy)2
zay^o|UO-}5-8hhIA!|K<C=i=VuN$T;U!$U~uA{l7oTm10(TDO^Wo{j?DtH{Cs=JND
zemLazCR>x_9^2r5?RMuE;>ZQ$%`#Dhx?aL#UEVKJ9?&@VaOyE7a2z2jPM*3jM}M2e
zHih~SZvRAzRh40fY+C=zTdW?I82~;}epnUD6^KEOMOcye6C{Od^UwAOCHd>3MQ^Dj
zz%ki(HHtB67P3>}^3=*5!>Z1QIrx~o;1}B333Ys;E1^_GU_ugb0pbLKo##)P;06r~
zZqeljf*os$hN*$PnPP==9<m11n*53im{GDT5hwtQ0Obbl$ccloQUa2r+aLotlt1Y>
z2Sbio5db4Q&QHW|=XLPGDIH`;1X8T8yv9}F8Dhi4@<6>daAJrKW_Q&9NzkxY03IG6
zdU(w<UO-SlAmkunvzxLNt-mhyaWMl9b}pz-as&e~t++k*$~!yw4GP1*#TR*oMPBwl
zJ{hDLj`XJ=UjRFzZ{WjmiGZ7o$f2Nj&vmkvkA8zlJKK*(07#z^>zOQTHEJ|jQ!mL8
zmM6JhxQ=2u{3|K+-~#H$g7(Rnx@m4kFRcV8GeB<2FjGHMl&g50`qQz)lG*Sn{$xEB
zXNV^gNUzyI)|Oj+W%h?7ljy}0y*Vwnjs!Q|e2ZR$d|aqlaab-p0?b<i=3COEVxkZB
zx{px=VGLmV)ST?LDzugPg<^sK<}-5Xsh*X5k5DfZsq^*oF!+rIKQSggTrf<Ij4NM=
z?>@O4wdl4^ziT-vRk@70;(E(Ercty|;7fNwBjkPZTq7#9W*NTiJmCzwIj<<DEVb>n
z<%Ukru&05wA>d~zofU>EYYU1CXT?A~jy^oY>&3i{NfUgdT+I1VW`3M1Zncg=C>!Fb
zs#qc4zA6zPX~Gzkau7R<F^P>O1y7NP9!+x>YN$S<fq@TLJMQSN6ruxX2|_Z?nSELv
z4PpbgNzze6rOHl03LqWYvPIN-cUAzbZQ8olQcCY)%hN*iux+g#dSZ5d^;8Xd2v$7Y
z*tu12%M5n~zz?dP_NB)kn6!wQTF9O*smag+cXA3cZ1u=UMWWNT9hZPH?s%?bv3F`S
zW+Y|~3bI1bi8Jb%s7p|l-_UI^7G4iO8d(&UtNXOUQaT8zKOngLxpm>8WKKXfAN4UU
zDK=n*_YyHgkj=mW_j{`xU0r6mX~+w025(aKB4R~006=Jd8iB<UI;YKtsRJBx_(WKM
z#g|iK8VjAwnT<0y0~DqV0o0xr%Y{P{+91tU==UQi8og(slpu76$F(kN*f`}Ly-w<5
zK2h+PvO2kkJpBm}QX7~2FpoB)BOR`84Nr}4fPaca2bEBd$kGt;-Sx;pIX{fDr`)l6
zg`I-Es4D*nkzWE=r+O3z+Lxssqs$J7<A-AN_pnN5;G(Zbq@qF})NOcB<yZU$^`!!S
z5qh^M6|c#K3Rh;<SnMaYKYrm~I*c3NP<(azo$i=0`R1mcZZ;)kL_p^9V!XmYG{hOO
z%s_z)$Nsj?^+%j*3?=~f52+B5BhyI#8-%b4`E^oyGZ%3a1~k4n`S#P$-)sM4Rm%;M
z-On<c_vx;$9sZ$awMX=S(XF9a*QGnj1-8;}P>6Bz-|BHuFaMnarN7mqAa?yb0R?}r
z7?1OR7V&FRiq|eJ$jmZO!rh-~YBn2b4sQfZnwg{bsoF`%a~g%CJ56R&aDT9)hl>>o
za&|2qE(Q7IZ5ea<iGeBj!o|obOoCMqgxj^3D~JOQ!pAj*JS*b5`~79lO2sh!A4k9k
zGuQ#BtldI>0x{|FC6ALb%alSly>Ber85by%jMOG<B8kZ+0>P6f*E3bEwpfF<tfkTK
zo)zkZ>7q3=!SzMK{nT_);$v|X<{!X>F_mK)y}muV58n@_t^5%1$1sm~{uW7c$=DdR
zPEIq*a37Kgs8O?MVMHJ{<LZba4BOWovy)LDAFWD`F(V&-O>!02F8wNoc>5@Z*Wooa
zF&pKQhYI`kEQyY?Qsj22fx$<*UHI3h@K}nN%~CElM9SOOhHQm=C5qmkk(>4y6|s$v
zkU!Wb@|);ewFFQK`wbt1qsavxDT|NlLNm0{RPE8&mB68F#eCGdBj}!m`?4Zn-8^Gp
z**1-6zeZ0NgW&4I08<SB<SVL6MH+`K0x1xB6sJIsNtFUVE~w|T8jFT><HWgcfOpVx
zs05)sMc^nLAR>_^sI22qT^3NydKGVgyJ7aB^<5$e<V<}{g*X1^Nn$(wW3bxO{>^6_
z;CgJUh(REZhm^fU8)L$^bc0eJYP=pmk`32ny8~bj=gY2F8Rn>m!^wx`Mv}NmsGsC-
z$r1|gD!Kx4zy^3PJe10>2mrp0#9{yf=X;MPQyK!a`32ME@zIxX0R9fEVv0D?uoG%)
zAs>81U}yG2Lc~Gp-?g!^Ko33?0Xb!O+=kc|;UKed3QS9FblSRS0M-Efiv~Iwkoyv0
z364;rKSqF7QQ2p@G~JLlWw_#SQxDBmDowWa&^w>I!ql4MFj;&d9nw~?1YZ|_ogOtb
zzin9rWmnEvBs$>+lBL59+J!*CLBP(-<qw2N&qS^w(eF7s2x!`)eF-+!83gf2jA;TF
zNgPNt1k_qCVJx%BdcCaS%pS<Kv5C~w{$GQE0>(6aHgsiK?jDQi`acY$nH>+^@dgRt
z^8ClxoF!>8x0%P@n${E!SR0*5*Q*+tu+|{(1!}nWm5Diw+z8imu+$*;r0|M};8FZ|
zpCBh&%9b!dv(d{Yd8ywjlAWK_nKoDWd~euT+uhMUVbEZozT)uKRq;TjXh;GjXBC#h
zp~bs4s3Q!@_bcI1*(k!G(CNT_T>xYjja{Ch39es+?Y4c9^f!1!?4M1+j>lU6-174<
zQbck@`MB&+Twc#02Ia3uiSOKWN_oBw+23`5Xbbp{hmJrbskHry`EaOprB#7LR#K}w
zGOEfgch8YJRfp*jB+f}zHjZDGaV0Hf<kprFx^;XmIlwmDz%l5MnAs^h_rgDJjYT(I
z{DMkdmuZlk#RmW^;JJ02u}AYq(XtcjC_%ch6Mz(>7;*Aj&=IQ$0AoU@M>{+Mh|Psi
zS%KU%!Etv`fRhm5S9b)B1o%XM2S`z4+0&=0^Xc84Ru~k}&!=btsJp~q+6+)uq2zcd
zAzZ9DR0oKl*Mkm%oa|x0W)F~oPvWf&ci5;WDc*f+e!9t)*_GJ?k{Amu6}(qZKcs+}
z&b9&f?Q<|-3tqw!@Ihyk7$e9+QNy1Fv4gBKQ;a*~#>SrVa)>;A94mECIbxt%+dtO<
zhz9b$^8!VyfSHH^HwL)?`OkYme69B(prSHSFj45IB2K`5sWGGE0CL7HK!SJ!b(Hff
z1ng}Qa#%eK!$lw-Vm4wD>sxVBT@Ot&Ko<i_T=*S`rj@52i`Qnr!w!|QM>7G>S%pO{
z97k$^_#$#MvbBmhL9dIL<Iq=#-8C?Pgh)@nqh5S2Rz_IC4RWzC2rZ8bE$4vv2<pPH
z{z0sFs$@P=>>ZbS%!<HL1B}H<$59yAfJo10YXw908Dl<}#fXDZqwaj@D7J}Ew}}uf
zt_NWh^Fvo&V|Z4~1J^M`>aAXa#R4x(!QKN4P(uAraH&b<1SaxHk2iqx0C0OUNlwt~
z@8wYC^OU>>jFW&S+x|FzPmk&4D^VJUobOJ*3L*MW)0MSR+ze1Aq(pw1ZHyL6iqJFu
zfe1}zG$_%0E)gZ7kR#VoO_i?IdRK*8RWTo2ZZciKsqeBFHiwU(OgH^$oV41fD(AK@
z_b^Fmz|y-Me;%#8hEWDmI^V2Id9soFHpex<vpAJISsR<*tT)5QP%)_7H|NOygPW4W
zH|kX)NJ<@f5U%-Ba<e+2;ZHwA+G8A~YStiLOm#hrLsTTD&Ic6b$3kbo$imbssf+F?
zf^q$>3$W_Pyn*rS8a_*lCuJ)jrw2eAKg^KEVY=AAcdqBW9y8W|H$5YP<zbu=Y_^m8
zVm%|0J1*m~>1Q_Vu=vSzFd@7B9UVB71`rj=REhXa)mOrAv*_91(ap^m0ho;sWc&pe
z=dPklj4EUHBj_9+Rd1`oEqQ`qf;|}2><au*uCR_V(5FE*1Q(AzBoqpG?^2|LKOvYF
z!rc`5@3_OD>!Bibk9%tA^o;f20;HfV83+ZgRel%Se-z0G3M%V@hvO_+r5B;!gXAYI
ztK<E|EoJTflATb*s-q6Z>X}!iq~an<rmb(zXMMedworCw=wj|QLUva#7JO3N&F-po
zt2~ffNo~?aL--Vl_OhXs01hLDD0d=|1QYVX(PSuCh>7|L+9d{lfp&(X%!c9b?yF;N
zAQ{#}@!3JvKMU!t!eZmJvLU!)gMh}mXI}QZHs~Xb{UAnl{P<XG2ZA<2{$jK(T#1uR
zRt#aOHJ~!$g^)M&*YgI1(3IV}ND?GxhR8Ga!n^8nM+sr7fcxc$uDcehxCd~C&qIZR
ztsxbDeModNv6T!$d@HK%(-N}2R(dJ{5=hZCJ0bay_?RIOn<>nrKo7(h@i07~PX#;)
zdL=w`njwbvI!c;##2t!HYQ0|e#LDdAQ+yv-wv$nCaF6_-t^}yS1OVJ|f4w#nNOi6j
zhb_f6C#b6vk$;CvGvsz9eA!-uKd&sJnaiCpS3z?Z)E#vju~c>IVT|kp(sEAb*CaBg
z5PA<nS$K9_*_BiYKD2%)P9FqC^J3T6z4+|?Rw6Tzi+V)ixv;f^lGm*Dh(44rX#3tr
zekVYYfD+4R4AY4S=Wq?u5Ktl{>r>VHypmGM_9ZON+m++3noVk>d=h8f$Qr2;+Lm$8
zYYhkcR5%Gide3u9gUn@vu)!xVnzLfS-9bajl?bGEzoRT=gwfIg>h>5t9E-<o5QFF&
zI9e2f`;+c|FPTZkFP4wJw<!+lo)wO(vOSDP#ic-DY!C3+?jrML_GzSQP4o-Bg_1tS
z-KKeSMNY>(tX`{uyQyeffpvjIZk?~dk*)=CAXWa`E#_IAHCS5%9bXlb8%#u?vJe&z
zH?PB$g96dU_t(=UQFrP{{6XFdJc8;_(Gop2b=|O}IuK&C>2)M_x)6{nGcycLv;czD
zJ+H_f13*Djk6{#y-E&F{m}ogNX0okLn8t*!6-%sp>;staP|>0-6e86_6-&UFmR%Gu
zAkpM<+|OPsddl+H*&l_boX-y+9aSIr8|W-|pTBe)-++vA?STUk0lK@u&6K7-gi6dj
zBNX=Rq##&{F(vXz55p%&D1Ukn8hhjm&IlF2RR9b`rOd!U2VFF5^Rmk7BgFnAeB4_a
zzy|+qCA}U*54MiOW^tPY#*o0jke&|3QI+@(2Z=gokO2u1G%(;G)4a}54=8^HS`mI*
z2Cl@)n-m-%h|V3>8%ETNi#P;~K)!`M;(*VOo3h^~EEHjB6^`(L5&_CRo)t?fT*|jy
zR6t(n5r%saT9R>ccifbv?C(UD3EOD%%OxFKdLo`bBXRHnoa4P=;#45-9eg>#LG2N=
zI3QUsC_tac@kWy5vNQ5yPJ`GFPQ9G}*5CDC3et#+>liKM2j%A$9^BKmRa6kA=j<~M
zPqI@Po=ze;gn#Hs#?Iy}!+gz(gg$;&RQpH_M$BQtrtJgZF|ib4ZmBgDlHP#}zzB76
zGZLXA?}9jely$iWXxDRmAc~Rn>kabuoh@P?Ly$_LAgS&94qs9But78;X0!gsp{-(t
z^ea&edX-WR&k~`OkQWT+P#O>DCn?eLM`UGsVKYq@#9xKqZ-hFmJgOJYwKa|`vRj(`
zykQ?d&@Wbo&Jp)kxlmM+3cLYE<46babnr5+>e(RO+-zdSXsQ|vxkjtgr$1)nu9r<C
z07OkHhPuZT;14TdxEkOW+z)Y|Dzc&PrLzpI(JibriK8T_<Si?ZZi4tS`T09~)7@`t
z`6<#~1JnF%W(qEhUma02$eH#K46=iKEWTCvzedP~$!<eO3W_KJEwv5`h{Feis8q4h
zGjy4?+2`3fx@wih$I$wU7$mjPQdQ!tX!(^DK-_>SMg;Q41B(APV@wWfy$}O#>DgfI
zRAD_+GH9QeO7y9lI0(z*{(1kevQz<@nS}MXokY_moNns!jaC_;@t&vMHdcrn&(lH*
zjs`&`Bt7A13Y)xj*l}f7X9rEz4~{X0H=!%9sMoXTwTkFHl(D4i3X15dG&Ny#n7U{M
zO!*~yPsh^LVNs_t{@JO}>7@=gSxIeWlxSfOLzG8HaZVfvi+w#T4%EZq252kg0m)lX
zkqrbS#vleHywl<xAato?CA|teEP{=ohn^AvN@6>&f#OPSX@PN_i9lqNVM$v(tgo@q
zt*}o@(1CysAR{?04kZFhw1+_lx&t{>^Z;CkQ-Pspgh9JIm?(Q7ynqIK2T=xO7okUC
zY$<2QEdl{kKq4XaG<E?64025p)h8Pz{*R`g@xk1FaHK%!O%vb>)HZr(O4O)TA5C_F
zs7CU<ik#_N_b&H`HDRhTDDk)KcGi*l_n9y7$JcrLm#-ZtXH-1nH%AqX;NLnh=2jvP
zP8&9vHE6F>qvQ$f=##(1zx;AZj@;pOVX^0SV$IdR?5DPU9Rw|E9^3ak)o-oM{gz|<
zXjFhy>OC8I77)cNSq|o}OB3sS%g&tl)WBv$=yd^%8O56lw#`E3fTV`l2!yr&86EOJ
z%Vngq*Q>_b{*&%=GZGZz;lk^=Ao7RQ4H?QctC#D@T56BUEG_rK7kkACxei8dv{*c&
z3;CHM97I@A93W>8CG_x=<%Yz%L|v%(h!TXjv$V$rR^U{6F`?i9sW91A`=#h;$!ACZ
zWn&y}xrM5J=0lCWhsA&ei|i)$(EK4*hPL&9_6_@ILpK$?Wa?D7o9S`+qd4$|{w2Xg
z{g6AB4iQD+p+=z{Y(7<4(!`<lF?>{z^=@C!X{+3gXWqdplScj3$ufTTL>F@7yyX&w
zm3V<!E%14aZT<@9-Q)$ri4P~uBKJv43tl^jaLxYkb{8IXZJ9Xx{$$OB5y>xP1^u?I
z7H4&LUT>dTU*&{cJFrcI0&`gAu<!LxB|+jA1B!`k#n{BRMHzrcrj9at95z@RvtjuC
z@`I77C*t@Y;7XS7{0jqi^2R+LpFWsg->gUozXcnSRkWw9!+>_Cg<*j^*Z|?|kDH7F
z&SY3VS`lFS+fs(b#;W3?hoLd1z2GZY2qVExhlCM>>%M?|A=;c_z#RqG72HY$y5Dna
zY*Mr@e!;WH-m1Y*%2syQ{fm?a$n!8kv0y}ZviLJHc=q}CUzC-ukCn5WW$Gb^+6+n%
zryf3PB1tl6G)9j++pw`5M;lu4Mk@y42-R!;0kgbFxdKi>Z_0d9FDEyu6EE;$!|bn&
zU@rq^`H<%_YwyPGwCiPlj_6V)_Ly9ToNvK>W*L~G!NikfyD1x3y?`Csi`bPJ9BC<X
z_tZ_(s6eKhjXd;AYYYS%Q;)m@z{7+_cwopvzzhyRd?tc=1IgoRPpPLuKY}JJok584
z>^47_HQdx+8$HM#fCrl&Vh<QWOMwB}#EUWi5s-d6J^*@-Q;&IvjnXm|?#PBv*9M;y
zE&<S#yL}#NX>3$PeGm$kuxY$O&`S8Dew;W{Kv0aKL;+`I4?d6w7oIL!i0%a@2a_^r
zKY`yvrGQL*5aC1uDR*#-lQJ+c!aZaq_2fQeeJ(%%Xk`J~q7XvdLAz~H3g$mN&rPVW
zO-Fx&z6)9Z23ha}Zjy3!Vv`#oheWni318coD9XR{UXZ+AYB%pna<sbm!AW<fw9g}y
z{}hwoIgNDi^03Y=?9)|{gT_Uxdb1Xly}>JZQFW#U#5{Fv{_WCnuU3)Gwy+&^e0R1x
znWfqCQ0AM-SB5A}Ke6F$S?=OWH$jMtM4hs(K=-@!9OBL<bjA2(^4Xrr3O}hya+&do
z`;{AF3Yh<1up~1%5z5E8wRF`5lm18&$2YC=NYGWFg?R*cxzHy)h-EUh!=JuiDpOb0
zzOiR<xibc-j=DKG`3M-}I}%xE_xi3{ZWHe5JdMB3_xlZ!0nj8C6_a-fhR@Gm-CRH_
zqTQqyeSqhrA&ZZLn)YvknSl#i>6)zTWzgc)RsU_()9v>2v-F#bf|pp+L2ZBiq%5S4
zS7u|~;&KzUCckd=$8+HJ*j5(2_(wM(i;78)sZo1$Gxze2j(D$+9rP!?AZF!3;>pnU
zHz^)!AKGW(Ro9q2GW*K~mmd1S#WS++`U=toH~ZXPw_lrAKalrpkZ#6j@#q)t^(o9f
z-z@&RBv@u`&*gGw#7hFOYVdl5EIvzme9tNbW)U(x-u`Q}=w___Hz@C8`+lw3OK;+5
zyr&HaZ0T8HnMFw6OjGpLjRr6txR`-u^Sw(0ZMvW5B0$R_^(VaNH+eUgkaKOb=+m|W
zfEz#}rNw9ImnJ$gKQDpb*uN)8&E75+6CowLIiYiI0{V!R*?-4Hf9V8FuUEI9rRTl=
z;$p|&KSrHy0;!<=fMP5gwzei3o?GUYb?LRy@T}1t2r5L?Q8zUpN;m`fq=y{7_~w}<
zJQWJ4$3W?6uDk~T%JD2>LN)~fc7`PKUOjAM1b8&De|-Zni{_wOV4s@Zcf1$9&0ieS
zse=3s8hNsPa|7q;dU<DVo9g8f`uc?*bd#q4-9li=jj_mvGFQY<`sMC9$`2J1)Gm(m
zG;m9%ZSf7QXWpXAP?({Xu-1O#hT4Y8_371SwV~u6g{*H_f6X2ttj<v(!-`j%M};?+
z!JdX7Ew(MDU@w`|qBCIZx4eepoEH5C{l#6y@hUUb`td?qL=D1+;m<mM_W9(_pa1`C
zoC0)I_#2d)cs}}9r#OG~R`3UvW7WX7t7Dg;WO1$Y&BZsIAU_lUgQ9q~^=D-XP}?7b
z1H}N@(0{Btg0=xnc&P_R1w*fH{6%gHdH<m$3WC3Vx43N?1aRPJv|Su%ccTCp9mb!v
zl>V#~O7ds_f3gYq_D7w+s(>W7LT>B)McO9l2;jmW0|GE<00Q22&<3{u(e`=!pKTBS
zL6K749|VQo+a<f@>z3r7#h`Zk>>K~a4W$_tpc&-(2-<0E`$l=NXV6=h^B(}Ebg;Y|
z|8}7K8BktzK0-lK2h3+FlcRz^8`>m~ynOr{boA{vNE~avgFZIr?#%?6mCLS^{7nc@
zrwv$6Gr%Gkyj=u7wx*6hd{8`7%~@ePrsEbcpzrn#W!clL-=K~18_#x!yqf?SfO|7Q
z_T${)^U>4t<nx{1Ab<?V4}U#r{^QB72@vlrPsm0yux1`d{a88yS`Y`h<gxz-W!}DB
zwLh4F_#0#Ym{Y(c`lERK(BaiV$h(#-{_82mFv9>x<AR{=%ZQVdF->GCZDTFn*?k1p
zi<^mYoU67!7OzgZ15?q?;;(NAtKXpK4)Qk^0DZv7X2oA!{RU~>4h<yJ0KhW>YZO@0
z?X%)}!1G~X4H2O|Rcw5hd>&&Ove*TkJkc4`1_sg|6=tYY<<M1J;4ZV9UqW)%k*T=M
zQ)age!4>%%^dcF^(uOFSy1s^lbSfc3Ms#l%8Tx0LWv0pYO?F_(hiK3pq;6XrgMm$<
zy*Tte$(iy=)djv;$l@3<4e=;|6gaYKRhJj`1PmenQYqw^{6AMxx?=X9y2j#Z-`_%T
zMcj}|Io^c7viNlj>@#?@e{T%$AKQlAwj6!c^GBtp|MbIu8Up^Oe=LK^EHnOX5NI~Q
zFdSw68pf-?_d7x%7t*N!nBo2$y2amz2Q~U9BJ61y|3<^Vur3d1`ky#d{}+dV!5aT<
zu(t&K$FGvll)du)<Ra8iW|?6!WRvxl*+g^)Dcdr~WPj&d{a-%=ni3zr#@q|pUq)Dk
zbjkt_R+<|(H&g(@KudUYz~t<L;{5fa_WfmG?|Bhn2)x`k!@#NY2WN<DfOjR4cX?+2
zPi0uZTOhK&3HOotbpUKao$|NL&;;fI6*lV?GgaT8odQe$MYIv!5!uJvITZ>_gWGQB
zfvNT%lmVS0Fm6a`?0}~JdZEO%w>?$?Jy!n<Y0-Zp&5(;1p!%;~ZmIrz5$GSrDYHyJ
zDYN;|<CaMvy4j2NYs^Z37X_Cg2EZU+{yE6&_HuxB)my&EG@c}1;F<ox79*U~n<{`3
z&fBryEKUG|(w_~XS>7!EtyNQ0z1n5-DEZ)bR58lCIgK6J!yB3H^{{T&(Tk`PWXx)y
z?N3VVFeG={UO9IbuX`CZ?xw`jk!^)ZNKz(U&(pY~*6P17^x$=V;`1>p%1@X{;@|UB
zxeNry%{aTAijA_Lpmun*+YX%Hay`DZjBC4*p0J&Ga%0HN{Bwn02}9xX!>3+X!JNZw
z)h<04^&Bc|!_Sb7rzb`yRPaf?)DtKDC4oXWWD6%gvt=mgBDb#pqw4<(eAQ;VQE!%u
z_R%oy%C{M}+?N?bK??rnLo!Rx^7lN))G~Zd*8zsf5+q-q!DJdvw17?Jk1ZqB`w<{B
zZTDTauj-fnyUdNGb2FyjJ~K5nX5w&Aa+N+#@u#MpApZ}TnSb^4)9r$JxTm#H{QQzq
zd2V#4TH!==lMS8YHgxMB{l@=q`--F?xRH~;?$U5pZ=>*>u<HEcq<AW!EPS4pcF(P2
zD{t$Gdy%(uNGDFk#q8~TydkGvz?)c#E%b1y_g|5}0awPK>T&=pPpC9xjQ4UMaFSKJ
zVls}t24vZ`-=K?eGo1IS4kTC(Q+M(1e5%X+Nq0~tvy3qr<OG=imw$usc_O|k|I>qi
z`ttuu|NrFGmx(%)Rto!-=AXZmo1=P%$>p+$jHbH}>E;_R8jmZ)qd$&mE#tFOcwLah
zpX`!p*N@e0@3Tz$jr7x{KX4uZ`Ps4-Jr7hm3<ksrJjHy(^@k8UzqmG~bUICvh1uM;
zW3;N(Vena>($s3*=tH;!Ka32G2Cgh%kgH0!VHiTPmiA+~XKe0jOvP&Gdnp~i<l#&x
z#4}4sjU?LJAQv}@%=dm$@fN;{sMH6`V2jQVba%FAqGNa`w;wA<j@vVNE5ER<nGfAR
z*T&IXdecUKVBWs$^XbhU^BKF!wabp+yf=c&-6}=Kj?(NOpo(vtts{KoIbGS%IbHDV
z(|pYby8ZWKHr|UTya&bF*q*<Pk&HSWR!x>Wa84|TU9^=+w{o}T6@Lb+a}(6X`O`GN
zm(D=5vWR-S0%Y7WV=_$o*{46hHYcel&>Vbfz)~x;NaE|G6`Cm6ofzbq9%B=>t8REg
zpXLz<`sjDCz9wnrvV8fVMnVmJ=tw!6LWqeuS4*{)F0@6abWW|r{)Q)MUk~anciKa$
zNF&##X^&1v7$}_A?XOtkcDh(=KK<2F%`Q0>(B<zOZlT`Sx!^SvJw=+DYM!FD_+9K1
z{N_BuNhGzgGz_5Ws;J}H)7JVoUbtW5skL6kmk>(O+8_LiDLI%snm0Bv>GwR`iG!uc
zqy)T==(WeFndIFA&g;m$nOug)0VgeZcCJ>B{m4E50P9_6E(rOmvjW?3?0^7e<7Hsz
zx!G=!S1Mq%)|JV_<0heRO%&yY=;XW4ySeC14~R!USSOtf2a1pGp5)y#>G7k?T)>1C
z%8!O8EREopEw_`Ip3Rz%2Z~fEVK#YvCB23}j5OHUY4<F{5L$EZ9aTMSTu|f)!3XWM
z0TSuwe!Jfw%T1YMs_8SsnK)mc?lc(gU@4cf<xJSn&MD?yD8Xv15X%8g)}hVa-yrAK
zem>Dioo><Z!Xr556Olz%(s_CZX1#5V0jvss^mRXfK6}LEVJCf)fhOO%K>DgX#QsC(
z9w&|TsE?_c+Nf0Mp2J#Sl%x<>L-iIdo8!0QkB0ZDMtY&F528#`p%j&EG{tWM-bv?|
zWe4$UN??sfuV)>0zS~GY8{@iqfZ&aM?zL;(tv6k?la$(<sc6xp{Nuq)DBW^M^yC6(
zyg^!@AxEYr>9@iN^uP^gzjKeiD&v9CKqj67;aaS@K|A4>&yNHse>rnrNhwO#p%oOL
z3Cf4y8E#(;$JmZdp4vs;<@$;sXLgq-(Nc}pKfuiNX>ysaH!*3I&`Y<u*S)JCowx<1
z-m5ARPJK7GwJ)hx-@GFr_f7E^c7xC`E{%|<(euj0@8aFEZ(Vp{2A{F~bF;u`jy=I>
zd0p_{Hzi%3E?74cMwkJv-<PWNP<6jS%cd}X2PyzWi6K)Wh@46AYeRPIv-|X*`40%f
zYr0PttTB=X6sbjyM;gy3cRA-;v*e0GqG8Uy`*1V6j*Fo7XgZTouZD4kI5)yr?wZKP
z0Gpf<%#sfOC+W7Bxon42Tdxf8X}GN#1;xuYO)#y>VY!53C&}bmg@%sdD>D*`A503P
z_boB%Dt))Nq13TUQ?MnObC7TQ%_pVv5ga<;yr9zt&Xo3{*5s9im)K=*ktgrbd<wa&
zY<2o0$M1w-<V}XM1yZ39wOL-}Jr)8#UEUi-(fD$G4@UpEO@g;|u{ISe#I2^EuGKJB
zY4}ks1d?LbY@>~7%o^coardl2icWQEE%O@^{^VnTT%=(LF=|oM!^rqmBrlhwd7yZE
zYgX~y?j>tqsRwkbpt-|)U1G;ocemgyc|oUJ^|%Ak_NSPiHa@N{6k34J+#Qb!P0qfs
zP6Yqxc8ZBDyxF_DxYWjy!s&1`Z+I_+RMg-D<k;<oPwVN!#R^((OWf?lseUdibTTRv
zEPMo2wNCcmY2wny4iWg1H=rybnrGaaJKtttxW;N>w7VN;atTT8#=R5AI4I9u@slbi
zJ2N?6A?w(mhe@JOr;nfn%lmhSmp@kWJl{nw>?xlrKt`kaJO<b#?Whf&R$7=kE)rNz
z1VT$^Vp_gA%4C%OWDO3r(Z*Vg^rvY{_p_~Vs7bE-LHu-V$!%yZ5M@0^jeoRHdbr}G
zMtAh{H)tpnzyUkkFb_K(q<a*;u<+1Ug)wQ~tWrht74x%7-%PvFV79rCPC+~8UjV)|
z1Fn~fk>YMavFsqo{w&kXsEzh(pd%@b#)d_;pj<ge)uEj|wf-QXqsmeu-W7=p?Z(Sj
zdM@lnGwQ!Vcc1=p%gr4}>$_6dkdm|(pMI2@62DR57<uG&WEtdWd-5O;-Iqi$TlubR
zsY6e3?)UD>S+r%vC1B4dp;mh8M-tVd;Rai)EFpGDo5ouBqNu!shk2v#(LH6X@vAdC
zAdybhOnlmbMM!<myW_{;+z5`~+p}52QOIY{c~VjCA_0mtuP}3?eo6MvlKA6iCFD&%
zW&#(~T^F@%(#(~pBGKjN!bV204I%=$9ams3gX?ck`3z^z_Zj8aNd<@Rla~c7akHQm
zn15QKc%=rs*!ZY=w!`ZE>fEe-IH(Gz>pb0HlKX_S7G_u2Yp#|`yZak753+F&t`AT<
zkckx1U^&u>?s>(?Nja&OvsLnaxMXk;Bwy^y+I5kbS^kcMR1^Lz=d<K*P<7hyhY%lc
zC026mtffIZlZi^ku|Vig*rfMOP|tW~<b5TC`qB9awoHuY{5>t<!MEE=dP({&e0A6?
z3E_OW%qQ<Ih<Z~U%PMB(bEY!m3eWkaI|pZ3$F`s}K0RXG8xdVT6b)`hJ<@PyLq^KS
zL&P{6rq5DY-Vm2z!t2PiY)S-r2(X?}E@1$9TzzEjRC2jU-dk=)%4jBQHNF~T#|+#B
zt)|Z%PQF=sqts{MeJXsxU8l3fC){sn!upCHXPu{;RFeqEch3cpl3xj~_pr4Ck4(?@
zOSP71rLO5*FRGuG9-CR{9<E1LecL|lb8X1nb;Df2lE+A1#zgA@`Q`AR*DBGvwT5qu
zv9!{zE<Z~XB0#cCbq}+YxZvbGP5b=T`u^o*41(BrqjSJzDZqs^sG?TOeS}i74O?Pw
z__%mrIWqZ^a4yrO5x0l0Q%f>$WQXM5aHG05_Ean|lsqC8p*O>EJXF^Vq+A$U5{9WM
zBG?_p15541uo;$@rH-^LxhONuX)S6MOB;-F8<AF!Qn9aue0CR(#e}4+(92pgp_Txy
zuaPw)0Cuyky*c5i*T<9CSLSdj5e<bD)9ZxHR4t2iZzVAVo^vpbf+&i0q&}m)c7lhf
zNsU>NF$s;8iM3ittK0nk!D+IK2Te(=@xFF_3e&7megfGdp_V9u*lN0i$6ADA7x(Ep
zrl^&~>O!la$;VL_>AU^H@r8-f^DYzEt4&HBlC}OLIIk*2h`!C{x|x^QdLhiNcn>K@
zJtOS7S7U8*pAm6YXo*QhS0jk^J~buB<zFh~MIyVj7)Y37(VI)R5~@YX$Icoir*sZ^
zBX_*)S2T?E^(TxU;!~#a=eTN0v0=#Z-udBl$a!=3K9}W9mmg7C^`*e>&%<A}2^(!5
z=C&(D;@q6wuP*%ge<JtpgHJjm`haQXoV{d$|HGeqy__VTCompM&*P2gn&8%vhyL+C
z)|vOE*7(VjL$1TE2GT;Sr&emx%h&VYs7EmUNy3jqNYCDC@TyN020UsMxGvijrcNq=
zIp?0gYgoYkC@mS?IaPa(ZD-ktpB-ja6oHnSvpN~din_XMEBq$pox@Cjb#?7W;ZLZ!
zFMA!Udl5Yf)=Q;Ad3UvJ3x6CA7teOHWNW7ibFheg9yXsw5DY9yJ<yL@lN4I99j`*B
zd~8UY?zX|G97~5ss~ngo%Dh;aI(ZXpCNAboh(<ax5H8h)52(7xbE10&3M$L**EtFX
z@?6#4luWr)H8|pQi$94d=^}&+H_c>ub23WR#1u;`rNcoZ2Ndy71EllI9Y)V5#WmV<
z4>qo?L-6w)TIss;?rVuT@vs%t%173Pf1(ny*voq*g9O2@q{m&DT*!Y|LF!vfo@H=!
zgA!|V!5eg{%JnAu_pT(Sc&^l_J*1*~-?#ErIJKl_U_miGuQo9RSFFNFq7FbqnOQxW
zPf4#RYGvz<PIQ}-pI?mNJX|5VdnNNu=*27VU+V9W>n3e$-|YaG^24Np<#}l}Cq_~F
zkC_^S-c|-F$$N<%OWHW{hYZ0}#b?eJKjxVxtGMIMD%>J@KhGEvUpp2bJd1Z+-Y0R$
zG~hs%&%@rEY8HCnTd!GE`GXE~i(BKDgQJ}gO)8~|#O3cylt&+Ry|&aJTHQ2UJwT9%
zjaGST$QU$zn5_FIL@w7$Nq(9i2#5srE`essfbhcKXvm<fQ$asVMaZ!si4WSA)CgKd
za!hGx5%q{x$7>~g88z;sCra|;(D1!%d1X-Xi|Df7=4*nY7;AbQ=!7Q3e7xm85@t*r
z;e-$C#m#zqhcl?6ae+&j`G(v@eUuR;EnvZt2YO_3?+qj9>HSxsIFq5G`;-m~5}BWk
z&R&~7tVsSyp=P*>nUwP`dBZ?bXcd?S1}dX<*^a(WB86IxG)>j7?FfgIht)^0>5`E<
z+<{JIUBw*S?*}?d=Jmo-q-x>K%k8)@|B@)4`dt&_S6bl<n<gs~781;%vP%H0SvzZ&
ziPs_3<^Zx>Xd-WhE2k?NvKsj`FXP7RWr?pe-417>wP|~hFm7e=*EsqkDn*?2Ri+})
zbD6LfCc5a=DU+TlnWt$X3>KG=?`?R!B6c<J4uGSifBA#gpjvnB%<P=zq5S*q&xY(v
zz$WDZy#os-MDbPtvPZ}rp|)h0D}B<EF-4jjRP0->c+$zxw)YbXs`O9AKj>yCKFJzg
zwPp=Jw*eAt{f~SSvHk}hTlSk4KS}BKs2qU|<+a7gWW12i!?rg^A)747oY8sOGJQai
zj5BZ+!f*8E85S6=xhSL$z=UscSo`l(IDbJ1yUN@F%#KVj0GnRxs8SL6+<kcE8EeeV
zZu>-g(bMtP!L2I>1;#-WmvhSJGTU__7c|cU73r@O+v&gim0v}IJn2X!&8oJ{S=2N{
z3_mtDQn;p4Yen{(!{990A<HEAjx)tAOkwq$I^^#lpNAzx#Lxg2dmw|bOmEHX%(giV
zeMOh6uX8H?C7%_Lreh_c!QD|ffcD!7dA&|)eGQi6-c^zylyU)oGNd}t$U3m^E}pQe
z;3vf;Ky$y-#=>0^Rc9zEQKEd`g@^Xkszj3PXB=o^WU3n#$T{+=OW1o#H@4&`(<`<i
zaW}y)`dnXzNJ~0)+2Y;+5;i$?@9yyZT$DQHbee#)Uh<bvk~_0M;HUY^*9KkpYD|p!
zX}KA>nAD7L9*fdwwHqh*8ao229e#>}6-Q1^k~{Hr%58&;nIF{?YO$F8;h^cZi3+Y@
zrh4Sb7ZZh27{T*KLRTiM7Ms(`fUZRnFD=5eBpOHKkG1t<Ty#SsjLXUy#9>gejYR-g
z#eY^~Nrv2!g6FBvq!pZ$823CQZmOYm%xfQWdx5cmooKfwCGG+gvWAfHC2@n&n!P)7
z|D}XFo>yE^^N!X{=*RD4F;b%Q_1@k$X(tZkycfe$<-AcZ68n_mY!@?3dX!<aFuh|H
z4ysw@#_2Pt5iUZ)%A9N+J}4LD^73FE<5N!T;}~fy<&nRU(Qm7dzO>M=jk90-QXn!v
z=i4WLK7!dK99!y3r+i2>6_d+m)m-oB%XBm`IFPdxinrS4ko<zH@W3v{GvL~IeS#an
z0>mF3_|RLyC*tp}REi+tB{_57?r2)iU(v8#j{CB3KlsbG0RKnt%Y3L<6;JAn5tX9Y
zGv@R?dyLvMz(&IKFz-470jPf+^_Gi}Yri4`wO}Ra&TEkr^p$tgbGuXHKp!ril$7|A
zIuUfIbN&#IL~%!!554cFlWdR`rAu^{Ze8Za=%F%C=UzZCk0CC}n00!cvVFTKm^Uf~
zj}e5gJ(aPlVy*~i=7byFP2=O4I3tb5N=>IkU`u8mUB<mx0p&_A&QTYT$Cq-=jZ#ou
zUj%Vq?-oho-U$)@!c*e%B4yNUS2LdH3wNB$$MaOYU|?QUkG?%vOKk1w)Z%-?L&{X`
z?2Nh_0SEPu0&$~t!??avyI&e+65P9YYF?gOc>B)wt(l1P6qKB$cZm<R6){d=zdrpn
zdVGXS$C-=-S<5aX4Z44WaO*hDSTNzrRnPoHR7>yr@b%9rZRwvhQ|6CAEI>1hGxH8E
z-nLnjrac5AXllU_v`&W$n+u+DkK(yH7opl6@7P9m>E2eE<-v(0<nm)Pg{hFJN{#OU
z$&Wdv>3&W9L=Z?5PV7o56l=Wufx-}u^x<CFlgzMGrhB7v<`Gct;BC&^GPar&q`L7u
zwXj*WmclwYFL@}G+gEqCyE;&-F1y>?`?OKONR`p>k*p!wW;9KZgUcQ1BAgdIEQg7$
zFI#(BQ7R>U81#O8-^3n^Zb8XUxUpBeb6K<~HiesfVz??-WFI2GehQ45K_e_odJS9y
z^>?MaG-}T(PT~5kMQyu`vdmAu3Rk=!Olqq`Tc_F&zmi=Z@M@Z<z-rOF)NeUkWZkUL
za6BWqfzt8<vcJxaCtBZZNZ+{3e$`Rc(@8I1t{Xs^VmDtcEMEDBykfQbvxL@?Jvtk&
zlT#b}nMap@VCXj(R!X%bB=Uh<=@rR!;fU|3r|(<%+1$;#m>juvYkPa&29)@qr#%LF
z!&I^Gur3_6q9N?UCBKHnfm3J)zSOYRc-6Idkfr&>%Z%k-?c=>6U&SjdkYV!_H)NR9
zS8-eWHM8Cp3+wB=4~EvPhn2wt%ehF~75jRL@&UHG_<%5Xm%%SDZHSt~ayH^X)1z1N
zwgDciZyY4-E0H$%{e-J)FO{g11D{p5(SMz3kE+vfQQ@RWLQCFpV?&>GzLpBoIziyI
zxK+-ZT}d3Au0wx)HeSo{V9>lL)Z?%~-8xz~FMPAE<{hsaJ@`U_WnHIlfhHM()_$<s
zS+jk#xTxWHI$F|A%PGUhI>RglPE*98o>(M>NehVRI0iG1X23GlOLj)*9x*){o<}nT
z{FX&pq*i4}&R+F3J@4xpo1L%PTh3S{O$Cvl35S`!iaoYJNmMeOgbeuZJU`xYb7gsx
z^Lef2=iwqMR6J3p)BiMbIs5(<f6>Fql2Vgc);3rHHmK_b^edm+!iP^wd0?3s6V1$r
z(U1AEUmEdfLn1_%!+Q6El;oi_{Pp9KB1=T(V?FPyvb`T0&ZvnG&_A7s!s{qEE`C++
zThNtD8-lG_jGCNNa>MlDB$Ma$wp2#o(+ZCc$J7^_JQz@6MR|5yy!_-*-E{IDBmL2g
z({R!i&fC5MKaa&N6U2uGfp>hmzAmqm*Z@%w>nlZ5aOCm_XHq}UZ0~#A4{RSLC>lsl
zUpuIL1~i78&jkdpOXq^YXjRK}_&<cr?=~J{#}_G;Is8iMWXXI);1HVn0h@`-?abi*
z3X-7P-)v>xHwjy1AJ1}EK-q#OcuRB}N}~i=I_rbVO)1!tJ)e`LB{vd&TcVi@_VQV|
z1DFn9aB78+aU<JfNS9)sHh2xcOj6^>Ar4|gZ%-7OEzgT;0zHHNTDmTp+E=NK9JKT0
z3R|+EM2E6W@Snp$GiMgp#S3_c_?sg=Qs<d6EWmVjF^yg=_MdQc5?moYYf?)kdRbh`
z9r3{R;H?&|TLSw`2(>mAZ!Fw4F~i@iYu3bY+u&|ZTW!7T=az#xQZ%7Y+BjZ^WE-d!
z#_}V&o!7lMm00c<rx9Muv)^l5((YxwHByd^ouRbBND)I2=k?goMJ)hub&;=RevKB}
zn-d(Sm=MXUr*P+ddPdgg`z|%2e-01x?qVFMxyo$8!Gf%1AN%ur3k6DSLo<C_FFLV-
zJGeJq9|MqQ5poB;CA@LAb&bs}F<p+<=)ILaZAaBwF2<mrCR%4#5jRb|M^*VJXh;B3
zy|)1CMKDFca}5tB#pJ}6rHJM#B*;}*+~KTox!K;Ia8%9Dx>SW@wx3?fw_ieeNP?<-
zCEMamNfM*r%;Bf6iSE5b4=#Su5Oz%{Ca@QqofWNcDhq8Vif&RlbtgU#HE*hU=`KON
znq0&ib8>0M+DS_m=R~48KBFNL-SmW;P7@m!LC{7jeziz2b#u2|0=74V*cnx=dHML1
zl-mDsFr|~ciVLO23X$};D(Rs#H3#o=r`&1z1OWDm;gHZL_hHS!Lcwx*%=707xyB~p
zOllIz?l^&7GLx@*W^BLa5;e6Q2E9qFtR;9iGmIJAw&^nmJ_7Fj8a_3tfC))aCpB-E
zp>G*VjU0~N+m)1pAqSWl7k(-xG0*wDY1QOU2o{>~RS!F8?m?ucS#@05V~};-kR4Lg
zSpAy0L+TWilsZ@8PU4P7Z6LoI`>^rgHz>L;V4!p_P~rL4XTIMFN7-c<1&Oa^I6zwv
z$95{%tGhpFdElrnwOX^Enl~zNR~+1m!W!n4mUP4M2l|sPLfDidf&g++7;q8U38d}_
z6gu=}M%U#Et}j0raTr|E@bVO8$i15DTkGSKE40p~iv!gxzZRNxq5eU#B4Oy{yoUPD
zVD+=}WiJk{+Y@P;ACtkX<Ifq#neoaulLj#>{BlY0UrMivAqbNRjLyocgj9}|cdB-5
zT7CDPJ3rKZur9+-ln|sOR7!uaTM`qfu<S_inXA3NqTsO$ALZK|JI{btF)*51r#Wf+
z0R@sbrbR`l{tdEFlSxOs*_59W4Ju@(JXLosvRwaiub^nwLYC*<+G>W~0OkhMoHiy^
z+nbfxHYVh0bZHOMAZ%m0pXnV}N&*QNan?qW-IdZ9_wI3`^e^qYy0I%+!Mkow4nMvO
zT(6+fQ-Qvm1(m~2j~v8>%lR7XCU&nPBs>$_U*tJ#wg@hh(D=HTMC`m`r%iFzRf~^t
zVY;hA$;6znv3IGB9ToqMIhf^esK6Q`h(;!<)m0E8$2sIROt~G*(3muHvlTpi|4SF4
z6=b+_VkK09#!;^Ma6h^Z@NXf3hyZHbc*krDvAe5G?O&D14Z5p|2s&mgUa))&*U&Yo
zl|1aDe~Rsk7PAVdFH#P%!+e%ob?Uai(WmL_TUn;NFC`>7)*Y}$IzQ-2J4@<QOMX1w
zvd=S#XQQB8Gx6C<aEA@OZuBgovyCO9hgNWOaMOn)Zi+vrO|%y^jj?4W6LF=qybA=V
zS+pk0e+c*Edb3BXcEk#A9k+3VQa?<5(;CsYiaqw$iH29*&9GlgfS-C*!WDW_dGx7`
zYa~!<wTzccE_dca4ThJsA7831y=e+!!}uc9Ur{L(9}BrutZU@)?nY(do~$>u$1{CB
zjC@;8_(mE;9Fz!hI5X&6;2XJ8uq(4LN4C90Bh~Qy_0EcaH8o+Qq~85v^5u1-N0o1~
z247dPtfzPML$y8D2Z}PZ^Y$232P=WHY7!5d(k5q>6XglEi^IP12MDWSB!yR>B=6GE
zo~`QSb-fQCRixk=T&pW>an%$**@ja2j4W=_5%GFffi;mqe&X4jN^fk^B)y#RI<6ca
z`KooB(~TgP4stVU%@$0en#L0jqs_YZSk@l|-wHdE$z}Hb@kT#JD4nZmGCo&*+JDJx
zP<SPNGb2OfC3JAza_`=6(2T<qv2gO)8uwtp*fw;Q?sxsBa*HZljnMyCr+k6CdGC>{
z??n)M9%cP~mMN|ZGAk1$hek9eN^BK8F0D3u^rBI^gTjg^#gO{h<TC;TqvexB<~#yO
z{lP?0{X|Eu7QdZg3;}@I;|9G?ONF}fatV3|p=cnBZu(XVt5u`aRVHe7_702sLE^9;
z%&>K#nY|KAk0Nx)CNGX-nM+kx;vfRKA8Y@s1HoUnPrym}f9sR6_`Rza(U(#1xvW60
zt#a#5i=TquLRuCt@XGBVK1%QO&9rW0f8Nl2ePU-6rT}u@vN&96->3R`8@BUruHw`r
z-7@F6Ga*xCSKF!OCe*cOS+OfJ%`PV{TZUe|9{wS?I?C?{0R#Tw&HmYh1K@29z+L^<
ze~8oj#Md(0D@^7MYhSO{#!)OZYpzjGiTrmI7^o~exeutIT4bF(hnUB?2=6$VhJV`!
ztXcE_iR33Q{O7+L6E<-LP)x#JpzgE_-r++7?6P^cJj9MnYV<1s&>w>udg8A2S*+TH
zM?69U@Q44ER-7(QSxFBztOCo9yS=i5#*;#hNDsYiomLLJ*J}+4XuTlxQ1<_VS!nsC
zjz0b`BNG!Aj6XfSudIKMMpH$)xOVg3dwF2UZJ4w)k3K(+6(%qdwoBkxO6yD0^8bS@
zPN<_#Jc8z~H-Tc_@N?d;CY0YVw6TN!)4=}=YX05p{BK_NlO(`Tj)CE3`8H^jK-{6`
zX^`-n0Ghc6fuCbJKiIhQUpovv-WeP|(-O*43%NWE>CF;JtcR`G&zbBxe@OTZO5|kH
za%3#IVs45qwD{Hc8)Tk*rs74S+`gZD(@G(C`kq^M)y#N-feKeRVnOgmi1msUK3vp(
zeJQi5kaArSaYHs}hg!V3&O0pl`^UGRrT$;ZzvV#!(!{uVvN*}4o!8T`?zySkWA>~@
z9}lwb5bwnwi>T@!v>O^SFW3r=MqoGhCa$#nxcv2-)JGJlG*Iub?w?q(H%)~H$|8*7
z2jxC<*EqF3*DZWUlFy;ZhJpRxtzuw#b86k(9W=F1Mha!bS|;52`WwW?eB2=k#JW8k
zm1ITwMu|NfbCUf0R@BcX)fyTk!jB<jVLVdLOR{jCuGe7ykn^J-Cf^CBs}d?yaPx0#
zaCRh+4tbV|R|#phNn^t`qE$)b!`lM96U=;hO=>50J<IY#&CJ5BbUg0sB|jkwJA;G9
zl9vtQi=1k{%}V%lA2S0kZ~FbFvJdOKGk=CM$A00%yqn_c2L^rl)A<TRPxfG;Y{evt
zyRMW1#;CF1qyW7)#RV;^AI9CLr4AQ6VatE?@{~PkMU2=jH%^lcy)9ySUaz~ccR-~<
zgpA`JldCr9gf(tBQ#i^Xw)Z||If)z_THU|y)7Zta%dIBey2^0ds_Zqw&pXTH!4=(T
znp1(i6~3-L*JZFZ<}Uuvm6eIdw#?cX5ISOqwuj9DW-Q~wT=r&dmif>V<}jnm#0T&B
zy4XL_9)mLeu?Hw-4zUVwfw&>>i)-sUt=buqY!@6jN*|ovu9aw4UB9nd^91vbtL*MS
z?5b48O3_|Up8eO^w#?h4_k1?r;h3iP=r-1WXl_z7@r4_`X?RE9&k`zpVu7(e!yU1(
zI4ciUN$fD~9s0QS5=W*;t>jATP`JdJXstf{T(5HTMfzJv#5g-W9JEpCR8spfZDyS%
z&i-|I0U<eK-My|~8q1keSJQ1@8`06TB!1n^IqsFA0u6xD!}^8Bjjluv^>-_L__qeA
zR0YEQR%2fWVAGFa)0P)HP(9o>a(GhZBz|H*qf{!@qxZMT@z;q$WTw<5-j-fu^uuSe
zO#}JsB8o>~{=$ywuv}a#-0RwTcsJAAfxEx6AJR`+bIPR%c#QP5)@Ez9WDYIM4HA&M
zR#2i#^<g(2??<;WPk2syJup!Y9u#E}N&5krKNG}&f-Mg!Ydk8|-gZ5IZ!UjR^F*Jc
z#ZIEOkuFgMc_v|ca9!bJ9QoSB@BeG>JHwjlwsnJmC{mPeXbK1_Riy?9APOP^(n~0c
zASFgbS||YorFR4br6@=>AiaYSs&taj353vVKuQF}yT0$c_wN0i=bU}pKKsXg&hItm
z9BYm_*I09ucMNXl9=0L@hruKvYE=$*zz%4-f)|^suPlw~ytx^r+~(M?{5DSKkrhnO
z-t9_duMXSL4O307>zYqgFZ^>f_CKEG-|F-0{i^?)>4(&RuRDzY)Ez6$ObC&t$TpLz
zp`VG`c^h|V7!pvc@)c_(G!$yf@@<--cGtq|n()Vqp&plZ=1;l)290Df)|W7#-m4B5
zOUF`;tro~c^tieTeZF6%KzeaFViQ^sDNXT)7qk<(D(ioiWQZ6P-!BdrS9*dQr0ofy
zry!4ws<Ebe(OC&?eahjsJ#01o1V-DQ9F;JS9$xkaa0b6t@fkT$wn8h-hjL_O7M&&B
z|7^ube^e7|1nlZemv2>4!<<_*rqXela1L8h!b(Z_j*#*2C@bUSc+tg`+<%YDT5Cg<
z?O5m2N&q=#`S9%oi=#z!`&iWYDNJvyQmM(jzl>zN3f)SYAID)xuw$GiYou>}cK5`*
zy&m8!e%}UV9+uV%QS8e>6tx1Kt@;-v`)9rxN|(^C7$Sv5xg#I{YR<RNLb+V-pEnkC
zCQUi)Wzh#PN!u_JmM!p|VF>nw(|3NvsH@dSW0rND(lmoUVLfKfJ55xUS)!{lepr0>
zkvieze*?Pf^f(Kt*J=XpD9L5C_zf~<7*rEvkCEo>hh!u<w2;)cy`s@X8^%dkYt>86
zN>kee`3`0gfxt`K8X3}S=_*IQEpf}9;;uis9VC{LROeBoy<$ldh_7=B-F<VD9`Bty
zGf2rskWbzlC2@rJESh6yvenb4V)Pbdx=;Ge14phOqlYF2NIlF_hyUf8<ZCH+hXh!a
zo9N3+c_fucmj<b&75v0WzANt1<SB^JPBu5}y*F3fH@njV7FN(v%|R$nw8ar--zRAn
zXFc7!5K0ZSy-OLf=FZf>9y>hF2}5KzoVRh>Y>$i8H<dW%&32Tt$P3&27_)Ro>a6tg
zyW_Hk;5Wxtslyv<2Mx8cxd4_`#^~j!|8oY-!K=k1KWuPH>f})UI-EVznlD(O@Qu+!
z2nh0!7S4Ect6hbDeYX+xrMk-N`GIF4(L&X&vQ|(qZ$X#xDpxO$LXCN`$`p&D)?$r2
ziim+)9sr(#N*|>YP~Aua74c|>d3nyan<ixSj)wt8$-$M4O0$E&VZ(&f#{333Vw153
zZ@VOP)t+|T_`o9SU`RGeVh*xs<@VsaR8|qQPi*;_YWAUI_9|7=1#)v}Wk9I&#OoUv
zl(*-|{=w_nY|?e+QbGP`+X;%Zq9h&i-Kxob8D1!or1@;&rgvv)79-+*029(djf@>2
z!bBHwVjNBjp7t-N?4o$Q3Vr%RiQ<yR!+TDo`jb|?c$#sDa8Jic>TVzC7Um#>KeNRr
zH`47(?z;oiSEO6Fcl_7g`zMx;^@*@phMs4A9*hg0=l$&~dqo!FHEr_rRk@qM)EMX)
zoYMUn@r4yx&pUgTzZThB6SRbO0BmC>)9#UuX{V2U^j_T*?EWU7@bZ|3;NGbUrAIts
z$?+6k>DQdJg7s$hhQen-r>1~LW1ID}jl#^dJq}LKn<wjxE96XHN_at<yqGZ&-?0_H
z=rDX2Dn>n`<(;e366MCx(-9pJiXS`xPwjOeq14YccQTd<W}&{IwFJv9)q-85ON`f^
zo0eKBdFqi#tM+MSkxW-pk>Z|^)o<k2S*-$mBt(>6>%LD(*yD&p9TmOC3Mq@m)>~f(
zEZ=-VKlwJwVZxD?oD{%R{=`c%Az|{PfzZi{VOJKjDp|;rz&5(>Cv(wUMls2$@+{qP
zd24#<(qusiA1+}23XuJ1l?k{DUm4ZY>L+xbYE-u(KTV_yny;v+)-QI7({%HkT}XxH
z0|!*{!El8NBl3Qa)rdQ`kgyunIy_i<rjY5TYqxMJim`D&`;(?hklo;M&oU_~>ZO?5
zwZ}Wh)7u%lRQ03*CcQl2kL<Cp#L(;PLF)Nmn&tZyI<~AK#-fvq%&jaI%dSWl0CeE&
zm~7tAH-Ak~(byEI@AezS=mdUVxy_>J+kCbuo!Y$qR|wHcUB2T?Kht$_)4Me~aE&Uw
z605_f7UfxkA?^BpUfgkK1jr_=i3#Y~kZ4Tay)2H%J5t^#bDOVzdI)aj52=u$lubBc
zFT>ff$6MteHpI&n)>JPZ${h=w$kt@4L-|6rny+pkT;YP}&DG>D9nm)u0~o|IXte89
z&Q8GV!Wt8<(e#_rGut!(sa4F5`@_uaR@O)j0(^~?AMmF-T*z3JP0&ND%{2;Qk}psK
z9^%|nhK1^$>~BD)V~AYK-&o@S`@ZwRhFihZN)s>V)9;&)E?2K^Dn6%1d-z`o-~A1u
zTW+G4kl1@ub4#1e!7ILAJ{H=ro5hs(tq4v;dCs(wZ}#MCv-ahGZ*IMPTpGNQ0q<~8
zA?*hoF3<D_N>Z#RxM$d%JgaYnKTg|}j%U^|zzEbghk*f<^Jhxvw-FPQvzN>&*-|8<
z$gcIx!+|2(Tfy#OewX2>?D5Zy(>WJ(+*ws1@<*M^cV#<U>U1(&Cy(ma<md5yc@s&K
ziXTnnZ5C87jwYY*k!7QEKktX~l@!{jN(ddP$HiY;2Jx<>HGpZ0lRi@hbcZhcY8UVk
zP_o3ajxvxp4G$?;70h?Fj=Y~pY!o<TKI4>YWpMgwkILqNu0p<q(#1SyuM?)S>Ng0J
z-yj|H&1pyz8;6Q@1r<H;_e-&|Vnga)wi`x$u=<H!a*&@{u^{k~p9X(k;}b3)6(3oh
zXgBH<<xgdAj$FM3*pg49XCqN69pH@j1+dQxg)75T7WZd7?{<DCvG-0?#XOU)#3G76
zkG2n|m@JB1{Hb~jL|1e64<AzAd6<mE!8I<Vi@JVGc`EDuu0y$~L&og}ezUI^`k>?K
z4F7?yHyaf+w9sP3R8NKwk0{8NpfRNlD%hJH%|ebSv+8B)4p(f^udGNEIPnBMvT=W~
z#N!ro^;B-PD^Z1ZCN~ra@8KX{2Cx4rT%6OODdl<7GWT%=QK52fHQ97zLJS;}`DOCU
zU<<PE-A!*$@Rlm8pJt-^P;q~B9>um*IAcPAV?lfCHsQk@dKxKu--fy}%CiSZMW<~{
zo9QAcMkIbG0bYcK;>1UQS6H&rq*C3N?^CJROdBV4>`3V@Q1fMQk1D^@eIlNEUFYIY
zs;#;}#mo>sAzTR*o2?}vo>Y~2?cUI-c~TMIVu9b)vJf3pbDVwZBe<ySHBpA2_AzqI
zFQ~HUP5n8CJkZ<*_+3WZoy|qeLCOt+279`m2~B`S9#Q<3>F8c0Y(>H@rblPYu0Dwz
zr_sp9pS7aO)_^FQT+5JX#);EyUP+u+A%LE!M8g5ATIi+A5xwL|uqs;<k6BATa!EI2
zfhImg!f5x3n&3}kW;VFoL}ys>X7Qq@M{3e(q$(4LFa0WY>4)DSmt&<tNg1&|SC@iw
z`a+??_quslaeKu@=H%x|r`814T4ncr>>Eqr@FUNsNi)>^S4En9=|MkAn4L`IY?zNV
z1C-0OJe+NJ?3E40Zbht<$O9=vUivL9vP|;V16cv>f`zvQCf|TRqsL;P!nBI^;ISg{
z)trvkgCmdj^(|g>_)9@mW}(Bma+~%Ex^$i)il{Eb`J%RAXscTb&yqA%<u(|>d^gp-
zi2vN!?NM3RUJ0WnN{^3Q48E7bgwet^m^;D0)S-)(sNOqzb$-v;@6KW@VY^EMPAX3E
zm>sJyZyKpdkhLw>{P091m|{Rl&=F}RDB+$F$<yh_41#7^Z>t35H-cxRseIdigT^5K
z@AFnVF;yHseUpOyUpZ79Zfj6=ufiQVticc?IXZ1Jd(?yvj81OR;Nn?pT{+LghjgiM
zapHdYSFn<o_6q>V``{M;+vX!EK6@5$9%Fodz%2@HIDrEOZpELJ)?eBa4uq71fm28Q
z7>PZLYTyA62^{YQAF7|mGu%!Y7ZDD6gL~r6JEJ{8@>xL>muw>iiyUgDsTP~A4CawX
z8r?oy%ExCt4z-TSvE%O=wH;iBUMF{kx@S_f5bKiS0(OqMZ@MHpbAd{J3b6FkB=v2f
zG>IZV>0{oq{*S4i8KKk(j2;u{4`8p>TC_Abe&ktBgSn)H>Gel~TvTVeXksd;l4jA0
z8V$H-k}{IO<@ViYo9p=a79sw(F3VJ3n)w=cn`H4<aEQBTOkX0G0_YLq(gBHQkj{|2
zr5bEV4k_~H4@Fm^HMrm2vrUhP&SCfwKWK+fche#Fhm+E84j*X55V(6@q{R8PUy84x
zJ-ZeG%JU^gt9(m?e)L6~tE1dp-R4&Yzw0tQNfhA&f>yr7Q8G=F5R`HFczN^nDe+;1
z;ApYqsm{;}MTB7s|0LKYXx&3`yu^|Ve7gwj2}<Un&m(Ryb!@KL_{K6?bfYbjcBBn&
zoZ+N|EhRuJsM?Sja|D!8n7dlGv@-`x|HJ9c7vAi#e`go}2hOm6Y+C)Jr{>tr_N2ko
zxSAF&I<~V#Mc`&~Jn+>*0skmUAJ9KM0Y3oLE8g@*Y#9o3!Xgi#bjO>w3IQN$><Q{U
z0M37>oVW=<DF2-?JT2>)FMz0LjLOe_m!nkxTS)@Iv3~%9^B0-zjzW)x0JCeL94G()
zU|ji+0l@p7rj)P9)Zga*8$X5MaVP+(`X8_J_Y8aax(@)@<NuudZyaMD?d#S2ZSKGE
zbEEtJY%$-#898-<QDXM84h8*g5g)bE+RcIbp!yROcpKP19D&C83T6|=em>l|{c;?d
zE*c?}k&_16MLc;(b_ti>xMW2M+3fJ;zVeEs%@M~!r!;TG0ac7uCEv|R&t7HyyjPXA
z^OfG%E>eW5p6Bo(Z4Wa$#kzBj&04T3`QD|P%7BVHU#mBB1*`LgYa>;-*JOb-C)}oo
zXb)w>K_cE#w_nO9kn4l1DmPY#Ojt96OcGYF)ix|)-!!TDEph+41tf}m#-3bmJ!}q_
zfu-7)Wt{Y4>;N-VP*Fou4-GQF`|0N*Bkv4v?p?F;*nk=Xm;COd2dY751#1pYDjPZ5
zbk3cyTIs^S9)QK?UF8`scBM(tr#I_Ifchdke5}z~HWO=dESkzExz$WGcLT%tH2J2C
z*GeKEcFuL>l@IN9OCP=b+U|BG@9^GhhV$GCyn1}QR2nWUl$Vpb+xF>EJ8bxi)_A`J
z#rfbBLyF5BM<)O4;)>}PwM`-TjrFB7W@S=GQ7GUx@FuyDnC{z}>?0CxUt3y(eJSp8
zDYBQAnVeLhBv**k#PCUm%whE#(z+(Jct%-N)d*7zSAoJ$DPE6fYc@ABR4&+9-~JI>
zAvqRh&ss~9I6yQ5m#wT6T1q4dm6_8)wI4{eqn4>_zK-~bqh=eqwQ+m(eUK&H)OnZk
z$y=-mt|7dTy@Ewb^k++C@vmZW;gq}0<*;kvzd;r)K3aS8gEZi$MYFMVebAz8>lx>?
zkPevHVsMXH86=UAnhW>8h-;v!<Et;68lOVdZkm6py7_!yMfGBn$hL681Y>nbg96)H
zp4zoE^CjX^fTbtfG`!8TdWLydheI^nVzFQImLPS-DP4jO7}SjuARjJ6d{+)Q^$KPE
z$9W=;Dt9M4DI@5-FlmZE*m~JTFmk+j#K<a?R<jZGaB$sE!S&fN9K6u?x`iwdG1zqx
z^%7_x?|vJh*g+5(I#zVxAv=0rs0$T3OCM-HwbvrgMuj0ndNid!JK%N|JQMjLp_U}F
zVkO%=$9?WM=oBSDt177vADX&UU;yeS(KUjcV3b(QD8%damp$%q8XwD&D`PwF4ro>P
zy4mv(+b^#*m`ab<LDSMu7-<7bP54QI?efj>ZW51e;0+Iz^UK|{TGGn0izewZ>aJ&R
zB>kF^S6GK@0Cl1dpO=z%Ry%fYt|gu!d71SKpF!H9TvZwCaSf~`j!pcr#`vvADcWUb
zSAw_=yfSWuEJKVL)!VFjw<Ve#5cz0)Xz03l??W;9o4%|qtB%y74@$|=_%F>9kiLbe
zHAQb<M8n|r{z??Mlq7k<=`blbCEIcCN#2S+o7Ph<Co__m=JU6p84h?A=OHxo^UCv%
zbLU(LA-9_wReVb<LVknPOTUt#?+d%I0gleBP~4R4nYSeQ4!2Anki#n(Umdy*AO8aX
z_2iNDg>j?OiyG9gHt#x)A&WK$QZb}9Nc*pjct)&lytd%?1{Nb)AyOsCor9m_ab{aJ
zbAc2183eVsoLduu<ve{;s{Xygb9^RWJyG1|HXr|pLv2+|tEnvKU>j^+BwNw~6pJ}b
zdF>}j;y%|Set&0i=yLpXk%UvtH@LWs+{_Q7Q-V(Lylqgyrh0P;wKU06uJ0Xsy=p=`
zKv0aq#Q)rT!{+ck*bz)hb8Jz{`*qCG+b4%6>06QRTg()n$vfmE+)kYpsu;fV>0Amm
z=sFK6YfY8&SY#V+5!rUWdrsL_OPY*cq^)|87z15hLfft0%eXWsW5WUGG}Li_ZbIlS
z#xX#SzcIbIp~}_7QZu}vq+{>#MN#|%^CA={W)`FR?V#O@Lxtlbuyn?P<-=zd9zdP<
z3~rm<kf?daz7LtXSE->hZBS0!?j7<;rK^_<-+RTX^!ye+*khr<o)I*h@8(c+NN}F|
z7iUOmDb8Tkx2FOzss8e`mXgmyT7FJFBI6UI=25W*EwBCB_q{P=V_&VlF)+3r`54zW
zXcJ>EH0GjFh0c)EwJzr)wAKf|#!*ED<(zlzv;Ym-gn_A%`vdyE_0?rn*fe~xeN=f@
z8)^<RDiz%714ZYpa`0WMYr0`2+Ld6=wO{OgX?|%%mD9A-q^kM646hos<#Jl$+qyT{
zUv>^n%&tH~MxreFCne`_X5uLrZq7xMp(VyNg(6f?-5!9ELtk?&9s&K#@g7d`*%4YF
zcd|ph41;l4>1i8`Pi)}F!M$|b#nwf|&KidZTu;3U5Ok<h370N8w_+_&KI{>hm$vfc
zi2Zxpkm7~Wk3Rz-&r6Gu@ODj2oQ!u~b+ds0?)Y78Je?*T*8l~;mSr+J6D|xTTEs9~
zY<9pxch<L{-|dnFwy<0x<S)K?gS&E8Fbv^pT^(K^?&jHNo+w?jc)(f&E&@&B@Eh>L
z6QL2I1WXU21(B#&y7LQvx^&Mm(VW~j6Db2v&BaMbDn&j++_k4pQ01Am4U|$+@ICz=
zND4e$h`_kG@g$tTc{_w=+e#!d_O%SGQQf3;n|&(9l*vxpq}1LrW3v3$RqE$@VHQ%`
z(D$+{!dRxXWUsEF48k<oIftN(-9q@8ly?j8-ewQ!ka`_h^o?)oA+5K}06nt&?$^jR
zQ&XSk*^ke^)?<J=e-F;pL^~j>+bncCVjmL+m~&>vlO={z^SHcolwrJ~B0sWdwoSvN
zfsAGg+UT4Z>{zS8i-&<{m1BLdHQUqa>056kpHb^h7za&GYjl~4fTU>lw)#NI!W0BY
znM`{l4hH|(lZVQr_IHau0jj0!Va1QV;NJWZ#WKQ<#WIdTlO2U-4DJ^<+Cdg%mo{GL
z00z5Kx9uiV!E*O8<fAQn2ZO_;d@)?7d2m-7A!<^VylCY-v8<(ZQ*yJ4Sv{sX8st9k
z1$p7v$j|)O$NaNBzp{^1)_;R`uvmZBs1d$uCx?9G)1~%mYiAp0pG0)Seas`Z9+n%4
z1H6vZ@x>TG&zIpnl)*o=lg7V1n~RRTYzYCh6c#Tdc?wyzDsE%c*o>7-RjmuTs?;9$
zDym>3t##Wg9PKe}&@FZej*`=fepIEYfXLd>t8EAJLxS6&{G)yWxvMM#m$8>4uB1n4
zD(H80jz)|I&#mjs9n@rMVoi19qx4=>Wv&SG1&e3+zopU`Fe3Y9aN!%2U5lnHV0{JL
zkxm9fIl@W|r+}stjO{5wv!%E@kjA~~;ZN^{k5rP2r!mx-OR6kLY8&jaN&#UY|6W(3
zc{dCD?QvWA6@8>!Tk-@`k$d$}LmKK>+~o!97OgUuZv+Sj=?w9cA#NB9U=FGw(S-aA
zZ7u23S0bP!-ag0G!<1kd1i#)eaa)Fv{sc}RQW>fDlv!k3ktgcIARXNo*h+GBhqCaA
z_8KB9-0pRo+Z3X^I8EiJU}C_Se!wmmv?z_RAgSXwDn$*$5mQnF>Z^cpl}Mp(lgl$o
zG-w(Y{;-*WNBH5VXW-Nwj;D$%=lUQ{715&iLSQIP?Nqzdx^_v`_UX;7b+t-QW{)IU
zXDAtyW;ltw+OYhu91G{uYkr2y3j``mqP;La4#6J0a$2=D_4W94zDMMrPN$jf@ipL!
zerdLxTqbY6f}={N$1I(aZ7;GY3dgZ_cz5k)(I5IZ65j(J!Rip1pPLEiw9>S>DYiWz
z-<zfibLU)`U}sp>n(LaL@k6wTJDpXuAREf75;J5ltf~$>?+~GQ^`mW<>Ec7Tkuqtv
zp?Mr$rn~r3`E|D|H9ex9@|%DM0e@`xQJvVQqF#n8kqHpk6FD-*-F<NN%ugep55{9k
zTC;1JVO|6woaB|UkN$1B<L_K1qvw4+*{EKjCEqO8q?m-Z%cZ?J>J_p}_}0On;g9I#
zF{zGnY2chgY3LaHX?oKv(%!A+U{JeWb2QEi9G%X!8%TdeMe6J+EA1CU0l8;W=u1up
zc(3<IlS)`9{+;U(lUxM)fMe`(bS-o}XOK>7PBPrfr#eVXdI6WAeejfX_FEPNu*~H<
z7a}AoOX!fvI~drAq>I?r>>b#9B>LdX0_6O3xhi{Qb@{e+dI|FTuFeJ4PQ+vSnr|dZ
z%~BNK_2kSav$vB|<kIkmkbHQgzB9~?DNE&%Bh_I>^+yV(J*D~(cL4Dy?_e4K6x(~P
z`c;irL9s_4wHi5kKQznA(wWuyS+iqp8viwY&aakx2U+y{9R4&mWmZ%0l<hiNC-mLi
zIWrHc!zVJSBxV3?(yBkF9j#+~<4%i#kg_q`BL_mDoYJ)~j^}34rm7<z*Ve&1a|$f?
z(7iS1f(KF8`FW7B+PHP;Mv?nzi<`U(XXL!pS?=jQ_x_f38u=U4mTWD65?nXzzFJXg
zmP5#Sd_+<Sl^*j`%ge)Ds}PPqnK}r1_?}7jm(Rn~$w@_&rw7D!C_lc^HyVm8PD2b?
zq`qX<oh#p6b2;7O?Y4-FHFz}_Z!x^jZF`QvEva8^z;BQy;Jlya<2hxE32kbz0>tyL
zmAyUTB7&5(TA+X3zYX60hlxS|`48~P1O=E|vF?qL>?0ojsb4xEe!I}4%!2IoxZfbR
zJ)jRkH}iptq2~DIlciNh(2<Y|K=T(#7WIFpK_&2S`sW5Rm#P06zzyx+3#dIz1O2PU
zVF3geVu(Yd<7Tm1JLuLw^Odj@>>kE1d-J8TYA^f01r7W!*C##E$_$9-5_9=ql5qS>
z5`BM2P<Lna`%aUKq{4Uu30jkN`d8W?LW}*A5~zjtCn@w#N>#e?kADdJNcs<1gZQ8R
zwI73j?gvl+?EhulhbJac8rp1zai9CCfd0Dve=^kkiy!DOhJWz`Huo=vfAI_a_m=bz
z$C1DlaPH?<^R9pHryKmw<Rs(gEJv-5U}HaGeIR@-#>;034SM)bUCRHd2<W`w3*I(z
lDe04P9~?&6X8y%{{I5P!|Hbd0`;GpWJpZrd8Ss1jzX8)}$*2GT

literal 0
HcmV?d00001

diff --git a/demo/demo-esm.html b/demo/demo-esm.html
index ffc7b87b..5cfea595 100644
--- a/demo/demo-esm.html
+++ b/demo/demo-esm.html
@@ -1,14 +1,15 @@
-<head>
-  <script src="../assets/quicksettings.js"></script>
-  <!-- <script src="../assets/tf.min.js"></script> -->
-  <!-- <script src="../assets/tf-backend-wasm.min.js"></script> -->
-  <script src="./demo-esm.js" type="module"></script>
-</head>
-<body style="margin: 0; background: black; color: white; font-family: 'Segoe UI'; font-variant: small-caps; overflow-x: hidden">
-  <div id="main">
+<html>
+  <head>
+    <script src="../assets/quicksettings.js"></script>
+    <!-- <script src="../assets/tf.min.js"></script> -->
+    <!-- <script src="../assets/tf-backend-wasm.min.js"></script> -->
+    <script src="./demo-esm.js" type="module"></script>
+  </head>
+  <body style="margin: 0; background: black; color: white; font-family: 'Segoe UI'; font-size: 16px; font-variant: small-caps; overflow-x: hidden">
     <video id="video" playsinline style="display: none"></video>
     <image id="image" src="" style="display: none"></video>
     <canvas id="canvas"></canvas>
+    <div id="samples" style="display: flex; flex-wrap: wrap"></div>
     <div id="log" style="position: fixed; bottom: 0">Human library</div>
-  </div>
-</body>
+  </body>
+</html>
diff --git a/demo/demo-esm.js b/demo/demo-esm.js
index 73b64bca..7a505050 100644
--- a/demo/demo-esm.js
+++ b/demo/demo-esm.js
@@ -4,9 +4,11 @@ import human from '../dist/human.esm.js';
 
 const ui = {
   baseColor: 'rgba(255, 200, 255, 0.3)',
-  baseLabel: 'rgba(255, 200, 255, 0.8)',
+  baseLabel: 'rgba(255, 200, 255, 0.9)',
   baseFont: 'small-caps 1.2rem "Segoe UI"',
   baseLineWidth: 16,
+  baseLineHeight: 2,
+  columns: 3,
   busy: false,
   facing: 'user',
 };
@@ -23,8 +25,8 @@ const config = {
     gender: { enabled: true },
     emotion: { enabled: true, minConfidence: 0.5, useGrayscale: true },
   },
-  body: { enabled: false, maxDetections: 10, scoreThreshold: 0.7, nmsRadius: 20 },
-  hand: { enabled: false, skipFrames: 10, minConfidence: 0.5, iouThreshold: 0.3, scoreThreshold: 0.7 },
+  body: { enabled: true, maxDetections: 10, scoreThreshold: 0.7, nmsRadius: 20 },
+  hand: { enabled: true, skipFrames: 10, minConfidence: 0.5, iouThreshold: 0.3, scoreThreshold: 0.7 },
 };
 let settings;
 let worker;
@@ -49,20 +51,23 @@ const log = (...msg) => {
 async function drawFace(result, canvas) {
   if (!result) return;
   const ctx = canvas.getContext('2d');
-  ctx.strokeStyle = ui.baseColor;
-  ctx.font = ui.baseFont;
   for (const face of result) {
+    ctx.font = ui.baseFont;
+    ctx.strokeStyle = ui.baseColor;
     ctx.fillStyle = ui.baseColor;
     ctx.lineWidth = ui.baseLineWidth;
     ctx.beginPath();
     if (settings.getValue('Draw Boxes')) {
       ctx.rect(face.box[0], face.box[1], face.box[2], face.box[3]);
     }
-    const labelAgeGender = `${face.gender || ''} ${face.age || ''}`;
-    const labelIris = face.iris ? `iris: ${face.iris}` : '';
-    const labelEmotion = face.emotion && face.emotion[0] ? `emotion: ${Math.trunc(100 * face.emotion[0].score)}% ${face.emotion[0].emotion}` : '';
+    // silly hack since fillText does not suport new line
+    const labels = [];
+    if (face.agConfidence) labels.push(`${Math.trunc(100 * face.agConfidence)}% ${face.gender || ''}`);
+    if (face.age) labels.push(`Age:${face.age || ''}`);
+    if (face.iris) labels.push(`iris: ${face.iris}`);
+    if (face.emotion && face.emotion[0]) labels.push(`${Math.trunc(100 * face.emotion[0].score)}% ${face.emotion[0].emotion}`);
     ctx.fillStyle = ui.baseLabel;
-    ctx.fillText(`${Math.trunc(100 * face.confidence)}% face ${labelAgeGender} ${labelIris} ${labelEmotion}`, face.box[0] + 2, face.box[1] + 22);
+    for (const i in labels) ctx.fillText(labels[i], face.box[0] + 6, face.box[1] + 24 + ((i + 1) * ui.baseLineHeight));
     ctx.stroke();
     ctx.lineWidth = 1;
     if (face.mesh) {
@@ -102,11 +107,11 @@ async function drawFace(result, canvas) {
 async function drawBody(result, canvas) {
   if (!result) return;
   const ctx = canvas.getContext('2d');
-  ctx.fillStyle = ui.baseColor;
-  ctx.strokeStyle = ui.baseColor;
-  ctx.font = ui.baseFont;
-  ctx.lineWidth = ui.baseLineWidth;
   for (const pose of result) {
+    ctx.fillStyle = ui.baseColor;
+    ctx.strokeStyle = ui.baseColor;
+    ctx.font = ui.baseFont;
+    ctx.lineWidth = ui.baseLineWidth;
     if (settings.getValue('Draw Points')) {
       for (const point of pose.keypoints) {
         ctx.beginPath();
@@ -164,13 +169,13 @@ async function drawBody(result, canvas) {
 async function drawHand(result, canvas) {
   if (!result) return;
   const ctx = canvas.getContext('2d');
-  ctx.font = ui.baseFont;
-  ctx.lineWidth = ui.baseLineWidth;
-  window.result = result;
   for (const hand of result) {
+    ctx.font = ui.baseFont;
+    ctx.lineWidth = ui.baseLineWidth;
     if (settings.getValue('Draw Boxes')) {
       ctx.lineWidth = ui.baseLineWidth;
       ctx.beginPath();
+      ctx.strokeStyle = ui.baseColor;
       ctx.fillStyle = ui.baseColor;
       ctx.rect(hand.box[0], hand.box[1], hand.box[2], hand.box[3]);
       ctx.fillStyle = ui.baseLabel;
@@ -398,34 +403,74 @@ async function setupCamera() {
   });
 }
 
-// eslint-disable-next-line no-unused-vars
-async function setupImage() {
-  const image = document.getElementById('image');
-  image.width = window.innerWidth;
-  image.height = window.innerHeight;
+async function processImage(input) {
+  ui.baseColor = 'rgba(200, 255, 255, 0.5)';
+  ui.baseLabel = 'rgba(200, 255, 255, 0.8)';
+  ui.baseFont = 'small-caps 3.5rem "Segoe UI"';
+  ui.baseLineWidth = 16;
+  ui.baseLineHeight = 5;
+  ui.columns = 3;
+  const cfg = {
+    backend: 'webgl',
+    console: true,
+    face: {
+      enabled: true,
+      detector: { maxFaces: 10, skipFrames: 0, minConfidence: 0.1, iouThreshold: 0.3, scoreThreshold: 0.3 },
+      mesh: { enabled: true },
+      iris: { enabled: true },
+      age: { enabled: true, skipFrames: 0 },
+      gender: { enabled: true },
+      emotion: { enabled: true, minConfidence: 0.1, useGrayscale: true },
+    },
+    body: { enabled: true, maxDetections: 10, scoreThreshold: 0.7, nmsRadius: 20 },
+    hand: { enabled: true, skipFrames: 0, minConfidence: 0.5, iouThreshold: 0.3, scoreThreshold: 0.5 },
+  };
   return new Promise((resolve) => {
-    image.onload = () => resolve(image);
-    image.src = 'sample.jpg';
+    const image = document.getElementById('image');
+    image.onload = async () => {
+      log('Processing image:', image.src);
+      const canvas = document.getElementById('canvas');
+      image.width = image.naturalWidth;
+      image.height = image.naturalHeight;
+      canvas.width = image.naturalWidth;
+      canvas.height = image.naturalHeight;
+      const result = await human.detect(image, cfg);
+      await drawResults(image, result, canvas);
+      const thumb = document.createElement('canvas');
+      thumb.width = window.innerWidth / (ui.columns + 0.02);
+      thumb.height = canvas.height / (window.innerWidth / thumb.width);
+      const ctx = thumb.getContext('2d');
+      ctx.drawImage(canvas, 0, 0, canvas.width, canvas.height, 0, 0, thumb.width, thumb.height);
+      document.getElementById('samples').appendChild(thumb);
+      image.src = '';
+      resolve(true);
+    };
+    image.src = input;
   });
 }
 
+// eslint-disable-next-line no-unused-vars
+async function detectSampleImages() {
+  ui.baseFont = 'small-caps 3rem "Segoe UI"';
+  document.getElementById('canvas').style.display = 'none';
+  log('Running detection of sample images');
+  const samples = ['../assets/sample1.jpg', '../assets/sample2.jpg', '../assets/sample3.jpg', '../assets/sample4.jpg', '../assets/sample5.jpg', '../assets/sample6.jpg'];
+  for (const sample of samples) await processImage(sample);
+}
+
 async function main() {
   log('Human demo starting ...');
 
   // setup ui control panel
   await setupUI();
-  // setup webcam
-  await setupCamera();
-
-  // or setup image
-  // const input = await setupImage();
 
   const msg = `Human ready: version: ${human.version} TensorFlow/JS version: ${human.tf.version_core}`;
   document.getElementById('log').innerText += '\n' + msg;
   log(msg);
 
-  // run actual detection. if input is video, it will run in a loop else it will run only once
-  // runHumanDetect(video, canvas);
+  // use one of the two:
+  await setupCamera();
+  // await detectSampleImages();
 }
 
 window.onload = main;
diff --git a/demo/sample-image.jpg b/demo/sample-image.jpg
deleted file mode 100644
index 83bc55327ec153442d31d2b9895ab03a187c3ed2..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 102644
zcmcG$2UHVZ`z|_kqzZ;AO+ZLMq$5ST^d5QuQ97X)K~W%pphS8Hk=`K)(oqEIz4szY
z?}#+%ckuiDzwey?IqR-_?z%UuNru@od$MQ0<$0g?-MN~)`T^WjRa8*~u&}UzC*U7&
z^%JmDcy4b402&$q7XSc+04^3ifCuhjftwW8?SJnpVnG3%zsIq`Mt%T%2PnWz6Wp#}
zPr&W^n!lf}PwkyO?>k$2vIyN5xX&*L-f0Gq2k>$6@bPf*@$v8p2=EDsZW0mQxIqLV
zBPY2@1EHm*flyP^Gu~yUzs<%#P0ezTg^hy~#s#Be=HcUk^4*2Ppx2#X5fBg%5fa@Z
zBDw{or>2MgAHS}e0186vL0kX_>kfcTfrUeXb=3+mgWnzxd@k3I=ARc9HV*iG2?%cx
z5rZdG-2||)aB#43aq#ePalx~F!FB+b0*~?zzbyVOou>q>ZV-Wh=uAR3x$<vR@S$Jq
zf)?(9H;AZdXzA$ha&SUnTtdPk4@AYp<sT_1Dk-a|>gwqm7#bNPEUm0<Z0+ok9-hy<
zyq~}D33?qI^5$(QIwm$QJ|QtFIVCGQCpRy@ps?suMP*fWO>JF$b4zPmdq-zi_wdN*
z*!aZc)b!7V#iiwy)wT7Fz2Ex>heyYMPEN1Kg$0iDzh3_u*nb-r1voBjTwEMng6nZ%
zVS9rc4h1gW9e#XDSsj9>Znszk0tg{;(V6AnZm<c$e^FVu4-rwb3;n#icRjSfBl~|g
zu)zOUBm4Kj{xhy=fCL8%oID%~KpNOL5A-Ku`4jC>tR{uwNI#+pmr=%je;gt?X?fdW
zgyzWX6N|u~Qg=y(m@8n<v&3<=dyDs6;4pgelCp6B3V5t`1+)$OxA06|0fS?1R{%=>
z3V4^*eDqcB<5rywE%P5$XX#&r{(H}|UCu-i&9zs+>llwKz}XT!Eoaf)bVut7AlY@e
z0^}iAz#l^CKlRc-KWzK2S)^Y9SOix9jarmxi}}&BD(jAOcJRU<ELXt7=Ka;9#}ywc
zu7DdF%~!xo6nJbSrtE#)&5MM49p`=C%zv;O&&^brw?5mY?+eW`pSd%io9VZzOWR1u
zUIBs+!CT>jw<6!_vh!cD18)Vq0dJ*D{_s!FuN3k(2*vc5)E{|H>QkF_GAOp2V^epe
z;+LU|E<#xB&Nyx+)UPn=)eMS{FQ$9h((EFddR&PrZ?T~>CnJ<K6!E8l4`PsoZ#3Ug
z)O0lrM3t+@{=u@!oego=FG|{dSJBCWc@a%82af*esWu)KO3XBQsy58g(||4Dp!iVz
zN5W4<7$GzRcgulO?<{{^pF$8ls<hikUu;Hf$(My9;zB_)u+fG<RuUc+3q+rV2nys|
zph$FN@E|~Tim5jCkPIFV0I39u)ntej0LUo-%`jjf74tj97H4FudPaQ(czt@ANQtz@
zA<~A@*5+o&Jy7i({konPu;rd;d)%?99yZ?D`<VbaS<ZmYWF(Dunqi8ge>50c)Gb|?
z_uhmgPV)Yx#28HK=hFul$OoL`5jND^s<Q@86T<iIJw?)%`s{}d*WK$*8gl9(Qjg6p
zKg;@PC^1_pkm561$)AUoQ5=l}Ml@CeW?@jQH<&vR{yd+?t#q4byf^+Vbg2fpm&-k%
zB<PJHQ;m9gB=+&q@Fn9P^n^xb8GS2tHH@*<#liYch%gx;11r?8!8O+>EH$=V@Xxz?
z7gW_t`@t)~PUvPLz7BP!Yk#*!^)i`j%UdftTr0lyekgCqxuxX6k%0LIa5wAYjDMvS
zTFjLA6<+(omz=35pCk5Dttcsd(%v)tS|2YrtM77;=IY)|6$AtnRYV60IZ6hxz~`y`
z_l;oS3aIAr|4A_j3cYmS6_C~Usx-L`5K|#FnSRkto(ppnedimVFC5qVVEAp1SJ(}V
z@~A}twhU0R9wMmvgWWwiY!t5CH=IPxgMT}v5q2>qtGseZ!z!<DDJ&okbJYAYyzog}
z)_ydE$Eb2eR#sqg{hlUJJf{9LVX<7<o=_EvMQV-o_Z)kqzHrEUIHp-oIoJy=Dg&iJ
zM+K!LCj-$1_xXrm99U`*FchyU0+a)6G~f!UtVijn<ALMQd>jW<YGbdU0a(wqs6goo
z+0L2+eLhe}38L43oW;k^R;p>N%e1x8ZC|(FDZxBC&r88~_crad3p41k6u79-F`@P?
z3~02TuI8D^<xZhlcxpgZn}^c^-_66`EtslEv%{|ZI9^d|64VDgoo6ETn8=Tnz1I_W
zK6<A4I5);KL4KLaF>IXNeQ?@-QzgtkD+gcK4vX|IAQT4IP9kNE2_Y!F>>S{odL_>Q
z7jQ|LrId;^SN4tAy!5|)hzaz+`Oj|lmbdx-pLq=Ko5`KV)xAc9RL3)LV!}@cv5q7@
zo}9akN0!g|4=nms0{`s_fZ4MC3P^9iusdU%IKO<bDMRqzyC6XMjfdzB3O&z_&zbSj
zl0NL6=-a!mAccax!>`(+A>y(wTX>-Dh+}lWuW?2wyG-El{ij9Y3X-OeLX&s46mOJ#
zE))+?cC@b4U|DZ7H_?8VkUyx57HLhQa@RT(=1GX~(58zi{s>9VVGO1E1;cPQ$iEYZ
zY;iTHicygojY$M7luTh;LxOoCQaYb1n78SQTGv*>Z|XvUk_Poxc}1+Yw+hMU>GRF_
zCwrlJ{}xgmSI{<O(5qvkZvpMF(I#jH9*z(NnwE%^hY*wm1Qf`{M=^pTT?y#OkY@#m
zKt-<wAb>4rSe$4*^Q<a%c8F3(Rs#=03DM>!wc0TE9Fenyd0f^GceFWH3-Zn!VcrQ2
zhX^@Or5aRy_RO(+SKy0$<_rsBH^`k=m!mG2DyXH(K@QZ4;iLF@O$thGx9l^F+)eiI
z-G~|3MJK3?7?a28*MHlw18u|blRHB3EJCRxpkV)<etJMvaZ{dNQte3oEM@n?yXI2s
zaq$9q$c>T>Ki0;C4dHbC6g-@+;4ifl89Y6YjLLt%CsN~NW5YPY&L8oei!}_ND_K6v
zR>Lu#o|Mdpc^Md6t<$9QiFYo%mrN%x;2am_P+dEmpV&&B*wXcTK>_dGE!H2(L?0PK
zzSk}Jet%P@Qul9l2cPMy<ulXLZEn>o;N!qWuDQ4z3YtGi5Zq&c$wCF@nULN@-l(;x
zpqxpwe?<4xVp<eZ)KMiu7`rWDh+kAhx3ZX-Bj`RpXA$Psr}X@fu2u${IQ|@uUDst+
zW?Y2V74Sw_J#~ajg3tpUHfX0ou9H$7gzmQaWm&y6HY40Z6u4YIi0~aU+(>TWX-3Nt
z@yn2wVAOW#3wAi%5b38Ix&LkMYYVZUMOZSqZ2)anJnFAd@IX+{AcNN~K^sd=hHMku
z0o@4{t2rT1zCn%rI|N(YiQSbuCunC}%fo0(MFH>c#1ks68`CHMEX-71zE8fZA<i}B
zXk%WpjTpDhC9#{S6fSl~YEbxytn0GbVuVPZBfXHXIX1%I52Ls{9i*0ixyjJheApyA
z-qpRE9Lt0>;r+^G%<Yk1aTZu&Q)UxnpO29tw1?7D6o$*R=47Zlz_nW+G~nmt_(X2K
zIToHj-E}!nYmje6lxC9Ey<{+~fzfx$6?Mzt?3V`G?%hUE9nRL9^4{(3U-eQGWB-l3
z0;Wr)S1Z$xhccqh^gGUvVlPjHe2W$@U#eUIzVv|9HAs=nxdOh1Gb2GXk~Zu=uZSYA
zGW8#@m)^}*`uEj1+lI{h(^_Y1Lr2AtC)=DC2X?@Jbtue(>D_)tXrIGUi%b1*UxtbI
z46&EH?tgf41ypXX(q3*0;STw3jMXsns2|a@eOcFw`dO&-#HwqcLEw3LXAxbZ#4Qdy
zr*%G5K)7;!;z|_PxbV_hC|&-bm4Q>@xr6cE(k0cs91ovltUkRko4|$8Ks_hgZRO$S
z_8g0nJ52@AB?NEiimSsDsHpF2$!AusaUF`dskzDIHs!E1<qcx<Euy<dFV~AA7Hp}C
z{4wXqcu$_NW;AFT*qg3pK^YG!RS9_KHs$?GG$7j85nqO6fHny5Q|%f)+=b?#Og$9w
z_-JUeP}Bj}*y4x1-7TU7<}XlUM#aePun7h7xe}AnVF}Pe$SUyTK}Z~Yr#nX@yX~{+
zx_Uie`E4d_B9Ci?F*ZJMOU1Fwg({~$)}H88{a~b>WsrwvpZ05xq*ZKCN|3?d7Et`V
z5({O!QhRx70(qOdMz+0F_u}hD5x(CH_>Cm{@P`@nfN53y*GMNkQlXH58yXq+?&K8G
ztzQA3<lp|7m|Y4^nN9U&`iiQp7;?6KkzeK=-)|i8Li&UA+w@iq?DeQeY$3x63TuG$
zBK$=H^ZqvH@)a<Vr~IG&?A|ia3)oBZUN!(s)^E<wR8IU4*Y@v8uTE?AfI2Tjko@!t
zkQ(+gzwmk3Lw<4;`}+#m9K>2N@*k*7PXwWGq(IB1Z_at+=>hPcZgW*ibb@%k1O#-7
zUY|}7<TtK+!&8Gp0T<`_^M}lbHeZovG6WdFv?5N>V7U%Y0p=(qvSg*NQJG!Vkq@aT
z;~&F{>`4DJK1#V2h9y%UJ;2XKS288bDa0Q$87Ao{+AAF_4`cY6BmF|VZ8}yYa4`8j
zcJ^F~;r*8JlwEXMZq}VUefI19YT8|Oqm!uxrv@mxLfeNmj0$(~gu=A3nc`!M$m7Ny
zDcvf(0!sf&P#m$*GXDS_05%5&vI?-#8YmqN&`;or*<sn+0euhvJ}h(pc}^c6<9Wd~
zus}eu?0vwG1TtFfrVXLUdhV<@JL4odDhf3Bu4Rso9t57{RnqnIsuU#4HdlbrJi_x?
z{bYz)jaZulVh>Hrh8!GE#U+aADX&t$jkX7V*-sT=J@C<`T|u!V6wGRKSebd3nDifX
zId;9rXyxzURim6;-}YN877obhwlBbfLFwrgVNlwU@u)#iL~`2i<m3<(_;vbwn1N$E
zx&pd3H;C2joh|I8hRR&5iEc6gt^TTOkF*wOgQN3RZu=*kd4yR#Vcs;oj9ugr)TuH-
zaRzey{?$t@yMNFXR%Y(nYRM*P$KJsC_8_VE>CUK&Jj>nNx8=BFG(~S4YieOkh7Y+l
z4EmfaPZI2IaM5Cm5#T=g4aHstA=$MC>BumPUQU?*CL9Ju0Sun(CwgMY-wTn_l=z_6
zwoVZ6Ne8zH2BO?F4fZVwGAh#9_qvg)nRHR-BP8R^<Z%|LvWvQHZ57&sOq2-+I|fFu
z6Pat%tt4DTo;r3{v{2lBOF}H0VKgO2IWMW@8IGO8JC4eOgBJ8{!5Hy(%=O$dByd}y
z8fwscfVjH>1dpy`N^T5_g_kvZMfxL*pn;&pgZzyyK&S}lq}gFLf!OdGHQJ+eG=D%c
zY(j7f3W3}r0JI4)D27T9H+)AC{RNgKGVaYG+#fV!j}<&!FS@_&AQ%-`o6#9WFx2?E
zQA*~C24N^e$^0fFoeuNXdQyBG^HE2ht{geX4aZ5jr{gY&F3U^7YEBet64bB^P`@vG
zSEqJeIhJ)ZbzE|CQ$4|2^F`so-dT47u}4AGS-j*EJAtT_fbg6QFN(rS5_`~@iA2}y
zq2l7Ou5E%K?z_h}qB@dwoWFcd5u{368*uJo-{9nswWR3S^j&hq8b^L>Ys3yo_wcM@
zADgp;UcnkcW3H{)`ZVKQL+Y)pN&_S{9;N6cKECL~14fVGK*ET3Klg8jX3yUQqak}b
zk4or|)SK8S0!KH5m-`e#=1z#sq}at$IGf+S*4UFlTrk9~2E9v<(hR0?^%Dp}H(gG)
zDt+r`E?@Eh%>OrSua3`*oa%-S^1^n^t6h~YJU*s;ZPmi4!5odNE4jlAmW!wmhJ*G<
z8dc?mqPGz~+9JFO$PWjOhUgp)gQ}`0=POk@gn*74rIXM^*pPdaEXBMbqrVrDw%;56
zx+0ZmKxB8#pDXscP}Fju?77hHoeNSmo32cf`x5XImo<CS;w**mH^>I7d_OnkP)5c?
zz#v@2UClKsDi94QuDKbDC!J{MdEXKZ*S&DkDGe@9(g6*FGN}K5qYG_lyB(I@wK)L6
zCd5(_3al#t;$VQ`p@;f=f{x}tjUe<WcRCKn-gxGte4^mVe=+@EGE<DwQD_34Nz2%C
zT@4DsxWo&$Lbgp4y2OlaL_$Lmx~|*-p-qqfE>stT!;b$!9Xn|bdW$8dU|={ZtT-*N
zeCIU9s)Ea<!JzWz{%D!W2jLn*sd!<J6q3AXIGi#j;H{#4C7~RwGU*QNcJ7@}6&a@V
zGdAe+s^zCqwl%nZ<mRfC21KoVD2h=^?u?Z5Teo^fu7LG77YESa2Uox;=mmU<2+4Ut
z;v$dvcrg3lt-ybpA2vu|Ci&0Hq5e-#%QY&YL+M~~X5eZQ1YeF)u&&ozun=YKqNF!J
zJm)egd>o%zb6<koi0UO#ydk8NpU5V*yYJ4srHC<xD6)fA`DMIyczbdVHOI}>T#3%?
zBs^s|ExSuoxTCdV#jO1sEv8WpiO?ewFNkl@bcmD<ffiGozip(dye@g!OTTz_0~3=x
zyE2ayzP3uegJcqZHD%m32=FG?8kz2S!Ij0~I{oo<x4qC%?mEtBg68}y)KG+CO)i94
zfLId9dJp;y6wE^%Ck2li*nk;fCqf3VsR|x|7lP9V@IU}<C@3Hr!A8|*3jvHa*Y<+K
z7o2+s#y7)o=(UX4;kwj25a>(8`rxd#+?)c;7!^iGo5w*_NE1&gdBujp9|m!f1uz;m
zVeu4vRZ<>4B1<OG4D~?yHtp0DTpQEll9#1<fyVK?--}c1PKu0p`V(gTw)Cp{!|`=h
z3h60h0t4U6grp8@<Ji;ZSirRj@-I6}s3o7512?vRYdJH0eSfCWcz*Qm@|0>&Lo4*%
zQL+EYE)6*SXZnrj$KYYeO=e^bsN8$t3`bU(Tmb{m8+S8h2&OSt!0H!sP$sqwu7LgT
zyk~3LSpVzQfN#<K<;%8fF_Ic{0nz9Jh(?2f|81X4q-UC+y7KKy!ss+6a2c+E9`RoV
zm)o{HJrON?YH)2`4|!|_gd=JYm&uqjB}FJ_@>bCg-M4KlH=pGXVp}Ajh_ppB@UX+Q
zaRSk%M1VF7DFZ&@s6lM{?lGzLqPNK{8~Mjoq*ZlsOV^Y7rJBK&D}Jux(I^(;uSUla
z0sQa|if9x97zt9fVdfKB_e{7EPi0yX;j3QA#$Rr*t)4qt4*GXB6#z>O{RDToomd19
zK&U;i?k5#+@QTD>uRIdc3&W$j1{7+k+BgN*2EcSk_KpG1*FPC#pH<JYYO>C%f@a4Y
zrDCFGLHK!B2*-RUrD-e@k@M(C%oaonepN#H^^?5~F4L9U>*4g(4aKUrh69K1D<Yc*
zaYjcf)PWQS&kXgCcZv)d*C-M_-BLR0cV&_jA0Qg@d^bXGID5MKcE^+{7E&mG^BJ?L
z(-(S8>>h@+mI$#Y<+rbd2&JeR-m)ad8c+VB-K0I0n<F^lZ>3L8{k23O@1?8KkHZ5A
zQ=^<?3Ka+it#+sSCrCH8EL()Dk_+<Mk@Se%aPW5Ip#E<3@Mi!4Ex}iEALkDo(S?fU
zS_#w3G<5t@TekVDvUrVU)1a)}EsE;N(eS;S3ZH2heKCL2Kgp7AagX@}pgCw%)ZNQU
zO#P}^6Z?m4x~5p6REwkka?Bw1MaazMa}4v-YnQ?e@)J&ttLUMl+b$;yAf53)wK3f~
z(`YYWuSY716x9FlpAr96`;WM+yS1svF~!q|rKl6zWjnDIv{7TBJcjcLMVoS{83Qz)
z_5mDXx!04U4QsMMK}d!JKr{z4#mkBUu?w6Y;O5jn%PW!yP5p})FpR7$LquyPH*184
zsThknAL|qGkH5$FSc1cEf@lXymqEabm>zU$jdh2b#6HLzS8~j7h4o?oSTckwuZs40
z39Gde4}TgE9^mPJ`wFV6Y6t~yn{-VxQDx^eu}4vg=G8!|#YMN{cv-V+D-FXxNf?X|
zu70lTHC0&dZW$b9x2TH<E{oHxZAL!<vB_U1<2tfUA*!&yEYoKDIPABE>Ac*2ax7v(
z)^Bf5q8SGQ78#P*9s1QOXB>T2b`R3yf$=<aJ$1YFjN5P(&zUa0u+sol_KkH}>q=p9
zrco@G+&(^%5u-tl?%yf-4}3uDvTsad{(O+bMwM?YGmi|(Aq2vmZ<@lPS<rc@R^eWA
zbBH63CfAC)7x&Bt&ni}|nZoN!B#}pW;JpTol4ScZNqdayVtKl+bYg5ur>`1mkAC}u
zM-MDk(qD0i_Bq13gLLoOTEy0EFn?m6oABRkX#DMSysfL~3%91p(jl!7O8Ba*UJ@Mo
zed^(2WMaPQcqi8@^~rLRkQxskh~_G9<&#?|!LU@%!&0|bOapkepBvD|m>(w3lMtv(
z12GB}1DhVdUojpLKvPCQHPelV@s}#3n-sVqa-q~?j-CCA0<L1R2eO7cuXvsKKSzCj
zxIeIp!*9>_Sa17xkJGoApfE~Oz{bCsz|mmQxBn=alCO_Y0=r#1U_g#s%h6yZAS;|z
zZQPh5xHs*oUgQ(qexDdEb5mD7JizfE2+V|c271nph!0nQx%rvq7cS|~#y#$fWjmGG
z!#Np9i!fmhyjEstW(UO7O6|C4_#gwXO+Z%pk!K|UKg=Kip=%Rp*09vz|3!0wtfcl+
zm80uem%PZov{>A98Ndgp)Mj8>=5f`VVvQBiJ`oY?w92?U*L<kyFjmu#@%A?!%j%L)
z6SV5u7RH$(&I%oIx^5s?LD%MKx?0|{AwYMZ@qH;WS+vy4Q|W1C9Em9$Bh-X|))J28
zXKS)W;po#83QTVNcxFYwn)M|fE*NCu$*;ubd6s9ntKZe<aU1hGD<Bv$h{wanApXzD
z@ema&3ONMkIgwEYOTOGRby<_XkdVJ&mVj2RL)zHorv9VQ)Q<B?;fEI2GPx#U5Z}O`
z4WB(I-eZcRe&?2wOv-nc1m+=Q`aqX<uPY;lKp1X0juLd%!v9oSa7)nY>@Yj`uw=Dt
zXt3kh9#LMGC3txARmDLmrw*>VhoO4E74`gP!e>d&3!|bUD81*f196Ix{BX8&94YN|
z)5(oeRo(s6@zEoJw3Rfp6mrwzr-Fh;cK|4haxbEu2i#y08ee2vJ3LU8bzuM0!7xF#
z=liQj^LxYdBVoIu^jpmK_D<e}3Evl#!VcIPv@YV(_ovPK198q^AS2WGfq4p~U?g+T
z*MQ4zt&?ruRiz_1$h!E81Ja*jFJG#i4E=KjihK(sXwrr+J^igY&yN!KGyaz@a9Y6d
z>G^B=$H}yJ1LvMe`Yk{A^!KVrSc#r=%_oX}JNj-shuH06NQF}s2=zNeJdQgBHmp9t
zu$X;d9N5DF>UjBB*OWF7m_qp%p^u|sgG_h5;L5H|Zb3F~7FPG6_`!^ez);*G2L8qI
zAg`%d;X2onOPhgLoIpzk!?Yefc$ymgW9Z?De5`?NfX`IaLa~9(TacKy!$;Aazz4D2
z4DS7THDVIF<s|RgUE3$J0N?C$@+B#;)M^>ke<xys_1Cn5p%qkU!y62$h@elXoO_a~
zbI4lMXk|^@4ocEi-aQ?K)Bcs9E~%fT-bT`}qL3x!75}iYx5b%siBC3wO@!Cl6)@-0
zUC8l`o{#$qc-F`@7;F+jzN{4=?>=m08FZr8^Z6|QcXoVmN7xnMd8xPio*8^UF)Dri
z?La!7zs7PWz57Fw+0D+|;BkHU6R<n!cr9+?Z1JC_tIICk6{W0Tjt$38&p62X`N2%_
z8z%Mp-_IPpH3pp*&(;(?MpGh!o5>VseH;T^M;G%JOLv>=7*#x#lq$@O4K`Kp=TI)r
z)0L2_wXnMVuA7_OUMZr=r70K4z3u#@Cu0uLP0U8c*tBUfZKYlwmhGqiC1<5H4D!M(
zIlC(IS9*7Jw#!b<vhh@+Ctvp8(S-*vP@ui+g!rpauPRK#<~K=Q)GT2;$Tk1xz780z
zdTy@w@i?RYR?>B)Aj|X8f9Y_ie*C>uuB!o`(?irrY){nt4AXR2b>yL6!^vU;f-AqZ
zZ(FFcy0m?MEB9@yQIl3237NQ?X<qWr4+h=CRMlrx@073=hfFj#Vmiw^AGUK9xf{nm
z^n9EjpdQ#4622gsU-^JO&QDbtVP*WGOP7rSyM>eRru)m9h2vjedi?zhWJPIW2kP``
z73u|axrVuRyNHzF_|K1kl2hl;PAgn<%F9`Ts+L|S$vXOaB7!mAF){<)`E(2K7`e@B
z133~ll&SJ&<;^<Hp6m-QFEV=VfBD(?dI2T;MLK_D+`IoYe0TsWSQ|&M?mIvSpnxBz
zF`6n{;za+7$$m*s`(Hlv+&w)7$<b#&!APs$B#~kf!y^y9y)&Y_3&)-&kAtfWqu4a7
z?8rbC=c7FKJzdmcA(k7O3k>`-Q98P=)c=a;a<p;mGVrTxK<?mgJaAgj+d*X@CUOlP
zJv5@*WO_A#yZ_J~8!<G{sSM&S?}&r0(%<4^=5mg3k1tCjNY(2=0mVF>Vh0pj_+m70
zc$GwF)Yw~`%(}|ubQAgnGUFidzft5pELBn=khN}tW}@_am=E8<37+Qpelt;Nq5^Xh
zWLZ(o*Ix@??U>XfrDG0R;QT1U%6HMgqV%<m-TLYUmo&!FWX4nw&zSZ@z}aG2VD)R$
zgg__GLa%8tVm$k|PM_Zn$cBx5cKyxG7hRS^gygL?^WwcHupHPGlLMnGd`)=;__KM<
zE?qwJ>o=pVOMEKzl<fCvDSwL3hoM5e(0)SFOE(qHQ6Ca9dpgUgD6xL~$DE`=_r*Fm
zH!mS4x5xO+?gtw`;C}M?aJ3xLWyxxf<vkrFI>3>=BEm9a9jCIKOuOW?TGN228Hynh
zZut_qUSC(P&nW#&5P4H~Ea7}|p^4<Ck4<i9%wybDX~e#1?q^%1vURI_TcFvIPf$vr
ze<c-@>Fd~$w5o@@wW3*H9vBj}DjNli>Yb@K`m@!t4E5b#ST=jxSIq5n&zNxOmT^r&
zq~larU3lG(0}QNRtuXEfrkwI5t(0IqX;}JrW*;*;??D3*>il!IeKO?M1ABNLbfWJe
zf#gP7ZMdGG-Ar4jU%H6)*{d2LqVIay*&F0>Pz^&*uKh$rN8V-U&oKHwVcRZJ^8_Pr
zKIYKQa^IZz6(}Br9wN(c?a1+ts8j2l^a;lQuw?E(0McOdZ0AsxhwpYzPR-Aj{DYYk
zj?XW6FMdlmqsUp$F8vZOPndryMf-x8Jh0H<RphS>oX)QxO`8iwOiF)r{)bM0-@J{h
z>;AjZmyj1U^ZN&l$DLOI5%9n5`k};8k}-|i2yQRq<ya3^*yYX_ao%ebvLKa^8OP?C
zBiqNK(0NLvLJG*`lN!V=X=9H(j&D~Q94RD(DGAH)@PIkhYsw=;Eh#(NGDgAB8I@M`
zU^t`%5Y6|>z#oz!oa26=t<milH$u_6@wBg_lT0uoK4nAiH(}}Powt^Dc`Vn)9#jGv
zq5NEcPBdE#0^p=v9{};q9Nt~vF37T;y<L~JK%$&L7gY(w>9c?;2(!cT00Sflpx;Br
z=vSFk{~;v#Siy;Njw+ytQ32;~mvj$KD@@1B{B-q6JEQw~eNx2N9<kd^`ge5uBDcQ`
z`}=<OK3GSX<z}R<gwasCowYB1$0)(pW4rn!%5&a1xZM%EKP7d@H7jklosN-u;Z3He
zvhw*pF(tWBmLY8mlemX|Z%aI(6-z=_SKdYDo4xe@+^*ngNDWCxA!jx6^1Y>Sk4yBF
zq1eOmXI!G(iG&^!x9^2IFn2|#iEfPXl)m%ulh;(|XZ~sa#h-qklk;Uo{mrj$$KEs_
zQT(=;^)+1@i~V_KA@vC7SLq9lid<dgKn<#>CnOjftsEgv&lgQ%Z5rR2vfUgb>x}vO
z#-p7WEv6xq*PKmn>PynNdXk^A?W6|k(s02w`=~)J$C!@~>07?m*qP6aSLX9LTIxLG
zdREBEkbf3*0@|#WLc}8yt_sARZG|ps3D-9P1~j=GJ8%O@iT@$zxD7JncEp9!WsllT
zrDp`NwBPrWysX`KOSO!AJXTMo75d&sI`&BLNd8U`+tXNyy(*v8)CnkN;ia#5hj-cp
z<%|k@D_>mkmPW?;yUL}So}dwo04a}*LyJ*EZH48LD1>&wGMaa%Uq<80c`L&%L6}7`
zqk~C^{yeVJDz#C9a<B1B!hE26Mpo)-ru90QS8G|2bl;d>F(H0l`ufRRW5*>OLENrO
z-bDKj^M*l#_V@lJU&|6E#yfe+Sp?Y0>bSP3KB)jy#O*4pBPqbz-}oDvL5_2ge+4Xd
zXo9(oj~DrmscxP)z?@B$hS3Sy`8y*t7#1GbiGqM)p`(?mAF?bT4S1c(dB_loj}+V@
zA`lt@(}qbvt}x48z0r^G_DN9NUS@qP%;YW?d{bW=v+|>do3i5oBt*zJc7}O!r@A^i
zR~ySQtB@gLk6c_bcc?wE`p{5#sPL9)=K6S&3oE0R@WZuNB^4BgbOX*~k2gl^KePt6
zFg(lIFcLEn#7Bp13^W(^3&uDQx0W?U;L6~c?~fP7E5VH5!O{ICHi}I@$W*pqB}xkG
zn`U_iw-vDeD#PD5DtnbS;2)R=Mh&7w2oP+AU@3qamGQ4ALX`~qgl^@73JcL^@S&8L
zW5}P_KFy1aGw42`3hpr>%L)v<b~J>mXL)rW>@GI09gQb_8k-qm%tb(3uRo$3NB@9h
zI`Jp6j^Z#6eoyiHAy>7;;2R5Tr)<uTxe~n#1@{yO(hhoKkaF#Lv6d?$$Nlp@-;>AZ
zsIZqKXT$u(e`4<28#4+mS~=LAIxDCBaE?15_~Vcr0&qt-Z*Aw5guZ@wkbmM^%n`1g
zCygw2bEeZY$c?w8+?k@e+p)yWab%Q$Z<JJcysP&%rMz!ei|v<*ORfS*n9&U-$t3;M
z-jO$AbW1)=V<!4l<P@%A9jkIavk8@Ad9*P(=|xvSt#rH>XJ-5H9X3c(hAdhY-^Y9_
z#`wt|R+zJ!@K=JumWiRT8j`m6TB!1+nC$fN8o9zf{=Kf3^xF~r?B)tT{KdVeg~6nl
z7FZVZQ5|)rm<i^}y1;ZOm_qy5g*%<6B!8&^u&TX6Q}M9rGXmfR=9Nh1{jGGc`06o=
z{6G7V|JnUNJp54UD8ZgtZG@zoTC2*Pz$NC3lyp5eYzLI-Q3ye<u=t=r(^)*^1D8y=
z_*mFM4vxI;w+uW+`^h$bexGN#c_;)7RT(cS6Y&C0o1#KSrKnO$W6=0=Giy64JJ?(n
z@y)^AsylgI_mL6b6GE0HW}nfZteUZ)jD4l@)dr)pCv3-7i<>Dw3YiQ%<yz?d=LIZN
zv0;(1O$@740hvSgdxB8*Czcq~h^cfB{#MJn0uTDoAPiEcU7gSRGc{EQQl~Q6`XuhK
ztDpNLSxKp-&b++v&<8KMaMTMx&uylAxw`P=1?TW*=P<&zD=|r+6ZjhySb!8ASjq8^
zKLOoGGf4XV{U4S=jebJ+w;{I=z&+L9mcDqBbA<1OaG9g})aYp_4IghFQqTD7bP}08
zCwNNI=!<Sgv!x-@#OEj<j|`lu`y*0r#TpKpsSQ2IH3=`win*saCQyz4a)kD)Er!XO
zhB4GqF$#qxh4RL1GLlr>N&ePJ@PtvCe!cnUW7g9tHiTZ?+`N32$@2^a{V-=7z!rn3
zDuXwxu#xhbO|F}b^sKXQGqV{RONxkK1`?!p929wS+{VB9-%b>#IEwN%k8Tj}>WPnS
zPc9tiUw^DKES!B%kaF)A(ov|}&;Th*(BSs?&ZpnB)i+Fs<x&Q}or&J(a}P<sn@!mD
zc+6YOSh8@gMiSY&L1%osb|FA8cJwEOdPkmyVd5X@Ip*&A_0O$~DU~Hk8f;Wn!FK^4
zm#xs_Xa6cW`1+nlM3XI&Zt=c<^Iu_q0J0VjL8i75tP~8*_5}Iee%`Y`{(n-e@WG7J
zYO4A1FR(&V$T|J;1xW4slV1V1Wd46_DH?IU9m6N_v+FVr>!)<(?)mfe`NPBO@`L8=
zv#d{P%N<V-s_?MZHMAOJsFmglI3yI5hxDy)7cHa)C_5S1ZR!%?ius`ZU~e_;?MCsk
z!ewALnli}LT2m4~+zI5km`;Ds*yoY6H1SA3`u&4S6Rd}RmY!{94v*v`ki!)3pTB2(
zdtO)_lW@nQ^Pq%FaV5&;qQvjnl4drS_Ns8ct>xD>w<d;(L^xt{16?=7E?Z9jE5ltu
z{`sd_SRhvQo#RN1Ra&l)p%%G3%f<1X`iOYmBgQyO3c~%7*p!bBoLMxiFpcc`3fka`
zi0+#>Mm|-qJ9DJcN^V?YwK=I`=H3g$F<ZtqRz|yjH@PnzM7!{bVd<dFCn`~(uJ`!1
zf>B`Nrp%4&NPOWRW%#QR2;kX2S_7uC;UoW6n!%CAaUKo=+dPn&sUOFGi-7r%T&Ju7
zC68lL_mrU%#I;(aKdusIFlbGz+`o)TN5pWnnafn?)}~?DbnEQOB6Vp_4Bkb2`z(A<
zK1xd}Fupx{Vy#+^_=z>~YV2oRcY<8!cH?+OK~5R&bvANq6<xEcL-EbFZ&235hCJHO
z?pHuJCNtWnqocfJ>3ip@6Hk|An!qF{K_02u+c1M?tkn@6;zjv|&fnqT`@hN85ft&!
zD)QzMNk6H?UaC6GN&Fg{O=PcOTpsay@^T{HvP$5~P)Lcdxt2+i5mrB!qroWyXY`Vp
z5T&o)=$_*3Wuf<C)77v2lq2Mn<m`FzXMGlhMg(m*%4w2``4}P`sDC)YE`O93@hV0j
zi{V^~>n3!e3LFHveq9O)jIXkGo(k9DOn~_Qa?BgF^R}0#rLVzE$(v~~++OVkQN5mT
z(eCA|fr~@X73j~OpQN8{L{@=G_<>{)xEFwRH2<F-`%4DddXEUTM~m>+FMFiEivKxu
zF=VU-(u${uK-9%F@>vFj?WOU!@A-Q+v0RHErM;iUzI4@(Z~43GCF&TtGK_wJ3NTfk
z=5RoCfvi{{Px*6Hsd8t#5>lD$w!sKTMdHdywsZKVUUt=dmnhR=eKbS;`xK|a%C{Wy
zWJ&lXUA_zrIurABMfx{WNj4k5o8E8vot889)M;s!8s73SQXYOtKe%}=7W&94`u8ep
zJ?;#}V&T$6v@t@-Niv++uF<>RJdSxEt+EqCQKdA3Zmkel__K;vzV$*F;!R=_A9ukA
z#n%7wd&c1g&Bf1-RRh1F6;2kuqujEv%^AY`Z+5;K^A9o@-oUJf#Zplarg}R*E&fcb
z|Gjs$m@HpR$VOz^BYJ{0`EhbD4u)GB8~zaiAk>L~$8rDSzOF$~#=<`YI=HV7zz6@u
z9e~NAO3&THz}$c-`$kuT`jploVg-$M@C+T9n{&)j6EI9F&UK)HN*(QXc*<@DIt@8r
z`@Jed*~5G7V>TigIiW*EPE*(cu_lGl%_yTIoQ7vU97#_8!_IMZTC;LeYlo*3nTc@s
z^ePACOJA~*d(00i*TQRA1(b<wH%-R1xp<{tj!snJ(J60yr0VJo&2>-Z{n>7xFyZL!
zOu&MYrP42$Gd($tXLy`*9DQ$aM<)cVfKcBQdg|`YNhi4(@R*q*ewRC2Z!c_53-0lo
zQnJnFN6LhrOYd&p@?<H#NY?}-()EEyw7GCti@0F@R9ZDtO=V>At;}tGYqP4R{@RL=
zy=VDx?RJ0IR%lCRUKkj`b)r9iEHAX{&RhPJ86UZ<_mKg}|DGaX<5-?(I8f?9bi@-G
zbk0JwmUGNkP#@RXCS2@kaTmfTDR`uCR{u{S<2D8vv#uNBc9dKDodUNsWVSfE;g|o|
zLDzPq&)TEWKgkVwCDAjVH26LYqO2HiFB7oEb_n+)o;UW4INq;<GWeFqBHM1kWZD4s
z2kh2Ov!ds&#4MjJ7m8mts7`kxX75$7);+Bn!@=jSEo7mD>`ml5Bj5k2R#PFL;YrYo
zQHi0d>QR1J_xcInalgAHVH-fa=<tkZ)o8AOXn5JsYZUlV^F(u`135i(s5ol#S+VBk
zTiW}t#$Kg)9IG%;Tmc*FGW7pQ4?)1?$o$@7zW>C;<;X|k$Q^XQP4r|;B7=b?Wy%LT
z?0#u)BcIlE7$mK|vMO3$#q_H~ra?10zFgn?Zulo8!iFs&T_zB$F?qcZ>zdVMQ2zxz
z=fP%C@!tH67ZtJ9!fVvGueAv~>pd}wK3-Jxu{D6JG@^`|KIyj!`yqDLy!$?GnA3nN
zken~FO3BzOeyZPD`lLSW7WIUDbVSI9rgVibDP5et+4k&(l7c1({IWz!6`qS$imbGo
zsg(ZBbq;X&mxPtF4<+G+hwr`R;0uCl-XuhAwPVpqIf78mVrM$x9wGr3J-e09QA(1j
z=0eBi7gQAn7UM!cyfDod%B<;QMUrrGd#ce<^iPg3AyP>8pt7KD`yN1iM}D`n)Lx-O
z5#fFBA+|ziM>*A-h1=!=V=0ejX$lGO98}?zLxKpwzAD<IzSpt6Ocx5;1HwBz5rB{`
zko&k2$cu#lIX2g^2hKR!8^{H-BDh&mL=E<%f6?C{9!QN{r|k5z;;#$)5D1skZ}6tA
z%EjMA6OZFVGj!m5S%rK=P;BClb*5>xvtYfb8lU!PfrklU-4p-e?9-8`ku_rXR;QD(
zx8I}fKA7hVb5$!L@V(NQhSJ9IzSPR~x41*WtcdH$Cn^KeOGfUEwULBd<n;E)L))^6
z@@#rF?_%NYpL(VMK`Ga)!|y|a8r6D@-JVn0QJ0wJK3Y>JiQ2AhG=`hNMsYnLVs=8E
zV<j#J<M$>UolnJ)wWDSKJ1{u(G@^3zwm--6oF-%+y<a?{lg=gTJf7EG<`xvDd#TS1
zTO1L7(?HcZR2p+!dV8XRInREmhn25|v2pIO!~q_*u&Cp5VLBd{5)((~BbK1B_|Z+@
zR?KqS{4OH3u&ddO+$U~yLO`q$EEAb$+%pYN?_r3J{E2Mka<)sBVf0z}<ZM(!kguBa
zlvPAp5NAE5Yb?ZB^LrTDUv)f|zpy-^IZ2T_r^y}?&*Edam9PMPcI^1mriPPQ>0@Th
zOVO1yqw|9oA?$awcNkI`j+XXz&7ff22($GITU#ImH%?KUgiJy=ZC}fu+qb}l`{J>s
zZpHfga5A>iv;0S$L(^PBFD4D{;(1IkCFk9!%;3as=i`*fWm3KTwJPRf-i~5E!btQ|
z`bNL{mii#BZt2PX0pJH&<1KkY8#v!XOZ;20xM@vvyX$-n+kD%>f2Ze!d@uV9F@N^y
zQvTcyxE!y!+-d~rdwS_AFdqNRyCmI?BL81c#A%hbw@>ipRgcCZnlQD!tebeND22FI
zWnU;&QWF}?%mJYTr}e9cL1}TIDu2C`qW2sa&q=1@emXXrFx7xud7ft_23tW9!7@pr
zqC|@x9J3O;pRAVk@riNwgx-~yO>5vHlz;KI)or?fI8Vf^uR|1fG$H!_7W3mcETI{L
zsa$Cvr6y)o6Lx6pIfi>Lb~+%s3;D@3rj=qg#rUZ;mOCqo$!)Aq1n=eMc-#wD64o#*
zeSN+LzRM5cd|a6GXm$C3<us!4xh%v0lRk~qSe78WgJ=UR+?M7kdN4{^U7eoO>giw*
zdZJ={m#2)Qb^UU1b-#rB=}&#6DVHcCS$gsGtbeg3N&2kcn{~q#qn&I|HoOo*$<59}
zRXuAPFGW=c`JOL2dBkeNY>K%1Bk0xl5!{ox_wDlBRE#H8`4N=37HGg5sZI~T|D~4S
z@B{f}Zh&PEfISq;5F`E%8U@$z<bdTf*GqLElX&*_>6>^g5<Jp7JPy3FMk&M!%FSSj
z#GZ;e(Q*1lpS$?fkFgGS<JWmP@`yavxqvG`RSR>^Hl;MkB!Qx+S&}tWr@<lXS4iST
zzRzR1he>?KT6;c38%=|!?W{s)S3t{q?KhPLS>79c4>z)<5Mh7>{X}IA|HWeL;m1r2
zOz*w@WZme+qGKb(qU4@Y7oF4?<a=X5zW#KJ>SC!>Tzj}y?z54&kLBNVFs;*i&K*@o
zHM*nd)wX_tLg)G99>`2RPqEYbaE>8<?Om@O2diMN22H_D4U6I($)bf-$s5O(1@AkT
zRhSO4d|Ot;kiR`tDL&lyY*;>F3Lz>EJ24qbN?8i(5i3>f3;M2^{U`j@f+W}bP!Fjt
zk;4Ih)#5unhnH-6VYIPs?gVG7g-Er)v<!=?Ac5ql#TDV(u+sSQzR{<UB(NGd{nLZl
z|H_$PC>c44JUtp>ZIKZ2%DYRtqD&ikFK950&*#~?8C2g-1UV`EEjW0is6U^i$Q`Rm
zbil7FI`2<~MnhTPUXLuvPI-H;foyCIR+x2QdJ+jgq_;==PpnGH>2#lS^_*$xT2>6F
z;?u5vS~4;+Nw+p#7g~EailcOKy#3XHsnMq-O1P=FTHI>rIN1VZKrdsK{i(1fO)W|l
zXQ^p7d`_Hv&|}*qN{!)HhtNGH7(|?JO_mNH#f;{c!H0r1``HkzAC3Q0RJV8?)q&|v
zUs2BjT3v$gIZdfy9Y*vq*0q^iPFn81BGPgCUD*q{j_Sv3p=HgLH1T}`1Rs%6XM@=A
zTWC84htO(8O&>#Ue;;QXX!1;2b0HSQ>!TD7V^yJ3evayc)FHP?m3v_&a!b7C7+p_3
ziC-Su>vGt=uEpQ5q+3#hW>b{A@0Ip9F`+#3aE}OMl!w&@UU^Y*WYX4ZW6`b^NsI>2
zOl~iPDKvMTJB6imj9T8{+zBE)wK~qj^M}&5sDANUOhI?AZcT`;<s1~h8x3WIXnrcy
zyQI3Kg_v`+Y3b&qvf-Bb;hbT3SEPq|7Dwt$1ET74ss=89*pB0?j-5;4a38n$2fb!u
z!aSsmh;B*Eu1`VRHwL4-K5~z4=LW8i>shacp_jh(?=6dbye$#8hZdi3Lj|Q()@I=D
zf@|TzhGcGNl6JUsqW~>8RV=nVJK%RJxP&k>>)BZo8gL#N25Z=`wp9iHE}Q!gGR{f?
zz&x$}y1!rSBw3FZGZdp^Gz-GuJHgk%Bvt7^w?otfIii-bSGW)|p2cIteDL~o7={ri
zb=G`sk*`<?jT0gDu?^X#ZjlQj@T}qNz59HT!7qN(>et0mk<`2W!Odoj_5IFGkW{$;
zfQ*f#c8O`Fv~83Dy;Fz>>-HY0Ioz;(U~N_ZY0HYi$|fJzOBClLmFvpTgJ+Ygu_~tI
zee+VOI+Z?#N`isq-{4s;Zm$|92j-|?*5UYsr7%uT>xq5F2S>>UY5||o7|GtD;Am~y
z$wje3U2PqWjRSE;nC3PQb?!Se8EHy#o(odb@3dqmg{fu_3U4Ci85S~anGT84AKtbZ
z`m+@l?yhy<?oYJn8TtSTzjd)?*k!R+L}g95<ulu_qp@?p=<M#Rh`{iiPnNYDWhOV<
z3uk|%@1*LwK*Z<{n7c8K^0TcUyi}V;?~9h!ZbcYWvF30@F~y&;^BEnkSOB+XVzh=f
z*ThOn^3q;#CXBnWRz9_Xe}lv*n<qVq@rig&H5^*FC$N$>h?qIi@cNX#E9m|lCpC6m
z|9)iy<+@|SPqt?jSWlQNfeb2^XRGm#>(<o+j=bm<QrCq)Q$;JzzJDWzBTAZZQXU(&
z%2=-RH8%5MjD!;PvC~$wB7v9D4lWa!Ssl~&Ab^bCIsB+#$AtXh6RwA`pQA@c&qz*>
zi;sq<W=z^}DHWq11yFd~65#_GH>9Rqz_Nzy^R+W;uxc<8ELPY~0NJ5w>Hf&->)JrO
zD?l9hUrr(?y2vxL8aWu3yDseWpAi2wdAU}SU*|R%1s{b#z8x3xC&-*|v;KZq#ilH&
zQj%!bRJbEOHV8xZ?s2fUzZ@^&<VTZqB~~fh(Z(~pUU<z$Zd$MWE4UHsd)ce)H?G>n
z>zjH4eQ|k;oCJ&&ZRYzNaH1b!K33XrAxEWYEPDz80%tB`t+>VpqOQRmaOGV^&U%8+
z>z>eRADro&ufK9+lUiJZPlE>#VgU+*cZFWIS>QVAdD3ob1}Tn`9jSTG&7F<Vz*I$-
zLR?^DRG(8<ijiFmw{CENOMvqPGzb%HmA3_`rvld3jVFlYc{Njm3>mwSD-j<&nz($t
zga?hwk`-JCGGi43TM3YUv-(EzT0(QRN5`4P#n|`0`LjDdIG?<{53Vy-d+vNE0*vY1
zWjD7?nOdes;U<h^uLdhbMcSX^lKZ#}*Y`eR8r>Saj$x0Nb`CS}gYTB%8KZ~Ni?I$e
zh{sRuBxO0jS7wfAlP8Owj?UiDX!v6O^2t$~JbO>4a^hsPLR0DZ@`)$y5fvoGX7B7v
zBt{}YaiQ5rYj$x|U;bMg9|`ECIeh@QJyr(ZKQt?iP#j=T8hcp=e~C3U={tn5C=-$!
zk^#LoG=r!n-E)GmG)=?wh7F4_Ys#oXG0rt%q##<nCS1prhO~Y-3{B%rM(t?gz2rz|
zVhNlD{QA*37X*JMs^W?jvk%OZC*J(A>ig{3`c%me>Djz0DZ6eK&Jp|4KRVpdHq%1P
z2Ou@o>ENi&iU~^9;=GN}seBo$6#l*PB6Wg0?t|i~A7b}U>TI`OoKOCY=kWXvr6Tbq
z6++I)!$kLuV0i;a_b^m9t+GYH(=WMW(d6ZQeS5atug$_=bYJz1=U2Gbg}-xpoV;n&
zPj8HzN<kzrw#l=hmzEO+wB(CKs)k~VX_w#i*iLK7IlLX7RC*!SFV_<miFt0^M@2&)
zx>-K2mNIVr!GplV|AoDA*#^O}+o@q#H7p*pimm2$*@LsITR!kOKyM6_%&7iBMJ9=2
z5DjJ^yy`P9Mm^LPxNr5{6=55ZwhAVt{!$`9a`YqUz~_|Aqz9UxCYDtl)U6mL)UPcb
zD9inOKTw=p$BYn!+V|}@UA&#|I_S`x-Q!aNj)CJmBdy%|)=$$Xj$y7jnP#)LVF789
zuqb1rdr7F`8$78`BHYF^=kiiL2<~-F)USFkS6=F2Jwce<Bt#34%cIT9XF%13oP%5y
zW8?XrR(It8^8+}|jrMUy2N%LRa$pH2`RVH8+st!RamV*1C=VfeU?zgkcLVq&wGFQ`
zty0mMb!6t&@n)e&A$mz$-c$d>a=iYtgtde6a*SuB8w>W0^#jP%P1Bjr)YOFjt2@$X
z%Z<%BeBdH`X&sPf<GG}IAhZ;@?`Nsb7kT0nd3oaUtH>ub<U6+C)J@B#liMGcW#ZV*
z=aakp92~v1+zMJ%M48$4PvIu(uHBbb6MGa>a?*cv;e%C#SHOc669;0?)$Ki&?cdUW
z^rG7w@kjib7a}hq=CWtt>Lvc{ycwmyt`}rO6GZbr;pVhNPG{==X#UH%QZoKam)`W2
zBck6yq}}`j*(82(^?N1`&5cgsl8WsQPY<$|p^Eb~D{{{SemObGv9x64`uVGd48UV<
zkO|XHd5nNR01!-bE6o4>{(UW~+|K>W5kjd<lXGsc07Db)Soif!so!F%U%bjEhU?L!
zd$>5?WSxIR@Wo|d-FSG)q-<6Djnr4r+6E(Hx3t*7ZrE!+Rjcp`6Us_TDXku0UK0L@
zkRd%cKL+c@g6-)fwjO0NublUzp+Br?zkoGSO(y4D5^uwFRc_r><%cU(A(P`WV(Puw
zh`*;a7X&1oD=4^8YTZC<>h1p-t)1nyj2Ww(!stbz2W=%1YsSkXB=jJx!wV>qePy3P
z?4Xnw(W!4Z`KE*TulaVlARee2hnRP2GDV{69oRW^zLv_mc_xKN1?+WOWO%@%oX)Aq
z_Pj~{eG=w8J8b1e>E<=1S7ExzcNJIol*%&jm8nKUf#XHml6P~$Rwvm_l6E+h3v(Hc
zd<6k*+C-XQO+Y>d0GDz9SKbE-V4T?feuH48yJZJG%E*12p5>oP8KT5_ui829UHW0%
z3JMKLho{tA<fnT62xL0K7t2{WFJCbS!E0YGbX=zRE~cMg^YKpy`u5%o@}&L8<9geg
zE+fBf^=Wnl;Zy|cYHY8cndR(=;y2I`&$TgBSv#prtfNyI3JrGK^lu<XhNRML<HO%P
z6q=2he`WRpCl5KIn)|%n0X7oH+;G7*kbjpm?+Pe1XNgTQFbzv17*2Je<2XLoQjIkj
zv8aFwe<-*!TT<kw6D70eQ25>FQ1O!wX?1ivZ8og`2B%nYTbaBqPJ=C+p}1UUR3#TD
z*YHIgNqL*ACam$r2e&XxW_Y3cSC`(+8ajgEI=jU)p$QrBT0v(#O@$XP#)JxC<~GlU
zou^|roW9O^OaMd<jk|9KC;GCcQmv-TnSStJI4FwDMQEx&8RJQ2RHae!8d-DGTV6MV
zT;58186Sz@n&JeDU9JG{f3GZ?ui~Mrf}tAr)kU_35pPvLCf>1l)vU4<!6*J~^RJN-
z-lAUYwDi&IU%jVyU8Nf_$BpezVA7EG-hpwf)om(HY}PonP(Dhf^DHPL^#ClNC$PVy
z3>nd@Y<Pf|HiyksVj=v44;QbgpLQvAH6jS@G#`X1WsjfsnrH75QS~`_Lx;;?qdOSX
zfdtlsW^^p`XAs4SH`~iY*l>SX{O89Do^x(XTmekt-WR_9+t>Bs|H2({o>>_gdE{Wa
zD0WBxMLSZ}Bq|_CPuRY?U4hWuhj!W_`M%)e(7gB7!=rP6X8Z5og8R<qZZkqF;t%Wh
zX^e%+X@4`)TWKW4-Z85^X-r?@BHnX4dg;Qh(8%1t)lsVDqHd9C`W+dg6<5Lke^K_9
zL2-56x^@#nkRZW=OK?wc_aF_y-4hybT!UMH;10pv-Ccq-?(XguJZR3!^X&KR@2fhe
z>g<2b>Q!C+qnT^YF|PX>Qwoc%d6B_5QmGX~qte@Zbw(C@-BZglgo?$~=&y9PMbica
z!F`7edLtBt+)u8+|0lapU$(Bk_Glz#9iC1SekCI&^2OHtY2gWtpP+?C$c9x`nDIw@
zh_^x<_jR80)vKcXCIjWGIaUio$BAVDR*oPG2JnyRyMWLwVmp{8eS?xHOQQK}U1WW3
z+8x6^YeWxQo&z;^bStp_6MEoepVUL8_|I6Xji90EwS#nqw$YI4#jFtgnUF8O%kp3E
zd=JuF$=uStc1F{<+PZTBC;6G#IF~amMX(=A;g-FD1mRw=4TSrq8ZU+`d{v7}TAI;E
zUBHKj%JDXfKVdt&;p8qgyve;GSu5gzB$UfNR+;*T`l`^zdK919Y9D;dOu~jw!Rn3z
z`TdNZth_l+T}|pWW3oq2?PtBN^RR)Tt6p_mj6sxB5(jNcijO`5Or@-JcX1&RufwEY
z7C`UQukDtV`NFv@ebAtX0*BmW*ApFSmJRF;&Hd7SI4PbM1C&uG_#_O>EroA(MO4f|
zk>Z-k1woss1k83gX)-q)$&I7xo2$~qr6H2&?25C-x@hn~EP!2u;@uZ=r`*o@sn5*#
zg0DZr>)5gN1t?`GXp5+|G!EU_p{K^4DT>=r#gM9+76}%<@UBL}7-4m>UVt%}8B(75
zdBt(PYyk=LHQrA?h7|0QKg_^sn{;mbpEA5Snm3&$je|qz!5lj|EM>&H__@xj6n{uh
zPKTY$-V1QsKwM`b6P_a{`QOfj@OSn0D7H;tl^gB!$gT~N>6t+IFPd7p{UNY%#O!e<
zPjDbQg1&RMi&^t8{$!px<A|PHr1B1zTK7oco#l{YEGbIVLZ8faiI+o!cb+?egyfbo
z<IpwVfkAYRBx|o?YfCb-{30CI0P4c_FV&E1JN)1cGrnNU>q{Nt7{4MG$)R!iyd>2f
z|IbB7>is<H>I`;8xY`MY>#4d%2gBu4vY^q|er~4G3Jt9x5~I$Q0)hCPQz%ubs<chL
zDyVl&>%4pDB4K;!7LS??+%DJUGBNV%fgQ62o^(8YtCVR%UhENV47JfgqY_Z$#zTu7
zPZV~lY@>sn3yd-WEbLcxMH)#&N(IYhR7sZC#TpIBU0fTn)6V7Z?K)_jinaCFQJxQ1
zjN)Xu`e9%b62YTaKQ9e0Uf{RpUGQH(VPi}=_yse`uoV9q7CqUEzKk;z|KZ_pdsP+b
zB7=C!$CwU}RBBgo;Tb2AhzF|mMo#$4U7q1eKknxwXtM(3`prs=NE07u$P>^!i2yQu
zeG4GN8FUr~I3_+$wR8{r;ND3IWLRVD{ImFx8Jn8Af|8MqjReu9g!4W%>(zKJ7zs;~
zZG7@wTX^H`Wiad=Kzs9vY%TGUY(wPrQ#9i3&ST$irT@H8g!|8{|LG;f@6VH!r2r~U
zh$XgSt6<Qi{|<<*14k{Yyh1uwU>k*p^MaXnxiFWYh#r%tbHpKeyG^8kq?h@-qx8UM
z&!hS;;X+e~e?XJLr#zN0YOr*I6uYZk!WBWey2Bpt))|)Oz3QjNcd<gfNNEZOD)R>#
z+?FpbL<BXb^%s-r>W?G3!M}dYqc+htl~;WF^;G-12XY7{P_;ZMkYFA;a>vghrh6<l
z@l_b|y;FBFnU&B9=V6Ooo2s*4-2Fjh-@GRPubx+wx?XCZ-`$^>1z&(`f~2OEXiFX;
zT6X4W$N8nfwJwZK?^orXV6nf603R!G;AEpO$YRPW5>PH0^aHP7l_Fb>r=ngsX#boa
zIF%s$v5hDkK={qej9SpelVNH)p*-;5KDR!GeDtah&@0KD8n*^+DK=j$zs0I#Tz1gu
zjhbs7U{4m4WK^M`;mu(PjS;yYM<6QGo?Hv2MrcE%sbOiAdIft^5*vZe3a*;pJFO$I
zh66MsDuvcN2=K`wvIL8FC9L_o{RanN-x{0uaYg>)Y8CZ&wl&^E;+;3SZwXo@knM%6
zntDZ&gBD`<`hF~*kQ9=HY@ULtm^9rtYc0=0-s20kZu)ADM=HN_Ri|9uO|@26XjV;i
zq)6*DY&AV2SMZvS)rrW2%aFfHc@Z6;)lZ9dbnqz#GmLu}?xYuFWMIBg{VVmsrL;g&
zJn$u9<{QO-NagHAp*WD^(LYe7OIaGR$KMrY@b8KOh5v}7_*b|jE`n1LIp`JN3k|vr
ztSx9_V?z*ho9$wn&$!CHpTgNH6Y`x-<{IeLtPLAstB{df&@H>-KLRn8ijuh#JVGaG
zJ;xyMtjoQxwho~klqn9WDk~&*lgxw?T_F{$NNQ_lO5<BJZF5M|{sj}Kc52q|e>72t
zagP&yixX#pHnZ{V)|5GcEq8Q0<xtqk62v}~GVXnd-%HfE!S%v>qOnnrvx{rWIe&i$
z8-*l^;*(z>aS|=T+zKn~Rpay;ns$FSly{3Pc7j+nvOeQrhlM1L8D~{1P5-E_zL4V7
zm(+JN$)?YL42PH+zECV@NLwzJpMgOowCXdii)uCK#Y`_$PjQ0^^}4ERV~lk&$!|{0
zVf0{_ETkM$Zr`;dl=FK9ttM068YcWO4@@D?S{s8=7AVv`G48O-6ql)QUhOMzbF2iP
z?kv6eGD>u>sx%OP+v{3rIW%pwDd4nQ8{e@te>?y00z);5C910H2X)iIS6#Nk%^|;9
zTOc|;>bPEAyAyQJuge=M*0rKNF0pAX`JNUI+oK)wz_|8g<KJz@<P@T@IVio*sTlw@
zm{Q(;w(X$9Cr0Nf*C%^z!A0nt2+h;)L;jYdD~%BcMqqmZ%hLU#>9Hknoi7TNo8bKF
zB+DNfdgnEU^<`bBFIuZ|fw>Lru}xaiKx|Td&5WU6Q`Z8OjjWaZI>MM&8wm$a(zRYj
zmkG-}=|YWIhtK_Wn&$cWHoQ=O$Y~GU#68qXH?YqY-ui}XJA!X^=0EHqqMz4P$Nz4W
zq5*aUmF33BJX{3y!j6!b7lWRI5Y$Nmgr8l6+jXA?xDd-!5hx8L8WuAjG~_<~0g2Y#
zcU<DWx8JIHryX0F8K^RoVVr}$=c~Ja2M3wDn{{wY7OBh9B6?V^+<;Cydb03P8bmpk
zGRW!xWKG`4yx~%q+Wce{-w?`=f>Liz(5Lh8_(*SyYI<}YL5w`4#HwB>7mWGfQBR5l
zniO$%o7pajPBg9Ykz}Gz#X<5{tW6`@psO;PL)Si!13DZrhiMznj~cAw=9W0Bq7$_(
zXG@-K^vu~K2-o+8z>TM^#WE@1H_Dkc-V7x%kX1Tx`{}E%y;{?!q-f5+b8u}0Ln@CQ
zbb1nh)tG|MQJOFfK;=BEd}L+TcfNt5{Q`cBs>7PIJV#o|S6<HPJ8lIdz1)=vdqj<u
z-DFw$AIR5djlb8XnZb4C_7S_}c6x}sg}IjTn>Cvzi{RNGThbjU2J3uAdphYER|`(x
zRUzn8nOmByN#KeR)4UylXOXD$QY8yF@D;lNs+q{f^Gg{qT=$hFGTEXJ<A~!9B7ykE
zx_91e87iL|TFeGCh~X!FSdcavKV@~_n<hh+0u0*a`M07)2<BKX6JWw=#`+TeeZytr
zq~@9IRFN;&A&WWrF919bNImP<)&f;xM7PC>pkblf^iF4WVMq0V&t`rhuK*-f+U!MO
zrfq&>n&tDa*|=K@N#$g><ay~T!+(0Sh#dDAMTSVBT(0FgVpZM9Mr5Lz>KROO=90$?
zjaHZ9Z=QcuC{9A?4cq-gmgtL%HnF^+3ay7Sw@lJrD(Rk5GLjPM!l`zLj7OBvi<$_#
z-H%+cU<BBtLi~)R(0)lnx@3t>wO)BiXaBZ3ek+=Q5wqaE@KIJcUv>BBepddX-ofOL
zo+IjlUnNFtPmC&~>f3L;0_$^v(g-mFzy^sXE#95TjnjpD2+(IEna+9zTw|)uCRepN
z%IV1w{jJAklnRa{i}*43csa)T5~{{CwOj_`D0{ajMhOi2C#42P8*y6oj)+kL<HpYv
z`dKY=KYr)*a`6R53O0*7k?bX-T9<OMxh(TN;|}~V^h^^rYNx#;QHgs;;$pU4+HX!V
zrq0W`&rC<?@!5^=&O=H(Ia2A8$IP=TxrDe^)ql>)AxAkdkCq;g$N4-~vL1(yZK&O7
z3ChBs2{#twmXU;6SmUS<4_|dGFtLkQe6uPwY4~}yakLL(W5l(RA@BA>28HbFV5@9G
zwnBE*R!A<IuLgZ2gpcPkE^q&UBBscNMu3xvu=np1GLu#yl+f|$?1NFG?>)Bm%rCW2
zA-s3?2Skzy|DXFOZlGg#4?y(M4D`Uc_xJ?B*R`I>m(N}RHn<4-KYW&e@J)h-9Q!9N
z1NYE!9sIG=qn>Ph?LqfM%;}BE%q)xdo0*u#LOUO(!~C|5p#){zL#ZxV{K$uvi(Qmq
z>feBm-Lbm%-lfMQNkW-UD~d8YQiusnN2IB=HeO!R$SlokDwV(NY_^s4=9AdI)H?=&
zSam0t&+>9Db+l`ciB*XxHXRh;cPqpqbkwtz6M}(nY}cznCXd!@F+sPj$}U!(i<vrC
zSUva#EU3MFsV`KUQvZFZwMIO4hf(MQGC{AsOEOx@C$04`)<YZzJt#7`SE5%FrnYx(
zpv+xa^&q^k5e}1gIi<b0DVP5P>e4trBz20G60>(9`SEy^P_J3QxG@u+rPu3qp1OQ<
zOdG?slCEiea>z(S@Y02;VxnF9BD>B!jY;EMX!JaK9dx*-kaf&KAqzFD5aRX$HcJ*S
zx(SkO^31q9{sS6q2<&M~l67`SuPexhVlTYOy#rd}wm0|c--i#GZ4BE(J(#amrd46s
z6u1Bm!-g*OAorDp1UCnt>T|Q7uy)CfE2|$6_y^cdLXWVj3={7&2>|oNdo=Hw>1e}y
zdNerCiT2sB*edzY<yHBSe=g^)|0I0tjMm@<8GCXxI4=ujq`t^-6`3DfBimBxhbg_)
zY9Xq3^90yU!rA9um86KV_axlt)2Z4ld_Nin2bZu1)c|`<=rgA_&v$`ZgWGFwSh7F1
z@sFIXQiEF{P#*rN3jE0YR&d1|79mM7{6RL1K*=ixWxO^-Of2cmlcrMt!23S4-s%c`
z%BnANjnxqncQd#ht@uj)UAwxoDogE3IY!sZ!mv`<v8rnx8_!Cs5%AWAo%x=4x5QD@
zR&sx14MUy-KH^i=&z$Ih;kjyOpe0v%w)_6~41MW}=9g>$iUq!&ZWxQyjp_51!R@&>
zEV4%U_TTFqclc-tZ^PA6%u%HgG)XfQ$|RI$(tPN9CV_+(e490}Tm1@D(dXwQ6Dk1o
zQqkN_q=ynG4p=h@79&?UyN?70v{jaa3H(ygEUC1s5KokzEJicFFn7cI;myqk4@h!D
zzzs%@>TYfyMWbG<B-H1;;5iy85;AdC31T3y5I_9z!mSbaxSkcGGCNX+l2>dp;vfxY
zW{H3UH8nlGPARo#0fx7f=NX8NRmFjGfBoFTo}SA~{pD0?3JtcAx<x*tggOy;p=(MW
zRz;TQ40Cqo-H`+9q9(??A48%dZFz84*6#ffR~-;+=jWenpLX4G^<5DvcJ<8%L>yw8
z+%IH}<}k>jf2AJ4%mK$sDX^JB)u=pd%fI`*h<XS&E7vmDbi}!WrspN;Zq@aO1qmaK
z<O=<cBH3(6bvQoXnDJd&+{TdW5M;zds$%7E@@3oxn@k~Y%2V>GVzW_$#EW+Ib_%E6
z!!hSCT!%8p4b*YY)uO~%UQ?;@BmvO=-{BgSO1f%#<C|q!0%H%6hIL}R=yQnWJf+!O
zD(#lpWNs_p|A2B}Qbq<l>d%<WBRq3Vg1yGgCx#dhjgNzcaq2KBpwQJvU+(>Hjxssp
z#<w~40XZLR`v+u6?V~QY_d?)&cZ-J-2*XU)12%+_HsUVJv4xkRuOxrcxWN6BhT*h+
zrktu>VOC!$x6U_$MpK28gW}L)ybgXk`1X;wzZJESYd?S4T$bwcwo{AEM2i$FyalZh
zgxvX(?qzgpdYeVr%-DCUQ(CV2sSz5ZRU?~~;=9au5-1MeFTU=PpT6Amp?Yu$4n&<D
z#7x{}6>aPb_`Z#{ypt??jJFg&Vf`q_JX6a(dw?TkSX!=_fP^{fFh%ukl-;k#PK3ns
zZx<H<=(lJ5fBWs-|5DUFZr25>IJkJ{ib(0gEz%Z}oPp*gZ_tuOWDqTCfSR~yJ+LRk
z!6ix3v-tE_+t0gVsBjg-KCbQ=XR`auGuO0R<DT4kLqfXLUmO0$`Kqjuc^iZ!G9ti*
zo=<biPg$L3HYu~y4*@+c&*Um1g5dOL$q-~h)HOWESlVeEy!3)0oXjqTO<Ji~@#R5f
zh&bgrjAhcZN;Ttw5~!@cS{lF5zA-`lr)TI}E;XT6!;GPJWcNd{_`*(ha~H)LI>w<0
zNp>@0Kn&wZON+5ym3k(J<x-}m`MWHAv^At(mkyz3p{asm=FQR?nBb)q&l9w_PcfXY
zSm~%~U)kQa<^4N0?b3o6yktKb)5NLO)@if;@`yVbXAs<{x$jo`ze!8@p8>sq_;0OL
zNqUCi0js?j0`8t2O&8o5vuE-M4b4-av2)jH4NJxr-Gm~^T{+k+#_VkKSA}6uMLfEc
zDt*C^;V>K}_%%t1N;ft&l>uvT%xR2e?6W~`OfzB%CbTAsK;CwDzgH`~`i)G+UT%n5
zs(=u|b1;=Gld5^jC&x!jQsYo*br-}u>Wwa}9cO;@@OUPq|H3`O^*}tg8I4_B&IdnA
zI#Y7^)jF@fE7fdK&0T`@Ps%LC%&iLRYZgA&h#gJKtwgOTg&s(B?GsOV^+gxmiA<+%
zU+2A3q2m_w2>D7R{|u${wM5usH*<{6yYHh68F6bC-vZPSq>{P(aIOX1Hj7!o1Z@Kg
zzBT5CU9kzmbe+5#o{fbYt<h3v5M5Q(`fxP7lO3%OQe@#~OR>?l)5zKP@+xNzS4E$G
zj-T+72GzY_h|=KqG8Pmq)KaBi{-#oi^SHqLBaVOTwDv%tq%pL2j*RUV$7dkL)`Z++
zXxc5Lz>#|=_96=-OQB*mMU}$&4=9oju!`k)Qy0?rzeo!aAwQDOe5sOv+mVCHHkr`r
z%q8qxKFc(LAQT0Q1^@lMCl%SE+1~$r+A)7YWeUrSgy-~{Q%F|kddGWRMm()HFTtJ7
zM?xzhZ5n>6YeXD?Ycs6rrKtj6XRlf`>ZDT8uV3x!cNeo~cU>@i$3ueP;KVH}now1n
z7~Ks2P}AJzBH4<i-RD1zV&$G6=Tg;y38uWO(&DavmOYJ(zbgCnBk4<VaWvq)<S0U*
zTS(e8GP#Q?y2iEUxh;h4!q2uBU?goENi+!o<hoaH{-;7jZK(8j*CV~7RX5ICjA%=Z
z7=;hNM2as8hUL$Or|k!4oAhBfVBAyci)owKOCH<|nk*{<lLY18B#7cF+HJ)AZRSLh
zsgQHx)r!qms%9s!<r}gs-JWF6ne$6)YL5*4uN4Ws5Yyna)zEB;+|(-_lzR092BRkG
zrAipPlCItgk~Y+=M;ssP!@*U#TFG?p0wX26bOhLmo+>G6+nAF1NGDOG^f;&zH{9mC
zYEP?u6gjSrm(>uQ|J0FUJ$-fC`bZ_^T%Og^-k^wZW5HE%VwcSFk}r7CvxM!J*YD`@
zshlQUX}Ypu44J2xp-@i`7@RlXnwj7zX?xh0ufwuhIx%iBK#N#;UQLKbkW5jcOS>y)
z{Ij>VOG=UVSgXO`990tB?xDfu8Qva{de0e|?$}#>KlYEh@V{?~<W>f@s$V;V|7s4c
z^8kAP+{+s<8m`Dbdu~1gdMiL)*qr$I(ED$8%jx~yZ786s`zt#9KYlbR<%0|X5fHjM
z{3moJdZ7_4bt_>kIQ0GcNWPGcfx^oIl+J!JZehs^%IhU3QD+g}f-M2nv|&Y6XJg|7
z1)G*n{_)efuQtU`p^#1>MrNISDwfS*)tc@wTig95ouYukRD#--%55j5Z>LZfLX4^Y
zs~aEpDt+YrMhTUv_Q1+DO<07i-Bppe=>fQqs?-+DO>>V`w3$jGrHnv0s$qgYp)Niv
zA&AD4-1t+H%kvX8%?b>Ai4)a_6dzu${%Ox7!57zx>q8yC5VDan2Kj{?q>tyW?9r!y
zjFEWn;6%>r<YCAmKh?GED9$QdGDa0{rYbcy9Xi(PkPA!x&2?e(Aa)nVwPnuv^M@o$
zqR^}$5_m{l7W^wAizXw39#a!Dh=)ppmbo;6T^6k?E}2afv8v)lYFvuswfP{XT#}a^
z{<4?VvT8dy**Rakps>fn<57{d!&59jPhLOeYT>Yqhv{wTH4Zt`s-E<-5vmJ$RCN0C
zjEH$#CtjIm7#)EiATdI2p!J=tWu6*(C&lEJA@j+2#0O&HLMAgA>B!Z=`g#q=r^t9u
zZv`Q^aZLM!;P!wS9q#ih&19I3iW^kYu3JWIRQK&(NI>l*K?uy&jnD#Gn2MC55Jy}L
zf2K52ApdF;T_|36KV4nb9X#VgPJTZrTBy5NPnNOE5ZlB|sAO@tB+F{xp|+i7th6(j
z`#@?DjfR-PWV&ZHbodIXgjM%cm+Xi4SEqam<OMk1*JGd=`i-wY6W~)l*BL|#FM||U
z&_1qn>2r3TiIIs5scu(a<<s>G)rnEr{Jw~MZsxww_Dl7lU&w_^)&}8tY&N6e+cTJ4
z!B^Hi)Y(^=_{wvj66J9LY$ZISiZFez{iL}D-Yr>+N`s)CYHoJ<V1cAlZWcmO8)T!d
z{#%RBFyC%@S*gAOmL+99tFB53eVc;Lp)7h#N*$~l2@kpUbhn_c`el%p9NHoxdjqTC
z#w<0=PHS*eqsw9Bqcuj=gn5D_k-7;PD?u3=t?~m^x^DP4TK@wboh(I<@;#S)11S!X
z68hT_gJ^-B6D`es>m6RJlWg87KTn1>Sl-Lm0i8EET5U*H9|s#dT3ZH}D&jI$DU+o)
zH3Lo4=C@{>>1T>$4zVnDHLR9Hwp0RfyW&Z=XPIxBV#GSOp*crigK+V<KgY{j8jrYL
z1V;ae5)O`(+29;|BR}_2oaRM3>uSwyF?i7WcrSHz1G?#zZk3s!$hQ~L_)Oa#uaJoI
zA<S+$goL`q=)na_r4*=D!P2eE8{HfOEeXehDj^@US<q&&S{)9jy#8cCpwvlIrq6N|
zjPF9O!Ld+^X`uGW625Xo(y%v#Or@9VXPM*(*{Q~A1Jjq6?+fB;x1VN_Ev$54c-C$R
zxY30)1G}P+%+XiHSxsqrA&&8rhHfs$q5sL)we@MXrVk#B#84j%5|0r{8XSiZ`*Qy!
zvk&e1+z$R8y-jXnu=u&&EPLj&`c<D_K4$Jk@Rm!7o~Iq(e4nda96iPK7pIxwef?t&
z;T9ZYn}i6TszN?JDWJr8GAMZy{rqgh@Sa6^2`QW;esm%bE_G{$^-PJSrKFp4`Ic+k
z+cfv0Ge+hk9V^WYbvG*vZh(###y9>F>+7nqN`o}@Ejf~PyfkX*8Y?-1O(Gn4V9ffd
z?61S_-wk6&?cRUK?z$oW#O`|j;#;_{0m37lCGOEA79Rl+@mZ0hs)hPK``i3Ijsu%T
z<+OJ_l1YeoZap<yEtvXQ8!fM81yOJp_b)&vjTguSRC6^kKY8VCvL8o_>9v(jTKPh{
zCbHSSEnP%kpXoT?>RXh%ZRNSer&tYG&9A3dPdDNB(>AneG&g^UdJ%ijYjR+pyQYFD
zsLe1|xST{FR%(HcNI;WY@1B%0B;4}SsGfJkhRjX9;SoMT>DF4O{=1*7bS?y*(*}~+
zA4{HYPydjC%#zvwnN-}$!as>HM-oJk4IE_z{2^?cO$JOMi62};M0w6e9Plh>C%3Kb
z(sLi{65liIcJv?UUsfUXv|Dv?_!9-GZCw>Bn|0}#Ox}Dh!;u!z->K=t?Mr&T__njS
zmb4N$D?^iar7_2fO&qv_W1Hm)2rj%ywHo5BoEKOeaDqsrx4q&ouJrOFWid%D6#CbV
zR9Yx!OOJZ~fUFF>=S-iS4w%OnkK^cy*-;$5kMIuZMEH)Lg^U4F1c2N{q%0%CUY(_c
zq)ESiJit+Sfh8N=6#-}s+%ww1zSeJzEej8(r_mfZhtsufleVXq%}FE^AvZdoRg&<X
zdacdxC1!f~!HZ=BBlhyn(<cG^`&K$%o$+(-evKR{gEM;$#VT+I54jv4IHI*4g+dxJ
z_kI=Qk&;9`&2$Pf%*gLrrwuoGu+o{ZwX=^KR@ITA!wP%B%x8LvW#fBkjKpQVNexk9
zggOw?vmfz7nVZc@!@62~k<r0;)4wcw@IrL>Wg4=I-^GsAn*|d;VQ{85L|Ov~4mBTP
zH0H|3vWIzlrRGQ(KZ3B&zPCXX-!#OCQVEdN7BP56M7r_&DmXvJNVA)g3hx0M!s6e3
z+V*gCx$uE>V;rFEhneyH?r0@JpQF;}6W!8qi98ygH|g|rpq!V#=-e3Nkz{h?C|d|`
zsu0$A@o$SAES!INRMHs=GA7&a3a!2VR*;~)n^0q{bxoVVmUrWSpayA_Te8=khM~_1
zL2<u3QJ=Bf$Tln=o+4f09JTy7`C?)S(hfSKif}wB8j>p-a6Em&&YqK6Z+Y7mJL=RU
z^YylD8As_&Y%D*0cGm~^#iobCetR@3>;CNKmRidt=}tH2606N8`4u~g7lNY2)kcF>
zuQ7s+nszroCml09@{PoJ3@<HBPwgB_3EomRopXrV23?efhMQ>B1X6s5b?(^*9+amt
zxo<-j%fPtpDHv&mgDpW6A1qq^7WGnHz6*Suvo!o>>7R+>Hpfb7W3kFw&1=l^MU^c|
zLtth|B7W_4?4emg7x#(b0|$9VD^x`xIwkCKwTb4`3deGOag+TW{y?GEuvSphhyyq!
zS9RtKg)U+8>3DvcrN~jp+b;KDI}?2la$5|}4v2X`C)vfYH-pWFp-!SlDZ@i%si2ab
zqr+NAGZ`jsVzY5Z%N56@En^Y!UV+zu)lPRs{cEe_<#i_kK~J2TkL9NcbJC;sOC*WH
z(E#=i^gM)C5Ba~|h@113JA`2aJ3FA7GFAHrbUVU^w0Ov!Mg!&jRON>oa*=u6x6Pn%
zLU!+*@q7p9w4|QX0MS9a+4IC7P=Wa$&{;A_xYqk%(feUH@_%+0#+#=+s!PWa&YSAp
zqZ}tP>aStmj_N>TBL-epQBhT1$O4p1Dj^&4ikX=abgX&{|MY6@aBW!GSjt3+7N=}b
zQ4}^>YJ)0!8}<W{^!hkaeO;%RRq29`r8A}nJ^soS^}T$_)suXoq<mcR1(K!u=J2Op
zaaf*MF^KQ}I&bQf26Q*IWBB|gmzYeRPos>dzBH}e^)gZ=yOu(2^ebhZkhGu2{J|*H
z-y}CZ-$9Qcc~Vzkh=Ptk#Tp;IpiOMZW5OEE&*#+PVmmgnUqg=t3T@xx*sO4@eLLr&
zJV>f;EQL~F4pZPQv+OVC=DAOgFU~A;B%K^nmQIcKZ7F~4Tw=k##yOxoQn<P{yQTb~
zj(?QSm^g#opIWbz{vpR-^I~bW9d@oe+JpM?7yU=f=GaSg2Q?@6n~BYUS$J0F{*${o
zQnS0tN$Aj(pa2DUsVJ-9XxxJJ1oOvygeY%0`zEoGxmgIDNh1#lzs8XA?};@nuS2Os
zU7hAnZ$HkjL|Vk^D>|iYjNgqt9!)8z^cx_jRX!+_G7Qhh?f(JIzMoe7_WxfNk>40i
z_Xk8EZYdAGFBoi*Qqizn2az6oYm;^=xV;deRA^nXQ?Hd1tUsk6D=32~>lV(Y>Lx88
z5L(7mb95q*loW?6e|Qn5k;Z#z7Ac=x#j>&{Q-$iv`PzQ9XAI5|BrgB(R%)czdAxJF
z`1DE&&C-fbW=LIv^BaN)o#rjo3{e`&pu9FnAz#kj@!}p6D&LnVV64SF>Z0@OI{xU=
z_ze=5A6uK?r6$_?3K{xEX<ry9(l?{1{H$0hW!GXgP=D4=x1niR|ErX8z}St$Q<3{S
zA^;UkYl%7RI+ca?FY<2j7}Mst##h$llq@w(5ENnStitA3y4`IpWWVuybbkEq9#J<u
zA@LQz<V_NTTN!m1s_S>8J|*ew)+D+xzNt6<vqi`;JiV!q=VV~(S&=?H|BWNgN1K2A
z<##WOeMd-tVS1<2thaqRO8>NeT(}6b1c>)d(@nm<E_Lq52!%XjUMS>{O@M=LDuJth
z2$Iy;*wk*>!Un^TkB|2D81FX>gAkj}#=Uc9>Z;R_Q%_u68hA6Gbh^!eTt4@4;)c##
znzo2jZ99jEaxXJA&lw8Ti&|Vk&bQ_8n$x5=R{r084~~?A0-knpO%j`>Uety~f8@rp
zMQ|qzyt=NO+=|41=XyUgW2`8VvB_JEm*538mVh{97CfcV8t0P`YBgRAdErc9%S8$M
zmDBKkr2jqMj@V9iG492zc7jPd5&_I;V%y-2F(Ok@V=1j&;Z7OqH6iQFFJ{|y{@8eN
zIP89gb^MBiW^HH;?uR`zX)Xq0wWm9LLNDM4wboa!Xqpp8JyI;v+RyI5PUW;1oH#XB
z6P@0yv+j!vr>%Z_5fJZiqTOtSVc$vdE!Zv9urjqgG-SHLleG8eBcrTn?XcxAFASQl
zLs&LibDUx+$+wwA63R~gS~tdIKc`ePqduUIhxS0Dywm||>7SkfqzYuUV{|ncbRu}a
zXjnmzvDO_bUwET<;!_vlgdrv%-b#Ev!oL(fORIDAqU7s94yP@!Q!|Nu(SkV)%pD5-
z-$N#D)n<(*cU8U`e^{=Of^NsX?8greZ?wfA<Cbp%MAyJ-YiYi_E;a-Zd_iapQRsk2
z(ruPlpJR$}UiTBN9~lF^EEt<bpmUGdH}$1J*j~MPl%^0vr6rR3CKy>6dyxE?3D5hX
ziD!h(07nA7&H+N*67MI9`E~SWSxM7aWS!)v=$JAcobh>JBxV$tzXbM5Ey_Q}E{t^j
znV3vGe92XL6KVJRX@mrN0e(Lvg%8z#!RP{bH$aRfl7#=cz2heBtL4ATZzQowfT$RD
zNT3u!Ci421`uvxQtoHQcfc3CZEi{nqOT*v7WeNj7n?I&J;!dt;9~Av2Ij7aA<P~bc
zWLp6^<3UBDuG!#6f5@IGB*O0ZZhJ-$O^3T&doUpGbUj$t;EN+u)-g}#cl{JzzsdGm
z(si|<V?!MXU$99+$c4ThGcM*A&`ihIJ$GOS)?!slj)b9vyog|!{?3PTS^RwJ>^FZ|
zP*NLKUoXv~tSD`4wQx*-v8q)ye*xc7&RG4;tXJ>s$?Fxag<QE>m4r#X*u_Uo`Q^Y^
zCV9GGw98Yg3-sY~RgA9eJk={5=MPFY>d-}P91zymljQ17x^lu8b-RQ=py(-iZu34v
z%8}29MZrzRC~p2EQjB0Qk>tsOBLRfi%Ayx1HumR%N`Ev53$4gO1RI|QsVOJ1Kpfxv
zb!F5-+o+S{e870ja=g;O6>gu$$5w3JpXLMupIRVQg2FO)Aq<!NtnU+I^+uIkhEwdx
z5Cg2UOTI=BTilLC-dU~%ajxJM?RdtLjAD?x8|FZgmj{C7BcBYPB#@%#49+j#KkVI2
zy-is*$1YB_Wq>+TkMX=7+fuSi<~o0A*+5fkNb$jys1HD9ycMQ;yP2x(;YJ0a)srV%
zA}EFH8VIiS$1aND@gwu2d}?u2(J*Mj;=Z$ixEMd!9^NNov_hvVcudy?2<Nh$YB{KD
z-w5cc8;_A{^Sc05Op@Puj7RqxbuKh_kt9;_)L6XOyLTEwS7Yx%X*v14Wz=*q{loc1
zR%^UihB334r)x6p;r2@PKOi-H5gvSkXU7fY!3*uGg@&B@{<i9L{7{b0hA$Z{XCYg4
z9q1Cp<y0~6>?WuP#<mU_-D*8^d8uys_Nnoq)0g28xkQ<1Y9Q!PYkF3&Vq+(pzg#^~
z*KQ<$%XwKxzCnvOK9Nwd^~PtPoj_2u`~8+;X|~F3@K%Rgzj-)oK=9PLfknPO5zaW<
z-;yWKz(HF_%#1}5k_<g?C_@*&)Q|$&PH3NQ;LCy9fFohgv=b*V7(z{!7)5S@a3c3`
zT6L>s-0^IiWoD;@?F3o3>3+9L`*5tXr&lIV08KEbY3xv=UHM1mk!FU8dG5nfE=2E5
zc>EX!Z!be#TbvMnT2RncUTx?wFz{nedoFRv27SvXC$n6u!<?UTBy>lLne%uKj|}?b
z`2-yoLtHCgvjKbC;rYCLpBfiRJlW#HZ}akip0i6AjQ%35f~Ji{t_Yp=evT8|I|QkY
zBZf4ICW$_Dlu4(E?fAeiPdi9em1Wr1^?m0U5qQ2=V-gOKmor-N5cfrXJDb-w&})w3
zv1SBwns)p&lCnv?Wja_N=G_^tl+ovw<6lRidylQdB*2%07o5UYUU&2`oRXA_yo^Xu
zyrh$oKPcjyUOo&}0Q4p#JgB%HUDX8`88Mvi>g%fYRu$hLYR++P^8eUNg1p%yPGv2P
zn~Gl)s^HBcMu$JV+BO!eOx#_zdtNB_moTGf4sMOdw|v@C<7MgV=@r_|Z`ppO^^L*B
zyVjyC{!!`uebzBN;ox~)sao1ClN7}YxM>rSq=E2|!|`JcWCv}ylN3rA8=^xf!GG#x
z2Mfe#=7DcB)9%GsBu9|4*m9z@`L++f{ZRLg2#S@KvqM@(iMxO?6eV~h!LwxOs&|WZ
zH({v0G@o9(<T6P5fon1@zBFpmQ%hcb7?axrV^1@AP9>sgL=DfJKvf%m(&18#F;lv{
zJMMaHfqs^!ZVwWmRW=~(KR0egnhqey7gT{ptd$8Azs0}|?Bhl^<?)rp9-nHy2&Sbq
z2AG<{x4fdohwqpy(beA9V9i!|DIZMOlbK`d^9JuAP{4hPoODjiIXrolc_IMC`9r5l
zl#tJ5q^_G<64ew{#0D|FPT(5dq8BhT)T>c#Z(b2W+&DAi?hBa+9Gou+&<OC2;pHJH
zqw1&$$HnX_4g=w=V+{I`JmmrT2tn`jsb?H-b8R4)V(<uL1p#IJqI$>W96%WeZ%Ez5
zS3Cke<UWPp04e!@x&N!n2{VIi{y_!7U<#1g_yb}(>Y@C%Y~e-k9hjMoyks$Jm(;py
z1~0iVx>&WV2)C=$?t>$!+30;m1Ob{k!>ZK@!$ZmT7&u3+!8j8%+#zKbF9!vKE0lh_
zFp6)q(0q>$;Hj{A=a)|1C3DbQ@S8Lp_`Y9H=4TUTd9D55$Upjzm}&Q56G@GXXK3FK
zP!lFdaWM$QxNw!_pk#>3Q?w3F#->2;55E$h$4>j1`J)oAUzeslNP*>g<crm$cO24r
zioRXT+z`*q3_tKJnh5DgQW^910~Klw35eBBSIOM;{j8JDjvO0#B3It87v&7KBf6K7
z4DwQH-0t%}9XzFNO5za8^8<+`u#ULeX-hVv+-zlFga{GuOVr8M^}$uUJ48jd7LAN#
z*qfa4)3sf<Mt!I&zV^<3b*+1Fzh>WkV`5nPLdCmT+632CS9v3Quu%N^#OhY>kXbxM
z;{XxNli0b0lp=abxlQAc#;TP6QcqUW?kP4SmX{zO!rN;Z_T#@c<|-c}o5&7;h)cV7
z8YfTwEg{oo?YHf%42uR+{0PBGdxun2)aGg<pNy|pHw6Y63@c)(SR(h61Z`QmXTErp
z`HYp57h)Caf{dfZDe=*3;N0@jWHg~qn>JDCD{>avKZym3uTI^PuYYNh8Qv9p7({Px
zv92!It1MkI!?hrG$i#v8t}5i`bcZihYO1MCRdr)<;08)dMj$qChjt;2WvQyG{{GwA
z@rg>U&SF(n>clhsBMXYgmZqub*TuPc)r{$bJu3Phn;(|GZ*P_st-PTO3!98{b<#<T
zdpFYF`?|!~+9A@dXPS8nvocuWhj*A=hNaqs1)pru9JQ&#_j9h$w^<OXa~`eF%Jass
zjf3k7F0&-!t#^GCt)-sSD5*n5Y%{u(5e9tyTLi=(IHO}qogGinY~Sft;cOSK^pELp
z#=&wSyn>=Rg^pNaLf!a3`=f!Ufec{R{Hv`0eLp9a9d=Tf-YsQdBBWi9U}VOZJPisg
zWto4sy5g>`rWYT5X2cO&dNgLn7-z2)+MzO8#ksVB6{kMHL3fz#yisDWQWlYZ2KiwR
zrn*p2(b2{Q!kt)86H|gm`FU}c9;HTmEGAY%&V|0IzWag*Y_inTwY=|l+as2+zkdml
z;E5$q-F5_~t#o1Tc)%t?3YTrHP|F|+Z4sQB>uh2<OuwQj{7*G_3LTm0Pj|GalqnDe
z`DUw=Wp=$rbq6@y&VsByzWp2>!SFQnSl2s!rcIeR(<%-%!ly(wCBby}ghoy~9UaXI
z8I=}#`^qCYjpxlu;kwP;8XB|5#@L%DU6<#?DLULrB9_S%mrRF^bLdvpY8kQV*<oG4
zovD{|@@x^2lkjXd?YOpKbPO>l4g8XEa!eg=37xZbrChC%ah-yXn7(sWx>$&S2sobJ
z=)5V_G2$)Ci7V}lh)<1pX@1D9ZxtuQH$!AAiy^A4Q~`CYk`xPTxGeAJE;Xr=gp4Ik
z{VI1mT4^d1G;(Tu>d6q*Ah$}yjGYG{<IC;Q+z5LyqiGg}rV>_?iPW1H4LStzb+4Y$
zJKpaoFI7DVc`tDEnDS#=Bt4&cYmd#?4!Wl$;X1Zc(jPmN@oZyHmJBddZqK<%(H{XL
ze|XSHoAF=#S6omvlw6u(8S^JT?)YCMq5BDOa_K@_c-)h@gF0WtJM&)0OIGS0Bd=k*
z7W6;;OQytV_(iC_f(R%R1No_7u)DxKOGNF#0ECKh?X+<-iKz5fJXIwKAc#sY_oa)0
z;NA8(!hEd7W=6daS}ib-5gylY!vXS_&0>&=7#>M}J1POT!F*MOCF^7z+X;!Xg*N;u
ztE?NhdMCxBHmX1{#(Aa7k=T%${@770Xqi_ys%&v1Sy#K1=Ou6Ft6fi@aF24&SCLS_
ze*7g}JMGRb7g{jrgCfRTL1@j~0NL4GnIPK2)tlyv>MrAL<D7GFvv|PyQiZu<t@3mG
ze6ntfr!bW_EgSjbus8a5akbmIV&?Ag2qiU1$O3;O;*lyR{%{aHe@7ZWxI{-h8w<Q_
zn2FSxYDI2X<3v=*@!P1SCGLy8RNL34ZkKO@-u>6Mr{hh#K>g6Jbj5Y#qO$-g>0nH|
zZyL|G4|KA9KbIt@mFu7y8*6Vu`6B#+OVUm4$;1+h_j4+ew<FClsUjebMstlg{(i=l
za$$<5h@hH_L*lv0;gDehHDNC$$=w6kG?ztLGxZRoPB6oQ1x89SMTF{DE=GJOb4|jF
zag<l^&##J`Qy)$#82eXh;Y4&#9%(mH3j7#DlOQ3qycd_3k%cG8SMIMl>P09bpAn`r
zU@e8~f+;sKGtW}gmZp5$+V6*_BbgfaY@Z|)ra-B1_j3mC)$I&)DAW=y*6ms@^TB))
zLzSfIR|>0O7Ju283(rqPCt9c0VYYrlsa7P1E9L2x(xe-()8foaJ=TEPZw@-*V$ilc
zW%AdM?6A$Op2-2qN~1_JM)uDk$6;1Yj@NEDo8ruSi70SL!u(Fl-d*)#;f<wGm`1fW
z=?GO08CzQy#swqI1(m)o7siIBB;lp>M9IFx+^X)y7nLFUG2*@Dxzyh+l3aLMQ#Oeo
zR#Oy2S8Ikdi<Q+V$K_aLoj+Jz5|{e?TCvF6HlcU%eytH0_Nwm#axLA;LO9C&GJmx1
zn9OH6yG!jX67hPVVywHH0_PwpL5>>2MhAE9d3aE(9em8$G`onAP_8cN`)LFuYwf#4
zn_0+`dug&LbU3S^MJ5OK)L`n+st$OX{7Dr1NYS`dwCF0E(WyQizZ4x|Lmkwu+*P42
zG&sLJn)!==wrys!Gey-SxGPfS<I{oWFTCb*aci{YZ==12qV1AbWSed%^Op#HfrQRP
zDkq&xknC(pTjBG={L;|eE3dGejTNpv<>^n+U@8&fL`m{z!#0*d3cYuZ;CQ#%&;;S{
zyOpKaRZ{`-3{S7`6H4+9qE4Dp`gt})SMHt4`E_Tl>*yAlDKtsBj_T#AF`th<KWi1`
z$=}4zH(!S8)~*(i2)52n;x<M|uQfW6Ki3i9AL$*Gn7LZBundQ15QFG{iBk0;<)tj9
zI1|5Ph*1ogNib>%L~?LocTMUHN2EP3++wX5j$VIAo%WEvqWA-<IpU<sqyCNbDTYgL
z;0iI1qpwy)rjLaZTLqJwFXFg@&Dg?|>%*7$ft^?0D8Xf@TTRMl(oxCbiZOj8jITzp
zsJ6R&_R3_jDi5Ty&Fpm|6h!sT0=}4?U>ZE!Juzmyd76@e>ZF<e8pM#xE?7%aAx)}a
z`*bYDOC!<K<HBi;9Zg$f*sKndU_4;8U3aJxDM9sUVeTu^&caFD^`rg+GMY8X>68@w
zp!%$)`b^1tY{yr-X4-6*?vQ#v!__uS9W>iUSCSK4v)W-4i2Sy7*!JbkcE-=6gUI?N
z?=j=DDAnITJnR3?HY2mD2Szsx0p35WhQ}Lip(QI|=bGSy@y~rZC-rNW3jqK6|NNei
z`%`H--UFpf4A)@FXbvDH`ex2e`WZKMFt`?tzs^K6*R%}q5`y@F36`6Q_y;1d5VU?^
zROxsc+=ZwE#f|!LGZ1>*z}psg4tUux`av8;d{4&x-*)fKxgJQ%l_v|&wsmU6-u>K1
zcR12ccdVb3z<GUXbxH7jE9BOWb#J#r&8GFTX7Cid!_>Pe(Q)8Yv1UJ+Kq_IIzM;QO
zyiniB<x>&MHv4Cz83h62s(nF5?UIGuVyET(?o|yt_6(&cj1M3Rv`7c-ggH`6nbjQ5
zF}g;J<}^$T>tV;pUUlaojM=LA?BjYazxH*>u`<(T@4imj*>d{ci}oN1TqmzRyv;P_
z)8U~TEdsqD+QNGc=Asn^2RQDS+JI*G#dZfWy}8oUoAKk=>Z&xGxBJ5DYT}CIx^Y+d
z<(eLnZ1cUb76T!5W!O(O?rXO|eYh~AdXUW+&>DuZ({I0^&taWp>p*4Qmyj%}S0i!7
zHA~>zdKLnH4u2bHs{E=G%s(+Av%L25Wbu?T26Ih%OO=Puf1)JBk9uWjpI8bA=O0Yu
zM&Cu+;^4U`E@?)r`n55=B4uv_JI?R+@Y*lZN@S-P#Q<i#mMIH{v_>11<h8&>=}zWb
zcdP9=_dI&NHSzdSR<MqKUgIe#8`rjX+lVaL=z?yP>%BU#h1v0ZnHjqqG86jF5Ls!=
zqF*g$^B47S1|6?nYm|3ac}QyobAZN1^>WNCkCjw^0qyTd6%A+C!x+|ip?#$%&!h6>
z@Dz3>7Y<B{0GBr)eME}_C)PA1q68X#N#2#dJf}$uxq~Nfb?$Gj+B8+)q~5@ZP&xR{
zmLrs-Jf3Q}R2@o!Dayu37{WqM8}}%xLJL>odlr}XD)jl?x6Lt)X#GHaK+ZE1IEmGD
zk-#}%8=a3%JxlHAs7Kaz50c1{*V>t;toCjz{9KO_#*~YBU#rz|yJfZJTahC9N>Oi3
zNnQ~^zjcWTKWnCSjBf#B<phcDnAK<nrlQa+WPoW|{8*y5M@17-GNYE}B*qmhB~D3S
z<dmAKz%7i4GATuz2-8&*m_$ql?zq7u$6k=0KGe1+DBbEvhE5XClSv?UuBAir%<ZBn
zE6m#&L+Cmg@r$17))4lt1Czq69ahEwq0Hv!c*d5^WMN<X;nxZKo^=1FY{nwk1EIEm
zDP|~-CP{%~DaNX|LgvBUY-``(FSYE?@Kj5e*G$4U+V%@Xu(_BEE)GuxxrwXm$Pu)6
z1Xli>xPL%lWH#}6`IBygPZAErY^miVq!b?_nOt}oQzq=3S1n;$DKW$5`+4n~*b$Am
z(}@gK-R=q5Gb|@t^RetwfMa(V(V%}@x$XoQ0dsz9RJ^v<U>|hBOBRJ=f8=OUyrS(^
z(erk2bSp(?$KZxV<;x>XYA+=1xT5Uy%1%S<s&wU`_quq9%K@cyZ{J$(*<B9H{@cZc
zZElJJ{LR9>7P4NB)C`N4;@l|a&8jKZVWzpo>e)JM+1Jbl;@?Q?bcPr$R&(b_M*YB<
z3olLQRaYh74P<Xggp<y>yH?!d{1TLNA=LHKMe6e#Cc+&tfM`oTS|u6j=X@>@f3Doh
z4q1=kiyi}$vm?-!WK8Dzty{(})aZfDzBn+KyXfgYsj59J^u-fmoQ~rmwf{1Zz)dLu
zvGH4vs0vZgR&^eMcYoBR)@ulD<$aM{h&A8uHYdG4RGHsM*qaot7Qy)M9q!H182)ds
zDXm1=WCDa~4ZDSoQ%QI}YmC&V(VqMBf2J*rpb7Vol1!V_2M+G8YFW#Ej~#lK^5xxc
zN@04H40-lxQ#NlxGp<{#+#3zMu4??8MySEcXqQ@WYt8hDGr_uP410(XUe-Bx;7<G-
zU7hzy6n&TsAGV7#y3Y#lss`X23#rb70%D|prpX9cQS>t}7r{k~jIM&&XjrG;NGXXB
zx+Etg+7yg@twP=ko@A@ad|7kC@ah;nqJX#8Os5xWXM;0<szar?6(E3_o<sip3P5<7
z0o4ZJu)y(5SbP532P{J&AfX|^t|I^e&d3sE#tp!u1*lrAQTSj~|Mhd=1C7gQsRiH>
zvlr%l+n<S%<_*nNop8*PK*_%Jc_qA)_VrcB{F2siu4AEn?fGO46ELLKEEC~Xn#?+`
z3BVTpL+=ho43|l~3{E~45GGFnORS{jmbklVv~^F)VokYM%#a4^P?cCf#RF_b??m?w
zH1Z#qj~EB2qAL-;ZiU=iTn7CVMAyBKE8%G@QxZ(QNvOQDYv0w~4iI>l&ObEO8_&0Q
zPxoumlU51OEcY1xFqV8xK(1kT*|5%#TXkpE4e~Ft090Qtapj{|O}H`Rg|l>*$nzcb
z7`RTv2yl32#D;Na)*3Ag$h0H#7X3yZAXlJqi`Qm4IPIF@;5f$_6*j+pRaz#tqvCCj
zbSGh|-)c7Dka0bKk|%$<ztO6>KJb#VazN;M2)v7n@)*T#K#S+(Y_8&ttpyuGNi|}n
zX?g+=9>!qzBNo(|%Wkr!!&K+7Nxzwr@6{R1`0&l#@;m|jAG_X1gB4@70BIh(tK5s5
z^rU=I_@`4_dP=fZMIcxGc0+g)U<v;|HjwVTWo&G5+tol*ClPzh(-8<%h}mEHC~0Cw
zzmN%}eEjKpFeMZ}u7H~0qI0TL<%5Fmf{LEso{0RaP*R<>FjRf+ZOr)#Fx+X#!joOB
zcD=(#yq)Np698u}!lnJ1>NTpgro7(z-3MOCGcKgT%5fYR-GcmO=?v~DB*S<(g9~P1
zof;bv{#M1ojs}BZ-6T&_O<oO8&=ohfmG-PmHcWLZzn6g0?DgQUpuOLsw4^_BBf!)3
zzq=T=UP*S+KK=nQTtVcOL%`<K+ESlmG`C`JC-Izh5rkb18K73TOnUXjnS(x$UTH@%
zHQ^{$QU=3I7;@9;<-PlOJTMFAP7>%q@s^j_p-ai)a2D@g1VfQCs;xZh-DCh*O0{n=
zDYW$C#|i159y#fnrU06$S_cO=MUaKXJo@x4A_Zx;IZ>D;rK5iey$zVBaBms|#WX?I
zyx8B&nM(2+hufk4q%ll<+}{{9_+u}^TP9|Z{ESNOQID<>**S%I91h6Mdc9Wd?^S{0
zIh%|no`h55Cag}#4M}7ql;r+cr5$fMC!U+X(p0*`7@5$g2ECQgrpSGn_pM@$$}rnT
ze7?W%LD$G;O1Zs=aY>x&;aRQ~1oZ<jHH-w5DuLu?#R=lObowzo-hydgSAm%l38Gu!
z6@93)E7Uz|<<r!_bAO5LvQ+MZ>&YJwnaYcI`iE_=y+wZ~t5;Xqt)*0j(qwNxh9J&1
zaJDpQh*pOi2{Uv_lY}CAKmAa=l~7q!=>Hj{(imBUDu`5qRy1Nq*p2!%I)bh3a3Z*4
zW9-$+njX#i8Le`3M25;@i_VaqaQ6h;sxDJQikFV2q{ttTYe!d8p3<QDcm2hjfF%FH
zQye--6rzQ{xe4j@xKE0$LcFWsQ!&)NsB9YpwSp)j0N-DgW6?z-ws_#(i<{T(nF<=8
zxSy^Xh0`rO%xhd4Yi&GaI>sYs^G7V9V+4=xK<JB=NH$qCh4)J{X|(VN@Ow`so2=Qv
z>U^LL-?9VF*Jiskj>~Gddj0wQIW$GNfdnxU>qR3N%5V4NY~Mnu;=JE%7D=aA{fO?|
zY%D8_h&K5=;iDi?#wn2~A0CuPn<bqhk75*oKq9THVWdh4^UwS&oMt~~Qagg*Rd+;p
zg=;d0!8E_mCSf-~lq$Fl;|+}{X^iaI%x#F{%sopb-fW(Q;ju#?d!I(0ela{owr>hJ
z_DT=gRXx~U%`=qOcC=7A^^v9VJW4CiZ*YouG)24h(tSDR`(Kp3WmH@3yFM67ftD61
zTAUX5;_ei;;_eQ`T|=QzBoL&y6(<COOK@7;-Q69EYu?E_=lsr@wf-|7=2O<%dG?)^
zWUoiA`x0FtF~gI>!19^H$@z_(JtQTh=F|W9D5_}wZYOS=K6Uw25oV(7#&RF``Z4DR
zBE9nH(D5^&7(Ktj`B+3<W{lvm&gYaVU}CQyna$G~c}{GOToCkmBKmff@;pmZ&C7Et
zFGuCiGkZ|AV7#KT0~v1OK5;TCt%q`GMCL{Z#y}}+EVZ!`0g}Sf@Us7{-shM2S?$W^
zQO?EnwC9x(g|$k${IUeGdD{F*<g!a*YHhbb?=%X5TS34~=$ukx)gw9gGA*jj#7)#|
zutiy^W>)F%IE<IuISM07jxl7&{DE&X)KM>}LFd2BFafD2H*1&k!Lr~+<mtuyo)jRq
z_U|#}$F|*L5c1SQm4Jl#{%b_Ci_=PI`L9G*bSyyN<)aC(p8**}R5|usVk)-(r^uEo
zVG^0Ncb#roi_DTE0MxB}NTBIwB19_ME1!DM<ZN^b83^$B8lFCRe{A6p^B(Ek%Y!|^
zY6@aX80t@M_dU(&^9<Os4f3=*&FgYJ7Bea4KZW5_L%J-ulT2-6qbG{Iy=-7?#jPVK
z&WR-6Lzg*Cyw0O&(Htlg0lJyYnwfJC)AxOheHLpk@|t|Lfq@c~*As8)1XzZL7DrY4
zM+5pTyv%8>$-boyK&d!Jaj@ofMV6D7f$N8!vP=t{(M811%reM`l5?0Bn=W7fVaXdP
zb<s(GtiXq)2VJpAzr<R8`gRZA$}!D_lKiJJ*`IzE?ud`f`aACE#3*`8pJToman26{
zasA~#^O>u?L@^5LsRR3-E^psL^IB}Q`Ixap!#vM)d1Ek~EUrMH;kD6bXIr3W(Q^)>
zIIBSe`}-7}x<0IGo0MkC(Er)KIEu>4Y>20Xs@I`#!cRKGr-hX1^e9(Z>pVgaU0`*3
zoV)DY`cM*D(RS7H)%sqZA>=vo>gvg{4vSfNU9{6_jq7JW!s|*fNY#PKRO=tWMu&Ri
zX}+f^=d|BnnmvF#_t+(2HDQ`>U+&JH_|c5QScCAhW1A$ex48xuOa7)Ir<l<;mnCPa
zz|s{#{QMAFUsk&wOQKKM*p_sRlUN^1=-hBeam4*o=TM!q=~zO>g6yyv$P=arnMzDF
zOrUTflKVhwj^mKWRvx>{C@f8}z>xRE(oXp4W&RcOkoRk|VW7V*>SLLBwm~Q^4$4oa
zlSz1)!itz9>2J4Mh4+>KO>62Vzk51cNR1OT|NAzTL)hR)cQ??)%QAjiCz6Xr){*Hc
z&onM(5S}oCFjZKp!G+#%$5nNr3wB84rmDOc?Vj@^>nMga{j)x2|GmbatiYMkPH)+)
zz=<lFwTyfF#nUC?As-9Jqo6Z>61ydF^)L40xeeGt?#wR@rpt%pdne-f8!rf@RK%qF
z$!GY@sD%RCBjOJ(z@~$#TrcFm8c`+(lgV9`+l)XGpVRvsCZ_Vu<cP~pjT(f6CBvaM
z<-^hIHt+i|2>bO`Ok?`$jcQJ}A27YuJPe4-(hQ=D7;HQZPKw8H6j*D%3anFJ?@hp1
z<n2Jr{0!d8^}>8K^BqxTR{2Vd>I*xWEKcu+5~d#*Z^btQw8?fKjsFbav91923#ur8
zhc~SzG`iD<c+QJ6A9y`lJ&KeL67<n>CY)A%Q#p+;ejWfCk7^&^b`16C)qgkSNKHI+
zXS5s0-uUbEl=0P#ktiWy*cG<&heGd~H*U63H{f(H0Zgpzyb1rQMdz_FRK?XRGX<7$
zkK<DHMTK0I29Et~I>=(Un^V~4nW`n?PC)oUe(;R{s0oA~WmNfVzTku69_l8@?W8ML
z<J{Xa;(gj)D5A~T`RyJyF4tU>F5g^pk@<Kts*7yZ2~wpngBo;dsjft&MAt4yHv)^F
z?oDS4wCwG)mgMe$ojml0OBD3td_;`!Kb50D5HGoy+q&<k)<L=1OKe6@Q1@jcPb(AW
zC*+1_G4`p+`-;p%f0)|t&O)49Gcv~;lyOXsII+G?0uRC+PF+xDeE@2#<^FFmCu*Eh
zG>6sBuVH>PT7ElM_uu9(+%b++8w2exL;MHPT*ikWD#b(1VNYJcB;2mK1}R)eN)4B5
z1Sf_uuKt&weI)7fey_5B;#i@T!r3sIS-&{TeEv4uW?+jYy}!2b{?zfmd>VpK1MG!Y
z28Zwo+2Y*{D#A54t*9*E$esJAdmY`#tCph}#@RRa6yHPQF}{*oku+mRK_p8HMr^r0
zXmH8ZBTv)+Js26C&60b+`41rZKezw?yQ1W1@7l=8{kaq4WmdpAMigT1O^f2>k7u!e
zO|L|mDau#`f5y%uPK0V%{|%)_x~CZc)OC=aWMCE-C~~QnR9{GjpuLO&X%ujASkZC_
zzVt54RNcp~Z5FN|y8q-v*l6>Gub*?ZAzx3KP7$?N{<?CKQ73{L<mlpccP(6fqE%SN
zUhw&k>&2#WP10Oty+o~|d4e*8Y|k6ei*d<2qETbWLWxn+LgRz{L-4DDN9J+-Vbt8p
zhuW}c5L41AxX5)7n%XwLA4`Wt;y}U9IOk@3X)BV0W)#~H(&|j%rcP&7*ikI-I6CYX
z2ssT>u~}I|iT?pMxa4M4t15*1q+Ts7MT9SX?S!{+4b;%UqfWHXTxhE6?QLC43z80l
zSat~vZMBMu7Irh*D|w%X6idD`Z#nz}FxY1evC)|SnCiOs>%^EF%q+j9x$DDjw5tuc
z;+0rnKi-UF!BpL>NMoC7Gnzi`ns!#jc<4oV`AqBF>gygW9Vp!pUKPuP6Db|28qzHb
zoJqaGYO4hdeOu1h^wQO7GVAd2*NJiE#u3p0g<-uld!9(riLxGu?;e!zROSn6nG4!D
zF;US$wG8lt1Q5M-=cH<vK3R^wcF$<74@&aZczOUPv{NuCC0z4pj@H8^bXf62Pf_}x
znqZ;CjXQbfjpP8IQsP654O0b7=h07j+UFP1)3pxeT4!QLUnP1~gIg~T3}mL!)zZ^k
zgL*!jok8|S?8L-b2K^*h^SrkkO@I$G4ikTRMXJMTqWj34?Jk^MTbP88RfRkj6J`2_
z1z-Y$N+&==(tx&(iou;wF@D#88NWoqb6WOSW?B_=L+{Uo=Vut`@>$C>bxG(mY0KWX
zS-{z+__vdjqlcHeK7g>Bj+6(M5{{JgmWF4uw6D2%3lLC8Gz*Wg6Mcg+5d+a7G?CuP
zw;s`_-~G&eZ-kf_NGRRP`uP?sRR*#iH>*&CZ@9!4P3NfFwU0eWRp^KI|GXfQSp?#$
zHGc2mzF}StVnV5OAr)cHZmgz?hl-oMw8D%!)sCPL!S6&q;kIFmMc7?80AZ~d+i#Oe
z(}kbrpUd=|8gi2P-ekNhCMqgMx2X(vk}5M#yD4ncYG??6B&#nw^V-FgaKjw&9jw*c
z_3%$CE8DCYD>Q0o{r_rogGC?&iu%jzD9ZN63-}IrJ@rm(qx%+utl~Qbw=N~-C)Z_v
zsn?(CK8~#HtGqNXiN(+{>qxIKw$3>@Ypu-ndPM6B#%5|8iv3A4dneu*zihy>n4-l0
zf+P6E5}Bo$B-wfy$+Z}P<~tS~QLHP+>Jm^ZYhQj%#xZHY%FVLpe*WsJN4@E%^lP7W
zlul7-;^^@+H|_6oO8i9?M>f+$fhpQX{6PYz@TJPqBDip|-dwmOC}7UWTU@MLLAAWF
zxyHAbdDvbHBsXw7AqR6Xva-@7?6&GJ{r#nvH;FgNi$sT^Uj7Z*%t&E;rv1(jP-2`e
zZ+^z5(eUqw*~LB72$pIj7E#m-B%j5TQ)U)`imF>ocrG<k8l@7&jfB*oP$Mi~2Fo_I
zO4>eWmx`9;67=#K)!BX^cNeSw*;vb`9ut;j+F!eo?j2O@F^57Bka76QcEJS`n==I_
z)RL@b0g3;PTZtB@^3_C{?MeaNEg^c3N94x-O>aQ#W_j-tD;3;!<g1MhoO(T=!BKHl
z+343TLE400?lTLoSn1ln+4b&EA4!J1p2I|f3<k4TWL>;NhjN~yTX)czeAj8QnCC>F
z`u29358~M_p#kKgL3G=P;+KD^c90C>ig+XeMkHGf5lr*c@i6Tt1b%vwy9ZuF;we=N
zkOmmI2-9bW|9v0uKLbiHyr}N)z!ykz#ut?TtyeHG4cyxlANmJSA%T%pio|icfvLz@
z2h5<3lqN?01Nelf2mHU5Mp;K<t>n(0u}sdyJfX+i;D=quKdhY+EP|={k~jp1!+9Ra
znH(8QqR6=#LO^9uiv)|L{-e+~rOuoxdKqEQECDZCYtQwhlZ30iZO?C6GECbJrwjik
zN0YTj+1p-1+vD*i8Kc$96f#9}sqVR^B5VBl^i59_*ExyluTq0I4OVvfObOXNjUHFL
zD}1XB0+gUC*0-^{M)JcWT>vx8sm3S^u*DN}AI)P`WKL}+$z~>lNao&c5~Ais&Zvc(
zFz&-UV$$kb<(@gcnk{aR1eP5FByGQ<EY*8ln2rSR3Y&jF-TPIU1kungJXAUKCsJ4B
zBxBna3TlnPw#mZDB%h<$R2pFa;8Lf!zAjX{<G)2%PYSfpN<K7Qk>gI<ewzbxEIvOz
z9uH@gM+G1?fcfLSY|Cr0HE1eb<q68#$X%S{7u1tEs<Xz+VvRd}bjJMcXp9O`0vB?h
z;+$V!{=QsQ#yi=F>hbHD817^EHR43Ez#1`a<ArhPa1FDaetkR=B3hQfI+ymoaan@F
zd6u$X)o%U|!OaDi%IkY_%AB8D^x5M{_RCkFu#UVRc9XZ&;LZia*_@!<GaQEl*7!w<
z7y$NRVd!3#yT9&=d6^upB5y$(IM}=Ui)P#nFVVe;z!1;ZAn$~D(K2+npM?g>+lmm$
zUX1Gl8AV90xQ^I$LZ$QjAy*9dVeXuuLt~LDl|<a1!rtH_x6QU4uj-v?qxv`h02op2
zEYysq+HbTrn?44YEg}Mf(FVH2y&Cg@dw0o{co%r85@Oj<{*gqI%s`b|1_1foXSkmw
zlNV3^Qmd{+=!w2UyijGE5*h2S?D?0vLf;LPTrPy)^LBIAB+9My5&Q+S#klqDDdKu0
z=k$Hnbs7!OMAxTfi-(4))q4bgYUNy{t79F!U=R>}KUs;(uY|p(V;`(n8#DLbtNWvP
zxMK8gSM@CkA&=bAcAvaK2NM`gw3P3?JZ~^CXW59U7oF_9kYmb*?FE0($y2yzLk(?d
zk*e=*aq$jy@5DY6AL?Q);|omht`iCLt`k}0%Nqp=V%cRAhw_EzWn221Cz(>7i!^H5
z&#k0&k^AOuiYyv(7Ig$<WXe~~O{R~xB#4Tq_PSh)jiPNWz01xg`;&{SVo8@jvu7Ta
zda}`mxBK$ffKbQjz)u^W>uN8(2*kclrW3>k*4VWdxsdrIeyEfzD7pCha4NgYKf+kj
z=Xg_6zqHjv<>%n{y4{M(sOU_4@^v!1;G$_S@=xz25}B?J7SW*YytEX%9`FtQ2XK-a
ztSpCHi9vCM>2UmvD%K~Jf<C?0liz4jL-^U|xFEU!Ww{5e;bF|MyR+ZR+(~;?9OAo`
zQ&%=m+`}aoab3tm@bN(Z5~hTz(FqCq_Ezr@9ZM&?VGFgwYDA|`WqnCIRbL+l;A@bd
z(+oHvd2^SANxHo(GBxZ$`>oK$UB#SjG`xs+ckTK<)BgjDtuq<4-v~hn_Ylu?eSi1m
z?HraIls9zTDbuf!4lLpO#I#D^-p#oL*(1w*&{1+<?c;+!rV8x;A+!#5#66YdE%-13
zzjvOdOy7hV@;_+#A+F!z`o=qplWSKXkD_@nwIbzUS{%!y8IBtt%s=rPKN`uAVB_`~
z`u_cKcW~ew^z%k(5ZU9$=Mf)73liOCs8s)p^Fu^mJm%VPgj$oZ{`cpPDUN-fAn4~{
zy)Z`99Sbc`Ru=zA^a>_2S~*h0k*LucZ%9C2ONk{Ss$e<Bp`!I<UoA%pHtEsJ*|T)^
z#u2<(VXI#|z5C+?Z|RQ<9$fh2b6>_fHvz^Q#xCXUJ)VAM3h~Xisr;RXYy@H+qHs+V
zVRvr`ovLn#Kk5i;GqC%kn<u2~)57dZ;tQIn>@<aoL{;>Q_Ab9eEP&^ig~MAhiVQ5J
z6Z(@`tPR^32Fx5YDFT#4GQsskw-8x=p2*@brTXY2m#l=c_SN(kA?*LED3PpPDv{t>
zyvVvXj@*Ckkrur!>)X!zNeHtVnYb-xcl7<BEFAR3J&AGlLmDq8|Mht%v;qD%bdS_h
z8brUwWR_sj{PO^{zm+saUzOtDMkW@_ed`tx^__lv7D|5Y-jdzZQdkvbc|Q>+i9thX
z!v{dEZW0)%gk<yFs3b5hoEqr9ba&G(X^A815_#06{Dl8jRLrlXeQj=+03tXzWGQs1
z^@4kJT=WP$^Q83l>j1uABIS!p-itJljIyz+AH_{9<3$0U!lKP(J=15mIB%O$ha8$0
zD1;F-A?L(iKo^HH&->hjaK+ZL73D}6{d>tmx0kFvhLwXsTu8O{#EhP?A<s>w()SP5
zqRsDg*Qf?tzPc{|{-FkM$#@gsE8Qd+n|-<bIOk=*(K~guKU0YABhLTpL6^=UTI^s-
zHYvm1l9Z@1$r5B4rU~hXs2*-QjQ*Lv)bYZ(Lo^}HgJPfdN$>R07zqX_R83y-d=RXB
zr?j_nZ&=;dPx)uI>FA``hh@rp;;Lc`&KmJf8c+DI0jcSbx7rsJDD`fZJF}6W)2&G&
zfwq3d68RP6h47ows}c9X`;|$GC7VU=<Wo+Q`Zi~vAO0@Md-s})SK`kO53QkPo<%CB
z*<%5W5K4VnV~jDL@qs-ORGLzi7C(~}Am&tU?8j))tXlG<gX`Ohzx<%3lZPB*e?>wA
zLsA{J$IJeqJrU26rLPLnr9%l}Noz~9HPAB+T@hs?KL<N|7T8ql)6ilPc8~l$KK#^?
zlqXx^=pVpaclbr6CrRgq|1B?mev@!nH9n|h5J7m`;InbiBt7!82vmX#yXYgoB5-;&
zZ<a0!ICPX(e_j(dJy&fK_y02#SVE3j!HrAxA$A#aFO}PVNo{h^2+tsEw8)c<&Dtj@
zFU$D>d#2ZcAGk=mmmbohEjzLFqUn;kp@j6Kfq`DnG`7pO`@FU?j=Lk%c>1@_-!RVm
zxmb^%jgTF)&!?8^S7$2K?JF*#EtbjszwE;+#*$uszPcpN9dI<Zb>jO=C0L)D^<Wc~
zm(3-%gXSu^tqH{oC#ZXi3=_(02}(H)xFUF+_58d)b^KI%pP{Pvpvz6OM|;hXG2_Dx
zrx4<0U7ay=IJoI2ggK*|#oPhgS5lvX%18?Bcm1pdy#?u0HC7YkLCE}8?7AZsg=@0p
zhzXK~ICQ3TK|{^CC;Sbl6OlY(9%Ct6K}Ym`^yKVyOdI`3ihzX0j-*?kYJOFTW_n9O
zzSjMRawHfY;_-S@#l9e|zyEJl=inAHU^bfyfqR9t)xP*^*%n8`WP}gzXgLjXl!eC$
zRVK1!t#w!88=KS@$gx4C%=ESly%tLWSJyD=Nxmg`bFP7#0^ev~Bt(c{f8YQs>JvEC
z`JI>Lm(G$nW{?Wn?`ZQ(4?w;A>sm!@rl^pw*PA)zSRa&qOp*<(d)=93t&Kr@l=W{}
zE<F~jZAy=ERkJtv)w^#`TiU#_C7!B)>t166A1?{f{nlp<vk5(Sz*gQuU%y#^DQJ{&
zDpdB|LRELS=dwjd3aazW!^xsL$&>VNcVP5fz%l8*Du`ihoT0}~Kj##BCV_Ip*S4;)
zBdsnP26Zx*R0@IqtIY~S$fHKCbOx;=dv#Ce!_+lIe7s6jhVe<BvX9@rbIfNAWVs4A
z{C2_LAf8Tm_(i~hN;bG3=<z@9;rqfvOOdBVTJhmnAsj4u$Qivqc+CrmNvmzMeCjsn
zd9KJ1dCv;60>on|Q#n|XEK+uq89BTmpL5>V$|9u9RXq65a7Msm(A?9tcznUht?A`7
zLMjBOaRoT0d5<af-v$4_N2JW&Y@Ax?{7hy?A*QTD+32Xh?*$iGB7ovI$w+}fa*q-_
zeRvyUngnST^9B2fG?UQMe>Uh)5Gt-XjE0^PLIfH|$*ARj@?H9ue}nm7eN`D2fVxab
zwSAC}%|=DBb>DeXvUs>5xcfFCcP*82-Aebfo7V;*vC?Pz*PrSH1NGl3D%V;*A42>#
ztA^6P1zzn{xDH~gWyt9^=+tn)u9aGj?<aq%Nw}g}4nP+Vwx6><Bm0xSSA?Dimq;j4
zg-t&ypv=g4*hC-Oqpx1<H%IqbBF{4SV^`Mg`pt;93|+p!_R>s-TX&)ql+~PIi*Z=5
z3UTrHINOs~yO=U2^gF!fqG|iv6!Fd3Y*wa{qzwJ>XrtV89gC8FaR2(($FUE8#5>wN
zGQe@PiCcUPC1&e>Kfo2Pzd<GRyZ*eKxZerv6Fx=Lw9=f!mw%daCy$=2Zw|G7UM^je
zHVEKHvrN<$%wBK`zSf@>-+I_WdRtsVm;zZtG!GI1HtDWQA8CUa4I_6dv7Z$;&N;sS
z5+8-1=uojGN{%P8cwXbmzEFy9YnhEm9jcA<UPq8XKji2uHhGNpc0um#lfh)+@`Uw_
zLrWuBuF{wX2=cq)#nrh-7LA1XS_=}x(^ABND7;dQsk>b4qoScGb~%NMrMCQ)E{GO@
zEiW2iQB+S)JS#Jx#o#J>bokdZ3Cy)ZS<D5pR!et5SPuyoKI&!|I&8U(sTJ9gWwI5e
zXi|+Ng>RE?h7}(ys`d?Uxn7n|MP9Xjdu5%dkQH<)6cofX=06!uvqQB)(Lu@}pb?`U
za4i5aw;uD0W+!-pqpKp-r}`N`qF9etPKNGN)9#%%@3X-rja&M%nBwiw#v_Dn!7D|V
zMAWN$XV?=Q_77kJVS0C~8815j@Td*x8nFI~+vB8D4ku~|SZYj|cqJqJF;mSsAje;u
zY$%M_9_65O>;=34gT&YgA!p?6l;g5lq5g@N(TQz7ZS#J+PuAl=|Hi6zMQts-5|u<-
zyL^WtPGRtL=gszD(z1u%1$X^s={j!bH9cH;;M!jGAT?AM^{U)DEsB2+Ltrykf~Vm=
zh12QuSX`#!QHEbgv5%{Toih>Mx20%p)R=e3gaz&L-Y`I|;}n&1fYVnyLw<U8?JEH#
zGje3-;pI9`et>&Ju^|%i;pi9yp6@u~uT@{G9-=t6R%(}r^?HabnaoCF_4_(PySN+%
zmwI(gh!-0Tf{F3ilQX|f(z;FyZ%zA@9P7hzTNuCa+eCTpJq``|_FwHxiX-dl<o=lV
zL6E67d;KMzOIpq?w^M(V<y;4Axm9nOmW7#{Kj<>8A`FjmEmR(e7qz?bC?^Vp3_f{|
za#PONF?FFUua|)UQI!Cf@2#AUd5aMr4YsIwe$;Eoep*d#re3Wt>w9Y@`z^XxkMrT%
zvt!N-uX(x5yBxjVxeqKk0Dzl*)1sW}Uz1!&uHR-GJz2KUU;r;-RF>X8>Dhb?_`7V)
z!JE{|ed&PIL3z%UBg2B?C%=T~L7m?r7bqiuGN9D1k&j;=t_#Ql-Hgbm5u^SpGo;|0
z(?}jlcNSkTVeXb-T()C$<n-9Dw;5d6jb~NV2Ln(Hri%CIh+|pyeMsfLIXlzne=;)6
z+b4rOvE<dA(?#|Bk=z{@h*!u2F1-%Tt^V$}cSkW@H5Yqwi!{r`Dc@G^VR;MMNqS=W
zITuJ0P{N|{ZtWd{ZW|a`!EyH_Ci*BSy)bgJkf8A+QBbp)gl`bDF^}-$5;gf{MpRc<
z-L1f)qpbO|tj0!CgFB0~Y}I@IiNcbvMy;{W)2OrvDY%u3igBT%<7N6;g$00p`Vtd6
z+Au5lO*Ig+TW?^Rz!J1uy+EezlWD6$cCU_-qcC=-ooTwNkTllO=BnV}gts*I*q<-`
zhbnDm%Qh{cy4RJl>eTpAjQ_0{q0Bp;nqnb5PK_&)5r-#}YEi9i?M$DM-~JpUf?uWB
zhJn6&Qv?oFqU}@rOnt5a<V91&u+y{vs%agSX?^UY!Dgz?Fq0yhXt{Cm8sW%|xs@jE
zoaNksqkXukYKpASCwLq(qYYO)(uQYD_u1jik|^cHM{<Pum!*i;<WsbvQ;OpiHh900
z(G7w&yaA-PP(&TTQu!J<3$o9td>BWG<(@&}JDkP-9Y+%A|CX@pO-EWlRRD;mL`|<t
zWEUNTno|d};_6V=f%VG}_VB{~AIvV;0{;MtZQ4!Pjx<K7>CY5XRyE1OB=aWhZ{eNv
zjpj{g=ql25c`BXPko<U2neR#&+!C)d>XDGo|CsXW^4NrarY}gb+ccgODy1y`Mp!s=
zTws(w{VZ5M{a1u6H&7+Sm!?f4%mPlH!weRugNti^>4+{GJ6)H&7A(85*yZ+OQ&chP
zBabGWJ9VN^G2vXcca@<)5WM5?`(RsL#0b?Z9Vqc2eL(2X=Y5sMLGeW|O&YN$E`?Il
zF)(;DgX$ghlIonvI*Gduu1eetb}Bhw?wCyC%t=#=4Ohth>Y?s=lqHBdA(@MhTYsh$
z0_~qoX43MGyblyEU3j~!mp<+jE#0Heu_)j3vc^y;>uVDURodQFX{<?*CB(gGpzw-m
z&3<aB)JHA;%1byNL0xL2XUeQ^IIS}PVS7vJ=4}&~U9lqRnqyTy(DdT<7~#H8H}CPb
zCp$3PQe8h4Cp-Z~6_pp2O500z!A>oYV5=#8tndA)l?C>#jmsSPNYCKzAAnqApOhY=
zzH*VaJNAaf66$tUQ9*(Fu8qHVN@qT8LDw^p#g$u!&CnIozfxNMGbqQ(SAb3xIa`ew
z_3bDhAd^!<PDI$Y?yv6hYo|Bg6DjhYnqEhXSNMCU2RQ-p`O^JfPMG0^Ty&_LTK?Rc
zrKcP$s7u`h2cR13f@>hnZ1Ezg64j}+cols}T?jk1_B=`M={8mB-wLRCVs3Msh!eJc
zyLO)AAW?ieJ6)_U(rl^DqWT(#<bbdIz~RKl4fN`lluv*&%ptxcqPF~uJ@RseL}kIw
zV#sTg!VmXqA3eb~i<|R)tZ7T&3F;y{al%v2-Iac}b~f~JNbpqF!Co`l{?Q<VJ#wXK
z>IbdZ*X4Q}>lJ!_e9)tIR#?)#j+ZUTm=?j7m87e80m{LMs~G3pmQ2c|ND}dYS@8a;
zSTVesmy$ARcV*v+{wMZ}UJV1s^w(@4gjcoA63=q(rf^2P?}oL4;mFIz!AvEp-!<hM
z_ApZK&R4!SH@`mXGFr3Gf8`vYuH9~63Tw={<1w-)ZD!PQ^i2A0PbT=l=7W0|CyryU
z(Pd8zPaXb{*g+k(#H*o;?y7sM=|BpqUBOvv-qGLyP^MEfl{z_F)ZbfadeD2+@exo}
z@SOaL5qE4(Znu`VJg$D~sMIt_d4J$Ub@oAjH24^9&66p`xcUla&057Ok#&MpGbZQ#
zT2-@K5?MLlOLkU7y3Z<2msY4=kZji1;OF6pWPKQQGqR23{h)m^p5pF)qV62BFp>e~
zDay3St+0z^)9<MZm>pi*CyK!B8SW&(xqV-13ONMg_*4ChLs<=pt@*cYoKOeZJbyGz
zWk~eLEbSRsz#KUako&=N#D*)=%9PSK#zdiF=a`Dm6KtHqkG2}rSMT)dad6u(g@G;L
z#i354y6hCIusVvzrB#F4#>I><@Qs0t)4N>tN70gurIiQ;S`>X9rR<D`8DZEOWb9Ck
zm}r$6F&v>2p!-4(i5`TD70k9Q<zgDzK`^UVTI{3JJVIF5Z4yAvlkS34k-7uCzn3<o
z-*Tf&o4FZ@jO=vxwP|_t$OyaTsCgvmz@0G^?Or>X1+<i&tZV4U0SHgT2+oIv*lKv9
zYfNxcytn0!MfbaBeFKHF#F*?lGz)OB?J3B|oSd9(n%qdY*p1dvjZVhmc%b^k5;i}-
zg{liYI}Q)kFfV+6uAsm>9FrVVH8jc@>myb9!z@3B4s6e&>eiKRV>ie9t)1{gy`Q>n
z_+nV~#p=}doI8!<{N)<uod#3+>P62Bu8aMSKc}M5A+zK}n+OwPa@_*8ch$bW7DSKY
zT5g*khd#7|IOQm0Je%1vo@0I7ZhGg{3Rx}%pE>VHcS=QsT01JGDR(v9$`o9m^<!`k
zk|wjxQ9Qw*w;p;*pLI?e#1t9_!moRrWJOx!tce%vhv*<>^FLJNTxi^Akq`T_P}zJ}
zA6*2HRD)c$AG1oVQ^&*2zuSj|7msyUM20FgD9iPCBsL!Z77glw^n5tne>*&=X{XA*
zHhaE*RF=^&&XByJ*p0388iOj_Qs&cxfLI9n)6q=Pc^2351KEuX6@9a?-_1XOfujXp
zmq`EMOqlV3yZt54&8>Q8H({5Y#H#H$9zh(Oy79BxChylk^MX|w?|pjHnx(ekOe8I+
znE!;<yc!wxsKtliXoz00n!Osz=906};ad>A2+^>n=jzgO`QkC2vz1}>vz{t4L#4Uz
z#!wgvN?|v^C;qJEvz91`F=WQMOd5pjXkZ(d3}J1xwJNGxX2>g+7iDM*`)=dV!Os>$
z*#B%9<Jk}6vMr75`dRSSx{nNnc3LK*^KS!vmTl&Kv}zXkMir8ZMC@25=ljUV&3r%|
z=pz}UtOMkp$4ft_^6t&AG3^k5+>Nb1d|bz0i$@XhsfZP`tNh9aD*!R8kq)Aue!$J5
z8VzB%n0kcXtM8QD>R$Xp9a1~3+>uMkT`?UD5p#O3=4fV5`u6Q+?YYR5joqFrpG`o6
zH_B$!Cbovfgyrc2ggpMbjjoBe#Kj~0NFo0Xl@nz?xB+aS@=9!8)M2G(oir%#_FoNK
z%tUhd{!RI_l97_@!21z)A#XzU*afUKyd!?tAFhN^-%ZGnNMfS3ci=<Dj0{^O{9RYU
zY#MI0U<n(*-tJyD$8%GD3Nz?6tXa#AJf(4h^`<&M)xuqlJXx%Dx4QJ2CwhEGM5d5}
z{&9fLgzwAWCUHMCX?<72A-50mmVU&4!$~S!iWZ1k#+T~HyYb5olZu*9J+RBgagY*i
z-ASHD-5<&gHcjT=M1ENX%;N7t3cZDxnx<M(woc!x#Qb4Y@=Hy9nX7}b+7KwdIrN5&
zeRXkf%-)hr;Z4G9O*YdsczBi2Tvcg&pY=(l(D+^Gwrt<8pH3wXixhDc*n2Xx;1n$+
z^fU)k)mKkmnDf<%P1svZ$pvl<g%UO{iKEgmu6AoHu%28CEUJhGwuF-e810Ge1B?7Q
zlP5Do2Bb#1o<Jlt^LITy4ULjT%^C|<*>l&>oaXet4eR+Gj5hJ%7d7(cd*5cu9Fj0<
zk}!T1H{x>}$QWAA!*Gs_cAJ3{6>946l)A^yX)0ijup>jAPdoP*(=&K>Ykv?4aJjee
zgD<UgWHiLoAsAvt(a&J?8s1XOVJix<teFqoZpjR7O4dH2$%AT+EInN0P^PXjP!??w
z#rTw)kDN<EploYY!{OgIg(tLTG~j*5wuwOFLWhHxfaC}n63&sw1}d=h9qB_Wy3<>!
z8||yq;M{YPlP|r}@7l7R9<!xJ%gd<Ves}GANg7Vm5Eb8>fDWd8`V|zQ3+2hYZ4`7P
zpd)dhz2&TFz^q~1@mM-gU+S%ipU8XL7PTJCP~kS6XmU7{H4$c*@nzc*I2_!@to%Se
zhh0yRg6tGNan@P5n|%7qe4zgGhoiD@*FnaGFl5nNZ-B1IX*3Ah)w?NLcTY5d$la9j
zI~t5Cq}1Z<76+a3dhHIgVCIDF!no6le0*xe>Eu~g+(t2q)SSEsIH*sy$+O~$L#6i7
zoqExPBP7PClC6v|R#LuXT~RPnN%cgs#`l(>ZJW1Dg~P(SabrB|r7-3zwpQUH{j13J
z&=9xJXoNO$G>e72CGHxl#R(NPY&D)Vf#T0>v8D!UAJ2EoWC5}K(1hwd2QeM<rcR90
zYJ_%Jpxoll;!;gthLRhP&GJ~|j;fkJTfvaWT$?@-zvp<vME&tPp>zYSc<(=eb)no&
zpE7<QHdfFD-e@1~USg2F7u$X!9`WTRzrjH(H>AhuLsakV*x+jYR77H|ihss<#`h`P
z!UKbH_lAKi7Tl2eBZG!jeK+z$Di?&XfmTf1Ms(5fo^>tjA>S;-;&jIx#9^@_go@zZ
z6QgYcYfPE(v8!}sZNd20u^pq02Rarmp5;Zv+D(d`&xr4{(??Z2DOvxX)F=P?V%Y}w
zZGdegk=)!8RCrpd3+q#w?x}duEE^?S$8YaK4zVVJ?DLMlhX<=NTS&2EXzG0TXZW2>
zp<gsz{3WpVd;gh9td|E`Y+)l;<LQ?1@xn8j-`x$=A}`#qFm(+oDunA+GVE6pZ9RSc
zvZ1|AlQ8R^@1SJ|05phrRBmWIzYgD29CLFWrocSsG|k)a<G8ZdD5yU{SZXf7%7YtA
z3=9u443}v0PdCN6Bvx<)h@qtDl|dU73QlLzMJ?}Wqkb%pxoJigk$V$qIXIuWr2GNF
z^DE&89>m*VHxfA`%8hT8=znk0pHpFq&~ZNQBU7{#6Aq}%RqbPTdJdflKcfN7Y8o<)
zZriL(bi~tQLxZgLHKmqAgU;Su3&#VpgT1Eh^|sV56VXutf<`u|Vhi$&C^{7~U-x*(
z7t*Jqs<^h2+cd|N&XW11oqv<y4O^ZNef=bCqf$1;fR>j0psT<zWb1;M`lik{7}w>^
zhgJ^he2@93<8Hb>&ju%&X0V45L5cOH#_VQ`e53gb0*`1nd;C73O_r$!8DW;zCVL~O
z;!6z6g^}&TeEG+A;=l+Q|0}PkF43_<pq0zD|K77tBSOvOF{s!#*zmrjd1^iJ06g{)
z%o=<Kyn6v$qi*=D<gO%NXDB;TU+9({ONILn;LSU|px3AuWkk(5;cRB|PIUqD`3iyS
zkXNWcRPlIJ3IWQ{v>2R{JFht`^Veh|t;9&<=eL|!e>$G896lD@nW@qKQEtQ=_>i0?
zH(;i>cj>cLG8|-wa(R$LmP_fKjk+GnO_k>^u1|${;I-(*!uu_8qZ5USVT*=O%nA_i
z|MKEqh$Cu?XhVI;4S+i;iLpecOPRJs=uP&>_)$KOj5<~-zige60`Ln{*MxbB&DPG^
zPa4MkouR>DrWkJR^(uwP{rkGLn*=`4!4T-zahBofAKRM9%=ERob2CKD_ql4i9bANh
zBs+9GDz=+NFT~OheG^|eXXg99Sy<OwF2`#jknyd-AXJ1j#r~=OSCYu#XUf+_q)3L+
zvV0TJE?3jQ5s22C`!&{&fJ=4^ZftbjD}<`Csl`_2+`Z_(q#}UlkkSR9eS#sSRRbxr
zE88}Rsrv1`(}8~g-BY^lJ?^q=cu&B7@G;W$<_Zrfv23UP2e4@`)`wKbNV1qdZmJ>Z
zv)7O^?EBO$gj9@h@M3c1^IwL40QiaxI{(jG>{^b=6A;$OQ(gR6)e_Zxlp$~p5BQho
ztZE7Q^**w={NJTw+Pem*c1^+e6gZ(*nE}w-sOvW#_#iyoJJIG+SchCyBYc`Z!tk!O
zaV@v)>?C>*S{{4QWe6O50>N3G*sb(nFsE$}_V>)lF8r8Ui-E4G`Gn`m#F!%ih?A;+
zIk(fMrnQJ&(@PcCXikpIzDTUXSBLD*Iyp}7B<cPH7IE8o(_o;S`kpk}ZxD`wRYx_%
zTX%TDO8_#LR1t&+TdTeG)e$wYFu0*{p<0_KCVy~e2~fZD{^nypSB7JphzcIYjM@~m
zwyEQZAmR!?gAA?NAvLZ*`in#2Wp69s)DEN$I#AcJVqY%xzV&Mjp?JjNN|6R%btt$d
z^<U@fgav(v2b~U_`P02d!ym+)8zEwc@-Sh8aWd%+UBmqZe&?vccAyN*fxn1f)mSxS
zB37O`A!W75$$*7y$FS+FqM+$O2rZBkaeg0gTYkU`Xs+LmueK{pP9wJznbXp&&y|iV
zVV?cup)>XcC(hy0c*l1@Y=&rhgl{mlk$X|-O<G~h@Rx%&#Z%33&`sG$X%s=82tT2^
z!XaL(ak_#FKWlYPnIfV-6X>?Pa7@*C4{gY2l+P^QYO2WO(Wzk|u@)>I;LS_03CH+b
zDXy*-PFTq7O8A51<U}5wM**IkTB5Gs91i!T1pzQY-&8mRXGjQ<8fWE~MSTP)kiw&f
z;>5Lm{>0v<-rw#W37aiooG00Y1@oB=9m|kfnG34EdVf8mNf?Fu&?<QA!q)n&CKxWN
zzIJfikaV1~y8p}ami8!PMlKR(r8(iKNw*3*-Yp~KMcBe`;}t-At!(L_b2O{`51?59
z=F=U_sPx9heFI$paDhx8*iCR;*f6Q|ZuSSJW&eWtu@{x8K#a(uq6Ebm=~Gof1H!^y
zqCw$)QTmjg*(<+y_lBqW`sNYZEP;6u?!(`caviK1w5QT&crLJ(d?kQwI*J%rv@Ox4
zV-zF8c<Rl`#hV=J?=N}A0aC|(4slm{uj}B$PA|U$hh%A%!l5}V)4@x=G^W=aWqD#H
z9Y<Q7zirsM2785Y6m8D3P{v)EP>LMpH8t_42tTlIVkyQkVR7jLIFb4*+1E}M`WP&n
zA`hBzJMu+uZ3RV=3>K{+ITCQpU(P<FgmRi%1JNy24nCpw?}yOLKwqOoF(9|1D!LI|
zh8`DoQqHG439C6`&OcGw@{j0W<B=64Oc?zGpfO3$cbJfVqvjWz0F(CS-S=RguAHH8
zc7sSlT=2nv_oC=$JRiNn2PJ6}ttERU7BQ}23iWR?k&zRF>eM7V_iD~~t<NMQEBBwq
zi}<R8#0N`h7&?lyL(%<O2aE<7AbHdz?g3%gGpJ(qvVI~9DP_Oe|AZ@oy$=A_Jd-U|
zG%KxGPTRiD^KWsu**)%cT)G1a2xHp{jb0PCpzMo3?82)&MPWfFlmi`Il*3kSiBXoF
zq=BT!>3yeRMsq|-AL!}xP78OkxtNrdqdk@U-Qes%n^XUQxbkf!kVL3`)pPwybuV4x
zX@okt7i_snTicSu<V5@Z1os7Zh`%lc%+4UCV3P&hh)bZX|9xyQoI(%=LTA(T&S`}#
zjcM#p@Azk%ey7LZCqh_a*j*9@kdy0X2U>>N<l(Oe_P<7$PPW@*Z1?F#FPV<+sTkWA
zB#)CF&-&>0g@z6Mc)mHiV^^Ca^nNB~VR=RK__s1=>$#*?{O@>O8Sx)^AhJETjsw@&
z<#WasBusW+IwJ#*!mTuf*zjJA?OI|ok2buS@59TninrZop|5AjYffaV<B)1S0paVM
z64%FNBc?MuvaQ)hs1*X2UpUflT0g22vex9l+O5-{WIPsC@pcLr8VNpy=Tq(;<@xsd
zmhMWKKWMQH1|XRsk<$gVAoN}`Y)3FH?>_OO4@!^!yvz4>DEun-I2HjlME3P>{ypsO
zvkbk7VlVWR(9WdUB^(DWyuWdpd%C2$9WQv~J3{z@ANMui0d0^yKqNF8gehL@+_9I2
zTtJZaQ45Wxm&3p{De&=9<^8VsS)K&O|8egZ4Bm2IOKteHbqbl6)mO9oXknBigGKh3
zAjgtI&Ci5gw6W{A$T-M<0RJh1bOpR+a>M5Wly&B&`Hy2n2G5}i=^z<?rn!%*?T%jn
zgshXo*cNEq&WAF1tkgyGXfJ4N-b|LR2{K8LWFI&ThkhIQ^;&PkwVu)rnZZu^6M!1c
z-2-?tTO9i~j^J%lXI?Bz3+1;$1G?-e%vVluI{!UF?*TwKL(DL{91X>@WBB0d9+0kq
zryh&A&=KeWn&?mS>|*`yAH8wPdhO()X{5jWa<qH-I~rxZXL)84y=zutv|Qm`fRqaR
zv~2XZG1i_=T7*~j7GqI+;*xS>2AYlVG6}k?DD)LSbJp-dGP|BT_|C_!A@qT$Iefu1
zy9j&*J9z1fIy^Svz#=?SQ`#eoc?>B#&>>BRN1wcTcvnEP?_uJ$`RPqh;jWevG}Gfr
zagguXmls9}joGpBI*?M_-v!MwcUDw%vGmisa^ADK^lSzwt?-t6Im(z7+-efpe27bj
zmHk|<plcTVsKe3fbeA<`Dl4{={Ok53_|oCqwP5_NK!xUk^N;)J$(3OWlL4Sf5meOT
zoIFZf?vb+$xwPwM>a-zO!L5rxe+=rP9-|@y?`3dr(g<&F?(Pt6gs^sGs)egoyo|+f
zkiOq3V?>vJqk03~WE!mGOYkNB!wk9r9yqUp4HADh-SCQhb~NshB)Olah#s{N>t!yz
zE4GYrw2PSNgqKNva)@uHH~%wYSkBZ^`&8b%DQM$iJY>&#3wHo6pd;8<B*|r)T=s%o
zEB^VI6p-TU%GF^|Awj_&w#i?wjx`{RvErh5<kYI-OoLV;#d&0u%p|-*@$-m9Km7p=
zqK<K2veEFjXnK6Szb1>l#Ur{>AJ_Q_afj9l)%#sZ7f8A0SC(&1@~KZ_^)25JLZVCt
z6d{SysUC?cLJyimn>ZtJD{W;_Ooed01XYY{sJr5+NTFpd$=5Gv-u>HIsGWp_S6%%P
zoZ%EfBLD>e=R)sEqI82eT`ApAEjRw5iXfm@xL|cH{BxVfnLO4j;$)~omfHw@Na|n^
zoF-a^nCBaw{QUQs2GlR-+cBV_*Q;Vb*@pH#&LOM-5@Znl>?2m?{Pd>mA0yoP8?`?j
z9g~>^#z{6(yRrjM%!g)f%HN~Vn4eD>NoW{@>&=Xo{V3(Kc3Gd7|Elmc;2)bz&$Gp4
zk{D5aXq;nTn<9K04?fN`0^d6^J>wNFTwj^yWqW^<BS<+4A1ITRAzJ_jM90xwhAAhm
z%A(QQ-u`+ZEq#(F+8Z~)N&^1)+HiScc|niVW<(`aDY1OJm9!FDIk3Li9{u__Kv{FA
zJ`*Th<s|;6I+d!mNzDF+0*bGzbX1AAv^rva!)WHM$>{nv<b@n%mTm=Um2XB;(wPhQ
z*q+fz)%jN}z7o-q<y)x>`mdeKvSmRnGoPX|3L`dIh?mPy+Rpv=e!;SM#qDDij$?}6
z%qV2Gzm8n;byyo-Z`XSv>Npns$*8=!&^0U(VoIYJc%MV>=mg}68<t5X@PN8BAO4Zo
zAdr>I3Ziv3%Kg=eWTMPOGQ4lD=$`-0ZLK>IfHOqciXh8s5e{tMG_1_9jLSx|G5#%c
zOF^tl8;c#cs_tS}4Q_o`U3y^?VCElmc{O6y=YC|A{<(6>ZIQ+MP;J%zyz=~$c(T<9
zP1T5nD<5@kT1K}Yw;Kp)EvZCmb;)<6d#y9#q08uW{c2TY>qBl4E;4Uh>rby}uMF>$
z`#a(2aJZT!juALGWr2jH{|U=gS8-z5FQ(Uz@X(y9#WJF1`a$kS8C@yJPrIQy1AtS4
zV=DI?y@vprUL#}R4rP^1tC4H}*b$;*OW|J?%rkKA$v#r7$KBtg;|Rgi@vlTCckUV3
z8-nwH4n+E^5RN?|{v8+;hpcJ%Z%x4eR+O8E0r!yCO#dRo33bIUACQ8^yNUmNj2&uc
zheQ7gj|vmTP*tm-<=U-HyX(C7G(QqHW9j!!QX)^Iup?o}mA4)9M|mfbe)dJ#2U94+
zP%51u*#-+ul#y0i<rkWP?2!c7Dt=>K;_ij9;`EB4q$6$k%W#erg6)6=ErZhuvE9tq
z{{VvJ#QUSO4VM@Hf(W-@1zR1FmQLXr$hddCe*k<H+1Q&3UM!w^74JAtoPS(X|D0>?
z!_2v;k#nS-b7Nr&`X0x}Xu8d`=(I8&Pv96;BldS$#&54+jAe08cuVO%ulgdxn@ipm
zeJjhSOzizM&k6kkjD6>m`!bq(BAO?o!eosg9q)+?y}i-hIg*C{xmEyi75mJ>zSfQZ
zrX^-Ui?@3sS%(5Ii{tQB`X+h0)R1OteO7(;(C5CC56_2(JZlsV-raAS5;rhN$a(EG
zKsJ#jCSQa`_<J9WZ}8;!CUYiT#f8?^2#14o<mbHw(OVyW?jZTx?=mu)MJ{`DNQssz
z%irVLvWMLT^|)a>cHoHlI{3{;x~oZ1b6=NkQ?WUFU^ND{2FyJ2a~n(J<rvS1b4RBb
zX)k_$*y)7FGGKg!F{ob+CLr0R+AMy-)(?|WUZ!nsyzkNda9sQDXDvB~0;1cCExJC+
zV$HicDw1@@Bjb?$0aHGg)T^y0hs>!N<JB4Sf1i|&G+ss=4R^!u{c~L^XiRBNP~%#s
z<T3G3XlnR_fA((<Zj~M>zhXB-?n1f>Tv7CO<03-n*fIZVrJ0SR=2__z^ta53fVO!3
zuTNm!Z-_<W3$jNNfu!i_R!i4Vo%p?^KVCgA*lNphQ!SA_od}BIqHTdycZI9MXm}px
z33wzvByF8EUvG%U`CxjIHjcQr&wR$0P{>-ne=Fp59*S|+=_Pl30k}fOCUbj~`_Lhc
z4Cj@Qvg-Tt@{R~JD^-3meB{2{jPG*Cob*aE>>8=?|7zwTkM&00K;fl2`|VlWObLOZ
z;a~`c*W*bJQ3#u~%O2l+ic-pHXXGZ;In8Xgml<`8hYG;$3AY+JN03oPL#@lOBmN7T
z@v_N&(-~B#Vfa(hA$Os3Ww=lnk1hF!CNd23*U8xxZ!Wjr4~WQihGUc3=X^18vs}dl
zY!Br`WU!Iu2tXc#q}chs=@$vK+{{FxeBrgrE6N24wL=wFWGoZjkeV8o+IBLFl_T5E
zvlpW>;boGx)lbI7CaA2e%oZ*OeNo7^O<K9$ndTyFcQ;M%`YgP;GPgv*YTK_Wdr(vk
z_!m}dGG#gAIp9>-xiZLqtLF)0wbl*qlY7_E^68J}Wa0fUSAWx}dY3n&NRN;7U!bj_
z;DUX}0x`bvYb%fgt{SSW&)W;`JKcE;t}~~vsw8KSaAo1(D=h85ViA&Ft>0}Sy{}=^
z7r2`)M1pm6C+$Y<0st>qAdiEY)?dQppPKaHg0i}{1)LG1L*%TkCHpMDugNALz8Sab
zX&8<5=th(vxfRo%$R}=7x>N>6t+I!XJ8{UAsc(MH&(ntH@TYR+%)lbc>DKfygb-z_
zY?_R#k7aU?x0;Q+=?2gAePXEC{=t`F)D<S@7a5xgRZGysNkB<D8FClnUD?OJDF-ME
zC(LRfWi@!WY5=THGMGgEtoSc!uZRgpZ+WALN<KVFaf)o#<Vf$n2x8)_Gm=}TGgZ$R
zntneVB6MwXwbhi<8y72>GD`2<wCL~v^8g!U7wUetTm{U|VQz&FJhOyM;|~s2FTWnl
z{o*+B``Ar)`GMg^S`hp{yuF!gwU<YDzH>UWS4%O%Uaj9OS81)&|JH6V<9nKYr*Vg)
z3ckYOU9=d&^EK?QRn4sb5&fbHdfbtf9kb&CauvL{5_9&+z7rrU*;vC`>mh<WZaUV@
zl)DsBGctImi_{o~!UWb!nO{?}`zJ}nIrMGx<`vF&5rz4or)`e1*vk~wNX7kT3TvMM
z;8}T^b=&TU3NN;wgQJI=bLX!Qlr@ZGZ=&d)1_|H$3~n<yqkNHS9BZTvlfY9z;*js8
zDaNa2ylF3ognxgsiBj;hHLokB<Ddn_w{n6p!ZM;-2`k?``jLH!@TF-A2o;-R6&Li=
zYv?f|ayl5;@pj7^$I$dDG*&2GF#gqO8^PFnKxeT()bi2Lo>pxAwSWKxmimkceWRM2
z2+<9b7w=eZ3z^ggW&!!9QAwMxlto}3lsS!m>D+;5kNV{H{-)2tNM+`IAC0vw^2B6g
z%s-+X_-C<HiU(weZBHtL`MTtJL$Y8#RED~sM}sqQS}{RLliSWwQ_kR}sH|NtwhF-0
z|H~tP1g~I6oNG1sm-QGT{6Pm-IXZcIws^wia{$Kz^uc<$A=OPWjE;<FvAH_SID1U3
zF-3UEO>Dg@H3Ua74|;gMf73*4zZBoGWyBf2Gw9Ei0rCS2O|$QIv~msiq=R>ql~I4E
z(4MovSy(7NoNA#HPYi6+FZMJn35fe8=P0k06eg3@XEGL7#^Q;y`(WdaXa={AR-TV!
z?rV^JU9opl&gHYb;DH2L!l}xwBO{yIUk;=~4LeBFvF>PYeRbH<PC|tNsmJHujoK#E
zYbc*k@=Dn5;I4dIZSoTo(xafCm5>zEER^|)On&wRyTl<ef|!k@7l>+P)lEWN`a7w!
zv`A^8(J30LyGR$*kF!dh?=ZjjVZXPYN8Sx0`&@wa>G`6lqloFNz1pVKJzAYk{|-A6
z`D_1)v|FNvO22-~d8F|}3}k&}rADu&D?SP_`T)Syoephx)0F!#{{`vO!&B}YD)2JL
zSMW^coP#5UUF*8YjHQn`c8p{xr_^6tff${Qr1Pp6tzupN`(-?~quP*Gnsu-oSxP)s
zBIoY^#n@X$#nD9F!UF_I2of~7yUXAX!QI{6-9rWmZUKS~9^Bm_5Zv9}U4lEg{k-eD
z@47#~f3sHCRQIZxt~zI*efD-RxvJhS;cN+&L_8<Xn(hA{gXtD_h-O>jg^{04wp#zb
zN){Cj3&sIaVa=bu*ZfQQs3@uVq)>>4@l{wxX4;mNZ~gS%5Bk82EI;LU<pgn@QRw4z
zc%C?dO7iN_GN7P2gmrUAx*v+M=CMX8Q(?BBT7OJ@taVUIh09J(QApJt-qzR(6-p8r
z?>}TT{9sUWE#s18cGQGDrngeMEuQ@Y2fRzcuJT(IW<mf!ZSnonR|jkI%H>{Mn_whJ
z`S@>UZl2ZaX3Y6(TKCTe<nLfhwK-Nz9DZq^cI|kt0zsSi^Ng`$4;q{s4yhTm%J0+r
zMYsaX_xlG*>*zCBi@1NNFOTGZDPbDv8z~g|7m95&eEBl?T$b2pCStI=h^D4UKUiAp
znzWQl|L78ewK#z_I>Im=eJz5K=Lx=57zG`BW5M|Kl3d}nfs830jeQ*+Aq7%y*{A99
zghW@6(+1!)VTm+aP@}f|;ejnHgVbm{6~!>#trkCD6db@IhG!;}<T~o6t}y0LqmnK)
zOWxjOiWJ|BbT@{k(=^YRyge5PvXs6a1{_1nkcSg1Lhj<%2>?Hc^X^%#I-*h=E@;E=
zmF6awE9oNS&KOtjZ!oS0rwj4br<AIw^jF`=lVPC+;aYK$ofIX7+09%NThC4i$7e_f
zNfC-l2L^>HCJqHF;T5Lu5%sT>c61H|#9wnstR!@a{A&8yeFwMSsJ%0p?AM3a!ZJyF
zHgBrfn#U^RoE#zZYuw8g|3F^9(Dotdib>p((WLjjooRiVS`K%NYiM;xT$*e6@~hH9
z?X)kUkWCFuxKOOUziD#YO}Q0Lthg%K<AtsAJxQB^V=i)T9yRgE-EskA8PqP3)AcES
z9=jR0<@;b8on3{jX4^)CkV%>n@|LA~vDZ%D4{v9db$g|<Ov}H`@!rVBB>rH2@xN5S
z=$j<4ZMPEOva?WR=a^M>QmKeuN_1OgaK-bfMAHZn8Z9nj!4r6jyy&h=s1<y^6a5F$
z0@$kyeVtdnPY6%j7VSS7Ye577xC=hxgF-XHFZU^L&tU&T$m0de>z(zd|NZm~z@*pf
zM$e~{Fn4Ky&Q*Ko{Zif;2ap{2KrZS;5SIZJp)8=oa>KIbMFHZ2k}pZn7C!V9qZWPf
zc@mxl#a3&^sCCTGjr`GktH6{m&iCls|GgXN|K7w_PcH&`{Aos7)cHN5&whY_(H&4>
z`0>Axjs2?U_Alvaer&4fI?G?IlMKjolFO1)x(Lh3SO-ufKeXm(`|=mF3YA1mydoC`
zylNAuZFF^^468xAD*bsPQ3%N0YoldKUU*eo`Bl$7><O*Ax=p=zzVhh(#_`~*O^vix
zNuNr*>w;Qyy{W^#h2URrPTliky%wm>Y8`R@>{Z!n5%pa7FXV}<Ael|G4zD+?zsp@X
z$G74NbILbHZygrlB_jmPg^S*9vim8~4RCfvQPzr7>3bkwO{bnkky&O%R}GO$?=lsq
zmMJFNd|$8?d4ahO1p|4g+Mjn?WQ#6_w10gEi4WumjLIYyfnz_JJ5o{x3#f&^bFY#@
zQ%lpya->e7tZXz!tDw-58J6tXN62$o4`i=jV{rT8Y+prnUVS#vVk{gFE~s%{O<q7L
zG@`kecm4Jy?iE^6@^P?4lqsHQuFG;Llw4-$<c(ADe9trF(-(;FWrccs1bzL<z&RO^
z&Ha$i-#PsYU^E*$N@<?h5xT~pUN%4fi_0=+L&o@eH@&{W0mx-unkP{tU;^DI|LOAW
z{$WM}cv~mlLjVKvG*8%F)d<tBrZ9#<;4ZZ(Y(qA7ziA%{e2LKxq}Wg49P>%6hX0_O
z^Ip7xe@SEsnMCmb1HmmR#_W|vpT;3BNS70c+4H$jcVr|{bj?%XZ;StRq&hsyoD@5V
zq!8_4nyEWD(FqOsjvI3iVH1BW&if9TgBM$gpM)1dgl2&g!WD!g)gK5;$67X{j)LDf
z8C@AR!^sDoz9Ihi0PqvMy5i5_jai6Bmj%r~EL`5{&<Oqm^#e~%Krz3sy^cGNJKlSL
z@;Gmw!q6u4#fbaAt7U1P=uGXu%^bAJ4_<-An6Dwr)ReQ1Oh)nr3hsm_)dcHw(|a{R
z!?=X+b-co_q>+6W1j!Qk-P%al2C&lGn#uG11lA_5s$6Z?8V*u-i?250vP~||ZBP<c
z7>?P4NmuNP=(B$Bx5{ZCfm8x`K+0>qw%z$G1zOE8KVxr6WA#)Nnn+^kdx_^I&*7O1
zux>Sm7lD_#70^Xf--PnpM8vx)iK_?sC67)ic2XceshJrnuuGkH<)Z;PK#W#>!*v6L
zbR5B+#KdeQo{Cn2C`33D;zq{1!NBG3YHJsq$u)99bB)!8?RPDF;>NDF(4mrnnFX}v
zuqPC+^%+DPBUy_tB$v&vNw6~AZdV@h#jGdsWOa*0^Hq9tR?ugSYpr?EW9)eeew26!
zvEXx<&GYjKyJTjCp#7~(Ug(7ybR#H;p&6wnarQ(EmC8KJ6SNyiO<sgp*hxEFMASza
zWnTQ6_+?QRbBP!?|Ha$g+&(^-L4AweDlDb3&t{IaEL=gCCJWwQO*4`q;=>TT0?oa$
z>%bL4-p}7v6~!W-Cw`FA9!5=~<>0=QmAbknCh}_RVYPLhTvN?e;_Scq00Ow`1m<Q_
zSSAn=`q~U;Dc!M(cWe7g-^OZ1X?V)6*{s`ZM!2#21joZqRkhXE7+r4Q68ND_lr}ii
znXwFXSTVQ=YwHi30{JB|oy*wcW6CGzqBPUB18UOdms|__zT27Z>jt~_e3)ZI666Ln
zAH5i@N;va37^7tMem2s{OaG=9j^wHmX#a7Zidwq#!l#w{otJ^+%iGXH(1yP02Uu%V
zM8h~E@s7U47OTdeHM^Oeuf7+&sT*7aZ~nl!^BeMX<I^a7eZume7T!XR=sxiCPn37w
z1Nnww-2XEO^#6>?8><088Xd?*_J3Bp4skC}K*#%6;9xywNWE!G`Y0G3o3jfm{B#3D
z)X*63K;{j;m!x<XxN==4^#2S#1YO%h4(5%X4*--#wr+_6K~F~%krYr^JDEfb{Wizk
zgfH}SlNI4251HiF^0(_%b8+Hq^C{b}VkpIYg2g93vj)~mIu7eou7GDH@HAE$1&>83
zdtB-q9*T<W`p%I8Iw_G4@KDnOZ?$O@saJY`%X|%*vOV5O#c_5g9&(6mS5AXD>0X}k
zs0lxuuLaz|NFu%o>d?f?*c(WBx!~S$e)K4`q(XMChbh{6E|Vjuaq<569Zv&JV1Df&
z#xdN)%CUy!K0Q%eF;1Y0DK=bVHzq!eE7!%J6nd;_HDM(CRc&oWI$_19CU&$4uR>i_
zlVd;8i_f2BfIIm?ZcrYO+s5Ub(tWs4`1&E)I`wvV7ip&?<OHrUi%xf%Gxw4!o_){B
zlz0@d1dFAbcFc>_<BdPBn@9nibAQF~#X~1Oiav-QySQup^2w7Plk5(Xy+%u@2iMg}
z01#SX;;*=zciH=N>{~5P3jR~;MfW+B-;mL6N~s=7j=5{;SB3cIK7YO0F2NuVFcr$V
z)HDg$Y{4spJ3CuGKBmn(q4-X`e%rBK;kxf&Hx@T|Wr%x+MkaJVb*4JE)K0ynuu>)|
z@RT(9*tn8(`z!!?Q4oJEa{=@$&y-%<q=3@?)NsUI?bUO<?}A#IdYObx`wrac6yWS(
zEm|VOR8|r+`KZtg$;;X4wB;3F{ylOok?T7ehBF^VL2@i1YE^<VfvLOfq$o)NXrPLP
z`x#M6l@I3PIGsLA9w5?hOqEepksrNgn)iB+2K)mZy6cS!{(94R_y<b;<Ntw#E6rx$
zSru~WKGCF?{TRCP4UOg>2wNiC8L&<LZ%~@$$8>`nKrBQraIS^|Agli_(L5iQxhLWx
zZRvX50d`&-|DG)2k;eq^odcOm5zV)RriUpz7mN0yJD=4xHntOBuKlp9H>zUuT8!?(
zVTA$8Kw-#mLBe*QrK<h@=1x=~umrP9F~cMy!2P=DY*T!{E>NwHpHO<X7y(rFu$lJ*
zOI3)k?R5V@OY{xcM|Tj&3+!9gvMvn0lshgV%znwl8@~Z+3~{FG7PjNN@-@7axas63
zV4!7olw<kWDILqQbMa?-oDzjR9RZgKmKFG)-(e3w^A3*lo+z!;(JS2Yy!x&)5V<}L
zkL6pFW$vgke>73#Eq2%#i`O!Jl=U^M;H<%vWw#-4(H}rU4pB1Zo4-qP@^iyYV*2g^
znUGl|ec_mgeo1FB1EZJ8Yaw$a9jS1Zg5mFty=b5c9|$cg*<3|PH*c|$W72&B-_FQe
zf8L`1tP#FB7rvn`*GrHu0#7QrQc)>NAynzDLy$W&+2v$i*IenEFv`txX_f;>@QlNI
zjadgd>$cWC{H*#f7}MC^4w+u*1fz;BtJDOpwH#}~D<pZ@FTR0NR<9l$i$tq)HWPpt
zej0KBEn<NHelDgYhnB;`i3ra102UqTvoc&Jj_7qUc`8#bVSuFXX1oR-&nbUB0`0<s
ztKhgz?eEWU0d%~){EZCJ4b&?#%`L(wP6R_r*0VtampmkaMTfydpHsr#w0kwso}I4Z
z262Q#%%3tpN|8`OW(!uZb`P!NGrSmkxF)$`{TUR*`y6!!E^wV0LJJS%Jt-r}>^`xK
zjt$+Frl@hSzGn~4I4CJ^vP`G&T|{h*`T4pP36*Z@;#gb|sJARLxd(R8Up0a$oGG~P
z0(=9rD?XZdP-*-@oSoHep*InXL0@$;b>J6XCJ4JkeoHE2+tS;>DrlbX<F7UA(7cT)
zn{X;p@uvt7_Ab!Y`i4&E(Igc8nyBY`;tj;i|5H1i|0;0_>%gKi%Vbhp>r`D|MPXq0
zNNnyhfEiCR=q3Gi6@Js1cfT!ue9RxQujT}RD0>LS9GXHUu3e`dIm%^m%_y85))z0d
z?}+(kx7wu}&WP7}vUGB<8JVM3klhAgEKy6Yl7F}l4cMAe#%+6MY`GNJ5iCzRVQXTU
zYV>QkmP>B8@BBUKllk0hCl1cC%=X&lP9{y@iuoRU+274^y~$<X|LuaX0i@Obkm7{p
zC5ghdEkhwT7&7J36t*rifGqMThdFDmC0A@$<z$CEWOv`b;E^CPs6CE~bPgY<%yIi?
zJAkng(EM51Ft*OS(&e&9G9OxPlTl^+(_?kWvED!n`p-rV?x1S9?oG7S5^`|y4`j12
zffVdu#lBPRqf?m5v0U3?&f;{YvA!{@L9J^c_kEc@&$5yqlg33kUeSE8l%oZ8m6&eD
zyA%_DB=AfD(Js5-U984xm*=sap8V@#YcsIitc^kyS8=cJv%3=)i@%CH2`RAu5$@QR
zdu-fzO(@2U{G<eFyTVnFR00p-Ko5AZV8G~UcpA!em|bwZl@k(wi$D1nsq-s;T{mj#
zFw-rfuU%b_=FMF1SNUYCd)$X?`-A^GQXUDD?{{B?v|5}oB-~kJWcNzjE0>y?u*jsP
zVY@7}1e6(@vEOV%(YG%Q32Cy@-Ct<CuHIKha$-yDn;0erU*Y}WgOQXu(aSiMmHp8C
zb3W$_M%>THx@-o;nz*+5gf~K#w_zLhtvco-#8(cd`|WsPFlW|4%P&BNjdr31!uCKB
zS!t7CP#remG5DySV3(UbhcoJqOYp5QfZpb=!hvXPLYV$h<sYd11#-wAgWnF5qR6mL
zNb)BY5~8GLRO58w#y{Nah-<+SS0LqNZV=&!<@TU~hupkc(Uv<7uUnJ2Hz@q|$1IFz
z{|-vSQxn|ZqOowNZD)iJq>~6~N_>c7aLDFb1f(m@qHnj9aJyn~P4?Sm#Y26pRH|~w
z?uUZMdpSla6qHuHIUnAWnmc<QV9g~CTL?G_9<k*c1O)mZ`0hN;uta=D$zq)iGh@E2
zV`+rfG-{`hV0uKZM7COQRYFz3G?ZkHqUK}WPElEGC>=U5lCHQ85J6jfUXUv>`O01J
zr6Sa3XVVvhLmWt}t^`(k<MwT5WckK%%x*2-0s6oWktxbprNx_V?j%ZTVv8j&_A9|k
zthFsdYQ*9^qhWvF@Xx^fEylP3Fk~kjmG>_njx^Jd8@@8*rBxDb2=~t3nh<A+IZy?z
z%9`EgMi6_HLI!5kH4p67vhVEJBbNiI`O*9!ubHLk{qF(4r69kbS;DD14N|c-3HnXc
z>}_#_-V+5f9IU1)UM|5R@THj<+1*Vq0-j?o)MthKHcI;;kULu?)1w#Zyo|Y#f&Oe`
z;U&l`@?;-T_U-Jii4fjWrTfzOba4p2EAo`fVjXUuez=R2at)VA!MhpU&BX$gtQ){#
za=vV^3LJq<!9M&4%Qvx)raGammZ)TmVD(X%%i(NR9R2Mtk=`Y*aEu)5GgpiqnpCr!
z^FPpeT$1pe+rz9UcgHI9trhRo=R`J4@TiVLO(ujJL70v3TDMjPOP(_Ak6=exP-T^P
zEdicyZ|xa|__f%YHetprzFI{5T%4e!zm48nvj<<9a^W~D4{5Z<ZVyYH7FNu*jB|R$
z`=cxT0}7^qo+|Jr2AHV}%DQ;aT<qZ;wM+qXulxO$V9_z-*(&^=horz}9OCg1Nxa!e
zcPtO1c}@Y@GjuAhq-JF3M4a8NycJ4#jepw*;wS?k98Fx9^x9Z*a*3)C!pq|QIyr&V
zNC`sO5$+2>@VzuBJ}^NU&xD`7XVszf;@BBWqR{YRn{gq`@SEjcj!koNydE*G#7c0%
zNK7_X+V3JMMo5ir<{X2)+alu=!}0z-`tc7Qu@)dVe0QX~a+6In{5BRgyi#YCknVZ5
z67A~%^5@Ax7zEpx#qxs7tYQV*uye#d!@<EdBEl^faqIfT@{A?e$Gr3!F*#RcoWs=&
z(FomuY7Di-S<<n84N$M8&vPO=75dXd&b%QlG)Djrrd@b7@h0!tkmOwENg3v~3hkcq
zm2S#5UWS}h3O*c;CkGATwLIH0r6BtW2<idVl6o}@Db9>37mL(YuM$_b`5KGdFvJRR
zA<&uFY9969VD-k`1+m4k1adbucKC>hn_LVl1k-Q2zIFdVI4H`g6%!32J1CLwbVED~
zmDF?*?{!&8`dqjmo3z4xaeFdJ#N`U=r*gReKy8ji*HoPqmNmKdn5!oFh0-q388c=G
zG9rc@v~bg31@l*QzQkR%_qy)hJgL2C6lYWO6Ic@4zYr1!clg|U&>Lih{G?mLHN=ag
zTF8Z;r6=6YFps>&&3-N-Po~+HFsY9aa9G^IPN{hKHlY#z7)g;*A6#Mfp}8&3W&i*}
z^osS;<%S$a5U&<1w<hO&ZxdTvb~&w#^ZVGf|JyQE-Ag<u_EvTJ{tI&_Asa@*Z)0*L
zE=$Sv5nRWO-*2wfNpg8OCI6--&Hn9*m$)Tal5|!X5cHkhU(F1ZqDj0dl$o;jG?Yyv
zlH@~*J@<h6SI6&FDM5Uh;h(25`j?sE&<|Me1rc{S9_H|V44GN6z2PJX4+aAsE9%hS
zWg{<PQ7)WmUB`>tWh0nh#6C&dMn{hvF6r%asu2E)X?GL(jGMX_(;AID7qPd7pS*iM
z269}K{slfL`SVhi0bTLTABpA2E}(?tL}v6hWS`6GaFPG0EU7D^Bymun(-ij^_vsA`
zBiKVfxQc8BTepiBf0@M#NUh~)GfoY~wZc42AAcbTl9Gf8Axo<R$<P$i<H{!Y+?1l%
z*`@~oKCUMQ2r-VT-P=3EsF--~tcnX9M8+CAG`*xG)-sR*L7?Lc#~gg73`FOw)nudr
zUKp9dmtsF`v<PCa5|-3E*Q_{u00s<r>3n8VC~A>HU?#{B%9h3X!xE7`zspSQ1Da4B
zfF9K=eOj^k2jXm~PXSE9Yx4s&Fm#)tO%YWMctU>KIlL#7fr^5U<wv(49~{=c%~+$i
z=DQ&X|0o2<njULHRmbK-_#G^tH2p=&zKVn?2;cMoX@D=Fc&g<38N@c>TS9*JjV{5k
zUWuKM*hK|T7lD3hCpQO?YB+HMq0!w(WaieYi(*iH@KiqP4I!*X{f(vbiVgv>!>~j6
zknfxDK|18nK_Mr0;_28zpLZNyvMpfbCihc6!#SrZo3!~cTVxX=n3TFG&#XZil~m0o
zl9`$_4y&4bDW|35jgnYh_;%_K-pgl~r|m)|v7@hAr&cRnR})xl$ZgdU#QRltYK*PX
zzg0%rV357fOSys{c7$V?^K?bi%6$rA0_>p#R5%-|=#n^8pnFb9xuT?dj8T6Dw`PXb
z^(P#r3(JTAT+&<I6J4LiAHnE2GqKXb?%78=!+eLj{^JRY!iy@YC<9rjw)-L?SYL<P
z8730ABOJ74oCZ$@4)l9_B?Q7dKA}I%6!PL!n6rLb(*lYoEb4-F@iPP=D7GKfgu3Y3
z8u4~0p)<`Udt8DNG|fzrx;dkiPm*&7e6bQUB{$n`)&7dI&)G8H2LFNFfQrKZNAt6*
z;IeU~^sxRrKml;SL%b*uBwiC;UJn5f>J3{>ebTYA+#;0wJHHu0>Q1rH&t9{mb<|mh
zDpRNolWgu#Pa`DB9kRqtvL1IWyanuU3Iq3fFB1<c|2qd32x7hp@f98gCMWs-J5Xpt
zK8tt9)B11uhd)A{vY@}6%O`K5d|Ln2^5`F1F+}i+<E3kBf6%@8TW;%`a*o4mX6?k*
z)7r8p{=OEAoJ^L(g{xq?gk8BTjNr&qsrQ{tEyetFbENX+xB2hxK%7gh29v!Zcp!pI
z{#2TMA2$lo9N3a8k=wG5GuTn4hXnD=T!}DnlmL6-Lux-b4{9Dkww;>ke+X+r`F%@Q
z%s)mqR-sL171x7zKS8?ecg_d{t1}*=S!{#t$C1Mn^tI8vNwxwD@<f-#MrdPUGwhqQ
z@OplVYDc{1kPm;H7q}zC53WqGQhilsKe>EVmzL6GJMlyWWmzYHaCbY#rE}ZOBd#?8
zi$r1*LT?~v^pPxR$9$^!4F?4ffthJiZI7!R-MIy~#+{~3W%*S2u?4nNYyU+oP@!5W
zTT3o{_dF0vdRGdyXlZMU70NOTP_n>kB-B`oNiWokim)IJ7}zzEAVixkTa^O{13!u-
zRX5|t0N?;qRpWexa(XBh`IdJ(V!=H(_0>%?oV@}42jCJELY$PWg%alrN_X)l5Y&QZ
z5wf3~b>ZcF<TF1t=Z|{_Y<=nT($ro9jNT>32-NwzgJo~yN;?{6&m*^`aNqVhBd3Vo
zPkdG9qLf~m5V3|Os7g<q;VGoIm=Hh$Wf_{N3Wesa9f*DU2P%(Cx+ngbI%T*8@kFIB
zjaEaiDEy4fMbfpPQn>NxGT7O5)I@TI1p3qHzg+lU-)UTGzNz6SgLibh>9Q87`*Sqn
zHC9&Nz_t`caUxZ0hF0N#($El)>xVGq*)MnU6Q`Tu(Ol-fyMK<sZfvz2?!lsYgEsEJ
zlsT?X3ivsBd|W~hp6lkSP)&$nF;<`+<Rz!uJ$P8*EVOEnGkZVdphf}H3t@e+)lYxt
zx?CP)mq!aDvlV~~n$r|Mbh9if9)>1M6sAIs02pE_PK9nYATl3wl_aUu&7)8zufPTU
zY4)CL>oEs;B+5UHf)k=ZE(rS=W#-?@7Jtlo@Ps7a>~sbB0cQN{ZcbvRc|ip`prqF0
zqwwcH!7cU0yt=p6v^_a41qmPO*+o9+mzY3*+vQ6U47%-#o_qg85p5#AYe|kXb+~|A
zFk4rC?;s`E4yD!`e7kcx(rV<R2j5B706|J=$#;)}EZpXDg$tBewL3j}gtunjO$>LI
zApR{cv{!EW@Rbf?H2)n#kk!fat|ak&t~6496XnbnVqjyfA3pRWT0)T4^X}JQrFD4b
zw+Gk7*YpEcW@L?ZO`%Gu#Jx8a2u^L}L@P@B3=>)9YxdSUpC~K@IrR5ya3i0b?y|wT
zEvkzSt0DG_UX1W|{o;M~PiEIx2i3nUbf{OFXP$GS=YQ@p{T51MI_9jAQ~$a&?swW%
zB`}-rF@IrV@E}yg<Ub+uV0TG=-Q~PdkQDvd$j;o7o!6kvZH7em46^mBA#X+!h=4jL
z44PPLOi7jz^%LKW=d<r0k8g|!Zwul06pH_W8mp9#wHdh}v8}6Sd(%v4uZ_MV*>@2O
z<n~-qR{ubdtPHDB26RY|)QQmz{Qt}*4Y46Ihm7J(ya5_Y+Bgmn>ljPOWAIYkv;D2h
zZ^08%t7r0=|E^K@#{*;zYe5UhJ{oXhLbCGyfkMOPA%`gcGjUX8H_0-E?pdA=P+~jC
zC#hF#R1~f$2ee+Vr<`j>UcNlU`|orJO)uYde>8Dof#P#pGP*>&>5CP@%i)U&k7%sR
zE34@0Omt|mC(Z}W1&<a*l>9j5wc1bfV7>vl=VA<MRUdG&2MHcWQ2e}%OkQ%%6Fh`{
z%Y5J58N~e@y#uL@RXT#&pD(10{{!`bzrR_gsJ;zCSA3uyTSl#x_Wm4w%8;L?A=O2f
z?7|(2{d5TuWFwk|VFf}(yc+!I->%ii^=n#`<G|$}++24;BsFXAOy!y$m78dU8sfIb
zxSzGugzd)X9>Z^WFZBPW@48vZ|7w{<KN4Q@tkVlO50Q;!i%kkH@n#}nvLF+GotdsK
z(&@{@rlZSCoONI6;JnlC*Q&O6wvBaVn2i3JG<6^gIOHkYbj?tTZGF3k&A=ldP}2Q?
zKfsCk8wF%A)#b0a;I)Q<AS^ocrjZ0lXjq!1m_ZHXo@t9)Xd<T;(9>0bWzUKr%#ILI
zy3Q+J=muV<J#<XCEc#u?gN#6l-mt!pql+;FuR8|BNw$riUYJeFzg4q6PoCf&rxWQ3
zR)=SqgMxCaJ4;IMYVmH`nrF@Ir+5Hc<iI!6&=8cjYfU`O_Zpuo=}b}Fn%!d*bz)Sg
zh$JTC2R_a|L5G)h{vzsX!Rla+xE3|{Nfv!<nrUFAsrl0yjk-iR`_zNe9EW%wX&}=7
zBgK;@@rzwcfMI>ttX>@Z0+HusWW@*b76M;%5u9ZTN^!y@ff>E`GPgPJu6g($5et-2
zHQgpi|61em%U1I4Ea%0*yd1*bQ1;bV36_cRH)#KgdBW(q+`$6UZW@pFqgcC~Sq~Df
z<-fb1^$c2w0zF7(CkKee{Zk<(A|gX_5s3R`Z&=9{Wc{O+{<LCQ`)85i3yiDJ^A#1t
z^`$k{lx_pIVtxeF$67JspIqJsOLdcu5;n{vgPkPduwbQFzMU}jB;l^Y`J?$xwAkxb
z-{zPaRGtWeAu~WWH&vy+HO#9v4gGMgs6}49icF|R*-_Z|nfNxJ7Yhcvmt2D9M0s86
z4uaOz=4{axUw2yE`J$3P5<2$y(Tw?_M(xBJe?e)&hz^)5uz^j=(!6|(WQ2z2(BHN_
z+EBj9*F1R(SGW58G93&#@?7ylE(~9`t~aL$B4P#G5i9k~ofJ654Y|iU%MI@3i*wmS
z4cv9PoR^QG%pO&iB;%-Wg754t)6u&aztCr{Kn!=k68WOn-bbxlrbo?pSKHq)4?8Qj
zS7J6e+*D(+Yx>M+O7fKt&1jPM8^O=;64wW~qLak8Kl82QWzQwEwrXZ_O3D*>2)<^s
z5g~TSH&!!^4bgOmIdwFFREi90+sJl_UGyvY@TuKZY9Fl-FO1QKYj3?@Mw$nodq4*r
z7Vsqwhia|l(tq0nHj(BUOzt}ZUTyPf&sD<%ixu9+CA3@Zf5qLE;?Ms6&du<9EwN=)
zSoXR@6lokFV4-1e1p8$)JYn`>a=$~@o%aN7#v7<6#gE(TEtjI=RM3O?v;>Ex@vuYR
zV<(nqGn9}??J2U`1oPSz<0J4AYFJj*pNCsi2$>KV&-WO>c37LqmoVncqTU2xi1^Ml
z1uW&mFXpS(3v?-KiIva9>E==Pe<#OrwP`7>)|a1-t?#8B0#^;d2AGy6E(nK6>%n!r
z`*GkG*_valj1Zgt+|0r(+=2t-aXt@kdZmIyUdN5bH`gh`naul9r@ww){FeH5DpNX^
z3?5y)QM_5Ah5hYQ#-=}p_Ee(7F){2SuC;2jV!G7=m5SA&@{0Gx;q4QefAiCc(i@*5
z-X(4jD4l1LS>G=gBA-h<H*zH2#BU;ry7hi)ces@u+S(k$A-4`wzhpWuur+SJYQ#(o
z*{{CV6+=QA9CmZ_GAj`+BQ@{E9+zYQ^CGvo^}$LMACSqoJt$Xn)LNq#6%bXpu)JIc
zi9g14K69GNrI7l%$dy1W-l@1H>iK4`)%^f@trQvExdM3<vYn;!^^V##d3Jv`aUTBW
z`>y+6y(y;+v|eAQ5^u1J$_#Sqn;Ntig*G@_4)u!O_!|=?D$;=;A@-#~e=6)r=vPra
z{YGNd!;ER@&XyZ_lG;L$KZ7PGq^`7>6m`9Rh+fHF330!4awVHJ!=GMcVwTFN-UQV%
zOW{k*wa3MnGLAjB;H)VXqu-`x#duVS0Sr~c&np&%;qQ@riwY|)KOW?CQ)5~k1_0Bb
z81Ak83X1dW;YPgU0m^T%q0=Jiegqia2p<V-C^6&&a;%83rj>BgolM@{eG*R_NCL~Z
zuvKq1H33e8s)~DymNhm`#IdK5Wx+p0`t80ut&fBp(nUnKJ^KEYQ2pWb*w3-5OqTm<
zt6`~@v>(98G)1IEoh*~qNN9!v)x<Nj`Fqw>RiR`lJ;RA3I1heV5eAm8_~Jgw?%KLZ
z<R55+cgN+1?9W@0okDHf?}4Q}q~NlaE-X?%3f~D4?OmzaN*VFjKYb2tGsXr`cV?cw
zj9V}};=#R7d56s1m)<8;I>H?xJx=&}fh=z<&~p%cGLY=*0|X?=4o#Dd4lw4%`AY~Q
zdC6)VG>x*6DL8UIdKc9ilaRi#9<r@dP<a}PF%l>L8-SD(nbBk*1}cj4N{r8%S%oVV
zV4!j<uk;pe%NL}}!7lQi@(x>$Kl`%Ooct<_O_fDd204qb10$#&blpdvF$p_^yHml{
zW!Xuy1lIEt?QN20rn*!45&W3ltMT!RkiK<JiYW-Lg_EHr|6Yah0lFobGby<gJ15Z^
zzv1-mjfUp3oD)Gu<RnpS>fLQRcso4QCED0Di`=1^yhT50OtU{Trj$M`BOov5wxod}
z$ZR<e2@$q-Ev2?443E)w?!*;FdA-3+Orm9%r5<i}dYSuTu1;IdH2FZo)(si&d@kFc
z`6$A$g}OQ-Z!wk>f1Jy(yf<F+bc4QZBlqQK$Qpe$nBrM-hiFXB_1l^F>ZGQi%q9$U
zlk+RpOW7jy^w0UY?kj33Kh|0n{cwLtR|P>!_-pr1){<(}RW=2C_RV;CWI@9*gAcf4
zuC8=a^nUoqg*Rjsk|m^<Y9G>bE8eC4J?J1a6X{1Foiw+fhE>lDC<5<BYQlcNi;)@v
zPa@37RnNe{CfF?Yfdt`c;13aGL2z01Lx74^gnf(Y9MEfX1aZsj?l|~4_${jIRZb1;
z76lMruE%*vuEcEIIgRiQL(K@M;fD&%Wxt)%x7Wj9z>b1;69Wx68$**DLoYP^<BPXB
zg*ixM$MYXGl24}r3+wg4Dzk!-QUixZgCD`yen_WQf0Hui>F}AC)cxNMPPAuBx2Gj<
zO6aJU29L8yT4y#6qH8@Zj&v?d5X;a^T^W91s%c4L1_##>yKO*4+|(^;!EMFpif6!s
z(e+Dl!b%>*i}ZcGc6C$hcM(Y|viG#nnt0tIG)iLvFczdix6r{{@~lCFTi@4v8>pYv
zuhwr0*{{$<LQlwo8da=<oC98h2xwwHjYcKcro@ZU_R2xbwHGy#m*#3rsdQE3jH5rN
zj*a(J5T4;!Yl-zBY9zWXE)%eMO_A8EoXj#@7k#7?o(`)P-~RFh_WSJSbpqY+U~UVB
z8*rRWqomCzV~J%Dx^t+hWg%#dKN2B49p+AefaupVi$+|#c5JQSV>AZLivO;~a#96t
zP`DtCZNQaOS=Q9y%ufyAb_dW?f+O+?^a$NFijrwLI%=RlRnnQRJ8^0fn)js$zrl65
zF6g@1q|_RfOWN@U>J?vR0>$A^r6N;d6{C&w<Qtw3;g%6kdrC^fEpqCAECt!1`QmhI
zLXT)i%H+wUUWH_JxiPU4jD*#FAe++6c9`ihgEQ7DWznI6S>*j)=wy@HYlK^Dl?(1m
z62SQhRp(~0VU<kA4}*$2(HS{SExg!z-}>ytXd+7@5(Ml=EjpQN&y2klI8!vn%H-2r
z5JFEX+`a1$y=)hdXPRE>#Mp)EF%~QEW$MLtvPRh=doBr8bz>4hA9g>9BK)9s>n^cH
z>kbq{Fq}QDbg1vZ@!2+ywI=Kzc&MR(gqI>IuG<v38%Q?~Pt?AIp{ABLnPH6%<~UR`
zSgzIn3S`%bJ5MK0((8pYv$=Fw*<0#I`@L->{BTRt;_{%ARTOxl>oxWkQ3MBoplEN=
zzI@+3Ek3+wgH~bey(P(y;rs+*lb_@fIrcmixUa>=0?*U#`Jk}@LUpKrb8AN|HNV+j
zzSDWmUfeDUO^u5Fv~>T1V`n|54HcjqcU5&SvWeDZtT9Hc5qWv8h_uu@ldaJ69SCwG
zr+pbde;!jy-#5l{?QI&B0eZj-GDnT9A0_BsC<oj$$A@jkcfP9)S#GM&bTrf^Mcq%V
z#mVlhWquTGe1ES7h_bu8+O(!UQaP#Bs~-PY@AohL8d~1GNPd>6LE}lOxGM9}^GMSj
zP)v{hVMk}L%7@O{RPW{VXxZ+sVCGYB&MUyaUmbL;$J8x6$#HJ$&)NJKr)VRcuNiZ1
zI6TlYrjr6#Fs|0>1*V3RtW_dv8Y~o1thfI4t4nCfBQ6)&ozB$vT`;O;ZTob*bRja=
zapXt&`SY{Zdfb_l@3%oM@^lr{eEj|lMAz>6XdjU{3+CloDVP&!7g2E+Nm>}Osc2;w
zNq7*J*>nI#BuIbL5}L-_2Wy^==B6Ue3h>p*dp~|Jv4%ab_0*>VAf*)Vd5Y7;5y(#O
ztglY@`q1l$C9sI$Mz7GK+^QAY_(+Wtb#sQlJf#VkP!O4E0)io~D_ViIf{UAb$z9sI
zgj#m$w2Q)O{#v>2^le+FFgb$qhw*v+U<3Hhm`aK+=oG=49(MF09sezNk)g^IlwpUT
zE|0;(g8KcRKL{;no;xzl8i@)rzf!Z)w2e&*GwEi01rb%tX)v*4;FenzRotWpK$Fac
zZVBfKIw;{T(}D4kzLN-1nNULUH{bD5$R<7LA80bhob55@qsfR%+J4p0^`Ub1<AL+e
z+mptB`VedbC0Vw2+4U6X7xIaD)7r0*dm|2-y$lam16M#PAy5O4$nZ$>6H>{e=*5L{
zRKgV5Q%K%B__$F?nNnW{mb-kmUw@zThK{}HNajmUX}?M{6;RDGx#Qf+7Q-0>cGuSC
zNlul&RS|g4F4WJ?vUVtWm^Q!x1x|LqAM%VthG?c+<+OFka7lO!ii4jA3g9W4{&FRh
zfb017Qr@GLoZC&_!AE6SHGFBJ!=jGz;ctGiE$9}D3Mg1P5*fZMF-s=m_tmi5WhfeN
z>2k)3(}!A1K_8Z<7+GS<kqxnNl_3FHIlc?L6h6+-`aw%OSZ5`1MbYe>&6#%!&XHys
zDRGj8e;m0{qH&dRz*sp7%_DHJ^qHepBc5;Q{j`|uHM#_m)+##>=Xy8)tyT$p-fW_{
z!^|oJhS$m<+l}hRbmft*n*kb^CwjM_-?U%cX<kXg97;$@m=gKB^PN!P7>&sLv2=9A
z=0ska|1tBVsw^@ocrbms8#GX$be4(^K$UsYLfN)7DHc#5b0HZ<Ws?cANh3nA6EY>1
zq};I{qiP-=5r`eE<!7%#`Y>ozphE-F!1RmNzgAvjRulROBkyddIraw*{2{qen3`S~
zJE*lsDkD5EmJtw7h!iehE*RK^u*?n{&DVQSL^*I%ya8awUEX9jKYI#1hYZF`f7<&7
z+yksrlNtqNZ(L5-b%S<%T%3=Nd4#GkF`9F~2Y!vQ@gyHU0)3bxB)w-Of%zk0nl2Y)
z>>R$(8F{@2uk72D8MC*gq8rAIhcjfdxkpGBgM@t%!2?3W*>2D@Yy3dRgsL5WMaRb=
zV#k6k@s?`PA)mVI7eyY54z~`x#^;oZ?4j?aw~n_)YIuQ&rnAYLcX;o#WBg@GZr|7J
zp@wL=<cP=1t!yI4kLk?}Tm%qFAL)z$y8o}%B(?d*uh+@t()zDCyEi=8tPiru`+Ln}
z>7+MfQjX86b$&m2F%g|FR25z7XW56=!je;O`ds*`@V}=wlX$jPXJhi(okkz)-i(`|
zHBd2tRMkELegi+<bK=n&aC5BLq<tq@`*3q$BSEtNks$EblgR(JBSSx!rN!M#XbFRZ
z0+_0Z=+Bd*m~&w}X$ky*C2!B3#8)!SsiKT*!l=KNp&&_Hta-6r`>|AL)OjBy_1Flx
z(<SbSYU?Z!xh*IZiS_0!jTylQElz`*_iB^x&W1(t`VlobRe1_qTeK_kDBMgWEQ`vn
zIQORXO3ENxCvES)gnT`$Nk-yT(~;$QopFD|R0M(&Q63(CwxxskV7c2eb|elF1b=#h
zFfx}Dk=7rQMFM~6&c@(Asa}iV>jo^@6!r3p>A&iv{sS>;e#!Ptp!bXUQ}NMRPoj-z
z+Ir(}Pw*?P3Cgf+dXw)C_;ADY6W<KwYDel4;F4Umu}Jg}wEhi}JoTVqXVuqbK~U}d
zdA=w&|2M<u2I>T9Zdnnb-~zLwnbW!Q!7WX5VWU))$<!BV=vI5$ZiE#59)v=zhh=GD
ze=j{US8%9M%o53M@Ljg-$r*X$OgraYGOZ&eZj4`|>igcwn~Qn_flAr_Joj>`BLmCC
zBD{>&D&Fn{{w%=o4ebj@`{XGggM%(I?x%T?^aH^bi7a6ezf1rIQB(|1&5>R9u6$&U
zgI47Qj_H;DyN^=@&9h<tw0UBd=$ZcOBQ^D=uNr0UkovE+{%wMmK4ojWd<GY5GpE~#
z={izzB7h!vQ{c>H8tp@Q22(et<$>~U?MezuKiPP&naVFXvc;3jhKug_YUuHd<i7;x
z6hNU0Ha?1=mN`!Ut&}(RJ%@)>$lfcFA(e^{!eeFa>8%+pm{t5;2(IFk{=HO;i4Bbo
z8RypQ_*)Z-ZfDbk<Rh*1@u5kx0zLEm)aFQK6=FW&=S7)q9^!(30LQua^nHK($_6*F
zrEW|<O~O)`nSoz9{LjL}BR2Xf66xmLoWmvDk*4%5bvlhK=$G$Yo8%uXIL@u;F-ltm
zxoTxWSKH2rRJjVH!939TE}Cfb^s(xLkfmxCgYGkm!+$}D(8e?WB>Ga?MF5}L4iQ~~
zPK*RYO#39Y0)D9O=L$H9Vg<o;C-CG43>i1D%NO=5Rt>lo@C25AQR<Y4+dCxE=-5pU
zr{3K+PC+?c(YiEH4G30d6KVeP{Z*A&z1r2%mui>SEr8y-(^q`V^TPDA!u870Ra&<b
z4r3H3R&pkZZCW4wbut~um}{BB4B5Ts9HA*GB^Q;LK1WCGosEVrGtEfmwG^J@GhUqu
zc1K#vJ4A<2sVqd}Pd7+Js*Z#*0_X0R0gU~ewjD7g(d-JhU(_<(!AU25^28w%Zn@JU
z9(d<$>XiJJ09n@#JDm<+8m^yhJuh_^-Z3cmC;yucO--d-Uk&cXu22p!L(zxJR4rXR
zb(qLoR8clj+?;p$z%Bq2O8X}#p6x^~5+T?o4-5If3kaM0Ku8hEH3<``cmrCs=qMFY
z_bU88SuRQceqc---Ix%m)A(x{|Izd5r;>X!pp6l~APHqy8{6lb0jkp7`DeiJQ4=^?
zicN?M@aYny*t+K&>tS4Z_q3<US8`AZ3%xlT3XW@*HX3@wJDm?v*>uJ~Yf4sUW-iU@
zJ=7_t>@`Pf4}6WDBNCUL<_x>82Lzxq;#w6s^rxba_|M&2x=U*UOKm<wU4xx-6YC>g
zQ7m^&TX(3a++BtN92RDbNrmjTSn0x50z@FmTZn!n0)zWB9V-DcT0mX7i-wF)O4qy{
z2SAkV=pM-v04|Pv3sL8=P*PsepL9V{Hed_U=%g{a%<m>hOJqsu+5toLKe`l45N00-
zvXI><vqMB|PJ`&r$n{#5&!1!nT9`M~w7@`M^|ht@AY*bq(6243O25CeHRhiCt0p5a
zCWS96m??tOqndo)3D#L)p5jo^mfr#;k!<B`J)}nUA+0*nQkeWNqQ2&MToLEh0G<>j
zN&7$u@oXwV{85mnoE$ULTpf7v`Rq_o<hJrap(YX+TVzWsPFK~~UZ)qGBidKv77rU9
z=}`81%wo+PgDy8|(Uet*XLp|S(#}t&@Qn0;s5Ib;Ng{Yneq@Hawn~!XV7jzlJ~#He
zXR$dh^2j9A*v^KV>MrBu<PY($a@+Y_3!S{f)m=|cQtiN|t`8S~%3Id9ttP5$KVh+$
z%TQ{HzzmGV?w7ASJaeUz6#W>LnwR<qLb5MD9da)V@?fwuj3UqQNAbY-ug5pI9&ruQ
zl6n7OC0I)J_mCIhv(@A%(bqrtEExuO6zLK3r&)!lpuAGaxx@CBO6Rl3nC<pT&w*kx
zPC(0UVoc8r_Ma;|UPm~+1}j@PJW6FqSfDI_@M3{#5A{s&biH)KtUU^2M8;h{Jl<zJ
zwX<{#S}@G)ga12dLjVojxT;b$Z@FBMd4i-{vLz(|xkF$92>|jaWX;fg>DVsFOf?Wn
zLIau7G~z}S<M^xT67&q=c`~BWMNk9V@N25Az%b;+{StKdg#jH!Bz1ZQM(%7?NuQrj
z-%o6|2<vMxl?dsm`zy$<w%A}ryl~@Zs#yJm!!B*;37$N|g9C`8F;=5z;L%e#{u)Mq
znY`k=NG5Uz4a3(HNlHqXShg_}4rvO-DOWDCost1txzLfVZ-=jTN$1wQh+}d61!<Cl
z*R!g4cJj(j<nP?~>ux31-1lN7Pv@ju%{4TI+)V2k9lto$nu2J8{P5khCNXOos3}8?
zN_Vg)WUK5bFg`?IYtI;<P}*4EgY@Z})*O?VOFqekk{WZ9|B@XosBTonyvk|@x(M8+
zb8{nOQbXL{Sp1vpn$59DV%$x>anN^u2=G<i5thA075S3~2(dckVS!9jEI=^41#J6)
ztXD^#lkd4yvF~zB?ei=^%%f_a%CP@^BJeLk3ZR8W)d?q@jl$uBS{q0XrYq075aQ#q
z8+%uL4BkEBs=2YrEOc@f(s)yNoUIwKR8c$8I(T%C;yj-oJUF@ve2GD(sH1zHRJ2LZ
zO1;(u$M7l0ZswcQm!4Y+f(mGcD|srdOlnU<En~T_VKZP6brTGPj(O&7Hl1UpYNXip
z;*vVTXFqwHN*yS!YGs8j&S030PRkJYdompAMOukQ%4ie(-c}17WBWki5WQ1UoO6-*
zGzmNCqLMrY;^!3n4zd2-^gW9*Uj2<au+Jc$>&>a&NM9bZmmj^6RvLF7@}8irr81U_
zWxC2n5qYNQe)GpSS*4p2iyn7VCGxh$^8mFow56N2^theWaLl87ATlBQ2ZAk>=Wfwm
zR`@067g>XqAYi{YTzqqElc1T*PQdscaXwygLsSIc_xZe2Q?Y{3>2{c|3~G=t99Tf+
zJppe0EF6)JGyTOPf?}jlng@IK!E!27U!N$s_w~=*Ys5`HMN7~WHdY8yj&qJiZ}>SJ
zp`SXJDmYKSLOzd3I1>l=&aAOmfp~uiJPq64^WJ+0d(5sD>?=P)vj)@lrrvdhL$)kF
zUsr|u=soUTn3hOay!L58e+Cbg(DOtcj(p=|m|@O>lH-YoO+zcUl9@EpWSiI+!ES)F
zGcUI0>ujU7h?(#Pr<W#4dwtw<<J4Q8*KHoaSunp-Y<#eHJ25b|!4K^y*N<{q>((1(
zR^$=5e4K1zIIWWa+c|(g>8{fu>R7$*B%bC4$`EMx@z|WR901u`vX#jTr~;3{AezYE
z)US@J-HQ&*Oh~5O2eof-Voyh~pH-&A3pKdILZ7e&EMZGigCc-RKSf#7$fo%eD2;S4
zbG7SPI_TW>Jc+2Ym4a|G+P_7}-g4LL@am!+EedW;euKY$MY^UVjBG1KZyXVT<gr|X
zJusFM9w7334Lzi$rFoP)9ZLRrV`gk(ZgIBb8V(HXd5V`U8VM2t%-?>F9SE~hoaoZH
z;#+@(`0<v}dKcm_vdB8R`rE8Zj5)m$^ZUW?3>pB<iVKpvZ{l!&IR{pS9#fxZE22s+
z^=b<2zm062M4+Mh(1t)(<vG3;2-D{wL%6nC`Bl$(nNs2%$H}jKcc3$|pWP;gf8S1U
zmz1Ns7^on-cpdv&_ba&7iQcsj%9NP%7p$B~e}sDXj`|vF`Y-QGQr_~G#qJIktB3LA
z&r(qHENaHsN{L&EH&bI`X7=FWohhUuu<MOXg$*mXHP{^5X17Lit>pe-Fn2i8yC-mT
zn0k(Fz+B(^GBFfsA$05LvHH#MyJells<~B##Y5}LXicccxadl}p0a0gbm>>dy*lU*
z7sHZ4zz-g9IHS!$45IVSc{hl)*3$qYIN}C?iAZ)@5Hc;LGMsj#d?bh)Mn!%}AN2+_
zM}Ua6DI;JSh@6_45f~ZBH%Tqq{sU2Wa$e8j^1FMJM%blF@Y8;ZYD)Jof1NIV;EGid
z%PT9>W1BLN@IkY6B^`MRWpu_B^Z*gH^$&R=LFIWU17K)t9ic(ySO@j|x`Ei7DkGhU
z;<=1a!yyB*1L}Z;ML(;Ai>6cRc}`Napj`KS-MOEB4^5N!gBDTgQR1RZqs2}gYG^~A
zUizA)MsdLfg|N`iw93Sv$~IDlv@nSP|0@;+#Ce@%6YX;$f{KU@`NA{%uJgK&SmjoO
z@>_HM9X$yKttGSvx3q>Tq=i0@6n}ckaqd>nyVuw!Z207LtLbUBBQmTvT5EhKTeqX^
z+nz%O;!Ab(XOgof8lY1&Hs?3aaB>$1tK)agd1!Ta{&<fyArc<7XD;6aL+TZ#fZv=l
zz8IPaKh)UXpcqm;Sk?kvlk{_z1q8_&Y<VIjb$!Gx6Uv54m?7w9;&Y^t&R3b$H2TOp
z>mA%Z3taf$t?$(G`h(AdC${OdwfF|pY0*q3HAcRgadOet5dem2YPcFcHi4RlmN>Tq
z*nJQEo6K=s7wzhwXwmm-jHS6zaWy?A*LRs+<U`qU<#Vz9e$(qkxgZa}=XV(c#p5+H
zE~?C$tKF<5q6|q1x0wZ>U5G5@l`{9D@;bH<*}-l*41o5F6B?(7M`5}I<_k+;{KfI$
z%vGv(qyD2gS%yXx4*0OjSZar|fAp%X)1RSkvWD{Cq1Vd<1W$>){hZ&c3W>@Wcd=-d
zv#jg*>WUYZ+9~2U|1FnkH2Ml+v#KYvX2{h1guPvHD3vz6?lU9(_4$ZOC_hm{cI@+0
zi+G&;yh9*)%@rZ`N|tmfGDo)Ur?k($u30wWJ_J74rk@9M5C_@mgDa<P0;jwU<@Ic>
zJZFf?%r@NUf%TnB?J6U_WpLg5n_C-Qp*{C`zk%fDjN~e>IHh~Msf+8IMxty_8o>>A
z!}doNN;&daDTEe!Veg~Dj#)Q=3}+l`V?M*pEJ?5&qEZp%tmG!N-Qy&pKBvKix%EgQ
zyN<cP+U@>m`2B>P8xOECor)KlTwpI|)nm!%x}o%+dZ>Y|cT#x&-G9%G`fV?LELoEA
zR1U*0Q5$7GE!}H&Zb2%MsUrNy2B{44dxPVf1kg+ZQyd>Q*qZ^soa#5=>;co;C|1=j
zkj%}Q=75-6f?a}N`WKJpKD`0#%t!)9H-IBSqoYU%Khij0ZFZ(!$!n(dPCnf8?`&-#
zrHB_1gP5dlOOORck!*)2{H(TG1lFbYnBM|3iyYQw>iN}qD6TVpL&f-~N>{#M6|lNt
zo)g7$c>^U=YN*=)b|NEaCz8eFSuqZ*f3AwFR(M9)#}p@wkm7P%WjcU~GV7QFziPuh
z3G+L}AH+Qh{%vz1l8(myqXQPxCx6}gV_2%q*I`8GYuamCe1i6A955(-jFz$yQ5SSt
zZ}cJJmyA=?*B*YDeuiIB-B8O(lxksByGV`Q^)P<{0f!PT)xQ?OddU&y;m=%l;d3`K
z;Fs2B?^l<>3c7iRWl0a`BWCNQdJUA4FXQa-mYRrDk{3rBlt~5gnL~`d3H}Q19=t}~
z0nWS?C?+*-SZ2ROF@W4JAbS?d7i4%s=eV60fAoWI9HFfx7WeHs1F5q_TPTw^WW{S3
zRb)UnjK%L~13{(9JTyN!<pU#q`7u|(TxAw}yHSS|`jff3Y-D;HIK&Fx<R9oS)F8Xf
z;gitryl2Js#|}jnW(&aI<53H$@_ztzL5jY^#xD09u??$O$uw5yI8>-TXaXsYUYn1W
zfEsBCpadeXY2#(4p(*VMjD1;$^ahI+4ayXfbgE>G%~;c9^0f<DKe;Nm<V8yyU79W1
z+bP#1j$xRgk={Z6%H)yyWAUmNGd0D;vq*CCF=O>Kky6oH%t(I!04Eh@Pu^i;KPaX_
z3!R6hB!4>q$MB#HTkC`Uou`<%P>&zfhbR94W)%`kjW$*%=58q@RQLP8A(~DZCyarX
zLs99sw-z@EHKdOes(Pu#GEZsnUyBn|NZ`FI9o{&{0sjE{_1i@!H4RwNGRSO-o;S$m
zfsf5HN&u;G{;jd;Tpy0WZ8q<ZxDXGiq_+{}>~SLJ_(p0yK0iL<Kn702YKa~b?D~?X
zK+6YTp{=hHSpy18o`4z<$~GZdGm-PEkPTs6U_VS8;;53;H=bpvSLK(A$GG`rBy>F}
z5hQNkXY#^u=dD}T7=LWuVU7f>paIw%QtDdxSX)ZF#^;%SiJIxY-eZeSWxCn>q6eCU
z4jriKHM2DD@1xXh6>jZeEW8VgLNoZ{v*xzAf)gf{t2Cn-F-Uf>{0~~D^d{X7cH&E`
zmL^%4NIK*JRSQhKOre)QcvLVQE1kaEa6RgbM=~7KLSrNwSkq~Uod)!#f@5Kt0K1&1
zHCkldk|+ZHkls#@smND5j8KE=mM#APvQu~~Np&t;NBZi>$jx8zEsVA|b_pX!##=aD
ze76n39sTHUQN}I7zeCV;J8!etK^XoO6{gCpFx-malWm=GR!E_BrKG&a!K}$Go(97M
zNVGvJEQ1UO<yg1Y0A-V<IJ+@TT^M(evMD$`Rf(>p)3j@AaC0ru`$VAjL6C$0075^A
zuDH9L&r7p9^59I1v=Qr4XfP!DtR`WYBFxhJq7B4;ILG5w_A-LJl4V)vdyAQ;mzgAT
z?0w12NoK6`FV~tL!cOwLH+-_F0IA}OA9(&1!rj@@g@|TS6dJIxkw_a6S7T<M9|t{X
zu!}Lb?%t{s>sJzF#*u(&6m_Z<#xms8i0nXN)NpAuQe@KvnLutN{&Yg-HQIB<X3)`X
zCEpnYcBPB%999yIjZ?VzSoAe0pU7{Ulx_*BOHgN=(O%iaA*9U~0L>*x1us!Rn3PjU
zL7D{(08(a<1r!d!NKPmio2geUa0X2PU4p>3GaaMQb*p=A?eVcP-HUTl@3K4NMiisZ
zQ*z1GZl8}nPoQH?k5Ihb@?6WG_swMKs~gdho`v{zO-WFLcOg^QoYsD=e>{;qa_wbd
z*BsVLE12%Y;nZl}#!a8@bnEH+uvi!V1af~x$KzB10=m)jlF=j^vT!=p{c*qIade}i
zKbAk>Jby|$f`04%yOg(cOk<Nwo0qg}p<#9|88?*&X!om9J<hFl`_J{xa0jJBNlTrV
z7kKicaCkp`m|-dVMl<bLSK4LmlWm?2!~4Kibn7WQGg_2fRnb6`oD?fgUbVT*CUZpq
zD)MRjv5F87vCC49l)y%$r(g=YKX;+Xd(Cj5eHBiB^dU%Vxt~pQnruTUdQ^eEtN#F_
z{{U#u(XS@jpYHJO`jxB-nCxp;;;^*vTeVS;*Ewpp1+xexkdBF(0J71R(5;F78;J+z
zss0sKX7cToJLe{nDtn8j$+GQmIBqdXc%DV$pcq`_QzXA@aPb9b*sCeXAmD>ux!`XW
zCaD_P-nd)U7*@~v!TvO)G|t{Psnx0t3DvVo$0C>%?g1eBR|n(cvq>Ozd9Z&97UDeG
zOf}HsugHHYPY$SEU?X%OgZWh?Nq=3k2F}v;0QpIc@%mP?NA}Gq1v$)nmp#oUibp4F
zb>s#Q$j%Quilcn0&UPO8`uo)xlvYx#anrpvatkW~!4v_1PPP)XEcqp+RA0o@mkcDn
zxpDkbGk#>!>S(?a&YroDS6>t>CYf!#2E)4YKixm$QYEt>mqNF>ka-BvOnNj&2lTAn
zde;8{NyK-n9nz8VvH4_Q#8upsV#9G1*&B#&M(a=V<HPPeV48=x%-fbL3WHT5IFMB#
zGE1fTvrz$&RKz+K!&LJo^~%x!x05S_+NHpt3o@yP!*U(p>#r1J{lq`~U01h#@eff+
z7e0%~b5^=r3$0sC1Y~&;llTf!(^0`)9*Y3PU=GzGt_F6ZSWi6E^V+4vL}QvoBb>jn
zcv$hpWlX)O7`7T4>laCD8ARQZ%)|^R<&Qqq#p?5m%|>bVg$&qL9Ti6kKL7x(st)Hh
zc(oM!{-5yXd)AGY`Zk4oaCm!$TV(ou=l)7R;8nXh>F;wfYgb21Z;*`k6sNH<+}@EG
z!k<bl(<}kx*By?MSkq*O)OD%b1FbZLa<q&yNU|+E1<NRN8v#|QNR|!Kk$cq$YjveV
zwW$pgGKXFd<w&v=&Uqg74Ha?AI2a?IaZ4QAWH>c015z&A${mI})VB$_ani9$%(KqB
z(O$vGAUUKxXaOjo0*V2KXw3wPMF131Yyyf!Av$wLXaQqF=Z>B0O48qI?#JUwPeZ2*
z4r+*=X#xymnxxEF<kF9GTx9PqR3d^~ioF7q!6QFfQ`oE;eAg~VHhTX6^~EysHIf|h
zp0$R|`WZD=lT6Ij7s%7>p&egz?f9Le^j!Wmm0dB|`5NleXtAcp=IS=F{{VhfU(AY2
z={(!n<o@)G-<hMSA7q0xQzV;c2&cbVddVR9zHUE=cA%tPx7BqUY<V|wZTjY}FNw7t
zlX){Qqa|kPMV6~dIu$Q<A#=#JiZ}IIrItLLeid<6Rw7RB;q{`b#EL1b+qzJY4WgI@
z6gC2pino9Jy)wc-?|G9S#O?n8*(!RK9hsgvrZLH?WQj{s2Iq?lrPZVk0P<~P2M4Qd
z8^0X!_*NAt>}gx$VyMZ+aaI=D&)p)KnKV|RTXdPNZ8=Pl+!h=XIQF1+888i{+#-;C
zx#Mr3D*66Gu`W3yoYF&<$YL=@Rv*H1SM?w!iDyp3nq-heFUkd2hT-9nWQoeUoQ4Mk
zPzSAeBgXOS3YL1Ki0;P=A>c{=2D{{vp{p}Q?oP7tP7j{+42EaQ{x!n*`{sdS>GKxP
zr3goxNHXPzP;=6aGvwBxbHcI4aa4&rds{EIX&1I~C^&LPe}wj~b#)Kys3eZ%TMNp4
zM^Qk|F7Dt@GsvT8W+xfxP|GP{gWIJt2$b?^Ny`2-0nzCTxBD|Ar0VDJIsTQ6Z!h<{
z(&}mUSM}J@vXv4{w(NTf>nv~K{>gHwD;VZp0_6O_{{ZnctPxi&SnGF|vNZOVAmDJ*
z90B>)FRp607q>4Z%+~wK=WA}JN-!jg%yQ`}7W4-c_@;b>m>#$dNr469W=v-|s5%g8
zkeO~lz8sFVRz4dBn-J*~#?@vf1t?%wU{X#vt1-4kAT)I(AMlRYF5ELqG@11uApZc_
zO=McRR@!<}MbzKqzarCDS)jwZ^3p^6Hj9t@>4)^GsqmFcbFqNsY3y;{xL)UC9~rF}
z4n;`Q)Jv!qXvkq)#payLHWXwU3$a(kL3F<k$jX0vaKZ&X-B|uuAMh&knSd49hxl%B
z;=Uw`l2)F%`-p5TNBppaNaOvvK4bcAABAl9F^-XM`u?;t5lavUQfqqIL{XyRl(rOi
zHf|+Y;ZR7<Dl2&ua!*ccn#R{AzL-TQb^ER|4O46N2*JM!X<U}<P@4B2#ZfKy%sH)j
zN#-<k`@m+wJTR+9LAZ{U7O>F0sZe5pjA8RqPy~_5wBXc?QHM%~wJ5Ge)5onZ@z5)h
zTRB$m*0p}tfp$zXI}y}YGLEIi=y}~0>73!W(vxYR1jQINAQ&{eeP{tONkEum6m_PU
zgvA)7I|URBF;0#%BW+Pd>@=L6k)>;Ey@HRrI-1lFg^cnr2XS2UmbN|I4jD!cqcf(A
zx#`ETsFu~+g$F*EtyGdn4SEu(KXn#ivt|3;YC@q~<|eaq=6W5}aMGa*NRDYELBZyd
z)y?I~K&JtZ%-xiCC)f^Zlytch#d1rMJxIJ#){U!5nsv|GZ6j5VDRvdy0)f!`=CLbC
z?5=qjhMh*^v3o+}R=0oo=1jl$$SwUUiKD5T_*eDVN^%UfGD}vliQj2#HgTHtD@_{N
zY<z&L^9(V@b6og|xY;A84;1fnoYlNHac7w=KY15Yrm!-h%DR}`cI{eIsm_e*)8t|h
zGI2pTtqg?csie;|0PK$R%4h<N(*UO}CIS&wwZE2Yc)LXeWAP(#{aDgll)e^cQW|WK
z#cKF-PMY@eNuz<KhR!o}HU~srdHS3W#-+O=>BY)PA)2$i@Yqh|mv~!o^*&?PZYWy>
z^>+X=a4O;tCNRfvG|586B(urO5<e>6wc57j2N?%6n8=}nW}B&Jsa0|KHb3@i=jUC$
zdeDb0kq1h@1fbf5BRJrXQA~oXs8;!MRP;0lB7h}H3XoTVI#+w)A0N+Sw%b+RJ+R?q
z{xtsp5l+`u3)uBNJ5sjPZPt4!l334mJddqh4o5>-*w99L<vd8gYPu`bVX^rNXlI@>
zc~?X?LP)A8+a*xk6o2bYqAYc?_=eX=hG%o;ys8LA<mcYFU0V6|HhHBW5=ja=4A6wk
zQ<dP*s$2g6uS`NHr@HNp53K-o8dNVT+iV|lIUnhSCb_+yRl!4sBhr^r{>uKl8@esq
zTF<>;Sy+q!MsZzkogo@s=w`V`eXc}oI%gi9^qbIe-|sSY3tJ|Sf7tGy)ufo3=bqjI
z$810fZIF=Pu*CfrA45&MjctE*oy<AhX(z3Uv6m^!W`-XqHAJ%2qi>9Z+Mu{ijgdfy
zf2@z5w2-kBz<i@T(j#qZEC$}M#g>&Gas{z$dKQt-^2K0Er6kCyTNhJ*yYeloW-vu|
z-WnN()wBNqF5rLT?@?FbE{^9*4nZ{M&TEOCXnnro{8r7JZbwQ_L<@4)+0SuRZlP6E
z$2E^N*|m>5_?<r6yDKF>>SHY5?>8s<55l;)c~EQ8qOAmZ_?LRDG@Dht)nJZJA>Fz|
z=!!Qt=nYrYE?V;X;^p{hWM?DOC#7rD&+L5v0I$rGO^-6}Zi~%dX~MYA12mrCv>A1O
z<}tzIn{R)$SQDR0o>V)U7J7s?P`$m-!2SpHHIaB8DGKh~I~q}qn47s+?ryoJVAY74
z-r=H<5t`mvyy(_MJogndT~N_#C(4ZXscSOCYHa`|89k@~B-6mbQ&V6Yl9CheOEjCg
z%_|P)E=FlF>0PEnjUnk$0%D9%1F1XnKnah^nrIUWC>S1;+!_FN6abjww5;P;;mdVA
z)^#>Iu#x7S-$PafNTLJ~JDRIK!E=H4f1P6$eNTF=MNX7y#AJA8#%n?{ksu(i<3#%v
zgHgMaA)5LTo<ko_wVyr2l8i(?ne9#yaeA3zadD~c($$o{#dSP^R2OzQ3cQkP)zmL#
zY}S^?F5~?oP<sCWusw%=Q}h+R)6nrHLZojg87z#41d3w{qm`oejq=TLC-*JoKbn;v
zsTDxd>JQ;TQ`*Avu@Q`w1&`2E5+#j{Ex-6pboM?h3AJQ{a047R4P#c9wCZh6r?WNT
z@n)MM;%_oPP|KRi)|*1TjiiR;B)vcgd90yMaFMj?LUB1x2TqiP*IC3RB>*PideVwu
z7^0oPS~(RUIwUemh&r<cAB|0|Fqc-lEB8c<tMms0^b{@1`z00_siTClgF_^D+0Ut~
zx<fss;*##L$P7~eak(3VNtXuWB+2y~7l%)RIX?U|%Mt$o$4LHWvo0+pirr;kLKA>#
zC!p#@saT2!Sk8K%da^D|pDS`bJ5?EqvnX-2a61n5xnm!lZ?%{LPbQNn(1lW4tE+wp
z`#WS0VjKSeWv*LRh^5Kk4roJ`$X!oQWJR73S~eLt{u5b>#d=d9ksEeY_ou83eQJQ2
zZJQL<h4X<xu3pmbH61c>HH@-E<NK$B^%dXfzY-k+$E?|t+s%jiiX3iIk=<z8=9#VW
zr(B60Mp+2M55lrF#4t@9r|yzLHJz=wXyJTLZLMW+k`Fg_;G>%6t}cwyv+WDJk}8DE
z`B}bg$J(o-?aH_3K_egUQpQ;DIiN(YIw&IplTf>EZ>LHC>a-x!TFHqvBHO-C)NTH?
ziG2hx$Ue;NBlkF^)W5Qc?AlcMyycJFkSn9m1hN~yD)tiX18`&+;D-K`J%<OV%f{4v
zg0jd=AoU!N(y-=#FP3gSahe%7862%E0}P!Hb*i`0GfwQ~Sob5PCAi8ga%tz7_^Kt3
zvu(0t)~sBvMFJ|vm0~JFxNOh>+_0r(^rz4beQCDd!xlFAcF?IF{#YmVHJLhYT9vVN
zJ~l&a*vgvkyfSfZYySXTfA)8%tMHdcbF7DvQnC)U#Ll8yShCf*YP&$Jrur8YO)KF|
zP{?`m{GtNvJwX2eIIPy@(cK<1<13cC*X|sgoZ_+iuQlsaR*F2Dj`M7}VxMKwbqlVI
zAQsX6TbBO-$5$WlD#Jf|tpv76yKdzpz@|z|^`s&&osCTmzT-d_BeKBar28Q~YHm9X
zdrhO7r8HYjGz1|5HK(SN8+xyCL)aGP;=-(o$?4P5lLS^piOHgX3QBeXDZuea2D5YP
zQNJ}d0|OL<S7Dd|OSYr~Z5W^fZ5w?k0NQBSXaLw~*k}L=%`cZg2bV!g02Vs8aq_@9
z<o=bM&dffFOl`%%3aC0%3dTnPMo*xsQArL}B{}F?5i)_i?I*T+P|L?GPwP;(JCbTP
z7AQBjPs*m)BjFBP++vfKwl$%LQBGya3u&+w2N#kl4u9v8p5N~t{+%lhX&a!bfz6C<
zE>e`sv}U5}D?`a;-8<#6)Sv#Cs(x<2;Z!LhOJprr!$TOsDqLWYdr-#})SLHk$QDoD
zD8VD9YQ@@F-5`4{(75W1f0Zq6JuPB^-oJcq#W*k{zvEP98KzJHS2XmCc{8T<#xY7)
zwuab#Uue1ZCYs9G{{Yo9E`KVn_A)MZ5Y(dkowe{O`_x{94I8K5Qa=iA4N>PxE-uXy
zD*Uu!f=ik*gR0vWA1XhX-$nV4@~RRsW@37ilS165@S+#1@vu|?j?^+q1CMG@!ljgS
z7ulP`a9#v1=4FV0GnWcS>J@8`y11HV#H?@+-RvntV(Pt9?uaQ=;MJKAjrOoLt#5iF
zO~bc1prY@FZcMtfyMN1A<Wcyc#X+h^=6uRA$0yd8QNM+MU5Ru}RTQi<aoA^}t!+=k
zHn*^wn<nyPAhy6Vr~E0c!q+*4(KQ1Qnk#3PKkp~aQ^OpwukRp{51~_3X3;LmSp1O4
z-$f>^+vz%l{{ZT&XK#IkQ}q>%n;!=1w*+rU;9pm4pU783W#9-C4S8`a$GWe|DLHm6
zN4d~hT0^5FeTvpNhB;t4KT}j@)vUE^8>rG%W`O0${u5DqyOxYAy)wsIv)}Tq%8Yr6
ze)pmGt|w8`<*=Edn0*`8m^IZDWQmI8p&;>)DuYJ5zla~D1TkY+LHUa};fkCjj-*2q
z<NcM->p+Ul9EJS7q#sI2p)p`LS^(D5B{1o_hEE~^$JRs@lsw5!dJoEyE$oYMTiXxy
zvhF^Dx*Z1JN_(JzrSl|5kNAIyNcH!o_6mOzDM@S~gte9Ik!PwtLZ9hbzi55iCjd8Q
z9StT*FGR~7w2l;g!`iUqY_DEVG?E)+H6ne}R7lk&TQz@hc)=c&0bRh|PXrbEPy<(%
zsW#++>p&O0OvQXbd^5Ytkj4+P{G<6*MLuHETNhK~ex+$t#~@+SyUz=d*4B;d5B}6@
z&ac8<9nQwkWL2gE6~@lAPinl@oK1?xW{?{+nNwMsgu>rTnnU-bI3Gpi{<!?ABe|l{
zHh87%hHI&0>RE}XI#;f=Nb%n0j<5TR9X`x;Sk1(P{inen_~>8Gv7Iu&;;C#^Pm092
z#U^>AA$Y4WI*d>>YFo&p4C1aRL(VHQh~_|7wPxHwoYf%|9x@GBwO}~!Oi0&)Va6)X
zLG-MP8rT&cLFj3Kv{L~oDnMzZB7iV+NW=lnbs2~bX>&jZXw66leQ84hKnCDw-HHG=
z(Y}BabuRCE05|*4Ulf=Q-H3}+aY$v8C-SK}AllIuA+~#o4`d(2YeQmNM~LyB=eenr
zukNGP!eP{4>rZ}#cFQ*;Q<4(RfI+3PoPGHunV}7`qWV^hFhye;&ftBaxrt9Z2lvnP
z{Obuw`Up~ribh4(mUm%^VsdlNO=imnUNc)$XNQlbr7b!YWq`i|s!176wY|?8ROFS}
zY9E(FyFK1`*YE+)_*DubWi63n))Op%4xLR~yolv<_h@^aFsA*Jkd_v2s9h`(bv;S#
ziZ84tv7E;!DC42*YI&D8DT9mU<{^$y$85YcI#_Y{d8#uY?}JM~3}cLA)}>hpAf`?z
z8g_R@sOJQ8mHw4xH&Q<EH4j7B!&OpRlc+K;7$TYWc=Pi|ocat@^H|od9aOZM&_A<L
z`Hn)r?4D_}=`x~jM^_a6r*j!)GL&4YJ-QL;vAP?%asIB|{{S}mKlW<OHKN?|rJ_iW
zA2{@@8@Qp8gC?{zdl_%;l@J}h%_i&(o`I<8@WbI9Lf$FkORIESsN>EVh{!nn39lt8
z#BLzt)~26RTCucZ{J&NhTz&3owWCHL4k^Wnqz1Q#XAtVvBR}%lLHZL`ZzmDj;zNv{
zoy`_0{3yV3H|29(Uxl^(g>5d_;G84+dl~{R$4}wb@+7*1@v>4t_B_{78-`z)1NES`
ziPhCc6an-UzH#-bxRMoa**coB7N`dY6b+-4_=Vzsv)TQQ(Ek2m^AY?D{x!>dJdi!@
z^|Rv>LQW5SAO5{Hu{#+WZml+vs7nhH(W{kZ&rW)Z==I4{P}7!F1-T3#+v`IqJLqER
zH<DUpoH_%_ALm^boi&Z}PK^|0*>by-W{Zl(7MUD28l*;8*JFH*mB<|MYeU4>Lf%6S
zxF5acaoas;xSBc1Exdu1`K3IQ=~OO|2f^h1<4jDJ&y6DfSN<s?Pp{rT`zWfrB$0Pu
zgzZTLwne*06yqzJ>^v)F44Rd<n*uCh!#t!Du}4!w*s5O|MpnCHZv@hdHjw~g0op>i
z&j$v%sW@g=05Cj({U}VAy^&#NEvL$i$199~v@1D|O*9N`Ng3qONY-9zYgfn0XaY%-
zxTYz<GytjAw6u6c`lXyiZW2t2N3K{>dldzFBtfb_pMP^OAL}G%{LX5uZMC_kg(r5|
zEQ%dDuG7LP{{RyELVHHT{{SCq&ac92M{}>WHE%V*_d3yL6Sl1^n5<>Ftac)q^V6+j
zc)nehlg6irW7+uk{c}VXrAtUN$Vu*M&Y5lY+xRW>fg^y;{nhS&0h;yRhm#d~BuA=U
zi_J#j;0!E|O0V}<u>1xEMv5Am*d+O+*s{P%<!ZAdEeuJKmd8^_B_pP2u*)#r5Jy_S
z9sdCPMw3Ih_PqzaG46v7w5%5^JBnhq4+j*)Scc%Qda{?Cb*XW=Ns8Bi_ok)JPX>~R
z%aF^vtxv@u5{y)gZadOY2Ru#%BXTRK%t|N#qND<f0C(`yXaJ&s3Mc@gfF8WYQ`FUq
zXpuQ?anMw?q18nu^(^_Wa2IN?(yGrZmd0vSRnDwMY4TW)dsXXFz;4Ia6y&rs!O6MZ
znzpX6LH4VVO6L(qf90cl{{XT({#8cvP<frCl1<LXEuJeC4Nq#WHO4k&6}ZXC{VGM3
zT;O%9JlNJ#dX`oo7bJi6>mO6VYzk0%n`7XkLgGKkp+WxuO)eb&0N*OV{86eyn%y20
zw@A{{WBuD7Q&L<-@}1cNp}kK{L;alzV)JSDD{*a>QC*C8RvPu6mpD|Idi5USv8xw;
zbaPW*Wt_gfpxjuU_}>xe;D0*Cj(3f|Zb!Xu38|~2mZTgVnL8rNyQ4Wil?lq6GUBL#
z;w|cNO<3JeUML-la7C6Ifx+uq_S#PGnNB@LV^UVs_cGi!wOX1H*eqD*pQS-|o>@8k
zIIJTk+3(h*snYjM)|Bq(K+R}cKKSs*+*Xfcj{`s6G*-Nf!KdCzoOzENf6b5nf-5Ew
zzli`~Ygo%mo;D*6qlt00gQu|pOGEgns(m;$QFlikD-Y}^ro-b>?i6xK#d<D>Zw`s1
zo2!rmV?5N)vCGuuZFOm{wVT_gIsMTpr=}HyWAa+f)*@S5iD8K5JcdK+r;vXmRF$2L
zUa(Bc13f9t8!F|Lf$vmeHi`hXZW`KnCm3L?t&!WBcCrfUw`fj6OzJ+FG+eK;L2VbC
zC|_EhYgV<m<n%cPfmrPH86=kb%?+LKsXtokZSIcWo2?f$TNak%cY2;Vf$Bv|W3f0@
zb?Z$wh{{hnphuGHGqBZemgRqVBqN`#ZTLe^XSx^3ghV8Ny~j!fwPc#+hJx-}fPkcp
zyCboz_*JaY(R1XtQR(g}Tx2$JFZP2+E<k=1{{SjW>r0y(wU#1BaKoXdD`|4X&bpPG
z%@6v}KX|>tte9?$k`}pr%m&phK$uHD&z2$PO5fe6TG5prOQ%(59Xgt5mJp|zq`EKN
zYn21v3g6>XUx1!mE6@&cLL8RGjWz=;h^h%B;A6dZ%D-c`7SR*tb(M#&_kR!PN<fv9
zXF1|)`<aM{%aA*I*Bvl2NLX<F#WYG?g=QaTKDp#oT4D-O)_^5pj2e$SUm}|V^kh0z
zGA`Dn0<TKg@X!2)tymqtSu*GS)h+!gJ&ULDuTwd`ZnXr9iewIjw`3ah9|^~u8rA;*
z-yit#{{R}wuflCZc65+-tFewNf?J)DC&m<0m9RS2OwkfsAKsZ;m56UYUf-@Om+>B3
zO}k^sC~3!ZI-ddM=tkaaQ%gY&mHp>NSp103`%l6D0LMdDuOrN-@m$MT0X*|o5)i*i
zW@MW{=}a((%_cNJ*1YtjHks{GF|BQZz%>kzk<&E5j}AR)GJR+PcN#bQ&;h*nG`o+f
zssT5jT9IYPdX@#aZgIwILJOSrH7yj2is6R_ma`~IC<B}$Gyz>m%xx3^QAvPNNCKCo
z00-KVfC?x8q%;7#whWq@-r-NRDQt95noi7--_on^+)u7)M7%__u_S85)}@4rZ*v5f
zFd5-i4DoK;ymuMxN_!gM(@&bh>epue@<~!L3`G9!EaXPLlzl3JwP`rrP@gH)Rt~GO
zC67!DxiSIxRI6gC&zZRTVxm<$9Tc+18%fAXrZ9{So`$h+CNj4?Rnw1CmoTpF?1}Gw
zT{lioRCzx)BmV$psH22O2$oL&0CalRw>-5Q%J0b2(^aL3y7AZw(EBRSYS`VmdsY*j
zneXAK`&ub?Mq;DEAzU{Gr1BDFCL+k)at12f>ToI-P7iWub)7=OKPupf?fo2XAN^{I
zc<isn_<wTe_)kyHg#{PQ6ON_VxjoU3ByVpa3)C8sE*4HC02<WhMvbf?;!>3Lwn*D7
za!LHF7e0rtPYV|q&6eVgTa1swuh~i<ZLY)XSkCPpt`<?JB-_x_XGc$!f&6PTQ7YKy
zb_YRG4bN8;*RpLHQryO!cH`?(+s?QQFg?X|IOuqITtr<>33XYNX^<!*GRp7Apb9^p
zYc3T9#SvDCHKeLmY?9_>iE*0Ak+;ZCtz0{veq%K$sIG?BgzW{^sHRR4IXSH#9%?Qf
zY}4gTXYi-HJl=*Df<V06-bmzM%^MXJ<9(}3x1adnZz2AjRQ~|zL+}*(mc9#rT@27j
z=~IC$k!>;!faeV2sK{b2*#LZF^Q(_zmIgjYS^#rIlCK~}<AT6oAE>AJhTW4@nX(U=
zwZ=Q;LH@M5mY2H!0IuVru|p-hD_|7_<JPU}@a@<@<JOpwO3rDaebGwk*ljhlX|LVH
zN18;--*CbF>R8ig>U6C^FYWxufl>|`Ph(n2H&ALwJjMI1RwKETeeOj7V{5uSmZ-*A
z6vaM2)-Qjrtya?uX;y~XMpsbTR$lalWKQ>D;R=C}eQK*&E3Pp?B`eDfxO4|ASD_1e
zXi{t(jC~C<qp{0sHnZvXMpX=R*Bk~FV@=a6^(#Rl{qN!%xy=G=TQffL($2<dBa(R)
zm!hwF%6XPqRZuqHRyj1$oVT&JrNTe5ZHsrFM?Zp~AJ(Z%##^UvdPPffsL^8k94T;k
zxR8G6e{_F){69ZR)mLRvhg|!LOSn%&qH`MHadda>=aII#>%_~3=zV=@l1D0&Z@q`!
z?mN|5q$VXT0wxD^UrLIb0%#;F!RDoy?p*hz0HfutO#|hc@<Is_+RA>JUAg}N$EiJw
zr|}~^P{Sgb1$G4+iuCUbuHOyVpYLx!tz%zgZ$ox&Th@##n&6q&5^}s^p>7RiYKbO3
zRfq$Im9l@`AN_p(Rn6<;I$hWMhN>SBRodQ1m0*s8irBqxv*|V#y&6DnWd8txaHH~E
z{{X<N*L$8^J{HPK+gF;R_f6)vJ?W519p|lR>F?(jBaums7MVtAqz_7x`Vl@Vvy9M5
z8%8Re%ZB2Z5#Z8dfEN^PGz?9mof>X50d6}@-*>HAhTW9&Qra3ct)n1h)cD0_P=r!Y
z2RTVyNX%^%08v8#rQ;PK0~8tnF-brY#RJNIT-Jm-OSs82m<swdfMgtfYd(8fVQett
z)`T>!RA6L%tJk2d7;MrQ!%9Zu>q<|d*@UGTw<&Cg0UI^|&*N7ih6tA-6qD^)&9t;T
zD8`(hGgl{Bk~Jk&2mA)8&voWEEDz~Q5qq;-Mj|mvD2-Tp)vH+Ya!xbr#Z<N@3e@I{
z+cG^xMIhVlT*td7p_O(?W9eBla>o_b4Ud(^zG=2-UqoJOC_v&nWY4y8-}^OLx`YX0
z+&)dX6r}bF6#0{H$ePKGte}h(CbaG&EBpI%!*vyew?2yp58ZOEg=?L&sBB_E)~OxH
z#pC=%Ty{AtuX^aa8pm}GSOdlmI@UGbpDZk1+?HSVYN0MyW@nDVs??QQd!m}mbGBiR
zdRC23BM0eQpFM4CUr9~1y6yF<^IjJ`n#D%$=c`h;Bvn5mA$=rg3TpMO+f9NvIjuQS
z$Mu;zMATUyXv;U}Oqy?y?>WHYvr*LgDb#+(?esELM3M~j&1tMF<>Vfsu4k2&#xinb
zdd?T@F=TwOWc^u-e<CX>CvE$2iXyZz#8q&c<sg;W<ma_EtH|P$ZQY$mhR2ns%O*Z)
z93s{yh^`IJuR9a>PSN;O$8#I>BhjC3v16W|as&C(w6)zm&5M7_btu2@j#u><{AsgT
zY>}QgJdSGhwUzb4W*DDmV~h%;A-9&-Z~(KbVD#@+^xrhb3&iaboNiv<jQ~F6Y}{0w
zd(vxjgC)D|8GCrnev7yN0A;1r(R?gspFS7{JOjmQ>dc13%Gd`O6v;VV?Vbq~?><rd
zZTZkEntD#Vai?WhYsSYU9<|@tYFpw*#|!}!nlyy3p>FS6vbB#hH_zD@mNk&xGqqR2
zw3sicBQ%EOTaR0`zef8!n3ErIxHX4ycG6vJIl%8ocQ<s@-WL_U4WKF7kqwYuS%9Xw
zyO}=4F=fu&<cfDHFG6_r8x2i-mSqlj+T0()xrwcAG<(UO2l)ZcJ*XQoHEY3pc4k!B
zk;wHmQXm6@PvumMUguXc2sA4^a9LcB#j*V=u90kwy5{75;v>la0G^H6NA>opX12Vw
z{<Jy^CVPntvY(l~xE_FeR)njCrjeYT-OgWBAL$KQ+g12r2%(J?wFUCkBTkLT0<OP>
zH8s*dv<ySZ194pZ*!9Ib*%2pZU=fN+U`qpGu}d-#awq|VAk?9Vs<01?;j!yl+FoPR
z^?3th_X{7_bc_7~sF8H{`~LtVJ2as2%`yf!2Q}(?BzrtNU_AqWO2WR#=)>;KD|*v}
z;<#pY!4n1sn95m|RAjNhtW8oP)ZI+XsD4Ps!S9drn&&)0Cfz3F{{XN>T|bDLuXE3A
zZOzr(7U_-BxIv!TIH&7VEzYGpv2hl5Mm-Tm7xV(Xey5nUhtK-_$_)5iR&)L1OoINO
zW#<s`ov1%r=`5`G$B0xce!jIWmlEn(T1Cyo<LEi2$tIMmjj}htN{vUMN>s3z{_UiU
z`|>KK#goOzA=CT9sR*pm5-Pjf7N7(chL`MSi-m^0t0>MYvkV+l5SXbUDjVkFfHo~1
zIITF*7<SENEz2emj@YKU3X6<|#woQxb9C0Oqc9^BlmNKml7I>-Kq#OE%}r|onm`9i
z0O{;30xu&KsT^vcU<!~V+P%eM_>*K-suuyA{*)IFh{Ct>;!t8hU;D?JrE@7~B%GRQ
z9aUOTQHYlNw{ce4q|Z4OEUZ2GbQ<!Iyn|3F-Ow5q##P&Lv2xzdX&4VIfP2;Df=IqY
z!5u|Rj;tDK+>up*<d4FmxG&HUYa4sVr7H4J(9e5$TObgB8l?8pE_WfV(?^j~E?XT2
zr*18!aBv+_BR;<>f&D6+Q^zcOB<j5lW|NkubCf8x23fUeojj>IIpEasPS(4la2_MZ
zbJC&3wt7@6D8))YGNt1cZ!!{Ddelie&Iv=c`{$)vPRAu`iB8X`TFzjXB;)xS)?|j^
ze8?nN<?D>_YAQv>%;~|_jNs|tRAXv!BRB7mES)zmdZw0GV94N#+7I14%uX6Jt$66k
zu48{FWez=YSyGHOI2@SN)~sk!YUQhK00Z@{4j8RdHj(mx3Xb7e$H2~OG|#1ya>KN{
z6=Q*msN>t3zN+Ck?e9d|IWqb(gj|_d#2@k&kyL-G+Y$c&fPlZ!vZiC^627$KvFGAG
zH|iwV2LrEDOs0-$9gd5Er&*%oagezmL0oOq7rK>?cPI6tJCMf1A&PX6byf^N3b(3e
z+giHvjtq0LLOWr&r_`nJ+<vLHE8}qz4VMZgLD_e8AA!jHD!ID2m3Fkxk4*7SI|_P9
z*o~EsQcu>dX}X2YgoUCx3@}HnRf!pOeDkr&h0ma;>Sukfn*OkrKZK3{0M@B{mmQeP
zG&~X4JX5thr+h$i0&&lJ2{WjmL0g(8x@;rzr;N)8<#2gg2A#Dt?=N2B`bow{_Q)QU
zX5Q^&`6U4TX;^99m!W|O5=2Nj?b@|0qYlNNlSt8>Mv|BS3hL~wx1h~SA&tJYVHr5C
zA6}1XY>E)mID5uXaq^s1mb;SnG)YN+EOqWhP;61RwKun6V^UW<<FKTf*d@y<U?@22
znrVvK8=8a-In!+;Awt{kc0CjpW+eV}!z#@xM*t$)A&9~1eOUXS{<SMy(pn>=(yhr*
zdt$U?gBZnCj1Es+h@`}x^^cr2ODP?vf;1a#N>;nOK<fbkLFrw^;-kjU1s+*axb2FT
zDJ^bzE~RheUtB}e%W^7}XiC@t!K*{(j+z+@DF&1r(*f13YC!Pa;}zxZB0zqM<o^Il
znIhr8$(gkAK>~F+73*4P#*1b>3lGt23i~6c57fI5tI;<V!6U9Ds%h}H&oX2G04$dy
z{o}y@03X7$HiqU*uUzkoX4$34zuq8Ll$GR+D!WCTmX(+zy}sx7FqA|7+x(;S+<ywj
z^YOv2SE=XA;Y<4dv>0U1tu;^?GfbNo<6=-?{JF(%U(REKG>_$7qoqkkR`z0WPt?<P
zvFkvGDw?qrvPbt$N2LH<ghw}E52Z+LSK5(CKF+<x1{L<H83gcTa4Rn2+mppo6GaD)
zXIf=4Nuij>-n8wnxa(5ma<(+B?p231X;>a>7Md82xu_X&R5O{RBQ?}!4Nag2e+@Q(
z6=`uvfZ8blR_}-4-^m^N(qMF|xF8OdayY>hfO|#7ao#RCx`hGfZ%Pq#*SuwtB8zYF
z5zSG?=QT$~T`Y|o7?(SV6{F-AP;tk&tS2o`OAAN$S`iCHcfl16z*r;+8seN;$yzCD
zY2HnnJ9SKnG07ZNOqE=-5>InZM!0CSuE-~ts5mvBCBu`}oMN{z#8r}fs9|%vh9mK<
z%NY-uZx%O(V5g3>svw}-+@HlG7t_eU#Umfo*EY+z=O;B()`nQAIZsk+c99>BYfH@D
zIrO4P?91f&T4Jdf#w$AR*apZuTiS@TOu{#&nH+L5e6n*_VGaiCc@BH6RIP1J94zI}
zqtsToGO3Q^C$>#z%CcLwP?k?j)W$Z@>BddFR9m9$jnguahX?zonq-37Nk4xdmfos3
zsd2NqHjPMDq`jihrFN0-+JqlMIL%j5VUTYeRi~ll)rwMbTOAIUG%f((AEjx@0(ng5
zpJQColT7-&CnV@!$jY>3mg@`mN2O^;<^7`G5s%(TtNPVRyO`meV?tB6bDX=M&AFO(
z<U6zeXEjrDnGQN0wX!~Tt4-=H*b6AfpfwOVrjB}=Na^g^VS8Mk`FD2u8syyN*!$5F
zHa4_1&UGh$`RIu!{q%tU09t)Pd@NP}07=n^{{WVjkDvGh{{Vvz{CcV&rxDAeCkw?m
zr?pW8G9_5K=SH({?#n8F=qLJ`mQ;3S?FCiFPf<p{X_oVma$Bw`bsN~hq^3b4#LP0h
z(lRpr;6($mr{SpKlUKJ(nAS*`?JC2LYqHe5A89%lw9UM|KnE&+3Qd;?-5jL$_gZY;
zRO&~+V_JG_#n@DuOGI}13hrTD?wJad9=NHJxs<jo-P#6>;8&gay35PS$6D1=l2;T@
zE7l|1G2^{cibjm(NnSC*sVrD7CASWU?5h26dey5h-`pauOJ{&P)3A-75>57dgt3fb
zPbM$4HhP_)gdW4M*XdDdsm^GQ#nt$l>~^{`se~Ub&mBOk@@>Xg72~+8FgafnMxr3#
z^P1(m)^b{v(DbhW5}Gu`e-yvcx--U0Da}jVZ*#@`TMI$rOT}ftX52o8u%sAcTCI%G
z+Lqn0*zHJUtSUZ+s0Gbx$Vzx-+-~AqsR<wMssaB1p&CrMZ2b(}`ctz*L<)oEU#J!9
znt1;JiA}H_oxhUTJ${EyAE|mFp0#B-t_f~+$6`flb?Ne~qfYW4F}|G#_5ABISiKuE
zWHrWk*5mz5AotuwbYcE9b60-hk$8U3q1r*w?isEh{n7HT@^kps6zhYLUWz!XZjlCR
zNZYkCQ!m&Rks33@XDjY~D?(8uu?U<IGIUeMMN816LYxc{gYQqF1&893p{HORk`!RW
zwP{Cmj}RxXdX_W|I%ll{RXq(&h9QxCY6*%pFfwko<korn)ku~h=|ZtyhNwl`d!o1m
z)}_7QsmQFPibb!stu?O{^Tkb=?sL>vNdi(Hv;dS)0ZYlGz)C3qgjY}Dq6~P?G?*L7
z!*f@nBNZSlk_AFZ;Bsrv^%>3G{Ij22FVcgeg4X0Q+q}e*aHG97?S5=2299I3;oPHg
z?w<pv-UO{``QjzP>(C13sjRd;N(<Q0yp0$vgwd<x41Fs;>q<Xp+)2e+x;&RIcV<l`
zvZAiwcpOz}tyPzBBOZpQncVcLLMj>=H#V$XggjHM^k=yYcY%Rj5qcg*CLZ%;+kX$l
z@$P9pVboT}oo^Mj{jL)`T}jWN=B7bKv}5as&QT}#g>$NQlbX0}Wm{KrXjq2?gNo6L
zCY3U(NgjfxZ7t61y*KWT@+Ai$8kutovSpO!rZQJMF;ryfHDZIKj>qz++Or4eAlA=g
z#i1ESqPe0%$Ub%bDvjJGGnM<boF1pIkEGo$b<nR8k27-otyb9YO|q{wQrzajKXJ#Q
zk$E4NCmF|TdHwtu{w1T#;<cwLVP_lpnG8lNV)F7;vDmpR+<q0DqNOBv;Ob6=3ANQ3
zR`+cNR5s9h@y%-KGWoZ@V=*E)q<sAp<MqWkN2#40OOjNZDQX^EP{shNNXT1(%{8%(
zDaJJ{jV(h+x{f56sBl+5FKWmjhQaGrnmDRHXLEMeLee5&f2laGdA{-b(uogYwV|r>
z_9;KVY@eoA{{UsD)NA2lX!0uFh~*$fI+$FZ!MF$TI6s9#V`!#$<ODhrtPf%;x`X&p
zW+ceXNw<!onIl>(uLhkoXZsPj{{U5TKk?;1{6!{IdXVWiA7{7)zUy#nsD~3sAFAUO
z5<`x<85*{Tz_iw5E7g#8`c@7A;lT!)WXV{1#@O8%&+`<9N9e@<G{CWCc3_<!10jI~
zcCTPXc`Z^w{7MZbX$Ep0Dv*TQ!@Xr`AnzO=X%{VzL-6>(TJ<jh>7Uw{{#$3tMcmk{
zrjvJl2t7<`7BR!+h!BiprFloi{{RfZZ*3LTxY~}zSP!RK>BGwMahA<8Y3nO9#-i~3
zjI(c@T3jBCc&=B+a7S;fB0`FZ<qwrc1#z0SWMC)+b6Yke`!3Mkw-+)m?pL}G@BaXY
z=jla-9q-tSR8n<W{a0*%_sW6&D`!Qud7?WA7#XUHCTrejYvH)u#~=g@FM73)F`Y={
z3ev_ebCB_JfUCERS0Z{=Qrx6^w}4;+!|3O$F(S4#8;CUcTH^9RsG~S0vwM{$c+ZQq
zi1n>yt|3jR7id1X?^ts#MQpYpiF(j7tw;-5-xwmJQMlCrgkrSy94?Wh$BY(9B|iCb
zzxyR8(75d{>#8#W#W)lAP{<^N=0qRu3iZuApX|7Oq(8`4HGYRyAE~1c-Kp0V!1p@V
z#IT`~RE>JE0-@BV5?nc7xIR_-4h2DfWXZUb8LvLpowXfxVu$8tVH%Heaa~wnisvpP
zRFW&}(<GXlq9d4M#4+Rhs5XQ6<3EjMNWk^3`0R2`-c6NT?;1hYsgPZ@>&IH@>~HPv
z5h2u&ot!MG!>9PxRTJoI8(&gQM@e#~=Gs-ei>Z=7yg%Mr%fTH+N2uva61p3Whbb7Q
zvF5ZPji$R5Q?`(bpKEN?i+I@YMXoU#?Wvz@Y*An$l2ZLeMfg)f3{yy^0gP0FM^Q?^
zC!Hr_N1ZRND-Oi8R)P@wQeuD-mj-|ok4iz-lK?$wDFD;&TACf-+2_INX)rqlVIqc7
z4r)bmDo0|LwR_~*Fb53(04T0THbUHtoYIliL0K2TETSm*9nD^f+94R^?j7-2$+Uae
zct)nIWx}*@{_a!x)_u8mx=`8cL{-(<$Cx=rT&d`HmN;8reW|U98<F0F&x<qlW<_NJ
zvSgSlkSjZCdlaO-nN{0(&T7SuL~a2!ymmag?(Ds$4scCqs($#m{s^RR{BYu?OjOm<
zIqRe^W~ck76^_vm=O^+NN>((DB_yspXVm*<uh_Dw=OUtR==3mfQJH^gllMhOaT57`
z59e5?sq9J-lv*<kcw>TpIubveb^Pn2n?uI-5mH8!ar~p^Z?0)BpdvHBEIS&>O3d~p
zNjhyl#$S-FlB5IgS0lFbW5z~DdeP0;^01h7&nWjYbz4_P83*f%qiZLW_;0#;)~VfG
z@-Z-UaT=Dq#8dJfDIW6-3D2OZOH%4rP>qol(=EmtH2{Iny<O2qm1n3-{{ZvNV;@q$
z{{SU4<*|f4>fGqBe8N#0uTfOImgm${j>O`hHAc~^qiZtQIEv#kTOJCY-%8unw7YxT
zU$og^?i+z$<x<tPM=~=sqz$V>9L=*amFL#5E*EU_4!B}7R)qE=tqbp;P19e|9y9e0
zPx4_v<FhD=-Fpf3OPgt}V`pom1-6X*<A(hD{xyA2<E&ac%@4ps%SXL%4^jus{{Ysm
zy5GXuTygC%f-%lO{{RZKiFP5p@LrE`86IB3d#UMMrnlfbn{6`iSU=hnx;RW>xZFRj
z4k@9*>-U~(J-b0ESrw6#;DRtkY3VwmhY?A@>`i4Q&=a#di+5JR2Q>zvqT1i(nS8l_
zzQ5L^RLf~Jb9bH}zO!P`k2wc9PQR6BG{swO#8^G}#Z|~?az}3=`CDnPSkg6^CDLFF
z56Xc_$E7AXZgT!3c<(PGNVyrwt6FZIw`s*Ev81*;Zw={_T`*V>oE}f5dar?EpF_TL
zI_5bP0!M1uB?Yfg<+42mwL3kV%&G@|Ys<bUYf<Vkt-Kt>hdK2$p!r?Q;iI|b5&V-C
z_pS@#69cX&9im#y)$C;HQN*F{ZpZpGL}Y)@M(m^d`U>tmJK;HX-M6}-F<Z(*Dz5~N
zz<LU}E4#42EOk3;p8{xiGfM2TqZTBvex{qJcpk#ma$vfgq2p-){<X5TwjxzM=b^i0
z;X6xU-bC{P!}z%sPG1aOS<58Qy4(ZKIvQ3nx^Cx~_{uworgXW5<2lIOF`DF<Rx?&*
zrDNHAE@RXDDQP6Xd6ROlu4~QyC0$-zUdCL%+3~;!jMi&$NaAB`FwI$RRMw&@o*JIZ
zmB^t0MEi0LO$jt09_R?Gn)<YV+3=RnmuVm2yD$F$vrzXgKMOND($L8r7Zb<+@cwn{
zwkQ52arXV^^cBsB=yhTGnou05sq2C6buEiFzq^iUzq-2@>VNv#tgDk!TiGKMSh(^n
zrIS1=*At=1w(jpxZ~1qLq*MOc`B(BMKZSPTm%7=>SL0uKku;d9?8U3F;uvTY*cR<h
zO=;j`Jt`jNh;%nn!zSotWj@BNeX1$jbil_U{xswtsHwKMACW3sG|QNF%s;&m<b774
zco|O|{{T9xlE!12M_woqMo3OcsP@x9AzA{{0dOeCwE!U$(kKBjONszODJUruQB+7$
zH2^eH=70>E4JHCnNCav{BpiwW+0!+177|0nZbzxc%bbkVjWZWmxHH@_{sjOEr)Ha^
zE_w=<cGFuBHRDFjv-7E<)QHC7X{e@YPZ=t1%VJ3{M29=7-k9Q9$G0>!WKCR0wRbxh
zWD96JQ5%9QBPg>MSzu%pgEYT3GgTCCDvl{wbTW;@cTDqI*6$gd`CO?LMs4#KtduN=
z6{`$q?DuWy7C$ltVLi@_JZ}@8pC4y+_pHg31Xdk$gH)N*4TR1|3yOWSjreY(nkBK*
zMoCE*E>NO}PMvX4dBRArGoP(wqtx|Wtu9toEHl!d=Egwd1XjnJ%#@wWv5zt^P=oAh
zw9?A}WpH--R4va!q?}>QwkY|5nD(pMdluXWUe$8g^YC(B)LZ1=6A1?+y>qhLktZCA
zDLpJ<n^IDoWVb9wYk&#nmUxi-uDGU~xy<nKjX!&-P3@T*F`SCkfT+-l$<|wpbN<Tz
z0QPmGcR482bmY_6%9wdFbIuPG#g{Er9ga%A?3_o)qDAAaYItW;HkTh~oaPA*IqYa^
z=*4bd*YxnWF<nKFxx-?)Ik=I&rjpoWD?(43LxBGPa^gYxs(*zx5!sRHwo>XBv0Feg
z@*EFudiMVS3u!j`UX*R(0J{9;$mE`f`hHcS?l>C@GaeJAO1qCzYRuUxsi`8|5>b=)
zf2gMR95*=M9QbA}YR%@hK$r22z>Yl3c0Tpuj*OBrFe>VBLF{T~x`H+IonfuYq{NjR
z1L<9bozhPjWFUj-STt!8{A3z!)|+d1xSAAid~-;n2J^yEF(%c@`!#52R?=y<GTU87
z-+OYf<QfE{W>iHV+ChNhr%LH9@Tw1b3X$wT4s9bE3~~TByApkCw2lyQTGsg3Rrt<3
zSbh4W=>a8kfCXi0S~cA8mJuIXQQuQJE3<;tZQjoNY=dd^HRe7f-KhAYc;hM+cxOEo
zK;)lcPNX>=rnYuSp>Z&|niy@-g{((xYDXtMSo@#w9joXq9xp1#Pe$Viyn0iDv`KH7
zqG0me$b6u7=aX96d_G#_^IB|T-lmc;sd|b;ncBXYX{kVByn!V`JC1A0JYDeSTXk!l
z7@1>JlL_hvwI;chnQ)icWyO`=&ebH~MG~N2PAe-^w~tcNZy}u(jAt>h>MNE~<z~==
zl4$aoqG%>$kPymoij_`RHL7<pYR<${Cpi>Gdm3UJBH^iGlWFFV7eNd-tNtcY{jDM@
z?efw%@#(b{f29rTXqUMDU#W`9R+UbA&`9|I0N2O<_^(^C{{WC$`$#|Z8t24xI&k~5
zO5g5^oOP}Txzx5S-G1)g7`x-lZ>j0}WAm(A$H41T)VXePJ}<r70EZ_oa&cI!!qda>
z#yES)Os)RnWmokt`B$SsS_t5%?<V^r8bgwEOs<2NDKS76ZCeJkl!i}wm${^NB7XGG
zFt2Qi23degR7(~|UO~YatqXl_WnGsjmnr`MO$Q_KrC}G~pGdr1A2Qj`ugs(T{c1y?
zysNs-NatVn8=uJ0w&8BZY`SH|yo|HN%b!t-&A5Q0A4+IVSgd%Ws05=l2xB!jaG(Qt
zq~AaWN$o~`s3{ZDRT31#)_@#7?KXfEoY9I+0JM$PkPRa=+5jn4*?JQ|(SA{~-B&qc
zGEFsP5nCSFsIlJ{XYw=D7jKkOx)tkGjz+qXwnw#X_;pDx#)q(z0q#E$z|`udVlBdh
z&MLK>(qx024)vXiPg0hw{$a|Wm=2YMMoGqVP9}}s!l@#!X|gw(dF@2DDhAz*pEotR
z44ARH&vJ?U&-AP$j=VgyW+TQVx@ed9hHDP-j7;&80aMLdT8=d=nOE&NINUMxr4t4`
z6Q9bo<uX{#mn|jRY})xfvzn(ouLI|*9^lmSBzjZC&Q#R<6^-G~1d4LqXS;q?Y6q0@
za=CiW=;H$<RJSbhMofKi$j@5MtxsB(CRJ!y#cQIih98|=vAzi?B!O2b-ZnhUJ}ywG
zn>JqICybAmIP03lk&JH2Mn-BWMeNU0Fsh{(&1zn<XylJ4ITa408W!Z^@T9&g6;;A`
zZe(XiBSSFUa-OxNu9ZzQPxAi&iCSJgOA-G7vWn5}k1jLjp)~hKERz<44ap{}_p3*j
zQhd(rmN`9X(vRX=WF5YZD(ZSZ*)Uy3#{QMWUqsN|%NzApBBDEuy$bcA56z+75AIJJ
z55zb>(y5LcolnDyWfD(x!!4bPeLZW^?B&}sAK)Aj>O0npR5<BoXsmu)vUS17)~&|1
zNQo;u69pKjqWsJ~RQ~`fr?5lfMG4a!j<w<+5v)GP;(LdUHpez4UO4Bk>q=V*9M}sW
z{_Fn$I-1V<dnq!EaotZtSx9GBr0bB}kdDGRJsZ7r*7EKGvw9k*a(rf)bTbk4pk`EY
z=HL@roD2$0f!}yXUi&q)ajSjNj8|7Dh}ogS5t2JrvuWxL-&R`ib-G^N&n&+)4tcK9
zOI7mmw0=~&8Ow9V{x?{83u8T?Q8SYn?Vn2UU2XxSUeBl9r<rhAHPa*YJwk)%I-h@~
zbYdECr*7YqXH4S#nxB!<_<v8*rnR<~J9&J<K`PR4Ij>cUBG*xlh?wK(D;Vs|BDF=>
zMieTH4@$opU;sN-hD@6t)ExFSt|7i@A9&PLNH9v!BhUP3v0KQqq=F@s;keH=p0@T-
zX#ks|UU8qhT$L1go!Q$*;XM1sdP~H@@*S@U;hBG}b1}DSx?c)0s;wqS!y>C^Yu1d+
znSf-f<Pk-N%UXV9yM`?N_f;Q>#cJ!a`TEuTfaDq92itL}>RR{==S<aO@}6ms*)6c$
za|r(cY=1iS%U(3!q^th`pql5zbUJX`vqs<Uin(hSn{bFu0X&~U-nbs;WL4dSyO+(D
zNimF`2cYd-{;4L>Ni9N7?X%4_$=df*DIQzq2%vW~-F?UQlR+5<D^*)%$3wJWpT)9&
z8ue+bB#$~y>Ay2KJo89LrE6m$D5)AGgyE}37T{F9%Z|c7dX9mT!21IXo<=E*+?;f(
zLTGpy#tscn1;xd?hKZaJ-AyYA9R-Jr?xRs1;zx5ALaw~lq&J=()YlWm_E&FSR8&eY
zbW(lDbq@u_G=6rOav6H^az8rdH7yrVu*seW*guIzMG>DyIWERge08RmtqjJO<E=2Z
zXaQ6f=~pdn*~rMDfv+8k?BohjVxEGtyC}yrR>X)#X|w=QKnezFL7{-Ol^`_nMF2GQ
zq{dhQLmJ$+;|LU-k=CFJ(;b7Se`=GbeG4Krjyx~oeQKnD?ljX1(dDukvvqV{ABzx<
zI+{}0#ygstgR)H?KN`HP3RuWS(&TsRSvSz$Fl}gZ&g@x<C#_Z+U5QhMt0nFs;NiN}
z%{L2{_2QE?iZo(!rHJiXQQZFk#I<+-037)L0KQlM0J5oQx^HqN`E^LM+*UTf2l`~B
z4)t*hUg*O)Qbj7ZIjs&BUxn61`MT5a9ZH^R8Smlbr%f35keFp8ebZ4ap0!Bu>qptt
zaV}ZP6@_Ue%h{?Wx#{5Mlq^)eg&gE@R6fq@lDvUbwm9ok{o`>Yis7R;J*ru*7E&|$
z(1F%&9E_SAqU8=pwHGr6kDUH=t#@;!RBDGSDyZD}WAd7_t~*0=w#~{%3~P@=v=jQ1
zP401XIKi_VoYGYsfm6oW7&HOUcrJaKgWIv<LC4mh@s+RH^)2|~L7tuJe$e_H{b*_b
z05+egC!|RK0F;V9_G)Z!-|ssA00Bq_n;;!ThP&%Cg%zS7Lv0TDqknCu@ag(hxVn1P
zM4F3RmJ{eIR14;=R~a~}Ye7TeMLUG~PjOr~#H}tny$SA${`%ME2h<aSC?+S2I2*oR
zl_C&$kbX6bG_b20_iTDF?OPT)we5z`&F011xcun_?0P1d=3Lk%#PCT7VUQS7Uu_Yc
zq#vQEv1#1JS%;sB)Q{)tY9up4$Qa*>>FgvVE=>}82<v<|e%5z^hejA8y9xY4f+O^;
zVD6C;X!4(lcj>A?i8ct|kTYHZt;VrxbE*s~o-10d+AFbF=88Kn1UpS<7-OBT2j*+o
z4-=ndKgyyzF|E-xnrG%Ur4sk83|THcX|6pe5ne&XP`Yd(pbihj!|m4bM%@IZiS$ui
z-Oa;EA;OT0@~FpQTsV!+>K_e>CWm%~aFZzg`N9VNb><RA(9G!lRN}6S`zCePyvTld
zsAgVrD|T~^iMYX}WzJ~~yW!aL;te=5?om234`6bCK~++G%;fZ9O)jNBfqMS{kz^c>
zDjB2CBAElU#oeh}ulr)XZrozQy`^8|)ys(89e7@)`)*dWVm?)@(sys&ujAMB{&mFn
zJ0ZQwF4#=hC*qs0BHl*M4lpUfUP&3v+AQJp6Sq*3`T1Ko+Ti~Hyf{z%e63}%EX0$N
zPDwTDt<DL2C`tjrs%B%jG=N~z8US$Bawp(YWki{qie(~#l9~|pY*ewyiefHXh~>h%
zgY~JOPDB2s%6lF~0v&5UrKZRl-bokUGsop!K9}PgX=7-0n^PkZfh;5=531t_^ry=g
zxp9|rdyfoh9wCg&q})ViJnSDM{$1;d);ueyX|gbQ-IRC62>dEnVoZ;+KfMcR#P_E3
z5oQZK9&1Vngky@9#)}3*I#U`(sIf*RC2l}daX<zr7@!5|+Ln_5qL2wmjCG&{Vd+9}
z+4Satt;ej}mA4A9V;C!+T1@Mq?@bn(EN2|?Rb+vWMKv*cn^&aT(1*!jc?P<D3Oq!B
z(Kn&V7&P@SUq(Fd!l^dkr{?0Yd!CgT+7~mVgtq;?s}@jKk=He>P6~;qa!4g+Azzx)
zw-~@3C}yzMH#Fc)%Hz2lRM!g8>Gq0x?~(rikp7hol}TRXV35su0X>c?9}x|gr>`Qa
zCb5mDW1pFF`NnhBqi9cW=UUw2sY&xRO=>o{&IdKA9j5;PnR0vjR#hcq+smnQLc?z?
z_s8=9J788V#ef{fe+nTpJjVpNxt8pT7|7s^8lD!x`Ki6mDlz0qGD&uvVA90wj50<(
zl=>9(sdClDgjZ4TjD>o{xVM(w@ajjs5tl|*V}s;Mtv#7FjmF6^+=J9rtNU~12_fYV
zQR!6lW)#=8pzXRT*(*zR6l1G6P(G%v_<;SRP`Q<Qd5w>zc%!L|S2ZS#YD%MwDa+P?
zFKP3bH4Aw`BMd8d#@BG(GKj_=7AnK~)V;^h<f}tR`J}p!`}Wd)gLm|%_8dMWbzTG5
z))yH609djO-1f&6-dVm>W~mgmHtlYmE#48gV~|fA@mKB6aau)KQ<IWO^e@CRDpg!E
zpF^6nHKda&+$Ppx?_7t(Sv;Q!-(&dpr{+ZgYf8t3s2SwY#;|3Mf+=T}V})7WxBw2d
z=^hdAqS^(z)V^s?A}-7IskoWHulR=BN3|h<c?iNp2l?W>!c9i%^6lq$+F0iR;}nW7
z+Dhu+vuE+HhQ{a;GI^<L<3{a^`GjJ+D@=tI3}&noZNUececk(r{`FiPj9R(FYd5A_
zpd@dUcQxg{Aay#V<wzi&qO_E`ja9i$*WvQ#Pl?FJ`M@8Adifc+e1CSFHf421$OZsa
ztMRS{YGO!7tucobi7UuD);^%viN2JW9&f7avPC(9%WoVO#z{0-+}uO^{YAz)tUE?q
z>*-umsNAp79r22ce8jrdp87!?mdNSRRTl(T3*${E`#Sk#LQHcE^ZX~$s+8Q{Ga|CI
zY;)>!L;c#^<si=1so}w(4c`gr_cywvjbjVTEEdyCgc&0dupg-viuzWdpL)jVePM6#
zsk*thvzw-xT5I|V7?w8KByOkCl!}#s_o_0pC()w~tlIwoy_)qX2l^DF{+g%qYURXk
zj=VD5ziYteuipHT#SixvbAP;f{{R8_R}<XpkHXoTaB#d=ocQYCnGzs<&1*q><`w<}
z8Gq5dJO2PpFJ}J$$Z7oxH~u?WWLKn@Q;K~Tm;4b8B?JICr{;m7r>#vBW347qECL9i
zpGrjtO&h)F0Nqbk8LAT`j$NmYwSEiMQO`6IUWV;k1QHEQ@mt*g0IiG@{o%nB0SVIY
zEwMd}ra-)m95Akv!~QiXB3k&419xf;)+dP+dmr(w@bHaFk%XPE)N!@-IvrcVx=yce
z_RkIK5Zxnx%RGgZ520T5z-oFm#+xfG#4G}yxU0re_BEY0tk?DTb}n(TM=Cl~!&eg}
z4+e|_K+GB|OOOVVm<VYF06L0H0*XKs(r5uGGtC1-B>7GSX;^%yrMgm<$6goX)W0R<
zA%0LZ^s38rGDdi;V-|W9D#a}a`H{Hwt<5X}3YDsA$0cd0EeJJ93eWOL7?H>wsw8JI
zg=IflmWNJEb+R;!%8E(FVBJ9@Trdbm52ajf)XJ^e(6y)7xj7)_wV<$dl>;dojZ1o(
zLO$uOabxy&1Jq}&E&Lgd(FawBKbtZBwVau)6s+1R+Q4JFR>yj;ugB-z4W|bb;`BD9
z9%{)WpSy+OJD6lt3pyS#iq$)jT0Y8rvP8kgMh$4&%<8>K6%`Lh10<Vhw6m*>IO3_?
zLaN^>$gJa|vDJ)9aOUhnkwqeq*w11rrNy~SkSj!au9}rTbZ0z;Sk%zY8?9+=4=Oc%
z>oZ~vzE~>Qq`V-;H({|@uZa2zje7=WHM~+bQ!3x3G|4@m$spW@wrKF8-8JnYv2Q^U
z&#POw`C2k%KkyN|`c_@YmL@9X{{W2(b1IAXUeX&jRidoo3UZ8^0J2wXk$r0Y%Kf7D
zV0hfattL^Md2IHwl#pQdKDBehQrN|-#@eitM`lt)c8PZ|$`RA?q3X08ET*}(FZeF#
zTWYQtpC!iA{SA72Cu%6}Dv?WLQU?SI)U|$<q9K})o+{F=M-^z#sN2{WNZ7&Wv93en
z7MXE<;cKYv;%1jBKm)D~OlKZjo6dyeQR#n8PX)NNxVDhTa|DsNJcd#$Lq^aqwK;;t
z6C_>aaeylul6o$Wqgd!r?gd(0eA2yS>z*)<Akk^WJe>@khti~0>~Qxp%Xc$LG>q;~
zB$~5iyO?Hzjbw}{&0N!@{qP?3C}(G-;KsGL6JbS5+zHZZj`8<KX|2qTCtn#2UT3J2
z_LmqPYSC+B)cg|$%N-&<?7@%Xb6%#rNi%*QtyDHNeLWd=&3;|StzJshz;kq_Fu<Ua
ztj7S>2DU)c=iD6Saq<poWwB9LXBVbHVH}vZjx#TP?X~{^QCFo`$rjLC2|v3i$v=&9
zt2OSOex_vlmBqW5uWc3}c-<OpBj{_A@uX%wJP>@?nS(3TLALeKsidOLAzGemsN+N~
zrx~Ux+C?!P4uNp?_FApPGBj*0yr4lKo$jOgX0w*+>NWk%%<=yKxgiviPjhD{7{<&O
zS=C`|rlWWd{dRBht5NueRF!tie{lZ*@4kPXE=b>DdwQ!*W@~tVTECZ8fgV`ddCuri
z{<yD1x4o0>(`6uIy*k3a{?*G_B+lGDX*I9pjV(H1G|T3}2F3ZC362LRr7g|ik(r*_
z0$AjqoBV5)?sWT4cVtkwy0-{-S|R#759?lI;!FKDO<G?*!gktr{Fgg_nwr}VuPdXU
z5jb+I$MGw8!R+4SO&Vpqw)Tr<8AOsz=OmKHKT>N5o-4i=<<#M=Zg<PD!YMISWF-Qm
z1vK<DnP{-B#}w?+C}}j@&;goIaa024p0A1sCXQ@(t8rbG??4xAbvYtnLmoR;lTBr*
zs9U89{hfFsm=xx<z0)0Lv6-YrISi)^E344_Z)>PTwq7QS%^eot51oF#wB6;qD*eWI
zhr%i#H`+w2acbO#U+d^=i(v6cQbj7j@(n3LrOfsZG2%dJ%GH?6N;6hKQB(s@T1o(Y
z6(AI5nm{)+5ljR<DTMZ=aiQp&xoV$r0|SbdbW+jW5+eN4$sWOxQ*BPE$K1Gpk=)ls
zVIf^QHus`SLyj%p!6YT5GO-x<t!)s;aW{~lNchNFgP}Viwt9uNo~<#ukITkEQP|cM
z_M(zq%JC1BvBqk-S(~V@bVnYf$a8^;o=tXUR0<a>(9y^zUcn}>J1*%CPinV!s;XZi
zNs+cF!T$ih0YA>7RPV8^JWZCRn_WUFrTJSvl}}KYH=G)aSy=7DO-iz8t96Emn!>t_
zWR(~-r6X+9qbjVg8Shu^ulU2&N;f?0CN0vIyB5~%ah@r&T`K|9{c2+kPfETqN)|<7
zIVU95nB`eMR_j_cdAMv#jQM4=E8SSJVpEY+e$L_e28*IG7-_<HCP}1^gPNJHgU5y^
zH7Y9RpIDNMi)1VFNI9!_4g0XJGxe&MLNM>$cO%kNBVNk9Y`0(!`@}E)%xay;l#iR{
z1HC1=le(tVN|cpn7*wzfxD?RQrR46qwo&{<3bm_1Z8a;!9b<e7n~aV}O->wc3o`aS
z>mF!c7uxDYD87T2b9X_vjV&a&m9Po?O?TQ&;Jee2z}oTrthwhMjWqQ%j>y`xmmmt)
zx04vHW3du7?PV>EgYh+fFz2fZ(HOD!5tiwLz^qMnH#$w&h#xQ}-~q=Xr|g<HYM1jk
z71yPVq;l?&j~h$mkHW0pYC5-*zRt=S6pjn1{{RZ|<ytY7ndvHZCa$&hIDL0g)#YuP
zmY)NsqK8rc0FJP%HMz0eM3xdE{{W_Qk^I4~m`O=qEaaEE?KqMv>5O<ck$YiZ{3}yV
zhRQ}BUF#3}=?NdmntF#U9?r(C#gdr6cXw=ltXKJpr>EH6{io$Hxc>lRWN+kYDR&QM
z-@?;7%`)2IB3;nS!0WW)v{}jzn<MK==$8pcbVU`&-PWa=T*ua$F=sEWa;;uzsf=AW
z)~yt_J%hqqB$rC^u*eV)N3gF#gzpXR!}`@nV@YrFG~pOfD#9Anz;<&@ev}eZxj3#*
z#H=Rr?UL{8(-XXV)>R+6Yfs`i^3spo>M+MPI1?zzU$1(KX<>|rFOkttG0LYn_XfFJ
z)80Dmu7qzj&7j20F6>){d}kGxu0*#s652db?R75`sK!k++Ios<a-LdOYmj=<;P$SQ
zkY-3YG>kinb`{NAX8S*e<|iZV7f7Ryp%jz<0HKPhpGpyOdoLm>5E_K#ThfaV&Cdy)
z`ovzqn)LfKn`@7%ALm?{kBIEUY-`!#D+hk^orCB(KhmI>j8`Mv>X$@i-mnoieJjWH
z89eLhpVi5(j4tft#Jv%FQ~|WjB0vxNhm4cli5UL?CKZ=m(b(pxzuox;d89pQkR=r$
z6y%_5a%J=s{?#niiAg|&&omQ?pbXMxf^I6Y2BcXqXb|<3=B!6P4s$>nc6ysf!yJm1
z>KjY(x)%{o*N%p)3;q)EW~1RfON>bi0<uFO-5iQ~jDK3|b*l{{#Wt;fpo0vXcS&we
zOlQ75R=R3VoUrOKR+8V(%+X15So6D!OPguhIU-=ndk(alCI0}PZ~p*%*8wI;pjDC-
zRRB^GiU0x46achP16q!}Qe#3I5jQX&og#;o`SN*Q!ldu@sb@qHW)N0CtwlM>0)wKi
zl*ozn=C*BQCrNb2%)E*vvBd?cPEAHvj7G<grC;#grn+{ZlfOU+q>!7rPg=Vdnw;|>
z+OF9Gv*h_tTCzvNz=4zArM&V}X+xS;0u_~~Dua)uPp&aKwb<;kvX8*xj-`E;CYLE$
zxi|u~mU!d`c?Z)SYbuLV(92|_JxJUG@{Im9n{^h}=bCiU)tLLxx+=74-9V;&mv<;@
zR(Q1I&n{F(R!I8ho#ru64MDY>bYx|lJXDc8D+;TbMLuhqw)2CH%~quH9z5ki_U5pa
z?tKM0LQwaLaz}3Cnu1L!<z4pftUtS)8oq8_S;agrx4MSqHn+Heu;{|B{{XFw2H}T#
zruv-pDasB|ny|1!W76)(`_fGue}~LZ>K357{n{_DN*3YE%T50PT}<g04NR*%*()%s
zHfxzhP$PcE5iYl<rZblx&aHU59h#=9*yHURMP^eu5XB#YK?k*H!*?Ch2(Je{1NWxT
z8<rMeLli)Qqmh&Mk7~J~G`3oSX}2((?Ql9O_BAVWRF1|HV|*T+jci#rn<}nrWuX!`
zp>3E8S{B2387CEHcDbPT=u~n+#bxRi=gVZ1lweexQ5Ss&HECS%4;Dpprg)BM70Bdm
zKBBH#_>WPsRE|jlpy&v%I<_7%(Vg(bN}EQsJ}I`i0dMX0=e{Xs(sWH#cS6XFe+uE{
zod<MvDO1ENGaCEA(<+y=lsN1vjlPw6V<s*lBOUQwv~fzN`<!*K(x<U%>-|k<>1X{j
z@vexRr=aOh$r|lqIP7Zt*VCZlX$SlYm9A80DD28{YLb}_Ol{A2)c*iVzAr;3&kRlY
z5&VrNhUz!*B+PAE;@HO9?VJ7qE<czx=bESXY?nkIwj+B5fAFe}!tiTr9*6L*)#AIs
zxs<@DSrTZru>6UyRklE|hTg4~#sz4ei$a_uX4Bak5RL^|9V=6mT82VJ0%bdv&2d`v
z3wwDcA`(d-1Ds@gA78CuRb5!xkHvE;6BXM1oX*i)FZ#BVh6m7|^<AzC=EGKxBR?#F
za=!lnO6I>)uH6RP$2OZYMD6BU%f{iz`qVd09yx8@0#6oM3iM%B2W=TL*Hg*n0j?xH
z31dtuM?!0);zVu^X|2UjSku%`m7-};bMuLA4?S|*NB)6SVq?H2k}iAvm*hP2Ol8dt
zl5F@o7ky!ND-uQq2U2@is9Her*v`@oxm13AH)HwNC4F9pP98}uO^cZPy|9swAwo7g
zG35TI^r})HrE)W_PVYk-UU43uGmbE(yn8uvYpV*hY~sZ3jaiQW01aA_*)d&8xbzID
zN9qk_H%jYU7|(U+grb51iU32>oQD(*6CtSq#Y;5?ITYFi2c<Ho0*VL_&<UymLt>sD
zfFI_XmKn-6M_K^VTc~cL@~%fe!rj57(zQFS6Hr+okc@X77*GeIi^z3-Kkc^41b}?%
zI~u@m5<vs9F?4k}Af<E4F8Uimq_sGDD|;A)6u?a4nrHy}Q9uTHr8EGPnWV;mnz0HM
zk0+_AX0(;q^jx(#p=kmE)YRPTj4x{%Qd%;`xa=y2luU<pOl8b<VyHMJ5><o_02P1B
zD{9#EqH?jFJTh@_v4$mxxmkWvJJg$F&Q-yxJmNV9JBo>q_o~qhZtr?N_ftV8w}s-4
zePZ4z7zJlh&(^8ycL>^q<vAtEC+cZ+G;Yu;TD!9nD)d~Bezc{rz7_k(u(B}kOEL2o
zpGt&!RbMqWieX7TD#fY|*v9W})uVbIeiEPW6E4YUjybH$yHgR$pvSFY89ScdI|R8s
zq!9u{Pf?1RC}J!(lZw@zRVq=v<Rs}9aR?{zs*Lh3Mlnez)K%Jx_e9+x?^dIhXq0U<
z<hn9&^=d`)L`T%*!AS;cFSdf+F-bzjPai42c+;5&_D!b8$&yJWIUhhjk*w(mX^=lb
zO(3Y!UbZXW01D!&Ri1rsjfGv)#$^=x8LiH{!SN->j5MgE`=(K!&aC+R$O$yCvEMb^
zo5AQXa4}P$oS!|q9(??{p+Vd}w8nI?&HK2HNb<()de!Tzmy=`>;7usW1NauRH;d5e
zv>UX#wJ@n>Yj78F{vlh|PE?E>9OPD!chn?~nl}p=-RWAodu2m6PSvf<k~E1W1yi0t
zG9T{I5!71!$Ax@T2`;#^L!G2;4eN^J1;pwAd`P3SRuZ(G&7`fPEk&rnep#X^^yE-A
z5d`13`Ox*nVJdve%p)#Wb8k-Z8%K;v$?V*U+8XJ&k87$(Jw<XV4l=pd^_#}b#MGQJ
zu~Cm|$<mMdXX9P**z?}!py|KrSLsgF8)_0MUr;e1nEA)+TW~()6xf-asUu3ISM&6&
zjWa`OZcN)yDC;!50raJJvM!u&t2^HWTR{}pO$&(%t_$ZmuI9ya3KQyorEe9Y9QS|M
zV^SO%uy%#53>qVM0Ay4ia6B~vu~O*Z=DhaplIv2H?Ie3U@ZWLH75<gaSl)`&vKf>6
zLQ@U;eVb8H!{Y)^uf0aQt|4nX<liJ<n4}#@?g6Y4zs%^l5EB0YW|xJ-m^lX>SGXde
zk;UrU-9(@|{_aH{p0w9uUd^6ae1UK03&0L>`P4(Lbghm=CsR#j^`?Nm;w9O0Z2$uh
z+Qhsb<wGCUjb^FE3u347vSmC{k(0$HN$Fkzu!Sxygv2H=+lj}`)MwkZcMwXFY7-b?
zfvpw~IH>uE8}pB<o;`uDDzvIPud%%-r&V8LHhZIY9P0}{`DG025YNmw1i9k}lTqAD
z=9{TjJd%LH1V1qzyNKhUB>iZN`IV)(b2vN4V++Oc%@&(?8n2kqs;lf7+RQQ+>t0ce
z6P~r*hL0*{b!MjSjcZ)V;k)(Y_jeG<@5psO(z0uG_Z0SCpQwhKX_<vBAP%$&21<se
z9Ez4`grqb8(rGGyD9r>2dQ)>$06C{;)_@cm0iXw~VDVQ238UKGdl~@H@Yb;=m3%j3
z!p9E3*0y|Urbh5bsoDL<``~{%&RRQ^jz<QJR)mHP7&HKc8cZ4hOj7M8fD>a<6(Xbs
z0QaP2Gz}A4#y4)0pzT+@vnL~g?^4d(JZ~KZjxZbpS7DANmv%Y#tmgDRY!xWWS#&H-
za;TWB_L@AC+|s8T6^^YAZeGM+YCzqp$-Y&D0&|`^RZcfJ=;BgM7}xQF-ePhOrBUZK
zq)#TQw5~2&)Y6Ris<9UkRsA;2?%B3>>qUf1r2`hRA3y51zJJ|Pf2CxxEPZoH?o-w)
z`s`hScE|5i!m^g&8kuZ#;nQ+S76LFheEwBsLhR%p#<Fbosl%j^0SXE3DMRfe0~LJ?
z;}scbXvH8XJ!-xCuady=#blPc9>sc*_AdR51(m_TH7sG9maXi3ojJKX1kL4)ih#--
z^Gyo0?CePsuHjlqwUl(FBclao+9}`LjP4k!D-Pp<O&r*YU%Y7=8mW6_vWJ{UZol~P
zFaH3uRTN_*4)m79$;+FOmM#-?JdQ<Y(?sH9%On+hPT{jv(C3clQQ)Tk0E)KF$MAf^
z@Td6M2(pQfdx?LYM^c*B=O7RsX;lDWO|cr+5lbbSNeSA@IrptSH&5~{g3kM0#_3zB
z$E8EDv=+B54Qe|X;_^bpY^{PjxbIz^q=DKejk2VI54AdO<xblj6w(sWpz3fdUI~J@
zLEC{_85=MUGCocQc_Y0t@5xmw-yQj;hdt5Zz9xiTy~>~YV<Thpu5xV}_Bg|&ODYkc
zl=EDxOPkhy@tYzlHreBN&$yyHcw3Zw8XIP0t7?Su)q2i9DTWy0rE0<KPki6IJoc=;
z9OnD+tr3~%`f&Zq{VF|F;xqK9Kyxrp`|>^Op|q3a6+llj#~gdzatA@`YcI>5^!e9Y
z)&cipLCbpq{c2^Yr8BeeOj0e4)4WRzc?ycBIQsk7X=bO$PTgywJ0VMXe_f48Bj94D
zXt?WE%!=~HM|!0;0r#MmQq*Ju-SXjnTxUHik^oHKWWIEf$1%KvYmf0B;MNs)vAq?g
z#JW=~W*IZ~$T<l8U44&pS+_R<6))|+aHnkX9;)88l(f{<6>cXMx2bh)lWB3)Pi*xS
zR_Gl%9YK)hJ-g7-io3U9llM<El3y}LFV&87OxQ;wkzH0e!?ix%BR#}XFIRA*@i?HQ
zuIyOx1cGg9`O^gq%oiOtgMs=A%rq^BJ!IM(gw@MEf@sXYb0ajFDCw;Aok6aoj>Ifc
z@Bsi4aB?fE((kos^gTk{ByV#f{I>`1DwZ3U9{J62&dJ#9qLXe`=>GsB6MK2FjrEZz
zkO-oVFSr<VW*?ny>37k|vIlm!ZzNALMB*r82ZjtsKXd{4)-@oeuD3aAsYW*FLE@=m
zwX@6&jTUfT6xk6V6(keOwhvyF<GQp43(_|2+yi6S*0@PKS((9mw<gkldmUaNxhB{+
z_S+&7e=OEsO?D$m?1q{#Ow2J$KnlHSJJ2#zsaJtg%>|~l0~tw82q_IR13Xh|fK{V3
z0A$mLy#P6CSlbi;vJ!AdTJ5|WdY9HS>6c$NSLMw`MWx1$OBCGIp$@>LJPH6iQ*l5H
zVNxXkI|7qvq`+OGl_T8P9co*#aEp|ZMvPFTDhU}i4a~gYa4R(K&u(y3DRQC|ljg~#
zn&%w}K9x>4GN)gXIu)GHCp>}fY7-o-)z9ZtwmjOBl%Alm9)#7UR`S9(PSi5!q06;?
zM;|K+qjLd;`Q=X_`qGOE(d<{vShbN5FqLu$tqxKa#!lXqHNDK1`SO^qdg6d5(%6|V
zCF)Y%L-~!r`!!aA-PndIarCIZrm=&Sl18j?DNuL>eswG*Lv-XCi9HWug*d^cXfX4X
zVK6;ut7yj~pgrrIyB_qQrK&S75XB+pqFFPHoZ`A8s=49Rnx{6IE6f2KO-C^=9#0hF
zmBopz%?A`x+A?b8suo-UPEN<4gO4>!63J=34n=0kWWzg6N}1J+!L?>JEoJNtNJO#r
zVMAv%I#nsFq(t`;5r(1}esBQiIIBi7sNAn3b6Qw#FK!dOL`FV}Hva%hrKH+G(oWOP
zu>@SMMt2`dg|&|A(TB8+L`m{8wlH!rSZpxFd((<A%sF2xh}ykka2mAedGR`b1mtV1
z<M?4Fm&Z?wTO{>of6SVPxldz|--=a}<`nEn)M?8s@W<swT&c)VI-b?0+FkGN<D3-p
zPUn8P_NZ#+@p>0;d^KrzW{O>-5#@^jGTc)5M^Q=be$Br!TOGXhWzPnx8Y>D~yPe&u
z`PY!J>5SHvo_y<)dv2{!EZvfSDjku4WXSv~q@YJ{BQYh2>S>~%L(2R~3AetukPJsE
zoS%Bhx1HqE<B=N$cqX}ems3_`>Y9}B+!dM>o+Hz8)ieE@ax&4&YW}OzkrrY7n?K&N
zxQ^j{GlNswiP5&)M9OLCCi)M*Z{1Z58~#7P8l|x(q-p2(L+Mp@w!{sON`<&1infuB
z(fZe2r{u;7t6<A>7~_x3o9F)O)_Y4NsBUjvlGt#ODZzhX?NH5Wo!5g|<k1Vro?iJr
zRRb6m+FK|I81!H3T@l#{^)%reRHX4#%vn})aaor`tq4aut;+I4<UnFGH$;=84@&17
zP>844^=mSLe(SD7C(|{{R!u#2I%vn;^B(cK#)~wbeVI9i>3;M_-j+*8Nxsm^rM~6D
z54$3t;uMl>rphmBAcFGP{LrcM7t~fCizR5Z$<<C(;y_JYV`3)mndeyIg(Xf!ZgP`t
z-n*8scyWK?cWUjOu(6oq(E)GiQr?D?pT3OOV$4~&B;?d=Dv1;6Qm@_SrGeG>HZR;u
zA?GiF?OuhW$|1LvJh<P^U8R_EyyuGGto`P9Vc#lDvQTY2R!*g4jT3n!#LO_;4mS@&
z?O3Te>QthntYB-G?Qpo2gm)$OvJbnnfE9Dw117kwa_C-algu%U?ErcT?4qvjjuyFn
z(cT_V*V4wsfi1krf#3ISf%$=1EGwg^+V@w_^cZ-enUe}A8XX^6PvSHUk|jzyRI<k&
zl-3+jGCE3VK}g*xkT!~-6b#TP0dY$}2n{`C#Q;jJ8T-c-Ye&2`ckxVzfO+~-mev~F
zk2GeOgEVG<4AGh_184^nnEDP4nlfn!HyY5>RY)wy9QUOyj;tbH)m*zPpi&KG+?=13
zX0oRf)W%ecjh5m?1cZX$DEF(2Zwih^=Bs)O35HE`Oi7h7)Agt@pWQ#5S>|Fb?Hz|K
zal3RQ=~IZ~4VEPF#(1fb*Mg@9tctgaqwZ7*BD7^C<$HTlTN&a~Q)z5hn%8GBoy+Y|
z1@#r8I2zQMWBGGYvKBumrWQIc3{O3Uj;-Y^BuxrL!brpQaC?tqPa1=u#FlohzxB^=
z{(?nf*0()uB$|R&0exuSe@|8a0IqT{{&}Vwoj5Qi8>IIVD}I<1tj(ucK`B_d1?HSr
zA85E=P0)k<nyd%6ll@iv1bQ;MpXFIfn_HfZ48oOF%d;}?tl8wucNVYz0BzNO^ej~|
zq~5dTVJ<)En+E<yw~I+w@o_22t82L=`gPPmZ?>5~-i(T=BD(NF6eej}5>Xi(IXNP=
z?BZMiahjTsOCs{RR;zgy{GfO>MtfFLae^y2v)jf{=Zd#N6>l%SN#{o(#s+Ig9;cT|
zydwRZD$N{Z{n46cnHou}TSioEGOm7^=BaFP;-eKcWH8KTo=0x3#fkcwWx)#{7|jH^
zb5ppG1~Io>R%3nR^{PiEB|cc(L00YCXsbD5bv_UM<;CC_<~2;`)rB|5%rj<yV`k;Y
z(Bh%)TiD}|TAJ1uxVVlel(Vy73S=Imq}wHoF<M8=#C+^OTGP0cjgg}821zxVW{Hq$
z&TaHr<jxMlYs@rT=F{~E?t@?k!w>Hc;;JC`J*PyuNqm{3Y{ugt`w?6EIolgJ<+pKK
zBAK@gdn0x@E=EZ_)>fef#JT<83-zF`iJV7?;gjsrDJsVY9M_y$YI4}S#Uk#HJgMzi
zH>gW%lIj|S6Dk<6FgXp@szrHh-z=jag+q(o6&9_i{np`2K`zG~IqO!eFC?4gcFFwe
zqkg1L!F+#qYd=Rn>4WQ5W_m`Re{w#RQ(QTPzLlEXYjcs)*LA_qt#%eDJh55{?oDwT
z5#~;D6880~OLX^BGk~`C(GQuik_TWt>LsDAHPrM^48NJATRsR@$pD<!Rcuv4?jty^
zg|a2=%d0AXcvQRBZgW+skyd$^wPxKN*`=_KD_d*yxM{6RHM;SX=Lg=Z#j8hn1RAs{
zAUp{G!!2=6ZZfpCXLMqmeci{~#+DZ*6_ki28QA>@?8@CgI<+jD5v`^GNUOme9y8p1
zD@Zl561&vHxVUq4nH)@}2mb(gcdk3d^O)@IZSr?G%V*f}RUOP{vC2{v$7<NXSes+>
z(ALJA7nL5PasmC_s{w<KT!j8iYGtu>{wtW6qa1NShT@!hPzQ10MA@S$?#xYgdT<+5
zdsm%Tc6QR)8V~ntHt6*jC+I7kqhrtfT`{z~B*u3-;R_C)rnwMuYt+N9aWjIW_nD*N
z>3qF^PbGi`WmBI*aDPKt(`Gqc4<%Ts{3~vxx9%@r@**cRQMs5?fW;x9$SJS)=o=oD
zD*9Bi#iTS0V^e9OTMA5OnIVd8Pz4kKP)z}FqJbg64)rMZp^UO=C;+3a1F%w1Af(4M
znEDGv77$hI5XK{pfYh_65Ao<eSuk<dn-DJ8#}p)MtmKh#VP+uIh8zM~vT{C(od*b6
zl_doCs5X(B)gL*CkG{!Z&r?s4vecV9aEYiDrWsR;s?oZA<v&^x&54V>#9iCC&MNhc
z5zgBsR_CD<H=%_%Uq)EA(}P?XNh6$dNvhfnK2?tcBxf`!Hta;;qp<5qHtAPK`_jh0
zqneGR%7SP+@Q>ZF-lnTf89a|#3&g?7IuV3D22awbw*_CP=SojQpAAj>M9oc+N}xy-
z{uLCOY(IJ;LG-Lt+)tscfu&hruW@apk_Nz@Sp7CJ6{8*1)REv_NW-u#xPG-#eMN_-
z87CH3YpaAmJXY>K;7{<ROX$OX^2XSoQMrGUVwJD4gS{DDt%*OiY;p72xIgsHY=1Cm
z8sfr1*6KX^?SJs0L8En#jb+M;Q`_0Qe3!1d{{Wtlk^G6G9W5EMx`nUM?qBA$r!z}q
z$c+wq*G!ePc?kI=f>Zv?>Hak?m!jUJ*H;48MoVZUk(>;L`A7#pQ&6Ye-n}1rHK~tk
z!_51jjCxgxp;u431vd3Agd;jOXsRC>P(HPa{{VdbYgnE>GEJzi!c)-Vtl>rbDP3)5
z2*r0Y=LWBMtnD^1?j*+m&(!l$_cePM1Fd#m2DOAmJ@kYbBl%EcryXfaVpiv6r9f5_
zEB^qknDz7(G`};ypdfw~2^M+(0E&D;aXrlTkiYsUJ8`$IbFoJy%!r7-+zO?lxi0c{
zJx9SFbUJCbnS&VHDt!U2#=`f@*rTx~oMfb~jbyF4vkszLc>e%on$FcVFDlubo|H-2
z?6j|BauZGGv1uD|KDCv7rdizrUL_2Fh;^<7Hl5D5O*LbmyYR<3w~*t|)^*m6G%3n4
zKD|Y68n@i%l_wohn$S3JTGP`3MjRU2_c`9hyYu_Hp`;KmIrX5E(ClX(ND4E>L#nIG
z0|gx7r7cFcFf_Xhk2sSV9QUrW+R||t`8cf{LQPz)e-zR>$BZwpVO8;MG|6cRX=9I)
z2g<qb%xfmK)Yg{ArRm72XK5TGY6KX?bT;q?19zbQRnoMvo7tCE&MbF7jWOYpKlNuT
z+*Mr6j5uz?HJ^O!kV?p;uu?!CwKCWeyGH=J#-ndDeC)jPthHsQi&KwqwS?ROmQT7{
z{nPy`&*yUYWOur0d(k1bYb(ea&f%UT)-wF;Z?C082cEP2g6!gGj_4fs_V)x;MXfAM
zx~5O4TOxT5FgHl7J&rlAD)D9Lyw%)C_agZR=Us7DXC&T+Qm$%Tam`VY3!v>=x_Hlq
z?%04@Nu!8*;09y=0J7A`xNq`jKO*gsm(bFv=79+7j+DZ9rhx7|D%-Sp*WA}v7Nc!x
z5cB0k!0vHgbwu2qj@UGkxk~fJ)+zGAzIW^qb5%yCu3n$?Ej-5==+f~^*rlwPomTXM
zF+5YH>en`6ThkMHUp5f(D~iEd?8C|~Hi^$wNlNk(34FVV<nD|c@-=V7QF&VSqJRY-
zXmSAbEJ*z;ZlryO>QJJY0ZybeKNQv+(k{c4)T`-I$_fntF}l(kp$SD&MA#{&fe211
zGeBGkq@Y8oLB#_dN~h+<09&Gf3Md+3^`xPIQf8zDqNxZf^^qb5`qarCIDd?8GUqjV
z*#{V@md9=eH>`_`jP|C?Kted?uv?!?Cb?Njy{R_w$gS>t-AMCElS39)zA;g}IXoZE
zvUAk+DbQSrBCOvd0q;$f*!-h7tr6!}R&5ZjLvvRw?pZD)>C|?i>~qtNw>7Q~btGjN
zB#O=P45jb>ZmKa#JD62l4MIkYm>+)Ozwy-{{8dR58`qj7?0Ok}K6zR8B@k^rs=dku
zJ@Zo;U7oFKQB>W?Rl@bDA!RaEOJmQZ*D^E=L6jc!DP#m=JXQ?%xoobA7Y7f|r7}B{
z8x5wmO6c(=U)fIO-?wg$EPo2Ab2jmj`cjN+&9POLNHVI#&+IEs-ran#Uz)kd=gRQ+
zlX4=wP#Ae@+MobJb0;|&2Q;@u;vk_}%c3Eg_$VO<<wlcrB1NgPyDn{rI5@~IqZBAu
z^RXDOX*J8S4ZPqEl{-t7KQ<|~skIv4yRiaHsI9rW)>9m@sz;rPnv#kNa-3D1rK^9L
zqe1*mGgo|2+0@AX>L~pwbu{Q?*IVFXWLt<&LR8Y0!j9*nB7{NcqOtBJd2Qubd6HQQ
zE;|nOn=~=6t#s=}^29ORT_3*f$XbrV#deuKSYThg_{ZQY3AZO>rkqk&H#EEW;%_eA
zJVo7)-fHM9NNGtQcc`r?PWls4Hdi$5+c!$Ss>Ad(&-lXQ?D|=3<K`8l_GHJW_1NU#
zy%H$HgP(ev?HD{9*9EM0C!uN!jyb2xA~5Hr0B6mo+gt)!6nl;hW<8j<d?Vy`&2LIg
zyEC4(HEW`+_xifcu~Ke9Us~un4@1z<m_~Zm!L#!hngE{O%>$_7wG1Is+Z9cj%X14^
z0d3p?%W;9;up+gUHrm^vGDtz0`w#YOIc{{)dLFM~9?=SabpZV<qJ(o>t<gOn_1v_)
zR8tls&{dh4R|;Zs%4<7bziWLd<%=7#0LZAQ?H0lc@<`?3*S^#Hy+RnDJP9@okMXGP
z*jh-k8)%9iQdo&x*9Si`x!=jk?1$V(YpkyOe8YO9C0OwP0R4KduJT)KxmFhzr*fYA
zeZ6s48(!;XM6d9%HR(&)q=>^{9$;^5XZhESF)~c<+?FJhT~XIlj(gaZI#a^tsLX4i
z#ckS+ngyhN<dR8<$8VKK{(@;fg~Mm*`lFw|7`%{s(kClONezDHpqS!-X3vLO_VJd=
z)=w!jlH;v*Hl7r=Mh$x;mk<8aaa`3W$n1`qaOPT_R-dNYTObzjMxR9$az)2MTurvl
zq>@?`Zy;HtD&Sy@5nK+x;WoodVb60J{*|p7zRl~gm0Eo0;;t>Ff<lrbD-NKHR)>g`
zHl!pO#IQpqPjX{i{<*J8>9o<tm%7)d>Sbb+2AP<kdeetAn9%c0YDJ2*B}*t;37`fV
zLsTIsswIsyln6sfgFu=T3S3YY3Q_Ms!#Mg=kmaZX<rDx>KnQ6?0}_&p2q|a+=9^<B
z^xIb<2i|W=)w;Ki*ci}ZRAaAN#&h?SQQwA+j!exh*A+Z&*d40g#@I({mqYg-T1i}m
z$*HnC>sFLup#scDQN>tERfaukta+5{UFu!7$Ckw7(xQ+J#}&-?K8`;MGcP!npkwoz
z+B+ULA^Yr6^BQpZBvbE0CNa|ILnhX}w_)fj3tMdJQhQMi&N{myX$X%>vIqE{NB%?l
zRItoXN>e>7zE+)yx96=_x~${5>sM*hWh?>I;+14j!x`eXwugu96j^#3y`W?d#M7Sf
zeaQCWvgAF8;}eFaeZ|J&b||Q0j!9f(^IIikcvI%BR?#%=?D))$f!?IJg#>@PTFJ`n
z_Wr7&N%L-1JZC*>#5V*c4;5`(<-+0>IoOKoX4~?Q&Z+#)K7M|+aatU>+R~L<bSq65
zE&||jQ(40CSn6y>N0Ois{{USR{3vIf;_lk>D#JW#K5h@vq-o-1$9frXuN4~zv8%7l
zR~@}YQkpXygnqQlYeLCM45g|$jSVd~Us{ZR7^T;neyu;T#V5I|_(~Pt{5Y~L9M>(t
zEwpFardtyo&erNOwvrHe#%q<czJ@Dl3$qtyV7q#Op0$*njiRo?-E5XKaXfO7pOiCw
z<4vCS3+<~cO#-hAu;dTIvT}-6Iud5ikE5p4l3S>wg}DUqO>f)Akd|1+6m%r#cfD&$
z9IoWlZKBPQ0`@L{hziB=bPGP5%H!qUGgm#C5a@GjrP*x4ji()sDy_zrQ<!EyLtG8_
zI<45M%Y6q;u04%6?b46+v9PMLCA$hW6${)|C%9>+<L=h9<7aa!v(mwMKX+Ahq}_Br
zwa{{&k))>BmbF?!5`hrKF${2O*V-Jf-ZaF~nSCY>^UDr^)?J*E3)p0ti_0s8iF+vS
zYGt{#J&#kd7?5}9D|#XG99HXbeU?sMHT%peCYB@oVyJ?=s}zbxL9S!OS1?!z;+`d)
zWH~FI)ty$O8?n<)QnD~^#kr5qw6lsxJN4Y3QJ>bCX{x~-Z3Om;&_7TZ@$NHRJjyR^
z4&@gmxXX(;;F3A)X8~L0kAEMAX>O&G(nzm7yxY0Nh<|~*<Ji^3a_mEQ+{p1JVeswx
zz;hS@<3H>%>0VzN=bGt@(BiA{6RP@Ds+DGAE)#0j+o4smgP+~oCPDRVzy8fhmmh_g
z+P0rOx`duYY{QO4GCdafuL`5sn#NUQ9nPv4Y1Vd#@uJ<Iy|7hly(U<j=5?v|=<JqJ
zRx#$2H9R!J#n`vr+zxBK7ppXALFTcj`JK#tpR@N_ucy!QfNJv)!;0r>%%t=yO1#xc
zBL<Mnty4#~zg$fi&t?OWT(^t7IddF#GTY{5TX2U1k+ql)(-qS~uP?gxFsVxMZ5&gn
zRw020(9&0W?r{l9J?H`1q$i3@X^tH!b5hG4LFq_CD5?Q7Nt&Vrq@V(d04SgWgFrDd
zY0>eLXc8|TXrQEmiU(mj(MsSHjZ92I)~{)nGG9&+0SZ3zdQ{3>s^g00b~{U3M74xR
z65&_9b9%hRrf?5(E0UGvE1m7~M##@zqOGf-L6S4xx|m^Pxtn5n(J1AGMKsa3-D;J4
zofwLdak3#&4?-$f$IL5Ehn)rQ7XWRY>NjUr=cQrP_f<xbpLF?<eQ7rXR<=CJ)#jUG
z#yQ9$q%oE}a%i>As(QrQvPPQn70)4gtW9epQj`InaY{Nn998!hP~0u_<mZSoe-~r^
z6t?fRc{E8~9T-{i)Vm&M^r$W*bKew3J<g0qCCWC5Q!k9<)UgMCfYyhZ6(;PB2<+u#
z2Lyd-{{UrFLS$i|PPLPg=zDbVua@#EvXa~Zk9w95k@@SN%Bk#f(6Xy?(Uf*>DIRcQ
zDl?kRMcnl(S6r<b%+0ugz^QGrGm-h$);@Za-?dGXN%u~1QA2KBPUgmaDK}%bbbYl|
z*n#Yw8~1yg@U1HdWNj}005P{a7pMDK{*;`I;e?!OxTSPu&0?*czbjQmwPceG)7qMn
zFt3E=l9WqnW?6YWRV(QBFb0~Cu{dW=pD|XHRi0tDr8Hvd%wv^eN&G81`M2Fl{{Y@%
zX)|eTL0*yIi5?9uJRAlks$Rs!YF*bgR!d2MYz?P1$>=HerY$B!OY<@G6@=Bz>6fmi
zGt0o*tHQ6Kt8qfMkcqAaS-D94eb2RHySdP`B(}MErrjeFf0b@Rw+V0Nf>tcx5^<lU
z5Wh9h=WbHwpLBlEoAZ!C6)%dAwT<9k;UtgFyB^0n)9*M)ZwOFWnvt)^_l0m+>KHBq
zHbYgI=1=3n`ic+?nmcJh;X$nNYKtM-LG<Rfrp|h(hfqvMMu`{ifya8>=RF3%0UG@(
zd7lbtl2e9UlUEW>DT$S?R7iRIz3M$3W1m}&O_@|UR&V4e>S-O1NwUIV_a?LijO<Ay
zE_zn0RzI>xW8MD%EQ$J>tuzYXdM96TRzX5rIn}cShi!9u!a;F6$tz1dko>NQK-u_K
z71HI@){AOaMGHMf+SX9;K_{UNCUQLw6=LE&I`$U0w~k1id65w31Jn;n<BNosEsmLU
zCwq}tM-0l=hT$F=dItGQw_{L3(pp?err%`2A=-1u9)wjcjS}9rWqbQlbPYz<Pnn`t
z3f`;N`HJ%RVJEe9M`Mnv*5#<}$)~h?7o}N|pK!)W0xwc<E2g<XHCvDe%M+N#IpGG;
z{HkPWABFm_)Y{XLMY^*AxrS<Kp^8p}4eBe3YV7Z&Hw!qFq_l8yrCffsld4-|C6uYh
zde2*))N!h{K3Q&J50ssxR;HP#v|qcMK7iHPw0T&Tu633gj9XVSDEgdMv^t%rA2q~Z
z_t>oE7GVh?@@iIC=H?&!=v8LbEC>9(MgIVOQc-=2a*uKc)^zdK>LK{lms-Y_+GVtj
zbr#<-;|D#))BS1k$@B#!W^-Db_;&OWG_e-`@`4Ru+;~{Z;k4OIkhk~Lf-A0*lF~;t
zNX6_i{3Cj!JbqP~f1_R6b0U1cwbGr;9Br{f?C<SP7VVy*lR`thuX+qe)KoIA8$Ibn
zdGA00kKT_cdYWhh$`qbZ_ogcWg?Z~olDO|q;;^C=^r!uuJt(+^SX0uM$d9n1<6^84
i`cu4!{{Xu~5Um_RKh=zymdf1;EXp>h$>O5mfB)IYY_VYg

diff --git a/demo/sample-video.jpg b/demo/sample-video.jpg
deleted file mode 100644
index 83330f43beb16e3177afe4d6fc6789668cb94799..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 84394
zcmc$_2UrtbyDmIQ=p6|Vnp7bHX%f0f1nC{5h^RCnbP$jx5<oyAy%PmO2~9y!q=<-!
z0YMQ6h*G5p2nb4-e#ZCxzWwcgU;lOOz0d#ebB1e%$(l7YS(#^^a^Lql{&GA8uo>za
z=mHQ31aJiZfa7TZs}t(s1^^}|fFu9_bO05E7oZ0BAmDlq!uN0cx)3A)`>P!a9+U^b
zEx-=0rr`Q#aRk>t=lu2G-^s%_NXGYikPuqtjEuYjc&9Ny8=#@0rlF>yp`oUsrKO=`
zU}IpQr)Pk(urjf6!x0E>I2RYMz$rmqK2d%yE}_#xqT&)LNfeKuw45|j?i3P*{Bscq
zEiEkr9Rmjg0|%0qix>IdUXEV_>~zpUDgXu%2B7Q^7(3+nEg%Sfdus4^`Sa8KV}U?n
z;P<7aqi0|QJ3MCtpb!`gN(G~)rlJCSM}o%zDt77<!t&>7I4qoKMFQYwqSK4%M78SL
zIV}f%i77bWh@ofT;zscBo)VWpq9oCZO3Es#YT6fcboKNN46QC&+t}J&vUhR4?&gm5
z@C*zJz8Ml479M*iE<WLIB0eKCD?2AQFP~6ST1G0bsI01f`mFwWLt|6(i#Hu_$(>#A
zx_c;tL!X8}e;FB_o|*kNH^1<GacOORV{>bJ=lAa3pK(FJasG4pdtm=zT<qYupj1>a
zD%wBef<QyS6~<0QEi6xS;=Bc|Qvip^nP@t=R(f$=JH4oa<u6X>8v_hnV(95pYk!9J
z*U0{B1B?0J8ri=M>|f&=1(;wE@ZrJO0Zrikn@Pm0H7SSM!tL*l4A(w7H-8LBf-PR5
zJ&^2;Kx}OO+}}F}4ooh4C{WW3$3{(CFMr+mYI+Q`Jw68B4z^9(c(rYe{qysuK;pKy
z*>7vY&cV)L+t1}=U=3aLt9t9_?$3&-ub+>B8SDaPQD5h`^<$Q2i(hxPj)C8VYrn0-
z|6Tv)en}JX=B*J&5&!I=`Tx8tWnO#R`#-iF18*LEZd-U&a12CUI|dRfKE@Nt$3Wbr
z$2W`q`SmDKfAwkjXCKmsJBv^L^r<}t`XV)tj@pg^4zP3F<-Z!#|8CGbLv509AUP2f
zRAly#aStgMF$cF8U0QV|yzXK^Cs8FofMgyh8e&cFtCabNhg8hum5_hxqF~(o0Y)iV
z-2!-jeoJX@1WmRMRlI`!`z9+W_g!THPugb^t3_`zc}UY^Q8F`bqGcv_7NTQIkk{&T
zs+^)aFXtaLzayhjd^S6l8hLnUJJ|_|8f6X1S}m<q&l~;b(}Q^b*NZ092{1FO(dc52
ziN(9>lvzBWs^Q_mGUK9w)d(<lN(rklhiq0@8K<8o6{pN*&9-`B_uh_1L}#qIXQ5}h
z)5jNI+~HpHi;j<Y9AHap>X{U`Nh+7?su4%g8Jp{Qx-i$rPU4uXrGCC_GsIjZD<g<R
zk^8Y2`V=Htj)SL^-z}5ol&mKeqLd`2o?x0)S|jhEW;vfntrkuHxrPd+Pm^cSa?!`P
zHQ7i<T)xIGsJ-$7lUl4+zfZEO=rQog;AymjNtSmRTKt2h&-Q$@ygc6IRTeXi2C5|z
zi@1*+qeRO=b9spA3`zaebRo=HEm+RCui?%(fmH85J@HM>X-QTb&e2`yI}9{3lc)gw
z5H_ZMn`j!K6uXF7wf@Z;MNur;5jpa+8`UH;+w=wxMuqr1?4wG*OQB9W4}{&kUU2zF
zbo#n}^?2}d(y8W`=Qu2GMLl*{YKV(^`e;Q)(9DJDsc(=$j>&@mR7?+`X*&PNqlkU$
zyvy~oqd~SdM;B;pPNbjBF!0{+FIL(zzPUVz*GcPLvtQ@)Q(UQcPZUpfEtHxWh{N7V
z5R{o0Re2LHD!}UJVAXtHfUXyih65HD>VIr({1Q#is#X{61?3(C7&Pph!=UTwm%}L#
z{sqpep13|Iqb*_xzx+=OS)PUe5J_~Y>#{f3Krz!@e!w7~X{=RAR*^A-;|3_7JsqAb
zuoL=ngzQ{leI!F*ooy4gUh&AVmD=J$0kt?*IE6$zjN`9A%IZ}07$H(Kn{i^)Niu5Y
z&_gVdsuw{FVOul8#QoF(kP`5@UO>h}N1n?^X09R4fHBKy@k;Hz8AR<iIrvWQ=ep_(
z3Pk?RMI%bSaH_pLh6X8uV@pBFe@Ef|HXCgn-OX}kLmCNBB;+L}a1lc-fCNe(8v;l&
zfFQ$xCKb|HG=_e%0FSr>U($l;O=G$&hSOeG7rq`eMuv*QfeufF_R7(b7RLdn@uW8$
z)dUaiaZ9#|`ujAhL^etm<xV6Lk=M(Ih1?9J<oXD4xWj$Qg~kk0?Ez!Ul`=@92C@qd
zu%~4C5UI#A<NfN1IuQR-_}kc%_3XXE4Q>ZqO(~-bwj1~Pa!v;nu3RnHZq+=&T$On4
z_xjP!=M*m<-=L0ZnwG&YbJ_H-dE^(O2cIh$6{yZ&zsPj*>C@>Ux-16q86PGXP5SGN
z!i4J#7~ak3c_oS!bOVqLRSBG&>P&-XeNEIt@iFkq1P~mxY1`;O2EK}?u1H<m{T}r*
z6GOY5{rF&0bEms>-|+KMxZwV`e#jxo>ln~rJzU2gQJx<Ig$YNyXdueN9&}gQ$G|PZ
zfAx}&exC4D>Ybjj(0*|L5@y$IhZ(r_p5*SX_;7R4{sFP6SLIhE7Fg*rwD@VPjqBne
zOT(Bf8=+~hZ(p2lFBk4P^VQxTL1Y#{N0m;@9)*3PQE{?tM64Dds~J=H_sr!Bv+l?s
z1+b8>AMq|HE251EK(M+>Icz9pbIB1oUx1;bF`;tCar2at06|0fMNAxA8m6_jXiAgE
zXr|jfHIuKTrosBw^0G7EqnB48t5Wq1g;IGo`gUguXY!c{nZ-nE5xgUO++2uk@royS
z&zxEbo`q+_Ya3H7>kly*JL9RSEN4@g0imly;B7$PzyY7`1!PT8DrG`lBqR}o8e*mv
zs6@0GnunDO57h&faH}DP9Bzc8_87eLxj-@(uc*=^HOC=o*dFg`7?U0)RdzIum7Nmi
zYA(;<oaqDQwMnbR0%PYLD3GQ-3V0Jg{(+P%fU^uV;LTF_YvTwDJU~}QWQm%_3Sx(%
zCQ^5whb!L&9~#T=n@f~2+pG?aeiw#QFEergoY}hy{wK>R&-<i5HhEj5b{4Wk-2Z*<
z>xpjXr|9YFbbxN`8dW88Y}b3j=l<scj8~tgEE^JdrqiB2rsk8{iMHv0o>5Dg9ZuJ|
zXMld+l22^BLebLqn|N}Sg$0qM#7avwwe_&bsR1+?DGcowpSA`3z&4-X?o2%p89#Ou
zS|j+sc`R^JIR+Ud)rNx3k9yA@jsm+~Mw%@(ir)h5A3O>r{RD%888k1p0Vxnc0or#e
zB9~|qAn8V<IPMC=o8eY4Gs9!RBIFowc&Penhu;!Wihent7x(>-_t{y|)j&cnhFeu4
z3UobsOaOKClH=K+3e-#<D@_3bP~R#eYHcm*)1K{z7G}9T2c)H80M=m;5?0}kAkw1d
z>+4%BXU($ECRwW$#nl%K1SZoTjxEV2-#OjdKU_c+hbWm6N6BM=ooY4QZtlq}I+B5L
zaZ?G7<7B-#l)Vql{<|OY5#_Je4~-bWFeVX<&7qW;v^&r&77ttrPIW$wT1K@TlHw7d
zPYunof?&&p;IB4;`uuX}c9{kic>^hSLelG@was)uIw6NzIYB`3Ocu~-=wuH#nR+~G
zDuV?WH4fuAd`{89;Ld6EczFz+O->$tLy|d^*W{%+ZvGV~|G1~7{xNXhw)A&w`?mc~
zVo<M@h0LN8QzE~CXlKZ?&i1ECXK3CiJ%}yxT*z6tc#eZtHa^^2?RKp*TSHByV!PMF
zJ07Ksr^`)VSt<3)<8H1<@VrxuYCp%(`s=XuK}h5F&OC=aQ@<)UdmBhsZ+S`_HK2O%
zEeK%H$vx1VZZGq+RYr-ai=A=Q;~I`!=l+nK*;MPV=E!k{6M&mTbpCDQ@HtiU8+Gt8
zaCIZ<I~uZaHRIfQ`w1_r^n3W5V9b8NVaTG^FLX8B@(A4-XT4CPxrib6%Gy)OC{S=3
zCPIqnfqPONMZ4~oYIYVA>t0jvGzr0FMC$E|lq@J$mboQV4oNij->!gR7zzF=<?$*f
zLMwt6jg9@C$Y;%k%4h!=0l^#z(cdbGj+fT~_$dpcIDS_6ugGv~x|~V(7|fcy4XrBY
zm+!YvT~Z?7y4i@?Jn@l)I5Vd&v?)@cA5C;lvTAzUrD$Z%Un7T#^e8fPM3j+XCt~%1
zJqx^}pdk|Bf|u9FsFbz>h?2ZGcoPvCy`oP|>O|pOhI8R}qyZ~P!1*%hEIbRScgCH-
z)2K$+w7!iXT8%Rpvuj<b{*l%HF5e@mK9BwBo)r%kavv*-$0+-GY9S%==lS~~cr7GA
z7MI7w!kwezQSvL8l*`k$2e<w*fNjsMNarx;rt|zsrEdPL&9z~^!5z~ok%5<niihU)
zAG=@R^N0OX;q>cJt2um|nYcQihD3~yblL0NKL)&;z}LQScVPQ{>h6zYpa2m3A8(<A
zep~3+;W-A1E=6rJ0J|*)2>Va=L$eVn{J~`!YUc9TPApXdY;m}dnVPPYr@{*25n(QL
za%d_N3$ry7T$64+2K;R-zV#c8M<UU18EhHs8Ns2S&ccvx(YSbu!NE$4%_1cXjSDgK
zpC>?G4_3wUtCH2xbPf*&Tg)LTn~=$9;@JTV1Cj_^)I)a)ek?qL6|-osh{Z!{m?t={
z5!z!6aufQuI4A5T+arJsS|+;1aq|y9F2Od+PHy5@IrYp1Z&pFHGC48OjOcM+l~R67
zpcIN@vWzySg680u0?D%B0CEu%TV&{94oM<71d)XqsKqtDL$Kx0nIua{00#cr!F3IM
zvd~#{9?{ul7^i_{dpO)U<R-)#VA<)ef8Q>4lo;Gzj!R59d(pc-{=%D5Co-S_r=Wm5
zFo=kcn$*8Vl=3X=!+6R>hjHhVEbx%KIwyx9J-4Kefz)KZ!!Lha!;@=vk=52w3U1tp
zF#_lW9L{kR4Bg`1e&1EJvajqiw;q_p6ct`Uxg~K}ND*w<z5iNpsV?a@`X0Kiu5GdH
zNc9-_T)NM91X{{$(P22C`G2Xk`lK3P5MTPDkK8fPg`~MPLV6(W`MAbAMY*ZIdA6|L
z&be}G&$7DcO<I@Zd_fWeAa7($jUerrL-7_X`aF*-Sbt|v7CBwIMyAB*>`?}rwKhpa
zXzmgc63!4Zjp;AX6)JjDgb|=b_X;`ps|`Y4Q{q>QX#+<KS!;l<)NqSlC^x#EnbsZ&
z2q5hpY4MP4A~l?4gs2@&RtMb`?Qrel=fM%<leAJx>byM60+aV@+iw?(iKw=+hL4g(
zE>wr7>XhR?o~zG$h{$q|rU#!F4Wbts&6xTo6G3#ulA&-02Xi^HG?W1e3D8B+Iha!e
zi1}PNKy-`#4DS>$gVzJdzUNftu)1`zGF*r)$<iE}qObU>Ehth-wMCe8<2fa;$i8xx
z-&$wF;euHBkTEJo#P>O$C8C3jPVK?*;pGX4J7feAnvq7MT6(e`aL$A?$?K~3)1La(
zc6hr)qx2XUc{nuxv_7t(H&oQyooR~S_H3hC)q%*Vnyi!)>fhXap$V2IW``L?8>5Cg
z3EPHIeICf8ZnNE$Q-I(I3d}vrO7~A`F4SiMQLYsS+Zm}pvOxDjF^Bwpisn<RqX+{q
zb^T9Id~)D~ZFnKDDsVrzzBOus$Xc-Z2oB?{x-6O<&n$yHCErm_jqDYPyVjE{(F*1M
zj?xoA#c;PGmoZ(gJ4BaOs9S5|Sm`ADQpwZ2gV+6H(kcwvT6k*nF37STC9R`IvL84D
z<_a{S@MgFWx9D&#`nAbaaM?zwfSoy2`a8rydAYT>8;(2Ol(rQtWzG||pXtCjBo;B{
z!X_|02Hsi;Ab#q4B;D3g4I>Mm=n~dts-WVJk*e-L(C==L8f*@y-7GC;7Pxb!Y#<>q
z7RAqQtth-+F`N~`UKqL9x?7_wLoHzl7>D;`fj!G@A`Bi#091Ox(xw#7FpYOG7up*b
zO=C<!O`wtt9h}nuQYRJ?fR~@dar0Zw;Fv7%REYAKJhCX>ml;si>%AOHd{mvw+}YET
zhA$RsFUjn;vT>E2BnA-rGsNMYiv~$fa0(IvmdK(6HZX|bnA}Nx9)Jtp*b0Lvg#&q{
zF@1NzUoBuRD!(YW;kn(baHM5>ZL3puehW-HtE1?}pC46^9ko1OSA74wC*t>@(LLjg
zpp2QH@1um%FXi^6*X3H$N~+&1yA&HXQ)8vu)Ht|GZvFH<@dF-^Es`-{d%)Q@Zmn)b
zmHBSbX7!d^hk2aOxo(XIx#tXDsJNy1+&w7u?~@s?^o<p|fraU{+aG>x^$R#P#72JR
z1XZ??wBs)YdgC3K7HErJNP({|9zgzM)BN~k)H{~tqt^LjpdSmHUy)k1A)d^x`FIKX
zEx`0UDkC~_F=FGq*1C@R@5t{Q@SxH*C{Al{Ft-9tO@7y9Xxc;udheY-22NaV+sS!)
zcL6ReY)iOlLM=qaVaVz!gGTO<_Q3Q66)_0$z#=S+p@fkxI8^e(<p&onZk0;>jG#1e
zam@{8-;mto&00nT`*)~(bIa~>{$^ixhUv=X%45Lu9P6BAAHUOuv?lCWaYX=|JG`}u
z$d3xd&?gPVVsJ1c47H4UzZ~g^-yEo4YhO0jeulq6U?V67MY=_vGt0iGGhq^({BZ1Q
zJD(_SVraWOor}ZY_k1tEbEPwdeFZ~@iZTR%(nSoTK`dBK6Vq>Jfull@#xOJ**mkh$
zUo=YUhgO!t&_F9_VRQ}_cnqQh$Cb%b4qYkbH-}napeZ^&DKT}`US|H)A@s3^nM1+0
zmJb=CyEyC8Icjv@q;;|zq#6Ry`Q%qwbjINWc)%R2#>G(qm^-4w$@NT111Tt&N@|=>
z{+8g5%aPC3`p*rWE+X#(__7&po;X7jaM0@{{qaQ<UHtk;K2zrStIly)ov5PnHmUJp
z)K|q*34iaPqdYH%Hy&<)PU5cv|9^j0jGL9JTn$q1lalax+5E2SS3m8?7uTfhJ1Xoy
z@h3UJjJkY1?$XUpX7TrPvt>}rir~DjMplgwos9Ic*Fd-<-0HlG^I$T|Lh20kDgAEc
zDaNgeNGy^Y@6azti=OU6+hM{;D`Y?ok98iP<gt*c%!QR;##qRrK{AuqHEq1lOH&bK
zXh-H&ugKu3%QrR1_IKbW#Bb(_uK4i5tWR|Q=&1lOj|CHA5L0nV8tAuzM3Q;7h_nHV
zhSX4QRZ=l9<<avtXeN(CFF-PK1aVu@P1<rXYOYxGs%IvLbXTn@>xb%39*UpNp1lys
z46Sq1xH|Hp?pozTRD|!L*wx7Ij9gqg0qajhzfk6*88JkynIv<euIDluSg_vp7nBAN
zr6M@)FBXlHI5?AKA(qW@CXJ3-28>EZDJa}zc@_h3rJfx_kC#)kf?T}Sv3!{`?rQu{
zBLhBo{P_cc{9vv)6nCb%+Gm50y6@91o>CBRURvGUHy6IyPc6g*#+)AzJXq7yAVK3C
zWuO3Kx}(6<ZTLn&RaDC#&r|gDGKg0oj)9+l5|%5PakL}S$G}2T)K0IUYa<v=<Ut?=
zeUJ5D-OhFU9Rp`wK&S)@?4N2oSW7+HE&Y*m3~Y32?$6`?hkGu(3}6bthTOK7oS*cb
za?z9qZq1d&HsXhYkeU=-2~!o)_n{N&i-xLyyeyFovTIq0f+x76aaIuhP14>r(d~og
zI;2SOD`$oWV<pSE`qssyYlaqNI!9EXvDRO9g})Zg#?lKTILdlDWT+6eqL)o!0v<p;
zOL?ZY;O83HXeBj#DvU^Lxk{vU17GwMk3i$|4=nM3Jb_wBU^18B2!J1MY>|jAogTKs
zciyV7+S6CF)#sz>Gal8pB2qgmQue0R48NwLUq3rpccazt$quKdv`<@#-0k=n?@Oun
zxWhRon@r2+w=;mq-^9~)DKip+r}4Mr%a~X}v~5iXkVX^(M5{V&Hi#|?a4x_*A`v@f
zWFdf6r2=9M;|JZv)uFukJVq0uj3iT2UyYXyXE>kbxR$?OE0wE_QQ##{JmnOoo3Lp5
zeNMy`-f8GflU2?M$C*5!a2{0Qyon@0yyr?Dk!nPDP+w~W?M)WObC2S(=r)S(%yRSQ
zM2dngXGDx$f2<7Cg;@AwI@dMt_>BmG27}tQ*w)uvz~8<IS`Fyv&swdQ{z!}JgCA4^
z`(lC{WKGF)-Lwa5W<8*-{C_>({@3*%2?`to?^qW85tV2_lVHsxeCpy+tn2LHDW4T5
zKqB@SZu+C-xJ6RxaSrB@NF*ys26Y!gMsA_wy~*pp!~MdjT~B)8$}BhGOf&}VWEFFq
z>Q78H1XD`Tf(MwE%k7Y8Z~~?unF06@b0K(mzx?Z&IHLBVo+%d5JX27Jgam?stQFp3
zvm9DPk_M35rJ(y)GY89%2|au68bAd^PnvcrSUss!+*0BzVXCAm4lXd`Y$Cxp^>dI1
zRbYC2F@1dK;a`3xl3DJwUey(egL;W21sl;}cfnB9fjoxGWw?ERP4M-ntWQ%KA*^rm
z?#VB4^--)3LaQjQQ7@(rnT1YKS?&<EK>Qy-!WjT_Q~;g^EMuu<G&~X$L12R=33MX%
z)^Wb^n|EskoqabKHY>D+ayWy(cPJ)4qj4BAF8uWJWP;RqVx}UGVZSy{WK|qWh{Lgb
zWs^wV8E#2kE=1-7!8YAdZMd||2MU`vX=E95^7^lebGJ!HKcadce!f*>bQICG7Zn%y
zKJaPLo0B#qFjt*D)O?dFWeFO{+ELN8{*(3Rvb#S87mwg!_PGZ;shh_D190}AC;sac
z9P7WGqnBM3)fajUnEfKcdWPB-UPe{?wwUw0wlS$Wcz^`{`xmq;lr_p8kUclFbPRl7
zU+MD8UBn1#dZ>)_dO|Kg@WwEsu6o=xsn44w<UkHlCc^lal))$c%|8u3lVI@8`P@c(
zVB{Fc4``j^Ox+}EEwpO>#v)vXvs@-|pg;KLGFeGok_idWk%L>NQljroK0wmQ1Q_)~
z;SpP<%|u$*vYtI~^#LXIYCW)Y6N_-mT{NLH4=|;{aX;2J<@t%RGuLF0$1v)_fs))-
zFvbD?E2jLx#W4a81}Hdw6qur#GJZv$0E3@)oe|!J8{Sj~p(I(cRkM=RNpjriIcK#K
zNa)4p{Bu8&G<lyAvMIAC+?O%;*q1p{Usiu5*_mLY`1O9Fsqe8bZTIdp$_`oI-C}fK
zluUgL|3S6Z%Xj$MYEuy@l13{~2_HD^zuZfc2WGQrjJyZ4%l4dcS3cs!l(prGijS=G
zr^ixrjNHx^_Z&)z>TW5SD8_q`^Ljnyb2<4WIG?(?de^_n#-M9aO%+b}o@T;<7A!ph
zPzul;MbHeIj#xsTp;xl#*4yH=xp{Lbn)9DlY=#FWJQWff)6#V>((_=tZ*87?(>&Ka
z!g}Xs>2>BOjiI;5E^=1+cHbPg#h)oIFSqCp&OFiQjc7>~<4I&FeIL@te+<xQ^6KR0
z9Lc8kG;u3g(bwqholns}c<!e-11LP*lCS&5qP9U6JEuaK)90~VR)|NoTwI{z<6%w-
ztPkv_ZA1eTfp2$s{qE*;`^1h(VkoL9R;g!nwT$ABM4!G7x}F}lT&^>>6)D!NUwEUG
z+<@BP(z@Uccu(h115<PH82*0PskMf})X<X2l~+B6&w%>|EbPtFaUCa(eE3hCkt%uG
zZ1i45#`zeqJwwkec9O2o8!R0qN^lHF0Nl(UfN-pkdePnv2nd`9tFoZ<b%sx(h??|w
zJ1nJm$vd<Jljda*rOK;9NF#rmFhfqZqAu=PLT=X+eU~wOrogY!so?v2Hp*yWc@MMZ
zayb}!HH*m}D2Gl3moDm4uNb0|01}aU;W@GwfUgYEnFs0s2|>$-%z$jTKhgP6r2<9<
z;BAJ^4E8jt9<OIfQ0^L83J%We5N-vKW?O~0(W~nH(Aa3Q3OLqh_z)NYpaTg?{R686
zLoKd9fr*DimdC-T)cS?Nj0Vfc4R67+wO|=DF$`QF&?~1Et&7xV3b(+!e(`_zJ-||I
zy<#H0YgP*yy2V&_4CIs@>UzEnC%@ir=lfO8%+Xa-G<$Dd(M`jRX1{0oywYiLlIH<E
z<`{^;KWSuZo5*JE@8^=Z6SJHY{P20-<-n4Zw;?cjr3l!;!<@$ZKN{0b3Lzbl%*NLJ
zCi`9XbVBSDe6-0G2V=`K!-@Iq*}ZbLUKifV)d<_ycS0hol=_8lS~QME<4@#Xm%<f?
zeY0F41u!-euGH5mbRTioj%=}(fe!DBjXQpEzOlLC%J<3(_gacwMiO=&(l&oKo2IWl
z)g({9TB9VI@J5(wR^z>(eeA_&gIoF~MxEm#QSuC?15X?1-dz{k6a!Hp5K3~vP8uXS
zC34WUiIL3Sk+Z(>e^Zd}v&tm6hO@-+U9cP;M>Ke=Z#25(UOmiDJitOPPe>TFwsGf5
zK31Wt*!jU3lkvV8GWta<J<U`0hcv5KRbzDa-a~zT7?T|2EWq=G7A64)W=pZVS$BY*
z8)RPuTVRGQ^LmD^0YYr6hR?#w%a5OH1@7Nj%&yG@AfaaGWOtVZ4>yq;Gob6P1Cv25
z9P59xO{!q+!`(fB;Z7U!%aQiJKSjW((e0l#Ua;T8-)CJC7SGS;#n$l5;2^!I?Xpf~
z#ER~y3_6aCcCRqh_jYj32J!pxrr8ejh!0WO&I{sqgd?1zjj>k`1B|}nbr|uIxR77_
zlxPesCAS8~l@-wspge$^Amd@ImGaUu1tbFMf%++p?^G%Q!$}($7MheHpd=@c0W~s%
z<Jb5um<^~zfMq_nMjRZldiDp_Mgyxu(-fS5A@WqJ#j+_#ZY-ODRVbVRB-8r4W<eWn
zO7J0x;Fto8@QyO5h<*rIEf|}FNj6!Ok`zd0a;o3!r|}S$8^dcd6w`&0h=92QhU0Rd
zm?e##<M}@O)ziMzpQ;Vp-l}dZW?PebtF}+2hv;?;IXCl_A#0O|0xW>S1zV6`JXsgj
zq?v7WDJ#3*!UWN6r=?QP`OrAM1@xRPca;VjQXE0f9XXF^!HyYD$&7-Q32DfZmWrOc
z#wvvMmF-}1mDupy%=i9LlBY#6brlu1HsTNbu92)gzFhGuar?5Ms)+23r<xP2JPt=*
zHc?N7jyBu!z*r38O&0_wb2kL1_5Qg77K#67n(VHgsZaT=)w=#(V3@KHa16NN&IMAw
ze&c+S0osde;dS+p;&NE*eA3!GqIT1sWh`0TUW=ofWC{E<03In&K!}eD-2|r_Z53QI
z8IAgEb?}h2!V`O+m1?2zj{>52C>RY#lD3HezvX;1$kbyGk%;<wCN;~HES&Wgy8bOc
z1uVD&j7FpH8NDNF4aOrOHo40f22~O$2oR`*oS%iS8PmF@s1Rux?16Jx^?+4ZE0muO
z4Km{_^{X$PW9+sNi)J9+nBp;I4$HEtBZMHDNFc1qPc5UM1;G%d<+$OFyj1GGI((5x
z2xgWbZ|ppPHwFo<dSN<hL^+gK{*Q)(NDZSjzElW)YSS3laih&X@nNZcivsD9^zAV(
z`Yiu(!J0tw7mDY6Z{tA(`Dp6M`jkjXJp)o9g8`L~hUZlT(4YcLrx5S}!zs9B(;pcF
z0onzd&Sq3>2bkiHo)3l+wI-dZ^0&m<BNN_DzirM;i(qwF!`~QRBK<0PbvIKlQe?S(
z?5()9tneJ~$$+_S#wE4+N59{zqub<`L<U!)76qLYXh-G@l6Ezflzp}rEwRp*Czl(B
zLnMyGHQkaqevH*5WjuYk4%j`BPUr1<5TkqJHQ{<g<4o>l#R$~-Y}*em-nws&0g*z?
z{v*;@3XCNNhgOVKupT;lS;2nX%+E`pwa(!7;2Dv|d{ksz72P)+adrDq%ZlyW@^J${
z=CO}=DQ6DT`}LGFl~-RQHrV7+Y10h&GjsuoS1;G7TR%;XJm%|WhU@((E5qq0M7`&8
zYN!LIQs}#=Qu+)CUdb{Pud>jE(CjkXD`oQh-)@5Z^^3ovFA}Kr>S|kuzBmRRd$eux
z0=uO~niA9YQHCS@sR24jH~3o(SF(oNOdiV#%5DM37lp){GWVjP@$DpLOZmW}k6DiW
z;=W(O%B1;vVN2ioBk`#SkSrM}>WrrgJ!_eQDv1-!mK~0k$E=vbp$rNH8fmrY1pBK)
zK+{$!njR0j6_Dk`apmCPn-#$zB8i0q(`1=IOb0)uU)WY)F|b$eq}qlKh?u_ahYnT(
zBmu1WOse6HEuzcfGA8k&TOqYPE3}XrECi;IkT0Y$5M%c#^#RoV{sdtk5U*lvz#-Y3
za|Jw-8Kd=ek%%U+h-9vJ&KGfcPK`wxn}ZMt=cyL7rPO0;Qyn4Wl^O|du0@d@e#(Sa
zBimV@aa*Z@0;7v+r$$bomWT|IA~AS+ERxtwEXn6$Kr&LpsblaoX%-j)1CsqsXhfpK
zp6pibo6BP;*Nu}Xqgft){@b_bAMB`k+(LRze0hYQK^dXmi7g1$*c}Zw9{p%bI|g_&
zG5^}vd|q=5w8(zb?8LFIctx$g-(_7Z-M9a`|MW=v01Nz2c6_sz{Vl|v`DYHory*B8
zVIL$`<8yn43%y(ep3o93KWTB)sK8Q~jRb0C0u1@!aP_c?x8yv?vI*S)SN``;r6nHp
z5tZ6haZgra@g_`5!K?}HJWpQq?Vn*?lFA=CEBnxHp|}3#wV!$$4D`+7Pm)i@Ry>TU
z$r0mS_;OI+FRX#&H}AypTC1ELMEnXP^+KIj=W~OhTC06<x0nQ2`=V}yx%~O9W!+UQ
zWTf=0xwyK&NER<mTrZr?-aJqr#3}YbmSu|s5*i8UavV#2Gg*#bV=)p7IZx`-z(N9b
zK~ld)<>U~Ze+2^-lB~iMh;eWjx{HOn&XcUS%I_GSD6BH3y)*NGV^Iszjvd20bK0U;
z>Y1QxAlJ**67~oD998OvG=e22V<RQG83frmcZfiW#$t=<pK?(F!`xm9zaxXflq>jk
ze7-CuCqLbMttLpZ*WywEd@@?+Jn@gv>O{ZAc_juUCum7eEI@!26N7ivVBzR!TXz3r
zJYAMU8J$bWGW@CJp2r2EoYrm=IDM<smZ81D!UYrAIAADYyQ*%Tm_B&Y>%!cfu<!Jv
zG?CA2f~EGE&v53N7cyce8l?0w)K5%lUqKG4P3UW`H@;|0iEzhbUHLLD2xli$D$0<L
zmX0n$ib5VP+Jtnw2Zwnv55`GxG%;zP;7+|kCurCrs?FC~YhkZDiksq6k%0Z8I)Vd-
z_YOZDRDNljxYmPX{;%#oB&^Lu6DOu1N6E_Pk9vW={Q8b{7AeyP<5}mFtefanfJKGG
ziV>H5Kw02{!+H9K%W+3*x9jsn!&J#Z=U7cfR~pp$2OYyJOQGfS*EO^a>V`x|LcB9%
zeN{Q%FI{7O<fgsNV^8}L<CkZi$DdL<6?yeTMtaQ~mbJnt$pv3&_l6tK$YMU<-{V_*
znwf=qVFD?tq&`1G8W}TB$V?1I{-eiOB@#08fEfmusGkkQ03iMfz?+HGR)2B4dz60Y
zJcxlUz-frIG^7a(gWy_)-@$j_tD#u09CsADr4pjOuxt4fL?gVdV)68X^9=rJ;cJ*5
zD*758Z_Y1c;#kxPX>waco%?<TF^|RTT^jtfF&s5@n-#lMR*(zlhF)3A%dLN3U%abi
z`Ge@}LzI0E#?GcPVI|5e!3A2+xmgU{H<3XCV_?=u#TQy8g?ZbFB?HQEpR(5uAobgS
z;=;VL%~j(3i)HobK;K%mS*nO(MVnm3vaLD|&giDU<$VNdM-tKI1fJT-1Ick-hsq9F
zZluRJy|zE{>fWJO<r-~~a*+#zP|7+Qf2mAsnfLlS2A1>9W!W@}y;U}j6D_4QFcS38
zd)(v5pCA>l^%VBe`uD<@BhJzpM)k3TLHEIoNrM`hTQiwG0}o+*6_X6(2fK@-W&I2H
z-*6XR`F7_^cfgIR^@f|}?irUj4J_po%;~KPUg977Oto4nKDBL2a5a1=67N?O(k+3Z
z-Fp%p%D+G)vczn3Bt||SpPcBG9dV7JyTjh|?vYCv;wP+f_w%~yBi&!55B|@_UmEcT
zqzgApuA9D5wf)$sIecO%pwF}d50FAimNFgePB-j7UcHq+*9<uZ2CjW@PZNbCIY!6m
z-h6y@_<noHjV_sE;K}QAy{7|o_@0RvTSI$$QtwL|%!;N|h~t!RDVI!m5DF{K_OZt=
zHX?1t->$l3+^jm1xl^~(NgZ_YZ~dgL!tBpoU?S>s{ntHE-_~93pIsjMjqqjGsZgct
zDiYG(1b?tuj`Lr9FrO0>*LUudVugaxP8lucC4{hamZ9bOyVRZYnk3~XCkya5<uCEv
zRhjz3%vndYNO{$d1ykL4yl+G(dp-YL7P6=uhNa5+7}az`Q`DiB^lRU|vHMwudfbn5
z%ho7(%GG_N);jMiV@B-9fJf-TQf_!jEZPmnG?Uzg28fiLc^rR5VSRBWqPYYo07?q_
zAun<R&~ysRrp#bz-~iftZZHt53Zi>%xsd>Q!r*~9?#c*JQI4N)I5(=+1Gs-wAMama
z$65n@GL1Z1KMD$4VqvU}R4}IJGMQl3SNPJ;#SdBa)PbJ}I?0OJ`(QXLO8c17(!BZm
zYZee;;g9V0crulAQJO~0TA00Ft5n_!(n8}wx`D-@Yxq2H_dS%rfm^N+sm6?w%mE$o
zQ9xdIydUyMtmGVngec9YNdtV59)E%TlcD9%X!{q^ec8%6U2{<jR*FF?EegbvW|k-a
zL>WM>Cz=jPG@*ubIF`iNt6YI4N_3SR1NJ{cmZHW5Ne)M%hSwOF_lkCC7j+NK_G33I
z(f|Gv5q7)cuzk;>&FX0C5D9#(0`(%dv;2mRfhJ8uK>4q0@NH?XvQlvW>1o;NGXa50
znKyD{u~=7&q>!sv7<uz2-EQ6d=mPDNVA6|Uuc~PEFFagI&?NZ=w0>xQR9j;8V$yo-
zvt($smHumg8V{I&@743|CX;t&io`Os=p&pY_LFs9yN2XYoGx9D>)VOfF)6y$FW>yo
zKA`t)*j7311)#QZde7ZX^G7jSzA#hA)eWA71C<9Q_XxrnWd3TKi?&una=nh8S6K!g
zKQ(z0*SDs#ew1yfJ*RrbX!^r<`oUGp^M@legATRD(<^4CeWNvE44QYDQa{$4&YZa`
zl3831WdN0lpaIe3b@_H(MQOYGV<8v!tfIl^xCzE`0fNa8(E1D%W<p54Fcyz6;`v@s
z1}lGa@Gennh17*+a!p}#-F#$X^>*7P;DJgQEuW6C=-p3et;0e<nmg2LzdFEf&@)%&
zcGz6h!%V}u^>hKPS$-374P{>Uu~p9K4cET?&Rn@0gmAyaU`FK+AuM{O{P>l6S-!G4
z%R{Y&<mJJ&@-mc6m!IF+KqB_i6*E<uM>Uty`+v(ty<L(8@xAZQvqs9;l8WQ$XsUa$
zEQYtOgBHdUyXv&`@rAmQG**{+#uF226!`no-j4MdYPH4v@xQ(D@uTvXT0ZO;21Hn9
zHIR(a_DU3l&mM)72=fsi=XDg93oV7)Q85ze${=`?W+K3F;{ktvzkpIf!1g_VgWcR%
zn};ceAC~UDK(PL#+peG>|H(uD%}7)AgUO{zByfA(?>FmKZK*@X5^c^ZQ@5WNufbUn
zt+=$@@ryAuy0q5}s}9Z^P3!oaJlyi&uz0c|N->T>zU;nKXtN-eG{y2|+2+6k%q^Wp
z#|&+Tb~o;7Uvb!Ff85iXv445LX>qcwpTv!iy-IpEcH!g4yY@WqiPGQJ=`GUte+7=S
z-WF+Di@6*cvtX9NcTHx}I{Dp;55JMMuImkvulX`t>cEQP?V8E`D`a|n=uT`)uhye{
z?3-`lg6P08+OrFEM>3TD7?Z9n1E>WfT>43OQpy~SUaj73BqVBDYLNWq^<q$_@5OX`
zj{S@eY<`>F*B1+ko81pbPu6+W*gq#H-b}pgJMZA$Cb%G4+)iYc{73Qfh!qj_4ipqE
zXBG9_VcIBSIR>gZ+we<9+R5O{iKtcRTkTtaERuJykg~;0mx@>U#-RNjV)gji8`8KO
zsO~r$TqT@c;{Sd)rBou*;)V3OKKBuf&|DgtYD4jTxgT~yyFdPFeYbkD@=O)P*6NAS
z4mNeO_L<4ra9k8l<{($I14(w9{i3Ji7s&jECI<V>s=pLw1xADO2DF|iu9%#dw`3|;
zN1Tp2zEih!x>w4)Z&X^8B`O(wk;*H6P4h{Zwn%-?NIH66<|@IFG4xX6;TP9#&vl_A
z<SQq;lw=;srJjyTQAD*^5`%&PSlAY)(2f>;9MyumlaRgg#b3lhC7=oeipd||B=!mT
zVNPpktd>HCcam3(Hp}#b9%-_L#xTpAhMsg*aTtVsd>+Mq^=o6Fy5unsY)b9ajS9>W
zE3)r>N)eSVtoQArDWChom!hL;x@0T)W~RI$?d|T0$r;y_$6CYg&;0w}1r%N^WDd7U
zsOhLg@-WJ%It*9kb=$4p;*FXK=H9Q^wJjbF9Dfl(&56qD9}%@hI9;r$?~-~~d)IjW
zfwyG_|4C1-q?Aw9LQaOhJ*T0ZTV-U3KbbEQg##&EoV*c8F6!&RjLqxwh<r@$HGEjc
zyB94Q@*IpU8bG<K5(Q}tDx6zLec#3bvz~F@?_Pz~Lz9`5x1r~5M)*`oiJJe*+y7%v
zC6YG()Sr~P#fH**_=ZPr`>SxMQT8Q<93dshyWJNN)@C!}xo_ghB*b&w7tx{(DO$XC
z&VhA#QC#Ex5jM%>N46&}zpOU8+&$kJn50B$KFAv)z{~m{iM2&^CMI{hJ0<<{aC&ZP
zR&%pkQ#R1X=FQ0MvHh3#Iww>lgg?G6+GB{u@F-v9d=|l%=aWzIn86xvsQ0w<VO8F{
z3yI7Y()HDPjfZM6CR>YVn4e|~@Vx(7dM@me*tv5*{fWO+XK=C=@%5i%jBm8OvnZIf
zal26~U?8?vc>qSuRw;w$l!FaVu?Yp5^l#|}_hGOMgDG?Bz5n-Dqm3nX0{^vBs)paV
zH0r2);iy^go7X!T_UvxS@S%FTgkbSN3JDIR59bDy8wzCpwv7FWVfh-}E7|zIn@H>1
zch%#zRabZ+@jX{=u9D9S$`UrUx<RUZ@K)qwySNj^xdw|u-3Q;XdbZ^UmW|#rBo{Hg
z7Tp_C4+n2uDe*+GUDK_NiZ{zTdd{MO*PT&&*Q{%Ek~m*rj*TlL)=Qs)fn^YyQ|lPe
z8~izRuf^^`;l6*A8cBp#HPUE$C27P9RY>!rW`kiaSQfU_;QI7JBb|C@;GT84$)mu*
z(Qe;4x5QGXp))fjV~fUGGa%pSLU6EVW+6c{A%DxIK@u^p%H`gz>$|U2Uh|Z6=!c|D
zS~jf);IARWVl<{2{6ovF<KNv!$e2cUN?Q58M{4OAudZOSOx@qU=&-Xh{FSvqqgAyO
zX)5)>FPXerl6bbbEQ@9$C7-9t#bT(Ja3;ftFHv*Sfp_4F;@LdTM2`CVuI>A!JYB=3
zfz~H#-pVUh6r6yM32P}lNa-xk>FU8bYsw$gi3%Jf?T1@F`C-%e+@nN>4p6IDu*o%c
zLNX#n17$_|qYYoDEk3$q#FKHQ<+bXlKCnONBR%i|oEvuLLfjLTiA6ng>`QhhuQ1VR
zF_STa?_L#@GX1!gH9q~3cP?-CK2=o7pbFc5&6;m)VJ*$8e>;}($8#>#S2UG8(mOja
z*7eGb?p1k?D%VA3J{qf!I+(5j6Lk(``g?{pi&*PZ4Wi?=#u9-9i)T`2tr_)(pDQ42
z;UC;^Y&GH<Sb&V^6^<V@28HNkC`?zQib)Bt^vxWCLn#e{Wr_(|h{@79Q{eRg9y(Lm
z&e?BgGC!77lpp#x60G5UsigTMC~N<l1p0@~mf%z(>29w(YH;s;b~(&y10{h~2Nx*6
z*`lt?{w4hbIo3kTMFQ<MNU{FW3HDt3TPJArw@wfh^_NcY-**Oef`45{^J(tiI>G<^
z%%?96u+l-h{y+9hZnriS!&I)ZL_Or}kCdwM4$%JyrhK!bpsaaN^dLF3Z0AH5YD6?j
z-bzTD+dT8gUWdQ1wc2h~IN*aH&$cQj?385EnGD$0Ov!kNmGOoAH~kf7o%6^79=d^;
z6HJsiiNfsDh%H+9frY=A!!fY8c(eEiC<Dc|IaMxki76KR=%3nTCI^lBz18p?iVk`*
zQBj4^BuR-|s?{jZ*c2#n?a|jf*b?&P9P-#Z%B{rt$U5DKy3@e=&67WZE5BmMHBiUE
zNe9M&JVhjx%+%Sh_KdFF9B=0@JtXx8hNngcy-&LDv)y;p{q7*yKAlBEwknj>qVfG@
zc+)i<5NQ24;SthM{A$ogEFydr3_|4>DV&P>u4`KMXAYz13ATkuM<I9#4sKPc#gOiB
z7tgYwTCV;jb44~oPH7ehHy1zY=pTFYMi?nzgS`bh0uxN}l-Am|@DtYI5d>AY3^EKh
zU)ENx*)?YzzU-&F($f?6F?={=tS%5fY1W4TZjH0iNb-S{QGH(4Sxhm1>ix5yvR*aG
zS1$yfE_9nw8yYVgk<D*aiY&Nr>9nmxM#SK7m{QS1s$AFH&ptnrRF<CaRRlo}7qhf`
z?vvw{9@i6RoeGS6RKJ4~Jnko_@4OaXzi}sCsBS^9=bqi9(k&s$CF@*`%BFjlFZH{z
z$C+1k@9JFJn7HDKRH}|{)QCfOZY!bNxZXR3`H30K-M#PM{*AE5+HqTr5*&FBRX8o2
zHmx6+MDVjxC)`Fl24!Ctidl|PA1H4*Ri<P1YcMHD;XyKDH9~X2#yXmZf5^bD@_jxv
zT*5KSS6htXP({)5@=c!JU@cqK>s<H`+3bc-byvvZDBK@$HmJM9JBou)Ifz~UnO)TM
z2jUc00MX1mVv2_y%?c>%#W^_FEK+u@k`9NbhI+4X(^fs#xgik!uSpaC*bCt0-WzF~
z(j3n1`uZ!aW5!Fpi1ir2uDOU|rDqI;!;<Q<FNMIK>oeydghB*Xqu9}};DI~5xo^26
zgDQuJcMX%hxhN8@&`U5$<(wM~cRw1w(~Vfa@Hs@2-S4NOROoc!Fg^Om>t-F{56Dpe
zmqzp4*V>}nN6)Rrp7-Xy_+;Sz5qqyP+1?~}qDbC1-*+`DOpgYg)qUp-HS>clF~=KR
z9%lo*b1mr}I~9g8C2#9zIQpLJ^<|fvil)!(W$Hi(-B9bN4Vd~gY|ps6wL9-oxN423
zN#+zuz7q3el$fTg>Kx8`Yx5r^`Cq_?4Q-O<H;^gc>M@(sq6zz%5cLWF9pD^pP$4zv
zSF>(j{<`-qDm!$%-%UGxGOK|~-luc*!QdU0@0x}myl-^HecM<xW3K_?)?Y97B8`K7
z>U0{^v&5w9f_5Z40<UML(Q8F+l`%Pd8{A02Qk{*kYINK%cHYHp5q&Hf<}MHM^J?qX
z7R?CQzfOI6?E^fDv7lqCw!-vxdZiZ7_8cF4w0^5R(-x&rIrvHl``M18KIdcjPlwbG
zwg&E%Ba^kuMtRo-e>_oS=5=VqltBo8Kxc+_;~eC?)-#b8g~jnY>WQVa7Qa$5!!m^B
zRDbexEjo0TPBG@Sd+|`5MlR1OmZUQyGQaOMu&IrY>BnjWr0))^<ID92jXo1FGRQ*@
zTx%sfk1_Voq#3CEeC5~oPe*M!Z`NXntb#+1H=EPPZ}eso9z8)PFq92@Pm0IYW?jqW
zd`OcKORV(%bfNKN+9NgVyPI}>*XFD~*rL?Fr?XDe<nA+w+yE;41H=<B_nLYgQnl|g
zQNaVDiE_VwYX(yrNF*q5OA$gi;uF-XLT}jZoT+i-YN}5IK9Wo^A0-FnGc?TJd}Zq$
zE@COF{-7<zT$ZBp*kxeOVDtA?B_>ihev<a|=JmsugIP~E(Wc7cXX>0)pLGQMv|<Uj
zRku-dk#Ut42DU1=olKQ4oVXnOuKKBV%JWY=4s)Jwr8#xG@pa5&cwcbB7l1%J<9I>w
zsWX$MJ06qI<Du|oZOOz4p|V@W6fCfapYd6!5q7Xdv;>0lgriM8l=`77y^^eiwBh-Q
zvO-s(!qQkPTmJ0)5$+(k<$oyMfQ@Ug<sAD#hx<IryG}-|#tR2hpQaKkvWE<wNLEd`
ztM>i=I|EI8^`@f+)Oo$qk^RDV6j8MQ?IMAGTZB%~<J%^mj`ELEo_)L5Dj#X>fq*(8
z*KszDoA`5Y15@iak`4zj&@VYWmt*uQE%bfl=#?>mzbP#oU_9F}5SY^$Q16!eLr*A@
z_~$x+ptEY%XxMUcN$wA3Fl6=}t|NhcSX5siIF%EB?)BG=agh5K!LdG3KG@ORc~|S+
zDasp>y5}yjUJN_C)6up>4pN=PFRSF|Rts)k{<$|Fl^uqDOQTwH87l1Z0F-{u76x8|
zE32NRtj^=K8(m0(ru8f{tEBeQVhug&ZPL%5`W_4@9T>L?+Lvb;BmaHm`}y)ZRbX!r
zYCVgdLq{`yZdEx)YmP)Dqodxw|4q2iiWmP7nD-?J`GweWKRhZ$QC*-!8q*W|`#N7-
zFyY$Ud2hz$Q%;B9@9$X#Fc#Pp%w9RM16#SOT{NBL+&I+~KMlM5p1(Sx6lOj%;E^$)
z@AStBipLvsPWw-BO5R`&4W2qY$V2d@Zq$gZx}2)ImXcM9ucI}*IwS<Lp1R&F`Q---
zzfHuU@T?S^wf@AZ4ej(ePzn}RgOLrk5~zH)s#kNZo?=M_<I(87E?PqUWyOL;-s-L%
zhT3i9#pw&Nh^lO0(fmpxbZ_V`P%^9koPmky(>vRA0!?}`LrTHY(pz~-UOqLsT0gm2
zZ_a@gj}B(JLaP|yrjXe2p08cjR42j@Gz@M%>8FdIGbG==6gRV9tdL0%u-npz^uNe%
z+UcbmNFQxJK50Px>7_w(?HMnW?tEJIy+x#-i`lYWD@)dHTK%wj_$OT(x4^h}rxGaD
zYV5-=W~EBAzia89d(Ok@<S#-%QNKwthcrgx<?+tvlLlYaN7n(R@MuQ7{Q2`_01upL
zE|tQNd-~~i%BWjt$HWKvhm!*@+PGhlSWW8-WNzErwK=D-!EehtBg>cf4IA`CPUXy#
zql=$-nl?zqA+~j?6~DVIxfxuGdL7mg!H@MR!HDNl7N_+d+<O1rON3Atj;H$91i#Rb
z_GFq!u4NmA7cYA1s3jvAqmm;b-u#D;^maZNHPZ4Qe5uZvsrXgew`KSI%G&o%uic*u
z4P%Wfcw1)Q{f;W{%=suThSV%q7tZ3~>od4?=_KXF6TPp5rAHMyOW(h{r4OM>q?>z9
z%cqCP=&DKn^l6^Jr%SL&am$!Bx49Q|<Mf9y-36v<vs2jyQaH#8;q0l<7xrVd>B^aW
zi6V)FfucV4Gx>tawfUl%8(Xsv0G2KXulq&JJyoRzT8jcM`wH-jH*J7dJUFC$6I3@K
zMJgw=gZG~Ou2Rk!vbxTGOI6`1Z05Tlxq_sCT-7&o7>!UO%fk+~Zr2^Q0yQIq%&D+x
zTv&vod0p*b-CgSYfxsEHZ?O(#g(nO57;~F`%}rE##rZw$I#tC|{WLDHIZuDY->zYB
zkZ0^2O?)!x>!SUvr>$B9x)*V|%7yLe`*q!Fkxl`B+qBTvDt(-#4;O8%ewXl7uAFVZ
zA1E;S4J^AZx}TAi2>C$A4`m7KL_!N6R&>kM?`08O8~P&&*XrN><nz<Ld#ep4`%(^u
z?QuLoFJ^m*#28H)xi@&fn1McZKSKBCXEekU4@lix%RLA+0H@e~JGv<3Uidrc=!fn3
z2humCQzV+R9|~^rZ5#u}AS!bzui(qFAx6ur+@o2d#`VoqS)rLd4{ixV1oR8WtxVo}
z4P``_pNml(-Q#t7#m`)^OmNmZY}&TirSLVZ;B8JPP3W6Z;hkQ%@z)(J);V~*aOyPC
z^+gnxr|W;F+z*n_2($sMt1n#Zv=;6OO}>8;{~Z-LgsW4-;fRxIZQ9`U5L3DIT;{yF
z9dFj9_lE<&`_ZO3W+^xE+0p`0NvtHf>y<OXo!3I9ZEIxQ8@$KN(8g+<-j=*-Qi6QT
zHg+50Mt8d`gcdexb4`Dz`u^tLzM&gY@9uw4rFbJ#&5M=8gZm5Hq%NpBXFRCbstLLt
z+#}~05le_CpyY=vdH5ID^dsTNfaI_J*!*kb1PYDScr4sHO&W@0$-3WF11PJe%P-p=
zyuc{Cb9Q(t*ZAfl6Xrv%A7ZJ*vpuD!H~8Ib7Zz=may~*HH9Hh!sLhs$Uv}u^^Y32L
zKrMT#*Bv}8p4#=f$?!Pc<;NP81MERyuZEMQ97OR(ht;dQsa`T=uN-b);=2(UNJ)33
zE!W|c+nmVIj=Qup;)~FoG?fhf?Bilf{~kWG+KZY|z5l;>d&{7>-mcBJ2^vBm5Zo=e
zyEnnz-Ge*9-QC?i(0JqS5!|hD2<{N1ame)leV=olx6YZVIj3r7zEGRy-n*--*=yfx
zt?T+Ng$^e@#?hFa1xGhWs+j&oattC($?WlKu<b}FJ5Dlay|=JJk|Ph>#Li1tNf(S!
z<%d^Y+we_)4M}ZP)P&LY_zT*n7shXkLPvu59*exG2D}+@Cj~9G(BYoO>i{jr;b%MD
z0p{qiuH1JaeC5^Y+#_`PAM&Hqzn5!C%XxlQmH+T%2&tu<krOp(c8`FH@jH%TCeEB|
zmQ9&Depn2r=&vc^BNyur=F^-qObFbp6DOkNfXA7(jgW9g`DPy#a(I}7_KSRssBRNk
zkKSypPglN1z{k;N_d4$VqT4_D^*`jk23%ekm|vw55#QFSB|i0yq&s;z<duIN@R|Bq
zzrW!D^aN<fxhax_V-rNf+%+(r3G4^l{_8u#Mnf{gK1h$;yY^zwFVyDsfW1ijzkqm+
zHXLAB;gb^ZV&E^pf99o}><*?c>phmoIOBt1T<~7*q;K;AH)CI)is1iWs}2FLbBRxv
zftzlDHbXD>eHWboVDrxFv+><J*$PZ96u|m)4igCc|G1F*_<g5Ds&cAXb-=O0?*n3$
z;5_6=-~!l<Z@UY3=pzSrNt@z4#tSEKwA;}x4+{@We~MD;h%PX1sRoRS88-8qD^R53
zgGs$XSA1OS%Y3w#MPi%*T)4<}C!*AJPCt@b1(KEaqW8@<vWA2bn!~0SM4F*oG0|Nm
zzgG*v2B{a!fRW#@mP?<W`_KkWFs9jDPRPMIF|4DroF^vIVm9%4A<bVAt(j|mH9^m7
z=ov*rTD?)>yfZ4WzY?C{rLDY9Kkb>rmiH33py!k2BmK8+$y~d8r{7j>Ok#P&ahgvY
z{EzB%^`f-N51M1lQn$l@0pAQK@ai2IY8dlL5))eV{oV!Rj6TJe-_#~5U%rI(zia>0
z=49ZAgU(nio10j3bf@Qc;p9#faMnJFA$R$Zpp&eLV=j?Dd?c`d#mr1jMeQAfL0MA~
zOXe1!0hjOi&?yq4fVzstE(5q?oUHxLdAx^AHBSBNcDj2STUN(NWy-ZmPJ3@_DHsi{
zeRO-EzY?At`7tA-Ijnk;lOEt$7*z_LKJXuL?jHg%<Pozf>nSTLTsS}Zm|a$?Cluz>
zo*xq(>KGw&LG5oqknf~>Sw3bx#u18D@aV#J0`m`B799)niqgXY#<qj=@N9f_>^2>)
z{LD(@4Z_X_0p)gkGD?Dw26n5uXT$|lo<%JM<(xh~1egbT-AL)q=*#fu8D+V%NZ0X-
z7ABF}2||Xday~(?3CS9+#`PsGL}lbUNDD_xyk@&~@!Amb!$K5mdObCwa^;~N{f{SH
zkm+_kA7ydO(aNoL;<{4$=(gKjwzCeQ`{$<SWW?EXxuO$8jrExUZYxvDtTMBlc`=76
zJ)RiAfB3{e18icZ!J_XZ5s_`_KB&p8QdtT|Sg%>}dv8aXBx^`_J@pedl-*}Qu8f(3
zqq`ird+dRYvTL=5EN&2K4tiwD@`c)rF@=tH*I;7vSK;c(>W(={YcK3JQolw)9(jKL
zsyoG2+JnnFtsI!Jt}hwXv#m%!KOS^F#eKek+x$cm59ohSz7a@xsc5b{--3Sweb11B
zLQB=kIOQj~BVV1yt{bo|;?jR*S6hHNStvZ|sW!XE{{l4VjtJxs%ocZlTAuJJk7#LL
za7YtD>?dP`-qniRFcP_`$oxl8I1oW6<DLv$er-7fF5d>2Bm%psz8XI7EaIrGX}rRE
zvdbPD@vxYn{%P)XE50=CG+s(K`wsw0Q`(Tx+UdR)U{N|E<UM#;ipF`Atk#-qTb(gX
z3aWq$Y&`2BdVMT;Vp*N(yQl&H|JVFp1H>CZFDzLbFK(^DYH}@jvma!Y=w&tJ!<vE6
z)Es!h?9m!PJ#&F~M;L}I0**O}<;%vcqjs?&43#=QR=Gl#mb;?=EoC;eBWDtGVOw5I
z`OVG}J_4zcF7otTFhG+C2|meHfr%G?&;|~QFpVpAU|p7h++NB>o=QSr9P3^LcvAcQ
z#~R<{m1|FGNSHoqb7m2D=tP@%Yw<)B_sd^^PM@Cc>t!i|XtIi-$zVVlT<kWo*Ny64
z13?n{d*@fPk7FO9T-h=l{x<!u?!0scg?e5z*>opDw1xF+f!&OijU)l-GWHEh*uPkO
zW`C@ysFv6(FO!{Uu7Wx6SJ*#O&a7k&>%Z8}FO~LBwQXDd1=L0~|4f!aGBCj3vb?QT
zmq*e+viNfR`3_81cFah$HRr0ur5j-HR%!GBoWG1btzFKi%S+oZXH-u0Nkb~O_qUCi
zig2L7s7pdTRkUJP%JA5aaQGMrks=edv^0=gv})r*DrW|N|Iv|}WRmL(y+DF`dm2`J
zbmz6&AcMh$U^6$Fr<7hWy6yf{=7r{1de;Dvlv<-gh@`_B7T#>tn%M3W>2|wwvbI_b
zZ<D+_&-NS6r1vhJ8+pW7crYQK!SF=#J?dq1AeO{xOymOx<3_0yuaY+gEfvnX{Fcz;
z*an9#mT!~^vW+J^H@Hqu2rtK;XZOqqqmSEqP%7hFgfR=gx7pE>&)in*M1F~%gAsd-
z?}ue`y^(jJ@Xvkgww3Saolo#EyWS7M+}xQwgGuS}_MyXE1+J_{l3zxh-ljCHlS=!5
z$ibDTR=ge(wj7LVNIOU=Suh#CDeV4$Nk%u~R8+c6b93OLCaX$Q03>!(LekqdwC!HZ
zoA$b^-%r$ixeTbrCtb(K%;%HkpEfY$P6tP)jmizE{4q5DDE?ftffXaDu{x@ZaomB8
zha=YD$o$5iFUdw4>qCleLOyHfQshuuizCW)SeX%8@_Kf~^(3)DDwT1Ie?7*Ba}(bk
zpn*B`vheOvJ4U08Z!>z)@*N+QdPFKP-P?SYjCrcX7aO?#dYxS77Xl;O4EQMjY~MWV
zF9=rjIxtU^{uWC9Pw5(dQV9Q759FWwynN>Z<3ukp3e4*EnW0fvxKo+rD0C!kjEJ!@
z5gPNRRb!V-e5Jg-_)p&1P=uc|<4Mh(cCe#`EJXr|;=F}LQhp&H1Bg(l!h8}#!K4<v
z#Y$xBR3q)}Rp;Oh%5@KS)cv+$qPZMlZ!Sa9hBJbkA&ih=-0_E}uEeHQi<GF-BU;Op
zI=G}BbB$dwRY0twks+wR5}BU>xtVY=<XPMoXDZv-*QH68t5IwvF9zIt%TIyA`vvGm
zO>EZcg9h;F{2Ke6*o~GC25S3vz0jcT*XP3boA7@DBB?O*5H{T9S(r@qpsODHHfPQ<
zXNJrT2F87=a_mo-?L=nm>jNq~q#<uF(Wh)g!eK5+QM$l)y9klow8R1_Yj{QIc0%G$
z1;NoxkvUqBdxhaw8#K{xd2nm=Q63zUf{rl%&nfy~UVG8|eAXA~R@?bdmO?#I-Joqz
z#gFwYU%iS@CbYZ$0<`?uMQ5V7(cM(1o||z-><J#?<?;Vi3BW7sd_QDYQ^QGhvY+n$
z3wZxVpbyor)syO_*ibRyYe2fSLP5R;1z`r$c9>v}N8jZU?bCu`=K|KR0UA0u>-`|g
zB)^jwgJiWq^;WWOv)5%X8mxgIRU71`>tBtbELKfhWQyqV+v$wRuy6RhzqM@TRG+a^
zAt#NPZb~Vl)AD@PU*?yHxlBpjftm@av=KB^PiNO~aJa+CR+3+2US2HO1RZ%6Wb`fA
zAYaPXTraJ*Zprh|u<geVx7{fEXCchlmasfBcK=9}SYoSkQ~H`NzN@=8&R4e}0oAHJ
z4*~3sxEHB4D$F*VaDRJh^_#j+Sd)6k(|ShY&QmZK;GH3sWapn#7mV%iwJXZ&eC$N&
zG@%C~TD*1R!qrNThut}TwMqwSp(i=Bf_bwRxwsS~>(_-znqbwH*(Gm+YvQX-xyr-w
zZ|9?Ae^Mx=Dx2KecMH<7#;jO+7$)Te0iIU9jMU&NZUG{%aUrC2f~Vg3hw8VE8ivuX
zQOL8EGAG9jMH(&h;botn#ft5shO}@hYNQsO;m2a)Tiyj;#Ju_CP`O%nnK)VEV7e6e
z;rS>rExKj(%S`5VMeeT;@ZYgR)9PrWf5WJWnxe0aiS@vAJi_RYUNu(XPNUxfu{)ls
z<_$b3MQHLn>7Nf@*YaPacjR^&wACTBxGq>r!vkLvoqPAb`a(SKW(BOpEB(p#zV9A@
z1#n+pXBMIW+8)4ryZH^&@!LYQ75|9Q6|}vOz236!q3yC#M<H<`@6jI!CV>K1yNbfP
z!$$`^53p+ZTwU%9L5~X~#Q1!Z;-w^TE4`xYlDQ~AQ$KK*ag%Y7MZZAul2HTCXE}s3
zt(AxmdSPHVpB*DUm#}cA>hV*5c;9XlPuA;KjB%kYbPC{*A`8+S#Cy}NUL|cKi6kXe
zxlO{W=%4v9l=kjp3WFi5UgLawYv#f02^t|__FIt9=J!x6!@&MPuun@frFSX%<eRD_
z?43#&b5eLfd5h^Kg!N(!wt<*4Zw%pE-c6B|B(P)+LI%s4F`u0|Qf)d3%4Jiz$R;O4
zKu-w9o0<Ql7^j$l0me_~->#TN=57O<$(rr2h3a_sSM#h$oC|K?a2n=nYc1PadvvXW
zwsmGrbOl`mHS6jnUmgNsxktzqBO{Sjdv^z$Xx2J)%9(0vaEkwgc3b<bBWrvfVVs&_
z_}UlQ?_b%_ofnJA@&U&GdfSqJy=xGqudha$)OF*qYALGucLw&<8v?Sdsu7Wt>j6|-
z_U73<sgSSo&&BN|uUhIiltHZi|N4ZB1ee%>1A+m-BS_%On;;Qj)8K}{LMuOf&iwb3
zf$#Xx+EC9o*>BEoxn`wBZh}Pp1?Gg0;@-1A7~2AB?kjXmrs18HJ{Q^|2-WhRNX?0&
zuO>*+orG|T66u-zB7R(mAmbnuuRl^+%6d{~zO3IVv&N*ud4KTObW>!B7v=gA0z4_I
zmDQD2oluBDOeH8Eyce2}OOgHj^evr+!VCMCyU0hunseFL``X&^ugWcIO9Z?=U%4gX
zh3RKV>5B<7`YsRep;Z$jWpDfy-vr@-k%DOE97Gl?B=KtPY!7}m0_0q8D&aHRTEF=G
z+1bZRj+xI@PDftj3Mz4iuObfBxwWQ>Z`-wqezBV#E1BTu-FqnMdfQwGZNH>=*&ghw
z=bX|@Ao^wDZtn0iFL=O1PSDV%a*AgxV>0o(L&s503J2$W&OsWM?ArG>k4U9a9Yh2S
zg?W?kW0YqbjXf6!W@JQ0-4yO|S^P;C8>&uEr}ZLxS8W^uSr=wEO4-tfY$je-Dz{aw
zZ^_-09c3l7Bmy@Gvpz3-_awn1)tCHKA+on1t~$47?(j3a6q$r$5tmXQjTyGL!~IZf
zetJns66}|C0=)-`q5^25GrHN)a2;u>xA~019`um6Jm9^Lb`Y(NA2(`&Tq~mwt9UF{
zX9h*qOQ)9bTgv7OEb<lYtS3?G*OnQCb7$G*cTB;SDLT<t3-{Fm;Cq>nip{~#Hxq`E
zWo2wWdA^KX;32<-e3`<tk+1wHeWMKmSiD^MMl#OC9p+Hs>XHrWI5vKy#++l6`XDYC
z1n_{{JO$UZM)XY?+6-~`V_0c&cTD-G`Jo!{10<7Sz?+3y*sxEa_KIDo0T?eBBJrM`
zwAnk(e?W1N&eTW!%HO(O2lf0!hU)V(W!VlXrjJR0MwjSO3`~z}nkSN$MWA=n9z-W>
zmrfM^!>mlwEYP2%S-aD&g1_=6OJN0C_3@|7w;l!CmlY_XN5}cCpIo8d9}GeTNB5^&
z+Cu)md~bD{*q+fG+d^_ym%;eYWW!WzZgqyNxB3qlXC#3?L<6^|qezE`Zt6@StcVer
z?rxh`6K~0k7PiywMEo<hY8`3PCw13SY{BR6YYG*#o8Gf+Qem_b3XykS8*TnZ5dw=y
zLCInD44GrQ^vk;_^!V}`t%Dba0K6ri>GT^;q2Rra<PBMlwcXT>PO^7K&Afkll@SPM
zLG<ePN%JM2&I=~XT|xxoW+A3R7eC}%^$TIzv^tGuwKKP%-_5pglfa%MO<<J9>&LjZ
zKim&Zm<A&@Pd)oMSRd<~L|xk{L!628&XD(aQ)DYyyNN1CJTaEd8xvOW0+}u-k_ly%
z#_fe1-0{&$LeQlc<hh7`>J`oDEjxi%;K`H01Z*kWD!W`+!-${NcnW$RwOBf~6s>C(
z*@jHN33_9GX-&T4$Rdysr*9A1C60iPJ3|WM0zcDN3{@?y4?2*`lwS*lr8UA@JrYPl
zdB5{Q@){<8vl~)By3#bS<d^d7yW0NAANrEq2=Zh@4sO1lgfZjT^nb3Tzk--@Y6!o!
z=_gG`Sy__GxD3d7n>acWB<!RP$xXKWVg3{Tt*0Z#ufvj+WsCl*Ol42PE@GYXW6Gvy
zWq%QR?&oQ&A?1#yZW+rE8!Tnk41=g0G#TNs65gGHmiWqhGtG|n9(e)>dy_N;HJNhP
zlwpm4qqG`_;jwDh&>=nm{+$u=y#4OnS;p!a+VziNiG(prgyTuo3~tqQWb!a#6UPx8
z%Y*L%HISLXCcVMxy=aOAk9S{3YthE_cz&jgIyb*;_T+tlo3-t<ETwMBReWN)mbCel
z7jkT=aIxgM%V-{w&&FBst@O`dK;+Jxp7_`LDazr9w;s=<2$)g<VOdj3OGWRGe+N(n
z4tE8@(0fh)xhaGeuH)E40g!S&DAm{Rwb9CeCzv%uNS*z`GDIiW1qQ%O1hE&RY-f}2
zPzI+LQj;=}x=v7_ztEeh_`o0}(31Q1BLyO28)PHbe}Xcs8vShRiSIsA$|0>b(+<y%
zgVX-YFV1x~80*y?adC|cZ4|<6*)=2Kan$qN|CKS{5MIsc!2=kL!&E9)Ksc}es$22c
zUOD&U?w~FW_*;{TCh^CjRZV^3`25ipzw@=h&^>(_Qg3cqpyi58gyLS-$KaTi2HLS_
zh8gD68HEh&2Dh@H`DWmuGAwMz<(Pb;<3AI97X!EuB{U-X3rO1QgGoRbE*b`?<@w&>
zAP<aVJU{1L?6t%XjmHF{GSQ4bfhpgr1hsE>w|a$nYGa~%biiNNDKjXqD6(u;2@@l(
z<aY&BgqMstxhPu>*_s5E3}eRZDfdeHkWOse&HpHsbH+-kL;Ni4s}=P%1^a!wRsD$p
zR)}n?8RI%G2=#tBHCT0L?=T?!<6c~@K7zWPY8@f2Dq+gfp`rjqG>(3FZIpU<9a&hs
zgH?yA5Iyw{d{MOu-1k?_fJu6rPTU?vKBVoEdu0U8d%tD_EBu{M-T6*~>U{@=EH)uV
zt3Som?zuK`4E4jNzVkmYJ$oY+=sN^C^0)0f*Hbe{1U1wEmspGwgCe$>S1eU~9@3mJ
z;Sat!;l(T2K-<rAnTuq(=<m_Xggl)Faq;~9i5Y|51FDe5z(#(m3o=NT4YQ+nI<<{b
z2niP*zo-3nhOaA^$&PJBKaRWROg?R;l!I{b7(Id+*8_L~n*H{qAu!#AvjOksNynl+
zNpZ<S8f?2fo=7@y+>I$)UCdhv!gvHdYDu-1j0N7=A$EKdn$-n*E&txnA+2b@-!J*1
zw8JR#)w}GX67Kud7bP?pqWAa2k$lD}9h^9ikLop<oV8fcM)Mi?L=>O!dfTD&1}CEY
zF2?@^_hqLy<xa79K)!5Z*N$g4_|8!I#Hg3^9#;M-II?s(^!KhS?3Q7#F$#tCQF9*_
zQ&sYKz|p3_nR{p@i#A&hH1nx#c=_CY6NPGho7U<4mv`;zcKqUUH_IV9+AQL;z+wzb
zd8{7&l`P3pj%AVDD%;(gp71iFU9;PfX7WMFlSQ11f{(GIvmb?Z_PfoUs4TmILpjOw
ze!3i?S-%DG{(!Bk=VPupGk+{+<lDZ+X7(B@_G=d>25d{=yH{IvNdA(2JuF|$aQsS&
zZ`inIH3Qp866@u6D~zYq>Hxfi`(E7LMJUwoRcUYMjjxws_+RP&>p(>(&_NRRsWM@x
z-+vzg{(boGqyN1xa*ui4?$bXD2JZ3M>rE8^m$=}rJ{3Pm*&g74#z`V7U#@l_ofc1_
zT$GZ7(p(TV0m5%OSwmbOFOp>_+3S%3bDy3%ipoTPK4`F%@?!rYD5s!SgZs(H|9r%c
zOc^45gOtusJV&gUFZe19^WbH7SUk?sXPil2Xp1ywqTq@{izLL$aS%5jJK^59Wq)aU
zPvcxLKUX>M0(q_?5DT;&S|&`g9f7e~rViAdBX7$Mq#x%3TZ|i-iTiaU_G+1z+5w>}
z&GsE){?(cd)HkT6Xzrj$d@i_38<^wSC9p?x^FkAlRf1oHsApgr#P!Sp-1c^W0!Mq`
zGO;Pmzv9BbG9tmMIjKgmtsydH2J}N9uZa_cdB>(a@A&5v$f$*Fs<K@`1+=mJS?4dn
z6Dp}&IyC+)KF4>W2{|%P!p0z2@icd<69aRBJUj9wEoD2U!la7)j%B%-OIb)_aVWfo
z`D+-}(o&p?hXf@SW4P@djBWj97!m?ecMK9hogeZ#6f7-lM71salDv*oo6-}pdLiO>
zoGYDMiY|u2z^xdEXwnLlI274yGF0BHl8rWp(H%YuF#Vi*I^?jBvcKJYa0tQkRMJQ{
zIbwG!H+ABRo<W<c)@$;jtiIYBR$D+oy1Stf(7;8y#Z}t3#0|;`RT$ig{_Vz%fJWWV
z4;O*N7>nPtM_%(#^UILv43Ey7jJ&o@tY8Jv*Vj{v@h9Dx!HtCtar%}Q5$Y&Y4x6l}
zdY(tkdm|C9xV)>MP@LmA=}?tMvNLU%XwJRrY8>o#$^J`@9m?(Q=7B1hG(Og|O~(G^
z>PdrBe{^LxR|;>8_!Vb>d9qoz1&QeQ^04N?iRR5}w=H^pb|SYNZ&dF53B_N4TA$oF
zB=X$&Wo7r+K9Q?k#oP(3&DaJ~HEW$$L$@YSQ|?m+uyRGbkpyAY^}xRAgJRINbaD9?
zxZIc29A!S}z*_)Q{8B*kUjX|Qh$kW;&>*gRh&E?T-$M(5+f{h--j8P7R_u>Yg3T7&
z@%qW?udZpDq1M9`Z^&)k%dYyPc9_?9zdlT3dYBjz{CqN0VHL3+HwT<211q`(eqA`8
z`wQ@RAOc36zP_h%w=27eeT;$JPw=b@3^Kj!TRsQX;R_U4SqyY0`5$F&a`|1!Yd7)Q
zGk3==4mx^iO~J^e(xd2#z8=YB>E&X*CBS^I!IZUB)pmXPcd?OtI@4`O#)8p8PbbkM
zBObKZ+;d^tRdH3%XibGo`M<t$>WkooO?YF$y1Uv^CTE{M3JJ_Kt!g-El7?cp91!M-
zMC4i)ZaP70XCm_2#w<5fI<3d63651hynP-fi#Jq`$}vWzdHPfb>`F}NU3mXzu7;VW
zv-_&JEgUUH&`py>wt|azMhE;ubT~v??700H4p2BB;MOE=fxyaCNjMl|m~{&<&jE}P
zZbpm^$NE}M5&k$WZWPUpN&v2$X*sgo*xC7dpIf#?5C!S+CY_<Z!J{oGWJsMdfk7sC
zXD{@=SWnA+F==!Fx}~FW{sq8U0OLgtZ6F28D+-w2vnVC#q8kU_64gb~np5Sw3#=XV
z2&luh7JOZ#cZR7FekUy0ooa3V1AhXD3^@FY3iUsGq1JV1S{fsg^||z2uI_chU%)4T
zU%1V*DVVj#deRy1<?G|3=8|d72RE)w5JCgKTiI``mN#N%Lh952XN7Obv@K%V!bbqa
z=fhjFcy{*tofy?AIf;bTL=X+za*_4wT;I+rDU>s?M$ZEG%sV@v`J@gSImZQO;H&BT
z{i#X(4QPlEw&x0H)PYVN>B#0VhpP!&i>Kz~MyFuWG3O-bbUN`&LBZ7&Pfa(qQTFl@
zOGR2?G4m1}%|Q#Wf8j*l#|4iQ4S3(#y9oe8j1jMziH(=lt2ks61|*n4DaV#b1lE&X
z>hiw$<=mJy&rrKyq%B4rBfrw{3c2pLUDuh>?P;2CIxUbw<j1B|#~uPT#Oc=I;;+`?
z;;!#Cb-)rFxXg^tt4h9!;zs$>lc~PhkhN)DaU|_{IC#s21CjJkEzQmhN$5JeF{9rS
zn-+@?wyDpG>#TDVL<eo!Y^M0@cxDPJy`ST@*WB!<(h!{3vC-H0xs@bzX~%PAMU}?|
z7mLq6hG<Q9y>DrVpl99_<tS|!IrcwsWEu@{LGqcGQjPr42Zl4uNI}WnlRVnU=EEEb
z=kKreR@yjIRBFeq*=6c?q->>%99LjBWj`@ER>GPg(#7wrI%|`8wG`$TmUKA6{S1bP
z@e{)x89aoo<rs3abahIl*JU_mz)ulF(9_@P-}mY)3M#oUPL{pwy3QNz2u7Kj*06E>
zB)F*}-2F_Zpw+)xxX-v!29=(eG8Lu}Ta%|a=CG;3)tNFKN|o!HuMt>F!O`h%bRMSS
zX3Df9ZJ}&9!Qc5)t6thv)_U8u#vwXfqw`reeP{zC^#O9TFq5{Py}FVQIrNvbIuXBP
zem$xKp6MLxliIPp>^N@}B{|<O;Nn-5Vxv`)KlBk&lQC1I&{agi*vD*~11t<R<nMBL
zrs0FO^AqSp*jqrqvY|MI1jt|)dDm%}EJtc>pW<vlwlMR^Xv}ZPDVf!E(GQO`w^XNN
z_jJIkE`As5K3ByYR%rtYT<%Med3*+a+~bAR7-Fo6DBD7h0=xXOtJlr~&_?kzIueDt
zC0-@eo^3mk9w!0bGGmxWsYYy&1B`#TAxAf5hQtqM3)RcuGP2#clFR&qkh6ECCg|oA
zb`VW5I?cn9H{GUt9S4Ot&tk2xLcPb?(+GcB!q3_6!3ESGz2&Kf?Ks}8X1Y(UWbA@s
z2{q1!e*r)3I6pmhB~TGnni(GJxM>#o#N!OA4%@^=Q&aWQU0#Ho3(42$8}{EaG_3Tr
zK1D5;2IojP@lXozEjdHM49^NV#&7tm6T^A0d)u!KZRU828RT0^7-~nD;s^#&Q(Hy(
zwa8fcCneg9-?i}9aSLWb$W!ck_*NM4TLJqBvH0*JMjE}G--S6x#-||*eiU_H@sv)O
z2G;u;aX-{%`4Q|Re7(KVWG5-U+$Z`JJN@bO@{=t$8uCZg(Nm-XRY?b<S$b2yBO`uO
ziy6eg)-!|+m>}CSoQUg$8ONMQAy=c00pd^hKdM$)Uz$(~>%gCBf5cXHw;vQLq>D^a
z+U$nse0dGObz&kd&~e&;J5~;mK}t{j8PVFV-+-IcrXOXllbB)3BZfi!8@|a4O_~@(
z@0^wEyQ`HM=@OQ{4aHZnKgGJT@7CbkV9XhnKfS>$u|iKEsj%7s6rF4lONWcX?eFh~
z%BRs2{w4n3dxbq!hdccLol(9d-Bo98gY!<~kuRkD?JATz8IpPNM4Gc<*mW8jVITL>
zfj45jMoVQ10;bs#=$tBS-;3a+g{)5G1YWtqhD@B5ijp6&i9Zg&^Kf0pIzub7wk03u
z2t1~;GfA5b6SEsmL`P3%t2dkpiZJC)Xyx%=p=K3Rep(;bW>5jb^!JCzazUhQ)}}}t
z^l;!Z7#_MNEYQoe_0&5KreGn$8fl7A@wXj;A7C<?hXui*rCnGIBwh%bdz@LUC_=oD
z<L81a-ztgsnD;H*Gm5kMMsTna_?!THAQ;p1kk3gp(o2usYeyW@3dB!txH7ulFE*qN
zLaCp@cLn>>$_1`BFER!Zj;>dUTj#bUK?B#EPRw_{S}B!y)oTa2UCgV$6l&NoT{=h<
zZNH>9Mv18E&%4A6<Cdq#+8#6KXt6(s3fnBQpA%5w96TO5{mO9(C^NV`2ze`NTO1Wk
zf$+hX)Xst>|Mo`UqYTc6iU+woXtgtXn1<`BMIE&An@n?pM>@%|%x}xbr+G~q>Ga2(
zUtI{ygc<Z*B%)v)&?;dbyMYyg3%wMBYR|;Ci{eagAvK1^yfbQ2;S-nRTV~duV2_`B
z!|){61|8x2lC&(Rv!%ui_Tw{9omxj+3qB8_g(tpW1sr@w%?;f9;@eW1>6Lxrb-kpT
z)f~>jc@VNFUcA}*Uxm4k+D)I=s|(qMBbR@SId^^IHJFwCfDt5Zl&+IM_VaYaEJ_{)
z!Ebao5v0~7&Kliak4#q%40rwH`fb0eTJO71dHjw^jbQ#5i#EtZC47#*N?yHX4&Ch@
z!#Jb*1fH$%PSKm=etDB5dA|+*z}{!Ulc0);U3CnSG}?%f6AAACPZDFvf28@r1U3jr
zZUIG<4*pHjyO^)o{bP$}<8nWbbZng=9k$k!kF4vg%OaZLI%ZeIKH7DWclAf!==%BE
z5zqQg<<=PQIQ`3nqUO;%YTrN=2YwDeFbz42wCs$m<v6#D3HbS{_O=MkoJR-4DtVDG
z(NWqR(YIr50p_c9opxQkd2`Go&6XFb3!QoDt{#WUgZkg7mofUO{q5n%^<`WE7V{{+
zIk)U&{ZDaUE=gN{`x<7lhw#PoYFkwpeo9YU<otle&E#st@8o)*==i9pt(#%<mPUGN
zT4RS;yF4hdb9u_}G=dJxu;Qe(Cva=8K4DgueFUJ5OKGrRK0#HgCWy+2ndg|wFhU*d
znj9CWty;Y{@%{&sk_OHs@;oFxfCh%6zz*1S{U-^&!b#m`HhqOaXhQa8Vt*$g7uQUn
z*^yz|&KkkE*~6NpDpJT)K)I9Ge=Wxb`1c;X`tt75J^4>P-#QJez}-F-KAdhQm7LuH
zub2BMM3Gm|0}q-4#1h+J!sIE-=Wqa-QwMC@JMn-j@;|*wMF2VRz2%?6hZ`7K4kl$!
zI(2RX=@L})FC`NXWAwQ7NY~-WB(3PBM*;}nevC8C@ft)#q=skmMYS&>ISSTNwhfP-
z;tIl>Ck&tuD#xy3pAA8#1HtDYIoNDop1y@v>XVv=p2&ExEME$2$JP?R*eqGNi=-pM
zdIzGE>EB@!z+h=vK%)$P4Q+=hTmeuoSqfRW3Q1T>ByA^jC22yD&vNE3U;~^>)^V>b
z!9yeQ+losw>HWZm8!k97OoTwH{0h}}PKE8QoaLm!06Nf_=!Xe^p1d(vr$layv78R4
zofk=b@I*6<G%_;whd~>kZo6uh68EU-QNy0<X=XpolC_<&-pE5GhHo^+uhUIAFtqZ!
z%jE&D_-Qi7cE&ru!9kIhGYd}VgSO-wT|{a81^LG3K?1(F!G*?pYBi|UTQWB%;<;&g
zIahpDc06OZ?|Lm^>2@4pONcT$x;}zsZFgal-i7PW0Yw2w1K8oJcGepnEXz8cHML`K
z*=^yl;+u?&TidK$oaxpZc~vzLp~4;_INw@QvMAnQBzGv3GU`$2r#95?JfQ%h`$T=m
z72DTo+ze<Fg5)tE@eRLf=FBYT6P1YJs(&;iI8y67UrO8(7LGRrpC{z?>@be4^EP3L
zg8Oh*2RAUfb&)CHR4)n`6WI>k7d-cbP9*U@OgVb6Va#pxXv;Re<+`g8aS4V=tTjq5
zN(zZs^b|Ursai^Ez3};}GCzvNf?9@|blQy_C4CxZ3b{LJ^7}2~G5DVk!{H)i1wMFl
zgC)JzayXi6IJDTkZHXdwpXPMgx8Z1fq)KjTOq&I$CmiIKa%EAd09H7YYb*pD=<KP6
zSytzHExZC?oS#PxIENhc+=0;du&oy9w-zlQm3z$n)W3ilC$Ls(`OfE};|Ry*2u<hV
z$JzSPY4}Bul%#>Efsd*JJT`3L%ZVI+S(wIx*{0=dNzR@*0%`h$wrgkP58e9peb+Uu
zgiSpt8~RO+K^r}Ozl(s$oPlf1m3hqyGxIy#ZX_J0k~@c+I<xn+WEY&iY1qyai2nR_
zI5~u<09Dn8Sduk`pX9Da5LP*!8@9+_KAS<c+g%yjadD?d8X_#mWt++&KX`u1dA5^=
zg1iQxLS+N*WM)a%P&-)P^}{<+KQ1<f88j>DZQKl5w_b1$Zu8UhO#OiLS<GxBUjWDL
zhElsaW*3^d4@+hD3y|lHUhVogxW#e>NINEc6rqtTAnj|O-)2OvdYbpyT)ha|uR~=2
zVM74E_>{mLi?$z1El-D9+N{Q=pfwYyxEshiER>mNd#!Bt-h7i7PYYcy3BUtbQ0NVO
zya5E7B|e-7UKGCekS*=}<D)o$RZw6|H<(Pa;NTV3*gcu~-?|)P^$Rocy%Fn+fD-U>
z6!7q7imsh_;nc4eYQ;$z`C@u=Vkm|Na9yeRKsK4s{5H<f$E~DpPxSo=34MaU_cI-)
zo|x<GHxKg1Is64Ouw1vZD7)xE-GTh1uL{Ku2yn%j^JZ|PfYZJY7PT($PDICA&F-ot
z7axB@%iiPMK5E*{JheV*gb<;<b2X{>RX6U9Lw3tWn@R<7kxVcog?m<13$$4M3ut+m
zU;40BuAfj9DBv}B7Wv%tV!bEceiuy<7B$FWyO-cm2(rU5r~jx1eO&lJ<*F2U{>l{u
zX(l^leT8a$xz-HY$hYyiD}_xL4bbSHg7CEh^#*(JyHj6|6$(JpoMmP$s<6i)Z#dv$
z(D_Vm+dLwW$wA@8Rg2;~{;7AJE1jId@#E1$!MQ>vv(jULNs##m{$oV}@fL~r$#Z&*
zSoJg?nsH~$-D<j9dq@HY^B`Mpgp%@}a{QYIE0?`Kxj8%xpH-cQob*7i)4IxP^O*6~
zLij~UX|fI)>Oq{*l!IIx<Yaq#XaRn(3r4U7k<2RfCvYl(=9)W~&Pr+=JMmP%+OFB;
zIU8Jc8?%~OAJ-}+pM43|nocK%dw`_G+2>D?uZk~qz?&iZK?n4`33d)_c8>IT@v!Ys
z4AXw6sADDL9=z{I-soI68arDtXYZiHkN9$uurQD<#jsY_v7jeZX}4DSDZCqK^$wq<
zn-yn<iBgcf{MarQz2~IB$PBo?R8``4+Jlpc)Z{+w@~BqvRI@p*UE(?bs;|5@E^ote
zSpo8%VTLAf-hlW1WS3tguZvANWBmoV!R1tYyc+d17d}{)HR%(M>gB_;cQmW?5P>ah
zaSz6*6Fb?p<qSFrG`NL?*vrUoXi66X=gErec<L^7kJMm6Clmf|0$Ozy2N%2%sv(Gj
z_A^JEp?*I-a!eHcP058ukm>-fv28ITHk!+|5;i$SH$iLakwJH%*ACbfvUd8j2f5!v
z;Bmj%<xPsrcGuMdi!6;DM`>J{LV~NJ`*58=wUR^a9%hOPNoh=sY<_2ha7Das#Z`Hc
zjsz{FbP`EXT7`w_k|VVn&1XSA8wZ4l?p3FaqxZ2i&1j!3R#{#9r_aqJO;IMM-zcB?
zl&YrCm+iML>xSp%iqUtfoO!5M6U-H>lFO)-CmR{RpNpln>MYC85A}z4jc(4}Nk24P
zpMNuO8NxYjC`#a#{3rc0^F}WLAv6v5;!ieBR5(zDC&DD5b?W~m%*@vP8j#=m{9gjY
zzr%>7^?wN@(*GrnOvC+m*8~2KE!PR)YI>rLbVuX&hctN%iinGn(X7T5<FFxW>_Rg5
z!4*k~8?2Dd;AszR1CRQ!>k+OG%5vuHS^U3ML?2a96s*8JpCLgEz5F;xE?#E}GQ7ld
z(+6O9`vA^iU^pOeKiI%0ufS(6dc~S<wH|mlrUt9@lm+#e0|UOLT)ccK?GMcXdjUQC
zon=tBrD6u;h>4+9e*q4#kc(Z)_GEq!lWkh-S0+9;o88X?ww&qtST^%X-3m7YCxawc
zJuF)teiisaK+zA%(^^sXh1mW!DahxuKRD>%_2yoM3SW+EH@)WI!X9ULX2afuO0g@T
zy_mfU>;I|Am!j1l94*XEzK~5BlLKz+YGRW$`~|G4KUay?Dk(9Xiq^>ZBl;{WzLaKW
z9w%DSzEWo<)ANP3ZN-4wM-Q{fHV(>h=G1U2$2Ombe@SPQ_9!JxT04T%6U#6fDsh);
zqQ8GBP_k>njMQ{c6IM<E^H1lhElr1qF%HsiIuM7ziEK9|5Yk^|{{YFdxC*Le`dO%0
zc_NFXPszjU6{V4K#`$QoP+ZH_?FsvF?8fw|OO5U@@egyiz`NC2*V=6s&(Mz9;h7os
z2g`QBZ-8iSER+1-YK9DDkQ2w`x7#0}+DYUZl}&pjvt*JzYS<-|@;*FQm(4qNr?n~g
zpF(SieY<CRtg3F9mirqyWOhDD<CKk(3K9Oui2Bnox1UL`CL@+4UCWRdMKu5^Yzyw_
z19{N8hGeRSaLXU|2{f3LpYc-rm}s3HnigIlkE9$VcEyY=d}H3G6orw9<>dr(6c%Se
ztjH+>2Qir{h8gGJDFPe}Ddy=H1kCcdzW}fKobtzrN5bvG!6~`S{zUE$vR+sLJQenU
zm{2KlR+)THk)Gu$?pB1n6BCtSI}J;oM5{hTg0O>BHwpw);H?aP^gn8UqWe{TqN4>1
zC#Srgl%>cy2VDX+z=ZjiZWT&;KBP*1LD(>px<oMl=3W_P?}c6i+?1Y)T>r-b+&@fh
z@WOaGo$Ln{$(()-_Pr{h1LC+Aox_KGI7en6Yq~rB3b3P3;m<DN+3cuo?ww=1zB;;@
zaO{t+BO-)0c!Y+NCAUc>UUBci1>t>zJu#ozoz}Z;+h3`v53X@Q#?j%d;qO;r1`FXY
zFpq}xUqDQW!_YG{^7V!6m7$62eH+;|gzV`i@V<&?EpSRF@B+4n->v}skIo4CKYGGJ
zF9obO@B4pOQm#G~GHRLc*{SDN5}QOC_$;af9D?S{xGRPf3K*h3@?+1*WN>~(%H@Q+
zDMEE&<NscfC^C<oR6<lG`dP@Wj1iTOkHlsaoxeyM1_{#?X+~RxH5Yc4Z37={TVxF7
zAyhBqgbvf=5y_GP%3CT%fz=}IrHH!qZhKRA)q)<v$J9Ki8}oZWRW%x6<sLp#9~G48
z7T_T`Q4iAaRDPrt)4^W?!wL|A1jxEX2_a^$pDtdZnt?rv<U}#XAU603`nw{u^{;15
zrglas@au;OV@ZFKT!gE1(FS{?XCdyIBj`(H@sd_=L}J}$Q<@kfB%!4UypXpTNz?49
zK6{aOippFxf?gv#R5?(N=&Nmh-?Btw+WZDo;W<g^Puj@s!&R9&Zi%CO0X6SEw$;>A
zUONeQz-Pwvw)ndtn9Yrqg7plKo0FOBOmT&5#E!3Sm6IT0yyUIzFK?e^X7z8@s9>|A
zNNfndo__uv6FpzCV+Ummqad(nkNz_0fBd!>tCkP8QRf@I^H<8^c0jFeCBT+0K*BFr
z=rVs$%~7`jdYUe*EV`<3;9Gce{dk*8o#&&KbVsC)ZMb5cY0KqB#V>qx#o}OWhQoJu
zqWxC*<GH6=lLw1UfLBi_K%gprVh(3vqI{@QAJN0Yf<Vh?;lRw>Le=G#5qI7wKXOIN
zVL_#VZ-=INCKx9-J~69OyTGB5-+9ostQ4VXnI2y73iyUmrqKpUT>w}Vlk>097wQj;
zf~>6&pWKIG6(`^gSt2!^tF(3#5xfk8$saR$jAp-GQKVSzWo=1=^pc{tG2&okXKC<g
zoLh(G#2X1wXhjPvl=jdrx-a0MMA5h6TE7kohV3=no^2;oPBl2)i!8K>r+4;B;Und~
z{9Y)2yE&V@#ISo?hRE)r+fJ;RwZ?5RRXKh1)JE1>l5<68z?k<@uk4G-$7r)ulHgjc
ztv)XqXbq#I`_>?%k9(zth1aAVeARA@fW}TChyrznk^W3~5fj@!L6RL49)|pMOq>4D
zhC|If!JdjoeyL?|sk0VI%zkpDr&3WluV${1$YmeTPRWrUM!MiD-xghJI2Lm_s1YyA
zqU~lO$3njboq9V&?BjhE!N1>n5*N)@I)S7l{eJ7#jQhI@+=EX5_cCk}Q{D5Ql;-M7
zPFv$Q7-yQWL<39prK<1xhuruyutodTOIG?8_F$ZPO9yyeFn-wjM7B~5YYkDuJYSlD
z2VrS}|8W7BmFvHr3i$tUW_s|BZ-e{Rgfz-XE5xzy90u>D853k!_21N+-}enO7%uWz
zRH_n-Nv<a^2+8>woU9?U=HpRe>0fg6Sy(AxN00IANJoHroX;Gqdngg&FnmW*Yvl6g
z+d1mH6Rr7B(RWh;wh|)Ezecc~lWcb40^<=-1pe4c3JWe^$FBolg|)e19Z0=qIG>@H
zuLF2~va)Gf$X8X1Yf`onTDh;ED+@2QydQkf8O+r&JB|ieIEM*JDYTE29Uh2mP-dIA
zrw<@4HN9JpsIyeGT$WiuH(ZAS7hGsd&9HHdPIpzQzj@`p>4XX^^2H#L(49>HEKf>!
z&;_;i_W-@5CITsaAbE;xZFa=YcJ&phS9h|M!gXw&57nrl)0+NDB#?m!mu{Fk;Lpcf
zuPC6P|0BtRyRubu==Jjrn+p=>FfvG@BEmDNgiE|)!dEA&&c=onn4T!jk;rUGkEQ%$
zFP;ApHO5Z2iKvWVcPDhVeU19%;-umCLW&+cMO%LDbK8Tiegy3eSYBt(?V<OsjlX8?
zlNrt@m)9Yg+k=xFM<}wDVAqU$^P|1q1tyD$isllTxR(7Unsbt%B?yG$*v6ynbHhX7
z-!M0wBIR#mbpA3LQ-4bobH?}69a=6F1Wfa?L~JnpU@-EjQ{Zsm-0tu%;P@`kH|E5x
z*3$>hNZ86JZVLB&SHOC5M@|2L7?NnVM|8JBItCjqXy#QQ72+UNOm!d(ME$xpBgmL}
zy4IOLM6Ho~JBe)DYEsZfLOxo5%3r1b83IR}NUixjM_8WeZ8!vO^Gz(ZMH}4^O_{=q
zARg0Sz<gn41(fWcTRQQKG#nUCYU9Gm>2HgCq(_K;rlJU|l}<p?Bea5uk@r-2jR$MV
zDJmTKqkGfjzYMwZtl$bYtcFhCfn|0cr05C=eiSzQ=Cjv;SR{Bvr5){nnys=$EaRG;
z@VY;=rPSlI$B#TSRW^m}$A_2IbEdt59c>xT-PEo=NVlE4s!J%l)hNckBkbGQF1!uQ
z+~W#L5S>Yo)UZMZvr*E4bHcYon$Ho?@I5Br+DR(We0r2g3F<|f;ffudZ)!88MaMgE
zWU@)Cw@R1GcQ#s@6t?qP2q#pV^A-ka(4`qcG|Ql8AlU7(wE}|oZvP8&z-Q{ccTB#q
z0DS}#zaIZX*!|s@*&p$<D%)MEHF&2ZhUXpA3<XL}U>J|L@bN}vrM^wF^rsyegv7m~
z1lsiMdZjlNv_-V#w&T+o?^eARP-Wmbyn~6czxINZP@X0ESxj8<Wxtt`e`5Q64qr*y
z3@352@^yaYdztgsMwDE(cF{d?#Lf5)@&wKT+c{fYJ1$kCEWZ)mWIV4ij=1{AVA$OB
z#f&jR8Al4t_D}j=Kh07T_58uVw6$go^ko`<GDJ$NCyn}Giek<D%c<(Ji3!}*?}e2^
zWd8zSGg9wg=dzABI7$i_)w9n9$TB4RFCd+u1MDPr!DQ8j0hOq{GlK=9#a~Q5YFKO6
zW0mrTf(QA_pso(iGum1PMJyKU>ko2P<=nFp%f#wp=3{EC+!D=l9L}Liq_Fu;^w4s2
zMU?kX=nhnsIsp&P8XPZ?k5HK0;rgNw3{MW<qtOGK2cm}X5$$MX$fr0_Rkvr9q@jzx
z93k+$a5O2h#-Z&z`Cje8X;}$o_X<E@wDm-XR+|G|0z#UjI1JQwAeB)V>-!mGT{4yn
zS(7<1DGlw3hCCtSKQws~YsW}f;}W2swQPQn(@UrzAvE|)ntp1%()^r7mg!f@FPNw`
zqXHQ&$Bqs*m4XT>Y|U^+)ep?&%Bymkz4%mCWTA=CvO4c@VXPYCoEo>0X`FrIotCy{
zu^wO<U)nkWpf<DK^(^mNsW8#Z96J#7xAXQzMmT#3lzGTgaK96iYPsHa4(%%J$|*pK
zGT^A)N(f>>bak28f5`L5Z#k96uG~vm&5l$}uqDb2*I?Vgzadvk`}&-DY=bD`AQ>Z>
zL6(&Yt@gN#L(|TubM+lrUpX-KnC>G*;_#V2=Oja@Ro+oY9NnRbH><Yx0=l*e%u#E5
zn3iyuy*tzUT=6u1L#JQaP@d8t74{|VOT&qwXXHbS_&K6ilg&i4vo_?6@FR^5p`;;L
zuhQCYPtPOQ+=XZHWTVkf5q?d+JZk7NB~T7haL4IWjeEM2!7F0Ds`W|1JorMBrmi?X
zT(xs8^e$fXi}4vE?FCAMm}QhK=1v;^$jO9WxN{eF=>&a`y|QB$<vUed;vWI&!@X1s
zF?Xu_i^eCpCz{6PJ9=rr*`h4J`3S94Fe7Y&x6e(5F<9-x29&sb*9pXLZ@|j|-e2EL
z`f8`Pp|I%I=PSK{)ic`1Ax+-GEef-l-PJLq3Y-j8s;kA|?;xs*e_S$c*u~$?(IhP?
zX}}sUuIGQ$tcd0PljjSe*~{8q6VRwC1Nyu*RS1SOSq52cgv6B7d6!yF^}v5Hndg*1
z;VVNxtwEG0{o@7eS6b^2FB2ej=3_Z+ut=)WNY`3RzkY&={i?^EwUYPeeUb%pf^%A9
zIN0q2A;2yDoLx`#?$-K`I6d<=#MdZe>7vG8RggGb1Rf#oy*!??_n6w!_5L&xb2UOk
zvj`zN-Al_xGza8)E`kSO>yu6~+rI!)U5gC9vxDp0(e!Q02a>M@C{!0dw>9zTfjWQO
z2aBt?wqLp00O3-Rt{=W!{W;5PpM1+COO?PN<43-^nq<p4;?N=2k${u(Nvvj^r>b2&
ze=P&K<01yo<i7PZmV2^i*Zxtdvn$~;WYF*Ta6?dWQ}DRS0qF)b86!xjhie5qE;k-Q
zljL0^_aMo64bw{QvheLpPa}>2kz1M=sg{!t?qaE(cs2_?REAGK^X<N%92xQ?+Cl^r
z^Gm}CilR+FA;I+ziWa>~`<HXfpzoLR10HMN$$#24L>iB@vV0-soYX<E9&F5Uq_XXY
ztU<29L+|};Ud9fp*+!_<KzH6{sEqaX-EOLC<$hB~Bg@f5D@?qQ@m54N##*5re4?yA
zg;R`{M#@(SZ&uEQ;!@DTRULOZ8%*t{=lSeI!=V1!p~xkOciOjz0phs0xSr$c4~^5*
zzvxnXGvtcc66fGbDHXQE<W-TYGYf0XBn|@wqH$?&H!_d{7BLQF+|QWM9d{}}MG1+6
z0;gsJ0JDDqKO{>N+QeZJ$o?jhQ@qv<i|R;)9qvl2*YYdy2Q0E5nJsAI`u`*@if|pU
zW(_%P*1b<f-$og{@ISb4us{&!8PS;a3w|0<#ZJ+h+NeeDNwcv&nX0QJfz~kw#&;NN
z5Ew4rF&SP*;m3XyH9x`qE|V?zXCb~{>-}Buga3Pf1O{INAzGG9V~@?^^U;9S_o#{P
z<?C&~Wj`<tiUqicGA_-oDOJ%RyR;P@$!QwGpn%{hnr81TNM{_Aq`Ba{NdjMW{o0Q?
zCvE2B6xo_`Rp#yl3&otTS)=EwkZj73<$jTcoV3F3TfH0?Q1P3gr=JN7&?a*LBmXBy
z$NgwG7AX;vjl6~K%I`b)6B@bc-SK8bI?+qnX$L4h{>OV!uRKC+<Qk@>CGyJR4Nvu4
z%x!qX#M;J1Z<uq1C+&u%KPkqGaWL>FAR3UCCr5|3(2w)FSJcHaW#|wHR##dl@(5&>
zB7oUJaL)6yYuSgxxT;Bj1FlpNZjGtut1<<19X0&9)V-s5h>MjZcCC(Ho&Rgm3nR+2
zbXBVm<TT*rwZ-UjKR6=%KldTC?&Z8|MaXq>!4dci<oLEkp{;ndv(fe%p8T=$i7clj
zq{KUywIK^9O$ZN_tM#VZOy|m7Hv7fbqYaOXg?vu}3AE}jdtBo$awncf>x@i{lgl2f
zWaVHp+DZ_&xlx<U^!qA?T`KR}3NB{2gtV6qY^=f4RgH7XpO(ScEZAl_Aj;%Bn>23Y
zI6`ShIyh|vSFJC`dPoz3Uh}_`4QZm^JVX!qlq$3e>no{bs0JUY3Hm@rNDW>o3EqoM
zKw2cSacMb55^WEkdz4>&FlHOTmrb3S{h?zpvH3yK3*8kI_yEI^4NnXG=id`q)nS=N
z=uHv)|6>K<f7yciZre6_SbA}$K=gv;&;x9@hTo0-IG|@arqzQUP&)kreO%NKIUEvF
ziB86j6oPy@ubJLN!KEB|Ry_Jlk8KXgTYE4Ct|x9!SlQ~OrMxK)5MxuM8H?eENPRcq
zE4>=tO=v+|z#43nrtAMnnqJm4TGie!e5gR!QHh-mSXf$RV3^~yP9syIA1mly_%K9f
z)Er?FBmV(#B4Z~Gn%UEBJ}fl~BPJWHA+^MjAt)|~2{nd(+hm3^1vK~91;>E>MD#>m
zd)Z-{edf9sx7YLmtDI4(C|<Smn@S|O2BkIS7IvNUZerlJJO#;K?cmspq;=~@YV3Si
zC9>CM`l?0L5tiN!)Aw!GJYaAE#(%r2tQqOACf-DsB);sw3Bp~uT{(D>xRgDudCzhF
ztZyXwPLhg|MZRL#7!2Sl&nEd%uCifD2B}?<qX{SWB%YT4Z@j%_P@G@8?%7Cy5F`oi
z?h-t>ySp@Q!L@M>kPztL?rse<?iPZ(H0}g<f;<1|cfWJa-XG3PO`VxN-xg0Tx{9ji
ze%8A0>$-mQheoDgVOGy<Nd=nB(Oe|w^EMfP@!z22a@%pb#+9O`B5U*}uZ*MF+dLxD
zi%3hx-Y}!iRcB~D83`H7Ot-4$w}WcatJjWLB~%7M_u?zka+pFhf`KfYhnSu{B|w8r
z0bEzr2pIQVnd&KMH3CZQzSr7c%l_S_Q|2@)iJE#9mbDMH94E1DBW!89<AIL8d8>7a
zEyC=)doGd^s<=$W=X}$zg2n>N*xd_W$S+VS*l~(c5v3okwY<DR%;vQ^^XU1^ZfNLx
zr2O?%B=Bgt4-!Inz-i*e{7Y!Zkh)y;_2|=5SgRbQs#%dLNbS}kuE(EPhcv9E;^*vU
zU(dyq*-z0~VetloeUzqs;d^!e@7>O8OpypH<DypPf?P?J0#4lm<6KHAcNUJ<DlNqm
zwlz>{nICrBxI0lO<I1HVpof4-QD`3g?5G)(oaHU_NN(p99Pn_%Yw#GvTl2$`(jqU(
z3WivJ^RHT9*}a_sZz98s!5E&qeR#fW$@nelZLtVAe4oiZ*tf-10y@sTKOQR->9Pgo
z(Av$ea*2rE1{u9q-^&Z7F7v(wWmI~w&Q8_9+FZYgJbupFy_bNMH;nikyzyNSvWU!2
zN({-;?N5L3EyAyLA<^E>kfG!_R}PmKH}BU;_tyN7?UT!<X&$eFD+^DY3`vS%s(JsB
zx$k4ZB2T<*XXaVIClGBc@}|NtebjR1#FVe%x^&3FC&%iJ3Z%2RSDE-(4c<ck2k_nL
z{>S8x$*%8H#aVW=aYMVh(xW>_IoL=Piapf>6Gx`#%(i+?m0U+ny(jbJuHh{1JHt8d
zlau7V7ZqL1b0sh*DDf)&)6WpB{m313K5iTPfWfov3-~M*bl){U;B)GdAlqWwyvLR&
z<hap5<+K?;<BGIiVv1)7?``}pRk9Amo6VGhXq)42NuDteR;?FOVzuO9ycqUh=_^!A
zv(flE8o#Mwm&-wC5%t-&rM+0e6Y|;WW}9JyUKL^2w8)?>S`?kJ0sygF)L&T-y69At
zG*hOb)=>m*$V-|D%;lU1pxkw=#k(r-|4f6eGTdT#{^`AOb13#2B2GcBpCfRr#|_Ln
zN%)BTH76Zv(AX{?GY=XhWh$=8Xf<*2+HkhO7+>_L3Oc*kP2X9%i*d_H<<MXXk59|)
z3_OJ5R-yI0(if&JNyiSmwtH3YYt&_ZXJj(+qBYx1Zi*39pQIG?K)8WNq_TciEkb<b
z_?wd%(+*!b-p=TQs$K=R8SKr3JYvm_e|U-vn1IABaU2l8;iHT<<%Do8Hqnsz;=Tnz
z0;XQWBPO38wb&=D#DEr5k2*@{t%{qVATEjvhI*vZwBTg7ETkFUq-GvNY(f$ItoNVu
zAONdHE`fZGwKAlZzdwk%|K%o8@!|_UVv#>lqCLJYuAqx{X|tOOoL0uvr$<s-dY#ku
z+WX2GpaO3xS=+f!i2eguQ-f4fw0C4&eo5Jqjh_^tPqdW2rS^f2pB8-&C_$)k)q|f5
zOgjIl$UT|LfPG!9pJLBs-iCeZecRkdBM`k10v9;Lom_Wn1b@ZGe$A)Tt*`3nkp}@q
z>7uqma16)lb^;g3v~MB5Z}OxnC=8q4NRLjrZm0q0zdbNB%v=j{%<Pt>XM8XOo$ZrS
z0CnOj9rOJz#{C-g>a(|@XHLdIm3O4^5sP@p`%r_XIHmyvxG>0)za&EzQ^?sU$}P@K
z%aOiVK=D)ZL738!h~@TjPTKF^rA!PqP9GZZlEZ>THHLZJmp8_L3lItnLaihE%GJ_8
z#kU2NEyq&%=Gk~R`W&WbWVnYe@zx@55$Q}y{4S?gl5!#x4A+X2l@ScfuBdp!dDY5!
zteu|uJ?@dI_1%*3DkG#-6;5nwDSl^9S>__}y(?+Em2c`wsrKoC`^diHWn-Y3()G5i
z<Y^YnJiGt~xi$cHceJs4iHoQkm(RalH6I%}*Yb|xI7up!7oj^D6JU<~?&!miF{+8{
zrLL*cVR%d5MmDb*oj$WNM5ivNrdYQc!?Fy)17@dcg|f}0E4mWh#tGoX2BH_`dgE<6
zp7@Pw%v-YZLnD&jAnUwMa2D!>bbsaF3*!?|E5@Ohl|hpsDZBiEm$aUJKurB*KO;se
zR}lrBxz8=s?@LHVibfRH-0Xo;q;5H_k|pd!m2Kfr4P?xC#c=Vywft<cL_tc8Fh#Ab
zDpqsGva20zKo?mURpG12Wc|f>krsvMX$`yBQg7SF^V4_KTX*<Be&#tJa{R?!rM$L6
z%(v)1n+?1m+s*VpzLMOMuU1yMZHqKw@ihM`A_Riu@t&S^<NXbqO7u$j77}!PI8obl
zK9eOPjA7ESyd0*2W|<Vx=#GRz+TGTN7zN6#IE6ga)k9+Au}Rk3sgBrRAe`aVa_B>O
z*)ucWR6~Nbs3au0IiapD&Ru~e>~beTl`|`A3VA-`OZJzG&dT?FmTUDcB<{GR*ah(4
zc)k7?eFWg5cK&H!_W!{eem|i{?e$)ygr`HPP5lF?g_oD5Oa2Pu&SS0Yki(qhbhj(+
zKnH<#yf^MVr7C^C^tG+!(Gh>lk97yvj+cKXi4RuEm%Y^l^$US}YXW`BipCLZi4zEQ
zJKM7h8bUtGv}}kRw4_zCg@u9Zh&dSAl9NNco{w0h+Dv?T5alX-Jg`p;&BWS*vrYCV
zk(L`Ga`wV=U)vYJF~=|F`E~NBzj!sD&S1m>8^l?5e3o5T&lNv0Pgt0ofq*`A4`(Eh
zmRoFsxStlxh#k<6hPBcI|3fSig7+a+U6W}l4rCElos8wQvWKeQ0ve|93;AQx_Msg7
zE|GA~sLe{*fz|$HHUH#@J*~D%F7}C_Xl;9SQc>`R&e}68?ZM1`Y4+f*B#z)yVVxNF
z6I@NHGkbQLudh5)<jy(f9}BD{fI8!EQ65O#l%+iogv7r6L=lZ#X9u0v#o($E<S5Hg
zN^7~Cceyh}2_bBYX)o{bTCoMSd(nkZ6uYg{&vwY&x<3A7%VKw}$^Y$gtVO+f!|P!P
zDQp10>$CO-RQTac=@vzQT6$#fUd^~=RGBZx(uEjgCpK&r$zk?ce17wpy!B9XWfibl
z0u$jLS8{`mzP=4J`I4Z#(n4sGr?XhhKk6p=)7|di`-jU3KG$G$Fuu3tz#AK2bA9QH
zvx~)U-B<W*>;{TG(xDd%tf=_Rr*mcyBdMJTE0!JAHFGeCYpEy-Ej?7-;>mBZZK&`@
zMRnp@FbY>HYEE;E50@CX1$B<=rSK4W>yYo`0gudpt9_fH_YbNdajsolFc|&I_P*0;
z)D+*3X$c(Q8wuanqB%Jb5gRr)Jd?zfKDfia;*teKt@Z~GdBZQ>+76TDW%XZLFA2(>
z7yCF28qph!+l!%ZTuj?7`Jpy!APbA+o+ymJ%nfE0R|6~=4q8oF4ZJ7L!d+^VH_(Vx
z4rFpnKYik?>778V0BOgE@S%{jmScmA(2v~UodL}qi0AB4uUZD*iIl@-e7`MAK<9DN
zT8!{E9;-t?wE|-FA|7=V*b1-2q@b2TrH;=ISJzy593bLLR({%M1LfeYoN>dOJE*Zv
zJ=P-|+wqvWPgH<ixcCNQXfp2VUny8`Mbp2!VM4FDhWwbb78Dn+j_8Z*8opkakw#4o
z$ryfk8Wux8u&%EItFi5*262ZB4{nM?Tk37ucv6I58XDrqYmFswlL91&p8NG*hnruo
zbJO5)J@Bzf9b7Bj$*>29X`f^UpDR4X!fDt~^czpGAj@aQ|J(`Bzxwan2$z(RF3UOm
z8J8Mm^#-bB559znQ&26%y#Y=bf@k+FT6(N9Ml+-4wH9G$F``LE{{SHTRY4_9ERsag
zMly*Mer0Gz?(Ud!yD|KtvsBnm6)`e0EKf6tt(lg7a~1x7jhHmKH+bD}^mx9t%gShV
zw8w>VaUSZq6LZrB>{S|+Y-G1E6eG%&`#4(4z>PNEy!pA;A=k%-Y!Ut`rzP))46dTI
zxs>+!PF!tkS<M85J^FP!cbO2mJ=dt64Aq+D$?)RquXI1&v){e?|9;4ru$@K^An}Et
zFE(H7MmO8Xf~&)B%4+xo0*JhGoGD*y$uZj}AMeG0UP$?{hW#0yKgcm;KhJ1AszMiD
z<V5GnC9Tp6p&AGbYdkf%318U;T4NbX?4sw>NSk9$$gkuqKc@I(2BDVvUHbA5ox=)b
zs85S0lS@V&&}&ovgm0e+BAB#qW(!=ep%CtpSh*XAoMuqdy9~8&Gq0gl(F`mHqypKs
z=neUXZdkPYtnc3$w&4V3h{H)JtTq(DYs0es(@aJo{mas+8R^=#z~1pM-WP(_`u^aH
zpPe?EO>_In&cz1OKg2$h1u}8Sf&Oez7*5*vYQL;bS9xQ_8ohk)>teqMa1jlfBx1*@
zD}y+GoITrtw|N?P=4~Ssh-Ixsm%6J=?Dl8q6G-~(2&S??$9l5ug)G-cF&XZ#Ube=m
zgg1@_pOLqPwFYX@N5of0u#gxB9BM5KHd{k_YXgoM&1Esp8B~fa5sPE><Lkef$O_Kt
z46-{6nrie2mo?hC(f(kEpHLhaDS%f-S=me0$95=Z4yt@ZL}kSET#ST}yf~!cB28X4
zs|jx1ozxxug>!*rrhJhP?6qE$4KW-(8b;ySp1fx@#{9p#MUx>D4#z0V>-1V(28-_c
z*Oej+`nif-vJ4((I}l+_<XGj#g~}=_1&mCj#~3le8Ck6SxNxm0Fy{LsMig9iCG^nq
z7lrbeJzo4D879HiiDK)#I*+;oXhA;ZSAj?9EFvs{z2ZLxVfl&Y#6?JNkOLY9wA#KZ
zOt2tnHeY7C$FL9;`l;muiE4jCg@ne_WS>e)8*!Qa1tT;6HB)tvqdq;>Ik#MIrAYQ`
z$xwufd4p2Et<rgEsxn)f$yZ}sD{15iKvGHkU^S82Z~(IFRyt;M0+$|dJVYuSC3HHI
z0aj6So%hxPiFEEb820tcKu#+qj&Vse{v754X-31BNtBDOg1auNjznQ1Gz#Up^Y3*O
z_exF!KFf0*2BW$nY6S(3dHI>eh5|#C!Sf2a8o(Szj_S1ECB_QUrskIdAa<)wtnCG1
zi<*jy`li%w@FR^1-F0*|0XfyjGM~v+ec!><#VWs?)`}bsr6u#3{F!f2m&Q)<n!Bf$
zU4}Wv2$WV>>!I~L95<N3x-*_s=5|iBqqs7J!fNb2R2#p7yydyjb}-jaDQ9n&Z>#BB
zZ5$|;O6r_i(@%G!Cf;#ad_|#PM9SLE`iqu_a@-bEkGWotnI6nU(~|r)I_)`5qES7R
zGsPR;I@IeGEIz%>hx`txXhQ9CjOVKQ%<-Ecdl88SsY-$p3^*vuL;N=;3c!+b9|>p7
zuHC)(!EcUv>ZpUdVY_#5<|c1`*ec<3X`{7LVq0C#Ks*r<;a*DcjZS3wBA68$*o+aG
z<2P+i(1;v+qO1DVI=TKlOaa1NiBD^WgQE6wZI|d?)i4t6(qPE};lHW$B}~jB_9FZ3
zq};KOSV@;=Tq+zB)aYIFv11XLxY+fnJc8yMinE(8QI{E0ftvQ;O}G%AF82k5QXPBK
zdk8G#Q=s>{I1<a|@gTyJ78xuXRD}56*`v!;xZK~V6)OsA*GuS;3C(=HE4I*g72v_w
zF&>lM0HT2Jl%d3RSy6K<CR7!YjtNQRb_u-Wq7-dRA=MaWhrgfBs3BZe8D_FW*vMAi
zRYC9$F0snaI8C?4N@8uTLI_{-ALCD|X-arqhp1FpXZ;>Zk~o$XVwyiOXpxv#Ic{({
zqy;gfi{V+(1=1hMR(=HA!(3KH&mX?51(r!#X+c8G-6$AW`6(t^PYrvPgNof>Eh3(`
zJLy#Orlw*AE_5^PijO2mM)b0015ScF{LzntNDN41W^K}(;<YI!ow?OD=I9G|^o!g!
zeeJ?s{PwDfUSU@zl7^^6BIUH>n{8<X_{GF&<yvG%m>(<z(8Ro;y{<biy|L!v<3ffh
zT6gkLA^(ImP=>g}@|Yp%=6sa5kRDV)KAzp1kNKv32aL`G`(rLSwm9h5SSsKv{NTg?
zAaTN6AiW+g25zJbk`~TkTBnd{tmXcU1&Y#1!==wIX|efoZqR(5E}mE9@rV>Yy<}KR
ztDVwTJgbFWC*O@US<E4*<bqo+_nOw)y;ID5-vhJxuq?$>m$TT73vR8q^MeV}!B1tb
zikmy6I;kz+vB+&_pK5SV<Tn~uf+zdL?iR~FePZiKe2vFeI|sV~)8~NXZr-d;Zd7_&
zSLERptq%^C+1O?DVxM_1D)+Kh2gR{s&%mK-UG{L?lLjme!+QLSy}EN$l(q^SW-&Gf
zCHWv<2byo4=o?pJ6UVImr3RekIvWtwtNm7SN@dd~U_Y3=3YPDE9=-dO%_&#*yNcC`
zTmb}4pOy?58H)#&L|+)`@>zeQebOEdoAfInqTC1Pm&<jL3u)ICm<h(KZ)cz=LYknG
zfi3w6wdTK3+C>C|SybBM$TjI{E-PdmA2y6aciH!R3(Vfxc=7Kwi4rw!-~)5q!k#Pi
z&n)#U4~Ami2Qsb!xuaOs&aEE78!49~2H&8hTjCr<`jA1<=raf6UpPW`0cdKC2EM~g
zN+}*s#XD@V<K>gL9TcmnmlWFuaJbF?xs>}K3%Gv$|DQ+jGfa#NP21n~!rgEZC;Ev$
zufO5887EXcSDZs2_Xj#Ls_jP05m3g6OrOOi^B1C6aNm?gHSRT$b7%&N-FgkN6L>og
z`VSyqo4EdUt6zM_@Cu$$M%)y&!hI9@Spf@MUOOmRV~4enK07=bO&e|2b1T6#{w;*H
z;Fv`g#+;D18LyL$o%*YmANnDYxD-CW@C$5TZFhRpbftoJlZ}aPhYyxTi*mJx<!;0l
zJ4Wq{D!k3y)e03T;lkUzDG{GL)~Oy^=~rSF#A0J?jFO1nam#ROTW6&d_{oq3>Q`d5
z`Z!89Z6NOD+>X#);a|#Su8n-d^G%N0SpK`Ww8*7h-8q_c>iaTM046>$`9FY*Q*u<)
z3|J#YE3b6z+KZhG4;97P%%7()JaK1ucWj6@xXVFckp3$Uk0sAnS2Cu}sc&q2(%3}0
z=Tfx834Gs2s02ti?;ic<emY5z67E0RJll$k*ch5$2{yKSPp(y};T>z(;Y%CVsj(J*
z7{G!@K$*F%6T<2}JlTv$_uXkczufRn7e!N1b2jSlY)8$r&a+68nYb{oN)i1Qv70Ze
z)t61%iOL13kH-U#DBZM8Hc)x#V#9?wXuYAC{9y=}7vjV<%N5w|dKNVjVIlJGB$q6W
zz|LHDcprTd$>(fNmC)U78Cz|-ygn+UY24{OJ4pKH<I;x32obkHG3d|SxfK^ub8K`b
zaP?F?;BK4&Z09-YLWaC?>q+{D2_a&1IkiF*6?pk~Q*LdK8NdhZE{7`P|4P_WNf)pj
zU}^a~hX3f)yzCsY9{7-t89VR!$CFleyxV7Zz<=T0_OTGu#L&0W+Gk&TdD{(rP9b9F
z^qI4~KDfjlh3?wm9r5~10GkWfNy65V9MU<F{5a6UDpmAdgMM1^&FV+XU25ekBa_w=
zpHwMRo>e2>DpxvP+8~}Y3#-AcPWROdzZ0?Ly=tlLQ*+L}2hETp4%?O_XWNwkY2kO7
zVh}AuuN*<x8G+;XKMYSj*}wY^Sa}Uu`YCK$q9^lsg@5doOGhyER7z6#P0W_cTd<y>
zyio>6_G}y(R$)zFY856aj9jwhC=ObYf4gYffZFP><2nkA1T(kc84EzL-1c<&@~OJv
zd<jwcuvn>MlW}?Gnr2W8muvFK2|6h`vwkMtA4PWdv2)d<-Lw_dL|u#DxZj4SFus&e
zhRYWHCd3MI7t*Hxlt~?Lo36Ket|iIgYAcWFKsSn^yO-}xmJ-4>;*~D@TQ?$pZ+SLe
z&5qTbtCHcY_PF212T|LUz<^e2XVq6k4X?;s;K(r$ljvoQg?MVuvf(Pd_z`^2D0J`0
zUKk@!rJq%vo=c^Z>BS|Pg`Ha6X$`-d8Gi~v`&9Tk$4C9nj);TKC#`NkO<Z+~`&74)
z1uCTc-TIdgWpt)R3Yj7DW)R?2Cp}-}*HRT_j4l3Wuna~eK^ZSYJi+s&+S86Y@;kaj
z6On>6Oc-Ty!M3ZWJSurQ>Luhk;@5sc@V_`C)Sn)9a_@!;hkm~LGW-iha!C!JYQY66
z0bt?(txZ*yTOXNyoZ&5z@;z8k7{zaK3NnK*aD)CW6w5!1VvY1sSGW?(&0uI5O`0Yo
zCy;q^Y2T_RztYio;3dO@$SoF9<}NGlGLkVR4N{|2=k6CYpy$j}=RC3>wjz*~qXHbx
z=l24*7?B;tMP=Zf-q$~6?DL;&DM`H7qKmFFAX}g6*g7m`oRzXy{B%bx8Co+DX$&!{
z)SwPNK4>{XzWn_>G%$~=On&*-0|;R3L4*@n$V3bb=4yUw#Or?fQ?L_f7F(<}%TKDw
zgb%NzvvRSDDfubf!-y^%I5Ar#15=m;?-yjuuP|QhLMISGJ2JV(1z$g2eOKq-t1_fm
z-CwVidV=kRV1h1ZHlISXyVsiG<1cx*Y<nIQV|ze_uwTqZ%L%%l=}ML)Wm0A0e7`2q
z=?++Jl-wF;-yR{A@4BZ>)xZ<0w{$FCkeF>k<IN1>86^r^>dpz`Y$wXEE%ws-Zq*Wa
znH4r)O+S6-hu)t_Yk7pZ$0<;u)tqb}JIl_j@_ucU*@`$^X25?oLtmQGU$-jfVe6}$
zf>lL5+kB<-L(LniST>z?iL4BC86l_!wGhres9OW*=b3nF_;dTJnK?V<n{sxv7W>>*
z-3?1U<z*7Q8vsoE3{mag+Y<Iby{rm9?dT_UMg1c_W;onLh199u{ATBlr;4|Bu5ard
zFT=gwFU4KYcZDD!P6jeag8uXtCWwRrI6q(gHr^7$3(jN=wZKLxHB4-F_RVwpF#H*6
zcW7zvHW#gFRaa6^MIat3PKtiQ79EqnIp5GFzHSTmK4aS#jqSwdG{&MI%|HH>Aj^s=
z?vm5^O#z&k?GS++{(Wf1ucA5Y%?i@y`@k7VIrH|qBz)RMwLN8Gm+t#WoFDq_IG=8+
zT5~cS3?Koewwg<K5!R{)*C`4{Uwg&X5BcjMR2<c9D2mg{<p9dQ#L4Kh7PnZVV_|x>
ziH85gq9vmBEKBm>*{i3VZAf=|YyGD1B#=cw6-!W8MgKr`FZug)#I^u=ay@Q=^a$xf
zaVXUDWNFd_KVZn<J^%jSa%fTjuPR^abB9#^r@G<7L|$hiPViOqDEJ=$n^yM4JD8=M
zAZ$<Srh@Z*pQcQT7o5S+m;0FpoNu%aJM|uFmyz0Ps!6KkHgV*{KwgPc;fX4_=5P@f
zDmub<JDdMVKOukl7}V(B(}g<2>gnJx^ep8>Lp3s>hrA%IF)YBnKs7TejRERAKF-qd
zuLx!!O8E@8GVNt@w~VvC-kSC_cZjpob1zAPX*!~5Qf#7HqPd<T0Hn<{6u^P%zKIc2
zMCJo8r$V@xq`ivgoJR%s6o(b;JMs%jbnq{f&0kTldJ$KJmc_+wUr2^Zj>fVIf&4@a
zWB$tzp~`zTjm~+Ky?2>De-<q<R)?txn@&N^ZX`8olTluibhqoc?JDjQ4_$-E?@w9Y
zw7;)0Pm9ZCmv`#rXB7Mkk0o9+g`-<mP3_%GYg^OCponG5=2qzCr3m$xvVPk_we$5J
zljx=<#&Ur>y_U}GQSr+VMk=JyhdA6J26Ms(p+H6?nC<_~9Sy&9x5AQ;2&x)X1j>RF
zFr|!^k|QbMgX&rff+0elT)Q;aIzG`xKK5<mtuIF>K_w{+?rSRaXemyso)zz0CCg~?
z@&VCISo#o>AH{kpM%yt-Cp4OCzoXR}xlZee##I`bq8bw@>eKd;w~_aE@$mE5q|w;2
zSNFv@`t6&ztfkW`{pIkyl{%kTTA4|_JJIAS?OcATG&u$y{D#cD1gp*!aDPeu@i+ia
z6{?sB^0WGIQ!W_a^Q5jWQzjW-p5TsJEZ-J|+WK4$Wa8Xb51W0}K5v#<gXj}h_#?cE
z!E)ID9FfEpBT~2I2zZ?c2tY|(Zp5PXv2eeHyrp`umk&C~!(0h|K>qYJP``7moKq;M
z89VX~DJ6SK3qcO|tN{@9TW-v!aujmK!M7QHwic~qQTwyf-b9}&*Z=w!?}sDQjc;#)
zd$@9E+$`GD_-6<#MXL@==t9%>iy;1l0MFCc8N0fIm&+n-sQh$nk+C2)y{ptG()m`W
z)!9rBYFAJ{TjxSN<64%#It`?@Lodo~?!4hm5m)j$6jw3Gu{p2uobO59QMv?fA?>gw
z8(;gR96BfcgG<@378?CPI#*=L;9hx0M5n!TCz_n;@f$&BQp7$|`cV--G3M7q6Fcn^
zK^5vIBhmQUkX33I^Li@&!_3TlQ6MI!-}cBi*-|LTx^-NGabeo<4y?jnrM9df!bqen
z(9b-ikQa#jsZC<s!^^BlBli|0!8$xgf5pe^a!_~v_j*Y5lEYlbofypqqs7@Dq+LZF
zvFhuHB*TDF5<)XBY`IA$+2iZXYK{Y>$Z`@6xvs*JUwVsy9{(Fp-Zrr9Az>h_f&=d?
z<?n-+*R(sLEL_psYbkK1+{w(++Q(5(-UEsAeXa0kdzqh7UH!}K&&B{k+$w7DXu@YA
z=Z6ky*&eFAt@|S5N%fWTmUrT48DAn{%0Ibs7snnX@DhuI=c*mkT9J}U&o?ag18Vb@
zUqQj+vz-FGHGTPsRqRk3=yX{Rnp8aPUx=138=M_TA~kf(ph=vs>QksfvKUO$;asmU
zQHhYh*-9kl*y*-Uz!%v1W)QSAJm8=@FLF9G&ZU~F`gIUVK**=NQjEeUh{O~%T9#wL
z<@287T;GV}Wi<7k($AWxASq+|FoB<aZJL|!w-bVi*BN-W{2ei$YSkEcxLH0PT7Fcz
z*VEZMGQAoQ8w-Z_rnhaZKZT%XLV_m=;rU#zJicwI&lj;*|8X(<Upy86Uv2$we#jfL
z9p37Kc_d~pV|uO7wEb%+@BnFtg!6K)<GrA6shCD0h$N8ZL*;4?!gD$8Pi2?I9g3b{
zX3-J&s8}pn=BFlh{qW1ogLZu300-b1!XLAfWA)V%>9UMY4wR_cv&SacjiKD$$$zLN
zJHOIpEAh;f&nHmagvB=)a$VrdW`*<?%t~NM`}9ucd+gp#Kc{8g<=@f7XGo$ue?i0A
zRf!BDi`yr_n{ggg$YEx1{b^I+2CGGpROBSAf%{8U#%K8r#n7FgEAR(+4*ao(3gXx=
zJ{3T^-V+~pP6K;7(b+~eb=*IIliAk`-1C#U*NZ%WUB%0`DzSGVF~EJ%ypHe!$VtI2
zOl*d6zcVLvyg?<qR`fLW#@17dHU!PY2U><b!ieNY&g1bPr=FV<`6REK5~Oe@5Eu~d
zU-pM|Cs&+XXG5d=588I8lBEWdg2!co{>7oXk#Z0Bqpnj1UH<5*bHX;e!d}D4jKyD(
zBgI;0v9uZXMc1yRv=M`ch(ifFLGN?L-|m$WF&iD~YE{hob0{1eAw|;#mg^#$Cp2Qf
zviZc6O{DV-sn(z+XA2`h;gS884C}L+TZ11P{<~DCQ?~qF_0}6_`)5s8RQsYxcQ!eC
z0s3s(s?qO%<aBwZxjLLl&fq}UqvC0y`Ug!5nHRb)(`x(Xvhli)9iZiOSz@viHQvtr
z{>xPJ93NgrYfN_yx?!RT7N(Yx9=aE{+ac^rD9O&1<!Zy*Gb=tlfg(-+AP-#G*sBX;
zmBKy3t2H;}@Tfp>C-lw=z22+SaQ(>(HpGvgd*l3x-GqvewP@F44g#~i%{DDFb_3)#
zoj#jl^B_2}PhK$o=_-JtT{*;}9tko;IOey;2B<KuP{$@Ml*x0r9kAG*NT;pNk*Nre
z+GrfA#JQ?hCzb)W3-z4rBP=yrb~I~7r$RDX$#|$uW-Y?Q)9F&aJ5;Vd=poq+VssD}
zRtAW4G_T%%WIYf7{B<Cbf;8BZ2JR-$MRgg^ej>WVjpuWq@lVM&6WW6g(xak;#MrsM
zWuPw~r}8Ws(8G71Sd8?3&t<=VYdmIRu+;84iR`AzR0ngiRdOKT+Xo+b42k))5aopx
zbdj56l;&UI1~qiCH+>Gq`h?R~LtQ0AH1zNfK&8<?P{JD^cy|kpW{S9Ak39>9KnpBI
zVx5~nyWm7m?vHy3?w3J$)5oB7Lj}4qgq_A_Y5$0AN6{Wr)~+Hmw^RMO`=vRhKQe(V
z8j0HJE*^JD2d`#e0Xi|B@d<gtKXUsS&08vV9!d_r22_^9J$T%zR5C&ab^3or&*9U%
z56zLRlpmm1Y1XzKGmT{1V(eS^;_dp|Sy*cDCS0!K=ak!M*NfH3ueGD^@{tn&jv);R
zSlm&!c__yBU?Kngh#ujHQY8a=-vqIB4oy5HicEwM5s^B3-iw%shY!ZWl@w<BdGSjz
zM0I)W@;#hG+h73wg;yeJx9^16<+2oibVnDT8Pfn^_7Mk{yPF<QC;SXxb6H96n>+e3
zOL~(z#d;_Ht9ulO@8%Q=^xqULfF9cg&Z|uO`Qvr&e`Kx1IKG}5qQV*l7V}WeI=S33
zR*X?-2H`NCxFQ0ACqxdP^j{w|hV#A#8T`6f{z$;lo8K&x7DS!-5=M2%<NHWVD#>+*
zB$0%DQ%li3gA&Hv<SiSFu1LqH05Zp4bu<y$s<Ww4_aQ8I`{>QAU;{4YF_hRV=wBVP
z5fIKW_qyTvYRcKEB0y)>n_$~QK3o)K=LH|<<tROu!;6B6+X7zs;&{fg+~fS@oqglv
z<NV`PgcX%<si&|yKEZO5s~v6wsTjJ+b+-mw%ZmWEijNfBzv>6*X3QbpxI7&!eA;?f
z+Rz3jjL}(OMH@aAhta~66V_RhsYRBmV5_^hFqBFU&fp<(u0WM9@<H}R;>!+0ILC^c
z$wL^s^^4k0?EbS?#6>c+nG%%`R&=#m?g_Nt6-Z}zF=P<fh6i^wg$Vf$OdNA^8p|pg
zWf@T-H<5`QYY1pV5>LhBrMm}|$=`@1(zv()#np2@OZw1i8w}ABqlnUd8JN-C3|9l{
zGVczZE_FTh8^!<x<Kge=9?^Az4Bhy6P!woA?dP%wCgTQ!$io4XCsETDk8rRc@_=NX
zET&eg7%x4Kot<#?!(QH8F&p3XlviFsQ7|WhL&f)?Nh|I{d7?=ZCSRgskxNEoQ<%72
zXYIIt%AfpB#=%{YviGBl5f4+Y_Hn~n;zNny*Ht;gSELM4AGE)XWy7s)aKd4hc9W*B
zJEN8aoVvA6K6~ww=C6hKQ|U9F-`gh33|kcfrQaT*0#p2`l<`yZx`m~}(!&Jf6qnF^
z8YZGS(sJ}IXo4H@X?3!H3x-X`#&m9ATiBv{_$?n5>0I)lpRIIjF%R0}-|FXYKb{(w
z>kt_o#GMDZw>r^9RC^?1vn7}VF@2Ip6kJ?~;Dn4S0~RumVh`qtK>P8A-CX2EO?#vG
z`@Q^cDT$;R;!X$l$Y*G;#IwwwcXVqOmr01`@t-m;s;X5YsbWQrx$%LsX(9;>AZj>B
zxA7U#6k($fi%8*Br^t&t{wfeg!luqaJ3xXeQX#vfOwItqjq@darEmDjl~H>-zi|X-
z{Ah8q*+N@)pIIRDJi{*g=zYAI$XDi9ioWeiWTogqrfESs$)=In+FiJEE=8S91=bBQ
zLWT;2!YRW{R;yx>veTjd^U1qOpWO|4j$2lS_wr8L9X^SL5sys`{`wn}8VKpU=^6Xt
zn4A0;_raeFi;Mq&`>QY6k-jjm4Sqh^d8Wr5OZ3^L)rbTw{&uZpclT)&f;e{9-<#~+
zu>|><awY0f9>#kGW4FeL<J@Gw4_O#{1+{;ha;+{115RtY?&E&Fy%aQao55Oh_XBn3
zp&xIKb4bQHw_Vai(Yg@@Y15dmqln&GY8}S;?97*>7OvRUDbS+mCdk7++c@s92w|I#
zHmeG%qa^=dl@B&7@d_Wf-bUA)S{}t(Gti+9X_xf&SX&AMhBMBi(m%n^Us!N8pNWhv
zm?7j3li|*qtL~cUcguN4yk9aOFcC3&A10k9UHKSu)@5_3uD~@sJ0!*&I*kxR?bpeW
z_~G^QbGd<}FK2MzFJ_S8x8;$(ut^?T)at`umsoHme-*q^`1}J{pl|~ger~NOCRO|g
za5m>q_|%&VM6(iy?ekFOkzfBA$FSIM6cSDPs>4%c)5$@WH4O&Dm?u}U^f#S-GEbKq
zAwUWJ(+bHWo7c!VgtkO-9?H{tz`I+NK3ygCq*uzr$t9V#4}Np>X(<AynoF)y-lWzl
zlxa9IEoH5F&lPjt;w04ZG7+h<nO|90bTh{i?XOAI-}u{J<N|B3#~4w7QEs|(d~&U^
zWO7jCB(HSZwHUEv+Q<}Pk32LJ&YavAzYwUU<8#Y6C25LnC12vXu0=%D=8MQ-4BQyw
zg_KcGRvwZ<yS)KLq|K3Mvn0j5(VV00Xd87Eid95$QJ`KC$F(H;0ksl(OE>vzSNtdZ
zi+MG!-6jmTeJa~d!#1qjNw;$c>d!K{IJe^Ea}^_01jD*-plhls5oP@3Nl0HT*>5T<
z;_D>2#R)u2Z?=EADL^4go$DPLR&|CX@w6sJKZGrLhVusC{4h*hLOf>&kWyj*0B7W`
z<lBI$B*y6jxCC0~DVo#=jE^~oDv-E=-$zcj5<y<45OBNGop0UkcZ<gsY;38hU+H-@
ztcGo!nsQ4=i$>z=(nEL1AF8;t85L`I>%2KRge6?MIJ(_2iTRf~PgOUarF3O95grPQ
z0gm_j{{Z^N0$G|SamRLU8141kWRSsU2+ZPmo3AwDsbg+;tyO-R**tt)-j<yP4NfI>
zf|UloO62za>AhD95gRah?wUa%P{zFUUbAMK?9^g)X&&L@{4)Q$g%JI;_(2vElR;9m
zWD|v|eccE{AK5>o9zSviAY8OJqP`chSyx#GgDr`}WaW*W(M$6U!^_-F^y+;%X}m>5
zSvPXEh9-t+12$2?-Q^=|hB=7?!J9fZ0+zjlZkc~y8|TKYUEZB$Jo2v)^~Pa|368*9
zC2gjN#3?Ct{L-`3HqMCQ$hq^C!B)XDAb3v|u1-DX7Jr3v5lj9@7E(O?xHI_zpVBPA
z1vWIVng0$<lDn0A`7Cxf34d39#XgCBbpCgu(h|k>Dx41u4`zZ3Z^RDaZ`6NfF8!BJ
zhchexcR$?BXL<6Zi)1#~$JDcEO_uW#?q<ds3uWHu^3(kxM1f-=Gbp@Z22BP$YT+yz
zJ!;x4Lx`~M#(5#p0Dj@Afv>HDKJoROs3kK-6`toBH2M)JOe3oH?<CqY6afO`MvD`*
z--#&IzuqV@X*fa${`A?4Df+Ea#$5r4>R!H|*q&#vi7<EeZ>7IoR-p#61lEN0%yJ)b
zQT2SkE1?xTP2h?VFFfT64uCOQthQQS)t3H@6cWg@?|q%27k$VxR?6+NeTQ}jLgZ`3
z`@)h=#N8I6k1^qwHK#PHf38it87Hv06`J_Y_Tw@~JR+_K5tNXfl=`|KaKoqjXnJyO
z!hubyBfxG*!AX>ti5oPfoEF4XrFs=}gPB3q9Y1r?FGVR7X(F0TLk-@Es?4(N)Nc;v
zc=Er^vr5(SRnH3gQj5T)s34GMTNg&E{S`?Z|8#*J0aC5H*c0;uikzJEm1*GXhd8I<
zau{|R+?xapapyl2C&*VP$nArTGe#?jwE3%}2-UV^x*aa+zGk9@lbr}gf%S+f+fv8l
z9<`Y{U;SnfrA2S>Hiuq4(ta%d`kOmZVbpY}iXsMgy)^#0ukqg8E1osTu)IvX_bfry
zy^vqisXYeg2fHw3M&!Oe9Ll5_q+B~Fd~4<85N$csI!?i9(X6c<#61}t8S^=*im%_z
z>1wRCL58^wuaMd9nhy;%l5rMlPywvA!<1pu`N{Di<O4tT(^;JD#tE&k<Y>Ts@TPFl
z_IjuWPxRA>{)P*@XlEcCGckbGco^8`LvVmJ#!3mUVTr&)L1eTn*<Vn!FFDyBd4FVs
zl3ny`UZ)!5n!ZMARiOgsxS41o=hzH1VA9OJwRoAORrA!9-y8;Y36>#%^J4OWGR)j}
zx%cVSQ}Y{9E~PUaLf#<96)Ty|NA+uIUax?AVNhk`6!+>6P2V>%o`Rh<v#gq1?4~4x
zogn$6l2%m{9rFdSei5@TZ~D|v&e3sOyEDSX`4Db@**!RUx!n-#>I0>!ZQ$cJ&mA2X
zefQfRv{RHO^R#$wkXBzZ6B#nq8x_{?e82uE{a2+gA-b;R-Jo9~%Td4z)OZZ*VApl6
zoA<{H)I)5yJ$3+s9CtI%KD(p*4<K8yi?Y+ST?l)ie9x$KhMY*i?jOLW*nC?sj-WVw
zi&DDNd4Ya(UTh-mx8ZLq&*FDwkbgxOptIPf5)d@It3ieOp{yg(R;nKfAAV~*i~j@g
z|66+hq~0=fa=q;oNEVlJKAK=)BTwLxvj1j@6ifW{RNdCn;J#N~MRav2&S({l%Us~9
zN<KPmByF=qkLcZ`2|u2T$_4LEs;n5_#G}y9U-c5aEu#8n@tjDB@eLzu!C$Y$u{$!2
zGy4ff%~?0B;9bi^4-1iYDZyk$qgtW?v>kBqr|7D!<`~oa)RAxb-jOYDk`mv)2{(3|
z;98Ahum>#tt&U)CjFx;Q`RP*-gD8%EWQ^wCZiGVn5yki2^|`SKoc;c9z8xHGr}lt<
z{_7a_A5ba2hC{tK_MqCS*maCs-bF+X&O=GMp9#LF3N1*p*aMYg5vzT_{G{Opul!T<
zGGq^aHaIWWBa7IXV|%|jJbW0E0e1Sbq&nS^Yc3TxYcOtK|5C77vuTN!G!=DSxxy#!
zw<7BoJ72^%9n({?tM9Tj<yE3Uj4*|cOf8o)XTNCvL$L_e$trkyCiV}6hG9&gg4=q?
zh>@4{hcN6N1}d-a!Sifi0@<FEMuz(k-=)3&K+i*Bh!#r}tmOn_o0Ox(_lGc_UH-?~
zAt^lzxUpLKjfm|eE}AZcny4FyVvzDhzfa!JU9DJR53x6EcBn|ktA%neo&>Nno1|Fj
z1rAQy^>zxd6cL0GGruz^NlN~Hk*f%6s!P|4FulM%87LS@y=~ScMN5l?RY;W>-$lgs
zX^aexj4G7C6$7sR&pjovimhxzrcXz?EEF`K3=ct3iS>sWO5<EM<|(EVF}mdA@tDje
zVm`@ShZw$G)w^nOBAKKf34U;dS<I|j5$5Vh@Og#+>PfflVQz-(^BdDfSib32uwOug
zNCsQ9=o=Gg%<3G;D2uP5?Ydl$-}K0jlc{47`qiOUU9o6;qvh`S)1?#N!es1tJl&Cv
zSKru)>|1@Y{e)LzrX}oTtnhWRRq5Ph*4qoSpi_~Sf*}F*I9mInZ>A%K+NK<TyPnm7
zvU^$iRGStRYwz`+a+o4K<Pafij9eHLet_mu4AR-=MPXvVaEmfIV>7C-`kR(rS>w8r
z?JUSO1hWH#5g$&5cEng`HB1lIlD?>?+BM9tRKOu%UC%zIuLL#Pm`AN-(#JKAhk%h;
z6l?cI>>1Om=n(?#(5C%ENG%xEDEj0;TFaP=c82zk>fB%EU-`w~Q%*vlhAJk~PcVd2
z-E=PL$5oD_HW0r$ySCMP!G-L~FD&7jXaWZg+G9%7SzDECKm>_`wCGrj;30*e<cbTV
z+_!L=UaQoi5@<J^%8E8iuk>E_-stedPN?MrrrAWnKxq+l^4zJ0qU~{k#h1iTURQ@%
z&eEV>zdd=oO?@>>H#>QmUDQ108Z2mxxNF@;O=)!c`XS>`cYQZp7gcnwh>s_1sy9}C
z_WX<O5@wqXa!yHG_NQJ>yW0nJr5ptj6_pq}c#_n%{8@no|0)6=<N{^U$j4qC6-`RV
zP!ebMp|lTQ69q&uKJE@0(9VJ>Psb%5_R}AEQ$CLNIHk1w1At4lfoBYfY8i*@#CkZC
zJ$0dI$1|}Y#nQf!;=g&H%oY^fXfGAGT#4>*VpX%57jBcQpKstEK(%{qV$}cg!$-B!
zQBTaiD4@O++VI-<N#yMkw*kM+TrF{&e*ktscr=UYGrTSEonl${*VV#505Y*ZaK+94
z{4kA|VtCZJNbtqXD=r}T|NH%V0^2KTRl>YwzHIGIMGu>?n>+{?bA^~)Hx9l~3}Vb3
z4~W&Q^Uz>cx29uKm85eO$+;3dFgrw>TYm7Ili^s>a#uc@xOg2^wSkW(y#2ZRQOjUB
zU@!SMwLmU}H`VnL<slNOQNa!)#@oc3c)BTfEQ(+&<oe^IZVd0+XX9gTf1IoYTiU6n
zD;;l@o?H5Asc`c!?U3wjNPio)x#nQ6P9dL4mgbj`d~81mXt;!4;j6YyO&$P+zsyoz
z|8Gf-Ta0fJ9~B@*R-lLxRndZ$A-SS9YfzhN6$VBY<X@KUt`J8Xm|kejug#AQH?gmw
zREnDA>W|wGQ$FpNA5Gy989g-#gPgGBA<#%iua<{I-9ahaGPR7Qn+YdN?dt~L{XW*V
z*KkZsR77EX|0iBc1{RW%_<j4T%*G5`Lp-7x(Jyu)xmf3wmm(UGXn#O(DD3r(M&m3^
zB?K-$((oV<Pj&q$=9ildR2p?fJ=epn`aTwM&I|YD#&X#F!pq5st`!T&RZD`oOYX}S
zC*QPaN;vzuDYwm;U5Q1-tZME_-bO^&$6J025-Atp_38dYa#ffmQh>jrqw8{`@-XdP
zzFlMr*LdB27vN1O14R1X6=3ShQlh0M9?S9jOBIB^N}#S8&3in4Tep#UhJb4QWd$0E
zbg3mwN9tCzT73D<hp!9Me=c)ds<q`J?Y%IyMbAFn$HDe2_Ty7ML7cHK4sPcPmOQ12
zG*{=6jz(h#=>p@TZx_Zm&q!)vji2e=(S8HK19Pm3_&&1Qm$JT)gjVk8xv|3Ta}A->
z{fx102v70Fd}Znr50%_U4tR#LD9RuM6sT}Ddm)Zbot+uc?|jJF`{Al;MKlzG#J4@S
zeGm_iWYC*!7&TVGsQDhViE%MCADON|+AcZODo1GEFQQ@}L6*T3G5>dXpeSdnDZqR3
zy9@jhAv}(4W|+&kN;(qMIO8Lx`9!)UfPGf9`9!?$cy!<g$h8PJ%2n2-hnD~LOk<}s
zUi?!#PG9xGi;HU7<O`in<_u!vHJsgA|M?syo|Db&Xqdio|7BiCprfo1tvyh1!2peU
zbG`;hEO8(^qkx<B=5ecfX=9M_vV_orpTF2LzAiG^B7MM*oTYfK$>;zFh32D7KZX)$
ze_WNUUJ<bQQP8tmY%WfYJIWlS0SXFNC?S3;H`nzx@Y4A1>DPQG;+a2-!w@uQO*vtc
zVNG3O0+kkYnE}Tn!S2<0ZX?T}QB5K2Xc&5oUSU&N<jfb1O@w0_7+F}TyeMCrn#Q<l
zrDff^ae1M25H?(KECot8@ee~yALQdlZ#}%9%KH#EBs5<VHeP)O;dikqgj=QHhi!_N
zE4d{$xWzjv7|ze2u!5^Ka^QyL4W7XN7aNTx`v?nv_5y#TfYTdt`3YZ<#3_h@ae7cA
zQ>g72p(kz+xB5UB;Q>>^)lDAC_kJM&I*RP<nAw%Qf>DLvNs!K=&O&l;7FZ=y2E*3(
zf$+n*)=F<f^lEj9ZOg-3cB+lgw*ZfGB^L}^cVFjKEoL5=x7BB=b+ry)BhNK+IX15b
zM}`my=qdRSR-}~z#W2||LIy|0`k_mbr>$x4iSbe)&med6erP2_D6_^kII<m!aV=N-
zqfBJ(RApFXL;kFNv8cMHSiP;<mr4VRMDhsxi)SxHD-o(t{(*U=51ycOIVUUaL*9~~
zN0Xv##fYZ@7a(J_<cHL_+P@9bUG>vWV*SPw=&#-8*}lf1^GB$;_^pwh@#5j>wmcPj
z7(mC@a4eNP83Kzt9`tci8gP{$ofnxX1x+c%c55p%jP25A!(|c9ZDN<RuekGFEvGC&
zf%d-XSjrNfu6@`7{{RGQ>c+dTC}RAky-3H>t>Zq8ohGh`Fn5d%GPpe-P@ja0bYChI
zyLwZIbIZRGMz`OF<h?bA8(6=OBXU<+vG3(Nr<*ZxW*qe}zW9`LDYPCU&<6pbOy!6U
ze0vE8&plsRMu+f3sxja+<ZJ$Kfz01l8xcupVp+ouwf_(WOreZtW*5#pXs+kHHXvkE
zVQ{K7KLC)`QPrZV9H{ScyL7d}nya^gqn36jN)BZYu}#j{;%gO>Qc1{H(#F&3$<x{X
zR*BK0--D+XJ_mO(@$oO)TeYQ6@XcW~zx$ZJH!ydA?N_~*Ucx?@HA|GDs|<D@zH+xE
za(|aXs6Qy$2u}wzFOA!WWqc63{btKaY&BQmu#_J1F630loz=cS5Ql!D`TL8vj?zv@
z&GbMm&dW@{EqweZQ^+!@am18dPC55?39bn*@@(C94)|1t{b*%t)<v_kl0)z4H}{jM
z%h7*<TX5oBe(r*JkaJ`^>`=#3nXX4M%{)isK(cn`@H5q>rl=NpF2?JyWA5XU*MKg8
zq7r%XJdLw@V1@jNgPM6_z%+}F*@~4^&JC)7?9pi#?1|8`-x0X3B+}v5zWd$`8yG$s
z<(K2RMH0;X_(#oMzWVb8W(&KTU)UGWr^`g;9WC0&)IZ1P30GZ+!NBtX<zwbmeKhj~
z!<H>+yDVm@P`DO=L14Q{XijYlPwYD$58AJ~JweAy8DF7^dQ1sgjF~8MiUcX_QM`&R
zHqm``Ck~*IlA}2sy=mKoFLSx(gH?&miYhgqH??ahcR|;Xasd<`tDQKTA?i8T&5^Yj
zsjNtdg40k|<cr#-8m~%}9QSUI7@8SzPWQdDrYTv+#z#;-T7zOUDapHuAJ4-{fD1sw
zNDSRC!)6w>17m~!9}%qSd}!ihVPz(EL6n{??>mpO2Ym^wB4aMK3(hsc1ffU_&MiBL
z=RX{I?;96#bu*;i#=g-to~B541+dM&Up*zd5vXeWuZC<6hF-l*Prk(1XZNq79F6t4
zE+06bnTH!&bSB|f@nqvDSuW*H4LBOnZe57unMwbBekR#rdt^a3Nko36h%KPxh9^9{
zL!xaaXriIgK09oP>{qe~)qa9)&BG4`{Gh{pOA%}!dQ=1M%5H1FQHbPrhV8(YYXbF%
z?saZ(Iy?2Me7x@+6o=D4_0-faJI$q(m#Q(njS6W!4a<<oFT3!fq$$ntyo>XI3CW@<
z9;t?lJMpGpjXe2F^lKT;>-0F_C+G42M{Uoqzjzfjhj*ITYlcI?oBZV(sMM=K%qPzX
zN;vj;uIyPvG{{Ed8h2J<k6kK{Id@%t@UY)7R;H0OiS35hp0>#_W?&v~GUY4QeU+){
ztgZouYm{V1puJ52sW->ZzxLFxHxrKF%g3l4P5djGyK{M~WzY6iQc^e+y9x~B3Fnnx
zKxdJ!DruNcrI~Yq4b0ZN?-xAIdxxJ{wc{u-4R5GEE_S*24UAu&)h79|JESC@gNDzw
zM1~+ObBD61_uB83&|?%J>RD6tswCi0soYW^GN<YrEO%P~67~1Hytf~kW{OhZ!FzY*
zAzT@`T#{~jP+7xY^C2neH3H><)_HFYTDB_WZ9j;#Yq7&YB{-bImUtGkd*Xsq+}5-&
z>mub992oAlACQZ*-9C?xBl0+HkZPS7HIE2xgAsRm$s0O3#V4|{hXhUyfWKcJKfEl%
z3`x}fa8F&T?dw9%a|pq?=|~=D%s914#@X=uovmD~ha+}#v39~5d-#&9+$GdoPWE88
z@jA(<QvG?p_J(F*Juk9`L!dIFHo9_H>J@i4z`KZgl?(UjSGFqYCwB2W9>BI+a|D6d
zDX)IGZ`1DXWD|dhg0(aHHP3Z|LpVZONKVoWO3bfbL3@RyENepVvH+QUzN~deF=5pF
zYv|^Lm)75l0SKPUeQz$t5z=cq@l^5!1)VX)dh>kBTrK4|Nd8n#$*Ok*wG)_N)EhQT
z3jQ`?yEM(qpTbG{i|RzhLtTY5jd^H83P@P5*&X(I(*aVK7}!SPs2jfs%k9zbnB8n>
z?(0igb$w%L<REmlOTR9dF`_)e*-q8KdfIw5?rf-x3n=|tIer3KXN?LhrW?d)VOa0D
zGWu>{HGGGQnMKxXAxp+gc0E)JKmYU+eI5>ug*O|PDrR3-eEgOvo;LqgXW-WC6&JC6
zxMz6-CAQflcoKfT?IO;b5?%BC2OwJddT;-qE5Ls)|8w=<{IKuh+m%}0#@j8<tz{!u
z*&wN7aSEQp8f#QV*c1QCodjN;Kqi9o|H0c^N5%EEYl4LY2=4CgUbqw7-66QUTM7%1
z3KAR&4=zP;CnUk$EkP38CAj|Pd|%(&cV<@i%o<t!-&vboFD&3}dEQ6Fh=|CVb%%CP
z&p-blb=&mpA%D@IEB`*<Uzou`G*L7->u#2YxLe|mKYh8gbx~&Hq4+>lwO`x&p?cQI
z%MsoDo#3&q4R=)1H&pAZDvG{rY0ioZ!dg>XioR4wMA>^ioH!QLoeZ3Ln=IB-mM4Zb
znRD%AxJmSnkc6>LFyLSOt<s6`1LP0p!j!W1Go>EI5AxZoM%ydv7r}Yu^!|T9F=nPZ
zllj75d0m*#*Wol=75{>~3Fno2CeF~E1M9VT@}2p2yb$CqOr4DyF}b=B&IWWpqKs^w
zuWdbVez+E`X$aCTylz8%L2PkhT<s!x4=2eH?p<gwvYxKB*>RtfdYaQA8f!3>CB0`*
zctFo`;c;n>Q4pv+(J#%zzCu7Uf6?D)L0FIAYk`~JJFtELgx!`ZYH1U`^C2)<fA*%v
z-HcKy1sfN^dcj9=VVKNQyzeE=d3nb2X5F!03cV7kaRj=UO^u}$Q{z@N-F{2W3`N^$
zlC4QQKlj0F2iwKwZu(^tqa{?Lyg-5ICB4#YkI13w6!VWXxhQ#~m5L$f&ff@<e6{UN
z9%~}+rYi{1vb=xk3<D}<0jOvZsxE8Mo7{*qgF42wZ$);Wm=}oLu@{PvGv!-9$L};P
zM&k4B!QW_VQBGr~3JnMuGPM`sj>8J3tuG58fVmpS$!2l-rusCJ`6=a`c7(6YN4}0j
zX>`9NP0=E>vCu0ZYKd4&Vj5M0!kXm3B&6@%we%X?zM=aCKj|1fzJM}t(3Jr(vNKck
zmf@4^-g+~nmElq9hn%^{H?pIvnveVLniiy)&VTb))3nzQA|pesTRJnY9}HB?XLULS
zcy*xlbY0teX=fpeikL>7XmW$XZC|YAhOKse)clJS;&=Au7vgsW5!ls(YFe{i*-<EJ
z2%<dKm2d`)4@jmjc8h&hA%hwQCdSbvIuqM6%2xag7Ay6{TO$sV%Bb0_3lj$ttjrVL
z!!c)XlT!f-`jYfw>cTrOE!7|DCm*JlpDuoT8-}&kj@3_;QoHy$SatRTY6XqNFF0x}
zzXmP79-LX^_Z;u%8Y*HOyBbSku?@*1kDqz%DmvUJ>%rP6a~QxM?HHXD_nh_+zpS*c
zrKCzao7=sklLvps&KBKm=Tb#}d+0k=#z!4mYZh#Tb6q>jr*0^fatHQlBXF@c3z(sX
z2z~KS^-FAR{p($rfdQcd0=l7A=oB~UFUK3V2Y_NOxw+!j0kFN+dLcrI{2vVinLTCF
zUBb@~kn@rgw&#`JvpE1c0ns>XolXE@H-%5EkE)OW&qrVx13;9YdXG(z4ZxQPU23*)
zE%)VF`VPqW`5%0@|J$-0s6%=CMpoA+!Y_ad8+_osBeL5fge?M}ROJ3XMc9RE$y(1$
zCwm|hgKmRFS|rD2%uN?}0R6(qAc%$C786`^4M%uI9~Yjd%{3GffRXela`IGDcfMOd
z-7h3Q{mFbqLsH55UjN9Z$PZ4a{)?K6q@M7siS>{@6_+$!FQD?O;w^4H;T-MfZ!EdW
z_Csa5h(d)Mf`f%gPI#(^TcAr50SoHjQvN3MLer{B%WU{^)%I0YS|bwHE|ZT6Sm&~f
z0q!I08w5vD4&*7Ww9OJ#ob(-ZQB%y!wOf)F*s-C~f_r2y*rEfl3W%uGFiWU{f+NT1
zXSGRj0_DVVNVqGtsHrnX$LhogbGcSAm}(8$*(&#?+}Eg^Px`?S)vt_dIBCzCY`wS4
z#|X*pRP60$wL=DR9}Yu`R*ZMS0?L~_;@$U01BiC-d#d@JKvP?qBLg57K`${FC`oBE
z2mQ=#8m$$Z4;tDwCWF_s<E=&k5-pMwKnyOO!~zZRP&u1)Y9qa_a$m;$2b4F>AAVK$
zn66DPxqihFGL4*~medSq)<7iWV^{~$lrazJ!uktkO<*%pZV-7>HCyK==Ae-HuN2AN
zCfQ`1%bY`PEU+J6vrb54vuKe9UNMx+v?wMTisv<6t*wT;mVH>5=f{=Zmz{7X@%_x-
zN&B@G>c@0mE5>_dK5(qpzvC75<t(~SH}z0K`I-z5oo~QICR_{}O&>JpX_{e<eTDtw
zLonHIgX~kjA1{7LoJ^Bam+zQs<*SgTxvi)k_1Zf;Tuohd5&4)mir{7w`WsG5Lj9+y
z%A?V<)dB<+@Tt6(3Hn=&PJbXe!7+9S^`e^i=o0f`dhElD^HKcaqWAQoJFK`TWG<4B
zwj~jQQ+SunE#2u_7Xw|SAn&uqN`8=!fW$AfYl<y$4*rDUA~8#HxGZ=TA;^Jx5=Pr4
z`JjJVj3qu4O0~#|;NY_f;42}YyXJeaB5iyH4eXd7NsD@BQui8tDG~m18@Fe`HC%BX
zd;N0$w7=fFPLr=<-2>^C7W?<tWKjNV97M77*mqeH0uXK90d6Q(XE>9IN6&xmrxE_I
z=GS=+#ZaLiA=6w1_g(=V*It!fu|?F&x)>imFRj_c&rvei8*qt)1Z_n3Ey5it;6I0B
zz%b>VM>?jl(e2nZ9N={pf73=`k6EEl+U!ckKzN2puMg-4qBb;rD=OfyizhYLBYQqD
zMq^B^>F$2i!y}j&GJRh$bfJwTH8TOOLxt0&ko^srC9A|!{A-D2J0ffj%N#Uv?l{u-
z+*!pA^{aJcQ11UsLyPX(%Ik_G8-qedogrBOn-eLJdGq##OM|2FU)vsE246%82>X@J
zd<!@-tnz%0X13X5_^0OZCQn&Fmjh#}7URMr3D-u8!&{Yi2ni0kMQvG?1N)k+W=Z1t
z<CI3=YOJ;<$H$eL9JmH_<dVE`*#NPLkA5S1KuRhd&6^Qs4-&;0n-C!fDOTq9@SVuu
zc`((#GD&dLIg-KARbUvFDME=-u5c7YV`T=ckjW;Pu|TLWpFC_<6lKQY-TW>pCxos}
zsKqiNB9R$~EJPKAWTzJxGA&TqCfyF4NgAay9FE*$;|u@eZP89!FG;ni<q$xkelhnn
zI5TNb%9=ppDgH*}v80nZzJsmTSr7{gAa@rqoYPL${Rd>|eOOTmH;58Wwq><mC13LM
zV}|j2HqRjN%U|;7_6*pmgip?+G`bApCuYwvCi!SeR{zKY2(F*vsw9C}oz(3&IgEY>
zGI*tD9QH2;S2Y;iGrB9KHp*EFp$qZ;#JZy3*UMjZ&u$U|akJ}Em=ezAF_)yPgIrZQ
z*O;n8oh3;!o@x}h(02GT$Bv=4oHVGFm3CIASEXa}gb#D^2QmUieS754WLt|Ez)1pt
zsVf@d`-Jf!-H?2f559x0#N#Vf_J8xp7e8DiXvgy~lfCnZFe2D=QqF<?eO-+-b)C|Y
z!&ktOk&kA>N+5apTa343F)+{=lDpLX56DjT`2fQALb$$}^!+&n^|tJ3=uYQ9A9Puz
z8);vzdb5E&=)ax-`ojP6!5<fic>$eNdzr?kZCT>UkAeu7U(<d-myXOz5<ez=O;7*k
zlLFO|`E<_DI<zN)KzX69wisCy^X|$pzDfZ@3yFb9j=KHCq9S&34g9V}dzE^yIK)*2
zx*Wxv`Z?7F*&x-^w1#Vt4MA6}z@^FjaxpTI1Xa>WnEEQhIalyHi~bYQ7vp2_M+STs
zyTKNF0p<g<!ma^*m~8LvZb#|FEdAtC8DrqHi=yS+)e2^TU&mi@9=Io~p*nW<fvJL%
zJK74VX<a!d6Ys!S8{~9J+M7)!us~(IpoKpqO*p{8QY82TO0!_rny!P{Jc_=7JNU-(
z$9fX00z;4Cy`pRW;{3;(Avpv%Qv+Nho=&^kc+So(@6sfjkKz7jIT9hr3nxV3aUKLj
zX%zi&!oX9$Mi8$=?8n>M1vB8yPa}~%v%PNuEUy#1C)8w6oW0j0(OwhwCW7cT)3p(I
z^RPhC0a}FAPPo3ws`5joD9p;GBiBXPpI4GsJ_9l%Lx-<E)=#b~Zk{|bgTZL~?8IqY
z(I(kBw?liC)=#fMTECc^JUxV!=`DNq5l>QtM!%#23!vH0fKFy>!VL~TtmOrT<#yG<
zQNT4s!=QWiPc9G52+O|9&yE@+Mb4R)rjrz9=}B0vS(*uH528s}^9akbfzv`6gkS!O
zIz4-q^HgLrrSv!X@K0999GP5{6sM3?`}WUjpDVpw(g+ktm>%ZwnMG0Zhv{gy%AL%i
zHsEe4fy(l5tN`B4*4V-38{Xy3O?5~Z06ey#;Fyl&@LzY(CKxueh)mEC+vnObV(t;W
zhezIcDeu9JlO>lJ3~aQ0LXDWb_~?d<)NguM9=4|f;d`HJr6-PNURZ)cE#{x(%MYb!
z;`~}~(#%LM5hM5b8T=(bH9hN5X$ZF$H@q))`YJF}C+Nx%>U!5~CLS28{I-CTKO>iv
z={ru>=n!hdswmO4Y}sA)?VtST*oNJSFWrURG|u{F@qKKpN$tzzym}%uLR)h%1<HOY
z;zf7?Mm|H|LwRC(*_;Y-u5;*X7|m6n-0v*b;{3E7tmT5;d=ClFPys6H1G3Na5?%N1
zhWwaEuj(0h()m#c5K)4^4IzQZZ(&+cPjao+^2RglidWzNo)#OICOY~)mJQRQjoR^%
zF%(PWOTYScS3vyt`g&m-BaF-;_zy*z?I~wu?rrd|Z=#G*L^6U14W4CMG>N<O_%jjd
zDNOwz36N+qi{*H!--^f40`9Bc76$)C6Ie=?-bVEYcu<#+eL*BC_t~jY5xB>COQ*=t
zUb(aAss<l6ko)KjeakxK;x$!9O9+j!>XDVjdH8NLs9YN1C3vAOKfhJx{}bb6F742z
zq<wwhACQ2wsjaH`VjJ!I-P`Kf2`H!pJSYGw0Z;u=M)z60)NWoo0CD!uN2x!|uiBjL
z!r!bAqS-I4?KW_mIjC%YR(La#GTKGywP5L%F<eX5Nm@sxi8IdOsnkI5KJ|cC?7ieZ
zw2Pblc<r&olk}=KT=(vcdcd~l>3T*<fMR2Sl4KxBL5xtK|48i!X|p@Q^SMZEg1R2t
z2rAsuNy*RTb^uY32gOz*pmxm1&^p;1ikz$v5i5JO*$BQK=yy-5fdYWf)E|L1YndT1
zX5yWp19JU&$AC0)Uco#FV1*M_2N%8K4Ee41&US+k0gt*olMs3ORqbzmd~DwWR)ho*
zWE~lNMezc5*^wgI1l^nMY1%RtK~Nl=Y)BOap`(l?SrwW2Sm#V;z+)ar@Wsv3(?App
z6?pl82dwpmj`27Tc{o;s&7l-ye|XoDLn3ow<NY<2bq40)Sh7goKx{>@NYq;P1$3nG
zch2_C^3&%?7+=XK|4_1O8>O{(@}5khIG4c*2><EgfnfN-@EZQZf;LrXv~afhUl&ND
z>CDY_@$_HpQR38_^`6I;^I2q0ao)d<-ehU#c~<k!=WLHmPA<Itqxi~ZnGu}()rC0h
zw$dR!K@wd0{X|!ne9m`3foU*A8#6n6=1ARkLWh{tOmMN9K=4WY9}peUP;uQP4crwg
zk*?uA&r_IM@)>yxs>?`@KhLiMS54vrW%*tAz5{(Q+D|j**H=%~)5(HMxg#mxAEM0@
zx4-`q?4NRfE|4KHU6<~{xeR&0(fOLXX#d85wl26xd%5cQi}2HB@^#6hrRV(~&5^94
zRLkaA7e<Wo@5pHGvuqkGxO~~&;NQ~+5zF;8gQip@#|SMr$P?lUT;3$Mo}FHna$kan
z9)N7K|6`$lIZ$j|PD(E(p9J^r<w4*enxB`DDG{JUFWs^W^sKyJnnC~5D%D@bM5np6
z;LSkei?WF9(jn8UWWD7tN6}^|ly&4m8!4J#fIT(jFIzn$6`QZ}+riJDF69yEr?UBs
zvW|3PBxqE9u6HYKm+D;0XEumcB@aL4c$Zcdm`*DMWEHgN26xUPNd@;DCd`ITFXNI-
z@D&s{rV<P9)^ddCBFau!M3*759p1{lLKhAAjsCsElMg*@<V;7%98Zx<3c{@#s?g9V
z7D)7=Dr87y`N*IBVnuqe4P>M}@*Qqdn-tH%z@f7cK!6MhEfFrW?LgkvS6NG82@n<q
zD{05$S_Iiw(eC&oJ<j&+y9V}Ybgl~9IN^0k^L>)!pd-1Blh9^pYsh*gCq|WKt@Ix>
zG&Vcr`1d|EkkUQ{Mo&HF5Wt_U1MTKV@yW#4gy?c>fLKRNT^)NsiU<KUc!|xRq?h^C
zZQlbKOfC&ev!4-6SOxiPLxCQwxHUtT&CMXBGu844Tx_R3iD{%BCyA>fdF8SbC<rN_
zj6U+i0CSKa=0R<6<HxUrhJo>l_}r_onPYdJm_F-*7R+S}AUw(8tzkRU$>?H8p0)sw
zeok3F-PCAzi-yhURNAy(kwd2x3C0A&WW(*4Hm%H5EaV$`kGuFU5hvs5RqR+8R?dv`
zLwcYLM_6hIV3(oVM0ZK?mNF1d&?#1QpE3S;!s^cB8Ug8eA&UQ%0j&q<b+yd3-)rz1
zrich*lFMXp1$@7k=b*OGrx4b$w;qpzu8Y*?Non=nd^PZgf59BWHTX5Ga8jPjXrx+J
zez-chxnd=V;ykFAJU1Cyf1_MfAF2vi>wu$WKg^Ozel^iN?bs}u2ceAyZNZvVb}ZQ$
z)!OFIT@%`+MJM<K)*z0_caKZbYjO8I;jgMq)O^L7L%m2S3q#zrmn)t#d|!s1ua#~L
zrXFc-7w8i-Li@<ypY(fzKDv}+TLG>-3_oBQb;Xpk@d4bvaZ@&|NS|LuJN^OLzC4sX
zvhO`7KOO>H*_y%|>9cC-#yZF{0M1r|{=ZE5vQ?anvl{*Lr{^<bgyTer(L?9F-dure
zF{PG|`ft0LuE+^A+YeNoa|?0rn(^c8sADtC2r>1}v{i#kzbR{-p|~mpGYHaBP0wX!
z<Fmon1UDDFH4vP$=QS5>Xj2+p68j<xoDkE;(OBp!RhTuHzrA(4DHokmjXG7KmL;ok
zp5P-WBC<DTzl?;+_;#ijxUYEmpdvO^IBZt}9=1Zz=}6_g2W4r~{i|GjJiAUBArU=Q
z_TyQ+E-*B~w#66LS~RU}6Mae$3puYzYXqk?3QUfO`Jf#s&m2p<O%56UrYJM2Jn51#
zse_Y~TF{Izx)Gt~4zBo<^~((f5<iPnMkDBWV3G=fiyo*e>68kc2=RDGn_5`&xm2`U
zn@Y!Q9B1pPL~6!3A{#m|m5rRtEf#1>?&HpSiTNrEBp=A2b<=1ayve)#89^S6HK3|R
zeTB&2QPNfhxD6!~D3|G^$b8U`=n<`OWPl-yr>Shv<by=ENC#+vb12U`DPjBYaFL9_
z!(4;RbhgNppe-Ck;R7K@zXW(geUS$$dDEzB))RTsrwcT=m_sD<Kmw0*7&AqLxV}L6
zx)|^i*^{}Emxa=#k~RF@qK^<AL7>Zi-UEjW3cg2D|E07TnT2)b{#Q<#97ZZ|+5WVT
z=M}#0d!ji+x*0*nS)l#HPT0+GGNr0n8UJW^>f)Bh)OmRkdwe5tVVeKbXebmOZcjD)
zSE$#43OI(&xt8R9ZO&Xc--@1zlh@A`IWh)2+Dbwvv>U3(C@Yd&nw1E?EJlErNFU~2
zBdx;-NUv#BUxNiPAx?v#zn9!`G;zSS0C{K-xmsbh_l&%N-+RE$&a7=R7+i)h$cm&$
z5I^Q^vgG5jss&b7g58zVUS>|IUzme`SWjXHS2GnHJWAkuv`pPvzj#{71e0ey-OvC!
zFw;n-w|~Jr_s7!GFD=xHWdMsRZtow^Vo}lma1?2HzIU(w2~a)G|5u3!90uTiThIrv
zOOAB5U92vb-$9;TKoW^+FQGU<lMAkt0%&!u?*VRmCFuWMSwo(Wvpa$mXDuS0SIz;K
zFC)Mf6`QWOx>$F(x@ns+^+&CDHs+heXpl|F9PW;Y;JJ@8rq@tarClkXjcAJZk<k+4
zqn;e2;OA@{OAb#>kycYsy%v}-F{#DIUUZvw+t6aZv_JzB9bd|Ke%>lk4Pig1!WOE)
z^?`x9#o8D_hHt}3$H4Q80GZA=!Q5%$+A+noGo?)9-@Yftxuwn~&WOwey$_RP2c`|G
zqS#)2i_OqRF$QXVM;DTvOF~qP)rUp%nb){`)GdIkV0u}AU98Y-y63fH@l_26o5IUq
zTUzLw)EbVbVcBaWxKTH|wTH|W3^@s9=c~<YNmn!3@Op0PZfAdTu`x)Fl4Ni-$;X|M
zlsi4VnLFFPdfLmDM6-oKH^w2IXw&C3|MR>i8ooOsXGKDrMqbWuL3`&FLaYezIk=+K
zEg|R}#T#mUZyJGlsl^;Y@7q~o3q=t`)KD&+W>zg>&WdotZ5NrUs^CF|f!@uuYb&YS
zM39g^RSsPOfGyHxg%=PHspc51C`+l$JHZ_8gwQ<Oj4<mw^o&jk38+5$cFzatdFD_$
z&&T=0hZA<Gtv2<q!{;j>MRqPlKBhK_kZDu_uK?-cjv=y2BFjw~8v4k5lze+aoih&&
z3u!)5N^LqN7QvSY7z%V4W7v4>XT`{YtW%R?I8zB0JY7=%sxE5shNE=cvEVQSTKP5h
zJp8r8n@s*jkp4AF$C{R^Qxj9~c>Yp*naFAfN4DpxKBi<1v4bq%X3@tSMJE~8$iL{;
zWTMpsw%EeD)KcmA&Q9hoIy7JY)-BSw%>D}-B=&yE0Djpn%E@~slH+6Au<5qlNOIbj
zGtU;siFCttkpVtx_syIPCQm(czg!j3T{!NP+jBl->&P5%U!d8&-rpb#)2j;dxK$7$
zuWyLbIN?;m7fh??XfYbs{qkg@VDFm}Ewj~NIW;8KpCy(7AI&zTQn6{xSdceY^n9aE
zXbYmx&`as!#8%S#@>`2yJt%YWx{R2qq(QA{(amf)U{#_@;0eO&S$r`4$$=#jJViEC
zw(nUtcJra-0}{h}usSgvLV6<_o~Xzm1*3r(4fLVsrO+E(mYHqIz!{NcQx0wwu-&nZ
z9cK3ll9U1L346jXc#Ipmo14>ATQZ$_$Wi>#>3M%4RkH<H<$ah^PpkhEyk*u`AlA76
zrc#{UX|4(8`m)uAG|u%`FrnC|G@1APZ>kEG3?$?Tw~5L%mL&QX;bM3$-bX=u%-$}C
zOEOgtj+BJhK`%)Ke|?b7e?(pRN=X4i(^nhMp&I}4LBzZd)XYOLrQebiGaRBCA-3p)
ztF&K4%M39%u9xb%#EbwK)etIFCS%A4)f6fZlff-q=W=435TsXN&J!s|io8BD3!u3;
zY^LQT|3bG1Yary#y*IR&vgp4^ii%(x?juEk0sj8U^6!g=#is>@r;c${v8G5TK=O>W
zRzwsdU9l7P%+7?0(|m8Ox6@`jvACz-6z?KUO7^fRSme$b1qvGfQ4v8RoA7-Y82T9r
zMU=Z`=+Vzl@eNQE(>!k-3H63`CRLD~atI)BbIk^w%S^c$x-_(<5An!;6GUMAgc#UA
zD14R_Te)tYo$+-s+VeJ!5Mgk`-N7bb!01WpS1~|rz|ghlN@`qJ4_PqV!(rV!Qco*B
z?{beaxT0MwFX}pWK<b1oxOvjV?9kDu)@Kg^n{xxTn4jRK3}%S(5+T|rZV2bR4<N2U
zK~H~fr+k=x6|mF(Fc*;IJ5uy&P)u*$HcbZG@V0Di3NVo*y)K(TB4L2~Bwc|@X9=Jn
zxj^dtgY`95tU*Uq8aUZ?k&Fhor;-#q@4A%I#|f!qPU(Tr_^J~8{Uw>^abva_^1AB|
zlaumdsz-tT(>srFEQhzeJQLSoJ+rs1W7MK44Jkh(L|YR}ve4ALVQo%_<3!(3MIPqr
znaxcyMZ+3prhxE}JF-JI%yi7i0R=vhpA|<k)0IVf*$&ZiLZs!Ti4-V_(E5D^W>aBK
zuzi+WAgR&uTB37WL56pr;!L2TVM$rppd}|`Z*iT&AxV^O8XcC&q|m&sg{lwc)w4_I
z1q+ywyaBT=WX}9)`M|)NpFg;?bVSgP=l-=qA!Ns*mW8NP2wc(d9xzfrNW-~FR3k79
zS6|;*_BJ|6E~6vIjh?^MKU>J-r2;+e%SrSNpnrD2x@}4^PxzK_x!z2#)cP{ndvOZ{
z`uzKjCwBP2;f$zEnJVzD$a^{Z(RAan@typWOKYxg#X&3ajUlfJQ7sxnBC5E)9~p}v
zW}NngL1F?lP033QtIs9+**Wta&)n%LF=<H$_R8Qk>2Jry2*KVKI8dHtTY510#1(Qn
za*uNXD8T-swA0IC+sknw;KD!-lK%f?qASfN2F{)08AaO)v-I5#Xezb&zOft6)O)0>
z$HQB4OnX>K?`JK?15AP5-}*?mig23m5W=28v7Ux+dyAwqDli9?vqpUw7=gc{hapR9
zu@Lq=vqz-3x@rN*)Bs6U!!Yl+0li5^u#kWW!K(kBU^d})o|jAhV3Q|^A=rWHy{yw%
z=LKW7lx}tMO%3-}zcAKf7^F+#$9UCq58vDB)A1luN!Pj=`gY<?*G>60_R_1$bf`{4
zc&Xc7WfcRpFS53Z<a^oT-il=m1E2>c81v4+eH5hj7Q1xug9#O$p(1s0(A?KyGd#bC
z1%>;LgX`K~U3_U>?JIOD`Wy+D21L{-*IGNusfW11mh4z^D-OvFXV~cog~$V1S#;Ze
z;QY-%NE3UdxG&I`^VaRaGW|;ai6TyO50#M6zc$A2s!~1JE#+sF#s&_h2E8qr`invH
zQ<dIyw5gJ92;}wH2|F)pi)uayoD<FY#WH{2VF1MOHbwJZ$rM?dfe4ARjSw>gQ(MXM
z3IXpU7;}S%5H!#DFh6#yT!gItPOw)MSyYR#<4h)wZ7)){&UXUDz0RgwWb1uNsq^gw
z2I9wd#*q{)LfK4@`~?Sr(4F@H<JsnhmuQmN;w97ven^jLZ_GaAb*xgE;%c5PboiUk
z(52tQ;yXAgKX)kRoY{vQeq^XtZq<Q$xAG#@;k%G7T<z?7`;FzruZ4chKMJt<B@a!x
z$g2an^i&YiV~=2I%9&cIEzG{MPPRCM#b>jh`oCY-WSXY^{KGh&^5B>v7huogu5lDx
z(w-opP6o-~|GUUhd868KB{|@9tvIId(^;OwR6<M^)Y}cD^HF8tBN?~Y3e{T0Jmu*&
zY^Go)7g{<YEk<c@0e{Q>QJ$fDBS*3U0F|+Bi1hsjM3WYB1{MUTt;Sk@%ouR`Svfp_
z3{(9jgIKfnmo!-w^cO<H`zXsu84cYw{~Nj(>m%88<CS80X7Ou<_FpGp*NEKGZ!Fs{
z<xwRK`7OPj5O$%qTdf_LNvi>!j_x?Lq1*7dj_P$9sBDd#Dld(&=Zyv^K-}!H;%|v7
z#P3BP<n9TiASS{Amlkb=|J~<N7GDmYDt~SOHZCBF|MT>;O<`*MULfC*4aj%20EJl>
znK#iHN{SgNVO-tU=wFF4ws8L#tX1?0nM3~QUBLb$V7NTa{dIcYquvsf47p>oB>izd
zVqf|4l?9~@jEw%a%xW+@j>mL$F4kRvm$QCAyW0lp663ox89d-n%Of#bm8x~1%m205
zyL5#w=|Oi~tkMl<wrOH2)<eJ2?tQZ2s)F)jndlQ2(LPs&3!ln<>5_C+kd;9rOBjm6
zv8`z7=tq|Gmkdt5vhkzl;?XLirLq^TvnA@SGP{+QkBf&6_IP3U2hjcIBc?WrB+QDc
zs~YH`Gun~Qb|cz!({ENv*j@{k%o?BZU4n12sZR)YX>F}zc$MDS4g|e{qK13NWxPK{
z6Vg9D){fgpAAt}4dv_?H8oE~ovO6T+Vj2XDWW(hR=!GMbb`H{HND0J*f(JB8deg!Z
zD83M4USX&ZJE7h-K;24Qy+n!>o?kVDl}O4G%~dBXep>!nFvR5}WUku7I|Nl8{9}^c
zGZN@toXe<9cN9yDFhTbXKw~cc!jkwdA!4AMB)06mBbold*GnzB?Ku|bthjee90Mv%
zOS6%Qj5sbDD+h~_*os4@X+X14d!+`n?LPn#!Gu%z%Jj2+A3*Q+VFej{pI{yZd0Dw=
zlmNMfl~h6Hzp2yF{{0TpY&{)fxj{aEX)-LdTYF+=uhR_G3kVH&6~H%U7`d%;)o*v^
zp)9@+(lNj{$(QJ$tp}DH#MXPCA%-F%JM)4dgRQ9wRp>%}z|{b=t}vH9ew08g;aHK4
z!!Oe75pXW}gNkP<R9!6-^?bmokAG}5DV&wlaPpL$pWjCVNe*ua&NnE@Zpu^CIydIZ
ze3wgI(g1&?QsnrGjc;}O@XTJq{a*Hpt23$;7fk}s{IuaS%6Mi3(wgME3|%9|R3k+>
zHie(RV<~Z+_fqCy{d=}JgdRm$SUCBrG+(aj7U?E{x03DH_tueQ4%kUGUP~Uabl~_r
zVVN36bInHbU&Xvaoh@>}q!Aqy*Jc=gGGXSn)Ao&q?&Kv75q%a&Ie|2Vuv@tPf^?t)
za4E;5qUZf(8lVzFM>j*gN!YmDV>x;ugg62Cr<BjT-aj&Z|20Fp8KA-xib`AkD}3O;
zf9`+n=w4X3&n4hcwSl3~IvVB(SUpuDZRV5RjNBQiA?=~mG4q>d{YjHNJm#6pTfm(O
zAU=h#Hzp6h;$=3!m;$rs`IkBj`In|<f0FM@hFe;hD)4Ljs^Wm1<Rf?6{Jy&Pk~@*P
zB`mmBh7E{T<3$I$k0DBBq*QXAFHhZQzo~6U4ydfBb`7-^=x`P$KGo26WfS(3&QJp<
zM$&dge*YNKW>emI!gTh&^3m;duikdJ_E$qGLx7bw4&!b`TdOcX+Dstdf$@gkJIgq&
zvV+gx1<0QNJPv*t@kya`evuJhdgG^<xuXa^c{Fb&dDfP|F|GK1Q|YnCnZGuuhMy9<
z9lBur2e;MMMtX$>D#ukj;N1gZ_iDH7>J>v7A=d7zv_8>rMjO_Dnm}Z=^5he`QC^}_
z83=wU*};fI|8;-JiX;*B?(oImW|r8*;8z0WW@P^Ryj>x^pDm^esl2jeMP%HWkDkR_
zcMTRU34wqLw@fn0)j+#5reuEivm7cBj<2W(+g^_fjUc{yFBf0Oq%#sxh&N0}h7LkG
zz2br*&-dd}pFWve%QNIP*C}04h2raN7z#4hnt*o7rJXj?cT&<LU5>rv#q1B-iEHfs
zH>F5Tv-!!nf}mqvTL5kn_i&a&kQq!-X1FdwX2qF(Z`n?WFzw7B#8wE|({W{ci5GwR
z(fr7K_J^+@!ZxL<@bRj4{B?FXY|%NM#fHGrYqr~0=tafmU4}RFn_1;XC1^>4GHO*d
zZF3{PS1=KBT@AC*Vv18H#|E>#|8`QzVyA5F{w-c5%pvh6VFzGU)0cF}U0j8CO+2(8
zwGeJETD<hO3Sbu$=sQO3WrCRp)HFhra}0lbT3K$kxqHG-Aq(x3V${u8#KJb&_YCW4
zY<w}4z8cmE?pp`CC<TSJME!y0Uu6CPZJ9A;03@Do4~7ka;DkD8n`xMl>m^nrb^=F!
zcT?u92H?=gyYH+isRBDRd?z;NDro2ARQMM^?&z7L?R_|hF~H)tMZbs4ISk!a(GjIF
zWx+o<d?$swT8c8KFy)jYzap|!TJdDyv1g7!g}SEZc$Tw78851AHuI;P`~zx&x^*ro
z9_QEFh4@^btaDwow8i?@a8QgnI@P&Xv<oAAd%eRgmy*1>FL`k!tFt^cE5*3HW-Kdt
zoj=js#Axrnso0(tdcVO<6pBVsxMK<%`iN<0st!ZRXKwPX9&~7VZ*Pnc&uVi#H5Y5i
zmj^=;e<9EiVk?Sz5bqMNNF+Djj!8jkB*7D}(-#F@6z+0RKW=&yG#IyL*#`$K37`#U
zl0=WmW2iHV>U!y@Q2It6$Pi3ja41H(+NZ12**g?$jX~J)>w^d6-TQ>l6R_WTR@S_}
zAnBW;YAU_uY}(6cP3f*xW)MG@G8aWEG`}iBwFr})=5n|1fVq~vXTmMXHQ^Ho+c=!$
z-MaN0emR?M>1{h<oIO1|$#i}jNvz<>Sv^jMR3#VmRvsZ`dOM&9jax#_y*V_u<3w)N
zY?7-qMkNDVHS!W6mG|3u#|>PP`rcoXd+{3g^AzbPHqpUQ?N_6$pwAKL+b>Lj4R%cx
zWGQ3lCBg~mPd@Lajs+p7uK@uYNs;fp!JT=up=>b)J}THl3YE(FLngx%S(W}QU6Lh=
z2G%e7>{Mduuk449x)hAN>6CuEwWvHqJpfKPRP-<%GZt5-*iYVt_Jx!ESwS~*#dnyr
z>@BN>GjY@2_<NO`%_i<<aWoqT?+8f^wdJyoiP2BBG5rzPD`m-?;WeplClwU*Yh^W4
ztZ&Gs4lEq<?Wsy3tYd}U>&<!=8k?P2j49R$-7&9-A|XDWG?-|~^)>081Ud&c$t!EG
zok){yRU1=Y8UotXVR+*+v~;n~swlsquStiG^7E2a>t=^$^}m-RNyM9<DS5vf>Ajo6
zla>H%Wv*!jt_AEW`#iQ@W)95zKfeBS@?C1apLqba=%ys{szyyl$nBROM{=1}>)W`r
zJAM8{3M2hYp{uv9_4pq90!B!xj*@klR(p@5>_sQKw*r~RK9uL|#t)aebf-gt<9Yt&
zYSbMA;(%=}C{Mc%8=xGP9p}C>Y-9Qb1+^vcfa!npPy!&Dt*B&mmtz0|N(3=qa*r;;
zf%==2R8j&XTXubazzpV<3h5Nkgr&4e?CJ_ABHgj{zMs2sYyrg{Bl}gj-Frhp)*q7x
z)0HA+X4=ZWJdh{*s#KS|i@cn}rAB&}`BOx$tLUOx?hPFfgAyf@aV$irBBb+!(+?T9
z4O89>$Gkg!)Z0eiSKXr3W2DQVQf5dQFzEO$Lgmc&Ia*meWxZ%UWFAIWyOAi&oMPwU
z<vcSyyrL_6Q&bYFI$!p8V^TaL#1Zaf>T<J$)fS|9<-sN>p5=o>7fHB|WA&r-Gc#2<
zdnm^NoB}<NIap6FAW$YhXJ*Ax!KbUljT%7ingpE%L#gE14zb)jg;w{#Ty&e`nfF1D
zU%X8uo+WFTYi*evoSZ6Nn0dn?WdZdC-bPp-D;na$fYk=b@Xg$Sjhw0{vVxG-o9Q0A
zb#eJ<{{~n8oZoPmVex11<wkLmbBtf_NDZ`|hrvzBly4v_86Pc!qNP`(fxZ&PrZlG0
z&!-EOS~;emdG=^Jok)6RpiLSAe&QFO)?NPM_y-GRc9X0~5&zBTuzLUR%5-etxW2A6
z9->our}pOSjNb|0&rMRlZQ~RB)~+Cj4{|5pQj|T60DB_Lx93)Ina1QIIryZr;P)O@
zW8%1<1-QM&tCcZ>pWchV;9;Ti<tJt@=s?0kII%!y&qLM#Qls*f=FY*<Q;5xyS+^X?
z@5NGGj{cW##DEOh(O}3zyaZX1yc5@g9V<N_{t~8&9&K8x)@}MbO29(MBka|mvD{$V
zbaPUej-cz!P(Ky`!`G1gz*(0@dfZU`6{kA4+0@yB!|=<TRWV5|_H8vTEk@ek1cbS-
zT6!`-zusn<!jo}IdgD^~x#*1Fj_Q7Dos{%utIVna`g3qmJTw4NaFYM=1LJn8maYnm
zAOgqgw`d+g7s+LGmvu8fkhz@7J`xm^Hj(@i!U8nsa%rBfwEop+69U3;3vU+yd<DOE
z+hD%;XmslzkOB;JXQc&njV+LaH>;U-1!*_GD4ZM_%D$ry`nFpq7w()orqf0hT&15T
zi(KR}0KWU&CpYxcpi&bRVY~D*PxNL$JU&Pie!>Yv!*S^%j_Ik?;LhnZM&IeY74R3|
z=cL;Z4@dPim#V5Bfv~-he*fc&p4tFsUCSWyk(@AZY|7l!p?v#T@S{W6G0aH5wxLPQ
zxV3_7^<m+|u3FH?VHujqHbdnlJ8SdR5r1a$Kals~Y&A=;fvzUhDt5TaP7Okbg06s5
zl$l1YDf3S)mpINid>-o0wXq)*7atP~*hTS-aletjW|Qpn7#@9Q$2g##B({IM>NmhE
zim|dD{92Su6l_H2WEO8RnB{!~2hGlGGm5={KL_o#$Rmp~zLiDksxgbONhmnP5~0u&
z@Si0QGMmTKm$@$E8ZPYT4k+N$jg!D<S0Ngi4Ao9>B4FC*m1@2KI?5*DF13yE->0l)
z#;Y7sgCRvVH#6yEl1_*}qC$0BKuT-_MytZ!>gpETw>5YKNlu8~ks}p=ltOq*83VLi
zkr38Rjv`GFT)|^LO>aPXp<DgO2^qcJG96iDGqO0WJIQJJ6(aa*0c9TH(D`RVaEnAB
z(U%qBM--od_Xmbtn`4F!By2$k0e2oUBCCfSWBCEsHE+^SUr$M($CJqZy(&jC8NTlf
znX~)Ee#*i!Tk15G(kA5GQ>}qJ8!{(8&6hZ)vjIit=?^eg4tM1;+bvG$b*-9m|Ixs)
z@9HpT29^H+30Uu`Hf7(<PBFtwAf-`Nq=kKwH#ADT?H@1>B${@HjgG~idZh+<4A$W{
zd=79JgpiNVH@Bbqod01w+jmv0`<U7(Ah>KvWlET#TJQ0E)DZ7Gp&r0YK^hoAai9FC
zf2UGgu<Plox}0=fOm^?gOa5_pWWY8|Cf&@LJ^C)P0^ek-F@H$%abB@@qG0L~Pyi3@
z#s9Yzg#ds7hh`xi&l<0|5`MCh{wK&yh!D3qb$1o_3}B4LVpdNvaL5rrb`t}g0*OG$
z@f&|x4{5Hqq+dQ3;lwq!j4f`<lMbm?=BQ7%k%7ZsIXPG)C-@rC(^5MTaLPwcsK-f)
zP0QK<aoyq1LFKymoLN-&HF*!S1k|N3gvT#H<ocxrajNtT;o}0`6&LS~VBaZ+oN}qx
z=X&rI?toLdh;s0h^|Wr93KmD6#{4bG92qxd!^w#XrO0l(Jx5b#9_PnJAs$~=<2xik
z>0i^-lC+4M^1{+l_G*3|ql#(#z@7S5T6T3w+Fv3|_w-$U#EGU>(e?wDCMp1GF~Fkn
z4TxC6F7;!kcO`svr1+-5DycJ?PT$1SGad{~s0|+BUFX#TE`y#oS8QIk8Oj{rX}t~n
zxl}n?I0z6y<qRmlao<OP<yj==<fz5;(#h+^AZSkHu{DCp_DEuirCkA4<?mC@`;fg{
zjMF%Vg_exUMs|(Ag*+AMU`<q=Dt!@JF7{5MAkt*ZI8gJ}^1j>6)1uVRftJoKouPyB
zDe5b<k?i(e=}$fi6N1e8W~xs^wUMiZt|jypzPin!$cy7$RoAW@OwgNfB60E=rQ7x(
z(L3HJm#KJc_qKqXe(-mItw95zh?D{sgYVx7zb7m9UCzzIBlp=}^V*-NA`HIXjEZ}|
zn`eU#yey2fvrEr3FUT$f50KA!A<%^|7Nuo&G_AB@ig0IlhL^BKN504lO;3itn?FH3
zA~eNcIGs)HZ&9kyR2aCZNiQ{%oz(pY#1pD~?oiu^av4s+OpJ<>`%~7+Gq3aXNCN@b
zPf@f8V@aARD*4RNF@$t_QR)iidnx};;_A$>M5_pulFMq_0^W;B?Bex_N^!Y`Ze53b
z%89ZfyJ|;GxyzJ$OcoExAxC*zfp~o1kMEIJC#C~ef*-$b!^xG0O=RswQQK>Dkn~OY
zc#NLKJOireNkzcXtKUz}pbA!Sv&6Ca$af23MpPFzkhSl)$K4C`COcXhHT;e(?<V-}
z14CZ^+f`o9HDAgBi)UPBP!)z@hIbEhyj6Ps>vSnNb(;LRJ<a<c4>P;Yg=;joBSlE#
zbJCsCXK6|~<q$u}26F_CfcncmK+@2lcF1|20VMe_z`sD?vE=1mF!(=;AyQ;FyJlcW
zsQ#vmkk{v1cETI3%N!=tl<o6!Lfre0LdpMG2lU?y_1#5#6Aiuh=fiMxPyc=KZ$}Ts
zEb8fSR|lr+nh!ArWsly`vtJZ@!6r=0gGv%Ma7FQH=8t^a^G8FAv8v4SfOx2KTU2U3
z>JM%oddF;Jvv?a}&nu~z`0+rHNL_^@MJ<V>uO@?AYafxkb~khdL;Njv3w%T*+_fKH
zlU!VGH2>FPJ)T9buifQ(hs)#xm0*#{8|n8}^^Kt&yd~&jZDqc2i=_Ufg2i=lrzk}c
zs%K}9)6_-DsS#;^rI3v`i9$+}-V#~(o}ZpUCo-rl)*aCysqDXP$~4*HKZVSmmd6OT
zzRSQE`&lW^wV8QyDDLXN`d4Uy=-`FwJn?H5Y34EK(i`8DPc5YkzGS2Q<hOl26y^@T
zBrA0kFQy7ZyBwdSkv%hg-p&^}HGQ|IU93YNq};vU`+j6yT<dLK*v2{%7FN2Heq@Dn
zU4s#8*qAa;O-|QNPKZi&%dvb%W6|wJH&>_Ck&wJ+8TrR5+9)kO%hcPQCTNHg@*_9#
zlWKUf%Zwle7LJygq!dvSFR!MkCp?HGS+MuGm%yU04Q<jwDh?LgOvn`A{Sd#x{d2!t
z>{s*r)OjkqZO4L^Pk4aDL4Q(lWZgR!)_FCrs*V8B`qjVet8QppuxP%3KOGfFE#a2~
zyo`y%o2A>A%4ny`#%ctS(BTh&NQXd%xT(QQwCWSz5Fwgh%oUj6D+*hrq0+FVN@=33
z#4KYkCOgWex3>9Y)u*_EbV4H~M+w+;lU1(7G)k>n1iSeckTwT&RFdruA*h^IZ^ANS
zM}ZDX0!sz2Dp9eL*(qH5@8;6&p|<dmo7%km2BI+xJ@oug&tLqjm)F8x1(@?l$A4ac
zghGe+L^eLI2tD{&v<Z-3g7s)rOA(dt9*4-Pcik)HreBHfa(f#93F`dQj!;Fa0s4Ux
zjkMEgw2u<Ak(b&+V-KkW$gT$_ib8~Z#R(N<AnNQ={)tL^8<{QAvbEUVNCKrmPGr($
z=v><(-u&w&)Or18U55nzA>MJ#^178ZHD~7iT-5esrygj>{6R>%ky`X$wKXkBnfo__
zk+t5vNukXa&4*ltHOd!Y+d8-eL7=e4tteNLj<nS@h@{JP(^VB79CHNI!6ciVb%ePa
zG8i(m0Cu(@y_Y%BInn>%C>UG{tO~6!6K<wQmQ%L?@@{#*$>rNuK$*EH>9WgV)WI_L
zimVX@!8RhGa(Ijqp`e+#6^wpRG>%~wjUM~SlZKXowoL*D4o=E!Ut_8q4UXo=HX>VF
zDo5D+8JXTKTkT{?GYe_Hb|$94jND`NWZkx-`9zR_-NVj>V{yLQK_QjvOBkuBoI=h#
zD3B+DS|Nx~K|04|Z@j*;68V-`Mg{b{peNDLbq&vht5}%0>(rotQpn&(V^yEu;V56h
z6#~=MgF=E&&-f?W4<#xfaVW{#N;JxZRJPAYiek~rAK(+StN|Kp2<{~;<-UVr(@3Sc
zs>wQPBtN;$?YIYrXg^`tf<C?eF1lH5e%Q$bUtL_n;8#N@Bc)KO*T<F8Yr`5tc+VnQ
zgvFukb_$;jDtep5kDSu%epdrh+GO{v2Qu1-ija<lHZBL89<kb<E=)mEeeM7e)Ucy4
zURw$0{WT98;_hVF)^nc>4UzG-VXX<H0mEx@+&ny34>gczjp20#bY)cLkl?)bfWX2u
zM>FKQI8{9k<4mVP5R^ib=TP@W5HslL<Va1Hw&v1q$)E(6ljKNKJWgJ`pZr+NiS2f0
z2%{iluQRjCw{jwf0v!ik4Bddm&O+qUMTtC0w8{WYs%3EH7op{xTfwQhpUxPK>LFdE
z`=-9a&)&)An2_8)8;YZa=7h8;KtsPI$vLaVmTEX6UB>n`X5@;p8$ClaWIm@+t4v&Q
zI--p*Y572d4p*=xu&>kh$h0}Y=?@0tKo6)!IrBIvc1r!D=EY<RR=apu{dU0~?CSnu
zNkb$rEztIlM1o<hbU<_J^uR|iZ@k(l{@r*uf`H~YR#YwQ!|MjDV3NTa%rW;+QZk*E
zsWz8fCoxaGL9k8Ox9}RrADX6%3M?XFFE)*4`B%3>Ee2~9<Gshb9n5an&C345G@iU%
z6D~g#`Pnf)s71A3RrVJOX;IJjt2ai79O(wf!0B>sQ=4cw$?0^+uYqG&-QMqur^Dkg
zC3Ran@3z@$bE#K|w7TADhVENyBMeTZbXjfhj7GaJCiLTjyIYWW`iQ4dICLi^{K<AB
z`A_-OL3ZTG>a{|bY7x4T#UVUIKG+N>+59ptYt;l$6w}wSVEriLw8b3ef`T&l%a;Ww
z73zJhKc+k}YhlRzio!V7I$uWEZ5Nw(Mb17#B_<>wjf34i!<fE5VR)K<x@Y-jD_4cS
zN6I=A1^OyaO{mD=2yU*F^BR>8`_6tEJ~Z?0G@!mhs0!(-&Xw2vmLLBVsaw{0`}w<%
zEq*m6`nYGei(<^XC>Sdnz5)Y-y>W7xrH1aNMA)ym_A=|$Ue7vK+Rn!MQ!IZ|>5uM7
z^Icg~(J4;{EDrki4nlwks>$_0b#+kD2Tf{zVg4j&VjOb2z|t;UB@xmR^?X!$dF*U=
zAyw$s%Nc7(g$R`s&G#KJ5=(F(k}{WGL;bSYk>x&EMQ=u`!K9OOYSk1p0C**?0GIku
z;eKh9^`OsNTChIdI4#&b|Az!@dyRgonNBex32f{g5i&|`)8!KBzLC;29clLlZ;1`3
zBo((n)ssEWF6QvQuwaWWMUkPR>e(<(6~PHP)UPq-=HJN~KX^9FjrlSjR}0%Isz(My
z-x|S@#MHFVDLmtvg6utv{94n@?*AB9=Lp-q=hQ^-$T=|)JL~IT;PI#%nCDv^pV`zy
zu`qC7$M2Z*QagR(lOZg>GeJ7~w5}A0ZpM`QGZ(@Xf&Av;smnq}uRFE<s33SaR{Fht
zvuA_FhRN`ldmu7rYY=Yp58gWdfe%)kd8x*yM*|@9tUrAt5P!hi1^tVFYC`LBM%(I;
z_-m`csbn6Pb6%W~fXMS3cLv?i!sY|tI)+4fKsYSGZcY~1v2m#~g3$gQwABLgyM=N7
zJ)H2x8dt+mZbK1H$h<T_=MkmhS!1+qvvl@sR_Z1>6Zka({`Ov94f}4vmaa8}<&&BB
zlTsPTe7YHkf9T;C*EsxT@@w0ox+}YNfhye*6oO#-Z@*}tUVQ0o<9X?$$$y(SvNRl+
ze%hn2{GqU~q;9aUV9MAUgUDz2*-ekGag<uY*fKrBOTs35FcO0VA9LY{l8(#}k4r%C
zZY0|5EA#Dem!1u1VoTPF+t8WlcwF1;#}#r#VNp@0RA;gshX9XKSSEP*s?K@x_#lS!
zBYSmG$U?5NX)q!!BW%j|jeo!=rdl~c8ofj3&cTxHH?`XWTQ2DPz3jq0DWf9-T6C%M
z?J%E0J>wh#M+Rt?giQ`-kPF#`z6wc>2Iu52ePWgfQ2Bdr*a>_wt76Do%#B}Y8b`;1
z=Eu!HAZUhh;l16)!)y2->T-dUS>Cm+HO>b)j;?lz-A<b%has~91y?z*X32!*w7fD0
zYSPWT0DQp{oYMhy<=X8Q0^9<Qgp0LK3sds}X1?uIf#vDU=%l@ghoLdAv4a7ZAn+yQ
z!7-NabII4-1uClFYsl51#WZDJ<#Iz%W(&C>6l24M-iO{kSTr;Dwc*}kMwGOr<Uzbk
z6b0Zm%~?A(`-$*;V2t`}^I0}P-VGp)2W=xA7~21+B>b_Xk}vnTV6k$fj9cn=RodQx
ztJLD1Hi3BP$RUxOwgj$m#zSJM2tntOCv!%)Ixx*VXv@0-kj?cLjIU%GAk_`HYDC5W
z60!Ar`Z}}1RUXas*QtXE#?%X6oJsMv=VZOl{ubX>C%m3yYaln2xfi*GGtqQ{UH=GM
zArP<iZ<QH}2(aF6Hi=H6ZoL_ll-H7h>||=~Sy~J1bY7RE`MC(a5aovDS_uj$XJ83~
znLvW2H$^tO1QQ@C_KApzAJdU(7UndfS#!{jCyf5p1eDxLMHTSb2VGnhKQc5*14N&z
z9Ay<TJ+a$TzLC63&39Nx6fkWq7elIn)o`bEr+So2#jPWSs4bNVvg>krTP;CHLj2H*
z;|+jWAa?dx4<I97impHGK>r~7@F(VcliH0|Xiz%IqPbiFF7Q$--T+ay@UqDIW-R&s
z7;ti>acc)Cuv37R_y5U=&SDEjd-KRT#|MLNq;wsWP|%KaTZ<LIlm)7Nod}q*T+RLj
zs9AK@BWV8Tb>%N`{{anfglRq9YyCS^cz~E?vBjD}!lCke;}a^h14X6gO%{VC`@C&r
z0o$h;vA0*7wLC6z@(-c2!YxRK$kHMcrbX^JQz1#AG4EgdfCv%)1<7hu*raSLJspV`
z2tWsY_Wgb^mVh1j@@zVO2`iLYEQ9R!%KRQQvQ}(XH;&BGPEJhcjJPp$7v@Khxj6D(
z(d7FGFKPJ;C_m%K?z2t&lAXAsMa5({@Ccb_;P(gGp6I2_+{fsNNxrbp#*fDK9r@O6
zim6H>=%}XG8_0L`TvCEv^YbbJigE>V20w%ye6!d^v=X7OYORA1eYL~55V~a~i#?x_
zmkers+Wh?aF1Na{#pI+j0tFU6EB(4A`a+Eeb(@su&O@896CAC8qv838H!FYr-)eiy
zpg5v#{daH)4ueaOpn(J!+y)44gS)%CCIm~632uRbzyx=KLx4bn2L^Xb@Zc`N^6&SY
zQ|ErTw{F$_aKBDhclA{D+I#J1t>?EK^=}dJdpJ8z1;GIXxn43RvbIl*PNfT7X@#2e
z=_76lw_<N2jA-+kg8f}tT_{)8I_v<7;=QvGxF*P<(Klc-WcFj!qv;&ZP4e7}MG5Dl
zM1;Yz@?{Or@DfkdLISx^nbEs9-ATpHKjuxMC<UzZvvh4fdVTNXX9-$S@>YXX7-C0c
z7@tq(6X9(j<oGH<orFq|Hpy`n?XeMO&=`aw+DvP!mLbHzIHf2#lWm-kRy?#pY)saq
z0%@&)K0n?Ro)|K-rB39GY@?6CSwtp(>XG6^4_zukFI`i3IiKj6?J|vXt;#?4MhDxX
znZ}C89eX~wLOHc_ICNY>&)h2nhV{T-pOk-W&}g}5xog6k`2K|8{hN`EUHuH<pHjZY
zxg+OE@$;cJ!w*?u)xn>BoO!*5L1gKRRS5NuM-KqqEw&(-9wt^~l-{KZvM9(>g1n=j
zS&=#Bzn~ZgJA_WwU6(jSL^}t<(DJxCcR`|qCE)J$h!2di+eO|fDYM}$!y(#TILT1J
zf5y4YNrF68R0=^~Lr`yrlhEfiddqTreH@iJy#y4s|7xrgxa~RAsr;MsvbBGC+{NIr
z9{{8O1Kj{E>J~`{p0@x}Ix*}LV4DB`@O$O2o7aZ{sj^qqIj85r!2@EKNCL|nXxAQ4
zPfM1$<Nl+oun7P5){FmI=Jo>AHUH}ZSUl)~9{7oe1OV9%jHVrX>;y6WU%$^<_FA|6
znA?~f1y+wl7)~FhT$A-xiLcafvOoAObE3=m2iWEY*hl6q_q(C=nRgr2%*=;Va1Y;U
z7%=%YTFPbgu$6)^qviz}7F&wh*5$=VGWy>{KMH*6A*FPCy2BnQ&B~O?wjA+Pj<`@4
zKgKBVvrpJD_T8=NVXegH)%5|9S{SompH?NXR3&;Eb>8V7%HI|59v>SWrWg&aWVZxR
zkBU!ME{T=AGnS9gBJ9(77Vj+OVoIl$pt#F>$m+~y8xE>?)v)x2c#S6%e_2hS>f@-i
z1MATp{{CQ}znrjW#puCFQi6=oS7PJ1vbSK2e;_=}wnGoVEr3^TEJakq^p_D(Kml$a
z_rZ?Gj51~-6iMc-qg6RTw)WOmme6ddbCxYdQ(575_Kis1Yf`Fg!P}=B=S7qdzXv85
zLY8-ApX^GXF81E_EiWHQpt^<Qk$Ho<Auq0t>{$UZvWB=nb*U3I3>bd&8J8V9*`KwY
zZ<qN7zi21Foz0IP9Ue@oQQ9?2T9<IX5cfBqG!!fQzO^!fb3JswJF-~PV->?H#K|Ts
z7;oV494P1D_2`kA9wr|`df(l3s+1f*Z>50{kYVp5x?FMPjK5BjG>BqE!K=>7_lJ<`
zU?mIZ2KAI$Oowpwp*)?#J`H=1$wB_9PAs}pPPkN)&Ezu&911d&=$xl2!zlk`R&3Vh
zWf9@vE_4Edk*c+(qkIE6EBn^w5}Tyj-8IATF<s`9R*qP^z-20DUHAp6L@J|u%A>HI
zIwWavF^T@^K2Un${Ux1IF#pFAcjdX`2z#0GZ?l)1$urrb!2^ABI7R^8C<uNYiN*mz
zS8yIjLljEb*`x_6yyfCne8x##f-9~=3#GZJrju=+Sb|$Q2>L2uP&KhTfmWcR5Kv9f
zIzmv{rhL2TksL~hqGxOby~7uVV`5R})JNdzO6}s)f?oC%i=<4$a5~<Oa)_85-vO<t
zBDDoF^}3I*Rb-K2cBS`MsYwR_cP?)dhW3z}f4->t-xmR=qH5TfGw0)w*5AWtAf^SN
z8&0aI@a~z+PGdUg|MQJCuLlA?xUMB=xi)w-<k#LshfDL<-*p(|vdypphlCc<eM8nA
znhL0S(i}@wzpQ1*XMrEKV91+m6pISpdZBF!bbRH`%jF45>8E*NX3<J?#t2PK-rA6j
zhGJ*%c3UaWCs;!ACFoSA7K6I#mNxS<^>))mj;t^HF3wT<`SIZ^{S~;HZi)VHD_e$2
zI@515r^lp#RGGA8P8Ogrv_S2)aNvv|iDf-KD7?lsk(616Ct~Ga@;6IJ&i)hM&rqFH
z_d{4my!TXu#s2TEC#FsY;o*#j17plg$m;U`+uTb>d)o?h#LQ$O`=#WI4bU<R)5T2E
z9(PL*c7MB1PgBD`5UWOg=71p`)@d)nYdM(<+CMLvzy{1>kSQ8@RNm%UK(T&Dru&c8
zK7zECQ^v?hW1oSZDqG?*<;gDd?7s>g3rlSHln{IzEAHfkO7AoL*YttNOg+Rc^^iWh
ziUgajcdg`UM(2F=wrdIn=}#Zh<=~i;qASc#V?@)1OJ%B3)bK8=H!%_aKuaK-Hv6_E
zLZWXgL<d0Y-}eNKm@@1*H&0!gugO$Vmt`$mt&+12=sU@;r>ofyyay?IBNVs)mLF7(
z+e%NPvAh^Q;@bLVmIjCFA=FRAwln!xqB+r@WM9QUgQaAhL)nD%b2sQ`x8AW+pFL<`
z@D<>Ax=gk}D|4rwr9pC(z~uZ-&!MsBX+tP{v3@674o!&(iHL~Ri21w~?|lPSvu1t`
zGC#`Q*zP_xL6X)ImTV4xA#8e}+tA6Gm0dk#PQgva59o*aU|S#q8fLCz!)`dTUE;T1
z-fj>};=1WL5L$Y>F~~GBmp8;bilZy5oDg9>R51s|T2JGpKmqmvicV(&@kv_>bZDom
z>^BX`yD8G_v`M%;nzCZZ6toWmAp=i-!oj1C^91^x7^Qm&(AP-)E|7T_c+t+j5zr_|
z8xjNJaR|ktVfcfw{14>FbaOlB{V*?+GgtLSdFrwqr7WBRKFHkA^2>HFIYY=^#nMce
zvJy>Hh0`aMA;)(vvYHY|t1l%OaC$B0VZX#;)HAJxmc5(4O)i<ppZ(1}e6GB?fmoWA
z6bvl8XK0$b#O2F3Zmu&_UT^hJam0)FP^3%ZC}#;5d;v)!y9qJ3e#5;OxsYv_xVTU!
z{+?1y1w)x;=sd$PbAaRNc9n%llKZ`Uu)C+!^w*GumNYNx_roG-K<8!Q2-ykt@0ZfY
zr!XvK-TsfH#x7V*%AW33+d2#^GQi1Fp;7z*gr4)2yBWvQREL;))dCX8^=5+i;stlT
zV#+}CDgNK*tqxjjWJ_`PVicfnL;&;cQ&Y*;Ihhtp71oAC_i_0z%8a($ob<u+Jm;YY
zwrS>4y@uXFpKIQZ`K_@F1jnV8%tMtjvAwi%Mppr1ggN`_nxtW%c#l6rL+`q!gy@S7
z@YBs`EZ=yvMuNaTzv=%rrWkUavOX$=%0G4-$z0%rZt@0CCz}cn)pzXYL(mL1{%Yuu
zdI%7E_Sf`-cJ<VrX}-;nw8F<s-mxN-g)C8tQQwa12zu1|CdVY1pi<>Ibd5$8$KN-x
zjE;M2#@}An1o*KT!(^CDg5r;;N1Z90ooADL1(5ez$eORzWVn-}(O?Bv4Sz_SseXs>
zmqV?~>ZzsGw&o!j5w<f8@{sv#DOtrJ$U_Dj(b;)-NkmH`7n0`bWqIR1%_UE7OOxEl
z>I(OIjl5&SjS{qFf;X}t`cus4!+vA)nrl6Oy{L=oL5@^;jW@i=CdjkT6XAqo`6+2%
z{~L9U(CZ&c_w6EQpEH}qxzfAh!G^WzIFAyh`F4v79?qYW{GQ3xs*X<~X0b-9NZT_h
zdWQw~P4u*LSC^7*sH<)y6-qUPhn*98u8h*_DyX=st;;pao=47t-?Of5eyU6Qk+<~B
zjS6wvNwpadv*cFYj9guta=2*zaF<NLG}`)t-bevFvTFJK%MsS7rn7QlY24mkSfgd2
zF3QyYY%OM*#5L9QllJ^Up%*J=AQ*KM|Gqt;tD{M0el-j)qk}PfEF{BN_@XKvk2F0H
zC1nhZIXmno`t_cDYiNiXiFKaNC*n}(j7O~t#ckVD+JKoY9}a3Guys}~*~TkILLMp5
z!nDFWKG#hl=B?YeC@}aIkxzb37e{+F4w*!_kr1+;bifE~E(gm=-@2_wkfW_44SorS
za&m9oXrWRJ-LdT%8+%c`)K6aR<Y@?~7Txz}=YF|FI%KBx$V?k55MdH2UArnDxHE;<
zK5-P?g=q@R5bP)=zcF_(_nG!zAq>7_r(;iX;!kX{*dP%5yr(0$M6;gAjW65JiNuik
z^bA8j*@ydbcgdM{C-0(rWOp(%8CQeVzXWZpntD5VKxPHVw<AMuvmO>P6-M{)Hde!}
z3THD~ktxn1HW4w}icTbU<8-s(Qgc^c|3FFAf&aCfOd#Ihnfxpx0j5r}N6FP*^#Fu@
z@PFLUkMm)F*SoG~K#u#29H~Dj{F{}Qm%NPNV~wXBQ!5Ij_3=1XZL2|JYfuK9|A)Jn
zmWTmy+VDSqYiod*4WxR%^gv;5GWVyz;EANe7$6zo0f_HFJ^-As&-=mvA9gd-V^HnG
zC6E;S-(TW?A6NA&wp5z3OS;)N|37G3&U#6MTrm!Wqm0hIBqR4836A6p^MOeW^?FrJ
zPz8Hm?3ijcHF1a>%h!)iL%;ERJ3D@*>qe-mmnm)$ZXzr)hS~i94!*lMe-z9~Gbh0!
zQpBe$w}BpQUq?@SZa34?zU9e4R<P~90bi+_ky_U{v97tZPoYN2$g=9x&oSm<rpeD9
zL4ClE?vp6Rr5I>CTlT7(shIy1?qtYdZ7*FI%+Ogt!{!??pn6428015UJ&umC^78Mm
zTl_!8bnRLC87b|apJp2Ho+-KPi!N~f-Vp+S$08P$ZcM?6z1Q-*`0(bM_kNEl+VYHi
zT5y!tq{K8cF>gUVPwc7RS<GXUx!D&#A!WPdrNaAa6cvKZ{MBHCz!CLsu0ETZqxd6Q
z9IO4EA>s7FQ27ND-uL3M?H*sg$82f_r!Q0DOAO+xvn-^7ty>@HH)lpbqSO@3!q2><
zrC<)N9*@ctQt$Fg;rj-?8uarU63%>CU`tsuA90(}f|=`uA3b032T$2dJ7R9(<!=6J
zYy94g=b8o#8g^D<FHr;ZutGjxS+*Or;EJ=Cz?eo_;!Y{?I~al6)S8!Qg~SJb4g_|T
zRE$X-K6%9SxS-9-oAhpaM3+nFC?~0{_b6qC_Z-5;1m6f_3ex|PgP{x?bnC0j<9)mu
zfya6ndv~j|#Hl|904Zm?Ub8e-D;@i4pjEy7T`?uf6ET~`^3<!X?EKtO6iN8$DUN}R
z>WXYEHL>c18xVNWXwBmAjbq|s2j9ZMw>;_Y0nga%t?Azi?Of9*a#6<8%C^)BSwM|o
z+WO<so4PKlaysMugcNvWrU{d^vnA7`G=}32SwCrz2>CCLENKu>+xmOJJ@D&jS`PzE
zjMAz5>x9nR9}l=|8`q%kfJ`Q@5mbGsfm9d2(_W2WJiQZ5y6F7xz#IQsT{K5t`Kk@g
zdz$EiHodFCqiY8!%zS&#w9sOCJ`(mzE?nSV7P&{2dCJ?t3~$yNEo_;sca!9)?0b!w
zYRakLrE*SH^0pCeCd^?aY>Xd>CVZLyeF%_D|IbsB4wir!kut}hfxvQ~yFX{T9xuET
z-hT{Ey88o!Uk(7wksNrq;Cw`V1%eVFBmjdm{$C$haj!sqQhg$&0A%#HSfDq$4$|5t
zux*4GK;fnyFlnQ~M$qvh2UJ^_)Zz6if!42Qf{*U=Lhf{|ZE5#zRf=|)efRZSj`n!)
z1cP3^aa=6b$C;mfzCr{npRw3$%J#9-QRWs=rPOI-IS<x%$SW|+&|cQXovg0k_(q?G
zy$vOIu%fCwDEk9G>8S=58Ly}u8iTZJspF$)h?f*s1y{@DcHnj-s9?X$t3eATH*R>@
z-hrB9OVEoSCpT}qCBvl-qWErh4|r5JKGxiN*NaxD9{9I{W~<oT!JpF@2hc0bgbgWJ
zv?y<~QGH#_$2&?riR{xx{&!+!uMw;z6Drc8AEgQ&GZwYR^_ls3$3}VdLjXG)!*?`7
z-K1j}?o`qd+!w~K@Vt_2Yaai%q_tNIxw$)2H_{=^DrLv;O2d2Yt|h@|LaqG_M`-Bp
zsvGNU8(S6Xx8GO2;>N+yXc_Jp%b_}h$Ni}=yoD>x*QlXQon!3!m-I38digIlpm^J)
z-C{+Cm+0X%bHxPRd}!O@Jh}Yei~4aWKu?FyZw>n5nMovwi!mTz!N}J$;ZVykRi{+w
zFTN0+V~T#E94SIYu`wo9%%P3%7NVcO5V9zywKKfr?llWQI=s(+E?Uhad+x}v`<a&D
zS9^s{vX-iqNDUUd;9|JY1ra6mwar)Dzlw}AMFAo)KNLvm#LAAa^$Y@DO1@Yx=Up#X
z%{@#o-0kz$Vjd-AhrB8j&>MN#fv*6bx3in(WmFkZrS3oYUd-=}kZ19rqgaMvp`-%U
zxnJ)a2BTp}>Y}Ha)L#kF$px>cyt2nFdD#aN{@97dJQ7qAh*U@0F7|AP#_lSBXi72q
z&<RczIYEH){0}*KH7~l9=kGbVUoC-FMRTo0FCQf{<NU&TV{-_xw?z}+s4eX2x**T_
z2STF*VD`vq%Nw&tfPsQc|8;}^>t2KTW@2q(PUnBG8_H|-cM%bN&HZ(Q3tVP8`Pw1&
z4<swrbwb*8gTTnfmlg{0x|rSu0L6*`1<jdm!!mlVrc~q+UF08#9O}z#xM3JWceLjl
z_M0{AR7g^Ws*$T<Yfm5-h&2VzNA~cxQtMOv{`>XCGsW9(rHg8&8`YTmdzpt@Ks3^A
zAV-)ccv2R`=K#B8m*s;?*`rhgRLy*7z;c=Cq~gklsxdA>q@Nd^iif@UC9u0{Rs+7c
zY0HzZZ^UJ`IsWG@bk=7#w&lWp?4JN#2U-8~PwZ}T){mq8d5Iq9UC1TtO^4}i4*=yx
z2gkGc>9*3SLB5|sE(ge4gU0`bT<=Wl_c0Sp(^#;jOMRlg%XsW|E8e+7U5$tOBx4h~
zL-H$Xj%qd9^)GI#y>RsN*&<AG*sy$-@7ZK5AdJh4G}4g^#uYQQTSPHv`19ZTRhL(?
zD*uTOH97hDs19U8*Gc77f95zCxSv(Fkp<elvz650WYECPN%h~IOb^G>W+QCN!Np_y
zjkekuxHk9xqectWbu8g#)D~vwcazD4g$xy+!j`lY<3Ld&*G|^x4VExn!1Cl3fdl!T
zfY_Jfk>1gcB=|eUa_5M^4CYAMGCn%|nHx`jJ5vkdn<_g5+(N$heRkKJE0c~!lCBb6
z_vZt<<|(bl-*&%WRFgoYe?3^CT$%s6#Mx0v7VE|^m!p-{lr!#FVd)_*i`<)s^MS0w
z)Mmi%!06VYm5G5LzWxW<4J%A4R^y^Zuru{s7ssrw0N>c$CpDL=7UCySds2P25LZf%
zKLoVBiF=!RU)PVO?5@5)xP&lNu8C<%?JmW*4DQgEj!#DSff%G40ge#%;kWEEHrqV$
zadf3oUwM)s^oWms`)``B4);xQbLK<vLG``1G{ybiaJ$ri@5S8c@i0us#k_a62vCaV
z#PaxCYbbO1b)5n00oVHkS0NBuQn8C15*Ek6%Fx>HQWZl#g+{fdfttQ3EtUE@E<uZs
zaL@kEm@z)^Gr!pEKLHLc+!P|uZJvVp=8>P$TD{SKBfb_sux9p}OLEel`7^P$S}G2N
zAg*nY;<L9722F4=2YO_R!l)^nlcv0Ipe^eWJX7;2YkH%}P%b28xFnLlQQsIc^W>b5
zi)cHvKerQF<t|73th<jny{r%G1KgP!sSYI+1=?7hkT%za?2!tiPACxjo*?uIf)m^%
zRA|;@o(Kx%q-7i@xcH3JX?m4W^bVg4v{KGbR^7@NS-H|P*w0DxEEjYMo^%qfcrz-x
zo_rMYv8R%A^P8Uackmu>k?Fg&;A*RQ7eOBFsWsUK4EDCyhvF289x+ko!UF0i4?Zgf
zXX-?P#=Pk8A0^upcD%TU`dhL1eTf@uW<=bFU6AljY?n>R?J|X830<KlEAAW^r={LS
zI&B}rLTXCta>Sp;z_S4XNMJ~qalRa^C<cCBB&93lt6nGa5lw7f%vR1YfY#(wMimeK
zd)fCGAp9u3uOzkU`TNv9O4g>Gq*aZI>=IMyfmI_h@5gR*z4sMjZR_F#N|aYzk`tHK
zcA`hFpOmeBKGWM~^IVs1+SewbFto?r^-b((s$&@W0}mwO$?xb&3N-oiQbDLoK$$gG
zu<en8hP}zExuaV!s~UEl@y+vR$}5xi`8j7jzTRH15C4*Vj)Gb{3Z%bR&;7{iOZ*vv
zm#4M<+LvG=Gte_DqSc@)LX<XBqUS-uaoW|^QiC+?hjD-hf1^^KtFC&xh)LzMZ*Q``
zFT0G7X8t;c&zLmIv$QeW4*h;TtNsxw%t9mG^qf?v!_#XaJMOBQddAl5DS9|*4*Xmu
zma0Iw%Ro0$5%*TdDqafTYn3tIN%H7X?;nU_w|VkUrij;XGbKv4?n9%I3Hq35OO~xr
zYTB7;?0uS#4^gmY3FI-x2i?@yx+z+<;FR$7i}nNz63Sy84>uTw*#2z6&~puR@&Kya
z=~);J6Gt(~eSlIA3&>F#Xb(oUL`%X$Sfxb=I@PH#rpK9$;?D=F6*m$~B4SG!Twdv5
zD}a;*<MX&8Vl9Y6Y#12TybW2IUKX)Btl-(L*Y3E(VwX3H4b##9(&cVmbUrmtgfgMV
z3v9!uSKv<f!KWwQCIhwwOt>F@(3b67XJn7Qx!hQ4WPCAwK6b2(vTi)BkEGV&WnM6E
z!!EGxicrSQpsw5g)D&ZaEg%nvEB)mn1Z#*n(Tv}O5FRpy#dX&l+KTpR=Pb+PvKB+f
zu?JZqmW%xz&vF4C?!t~l3`-jKjtrZq4fl)Z5z3Vb_|CXun=o{pTNs*Fw7>{9Cpq5-
zcA1`A9id9JYbWraMmy2KSqX$=>H^6pIDA2o!#@qfz}-&Ug`!#qQ1DK?0TU~gb_TIR
z?*Yh=y^5wuK21LsJBvlh=e9#AQ6mSvq4wJHVcN;QvSY=)Fq%Jo*L6^>Zr~F}m0i=R
z<^FQiURrcUzVweTwTkXrZee(j`r|f9cyq+Je)%TDo}+DhN1lYSbe;M+#WpXK7OUlR
z;ZVz%r!z%XO;7>jxTjndlY@c&0FzrNJ|yL+)JjCPFNekG`;&cL4oUo+^r4JU7mc>=
z2WLXc>Zid)MjK`g-sb-A5sTfd7Fxc0uBzU9&gg2x@0;GWoaRxyo|9hsl)VsV<Kv9l
zx7L7@AVLydv~z5)Zl^R2s|AdP=-sQw@SKP%maDGu4JL?n@BO_D1K2fx?o`S@7f!C&
z($=#=gPHB;_khMa0|yjrqcsnDI%a5wJwto_K_@&64s9Jdq^#l?qpLVOn(K8l&RhY>
zxTy*e#FkuMTGN-+cWQc;x#fY6=RVFzhe$<ngj)P_zTV$x2s=j7jh4QLdW;Ni)62?7
z90`NFRsQ=vhihKS62qL{hCZ6bv*hqc{Z;2%Hxv8R9;)ZT0Z0Ai7<^D^FD>*cwNQGV
zW|>b9<?uc+GS|Xvql#NS3zPm{G$_^cC9r1z7hZxb;`~=EcY=L@5pf507(O+duQ)E=
z+Q7z*sZ6o+NP#(Fq15`ou2I9Q;ISC{C$GJ!r3__KYT}ZI=!o@MKDz}>-Dq-sbly%*
z6$Pgx5B6KKP5%aok28ynX>zXQpCGY4r7DEo#xcKq-xkJl?IBUnH+wxUu<Yn^;-Y;{
zzm0g7R%jjSjFA|URE<tv`*i!G*DkatZK_zg^bCn3=na3Li~W69J#~Z?y~Z?QRI~{D
zf|a}T3%3f<b_F^+W}-K&o?+Qn8At9{$9xc2f@{<X{W0mW>03sZXEI==aAin@hvC{j
z0j4pCr7`{|#D+GwI*!R*w1BGH(%)IeW7VivQXKZb$>9xz^GrdybiB_eTI4uWx-B?K
z$=TOX2CcuXbZDQ_UzXOI;p0ol;B~XLzGGU`>2J4uP83<5RVP{HS3)79tL$W}&|X&Y
zk&;yHI$4*qr}Jd8fiebTvEJe2V0c2MNFe^}^WA&wilI`~PS~TKQ6RIc63tJ5%On}K
zt87>9fyczB)#++vRC6*Sql)LSEUcE8OY3V?Y9+&sslbllzZR_+E3WjSj1a(?)XRIP
z?+l4pjL7(<Vx5ZZ(ui>Tuo&JZz?3yJjp0c#9Ps5)i=wsPIlI2$^4^I)un6;~ueRBS
z(|`#mG@qfe;^Z%#-Qnar9jUantz`bVF_(j;ICTR@&l;Wr#Gda=$oytWMv<6HZNR{R
zF6l#$DNSLI&_>RJ=$KFjGdmEMZp}QMhUx2d41Mw`eWu})w0v9xknJ3fh(7+}fR|sS
zxP=Eoerj&-qK7LgmsXtjISj|yNZKM0Qe6owLuf;lB!tUohfoHKq|%2xsZp7bmC%@C
z=fPnDw=;Uq2iF3g4BBea(G~SMK`N<RVp%9CV7oq~J)7T)NPKc*GLkypi+>dv%s*RB
z=6>TT>n>-HIH5jN5|y%OPYY(~a3w@wDNus|@r1DiY(M@EkInX3&mmNBsOq>$rA*!Z
zVtye;ZX@FFoT^xMb!m_@3tFw&$U^|}SDF4M%4rVLGW`pnEBCWg4u6WN7g__n-k3d<
zho|==zv-?BwlUbJb#pyZE5Ad{YmT<p<^3XrT2VS<=8g0RRO&y_NN5G~*rtNbYH=a5
z%hdifI_D~Y<Nb@2KL5ZL<rs`@=@}crMdRYVfLdf_fECZ{DFv$GHD5iNv;GLO9zK2p
zlU-Jg$(Q&N2(&^f*qj=AW(QHu*hP&*p9rHv`ch7?gP7R9>YA;D2A!W6|CF;#S+%K6
z{en`WER;u#GU>(C)Zpo>lPiK@`2IMz(pbp50G%66e4y<j?L}eA?er6NS8^7oTffqV
zvh@hU7eG&XNkig~h!oFaJ-%*UklVsiJ+dF~*iauo-ij7egEs8-MLAvxK}gH+C_mr{
zPLpL7Y|bUxNSTc&uoo+4l?e!^?F|$n$QRK}$L+Kc#AWDnHtAQ>C9QcEd%7AKCT_*Q
z1g3Tqfq>$!Tg&yw<ny@DDvO9$=^1C*@mLxY2VO~&$Y-fk@crtr{liV<-*z_obqN4P
zTxS0RWmhe)BWJLEp-7ub($&0Y)~gYewqFN(@S`18mtnZ^1bX9yqoB)@7~cs=8XOc)
z$~9EmT-zd)JKhKl?(sv5r3gEtL?CK0+mIF(%9TI_tT-YOuOt5Iks0?6|HQb_p)tQ`
zr4OrQ%FqubLl2@x;&#>9=zeM&2&r$z&H6Ep*$#fKUFWTgwKPC~r{|WtQ;bdv3&Bqs
zBpipa;C=2P(4}1oIK$b3(wEVA1XZR)G&*F5;&|tAQJ<=(wu?wkP31L)C-cgCImy{&
zMt`j(BKUZXEPA~YZoN=c7#7mp^hJ6&3gtH`ws+(++}PHqKRvo65g+C?`;PU8f96Us
z^L14-H83J(dBIKonc>GKyMAu8#j5`KB8Xa7erA43qfVMoBslo;a~*sLF>iBONt!!d
z*ZT=}0Tx%<SFYP<LXCkZ{!oH!*7bpbzDzs%SOKEdnz2@qN{l66nw~WH3s3(8>BV3G
zthuQG1NxN|OL{BabX&yW=*DRZN(84Eq8_d4C5Tu#If`_668yIGCw)=a>a~!!p`FSC
zgPGUR3)#95v|R%Owg}$@)UkPrffC!NpJ~R+F239z3-0$pd!dzOlNUGD%tnVB_}_Y}
zbp#!f7FvPT)I26J!zz~Wva{moYo_SjLrI@*>*=Rl*ZQAa9^Na8GxsQrx_tJQ8W%Cj
zz2?)riJFbSh?>P`8?UZKyw$w)056uK%zN|zrR;rQkLiU?Ex_9s=)rG}6X~IEazYd+
z2fg2CEM{QLx4He+!JhD=_-oB^b%2ar;<KT@In%R6>BTZa(p1}#DbJ^@LuT01U$DWc
zX|7FvC@xt&Y&=GmbzD=qX9qIL&g(7Xa<sNInyqS4@H|dMn0dw?U*rq^tec+6NK`!<
z#4i5X9oR*1;ZCPLh1S__{40MYIE%Wf(?ZRa*JSUB@QPcF9Eoptd7FXL`JoQhhrTb%
zif>gCHKjsmm)hQVag)vfjq-e7@=qc8iPZ=@;k!0)&hp1L>z}>3Ga5^YA(Lbbtw27%
z0q8aJXelR1#8P96Q8&kN{$zMtAZF-dhTy*XE5Id5KgrJ;1tqqKqWEZrbuEg_RD6(5
zpKo!j27MTg?6@qs7(!t%kNOberOrceCz_~P83^4>YH-&Pq~1x?7eytQJ_PW1v|o^D
z4YE_sC($LWN-Be`ZhlLqR|m8;5_D5HinXL9Jkoz+Cmk>3!c?O^G4Z&NdA4WAjqY$<
zR>UkKSFoLs=x`901O#+yaIv}K^DTTEK`tfp<K^Mh0w7gxx2T>@K^Bs;sCVF6{oy3F
zl9S*oB=*ZO*~=29(G)pbe`f4(7M?yjL1^zhS36M%j5&dN=um?{F`d9Vsf^;+3rMum
z){c}jB?}fmL8{>QeGTq_Q{_yro4<7iHiH(yCD#xp4PSKyAKy7d&{J>JQ+91_GyGSL
zye%G0vq6Cg8+qP(I?|CRfcP25CZjqr`^|}e^(&W~@67Tx+5#V;&g>GUuI*iP6Eh>Y
zis_<FB6rp{<lO91MCcVSO7?|NoLnsf*Y`LGT)BZs-N(h=%)e?A^*hGjf~%Dmrp@MU
z2d^(rY|K_p`{UBfwbG#M(41j?)fV+qTIb<7%w?ZB9wZ!()(!3y-St(jCU`J7;IoRo
zyc@VRa9Czi&6cQVAJ|yT?P0|(3jZSzqHA`l7w0hW?m}e7<J}Y=uPvY{1<e)Y2$_LI
z89>#$O66k)<~V;v>G5lY+Wfos%?)1i*EShW7J**fJYI>GRX?&Y3H~&-PU$y=3t`sq
z?)1`qX@yh!E`bm^(CMAoj$I*%_E=kttwgr(wtr!ib@Yt%t?l_feOo0#!K2&vl~q&s
z&z#`ZUML=-6)b*XKZDuiC8Uw9p7tLuJ_k6MU6D!4#7=PV8{5)%c0SfP?f9%ghB{*l
zqBsM!sHET2_xmDkPDhd?-k}x5S2340sk>hYO~@4`Q|)(OeGV@6E=ag4rMKEhmAU~B
zx?1o|9iH(WQ$4oz`-G`#Bo1(do>Fqwe|66bWt4y5$(io(uC_q$a(Vgrkve*DXt0U=
zQ{GkZeB-$aE8{`$wCUe+w@O+{@#+;seuEa<t9W}Oc!H2bW17#hD&1Ugu@6G%VrgN~
z$?j{$rzrIHB%vkcyF>+2;o|9?ODCCi`}|{0Ap=8FA}smkI1fh@bEYPjoK14{#{m=8
zgktv!K`5QD!8&-8aP?b;H$Tv-pRAXYz)x04BCdy7m*o&qKaP1n{qfabd5%uZ`nCXv
z?n)A_jbf0|&SF8zRW1yJDlPF+Qe}A@JEO|-<x_{|-!L)+$!|sHs!*=@xOpxVzvedb
z)jYHB*knnpC%r#VSk!P+Rnl#4c0bbs7X^rO-Lrg_wv^O=vU%>l6Cb*r5H9l`C`oLo
zZE|_D<W=<Qo29_qqF6pcfNFGK@hyx68dS~8EO4+IMKK{DWebh@IqBX_e~5B=aBg6x
z@l{`1hP_-;mhA&U;uBq@-d**9*q1R98%Odk;BvI;j2EC)*^AE+NbHn-jir5cLy~G{
zOI3N28Yi9s6I*ISnIL>?bw7%CV^RZSeOfdJPJF%Zd{eYlonzm{(<1_<%19x0*JV5O
z-pcY-1J(X-8UM^*6>Js$VZv-)!6!e|oc4I4ynJd7xe}}Ju%j6E;6R|};_fhS9@nbX
z#|UQ7?$opP(-32x@_oK~%c~M3Lj$vwLwl|){hK>rj}<B!Au@*)g#@HuY55p9O-SBH
zWuw@rMLW@1Sit)GU?~sfe!LF4@*MA>iOCy6f6)XP@;_3}^Noq$Rxw-RSYm<JO^3qZ
zE9HRN+u7a2V{@u{p%7Nm;33~s=RLd~A-LEAcLW@_M1?*H-tu|Z*X6o^4a<fp^a0CE
zf5_wKPL=bOxA9^T<sK!3n(@+IR}yg;SueVIH1OAo4>*6E8?d};f!oH}8Zd4^uErs4
zZyoF3C)F$#v8yn~mCDHk+M@$1Fcw=K+Zz_{Bq+ZMQayFv*W|5_JRlWv;VaP8ZhnHB
za@|g`*7Kfg_9c}#rv{Jpo={3vc9RhpP({thv~rYuzu#lr8nA6V0lb8zF8RkqVdAUb
za;jfC@>lK8(yD7bJM4A-)3*5t-Vs0UJ={vu3ePwg^}f{;1A~2|XWnR(yNI1188JU%
zN+?)&RP}DA8Mm3uFuyMsM0|)>pFAq|f?8sv89=6~$oXabu{C%SK2qv%Eb~n8=Bg05
zNt6k}gp0IK8JS7C<n-X%_UVJmI+>eV-X}eyCUMFKE-EwxT3$EW`MRdU$<JAz3=Ob{
z+h^R(qZet9sXD#-ZTF_@6L-z320#DONhepCwA&<&UDC(JvK3Y1$n!p+QZ>-lhBEuZ
zN#co7Ez8fhOpC2Xn~Kb9asfX$VHos+88a%0mZ#)cF?noPqh&<sb<t_+5}PR8QBQeG
zRHw!d+`QGFd<}m1J9h1AF)PP*Rn1!0Df2En=UM~4-C<l&lT@%`UWXP5)(Q~Hj$Hm*
zu{Jz7x}3?D2BVF#W1moe3hXuJtBt>%K*bbzskN%B6D-T5N`_wC1?owbO)>TzE;uXN
zI@IU#0(4*C8m))Qw#G8z&Spf;;j?VjFLMKs*eA|ejMH#sYZ`iOpjHhAG0R~=l1$Lz
zP$D!CTsZ~4sT*BzFIQq?e3yy5oI6e@RuIzdTud~1yX1YtvGN32VA)Enq&|tFX2>L;
zCKsB$^q;aCVi!<o`9_o9BNK<OZ3(A>+4VmVcB-FfOp|Px%GOBtneTtXsq_?_QIA$N
zM)uO29Z>Nb;NxjO{PFPG()RUo=cF`HHgemQV#8Jj?6C+Tj9MY_NFHGn>_qJM)HE=<
zFODQd_w%w8w<^=9RQW{w6@-w|ijXaWgkd0D%C&wH09KRWVjKKtvr3<Bfkf_^wPy1c
z2#9z@P0vX|P`><ulKHCT-y%51jczz3r}xM8viW`I!1PIp3J_W8lu|R7YgfXrxsKZ&
zKY3;$<}3^X6IKzykL;I16RIK(LiX;2uRbuiDA3QDh}DS24C}MW2!XPNT2{g3Oe<hz
zb;I8V{!8y@Y$H^F_4%PAa|zH^OqwE9=7oC54|Q~fZ1fYnDnP{!70h|Yr=s4>W5yqu
z0O~uHn^3ppo2b9DKFehXmQ(lk#X)b)Wox!qkA6LO9un%i$zM<WTqZbB`m{0TZpCc+
zr-IS>{_w)lrhu91nb@Aub|n?cW}!+ydIuq>>oXpm1_@WR>U3exABPE0`mDHy^?0XY
zZYI)EOE-oil&_R!YmVifUEJA(wc>2)6QZ1xZ|a^bb+0DCcdWzxr9wG@I9aFMvx;l<
zhh%2@XvL&$CjIJEVK1__)#7{BeR)~_c?bs%$^%9RN=Rs-XT?QX!A4t0gOS)=Lqnrn
zJ!#(|xNcJvX6W9NmCWZ%0OP6@?`+yCN-Nc@8lAggJD-&8G=7}nNi3d86G6~ZIHofc
z;3?GA=V`#OzN1B$gtq`gVF}{Zyk9x`1|sc?e#QTWhh>fgLwYS5E%{(t-td{&HQCC<
z0R@?NKGHWtL7VV1%LH*Gs^$}&-@^tl1z;%aS@~wwii@C*+q;5ra+;!nVh~-dB-{&v
z-+Nl)#Zz<A!RS!xOmcWmTb9B)qX;eJVBi!ahj=HcLZ2x;{kbt4PHV_eA#CQ5twNfj
z-p@s#U?xQUbaNb6kpq&2Q;m*}1+eg(lS(NjTPb>hiV^eRK8iUmK2ZUa;1rs@?Q?cN
zRk+6~%qV^N!o`lH<rhU@5e9I65w|Nyj&l#0(^@+VoOxCF>z?T!=xNZ~3@tz!PHMTG
zy|@zFd-DpT*zTqXD1IR<i=9-~LSk-rZf6WOz$ht{b4G~_e~F?P$^oS<9$3!9?jpj$
zfyfP=fc3|HJevM+{hnpPjwpS5gBnSgPYCjihDy@zQ=FU1+Zr9%xQ_Pk!qOTEBJ>Tz
z`BhD`)qIS07bH$K$oktP^@uco3#zBhy<u=CpnauCJ)=fY$DrW{wGOKJswSW6slCxe
zU7eDHy{or$Olz06<o}-!o}#}m`SDM_%T}#<QErTlqO-Dc=89ZO1bntp=ouA`s;eKs
zWj8#9rgSl#&xB<JW4DxBKubjV7$Me?RPnPyPfC2^4@T7ZIiyhR8g?ky?4ENIL8yEu
zFr7%%ttg@oq;I3LIjKWpXj(v^fXg5}E-;EL3|em|CJOfp3ImRa3@GO#rS%toagwh4
ze2#h@^QP6j^_KHB6U>LjqgmT3b?Zh22Z~j)UWMonDh;dSl$Ua`oy)RKhE-PlRPg}d
zAplCI>>T=Y?ztLB3;ZnnZ#fz2xwQ+P)(BT@{KH;@ryOTTS3s3X?gj)8fdM$rqi9v0
z3+mWzm9Vr}G;zl>azc%C0(8?)m7h=^yt$;=z;4?3)1^B3U#Adsg|>lfR1k6RM`uZ`
z0JBtCp!L#6z%=)DQq~l=28j;rWkEP)&O{mdOs9VywvE5PO>l*xuavBL-E5IXI!%%k
z7UWa|gw`j>)KQK`hj`yU&i_eJgLNBeL)@wSdYnHUE(pzPEwZ@4m7Ru#RllUYe?g&x
NDO8^5MDuUq{{WD+B%1&L

diff --git a/dist/human-nobundle.cjs b/dist/human-nobundle.cjs
index 056381bd..fd88beb6 100644
--- a/dist/human-nobundle.cjs
+++ b/dist/human-nobundle.cjs
@@ -3886,11 +3886,13 @@ var require_ssrnet = __commonJS((exports2) => {
     return models2.gender;
   }
   async function predict(image, config2) {
-    frame += 1;
-    if (frame >= config2.face.age.skipFrames) {
+    if (frame > config2.face.age.skipFrames) {
       frame = 0;
-      return last;
+    } else {
+      frame += 1;
     }
+    if (frame === 0)
+      return last;
     let enhance;
     if (image instanceof tf2.Tensor) {
       const resize = tf2.image.resizeBilinear(image, [config2.face.age.inputSize, config2.face.age.inputSize], false);
@@ -3909,7 +3911,11 @@ var require_ssrnet = __commonJS((exports2) => {
     if (config2.face.gender.enabled) {
       const genderT = await models2.gender.predict(enhance);
       const data = await genderT.data();
-      obj.gender = Math.trunc(100 * data[0]) < 50 ? "female" : "male";
+      const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;
+      if (confidence > config2.face.gender.minConfidence) {
+        obj.gender = data[0] <= 0.5 ? "female" : "male";
+        obj.confidence = confidence;
+      }
       tf2.dispose(genderT);
     }
     tf2.dispose(enhance);
@@ -5077,6 +5083,7 @@ var require_config = __commonJS((exports2) => {
       },
       gender: {
         enabled: true,
+        minConfidence: 0.8,
         modelPath: "../models/ssrnet-gender/imdb/model.json"
       },
       emotion: {
@@ -5093,7 +5100,7 @@ var require_config = __commonJS((exports2) => {
       modelPath: "../models/posenet/model.json",
       inputResolution: 257,
       outputStride: 16,
-      maxDetections: 5,
+      maxDetections: 10,
       scoreThreshold: 0.7,
       nmsRadius: 20
     },
@@ -5105,7 +5112,7 @@ var require_config = __commonJS((exports2) => {
       iouThreshold: 0.3,
       scoreThreshold: 0.7,
       enlargeFactor: 1.65,
-      maxHands: 2,
+      maxHands: 10,
       detector: {
         anchors: "../models/handdetect/anchors.json",
         modelPath: "../models/handdetect/model.json"
@@ -5256,11 +5263,6 @@ async function detect(input, userConfig) {
       await tf.setBackend(config.backend);
       await tf.ready();
     }
-    let savedWebglPackDepthwiseConvFlag;
-    if (tf.getBackend() === "webgl") {
-      savedWebglPackDepthwiseConvFlag = tf.env().get("WEBGL_PACK_DEPTHWISECONV");
-      tf.env().set("WEBGL_PACK_DEPTHWISECONV", true);
-    }
     if (config.face.enabled && !models.facemesh)
       models.facemesh = await facemesh.load(config.face);
     if (config.body.enabled && !models.posenet)
@@ -5311,13 +5313,13 @@ async function detect(input, userConfig) {
           annotations: face.annotations,
           age: ssrData.age,
           gender: ssrData.gender,
+          agConfidence: ssrData.confidence,
           emotion: emotionData,
           iris: iris !== 0 ? Math.trunc(100 * 11.7 / iris) / 100 : 0
         });
       }
       tf.engine().endScope();
     }
-    tf.env().set("WEBGL_PACK_DEPTHWISECONV", savedWebglPackDepthwiseConvFlag);
     perf.total = Object.values(perf).reduce((a, b) => a + b);
     resolve({face: faceRes, body: poseRes, hand: handRes, performance: perf});
   });
diff --git a/dist/human-nobundle.cjs.map b/dist/human-nobundle.cjs.map
index efa922c1..afd3b629 100644
--- a/dist/human-nobundle.cjs.map
+++ b/dist/human-nobundle.cjs.map
@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../src/facemesh/blazeface.js", "../src/facemesh/keypoints.js", "../src/facemesh/box.js", "../src/facemesh/util.js", "../src/facemesh/pipeline.js", "../src/facemesh/uvcoords.js", "../src/facemesh/triangulation.js", "../src/facemesh/facemesh.js", "../src/ssrnet/ssrnet.js", "../src/emotion/emotion.js", "../src/posenet/modelBase.js", "../src/posenet/modelMobileNet.js", "../src/posenet/heapSort.js", "../src/posenet/buildParts.js", "../src/posenet/keypoints.js", "../src/posenet/vectors.js", "../src/posenet/decodePose.js", "../src/posenet/decodeMultiple.js", "../src/posenet/util.js", "../src/posenet/modelPoseNet.js", "../src/posenet/posenet.js", "../src/handpose/box.js", "../src/handpose/handdetector.js", "../src/handpose/keypoints.js", "../src/handpose/util.js", "../src/handpose/pipeline.js", "../src/handpose/handpose.js", "../src/config.js", "../src/index.js"],
-  "sourcesContent": ["const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.age.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    obj.gender = Math.trunc(100 * data[0]) < 50 ? 'female' : 'male';\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "export default {\n  backend: 'webgl',\n  console: true,\n  face: {\n    enabled: true, // refers to detector, but since all other face modules rely on detector, it should be a global\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be blazeface-front or blazeface-back\n      inputSize: 256, // fixed value: 128 for front and tfhub and 256 for back\n      maxFaces: 10, // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10, // how many frames to go without running the bounding box detector\n      minConfidence: 0.5, // threshold for discarding a prediction\n      iouThreshold: 0.3, // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7, // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192, // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3, // empiric tuning\n      inputSize: 64, // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json',\n      inputSize: 64, // fixed value\n      skipFrames: 10,\n    },\n    gender: {\n      enabled: true,\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64, // fixed value\n      minConfidence: 0.5,\n      skipFrames: 10,\n      useGrayscale: true,\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257, // fixed value\n    outputStride: 16, // fixed value\n    maxDetections: 5,\n    scoreThreshold: 0.7,\n    nmsRadius: 20,\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256, // fixed value\n    skipFrames: 10,\n    minConfidence: 0.5,\n    iouThreshold: 0.3,\n    scoreThreshold: 0.7,\n    enlargeFactor: 1.65, // empiric tuning\n    maxHands: 2,\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    let savedWebglPackDepthwiseConvFlag;\n    if (tf.getBackend() === 'webgl') {\n      savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n      tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
-  "mappings": ";;;;;;;;;;;;;;;;;;AAAA;AAAA,QAAM,MAAK;AAEX,QAAM,gBAAgB;AAEtB,2BAAyB;AACvB,UAAM,OAAO,CAAE,SAAS,CAAC,YAAY,IAAI,YAAY,IAAI,SAAS,CAAC,GAAG;AACtE,UAAM,UAAU;AAChB,aAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,QAAQ;AACvC,YAAM,SAAS,KAAK,QAAQ;AAC5B,YAAM,WAAW,KAAK,MAAO,aAAY,SAAS,KAAK;AACvD,YAAM,WAAW,KAAK,MAAO,aAAY,SAAS,KAAK;AACvD,YAAM,aAAa,KAAK,QAAQ;AAChC,eAAS,QAAQ,GAAG,QAAQ,UAAU;AACpC,cAAM,UAAU,SAAU,SAAQ;AAClC,iBAAS,QAAQ,GAAG,QAAQ,UAAU;AACpC,gBAAM,UAAU,SAAU,SAAQ;AAClC,mBAAS,IAAI,GAAG,IAAI,YAAY;AAC9B,oBAAQ,KAAK,CAAC,SAAS;AAAA;AAAA;AAAA;AAAA;AAK/B,WAAO;AAAA;AAGT,QAAM,aAAa,CAAC;AAClB,QAAI,eAAe;AACnB,QAAI,WAAW;AACf,QAAI,SAAS;AAAA;AAGf,QAAM,YAAY,CAAC,mBAAoB;AAAA,IACrC;AAAA,IACA,YAAY,IAAG,MAAM,gBAAgB,CAAC,GAAG,IAAI,CAAC,IAAI;AAAA,IAClD,UAAU,IAAG,MAAM,gBAAgB,CAAC,GAAG,IAAI,CAAC,IAAI;AAAA;AAGlD,QAAM,WAAW,CAAC,KAAK;AACrB,UAAM,SAAS,IAAG,IAAI,IAAI,YAAY;AACtC,UAAM,OAAO,IAAG,IAAI,IAAI,UAAU;AAClC,UAAM,iBAAiB,IAAG,SAAS,CAAC,QAAQ,OAAO;AACnD,WAAO,UAAU;AAAA;AAGnB,wBAAsB,YAAY,SAAS;AACzC,UAAM,YAAY,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACpD,UAAM,UAAU,IAAG,IAAI,WAAW;AAClC,UAAM,WAAW,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACnD,UAAM,qBAAqB,IAAG,IAAI,UAAU;AAC5C,UAAM,oBAAoB,IAAG,IAAI,SAAS;AAC1C,UAAM,cAAc,IAAG,IAAI,oBAAoB;AAC/C,UAAM,SAAS,IAAG,IAAI,mBAAmB;AACzC,UAAM,OAAO,IAAG,IAAI,mBAAmB;AACvC,UAAM,kBAAkB,IAAG,IAAI,QAAQ;AACvC,UAAM,gBAAgB,IAAG,IAAI,MAAM;AACnC,UAAM,aAAa;AACnB,WAAO,IAAG,SAAS,CAAC,iBAAiB,gBAAgB;AAAA;AAGvD,kCAAgC,MAAM;AACpC,WAAO,IAAG,KAAK;AACb,YAAM,MAAM,KAAK,SAAS,KAAK,SAAS;AACxC,aAAO,SAAS,KAAK,aAAa,eAAe;AAAA;AAAA;AAIrD;AAAA,IACE,YAAY,OAAO;AACjB,WAAK,iBAAiB;AACtB,WAAK,QAAQ,QAAO,SAAS;AAC7B,WAAK,SAAS,QAAO,SAAS;AAC9B,WAAK,WAAW,QAAO,SAAS;AAChC,WAAK,cAAc,gBAAgB,QAAO,SAAS;AACnD,WAAK,UAAU,IAAG,SAAS,KAAK;AAChC,WAAK,YAAY,IAAG,SAAS,CAAC,KAAK,OAAO,KAAK;AAC/C,WAAK,eAAe,QAAO,SAAS;AACpC,WAAK,aAAa;AAClB,WAAK,iBAAiB,QAAO,SAAS;AAAA;AAAA,UAGlC,iBAAiB;AAErB,UAAK,CAAC,cAAgB,WAAW,sBAAwB,WAAW,MAAM,WAAW,KAAO,WAAW,MAAM,KAAK,KAAO,WAAW,MAAM,KAAK;AAAI,eAAO;AAC1J,YAAM,CAAC,iBAAiB,OAAO,UAAU,IAAG,KAAK;AAC/C,cAAM,eAAe,WAAW,eAAe,CAAC,KAAK,OAAO,KAAK;AACjE,cAAM,kBAAkB,IAAG,IAAI,IAAG,IAAI,aAAa,IAAI,MAAM,MAAM;AACnE,cAAM,oBAAoB,KAAK,eAAe,QAAQ;AACtD,YAAI;AAEJ,YAAI,MAAM,QAAQ;AAChB,gBAAM,SAAS,kBAAkB,KAAK,CAAC,GAAG,MAAM,EAAE,OAAO,EAAE;AAC3D,gBAAM,YAAY,IAAG,OAAO,CAAC,OAAO,IAAI,OAAO,KAAK;AACpD,gBAAM,YAAY,IAAG,OAAO,CAAC,OAAO,IAAI,OAAO,KAAK;AACpD,gBAAM,SAAS,IAAG,OAAO,CAAC,WAAW,YAAY;AACjD,uBAAa,OAAO,QAAQ;AAAA;AAE5B,uBAAa,kBAAkB;AAAA;AAEjC,cAAM,gBAAgB,aAAa,YAAY,KAAK,SAAS,KAAK;AAClE,cAAM,SAAS,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACjD,cAAM,YAAY,IAAG,QAAQ,QAAQ;AACrC,eAAO,CAAC,YAAY,eAAe;AAAA;AAGrC,YAAM,mBAAmB,MAAM,IAAG,MAAM,uBAAuB,OAAO,QAAQ,KAAK,UAAU,KAAK,cAAc,KAAK;AACrH,YAAM,aAAa,MAAM,iBAAiB;AAC1C,uBAAiB;AACjB,UAAI,gBAAgB,WAAW,IAAI,CAAC,aAAa,IAAG,MAAM,OAAO,CAAC,UAAU,IAAI,CAAC,GAAG;AACpF,sBAAgB,MAAM,QAAQ,IAAI,cAAc,IAAI,OAAO;AACzD,cAAM,OAAO,MAAM,YAAY;AAC/B,oBAAY;AACZ,eAAO;AAAA;AAET,YAAM,iBAAiB;AACvB,eAAS,IAAI,GAAG,IAAI,cAAc,QAAQ;AACxC,cAAM,cAAc,cAAc;AAClC,cAAM,eAAe,IAAG,KAAK;AAC3B,gBAAM,MAAM,UAAU;AACtB,gBAAM,WAAW,WAAW;AAC5B,gBAAM,SAAS,KAAK,YAAY;AAChC,gBAAM,YAAY,IACf,MAAM,iBAAiB,CAAC,UAAU,gBAAgB,IAAI,CAAC,GAAG,KAC1D,UACA,QAAQ,CAAC,eAAe;AAC3B,gBAAM,cAAc,IAAG,MAAM,QAAQ,CAAC,WAAW,CAAC;AAClD,iBAAO,CAAE,KAAK,WAAW,aAAa;AAAA;AAExC,uBAAe,KAAK;AAAA;AAEtB,YAAM;AACN,aAAO;AACP,sBAAgB;AAChB,aAAO;AAAA,QACL,OAAO;AAAA,QACP,aAAa,CAAC,WAAW,MAAM,KAAK,KAAK,OAAO,WAAW,MAAM,KAAK,KAAK;AAAA;AAAA;AAAA,UAIzE,cAAc;AAClB,YAAM,QAAQ,IAAG,KAAK;AACpB,YAAI,CAAE,kBAAiB,IAAG;AACxB,kBAAQ,IAAG,QAAQ,WAAW;AAAA;AAEhC,eAAO,MAAM,UAAU,WAAW;AAAA;AAEpC,YAAM,CAAE,OAAO,eAAgB,MAAM,KAAK,iBAAiB;AAC3D,YAAM;AACN,aAAO,QAAQ,IAAI,MAAM,IAAI,OAAO;AAClC,cAAM,YAAY,uBAAuB,MAAM;AAC/C,cAAM,CAAC,cAAc,SAAS,mBAAmB,MAAM,QAAQ,IAAI,CAAC,KAAK,WAAW,WAAW,KAAK,aAAa,IAAI,OAAO,MAAM,EAAE;AACpI,cAAM,SAAS,KAAK;AACpB,cAAM,CAAC,cAAc,gBAAgB;AACrC,cAAM,kBAAkB,aACrB,IAAI,CAAC,aAAc;AAAA,UACjB,UAAS,KAAK,OAAO,MAAM;AAAA,UAC3B,UAAS,KAAK,OAAO,MAAM;AAAA;AAEhC,cAAM,iBAAiB;AAAA,UACrB,SAAS,QAAQ,MAAM,GAAG;AAAA,UAC1B,aAAa,QAAQ,MAAM;AAAA,UAC3B,WAAW;AAAA,UACX,aAAa;AAAA;AAEf,mBAAW,KAAK;AAChB,aAAK,UAAU;AACf,aAAK,YAAY;AACjB,kBAAU;AACV,eAAO;AAAA;AAAA;AAAA;AAKb,sBAAoB;AAClB,UAAM,YAAY,MAAM,IAAG,eAAe,QAAO,SAAS,WAAW,CAAE,WAAW,QAAO,SAAS,UAAU,SAAS;AACrH,UAAM,QAAQ,IAAI,eAAe,WAAW;AAC5C,WAAO;AAAA;AAGT,WAAQ,OAAO;AACf,WAAQ,iBAAiB;AACzB,WAAQ,aAAa;AAAA;;;ACpLrB;AAAA,WAAQ,mBAAmB;AAAA,IACzB,YAAY;AAAA,MACV;AAAA,MAAI;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MACtD;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MACvD;AAAA,MAAK;AAAA,MAAI;AAAA,MAAK;AAAA,MAAI;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAI;AAAA,MAAI;AAAA,MAAK;AAAA,MAAI;AAAA;AAAA,IAEpD,gBAAgB,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI,GAAG,KAAK,KAAK,KAAK,KAAK;AAAA,IAC7D,gBAAgB,CAAC,KAAK,IAAI,KAAK,IAAI,IAAI,KAAK,KAAK,KAAK,KAAK;AAAA,IAC3D,gBAAgB,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI,IAAI,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9D,gBAAgB,CAAC,IAAI,IAAI,IAAI,KAAK,IAAI,IAAI,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9D,gBAAgB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC/C,gBAAgB,CAAC,IAAI,GAAG,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACtD,gBAAgB,CAAC,KAAK,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC1C,gBAAgB,CAAC,KAAK,IAAI,KAAK,IAAI,IAAI,IAAI,IAAI,KAAK;AAAA,IACpD,gBAAgB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC/C,gBAAgB,CAAC,KAAK,IAAI,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,gBAAgB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACzD,mBAAmB,CAAC,KAAK,IAAI,IAAI,KAAK,IAAI,KAAK,IAAI;AAAA,IACnD,mBAAmB,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI;AAAA,IACzC,cAAc,CAAC,KAAK,KAAK,KAAK,KAAK;AAAA,IACnC,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9C,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9C,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9C,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,kBAAkB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACtD,kBAAkB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC5C,aAAa,CAAC,KAAK,KAAK,KAAK,KAAK;AAAA,IAClC,mBAAmB,CAAC;AAAA,IACpB,SAAS,CAAC;AAAA,IACV,YAAY,CAAC;AAAA,IACb,iBAAiB,CAAC;AAAA,IAClB,gBAAgB,CAAC;AAAA,IACjB,YAAY,CAAC;AAAA,IACb,WAAW,CAAC;AAAA;AAEd,WAAQ,2BAA2B;AAAA,IACjC,CAAE,KAAK,aAAa,SAAS,CAAC,GAAG,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IACrD,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IACtD,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IACtD,CAAE,KAAK,aAAa,SAAS,CAAC,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG;AAAA,IACtD,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC9D,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC9D,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC9D,CAAE,KAAK,gBAAgB,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC7D,CAAE,KAAK,gBAAgB,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA;AAAA;;;AC/CvD;AAAA,QAAM,MAAK;AAEX,+BAA6B,KAAK;AAChC,UAAM,aAAa,CAAC,IAAI,WAAW,KAAK,OAAO,IAAI,IAAI,WAAW,KAAK,OAAO;AAC9E,UAAM,WAAW,CAAC,IAAI,SAAS,KAAK,OAAO,IAAI,IAAI,SAAS,KAAK,OAAO;AACxE,WAAO,CAAE,YAAY;AAAA;AAEvB,WAAQ,sBAAsB;AAC9B,sBAAoB;AAClB,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA,MAC1C,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA;AAAA;AAG9C,WAAQ,aAAa;AACrB,wBAAsB;AACpB,WAAO;AAAA,MACL,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA,MAC5D,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA;AAAA;AAGhE,WAAQ,eAAe;AACvB,oCAAkC,KAAK,OAAO;AAC5C,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,QAAQ,CAAC;AAAA,MACb,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,SAAS,KAAK;AAAA,MAChE,IAAI,SAAS,KAAK;AAAA;AAEpB,WAAO,IAAG,MAAM,cAAc,OAAO,OAAO,CAAC,IAAI;AAAA;AAEnD,WAAQ,2BAA2B;AACnC,sBAAoB,KAAK,SAAS;AAChC,UAAM,SAAS,aAAa;AAC5B,UAAM,OAAO,WAAW;AACxB,UAAM,cAAc,CAAC,SAAS,KAAK,KAAK,GAAG,SAAS,KAAK,KAAK;AAC9D,UAAM,aAAa,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACxE,UAAM,WAAW,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACtE,WAAO,CAAE,YAAY,UAAU,WAAW,IAAI;AAAA;AAEhD,WAAQ,aAAa;AACrB,uBAAqB;AACnB,UAAM,UAAU,aAAa;AAC7B,UAAM,OAAO,WAAW;AACxB,UAAM,UAAU,KAAK,IAAI,GAAG;AAC5B,UAAM,WAAW,UAAU;AAC3B,UAAM,aAAa,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACxD,UAAM,WAAW,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACtD,WAAO,CAAE,YAAY,UAAU,WAAW,IAAI;AAAA;AAEhD,WAAQ,cAAc;AAAA;;;AClDtB;AAAA,WAAQ,kBAAkB,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG;AAKxD,4BAA0B;AACxB,WAAO,QAAQ,IAAI,KAAK,KAAK,KAAK,MAAO,SAAQ,KAAK,MAAO,KAAI,KAAK;AAAA;AAExE,WAAQ,mBAAmB;AAM3B,2BAAyB,QAAQ;AAC/B,UAAM,UAAU,KAAK,KAAK,IAAI,KAAK,MAAM,CAAE,QAAO,KAAK,OAAO,KAAK,OAAO,KAAK,OAAO;AACtF,WAAO,iBAAiB;AAAA;AAE1B,WAAQ,kBAAkB;AAC1B,wBAAsB;AACpB,WAAO,MAAM,MAAM,KAAK;AAAA;AAE1B,WAAQ,eAAe;AACvB,kCAAgC,GAAG;AACjC,WAAO,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG;AAAA;AAEvC,eAAa,IAAI;AACf,QAAI,UAAU;AACd,aAAS,IAAI,GAAG,IAAI,GAAG,QAAQ;AAC7B,iBAAW,GAAG,KAAK,GAAG;AAAA;AAExB,WAAO;AAAA;AAET,WAAQ,MAAM;AACd,8BAA4B,KAAK;AAC/B,UAAM,SAAS;AACf,aAAS,IAAI,GAAG,IAAI,IAAI,QAAQ;AAC9B,aAAO,KAAK,IAAI,GAAG;AAAA;AAErB,WAAO;AAAA;AAET,WAAQ,qBAAqB;AAC7B,qCAAmC,MAAM;AACvC,UAAM,UAAU;AAChB,UAAM,OAAO,KAAK;AAClB,aAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,cAAQ,KAAK;AACb,eAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,gBAAQ,KAAK,KAAK,IAAI,KAAK,MAAM,mBAAmB,MAAM;AAAA;AAAA;AAG9D,WAAO;AAAA;AAET,+BAA6B,UAAU;AACrC,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,iBAAiB,CAAC,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,GAAG,GAAG;AAClE,UAAM,oBAAoB,uBAAuB,OAAO,IAAI,OAAO;AACnE,UAAM,2BAA2B,0BAA0B,mBAAmB;AAC9E,UAAM,4BAA4B,uBAAuB,CAAC,OAAO,IAAI,CAAC,OAAO;AAC7E,WAAO,0BAA0B,0BAA0B;AAAA;AAE7D,WAAQ,sBAAsB;AAC9B,iCAA+B;AAC7B,UAAM,oBAAoB,CAAC,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG,KAAK,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AAClF,UAAM,uBAAuB,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AACtD,UAAM,sBAAsB;AAAA,MAC1B,CAAC,IAAI,kBAAkB,IAAI;AAAA,MAC3B,CAAC,IAAI,kBAAkB,IAAI;AAAA;AAE7B,WAAO;AAAA,MACL,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,CAAC,GAAG,GAAG;AAAA;AAAA;AAGX,WAAQ,wBAAwB;AAChC,uBAAqB,uBAAuB;AAC1C,WAAO;AAAA,MACL,IAAI,uBAAuB,eAAe;AAAA,MAC1C,IAAI,uBAAuB,eAAe;AAAA;AAAA;AAG9C,WAAQ,cAAc;AACtB,mCAAiC,GAAG;AAClC,WAAO,KAAK,KAAO,GAAE,KAAK,EAAE,OAAO,IAAO,GAAE,KAAK,EAAE,OAAO;AAAA;AAE5D,WAAQ,0BAA0B;AAAA;;;ACvFlC;AACA,QAAM,MAAK;AACX,QAAM,WAAW;AACjB,QAAM,YAAY;AAClB,QAAM,OAAO;AAEb,QAAM,kBAAkB;AACxB,QAAM,0CAA0C;AAChD,QAAM,mBAAmB;AACzB,QAAM,0CAA0C,CAAC,kBAAkB,UAAU,iBAAiB,qBAAqB;AACnH,QAAM,wBAAwB;AAC9B,QAAM,uBAAuB;AAC7B,QAAM,+CAA+C,CAAC,uBAAuB;AAC7E,QAAM,mBAAmB,UAAU,iBAAiB;AACpD,QAAM,kBAAkB,CAAC,iBAAiB,IAAI,iBAAiB,iBAAiB,SAAS;AACzF,QAAM,oBAAoB,UAAU,iBAAiB;AACrD,QAAM,mBAAmB,CAAC,kBAAkB,IAAI,kBAAkB,kBAAkB,SAAS;AAC7F,QAAM,0BAA0B;AAChC,QAAM,0BAA0B;AAChC,QAAM,kBAAkB;AACxB,QAAM,uBAAuB;AAG7B,iCAA+B,WAAW,WAAW,QAAQ;AAC3D,aAAS,IAAI,GAAG,IAAI,UAAU,yBAAyB,QAAQ;AAC7D,YAAM,CAAE,KAAK,WAAY,UAAU,yBAAyB;AAC5D,YAAM,kBAAkB,UAAU,iBAAiB,GAAG,SAAS;AAC/D,YAAM,uBAAuB,QAAQ;AACrC,UAAI,wBAAwB,KAAK,SAAS;AACxC,iBAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ;AAClC,gBAAM,QAAQ,QAAQ;AACtB,oBAAU,gBAAgB,MAAM;AAAA,YAC9B,UAAU,OAAO;AAAA,YAAI,UAAU,OAAO;AAAA,YACrC,WAAU,OAAO,KAAK,UAAU,gBAAgB,IAAI,MAAM;AAAA;AAAA;AAAA;AAAA;AAAA;AAOrE;AAAA,IACE,YAAY,qBAAqB,cAAc,WAAW;AAExD,WAAK,oBAAoB;AACzB,WAAK,0BAA0B;AAC/B,WAAK,sBAAsB;AAC3B,WAAK,eAAe;AACpB,WAAK,YAAY;AACjB,WAAK,YAAY,QAAO,KAAK;AAC7B,WAAK,aAAa,QAAO,KAAK;AAC9B,WAAK,WAAW,QAAO,KAAK;AAC5B,WAAK,cAAc,QAAO,KAAK;AAAA;AAAA,IAGjC,mBAAmB,WAAW,KAAK,OAAO;AACxC,YAAM,UAAU,SAAS,WAAW,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI;AAChF,YAAM,cAAc,CAAC,QAAQ,KAAK,KAAK,WAAW,QAAQ,KAAK,KAAK;AACpE,YAAM,eAAe,UAAU,IAAI,CAAC,UAAW;AAAA,QAC7C,YAAY,KAAM,OAAM,KAAK,KAAK,YAAY;AAAA,QAC9C,YAAY,KAAM,OAAM,KAAK,KAAK,aAAa;AAAA,QAAI,MAAM;AAAA;AAE3D,YAAM,uBAAuB,KAAK,oBAAoB,OAAO,CAAC,GAAG;AACjE,YAAM,gBAAgB,aAAa,IAAI,CAAC,UAAW,CAAC,GAAG,KAAK,YAAY,OAAO,uBAAuB,MAAM;AAC5G,YAAM,wBAAwB,KAAK,sBAAsB;AACzD,YAAM,YAAY,CAAC,GAAG,SAAS,aAAa,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI,YAAa;AACrG,YAAM,oBAAoB;AAAA,QACxB,KAAK,IAAI,WAAW,sBAAsB;AAAA,QAC1C,KAAK,IAAI,WAAW,sBAAsB;AAAA;AAE5C,aAAO,cAAc,IAAI,CAAC,UAAW;AAAA,QACnC,MAAM,KAAK,kBAAkB;AAAA,QAC7B,MAAM,KAAK,kBAAkB;AAAA,QAAI,MAAM;AAAA;AAAA;AAAA,IAI3C,iCAAiC;AAC/B,YAAM,WAAW,UAAU,gBAAgB,IAAI;AAC/C,YAAM,YAAY,UAAU,iBAAiB,IAAI;AACjD,aAAO,WAAW;AAAA;AAAA,IAIpB,UAAU,WAAW,MAAM,qBAAqB,qBAAqB,OAAO;AAC1E,YAAM,MAAM,SAAS,YAAY,SAAS,WAAW,KAAK,8BAA8B,CAAC,UAAU,sBAAsB,UAAU,wBAAwB,KAAK;AAChK,YAAM,UAAU,SAAS,WAAW;AACpC,UAAI,OAAO,IAAG,MAAM,cAAc,MAAM,CAAC;AAAA,QACvC,IAAI,WAAW,KAAK,KAAK;AAAA,QACzB,IAAI,WAAW,KAAK,KAAK;AAAA,QAAW,IAAI,SAAS,KAAK,KAAK;AAAA,QAC3D,IAAI,SAAS,KAAK,KAAK;AAAA,UACrB,CAAC,IAAI,CAAC,KAAK,UAAU,KAAK;AAC9B,UAAI;AACF,eAAO,IAAG,MAAM,cAAc;AAAA;AAEhC,aAAO,CAAE,KAAK,SAAS;AAAA;AAAA,IAIzB,aAAa,SAAS,QAAQ,YAAY,OAAO;AAC/C,YAAM,eAAe;AACrB,eAAS,IAAI,GAAG,IAAI,sBAAsB;AACxC,cAAM,IAAI,QAAQ,IAAI;AACtB,cAAM,IAAI,QAAQ,IAAI,IAAI;AAC1B,cAAM,IAAI,QAAQ,IAAI,IAAI;AAC1B,qBAAa,KAAK;AAAA,UACf,QACI,IAAK,IAAI,KAAK,WACd,IAAI,KAAK,YAAa,WAAW,KAAK,OAAO,WAAW;AAAA,UAC5D,IAAI,KAAK,WAAY,WAAW,KAAK,OAAO,WAAW;AAAA,UAAI;AAAA;AAAA;AAGhE,aAAO,CAAE,WAAW,cAAc,MAAM,aAAa,MAAM;AAAA;AAAA,IAI7D,sBAAsB,WAAW,YAAY;AAC3C,YAAM,eAAe,UAAU,UAAU,iBAAiB,GAAG,sBAAsB,0BAA0B;AAC7G,YAAM,eAAe,UAAU,UAAU,iBAAiB,GAAG,sBAAsB,0BAA0B;AAC7G,YAAM,WAAY,gBAAe,gBAAgB;AAEjD,aAAO,WAAW,IAAI,CAAC,OAAO;AAC5B,YAAI,IAAI;AACR,YAAI,MAAM;AACR,cAAI;AAAA,mBACK,MAAM;AACf,cAAI;AAAA;AAEN,eAAO,CAAC,MAAM,IAAI,MAAM,IAAI;AAAA;AAAA;AAAA,UAI1B,QAAQ,OAAO;AACnB,WAAK,aAAa,QAAO,SAAS;AAClC,WAAK,WAAW,QAAO,SAAS;AAChC,UAAI,KAAK;AACP,cAAM,CAAE,OAAO,eAAgB,MAAM,KAAK,oBAAoB,iBAAiB;AAC/E,YAAI,MAAM,WAAW;AACnB,eAAK,oBAAoB;AACzB,iBAAO;AAAA;AAET,cAAM,cAAc,MAAM,IAAI,CAAC;AAC7B,gBAAM,gBAAgB;AAAA,YACpB,YAAY,WAAW,IAAI,WAAW,UAAU;AAAA,YAChD,UAAU,WAAW,IAAI,SAAS,UAAU;AAAA;AAE9C,qBAAW,IAAI,WAAW;AAC1B,qBAAW,IAAI,SAAS;AACxB,gBAAM,YAAY,SAAS,oBAAoB,eAAe;AAC9D,gBAAM,cAAc,SAAS,WAAW;AACxC,gBAAM,YAAY,WAAW,UAAU;AACvC,qBAAW,UAAU;AACrB,qBAAW,YAAY;AACvB,iBAAO,IAAK,aAAa;AAAA;AAE3B,aAAK,wBAAwB;AAC7B,aAAK,0BAA0B;AAAA;AAE/B,aAAK;AAAA;AAEP,YAAM,UAAU,IAAG,KAAK,MAAM,KAAK,kBAAkB,IAAI,CAAC,KAAK;AAC7D,YAAI,QAAQ;AAEZ,cAAM,4BAA4B,IAAI,UAAU,UAAU;AAC1D,YAAI,CAAC,cAAc,mBAAmB;AACtC,YAAI,8BAA8B;AAChC,WAAC,cAAc,mBAAmB;AAAA;AAEpC,gBAAQ,KAAK,gBAAgB,IAAI,UAAU,eAAe,IAAI,UAAU;AACxE,cAAM,aAAa,SAAS,aAAa,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI;AACrF,cAAM,uBAAuB,CAAC,WAAW,KAAK,MAAM,MAAM,IAAI,WAAW,KAAK,MAAM,MAAM;AAC1F,YAAI,eAAe;AACnB,YAAI,iBAAiB,KAAK;AAC1B,YAAI,UAAU;AACZ,yBAAe,IAAG,MAAM,iBAAiB,OAAO,OAAO,GAAG;AAC1D,2BAAiB,KAAK,oBAAoB,CAAC,OAAO;AAAA;AAEpD,cAAM,SAAS,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI;AAC3D,cAAM,OAAO,SAAS,yBAAyB,QAAQ,cAAc,CAAC,KAAK,YAAY,KAAK,YAAY,IAAI;AAE5G,cAAM,CAAC,EAAE,MAAM,UAAU,KAAK,aAAa,QAAQ;AACnD,cAAM,iBAAiB,IAAG,QAAQ,QAAQ,CAAC,IAAI;AAC/C,YAAI,YAAY,eAAe;AAC/B,YAAI,QAAO,KAAK;AACd,gBAAM,CAAE,KAAK,YAAY,SAAS,gBAAgB,MAAM,eAAgB,KAAK,UAAU,WAAW,MAAM,gBAAgB,IAAI,gBAAgB,IAAI;AAChJ,gBAAM,CAAE,KAAK,aAAa,SAAS,iBAAiB,MAAM,gBAAiB,KAAK,UAAU,WAAW,MAAM,iBAAiB,IAAI,iBAAiB;AACjJ,gBAAM,iBAAkB,KAAK,UAAU,QAAQ,IAAG,OAAO,CAAC,aAAa;AACvE,gBAAM,qBAAqB,eAAe;AAC1C,yBAAe;AACf,gBAAM,cAAc,mBAAmB,MAAM,GAAG,uBAAuB;AACvE,gBAAM,CAAE,WAAW,kBAAkB,MAAM,qBAAsB,KAAK,aAAa,aAAa,YAAY,gBAAgB;AAC5H,gBAAM,eAAe,mBAAmB,MAAM,uBAAuB;AACrE,gBAAM,CAAE,WAAW,mBAAmB,MAAM,sBAAuB,KAAK,aAAa,cAAc,aAAa;AAChH,gBAAM,gCAAgC,KAAK,iCAAiC;AAC5E,cAAI,KAAK,IAAI,iCAAiC;AAC5C,kCAAsB,WAAW,kBAAkB;AACnD,kCAAsB,WAAW,mBAAmB;AAAA,qBAE3C,gCAAgC;AACzC,kCAAsB,WAAW,kBAAkB,QAAQ,CAAC,aAAa;AAAA;AAEzE,kCAAsB,WAAW,mBAAmB,SAAS,CAAC,aAAa;AAAA;AAE7E,gBAAM,yBAAyB,KAAK,sBAAsB,WAAW,mBAAmB;AACxF,gBAAM,0BAA0B,KAAK,sBAAsB,WAAW,oBAAoB;AAC1F,sBAAY,UAAU,OAAO,wBAAwB,OAAO;AAAA;AAE9D,cAAM,wBAAwB,KAAK,mBAAmB,WAAW,KAAK,OAAO;AAC7E,YAAG,QAAQ;AACX,cAAM,eAAe,SAAS,WAAW,KAAK,8BAA8B;AAC5E,YAAI,QAAO,KAAK;AACd,gBAAM,oBAAoB,IAAG,SAAS;AACtC,eAAK,kBAAkB,KAAK,IAAK,cAAc,WAAW,kBAAkB;AAC5E,gBAAM,cAAa;AAAA,YACjB,QAAQ;AAAA,YACR,KAAK;AAAA,YACL,YAAY,KAAK;AAAA,YACjB,OAAO;AAAA;AAET,iBAAO;AAAA;AAET,cAAM,aAAa;AAAA,UACjB,QAAQ;AAAA,UACR,KAAK;AAAA,UACL,YAAY,KAAK;AAAA,UACjB,OAAO;AAAA;AAET,eAAO;AAAA;AAET,aAAO;AAAA;AAAA,IAIT,wBAAwB;AACtB,eAAS,IAAI,GAAG,IAAI,MAAM,QAAQ;AAChC,cAAM,MAAM,MAAM;AAClB,cAAM,cAAc,KAAK,kBAAkB;AAC3C,YAAI,MAAM;AACV,YAAI,eAAe,YAAY;AAC7B,gBAAM,CAAC,WAAW,aAAa,IAAI;AACnC,gBAAM,CAAC,SAAS,WAAW,IAAI;AAC/B,gBAAM,CAAC,mBAAmB,qBAAqB,YAAY;AAC3D,gBAAM,CAAC,iBAAiB,mBAAmB,YAAY;AACvD,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,eAAgB,WAAU,aAAc,WAAU;AACxD,gBAAM,UAAW,WAAU,aAAc,WAAU;AACnD,gBAAM,kBAAmB,mBAAkB,qBAAsB,mBAAkB;AACnF,gBAAM,eAAgB,WAAU,kBAAkB;AAAA;AAEpD,YAAI,MAAM;AACR,eAAK,kBAAkB,KAAK;AAAA;AAAA;AAGhC,WAAK,oBAAoB,KAAK,kBAAkB,MAAM,GAAG,MAAM;AAAA;AAAA,IAGjE,sBAAsB;AACpB,UAAI,KAAK,kBAAkB,UAAU;AACnC,aAAK,oBAAoB;AAAA,UACvB,GAAG,KAAK,kBAAkB,MAAM,GAAG;AAAA,UACnC,GAAG,KAAK,kBAAkB,MAAM,QAAQ;AAAA;AAAA;AAAA;AAAA,IAK9C;AACE,YAAM,YAAY,KAAK,kBAAkB;AACzC,YAAM,SAAS,cAAc;AAC7B,UAAI,KAAK,aAAa,KAAK;AACzB,eAAO;AAAA;AAET,aAAO,cAAc,KAAK,YAAY,KAAK,2BAA2B,KAAK;AAAA;AAAA,IAG7E,8BAA8B;AAC5B,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,aAAa,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AACjD,YAAM,WAAW,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AAC/C,aAAO,CAAE,YAAY;AAAA;AAAA;AAGzB,WAAQ,WAAW;AAAA;;;AC3RnB;AAAA,WAAQ,YAAY;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,kBAAkB;AAAA,IACnB,CAAC,gBAAgB;AAAA,IACjB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA;AAAA;;;ACpdtB;AAAA;AAAA;AAAA;AAAA,MAAO,wBAAQ;AAAA,IACb;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IACpE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACtE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACxE;AAAA,IAAG;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACvE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAC1E;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACzE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACxE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACzE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACpE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA;AAAA;;;ACxKnE;AAAA,QAAM,MAAK;AACX,QAAM,YAAY;AAClB,QAAM,YAAY;AAClB,QAAM,OAAO;AACb,QAAM,YAAY;AAClB,QAAM,gBAAgB,wBAA2B;AAEjD;AAAA,IACE,YAAY,WAAW,gBAAgB,WAAW;AAChD,WAAK,WAAW,IAAI,KAAK,SAAS,WAAW,gBAAgB,WAAW;AACxE,UAAI;AAAQ,aAAK,SAAS;AAAA;AAAA,UAGtB,cAAc,OAAO;AACzB,UAAI;AAAQ,aAAK,SAAS;AAC1B,YAAM,QAAQ,IAAG,KAAK;AACpB,YAAI,CAAE,kBAAiB,IAAG;AAAS,kBAAQ,IAAG,QAAQ,WAAW;AACjE,eAAO,MAAM,UAAU,WAAW;AAAA;AAEpC,YAAM,cAAc,MAAM,KAAK,SAAS,QAAQ,OAAO;AACvD,UAAG,QAAQ;AACX,YAAM,UAAU;AAChB,iBAAW,cAAe,eAAe;AACvC,cAAM,aAAa,WAAW,WAAW;AACzC,YAAI,cAAc,KAAK,OAAO,SAAS;AACrC,gBAAM,OAAO,WAAW,SAAS,WAAW,OAAO,cAAc;AACjE,gBAAM,cAAc;AACpB,cAAI,QAAQ,KAAK,SAAS;AACxB,uBAAW,OAAO,UAAU;AAC1B,kBAAI,KAAK,OAAO,KAAK,WAAW,IAAI,SAAS,YAAY;AACvD,4BAAY,OAAO,UAAU,iBAAiB,KAAK,IAAI,CAAC,UAAU,KAAK;AAAA;AAAA;AAAA;AAI7E,kBAAQ,KAAK;AAAA,YACX,YAAY,cAAc;AAAA,YAC1B,KAAK,WAAW,MAAM,CAAC,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,SAAS,KAAK,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,SAAS,KAAK,WAAW,IAAI,WAAW,MAAM;AAAA,YAC3M;AAAA,YACA;AAAA,YACA,OAAO,WAAW,QAAQ,IAAG,MAAM,WAAW,SAAS;AAAA;AAAA;AAG3D,mBAAW,WAAW;AACtB,mBAAW,MAAM;AAAA;AAEnB,aAAO;AAAA;AAAA;AAIX,sBAAoB;AAClB,UAAM,UAAS,MAAM,QAAQ,IAAI;AAAA,MAC/B,UAAU,KAAK;AAAA,MACf,IAAG,eAAe,QAAO,KAAK,WAAW,CAAE,WAAW,QAAO,KAAK,UAAU,SAAS;AAAA,MACrF,IAAG,eAAe,QAAO,KAAK,WAAW,CAAE,WAAW,QAAO,KAAK,UAAU,SAAS;AAAA;AAEvF,UAAM,WAAW,IAAI,kBAAkB,QAAO,IAAI,QAAO,IAAI,QAAO,IAAI;AACxE,WAAO;AAAA;AAGT,WAAQ,OAAO;AACf,WAAQ,oBAAoB;AAC5B,WAAQ,YAAY;AACpB,WAAQ,gBAAgB;AAAA;;;AC9DxB;AAAA,QAAM,MAAK;AAEX,QAAM,UAAS;AACf,MAAI,OAAO,CAAE,KAAK,GAAG,QAAQ;AAC7B,MAAI,QAAQ;AAEZ,0BAAwB,OAAO;AAC7B,UAAM,SAAS,IAAG,QAAQ,WAAW;AACrC,UAAM,SAAS,IAAG,MAAM,eAAe,QAAQ,CAAC,MAAM;AACtD,UAAM,SAAS,IAAG,KAAK,IAAG,WAAW,QAAQ,IAAI;AACjD,WAAO;AAAA;AAGT,yBAAuB;AACrB,QAAI,CAAC,QAAO;AAAK,cAAO,MAAM,MAAM,IAAG,eAAe,QAAO,KAAK,IAAI;AACtE,WAAO,QAAO;AAAA;AAGhB,4BAA0B;AACxB,QAAI,CAAC,QAAO;AAAQ,cAAO,SAAS,MAAM,IAAG,eAAe,QAAO,KAAK,OAAO;AAC/E,WAAO,QAAO;AAAA;AAGhB,yBAAuB,OAAO;AAC5B,aAAS;AACT,QAAI,SAAS,QAAO,KAAK,IAAI;AAC3B,cAAQ;AACR,aAAO;AAAA;AAET,QAAI;AACJ,QAAI,iBAAiB,IAAG;AACtB,YAAM,SAAS,IAAG,MAAM,eAAe,OAAO,CAAC,QAAO,KAAK,IAAI,WAAW,QAAO,KAAK,IAAI,YAAY;AACtG,gBAAU,IAAG,IAAI,QAAQ,CAAC;AAC1B,UAAG,QAAQ;AAAA;AAEX,gBAAU,MAAM,SAAS,OAAO,QAAO,KAAK,IAAI;AAAA;AAElD,UAAM,MAAM;AACZ,QAAI,QAAO,KAAK,IAAI;AAClB,YAAM,OAAO,MAAM,QAAO,IAAI,QAAQ;AACtC,YAAM,OAAO,MAAM,KAAK;AACxB,UAAI,MAAM,KAAK,MAAM,KAAK,KAAK,MAAM;AACrC,UAAG,QAAQ;AAAA;AAEb,QAAI,QAAO,KAAK,OAAO;AACrB,YAAM,UAAU,MAAM,QAAO,OAAO,QAAQ;AAC5C,YAAM,OAAO,MAAM,QAAQ;AAC3B,UAAI,SAAS,KAAK,MAAM,MAAM,KAAK,MAAM,KAAK,WAAW;AACzD,UAAG,QAAQ;AAAA;AAEb,QAAG,QAAQ;AACX,WAAO;AACP,WAAO;AAAA;AAGT,WAAQ,UAAU;AAClB,WAAQ,UAAU;AAClB,WAAQ,aAAa;AAAA;;;ACzDrB;AAAA,QAAM,MAAK;AAEX,QAAM,cAAc,CAAC,SAAS,WAAW,QAAQ,SAAS,OAAO,WAAW;AAC5E,QAAM,UAAS;AACf,MAAI,OAAO;AACX,MAAI,QAAQ;AACZ,QAAM,aAAa;AAEnB,oBAAkB,OAAO;AACvB,UAAM,SAAS,IAAG,KAAK;AACrB,YAAM,SAAS,IAAG,QAAQ,WAAW,OAAO;AAC5C,YAAM,SAAS,IAAG,MAAM,eAAe,QAAQ,CAAC,MAAM;AACtD,YAAM,SAAS,IAAG,KAAK,IAAG,WAAW,QAAQ,IAAI;AACjD,aAAO;AAAA;AAET,WAAO;AAAA;AAGT,sBAAoB;AAClB,QAAI,CAAC,QAAO;AAAS,cAAO,UAAU,MAAM,IAAG,eAAe,QAAO,KAAK,QAAQ;AAClF,WAAO,QAAO;AAAA;AAGhB,yBAAuB,OAAO;AAC5B,aAAS;AACT,QAAI,SAAS,QAAO,KAAK,QAAQ;AAC/B,cAAQ;AACR,aAAO;AAAA;AAET,UAAM,UAAU,IAAG,KAAK;AACtB,UAAI,iBAAiB,IAAG;AACtB,cAAM,SAAS,IAAG,MAAM,eAAe,OAAO,CAAC,QAAO,KAAK,QAAQ,WAAW,QAAO,KAAK,QAAQ,YAAY;AAC9G,cAAM,CAAC,GAAG,GAAG,KAAK,IAAG,MAAM,QAAQ,GAAG;AACtC,YAAI,QAAO,KAAK,QAAQ;AAEtB,gBAAM,KAAK,IAAG,IAAI,GAAG,CAAC;AACtB,gBAAM,KAAK,IAAG,IAAI,GAAG,CAAC;AACtB,gBAAM,KAAK,IAAG,IAAI,GAAG,CAAC;AACtB,gBAAM,YAAY,IAAG,KAAK,CAAC,IAAI,IAAI;AACnC,iBAAO;AAAA;AAET,eAAO;AAAA;AAET,aAAO,SAAS,OAAO,QAAO,KAAK,QAAQ;AAAA;AAE7C,UAAM,MAAM;AACZ,QAAI,QAAO,KAAK,QAAQ;AACtB,YAAM,WAAW,MAAM,QAAO,QAAQ,QAAQ;AAC9C,YAAM,OAAO,MAAM,SAAS;AAC5B,eAAS,IAAI,GAAG,IAAI,KAAK,QAAQ;AAC/B,YAAI,aAAa,KAAK,KAAK,QAAO,KAAK,QAAQ;AAAe,cAAI,KAAK,CAAE,OAAO,KAAK,IAAI,MAAM,KAAK,MAAM,MAAM,aAAa,KAAK,MAAM,MAAM,SAAS,YAAY;AAAA;AAErK,UAAI,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE;AAC/B,UAAG,QAAQ;AAAA;AAEb,QAAG,QAAQ;AACX,WAAO;AACP,WAAO;AAAA;AAGT,WAAQ,UAAU;AAClB,WAAQ,OAAO;AAAA;;;AC7Df;AAAA,QAAM,MAAK;AAEX;AAAA,IACE,YAAY,OAAO;AACjB,WAAK,QAAQ;AACb,WAAK,eAAe;AACpB,YAAM,aAAa,KAAK,MAAM,OAAO,GAAG;AACxC,UAAG,KAAK,OAAQ,WAAW,OAAO,MAAQ,WAAW,OAAO,IAAK,MAAM,gBAAgB,WAAW,OAAO,WAAW;AAAA;AAAA,IAgBtH,QAAQ;AACN,aAAO,IAAG,KAAK;AACb,cAAM,UAAU,KAAK,gBAAgB,MAAM;AAC3C,cAAM,UAAU,QAAQ,WAAW;AACnC,cAAM,UAAU,KAAK,MAAM,QAAQ;AACnC,cAAM,YAAY,QAAQ,IAAI,CAAC,MAAM,EAAE,QAAQ,CAAC;AAChD,cAAM,eAAe,KAAK,kBAAkB;AAC5C,eAAO;AAAA,UACL,eAAe,aAAa,QAAQ;AAAA,UACpC,SAAS,aAAa;AAAA,UACtB,iBAAiB,aAAa;AAAA,UAC9B,iBAAiB,aAAa;AAAA;AAAA;AAAA;AAAA,IAQpC;AACE,WAAK,MAAM;AAAA;AAAA;AAGf,WAAQ,YAAY;AAAA;;;AC9CpB;AAAA,QAAM,MAAK;AACX,QAAM,YAAY;AAElB,0BAAwB,UAAU;AAAA,IAEhC,gBAAgB;AAEd,aAAO,IAAG,KAAK,MAAM,IAAG,IAAI,OAAO,OAAO,IAAI;AAAA;AAAA,IAIhD,kBAAkB;AAChB,YAAM,CAAC,SAAS,SAAS,iBAAiB,mBAAmB;AAC7D,aAAO,CAAE,SAAS,SAAS,iBAAiB;AAAA;AAAA;AAGhD,WAAQ,YAAY;AAAA;;;AChBpB;AACA,gBAAc;AACZ,WAAO,KAAK,MAAM,IAAI;AAAA;AAExB;AAAA,IACE,YAAY,SAAS;AACnB,WAAK,gBAAgB,IAAI,MAAM;AAC/B,WAAK,mBAAmB;AACxB,WAAK,kBAAkB;AAAA;AAAA,IAGzB,QAAQ;AACN,WAAK,cAAc,EAAE,KAAK,oBAAoB;AAC9C,WAAK,KAAK,KAAK;AAAA;AAAA,IAGjB;AACE,YAAM,MAAM,KAAK,cAAc;AAC/B,WAAK,SAAS,GAAG,KAAK;AACtB,WAAK,KAAK;AACV,WAAK,cAAc,KAAK,mBAAmB,KAAK;AAChD,aAAO;AAAA;AAAA,IAGT;AACE,aAAO,KAAK,qBAAqB;AAAA;AAAA,IAGnC;AACE,aAAO,KAAK,mBAAmB;AAAA;AAAA,IAGjC;AACE,aAAO,KAAK,cAAc,MAAM,GAAG,KAAK,mBAAmB;AAAA;AAAA,IAG7D;AACE,aAAO,KAAK,cAAc;AAAA;AAAA,IAG5B,KAAK;AACH,aAAO,IAAI,KAAK,KAAK,KAAK,KAAK,IAAI;AACjC,aAAK,SAAS,GAAG,KAAK;AACtB,YAAI,KAAK;AAAA;AAAA;AAAA,IAIb,KAAK;AACH,aAAO,IAAI,KAAK,KAAK;AACnB,YAAI,IAAI,IAAI;AACZ,YAAI,IAAI,KAAK,oBAAoB,KAAK,KAAK,GAAG,IAAI;AAAI;AACtD,YAAI,CAAC,KAAK,KAAK,GAAG;AAAI;AACtB,aAAK,SAAS,GAAG;AACjB,YAAI;AAAA;AAAA;AAAA,IAIR,WAAW;AACT,aAAO,KAAK,gBAAgB,KAAK,cAAc;AAAA;AAAA,IAGjD,KAAK,GAAG;AACN,aAAO,KAAK,WAAW,KAAK,KAAK,WAAW;AAAA;AAAA,IAG9C,SAAS,GAAG;AACV,YAAM,IAAI,KAAK,cAAc;AAC7B,WAAK,cAAc,KAAK,KAAK,cAAc;AAC3C,WAAK,cAAc,KAAK;AAAA;AAAA;AAG5B,WAAQ,UAAU;AAAA;;;ACvElB;AAAA,QAAM,WAAW;AAEjB,uCAAqC,YAAY,OAAO,UAAU,UAAU,oBAAoB;AAC9F,UAAM,CAAC,QAAQ,SAAS,OAAO;AAC/B,QAAI,eAAe;AACnB,UAAM,SAAS,KAAK,IAAI,WAAW,oBAAoB;AACvD,UAAM,OAAO,KAAK,IAAI,WAAW,qBAAqB,GAAG;AACzD,aAAS,WAAW,QAAQ,WAAW,MAAM,EAAE;AAC7C,YAAM,SAAS,KAAK,IAAI,WAAW,oBAAoB;AACvD,YAAM,OAAO,KAAK,IAAI,WAAW,qBAAqB,GAAG;AACzD,eAAS,WAAW,QAAQ,WAAW,MAAM,EAAE;AAC7C,YAAI,OAAO,IAAI,UAAU,UAAU,cAAc;AAC/C,yBAAe;AACf;AAAA;AAAA;AAGJ,UAAI,CAAC;AACH;AAAA;AAAA;AAGJ,WAAO;AAAA;AAOT,mCAAiC,gBAAgB,oBAAoB;AACnE,UAAM,CAAC,QAAQ,OAAO,gBAAgB,OAAO;AAC7C,UAAM,QAAQ,IAAI,SAAS,QAAQ,SAAS,QAAQ,cAAc,CAAC,CAAE,WAAY;AACjF,aAAS,WAAW,GAAG,WAAW,QAAQ,EAAE;AAC1C,eAAS,WAAW,GAAG,WAAW,OAAO,EAAE;AACzC,iBAAS,aAAa,GAAG,aAAa,cAAc,EAAE;AACpD,gBAAM,QAAQ,OAAO,IAAI,UAAU,UAAU;AAE7C,cAAI,QAAQ;AAAgB;AAE5B,cAAI,4BAA4B,YAAY,OAAO,UAAU,UAAU,oBAAoB;AACzF,kBAAM,QAAQ,CAAE,OAAO,MAAM,CAAE,UAAU,UAAU,IAAI;AAAA;AAAA;AAAA;AAAA;AAK/D,WAAO;AAAA;AAET,WAAQ,0BAA0B;AAAA;;;AC7ClC;AAAA,WAAQ,YAAY;AAAA,IAClB;AAAA,IAAQ;AAAA,IAAW;AAAA,IAAY;AAAA,IAAW;AAAA,IAAY;AAAA,IACtD;AAAA,IAAiB;AAAA,IAAa;AAAA,IAAc;AAAA,IAAa;AAAA,IACzD;AAAA,IAAW;AAAA,IAAY;AAAA,IAAY;AAAA,IAAa;AAAA,IAAa;AAAA;AAE/D,WAAQ,gBAAgB,SAAQ,UAAU;AAC1C,WAAQ,UAAU,SAAQ,UAAU,OAAO,CAAC,QAAQ,WAAW;AAC7D,WAAO,aAAa;AACpB,WAAO;AAAA,KACN;AACH,QAAM,qBAAqB;AAAA,IACzB,CAAC,WAAW;AAAA,IAAiB,CAAC,aAAa;AAAA,IAC3C,CAAC,aAAa;AAAA,IAAc,CAAC,WAAW;AAAA,IACxC,CAAC,YAAY;AAAA,IAAc,CAAC,YAAY;AAAA,IACxC,CAAC,cAAc;AAAA,IAAkB,CAAC,cAAc;AAAA,IAChD,CAAC,YAAY;AAAA,IAAc,CAAC,aAAa;AAAA,IACzC,CAAC,gBAAgB;AAAA,IAAkB,CAAC,WAAW;AAAA;AAQjD,WAAQ,YAAY;AAAA,IAClB,CAAC,QAAQ;AAAA,IAAY,CAAC,WAAW;AAAA,IAAY,CAAC,QAAQ;AAAA,IACtD,CAAC,YAAY;AAAA,IAAa,CAAC,QAAQ;AAAA,IACnC,CAAC,gBAAgB;AAAA,IAAc,CAAC,aAAa;AAAA,IAC7C,CAAC,gBAAgB;AAAA,IAAY,CAAC,WAAW;AAAA,IACzC,CAAC,YAAY;AAAA,IAAc,CAAC,QAAQ;AAAA,IACpC,CAAC,iBAAiB;AAAA,IAAe,CAAC,cAAc;AAAA,IAChD,CAAC,iBAAiB;AAAA,IAAa,CAAC,YAAY;AAAA,IAC5C,CAAC,aAAa;AAAA;AAEhB,WAAQ,uBAAuB,mBAAmB,IAAI,CAAC,CAAC,YAAY,gBAAiB,CAAC,SAAQ,QAAQ,aAAa,SAAQ,QAAQ;AACnI,WAAQ,eAAe;AAAA,IACrB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA;AAAA;;;AC3DF;AAAA,QAAM,MAAM;AAEZ,0BAAwB,GAAG,GAAG,UAAU;AACtC,WAAO;AAAA,MACL,GAAG,QAAQ,IAAI,GAAG,GAAG;AAAA,MACrB,GAAG,QAAQ,IAAI,GAAG,GAAG,WAAW,IAAI;AAAA;AAAA;AAGxC,WAAQ,iBAAiB;AAEzB,0BAAwB,MAAM,cAAc;AAC1C,UAAM,CAAE,UAAU,UAAU,IAAI,YAAa;AAC7C,UAAM,CAAE,GAAG,KAAM,eAAe,UAAU,UAAU,UAAU;AAC9D,WAAO;AAAA,MACL,GAAG,KAAK,WAAW,eAAe;AAAA,MAClC,GAAG,KAAK,WAAW,eAAe;AAAA;AAAA;AAGtC,WAAQ,iBAAiB;AAEzB,qBAAmB,SAAS;AAC1B,UAAM,SAAS,IAAI,MAAM;AACzB,aAAS,IAAI,GAAG,IAAI,MAAM;AACxB,aAAO,KAAK;AAAA;AAEd,WAAO;AAAA;AAET,WAAQ,YAAY;AAEpB,iBAAe,GAAG,KAAK;AACrB,QAAI,IAAI;AAAK,aAAO;AACpB,QAAI,IAAI;AAAK,aAAO;AACpB,WAAO;AAAA;AAET,WAAQ,QAAQ;AAEhB,2BAAyB,IAAI,IAAI,IAAI;AACnC,UAAM,KAAK,KAAK;AAChB,UAAM,KAAK,KAAK;AAChB,WAAO,KAAK,KAAK,KAAK;AAAA;AAExB,WAAQ,kBAAkB;AAE1B,sBAAoB,GAAG;AACrB,WAAO,CAAE,GAAG,EAAE,IAAI,EAAE,GAAG,GAAG,EAAE,IAAI,EAAE;AAAA;AAEpC,WAAQ,aAAa;AAErB,uBAAqB,GAAG,KAAK;AAC3B,WAAO,CAAE,GAAG,MAAM,EAAE,GAAG,KAAK,MAAM,GAAG,MAAM,EAAE,GAAG,KAAK;AAAA;AAEvD,WAAQ,cAAc;AAAA;;;ACnDtB;AAAA,QAAM,YAAY;AAClB,QAAM,UAAU;AAEhB,QAAM,uBAAuB,UAAU,UAAU,IAAI,CAAC,CAAC,gBAAgB,mBAAoB,CAAC,UAAU,QAAQ,iBAAiB,UAAU,QAAQ;AACjJ,QAAM,qBAAqB,qBAAqB,IAAI,CAAC,CAAC,EAAE,kBAAkB;AAC1E,QAAM,qBAAqB,qBAAqB,IAAI,CAAC,CAAC,mBAAmB;AACzE,2BAAyB,QAAQ,OAAO;AACtC,UAAM,WAAW,cAAc,MAAM,KAAK;AAC1C,WAAO;AAAA,MACL,GAAG,cAAc,IAAI,MAAM,GAAG,MAAM,GAAG;AAAA,MACvC,GAAG,cAAc,IAAI,MAAM,GAAG,MAAM,GAAG,WAAW;AAAA;AAAA;AAGtD,oCAAkC,OAAO,cAAc,QAAQ;AAC7D,WAAO;AAAA,MACL,GAAG,QAAQ,MAAM,KAAK,MAAM,MAAM,IAAI,eAAe,GAAG,SAAS;AAAA,MACjE,GAAG,QAAQ,MAAM,KAAK,MAAM,MAAM,IAAI,eAAe,GAAG,QAAQ;AAAA;AAAA;AAUpE,oCAAkC,QAAQ,gBAAgB,kBAAkB,cAAc,SAAS,cAAc,eAAe,mBAAmB;AACjJ,UAAM,CAAC,QAAQ,SAAS,aAAa;AAErC,UAAM,wBAAwB,yBAAyB,eAAe,UAAU,cAAc,QAAQ;AACtG,UAAM,eAAe,gBAAgB,QAAQ,uBAAuB;AACpE,UAAM,iBAAiB,QAAQ,WAAW,eAAe,UAAU;AACnE,QAAI,iBAAiB;AACrB,aAAS,IAAI,GAAG,IAAI,kBAAkB;AACpC,YAAM,wBAAwB,yBAAyB,gBAAgB,cAAc,QAAQ;AAC7F,YAAM,cAAc,QAAQ,eAAe,sBAAsB,GAAG,sBAAsB,GAAG,kBAAkB;AAC/G,uBAAiB,QAAQ,WAAW;AAAA,QAClC,GAAG,sBAAsB,IAAI;AAAA,QAC7B,GAAG,sBAAsB,IAAI;AAAA,SAC5B,CAAE,GAAG,YAAY,GAAG,GAAG,YAAY;AAAA;AAExC,UAAM,wBAAwB,yBAAyB,gBAAgB,cAAc,QAAQ;AAC7F,UAAM,QAAQ,aAAa,IAAI,sBAAsB,GAAG,sBAAsB,GAAG;AACjF,WAAO,CAAE,UAAU,gBAAgB,MAAM,UAAU,UAAU,mBAAmB;AAAA;AAQlF,sBAAoB,MAAM,QAAQ,SAAS,cAAc,kBAAkB;AACzE,UAAM,WAAW,OAAO,MAAM;AAC9B,UAAM,WAAW,mBAAmB;AACpC,UAAM,oBAAoB,IAAI,MAAM;AAEpC,UAAM,CAAE,MAAM,UAAU,OAAO,aAAc;AAC7C,UAAM,YAAY,QAAQ,eAAe,UAAU,cAAc;AACjE,sBAAkB,SAAS,MAAM;AAAA,MAC/B,OAAO;AAAA,MACP,MAAM,UAAU,UAAU,SAAS;AAAA,MACnC,UAAU;AAAA;AAIZ,aAAS,OAAO,WAAW,GAAG,QAAQ,GAAG,EAAE;AACzC,YAAM,mBAAmB,mBAAmB;AAC5C,YAAM,mBAAmB,mBAAmB;AAC5C,UAAI,kBAAkB,qBAAqB,CAAC,kBAAkB;AAC5D,0BAAkB,oBAAoB,yBAAyB,MAAM,kBAAkB,mBAAmB,kBAAkB,QAAQ,SAAS,cAAc;AAAA;AAAA;AAK/J,aAAS,OAAO,GAAG,OAAO,UAAU,EAAE;AACpC,YAAM,mBAAmB,mBAAmB;AAC5C,YAAM,mBAAmB,mBAAmB;AAC5C,UAAI,kBAAkB,qBAAqB,CAAC,kBAAkB;AAC5D,0BAAkB,oBAAoB,yBAAyB,MAAM,kBAAkB,mBAAmB,kBAAkB,QAAQ,SAAS,cAAc;AAAA;AAAA;AAG/J,WAAO;AAAA;AAET,WAAQ,aAAa;AAAA;;;ACnFrB;AAAA,QAAM,aAAa;AACnB,QAAM,aAAa;AACnB,QAAM,UAAU;AAEhB,+CAA6C,OAAO,kBAAkB,CAAE,GAAG,IAAK;AAC9E,WAAO,MAAM,KAAK,CAAC,CAAE;AACnB,YAAM,wBAAwB,UAAU,YAAY;AACpD,aAAO,QAAQ,gBAAgB,GAAG,GAAG,sBAAsB,GAAG,sBAAsB,MAAM;AAAA;AAAA;AAO9F,4BAA0B,eAAe,kBAAkB;AACzD,UAAM,8BAA8B,kBAAkB,OAAO,CAAC,QAAQ,CAAE,UAAU,QAAS;AACzF,UAAI,CAAC,oCAAoC,eAAe,kBAAkB,UAAU;AAClF,kBAAU;AAAA;AAEZ,aAAO;AAAA,OACN;AACH,WAAO,8BAA8B,kBAAkB;AAAA;AAKzD,QAAM,sBAAsB;AAwD5B,+BAA6B,cAAc,eAAe,wBAAwB,wBAAwB,cAAc,mBAAmB,iBAAiB,KAAK,YAAY;AAC3K,UAAM,QAAQ;AACd,UAAM,QAAQ,WAAW,wBAAwB,gBAAgB,qBAAqB;AACtF,UAAM,mBAAmB,YAAY;AAGrC,WAAO,MAAM,SAAS,qBAAqB,CAAC,MAAM;AAEhD,YAAM,OAAO,MAAM;AAInB,YAAM,kBAAkB,QAAQ,eAAe,KAAK,MAAM,cAAc;AACxE,UAAI,oCAAoC,OAAO,kBAAkB,iBAAiB,KAAK,KAAK;AAAK;AAEjG,YAAM,YAAY,WAAW,WAAW,MAAM,cAAc,eAAe,cAAc,wBAAwB;AACjH,YAAM,QAAQ,iBAAiB,OAAO,kBAAkB;AACxD,YAAM,KAAK,CAAE,WAAW;AAAA;AAE1B,WAAO;AAAA;AAET,WAAQ,sBAAsB;AAAA;;;ACvG9B;AAAA,QAAM,MAAK;AACX,QAAM,MAAM;AAEZ,2CAAyC,GAAG,GAAG;AAC7C,WAAQ,IAAI,iBAAiB,IAAI;AAAA;AAGnC,gCAA8B,WAAW;AACvC,WAAO,IAAI,qBAAqB,OAAO,CAAC,QAAQ,CAAC,WAAW;AAC1D,UAAI,gCAAgC,UAAU,WAAW,OAAO,UAAU,YAAY,OAAO;AAC3F,eAAO;AAAA;AAET,aAAO,KAAK,CAAC,UAAU,YAAY,UAAU;AAC7C,aAAO;AAAA,OACN;AAAA;AAEL,WAAQ,uBAAuB;AAE/B,QAAM,CAAE,mBAAmB,qBAAsB;AACjD,0BAAwB;AACtB,WAAO,UAAU,OAAO,CAAC,CAAE,MAAM,MAAM,MAAM,OAAQ,CAAE,UAAU,CAAE,GAAG,QAAW;AAAA,MAC/E,MAAM,KAAK,IAAI,MAAM;AAAA,MACrB,MAAM,KAAK,IAAI,MAAM;AAAA,MACrB,MAAM,KAAK,IAAI,MAAM;AAAA,MACrB,MAAM,KAAK,IAAI,MAAM;AAAA,QACnB;AAAA,MACF,MAAM;AAAA,MACN,MAAM;AAAA,MACN,MAAM;AAAA,MACN,MAAM;AAAA;AAAA;AAGV,WAAQ,iBAAiB;AACzB,gCAA8B;AAC5B,UAAM,CAAE,MAAM,MAAM,MAAM,QAAS,eAAe;AAClD,WAAO,CAAC,CAAE,GAAG,MAAM,GAAG,OAAQ,CAAE,GAAG,MAAM,GAAG,OAAQ,CAAE,GAAG,MAAM,GAAG,OAAQ,CAAE,GAAG,MAAM,GAAG;AAAA;AAE1F,WAAQ,uBAAuB;AAC/B,mCAAiC;AAC/B,WAAO,QAAQ,IAAI,QAAQ,IAAI,CAAC,WAAW,OAAO;AAAA;AAEpD,WAAQ,oBAAoB;AAE5B,qBAAmB,MAAM,QAAQ,QAAQ,UAAU,GAAG,UAAU;AAC9D,WAAO;AAAA,MACL,OAAO,KAAK;AAAA,MACZ,WAAW,KAAK,UAAU,IAAI,CAAC,CAAE,OAAO,MAAM,cAAgB;AAAA,QAC5D;AAAA,QACA;AAAA,QACA,UAAU;AAAA,UACR,GAAG,SAAS,IAAI,SAAS;AAAA,UACzB,GAAG,SAAS,IAAI,SAAS;AAAA;AAAA;AAAA;AAAA;AAKjC,WAAQ,YAAY;AAEpB,sBAAoB,OAAO,QAAQ,QAAQ,UAAU,GAAG,UAAU;AAChE,QAAI,WAAW,KAAK,WAAW,KAAK,YAAY,KAAK,YAAY;AAC/D,aAAO;AAAA;AAET,WAAO,MAAM,IAAI,CAAC,SAAS,UAAU,MAAM,QAAQ,QAAQ,SAAS;AAAA;AAEtE,WAAQ,aAAa;AAErB,oCAAkC;AAChC,WAAO,iBAAiB,IAAG,SAAS,CAAC,MAAM,MAAM,IAAI,MAAM,MAAM,MAAM,CAAC,MAAM,QAAQ,MAAM;AAAA;AAE9F,WAAQ,2BAA2B;AAEnC,yBAAuB;AACrB,WAAO,iBAAiB,IAAG,SAAS,QAAQ,IAAG,QAAQ,WAAW;AAAA;AAEpE,WAAQ,gBAAgB;AAExB,gCAA8B,OAAO,cAAc;AACjD,WAAO,IAAG,KAAK;AACb,YAAM,cAAc,cAAc;AAClC,aAAO,YAAY,eAAe,CAAC,cAAc;AAAA;AAAA;AAGrD,WAAQ,uBAAuB;AAE/B,0BAAwB,OAAO,CAAC,SAAS;AACvC,UAAM,CAAC,QAAQ,SAAS,yBAAyB;AACjD,UAAM,eAAe,UAAU;AAC/B,UAAM,SAAS,QAAQ;AACvB,QAAI,CAAC,MAAM,MAAM,MAAM,QAAQ,CAAC,GAAG,GAAG,GAAG;AACzC,QAAI,SAAS;AAEX,aAAO;AACP,aAAO;AACP,aAAO,KAAK,MAAM,MAAO,gBAAe,SAAS;AACjD,aAAO,KAAK,MAAM,MAAO,gBAAe,SAAS;AAAA;AAGjD,aAAO,KAAK,MAAM,MAAQ,KAAM,eAAgB,QAAQ;AACxD,aAAO,KAAK,MAAM,MAAQ,KAAM,eAAgB,QAAQ;AACxD,aAAO;AACP,aAAO;AAAA;AAET,UAAM,UAAU,IAAG,KAAK;AACtB,UAAI,cAAc,cAAc;AAChC,oBAAc,IAAG,MAAM,aAAa,CAAC,CAAC,MAAM,OAAO,CAAC,MAAM,OAAO,CAAC,GAAG;AACrE,aAAO,YAAY,eAAe,CAAC,SAAS;AAAA;AAE9C,WAAO,CAAE,SAAS,SAAS,CAAE,KAAK,MAAM,MAAM,MAAM,OAAO,MAAM,QAAQ;AAAA;AAE3E,WAAQ,iBAAiB;AAEzB,6BAA2B,OAAO,CAAC,QAAQ,QAAQ,CAAC,uBAAuB,uBAAuB;AAChG,UAAM,SAAU,UAAS,QAAQ,MAAM,QAAQ,UAAW;AAC1D,UAAM,SAAU,SAAQ,QAAQ,OAAO,QAAQ,SAAU;AACzD,UAAM,cAAc,WAAW,OAAO,QAAQ,QAAQ,CAAC,QAAQ,KAAK,CAAC,QAAQ;AAC7E,WAAO;AAAA;AAET,WAAQ,oBAAoB;AAAA;;;ACrH5B;AAAA,QAAM,MAAK;AACX,QAAM,iBAAiB;AACvB,QAAM,iBAAiB;AACvB,QAAM,OAAO;AAEb;AAAA,IACE,YAAY;AACV,WAAK,YAAY;AAAA;AAAA,UAuBb,cAAc,OAAO;AACzB,YAAM,eAAe,QAAO;AAE5B,YAAM,CAAC,QAAQ,SAAS,KAAK,yBAAyB;AACtD,YAAM,CAAE,SAAS,WAAY,KAAK,eAAe,OAAO,CAAC,QAAO,iBAAiB,QAAO;AACxF,YAAM,CAAE,eAAe,SAAS,iBAAiB,mBAAoB,KAAK,UAAU,QAAQ;AAC5F,YAAM,mBAAmB,MAAM,KAAK,kBAAkB,CAAC,eAAe,SAAS,iBAAiB;AAChG,YAAM,eAAe,iBAAiB;AACtC,YAAM,gBAAgB,iBAAiB;AACvC,YAAM,yBAAyB,iBAAiB;AAChD,YAAM,yBAAyB,iBAAiB;AAChD,YAAM,QAAQ,MAAM,eAAe,oBAAoB,cAAc,eAAe,wBAAwB,wBAAwB,cAAc,QAAO,eAAe,QAAO,gBAAgB,QAAO;AACtM,YAAM,cAAc,KAAK,kBAAkB,OAAO,CAAC,QAAQ,QAAQ,CAAC,QAAO,iBAAiB,QAAO,kBAAkB;AACrH,oBAAc;AACd,cAAQ;AACR,sBAAgB;AAChB,sBAAgB;AAChB,cAAQ;AACR,aAAO;AAAA;AAAA,IAGT;AACE,WAAK,UAAU;AAAA;AAAA;AAGnB,WAAQ,UAAU;AAClB,+BAA6B;AAC3B,UAAM,aAAa,MAAM,IAAG,eAAe,QAAO;AAClD,UAAM,YAAY,IAAI,eAAe,UAAU,YAAY,QAAO;AAClE,WAAO,IAAI,QAAQ;AAAA;AAYrB,sBAAoB;AAClB,WAAO,cAAc;AAAA;AAEvB,WAAQ,OAAO;AAAA;;;AC1Ef;AAAA,QAAM,iBAAiB;AACvB,QAAM,eAAe;AACrB,QAAM,iBAAiB;AACvB,QAAM,YAAY;AAClB,QAAM,OAAO;AAEb,WAAQ,OAAO,aAAa;AAC5B,WAAQ,UAAU,aAAa;AAE/B,WAAQ,YAAY,eAAe;AACnC,WAAQ,sBAAsB,eAAe;AAC7C,WAAQ,eAAe,UAAU;AACjC,WAAQ,UAAU,UAAU;AAC5B,WAAQ,YAAY,UAAU;AAC9B,WAAQ,YAAY,UAAU;AAC9B,WAAQ,uBAAuB,KAAK;AACpC,WAAQ,iBAAiB,KAAK;AAC9B,WAAQ,uBAAuB,KAAK;AACpC,WAAQ,oBAAoB,KAAK;AACjC,WAAQ,YAAY,KAAK;AAAA;;;ACnBzB;AAAA,QAAM,MAAK;AAEX,sBAAoB;AAClB,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA,MAC1C,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA;AAAA;AAG9C,WAAQ,aAAa;AAErB,wBAAsB;AACpB,WAAO;AAAA,MACL,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA,MAC5D,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA;AAAA;AAGhE,WAAQ,eAAe;AAEvB,oCAAkC,KAAK,OAAO;AAC5C,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,QAAQ,CAAC;AAAA,MACb,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,SAAS,KAAK;AAAA,MAChE,IAAI,SAAS,KAAK;AAAA;AAEpB,WAAO,IAAG,MAAM,cAAc,OAAO,OAAO,CAAC,IAAI;AAAA;AAEnD,WAAQ,2BAA2B;AAEnC,+BAA6B,KAAK;AAChC,UAAM,aAAa,CAAC,IAAI,WAAW,KAAK,OAAO,IAAI,IAAI,WAAW,KAAK,OAAO;AAC9E,UAAM,WAAW,CAAC,IAAI,SAAS,KAAK,OAAO,IAAI,IAAI,SAAS,KAAK,OAAO;AACxE,UAAM,gBAAgB,IAAI,cAAc,IAAI,CAAC;AAC3C,YAAM,cAAc,CAAC,MAAM,KAAK,OAAO,IAAI,MAAM,KAAK,OAAO;AAC7D,aAAO;AAAA;AAET,WAAO,CAAE,YAAY,UAAU;AAAA;AAEjC,WAAQ,sBAAsB;AAE9B,sBAAoB,KAAK,SAAS;AAChC,UAAM,SAAS,aAAa;AAC5B,UAAM,OAAO,WAAW;AACxB,UAAM,cAAc,CAAC,SAAS,KAAK,KAAK,GAAG,SAAS,KAAK,KAAK;AAC9D,UAAM,aAAa,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACxE,UAAM,WAAW,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACtE,WAAO,CAAE,YAAY,UAAU,eAAe,IAAI;AAAA;AAEpD,WAAQ,aAAa;AAErB,uBAAqB;AACnB,UAAM,UAAU,aAAa;AAC7B,UAAM,OAAO,WAAW;AACxB,UAAM,UAAU,KAAK,IAAI,GAAG;AAC5B,UAAM,WAAW,UAAU;AAC3B,UAAM,aAAa,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACxD,UAAM,WAAW,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACtD,WAAO,CAAE,YAAY,UAAU,eAAe,IAAI;AAAA;AAEpD,WAAQ,cAAc;AAEtB,oBAAkB,KAAK;AACrB,UAAM,UAAU;AAAA,MACd,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA,MAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA;AAExE,UAAM,cAAc,CAAC,QAAQ,KAAK,YAAY,IAAI,QAAQ,KAAK,YAAY;AAC3E,UAAM,aAAa,CAAC,IAAI,WAAW,KAAK,YAAY,IAAI,IAAI,WAAW,KAAK,YAAY;AACxF,UAAM,WAAW,CAAC,IAAI,SAAS,KAAK,YAAY,IAAI,IAAI,SAAS,KAAK,YAAY;AAClF,WAAO,CAAE,YAAY,UAAU,eAAe,IAAI;AAAA;AAEpD,WAAQ,WAAW;AAAA;;;ACtEnB;AAAA,QAAM,MAAK;AACX,QAAM,WAAW;AAEjB;AAAA,IACE,YAAY,OAAO,SAAS;AAC1B,WAAK,QAAQ;AACb,WAAK,QAAQ,QAAO;AACpB,WAAK,SAAS,QAAO;AACrB,WAAK,UAAU,QAAQ,IAAI,CAAC,WAAW,CAAC,OAAO,UAAU,OAAO;AAChE,WAAK,gBAAgB,IAAG,SAAS,KAAK;AACtC,WAAK,kBAAkB,IAAG,SAAS,CAAC,QAAO,WAAW,QAAO;AAC7D,WAAK,wBAAwB,IAAG,SAAS,CAAC,QAAO,YAAY,GAAG,QAAO,YAAY;AAAA;AAAA,IAGrF,eAAe;AACb,aAAO,IAAG,KAAK;AACb,cAAM,aAAa,IAAG,MAAM,OAAO,CAAC,GAAG,IAAI,CAAC,IAAI;AAChD,cAAM,WAAW,IAAG,MAAM,OAAO,CAAC,GAAG,IAAI,CAAC,IAAI;AAC9C,cAAM,kBAAkB,IAAG,IAAI,IAAG,IAAI,YAAY,KAAK,kBAAkB,KAAK;AAC9E,cAAM,eAAe,IAAG,IAAI,UAAU,KAAK;AAC3C,cAAM,cAAc,IAAG,IAAI,IAAG,IAAI,iBAAiB,eAAe,KAAK;AACvE,cAAM,YAAY,IAAG,IAAI,IAAG,IAAI,iBAAiB,eAAe,KAAK;AACrE,eAAO,IAAG,SAAS,CAAC,aAAa,YAAY;AAAA;AAAA;AAAA,IAIjD,mBAAmB,kBAAkB;AACnC,aAAO,IAAG,KAAK;AACb,cAAM,YAAY,IAAG,IAAI,IAAG,IAAI,iBAAiB,QAAQ,CAAC,IAAI,GAAG,KAAK,KAAK,kBAAkB,KAAK,QAAQ;AAC1G,eAAO,IAAG,IAAI,WAAW,KAAK;AAAA;AAAA;AAAA,UAI5B,iBAAiB;AACrB,YAAM,kBAAkB,IAAG,KAAK,MAAM,IAAG,IAAI,IAAG,IAAI,OAAO,MAAM;AACjE,YAAM,oBAAoB,KAAK,MAAM,QAAQ;AAC7C,YAAM,aAAa,kBAAkB;AAErC,YAAM,SAAS,IAAG,KAAK,MAAM,IAAG,QAAQ,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI,KAAK;AAE/E,YAAM,WAAW,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACnD,YAAM,QAAQ,KAAK,eAAe;AAClC,YAAM,uBAAuB,MAAM,IAAG,MAAM,uBAAuB,OAAO,QAAQ,KAAK,UAAU,KAAK,cAAc,KAAK;AACzH,YAAM,iBAAiB,MAAM,qBAAqB;AAClD,YAAM,YAAY;AAAA,QAChB;AAAA,QAAiB;AAAA,QAAmB;AAAA,QAAsB;AAAA,QAC1D;AAAA,QAAO;AAAA,QAAU;AAAA;AAEnB,UAAI,eAAe,WAAW;AAC5B,kBAAU,QAAQ,CAAC,WAAW,OAAO;AACrC,eAAO;AAAA;AAET,YAAM,gBAAgB,IAAG,KAAK;AAC5B,cAAM,gBAAgB;AACtB,mBAAW,KAAK;AACd,gBAAM,WAAW,eAAe;AAChC,gBAAM,cAAc,IAAG,MAAM,OAAO,CAAC,UAAU,IAAI,CAAC,GAAG;AACvD,gBAAM,mBAAmB,IAAG,MAAM,YAAY,CAAC,UAAU,IAAI,CAAC,GAAG;AACjE,gBAAM,gBAAgB,IAAG,KAAK,MAAM,KAAK,mBAAmB,kBAAkB,UAAU,QAAQ,CAAC,IAAI;AACrG,wBAAc,KAAK,CAAE,OAAO,aAAa;AAAA;AAE3C,eAAO;AAAA;AAET,aAAO;AAAA;AAAA,UASH,mBAAmB,OAAO;AAC9B,YAAM,cAAc,MAAM,MAAM;AAChC,YAAM,aAAa,MAAM,MAAM;AAC/B,WAAK,eAAe,QAAO;AAC3B,WAAK,iBAAiB,QAAO;AAC7B,WAAK,WAAW,QAAO;AACvB,YAAM,QAAQ,IAAG,KAAK,MAAM,MAAM,eAAe,CAAC,KAAK,OAAO,KAAK,SAAS,IAAI;AAChF,YAAM,cAAc,MAAM,KAAK,iBAAiB;AAChD,YAAM;AACN,UAAI,CAAC,eAAgB,YAAY,WAAW;AAAI,eAAO;AACvD,YAAM,QAAQ;AACd,iBAAW,KAAK;AACd,cAAM,aAAa,YAAY;AAC/B,cAAM,gBAAgB,MAAM,WAAW,MAAM;AAC7C,cAAM,aAAa,cAAc,GAAG,MAAM,GAAG;AAC7C,cAAM,WAAW,cAAc,GAAG,MAAM,GAAG;AAC3C,cAAM,gBAAgB,MAAM,WAAW,cAAc;AACrD,mBAAW,MAAM;AACjB,mBAAW,cAAc;AACzB,cAAM,KAAK,SAAS,oBAAoB,CAAE,YAAY,UAAU,gBAAiB,CAAC,aAAa,KAAK,OAAO,cAAc,KAAK;AAAA;AAEhI,aAAO;AAAA;AAAA;AAGX,WAAQ,eAAe;AAAA;;;AChGvB;AAAA,WAAQ,mBAAmB;AAAA,IACzB,OAAO,CAAC,GAAG,GAAG,GAAG;AAAA,IACjB,aAAa,CAAC,GAAG,GAAG,GAAG;AAAA,IACvB,cAAc,CAAC,GAAG,IAAI,IAAI;AAAA,IAC1B,YAAY,CAAC,IAAI,IAAI,IAAI;AAAA,IACzB,OAAO,CAAC,IAAI,IAAI,IAAI;AAAA,IACpB,UAAU,CAAC;AAAA;AAAA;;;ACNb;AAAA,4BAA0B;AACxB,WAAO,QAAQ,IAAI,KAAK,KAAK,KAAK,MAAO,SAAQ,KAAK,MAAO,KAAI,KAAK;AAAA;AAExE,WAAQ,mBAAmB;AAE3B,2BAAyB,QAAQ;AAC/B,UAAM,UAAU,KAAK,KAAK,IAAI,KAAK,MAAM,CAAE,QAAO,KAAK,OAAO,KAAK,OAAO,KAAK,OAAO;AACtF,WAAO,iBAAiB;AAAA;AAE1B,WAAQ,kBAAkB;AAE1B,QAAM,yBAAyB,CAAC,GAAG,MAAO,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG;AACxE,eAAa,IAAI;AACf,QAAI,UAAU;AACd,aAAS,IAAI,GAAG,IAAI,GAAG,QAAQ;AAC7B,iBAAW,GAAG,KAAK,GAAG;AAAA;AAExB,WAAO;AAAA;AAET,WAAQ,MAAM;AAEd,8BAA4B,KAAK;AAC/B,UAAM,SAAS;AACf,aAAS,IAAI,GAAG,IAAI,IAAI,QAAQ;AAC9B,aAAO,KAAK,IAAI,GAAG;AAAA;AAErB,WAAO;AAAA;AAET,WAAQ,qBAAqB;AAE7B,qCAAmC,MAAM;AACvC,UAAM,UAAU;AAChB,UAAM,OAAO,KAAK;AAClB,aAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,cAAQ,KAAK;AACb,eAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,gBAAQ,KAAK,KAAK,IAAI,KAAK,MAAM,mBAAmB,MAAM;AAAA;AAAA;AAG9D,WAAO;AAAA;AAET,+BAA6B,UAAU;AACrC,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,iBAAiB,CAAC,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,GAAG,GAAG;AAClE,UAAM,oBAAoB,uBAAuB,OAAO,IAAI,OAAO;AACnE,UAAM,2BAA2B,0BAA0B,mBAAmB;AAC9E,UAAM,4BAA4B,uBAAuB,CAAC,OAAO,IAAI,CAAC,OAAO;AAC7E,WAAO,0BAA0B,0BAA0B;AAAA;AAE7D,WAAQ,sBAAsB;AAE9B,iCAA+B;AAC7B,UAAM,oBAAoB,CAAC,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG,KAAK,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AAClF,UAAM,uBAAuB,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AACtD,UAAM,sBAAsB;AAAA,MAC1B,CAAC,IAAI,kBAAkB,IAAI;AAAA,MAC3B,CAAC,IAAI,kBAAkB,IAAI;AAAA;AAE7B,WAAO;AAAA,MACL,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,CAAC,GAAG,GAAG;AAAA;AAAA;AAGX,WAAQ,wBAAwB;AAEhC,uBAAqB,uBAAuB;AAC1C,WAAO;AAAA,MACL,IAAI,uBAAuB,eAAe;AAAA,MAC1C,IAAI,uBAAuB,eAAe;AAAA;AAAA;AAG9C,WAAQ,cAAc;AAAA;;;ACzEtB;AAAA,QAAM,MAAK;AACX,QAAM,WAAW;AACjB,QAAM,OAAO;AAEb,QAAM,0CAA0C;AAChD,QAAM,wBAAwB,CAAC,GAAG;AAClC,QAAM,wBAAwB,CAAC,GAAG;AAClC,QAAM,0BAA0B;AAChC,QAAM,oBAAoB,CAAC,GAAG,GAAG,GAAG,IAAI,IAAI,GAAG;AAC/C,QAAM,oCAAoC;AAC1C,QAAM,6CAA6C;AAGnD;AAAA,IACE,YAAY,qBAAqB,cAAc;AAC7C,WAAK,oBAAoB;AACzB,WAAK,0BAA0B;AAC/B,WAAK,sBAAsB;AAC3B,WAAK,eAAe;AACpB,WAAK,YAAY,QAAO;AACxB,WAAK,aAAa,QAAO;AACzB,WAAK,gBAAgB,QAAO;AAAA;AAAA,IAI9B,uBAAuB,eAAe;AACpC,YAAM,uBAAuB,cAAc,IAAI,CAAC;AAC9C,cAAM,wBAAwB,CAAC,GAAG,OAAO;AACzC,eAAO,KAAK,YAAY,uBAAuB;AAAA;AAEjD,YAAM,gBAAgB,KAAK,8BAA8B;AAGzD,aAAO,SAAS,WAAW,SAAS,YAAY,SAAS,SAAS,eAAe,yBAAyB,KAAK;AAAA;AAAA,IAIjH,uBAAuB;AAIrB,YAAM,cAAc,KAAK,8BAA8B;AACvD,YAAM,gBAAgB,SAAS,WAAW,SAAS,YAAY,SAAS,SAAS,aAAa,yBAAyB;AACvH,YAAM,gBAAgB;AACtB,eAAS,IAAI,GAAG,IAAI,kBAAkB,QAAQ;AAC5C,sBAAc,KAAK,UAAU,kBAAkB,IAAI,MAAM,GAAG;AAAA;AAE9D,oBAAc,gBAAgB;AAC9B,aAAO;AAAA;AAAA,IAKT,mBAAmB,WAAW,KAAK,OAAO;AACxC,YAAM,UAAU,SAAS,WAAW;AACpC,YAAM,cAAc,CAAC,QAAQ,KAAK,KAAK,WAAW,QAAQ,KAAK,KAAK;AACpE,YAAM,eAAe,UAAU,IAAI,CAAC,UAAU;AAAA,QAC5C,YAAY,KAAM,OAAM,KAAK,KAAK,YAAY;AAAA,QAC9C,YAAY,KAAM,OAAM,KAAK,KAAK,aAAa;AAAA,QAAI,MAAM;AAAA;AAE3D,YAAM,uBAAuB,KAAK,oBAAoB,OAAO,CAAC,GAAG;AACjE,YAAM,gBAAgB,aAAa,IAAI,CAAC;AACtC,cAAM,UAAU,KAAK,YAAY,OAAO;AACxC,eAAO,CAAC,GAAG,SAAS,MAAM;AAAA;AAE5B,YAAM,wBAAwB,KAAK,sBAAsB;AACzD,YAAM,YAAY,CAAC,GAAG,SAAS,aAAa,MAAM;AAClD,YAAM,oBAAoB;AAAA,QACxB,KAAK,IAAI,WAAW,sBAAsB;AAAA,QAC1C,KAAK,IAAI,WAAW,sBAAsB;AAAA;AAE5C,aAAO,cAAc,IAAI,CAAC,UAAU;AAAA,QAClC,MAAM,KAAK,kBAAkB;AAAA,QAAI,MAAM,KAAK,kBAAkB;AAAA,QAC9D,MAAM;AAAA;AAAA;AAAA,UAIJ,cAAc,OAAO;AACzB,WAAK,sBAAsB,QAAO;AAClC,WAAK,sBAAsB,QAAO;AAClC,WAAK,WAAW,QAAO;AACvB,YAAM,cAAc,KAAK;AACzB,UAAI,gBAAgB;AAClB,cAAM,yBAAyB,MAAM,KAAK,oBAAoB,mBAAmB,OAAO;AACxF,aAAK,oBAAoB;AACzB,mBAAW,KAAK;AACd,eAAK,wBAAwB,uBAAuB,IAAI,MAAyB;AAAA;AAEnF,aAAK,0BAA0B;AAAA;AAE/B,aAAK;AAAA;AAGP,YAAM,QAAQ;AACd,UAAI,CAAC,KAAK;AAAmB,eAAO;AACpC,iBAAW,KAAK,KAAK;AACnB,cAAM,aAAa,KAAK,kBAAkB,GAAG;AAC7C,YAAI,CAAC;AAAY,iBAAO;AACxB,cAAM,QAAQ,KAAK,gBAAgB,WAAW,cAAc,oCAAoC,WAAW,cAAc;AACzH,cAAM,aAAa,SAAS,aAAa;AACzC,cAAM,uBAAuB,CAAC,WAAW,KAAK,MAAM,MAAM,IAAI,WAAW,KAAK,MAAM,MAAM;AAC1F,cAAM,eAAe,IAAG,MAAM,iBAAiB,OAAO,OAAO,GAAG;AAChE,cAAM,iBAAiB,KAAK,oBAAoB,CAAC,OAAO;AACxD,cAAM,MAAM,cAAc,KAAK,uBAAuB,WAAW,eAAe,kBAAkB;AAClG,cAAM,eAAe,SAAS,yBAAyB,KAAK,cAAc,CAAC,KAAK,WAAW,KAAK;AAChG,cAAM,YAAY,aAAa,IAAI;AACnC,qBAAa;AACb,qBAAa;AACb,cAAM,aAAa,KAAK,aAAa,QAAQ;AAC7C,cAAM,CAAC,MAAM,aAAa;AAC1B,kBAAU;AACV,cAAM,YAAY,KAAK,WAAW;AAClC,aAAK;AACL,YAAI,YAAY,QAAO;AACrB,oBAAU;AACV,eAAK,kBAAkB,KAAK;AAC5B,iBAAO;AAAA;AAET,cAAM,oBAAoB,IAAG,QAAQ,WAAW,CAAC,IAAI;AACrD,cAAM,YAAY,MAAM,kBAAkB;AAC1C,kBAAU;AACV,0BAAkB;AAClB,cAAM,SAAS,KAAK,mBAAmB,WAAW,KAAK,OAAO;AAC9D,cAAM,kBAAkB,KAAK,uBAAuB;AACpD,aAAK,wBAAwB,iBAAiB,OAA2B;AACzE,cAAM,SAAS;AAAA,UACb,WAAW;AAAA,UACX,YAAY;AAAA,UACZ,KAAK;AAAA,YACH,SAAS,gBAAgB;AAAA,YACzB,aAAa,gBAAgB;AAAA;AAAA;AAGjC,cAAM,KAAK;AAAA;AAEb,aAAO;AAAA;AAAA,IAIT,8BAA8B;AAC5B,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,aAAa,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AACjD,YAAM,WAAW,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AAC/C,aAAO,CAAE,YAAY;AAAA;AAAA,IAKvB,wBAAwB,KAAK,aAAa;AACxC,UAAI;AACF,aAAK,kBAAkB,SAAS,CAAC;AAAA;AAEjC,cAAM,cAAc,KAAK,kBAAkB,OAAO;AAClD,YAAI,MAAM;AACV,YAAI,eAAe,QAAQ,YAAY,cAAc;AACnD,gBAAM,CAAC,WAAW,aAAa,IAAI;AACnC,gBAAM,CAAC,SAAS,WAAW,IAAI;AAC/B,gBAAM,CAAC,mBAAmB,qBAAqB,YAAY;AAC3D,gBAAM,CAAC,iBAAiB,mBAAmB,YAAY;AACvD,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,eAAgB,WAAU,aAAc,WAAU;AACxD,gBAAM,UAAW,WAAU,aAAc,WAAU;AACnD,gBAAM,kBAAmB,mBAAkB,qBAAsB,mBAAkB;AACnF,gBAAM,eAAgB,WAAU,kBAAkB;AAAA;AAEpD,aAAK,kBAAkB,OAAO,KAAK,MAAM,0CAA0C,cAAc;AAAA;AAAA;AAAA,IAIrG;AACE,aAAO,CAAC,KAAK,qBAAsB,KAAK,kBAAkB,WAAW,KAAO,KAAK,2BAA2B,KAAK;AAAA;AAAA;AAGrH,WAAQ,eAAe;AAAA;;;ACjLvB;AAAA,QAAM,MAAK;AACX,QAAM,OAAO;AACb,QAAM,YAAY;AAClB,QAAM,OAAO;AAEb;AAAA,IACE,YAAY;AACV,WAAK,WAAW;AAAA;AAAA,UAGZ,cAAc,OAAO;AACzB,WAAK,sBAAsB,QAAO;AAClC,WAAK,sBAAsB,QAAO;AAClC,WAAK,WAAW,QAAO;AACvB,YAAM,QAAQ,IAAG,KAAK;AACpB,YAAI,CAAE,kBAAiB,IAAG;AACxB,kBAAQ,IAAG,QAAQ,WAAW;AAAA;AAEhC,eAAO,MAAM,UAAU,WAAW;AAAA;AAEpC,YAAM,cAAc,MAAM,KAAK,SAAS,cAAc,OAAO;AAC7D,YAAM;AACN,YAAM,QAAQ;AACd,UAAI,CAAC;AAAa,eAAO;AACzB,iBAAW,cAAc;AACvB,YAAI,CAAC;AAAY,iBAAO;AACxB,cAAM,cAAc;AACpB,mBAAW,OAAO,OAAO,KAAK,UAAU;AACtC,sBAAY,OAAO,UAAU,iBAAiB,KAAK,IAAI,CAAC,UAAU,WAAW,UAAU;AAAA;AAEzF,cAAM,KAAK;AAAA,UACT,YAAY,WAAW,cAAc;AAAA,UACrC,KAAK,WAAW,MAAM,CAAC,WAAW,IAAI,QAAQ,IAAI,WAAW,IAAI,QAAQ,IAAI,WAAW,IAAI,YAAY,KAAK,WAAW,IAAI,QAAQ,IAAI,WAAW,IAAI,YAAY,KAAK,WAAW,IAAI,QAAQ,MAAM;AAAA,UACrM,WAAW,WAAW;AAAA,UACtB;AAAA;AAAA;AAGJ,aAAO;AAAA;AAAA;AAGX,WAAQ,WAAW;AAEnB,6BAA2B;AACzB,QAAI,IAAG,MAAM,SAAS;AAEpB,YAAM,KAAK;AACX,YAAM,OAAO,MAAM,GAAG,aAAa,IAAI,QAAQ,WAAW;AAC1D,aAAO,KAAK,MAAM;AAAA;AAEpB,WAAO,IAAG,KAAK,MAAM,KAAK,KAAK,CAAC,MAAM,EAAE;AAAA;AAG1C,sBAAoB;AAClB,UAAM,CAAC,SAAS,mBAAmB,iBAAiB,MAAM,QAAQ,IAAI;AAAA,MACpE,YAAY,QAAO,SAAS;AAAA,MAC5B,IAAG,eAAe,QAAO,SAAS,WAAW,CAAE,WAAW,QAAO,SAAS,UAAU,SAAS;AAAA,MAC7F,IAAG,eAAe,QAAO,SAAS,WAAW,CAAE,WAAW,QAAO,SAAS,UAAU,SAAS;AAAA;AAE/F,UAAM,WAAW,IAAI,KAAK,aAAa,mBAAmB,SAAS;AACnE,UAAM,WAAW,IAAI,KAAK,aAAa,UAAU,eAAe;AAChE,UAAM,YAAW,IAAI,SAAS;AAC9B,WAAO;AAAA;AAET,WAAQ,OAAO;AAAA;;;AC/Df;AAAA;AAAA;AAAA;AAAA,MAAO,iBAAQ;AAAA,IACb,SAAS;AAAA,IACT,SAAS;AAAA,IACT,MAAM;AAAA,MACJ,SAAS;AAAA,MACT,UAAU;AAAA,QACR,WAAW;AAAA,QACX,WAAW;AAAA,QACX,UAAU;AAAA,QACV,YAAY;AAAA,QACZ,eAAe;AAAA,QACf,cAAc;AAAA,QACd,gBAAgB;AAAA;AAAA,MAElB,MAAM;AAAA,QACJ,SAAS;AAAA,QACT,WAAW;AAAA,QACX,WAAW;AAAA;AAAA,MAEb,MAAM;AAAA,QACJ,SAAS;AAAA,QACT,WAAW;AAAA,QACX,eAAe;AAAA,QACf,WAAW;AAAA;AAAA,MAEb,KAAK;AAAA,QACH,SAAS;AAAA,QACT,WAAW;AAAA,QACX,WAAW;AAAA,QACX,YAAY;AAAA;AAAA,MAEd,QAAQ;AAAA,QACN,SAAS;AAAA,QACT,WAAW;AAAA;AAAA,MAEb,SAAS;AAAA,QACP,SAAS;AAAA,QACT,WAAW;AAAA,QACX,eAAe;AAAA,QACf,YAAY;AAAA,QACZ,cAAc;AAAA,QACd,WAAW;AAAA;AAAA;AAAA,IAGf,MAAM;AAAA,MACJ,SAAS;AAAA,MACT,WAAW;AAAA,MACX,iBAAiB;AAAA,MACjB,cAAc;AAAA,MACd,eAAe;AAAA,MACf,gBAAgB;AAAA,MAChB,WAAW;AAAA;AAAA,IAEb,MAAM;AAAA,MACJ,SAAS;AAAA,MACT,WAAW;AAAA,MACX,YAAY;AAAA,MACZ,eAAe;AAAA,MACf,cAAc;AAAA,MACd,gBAAgB;AAAA,MAChB,eAAe;AAAA,MACf,UAAU;AAAA,MACV,UAAU;AAAA,QACR,SAAS;AAAA,QACT,WAAW;AAAA;AAAA,MAEb,UAAU;AAAA,QACR,WAAW;AAAA;AAAA;AAAA;AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACnEjB,MAAM,KAAK;AACX,MAAM,WAAW;AACjB,MAAM,SAAS;AACf,MAAM,UAAU;AAChB,MAAM,UAAU;AAChB,MAAM,WAAW;AACjB,MAAM,WAAW,iBAAuB;AACxC,MAAM,MAAM;AAEZ,IAAI;AAGJ,MAAM,SAAS;AAAA,EACb,UAAU;AAAA,EACV,SAAS;AAAA,EACT,UAAU;AAAA,EACV,MAAM;AAAA,EACN,KAAK;AAAA,EACL,QAAQ;AAAA,EACR,SAAS;AAAA;AAEX,MAAM,MAAM;AACV,MAAI,OAAO,gBAAgB;AAAa,WAAO,YAAY;AAC3D,SAAO,SAAS,OAAO,QAAQ,OAAO,YAAY,MAAO;AAAA;AAG3D,MAAM,MAAM,IAAI;AAEd,MAAI,OAAO;AAAS,YAAQ,IAAI,GAAG;AAAA;AAIrC,sBAAsB;AACpB,QAAM,WAAW,CAAC,QAAQ,OAAO,OAAO,QAAQ;AAChD,SAAO,QAAQ,OAAO,CAAC,MAAM;AAC3B,WAAO,KAAK,OAAO,IAAI,QAAQ,CAAC;AAC9B,YAAM,OAAO,KAAK;AAClB,YAAM,OAAO,IAAI;AACjB,UAAI,MAAM,QAAQ,SAAS,MAAM,QAAQ;AACvC,aAAK,OAAO,KAAK,OAAO,GAAG;AAAA,iBAClB,SAAS,SAAS,SAAS;AACpC,aAAK,OAAO,UAAU,MAAM;AAAA;AAE5B,aAAK,OAAO;AAAA;AAAA;AAGhB,WAAO;AAAA,KACN;AAAA;AAGL,gBAAgB;AACd,MAAI,CAAC;AAAO,WAAO;AACnB,QAAM,QAAQ,MAAM,gBAAgB,MAAM,cAAc,MAAM,SAAU,MAAM,SAAU,MAAM,MAAM,KAAK;AACzG,MAAI,CAAC,SAAU,UAAU;AAAI,WAAO;AACpC,MAAI,MAAM,cAAe,MAAM,cAAc;AAAI,WAAO;AACxD;AACE,OAAG;AAAA;AAEH,WAAO;AAAA;AAET,SAAO;AAAA;AAGT,sBAAsB,OAAO;AAC3B,WAAS,UAAU,UAAU;AAG7B,QAAM,QAAQ,OAAO;AACrB,MAAI;AACF,QAAI,OAAO;AACX,WAAO,CAAE;AAAA;AAIX,SAAO,IAAI,QAAQ,OAAO;AAExB,UAAM,eAAe,OAAO,OAAO,QAAQ,OAAO,CAAC,MAAM,GAAG;AAC5D,QAAI,iBAAiB;AAAG,UAAI;AAG5B,QAAI,GAAG,iBAAiB,OAAO;AAC7B,UAAI,kCAAkC,OAAO;AAC7C,YAAM,GAAG,WAAW,OAAO;AAC3B,YAAM,GAAG;AAAA;AAGX,QAAI;AACJ,QAAI,GAAG,iBAAiB;AACtB,wCAAkC,GAAG,MAAM,IAAI;AAC/C,SAAG,MAAM,IAAI,4BAA4B;AAAA;AAI3C,QAAI,OAAO,KAAK,WAAW,CAAC,OAAO;AAAU,aAAO,WAAW,MAAM,SAAS,KAAK,OAAO;AAC1F,QAAI,OAAO,KAAK,WAAW,CAAC,OAAO;AAAS,aAAO,UAAU,MAAM,QAAQ,KAAK,OAAO;AACvF,QAAI,OAAO,KAAK,WAAW,CAAC,OAAO;AAAU,aAAO,WAAW,MAAM,SAAS,KAAK,OAAO;AAC1F,QAAI,OAAO,KAAK,WAAW,OAAO,KAAK,IAAI,WAAW,CAAC,OAAO;AAAK,aAAO,MAAM,MAAM,OAAO,QAAQ;AACrG,QAAI,OAAO,KAAK,WAAW,OAAO,KAAK,OAAO,WAAW,CAAC,OAAO;AAAQ,aAAO,SAAS,MAAM,OAAO,WAAW;AACjH,QAAI,OAAO,KAAK,WAAW,OAAO,KAAK,QAAQ,WAAW,CAAC,OAAO;AAAS,aAAO,UAAU,MAAM,QAAQ,KAAK;AAE/G,UAAM,OAAO;AACb,QAAI;AAGJ,gBAAY;AACZ,OAAG,SAAS;AACZ,UAAM,UAAU,OAAO,KAAK,UAAU,MAAM,OAAO,QAAQ,cAAc,OAAO,OAAO,QAAQ;AAC/F,OAAG,SAAS;AACZ,SAAK,OAAO,KAAK,MAAM,QAAQ;AAG/B,gBAAY;AACZ,OAAG,SAAS;AACZ,UAAM,UAAU,OAAO,KAAK,UAAU,MAAM,OAAO,SAAS,cAAc,OAAO,OAAO,QAAQ;AAChG,OAAG,SAAS;AACZ,SAAK,OAAO,KAAK,MAAM,QAAQ;AAG/B,UAAM,UAAU;AAChB,QAAI,OAAO,KAAK;AACd,kBAAY;AACZ,SAAG,SAAS;AACZ,YAAM,QAAQ,MAAM,OAAO,SAAS,cAAc,OAAO,OAAO;AAChE,WAAK,OAAO,KAAK,MAAM,QAAQ;AAC/B,iBAAW,QAAQ;AAEjB,YAAI,CAAC,KAAK,SAAS,KAAK,MAAM;AAC5B,cAAI,4BAA4B,KAAK;AACrC;AAAA;AAGF,oBAAY;AACZ,cAAM,UAAW,OAAO,KAAK,IAAI,WAAW,OAAO,KAAK,OAAO,UAAW,MAAM,OAAO,QAAQ,KAAK,OAAO,UAAU;AACrH,aAAK,YAAY,KAAK,MAAM,QAAQ;AAEpC,oBAAY;AACZ,cAAM,cAAc,OAAO,KAAK,QAAQ,UAAU,MAAM,QAAQ,QAAQ,KAAK,OAAO,UAAU;AAC9F,aAAK,UAAU,KAAK,MAAM,QAAQ;AAClC,aAAK,MAAM;AAGX,cAAM,OAAQ,KAAK,YAAY,eAAe,KAAK,YAAY,eAC3D,KAAK,IAAI,KAAK,YAAY,YAAY,GAAG,KAAK,KAAK,YAAY,YAAY,GAAG,IAAI,KAAK,YAAY,aAAa,GAAG,KAAK,KAAK,YAAY,aAAa,GAAG,MACzJ;AACJ,gBAAQ,KAAK;AAAA,UACX,YAAY,KAAK;AAAA,UACjB,KAAK,KAAK;AAAA,UACV,MAAM,KAAK;AAAA,UACX,aAAa,KAAK;AAAA,UAClB,KAAK,QAAQ;AAAA,UACb,QAAQ,QAAQ;AAAA,UAChB,SAAS;AAAA,UACT,MAAO,SAAS,IAAK,KAAK,MAAM,MAAM,OAAmC,QAAQ,MAAM;AAAA;AAAA;AAG3F,SAAG,SAAS;AAAA;AAId,OAAG,MAAM,IAAI,4BAA4B;AAGzC,SAAK,QAAQ,OAAO,OAAO,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI;AACtD,YAAQ,CAAE,MAAM,SAAS,MAAM,SAAS,MAAM,SAAS,aAAa;AAAA;AAAA;AAIxE,QAAQ,SAAS;AACjB,QAAQ,WAAW;AACnB,QAAQ,SAAS;AACjB,QAAQ,SAAS;AACjB,QAAQ,WAAW;AACnB,QAAQ,SAAS;AACjB,QAAQ,UAAU;AAClB,QAAQ,WAAW;AACnB,QAAQ,KAAK;AACb,QAAQ,UAAU,IAAI;",
+  "sourcesContent": ["const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  if (frame > config.face.age.skipFrames) {\n    frame = 0;\n  } else {\n    frame += 1;\n  }\n  if (frame === 0) return last;\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;\n    if (confidence > config.face.gender.minConfidence) {\n      obj.gender = data[0] <= 0.5 ? 'female' : 'male';\n      obj.confidence = confidence;\n    }\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "/* eslint-disable indent */\n/* eslint-disable no-multi-spaces */\n\nexport default {\n  backend: 'webgl',          // select tfjs backend to use\n  console: true,             // enable debugging output to console\n  face: {\n    enabled: true,           // controls if specified modul is enabled\n                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion\n                             // (note: module is not loaded until it is required)\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.\n                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.\n      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'\n      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192,        // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3,    // empiric tuning\n      inputSize: 64,         // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'\n                                                         // which determines training set for model\n      inputSize: 64,         // fixed value\n      skipFrames: 10,        // how many frames to go without re-running the detector\n    },\n    gender: {\n      enabled: true,\n      minConfidence: 0.8,    // threshold for discarding a prediction\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64,         // fixed value\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      skipFrames: 10,        // how many frames to go without re-running the detector\n      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257,    // fixed value\n    outputStride: 16,        // fixed value\n    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256,          // fixed value\n    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n    minConfidence: 0.5,      // threshold for discarding a prediction\n    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace\n    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    // let savedWebglPackDepthwiseConvFlag;\n    // if (tf.getBackend() === 'webgl') {\n    //   savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n    //  tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    // }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          agConfidence: ssrData.confidence,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    // tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
+  "mappings": ";;;;;;;;;;;;;;;;;;AAAA;AAAA,QAAM,MAAK;AAEX,QAAM,gBAAgB;AAEtB,2BAAyB;AACvB,UAAM,OAAO,CAAE,SAAS,CAAC,YAAY,IAAI,YAAY,IAAI,SAAS,CAAC,GAAG;AACtE,UAAM,UAAU;AAChB,aAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,QAAQ;AACvC,YAAM,SAAS,KAAK,QAAQ;AAC5B,YAAM,WAAW,KAAK,MAAO,aAAY,SAAS,KAAK;AACvD,YAAM,WAAW,KAAK,MAAO,aAAY,SAAS,KAAK;AACvD,YAAM,aAAa,KAAK,QAAQ;AAChC,eAAS,QAAQ,GAAG,QAAQ,UAAU;AACpC,cAAM,UAAU,SAAU,SAAQ;AAClC,iBAAS,QAAQ,GAAG,QAAQ,UAAU;AACpC,gBAAM,UAAU,SAAU,SAAQ;AAClC,mBAAS,IAAI,GAAG,IAAI,YAAY;AAC9B,oBAAQ,KAAK,CAAC,SAAS;AAAA;AAAA;AAAA;AAAA;AAK/B,WAAO;AAAA;AAGT,QAAM,aAAa,CAAC;AAClB,QAAI,eAAe;AACnB,QAAI,WAAW;AACf,QAAI,SAAS;AAAA;AAGf,QAAM,YAAY,CAAC,mBAAoB;AAAA,IACrC;AAAA,IACA,YAAY,IAAG,MAAM,gBAAgB,CAAC,GAAG,IAAI,CAAC,IAAI;AAAA,IAClD,UAAU,IAAG,MAAM,gBAAgB,CAAC,GAAG,IAAI,CAAC,IAAI;AAAA;AAGlD,QAAM,WAAW,CAAC,KAAK;AACrB,UAAM,SAAS,IAAG,IAAI,IAAI,YAAY;AACtC,UAAM,OAAO,IAAG,IAAI,IAAI,UAAU;AAClC,UAAM,iBAAiB,IAAG,SAAS,CAAC,QAAQ,OAAO;AACnD,WAAO,UAAU;AAAA;AAGnB,wBAAsB,YAAY,SAAS;AACzC,UAAM,YAAY,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACpD,UAAM,UAAU,IAAG,IAAI,WAAW;AAClC,UAAM,WAAW,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACnD,UAAM,qBAAqB,IAAG,IAAI,UAAU;AAC5C,UAAM,oBAAoB,IAAG,IAAI,SAAS;AAC1C,UAAM,cAAc,IAAG,IAAI,oBAAoB;AAC/C,UAAM,SAAS,IAAG,IAAI,mBAAmB;AACzC,UAAM,OAAO,IAAG,IAAI,mBAAmB;AACvC,UAAM,kBAAkB,IAAG,IAAI,QAAQ;AACvC,UAAM,gBAAgB,IAAG,IAAI,MAAM;AACnC,UAAM,aAAa;AACnB,WAAO,IAAG,SAAS,CAAC,iBAAiB,gBAAgB;AAAA;AAGvD,kCAAgC,MAAM;AACpC,WAAO,IAAG,KAAK;AACb,YAAM,MAAM,KAAK,SAAS,KAAK,SAAS;AACxC,aAAO,SAAS,KAAK,aAAa,eAAe;AAAA;AAAA;AAIrD;AAAA,IACE,YAAY,OAAO;AACjB,WAAK,iBAAiB;AACtB,WAAK,QAAQ,QAAO,SAAS;AAC7B,WAAK,SAAS,QAAO,SAAS;AAC9B,WAAK,WAAW,QAAO,SAAS;AAChC,WAAK,cAAc,gBAAgB,QAAO,SAAS;AACnD,WAAK,UAAU,IAAG,SAAS,KAAK;AAChC,WAAK,YAAY,IAAG,SAAS,CAAC,KAAK,OAAO,KAAK;AAC/C,WAAK,eAAe,QAAO,SAAS;AACpC,WAAK,aAAa;AAClB,WAAK,iBAAiB,QAAO,SAAS;AAAA;AAAA,UAGlC,iBAAiB;AAErB,UAAK,CAAC,cAAgB,WAAW,sBAAwB,WAAW,MAAM,WAAW,KAAO,WAAW,MAAM,KAAK,KAAO,WAAW,MAAM,KAAK;AAAI,eAAO;AAC1J,YAAM,CAAC,iBAAiB,OAAO,UAAU,IAAG,KAAK;AAC/C,cAAM,eAAe,WAAW,eAAe,CAAC,KAAK,OAAO,KAAK;AACjE,cAAM,kBAAkB,IAAG,IAAI,IAAG,IAAI,aAAa,IAAI,MAAM,MAAM;AACnE,cAAM,oBAAoB,KAAK,eAAe,QAAQ;AACtD,YAAI;AAEJ,YAAI,MAAM,QAAQ;AAChB,gBAAM,SAAS,kBAAkB,KAAK,CAAC,GAAG,MAAM,EAAE,OAAO,EAAE;AAC3D,gBAAM,YAAY,IAAG,OAAO,CAAC,OAAO,IAAI,OAAO,KAAK;AACpD,gBAAM,YAAY,IAAG,OAAO,CAAC,OAAO,IAAI,OAAO,KAAK;AACpD,gBAAM,SAAS,IAAG,OAAO,CAAC,WAAW,YAAY;AACjD,uBAAa,OAAO,QAAQ;AAAA;AAE5B,uBAAa,kBAAkB;AAAA;AAEjC,cAAM,gBAAgB,aAAa,YAAY,KAAK,SAAS,KAAK;AAClE,cAAM,SAAS,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACjD,cAAM,YAAY,IAAG,QAAQ,QAAQ;AACrC,eAAO,CAAC,YAAY,eAAe;AAAA;AAGrC,YAAM,mBAAmB,MAAM,IAAG,MAAM,uBAAuB,OAAO,QAAQ,KAAK,UAAU,KAAK,cAAc,KAAK;AACrH,YAAM,aAAa,MAAM,iBAAiB;AAC1C,uBAAiB;AACjB,UAAI,gBAAgB,WAAW,IAAI,CAAC,aAAa,IAAG,MAAM,OAAO,CAAC,UAAU,IAAI,CAAC,GAAG;AACpF,sBAAgB,MAAM,QAAQ,IAAI,cAAc,IAAI,OAAO;AACzD,cAAM,OAAO,MAAM,YAAY;AAC/B,oBAAY;AACZ,eAAO;AAAA;AAET,YAAM,iBAAiB;AACvB,eAAS,IAAI,GAAG,IAAI,cAAc,QAAQ;AACxC,cAAM,cAAc,cAAc;AAClC,cAAM,eAAe,IAAG,KAAK;AAC3B,gBAAM,MAAM,UAAU;AACtB,gBAAM,WAAW,WAAW;AAC5B,gBAAM,SAAS,KAAK,YAAY;AAChC,gBAAM,YAAY,IACf,MAAM,iBAAiB,CAAC,UAAU,gBAAgB,IAAI,CAAC,GAAG,KAC1D,UACA,QAAQ,CAAC,eAAe;AAC3B,gBAAM,cAAc,IAAG,MAAM,QAAQ,CAAC,WAAW,CAAC;AAClD,iBAAO,CAAE,KAAK,WAAW,aAAa;AAAA;AAExC,uBAAe,KAAK;AAAA;AAEtB,YAAM;AACN,aAAO;AACP,sBAAgB;AAChB,aAAO;AAAA,QACL,OAAO;AAAA,QACP,aAAa,CAAC,WAAW,MAAM,KAAK,KAAK,OAAO,WAAW,MAAM,KAAK,KAAK;AAAA;AAAA;AAAA,UAIzE,cAAc;AAClB,YAAM,QAAQ,IAAG,KAAK;AACpB,YAAI,CAAE,kBAAiB,IAAG;AACxB,kBAAQ,IAAG,QAAQ,WAAW;AAAA;AAEhC,eAAO,MAAM,UAAU,WAAW;AAAA;AAEpC,YAAM,CAAE,OAAO,eAAgB,MAAM,KAAK,iBAAiB;AAC3D,YAAM;AACN,aAAO,QAAQ,IAAI,MAAM,IAAI,OAAO;AAClC,cAAM,YAAY,uBAAuB,MAAM;AAC/C,cAAM,CAAC,cAAc,SAAS,mBAAmB,MAAM,QAAQ,IAAI,CAAC,KAAK,WAAW,WAAW,KAAK,aAAa,IAAI,OAAO,MAAM,EAAE;AACpI,cAAM,SAAS,KAAK;AACpB,cAAM,CAAC,cAAc,gBAAgB;AACrC,cAAM,kBAAkB,aACrB,IAAI,CAAC,aAAc;AAAA,UACjB,UAAS,KAAK,OAAO,MAAM;AAAA,UAC3B,UAAS,KAAK,OAAO,MAAM;AAAA;AAEhC,cAAM,iBAAiB;AAAA,UACrB,SAAS,QAAQ,MAAM,GAAG;AAAA,UAC1B,aAAa,QAAQ,MAAM;AAAA,UAC3B,WAAW;AAAA,UACX,aAAa;AAAA;AAEf,mBAAW,KAAK;AAChB,aAAK,UAAU;AACf,aAAK,YAAY;AACjB,kBAAU;AACV,eAAO;AAAA;AAAA;AAAA;AAKb,sBAAoB;AAClB,UAAM,YAAY,MAAM,IAAG,eAAe,QAAO,SAAS,WAAW,CAAE,WAAW,QAAO,SAAS,UAAU,SAAS;AACrH,UAAM,QAAQ,IAAI,eAAe,WAAW;AAC5C,WAAO;AAAA;AAGT,WAAQ,OAAO;AACf,WAAQ,iBAAiB;AACzB,WAAQ,aAAa;AAAA;;;ACpLrB;AAAA,WAAQ,mBAAmB;AAAA,IACzB,YAAY;AAAA,MACV;AAAA,MAAI;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MACtD;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MACvD;AAAA,MAAK;AAAA,MAAI;AAAA,MAAK;AAAA,MAAI;AAAA,MAAK;AAAA,MAAK;AAAA,MAAK;AAAA,MAAI;AAAA,MAAI;AAAA,MAAK;AAAA,MAAI;AAAA;AAAA,IAEpD,gBAAgB,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI,GAAG,KAAK,KAAK,KAAK,KAAK;AAAA,IAC7D,gBAAgB,CAAC,KAAK,IAAI,KAAK,IAAI,IAAI,KAAK,KAAK,KAAK,KAAK;AAAA,IAC3D,gBAAgB,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI,IAAI,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9D,gBAAgB,CAAC,IAAI,IAAI,IAAI,KAAK,IAAI,IAAI,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9D,gBAAgB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC/C,gBAAgB,CAAC,IAAI,GAAG,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACtD,gBAAgB,CAAC,KAAK,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC1C,gBAAgB,CAAC,KAAK,IAAI,KAAK,IAAI,IAAI,IAAI,IAAI,KAAK;AAAA,IACpD,gBAAgB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC/C,gBAAgB,CAAC,KAAK,IAAI,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,gBAAgB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACzD,mBAAmB,CAAC,KAAK,IAAI,IAAI,KAAK,IAAI,KAAK,IAAI;AAAA,IACnD,mBAAmB,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI;AAAA,IACzC,cAAc,CAAC,KAAK,KAAK,KAAK,KAAK;AAAA,IACnC,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9C,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9C,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC9C,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,eAAe,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACxD,kBAAkB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IACtD,kBAAkB,CAAC,KAAK,KAAK,KAAK,KAAK,KAAK;AAAA,IAC5C,aAAa,CAAC,KAAK,KAAK,KAAK,KAAK;AAAA,IAClC,mBAAmB,CAAC;AAAA,IACpB,SAAS,CAAC;AAAA,IACV,YAAY,CAAC;AAAA,IACb,iBAAiB,CAAC;AAAA,IAClB,gBAAgB,CAAC;AAAA,IACjB,YAAY,CAAC;AAAA,IACb,WAAW,CAAC;AAAA;AAEd,WAAQ,2BAA2B;AAAA,IACjC,CAAE,KAAK,aAAa,SAAS,CAAC,GAAG,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IACrD,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IACtD,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IACtD,CAAE,KAAK,aAAa,SAAS,CAAC,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG;AAAA,IACtD,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC9D,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC9D,CAAE,KAAK,aAAa,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC9D,CAAE,KAAK,gBAAgB,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA,IAC7D,CAAE,KAAK,gBAAgB,SAAS,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI;AAAA;AAAA;;;AC/CvD;AAAA,QAAM,MAAK;AAEX,+BAA6B,KAAK;AAChC,UAAM,aAAa,CAAC,IAAI,WAAW,KAAK,OAAO,IAAI,IAAI,WAAW,KAAK,OAAO;AAC9E,UAAM,WAAW,CAAC,IAAI,SAAS,KAAK,OAAO,IAAI,IAAI,SAAS,KAAK,OAAO;AACxE,WAAO,CAAE,YAAY;AAAA;AAEvB,WAAQ,sBAAsB;AAC9B,sBAAoB;AAClB,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA,MAC1C,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA;AAAA;AAG9C,WAAQ,aAAa;AACrB,wBAAsB;AACpB,WAAO;AAAA,MACL,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA,MAC5D,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA;AAAA;AAGhE,WAAQ,eAAe;AACvB,oCAAkC,KAAK,OAAO;AAC5C,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,QAAQ,CAAC;AAAA,MACb,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,SAAS,KAAK;AAAA,MAChE,IAAI,SAAS,KAAK;AAAA;AAEpB,WAAO,IAAG,MAAM,cAAc,OAAO,OAAO,CAAC,IAAI;AAAA;AAEnD,WAAQ,2BAA2B;AACnC,sBAAoB,KAAK,SAAS;AAChC,UAAM,SAAS,aAAa;AAC5B,UAAM,OAAO,WAAW;AACxB,UAAM,cAAc,CAAC,SAAS,KAAK,KAAK,GAAG,SAAS,KAAK,KAAK;AAC9D,UAAM,aAAa,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACxE,UAAM,WAAW,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACtE,WAAO,CAAE,YAAY,UAAU,WAAW,IAAI;AAAA;AAEhD,WAAQ,aAAa;AACrB,uBAAqB;AACnB,UAAM,UAAU,aAAa;AAC7B,UAAM,OAAO,WAAW;AACxB,UAAM,UAAU,KAAK,IAAI,GAAG;AAC5B,UAAM,WAAW,UAAU;AAC3B,UAAM,aAAa,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACxD,UAAM,WAAW,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACtD,WAAO,CAAE,YAAY,UAAU,WAAW,IAAI;AAAA;AAEhD,WAAQ,cAAc;AAAA;;;AClDtB;AAAA,WAAQ,kBAAkB,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG;AAKxD,4BAA0B;AACxB,WAAO,QAAQ,IAAI,KAAK,KAAK,KAAK,MAAO,SAAQ,KAAK,MAAO,KAAI,KAAK;AAAA;AAExE,WAAQ,mBAAmB;AAM3B,2BAAyB,QAAQ;AAC/B,UAAM,UAAU,KAAK,KAAK,IAAI,KAAK,MAAM,CAAE,QAAO,KAAK,OAAO,KAAK,OAAO,KAAK,OAAO;AACtF,WAAO,iBAAiB;AAAA;AAE1B,WAAQ,kBAAkB;AAC1B,wBAAsB;AACpB,WAAO,MAAM,MAAM,KAAK;AAAA;AAE1B,WAAQ,eAAe;AACvB,kCAAgC,GAAG;AACjC,WAAO,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG;AAAA;AAEvC,eAAa,IAAI;AACf,QAAI,UAAU;AACd,aAAS,IAAI,GAAG,IAAI,GAAG,QAAQ;AAC7B,iBAAW,GAAG,KAAK,GAAG;AAAA;AAExB,WAAO;AAAA;AAET,WAAQ,MAAM;AACd,8BAA4B,KAAK;AAC/B,UAAM,SAAS;AACf,aAAS,IAAI,GAAG,IAAI,IAAI,QAAQ;AAC9B,aAAO,KAAK,IAAI,GAAG;AAAA;AAErB,WAAO;AAAA;AAET,WAAQ,qBAAqB;AAC7B,qCAAmC,MAAM;AACvC,UAAM,UAAU;AAChB,UAAM,OAAO,KAAK;AAClB,aAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,cAAQ,KAAK;AACb,eAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,gBAAQ,KAAK,KAAK,IAAI,KAAK,MAAM,mBAAmB,MAAM;AAAA;AAAA;AAG9D,WAAO;AAAA;AAET,+BAA6B,UAAU;AACrC,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,iBAAiB,CAAC,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,GAAG,GAAG;AAClE,UAAM,oBAAoB,uBAAuB,OAAO,IAAI,OAAO;AACnE,UAAM,2BAA2B,0BAA0B,mBAAmB;AAC9E,UAAM,4BAA4B,uBAAuB,CAAC,OAAO,IAAI,CAAC,OAAO;AAC7E,WAAO,0BAA0B,0BAA0B;AAAA;AAE7D,WAAQ,sBAAsB;AAC9B,iCAA+B;AAC7B,UAAM,oBAAoB,CAAC,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG,KAAK,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AAClF,UAAM,uBAAuB,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AACtD,UAAM,sBAAsB;AAAA,MAC1B,CAAC,IAAI,kBAAkB,IAAI;AAAA,MAC3B,CAAC,IAAI,kBAAkB,IAAI;AAAA;AAE7B,WAAO;AAAA,MACL,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,CAAC,GAAG,GAAG;AAAA;AAAA;AAGX,WAAQ,wBAAwB;AAChC,uBAAqB,uBAAuB;AAC1C,WAAO;AAAA,MACL,IAAI,uBAAuB,eAAe;AAAA,MAC1C,IAAI,uBAAuB,eAAe;AAAA;AAAA;AAG9C,WAAQ,cAAc;AACtB,mCAAiC,GAAG;AAClC,WAAO,KAAK,KAAO,GAAE,KAAK,EAAE,OAAO,IAAO,GAAE,KAAK,EAAE,OAAO;AAAA;AAE5D,WAAQ,0BAA0B;AAAA;;;ACvFlC;AACA,QAAM,MAAK;AACX,QAAM,WAAW;AACjB,QAAM,YAAY;AAClB,QAAM,OAAO;AAEb,QAAM,kBAAkB;AACxB,QAAM,0CAA0C;AAChD,QAAM,mBAAmB;AACzB,QAAM,0CAA0C,CAAC,kBAAkB,UAAU,iBAAiB,qBAAqB;AACnH,QAAM,wBAAwB;AAC9B,QAAM,uBAAuB;AAC7B,QAAM,+CAA+C,CAAC,uBAAuB;AAC7E,QAAM,mBAAmB,UAAU,iBAAiB;AACpD,QAAM,kBAAkB,CAAC,iBAAiB,IAAI,iBAAiB,iBAAiB,SAAS;AACzF,QAAM,oBAAoB,UAAU,iBAAiB;AACrD,QAAM,mBAAmB,CAAC,kBAAkB,IAAI,kBAAkB,kBAAkB,SAAS;AAC7F,QAAM,0BAA0B;AAChC,QAAM,0BAA0B;AAChC,QAAM,kBAAkB;AACxB,QAAM,uBAAuB;AAG7B,iCAA+B,WAAW,WAAW,QAAQ;AAC3D,aAAS,IAAI,GAAG,IAAI,UAAU,yBAAyB,QAAQ;AAC7D,YAAM,CAAE,KAAK,WAAY,UAAU,yBAAyB;AAC5D,YAAM,kBAAkB,UAAU,iBAAiB,GAAG,SAAS;AAC/D,YAAM,uBAAuB,QAAQ;AACrC,UAAI,wBAAwB,KAAK,SAAS;AACxC,iBAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ;AAClC,gBAAM,QAAQ,QAAQ;AACtB,oBAAU,gBAAgB,MAAM;AAAA,YAC9B,UAAU,OAAO;AAAA,YAAI,UAAU,OAAO;AAAA,YACrC,WAAU,OAAO,KAAK,UAAU,gBAAgB,IAAI,MAAM;AAAA;AAAA;AAAA;AAAA;AAAA;AAOrE;AAAA,IACE,YAAY,qBAAqB,cAAc,WAAW;AAExD,WAAK,oBAAoB;AACzB,WAAK,0BAA0B;AAC/B,WAAK,sBAAsB;AAC3B,WAAK,eAAe;AACpB,WAAK,YAAY;AACjB,WAAK,YAAY,QAAO,KAAK;AAC7B,WAAK,aAAa,QAAO,KAAK;AAC9B,WAAK,WAAW,QAAO,KAAK;AAC5B,WAAK,cAAc,QAAO,KAAK;AAAA;AAAA,IAGjC,mBAAmB,WAAW,KAAK,OAAO;AACxC,YAAM,UAAU,SAAS,WAAW,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI;AAChF,YAAM,cAAc,CAAC,QAAQ,KAAK,KAAK,WAAW,QAAQ,KAAK,KAAK;AACpE,YAAM,eAAe,UAAU,IAAI,CAAC,UAAW;AAAA,QAC7C,YAAY,KAAM,OAAM,KAAK,KAAK,YAAY;AAAA,QAC9C,YAAY,KAAM,OAAM,KAAK,KAAK,aAAa;AAAA,QAAI,MAAM;AAAA;AAE3D,YAAM,uBAAuB,KAAK,oBAAoB,OAAO,CAAC,GAAG;AACjE,YAAM,gBAAgB,aAAa,IAAI,CAAC,UAAW,CAAC,GAAG,KAAK,YAAY,OAAO,uBAAuB,MAAM;AAC5G,YAAM,wBAAwB,KAAK,sBAAsB;AACzD,YAAM,YAAY,CAAC,GAAG,SAAS,aAAa,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI,YAAa;AACrG,YAAM,oBAAoB;AAAA,QACxB,KAAK,IAAI,WAAW,sBAAsB;AAAA,QAC1C,KAAK,IAAI,WAAW,sBAAsB;AAAA;AAE5C,aAAO,cAAc,IAAI,CAAC,UAAW;AAAA,QACnC,MAAM,KAAK,kBAAkB;AAAA,QAC7B,MAAM,KAAK,kBAAkB;AAAA,QAAI,MAAM;AAAA;AAAA;AAAA,IAI3C,iCAAiC;AAC/B,YAAM,WAAW,UAAU,gBAAgB,IAAI;AAC/C,YAAM,YAAY,UAAU,iBAAiB,IAAI;AACjD,aAAO,WAAW;AAAA;AAAA,IAIpB,UAAU,WAAW,MAAM,qBAAqB,qBAAqB,OAAO;AAC1E,YAAM,MAAM,SAAS,YAAY,SAAS,WAAW,KAAK,8BAA8B,CAAC,UAAU,sBAAsB,UAAU,wBAAwB,KAAK;AAChK,YAAM,UAAU,SAAS,WAAW;AACpC,UAAI,OAAO,IAAG,MAAM,cAAc,MAAM,CAAC;AAAA,QACvC,IAAI,WAAW,KAAK,KAAK;AAAA,QACzB,IAAI,WAAW,KAAK,KAAK;AAAA,QAAW,IAAI,SAAS,KAAK,KAAK;AAAA,QAC3D,IAAI,SAAS,KAAK,KAAK;AAAA,UACrB,CAAC,IAAI,CAAC,KAAK,UAAU,KAAK;AAC9B,UAAI;AACF,eAAO,IAAG,MAAM,cAAc;AAAA;AAEhC,aAAO,CAAE,KAAK,SAAS;AAAA;AAAA,IAIzB,aAAa,SAAS,QAAQ,YAAY,OAAO;AAC/C,YAAM,eAAe;AACrB,eAAS,IAAI,GAAG,IAAI,sBAAsB;AACxC,cAAM,IAAI,QAAQ,IAAI;AACtB,cAAM,IAAI,QAAQ,IAAI,IAAI;AAC1B,cAAM,IAAI,QAAQ,IAAI,IAAI;AAC1B,qBAAa,KAAK;AAAA,UACf,QACI,IAAK,IAAI,KAAK,WACd,IAAI,KAAK,YAAa,WAAW,KAAK,OAAO,WAAW;AAAA,UAC5D,IAAI,KAAK,WAAY,WAAW,KAAK,OAAO,WAAW;AAAA,UAAI;AAAA;AAAA;AAGhE,aAAO,CAAE,WAAW,cAAc,MAAM,aAAa,MAAM;AAAA;AAAA,IAI7D,sBAAsB,WAAW,YAAY;AAC3C,YAAM,eAAe,UAAU,UAAU,iBAAiB,GAAG,sBAAsB,0BAA0B;AAC7G,YAAM,eAAe,UAAU,UAAU,iBAAiB,GAAG,sBAAsB,0BAA0B;AAC7G,YAAM,WAAY,gBAAe,gBAAgB;AAEjD,aAAO,WAAW,IAAI,CAAC,OAAO;AAC5B,YAAI,IAAI;AACR,YAAI,MAAM;AACR,cAAI;AAAA,mBACK,MAAM;AACf,cAAI;AAAA;AAEN,eAAO,CAAC,MAAM,IAAI,MAAM,IAAI;AAAA;AAAA;AAAA,UAI1B,QAAQ,OAAO;AACnB,WAAK,aAAa,QAAO,SAAS;AAClC,WAAK,WAAW,QAAO,SAAS;AAChC,UAAI,KAAK;AACP,cAAM,CAAE,OAAO,eAAgB,MAAM,KAAK,oBAAoB,iBAAiB;AAC/E,YAAI,MAAM,WAAW;AACnB,eAAK,oBAAoB;AACzB,iBAAO;AAAA;AAET,cAAM,cAAc,MAAM,IAAI,CAAC;AAC7B,gBAAM,gBAAgB;AAAA,YACpB,YAAY,WAAW,IAAI,WAAW,UAAU;AAAA,YAChD,UAAU,WAAW,IAAI,SAAS,UAAU;AAAA;AAE9C,qBAAW,IAAI,WAAW;AAC1B,qBAAW,IAAI,SAAS;AACxB,gBAAM,YAAY,SAAS,oBAAoB,eAAe;AAC9D,gBAAM,cAAc,SAAS,WAAW;AACxC,gBAAM,YAAY,WAAW,UAAU;AACvC,qBAAW,UAAU;AACrB,qBAAW,YAAY;AACvB,iBAAO,IAAK,aAAa;AAAA;AAE3B,aAAK,wBAAwB;AAC7B,aAAK,0BAA0B;AAAA;AAE/B,aAAK;AAAA;AAEP,YAAM,UAAU,IAAG,KAAK,MAAM,KAAK,kBAAkB,IAAI,CAAC,KAAK;AAC7D,YAAI,QAAQ;AAEZ,cAAM,4BAA4B,IAAI,UAAU,UAAU;AAC1D,YAAI,CAAC,cAAc,mBAAmB;AACtC,YAAI,8BAA8B;AAChC,WAAC,cAAc,mBAAmB;AAAA;AAEpC,gBAAQ,KAAK,gBAAgB,IAAI,UAAU,eAAe,IAAI,UAAU;AACxE,cAAM,aAAa,SAAS,aAAa,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI;AACrF,cAAM,uBAAuB,CAAC,WAAW,KAAK,MAAM,MAAM,IAAI,WAAW,KAAK,MAAM,MAAM;AAC1F,YAAI,eAAe;AACnB,YAAI,iBAAiB,KAAK;AAC1B,YAAI,UAAU;AACZ,yBAAe,IAAG,MAAM,iBAAiB,OAAO,OAAO,GAAG;AAC1D,2BAAiB,KAAK,oBAAoB,CAAC,OAAO;AAAA;AAEpD,cAAM,SAAS,CAAE,YAAY,IAAI,YAAY,UAAU,IAAI;AAC3D,cAAM,OAAO,SAAS,yBAAyB,QAAQ,cAAc,CAAC,KAAK,YAAY,KAAK,YAAY,IAAI;AAE5G,cAAM,CAAC,EAAE,MAAM,UAAU,KAAK,aAAa,QAAQ;AACnD,cAAM,iBAAiB,IAAG,QAAQ,QAAQ,CAAC,IAAI;AAC/C,YAAI,YAAY,eAAe;AAC/B,YAAI,QAAO,KAAK;AACd,gBAAM,CAAE,KAAK,YAAY,SAAS,gBAAgB,MAAM,eAAgB,KAAK,UAAU,WAAW,MAAM,gBAAgB,IAAI,gBAAgB,IAAI;AAChJ,gBAAM,CAAE,KAAK,aAAa,SAAS,iBAAiB,MAAM,gBAAiB,KAAK,UAAU,WAAW,MAAM,iBAAiB,IAAI,iBAAiB;AACjJ,gBAAM,iBAAkB,KAAK,UAAU,QAAQ,IAAG,OAAO,CAAC,aAAa;AACvE,gBAAM,qBAAqB,eAAe;AAC1C,yBAAe;AACf,gBAAM,cAAc,mBAAmB,MAAM,GAAG,uBAAuB;AACvE,gBAAM,CAAE,WAAW,kBAAkB,MAAM,qBAAsB,KAAK,aAAa,aAAa,YAAY,gBAAgB;AAC5H,gBAAM,eAAe,mBAAmB,MAAM,uBAAuB;AACrE,gBAAM,CAAE,WAAW,mBAAmB,MAAM,sBAAuB,KAAK,aAAa,cAAc,aAAa;AAChH,gBAAM,gCAAgC,KAAK,iCAAiC;AAC5E,cAAI,KAAK,IAAI,iCAAiC;AAC5C,kCAAsB,WAAW,kBAAkB;AACnD,kCAAsB,WAAW,mBAAmB;AAAA,qBAE3C,gCAAgC;AACzC,kCAAsB,WAAW,kBAAkB,QAAQ,CAAC,aAAa;AAAA;AAEzE,kCAAsB,WAAW,mBAAmB,SAAS,CAAC,aAAa;AAAA;AAE7E,gBAAM,yBAAyB,KAAK,sBAAsB,WAAW,mBAAmB;AACxF,gBAAM,0BAA0B,KAAK,sBAAsB,WAAW,oBAAoB;AAC1F,sBAAY,UAAU,OAAO,wBAAwB,OAAO;AAAA;AAE9D,cAAM,wBAAwB,KAAK,mBAAmB,WAAW,KAAK,OAAO;AAC7E,YAAG,QAAQ;AACX,cAAM,eAAe,SAAS,WAAW,KAAK,8BAA8B;AAC5E,YAAI,QAAO,KAAK;AACd,gBAAM,oBAAoB,IAAG,SAAS;AACtC,eAAK,kBAAkB,KAAK,IAAK,cAAc,WAAW,kBAAkB;AAC5E,gBAAM,cAAa;AAAA,YACjB,QAAQ;AAAA,YACR,KAAK;AAAA,YACL,YAAY,KAAK;AAAA,YACjB,OAAO;AAAA;AAET,iBAAO;AAAA;AAET,cAAM,aAAa;AAAA,UACjB,QAAQ;AAAA,UACR,KAAK;AAAA,UACL,YAAY,KAAK;AAAA,UACjB,OAAO;AAAA;AAET,eAAO;AAAA;AAET,aAAO;AAAA;AAAA,IAIT,wBAAwB;AACtB,eAAS,IAAI,GAAG,IAAI,MAAM,QAAQ;AAChC,cAAM,MAAM,MAAM;AAClB,cAAM,cAAc,KAAK,kBAAkB;AAC3C,YAAI,MAAM;AACV,YAAI,eAAe,YAAY;AAC7B,gBAAM,CAAC,WAAW,aAAa,IAAI;AACnC,gBAAM,CAAC,SAAS,WAAW,IAAI;AAC/B,gBAAM,CAAC,mBAAmB,qBAAqB,YAAY;AAC3D,gBAAM,CAAC,iBAAiB,mBAAmB,YAAY;AACvD,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,eAAgB,WAAU,aAAc,WAAU;AACxD,gBAAM,UAAW,WAAU,aAAc,WAAU;AACnD,gBAAM,kBAAmB,mBAAkB,qBAAsB,mBAAkB;AACnF,gBAAM,eAAgB,WAAU,kBAAkB;AAAA;AAEpD,YAAI,MAAM;AACR,eAAK,kBAAkB,KAAK;AAAA;AAAA;AAGhC,WAAK,oBAAoB,KAAK,kBAAkB,MAAM,GAAG,MAAM;AAAA;AAAA,IAGjE,sBAAsB;AACpB,UAAI,KAAK,kBAAkB,UAAU;AACnC,aAAK,oBAAoB;AAAA,UACvB,GAAG,KAAK,kBAAkB,MAAM,GAAG;AAAA,UACnC,GAAG,KAAK,kBAAkB,MAAM,QAAQ;AAAA;AAAA;AAAA;AAAA,IAK9C;AACE,YAAM,YAAY,KAAK,kBAAkB;AACzC,YAAM,SAAS,cAAc;AAC7B,UAAI,KAAK,aAAa,KAAK;AACzB,eAAO;AAAA;AAET,aAAO,cAAc,KAAK,YAAY,KAAK,2BAA2B,KAAK;AAAA;AAAA,IAG7E,8BAA8B;AAC5B,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,aAAa,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AACjD,YAAM,WAAW,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AAC/C,aAAO,CAAE,YAAY;AAAA;AAAA;AAGzB,WAAQ,WAAW;AAAA;;;AC3RnB;AAAA,WAAQ,YAAY;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,kBAAkB;AAAA,IACnB,CAAC,gBAAgB;AAAA,IACjB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,kBAAkB;AAAA,IACnB,CAAC,iBAAiB;AAAA,IAClB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA,IACpB,CAAC,mBAAmB;AAAA;AAAA;;;ACpdtB;AAAA;AAAA;AAAA;AAAA,MAAO,wBAAQ;AAAA,IACb;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IACpE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACtE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACxE;AAAA,IAAG;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACvE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAC1E;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACzE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACxE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACzE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACzE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAC1E;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IACpE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAI;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACpE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACrE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACvE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAI;AAAA,IAAG;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACxE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IACtE;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA,IAAK;AAAA;AAAA;;;ACxKnE;AAAA,QAAM,MAAK;AACX,QAAM,YAAY;AAClB,QAAM,YAAY;AAClB,QAAM,OAAO;AACb,QAAM,YAAY;AAClB,QAAM,gBAAgB,wBAA2B;AAEjD;AAAA,IACE,YAAY,WAAW,gBAAgB,WAAW;AAChD,WAAK,WAAW,IAAI,KAAK,SAAS,WAAW,gBAAgB,WAAW;AACxE,UAAI;AAAQ,aAAK,SAAS;AAAA;AAAA,UAGtB,cAAc,OAAO;AACzB,UAAI;AAAQ,aAAK,SAAS;AAC1B,YAAM,QAAQ,IAAG,KAAK;AACpB,YAAI,CAAE,kBAAiB,IAAG;AAAS,kBAAQ,IAAG,QAAQ,WAAW;AACjE,eAAO,MAAM,UAAU,WAAW;AAAA;AAEpC,YAAM,cAAc,MAAM,KAAK,SAAS,QAAQ,OAAO;AACvD,UAAG,QAAQ;AACX,YAAM,UAAU;AAChB,iBAAW,cAAe,eAAe;AACvC,cAAM,aAAa,WAAW,WAAW;AACzC,YAAI,cAAc,KAAK,OAAO,SAAS;AACrC,gBAAM,OAAO,WAAW,SAAS,WAAW,OAAO,cAAc;AACjE,gBAAM,cAAc;AACpB,cAAI,QAAQ,KAAK,SAAS;AACxB,uBAAW,OAAO,UAAU;AAC1B,kBAAI,KAAK,OAAO,KAAK,WAAW,IAAI,SAAS,YAAY;AACvD,4BAAY,OAAO,UAAU,iBAAiB,KAAK,IAAI,CAAC,UAAU,KAAK;AAAA;AAAA;AAAA;AAI7E,kBAAQ,KAAK;AAAA,YACX,YAAY,cAAc;AAAA,YAC1B,KAAK,WAAW,MAAM,CAAC,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,SAAS,KAAK,WAAW,IAAI,WAAW,IAAI,WAAW,IAAI,SAAS,KAAK,WAAW,IAAI,WAAW,MAAM;AAAA,YAC3M;AAAA,YACA;AAAA,YACA,OAAO,WAAW,QAAQ,IAAG,MAAM,WAAW,SAAS;AAAA;AAAA;AAG3D,mBAAW,WAAW;AACtB,mBAAW,MAAM;AAAA;AAEnB,aAAO;AAAA;AAAA;AAIX,sBAAoB;AAClB,UAAM,UAAS,MAAM,QAAQ,IAAI;AAAA,MAC/B,UAAU,KAAK;AAAA,MACf,IAAG,eAAe,QAAO,KAAK,WAAW,CAAE,WAAW,QAAO,KAAK,UAAU,SAAS;AAAA,MACrF,IAAG,eAAe,QAAO,KAAK,WAAW,CAAE,WAAW,QAAO,KAAK,UAAU,SAAS;AAAA;AAEvF,UAAM,WAAW,IAAI,kBAAkB,QAAO,IAAI,QAAO,IAAI,QAAO,IAAI;AACxE,WAAO;AAAA;AAGT,WAAQ,OAAO;AACf,WAAQ,oBAAoB;AAC5B,WAAQ,YAAY;AACpB,WAAQ,gBAAgB;AAAA;;;AC9DxB;AAAA,QAAM,MAAK;AAEX,QAAM,UAAS;AACf,MAAI,OAAO,CAAE,KAAK,GAAG,QAAQ;AAC7B,MAAI,QAAQ;AAEZ,0BAAwB,OAAO;AAC7B,UAAM,SAAS,IAAG,QAAQ,WAAW;AACrC,UAAM,SAAS,IAAG,MAAM,eAAe,QAAQ,CAAC,MAAM;AACtD,UAAM,SAAS,IAAG,KAAK,IAAG,WAAW,QAAQ,IAAI;AACjD,WAAO;AAAA;AAGT,yBAAuB;AACrB,QAAI,CAAC,QAAO;AAAK,cAAO,MAAM,MAAM,IAAG,eAAe,QAAO,KAAK,IAAI;AACtE,WAAO,QAAO;AAAA;AAGhB,4BAA0B;AACxB,QAAI,CAAC,QAAO;AAAQ,cAAO,SAAS,MAAM,IAAG,eAAe,QAAO,KAAK,OAAO;AAC/E,WAAO,QAAO;AAAA;AAGhB,yBAAuB,OAAO;AAC5B,QAAI,QAAQ,QAAO,KAAK,IAAI;AAC1B,cAAQ;AAAA;AAER,eAAS;AAAA;AAEX,QAAI,UAAU;AAAG,aAAO;AACxB,QAAI;AACJ,QAAI,iBAAiB,IAAG;AACtB,YAAM,SAAS,IAAG,MAAM,eAAe,OAAO,CAAC,QAAO,KAAK,IAAI,WAAW,QAAO,KAAK,IAAI,YAAY;AACtG,gBAAU,IAAG,IAAI,QAAQ,CAAC;AAC1B,UAAG,QAAQ;AAAA;AAEX,gBAAU,MAAM,SAAS,OAAO,QAAO,KAAK,IAAI;AAAA;AAElD,UAAM,MAAM;AACZ,QAAI,QAAO,KAAK,IAAI;AAClB,YAAM,OAAO,MAAM,QAAO,IAAI,QAAQ;AACtC,YAAM,OAAO,MAAM,KAAK;AACxB,UAAI,MAAM,KAAK,MAAM,KAAK,KAAK,MAAM;AACrC,UAAG,QAAQ;AAAA;AAEb,QAAI,QAAO,KAAK,OAAO;AACrB,YAAM,UAAU,MAAM,QAAO,OAAO,QAAQ;AAC5C,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,aAAa,KAAK,MAAM,KAAK,IAAI,MAAM,MAAO,MAAK,KAAK,SAAS;AACvE,UAAI,aAAa,QAAO,KAAK,OAAO;AAClC,YAAI,SAAS,KAAK,MAAM,MAAM,WAAW;AACzC,YAAI,aAAa;AAAA;AAEnB,UAAG,QAAQ;AAAA;AAEb,QAAG,QAAQ;AACX,WAAO;AACP,WAAO;AAAA;AAGT,WAAQ,UAAU;AAClB,WAAQ,UAAU;AAClB,WAAQ,aAAa;AAAA;;;AC9DrB;AAAA,QAAM,MAAK;AAEX,QAAM,cAAc,CAAC,SAAS,WAAW,QAAQ,SAAS,OAAO,WAAW;AAC5E,QAAM,UAAS;AACf,MAAI,OAAO;AACX,MAAI,QAAQ;AACZ,QAAM,aAAa;AAEnB,oBAAkB,OAAO;AACvB,UAAM,SAAS,IAAG,KAAK;AACrB,YAAM,SAAS,IAAG,QAAQ,WAAW,OAAO;AAC5C,YAAM,SAAS,IAAG,MAAM,eAAe,QAAQ,CAAC,MAAM;AACtD,YAAM,SAAS,IAAG,KAAK,IAAG,WAAW,QAAQ,IAAI;AACjD,aAAO;AAAA;AAET,WAAO;AAAA;AAGT,sBAAoB;AAClB,QAAI,CAAC,QAAO;AAAS,cAAO,UAAU,MAAM,IAAG,eAAe,QAAO,KAAK,QAAQ;AAClF,WAAO,QAAO;AAAA;AAGhB,yBAAuB,OAAO;AAC5B,aAAS;AACT,QAAI,SAAS,QAAO,KAAK,QAAQ;AAC/B,cAAQ;AACR,aAAO;AAAA;AAET,UAAM,UAAU,IAAG,KAAK;AACtB,UAAI,iBAAiB,IAAG;AACtB,cAAM,SAAS,IAAG,MAAM,eAAe,OAAO,CAAC,QAAO,KAAK,QAAQ,WAAW,QAAO,KAAK,QAAQ,YAAY;AAC9G,cAAM,CAAC,GAAG,GAAG,KAAK,IAAG,MAAM,QAAQ,GAAG;AACtC,YAAI,QAAO,KAAK,QAAQ;AAEtB,gBAAM,KAAK,IAAG,IAAI,GAAG,CAAC;AACtB,gBAAM,KAAK,IAAG,IAAI,GAAG,CAAC;AACtB,gBAAM,KAAK,IAAG,IAAI,GAAG,CAAC;AACtB,gBAAM,YAAY,IAAG,KAAK,CAAC,IAAI,IAAI;AACnC,iBAAO;AAAA;AAET,eAAO;AAAA;AAET,aAAO,SAAS,OAAO,QAAO,KAAK,QAAQ;AAAA;AAE7C,UAAM,MAAM;AACZ,QAAI,QAAO,KAAK,QAAQ;AACtB,YAAM,WAAW,MAAM,QAAO,QAAQ,QAAQ;AAC9C,YAAM,OAAO,MAAM,SAAS;AAC5B,eAAS,IAAI,GAAG,IAAI,KAAK,QAAQ;AAC/B,YAAI,aAAa,KAAK,KAAK,QAAO,KAAK,QAAQ;AAAe,cAAI,KAAK,CAAE,OAAO,KAAK,IAAI,MAAM,KAAK,MAAM,MAAM,aAAa,KAAK,MAAM,MAAM,SAAS,YAAY;AAAA;AAErK,UAAI,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE;AAC/B,UAAG,QAAQ;AAAA;AAEb,QAAG,QAAQ;AACX,WAAO;AACP,WAAO;AAAA;AAGT,WAAQ,UAAU;AAClB,WAAQ,OAAO;AAAA;;;AC7Df;AAAA,QAAM,MAAK;AAEX;AAAA,IACE,YAAY,OAAO;AACjB,WAAK,QAAQ;AACb,WAAK,eAAe;AACpB,YAAM,aAAa,KAAK,MAAM,OAAO,GAAG;AACxC,UAAG,KAAK,OAAQ,WAAW,OAAO,MAAQ,WAAW,OAAO,IAAK,MAAM,gBAAgB,WAAW,OAAO,WAAW;AAAA;AAAA,IAgBtH,QAAQ;AACN,aAAO,IAAG,KAAK;AACb,cAAM,UAAU,KAAK,gBAAgB,MAAM;AAC3C,cAAM,UAAU,QAAQ,WAAW;AACnC,cAAM,UAAU,KAAK,MAAM,QAAQ;AACnC,cAAM,YAAY,QAAQ,IAAI,CAAC,MAAM,EAAE,QAAQ,CAAC;AAChD,cAAM,eAAe,KAAK,kBAAkB;AAC5C,eAAO;AAAA,UACL,eAAe,aAAa,QAAQ;AAAA,UACpC,SAAS,aAAa;AAAA,UACtB,iBAAiB,aAAa;AAAA,UAC9B,iBAAiB,aAAa;AAAA;AAAA;AAAA;AAAA,IAQpC;AACE,WAAK,MAAM;AAAA;AAAA;AAGf,WAAQ,YAAY;AAAA;;;AC9CpB;AAAA,QAAM,MAAK;AACX,QAAM,YAAY;AAElB,0BAAwB,UAAU;AAAA,IAEhC,gBAAgB;AAEd,aAAO,IAAG,KAAK,MAAM,IAAG,IAAI,OAAO,OAAO,IAAI;AAAA;AAAA,IAIhD,kBAAkB;AAChB,YAAM,CAAC,SAAS,SAAS,iBAAiB,mBAAmB;AAC7D,aAAO,CAAE,SAAS,SAAS,iBAAiB;AAAA;AAAA;AAGhD,WAAQ,YAAY;AAAA;;;AChBpB;AACA,gBAAc;AACZ,WAAO,KAAK,MAAM,IAAI;AAAA;AAExB;AAAA,IACE,YAAY,SAAS;AACnB,WAAK,gBAAgB,IAAI,MAAM;AAC/B,WAAK,mBAAmB;AACxB,WAAK,kBAAkB;AAAA;AAAA,IAGzB,QAAQ;AACN,WAAK,cAAc,EAAE,KAAK,oBAAoB;AAC9C,WAAK,KAAK,KAAK;AAAA;AAAA,IAGjB;AACE,YAAM,MAAM,KAAK,cAAc;AAC/B,WAAK,SAAS,GAAG,KAAK;AACtB,WAAK,KAAK;AACV,WAAK,cAAc,KAAK,mBAAmB,KAAK;AAChD,aAAO;AAAA;AAAA,IAGT;AACE,aAAO,KAAK,qBAAqB;AAAA;AAAA,IAGnC;AACE,aAAO,KAAK,mBAAmB;AAAA;AAAA,IAGjC;AACE,aAAO,KAAK,cAAc,MAAM,GAAG,KAAK,mBAAmB;AAAA;AAAA,IAG7D;AACE,aAAO,KAAK,cAAc;AAAA;AAAA,IAG5B,KAAK;AACH,aAAO,IAAI,KAAK,KAAK,KAAK,KAAK,IAAI;AACjC,aAAK,SAAS,GAAG,KAAK;AACtB,YAAI,KAAK;AAAA;AAAA;AAAA,IAIb,KAAK;AACH,aAAO,IAAI,KAAK,KAAK;AACnB,YAAI,IAAI,IAAI;AACZ,YAAI,IAAI,KAAK,oBAAoB,KAAK,KAAK,GAAG,IAAI;AAAI;AACtD,YAAI,CAAC,KAAK,KAAK,GAAG;AAAI;AACtB,aAAK,SAAS,GAAG;AACjB,YAAI;AAAA;AAAA;AAAA,IAIR,WAAW;AACT,aAAO,KAAK,gBAAgB,KAAK,cAAc;AAAA;AAAA,IAGjD,KAAK,GAAG;AACN,aAAO,KAAK,WAAW,KAAK,KAAK,WAAW;AAAA;AAAA,IAG9C,SAAS,GAAG;AACV,YAAM,IAAI,KAAK,cAAc;AAC7B,WAAK,cAAc,KAAK,KAAK,cAAc;AAC3C,WAAK,cAAc,KAAK;AAAA;AAAA;AAG5B,WAAQ,UAAU;AAAA;;;ACvElB;AAAA,QAAM,WAAW;AAEjB,uCAAqC,YAAY,OAAO,UAAU,UAAU,oBAAoB;AAC9F,UAAM,CAAC,QAAQ,SAAS,OAAO;AAC/B,QAAI,eAAe;AACnB,UAAM,SAAS,KAAK,IAAI,WAAW,oBAAoB;AACvD,UAAM,OAAO,KAAK,IAAI,WAAW,qBAAqB,GAAG;AACzD,aAAS,WAAW,QAAQ,WAAW,MAAM,EAAE;AAC7C,YAAM,SAAS,KAAK,IAAI,WAAW,oBAAoB;AACvD,YAAM,OAAO,KAAK,IAAI,WAAW,qBAAqB,GAAG;AACzD,eAAS,WAAW,QAAQ,WAAW,MAAM,EAAE;AAC7C,YAAI,OAAO,IAAI,UAAU,UAAU,cAAc;AAC/C,yBAAe;AACf;AAAA;AAAA;AAGJ,UAAI,CAAC;AACH;AAAA;AAAA;AAGJ,WAAO;AAAA;AAOT,mCAAiC,gBAAgB,oBAAoB;AACnE,UAAM,CAAC,QAAQ,OAAO,gBAAgB,OAAO;AAC7C,UAAM,QAAQ,IAAI,SAAS,QAAQ,SAAS,QAAQ,cAAc,CAAC,CAAE,WAAY;AACjF,aAAS,WAAW,GAAG,WAAW,QAAQ,EAAE;AAC1C,eAAS,WAAW,GAAG,WAAW,OAAO,EAAE;AACzC,iBAAS,aAAa,GAAG,aAAa,cAAc,EAAE;AACpD,gBAAM,QAAQ,OAAO,IAAI,UAAU,UAAU;AAE7C,cAAI,QAAQ;AAAgB;AAE5B,cAAI,4BAA4B,YAAY,OAAO,UAAU,UAAU,oBAAoB;AACzF,kBAAM,QAAQ,CAAE,OAAO,MAAM,CAAE,UAAU,UAAU,IAAI;AAAA;AAAA;AAAA;AAAA;AAK/D,WAAO;AAAA;AAET,WAAQ,0BAA0B;AAAA;;;AC7ClC;AAAA,WAAQ,YAAY;AAAA,IAClB;AAAA,IAAQ;AAAA,IAAW;AAAA,IAAY;AAAA,IAAW;AAAA,IAAY;AAAA,IACtD;AAAA,IAAiB;AAAA,IAAa;AAAA,IAAc;AAAA,IAAa;AAAA,IACzD;AAAA,IAAW;AAAA,IAAY;AAAA,IAAY;AAAA,IAAa;AAAA,IAAa;AAAA;AAE/D,WAAQ,gBAAgB,SAAQ,UAAU;AAC1C,WAAQ,UAAU,SAAQ,UAAU,OAAO,CAAC,QAAQ,WAAW;AAC7D,WAAO,aAAa;AACpB,WAAO;AAAA,KACN;AACH,QAAM,qBAAqB;AAAA,IACzB,CAAC,WAAW;AAAA,IAAiB,CAAC,aAAa;AAAA,IAC3C,CAAC,aAAa;AAAA,IAAc,CAAC,WAAW;AAAA,IACxC,CAAC,YAAY;AAAA,IAAc,CAAC,YAAY;AAAA,IACxC,CAAC,cAAc;AAAA,IAAkB,CAAC,cAAc;AAAA,IAChD,CAAC,YAAY;AAAA,IAAc,CAAC,aAAa;AAAA,IACzC,CAAC,gBAAgB;AAAA,IAAkB,CAAC,WAAW;AAAA;AAQjD,WAAQ,YAAY;AAAA,IAClB,CAAC,QAAQ;AAAA,IAAY,CAAC,WAAW;AAAA,IAAY,CAAC,QAAQ;AAAA,IACtD,CAAC,YAAY;AAAA,IAAa,CAAC,QAAQ;AAAA,IACnC,CAAC,gBAAgB;AAAA,IAAc,CAAC,aAAa;AAAA,IAC7C,CAAC,gBAAgB;AAAA,IAAY,CAAC,WAAW;AAAA,IACzC,CAAC,YAAY;AAAA,IAAc,CAAC,QAAQ;AAAA,IACpC,CAAC,iBAAiB;AAAA,IAAe,CAAC,cAAc;AAAA,IAChD,CAAC,iBAAiB;AAAA,IAAa,CAAC,YAAY;AAAA,IAC5C,CAAC,aAAa;AAAA;AAEhB,WAAQ,uBAAuB,mBAAmB,IAAI,CAAC,CAAC,YAAY,gBAAiB,CAAC,SAAQ,QAAQ,aAAa,SAAQ,QAAQ;AACnI,WAAQ,eAAe;AAAA,IACrB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA;AAAA;;;AC3DF;AAAA,QAAM,MAAM;AAEZ,0BAAwB,GAAG,GAAG,UAAU;AACtC,WAAO;AAAA,MACL,GAAG,QAAQ,IAAI,GAAG,GAAG;AAAA,MACrB,GAAG,QAAQ,IAAI,GAAG,GAAG,WAAW,IAAI;AAAA;AAAA;AAGxC,WAAQ,iBAAiB;AAEzB,0BAAwB,MAAM,cAAc;AAC1C,UAAM,CAAE,UAAU,UAAU,IAAI,YAAa;AAC7C,UAAM,CAAE,GAAG,KAAM,eAAe,UAAU,UAAU,UAAU;AAC9D,WAAO;AAAA,MACL,GAAG,KAAK,WAAW,eAAe;AAAA,MAClC,GAAG,KAAK,WAAW,eAAe;AAAA;AAAA;AAGtC,WAAQ,iBAAiB;AAEzB,qBAAmB,SAAS;AAC1B,UAAM,SAAS,IAAI,MAAM;AACzB,aAAS,IAAI,GAAG,IAAI,MAAM;AACxB,aAAO,KAAK;AAAA;AAEd,WAAO;AAAA;AAET,WAAQ,YAAY;AAEpB,iBAAe,GAAG,KAAK;AACrB,QAAI,IAAI;AAAK,aAAO;AACpB,QAAI,IAAI;AAAK,aAAO;AACpB,WAAO;AAAA;AAET,WAAQ,QAAQ;AAEhB,2BAAyB,IAAI,IAAI,IAAI;AACnC,UAAM,KAAK,KAAK;AAChB,UAAM,KAAK,KAAK;AAChB,WAAO,KAAK,KAAK,KAAK;AAAA;AAExB,WAAQ,kBAAkB;AAE1B,sBAAoB,GAAG;AACrB,WAAO,CAAE,GAAG,EAAE,IAAI,EAAE,GAAG,GAAG,EAAE,IAAI,EAAE;AAAA;AAEpC,WAAQ,aAAa;AAErB,uBAAqB,GAAG,KAAK;AAC3B,WAAO,CAAE,GAAG,MAAM,EAAE,GAAG,KAAK,MAAM,GAAG,MAAM,EAAE,GAAG,KAAK;AAAA;AAEvD,WAAQ,cAAc;AAAA;;;ACnDtB;AAAA,QAAM,YAAY;AAClB,QAAM,UAAU;AAEhB,QAAM,uBAAuB,UAAU,UAAU,IAAI,CAAC,CAAC,gBAAgB,mBAAoB,CAAC,UAAU,QAAQ,iBAAiB,UAAU,QAAQ;AACjJ,QAAM,qBAAqB,qBAAqB,IAAI,CAAC,CAAC,EAAE,kBAAkB;AAC1E,QAAM,qBAAqB,qBAAqB,IAAI,CAAC,CAAC,mBAAmB;AACzE,2BAAyB,QAAQ,OAAO;AACtC,UAAM,WAAW,cAAc,MAAM,KAAK;AAC1C,WAAO;AAAA,MACL,GAAG,cAAc,IAAI,MAAM,GAAG,MAAM,GAAG;AAAA,MACvC,GAAG,cAAc,IAAI,MAAM,GAAG,MAAM,GAAG,WAAW;AAAA;AAAA;AAGtD,oCAAkC,OAAO,cAAc,QAAQ;AAC7D,WAAO;AAAA,MACL,GAAG,QAAQ,MAAM,KAAK,MAAM,MAAM,IAAI,eAAe,GAAG,SAAS;AAAA,MACjE,GAAG,QAAQ,MAAM,KAAK,MAAM,MAAM,IAAI,eAAe,GAAG,QAAQ;AAAA;AAAA;AAUpE,oCAAkC,QAAQ,gBAAgB,kBAAkB,cAAc,SAAS,cAAc,eAAe,mBAAmB;AACjJ,UAAM,CAAC,QAAQ,SAAS,aAAa;AAErC,UAAM,wBAAwB,yBAAyB,eAAe,UAAU,cAAc,QAAQ;AACtG,UAAM,eAAe,gBAAgB,QAAQ,uBAAuB;AACpE,UAAM,iBAAiB,QAAQ,WAAW,eAAe,UAAU;AACnE,QAAI,iBAAiB;AACrB,aAAS,IAAI,GAAG,IAAI,kBAAkB;AACpC,YAAM,wBAAwB,yBAAyB,gBAAgB,cAAc,QAAQ;AAC7F,YAAM,cAAc,QAAQ,eAAe,sBAAsB,GAAG,sBAAsB,GAAG,kBAAkB;AAC/G,uBAAiB,QAAQ,WAAW;AAAA,QAClC,GAAG,sBAAsB,IAAI;AAAA,QAC7B,GAAG,sBAAsB,IAAI;AAAA,SAC5B,CAAE,GAAG,YAAY,GAAG,GAAG,YAAY;AAAA;AAExC,UAAM,wBAAwB,yBAAyB,gBAAgB,cAAc,QAAQ;AAC7F,UAAM,QAAQ,aAAa,IAAI,sBAAsB,GAAG,sBAAsB,GAAG;AACjF,WAAO,CAAE,UAAU,gBAAgB,MAAM,UAAU,UAAU,mBAAmB;AAAA;AAQlF,sBAAoB,MAAM,QAAQ,SAAS,cAAc,kBAAkB;AACzE,UAAM,WAAW,OAAO,MAAM;AAC9B,UAAM,WAAW,mBAAmB;AACpC,UAAM,oBAAoB,IAAI,MAAM;AAEpC,UAAM,CAAE,MAAM,UAAU,OAAO,aAAc;AAC7C,UAAM,YAAY,QAAQ,eAAe,UAAU,cAAc;AACjE,sBAAkB,SAAS,MAAM;AAAA,MAC/B,OAAO;AAAA,MACP,MAAM,UAAU,UAAU,SAAS;AAAA,MACnC,UAAU;AAAA;AAIZ,aAAS,OAAO,WAAW,GAAG,QAAQ,GAAG,EAAE;AACzC,YAAM,mBAAmB,mBAAmB;AAC5C,YAAM,mBAAmB,mBAAmB;AAC5C,UAAI,kBAAkB,qBAAqB,CAAC,kBAAkB;AAC5D,0BAAkB,oBAAoB,yBAAyB,MAAM,kBAAkB,mBAAmB,kBAAkB,QAAQ,SAAS,cAAc;AAAA;AAAA;AAK/J,aAAS,OAAO,GAAG,OAAO,UAAU,EAAE;AACpC,YAAM,mBAAmB,mBAAmB;AAC5C,YAAM,mBAAmB,mBAAmB;AAC5C,UAAI,kBAAkB,qBAAqB,CAAC,kBAAkB;AAC5D,0BAAkB,oBAAoB,yBAAyB,MAAM,kBAAkB,mBAAmB,kBAAkB,QAAQ,SAAS,cAAc;AAAA;AAAA;AAG/J,WAAO;AAAA;AAET,WAAQ,aAAa;AAAA;;;ACnFrB;AAAA,QAAM,aAAa;AACnB,QAAM,aAAa;AACnB,QAAM,UAAU;AAEhB,+CAA6C,OAAO,kBAAkB,CAAE,GAAG,IAAK;AAC9E,WAAO,MAAM,KAAK,CAAC,CAAE;AACnB,YAAM,wBAAwB,UAAU,YAAY;AACpD,aAAO,QAAQ,gBAAgB,GAAG,GAAG,sBAAsB,GAAG,sBAAsB,MAAM;AAAA;AAAA;AAO9F,4BAA0B,eAAe,kBAAkB;AACzD,UAAM,8BAA8B,kBAAkB,OAAO,CAAC,QAAQ,CAAE,UAAU,QAAS;AACzF,UAAI,CAAC,oCAAoC,eAAe,kBAAkB,UAAU;AAClF,kBAAU;AAAA;AAEZ,aAAO;AAAA,OACN;AACH,WAAO,8BAA8B,kBAAkB;AAAA;AAKzD,QAAM,sBAAsB;AAwD5B,+BAA6B,cAAc,eAAe,wBAAwB,wBAAwB,cAAc,mBAAmB,iBAAiB,KAAK,YAAY;AAC3K,UAAM,QAAQ;AACd,UAAM,QAAQ,WAAW,wBAAwB,gBAAgB,qBAAqB;AACtF,UAAM,mBAAmB,YAAY;AAGrC,WAAO,MAAM,SAAS,qBAAqB,CAAC,MAAM;AAEhD,YAAM,OAAO,MAAM;AAInB,YAAM,kBAAkB,QAAQ,eAAe,KAAK,MAAM,cAAc;AACxE,UAAI,oCAAoC,OAAO,kBAAkB,iBAAiB,KAAK,KAAK;AAAK;AAEjG,YAAM,YAAY,WAAW,WAAW,MAAM,cAAc,eAAe,cAAc,wBAAwB;AACjH,YAAM,QAAQ,iBAAiB,OAAO,kBAAkB;AACxD,YAAM,KAAK,CAAE,WAAW;AAAA;AAE1B,WAAO;AAAA;AAET,WAAQ,sBAAsB;AAAA;;;ACvG9B;AAAA,QAAM,MAAK;AACX,QAAM,MAAM;AAEZ,2CAAyC,GAAG,GAAG;AAC7C,WAAQ,IAAI,iBAAiB,IAAI;AAAA;AAGnC,gCAA8B,WAAW;AACvC,WAAO,IAAI,qBAAqB,OAAO,CAAC,QAAQ,CAAC,WAAW;AAC1D,UAAI,gCAAgC,UAAU,WAAW,OAAO,UAAU,YAAY,OAAO;AAC3F,eAAO;AAAA;AAET,aAAO,KAAK,CAAC,UAAU,YAAY,UAAU;AAC7C,aAAO;AAAA,OACN;AAAA;AAEL,WAAQ,uBAAuB;AAE/B,QAAM,CAAE,mBAAmB,qBAAsB;AACjD,0BAAwB;AACtB,WAAO,UAAU,OAAO,CAAC,CAAE,MAAM,MAAM,MAAM,OAAQ,CAAE,UAAU,CAAE,GAAG,QAAW;AAAA,MAC/E,MAAM,KAAK,IAAI,MAAM;AAAA,MACrB,MAAM,KAAK,IAAI,MAAM;AAAA,MACrB,MAAM,KAAK,IAAI,MAAM;AAAA,MACrB,MAAM,KAAK,IAAI,MAAM;AAAA,QACnB;AAAA,MACF,MAAM;AAAA,MACN,MAAM;AAAA,MACN,MAAM;AAAA,MACN,MAAM;AAAA;AAAA;AAGV,WAAQ,iBAAiB;AACzB,gCAA8B;AAC5B,UAAM,CAAE,MAAM,MAAM,MAAM,QAAS,eAAe;AAClD,WAAO,CAAC,CAAE,GAAG,MAAM,GAAG,OAAQ,CAAE,GAAG,MAAM,GAAG,OAAQ,CAAE,GAAG,MAAM,GAAG,OAAQ,CAAE,GAAG,MAAM,GAAG;AAAA;AAE1F,WAAQ,uBAAuB;AAC/B,mCAAiC;AAC/B,WAAO,QAAQ,IAAI,QAAQ,IAAI,CAAC,WAAW,OAAO;AAAA;AAEpD,WAAQ,oBAAoB;AAE5B,qBAAmB,MAAM,QAAQ,QAAQ,UAAU,GAAG,UAAU;AAC9D,WAAO;AAAA,MACL,OAAO,KAAK;AAAA,MACZ,WAAW,KAAK,UAAU,IAAI,CAAC,CAAE,OAAO,MAAM,cAAgB;AAAA,QAC5D;AAAA,QACA;AAAA,QACA,UAAU;AAAA,UACR,GAAG,SAAS,IAAI,SAAS;AAAA,UACzB,GAAG,SAAS,IAAI,SAAS;AAAA;AAAA;AAAA;AAAA;AAKjC,WAAQ,YAAY;AAEpB,sBAAoB,OAAO,QAAQ,QAAQ,UAAU,GAAG,UAAU;AAChE,QAAI,WAAW,KAAK,WAAW,KAAK,YAAY,KAAK,YAAY;AAC/D,aAAO;AAAA;AAET,WAAO,MAAM,IAAI,CAAC,SAAS,UAAU,MAAM,QAAQ,QAAQ,SAAS;AAAA;AAEtE,WAAQ,aAAa;AAErB,oCAAkC;AAChC,WAAO,iBAAiB,IAAG,SAAS,CAAC,MAAM,MAAM,IAAI,MAAM,MAAM,MAAM,CAAC,MAAM,QAAQ,MAAM;AAAA;AAE9F,WAAQ,2BAA2B;AAEnC,yBAAuB;AACrB,WAAO,iBAAiB,IAAG,SAAS,QAAQ,IAAG,QAAQ,WAAW;AAAA;AAEpE,WAAQ,gBAAgB;AAExB,gCAA8B,OAAO,cAAc;AACjD,WAAO,IAAG,KAAK;AACb,YAAM,cAAc,cAAc;AAClC,aAAO,YAAY,eAAe,CAAC,cAAc;AAAA;AAAA;AAGrD,WAAQ,uBAAuB;AAE/B,0BAAwB,OAAO,CAAC,SAAS;AACvC,UAAM,CAAC,QAAQ,SAAS,yBAAyB;AACjD,UAAM,eAAe,UAAU;AAC/B,UAAM,SAAS,QAAQ;AACvB,QAAI,CAAC,MAAM,MAAM,MAAM,QAAQ,CAAC,GAAG,GAAG,GAAG;AACzC,QAAI,SAAS;AAEX,aAAO;AACP,aAAO;AACP,aAAO,KAAK,MAAM,MAAO,gBAAe,SAAS;AACjD,aAAO,KAAK,MAAM,MAAO,gBAAe,SAAS;AAAA;AAGjD,aAAO,KAAK,MAAM,MAAQ,KAAM,eAAgB,QAAQ;AACxD,aAAO,KAAK,MAAM,MAAQ,KAAM,eAAgB,QAAQ;AACxD,aAAO;AACP,aAAO;AAAA;AAET,UAAM,UAAU,IAAG,KAAK;AACtB,UAAI,cAAc,cAAc;AAChC,oBAAc,IAAG,MAAM,aAAa,CAAC,CAAC,MAAM,OAAO,CAAC,MAAM,OAAO,CAAC,GAAG;AACrE,aAAO,YAAY,eAAe,CAAC,SAAS;AAAA;AAE9C,WAAO,CAAE,SAAS,SAAS,CAAE,KAAK,MAAM,MAAM,MAAM,OAAO,MAAM,QAAQ;AAAA;AAE3E,WAAQ,iBAAiB;AAEzB,6BAA2B,OAAO,CAAC,QAAQ,QAAQ,CAAC,uBAAuB,uBAAuB;AAChG,UAAM,SAAU,UAAS,QAAQ,MAAM,QAAQ,UAAW;AAC1D,UAAM,SAAU,SAAQ,QAAQ,OAAO,QAAQ,SAAU;AACzD,UAAM,cAAc,WAAW,OAAO,QAAQ,QAAQ,CAAC,QAAQ,KAAK,CAAC,QAAQ;AAC7E,WAAO;AAAA;AAET,WAAQ,oBAAoB;AAAA;;;ACrH5B;AAAA,QAAM,MAAK;AACX,QAAM,iBAAiB;AACvB,QAAM,iBAAiB;AACvB,QAAM,OAAO;AAEb;AAAA,IACE,YAAY;AACV,WAAK,YAAY;AAAA;AAAA,UAuBb,cAAc,OAAO;AACzB,YAAM,eAAe,QAAO;AAE5B,YAAM,CAAC,QAAQ,SAAS,KAAK,yBAAyB;AACtD,YAAM,CAAE,SAAS,WAAY,KAAK,eAAe,OAAO,CAAC,QAAO,iBAAiB,QAAO;AACxF,YAAM,CAAE,eAAe,SAAS,iBAAiB,mBAAoB,KAAK,UAAU,QAAQ;AAC5F,YAAM,mBAAmB,MAAM,KAAK,kBAAkB,CAAC,eAAe,SAAS,iBAAiB;AAChG,YAAM,eAAe,iBAAiB;AACtC,YAAM,gBAAgB,iBAAiB;AACvC,YAAM,yBAAyB,iBAAiB;AAChD,YAAM,yBAAyB,iBAAiB;AAChD,YAAM,QAAQ,MAAM,eAAe,oBAAoB,cAAc,eAAe,wBAAwB,wBAAwB,cAAc,QAAO,eAAe,QAAO,gBAAgB,QAAO;AACtM,YAAM,cAAc,KAAK,kBAAkB,OAAO,CAAC,QAAQ,QAAQ,CAAC,QAAO,iBAAiB,QAAO,kBAAkB;AACrH,oBAAc;AACd,cAAQ;AACR,sBAAgB;AAChB,sBAAgB;AAChB,cAAQ;AACR,aAAO;AAAA;AAAA,IAGT;AACE,WAAK,UAAU;AAAA;AAAA;AAGnB,WAAQ,UAAU;AAClB,+BAA6B;AAC3B,UAAM,aAAa,MAAM,IAAG,eAAe,QAAO;AAClD,UAAM,YAAY,IAAI,eAAe,UAAU,YAAY,QAAO;AAClE,WAAO,IAAI,QAAQ;AAAA;AAYrB,sBAAoB;AAClB,WAAO,cAAc;AAAA;AAEvB,WAAQ,OAAO;AAAA;;;AC1Ef;AAAA,QAAM,iBAAiB;AACvB,QAAM,eAAe;AACrB,QAAM,iBAAiB;AACvB,QAAM,YAAY;AAClB,QAAM,OAAO;AAEb,WAAQ,OAAO,aAAa;AAC5B,WAAQ,UAAU,aAAa;AAE/B,WAAQ,YAAY,eAAe;AACnC,WAAQ,sBAAsB,eAAe;AAC7C,WAAQ,eAAe,UAAU;AACjC,WAAQ,UAAU,UAAU;AAC5B,WAAQ,YAAY,UAAU;AAC9B,WAAQ,YAAY,UAAU;AAC9B,WAAQ,uBAAuB,KAAK;AACpC,WAAQ,iBAAiB,KAAK;AAC9B,WAAQ,uBAAuB,KAAK;AACpC,WAAQ,oBAAoB,KAAK;AACjC,WAAQ,YAAY,KAAK;AAAA;;;ACnBzB;AAAA,QAAM,MAAK;AAEX,sBAAoB;AAClB,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA,MAC1C,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA;AAAA;AAG9C,WAAQ,aAAa;AAErB,wBAAsB;AACpB,WAAO;AAAA,MACL,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA,MAC5D,IAAI,WAAW,KAAM,KAAI,SAAS,KAAK,IAAI,WAAW,MAAM;AAAA;AAAA;AAGhE,WAAQ,eAAe;AAEvB,oCAAkC,KAAK,OAAO;AAC5C,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,IAAI,MAAM,MAAM;AACtB,UAAM,QAAQ,CAAC;AAAA,MACb,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,WAAW,KAAK;AAAA,MAAG,IAAI,SAAS,KAAK;AAAA,MAChE,IAAI,SAAS,KAAK;AAAA;AAEpB,WAAO,IAAG,MAAM,cAAc,OAAO,OAAO,CAAC,IAAI;AAAA;AAEnD,WAAQ,2BAA2B;AAEnC,+BAA6B,KAAK;AAChC,UAAM,aAAa,CAAC,IAAI,WAAW,KAAK,OAAO,IAAI,IAAI,WAAW,KAAK,OAAO;AAC9E,UAAM,WAAW,CAAC,IAAI,SAAS,KAAK,OAAO,IAAI,IAAI,SAAS,KAAK,OAAO;AACxE,UAAM,gBAAgB,IAAI,cAAc,IAAI,CAAC;AAC3C,YAAM,cAAc,CAAC,MAAM,KAAK,OAAO,IAAI,MAAM,KAAK,OAAO;AAC7D,aAAO;AAAA;AAET,WAAO,CAAE,YAAY,UAAU;AAAA;AAEjC,WAAQ,sBAAsB;AAE9B,sBAAoB,KAAK,SAAS;AAChC,UAAM,SAAS,aAAa;AAC5B,UAAM,OAAO,WAAW;AACxB,UAAM,cAAc,CAAC,SAAS,KAAK,KAAK,GAAG,SAAS,KAAK,KAAK;AAC9D,UAAM,aAAa,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACxE,UAAM,WAAW,CAAC,OAAO,KAAK,YAAY,IAAI,OAAO,KAAK,YAAY;AACtE,WAAO,CAAE,YAAY,UAAU,eAAe,IAAI;AAAA;AAEpD,WAAQ,aAAa;AAErB,uBAAqB;AACnB,UAAM,UAAU,aAAa;AAC7B,UAAM,OAAO,WAAW;AACxB,UAAM,UAAU,KAAK,IAAI,GAAG;AAC5B,UAAM,WAAW,UAAU;AAC3B,UAAM,aAAa,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACxD,UAAM,WAAW,CAAC,QAAQ,KAAK,UAAU,QAAQ,KAAK;AACtD,WAAO,CAAE,YAAY,UAAU,eAAe,IAAI;AAAA;AAEpD,WAAQ,cAAc;AAEtB,oBAAkB,KAAK;AACrB,UAAM,UAAU;AAAA,MACd,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA,MAAI,IAAI,SAAS,KAAK,IAAI,WAAW;AAAA;AAExE,UAAM,cAAc,CAAC,QAAQ,KAAK,YAAY,IAAI,QAAQ,KAAK,YAAY;AAC3E,UAAM,aAAa,CAAC,IAAI,WAAW,KAAK,YAAY,IAAI,IAAI,WAAW,KAAK,YAAY;AACxF,UAAM,WAAW,CAAC,IAAI,SAAS,KAAK,YAAY,IAAI,IAAI,SAAS,KAAK,YAAY;AAClF,WAAO,CAAE,YAAY,UAAU,eAAe,IAAI;AAAA;AAEpD,WAAQ,WAAW;AAAA;;;ACtEnB;AAAA,QAAM,MAAK;AACX,QAAM,WAAW;AAEjB;AAAA,IACE,YAAY,OAAO,SAAS;AAC1B,WAAK,QAAQ;AACb,WAAK,QAAQ,QAAO;AACpB,WAAK,SAAS,QAAO;AACrB,WAAK,UAAU,QAAQ,IAAI,CAAC,WAAW,CAAC,OAAO,UAAU,OAAO;AAChE,WAAK,gBAAgB,IAAG,SAAS,KAAK;AACtC,WAAK,kBAAkB,IAAG,SAAS,CAAC,QAAO,WAAW,QAAO;AAC7D,WAAK,wBAAwB,IAAG,SAAS,CAAC,QAAO,YAAY,GAAG,QAAO,YAAY;AAAA;AAAA,IAGrF,eAAe;AACb,aAAO,IAAG,KAAK;AACb,cAAM,aAAa,IAAG,MAAM,OAAO,CAAC,GAAG,IAAI,CAAC,IAAI;AAChD,cAAM,WAAW,IAAG,MAAM,OAAO,CAAC,GAAG,IAAI,CAAC,IAAI;AAC9C,cAAM,kBAAkB,IAAG,IAAI,IAAG,IAAI,YAAY,KAAK,kBAAkB,KAAK;AAC9E,cAAM,eAAe,IAAG,IAAI,UAAU,KAAK;AAC3C,cAAM,cAAc,IAAG,IAAI,IAAG,IAAI,iBAAiB,eAAe,KAAK;AACvE,cAAM,YAAY,IAAG,IAAI,IAAG,IAAI,iBAAiB,eAAe,KAAK;AACrE,eAAO,IAAG,SAAS,CAAC,aAAa,YAAY;AAAA;AAAA;AAAA,IAIjD,mBAAmB,kBAAkB;AACnC,aAAO,IAAG,KAAK;AACb,cAAM,YAAY,IAAG,IAAI,IAAG,IAAI,iBAAiB,QAAQ,CAAC,IAAI,GAAG,KAAK,KAAK,kBAAkB,KAAK,QAAQ;AAC1G,eAAO,IAAG,IAAI,WAAW,KAAK;AAAA;AAAA;AAAA,UAI5B,iBAAiB;AACrB,YAAM,kBAAkB,IAAG,KAAK,MAAM,IAAG,IAAI,IAAG,IAAI,OAAO,MAAM;AACjE,YAAM,oBAAoB,KAAK,MAAM,QAAQ;AAC7C,YAAM,aAAa,kBAAkB;AAErC,YAAM,SAAS,IAAG,KAAK,MAAM,IAAG,QAAQ,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI,KAAK;AAE/E,YAAM,WAAW,IAAG,MAAM,YAAY,CAAC,GAAG,IAAI,CAAC,IAAI;AACnD,YAAM,QAAQ,KAAK,eAAe;AAClC,YAAM,uBAAuB,MAAM,IAAG,MAAM,uBAAuB,OAAO,QAAQ,KAAK,UAAU,KAAK,cAAc,KAAK;AACzH,YAAM,iBAAiB,MAAM,qBAAqB;AAClD,YAAM,YAAY;AAAA,QAChB;AAAA,QAAiB;AAAA,QAAmB;AAAA,QAAsB;AAAA,QAC1D;AAAA,QAAO;AAAA,QAAU;AAAA;AAEnB,UAAI,eAAe,WAAW;AAC5B,kBAAU,QAAQ,CAAC,WAAW,OAAO;AACrC,eAAO;AAAA;AAET,YAAM,gBAAgB,IAAG,KAAK;AAC5B,cAAM,gBAAgB;AACtB,mBAAW,KAAK;AACd,gBAAM,WAAW,eAAe;AAChC,gBAAM,cAAc,IAAG,MAAM,OAAO,CAAC,UAAU,IAAI,CAAC,GAAG;AACvD,gBAAM,mBAAmB,IAAG,MAAM,YAAY,CAAC,UAAU,IAAI,CAAC,GAAG;AACjE,gBAAM,gBAAgB,IAAG,KAAK,MAAM,KAAK,mBAAmB,kBAAkB,UAAU,QAAQ,CAAC,IAAI;AACrG,wBAAc,KAAK,CAAE,OAAO,aAAa;AAAA;AAE3C,eAAO;AAAA;AAET,aAAO;AAAA;AAAA,UASH,mBAAmB,OAAO;AAC9B,YAAM,cAAc,MAAM,MAAM;AAChC,YAAM,aAAa,MAAM,MAAM;AAC/B,WAAK,eAAe,QAAO;AAC3B,WAAK,iBAAiB,QAAO;AAC7B,WAAK,WAAW,QAAO;AACvB,YAAM,QAAQ,IAAG,KAAK,MAAM,MAAM,eAAe,CAAC,KAAK,OAAO,KAAK,SAAS,IAAI;AAChF,YAAM,cAAc,MAAM,KAAK,iBAAiB;AAChD,YAAM;AACN,UAAI,CAAC,eAAgB,YAAY,WAAW;AAAI,eAAO;AACvD,YAAM,QAAQ;AACd,iBAAW,KAAK;AACd,cAAM,aAAa,YAAY;AAC/B,cAAM,gBAAgB,MAAM,WAAW,MAAM;AAC7C,cAAM,aAAa,cAAc,GAAG,MAAM,GAAG;AAC7C,cAAM,WAAW,cAAc,GAAG,MAAM,GAAG;AAC3C,cAAM,gBAAgB,MAAM,WAAW,cAAc;AACrD,mBAAW,MAAM;AACjB,mBAAW,cAAc;AACzB,cAAM,KAAK,SAAS,oBAAoB,CAAE,YAAY,UAAU,gBAAiB,CAAC,aAAa,KAAK,OAAO,cAAc,KAAK;AAAA;AAEhI,aAAO;AAAA;AAAA;AAGX,WAAQ,eAAe;AAAA;;;AChGvB;AAAA,WAAQ,mBAAmB;AAAA,IACzB,OAAO,CAAC,GAAG,GAAG,GAAG;AAAA,IACjB,aAAa,CAAC,GAAG,GAAG,GAAG;AAAA,IACvB,cAAc,CAAC,GAAG,IAAI,IAAI;AAAA,IAC1B,YAAY,CAAC,IAAI,IAAI,IAAI;AAAA,IACzB,OAAO,CAAC,IAAI,IAAI,IAAI;AAAA,IACpB,UAAU,CAAC;AAAA;AAAA;;;ACNb;AAAA,4BAA0B;AACxB,WAAO,QAAQ,IAAI,KAAK,KAAK,KAAK,MAAO,SAAQ,KAAK,MAAO,KAAI,KAAK;AAAA;AAExE,WAAQ,mBAAmB;AAE3B,2BAAyB,QAAQ;AAC/B,UAAM,UAAU,KAAK,KAAK,IAAI,KAAK,MAAM,CAAE,QAAO,KAAK,OAAO,KAAK,OAAO,KAAK,OAAO;AACtF,WAAO,iBAAiB;AAAA;AAE1B,WAAQ,kBAAkB;AAE1B,QAAM,yBAAyB,CAAC,GAAG,MAAO,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,GAAG;AACxE,eAAa,IAAI;AACf,QAAI,UAAU;AACd,aAAS,IAAI,GAAG,IAAI,GAAG,QAAQ;AAC7B,iBAAW,GAAG,KAAK,GAAG;AAAA;AAExB,WAAO;AAAA;AAET,WAAQ,MAAM;AAEd,8BAA4B,KAAK;AAC/B,UAAM,SAAS;AACf,aAAS,IAAI,GAAG,IAAI,IAAI,QAAQ;AAC9B,aAAO,KAAK,IAAI,GAAG;AAAA;AAErB,WAAO;AAAA;AAET,WAAQ,qBAAqB;AAE7B,qCAAmC,MAAM;AACvC,UAAM,UAAU;AAChB,UAAM,OAAO,KAAK;AAClB,aAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,cAAQ,KAAK;AACb,eAAS,MAAM,GAAG,MAAM,MAAM;AAC5B,gBAAQ,KAAK,KAAK,IAAI,KAAK,MAAM,mBAAmB,MAAM;AAAA;AAAA;AAG9D,WAAO;AAAA;AAET,+BAA6B,UAAU;AACrC,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,OAAO,KAAK,IAAI;AACtB,UAAM,iBAAiB,CAAC,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,GAAG,GAAG;AAClE,UAAM,oBAAoB,uBAAuB,OAAO,IAAI,OAAO;AACnE,UAAM,2BAA2B,0BAA0B,mBAAmB;AAC9E,UAAM,4BAA4B,uBAAuB,CAAC,OAAO,IAAI,CAAC,OAAO;AAC7E,WAAO,0BAA0B,0BAA0B;AAAA;AAE7D,WAAQ,sBAAsB;AAE9B,iCAA+B;AAC7B,UAAM,oBAAoB,CAAC,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG,KAAK,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AAClF,UAAM,uBAAuB,CAAC,OAAO,GAAG,IAAI,OAAO,GAAG;AACtD,UAAM,sBAAsB;AAAA,MAC1B,CAAC,IAAI,kBAAkB,IAAI;AAAA,MAC3B,CAAC,IAAI,kBAAkB,IAAI;AAAA;AAE7B,WAAO;AAAA,MACL,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,kBAAkB,GAAG,OAAO,oBAAoB;AAAA,MAChD,CAAC,GAAG,GAAG;AAAA;AAAA;AAGX,WAAQ,wBAAwB;AAEhC,uBAAqB,uBAAuB;AAC1C,WAAO;AAAA,MACL,IAAI,uBAAuB,eAAe;AAAA,MAC1C,IAAI,uBAAuB,eAAe;AAAA;AAAA;AAG9C,WAAQ,cAAc;AAAA;;;ACzEtB;AAAA,QAAM,MAAK;AACX,QAAM,WAAW;AACjB,QAAM,OAAO;AAEb,QAAM,0CAA0C;AAChD,QAAM,wBAAwB,CAAC,GAAG;AAClC,QAAM,wBAAwB,CAAC,GAAG;AAClC,QAAM,0BAA0B;AAChC,QAAM,oBAAoB,CAAC,GAAG,GAAG,GAAG,IAAI,IAAI,GAAG;AAC/C,QAAM,oCAAoC;AAC1C,QAAM,6CAA6C;AAGnD;AAAA,IACE,YAAY,qBAAqB,cAAc;AAC7C,WAAK,oBAAoB;AACzB,WAAK,0BAA0B;AAC/B,WAAK,sBAAsB;AAC3B,WAAK,eAAe;AACpB,WAAK,YAAY,QAAO;AACxB,WAAK,aAAa,QAAO;AACzB,WAAK,gBAAgB,QAAO;AAAA;AAAA,IAI9B,uBAAuB,eAAe;AACpC,YAAM,uBAAuB,cAAc,IAAI,CAAC;AAC9C,cAAM,wBAAwB,CAAC,GAAG,OAAO;AACzC,eAAO,KAAK,YAAY,uBAAuB;AAAA;AAEjD,YAAM,gBAAgB,KAAK,8BAA8B;AAGzD,aAAO,SAAS,WAAW,SAAS,YAAY,SAAS,SAAS,eAAe,yBAAyB,KAAK;AAAA;AAAA,IAIjH,uBAAuB;AAIrB,YAAM,cAAc,KAAK,8BAA8B;AACvD,YAAM,gBAAgB,SAAS,WAAW,SAAS,YAAY,SAAS,SAAS,aAAa,yBAAyB;AACvH,YAAM,gBAAgB;AACtB,eAAS,IAAI,GAAG,IAAI,kBAAkB,QAAQ;AAC5C,sBAAc,KAAK,UAAU,kBAAkB,IAAI,MAAM,GAAG;AAAA;AAE9D,oBAAc,gBAAgB;AAC9B,aAAO;AAAA;AAAA,IAKT,mBAAmB,WAAW,KAAK,OAAO;AACxC,YAAM,UAAU,SAAS,WAAW;AACpC,YAAM,cAAc,CAAC,QAAQ,KAAK,KAAK,WAAW,QAAQ,KAAK,KAAK;AACpE,YAAM,eAAe,UAAU,IAAI,CAAC,UAAU;AAAA,QAC5C,YAAY,KAAM,OAAM,KAAK,KAAK,YAAY;AAAA,QAC9C,YAAY,KAAM,OAAM,KAAK,KAAK,aAAa;AAAA,QAAI,MAAM;AAAA;AAE3D,YAAM,uBAAuB,KAAK,oBAAoB,OAAO,CAAC,GAAG;AACjE,YAAM,gBAAgB,aAAa,IAAI,CAAC;AACtC,cAAM,UAAU,KAAK,YAAY,OAAO;AACxC,eAAO,CAAC,GAAG,SAAS,MAAM;AAAA;AAE5B,YAAM,wBAAwB,KAAK,sBAAsB;AACzD,YAAM,YAAY,CAAC,GAAG,SAAS,aAAa,MAAM;AAClD,YAAM,oBAAoB;AAAA,QACxB,KAAK,IAAI,WAAW,sBAAsB;AAAA,QAC1C,KAAK,IAAI,WAAW,sBAAsB;AAAA;AAE5C,aAAO,cAAc,IAAI,CAAC,UAAU;AAAA,QAClC,MAAM,KAAK,kBAAkB;AAAA,QAAI,MAAM,KAAK,kBAAkB;AAAA,QAC9D,MAAM;AAAA;AAAA;AAAA,UAIJ,cAAc,OAAO;AACzB,WAAK,sBAAsB,QAAO;AAClC,WAAK,sBAAsB,QAAO;AAClC,WAAK,WAAW,QAAO;AACvB,YAAM,cAAc,KAAK;AACzB,UAAI,gBAAgB;AAClB,cAAM,yBAAyB,MAAM,KAAK,oBAAoB,mBAAmB,OAAO;AACxF,aAAK,oBAAoB;AACzB,mBAAW,KAAK;AACd,eAAK,wBAAwB,uBAAuB,IAAI,MAAyB;AAAA;AAEnF,aAAK,0BAA0B;AAAA;AAE/B,aAAK;AAAA;AAGP,YAAM,QAAQ;AACd,UAAI,CAAC,KAAK;AAAmB,eAAO;AACpC,iBAAW,KAAK,KAAK;AACnB,cAAM,aAAa,KAAK,kBAAkB,GAAG;AAC7C,YAAI,CAAC;AAAY,iBAAO;AACxB,cAAM,QAAQ,KAAK,gBAAgB,WAAW,cAAc,oCAAoC,WAAW,cAAc;AACzH,cAAM,aAAa,SAAS,aAAa;AACzC,cAAM,uBAAuB,CAAC,WAAW,KAAK,MAAM,MAAM,IAAI,WAAW,KAAK,MAAM,MAAM;AAC1F,cAAM,eAAe,IAAG,MAAM,iBAAiB,OAAO,OAAO,GAAG;AAChE,cAAM,iBAAiB,KAAK,oBAAoB,CAAC,OAAO;AACxD,cAAM,MAAM,cAAc,KAAK,uBAAuB,WAAW,eAAe,kBAAkB;AAClG,cAAM,eAAe,SAAS,yBAAyB,KAAK,cAAc,CAAC,KAAK,WAAW,KAAK;AAChG,cAAM,YAAY,aAAa,IAAI;AACnC,qBAAa;AACb,qBAAa;AACb,cAAM,aAAa,KAAK,aAAa,QAAQ;AAC7C,cAAM,CAAC,MAAM,aAAa;AAC1B,kBAAU;AACV,cAAM,YAAY,KAAK,WAAW;AAClC,aAAK;AACL,YAAI,YAAY,QAAO;AACrB,oBAAU;AACV,eAAK,kBAAkB,KAAK;AAC5B,iBAAO;AAAA;AAET,cAAM,oBAAoB,IAAG,QAAQ,WAAW,CAAC,IAAI;AACrD,cAAM,YAAY,MAAM,kBAAkB;AAC1C,kBAAU;AACV,0BAAkB;AAClB,cAAM,SAAS,KAAK,mBAAmB,WAAW,KAAK,OAAO;AAC9D,cAAM,kBAAkB,KAAK,uBAAuB;AACpD,aAAK,wBAAwB,iBAAiB,OAA2B;AACzE,cAAM,SAAS;AAAA,UACb,WAAW;AAAA,UACX,YAAY;AAAA,UACZ,KAAK;AAAA,YACH,SAAS,gBAAgB;AAAA,YACzB,aAAa,gBAAgB;AAAA;AAAA;AAGjC,cAAM,KAAK;AAAA;AAEb,aAAO;AAAA;AAAA,IAIT,8BAA8B;AAC5B,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE;AAClC,YAAM,aAAa,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AACjD,YAAM,WAAW,CAAC,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,GAAG;AAC/C,aAAO,CAAE,YAAY;AAAA;AAAA,IAKvB,wBAAwB,KAAK,aAAa;AACxC,UAAI;AACF,aAAK,kBAAkB,SAAS,CAAC;AAAA;AAEjC,cAAM,cAAc,KAAK,kBAAkB,OAAO;AAClD,YAAI,MAAM;AACV,YAAI,eAAe,QAAQ,YAAY,cAAc;AACnD,gBAAM,CAAC,WAAW,aAAa,IAAI;AACnC,gBAAM,CAAC,SAAS,WAAW,IAAI;AAC/B,gBAAM,CAAC,mBAAmB,qBAAqB,YAAY;AAC3D,gBAAM,CAAC,iBAAiB,mBAAmB,YAAY;AACvD,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,YAAY,KAAK,IAAI,WAAW;AACtC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,UAAU,KAAK,IAAI,SAAS;AAClC,gBAAM,eAAgB,WAAU,aAAc,WAAU;AACxD,gBAAM,UAAW,WAAU,aAAc,WAAU;AACnD,gBAAM,kBAAmB,mBAAkB,qBAAsB,mBAAkB;AACnF,gBAAM,eAAgB,WAAU,kBAAkB;AAAA;AAEpD,aAAK,kBAAkB,OAAO,KAAK,MAAM,0CAA0C,cAAc;AAAA;AAAA;AAAA,IAIrG;AACE,aAAO,CAAC,KAAK,qBAAsB,KAAK,kBAAkB,WAAW,KAAO,KAAK,2BAA2B,KAAK;AAAA;AAAA;AAGrH,WAAQ,eAAe;AAAA;;;ACjLvB;AAAA,QAAM,MAAK;AACX,QAAM,OAAO;AACb,QAAM,YAAY;AAClB,QAAM,OAAO;AAEb;AAAA,IACE,YAAY;AACV,WAAK,WAAW;AAAA;AAAA,UAGZ,cAAc,OAAO;AACzB,WAAK,sBAAsB,QAAO;AAClC,WAAK,sBAAsB,QAAO;AAClC,WAAK,WAAW,QAAO;AACvB,YAAM,QAAQ,IAAG,KAAK;AACpB,YAAI,CAAE,kBAAiB,IAAG;AACxB,kBAAQ,IAAG,QAAQ,WAAW;AAAA;AAEhC,eAAO,MAAM,UAAU,WAAW;AAAA;AAEpC,YAAM,cAAc,MAAM,KAAK,SAAS,cAAc,OAAO;AAC7D,YAAM;AACN,YAAM,QAAQ;AACd,UAAI,CAAC;AAAa,eAAO;AACzB,iBAAW,cAAc;AACvB,YAAI,CAAC;AAAY,iBAAO;AACxB,cAAM,cAAc;AACpB,mBAAW,OAAO,OAAO,KAAK,UAAU;AACtC,sBAAY,OAAO,UAAU,iBAAiB,KAAK,IAAI,CAAC,UAAU,WAAW,UAAU;AAAA;AAEzF,cAAM,KAAK;AAAA,UACT,YAAY,WAAW,cAAc;AAAA,UACrC,KAAK,WAAW,MAAM,CAAC,WAAW,IAAI,QAAQ,IAAI,WAAW,IAAI,QAAQ,IAAI,WAAW,IAAI,YAAY,KAAK,WAAW,IAAI,QAAQ,IAAI,WAAW,IAAI,YAAY,KAAK,WAAW,IAAI,QAAQ,MAAM;AAAA,UACrM,WAAW,WAAW;AAAA,UACtB;AAAA;AAAA;AAGJ,aAAO;AAAA;AAAA;AAGX,WAAQ,WAAW;AAEnB,6BAA2B;AACzB,QAAI,IAAG,MAAM,SAAS;AAEpB,YAAM,KAAK;AACX,YAAM,OAAO,MAAM,GAAG,aAAa,IAAI,QAAQ,WAAW;AAC1D,aAAO,KAAK,MAAM;AAAA;AAEpB,WAAO,IAAG,KAAK,MAAM,KAAK,KAAK,CAAC,MAAM,EAAE;AAAA;AAG1C,sBAAoB;AAClB,UAAM,CAAC,SAAS,mBAAmB,iBAAiB,MAAM,QAAQ,IAAI;AAAA,MACpE,YAAY,QAAO,SAAS;AAAA,MAC5B,IAAG,eAAe,QAAO,SAAS,WAAW,CAAE,WAAW,QAAO,SAAS,UAAU,SAAS;AAAA,MAC7F,IAAG,eAAe,QAAO,SAAS,WAAW,CAAE,WAAW,QAAO,SAAS,UAAU,SAAS;AAAA;AAE/F,UAAM,WAAW,IAAI,KAAK,aAAa,mBAAmB,SAAS;AACnE,UAAM,WAAW,IAAI,KAAK,aAAa,UAAU,eAAe;AAChE,UAAM,YAAW,IAAI,SAAS;AAC9B,WAAO;AAAA;AAET,WAAQ,OAAO;AAAA;;;AC/Df;AAAA;AAAA;AAAA;AAGA,MAAO,iBAAQ;AAAA,IACb,SAAS;AAAA,IACT,SAAS;AAAA,IACT,MAAM;AAAA,MACJ,SAAS;AAAA,MAGT,UAAU;AAAA,QACR,WAAW;AAAA,QAEX,WAAW;AAAA,QACX,UAAU;AAAA,QACV,YAAY;AAAA,QAGZ,eAAe;AAAA,QACf,cAAc;AAAA,QACd,gBAAgB;AAAA;AAAA,MAElB,MAAM;AAAA,QACJ,SAAS;AAAA,QACT,WAAW;AAAA,QACX,WAAW;AAAA;AAAA,MAEb,MAAM;AAAA,QACJ,SAAS;AAAA,QACT,WAAW;AAAA,QACX,eAAe;AAAA,QACf,WAAW;AAAA;AAAA,MAEb,KAAK;AAAA,QACH,SAAS;AAAA,QACT,WAAW;AAAA,QAEX,WAAW;AAAA,QACX,YAAY;AAAA;AAAA,MAEd,QAAQ;AAAA,QACN,SAAS;AAAA,QACT,eAAe;AAAA,QACf,WAAW;AAAA;AAAA,MAEb,SAAS;AAAA,QACP,SAAS;AAAA,QACT,WAAW;AAAA,QACX,eAAe;AAAA,QACf,YAAY;AAAA,QACZ,cAAc;AAAA,QACd,WAAW;AAAA;AAAA;AAAA,IAGf,MAAM;AAAA,MACJ,SAAS;AAAA,MACT,WAAW;AAAA,MACX,iBAAiB;AAAA,MACjB,cAAc;AAAA,MACd,eAAe;AAAA,MACf,gBAAgB;AAAA,MAChB,WAAW;AAAA;AAAA,IAEb,MAAM;AAAA,MACJ,SAAS;AAAA,MACT,WAAW;AAAA,MACX,YAAY;AAAA,MAGZ,eAAe;AAAA,MACf,cAAc;AAAA,MACd,gBAAgB;AAAA,MAChB,eAAe;AAAA,MACf,UAAU;AAAA,MACV,UAAU;AAAA,QACR,SAAS;AAAA,QACT,WAAW;AAAA;AAAA,MAEb,UAAU;AAAA,QACR,WAAW;AAAA;AAAA;AAAA;AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AC/EjB,MAAM,KAAK;AACX,MAAM,WAAW;AACjB,MAAM,SAAS;AACf,MAAM,UAAU;AAChB,MAAM,UAAU;AAChB,MAAM,WAAW;AACjB,MAAM,WAAW,iBAAuB;AACxC,MAAM,MAAM;AAEZ,IAAI;AAGJ,MAAM,SAAS;AAAA,EACb,UAAU;AAAA,EACV,SAAS;AAAA,EACT,UAAU;AAAA,EACV,MAAM;AAAA,EACN,KAAK;AAAA,EACL,QAAQ;AAAA,EACR,SAAS;AAAA;AAEX,MAAM,MAAM;AACV,MAAI,OAAO,gBAAgB;AAAa,WAAO,YAAY;AAC3D,SAAO,SAAS,OAAO,QAAQ,OAAO,YAAY,MAAO;AAAA;AAG3D,MAAM,MAAM,IAAI;AAEd,MAAI,OAAO;AAAS,YAAQ,IAAI,GAAG;AAAA;AAIrC,sBAAsB;AACpB,QAAM,WAAW,CAAC,QAAQ,OAAO,OAAO,QAAQ;AAChD,SAAO,QAAQ,OAAO,CAAC,MAAM;AAC3B,WAAO,KAAK,OAAO,IAAI,QAAQ,CAAC;AAC9B,YAAM,OAAO,KAAK;AAClB,YAAM,OAAO,IAAI;AACjB,UAAI,MAAM,QAAQ,SAAS,MAAM,QAAQ;AACvC,aAAK,OAAO,KAAK,OAAO,GAAG;AAAA,iBAClB,SAAS,SAAS,SAAS;AACpC,aAAK,OAAO,UAAU,MAAM;AAAA;AAE5B,aAAK,OAAO;AAAA;AAAA;AAGhB,WAAO;AAAA,KACN;AAAA;AAGL,gBAAgB;AACd,MAAI,CAAC;AAAO,WAAO;AACnB,QAAM,QAAQ,MAAM,gBAAgB,MAAM,cAAc,MAAM,SAAU,MAAM,SAAU,MAAM,MAAM,KAAK;AACzG,MAAI,CAAC,SAAU,UAAU;AAAI,WAAO;AACpC,MAAI,MAAM,cAAe,MAAM,cAAc;AAAI,WAAO;AACxD;AACE,OAAG;AAAA;AAEH,WAAO;AAAA;AAET,SAAO;AAAA;AAGT,sBAAsB,OAAO;AAC3B,WAAS,UAAU,UAAU;AAG7B,QAAM,QAAQ,OAAO;AACrB,MAAI;AACF,QAAI,OAAO;AACX,WAAO,CAAE;AAAA;AAIX,SAAO,IAAI,QAAQ,OAAO;AAExB,UAAM,eAAe,OAAO,OAAO,QAAQ,OAAO,CAAC,MAAM,GAAG;AAC5D,QAAI,iBAAiB;AAAG,UAAI;AAG5B,QAAI,GAAG,iBAAiB,OAAO;AAC7B,UAAI,kCAAkC,OAAO;AAC7C,YAAM,GAAG,WAAW,OAAO;AAC3B,YAAM,GAAG;AAAA;AAUX,QAAI,OAAO,KAAK,WAAW,CAAC,OAAO;AAAU,aAAO,WAAW,MAAM,SAAS,KAAK,OAAO;AAC1F,QAAI,OAAO,KAAK,WAAW,CAAC,OAAO;AAAS,aAAO,UAAU,MAAM,QAAQ,KAAK,OAAO;AACvF,QAAI,OAAO,KAAK,WAAW,CAAC,OAAO;AAAU,aAAO,WAAW,MAAM,SAAS,KAAK,OAAO;AAC1F,QAAI,OAAO,KAAK,WAAW,OAAO,KAAK,IAAI,WAAW,CAAC,OAAO;AAAK,aAAO,MAAM,MAAM,OAAO,QAAQ;AACrG,QAAI,OAAO,KAAK,WAAW,OAAO,KAAK,OAAO,WAAW,CAAC,OAAO;AAAQ,aAAO,SAAS,MAAM,OAAO,WAAW;AACjH,QAAI,OAAO,KAAK,WAAW,OAAO,KAAK,QAAQ,WAAW,CAAC,OAAO;AAAS,aAAO,UAAU,MAAM,QAAQ,KAAK;AAE/G,UAAM,OAAO;AACb,QAAI;AAGJ,gBAAY;AACZ,OAAG,SAAS;AACZ,UAAM,UAAU,OAAO,KAAK,UAAU,MAAM,OAAO,QAAQ,cAAc,OAAO,OAAO,QAAQ;AAC/F,OAAG,SAAS;AACZ,SAAK,OAAO,KAAK,MAAM,QAAQ;AAG/B,gBAAY;AACZ,OAAG,SAAS;AACZ,UAAM,UAAU,OAAO,KAAK,UAAU,MAAM,OAAO,SAAS,cAAc,OAAO,OAAO,QAAQ;AAChG,OAAG,SAAS;AACZ,SAAK,OAAO,KAAK,MAAM,QAAQ;AAG/B,UAAM,UAAU;AAChB,QAAI,OAAO,KAAK;AACd,kBAAY;AACZ,SAAG,SAAS;AACZ,YAAM,QAAQ,MAAM,OAAO,SAAS,cAAc,OAAO,OAAO;AAChE,WAAK,OAAO,KAAK,MAAM,QAAQ;AAC/B,iBAAW,QAAQ;AAEjB,YAAI,CAAC,KAAK,SAAS,KAAK,MAAM;AAC5B,cAAI,4BAA4B,KAAK;AACrC;AAAA;AAGF,oBAAY;AACZ,cAAM,UAAW,OAAO,KAAK,IAAI,WAAW,OAAO,KAAK,OAAO,UAAW,MAAM,OAAO,QAAQ,KAAK,OAAO,UAAU;AACrH,aAAK,YAAY,KAAK,MAAM,QAAQ;AAEpC,oBAAY;AACZ,cAAM,cAAc,OAAO,KAAK,QAAQ,UAAU,MAAM,QAAQ,QAAQ,KAAK,OAAO,UAAU;AAC9F,aAAK,UAAU,KAAK,MAAM,QAAQ;AAClC,aAAK,MAAM;AAGX,cAAM,OAAQ,KAAK,YAAY,eAAe,KAAK,YAAY,eAC3D,KAAK,IAAI,KAAK,YAAY,YAAY,GAAG,KAAK,KAAK,YAAY,YAAY,GAAG,IAAI,KAAK,YAAY,aAAa,GAAG,KAAK,KAAK,YAAY,aAAa,GAAG,MACzJ;AACJ,gBAAQ,KAAK;AAAA,UACX,YAAY,KAAK;AAAA,UACjB,KAAK,KAAK;AAAA,UACV,MAAM,KAAK;AAAA,UACX,aAAa,KAAK;AAAA,UAClB,KAAK,QAAQ;AAAA,UACb,QAAQ,QAAQ;AAAA,UAChB,cAAc,QAAQ;AAAA,UACtB,SAAS;AAAA,UACT,MAAO,SAAS,IAAK,KAAK,MAAM,MAAM,OAAmC,QAAQ,MAAM;AAAA;AAAA;AAG3F,SAAG,SAAS;AAAA;AAOd,SAAK,QAAQ,OAAO,OAAO,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI;AACtD,YAAQ,CAAE,MAAM,SAAS,MAAM,SAAS,MAAM,SAAS,aAAa;AAAA;AAAA;AAIxE,QAAQ,SAAS;AACjB,QAAQ,WAAW;AACnB,QAAQ,SAAS;AACjB,QAAQ,SAAS;AACjB,QAAQ,WAAW;AACnB,QAAQ,SAAS;AACjB,QAAQ,UAAU;AAClB,QAAQ,WAAW;AACnB,QAAQ,KAAK;AACb,QAAQ,UAAU,IAAI;",
   "names": []
 }
diff --git a/dist/human.cjs b/dist/human.cjs
index 53ce8fd8..0dcfd5e0 100644
--- a/dist/human.cjs
+++ b/dist/human.cjs
@@ -1,39 +1,39 @@
-var Es=Object.defineProperty,c2=Object.prototype.hasOwnProperty;var Le=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),rf=n=>Es(n,"__esModule",{value:!0}),Sl=(n,t)=>{rf(n);for(var e in t)Es(n,e,{get:t[e],enumerable:!0})},h2=(n,t)=>{if(rf(n),typeof t=="object"||typeof t=="function")for(let e in t)!c2.call(n,e)&&e!=="default"&&Es(n,e,{get:()=>t[e],enumerable:!0});return n},Na=n=>n&&n.__esModule?n:h2(Es({},"default",{value:n,enumerable:!0}),n);var ff=Le(d2=>{Sl(d2,{FetchError:()=>Ut,Headers:()=>vn,Request:()=>Gi,Response:()=>Dn,default:()=>A2});const hn=Na(require("stream")),Ll=Na(require("http")),Ds=Na(require("url")),af=Na(require("https")),ci=Na(require("zlib")),p2=hn.default.Readable,$n=Symbol("buffer"),Il=Symbol("type");class xa{constructor(){this[Il]="";const n=arguments[0],t=arguments[1],e=[];let i=0;if(n){const a=n,s=Number(a.length);for(let o=0;o<s;o++){const l=a[o];let u;l instanceof Buffer?u=l:ArrayBuffer.isView(l)?u=Buffer.from(l.buffer,l.byteOffset,l.byteLength):l instanceof ArrayBuffer?u=Buffer.from(l):l instanceof xa?u=l[$n]:u=Buffer.from(typeof l=="string"?l:String(l)),i+=u.length,e.push(u)}}this[$n]=Buffer.concat(e);let r=t&&t.type!==void 0&&String(t.type).toLowerCase();r&&!/[^\u0020-\u007E]/.test(r)&&(this[Il]=r)}get size(){return this[$n].length}get type(){return this[Il]}text(){return Promise.resolve(this[$n].toString())}arrayBuffer(){const n=this[$n],t=n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength);return Promise.resolve(t)}stream(){const n=new p2;return n._read=function(){},n.push(this[$n]),n.push(null),n}toString(){return"[object Blob]"}slice(){const n=this.size,t=arguments[0],e=arguments[1];let i,r;t===void 0?i=0:t<0?i=Math.max(n+t,0):i=Math.min(t,n),e===void 0?r=n:e<0?r=Math.max(n+e,0):r=Math.min(e,n);const a=Math.max(r-i,0),s=this[$n],o=s.slice(i,i+a),l=new xa([],{type:arguments[2]});return l[$n]=o,l}}Object.defineProperties(xa.prototype,{size:{enumerable:!0},type:{enumerable:!0},slice:{enumerable:!0}});Object.defineProperty(xa.prototype,Symbol.toStringTag,{value:"Blob",writable:!1,enumerable:!1,configurable:!0});function Ut(n,t,e){Error.call(this,n),this.message=n,this.type=t,e&&(this.code=this.errno=e.code),Error.captureStackTrace(this,this.constructor)}Ut.prototype=Object.create(Error.prototype);Ut.prototype.constructor=Ut;Ut.prototype.name="FetchError";let Al;try{Al=require("encoding").convert}catch(n){}const Xn=Symbol("Body internals"),sf=hn.default.PassThrough;function Ct(n){var t=this,e=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{},i=e.size;let r=i===void 0?0:i;var a=e.timeout;let s=a===void 0?0:a;n==null?n=null:of(n)?n=Buffer.from(n.toString()):Ca(n)||(Buffer.isBuffer(n)||(Object.prototype.toString.call(n)==="[object ArrayBuffer]"?n=Buffer.from(n):ArrayBuffer.isView(n)?n=Buffer.from(n.buffer,n.byteOffset,n.byteLength):n instanceof hn.default||(n=Buffer.from(String(n))))),this[Xn]={body:n,disturbed:!1,error:null},this.size=r,this.timeout=s,n instanceof hn.default&&n.on("error",function(o){const l=o.name==="AbortError"?o:new Ut(`Invalid response body while trying to fetch ${t.url}: ${o.message}`,"system",o);t[Xn].error=l})}Ct.prototype={get body(){return this[Xn].body},get bodyUsed(){return this[Xn].disturbed},arrayBuffer(){return Er.call(this).then(function(n){return n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength)})},blob(){let n=this.headers&&this.headers.get("content-type")||"";return Er.call(this).then(function(t){return Object.assign(new xa([],{type:n.toLowerCase()}),{[$n]:t})})},json(){var n=this;return Er.call(this).then(function(t){try{return JSON.parse(t.toString())}catch(e){return Ct.Promise.reject(new Ut(`invalid json response body at ${n.url} reason: ${e.message}`,"invalid-json"))}})},text(){return Er.call(this).then(function(n){return n.toString()})},buffer(){return Er.call(this)},textConverted(){var n=this;return Er.call(this).then(function(t){return f2(t,n.headers)})}};Object.defineProperties(Ct.prototype,{body:{enumerable:!0},bodyUsed:{enumerable:!0},arrayBuffer:{enumerable:!0},blob:{enumerable:!0},json:{enumerable:!0},text:{enumerable:!0}});Ct.mixIn=function(n){for(const t of Object.getOwnPropertyNames(Ct.prototype))if(!(t in n)){const e=Object.getOwnPropertyDescriptor(Ct.prototype,t);Object.defineProperty(n,t,e)}};function Er(){var n=this;if(this[Xn].disturbed)return Ct.Promise.reject(new TypeError(`body used already for: ${this.url}`));if(this[Xn].disturbed=!0,this[Xn].error)return Ct.Promise.reject(this[Xn].error);let t=this.body;if(t===null)return Ct.Promise.resolve(Buffer.alloc(0));if(Ca(t)&&(t=t.stream()),Buffer.isBuffer(t))return Ct.Promise.resolve(t);if(!(t instanceof hn.default))return Ct.Promise.resolve(Buffer.alloc(0));let e=[],i=0,r=!1;return new Ct.Promise(function(a,s){let o;n.timeout&&(o=setTimeout(function(){r=!0,s(new Ut(`Response timeout while trying to fetch ${n.url} (over ${n.timeout}ms)`,"body-timeout"))},n.timeout)),t.on("error",function(l){l.name==="AbortError"?(r=!0,s(l)):s(new Ut(`Invalid response body while trying to fetch ${n.url}: ${l.message}`,"system",l))}),t.on("data",function(l){if(r||l===null)return;if(n.size&&i+l.length>n.size){r=!0,s(new Ut(`content size at ${n.url} over limit: ${n.size}`,"max-size"));return}i+=l.length,e.push(l)}),t.on("end",function(){if(r)return;clearTimeout(o);try{a(Buffer.concat(e,i))}catch(l){s(new Ut(`Could not create Buffer from response body for ${n.url}: ${l.message}`,"system",l))}})})}function f2(n,t){if(typeof Al!="function")throw new Error("The package `encoding` must be installed to use the textConverted() function");const e=t.get("content-type");let i="utf-8",r,a;return e&&(r=/charset=([^;]*)/i.exec(e)),a=n.slice(0,1024).toString(),!r&&a&&(r=/<meta.+?charset=(['"])(.+?)\1/i.exec(a)),!r&&a&&(r=/<meta[\s]+?http-equiv=(['"])content-type\1[\s]+?content=(['"])(.+?)\2/i.exec(a),r||(r=/<meta[\s]+?content=(['"])(.+?)\1[\s]+?http-equiv=(['"])content-type\3/i.exec(a),r&&r.pop()),r&&(r=/charset=(.*)/i.exec(r.pop()))),!r&&a&&(r=/<\?xml.+?encoding=(['"])(.+?)\1/i.exec(a)),r&&(i=r.pop(),(i==="gb2312"||i==="gbk")&&(i="gb18030")),Al(n,"UTF-8",i).toString()}function of(n){return typeof n!="object"||typeof n.append!="function"||typeof n.delete!="function"||typeof n.get!="function"||typeof n.getAll!="function"||typeof n.has!="function"||typeof n.set!="function"?!1:n.constructor.name==="URLSearchParams"||Object.prototype.toString.call(n)==="[object URLSearchParams]"||typeof n.sort=="function"}function Ca(n){return typeof n=="object"&&typeof n.arrayBuffer=="function"&&typeof n.type=="string"&&typeof n.stream=="function"&&typeof n.constructor=="function"&&typeof n.constructor.name=="string"&&/^(Blob|File)$/.test(n.constructor.name)&&/^(Blob|File)$/.test(n[Symbol.toStringTag])}function lf(n){let t,e,i=n.body;if(n.bodyUsed)throw new Error("cannot clone body after it is used");return i instanceof hn.default&&typeof i.getBoundary!="function"&&(t=new sf,e=new sf,i.pipe(t),i.pipe(e),n[Xn].body=t,i=e),i}function uf(n){return n===null?null:typeof n=="string"?"text/plain;charset=UTF-8":of(n)?"application/x-www-form-urlencoded;charset=UTF-8":Ca(n)?n.type||null:Buffer.isBuffer(n)||(Object.prototype.toString.call(n)==="[object ArrayBuffer]"||ArrayBuffer.isView(n))?null:typeof n.getBoundary=="function"?`multipart/form-data;boundary=${n.getBoundary()}`:n instanceof hn.default?null:"text/plain;charset=UTF-8"}function cf(n){const t=n.body;return t===null?0:Ca(t)?t.size:Buffer.isBuffer(t)?t.length:t&&typeof t.getLengthSync=="function"&&(t._lengthRetrievers&&t._lengthRetrievers.length==0||t.hasKnownLength&&t.hasKnownLength())?t.getLengthSync():null}function m2(n,t){const e=t.body;e===null?n.end():Ca(e)?e.stream().pipe(n):Buffer.isBuffer(e)?(n.write(e),n.end()):e.pipe(n)}Ct.Promise=global.Promise;const hf=/[^\^_`a-zA-Z\-0-9!#$%&'*+.|~]/,Tl=/[^\t\x20-\x7e\x80-\xff]/;function Ra(n){if(n=`${n}`,hf.test(n)||n==="")throw new TypeError(`${n} is not a legal HTTP header name`)}function df(n){if(n=`${n}`,Tl.test(n))throw new TypeError(`${n} is not a legal HTTP header value`)}function Dr(n,t){t=t.toLowerCase();for(const e in n)if(e.toLowerCase()===t)return e;return}const ct=Symbol("map");class vn{constructor(){let n=arguments.length>0&&arguments[0]!==void 0?arguments[0]:void 0;if(this[ct]=Object.create(null),n instanceof vn){const t=n.raw(),e=Object.keys(t);for(const i of e)for(const r of t[i])this.append(i,r);return}if(!(n==null))if(typeof n=="object"){const t=n[Symbol.iterator];if(t!=null){if(typeof t!="function")throw new TypeError("Header pairs must be iterable");const e=[];for(const i of n){if(typeof i!="object"||typeof i[Symbol.iterator]!="function")throw new TypeError("Each header pair must be iterable");e.push(Array.from(i))}for(const i of e){if(i.length!==2)throw new TypeError("Each header pair must be a name/value tuple");this.append(i[0],i[1])}}else for(const e of Object.keys(n)){const i=n[e];this.append(e,i)}}else throw new TypeError("Provided initializer must be an object")}get(n){n=`${n}`,Ra(n);const t=Dr(this[ct],n);return t===void 0?null:this[ct][t].join(", ")}forEach(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:void 0,e=Nl(this),i=0;for(;i<e.length;){var r=e[i];const a=r[0],s=r[1];n.call(t,s,a,this),e=Nl(this),i++}}set(n,t){n=`${n}`,t=`${t}`,Ra(n),df(t);const e=Dr(this[ct],n);this[ct][e!==void 0?e:n]=[t]}append(n,t){n=`${n}`,t=`${t}`,Ra(n),df(t);const e=Dr(this[ct],n);e!==void 0?this[ct][e].push(t):this[ct][n]=[t]}has(n){return n=`${n}`,Ra(n),Dr(this[ct],n)!==void 0}delete(n){n=`${n}`,Ra(n);const t=Dr(this[ct],n);t!==void 0&&delete this[ct][t]}raw(){return this[ct]}keys(){return xl(this,"key")}values(){return xl(this,"value")}[Symbol.iterator](){return xl(this,"key+value")}}vn.prototype.entries=vn.prototype[Symbol.iterator];Object.defineProperty(vn.prototype,Symbol.toStringTag,{value:"Headers",writable:!1,enumerable:!1,configurable:!0});Object.defineProperties(vn.prototype,{get:{enumerable:!0},forEach:{enumerable:!0},set:{enumerable:!0},append:{enumerable:!0},has:{enumerable:!0},delete:{enumerable:!0},keys:{enumerable:!0},values:{enumerable:!0},entries:{enumerable:!0}});function Nl(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:"key+value";const e=Object.keys(n[ct]).sort();return e.map(t==="key"?function(i){return i.toLowerCase()}:t==="value"?function(i){return n[ct][i].join(", ")}:function(i){return[i.toLowerCase(),n[ct][i].join(", ")]})}const Cl=Symbol("internal");function xl(n,t){const e=Object.create(Rl);return e[Cl]={target:n,kind:t,index:0},e}const Rl=Object.setPrototypeOf({next(){if(!this||Object.getPrototypeOf(this)!==Rl)throw new TypeError("Value of `this` is not a HeadersIterator");var n=this[Cl];const t=n.target,e=n.kind,i=n.index,r=Nl(t,e),a=r.length;return i>=a?{value:void 0,done:!0}:(this[Cl].index=i+1,{value:r[i],done:!1})}},Object.getPrototypeOf(Object.getPrototypeOf([][Symbol.iterator]())));Object.defineProperty(Rl,Symbol.toStringTag,{value:"HeadersIterator",writable:!1,enumerable:!1,configurable:!0});function g2(n){const t=Object.assign({__proto__:null},n[ct]),e=Dr(n[ct],"Host");return e!==void 0&&(t[e]=t[e][0]),t}function v2(n){const t=new vn;for(const e of Object.keys(n)){if(hf.test(e))continue;if(Array.isArray(n[e]))for(const i of n[e]){if(Tl.test(i))continue;t[ct][e]===void 0?t[ct][e]=[i]:t[ct][e].push(i)}else Tl.test(n[e])||(t[ct][e]=[n[e]])}return t}const hi=Symbol("Response internals"),y2=Ll.default.STATUS_CODES;class Dn{constructor(){let n=arguments.length>0&&arguments[0]!==void 0?arguments[0]:null,t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{};Ct.call(this,n,t);const e=t.status||200,i=new vn(t.headers);if(n!=null&&!i.has("Content-Type")){const r=uf(n);r&&i.append("Content-Type",r)}this[hi]={url:t.url,status:e,statusText:t.statusText||y2[e],headers:i,counter:t.counter}}get url(){return this[hi].url||""}get status(){return this[hi].status}get ok(){return this[hi].status>=200&&this[hi].status<300}get redirected(){return this[hi].counter>0}get statusText(){return this[hi].statusText}get headers(){return this[hi].headers}clone(){return new Dn(lf(this),{url:this.url,status:this.status,statusText:this.statusText,headers:this.headers,ok:this.ok,redirected:this.redirected})}}Ct.mixIn(Dn.prototype);Object.defineProperties(Dn.prototype,{url:{enumerable:!0},status:{enumerable:!0},ok:{enumerable:!0},redirected:{enumerable:!0},statusText:{enumerable:!0},headers:{enumerable:!0},clone:{enumerable:!0}});Object.defineProperty(Dn.prototype,Symbol.toStringTag,{value:"Response",writable:!1,enumerable:!1,configurable:!0});const Jn=Symbol("Request internals"),Ol=Ds.default.parse,b2=Ds.default.format,w2="destroy"in hn.default.Readable.prototype;function ks(n){return typeof n=="object"&&typeof n[Jn]=="object"}function S2(n){const t=n&&typeof n=="object"&&Object.getPrototypeOf(n);return!!(t&&t.constructor.name==="AbortSignal")}class Gi{constructor(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{},e;ks(n)?e=Ol(n.url):(n&&n.href?e=Ol(n.href):e=Ol(`${n}`),n={});let i=t.method||n.method||"GET";if(i=i.toUpperCase(),(t.body!=null||ks(n)&&n.body!==null)&&(i==="GET"||i==="HEAD"))throw new TypeError("Request with GET/HEAD method cannot have body");let r=t.body!=null?t.body:ks(n)&&n.body!==null?lf(n):null;Ct.call(this,r,{timeout:t.timeout||n.timeout||0,size:t.size||n.size||0});const a=new vn(t.headers||n.headers||{});if(r!=null&&!a.has("Content-Type")){const o=uf(r);o&&a.append("Content-Type",o)}let s=ks(n)?n.signal:null;if("signal"in t&&(s=t.signal),s!=null&&!S2(s))throw new TypeError("Expected signal to be an instanceof AbortSignal");this[Jn]={method:i,redirect:t.redirect||n.redirect||"follow",headers:a,parsedURL:e,signal:s},this.follow=t.follow!==void 0?t.follow:n.follow!==void 0?n.follow:20,this.compress=t.compress!==void 0?t.compress:n.compress!==void 0?n.compress:!0,this.counter=t.counter||n.counter||0,this.agent=t.agent||n.agent}get method(){return this[Jn].method}get url(){return b2(this[Jn].parsedURL)}get headers(){return this[Jn].headers}get redirect(){return this[Jn].redirect}get signal(){return this[Jn].signal}clone(){return new Gi(this)}}Ct.mixIn(Gi.prototype);Object.defineProperty(Gi.prototype,Symbol.toStringTag,{value:"Request",writable:!1,enumerable:!1,configurable:!0});Object.defineProperties(Gi.prototype,{method:{enumerable:!0},url:{enumerable:!0},headers:{enumerable:!0},redirect:{enumerable:!0},clone:{enumerable:!0},signal:{enumerable:!0}});function L2(n){const t=n[Jn].parsedURL,e=new vn(n[Jn].headers);if(e.has("Accept")||e.set("Accept","*/*"),!t.protocol||!t.hostname)throw new TypeError("Only absolute URLs are supported");if(!/^https?:$/.test(t.protocol))throw new TypeError("Only HTTP(S) protocols are supported");if(n.signal&&n.body instanceof hn.default.Readable&&!w2)throw new Error("Cancellation of streamed requests with AbortSignal is not supported in node < 8");let i=null;if(n.body==null&&/^(POST|PUT)$/i.test(n.method)&&(i="0"),n.body!=null){const a=cf(n);typeof a=="number"&&(i=String(a))}i&&e.set("Content-Length",i),e.has("User-Agent")||e.set("User-Agent","node-fetch/1.0 (+https://github.com/bitinn/node-fetch)"),n.compress&&!e.has("Accept-Encoding")&&e.set("Accept-Encoding","gzip,deflate");let r=n.agent;return typeof r=="function"&&(r=r(t)),!e.has("Connection")&&!r&&e.set("Connection","close"),Object.assign({},t,{method:n.method,headers:g2(e),agent:r})}function Oa(n){Error.call(this,n),this.type="aborted",this.message=n,Error.captureStackTrace(this,this.constructor)}Oa.prototype=Object.create(Error.prototype);Oa.prototype.constructor=Oa;Oa.prototype.name="AbortError";const pf=hn.default.PassThrough,I2=Ds.default.resolve;function di(n,t){if(!di.Promise)throw new Error("native promise missing, set fetch.Promise to your favorite alternative");return Ct.Promise=di.Promise,new di.Promise(function(e,i){const r=new Gi(n,t),a=L2(r),s=(a.protocol==="https:"?af.default:Ll.default).request,o=r.signal;let l=null;const u=function(){let m=new Oa("The user aborted a request.");if(i(m),r.body&&r.body instanceof hn.default.Readable&&r.body.destroy(m),!l||!l.body)return;l.body.emit("error",m)};if(o&&o.aborted){u();return}const c=function(){u(),p()},h=s(a);let d;o&&o.addEventListener("abort",c);function p(){h.abort(),o&&o.removeEventListener("abort",c),clearTimeout(d)}r.timeout&&h.once("socket",function(f){d=setTimeout(function(){i(new Ut(`network timeout at: ${r.url}`,"request-timeout")),p()},r.timeout)}),h.on("error",function(f){i(new Ut(`request to ${r.url} failed, reason: ${f.message}`,"system",f)),p()}),h.on("response",function(f){clearTimeout(d);const m=v2(f.headers);if(di.isRedirect(f.statusCode)){const S=m.get("Location"),L=S===null?null:I2(r.url,S);switch(r.redirect){case"error":i(new Ut(`uri requested responds with a redirect, redirect mode is set to error: ${r.url}`,"no-redirect")),p();return;case"manual":if(L!==null)try{m.set("Location",L)}catch(C){i(C)}break;case"follow":if(L===null)break;if(r.counter>=r.follow){i(new Ut(`maximum redirect reached at: ${r.url}`,"max-redirect")),p();return}const N={headers:new vn(r.headers),follow:r.follow,counter:r.counter+1,agent:r.agent,compress:r.compress,method:r.method,body:r.body,signal:r.signal,timeout:r.timeout,size:r.size};if(f.statusCode!==303&&r.body&&cf(r)===null){i(new Ut("Cannot follow redirect with body being a readable stream","unsupported-redirect")),p();return}(f.statusCode===303||(f.statusCode===301||f.statusCode===302)&&r.method==="POST")&&(N.method="GET",N.body=void 0,N.headers.delete("content-length")),e(di(new Gi(L,N))),p();return}}f.once("end",function(){o&&o.removeEventListener("abort",c)});let g=f.pipe(new pf);const v={url:r.url,status:f.statusCode,statusText:f.statusMessage,headers:m,size:r.size,timeout:r.timeout,counter:r.counter},b=m.get("Content-Encoding");if(!r.compress||r.method==="HEAD"||b===null||f.statusCode===204||f.statusCode===304){l=new Dn(g,v),e(l);return}const w={flush:ci.default.Z_SYNC_FLUSH,finishFlush:ci.default.Z_SYNC_FLUSH};if(b=="gzip"||b=="x-gzip"){g=g.pipe(ci.default.createGunzip(w)),l=new Dn(g,v),e(l);return}if(b=="deflate"||b=="x-deflate"){const S=f.pipe(new pf);S.once("data",function(L){(L[0]&15)===8?g=g.pipe(ci.default.createInflate()):g=g.pipe(ci.default.createInflateRaw()),l=new Dn(g,v),e(l)});return}if(b=="br"&&typeof ci.default.createBrotliDecompress=="function"){g=g.pipe(ci.default.createBrotliDecompress()),l=new Dn(g,v),e(l);return}l=new Dn(g,v),e(l)}),m2(h,r)})}di.isRedirect=function(n){return n===301||n===302||n===303||n===307||n===308};di.Promise=global.Promise;var A2=di});var lr=Le(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var El=function(n,t){return El=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},El(n,t)};function Zn(n,t){El(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var T2=1e-7,N2=1e-4,x2=function(){function n(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}return n.prototype.get=function(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)},n.prototype.set=function(t,e){this.dataIdsCount++,this.data.set(t,e)},n.prototype.has=function(t){return this.data.has(t)},n.prototype.delete=function(t){return this.dataIdsCount--,this.data.delete(t)},n.prototype.numDataIds=function(){return this.dataIdsCount},n}(),mf=function(){function n(){}return n.prototype.time=function(t){return $("time")},n.prototype.read=function(t){return $("read")},n.prototype.readSync=function(t){return $("readSync")},n.prototype.numDataIds=function(){return $("numDataIds")},n.prototype.disposeData=function(t){return $("disposeData")},n.prototype.write=function(t,e,i){return $("write")},n.prototype.move=function(t,e,i,r){return $("move")},n.prototype.memory=function(){return $("memory")},n.prototype.floatPrecision=function(){return $("floatPrecision")},n.prototype.epsilon=function(){return this.floatPrecision()===32?T2:N2},n.prototype.batchMatMul=function(t,e,i,r){return $("batchMatMul")},n.prototype.fusedBatchMatMul=function(t){var e=t.a,i=t.b,r=t.transposeA,a=t.transposeB,s=t.bias,o=t.activation,l=t.preluActivationWeights;return $("fusedBatchMatMul")},n.prototype.slice=function(t,e,i){return $("slice")},n.prototype.stridedSlice=function(t,e,i,r){return $("stridedSlice")},n.prototype.unstack=function(t,e){return $("unstack")},n.prototype.reverse=function(t,e){return $("reverse")},n.prototype.concat=function(t,e){return $("concat")},n.prototype.neg=function(t){return $("neg")},n.prototype.add=function(t,e){return $("add")},n.prototype.addN=function(t){return $("addN")},n.prototype.subtract=function(t,e){return $("subtract")},n.prototype.multiply=function(t,e){return $("multiply")},n.prototype.realDivide=function(t,e){return $("realDivide")},n.prototype.floorDiv=function(t,e){return $("floorDiv")},n.prototype.sum=function(t,e){return $("sum")},n.prototype.prod=function(t,e){return $("prod")},n.prototype.unsortedSegmentSum=function(t,e,i){return $("unsortedSegmentSum")},n.prototype.argMin=function(t,e){return $("argMin")},n.prototype.argMax=function(t,e){return $("argMax")},n.prototype.equal=function(t,e){return $("equal")},n.prototype.notEqual=function(t,e){return $("notEqual")},n.prototype.less=function(t,e){return $("less")},n.prototype.lessEqual=function(t,e){return $("lessEqual")},n.prototype.greater=function(t,e){return $("greater")},n.prototype.greaterEqual=function(t,e){return $("greaterEqual")},n.prototype.logicalNot=function(t){return $("logicalNot")},n.prototype.logicalAnd=function(t,e){return $("logicalAnd")},n.prototype.logicalOr=function(t,e){return $("logicalOr")},n.prototype.where=function(t){return $("where")},n.prototype.select=function(t,e,i){return $("select")},n.prototype.topk=function(t,e,i){return $("topk")},n.prototype.min=function(t,e){return $("min")},n.prototype.minimum=function(t,e){return $("minimum")},n.prototype.mod=function(t,e){return $("mod")},n.prototype.max=function(t,e){return $("max")},n.prototype.maximum=function(t,e){return $("maximum")},n.prototype.all=function(t,e){return $("all")},n.prototype.any=function(t,e){return $("any")},n.prototype.squaredDifference=function(t,e){return $("squaredDifference")},n.prototype.ceil=function(t){return $("ceil")},n.prototype.floor=function(t){return $("floor")},n.prototype.round=function(t){return $("round")},n.prototype.sign=function(t){return $("sign")},n.prototype.isNaN=function(t){return $("isNaN")},n.prototype.isInf=function(t){return $("isInf")},n.prototype.isFinite=function(t){return $("isFinite")},n.prototype.pow=function(t,e){return $("pow")},n.prototype.exp=function(t){return $("exp")},n.prototype.expm1=function(t){return $("expm1")},n.prototype.softmax=function(t,e){return $("softmax")},n.prototype.log=function(t){return $("log")},n.prototype.log1p=function(t){return $("log1p")},n.prototype.sqrt=function(t){return $("sqrt")},n.prototype.rsqrt=function(t){return $("rsqrt")},n.prototype.square=function(t){return $("square")},n.prototype.reciprocal=function(t){return $("reciprocal")},n.prototype.relu=function(t){return $("relu")},n.prototype.relu6=function(t){return $("relu6")},n.prototype.prelu=function(t,e){return $("prelu")},n.prototype.elu=function(t){return $("elu")},n.prototype.eluDer=function(t,e){return $("eluDer")},n.prototype.selu=function(t){return $("selu")},n.prototype.int=function(t){return $("int")},n.prototype.clip=function(t,e,i){return $("clip")},n.prototype.abs=function(t){return $("abs")},n.prototype.complexAbs=function(t){return $("complexAbs")},n.prototype.sigmoid=function(t){return $("sigmoid")},n.prototype.softplus=function(t){return $("softplus")},n.prototype.sin=function(t){return $("sin")},n.prototype.cos=function(t){return $("cos")},n.prototype.tan=function(t){return $("tan")},n.prototype.asin=function(t){return $("asin")},n.prototype.acos=function(t){return $("acos")},n.prototype.atan=function(t){return $("atan")},n.prototype.atan2=function(t,e){return $("atan2")},n.prototype.sinh=function(t){return $("sinh")},n.prototype.cosh=function(t){return $("cosh")},n.prototype.tanh=function(t){return $("tanh")},n.prototype.asinh=function(t){return $("asinh")},n.prototype.acosh=function(t){return $("acosh")},n.prototype.atanh=function(t){return $("atanh")},n.prototype.erf=function(t){return $("erf")},n.prototype.step=function(t,e){return $("step")},n.prototype.fusedConv2d=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedConv2d")},n.prototype.conv2d=function(t,e,i){return $("conv2d")},n.prototype.conv2dDerInput=function(t,e,i){return $("conv2dDerInput")},n.prototype.conv2dDerFilter=function(t,e,i){return $("conv2dDerFilter")},n.prototype.fusedDepthwiseConv2D=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedDepthwiseConv2D")},n.prototype.depthwiseConv2D=function(t,e,i){return $("depthwiseConv2D")},n.prototype.depthwiseConv2DDerInput=function(t,e,i){return $("depthwiseConv2DDerInput")},n.prototype.depthwiseConv2DDerFilter=function(t,e,i){return $("depthwiseConv2DDerFilter")},n.prototype.conv3d=function(t,e,i){return $("conv3d")},n.prototype.conv3dDerInput=function(t,e,i){return $("conv3dDerInput")},n.prototype.conv3dDerFilter=function(t,e,i){return $("conv3dDerFilter")},n.prototype.maxPool=function(t,e){return $("maxPool")},n.prototype.maxPoolBackprop=function(t,e,i,r){return $("maxPoolBackprop")},n.prototype.avgPool=function(t,e){return $("avgPool")},n.prototype.avgPoolBackprop=function(t,e,i){return $("avgPoolBackprop")},n.prototype.avgPool3d=function(t,e){return $("avgPool3d")},n.prototype.avgPool3dBackprop=function(t,e,i){return $("avgPool3dBackprop")},n.prototype.maxPool3d=function(t,e){return $("maxPool3d")},n.prototype.maxPool3dBackprop=function(t,e,i,r){return $("maxPool3dBackprop")},n.prototype.reshape=function(t,e){return $("reshape")},n.prototype.cast=function(t,e){return $("cast")},n.prototype.tile=function(t,e){return $("tile")},n.prototype.pad=function(t,e,i){return $("pad")},n.prototype.transpose=function(t,e){return $("transpose")},n.prototype.gather=function(t,e,i){return $("gather")},n.prototype.gatherND=function(t,e){return $("gatherND")},n.prototype.scatterND=function(t,e,i){return $("scatterND")},n.prototype.batchToSpaceND=function(t,e,i){return $("batchToSpaceND")},n.prototype.spaceToBatchND=function(t,e,i){return $("spaceToBatchND")},n.prototype.resizeBilinear=function(t,e,i,r){return $("resizeBilinear")},n.prototype.resizeBilinearBackprop=function(t,e,i){return $("resizeBilinearBackprop")},n.prototype.resizeNearestNeighbor=function(t,e,i,r){return $("resizeNearestNeighbor")},n.prototype.resizeNearestNeighborBackprop=function(t,e,i){return $("resizeNearestNeighborBackprop")},n.prototype.batchNorm=function(t,e,i,r,a,s){return $("batchNorm")},n.prototype.localResponseNormalization4D=function(t,e,i,r,a){return $("localResponseNormalization4D")},n.prototype.LRNGrad=function(t,e,i,r,a,s,o){return $("LRNGrad")},n.prototype.multinomial=function(t,e,i,r){return $("multinomial")},n.prototype.oneHot=function(t,e,i,r){return $("oneHot")},n.prototype.cumsum=function(t,e,i,r){return $("cumsum")},n.prototype.nonMaxSuppression=function(t,e,i,r,a){return $("nonMaxSuppression")},n.prototype.fft=function(t){return $("fft")},n.prototype.ifft=function(t){return $("ifft")},n.prototype.complex=function(t,e){return $("complex")},n.prototype.real=function(t){return $("real")},n.prototype.imag=function(t){return $("imag")},n.prototype.cropAndResize=function(t,e,i,r,a,s){return $("cropAndResize")},n.prototype.depthToSpace=function(t,e,i){return $("depthToSpace")},n.prototype.split=function(t,e,i){return $("split")},n.prototype.sparseToDense=function(t,e,i,r){return $("sparseToDense")},n.prototype.diag=function(t){return $("diag")},n.prototype.fill=function(t,e,i){return $("fill")},n.prototype.onesLike=function(t){return $("onesLike")},n.prototype.zerosLike=function(t){return $("zerosLike")},n.prototype.linspace=function(t,e,i){return $("linspace")},n.prototype.dispose=function(){return $("dispose")},n}();function $(n){throw new Error("'"+n+"' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen")}var gf="tfjsflags",vf=function(){function n(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}return n.prototype.setPlatform=function(t,e){this.platform!=null&&console.warn("Platform "+this.platformName+" has already been set. "+("Overwriting the platform with "+e+".")),this.platformName=t,this.platform=e},n.prototype.registerFlag=function(t,e,i){if(this.flagRegistry[t]={evaluationFn:e,setHook:i},this.urlFlags[t]!=null){var r=this.urlFlags[t];console.warn("Setting feature override from URL "+t+": "+r+"."),this.set(t,r)}},n.prototype.getAsync=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return t in this.flags?[2,this.flags[t]]:(e=this.flags,i=t,[4,this.evaluateFlag(t)]);case 1:return e[i]=r.sent(),[2,this.flags[t]]}})})},n.prototype.get=function(t){if(t in this.flags)return this.flags[t];var e=this.evaluateFlag(t);if(e instanceof Promise)throw new Error("Flag "+t+" cannot be synchronously evaluated. Please use getAsync() instead.");return this.flags[t]=e,this.flags[t]},n.prototype.getNumber=function(t){return this.get(t)},n.prototype.getBool=function(t){return this.get(t)},n.prototype.getFlags=function(){return this.flags},Object.defineProperty(n.prototype,"features",{get:function(){return this.flags},enumerable:!0,configurable:!0}),n.prototype.set=function(t,e){if(this.flagRegistry[t]==null)throw new Error("Cannot set flag "+t+" as it has not been registered.");this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)},n.prototype.evaluateFlag=function(t){if(this.flagRegistry[t]==null)throw new Error("Cannot evaluate flag '"+t+"': no evaluation function found.");return this.flagRegistry[t].evaluationFn()},n.prototype.setFlags=function(t){this.flags=Object.assign({},t)},n.prototype.reset=function(){this.flags={},this.urlFlags={},this.populateURLFlags()},n.prototype.populateURLFlags=function(){var t=this;if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;var e=C2(this.global.location.search);if(gf in e){var i=e[gf].split(",");i.forEach(function(r){var a=r.split(":"),s=a[0],o=a[1];t.urlFlags[s]=R2(s,o)})}},n}();function C2(n){var t={};return n.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,function(e){for(var i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];return O2(t,i[0],i[1]),i.join("=")}),t}function O2(n,t,e){n[decodeURIComponent(t)]=decodeURIComponent(e||"")}function R2(n,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(""+ +t===t)return+t;throw new Error("Could not parse value flag value "+t+" for flag "+n+".")}function He(){return I.ENV}I.ENV=null;function E2(n){I.ENV=n}var Dl;function yf(){if(Dl==null){var n=void 0;if(typeof window!="undefined")n=window;else if(typeof global!="undefined")n=global;else if(typeof process!="undefined")n=process;else if(typeof self!="undefined")n=self;else throw new Error("Could not find a global object");Dl=n}return Dl}function D2(){var n=yf();return n._tfGlobals==null&&(n._tfGlobals=new Map),n._tfGlobals}function bf(n,t){var e=D2();if(e.has(n))return e.get(n);var i=t();return e.set(n,i),e.get(n)}var kl="Abs",Fl="Acos",Wl="Acosh",Fs="Add",Ul="AddN",wf="All",Sf="Any",Bl="ArgMax",zl="ArgMin",Pl="Asin",_l="Asinh",Ml="Atan",Hl="Atanh",Vl="Atan2",ql="AvgPool",Lf="AvgPoolBackprop",Gl="AvgPool3D",If="AvgPool3DBackprop",Yl="BatchMatMul",jl="BatchToSpaceND",Kl="BroadcastTo",Ws="Cast",$l="Ceil",Xl="ClipByValue",Af="Complex",Jl="Concat",Zl="Conv2D",Tf="Conv2DBackpropFilter",Ql="Conv2DBackpropInput",eu="Conv3D",Nf="Conv3DBackpropFilterV2",xf="Conv3DBackpropInputV2",tu="Cos",nu="Cosh",iu="Cumsum",Cf="CropAndResize",Rf="DepthToSpace",ru="DepthwiseConv2dNative",Of="DepthwiseConv2dNativeBackpropFilter",Ef="DepthwiseConv2dNativeBackpropInput",Df="Diag",au="Dilation2D",kf="Dilation2DBackpropInput",Ff="Dilation2DBackpropFilter",su="Div",ou="Elu",Wf="EluGrad",lu="Erf",Uf="Equal",uu="Exp",cu="Expm1",Bf="FFT",zf="Fill",Pf="FlipLeftRight",hu="Floor",du="FloorDiv",pu="FusedBatchNorm",fu="GatherV2",_f="GatherNd",Mf="Greater",mu="GreaterEqual",gu="Identity",Hf="IFFT",Vf="Imag",vu="IsFinite",yu="IsInf",bu="IsNan",qf="Less",Gf="LessEqual",Yf="LinSpace",wu="Log",Su="Log1p",jf="LogicalAnd",Kf="LogicalNot",$f="LogicalOr",Lu="LogSoftmax",Iu="LRN",Xf="LRNBackprop",Au="Max",Tu="Maximum",Nu="MaxPool",Jf="MaxPoolBackprop",xu="MaxPool3D",Zf="MaxPool3DBackprop",Qf="MaxPoolWithArgmax",k2="Mean",Cu="Min",Ru="Minimum",Ou="Mod",Eu="Multiply",Du="Negate",em="NotEqual",tm="NonMaxSuppressionV3",nm="NonMaxSuppressionV4",im="NonMaxSuppressionV5",ku="OnesLike",Fu="OneHot",Wu="PadV2",F2="Pool",Uu="Pow",Bu="Prelu",rm="Prod",am="Range",sm="Real",zu="Reciprocal",Pu="Relu",_u="Reshape",Mu="ResizeNearestNeighbor",om="ResizeNearestNeighborGrad",Hu="ResizeBilinear",lm="ResizeBilinearGrad",Vu="Relu6",qu="Reverse",Gu="Round",Yu="Rsqrt",um="ScatterNd",ju="SelectV2",Ku="Selu",$u="Slice",Xu="Sin",Ju="Sinh",Zu="Sign",Qu="Sigmoid",ec="Softplus",tc="Sqrt",nc="Sum",ic="SpaceToBatchND",rc="SplitV",ac="Softmax",sc="SquaredDifference",cm="Square",oc="Sub",hm="SparseToDense",dm="StridedSlice",lc="Tan",uc="Tanh",cc="Tile",pm="TopK",hc="Transpose",fm="Unique",dc="Unpack",pc="UnsortedSegmentSum",fc="ZerosLike",mc="Step",gc="FromPixels",mm="RotateWithOffset",vc="_FusedMatMul",yc="FusedConv2D",bc="FusedDepthwiseConv2D";var kr=bf("kernelRegistry",function(){return new Map}),Ea=bf("gradRegistry",function(){return new Map});function Sc(n,t){var e=wc(n,t);return kr.get(e)}function Lc(n){return Ea.get(n)}function Us(n){for(var t=kr.entries(),e=[];;){var i=t.next(),r=i.done,a=i.value;if(r)break;var s=a[0],o=a[1],l=s.split("_")[0];l===n&&e.push(o)}return e}function gm(n){var t=n.kernelName,e=n.backendName,i=wc(t,e);kr.has(i)&&console.warn("The kernel '"+t+"' for backend "+("'"+e+"' is already registered")),kr.set(i,n)}function vm(n){var t=n.kernelName;Ea.has(t)&&(He().getBool("DEBUG")&&console.warn("Overriding the gradient for '"+t+"'")),Ea.set(t,n)}function W2(n,t){var e=wc(n,t);if(!kr.has(e))throw new Error("The kernel '"+n+"' for backend "+("'"+t+"' is not registered"));kr.delete(e)}function U2(n){if(!Ea.has(n))throw new Error("The gradient '"+n+"' for backend is not registered");Ea.delete(n)}function B2(n,t){var e=Us(n);e.forEach(function(i){var r=Object.assign({},i,{backendName:t});gm(r)})}function wc(n,t){return t+"_"+n}function ym(n){for(var t=n.length,e=0,i=0;t>0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function Da(n,t,e){return Math.max(n,Math.min(t,e))}function z2(n){return n%2===0?n:n+1}function P2(n){for(var t=0,e=0;e<n.length;e++)t+=n[e];return t}function _2(n,t){var e=Math.random();return t*e+(1-e)*n}function M2(n,t){for(var e=0,i=0;i<n.length;i++){var r=Number(n[i])-Number(t[i]);e+=r*r}return e}function E(n,t){if(!n)throw new Error(typeof t=="string"?t:t())}function Be(n,t,e){e===void 0&&(e=""),E(dn(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function Yi(n){E(n!=null,function(){return"The input to the tensor constructor must be a non-null value."})}function ji(n,t,e){if(t===void 0&&(t=[]),e===void 0&&(e=!1),t==null&&(t=[]),Array.isArray(n)||Dt(n)&&!e)for(var i=0;i<n.length;++i)ji(n[i],t,e);else t.push(n);return t}function ot(n){if(n.length===0)return 1;for(var t=n[0],e=1;e<n.length;e++)t*=n[e];return t}function H2(n){return n.length===0}function dn(n,t){if(n===t)return!0;if(n==null||t==null)return!1;if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==t[e])return!1;return!0}function rt(n){return n%1===0}function V2(n){if(Math.tanh!=null)return Math.tanh(n);if(n===Infinity)return 1;if(n===-Infinity)return-1;var t=Math.exp(2*n);return(t-1)/(t+1)}function q2(n){var t=Math.ceil(Math.sqrt(n));return[t,Math.ceil(n/t)]}function G2(n){for(var t=new Uint32Array(n),e=0;e<n;++e)t[e]=e;return ym(t),t}function ka(n,t){return t<=n.length?n:n+" ".repeat(t-n.length)}function Y2(n,t,e){return t===void 0&&(t=function(i){return 0}),new Promise(function(i,r){var a=0,s=function(){if(n()){i();return}a++;var o=t(a);if(e!=null&&a>=e){r();return}setTimeout(s,o)};s()})}function bm(n,t){for(var e=1,i=-1,r=0;r<n.length;++r)if(n[r]>=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Qe(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&i<e}),function(){return"All values in axis param must be in range [-"+e+", "+e+") but "+("got axis "+n)}),E(n.every(function(i){return rt(i)}),function(){return"All values in axis param must be integers but "+("got axis "+n)}),n.map(function(i){return i<0?e+i:i})}function wm(n,t){for(var e=[],i=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Qe(t,n).sort(),s=0,o=0;o<n.length;++o){if(a!=null){if(a[s]===o&&n[o]!==1)throw new Error("Can't squeeze axis "+o+" since its dim '"+n[o]+"' is not 1");(a[s]==null||a[s]>o)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function Bs(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function Sm(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function Lm(n,t){for(var e=0;e<n.length;e++){var i=n[e];if(isNaN(i)||!isFinite(i))throw Error("A tensor of type "+t+" being uploaded contains "+i+".")}}function Im(n){return n==="bool"||n==="complex64"||n==="float32"||n==="int32"||n==="string"}function Am(n,t){return t==="complex64"||(t==="float32"&&n!=="complex64"||t==="int32"&&n!=="float32"&&n!=="complex64")?!1:!(t==="bool"&&n==="bool")}function Dt(n){return n instanceof Float32Array||n instanceof Int32Array||n instanceof Uint8Array}function Tm(n){if(n==="float32"||n==="int32")return 4;if(n==="complex64")return 8;if(n==="bool")return 1;throw new Error("Unknown dtype "+n)}function Nm(n){if(n==null)return 0;var t=0;return n.forEach(function(e){return t+=e.length}),t}function pi(n){return typeof n=="string"||n instanceof String}function xm(n){return typeof n=="boolean"}function Cm(n){return typeof n=="number"}function zs(n){return Array.isArray(n)?zs(n[0]):n instanceof Float32Array?"float32":n instanceof Int32Array||n instanceof Uint8Array?"int32":Cm(n)?"float32":pi(n)?"string":xm(n)?"bool":"float32"}function fi(n){return!!(n&&n.constructor&&n.call&&n.apply)}function Ps(n,t){for(var e=t;e<n;++e)if(n%e===0)return e;return n}function Fr(n){var t=n.length;if(t<2)return[];var e=new Array(t-1);e[t-2]=n[t-1];for(var i=t-3;i>=0;--i)e[i]=e[i+1]*n[i+1];return e}function j2(n,t){return t==="string"?Ic(n):_s([n],t)}function _s(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=ji(n)),He().getBool("DEBUG")&&Lm(n,t),K2(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i<e.length;++i)Math.round(n[i])!==0&&(e[i]=1);return e}else throw new Error("Unknown data type "+t)}function Rm(n,t,e){var i=new Array;if(t.length===1)for(var r=t[0],a=0;a<r;a++)i[a]=e[n+a];else for(var r=t[0],s=t.slice(1),o=s.reduce(function(u,c){return u*c}),a=0;a<r;a++)i[a]=Rm(n+a*o,s,e);return i}function Wr(n,t){if(n.length===0)return t[0];var e=n.reduce(function(i,r){return i*r});if(e===0)return[];if(e!==t.length)throw new Error("["+n+"] does not match the input size "+t.length+".");return Rm(0,n,t)}function K2(n,t){return n instanceof Float32Array&&t==="float32"||n instanceof Int32Array&&t==="int32"||n instanceof Uint8Array&&t==="bool"}function Ac(n,t){for(var e=Ur(n,t),i=0;i<e.length;i++)e[i]=1;return e}function Ur(n,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool")return new Uint8Array(n);throw new Error("Unknown data type "+t)}function $2(n,t){var e=n.reduce(function(i,r){return i*r},1);if(t==null||t==="float32")return Wr(n,new Float32Array(e));if(t==="int32")return Wr(n,new Int32Array(e));if(t==="bool")return Wr(n,new Uint8Array(e));throw new Error("Unknown data type "+t)}function Tc(){return He().platform.now()}function Nc(n){n.forEach(function(t){E(Number.isInteger(t)&&t>=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function X2(n,t){return He().platform.fetch(n,t)}function Ic(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.encode(n,t)}function xc(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.decode(n,t)}function J2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r<n.length-1;++r)i+=e[r]*n[r];return i}function Z2(n,t,e){if(t===0)return[];if(t===1)return[n];for(var i=new Array(t),r=0;r<i.length-1;++r)i[r]=Math.floor(n/e[r]),n-=i[r]*e[r];return i[i.length-1]=n,i}var Q2={__proto__:null,shuffle:ym,clamp:Da,nearestLargerEven:z2,sum:P2,randUniform:_2,distSquared:M2,assert:E,assertShapesMatch:Be,assertNonNull:Yi,flatten:ji,sizeFromShape:ot,isScalarShape:H2,arraysEqual:dn,isInt:rt,tanh:V2,sizeToSquarishShape:q2,createShuffledIndices:G2,rightPad:ka,repeatedTry:Y2,inferFromImplicitShape:bm,parseAxisParam:Qe,squeezeShape:wm,getTypedArrayFromDType:Bs,getArrayFromDType:Sm,checkConversionForErrors:Lm,isValidDtype:Im,hasEncodingLoss:Am,isTypedArray:Dt,bytesPerElement:Tm,bytesFromStringArray:Nm,isString:pi,isBoolean:xm,isNumber:Cm,inferDtype:zs,isFunction:fi,nearestDivisor:Ps,computeStrides:Fr,createScalarValue:j2,toTypedArray:_s,toNestedArray:Wr,makeOnesTypedArray:Ac,makeZerosTypedArray:Ur,makeZerosNestedTypedArray:$2,now:Tc,assertNonNegativeIntegerDimensions:Nc,fetch:X2,encodeString:Ic,decodeString:xc,locToIndex:J2,indexToLoc:Z2};var nA=function(){function n(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new tA)}return n.prototype.profileKernel=function(t,e,i){for(var r,a=function(){r=i()},s=this.backendTimer.time(a),o=function(c){var h=r[c];h.data().then(function(d){eA(d,h.dtype,t)})},l=0;l<r.length;l++)o(l);var u={kernelName:t,outputs:r,inputs:e,timeMs:s.then(function(c){return c.kernelMs}),extraInfo:s.then(function(c){return c.getExtraProfileInfo!=null?c.getExtraProfileInfo():""})};return u},n.prototype.logKernelProfile=function(t){var e=this,i=t.kernelName,r=t.outputs,a=t.timeMs,s=t.inputs,o=t.extraInfo;r.forEach(function(l){Promise.all([l.data(),a,o]).then(function(u){e.logger.logKernelProfile(i,l,u[0],u[1],s,u[2])})})},n}();function eA(n,t,e){if(t!=="float32")return!1;for(var i=0;i<n.length;i++){var r=n[i];if(isNaN(r)||!isFinite(r))return console.warn("Found "+r+" in the result of '"+e+"'"),!0}return!1}var tA=function(){function n(){}return n.prototype.logKernelProfile=function(t,e,i,r,a,s){var o=typeof r=="number"?ka(r+"ms",9):r.error,l=ka(t,25),u=e.rank,c=e.size,h=ka(e.shape.toString(),14),d="";for(var p in a){var f=a[p];if(f!=null){var m=f.shape||e.shape,g=m.length;d+=p+": "+g+"D "+(g>0?m:"")+" "}}console.log("%c"+l+"	%c"+o+"	%c"+u+"D "+h+"	%c"+c+"	%c"+d+"	%c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function iA(n,t,e){for(var i={},r={},a=0;a<t.length;a++)i[t[a].id]=!0;for(var a=0;a<n.length;a++){var s=n[a],o=s.inputs;for(var l in o){for(var u=o[l],c=!1,h=0;h<t.length;h++)if(i[u.id]){s.outputs.forEach(function(w){return i[w.id]=!0}),c=!0,r[s.id]=!0;break}if(c)break}}var d={};d[e.id]=!0;for(var p={},a=n.length-1;a>=0;a--)for(var s=n[a],o=s.inputs,h=0;h<s.outputs.length;h++)if(d[s.outputs[h].id]){for(var l in o)d[o[l].id]=!0,p[s.id]=!0;break}for(var f=[],a=0;a<n.length;a++){var s=n[a];if(r[s.id]&&p[s.id]){var m={};for(var l in s.inputs){var g=s.inputs[l];i[g.id]&&(m[l]=g)}var v=Object.assign({},s);v.inputs=m,v.outputs=s.outputs,f.push(v)}}return f}function rA(n,t,e,i){for(var r=function(s){var o=t[s],l=[];if(o.outputs.forEach(function(d){var p=n[d.id];p!=null?l.push(p):l.push(null)}),o.gradient==null)throw new Error("Cannot compute gradient: gradient function not found "+("for "+o.kernelName+"."));var u=o.gradient(l),c=function(d){if(!(d in u))throw new Error("Cannot backprop through input "+d+". "+("Available gradients found: "+Object.keys(u)+"."));var p=e(function(){return u[d]()});if(p.dtype!=="float32")throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+(d+" must have 'float32' dtype, but has '"+p.dtype+"'"));var f=o.inputs[d];if(!dn(p.shape,f.shape))throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+("'"+d+"' has shape '"+p.shape+"', which does not match ")+("the shape of the input '"+f.shape+"'"));if(n[f.id]==null)n[f.id]=p;else{var m=n[f.id];n[f.id]=i(m,p),m.dispose()}};for(var h in o.inputs)c(h)},a=t.length-1;a>=0;a--)r(a)}var Om=20,Fa=3,Cc=7;function sA(n,t,e,i){var r=Fr(t),a=aA(n,t,e,r),s=t.length,o=Ms(n,t,e,r,a),l=["Tensor"];return i&&(l.push("  dtype: "+e),l.push("  rank: "+s),l.push("  shape: ["+t+"]"),l.push("  values:")),l.push(o.map(function(u){return"    "+u}).join(`
+var Es=Object.defineProperty,c2=Object.prototype.hasOwnProperty;var Le=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),rf=n=>Es(n,"__esModule",{value:!0}),Ll=(n,t)=>{rf(n);for(var e in t)Es(n,e,{get:t[e],enumerable:!0})},h2=(n,t)=>{if(rf(n),typeof t=="object"||typeof t=="function")for(let e in t)!c2.call(n,e)&&e!=="default"&&Es(n,e,{get:()=>t[e],enumerable:!0});return n},Na=n=>n&&n.__esModule?n:h2(Es({},"default",{value:n,enumerable:!0}),n);var ff=Le(d2=>{Ll(d2,{FetchError:()=>Ut,Headers:()=>gn,Request:()=>Gi,Response:()=>Dn,default:()=>A2});const cn=Na(require("stream")),Il=Na(require("http")),Ds=Na(require("url")),af=Na(require("https")),ci=Na(require("zlib")),p2=cn.default.Readable,$n=Symbol("buffer"),Al=Symbol("type");class xa{constructor(){this[Al]="";const n=arguments[0],t=arguments[1],e=[];let i=0;if(n){const a=n,s=Number(a.length);for(let o=0;o<s;o++){const l=a[o];let u;l instanceof Buffer?u=l:ArrayBuffer.isView(l)?u=Buffer.from(l.buffer,l.byteOffset,l.byteLength):l instanceof ArrayBuffer?u=Buffer.from(l):l instanceof xa?u=l[$n]:u=Buffer.from(typeof l=="string"?l:String(l)),i+=u.length,e.push(u)}}this[$n]=Buffer.concat(e);let r=t&&t.type!==void 0&&String(t.type).toLowerCase();r&&!/[^\u0020-\u007E]/.test(r)&&(this[Al]=r)}get size(){return this[$n].length}get type(){return this[Al]}text(){return Promise.resolve(this[$n].toString())}arrayBuffer(){const n=this[$n],t=n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength);return Promise.resolve(t)}stream(){const n=new p2;return n._read=function(){},n.push(this[$n]),n.push(null),n}toString(){return"[object Blob]"}slice(){const n=this.size,t=arguments[0],e=arguments[1];let i,r;t===void 0?i=0:t<0?i=Math.max(n+t,0):i=Math.min(t,n),e===void 0?r=n:e<0?r=Math.max(n+e,0):r=Math.min(e,n);const a=Math.max(r-i,0),s=this[$n],o=s.slice(i,i+a),l=new xa([],{type:arguments[2]});return l[$n]=o,l}}Object.defineProperties(xa.prototype,{size:{enumerable:!0},type:{enumerable:!0},slice:{enumerable:!0}});Object.defineProperty(xa.prototype,Symbol.toStringTag,{value:"Blob",writable:!1,enumerable:!1,configurable:!0});function Ut(n,t,e){Error.call(this,n),this.message=n,this.type=t,e&&(this.code=this.errno=e.code),Error.captureStackTrace(this,this.constructor)}Ut.prototype=Object.create(Error.prototype);Ut.prototype.constructor=Ut;Ut.prototype.name="FetchError";let Tl;try{Tl=require("encoding").convert}catch(n){}const Xn=Symbol("Body internals"),sf=cn.default.PassThrough;function Ct(n){var t=this,e=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{},i=e.size;let r=i===void 0?0:i;var a=e.timeout;let s=a===void 0?0:a;n==null?n=null:of(n)?n=Buffer.from(n.toString()):Ca(n)||(Buffer.isBuffer(n)||(Object.prototype.toString.call(n)==="[object ArrayBuffer]"?n=Buffer.from(n):ArrayBuffer.isView(n)?n=Buffer.from(n.buffer,n.byteOffset,n.byteLength):n instanceof cn.default||(n=Buffer.from(String(n))))),this[Xn]={body:n,disturbed:!1,error:null},this.size=r,this.timeout=s,n instanceof cn.default&&n.on("error",function(o){const l=o.name==="AbortError"?o:new Ut(`Invalid response body while trying to fetch ${t.url}: ${o.message}`,"system",o);t[Xn].error=l})}Ct.prototype={get body(){return this[Xn].body},get bodyUsed(){return this[Xn].disturbed},arrayBuffer(){return Er.call(this).then(function(n){return n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength)})},blob(){let n=this.headers&&this.headers.get("content-type")||"";return Er.call(this).then(function(t){return Object.assign(new xa([],{type:n.toLowerCase()}),{[$n]:t})})},json(){var n=this;return Er.call(this).then(function(t){try{return JSON.parse(t.toString())}catch(e){return Ct.Promise.reject(new Ut(`invalid json response body at ${n.url} reason: ${e.message}`,"invalid-json"))}})},text(){return Er.call(this).then(function(n){return n.toString()})},buffer(){return Er.call(this)},textConverted(){var n=this;return Er.call(this).then(function(t){return f2(t,n.headers)})}};Object.defineProperties(Ct.prototype,{body:{enumerable:!0},bodyUsed:{enumerable:!0},arrayBuffer:{enumerable:!0},blob:{enumerable:!0},json:{enumerable:!0},text:{enumerable:!0}});Ct.mixIn=function(n){for(const t of Object.getOwnPropertyNames(Ct.prototype))if(!(t in n)){const e=Object.getOwnPropertyDescriptor(Ct.prototype,t);Object.defineProperty(n,t,e)}};function Er(){var n=this;if(this[Xn].disturbed)return Ct.Promise.reject(new TypeError(`body used already for: ${this.url}`));if(this[Xn].disturbed=!0,this[Xn].error)return Ct.Promise.reject(this[Xn].error);let t=this.body;if(t===null)return Ct.Promise.resolve(Buffer.alloc(0));if(Ca(t)&&(t=t.stream()),Buffer.isBuffer(t))return Ct.Promise.resolve(t);if(!(t instanceof cn.default))return Ct.Promise.resolve(Buffer.alloc(0));let e=[],i=0,r=!1;return new Ct.Promise(function(a,s){let o;n.timeout&&(o=setTimeout(function(){r=!0,s(new Ut(`Response timeout while trying to fetch ${n.url} (over ${n.timeout}ms)`,"body-timeout"))},n.timeout)),t.on("error",function(l){l.name==="AbortError"?(r=!0,s(l)):s(new Ut(`Invalid response body while trying to fetch ${n.url}: ${l.message}`,"system",l))}),t.on("data",function(l){if(r||l===null)return;if(n.size&&i+l.length>n.size){r=!0,s(new Ut(`content size at ${n.url} over limit: ${n.size}`,"max-size"));return}i+=l.length,e.push(l)}),t.on("end",function(){if(r)return;clearTimeout(o);try{a(Buffer.concat(e,i))}catch(l){s(new Ut(`Could not create Buffer from response body for ${n.url}: ${l.message}`,"system",l))}})})}function f2(n,t){if(typeof Tl!="function")throw new Error("The package `encoding` must be installed to use the textConverted() function");const e=t.get("content-type");let i="utf-8",r,a;return e&&(r=/charset=([^;]*)/i.exec(e)),a=n.slice(0,1024).toString(),!r&&a&&(r=/<meta.+?charset=(['"])(.+?)\1/i.exec(a)),!r&&a&&(r=/<meta[\s]+?http-equiv=(['"])content-type\1[\s]+?content=(['"])(.+?)\2/i.exec(a),r||(r=/<meta[\s]+?content=(['"])(.+?)\1[\s]+?http-equiv=(['"])content-type\3/i.exec(a),r&&r.pop()),r&&(r=/charset=(.*)/i.exec(r.pop()))),!r&&a&&(r=/<\?xml.+?encoding=(['"])(.+?)\1/i.exec(a)),r&&(i=r.pop(),(i==="gb2312"||i==="gbk")&&(i="gb18030")),Tl(n,"UTF-8",i).toString()}function of(n){return typeof n!="object"||typeof n.append!="function"||typeof n.delete!="function"||typeof n.get!="function"||typeof n.getAll!="function"||typeof n.has!="function"||typeof n.set!="function"?!1:n.constructor.name==="URLSearchParams"||Object.prototype.toString.call(n)==="[object URLSearchParams]"||typeof n.sort=="function"}function Ca(n){return typeof n=="object"&&typeof n.arrayBuffer=="function"&&typeof n.type=="string"&&typeof n.stream=="function"&&typeof n.constructor=="function"&&typeof n.constructor.name=="string"&&/^(Blob|File)$/.test(n.constructor.name)&&/^(Blob|File)$/.test(n[Symbol.toStringTag])}function lf(n){let t,e,i=n.body;if(n.bodyUsed)throw new Error("cannot clone body after it is used");return i instanceof cn.default&&typeof i.getBoundary!="function"&&(t=new sf,e=new sf,i.pipe(t),i.pipe(e),n[Xn].body=t,i=e),i}function uf(n){return n===null?null:typeof n=="string"?"text/plain;charset=UTF-8":of(n)?"application/x-www-form-urlencoded;charset=UTF-8":Ca(n)?n.type||null:Buffer.isBuffer(n)||(Object.prototype.toString.call(n)==="[object ArrayBuffer]"||ArrayBuffer.isView(n))?null:typeof n.getBoundary=="function"?`multipart/form-data;boundary=${n.getBoundary()}`:n instanceof cn.default?null:"text/plain;charset=UTF-8"}function cf(n){const t=n.body;return t===null?0:Ca(t)?t.size:Buffer.isBuffer(t)?t.length:t&&typeof t.getLengthSync=="function"&&(t._lengthRetrievers&&t._lengthRetrievers.length==0||t.hasKnownLength&&t.hasKnownLength())?t.getLengthSync():null}function m2(n,t){const e=t.body;e===null?n.end():Ca(e)?e.stream().pipe(n):Buffer.isBuffer(e)?(n.write(e),n.end()):e.pipe(n)}Ct.Promise=global.Promise;const hf=/[^\^_`a-zA-Z\-0-9!#$%&'*+.|~]/,Nl=/[^\t\x20-\x7e\x80-\xff]/;function Ra(n){if(n=`${n}`,hf.test(n)||n==="")throw new TypeError(`${n} is not a legal HTTP header name`)}function df(n){if(n=`${n}`,Nl.test(n))throw new TypeError(`${n} is not a legal HTTP header value`)}function Dr(n,t){t=t.toLowerCase();for(const e in n)if(e.toLowerCase()===t)return e;return}const ct=Symbol("map");class gn{constructor(){let n=arguments.length>0&&arguments[0]!==void 0?arguments[0]:void 0;if(this[ct]=Object.create(null),n instanceof gn){const t=n.raw(),e=Object.keys(t);for(const i of e)for(const r of t[i])this.append(i,r);return}if(!(n==null))if(typeof n=="object"){const t=n[Symbol.iterator];if(t!=null){if(typeof t!="function")throw new TypeError("Header pairs must be iterable");const e=[];for(const i of n){if(typeof i!="object"||typeof i[Symbol.iterator]!="function")throw new TypeError("Each header pair must be iterable");e.push(Array.from(i))}for(const i of e){if(i.length!==2)throw new TypeError("Each header pair must be a name/value tuple");this.append(i[0],i[1])}}else for(const e of Object.keys(n)){const i=n[e];this.append(e,i)}}else throw new TypeError("Provided initializer must be an object")}get(n){n=`${n}`,Ra(n);const t=Dr(this[ct],n);return t===void 0?null:this[ct][t].join(", ")}forEach(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:void 0,e=xl(this),i=0;for(;i<e.length;){var r=e[i];const a=r[0],s=r[1];n.call(t,s,a,this),e=xl(this),i++}}set(n,t){n=`${n}`,t=`${t}`,Ra(n),df(t);const e=Dr(this[ct],n);this[ct][e!==void 0?e:n]=[t]}append(n,t){n=`${n}`,t=`${t}`,Ra(n),df(t);const e=Dr(this[ct],n);e!==void 0?this[ct][e].push(t):this[ct][n]=[t]}has(n){return n=`${n}`,Ra(n),Dr(this[ct],n)!==void 0}delete(n){n=`${n}`,Ra(n);const t=Dr(this[ct],n);t!==void 0&&delete this[ct][t]}raw(){return this[ct]}keys(){return Cl(this,"key")}values(){return Cl(this,"value")}[Symbol.iterator](){return Cl(this,"key+value")}}gn.prototype.entries=gn.prototype[Symbol.iterator];Object.defineProperty(gn.prototype,Symbol.toStringTag,{value:"Headers",writable:!1,enumerable:!1,configurable:!0});Object.defineProperties(gn.prototype,{get:{enumerable:!0},forEach:{enumerable:!0},set:{enumerable:!0},append:{enumerable:!0},has:{enumerable:!0},delete:{enumerable:!0},keys:{enumerable:!0},values:{enumerable:!0},entries:{enumerable:!0}});function xl(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:"key+value";const e=Object.keys(n[ct]).sort();return e.map(t==="key"?function(i){return i.toLowerCase()}:t==="value"?function(i){return n[ct][i].join(", ")}:function(i){return[i.toLowerCase(),n[ct][i].join(", ")]})}const Rl=Symbol("internal");function Cl(n,t){const e=Object.create(Ol);return e[Rl]={target:n,kind:t,index:0},e}const Ol=Object.setPrototypeOf({next(){if(!this||Object.getPrototypeOf(this)!==Ol)throw new TypeError("Value of `this` is not a HeadersIterator");var n=this[Rl];const t=n.target,e=n.kind,i=n.index,r=xl(t,e),a=r.length;return i>=a?{value:void 0,done:!0}:(this[Rl].index=i+1,{value:r[i],done:!1})}},Object.getPrototypeOf(Object.getPrototypeOf([][Symbol.iterator]())));Object.defineProperty(Ol,Symbol.toStringTag,{value:"HeadersIterator",writable:!1,enumerable:!1,configurable:!0});function g2(n){const t=Object.assign({__proto__:null},n[ct]),e=Dr(n[ct],"Host");return e!==void 0&&(t[e]=t[e][0]),t}function v2(n){const t=new gn;for(const e of Object.keys(n)){if(hf.test(e))continue;if(Array.isArray(n[e]))for(const i of n[e]){if(Nl.test(i))continue;t[ct][e]===void 0?t[ct][e]=[i]:t[ct][e].push(i)}else Nl.test(n[e])||(t[ct][e]=[n[e]])}return t}const hi=Symbol("Response internals"),y2=Il.default.STATUS_CODES;class Dn{constructor(){let n=arguments.length>0&&arguments[0]!==void 0?arguments[0]:null,t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{};Ct.call(this,n,t);const e=t.status||200,i=new gn(t.headers);if(n!=null&&!i.has("Content-Type")){const r=uf(n);r&&i.append("Content-Type",r)}this[hi]={url:t.url,status:e,statusText:t.statusText||y2[e],headers:i,counter:t.counter}}get url(){return this[hi].url||""}get status(){return this[hi].status}get ok(){return this[hi].status>=200&&this[hi].status<300}get redirected(){return this[hi].counter>0}get statusText(){return this[hi].statusText}get headers(){return this[hi].headers}clone(){return new Dn(lf(this),{url:this.url,status:this.status,statusText:this.statusText,headers:this.headers,ok:this.ok,redirected:this.redirected})}}Ct.mixIn(Dn.prototype);Object.defineProperties(Dn.prototype,{url:{enumerable:!0},status:{enumerable:!0},ok:{enumerable:!0},redirected:{enumerable:!0},statusText:{enumerable:!0},headers:{enumerable:!0},clone:{enumerable:!0}});Object.defineProperty(Dn.prototype,Symbol.toStringTag,{value:"Response",writable:!1,enumerable:!1,configurable:!0});const Jn=Symbol("Request internals"),El=Ds.default.parse,b2=Ds.default.format,w2="destroy"in cn.default.Readable.prototype;function ks(n){return typeof n=="object"&&typeof n[Jn]=="object"}function S2(n){const t=n&&typeof n=="object"&&Object.getPrototypeOf(n);return!!(t&&t.constructor.name==="AbortSignal")}class Gi{constructor(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{},e;ks(n)?e=El(n.url):(n&&n.href?e=El(n.href):e=El(`${n}`),n={});let i=t.method||n.method||"GET";if(i=i.toUpperCase(),(t.body!=null||ks(n)&&n.body!==null)&&(i==="GET"||i==="HEAD"))throw new TypeError("Request with GET/HEAD method cannot have body");let r=t.body!=null?t.body:ks(n)&&n.body!==null?lf(n):null;Ct.call(this,r,{timeout:t.timeout||n.timeout||0,size:t.size||n.size||0});const a=new gn(t.headers||n.headers||{});if(r!=null&&!a.has("Content-Type")){const o=uf(r);o&&a.append("Content-Type",o)}let s=ks(n)?n.signal:null;if("signal"in t&&(s=t.signal),s!=null&&!S2(s))throw new TypeError("Expected signal to be an instanceof AbortSignal");this[Jn]={method:i,redirect:t.redirect||n.redirect||"follow",headers:a,parsedURL:e,signal:s},this.follow=t.follow!==void 0?t.follow:n.follow!==void 0?n.follow:20,this.compress=t.compress!==void 0?t.compress:n.compress!==void 0?n.compress:!0,this.counter=t.counter||n.counter||0,this.agent=t.agent||n.agent}get method(){return this[Jn].method}get url(){return b2(this[Jn].parsedURL)}get headers(){return this[Jn].headers}get redirect(){return this[Jn].redirect}get signal(){return this[Jn].signal}clone(){return new Gi(this)}}Ct.mixIn(Gi.prototype);Object.defineProperty(Gi.prototype,Symbol.toStringTag,{value:"Request",writable:!1,enumerable:!1,configurable:!0});Object.defineProperties(Gi.prototype,{method:{enumerable:!0},url:{enumerable:!0},headers:{enumerable:!0},redirect:{enumerable:!0},clone:{enumerable:!0},signal:{enumerable:!0}});function L2(n){const t=n[Jn].parsedURL,e=new gn(n[Jn].headers);if(e.has("Accept")||e.set("Accept","*/*"),!t.protocol||!t.hostname)throw new TypeError("Only absolute URLs are supported");if(!/^https?:$/.test(t.protocol))throw new TypeError("Only HTTP(S) protocols are supported");if(n.signal&&n.body instanceof cn.default.Readable&&!w2)throw new Error("Cancellation of streamed requests with AbortSignal is not supported in node < 8");let i=null;if(n.body==null&&/^(POST|PUT)$/i.test(n.method)&&(i="0"),n.body!=null){const a=cf(n);typeof a=="number"&&(i=String(a))}i&&e.set("Content-Length",i),e.has("User-Agent")||e.set("User-Agent","node-fetch/1.0 (+https://github.com/bitinn/node-fetch)"),n.compress&&!e.has("Accept-Encoding")&&e.set("Accept-Encoding","gzip,deflate");let r=n.agent;return typeof r=="function"&&(r=r(t)),!e.has("Connection")&&!r&&e.set("Connection","close"),Object.assign({},t,{method:n.method,headers:g2(e),agent:r})}function Oa(n){Error.call(this,n),this.type="aborted",this.message=n,Error.captureStackTrace(this,this.constructor)}Oa.prototype=Object.create(Error.prototype);Oa.prototype.constructor=Oa;Oa.prototype.name="AbortError";const pf=cn.default.PassThrough,I2=Ds.default.resolve;function di(n,t){if(!di.Promise)throw new Error("native promise missing, set fetch.Promise to your favorite alternative");return Ct.Promise=di.Promise,new di.Promise(function(e,i){const r=new Gi(n,t),a=L2(r),s=(a.protocol==="https:"?af.default:Il.default).request,o=r.signal;let l=null;const u=function(){let m=new Oa("The user aborted a request.");if(i(m),r.body&&r.body instanceof cn.default.Readable&&r.body.destroy(m),!l||!l.body)return;l.body.emit("error",m)};if(o&&o.aborted){u();return}const c=function(){u(),p()},h=s(a);let d;o&&o.addEventListener("abort",c);function p(){h.abort(),o&&o.removeEventListener("abort",c),clearTimeout(d)}r.timeout&&h.once("socket",function(f){d=setTimeout(function(){i(new Ut(`network timeout at: ${r.url}`,"request-timeout")),p()},r.timeout)}),h.on("error",function(f){i(new Ut(`request to ${r.url} failed, reason: ${f.message}`,"system",f)),p()}),h.on("response",function(f){clearTimeout(d);const m=v2(f.headers);if(di.isRedirect(f.statusCode)){const S=m.get("Location"),L=S===null?null:I2(r.url,S);switch(r.redirect){case"error":i(new Ut(`uri requested responds with a redirect, redirect mode is set to error: ${r.url}`,"no-redirect")),p();return;case"manual":if(L!==null)try{m.set("Location",L)}catch(C){i(C)}break;case"follow":if(L===null)break;if(r.counter>=r.follow){i(new Ut(`maximum redirect reached at: ${r.url}`,"max-redirect")),p();return}const N={headers:new gn(r.headers),follow:r.follow,counter:r.counter+1,agent:r.agent,compress:r.compress,method:r.method,body:r.body,signal:r.signal,timeout:r.timeout,size:r.size};if(f.statusCode!==303&&r.body&&cf(r)===null){i(new Ut("Cannot follow redirect with body being a readable stream","unsupported-redirect")),p();return}(f.statusCode===303||(f.statusCode===301||f.statusCode===302)&&r.method==="POST")&&(N.method="GET",N.body=void 0,N.headers.delete("content-length")),e(di(new Gi(L,N))),p();return}}f.once("end",function(){o&&o.removeEventListener("abort",c)});let g=f.pipe(new pf);const v={url:r.url,status:f.statusCode,statusText:f.statusMessage,headers:m,size:r.size,timeout:r.timeout,counter:r.counter},b=m.get("Content-Encoding");if(!r.compress||r.method==="HEAD"||b===null||f.statusCode===204||f.statusCode===304){l=new Dn(g,v),e(l);return}const w={flush:ci.default.Z_SYNC_FLUSH,finishFlush:ci.default.Z_SYNC_FLUSH};if(b=="gzip"||b=="x-gzip"){g=g.pipe(ci.default.createGunzip(w)),l=new Dn(g,v),e(l);return}if(b=="deflate"||b=="x-deflate"){const S=f.pipe(new pf);S.once("data",function(L){(L[0]&15)===8?g=g.pipe(ci.default.createInflate()):g=g.pipe(ci.default.createInflateRaw()),l=new Dn(g,v),e(l)});return}if(b=="br"&&typeof ci.default.createBrotliDecompress=="function"){g=g.pipe(ci.default.createBrotliDecompress()),l=new Dn(g,v),e(l);return}l=new Dn(g,v),e(l)}),m2(h,r)})}di.isRedirect=function(n){return n===301||n===302||n===303||n===307||n===308};di.Promise=global.Promise;var A2=di});var lr=Le(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var Dl=function(n,t){return Dl=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Dl(n,t)};function Zn(n,t){Dl(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var T2=1e-7,N2=1e-4,x2=function(){function n(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}return n.prototype.get=function(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)},n.prototype.set=function(t,e){this.dataIdsCount++,this.data.set(t,e)},n.prototype.has=function(t){return this.data.has(t)},n.prototype.delete=function(t){return this.dataIdsCount--,this.data.delete(t)},n.prototype.numDataIds=function(){return this.dataIdsCount},n}(),mf=function(){function n(){}return n.prototype.time=function(t){return $("time")},n.prototype.read=function(t){return $("read")},n.prototype.readSync=function(t){return $("readSync")},n.prototype.numDataIds=function(){return $("numDataIds")},n.prototype.disposeData=function(t){return $("disposeData")},n.prototype.write=function(t,e,i){return $("write")},n.prototype.move=function(t,e,i,r){return $("move")},n.prototype.memory=function(){return $("memory")},n.prototype.floatPrecision=function(){return $("floatPrecision")},n.prototype.epsilon=function(){return this.floatPrecision()===32?T2:N2},n.prototype.batchMatMul=function(t,e,i,r){return $("batchMatMul")},n.prototype.fusedBatchMatMul=function(t){var e=t.a,i=t.b,r=t.transposeA,a=t.transposeB,s=t.bias,o=t.activation,l=t.preluActivationWeights;return $("fusedBatchMatMul")},n.prototype.slice=function(t,e,i){return $("slice")},n.prototype.stridedSlice=function(t,e,i,r){return $("stridedSlice")},n.prototype.unstack=function(t,e){return $("unstack")},n.prototype.reverse=function(t,e){return $("reverse")},n.prototype.concat=function(t,e){return $("concat")},n.prototype.neg=function(t){return $("neg")},n.prototype.add=function(t,e){return $("add")},n.prototype.addN=function(t){return $("addN")},n.prototype.subtract=function(t,e){return $("subtract")},n.prototype.multiply=function(t,e){return $("multiply")},n.prototype.realDivide=function(t,e){return $("realDivide")},n.prototype.floorDiv=function(t,e){return $("floorDiv")},n.prototype.sum=function(t,e){return $("sum")},n.prototype.prod=function(t,e){return $("prod")},n.prototype.unsortedSegmentSum=function(t,e,i){return $("unsortedSegmentSum")},n.prototype.argMin=function(t,e){return $("argMin")},n.prototype.argMax=function(t,e){return $("argMax")},n.prototype.equal=function(t,e){return $("equal")},n.prototype.notEqual=function(t,e){return $("notEqual")},n.prototype.less=function(t,e){return $("less")},n.prototype.lessEqual=function(t,e){return $("lessEqual")},n.prototype.greater=function(t,e){return $("greater")},n.prototype.greaterEqual=function(t,e){return $("greaterEqual")},n.prototype.logicalNot=function(t){return $("logicalNot")},n.prototype.logicalAnd=function(t,e){return $("logicalAnd")},n.prototype.logicalOr=function(t,e){return $("logicalOr")},n.prototype.where=function(t){return $("where")},n.prototype.select=function(t,e,i){return $("select")},n.prototype.topk=function(t,e,i){return $("topk")},n.prototype.min=function(t,e){return $("min")},n.prototype.minimum=function(t,e){return $("minimum")},n.prototype.mod=function(t,e){return $("mod")},n.prototype.max=function(t,e){return $("max")},n.prototype.maximum=function(t,e){return $("maximum")},n.prototype.all=function(t,e){return $("all")},n.prototype.any=function(t,e){return $("any")},n.prototype.squaredDifference=function(t,e){return $("squaredDifference")},n.prototype.ceil=function(t){return $("ceil")},n.prototype.floor=function(t){return $("floor")},n.prototype.round=function(t){return $("round")},n.prototype.sign=function(t){return $("sign")},n.prototype.isNaN=function(t){return $("isNaN")},n.prototype.isInf=function(t){return $("isInf")},n.prototype.isFinite=function(t){return $("isFinite")},n.prototype.pow=function(t,e){return $("pow")},n.prototype.exp=function(t){return $("exp")},n.prototype.expm1=function(t){return $("expm1")},n.prototype.softmax=function(t,e){return $("softmax")},n.prototype.log=function(t){return $("log")},n.prototype.log1p=function(t){return $("log1p")},n.prototype.sqrt=function(t){return $("sqrt")},n.prototype.rsqrt=function(t){return $("rsqrt")},n.prototype.square=function(t){return $("square")},n.prototype.reciprocal=function(t){return $("reciprocal")},n.prototype.relu=function(t){return $("relu")},n.prototype.relu6=function(t){return $("relu6")},n.prototype.prelu=function(t,e){return $("prelu")},n.prototype.elu=function(t){return $("elu")},n.prototype.eluDer=function(t,e){return $("eluDer")},n.prototype.selu=function(t){return $("selu")},n.prototype.int=function(t){return $("int")},n.prototype.clip=function(t,e,i){return $("clip")},n.prototype.abs=function(t){return $("abs")},n.prototype.complexAbs=function(t){return $("complexAbs")},n.prototype.sigmoid=function(t){return $("sigmoid")},n.prototype.softplus=function(t){return $("softplus")},n.prototype.sin=function(t){return $("sin")},n.prototype.cos=function(t){return $("cos")},n.prototype.tan=function(t){return $("tan")},n.prototype.asin=function(t){return $("asin")},n.prototype.acos=function(t){return $("acos")},n.prototype.atan=function(t){return $("atan")},n.prototype.atan2=function(t,e){return $("atan2")},n.prototype.sinh=function(t){return $("sinh")},n.prototype.cosh=function(t){return $("cosh")},n.prototype.tanh=function(t){return $("tanh")},n.prototype.asinh=function(t){return $("asinh")},n.prototype.acosh=function(t){return $("acosh")},n.prototype.atanh=function(t){return $("atanh")},n.prototype.erf=function(t){return $("erf")},n.prototype.step=function(t,e){return $("step")},n.prototype.fusedConv2d=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedConv2d")},n.prototype.conv2d=function(t,e,i){return $("conv2d")},n.prototype.conv2dDerInput=function(t,e,i){return $("conv2dDerInput")},n.prototype.conv2dDerFilter=function(t,e,i){return $("conv2dDerFilter")},n.prototype.fusedDepthwiseConv2D=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedDepthwiseConv2D")},n.prototype.depthwiseConv2D=function(t,e,i){return $("depthwiseConv2D")},n.prototype.depthwiseConv2DDerInput=function(t,e,i){return $("depthwiseConv2DDerInput")},n.prototype.depthwiseConv2DDerFilter=function(t,e,i){return $("depthwiseConv2DDerFilter")},n.prototype.conv3d=function(t,e,i){return $("conv3d")},n.prototype.conv3dDerInput=function(t,e,i){return $("conv3dDerInput")},n.prototype.conv3dDerFilter=function(t,e,i){return $("conv3dDerFilter")},n.prototype.maxPool=function(t,e){return $("maxPool")},n.prototype.maxPoolBackprop=function(t,e,i,r){return $("maxPoolBackprop")},n.prototype.avgPool=function(t,e){return $("avgPool")},n.prototype.avgPoolBackprop=function(t,e,i){return $("avgPoolBackprop")},n.prototype.avgPool3d=function(t,e){return $("avgPool3d")},n.prototype.avgPool3dBackprop=function(t,e,i){return $("avgPool3dBackprop")},n.prototype.maxPool3d=function(t,e){return $("maxPool3d")},n.prototype.maxPool3dBackprop=function(t,e,i,r){return $("maxPool3dBackprop")},n.prototype.reshape=function(t,e){return $("reshape")},n.prototype.cast=function(t,e){return $("cast")},n.prototype.tile=function(t,e){return $("tile")},n.prototype.pad=function(t,e,i){return $("pad")},n.prototype.transpose=function(t,e){return $("transpose")},n.prototype.gather=function(t,e,i){return $("gather")},n.prototype.gatherND=function(t,e){return $("gatherND")},n.prototype.scatterND=function(t,e,i){return $("scatterND")},n.prototype.batchToSpaceND=function(t,e,i){return $("batchToSpaceND")},n.prototype.spaceToBatchND=function(t,e,i){return $("spaceToBatchND")},n.prototype.resizeBilinear=function(t,e,i,r){return $("resizeBilinear")},n.prototype.resizeBilinearBackprop=function(t,e,i){return $("resizeBilinearBackprop")},n.prototype.resizeNearestNeighbor=function(t,e,i,r){return $("resizeNearestNeighbor")},n.prototype.resizeNearestNeighborBackprop=function(t,e,i){return $("resizeNearestNeighborBackprop")},n.prototype.batchNorm=function(t,e,i,r,a,s){return $("batchNorm")},n.prototype.localResponseNormalization4D=function(t,e,i,r,a){return $("localResponseNormalization4D")},n.prototype.LRNGrad=function(t,e,i,r,a,s,o){return $("LRNGrad")},n.prototype.multinomial=function(t,e,i,r){return $("multinomial")},n.prototype.oneHot=function(t,e,i,r){return $("oneHot")},n.prototype.cumsum=function(t,e,i,r){return $("cumsum")},n.prototype.nonMaxSuppression=function(t,e,i,r,a){return $("nonMaxSuppression")},n.prototype.fft=function(t){return $("fft")},n.prototype.ifft=function(t){return $("ifft")},n.prototype.complex=function(t,e){return $("complex")},n.prototype.real=function(t){return $("real")},n.prototype.imag=function(t){return $("imag")},n.prototype.cropAndResize=function(t,e,i,r,a,s){return $("cropAndResize")},n.prototype.depthToSpace=function(t,e,i){return $("depthToSpace")},n.prototype.split=function(t,e,i){return $("split")},n.prototype.sparseToDense=function(t,e,i,r){return $("sparseToDense")},n.prototype.diag=function(t){return $("diag")},n.prototype.fill=function(t,e,i){return $("fill")},n.prototype.onesLike=function(t){return $("onesLike")},n.prototype.zerosLike=function(t){return $("zerosLike")},n.prototype.linspace=function(t,e,i){return $("linspace")},n.prototype.dispose=function(){return $("dispose")},n}();function $(n){throw new Error("'"+n+"' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen")}var gf="tfjsflags",vf=function(){function n(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}return n.prototype.setPlatform=function(t,e){this.platform!=null&&console.warn("Platform "+this.platformName+" has already been set. "+("Overwriting the platform with "+e+".")),this.platformName=t,this.platform=e},n.prototype.registerFlag=function(t,e,i){if(this.flagRegistry[t]={evaluationFn:e,setHook:i},this.urlFlags[t]!=null){var r=this.urlFlags[t];console.warn("Setting feature override from URL "+t+": "+r+"."),this.set(t,r)}},n.prototype.getAsync=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return t in this.flags?[2,this.flags[t]]:(e=this.flags,i=t,[4,this.evaluateFlag(t)]);case 1:return e[i]=r.sent(),[2,this.flags[t]]}})})},n.prototype.get=function(t){if(t in this.flags)return this.flags[t];var e=this.evaluateFlag(t);if(e instanceof Promise)throw new Error("Flag "+t+" cannot be synchronously evaluated. Please use getAsync() instead.");return this.flags[t]=e,this.flags[t]},n.prototype.getNumber=function(t){return this.get(t)},n.prototype.getBool=function(t){return this.get(t)},n.prototype.getFlags=function(){return this.flags},Object.defineProperty(n.prototype,"features",{get:function(){return this.flags},enumerable:!0,configurable:!0}),n.prototype.set=function(t,e){if(this.flagRegistry[t]==null)throw new Error("Cannot set flag "+t+" as it has not been registered.");this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)},n.prototype.evaluateFlag=function(t){if(this.flagRegistry[t]==null)throw new Error("Cannot evaluate flag '"+t+"': no evaluation function found.");return this.flagRegistry[t].evaluationFn()},n.prototype.setFlags=function(t){this.flags=Object.assign({},t)},n.prototype.reset=function(){this.flags={},this.urlFlags={},this.populateURLFlags()},n.prototype.populateURLFlags=function(){var t=this;if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;var e=C2(this.global.location.search);if(gf in e){var i=e[gf].split(",");i.forEach(function(r){var a=r.split(":"),s=a[0],o=a[1];t.urlFlags[s]=R2(s,o)})}},n}();function C2(n){var t={};return n.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,function(e){for(var i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];return O2(t,i[0],i[1]),i.join("=")}),t}function O2(n,t,e){n[decodeURIComponent(t)]=decodeURIComponent(e||"")}function R2(n,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(""+ +t===t)return+t;throw new Error("Could not parse value flag value "+t+" for flag "+n+".")}function He(){return I.ENV}I.ENV=null;function E2(n){I.ENV=n}var kl;function yf(){if(kl==null){var n=void 0;if(typeof window!="undefined")n=window;else if(typeof global!="undefined")n=global;else if(typeof process!="undefined")n=process;else if(typeof self!="undefined")n=self;else throw new Error("Could not find a global object");kl=n}return kl}function D2(){var n=yf();return n._tfGlobals==null&&(n._tfGlobals=new Map),n._tfGlobals}function bf(n,t){var e=D2();if(e.has(n))return e.get(n);var i=t();return e.set(n,i),e.get(n)}var Fl="Abs",Wl="Acos",Ul="Acosh",Fs="Add",Bl="AddN",wf="All",Sf="Any",zl="ArgMax",Pl="ArgMin",_l="Asin",Ml="Asinh",Hl="Atan",Vl="Atanh",ql="Atan2",Gl="AvgPool",Lf="AvgPoolBackprop",Yl="AvgPool3D",If="AvgPool3DBackprop",jl="BatchMatMul",Kl="BatchToSpaceND",$l="BroadcastTo",Ws="Cast",Xl="Ceil",Jl="ClipByValue",Af="Complex",Zl="Concat",Ql="Conv2D",Tf="Conv2DBackpropFilter",eu="Conv2DBackpropInput",tu="Conv3D",Nf="Conv3DBackpropFilterV2",xf="Conv3DBackpropInputV2",nu="Cos",iu="Cosh",ru="Cumsum",Cf="CropAndResize",Rf="DepthToSpace",au="DepthwiseConv2dNative",Of="DepthwiseConv2dNativeBackpropFilter",Ef="DepthwiseConv2dNativeBackpropInput",Df="Diag",su="Dilation2D",kf="Dilation2DBackpropInput",Ff="Dilation2DBackpropFilter",ou="Div",lu="Elu",Wf="EluGrad",uu="Erf",Uf="Equal",cu="Exp",hu="Expm1",Bf="FFT",zf="Fill",Pf="FlipLeftRight",du="Floor",pu="FloorDiv",fu="FusedBatchNorm",mu="GatherV2",_f="GatherNd",Mf="Greater",gu="GreaterEqual",vu="Identity",Hf="IFFT",Vf="Imag",yu="IsFinite",bu="IsInf",wu="IsNan",qf="Less",Gf="LessEqual",Yf="LinSpace",Su="Log",Lu="Log1p",jf="LogicalAnd",Kf="LogicalNot",$f="LogicalOr",Iu="LogSoftmax",Au="LRN",Xf="LRNBackprop",Tu="Max",Nu="Maximum",xu="MaxPool",Jf="MaxPoolBackprop",Cu="MaxPool3D",Zf="MaxPool3DBackprop",Qf="MaxPoolWithArgmax",k2="Mean",Ru="Min",Ou="Minimum",Eu="Mod",Du="Multiply",ku="Negate",em="NotEqual",tm="NonMaxSuppressionV3",nm="NonMaxSuppressionV4",im="NonMaxSuppressionV5",Fu="OnesLike",Wu="OneHot",Uu="PadV2",F2="Pool",Bu="Pow",zu="Prelu",rm="Prod",am="Range",sm="Real",Pu="Reciprocal",_u="Relu",Mu="Reshape",Hu="ResizeNearestNeighbor",om="ResizeNearestNeighborGrad",Vu="ResizeBilinear",lm="ResizeBilinearGrad",qu="Relu6",Gu="Reverse",Yu="Round",ju="Rsqrt",um="ScatterNd",Ku="SelectV2",$u="Selu",Xu="Slice",Ju="Sin",Zu="Sinh",Qu="Sign",ec="Sigmoid",tc="Softplus",nc="Sqrt",ic="Sum",rc="SpaceToBatchND",ac="SplitV",sc="Softmax",oc="SquaredDifference",cm="Square",lc="Sub",hm="SparseToDense",dm="StridedSlice",uc="Tan",cc="Tanh",hc="Tile",pm="TopK",dc="Transpose",fm="Unique",pc="Unpack",fc="UnsortedSegmentSum",mc="ZerosLike",gc="Step",vc="FromPixels",mm="RotateWithOffset",yc="_FusedMatMul",bc="FusedConv2D",wc="FusedDepthwiseConv2D";var kr=bf("kernelRegistry",function(){return new Map}),Ea=bf("gradRegistry",function(){return new Map});function Lc(n,t){var e=Sc(n,t);return kr.get(e)}function Ic(n){return Ea.get(n)}function Us(n){for(var t=kr.entries(),e=[];;){var i=t.next(),r=i.done,a=i.value;if(r)break;var s=a[0],o=a[1],l=s.split("_")[0];l===n&&e.push(o)}return e}function gm(n){var t=n.kernelName,e=n.backendName,i=Sc(t,e);kr.has(i)&&console.warn("The kernel '"+t+"' for backend "+("'"+e+"' is already registered")),kr.set(i,n)}function vm(n){var t=n.kernelName;Ea.has(t)&&(He().getBool("DEBUG")&&console.warn("Overriding the gradient for '"+t+"'")),Ea.set(t,n)}function W2(n,t){var e=Sc(n,t);if(!kr.has(e))throw new Error("The kernel '"+n+"' for backend "+("'"+t+"' is not registered"));kr.delete(e)}function U2(n){if(!Ea.has(n))throw new Error("The gradient '"+n+"' for backend is not registered");Ea.delete(n)}function B2(n,t){var e=Us(n);e.forEach(function(i){var r=Object.assign({},i,{backendName:t});gm(r)})}function Sc(n,t){return t+"_"+n}function ym(n){for(var t=n.length,e=0,i=0;t>0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function Da(n,t,e){return Math.max(n,Math.min(t,e))}function z2(n){return n%2===0?n:n+1}function P2(n){for(var t=0,e=0;e<n.length;e++)t+=n[e];return t}function _2(n,t){var e=Math.random();return t*e+(1-e)*n}function M2(n,t){for(var e=0,i=0;i<n.length;i++){var r=Number(n[i])-Number(t[i]);e+=r*r}return e}function E(n,t){if(!n)throw new Error(typeof t=="string"?t:t())}function Be(n,t,e){e===void 0&&(e=""),E(hn(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function Yi(n){E(n!=null,function(){return"The input to the tensor constructor must be a non-null value."})}function ji(n,t,e){if(t===void 0&&(t=[]),e===void 0&&(e=!1),t==null&&(t=[]),Array.isArray(n)||Dt(n)&&!e)for(var i=0;i<n.length;++i)ji(n[i],t,e);else t.push(n);return t}function ot(n){if(n.length===0)return 1;for(var t=n[0],e=1;e<n.length;e++)t*=n[e];return t}function H2(n){return n.length===0}function hn(n,t){if(n===t)return!0;if(n==null||t==null)return!1;if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==t[e])return!1;return!0}function rt(n){return n%1===0}function V2(n){if(Math.tanh!=null)return Math.tanh(n);if(n===Infinity)return 1;if(n===-Infinity)return-1;var t=Math.exp(2*n);return(t-1)/(t+1)}function q2(n){var t=Math.ceil(Math.sqrt(n));return[t,Math.ceil(n/t)]}function G2(n){for(var t=new Uint32Array(n),e=0;e<n;++e)t[e]=e;return ym(t),t}function ka(n,t){return t<=n.length?n:n+" ".repeat(t-n.length)}function Y2(n,t,e){return t===void 0&&(t=function(i){return 0}),new Promise(function(i,r){var a=0,s=function(){if(n()){i();return}a++;var o=t(a);if(e!=null&&a>=e){r();return}setTimeout(s,o)};s()})}function bm(n,t){for(var e=1,i=-1,r=0;r<n.length;++r)if(n[r]>=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Qe(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&i<e}),function(){return"All values in axis param must be in range [-"+e+", "+e+") but "+("got axis "+n)}),E(n.every(function(i){return rt(i)}),function(){return"All values in axis param must be integers but "+("got axis "+n)}),n.map(function(i){return i<0?e+i:i})}function wm(n,t){for(var e=[],i=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Qe(t,n).sort(),s=0,o=0;o<n.length;++o){if(a!=null){if(a[s]===o&&n[o]!==1)throw new Error("Can't squeeze axis "+o+" since its dim '"+n[o]+"' is not 1");(a[s]==null||a[s]>o)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function Bs(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function Sm(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function Lm(n,t){for(var e=0;e<n.length;e++){var i=n[e];if(isNaN(i)||!isFinite(i))throw Error("A tensor of type "+t+" being uploaded contains "+i+".")}}function Im(n){return n==="bool"||n==="complex64"||n==="float32"||n==="int32"||n==="string"}function Am(n,t){return t==="complex64"||(t==="float32"&&n!=="complex64"||t==="int32"&&n!=="float32"&&n!=="complex64")?!1:!(t==="bool"&&n==="bool")}function Dt(n){return n instanceof Float32Array||n instanceof Int32Array||n instanceof Uint8Array}function Tm(n){if(n==="float32"||n==="int32")return 4;if(n==="complex64")return 8;if(n==="bool")return 1;throw new Error("Unknown dtype "+n)}function Nm(n){if(n==null)return 0;var t=0;return n.forEach(function(e){return t+=e.length}),t}function pi(n){return typeof n=="string"||n instanceof String}function xm(n){return typeof n=="boolean"}function Cm(n){return typeof n=="number"}function zs(n){return Array.isArray(n)?zs(n[0]):n instanceof Float32Array?"float32":n instanceof Int32Array||n instanceof Uint8Array?"int32":Cm(n)?"float32":pi(n)?"string":xm(n)?"bool":"float32"}function fi(n){return!!(n&&n.constructor&&n.call&&n.apply)}function Ps(n,t){for(var e=t;e<n;++e)if(n%e===0)return e;return n}function Fr(n){var t=n.length;if(t<2)return[];var e=new Array(t-1);e[t-2]=n[t-1];for(var i=t-3;i>=0;--i)e[i]=e[i+1]*n[i+1];return e}function j2(n,t){return t==="string"?Ac(n):_s([n],t)}function _s(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=ji(n)),He().getBool("DEBUG")&&Lm(n,t),K2(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i<e.length;++i)Math.round(n[i])!==0&&(e[i]=1);return e}else throw new Error("Unknown data type "+t)}function Rm(n,t,e){var i=new Array;if(t.length===1)for(var r=t[0],a=0;a<r;a++)i[a]=e[n+a];else for(var r=t[0],s=t.slice(1),o=s.reduce(function(u,c){return u*c}),a=0;a<r;a++)i[a]=Rm(n+a*o,s,e);return i}function Wr(n,t){if(n.length===0)return t[0];var e=n.reduce(function(i,r){return i*r});if(e===0)return[];if(e!==t.length)throw new Error("["+n+"] does not match the input size "+t.length+".");return Rm(0,n,t)}function K2(n,t){return n instanceof Float32Array&&t==="float32"||n instanceof Int32Array&&t==="int32"||n instanceof Uint8Array&&t==="bool"}function Tc(n,t){for(var e=Ur(n,t),i=0;i<e.length;i++)e[i]=1;return e}function Ur(n,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool")return new Uint8Array(n);throw new Error("Unknown data type "+t)}function $2(n,t){var e=n.reduce(function(i,r){return i*r},1);if(t==null||t==="float32")return Wr(n,new Float32Array(e));if(t==="int32")return Wr(n,new Int32Array(e));if(t==="bool")return Wr(n,new Uint8Array(e));throw new Error("Unknown data type "+t)}function Nc(){return He().platform.now()}function xc(n){n.forEach(function(t){E(Number.isInteger(t)&&t>=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function X2(n,t){return He().platform.fetch(n,t)}function Ac(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.encode(n,t)}function Cc(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.decode(n,t)}function J2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r<n.length-1;++r)i+=e[r]*n[r];return i}function Z2(n,t,e){if(t===0)return[];if(t===1)return[n];for(var i=new Array(t),r=0;r<i.length-1;++r)i[r]=Math.floor(n/e[r]),n-=i[r]*e[r];return i[i.length-1]=n,i}var Q2={__proto__:null,shuffle:ym,clamp:Da,nearestLargerEven:z2,sum:P2,randUniform:_2,distSquared:M2,assert:E,assertShapesMatch:Be,assertNonNull:Yi,flatten:ji,sizeFromShape:ot,isScalarShape:H2,arraysEqual:hn,isInt:rt,tanh:V2,sizeToSquarishShape:q2,createShuffledIndices:G2,rightPad:ka,repeatedTry:Y2,inferFromImplicitShape:bm,parseAxisParam:Qe,squeezeShape:wm,getTypedArrayFromDType:Bs,getArrayFromDType:Sm,checkConversionForErrors:Lm,isValidDtype:Im,hasEncodingLoss:Am,isTypedArray:Dt,bytesPerElement:Tm,bytesFromStringArray:Nm,isString:pi,isBoolean:xm,isNumber:Cm,inferDtype:zs,isFunction:fi,nearestDivisor:Ps,computeStrides:Fr,createScalarValue:j2,toTypedArray:_s,toNestedArray:Wr,makeOnesTypedArray:Tc,makeZerosTypedArray:Ur,makeZerosNestedTypedArray:$2,now:Nc,assertNonNegativeIntegerDimensions:xc,fetch:X2,encodeString:Ac,decodeString:Cc,locToIndex:J2,indexToLoc:Z2};var nA=function(){function n(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new tA)}return n.prototype.profileKernel=function(t,e,i){for(var r,a=function(){r=i()},s=this.backendTimer.time(a),o=function(c){var h=r[c];h.data().then(function(d){eA(d,h.dtype,t)})},l=0;l<r.length;l++)o(l);var u={kernelName:t,outputs:r,inputs:e,timeMs:s.then(function(c){return c.kernelMs}),extraInfo:s.then(function(c){return c.getExtraProfileInfo!=null?c.getExtraProfileInfo():""})};return u},n.prototype.logKernelProfile=function(t){var e=this,i=t.kernelName,r=t.outputs,a=t.timeMs,s=t.inputs,o=t.extraInfo;r.forEach(function(l){Promise.all([l.data(),a,o]).then(function(u){e.logger.logKernelProfile(i,l,u[0],u[1],s,u[2])})})},n}();function eA(n,t,e){if(t!=="float32")return!1;for(var i=0;i<n.length;i++){var r=n[i];if(isNaN(r)||!isFinite(r))return console.warn("Found "+r+" in the result of '"+e+"'"),!0}return!1}var tA=function(){function n(){}return n.prototype.logKernelProfile=function(t,e,i,r,a,s){var o=typeof r=="number"?ka(r+"ms",9):r.error,l=ka(t,25),u=e.rank,c=e.size,h=ka(e.shape.toString(),14),d="";for(var p in a){var f=a[p];if(f!=null){var m=f.shape||e.shape,g=m.length;d+=p+": "+g+"D "+(g>0?m:"")+" "}}console.log("%c"+l+"	%c"+o+"	%c"+u+"D "+h+"	%c"+c+"	%c"+d+"	%c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function iA(n,t,e){for(var i={},r={},a=0;a<t.length;a++)i[t[a].id]=!0;for(var a=0;a<n.length;a++){var s=n[a],o=s.inputs;for(var l in o){for(var u=o[l],c=!1,h=0;h<t.length;h++)if(i[u.id]){s.outputs.forEach(function(w){return i[w.id]=!0}),c=!0,r[s.id]=!0;break}if(c)break}}var d={};d[e.id]=!0;for(var p={},a=n.length-1;a>=0;a--)for(var s=n[a],o=s.inputs,h=0;h<s.outputs.length;h++)if(d[s.outputs[h].id]){for(var l in o)d[o[l].id]=!0,p[s.id]=!0;break}for(var f=[],a=0;a<n.length;a++){var s=n[a];if(r[s.id]&&p[s.id]){var m={};for(var l in s.inputs){var g=s.inputs[l];i[g.id]&&(m[l]=g)}var v=Object.assign({},s);v.inputs=m,v.outputs=s.outputs,f.push(v)}}return f}function rA(n,t,e,i){for(var r=function(s){var o=t[s],l=[];if(o.outputs.forEach(function(d){var p=n[d.id];p!=null?l.push(p):l.push(null)}),o.gradient==null)throw new Error("Cannot compute gradient: gradient function not found "+("for "+o.kernelName+"."));var u=o.gradient(l),c=function(d){if(!(d in u))throw new Error("Cannot backprop through input "+d+". "+("Available gradients found: "+Object.keys(u)+"."));var p=e(function(){return u[d]()});if(p.dtype!=="float32")throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+(d+" must have 'float32' dtype, but has '"+p.dtype+"'"));var f=o.inputs[d];if(!hn(p.shape,f.shape))throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+("'"+d+"' has shape '"+p.shape+"', which does not match ")+("the shape of the input '"+f.shape+"'"));if(n[f.id]==null)n[f.id]=p;else{var m=n[f.id];n[f.id]=i(m,p),m.dispose()}};for(var h in o.inputs)c(h)},a=t.length-1;a>=0;a--)r(a)}var Om=20,Fa=3,Rc=7;function sA(n,t,e,i){var r=Fr(t),a=aA(n,t,e,r),s=t.length,o=Ms(n,t,e,r,a),l=["Tensor"];return i&&(l.push("  dtype: "+e),l.push("  rank: "+s),l.push("  shape: ["+t+"]"),l.push("  values:")),l.push(o.map(function(u){return"    "+u}).join(`
 `)),l.join(`
-`)}function aA(n,t,e,i){var r=ot(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?Ua(n):n;if(o>1)for(var u=0;u<r/a;u++)for(var c=u*a,h=0;h<a;h++)s[h]=Math.max(s[h],Wa(l[c+h],0,e).length);return s}function Wa(n,t,e){var i;return Array.isArray(n)?i=parseFloat(n[0].toFixed(Cc))+" + "+(parseFloat(n[1].toFixed(Cc))+"j"):pi(n)?i="'"+n+"'":e==="bool"?i=Em(n):i=parseFloat(n.toFixed(Cc)).toString(),ka(i,t)}function Em(n){return n===0?"false":"true"}function Ms(n,t,e,i,r,a){a===void 0&&(a=!0);var s=e==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(e==="complex64"){var u=Ua(n);return[Wa(u[0],0,e)]}return e==="bool"?[Em(n[0])]:[n[0].toString()]}if(l===1){if(o>Om){var c=Fa*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-Fa)*s,o*s));return e==="complex64"&&(h=Ua(h),d=Ua(d)),["["+h.map(function(C,R){return Wa(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return Wa(C,r[o-Fa+R],e)}).join(", ")+"]"]}var p=e==="complex64"?Ua(n):Array.from(n);return["["+p.map(function(C,R){return Wa(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>Om){for(var b=0;b<Fa;b++){var w=b*g,S=w+g;v.push.apply(v,Ms(n.slice(w,S),f,e,m,r,!1))}v.push("...");for(var b=o-Fa;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,Ms(n.slice(w,S),f,e,m,r,b===o-1))}}else for(var b=0;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,Ms(n.slice(w,S),f,e,m,r,b===o-1))}var L=l===2?",":"";v[0]="["+v[0]+L;for(var b=1;b<v.length-1;b++)v[b]=" "+v[b]+L;for(var N=`,
+`)}function aA(n,t,e,i){var r=ot(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?Ua(n):n;if(o>1)for(var u=0;u<r/a;u++)for(var c=u*a,h=0;h<a;h++)s[h]=Math.max(s[h],Wa(l[c+h],0,e).length);return s}function Wa(n,t,e){var i;return Array.isArray(n)?i=parseFloat(n[0].toFixed(Rc))+" + "+(parseFloat(n[1].toFixed(Rc))+"j"):pi(n)?i="'"+n+"'":e==="bool"?i=Em(n):i=parseFloat(n.toFixed(Rc)).toString(),ka(i,t)}function Em(n){return n===0?"false":"true"}function Ms(n,t,e,i,r,a){a===void 0&&(a=!0);var s=e==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(e==="complex64"){var u=Ua(n);return[Wa(u[0],0,e)]}return e==="bool"?[Em(n[0])]:[n[0].toString()]}if(l===1){if(o>Om){var c=Fa*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-Fa)*s,o*s));return e==="complex64"&&(h=Ua(h),d=Ua(d)),["["+h.map(function(C,R){return Wa(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return Wa(C,r[o-Fa+R],e)}).join(", ")+"]"]}var p=e==="complex64"?Ua(n):Array.from(n);return["["+p.map(function(C,R){return Wa(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>Om){for(var b=0;b<Fa;b++){var w=b*g,S=w+g;v.push.apply(v,Ms(n.slice(w,S),f,e,m,r,!1))}v.push("...");for(var b=o-Fa;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,Ms(n.slice(w,S),f,e,m,r,b===o-1))}}else for(var b=0;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,Ms(n.slice(w,S),f,e,m,r,b===o-1))}var L=l===2?",":"";v[0]="["+v[0]+L;for(var b=1;b<v.length-1;b++)v[b]=" "+v[b]+L;for(var N=`,
 `,b=2;b<l;b++)N+=`
-`;return v[v.length-1]=" "+v[v.length-1]+"]"+(a?"":N),v}function Ua(n){for(var t=[],e=0;e<n.length;e+=2)t.push([n[e],n[e+1]]);return t}var Hs=function(){function n(t,e,i){var r=this;if(this.dtype=e,this.shape=t.slice(),this.size=ot(t),i!=null){var a=i.length;E(a===this.size,function(){return"Length of values '"+a+"' does not match the size "+("inferred by the shape '"+r.size+"'.")})}if(e==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=i||Sm(e,this.size),this.strides=Fr(t)}return n.prototype.set=function(t){for(var e=this,i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];i.length===0&&(i=[0]),E(i.length===this.rank,function(){return"The number of provided coordinates ("+i.length+") must "+("match the rank ("+e.rank+")")});var a=this.locToIndex(i);this.values[a]=t},n.prototype.get=function(){for(var t=[],e=0;e<arguments.length;e++)t[e]=arguments[e];t.length===0&&(t=[0]);for(var i=0,r=0,a=t;r<a.length;r++){var s=a[r];if(s<0||s>=this.shape[i]){var o="Requested out of range element at "+t+". "+("  Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u<t.length-1;++u)l+=this.strides[u]*t[u];return this.values[l]},n.prototype.locToIndex=function(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];for(var e=t[t.length-1],i=0;i<t.length-1;++i)e+=this.strides[i]*t[i];return e},n.prototype.indexToLoc=function(t){if(this.rank===0)return[];if(this.rank===1)return[t];for(var e=new Array(this.shape.length),i=0;i<e.length-1;++i)e[i]=Math.floor(t/this.strides[i]),t-=e[i]*this.strides[i];return e[e.length-1]=t,e},Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.toTensor=function(){return kn().makeTensor(this.values,this.shape,this.dtype)},n}(),kn=null,Br=null;function oA(n){kn=n}function lA(n){Br=n}var Y=function(){function n(t,e,i,r){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=ot(t),this.strides=Fr(t),this.dataId=i,this.id=r,this.rankType=this.rank<5?this.rank.toString():"higher"}return Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.buffer=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Br.buffer(this.shape,this.dtype,t)]}})})},n.prototype.bufferSync=function(){return Br.buffer(this.shape,this.dtype,this.dataSync())},n.prototype.array=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Wr(this.shape,t)]}})})},n.prototype.arraySync=function(){return Wr(this.shape,this.dataSync())},n.prototype.data=function(){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){switch(i.label){case 0:return this.throwIfDisposed(),t=kn().read(this.dataId),this.dtype==="string"?[4,t]:[3,2];case 1:e=i.sent();try{return[2,e.map(function(r){return xc(r)})]}catch(r){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}i.label=2;case 2:return[2,t]}})})},n.prototype.dataSync=function(){this.throwIfDisposed();var t=kn().readSync(this.dataId);if(this.dtype==="string")try{return t.map(function(e){return xc(e)})}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t},n.prototype.bytes=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return this.throwIfDisposed(),[4,kn().read(this.dataId)];case 1:return t=e.sent(),this.dtype==="string"?[2,t]:[2,new Uint8Array(t.buffer)]}})})},n.prototype.dispose=function(){if(this.isDisposed)return;kn().disposeTensor(this),this.isDisposedInternal=!0},Object.defineProperty(n.prototype,"isDisposed",{get:function(){return this.isDisposedInternal},enumerable:!0,configurable:!0}),n.prototype.throwIfDisposed=function(){if(this.isDisposed)throw new Error("Tensor is disposed.")},n.prototype.print=function(t){return t===void 0&&(t=!1),Br.print(this,t)},n.prototype.clone=function(){return this.throwIfDisposed(),Br.clone(this)},n.prototype.toString=function(t){t===void 0&&(t=!1);var e=this.dataSync();return sA(e,this.shape,this.dtype,t)},n.prototype.cast=function(t){return this.throwIfDisposed(),Br.cast(this,t)},n.prototype.variable=function(t,e,i){return t===void 0&&(t=!0),this.throwIfDisposed(),kn().makeVariable(this,t,e,i)},n}();Object.defineProperty(Y,Symbol.hasInstance,{value:function(n){return!!n&&n.data!=null&&n.dataSync!=null&&n.throwIfDisposed!=null}});var Ba=function(n){Zn(t,n);function t(e,i,r,a){var s=n.call(this,e.shape,e.dtype,e.dataId,a)||this;return s.trainable=i,s.name=r,s}return t.prototype.assign=function(e){if(e.dtype!==this.dtype)throw new Error("dtype of the new value ("+e.dtype+") and "+("previous value ("+this.dtype+") must match"));if(!dn(e.shape,this.shape))throw new Error("shape of the new value ("+e.shape+") and "+("previous value ("+this.shape+") must match"));kn().disposeTensor(this),this.dataId=e.dataId,kn().incRef(this,null)},t.prototype.dispose=function(){kn().disposeVariable(this),this.isDisposedInternal=!0},t}(Y);Object.defineProperty(Ba,Symbol.hasInstance,{value:function(n){return n instanceof Y&&n.assign!=null&&n.assign instanceof Function}});(function(n){n.R0="R0",n.R1="R1",n.R2="R2",n.R3="R3",n.R4="R4",n.R5="R5",n.R6="R6"})(I.Rank||(I.Rank={}));var Rc;(function(n){n.float32="float32",n.int32="int32",n.bool="int32",n.complex64="complex64"})(Rc||(Rc={}));var Oc;(function(n){n.float32="float32",n.int32="int32",n.bool="bool",n.complex64="complex64"})(Oc||(Oc={}));var Ec;(function(n){n.float32="float32",n.int32="float32",n.bool="float32",n.complex64="complex64"})(Ec||(Ec={}));var Dc;(function(n){n.float32="complex64",n.int32="complex64",n.bool="complex64",n.complex64="complex64"})(Dc||(Dc={}));var uA={float32:Ec,int32:Rc,bool:Oc,complex64:Dc};function Vs(n,t){if(n==="string"||t==="string"){if(n==="string"&&t==="string")return"string";throw new Error("Can not upcast "+n+" with "+t)}return uA[n][t]}function cA(n){return Vs(n,"int32")}function at(n,t){if(n.dtype===t.dtype)return[n,t];var e=Vs(n.dtype,t.dtype);return[n.cast(e),t.cast(e)]}function Dm(n,t){E(n.dtype===t.dtype,function(){return"The dtypes of the first("+n.dtype+") and"+(" second("+t.dtype+") input must match")})}function hA(n,t){return t.some(function(e){return e.id===n.id})}function kc(n){var t=[],e=new Set;return km(n,t,e),t}function km(n,t,e){if(n==null)return;if(n instanceof Y){t.push(n);return}if(!dA(n))return;var i=n;for(var r in i){var a=i[r];e.has(a)||(e.add(a),km(a,t,e))}}function dA(n){return Array.isArray(n)||typeof n=="object"}var pA={__proto__:null,makeTypesMatch:at,assertTypesMatch:Dm,isTensorInList:hA,getTensorsInContainer:kc};var Fm=function(){function n(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null}}return n.prototype.dispose=function(){for(var t in this.registeredVariables)this.registeredVariables[t].dispose()},n}(),gA=function(){function n(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Fm}return n.prototype.ready=function(){return de(this,void 0,void 0,function(){var t,e,i,r;return pe(this,function(a){switch(a.label){case 0:if(this.pendingBackendInit!=null)return[2,this.pendingBackendInit.then(function(){})];if(this.backendInstance!=null)return[2];t=this.getSortedBackends(),e=0,a.label=1;case 1:return e<t.length?(i=t[e],[4,this.initializeBackend(i).success]):[3,5];case 2:return r=a.sent(),r?[4,this.setBackend(i)]:[3,4];case 3:return a.sent(),[2];case 4:return e++,[3,1];case 5:throw new Error("Could not initialize any backends, all backend initializations failed.")}})})},Object.defineProperty(n.prototype,"backend",{get:function(){if(this.pendingBackendInit!=null)throw new Error("Backend '"+this.backendName+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");if(this.backendInstance==null){var t=this.initializeBackendsAndReturnBest(),e=t.name,i=t.asyncInit;if(i)throw new Error("The highest priority backend '"+e+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");this.setBackend(e)}return this.backendInstance},enumerable:!0,configurable:!0}),n.prototype.backendNames=function(){return Object.keys(this.registryFactory)},n.prototype.findBackend=function(t){if(!(t in this.registry))if(t in this.registryFactory){var e=this.initializeBackend(t).asyncInit;if(e)return null}else return null;return this.registry[t]},n.prototype.findBackendFactory=function(t){return t in this.registryFactory?this.registryFactory[t].factory:null},n.prototype.registerBackend=function(t,e,i){return i===void 0&&(i=1),t in this.registryFactory?(console.warn(t+" backend was already registered. Reusing existing backend factory."),!1):(this.registryFactory[t]={factory:e,priority:i},!0)},n.prototype.setBackend=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s;return pe(this,function(o){switch(o.label){case 0:if(this.registryFactory[t]==null)throw new Error("Backend name '"+t+"' not found in registry");return this.backendName=t,this.registry[t]==null?(this.backendInstance=null,e=this.initializeBackend(t),i=e.success,r=e.asyncInit,r?[4,i]:[3,2]):[3,4];case 1:return s=o.sent(),[3,3];case 2:s=i,o.label=3;case 3:if(a=s,!a)return[2,!1];o.label=4;case 4:return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new nA(this.backendInstance),[2,!0]}})})},n.prototype.setupRegisteredKernels=function(){var t=this,e=Us(this.backendName);e.forEach(function(i){i.setupFunc!=null&&i.setupFunc(t.backendInstance)})},n.prototype.disposeRegisteredKernels=function(t){var e=this,i=Us(t);i.forEach(function(r){r.disposeFunc!=null&&r.disposeFunc(e.registry[t])})},n.prototype.initializeBackend=function(t){var e=this,i=this.registryFactory[t];if(i==null)throw new Error("Cannot initialize backend "+t+", no registration found.");try{var r=i.factory();if(r&&!(r instanceof mf)&&typeof r.then=="function"){var a=++this.pendingBackendInitId,s=r.then(function(o){return a<e.pendingBackendInitId?!1:(e.registry[t]=o,e.pendingBackendInit=null,!0)}).catch(function(o){return a<e.pendingBackendInitId||(e.pendingBackendInit=null,console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message)),!1});return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=r,{success:!0,asyncInit:!1}}catch(o){return console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message),{success:!1,asyncInit:!1}}},n.prototype.removeBackend=function(t){if(!(t in this.registryFactory))throw new Error(t+" backend not found in registry");this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)},n.prototype.getSortedBackends=function(){var t=this;if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(function(e,i){return t.registryFactory[i].priority-t.registryFactory[e].priority})},n.prototype.initializeBackendsAndReturnBest=function(){for(var t=this.getSortedBackends(),e=0;e<t.length;e++){var i=t[e],r=this.initializeBackend(i),a=r.success,s=r.asyncInit;if(s||a)return{name:i,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")},n.prototype.moveData=function(t,e){var i=this.state.tensorInfo.get(e),r=i.backend,a=this.readSync(e);r.disposeData(e),i.backend=t,t.move(e,a,i.shape,i.dtype),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++},n.prototype.tidy=function(t,e){var i=this,r=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");r=t}var a;return this.scopedRun(function(){return i.startScope(r)},function(){return i.endScope(a)},function(){return a=e(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a})},n.prototype.scopedRun=function(t,e,i){t();try{var r=i();return e(),r}catch(a){throw e(),a}},n.prototype.nextTensorId=function(){return n.nextTensorId++},n.prototype.nextVariableId=function(){return n.nextVariableId++},n.prototype.clone=function(t){var e=this.makeTensorFromDataId(t.dataId,t.shape,t.dtype),i={x:t},r=function(s){return{x:function(){var o="float32",l={x:s},u={dtype:o};return z.runKernelFunc(function(c){return c.cast(s,o)},l,null,Ws,u)}}},a=[];return this.addTapeNode(this.state.activeScope.name,i,[e],r,a,{}),e},n.prototype.runKernel=function(t,e,i,r,a){var s=null,o=null;return this.runKernelFunc(s,e,o,t,i,r,a)},n.prototype.shouldCheckForMemLeaks=function(){return this.ENV.getBool("IS_TEST")},n.prototype.checkKernelForMemLeak=function(t,e,i){var r=this.backend.numDataIds(),a=0;i.forEach(function(l){a+=l.dtype==="complex64"?3:1});var s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=r-e-a-s;if(o>0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=Sc(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var N=l.getTensorsForGradient(r,e,L);if(N==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});N=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(N)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=Lc(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&pi(t[0])&&(a=t.map(function(c){return Ic(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Nm(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new Ba(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Tm(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof Ba||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a<s.length?(o=s[a],l=o,[4,o.kernelTimeMs]):[3,6];case 3:return l.kernelTimeMs=c.sent(),u=o,[4,o.extraInfo];case 4:u.extraInfo=c.sent(),c.label=5;case 5:return a++,[3,2];case 6:return[2,this.state.activeProfile]}})})},n.prototype.isTapeOn=function(){return this.state.gradientDepth>0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=Lc(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=Ur(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=kc(t),r=new Set(i.map(function(l){return l.id})),a=0;a<this.state.activeScope.track.length;a++){var s=this.state.activeScope.track[a];!s.kept&&!r.has(s.id)&&s.dispose()}var o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],i.forEach(function(l){!l.kept&&l.scopeId===o.id&&e.track(l)})},n.prototype.gradients=function(t,e,i,r){var a=this;if(r===void 0&&(r=!1),E(e.length>0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=iA(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??fA(s.shape),rA(l,o,function(c){return a.tidy(c)},mA);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h<d.length;h++){var p=d[h];p.dispose()}}),a.state.activeTape=null),{value:s,grads:u}})},n.prototype.customGrad=function(t){var e=this;return E(fi(t),function(){return"The f passed in customGrad(f) must be a function."}),function(){for(var i=[],r=0;r<arguments.length;r++)i[r]=arguments[r];E(i.every(function(o){return o instanceof Y}),function(){return"The args passed in customGrad(f)(x1, x2,...) must all be tensors"});var a,s={};return i.forEach(function(o,l){s[l]=o}),e.runKernelFunc(function(o,l){return a=t.apply(void 0,i.concat([l])),E(a.value instanceof Y,function(){return"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"}),E(fi(a.gradFunc),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."}),a.value},s,function(o,l){var u=a.gradFunc(o,l),c=Array.isArray(u)?u:[u];E(c.length===i.length,function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."}),E(c.every(function(d){return d instanceof Y}),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."});var h={};return c.forEach(function(d,p){h[p]=function(){return d}}),h})}},n.prototype.readSync=function(t){var e=this.state.tensorInfo.get(t);return e.backend.readSync(t)},n.prototype.read=function(t){var e=this.state.tensorInfo.get(t);return e.backend.read(t)},n.prototype.time=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return e=Tc(),[4,this.backend.time(t)];case 1:return i=r.sent(),i.wallMs=Tc()-e,[2,i]}})})},n.prototype.track=function(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t},Object.defineProperty(n.prototype,"registeredVariables",{get:function(){return this.state.registeredVariables},enumerable:!0,configurable:!0}),n.prototype.reset=function(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Fm;for(var t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null},n.nextTensorId=0,n.nextVariableId=0,n}();function fA(n){var t=Ac(ot(n),"float32");return z.makeTensor(t,n,"float32")}function Wm(){var n=yf();if(n._tfengine==null){var t=new vf(n);n._tfengine=new gA(t)}return E2(n._tfengine.ENV),oA(function(){return n._tfengine}),n._tfengine}var z=Wm();function mA(n,t){var e={a:n,b:t};return z.runKernelFunc(function(i,r){var a=i.add(n,t);return r([n,t]),a},e,null,Fs)}function vA(){return typeof navigator!="undefined"&&navigator!=null}function yA(){if(vA()){var n=navigator.userAgent||navigator.vendor||window.opera;return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(n)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(n.substr(0,4))}return!1}function Um(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var bA={__proto__:null,isMobile:yA,isBrowser:Um};var Qn=He();Qn.registerFlag("DEBUG",function(){return!1},function(n){n&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});Qn.registerFlag("IS_BROWSER",function(){return Um()});Qn.registerFlag("IS_NODE",function(){return typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined"});Qn.registerFlag("IS_CHROME",function(){return typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor)});Qn.registerFlag("PROD",function(){return!1});Qn.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",function(){return Qn.getBool("DEBUG")});Qn.registerFlag("DEPRECATION_WARNINGS_ENABLED",function(){return!0});Qn.registerFlag("IS_TEST",function(){return!1});function Fn(n,t){var e=n;if(Dt(n))return t==="string"?[]:[n.length];if(!Array.isArray(n))return[];for(var i=[];Array.isArray(e)||Dt(e)&&t!=="string";)i.push(e.length),e=e[0];return Array.isArray(n)&&He().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Bm(n,i,[]),i}function Bm(n,t,e){if(e=e||[],!Array.isArray(n)&&!Dt(n)){E(t.length===0,function(){return"Element arr["+e.join("][")+"] is a primitive, "+("but should be an array/TypedArray of "+t[0]+" elements")});return}E(t.length>0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r<n.length;++r)Bm(n[r],i,e.concat(r))}function zm(n,t,e,i){if(n==null)return;if(n!=="numeric"&&n!==t||n==="numeric"&&t==="string")throw new Error("Argument '"+e+"' passed to '"+i+"' must "+("be "+n+" tensor, but got "+t+" tensor"))}function O(n,t,e,i){if(i===void 0&&(i="numeric"),n instanceof Y)return zm(i,n.dtype,t,e),n;var r=zs(n);if(r!=="string"&&["bool","int32","float32"].indexOf(i)>=0&&(r=i),zm(i,r,t,e),n==null||!Dt(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=Fn(n,r);!Dt(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?_s(n,r):ji(n,[],o);return z.makeTensor(l,s,r)}function za(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var Pm="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+Pm;var r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];z.startScope(e);try{var o=i.apply(void 0,a);return o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),z.endScope(o),o}catch(l){throw z.endScope(null),l}};return Object.defineProperty(r,"name",{value:e,configurable:!0}),r}function wA(n,t){var e=O(n,"real","complex"),i=O(t,"imag","complex");Be(e.shape,i.shape,"real and imag shapes, "+e.shape+" and "+i.shape+", must match in call to tf.complex().");var r=function(s){return s.complex(e,i)},a={real:e,imag:i};return z.runKernelFunc(r,a,null,Af)}var mi=U({complex_:wA});function gi(n,t,e,i){if(i==null&&(i=zs(n)),i==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!Dt(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){Nc(t);var r=ot(t),a=ot(e);E(r===a,function(){return"Based on the provided shape, ["+t+"], the tensor should have "+(r+" values but has "+a)});for(var s=0;s<e.length;++s){var o=e[s],l=s===e.length-1?o!==ot(t.slice(s)):!0;E(e[s]===t[s]||!l,function(){return"Error creating a new Tensor. Inferred shape "+("("+e+") does not match the provided ")+("shape ("+t+"). ")})}}return!Dt(n)&&!Array.isArray(n)&&(n=[n]),t=t||e,n=i!=="string"?_s(n,i):ji(n,[],!0),z.makeTensor(n,t,i)}function vi(n,t,e){var i=Fn(n,e);return gi(n,t,i,e)}var Fc={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var qs=4;function LA(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l=this;return pe(this,function(u){switch(u.label){case 0:for(e=[],i=[],r=Array.isArray(n)?n.map(function(c){return c.name}):Object.keys(n),a=function(c){var h=r[c],d=Array.isArray(n)?n[c].tensor:n[h];if(d.dtype!=="float32"&&d.dtype!=="int32"&&d.dtype!=="bool"&&d.dtype!=="string"&&d.dtype!=="complex64")throw new Error("Unsupported dtype in weight '"+h+"': "+d.dtype);var p={name:h,shape:d.shape,dtype:d.dtype};if(d.dtype==="string"){var f=new Promise(function(m){return de(l,void 0,void 0,function(){var g,v,b,w,S,L,N;return pe(this,function(C){switch(C.label){case 0:return[4,d.bytes()];case 1:for(g=C.sent(),v=g.reduce(function(R,D){return R+D.length},0)+qs*g.length,b=new Uint8Array(v),w=0,S=0;S<g.length;S++)L=g[S],N=new Uint8Array(new Uint32Array([L.length]).buffer),b.set(N,w),w+=qs,b.set(L,w),w+=L.length;return m(b),[2]}})})});i.push(f)}else i.push(d.data());t!=null&&(p.group=t),e.push(p)},s=0;s<r.length;++s)a(s);return[4,Promise.all(i)];case 1:return o=u.sent(),[2,{data:SA(o),specs:e}]}})})}function _m(n,t){for(var e={},i,r=0,a=0,s=t;a<s.length;a++){var o=s[a],l=o.name,u=o.dtype,c=o.shape,h=ot(c),d=void 0;if("quantization"in o){var p=o.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error("Weight "+o.name+" with quantization "+p.dtype+" doesn't have corresponding metadata min and scale.")}else if(p.dtype==="float16"){if(u!=="float32")throw new Error("Weight "+o.name+" is quantized with "+p.dtype+" "+("which only supports weights of type float32 not "+u+"."))}else throw new Error("Weight "+o.name+" has unknown "+("quantization dtype "+p.dtype+". ")+"Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.");var f=Fc[p.dtype],m=n.slice(r,r+h*f),g=p.dtype==="uint8"?new Uint8Array(m):new Uint16Array(m);if(u==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){d=new Float32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=b*p.scale+p.min}}else if(p.dtype==="float16")i===void 0&&(i=IA()),d=i(g);else throw new Error("Unsupported quantization type "+p.dtype+" for weight type float32.");else if(u==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error("Unsupported quantization type "+p.dtype+" for weight type int32.");d=new Int32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=Math.round(b*p.scale+p.min)}}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*f}else if(u==="string"){var w=ot(o.shape);d=[];for(var v=0;v<w;v++){var S=new Uint32Array(n.slice(r,r+qs))[0];r+=qs;var L=new Uint8Array(n.slice(r,r+S));d.push(L),r+=S}}else{var N=Fc[u],m=n.slice(r,r+h*N);if(u==="float32")d=new Float32Array(m);else if(u==="int32")d=new Int32Array(m);else if(u==="bool")d=new Uint8Array(m);else if(u==="complex64"){d=new Float32Array(m);for(var C=new Float32Array(d.length/2),R=new Float32Array(d.length/2),v=0;v<C.length;v++)C[v]=d[v*2],R[v]=d[v*2+1];var D=vi(C,c,"float32"),k=vi(R,c,"float32");e[l]=mi(D,k)}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*N}u!=="complex64"&&(e[l]=vi(d,c,u))}return e}function SA(n){if(n===null)throw new Error("Invalid input value: "+JSON.stringify(n));var t=0,e=[];n.forEach(function(a){if(t+=a.byteLength,e.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error("Unsupported TypedArray subtype: "+a.constructor.name)});var i=new Uint8Array(t),r=0;return e.forEach(function(a){i.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),i.buffer}var Wc=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function Mm(n){return Wc?Buffer.byteLength(n):new Blob([n]).size}function AA(n){if(Wc)return Buffer.from(n).toString("base64");for(var t=new Uint8Array(n),e="",i=0,r=t.length;i<r;i++)e+=String.fromCharCode(t[i]);return btoa(e)}function TA(n){if(Wc){var t=Buffer.from(n,"base64");return t.buffer.slice(t.byteOffset,t.byteOffset+t.byteLength)}for(var e=atob(n),i=new Uint8Array(e.length),r=0;r<e.length;++r)i.set([e.charCodeAt(r)],r);return i.buffer}function Uc(n){if(n.length===1)return n[0];var t=0;n.forEach(function(r){t+=r.byteLength});var e=new Uint8Array(t),i=0;return n.forEach(function(r){e.set(new Uint8Array(r),i),i+=r.byteLength}),e.buffer}function Hm(n){var t="/";for(n=n.trim();n.endsWith(t);)n=n.slice(0,n.length-1);var e=n.split(t);return e[e.length-1]}function Pa(n){if(n.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:n.modelTopology==null?0:Mm(JSON.stringify(n.modelTopology)),weightSpecsBytes:n.weightSpecs==null?0:Mm(JSON.stringify(n.weightSpecs)),weightDataBytes:n.weightData==null?0:n.weightData.byteLength}}function NA(){var n=function(i){for(var r=i<<13,a=0;(r&8388608)===0;)a-=8388608,r<<=1;return r&=~8388608,a+=947912704,r|a},t=new Uint32Array(2048);t[0]=0;for(var e=1;e<1024;e++)t[e]=n(e);for(var e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function xA(){var n=new Uint32Array(64);n[0]=0,n[31]=1199570944,n[32]=2147483648,n[63]=3347054592;for(var t=1;t<31;t++)n[t]=t<<23;for(var t=33;t<63;t++)n[t]=2147483648+(t-32<<23);return n}function CA(){for(var n=new Uint32Array(64),t=0;t<64;t++)n[t]=1024;return n[0]=n[32]=0,n}function IA(){var n=NA(),t=xA(),e=CA();return function(i){for(var r=new ArrayBuffer(4*i.length),a=new Uint32Array(r),s=0;s<i.length;s++){var o=i[s],l=n[e[o>>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var nn=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),RA=function(n){return nn.registerSaveRouter(n)},OA=function(n){return nn.registerLoadRouter(n)},EA=function(n){return nn.getSaveHandlers(n)},DA=function(n,t){return nn.getLoadHandlers(n,t)};var Bc="tensorflowjs",zc=1,Ki="models_store",yi="model_info_store";function Vm(){if(!He().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function Pc(n){var t=n.result;t.createObjectStore(Ki,{keyPath:"modelPath"}),t.createObjectStore(yi,{keyPath:"modelPath"})}var zr=function(){function n(t){if(this.indexedDB=Vm(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(Bc,zc);s.onupgradeneeded=function(){return Pc(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(Ki,"readonly"),u=l.objectStore(Ki),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=Pa(e),d=o.transaction(yi,"readwrite"),p=d.objectStore(yi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(Ki,"readwrite");var g=m.objectStore(Ki),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(yi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),qm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(zr.URL_SCHEME))?kA(n.slice(zr.URL_SCHEME.length)):null};nn.registerSaveRouter(qm);nn.registerLoadRouter(qm);function kA(n){return new zr(n)}function FA(n){return n.startsWith(zr.URL_SCHEME)?n.slice(zr.URL_SCHEME.length):n}var WA=function(){function n(){this.indexedDB=Vm()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(Bc,zc);a.onupgradeneeded=function(){return Pc(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(yi,"readonly"),l=o.objectStore(yi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h<d.length;h++){var p=d[h];c[p.modelPath]=p.modelArtifactsInfo}i(c)},u.onerror=function(c){return s.close(),r(u.error)},o.oncomplete=function(){return s.close()}},a.onerror=function(s){return r(a.error)}})]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e=this;return pe(this,function(i){return t=FA(t),[2,new Promise(function(r,a){var s=e.indexedDB.open(Bc,zc);s.onupgradeneeded=function(){return Pc(s)},s.onsuccess=function(){var o=s.result,l=o.transaction(yi,"readwrite"),u=l.objectStore(yi),c=u.get(t),h;c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+t+"' in IndexedDB."));var d=u.delete(t),p=function(){h=o.transaction(Ki,"readwrite");var f=h.objectStore(Ki),m=f.delete(t);m.onsuccess=function(){return r(c.result.modelArtifactsInfo)},m.onerror=function(g){return a(c.error)}};d.onsuccess=p,d.onerror=function(f){return p(),o.close(),a(c.error)}},c.onerror=function(d){return o.close(),a(c.error)},l.oncomplete=function(){h==null?o.close():h.oncomplete=function(){return o.close()}}},s.onerror=function(o){return a(s.error)}})]})})},n}();var ei="/",Pr="tensorflowjs_models",Gm="info",UA="model_topology",BA="weight_specs",zA="weight_data",PA="model_metadata";function Ym(n){return{info:[Pr,n,Gm].join(ei),topology:[Pr,n,UA].join(ei),weightSpecs:[Pr,n,BA].join(ei),weightData:[Pr,n,zA].join(ei),modelMetadata:[Pr,n,PA].join(ei)}}function _A(n){var t=n.split(ei);if(t.length<3)throw new Error("Invalid key format: "+n);return t.slice(1,t.length-1).join(ei)}function MA(n){return n.startsWith(_r.URL_SCHEME)?n.slice(_r.URL_SCHEME.length):n}var _r=function(){function n(t){if(!He().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=Ym(this.modelPath)}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r;return pe(this,function(a){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");e=JSON.stringify(t.modelTopology),i=JSON.stringify(t.weightSpecs),r=Pa(t);try{return this.LS.setItem(this.keys.info,JSON.stringify(r)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,i),this.LS.setItem(this.keys.weightData,AA(t.weightData)),this.LS.setItem(this.keys.modelMetadata,JSON.stringify({format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata})),[2,{modelArtifactsInfo:r}]}catch(s){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error("Failed to save model '"+this.modelPath+"' to local storage: size quota being exceeded is a possible cause of this failure: "+("modelTopologyBytes="+r.modelTopologyBytes+", ")+("weightSpecsBytes="+r.weightSpecsBytes+", ")+("weightDataBytes="+r.weightDataBytes+"."))}return[2]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o;return pe(this,function(l){if(t=JSON.parse(this.LS.getItem(this.keys.info)),t==null)throw new Error("In local storage, there is no model with name '"+this.modelPath+"'");if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");if(e={},i=JSON.parse(this.LS.getItem(this.keys.topology)),i==null)throw new Error("In local storage, the topology of model '"+this.modelPath+"' is missing.");if(e.modelTopology=i,r=JSON.parse(this.LS.getItem(this.keys.weightSpecs)),r==null)throw new Error("In local storage, the weight specs of model '"+this.modelPath+"' are missing.");if(e.weightSpecs=r,a=this.LS.getItem(this.keys.modelMetadata),a!=null&&(s=JSON.parse(a),e.format=s.format,e.generatedBy=s.generatedBy,e.convertedBy=s.convertedBy,e.userDefinedMetadata=s.userDefinedMetadata),o=this.LS.getItem(this.keys.weightData),o==null)throw new Error("In local storage, the binary weight values of model "+("'"+this.modelPath+"' are missing."));return e.weightData=TA(o),[2,e]})})},n.URL_SCHEME="localstorage://",n}(),jm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(_r.URL_SCHEME))?HA(n.slice(_r.URL_SCHEME.length)):null};nn.registerSaveRouter(jm);nn.registerLoadRouter(jm);function HA(n){return new _r(n)}var VA=function(){function n(){E(He().getBool("IS_BROWSER"),function(){return"Current environment is not a web browser"}),E(typeof window=="undefined"||typeof window.localStorage!="undefined",function(){return"Current browser does not appear to support localStorage"}),this.LS=window.localStorage}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s;return pe(this,function(o){for(t={},e=Pr+ei,i=ei+Gm,r=0;r<this.LS.length;++r)a=this.LS.key(r),a.startsWith(e)&&a.endsWith(i)&&(s=_A(a),t[s]=JSON.parse(this.LS.getItem(a)));return[2,t]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){if(t=MA(t),e=Ym(t),this.LS.getItem(e.info)==null)throw new Error("Cannot find model at path '"+t+"'");return i=JSON.parse(this.LS.getItem(e.info)),this.LS.removeItem(e.info),this.LS.removeItem(e.topology),this.LS.removeItem(e.weightSpecs),this.LS.removeItem(e.weightData),[2,i]})})},n}();var Mr="://",bi=function(){function n(){this.managers={}}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerManager=function(t,e){E(t!=null,function(){return"scheme must not be undefined or null."}),t.endsWith(Mr)&&(t=t.slice(0,t.indexOf(Mr))),E(t.length>0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function Gs(n){if(n.indexOf(Mr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+bi.getSchemes().join(",")));return{scheme:n.split(Mr)[0],path:n.split(Mr)[1]}}function Km(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=nn.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=nn.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=Gs(n).scheme,l=Gs(n).path,u=o===Gs(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,bi.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,bi.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function qA(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=bi.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e<i.length?(r=i[e],[4,bi.getManager(r).listModels()]):[3,4];case 2:a=l.sent();for(s in a)o=r+Mr+s,t[o]=a[s];l.label=3;case 3:return e++,[3,1];case 4:return[2,t]}})})}function GA(n){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){return t=Gs(n),e=bi.getManager(t.scheme),[2,e.removeModel(t.path)]})})}function YA(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!1,[2,Km(n,t,e)]})})}function jA(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!0,[2,Km(n,t,e)]})})}var KA=function(){function n(){}return n.prototype.fetch=function(t,e){return fetch(t,e)},n.prototype.now=function(){return performance.now()},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Browser's encoder only supports utf-8, but got "+e);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return new TextDecoder(e).decode(t)},n}();if(He().get("IS_BROWSER")){He().setPlatform("browser",new KA);try{bi.registerManager(_r.URL_SCHEME,new VA)}catch(n){}try{bi.registerManager(zr.URL_SCHEME,new WA)}catch(n){}}var $A={importFetch:function(){return ff()}},_c,XA=function(){function n(){this.util=require("util"),this.textEncoder=new this.util.TextEncoder}return n.prototype.fetch=function(t,e){return He().global.fetch!=null?He().global.fetch(t,e):(_c==null&&(_c=$A.importFetch()),_c(t,e))},n.prototype.now=function(){var t=process.hrtime();return t[0]*1e3+t[1]/1e6},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Node built-in encoder only supports utf-8, but got "+e);return this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)},n}();He().get("IS_NODE")&&He().setPlatform("node",new XA);function Wn(n,t,e){return t===void 0&&(t="float32"),t=t||"float32",Nc(n),new Hs(n,t,e)}function JA(n,t){var e=O(n,"x","cast");if(!Im(t))throw new Error("Failed to cast to unknown dtype "+t);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");var i={x:e},r={dtype:t};return z.runKernelFunc(function(a){return a.cast(e,t)},i,null,Ws,r)}var ue=U({cast_:JA});function ZA(n){var t=O(n,"x","clone",null),e=function(){return z.makeTensorFromDataId(t.dataId,t.shape,t.dtype)},i={x:t};return z.runKernelFunc(e,i,null,gu)}var $i=U({clone_:ZA});function $m(n,t){t===void 0&&(t=!1),console.log(n.toString(t))}Wm();var QA={buffer:Wn,cast:ue,clone:$i,print:$m};lA(QA);var eT="model",tT=".json",nT=".weights.bin";function Xm(n){return new Promise(function(t){return setTimeout(t)}).then(n)}var Mc=function(){function n(t){if(!He().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(n.URL_SCHEME)&&(t=t.slice(n.URL_SCHEME.length)),(t==null||t.length===0)&&(t=eT),this.modelTopologyFileName=t+tT,this.weightDataFileName=t+nT}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");if(e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"})),!(t.modelTopology instanceof ArrayBuffer))return[3,1];throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");case 1:return i=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:i},a=window.URL.createObjectURL(new Blob([JSON.stringify(r)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor,s.download=this.modelTopologyFileName,s.href=a,[4,Xm(function(){return s.dispatchEvent(new MouseEvent("click"))})];case 2:return l.sent(),t.weightData!=null?(o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor,o.download=this.weightDataFileName,o.href=e,[4,Xm(function(){return o.dispatchEvent(new MouseEvent("click"))})]):[3,4];case 3:l.sent(),l.label=4;case 4:return[2,{modelArtifactsInfo:Pa(t)}]}})})},n.URL_SCHEME="downloads://",n}(),iT=function(){function n(t){if(t==null||t.length<1)throw new Error("When calling browserFiles, at least 1 file is required, "+("but received "+t));this.files=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i=this;return pe(this,function(r){return t=this.files[0],e=this.files.slice(1),[2,new Promise(function(a,s){var o=new FileReader;o.onload=function(l){var u=JSON.parse(l.target.result),c=u.modelTopology;if(c==null){s(new Error("modelTopology field is missing from file "+t.name));return}e.length===0&&a({modelTopology:c});var h=u.weightsManifest;if(h==null){s(new Error("weightManifest field is missing from file "+t.name));return}var d;try{d=i.checkManifestAndWeightFiles(h,e)}catch(g){s(g);return}var p=[],f=[],m=[];h.forEach(function(g){g.paths.forEach(function(v){f.push(v),m.push(null)}),p.push.apply(p,g.weights)}),h.forEach(function(g){g.paths.forEach(function(v){var b=new FileReader;b.onload=function(w){var S=w.target.result,L=f.indexOf(v);m[L]=S,m.indexOf(null)===-1&&a({modelTopology:c,weightSpecs:p,weightData:Uc(m),format:u.format,generatedBy:u.generatedBy,convertedBy:u.convertedBy,userDefinedMetadata:u.userDefinedMetadata})},b.onerror=function(w){return s("Failed to weights data from file of path '"+v+"'.")},b.readAsArrayBuffer(d[v])})})},o.onerror=function(l){return s("Failed to read model topology and weights manifest JSON "+("from file '"+t.name+"'. BrowserFiles supports loading ")+"Keras-style tf.Model artifacts only.")},o.readAsText(t)})]})})},n.prototype.checkManifestAndWeightFiles=function(t,e){for(var i=[],r=e.map(function(u){return Hm(u.name)}),a={},s=0,o=t;s<o.length;s++){var l=o[s];l.paths.forEach(function(u){var c=Hm(u);if(i.indexOf(c)!==-1)throw new Error("Duplicate file basename found in weights manifest: "+("'"+c+"'"));if(i.push(c),r.indexOf(c)===-1)throw new Error("Weight file with basename '"+c+"' is not provided.");a[u]=e[r.indexOf(c)]})}if(i.length!==e.length)throw new Error("Mismatch in the number of files in weights manifest "+("("+i.length+") and the number of weight files provided ")+("("+e.length+")."));return a},n}(),aT=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Mc.URL_SCHEME))?rT(n.slice(Mc.URL_SCHEME.length)):null};nn.registerSaveRouter(aT);function rT(n){return n===void 0&&(n="model"),new Mc(n)}function sT(n){return new iT(n)}function Jm(n,t,e,i){s(n),e=e??0,i=i??1,o(e,i);var r=0,a=function(l){return l.then(function(u){var c=e+ ++r/n.length*(i-e);return t(c),u}),l};function s(l){E(l!=null&&Array.isArray(l)&&l.length>0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function Zm(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?He().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,Jm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,Jm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function oT(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return Zm(o,{requestInit:i})},a=Qm(r),[2,a(n,t,e)]})})}function Qm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=Fc[S]*ot(w.shape),N=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(N(),o[R]=!0)}):N(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. 
-`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w<v;w++)b+=d[f+w].byteLength;for(var S=new ArrayBuffer(b),L=new Uint8Array(S),N=0,C=0;C<v;C++){var R=new Uint8Array(d[f+C]);L.set(R,N),N+=R.byteLength}var D=s[g];D.forEach(function(k){var W=S.slice(k.groupOffset,k.groupOffset+k.sizeBytes),F=_m(W,[k.manifestEntry]);for(var P in F)p[P]=F[P]}),f+=v}),[2,p]}})})}}var lT="application/octet-stream",uT="application/json",eg=function(){function n(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(E(typeof e.fetchFunc=="function",function(){return"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"}),this.fetch=e.fetchFunc):this.fetch=He().platform.fetch,E(t!=null&&t.length>0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:uT}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:lT}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:Pa(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=cT(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;l<u.length;l++)c=u[l],o.push.apply(o,c.weights);for(h=[],d=[],p=0,f=t;p<f.length;p++)for(m=f[p],g=0,v=m.paths;g<v.length;g++)b=v[g],this.weightUrlConverter!=null?d.push(this.weightUrlConverter(b)):h.push(s+b+a);return this.weightUrlConverter?(S=(w=h.push).apply,L=[h],[4,Promise.all(d)]):[3,2];case 1:S.apply(w,L.concat([C.sent()])),C.label=2;case 2:return[4,Zm(h,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress})];case 3:return N=C.sent(),[2,[o,Uc(N)]]}})})},n.URL_SCHEME_REGEX=/^https?:\/\//,n}();function cT(n){var t=n.lastIndexOf("/"),e=n.lastIndexOf("?"),i=n.substring(0,t),r=e>t?n.substring(e):"";return[i+"/",r]}function Hc(n){return n.match(eg.URL_SCHEME_REGEX)!=null}var tg=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return Hc(i)}):e=Hc(n),e?Vc(n,t):null};nn.registerSaveRouter(tg);nn.registerLoadRouter(tg);function Vc(n,t){return new eg(n,t)}function hT(n,t){return Vc(n,t)}var qc=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),dT=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function pT(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new qc(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new qc({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new qc({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function fT(n){return new dT(n)}var mT={__proto__:null,browserFiles:sT,browserHTTPRequest:hT,concatenateArrayBuffers:Uc,decodeWeights:_m,encodeWeights:LA,fromMemory:pT,getLoadHandlers:DA,getModelArtifactsInfoForJSON:Pa,getSaveHandlers:EA,http:Vc,isHTTPScheme:Hc,loadWeights:oT,registerLoadRouter:OA,registerSaveRouter:RA,weightsLoaderFactory:Qm,withSaveHandler:fT,copyModel:YA,listModels:qA,moveModel:jA,removeModel:GA};function gT(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=bm(t,e.size),E(e.size===ot(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,_u,r)}var V=U({reshape_:gT});function vT(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=at(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=ot(h),f=ot(d);E(dn(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(N,C){return C([g,v]),N.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,Yl,S);return V(L,m)}var We=U({matMul_:vT});function yT(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,Fu,l)}var Ys=U({oneHot_:yT});function bT(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a<e.rank,function(){return"All entries in 'perm' must be between 0 and "+(e.rank-1)+(" but got "+t)})}),e.rank<=1)return e.clone();var i={x:e},r={perm:t};return z.runKernelFunc(function(a){return a.transpose(e,t)},i,null,hc,r)}var ut=U({transpose_:bT});function wT(n,t,e){var i=O(n,"labels","confusionMatrix"),r=O(t,"predictions","confusionMatrix");E(e==null||e>0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=Ys(ue(i,"int32"),e),s=Ys(ue(r,"int32"),e),o=ut(a);return ue(We(o,s),"int32")}var ST=U({confusionMatrix_:wT});var LT={__proto__:null,confusionMatrix:ST};function ng(n,t,e){if(Yi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=Fn(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return gi(n,t,i,e)}var Hr;function IT(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState<o)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}var l=Sc(gc,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(gc,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Hr==null&&(Hr=document.createElement("canvas").getContext("2d")),Hr.canvas.width=d,Hr.canvas.height=p,Hr.drawImage(n,0,0,d,p),f=Hr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v<g;v++)for(var b=0;b<t;++b)m[v*t+b]=f[v*4+b]}var w=[p,d,t];return ng(m,w,"int32")}function AT(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v;return pe(this,function(b){switch(b.label){case 0:if(e=O(n,"img","toPixels"),n instanceof Y||(i=e,e=ue(i,"int32"),i.dispose()),e.rank!==2&&e.rank!==3)throw new Error("toPixels only supports rank 2 or 3 tensors, got rank "+e.rank+".");if(r=e.shape.slice(0,2),a=r[0],s=r[1],o=e.rank===2?1:e.shape[2],o>4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h<a*s;++h){for(d=[0,0,0,255],p=0;p<o;p++){if(f=l[h*o+p],e.dtype==="float32"){if(f<0||f>1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var TT=U({fromPixels_:IT}),NT={__proto__:null,toPixels:AT,fromPixels:TT};function ig(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a<e.length-1;++a)r*=e[a];var s=n.shape,o=e.slice();o.pop();for(var l=1,a=i;a<n.rank;++a)l*=s[a],o.push(s[a]);var u=Fr(n.shape).map(function(c){return c/l}).concat([1]).slice(0,i);return[o,r,l,u]}var xT={__proto__:null,prepareAndValidate:ig};function Gc(n,t,e){var i=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank<r)throw new Error(a+(" update.rank < "+r+". "));if(n.length<i+(e.rank-r))throw new Error(a+(" Output shape length < "+(i+(e.rank-r))));if(e.rank!==r+n.length-i)throw new Error(a+(" update.rank != "+(r+n.length-i)));for(var s=0;s<r;++s)if(e.shape[s]!==t.shape[s])throw new Error(a+(" updates.shape["+s+"] ("+e.shape[s]+") != indices.shape["+s+"] ("+t.shape[s]+")."));for(var s=0;s<e.rank-r;++s)if(e.shape[s+r]!==n[s+i])throw new Error(a+(" updates.shape["+(s+r)+"] ("+e.shape[s+r]+") != shape["+(s+r)+"] ("+n[s+r]+")"))}function Yc(n,t,e){if(t.rank<1)throw new Error("tf.scatterND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(n.rank<1)throw new Error("tf.scatterND() expects the updates to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.dtype!=="int32")throw new Error("The dtype of 'indices' should be int32, but got dtype: "+t.dtype);if(e.length<1)throw new Error("Output rank must be greater or equal to 1, but got shape: "+e);if(e.length===0){if(t.size===0)throw new Error("Indices specified for empty output. indices shape: "+t.shape);if(n.size===0)throw new Error("Updates specified for empty output. updates shape: "+n.shape)}Gc(e,t,n)}function rg(n,t,e){for(var i=t.shape.length,r=i>1?t.shape[i-1]:1,a=e.length,s=1,o=r;o<a;++o)s*=e[o];var l=r<1?1:r,u=ot(t.shape)/l,c=Fr(e.slice(0,r)).concat([1]),h=ot(e);return{sliceRank:r,numUpdates:u,sliceSize:s,strides:c,outputSize:h}}var CT={__proto__:null,validateUpdateShape:Gc,validateInput:Yc,calculateShapes:rg};function ag(n,t,e){var i=n.shape.length;E(i===t.length,function(){return"Error in slice"+i+"D: Length of begin "+t+" must "+("match the rank of the array ("+i+").")}),E(i===e.length,function(){return"Error in slice"+i+"D: Length of size "+e+" must "+("match the rank of the array ("+i+").")});for(var r=function(s){E(t[s]+e[s]<=n.shape[s],function(){return"Error in slice"+i+"D: begin["+s+"] + size["+s+"] "+("("+(t[s]+e[s])+") would overflow input.shape["+s+"] ("+n.shape[s]+")")})},a=0;a<i;++a)r(a)}function js(n){for(var t=[],e=0;n>0;)n&1&&t.push(e),n/=2,e++;return t}function sg(n,t,e){for(var i=[],r=0;r<n.length;r++)i[r]=Math.ceil((t[r]-n[r])/e[r]);return i}function og(n,t,e,i){for(var r=n.slice(),a=r.length;a<i.length;a++)r.push(1);for(var a=0;a<e;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function lg(n,t,e){return e<=n?e:e-(t-1)}function ug(n,t){for(var e=[],i=0;i<n;i++)e.push(t+i);return e}function mg(n,t,e,i,r,a,s,o,l){var u=n.length,c=new Array(u),h=new Array(u),d=new Array(u);if(t.length&&e>0){var p=t[0],f=e+1;c=cg(s,p,f,i,n),h=hg(o,p,f,r,n),d=og(a,p,f,n)}else for(var m=0;m<u;m++)c[m]=pg(s,i,a,n,m,l),h[m]=fg(o,r,a,n,m,l),d[m]=dg(a,m,l);return{begin:c,end:h,strides:d}}function cg(n,t,e,i,r){for(var a=r.slice(),s=ug(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=0;else{var l=lg(t,e,o),u=i[l];n&1<<l&&(u=0),a[o]=u}return a}function hg(n,t,e,i,r){for(var a=r.slice(),s=ug(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=lg(t,e,o),u=i[l];n&1<<l&&(u=Number.MAX_SAFE_INTEGER),a[o]=u}for(var c=0;c<a.length;c++){var h=r[c];a[c]<0&&(a[c]+=h),a[c]=Da(0,a[c],r[c])}return a}function dg(n,t,e){var i=n[t];return(e&1<<t||i==null)&&(i=1),i}function pg(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=Da(0,s,l-1),s}function fg(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=Da(0,s,l):s=Da(-1,s,l-1),s}function RT(n,t,e){for(var i=e.length,r=0;r<e.length;r++)if(e[r]>1){i=r;break}for(var r=i+1;r<e.length;r++)if(t[r]>0||e[r]!==n[r])return!1;return!0}function OT(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i<n.length-1;i++)e+=n[i]*t[i];return e}function jc(n,t,e){var i,r=n.shape.length;typeof t=="number"?i=[t].concat(new Array(r-1).fill(0)):t.length<r?i=t.concat(new Array(r-t.length).fill(0)):i=t.slice(),i.forEach(function(s){E(s!==-1,function(){return"slice() does not support negative begin indexing."})});var a;return e==null?a=new Array(r).fill(-1):typeof e=="number"?a=[e].concat(new Array(r-1).fill(-1)):e.length<r?a=e.concat(new Array(r-e.length).fill(-1)):a=e,a=a.map(function(s,o){return s>=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var gg={__proto__:null,assertParamsValid:ag,maskToAxes:js,computeOutShape:sg,stridesWithElidedDims:og,getNormalizedAxes:mg,startIndicesWithElidedDims:cg,stopIndicesWithElidedDims:hg,stridesForAxis:dg,startForAxis:pg,stopForAxis:fg,isSliceContinous:RT,computeFlatOffset:OT,parseSliceParams:jc};var vg=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),yg=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function wi(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),yg.register(n)}var ET={__proto__:null,Serializable:vg,SerializationMap:yg,registerClass:wi};var DT=.001,bg=.1;function kT(n,t,e){return e==null&&(e=Kc()),$c(n,t,function(i,r){return Xc(i,r,e)})}function Kc(){return z.backend.floatPrecision()===32?DT:bg}function $c(n,t,e){var i=!0;if((Dt(n)||Dt(t))&&(i=!1),Dt(n)&&Dt(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=Fn(n),o=Fn(t);if(!dn(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Dt(n)?n:ji(n),u=Dt(t)?t:ji(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`.
+`;return v[v.length-1]=" "+v[v.length-1]+"]"+(a?"":N),v}function Ua(n){for(var t=[],e=0;e<n.length;e+=2)t.push([n[e],n[e+1]]);return t}var Hs=function(){function n(t,e,i){var r=this;if(this.dtype=e,this.shape=t.slice(),this.size=ot(t),i!=null){var a=i.length;E(a===this.size,function(){return"Length of values '"+a+"' does not match the size "+("inferred by the shape '"+r.size+"'.")})}if(e==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=i||Sm(e,this.size),this.strides=Fr(t)}return n.prototype.set=function(t){for(var e=this,i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];i.length===0&&(i=[0]),E(i.length===this.rank,function(){return"The number of provided coordinates ("+i.length+") must "+("match the rank ("+e.rank+")")});var a=this.locToIndex(i);this.values[a]=t},n.prototype.get=function(){for(var t=[],e=0;e<arguments.length;e++)t[e]=arguments[e];t.length===0&&(t=[0]);for(var i=0,r=0,a=t;r<a.length;r++){var s=a[r];if(s<0||s>=this.shape[i]){var o="Requested out of range element at "+t+". "+("  Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u<t.length-1;++u)l+=this.strides[u]*t[u];return this.values[l]},n.prototype.locToIndex=function(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];for(var e=t[t.length-1],i=0;i<t.length-1;++i)e+=this.strides[i]*t[i];return e},n.prototype.indexToLoc=function(t){if(this.rank===0)return[];if(this.rank===1)return[t];for(var e=new Array(this.shape.length),i=0;i<e.length-1;++i)e[i]=Math.floor(t/this.strides[i]),t-=e[i]*this.strides[i];return e[e.length-1]=t,e},Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.toTensor=function(){return kn().makeTensor(this.values,this.shape,this.dtype)},n}(),kn=null,Br=null;function oA(n){kn=n}function lA(n){Br=n}var Y=function(){function n(t,e,i,r){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=ot(t),this.strides=Fr(t),this.dataId=i,this.id=r,this.rankType=this.rank<5?this.rank.toString():"higher"}return Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.buffer=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Br.buffer(this.shape,this.dtype,t)]}})})},n.prototype.bufferSync=function(){return Br.buffer(this.shape,this.dtype,this.dataSync())},n.prototype.array=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Wr(this.shape,t)]}})})},n.prototype.arraySync=function(){return Wr(this.shape,this.dataSync())},n.prototype.data=function(){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){switch(i.label){case 0:return this.throwIfDisposed(),t=kn().read(this.dataId),this.dtype==="string"?[4,t]:[3,2];case 1:e=i.sent();try{return[2,e.map(function(r){return Cc(r)})]}catch(r){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}i.label=2;case 2:return[2,t]}})})},n.prototype.dataSync=function(){this.throwIfDisposed();var t=kn().readSync(this.dataId);if(this.dtype==="string")try{return t.map(function(e){return Cc(e)})}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t},n.prototype.bytes=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return this.throwIfDisposed(),[4,kn().read(this.dataId)];case 1:return t=e.sent(),this.dtype==="string"?[2,t]:[2,new Uint8Array(t.buffer)]}})})},n.prototype.dispose=function(){if(this.isDisposed)return;kn().disposeTensor(this),this.isDisposedInternal=!0},Object.defineProperty(n.prototype,"isDisposed",{get:function(){return this.isDisposedInternal},enumerable:!0,configurable:!0}),n.prototype.throwIfDisposed=function(){if(this.isDisposed)throw new Error("Tensor is disposed.")},n.prototype.print=function(t){return t===void 0&&(t=!1),Br.print(this,t)},n.prototype.clone=function(){return this.throwIfDisposed(),Br.clone(this)},n.prototype.toString=function(t){t===void 0&&(t=!1);var e=this.dataSync();return sA(e,this.shape,this.dtype,t)},n.prototype.cast=function(t){return this.throwIfDisposed(),Br.cast(this,t)},n.prototype.variable=function(t,e,i){return t===void 0&&(t=!0),this.throwIfDisposed(),kn().makeVariable(this,t,e,i)},n}();Object.defineProperty(Y,Symbol.hasInstance,{value:function(n){return!!n&&n.data!=null&&n.dataSync!=null&&n.throwIfDisposed!=null}});var Ba=function(n){Zn(t,n);function t(e,i,r,a){var s=n.call(this,e.shape,e.dtype,e.dataId,a)||this;return s.trainable=i,s.name=r,s}return t.prototype.assign=function(e){if(e.dtype!==this.dtype)throw new Error("dtype of the new value ("+e.dtype+") and "+("previous value ("+this.dtype+") must match"));if(!hn(e.shape,this.shape))throw new Error("shape of the new value ("+e.shape+") and "+("previous value ("+this.shape+") must match"));kn().disposeTensor(this),this.dataId=e.dataId,kn().incRef(this,null)},t.prototype.dispose=function(){kn().disposeVariable(this),this.isDisposedInternal=!0},t}(Y);Object.defineProperty(Ba,Symbol.hasInstance,{value:function(n){return n instanceof Y&&n.assign!=null&&n.assign instanceof Function}});(function(n){n.R0="R0",n.R1="R1",n.R2="R2",n.R3="R3",n.R4="R4",n.R5="R5",n.R6="R6"})(I.Rank||(I.Rank={}));var Oc;(function(n){n.float32="float32",n.int32="int32",n.bool="int32",n.complex64="complex64"})(Oc||(Oc={}));var Ec;(function(n){n.float32="float32",n.int32="int32",n.bool="bool",n.complex64="complex64"})(Ec||(Ec={}));var Dc;(function(n){n.float32="float32",n.int32="float32",n.bool="float32",n.complex64="complex64"})(Dc||(Dc={}));var kc;(function(n){n.float32="complex64",n.int32="complex64",n.bool="complex64",n.complex64="complex64"})(kc||(kc={}));var uA={float32:Dc,int32:Oc,bool:Ec,complex64:kc};function Vs(n,t){if(n==="string"||t==="string"){if(n==="string"&&t==="string")return"string";throw new Error("Can not upcast "+n+" with "+t)}return uA[n][t]}function cA(n){return Vs(n,"int32")}function at(n,t){if(n.dtype===t.dtype)return[n,t];var e=Vs(n.dtype,t.dtype);return[n.cast(e),t.cast(e)]}function Dm(n,t){E(n.dtype===t.dtype,function(){return"The dtypes of the first("+n.dtype+") and"+(" second("+t.dtype+") input must match")})}function hA(n,t){return t.some(function(e){return e.id===n.id})}function Fc(n){var t=[],e=new Set;return km(n,t,e),t}function km(n,t,e){if(n==null)return;if(n instanceof Y){t.push(n);return}if(!dA(n))return;var i=n;for(var r in i){var a=i[r];e.has(a)||(e.add(a),km(a,t,e))}}function dA(n){return Array.isArray(n)||typeof n=="object"}var pA={__proto__:null,makeTypesMatch:at,assertTypesMatch:Dm,isTensorInList:hA,getTensorsInContainer:Fc};var Fm=function(){function n(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null}}return n.prototype.dispose=function(){for(var t in this.registeredVariables)this.registeredVariables[t].dispose()},n}(),gA=function(){function n(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Fm}return n.prototype.ready=function(){return de(this,void 0,void 0,function(){var t,e,i,r;return pe(this,function(a){switch(a.label){case 0:if(this.pendingBackendInit!=null)return[2,this.pendingBackendInit.then(function(){})];if(this.backendInstance!=null)return[2];t=this.getSortedBackends(),e=0,a.label=1;case 1:return e<t.length?(i=t[e],[4,this.initializeBackend(i).success]):[3,5];case 2:return r=a.sent(),r?[4,this.setBackend(i)]:[3,4];case 3:return a.sent(),[2];case 4:return e++,[3,1];case 5:throw new Error("Could not initialize any backends, all backend initializations failed.")}})})},Object.defineProperty(n.prototype,"backend",{get:function(){if(this.pendingBackendInit!=null)throw new Error("Backend '"+this.backendName+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");if(this.backendInstance==null){var t=this.initializeBackendsAndReturnBest(),e=t.name,i=t.asyncInit;if(i)throw new Error("The highest priority backend '"+e+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");this.setBackend(e)}return this.backendInstance},enumerable:!0,configurable:!0}),n.prototype.backendNames=function(){return Object.keys(this.registryFactory)},n.prototype.findBackend=function(t){if(!(t in this.registry))if(t in this.registryFactory){var e=this.initializeBackend(t).asyncInit;if(e)return null}else return null;return this.registry[t]},n.prototype.findBackendFactory=function(t){return t in this.registryFactory?this.registryFactory[t].factory:null},n.prototype.registerBackend=function(t,e,i){return i===void 0&&(i=1),t in this.registryFactory?(console.warn(t+" backend was already registered. Reusing existing backend factory."),!1):(this.registryFactory[t]={factory:e,priority:i},!0)},n.prototype.setBackend=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s;return pe(this,function(o){switch(o.label){case 0:if(this.registryFactory[t]==null)throw new Error("Backend name '"+t+"' not found in registry");return this.backendName=t,this.registry[t]==null?(this.backendInstance=null,e=this.initializeBackend(t),i=e.success,r=e.asyncInit,r?[4,i]:[3,2]):[3,4];case 1:return s=o.sent(),[3,3];case 2:s=i,o.label=3;case 3:if(a=s,!a)return[2,!1];o.label=4;case 4:return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new nA(this.backendInstance),[2,!0]}})})},n.prototype.setupRegisteredKernels=function(){var t=this,e=Us(this.backendName);e.forEach(function(i){i.setupFunc!=null&&i.setupFunc(t.backendInstance)})},n.prototype.disposeRegisteredKernels=function(t){var e=this,i=Us(t);i.forEach(function(r){r.disposeFunc!=null&&r.disposeFunc(e.registry[t])})},n.prototype.initializeBackend=function(t){var e=this,i=this.registryFactory[t];if(i==null)throw new Error("Cannot initialize backend "+t+", no registration found.");try{var r=i.factory();if(r&&!(r instanceof mf)&&typeof r.then=="function"){var a=++this.pendingBackendInitId,s=r.then(function(o){return a<e.pendingBackendInitId?!1:(e.registry[t]=o,e.pendingBackendInit=null,!0)}).catch(function(o){return a<e.pendingBackendInitId||(e.pendingBackendInit=null,console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message)),!1});return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=r,{success:!0,asyncInit:!1}}catch(o){return console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message),{success:!1,asyncInit:!1}}},n.prototype.removeBackend=function(t){if(!(t in this.registryFactory))throw new Error(t+" backend not found in registry");this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)},n.prototype.getSortedBackends=function(){var t=this;if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(function(e,i){return t.registryFactory[i].priority-t.registryFactory[e].priority})},n.prototype.initializeBackendsAndReturnBest=function(){for(var t=this.getSortedBackends(),e=0;e<t.length;e++){var i=t[e],r=this.initializeBackend(i),a=r.success,s=r.asyncInit;if(s||a)return{name:i,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")},n.prototype.moveData=function(t,e){var i=this.state.tensorInfo.get(e),r=i.backend,a=this.readSync(e);r.disposeData(e),i.backend=t,t.move(e,a,i.shape,i.dtype),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++},n.prototype.tidy=function(t,e){var i=this,r=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");r=t}var a;return this.scopedRun(function(){return i.startScope(r)},function(){return i.endScope(a)},function(){return a=e(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a})},n.prototype.scopedRun=function(t,e,i){t();try{var r=i();return e(),r}catch(a){throw e(),a}},n.prototype.nextTensorId=function(){return n.nextTensorId++},n.prototype.nextVariableId=function(){return n.nextVariableId++},n.prototype.clone=function(t){var e=this.makeTensorFromDataId(t.dataId,t.shape,t.dtype),i={x:t},r=function(s){return{x:function(){var o="float32",l={x:s},u={dtype:o};return z.runKernelFunc(function(c){return c.cast(s,o)},l,null,Ws,u)}}},a=[];return this.addTapeNode(this.state.activeScope.name,i,[e],r,a,{}),e},n.prototype.runKernel=function(t,e,i,r,a){var s=null,o=null;return this.runKernelFunc(s,e,o,t,i,r,a)},n.prototype.shouldCheckForMemLeaks=function(){return this.ENV.getBool("IS_TEST")},n.prototype.checkKernelForMemLeak=function(t,e,i){var r=this.backend.numDataIds(),a=0;i.forEach(function(l){a+=l.dtype==="complex64"?3:1});var s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=r-e-a-s;if(o>0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=Lc(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var N=l.getTensorsForGradient(r,e,L);if(N==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});N=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(N)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=Ic(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&pi(t[0])&&(a=t.map(function(c){return Ac(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Nm(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new Ba(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Tm(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof Ba||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a<s.length?(o=s[a],l=o,[4,o.kernelTimeMs]):[3,6];case 3:return l.kernelTimeMs=c.sent(),u=o,[4,o.extraInfo];case 4:u.extraInfo=c.sent(),c.label=5;case 5:return a++,[3,2];case 6:return[2,this.state.activeProfile]}})})},n.prototype.isTapeOn=function(){return this.state.gradientDepth>0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=Ic(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=Ur(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=Fc(t),r=new Set(i.map(function(l){return l.id})),a=0;a<this.state.activeScope.track.length;a++){var s=this.state.activeScope.track[a];!s.kept&&!r.has(s.id)&&s.dispose()}var o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],i.forEach(function(l){!l.kept&&l.scopeId===o.id&&e.track(l)})},n.prototype.gradients=function(t,e,i,r){var a=this;if(r===void 0&&(r=!1),E(e.length>0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=iA(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??fA(s.shape),rA(l,o,function(c){return a.tidy(c)},mA);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h<d.length;h++){var p=d[h];p.dispose()}}),a.state.activeTape=null),{value:s,grads:u}})},n.prototype.customGrad=function(t){var e=this;return E(fi(t),function(){return"The f passed in customGrad(f) must be a function."}),function(){for(var i=[],r=0;r<arguments.length;r++)i[r]=arguments[r];E(i.every(function(o){return o instanceof Y}),function(){return"The args passed in customGrad(f)(x1, x2,...) must all be tensors"});var a,s={};return i.forEach(function(o,l){s[l]=o}),e.runKernelFunc(function(o,l){return a=t.apply(void 0,i.concat([l])),E(a.value instanceof Y,function(){return"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"}),E(fi(a.gradFunc),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."}),a.value},s,function(o,l){var u=a.gradFunc(o,l),c=Array.isArray(u)?u:[u];E(c.length===i.length,function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."}),E(c.every(function(d){return d instanceof Y}),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."});var h={};return c.forEach(function(d,p){h[p]=function(){return d}}),h})}},n.prototype.readSync=function(t){var e=this.state.tensorInfo.get(t);return e.backend.readSync(t)},n.prototype.read=function(t){var e=this.state.tensorInfo.get(t);return e.backend.read(t)},n.prototype.time=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return e=Nc(),[4,this.backend.time(t)];case 1:return i=r.sent(),i.wallMs=Nc()-e,[2,i]}})})},n.prototype.track=function(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t},Object.defineProperty(n.prototype,"registeredVariables",{get:function(){return this.state.registeredVariables},enumerable:!0,configurable:!0}),n.prototype.reset=function(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Fm;for(var t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null},n.nextTensorId=0,n.nextVariableId=0,n}();function fA(n){var t=Tc(ot(n),"float32");return z.makeTensor(t,n,"float32")}function Wm(){var n=yf();if(n._tfengine==null){var t=new vf(n);n._tfengine=new gA(t)}return E2(n._tfengine.ENV),oA(function(){return n._tfengine}),n._tfengine}var z=Wm();function mA(n,t){var e={a:n,b:t};return z.runKernelFunc(function(i,r){var a=i.add(n,t);return r([n,t]),a},e,null,Fs)}function vA(){return typeof navigator!="undefined"&&navigator!=null}function yA(){if(vA()){var n=navigator.userAgent||navigator.vendor||window.opera;return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(n)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(n.substr(0,4))}return!1}function Um(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var bA={__proto__:null,isMobile:yA,isBrowser:Um};var Qn=He();Qn.registerFlag("DEBUG",function(){return!1},function(n){n&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});Qn.registerFlag("IS_BROWSER",function(){return Um()});Qn.registerFlag("IS_NODE",function(){return typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined"});Qn.registerFlag("IS_CHROME",function(){return typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor)});Qn.registerFlag("PROD",function(){return!1});Qn.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",function(){return Qn.getBool("DEBUG")});Qn.registerFlag("DEPRECATION_WARNINGS_ENABLED",function(){return!0});Qn.registerFlag("IS_TEST",function(){return!1});function Fn(n,t){var e=n;if(Dt(n))return t==="string"?[]:[n.length];if(!Array.isArray(n))return[];for(var i=[];Array.isArray(e)||Dt(e)&&t!=="string";)i.push(e.length),e=e[0];return Array.isArray(n)&&He().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Bm(n,i,[]),i}function Bm(n,t,e){if(e=e||[],!Array.isArray(n)&&!Dt(n)){E(t.length===0,function(){return"Element arr["+e.join("][")+"] is a primitive, "+("but should be an array/TypedArray of "+t[0]+" elements")});return}E(t.length>0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r<n.length;++r)Bm(n[r],i,e.concat(r))}function zm(n,t,e,i){if(n==null)return;if(n!=="numeric"&&n!==t||n==="numeric"&&t==="string")throw new Error("Argument '"+e+"' passed to '"+i+"' must "+("be "+n+" tensor, but got "+t+" tensor"))}function O(n,t,e,i){if(i===void 0&&(i="numeric"),n instanceof Y)return zm(i,n.dtype,t,e),n;var r=zs(n);if(r!=="string"&&["bool","int32","float32"].indexOf(i)>=0&&(r=i),zm(i,r,t,e),n==null||!Dt(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=Fn(n,r);!Dt(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?_s(n,r):ji(n,[],o);return z.makeTensor(l,s,r)}function za(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var Pm="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+Pm;var r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];z.startScope(e);try{var o=i.apply(void 0,a);return o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),z.endScope(o),o}catch(l){throw z.endScope(null),l}};return Object.defineProperty(r,"name",{value:e,configurable:!0}),r}function wA(n,t){var e=O(n,"real","complex"),i=O(t,"imag","complex");Be(e.shape,i.shape,"real and imag shapes, "+e.shape+" and "+i.shape+", must match in call to tf.complex().");var r=function(s){return s.complex(e,i)},a={real:e,imag:i};return z.runKernelFunc(r,a,null,Af)}var mi=U({complex_:wA});function gi(n,t,e,i){if(i==null&&(i=zs(n)),i==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!Dt(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){xc(t);var r=ot(t),a=ot(e);E(r===a,function(){return"Based on the provided shape, ["+t+"], the tensor should have "+(r+" values but has "+a)});for(var s=0;s<e.length;++s){var o=e[s],l=s===e.length-1?o!==ot(t.slice(s)):!0;E(e[s]===t[s]||!l,function(){return"Error creating a new Tensor. Inferred shape "+("("+e+") does not match the provided ")+("shape ("+t+"). ")})}}return!Dt(n)&&!Array.isArray(n)&&(n=[n]),t=t||e,n=i!=="string"?_s(n,i):ji(n,[],!0),z.makeTensor(n,t,i)}function vi(n,t,e){var i=Fn(n,e);return gi(n,t,i,e)}var Wc={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var qs=4;function LA(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l=this;return pe(this,function(u){switch(u.label){case 0:for(e=[],i=[],r=Array.isArray(n)?n.map(function(c){return c.name}):Object.keys(n),a=function(c){var h=r[c],d=Array.isArray(n)?n[c].tensor:n[h];if(d.dtype!=="float32"&&d.dtype!=="int32"&&d.dtype!=="bool"&&d.dtype!=="string"&&d.dtype!=="complex64")throw new Error("Unsupported dtype in weight '"+h+"': "+d.dtype);var p={name:h,shape:d.shape,dtype:d.dtype};if(d.dtype==="string"){var f=new Promise(function(m){return de(l,void 0,void 0,function(){var g,v,b,w,S,L,N;return pe(this,function(C){switch(C.label){case 0:return[4,d.bytes()];case 1:for(g=C.sent(),v=g.reduce(function(R,D){return R+D.length},0)+qs*g.length,b=new Uint8Array(v),w=0,S=0;S<g.length;S++)L=g[S],N=new Uint8Array(new Uint32Array([L.length]).buffer),b.set(N,w),w+=qs,b.set(L,w),w+=L.length;return m(b),[2]}})})});i.push(f)}else i.push(d.data());t!=null&&(p.group=t),e.push(p)},s=0;s<r.length;++s)a(s);return[4,Promise.all(i)];case 1:return o=u.sent(),[2,{data:SA(o),specs:e}]}})})}function _m(n,t){for(var e={},i,r=0,a=0,s=t;a<s.length;a++){var o=s[a],l=o.name,u=o.dtype,c=o.shape,h=ot(c),d=void 0;if("quantization"in o){var p=o.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error("Weight "+o.name+" with quantization "+p.dtype+" doesn't have corresponding metadata min and scale.")}else if(p.dtype==="float16"){if(u!=="float32")throw new Error("Weight "+o.name+" is quantized with "+p.dtype+" "+("which only supports weights of type float32 not "+u+"."))}else throw new Error("Weight "+o.name+" has unknown "+("quantization dtype "+p.dtype+". ")+"Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.");var f=Wc[p.dtype],m=n.slice(r,r+h*f),g=p.dtype==="uint8"?new Uint8Array(m):new Uint16Array(m);if(u==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){d=new Float32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=b*p.scale+p.min}}else if(p.dtype==="float16")i===void 0&&(i=IA()),d=i(g);else throw new Error("Unsupported quantization type "+p.dtype+" for weight type float32.");else if(u==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error("Unsupported quantization type "+p.dtype+" for weight type int32.");d=new Int32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=Math.round(b*p.scale+p.min)}}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*f}else if(u==="string"){var w=ot(o.shape);d=[];for(var v=0;v<w;v++){var S=new Uint32Array(n.slice(r,r+qs))[0];r+=qs;var L=new Uint8Array(n.slice(r,r+S));d.push(L),r+=S}}else{var N=Wc[u],m=n.slice(r,r+h*N);if(u==="float32")d=new Float32Array(m);else if(u==="int32")d=new Int32Array(m);else if(u==="bool")d=new Uint8Array(m);else if(u==="complex64"){d=new Float32Array(m);for(var C=new Float32Array(d.length/2),R=new Float32Array(d.length/2),v=0;v<C.length;v++)C[v]=d[v*2],R[v]=d[v*2+1];var D=vi(C,c,"float32"),k=vi(R,c,"float32");e[l]=mi(D,k)}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*N}u!=="complex64"&&(e[l]=vi(d,c,u))}return e}function SA(n){if(n===null)throw new Error("Invalid input value: "+JSON.stringify(n));var t=0,e=[];n.forEach(function(a){if(t+=a.byteLength,e.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error("Unsupported TypedArray subtype: "+a.constructor.name)});var i=new Uint8Array(t),r=0;return e.forEach(function(a){i.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),i.buffer}var Uc=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function Mm(n){return Uc?Buffer.byteLength(n):new Blob([n]).size}function AA(n){if(Uc)return Buffer.from(n).toString("base64");for(var t=new Uint8Array(n),e="",i=0,r=t.length;i<r;i++)e+=String.fromCharCode(t[i]);return btoa(e)}function TA(n){if(Uc){var t=Buffer.from(n,"base64");return t.buffer.slice(t.byteOffset,t.byteOffset+t.byteLength)}for(var e=atob(n),i=new Uint8Array(e.length),r=0;r<e.length;++r)i.set([e.charCodeAt(r)],r);return i.buffer}function Bc(n){if(n.length===1)return n[0];var t=0;n.forEach(function(r){t+=r.byteLength});var e=new Uint8Array(t),i=0;return n.forEach(function(r){e.set(new Uint8Array(r),i),i+=r.byteLength}),e.buffer}function Hm(n){var t="/";for(n=n.trim();n.endsWith(t);)n=n.slice(0,n.length-1);var e=n.split(t);return e[e.length-1]}function Pa(n){if(n.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:n.modelTopology==null?0:Mm(JSON.stringify(n.modelTopology)),weightSpecsBytes:n.weightSpecs==null?0:Mm(JSON.stringify(n.weightSpecs)),weightDataBytes:n.weightData==null?0:n.weightData.byteLength}}function NA(){var n=function(i){for(var r=i<<13,a=0;(r&8388608)===0;)a-=8388608,r<<=1;return r&=~8388608,a+=947912704,r|a},t=new Uint32Array(2048);t[0]=0;for(var e=1;e<1024;e++)t[e]=n(e);for(var e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function xA(){var n=new Uint32Array(64);n[0]=0,n[31]=1199570944,n[32]=2147483648,n[63]=3347054592;for(var t=1;t<31;t++)n[t]=t<<23;for(var t=33;t<63;t++)n[t]=2147483648+(t-32<<23);return n}function CA(){for(var n=new Uint32Array(64),t=0;t<64;t++)n[t]=1024;return n[0]=n[32]=0,n}function IA(){var n=NA(),t=xA(),e=CA();return function(i){for(var r=new ArrayBuffer(4*i.length),a=new Uint32Array(r),s=0;s<i.length;s++){var o=i[s],l=n[e[o>>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var tn=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),RA=function(n){return tn.registerSaveRouter(n)},OA=function(n){return tn.registerLoadRouter(n)},EA=function(n){return tn.getSaveHandlers(n)},DA=function(n,t){return tn.getLoadHandlers(n,t)};var zc="tensorflowjs",Pc=1,Ki="models_store",yi="model_info_store";function Vm(){if(!He().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function _c(n){var t=n.result;t.createObjectStore(Ki,{keyPath:"modelPath"}),t.createObjectStore(yi,{keyPath:"modelPath"})}var zr=function(){function n(t){if(this.indexedDB=Vm(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(zc,Pc);s.onupgradeneeded=function(){return _c(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(Ki,"readonly"),u=l.objectStore(Ki),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=Pa(e),d=o.transaction(yi,"readwrite"),p=d.objectStore(yi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(Ki,"readwrite");var g=m.objectStore(Ki),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(yi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),qm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(zr.URL_SCHEME))?kA(n.slice(zr.URL_SCHEME.length)):null};tn.registerSaveRouter(qm);tn.registerLoadRouter(qm);function kA(n){return new zr(n)}function FA(n){return n.startsWith(zr.URL_SCHEME)?n.slice(zr.URL_SCHEME.length):n}var WA=function(){function n(){this.indexedDB=Vm()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(zc,Pc);a.onupgradeneeded=function(){return _c(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(yi,"readonly"),l=o.objectStore(yi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h<d.length;h++){var p=d[h];c[p.modelPath]=p.modelArtifactsInfo}i(c)},u.onerror=function(c){return s.close(),r(u.error)},o.oncomplete=function(){return s.close()}},a.onerror=function(s){return r(a.error)}})]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e=this;return pe(this,function(i){return t=FA(t),[2,new Promise(function(r,a){var s=e.indexedDB.open(zc,Pc);s.onupgradeneeded=function(){return _c(s)},s.onsuccess=function(){var o=s.result,l=o.transaction(yi,"readwrite"),u=l.objectStore(yi),c=u.get(t),h;c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+t+"' in IndexedDB."));var d=u.delete(t),p=function(){h=o.transaction(Ki,"readwrite");var f=h.objectStore(Ki),m=f.delete(t);m.onsuccess=function(){return r(c.result.modelArtifactsInfo)},m.onerror=function(g){return a(c.error)}};d.onsuccess=p,d.onerror=function(f){return p(),o.close(),a(c.error)}},c.onerror=function(d){return o.close(),a(c.error)},l.oncomplete=function(){h==null?o.close():h.oncomplete=function(){return o.close()}}},s.onerror=function(o){return a(s.error)}})]})})},n}();var ei="/",Pr="tensorflowjs_models",Gm="info",UA="model_topology",BA="weight_specs",zA="weight_data",PA="model_metadata";function Ym(n){return{info:[Pr,n,Gm].join(ei),topology:[Pr,n,UA].join(ei),weightSpecs:[Pr,n,BA].join(ei),weightData:[Pr,n,zA].join(ei),modelMetadata:[Pr,n,PA].join(ei)}}function _A(n){var t=n.split(ei);if(t.length<3)throw new Error("Invalid key format: "+n);return t.slice(1,t.length-1).join(ei)}function MA(n){return n.startsWith(_r.URL_SCHEME)?n.slice(_r.URL_SCHEME.length):n}var _r=function(){function n(t){if(!He().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=Ym(this.modelPath)}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r;return pe(this,function(a){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");e=JSON.stringify(t.modelTopology),i=JSON.stringify(t.weightSpecs),r=Pa(t);try{return this.LS.setItem(this.keys.info,JSON.stringify(r)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,i),this.LS.setItem(this.keys.weightData,AA(t.weightData)),this.LS.setItem(this.keys.modelMetadata,JSON.stringify({format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata})),[2,{modelArtifactsInfo:r}]}catch(s){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error("Failed to save model '"+this.modelPath+"' to local storage: size quota being exceeded is a possible cause of this failure: "+("modelTopologyBytes="+r.modelTopologyBytes+", ")+("weightSpecsBytes="+r.weightSpecsBytes+", ")+("weightDataBytes="+r.weightDataBytes+"."))}return[2]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o;return pe(this,function(l){if(t=JSON.parse(this.LS.getItem(this.keys.info)),t==null)throw new Error("In local storage, there is no model with name '"+this.modelPath+"'");if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");if(e={},i=JSON.parse(this.LS.getItem(this.keys.topology)),i==null)throw new Error("In local storage, the topology of model '"+this.modelPath+"' is missing.");if(e.modelTopology=i,r=JSON.parse(this.LS.getItem(this.keys.weightSpecs)),r==null)throw new Error("In local storage, the weight specs of model '"+this.modelPath+"' are missing.");if(e.weightSpecs=r,a=this.LS.getItem(this.keys.modelMetadata),a!=null&&(s=JSON.parse(a),e.format=s.format,e.generatedBy=s.generatedBy,e.convertedBy=s.convertedBy,e.userDefinedMetadata=s.userDefinedMetadata),o=this.LS.getItem(this.keys.weightData),o==null)throw new Error("In local storage, the binary weight values of model "+("'"+this.modelPath+"' are missing."));return e.weightData=TA(o),[2,e]})})},n.URL_SCHEME="localstorage://",n}(),jm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(_r.URL_SCHEME))?HA(n.slice(_r.URL_SCHEME.length)):null};tn.registerSaveRouter(jm);tn.registerLoadRouter(jm);function HA(n){return new _r(n)}var VA=function(){function n(){E(He().getBool("IS_BROWSER"),function(){return"Current environment is not a web browser"}),E(typeof window=="undefined"||typeof window.localStorage!="undefined",function(){return"Current browser does not appear to support localStorage"}),this.LS=window.localStorage}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s;return pe(this,function(o){for(t={},e=Pr+ei,i=ei+Gm,r=0;r<this.LS.length;++r)a=this.LS.key(r),a.startsWith(e)&&a.endsWith(i)&&(s=_A(a),t[s]=JSON.parse(this.LS.getItem(a)));return[2,t]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){if(t=MA(t),e=Ym(t),this.LS.getItem(e.info)==null)throw new Error("Cannot find model at path '"+t+"'");return i=JSON.parse(this.LS.getItem(e.info)),this.LS.removeItem(e.info),this.LS.removeItem(e.topology),this.LS.removeItem(e.weightSpecs),this.LS.removeItem(e.weightData),[2,i]})})},n}();var Mr="://",bi=function(){function n(){this.managers={}}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerManager=function(t,e){E(t!=null,function(){return"scheme must not be undefined or null."}),t.endsWith(Mr)&&(t=t.slice(0,t.indexOf(Mr))),E(t.length>0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function Gs(n){if(n.indexOf(Mr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+bi.getSchemes().join(",")));return{scheme:n.split(Mr)[0],path:n.split(Mr)[1]}}function Km(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=tn.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=tn.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=Gs(n).scheme,l=Gs(n).path,u=o===Gs(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,bi.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,bi.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function qA(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=bi.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e<i.length?(r=i[e],[4,bi.getManager(r).listModels()]):[3,4];case 2:a=l.sent();for(s in a)o=r+Mr+s,t[o]=a[s];l.label=3;case 3:return e++,[3,1];case 4:return[2,t]}})})}function GA(n){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){return t=Gs(n),e=bi.getManager(t.scheme),[2,e.removeModel(t.path)]})})}function YA(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!1,[2,Km(n,t,e)]})})}function jA(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!0,[2,Km(n,t,e)]})})}var KA=function(){function n(){}return n.prototype.fetch=function(t,e){return fetch(t,e)},n.prototype.now=function(){return performance.now()},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Browser's encoder only supports utf-8, but got "+e);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return new TextDecoder(e).decode(t)},n}();if(He().get("IS_BROWSER")){He().setPlatform("browser",new KA);try{bi.registerManager(_r.URL_SCHEME,new VA)}catch(n){}try{bi.registerManager(zr.URL_SCHEME,new WA)}catch(n){}}var $A={importFetch:function(){return ff()}},Mc,XA=function(){function n(){this.util=require("util"),this.textEncoder=new this.util.TextEncoder}return n.prototype.fetch=function(t,e){return He().global.fetch!=null?He().global.fetch(t,e):(Mc==null&&(Mc=$A.importFetch()),Mc(t,e))},n.prototype.now=function(){var t=process.hrtime();return t[0]*1e3+t[1]/1e6},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Node built-in encoder only supports utf-8, but got "+e);return this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)},n}();He().get("IS_NODE")&&He().setPlatform("node",new XA);function Wn(n,t,e){return t===void 0&&(t="float32"),t=t||"float32",xc(n),new Hs(n,t,e)}function JA(n,t){var e=O(n,"x","cast");if(!Im(t))throw new Error("Failed to cast to unknown dtype "+t);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");var i={x:e},r={dtype:t};return z.runKernelFunc(function(a){return a.cast(e,t)},i,null,Ws,r)}var ue=U({cast_:JA});function ZA(n){var t=O(n,"x","clone",null),e=function(){return z.makeTensorFromDataId(t.dataId,t.shape,t.dtype)},i={x:t};return z.runKernelFunc(e,i,null,vu)}var $i=U({clone_:ZA});function $m(n,t){t===void 0&&(t=!1),console.log(n.toString(t))}Wm();var QA={buffer:Wn,cast:ue,clone:$i,print:$m};lA(QA);var eT="model",tT=".json",nT=".weights.bin";function Xm(n){return new Promise(function(t){return setTimeout(t)}).then(n)}var Hc=function(){function n(t){if(!He().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(n.URL_SCHEME)&&(t=t.slice(n.URL_SCHEME.length)),(t==null||t.length===0)&&(t=eT),this.modelTopologyFileName=t+tT,this.weightDataFileName=t+nT}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");if(e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"})),!(t.modelTopology instanceof ArrayBuffer))return[3,1];throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");case 1:return i=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:i},a=window.URL.createObjectURL(new Blob([JSON.stringify(r)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor,s.download=this.modelTopologyFileName,s.href=a,[4,Xm(function(){return s.dispatchEvent(new MouseEvent("click"))})];case 2:return l.sent(),t.weightData!=null?(o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor,o.download=this.weightDataFileName,o.href=e,[4,Xm(function(){return o.dispatchEvent(new MouseEvent("click"))})]):[3,4];case 3:l.sent(),l.label=4;case 4:return[2,{modelArtifactsInfo:Pa(t)}]}})})},n.URL_SCHEME="downloads://",n}(),iT=function(){function n(t){if(t==null||t.length<1)throw new Error("When calling browserFiles, at least 1 file is required, "+("but received "+t));this.files=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i=this;return pe(this,function(r){return t=this.files[0],e=this.files.slice(1),[2,new Promise(function(a,s){var o=new FileReader;o.onload=function(l){var u=JSON.parse(l.target.result),c=u.modelTopology;if(c==null){s(new Error("modelTopology field is missing from file "+t.name));return}e.length===0&&a({modelTopology:c});var h=u.weightsManifest;if(h==null){s(new Error("weightManifest field is missing from file "+t.name));return}var d;try{d=i.checkManifestAndWeightFiles(h,e)}catch(g){s(g);return}var p=[],f=[],m=[];h.forEach(function(g){g.paths.forEach(function(v){f.push(v),m.push(null)}),p.push.apply(p,g.weights)}),h.forEach(function(g){g.paths.forEach(function(v){var b=new FileReader;b.onload=function(w){var S=w.target.result,L=f.indexOf(v);m[L]=S,m.indexOf(null)===-1&&a({modelTopology:c,weightSpecs:p,weightData:Bc(m),format:u.format,generatedBy:u.generatedBy,convertedBy:u.convertedBy,userDefinedMetadata:u.userDefinedMetadata})},b.onerror=function(w){return s("Failed to weights data from file of path '"+v+"'.")},b.readAsArrayBuffer(d[v])})})},o.onerror=function(l){return s("Failed to read model topology and weights manifest JSON "+("from file '"+t.name+"'. BrowserFiles supports loading ")+"Keras-style tf.Model artifacts only.")},o.readAsText(t)})]})})},n.prototype.checkManifestAndWeightFiles=function(t,e){for(var i=[],r=e.map(function(u){return Hm(u.name)}),a={},s=0,o=t;s<o.length;s++){var l=o[s];l.paths.forEach(function(u){var c=Hm(u);if(i.indexOf(c)!==-1)throw new Error("Duplicate file basename found in weights manifest: "+("'"+c+"'"));if(i.push(c),r.indexOf(c)===-1)throw new Error("Weight file with basename '"+c+"' is not provided.");a[u]=e[r.indexOf(c)]})}if(i.length!==e.length)throw new Error("Mismatch in the number of files in weights manifest "+("("+i.length+") and the number of weight files provided ")+("("+e.length+")."));return a},n}(),aT=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Hc.URL_SCHEME))?rT(n.slice(Hc.URL_SCHEME.length)):null};tn.registerSaveRouter(aT);function rT(n){return n===void 0&&(n="model"),new Hc(n)}function sT(n){return new iT(n)}function Jm(n,t,e,i){s(n),e=e??0,i=i??1,o(e,i);var r=0,a=function(l){return l.then(function(u){var c=e+ ++r/n.length*(i-e);return t(c),u}),l};function s(l){E(l!=null&&Array.isArray(l)&&l.length>0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function Zm(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?He().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,Jm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,Jm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function oT(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return Zm(o,{requestInit:i})},a=Qm(r),[2,a(n,t,e)]})})}function Qm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=Wc[S]*ot(w.shape),N=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(N(),o[R]=!0)}):N(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. 
+`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w<v;w++)b+=d[f+w].byteLength;for(var S=new ArrayBuffer(b),L=new Uint8Array(S),N=0,C=0;C<v;C++){var R=new Uint8Array(d[f+C]);L.set(R,N),N+=R.byteLength}var D=s[g];D.forEach(function(k){var W=S.slice(k.groupOffset,k.groupOffset+k.sizeBytes),F=_m(W,[k.manifestEntry]);for(var P in F)p[P]=F[P]}),f+=v}),[2,p]}})})}}var lT="application/octet-stream",uT="application/json",eg=function(){function n(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(E(typeof e.fetchFunc=="function",function(){return"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"}),this.fetch=e.fetchFunc):this.fetch=He().platform.fetch,E(t!=null&&t.length>0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:uT}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:lT}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:Pa(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=cT(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;l<u.length;l++)c=u[l],o.push.apply(o,c.weights);for(h=[],d=[],p=0,f=t;p<f.length;p++)for(m=f[p],g=0,v=m.paths;g<v.length;g++)b=v[g],this.weightUrlConverter!=null?d.push(this.weightUrlConverter(b)):h.push(s+b+a);return this.weightUrlConverter?(S=(w=h.push).apply,L=[h],[4,Promise.all(d)]):[3,2];case 1:S.apply(w,L.concat([C.sent()])),C.label=2;case 2:return[4,Zm(h,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress})];case 3:return N=C.sent(),[2,[o,Bc(N)]]}})})},n.URL_SCHEME_REGEX=/^https?:\/\//,n}();function cT(n){var t=n.lastIndexOf("/"),e=n.lastIndexOf("?"),i=n.substring(0,t),r=e>t?n.substring(e):"";return[i+"/",r]}function Vc(n){return n.match(eg.URL_SCHEME_REGEX)!=null}var tg=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return Vc(i)}):e=Vc(n),e?qc(n,t):null};tn.registerSaveRouter(tg);tn.registerLoadRouter(tg);function qc(n,t){return new eg(n,t)}function hT(n,t){return qc(n,t)}var Gc=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),dT=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function pT(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new Gc(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new Gc({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new Gc({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function fT(n){return new dT(n)}var mT={__proto__:null,browserFiles:sT,browserHTTPRequest:hT,concatenateArrayBuffers:Bc,decodeWeights:_m,encodeWeights:LA,fromMemory:pT,getLoadHandlers:DA,getModelArtifactsInfoForJSON:Pa,getSaveHandlers:EA,http:qc,isHTTPScheme:Vc,loadWeights:oT,registerLoadRouter:OA,registerSaveRouter:RA,weightsLoaderFactory:Qm,withSaveHandler:fT,copyModel:YA,listModels:qA,moveModel:jA,removeModel:GA};function gT(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=bm(t,e.size),E(e.size===ot(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,Mu,r)}var V=U({reshape_:gT});function vT(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=at(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=ot(h),f=ot(d);E(hn(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(N,C){return C([g,v]),N.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,jl,S);return V(L,m)}var We=U({matMul_:vT});function yT(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,Wu,l)}var Ys=U({oneHot_:yT});function bT(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a<e.rank,function(){return"All entries in 'perm' must be between 0 and "+(e.rank-1)+(" but got "+t)})}),e.rank<=1)return e.clone();var i={x:e},r={perm:t};return z.runKernelFunc(function(a){return a.transpose(e,t)},i,null,dc,r)}var ut=U({transpose_:bT});function wT(n,t,e){var i=O(n,"labels","confusionMatrix"),r=O(t,"predictions","confusionMatrix");E(e==null||e>0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=Ys(ue(i,"int32"),e),s=Ys(ue(r,"int32"),e),o=ut(a);return ue(We(o,s),"int32")}var ST=U({confusionMatrix_:wT});var LT={__proto__:null,confusionMatrix:ST};function ng(n,t,e){if(Yi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=Fn(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return gi(n,t,i,e)}var Hr;function IT(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState<o)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}var l=Lc(vc,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(vc,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Hr==null&&(Hr=document.createElement("canvas").getContext("2d")),Hr.canvas.width=d,Hr.canvas.height=p,Hr.drawImage(n,0,0,d,p),f=Hr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v<g;v++)for(var b=0;b<t;++b)m[v*t+b]=f[v*4+b]}var w=[p,d,t];return ng(m,w,"int32")}function AT(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v;return pe(this,function(b){switch(b.label){case 0:if(e=O(n,"img","toPixels"),n instanceof Y||(i=e,e=ue(i,"int32"),i.dispose()),e.rank!==2&&e.rank!==3)throw new Error("toPixels only supports rank 2 or 3 tensors, got rank "+e.rank+".");if(r=e.shape.slice(0,2),a=r[0],s=r[1],o=e.rank===2?1:e.shape[2],o>4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h<a*s;++h){for(d=[0,0,0,255],p=0;p<o;p++){if(f=l[h*o+p],e.dtype==="float32"){if(f<0||f>1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var TT=U({fromPixels_:IT}),NT={__proto__:null,toPixels:AT,fromPixels:TT};function ig(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a<e.length-1;++a)r*=e[a];var s=n.shape,o=e.slice();o.pop();for(var l=1,a=i;a<n.rank;++a)l*=s[a],o.push(s[a]);var u=Fr(n.shape).map(function(c){return c/l}).concat([1]).slice(0,i);return[o,r,l,u]}var xT={__proto__:null,prepareAndValidate:ig};function Yc(n,t,e){var i=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank<r)throw new Error(a+(" update.rank < "+r+". "));if(n.length<i+(e.rank-r))throw new Error(a+(" Output shape length < "+(i+(e.rank-r))));if(e.rank!==r+n.length-i)throw new Error(a+(" update.rank != "+(r+n.length-i)));for(var s=0;s<r;++s)if(e.shape[s]!==t.shape[s])throw new Error(a+(" updates.shape["+s+"] ("+e.shape[s]+") != indices.shape["+s+"] ("+t.shape[s]+")."));for(var s=0;s<e.rank-r;++s)if(e.shape[s+r]!==n[s+i])throw new Error(a+(" updates.shape["+(s+r)+"] ("+e.shape[s+r]+") != shape["+(s+r)+"] ("+n[s+r]+")"))}function jc(n,t,e){if(t.rank<1)throw new Error("tf.scatterND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(n.rank<1)throw new Error("tf.scatterND() expects the updates to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.dtype!=="int32")throw new Error("The dtype of 'indices' should be int32, but got dtype: "+t.dtype);if(e.length<1)throw new Error("Output rank must be greater or equal to 1, but got shape: "+e);if(e.length===0){if(t.size===0)throw new Error("Indices specified for empty output. indices shape: "+t.shape);if(n.size===0)throw new Error("Updates specified for empty output. updates shape: "+n.shape)}Yc(e,t,n)}function rg(n,t,e){for(var i=t.shape.length,r=i>1?t.shape[i-1]:1,a=e.length,s=1,o=r;o<a;++o)s*=e[o];var l=r<1?1:r,u=ot(t.shape)/l,c=Fr(e.slice(0,r)).concat([1]),h=ot(e);return{sliceRank:r,numUpdates:u,sliceSize:s,strides:c,outputSize:h}}var CT={__proto__:null,validateUpdateShape:Yc,validateInput:jc,calculateShapes:rg};function ag(n,t,e){var i=n.shape.length;E(i===t.length,function(){return"Error in slice"+i+"D: Length of begin "+t+" must "+("match the rank of the array ("+i+").")}),E(i===e.length,function(){return"Error in slice"+i+"D: Length of size "+e+" must "+("match the rank of the array ("+i+").")});for(var r=function(s){E(t[s]+e[s]<=n.shape[s],function(){return"Error in slice"+i+"D: begin["+s+"] + size["+s+"] "+("("+(t[s]+e[s])+") would overflow input.shape["+s+"] ("+n.shape[s]+")")})},a=0;a<i;++a)r(a)}function js(n){for(var t=[],e=0;n>0;)n&1&&t.push(e),n/=2,e++;return t}function sg(n,t,e){for(var i=[],r=0;r<n.length;r++)i[r]=Math.ceil((t[r]-n[r])/e[r]);return i}function og(n,t,e,i){for(var r=n.slice(),a=r.length;a<i.length;a++)r.push(1);for(var a=0;a<e;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function lg(n,t,e){return e<=n?e:e-(t-1)}function ug(n,t){for(var e=[],i=0;i<n;i++)e.push(t+i);return e}function mg(n,t,e,i,r,a,s,o,l){var u=n.length,c=new Array(u),h=new Array(u),d=new Array(u);if(t.length&&e>0){var p=t[0],f=e+1;c=cg(s,p,f,i,n),h=hg(o,p,f,r,n),d=og(a,p,f,n)}else for(var m=0;m<u;m++)c[m]=pg(s,i,a,n,m,l),h[m]=fg(o,r,a,n,m,l),d[m]=dg(a,m,l);return{begin:c,end:h,strides:d}}function cg(n,t,e,i,r){for(var a=r.slice(),s=ug(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=0;else{var l=lg(t,e,o),u=i[l];n&1<<l&&(u=0),a[o]=u}return a}function hg(n,t,e,i,r){for(var a=r.slice(),s=ug(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=lg(t,e,o),u=i[l];n&1<<l&&(u=Number.MAX_SAFE_INTEGER),a[o]=u}for(var c=0;c<a.length;c++){var h=r[c];a[c]<0&&(a[c]+=h),a[c]=Da(0,a[c],r[c])}return a}function dg(n,t,e){var i=n[t];return(e&1<<t||i==null)&&(i=1),i}function pg(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=Da(0,s,l-1),s}function fg(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=Da(0,s,l):s=Da(-1,s,l-1),s}function RT(n,t,e){for(var i=e.length,r=0;r<e.length;r++)if(e[r]>1){i=r;break}for(var r=i+1;r<e.length;r++)if(t[r]>0||e[r]!==n[r])return!1;return!0}function OT(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i<n.length-1;i++)e+=n[i]*t[i];return e}function Kc(n,t,e){var i,r=n.shape.length;typeof t=="number"?i=[t].concat(new Array(r-1).fill(0)):t.length<r?i=t.concat(new Array(r-t.length).fill(0)):i=t.slice(),i.forEach(function(s){E(s!==-1,function(){return"slice() does not support negative begin indexing."})});var a;return e==null?a=new Array(r).fill(-1):typeof e=="number"?a=[e].concat(new Array(r-1).fill(-1)):e.length<r?a=e.concat(new Array(r-e.length).fill(-1)):a=e,a=a.map(function(s,o){return s>=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var gg={__proto__:null,assertParamsValid:ag,maskToAxes:js,computeOutShape:sg,stridesWithElidedDims:og,getNormalizedAxes:mg,startIndicesWithElidedDims:cg,stopIndicesWithElidedDims:hg,stridesForAxis:dg,startForAxis:pg,stopForAxis:fg,isSliceContinous:RT,computeFlatOffset:OT,parseSliceParams:Kc};var vg=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),yg=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function wi(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),yg.register(n)}var ET={__proto__:null,Serializable:vg,SerializationMap:yg,registerClass:wi};var DT=.001,bg=.1;function kT(n,t,e){return e==null&&(e=$c()),Xc(n,t,function(i,r){return Jc(i,r,e)})}function $c(){return z.backend.floatPrecision()===32?DT:bg}function Xc(n,t,e){var i=!0;if((Dt(n)||Dt(t))&&(i=!1),Dt(n)&&Dt(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=Fn(n),o=Fn(t);if(!hn(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Dt(n)?n:ji(n),u=Dt(t)?t:ji(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`.
 `)+("Actual:   "+l+`.
 `)+("Expected: "+u+"."));for(var c=0;c<u.length;++c){var h=l[c],d=u[c];if(!e(h,d))throw new Error("Arrays differ: actual["+c+"] = "+h+", expected["+c+"] = "+d+`.
 `+("Actual:   "+l+`.
-`)+("Expected: "+u+"."))}}function FT(n,t){n().then(function(){return t.fail()},function(){return t()})}function WT(n,t){var e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return pi(n)||pi(n[0])||pi(t)||pi(t[0])?$c(n,e,function(i,r){return i==r}):$c(n,t,function(i,r){return Xc(i,r,0)})}function UT(n,t,e){if(e==null&&(e=Kc()),!Xc(n,t,e))throw new Error("Numbers differ: actual === "+n+", expected === "+t)}function Xc(n,t,e){return!isFinite(n)&&!isFinite(t)?!0:!(isNaN(n)||isNaN(t)||Math.abs(n-t)>e)}function BT(n,t,e){for(var i=0;i<n.length;i++)if(n[i]<t||n[i]>e)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function zT(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var PT={__proto__:null,TEST_EPSILON_FLOAT16:bg,expectArraysClose:kT,testEpsilon:Kc,expectPromiseToFail:FT,expectArraysEqual:WT,expectNumbersClose:UT,expectValuesInRange:BT,expectArrayBuffersEqual:zT};var _T="2.6.0";function MT(){He().set("PROD",!0)}function HT(){He().set("DEBUG",!0)}function VT(){He().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function At(n){He().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function qT(){z.disposeVariables()}function GT(){return z}function YT(){return z.memory()}function jT(n){return z.profile(n)}function mt(n,t){return z.tidy(n,t)}function Mt(n){var t=kc(n);t.forEach(function(e){return e.dispose()})}function wg(n){return z.keep(n)}function KT(n){return z.time(n)}function $T(n){return z.setBackend(n)}function XT(){return z.ready()}function JT(){return z.backendName}function ZT(n){z.removeBackend(n)}function QT(n){return z.findBackend(n)}function e1(n){return z.findBackendFactory(n)}function t1(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function n1(){return z.backend}function i1(n,t){He().setPlatform(n,t)}function r1(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Fs)}var fe=U({add_:r1});function a1(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,du)}var Jc=U({floorDiv_:a1});function s1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=at(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return Jc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,su,o)}var Ie=U({div_:s1});function o1(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Eu)}var J=U({mul_:o1});function l1(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,kl)}var Kt=U({abs_:l1});function u1(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,Fl)}var Sg=U({acos_:u1});function c1(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,Wl)}var Lg=U({acosh_:c1});function h1(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!dn(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,Ul)}var d1=U({addN_:h1});function Zc(n,t){for(var e=0;e<n.length;++e)if(n[n.length-e-1]!==t-1-e)return!1;return!0}function Ig(n,t,e){for(var i=n.length+t.length,r=[],a=0,s=0,o=0;o<i;o++)e.indexOf(o)===-1?r.push(n[a++]):r.push(t[s++]);return r}function Ag(n,t){for(var e=[],i=n.length,r=0;r<i;r++)t.indexOf(r)===-1&&e.push(n[r]);var a=t.map(function(s){return n[s]});return[e,a]}function rn(n,t){var e=t.map(function(i){return 1});return Ig(n,e,t)}function p1(n,t,e){E(Zc(t,e),function(){return n+" supports only inner-most axes for now. "+("Got axes "+t+" and rank-"+e+" input.")})}function an(n,t){if(Zc(n,t))return null;for(var e=[],i=0;i<t;++i)n.indexOf(i)===-1&&e.push(i);return n.forEach(function(r){return e.push(r)}),e}function Ks(n){return n.map(function(t,e){return[e,t]}).sort(function(t,e){return t[1]-e[1]}).map(function(t){return t[0]})}function Un(n,t){for(var e=[],i=t-n;i<t;++i)e.push(i);return e}function f1(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","all","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=an(u,i.rank);c!=null&&(i=ut(i,c),u=Un(u.length,i.rank));var h=o.all(i,u);if(e){var d=rn(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,wf,s)}var Tg=U({all_:f1});function m1(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","any","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=an(u,i.rank);c!=null&&(i=ut(i,c),u=Un(u.length,i.rank));var h=o.any(i,u);if(e){var d=rn(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Sf,s)}var Ng=U({any_:m1});function g1(n,t){t===void 0&&(t=0);var e=O(n,"x","argMax"),i=function(s,o){o([e]);var l=Qe(t,e.shape),u=an(l,e.rank);return u!=null&&(e=ut(e,u),l=Un(l.length,e.rank)),s.argMax(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,Bl,a)}var xg=U({argMax_:g1});function v1(n,t){t===void 0&&(t=0);var e=O(n,"x","argMin"),i=function(s,o){o([e]),t==null&&(t=0);var l=Qe(t,e.shape),u=an(l,e.rank);return u!=null&&(e=ut(e,u),l=Un(l.length,e.rank)),s.argMin(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,zl,a)}var Cg=U({argMin_:v1});function y1(n){var t=O(n,"x","asin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asin(t);return r([t]),a},e,null,Pl)}var Rg=U({asin_:y1});function b1(n){var t=O(n,"x","asinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asinh(t);return r([t]),a},e,null,_l)}var Og=U({asinh_:b1});function w1(n){var t=O(n,"x","atan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atan(t);return r([t]),a},e,null,Ml)}var Eg=U({atan_:w1});function S1(n,t){var e,i=O(n,"a","atan2"),r=O(t,"b","atan2");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.atan2(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Vl)}var Dg=U({atan2_:S1});function L1(n){var t=O(n,"x","atanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atanh(t);return r([t]),a},e,null,Hl)}var kg=U({atanh_:L1});function I1(n,t,e,i,r,a){r===void 0&&(r="NHWC");var s=n[3],o=t.concat([s]),l=_a(r);return Bn(n,o,e,a,i,null,null,l)}function Vr(n,t,e,i,r,a,s){s===void 0&&(s="channelsLast");var o=$s(t),l=o[0],u=o[1],c;if(s==="channelsLast")c=[l,u,n[3],n[3]];else if(s==="channelsFirst")c=[l,u,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Bn(n,c,e,i,r,a,!1,s)}function Ha(n,t,e,i,r,a,s){s===void 0&&(s="NDHWC");var o=Qc(t),l=o[0],u=o[1],c=o[2],h,d;if(s==="NDHWC")d="channelsLast",h=[l,u,c,n[4],n[4]];else if(s==="NCDHW")d="channelsFirst",h=[l,u,c,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Ma(n,h,e,i,r,!1,d,a)}function Bn(n,t,e,i,r,a,s,o){s===void 0&&(s=!1),o===void 0&&(o="channelsLast");var l=[-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3];if(o==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3];else if(o==="channelsFirst")u=n[0],d=n[1],c=n[2],h=n[3];else throw new Error("Unknown dataFormat "+o);var p=t[0],f=t[1],m=t[3],g=$s(e),v=g[0],b=g[1],w=$s(i),S=w[0],L=w[1],N=qr(p,S),C=qr(f,L),R=A1(r,c,h,v,b,N,C,a,o),D=R.padInfo,k=R.outHeight,W=R.outWidth,F=s?m*d:m,P;return o==="channelsFirst"?P=[u,F,k,W]:o==="channelsLast"&&(P=[u,k,W,F]),{batchSize:u,dataFormat:o,inHeight:c,inWidth:h,inChannels:d,outHeight:k,outWidth:W,outChannels:F,padInfo:D,strideHeight:v,strideWidth:b,filterHeight:p,filterWidth:f,effectiveFilterHeight:N,effectiveFilterWidth:C,dilationHeight:S,dilationWidth:L,inShape:n,outShape:P,filterShape:t}}function Ma(n,t,e,i,r,a,s,o){a===void 0&&(a=!1),s===void 0&&(s="channelsLast");var l=[-1,-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3],p=l[4];if(s==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3],p=n[4];else if(s==="channelsFirst")u=n[0],p=n[1],c=n[2],h=n[3],d=n[4];else throw new Error("Unknown dataFormat "+s);var f=t[0],m=t[1],g=t[2],v=t[4],b=Qc(e),w=b[0],S=b[1],L=b[2],N=Qc(i),C=N[0],R=N[1],D=N[2],k=qr(f,C),W=qr(m,R),F=qr(g,D),P=T1(r,c,h,d,w,S,L,k,W,F,o),H=P.padInfo,_=P.outDepth,j=P.outHeight,K=P.outWidth,q=a?v*p:v,G;return s==="channelsFirst"?G=[u,q,_,j,K]:s==="channelsLast"&&(G=[u,_,j,K,q]),{batchSize:u,dataFormat:s,inDepth:c,inHeight:h,inWidth:d,inChannels:p,outDepth:_,outHeight:j,outWidth:K,outChannels:q,padInfo:H,strideDepth:w,strideHeight:S,strideWidth:L,filterDepth:f,filterHeight:m,filterWidth:g,effectiveFilterDepth:k,effectiveFilterHeight:W,effectiveFilterWidth:F,dilationDepth:C,dilationHeight:R,dilationWidth:D,inShape:n,outShape:G,filterShape:t}}function N1(n,t,e,i,r){i==null&&(i=eh(n,t,e));var a=n[0],s=n[1],o=Xi((a-t+2*i)/e+1,r);E(rt(o),function(){return"The output # of rows ("+o+") must be an integer. Change the stride and/or zero pad parameters"});var l=Xi((s-t+2*i)/e+1,r);return E(rt(l),function(){return"The output # of columns ("+l+") must be an integer. Change the stride and/or zero pad parameters"}),[o,l]}function x1(n,t,e,i,r,a){r==null&&(r=eh(n,t,i));var s=n[0],o=n[1],l=n[2],u=Xi((s-t+2*r)/i+1,a);E(rt(u),function(){return"The output # of depths ("+u+") must be an integer. Change the stride and/or zero pad parameters"});var c=Xi((o-t+2*r)/i+1,a);E(rt(c),function(){return"The output # of rows ("+c+") must be an integer. Change the stride and/or zero pad parameters"});var h=Xi((l-t+2*r)/i+1,a);return E(rt(h),function(){return"The output # of columns ("+h+") must be an integer. Change the stride and/or zero pad parameters"}),[u,c,h,e]}function eh(n,t,e,i){i===void 0&&(i=1);var r=qr(t,i);return Math.floor((n[0]*(e-1)-e+r)/2)}function $s(n){return typeof n=="number"?[n,n,n]:n.length===2?[n[0],n[1],1]:n}function Qc(n){return typeof n=="number"?[n,n,n]:n}function qr(n,t){return t<=1?n:n+(n-1)*(t-1)}function A1(n,t,e,i,r,a,s,o,l){var u,c,h;if(typeof n=="number"){var d=n===0?"VALID":"NUMBER";u={top:n,bottom:n,left:n,right:n,type:d};var p=N1([t,e],a,i,n,o);c=p[0],h=p[1]}else if(n==="same"){c=Math.ceil(t/i),h=Math.ceil(e/r);var f=Math.max(0,(c-1)*i+a-t),m=Math.max(0,(h-1)*r+s-e),g=Math.floor(f/2),v=f-g,b=Math.floor(m/2),w=m-b;u={top:g,bottom:v,left:b,right:w,type:"SAME"}}else if(n==="valid")u={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-a+1)/i),h=Math.ceil((e-s+1)/r);else if(typeof n=="object"){var S=l==="channelsLast"?n[1][0]:n[2][0],v=l==="channelsLast"?n[1][1]:n[2][1],b=l==="channelsLast"?n[2][0]:n[3][0],w=l==="channelsLast"?n[2][1]:n[3][1],d=S===0&&v===0&&b===0&&w===0?"VALID":"EXPLICIT";u={top:S,bottom:v,left:b,right:w,type:d},c=Xi((t-a+S+v)/i+1,o),h=Xi((e-s+b+w)/r+1,o)}else throw Error("Unknown padding parameter: "+n);return{padInfo:u,outHeight:c,outWidth:h}}function T1(n,t,e,i,r,a,s,o,l,u,c){var h,d,p,f;if(typeof n=="number"){var m=n===0?"VALID":"NUMBER";h={top:n,bottom:n,left:n,right:n,front:n,back:n,type:m};var g=x1([t,e,i,1],o,1,r,n,c);d=g[0],p=g[1],f=g[2]}else if(n==="same"){d=Math.ceil(t/r),p=Math.ceil(e/a),f=Math.ceil(i/s);var v=(d-1)*r+o-t,b=(p-1)*a+l-e,w=(f-1)*s+u-i,S=Math.floor(v/2),L=v-S,N=Math.floor(b/2),C=b-N,R=Math.floor(w/2),D=w-R;h={top:N,bottom:C,left:R,right:D,front:S,back:L,type:"SAME"}}else if(n==="valid")h={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),p=Math.ceil((e-l+1)/a),f=Math.ceil((i-u+1)/s);else throw Error("Unknown padding parameter: "+n);return{padInfo:h,outDepth:d,outHeight:p,outWidth:f}}function Xi(n,t){if(!t)return n;switch(t){case"round":return Math.round(n);case"ceil":return Math.ceil(n);case"floor":return Math.floor(n);default:throw new Error("Unknown roundingMode "+t)}}function Si(n){var t=$s(n),e=t[0],i=t[1],r=t[2];return e===1&&i===1&&r===1}function Ht(n,t){return Si(n)||Si(t)}function _a(n){if(n==="NHWC")return"channelsLast";if(n==="NCHW")return"channelsFirst";throw new Error("Unknown dataFormat "+n)}function C1(n,t,e,i,r){var a=O(n,"x","avgPool","float32"),s=1;E(Ht(e,s),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")});var o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in avgPool: x must be rank 4 but got rank "+o.rank+"."}),r!=null&&E(rt(i),function(){return"Error in avgPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Vr(o.shape,t,e,1,i,r);return f([o]),m.filterWidth===1&&m.filterHeight===1&&dn(m.inShape,m.outShape)?o.clone():p.avgPool(o,m)},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,ql,h);return d=ue(d,a.dtype),l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var th=U({avgPool_:C1});function R1(n,t,e,i,r,a,s){a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:At("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","avgPool3d","float32"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in avgPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in avgPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Ht(e,s),function(){return"Error in avgPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in avgPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ha(l.shape,t,e,s,i,r,a);return m([l]),f.avgPool3d(l,g)},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Gl,d);return p=ue(p,l.dtype),u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var O1=U({avgPool3d_:R1});function Fg(n,t){var e=n[0].length;n.forEach(function(r,a){E(r.length===e,function(){return"Error in concat"+e+"D: rank of tensors["+a+"] must be the same "+("as the rank of the rest ("+e+")")})}),E(t>=0&&t<e,function(){return"Error in concat"+e+"D: axis must be between 0 and "+(e-1)+"."});var i=n[0];n.forEach(function(r,a){for(var s=0;s<e;s++)E(s===t||r[s]===i[s],function(){return"Error in concat"+e+"D: Shape of tensors["+a+"] ("+r+") "+("does not match the shape of the rest ("+i+") ")+("along the non-concatenated axis "+a+".")})})}function Wg(n,t){for(var e=n[0].slice(),i=1;i<n.length;i++)e[t]+=n[i][t];return e}function E1(n,t){t===void 0&&(t=0),E(n.length>=1,function(){return"Pass at least one tensor to concat"});var e=za(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
-          with dtype `+s.dtype+". ")});var i=function(s,o){var l=Qe(t,e[0].shape)[0],u=Wg(e.map(function(d){return d.shape}),l);if(ot(u)===0)return vi([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Fg(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,Jl,a)}var Rt=U({concat_:E1});function D1(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,Qu)}var Ji=U({sigmoid_:D1});function k1(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=jc(i,t,e),c=u[0],h=u[1];return ag(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,$u,s)}var ze=U({slice_:k1});function F1(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,uc)}var Xs=U({tanh_:F1});function W1(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=Rt([u,h],1),p=We(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),N=fe(J(Ji(b),Xs(w)),J(c,Ji(fe(s,S)))),C=J(Xs(N),Ji(L));return[N,C]}var U1=U({basicLSTMCell_:W1});function B1(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length  "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,jl,o)}var Js=U({batchToSpaceND_:B1});function z1(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function P1(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=z1(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Zs(o),Zs(l),Zs(c),Zs(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,pu,f);return V(m,s.shape)}function Zs(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Va=U({batchNorm_:P1});function _1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Va(s,o,l,c,u,a)}var M1=U({batchNorm2d_:_1});function H1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Va(s,o,l,c,u,a)}var V1=U({batchNorm3d_:H1});function q1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Va(s,o,l,c,u,a)}var G1=U({batchNorm4d_:q1});function Y1(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.length<e.rank)throw new Error("broadcastTo(): shape.length="+t.length+" < input.rank="+e.rank+".");if(t.length>e.rank){for(var r=e.shape.slice();r.length<t.length;)r.unshift(1);e=V(e,r)}for(var a=e.shape,s=Array.from(t),o=t.length-1;o>=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return $i(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,Kl,h)}var Qs=U({broadcastTo_:Y1});function j1(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,$l)}var Ug=U({ceil_:j1});function K1(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,Xl,a)}var Bg=U({clipByValue_:K1});function $1(n){return Rt(n,0)}var X1=U({concat1d_:$1});function J1(n,t){return Rt(n,t)}var Z1=U({concat2d_:J1});function Q1(n,t){return Rt(n,t)}var eN=U({concat3d_:Q1});function tN(n,t){return Rt(n,t)}var nN=U({concat4d_:tN});function iN(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(Ht(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=_a(r),w=Bn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,Zl,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Gr=U({conv2d_:iN});function rN(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(Ht(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Gr(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var zg=U({conv1d_:rN});function aN(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=_a(a),S=Bn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,Ql,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var nh=U({conv2DBackpropInput_:aN});function sN(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return nh(e,s,o,i,r,"NHWC",a)}var Pg=U({conv2dTranspose_:sN});function oN(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(Ht(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=Ma(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,eu,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var lN=U({conv3d_:oN});function uN(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=Ma(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,xf,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var _g=U({conv3DBackpropInput_:uN});function cN(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return _g(e,a,s,i,r)}var hN=U({conv3dTranspose_:cN});function dN(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,tu)}var eo=U({cos_:dN});function pN(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,nu)}var ih=U({cosh_:pN});function fN(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=an([t],r.rank),h=r;c!=null&&(h=ut(r,c));var d=Un(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=Ks(c);p=ut(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,iu,o)}var rh=U({cumsum_:fN});function mN(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
+`)+("Expected: "+u+"."))}}function FT(n,t){n().then(function(){return t.fail()},function(){return t()})}function WT(n,t){var e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return pi(n)||pi(n[0])||pi(t)||pi(t[0])?Xc(n,e,function(i,r){return i==r}):Xc(n,t,function(i,r){return Jc(i,r,0)})}function UT(n,t,e){if(e==null&&(e=$c()),!Jc(n,t,e))throw new Error("Numbers differ: actual === "+n+", expected === "+t)}function Jc(n,t,e){return!isFinite(n)&&!isFinite(t)?!0:!(isNaN(n)||isNaN(t)||Math.abs(n-t)>e)}function BT(n,t,e){for(var i=0;i<n.length;i++)if(n[i]<t||n[i]>e)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function zT(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var PT={__proto__:null,TEST_EPSILON_FLOAT16:bg,expectArraysClose:kT,testEpsilon:$c,expectPromiseToFail:FT,expectArraysEqual:WT,expectNumbersClose:UT,expectValuesInRange:BT,expectArrayBuffersEqual:zT};var _T="2.6.0";function MT(){He().set("PROD",!0)}function HT(){He().set("DEBUG",!0)}function VT(){He().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function At(n){He().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function qT(){z.disposeVariables()}function GT(){return z}function YT(){return z.memory()}function jT(n){return z.profile(n)}function mt(n,t){return z.tidy(n,t)}function Mt(n){var t=Fc(n);t.forEach(function(e){return e.dispose()})}function wg(n){return z.keep(n)}function KT(n){return z.time(n)}function $T(n){return z.setBackend(n)}function XT(){return z.ready()}function JT(){return z.backendName}function ZT(n){z.removeBackend(n)}function QT(n){return z.findBackend(n)}function e1(n){return z.findBackendFactory(n)}function t1(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function n1(){return z.backend}function i1(n,t){He().setPlatform(n,t)}function r1(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Fs)}var fe=U({add_:r1});function a1(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,pu)}var Zc=U({floorDiv_:a1});function s1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=at(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return Zc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,ou,o)}var Ie=U({div_:s1});function o1(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Du)}var J=U({mul_:o1});function l1(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,Fl)}var jt=U({abs_:l1});function u1(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,Wl)}var Sg=U({acos_:u1});function c1(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,Ul)}var Lg=U({acosh_:c1});function h1(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!hn(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,Bl)}var d1=U({addN_:h1});function Qc(n,t){for(var e=0;e<n.length;++e)if(n[n.length-e-1]!==t-1-e)return!1;return!0}function Ig(n,t,e){for(var i=n.length+t.length,r=[],a=0,s=0,o=0;o<i;o++)e.indexOf(o)===-1?r.push(n[a++]):r.push(t[s++]);return r}function Ag(n,t){for(var e=[],i=n.length,r=0;r<i;r++)t.indexOf(r)===-1&&e.push(n[r]);var a=t.map(function(s){return n[s]});return[e,a]}function nn(n,t){var e=t.map(function(i){return 1});return Ig(n,e,t)}function p1(n,t,e){E(Qc(t,e),function(){return n+" supports only inner-most axes for now. "+("Got axes "+t+" and rank-"+e+" input.")})}function rn(n,t){if(Qc(n,t))return null;for(var e=[],i=0;i<t;++i)n.indexOf(i)===-1&&e.push(i);return n.forEach(function(r){return e.push(r)}),e}function Ks(n){return n.map(function(t,e){return[e,t]}).sort(function(t,e){return t[1]-e[1]}).map(function(t){return t[0]})}function Un(n,t){for(var e=[],i=t-n;i<t;++i)e.push(i);return e}function f1(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","all","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=rn(u,i.rank);c!=null&&(i=ut(i,c),u=Un(u.length,i.rank));var h=o.all(i,u);if(e){var d=nn(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,wf,s)}var Tg=U({all_:f1});function m1(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","any","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=rn(u,i.rank);c!=null&&(i=ut(i,c),u=Un(u.length,i.rank));var h=o.any(i,u);if(e){var d=nn(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Sf,s)}var Ng=U({any_:m1});function g1(n,t){t===void 0&&(t=0);var e=O(n,"x","argMax"),i=function(s,o){o([e]);var l=Qe(t,e.shape),u=rn(l,e.rank);return u!=null&&(e=ut(e,u),l=Un(l.length,e.rank)),s.argMax(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,zl,a)}var xg=U({argMax_:g1});function v1(n,t){t===void 0&&(t=0);var e=O(n,"x","argMin"),i=function(s,o){o([e]),t==null&&(t=0);var l=Qe(t,e.shape),u=rn(l,e.rank);return u!=null&&(e=ut(e,u),l=Un(l.length,e.rank)),s.argMin(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,Pl,a)}var Cg=U({argMin_:v1});function y1(n){var t=O(n,"x","asin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asin(t);return r([t]),a},e,null,_l)}var Rg=U({asin_:y1});function b1(n){var t=O(n,"x","asinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asinh(t);return r([t]),a},e,null,Ml)}var Og=U({asinh_:b1});function w1(n){var t=O(n,"x","atan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atan(t);return r([t]),a},e,null,Hl)}var Eg=U({atan_:w1});function S1(n,t){var e,i=O(n,"a","atan2"),r=O(t,"b","atan2");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.atan2(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,ql)}var Dg=U({atan2_:S1});function L1(n){var t=O(n,"x","atanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atanh(t);return r([t]),a},e,null,Vl)}var kg=U({atanh_:L1});function I1(n,t,e,i,r,a){r===void 0&&(r="NHWC");var s=n[3],o=t.concat([s]),l=_a(r);return Bn(n,o,e,a,i,null,null,l)}function Vr(n,t,e,i,r,a,s){s===void 0&&(s="channelsLast");var o=$s(t),l=o[0],u=o[1],c;if(s==="channelsLast")c=[l,u,n[3],n[3]];else if(s==="channelsFirst")c=[l,u,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Bn(n,c,e,i,r,a,!1,s)}function Ha(n,t,e,i,r,a,s){s===void 0&&(s="NDHWC");var o=eh(t),l=o[0],u=o[1],c=o[2],h,d;if(s==="NDHWC")d="channelsLast",h=[l,u,c,n[4],n[4]];else if(s==="NCDHW")d="channelsFirst",h=[l,u,c,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Ma(n,h,e,i,r,!1,d,a)}function Bn(n,t,e,i,r,a,s,o){s===void 0&&(s=!1),o===void 0&&(o="channelsLast");var l=[-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3];if(o==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3];else if(o==="channelsFirst")u=n[0],d=n[1],c=n[2],h=n[3];else throw new Error("Unknown dataFormat "+o);var p=t[0],f=t[1],m=t[3],g=$s(e),v=g[0],b=g[1],w=$s(i),S=w[0],L=w[1],N=qr(p,S),C=qr(f,L),R=A1(r,c,h,v,b,N,C,a,o),D=R.padInfo,k=R.outHeight,W=R.outWidth,F=s?m*d:m,P;return o==="channelsFirst"?P=[u,F,k,W]:o==="channelsLast"&&(P=[u,k,W,F]),{batchSize:u,dataFormat:o,inHeight:c,inWidth:h,inChannels:d,outHeight:k,outWidth:W,outChannels:F,padInfo:D,strideHeight:v,strideWidth:b,filterHeight:p,filterWidth:f,effectiveFilterHeight:N,effectiveFilterWidth:C,dilationHeight:S,dilationWidth:L,inShape:n,outShape:P,filterShape:t}}function Ma(n,t,e,i,r,a,s,o){a===void 0&&(a=!1),s===void 0&&(s="channelsLast");var l=[-1,-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3],p=l[4];if(s==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3],p=n[4];else if(s==="channelsFirst")u=n[0],p=n[1],c=n[2],h=n[3],d=n[4];else throw new Error("Unknown dataFormat "+s);var f=t[0],m=t[1],g=t[2],v=t[4],b=eh(e),w=b[0],S=b[1],L=b[2],N=eh(i),C=N[0],R=N[1],D=N[2],k=qr(f,C),W=qr(m,R),F=qr(g,D),P=T1(r,c,h,d,w,S,L,k,W,F,o),H=P.padInfo,_=P.outDepth,j=P.outHeight,K=P.outWidth,q=a?v*p:v,G;return s==="channelsFirst"?G=[u,q,_,j,K]:s==="channelsLast"&&(G=[u,_,j,K,q]),{batchSize:u,dataFormat:s,inDepth:c,inHeight:h,inWidth:d,inChannels:p,outDepth:_,outHeight:j,outWidth:K,outChannels:q,padInfo:H,strideDepth:w,strideHeight:S,strideWidth:L,filterDepth:f,filterHeight:m,filterWidth:g,effectiveFilterDepth:k,effectiveFilterHeight:W,effectiveFilterWidth:F,dilationDepth:C,dilationHeight:R,dilationWidth:D,inShape:n,outShape:G,filterShape:t}}function N1(n,t,e,i,r){i==null&&(i=th(n,t,e));var a=n[0],s=n[1],o=Xi((a-t+2*i)/e+1,r);E(rt(o),function(){return"The output # of rows ("+o+") must be an integer. Change the stride and/or zero pad parameters"});var l=Xi((s-t+2*i)/e+1,r);return E(rt(l),function(){return"The output # of columns ("+l+") must be an integer. Change the stride and/or zero pad parameters"}),[o,l]}function x1(n,t,e,i,r,a){r==null&&(r=th(n,t,i));var s=n[0],o=n[1],l=n[2],u=Xi((s-t+2*r)/i+1,a);E(rt(u),function(){return"The output # of depths ("+u+") must be an integer. Change the stride and/or zero pad parameters"});var c=Xi((o-t+2*r)/i+1,a);E(rt(c),function(){return"The output # of rows ("+c+") must be an integer. Change the stride and/or zero pad parameters"});var h=Xi((l-t+2*r)/i+1,a);return E(rt(h),function(){return"The output # of columns ("+h+") must be an integer. Change the stride and/or zero pad parameters"}),[u,c,h,e]}function th(n,t,e,i){i===void 0&&(i=1);var r=qr(t,i);return Math.floor((n[0]*(e-1)-e+r)/2)}function $s(n){return typeof n=="number"?[n,n,n]:n.length===2?[n[0],n[1],1]:n}function eh(n){return typeof n=="number"?[n,n,n]:n}function qr(n,t){return t<=1?n:n+(n-1)*(t-1)}function A1(n,t,e,i,r,a,s,o,l){var u,c,h;if(typeof n=="number"){var d=n===0?"VALID":"NUMBER";u={top:n,bottom:n,left:n,right:n,type:d};var p=N1([t,e],a,i,n,o);c=p[0],h=p[1]}else if(n==="same"){c=Math.ceil(t/i),h=Math.ceil(e/r);var f=Math.max(0,(c-1)*i+a-t),m=Math.max(0,(h-1)*r+s-e),g=Math.floor(f/2),v=f-g,b=Math.floor(m/2),w=m-b;u={top:g,bottom:v,left:b,right:w,type:"SAME"}}else if(n==="valid")u={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-a+1)/i),h=Math.ceil((e-s+1)/r);else if(typeof n=="object"){var S=l==="channelsLast"?n[1][0]:n[2][0],v=l==="channelsLast"?n[1][1]:n[2][1],b=l==="channelsLast"?n[2][0]:n[3][0],w=l==="channelsLast"?n[2][1]:n[3][1],d=S===0&&v===0&&b===0&&w===0?"VALID":"EXPLICIT";u={top:S,bottom:v,left:b,right:w,type:d},c=Xi((t-a+S+v)/i+1,o),h=Xi((e-s+b+w)/r+1,o)}else throw Error("Unknown padding parameter: "+n);return{padInfo:u,outHeight:c,outWidth:h}}function T1(n,t,e,i,r,a,s,o,l,u,c){var h,d,p,f;if(typeof n=="number"){var m=n===0?"VALID":"NUMBER";h={top:n,bottom:n,left:n,right:n,front:n,back:n,type:m};var g=x1([t,e,i,1],o,1,r,n,c);d=g[0],p=g[1],f=g[2]}else if(n==="same"){d=Math.ceil(t/r),p=Math.ceil(e/a),f=Math.ceil(i/s);var v=(d-1)*r+o-t,b=(p-1)*a+l-e,w=(f-1)*s+u-i,S=Math.floor(v/2),L=v-S,N=Math.floor(b/2),C=b-N,R=Math.floor(w/2),D=w-R;h={top:N,bottom:C,left:R,right:D,front:S,back:L,type:"SAME"}}else if(n==="valid")h={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),p=Math.ceil((e-l+1)/a),f=Math.ceil((i-u+1)/s);else throw Error("Unknown padding parameter: "+n);return{padInfo:h,outDepth:d,outHeight:p,outWidth:f}}function Xi(n,t){if(!t)return n;switch(t){case"round":return Math.round(n);case"ceil":return Math.ceil(n);case"floor":return Math.floor(n);default:throw new Error("Unknown roundingMode "+t)}}function Si(n){var t=$s(n),e=t[0],i=t[1],r=t[2];return e===1&&i===1&&r===1}function Ht(n,t){return Si(n)||Si(t)}function _a(n){if(n==="NHWC")return"channelsLast";if(n==="NCHW")return"channelsFirst";throw new Error("Unknown dataFormat "+n)}function C1(n,t,e,i,r){var a=O(n,"x","avgPool","float32"),s=1;E(Ht(e,s),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")});var o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in avgPool: x must be rank 4 but got rank "+o.rank+"."}),r!=null&&E(rt(i),function(){return"Error in avgPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Vr(o.shape,t,e,1,i,r);return f([o]),m.filterWidth===1&&m.filterHeight===1&&hn(m.inShape,m.outShape)?o.clone():p.avgPool(o,m)},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,Gl,h);return d=ue(d,a.dtype),l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var nh=U({avgPool_:C1});function R1(n,t,e,i,r,a,s){a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:At("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","avgPool3d","float32"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in avgPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in avgPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Ht(e,s),function(){return"Error in avgPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in avgPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ha(l.shape,t,e,s,i,r,a);return m([l]),f.avgPool3d(l,g)},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Yl,d);return p=ue(p,l.dtype),u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var O1=U({avgPool3d_:R1});function Fg(n,t){var e=n[0].length;n.forEach(function(r,a){E(r.length===e,function(){return"Error in concat"+e+"D: rank of tensors["+a+"] must be the same "+("as the rank of the rest ("+e+")")})}),E(t>=0&&t<e,function(){return"Error in concat"+e+"D: axis must be between 0 and "+(e-1)+"."});var i=n[0];n.forEach(function(r,a){for(var s=0;s<e;s++)E(s===t||r[s]===i[s],function(){return"Error in concat"+e+"D: Shape of tensors["+a+"] ("+r+") "+("does not match the shape of the rest ("+i+") ")+("along the non-concatenated axis "+a+".")})})}function Wg(n,t){for(var e=n[0].slice(),i=1;i<n.length;i++)e[t]+=n[i][t];return e}function E1(n,t){t===void 0&&(t=0),E(n.length>=1,function(){return"Pass at least one tensor to concat"});var e=za(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
+          with dtype `+s.dtype+". ")});var i=function(s,o){var l=Qe(t,e[0].shape)[0],u=Wg(e.map(function(d){return d.shape}),l);if(ot(u)===0)return vi([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Fg(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,Zl,a)}var Rt=U({concat_:E1});function D1(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,ec)}var Ji=U({sigmoid_:D1});function k1(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=Kc(i,t,e),c=u[0],h=u[1];return ag(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,Xu,s)}var ze=U({slice_:k1});function F1(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,cc)}var Xs=U({tanh_:F1});function W1(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=Rt([u,h],1),p=We(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),N=fe(J(Ji(b),Xs(w)),J(c,Ji(fe(s,S)))),C=J(Xs(N),Ji(L));return[N,C]}var U1=U({basicLSTMCell_:W1});function B1(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length  "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,Kl,o)}var Js=U({batchToSpaceND_:B1});function z1(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function P1(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=z1(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Zs(o),Zs(l),Zs(c),Zs(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,fu,f);return V(m,s.shape)}function Zs(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Va=U({batchNorm_:P1});function _1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Va(s,o,l,c,u,a)}var M1=U({batchNorm2d_:_1});function H1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Va(s,o,l,c,u,a)}var V1=U({batchNorm3d_:H1});function q1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Va(s,o,l,c,u,a)}var G1=U({batchNorm4d_:q1});function Y1(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.length<e.rank)throw new Error("broadcastTo(): shape.length="+t.length+" < input.rank="+e.rank+".");if(t.length>e.rank){for(var r=e.shape.slice();r.length<t.length;)r.unshift(1);e=V(e,r)}for(var a=e.shape,s=Array.from(t),o=t.length-1;o>=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return $i(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,$l,h)}var Qs=U({broadcastTo_:Y1});function j1(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,Xl)}var Ug=U({ceil_:j1});function K1(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,Jl,a)}var Bg=U({clipByValue_:K1});function $1(n){return Rt(n,0)}var X1=U({concat1d_:$1});function J1(n,t){return Rt(n,t)}var Z1=U({concat2d_:J1});function Q1(n,t){return Rt(n,t)}var eN=U({concat3d_:Q1});function tN(n,t){return Rt(n,t)}var nN=U({concat4d_:tN});function iN(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(Ht(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=_a(r),w=Bn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,Ql,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Gr=U({conv2d_:iN});function rN(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(Ht(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Gr(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var zg=U({conv1d_:rN});function aN(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=_a(a),S=Bn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,eu,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var ih=U({conv2DBackpropInput_:aN});function sN(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return ih(e,s,o,i,r,"NHWC",a)}var Pg=U({conv2dTranspose_:sN});function oN(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(Ht(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=Ma(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,tu,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var lN=U({conv3d_:oN});function uN(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=Ma(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,xf,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var _g=U({conv3DBackpropInput_:uN});function cN(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return _g(e,a,s,i,r)}var hN=U({conv3dTranspose_:cN});function dN(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,nu)}var eo=U({cos_:dN});function pN(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,iu)}var rh=U({cosh_:pN});function fN(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=rn([t],r.rank),h=r;c!=null&&(h=ut(r,c));var d=Un(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=Ks(c);p=ut(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,ru,o)}var ah=U({cumsum_:fN});function mN(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
     `+r+" and "+t+`  for depthToSpace with input shape
     `+i.shape}),E(a*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
     `+a+" and "+t+` for depthToSpace with input shape
-        `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,Rf,u)}var Mg=U({depthToSpace_:mN});function gN(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(rt(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(Ht(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Bn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,ru,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var qa=U({depthwiseConv2d_:gN});function vN(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Df)}var yN=U({diag_:vN});function bN(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(au,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Hg=U({dilation2d_:bN});function wN(n,t){for(var e=n.length,i=[],r=0;r<e;r++){var a=e-1-r,s=n[a]||1,o=t[t.length-1-r]||1;o>1&&s===1&&i.unshift(a)}return i}function gt(n,t){for(var e=[],i=0;i<t.length;i++){var r=n[n.length-i-1],a=t.length-i-1,s=t[a];(r==null||r===1&&s>1)&&e.unshift(a)}return e}function et(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;r<i;r++){var a=n[n.length-r-1];a==null&&(a=1);var s=t[t.length-r-1];if(s==null&&(s=1),a===1)e.unshift(s);else if(s===1)e.unshift(a);else if(a!==s){var o="Operands could not be broadcast together with shapes "+(n+" and "+t+".");throw Error(o)}else e.unshift(a)}return e}function SN(n,t){var e,i=O(n,"a","equal"),r=O(t,"b","equal");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.equal(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Uf)}var Ga=U({equal_:SN});function LN(n,t,e){var i=O(t,"a","where"),r=O(e,"b","where"),a=O(n,"condition","where","bool"),s=et(i.shape,r.shape),o=Qs(i,s),l=Qs(r,s);a.rank===1&&E(a.shape[0]===i.shape[0],function(){return"The first dimension of `a` must match the size of `condition`."}),a.rank!==1&&Be(a.shape,l.shape,"Error in where: ");var u=function(h,d){var p=h.select(a,o,l);return d([a]),p},c={condition:a,t:o,e:l};return z.runKernelFunc(u,c,null,ju)}var yn=U({where_:LN});function IN(n){var t=O(n,"x","zerosLike"),e={x:t};return z.runKernelFunc(function(i){return i.zerosLike(t)},e,null,fc)}var Ee=U({zerosLike_:IN});function AN(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");e=at(i,r),i=e[0],r=e[1];var a=Ie(i,r),s=Ee(a),o=Ga(r,s);return yn(o,s,a)}var Vg=U({divNoNan_:AN});function TN(n,t){var e=O(n,"t1","dot"),i=O(t,"t2","dot");E((e.rank===1||e.rank===2)&&(i.rank===1||i.rank===2),function(){return"Error in dot: inputs must all be rank 1 or 2, but got ranks "+(e.rank+" and "+i.rank+".")});var r=e.rank===1?e.size:e.shape[1],a=i.rank===1?i.size:i.shape[0];if(E(r===a,function(){return"Error in dot: inner dimensions of inputs must match, but got "+(r+" and "+a+".")}),e.rank===1&&i.rank===1){var s=V(e,[1,-1]),o=V(i,[-1,1]),l=We(s,o);return V(l,[])}else if(e.rank===1&&i.rank===2){var s=V(e,[1,-1]),o=V(i,[i.shape[0],i.shape[1]]),l=We(s,o);return V(l,[l.size])}else if(e.rank===2&&i.rank===1){var o=V(i,[-1,1]),l=We(e,o);return V(l,[l.size])}else{var o=V(i,[i.shape[0],i.shape[1]]),l=We(e,o);return l}}var qg=U({dot_:TN});function NN(n){var t=O(n,"x","elu"),e=function(r,a){var s=r.elu(t);return a([s]),s},i={x:t};return z.runKernelFunc(e,i,null,ou)}var ah=U({elu_:NN});function xN(n){var t=O(n,"x","erf");E(t.dtype==="int32"||t.dtype==="float32",function(){return"Input dtype must be `int32` or `float32`."}),t.dtype==="int32"&&(t=ue(t,"float32"));var e={x:t};return z.runKernelFunc(function(i,r){var a=i.erf(t);return r([t]),a},e,null,lu)}var Gg=U({erf_:xN});function CN(n){var t=O(n,"x","exp"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.exp(t);return r([a]),a},e,null,uu)}var bn=U({exp_:CN});function RN(n,t){t===void 0&&(t=0);var e=null,i=O(n,"x","expandDims",e);E(t<=i.rank,function(){return"Axis must be <= rank of the tensor"});var r=i.shape.slice();return t<0&&(E(-(i.rank+1)<=t,function(){return"Axis must be in the interval ["+-(i.rank+1)+", "+i.rank+"]"}),t=i.rank+t+1),r.splice(t,0,1),V(i,r)}var wn=U({expandDims_:RN});function ON(n){var t=O(n,"x","expm1"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.expm1(t);return r([t]),a},e,null,cu)}var Yg=U({expm1_:ON});function EN(n,t){var e=null,i=O(n,"x","tile",e);E(i.rank===t.length,function(){return"Error in transpose: rank of input "+i.rank+" "+("must match length of reps "+t+".")});var r=function(l,u){var c=l.tile(i,t);return u([i]),c},a=[i],s={x:i},o={reps:t};return z.runKernelFunc(r,s,null,cc,o,a)}var Yr=U({tile_:EN});function DN(n,t,e,i){i===void 0&&(i="float32"),t==null&&(t=n);for(var r=Wn([n,t],i),a=n<=t?n:t,s=0;s<a;++s)r.set(1,s,s);var o=V(r.toTensor(),[n,t]);if(e==null)return o;if(e.length===1)return Yr(wn(o,0),[e[0],1,1]);if(e.length===2)return Yr(wn(wn(o,0),0),[e[0],e[1],1,1]);if(e.length===3)return Yr(wn(wn(wn(o,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error("eye() currently supports only 1D and 2D "+("batchShapes, but received "+e.length+"D."))}var jg=U({eye_:DN});function sh(n,t,e){var i={shape:n,value:t,dtype:e};return z.runKernelFunc(function(r){return r.fill(n,t,e)},{},null,zf,i)}function kN(n){var t=O(n,"x","floor"),e={x:t};return z.runKernelFunc(function(i){return i.floor(t)},e,null,hu)}var to=U({floor_:kN});var oh=30;function FN(n){return n<=oh?n:Ps(n,Math.floor(Math.sqrt(n)))}function WN(n,t){var e=!1,i;for(n<=oh?(i=n,e=!0):i=Ps(n,Math.floor(Math.sqrt(n)));!e;)i>t||i===n?e=!0:i=Ps(n,i+1);return i}function UN(n,t,e){for(var i=[],r=n.length,a=0;a<r;a++)a!==t?i.push(n[a]):i.push(e);return i}function Kg(n,t,e){for(var i=n.shape[e],r=[],a=1,s=1,o=0;o<e;o++)r.push(n.shape[o]),a*=n.shape[o];for(var o=0;o<t.rank;o++)r.push(t.shape[o]);for(var o=e+1;o<n.rank;o++)r.push(n.shape[o]),s*=n.shape[o];return{batchSize:a,sliceSize:s,dimSize:i,outputShape:r}}var BN={__proto__:null,segOpComputeOptimalWindowSize:WN,computeOutShape:UN,collectGatherOpShapeInfo:Kg};function zN(n,t,e){e===void 0&&(e=0);var i=O(n,"x","gather"),r=O(t,"indices","gather","int32"),a={x:i,indices:r},s={axis:e},o=function(l,u){var c=Qe(e,i.shape)[0],h=Kg(i,r,c),d=l.gather(i,V(r,[r.size]),c);return u([i,r]),V(d,h.outputShape)};return z.runKernelFunc(o,a,null,fu,s)}var no=U({gather_:zN});function PN(n,t){var e,i=O(n,"a","greater"),r=O(t,"b","greater");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.greater(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Mf)}var Li=U({greater_:PN});function _N(n,t){var e,i=O(n,"a","greaterEqual"),r=O(t,"b","greaterEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.greaterEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,mu)}var Zi=U({greaterEqual_:_N});function MN(n){var t=O(n,"input","imag"),e=function(r){return r.imag(t)},i={input:t};return z.runKernelFunc(e,i,null,Vf)}var io=U({imag_:MN});function HN(n){var t=O(n,"x","isFinite"),e={x:t};return z.runKernelFunc(function(i){return i.isFinite(t)},e,null,vu)}var $g=U({isFinite_:HN});function VN(n){var t=O(n,"x","isInf"),e={x:t};return z.runKernelFunc(function(i){return i.isInf(t)},e,null,yu)}var Xg=U({isInf_:VN});function qN(n){var t=O(n,"x","isNaN"),e={x:t};return z.runKernelFunc(function(i){return i.isNaN(t)},e,null,bu)}var Jg=U({isNaN_:qN});function GN(n,t){var e,i=O(n,"a","maximum"),r=O(t,"b","maximum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.maximum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Tu)}var jr=U({maximum_:GN});function be(n,t){if((Dt(n)&&t!=="string"||Array.isArray(n))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&Dt(n)&&!(n instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");var e=[],i=[];return gi(n,e,i,t)}function YN(n,t){t===void 0&&(t=.2);var e=O(n,"x","leakyRelu");return jr(J(be(t),e),e)}var Zg=U({leakyRelu_:YN});function jN(n,t){var e,i=O(n,"a","less"),r=O(t,"b","less");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.less(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,qf)}var ro=U({less_:jN});function KN(n,t){var e,i=O(n,"a","lessEqual"),r=O(t,"b","lessEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.lessEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Gf)}var Qi=U({lessEqual_:KN});function $N(n,t,e){if(e<=0)throw new Error("The number of values should be positive.");var i={start:n,stop:t,num:e};return z.runKernelFunc(function(r){return r.linspace(n,t,e)},{},null,Yf,i)}function XN(n,t,e,i,r){t===void 0&&(t=5),e===void 0&&(e=1),i===void 0&&(i=1),r===void 0&&(r=.5);var a=O(n,"x","localResponseNormalization");E(a.rank===4||a.rank===3,function(){return`Error in localResponseNormalization: x must be rank 3 or 4 but got
-               rank `+a.rank+"."}),E(rt(t),function(){return"Error in localResponseNormalization: depthRadius must be an "+("integer but got depthRadius "+t+".")});var s=a,o=!1;a.rank===3&&(o=!0,s=V(a,[1,a.shape[0],a.shape[1],a.shape[2]]));var l=function(d,p){var f=d.localResponseNormalization4D(s,t,e,i,r);return p([s,f]),f},u={x:s},c={depthRadius:t,bias:e,alpha:i,beta:r},h=z.runKernelFunc(l,u,null,Iu,c);return o?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Qg=U({localResponseNormalization_:XN});function JN(n){var t=O(n,"x","log"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log(t);return r([t]),a},e,null,wu)}var er=U({log_:JN});function ZN(n){var t=O(n,"x","log1p"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log1p(t);return r([t]),a},e,null,Su)}var lh=U({log1p_:ZN});function QN(n){return E(fi(n),function(){return"The f passed in grad(f) must be a function"}),function(t,e){var i=O(t,"x","tf.grad",null),r=e!=null?O(e,"dy","tf.grad"):null;return z.tidy(function(){var a=z.gradients(function(){return n(i)},[i],r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),ao(o),o[0]})}}function ex(n){return E(fi(n),function(){return"The f passed in grads(f) must be a function"}),function(t,e){E(Array.isArray(t),function(){return"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s"});var i=za(t,"args","tf.grads",null),r=e!=null?O(e,"dy","tf.grads"):null;return z.tidy(function(){var a=z.gradients(function(){return n.apply(void 0,i)},i,r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),ao(o),o})}}function tx(n){return E(fi(n),function(){return"The f passed in valueAndGrad(f) must be a function"}),function(t,e){E(t instanceof Y,function(){return"The x passed in valueAndGrad(f)(x) must be a tensor"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrad(f)(x, dy) must be a tensor"});var i=z.gradients(function(){return n(t)},[t],e),r=i.grads,a=i.value;return ao(r),{grad:r[0],value:a}}}function nx(n){return E(fi(n),function(){return"The f passed in valueAndGrads(f) must be a function"}),function(t,e){E(Array.isArray(t)&&t.every(function(r){return r instanceof Y}),function(){return"The args passed in valueAndGrads(f)(args) must be array of tensors"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrads(f)(args, dy) must be a tensor"});var i=z.gradients(function(){return n.apply(void 0,t)},t,e);return e!=null&&Be(i.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),ao(i.grads),i}}function ev(n,t){E(fi(n),function(){return"The f passed in variableGrads(f) must be a function"}),E(t==null||Array.isArray(t)&&t.every(function(h){return h instanceof Ba}),function(){return"The varList passed in variableGrads(f, varList) must be an array of variables"});var e=t!=null;if(!e){t=[];for(var i in z.registeredVariables)t.push(z.registeredVariables[i])}var r=e?t.filter(function(h){return!h.trainable}):null,a=t.length;t=t.filter(function(h){return h.trainable}),E(t.length>0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function zn(n){return z.customGrad(n)}function ao(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
-    the f you passed encloses all operations that lead from x to y.`)}function ix(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,Du)}var vt=U({neg_:ix});function rx(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,ec)}var uh=U({softplus_:rx});function ax(n){var t=O(n,"x","logSigmoid"),e=zn(function(i){var r=vt(uh(vt(i))),a=function(s){var o=J(s,Ji(vt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var tv=U({logSigmoid_:ax});function sx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=an(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=Un(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=rn(f.shape,Qe(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,Au,s)}var tr=U({max_:sx});function ox(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,oc)}var ge=U({sub_:ox});function lx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Qe(t,i.shape),c=an(u,i.rank),h=u,d=i;c!=null&&(d=ut(i,c),h=Un(h.length,i.rank));var p=o.sum(d,h);if(e){var f=rn(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,nc,s)}var Ae=U({sum_:lx});function ux(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=tr(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),er(Ae(bn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,Lu,a)}var nv=U({logSoftmax_:ux});function cx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Qe(t,i.shape),a=tr(i,r,!0),s=ge(i,a),o=bn(s),l=Ae(o,r),u=er(l),c=fe(V(a,u.shape),u);if(e){var h=rn(c.shape,r);return V(c,h)}return c}var ch=U({logSumExp_:cx});function hx(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,jf)}var nr=U({logicalAnd_:hx});function dx(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,Kf)}var so=U({logicalNot_:dx});function px(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,$f)}var hh=U({logicalOr_:px});function fx(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return et(e.shape,i.shape),nr(hh(n,t),so(nr(n,t)))}var iv=U({logicalXor_:fx});function mx(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(Ht(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Vr(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&dn(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,Nu,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var dh=U({maxPool_:mx});function gx(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:At("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Ht(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ha(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,xu,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var vx=U({maxPool3d_:gx});function yx(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(Qf,s,o);return{result:l[0],indexes:l[1]}}var bx=U({maxPoolWithArgmax_:yx});function ti(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=ti(n,"float32"),i=ti(n,"float32");return mi(e,i)}var r=Ur(ot(n),t);return z.makeTensor(r,n,t)}function Kr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Kr(n,"float32"),i=ti(n,"float32");return mi(e,i)}var r=Ac(ot(n),t);return z.makeTensor(r,n,t)}function wx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Qe(t,i.shape),a=Ag(i.shape,r),s=a[1],o=ot(s),l=zn(function(u){var c=be(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Ie(h,c),p=Ae(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Ie(J(v,Kr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var Ya=U({mean_:wx});function Sx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=an(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=Un(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=rn(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Cu,s)}var oo=U({min_:Sx});function Lx(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ru)}var lo=U({minimum_:Lx});function Ix(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ou)}var ph=U({mod_:Ix});function Ax(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ye=U({square_:Ax});function Tx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Qe(t,n.shape),r=Ya(n,i,e),a=r.shape;e||(a=rn(r.shape,i));var s=Ye(ge(ue(n,"float32"),V(r,a))),o=Ya(s,i,e);return{mean:r,variance:o}}var Nx=U({moments_:Tx});function xx(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=za(e,"c","multiRNNCell"),s=za(i,"h","multiRNNCell"),o=r,l=[],u=0;u<n.length;u++){var c=n[u](o,a[u],s[u]);l.push(c[0]),l.push(c[1]),o=c[1]}for(var h=[],d=[],u=0;u<l.length;u+=2)h.push(l[u]),d.push(l[u+1]);return[h,d]}var Cx=U({multiRNNCell_:xx});function Rx(n,t,e,i){i===void 0&&(i=!1);var r=O(n,"logits","multinomial"),a=r.size,s=r.rank;if(a<2)throw new Error("Error in multinomial: you need at least 2 outcomes, but got "+(a+"."));if(s>2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var Ox=U({multinomial_:Rx});function Ex(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,em)}var uo=U({notEqual_:Ex});function Dx(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,sm)}var ja=U({real_:Dx});function kx(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=fh(ja(t)),o=Ee(io(t));return mi(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,ku)}var fh=U({onesLike_:kx});function Fx(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return We(r,a)}var Wx=U({outerProduct_:Fx});function Ux(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,Wu,a)}var ir=U({pad_:Ux});function Bx(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),ir(n,[t],e)}var zx=U({pad1d_:Bx});function Px(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),ir(n,t,e)}var _x=U({pad2d_:Px});function Mx(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),ir(n,t,e)}var Hx=U({pad3d_:Mx});function Vx(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),ir(n,t,e)}var qx=U({pad4d_:Vx});function Gx(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,ic,s)}var co=U({spaceToBatchND_:Gx});function Kx(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(Ht(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=Vr(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=jx([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=Yx([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:co(o,c,f),b=e==="avg"?function(){return th(v,t,a,g)}:function(){return dh(v,t,a,g)},w=b(),S=d?w:Js(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function Yx(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function jx(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var rv=U({pool_:Kx});function $x(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=at(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,Uu)}var ni=U({pow_:$x});function Xx(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,Bu)}var mh=U({prelu_:Xx});function Jx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Qe(t,i.shape),u=an(l,i.rank),c=l,h=i;u!=null&&(h=ut(i,u),c=Un(c.length,i.rank));var d=o.prod(h,c);if(e){var p=rn(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,rm,s)}var av=U({prod_:Jx});function Zx(n,t,e){var i=ot(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a<i;a++)r[a]=t();return z.makeTensor(r,n,e)}var Qx=U({rand_:Zx}),$r=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function rr(n,t){return t={exports:{}},n(t,t.exports),t.exports}var eC=rr(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})($r,n,!1)}),tC=rr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})($r,n,!1)}),nC=rr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})($r,n,!1)}),iC=rr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})($r,n,!1)}),rC=rr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})($r,n,!1)}),aC=rr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})($r,n,!1)}),ar=rr(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,N){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||N||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,N=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(N||(S=[N++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%N]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,j=C.S;W--;)F=j[H=h&H+1],P=P*r+j[h&(j[H]=j[_=h&_+F])+(j[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var N=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{N.push(g(S[R],L-1))}catch(D){}return N.length?N:C=="string"?S:S+"\0"}function v(S,L){for(var N=S+"",C,R=0;R<N.length;)L[h&R]=h&(C^=L[h&R]*19)+N.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,N=L&&L.plugins;return[+new Date,i,N,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=require("crypto")}catch(S){}}})([],Math)});ar.alea=eC;ar.xor128=tC;ar.xorwow=nC;ar.xorshift7=iC;ar.xor4096=rC;ar.tychei=aC;var sC=ar,gh=sC.alea;var vh=function(){function n(t,e,i,r,a){this.mean=t,this.stdDev=e,this.dtype=i,this.nextVal=NaN,this.truncated=r,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);var s=a||Math.random();this.random=gh(s.toString())}return n.prototype.nextValue=function(){if(!isNaN(this.nextVal)){var t=this.nextVal;return this.nextVal=NaN,t}for(var e,i,r=!1;!r;){var a=void 0,s=void 0,o=void 0;do a=2*this.random()-1,s=2*this.random()-1,o=a*a+s*s;while(o>=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),oC=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=gh(a.toString()),this.randn=new vh(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a<e||Math.log(a)<i)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)},n.prototype.convertValue=function(t){return this.dtype==="float32"?t:Math.round(t)},n}(),lC=function(){function n(t,e,i,r){var a=this;if(t===void 0&&(t=0),e===void 0&&(e=1),this.canReturnFloat=function(){return a.dtype==null||a.dtype==="float32"},this.min=t,this.range=e-t,this.dtype=i,r==null&&(r=Math.random()),typeof r=="number"&&(r=r.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error("The difference between "+t+" - "+e+" <= 1 and dtype is not float");this.random=gh(r)}return n.prototype.convertValue=function(t){return this.canReturnFloat()?t:Math.round(t)},n.prototype.nextValue=function(){return this.convertValue(this.min+this.range*this.random())},n}();function uC(n,t,e,i,r){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e==null&&(e=1),i==null&&(i="float32"),i!=="float32"&&i!=="int32")throw new Error("Unsupported data type "+i);for(var a=new oC(t,e,i,r),s=Wn(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var cC=U({randomGamma_:uC});function hC(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type "+i);for(var a=new vh(t,e,i,!1,r),s=Wn(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var dC=U({randomNormal_:hC});function pC(n,t,e,i,r){t===void 0&&(t=0),e===void 0&&(e=1),i===void 0&&(i="float32");for(var a=Wn(n,i),s=new lC(t,e,null,r),o=0;o<a.values.length;o++)a.values[o]=s.nextValue();return a.toTensor()}var sv=U({randomUniform_:pC});function Xr(n,t){Yi(n);var e=Fn(n,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");var i=null;return gi(n,i,e,t)}function yh(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e===0)throw new Error("Cannot have a step of zero");var r=function(){var s=n===t,o=n<t&&e<0,l=t<n&&e>1;if(s||o||l)return ti([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=Ur(u,i);t<n&&e===1&&(e=-1),c[0]=n;for(var h=1;h<c.length;h++)c[h]=c[h-1]+e;return Xr(c,i)},a={start:n,stop:t,step:e,dtype:i};return z.runKernelFunc(r,{},null,am,a)}function fC(n){var t=O(n,"x","reciprocal"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.reciprocal(t);return r([t]),a},e,null,zu)}var ov=U({reciprocal_:fC});function mC(n){var t=O(n,"x","relu"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu(t)},i={x:t};return z.runKernelFunc(e,i,null,Pu)}var Ka=U({relu_:mC});function gC(n){var t=O(n,"x","relu6"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu6(t)},i={x:t};return z.runKernelFunc(e,i,null,Vu)}var bh=U({relu6_:gC});function vC(n,t){var e=O(n,"x","reverse"),i=function(s){var o=Qe(t,e.shape);if(e.rank===0)return $i(e);var l=s.reverse(e,o);return V(l,e.shape)},r={x:e},a={dims:t};return z.runKernelFunc(i,r,null,qu,a)}var ii=U({reverse_:vC});function yC(n){var t=O(n,"x","reverse");return E(t.rank===1,function(){return"Error in reverse1D: x must be rank 1 but got rank "+t.rank+"."}),ii(t,0)}var bC=U({reverse1d_:yC});function wC(n,t){var e=O(n,"x","reverse");return E(e.rank===2,function(){return"Error in reverse2D: x must be rank 2 but got rank "+e.rank+"."}),ii(e,t)}var SC=U({reverse2d_:wC});function LC(n,t){var e=O(n,"x","reverse");return E(e.rank===3,function(){return"Error in reverse3D: x must be rank 3 but got rank "+e.rank+"."}),ii(e,t)}var IC=U({reverse3d_:LC});function AC(n,t){var e=O(n,"x","reverse");return E(e.rank===4,function(){return"Error in reverse4D: x must be rank 4 but got rank "+e.rank+"."}),ii(e,t)}var TC=U({reverse4d_:AC});function NC(n){var t=O(n,"x","round"),e={x:t};return z.runKernelFunc(function(i){return i.round(t)},e,null,Gu)}var lv=U({round_:NC});function xC(n){var t=O(n,"x","rsqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.rsqrt(t);return r([t]),a},e,null,Yu)}var wh=U({rsqrt_:xC});function CC(n){var t=O(n,"x","selu"),e=function(r,a){var s=r.selu(t);return a([t]),s},i={x:t};return z.runKernelFunc(e,i,null,Ku)}var uv=U({selu_:CC});function RC(n,t,e,i,r,a,s){a===void 0&&(a=[1,1]),s===void 0&&(s="NHWC");var o=O(n,"x","separableConv2d"),l=O(t,"depthwiseFilter","separableConv2d"),u=O(e,"pointwiseFilter","separableConv2d"),c=o,h=!1;if(o.rank===3&&(h=!0,c=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),s==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");E(c.rank===4,function(){return"Error in separableConv2d: input must be rank 4, but got "+("rank "+c.rank+".")}),E(l.rank===4,function(){return"Error in separableConv2d: depthwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.rank===4,function(){return"Error in separableConv2d: pointwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.shape[0]===1,function(){return"Error in separableConv2d: the first dimension of pointwise filter "+(" must be 1, but got "+u.shape[0]+".")}),E(u.shape[1]===1,function(){return"Error in separableConv2d: the second dimension of pointwise "+("filter must be 1, but got "+u.shape[1]+".")});var d=l.shape[2],p=l.shape[3];E(u.shape[2]===d*p,function(){return"Error in separableConv2d: the third dimension of pointwise filter "+("must be "+d*p+", ")+("but got "+u.shape[2]+".")});var f=qa(c,l,i,r,s,a),m=1,g=Gr(f,u,m,"valid",s);return h?V(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var cv=U({separableConv2d_:RC});function OC(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,l,h;return pe(this,function(d){switch(d.label){case 0:return e=O(n,"x","setdiff1d"),i=O(t,"y","setdiff1d"),E(e.dtype===i.dtype,function(){return"x and y should have the same dtype, but got x ("+e.dtype+") and y ("+i.dtype+")."}),E(e.rank===1,function(){return"x should be 1D tensor, but got x ("+e.shape+")."}),E(i.rank===1,function(){return"y should be 1D tensor, but got y ("+i.shape+")."}),[4,e.data()];case 1:return r=d.sent(),[4,i.data()];case 2:for(a=d.sent(),s=new Set(a),o=0,l=0;l<r.length;l++)s.has(r[l])||o++;for(u=new Hs([o],e.dtype),c=new Hs([o],"int32"),l=0,h=0;l<r.length;l++)s.has(r[l])||(u.values[h]=r[l],c.values[h]=l,h++);return[2,[u.toTensor(),c.toTensor()]]}})})}var EC=OC;function DC(n){var t=O(n,"x","sign"),e={x:t};return z.runKernelFunc(function(i){return i.sign(t)},e,null,Zu)}var hv=U({sign_:DC});function kC(n){var t=O(n,"x","sin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sin(t);return r([t]),a},e,null,Xu)}var Sh=U({sin_:kC});function FC(n){var t=O(n,"x","sinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sinh(t);return r([t]),a},e,null,Ju)}var Lh=U({sinh_:FC});function WC(n,t,e){var i=O(n,"x","slice1d");return E(i.rank===1,function(){return"slice1d expects a rank-1 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,[t],[e])}var UC=U({slice1d_:WC});function BC(n,t,e){var i=O(n,"x","slice2d");return E(i.rank===2,function(){return"slice2d expects a rank-2 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var zC=U({slice2d_:BC});function PC(n,t,e){var i=O(n,"x","slice3d");return E(i.rank===3,function(){return"slice3d expects a rank-3 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var _C=U({slice3d_:PC});function MC(n,t,e){var i=O(n,"x","slice4d");return E(i.rank===4,function(){return"slice4d expects a rank-4 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var HC=U({slice4d_:MC});function VC(n,t){t===void 0&&(t=-1);var e=O(n,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and dim was "+t));var i={logits:e},r={dim:t};return z.runKernelFunc(function(a,s){var o=a.softmax(e,t);return s([o]),o},i,null,ac,r)}var dv=U({softmax_:VC});function qC(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.fft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=n.as2D(r,i),s=e.fft(a);return s.reshape(n.shape)},t,null,Bf)}var ho=U({fft_:qC});function GC(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.ifft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=V(n,[r,i]),s=e.ifft(a);return V(s,n.shape)},t,null,Hf)}var $a=U({ifft_:GC});function YC(n){var t=n.shape[n.shape.length-1],e=n.size/t,i;if(t<=2){var r=V(n,[e,t]);i=$a(r)}else{var a=[e,2*(t-1)],s=V(ja(n),[e,t]),o=V(io(n),[e,t]),l=ii(ze(s,[0,1],[e,t-2]),1),u=J(ii(ze(o,[0,1],[e,t-2]),1),be(-1)),c=Rt([s,l],1),h=Rt([o,u],1),r=V(mi(c,h),[a[0],a[1]]);i=$a(r)}if(i=ja(i),n.rank===3&&n.shape[0]!==0){var d=i,p=n.shape[0];i=V(i,[p,i.shape[0]/p,i.shape[1]]),d.dispose()}return i}var Ih=U({irfft_:YC});function pv(n,t,e){e===void 0&&(e=0);var i=[];if(typeof t=="number")E(n.shape[e]%t===0,function(){return"Number of splits must evenly divide the axis."}),i=new Array(t).fill(n.shape[e]/t);else{var r=t.reduce(function(o,l){return l===-1&&(o+=1),o},0);E(r<=1,function(){return"There should be only one negative value in split array."});var a=t.indexOf(-1);if(a!==-1){var s=t.reduce(function(o,l){return l>0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function jC(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Qe(e,i.shape)[0],c=pv(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,rc,s)}var Jr=U({split_:jC});function KC(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&t<e){var a=n.shape.map(function(v){return 0}),s=n.shape.map(function(v){return v});s[n.shape.length-1]=t,r=ze(n,a,s),e=t}else if(t!=null&&t>e){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=Rt([n,ti(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(mi(r,l),[i,e]),c=ho(u),h=Math.floor(e/2)+1,d=ja(c),p=io(c),f=Jr(d,[h,e-h],d.shape.length-1),m=Jr(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(mi(f[0],m[0]),g)}var po=U({rfft_:KC});function $C(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,tc)}var Vt=U({sqrt_:$C});function XC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,sc,o)}var fo=U({squaredDifference_:XC});function JC(n,t){var e=O(n,"x","squeeze");return V(e,wm(e.shape,t).newShape)}var mo=U({squeeze_:JC});function ZC(n,t){t===void 0&&(t=0);var e=za(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return wn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Be(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return wn(o,t)});return Rt(s,t)}var sr=U({stack_:ZC});function QC(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,mc,r)}var Zr=U({step_:QC});function eR(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=js(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=js(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=mg(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var N=js(l);N.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=sg(t,e,i),R=C.filter(function(W,F){return N.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,dm,d)}var fv=U({stridedSlice_:eR});function tR(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,lc)}var mv=U({tan_:tR});function Xa(n,t,e){if(Yi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=Fn(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return gi(n,t,i,e)}function nR(n,t,e){if(Yi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=Fn(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return gi(n,t,i,e)}function iR(n,t,e){if(Yi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=Fn(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return gi(n,t,i,e)}function rR(n,t,e){if(Yi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=Fn(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,gi(n,t,i,e)}function aR(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,pm,s),l=o[0],u=o[1];return{values:l,indices:u}}var gv=U({topk_:aR});function sR(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new vh(t,e,i,!0,r),s=Wn(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var oR=U({truncatedNormal_:sR});function lR(n,t){t===void 0&&(t=0);var e=O(n,"x","unique",null);E(e.rank>0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(fm,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var vv=U({unique_:lR});function uR(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(rt(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,pc,s)}var Ah=U({unsortedSegmentSum_:uR});function cR(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t<e.shape.length,function(){return"Axis = "+t+" is not in [-"+e.shape.length+", "+e.shape.length+")"}),t<0&&(t+=e.shape.length);var i={value:e},r={axis:t},a=function(s){return s.unstack(e,t)};return z.runKernelFunc(a,i,null,dc,r)}var go=U({unstack_:cR});function hR(n,t,e,i){return t===void 0&&(t=!0),z.makeVariable(n,t,e,i)}function yv(n,t){for(var e=[],i=0;i<t.length;i++)t[i]&&e.push(i);for(var r=Wn(n,"int32"),a=Wn([e.length,n.length],"int32"),i=0;i<e.length;i++){var s=r.indexToLoc(e[i]),o=i*n.length;a.values.set(s,o)}return a.toTensor()}function dR(n){return de(this,void 0,void 0,function(){var t,e,i;return pe(this,function(r){switch(r.label){case 0:return t=O(n,"condition","whereAsync","bool"),[4,t.data()];case 1:return e=r.sent(),i=yv(t.shape,e),n!==t&&t.dispose(),[2,i]}})})}var bv=dR;function pR(n,t,e){return de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:for(i=O(n,"tensor","boolMask"),r=O(t,"mask","boolMask","bool"),a=e??0,s=r.rank,o=i.shape,E(s>0,function(){return"mask cannot be scalar"}),Be(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u<a+s;u++)l*=o[u];return c=o.slice(0,a).concat([l],o.slice(a+s)),h=V(i,c),d=V(r,[-1]),[4,bv(d)];case 1:return p=g.sent(),f=mo(p,[1]),m=no(h,f,a),n!==i&&i.dispose(),t!==r&&r.dispose(),f.dispose(),h.dispose(),d.dispose(),p.dispose(),[2,m]}})})}var fR=pR;function mR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","notEqualStrict"),i=O(t,"b","notEqualStrict");return Be(e.shape,i.shape,"Error in notEqualStrict: "),uo(e,i)}function gR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessStrict"),i=O(t,"b","lessStrict");return Be(e.shape,i.shape,"Error in lessStrict: "),ro(e,i)}function vR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","equalStrict"),i=O(t,"b","equalStrict");return Be(e.shape,i.shape,"Error in equalStrict: "),Ga(e,i)}function yR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessEqualStrict"),i=O(t,"b","lessEqualStrict");return Be(e.shape,i.shape,"Error in lessEqualStrict: "),Qi(e,i)}function bR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterStrict"),i=O(t,"b","greaterStrict");return Be(e.shape,i.shape,"Error in greaterStrict: "),Li(e,i)}function wR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterEqualStrict"),i=O(t,"b","greaterEqualStrict");return Be(e.shape,i.shape,"Error in greaterEqualStrict: "),Zi(e,i)}var wv=U({equalStrict_:vR}),Sv=U({greaterEqualStrict_:wR}),Lv=U({greaterStrict_:bR}),Iv=U({lessEqualStrict_:yR}),Av=U({lessStrict_:gR}),Tv=U({notEqualStrict_:mR});function SR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","addStrict"),i=O(t,"b","addStrict");return Be(e.shape,i.shape,"Error in addStrict: "),fe(e,i)}function LR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","subStrict"),i=O(t,"b","subStrict");return Be(e.shape,i.shape,"Error in subStrict: "),ge(e,i)}function IR(n,t){return At("strict variants of ops have been deprecated and will be removed in future"),Be(n.shape,t.shape,"Error in powStrict: "),ni(n,t)}function AR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","mul"),i=O(t,"b","mul");return Be(e.shape,i.shape,"Error in multiplyStrict: "),J(e,i)}function TR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","div"),i=O(t,"b","div");return Be(e.shape,i.shape,"Error in divideStrict: "),Ie(e,i)}function NR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","modStrict"),i=O(t,"b","modStrict");return Be(e.shape,i.shape,"Error in modStrict: "),ph(e,i)}function xR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","minimumStrict"),i=O(t,"b","minimumStrict");return Be(e.shape,i.shape,"Error in minimumStrict: "),lo(e,i)}function CR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","maximumStrict"),i=O(t,"b","maximumStrict");return Be(e.shape,i.shape,"Error in maximumStrict: "),jr(e,i)}function RR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","squaredDifferenceStrict"),i=O(t,"b","squaredDifferenceStrict");return Be(e.shape,i.shape,"Error in squaredDifferenceStrict: "),fo(e,i)}var Nv=U({addStrict_:SR}),xv=U({divStrict_:TR}),Cv=U({maximumStrict_:CR}),Rv=U({minimumStrict_:xR}),Ov=U({modStrict_:NR}),Ev=U({mulStrict_:AR}),Dv=U({powStrict_:IR}),kv=U({squaredDifferenceStrict_:RR}),Fv=U({subStrict_:LR});function OR(n,t,e,i){t===void 0&&(t="euclidean"),e===void 0&&(e=null),i===void 0&&(i=!1),n=O(n,"x","norm");var r=Wv(n,t,e),a=r.shape;if(i){var s=Qe(e,n.shape);a=rn(r.shape,s)}return V(r,a)}function Wv(n,t,e){if(e===void 0&&(e=null),n.rank===0)return Kt(n);if(n.rank!==1&&e===null)return Wv(V(n,[-1]),t,e);if(n.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return Ae(Kt(n),e);if(t===Infinity)return tr(Kt(n),e);if(t===-Infinity)return oo(Kt(n),e);if(t==="euclidean"||t===2)return Vt(Ae(ni(Kt(n),be(2,"int32")),e));throw new Error("Error in norm: invalid ord value: "+t)}if(Array.isArray(e)&&e.length===2){if(t===1)return tr(Ae(Kt(n),e[0]),e[1]-1);if(t===Infinity)return tr(Ae(Kt(n),e[1]),e[0]);if(t===-Infinity)return oo(Ae(Kt(n),e[1]),e[0]);if(t==="fro"||t==="euclidean")return Vt(Ae(Ye(n),e));throw new Error("Error in norm: invalid ord value: "+t)}throw new Error("Error in norm: invalid axis: "+e)}var vo=U({norm_:OR});function ER(n,t,e,i,r){r===void 0&&(r=!0);var a=O(n,"v","movingAverage"),s=O(t,"x","movingAverage"),o=O(e,"decay","movingAverage");Dm(a,s),E(dn(a.shape,s.shape),function(){return"Shape mismatch in v and x"});var l=be(1),u=ge(l,o),c=J(ge(s,a),u);if(r){E(i!=null,function(){return"When using zeroDebias: true, step is required."});var h=O(i,"step","movingAverage");c=Ie(c,ge(l,ni(o,h)))}return fe(a,c)}var DR=U({movingAverage_:ER});function kR(n,t,e){var i=O(n,"indices","scatterND","int32"),r=O(t,"updates","scatterND");Yc(r,i,e);var a=function(l){return l.scatterND(i,r,e)},s={indices:i,updates:r},o={shape:e};return z.runKernelFunc(a,s,null,um,o)}var FR=U({scatterND_:kR});function WR(n,t,e,i){if(n.dtype!=="int32")throw new Error("tf.sparseToDense() expects the indices to be int32 type,"+(" but the dtype was "+n.dtype+"."));if(n.rank>2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function UR(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);WR(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,hm,l)}var BR=U({sparseToDense_:UR});function zR(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,_f)}var PR=U({gatherND_:zR});function _R(n,t){if(t==null)return n.shape.slice();if(dn(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i<n.shape.length;i++)t[i]==null&&n.shape[i]!=null?e.push(n.shape[i]):e.push(t[i]);return e}return t}function MR(n,t,e,i){var r=O(n,"x","dropout");if(E(r.dtype==="float32",function(){return"x has to be a floating point tensor since it's going to be "+("scaled, but got a "+r.dtype+" tensor instead.")}),E(t>=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=_R(r,e),s=1-t,o=Ie(to(fe(sv(a,0,1,"float32",i),s)),s);return J(r,o)}var HR=U({dropout_:MR});function Uv(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function Th(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a<n;++a){var s=2*Math.PI*a/(n+i-1);r[a]=t-e*Math.cos(s)}return Xr(r,"float32")}function VR(n,t,e){return e===void 0&&(e=1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,g;return pe(this,function(v){switch(v.label){case 0:return i=O(n,"predictions","inTopK"),r=O(t,"targets","inTopK"),E(i.rank>1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Be(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=Bs("bool",u),d=0;d<u;d++){for(p=d*c,f=s.subarray(p,p+c),m=[],g=0;g<f.length;g++)m.push({value:f[g],index:g});for(m.sort(function(b,w){return w.value-b.value}),h[d]=0,g=0;g<e;g++)if(m[g].index===o[d]){h[d]=1;break}}return n!==i&&i.dispose(),t!==r&&r.dispose(),[2,vi(h,r.shape,"bool")]}})})}var qR=VR;function GR(n,t,e,i,r,a,s){a===void 0&&(a="NHWC");var o=n;n.rank===3&&(o=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var l=t;l.rank===3&&(l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]])),E(o.rank===4,function(){return"Error in conv2dDerFilter: input must be rank 4, but got shape "+(o.shape+".")}),E(l.rank===4,function(){return"Error in conv2dDerFilter: dy must be rank 4, but got shape "+(l.shape+".")}),E(e.length===4,function(){return"Error in conv2dDerFilter: filterShape must be length 4, but got "+(e+".")});var u=a==="NHWC"?o.shape[3]:o.shape[1],c=a==="NHWC"?l.shape[3]:l.shape[1];E(u===e[2],function(){return"Error in conv2dDerFilter: depth of input "+u+") must "+("match input depth in filter ("+e[2]+".")}),E(c===e[3],function(){return"Error in conv2dDerFilter: depth of dy ("+c+") must "+("match output depth for filter ("+e[3]+").")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerFilter: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var h=function(f){var m=1,g=_a(a),v=Bn(o.shape,e,i,m,r,s,!1,g);return f.conv2dDerFilter(o,l,v)},d={x:o,dy:l},p={strides:i,pad:r,dataFormat:a,dimRoundingMode:s};return z.runKernelFunc(h,d,null,Tf,p)}var Nh=U({conv2DBackpropFilter_:GR});function yo(n,t,e){if(e==null||e==="linear")return n;if(e==="relu")return J(n,Zr(t));throw new Error("Cannot compute gradient for fused activation "+e+".")}function bo(n,t){var e=t,i=gt(n.shape,t.shape);return i.length>0&&(e=Ae(e,i)),V(e,n.shape)}function wo(n,t,e){if(t==="linear")return n;if(t==="relu")return Ka(n);if(t==="elu")return ah(n);if(t==="relu6")return bh(n);if(t==="prelu")return mh(n,e);throw new Error("Unknown fused activation "+t+".")}var So=function(n,t){var e=n>0;return!e||t==="linear"};function YR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",So(z.state.gradientDepth,d)===!1){var f=Gr(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),wo(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(rt(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(Ht(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Bn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var N=function(F,P){var H=P,_=H[0],j=H[1],K=H[2],q=H[3],G=yo(F,K,d);E(Si(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=nh(j.shape,G,_,i,r),X=Nh(j,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=bo(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=zn(function(F,P,H){var _=z.runKernelFunc(C,R,null,yc,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:N}});return k(v,g)}else{var W=zn(function(F,P,H,_){var j=z.runKernelFunc(C,R,null,yc,D);return _([P,F,j,H]),b&&(j=V(j,[j.shape[1],j.shape[2],j.shape[3]])),{value:j,gradFunc:N}});return W(v,g,S)}}var jR=U({fusedConv2d_:YR});function KR(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,Of)}var Bv=U({depthwiseConv2dNativeBackpropFilter_:KR});function $R(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Ef);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var zv=U({depthwiseConv2dNativeBackpropInput_:$R});function XR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(So(z.state.gradientDepth,d)===!1){var f=qa(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),wo(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(Ht(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(rt(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Bn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var N=function(F,P){E(Si(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],j=P[2],K=P[3],q=yo(F,j,d),G=zv(_.shape,q,H,w),Z=Bv(_,q,H.shape,w);if(K!=null){var X=bo(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=zn(function(F,P,H){var _=z.runKernelFunc(C,R,null,bc,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:N}});return k(v,g)}else{var W=zn(function(F,P,H,_){var j=z.runKernelFunc(C,R,null,bc,D);return _([P,F,j,H]),b&&(j=V(j,[j.shape[1],j.shape[2],j.shape[3]])),{value:j,gradFunc:N}});return W(v,g,S)}}var JR=U({fusedDepthwiseConv2d_:XR});function ZR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(So(z.state.gradientDepth,c)===!1){var d=We(e,i,a,o);return l!=null&&(d=fe(d,l)),wo(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=at(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=ot(w),N=ot(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(dn(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[N,b,g]):V(f,[N,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=at(k,p)[0],et(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=yo(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=We(ie,X,!1,!0),re=We(Z,ie,!0,!1)):!a&&o?(te=We(ie,X,!1,!1),re=We(ie,Z,!0,!1)):a&&!o?(te=We(X,ie,!1,!0),re=We(Z,ie,!1,!1)):(te=We(X,ie,!0,!0),re=We(ie,Z,!0,!0)),l!=null){var le=bo(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var j=zn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,vc,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return j(R,D)}else{var K=zn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,vc,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return K(R,D,k)}}var QR=U({fusedMatMul_:ZR});var eO={__proto__:null,conv2d:jR,depthwiseConv2d:JR,matMul:QR};function tO(n){return Th(n,.54,.46)}var nO=U({hammingWindow_:tO});function iO(n){return Th(n,.5,.5)}var Pv=U({hannWindow_:iO});function rO(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a<n.size;){var o=a+t-n.size,l=Rt([ze(n,a,t-o),sh([o],r)]);s.push(l),a+=e}return s.length===0?Xa([],[0,t]):V(Rt(s),[s.length,t])}var _v=U({frame_:rO});function aO(n,t,e,i,r){r===void 0&&(r=Pv),i==null&&(i=Uv(t));for(var a=_v(n,t,e),s=J(a,r(t)),o=[],l=0;l<a.shape[0];l++)o.push(po(ze(s,[l,0],[1,t]),i));return Rt(o)}var sO=U({stft_:aO});function oO(n,t,e,i,r,a){var s=O(n,"image","cropAndResize"),o=O(t,"boxes","cropAndResize","float32"),l=O(e,"boxInd","cropAndResize","int32");r=r||"bilinear",a=a||0;var u=o.shape[0];E(s.rank===4,function(){return"Error in cropAndResize: image must be rank 4,"+("but got rank "+s.rank+".")}),E(o.rank===2&&o.shape[1]===4,function(){return"Error in cropAndResize: boxes must be have size ["+u+",4] "+("but had shape "+o.shape+".")}),E(l.rank===1&&l.shape[0]===u,function(){return"Error in cropAndResize: boxInd must be have size ["+u+"] "+("but had shape "+o.shape+".")}),E(i.length===2,function(){return"Error in cropAndResize: cropSize must be of length 2, but got "+("length "+i.length+".")}),E(i[0]>=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Cf,d);return p}var lO=U({cropAndResize_:oO});function uO(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(Pf,e,{});return i}var cO=U({flipLeftRight_:uO});function hO(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(mm,a,s);return o}var dO=U({rotateWithOffset_:hO});function Qr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function pO(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Qr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,tm,l)}var fO=U({nonMaxSuppression_:pO});function gO(n,t,e){var i=mO(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function mO(n,t,e){return yO(n,t,e||vO)}function vO(n,t){return n>t?1:n<t?-1:0}function yO(n,t,e){for(var i=0,r=n.length,a=0,s=!1;i<r;){a=i+(r-i>>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function Mv(n,t,e,i,r){return xh(n,t,e,i,r,0).selectedIndices}function Hv(n,t,e,i,r,a){return xh(n,t,e,i,r,0,!1,a,!0)}function Vv(n,t,e,i,r,a){return xh(n,t,e,i,r,a,!0)}function xh(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;c<t.length;c++)t[c]>r&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(qv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length<e&&u.length>0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m<r)break;for(var b=!1,w=d.length-1;w>=v;--w){var S=bO(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*wO(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&gO(u,f,qv))}var L=d.length,N=e-L;o&&N>0&&(d.push.apply(d,new Array(N).fill(0)),p.push.apply(p,new Array(N).fill(0)));var C={selectedIndices:Xr(d,"int32")};return s&&(C.selectedScores=Xr(p,"float32")),l&&(C.validOutputs=be(L,"int32")),C}function bO(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function wO(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function qv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function SO(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Qr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=Mv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var LO=SO;function IO(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Qr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(im,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var AO=U({nonMaxSuppressionWithScore_:IO});function TO(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Qr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=Vv(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var NO=TO;function xO(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Qr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(nm,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var CO=U({nonMaxSuppressionPadded_:xO});function RO(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Qr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=Hv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var OO=RO;function EO(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,Hu,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Gv=U({resizeBilinear_:EO});function DO(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,Mu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Yv=U({resizeNearestNeighbor_:DO});function kO(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(yh(0,s,1,"int32"),[-1,1]),u=yh(0,o,1,"int32"),c=ge(l,u),h=nr(Qi(c,be(+t,"int32")),Zi(c,be(-e,"int32"))),d=ti([s,o],i.dtype);return V(sr(go(V(i,[-1,s,o])).map(function(p){return yn(h,p,d)})),r)}var FO=U({bandPart_:kO});function WO(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r<n.length;++r)i(r)}else t=!0,n=Jr(n,n.shape[0],0).map(function(l){return mo(l,[0])});E(n.length<=n[0].shape[0],function(){return"Gram-Schmidt: Number of vectors ("+n.length+") exceeds "+("number of dimensions ("+n[0].shape[0]+").")});for(var a=[],s=n,o=function(l){a.push(z.tidy(function(){var u=s[l];if(l>0)for(var c=0;c<l;++c){var h=J(Ae(J(a[c],u)),a[c]);u=ge(u,h)}return Ie(u,vo(u,"euclidean"))}))},r=0;r<n.length;++r)o(r);return t?sr(a,0):a}var UO=U({gramSchmidt_:WO});function BO(n,t){if(t===void 0&&(t=!1),E(n.rank>=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return jv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=go(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=jv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(sr(r,0),n.shape),o=V(sr(a,0),n.shape);return[s,o]}function jv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=jg(e),a=$i(n),s=Xa([[1]],[1,1]),o=$i(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=vo(g),b=ze(a,[h,h],[1,1]),w=yn(Li(b,0),Xa([[-1]]),Xa([[1]])),S=ge(b,J(w,v)),L=Ie(g,S);L.shape[0]===1?o=$i(s):o=Rt([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var N=vt(Ie(We(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(N,o),D=ut(o);if(h===0)a=ge(C,We(R,We(D,C)));else{var k=ge(C,We(R,We(D,C)));a=Rt([ze(a,[0,0],[h,i]),k],0)}var W=ut(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,We(We(F,o),W));else{var P=ge(F,We(We(F,o),W));r=Rt([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],Mt([p,f,m])},c=0;c<l;++c)u(c);return!t&&e>i&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var zO=U({qr_:BO});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function PO(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ae(a);if(e===I.Reduction.MEAN){if(r==null)return Ya(a);var s=i.size/r.size,o=Ie(Ae(a),Ae(r));return s>1?Ie(o,be(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Ie(Ae(a),be(i.size));var l=J(r,Kr(i.shape)),u=ue(Ae(uo(l,be(0))),"float32");return Ie(Ae(a),u)}throw Error("Unknown reduction: "+e)}var ri=U({computeWeightedLoss_:PO});function _O(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Be(r.shape,a.shape,"Error in absoluteDifference: ");var o=Kt(ge(r,a));return ri(o,s,i)}var MO=U({absoluteDifference_:_O});function HO(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Be(a.shape,s.shape,"Error in cosineDistance: ");var l=be(1),u=ge(l,Ae(J(a,s),e,!0));return ri(u,o,r)}var VO=U({cosineDistance_:HO});function qO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Be(r.shape,a.shape,"Error in hingeLoss: ");var o=be(1);r=ge(J(be(2),r),o);var l=Ka(ge(o,J(r,a)));return ri(l,s,i)}var GO=U({hingeLoss_:qO});function YO(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Be(a.shape,s.shape,"Error in huberLoss: ");var l=be(i),u=Kt(ge(s,a)),c=lo(u,l),h=ge(u,c),d=fe(J(be(.5),Ye(c)),J(l,h));return ri(d,o,r)}var jO=U({huberLoss_:YO});function KO(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Be(a.shape,s.shape,"Error in logLoss: ");var l=be(1),u=be(i),c=vt(J(a,er(fe(s,u)))),h=J(ge(l,a),er(fe(ge(l,s),u))),d=ge(c,h);return ri(d,o,r)}var $O=U({logLoss_:KO});function XO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Be(r.shape,a.shape,"Error in meanSquaredError: ");var o=fo(r,a);return ri(o,s,i)}var JO=U({meanSquaredError_:XO});function ZO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Be(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ka(i),a=J(i,e),s=lh(bn(vt(Kt(i))));return fe(ge(r,a),s)}function QO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Be(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=be(i),u=be(1),c=be(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=ZO(a,s);return ri(h,o,r)}var eE=U({sigmoidCrossEntropy_:QO});function tE(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=zn(function(r,a,s){var o=!0,l=ch(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=vt(J(u,r)),h=Ae(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=rn(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),bn(g))),J(V(p,v),ge(bn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function nE(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Be(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=be(i),u=be(1),c=be(a.shape[1]);a=fe(J(a,ge(u,l)),Ie(l,c))}var h=tE(a,s);return ri(h,o,r)}var iE=U({softmaxCrossEntropy_:nE});var rE={fft:ho,ifft:$a,rfft:po,irfft:Ih},aE={hammingWindow:nO,hannWindow:Pv,frame:_v,stft:sO},sE={flipLeftRight:cO,resizeNearestNeighbor:Yv,resizeBilinear:Gv,rotateWithOffset:dO,cropAndResize:lO,nonMaxSuppression:fO,nonMaxSuppressionAsync:LO,nonMaxSuppressionWithScore:AO,nonMaxSuppressionWithScoreAsync:NO,nonMaxSuppressionPadded:CO,nonMaxSuppressionPaddedAsync:OO},oE={bandPart:FO,gramSchmidt:UO,qr:zO},lE={absoluteDifference:MO,computeWeightedLoss:ri,cosineDistance:VO,hingeLoss:GO,huberLoss:jO,logLoss:$O,meanSquaredError:JO,sigmoidCrossEntropy:eE,softmaxCrossEntropy:iE};var Ii=function(n){Zn(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return Mt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return ev(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&Mt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:be(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(vg);Object.defineProperty(Ii,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Ch=function(n){Zn(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:mt(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:mt(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;mt(function(){var d=fe(J(c,i.rho),J(Ye(u),1-i.rho)),p=J(Ie(Vt(fe(h,i.epsilon)),Vt(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ye(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(Mt(this.accumulatedGrads.map(function(e){return e.variable})),Mt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(Ii);wi(Ch);var Rh=function(n){Zn(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:mt(function(){return sh(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;mt(function(){var h=fe(c,Ye(u));c.assign(h);var d=fe(J(Ie(u,Vt(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&Mt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(Ii);wi(Rh);var Oh=function(n){Zn(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],mt(function(){s.accBeta1=be(i).variable(),s.accBeta2=be(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);mt(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:mt(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:mt(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ye(h),1-i.beta2)),g=Ie(f,a),v=Ie(m,s);d.assign(f),p.assign(m);var b=fe(J(Ie(g,fe(Vt(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&Mt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&Mt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),mt(function(){a.accBeta1.assign(ni(a.beta1,a.iterations_+1)),a.accBeta2.assign(ni(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(Ii);wi(Oh);var Eh=function(n){Zn(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],mt(function(){o.iteration=be(0).variable(),o.accBeta1=be(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);mt(function(){var a=ge(1,i.accBeta1),s=Ie(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=Kt(h),v=jr(m,g);d.assign(f),p.assign(v);var b=fe(J(Ie(s,a),Ie(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&Mt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&Mt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(Ii);wi(Eh);var Lo=function(n){Zn(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];mt(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=wg(be(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(Ii);wi(Lo);var Dh=function(n){Zn(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=be(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:mt(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;mt(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&Mt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(Lo);wi(Dh);var kh=function(n){Zn(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:mt(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:mt(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:mt(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;mt(function(){var d=fe(J(c,i.decay),J(Ye(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Ie(J(u,i.learningRate),Vt(ge(d,fe(Ye(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ye(u),1-i.decay)),g=fe(J(h,i.momentum),Ie(J(u,i.learningRate),Vt(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&Mt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&Mt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&Mt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(Ii);wi(kh);var or=function(){function n(){}return n.sgd=function(t){return new Lo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new Dh(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new kh(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new Oh(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Ch(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new Eh(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Rh(t,e)},n}();var uE={sgd:or.sgd,momentum:or.momentum,adadelta:or.adadelta,adagrad:or.adagrad,rmsprop:or.rmsprop,adamax:or.adamax,adam:or.adam};var cE=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function hE(){return new Promise(function(n){return cE(function(){return n()})})}function dE(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function pE(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s<a;++s)r=r.concat([n[s+1]/t[s],t[s]]);r=r.concat(n.slice(a+1))}return r}function fE(n,t,e){e===void 0&&(e=!0);var i=[];if(e){i.push(t);for(var r=t+1;r<n;++r)r<=2*t?(i.push(r),i.push(r-(t+1))):i.push(r)}else{for(var a=[],s=[],r=1;r<n;++r)r>=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function mE(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a<n.length;++a)a<=t.length?i?r.push(t[a-1]*n[a]):r.push(n[a]/t[a-1]):r.push(n[a]);return r}function gE(n,t){for(var e=[0],i=0;i<t;++i)e.push(n[i][0]);return e}function vE(n,t,e){for(var i=n.slice(0,1),r=0;r<e;++r)i.push(n[r+1]-t[r][0]-t[r][1]);return i}var Kv=1.7580993408473768,$v=1.0507009873554805;var yE=.3275911,bE=.254829592,wE=-.284496736,SE=1.421413741,LE=-1.453152027,IE=1.061405429;function AE(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.warn.apply(console,n)}function TE(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.log.apply(console,n)}function NE(n,t){if(n.length!==t.length)throw new Error("Cannot merge real and imag arrays of different lengths. real:"+(n.length+", imag: "+t.length+"."));for(var e=new Float32Array(n.length*2),i=0;i<e.length;i+=2)e[i]=n[i/2],e[i+1]=t[i/2];return e}function xE(n){for(var t=new Float32Array(n.length/2),e=new Float32Array(n.length/2),i=0;i<n.length;i+=2)t[i/2]=n[i],e[i/2]=n[i+1];return{real:t,imag:e}}function CE(n){for(var t=Math.ceil(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=0;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function RE(n){for(var t=Math.floor(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=2;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function OE(n,t){var e=n[t*2],i=n[t*2+1];return{real:e,imag:i}}function EE(n,t,e,i){n[i*2]=t,n[i*2+1]=e}function DE(n,t){for(var e=new Float32Array(n/2),i=new Float32Array(n/2),r=0;r<Math.ceil(n/2);r++){var a=(t?2:-2)*Math.PI*(r/n);e[r]=Math.cos(a),i[r]=Math.sin(a)}return{real:e,imag:i}}function kE(n,t,e){var i=(e?2:-2)*Math.PI*(n/t),r=Math.cos(i),a=Math.sin(i);return{real:r,imag:a}}function FE(n,t,e){if(t==="complex64"){if(n.dtype==="complex64")return n.clone();var i=ti(n.shape),r=ue(n,"float32"),a=e.complex(r,i);return i.dispose(),r.dispose(),a}if(!Am(n.dtype,t))return z.makeTensorFromDataId(n.dataId,n.shape,t);if(n.dtype==="complex64"){var s=e.real(n),a=ue(s,t);return s.dispose(),a}if(t==="int32")return e.int(n);if(t==="bool"){var o=be(0,n.dtype),a=e.notEqual(n,o);return o.dispose(),a}else throw new Error("Error in Cast: failed to cast "+n.dtype+" to "+t)}function WE(n,t){return z.makeTensorFromDataId(n.dataId,t,n.dtype)}function UE(n,t,e){var i=(t-n)/(e-1),r=Ur(e,"float32");r[0]=n;for(var a=1;a<r.length;a++)r[a]=r[a-1]+i;return Xr(r,"float32")}var BE={__proto__:null,slice_util:gg,segment_util:BN,castTensor:FE,reshapeTensor:WE,linspaceImpl:UE,upcastType:Vs,axesAreInnerMostDims:Zc,combineLocations:Ig,computeOutAndReduceShapes:Ag,expandShapeToKeepDim:rn,assertAxesAreInnerMostDims:p1,getAxesPermutation:an,getUndoAxesPermutation:Ks,getInnerMostAxes:Un,getBroadcastDims:wN,getReductionAxes:gt,assertAndGetBroadcastShape:et,assertParamsConsistent:Fg,computeOutShape:Wg,computeDilation2DInfo:I1,computePool2DInfo:Vr,computePool3DInfo:Ha,computeConv2DInfo:Bn,computeConv3DInfo:Ma,computeDefaultPad:eh,tupleValuesAreOne:Si,eitherStridesOrDilationsAreOne:Ht,convertConv2DDataFormat:_a,getFusedDyActivation:yo,getFusedBiasGradient:bo,applyActivation:wo,shouldFuse:So,PARALLELIZE_THRESHOLD:oh,computeOptimalWindowSize:FN,getImageCenter:dE,getReshaped:pE,getPermuted:fE,getReshapedPermuted:mE,getSliceBeginCoords:gE,getSliceSize:vE,prepareAndValidate:ig,validateUpdateShape:Gc,validateInput:Yc,calculateShapes:rg,SELU_SCALEALPHA:Kv,SELU_SCALE:$v,ERF_P:yE,ERF_A1:bE,ERF_A2:wE,ERF_A3:SE,ERF_A4:LE,ERF_A5:IE,warn:AE,log:TE,mergeRealAndImagArrays:NE,splitRealAndImagArrays:xE,complexWithEvenIndex:CE,complexWithOddIndex:RE,getComplexWithIndex:OE,assignToTypedArray:EE,exponents:DE,exponent:kE,prepareSplitSize:pv};function zE(n,t,e){var i=new Array(n.rank).fill(0),r=n.shape.slice();return t.map(function(a){var s=r.slice();s[e]=a;var o=ze(n,i,s);return i[e]+=a,o})}function PE(n,t){for(var e=new Array(n.rank),i=0;i<e.length;i++)e[i]=n.shape[i]*t[i];for(var r=Wn(e,n.dtype),i=0;i<r.values.length;++i){for(var a=r.indexToLoc(i),s=new Array(n.rank),o=0;o<s.length;o++)s[o]=a[o]%n.shape[o];var l=n.locToIndex(s);r.values[i]=n.values[l]}return r.toTensor()}function _E(n,t,e,i,r){for(var a=t[t.length-1],s=[n.length/a,a],o=s[0],l=s[1],u=Bs(e,o*i),c=Bs("int32",o*i),h=0;h<o;h++){for(var d=h*l,p=n.subarray(d,d+l),f=[],m=0;m<p.length;m++)f.push({value:p[m],index:m});f.sort(function(S,L){return L.value-S.value});for(var g=h*i,v=u.subarray(g,g+i),b=c.subarray(g,g+i),m=0;m<i;m++)v[m]=f[m].value,b[m]=f[m].index}var w=t.slice();return w[w.length-1]=i,[vi(u,w,e),vi(c,w,"int32")]}var ME={__proto__:null,nonMaxSuppressionV3Impl:Mv,nonMaxSuppressionV4Impl:Hv,nonMaxSuppressionV5Impl:Vv,split:zE,tile:PE,topkImpl:_E,whereImpl:yv};var HE={kernelName:kl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Zr(ue(e,"float32"),-1))}}}};var VE={kernelName:Fl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Ye(ue(e,"float32")),r=Vt(ge(be(1),i));return vt(Ie(n,r))}}}};var qE={kernelName:Wl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Vt(ge(Ye(ue(e,"float32")),1));return Ie(n,i)}}}};var GE={kernelName:Fs,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=gt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=gt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,i.shape)};return{a,b:s}}};var YE={kernelName:Ul,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var jE={kernelName:Bl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var KE={kernelName:zl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var $E={kernelName:Pl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Vt(ge(be(1),Ye(ue(e,"float32")))))}}}};var XE={kernelName:_l,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Vt(fe(be(1),Ye(ue(e,"float32"))));return Ie(n,i)}}}};var JE={kernelName:Vl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=fe(Ye(e),Ye(i)),l=J(n,Ie(i,o)),u=gt(e.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,e.shape)},s=function(){var o=fe(Ye(e),Ye(i)),l=vt(J(n,Ie(e,o))),u=gt(i.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,i.shape)};return{a,b:s}}};var ZE={kernelName:Ml,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(Ye(ue(e,"float32")),1))}}}};var QE={kernelName:Hl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ge(be(1),Ye(ue(e,"float32"))))}}}};function eD(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(Ht(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(rt(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=Ha(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,If,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var tD=U({avgPool3dBackprop_:eD});var nD={kernelName:Gl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return tD(n,i,a,s,c,l,u)}}}};function iD(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=Vr(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,Lf,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var rD=U({avgPoolBackprop_:iD});var aD={kernelName:ql,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return rD(n,i,a,s,o)}}}};var sD={kernelName:Yl,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return We(n,a,!1,!0)},b:function(){return We(r,n,!0,!1)}}:!o&&l?{a:function(){return We(n,a,!1,!1)},b:function(){return We(n,r,!0,!1)}}:o&&!l?{a:function(){return We(a,n,!1,!0)},b:function(){return We(r,n,!1,!1)}}:{a:function(){return We(a,n,!0,!0)},b:function(){return We(n,r,!0,!0)}}}};var oD={kernelName:jl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return co(n,r,a)}}}};var lD={kernelName:Kl,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o<s.length;o++)s[o]>1&&l.push(o);return{x:function(){return Ae(n,l,!0)}}}};var uD={kernelName:Ws,gradFunc:function(n){return{x:function(){return n.clone()}}}};var cD={kernelName:$l,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var hD={kernelName:Xl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return yn(nr(Zi(i,a),Qi(i,s)),n,Ee(n))}}}};var dD={kernelName:Jl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Qe(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=Jr(n,s,a);return o.map(function(l){return function(){return l}})}};var pD={kernelName:Zl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(Si(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return nh(r.shape,n,a,l,u,c)},filter:function(){return Nh(r,n,a.shape,l,u,c)}}}};var fD={kernelName:Ql,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Gr(n,a,o,l,u,1,c)},filter:function(){return Nh(n,r,a.shape,o,l,u,c)}}}};function mD(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=Ma(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,Nf,u)}var gD=U({conv3DBackpropFilter_:mD});var vD={kernelName:eu,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(Si(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return _g(o.shape,n,l,a,s)},filter:function(){return gD(o,n,l.shape,a,s)}}}};var yD={kernelName:tu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(vt(Sh(ue(e,"float32"))),n)}}}};var bD={kernelName:nu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Lh(ue(e,"float32")),n)}}}};var wD={kernelName:iu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=an([a],i.rank),u=rh(n,a,s,!o);return l!=null&&(u=ut(u,l)),u}}}};var SD={kernelName:ru,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(Si(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(Ht(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be  1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(rt(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Bn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return zv(c.shape,n,h,d)},filter:function(){return Bv(c,n,h.shape,d)}}}};var LD={kernelName:au,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(kf,s,e)},filter:function(){return z.runKernel(Ff,o,e)}}}};var ID={kernelName:su,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=gt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=gt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return vt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var AD={kernelName:ou,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Wf)}}}};var TD={kernelName:lu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(bn(vt(Ye(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var ND={kernelName:uu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var xD={kernelName:cu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,bn(e))}}}};var CD={kernelName:hu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var RD={kernelName:du,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=gt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=gt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return vt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var OD={kernelName:pu,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??be(1),u=gt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h<r.shape.length-1;++h)c.push(r.shape[h]);c.push(1)}var d=ge(r,a),p=J(n,l),f=wh(fe(s,be(i))),m=J(J(J(f,f),f),be(-.5)),g=function(){return a.rank===1?V(J(J(n,Yr(V(f,[1,1,1,a.shape[0]]),c)),l),r.shape):V(J(J(n,f),l),r.shape)},v=function(){var L=J(J(f,be(-1)),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},b=function(){var L=J(J(m,d),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},w=function(){var L=J(d,f),N=J(n,L);return a.rank===1&&(N=Ae(N,u)),V(N,a.shape)},S=function(){var L=n;return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)};return{x:g,mean:v,variance:b,scale:w,offset:S}}};var ED={kernelName:fu,inputsToSave:["x","indices"],gradFunc:function(n,t,e){var i=t[0],r=t[1],a=e.axis,s=Qe(a,i.shape)[0],o=function(){var l=i.shape,u=r.size,c=l.slice(0,s),h=c.length,d=l.slice(a,l.length).slice(1),p=d.length,f=Xv(0,h),m=Xv(h+1,h+1+p),g=Jv([c,[u],d]),v=V(n,g),b=V(r,[u]),w=Jv([[h],f,m]),S=ut(v,w),L=Ah(S,b,i.shape[s]),N=Ks(w);return L=ut(L,N),L};return{x:o,indices:function(){return r}}}};function Xv(n,t){for(var e=[],i=n;i<t;++i)e.push(i);return e}function Jv(n){for(var t=[],e=0;e<n.length;++e)for(var i=0;i<n[e].length;++i)t.push(n[e][i]);return t}var DD={kernelName:mu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1];return{a:function(){return Ee(e)},b:function(){return Ee(i)}}}};var kD={kernelName:gu,gradFunc:function(n){return{x:function(){return ue(n,"float32")}}}};var FD={kernelName:vu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var WD={kernelName:yu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var UD={kernelName:bu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var BD={kernelName:Su,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(e,1))}}}};var zD={kernelName:wu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ue(e,"float32"))}}}};var PD={kernelName:Lu,inputsToSave:[],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.axis;return{logits:function(){var a=!0,s=bn(i);return ge(n,J(Ae(n,r,a),s))}}}};function _D(n,t,e,i,r,a,s){i===void 0&&(i=5),r===void 0&&(r=1),a===void 0&&(a=1),s===void 0&&(s=.5);var o=function(c){return c.LRNGrad(e,n,t,i,r,a,s)},l={x:n,y:t,dy:e},u={depthRadius:i,bias:r,alpha:a,beta:s};return z.runKernelFunc(o,l,null,Xf,u)}var MD=U({localResponseNormalizationBackprop_:_D});var HD={kernelName:Iu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.depthRadius,l=s.bias,u=s.alpha,c=s.beta;return{x:function(){return MD(r,a,n,o,l,u,c)}}}};function Zv(n,t,e,i,r){return t.rank<e.rank&&(t=V(t,rn(t.shape,i))),n.rank<e.rank&&(n=V(n,rn(n.shape,i))),{x:function(){var a=J(n,ue(Ga(e,t),n.dtype));return r==null?a:ut(a,r)}}}var Qv={kernelName:Au,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.reductionIndices,a=t[0],s=t[1],o=Qe(r,a.shape),l=an(o,a.rank),u=Zv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var VD={kernelName:Tu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Zi(e,i),"float32"))},a=function(){return J(n,ue(ro(e,i),"float32"))};return{a:r,b:a}}};function qD(n,t,e,i,r,a,s,o){a===void 0&&(a=[1,1,1]);var l=O(n,"dy","maxPool3dBackprop"),u=O(t,"input","maxPool3dBackprop"),c=O(e,"output","maxPool3dBackprop"),h=l,d=u,p=c,f=!1;u.rank===4&&(f=!0,h=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=V(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),p=V(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),E(h.rank===5,function(){return"Error in maxPool3dBackprop: dy must be rank 5 but got rank "+(h.rank+".")}),E(d.rank===5,function(){return"Error in maxPool3dBackprop: input must be rank 5 but got rank "+(d.rank+".")}),E(p.rank===5,function(){return"Error in maxPool3dBackprop: output must be rank 5 but got rank "+(p.rank+".")}),E(Ht(r,a),function(){return"Error in maxPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+r+" and dilations '"+a+"'")}),o!=null&&E(rt(s),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+o+" but got pad "+s+".")});var m=function(w){var S=Ha(d.shape,i,r,a,s,o);return w.maxPool3dBackprop(h,d,p,S)},g={dy:h,input:d,output:p},v={filterSize:i,strides:r,dilations:a,pad:s,dimRoundingMode:o},b=z.runKernelFunc(m,g,null,Zf,v);return f?V(b,[b.shape[1],b.shape[2],b.shape[3],b.shape[4]]):b}var GD=U({maxPool3dBackprop_:qD});var YD={kernelName:xu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.dilations,c=s.pad,h=s.dimRoundingMode,d=u??[1,1,1];return{x:function(){return GD(n,r,a,o,l,d,c,h)}}}};function jD(n,t,e,i,r,a,s){var o=O(n,"dy","maxPoolBackprop"),l=O(t,"input","maxPoolBackprop"),u=O(e,"output","maxPoolBackprop");E(l.rank===o.rank,function(){return"Rank of input ("+l.rank+") does not match rank of dy "+("("+o.rank+")")}),E(o.rank===4,function(){return"Error in maxPoolBackprop: dy must be rank 4 but got rank "+(o.rank+".")}),E(l.rank===4,function(){return"Error in maxPoolBackprop: input must be rank 4 but got rank "+(l.rank+".")}),s!=null&&E(rt(a),function(){return"Error in maxPoolBackprop: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+a+".")});var c=function(p){var f=Vr(l.shape,i,r,1,a,s);return p.maxPoolBackprop(o,l,u,f)},h={dy:o,input:l,output:u},d={filterSize:i,strides:r,pad:a,dimRoundingMode:s};return z.runKernelFunc(c,h,null,Jf,d)}var KD=U({maxPoolBackprop_:jD});var $D={kernelName:Nu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.pad;return{x:function(){return KD(n,r,a,o,l,u)}}}};var XD={kernelName:Cu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.axis,a=t[0],s=t[1],o=Qe(r,a.shape),l=an(o,a.rank),u=Zv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var JD={kernelName:Ru,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Qi(e,i),"float32"))},a=function(){return J(n,ue(Li(e,i),"float32"))};return{a:r,b:a}}};var ZD={kernelName:Ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=gt(e.shape,r);return o.length>0?V(Ae(n,o),e.shape):n},s=function(){var o=J(n,vt(to(Ie(e,i)))),l=gt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var QD={kernelName:Eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=gt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=gt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var ek={kernelName:Du,gradFunc:function(n){return{x:function(){return vt(n)}}}};var tk={kernelName:Fu,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return ti(e.shape,"float32")}}}};var nk={kernelName:ku,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ey={kernelName:Wu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var ik={kernelName:Uu,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=et(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,ni(a,ge(c,be(1))))),d=gt(a.shape,o);return d.length>0&&(h=Ae(h,d)),V(h,a.shape)},u=function(){var c=Li(a,0),h=yn(c,er(a),Ee(a)),d=J(n,J(r,h)),p=gt(s.shape,o);return p.length>0&&(d=Ae(d,p)),V(d,s.shape)};return{a:l,b:u}}};var rk={kernelName:Bu,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=Li(e,0);return{x:function(){return yn(r,n,J(n,i))},alpha:function(){var a=yn(r,Ee(n),J(n,e)),s=gt(i.shape,n.shape);return s.length>0&&(a=Ae(a,s)),V(a,i.shape)}}}};var ak={kernelName:zu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,vt(Ye(e)))}}}};var sk={kernelName:Vu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(Qi(e,6),Zr(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var ok={kernelName:Pu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue(Zr(e),"float32"))}}}};var lk={kernelName:_u,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var uk={kernelName:Hu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,lm,e)};return{images:s}}};var ck={kernelName:Mu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,om,e)};return{images:s}}};var hk={kernelName:qu,gradFunc:function(n,t,e){var i=e.dims,r=Qe(i,n.shape);return{x:function(){return ii(n,r)}}}};var dk={kernelName:Gu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var pk={kernelName:Yu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return vt(Ie(n,J(ni(e,1.5),2)))}}}};var fk={kernelName:ju,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(so(e),n.dtype))}}}};var mk={kernelName:Ku,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Li(e,be(0)),r=be(Kv),a=be($v),s=J(n,a),o=J(J(n,r),bn(ue(e,"float32")));return yn(i,s,o)}}}};var gk={kernelName:Qu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(be(1),e)))}}}};var vk={kernelName:Zu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var yk={kernelName:Xu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(eo(ue(e,"float32")),n)}}}};var bk={kernelName:Ju,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ih(ue(e,"float32")),n)}}}};var wk={kernelName:$u,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=jc(i,a,s),u=l[0],c=l[1],h=[],d=0;d<n.rank;d++)h.push([u[d],o[d]-u[d]-c[d]]);return{x:function(){return ir(n,h)}}}};var Sk={kernelName:ac,outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.dim,a=!0,s=J(n,i);return{logits:function(){return ge(s,J(Ae(s,[r],a),i))}}}};var Lk={kernelName:ec,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Ji(e))}}}};var ty={kernelName:ic,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.paddings;return{x:function(){return Js(n,r,a)}}}};var ny={kernelName:rc,gradFunc:function(n,t,e){var i=e.axis;return{x:function(){return Rt(n,i)}}}};var Ik={kernelName:tc,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,J(Vt(ue(e,"float32")),2))}}}};var Ak={kernelName:cm,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(ue(e,"float32"),2))}}}};var Tk={kernelName:sc,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=be(2),a=function(){return J(n,J(r,ge(e,i)))},s=function(){return J(n,J(r,ge(i,e)))};return{a,b:s}}};var Nk={kernelName:mc,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var xk={kernelName:oc,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=gt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=gt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(vt(o),i.shape)};return{a,b:s}}};var Ck={kernelName:nc,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Qe(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Kr(i.shape,"float32"));return{x:function(){return l}}}};var Rk={kernelName:lc,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Ye(eo(e)))}}}};var Ok={kernelName:uc,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(be(1),Ye(e)),n)}}}};var Ek={kernelName:cc,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o<r[0];++o)s=fe(s,ze(n,[o*i.shape[0]],[i.shape[0]]));else if(i.rank===2)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1]],[i.shape[0],i.shape[1]]));else if(i.rank===3)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2]],[i.shape[0],i.shape[1],i.shape[2]]));else if(i.rank===4)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)for(var c=0;c<r[3];++c)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2],c*i.shape[3]],[i.shape[0],i.shape[1],i.shape[2],i.shape[3]]));else throw new Error("Gradient for tile operation is not implemented for rank-"+(i.rank+" tensors yet."));return s};return{x:a}}};var Dk={kernelName:hc,gradFunc:function(n,t,e){var i=e,r=i.perm,a=Ks(r);return{x:function(){return ut(n,a)}}}};var kk={kernelName:dc,gradFunc:function(n,t,e){var i=e,r=i.axis;return{value:function(){return sr(n,r)}}}};var Wk={kernelName:pc,inputsToSave:["segmentIds"],gradFunc:function(n,t){var e=t[0],i=function(){return Fk(n,e)};return{x:i}}};function Fk(n,t){for(var e=jr(t,Ee(t)),i=no(n,e),r=Zi(t,be(0,"int32")),a=i.rank-r.rank,s=0;s<a;++s)r=wn(r,s+1);r=nr(r,Kr(i.shape,"bool"));var o=Ee(i);return yn(r,i,o)}var Uk={kernelName:fc,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var Bk=[HE,VE,qE,GE,YE,jE,KE,$E,XE,JE,ZE,QE,nD,aD,sD,oD,lD,uD,cD,hD,dD,fD,pD,vD,yD,bD,wD,SD,LD,ID,AD,TD,ND,xD,RD,CD,OD,ED,DD,kD,FD,WD,UD,BD,zD,PD,HD,Qv,Qv,VD,YD,$D,XD,JD,ZD,QD,ek,tk,nk,ey,ey,ik,rk,ak,sk,ok,lk,uk,ck,hk,dk,pk,fk,mk,gk,vk,yk,bk,wk,Sk,Lk,ty,ty,ny,ny,Ik,Tk,Ak,Nk,xk,Ck,Rk,Ok,Ek,Dk,kk,Wk,Uk];for(var Fh=0,iy=Bk;Fh<iy.length;Fh++){var zk=iy[Fh];vm(zk)}Y.prototype.abs=function(){return this.throwIfDisposed(),Kt(this)};Y.prototype.acos=function(){return this.throwIfDisposed(),Sg(this)};Y.prototype.acosh=function(){return this.throwIfDisposed(),Lg(this)};Y.prototype.addStrict=function(n){return this.throwIfDisposed(),Nv(this,n)};Y.prototype.add=function(n){return this.throwIfDisposed(),fe(this,n)};Y.prototype.all=function(n,t){return this.throwIfDisposed(),Tg(this,n,t)};Y.prototype.any=function(n,t){return this.throwIfDisposed(),Ng(this,n,t)};Y.prototype.argMax=function(n){return this.throwIfDisposed(),xg(this,n)};Y.prototype.argMin=function(n){return this.throwIfDisposed(),Cg(this,n)};Y.prototype.asScalar=function(){return this.throwIfDisposed(),E(this.size===1,function(){return"The array must have only 1 element."}),V(this,[])};Y.prototype.asType=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.as1D=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.as2D=function(n,t){return this.throwIfDisposed(),V(this,[n,t])};Y.prototype.as3D=function(n,t,e){return this.throwIfDisposed(),V(this,[n,t,e])};Y.prototype.as4D=function(n,t,e,i){return this.throwIfDisposed(),V(this,[n,t,e,i])};Y.prototype.as5D=function(n,t,e,i,r){return this.throwIfDisposed(),V(this,[n,t,e,i,r])};Y.prototype.asin=function(){return this.throwIfDisposed(),Rg(this)};Y.prototype.asinh=function(){return this.throwIfDisposed(),Og(this)};Y.prototype.atan=function(){return this.throwIfDisposed(),Eg(this)};Y.prototype.atan2=function(n){return this.throwIfDisposed(),Dg(this,n)};Y.prototype.atanh=function(){return this.throwIfDisposed(),kg(this)};Y.prototype.avgPool=function(n,t,e,i){return this.throwIfDisposed(),th(this,n,t,e,i)};Y.prototype.batchToSpaceND=function(n,t){return this.throwIfDisposed(),Js(this,n,t)};Y.prototype.batchNorm=function(n,t,e,i,r){return this.throwIfDisposed(),Va(this,n,t,e,i,r)};Y.prototype.broadcastTo=function(n){return this.throwIfDisposed(),Qs(this,n)};Y.prototype.cast=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.ceil=function(){return this.throwIfDisposed(),Ug(this)};Y.prototype.clipByValue=function(n,t){return this.throwIfDisposed(),Bg(this,n,t)};Y.prototype.concat=function(n,t){return this.throwIfDisposed(),n instanceof Y&&(n=[n]),Rt([this].concat(n),t)};Y.prototype.conv1d=function(n,t,e,i,r,a){return this.throwIfDisposed(),zg(this,n,t,e,i,r,a)};Y.prototype.conv2dTranspose=function(n,t,e,i,r){return this.throwIfDisposed(),Pg(this,n,t,e,i,r)};Y.prototype.conv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Gr(this,n,t,e,i,r,a)};Y.prototype.cos=function(){return this.throwIfDisposed(),eo(this)};Y.prototype.cosh=function(){return this.throwIfDisposed(),ih(this)};Y.prototype.cumsum=function(n,t,e){return this.throwIfDisposed(),rh(this,n,t,e)};Y.prototype.depthToSpace=function(n,t){return this.throwIfDisposed(),Mg(this,n,t)};Y.prototype.depthwiseConv2D=function(n,t,e,i,r,a){return At("depthwiseConv2D is deprecated, use depthwiseConv2d instead"),this.throwIfDisposed(),qa(this,n,t,e,i,r,a)};Y.prototype.depthwiseConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),qa(this,n,t,e,i,r,a)};Y.prototype.dilation2d=function(n,t,e,i,r){return this.throwIfDisposed(),Hg(this,n,t,e,i,r)};Y.prototype.divNoNan=function(n){return this.throwIfDisposed(),Vg(this,n)};Y.prototype.divStrict=function(n){return this.throwIfDisposed(),xv(this,n)};Y.prototype.div=function(n){return this.throwIfDisposed(),Ie(this,n)};Y.prototype.dot=function(n){return this.throwIfDisposed(),qg(this,n)};Y.prototype.elu=function(){return this.throwIfDisposed(),ah(this)};Y.prototype.equalStrict=function(n){return this.throwIfDisposed(),wv(this,n)};Y.prototype.equal=function(n){return this.throwIfDisposed(),Ga(this,n)};Y.prototype.erf=function(){return this.throwIfDisposed(),Gg(this)};Y.prototype.exp=function(){return this.throwIfDisposed(),bn(this)};Y.prototype.expandDims=function(n){return this.throwIfDisposed(),wn(this,n)};Y.prototype.expm1=function(){return this.throwIfDisposed(),Yg(this)};Y.prototype.fft=function(){return this.throwIfDisposed(),ho(this)};Y.prototype.flatten=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.floor=function(){return this.throwIfDisposed(),to(this)};Y.prototype.floorDiv=function(n){return this.throwIfDisposed(),Jc(this,n)};Y.prototype.gather=function(n,t){return this.throwIfDisposed(),no(this,n,t)};Y.prototype.greaterEqualStrict=function(n){return this.throwIfDisposed(),Sv(this,n)};Y.prototype.greaterEqual=function(n){return this.throwIfDisposed(),Zi(this,n)};Y.prototype.greaterStrict=function(n){return this.throwIfDisposed(),Lv(this,n)};Y.prototype.greater=function(n){return this.throwIfDisposed(),Li(this,n)};Y.prototype.ifft=function(){return this.throwIfDisposed(),$a(this)};Y.prototype.irfft=function(){return this.throwIfDisposed(),Ih(this)};Y.prototype.isFinite=function(){return this.throwIfDisposed(),$g(this)};Y.prototype.isInf=function(){return this.throwIfDisposed(),Xg(this)};Y.prototype.isNaN=function(){return this.throwIfDisposed(),Jg(this)};Y.prototype.leakyRelu=function(n){return this.throwIfDisposed(),Zg(this,n)};Y.prototype.lessEqualStrict=function(n){return this.throwIfDisposed(),Iv(this,n)};Y.prototype.lessEqual=function(n){return this.throwIfDisposed(),Qi(this,n)};Y.prototype.lessStrict=function(n){return this.throwIfDisposed(),Av(this,n)};Y.prototype.less=function(n){return this.throwIfDisposed(),ro(this,n)};Y.prototype.localResponseNormalization=function(n,t,e,i){return this.throwIfDisposed(),Qg(this,n,t,e,i)};Y.prototype.logSigmoid=function(){return this.throwIfDisposed(),tv(this)};Y.prototype.logSoftmax=function(n){return this.throwIfDisposed(),nv(this,n)};Y.prototype.logSumExp=function(n,t){return this.throwIfDisposed(),ch(this,n,t)};Y.prototype.log=function(){return this.throwIfDisposed(),er(this)};Y.prototype.log1p=function(){return this.throwIfDisposed(),lh(this)};Y.prototype.logicalAnd=function(n){return this.throwIfDisposed(),nr(this,n)};Y.prototype.logicalNot=function(){return this.throwIfDisposed(),so(this)};Y.prototype.logicalOr=function(n){return this.throwIfDisposed(),hh(this,n)};Y.prototype.logicalXor=function(n){return this.throwIfDisposed(),iv(this,n)};Y.prototype.matMul=function(n,t,e){return this.throwIfDisposed(),We(this,n,t,e)};Y.prototype.maxPool=function(n,t,e,i){return this.throwIfDisposed(),dh(this,n,t,e,i)};Y.prototype.max=function(n,t){return this.throwIfDisposed(),tr(this,n,t)};Y.prototype.maximumStrict=function(n){return this.throwIfDisposed(),Cv(this,n)};Y.prototype.maximum=function(n){return this.throwIfDisposed(),jr(this,n)};Y.prototype.mean=function(n,t){return this.throwIfDisposed(),Ya(this,n,t)};Y.prototype.min=function(n,t){return this.throwIfDisposed(),oo(this,n,t)};Y.prototype.minimumStrict=function(n){return this.throwIfDisposed(),Rv(this,n)};Y.prototype.minimum=function(n){return this.throwIfDisposed(),lo(this,n)};Y.prototype.modStrict=function(n){return this.throwIfDisposed(),Ov(this,n)};Y.prototype.mod=function(n){return this.throwIfDisposed(),ph(this,n)};Y.prototype.mulStrict=function(n){return this.throwIfDisposed(),Ev(this,n)};Y.prototype.mul=function(n){return this.throwIfDisposed(),J(this,n)};Y.prototype.neg=function(){return this.throwIfDisposed(),vt(this)};Y.prototype.norm=function(n,t,e){return this.throwIfDisposed(),vo(this,n,t,e)};Y.prototype.notEqualStrict=function(n){return this.throwIfDisposed(),Tv(this,n)};Y.prototype.notEqual=function(n){return this.throwIfDisposed(),uo(this,n)};Y.prototype.oneHot=function(n,t,e){return t===void 0&&(t=1),e===void 0&&(e=0),this.throwIfDisposed(),Ys(this,n,t,e)};Y.prototype.onesLike=function(){return this.throwIfDisposed(),fh(this)};Y.prototype.pad=function(n,t){return this.throwIfDisposed(),ir(this,n,t)};Y.prototype.pool=function(n,t,e,i,r){return this.throwIfDisposed(),rv(this,n,t,e,i,r)};Y.prototype.powStrict=function(n){return this.throwIfDisposed(),Dv(this,n)};Y.prototype.pow=function(n){return this.throwIfDisposed(),ni(this,n)};Y.prototype.prelu=function(n){return this.throwIfDisposed(),mh(this,n)};Y.prototype.prod=function(n,t){return this.throwIfDisposed(),av(this,n,t)};Y.prototype.reciprocal=function(){return this.throwIfDisposed(),ov(this)};Y.prototype.relu=function(){return this.throwIfDisposed(),Ka(this)};Y.prototype.relu6=function(){return this.throwIfDisposed(),bh(this)};Y.prototype.reshapeAs=function(n){return this.throwIfDisposed(),V(this,n.shape)};Y.prototype.reshape=function(n){return this.throwIfDisposed(),V(this,n)};Y.prototype.resizeBilinear=function(n,t){return this.throwIfDisposed(),Gv(this,n,t)};Y.prototype.resizeNearestNeighbor=function(n,t){return this.throwIfDisposed(),Yv(this,n,t)};Y.prototype.reverse=function(n){return this.throwIfDisposed(),ii(this,n)};Y.prototype.rfft=function(){return this.throwIfDisposed(),po(this)};Y.prototype.round=function(){return this.throwIfDisposed(),lv(this)};Y.prototype.rsqrt=function(){return this.throwIfDisposed(),wh(this)};Y.prototype.selu=function(){return this.throwIfDisposed(),uv(this)};Y.prototype.separableConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),cv(this,n,t,e,i,r,a)};Y.prototype.sigmoid=function(){return this.throwIfDisposed(),Ji(this)};Y.prototype.sign=function(){return this.throwIfDisposed(),hv(this)};Y.prototype.sin=function(){return this.throwIfDisposed(),Sh(this)};Y.prototype.sinh=function(){return this.throwIfDisposed(),Lh(this)};Y.prototype.slice=function(n,t){return this.throwIfDisposed(),ze(this,n,t)};Y.prototype.softmax=function(n){return this.throwIfDisposed(),dv(this,n)};Y.prototype.softplus=function(){return this.throwIfDisposed(),uh(this)};Y.prototype.spaceToBatchND=function(n,t){return this.throwIfDisposed(),co(this,n,t)};Y.prototype.split=function(n,t){return this.throwIfDisposed(),Jr(this,n,t)};Y.prototype.sqrt=function(){return this.throwIfDisposed(),Vt(this)};Y.prototype.square=function(){return this.throwIfDisposed(),Ye(this)};Y.prototype.squaredDifference=function(n){return this.throwIfDisposed(),fo(this,n)};Y.prototype.squaredDifferenceStrict=function(n){return this.throwIfDisposed(),kv(this,n)};Y.prototype.squeeze=function(n){return this.throwIfDisposed(),mo(this,n)};Y.prototype.stack=function(n,t){this.throwIfDisposed();var e=n instanceof Y?[this,n]:[this].concat(n);return sr(e,t)};Y.prototype.step=function(n){return this.throwIfDisposed(),Zr(this,n)};Y.prototype.stridedSlice=function(n,t,e,i,r,a,s,o){return this.throwIfDisposed(),fv(this,n,t,e,i,r,a,s,o)};Y.prototype.subStrict=function(n){return this.throwIfDisposed(),Fv(this,n)};Y.prototype.sub=function(n){return this.throwIfDisposed(),ge(this,n)};Y.prototype.sum=function(n,t){return this.throwIfDisposed(),Ae(this,n,t)};Y.prototype.tan=function(){return this.throwIfDisposed(),mv(this)};Y.prototype.tanh=function(){return this.throwIfDisposed(),Xs(this)};Y.prototype.tile=function(n){return this.throwIfDisposed(),Yr(this,n)};Y.prototype.toBool=function(){return this.throwIfDisposed(),ue(this,"bool")};Y.prototype.toFloat=function(){return this.throwIfDisposed(),ue(this,"float32")};Y.prototype.toInt=function(){return this.throwIfDisposed(),ue(this,"int32")};Y.prototype.topk=function(n,t){return this.throwIfDisposed(),gv(this,n,t)};Y.prototype.transpose=function(n){return this.throwIfDisposed(),ut(this,n)};Y.prototype.unique=function(n){return this.throwIfDisposed(),vv(this,n)};Y.prototype.unsortedSegmentSum=function(n,t){return this.throwIfDisposed(),Ah(this,n,t)};Y.prototype.unstack=function(n){return this.throwIfDisposed(),go(this,n)};Y.prototype.where=function(n,t){return this.throwIfDisposed(),yn(n,this,t)};Y.prototype.zerosLike=function(){return this.throwIfDisposed(),Ee(this)};I.Abs=kl;I.Acos=Fl;I.Acosh=Wl;I.AdadeltaOptimizer=Ch;I.AdagradOptimizer=Rh;I.AdamOptimizer=Oh;I.AdamaxOptimizer=Eh;I.Add=Fs;I.AddN=Ul;I.All=wf;I.Any=Sf;I.ArgMax=Bl;I.ArgMin=zl;I.Asin=Pl;I.Asinh=_l;I.Atan=Ml;I.Atan2=Vl;I.Atanh=Hl;I.AvgPool=ql;I.AvgPool3D=Gl;I.AvgPool3DBackprop=If;I.AvgPoolBackprop=Lf;I.BatchMatMul=Yl;I.BatchToSpaceND=jl;I.BroadcastTo=Kl;I.Cast=Ws;I.Ceil=$l;I.ClipByValue=Xl;I.Complex=Af;I.Concat=Jl;I.Conv2D=Zl;I.Conv2DBackpropFilter=Tf;I.Conv2DBackpropInput=Ql;I.Conv3D=eu;I.Conv3DBackpropFilterV2=Nf;I.Conv3DBackpropInputV2=xf;I.Cos=tu;I.Cosh=nu;I.CropAndResize=Cf;I.Cumsum=iu;I.DataStorage=x2;I.DepthToSpace=Rf;I.DepthwiseConv2dNative=ru;I.DepthwiseConv2dNativeBackpropFilter=Of;I.DepthwiseConv2dNativeBackpropInput=Ef;I.Diag=Df;I.Dilation2D=au;I.Dilation2DBackpropFilter=Ff;I.Dilation2DBackpropInput=kf;I.Div=su;I.Elu=ou;I.EluGrad=Wf;I.Environment=vf;I.Equal=Uf;I.Erf=lu;I.Exp=uu;I.Expm1=cu;I.FFT=Bf;I.Fill=zf;I.FlipLeftRight=Pf;I.Floor=hu;I.FloorDiv=du;I.FromPixels=gc;I.FusedBatchNorm=pu;I.FusedConv2D=yc;I.FusedDepthwiseConv2D=bc;I.GatherNd=_f;I.GatherV2=fu;I.Greater=Mf;I.GreaterEqual=mu;I.IFFT=Hf;I.Identity=gu;I.Imag=Vf;I.IsFinite=vu;I.IsInf=yu;I.IsNan=bu;I.KernelBackend=mf;I.LRN=Iu;I.LRNBackprop=Xf;I.Less=qf;I.LessEqual=Gf;I.LinSpace=Yf;I.Log=wu;I.Log1p=Su;I.LogSoftmax=Lu;I.LogicalAnd=jf;I.LogicalNot=Kf;I.LogicalOr=$f;I.Max=Au;I.MaxPool=Nu;I.MaxPool3D=xu;I.MaxPool3DBackprop=Zf;I.MaxPoolBackprop=Jf;I.MaxPoolWithArgmax=Qf;I.Maximum=Tu;I.Mean=k2;I.Min=Cu;I.Minimum=Ru;I.Mod=Ou;I.MomentumOptimizer=Dh;I.Multiply=Eu;I.Negate=Du;I.NonMaxSuppressionV3=tm;I.NonMaxSuppressionV4=nm;I.NonMaxSuppressionV5=im;I.NotEqual=em;I.OP_SCOPE_SUFFIX=Pm;I.OneHot=Fu;I.OnesLike=ku;I.Optimizer=Ii;I.PadV2=Wu;I.Pool=F2;I.Pow=Uu;I.Prelu=Bu;I.Prod=rm;I.RMSPropOptimizer=kh;I.Range=am;I.Real=sm;I.Reciprocal=zu;I.Relu=Pu;I.Relu6=Vu;I.Reshape=_u;I.ResizeBilinear=Hu;I.ResizeBilinearGrad=lm;I.ResizeNearestNeighbor=Mu;I.ResizeNearestNeighborGrad=om;I.Reverse=qu;I.RotateWithOffset=mm;I.Round=Gu;I.Rsqrt=Yu;I.SGDOptimizer=Lo;I.ScatterNd=um;I.SelectV2=ju;I.Selu=Ku;I.Sigmoid=Qu;I.Sign=Zu;I.Sin=Xu;I.Sinh=Ju;I.Slice=$u;I.Softmax=ac;I.Softplus=ec;I.SpaceToBatchND=ic;I.SparseToDense=hm;I.SplitV=rc;I.Sqrt=tc;I.Square=cm;I.SquaredDifference=sc;I.Step=mc;I.StridedSlice=dm;I.Sub=oc;I.Sum=nc;I.Tan=lc;I.Tanh=uc;I.Tensor=Y;I.TensorBuffer=Hs;I.Tile=cc;I.TopK=pm;I.Transpose=hc;I.Unique=fm;I.Unpack=dc;I.UnsortedSegmentSum=pc;I.Variable=Ba;I.ZerosLike=fc;I._FusedMatMul=vc;I.abs=Kt;I.acos=Sg;I.acosh=Lg;I.add=fe;I.addN=d1;I.addStrict=Nv;I.all=Tg;I.any=Ng;I.argMax=xg;I.argMin=Cg;I.asin=Rg;I.asinh=Og;I.atan=Eg;I.atan2=Dg;I.atanh=kg;I.avgPool=th;I.avgPool3d=O1;I.backend=n1;I.backend_util=BE;I.basicLSTMCell=U1;I.batchNorm=Va;I.batchNorm2d=M1;I.batchNorm3d=V1;I.batchNorm4d=G1;I.batchToSpaceND=Js;I.booleanMaskAsync=fR;I.broadcastTo=Qs;I.browser=NT;I.buffer=Wn;I.cast=ue;I.ceil=Ug;I.clipByValue=Bg;I.clone=$i;I.complex=mi;I.concat=Rt;I.concat1d=X1;I.concat2d=Z1;I.concat3d=eN;I.concat4d=nN;I.conv1d=zg;I.conv2d=Gr;I.conv2dTranspose=Pg;I.conv3d=lN;I.conv3dTranspose=hN;I.copyRegisteredKernels=B2;I.cos=eo;I.cosh=ih;I.cosineWindow=Th;I.cumsum=rh;I.customGrad=zn;I.deprecationWarn=At;I.depthToSpace=Mg;I.depthwiseConv2d=qa;I.device_util=bA;I.diag=yN;I.dilation2d=Hg;I.disableDeprecationWarnings=VT;I.dispose=Mt;I.disposeVariables=qT;I.div=Ie;I.divNoNan=Vg;I.divStrict=xv;I.dot=qg;I.dropout=HR;I.elu=ah;I.enableDebugMode=HT;I.enableProdMode=MT;I.enclosingPowerOfTwo=Uv;I.engine=GT;I.env=He;I.equal=Ga;I.equalStrict=wv;I.erf=Gg;I.exp=bn;I.expandDims=wn;I.expm1=Yg;I.eye=jg;I.fft=ho;I.fill=sh;I.findBackend=QT;I.findBackendFactory=e1;I.floor=to;I.floorDiv=Jc;I.fused=eO;I.gather=no;I.gatherND=PR;I.gather_util=xT;I.getBackend=JT;I.getGradient=Lc;I.getKernel=Sc;I.getKernelsForBackend=Us;I.grad=QN;I.grads=ex;I.greater=Li;I.greaterEqual=Zi;I.greaterEqualStrict=Sv;I.greaterStrict=Lv;I.ifft=$a;I.imag=io;I.image=sE;I.inTopKAsync=qR;I.io=mT;I.irfft=Ih;I.isFinite=$g;I.isInf=Xg;I.isNaN=Jg;I.keep=wg;I.kernel_impls=ME;I.leakyRelu=Zg;I.less=ro;I.lessEqual=Qi;I.lessEqualStrict=Iv;I.lessStrict=Av;I.linalg=oE;I.linspace=$N;I.localResponseNormalization=Qg;I.log=er;I.log1p=lh;I.logSigmoid=tv;I.logSoftmax=nv;I.logSumExp=ch;I.logicalAnd=nr;I.logicalNot=so;I.logicalOr=hh;I.logicalXor=iv;I.losses=lE;I.matMul=We;I.math=LT;I.max=tr;I.maxPool=dh;I.maxPool3d=vx;I.maxPoolWithArgmax=bx;I.maximum=jr;I.maximumStrict=Cv;I.mean=Ya;I.memory=YT;I.min=oo;I.minimum=lo;I.minimumStrict=Rv;I.mod=ph;I.modStrict=Ov;I.moments=Nx;I.movingAverage=DR;I.mul=J;I.mulStrict=Ev;I.multiRNNCell=Cx;I.multinomial=Ox;I.neg=vt;I.nextFrame=hE;I.norm=vo;I.notEqual=uo;I.notEqualStrict=Tv;I.oneHot=Ys;I.ones=Kr;I.onesLike=fh;I.op=U;I.outerProduct=Wx;I.pad=ir;I.pad1d=zx;I.pad2d=_x;I.pad3d=Hx;I.pad4d=qx;I.pool=rv;I.pow=ni;I.powStrict=Dv;I.prelu=mh;I.print=$m;I.prod=av;I.profile=jT;I.rand=Qx;I.randomGamma=cC;I.randomNormal=dC;I.randomUniform=sv;I.range=yh;I.ready=XT;I.real=ja;I.reciprocal=ov;I.registerBackend=t1;I.registerGradient=vm;I.registerKernel=gm;I.relu=Ka;I.relu6=bh;I.removeBackend=ZT;I.reshape=V;I.reverse=ii;I.reverse1d=bC;I.reverse2d=SC;I.reverse3d=IC;I.reverse4d=TC;I.rfft=po;I.round=lv;I.rsqrt=wh;I.scalar=be;I.scatterND=FR;I.scatter_util=CT;I.selu=uv;I.separableConv2d=cv;I.serialization=ET;I.setBackend=$T;I.setPlatform=i1;I.setdiff1dAsync=EC;I.sigmoid=Ji;I.sign=hv;I.signal=aE;I.sin=Sh;I.sinh=Lh;I.slice=ze;I.slice1d=UC;I.slice2d=zC;I.slice3d=_C;I.slice4d=HC;I.slice_util=gg;I.softmax=dv;I.softplus=uh;I.spaceToBatchND=co;I.sparseToDense=BR;I.spectral=rE;I.split=Jr;I.sqrt=Vt;I.square=Ye;I.squaredDifference=fo;I.squaredDifferenceStrict=kv;I.squeeze=mo;I.stack=sr;I.step=Zr;I.stridedSlice=fv;I.sub=ge;I.subStrict=Fv;I.sum=Ae;I.sumOutType=cA;I.tan=mv;I.tanh=Xs;I.tensor=vi;I.tensor1d=Xr;I.tensor2d=Xa;I.tensor3d=ng;I.tensor4d=nR;I.tensor5d=iR;I.tensor6d=rR;I.tensor_util=pA;I.test_util=PT;I.tidy=mt;I.tile=Yr;I.time=KT;I.topk=gv;I.train=uE;I.transpose=ut;I.truncatedNormal=oR;I.unique=vv;I.unregisterGradient=U2;I.unregisterKernel=W2;I.unsortedSegmentSum=Ah;I.unstack=go;I.upcastType=Vs;I.util=Q2;I.valueAndGrad=tx;I.valueAndGrads=nx;I.variable=hR;I.variableGrads=ev;I.version_core=_T;I.where=yn;I.whereAsync=bv;I.zeros=ti;I.zerosLike=Ee});var Ow=Le(Xe=>{"use strict";Object.defineProperty(Xe,"__esModule",{value:!0});var y=lr();var Wh=function(n,t){return Wh=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Wh(n,t)};function Q(n,t){Wh(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var $t=function(){return $t=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t},$t.apply(this,arguments)};function Pk(n,t){var e={};for(var i in n)Object.prototype.hasOwnProperty.call(n,i)&&t.indexOf(i)<0&&(e[i]=n[i]);if(n!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,i=Object.getOwnPropertySymbols(n);r<i.length;r++)t.indexOf(i[r])<0&&(e[i[r]]=n[i[r]]);return e}function Se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function ve(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Uh;function yt(){return Uh==null&&(Uh=y.backend().epsilon()),Uh}function Sn(){return"channelsLast"}var Ai=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Pn=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),M=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Te=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),_k=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),dV=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error);function ur(n,t){if(Array.isArray(n)){for(var e=[],i=0;i<t;i++)e=e.concat(n);return e}else{var e=new Array(t);return e.fill(n),e}}function _n(n,t){if(!n)throw new _k(t)}function ry(n,t){for(var e=0,i=0,r=n;i<r.length;i++){var a=r[i];a===t&&e++}return e}function qt(n){return n.length===1?n[0]:n}function Je(n){return Array.isArray(n)?n:[n]}function ai(n){var t=n.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2"),e=t.replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function cr(n){return n.length<=1||n.indexOf("_")===-1?n:n.replace(/[_]+(\w|$)/g,function(t,e){return e.toUpperCase()})}var pn={};function Bh(n){if(n==null)return null;var t={};return t.className=n.getClassName(),t.config=n.getConfig(),t}function zh(n){if(n==null||typeof n!="object")return;if(Array.isArray(n))n.forEach(function(s){return zh(s)});else for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e],a=n[r];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?n[r]=a.value:zh(a))}}function Ja(n,t,e,i,r){var a,s,o;if(t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i="object"),r===void 0&&(r=!1),typeof n=="string"){var l=n,u=void 0;if(l in e)u=e[l];else if(l in pn)u=pn[l];else if(u=t[l],u==null)throw new M("Unknown "+i+": "+n+`. This may be due to one of the following reasons:
+        `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,Rf,u)}var Mg=U({depthToSpace_:mN});function gN(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(rt(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(Ht(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Bn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,au,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var qa=U({depthwiseConv2d_:gN});function vN(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Df)}var yN=U({diag_:vN});function bN(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(su,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Hg=U({dilation2d_:bN});function wN(n,t){for(var e=n.length,i=[],r=0;r<e;r++){var a=e-1-r,s=n[a]||1,o=t[t.length-1-r]||1;o>1&&s===1&&i.unshift(a)}return i}function gt(n,t){for(var e=[],i=0;i<t.length;i++){var r=n[n.length-i-1],a=t.length-i-1,s=t[a];(r==null||r===1&&s>1)&&e.unshift(a)}return e}function et(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;r<i;r++){var a=n[n.length-r-1];a==null&&(a=1);var s=t[t.length-r-1];if(s==null&&(s=1),a===1)e.unshift(s);else if(s===1)e.unshift(a);else if(a!==s){var o="Operands could not be broadcast together with shapes "+(n+" and "+t+".");throw Error(o)}else e.unshift(a)}return e}function SN(n,t){var e,i=O(n,"a","equal"),r=O(t,"b","equal");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.equal(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Uf)}var Ga=U({equal_:SN});function LN(n,t,e){var i=O(t,"a","where"),r=O(e,"b","where"),a=O(n,"condition","where","bool"),s=et(i.shape,r.shape),o=Qs(i,s),l=Qs(r,s);a.rank===1&&E(a.shape[0]===i.shape[0],function(){return"The first dimension of `a` must match the size of `condition`."}),a.rank!==1&&Be(a.shape,l.shape,"Error in where: ");var u=function(h,d){var p=h.select(a,o,l);return d([a]),p},c={condition:a,t:o,e:l};return z.runKernelFunc(u,c,null,Ku)}var vn=U({where_:LN});function IN(n){var t=O(n,"x","zerosLike"),e={x:t};return z.runKernelFunc(function(i){return i.zerosLike(t)},e,null,mc)}var Ee=U({zerosLike_:IN});function AN(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");e=at(i,r),i=e[0],r=e[1];var a=Ie(i,r),s=Ee(a),o=Ga(r,s);return vn(o,s,a)}var Vg=U({divNoNan_:AN});function TN(n,t){var e=O(n,"t1","dot"),i=O(t,"t2","dot");E((e.rank===1||e.rank===2)&&(i.rank===1||i.rank===2),function(){return"Error in dot: inputs must all be rank 1 or 2, but got ranks "+(e.rank+" and "+i.rank+".")});var r=e.rank===1?e.size:e.shape[1],a=i.rank===1?i.size:i.shape[0];if(E(r===a,function(){return"Error in dot: inner dimensions of inputs must match, but got "+(r+" and "+a+".")}),e.rank===1&&i.rank===1){var s=V(e,[1,-1]),o=V(i,[-1,1]),l=We(s,o);return V(l,[])}else if(e.rank===1&&i.rank===2){var s=V(e,[1,-1]),o=V(i,[i.shape[0],i.shape[1]]),l=We(s,o);return V(l,[l.size])}else if(e.rank===2&&i.rank===1){var o=V(i,[-1,1]),l=We(e,o);return V(l,[l.size])}else{var o=V(i,[i.shape[0],i.shape[1]]),l=We(e,o);return l}}var qg=U({dot_:TN});function NN(n){var t=O(n,"x","elu"),e=function(r,a){var s=r.elu(t);return a([s]),s},i={x:t};return z.runKernelFunc(e,i,null,lu)}var sh=U({elu_:NN});function xN(n){var t=O(n,"x","erf");E(t.dtype==="int32"||t.dtype==="float32",function(){return"Input dtype must be `int32` or `float32`."}),t.dtype==="int32"&&(t=ue(t,"float32"));var e={x:t};return z.runKernelFunc(function(i,r){var a=i.erf(t);return r([t]),a},e,null,uu)}var Gg=U({erf_:xN});function CN(n){var t=O(n,"x","exp"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.exp(t);return r([a]),a},e,null,cu)}var yn=U({exp_:CN});function RN(n,t){t===void 0&&(t=0);var e=null,i=O(n,"x","expandDims",e);E(t<=i.rank,function(){return"Axis must be <= rank of the tensor"});var r=i.shape.slice();return t<0&&(E(-(i.rank+1)<=t,function(){return"Axis must be in the interval ["+-(i.rank+1)+", "+i.rank+"]"}),t=i.rank+t+1),r.splice(t,0,1),V(i,r)}var bn=U({expandDims_:RN});function ON(n){var t=O(n,"x","expm1"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.expm1(t);return r([t]),a},e,null,hu)}var Yg=U({expm1_:ON});function EN(n,t){var e=null,i=O(n,"x","tile",e);E(i.rank===t.length,function(){return"Error in transpose: rank of input "+i.rank+" "+("must match length of reps "+t+".")});var r=function(l,u){var c=l.tile(i,t);return u([i]),c},a=[i],s={x:i},o={reps:t};return z.runKernelFunc(r,s,null,hc,o,a)}var Yr=U({tile_:EN});function DN(n,t,e,i){i===void 0&&(i="float32"),t==null&&(t=n);for(var r=Wn([n,t],i),a=n<=t?n:t,s=0;s<a;++s)r.set(1,s,s);var o=V(r.toTensor(),[n,t]);if(e==null)return o;if(e.length===1)return Yr(bn(o,0),[e[0],1,1]);if(e.length===2)return Yr(bn(bn(o,0),0),[e[0],e[1],1,1]);if(e.length===3)return Yr(bn(bn(bn(o,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error("eye() currently supports only 1D and 2D "+("batchShapes, but received "+e.length+"D."))}var jg=U({eye_:DN});function oh(n,t,e){var i={shape:n,value:t,dtype:e};return z.runKernelFunc(function(r){return r.fill(n,t,e)},{},null,zf,i)}function kN(n){var t=O(n,"x","floor"),e={x:t};return z.runKernelFunc(function(i){return i.floor(t)},e,null,du)}var to=U({floor_:kN});var lh=30;function FN(n){return n<=lh?n:Ps(n,Math.floor(Math.sqrt(n)))}function WN(n,t){var e=!1,i;for(n<=lh?(i=n,e=!0):i=Ps(n,Math.floor(Math.sqrt(n)));!e;)i>t||i===n?e=!0:i=Ps(n,i+1);return i}function UN(n,t,e){for(var i=[],r=n.length,a=0;a<r;a++)a!==t?i.push(n[a]):i.push(e);return i}function Kg(n,t,e){for(var i=n.shape[e],r=[],a=1,s=1,o=0;o<e;o++)r.push(n.shape[o]),a*=n.shape[o];for(var o=0;o<t.rank;o++)r.push(t.shape[o]);for(var o=e+1;o<n.rank;o++)r.push(n.shape[o]),s*=n.shape[o];return{batchSize:a,sliceSize:s,dimSize:i,outputShape:r}}var BN={__proto__:null,segOpComputeOptimalWindowSize:WN,computeOutShape:UN,collectGatherOpShapeInfo:Kg};function zN(n,t,e){e===void 0&&(e=0);var i=O(n,"x","gather"),r=O(t,"indices","gather","int32"),a={x:i,indices:r},s={axis:e},o=function(l,u){var c=Qe(e,i.shape)[0],h=Kg(i,r,c),d=l.gather(i,V(r,[r.size]),c);return u([i,r]),V(d,h.outputShape)};return z.runKernelFunc(o,a,null,mu,s)}var no=U({gather_:zN});function PN(n,t){var e,i=O(n,"a","greater"),r=O(t,"b","greater");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.greater(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Mf)}var Li=U({greater_:PN});function _N(n,t){var e,i=O(n,"a","greaterEqual"),r=O(t,"b","greaterEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.greaterEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,gu)}var Zi=U({greaterEqual_:_N});function MN(n){var t=O(n,"input","imag"),e=function(r){return r.imag(t)},i={input:t};return z.runKernelFunc(e,i,null,Vf)}var io=U({imag_:MN});function HN(n){var t=O(n,"x","isFinite"),e={x:t};return z.runKernelFunc(function(i){return i.isFinite(t)},e,null,yu)}var $g=U({isFinite_:HN});function VN(n){var t=O(n,"x","isInf"),e={x:t};return z.runKernelFunc(function(i){return i.isInf(t)},e,null,bu)}var Xg=U({isInf_:VN});function qN(n){var t=O(n,"x","isNaN"),e={x:t};return z.runKernelFunc(function(i){return i.isNaN(t)},e,null,wu)}var Jg=U({isNaN_:qN});function GN(n,t){var e,i=O(n,"a","maximum"),r=O(t,"b","maximum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.maximum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Nu)}var jr=U({maximum_:GN});function be(n,t){if((Dt(n)&&t!=="string"||Array.isArray(n))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&Dt(n)&&!(n instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");var e=[],i=[];return gi(n,e,i,t)}function YN(n,t){t===void 0&&(t=.2);var e=O(n,"x","leakyRelu");return jr(J(be(t),e),e)}var Zg=U({leakyRelu_:YN});function jN(n,t){var e,i=O(n,"a","less"),r=O(t,"b","less");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.less(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,qf)}var ro=U({less_:jN});function KN(n,t){var e,i=O(n,"a","lessEqual"),r=O(t,"b","lessEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.lessEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Gf)}var Qi=U({lessEqual_:KN});function $N(n,t,e){if(e<=0)throw new Error("The number of values should be positive.");var i={start:n,stop:t,num:e};return z.runKernelFunc(function(r){return r.linspace(n,t,e)},{},null,Yf,i)}function XN(n,t,e,i,r){t===void 0&&(t=5),e===void 0&&(e=1),i===void 0&&(i=1),r===void 0&&(r=.5);var a=O(n,"x","localResponseNormalization");E(a.rank===4||a.rank===3,function(){return`Error in localResponseNormalization: x must be rank 3 or 4 but got
+               rank `+a.rank+"."}),E(rt(t),function(){return"Error in localResponseNormalization: depthRadius must be an "+("integer but got depthRadius "+t+".")});var s=a,o=!1;a.rank===3&&(o=!0,s=V(a,[1,a.shape[0],a.shape[1],a.shape[2]]));var l=function(d,p){var f=d.localResponseNormalization4D(s,t,e,i,r);return p([s,f]),f},u={x:s},c={depthRadius:t,bias:e,alpha:i,beta:r},h=z.runKernelFunc(l,u,null,Au,c);return o?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Qg=U({localResponseNormalization_:XN});function JN(n){var t=O(n,"x","log"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log(t);return r([t]),a},e,null,Su)}var er=U({log_:JN});function ZN(n){var t=O(n,"x","log1p"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log1p(t);return r([t]),a},e,null,Lu)}var uh=U({log1p_:ZN});function QN(n){return E(fi(n),function(){return"The f passed in grad(f) must be a function"}),function(t,e){var i=O(t,"x","tf.grad",null),r=e!=null?O(e,"dy","tf.grad"):null;return z.tidy(function(){var a=z.gradients(function(){return n(i)},[i],r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),ao(o),o[0]})}}function ex(n){return E(fi(n),function(){return"The f passed in grads(f) must be a function"}),function(t,e){E(Array.isArray(t),function(){return"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s"});var i=za(t,"args","tf.grads",null),r=e!=null?O(e,"dy","tf.grads"):null;return z.tidy(function(){var a=z.gradients(function(){return n.apply(void 0,i)},i,r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),ao(o),o})}}function tx(n){return E(fi(n),function(){return"The f passed in valueAndGrad(f) must be a function"}),function(t,e){E(t instanceof Y,function(){return"The x passed in valueAndGrad(f)(x) must be a tensor"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrad(f)(x, dy) must be a tensor"});var i=z.gradients(function(){return n(t)},[t],e),r=i.grads,a=i.value;return ao(r),{grad:r[0],value:a}}}function nx(n){return E(fi(n),function(){return"The f passed in valueAndGrads(f) must be a function"}),function(t,e){E(Array.isArray(t)&&t.every(function(r){return r instanceof Y}),function(){return"The args passed in valueAndGrads(f)(args) must be array of tensors"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrads(f)(args, dy) must be a tensor"});var i=z.gradients(function(){return n.apply(void 0,t)},t,e);return e!=null&&Be(i.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),ao(i.grads),i}}function ev(n,t){E(fi(n),function(){return"The f passed in variableGrads(f) must be a function"}),E(t==null||Array.isArray(t)&&t.every(function(h){return h instanceof Ba}),function(){return"The varList passed in variableGrads(f, varList) must be an array of variables"});var e=t!=null;if(!e){t=[];for(var i in z.registeredVariables)t.push(z.registeredVariables[i])}var r=e?t.filter(function(h){return!h.trainable}):null,a=t.length;t=t.filter(function(h){return h.trainable}),E(t.length>0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function zn(n){return z.customGrad(n)}function ao(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
+    the f you passed encloses all operations that lead from x to y.`)}function ix(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,ku)}var vt=U({neg_:ix});function rx(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,tc)}var ch=U({softplus_:rx});function ax(n){var t=O(n,"x","logSigmoid"),e=zn(function(i){var r=vt(ch(vt(i))),a=function(s){var o=J(s,Ji(vt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var tv=U({logSigmoid_:ax});function sx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=rn(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=Un(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=nn(f.shape,Qe(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,Tu,s)}var tr=U({max_:sx});function ox(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,lc)}var ge=U({sub_:ox});function lx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Qe(t,i.shape),c=rn(u,i.rank),h=u,d=i;c!=null&&(d=ut(i,c),h=Un(h.length,i.rank));var p=o.sum(d,h);if(e){var f=nn(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,ic,s)}var Ae=U({sum_:lx});function ux(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=tr(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),er(Ae(yn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,Iu,a)}var nv=U({logSoftmax_:ux});function cx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Qe(t,i.shape),a=tr(i,r,!0),s=ge(i,a),o=yn(s),l=Ae(o,r),u=er(l),c=fe(V(a,u.shape),u);if(e){var h=nn(c.shape,r);return V(c,h)}return c}var hh=U({logSumExp_:cx});function hx(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,jf)}var nr=U({logicalAnd_:hx});function dx(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,Kf)}var so=U({logicalNot_:dx});function px(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,$f)}var dh=U({logicalOr_:px});function fx(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return et(e.shape,i.shape),nr(dh(n,t),so(nr(n,t)))}var iv=U({logicalXor_:fx});function mx(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(Ht(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Vr(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&hn(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,xu,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var ph=U({maxPool_:mx});function gx(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:At("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Ht(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ha(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Cu,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var vx=U({maxPool3d_:gx});function yx(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(Qf,s,o);return{result:l[0],indexes:l[1]}}var bx=U({maxPoolWithArgmax_:yx});function ti(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=ti(n,"float32"),i=ti(n,"float32");return mi(e,i)}var r=Ur(ot(n),t);return z.makeTensor(r,n,t)}function Kr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Kr(n,"float32"),i=ti(n,"float32");return mi(e,i)}var r=Tc(ot(n),t);return z.makeTensor(r,n,t)}function wx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Qe(t,i.shape),a=Ag(i.shape,r),s=a[1],o=ot(s),l=zn(function(u){var c=be(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Ie(h,c),p=Ae(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Ie(J(v,Kr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var Ya=U({mean_:wx});function Sx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=rn(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=Un(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=nn(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Ru,s)}var oo=U({min_:Sx});function Lx(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ou)}var lo=U({minimum_:Lx});function Ix(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Eu)}var fh=U({mod_:Ix});function Ax(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ye=U({square_:Ax});function Tx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Qe(t,n.shape),r=Ya(n,i,e),a=r.shape;e||(a=nn(r.shape,i));var s=Ye(ge(ue(n,"float32"),V(r,a))),o=Ya(s,i,e);return{mean:r,variance:o}}var Nx=U({moments_:Tx});function xx(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=za(e,"c","multiRNNCell"),s=za(i,"h","multiRNNCell"),o=r,l=[],u=0;u<n.length;u++){var c=n[u](o,a[u],s[u]);l.push(c[0]),l.push(c[1]),o=c[1]}for(var h=[],d=[],u=0;u<l.length;u+=2)h.push(l[u]),d.push(l[u+1]);return[h,d]}var Cx=U({multiRNNCell_:xx});function Rx(n,t,e,i){i===void 0&&(i=!1);var r=O(n,"logits","multinomial"),a=r.size,s=r.rank;if(a<2)throw new Error("Error in multinomial: you need at least 2 outcomes, but got "+(a+"."));if(s>2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var Ox=U({multinomial_:Rx});function Ex(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,em)}var uo=U({notEqual_:Ex});function Dx(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,sm)}var ja=U({real_:Dx});function kx(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=mh(ja(t)),o=Ee(io(t));return mi(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,Fu)}var mh=U({onesLike_:kx});function Fx(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return We(r,a)}var Wx=U({outerProduct_:Fx});function Ux(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,Uu,a)}var ir=U({pad_:Ux});function Bx(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),ir(n,[t],e)}var zx=U({pad1d_:Bx});function Px(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),ir(n,t,e)}var _x=U({pad2d_:Px});function Mx(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),ir(n,t,e)}var Hx=U({pad3d_:Mx});function Vx(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),ir(n,t,e)}var qx=U({pad4d_:Vx});function Gx(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,rc,s)}var co=U({spaceToBatchND_:Gx});function Kx(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(Ht(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=Vr(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=jx([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=Yx([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:co(o,c,f),b=e==="avg"?function(){return nh(v,t,a,g)}:function(){return ph(v,t,a,g)},w=b(),S=d?w:Js(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function Yx(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function jx(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var rv=U({pool_:Kx});function $x(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=at(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,Bu)}var ni=U({pow_:$x});function Xx(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,zu)}var gh=U({prelu_:Xx});function Jx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Qe(t,i.shape),u=rn(l,i.rank),c=l,h=i;u!=null&&(h=ut(i,u),c=Un(c.length,i.rank));var d=o.prod(h,c);if(e){var p=nn(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,rm,s)}var av=U({prod_:Jx});function Zx(n,t,e){var i=ot(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a<i;a++)r[a]=t();return z.makeTensor(r,n,e)}var Qx=U({rand_:Zx}),$r=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function rr(n,t){return t={exports:{}},n(t,t.exports),t.exports}var eC=rr(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})($r,n,!1)}),tC=rr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})($r,n,!1)}),nC=rr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})($r,n,!1)}),iC=rr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})($r,n,!1)}),rC=rr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})($r,n,!1)}),aC=rr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})($r,n,!1)}),ar=rr(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,N){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||N||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,N=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(N||(S=[N++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%N]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,j=C.S;W--;)F=j[H=h&H+1],P=P*r+j[h&(j[H]=j[_=h&_+F])+(j[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var N=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{N.push(g(S[R],L-1))}catch(D){}return N.length?N:C=="string"?S:S+"\0"}function v(S,L){for(var N=S+"",C,R=0;R<N.length;)L[h&R]=h&(C^=L[h&R]*19)+N.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,N=L&&L.plugins;return[+new Date,i,N,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=require("crypto")}catch(S){}}})([],Math)});ar.alea=eC;ar.xor128=tC;ar.xorwow=nC;ar.xorshift7=iC;ar.xor4096=rC;ar.tychei=aC;var sC=ar,vh=sC.alea;var yh=function(){function n(t,e,i,r,a){this.mean=t,this.stdDev=e,this.dtype=i,this.nextVal=NaN,this.truncated=r,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);var s=a||Math.random();this.random=vh(s.toString())}return n.prototype.nextValue=function(){if(!isNaN(this.nextVal)){var t=this.nextVal;return this.nextVal=NaN,t}for(var e,i,r=!1;!r;){var a=void 0,s=void 0,o=void 0;do a=2*this.random()-1,s=2*this.random()-1,o=a*a+s*s;while(o>=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),oC=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=vh(a.toString()),this.randn=new yh(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a<e||Math.log(a)<i)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)},n.prototype.convertValue=function(t){return this.dtype==="float32"?t:Math.round(t)},n}(),lC=function(){function n(t,e,i,r){var a=this;if(t===void 0&&(t=0),e===void 0&&(e=1),this.canReturnFloat=function(){return a.dtype==null||a.dtype==="float32"},this.min=t,this.range=e-t,this.dtype=i,r==null&&(r=Math.random()),typeof r=="number"&&(r=r.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error("The difference between "+t+" - "+e+" <= 1 and dtype is not float");this.random=vh(r)}return n.prototype.convertValue=function(t){return this.canReturnFloat()?t:Math.round(t)},n.prototype.nextValue=function(){return this.convertValue(this.min+this.range*this.random())},n}();function uC(n,t,e,i,r){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e==null&&(e=1),i==null&&(i="float32"),i!=="float32"&&i!=="int32")throw new Error("Unsupported data type "+i);for(var a=new oC(t,e,i,r),s=Wn(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var cC=U({randomGamma_:uC});function hC(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type "+i);for(var a=new yh(t,e,i,!1,r),s=Wn(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var dC=U({randomNormal_:hC});function pC(n,t,e,i,r){t===void 0&&(t=0),e===void 0&&(e=1),i===void 0&&(i="float32");for(var a=Wn(n,i),s=new lC(t,e,null,r),o=0;o<a.values.length;o++)a.values[o]=s.nextValue();return a.toTensor()}var sv=U({randomUniform_:pC});function Xr(n,t){Yi(n);var e=Fn(n,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");var i=null;return gi(n,i,e,t)}function bh(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e===0)throw new Error("Cannot have a step of zero");var r=function(){var s=n===t,o=n<t&&e<0,l=t<n&&e>1;if(s||o||l)return ti([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=Ur(u,i);t<n&&e===1&&(e=-1),c[0]=n;for(var h=1;h<c.length;h++)c[h]=c[h-1]+e;return Xr(c,i)},a={start:n,stop:t,step:e,dtype:i};return z.runKernelFunc(r,{},null,am,a)}function fC(n){var t=O(n,"x","reciprocal"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.reciprocal(t);return r([t]),a},e,null,Pu)}var ov=U({reciprocal_:fC});function mC(n){var t=O(n,"x","relu"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu(t)},i={x:t};return z.runKernelFunc(e,i,null,_u)}var Ka=U({relu_:mC});function gC(n){var t=O(n,"x","relu6"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu6(t)},i={x:t};return z.runKernelFunc(e,i,null,qu)}var wh=U({relu6_:gC});function vC(n,t){var e=O(n,"x","reverse"),i=function(s){var o=Qe(t,e.shape);if(e.rank===0)return $i(e);var l=s.reverse(e,o);return V(l,e.shape)},r={x:e},a={dims:t};return z.runKernelFunc(i,r,null,Gu,a)}var ii=U({reverse_:vC});function yC(n){var t=O(n,"x","reverse");return E(t.rank===1,function(){return"Error in reverse1D: x must be rank 1 but got rank "+t.rank+"."}),ii(t,0)}var bC=U({reverse1d_:yC});function wC(n,t){var e=O(n,"x","reverse");return E(e.rank===2,function(){return"Error in reverse2D: x must be rank 2 but got rank "+e.rank+"."}),ii(e,t)}var SC=U({reverse2d_:wC});function LC(n,t){var e=O(n,"x","reverse");return E(e.rank===3,function(){return"Error in reverse3D: x must be rank 3 but got rank "+e.rank+"."}),ii(e,t)}var IC=U({reverse3d_:LC});function AC(n,t){var e=O(n,"x","reverse");return E(e.rank===4,function(){return"Error in reverse4D: x must be rank 4 but got rank "+e.rank+"."}),ii(e,t)}var TC=U({reverse4d_:AC});function NC(n){var t=O(n,"x","round"),e={x:t};return z.runKernelFunc(function(i){return i.round(t)},e,null,Yu)}var lv=U({round_:NC});function xC(n){var t=O(n,"x","rsqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.rsqrt(t);return r([t]),a},e,null,ju)}var Sh=U({rsqrt_:xC});function CC(n){var t=O(n,"x","selu"),e=function(r,a){var s=r.selu(t);return a([t]),s},i={x:t};return z.runKernelFunc(e,i,null,$u)}var uv=U({selu_:CC});function RC(n,t,e,i,r,a,s){a===void 0&&(a=[1,1]),s===void 0&&(s="NHWC");var o=O(n,"x","separableConv2d"),l=O(t,"depthwiseFilter","separableConv2d"),u=O(e,"pointwiseFilter","separableConv2d"),c=o,h=!1;if(o.rank===3&&(h=!0,c=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),s==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");E(c.rank===4,function(){return"Error in separableConv2d: input must be rank 4, but got "+("rank "+c.rank+".")}),E(l.rank===4,function(){return"Error in separableConv2d: depthwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.rank===4,function(){return"Error in separableConv2d: pointwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.shape[0]===1,function(){return"Error in separableConv2d: the first dimension of pointwise filter "+(" must be 1, but got "+u.shape[0]+".")}),E(u.shape[1]===1,function(){return"Error in separableConv2d: the second dimension of pointwise "+("filter must be 1, but got "+u.shape[1]+".")});var d=l.shape[2],p=l.shape[3];E(u.shape[2]===d*p,function(){return"Error in separableConv2d: the third dimension of pointwise filter "+("must be "+d*p+", ")+("but got "+u.shape[2]+".")});var f=qa(c,l,i,r,s,a),m=1,g=Gr(f,u,m,"valid",s);return h?V(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var cv=U({separableConv2d_:RC});function OC(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,l,h;return pe(this,function(d){switch(d.label){case 0:return e=O(n,"x","setdiff1d"),i=O(t,"y","setdiff1d"),E(e.dtype===i.dtype,function(){return"x and y should have the same dtype, but got x ("+e.dtype+") and y ("+i.dtype+")."}),E(e.rank===1,function(){return"x should be 1D tensor, but got x ("+e.shape+")."}),E(i.rank===1,function(){return"y should be 1D tensor, but got y ("+i.shape+")."}),[4,e.data()];case 1:return r=d.sent(),[4,i.data()];case 2:for(a=d.sent(),s=new Set(a),o=0,l=0;l<r.length;l++)s.has(r[l])||o++;for(u=new Hs([o],e.dtype),c=new Hs([o],"int32"),l=0,h=0;l<r.length;l++)s.has(r[l])||(u.values[h]=r[l],c.values[h]=l,h++);return[2,[u.toTensor(),c.toTensor()]]}})})}var EC=OC;function DC(n){var t=O(n,"x","sign"),e={x:t};return z.runKernelFunc(function(i){return i.sign(t)},e,null,Qu)}var hv=U({sign_:DC});function kC(n){var t=O(n,"x","sin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sin(t);return r([t]),a},e,null,Ju)}var Lh=U({sin_:kC});function FC(n){var t=O(n,"x","sinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sinh(t);return r([t]),a},e,null,Zu)}var Ih=U({sinh_:FC});function WC(n,t,e){var i=O(n,"x","slice1d");return E(i.rank===1,function(){return"slice1d expects a rank-1 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,[t],[e])}var UC=U({slice1d_:WC});function BC(n,t,e){var i=O(n,"x","slice2d");return E(i.rank===2,function(){return"slice2d expects a rank-2 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var zC=U({slice2d_:BC});function PC(n,t,e){var i=O(n,"x","slice3d");return E(i.rank===3,function(){return"slice3d expects a rank-3 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var _C=U({slice3d_:PC});function MC(n,t,e){var i=O(n,"x","slice4d");return E(i.rank===4,function(){return"slice4d expects a rank-4 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var HC=U({slice4d_:MC});function VC(n,t){t===void 0&&(t=-1);var e=O(n,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and dim was "+t));var i={logits:e},r={dim:t};return z.runKernelFunc(function(a,s){var o=a.softmax(e,t);return s([o]),o},i,null,sc,r)}var dv=U({softmax_:VC});function qC(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.fft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=n.as2D(r,i),s=e.fft(a);return s.reshape(n.shape)},t,null,Bf)}var ho=U({fft_:qC});function GC(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.ifft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=V(n,[r,i]),s=e.ifft(a);return V(s,n.shape)},t,null,Hf)}var $a=U({ifft_:GC});function YC(n){var t=n.shape[n.shape.length-1],e=n.size/t,i;if(t<=2){var r=V(n,[e,t]);i=$a(r)}else{var a=[e,2*(t-1)],s=V(ja(n),[e,t]),o=V(io(n),[e,t]),l=ii(ze(s,[0,1],[e,t-2]),1),u=J(ii(ze(o,[0,1],[e,t-2]),1),be(-1)),c=Rt([s,l],1),h=Rt([o,u],1),r=V(mi(c,h),[a[0],a[1]]);i=$a(r)}if(i=ja(i),n.rank===3&&n.shape[0]!==0){var d=i,p=n.shape[0];i=V(i,[p,i.shape[0]/p,i.shape[1]]),d.dispose()}return i}var Ah=U({irfft_:YC});function pv(n,t,e){e===void 0&&(e=0);var i=[];if(typeof t=="number")E(n.shape[e]%t===0,function(){return"Number of splits must evenly divide the axis."}),i=new Array(t).fill(n.shape[e]/t);else{var r=t.reduce(function(o,l){return l===-1&&(o+=1),o},0);E(r<=1,function(){return"There should be only one negative value in split array."});var a=t.indexOf(-1);if(a!==-1){var s=t.reduce(function(o,l){return l>0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function jC(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Qe(e,i.shape)[0],c=pv(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,ac,s)}var Jr=U({split_:jC});function KC(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&t<e){var a=n.shape.map(function(v){return 0}),s=n.shape.map(function(v){return v});s[n.shape.length-1]=t,r=ze(n,a,s),e=t}else if(t!=null&&t>e){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=Rt([n,ti(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(mi(r,l),[i,e]),c=ho(u),h=Math.floor(e/2)+1,d=ja(c),p=io(c),f=Jr(d,[h,e-h],d.shape.length-1),m=Jr(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(mi(f[0],m[0]),g)}var po=U({rfft_:KC});function $C(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,nc)}var Vt=U({sqrt_:$C});function XC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,oc,o)}var fo=U({squaredDifference_:XC});function JC(n,t){var e=O(n,"x","squeeze");return V(e,wm(e.shape,t).newShape)}var mo=U({squeeze_:JC});function ZC(n,t){t===void 0&&(t=0);var e=za(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return bn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Be(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return bn(o,t)});return Rt(s,t)}var sr=U({stack_:ZC});function QC(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,gc,r)}var Zr=U({step_:QC});function eR(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=js(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=js(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=mg(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var N=js(l);N.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=sg(t,e,i),R=C.filter(function(W,F){return N.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,dm,d)}var fv=U({stridedSlice_:eR});function tR(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,uc)}var mv=U({tan_:tR});function Xa(n,t,e){if(Yi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=Fn(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return gi(n,t,i,e)}function nR(n,t,e){if(Yi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=Fn(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return gi(n,t,i,e)}function iR(n,t,e){if(Yi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=Fn(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return gi(n,t,i,e)}function rR(n,t,e){if(Yi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=Fn(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,gi(n,t,i,e)}function aR(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,pm,s),l=o[0],u=o[1];return{values:l,indices:u}}var gv=U({topk_:aR});function sR(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new yh(t,e,i,!0,r),s=Wn(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var oR=U({truncatedNormal_:sR});function lR(n,t){t===void 0&&(t=0);var e=O(n,"x","unique",null);E(e.rank>0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(fm,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var vv=U({unique_:lR});function uR(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(rt(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,fc,s)}var Th=U({unsortedSegmentSum_:uR});function cR(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t<e.shape.length,function(){return"Axis = "+t+" is not in [-"+e.shape.length+", "+e.shape.length+")"}),t<0&&(t+=e.shape.length);var i={value:e},r={axis:t},a=function(s){return s.unstack(e,t)};return z.runKernelFunc(a,i,null,pc,r)}var go=U({unstack_:cR});function hR(n,t,e,i){return t===void 0&&(t=!0),z.makeVariable(n,t,e,i)}function yv(n,t){for(var e=[],i=0;i<t.length;i++)t[i]&&e.push(i);for(var r=Wn(n,"int32"),a=Wn([e.length,n.length],"int32"),i=0;i<e.length;i++){var s=r.indexToLoc(e[i]),o=i*n.length;a.values.set(s,o)}return a.toTensor()}function dR(n){return de(this,void 0,void 0,function(){var t,e,i;return pe(this,function(r){switch(r.label){case 0:return t=O(n,"condition","whereAsync","bool"),[4,t.data()];case 1:return e=r.sent(),i=yv(t.shape,e),n!==t&&t.dispose(),[2,i]}})})}var bv=dR;function pR(n,t,e){return de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:for(i=O(n,"tensor","boolMask"),r=O(t,"mask","boolMask","bool"),a=e??0,s=r.rank,o=i.shape,E(s>0,function(){return"mask cannot be scalar"}),Be(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u<a+s;u++)l*=o[u];return c=o.slice(0,a).concat([l],o.slice(a+s)),h=V(i,c),d=V(r,[-1]),[4,bv(d)];case 1:return p=g.sent(),f=mo(p,[1]),m=no(h,f,a),n!==i&&i.dispose(),t!==r&&r.dispose(),f.dispose(),h.dispose(),d.dispose(),p.dispose(),[2,m]}})})}var fR=pR;function mR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","notEqualStrict"),i=O(t,"b","notEqualStrict");return Be(e.shape,i.shape,"Error in notEqualStrict: "),uo(e,i)}function gR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessStrict"),i=O(t,"b","lessStrict");return Be(e.shape,i.shape,"Error in lessStrict: "),ro(e,i)}function vR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","equalStrict"),i=O(t,"b","equalStrict");return Be(e.shape,i.shape,"Error in equalStrict: "),Ga(e,i)}function yR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessEqualStrict"),i=O(t,"b","lessEqualStrict");return Be(e.shape,i.shape,"Error in lessEqualStrict: "),Qi(e,i)}function bR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterStrict"),i=O(t,"b","greaterStrict");return Be(e.shape,i.shape,"Error in greaterStrict: "),Li(e,i)}function wR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterEqualStrict"),i=O(t,"b","greaterEqualStrict");return Be(e.shape,i.shape,"Error in greaterEqualStrict: "),Zi(e,i)}var wv=U({equalStrict_:vR}),Sv=U({greaterEqualStrict_:wR}),Lv=U({greaterStrict_:bR}),Iv=U({lessEqualStrict_:yR}),Av=U({lessStrict_:gR}),Tv=U({notEqualStrict_:mR});function SR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","addStrict"),i=O(t,"b","addStrict");return Be(e.shape,i.shape,"Error in addStrict: "),fe(e,i)}function LR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","subStrict"),i=O(t,"b","subStrict");return Be(e.shape,i.shape,"Error in subStrict: "),ge(e,i)}function IR(n,t){return At("strict variants of ops have been deprecated and will be removed in future"),Be(n.shape,t.shape,"Error in powStrict: "),ni(n,t)}function AR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","mul"),i=O(t,"b","mul");return Be(e.shape,i.shape,"Error in multiplyStrict: "),J(e,i)}function TR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","div"),i=O(t,"b","div");return Be(e.shape,i.shape,"Error in divideStrict: "),Ie(e,i)}function NR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","modStrict"),i=O(t,"b","modStrict");return Be(e.shape,i.shape,"Error in modStrict: "),fh(e,i)}function xR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","minimumStrict"),i=O(t,"b","minimumStrict");return Be(e.shape,i.shape,"Error in minimumStrict: "),lo(e,i)}function CR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","maximumStrict"),i=O(t,"b","maximumStrict");return Be(e.shape,i.shape,"Error in maximumStrict: "),jr(e,i)}function RR(n,t){At("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","squaredDifferenceStrict"),i=O(t,"b","squaredDifferenceStrict");return Be(e.shape,i.shape,"Error in squaredDifferenceStrict: "),fo(e,i)}var Nv=U({addStrict_:SR}),xv=U({divStrict_:TR}),Cv=U({maximumStrict_:CR}),Rv=U({minimumStrict_:xR}),Ov=U({modStrict_:NR}),Ev=U({mulStrict_:AR}),Dv=U({powStrict_:IR}),kv=U({squaredDifferenceStrict_:RR}),Fv=U({subStrict_:LR});function OR(n,t,e,i){t===void 0&&(t="euclidean"),e===void 0&&(e=null),i===void 0&&(i=!1),n=O(n,"x","norm");var r=Wv(n,t,e),a=r.shape;if(i){var s=Qe(e,n.shape);a=nn(r.shape,s)}return V(r,a)}function Wv(n,t,e){if(e===void 0&&(e=null),n.rank===0)return jt(n);if(n.rank!==1&&e===null)return Wv(V(n,[-1]),t,e);if(n.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return Ae(jt(n),e);if(t===Infinity)return tr(jt(n),e);if(t===-Infinity)return oo(jt(n),e);if(t==="euclidean"||t===2)return Vt(Ae(ni(jt(n),be(2,"int32")),e));throw new Error("Error in norm: invalid ord value: "+t)}if(Array.isArray(e)&&e.length===2){if(t===1)return tr(Ae(jt(n),e[0]),e[1]-1);if(t===Infinity)return tr(Ae(jt(n),e[1]),e[0]);if(t===-Infinity)return oo(Ae(jt(n),e[1]),e[0]);if(t==="fro"||t==="euclidean")return Vt(Ae(Ye(n),e));throw new Error("Error in norm: invalid ord value: "+t)}throw new Error("Error in norm: invalid axis: "+e)}var vo=U({norm_:OR});function ER(n,t,e,i,r){r===void 0&&(r=!0);var a=O(n,"v","movingAverage"),s=O(t,"x","movingAverage"),o=O(e,"decay","movingAverage");Dm(a,s),E(hn(a.shape,s.shape),function(){return"Shape mismatch in v and x"});var l=be(1),u=ge(l,o),c=J(ge(s,a),u);if(r){E(i!=null,function(){return"When using zeroDebias: true, step is required."});var h=O(i,"step","movingAverage");c=Ie(c,ge(l,ni(o,h)))}return fe(a,c)}var DR=U({movingAverage_:ER});function kR(n,t,e){var i=O(n,"indices","scatterND","int32"),r=O(t,"updates","scatterND");jc(r,i,e);var a=function(l){return l.scatterND(i,r,e)},s={indices:i,updates:r},o={shape:e};return z.runKernelFunc(a,s,null,um,o)}var FR=U({scatterND_:kR});function WR(n,t,e,i){if(n.dtype!=="int32")throw new Error("tf.sparseToDense() expects the indices to be int32 type,"+(" but the dtype was "+n.dtype+"."));if(n.rank>2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function UR(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);WR(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,hm,l)}var BR=U({sparseToDense_:UR});function zR(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,_f)}var PR=U({gatherND_:zR});function _R(n,t){if(t==null)return n.shape.slice();if(hn(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i<n.shape.length;i++)t[i]==null&&n.shape[i]!=null?e.push(n.shape[i]):e.push(t[i]);return e}return t}function MR(n,t,e,i){var r=O(n,"x","dropout");if(E(r.dtype==="float32",function(){return"x has to be a floating point tensor since it's going to be "+("scaled, but got a "+r.dtype+" tensor instead.")}),E(t>=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=_R(r,e),s=1-t,o=Ie(to(fe(sv(a,0,1,"float32",i),s)),s);return J(r,o)}var HR=U({dropout_:MR});function Uv(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function Nh(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a<n;++a){var s=2*Math.PI*a/(n+i-1);r[a]=t-e*Math.cos(s)}return Xr(r,"float32")}function VR(n,t,e){return e===void 0&&(e=1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,g;return pe(this,function(v){switch(v.label){case 0:return i=O(n,"predictions","inTopK"),r=O(t,"targets","inTopK"),E(i.rank>1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Be(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=Bs("bool",u),d=0;d<u;d++){for(p=d*c,f=s.subarray(p,p+c),m=[],g=0;g<f.length;g++)m.push({value:f[g],index:g});for(m.sort(function(b,w){return w.value-b.value}),h[d]=0,g=0;g<e;g++)if(m[g].index===o[d]){h[d]=1;break}}return n!==i&&i.dispose(),t!==r&&r.dispose(),[2,vi(h,r.shape,"bool")]}})})}var qR=VR;function GR(n,t,e,i,r,a,s){a===void 0&&(a="NHWC");var o=n;n.rank===3&&(o=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var l=t;l.rank===3&&(l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]])),E(o.rank===4,function(){return"Error in conv2dDerFilter: input must be rank 4, but got shape "+(o.shape+".")}),E(l.rank===4,function(){return"Error in conv2dDerFilter: dy must be rank 4, but got shape "+(l.shape+".")}),E(e.length===4,function(){return"Error in conv2dDerFilter: filterShape must be length 4, but got "+(e+".")});var u=a==="NHWC"?o.shape[3]:o.shape[1],c=a==="NHWC"?l.shape[3]:l.shape[1];E(u===e[2],function(){return"Error in conv2dDerFilter: depth of input "+u+") must "+("match input depth in filter ("+e[2]+".")}),E(c===e[3],function(){return"Error in conv2dDerFilter: depth of dy ("+c+") must "+("match output depth for filter ("+e[3]+").")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerFilter: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var h=function(f){var m=1,g=_a(a),v=Bn(o.shape,e,i,m,r,s,!1,g);return f.conv2dDerFilter(o,l,v)},d={x:o,dy:l},p={strides:i,pad:r,dataFormat:a,dimRoundingMode:s};return z.runKernelFunc(h,d,null,Tf,p)}var xh=U({conv2DBackpropFilter_:GR});function yo(n,t,e){if(e==null||e==="linear")return n;if(e==="relu")return J(n,Zr(t));throw new Error("Cannot compute gradient for fused activation "+e+".")}function bo(n,t){var e=t,i=gt(n.shape,t.shape);return i.length>0&&(e=Ae(e,i)),V(e,n.shape)}function wo(n,t,e){if(t==="linear")return n;if(t==="relu")return Ka(n);if(t==="elu")return sh(n);if(t==="relu6")return wh(n);if(t==="prelu")return gh(n,e);throw new Error("Unknown fused activation "+t+".")}var So=function(n,t){var e=n>0;return!e||t==="linear"};function YR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",So(z.state.gradientDepth,d)===!1){var f=Gr(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),wo(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(rt(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(Ht(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Bn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var N=function(F,P){var H=P,_=H[0],j=H[1],K=H[2],q=H[3],G=yo(F,K,d);E(Si(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=ih(j.shape,G,_,i,r),X=xh(j,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=bo(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=zn(function(F,P,H){var _=z.runKernelFunc(C,R,null,bc,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:N}});return k(v,g)}else{var W=zn(function(F,P,H,_){var j=z.runKernelFunc(C,R,null,bc,D);return _([P,F,j,H]),b&&(j=V(j,[j.shape[1],j.shape[2],j.shape[3]])),{value:j,gradFunc:N}});return W(v,g,S)}}var jR=U({fusedConv2d_:YR});function KR(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,Of)}var Bv=U({depthwiseConv2dNativeBackpropFilter_:KR});function $R(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Ef);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var zv=U({depthwiseConv2dNativeBackpropInput_:$R});function XR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(So(z.state.gradientDepth,d)===!1){var f=qa(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),wo(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(Ht(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(rt(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Bn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var N=function(F,P){E(Si(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],j=P[2],K=P[3],q=yo(F,j,d),G=zv(_.shape,q,H,w),Z=Bv(_,q,H.shape,w);if(K!=null){var X=bo(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=zn(function(F,P,H){var _=z.runKernelFunc(C,R,null,wc,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:N}});return k(v,g)}else{var W=zn(function(F,P,H,_){var j=z.runKernelFunc(C,R,null,wc,D);return _([P,F,j,H]),b&&(j=V(j,[j.shape[1],j.shape[2],j.shape[3]])),{value:j,gradFunc:N}});return W(v,g,S)}}var JR=U({fusedDepthwiseConv2d_:XR});function ZR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(So(z.state.gradientDepth,c)===!1){var d=We(e,i,a,o);return l!=null&&(d=fe(d,l)),wo(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=at(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=ot(w),N=ot(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(hn(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[N,b,g]):V(f,[N,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=at(k,p)[0],et(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=yo(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=We(ie,X,!1,!0),re=We(Z,ie,!0,!1)):!a&&o?(te=We(ie,X,!1,!1),re=We(ie,Z,!0,!1)):a&&!o?(te=We(X,ie,!1,!0),re=We(Z,ie,!1,!1)):(te=We(X,ie,!0,!0),re=We(ie,Z,!0,!0)),l!=null){var le=bo(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var j=zn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,yc,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return j(R,D)}else{var K=zn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,yc,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return K(R,D,k)}}var QR=U({fusedMatMul_:ZR});var eO={__proto__:null,conv2d:jR,depthwiseConv2d:JR,matMul:QR};function tO(n){return Nh(n,.54,.46)}var nO=U({hammingWindow_:tO});function iO(n){return Nh(n,.5,.5)}var Pv=U({hannWindow_:iO});function rO(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a<n.size;){var o=a+t-n.size,l=Rt([ze(n,a,t-o),oh([o],r)]);s.push(l),a+=e}return s.length===0?Xa([],[0,t]):V(Rt(s),[s.length,t])}var _v=U({frame_:rO});function aO(n,t,e,i,r){r===void 0&&(r=Pv),i==null&&(i=Uv(t));for(var a=_v(n,t,e),s=J(a,r(t)),o=[],l=0;l<a.shape[0];l++)o.push(po(ze(s,[l,0],[1,t]),i));return Rt(o)}var sO=U({stft_:aO});function oO(n,t,e,i,r,a){var s=O(n,"image","cropAndResize"),o=O(t,"boxes","cropAndResize","float32"),l=O(e,"boxInd","cropAndResize","int32");r=r||"bilinear",a=a||0;var u=o.shape[0];E(s.rank===4,function(){return"Error in cropAndResize: image must be rank 4,"+("but got rank "+s.rank+".")}),E(o.rank===2&&o.shape[1]===4,function(){return"Error in cropAndResize: boxes must be have size ["+u+",4] "+("but had shape "+o.shape+".")}),E(l.rank===1&&l.shape[0]===u,function(){return"Error in cropAndResize: boxInd must be have size ["+u+"] "+("but had shape "+o.shape+".")}),E(i.length===2,function(){return"Error in cropAndResize: cropSize must be of length 2, but got "+("length "+i.length+".")}),E(i[0]>=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Cf,d);return p}var lO=U({cropAndResize_:oO});function uO(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(Pf,e,{});return i}var cO=U({flipLeftRight_:uO});function hO(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(mm,a,s);return o}var dO=U({rotateWithOffset_:hO});function Qr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function pO(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Qr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,tm,l)}var fO=U({nonMaxSuppression_:pO});function gO(n,t,e){var i=mO(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function mO(n,t,e){return yO(n,t,e||vO)}function vO(n,t){return n>t?1:n<t?-1:0}function yO(n,t,e){for(var i=0,r=n.length,a=0,s=!1;i<r;){a=i+(r-i>>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function Mv(n,t,e,i,r){return Ch(n,t,e,i,r,0).selectedIndices}function Hv(n,t,e,i,r,a){return Ch(n,t,e,i,r,0,!1,a,!0)}function Vv(n,t,e,i,r,a){return Ch(n,t,e,i,r,a,!0)}function Ch(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;c<t.length;c++)t[c]>r&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(qv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length<e&&u.length>0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m<r)break;for(var b=!1,w=d.length-1;w>=v;--w){var S=bO(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*wO(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&gO(u,f,qv))}var L=d.length,N=e-L;o&&N>0&&(d.push.apply(d,new Array(N).fill(0)),p.push.apply(p,new Array(N).fill(0)));var C={selectedIndices:Xr(d,"int32")};return s&&(C.selectedScores=Xr(p,"float32")),l&&(C.validOutputs=be(L,"int32")),C}function bO(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function wO(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function qv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function SO(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Qr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=Mv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var LO=SO;function IO(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Qr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(im,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var AO=U({nonMaxSuppressionWithScore_:IO});function TO(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Qr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=Vv(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var NO=TO;function xO(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Qr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(nm,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var CO=U({nonMaxSuppressionPadded_:xO});function RO(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Qr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=Hv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var OO=RO;function EO(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,Vu,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Gv=U({resizeBilinear_:EO});function DO(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,Hu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Yv=U({resizeNearestNeighbor_:DO});function kO(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(bh(0,s,1,"int32"),[-1,1]),u=bh(0,o,1,"int32"),c=ge(l,u),h=nr(Qi(c,be(+t,"int32")),Zi(c,be(-e,"int32"))),d=ti([s,o],i.dtype);return V(sr(go(V(i,[-1,s,o])).map(function(p){return vn(h,p,d)})),r)}var FO=U({bandPart_:kO});function WO(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r<n.length;++r)i(r)}else t=!0,n=Jr(n,n.shape[0],0).map(function(l){return mo(l,[0])});E(n.length<=n[0].shape[0],function(){return"Gram-Schmidt: Number of vectors ("+n.length+") exceeds "+("number of dimensions ("+n[0].shape[0]+").")});for(var a=[],s=n,o=function(l){a.push(z.tidy(function(){var u=s[l];if(l>0)for(var c=0;c<l;++c){var h=J(Ae(J(a[c],u)),a[c]);u=ge(u,h)}return Ie(u,vo(u,"euclidean"))}))},r=0;r<n.length;++r)o(r);return t?sr(a,0):a}var UO=U({gramSchmidt_:WO});function BO(n,t){if(t===void 0&&(t=!1),E(n.rank>=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return jv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=go(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=jv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(sr(r,0),n.shape),o=V(sr(a,0),n.shape);return[s,o]}function jv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=jg(e),a=$i(n),s=Xa([[1]],[1,1]),o=$i(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=vo(g),b=ze(a,[h,h],[1,1]),w=vn(Li(b,0),Xa([[-1]]),Xa([[1]])),S=ge(b,J(w,v)),L=Ie(g,S);L.shape[0]===1?o=$i(s):o=Rt([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var N=vt(Ie(We(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(N,o),D=ut(o);if(h===0)a=ge(C,We(R,We(D,C)));else{var k=ge(C,We(R,We(D,C)));a=Rt([ze(a,[0,0],[h,i]),k],0)}var W=ut(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,We(We(F,o),W));else{var P=ge(F,We(We(F,o),W));r=Rt([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],Mt([p,f,m])},c=0;c<l;++c)u(c);return!t&&e>i&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var zO=U({qr_:BO});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function PO(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ae(a);if(e===I.Reduction.MEAN){if(r==null)return Ya(a);var s=i.size/r.size,o=Ie(Ae(a),Ae(r));return s>1?Ie(o,be(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Ie(Ae(a),be(i.size));var l=J(r,Kr(i.shape)),u=ue(Ae(uo(l,be(0))),"float32");return Ie(Ae(a),u)}throw Error("Unknown reduction: "+e)}var ri=U({computeWeightedLoss_:PO});function _O(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Be(r.shape,a.shape,"Error in absoluteDifference: ");var o=jt(ge(r,a));return ri(o,s,i)}var MO=U({absoluteDifference_:_O});function HO(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Be(a.shape,s.shape,"Error in cosineDistance: ");var l=be(1),u=ge(l,Ae(J(a,s),e,!0));return ri(u,o,r)}var VO=U({cosineDistance_:HO});function qO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Be(r.shape,a.shape,"Error in hingeLoss: ");var o=be(1);r=ge(J(be(2),r),o);var l=Ka(ge(o,J(r,a)));return ri(l,s,i)}var GO=U({hingeLoss_:qO});function YO(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Be(a.shape,s.shape,"Error in huberLoss: ");var l=be(i),u=jt(ge(s,a)),c=lo(u,l),h=ge(u,c),d=fe(J(be(.5),Ye(c)),J(l,h));return ri(d,o,r)}var jO=U({huberLoss_:YO});function KO(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Be(a.shape,s.shape,"Error in logLoss: ");var l=be(1),u=be(i),c=vt(J(a,er(fe(s,u)))),h=J(ge(l,a),er(fe(ge(l,s),u))),d=ge(c,h);return ri(d,o,r)}var $O=U({logLoss_:KO});function XO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Be(r.shape,a.shape,"Error in meanSquaredError: ");var o=fo(r,a);return ri(o,s,i)}var JO=U({meanSquaredError_:XO});function ZO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Be(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ka(i),a=J(i,e),s=uh(yn(vt(jt(i))));return fe(ge(r,a),s)}function QO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Be(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=be(i),u=be(1),c=be(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=ZO(a,s);return ri(h,o,r)}var eE=U({sigmoidCrossEntropy_:QO});function tE(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=zn(function(r,a,s){var o=!0,l=hh(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=vt(J(u,r)),h=Ae(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=nn(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),yn(g))),J(V(p,v),ge(yn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function nE(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Be(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=be(i),u=be(1),c=be(a.shape[1]);a=fe(J(a,ge(u,l)),Ie(l,c))}var h=tE(a,s);return ri(h,o,r)}var iE=U({softmaxCrossEntropy_:nE});var rE={fft:ho,ifft:$a,rfft:po,irfft:Ah},aE={hammingWindow:nO,hannWindow:Pv,frame:_v,stft:sO},sE={flipLeftRight:cO,resizeNearestNeighbor:Yv,resizeBilinear:Gv,rotateWithOffset:dO,cropAndResize:lO,nonMaxSuppression:fO,nonMaxSuppressionAsync:LO,nonMaxSuppressionWithScore:AO,nonMaxSuppressionWithScoreAsync:NO,nonMaxSuppressionPadded:CO,nonMaxSuppressionPaddedAsync:OO},oE={bandPart:FO,gramSchmidt:UO,qr:zO},lE={absoluteDifference:MO,computeWeightedLoss:ri,cosineDistance:VO,hingeLoss:GO,huberLoss:jO,logLoss:$O,meanSquaredError:JO,sigmoidCrossEntropy:eE,softmaxCrossEntropy:iE};var Ii=function(n){Zn(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return Mt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return ev(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&Mt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:be(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(vg);Object.defineProperty(Ii,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Rh=function(n){Zn(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:mt(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:mt(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;mt(function(){var d=fe(J(c,i.rho),J(Ye(u),1-i.rho)),p=J(Ie(Vt(fe(h,i.epsilon)),Vt(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ye(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(Mt(this.accumulatedGrads.map(function(e){return e.variable})),Mt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(Ii);wi(Rh);var Oh=function(n){Zn(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:mt(function(){return oh(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;mt(function(){var h=fe(c,Ye(u));c.assign(h);var d=fe(J(Ie(u,Vt(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&Mt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(Ii);wi(Oh);var Eh=function(n){Zn(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],mt(function(){s.accBeta1=be(i).variable(),s.accBeta2=be(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);mt(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:mt(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:mt(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ye(h),1-i.beta2)),g=Ie(f,a),v=Ie(m,s);d.assign(f),p.assign(m);var b=fe(J(Ie(g,fe(Vt(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&Mt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&Mt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),mt(function(){a.accBeta1.assign(ni(a.beta1,a.iterations_+1)),a.accBeta2.assign(ni(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(Ii);wi(Eh);var Dh=function(n){Zn(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],mt(function(){o.iteration=be(0).variable(),o.accBeta1=be(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);mt(function(){var a=ge(1,i.accBeta1),s=Ie(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=jt(h),v=jr(m,g);d.assign(f),p.assign(v);var b=fe(J(Ie(s,a),Ie(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&Mt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&Mt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(Ii);wi(Dh);var Lo=function(n){Zn(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];mt(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=wg(be(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(Ii);wi(Lo);var kh=function(n){Zn(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=be(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:mt(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;mt(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&Mt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(Lo);wi(kh);var Fh=function(n){Zn(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:mt(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:mt(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:mt(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;mt(function(){var d=fe(J(c,i.decay),J(Ye(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Ie(J(u,i.learningRate),Vt(ge(d,fe(Ye(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ye(u),1-i.decay)),g=fe(J(h,i.momentum),Ie(J(u,i.learningRate),Vt(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&Mt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&Mt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&Mt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(Ii);wi(Fh);var or=function(){function n(){}return n.sgd=function(t){return new Lo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new kh(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new Fh(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new Eh(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Rh(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new Dh(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Oh(t,e)},n}();var uE={sgd:or.sgd,momentum:or.momentum,adadelta:or.adadelta,adagrad:or.adagrad,rmsprop:or.rmsprop,adamax:or.adamax,adam:or.adam};var cE=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function hE(){return new Promise(function(n){return cE(function(){return n()})})}function dE(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function pE(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s<a;++s)r=r.concat([n[s+1]/t[s],t[s]]);r=r.concat(n.slice(a+1))}return r}function fE(n,t,e){e===void 0&&(e=!0);var i=[];if(e){i.push(t);for(var r=t+1;r<n;++r)r<=2*t?(i.push(r),i.push(r-(t+1))):i.push(r)}else{for(var a=[],s=[],r=1;r<n;++r)r>=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function mE(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a<n.length;++a)a<=t.length?i?r.push(t[a-1]*n[a]):r.push(n[a]/t[a-1]):r.push(n[a]);return r}function gE(n,t){for(var e=[0],i=0;i<t;++i)e.push(n[i][0]);return e}function vE(n,t,e){for(var i=n.slice(0,1),r=0;r<e;++r)i.push(n[r+1]-t[r][0]-t[r][1]);return i}var Kv=1.7580993408473768,$v=1.0507009873554805;var yE=.3275911,bE=.254829592,wE=-.284496736,SE=1.421413741,LE=-1.453152027,IE=1.061405429;function AE(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.warn.apply(console,n)}function TE(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.log.apply(console,n)}function NE(n,t){if(n.length!==t.length)throw new Error("Cannot merge real and imag arrays of different lengths. real:"+(n.length+", imag: "+t.length+"."));for(var e=new Float32Array(n.length*2),i=0;i<e.length;i+=2)e[i]=n[i/2],e[i+1]=t[i/2];return e}function xE(n){for(var t=new Float32Array(n.length/2),e=new Float32Array(n.length/2),i=0;i<n.length;i+=2)t[i/2]=n[i],e[i/2]=n[i+1];return{real:t,imag:e}}function CE(n){for(var t=Math.ceil(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=0;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function RE(n){for(var t=Math.floor(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=2;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function OE(n,t){var e=n[t*2],i=n[t*2+1];return{real:e,imag:i}}function EE(n,t,e,i){n[i*2]=t,n[i*2+1]=e}function DE(n,t){for(var e=new Float32Array(n/2),i=new Float32Array(n/2),r=0;r<Math.ceil(n/2);r++){var a=(t?2:-2)*Math.PI*(r/n);e[r]=Math.cos(a),i[r]=Math.sin(a)}return{real:e,imag:i}}function kE(n,t,e){var i=(e?2:-2)*Math.PI*(n/t),r=Math.cos(i),a=Math.sin(i);return{real:r,imag:a}}function FE(n,t,e){if(t==="complex64"){if(n.dtype==="complex64")return n.clone();var i=ti(n.shape),r=ue(n,"float32"),a=e.complex(r,i);return i.dispose(),r.dispose(),a}if(!Am(n.dtype,t))return z.makeTensorFromDataId(n.dataId,n.shape,t);if(n.dtype==="complex64"){var s=e.real(n),a=ue(s,t);return s.dispose(),a}if(t==="int32")return e.int(n);if(t==="bool"){var o=be(0,n.dtype),a=e.notEqual(n,o);return o.dispose(),a}else throw new Error("Error in Cast: failed to cast "+n.dtype+" to "+t)}function WE(n,t){return z.makeTensorFromDataId(n.dataId,t,n.dtype)}function UE(n,t,e){var i=(t-n)/(e-1),r=Ur(e,"float32");r[0]=n;for(var a=1;a<r.length;a++)r[a]=r[a-1]+i;return Xr(r,"float32")}var BE={__proto__:null,slice_util:gg,segment_util:BN,castTensor:FE,reshapeTensor:WE,linspaceImpl:UE,upcastType:Vs,axesAreInnerMostDims:Qc,combineLocations:Ig,computeOutAndReduceShapes:Ag,expandShapeToKeepDim:nn,assertAxesAreInnerMostDims:p1,getAxesPermutation:rn,getUndoAxesPermutation:Ks,getInnerMostAxes:Un,getBroadcastDims:wN,getReductionAxes:gt,assertAndGetBroadcastShape:et,assertParamsConsistent:Fg,computeOutShape:Wg,computeDilation2DInfo:I1,computePool2DInfo:Vr,computePool3DInfo:Ha,computeConv2DInfo:Bn,computeConv3DInfo:Ma,computeDefaultPad:th,tupleValuesAreOne:Si,eitherStridesOrDilationsAreOne:Ht,convertConv2DDataFormat:_a,getFusedDyActivation:yo,getFusedBiasGradient:bo,applyActivation:wo,shouldFuse:So,PARALLELIZE_THRESHOLD:lh,computeOptimalWindowSize:FN,getImageCenter:dE,getReshaped:pE,getPermuted:fE,getReshapedPermuted:mE,getSliceBeginCoords:gE,getSliceSize:vE,prepareAndValidate:ig,validateUpdateShape:Yc,validateInput:jc,calculateShapes:rg,SELU_SCALEALPHA:Kv,SELU_SCALE:$v,ERF_P:yE,ERF_A1:bE,ERF_A2:wE,ERF_A3:SE,ERF_A4:LE,ERF_A5:IE,warn:AE,log:TE,mergeRealAndImagArrays:NE,splitRealAndImagArrays:xE,complexWithEvenIndex:CE,complexWithOddIndex:RE,getComplexWithIndex:OE,assignToTypedArray:EE,exponents:DE,exponent:kE,prepareSplitSize:pv};function zE(n,t,e){var i=new Array(n.rank).fill(0),r=n.shape.slice();return t.map(function(a){var s=r.slice();s[e]=a;var o=ze(n,i,s);return i[e]+=a,o})}function PE(n,t){for(var e=new Array(n.rank),i=0;i<e.length;i++)e[i]=n.shape[i]*t[i];for(var r=Wn(e,n.dtype),i=0;i<r.values.length;++i){for(var a=r.indexToLoc(i),s=new Array(n.rank),o=0;o<s.length;o++)s[o]=a[o]%n.shape[o];var l=n.locToIndex(s);r.values[i]=n.values[l]}return r.toTensor()}function _E(n,t,e,i,r){for(var a=t[t.length-1],s=[n.length/a,a],o=s[0],l=s[1],u=Bs(e,o*i),c=Bs("int32",o*i),h=0;h<o;h++){for(var d=h*l,p=n.subarray(d,d+l),f=[],m=0;m<p.length;m++)f.push({value:p[m],index:m});f.sort(function(S,L){return L.value-S.value});for(var g=h*i,v=u.subarray(g,g+i),b=c.subarray(g,g+i),m=0;m<i;m++)v[m]=f[m].value,b[m]=f[m].index}var w=t.slice();return w[w.length-1]=i,[vi(u,w,e),vi(c,w,"int32")]}var ME={__proto__:null,nonMaxSuppressionV3Impl:Mv,nonMaxSuppressionV4Impl:Hv,nonMaxSuppressionV5Impl:Vv,split:zE,tile:PE,topkImpl:_E,whereImpl:yv};var HE={kernelName:Fl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Zr(ue(e,"float32"),-1))}}}};var VE={kernelName:Wl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Ye(ue(e,"float32")),r=Vt(ge(be(1),i));return vt(Ie(n,r))}}}};var qE={kernelName:Ul,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Vt(ge(Ye(ue(e,"float32")),1));return Ie(n,i)}}}};var GE={kernelName:Fs,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=gt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=gt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,i.shape)};return{a,b:s}}};var YE={kernelName:Bl,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var jE={kernelName:zl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var KE={kernelName:Pl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var $E={kernelName:_l,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Vt(ge(be(1),Ye(ue(e,"float32")))))}}}};var XE={kernelName:Ml,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Vt(fe(be(1),Ye(ue(e,"float32"))));return Ie(n,i)}}}};var JE={kernelName:ql,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=fe(Ye(e),Ye(i)),l=J(n,Ie(i,o)),u=gt(e.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,e.shape)},s=function(){var o=fe(Ye(e),Ye(i)),l=vt(J(n,Ie(e,o))),u=gt(i.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,i.shape)};return{a,b:s}}};var ZE={kernelName:Hl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(Ye(ue(e,"float32")),1))}}}};var QE={kernelName:Vl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ge(be(1),Ye(ue(e,"float32"))))}}}};function eD(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(Ht(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(rt(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=Ha(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,If,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var tD=U({avgPool3dBackprop_:eD});var nD={kernelName:Yl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return tD(n,i,a,s,c,l,u)}}}};function iD(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=Vr(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,Lf,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var rD=U({avgPoolBackprop_:iD});var aD={kernelName:Gl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return rD(n,i,a,s,o)}}}};var sD={kernelName:jl,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return We(n,a,!1,!0)},b:function(){return We(r,n,!0,!1)}}:!o&&l?{a:function(){return We(n,a,!1,!1)},b:function(){return We(n,r,!0,!1)}}:o&&!l?{a:function(){return We(a,n,!1,!0)},b:function(){return We(r,n,!1,!1)}}:{a:function(){return We(a,n,!0,!0)},b:function(){return We(n,r,!0,!0)}}}};var oD={kernelName:Kl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return co(n,r,a)}}}};var lD={kernelName:$l,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o<s.length;o++)s[o]>1&&l.push(o);return{x:function(){return Ae(n,l,!0)}}}};var uD={kernelName:Ws,gradFunc:function(n){return{x:function(){return n.clone()}}}};var cD={kernelName:Xl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var hD={kernelName:Jl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return vn(nr(Zi(i,a),Qi(i,s)),n,Ee(n))}}}};var dD={kernelName:Zl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Qe(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=Jr(n,s,a);return o.map(function(l){return function(){return l}})}};var pD={kernelName:Ql,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(Si(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return ih(r.shape,n,a,l,u,c)},filter:function(){return xh(r,n,a.shape,l,u,c)}}}};var fD={kernelName:eu,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Gr(n,a,o,l,u,1,c)},filter:function(){return xh(n,r,a.shape,o,l,u,c)}}}};function mD(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=Ma(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,Nf,u)}var gD=U({conv3DBackpropFilter_:mD});var vD={kernelName:tu,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(Si(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return _g(o.shape,n,l,a,s)},filter:function(){return gD(o,n,l.shape,a,s)}}}};var yD={kernelName:nu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(vt(Lh(ue(e,"float32"))),n)}}}};var bD={kernelName:iu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Ih(ue(e,"float32")),n)}}}};var wD={kernelName:ru,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=rn([a],i.rank),u=ah(n,a,s,!o);return l!=null&&(u=ut(u,l)),u}}}};var SD={kernelName:au,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(Si(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(Ht(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be  1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(rt(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Bn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return zv(c.shape,n,h,d)},filter:function(){return Bv(c,n,h.shape,d)}}}};var LD={kernelName:su,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(kf,s,e)},filter:function(){return z.runKernel(Ff,o,e)}}}};var ID={kernelName:ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=gt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=gt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return vt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var AD={kernelName:lu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Wf)}}}};var TD={kernelName:uu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(yn(vt(Ye(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var ND={kernelName:cu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var xD={kernelName:hu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,yn(e))}}}};var CD={kernelName:du,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var RD={kernelName:pu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=gt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=gt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return vt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var OD={kernelName:fu,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??be(1),u=gt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h<r.shape.length-1;++h)c.push(r.shape[h]);c.push(1)}var d=ge(r,a),p=J(n,l),f=Sh(fe(s,be(i))),m=J(J(J(f,f),f),be(-.5)),g=function(){return a.rank===1?V(J(J(n,Yr(V(f,[1,1,1,a.shape[0]]),c)),l),r.shape):V(J(J(n,f),l),r.shape)},v=function(){var L=J(J(f,be(-1)),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},b=function(){var L=J(J(m,d),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},w=function(){var L=J(d,f),N=J(n,L);return a.rank===1&&(N=Ae(N,u)),V(N,a.shape)},S=function(){var L=n;return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)};return{x:g,mean:v,variance:b,scale:w,offset:S}}};var ED={kernelName:mu,inputsToSave:["x","indices"],gradFunc:function(n,t,e){var i=t[0],r=t[1],a=e.axis,s=Qe(a,i.shape)[0],o=function(){var l=i.shape,u=r.size,c=l.slice(0,s),h=c.length,d=l.slice(a,l.length).slice(1),p=d.length,f=Xv(0,h),m=Xv(h+1,h+1+p),g=Jv([c,[u],d]),v=V(n,g),b=V(r,[u]),w=Jv([[h],f,m]),S=ut(v,w),L=Th(S,b,i.shape[s]),N=Ks(w);return L=ut(L,N),L};return{x:o,indices:function(){return r}}}};function Xv(n,t){for(var e=[],i=n;i<t;++i)e.push(i);return e}function Jv(n){for(var t=[],e=0;e<n.length;++e)for(var i=0;i<n[e].length;++i)t.push(n[e][i]);return t}var DD={kernelName:gu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1];return{a:function(){return Ee(e)},b:function(){return Ee(i)}}}};var kD={kernelName:vu,gradFunc:function(n){return{x:function(){return ue(n,"float32")}}}};var FD={kernelName:yu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var WD={kernelName:bu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var UD={kernelName:wu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var BD={kernelName:Lu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(e,1))}}}};var zD={kernelName:Su,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ue(e,"float32"))}}}};var PD={kernelName:Iu,inputsToSave:[],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.axis;return{logits:function(){var a=!0,s=yn(i);return ge(n,J(Ae(n,r,a),s))}}}};function _D(n,t,e,i,r,a,s){i===void 0&&(i=5),r===void 0&&(r=1),a===void 0&&(a=1),s===void 0&&(s=.5);var o=function(c){return c.LRNGrad(e,n,t,i,r,a,s)},l={x:n,y:t,dy:e},u={depthRadius:i,bias:r,alpha:a,beta:s};return z.runKernelFunc(o,l,null,Xf,u)}var MD=U({localResponseNormalizationBackprop_:_D});var HD={kernelName:Au,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.depthRadius,l=s.bias,u=s.alpha,c=s.beta;return{x:function(){return MD(r,a,n,o,l,u,c)}}}};function Zv(n,t,e,i,r){return t.rank<e.rank&&(t=V(t,nn(t.shape,i))),n.rank<e.rank&&(n=V(n,nn(n.shape,i))),{x:function(){var a=J(n,ue(Ga(e,t),n.dtype));return r==null?a:ut(a,r)}}}var Qv={kernelName:Tu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.reductionIndices,a=t[0],s=t[1],o=Qe(r,a.shape),l=rn(o,a.rank),u=Zv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var VD={kernelName:Nu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Zi(e,i),"float32"))},a=function(){return J(n,ue(ro(e,i),"float32"))};return{a:r,b:a}}};function qD(n,t,e,i,r,a,s,o){a===void 0&&(a=[1,1,1]);var l=O(n,"dy","maxPool3dBackprop"),u=O(t,"input","maxPool3dBackprop"),c=O(e,"output","maxPool3dBackprop"),h=l,d=u,p=c,f=!1;u.rank===4&&(f=!0,h=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=V(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),p=V(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),E(h.rank===5,function(){return"Error in maxPool3dBackprop: dy must be rank 5 but got rank "+(h.rank+".")}),E(d.rank===5,function(){return"Error in maxPool3dBackprop: input must be rank 5 but got rank "+(d.rank+".")}),E(p.rank===5,function(){return"Error in maxPool3dBackprop: output must be rank 5 but got rank "+(p.rank+".")}),E(Ht(r,a),function(){return"Error in maxPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+r+" and dilations '"+a+"'")}),o!=null&&E(rt(s),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+o+" but got pad "+s+".")});var m=function(w){var S=Ha(d.shape,i,r,a,s,o);return w.maxPool3dBackprop(h,d,p,S)},g={dy:h,input:d,output:p},v={filterSize:i,strides:r,dilations:a,pad:s,dimRoundingMode:o},b=z.runKernelFunc(m,g,null,Zf,v);return f?V(b,[b.shape[1],b.shape[2],b.shape[3],b.shape[4]]):b}var GD=U({maxPool3dBackprop_:qD});var YD={kernelName:Cu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.dilations,c=s.pad,h=s.dimRoundingMode,d=u??[1,1,1];return{x:function(){return GD(n,r,a,o,l,d,c,h)}}}};function jD(n,t,e,i,r,a,s){var o=O(n,"dy","maxPoolBackprop"),l=O(t,"input","maxPoolBackprop"),u=O(e,"output","maxPoolBackprop");E(l.rank===o.rank,function(){return"Rank of input ("+l.rank+") does not match rank of dy "+("("+o.rank+")")}),E(o.rank===4,function(){return"Error in maxPoolBackprop: dy must be rank 4 but got rank "+(o.rank+".")}),E(l.rank===4,function(){return"Error in maxPoolBackprop: input must be rank 4 but got rank "+(l.rank+".")}),s!=null&&E(rt(a),function(){return"Error in maxPoolBackprop: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+a+".")});var c=function(p){var f=Vr(l.shape,i,r,1,a,s);return p.maxPoolBackprop(o,l,u,f)},h={dy:o,input:l,output:u},d={filterSize:i,strides:r,pad:a,dimRoundingMode:s};return z.runKernelFunc(c,h,null,Jf,d)}var KD=U({maxPoolBackprop_:jD});var $D={kernelName:xu,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.pad;return{x:function(){return KD(n,r,a,o,l,u)}}}};var XD={kernelName:Ru,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.axis,a=t[0],s=t[1],o=Qe(r,a.shape),l=rn(o,a.rank),u=Zv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var JD={kernelName:Ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Qi(e,i),"float32"))},a=function(){return J(n,ue(Li(e,i),"float32"))};return{a:r,b:a}}};var ZD={kernelName:Eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=gt(e.shape,r);return o.length>0?V(Ae(n,o),e.shape):n},s=function(){var o=J(n,vt(to(Ie(e,i)))),l=gt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var QD={kernelName:Du,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=gt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=gt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var ek={kernelName:ku,gradFunc:function(n){return{x:function(){return vt(n)}}}};var tk={kernelName:Wu,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return ti(e.shape,"float32")}}}};var nk={kernelName:Fu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ey={kernelName:Uu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var ik={kernelName:Bu,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=et(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,ni(a,ge(c,be(1))))),d=gt(a.shape,o);return d.length>0&&(h=Ae(h,d)),V(h,a.shape)},u=function(){var c=Li(a,0),h=vn(c,er(a),Ee(a)),d=J(n,J(r,h)),p=gt(s.shape,o);return p.length>0&&(d=Ae(d,p)),V(d,s.shape)};return{a:l,b:u}}};var rk={kernelName:zu,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=Li(e,0);return{x:function(){return vn(r,n,J(n,i))},alpha:function(){var a=vn(r,Ee(n),J(n,e)),s=gt(i.shape,n.shape);return s.length>0&&(a=Ae(a,s)),V(a,i.shape)}}}};var ak={kernelName:Pu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,vt(Ye(e)))}}}};var sk={kernelName:qu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(Qi(e,6),Zr(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var ok={kernelName:_u,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue(Zr(e),"float32"))}}}};var lk={kernelName:Mu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var uk={kernelName:Vu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,lm,e)};return{images:s}}};var ck={kernelName:Hu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,om,e)};return{images:s}}};var hk={kernelName:Gu,gradFunc:function(n,t,e){var i=e.dims,r=Qe(i,n.shape);return{x:function(){return ii(n,r)}}}};var dk={kernelName:Yu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var pk={kernelName:ju,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return vt(Ie(n,J(ni(e,1.5),2)))}}}};var fk={kernelName:Ku,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(so(e),n.dtype))}}}};var mk={kernelName:$u,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Li(e,be(0)),r=be(Kv),a=be($v),s=J(n,a),o=J(J(n,r),yn(ue(e,"float32")));return vn(i,s,o)}}}};var gk={kernelName:ec,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(be(1),e)))}}}};var vk={kernelName:Qu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var yk={kernelName:Ju,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(eo(ue(e,"float32")),n)}}}};var bk={kernelName:Zu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(rh(ue(e,"float32")),n)}}}};var wk={kernelName:Xu,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=Kc(i,a,s),u=l[0],c=l[1],h=[],d=0;d<n.rank;d++)h.push([u[d],o[d]-u[d]-c[d]]);return{x:function(){return ir(n,h)}}}};var Sk={kernelName:sc,outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.dim,a=!0,s=J(n,i);return{logits:function(){return ge(s,J(Ae(s,[r],a),i))}}}};var Lk={kernelName:tc,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Ji(e))}}}};var ty={kernelName:rc,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.paddings;return{x:function(){return Js(n,r,a)}}}};var ny={kernelName:ac,gradFunc:function(n,t,e){var i=e.axis;return{x:function(){return Rt(n,i)}}}};var Ik={kernelName:nc,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,J(Vt(ue(e,"float32")),2))}}}};var Ak={kernelName:cm,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(ue(e,"float32"),2))}}}};var Tk={kernelName:oc,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=be(2),a=function(){return J(n,J(r,ge(e,i)))},s=function(){return J(n,J(r,ge(i,e)))};return{a,b:s}}};var Nk={kernelName:gc,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var xk={kernelName:lc,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=gt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=gt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(vt(o),i.shape)};return{a,b:s}}};var Ck={kernelName:ic,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Qe(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Kr(i.shape,"float32"));return{x:function(){return l}}}};var Rk={kernelName:uc,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Ye(eo(e)))}}}};var Ok={kernelName:cc,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(be(1),Ye(e)),n)}}}};var Ek={kernelName:hc,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o<r[0];++o)s=fe(s,ze(n,[o*i.shape[0]],[i.shape[0]]));else if(i.rank===2)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1]],[i.shape[0],i.shape[1]]));else if(i.rank===3)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2]],[i.shape[0],i.shape[1],i.shape[2]]));else if(i.rank===4)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)for(var c=0;c<r[3];++c)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2],c*i.shape[3]],[i.shape[0],i.shape[1],i.shape[2],i.shape[3]]));else throw new Error("Gradient for tile operation is not implemented for rank-"+(i.rank+" tensors yet."));return s};return{x:a}}};var Dk={kernelName:dc,gradFunc:function(n,t,e){var i=e,r=i.perm,a=Ks(r);return{x:function(){return ut(n,a)}}}};var kk={kernelName:pc,gradFunc:function(n,t,e){var i=e,r=i.axis;return{value:function(){return sr(n,r)}}}};var Wk={kernelName:fc,inputsToSave:["segmentIds"],gradFunc:function(n,t){var e=t[0],i=function(){return Fk(n,e)};return{x:i}}};function Fk(n,t){for(var e=jr(t,Ee(t)),i=no(n,e),r=Zi(t,be(0,"int32")),a=i.rank-r.rank,s=0;s<a;++s)r=bn(r,s+1);r=nr(r,Kr(i.shape,"bool"));var o=Ee(i);return vn(r,i,o)}var Uk={kernelName:mc,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var Bk=[HE,VE,qE,GE,YE,jE,KE,$E,XE,JE,ZE,QE,nD,aD,sD,oD,lD,uD,cD,hD,dD,fD,pD,vD,yD,bD,wD,SD,LD,ID,AD,TD,ND,xD,RD,CD,OD,ED,DD,kD,FD,WD,UD,BD,zD,PD,HD,Qv,Qv,VD,YD,$D,XD,JD,ZD,QD,ek,tk,nk,ey,ey,ik,rk,ak,sk,ok,lk,uk,ck,hk,dk,pk,fk,mk,gk,vk,yk,bk,wk,Sk,Lk,ty,ty,ny,ny,Ik,Tk,Ak,Nk,xk,Ck,Rk,Ok,Ek,Dk,kk,Wk,Uk];for(var Wh=0,iy=Bk;Wh<iy.length;Wh++){var zk=iy[Wh];vm(zk)}Y.prototype.abs=function(){return this.throwIfDisposed(),jt(this)};Y.prototype.acos=function(){return this.throwIfDisposed(),Sg(this)};Y.prototype.acosh=function(){return this.throwIfDisposed(),Lg(this)};Y.prototype.addStrict=function(n){return this.throwIfDisposed(),Nv(this,n)};Y.prototype.add=function(n){return this.throwIfDisposed(),fe(this,n)};Y.prototype.all=function(n,t){return this.throwIfDisposed(),Tg(this,n,t)};Y.prototype.any=function(n,t){return this.throwIfDisposed(),Ng(this,n,t)};Y.prototype.argMax=function(n){return this.throwIfDisposed(),xg(this,n)};Y.prototype.argMin=function(n){return this.throwIfDisposed(),Cg(this,n)};Y.prototype.asScalar=function(){return this.throwIfDisposed(),E(this.size===1,function(){return"The array must have only 1 element."}),V(this,[])};Y.prototype.asType=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.as1D=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.as2D=function(n,t){return this.throwIfDisposed(),V(this,[n,t])};Y.prototype.as3D=function(n,t,e){return this.throwIfDisposed(),V(this,[n,t,e])};Y.prototype.as4D=function(n,t,e,i){return this.throwIfDisposed(),V(this,[n,t,e,i])};Y.prototype.as5D=function(n,t,e,i,r){return this.throwIfDisposed(),V(this,[n,t,e,i,r])};Y.prototype.asin=function(){return this.throwIfDisposed(),Rg(this)};Y.prototype.asinh=function(){return this.throwIfDisposed(),Og(this)};Y.prototype.atan=function(){return this.throwIfDisposed(),Eg(this)};Y.prototype.atan2=function(n){return this.throwIfDisposed(),Dg(this,n)};Y.prototype.atanh=function(){return this.throwIfDisposed(),kg(this)};Y.prototype.avgPool=function(n,t,e,i){return this.throwIfDisposed(),nh(this,n,t,e,i)};Y.prototype.batchToSpaceND=function(n,t){return this.throwIfDisposed(),Js(this,n,t)};Y.prototype.batchNorm=function(n,t,e,i,r){return this.throwIfDisposed(),Va(this,n,t,e,i,r)};Y.prototype.broadcastTo=function(n){return this.throwIfDisposed(),Qs(this,n)};Y.prototype.cast=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.ceil=function(){return this.throwIfDisposed(),Ug(this)};Y.prototype.clipByValue=function(n,t){return this.throwIfDisposed(),Bg(this,n,t)};Y.prototype.concat=function(n,t){return this.throwIfDisposed(),n instanceof Y&&(n=[n]),Rt([this].concat(n),t)};Y.prototype.conv1d=function(n,t,e,i,r,a){return this.throwIfDisposed(),zg(this,n,t,e,i,r,a)};Y.prototype.conv2dTranspose=function(n,t,e,i,r){return this.throwIfDisposed(),Pg(this,n,t,e,i,r)};Y.prototype.conv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Gr(this,n,t,e,i,r,a)};Y.prototype.cos=function(){return this.throwIfDisposed(),eo(this)};Y.prototype.cosh=function(){return this.throwIfDisposed(),rh(this)};Y.prototype.cumsum=function(n,t,e){return this.throwIfDisposed(),ah(this,n,t,e)};Y.prototype.depthToSpace=function(n,t){return this.throwIfDisposed(),Mg(this,n,t)};Y.prototype.depthwiseConv2D=function(n,t,e,i,r,a){return At("depthwiseConv2D is deprecated, use depthwiseConv2d instead"),this.throwIfDisposed(),qa(this,n,t,e,i,r,a)};Y.prototype.depthwiseConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),qa(this,n,t,e,i,r,a)};Y.prototype.dilation2d=function(n,t,e,i,r){return this.throwIfDisposed(),Hg(this,n,t,e,i,r)};Y.prototype.divNoNan=function(n){return this.throwIfDisposed(),Vg(this,n)};Y.prototype.divStrict=function(n){return this.throwIfDisposed(),xv(this,n)};Y.prototype.div=function(n){return this.throwIfDisposed(),Ie(this,n)};Y.prototype.dot=function(n){return this.throwIfDisposed(),qg(this,n)};Y.prototype.elu=function(){return this.throwIfDisposed(),sh(this)};Y.prototype.equalStrict=function(n){return this.throwIfDisposed(),wv(this,n)};Y.prototype.equal=function(n){return this.throwIfDisposed(),Ga(this,n)};Y.prototype.erf=function(){return this.throwIfDisposed(),Gg(this)};Y.prototype.exp=function(){return this.throwIfDisposed(),yn(this)};Y.prototype.expandDims=function(n){return this.throwIfDisposed(),bn(this,n)};Y.prototype.expm1=function(){return this.throwIfDisposed(),Yg(this)};Y.prototype.fft=function(){return this.throwIfDisposed(),ho(this)};Y.prototype.flatten=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.floor=function(){return this.throwIfDisposed(),to(this)};Y.prototype.floorDiv=function(n){return this.throwIfDisposed(),Zc(this,n)};Y.prototype.gather=function(n,t){return this.throwIfDisposed(),no(this,n,t)};Y.prototype.greaterEqualStrict=function(n){return this.throwIfDisposed(),Sv(this,n)};Y.prototype.greaterEqual=function(n){return this.throwIfDisposed(),Zi(this,n)};Y.prototype.greaterStrict=function(n){return this.throwIfDisposed(),Lv(this,n)};Y.prototype.greater=function(n){return this.throwIfDisposed(),Li(this,n)};Y.prototype.ifft=function(){return this.throwIfDisposed(),$a(this)};Y.prototype.irfft=function(){return this.throwIfDisposed(),Ah(this)};Y.prototype.isFinite=function(){return this.throwIfDisposed(),$g(this)};Y.prototype.isInf=function(){return this.throwIfDisposed(),Xg(this)};Y.prototype.isNaN=function(){return this.throwIfDisposed(),Jg(this)};Y.prototype.leakyRelu=function(n){return this.throwIfDisposed(),Zg(this,n)};Y.prototype.lessEqualStrict=function(n){return this.throwIfDisposed(),Iv(this,n)};Y.prototype.lessEqual=function(n){return this.throwIfDisposed(),Qi(this,n)};Y.prototype.lessStrict=function(n){return this.throwIfDisposed(),Av(this,n)};Y.prototype.less=function(n){return this.throwIfDisposed(),ro(this,n)};Y.prototype.localResponseNormalization=function(n,t,e,i){return this.throwIfDisposed(),Qg(this,n,t,e,i)};Y.prototype.logSigmoid=function(){return this.throwIfDisposed(),tv(this)};Y.prototype.logSoftmax=function(n){return this.throwIfDisposed(),nv(this,n)};Y.prototype.logSumExp=function(n,t){return this.throwIfDisposed(),hh(this,n,t)};Y.prototype.log=function(){return this.throwIfDisposed(),er(this)};Y.prototype.log1p=function(){return this.throwIfDisposed(),uh(this)};Y.prototype.logicalAnd=function(n){return this.throwIfDisposed(),nr(this,n)};Y.prototype.logicalNot=function(){return this.throwIfDisposed(),so(this)};Y.prototype.logicalOr=function(n){return this.throwIfDisposed(),dh(this,n)};Y.prototype.logicalXor=function(n){return this.throwIfDisposed(),iv(this,n)};Y.prototype.matMul=function(n,t,e){return this.throwIfDisposed(),We(this,n,t,e)};Y.prototype.maxPool=function(n,t,e,i){return this.throwIfDisposed(),ph(this,n,t,e,i)};Y.prototype.max=function(n,t){return this.throwIfDisposed(),tr(this,n,t)};Y.prototype.maximumStrict=function(n){return this.throwIfDisposed(),Cv(this,n)};Y.prototype.maximum=function(n){return this.throwIfDisposed(),jr(this,n)};Y.prototype.mean=function(n,t){return this.throwIfDisposed(),Ya(this,n,t)};Y.prototype.min=function(n,t){return this.throwIfDisposed(),oo(this,n,t)};Y.prototype.minimumStrict=function(n){return this.throwIfDisposed(),Rv(this,n)};Y.prototype.minimum=function(n){return this.throwIfDisposed(),lo(this,n)};Y.prototype.modStrict=function(n){return this.throwIfDisposed(),Ov(this,n)};Y.prototype.mod=function(n){return this.throwIfDisposed(),fh(this,n)};Y.prototype.mulStrict=function(n){return this.throwIfDisposed(),Ev(this,n)};Y.prototype.mul=function(n){return this.throwIfDisposed(),J(this,n)};Y.prototype.neg=function(){return this.throwIfDisposed(),vt(this)};Y.prototype.norm=function(n,t,e){return this.throwIfDisposed(),vo(this,n,t,e)};Y.prototype.notEqualStrict=function(n){return this.throwIfDisposed(),Tv(this,n)};Y.prototype.notEqual=function(n){return this.throwIfDisposed(),uo(this,n)};Y.prototype.oneHot=function(n,t,e){return t===void 0&&(t=1),e===void 0&&(e=0),this.throwIfDisposed(),Ys(this,n,t,e)};Y.prototype.onesLike=function(){return this.throwIfDisposed(),mh(this)};Y.prototype.pad=function(n,t){return this.throwIfDisposed(),ir(this,n,t)};Y.prototype.pool=function(n,t,e,i,r){return this.throwIfDisposed(),rv(this,n,t,e,i,r)};Y.prototype.powStrict=function(n){return this.throwIfDisposed(),Dv(this,n)};Y.prototype.pow=function(n){return this.throwIfDisposed(),ni(this,n)};Y.prototype.prelu=function(n){return this.throwIfDisposed(),gh(this,n)};Y.prototype.prod=function(n,t){return this.throwIfDisposed(),av(this,n,t)};Y.prototype.reciprocal=function(){return this.throwIfDisposed(),ov(this)};Y.prototype.relu=function(){return this.throwIfDisposed(),Ka(this)};Y.prototype.relu6=function(){return this.throwIfDisposed(),wh(this)};Y.prototype.reshapeAs=function(n){return this.throwIfDisposed(),V(this,n.shape)};Y.prototype.reshape=function(n){return this.throwIfDisposed(),V(this,n)};Y.prototype.resizeBilinear=function(n,t){return this.throwIfDisposed(),Gv(this,n,t)};Y.prototype.resizeNearestNeighbor=function(n,t){return this.throwIfDisposed(),Yv(this,n,t)};Y.prototype.reverse=function(n){return this.throwIfDisposed(),ii(this,n)};Y.prototype.rfft=function(){return this.throwIfDisposed(),po(this)};Y.prototype.round=function(){return this.throwIfDisposed(),lv(this)};Y.prototype.rsqrt=function(){return this.throwIfDisposed(),Sh(this)};Y.prototype.selu=function(){return this.throwIfDisposed(),uv(this)};Y.prototype.separableConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),cv(this,n,t,e,i,r,a)};Y.prototype.sigmoid=function(){return this.throwIfDisposed(),Ji(this)};Y.prototype.sign=function(){return this.throwIfDisposed(),hv(this)};Y.prototype.sin=function(){return this.throwIfDisposed(),Lh(this)};Y.prototype.sinh=function(){return this.throwIfDisposed(),Ih(this)};Y.prototype.slice=function(n,t){return this.throwIfDisposed(),ze(this,n,t)};Y.prototype.softmax=function(n){return this.throwIfDisposed(),dv(this,n)};Y.prototype.softplus=function(){return this.throwIfDisposed(),ch(this)};Y.prototype.spaceToBatchND=function(n,t){return this.throwIfDisposed(),co(this,n,t)};Y.prototype.split=function(n,t){return this.throwIfDisposed(),Jr(this,n,t)};Y.prototype.sqrt=function(){return this.throwIfDisposed(),Vt(this)};Y.prototype.square=function(){return this.throwIfDisposed(),Ye(this)};Y.prototype.squaredDifference=function(n){return this.throwIfDisposed(),fo(this,n)};Y.prototype.squaredDifferenceStrict=function(n){return this.throwIfDisposed(),kv(this,n)};Y.prototype.squeeze=function(n){return this.throwIfDisposed(),mo(this,n)};Y.prototype.stack=function(n,t){this.throwIfDisposed();var e=n instanceof Y?[this,n]:[this].concat(n);return sr(e,t)};Y.prototype.step=function(n){return this.throwIfDisposed(),Zr(this,n)};Y.prototype.stridedSlice=function(n,t,e,i,r,a,s,o){return this.throwIfDisposed(),fv(this,n,t,e,i,r,a,s,o)};Y.prototype.subStrict=function(n){return this.throwIfDisposed(),Fv(this,n)};Y.prototype.sub=function(n){return this.throwIfDisposed(),ge(this,n)};Y.prototype.sum=function(n,t){return this.throwIfDisposed(),Ae(this,n,t)};Y.prototype.tan=function(){return this.throwIfDisposed(),mv(this)};Y.prototype.tanh=function(){return this.throwIfDisposed(),Xs(this)};Y.prototype.tile=function(n){return this.throwIfDisposed(),Yr(this,n)};Y.prototype.toBool=function(){return this.throwIfDisposed(),ue(this,"bool")};Y.prototype.toFloat=function(){return this.throwIfDisposed(),ue(this,"float32")};Y.prototype.toInt=function(){return this.throwIfDisposed(),ue(this,"int32")};Y.prototype.topk=function(n,t){return this.throwIfDisposed(),gv(this,n,t)};Y.prototype.transpose=function(n){return this.throwIfDisposed(),ut(this,n)};Y.prototype.unique=function(n){return this.throwIfDisposed(),vv(this,n)};Y.prototype.unsortedSegmentSum=function(n,t){return this.throwIfDisposed(),Th(this,n,t)};Y.prototype.unstack=function(n){return this.throwIfDisposed(),go(this,n)};Y.prototype.where=function(n,t){return this.throwIfDisposed(),vn(n,this,t)};Y.prototype.zerosLike=function(){return this.throwIfDisposed(),Ee(this)};I.Abs=Fl;I.Acos=Wl;I.Acosh=Ul;I.AdadeltaOptimizer=Rh;I.AdagradOptimizer=Oh;I.AdamOptimizer=Eh;I.AdamaxOptimizer=Dh;I.Add=Fs;I.AddN=Bl;I.All=wf;I.Any=Sf;I.ArgMax=zl;I.ArgMin=Pl;I.Asin=_l;I.Asinh=Ml;I.Atan=Hl;I.Atan2=ql;I.Atanh=Vl;I.AvgPool=Gl;I.AvgPool3D=Yl;I.AvgPool3DBackprop=If;I.AvgPoolBackprop=Lf;I.BatchMatMul=jl;I.BatchToSpaceND=Kl;I.BroadcastTo=$l;I.Cast=Ws;I.Ceil=Xl;I.ClipByValue=Jl;I.Complex=Af;I.Concat=Zl;I.Conv2D=Ql;I.Conv2DBackpropFilter=Tf;I.Conv2DBackpropInput=eu;I.Conv3D=tu;I.Conv3DBackpropFilterV2=Nf;I.Conv3DBackpropInputV2=xf;I.Cos=nu;I.Cosh=iu;I.CropAndResize=Cf;I.Cumsum=ru;I.DataStorage=x2;I.DepthToSpace=Rf;I.DepthwiseConv2dNative=au;I.DepthwiseConv2dNativeBackpropFilter=Of;I.DepthwiseConv2dNativeBackpropInput=Ef;I.Diag=Df;I.Dilation2D=su;I.Dilation2DBackpropFilter=Ff;I.Dilation2DBackpropInput=kf;I.Div=ou;I.Elu=lu;I.EluGrad=Wf;I.Environment=vf;I.Equal=Uf;I.Erf=uu;I.Exp=cu;I.Expm1=hu;I.FFT=Bf;I.Fill=zf;I.FlipLeftRight=Pf;I.Floor=du;I.FloorDiv=pu;I.FromPixels=vc;I.FusedBatchNorm=fu;I.FusedConv2D=bc;I.FusedDepthwiseConv2D=wc;I.GatherNd=_f;I.GatherV2=mu;I.Greater=Mf;I.GreaterEqual=gu;I.IFFT=Hf;I.Identity=vu;I.Imag=Vf;I.IsFinite=yu;I.IsInf=bu;I.IsNan=wu;I.KernelBackend=mf;I.LRN=Au;I.LRNBackprop=Xf;I.Less=qf;I.LessEqual=Gf;I.LinSpace=Yf;I.Log=Su;I.Log1p=Lu;I.LogSoftmax=Iu;I.LogicalAnd=jf;I.LogicalNot=Kf;I.LogicalOr=$f;I.Max=Tu;I.MaxPool=xu;I.MaxPool3D=Cu;I.MaxPool3DBackprop=Zf;I.MaxPoolBackprop=Jf;I.MaxPoolWithArgmax=Qf;I.Maximum=Nu;I.Mean=k2;I.Min=Ru;I.Minimum=Ou;I.Mod=Eu;I.MomentumOptimizer=kh;I.Multiply=Du;I.Negate=ku;I.NonMaxSuppressionV3=tm;I.NonMaxSuppressionV4=nm;I.NonMaxSuppressionV5=im;I.NotEqual=em;I.OP_SCOPE_SUFFIX=Pm;I.OneHot=Wu;I.OnesLike=Fu;I.Optimizer=Ii;I.PadV2=Uu;I.Pool=F2;I.Pow=Bu;I.Prelu=zu;I.Prod=rm;I.RMSPropOptimizer=Fh;I.Range=am;I.Real=sm;I.Reciprocal=Pu;I.Relu=_u;I.Relu6=qu;I.Reshape=Mu;I.ResizeBilinear=Vu;I.ResizeBilinearGrad=lm;I.ResizeNearestNeighbor=Hu;I.ResizeNearestNeighborGrad=om;I.Reverse=Gu;I.RotateWithOffset=mm;I.Round=Yu;I.Rsqrt=ju;I.SGDOptimizer=Lo;I.ScatterNd=um;I.SelectV2=Ku;I.Selu=$u;I.Sigmoid=ec;I.Sign=Qu;I.Sin=Ju;I.Sinh=Zu;I.Slice=Xu;I.Softmax=sc;I.Softplus=tc;I.SpaceToBatchND=rc;I.SparseToDense=hm;I.SplitV=ac;I.Sqrt=nc;I.Square=cm;I.SquaredDifference=oc;I.Step=gc;I.StridedSlice=dm;I.Sub=lc;I.Sum=ic;I.Tan=uc;I.Tanh=cc;I.Tensor=Y;I.TensorBuffer=Hs;I.Tile=hc;I.TopK=pm;I.Transpose=dc;I.Unique=fm;I.Unpack=pc;I.UnsortedSegmentSum=fc;I.Variable=Ba;I.ZerosLike=mc;I._FusedMatMul=yc;I.abs=jt;I.acos=Sg;I.acosh=Lg;I.add=fe;I.addN=d1;I.addStrict=Nv;I.all=Tg;I.any=Ng;I.argMax=xg;I.argMin=Cg;I.asin=Rg;I.asinh=Og;I.atan=Eg;I.atan2=Dg;I.atanh=kg;I.avgPool=nh;I.avgPool3d=O1;I.backend=n1;I.backend_util=BE;I.basicLSTMCell=U1;I.batchNorm=Va;I.batchNorm2d=M1;I.batchNorm3d=V1;I.batchNorm4d=G1;I.batchToSpaceND=Js;I.booleanMaskAsync=fR;I.broadcastTo=Qs;I.browser=NT;I.buffer=Wn;I.cast=ue;I.ceil=Ug;I.clipByValue=Bg;I.clone=$i;I.complex=mi;I.concat=Rt;I.concat1d=X1;I.concat2d=Z1;I.concat3d=eN;I.concat4d=nN;I.conv1d=zg;I.conv2d=Gr;I.conv2dTranspose=Pg;I.conv3d=lN;I.conv3dTranspose=hN;I.copyRegisteredKernels=B2;I.cos=eo;I.cosh=rh;I.cosineWindow=Nh;I.cumsum=ah;I.customGrad=zn;I.deprecationWarn=At;I.depthToSpace=Mg;I.depthwiseConv2d=qa;I.device_util=bA;I.diag=yN;I.dilation2d=Hg;I.disableDeprecationWarnings=VT;I.dispose=Mt;I.disposeVariables=qT;I.div=Ie;I.divNoNan=Vg;I.divStrict=xv;I.dot=qg;I.dropout=HR;I.elu=sh;I.enableDebugMode=HT;I.enableProdMode=MT;I.enclosingPowerOfTwo=Uv;I.engine=GT;I.env=He;I.equal=Ga;I.equalStrict=wv;I.erf=Gg;I.exp=yn;I.expandDims=bn;I.expm1=Yg;I.eye=jg;I.fft=ho;I.fill=oh;I.findBackend=QT;I.findBackendFactory=e1;I.floor=to;I.floorDiv=Zc;I.fused=eO;I.gather=no;I.gatherND=PR;I.gather_util=xT;I.getBackend=JT;I.getGradient=Ic;I.getKernel=Lc;I.getKernelsForBackend=Us;I.grad=QN;I.grads=ex;I.greater=Li;I.greaterEqual=Zi;I.greaterEqualStrict=Sv;I.greaterStrict=Lv;I.ifft=$a;I.imag=io;I.image=sE;I.inTopKAsync=qR;I.io=mT;I.irfft=Ah;I.isFinite=$g;I.isInf=Xg;I.isNaN=Jg;I.keep=wg;I.kernel_impls=ME;I.leakyRelu=Zg;I.less=ro;I.lessEqual=Qi;I.lessEqualStrict=Iv;I.lessStrict=Av;I.linalg=oE;I.linspace=$N;I.localResponseNormalization=Qg;I.log=er;I.log1p=uh;I.logSigmoid=tv;I.logSoftmax=nv;I.logSumExp=hh;I.logicalAnd=nr;I.logicalNot=so;I.logicalOr=dh;I.logicalXor=iv;I.losses=lE;I.matMul=We;I.math=LT;I.max=tr;I.maxPool=ph;I.maxPool3d=vx;I.maxPoolWithArgmax=bx;I.maximum=jr;I.maximumStrict=Cv;I.mean=Ya;I.memory=YT;I.min=oo;I.minimum=lo;I.minimumStrict=Rv;I.mod=fh;I.modStrict=Ov;I.moments=Nx;I.movingAverage=DR;I.mul=J;I.mulStrict=Ev;I.multiRNNCell=Cx;I.multinomial=Ox;I.neg=vt;I.nextFrame=hE;I.norm=vo;I.notEqual=uo;I.notEqualStrict=Tv;I.oneHot=Ys;I.ones=Kr;I.onesLike=mh;I.op=U;I.outerProduct=Wx;I.pad=ir;I.pad1d=zx;I.pad2d=_x;I.pad3d=Hx;I.pad4d=qx;I.pool=rv;I.pow=ni;I.powStrict=Dv;I.prelu=gh;I.print=$m;I.prod=av;I.profile=jT;I.rand=Qx;I.randomGamma=cC;I.randomNormal=dC;I.randomUniform=sv;I.range=bh;I.ready=XT;I.real=ja;I.reciprocal=ov;I.registerBackend=t1;I.registerGradient=vm;I.registerKernel=gm;I.relu=Ka;I.relu6=wh;I.removeBackend=ZT;I.reshape=V;I.reverse=ii;I.reverse1d=bC;I.reverse2d=SC;I.reverse3d=IC;I.reverse4d=TC;I.rfft=po;I.round=lv;I.rsqrt=Sh;I.scalar=be;I.scatterND=FR;I.scatter_util=CT;I.selu=uv;I.separableConv2d=cv;I.serialization=ET;I.setBackend=$T;I.setPlatform=i1;I.setdiff1dAsync=EC;I.sigmoid=Ji;I.sign=hv;I.signal=aE;I.sin=Lh;I.sinh=Ih;I.slice=ze;I.slice1d=UC;I.slice2d=zC;I.slice3d=_C;I.slice4d=HC;I.slice_util=gg;I.softmax=dv;I.softplus=ch;I.spaceToBatchND=co;I.sparseToDense=BR;I.spectral=rE;I.split=Jr;I.sqrt=Vt;I.square=Ye;I.squaredDifference=fo;I.squaredDifferenceStrict=kv;I.squeeze=mo;I.stack=sr;I.step=Zr;I.stridedSlice=fv;I.sub=ge;I.subStrict=Fv;I.sum=Ae;I.sumOutType=cA;I.tan=mv;I.tanh=Xs;I.tensor=vi;I.tensor1d=Xr;I.tensor2d=Xa;I.tensor3d=ng;I.tensor4d=nR;I.tensor5d=iR;I.tensor6d=rR;I.tensor_util=pA;I.test_util=PT;I.tidy=mt;I.tile=Yr;I.time=KT;I.topk=gv;I.train=uE;I.transpose=ut;I.truncatedNormal=oR;I.unique=vv;I.unregisterGradient=U2;I.unregisterKernel=W2;I.unsortedSegmentSum=Th;I.unstack=go;I.upcastType=Vs;I.util=Q2;I.valueAndGrad=tx;I.valueAndGrads=nx;I.variable=hR;I.variableGrads=ev;I.version_core=_T;I.where=vn;I.whereAsync=bv;I.zeros=ti;I.zerosLike=Ee});var Ow=Le(Xe=>{"use strict";Object.defineProperty(Xe,"__esModule",{value:!0});var y=lr();var Uh=function(n,t){return Uh=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Uh(n,t)};function Q(n,t){Uh(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var Kt=function(){return Kt=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t},Kt.apply(this,arguments)};function Pk(n,t){var e={};for(var i in n)Object.prototype.hasOwnProperty.call(n,i)&&t.indexOf(i)<0&&(e[i]=n[i]);if(n!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,i=Object.getOwnPropertySymbols(n);r<i.length;r++)t.indexOf(i[r])<0&&(e[i[r]]=n[i[r]]);return e}function Se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function ve(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Bh;function yt(){return Bh==null&&(Bh=y.backend().epsilon()),Bh}function wn(){return"channelsLast"}var Ai=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Pn=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),M=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Te=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),_k=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),dV=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error);function ur(n,t){if(Array.isArray(n)){for(var e=[],i=0;i<t;i++)e=e.concat(n);return e}else{var e=new Array(t);return e.fill(n),e}}function _n(n,t){if(!n)throw new _k(t)}function ry(n,t){for(var e=0,i=0,r=n;i<r.length;i++){var a=r[i];a===t&&e++}return e}function qt(n){return n.length===1?n[0]:n}function Je(n){return Array.isArray(n)?n:[n]}function ai(n){var t=n.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2"),e=t.replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function cr(n){return n.length<=1||n.indexOf("_")===-1?n:n.replace(/[_]+(\w|$)/g,function(t,e){return e.toUpperCase()})}var dn={};function zh(n){if(n==null)return null;var t={};return t.className=n.getClassName(),t.config=n.getConfig(),t}function Ph(n){if(n==null||typeof n!="object")return;if(Array.isArray(n))n.forEach(function(s){return Ph(s)});else for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e],a=n[r];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?n[r]=a.value:Ph(a))}}function Ja(n,t,e,i,r){var a,s,o;if(t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i="object"),r===void 0&&(r=!1),typeof n=="string"){var l=n,u=void 0;if(l in e)u=e[l];else if(l in dn)u=dn[l];else if(u=t[l],u==null)throw new M("Unknown "+i+": "+n+`. This may be due to one of the following reasons:
 `+("1. The "+i+" is defined in Python, in which ")+`case it needs to be ported to TensorFlow.js or your JavaScript code.
 `+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");return u}else{var c=n;if(c.className==null||c.config==null)throw new M(i+": Improper config format: "+(JSON.stringify(c)+`.
-`)+"'className' and 'config' must set.");var h=c.className,d=void 0,p=void 0;if(h in e?(a=e[h],d=a[0],p=a[1]):h in pn?(s=pn.className,d=s[0],p=s[1]):h in t&&(o=t[h],d=o[0],p=o[1]),d==null)throw new M("Unknown "+i+": "+h+`. This may be due to one of the following reasons:
+`)+"'className' and 'config' must set.");var h=c.className,d=void 0,p=void 0;if(h in e?(a=e[h],d=a[0],p=a[1]):h in dn?(s=dn.className,d=s[0],p=s[1]):h in t&&(o=t[h],d=o[0],p=o[1]),d==null)throw new M("Unknown "+i+": "+h+`. This may be due to one of the following reasons:
 `+("1. The "+i+" is defined in Python, in which ")+`case it needs to be ported to TensorFlow.js or your JavaScript code.
-`+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");if(p!=null){for(var f={},m=0,g=Object.keys(pn);m<g.length;m++){var v=g[m];f[v]=pn[v]}for(var b=0,w=Object.keys(e);b<w.length;b++){var v=w[b];f[v]=e[v]}var S=c.config;S.customObjects=f;for(var L=$t({},pn),N=0,C=Object.keys(e);N<C.length;N++){var v=C[N];pn[v]=e[v]}zh(c.config);var R=p(d,c.config,e,r);return pn=$t({},L),R}else{for(var L=$t({},pn),D=0,k=Object.keys(e);D<k.length;D++){var v=k[D];pn[v]=e[v]}var R=new d(c.config);return pn=$t({},L),R}}}function Mk(n,t){return n<t?-1:n>t?1:0}function Io(n,t){return-1*Mk(n,t)}function Ti(n){if(n==null)return n;for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.indexOf(r)===-1&&t.push(r)}return t}function Hk(n){if(n==null)throw new M("Invalid value in obj: "+JSON.stringify(n));for(var t in n)if(n.hasOwnProperty(t))return!1;return!0}function ea(n,t,e){if(e==null)return;if(n.indexOf(e)<0)throw new M(e+" is not a valid "+t+".  Valid values are "+n+" or null/undefined.")}function Ph(n,t,e,i){return e===void 0&&(e=0),i===void 0&&(i=Infinity),_n(e>=0),_n(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function Tt(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return Tt(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(ay(n)+".")})}function ay(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return ay(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function Vk(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];var o=y.util.now();return o-e<t||(e=o,i=n.apply(void 0,a)),i};return r}function sy(n){return n==="relu"?"relu":n==="linear"?"linear":n==="elu"?"elu":null}function _h(n,t){return y.tidy(function(){return y.sqrt(y.sum(y.mul(n,n),t,!0))})}var Za=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),oy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMaxValue=2,i.defaultAxis=0,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=_h(e,i.axis),a=y.clipByValue(r,0,i.maxValue);return y.mul(e,y.div(a,y.add(yt(),r)))})},t.prototype.getConfig=function(){return{maxValue:this.maxValue,axis:this.axis}},t.className="MaxNorm",t}(Za);y.serialization.registerClass(oy);var ly=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultAxis=0,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){return y.div(e,y.add(yt(),_h(e,i.axis)))})},t.prototype.getConfig=function(){return{axis:this.axis}},t.className="UnitNorm",t}(Za);y.serialization.registerClass(ly);var uy=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="NonNeg",t}(Za);y.serialization.registerClass(uy);var cy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMinValue=0,i.defaultMaxValue=1,i.defaultRate=1,i.defaultAxis=0,i.minValue=e.minValue!=null?e.minValue:i.defaultMinValue,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.rate=e.rate!=null?e.rate:i.defaultRate,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=_h(e,i.axis),a=y.add(y.mul(i.rate,y.clipByValue(r,i.minValue,i.maxValue)),y.mul(1-i.rate,r));return y.mul(e,y.div(a,y.add(yt(),r)))})},t.prototype.getConfig=function(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}},t.className="MinMaxNorm",t}(Za);y.serialization.registerClass(cy);var hy={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function bt(n){return Bh(n)}function dy(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"constraint")}function wt(n){if(n==null)return null;if(typeof n=="string"){var t=n in hy?hy[n]:n,e={className:t,config:{}};return dy(e)}else return n instanceof Za?n:dy(n)}function qk(n){return new oy(n)}function Gk(n){return new ly(n)}function Yk(){return new uy}function jk(n){return new cy(n)}var Kk={__proto__:null,maxNorm:qk,unitNorm:Gk,nonNeg:Yk,minMaxNorm:jk};var $k=["channelsFirst","channelsLast"],Xk=["valid","same","causal"],Jk=["max","avg"],Zk=["sum","mul","concat","ave"];var ta=new Map;function ht(n){ea($k,"DataFormat",n)}function sn(n){ea(Xk,"PaddingMode",n)}function py(n){ea(Jk,"PoolMode",n)}var Qa=[],fy="/";function hr(n,t){Qa.push(n);try{var e=t();return Qa.pop(),e}catch(i){throw Qa.pop(),i}}function Qk(){return Qa.length===0?"":Qa.join(fy)+fy}function gy(n){if(!my(n))throw new Error("Not a valid tensor name: '"+n+"'");return Qk()+n}function vy(n){if(!my(n))throw new Error("Not a valid tensor name: '"+n+"'");ta.has(n)||ta.set(n,0);var t=ta.get(n);if(ta.set(n,ta.get(n)+1),t>0){var e=n+"_"+t;return ta.set(e,1),e}else return n}var e3=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function my(n){return!!n.match(e3)}function t3(n){return n===parseInt(n.toString(),10)}function Ni(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r<e;++r)i*=n[r];return i}function yy(n){return n=Array.isArray(n)?new Float32Array(n):n,y.tensor1d(n)}function na(n){return y.min(yy(n)).dataSync()[0]}function xi(n){return y.max(yy(n)).dataSync()[0]}function Ln(n,t){if(t<n)throw new M("end ("+t+") < begin ("+n+") is forbidden.");for(var e=[],i=n;i<t;++i)e.push(i);return e}function es(n,t){return n.asType(t)}function ts(n,t){t===void 0&&(t=-1);var e=n.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),n.reshape(e)}function n3(n,t){return y.tidy(function(){if(n.shape.length!==2)throw new M("repeat() expects a rank-2 tensor, but received a "+("rank-"+n.shape.length+" tensor."));var e=ts(n,1);return Mh(e,[1,t,1])})}function i3(n){var t=[Ni(n.shape)];return n.reshape(t)}function r3(n){if(n.rank<=1)throw new M("batchFlatten requires a minimum rank of 2. Got rank: "+n.rank+".");var t=[n.shape[0],Ni(n.shape,1)];return n.reshape(t)}function dr(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[t,0],[e,n.shape[1]]);case 3:return y.slice3d(n,[t,0,0],[e,n.shape[1],n.shape[2]]);case 4:return y.slice4d(n,[t,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3]]);case 5:return y.slice(n,[t,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4]]);case 6:return y.slice(n,[t,0,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4],n.shape[5]]);default:throw new M("sliceAlongFirstAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function Hh(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[0,t],[n.shape[0],e]);case 3:return y.slice3d(n,[0,0,t],[n.shape[0],n.shape[1],e]);case 4:return y.slice4d(n,[0,0,0,t],[n.shape[0],n.shape[1],n.shape[2],e]);default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function Ao(n,t,e,i){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:switch(i){case 1:return dr(n,t,e);case 2:return Hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 3:switch(i){case 1:return dr(n,t,e);case 2:return y.slice3d(n,[0,t,0],[n.shape[0],e,n.shape[2]]);case 3:return Hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 4:switch(i){case 1:return dr(n,t,e);case 2:return y.slice4d(n,[0,t,0,0],[n.shape[0],e,n.shape[2],n.shape[3]]);case 3:return y.slice4d(n,[0,0,t,0],[n.shape[0],n.shape[1],e,n.shape[3]]);case 4:return Hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function Vh(n,t){t===void 0&&(t=-1);var e;return t<0&&(e=n[0].rank,e!==0?t=e:t=0),t===n[0].rank&&(t=-1),y.concat(n,t)}function by(n,t){switch(n.rank){case 1:return y.concat1d([n,t]);case 2:return y.concat2d([n,t],0);case 3:return y.concat3d([n,t],0);case 4:return y.concat4d([n,t],0);default:throw new M("concatAlongFirstAxis() received an unsupported "+("tensor rank: "+n.rank))}}function Mh(n,t){if(Array.isArray(t)||(t=[t]),n.rank!==t.length)throw new M("The length of input n ("+t.length+") does not match "+("the number of dimensions in input x ("+n.rank+")"));return y.tile(n,t)}function To(n,t,e,i,r){return t===void 0&&(t=0),e===void 0&&(e=1),y.randomNormal(n,t,e,i,r)}function Mn(n,t,e,i){if(n.rank<2||t.rank<2)throw new Te("dot requires both inputs to be rank >= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?qh(n.rank,i,Sn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?qh(n.rank,i,Sn()):null,activation:e}).reshape(f)}}function wy(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function ns(n){return y.mul(n,n)}function qh(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Hn(n,t,e){return y.tidy(function(){return e==null&&(e=Sn()),ht(e),n.add(qh(n.rank,t,e))})}function a3(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function s3(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function Sy(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function o3(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function is(n,t,e){return e===void 0&&(e=!1),e?n():t()}var l3=["fanIn","fanOut","fanAvg"],u3=["normal","uniform","truncatedNormal"];function c3(n){ea(l3,"FanMode",n)}function h3(n){ea(u3,"Distribution",n)}var fn=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),Ly=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(fn);y.serialization.registerClass(Ly);var Gh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(fn);y.serialization.registerClass(Gh);var Iy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(fn);y.serialization.registerClass(Iy);var Ay=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(fn);y.serialization.registerClass(Ay);var Ty=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return To(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(fn);y.serialization.registerClass(Ty);var Ny=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(fn);y.serialization.registerClass(Ny);var xy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(fn);y.serialization.registerClass(xy);function d3(n,t){t===void 0&&(t="channelsLast");var e,i;if(ht(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=Ni(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=Ni(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=Ni(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var Xt=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,c3(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,h3(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=d3(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(fn);y.serialization.registerClass(Xt);var Yh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Xt.className},t.className="GlorotUniform",t}(Xt);y.serialization.registerClass(Yh);var jh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Xt.className},t.className="GlorotNormal",t}(Xt);y.serialization.registerClass(jh);var Kh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Xt.className},t.className="HeNormal",t}(Xt);y.serialization.registerClass(Kh);var $h=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Xt.className},t.className="HeUniform",t}(Xt);y.serialization.registerClass($h);var Xh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Xt.className},t.className="LeCunNormal",t}(Xt);y.serialization.registerClass(Xh);var Jh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Xt.className},t.className="LeCunNormal",t}(Xt);y.serialization.registerClass(Jh);var Cy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=To(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(fn);y.serialization.registerClass(Cy);var Ry={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function Oy(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function st(n){return Bh(n)}function tt(n){if(typeof n=="string"){var t=n in Ry?Ry[n]:n;if(t==="GlorotNormal")return new jh;if(t==="GlorotUniform")return new Yh;if(t==="HeNormal")return new Kh;if(t==="HeUniform")return new $h;if(t==="LeCunNormal")return new Xh;if(t==="LeCunUniform")return new Jh;var e={};return e.className=t,e.config={},Oy(e)}else return n instanceof fn?n:Oy(n)}function p3(){return new Ly}function f3(){return new Gh}function m3(n){return new Iy(n)}function g3(n){return new Ay(n)}function v3(n){return new Ty(n)}function y3(n){return new Ny(n)}function b3(n){return new xy(n)}function w3(n){return new Xt(n)}function S3(n){return new Yh(n)}function L3(n){return new jh(n)}function I3(n){return new Kh(n)}function A3(n){return new $h(n)}function T3(n){return new Xh(n)}function N3(n){return new Jh(n)}function x3(n){return new Cy(n)}var C3={__proto__:null,zeros:p3,ones:f3,constant:m3,randomUniform:g3,randomNormal:v3,truncatedNormal:y3,identity:b3,varianceScaling:w3,glorotUniform:S3,glorotNormal:L3,heNormal:I3,heUniform:A3,leCunNormal:T3,leCunUniform:N3,orthogonal:x3};var R3=0;function Ey(){return R3++}var No={};function xo(n){return n===void 0&&(n=""),n in No||(No[n]=0),No[n]+=1,n+No[n].toString()}function Zh(n){return Array.isArray(n)&&Array.isArray(n[0])}function Co(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function je(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function Ro(n){for(var t=0,e=0,i=n;e<i.length;e++){var r=i[e];r.shape.length===0?t+=1:t+=r.shape.reduce(function(a,s){return a*s})}return t}var Dy="Variable",ky=function(){function n(t,e,i,r,a){e===void 0&&(e="float32"),i===void 0&&(i=Dy),r===void 0&&(r=!0),a===void 0&&(a=null),this.dtype=e??"float32",this.shape=t.shape,this.id=Ey(),i=i??Dy,this.originalName=gy(i),this.name=vy(this.originalName),this.trainable_=r,this.constraint=a,this.val=y.variable(t,this.trainable_,this.name,this.dtype)}return n.prototype.read=function(){return this.assertNotDisposed(),this.val},n.prototype.write=function(t){return this.assertNotDisposed(),O3(this.val,t),this.val.id!==t.id&&(this.val.assign(t),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this},n.prototype.dispose=function(){this.assertNotDisposed(),this.val.dispose()},n.prototype.assertNotDisposed=function(){if(this.val.isDisposed)throw new Error("LayersVariable "+this.name+" is already disposed.")},Object.defineProperty(n.prototype,"trainable",{get:function(){return this.trainable_},set:function(t){this.trainable_=t,this.val.trainable=t},enumerable:!0,configurable:!0}),n}();function O3(n,t){if(n.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(n.shape)+" vs. "+JSON.stringify(t.shape))}function Qh(n){return n.map(function(t){return t.read()})}function ed(n){n.forEach(function(t){var e=t[0];e.write(t[1])})}var Nt=function(){function n(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}return n}(),In=function(){function n(t,e,i,r,a,s,o){this.dtype=t,this.shape=e,this.sourceLayer=i,this.inputs=r,this.callArgs=a,this.outputTensorIndex=o,this.id=Ey(),s!=null&&(this.originalName=gy(s),this.name=vy(this.originalName)),this.rank=e.length}return n}(),E3=0,Oo=function(){function n(t,e){this.callArgs=e,this.id=E3++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(var i=0,r=t.inboundLayers;i<r.length;i++){var a=r[i];a!=null&&a.outboundNodes.push(this)}t.outboundLayer.inboundNodes.push(this)}return n.prototype.getConfig=function(){for(var t=[],e=0,i=this.inboundLayers;e<i.length;e++){var r=i[e];r!=null?t.push(r.name):t.push(null)}return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}},n}(),D3=0,De=function(n){Q(t,n);function t(e){e===void 0&&(e={});var i=n.call(this)||this;i._callHook=null,i._addedWeightNames=[],i._stateful=!1,i.id=D3++,i.activityRegularizer=null,i.inputSpec=null,i.supportsMasking=!1,i._trainableWeights=[],i._nonTrainableWeights=[],i._losses=[],i._updates=[],i._built=!1,i.inboundNodes=[],i.outboundNodes=[];var r=e.name;if(!r){var a=i.getClassName();r=ai(a)+"_"+xo(a)}if(i.name=r,i.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){var s=void 0;if(e.batchInputShape!=null)s=e.batchInputShape;else if(e.inputShape!=null){var o=null;e.batchSize!=null&&(o=e.batchSize),s=[o].concat(e.inputShape)}i.batchInputShape=s;var l=e.dtype;l==null&&(l=e.inputDType),l==null&&(l="float32"),i.dtype=l}return e.weights!=null?i.initialWeights=e.weights:i.initialWeights=null,i._refCount=null,i.fastWeightInitDuringBuild=!1,i}return t.nodeKey=function(e,i){return e.name+"_ib-"+i.toString()},t.prototype.getNodeAtIndex=function(e,i){if(this.inboundNodes.length===0)throw new Pn("The layer has never been called "+("and thus has no defined "+i+"."));if(this.inboundNodes.length<=e)throw new M("Asked to get "+i+" at node "+e+", "+("but the layer has only "+this.inboundNodes.length+" inbound nodes."));return this.inboundNodes[e]},t.prototype.getInputAt=function(e){return qt(this.getNodeAtIndex(e,"input").inputTensors)},t.prototype.getOutputAt=function(e){return qt(this.getNodeAtIndex(e,"output").outputTensors)},Object.defineProperty(t.prototype,"input",{get:function(){if(this.inboundNodes.length>1)throw new Ai("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new Ai("Layer "+this.name+" is not connected, no input to return.");return qt(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new Ai("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new Ai("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return qt(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Je(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Je(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r<e.length;r++){var a=e[r],s=i[r];if(s==null)continue;var o=a.rank;if(s.ndim!=null&&o!==s.ndim)throw new M("Input "+r+" is incompatible with layer "+this.name+": "+("expected ndim="+s.ndim+", found ndim="+o));if(s.maxNDim!=null&&o>s.maxNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected max_ndim="+s.maxNDim+", found ndim="+o));if(s.minNDim!=null&&o<s.minNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected min_ndim="+s.minNDim+", found ndim="+o+"."));if(s.dtype!=null&&a.dtype!==s.dtype)throw new M("Input "+r+" is incompatible with layer "+this.name+" "+(": expected dtype="+s.dtype+", found dtype="+a.dtype+"."));if(s.axes){var l=a.shape;for(var u in s.axes){var c=Number(u),h=s.axes[u],d=c>=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p<s.shape.length;++p){var f=s.shape[p],m=a.shape[p];if(f!=null&&m!=null&&f!==m)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected shape="+s.shape+", ")+("found shape="+a.shape+"."))}}},t.prototype.call=function(e,i){return e},t.prototype.invokeCallHook=function(e,i){this._callHook!=null&&this._callHook(e,i)},t.prototype.setCallHook=function(e){this._callHook=e},t.prototype.clearCallHook=function(){this._callHook=null},t.prototype.apply=function(e,i){var r=this;i=i||{},this.assertNotDisposed();for(var a=Je(e),s=!0,o=0,l=a;o<l.length;o++){var u=l[o];if(!(u instanceof In)){s=!1;break}}for(var c=!0,h=0,d=a;h<d.length;h++){var u=d[h];if(u instanceof In){c=!1;break}}if(s===c)throw new M("Arguments to apply() must be all SymbolicTensors or all Tensors");return hr(this.name,function(){if(!r.built){r.assertInputCompatibility(e);for(var p=[],f=0,m=Je(e);f<m.length;f++){var g=m[f];p.push(g.shape)}r.build(qt(p)),r.built=!0,r.initialWeights&&r.setWeights(r.initialWeights),r._refCount===null&&c&&(r._refCount=1)}if(r.assertInputCompatibility(e),c){for(var v=r.call(e,i),b=Je(v),w=[],S=0,L=b;S<L.length;S++){var N=L[S];a.indexOf(N)!==-1&&(N=N.clone()),w.push(N)}if(v=qt(w),r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}else{var C=k3(e),R=r.computeOutputShape(C),v=void 0,D=F3(e);if(r.warnOnIncompatibleInputShape(Array.isArray(e)?C[0]:C),R!=null&&R.length>0&&Array.isArray(R[0])?v=R.map(function(W,F){return new In(D,W,r,Je(e),i,r.name,F)}):v=new In(D,R,r,Je(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new Ai("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i<r.length;i++){var a=r[i],s=JSON.stringify(a.outputShapes);e.indexOf(s)===-1&&e.push(s)}if(e.length===1){var o=this.inboundNodes[0].outputShapes;return Array.isArray(o)&&Array.isArray(o[0])&&o.length===1?o[0]:o}else throw new Ai("The layer "+this.name+' has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.')},enumerable:!0,configurable:!0}),t.prototype.countParams=function(){if(!this.built)throw new Pn("You tried to call countParams() on "+this.name+", but the layer is not built yet. Build it first by calling build(batchInputShape).");return Ro(this.weights)},t.prototype.build=function(e){this.built=!0},t.prototype.getWeights=function(e){return e===void 0&&(e=!1),Qh(e?this.trainableWeights:this.weights)},t.prototype.setWeights=function(e){var i=this;y.tidy(function(){var r=i.weights;if(r.length!==e.length)throw new M('You called setWeights(weights) on layer "'+i.name+'" '+("with a weight list of length "+e.length+", ")+("but the layer was expecting "+r.length+" weights. ")+("Provided weights: "+e+"..."));if(r.length===0)return;for(var a=[],s=Qh(r),o=0;o<s.length;++o){var l=s[o],u=r[o],c=e[o];if(!y.util.arraysEqual(l.shape,c.shape))throw new M("Layer weight shape "+l.shape+" "+("not compatible with provided weight shape "+c.shape));a.push([u,c])}ed(a)})},t.prototype.addWeight=function(e,i,r,a,s,o,l){if(this._addedWeightNames.indexOf(e)!==-1)throw new M("Duplicate weight name "+e+" for layer "+this.name);this._addedWeightNames.push(e),r==null&&(r="float32"),this.fastWeightInitDuringBuild&&(a=tt("zeros"));var u=a.apply(i,r),c=new ky(u,r,e,o,l);return u.dispose(),s!=null&&this.addLoss(function(){return s.apply(c.read())}),o==null&&(o=!0),o?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c},t.prototype.setFastWeightInitDuringBuild=function(e){this.fastWeightInitDuringBuild=e},t.prototype.addLoss=function(e){var i;if(e==null||Array.isArray(e)&&e.length===0)return;e=Je(e),this._losses!==void 0&&this._losses!==null&&(i=this.losses).push.apply(i,e)},t.prototype.computeOutputShape=function(e){return e},t.prototype.computeMask=function(e,i){var r=this;if(!this.supportsMasking){if(i!=null)if(Array.isArray(i))i.forEach(function(a){if(a!=null)throw new TypeError("Layer "+r.name+" does not support masking, but was passed an inputMask.")});else throw new TypeError("Layer "+this.name+" does not support masking, but was passed an inputMask.");return null}return i},t.prototype.addInboundNode=function(e,i,r,a,s,o,l){l===void 0&&(l=null);var u=Je(e);i=Je(i),r=Je(r),a=Je(a),s=Co(s),o=Co(o);for(var c=[],h=[],d=[],p=0,f=u;p<f.length;p++){var m=f[p];c.push(m.sourceLayer),h.push(m.nodeIndex),d.push(m.tensorIndex)}new Oo({outboundLayer:this,inboundLayers:c,nodeIndices:h,tensorIndices:d,inputTensors:u,outputTensors:i,inputMasks:r,outputMasks:a,inputShapes:s,outputShapes:o},l);for(var g=0;g<i.length;g++)i[g].sourceLayer=this,i[g].nodeIndex=this.inboundNodes.length-1,i[g].tensorIndex=g},t.prototype.getConfig=function(){var e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e},t.prototype.disposeWeights=function(){return this.weights.forEach(function(e){return e.dispose()}),this.weights.length},t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Layer '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){if(!this.built)throw new Error("Cannot dispose Layer "+this.name+" because it has not been built yet.");if(this._refCount===null)throw new Error("Cannot dispose Layer "+this.name+" because it has not been used yet.");this.assertNotDisposed();var e=0;return--this._refCount===0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}},t}(y.serialization.Serializable);function k3(n){n=Je(n);for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.push(r.shape)}return qt(t)}function F3(n){return"float32"}function Fy(n,t,e){if((t==null||e!=null&&e>0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a<i.inboundLayers.length;a++)for(var s=i.inputTensors[a],o=i.inboundLayers[a],l=i.nodeIndices[a],u=Fy(s,o,l),c=0,h=u;c<h.length;c++){var d=h[c];r.indexOf(d)===-1&&r.push(d)}return r}var rs=function(n){Q(t,n);function t(e){var i=n.call(this,{dtype:e.dtype,name:e.name!=null?e.name:xo("input").toString()})||this;if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),i.trainable=!1,i.built=!0,i.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new M("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");var r=e.batchInputShape;if(r==null){if(e.inputShape==null)throw new M("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");r=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new M("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");var a=e.dtype||"float32";i.batchInputShape=r,i.dtype=a,i.inputSpec=[{shape:r}];var s=new In(i.dtype,i.batchInputShape,i,[],{},i.name);return s.nodeIndex=0,s.tensorIndex=0,new Oo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[r],outputShapes:[r]}),i}return t.prototype.apply=function(e,i){throw new M("Cannot pass any input to an "+("InputLayer's apply() method. InputLayer name: "+this.name))},t.prototype.dispose=function(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}},t.prototype.getConfig=function(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}},t.className="InputLayer",t}(De);y.serialization.registerClass(rs);function Wy(n){if(n.batchShape==null&&n.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(n.batchShape!=null&&n.shape!=null)throw new M("Please provide either a `shape` or `batchShape` argument to Input, but not both.");var t=n.batchShape;n.shape!=null&&t==null&&(t=[null].concat(n.shape));var e=n.dtype;e==null&&(e="float32");var i=new rs({batchInputShape:t,name:n.name,dtype:e,sparse:n.sparse}),r=i.inboundNodes[0].outputTensors;return r[0]}function Ci(n){return Se(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(n==null)return[2];t=[],e=[],i=[];for(r in n)a=n[r],typeof a!="number"&&(s=a,t.push(s.data()),e.push(r),i.push(s));return t.length>0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l<o.length;++l)n[e[l]]=o[l][0];y.dispose(i),u.label=2;case 2:return[2]}})})}function Uy(n){if(n==null)return;for(var t in n){var e=n[t];typeof e!="number"&&e.dispose()}}var By;(function(n){n[n.SILENT=0]="SILENT",n[n.VERBOSE=1]="VERBOSE"})(By||(By={}));var W3=125,ia=function(){function n(){this.validationData=null}return n.prototype.setParams=function(t){this.params=t},n.prototype.onEpochBegin=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onEpochEnd=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchBegin=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchEnd=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onTrainBegin=function(t){return Se(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.onTrainEnd=function(t){return Se(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.setModel=function(t){},n}(),zy=function(){function n(t,e){e===void 0&&(e=10),t==null&&(t=[]),this.callbacks=t,this.queueLength=e}return n.prototype.append=function(t){this.callbacks.push(t)},n.prototype.setParams=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setParams(t)}},n.prototype.setModel=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setModel(t)}},n.prototype.onEpochBegin=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onEpochEnd=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchBegin=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchEnd=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onTrainBegin=function(t){return Se(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainBegin(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n.prototype.onTrainEnd=function(t){return Se(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainEnd(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n}(),U3=function(n){Q(t,n);function t(){return n.call(this)||this}return t.prototype.onEpochBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.seen=0,this.totals={},[2]})})},t.prototype.onBatchEnd=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l=this;return ve(this,function(u){i==null&&(i={}),r=i.size==null?0:i.size,this.seen+=r,a=function(c){var h=i[c];if(typeof h=="number")s.totals.hasOwnProperty(c)||(s.totals[c]=0),s.totals[c]=s.totals[c]+h*r;else{var d=void 0;c in s.totals?d=s.totals[c]:s.totals[c]=0;var p=y.tidy(function(){return y.add(l.totals[c],y.mul(h,r))});s.totals[c]=p,d!=null&&d.dispose()}},s=this;for(o in i)a(o);return[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l,u=this;return ve(this,function(c){if(i!=null)for(r=function(h){if(a.totals[h]==null)return"continue";typeof a.totals[h]=="number"?i[h]=a.totals[h]/a.seen:y.tidy(function(){var d=y.mul(y.div(1,u.seen),u.totals[h]);i[h]=d,u.totals[h].dispose(),y.keep(i[h])})},a=this,s=0,o=this.params.metrics;s<o.length;s++)l=o[s],r(l);return[2]})})},t}(ia),Py=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.epoch=[],this.history={},[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){i==null&&(i={}),this.epoch.push(e);for(r in i)this.history[r]==null&&(this.history[r]=[]),this.history[r].push(i[r]);return[2]})})},t.prototype.syncData=function(){return Se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:e=[],i=[],r=[];for(a in this.history)for(s=this.history[a],o=0;o<s.length;++o)typeof s[o]!="number"&&(l=s[o],e.push(l.data()),i.push(a),r.push(o));return[4,Promise.all(e)];case 1:for(u=d.sent(),c=0;c<u.length;++c)h=this.history[i[c]][r[c]],h.dispose(),this.history[i[c]][r[c]]=u[c][0];return[2]}})})},t}(ia),_y=function(n){Q(t,n);function t(e,i){var r=n.call(this)||this;if(r.currentEpoch=0,r.yieldEvery=i||"auto",r.yieldEvery==="auto"&&(r.yieldEvery=W3),r.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");return y.util.isNumber(r.yieldEvery)&&(r.maybeWait=Vk(r.maybeWait.bind(r),r.yieldEvery)),r.trainBegin=e.onTrainBegin,r.trainEnd=e.onTrainEnd,r.epochBegin=e.onEpochBegin,r.epochEnd=e.onEpochEnd,r.batchBegin=e.onBatchBegin,r.batchEnd=e.onBatchEnd,r.yield=e.onYield,r}return t.prototype.maybeWait=function(e,i,r){return Se(this,void 0,void 0,function(){var a;return ve(this,function(s){switch(s.label){case 0:return a=[],this.yield!=null?[4,Ci(r)]:[3,2];case 1:s.sent(),a.push(this.yield(e,i,r)),s.label=2;case 2:return a.push(y.nextFrame()),[4,Promise.all(a)];case 3:return s.sent(),[2]}})})},t.prototype.onEpochBegin=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.currentEpoch=e,this.epochBegin!=null?[4,Ci(i)]:[3,3];case 1:return r.sent(),[4,this.epochBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.epochEnd!=null?[4,Ci(i)]:[3,2];case 1:a.sent(),r.push(this.epochEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="epoch"&&r.push(y.nextFrame()),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onBatchBegin=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.batchBegin!=null?[4,Ci(i)]:[3,3];case 1:return r.sent(),[4,this.batchBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onBatchEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.batchEnd!=null?[4,Ci(i)]:[3,2];case 1:a.sent(),r.push(this.batchEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="batch"?r.push(y.nextFrame()):y.util.isNumber(this.yieldEvery)&&r.push(this.maybeWait(this.currentEpoch,e,i)),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainBegin!=null?[4,Ci(e)]:[3,3];case 1:return i.sent(),[4,this.trainBegin(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t.prototype.onTrainEnd=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainEnd!=null?[4,Ci(e)]:[3,3];case 1:return i.sent(),[4,this.trainEnd(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t}(ia);function My(n,t){if(n==null&&(n={}),n instanceof ia)return[n];if(Array.isArray(n)&&n[0]instanceof ia)return n;var e=Je(n);return e.map(function(i){return new _y(i,t)})}var Hy=function(){function n(){}return n.registerCallbackConstructor=function(t,e){y.util.assert(t>=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function Vy(n,t,e,i,r,a,s,o,l){var u=new Py,c=[new U3].concat(Hy.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new zy(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function An(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function Eo(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(ns(n),t,!0),i=y.fill(e.shape,yt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function pr(n,t){return y.tidy(function(){return y.mean(ns(y.sub(t,n)),-1)})}function Do(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function ra(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),yt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function B3(n,t){return y.tidy(function(){var e=y.clipByValue(t,yt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,yt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(ns(y.sub(i,a)),-1)})}function z3(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(ns(e),-1)})}function P3(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function _3(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function M3(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function as(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,yt(),1-yt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function ko(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(i3(n)).toInt();t=y.clipByValue(t,yt(),1-yt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return as(a,t,e)})}function H3(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function Fo(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,yt(),1-yt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(H3(n,e),-1)})}function V3(n,t){return y.tidy(function(){var e=y.clipByValue(n,yt(),1),i=y.clipByValue(t,yt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function q3(n,t){return y.tidy(function(){var e=y.log(y.add(yt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function td(n,t){return y.tidy(function(){var e=Eo(n,-1),i=Eo(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var Wo={meanSquaredError:pr,meanAbsoluteError:Do,meanAbsolutePercentageError:ra,meanSquaredLogarithmicError:B3,squaredHinge:z3,hinge:P3,categoricalHinge:_3,logcosh:M3,categoricalCrossentropy:as,sparseCategoricalCrossentropy:ko,binaryCrossentropy:Fo,kullbackLeiblerDivergence:V3,poisson:q3,cosineProximity:td};function nd(n){if(typeof n=="string"){if(n in Wo)return Wo[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function id(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=es(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function rd(n,t){return y.tidy(function(){return es(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function qy(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function G3(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function Y3(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function Gy(n,t){return y.tidy(function(){var e=qy(n,t),i=Y3(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function j3(n,t){return y.tidy(function(){var e=qy(n,t),i=G3(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Yy(n,t){return Fo(n,t)}function jy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var K3=pr,$3=pr,X3=Do,J3=Do,Z3=ra,Q3=ra,ad=as,e4=td,Ky=ko,Uo={binaryAccuracy:id,categoricalAccuracy:rd,precision:Gy,categoricalCrossentropy:ad,sparseCategoricalCrossentropy:Ky,mse:K3,MSE:$3,mae:X3,MAE:J3,mape:Z3,MAPE:Q3,cosine:e4};function t4(n){if(typeof n=="string"&&n in Uo)return Uo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function Bo(n){if(_n(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(Wo);e<i.length;e++){var r=i[e];if(Wo[r]===n){t=r;break}}if(t!==void 0)return t;for(var a=0,s=Object.keys(Uo);a<s.length;a++){var r=s[a];if(Uo[r]===n){t=r;break}}return t!==void 0?t:n.name}function n4(n){var t={Adagrad:function(){return y.train.adagrad(.01)},Adadelta:function(){return y.train.adadelta(1,.95,yt())},Adam:function(){return y.train.adam(.001,.9,.999,yt())},Adamax:function(){return y.train.adamax(.002,.9,.999,yt(),0)},RMSProp:function(){return y.train.rmsprop(.001,.9,0,yt())},SGD:function(){return y.train.sgd(.01)}};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,n in t)return t[n]();throw new M("Unknown Optimizer "+n)}var $y=1*1024*1024;function Xy(n,t,e){if(e===void 0&&(e=!1),n==null||typeof n!="object"||Object.getPrototypeOf(n)!==Object.prototype||!sd(n))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){var i=JSON.stringify(n);i.length>$y&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+($y+"."))}}function sd(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e];if(typeof r!="string")return!1;if(!sd(n[r]))return!1}return!0}else if(Array.isArray(n)){for(var a=0,s=n;a<s.length;a++){var o=s[a];if(!sd(o))return!1}return!0}else return!1;else{var l=typeof n;return l==="string"||l==="number"||l==="boolean"}}function o4(n,t,e,i){i===void 0&&(i=console.log);var r=r4(n),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,e=e||[.45,.85,1]):(t=t||98,e=e||[.33,.55,.67,1]),e[e.length-1]<=1&&(e=e.map(function(d){return Math.floor(t*d)}));var s;if(!r){a.push("Receives inputs"),s=[];for(var o in n.nodesByDepth)s.push.apply(s,n.nodesByDepth[o])}i("_".repeat(t)),zo(a,e,i),i("=".repeat(t));for(var l=n.layers,u=0;u<l.length;++u)r?a4(l[u],e,i):s4(l[u],e,s,i),i((u===l.length-1?"=":"_").repeat(t));n.checkTrainableWeightsConsistency();var c=i4(n),h=Ro(n.nonTrainableWeights);i("Total params: "+(c+h)),i("Trainable params: "+c),i("Non-trainable params: "+h),i("_".repeat(t))}function i4(n){var t;return n.collectedTrainableWeights!=null?t=Ro(n.collectedTrainableWeights):t=Ro(n.trainableWeights),t}function r4(n){var t=!0,e=[],i=[];for(var r in n.nodesByDepth)e.push(n.nodesByDepth[r]);for(var a=0,s=e;a<s.length;a++){var o=s[a];if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l<u.length;l++){for(var c=u[l],h=!1,d=0,p=c.inboundNodes;d<p.length;d++){var f=p[d];if(i.indexOf(f)!==-1)if(h){t=!1;break}else h=!0}if(!t)break}return t}function zo(n,t,e){e===void 0&&(e=console.log);for(var i="",r=0;r<n.length;++r)r>0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function a4(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];zo(s,t,e)}function s4(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s<o.length;s++){var l=o[s];if(e!=null&&e.length>0&&e.indexOf(l)===-1)continue;for(var u=0;u<l.inboundLayers.length;++u){var c=l.inboundLayers[u].name,h=l.nodeIndices[u],d=l.tensorIndices[u];a.push(c+"["+h+"]["+d+"]")}}var p=n.name,f=n.getClassName(),m=a.length===0?"":a[0],g=[p+" ("+f+")",r,n.countParams().toString(),m];zo(g,t,i);for(var u=1;u<a.length;++u)zo(["","","",a[u]],t,i)}function Jy(n,t,e){return(n==="inboundNodes"||n==="outputLayers"||n==="inputLayers")&&t===0&&typeof e=="string"}function ss(n,t){if(n===null)return null;if(typeof n=="string")return cr(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];Jy(t,r,a)?e.push(a):e.push(ss(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u];if(u==="name"&&typeof c=="string")s[u]=c;else{var h=cr(u);s[h]=ss(c,h)}}return s}}function od(n,t){if(n==null)return null;if(typeof n=="string")return ai(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];Jy(t,r,a)?e.push(a):e.push(od(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u],h=ai(u);(u==="name"||u==="className")&&typeof c=="string"?s[h]=c:s[h]=od(c,u)}return s}}var ld="2.6.0";function l4(n,t){if(n.dtype==null||n.dtype===t.dtype)return t;try{return y.cast(t,n.dtype)}catch(e){throw new M("The dtype of the feed ("+t.dtype+") can not be cast to the dtype "+("of the key '"+n.name+"' ("+n.dtype+")."))}}var aa=function(){function n(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof n)for(var e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(var i=0,r=t;i<r.length;i++){var a=r[i];this.add(a.key,a.value)}}}return n.prototype.add=function(t,e,i){if(this.id2Value[t.id]==null)this.id2Value[t.id]=l4(t,e),this.name2Id[t.name]=t.id,i!=null&&(this.id2Mask[t.id]=i);else throw new M("Duplicate key: name="+t.name+", id="+t.id);return this},n.prototype.addFeed=function(t){this.add(t.key,t.value)},n.prototype.hasKey=function(t){return this.id2Value[t.id]!=null},n.prototype.names=function(){return Object.keys(this.name2Id)},n.prototype.getValue=function(t){if(t instanceof In){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Value[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Value[e]}},n.prototype.getMask=function(t){if(t instanceof In){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Mask[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Mask[e]}},n.prototype.disposeMasks=function(){this.id2Mask!=null&&y.dispose(this.id2Mask)},n}(),ud={},Zy={};function os(n,t,e,i){for(var r=e==null?!1:e.training,a=Array.isArray(n),s=a?n:[n],o=s.map(function(X){return X.name}),l=[],u=t.names(),c=0,h=o;c<h.length;c++){var d=h[c];u.indexOf(d)!==-1?l.push(t.getValue(d)):l.push(null)}i!=null&&(i.maxNumTensors=-Infinity,i.minNumTensors=Infinity);var p=o.join(",")+"|"+t.names().join(","),f,m;if(ud[p]==null){var g=u4(s,t);f=g.sorted,m=g.recipientCounts,ud[p]=f,Zy[p]=m}f=ud[p],m={},r||Object.assign(m,Zy[p]);for(var v=new aa(t),b=0;b<f.length;++b){if(i!=null){var w=y.memory().numTensors;w>i.maxNumTensors&&(i.maxNumTensors=w),w<i.minNumTensors&&(i.minNumTensors=w)}var S=f[b],L=S.sourceLayer;if(L instanceof rs)continue;for(var N=[],C=[],R=[],D=!1,k=0,W=S.inputs;k<W.length;k++){var F=W[k],P=v.getValue(F),H=v.getMask(F);N.push(P),C.push(H),H!=null&&(D=!0),r||(m[F.name]--,m[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!P.isDisposed&&F.sourceLayer.stateful!==!0&&R.push(P))}D&&(e=e||{},e.mask=C[0]);var _=Je(L.apply(N,e)),j=null;L.supportsMasking&&(j=L.computeMask(N,C));for(var K=c4(S),q=Array.isArray(K)?K:[K],G=0;G<q.length;++G){v.hasKey(q[G])||v.add(q[G],_[G],Array.isArray(j)?j[0]:j);var Z=o.indexOf(q[G].name);Z!==-1&&(l[Z]=_[G])}r||y.dispose(R)}return v.disposeMasks(),a?l:l[0]}function u4(n,t){y.util.assert(n!=null&&n.length>0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=Qy(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s<o.length;s++){for(var l=o[s],u=Qy(l,t),c=u.sorted,h=u.recipientMap,d=0,p=c;d<p.length;d++){var f=p[d];a.has(f.name)||(e.push(f),a.add(f.name))}var m=function(v){i[v]==null&&(i[v]=new Set),h[v].forEach(function(b){return i[v].add(b)})};for(var g in h)m(g)}return{sorted:e,recipientCounts:h4(i)}}function h4(n){var t={};for(var e in n)t[e]=n[e].size;return t}function Qy(n,t){for(var e=new Set,i=[],r={},a=0,s=t.names();a<s.length;a++){var o=s[a];e.add(o)}var l=[],u=[];for(l.push(n);l.length>0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d<p.length;d++){var f=p[d];if(r[f.name]==null&&(r[f.name]=new Set),r[f.name].add(c.name),e.has(f.name))continue;l.push(f)}}}return{sorted:i,recipientMap:r}}function c4(n){var t;if(n.sourceLayer.inboundNodes.length===1)t=n.sourceLayer.output;else{for(var e=null,i=0;i<n.sourceLayer.inboundNodes.length;++i)for(var r=0,a=n.sourceLayer.inboundNodes[i].outputTensors;r<a.length;r++){var s=a[r];if(s.id===n.id){e=i;break}}t=n.sourceLayer.getOutputAt(e)}return t}var d4=function(n){Q(t,n);function t(e){var i=n.call(this,{})||this;if(i.containerNodes=new Set,i.name=e.name,i.name==null){var r=i.getClassName().toLowerCase();i.name=xo(r)}if(i.supportsMasking=!1,i.trainable_=!0,Array.isArray(e.inputs)?i.inputs=e.inputs.slice():i.inputs=[e.inputs],Array.isArray(e.outputs)?i.outputs=e.outputs.slice():i.outputs=[e.outputs],Ti(i.inputs).length!==i.inputs.length)throw new M("The list of inputs passed to the model is redundant. All inputs should only appear once. Found: "+(""+i.inputs.map(function(Ce){return Ce.name})));Ti(i.outputs).length!==i.outputs.length&&console.warn("The list of outputs passed to the model is redundant. All outputs should only appear once. Found: "+(""+i.outputs.map(function(Ce){return Ce.name}))),i.inputLayers=[],i.inputLayersNodeIndices=[],i.inputLayersTensorIndices=[],i.outputLayers=[],i.outputLayersNodeIndices=[],i.outputLayersTensorIndices=[],i.layers=[],i.internalContainerRefs=[];for(var a=0,s=i.outputs;a<s.length;a++){var o=s[a],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;i.outputLayers.push(l),i.outputLayersNodeIndices.push(u),i.outputLayersTensorIndices.push(c)}for(var h=0,d=i.inputs;h<d.length;h++){var o=d[h],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;_n(u===0,"input layer has >1 nodes"),_n(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p<i.inputLayers.length;p++){var l=i.inputLayers[p];if(!(l instanceof rs))throw new TypeError("Input layers to a LayersModel must be InputLayer objects. "+("Received inputs: "+e.inputs+". ")+("Input "+p+" (0-based) originates ")+("from layer type "+l.getClassName()+"."));i.inputNames.push(l.name),i.feedInputShapes.push(l.batchInputShape),i.feedInputNames.push(l.name)}for(var f=0,m=i.outputLayers;f<m.length;f++){var l=m[f];i.outputNames.push(l.name)}i.internalInputShapes=i.inputs.map(function(Ce){return Ce.shape}),i.internalOutputShapes=i.outputs.map(function(Ce){return Ce.shape});for(var g={},v={},b={},w={},S={},L=[],N=function(Ce,Wt,ft,it,cn,Os){(it==null||cn==null||Os==null)&&(it=Ce.sourceLayer,cn=Ce.nodeIndex,Os=Ce.tensorIndex);var _t=it.inboundNodes[cn];if(ft.indexOf(_t)!==-1)throw new Pn("The tensor "+Ce.name+' at layer "'+it.name+'" is part of a cycle.');if(Wt.indexOf(_t)!==-1)return;i.containerNodes.add(t.nodeKey(it,cn)),it.id in S||(S[it.id]=Object.keys(S).length),ft.indexOf(_t)===-1&&ft.push(_t);for(var a2=_t.inboundLayers.length,Or=0;Or<a2;Or++){var s2=_t.inputTensors[Or],o2=_t.inboundLayers[Or],l2=_t.nodeIndices[Or],u2=_t.tensorIndices[Or];N(s2,Wt,ft,o2,l2,u2)}for(Wt.push(_t);ft.indexOf(_t)>=0;)ft.splice(ft.indexOf(_t),1);L.push(_t)},C=[],R=[],D=0,k=i.outputs;D<k.length;D++){var o=k[D];N(o,C,R)}for(var W=L.slice().reverse(),F=0,P=W;F<P.length;F++){var H=P[F];v[H.id]=H,H.id in g||(g[H.id]=0);var _=g[H.id],j=b[H.outboundLayer.id]==null?0:b[H.outboundLayer.id];_=Math.max(_,j),b[H.outboundLayer.id]=_,w[H.outboundLayer.id]=H.outboundLayer,g[H.id]=_;for(var p=0;p<H.inboundLayers.length;p++){var K=H.inboundLayers[p],u=H.nodeIndices[p],q=K.inboundNodes[u],G=g[q.id]==null?0:g[q.id];g[q.id]=Math.max(_+1,G),v[q.id]=q}}var Z={};for(var X in g){var _=g[X];_ in Z||(Z[_]=[]),Z[_].push(v[X])}var ee={};for(var ne in b){var _=b[ne];_ in ee||(ee[_]=[]),ee[_].push(w[ne])}var ie=Object.keys(ee).map(function(Ce){return parseInt(Ce,10)}).sort(Io);i.layers=[];for(var te=0,re=ie;te<re.length;te++){var _=re[te],le=ee[_];le.sort(function(Wt,ft){var it=S[Wt.id],cn=S[ft.id];return it<cn?-1:it>cn?1:0});for(var he=0,ye=le;he<ye.length;he++){var l=ye[he];l instanceof t&&i.internalContainerRefs.push(l),i.layers.push(l)}}i.layersByDepth=ee,ie=Object.keys(Z).map(function(Ce){return parseInt(Ce,10)}).sort(Io);for(var Oe=i.inputs.slice(),Fe=[],_e=0,Me=ie;_e<Me.length;_e++)for(var _=Me[_e],dt=0,Lt=Z[_];dt<Lt.length;dt++){var H=Lt[dt],l=H.outboundLayer;if(l!=null){for(var Ge=0,Ue=H.inputTensors;Ge<Ue.length;Ge++){var o=Ue[Ge];if(Oe.indexOf(o)===-1)throw new Pn("Graph disconnected: cannot obtain value for tensor "+o+(' at layer "'+l.name+'". ')+"The following previous layers were accessed without "+("issue: "+Fe))}for(var me=0,Ft=H.outputTensors;me<Ft.length;me++){var o=Ft[me];Oe.push(o)}Fe.push(l.name)}}i.nodesByDepth=Z;for(var lt=i.layers.map(function(Ce){return Ce.name}),pt=function(Ce){var Wt=lt.filter(function(ft){return ft===Ce}).length;if(Wt!==1)throw new Pn('The name "'+Ce+'" is used '+Wt+" times in the model. All layer names should be unique. Layer names: "+JSON.stringify(lt))},En=0,Et=lt;En<Et.length;En++){var It=Et[En];pt(It)}return i.outboundNodes=[],i.inboundNodes=[],new Oo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:i.inputs,outputTensors:i.outputs,inputMasks:i.inputs.map(function(Ce){return null}),outputMasks:i.outputs.map(function(Ce){return null}),inputShapes:i.inputs.map(function(Ce){return Ce.shape}),outputShapes:i.outputs.map(function(Ce){return Ce.shape})}),i.built=!0,i._refCount=1,i}return t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Container '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){this.assertNotDisposed();var e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(var i=0,r=this.layers;i<r.length;i++){var a=r[i];e.numDisposedVariables+=a.dispose().numDisposedVariables}for(var s=0,o=this.internalContainerRefs;s<o.length;s++){var l=o[s];e.numDisposedVariables+=l.dispose().numDisposedVariables}}return e.refCountAfterDispose=this._refCount,e},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this.layers.forEach(function(i){i._trainableWeights.forEach(function(r){return r.trainable=e})}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(this._trainableWeights.length>0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e=e.concat(a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.layers;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.loadWeights=function(e,i){i===void 0&&(i=!0);for(var r={},a=0,s=0,o=this.layers;s<o.length;s++)for(var l=o[s],u=0,c=l.weights;u<c.length;u++){var h=c[u];if(r[h.originalName]!=null)throw new M("Duplicate weight name: "+h.originalName);r[h.originalName]=h,a++}var d=[];for(var p in e){var f=p;if(r[p]==null){var m=p.split("/"),g=m.slice(0,-2).concat([m[m.length-1]]);f=g.join("/")}if(r[f]!=null)d.push([r[f],e[p]]);else if(i)throw new M("Provided weight data has no target variable: "+p);delete r[f]}if(i){var v=[];for(var b in r)v.push(b);if(v.length>0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}ed(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+ld,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=od(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Je(e);for(var a=new aa,s=0;s<r.inputs.length;++s)a.add(r.inputs[s],e[s]);return os(r.outputs,a,i)})},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){e=Je(e);var a;return i==null?a=ur(null,e.length):a=Je(i),r.runInternalGraph(e,a)[1]})},t.prototype.computeOutputShape=function(e){var i=Co(e);if(i.length!==this.inputLayers.length)throw new M("Invalid inputShape argument "+e+": "+("model has "+this.inputLayers.length+" tensor inputs."));for(var r={},a=0;a<i.length;a++){var s=this.inputLayers[a],o=i[a],l=s.name+"_0_0";r[l]=o}var u=Object.keys(this.nodesByDepth).map(function(P){return parseInt(P,10)}).sort(Io);if(u.length>1)for(var c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){var g=m[f],s=g.outboundLayer;if(this.inputLayers.map(function(H){return H.id}).indexOf(s.id)!==-1)continue;for(var v=[],b=0;b<g.inboundLayers.length;b++){var w=g.inboundLayers[b],S=g.nodeIndices[b],L=g.tensorIndices[b],l=w.name+"_"+S+"_"+L,N=r[l];v.push(N)}for(var C=s.computeOutputShape(qt(v)),R=Co(C),D=s.inboundNodes.indexOf(g),b=0;b<R.length;b++){var l=s.name+"_"+D+"_"+b;r[l]=R[b]}}for(var k=[],W=[],a=0;a<this.outputLayers.length;a++){var s=this.outputLayers[a],D=this.outputLayersNodeIndices[a],L=this.outputLayersTensorIndices[a],l=s.name+"_"+D+"_"+L;W.push(l)}for(var a=0;a<W.length;a++){var F=W[a];_n(F in r),k.push(r[F])}return qt(k)},t.prototype.runInternalGraph=function(e,i){i==null&&(i=ur(null,e.length));for(var r={},a=0;a<this.inputs.length;++a){var s=this.inputs[a],o=e[a],l=i[a];r[s.id]=[o,l]}for(var u=Object.keys(this.nodesByDepth).map(function(ee){return parseInt(ee,10)}).sort(Io),c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){for(var g=m[f],v=g.outboundLayer,b=g.inputTensors,w=g.outputTensors,S=new Array,L=0,N=b;L<N.length;L++){var s=N[L];s.id in r&&S.push(r[s.id])}if(S.length===b.length){var C={},R=void 0,D=void 0,k=void 0,W=void 0;if(g.callArgs!=null&&(C=g.callArgs),S.length===1){var F=S[0],P=F[0],H=F[1];C.mask==null&&(C.mask=H),k=Je(v.call(P,C)),W=Je(v.computeMask(P,H)),R=[P],D=[H]}else R=S.map(function(ee){return ee[0]}),D=S.map(function(ee){return ee[1]}),C.mask==null&&(C.mask=D),k=Je(v.call(R,C)),W=Je(v.computeMask(R,D));if(v.activityRegularizer)throw new Te("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(var a=0;a<w.length;++a){var s=w[a],o=k[a],l=W[a];r[s.id]=[o,l]}}}for(var _=[],j=[],K=[],q=0,G=this.outputs;q<G.length;q++){var s=G[q];_n(s.id in r,"Could not compute output "+s.name+" : "+s.id);var Z=r[s.id],X=Z[0],l=Z[1];K.push(X.shape),_.push(X),j.push(l)}return[_,j,K]},t.prototype.buildNodeConversionMap=function(e){for(var i={},r,a=0,s=this.layers;a<s.length;a++){var o=s[a];r=o instanceof t?1:0;for(var l=0;l<o.inboundNodes.length;l++){var u=t.nodeKey(o,l);this.containerNodes.has(u)&&(i[u]=r,r+=1)}}return i},t.prototype.getLayer=function(e,i){if(i!=null){if(this.layers.length<=i)throw new M("Was asked to retrieve layer at index "+i+", but model only "+("has "+this.layers.length+" layer(s)."));return this.layers[i]}else if(e==null)throw new M("Provide either a layer name or layer index");for(var r=0,a=this.layers;r<a.length;r++){var s=a[r];if(s.name===e)return s}throw new M("No such layer: "+e)},t.prototype.calculateLosses=function(){var e=this;return y.tidy(function(){for(var i=[],r=0,a=e.layers;r<a.length;r++)for(var s=a[r],o=0;o<s.inboundNodes.length;++o){var l=t.nodeKey(s,o);e.containerNodes.has(l)&&i.push.apply(i,s.calculateLosses())}return i})},t.prototype.getConfig=function(){for(var e={name:this.name},i=this.buildNodeConversionMap(this.layers),r=[],a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=o.getClassName(),u=o.getConfig(),c=[],h=0;h<o.inboundNodes.length;h++){var d=o.inboundNodes[h],p=t.nodeKey(o,h),f={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),f=d.callArgs}catch(D){console.warn("Layer "+o.name+" was passed non-serializable keyword arguments: "+(d.callArgs+". They will not be included ")+"in the serialized model (and thus will be missing at deserialization time)."),f={}}if(d.inboundLayers.length>0){for(var m=[],g=0;g<d.inboundLayers.length;g++){var v=d.inboundLayers[g],b=d.nodeIndices[g],w=d.tensorIndices[g],S=t.nodeKey(v,b),L=i[S];L==null&&(L=0),m.push([v.name,L,w,f])}c.push(m)}}}var N={};N.name=o.name,N.className=l,N.config=u,N.inboundNodes=c,r.push(N)}e.layers=r;for(var C=[],g=0;g<this.inputLayers.length;g++){var o=this.inputLayers[g],b=this.inputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.inputLayersTensorIndices[g];C.push([o.name,L,w])}e.inputLayers=C;for(var R=[],g=0;g<this.outputLayers.length;g++){var o=this.outputLayers[g],b=this.outputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.outputLayersTensorIndices[g];R.push([o.name,L,w])}return e.outputLayers=R,e},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s={},o={};function l(G,Z){G.name in o?o[G.name].push(Z):o[G.name]=[Z]}function u(G,Z){for(var X=[],ee,ne=0,ie=Z;ne<ie.length;ne++){var te=ie[ne],re=te[0],le=te[1],he=te[2];if(ee=te[3]==null?{}:te[3],!(re in s)){l(G,Z);return}var ye=s[re];if(ye.inboundNodes.length<=le){l(G,Z);return}var Oe=ye.inboundNodes[le];X.push(Oe.outputTensors[he])}X.length>0&&G.apply(qt(X),ee)}function c(G){var Z=G.name,X=An(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p<f.length;p++){var m=f[p];c(m)}for(;!Hk(o);)for(var g=0,v=d;g<v.length;g++){var m=v[g],b=s[m.name];if(b.name in o){var w=o[b.name];delete o[b.name];for(var S=0,L=w;S<L.length;S++){var N=L[S];u(b,N)}}}for(var C=[],R=[],D=i.inputLayers,k=0,W=D;k<W.length;k++){var m=W[k],F=m[0],P=m[1],H=m[2];_n(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;C.push(_[H])}for(var j=i.outputLayers,K=0,q=j;K<q.length;K++){var m=q[K],F=m[0],P=m[1],H=m[2];_n(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;R.push(_[H])}return new e({inputs:C,outputs:R,name:h})},Object.defineProperty(t.prototype,"stateful",{get:function(){if(this._stateful)throw new M("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(var e=0,i=this.layers;e<i.length;e++){var r=i[e];if(r.stateful)return!0}return!1},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){var e=this;y.tidy(function(){e.layers.forEach(function(i){i.stateful&&i.resetStates()})})},t}(De);function p4(n,t,e){var i=t.length;if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(a){return null});if(i===1)return Array.isArray(n)&&n.length===1?n:typeof n=="object"&&t[0]in n?[n[t[0]]]:[n];if(Array.isArray(n)){if(n.length!==i)throw new Error("Provided "+e+" is an array of "+n.length+" "+("element(s), but the model has "+i+" outputs. ")+"Make sure a set of weights is provided for each model output.");return n}else if(typeof n=="object"&&Object.keys(n).length>0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function eb(n,t){return p4(n,t,"classWeight")}function tb(n,t,e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function f4(n,t){return y.mul(n,t)}var m4=32;function ib(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor.  The provided Dataset instead generates "+(""+t)});var a=nb("input",n.inputNames,e),s=nb("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs.  (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs.  (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},u=0;u<a.length;u++)l(u);for(var c=function(d){y.util.assert(s[d].shape[0]===o,function(){return"Batch size mismatch: output "+(n.outputNames[d]+" has "+s[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},h=0;h<s.length;h++)c(h);return{xs:a,ys:s}}function nb(n,t,e){if(e instanceof y.Tensor)return[e];if(Array.isArray(e))return y.util.assert(e.length===t.length,function(){return"Received an array of "+e.length+" Tensors, but expected "+t.length+" to match the "+n+" keys "+t+"."}),e;for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r];if(e[s]==null)throw new M("The feature data generated by the dataset lacks the required "+(n+" key '"+s+"'."));i.push(e[s])}return i}function g4(n){if(n.length===3)throw new Te("Validation with sample weights is not implemented yet.");return{xs:n[0],ys:n[1]}}function y4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R,D,k,W,F,P,H,_,j,F,K,q,G,Z,F;return ve(this,function(X){switch(X.label){case 0:if(i=e.batchesPerEpoch!=null,y.util.assert(n.optimizer!=null,function(){return"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."}),y.util.assert(e!=null,function(){return"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."}),y.util.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(rb(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=g4(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=My(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=Vy(h,d,e.epochs,null,null,v4(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g<e.epochs?(b={},[4,f.onEpochBegin(g)]):[3,23];case 5:return X.sent(),w=0,S=0,i?[3,7]:[4,t.iterator()];case 6:v=X.sent(),X.label=7;case 7:return(i?w<e.batchesPerEpoch:!0)?[4,v.next()]:[3,21];case 8:return L=X.sent(),i&&L.done?(console.warn("You provided `batchesPerEpoch` as "+(e.batchesPerEpoch+", ")+"but your dataset iterator ran out of data after "+(w+" batches; ")+"interrupting training. Make sure that your dataset can generate at least `batchesPerEpoch * epochs` batches (in this case, "+(e.batchesPerEpoch*e.epochs+" batches). ")+"You may need to use the repeat() function when building your dataset."),[3,21]):L.value!=null?(N=ib(n,L.value),C=N.xs,R=N.ys,D={},D.batch=S,D.size=C[0].shape[0],[4,f.onBatchBegin(S,D)]):[3,15];case 9:if(X.sent(),k=[],!(e.classWeight!=null))return[3,13];W=eb(e.classWeight,n.outputNames),F=0,X.label=10;case 10:return F<W.length?(H=(P=k).push,[4,tb(R[F],null,W[F])]):[3,13];case 11:H.apply(P,[X.sent()]),X.label=12;case 12:return++F,[3,10];case 13:for(_=C.concat(R).concat(k),j=l(_),y.dispose(_),F=0;F<u.length;++F)K=u[F],q=j[F],D[K]=q,y.keep(q);return[4,f.onBatchEnd(S,D)];case 14:X.sent(),Uy(D),S++,w++,X.label=15;case 15:return(i?w>=e.batchesPerEpoch:L.done)?r?(G=void 0,rb(e.validationData)?(Z=Je,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Je(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?m4:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F<n.metricsNames.length;++F)b["val_"+n.metricsNames[F]]=G[F];X.label=19;case 19:return[3,21];case 20:return n.stopTraining_?[3,21]:[3,7];case 21:return[4,f.onEpochEnd(g,b)];case 22:return X.sent(),g++,n.stopTraining_?[3,23]:[3,4];case 23:return[4,f.onTrainEnd()];case 24:return X.sent(),[4,n.history.syncData()];case 25:return X.sent(),[2,n.history];case 26:return n.isTraining=!1,[7];case 27:return[2]}})})}function v4(n,t){var e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(n.size)&&(e=n.size),e}function rb(n){return typeof n.iterator=="function"}function b4(n){return typeof n.next=="function"}function w4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:if(e=e||{},i=e.batches!=null,r=n.testFunction,a=[],e.verbose>0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),b4(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=ib(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var N=0;N<L.length;++N)a.push(y.scalar(0));for(var C=S[0].shape[0],R=function(D){var k=L[D],W=a[D];a[D]=y.tidy(function(){return y.add(a[D],y.mul(C,k))}),u>0&&y.dispose(W)},N=0;N<L.length;++N)R(N);y.dispose(L),l+=C,++u}return a}),m.done?(i&&console.warn("Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least `batches` "+("batches (in this case, "+e.batches+" batches). ")+"You may need to use the repeat() function when building your dataset."),[2,"break"]):[2]}})},f.label=4;case 4:return(i?u<e.batches:!0)?[5,c()]:[3,6];case 5:return h=f.sent(),h==="break"?[3,6]:[3,4];case 6:for(d=0;d<a.length;++d)p=a[d],a[d]=y.div(a[d],l),y.dispose(p);return[2,qt(a)]}})})}function cd(n){y.util.assert(n>0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function ls(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return dr(i,t,e-t)}):dr(n,t,e-t)}function hd(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return hd(e,t)}):wy(n,t.dtype==="int32"?t:t.toInt())})}function dd(n,t){for(var e=[],i=0,r=null;i<n;)r=i+t,r>=n&&(r=n),e.push([i,r]),i=r;return e}function S4(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Se(this,void 0,void 0,function(){var m,g,v,b,w,S,L,N,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=Ln(0,g)),s==null&&(s=1),b=Vy(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(j){switch(j.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(j.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=dd(g,r),P=function(K){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(K,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[K][0],X=F[K][1],ee=dr(W,Z,X-Z);q.batch=K,q.size=X-Z;for(var ne=hd(e,ee),ie=t(ne),te=0;te<i.length;++te){var re=i[te],le=ie[te];q[re]=le,y.keep(le)}if(K===F.length-1&&m)for(var he=n.testLoop(l,u,r),te=0;te<i.length;++te){var re=i[te],le=he[te];y.keep(le),k["val_"+re]=le}}),[4,w.onBatchEnd(K,q)];case 2:return G.sent(),Uy(q),n.stopTraining_?[2,"break"]:[2]}})},H=0,j.label=3;case 3:return H<F.length?[5,P(H)]:[3,6];case 4:if(_=j.sent(),_==="break")return[3,6];j.label=5;case 5:return++H,[3,3];case 6:W.dispose(),j.label=7;case 7:return[4,w.onEpochEnd(D,k)];case 8:return j.sent(),n.stopTraining_?[2,"break"]:[2]}})},N=d,R.label=2;case 2:return N<a?[5,L(N)]:[3,5];case 3:if(C=R.sent(),C==="break")return[3,5];R.label=4;case 4:return++N,[3,2];case 5:return[4,w.onTrainEnd()];case 6:return R.sent(),[4,n.history.syncData()];case 7:return R.sent(),[2,n.history]}})})}function L4(n,t,e,i){return i===void 0&&(i={}),Se(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R,D,k;return ve(this,function(W){switch(W.label){case 0:if(n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,W.label=1;case 1:return W.trys.push([1,,7,8]),h=i.batchSize==null?32:i.batchSize,cd(h),d=!1,[4,n.standardizeUserData(t,e,i.sampleWeight,i.classWeight,d,h)];case 2:if(p=W.sent(),r=p[0],a=p[1],c=p[2],f=!1,m=void 0,!(i.validationData!=null&&i.validationData.length>0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=ls(r,b,w),r=ls(r,0,b),u=ls(a,b,w),a=ls(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),N=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=N.slice().concat(N.map(function(F){return"val_"+F}))):(C=null,m=[],R=N.slice()),D=My(i.callbacks,i.yieldEvery),[4,S4(n,L,S,N,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,fr(r,t),fr(a,e),fr(l,s),fr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function ab(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e<n.length;++e){var i=n[e];if(i.rank===1)t.push(ts(i,1));else{if(i.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(i)}}return t}function fr(n,t){if(n==null)return;var e=[];if(t instanceof y.Tensor)e.push(t.id);else if(Array.isArray(t))t.forEach(function(l){return e.push(l.id)});else if(t!=null)for(var i in t){var r=t[i];e.push(r.id)}var a=[];if(n instanceof y.Tensor)e.indexOf(n.id)===-1&&a.push(n);else if(Array.isArray(n))n.forEach(function(l){e.indexOf(l.id)===-1&&a.push(l)});else if(n!=null)for(var s in n){var o=n[s];e.indexOf(o.id)===-1&&a.push(o)}a.forEach(function(l){l.isDisposed||l.dispose()})}function I4(n){return n instanceof y.Tensor}function pd(n){return Array.isArray(n)}function sb(n){return!I4(n)&&!pd(n)}function ob(n,t,e,i,r){if(i===void 0&&(i=!0),r===void 0&&(r=""),t==null||t.length===0){if(n!=null){var a=!1;if(pd(n)&&n.length>0)a=!0;else if(sb(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(sb(n)){n=n,o=[];for(var l=0,u=t;l<u.length;l++){var c=u[l];if(n[c]==null)throw new M('No data provided for "'+c+'". Need data for each key in: '+(""+t));o.push(n[c])}}else if(pd(n)){if(n=n,n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("model expected. Expected to see "+t.length+" Tensor(s), but ")+("instead got the following list of Tensor(s): "+n));o=n}else{if(n=n,t.length>1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=ab(o),e!=null)for(var h=0;h<t.length;++h){if(e[h]==null)continue;var d=o[h];if(d.shape.length!==e[h].length)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have "+e[h].length+" dimension(s). but got array with ")+("shape "+d.shape));for(var p=0;p<e[h].length;++p){if(p===0&&!i)continue;var f=d.shape[p],m=e[h][p];if(m!=null&&m>=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function A4(n,t,e){var i=Ti(n.map(function(a){return a.shape[0]}));i.sort();var r=Ti(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function T4(n,t,e){for(var i=[pr,Fo,as],r=0;r<n.length;++r){var a=n[r],s=t[r],o=e[r];if(s==null)continue;if(s===as&&a.shape[a.shape.length-1]===1)throw new M("You are passing a target array of shape "+a.shape+" while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].");if(i.indexOf(s)!==-1)for(var l=a.shape.slice(1),u=o.slice(1),c=0;c<l.length;++c){var h=l[c],d=u[c];if(d!=null&&h!==d)throw new M("A target Tensor with shape "+a.shape+" was passed for an "+("output of shape "+o+", while using a loss function that ")+"expects targets to have the same shape as the output.")}}}function lb(n,t,e,i,r){i===void 0&&(i=!0),r===void 0&&(r="");var a;if(Array.isArray(n)){if(n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("the model expected. Expected to see "+t.length+" Tensor(s),")+(" but instead got "+n.length+" Tensors(s)."));a=n}else{if(t.length>1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s<t.length;++s){if(e[s]==null)continue;var o=a[s];if(o.shape.length!==e[s].length)throw new M("Error when checking "+r+": expected "+t[s]+" "+("to have "+e[s].length+" dimension(s), but got array with ")+("shape "+JSON.stringify(o.shape)));for(var l=0;l<e[s].length;++l){if(l===0&&!i)continue;var u=o.shape[l],c=e[s][l];if(c!=null&&c!==u)throw new M("Error when checking "+r+": expected "+(t[s]+" to have shape "+JSON.stringify(e[s])+" but ")+("got array with shape "+JSON.stringify(o.shape)+"."))}}}function N4(n,t){if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(l){return[]});var e;if(typeof n=="string"||typeof n=="function")e=[n];else if(Array.isArray(n)||typeof n=="object")e=n;else throw new TypeError("Type of metrics argument not understood. Expected an string,"+("function, Array, or Object, found: "+n));if(Array.isArray(e))return t.map(function(l){return e});for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r],o=e.hasOwnProperty(s)?e[s]:[];Array.isArray(o)||(o=[o]),i.push(o)}return i}var x4="layers-model",Ri=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.isTraining=!1,i}return t.prototype.summary=function(e,i,r){if(r===void 0&&(r=console.log),!this.built)throw new M("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");o4(this,e,i,r)},t.prototype.compile=function(e){var i=this;if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=n4(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof y.Optimizer))throw new M("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}var r=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(var a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new M('Unknown entry in loss dictionary: "'+a+'". '+("Only expected the following keys: "+this.outputNames));for(var s=0,o=this.outputNames;s<o.length;s++){var l=o[s];e.loss[l]==null&&console.warn('Output "'+l+'" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data '+("to be passed to "+l+" during training")),r.push(nd(e.loss[l]))}}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new M("When passing an Array as loss, it should have one entry per "+("model output. The model has "+this.outputs.length+" output(s), ")+("but you passed loss="+e.loss+"."));var u=e.loss;r=u.map(function(v){return nd(v)})}else{var c=nd(e.loss);this.outputs.forEach(function(v){r.push(c)})}this.lossFunctions=r,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(var h=0;h<this.outputs.length;++h){var d=this.internalOutputShapes[h],p=this.outputNames[h];this.feedOutputNames.push(p),this.feedOutputShapes.push(d),this.feedLossFns.push(this.lossFunctions[h])}var f=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],hr("loss",function(){for(var v=0;v<i.outputs.length;++v){if(f.indexOf(v)!==-1)continue;var b=i.lossFunctions[v];i.outputs.length>1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=N4(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};hr("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(N){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var j=i.internalOutputShapes[w];j[j.length-1]===1||i.lossFunctions[w]===Fo?["accuracy","acc"].indexOf(_)!==-1?D=id:["crossentropy","ce"].indexOf(_)!==-1&&(D=Yy):i.lossFunctions[w]===ko?["accuracy","acc"].indexOf(_)!==-1?D=jy:["crossentropy","ce"].indexOf(_)!==-1&&(D=Ky):["accuracy","acc"].indexOf(_)!==-1?D=rd:["crossentropy","ce"].indexOf(_)!==-1&&(D=ad);var K=void 0;["accuracy","acc"].indexOf(_)!==-1?K="acc":["crossentropy","ce"].indexOf(_)!==-1&&(K="ce"),k=D,R=C+K}else{var q=t4(_);k=q,R=C+Bo(_)}var G;hr(R,function(){G=k}),g(w,R,G)},F=0,P=N;F<P.length;F++){var H=P[F];W(H)}};L(S)},b=0;b<i.outputs.length;++b)v(b)}),this.collectedTrainableWeights=this.trainableWeights},t.prototype.checkTrainableWeightsConsistency=function(){if(this.collectedTrainableWeights==null)return;this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")},t.prototype.evaluate=function(e,i,r){r===void 0&&(r={});var a=r.batchSize==null?32:r.batchSize;cd(a);var s=!0,o=this.standardizeUserDataXY(e,i,s,a);try{var l=o[0].concat(o[1]);this.makeTestFunction();var u=this.testFunction,c=this.testLoop(u,l,a,r.verbose,r.steps);return qt(c)}finally{fr(o[0],e),fr(o[1],i)}},t.prototype.evaluateDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){return this.makeTestFunction(),[2,w4(this,e,i)]})})},t.prototype.checkNumSamples=function(e,i,r,a){a===void 0&&(a="steps");var s;if(r!=null){if(s=null,i!=null)throw new M("If "+a+" is set, batchSize must be null or undefined."+("Got batchSize = "+i))}else if(e!=null)Array.isArray(e)?s=e[0].shape[0]:s=e.shape[0];else throw new M("Either the input data should have a defined shape, or "+(a+" shoud be specified."));return s},t.prototype.execute=function(e,i){if(Array.isArray(i)&&i.length===0)throw new M("`outputs` is an empty Array, which is not allowed.");var r=Array.isArray(i),a=r?i:[i],s=this.retrieveSymbolicTensors(a),o=new aa;if(e instanceof y.Tensor&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new M("The number of inputs provided ("+e.length+") does not match the number of inputs of this model "+("("+this.inputs.length+")."));for(var l=0;l<this.inputs.length;++l)o.add(this.inputs[l],e[l])}else for(var u=0,c=this.inputs;u<c.length;u++){var h=c[u],d=e[h.name];if(d==null)throw new M("No value is provided for the model's input "+h.name);o.add(h,d)}var p=os(s,o);return r?p:p[0]},t.prototype.retrieveSymbolicTensors=function(e){for(var i=ur(null,e.length),r=e.length,a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=Array.isArray(o.output)?o.output:[o.output],u=l.map(function(p){return p.name}),c=0;c<e.length;++c){var h=u.indexOf(e[c]);if(h!==-1&&(i[c]=l[h],r--),r===0)break}if(r===0)break}if(r>0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=dd(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=ls(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v<m.length;++v)g.push({key:a.inputs[v],value:m[v]});else g.push({key:a.inputs[0],value:m});var b=new aa(g);return os(a.outputs,b)});d.forEach(function(p,f){return l[f].push(p)})},c=0;c<o.length;++c)u(c);return qt(l.map(function(h){return y.concat(h,0)}))})},t.prototype.predict=function(e,i){i===void 0&&(i={});var r=ab(e);lb(r,this.inputNames,this.feedInputShapes,!1);try{var a=i.batchSize==null?32:i.batchSize;return cd(a),this.predictLoop(r,a)}finally{fr(r,e)}},t.prototype.predictOnBatch=function(e){lb(e,this.inputNames,this.feedInputShapes,!0);var i=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,i)},t.prototype.standardizeUserDataXY=function(e,i,r,a){if(this.optimizer_==null)throw new Pn("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");for(var s=[],o=0;o<this.feedOutputShapes.length;++o){var l=this.feedOutputShapes[o],u=this.feedLossFns[o];u===ko?s.push(l.slice(0,l.length-1).concat([1])):s.push(l)}if(e=ob(e,this.feedInputNames,this.feedInputShapes,!1,"input"),i=ob(i,this.feedOutputNames,s,!1,"target"),A4(e,i),T4(i,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Se(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=eb(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p<d.length?(m=(f=h).push,[4,tb(c[p],null,d[p])]):[3,4];case 2:m.apply(f,[g.sent()]),g.label=3;case 3:return++p,[3,1];case 4:return[2,[u,c,h]]}})})},t.prototype.testLoop=function(e,i,r,a,s){var o=this;return a===void 0&&(a=0),y.tidy(function(){var l=o.checkNumSamples(i,r,s,"steps"),u=[];if(a>0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=dd(l,r),h=y.tensor1d(Ln(0,l)),d=0;d<c.length;++d){var p=c[d][0],f=c[d][1],m=dr(h,p,f-p),g=hd(i,m),v=e(g);if(d===0)for(var b=0;b<v.length;++b)u.push(y.scalar(0));for(var b=0;b<v.length;++b){var w=v[b];u[b]=y.add(u[b],y.mul(f-p,w))}}for(var b=0;b<u.length;++b)u[b]=y.div(u[b],l);return u})},t.prototype.getDedupedMetricsNames=function(){for(var e=this.metricsNames,i=[],r=0;r<e.length;++r){var a=e[r],s=a;if(ry(e,a)>1){var o=ry(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f<e.inputs.length;++f)p.push({key:e.inputs[f],value:a[f]});for(var m=new aa(p),g=os(e.outputs,m,{training:!0}),v,f=0;f<e.lossFunctions.length;++f){var b=e.lossFunctions[f],w=b(s[f],g[f]);o[f]!=null&&(w=f4(w,o[f]));var S=y.mean(w);r.push(S),f===0?v=w:v=y.add(v,w)}for(var f=0;f<e.metricsTensors.length;++f){var L=void 0;if(e.outputs.length>1&&f<e.outputs.length)L=r[f];else{var N=e.metricsTensors[f][0],C=e.metricsTensors[f][1];L=y.mean(N(s[C],g[C]))}y.keep(L),l.push(L)}return v=y.mean(v),e.calculateLosses().forEach(function(R){v=y.add(v,R)}),v},c=e.collectedTrainableWeights.map(function(p){return p.read()}),h=!0,d=e.optimizer_.minimize(u,h,c);return[d].concat(l)}},t.prototype.makeTestFunction=function(){var e=this;this.testFunction=function(i){return y.tidy(function(){for(var r=[],a,s=i.slice(0,e.inputs.length),o=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),l=[],u=0;u<e.inputs.length;++u)l.push({key:e.inputs[u],value:s[u]});for(var c=new aa(l),h=os(e.outputs,c),u=0;u<e.lossFunctions.length;++u){var d=e.lossFunctions[u],p=y.mean(d(o[u],h[u]));u===0?a=p:a=y.add(a,p),r.push(a)}for(var u=0;u<e.metricsTensors.length;++u){var f=e.metricsTensors[u][0],m=e.metricsTensors[u][1],g=y.mean(f(o[m],h[m]));r.push(g)}return r})}},t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Se(this,void 0,void 0,function(){return ve(this,function(a){return[2,L4(this,e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){return[2,y4(this,e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:return[4,this.standardizeUserData(e,i)];case 1:r=f.sent(),a=r[0],s=r[1],o=this.makeTrainFunction(),l=o(a.concat(s)),u=[],c=0,h=l,f.label=2;case 2:return c<h.length?(d=h[c],[4,d.data()]):[3,5];case 3:p=f.sent(),u.push(p[0]),f.label=4;case 4:return c++,[3,2];case 5:return y.dispose(l),[2,qt(u)]}})})},t.prototype.getNamedWeights=function(e){for(var i=[],r=e!=null&&e.trainableOnly,a=r?this.trainableWeights:this.weights,s=this.getWeights(r),o=0;o<a.length;++o){if(r&&!a[o].trainable)continue;i.push({name:a[o].originalName,tensor:s[o]})}return i},Object.defineProperty(t.prototype,"stopTraining",{get:function(){return this.stopTraining_},set:function(e){this.stopTraining_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.optimizer_},set:function(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)},enumerable:!0,configurable:!0}),t.prototype.dispose=function(){var e=n.prototype.dispose.call(this);if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){var i=y.memory().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=i-y.memory().numTensors}return e},t.prototype.getLossIdentifiers=function(){var e;if(typeof this.loss=="string")e=ai(this.loss);else if(Array.isArray(this.loss)){for(var i=0,r=this.loss;i<r.length;i++){var a=r[i];if(typeof a!="string")throw new Error("Serialization of non-string loss is not supported.")}e=this.loss.map(function(h){return ai(h)})}else{var s=Object.keys(this.loss);e={};for(var o=this.loss,l=0,u=s;l<u.length;l++){var c=u[l];if(typeof o[c]=="string")e[c]=ai(o[c]);else throw new Error("Serialization of non-string loss is not supported.")}}return e},t.prototype.getMetricIdentifiers=function(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[ai(Bo(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(function(r){return ai(Bo(r))});var e={};for(var i in this.metrics)e[i]=ai(Bo(this.metrics[i]));return e},t.prototype.getTrainingConfig=function(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}},t.prototype.loadTrainingConfig=function(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");var i=ss(e.optimizer_config),r=An(i),a;if(typeof e.loss=="string")a=cr(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(function(l){return cr(l)});else if(e.loss!=null){a={};for(var s in e.loss)a[s]=cr(e.loss[s])}var o;if(Array.isArray(e.metrics))o=e.metrics.map(function(l){return cr(l)});else if(e.metrics!=null){o={};for(var s in e.metrics)o[s]=cr(e.metrics[s])}this.compile({loss:a,metrics:o,optimizer:r})},t.prototype.save=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b;return ve(this,function(w){switch(w.label){case 0:if(typeof e=="string"){if(r=y.io.getSaveHandlers(e),r.length===0)throw new M("Cannot find any save handlers for URL '"+e+"'");if(r.length>1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:x4,generatedBy:"TensorFlow.js tfjs-layers v"+ld,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,Xy(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){Xy(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(d4);y.serialization.registerClass(Ri);var C4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(Ri);y.serialization.registerClass(C4);function R4(n,t){return Se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=ss(e),r=An(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o<l.length;o++)u=l[o],s[u.originalName]=a[u.originalName];r.loadWeights(s),y.dispose(a),c.label=2;case 2:return[2,r]}})})}function E4(n,t){return Se(this,void 0,void 0,function(){var e;return ve(this,function(i){if(t==null&&(t={}),typeof n=="string"){if(e=y.io.getLoadHandlers(n,t),e.length===0)e.push(y.io.browserHTTPRequest(n,t));else if(e.length>1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,O4(n,void 0,t)]})})}function O4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=An(ss(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=D4(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function D4(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var fd=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:xo("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r<a.length;r++){var s=a[r];i.add(s)}return i}return t.prototype.checkShape=function(e){var i=e.inboundNodes[0].outputTensors[0].shape;if(i.some(function(r){return r<0}))throw new M("Negative dimension size caused by adding layer "+(e.name+" with input shape [")+(e.inboundNodes[0].inputTensors[0].shape+"]"))},t.prototype.add=function(e){var i=e instanceof t||e instanceof Ri,r;if(i){if(r=e,r.outputs.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(r.inputs.length!==1)throw new M("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new M("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");var a=Wy({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(i)this.outputs=r.outputs,this.inputs=r.inputs;else{if(e.inboundNodes.length!==1)throw new M("A layer added to a Sequential model must not already be "+("connected somewhere else. LayersModel received layer "+e.name+" ")+("which has "+e.inboundNodes.length+" pre-existing inbound ")+"connections.");if(e.inboundNodes[0].outputTensors.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Fy(this.outputs[0])}this.inboundNodes=[],new Oo({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:ur(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(function(o){return o.shape}),outputShapes:this.outputs[0].shape})}else{var s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1},t.prototype.pop=function(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{var e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}},t.prototype.call=function(e,i){return this.model==null&&this.build(),this.model.call(e,i)},t.prototype.build=function(e){if(je(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new Ri({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0},t.prototype.countParams=function(){return this.built||this.build(),n.prototype.countParams.call(this)},t.prototype.summary=function(e,i,r){r===void 0&&(r=console.log),this.built||this.build(),n.prototype.summary.call(this,e,i,r)},t.prototype.setWeights=function(e){this.model==null&&this.build(),this.model.setWeights(e)},t.prototype.evaluate=function(e,i,r){if(r===void 0&&(r={}),!this.built)throw new Pn("The model needs to be compiled before being used.");return this.model.evaluate(e,i,r)},t.prototype.evaluateDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Pn("The model needs to be compiled before being used.");return[2,this.model.evaluateDataset(e,i)]})})},t.prototype.predict=function(e,i){return i===void 0&&(i={}),this.model==null&&this.build(),this.model.predict(e,i)},t.prototype.predictOnBatch=function(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)},t.prototype.compile=function(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames},Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.model==null?void 0:this.model.optimizer},set:function(e){this.model.optimizer=e},enumerable:!0,configurable:!0}),t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Se(this,void 0,void 0,function(){return ve(this,function(a){if(!this.built)throw new Pn("The model needs to be compiled before being used.");return[2,this.model.fit(e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Pn("The model needs to be compiled before being used.");return[2,this.model.fitDataset(e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){return[2,this.model.trainOnBatch(e,i)]})})},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s,o={};if(i instanceof Array){if(!(i[0].className!=null)||i[0].className==="Merge")throw new M("Legacy serialization format not supported yet.");s=i}else y.util.assert(i.layers!=null,function(){return"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."}),s=i.layers,delete i.layers,o=i;var l=new e(o);if(!(l instanceof t))throw new Te("Sequential.fromConfig called on non-Sequential input: "+l);for(var u=0,c=s;u<c.length;u++){var h=c[u],d=void 0,p=An(h,d,a);a&&p.setFastWeightInitDuringBuild(!0),l.add(p)}return l},Object.defineProperty(t.prototype,"stopTraining",{get:function(){if(this.model==null)throw new M("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining},set:function(e){if(this.model==null)throw new M("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e},enumerable:!0,configurable:!0}),t.prototype.getConfig=function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i],s={};s.className=a.getClassName(),s.config=a.getConfig(),e.push(s)}return{name:this.name,layers:e}},t.className="Sequential",t}(Ri);y.serialization.registerClass(fd);function k4(n){return new Ri(n)}function F4(n){return new fd(n)}function W4(n,t){return t==null&&(t={}),E4(n,t)}function ub(n){return Wy(n)}function U4(n,t){Hy.registerCallbackConstructor(n,t)}var Jt=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),B4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),a3(e,i)},t.className="elu",t}(Jt);y.serialization.registerClass(B4);var z4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.selu(e)},t.className="selu",t}(Jt);y.serialization.registerClass(z4);var P4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="relu",t}(Jt);y.serialization.registerClass(P4);var _4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tidy(function(){return y.minimum(6,y.relu(e))})},t.className="relu6",t}(Jt);y.serialization.registerClass(_4);var M4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return e},t.className="linear",t}(Jt);y.serialization.registerClass(M4);var H4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.sigmoid(e)},t.className="sigmoid",t}(Jt);y.serialization.registerClass(H4);var V4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return o3(e)},t.className="hardSigmoid",t}(Jt);y.serialization.registerClass(V4);var q4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.softplus(e)},t.className="softplus",t}(Jt);y.serialization.registerClass(q4);var G4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return s3(e)},t.className="softsign",t}(Jt);y.serialization.registerClass(G4);var Y4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tanh(e)},t.className="tanh",t}(Jt);y.serialization.registerClass(Y4);var cb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.softmax(e,i)},t.className="softmax",t}(Jt);y.serialization.registerClass(cb);var j4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.logSoftmax(e,i)},t.className="logSoftmax",t}(Jt);y.serialization.registerClass(j4);var K4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),y.tidy(function(){return y.sigmoid(e.mul(i)).mul(e)})},t.className="swish",t}(Jt);y.serialization.registerClass(K4);function Oi(n){return n.getClassName()}function md(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"activation")}function Ei(n){if(n==null){var t={};return t.className="linear",t.config={},md(t)}if(typeof n=="string"){var t={};return t.className=n,t.config={},md(t)}else return n instanceof Jt?n:md(n)}function gd(n){if(n!=null&&typeof n!="object")throw new Error("Argument to L1L2 regularizer's constructor is expected to be an "+("object, but received: "+n))}var hb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(y.serialization.Serializable),Po=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return gd(e),i.l1=e==null||e.l1==null?.01:e.l1,i.l2=e==null||e.l2==null?.01:e.l2,i.hasL1=i.l1!==0,i.hasL2=i.l2!==0,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=y.zeros([1]);return i.hasL1&&(r=y.add(r,y.sum(y.mul(i.l1,y.abs(e))))),i.hasL2&&(r=y.add(r,y.sum(y.mul(i.l2,ns(e))))),r.asScalar()})},t.prototype.getConfig=function(){return{l1:this.l1,l2:this.l2}},t.fromConfig=function(e,i){return new e({l1:i.l1,l2:i.l2})},t.className="L1L2",t}(hb);y.serialization.registerClass(Po);function $4(n){return gd(n),new Po({l1:n!=null?n.l1:null,l2:0})}function X4(n){return gd(n),new Po({l2:n!=null?n.l2:null,l1:0})}var db={l1l2:"L1L2"};function Ke(n){return Bh(n)}function pb(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"regularizer")}function nt(n){if(n==null)return null;if(typeof n=="string"){var t=n in db?db[n]:n,e={className:t,config:{}};return pb(e)}else return n instanceof hb?n:pb(n)}var fb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null&&(i.maxValue=e.maxValue),i}return t.prototype.call=function(e,i){e=xe(e);var r=y.relu(e);return this.maxValue!=null&&(r=y.clipByValue(r,0,this.maxValue)),r},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={maxValue:this.maxValue},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ReLU",t}(De);y.serialization.registerClass(fb);var mb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_ALPHA=.3,e==null&&(e={}),i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.leakyRelu(r,this.alpha)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LeakyReLU",t}(De);y.serialization.registerClass(mb);var gb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),i.supportsMasking=!0,i.alphaInitializer=tt(e.alphaInitializer||i.DEFAULT_ALPHA_INITIALIZER),i.alphaRegularizer=nt(e.alphaRegularizer),i.alphaConstraint=wt(e.alphaConstraint),e.sharedAxes==null)i.sharedAxes=null;else if(Array.isArray(e.sharedAxes))i.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")i.sharedAxes=[e.sharedAxes];else throw new M("Expected sharedAxes to be a number or an array of numbers, "+("but got "+e.sharedAxes));return i}return t.prototype.build=function(e){e=je(e);var i=e.slice(1);if(this.sharedAxes!=null)for(var r=0,a=this.sharedAxes;r<a.length;r++){var s=a[r];i[s-1]=1}this.alpha=this.addWeight("alpha",i,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);var o={};if(this.sharedAxes!=null)for(var s=1;s<e.length;++s)o[s]=e[s];this.inputSpec=[new Nt({ndim:e.length,axes:o})],this.built=!0},t.prototype.call=function(e,i){return e=xe(e),y.prelu(e,this.alpha.read())},t.prototype.getConfig=function(){var e={alphaInitializer:st(this.alphaInitializer),alphaRegularizer:Ke(this.alphaRegularizer),alphaConstraint:bt(this.alphaConstraint),sharedAxes:this.sharedAxes},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="PReLU",t}(De);y.serialization.registerClass(gb);var vb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==i.DEFAULT_ALPHA)throw new Te("Non-default alpha value ("+e.alpha+") is not supported by the ELU layer yet.");return i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.elu(r)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ELU",t}(De);y.serialization.registerClass(vb);var yb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_THETA=1,e==null&&(e={}),i.theta=e.theta==null?i.DEFAULT_THETA:e.theta,i}return t.prototype.call=function(e,i){var r=xe(e);return r.mul(es(r.greater(this.theta),"float32"))},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={theta:this.theta},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ThresholdedReLU",t}(De);y.serialization.registerClass(yb);var bb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_AXIS=1,e==null&&(e={}),i.softmax=new cb().apply,i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i}return t.prototype.call=function(e,i){var r=xe(e);return this.softmax(r,this.axis)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Softmax",t}(De);y.serialization.registerClass(bb);function sa(n,t,e){if(typeof n=="number")return ur(n,t);if(n.length!==t)throw new M("The "+e+" argument must be an integer or tuple of "+t+" integers."+(" Received: "+n.length+" elements."));for(var i=0;i<t;++i){var r=n[i];if(!t3(r))throw new M("The "+e+" argument must be an integer or tuple of "+t+(" integers. Received: "+JSON.stringify(n)+" including a")+(" non-integer number "+r))}return n}function Tn(n,t,e,i,r){if(r===void 0&&(r=1),n==null)return n;var a=t+(t-1)*(r-1),s;return e==="same"?s=n:s=n-a+1,Math.floor((s+i-1)/i)}function _o(n,t,e,i){if(n==null)return null;if(i==="valid")n=n*t+xi([e-t,0]);else if(i==="same")n=n*t;else throw new M("Unsupport padding mode: "+i+".");return n}function vd(n,t){return y.tidy(function(){return ht(t),t==="channelsFirst"?y.transpose(n,[0,2,3,1]):n})}function wb(n,t){return y.tidy(function(){return ht(t),t==="channelsFirst"?y.transpose(n,[0,2,3,4,1]):n})}function J4(n,t,e,i,r,a,s){return i===void 0&&(i=1),r===void 0&&(r="valid"),s===void 0&&(s=1),y.tidy(function(){if(a==null&&(a=Sn()),ht(a),n.shape.length!==3)throw new M("The input of a conv1dWithBias operation should be 3, but is "+(n.shape.length+" instead."));if(t.shape.length!==3)throw new M("The kernel for a conv1dWithBias operation should be 3, but is "+(t.shape.length+" instead"));if(e!=null&&e.shape.length!==1)throw new M("The bias for a conv1dWithBias operation should be 1, but is "+(t.shape.length+" instead"));if(a==="channelsFirst"&&(n=y.transpose(n,[0,2,1])),r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");var o=y.conv1d(n,t,i,r==="same"?"same":"valid","NWC",s);return e!=null&&(o=Hn(o,e)),o})}function Sb(n,t,e,i,r,a,s,o){return i===void 0&&(i=[1,1]),r===void 0&&(r="valid"),o===void 0&&(o=null),y.tidy(function(){if(a==null&&(a=Sn()),ht(a),n.rank!==3&&n.rank!==4)throw new M("conv2dWithBiasActivation expects input to be of rank 3 or 4, "+("but received "+n.rank+"."));if(t.rank!==3&&t.rank!==4)throw new M("conv2dWithBiasActivation expects kernel to be of rank 3 or 4, "+("but received "+n.rank+"."));var l=vd(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=y.fused.conv2d({x:l,filter:t,strides:i,pad:r==="same"?"same":"valid",dilations:s,dataFormat:"NHWC",bias:e,activation:o}),a==="channelsFirst"&&(l=y.transpose(l,[0,3,1,2])),l})}function Z4(n,t,e,i,r,a,s){return i===void 0&&(i=[1,1,1]),r===void 0&&(r="valid"),y.tidy(function(){if(a==null&&(a=Sn()),ht(a),n.rank!==4&&n.rank!==5)throw new M("conv3dWithBias expects input to be of rank 4 or 5, but received "+(n.rank+"."));if(t.rank!==4&&t.rank!==5)throw new M("conv3dWithBias expects kernel to be of rank 4 or 5, but received "+(n.rank+"."));var o=wb(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=y.conv3d(o,t,i,r==="same"?"same":"valid","NDHWC",s),e!=null&&(o=Hn(o,e)),a==="channelsFirst"&&(o=y.transpose(o,[0,4,1,2,3])),o})}var Lb=function(n){Q(t,n);function t(e,i){var r=n.call(this,i)||this;if(r.bias=null,r.DEFAULT_KERNEL_INITIALIZER="glorotNormal",r.DEFAULT_BIAS_INITIALIZER="zeros",t.verifyArgs(i),r.rank=e,Tt(r.rank,"rank"),r.rank!==1&&r.rank!==2&&r.rank!==3)throw new Te("Convolution layer for rank other than 1, 2, or 3 ("+r.rank+") is not implemented yet.");if(r.kernelSize=sa(i.kernelSize,e,"kernelSize"),r.strides=sa(i.strides==null?1:i.strides,e,"strides"),r.padding=i.padding==null?"valid":i.padding,sn(r.padding),r.dataFormat=i.dataFormat==null?"channelsLast":i.dataFormat,ht(r.dataFormat),r.activation=Ei(i.activation),r.useBias=i.useBias==null?!0:i.useBias,r.biasInitializer=tt(i.biasInitializer||r.DEFAULT_BIAS_INITIALIZER),r.biasConstraint=wt(i.biasConstraint),r.biasRegularizer=nt(i.biasRegularizer),r.activityRegularizer=nt(i.activityRegularizer),r.dilationRate=sa(i.dilationRate==null?1:i.dilationRate,e,"dilationRate"),r.rank===1&&Array.isArray(r.dilationRate)&&r.dilationRate.length!==1)throw new M("dilationRate must be a number or an array of a single number for 1D convolution, but received "+(""+JSON.stringify(r.dilationRate)));if(r.rank===2){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==2)throw new M("dilationRate must be a number or array of two numbers for 2D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}else if(r.rank===3){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==3)throw new M("dilationRate must be a number or array of three numbers for 3D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}return r}return t.verifyArgs=function(e){if(_n("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!Ph(e.kernelSize,"number",1,3))throw new M("BaseConv expects config.kernelSize to be number or number[] with "+("length 1, 2, or 3, but received "+JSON.stringify(e.kernelSize)+"."))},t.prototype.getConfig=function(){var e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:Oi(this.activation),useBias:this.useBias,biasInitializer:st(this.biasInitializer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),biasConstraint:bt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Mo=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;return r.kernel=null,t.verifyArgs(i),r.filters=i.filters,Tt(r.filters,"filters"),r.kernelInitializer=tt(i.kernelInitializer||r.DEFAULT_KERNEL_INITIALIZER),r.kernelConstraint=wt(i.kernelConstraint),r.kernelRegularizer=nt(i.kernelRegularizer),r}return t.prototype.build=function(e){var i;e=je(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=this.kernelSize.concat([a,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:(i={},i[r]=a,i)}],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a,s=r.bias==null?null:r.bias.read(),o=sy(r.activation.getClassName());if(o!=null&&r.rank===2)a=Sb(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate,o);else{if(r.rank===1)a=J4(e,r.kernel.read(),s,r.strides[0],r.padding,r.dataFormat,r.dilationRate[0]);else if(r.rank===2)a=Sb(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else if(r.rank===3)a=Z4(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else throw new Te("convolutions greater than 3D are not implemented yet.");r.activation!=null&&(a=r.activation.apply(a))}return a})},t.prototype.computeOutputShape=function(e){e=je(e);for(var i=[],r=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2),a=0;a<r.length;++a){var s=Tn(r[a],this.kernelSize[a],this.padding,this.strides[a],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[a]);i.push(s)}var o=[e[0]];return this.dataFormat==="channelsLast"?(o=o.concat(i),o.push(this.filters)):(o.push(this.filters),o=o.concat(i)),o},t.prototype.getConfig=function(){var e={filters:this.filters,kernelInitializer:st(this.kernelInitializer),kernelRegularizer:Ke(this.kernelRegularizer),kernelConstraint:bt(this.kernelConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.verifyArgs=function(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new M("Convolution layer expected config.filters to be a 'number' > 0 "+("but got "+JSON.stringify(e.filters)))},t}(Lb),yd=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!Ph(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(Mo);y.serialization.registerClass(yd);var Ib=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(Mo);y.serialization.registerClass(Ib);var Ab=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Nt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=je(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Nt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=_o(c,f,d,r.padding),v=_o(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Hn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=je(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=_o(i[a],u,o,this.padding),i[s]=_o(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(yd);y.serialization.registerClass(Ab);var Q4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=nt(i.depthwiseRegularizer),r.depthwiseConstraint=wt(i.depthwiseConstraint),r.pointwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=nt(i.pointwiseRegularizer),r.pointwiseConstraint=wt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=je(e),e.length<this.rank+2)throw new M("Inputs to SeparableConv"+this.rank+"D should have rank "+(this.rank+2+", but received input shape: ")+(""+JSON.stringify(e)));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null||e[r]<0)throw new M("The channel dimension of the inputs should be defined, "+("but found "+JSON.stringify(e[r])));for(var a=e[r],s=this.kernelSize.concat([a,this.depthMultiplier]),o=[],l=0;l<this.rank;++l)o.push(1);o.push(a*this.depthMultiplier,this.filters);var u=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,u,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",o,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,u,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,u,this.biasConstraint):this.bias=null,this.inputSpec=[new Nt({ndim:this.rank+2,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a;if(r.rank===1)throw new Te("1D separable convolution is not implemented yet.");return r.rank===2&&(r.dataFormat==="channelsFirst"&&(e=y.transpose(e,[0,2,3,1])),a=y.separableConv2d(e,r.depthwiseKernel.read(),r.pointwiseKernel.read(),r.strides,r.padding,r.dilationRate,"NHWC")),r.useBias&&(a=Hn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),r.dataFormat==="channelsFirst"&&(a=y.transpose(a,[0,3,1,2])),a})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=st(this.depthwiseInitializer),e.pointwiseInitializer=st(this.pointwiseInitializer),e.depthwiseRegularizer=Ke(this.depthwiseRegularizer),e.pointwiseRegularizer=Ke(this.pointwiseRegularizer),e.depthwiseConstraint=bt(this.depthwiseConstraint),e.pointwiseConstraint=bt(this.pointwiseConstraint),e},t.className="SeparableConv",t}(Mo),Tb=function(n){Q(t,n);function t(e){return n.call(this,2,e)||this}return t.className="SeparableConv2D",t}(Q4);y.serialization.registerClass(Tb);var Nb=function(n){Q(t,n);function t(e){var i=n.call(this,1,e)||this;return t.verifyArgs(e),i.inputSpec=[{ndim:3}],i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.dataFormat,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!Ph(e.kernelSize,"number",1,1))throw new M("Conv1D expects config.kernelSize to be number or number[] with "+("length 1, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv1D",t}(Mo);y.serialization.registerClass(Nb);var xb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return typeof e.cropping=="number"?i.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?i.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:i.cropping=e.cropping,i.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,i.inputSpec=[{ndim:4}],i}return t.prototype.computeOutputShape=function(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=xe(e),r.dataFormat==="channelsLast"){var a=Ao(e,r.cropping[0][0],e.shape[1]-r.cropping[0][0]-r.cropping[0][1],2);return Ao(a,r.cropping[1][0],e.shape[2]-r.cropping[1][1]-r.cropping[1][0],3)}else{var a=Ao(e,r.cropping[0][0],e.shape[2]-r.cropping[0][0]-r.cropping[0][1],3);return Ao(a,r.cropping[1][0],e.shape[3]-r.cropping[1][1]-r.cropping[1][0],4)}})},t.prototype.getConfig=function(){var e={cropping:this.cropping,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Cropping2D",t}(De);y.serialization.registerClass(xb);var Cb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_SIZE=[2,2],i.inputSpec=[{ndim:4}],i.size=e.size==null?i.DEFAULT_SIZE:e.size,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,i}return t.prototype.computeOutputShape=function(e){if(this.dataFormat==="channelsFirst"){var i=e[2]==null?null:this.size[0]*e[2],r=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],i,r]}else{var i=e[1]==null?null:this.size[0]*e[1],r=e[2]==null?null:this.size[1]*e[2];return[e[0],i,r,e[3]]}},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e),s=a.shape;if(r.dataFormat==="channelsFirst"){a=y.transpose(a,[0,2,3,1]);var o=r.size[0]*s[2],l=r.size[1]*s[3],u=a.resizeNearestNeighbor([o,l]);return y.transpose(u,[0,3,1,2])}else{var o=r.size[0]*s[1],l=r.size[1]*s[2];return a.resizeNearestNeighbor([o,l])}})},t.prototype.getConfig=function(){var e={size:this.size,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="UpSampling2D",t}(De);y.serialization.registerClass(Cb);function eF(n,t,e,i,r,a){return e===void 0&&(e=[1,1]),i===void 0&&(i="valid"),y.tidy(function(){r==null&&(r=Sn()),ht(r);var s=vd(n,r);if(n.rank!==4)throw new M("Input for depthwiseConv2d is required to be 4-D, but is instead "+(n.rank+"-D"));if(t.rank!==4)throw new M("depthwiseKernel is required to be 4-D, but is instead "+(t.rank+"-D"));return s=y.depthwiseConv2d(s,t,e,i==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}var Rb=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return i.depthwiseKernel=null,i.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,i.depthwiseInitializer=tt(e.depthwiseInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.depthwiseConstraint=wt(e.depthwiseConstraint),i.depthwiseRegularizer=nt(e.depthwiseRegularizer),i}return t.prototype.build=function(e){if(e=je(e),e.length<4)throw new M("Inputs to DepthwiseConv2D should have rank 4. "+("Received input shape: "+JSON.stringify(e)+"."));var i=this.dataFormat==="channelsFirst"?1:3;if(e[i]==null||e[i]<0)throw new M("The channel dimension of the inputs to DepthwiseConv2D should "+("be defined, but is not ("+e[i]+")."));var r=e[i],a=[this.kernelSize[0],this.kernelSize[1],r,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[r*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=eF(e,r.depthwiseKernel.read(),r.strides,r.padding,r.dataFormat,null);return r.useBias&&(a=Hn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),a})},t.prototype.computeOutputShape=function(e){e=je(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,s=Tn(i,this.kernelSize[0],this.padding,this.strides[0]),o=Tn(r,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,s,o]:[e[0],s,o,a]},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=st(this.depthwiseInitializer),e.depthwiseRegularizer=Ke(this.depthwiseRegularizer),e.depthwiseConstraint=bt(this.depthwiseRegularizer),e},t.className="DepthwiseConv2D",t}(Lb);y.serialization.registerClass(Rb);function Ob(n,t,e,i){if(Array.isArray(n)){if(t!=null||e!=null)throw new M("When inputs is an array, neither initialState or constants should be provided");i!=null&&(e=n.slice(n.length-i,n.length),n=n.slice(0,n.length-i)),n.length>1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Eb(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(Ln(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],N=y.tidy(function(){return n(L,d)});if(r==null)h=N[0],d=N[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=N[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return N[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v<p;++v)g(v);var b;if(o){var w=1;b=y.stack(c,w)}return[h,b,d]})}var Di=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r;if(e.cell==null)throw new M("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?r=new bd({cells:e.cell}):r=e.cell,r.stateSize==null)throw new M("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");return i.cell=r,i.returnSequences=e.returnSequences==null?!1:e.returnSequences,i.returnState=e.returnState==null?!1:e.returnState,i.goBackwards=e.goBackwards==null?!1:e.goBackwards,i._stateful=e.stateful==null?!1:e.stateful,i.unroll=e.unroll==null?!1:e.unroll,i.supportsMasking=!0,i.inputSpec=[new Nt({ndim:3})],i.stateSpec=null,i.states_=null,i.numConstants=null,i.keptStates=[],i}return t.prototype.getStates=function(){if(this.states_==null){var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return Ln(0,e).map(function(i){return null})}else return this.states_},t.prototype.setStates=function(e){this.states_=e},t.prototype.computeOutputShape=function(e){Zh(e)&&(e=e[0]),e=e;var i=this.cell.stateSize;Array.isArray(i)||(i=[i]);var r=i[0],a;if(this.returnSequences?a=[e[0],e[1],r]:a=[e[0],r],this.returnState){for(var s=[],o=0,l=i;o<l.length;o++){var u=l[o];s.push([e[0],u])}return[a].concat(s)}else return a},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){Array.isArray(i)&&(i=i[0]);var a=r.returnSequences?i:null;if(r.returnState){var s=r.states.map(function(o){return null});return[a].concat(s)}else return a})},Object.defineProperty(t.prototype,"states",{get:function(){if(this.states_==null){for(var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,i=[],r=0;r<e;++r)i.push(null);return i}else return this.states_},set:function(e){this.states_=e},enumerable:!0,configurable:!0}),t.prototype.build=function(e){if(this.numConstants!=null)throw new Te("Constants support is not implemented in RNN yet.");Zh(e)&&(e=e[0]),e=e;var i=this.stateful?e[0]:null,r=e.slice(2);this.inputSpec[0]=new Nt({shape:[i,null].concat(r)});var a=[e[0]].concat(e.slice(2));this.cell.build(a);var s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!y.util.arraysEqual(this.stateSpec.map(function(o){return o.shape[o.shape.length-1]}),s))throw new M("An initialState was passed that is not compatible with "+("cell.stateSize. Received stateSpec="+this.stateSpec+"; ")+("However cell.stateSize is "+this.cell.stateSize))}else this.stateSpec=s.map(function(o){return new Nt({shape:[null,o]})});this.stateful&&this.resetStates()},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new Ai("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape[0];if(a==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.states_==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_=[y.zeros([a,r.cell.stateSize])];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_[0]=y.zeros([a,r.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i===!0?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var s=0;s<r.states_.length;++s){var o=e[s],l=Array.isArray(r.cell.stateSize)?r.cell.stateSize[s]:r.cell.stateSize,u=[a,l];if(!y.util.arraysEqual(o.shape,u))throw new M("State "+s+" is incompatible with layer "+r.name+": "+("expected shape="+u+", received shape="+o.shape));r.states_[s]=o}}r.states_=r.states_.map(function(c){return y.keep(c.clone())})})},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Ob(e,r,a,this.numConstants);e=s.inputs,r=s.initialState,a=s.constants;var o=[],l=[];if(r!=null){i.initialState=r,o=o.concat(r),this.stateSpec=[];for(var u=0,c=r;u<c.length;u++){var h=c[u];this.stateSpec.push(new Nt({shape:h.shape}))}l=l.concat(this.stateSpec)}a!=null&&(i.constants=a,o=o.concat(a),this.numConstants=a.length);var d=o[0]instanceof In;if(d){var p=[e].concat(o),f=this.inputSpec.concat(l),m=this.inputSpec;this.inputSpec=f;var g=n.prototype.apply.call(this,p,i);return this.inputSpec=m,g}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;e=xe(e),o==null&&(r.stateful?o=r.states_:o=r.getInitialState(e));var l=Array.isArray(r.cell.stateSize)?r.cell.stateSize.length:1;if(o.length!==l)throw new M("RNN Layer has "+l+" state(s) but was passed "+(o.length+" initial state(s)."));r.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");var u={training:s},c=function(g,v){var b=r.cell.call([g].concat(v),u);return[b[0],b.slice(1)]},h=Eb(c,e,o,r.goBackwards,a,null,r.unroll,r.returnSequences),d=h[0],p=h[1],f=h[2];r.stateful&&r.resetStates(f,s);var m=r.returnSequences?p:d;return r.returnState?[m].concat(f):m})},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=y.zeros(e.shape);return r=y.sum(r,[1,2]),r=ts(r),Array.isArray(i.cell.stateSize)?i.cell.stateSize.map(function(a){return a>1?Mh(r,[1,a]):r}):i.cell.stateSize>1?[Mh(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),$t({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=An(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Di);var oa=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),wd=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=wt(e.kernelConstraint),i.recurrentConstraint=wt(e.recurrentConstraint),i.biasConstraint=wt(e.biasConstraint),i.dropout=na([1,xi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=na([1,xi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=je(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(e)},rate:r.dropout,training:s})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(a)},rate:r.recurrentDropout,training:s}));var o,l=r.dropoutMask,u=r.recurrentDropoutMask;l!=null?o=Mn(y.mul(e,l),r.kernel.read()):o=Mn(e,r.kernel.read()),r.bias!=null&&(o=Hn(o,r.bias.read())),u!=null&&(a=y.mul(a,u));var c=y.add(o,Mn(a,r.recurrentKernel.read()));return r.activation!=null&&(c=r.activation.apply(c)),[c,c]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:Oi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:Ke(this.kernelRegularizer),recurrentRegularizer:Ke(this.recurrentRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),recurrentConstraint:bt(this.recurrentConstraint),biasConstraint:bt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return $t({},e,i)},t.className="SimpleRNNCell",t}(oa);y.serialization.registerClass(wd);var Db=function(n){Q(t,n);function t(e){var i=this;return e.cell=new wd(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return new e(i)},t.className="SimpleRNN",t}(Di);y.serialization.registerClass(Db);var Sd=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new M("GRUCell does not support reset_after parameter set to true.");return i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=Ei(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=wt(e.kernelConstraint),i.recurrentConstraint=wt(e.recurrentConstraint),i.biasConstraint=wt(e.biasConstraint),i.dropout=na([1,xi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=na([1,xi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=je(e);var i=e[e.length-1];this.kernel=this.addWeight("kernel",[i,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("GRUCell expects 2 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training==null?!1:i.training,s=e[1];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:3})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:3}));var o=r.dropoutMask,l=r.recurrentDropoutMask,u,c,h;0<r.dropout&&r.dropout<1&&(e=y.mul(e,o[0]));var d=Mn(e,r.kernel.read());r.useBias&&(d=Hn(d,r.bias.read())),0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,l[0]));var p=r.recurrentKernel.read(),f=y.split(p,[2*r.units,r.units],p.rank-1),m=f[0],g=f[1],v=Mn(s,m),b=y.split(d,3,d.rank-1),w=b[0],S=b[1],L=b[2],N=y.split(v,2,v.rank-1),C=N[0],R=N[1];u=r.recurrentActivation.apply(y.add(w,C)),c=r.recurrentActivation.apply(y.add(S,R));var D=Mn(y.mul(c,s),g);h=r.activation.apply(y.add(L,D));var k=y.add(y.mul(u,s),y.mul(y.add(1,y.neg(u)),h));return[k,k]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:Oi(this.activation),recurrentActivation:Oi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:Ke(this.kernelRegularizer),recurrentRegularizer:Ke(this.recurrentRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),recurrentConstraint:bt(this.recurrentConstraint),biasConstraint:bt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return $t({},e,i)},t.className="GRUCell",t}(oa);y.serialization.registerClass(Sd);var kb=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Sd(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="GRU",t}(Di);y.serialization.registerClass(kb);var Ho=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=Ei(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.unitForgetBias=e.unitForgetBias,i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=wt(e.kernelConstraint),i.recurrentConstraint=wt(e.recurrentConstraint),i.biasConstraint=wt(e.biasConstraint),i.dropout=na([1,xi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=na([1,xi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=[i.units,i.units],i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){var i;e=je(e);var r=e[e.length-1];this.kernel=this.addWeight("kernel",[r,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);var a;if(this.useBias){if(this.unitForgetBias){var s=this.biasInitializer,o=this.units;a=new(i=function(l){Q(u,l);function u(){return l!==null&&l.apply(this,arguments)||this}return u.prototype.apply=function(c,h){var d=s.apply([o]),p=new Gh().apply([o]),f=s.apply([o*2]);return by(by(d,p),f)},u}(fn),i.className="CustomInit",i)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training;if(e=e,e.length!==3)throw new M("LSTMCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var s=e[1],o=e[2];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:4})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:4}));var l=r.dropoutMask,u=r.recurrentDropoutMask,c,h,d,p;0<r.dropout&&r.dropout<1&&(e=y.mul(e,l[0]));var f=Mn(e,r.kernel.read());0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,u[0])),f=y.add(f,Mn(s,r.recurrentKernel.read())),r.useBias&&(f=Hn(f,r.bias.read()));var m=y.split(f,4,f.rank-1),g=m[0],v=m[1],b=m[2],w=m[3];c=r.recurrentActivation.apply(g),h=r.recurrentActivation.apply(v),d=y.add(y.mul(h,o),y.mul(c,r.activation.apply(b))),p=r.recurrentActivation.apply(w);var S=y.mul(p,r.activation.apply(d));return[S,S,d]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:Oi(this.activation),recurrentActivation:Oi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:Ke(this.kernelRegularizer),recurrentRegularizer:Ke(this.recurrentRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),recurrentConstraint:bt(this.recurrentConstraint),biasConstraint:bt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return $t({},e,i)},t.className="LSTMCell",t}(oa);y.serialization.registerClass(Ho);var Fb=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Ho(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="LSTM",t}(Di);y.serialization.registerClass(Fb);var bd=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.cells=e.cells,i}return Object.defineProperty(t.prototype,"stateSize",{get:function(){for(var e=[],i=0,r=this.cells.slice().reverse();i<r.length;i++){var a=r[i];Array.isArray(a.stateSize)?e.push.apply(e,a.stateSize):e.push(a.stateSize)}return e},enumerable:!0,configurable:!0}),t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=e;for(var a=e.slice(1),s=[],o=0,l=r.cells.slice().reverse();o<l.length;o++){var u=l[o];Array.isArray(u.stateSize)?s.push(a.splice(0,u.stateSize.length)):s.push(a.splice(0,1))}s.reverse();for(var c=[],h,d=0;d<r.cells.length;++d){var u=r.cells[d];a=s[d],d===0?h=[e[0]].concat(a):h=[h[0]].concat(a),h=u.call(h,i),c.push(h.slice(1))}a=[];for(var p=0,f=c.slice().reverse();p<f.length;p++){var m=f[p];a.push.apply(a,m)}return[h[0]].concat(a)})},t.prototype.build=function(e){Zh(e)&&(e=e[0]),e=e;var i;this.cells.forEach(function(r,a){hr("RNNCell_"+a,function(){r.build(e),Array.isArray(r.stateSize)?i=r.stateSize[0]:i=r.stateSize,e=[e[0],i]})}),this.built=!0},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=function(s){return{className:s.getClassName(),config:s.getConfig()}},r=this.cells.map(i),a={cells:r};return $t({},e,a)},t.fromConfig=function(e,i,r){r===void 0&&(r={});for(var a=[],s=0,o=i.cells;s<o.length;s++){var l=o[s];a.push(An(l,r))}return new e({cells:a})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(!this.trainable)return[];for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.cells;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.weights)}return Qh(e)},t.prototype.setWeights=function(e){for(var i=[],r=0,a=this.cells;r<a.length;r++)for(var s=a[r],o=s.weights.length,l=e.splice(o),u=0;u<s.weights.length;++u)i.push([s.weights[u],l[u]]);ed(i)},t.className="StackedRNNCells",t}(oa);y.serialization.registerClass(bd);function ki(n){var t=n.ones,e=n.rate,i=n.training,r=i===void 0?!1:i,a=n.count,s=a===void 0?1:a,o=function(){return Sy(t(),e)},l=function(){return is(o,t,r)};if(!s||s<=1)return y.keep(l().clone());var u=Array(s).fill(void 0).map(l);return u.map(function(c){return y.keep(c.clone())})}var pV=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(oa),tF=function(n){Q(t,n);function t(e){var i=this;if(e.unroll)throw new Te("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Te("It is not possible at the moment to stack convolutional cells.");return i=n.call(this,e)||this,i.inputSpec=[new Nt({ndim:5})],i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null),i&&i.constants)throw new M("ConvRNN2D cell does not support constants");var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.prototype.computeOutputShape=function(e){var i=this.computeSingleOutputShape(e);return this.returnSequences||(i=[i[0]].concat(i.slice(2))),this.returnState&&(i=[i].concat(Array(2).fill([e[0]].concat(i.slice(-3))))),i},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=i.cell.stateSize,a=e.shape,s=i.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=y.zeros(o);return Array.isArray(r)?Array(r.length).fill(l):[l]})},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new Ai("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape,s=r.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=a[0];if(l==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.getStates()==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_=[y.zeros(o)];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_[0]=y.zeros(o);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var u=0;u<r.states_.length;++u){var c=e[u],h=o;if(!y.util.arraysEqual(c.shape,h))throw new M("State "+u+" is incompatible with layer "+r.name+": "+("expected shape="+h+", received shape="+c.shape));r.states_[u]=c}}r.states_=r.states_.map(function(d){return y.keep(d.clone())})})},t.prototype.computeSingleOutputShape=function(e){var i=this.cell,r=i.dataFormat,a=i.filters,s=i.kernelSize,o=i.padding,l=i.strides,u=i.dilationRate,c=r==="channelsFirst",h=e[c?3:2],d=e[c?4:3],p=Tn(h,s[0],o,l[0],u[0]),f=Tn(d,s[1],o,l[1],u[1]),m=e.slice(0,2).concat(c?[a,p,f]:[p,f,a]);return m},t.className="ConvRNN2D",t}(Di),Ld=function(n){Q(t,n);function t(e){var i=this,r=e.filters,a=e.kernelSize,s=e.strides,o=e.padding,l=e.dataFormat,u=e.dilationRate;return i=n.call(this,$t({},e,{units:r}))||this,i.filters=r,Tt(i.filters,"filters"),i.kernelSize=sa(a,2,"kernelSize"),i.kernelSize.forEach(function(c){return Tt(c,"kernelSize")}),i.strides=sa(s||1,2,"strides"),i.strides.forEach(function(c){return Tt(c,"strides")}),i.padding=o||"valid",sn(i.padding),i.dataFormat=l||"channelsLast",ht(i.dataFormat),i.dilationRate=sa(u||1,2,"dilationRate"),i.dilationRate.forEach(function(c){return Tt(c,"dilationRate")}),i}return t.prototype.build=function(e){var i;e=je(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=4,o=this.kernelSize.concat([a,this.filters*s]);this.kernel=this.addWeight("kernel",o,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);var l=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",l,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){var u=void 0;if(this.unitForgetBias){var c=this.biasInitializer,h=this.filters;u=new(i=function(d){Q(p,d);function p(){return d!==null&&d.apply(this,arguments)||this}return p.prototype.apply=function(f,m){var g=c.apply([h]),v=y.ones([h]),b=c.apply([h*2]);return Vh([g,v,b])},p}(fn),i.className="CustomInit",i)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e.length!==3)throw new M("ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training||!1,s=e[0],o=e[1],l=e[2],u=4;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(s)},rate:r.dropout,training:a,count:u}));var c=r.dropoutMask,h=function(te,re,le){return!re||!re[le]?te:y.mul(re[le],te)},d=h(s,c,0),p=h(s,c,1),f=h(s,c,2),m=h(s,c,3);0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(o)},rate:r.recurrentDropout,training:a,count:u}));var g=r.recurrentDropoutMask,v=h(o,g,0),b=h(o,g,1),w=h(o,g,2),S=h(o,g,3),L=3,N=y.split(r.kernel.read(),u,L),C=N[0],R=N[1],D=N[2],k=N[3],W=r.useBias?y.split(r.bias.read(),u):[null,null,null,null],F=W[0],P=W[1],H=W[2],_=W[3];d=r.inputConv(d,C,F,r.padding),p=r.inputConv(p,R,P,r.padding),f=r.inputConv(f,D,H,r.padding),m=r.inputConv(m,k,_,r.padding);var j=y.split(r.recurrentKernel.read(),u,L),K=j[0],q=j[1],G=j[2],Z=j[3];v=r.recurrentConv(v,K),b=r.recurrentConv(b,q),w=r.recurrentConv(w,G),S=r.recurrentConv(S,Z);var X=r.recurrentActivation.apply(y.add(d,v)),ee=r.recurrentActivation.apply(y.add(p,b)),ne=y.add(y.mul(ee,l),y.mul(X,r.activation.apply(y.add(f,w)))),ie=y.mul(r.recurrentActivation.apply(y.add(m,S)),r.activation.apply(ne));return[ie,ie,ne]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=e.units,r=Pk(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return $t({},r,a)},t.prototype.inputConv=function(e,i,r,a){var s=y.conv2d(e,i,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return r?Hn(s,r,this.dataFormat):s},t.prototype.recurrentConv=function(e,i){var r=1;return y.conv2d(e,i,r,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")},t.className="ConvLSTM2DCell",t}(Ho);y.serialization.registerClass(Ld);var Wb=function(n){Q(t,n);function t(e){var i=this,r=new Ld(e);return i=n.call(this,$t({},e,{cell:r}))||this,i}return t.fromConfig=function(e,i){return new e(i)},t.className="ConvLSTM2D",t}(tF);y.serialization.registerClass(Wb);var Id=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.rate=Math.max(Math.min(e.rate,1),0),i.noiseShape=e.noiseShape,i.seed=e.seed,i.supportsMasking=!0,i}return t.prototype.getNoiseShape=function(e){if(this.noiseShape==null)return this.noiseShape;for(var i=e.shape,r=[],a=0;a<this.noiseShape.length;++a)r.push(this.noiseShape[a]==null?i[a]:this.noiseShape[a]);return r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(0<r.rate&&r.rate<1){var s=i.training==null?!1:i.training,o=r.getNoiseShape(a),l=is(function(){return Sy(a,r.rate,o,r.seed)},function(){return a},s);return l}return e})},t.prototype.getConfig=function(){var e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.dispose=function(){return n.prototype.dispose.call(this)},t.className="Dropout",t}(De);y.serialization.registerClass(Id);var Ub=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[{ndim:3}],i}return t.prototype.getNoiseShape=function(e){var i=e.shape;return[i[0],1,i[2]]},t.className="SpatialDropout1D",t}(Id);y.serialization.registerClass(Ub);var Bb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.activation=null,i.useBias=!0,i.kernel=null,i.bias=null,i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){var r=null;e.batchSize!=null&&(r=e.batchSize),i.batchInputShape=[r,e.inputDim]}return i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation),e.useBias!=null&&(i.useBias=e.useBias),i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelConstraint=wt(e.kernelConstraint),i.biasConstraint=wt(e.biasConstraint),i.kernelRegularizer=nt(e.kernelRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.supportsMasking=!0,i.inputSpec=[{minNDim:2}],i}return t.prototype.build=function(e){var i;e=je(e);var r=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[r,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:(i={},i[-1]=r,i)}],this.built=!0},t.prototype.computeOutputShape=function(e){e=je(e);var i=e.slice();return i[i.length-1]=this.units,i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=sy(r.activation.getClassName()),o;return s!=null?o=Mn(a,r.kernel.read(),s,r.bias?r.bias.read():null):(o=Mn(a,r.kernel.read()),r.bias!=null&&(o=Hn(o,r.bias.read())),r.activation!=null&&(o=r.activation.apply(o))),o})},t.prototype.getConfig=function(){var e={units:this.units,activation:Oi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:Ke(this.kernelRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),biasConstraint:bt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dense",t}(De);y.serialization.registerClass(Bb);var zb=function(n){Q(t,n);function t(e){var i=this;return e=e||{},i=n.call(this,e)||this,i.inputSpec=[{minNDim:3}],i.dataFormat=e.dataFormat,i}return t.prototype.computeOutputShape=function(e){e=je(e);for(var i=0,r=e.slice(1);i<r.length;i++){var a=r[i];if(a==null)throw new M('The shape of the input to "Flatten" is not fully defined '+("(got "+e.slice(1)+"). Make sure to pass a complete ")+'"input_shape" or "batch_input_shape" argument to the first layer in your model.')}return[e[0],Ni(e,1)]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.dataFormat==="channelsFirst"&&a.rank>1){for(var s=[0],o=2;o<a.rank;++o)s.push(o);s.push(1),a=a.transpose(s)}return r3(a)})},t.prototype.getConfig=function(){var e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);var i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Flatten",t}(De);y.serialization.registerClass(zb);var Pb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.activation=Ei(e.activation),i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);return r.activation.apply(a)})},t.prototype.getConfig=function(){var e={activation:Oi(this.activation)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Activation",t}(De);y.serialization.registerClass(Pb);var _b=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.n=e.n,i.inputSpec=[{ndim:2}],i}return t.prototype.computeOutputShape=function(e){return[e[0],this.n,e[1]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return e=xe(e),n3(e,r.n)})},t.prototype.getConfig=function(){var e={n:this.n},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="RepeatVector",t}(De);y.serialization.registerClass(_b);var Mb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;i.targetShape=e.targetShape;for(var r=0;r<i.targetShape.length;++r)i.isUnknown(i.targetShape[r])&&(i.targetShape[r]=null);return i}return t.prototype.isUnknown=function(e){return e<0||e==null},t.prototype.fixUnknownDimension=function(e,i){for(var r="Total size of new array must be unchanged.",a=i.slice(),s=1,o=null,l=0;l<a.length;++l){var u=a[l];if(this.isUnknown(u))if(o===null)o=l;else throw new M("Can only specifiy one unknown dimension.");else s*=u}var c=Ni(e);if(o!==null){if(s===0||c%s!==0)throw new M(r);a[o]=c/s}else if(c!==s)throw new M(r);return a},t.prototype.computeOutputShape=function(e){for(var i=!1,r=0;r<e.length;++r)if(this.isUnknown(e[r])){i=!0;break}return i?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=a.shape,o=s.slice(0,1).concat(r.fixUnknownDimension(s.slice(1),r.targetShape));return a.reshape(o)})},t.prototype.getConfig=function(){var e={targetShape:this.targetShape},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Reshape",t}(De);y.serialization.registerClass(Mb);var Hb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error("Permute constructor requires `dims` to be an Array, but received "+(e.dims+" instead."));var r=Ln(1,e.dims.length+1);if(!y.util.arraysEqual(e.dims.slice().sort(),r))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");return i.dims=e.dims,i.dimsIncludingBatch=[0].concat(i.dims),i.inputSpec=[new Nt({ndim:i.dims.length+1})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=e.slice();return this.dims.forEach(function(r,a){i[a+1]=e[r]}),i},t.prototype.call=function(e,i){return y.transpose(xe(e),this.dimsIncludingBatch)},t.prototype.getConfig=function(){var e={dims:this.dims},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Permute",t}(De);y.serialization.registerClass(Hb);var Vb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null?i.maskValue=e.maskValue==null?0:e.maskValue:i.maskValue=0,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={maskValue:this.maskValue};return Object.assign(i,e),i},t.prototype.computeMask=function(e,i){var r=xe(e),a=-1;return y.any(y.notEqual(r,this.maskValue),a)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=-1,o=!0,l=y.any(y.notEqual(a,r.maskValue),s,o),u=a.mul(l.asType(a.dtype));return u})},t.className="Masking",t}(De);y.serialization.registerClass(Vb);var qb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.embeddings=null,i.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){var r=null;e.batchSize!=null&&(r=e.batchSize),e.inputLength==null?i.batchInputShape=[r,null]:i.batchInputShape=[r].concat(Je(e.inputLength))}return i.inputDim=e.inputDim,Tt(i.inputDim,"inputDim"),i.outputDim=e.outputDim,Tt(i.outputDim,"outputDim"),i.embeddingsInitializer=tt(e.embeddingsInitializer||i.DEFAULT_EMBEDDINGS_INITIALIZER),i.embeddingsRegularizer=nt(e.embeddingsRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.embeddingsConstraint=wt(e.embeddingsConstraint),i.maskZero=e.maskZero,i.supportsMasking=e.maskZero,i.inputLength=e.inputLength,i}return t.prototype.build=function(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0},t.prototype.warnOnIncompatibleInputShape=function(e){},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){return r.maskZero?(e=xe(e),y.notEqual(e,y.zerosLike(e))):null})},t.prototype.computeOutputShape=function(e){if(e=je(e),this.inputLength==null)return e.concat([this.outputDim]);var i=Je(this.inputLength);if(i.length!==e.length-1)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));for(var r=0,a=0;a<i.length;++a){var s=i[a],o=e[a+1];if(s!=null&&o!=null&&s!==o)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));s==null&&(i[r]=o),r++}return[e[0]].concat(i,[this.outputDim])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);a.dtype!=="int32"&&(a=es(a,"int32"));var s=wy(r.embeddings.read(),a.as1D());return s.reshape(je(r.computeOutputShape(a.shape)))})},t.prototype.getConfig=function(){var e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:st(this.embeddingsInitializer),embeddingsRegularizer:Ke(this.embeddingsRegularizer),activityRegularizer:Ke(this.activityRegularizer),embeddingsConstraint:bt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Embedding",t}(De);y.serialization.registerClass(qb);var mr=function(n){Q(t,n);function t(e){var i=n.call(this,e||{})||this;return i.supportsMasking=!0,i}return t.prototype.mergeFunction=function(e){throw new Te},t.prototype.computeElementwiseOpOutputShape=function(e,i){if(e==null||i==null)return null;if(e.length<i.length)return this.computeElementwiseOpOutputShape(i,e);if(i.length===0)return e;for(var r=e.slice(0,e.length-i.length),a=0;a<i.length;++a){var s=e[e.length-i.length+a],o=i[a];if(s==null||o==null||s<0||o<0)r.push(null);else if(s===1)r.push(o);else if(o===1)r.push(s);else{if(s!==o)throw new M("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(i));r.push(s)}}return r},t.prototype.build=function(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[je(e)]),e=e,e.length<2)throw new M("A merge layer should be called on an Array of at least 2 inputs."+(" Got "+e.length+" input(s)."));for(var i=[],r=0,a=e;r<a.length;r++){var s=a[r];s!=null&&s[0]!==null&&i.push(s[0])}if(i=Ti(i),i.length>1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l<e.length;++l){var s=e[l]==null?null:e[l].slice(1);o=this.computeElementwiseOpOutputShape(o,s)}var u=e.map(function(c){return c.length});e.indexOf(null)===-1&&Ti(u).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,r.reshapeRequired){var a=[],s=e.map(function(D){return D.rank});if(s.indexOf(null)===-1){for(var o=xi(s),l=0,u=e;l<u.length;l++){for(var c=u[l],h=c.rank,d=0;d<o-h;++d)c=ts(c,1);a.push(c)}return r.mergeFunction(a)}else{for(var p=!1,f=0,m=e;f<m.length;f++){var c=m[f],h=c.rank;if(h==null){var g=c.shape,v=g[0],b=g.slice(1).concat([v]),w=c.reshape([v].concat(Ni(g.slice(1))));w=y.transpose(w,[1,0]),w=w.reshape(b),a.push(w),p=!0}else if(h>1){var S=Ln(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),N=L.rank;if(p){if(N==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(N>1){var S=[N-1].concat(Ln(0,N-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r<e.length;++r){var a=e[r]==null?null:e[r].slice(1);i=this.computeElementwiseOpOutputShape(i,a)}for(var s=[],o=0,l=e;o<l.length;o++){var a=l[o];a!=null&&a[0]!==null&&s.push(a[0])}return s=Ti(s),s.length===1?i=s.concat(i):i=[null].concat(i),i},t.prototype.computeMask=function(e,i){return y.tidy(function(){if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an Array");if(!Array.isArray(e))throw new M("`inputs` should be an Array");if(i.length!==e.length)throw new M("The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths "+("("+e.length+" vs "+i.length+")"));if(i.every(function(s){return s==null}))return null;i=i.map(function(s){return s==null?s:y.expandDims(s,0)});for(var r=i[0],a=1;a<i.length-1;++a)r=y.logicalAnd(r,i[a]);return r})},t}(De),Gb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return i})},t.className="Add",t}(mr);y.serialization.registerClass(Gb);var Yb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.mul(i,e[r]);return i})},t.className="Multiply",t}(mr);y.serialization.registerClass(Yb);var jb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return y.mul(1/e.length,i)})},t.className="Average",t}(mr);y.serialization.registerClass(jb);var Kb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.maximum(i,e[r]);return i})},t.className="Maximum",t}(mr);y.serialization.registerClass(Kb);var $b=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.minimum(i,e[r]);return i})},t.className="Minimum",t}(mr);y.serialization.registerClass($b);var Xb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_AXIS=-1,e==null&&(e={}),i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new M("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;for(var i=!0,r=0,a=e;r<a.length;r++){var s=a[r];if(s!=null){i=!1;break}}if(i)return;for(var o=[],l=0;l<e.length;++l){var u=e[l].slice();u.splice(this.axis,1);for(var c=!1,h=0,d=o;h<d.length;h++){var s=d[h];if(y.util.arraysEqual(s,u)){c=!0;break}}c||o.push(u)}if(o.length>1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return Vh(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s<o.length;s++){var l=o[s];if(r[a]==null||l[a]==null){r[a]=null;break}r[a]+=l[a]}return r},t.prototype.computeMask=function(e,i){var r=this;if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new M("`inputs` should be an array for Concatenate");if(i.length!==e.length)throw new M("Mismatch in the length of mask ("+i.length+") "+("and the legnth of inputs ("+e.length+")"));return y.tidy(function(){var a=!0;if(i.forEach(function(u){if(u!=null){a=!1;return}}),a)return null;for(var s=[],o=0;o<e.length;++o)i[o]==null?s.push(y.onesLike(e[o]).asType("bool")):i[o].rank<e[o].rank?s.push(y.expandDims(i[o],-1)):s.push(i[o]);var l=y.concat(s,r.axis);return y.all(l,-1,!1)})},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Concatenate",t}(mr);y.serialization.registerClass(Xb);function us(n,t){for(;n<0;)n+=t;return n}function nF(n,t,e){if(n.shape.length>3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;l<s;++l)o.push(1);t=t.reshape(t.shape.concat(o))}else if(r>i){s=r-i;for(var o=[],l=0;l<s;++l)o.push(1);n=n.reshape(n.shape.concat(o))}else s=0;var u;if(n.shape.length===2&&t.shape.length===2)a[0]===a[1]?u=n.mul(t).sum(a[0]):u=n.transpose([1,0]).mul(t).sum(a[1]);else{var c=a[0]!==n.shape.length-1,h=a[1]===t.shape.length-1;u=n.matMul(t,c,h)}if(s>0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l<d+s;++l)p.push(l);u=u.squeeze(p)}return u.shape.length===1&&(u=u.expandDims(1)),u})}var Jb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.axes=e.axes,i.normalize=e.normalize==null?!1:e.normalize,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0],r=e[1];if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return us(s,e[o].shape.length)}):a=[us(this.axes,i.shape.length),us(this.axes,r.shape.length)],this.normalize&&(i=Eo(i,a[0]),r=Eo(r,a[1])),nF(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[us(this.axes,e.length),us(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(mr);y.serialization.registerClass(Jb);var Zb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return To(a.shape,0,r.stddev).add(a)},o=is(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass(Zb);var Qb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(To(a.shape,1,o))};return is(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(Qb);var ew=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=es(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return is(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(ew);function cs(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function iF(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=cs(n,s,o,e,t,r);return[l,s,o]})}function rF(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=Ln(0,n.rank);u<c.length;u++){var h=c[u];i.indexOf(h)!==-1?l.push(1):l.push(n.shape[h])}var d=s.reshape(l),p=o.reshape(l),f=t==null?null:t.reshape(l),m=e==null?null:e.reshape(l),g=cs(n,d,p,m,f,r);return[g,s,o]})}function aF(n,t,e,i,r){return r===void 0&&(r=.001),y.util.arraysEqual(i.slice().sort(),Ln(0,n.rank-1))?iF(n,t,e,i,r):rF(n,t,e,i,r)}var tw=function(n){Q(t,n);function t(e){var i=this;return e==null&&(e={}),i=n.call(this,e)||this,i.supportsMasking=!0,i.axis=e.axis==null?-1:e.axis,i.momentum=e.momentum==null?.99:e.momentum,i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.movingMeanInitializer=tt(e.movingMeanInitializer||"zeros"),i.movingVarianceInitializer=tt(e.movingVarianceInitializer||"ones"),i.betaConstraint=wt(e.betaConstraint),i.gammaConstraint=wt(e.gammaConstraint),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i}return t.prototype.build=function(e){var i;e=je(e);var r=this.axis>=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Nt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=Ln(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=ur(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,Ln(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),N=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return cs(s,L,N,C,R,r.epsilon)}else return cs(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=aF(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,N,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(N).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),movingMeanInitializer:st(this.movingMeanInitializer),movingVarianceInitializer:st(this.movingVarianceInitializer),betaRegularizer:Ke(this.betaRegularizer),gammaRegularizer:Ke(this.gammaRegularizer),betaConstraint:bt(this.betaConstraint),gammaConstraint:bt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(tw);var nw=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r<a.length;r++){var s=a[r];if(!Number.isInteger(s))throw new Error("Expected axis to be an array of integers, "+("but received "+JSON.stringify(i.axis)))}else throw new Error("Expected axis to be an integer or an array of integers, "+("but received "+JSON.stringify(i.axis)));return i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i.supportsMasking=!0,i}return t.prototype.build=function(e){e=je(e);var i=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(var r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=i);for(var a=0,s=this.axis;a<s.length;a++){var o=s[a];if(o<0||o>=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==Ti(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=ur(1,o),p=0,f=r.axis;p<f.length;p++){var m=f[p];d[m]=s[m]}for(var g=function(N){return N!=null&&N.shape.length!==o&&r.axis!==[o-1]?N.reshape(d):N},v=g(r.gamma.read()),b=g(r.beta.read()),w=[],S=[],L=0;L<o;++L)r.axis.indexOf(L)!==-1?(w.push(s[L]),S.push(1)):(w.push(1),S.push(s[L]));return c=c.tile(w),h=h.tile(w),v=v.tile(S),b=b.tile(S),cs(a,c,h,b,v,r.epsilon)})},t.prototype.getConfig=function(){var e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),betaRegularizer:Ke(this.betaRegularizer),gammaRegularizer:Ke(this.gammaRegularizer)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LayerNormalization",t}(De);y.serialization.registerClass(nw);function sF(n,t,e){return y.tidy(function(){if(n.rank!==4)throw new M("temporalPadding expects input tensor to be 4-D, but received a "+(n.rank+"-D tensor."));if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new M("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(e==null&&(e=Sn()),e!=="channelsLast"&&e!=="channelsFirst")throw new M("Unknown data format: "+e+". Supported data formats are 'channelsLast' and 'channelsFirst.");var i;return e==="channelsFirst"?i=[[0,0],[0,0],t[0],t[1]]:i=[[0,0],t[0],t[1],[0,0]],y.pad(n,i)})}var iw=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.dataFormat=e.dataFormat==null?Sn():e.dataFormat,e.padding==null)i.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")i.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new M("ZeroPadding2D expects padding to be a length-2 array, but "+("received a length-"+e.padding.length+" array."));var r=void 0,a=void 0;if(typeof e.padding[0]=="number")r=[e.padding[0],e.padding[0]],a=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new M("ZeroPadding2D expects height padding to be a length-2 array, "+("but received a length-"+e.padding[0].length+" array."));if(r=e.padding[0],e.padding[1].length!==2)throw new M("ZeroPadding2D expects width padding to be a length-2 array, "+("but received a length-"+e.padding[1].length+" array."));a=e.padding[1]}i.padding=[r,a]}return i.inputSpec=[new Nt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i,r;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return sF(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(iw);function Vo(n,t,e,i,r,a){return y.tidy(function(){ht(r),py(a),sn(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=Sn()),a==null&&(a="max"),n=vd(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function rw(n,t,e,i,r,a){return y.tidy(function(){ht(r),py(a),sn(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=Sn()),a==null&&(a="max"),n=wb(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var aw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(Tt(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return Tt(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,sn(i.padding),i.inputSpec=[new Nt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=Tn(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=ts(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),sw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),sn(a),Vo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(aw);y.serialization.registerClass(sw);var ow=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),sn(a),Vo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(aw);y.serialization.registerClass(ow);var lw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return Tt(i.poolSize,"poolSize"),Tt(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ht(i.dataFormat),sn(i.padding),i.inputSpec=[new Nt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=Tn(i,this.poolSize[0],this.padding,this.strides[0]),r=Tn(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),uw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),sn(a),Vo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(lw);y.serialization.registerClass(uw);var cw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),sn(a),Vo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(lw);y.serialization.registerClass(cw);var hw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return Tt(i.poolSize,"poolSize"),Tt(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ht(i.dataFormat),sn(i.padding),i.inputSpec=[new Nt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=Tn(i,this.poolSize[0],this.padding,this.strides[0]),r=Tn(r,this.poolSize[1],this.padding,this.strides[1]),a=Tn(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),dw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),sn(a),rw(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(hw);y.serialization.registerClass(dw);var pw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),sn(a),rw(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(hw);y.serialization.registerClass(pw);var fw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Nt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),mw=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(fw);y.serialization.registerClass(mw);var gw=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(fw);y.serialization.registerClass(gw);var vw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ht(i.dataFormat),i.inputSpec=[new Nt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),yw=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(vw);y.serialization.registerClass(yw);var bw=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(vw);y.serialization.registerClass(bw);var ww=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=An(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),Sw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=je(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=je(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Eb(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(ww);y.serialization.registerClass(Sw);function oF(n){ea(Zk,"BidirectionalMergeMode",n)}var lF="concat",Lw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=An(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=An(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?lF:e.mergeMode,oF(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):qt(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Ob(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Nt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof In,d=0,p=o;d<p.length;d++){var f=p[d];if(f instanceof In!==h)throw new M("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors")}if(h){var m=[e].concat(o),g=this.inputSpec.concat(l),v=this.inputSpec;this.inputSpec=g;var b=n.prototype.apply.call(this,m,i);return this.inputSpec=v,b}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.initialState,s,o;if(a==null)s=r.forwardLayer.call(e,i),o=r.backwardLayer.call(e,i);else{var l=a.slice(0,a.length/2),u=a.slice(a.length/2);s=r.forwardLayer.call(e,Object.assign(i,{initialState:l})),o=r.backwardLayer.call(e,Object.assign(i,{initialState:u}))}var c;r.returnState&&(Array.isArray(s)&&(c=s.slice(1).concat(o.slice(1))),s=s[0],o=o[0]),r.returnSequences&&(o=y.reverse(o,1));var h;return r.mergeMode==="concat"?h=Vh([s,o]):r.mergeMode==="sum"?h=y.add(s,o):r.mergeMode==="ave"?h=y.mul(.5,y.add(s,o)):r.mergeMode==="mul"?h=y.mul(s,o):r.mergeMode==null&&(h=[s,o]),r.returnState?r.mergeMode==null?h.concat(c):[h].concat(c):h})},t.prototype.resetStates=function(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()},t.prototype.build=function(e){var i=this;hr(this.forwardLayer.name,function(){i.forwardLayer.build(e)}),hr(this.backwardLayer.name,function(){i.backwardLayer.build(e)}),this.built=!0},t.prototype.computeMask=function(e,i){Array.isArray(i)&&(i=i[0]);var r;if(this.returnSequences?this.mergeMode==null?r=[i,i]:r=i:this.mergeMode==null?r=[null,null]:r=null,this.returnState){var a=this.forwardLayer.states,s=a.map(function(o){return null});return Array.isArray(r)?r.concat(s).concat(s):[r].concat(s).concat(s)}else return r},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e={mergeMode:this.mergeMode},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.fromConfig=function(e,i){var r=An(i.layer);if(delete i.layer,i.numConstants!=null)throw new Te("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");var a=i;return a.layer=r,new e(a)},t.className="Bidirectional",t}(ww);y.serialization.registerClass(Lw);function uF(n){return new rs(n)}function cF(n){return new vb(n)}function hF(n){return new fb(n)}function dF(n){return new mb(n)}function pF(n){return new gb(n)}function fF(n){return new bb(n)}function mF(n){return new yb(n)}function gF(n){return new Nb(n)}function vF(n){return new yd(n)}function yF(n){return new Ab(n)}function bF(n){return new Ib(n)}function wF(n){return new Tb(n)}function SF(n){return new xb(n)}function LF(n){return new Cb(n)}function IF(n){return new Rb(n)}function AF(n){return new Pb(n)}function TF(n){return new Bb(n)}function NF(n){return new Id(n)}function xF(n){return new Ub(n)}function CF(n){return new zb(n)}function RF(n){return new _b(n)}function OF(n){return new Mb(n)}function EF(n){return new Hb(n)}function DF(n){return new qb(n)}function kF(n){return new Gb(n)}function FF(n){return new jb(n)}function WF(n){return new Xb(n)}function UF(n){return new Kb(n)}function BF(n){return new $b(n)}function zF(n){return new Yb(n)}function PF(n){return new Jb(n)}function _F(n){return new tw(n)}function MF(n){return new nw(n)}function HF(n){return new iw(n)}function Ad(n){return new ow(n)}function VF(n){return Ad(n)}function qF(n){return Ad(n)}function Td(n){return new cw(n)}function GF(n){return Td(n)}function YF(n){return Td(n)}function Nd(n){return new pw(n)}function jF(n){return Nd(n)}function KF(n){return Nd(n)}function $F(n){return new mw(n)}function XF(n){return new yw(n)}function Iw(n){return new gw(n)}function Aw(n){return new bw(n)}function Tw(n){return new sw(n)}function Nw(n){return new uw(n)}function JF(n){return new dw(n)}function ZF(n){return new kb(n)}function QF(n){return new Sd(n)}function eW(n){return new Fb(n)}function tW(n){return new Ho(n)}function nW(n){return new Db(n)}function iW(n){return new wd(n)}function rW(n){return new Wb(n)}function aW(n){return new Ld(n)}function sW(n){return new Di(n)}function oW(n){return new bd(n)}function lW(n){return new Lw(n)}function uW(n){return new Sw(n)}var cW=Iw,hW=Aw,dW=Tw,pW=Nw;function fW(n){return new Zb(n)}function mW(n){return new Qb(n)}function gW(n){return new ew(n)}function vW(n){return new Vb(n)}var yW={__proto__:null,inputLayer:uF,elu:cF,reLU:hF,leakyReLU:dF,prelu:pF,softmax:fF,thresholdedReLU:mF,conv1d:gF,conv2d:vF,conv2dTranspose:yF,conv3d:bF,separableConv2d:wF,cropping2D:SF,upSampling2d:LF,depthwiseConv2d:IF,activation:AF,dense:TF,dropout:NF,spatialDropout1d:xF,flatten:CF,repeatVector:RF,reshape:OF,permute:EF,embedding:DF,add:kF,average:FF,concatenate:WF,maximum:UF,minimum:BF,multiply:zF,dot:PF,batchNormalization:_F,layerNormalization:MF,zeroPadding2d:HF,averagePooling1d:Ad,avgPool1d:VF,avgPooling1d:qF,averagePooling2d:Td,avgPool2d:GF,avgPooling2d:YF,averagePooling3d:Nd,avgPool3d:jF,avgPooling3d:KF,globalAveragePooling1d:$F,globalAveragePooling2d:XF,globalMaxPooling1d:Iw,globalMaxPooling2d:Aw,maxPooling1d:Tw,maxPooling2d:Nw,maxPooling3d:JF,gru:ZF,gruCell:QF,lstm:eW,lstmCell:tW,simpleRNN:nW,simpleRNNCell:iW,convLstm2d:rW,convLstm2dCell:aW,rnn:sW,stackedRNNCells:oW,bidirectional:lW,timeDistributed:uW,globalMaxPool1d:cW,globalMaxPool2d:hW,maxPool1d:dW,maxPool2d:pW,Layer:De,RNN:Di,RNNCell:oa,input:ub,gaussianNoise:fW,gaussianDropout:mW,alphaDropout:gW,masking:vW};function bW(n,t){return id(n,t)}function wW(n,t){return Yy(n,t)}function SW(n,t){return jy(n,t)}function LW(n,t){return rd(n,t)}function IW(n,t){return ad(n,t)}function AW(n,t){return Gy(n,t)}function TW(n,t){return j3(n,t)}function NW(n,t){return td(n,t)}function xW(n,t){return Do(n,t)}function CW(n,t){return ra(n,t)}function RW(n,t){return ra(n,t)}function OW(n,t){return ra(n,t)}function EW(n,t){return pr(n,t)}function DW(n,t){return pr(n,t)}function kW(n,t){return pr(n,t)}var FW={__proto__:null,binaryAccuracy:bW,binaryCrossentropy:wW,sparseCategoricalAccuracy:SW,categoricalAccuracy:LW,categoricalCrossentropy:IW,precision:AW,recall:TW,cosineProximity:NW,meanAbsoluteError:xW,meanAbsolutePercentageError:CW,MAPE:RW,mape:OW,meanSquaredError:EW,MSE:DW,mse:kW};var WW={__proto__:null,modelFromJSON:R4};function UW(n){return new Po(n)}function BW(n){return $4(n)}function zW(n){return X4(n)}var PW={__proto__:null,l1l2:UW,l1:BW,l2:zW};var xw=function(n){Q(t,n);function t(){var e=n!==null&&n.apply(this,arguments)||this;return e.model=null,e}return t.prototype.setModel=function(e){if(!(e instanceof Ri))throw new Error("model must be a LayersModel, not some other Container");this.model=e},t}(ia);function qo(n,t){return n<t}function Cw(n,t){return n>t}var Rw=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=qo:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=Cw:i.monitorFunc=qo,i.monitorFunc===qo&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===qo?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,Ci(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(xw);function _W(n){return new Rw(n)}var MW={earlyStopping:_W};Xe.Callback=xw;Xe.CallbackList=zy;Xe.CustomCallback=_y;Xe.EarlyStopping=Rw;Xe.History=Py;Xe.InputSpec=Nt;Xe.LayerVariable=ky;Xe.LayersModel=Ri;Xe.RNN=Di;Xe.Sequential=fd;Xe.SymbolicTensor=In;Xe.callbacks=MW;Xe.constraints=Kk;Xe.initializers=C3;Xe.input=ub;Xe.layers=yW;Xe.loadLayersModel=W4;Xe.metrics=FW;Xe.model=k4;Xe.models=WW;Xe.registerCallbackConstructor=U4;Xe.regularizers=PW;Xe.sequential=F4;Xe.version_layers=ld});var Gw=Le(gr=>{"use strict";Object.defineProperty(gr,"__esModule",{value:!0});var B=lr();var Ew=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t};function Vn(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function Nn(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=r[u[0]&2?"return":u[0]?"throw":"next"])&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[0,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var xn;(function(n){n[n.DT_INVALID=0]="DT_INVALID",n[n.DT_FLOAT=1]="DT_FLOAT",n[n.DT_DOUBLE=2]="DT_DOUBLE",n[n.DT_INT32=3]="DT_INT32",n[n.DT_UINT8=4]="DT_UINT8",n[n.DT_INT16=5]="DT_INT16",n[n.DT_INT8=6]="DT_INT8",n[n.DT_STRING=7]="DT_STRING",n[n.DT_COMPLEX64=8]="DT_COMPLEX64",n[n.DT_INT64=9]="DT_INT64",n[n.DT_BOOL=10]="DT_BOOL",n[n.DT_QINT8=11]="DT_QINT8",n[n.DT_QUINT8=12]="DT_QUINT8",n[n.DT_QINT32=13]="DT_QINT32",n[n.DT_BFLOAT16=14]="DT_BFLOAT16",n[n.DT_FLOAT_REF=101]="DT_FLOAT_REF",n[n.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",n[n.DT_INT32_REF=103]="DT_INT32_REF",n[n.DT_UINT8_REF=104]="DT_UINT8_REF",n[n.DT_INT16_REF=105]="DT_INT16_REF",n[n.DT_INT8_REF=106]="DT_INT8_REF",n[n.DT_STRING_REF=107]="DT_STRING_REF",n[n.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",n[n.DT_INT64_REF=109]="DT_INT64_REF",n[n.DT_BOOL_REF=110]="DT_BOOL_REF",n[n.DT_QINT8_REF=111]="DT_QINT8_REF",n[n.DT_QUINT8_REF=112]="DT_QUINT8_REF",n[n.DT_QINT32_REF=113]="DT_QINT32_REF",n[n.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(xn||(xn={}));var Dw;(function(n){var t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=n.CheckpointFormatVersion||(n.CheckpointFormatVersion={}))})(Dw||(Dw={}));var xd={};function HW(n,t){var e={tfOpName:n,category:"custom",inputs:[],attrs:[],customExecutor:t};xd[n]=e}function kw(n){return xd[n]}function VW(n){delete xd[n]}function A(n,t,e,i){var r=t.inputParams[n];if(r&&r.inputIndexStart!==void 0){var a=r.inputIndexStart,s=r.inputIndexEnd===0?void 0:r.inputIndexEnd===void 0?a+1:r.inputIndexEnd;if(r.type==="tensor")return Gt(t.inputNames[r.inputIndexStart],e,i);if(r.type==="tensors"){var o=t.inputNames.slice(a,s);return o.map(function(h){return Gt(h,e,i)})}var l=Gt(t.inputNames.slice(a)[0],e,i),u=l.dataSync();return r.type==="number"?u[0]:B.util.toNestedArray(l.shape,u)}var c=t.attrParams[n];return c&&c.value}function Gt(n,t,e){var i=Zt(n),r=i[0],a=i[1],s=e.currentContextIds.find(function(o){return!!t[Go(r,o)]});return s!==void 0?t[Go(r,s)][a]:void 0}function qW(n,t,e){return t[Go(n,e.currentContextId)]}function si(n,t){var e=Zt(n),i=e[0],r=e[1];return[Go(i,t&&t.currentContextId),r]}function Go(n,t){return t?n+"-"+t:n}function Zt(n){var t=n.split(":");if(t.length===1)return[n,0];var e=t[0];return[e,Number(t[t.length-1])]}function Yo(n,t,e){var i=A("pad",n,t,e);if(i==="explicit"){i=A("explicitPaddings",n,t,e);for(var r=[[0,0],[0,0],[0,0],[0,0]],a=0;a<4;a++)r[a][0]=i[a*2],r[a][1]=i[a*2+1];return r}return i}function oi(n){return n.kept?n:B.clone(n)}var GW=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],YW={__proto__:null,json:GW};var jW=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"clip_value_min",name:"clipValueMin",type:"number"},{tfName:"clip_value_max",name:"clipValueMax",type:"number"}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"clipValueMin",name:"clipValueMin",type:"number",defaultValue:0},{tfName:"clipValueMax",name:"clipValueMax",type:"number",defaultValue:6}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],KW={__proto__:null,json:jW};var $W=[{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],XW={__proto__:null,json:$W};var JW=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],ZW={__proto__:null,json:JW};var QW=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],eU={__proto__:null,json:QW};var tU=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],nU={__proto__:null,json:tU};var iU=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],rU={__proto__:null,json:iU};var aU=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],sU={__proto__:null,json:aU};var oU=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],lU={__proto__:null,json:oU};var uU=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],cU={__proto__:null,json:uU};var hU=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],dU={__proto__:null,json:hU};var pU=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],fU={__proto__:null,json:pU};var mU=[{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],gU={__proto__:null,json:mU};var vU=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool",notSupported:!0}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],yU={__proto__:null,json:vU};var bU=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],wU={__proto__:null,json:bU};var SU=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]}],LU={__proto__:null,json:SU};var Ww=function(){function n(){var t=[YW,KW,XW,ZW,eU,nU,rU,cU,lU,sU,dU,fU,gU,yU,wU,LU],e=[].concat.apply([],t.map(function(i){return i.json}));this.opMappers=e.reduce(function(i,r){return i[r.tfOpName]=r,i},{})}return Object.defineProperty(n,"Instance",{get:function(){return this._instance||(this._instance=new this)},enumerable:!0,configurable:!0}),n.prototype.transformGraph=function(t,e){var i=this;e===void 0&&(e={});var r=t.node,a=[],s=[],o=[],l=r.reduce(function(g,v){return g[v.name]=i.mapNode(v),v.op.startsWith("Placeholder")?a.push(g[v.name]):v.op==="Const"?s.push(g[v.name]):(v.input==null||v.input.length===0)&&o.push(g[v.name]),g},{}),u=[],c=[],h={},d={};e!=null&&(h=this.mapSignatureEntries(e.inputs),d=this.mapSignatureEntries(e.outputs));var p=Object.keys(l);p.forEach(function(g){var v=l[g];v.inputNames.forEach(function(b){var w=si(b)[0];v.inputs.push(l[w]),l[w].children.push(v)})}),Object.keys(d).length===0?p.forEach(function(g){var v=l[g];v.children.length===0&&c.push(v)}):Object.keys(d).forEach(function(g){var v=si(g)[0],b=l[v];b!=null&&(b.signatureKey=d[g],c.push(b))}),Object.keys(h).length>0?Object.keys(h).forEach(function(g){var v=si(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=kw(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Cd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Cd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=Ud(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Ud(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=Od(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=Od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=Wd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Wd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=zd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=zd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=Fd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Fd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=Bd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Bd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=Dd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Dd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=kd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=kd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Fw(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Fw(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=si(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:Ed(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=si(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=si(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function IU(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function Uw(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):IU(n);return t?e:e.toLowerCase()}function Cd(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?Uw(r.s,i):e}function Rd(n,t,e){var i=n[t];return i?i.b:e}function Od(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function Ed(n){typeof n=="string"&&(n=xn[n]);switch(n){case xn.DT_FLOAT:return"float32";case xn.DT_INT32:case xn.DT_INT64:case xn.DT_INT8:case xn.DT_UINT8:return"int32";case xn.DT_BOOL:return"bool";case xn.DT_DOUBLE:return"float32";case xn.DT_STRING:return"string";default:return null}}function Fw(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function Dd(n,t,e){var i=n[t];return i&&i.type?Ed(i.type):e}function kd(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return Ed(r)}):e}function Bw(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function Fd(n,t,e){var i=n[t];return i&&i.shape?Bw(i.shape):e}function Wd(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function Ud(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return Uw(a,i)}):e}function Bd(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Bw(r)}):e}function zd(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var AU=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Gt(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Gt(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return Od(this.node.rawAttrs,t,e);if(i.s!=null)return Cd(this.node.rawAttrs,t,e);if(i.b!=null)return Rd(this.node.rawAttrs,t,e);if(i.shape!=null)return Fd(this.node.rawAttrs,t,e);if(i.type!=null)return Dd(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return Wd(this.node.rawAttrs,t,e);if(i.list.s!=null)return Ud(this.node.rawAttrs,t,e);if(i.list.shape!=null)return Bd(this.node.rawAttrs,t,e);if(i.list.b!=null)return zd(this.node.rawAttrs,t,e);if(i.list.type!=null)return kd(this.node.rawAttrs,t,e)}return e},n}();var TU=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var NU=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Gt(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function mn(n,t,e){e===void 0&&(e=""),B.util.assert(xU(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function xU(n,t){if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==-1&&t[e]!==-1&&n[e]!==t[e])return!1;return!0}var CU=function(){function n(t,e,i,r,a,s,o){this.name=t,this.dtype=e,this.maxSize=i,this.elementShape=r,this.identicalElementShapes=a,this.dynamicSize=s,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=B.scalar(0),B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"closed",{get:function(){return this.closed_},enumerable:!0,configurable:!0}),n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.read=function(t){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||t>=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`,
-          because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),mn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i<this.size();i++)t.push(i)}if(t.length===0)return B.tensor([],[0].concat(this.elementShape));var r=this.readMany(t);return mn(this.elementShape,r[0].shape,"TensorArray shape mismatch: "),B.stack(r,0)},n.prototype.concat=function(t){if(!!t&&t!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but concat requested dtype "+t);if(this.size()===0)return B.tensor([],[0].concat(this.elementShape));for(var e=[],i=0;i<this.size();i++)e.push(i);var r=this.readMany(e);return mn(this.elementShape,r[0].shape,"TensorArray shape mismatch: tensor array shape ("+this.elementShape+") vs first tensor shape ("+r[0].shape+")"),B.concat(r,0)},n.prototype.scatter=function(t,e){if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);if(t.length!==e.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+e.shape[0]);var i=Math.max.apply(Math,t);if(!this.dynamicSize&&i>=this.maxSize)throw new Error("Max index must be < array size ("+i+"  vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
+`+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");if(p!=null){for(var f={},m=0,g=Object.keys(dn);m<g.length;m++){var v=g[m];f[v]=dn[v]}for(var b=0,w=Object.keys(e);b<w.length;b++){var v=w[b];f[v]=e[v]}var S=c.config;S.customObjects=f;for(var L=Kt({},dn),N=0,C=Object.keys(e);N<C.length;N++){var v=C[N];dn[v]=e[v]}Ph(c.config);var R=p(d,c.config,e,r);return dn=Kt({},L),R}else{for(var L=Kt({},dn),D=0,k=Object.keys(e);D<k.length;D++){var v=k[D];dn[v]=e[v]}var R=new d(c.config);return dn=Kt({},L),R}}}function Mk(n,t){return n<t?-1:n>t?1:0}function Io(n,t){return-1*Mk(n,t)}function Ti(n){if(n==null)return n;for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.indexOf(r)===-1&&t.push(r)}return t}function Hk(n){if(n==null)throw new M("Invalid value in obj: "+JSON.stringify(n));for(var t in n)if(n.hasOwnProperty(t))return!1;return!0}function ea(n,t,e){if(e==null)return;if(n.indexOf(e)<0)throw new M(e+" is not a valid "+t+".  Valid values are "+n+" or null/undefined.")}function _h(n,t,e,i){return e===void 0&&(e=0),i===void 0&&(i=Infinity),_n(e>=0),_n(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function Tt(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return Tt(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(ay(n)+".")})}function ay(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return ay(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function Vk(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];var o=y.util.now();return o-e<t||(e=o,i=n.apply(void 0,a)),i};return r}function sy(n){return n==="relu"?"relu":n==="linear"?"linear":n==="elu"?"elu":null}function Mh(n,t){return y.tidy(function(){return y.sqrt(y.sum(y.mul(n,n),t,!0))})}var Za=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),oy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMaxValue=2,i.defaultAxis=0,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=Mh(e,i.axis),a=y.clipByValue(r,0,i.maxValue);return y.mul(e,y.div(a,y.add(yt(),r)))})},t.prototype.getConfig=function(){return{maxValue:this.maxValue,axis:this.axis}},t.className="MaxNorm",t}(Za);y.serialization.registerClass(oy);var ly=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultAxis=0,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){return y.div(e,y.add(yt(),Mh(e,i.axis)))})},t.prototype.getConfig=function(){return{axis:this.axis}},t.className="UnitNorm",t}(Za);y.serialization.registerClass(ly);var uy=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="NonNeg",t}(Za);y.serialization.registerClass(uy);var cy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMinValue=0,i.defaultMaxValue=1,i.defaultRate=1,i.defaultAxis=0,i.minValue=e.minValue!=null?e.minValue:i.defaultMinValue,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.rate=e.rate!=null?e.rate:i.defaultRate,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=Mh(e,i.axis),a=y.add(y.mul(i.rate,y.clipByValue(r,i.minValue,i.maxValue)),y.mul(1-i.rate,r));return y.mul(e,y.div(a,y.add(yt(),r)))})},t.prototype.getConfig=function(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}},t.className="MinMaxNorm",t}(Za);y.serialization.registerClass(cy);var hy={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function bt(n){return zh(n)}function dy(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"constraint")}function wt(n){if(n==null)return null;if(typeof n=="string"){var t=n in hy?hy[n]:n,e={className:t,config:{}};return dy(e)}else return n instanceof Za?n:dy(n)}function qk(n){return new oy(n)}function Gk(n){return new ly(n)}function Yk(){return new uy}function jk(n){return new cy(n)}var Kk={__proto__:null,maxNorm:qk,unitNorm:Gk,nonNeg:Yk,minMaxNorm:jk};var $k=["channelsFirst","channelsLast"],Xk=["valid","same","causal"],Jk=["max","avg"],Zk=["sum","mul","concat","ave"];var ta=new Map;function ht(n){ea($k,"DataFormat",n)}function an(n){ea(Xk,"PaddingMode",n)}function py(n){ea(Jk,"PoolMode",n)}var Qa=[],fy="/";function hr(n,t){Qa.push(n);try{var e=t();return Qa.pop(),e}catch(i){throw Qa.pop(),i}}function Qk(){return Qa.length===0?"":Qa.join(fy)+fy}function gy(n){if(!my(n))throw new Error("Not a valid tensor name: '"+n+"'");return Qk()+n}function vy(n){if(!my(n))throw new Error("Not a valid tensor name: '"+n+"'");ta.has(n)||ta.set(n,0);var t=ta.get(n);if(ta.set(n,ta.get(n)+1),t>0){var e=n+"_"+t;return ta.set(e,1),e}else return n}var e3=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function my(n){return!!n.match(e3)}function t3(n){return n===parseInt(n.toString(),10)}function Ni(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r<e;++r)i*=n[r];return i}function yy(n){return n=Array.isArray(n)?new Float32Array(n):n,y.tensor1d(n)}function na(n){return y.min(yy(n)).dataSync()[0]}function xi(n){return y.max(yy(n)).dataSync()[0]}function Sn(n,t){if(t<n)throw new M("end ("+t+") < begin ("+n+") is forbidden.");for(var e=[],i=n;i<t;++i)e.push(i);return e}function es(n,t){return n.asType(t)}function ts(n,t){t===void 0&&(t=-1);var e=n.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),n.reshape(e)}function n3(n,t){return y.tidy(function(){if(n.shape.length!==2)throw new M("repeat() expects a rank-2 tensor, but received a "+("rank-"+n.shape.length+" tensor."));var e=ts(n,1);return Hh(e,[1,t,1])})}function i3(n){var t=[Ni(n.shape)];return n.reshape(t)}function r3(n){if(n.rank<=1)throw new M("batchFlatten requires a minimum rank of 2. Got rank: "+n.rank+".");var t=[n.shape[0],Ni(n.shape,1)];return n.reshape(t)}function dr(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[t,0],[e,n.shape[1]]);case 3:return y.slice3d(n,[t,0,0],[e,n.shape[1],n.shape[2]]);case 4:return y.slice4d(n,[t,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3]]);case 5:return y.slice(n,[t,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4]]);case 6:return y.slice(n,[t,0,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4],n.shape[5]]);default:throw new M("sliceAlongFirstAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function Vh(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[0,t],[n.shape[0],e]);case 3:return y.slice3d(n,[0,0,t],[n.shape[0],n.shape[1],e]);case 4:return y.slice4d(n,[0,0,0,t],[n.shape[0],n.shape[1],n.shape[2],e]);default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function Ao(n,t,e,i){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:switch(i){case 1:return dr(n,t,e);case 2:return Vh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 3:switch(i){case 1:return dr(n,t,e);case 2:return y.slice3d(n,[0,t,0],[n.shape[0],e,n.shape[2]]);case 3:return Vh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 4:switch(i){case 1:return dr(n,t,e);case 2:return y.slice4d(n,[0,t,0,0],[n.shape[0],e,n.shape[2],n.shape[3]]);case 3:return y.slice4d(n,[0,0,t,0],[n.shape[0],n.shape[1],e,n.shape[3]]);case 4:return Vh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function qh(n,t){t===void 0&&(t=-1);var e;return t<0&&(e=n[0].rank,e!==0?t=e:t=0),t===n[0].rank&&(t=-1),y.concat(n,t)}function by(n,t){switch(n.rank){case 1:return y.concat1d([n,t]);case 2:return y.concat2d([n,t],0);case 3:return y.concat3d([n,t],0);case 4:return y.concat4d([n,t],0);default:throw new M("concatAlongFirstAxis() received an unsupported "+("tensor rank: "+n.rank))}}function Hh(n,t){if(Array.isArray(t)||(t=[t]),n.rank!==t.length)throw new M("The length of input n ("+t.length+") does not match "+("the number of dimensions in input x ("+n.rank+")"));return y.tile(n,t)}function To(n,t,e,i,r){return t===void 0&&(t=0),e===void 0&&(e=1),y.randomNormal(n,t,e,i,r)}function Mn(n,t,e,i){if(n.rank<2||t.rank<2)throw new Te("dot requires both inputs to be rank >= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?Gh(n.rank,i,wn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?Gh(n.rank,i,wn()):null,activation:e}).reshape(f)}}function wy(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function ns(n){return y.mul(n,n)}function Gh(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Hn(n,t,e){return y.tidy(function(){return e==null&&(e=wn()),ht(e),n.add(Gh(n.rank,t,e))})}function a3(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function s3(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function Sy(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function o3(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function is(n,t,e){return e===void 0&&(e=!1),e?n():t()}var l3=["fanIn","fanOut","fanAvg"],u3=["normal","uniform","truncatedNormal"];function c3(n){ea(l3,"FanMode",n)}function h3(n){ea(u3,"Distribution",n)}var pn=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),Ly=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(pn);y.serialization.registerClass(Ly);var Yh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(pn);y.serialization.registerClass(Yh);var Iy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(pn);y.serialization.registerClass(Iy);var Ay=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(pn);y.serialization.registerClass(Ay);var Ty=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return To(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(pn);y.serialization.registerClass(Ty);var Ny=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(pn);y.serialization.registerClass(Ny);var xy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(pn);y.serialization.registerClass(xy);function d3(n,t){t===void 0&&(t="channelsLast");var e,i;if(ht(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=Ni(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=Ni(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=Ni(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var $t=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,c3(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,h3(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=d3(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(pn);y.serialization.registerClass($t);var jh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return $t.className},t.className="GlorotUniform",t}($t);y.serialization.registerClass(jh);var Kh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return $t.className},t.className="GlorotNormal",t}($t);y.serialization.registerClass(Kh);var $h=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return $t.className},t.className="HeNormal",t}($t);y.serialization.registerClass($h);var Xh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return $t.className},t.className="HeUniform",t}($t);y.serialization.registerClass(Xh);var Jh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return $t.className},t.className="LeCunNormal",t}($t);y.serialization.registerClass(Jh);var Zh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return $t.className},t.className="LeCunNormal",t}($t);y.serialization.registerClass(Zh);var Cy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=To(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(pn);y.serialization.registerClass(Cy);var Ry={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function Oy(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function st(n){return zh(n)}function tt(n){if(typeof n=="string"){var t=n in Ry?Ry[n]:n;if(t==="GlorotNormal")return new Kh;if(t==="GlorotUniform")return new jh;if(t==="HeNormal")return new $h;if(t==="HeUniform")return new Xh;if(t==="LeCunNormal")return new Jh;if(t==="LeCunUniform")return new Zh;var e={};return e.className=t,e.config={},Oy(e)}else return n instanceof pn?n:Oy(n)}function p3(){return new Ly}function f3(){return new Yh}function m3(n){return new Iy(n)}function g3(n){return new Ay(n)}function v3(n){return new Ty(n)}function y3(n){return new Ny(n)}function b3(n){return new xy(n)}function w3(n){return new $t(n)}function S3(n){return new jh(n)}function L3(n){return new Kh(n)}function I3(n){return new $h(n)}function A3(n){return new Xh(n)}function T3(n){return new Jh(n)}function N3(n){return new Zh(n)}function x3(n){return new Cy(n)}var C3={__proto__:null,zeros:p3,ones:f3,constant:m3,randomUniform:g3,randomNormal:v3,truncatedNormal:y3,identity:b3,varianceScaling:w3,glorotUniform:S3,glorotNormal:L3,heNormal:I3,heUniform:A3,leCunNormal:T3,leCunUniform:N3,orthogonal:x3};var R3=0;function Ey(){return R3++}var No={};function xo(n){return n===void 0&&(n=""),n in No||(No[n]=0),No[n]+=1,n+No[n].toString()}function Qh(n){return Array.isArray(n)&&Array.isArray(n[0])}function Co(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function je(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function Ro(n){for(var t=0,e=0,i=n;e<i.length;e++){var r=i[e];r.shape.length===0?t+=1:t+=r.shape.reduce(function(a,s){return a*s})}return t}var Dy="Variable",ky=function(){function n(t,e,i,r,a){e===void 0&&(e="float32"),i===void 0&&(i=Dy),r===void 0&&(r=!0),a===void 0&&(a=null),this.dtype=e??"float32",this.shape=t.shape,this.id=Ey(),i=i??Dy,this.originalName=gy(i),this.name=vy(this.originalName),this.trainable_=r,this.constraint=a,this.val=y.variable(t,this.trainable_,this.name,this.dtype)}return n.prototype.read=function(){return this.assertNotDisposed(),this.val},n.prototype.write=function(t){return this.assertNotDisposed(),O3(this.val,t),this.val.id!==t.id&&(this.val.assign(t),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this},n.prototype.dispose=function(){this.assertNotDisposed(),this.val.dispose()},n.prototype.assertNotDisposed=function(){if(this.val.isDisposed)throw new Error("LayersVariable "+this.name+" is already disposed.")},Object.defineProperty(n.prototype,"trainable",{get:function(){return this.trainable_},set:function(t){this.trainable_=t,this.val.trainable=t},enumerable:!0,configurable:!0}),n}();function O3(n,t){if(n.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(n.shape)+" vs. "+JSON.stringify(t.shape))}function ed(n){return n.map(function(t){return t.read()})}function td(n){n.forEach(function(t){var e=t[0];e.write(t[1])})}var Nt=function(){function n(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}return n}(),Ln=function(){function n(t,e,i,r,a,s,o){this.dtype=t,this.shape=e,this.sourceLayer=i,this.inputs=r,this.callArgs=a,this.outputTensorIndex=o,this.id=Ey(),s!=null&&(this.originalName=gy(s),this.name=vy(this.originalName)),this.rank=e.length}return n}(),E3=0,Oo=function(){function n(t,e){this.callArgs=e,this.id=E3++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(var i=0,r=t.inboundLayers;i<r.length;i++){var a=r[i];a!=null&&a.outboundNodes.push(this)}t.outboundLayer.inboundNodes.push(this)}return n.prototype.getConfig=function(){for(var t=[],e=0,i=this.inboundLayers;e<i.length;e++){var r=i[e];r!=null?t.push(r.name):t.push(null)}return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}},n}(),D3=0,De=function(n){Q(t,n);function t(e){e===void 0&&(e={});var i=n.call(this)||this;i._callHook=null,i._addedWeightNames=[],i._stateful=!1,i.id=D3++,i.activityRegularizer=null,i.inputSpec=null,i.supportsMasking=!1,i._trainableWeights=[],i._nonTrainableWeights=[],i._losses=[],i._updates=[],i._built=!1,i.inboundNodes=[],i.outboundNodes=[];var r=e.name;if(!r){var a=i.getClassName();r=ai(a)+"_"+xo(a)}if(i.name=r,i.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){var s=void 0;if(e.batchInputShape!=null)s=e.batchInputShape;else if(e.inputShape!=null){var o=null;e.batchSize!=null&&(o=e.batchSize),s=[o].concat(e.inputShape)}i.batchInputShape=s;var l=e.dtype;l==null&&(l=e.inputDType),l==null&&(l="float32"),i.dtype=l}return e.weights!=null?i.initialWeights=e.weights:i.initialWeights=null,i._refCount=null,i.fastWeightInitDuringBuild=!1,i}return t.nodeKey=function(e,i){return e.name+"_ib-"+i.toString()},t.prototype.getNodeAtIndex=function(e,i){if(this.inboundNodes.length===0)throw new Pn("The layer has never been called "+("and thus has no defined "+i+"."));if(this.inboundNodes.length<=e)throw new M("Asked to get "+i+" at node "+e+", "+("but the layer has only "+this.inboundNodes.length+" inbound nodes."));return this.inboundNodes[e]},t.prototype.getInputAt=function(e){return qt(this.getNodeAtIndex(e,"input").inputTensors)},t.prototype.getOutputAt=function(e){return qt(this.getNodeAtIndex(e,"output").outputTensors)},Object.defineProperty(t.prototype,"input",{get:function(){if(this.inboundNodes.length>1)throw new Ai("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new Ai("Layer "+this.name+" is not connected, no input to return.");return qt(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new Ai("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new Ai("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return qt(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Je(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Je(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r<e.length;r++){var a=e[r],s=i[r];if(s==null)continue;var o=a.rank;if(s.ndim!=null&&o!==s.ndim)throw new M("Input "+r+" is incompatible with layer "+this.name+": "+("expected ndim="+s.ndim+", found ndim="+o));if(s.maxNDim!=null&&o>s.maxNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected max_ndim="+s.maxNDim+", found ndim="+o));if(s.minNDim!=null&&o<s.minNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected min_ndim="+s.minNDim+", found ndim="+o+"."));if(s.dtype!=null&&a.dtype!==s.dtype)throw new M("Input "+r+" is incompatible with layer "+this.name+" "+(": expected dtype="+s.dtype+", found dtype="+a.dtype+"."));if(s.axes){var l=a.shape;for(var u in s.axes){var c=Number(u),h=s.axes[u],d=c>=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p<s.shape.length;++p){var f=s.shape[p],m=a.shape[p];if(f!=null&&m!=null&&f!==m)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected shape="+s.shape+", ")+("found shape="+a.shape+"."))}}},t.prototype.call=function(e,i){return e},t.prototype.invokeCallHook=function(e,i){this._callHook!=null&&this._callHook(e,i)},t.prototype.setCallHook=function(e){this._callHook=e},t.prototype.clearCallHook=function(){this._callHook=null},t.prototype.apply=function(e,i){var r=this;i=i||{},this.assertNotDisposed();for(var a=Je(e),s=!0,o=0,l=a;o<l.length;o++){var u=l[o];if(!(u instanceof Ln)){s=!1;break}}for(var c=!0,h=0,d=a;h<d.length;h++){var u=d[h];if(u instanceof Ln){c=!1;break}}if(s===c)throw new M("Arguments to apply() must be all SymbolicTensors or all Tensors");return hr(this.name,function(){if(!r.built){r.assertInputCompatibility(e);for(var p=[],f=0,m=Je(e);f<m.length;f++){var g=m[f];p.push(g.shape)}r.build(qt(p)),r.built=!0,r.initialWeights&&r.setWeights(r.initialWeights),r._refCount===null&&c&&(r._refCount=1)}if(r.assertInputCompatibility(e),c){for(var v=r.call(e,i),b=Je(v),w=[],S=0,L=b;S<L.length;S++){var N=L[S];a.indexOf(N)!==-1&&(N=N.clone()),w.push(N)}if(v=qt(w),r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}else{var C=k3(e),R=r.computeOutputShape(C),v=void 0,D=F3(e);if(r.warnOnIncompatibleInputShape(Array.isArray(e)?C[0]:C),R!=null&&R.length>0&&Array.isArray(R[0])?v=R.map(function(W,F){return new Ln(D,W,r,Je(e),i,r.name,F)}):v=new Ln(D,R,r,Je(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new Ai("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i<r.length;i++){var a=r[i],s=JSON.stringify(a.outputShapes);e.indexOf(s)===-1&&e.push(s)}if(e.length===1){var o=this.inboundNodes[0].outputShapes;return Array.isArray(o)&&Array.isArray(o[0])&&o.length===1?o[0]:o}else throw new Ai("The layer "+this.name+' has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.')},enumerable:!0,configurable:!0}),t.prototype.countParams=function(){if(!this.built)throw new Pn("You tried to call countParams() on "+this.name+", but the layer is not built yet. Build it first by calling build(batchInputShape).");return Ro(this.weights)},t.prototype.build=function(e){this.built=!0},t.prototype.getWeights=function(e){return e===void 0&&(e=!1),ed(e?this.trainableWeights:this.weights)},t.prototype.setWeights=function(e){var i=this;y.tidy(function(){var r=i.weights;if(r.length!==e.length)throw new M('You called setWeights(weights) on layer "'+i.name+'" '+("with a weight list of length "+e.length+", ")+("but the layer was expecting "+r.length+" weights. ")+("Provided weights: "+e+"..."));if(r.length===0)return;for(var a=[],s=ed(r),o=0;o<s.length;++o){var l=s[o],u=r[o],c=e[o];if(!y.util.arraysEqual(l.shape,c.shape))throw new M("Layer weight shape "+l.shape+" "+("not compatible with provided weight shape "+c.shape));a.push([u,c])}td(a)})},t.prototype.addWeight=function(e,i,r,a,s,o,l){if(this._addedWeightNames.indexOf(e)!==-1)throw new M("Duplicate weight name "+e+" for layer "+this.name);this._addedWeightNames.push(e),r==null&&(r="float32"),this.fastWeightInitDuringBuild&&(a=tt("zeros"));var u=a.apply(i,r),c=new ky(u,r,e,o,l);return u.dispose(),s!=null&&this.addLoss(function(){return s.apply(c.read())}),o==null&&(o=!0),o?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c},t.prototype.setFastWeightInitDuringBuild=function(e){this.fastWeightInitDuringBuild=e},t.prototype.addLoss=function(e){var i;if(e==null||Array.isArray(e)&&e.length===0)return;e=Je(e),this._losses!==void 0&&this._losses!==null&&(i=this.losses).push.apply(i,e)},t.prototype.computeOutputShape=function(e){return e},t.prototype.computeMask=function(e,i){var r=this;if(!this.supportsMasking){if(i!=null)if(Array.isArray(i))i.forEach(function(a){if(a!=null)throw new TypeError("Layer "+r.name+" does not support masking, but was passed an inputMask.")});else throw new TypeError("Layer "+this.name+" does not support masking, but was passed an inputMask.");return null}return i},t.prototype.addInboundNode=function(e,i,r,a,s,o,l){l===void 0&&(l=null);var u=Je(e);i=Je(i),r=Je(r),a=Je(a),s=Co(s),o=Co(o);for(var c=[],h=[],d=[],p=0,f=u;p<f.length;p++){var m=f[p];c.push(m.sourceLayer),h.push(m.nodeIndex),d.push(m.tensorIndex)}new Oo({outboundLayer:this,inboundLayers:c,nodeIndices:h,tensorIndices:d,inputTensors:u,outputTensors:i,inputMasks:r,outputMasks:a,inputShapes:s,outputShapes:o},l);for(var g=0;g<i.length;g++)i[g].sourceLayer=this,i[g].nodeIndex=this.inboundNodes.length-1,i[g].tensorIndex=g},t.prototype.getConfig=function(){var e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e},t.prototype.disposeWeights=function(){return this.weights.forEach(function(e){return e.dispose()}),this.weights.length},t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Layer '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){if(!this.built)throw new Error("Cannot dispose Layer "+this.name+" because it has not been built yet.");if(this._refCount===null)throw new Error("Cannot dispose Layer "+this.name+" because it has not been used yet.");this.assertNotDisposed();var e=0;return--this._refCount===0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}},t}(y.serialization.Serializable);function k3(n){n=Je(n);for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.push(r.shape)}return qt(t)}function F3(n){return"float32"}function Fy(n,t,e){if((t==null||e!=null&&e>0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a<i.inboundLayers.length;a++)for(var s=i.inputTensors[a],o=i.inboundLayers[a],l=i.nodeIndices[a],u=Fy(s,o,l),c=0,h=u;c<h.length;c++){var d=h[c];r.indexOf(d)===-1&&r.push(d)}return r}var rs=function(n){Q(t,n);function t(e){var i=n.call(this,{dtype:e.dtype,name:e.name!=null?e.name:xo("input").toString()})||this;if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),i.trainable=!1,i.built=!0,i.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new M("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");var r=e.batchInputShape;if(r==null){if(e.inputShape==null)throw new M("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");r=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new M("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");var a=e.dtype||"float32";i.batchInputShape=r,i.dtype=a,i.inputSpec=[{shape:r}];var s=new Ln(i.dtype,i.batchInputShape,i,[],{},i.name);return s.nodeIndex=0,s.tensorIndex=0,new Oo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[r],outputShapes:[r]}),i}return t.prototype.apply=function(e,i){throw new M("Cannot pass any input to an "+("InputLayer's apply() method. InputLayer name: "+this.name))},t.prototype.dispose=function(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}},t.prototype.getConfig=function(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}},t.className="InputLayer",t}(De);y.serialization.registerClass(rs);function Wy(n){if(n.batchShape==null&&n.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(n.batchShape!=null&&n.shape!=null)throw new M("Please provide either a `shape` or `batchShape` argument to Input, but not both.");var t=n.batchShape;n.shape!=null&&t==null&&(t=[null].concat(n.shape));var e=n.dtype;e==null&&(e="float32");var i=new rs({batchInputShape:t,name:n.name,dtype:e,sparse:n.sparse}),r=i.inboundNodes[0].outputTensors;return r[0]}function Ci(n){return Se(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(n==null)return[2];t=[],e=[],i=[];for(r in n)a=n[r],typeof a!="number"&&(s=a,t.push(s.data()),e.push(r),i.push(s));return t.length>0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l<o.length;++l)n[e[l]]=o[l][0];y.dispose(i),u.label=2;case 2:return[2]}})})}function Uy(n){if(n==null)return;for(var t in n){var e=n[t];typeof e!="number"&&e.dispose()}}var By;(function(n){n[n.SILENT=0]="SILENT",n[n.VERBOSE=1]="VERBOSE"})(By||(By={}));var W3=125,ia=function(){function n(){this.validationData=null}return n.prototype.setParams=function(t){this.params=t},n.prototype.onEpochBegin=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onEpochEnd=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchBegin=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchEnd=function(t,e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onTrainBegin=function(t){return Se(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.onTrainEnd=function(t){return Se(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.setModel=function(t){},n}(),zy=function(){function n(t,e){e===void 0&&(e=10),t==null&&(t=[]),this.callbacks=t,this.queueLength=e}return n.prototype.append=function(t){this.callbacks.push(t)},n.prototype.setParams=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setParams(t)}},n.prototype.setModel=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setModel(t)}},n.prototype.onEpochBegin=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onEpochEnd=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchBegin=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchEnd=function(t,e){return Se(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onTrainBegin=function(t){return Se(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainBegin(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n.prototype.onTrainEnd=function(t){return Se(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainEnd(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n}(),U3=function(n){Q(t,n);function t(){return n.call(this)||this}return t.prototype.onEpochBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.seen=0,this.totals={},[2]})})},t.prototype.onBatchEnd=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l=this;return ve(this,function(u){i==null&&(i={}),r=i.size==null?0:i.size,this.seen+=r,a=function(c){var h=i[c];if(typeof h=="number")s.totals.hasOwnProperty(c)||(s.totals[c]=0),s.totals[c]=s.totals[c]+h*r;else{var d=void 0;c in s.totals?d=s.totals[c]:s.totals[c]=0;var p=y.tidy(function(){return y.add(l.totals[c],y.mul(h,r))});s.totals[c]=p,d!=null&&d.dispose()}},s=this;for(o in i)a(o);return[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l,u=this;return ve(this,function(c){if(i!=null)for(r=function(h){if(a.totals[h]==null)return"continue";typeof a.totals[h]=="number"?i[h]=a.totals[h]/a.seen:y.tidy(function(){var d=y.mul(y.div(1,u.seen),u.totals[h]);i[h]=d,u.totals[h].dispose(),y.keep(i[h])})},a=this,s=0,o=this.params.metrics;s<o.length;s++)l=o[s],r(l);return[2]})})},t}(ia),Py=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.epoch=[],this.history={},[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){i==null&&(i={}),this.epoch.push(e);for(r in i)this.history[r]==null&&(this.history[r]=[]),this.history[r].push(i[r]);return[2]})})},t.prototype.syncData=function(){return Se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:e=[],i=[],r=[];for(a in this.history)for(s=this.history[a],o=0;o<s.length;++o)typeof s[o]!="number"&&(l=s[o],e.push(l.data()),i.push(a),r.push(o));return[4,Promise.all(e)];case 1:for(u=d.sent(),c=0;c<u.length;++c)h=this.history[i[c]][r[c]],h.dispose(),this.history[i[c]][r[c]]=u[c][0];return[2]}})})},t}(ia),_y=function(n){Q(t,n);function t(e,i){var r=n.call(this)||this;if(r.currentEpoch=0,r.yieldEvery=i||"auto",r.yieldEvery==="auto"&&(r.yieldEvery=W3),r.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");return y.util.isNumber(r.yieldEvery)&&(r.maybeWait=Vk(r.maybeWait.bind(r),r.yieldEvery)),r.trainBegin=e.onTrainBegin,r.trainEnd=e.onTrainEnd,r.epochBegin=e.onEpochBegin,r.epochEnd=e.onEpochEnd,r.batchBegin=e.onBatchBegin,r.batchEnd=e.onBatchEnd,r.yield=e.onYield,r}return t.prototype.maybeWait=function(e,i,r){return Se(this,void 0,void 0,function(){var a;return ve(this,function(s){switch(s.label){case 0:return a=[],this.yield!=null?[4,Ci(r)]:[3,2];case 1:s.sent(),a.push(this.yield(e,i,r)),s.label=2;case 2:return a.push(y.nextFrame()),[4,Promise.all(a)];case 3:return s.sent(),[2]}})})},t.prototype.onEpochBegin=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.currentEpoch=e,this.epochBegin!=null?[4,Ci(i)]:[3,3];case 1:return r.sent(),[4,this.epochBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.epochEnd!=null?[4,Ci(i)]:[3,2];case 1:a.sent(),r.push(this.epochEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="epoch"&&r.push(y.nextFrame()),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onBatchBegin=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.batchBegin!=null?[4,Ci(i)]:[3,3];case 1:return r.sent(),[4,this.batchBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onBatchEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.batchEnd!=null?[4,Ci(i)]:[3,2];case 1:a.sent(),r.push(this.batchEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="batch"?r.push(y.nextFrame()):y.util.isNumber(this.yieldEvery)&&r.push(this.maybeWait(this.currentEpoch,e,i)),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainBegin!=null?[4,Ci(e)]:[3,3];case 1:return i.sent(),[4,this.trainBegin(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t.prototype.onTrainEnd=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainEnd!=null?[4,Ci(e)]:[3,3];case 1:return i.sent(),[4,this.trainEnd(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t}(ia);function My(n,t){if(n==null&&(n={}),n instanceof ia)return[n];if(Array.isArray(n)&&n[0]instanceof ia)return n;var e=Je(n);return e.map(function(i){return new _y(i,t)})}var Hy=function(){function n(){}return n.registerCallbackConstructor=function(t,e){y.util.assert(t>=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function Vy(n,t,e,i,r,a,s,o,l){var u=new Py,c=[new U3].concat(Hy.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new zy(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function In(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function Eo(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(ns(n),t,!0),i=y.fill(e.shape,yt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function pr(n,t){return y.tidy(function(){return y.mean(ns(y.sub(t,n)),-1)})}function Do(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function ra(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),yt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function B3(n,t){return y.tidy(function(){var e=y.clipByValue(t,yt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,yt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(ns(y.sub(i,a)),-1)})}function z3(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(ns(e),-1)})}function P3(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function _3(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function M3(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function as(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,yt(),1-yt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function ko(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(i3(n)).toInt();t=y.clipByValue(t,yt(),1-yt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return as(a,t,e)})}function H3(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function Fo(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,yt(),1-yt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(H3(n,e),-1)})}function V3(n,t){return y.tidy(function(){var e=y.clipByValue(n,yt(),1),i=y.clipByValue(t,yt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function q3(n,t){return y.tidy(function(){var e=y.log(y.add(yt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function nd(n,t){return y.tidy(function(){var e=Eo(n,-1),i=Eo(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var Wo={meanSquaredError:pr,meanAbsoluteError:Do,meanAbsolutePercentageError:ra,meanSquaredLogarithmicError:B3,squaredHinge:z3,hinge:P3,categoricalHinge:_3,logcosh:M3,categoricalCrossentropy:as,sparseCategoricalCrossentropy:ko,binaryCrossentropy:Fo,kullbackLeiblerDivergence:V3,poisson:q3,cosineProximity:nd};function id(n){if(typeof n=="string"){if(n in Wo)return Wo[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function rd(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=es(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function ad(n,t){return y.tidy(function(){return es(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function qy(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function G3(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function Y3(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function Gy(n,t){return y.tidy(function(){var e=qy(n,t),i=Y3(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function j3(n,t){return y.tidy(function(){var e=qy(n,t),i=G3(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Yy(n,t){return Fo(n,t)}function jy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var K3=pr,$3=pr,X3=Do,J3=Do,Z3=ra,Q3=ra,sd=as,e4=nd,Ky=ko,Uo={binaryAccuracy:rd,categoricalAccuracy:ad,precision:Gy,categoricalCrossentropy:sd,sparseCategoricalCrossentropy:Ky,mse:K3,MSE:$3,mae:X3,MAE:J3,mape:Z3,MAPE:Q3,cosine:e4};function t4(n){if(typeof n=="string"&&n in Uo)return Uo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function Bo(n){if(_n(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(Wo);e<i.length;e++){var r=i[e];if(Wo[r]===n){t=r;break}}if(t!==void 0)return t;for(var a=0,s=Object.keys(Uo);a<s.length;a++){var r=s[a];if(Uo[r]===n){t=r;break}}return t!==void 0?t:n.name}function n4(n){var t={Adagrad:function(){return y.train.adagrad(.01)},Adadelta:function(){return y.train.adadelta(1,.95,yt())},Adam:function(){return y.train.adam(.001,.9,.999,yt())},Adamax:function(){return y.train.adamax(.002,.9,.999,yt(),0)},RMSProp:function(){return y.train.rmsprop(.001,.9,0,yt())},SGD:function(){return y.train.sgd(.01)}};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,n in t)return t[n]();throw new M("Unknown Optimizer "+n)}var $y=1*1024*1024;function Xy(n,t,e){if(e===void 0&&(e=!1),n==null||typeof n!="object"||Object.getPrototypeOf(n)!==Object.prototype||!od(n))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){var i=JSON.stringify(n);i.length>$y&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+($y+"."))}}function od(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e];if(typeof r!="string")return!1;if(!od(n[r]))return!1}return!0}else if(Array.isArray(n)){for(var a=0,s=n;a<s.length;a++){var o=s[a];if(!od(o))return!1}return!0}else return!1;else{var l=typeof n;return l==="string"||l==="number"||l==="boolean"}}function o4(n,t,e,i){i===void 0&&(i=console.log);var r=r4(n),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,e=e||[.45,.85,1]):(t=t||98,e=e||[.33,.55,.67,1]),e[e.length-1]<=1&&(e=e.map(function(d){return Math.floor(t*d)}));var s;if(!r){a.push("Receives inputs"),s=[];for(var o in n.nodesByDepth)s.push.apply(s,n.nodesByDepth[o])}i("_".repeat(t)),zo(a,e,i),i("=".repeat(t));for(var l=n.layers,u=0;u<l.length;++u)r?a4(l[u],e,i):s4(l[u],e,s,i),i((u===l.length-1?"=":"_").repeat(t));n.checkTrainableWeightsConsistency();var c=i4(n),h=Ro(n.nonTrainableWeights);i("Total params: "+(c+h)),i("Trainable params: "+c),i("Non-trainable params: "+h),i("_".repeat(t))}function i4(n){var t;return n.collectedTrainableWeights!=null?t=Ro(n.collectedTrainableWeights):t=Ro(n.trainableWeights),t}function r4(n){var t=!0,e=[],i=[];for(var r in n.nodesByDepth)e.push(n.nodesByDepth[r]);for(var a=0,s=e;a<s.length;a++){var o=s[a];if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l<u.length;l++){for(var c=u[l],h=!1,d=0,p=c.inboundNodes;d<p.length;d++){var f=p[d];if(i.indexOf(f)!==-1)if(h){t=!1;break}else h=!0}if(!t)break}return t}function zo(n,t,e){e===void 0&&(e=console.log);for(var i="",r=0;r<n.length;++r)r>0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function a4(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];zo(s,t,e)}function s4(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s<o.length;s++){var l=o[s];if(e!=null&&e.length>0&&e.indexOf(l)===-1)continue;for(var u=0;u<l.inboundLayers.length;++u){var c=l.inboundLayers[u].name,h=l.nodeIndices[u],d=l.tensorIndices[u];a.push(c+"["+h+"]["+d+"]")}}var p=n.name,f=n.getClassName(),m=a.length===0?"":a[0],g=[p+" ("+f+")",r,n.countParams().toString(),m];zo(g,t,i);for(var u=1;u<a.length;++u)zo(["","","",a[u]],t,i)}function Jy(n,t,e){return(n==="inboundNodes"||n==="outputLayers"||n==="inputLayers")&&t===0&&typeof e=="string"}function ss(n,t){if(n===null)return null;if(typeof n=="string")return cr(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];Jy(t,r,a)?e.push(a):e.push(ss(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u];if(u==="name"&&typeof c=="string")s[u]=c;else{var h=cr(u);s[h]=ss(c,h)}}return s}}function ld(n,t){if(n==null)return null;if(typeof n=="string")return ai(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];Jy(t,r,a)?e.push(a):e.push(ld(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u],h=ai(u);(u==="name"||u==="className")&&typeof c=="string"?s[h]=c:s[h]=ld(c,u)}return s}}var ud="2.6.0";function l4(n,t){if(n.dtype==null||n.dtype===t.dtype)return t;try{return y.cast(t,n.dtype)}catch(e){throw new M("The dtype of the feed ("+t.dtype+") can not be cast to the dtype "+("of the key '"+n.name+"' ("+n.dtype+")."))}}var aa=function(){function n(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof n)for(var e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(var i=0,r=t;i<r.length;i++){var a=r[i];this.add(a.key,a.value)}}}return n.prototype.add=function(t,e,i){if(this.id2Value[t.id]==null)this.id2Value[t.id]=l4(t,e),this.name2Id[t.name]=t.id,i!=null&&(this.id2Mask[t.id]=i);else throw new M("Duplicate key: name="+t.name+", id="+t.id);return this},n.prototype.addFeed=function(t){this.add(t.key,t.value)},n.prototype.hasKey=function(t){return this.id2Value[t.id]!=null},n.prototype.names=function(){return Object.keys(this.name2Id)},n.prototype.getValue=function(t){if(t instanceof Ln){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Value[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Value[e]}},n.prototype.getMask=function(t){if(t instanceof Ln){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Mask[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Mask[e]}},n.prototype.disposeMasks=function(){this.id2Mask!=null&&y.dispose(this.id2Mask)},n}(),cd={},Zy={};function os(n,t,e,i){for(var r=e==null?!1:e.training,a=Array.isArray(n),s=a?n:[n],o=s.map(function(X){return X.name}),l=[],u=t.names(),c=0,h=o;c<h.length;c++){var d=h[c];u.indexOf(d)!==-1?l.push(t.getValue(d)):l.push(null)}i!=null&&(i.maxNumTensors=-Infinity,i.minNumTensors=Infinity);var p=o.join(",")+"|"+t.names().join(","),f,m;if(cd[p]==null){var g=u4(s,t);f=g.sorted,m=g.recipientCounts,cd[p]=f,Zy[p]=m}f=cd[p],m={},r||Object.assign(m,Zy[p]);for(var v=new aa(t),b=0;b<f.length;++b){if(i!=null){var w=y.memory().numTensors;w>i.maxNumTensors&&(i.maxNumTensors=w),w<i.minNumTensors&&(i.minNumTensors=w)}var S=f[b],L=S.sourceLayer;if(L instanceof rs)continue;for(var N=[],C=[],R=[],D=!1,k=0,W=S.inputs;k<W.length;k++){var F=W[k],P=v.getValue(F),H=v.getMask(F);N.push(P),C.push(H),H!=null&&(D=!0),r||(m[F.name]--,m[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!P.isDisposed&&F.sourceLayer.stateful!==!0&&R.push(P))}D&&(e=e||{},e.mask=C[0]);var _=Je(L.apply(N,e)),j=null;L.supportsMasking&&(j=L.computeMask(N,C));for(var K=c4(S),q=Array.isArray(K)?K:[K],G=0;G<q.length;++G){v.hasKey(q[G])||v.add(q[G],_[G],Array.isArray(j)?j[0]:j);var Z=o.indexOf(q[G].name);Z!==-1&&(l[Z]=_[G])}r||y.dispose(R)}return v.disposeMasks(),a?l:l[0]}function u4(n,t){y.util.assert(n!=null&&n.length>0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=Qy(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s<o.length;s++){for(var l=o[s],u=Qy(l,t),c=u.sorted,h=u.recipientMap,d=0,p=c;d<p.length;d++){var f=p[d];a.has(f.name)||(e.push(f),a.add(f.name))}var m=function(v){i[v]==null&&(i[v]=new Set),h[v].forEach(function(b){return i[v].add(b)})};for(var g in h)m(g)}return{sorted:e,recipientCounts:h4(i)}}function h4(n){var t={};for(var e in n)t[e]=n[e].size;return t}function Qy(n,t){for(var e=new Set,i=[],r={},a=0,s=t.names();a<s.length;a++){var o=s[a];e.add(o)}var l=[],u=[];for(l.push(n);l.length>0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d<p.length;d++){var f=p[d];if(r[f.name]==null&&(r[f.name]=new Set),r[f.name].add(c.name),e.has(f.name))continue;l.push(f)}}}return{sorted:i,recipientMap:r}}function c4(n){var t;if(n.sourceLayer.inboundNodes.length===1)t=n.sourceLayer.output;else{for(var e=null,i=0;i<n.sourceLayer.inboundNodes.length;++i)for(var r=0,a=n.sourceLayer.inboundNodes[i].outputTensors;r<a.length;r++){var s=a[r];if(s.id===n.id){e=i;break}}t=n.sourceLayer.getOutputAt(e)}return t}var d4=function(n){Q(t,n);function t(e){var i=n.call(this,{})||this;if(i.containerNodes=new Set,i.name=e.name,i.name==null){var r=i.getClassName().toLowerCase();i.name=xo(r)}if(i.supportsMasking=!1,i.trainable_=!0,Array.isArray(e.inputs)?i.inputs=e.inputs.slice():i.inputs=[e.inputs],Array.isArray(e.outputs)?i.outputs=e.outputs.slice():i.outputs=[e.outputs],Ti(i.inputs).length!==i.inputs.length)throw new M("The list of inputs passed to the model is redundant. All inputs should only appear once. Found: "+(""+i.inputs.map(function(Ce){return Ce.name})));Ti(i.outputs).length!==i.outputs.length&&console.warn("The list of outputs passed to the model is redundant. All outputs should only appear once. Found: "+(""+i.outputs.map(function(Ce){return Ce.name}))),i.inputLayers=[],i.inputLayersNodeIndices=[],i.inputLayersTensorIndices=[],i.outputLayers=[],i.outputLayersNodeIndices=[],i.outputLayersTensorIndices=[],i.layers=[],i.internalContainerRefs=[];for(var a=0,s=i.outputs;a<s.length;a++){var o=s[a],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;i.outputLayers.push(l),i.outputLayersNodeIndices.push(u),i.outputLayersTensorIndices.push(c)}for(var h=0,d=i.inputs;h<d.length;h++){var o=d[h],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;_n(u===0,"input layer has >1 nodes"),_n(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p<i.inputLayers.length;p++){var l=i.inputLayers[p];if(!(l instanceof rs))throw new TypeError("Input layers to a LayersModel must be InputLayer objects. "+("Received inputs: "+e.inputs+". ")+("Input "+p+" (0-based) originates ")+("from layer type "+l.getClassName()+"."));i.inputNames.push(l.name),i.feedInputShapes.push(l.batchInputShape),i.feedInputNames.push(l.name)}for(var f=0,m=i.outputLayers;f<m.length;f++){var l=m[f];i.outputNames.push(l.name)}i.internalInputShapes=i.inputs.map(function(Ce){return Ce.shape}),i.internalOutputShapes=i.outputs.map(function(Ce){return Ce.shape});for(var g={},v={},b={},w={},S={},L=[],N=function(Ce,Wt,ft,it,un,Os){(it==null||un==null||Os==null)&&(it=Ce.sourceLayer,un=Ce.nodeIndex,Os=Ce.tensorIndex);var _t=it.inboundNodes[un];if(ft.indexOf(_t)!==-1)throw new Pn("The tensor "+Ce.name+' at layer "'+it.name+'" is part of a cycle.');if(Wt.indexOf(_t)!==-1)return;i.containerNodes.add(t.nodeKey(it,un)),it.id in S||(S[it.id]=Object.keys(S).length),ft.indexOf(_t)===-1&&ft.push(_t);for(var a2=_t.inboundLayers.length,Or=0;Or<a2;Or++){var s2=_t.inputTensors[Or],o2=_t.inboundLayers[Or],l2=_t.nodeIndices[Or],u2=_t.tensorIndices[Or];N(s2,Wt,ft,o2,l2,u2)}for(Wt.push(_t);ft.indexOf(_t)>=0;)ft.splice(ft.indexOf(_t),1);L.push(_t)},C=[],R=[],D=0,k=i.outputs;D<k.length;D++){var o=k[D];N(o,C,R)}for(var W=L.slice().reverse(),F=0,P=W;F<P.length;F++){var H=P[F];v[H.id]=H,H.id in g||(g[H.id]=0);var _=g[H.id],j=b[H.outboundLayer.id]==null?0:b[H.outboundLayer.id];_=Math.max(_,j),b[H.outboundLayer.id]=_,w[H.outboundLayer.id]=H.outboundLayer,g[H.id]=_;for(var p=0;p<H.inboundLayers.length;p++){var K=H.inboundLayers[p],u=H.nodeIndices[p],q=K.inboundNodes[u],G=g[q.id]==null?0:g[q.id];g[q.id]=Math.max(_+1,G),v[q.id]=q}}var Z={};for(var X in g){var _=g[X];_ in Z||(Z[_]=[]),Z[_].push(v[X])}var ee={};for(var ne in b){var _=b[ne];_ in ee||(ee[_]=[]),ee[_].push(w[ne])}var ie=Object.keys(ee).map(function(Ce){return parseInt(Ce,10)}).sort(Io);i.layers=[];for(var te=0,re=ie;te<re.length;te++){var _=re[te],le=ee[_];le.sort(function(Wt,ft){var it=S[Wt.id],un=S[ft.id];return it<un?-1:it>un?1:0});for(var he=0,ye=le;he<ye.length;he++){var l=ye[he];l instanceof t&&i.internalContainerRefs.push(l),i.layers.push(l)}}i.layersByDepth=ee,ie=Object.keys(Z).map(function(Ce){return parseInt(Ce,10)}).sort(Io);for(var Oe=i.inputs.slice(),Fe=[],_e=0,Me=ie;_e<Me.length;_e++)for(var _=Me[_e],dt=0,Lt=Z[_];dt<Lt.length;dt++){var H=Lt[dt],l=H.outboundLayer;if(l!=null){for(var Ge=0,Ue=H.inputTensors;Ge<Ue.length;Ge++){var o=Ue[Ge];if(Oe.indexOf(o)===-1)throw new Pn("Graph disconnected: cannot obtain value for tensor "+o+(' at layer "'+l.name+'". ')+"The following previous layers were accessed without "+("issue: "+Fe))}for(var me=0,Ft=H.outputTensors;me<Ft.length;me++){var o=Ft[me];Oe.push(o)}Fe.push(l.name)}}i.nodesByDepth=Z;for(var lt=i.layers.map(function(Ce){return Ce.name}),pt=function(Ce){var Wt=lt.filter(function(ft){return ft===Ce}).length;if(Wt!==1)throw new Pn('The name "'+Ce+'" is used '+Wt+" times in the model. All layer names should be unique. Layer names: "+JSON.stringify(lt))},En=0,Et=lt;En<Et.length;En++){var It=Et[En];pt(It)}return i.outboundNodes=[],i.inboundNodes=[],new Oo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:i.inputs,outputTensors:i.outputs,inputMasks:i.inputs.map(function(Ce){return null}),outputMasks:i.outputs.map(function(Ce){return null}),inputShapes:i.inputs.map(function(Ce){return Ce.shape}),outputShapes:i.outputs.map(function(Ce){return Ce.shape})}),i.built=!0,i._refCount=1,i}return t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Container '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){this.assertNotDisposed();var e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(var i=0,r=this.layers;i<r.length;i++){var a=r[i];e.numDisposedVariables+=a.dispose().numDisposedVariables}for(var s=0,o=this.internalContainerRefs;s<o.length;s++){var l=o[s];e.numDisposedVariables+=l.dispose().numDisposedVariables}}return e.refCountAfterDispose=this._refCount,e},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this.layers.forEach(function(i){i._trainableWeights.forEach(function(r){return r.trainable=e})}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(this._trainableWeights.length>0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e=e.concat(a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.layers;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.loadWeights=function(e,i){i===void 0&&(i=!0);for(var r={},a=0,s=0,o=this.layers;s<o.length;s++)for(var l=o[s],u=0,c=l.weights;u<c.length;u++){var h=c[u];if(r[h.originalName]!=null)throw new M("Duplicate weight name: "+h.originalName);r[h.originalName]=h,a++}var d=[];for(var p in e){var f=p;if(r[p]==null){var m=p.split("/"),g=m.slice(0,-2).concat([m[m.length-1]]);f=g.join("/")}if(r[f]!=null)d.push([r[f],e[p]]);else if(i)throw new M("Provided weight data has no target variable: "+p);delete r[f]}if(i){var v=[];for(var b in r)v.push(b);if(v.length>0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}td(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+ud,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=ld(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Je(e);for(var a=new aa,s=0;s<r.inputs.length;++s)a.add(r.inputs[s],e[s]);return os(r.outputs,a,i)})},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){e=Je(e);var a;return i==null?a=ur(null,e.length):a=Je(i),r.runInternalGraph(e,a)[1]})},t.prototype.computeOutputShape=function(e){var i=Co(e);if(i.length!==this.inputLayers.length)throw new M("Invalid inputShape argument "+e+": "+("model has "+this.inputLayers.length+" tensor inputs."));for(var r={},a=0;a<i.length;a++){var s=this.inputLayers[a],o=i[a],l=s.name+"_0_0";r[l]=o}var u=Object.keys(this.nodesByDepth).map(function(P){return parseInt(P,10)}).sort(Io);if(u.length>1)for(var c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){var g=m[f],s=g.outboundLayer;if(this.inputLayers.map(function(H){return H.id}).indexOf(s.id)!==-1)continue;for(var v=[],b=0;b<g.inboundLayers.length;b++){var w=g.inboundLayers[b],S=g.nodeIndices[b],L=g.tensorIndices[b],l=w.name+"_"+S+"_"+L,N=r[l];v.push(N)}for(var C=s.computeOutputShape(qt(v)),R=Co(C),D=s.inboundNodes.indexOf(g),b=0;b<R.length;b++){var l=s.name+"_"+D+"_"+b;r[l]=R[b]}}for(var k=[],W=[],a=0;a<this.outputLayers.length;a++){var s=this.outputLayers[a],D=this.outputLayersNodeIndices[a],L=this.outputLayersTensorIndices[a],l=s.name+"_"+D+"_"+L;W.push(l)}for(var a=0;a<W.length;a++){var F=W[a];_n(F in r),k.push(r[F])}return qt(k)},t.prototype.runInternalGraph=function(e,i){i==null&&(i=ur(null,e.length));for(var r={},a=0;a<this.inputs.length;++a){var s=this.inputs[a],o=e[a],l=i[a];r[s.id]=[o,l]}for(var u=Object.keys(this.nodesByDepth).map(function(ee){return parseInt(ee,10)}).sort(Io),c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){for(var g=m[f],v=g.outboundLayer,b=g.inputTensors,w=g.outputTensors,S=new Array,L=0,N=b;L<N.length;L++){var s=N[L];s.id in r&&S.push(r[s.id])}if(S.length===b.length){var C={},R=void 0,D=void 0,k=void 0,W=void 0;if(g.callArgs!=null&&(C=g.callArgs),S.length===1){var F=S[0],P=F[0],H=F[1];C.mask==null&&(C.mask=H),k=Je(v.call(P,C)),W=Je(v.computeMask(P,H)),R=[P],D=[H]}else R=S.map(function(ee){return ee[0]}),D=S.map(function(ee){return ee[1]}),C.mask==null&&(C.mask=D),k=Je(v.call(R,C)),W=Je(v.computeMask(R,D));if(v.activityRegularizer)throw new Te("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(var a=0;a<w.length;++a){var s=w[a],o=k[a],l=W[a];r[s.id]=[o,l]}}}for(var _=[],j=[],K=[],q=0,G=this.outputs;q<G.length;q++){var s=G[q];_n(s.id in r,"Could not compute output "+s.name+" : "+s.id);var Z=r[s.id],X=Z[0],l=Z[1];K.push(X.shape),_.push(X),j.push(l)}return[_,j,K]},t.prototype.buildNodeConversionMap=function(e){for(var i={},r,a=0,s=this.layers;a<s.length;a++){var o=s[a];r=o instanceof t?1:0;for(var l=0;l<o.inboundNodes.length;l++){var u=t.nodeKey(o,l);this.containerNodes.has(u)&&(i[u]=r,r+=1)}}return i},t.prototype.getLayer=function(e,i){if(i!=null){if(this.layers.length<=i)throw new M("Was asked to retrieve layer at index "+i+", but model only "+("has "+this.layers.length+" layer(s)."));return this.layers[i]}else if(e==null)throw new M("Provide either a layer name or layer index");for(var r=0,a=this.layers;r<a.length;r++){var s=a[r];if(s.name===e)return s}throw new M("No such layer: "+e)},t.prototype.calculateLosses=function(){var e=this;return y.tidy(function(){for(var i=[],r=0,a=e.layers;r<a.length;r++)for(var s=a[r],o=0;o<s.inboundNodes.length;++o){var l=t.nodeKey(s,o);e.containerNodes.has(l)&&i.push.apply(i,s.calculateLosses())}return i})},t.prototype.getConfig=function(){for(var e={name:this.name},i=this.buildNodeConversionMap(this.layers),r=[],a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=o.getClassName(),u=o.getConfig(),c=[],h=0;h<o.inboundNodes.length;h++){var d=o.inboundNodes[h],p=t.nodeKey(o,h),f={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),f=d.callArgs}catch(D){console.warn("Layer "+o.name+" was passed non-serializable keyword arguments: "+(d.callArgs+". They will not be included ")+"in the serialized model (and thus will be missing at deserialization time)."),f={}}if(d.inboundLayers.length>0){for(var m=[],g=0;g<d.inboundLayers.length;g++){var v=d.inboundLayers[g],b=d.nodeIndices[g],w=d.tensorIndices[g],S=t.nodeKey(v,b),L=i[S];L==null&&(L=0),m.push([v.name,L,w,f])}c.push(m)}}}var N={};N.name=o.name,N.className=l,N.config=u,N.inboundNodes=c,r.push(N)}e.layers=r;for(var C=[],g=0;g<this.inputLayers.length;g++){var o=this.inputLayers[g],b=this.inputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.inputLayersTensorIndices[g];C.push([o.name,L,w])}e.inputLayers=C;for(var R=[],g=0;g<this.outputLayers.length;g++){var o=this.outputLayers[g],b=this.outputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.outputLayersTensorIndices[g];R.push([o.name,L,w])}return e.outputLayers=R,e},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s={},o={};function l(G,Z){G.name in o?o[G.name].push(Z):o[G.name]=[Z]}function u(G,Z){for(var X=[],ee,ne=0,ie=Z;ne<ie.length;ne++){var te=ie[ne],re=te[0],le=te[1],he=te[2];if(ee=te[3]==null?{}:te[3],!(re in s)){l(G,Z);return}var ye=s[re];if(ye.inboundNodes.length<=le){l(G,Z);return}var Oe=ye.inboundNodes[le];X.push(Oe.outputTensors[he])}X.length>0&&G.apply(qt(X),ee)}function c(G){var Z=G.name,X=In(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p<f.length;p++){var m=f[p];c(m)}for(;!Hk(o);)for(var g=0,v=d;g<v.length;g++){var m=v[g],b=s[m.name];if(b.name in o){var w=o[b.name];delete o[b.name];for(var S=0,L=w;S<L.length;S++){var N=L[S];u(b,N)}}}for(var C=[],R=[],D=i.inputLayers,k=0,W=D;k<W.length;k++){var m=W[k],F=m[0],P=m[1],H=m[2];_n(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;C.push(_[H])}for(var j=i.outputLayers,K=0,q=j;K<q.length;K++){var m=q[K],F=m[0],P=m[1],H=m[2];_n(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;R.push(_[H])}return new e({inputs:C,outputs:R,name:h})},Object.defineProperty(t.prototype,"stateful",{get:function(){if(this._stateful)throw new M("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(var e=0,i=this.layers;e<i.length;e++){var r=i[e];if(r.stateful)return!0}return!1},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){var e=this;y.tidy(function(){e.layers.forEach(function(i){i.stateful&&i.resetStates()})})},t}(De);function p4(n,t,e){var i=t.length;if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(a){return null});if(i===1)return Array.isArray(n)&&n.length===1?n:typeof n=="object"&&t[0]in n?[n[t[0]]]:[n];if(Array.isArray(n)){if(n.length!==i)throw new Error("Provided "+e+" is an array of "+n.length+" "+("element(s), but the model has "+i+" outputs. ")+"Make sure a set of weights is provided for each model output.");return n}else if(typeof n=="object"&&Object.keys(n).length>0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function eb(n,t){return p4(n,t,"classWeight")}function tb(n,t,e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function f4(n,t){return y.mul(n,t)}var m4=32;function ib(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor.  The provided Dataset instead generates "+(""+t)});var a=nb("input",n.inputNames,e),s=nb("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs.  (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs.  (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},u=0;u<a.length;u++)l(u);for(var c=function(d){y.util.assert(s[d].shape[0]===o,function(){return"Batch size mismatch: output "+(n.outputNames[d]+" has "+s[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},h=0;h<s.length;h++)c(h);return{xs:a,ys:s}}function nb(n,t,e){if(e instanceof y.Tensor)return[e];if(Array.isArray(e))return y.util.assert(e.length===t.length,function(){return"Received an array of "+e.length+" Tensors, but expected "+t.length+" to match the "+n+" keys "+t+"."}),e;for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r];if(e[s]==null)throw new M("The feature data generated by the dataset lacks the required "+(n+" key '"+s+"'."));i.push(e[s])}return i}function g4(n){if(n.length===3)throw new Te("Validation with sample weights is not implemented yet.");return{xs:n[0],ys:n[1]}}function y4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R,D,k,W,F,P,H,_,j,F,K,q,G,Z,F;return ve(this,function(X){switch(X.label){case 0:if(i=e.batchesPerEpoch!=null,y.util.assert(n.optimizer!=null,function(){return"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."}),y.util.assert(e!=null,function(){return"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."}),y.util.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(rb(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=g4(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=My(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=Vy(h,d,e.epochs,null,null,v4(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g<e.epochs?(b={},[4,f.onEpochBegin(g)]):[3,23];case 5:return X.sent(),w=0,S=0,i?[3,7]:[4,t.iterator()];case 6:v=X.sent(),X.label=7;case 7:return(i?w<e.batchesPerEpoch:!0)?[4,v.next()]:[3,21];case 8:return L=X.sent(),i&&L.done?(console.warn("You provided `batchesPerEpoch` as "+(e.batchesPerEpoch+", ")+"but your dataset iterator ran out of data after "+(w+" batches; ")+"interrupting training. Make sure that your dataset can generate at least `batchesPerEpoch * epochs` batches (in this case, "+(e.batchesPerEpoch*e.epochs+" batches). ")+"You may need to use the repeat() function when building your dataset."),[3,21]):L.value!=null?(N=ib(n,L.value),C=N.xs,R=N.ys,D={},D.batch=S,D.size=C[0].shape[0],[4,f.onBatchBegin(S,D)]):[3,15];case 9:if(X.sent(),k=[],!(e.classWeight!=null))return[3,13];W=eb(e.classWeight,n.outputNames),F=0,X.label=10;case 10:return F<W.length?(H=(P=k).push,[4,tb(R[F],null,W[F])]):[3,13];case 11:H.apply(P,[X.sent()]),X.label=12;case 12:return++F,[3,10];case 13:for(_=C.concat(R).concat(k),j=l(_),y.dispose(_),F=0;F<u.length;++F)K=u[F],q=j[F],D[K]=q,y.keep(q);return[4,f.onBatchEnd(S,D)];case 14:X.sent(),Uy(D),S++,w++,X.label=15;case 15:return(i?w>=e.batchesPerEpoch:L.done)?r?(G=void 0,rb(e.validationData)?(Z=Je,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Je(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?m4:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F<n.metricsNames.length;++F)b["val_"+n.metricsNames[F]]=G[F];X.label=19;case 19:return[3,21];case 20:return n.stopTraining_?[3,21]:[3,7];case 21:return[4,f.onEpochEnd(g,b)];case 22:return X.sent(),g++,n.stopTraining_?[3,23]:[3,4];case 23:return[4,f.onTrainEnd()];case 24:return X.sent(),[4,n.history.syncData()];case 25:return X.sent(),[2,n.history];case 26:return n.isTraining=!1,[7];case 27:return[2]}})})}function v4(n,t){var e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(n.size)&&(e=n.size),e}function rb(n){return typeof n.iterator=="function"}function b4(n){return typeof n.next=="function"}function w4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:if(e=e||{},i=e.batches!=null,r=n.testFunction,a=[],e.verbose>0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),b4(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=ib(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var N=0;N<L.length;++N)a.push(y.scalar(0));for(var C=S[0].shape[0],R=function(D){var k=L[D],W=a[D];a[D]=y.tidy(function(){return y.add(a[D],y.mul(C,k))}),u>0&&y.dispose(W)},N=0;N<L.length;++N)R(N);y.dispose(L),l+=C,++u}return a}),m.done?(i&&console.warn("Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least `batches` "+("batches (in this case, "+e.batches+" batches). ")+"You may need to use the repeat() function when building your dataset."),[2,"break"]):[2]}})},f.label=4;case 4:return(i?u<e.batches:!0)?[5,c()]:[3,6];case 5:return h=f.sent(),h==="break"?[3,6]:[3,4];case 6:for(d=0;d<a.length;++d)p=a[d],a[d]=y.div(a[d],l),y.dispose(p);return[2,qt(a)]}})})}function hd(n){y.util.assert(n>0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function ls(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return dr(i,t,e-t)}):dr(n,t,e-t)}function dd(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return dd(e,t)}):wy(n,t.dtype==="int32"?t:t.toInt())})}function pd(n,t){for(var e=[],i=0,r=null;i<n;)r=i+t,r>=n&&(r=n),e.push([i,r]),i=r;return e}function S4(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Se(this,void 0,void 0,function(){var m,g,v,b,w,S,L,N,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=Sn(0,g)),s==null&&(s=1),b=Vy(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(j){switch(j.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(j.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=pd(g,r),P=function(K){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(K,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[K][0],X=F[K][1],ee=dr(W,Z,X-Z);q.batch=K,q.size=X-Z;for(var ne=dd(e,ee),ie=t(ne),te=0;te<i.length;++te){var re=i[te],le=ie[te];q[re]=le,y.keep(le)}if(K===F.length-1&&m)for(var he=n.testLoop(l,u,r),te=0;te<i.length;++te){var re=i[te],le=he[te];y.keep(le),k["val_"+re]=le}}),[4,w.onBatchEnd(K,q)];case 2:return G.sent(),Uy(q),n.stopTraining_?[2,"break"]:[2]}})},H=0,j.label=3;case 3:return H<F.length?[5,P(H)]:[3,6];case 4:if(_=j.sent(),_==="break")return[3,6];j.label=5;case 5:return++H,[3,3];case 6:W.dispose(),j.label=7;case 7:return[4,w.onEpochEnd(D,k)];case 8:return j.sent(),n.stopTraining_?[2,"break"]:[2]}})},N=d,R.label=2;case 2:return N<a?[5,L(N)]:[3,5];case 3:if(C=R.sent(),C==="break")return[3,5];R.label=4;case 4:return++N,[3,2];case 5:return[4,w.onTrainEnd()];case 6:return R.sent(),[4,n.history.syncData()];case 7:return R.sent(),[2,n.history]}})})}function L4(n,t,e,i){return i===void 0&&(i={}),Se(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R,D,k;return ve(this,function(W){switch(W.label){case 0:if(n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,W.label=1;case 1:return W.trys.push([1,,7,8]),h=i.batchSize==null?32:i.batchSize,hd(h),d=!1,[4,n.standardizeUserData(t,e,i.sampleWeight,i.classWeight,d,h)];case 2:if(p=W.sent(),r=p[0],a=p[1],c=p[2],f=!1,m=void 0,!(i.validationData!=null&&i.validationData.length>0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=ls(r,b,w),r=ls(r,0,b),u=ls(a,b,w),a=ls(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),N=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=N.slice().concat(N.map(function(F){return"val_"+F}))):(C=null,m=[],R=N.slice()),D=My(i.callbacks,i.yieldEvery),[4,S4(n,L,S,N,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,fr(r,t),fr(a,e),fr(l,s),fr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function ab(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e<n.length;++e){var i=n[e];if(i.rank===1)t.push(ts(i,1));else{if(i.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(i)}}return t}function fr(n,t){if(n==null)return;var e=[];if(t instanceof y.Tensor)e.push(t.id);else if(Array.isArray(t))t.forEach(function(l){return e.push(l.id)});else if(t!=null)for(var i in t){var r=t[i];e.push(r.id)}var a=[];if(n instanceof y.Tensor)e.indexOf(n.id)===-1&&a.push(n);else if(Array.isArray(n))n.forEach(function(l){e.indexOf(l.id)===-1&&a.push(l)});else if(n!=null)for(var s in n){var o=n[s];e.indexOf(o.id)===-1&&a.push(o)}a.forEach(function(l){l.isDisposed||l.dispose()})}function I4(n){return n instanceof y.Tensor}function fd(n){return Array.isArray(n)}function sb(n){return!I4(n)&&!fd(n)}function ob(n,t,e,i,r){if(i===void 0&&(i=!0),r===void 0&&(r=""),t==null||t.length===0){if(n!=null){var a=!1;if(fd(n)&&n.length>0)a=!0;else if(sb(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(sb(n)){n=n,o=[];for(var l=0,u=t;l<u.length;l++){var c=u[l];if(n[c]==null)throw new M('No data provided for "'+c+'". Need data for each key in: '+(""+t));o.push(n[c])}}else if(fd(n)){if(n=n,n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("model expected. Expected to see "+t.length+" Tensor(s), but ")+("instead got the following list of Tensor(s): "+n));o=n}else{if(n=n,t.length>1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=ab(o),e!=null)for(var h=0;h<t.length;++h){if(e[h]==null)continue;var d=o[h];if(d.shape.length!==e[h].length)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have "+e[h].length+" dimension(s). but got array with ")+("shape "+d.shape));for(var p=0;p<e[h].length;++p){if(p===0&&!i)continue;var f=d.shape[p],m=e[h][p];if(m!=null&&m>=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function A4(n,t,e){var i=Ti(n.map(function(a){return a.shape[0]}));i.sort();var r=Ti(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function T4(n,t,e){for(var i=[pr,Fo,as],r=0;r<n.length;++r){var a=n[r],s=t[r],o=e[r];if(s==null)continue;if(s===as&&a.shape[a.shape.length-1]===1)throw new M("You are passing a target array of shape "+a.shape+" while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].");if(i.indexOf(s)!==-1)for(var l=a.shape.slice(1),u=o.slice(1),c=0;c<l.length;++c){var h=l[c],d=u[c];if(d!=null&&h!==d)throw new M("A target Tensor with shape "+a.shape+" was passed for an "+("output of shape "+o+", while using a loss function that ")+"expects targets to have the same shape as the output.")}}}function lb(n,t,e,i,r){i===void 0&&(i=!0),r===void 0&&(r="");var a;if(Array.isArray(n)){if(n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("the model expected. Expected to see "+t.length+" Tensor(s),")+(" but instead got "+n.length+" Tensors(s)."));a=n}else{if(t.length>1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s<t.length;++s){if(e[s]==null)continue;var o=a[s];if(o.shape.length!==e[s].length)throw new M("Error when checking "+r+": expected "+t[s]+" "+("to have "+e[s].length+" dimension(s), but got array with ")+("shape "+JSON.stringify(o.shape)));for(var l=0;l<e[s].length;++l){if(l===0&&!i)continue;var u=o.shape[l],c=e[s][l];if(c!=null&&c!==u)throw new M("Error when checking "+r+": expected "+(t[s]+" to have shape "+JSON.stringify(e[s])+" but ")+("got array with shape "+JSON.stringify(o.shape)+"."))}}}function N4(n,t){if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(l){return[]});var e;if(typeof n=="string"||typeof n=="function")e=[n];else if(Array.isArray(n)||typeof n=="object")e=n;else throw new TypeError("Type of metrics argument not understood. Expected an string,"+("function, Array, or Object, found: "+n));if(Array.isArray(e))return t.map(function(l){return e});for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r],o=e.hasOwnProperty(s)?e[s]:[];Array.isArray(o)||(o=[o]),i.push(o)}return i}var x4="layers-model",Ri=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.isTraining=!1,i}return t.prototype.summary=function(e,i,r){if(r===void 0&&(r=console.log),!this.built)throw new M("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");o4(this,e,i,r)},t.prototype.compile=function(e){var i=this;if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=n4(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof y.Optimizer))throw new M("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}var r=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(var a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new M('Unknown entry in loss dictionary: "'+a+'". '+("Only expected the following keys: "+this.outputNames));for(var s=0,o=this.outputNames;s<o.length;s++){var l=o[s];e.loss[l]==null&&console.warn('Output "'+l+'" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data '+("to be passed to "+l+" during training")),r.push(id(e.loss[l]))}}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new M("When passing an Array as loss, it should have one entry per "+("model output. The model has "+this.outputs.length+" output(s), ")+("but you passed loss="+e.loss+"."));var u=e.loss;r=u.map(function(v){return id(v)})}else{var c=id(e.loss);this.outputs.forEach(function(v){r.push(c)})}this.lossFunctions=r,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(var h=0;h<this.outputs.length;++h){var d=this.internalOutputShapes[h],p=this.outputNames[h];this.feedOutputNames.push(p),this.feedOutputShapes.push(d),this.feedLossFns.push(this.lossFunctions[h])}var f=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],hr("loss",function(){for(var v=0;v<i.outputs.length;++v){if(f.indexOf(v)!==-1)continue;var b=i.lossFunctions[v];i.outputs.length>1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=N4(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};hr("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(N){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var j=i.internalOutputShapes[w];j[j.length-1]===1||i.lossFunctions[w]===Fo?["accuracy","acc"].indexOf(_)!==-1?D=rd:["crossentropy","ce"].indexOf(_)!==-1&&(D=Yy):i.lossFunctions[w]===ko?["accuracy","acc"].indexOf(_)!==-1?D=jy:["crossentropy","ce"].indexOf(_)!==-1&&(D=Ky):["accuracy","acc"].indexOf(_)!==-1?D=ad:["crossentropy","ce"].indexOf(_)!==-1&&(D=sd);var K=void 0;["accuracy","acc"].indexOf(_)!==-1?K="acc":["crossentropy","ce"].indexOf(_)!==-1&&(K="ce"),k=D,R=C+K}else{var q=t4(_);k=q,R=C+Bo(_)}var G;hr(R,function(){G=k}),g(w,R,G)},F=0,P=N;F<P.length;F++){var H=P[F];W(H)}};L(S)},b=0;b<i.outputs.length;++b)v(b)}),this.collectedTrainableWeights=this.trainableWeights},t.prototype.checkTrainableWeightsConsistency=function(){if(this.collectedTrainableWeights==null)return;this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")},t.prototype.evaluate=function(e,i,r){r===void 0&&(r={});var a=r.batchSize==null?32:r.batchSize;hd(a);var s=!0,o=this.standardizeUserDataXY(e,i,s,a);try{var l=o[0].concat(o[1]);this.makeTestFunction();var u=this.testFunction,c=this.testLoop(u,l,a,r.verbose,r.steps);return qt(c)}finally{fr(o[0],e),fr(o[1],i)}},t.prototype.evaluateDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){return this.makeTestFunction(),[2,w4(this,e,i)]})})},t.prototype.checkNumSamples=function(e,i,r,a){a===void 0&&(a="steps");var s;if(r!=null){if(s=null,i!=null)throw new M("If "+a+" is set, batchSize must be null or undefined."+("Got batchSize = "+i))}else if(e!=null)Array.isArray(e)?s=e[0].shape[0]:s=e.shape[0];else throw new M("Either the input data should have a defined shape, or "+(a+" shoud be specified."));return s},t.prototype.execute=function(e,i){if(Array.isArray(i)&&i.length===0)throw new M("`outputs` is an empty Array, which is not allowed.");var r=Array.isArray(i),a=r?i:[i],s=this.retrieveSymbolicTensors(a),o=new aa;if(e instanceof y.Tensor&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new M("The number of inputs provided ("+e.length+") does not match the number of inputs of this model "+("("+this.inputs.length+")."));for(var l=0;l<this.inputs.length;++l)o.add(this.inputs[l],e[l])}else for(var u=0,c=this.inputs;u<c.length;u++){var h=c[u],d=e[h.name];if(d==null)throw new M("No value is provided for the model's input "+h.name);o.add(h,d)}var p=os(s,o);return r?p:p[0]},t.prototype.retrieveSymbolicTensors=function(e){for(var i=ur(null,e.length),r=e.length,a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=Array.isArray(o.output)?o.output:[o.output],u=l.map(function(p){return p.name}),c=0;c<e.length;++c){var h=u.indexOf(e[c]);if(h!==-1&&(i[c]=l[h],r--),r===0)break}if(r===0)break}if(r>0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=pd(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=ls(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v<m.length;++v)g.push({key:a.inputs[v],value:m[v]});else g.push({key:a.inputs[0],value:m});var b=new aa(g);return os(a.outputs,b)});d.forEach(function(p,f){return l[f].push(p)})},c=0;c<o.length;++c)u(c);return qt(l.map(function(h){return y.concat(h,0)}))})},t.prototype.predict=function(e,i){i===void 0&&(i={});var r=ab(e);lb(r,this.inputNames,this.feedInputShapes,!1);try{var a=i.batchSize==null?32:i.batchSize;return hd(a),this.predictLoop(r,a)}finally{fr(r,e)}},t.prototype.predictOnBatch=function(e){lb(e,this.inputNames,this.feedInputShapes,!0);var i=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,i)},t.prototype.standardizeUserDataXY=function(e,i,r,a){if(this.optimizer_==null)throw new Pn("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");for(var s=[],o=0;o<this.feedOutputShapes.length;++o){var l=this.feedOutputShapes[o],u=this.feedLossFns[o];u===ko?s.push(l.slice(0,l.length-1).concat([1])):s.push(l)}if(e=ob(e,this.feedInputNames,this.feedInputShapes,!1,"input"),i=ob(i,this.feedOutputNames,s,!1,"target"),A4(e,i),T4(i,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Se(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=eb(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p<d.length?(m=(f=h).push,[4,tb(c[p],null,d[p])]):[3,4];case 2:m.apply(f,[g.sent()]),g.label=3;case 3:return++p,[3,1];case 4:return[2,[u,c,h]]}})})},t.prototype.testLoop=function(e,i,r,a,s){var o=this;return a===void 0&&(a=0),y.tidy(function(){var l=o.checkNumSamples(i,r,s,"steps"),u=[];if(a>0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=pd(l,r),h=y.tensor1d(Sn(0,l)),d=0;d<c.length;++d){var p=c[d][0],f=c[d][1],m=dr(h,p,f-p),g=dd(i,m),v=e(g);if(d===0)for(var b=0;b<v.length;++b)u.push(y.scalar(0));for(var b=0;b<v.length;++b){var w=v[b];u[b]=y.add(u[b],y.mul(f-p,w))}}for(var b=0;b<u.length;++b)u[b]=y.div(u[b],l);return u})},t.prototype.getDedupedMetricsNames=function(){for(var e=this.metricsNames,i=[],r=0;r<e.length;++r){var a=e[r],s=a;if(ry(e,a)>1){var o=ry(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f<e.inputs.length;++f)p.push({key:e.inputs[f],value:a[f]});for(var m=new aa(p),g=os(e.outputs,m,{training:!0}),v,f=0;f<e.lossFunctions.length;++f){var b=e.lossFunctions[f],w=b(s[f],g[f]);o[f]!=null&&(w=f4(w,o[f]));var S=y.mean(w);r.push(S),f===0?v=w:v=y.add(v,w)}for(var f=0;f<e.metricsTensors.length;++f){var L=void 0;if(e.outputs.length>1&&f<e.outputs.length)L=r[f];else{var N=e.metricsTensors[f][0],C=e.metricsTensors[f][1];L=y.mean(N(s[C],g[C]))}y.keep(L),l.push(L)}return v=y.mean(v),e.calculateLosses().forEach(function(R){v=y.add(v,R)}),v},c=e.collectedTrainableWeights.map(function(p){return p.read()}),h=!0,d=e.optimizer_.minimize(u,h,c);return[d].concat(l)}},t.prototype.makeTestFunction=function(){var e=this;this.testFunction=function(i){return y.tidy(function(){for(var r=[],a,s=i.slice(0,e.inputs.length),o=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),l=[],u=0;u<e.inputs.length;++u)l.push({key:e.inputs[u],value:s[u]});for(var c=new aa(l),h=os(e.outputs,c),u=0;u<e.lossFunctions.length;++u){var d=e.lossFunctions[u],p=y.mean(d(o[u],h[u]));u===0?a=p:a=y.add(a,p),r.push(a)}for(var u=0;u<e.metricsTensors.length;++u){var f=e.metricsTensors[u][0],m=e.metricsTensors[u][1],g=y.mean(f(o[m],h[m]));r.push(g)}return r})}},t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Se(this,void 0,void 0,function(){return ve(this,function(a){return[2,L4(this,e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){return[2,y4(this,e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:return[4,this.standardizeUserData(e,i)];case 1:r=f.sent(),a=r[0],s=r[1],o=this.makeTrainFunction(),l=o(a.concat(s)),u=[],c=0,h=l,f.label=2;case 2:return c<h.length?(d=h[c],[4,d.data()]):[3,5];case 3:p=f.sent(),u.push(p[0]),f.label=4;case 4:return c++,[3,2];case 5:return y.dispose(l),[2,qt(u)]}})})},t.prototype.getNamedWeights=function(e){for(var i=[],r=e!=null&&e.trainableOnly,a=r?this.trainableWeights:this.weights,s=this.getWeights(r),o=0;o<a.length;++o){if(r&&!a[o].trainable)continue;i.push({name:a[o].originalName,tensor:s[o]})}return i},Object.defineProperty(t.prototype,"stopTraining",{get:function(){return this.stopTraining_},set:function(e){this.stopTraining_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.optimizer_},set:function(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)},enumerable:!0,configurable:!0}),t.prototype.dispose=function(){var e=n.prototype.dispose.call(this);if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){var i=y.memory().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=i-y.memory().numTensors}return e},t.prototype.getLossIdentifiers=function(){var e;if(typeof this.loss=="string")e=ai(this.loss);else if(Array.isArray(this.loss)){for(var i=0,r=this.loss;i<r.length;i++){var a=r[i];if(typeof a!="string")throw new Error("Serialization of non-string loss is not supported.")}e=this.loss.map(function(h){return ai(h)})}else{var s=Object.keys(this.loss);e={};for(var o=this.loss,l=0,u=s;l<u.length;l++){var c=u[l];if(typeof o[c]=="string")e[c]=ai(o[c]);else throw new Error("Serialization of non-string loss is not supported.")}}return e},t.prototype.getMetricIdentifiers=function(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[ai(Bo(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(function(r){return ai(Bo(r))});var e={};for(var i in this.metrics)e[i]=ai(Bo(this.metrics[i]));return e},t.prototype.getTrainingConfig=function(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}},t.prototype.loadTrainingConfig=function(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");var i=ss(e.optimizer_config),r=In(i),a;if(typeof e.loss=="string")a=cr(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(function(l){return cr(l)});else if(e.loss!=null){a={};for(var s in e.loss)a[s]=cr(e.loss[s])}var o;if(Array.isArray(e.metrics))o=e.metrics.map(function(l){return cr(l)});else if(e.metrics!=null){o={};for(var s in e.metrics)o[s]=cr(e.metrics[s])}this.compile({loss:a,metrics:o,optimizer:r})},t.prototype.save=function(e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b;return ve(this,function(w){switch(w.label){case 0:if(typeof e=="string"){if(r=y.io.getSaveHandlers(e),r.length===0)throw new M("Cannot find any save handlers for URL '"+e+"'");if(r.length>1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:x4,generatedBy:"TensorFlow.js tfjs-layers v"+ud,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,Xy(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){Xy(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(d4);y.serialization.registerClass(Ri);var C4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(Ri);y.serialization.registerClass(C4);function R4(n,t){return Se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=ss(e),r=In(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o<l.length;o++)u=l[o],s[u.originalName]=a[u.originalName];r.loadWeights(s),y.dispose(a),c.label=2;case 2:return[2,r]}})})}function E4(n,t){return Se(this,void 0,void 0,function(){var e;return ve(this,function(i){if(t==null&&(t={}),typeof n=="string"){if(e=y.io.getLoadHandlers(n,t),e.length===0)e.push(y.io.browserHTTPRequest(n,t));else if(e.length>1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,O4(n,void 0,t)]})})}function O4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=In(ss(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=D4(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function D4(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var md=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:xo("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r<a.length;r++){var s=a[r];i.add(s)}return i}return t.prototype.checkShape=function(e){var i=e.inboundNodes[0].outputTensors[0].shape;if(i.some(function(r){return r<0}))throw new M("Negative dimension size caused by adding layer "+(e.name+" with input shape [")+(e.inboundNodes[0].inputTensors[0].shape+"]"))},t.prototype.add=function(e){var i=e instanceof t||e instanceof Ri,r;if(i){if(r=e,r.outputs.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(r.inputs.length!==1)throw new M("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new M("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");var a=Wy({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(i)this.outputs=r.outputs,this.inputs=r.inputs;else{if(e.inboundNodes.length!==1)throw new M("A layer added to a Sequential model must not already be "+("connected somewhere else. LayersModel received layer "+e.name+" ")+("which has "+e.inboundNodes.length+" pre-existing inbound ")+"connections.");if(e.inboundNodes[0].outputTensors.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Fy(this.outputs[0])}this.inboundNodes=[],new Oo({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:ur(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(function(o){return o.shape}),outputShapes:this.outputs[0].shape})}else{var s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1},t.prototype.pop=function(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{var e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}},t.prototype.call=function(e,i){return this.model==null&&this.build(),this.model.call(e,i)},t.prototype.build=function(e){if(je(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new Ri({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0},t.prototype.countParams=function(){return this.built||this.build(),n.prototype.countParams.call(this)},t.prototype.summary=function(e,i,r){r===void 0&&(r=console.log),this.built||this.build(),n.prototype.summary.call(this,e,i,r)},t.prototype.setWeights=function(e){this.model==null&&this.build(),this.model.setWeights(e)},t.prototype.evaluate=function(e,i,r){if(r===void 0&&(r={}),!this.built)throw new Pn("The model needs to be compiled before being used.");return this.model.evaluate(e,i,r)},t.prototype.evaluateDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Pn("The model needs to be compiled before being used.");return[2,this.model.evaluateDataset(e,i)]})})},t.prototype.predict=function(e,i){return i===void 0&&(i={}),this.model==null&&this.build(),this.model.predict(e,i)},t.prototype.predictOnBatch=function(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)},t.prototype.compile=function(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames},Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.model==null?void 0:this.model.optimizer},set:function(e){this.model.optimizer=e},enumerable:!0,configurable:!0}),t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Se(this,void 0,void 0,function(){return ve(this,function(a){if(!this.built)throw new Pn("The model needs to be compiled before being used.");return[2,this.model.fit(e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Pn("The model needs to be compiled before being used.");return[2,this.model.fitDataset(e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Se(this,void 0,void 0,function(){return ve(this,function(r){return[2,this.model.trainOnBatch(e,i)]})})},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s,o={};if(i instanceof Array){if(!(i[0].className!=null)||i[0].className==="Merge")throw new M("Legacy serialization format not supported yet.");s=i}else y.util.assert(i.layers!=null,function(){return"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."}),s=i.layers,delete i.layers,o=i;var l=new e(o);if(!(l instanceof t))throw new Te("Sequential.fromConfig called on non-Sequential input: "+l);for(var u=0,c=s;u<c.length;u++){var h=c[u],d=void 0,p=In(h,d,a);a&&p.setFastWeightInitDuringBuild(!0),l.add(p)}return l},Object.defineProperty(t.prototype,"stopTraining",{get:function(){if(this.model==null)throw new M("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining},set:function(e){if(this.model==null)throw new M("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e},enumerable:!0,configurable:!0}),t.prototype.getConfig=function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i],s={};s.className=a.getClassName(),s.config=a.getConfig(),e.push(s)}return{name:this.name,layers:e}},t.className="Sequential",t}(Ri);y.serialization.registerClass(md);function k4(n){return new Ri(n)}function F4(n){return new md(n)}function W4(n,t){return t==null&&(t={}),E4(n,t)}function ub(n){return Wy(n)}function U4(n,t){Hy.registerCallbackConstructor(n,t)}var Xt=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),B4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),a3(e,i)},t.className="elu",t}(Xt);y.serialization.registerClass(B4);var z4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.selu(e)},t.className="selu",t}(Xt);y.serialization.registerClass(z4);var P4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="relu",t}(Xt);y.serialization.registerClass(P4);var _4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tidy(function(){return y.minimum(6,y.relu(e))})},t.className="relu6",t}(Xt);y.serialization.registerClass(_4);var M4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return e},t.className="linear",t}(Xt);y.serialization.registerClass(M4);var H4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.sigmoid(e)},t.className="sigmoid",t}(Xt);y.serialization.registerClass(H4);var V4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return o3(e)},t.className="hardSigmoid",t}(Xt);y.serialization.registerClass(V4);var q4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.softplus(e)},t.className="softplus",t}(Xt);y.serialization.registerClass(q4);var G4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return s3(e)},t.className="softsign",t}(Xt);y.serialization.registerClass(G4);var Y4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tanh(e)},t.className="tanh",t}(Xt);y.serialization.registerClass(Y4);var cb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.softmax(e,i)},t.className="softmax",t}(Xt);y.serialization.registerClass(cb);var j4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.logSoftmax(e,i)},t.className="logSoftmax",t}(Xt);y.serialization.registerClass(j4);var K4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),y.tidy(function(){return y.sigmoid(e.mul(i)).mul(e)})},t.className="swish",t}(Xt);y.serialization.registerClass(K4);function Oi(n){return n.getClassName()}function gd(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"activation")}function Ei(n){if(n==null){var t={};return t.className="linear",t.config={},gd(t)}if(typeof n=="string"){var t={};return t.className=n,t.config={},gd(t)}else return n instanceof Xt?n:gd(n)}function vd(n){if(n!=null&&typeof n!="object")throw new Error("Argument to L1L2 regularizer's constructor is expected to be an "+("object, but received: "+n))}var hb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(y.serialization.Serializable),Po=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return vd(e),i.l1=e==null||e.l1==null?.01:e.l1,i.l2=e==null||e.l2==null?.01:e.l2,i.hasL1=i.l1!==0,i.hasL2=i.l2!==0,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=y.zeros([1]);return i.hasL1&&(r=y.add(r,y.sum(y.mul(i.l1,y.abs(e))))),i.hasL2&&(r=y.add(r,y.sum(y.mul(i.l2,ns(e))))),r.asScalar()})},t.prototype.getConfig=function(){return{l1:this.l1,l2:this.l2}},t.fromConfig=function(e,i){return new e({l1:i.l1,l2:i.l2})},t.className="L1L2",t}(hb);y.serialization.registerClass(Po);function $4(n){return vd(n),new Po({l1:n!=null?n.l1:null,l2:0})}function X4(n){return vd(n),new Po({l2:n!=null?n.l2:null,l1:0})}var db={l1l2:"L1L2"};function Ke(n){return zh(n)}function pb(n,t){return t===void 0&&(t={}),Ja(n,y.serialization.SerializationMap.getMap().classNameMap,t,"regularizer")}function nt(n){if(n==null)return null;if(typeof n=="string"){var t=n in db?db[n]:n,e={className:t,config:{}};return pb(e)}else return n instanceof hb?n:pb(n)}var fb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null&&(i.maxValue=e.maxValue),i}return t.prototype.call=function(e,i){e=xe(e);var r=y.relu(e);return this.maxValue!=null&&(r=y.clipByValue(r,0,this.maxValue)),r},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={maxValue:this.maxValue},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ReLU",t}(De);y.serialization.registerClass(fb);var mb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_ALPHA=.3,e==null&&(e={}),i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.leakyRelu(r,this.alpha)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LeakyReLU",t}(De);y.serialization.registerClass(mb);var gb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),i.supportsMasking=!0,i.alphaInitializer=tt(e.alphaInitializer||i.DEFAULT_ALPHA_INITIALIZER),i.alphaRegularizer=nt(e.alphaRegularizer),i.alphaConstraint=wt(e.alphaConstraint),e.sharedAxes==null)i.sharedAxes=null;else if(Array.isArray(e.sharedAxes))i.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")i.sharedAxes=[e.sharedAxes];else throw new M("Expected sharedAxes to be a number or an array of numbers, "+("but got "+e.sharedAxes));return i}return t.prototype.build=function(e){e=je(e);var i=e.slice(1);if(this.sharedAxes!=null)for(var r=0,a=this.sharedAxes;r<a.length;r++){var s=a[r];i[s-1]=1}this.alpha=this.addWeight("alpha",i,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);var o={};if(this.sharedAxes!=null)for(var s=1;s<e.length;++s)o[s]=e[s];this.inputSpec=[new Nt({ndim:e.length,axes:o})],this.built=!0},t.prototype.call=function(e,i){return e=xe(e),y.prelu(e,this.alpha.read())},t.prototype.getConfig=function(){var e={alphaInitializer:st(this.alphaInitializer),alphaRegularizer:Ke(this.alphaRegularizer),alphaConstraint:bt(this.alphaConstraint),sharedAxes:this.sharedAxes},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="PReLU",t}(De);y.serialization.registerClass(gb);var vb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==i.DEFAULT_ALPHA)throw new Te("Non-default alpha value ("+e.alpha+") is not supported by the ELU layer yet.");return i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.elu(r)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ELU",t}(De);y.serialization.registerClass(vb);var yb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_THETA=1,e==null&&(e={}),i.theta=e.theta==null?i.DEFAULT_THETA:e.theta,i}return t.prototype.call=function(e,i){var r=xe(e);return r.mul(es(r.greater(this.theta),"float32"))},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={theta:this.theta},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ThresholdedReLU",t}(De);y.serialization.registerClass(yb);var bb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_AXIS=1,e==null&&(e={}),i.softmax=new cb().apply,i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i}return t.prototype.call=function(e,i){var r=xe(e);return this.softmax(r,this.axis)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Softmax",t}(De);y.serialization.registerClass(bb);function sa(n,t,e){if(typeof n=="number")return ur(n,t);if(n.length!==t)throw new M("The "+e+" argument must be an integer or tuple of "+t+" integers."+(" Received: "+n.length+" elements."));for(var i=0;i<t;++i){var r=n[i];if(!t3(r))throw new M("The "+e+" argument must be an integer or tuple of "+t+(" integers. Received: "+JSON.stringify(n)+" including a")+(" non-integer number "+r))}return n}function An(n,t,e,i,r){if(r===void 0&&(r=1),n==null)return n;var a=t+(t-1)*(r-1),s;return e==="same"?s=n:s=n-a+1,Math.floor((s+i-1)/i)}function _o(n,t,e,i){if(n==null)return null;if(i==="valid")n=n*t+xi([e-t,0]);else if(i==="same")n=n*t;else throw new M("Unsupport padding mode: "+i+".");return n}function yd(n,t){return y.tidy(function(){return ht(t),t==="channelsFirst"?y.transpose(n,[0,2,3,1]):n})}function wb(n,t){return y.tidy(function(){return ht(t),t==="channelsFirst"?y.transpose(n,[0,2,3,4,1]):n})}function J4(n,t,e,i,r,a,s){return i===void 0&&(i=1),r===void 0&&(r="valid"),s===void 0&&(s=1),y.tidy(function(){if(a==null&&(a=wn()),ht(a),n.shape.length!==3)throw new M("The input of a conv1dWithBias operation should be 3, but is "+(n.shape.length+" instead."));if(t.shape.length!==3)throw new M("The kernel for a conv1dWithBias operation should be 3, but is "+(t.shape.length+" instead"));if(e!=null&&e.shape.length!==1)throw new M("The bias for a conv1dWithBias operation should be 1, but is "+(t.shape.length+" instead"));if(a==="channelsFirst"&&(n=y.transpose(n,[0,2,1])),r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");var o=y.conv1d(n,t,i,r==="same"?"same":"valid","NWC",s);return e!=null&&(o=Hn(o,e)),o})}function Sb(n,t,e,i,r,a,s,o){return i===void 0&&(i=[1,1]),r===void 0&&(r="valid"),o===void 0&&(o=null),y.tidy(function(){if(a==null&&(a=wn()),ht(a),n.rank!==3&&n.rank!==4)throw new M("conv2dWithBiasActivation expects input to be of rank 3 or 4, "+("but received "+n.rank+"."));if(t.rank!==3&&t.rank!==4)throw new M("conv2dWithBiasActivation expects kernel to be of rank 3 or 4, "+("but received "+n.rank+"."));var l=yd(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=y.fused.conv2d({x:l,filter:t,strides:i,pad:r==="same"?"same":"valid",dilations:s,dataFormat:"NHWC",bias:e,activation:o}),a==="channelsFirst"&&(l=y.transpose(l,[0,3,1,2])),l})}function Z4(n,t,e,i,r,a,s){return i===void 0&&(i=[1,1,1]),r===void 0&&(r="valid"),y.tidy(function(){if(a==null&&(a=wn()),ht(a),n.rank!==4&&n.rank!==5)throw new M("conv3dWithBias expects input to be of rank 4 or 5, but received "+(n.rank+"."));if(t.rank!==4&&t.rank!==5)throw new M("conv3dWithBias expects kernel to be of rank 4 or 5, but received "+(n.rank+"."));var o=wb(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=y.conv3d(o,t,i,r==="same"?"same":"valid","NDHWC",s),e!=null&&(o=Hn(o,e)),a==="channelsFirst"&&(o=y.transpose(o,[0,4,1,2,3])),o})}var Lb=function(n){Q(t,n);function t(e,i){var r=n.call(this,i)||this;if(r.bias=null,r.DEFAULT_KERNEL_INITIALIZER="glorotNormal",r.DEFAULT_BIAS_INITIALIZER="zeros",t.verifyArgs(i),r.rank=e,Tt(r.rank,"rank"),r.rank!==1&&r.rank!==2&&r.rank!==3)throw new Te("Convolution layer for rank other than 1, 2, or 3 ("+r.rank+") is not implemented yet.");if(r.kernelSize=sa(i.kernelSize,e,"kernelSize"),r.strides=sa(i.strides==null?1:i.strides,e,"strides"),r.padding=i.padding==null?"valid":i.padding,an(r.padding),r.dataFormat=i.dataFormat==null?"channelsLast":i.dataFormat,ht(r.dataFormat),r.activation=Ei(i.activation),r.useBias=i.useBias==null?!0:i.useBias,r.biasInitializer=tt(i.biasInitializer||r.DEFAULT_BIAS_INITIALIZER),r.biasConstraint=wt(i.biasConstraint),r.biasRegularizer=nt(i.biasRegularizer),r.activityRegularizer=nt(i.activityRegularizer),r.dilationRate=sa(i.dilationRate==null?1:i.dilationRate,e,"dilationRate"),r.rank===1&&Array.isArray(r.dilationRate)&&r.dilationRate.length!==1)throw new M("dilationRate must be a number or an array of a single number for 1D convolution, but received "+(""+JSON.stringify(r.dilationRate)));if(r.rank===2){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==2)throw new M("dilationRate must be a number or array of two numbers for 2D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}else if(r.rank===3){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==3)throw new M("dilationRate must be a number or array of three numbers for 3D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}return r}return t.verifyArgs=function(e){if(_n("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!_h(e.kernelSize,"number",1,3))throw new M("BaseConv expects config.kernelSize to be number or number[] with "+("length 1, 2, or 3, but received "+JSON.stringify(e.kernelSize)+"."))},t.prototype.getConfig=function(){var e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:Oi(this.activation),useBias:this.useBias,biasInitializer:st(this.biasInitializer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),biasConstraint:bt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Mo=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;return r.kernel=null,t.verifyArgs(i),r.filters=i.filters,Tt(r.filters,"filters"),r.kernelInitializer=tt(i.kernelInitializer||r.DEFAULT_KERNEL_INITIALIZER),r.kernelConstraint=wt(i.kernelConstraint),r.kernelRegularizer=nt(i.kernelRegularizer),r}return t.prototype.build=function(e){var i;e=je(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=this.kernelSize.concat([a,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:(i={},i[r]=a,i)}],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a,s=r.bias==null?null:r.bias.read(),o=sy(r.activation.getClassName());if(o!=null&&r.rank===2)a=Sb(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate,o);else{if(r.rank===1)a=J4(e,r.kernel.read(),s,r.strides[0],r.padding,r.dataFormat,r.dilationRate[0]);else if(r.rank===2)a=Sb(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else if(r.rank===3)a=Z4(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else throw new Te("convolutions greater than 3D are not implemented yet.");r.activation!=null&&(a=r.activation.apply(a))}return a})},t.prototype.computeOutputShape=function(e){e=je(e);for(var i=[],r=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2),a=0;a<r.length;++a){var s=An(r[a],this.kernelSize[a],this.padding,this.strides[a],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[a]);i.push(s)}var o=[e[0]];return this.dataFormat==="channelsLast"?(o=o.concat(i),o.push(this.filters)):(o.push(this.filters),o=o.concat(i)),o},t.prototype.getConfig=function(){var e={filters:this.filters,kernelInitializer:st(this.kernelInitializer),kernelRegularizer:Ke(this.kernelRegularizer),kernelConstraint:bt(this.kernelConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.verifyArgs=function(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new M("Convolution layer expected config.filters to be a 'number' > 0 "+("but got "+JSON.stringify(e.filters)))},t}(Lb),bd=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!_h(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(Mo);y.serialization.registerClass(bd);var Ib=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(Mo);y.serialization.registerClass(Ib);var Ab=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Nt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=je(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Nt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=_o(c,f,d,r.padding),v=_o(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Hn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=je(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=_o(i[a],u,o,this.padding),i[s]=_o(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(bd);y.serialization.registerClass(Ab);var Q4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=nt(i.depthwiseRegularizer),r.depthwiseConstraint=wt(i.depthwiseConstraint),r.pointwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=nt(i.pointwiseRegularizer),r.pointwiseConstraint=wt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=je(e),e.length<this.rank+2)throw new M("Inputs to SeparableConv"+this.rank+"D should have rank "+(this.rank+2+", but received input shape: ")+(""+JSON.stringify(e)));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null||e[r]<0)throw new M("The channel dimension of the inputs should be defined, "+("but found "+JSON.stringify(e[r])));for(var a=e[r],s=this.kernelSize.concat([a,this.depthMultiplier]),o=[],l=0;l<this.rank;++l)o.push(1);o.push(a*this.depthMultiplier,this.filters);var u=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,u,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",o,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,u,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,u,this.biasConstraint):this.bias=null,this.inputSpec=[new Nt({ndim:this.rank+2,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a;if(r.rank===1)throw new Te("1D separable convolution is not implemented yet.");return r.rank===2&&(r.dataFormat==="channelsFirst"&&(e=y.transpose(e,[0,2,3,1])),a=y.separableConv2d(e,r.depthwiseKernel.read(),r.pointwiseKernel.read(),r.strides,r.padding,r.dilationRate,"NHWC")),r.useBias&&(a=Hn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),r.dataFormat==="channelsFirst"&&(a=y.transpose(a,[0,3,1,2])),a})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=st(this.depthwiseInitializer),e.pointwiseInitializer=st(this.pointwiseInitializer),e.depthwiseRegularizer=Ke(this.depthwiseRegularizer),e.pointwiseRegularizer=Ke(this.pointwiseRegularizer),e.depthwiseConstraint=bt(this.depthwiseConstraint),e.pointwiseConstraint=bt(this.pointwiseConstraint),e},t.className="SeparableConv",t}(Mo),Tb=function(n){Q(t,n);function t(e){return n.call(this,2,e)||this}return t.className="SeparableConv2D",t}(Q4);y.serialization.registerClass(Tb);var Nb=function(n){Q(t,n);function t(e){var i=n.call(this,1,e)||this;return t.verifyArgs(e),i.inputSpec=[{ndim:3}],i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.dataFormat,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!_h(e.kernelSize,"number",1,1))throw new M("Conv1D expects config.kernelSize to be number or number[] with "+("length 1, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv1D",t}(Mo);y.serialization.registerClass(Nb);var xb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return typeof e.cropping=="number"?i.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?i.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:i.cropping=e.cropping,i.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,i.inputSpec=[{ndim:4}],i}return t.prototype.computeOutputShape=function(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=xe(e),r.dataFormat==="channelsLast"){var a=Ao(e,r.cropping[0][0],e.shape[1]-r.cropping[0][0]-r.cropping[0][1],2);return Ao(a,r.cropping[1][0],e.shape[2]-r.cropping[1][1]-r.cropping[1][0],3)}else{var a=Ao(e,r.cropping[0][0],e.shape[2]-r.cropping[0][0]-r.cropping[0][1],3);return Ao(a,r.cropping[1][0],e.shape[3]-r.cropping[1][1]-r.cropping[1][0],4)}})},t.prototype.getConfig=function(){var e={cropping:this.cropping,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Cropping2D",t}(De);y.serialization.registerClass(xb);var Cb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_SIZE=[2,2],i.inputSpec=[{ndim:4}],i.size=e.size==null?i.DEFAULT_SIZE:e.size,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,i}return t.prototype.computeOutputShape=function(e){if(this.dataFormat==="channelsFirst"){var i=e[2]==null?null:this.size[0]*e[2],r=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],i,r]}else{var i=e[1]==null?null:this.size[0]*e[1],r=e[2]==null?null:this.size[1]*e[2];return[e[0],i,r,e[3]]}},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e),s=a.shape;if(r.dataFormat==="channelsFirst"){a=y.transpose(a,[0,2,3,1]);var o=r.size[0]*s[2],l=r.size[1]*s[3],u=a.resizeNearestNeighbor([o,l]);return y.transpose(u,[0,3,1,2])}else{var o=r.size[0]*s[1],l=r.size[1]*s[2];return a.resizeNearestNeighbor([o,l])}})},t.prototype.getConfig=function(){var e={size:this.size,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="UpSampling2D",t}(De);y.serialization.registerClass(Cb);function eF(n,t,e,i,r,a){return e===void 0&&(e=[1,1]),i===void 0&&(i="valid"),y.tidy(function(){r==null&&(r=wn()),ht(r);var s=yd(n,r);if(n.rank!==4)throw new M("Input for depthwiseConv2d is required to be 4-D, but is instead "+(n.rank+"-D"));if(t.rank!==4)throw new M("depthwiseKernel is required to be 4-D, but is instead "+(t.rank+"-D"));return s=y.depthwiseConv2d(s,t,e,i==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}var Rb=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return i.depthwiseKernel=null,i.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,i.depthwiseInitializer=tt(e.depthwiseInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.depthwiseConstraint=wt(e.depthwiseConstraint),i.depthwiseRegularizer=nt(e.depthwiseRegularizer),i}return t.prototype.build=function(e){if(e=je(e),e.length<4)throw new M("Inputs to DepthwiseConv2D should have rank 4. "+("Received input shape: "+JSON.stringify(e)+"."));var i=this.dataFormat==="channelsFirst"?1:3;if(e[i]==null||e[i]<0)throw new M("The channel dimension of the inputs to DepthwiseConv2D should "+("be defined, but is not ("+e[i]+")."));var r=e[i],a=[this.kernelSize[0],this.kernelSize[1],r,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[r*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=eF(e,r.depthwiseKernel.read(),r.strides,r.padding,r.dataFormat,null);return r.useBias&&(a=Hn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),a})},t.prototype.computeOutputShape=function(e){e=je(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,s=An(i,this.kernelSize[0],this.padding,this.strides[0]),o=An(r,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,s,o]:[e[0],s,o,a]},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=st(this.depthwiseInitializer),e.depthwiseRegularizer=Ke(this.depthwiseRegularizer),e.depthwiseConstraint=bt(this.depthwiseRegularizer),e},t.className="DepthwiseConv2D",t}(Lb);y.serialization.registerClass(Rb);function Ob(n,t,e,i){if(Array.isArray(n)){if(t!=null||e!=null)throw new M("When inputs is an array, neither initialState or constants should be provided");i!=null&&(e=n.slice(n.length-i,n.length),n=n.slice(0,n.length-i)),n.length>1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Eb(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(Sn(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],N=y.tidy(function(){return n(L,d)});if(r==null)h=N[0],d=N[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=N[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return N[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v<p;++v)g(v);var b;if(o){var w=1;b=y.stack(c,w)}return[h,b,d]})}var Di=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r;if(e.cell==null)throw new M("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?r=new wd({cells:e.cell}):r=e.cell,r.stateSize==null)throw new M("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");return i.cell=r,i.returnSequences=e.returnSequences==null?!1:e.returnSequences,i.returnState=e.returnState==null?!1:e.returnState,i.goBackwards=e.goBackwards==null?!1:e.goBackwards,i._stateful=e.stateful==null?!1:e.stateful,i.unroll=e.unroll==null?!1:e.unroll,i.supportsMasking=!0,i.inputSpec=[new Nt({ndim:3})],i.stateSpec=null,i.states_=null,i.numConstants=null,i.keptStates=[],i}return t.prototype.getStates=function(){if(this.states_==null){var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return Sn(0,e).map(function(i){return null})}else return this.states_},t.prototype.setStates=function(e){this.states_=e},t.prototype.computeOutputShape=function(e){Qh(e)&&(e=e[0]),e=e;var i=this.cell.stateSize;Array.isArray(i)||(i=[i]);var r=i[0],a;if(this.returnSequences?a=[e[0],e[1],r]:a=[e[0],r],this.returnState){for(var s=[],o=0,l=i;o<l.length;o++){var u=l[o];s.push([e[0],u])}return[a].concat(s)}else return a},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){Array.isArray(i)&&(i=i[0]);var a=r.returnSequences?i:null;if(r.returnState){var s=r.states.map(function(o){return null});return[a].concat(s)}else return a})},Object.defineProperty(t.prototype,"states",{get:function(){if(this.states_==null){for(var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,i=[],r=0;r<e;++r)i.push(null);return i}else return this.states_},set:function(e){this.states_=e},enumerable:!0,configurable:!0}),t.prototype.build=function(e){if(this.numConstants!=null)throw new Te("Constants support is not implemented in RNN yet.");Qh(e)&&(e=e[0]),e=e;var i=this.stateful?e[0]:null,r=e.slice(2);this.inputSpec[0]=new Nt({shape:[i,null].concat(r)});var a=[e[0]].concat(e.slice(2));this.cell.build(a);var s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!y.util.arraysEqual(this.stateSpec.map(function(o){return o.shape[o.shape.length-1]}),s))throw new M("An initialState was passed that is not compatible with "+("cell.stateSize. Received stateSpec="+this.stateSpec+"; ")+("However cell.stateSize is "+this.cell.stateSize))}else this.stateSpec=s.map(function(o){return new Nt({shape:[null,o]})});this.stateful&&this.resetStates()},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new Ai("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape[0];if(a==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.states_==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_=[y.zeros([a,r.cell.stateSize])];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_[0]=y.zeros([a,r.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i===!0?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var s=0;s<r.states_.length;++s){var o=e[s],l=Array.isArray(r.cell.stateSize)?r.cell.stateSize[s]:r.cell.stateSize,u=[a,l];if(!y.util.arraysEqual(o.shape,u))throw new M("State "+s+" is incompatible with layer "+r.name+": "+("expected shape="+u+", received shape="+o.shape));r.states_[s]=o}}r.states_=r.states_.map(function(c){return y.keep(c.clone())})})},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Ob(e,r,a,this.numConstants);e=s.inputs,r=s.initialState,a=s.constants;var o=[],l=[];if(r!=null){i.initialState=r,o=o.concat(r),this.stateSpec=[];for(var u=0,c=r;u<c.length;u++){var h=c[u];this.stateSpec.push(new Nt({shape:h.shape}))}l=l.concat(this.stateSpec)}a!=null&&(i.constants=a,o=o.concat(a),this.numConstants=a.length);var d=o[0]instanceof Ln;if(d){var p=[e].concat(o),f=this.inputSpec.concat(l),m=this.inputSpec;this.inputSpec=f;var g=n.prototype.apply.call(this,p,i);return this.inputSpec=m,g}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;e=xe(e),o==null&&(r.stateful?o=r.states_:o=r.getInitialState(e));var l=Array.isArray(r.cell.stateSize)?r.cell.stateSize.length:1;if(o.length!==l)throw new M("RNN Layer has "+l+" state(s) but was passed "+(o.length+" initial state(s)."));r.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");var u={training:s},c=function(g,v){var b=r.cell.call([g].concat(v),u);return[b[0],b.slice(1)]},h=Eb(c,e,o,r.goBackwards,a,null,r.unroll,r.returnSequences),d=h[0],p=h[1],f=h[2];r.stateful&&r.resetStates(f,s);var m=r.returnSequences?p:d;return r.returnState?[m].concat(f):m})},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=y.zeros(e.shape);return r=y.sum(r,[1,2]),r=ts(r),Array.isArray(i.cell.stateSize)?i.cell.stateSize.map(function(a){return a>1?Hh(r,[1,a]):r}):i.cell.stateSize>1?[Hh(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),Kt({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=In(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Di);var oa=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),Sd=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=wt(e.kernelConstraint),i.recurrentConstraint=wt(e.recurrentConstraint),i.biasConstraint=wt(e.biasConstraint),i.dropout=na([1,xi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=na([1,xi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=je(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(e)},rate:r.dropout,training:s})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(a)},rate:r.recurrentDropout,training:s}));var o,l=r.dropoutMask,u=r.recurrentDropoutMask;l!=null?o=Mn(y.mul(e,l),r.kernel.read()):o=Mn(e,r.kernel.read()),r.bias!=null&&(o=Hn(o,r.bias.read())),u!=null&&(a=y.mul(a,u));var c=y.add(o,Mn(a,r.recurrentKernel.read()));return r.activation!=null&&(c=r.activation.apply(c)),[c,c]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:Oi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:Ke(this.kernelRegularizer),recurrentRegularizer:Ke(this.recurrentRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),recurrentConstraint:bt(this.recurrentConstraint),biasConstraint:bt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Kt({},e,i)},t.className="SimpleRNNCell",t}(oa);y.serialization.registerClass(Sd);var Db=function(n){Q(t,n);function t(e){var i=this;return e.cell=new Sd(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return new e(i)},t.className="SimpleRNN",t}(Di);y.serialization.registerClass(Db);var Ld=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new M("GRUCell does not support reset_after parameter set to true.");return i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=Ei(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=wt(e.kernelConstraint),i.recurrentConstraint=wt(e.recurrentConstraint),i.biasConstraint=wt(e.biasConstraint),i.dropout=na([1,xi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=na([1,xi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=je(e);var i=e[e.length-1];this.kernel=this.addWeight("kernel",[i,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("GRUCell expects 2 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training==null?!1:i.training,s=e[1];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:3})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:3}));var o=r.dropoutMask,l=r.recurrentDropoutMask,u,c,h;0<r.dropout&&r.dropout<1&&(e=y.mul(e,o[0]));var d=Mn(e,r.kernel.read());r.useBias&&(d=Hn(d,r.bias.read())),0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,l[0]));var p=r.recurrentKernel.read(),f=y.split(p,[2*r.units,r.units],p.rank-1),m=f[0],g=f[1],v=Mn(s,m),b=y.split(d,3,d.rank-1),w=b[0],S=b[1],L=b[2],N=y.split(v,2,v.rank-1),C=N[0],R=N[1];u=r.recurrentActivation.apply(y.add(w,C)),c=r.recurrentActivation.apply(y.add(S,R));var D=Mn(y.mul(c,s),g);h=r.activation.apply(y.add(L,D));var k=y.add(y.mul(u,s),y.mul(y.add(1,y.neg(u)),h));return[k,k]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:Oi(this.activation),recurrentActivation:Oi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:Ke(this.kernelRegularizer),recurrentRegularizer:Ke(this.recurrentRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),recurrentConstraint:bt(this.recurrentConstraint),biasConstraint:bt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Kt({},e,i)},t.className="GRUCell",t}(oa);y.serialization.registerClass(Ld);var kb=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Ld(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="GRU",t}(Di);y.serialization.registerClass(kb);var Ho=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=Ei(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.unitForgetBias=e.unitForgetBias,i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=wt(e.kernelConstraint),i.recurrentConstraint=wt(e.recurrentConstraint),i.biasConstraint=wt(e.biasConstraint),i.dropout=na([1,xi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=na([1,xi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=[i.units,i.units],i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){var i;e=je(e);var r=e[e.length-1];this.kernel=this.addWeight("kernel",[r,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);var a;if(this.useBias){if(this.unitForgetBias){var s=this.biasInitializer,o=this.units;a=new(i=function(l){Q(u,l);function u(){return l!==null&&l.apply(this,arguments)||this}return u.prototype.apply=function(c,h){var d=s.apply([o]),p=new Yh().apply([o]),f=s.apply([o*2]);return by(by(d,p),f)},u}(pn),i.className="CustomInit",i)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training;if(e=e,e.length!==3)throw new M("LSTMCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var s=e[1],o=e[2];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:4})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:4}));var l=r.dropoutMask,u=r.recurrentDropoutMask,c,h,d,p;0<r.dropout&&r.dropout<1&&(e=y.mul(e,l[0]));var f=Mn(e,r.kernel.read());0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,u[0])),f=y.add(f,Mn(s,r.recurrentKernel.read())),r.useBias&&(f=Hn(f,r.bias.read()));var m=y.split(f,4,f.rank-1),g=m[0],v=m[1],b=m[2],w=m[3];c=r.recurrentActivation.apply(g),h=r.recurrentActivation.apply(v),d=y.add(y.mul(h,o),y.mul(c,r.activation.apply(b))),p=r.recurrentActivation.apply(w);var S=y.mul(p,r.activation.apply(d));return[S,S,d]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:Oi(this.activation),recurrentActivation:Oi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:Ke(this.kernelRegularizer),recurrentRegularizer:Ke(this.recurrentRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),recurrentConstraint:bt(this.recurrentConstraint),biasConstraint:bt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Kt({},e,i)},t.className="LSTMCell",t}(oa);y.serialization.registerClass(Ho);var Fb=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Ho(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="LSTM",t}(Di);y.serialization.registerClass(Fb);var wd=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.cells=e.cells,i}return Object.defineProperty(t.prototype,"stateSize",{get:function(){for(var e=[],i=0,r=this.cells.slice().reverse();i<r.length;i++){var a=r[i];Array.isArray(a.stateSize)?e.push.apply(e,a.stateSize):e.push(a.stateSize)}return e},enumerable:!0,configurable:!0}),t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=e;for(var a=e.slice(1),s=[],o=0,l=r.cells.slice().reverse();o<l.length;o++){var u=l[o];Array.isArray(u.stateSize)?s.push(a.splice(0,u.stateSize.length)):s.push(a.splice(0,1))}s.reverse();for(var c=[],h,d=0;d<r.cells.length;++d){var u=r.cells[d];a=s[d],d===0?h=[e[0]].concat(a):h=[h[0]].concat(a),h=u.call(h,i),c.push(h.slice(1))}a=[];for(var p=0,f=c.slice().reverse();p<f.length;p++){var m=f[p];a.push.apply(a,m)}return[h[0]].concat(a)})},t.prototype.build=function(e){Qh(e)&&(e=e[0]),e=e;var i;this.cells.forEach(function(r,a){hr("RNNCell_"+a,function(){r.build(e),Array.isArray(r.stateSize)?i=r.stateSize[0]:i=r.stateSize,e=[e[0],i]})}),this.built=!0},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=function(s){return{className:s.getClassName(),config:s.getConfig()}},r=this.cells.map(i),a={cells:r};return Kt({},e,a)},t.fromConfig=function(e,i,r){r===void 0&&(r={});for(var a=[],s=0,o=i.cells;s<o.length;s++){var l=o[s];a.push(In(l,r))}return new e({cells:a})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(!this.trainable)return[];for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.cells;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.weights)}return ed(e)},t.prototype.setWeights=function(e){for(var i=[],r=0,a=this.cells;r<a.length;r++)for(var s=a[r],o=s.weights.length,l=e.splice(o),u=0;u<s.weights.length;++u)i.push([s.weights[u],l[u]]);td(i)},t.className="StackedRNNCells",t}(oa);y.serialization.registerClass(wd);function ki(n){var t=n.ones,e=n.rate,i=n.training,r=i===void 0?!1:i,a=n.count,s=a===void 0?1:a,o=function(){return Sy(t(),e)},l=function(){return is(o,t,r)};if(!s||s<=1)return y.keep(l().clone());var u=Array(s).fill(void 0).map(l);return u.map(function(c){return y.keep(c.clone())})}var pV=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(oa),tF=function(n){Q(t,n);function t(e){var i=this;if(e.unroll)throw new Te("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Te("It is not possible at the moment to stack convolutional cells.");return i=n.call(this,e)||this,i.inputSpec=[new Nt({ndim:5})],i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null),i&&i.constants)throw new M("ConvRNN2D cell does not support constants");var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.prototype.computeOutputShape=function(e){var i=this.computeSingleOutputShape(e);return this.returnSequences||(i=[i[0]].concat(i.slice(2))),this.returnState&&(i=[i].concat(Array(2).fill([e[0]].concat(i.slice(-3))))),i},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=i.cell.stateSize,a=e.shape,s=i.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=y.zeros(o);return Array.isArray(r)?Array(r.length).fill(l):[l]})},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new Ai("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape,s=r.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=a[0];if(l==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.getStates()==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_=[y.zeros(o)];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_[0]=y.zeros(o);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var u=0;u<r.states_.length;++u){var c=e[u],h=o;if(!y.util.arraysEqual(c.shape,h))throw new M("State "+u+" is incompatible with layer "+r.name+": "+("expected shape="+h+", received shape="+c.shape));r.states_[u]=c}}r.states_=r.states_.map(function(d){return y.keep(d.clone())})})},t.prototype.computeSingleOutputShape=function(e){var i=this.cell,r=i.dataFormat,a=i.filters,s=i.kernelSize,o=i.padding,l=i.strides,u=i.dilationRate,c=r==="channelsFirst",h=e[c?3:2],d=e[c?4:3],p=An(h,s[0],o,l[0],u[0]),f=An(d,s[1],o,l[1],u[1]),m=e.slice(0,2).concat(c?[a,p,f]:[p,f,a]);return m},t.className="ConvRNN2D",t}(Di),Id=function(n){Q(t,n);function t(e){var i=this,r=e.filters,a=e.kernelSize,s=e.strides,o=e.padding,l=e.dataFormat,u=e.dilationRate;return i=n.call(this,Kt({},e,{units:r}))||this,i.filters=r,Tt(i.filters,"filters"),i.kernelSize=sa(a,2,"kernelSize"),i.kernelSize.forEach(function(c){return Tt(c,"kernelSize")}),i.strides=sa(s||1,2,"strides"),i.strides.forEach(function(c){return Tt(c,"strides")}),i.padding=o||"valid",an(i.padding),i.dataFormat=l||"channelsLast",ht(i.dataFormat),i.dilationRate=sa(u||1,2,"dilationRate"),i.dilationRate.forEach(function(c){return Tt(c,"dilationRate")}),i}return t.prototype.build=function(e){var i;e=je(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=4,o=this.kernelSize.concat([a,this.filters*s]);this.kernel=this.addWeight("kernel",o,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);var l=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",l,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){var u=void 0;if(this.unitForgetBias){var c=this.biasInitializer,h=this.filters;u=new(i=function(d){Q(p,d);function p(){return d!==null&&d.apply(this,arguments)||this}return p.prototype.apply=function(f,m){var g=c.apply([h]),v=y.ones([h]),b=c.apply([h*2]);return qh([g,v,b])},p}(pn),i.className="CustomInit",i)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e.length!==3)throw new M("ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training||!1,s=e[0],o=e[1],l=e[2],u=4;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=ki({ones:function(){return y.onesLike(s)},rate:r.dropout,training:a,count:u}));var c=r.dropoutMask,h=function(te,re,le){return!re||!re[le]?te:y.mul(re[le],te)},d=h(s,c,0),p=h(s,c,1),f=h(s,c,2),m=h(s,c,3);0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=ki({ones:function(){return y.onesLike(o)},rate:r.recurrentDropout,training:a,count:u}));var g=r.recurrentDropoutMask,v=h(o,g,0),b=h(o,g,1),w=h(o,g,2),S=h(o,g,3),L=3,N=y.split(r.kernel.read(),u,L),C=N[0],R=N[1],D=N[2],k=N[3],W=r.useBias?y.split(r.bias.read(),u):[null,null,null,null],F=W[0],P=W[1],H=W[2],_=W[3];d=r.inputConv(d,C,F,r.padding),p=r.inputConv(p,R,P,r.padding),f=r.inputConv(f,D,H,r.padding),m=r.inputConv(m,k,_,r.padding);var j=y.split(r.recurrentKernel.read(),u,L),K=j[0],q=j[1],G=j[2],Z=j[3];v=r.recurrentConv(v,K),b=r.recurrentConv(b,q),w=r.recurrentConv(w,G),S=r.recurrentConv(S,Z);var X=r.recurrentActivation.apply(y.add(d,v)),ee=r.recurrentActivation.apply(y.add(p,b)),ne=y.add(y.mul(ee,l),y.mul(X,r.activation.apply(y.add(f,w)))),ie=y.mul(r.recurrentActivation.apply(y.add(m,S)),r.activation.apply(ne));return[ie,ie,ne]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=e.units,r=Pk(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Kt({},r,a)},t.prototype.inputConv=function(e,i,r,a){var s=y.conv2d(e,i,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return r?Hn(s,r,this.dataFormat):s},t.prototype.recurrentConv=function(e,i){var r=1;return y.conv2d(e,i,r,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")},t.className="ConvLSTM2DCell",t}(Ho);y.serialization.registerClass(Id);var Wb=function(n){Q(t,n);function t(e){var i=this,r=new Id(e);return i=n.call(this,Kt({},e,{cell:r}))||this,i}return t.fromConfig=function(e,i){return new e(i)},t.className="ConvLSTM2D",t}(tF);y.serialization.registerClass(Wb);var Ad=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.rate=Math.max(Math.min(e.rate,1),0),i.noiseShape=e.noiseShape,i.seed=e.seed,i.supportsMasking=!0,i}return t.prototype.getNoiseShape=function(e){if(this.noiseShape==null)return this.noiseShape;for(var i=e.shape,r=[],a=0;a<this.noiseShape.length;++a)r.push(this.noiseShape[a]==null?i[a]:this.noiseShape[a]);return r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(0<r.rate&&r.rate<1){var s=i.training==null?!1:i.training,o=r.getNoiseShape(a),l=is(function(){return Sy(a,r.rate,o,r.seed)},function(){return a},s);return l}return e})},t.prototype.getConfig=function(){var e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.dispose=function(){return n.prototype.dispose.call(this)},t.className="Dropout",t}(De);y.serialization.registerClass(Ad);var Ub=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[{ndim:3}],i}return t.prototype.getNoiseShape=function(e){var i=e.shape;return[i[0],1,i[2]]},t.className="SpatialDropout1D",t}(Ad);y.serialization.registerClass(Ub);var Bb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.activation=null,i.useBias=!0,i.kernel=null,i.bias=null,i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){var r=null;e.batchSize!=null&&(r=e.batchSize),i.batchInputShape=[r,e.inputDim]}return i.units=e.units,Tt(i.units,"units"),i.activation=Ei(e.activation),e.useBias!=null&&(i.useBias=e.useBias),i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelConstraint=wt(e.kernelConstraint),i.biasConstraint=wt(e.biasConstraint),i.kernelRegularizer=nt(e.kernelRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.supportsMasking=!0,i.inputSpec=[{minNDim:2}],i}return t.prototype.build=function(e){var i;e=je(e);var r=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[r,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:(i={},i[-1]=r,i)}],this.built=!0},t.prototype.computeOutputShape=function(e){e=je(e);var i=e.slice();return i[i.length-1]=this.units,i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=sy(r.activation.getClassName()),o;return s!=null?o=Mn(a,r.kernel.read(),s,r.bias?r.bias.read():null):(o=Mn(a,r.kernel.read()),r.bias!=null&&(o=Hn(o,r.bias.read())),r.activation!=null&&(o=r.activation.apply(o))),o})},t.prototype.getConfig=function(){var e={units:this.units,activation:Oi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:Ke(this.kernelRegularizer),biasRegularizer:Ke(this.biasRegularizer),activityRegularizer:Ke(this.activityRegularizer),kernelConstraint:bt(this.kernelConstraint),biasConstraint:bt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dense",t}(De);y.serialization.registerClass(Bb);var zb=function(n){Q(t,n);function t(e){var i=this;return e=e||{},i=n.call(this,e)||this,i.inputSpec=[{minNDim:3}],i.dataFormat=e.dataFormat,i}return t.prototype.computeOutputShape=function(e){e=je(e);for(var i=0,r=e.slice(1);i<r.length;i++){var a=r[i];if(a==null)throw new M('The shape of the input to "Flatten" is not fully defined '+("(got "+e.slice(1)+"). Make sure to pass a complete ")+'"input_shape" or "batch_input_shape" argument to the first layer in your model.')}return[e[0],Ni(e,1)]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.dataFormat==="channelsFirst"&&a.rank>1){for(var s=[0],o=2;o<a.rank;++o)s.push(o);s.push(1),a=a.transpose(s)}return r3(a)})},t.prototype.getConfig=function(){var e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);var i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Flatten",t}(De);y.serialization.registerClass(zb);var Pb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.activation=Ei(e.activation),i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);return r.activation.apply(a)})},t.prototype.getConfig=function(){var e={activation:Oi(this.activation)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Activation",t}(De);y.serialization.registerClass(Pb);var _b=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.n=e.n,i.inputSpec=[{ndim:2}],i}return t.prototype.computeOutputShape=function(e){return[e[0],this.n,e[1]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return e=xe(e),n3(e,r.n)})},t.prototype.getConfig=function(){var e={n:this.n},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="RepeatVector",t}(De);y.serialization.registerClass(_b);var Mb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;i.targetShape=e.targetShape;for(var r=0;r<i.targetShape.length;++r)i.isUnknown(i.targetShape[r])&&(i.targetShape[r]=null);return i}return t.prototype.isUnknown=function(e){return e<0||e==null},t.prototype.fixUnknownDimension=function(e,i){for(var r="Total size of new array must be unchanged.",a=i.slice(),s=1,o=null,l=0;l<a.length;++l){var u=a[l];if(this.isUnknown(u))if(o===null)o=l;else throw new M("Can only specifiy one unknown dimension.");else s*=u}var c=Ni(e);if(o!==null){if(s===0||c%s!==0)throw new M(r);a[o]=c/s}else if(c!==s)throw new M(r);return a},t.prototype.computeOutputShape=function(e){for(var i=!1,r=0;r<e.length;++r)if(this.isUnknown(e[r])){i=!0;break}return i?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=a.shape,o=s.slice(0,1).concat(r.fixUnknownDimension(s.slice(1),r.targetShape));return a.reshape(o)})},t.prototype.getConfig=function(){var e={targetShape:this.targetShape},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Reshape",t}(De);y.serialization.registerClass(Mb);var Hb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error("Permute constructor requires `dims` to be an Array, but received "+(e.dims+" instead."));var r=Sn(1,e.dims.length+1);if(!y.util.arraysEqual(e.dims.slice().sort(),r))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");return i.dims=e.dims,i.dimsIncludingBatch=[0].concat(i.dims),i.inputSpec=[new Nt({ndim:i.dims.length+1})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=e.slice();return this.dims.forEach(function(r,a){i[a+1]=e[r]}),i},t.prototype.call=function(e,i){return y.transpose(xe(e),this.dimsIncludingBatch)},t.prototype.getConfig=function(){var e={dims:this.dims},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Permute",t}(De);y.serialization.registerClass(Hb);var Vb=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null?i.maskValue=e.maskValue==null?0:e.maskValue:i.maskValue=0,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={maskValue:this.maskValue};return Object.assign(i,e),i},t.prototype.computeMask=function(e,i){var r=xe(e),a=-1;return y.any(y.notEqual(r,this.maskValue),a)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=-1,o=!0,l=y.any(y.notEqual(a,r.maskValue),s,o),u=a.mul(l.asType(a.dtype));return u})},t.className="Masking",t}(De);y.serialization.registerClass(Vb);var qb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.embeddings=null,i.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){var r=null;e.batchSize!=null&&(r=e.batchSize),e.inputLength==null?i.batchInputShape=[r,null]:i.batchInputShape=[r].concat(Je(e.inputLength))}return i.inputDim=e.inputDim,Tt(i.inputDim,"inputDim"),i.outputDim=e.outputDim,Tt(i.outputDim,"outputDim"),i.embeddingsInitializer=tt(e.embeddingsInitializer||i.DEFAULT_EMBEDDINGS_INITIALIZER),i.embeddingsRegularizer=nt(e.embeddingsRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.embeddingsConstraint=wt(e.embeddingsConstraint),i.maskZero=e.maskZero,i.supportsMasking=e.maskZero,i.inputLength=e.inputLength,i}return t.prototype.build=function(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0},t.prototype.warnOnIncompatibleInputShape=function(e){},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){return r.maskZero?(e=xe(e),y.notEqual(e,y.zerosLike(e))):null})},t.prototype.computeOutputShape=function(e){if(e=je(e),this.inputLength==null)return e.concat([this.outputDim]);var i=Je(this.inputLength);if(i.length!==e.length-1)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));for(var r=0,a=0;a<i.length;++a){var s=i[a],o=e[a+1];if(s!=null&&o!=null&&s!==o)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));s==null&&(i[r]=o),r++}return[e[0]].concat(i,[this.outputDim])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);a.dtype!=="int32"&&(a=es(a,"int32"));var s=wy(r.embeddings.read(),a.as1D());return s.reshape(je(r.computeOutputShape(a.shape)))})},t.prototype.getConfig=function(){var e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:st(this.embeddingsInitializer),embeddingsRegularizer:Ke(this.embeddingsRegularizer),activityRegularizer:Ke(this.activityRegularizer),embeddingsConstraint:bt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Embedding",t}(De);y.serialization.registerClass(qb);var mr=function(n){Q(t,n);function t(e){var i=n.call(this,e||{})||this;return i.supportsMasking=!0,i}return t.prototype.mergeFunction=function(e){throw new Te},t.prototype.computeElementwiseOpOutputShape=function(e,i){if(e==null||i==null)return null;if(e.length<i.length)return this.computeElementwiseOpOutputShape(i,e);if(i.length===0)return e;for(var r=e.slice(0,e.length-i.length),a=0;a<i.length;++a){var s=e[e.length-i.length+a],o=i[a];if(s==null||o==null||s<0||o<0)r.push(null);else if(s===1)r.push(o);else if(o===1)r.push(s);else{if(s!==o)throw new M("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(i));r.push(s)}}return r},t.prototype.build=function(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[je(e)]),e=e,e.length<2)throw new M("A merge layer should be called on an Array of at least 2 inputs."+(" Got "+e.length+" input(s)."));for(var i=[],r=0,a=e;r<a.length;r++){var s=a[r];s!=null&&s[0]!==null&&i.push(s[0])}if(i=Ti(i),i.length>1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l<e.length;++l){var s=e[l]==null?null:e[l].slice(1);o=this.computeElementwiseOpOutputShape(o,s)}var u=e.map(function(c){return c.length});e.indexOf(null)===-1&&Ti(u).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,r.reshapeRequired){var a=[],s=e.map(function(D){return D.rank});if(s.indexOf(null)===-1){for(var o=xi(s),l=0,u=e;l<u.length;l++){for(var c=u[l],h=c.rank,d=0;d<o-h;++d)c=ts(c,1);a.push(c)}return r.mergeFunction(a)}else{for(var p=!1,f=0,m=e;f<m.length;f++){var c=m[f],h=c.rank;if(h==null){var g=c.shape,v=g[0],b=g.slice(1).concat([v]),w=c.reshape([v].concat(Ni(g.slice(1))));w=y.transpose(w,[1,0]),w=w.reshape(b),a.push(w),p=!0}else if(h>1){var S=Sn(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),N=L.rank;if(p){if(N==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(N>1){var S=[N-1].concat(Sn(0,N-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r<e.length;++r){var a=e[r]==null?null:e[r].slice(1);i=this.computeElementwiseOpOutputShape(i,a)}for(var s=[],o=0,l=e;o<l.length;o++){var a=l[o];a!=null&&a[0]!==null&&s.push(a[0])}return s=Ti(s),s.length===1?i=s.concat(i):i=[null].concat(i),i},t.prototype.computeMask=function(e,i){return y.tidy(function(){if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an Array");if(!Array.isArray(e))throw new M("`inputs` should be an Array");if(i.length!==e.length)throw new M("The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths "+("("+e.length+" vs "+i.length+")"));if(i.every(function(s){return s==null}))return null;i=i.map(function(s){return s==null?s:y.expandDims(s,0)});for(var r=i[0],a=1;a<i.length-1;++a)r=y.logicalAnd(r,i[a]);return r})},t}(De),Gb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return i})},t.className="Add",t}(mr);y.serialization.registerClass(Gb);var Yb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.mul(i,e[r]);return i})},t.className="Multiply",t}(mr);y.serialization.registerClass(Yb);var jb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return y.mul(1/e.length,i)})},t.className="Average",t}(mr);y.serialization.registerClass(jb);var Kb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.maximum(i,e[r]);return i})},t.className="Maximum",t}(mr);y.serialization.registerClass(Kb);var $b=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.minimum(i,e[r]);return i})},t.className="Minimum",t}(mr);y.serialization.registerClass($b);var Xb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_AXIS=-1,e==null&&(e={}),i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new M("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;for(var i=!0,r=0,a=e;r<a.length;r++){var s=a[r];if(s!=null){i=!1;break}}if(i)return;for(var o=[],l=0;l<e.length;++l){var u=e[l].slice();u.splice(this.axis,1);for(var c=!1,h=0,d=o;h<d.length;h++){var s=d[h];if(y.util.arraysEqual(s,u)){c=!0;break}}c||o.push(u)}if(o.length>1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return qh(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s<o.length;s++){var l=o[s];if(r[a]==null||l[a]==null){r[a]=null;break}r[a]+=l[a]}return r},t.prototype.computeMask=function(e,i){var r=this;if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new M("`inputs` should be an array for Concatenate");if(i.length!==e.length)throw new M("Mismatch in the length of mask ("+i.length+") "+("and the legnth of inputs ("+e.length+")"));return y.tidy(function(){var a=!0;if(i.forEach(function(u){if(u!=null){a=!1;return}}),a)return null;for(var s=[],o=0;o<e.length;++o)i[o]==null?s.push(y.onesLike(e[o]).asType("bool")):i[o].rank<e[o].rank?s.push(y.expandDims(i[o],-1)):s.push(i[o]);var l=y.concat(s,r.axis);return y.all(l,-1,!1)})},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Concatenate",t}(mr);y.serialization.registerClass(Xb);function us(n,t){for(;n<0;)n+=t;return n}function nF(n,t,e){if(n.shape.length>3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;l<s;++l)o.push(1);t=t.reshape(t.shape.concat(o))}else if(r>i){s=r-i;for(var o=[],l=0;l<s;++l)o.push(1);n=n.reshape(n.shape.concat(o))}else s=0;var u;if(n.shape.length===2&&t.shape.length===2)a[0]===a[1]?u=n.mul(t).sum(a[0]):u=n.transpose([1,0]).mul(t).sum(a[1]);else{var c=a[0]!==n.shape.length-1,h=a[1]===t.shape.length-1;u=n.matMul(t,c,h)}if(s>0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l<d+s;++l)p.push(l);u=u.squeeze(p)}return u.shape.length===1&&(u=u.expandDims(1)),u})}var Jb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.axes=e.axes,i.normalize=e.normalize==null?!1:e.normalize,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0],r=e[1];if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return us(s,e[o].shape.length)}):a=[us(this.axes,i.shape.length),us(this.axes,r.shape.length)],this.normalize&&(i=Eo(i,a[0]),r=Eo(r,a[1])),nF(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[us(this.axes,e.length),us(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(mr);y.serialization.registerClass(Jb);var Zb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return To(a.shape,0,r.stddev).add(a)},o=is(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass(Zb);var Qb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(To(a.shape,1,o))};return is(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(Qb);var ew=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=es(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return is(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(ew);function cs(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function iF(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=cs(n,s,o,e,t,r);return[l,s,o]})}function rF(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=Sn(0,n.rank);u<c.length;u++){var h=c[u];i.indexOf(h)!==-1?l.push(1):l.push(n.shape[h])}var d=s.reshape(l),p=o.reshape(l),f=t==null?null:t.reshape(l),m=e==null?null:e.reshape(l),g=cs(n,d,p,m,f,r);return[g,s,o]})}function aF(n,t,e,i,r){return r===void 0&&(r=.001),y.util.arraysEqual(i.slice().sort(),Sn(0,n.rank-1))?iF(n,t,e,i,r):rF(n,t,e,i,r)}var tw=function(n){Q(t,n);function t(e){var i=this;return e==null&&(e={}),i=n.call(this,e)||this,i.supportsMasking=!0,i.axis=e.axis==null?-1:e.axis,i.momentum=e.momentum==null?.99:e.momentum,i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.movingMeanInitializer=tt(e.movingMeanInitializer||"zeros"),i.movingVarianceInitializer=tt(e.movingVarianceInitializer||"ones"),i.betaConstraint=wt(e.betaConstraint),i.gammaConstraint=wt(e.gammaConstraint),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i}return t.prototype.build=function(e){var i;e=je(e);var r=this.axis>=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Nt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=Sn(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=ur(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,Sn(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),N=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return cs(s,L,N,C,R,r.epsilon)}else return cs(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=aF(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,N,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(N).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),movingMeanInitializer:st(this.movingMeanInitializer),movingVarianceInitializer:st(this.movingVarianceInitializer),betaRegularizer:Ke(this.betaRegularizer),gammaRegularizer:Ke(this.gammaRegularizer),betaConstraint:bt(this.betaConstraint),gammaConstraint:bt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(tw);var nw=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r<a.length;r++){var s=a[r];if(!Number.isInteger(s))throw new Error("Expected axis to be an array of integers, "+("but received "+JSON.stringify(i.axis)))}else throw new Error("Expected axis to be an integer or an array of integers, "+("but received "+JSON.stringify(i.axis)));return i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i.supportsMasking=!0,i}return t.prototype.build=function(e){e=je(e);var i=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(var r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=i);for(var a=0,s=this.axis;a<s.length;a++){var o=s[a];if(o<0||o>=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==Ti(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=ur(1,o),p=0,f=r.axis;p<f.length;p++){var m=f[p];d[m]=s[m]}for(var g=function(N){return N!=null&&N.shape.length!==o&&r.axis!==[o-1]?N.reshape(d):N},v=g(r.gamma.read()),b=g(r.beta.read()),w=[],S=[],L=0;L<o;++L)r.axis.indexOf(L)!==-1?(w.push(s[L]),S.push(1)):(w.push(1),S.push(s[L]));return c=c.tile(w),h=h.tile(w),v=v.tile(S),b=b.tile(S),cs(a,c,h,b,v,r.epsilon)})},t.prototype.getConfig=function(){var e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),betaRegularizer:Ke(this.betaRegularizer),gammaRegularizer:Ke(this.gammaRegularizer)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LayerNormalization",t}(De);y.serialization.registerClass(nw);function sF(n,t,e){return y.tidy(function(){if(n.rank!==4)throw new M("temporalPadding expects input tensor to be 4-D, but received a "+(n.rank+"-D tensor."));if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new M("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(e==null&&(e=wn()),e!=="channelsLast"&&e!=="channelsFirst")throw new M("Unknown data format: "+e+". Supported data formats are 'channelsLast' and 'channelsFirst.");var i;return e==="channelsFirst"?i=[[0,0],[0,0],t[0],t[1]]:i=[[0,0],t[0],t[1],[0,0]],y.pad(n,i)})}var iw=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.dataFormat=e.dataFormat==null?wn():e.dataFormat,e.padding==null)i.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")i.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new M("ZeroPadding2D expects padding to be a length-2 array, but "+("received a length-"+e.padding.length+" array."));var r=void 0,a=void 0;if(typeof e.padding[0]=="number")r=[e.padding[0],e.padding[0]],a=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new M("ZeroPadding2D expects height padding to be a length-2 array, "+("but received a length-"+e.padding[0].length+" array."));if(r=e.padding[0],e.padding[1].length!==2)throw new M("ZeroPadding2D expects width padding to be a length-2 array, "+("but received a length-"+e.padding[1].length+" array."));a=e.padding[1]}i.padding=[r,a]}return i.inputSpec=[new Nt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i,r;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return sF(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(iw);function Vo(n,t,e,i,r,a){return y.tidy(function(){ht(r),py(a),an(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=wn()),a==null&&(a="max"),n=yd(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function rw(n,t,e,i,r,a){return y.tidy(function(){ht(r),py(a),an(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=wn()),a==null&&(a="max"),n=wb(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var aw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(Tt(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return Tt(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,an(i.padding),i.inputSpec=[new Nt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=An(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=ts(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),sw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),an(a),Vo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(aw);y.serialization.registerClass(sw);var ow=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),an(a),Vo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(aw);y.serialization.registerClass(ow);var lw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return Tt(i.poolSize,"poolSize"),Tt(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ht(i.dataFormat),an(i.padding),i.inputSpec=[new Nt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=An(i,this.poolSize[0],this.padding,this.strides[0]),r=An(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),uw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),an(a),Vo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(lw);y.serialization.registerClass(uw);var cw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),an(a),Vo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(lw);y.serialization.registerClass(cw);var hw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return Tt(i.poolSize,"poolSize"),Tt(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ht(i.dataFormat),an(i.padding),i.inputSpec=[new Nt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=je(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=An(i,this.poolSize[0],this.padding,this.strides[0]),r=An(r,this.poolSize[1],this.padding,this.strides[1]),a=An(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),dw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),an(a),rw(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(hw);y.serialization.registerClass(dw);var pw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ht(s),an(a),rw(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(hw);y.serialization.registerClass(pw);var fw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Nt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),mw=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(fw);y.serialization.registerClass(mw);var gw=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(fw);y.serialization.registerClass(gw);var vw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ht(i.dataFormat),i.inputSpec=[new Nt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),yw=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(vw);y.serialization.registerClass(yw);var bw=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(vw);y.serialization.registerClass(bw);var ww=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=In(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),Sw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=je(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=je(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Eb(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(ww);y.serialization.registerClass(Sw);function oF(n){ea(Zk,"BidirectionalMergeMode",n)}var lF="concat",Lw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=In(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=In(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?lF:e.mergeMode,oF(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):qt(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Ob(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Nt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof Ln,d=0,p=o;d<p.length;d++){var f=p[d];if(f instanceof Ln!==h)throw new M("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors")}if(h){var m=[e].concat(o),g=this.inputSpec.concat(l),v=this.inputSpec;this.inputSpec=g;var b=n.prototype.apply.call(this,m,i);return this.inputSpec=v,b}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.initialState,s,o;if(a==null)s=r.forwardLayer.call(e,i),o=r.backwardLayer.call(e,i);else{var l=a.slice(0,a.length/2),u=a.slice(a.length/2);s=r.forwardLayer.call(e,Object.assign(i,{initialState:l})),o=r.backwardLayer.call(e,Object.assign(i,{initialState:u}))}var c;r.returnState&&(Array.isArray(s)&&(c=s.slice(1).concat(o.slice(1))),s=s[0],o=o[0]),r.returnSequences&&(o=y.reverse(o,1));var h;return r.mergeMode==="concat"?h=qh([s,o]):r.mergeMode==="sum"?h=y.add(s,o):r.mergeMode==="ave"?h=y.mul(.5,y.add(s,o)):r.mergeMode==="mul"?h=y.mul(s,o):r.mergeMode==null&&(h=[s,o]),r.returnState?r.mergeMode==null?h.concat(c):[h].concat(c):h})},t.prototype.resetStates=function(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()},t.prototype.build=function(e){var i=this;hr(this.forwardLayer.name,function(){i.forwardLayer.build(e)}),hr(this.backwardLayer.name,function(){i.backwardLayer.build(e)}),this.built=!0},t.prototype.computeMask=function(e,i){Array.isArray(i)&&(i=i[0]);var r;if(this.returnSequences?this.mergeMode==null?r=[i,i]:r=i:this.mergeMode==null?r=[null,null]:r=null,this.returnState){var a=this.forwardLayer.states,s=a.map(function(o){return null});return Array.isArray(r)?r.concat(s).concat(s):[r].concat(s).concat(s)}else return r},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e={mergeMode:this.mergeMode},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.fromConfig=function(e,i){var r=In(i.layer);if(delete i.layer,i.numConstants!=null)throw new Te("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");var a=i;return a.layer=r,new e(a)},t.className="Bidirectional",t}(ww);y.serialization.registerClass(Lw);function uF(n){return new rs(n)}function cF(n){return new vb(n)}function hF(n){return new fb(n)}function dF(n){return new mb(n)}function pF(n){return new gb(n)}function fF(n){return new bb(n)}function mF(n){return new yb(n)}function gF(n){return new Nb(n)}function vF(n){return new bd(n)}function yF(n){return new Ab(n)}function bF(n){return new Ib(n)}function wF(n){return new Tb(n)}function SF(n){return new xb(n)}function LF(n){return new Cb(n)}function IF(n){return new Rb(n)}function AF(n){return new Pb(n)}function TF(n){return new Bb(n)}function NF(n){return new Ad(n)}function xF(n){return new Ub(n)}function CF(n){return new zb(n)}function RF(n){return new _b(n)}function OF(n){return new Mb(n)}function EF(n){return new Hb(n)}function DF(n){return new qb(n)}function kF(n){return new Gb(n)}function FF(n){return new jb(n)}function WF(n){return new Xb(n)}function UF(n){return new Kb(n)}function BF(n){return new $b(n)}function zF(n){return new Yb(n)}function PF(n){return new Jb(n)}function _F(n){return new tw(n)}function MF(n){return new nw(n)}function HF(n){return new iw(n)}function Td(n){return new ow(n)}function VF(n){return Td(n)}function qF(n){return Td(n)}function Nd(n){return new cw(n)}function GF(n){return Nd(n)}function YF(n){return Nd(n)}function xd(n){return new pw(n)}function jF(n){return xd(n)}function KF(n){return xd(n)}function $F(n){return new mw(n)}function XF(n){return new yw(n)}function Iw(n){return new gw(n)}function Aw(n){return new bw(n)}function Tw(n){return new sw(n)}function Nw(n){return new uw(n)}function JF(n){return new dw(n)}function ZF(n){return new kb(n)}function QF(n){return new Ld(n)}function eW(n){return new Fb(n)}function tW(n){return new Ho(n)}function nW(n){return new Db(n)}function iW(n){return new Sd(n)}function rW(n){return new Wb(n)}function aW(n){return new Id(n)}function sW(n){return new Di(n)}function oW(n){return new wd(n)}function lW(n){return new Lw(n)}function uW(n){return new Sw(n)}var cW=Iw,hW=Aw,dW=Tw,pW=Nw;function fW(n){return new Zb(n)}function mW(n){return new Qb(n)}function gW(n){return new ew(n)}function vW(n){return new Vb(n)}var yW={__proto__:null,inputLayer:uF,elu:cF,reLU:hF,leakyReLU:dF,prelu:pF,softmax:fF,thresholdedReLU:mF,conv1d:gF,conv2d:vF,conv2dTranspose:yF,conv3d:bF,separableConv2d:wF,cropping2D:SF,upSampling2d:LF,depthwiseConv2d:IF,activation:AF,dense:TF,dropout:NF,spatialDropout1d:xF,flatten:CF,repeatVector:RF,reshape:OF,permute:EF,embedding:DF,add:kF,average:FF,concatenate:WF,maximum:UF,minimum:BF,multiply:zF,dot:PF,batchNormalization:_F,layerNormalization:MF,zeroPadding2d:HF,averagePooling1d:Td,avgPool1d:VF,avgPooling1d:qF,averagePooling2d:Nd,avgPool2d:GF,avgPooling2d:YF,averagePooling3d:xd,avgPool3d:jF,avgPooling3d:KF,globalAveragePooling1d:$F,globalAveragePooling2d:XF,globalMaxPooling1d:Iw,globalMaxPooling2d:Aw,maxPooling1d:Tw,maxPooling2d:Nw,maxPooling3d:JF,gru:ZF,gruCell:QF,lstm:eW,lstmCell:tW,simpleRNN:nW,simpleRNNCell:iW,convLstm2d:rW,convLstm2dCell:aW,rnn:sW,stackedRNNCells:oW,bidirectional:lW,timeDistributed:uW,globalMaxPool1d:cW,globalMaxPool2d:hW,maxPool1d:dW,maxPool2d:pW,Layer:De,RNN:Di,RNNCell:oa,input:ub,gaussianNoise:fW,gaussianDropout:mW,alphaDropout:gW,masking:vW};function bW(n,t){return rd(n,t)}function wW(n,t){return Yy(n,t)}function SW(n,t){return jy(n,t)}function LW(n,t){return ad(n,t)}function IW(n,t){return sd(n,t)}function AW(n,t){return Gy(n,t)}function TW(n,t){return j3(n,t)}function NW(n,t){return nd(n,t)}function xW(n,t){return Do(n,t)}function CW(n,t){return ra(n,t)}function RW(n,t){return ra(n,t)}function OW(n,t){return ra(n,t)}function EW(n,t){return pr(n,t)}function DW(n,t){return pr(n,t)}function kW(n,t){return pr(n,t)}var FW={__proto__:null,binaryAccuracy:bW,binaryCrossentropy:wW,sparseCategoricalAccuracy:SW,categoricalAccuracy:LW,categoricalCrossentropy:IW,precision:AW,recall:TW,cosineProximity:NW,meanAbsoluteError:xW,meanAbsolutePercentageError:CW,MAPE:RW,mape:OW,meanSquaredError:EW,MSE:DW,mse:kW};var WW={__proto__:null,modelFromJSON:R4};function UW(n){return new Po(n)}function BW(n){return $4(n)}function zW(n){return X4(n)}var PW={__proto__:null,l1l2:UW,l1:BW,l2:zW};var xw=function(n){Q(t,n);function t(){var e=n!==null&&n.apply(this,arguments)||this;return e.model=null,e}return t.prototype.setModel=function(e){if(!(e instanceof Ri))throw new Error("model must be a LayersModel, not some other Container");this.model=e},t}(ia);function qo(n,t){return n<t}function Cw(n,t){return n>t}var Rw=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=qo:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=Cw:i.monitorFunc=qo,i.monitorFunc===qo&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===qo?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,Ci(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(xw);function _W(n){return new Rw(n)}var MW={earlyStopping:_W};Xe.Callback=xw;Xe.CallbackList=zy;Xe.CustomCallback=_y;Xe.EarlyStopping=Rw;Xe.History=Py;Xe.InputSpec=Nt;Xe.LayerVariable=ky;Xe.LayersModel=Ri;Xe.RNN=Di;Xe.Sequential=md;Xe.SymbolicTensor=Ln;Xe.callbacks=MW;Xe.constraints=Kk;Xe.initializers=C3;Xe.input=ub;Xe.layers=yW;Xe.loadLayersModel=W4;Xe.metrics=FW;Xe.model=k4;Xe.models=WW;Xe.registerCallbackConstructor=U4;Xe.regularizers=PW;Xe.sequential=F4;Xe.version_layers=ud});var Gw=Le(gr=>{"use strict";Object.defineProperty(gr,"__esModule",{value:!0});var B=lr();var Ew=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t};function Vn(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function Tn(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=r[u[0]&2?"return":u[0]?"throw":"next"])&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[0,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Nn;(function(n){n[n.DT_INVALID=0]="DT_INVALID",n[n.DT_FLOAT=1]="DT_FLOAT",n[n.DT_DOUBLE=2]="DT_DOUBLE",n[n.DT_INT32=3]="DT_INT32",n[n.DT_UINT8=4]="DT_UINT8",n[n.DT_INT16=5]="DT_INT16",n[n.DT_INT8=6]="DT_INT8",n[n.DT_STRING=7]="DT_STRING",n[n.DT_COMPLEX64=8]="DT_COMPLEX64",n[n.DT_INT64=9]="DT_INT64",n[n.DT_BOOL=10]="DT_BOOL",n[n.DT_QINT8=11]="DT_QINT8",n[n.DT_QUINT8=12]="DT_QUINT8",n[n.DT_QINT32=13]="DT_QINT32",n[n.DT_BFLOAT16=14]="DT_BFLOAT16",n[n.DT_FLOAT_REF=101]="DT_FLOAT_REF",n[n.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",n[n.DT_INT32_REF=103]="DT_INT32_REF",n[n.DT_UINT8_REF=104]="DT_UINT8_REF",n[n.DT_INT16_REF=105]="DT_INT16_REF",n[n.DT_INT8_REF=106]="DT_INT8_REF",n[n.DT_STRING_REF=107]="DT_STRING_REF",n[n.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",n[n.DT_INT64_REF=109]="DT_INT64_REF",n[n.DT_BOOL_REF=110]="DT_BOOL_REF",n[n.DT_QINT8_REF=111]="DT_QINT8_REF",n[n.DT_QUINT8_REF=112]="DT_QUINT8_REF",n[n.DT_QINT32_REF=113]="DT_QINT32_REF",n[n.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(Nn||(Nn={}));var Dw;(function(n){var t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=n.CheckpointFormatVersion||(n.CheckpointFormatVersion={}))})(Dw||(Dw={}));var Cd={};function HW(n,t){var e={tfOpName:n,category:"custom",inputs:[],attrs:[],customExecutor:t};Cd[n]=e}function kw(n){return Cd[n]}function VW(n){delete Cd[n]}function A(n,t,e,i){var r=t.inputParams[n];if(r&&r.inputIndexStart!==void 0){var a=r.inputIndexStart,s=r.inputIndexEnd===0?void 0:r.inputIndexEnd===void 0?a+1:r.inputIndexEnd;if(r.type==="tensor")return Gt(t.inputNames[r.inputIndexStart],e,i);if(r.type==="tensors"){var o=t.inputNames.slice(a,s);return o.map(function(h){return Gt(h,e,i)})}var l=Gt(t.inputNames.slice(a)[0],e,i),u=l.dataSync();return r.type==="number"?u[0]:B.util.toNestedArray(l.shape,u)}var c=t.attrParams[n];return c&&c.value}function Gt(n,t,e){var i=Jt(n),r=i[0],a=i[1],s=e.currentContextIds.find(function(o){return!!t[Go(r,o)]});return s!==void 0?t[Go(r,s)][a]:void 0}function qW(n,t,e){return t[Go(n,e.currentContextId)]}function si(n,t){var e=Jt(n),i=e[0],r=e[1];return[Go(i,t&&t.currentContextId),r]}function Go(n,t){return t?n+"-"+t:n}function Jt(n){var t=n.split(":");if(t.length===1)return[n,0];var e=t[0];return[e,Number(t[t.length-1])]}function Yo(n,t,e){var i=A("pad",n,t,e);if(i==="explicit"){i=A("explicitPaddings",n,t,e);for(var r=[[0,0],[0,0],[0,0],[0,0]],a=0;a<4;a++)r[a][0]=i[a*2],r[a][1]=i[a*2+1];return r}return i}function oi(n){return n.kept?n:B.clone(n)}var GW=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],YW={__proto__:null,json:GW};var jW=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"clip_value_min",name:"clipValueMin",type:"number"},{tfName:"clip_value_max",name:"clipValueMax",type:"number"}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"clipValueMin",name:"clipValueMin",type:"number",defaultValue:0},{tfName:"clipValueMax",name:"clipValueMax",type:"number",defaultValue:6}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],KW={__proto__:null,json:jW};var $W=[{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],XW={__proto__:null,json:$W};var JW=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],ZW={__proto__:null,json:JW};var QW=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],eU={__proto__:null,json:QW};var tU=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],nU={__proto__:null,json:tU};var iU=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],rU={__proto__:null,json:iU};var aU=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],sU={__proto__:null,json:aU};var oU=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],lU={__proto__:null,json:oU};var uU=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],cU={__proto__:null,json:uU};var hU=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],dU={__proto__:null,json:hU};var pU=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],fU={__proto__:null,json:pU};var mU=[{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],gU={__proto__:null,json:mU};var vU=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool",notSupported:!0}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],yU={__proto__:null,json:vU};var bU=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],wU={__proto__:null,json:bU};var SU=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]}],LU={__proto__:null,json:SU};var Ww=function(){function n(){var t=[YW,KW,XW,ZW,eU,nU,rU,cU,lU,sU,dU,fU,gU,yU,wU,LU],e=[].concat.apply([],t.map(function(i){return i.json}));this.opMappers=e.reduce(function(i,r){return i[r.tfOpName]=r,i},{})}return Object.defineProperty(n,"Instance",{get:function(){return this._instance||(this._instance=new this)},enumerable:!0,configurable:!0}),n.prototype.transformGraph=function(t,e){var i=this;e===void 0&&(e={});var r=t.node,a=[],s=[],o=[],l=r.reduce(function(g,v){return g[v.name]=i.mapNode(v),v.op.startsWith("Placeholder")?a.push(g[v.name]):v.op==="Const"?s.push(g[v.name]):(v.input==null||v.input.length===0)&&o.push(g[v.name]),g},{}),u=[],c=[],h={},d={};e!=null&&(h=this.mapSignatureEntries(e.inputs),d=this.mapSignatureEntries(e.outputs));var p=Object.keys(l);p.forEach(function(g){var v=l[g];v.inputNames.forEach(function(b){var w=si(b)[0];v.inputs.push(l[w]),l[w].children.push(v)})}),Object.keys(d).length===0?p.forEach(function(g){var v=l[g];v.children.length===0&&c.push(v)}):Object.keys(d).forEach(function(g){var v=si(g)[0],b=l[v];b!=null&&(b.signatureKey=d[g],c.push(b))}),Object.keys(h).length>0?Object.keys(h).forEach(function(g){var v=si(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=kw(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=Bd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Bd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=Ed(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=Ed(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=Ud(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Ud(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Od(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=Pd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Pd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=Wd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Wd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=zd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=zd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=kd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=kd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=Fd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Fd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Fw(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Fw(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=si(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:Dd(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=si(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=si(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function IU(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function Uw(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):IU(n);return t?e:e.toLowerCase()}function Rd(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?Uw(r.s,i):e}function Od(n,t,e){var i=n[t];return i?i.b:e}function Ed(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function Dd(n){typeof n=="string"&&(n=Nn[n]);switch(n){case Nn.DT_FLOAT:return"float32";case Nn.DT_INT32:case Nn.DT_INT64:case Nn.DT_INT8:case Nn.DT_UINT8:return"int32";case Nn.DT_BOOL:return"bool";case Nn.DT_DOUBLE:return"float32";case Nn.DT_STRING:return"string";default:return null}}function Fw(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function kd(n,t,e){var i=n[t];return i&&i.type?Dd(i.type):e}function Fd(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return Dd(r)}):e}function Bw(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function Wd(n,t,e){var i=n[t];return i&&i.shape?Bw(i.shape):e}function Ud(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function Bd(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return Uw(a,i)}):e}function zd(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Bw(r)}):e}function Pd(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var AU=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Gt(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Gt(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return Ed(this.node.rawAttrs,t,e);if(i.s!=null)return Rd(this.node.rawAttrs,t,e);if(i.b!=null)return Od(this.node.rawAttrs,t,e);if(i.shape!=null)return Wd(this.node.rawAttrs,t,e);if(i.type!=null)return kd(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return Ud(this.node.rawAttrs,t,e);if(i.list.s!=null)return Bd(this.node.rawAttrs,t,e);if(i.list.shape!=null)return zd(this.node.rawAttrs,t,e);if(i.list.b!=null)return Pd(this.node.rawAttrs,t,e);if(i.list.type!=null)return Fd(this.node.rawAttrs,t,e)}return e},n}();var TU=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var NU=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Gt(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function fn(n,t,e){e===void 0&&(e=""),B.util.assert(xU(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function xU(n,t){if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==-1&&t[e]!==-1&&n[e]!==t[e])return!1;return!0}var CU=function(){function n(t,e,i,r,a,s,o){this.name=t,this.dtype=e,this.maxSize=i,this.elementShape=r,this.identicalElementShapes=a,this.dynamicSize=s,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=B.scalar(0),B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"closed",{get:function(){return this.closed_},enumerable:!0,configurable:!0}),n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.read=function(t){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||t>=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`,
+          because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),fn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i<this.size();i++)t.push(i)}if(t.length===0)return B.tensor([],[0].concat(this.elementShape));var r=this.readMany(t);return fn(this.elementShape,r[0].shape,"TensorArray shape mismatch: "),B.stack(r,0)},n.prototype.concat=function(t){if(!!t&&t!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but concat requested dtype "+t);if(this.size()===0)return B.tensor([],[0].concat(this.elementShape));for(var e=[],i=0;i<this.size();i++)e.push(i);var r=this.readMany(e);return fn(this.elementShape,r[0].shape,"TensorArray shape mismatch: tensor array shape ("+this.elementShape+") vs first tensor shape ("+r[0].shape+")"),B.concat(r,0)},n.prototype.scatter=function(t,e){if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);if(t.length!==e.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+e.shape[0]);var i=Math.max.apply(Math,t);if(!this.dynamicSize&&i>=this.maxSize)throw new Error("Max index must be < array size ("+i+"  vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
           tensor.shape[0], but sum of lengths is
-        `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;c<t.length;++c){var h=c===0?0:a[c-1],d=[0,h,0],p=[1,t[c],s];o[c]=B.reshape(B.slice(e,d,p),i.elementShape)}return o});for(var l=[],u=0;u<t.length;u++)l[u]=u;this.writeMany(l,o)},n}();var jo=function(){function n(t,e,i,r){r===void 0&&(r=-1),this.tensors=t,this.elementShape=e,this.elementDtype=i,t!=null&&t.forEach(function(a){if(i!==a.dtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+a.dtype);mn(e,a.shape,"TensorList shape mismatch: "),B.keep(a)}),this.idTensor=B.scalar(0),this.maxNumElements=r,B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),n.prototype.copy=function(){return new n(this.tensors.slice(),this.elementShape,this.elementDtype)},n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.stack=function(t,e,i){var r=this;if(i===void 0&&(i=-1),e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(i!==-1&&this.tensors.length!==i)throw new Error("Operation expected a list with "+i+" elements but got a list with "+this.tensors.length+" elements.");return mn(t,this.elementShape,"TensorList shape mismatch: "),B.tidy(function(){var a=r.tensors.map(function(s){return B.reshape(s,t)});return B.stack(a,0)})},n.prototype.popBack=function(t,e){if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(this.size()===0)throw new Error("Trying to pop from an empty list.");var i=this.tensors.pop();return mn(i.shape,t,"TensorList shape mismatch: "),B.reshape(i,t)},n.prototype.pushBack=function(t){if(t.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+t.dtype+", but list elements "+this.elementDtype);if(mn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");B.keep(t),this.tensors.push(t)},n.prototype.resize=function(t){if(t<0)throw new Error("TensorListResize expects size to be non-negative. Got: "+t);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return mn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");mn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return mn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return mn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function RU(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);mn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new jo(a,t,i)}function OU(n,t,e){return new jo([],n,t,e)}function EU(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+"  vs. "+i+")");var a=new jo([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function DU(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to
+        `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;c<t.length;++c){var h=c===0?0:a[c-1],d=[0,h,0],p=[1,t[c],s];o[c]=B.reshape(B.slice(e,d,p),i.elementShape)}return o});for(var l=[],u=0;u<t.length;u++)l[u]=u;this.writeMany(l,o)},n}();var jo=function(){function n(t,e,i,r){r===void 0&&(r=-1),this.tensors=t,this.elementShape=e,this.elementDtype=i,t!=null&&t.forEach(function(a){if(i!==a.dtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+a.dtype);fn(e,a.shape,"TensorList shape mismatch: "),B.keep(a)}),this.idTensor=B.scalar(0),this.maxNumElements=r,B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),n.prototype.copy=function(){return new n(this.tensors.slice(),this.elementShape,this.elementDtype)},n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.stack=function(t,e,i){var r=this;if(i===void 0&&(i=-1),e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(i!==-1&&this.tensors.length!==i)throw new Error("Operation expected a list with "+i+" elements but got a list with "+this.tensors.length+" elements.");return fn(t,this.elementShape,"TensorList shape mismatch: "),B.tidy(function(){var a=r.tensors.map(function(s){return B.reshape(s,t)});return B.stack(a,0)})},n.prototype.popBack=function(t,e){if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(this.size()===0)throw new Error("Trying to pop from an empty list.");var i=this.tensors.pop();return fn(i.shape,t,"TensorList shape mismatch: "),B.reshape(i,t)},n.prototype.pushBack=function(t){if(t.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+t.dtype+", but list elements "+this.elementDtype);if(fn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");B.keep(t),this.tensors.push(t)},n.prototype.resize=function(t){if(t<0)throw new Error("TensorListResize expects size to be non-negative. Got: "+t);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return fn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");fn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return fn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return fn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function RU(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);fn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new jo(a,t,i)}function OU(n,t,e){return new jo([],n,t,e)}function EU(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+"  vs. "+i+")");var a=new jo([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function DU(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to
           tensor.shape[0], but sum of lengths is
-        `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c<t.length;++c){var h=c===0?0:r[c-1],d=[0,h,0],p=[1,t[c],a];u[c]=B.reshape(B.slice(n,d,p),e)}return n.dispose(),u}),o=new jo([],e,n.dtype,t.length),l=0;l<s.length;l++)o.setItem(l,s[l]);return o}var kU=void 0,FU=function(n,t,e){return Vn(kU,void 0,void 0,function(){var i,r,a,s,o,l,u,c,o,h,d,l,p,f,m,m,g,v,g,b,g,g,g,w,S,L,N,C,R,D,k,W,F,P,H,_,j,K,q,G,Z,X,ee,ne,ie,te,re,le,he,ye,Oe,Fe,_e,Me,dt,Lt,Ge,Ue,F,P,me,Ue,j,L,Ft,me,ne,ie,L,lt,me,L,pt,lt,me,q,G,L,pt,me,Ue,L,pt,lt,me,En,L,pt,me,re,me,he,L,Ue,P,me,Ue,L,Ft,me,Oe,L,Fe,me;return Nn(this,function(Et){switch(Et.label){case 0:i=n.op;switch(i){case"If":return[3,1];case"StatelessIf":return[3,1];case"While":return[3,3];case"StatelessWhile":return[3,3];case"LoopCond":return[3,9];case"Switch":return[3,10];case"Merge":return[3,12];case"Enter":return[3,13];case"Exit":return[3,14];case"NextIteration":return[3,15];case"TensorArrayV3":return[3,16];case"TensorArrayWriteV3":return[3,17];case"TensorArrayReadV3":return[3,18];case"TensorArrayGatherV3":return[3,19];case"TensorArrayScatterV3":return[3,20];case"TensorArrayConcatV3":return[3,21];case"TensorArraySplitV3":return[3,22];case"TensorArraySizeV3":return[3,23];case"TensorArrayCloseV3":return[3,24];case"TensorListSetItem":return[3,25];case"TensorListGetItem":return[3,26];case"TensorListScatterV2":return[3,27];case"TensorListScatter":return[3,27];case"TensorListReserve":return[3,28];case"TensorListGather":return[3,29];case"TensorListStack":return[3,30];case"TensorListFromTensor":return[3,31];case"TensorListConcat":return[3,32];case"TensorListPushBack":return[3,33];case"TensorListPopBack":return[3,34];case"TensorListSplit":return[3,35]}return[3,36];case 1:return r=A("thenBranch",n,t,e),a=A("elseBranch",n,t,e),s=A("cond",n,t,e),o=A("args",n,t,e),[4,s.data()];case 2:return l=Et.sent(),l[0]?[2,e.functionMap[r].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)]:[2,e.functionMap[a].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 3:return u=A("body",n,t,e),c=A("cond",n,t,e),o=A("args",n,t,e),[4,e.functionMap[c].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 4:return h=Et.sent(),d=o.map(function(It){return It.id}),[4,h[0].data()];case 5:l=Et.sent(),h.forEach(function(It){!It.kept&&d.indexOf(It.id)===-1&&It.dispose()}),p=o,f=function(){var It,Ce,Wt;return Nn(this,function(ft){switch(ft.label){case 0:return It=p,[4,e.functionMap[u].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 1:return p=ft.sent(),Ce=p.map(function(it){return it.id}),It.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[4,e.functionMap[c].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 2:return Wt=ft.sent(),[4,Wt[0].data()];case 3:return l=ft.sent(),Wt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[2]}})},Et.label=6;case 6:return l[0]?[5,f()]:[3,8];case 7:return Et.sent(),[3,6];case 8:return[2,p];case 9:return m=A("pred",n,t,e),[2,[oi(m)]];case 10:return m=A("pred",n,t,e),g=A("data",n,t,e),g.kept||(g=oi(g)),[4,m.data()];case 11:return[2,Et.sent()[0]?[void 0,g]:[g,void 0]];case 12:return v=n.inputNames.find(function(It){return Gt(It,t,e)!==void 0}),v?(g=Gt(v,t,e),[2,[oi(g)]]):[2,void 0];case 13:return b=A("frameName",n,t,e),g=A("tensor",n,t,e),e.enterFrame(b),[2,[oi(g)]];case 14:return g=A("tensor",n,t,e),e.exitFrame(),[2,[oi(g)]];case 15:return g=A("tensor",n,t,e),e.nextIteration(),[2,[oi(g)]];case 16:return w=A("size",n,t,e),S=A("dtype",n,t,e),L=A("elementShape",n,t,e),N=A("dynamicSize",n,t,e),C=A("clearAfterRead",n,t,e),R=A("identicalElementShapes",n,t,e),D=A("name",n,t,e),k=new CU(D,S,w,L,R,N,C),e.addTensorArray(k),[2,[k.idTensor,B.scalar(1)]];case 17:return W=A("tensorArrayId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),H=e.getTensorArray(W.id),H.write(F,P),[2,[H.idTensor]];case 18:return _=A("tensorArrayId",n,t,e),j=A("index",n,t,e),K=e.getTensorArray(_.id),[2,[K.read(j)]];case 19:return q=A("tensorArrayId",n,t,e),G=A("indices",n,t,e),Z=A("dtype",n,t,e),X=e.getTensorArray(q.id),[2,[X.gather(G,Z)]];case 20:return ee=A("tensorArrayId",n,t,e),ne=A("indices",n,t,e),ie=A("tensor",n,t,e),te=e.getTensorArray(ee.id),te.scatter(ne,ie),[2,[te.idTensor]];case 21:return re=A("tensorArrayId",n,t,e),le=e.getTensorArray(re.id),he=A("dtype",n,t,e),[2,[le.concat(he)]];case 22:return ye=A("tensorArrayId",n,t,e),Oe=A("tensor",n,t,e),Fe=A("lengths",n,t,e),_e=e.getTensorArray(ye.id),_e.split(Fe,Oe),[2,[_e.idTensor]];case 23:return Me=A("tensorArrayId",n,t,e),dt=e.getTensorArray(Me.id),[2,[B.scalar(dt.size(),"int32")]];case 24:return Lt=A("tensorArrayId",n,t,e),Ge=e.getTensorArray(Lt.id),Ge.clearAndClose(),[2,[Ge.idTensor]];case 25:return Ue=A("tensorListId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.setItem(F,P),[2,[me.idTensor]];case 26:return Ue=A("tensorListId",n,t,e),j=A("index",n,t,e),L=A("elementShape",n,t,e),Ft=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.getItem(j,L,Ft)]];case 27:return ne=A("indices",n,t,e),ie=A("tensor",n,t,e),L=A("elementShape",n,t,e),lt=A("numElements",n,t,e),me=EU(ie,ne,L,lt),e.addTensorList(me),[2,[me.idTensor]];case 28:return L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=OU(L,pt,lt),e.addTensorList(me),[2,[me.idTensor]];case 29:return q=A("tensorListId",n,t,e),G=A("indices",n,t,e),L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),me=e.getTensorList(q.id),[2,[me.gather(G,pt,L)]];case 30:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=e.getTensorList(Ue.id),[2,[me.stack(L,pt,lt)]];case 31:return En=A("tensor",n,t,e),L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),me=RU(En,L,pt),e.addTensorList(me),[2,[me.idTensor]];case 32:return re=A("tensorListId",n,t,e),me=e.getTensorList(re.id),he=A("dtype",n,t,e),L=A("elementShape",n,t,e),[2,[me.concat(he,L)]];case 33:return Ue=A("tensorListId",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.pushBack(P),[2,[me.idTensor]];case 34:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),Ft=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.popBack(L,Ft)]];case 35:return Oe=A("tensor",n,t,e),L=A("elementShape",n,t,e),Fe=A("lengths",n,t,e),me=DU(Oe,Fe,L),e.addTensorList(me),[2,[me.idTensor]];case 36:throw TypeError("Node type "+n.op+" is not implemented")}})})};function zw(n,t,e){var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=r==="fusedbatchnorm",u=A("numArgs",n,t,e);if(s){if(o&&u!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&u!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");var c=A("strides",n,t,e),h=Yo(n,t,e),d=A("dataFormat",n,t,e).toUpperCase(),p=A("dilations",n,t,e),f=A("args",n,t,e),m=f[0],g=f[1];return{stride:c,pad:h,dataFormat:d,dilations:p,biasArg:m,preluArg:g,activationFunc:a}}var WU=function(n,t,e){switch(n.op){case"Conv1D":{var i=A("stride",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),s=A("dilation",n,t,e);return[B.conv1d(A("x",n,t,e),A("filter",n,t,e),i,r,a,s)]}case"Conv2D":{var i=A("strides",n,t,e),r=Yo(n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv2d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{var l=zw(n,t,e),i=l.stride,r=l.pad,a=l.dataFormat,o=l.dilations,u=l.biasArg,c=l.preluArg,h=l.activationFunc;return[B.fused.conv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"FusedDepthwiseConv2dNative":{var d=zw(n,t,e),i=d.stride,r=d.pad,a=d.dataFormat,o=d.dilations,u=d.biasArg,c=d.preluArg,h=d.activationFunc;return[B.fused.depthwiseConv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{var p=A("outputShape",n,t,e),i=A("strides",n,t,e),r=Yo(n,t,e);return[B.conv2dTranspose(A("x",n,t,e),A("filter",n,t,e),p,[i[1],i[2]],r)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{var i=A("strides",n,t,e),r=Yo(n,t,e),o=A("dilations",n,t,e),a=A("dataFormat",n,t,e).toUpperCase();return[B.depthwiseConv2d(A("input",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"Conv3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv3d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2],i[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPoolWithArgmax":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e),m=A("includeBatchInIndex",n,t,e),g=B.maxPoolWithArgmax(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r,m),v=g.result,b=g.indexes;return[v,b]}case"AvgPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"MaxPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"Dilation2D":{var w=A("strides",n,t,e),r=A("pad",n,t,e),o=A("dilations",n,t,e),S=w[1],L=w[2],N=o[1],C=o[2];return[B.dilation2d(A("x",n,t,e),A("filter",n,t,e),[S,L],r,[N,C],"NHWC")]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var UU=function(n,t,e){switch(n.op){case"Fill":{var i=A("shape",n,t,e),r=A("dtype",n,t,e),a=A("value",n,t,e);return[B.fill(i,a,r)]}case"LinSpace":{var s=A("start",n,t,e),o=A("stop",n,t,e),l=A("num",n,t,e);return[B.linspace(s,o,l)]}case"Multinomial":{var u=A("logits",n,t,e),c=A("numSamples",n,t,e),h=A("seed",n,t,e);return[B.multinomial(u,c,h)]}case"OneHot":{var d=A("indices",n,t,e),p=A("depth",n,t,e),f=A("onValue",n,t,e),m=A("offValue",n,t,e);return[B.oneHot(d,p,f,m)]}case"Ones":return[B.ones(A("shape",n,t,e),A("dtype",n,t,e))];case"OnesLike":return[B.onesLike(A("x",n,t,e))];case"RandomUniform":return[B.randomUniform(A("shape",n,t,e),A("minval",n,t,e),A("maxval",n,t,e),A("dtype",n,t,e))];case"Range":{var s=A("start",n,t,e),g=A("stop",n,t,e),v=A("step",n,t,e);return[B.range(s,g,v,A("dtype",n,t,e))]}case"TruncatedNormal":{var i=A("shape",n,t,e),b=A("mean",n,t,e),w=A("stdDev",n,t,e),h=A("seed",n,t,e);return[B.truncatedNormal(i,b,w,A("dtype",n,t,e),h)]}case"Zeros":return[B.zeros(A("shape",n,t,e),A("dtype",n,t,e))];case"ZerosLike":return[B.zerosLike(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var BU=void 0;function Pd(n,t,e){var i=A("boxes",n,t,e),r=A("scores",n,t,e),a=A("maxOutputSize",n,t,e),s=A("iouThreshold",n,t,e),o=A("scoreThreshold",n,t,e),l=A("softNmsSigma",n,t,e);return{boxes:i,scores:r,maxOutputSize:a,iouThreshold:s,scoreThreshold:o,softNmsSigma:l}}var zU=function(n,t,e){return Vn(BU,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,a,s,o,l,u,p,h,f,a,s,o,l,u,m,h;return Nn(this,function(g){switch(g.label){case 0:i=n.op;switch(i){case"NonMaxSuppressionV5":return[3,1];case"NonMaxSuppressionV4":return[3,3];case"NonMaxSuppressionV3":return[3,5];case"NonMaxSuppressionV2":return[3,5];case"Where":return[3,7];case"ListDiff":return[3,9]}return[3,10];case 1:return r=Pd(n,t,e),a=r.boxes,s=r.scores,o=r.maxOutputSize,l=r.iouThreshold,u=r.scoreThreshold,c=r.softNmsSigma,[4,B.image.nonMaxSuppressionWithScoreAsync(a,s,o,l,u,c)];case 2:return h=g.sent(),[2,[h.selectedIndices,h.selectedScores]];case 3:return d=Pd(n,t,e),a=d.boxes,s=d.scores,o=d.maxOutputSize,l=d.iouThreshold,u=d.scoreThreshold,p=A("padToMaxOutputSize",n,t,e),[4,B.image.nonMaxSuppressionPaddedAsync(a,s,o,l,u,p)];case 4:return h=g.sent(),[2,[h.selectedIndices,h.validOutputs]];case 5:return f=Pd(n,t,e),a=f.boxes,s=f.scores,o=f.maxOutputSize,l=f.iouThreshold,u=f.scoreThreshold,[4,B.image.nonMaxSuppressionAsync(a,s,o,l,u)];case 6:return[2,[g.sent()]];case 7:return m=B.cast(A("condition",n,t,e),"bool"),[4,B.whereAsync(m)];case 8:return h=[g.sent()],m.dispose(),[2,h];case 9:return[2,B.setdiff1dAsync(A("x",n,t,e),A("y",n,t,e))];case 10:throw TypeError("Node type "+n.op+" is not implemented")}})})};var PU=function(n,t,e){switch(n.op){case"TopKV2":{var i=A("x",n,t,e),r=A("k",n,t,e),a=A("sorted",n,t,e),s=B.topk(i,r,a);return[s.values,s.indices]}case"Unique":{var i=A("x",n,t,e),s=B.unique(i);return[s.values,s.indices]}case"UniqueV2":{var i=A("x",n,t,e),o=A("axis",n,t,e),s=B.unique(i,o);return[s.values,s.indices]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var _U=function(n,t,e){switch(n.op){case"Const":return t[n.name];case"PlaceholderWithDefault":var i=A("default",n,t,e);return[Gt(n.name,t,e)||i];case"Placeholder":return[Gt(n.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{var r=A("x",n,t,e);return[oi(r)]}case"IdentityN":return A("x",n,t,e).map(function(h){return oi(h)});case"Snapshot":var a=A("x",n,t,e);return[oi(a)];case"Shape":return[B.tensor1d(A("x",n,t,e).shape,"int32")];case"ShapeN":return A("x",n,t,e).map(function(h){return B.tensor1d(h.shape)});case"Size":return[B.scalar(A("x",n,t,e).size,"int32")];case"Rank":return[B.scalar(A("x",n,t,e).rank,"int32")];case"NoOp":return[B.scalar(1)];case"Print":var s=A("x",n,t,e),o=A("data",n,t,e),l=A("message",n,t,e),u=A("summarize",n,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(l);for(var c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,u));return[s];default:throw TypeError("Node type "+n.op+" is not implemented")}};var MU=function(n,t,e){switch(n.op){case"ResizeBilinear":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeBilinear(i,[r[0],r[1]],a)]}case"ResizeNearestNeighbor":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeNearestNeighbor(i,[r[0],r[1]],a)]}case"CropAndResize":{var s=A("image",n,t,e),o=A("boxes",n,t,e),l=A("boxInd",n,t,e),u=A("cropSize",n,t,e),c=A("method",n,t,e),h=A("extrapolationValue",n,t,e);return[B.image.cropAndResize(s,o,l,u,c,h)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var HU=function(n,t,e){switch(n.op){case"Equal":return[B.equal(A("a",n,t,e),A("b",n,t,e))];case"NotEqual":return[B.notEqual(A("a",n,t,e),A("b",n,t,e))];case"Greater":return[B.greater(A("a",n,t,e),A("b",n,t,e))];case"GreaterEqual":return[B.greaterEqual(A("a",n,t,e),A("b",n,t,e))];case"Less":return[B.less(A("a",n,t,e),A("b",n,t,e))];case"LessEqual":return[B.lessEqual(A("a",n,t,e),A("b",n,t,e))];case"LogicalAnd":return[B.logicalAnd(A("a",n,t,e),A("b",n,t,e))];case"LogicalNot":return[B.logicalNot(A("a",n,t,e))];case"LogicalOr":return[B.logicalOr(A("a",n,t,e),A("b",n,t,e))];case"Select":case"SelectV2":return[B.where(A("condition",n,t,e),A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var VU=function(n,t,e){switch(n.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[B.matMul(A("a",n,t,e),A("b",n,t,e),A("transposeA",n,t,e),A("transposeB",n,t,e))];case"Transpose":return[B.transpose(A("x",n,t,e),A("perm",n,t,e))];case"_FusedMatMul":var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=A("numArgs",n,t,e);if(s){if(o&&l!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&l!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}var u=A("args",n,t,e),c=u[0],h=u[1];return[B.fused.matMul({a:A("a",n,t,e),b:A("b",n,t,e),transposeA:A("transposeA",n,t,e),transposeB:A("transposeB",n,t,e),bias:c,activation:a,preluActivationWeights:h})];default:throw TypeError("Node type "+n.op+" is not implemented")}};var qU=function(n,t,e){switch(n.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"FusedBatchNormV3":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"LRN":return[B.localResponseNormalization(A("x",n,t,e),A("radius",n,t,e),A("bias",n,t,e),A("alpha",n,t,e),A("beta",n,t,e))];case"Softmax":return[B.softmax(A("x",n,t,e))];case"LogSoftmax":return[B.logSoftmax(A("x",n,t,e))];case"SparseToDense":return[B.sparseToDense(A("sparseIndices",n,t,e),A("outputShape",n,t,e),A("sparseValues",n,t,e),A("defaultValue",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var GU=function(n,t,e){switch(n.op){case"Max":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.max(A("x",n,t,e),i,r)]}case"Mean":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.mean(A("x",n,t,e),i,r)]}case"Min":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.min(A("x",n,t,e),i,r)]}case"Sum":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.sum(A("x",n,t,e),i,r)]}case"All":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.all(A("x",n,t,e),i,r)]}case"Any":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.any(A("x",n,t,e),i,r)]}case"ArgMax":{var i=A("axis",n,t,e);return[B.argMax(A("x",n,t,e),i)]}case"ArgMin":{var i=A("axis",n,t,e);return[B.argMin(A("x",n,t,e),i)]}case"Prod":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.prod(A("x",n,t,e),i,r)]}case"Cumsum":{var i=A("axis",n,t,e),a=A("exclusive",n,t,e),s=A("reverse",n,t,e);return[B.cumsum(A("x",n,t,e),i,a,s)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var YU=function(n,t,e){switch(n.op){case"ConcatV2":case"Concat":{var i=A("n",n,t,e),r=A("axis",n,t,e),a=A("tensors",n,t,e);return a=a.slice(0,i),[B.concat(a,r)]}case"GatherV2":case"Gather":{var r=A("axis",n,t,e),s=A("x",n,t,e),o=A("indices",n,t,e);return[B.gather(s,B.cast(o,"int32"),r)]}case"ReverseV2":case"Reverse":{var r=A("axis",n,t,e),s=A("x",n,t,e);return[B.reverse(s,r)]}case"Slice":{var l=A("begin",n,t,e),u=A("size",n,t,e);return[B.slice(A("x",n,t,e),l,u)]}case"StridedSlice":{var l=A("begin",n,t,e),c=A("end",n,t,e),h=A("strides",n,t,e),d=A("beginMask",n,t,e),p=A("endMask",n,t,e),f=A("ellipsisMask",n,t,e),m=A("newAxisMask",n,t,e),g=A("shrinkAxisMask",n,t,e),v=A("x",n,t,e);return[B.stridedSlice(v,l,c,h,d,p,f,m,g)]}case"Pack":return B.tidy(function(){var D=A("axis",n,t,e),k=A("tensors",n,t,e),W=k[0].shape,F=B.squeeze(k[0]).shape,P=k.map(function(H){var _=B.util.arraysEqual(H.shape,W);if(!_&&!B.util.arraysEqual(B.squeeze(H).shape,F))throw new Error("the input tensors shape does not match");return _?H:B.reshape(H,W)});return[B.stack(P,D)]});case"Unpack":{var r=A("axis",n,t,e),v=A("tensor",n,t,e);return B.unstack(v,r)}case"Tile":{var b=A("reps",n,t,e);return[B.tile(A("x",n,t,e),b)]}case"Split":case"SplitV":{var r=A("axis",n,t,e),w=A("numOrSizeSplits",n,t,e),v=A("x",n,t,e);return B.split(v,w,r)}case"ScatterNd":{var o=A("indices",n,t,e),S=A("values",n,t,e),L=A("shape",n,t,e);return[B.scatterND(o,S,L)]}case"GatherNd":{var N=A("x",n,t,e),o=A("indices",n,t,e);return[B.gatherND(N,o)]}case"SparseToDense":{var o=A("sparseIndices",n,t,e),L=A("outputShape",n,t,e),C=A("sparseValues",n,t,e),R=A("defaultValue",n,t,e);return[B.sparseToDense(o,C,L,C.dtype===R.dtype?R:B.cast(R,C.dtype))]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var jU=function(n,t,e){switch(n.op){case"FFT":return[B.fft(A("x",n,t,e))];case"IFFT":return[B.ifft(A("x",n,t,e))];case"RFFT":return[B.rfft(A("x",n,t,e))];case"IRFFT":return[B.irfft(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var KU=function(n,t,e){switch(n.op){case"Cast":return[B.cast(A("x",n,t,e),A("dtype",n,t,e))];case"ExpandDims":{var i=A("axis",n,t,e);return[B.expandDims(A("x",n,t,e),i)]}case"Squeeze":{var i=A("axis",n,t,e);return[B.squeeze(A("x",n,t,e),i)]}case"Reshape":return[B.reshape(A("x",n,t,e),A("shape",n,t,e))];case"PadV2":case"Pad":return[B.pad(A("x",n,t,e),A("padding",n,t,e),A("constantValue",n,t,e))];case"SpaceToBatchND":{var r=A("blockShape",n,t,e),a=A("paddings",n,t,e);return[B.spaceToBatchND(A("x",n,t,e),r,a)]}case"BatchToSpaceND":{var r=A("blockShape",n,t,e),s=A("crops",n,t,e);return[B.batchToSpaceND(A("x",n,t,e),r,s)]}case"DepthToSpace":{var o=A("blockSize",n,t,e),l=A("dataFormat",n,t,e).toUpperCase();return[B.depthToSpace(A("x",n,t,e),o,l)]}case"BroadcastTo":return[B.broadcastTo(A("x",n,t,e),A("shape",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function Pw(n,t,e){var i=function(r,a,s){switch(r.category){case"arithmetic":return B.tidy(function(){return TU(r,a,s)});case"basic_math":return B.tidy(function(){return NU(r,a,s)});case"control":return FU(r,a,s);case"convolution":return B.tidy(function(){return WU(r,a,s)});case"creation":return B.tidy(function(){return UU(r,a,s)});case"dynamic":return zU(r,a,s);case"evaluation":return B.tidy(function(){return PU(r,a,s)});case"image":return B.tidy(function(){return MU(r,a,s)});case"graph":return B.tidy(function(){return _U(r,a,s)});case"logical":return B.tidy(function(){return HU(r,a,s)});case"matrices":return B.tidy(function(){return VU(r,a,s)});case"normalization":return B.tidy(function(){return qU(r,a,s)});case"reduction":return B.tidy(function(){return GU(r,a,s)});case"slice_join":return B.tidy(function(){return YU(r,a,s)});case"spectral":return B.tidy(function(){return jU(r,a,s)});case"transformation":return B.tidy(function(){return KU(r,a,s)});case"custom":var o=kw(r.op);if(o&&o.customExecutor)return o.customExecutor(new AU(r,a,s));throw TypeError("Custom op "+r.op+" is not registered.");default:throw TypeError("Unknown op '"+r.op+"'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()")}}(n,t,e);return i instanceof Promise?i.then(function(r){return[].concat(r)}):[].concat(i)}var _w=function(){function n(t,e,i,r){t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i={}),r===void 0&&(r={}),this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=i,this.functionMap=r,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}return n.prototype.newFrame=function(t,e){return{id:t,frameName:e,iterationId:0}},Object.defineProperty(n.prototype,"currentContext",{get:function(){return this.contexts},set:function(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextId",{get:function(){return this._currentContextIds[0]},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextIds",{get:function(){return this._currentContextIds},enumerable:!0,configurable:!0}),n.prototype.generateCurrentContextIds=function(){for(var t=[],e=0;e<this.contexts.length-1;e++){var i=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(i))}t.push(""),this._currentContextIds=t},n.prototype.contextIdforContexts=function(t){return t?t.map(function(e){return e.id===0&&e.iterationId===0?"":e.frameName+"-"+e.iterationId}).join("/"):""},n.prototype.enterFrame=function(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))},n.prototype.exitFrame=function(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function Hw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return Zt(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return Zt(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((Mw(d)||$U(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function XU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return Zt(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var JU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],ZU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function Mw(n){return JU.indexOf(n.op)>=0}function $U(n){return ZU.indexOf(n.op)>=0}var Vw=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=Hw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return XU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[Zt(d)[0]]}),s=e.map(function(d){return Zt(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new _w(i.weightMap,c,h,i.functionExecutorMap),p=Ew({},i.weightMap);Object.keys(t).forEach(function(w){var S=Zt(w),L=S[0],N=S[1],C=[];C[N]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g<u.length;g++){var v=u[g];if(!p[v.name]){var b=Pw(v,p,d);if(b instanceof Promise)throw new Error("The execution of the op '"+v.op+"' returned a promise. Please use model.executeAsync() instead.");p[v.name]=b,i.checkTensorForDisposal(v.name,v,p,d,f,s,m)}}return i.parent==null&&d.dispose(f),e.map(function(w){return Gt(w,p,d)})})},n.prototype.getFrozenTensorIds=function(t){var e=[].concat.apply([],Object.keys(t).map(function(i){return t[i]}).map(function(i){return i.map(function(r){return r.id})}));return new Set(e)},n.prototype.checkTensorForDisposal=function(t,e,i,r,a,s,o){if(e.category==="control"||s.indexOf(t)!==-1)return;i[t].forEach(function(l){l!=null&&(o[l.id]=(o[l.id]||0)+e.children.length)}),e.inputs.forEach(function(l){if(l.category!=="control"){var u=qW(l.name,i,r);u!=null&&u.forEach(function(c){if(c&&!a.has(c.id)){var h=o[c.id];h===1?(c.dispose(),delete o[c.id]):h!=null&&o[c.id]--}})}})},n.prototype.executeAsync=function(t,e){return Vn(this,void 0,void 0,function(){return Nn(this,function(i){return[2,this._executeAsync(t,e)]})})},n.prototype._executeAsync=function(t,e,i,r,a){return i===void 0&&(i=!1),r===void 0&&(r={}),a===void 0&&(a={}),Vn(this,void 0,void 0,function(){var s,o,l,u,c,h;return Nn(this,function(d){switch(d.label){case 0:return i||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e)),s=new _w(this.weightMap,r,a,this.functionExecutorMap),[4,this.executeWithControlFlow(t,s,e,i)];case 1:return o=d.sent(),l=e.map(function(p){return Gt(p,o,s)}),u=l.map(function(p){return p.id}),c=Object.keys(t).map(function(p){return t[p].id}),h=new Set(u.concat(c,this.weightIds)),Object.keys(o).forEach(function(p){var f=o[p];f.forEach(function(m){m&&!m.isDisposed&&!h.has(m.id)&&m.dispose()})}),this.parent==null&&s.dispose(h),[2,l]}})})},n.prototype.executeFunctionAsync=function(t,e,i){return Vn(this,void 0,void 0,function(){var r,a=this;return Nn(this,function(s){return r=t.reduce(function(o,l,u){return o[a.inputs[u].name]=l,o},{}),[2,this._executeAsync(r,this.outputNodes,!0,e,i)]})})},n.prototype.executeWithControlFlow=function(t,e,i,r){return Vn(this,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N=this;return Nn(this,function(C){switch(C.label){case 0:a=Object.keys(t),s=a.map(function(R){return N.graph.nodes[Zt(R)[0]]}),o=i.map(function(R){return Zt(R)[0]}),l=o.map(function(R){return N.graph.nodes[R]}),u=Hw(t,l,this.weightMap),c=u.usedNodes,h=u.missingInputs,d=u.dynamicNode,p=u.syncInputs,f=s.concat(this.graph.weights).map(function(R){return{node:R,contexts:e.currentContext}}),m=Ew({},this.weightMap),Object.keys(t).forEach(function(R){var D=Zt(R),k=D[0],W=D[1],F=[];F[W]=t[R],m[k]=F}),g={},v=this.getFrozenTensorIds(m),b={},C.label=1;case 1:return f.length>0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!Mw(R)&&!Gt(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=si(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=Pw(f.node,r,i);m||(m=si(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=si(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Gt(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Gt(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=Zt(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=Zt(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=Zt(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var QU="?tfjs-format=file",eB="model.json",qw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return Vn(this,void 0,void 0,function(){var t;return Nn(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new Vw(Ww.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Ww.Instance.transformGraph(t.modelInitializer);this.initializer=new Vw(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return Vn(this,void 0,void 0,function(){var i;return Nn(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return Vn(this,void 0,void 0,function(){var i;return Nn(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function tB(n,t){return t===void 0&&(t={}),Vn(this,void 0,void 0,function(){var e;return Nn(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+eB+QU)),e=new qw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var nB="2.6.0";gr.GraphModel=qw;gr.deregisterOp=VW;gr.loadGraphModel=tB;gr.registerOp=HW;gr.version_converter=nB});var h0=Le(Yt=>{"use strict";Object.defineProperty(Yt,"__esModule",{value:!0});var Ne=lr();var _d=function(n,t){return _d=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},_d(n,t)};function Ve(n,t){_d(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var la=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function vr(n,t){return t={exports:{}},n(t,t.exports),t.exports}var iB=vr(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(la,n,!1)}),rB=vr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(la,n,!1)}),aB=vr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(la,n,!1)}),sB=vr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(la,n,!1)}),oB=vr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(la,n,!1)}),lB=vr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(la,n,!1)}),yr=vr(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,N){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||N||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,N=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(N||(S=[N++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%N]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,j=C.S;W--;)F=j[H=h&H+1],P=P*r+j[h&(j[H]=j[_=h&_+F])+(j[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var N=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{N.push(g(S[R],L-1))}catch(D){}return N.length?N:C=="string"?S:S+"\0"}function v(S,L){for(var N=S+"",C,R=0;R<N.length;)L[h&R]=h&(C^=L[h&R]*19)+N.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,N=L&&L.plugins;return[+new Date,i,N,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=require("crypto")}catch(S){}}})([],Math)});yr.alea=iB;yr.xor128=rB;yr.xorwow=aB;yr.xorshift7=sB;yr.xor4096=oB;yr.tychei=lB;var uB=yr,Yw=uB.alea;function cB(n,t){return Ko(n,t)}function Ko(n,t,e,i){if(e===void 0&&(e=new Map),i===void 0&&(i=new Set),n==null)return null;if(i.has(n))throw new Error("Circular references are not supported.");if(e.has(n))return e.get(n);var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if(ua(n)){var a=Array.isArray(n)?[]:{};i.add(n);for(var s in n){var o=n[s],l=Ko(o,t,e,i);a[s]=l}return i.delete(n),a}else throw new Error("Can't recurse into non-iterable type: "+n);else return e.set(n,r.value),r.value}function hB(n,t){return t===void 0&&(t=Kw),jw(n,t)}function jw(n,t,e){e===void 0&&(e=new Set);var i=n[0];if(e.has(i))throw new Error("Circular references are not supported.");var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if(ua(i)){var a=Array.isArray(i)?[]:{};e.add(i);var s=function(l){var u=n.map(function(h){return h[l]}),c=jw(u,t,e);a[l]=c};for(var o in i)s(o);return e.delete(i),a}else throw new Error("Can't recurse into non-iterable type: "+i);else return r.value}function Kw(n){return n===null?null:ua(n[0])?{value:null,recurse:!0}:{value:n,recurse:!1}}function $w(n,t){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:e=new Map,Ko(n,t,e),i=0,r=Array.from(e.keys()),u.label=1;case 1:return i<r.length?(a=r[i],s=e.get(a),s instanceof Promise?[4,s]:[3,3]):[3,4];case 2:o=u.sent(),e.set(a,o),u.label=3;case 3:return i++,[3,1];case 4:return l=Ko(n,t,e),[2,l]}})})}function ua(n){return n!=null&&!ArrayBuffer.isView(n)&&(Array.isArray(n)||typeof n=="object"&&!(n instanceof Ne.Tensor))}function pB(n){return n==null||dB(n)||Array.isArray(n)||typeof n=="object"&&n instanceof Ne.Tensor||Ne.util.isTypedArray(n)}function dB(n){return n===null||typeof n!="object"&&typeof n!="function"}function mB(n){return cB(n,fB)}function fB(n){return n instanceof Ne.Tensor?{value:n.clone(),recurse:!1}:ua(n)?{value:null,recurse:!0}:{value:n,recurse:!1}}var Xw=function(){function n(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}return n.prototype.wrap=function(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity},n.prototype.get=function(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]},n.prototype.set=function(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e},n.prototype.length=function(){var t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t},n.prototype.isFull=function(){return this.length()===this.capacity},n.prototype.isEmpty=function(){return this.length()===0},n.prototype.push=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)},n.prototype.pushAll=function(t){for(var e=0,i=t;e<i.length;e++){var r=i[e];this.push(r)}},n.prototype.pop=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);var t=this.get(this.end);return this.set(this.end,void 0),t},n.prototype.unshift=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)},n.prototype.shift=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t},n.prototype.shuffleExcise=function(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var e=this.wrap(this.begin+t),i=this.get(e);return this.set(e,this.pop()),i},n}();var gB=function(n){Ve(t,n);function t(){return n.call(this,t.INITIAL_CAPACITY)||this}return t.prototype.isFull=function(){return!1},t.prototype.push=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.push.call(this,e)},t.prototype.unshift=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.unshift.call(this,e)},t.prototype.expand=function(){for(var e=this.capacity*2,i=new Array(e),r=this.length(),a=0;a<r;a++)i[a]=this.get(this.wrap(this.begin+a));this.data=i,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=r},t.INITIAL_CAPACITY=32,t}(Xw);function Jw(n){return new vB(n)}function Md(n){return new yB(n)}function bB(n,t){return new Zw(n,t)}function SB(n,t){return t===void 0&&(t=Fi.FAIL),new wB(n,t)}var xt=function(){function n(){}return n.prototype.toArray=function(){return se(this,void 0,void 0,function(){var t,e;return oe(this,function(i){switch(i.label){case 0:return t=[],[4,this.next()];case 1:e=i.sent(),i.label=2;case 2:return e.done?[3,4]:(t.push(e.value),[4,this.next()]);case 3:return e=i.sent(),[3,2];case 4:return[2,t]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){var t,e,i;return oe(this,function(r){switch(r.label){case 0:return t=this.prefetch(100),e=[],[4,t.next()];case 1:i=r.sent(),r.label=2;case 2:return i.done?[3,4]:(e.push(i.value),[4,t.next()]);case 3:return i=r.sent(),[3,2];case 4:return[2,e]}})})},n.prototype.resolveFully=function(){return se(this,void 0,void 0,function(){var t;return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:t=e.sent(),e.label=2;case 2:return t.done?[3,4]:[4,this.next()];case 3:return t=e.sent(),[3,2];case 4:return[2]}})})},n.prototype.resolveWhile=function(t){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:return[4,this.next()];case 1:e=r.sent(),i=t(e.value),r.label=2;case 2:return!e.done&&i?[4,this.next()]:[3,4];case 3:return e=r.sent(),i=t(e.value),[3,2];case 4:return[2]}})})},n.prototype.handleErrors=function(t){return new CB(this,t)},n.prototype.filter=function(t){return new NB(this,t)},n.prototype.map=function(t){return new xB(this,t)},n.prototype.mapAsync=function(t){return new Qw(this,t)},n.prototype.serialMapAsync=function(t){return new Qw(this,t).serial()},n.prototype.flatmap=function(t){return new RB(this,t)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.map(t).resolveFully()]})})},n.prototype.serialForEach=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.serialMapAsync(t).resolveWhile(function(i){return i===!0})]})})},n.prototype.rowMajorBatch=function(t,e){return e===void 0&&(e=!0),new TB(this,t,e)},n.prototype.columnMajorBatch=function(t,e,i){e===void 0&&(e=!0),i===void 0&&(i=Kw);var r=this.rowMajorBatch(t,e);return r.map(function(a){return hB(a,i)})},n.prototype.concatenate=function(t,e){return new Zw(Jw([this,t]),e)},n.prototype.take=function(t){return t<0||t==null?this:new AB(this,t)},n.prototype.skip=function(t){return t<0||t==null?this:new IB(this,t)},n.prototype.prefetch=function(t){return new e0(this,t)},n.prototype.shuffle=function(t,e){return new OB(this,t,e)},n.prototype.serial=function(){return new LB(this)},n}(),vB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.items=e,i.trav=0,i}return t.prototype.summary=function(){return"Array of "+this.items.length+" items"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return this.trav>=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:mB(e),done:!1}])})})},t}(xt),yB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(xt),LB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(xt),IB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++<this.maxCount?[4,this.upstream.next()]:[3,2];case 1:return e=i.sent(),e.done?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2,this.upstream.next()]}})})},t}(xt),AB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(xt),TB=function(n){Ve(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length<this.batchSize?[4,this.upstream.next()]:[3,3];case 2:return i=r.sent(),i.done?this.enableSmallLastBatch&&e.length>0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(xt),NB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(xt),xB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(xt),CB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.handler=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(xt),Qw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(xt),Hd=function(n){Ve(t,n);function t(){var e=n.call(this)||this;return e.outputQueue=new gB,e.lastRead=Promise.resolve({value:null,done:!1}),e}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.outputQueue.length()===0?[4,this.pump()]:[3,2];case 1:return e.sent()?[3,0]:[2,{value:null,done:!0}];case 2:return[2,{value:this.outputQueue.shift(),done:!1}]}})})},t}(xt),RB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,!0]}})})},t}(Hd),Zw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.baseErrorHandler=i,r.lastRead=null,r.iterator=null,r.moreIterators=e,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of chained summaries";return e+" -> Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(xt),Fi;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Fi||(Fi={}));var wB=function(n){Ve(t,n);function t(e,i){i===void 0&&(i=Fi.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof xt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,$w(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Fi.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Fi.SHORTEST:return[2,{value:null,done:!0}];case Fi.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(xt),e0=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new Xw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(xt),OB=function(n){Ve(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=Yw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(e0);var hs=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is
-      `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),Qt(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,EB)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,Qt(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,Qt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return Qt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return Qt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return Qt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,Qt(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=Md(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,bB(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?r=0:r=null,Qt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().skip(t)]}})})},r)},n.prototype.shuffle=function(t,e,i){var r=this;if(i===void 0&&(i=!0),t==null||t<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError("`Dataset.shuffle()` requires bufferSize to be specified.  If your data fits in main memory (for regular JS objects), and/or GPU memory (for `tf.Tensor`s), consider setting "+("bufferSize to the dataset size ("+this.size+" elements)"));var a=this,s=Yw(e||Ne.util.now().toString());return Qt(function(){return se(r,void 0,void 0,function(){var o;return oe(this,function(l){switch(l.label){case 0:return o=s.int32(),i&&(o+=s.int32()),[4,a.iterator()];case 1:return[2,l.sent().shuffle(t,o.toString())]}})})},this.size)},n.prototype.take=function(t){var e=this,i=this,r;return this.size!=null&&this.size>t?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,Qt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function Qt(n,t){return t===void 0&&(t=null),new(function(e){Ve(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(hs))}function DB(n){var t=this;return Qt(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,Jw(n)]})})},n.length)}function kB(n){var t=this;if(!ua(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i<n.length;i++)e=e==null?n[i].size:Math.min(e,n[i].size);else if(n instanceof Object)for(var r in n)e=e==null?n[r].size:Math.min(e,n[r].size);return Qt(function(){return se(t,void 0,void 0,function(){var a;return oe(this,function(s){switch(s.label){case 0:return[4,$w(n,function(o){if(o instanceof hs)return{value:o.iterator(),recurse:!1};if(ua(o))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")})];case 1:return a=s.sent(),[2,SB(a,Fi.SHORTEST)]}})})},e)}function EB(n){if(n===null)return null;var t=n[0];if(pB(t)){var e=FB(n);return{value:e,recurse:!1}}return{value:null,recurse:!0}}function FB(n){if(n.length===0)throw new Error("Can't make a batch of zero elements.");return n[0]instanceof Ne.Tensor?Ne.stack(n):Ne.tensor(n)}var t0=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.input=e,i}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.input.iterator()];case 1:return e=a.sent(),i=e.decodeUTF8(),r=i.split(`
-`).map(function(s){return s.endsWith("\r")&&(s=s.slice(0,-1)),s}),[2,r]}})})},t}(hs);var $o='"',ds=Symbol("out"),n0=Symbol("field"),Xo=Symbol("quote"),Vd=Symbol("quoteafterquote"),i0=Symbol("quoteinquote"),r0=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.input=e,r.hasHeader=!0,r.fullColumnNames=null,r.columnNamesValidated=!1,r.columnConfigs=null,r.configuredColumnsOnly=!1,r.delimiter=",",r.delimWhitespace=!1,r.base=new t0(e),i||(i={}),r.hasHeader=!(i.hasHeader===!1),r.fullColumnNames=i.columnNames,r.columnConfigs=i.columnConfigs,r.configuredColumnsOnly=i.configuredColumnsOnly,i.delimWhitespace?(Ne.util.assert(i.delimiter==null,function(){return"Delimiter should not be provided when delimWhitespace is true."}),r.delimWhitespace=!0,r.delimiter=" "):r.delimiter=i.delimiter?i.delimiter:",",r}return t.prototype.columnNames=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:e.sent(),e.label=2;case 2:return[2,this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames]}})})},t.prototype.setColumnNames=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u=this;return oe(this,function(c){switch(c.label){case 0:return[4,this.maybeReadHeaderLine()];case 1:if(e=c.sent(),!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");if(this.fullColumnNames&&e&&Ne.util.assert(e.length===this.fullColumnNames.length,function(){return"The length of provided columnNames ("+u.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."}),this.fullColumnNames||(this.fullColumnNames=e),i=this.fullColumnNames.reduce(function(h,d){return h[d]=h[d]+1||1,h},{}),r=Object.keys(i).filter(function(h){return i[h]>1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a<s.length;a++)if(o=s[a],l=this.fullColumnNames.indexOf(o),l===-1)throw new Error('The key "'+o+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}return this.columnNamesValidated=!0,[2]}})})},t.prototype.maybeReadHeaderLine=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return this.hasHeader?[4,this.base.iterator()]:[3,3];case 1:return e=s.sent(),[4,e.next()];case 2:if(i=s.sent(),i.done)throw new Error("No data was found for CSV parsing.");return r=i.value,a=this.parseRow(r,!1),[2,a];case 3:return[2,null]}})})},t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i=this;return oe(this,function(r){switch(r.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:r.sent(),r.label=2;case 2:return[4,this.base.iterator()];case 3:return e=r.sent(),this.hasHeader&&(e=e.skip(1)),[2,e.map(function(a){return i.makeDataElement(a)})]}})})},t.prototype.makeDataElement=function(e){for(var i=this.parseRow(e),r={},a={},s=0;s<this.fullColumnNames.length;s++){var o=this.fullColumnNames[s],l=this.columnConfigs?this.columnConfigs[o]:null;if(this.configuredColumnsOnly&&!l)continue;var u=i[s],c=null;if(u==="")if(l&&l.default!==void 0)c=l.default;else{if(l&&(l.required||l.isLabel))throw new Error("Required column "+o+" is empty in this line: "+e);c=void 0}else{var h=Number(u);if(isNaN(h))l&&l.dtype==="bool"?c=this.getBoolean(u):c=u;else if(!l||!l.dtype)c=h;else switch(l.dtype){case"float32":c=h;break;case"int32":c=Math.floor(h);break;case"bool":c=this.getBoolean(u);break;default:c=h}}l&&l.isLabel?a[o]=c:r[o]=c}return Object.keys(a).length===0?r:{xs:r,ys:a}},t.prototype.getBoolean=function(e){return e==="1"||e.toLowerCase()==="true"?1:0},t.prototype.parseRow=function(e,i){i===void 0&&(i=!0);for(var r=[],a=0,s=e.length,o=ds,l=0;l<s;l++)switch(o){case ds:switch(e.charAt(l)){case $o:a=l+1,o=Xo;break;case this.delimiter:if(a=l+1,this.delimiter===" "&&this.delimWhitespace)break;r.push(""),o=ds;break;default:o=n0,a=l;break}break;case n0:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l)),o=ds,a=l+1;break}break;case Xo:switch(e.charAt(l)){case $o:o=Vd;break}break;case Vd:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l-1)),o=ds,a=l+1;break;case $o:o=Xo;break;default:o=i0;break}break;case i0:switch(e.charAt(l)){case $o:o=Xo;break}break}if(o===Vd?r.push(e.substring(a,s-1)):r.push(e.substring(a)),i&&r.length!==this.fullColumnNames.length)throw new Error("Invalid row in csv file. Should have "+this.fullColumnNames.length+" elements in a row, but got "+r);return r},t}(hs);var WB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;i.microphoneConfig=e,i.isClosed=!1,i.fftSize=e.fftSize||1024;var r=Math.log2(i.fftSize);if(i.fftSize<0||r<4||r>14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(xt);var UB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(xt);var a0=function(){function n(){}return n}();var s0=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new BB(this,e)},t}(xt),BB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new zB(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(s0),zB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r<a.length;r++)s=a[r],this.outputQueue.push(s);return this.carryover=i[i.length-1],[2,!0]}})})},t}(Hd);var _B=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.decodeUTF8=function(){return new PB(this)},t}(xt),PB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.impl=new MB(e),i}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(s0),MB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;if(i.upstream=e,Ne.env().get("IS_BROWSER"))i.decoder=new TextDecoder("utf-8");else{var r=require("string_decoder").StringDecoder;i.decoder=new r("utf8")}return i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(Hd);var o0=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(_B);function VB(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=HB(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new o0(a,t)];case 3:throw new Error(r.statusText)}})})}var HB=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function l0(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var u0=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return l0(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new o0(this.input,this.options)]})})},t}(a0);var c0=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return l0(this.url)?[2,new u0(this.url,this.fileOptions).iterator()]:[2,VB(this.url,this.fileOptions)]})})},t}(a0);function qB(n,t){return t===void 0&&(t={}),new r0(new c0(n),t)}function GB(n){var t=this,e=Md(n);return Qt(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function YB(n){var t=this;return Qt(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,Md(function(){return e.next()})]}})})})}function jB(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,UB.create(n,t)]})})}function KB(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,WB.create(n)]})})}var $B="2.6.0";Yt.CSVDataset=r0;Yt.Dataset=hs;Yt.FileDataSource=u0;Yt.TextLineDataset=t0;Yt.URLDataSource=c0;Yt.array=DB;Yt.csv=qB;Yt.func=GB;Yt.generator=YB;Yt.microphone=KB;Yt.version_data=$B;Yt.webcam=jB;Yt.zip=kB});var p0=Le((d0,qd)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c<u.length;c++){o+=u.charCodeAt(c);var h=.02519603282416938*o;o=h>>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(d0,typeof qd=="object"&&qd,typeof define=="function"&&define)});var m0=Le((f0,Gd)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(f0,typeof Gd=="object"&&Gd,typeof define=="function"&&define)});var v0=Le((g0,Yd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,u==l.length&&(o.d=o.x<<10^o.x>>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(g0,typeof Yd=="object"&&Yd,typeof define=="function"&&define)});var b0=Le((y0,jd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h<c.length;++h)p[h&7]=p[h&7]<<15^c.charCodeAt(h)+p[h+1&7]<<13;for(;p.length<8;)p.push(0);for(h=0;h<8&&p[h]===0;++h);for(h==8?d=p[7]=-1:d=p[h],u.x=p,u.i=0,h=256;h>0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(y0,typeof jd=="object"&&jd,typeof define=="function"&&define)});var S0=Le((w0,Kd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f<v;++f)c&&(d^=c.charCodeAt((f+32)%c.length)),f===0&&(m=d),d^=d<<10,d^=d>>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(w0,typeof Kd=="object"&&Kd,typeof define=="function"&&define)});var I0=Le((L0,$d)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u<l.length+20;u++)o.b^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.a=s.a,o.b=s.b,o.c=s.c,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(L0,typeof $d=="object"&&$d,typeof define=="function"&&define)});var A0=Le((vV,Jo)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var N=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),N),R=new p(N),D=function(){for(var k=R.g(r),W=o,F=0;k<l;)k=(k+F)*i,W*=i,F=R.g(1);for(;k>=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,N=this,C=0,R=N.i=N.j=0,D=N.S=[];for(L||(w=[L++]);C<i;)D[C]=C++;for(C=0;C<i;C++)D[C]=D[R=c&R+w[C%L]+(S=D[C])],D[R]=S;(N.g=function(k){for(var W,F=0,P=N.i,H=N.j,_=N.S;k--;)W=_[P=c&P+1],F=F*i+_[c&(_[P]=_[H=c&H+W])+(_[H]=W)];return N.i=P,N.j=H,F})(i)}function f(w,S){return S.i=w.i,S.j=w.j,S.S=w.S.slice(),S}function m(w,S){var L=[],N=typeof w,C;if(S&&N=="object")for(C in w)try{L.push(m(w[C],S-1))}catch(R){}return L.length?L:N=="string"?w:w+"\0"}function g(w,S){for(var L=w+"",N,C=0;C<L.length;)S[c&C]=c&(N^=S[c&C]*19)+L.charCodeAt(C++);return b(S)}function v(){try{var w;return h&&(w=h.randomBytes)?w=w(i):(w=new Uint8Array(i),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(N){var S=e.navigator,L=S&&S.plugins;return[+new Date,e,L,e.screen,b(n)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),n),typeof Jo=="object"&&Jo.exports){Jo.exports=d;try{h=require("crypto")}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return d})})([],Math)});var N0=Le((yV,T0)=>{var XB=p0(),JB=m0(),ZB=v0(),QB=b0(),ez=S0(),tz=I0(),br=A0();br.alea=XB;br.xor128=JB;br.xorwow=ZB;br.xorshift7=QB;br.xor4096=ez;br.tychei=tz;T0.exports=br});var Z0=Le(ps=>{"use strict";Object.defineProperty(ps,"__esModule",{value:!0});var T=lr(),nz=N0();var Xd=function(n,t){return Xd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Xd(n,t)};function iz(n,t){Xd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function x0(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function C0(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}function ae(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&T.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the CPU backend."})})}var rz=T.kernel_impls.nonMaxSuppressionV3Impl,az=T.kernel_impls.split,sz=T.kernel_impls.tile,oz=T.kernel_impls.topkImpl,lz=T.kernel_impls.whereImpl;function Jd(n,t,e,i){if(e==="linear")return n.linear(t);if(e==="relu")return n.relu(t);if(e==="elu")return T.elu(t);if(e==="relu6")return n.relu6(t);if(e==="prelu")return n.prelu(t,i);throw new Error("Activation "+e+" has not been implemented for the CPU backend.")}var R0=function(n){iz(t,n);function t(){var e=n.call(this)||this;return e.blockSize=48,e.firstUse=!0,e.data=new T.DataStorage(e,T.engine()),e}return t.prototype.write=function(e,i,r){this.firstUse&&(this.firstUse=!1,T.env().get("IS_NODE")&&T.backend_util.warn(`
+        `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c<t.length;++c){var h=c===0?0:r[c-1],d=[0,h,0],p=[1,t[c],a];u[c]=B.reshape(B.slice(n,d,p),e)}return n.dispose(),u}),o=new jo([],e,n.dtype,t.length),l=0;l<s.length;l++)o.setItem(l,s[l]);return o}var kU=void 0,FU=function(n,t,e){return Vn(kU,void 0,void 0,function(){var i,r,a,s,o,l,u,c,o,h,d,l,p,f,m,m,g,v,g,b,g,g,g,w,S,L,N,C,R,D,k,W,F,P,H,_,j,K,q,G,Z,X,ee,ne,ie,te,re,le,he,ye,Oe,Fe,_e,Me,dt,Lt,Ge,Ue,F,P,me,Ue,j,L,Ft,me,ne,ie,L,lt,me,L,pt,lt,me,q,G,L,pt,me,Ue,L,pt,lt,me,En,L,pt,me,re,me,he,L,Ue,P,me,Ue,L,Ft,me,Oe,L,Fe,me;return Tn(this,function(Et){switch(Et.label){case 0:i=n.op;switch(i){case"If":return[3,1];case"StatelessIf":return[3,1];case"While":return[3,3];case"StatelessWhile":return[3,3];case"LoopCond":return[3,9];case"Switch":return[3,10];case"Merge":return[3,12];case"Enter":return[3,13];case"Exit":return[3,14];case"NextIteration":return[3,15];case"TensorArrayV3":return[3,16];case"TensorArrayWriteV3":return[3,17];case"TensorArrayReadV3":return[3,18];case"TensorArrayGatherV3":return[3,19];case"TensorArrayScatterV3":return[3,20];case"TensorArrayConcatV3":return[3,21];case"TensorArraySplitV3":return[3,22];case"TensorArraySizeV3":return[3,23];case"TensorArrayCloseV3":return[3,24];case"TensorListSetItem":return[3,25];case"TensorListGetItem":return[3,26];case"TensorListScatterV2":return[3,27];case"TensorListScatter":return[3,27];case"TensorListReserve":return[3,28];case"TensorListGather":return[3,29];case"TensorListStack":return[3,30];case"TensorListFromTensor":return[3,31];case"TensorListConcat":return[3,32];case"TensorListPushBack":return[3,33];case"TensorListPopBack":return[3,34];case"TensorListSplit":return[3,35]}return[3,36];case 1:return r=A("thenBranch",n,t,e),a=A("elseBranch",n,t,e),s=A("cond",n,t,e),o=A("args",n,t,e),[4,s.data()];case 2:return l=Et.sent(),l[0]?[2,e.functionMap[r].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)]:[2,e.functionMap[a].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 3:return u=A("body",n,t,e),c=A("cond",n,t,e),o=A("args",n,t,e),[4,e.functionMap[c].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 4:return h=Et.sent(),d=o.map(function(It){return It.id}),[4,h[0].data()];case 5:l=Et.sent(),h.forEach(function(It){!It.kept&&d.indexOf(It.id)===-1&&It.dispose()}),p=o,f=function(){var It,Ce,Wt;return Tn(this,function(ft){switch(ft.label){case 0:return It=p,[4,e.functionMap[u].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 1:return p=ft.sent(),Ce=p.map(function(it){return it.id}),It.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[4,e.functionMap[c].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 2:return Wt=ft.sent(),[4,Wt[0].data()];case 3:return l=ft.sent(),Wt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[2]}})},Et.label=6;case 6:return l[0]?[5,f()]:[3,8];case 7:return Et.sent(),[3,6];case 8:return[2,p];case 9:return m=A("pred",n,t,e),[2,[oi(m)]];case 10:return m=A("pred",n,t,e),g=A("data",n,t,e),g.kept||(g=oi(g)),[4,m.data()];case 11:return[2,Et.sent()[0]?[void 0,g]:[g,void 0]];case 12:return v=n.inputNames.find(function(It){return Gt(It,t,e)!==void 0}),v?(g=Gt(v,t,e),[2,[oi(g)]]):[2,void 0];case 13:return b=A("frameName",n,t,e),g=A("tensor",n,t,e),e.enterFrame(b),[2,[oi(g)]];case 14:return g=A("tensor",n,t,e),e.exitFrame(),[2,[oi(g)]];case 15:return g=A("tensor",n,t,e),e.nextIteration(),[2,[oi(g)]];case 16:return w=A("size",n,t,e),S=A("dtype",n,t,e),L=A("elementShape",n,t,e),N=A("dynamicSize",n,t,e),C=A("clearAfterRead",n,t,e),R=A("identicalElementShapes",n,t,e),D=A("name",n,t,e),k=new CU(D,S,w,L,R,N,C),e.addTensorArray(k),[2,[k.idTensor,B.scalar(1)]];case 17:return W=A("tensorArrayId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),H=e.getTensorArray(W.id),H.write(F,P),[2,[H.idTensor]];case 18:return _=A("tensorArrayId",n,t,e),j=A("index",n,t,e),K=e.getTensorArray(_.id),[2,[K.read(j)]];case 19:return q=A("tensorArrayId",n,t,e),G=A("indices",n,t,e),Z=A("dtype",n,t,e),X=e.getTensorArray(q.id),[2,[X.gather(G,Z)]];case 20:return ee=A("tensorArrayId",n,t,e),ne=A("indices",n,t,e),ie=A("tensor",n,t,e),te=e.getTensorArray(ee.id),te.scatter(ne,ie),[2,[te.idTensor]];case 21:return re=A("tensorArrayId",n,t,e),le=e.getTensorArray(re.id),he=A("dtype",n,t,e),[2,[le.concat(he)]];case 22:return ye=A("tensorArrayId",n,t,e),Oe=A("tensor",n,t,e),Fe=A("lengths",n,t,e),_e=e.getTensorArray(ye.id),_e.split(Fe,Oe),[2,[_e.idTensor]];case 23:return Me=A("tensorArrayId",n,t,e),dt=e.getTensorArray(Me.id),[2,[B.scalar(dt.size(),"int32")]];case 24:return Lt=A("tensorArrayId",n,t,e),Ge=e.getTensorArray(Lt.id),Ge.clearAndClose(),[2,[Ge.idTensor]];case 25:return Ue=A("tensorListId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.setItem(F,P),[2,[me.idTensor]];case 26:return Ue=A("tensorListId",n,t,e),j=A("index",n,t,e),L=A("elementShape",n,t,e),Ft=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.getItem(j,L,Ft)]];case 27:return ne=A("indices",n,t,e),ie=A("tensor",n,t,e),L=A("elementShape",n,t,e),lt=A("numElements",n,t,e),me=EU(ie,ne,L,lt),e.addTensorList(me),[2,[me.idTensor]];case 28:return L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=OU(L,pt,lt),e.addTensorList(me),[2,[me.idTensor]];case 29:return q=A("tensorListId",n,t,e),G=A("indices",n,t,e),L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),me=e.getTensorList(q.id),[2,[me.gather(G,pt,L)]];case 30:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=e.getTensorList(Ue.id),[2,[me.stack(L,pt,lt)]];case 31:return En=A("tensor",n,t,e),L=A("elementShape",n,t,e),pt=A("elementDType",n,t,e),me=RU(En,L,pt),e.addTensorList(me),[2,[me.idTensor]];case 32:return re=A("tensorListId",n,t,e),me=e.getTensorList(re.id),he=A("dtype",n,t,e),L=A("elementShape",n,t,e),[2,[me.concat(he,L)]];case 33:return Ue=A("tensorListId",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.pushBack(P),[2,[me.idTensor]];case 34:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),Ft=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.popBack(L,Ft)]];case 35:return Oe=A("tensor",n,t,e),L=A("elementShape",n,t,e),Fe=A("lengths",n,t,e),me=DU(Oe,Fe,L),e.addTensorList(me),[2,[me.idTensor]];case 36:throw TypeError("Node type "+n.op+" is not implemented")}})})};function zw(n,t,e){var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=r==="fusedbatchnorm",u=A("numArgs",n,t,e);if(s){if(o&&u!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&u!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");var c=A("strides",n,t,e),h=Yo(n,t,e),d=A("dataFormat",n,t,e).toUpperCase(),p=A("dilations",n,t,e),f=A("args",n,t,e),m=f[0],g=f[1];return{stride:c,pad:h,dataFormat:d,dilations:p,biasArg:m,preluArg:g,activationFunc:a}}var WU=function(n,t,e){switch(n.op){case"Conv1D":{var i=A("stride",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),s=A("dilation",n,t,e);return[B.conv1d(A("x",n,t,e),A("filter",n,t,e),i,r,a,s)]}case"Conv2D":{var i=A("strides",n,t,e),r=Yo(n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv2d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{var l=zw(n,t,e),i=l.stride,r=l.pad,a=l.dataFormat,o=l.dilations,u=l.biasArg,c=l.preluArg,h=l.activationFunc;return[B.fused.conv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"FusedDepthwiseConv2dNative":{var d=zw(n,t,e),i=d.stride,r=d.pad,a=d.dataFormat,o=d.dilations,u=d.biasArg,c=d.preluArg,h=d.activationFunc;return[B.fused.depthwiseConv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{var p=A("outputShape",n,t,e),i=A("strides",n,t,e),r=Yo(n,t,e);return[B.conv2dTranspose(A("x",n,t,e),A("filter",n,t,e),p,[i[1],i[2]],r)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{var i=A("strides",n,t,e),r=Yo(n,t,e),o=A("dilations",n,t,e),a=A("dataFormat",n,t,e).toUpperCase();return[B.depthwiseConv2d(A("input",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"Conv3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv3d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2],i[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPoolWithArgmax":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e),m=A("includeBatchInIndex",n,t,e),g=B.maxPoolWithArgmax(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r,m),v=g.result,b=g.indexes;return[v,b]}case"AvgPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"MaxPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"Dilation2D":{var w=A("strides",n,t,e),r=A("pad",n,t,e),o=A("dilations",n,t,e),S=w[1],L=w[2],N=o[1],C=o[2];return[B.dilation2d(A("x",n,t,e),A("filter",n,t,e),[S,L],r,[N,C],"NHWC")]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var UU=function(n,t,e){switch(n.op){case"Fill":{var i=A("shape",n,t,e),r=A("dtype",n,t,e),a=A("value",n,t,e);return[B.fill(i,a,r)]}case"LinSpace":{var s=A("start",n,t,e),o=A("stop",n,t,e),l=A("num",n,t,e);return[B.linspace(s,o,l)]}case"Multinomial":{var u=A("logits",n,t,e),c=A("numSamples",n,t,e),h=A("seed",n,t,e);return[B.multinomial(u,c,h)]}case"OneHot":{var d=A("indices",n,t,e),p=A("depth",n,t,e),f=A("onValue",n,t,e),m=A("offValue",n,t,e);return[B.oneHot(d,p,f,m)]}case"Ones":return[B.ones(A("shape",n,t,e),A("dtype",n,t,e))];case"OnesLike":return[B.onesLike(A("x",n,t,e))];case"RandomUniform":return[B.randomUniform(A("shape",n,t,e),A("minval",n,t,e),A("maxval",n,t,e),A("dtype",n,t,e))];case"Range":{var s=A("start",n,t,e),g=A("stop",n,t,e),v=A("step",n,t,e);return[B.range(s,g,v,A("dtype",n,t,e))]}case"TruncatedNormal":{var i=A("shape",n,t,e),b=A("mean",n,t,e),w=A("stdDev",n,t,e),h=A("seed",n,t,e);return[B.truncatedNormal(i,b,w,A("dtype",n,t,e),h)]}case"Zeros":return[B.zeros(A("shape",n,t,e),A("dtype",n,t,e))];case"ZerosLike":return[B.zerosLike(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var BU=void 0;function _d(n,t,e){var i=A("boxes",n,t,e),r=A("scores",n,t,e),a=A("maxOutputSize",n,t,e),s=A("iouThreshold",n,t,e),o=A("scoreThreshold",n,t,e),l=A("softNmsSigma",n,t,e);return{boxes:i,scores:r,maxOutputSize:a,iouThreshold:s,scoreThreshold:o,softNmsSigma:l}}var zU=function(n,t,e){return Vn(BU,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,a,s,o,l,u,p,h,f,a,s,o,l,u,m,h;return Tn(this,function(g){switch(g.label){case 0:i=n.op;switch(i){case"NonMaxSuppressionV5":return[3,1];case"NonMaxSuppressionV4":return[3,3];case"NonMaxSuppressionV3":return[3,5];case"NonMaxSuppressionV2":return[3,5];case"Where":return[3,7];case"ListDiff":return[3,9]}return[3,10];case 1:return r=_d(n,t,e),a=r.boxes,s=r.scores,o=r.maxOutputSize,l=r.iouThreshold,u=r.scoreThreshold,c=r.softNmsSigma,[4,B.image.nonMaxSuppressionWithScoreAsync(a,s,o,l,u,c)];case 2:return h=g.sent(),[2,[h.selectedIndices,h.selectedScores]];case 3:return d=_d(n,t,e),a=d.boxes,s=d.scores,o=d.maxOutputSize,l=d.iouThreshold,u=d.scoreThreshold,p=A("padToMaxOutputSize",n,t,e),[4,B.image.nonMaxSuppressionPaddedAsync(a,s,o,l,u,p)];case 4:return h=g.sent(),[2,[h.selectedIndices,h.validOutputs]];case 5:return f=_d(n,t,e),a=f.boxes,s=f.scores,o=f.maxOutputSize,l=f.iouThreshold,u=f.scoreThreshold,[4,B.image.nonMaxSuppressionAsync(a,s,o,l,u)];case 6:return[2,[g.sent()]];case 7:return m=B.cast(A("condition",n,t,e),"bool"),[4,B.whereAsync(m)];case 8:return h=[g.sent()],m.dispose(),[2,h];case 9:return[2,B.setdiff1dAsync(A("x",n,t,e),A("y",n,t,e))];case 10:throw TypeError("Node type "+n.op+" is not implemented")}})})};var PU=function(n,t,e){switch(n.op){case"TopKV2":{var i=A("x",n,t,e),r=A("k",n,t,e),a=A("sorted",n,t,e),s=B.topk(i,r,a);return[s.values,s.indices]}case"Unique":{var i=A("x",n,t,e),s=B.unique(i);return[s.values,s.indices]}case"UniqueV2":{var i=A("x",n,t,e),o=A("axis",n,t,e),s=B.unique(i,o);return[s.values,s.indices]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var _U=function(n,t,e){switch(n.op){case"Const":return t[n.name];case"PlaceholderWithDefault":var i=A("default",n,t,e);return[Gt(n.name,t,e)||i];case"Placeholder":return[Gt(n.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{var r=A("x",n,t,e);return[oi(r)]}case"IdentityN":return A("x",n,t,e).map(function(h){return oi(h)});case"Snapshot":var a=A("x",n,t,e);return[oi(a)];case"Shape":return[B.tensor1d(A("x",n,t,e).shape,"int32")];case"ShapeN":return A("x",n,t,e).map(function(h){return B.tensor1d(h.shape)});case"Size":return[B.scalar(A("x",n,t,e).size,"int32")];case"Rank":return[B.scalar(A("x",n,t,e).rank,"int32")];case"NoOp":return[B.scalar(1)];case"Print":var s=A("x",n,t,e),o=A("data",n,t,e),l=A("message",n,t,e),u=A("summarize",n,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(l);for(var c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,u));return[s];default:throw TypeError("Node type "+n.op+" is not implemented")}};var MU=function(n,t,e){switch(n.op){case"ResizeBilinear":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeBilinear(i,[r[0],r[1]],a)]}case"ResizeNearestNeighbor":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeNearestNeighbor(i,[r[0],r[1]],a)]}case"CropAndResize":{var s=A("image",n,t,e),o=A("boxes",n,t,e),l=A("boxInd",n,t,e),u=A("cropSize",n,t,e),c=A("method",n,t,e),h=A("extrapolationValue",n,t,e);return[B.image.cropAndResize(s,o,l,u,c,h)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var HU=function(n,t,e){switch(n.op){case"Equal":return[B.equal(A("a",n,t,e),A("b",n,t,e))];case"NotEqual":return[B.notEqual(A("a",n,t,e),A("b",n,t,e))];case"Greater":return[B.greater(A("a",n,t,e),A("b",n,t,e))];case"GreaterEqual":return[B.greaterEqual(A("a",n,t,e),A("b",n,t,e))];case"Less":return[B.less(A("a",n,t,e),A("b",n,t,e))];case"LessEqual":return[B.lessEqual(A("a",n,t,e),A("b",n,t,e))];case"LogicalAnd":return[B.logicalAnd(A("a",n,t,e),A("b",n,t,e))];case"LogicalNot":return[B.logicalNot(A("a",n,t,e))];case"LogicalOr":return[B.logicalOr(A("a",n,t,e),A("b",n,t,e))];case"Select":case"SelectV2":return[B.where(A("condition",n,t,e),A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var VU=function(n,t,e){switch(n.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[B.matMul(A("a",n,t,e),A("b",n,t,e),A("transposeA",n,t,e),A("transposeB",n,t,e))];case"Transpose":return[B.transpose(A("x",n,t,e),A("perm",n,t,e))];case"_FusedMatMul":var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=A("numArgs",n,t,e);if(s){if(o&&l!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&l!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}var u=A("args",n,t,e),c=u[0],h=u[1];return[B.fused.matMul({a:A("a",n,t,e),b:A("b",n,t,e),transposeA:A("transposeA",n,t,e),transposeB:A("transposeB",n,t,e),bias:c,activation:a,preluActivationWeights:h})];default:throw TypeError("Node type "+n.op+" is not implemented")}};var qU=function(n,t,e){switch(n.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"FusedBatchNormV3":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"LRN":return[B.localResponseNormalization(A("x",n,t,e),A("radius",n,t,e),A("bias",n,t,e),A("alpha",n,t,e),A("beta",n,t,e))];case"Softmax":return[B.softmax(A("x",n,t,e))];case"LogSoftmax":return[B.logSoftmax(A("x",n,t,e))];case"SparseToDense":return[B.sparseToDense(A("sparseIndices",n,t,e),A("outputShape",n,t,e),A("sparseValues",n,t,e),A("defaultValue",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var GU=function(n,t,e){switch(n.op){case"Max":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.max(A("x",n,t,e),i,r)]}case"Mean":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.mean(A("x",n,t,e),i,r)]}case"Min":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.min(A("x",n,t,e),i,r)]}case"Sum":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.sum(A("x",n,t,e),i,r)]}case"All":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.all(A("x",n,t,e),i,r)]}case"Any":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.any(A("x",n,t,e),i,r)]}case"ArgMax":{var i=A("axis",n,t,e);return[B.argMax(A("x",n,t,e),i)]}case"ArgMin":{var i=A("axis",n,t,e);return[B.argMin(A("x",n,t,e),i)]}case"Prod":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.prod(A("x",n,t,e),i,r)]}case"Cumsum":{var i=A("axis",n,t,e),a=A("exclusive",n,t,e),s=A("reverse",n,t,e);return[B.cumsum(A("x",n,t,e),i,a,s)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var YU=function(n,t,e){switch(n.op){case"ConcatV2":case"Concat":{var i=A("n",n,t,e),r=A("axis",n,t,e),a=A("tensors",n,t,e);return a=a.slice(0,i),[B.concat(a,r)]}case"GatherV2":case"Gather":{var r=A("axis",n,t,e),s=A("x",n,t,e),o=A("indices",n,t,e);return[B.gather(s,B.cast(o,"int32"),r)]}case"ReverseV2":case"Reverse":{var r=A("axis",n,t,e),s=A("x",n,t,e);return[B.reverse(s,r)]}case"Slice":{var l=A("begin",n,t,e),u=A("size",n,t,e);return[B.slice(A("x",n,t,e),l,u)]}case"StridedSlice":{var l=A("begin",n,t,e),c=A("end",n,t,e),h=A("strides",n,t,e),d=A("beginMask",n,t,e),p=A("endMask",n,t,e),f=A("ellipsisMask",n,t,e),m=A("newAxisMask",n,t,e),g=A("shrinkAxisMask",n,t,e),v=A("x",n,t,e);return[B.stridedSlice(v,l,c,h,d,p,f,m,g)]}case"Pack":return B.tidy(function(){var D=A("axis",n,t,e),k=A("tensors",n,t,e),W=k[0].shape,F=B.squeeze(k[0]).shape,P=k.map(function(H){var _=B.util.arraysEqual(H.shape,W);if(!_&&!B.util.arraysEqual(B.squeeze(H).shape,F))throw new Error("the input tensors shape does not match");return _?H:B.reshape(H,W)});return[B.stack(P,D)]});case"Unpack":{var r=A("axis",n,t,e),v=A("tensor",n,t,e);return B.unstack(v,r)}case"Tile":{var b=A("reps",n,t,e);return[B.tile(A("x",n,t,e),b)]}case"Split":case"SplitV":{var r=A("axis",n,t,e),w=A("numOrSizeSplits",n,t,e),v=A("x",n,t,e);return B.split(v,w,r)}case"ScatterNd":{var o=A("indices",n,t,e),S=A("values",n,t,e),L=A("shape",n,t,e);return[B.scatterND(o,S,L)]}case"GatherNd":{var N=A("x",n,t,e),o=A("indices",n,t,e);return[B.gatherND(N,o)]}case"SparseToDense":{var o=A("sparseIndices",n,t,e),L=A("outputShape",n,t,e),C=A("sparseValues",n,t,e),R=A("defaultValue",n,t,e);return[B.sparseToDense(o,C,L,C.dtype===R.dtype?R:B.cast(R,C.dtype))]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var jU=function(n,t,e){switch(n.op){case"FFT":return[B.fft(A("x",n,t,e))];case"IFFT":return[B.ifft(A("x",n,t,e))];case"RFFT":return[B.rfft(A("x",n,t,e))];case"IRFFT":return[B.irfft(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var KU=function(n,t,e){switch(n.op){case"Cast":return[B.cast(A("x",n,t,e),A("dtype",n,t,e))];case"ExpandDims":{var i=A("axis",n,t,e);return[B.expandDims(A("x",n,t,e),i)]}case"Squeeze":{var i=A("axis",n,t,e);return[B.squeeze(A("x",n,t,e),i)]}case"Reshape":return[B.reshape(A("x",n,t,e),A("shape",n,t,e))];case"PadV2":case"Pad":return[B.pad(A("x",n,t,e),A("padding",n,t,e),A("constantValue",n,t,e))];case"SpaceToBatchND":{var r=A("blockShape",n,t,e),a=A("paddings",n,t,e);return[B.spaceToBatchND(A("x",n,t,e),r,a)]}case"BatchToSpaceND":{var r=A("blockShape",n,t,e),s=A("crops",n,t,e);return[B.batchToSpaceND(A("x",n,t,e),r,s)]}case"DepthToSpace":{var o=A("blockSize",n,t,e),l=A("dataFormat",n,t,e).toUpperCase();return[B.depthToSpace(A("x",n,t,e),o,l)]}case"BroadcastTo":return[B.broadcastTo(A("x",n,t,e),A("shape",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function Pw(n,t,e){var i=function(r,a,s){switch(r.category){case"arithmetic":return B.tidy(function(){return TU(r,a,s)});case"basic_math":return B.tidy(function(){return NU(r,a,s)});case"control":return FU(r,a,s);case"convolution":return B.tidy(function(){return WU(r,a,s)});case"creation":return B.tidy(function(){return UU(r,a,s)});case"dynamic":return zU(r,a,s);case"evaluation":return B.tidy(function(){return PU(r,a,s)});case"image":return B.tidy(function(){return MU(r,a,s)});case"graph":return B.tidy(function(){return _U(r,a,s)});case"logical":return B.tidy(function(){return HU(r,a,s)});case"matrices":return B.tidy(function(){return VU(r,a,s)});case"normalization":return B.tidy(function(){return qU(r,a,s)});case"reduction":return B.tidy(function(){return GU(r,a,s)});case"slice_join":return B.tidy(function(){return YU(r,a,s)});case"spectral":return B.tidy(function(){return jU(r,a,s)});case"transformation":return B.tidy(function(){return KU(r,a,s)});case"custom":var o=kw(r.op);if(o&&o.customExecutor)return o.customExecutor(new AU(r,a,s));throw TypeError("Custom op "+r.op+" is not registered.");default:throw TypeError("Unknown op '"+r.op+"'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()")}}(n,t,e);return i instanceof Promise?i.then(function(r){return[].concat(r)}):[].concat(i)}var _w=function(){function n(t,e,i,r){t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i={}),r===void 0&&(r={}),this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=i,this.functionMap=r,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}return n.prototype.newFrame=function(t,e){return{id:t,frameName:e,iterationId:0}},Object.defineProperty(n.prototype,"currentContext",{get:function(){return this.contexts},set:function(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextId",{get:function(){return this._currentContextIds[0]},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextIds",{get:function(){return this._currentContextIds},enumerable:!0,configurable:!0}),n.prototype.generateCurrentContextIds=function(){for(var t=[],e=0;e<this.contexts.length-1;e++){var i=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(i))}t.push(""),this._currentContextIds=t},n.prototype.contextIdforContexts=function(t){return t?t.map(function(e){return e.id===0&&e.iterationId===0?"":e.frameName+"-"+e.iterationId}).join("/"):""},n.prototype.enterFrame=function(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))},n.prototype.exitFrame=function(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function Hw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return Jt(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return Jt(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((Mw(d)||$U(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function XU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return Jt(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var JU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],ZU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function Mw(n){return JU.indexOf(n.op)>=0}function $U(n){return ZU.indexOf(n.op)>=0}var Vw=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=Hw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return XU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[Jt(d)[0]]}),s=e.map(function(d){return Jt(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new _w(i.weightMap,c,h,i.functionExecutorMap),p=Ew({},i.weightMap);Object.keys(t).forEach(function(w){var S=Jt(w),L=S[0],N=S[1],C=[];C[N]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g<u.length;g++){var v=u[g];if(!p[v.name]){var b=Pw(v,p,d);if(b instanceof Promise)throw new Error("The execution of the op '"+v.op+"' returned a promise. Please use model.executeAsync() instead.");p[v.name]=b,i.checkTensorForDisposal(v.name,v,p,d,f,s,m)}}return i.parent==null&&d.dispose(f),e.map(function(w){return Gt(w,p,d)})})},n.prototype.getFrozenTensorIds=function(t){var e=[].concat.apply([],Object.keys(t).map(function(i){return t[i]}).map(function(i){return i.map(function(r){return r.id})}));return new Set(e)},n.prototype.checkTensorForDisposal=function(t,e,i,r,a,s,o){if(e.category==="control"||s.indexOf(t)!==-1)return;i[t].forEach(function(l){l!=null&&(o[l.id]=(o[l.id]||0)+e.children.length)}),e.inputs.forEach(function(l){if(l.category!=="control"){var u=qW(l.name,i,r);u!=null&&u.forEach(function(c){if(c&&!a.has(c.id)){var h=o[c.id];h===1?(c.dispose(),delete o[c.id]):h!=null&&o[c.id]--}})}})},n.prototype.executeAsync=function(t,e){return Vn(this,void 0,void 0,function(){return Tn(this,function(i){return[2,this._executeAsync(t,e)]})})},n.prototype._executeAsync=function(t,e,i,r,a){return i===void 0&&(i=!1),r===void 0&&(r={}),a===void 0&&(a={}),Vn(this,void 0,void 0,function(){var s,o,l,u,c,h;return Tn(this,function(d){switch(d.label){case 0:return i||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e)),s=new _w(this.weightMap,r,a,this.functionExecutorMap),[4,this.executeWithControlFlow(t,s,e,i)];case 1:return o=d.sent(),l=e.map(function(p){return Gt(p,o,s)}),u=l.map(function(p){return p.id}),c=Object.keys(t).map(function(p){return t[p].id}),h=new Set(u.concat(c,this.weightIds)),Object.keys(o).forEach(function(p){var f=o[p];f.forEach(function(m){m&&!m.isDisposed&&!h.has(m.id)&&m.dispose()})}),this.parent==null&&s.dispose(h),[2,l]}})})},n.prototype.executeFunctionAsync=function(t,e,i){return Vn(this,void 0,void 0,function(){var r,a=this;return Tn(this,function(s){return r=t.reduce(function(o,l,u){return o[a.inputs[u].name]=l,o},{}),[2,this._executeAsync(r,this.outputNodes,!0,e,i)]})})},n.prototype.executeWithControlFlow=function(t,e,i,r){return Vn(this,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N=this;return Tn(this,function(C){switch(C.label){case 0:a=Object.keys(t),s=a.map(function(R){return N.graph.nodes[Jt(R)[0]]}),o=i.map(function(R){return Jt(R)[0]}),l=o.map(function(R){return N.graph.nodes[R]}),u=Hw(t,l,this.weightMap),c=u.usedNodes,h=u.missingInputs,d=u.dynamicNode,p=u.syncInputs,f=s.concat(this.graph.weights).map(function(R){return{node:R,contexts:e.currentContext}}),m=Ew({},this.weightMap),Object.keys(t).forEach(function(R){var D=Jt(R),k=D[0],W=D[1],F=[];F[W]=t[R],m[k]=F}),g={},v=this.getFrozenTensorIds(m),b={},C.label=1;case 1:return f.length>0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!Mw(R)&&!Gt(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=si(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=Pw(f.node,r,i);m||(m=si(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=si(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Gt(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Gt(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=Jt(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=Jt(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=Jt(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var QU="?tfjs-format=file",eB="model.json",qw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return Vn(this,void 0,void 0,function(){var t;return Tn(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new Vw(Ww.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Ww.Instance.transformGraph(t.modelInitializer);this.initializer=new Vw(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return Vn(this,void 0,void 0,function(){var i;return Tn(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return Vn(this,void 0,void 0,function(){var i;return Tn(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function tB(n,t){return t===void 0&&(t={}),Vn(this,void 0,void 0,function(){var e;return Tn(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+eB+QU)),e=new qw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var nB="2.6.0";gr.GraphModel=qw;gr.deregisterOp=VW;gr.loadGraphModel=tB;gr.registerOp=HW;gr.version_converter=nB});var h0=Le(Yt=>{"use strict";Object.defineProperty(Yt,"__esModule",{value:!0});var Ne=lr();var Md=function(n,t){return Md=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Md(n,t)};function Ve(n,t){Md(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var la=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function vr(n,t){return t={exports:{}},n(t,t.exports),t.exports}var iB=vr(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(la,n,!1)}),rB=vr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(la,n,!1)}),aB=vr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(la,n,!1)}),sB=vr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(la,n,!1)}),oB=vr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(la,n,!1)}),lB=vr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(la,n,!1)}),yr=vr(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,N){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||N||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,N=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(N||(S=[N++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%N]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,j=C.S;W--;)F=j[H=h&H+1],P=P*r+j[h&(j[H]=j[_=h&_+F])+(j[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var N=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{N.push(g(S[R],L-1))}catch(D){}return N.length?N:C=="string"?S:S+"\0"}function v(S,L){for(var N=S+"",C,R=0;R<N.length;)L[h&R]=h&(C^=L[h&R]*19)+N.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,N=L&&L.plugins;return[+new Date,i,N,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=require("crypto")}catch(S){}}})([],Math)});yr.alea=iB;yr.xor128=rB;yr.xorwow=aB;yr.xorshift7=sB;yr.xor4096=oB;yr.tychei=lB;var uB=yr,Yw=uB.alea;function cB(n,t){return Ko(n,t)}function Ko(n,t,e,i){if(e===void 0&&(e=new Map),i===void 0&&(i=new Set),n==null)return null;if(i.has(n))throw new Error("Circular references are not supported.");if(e.has(n))return e.get(n);var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if(ua(n)){var a=Array.isArray(n)?[]:{};i.add(n);for(var s in n){var o=n[s],l=Ko(o,t,e,i);a[s]=l}return i.delete(n),a}else throw new Error("Can't recurse into non-iterable type: "+n);else return e.set(n,r.value),r.value}function hB(n,t){return t===void 0&&(t=Kw),jw(n,t)}function jw(n,t,e){e===void 0&&(e=new Set);var i=n[0];if(e.has(i))throw new Error("Circular references are not supported.");var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if(ua(i)){var a=Array.isArray(i)?[]:{};e.add(i);var s=function(l){var u=n.map(function(h){return h[l]}),c=jw(u,t,e);a[l]=c};for(var o in i)s(o);return e.delete(i),a}else throw new Error("Can't recurse into non-iterable type: "+i);else return r.value}function Kw(n){return n===null?null:ua(n[0])?{value:null,recurse:!0}:{value:n,recurse:!1}}function $w(n,t){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:e=new Map,Ko(n,t,e),i=0,r=Array.from(e.keys()),u.label=1;case 1:return i<r.length?(a=r[i],s=e.get(a),s instanceof Promise?[4,s]:[3,3]):[3,4];case 2:o=u.sent(),e.set(a,o),u.label=3;case 3:return i++,[3,1];case 4:return l=Ko(n,t,e),[2,l]}})})}function ua(n){return n!=null&&!ArrayBuffer.isView(n)&&(Array.isArray(n)||typeof n=="object"&&!(n instanceof Ne.Tensor))}function pB(n){return n==null||dB(n)||Array.isArray(n)||typeof n=="object"&&n instanceof Ne.Tensor||Ne.util.isTypedArray(n)}function dB(n){return n===null||typeof n!="object"&&typeof n!="function"}function mB(n){return cB(n,fB)}function fB(n){return n instanceof Ne.Tensor?{value:n.clone(),recurse:!1}:ua(n)?{value:null,recurse:!0}:{value:n,recurse:!1}}var Xw=function(){function n(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}return n.prototype.wrap=function(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity},n.prototype.get=function(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]},n.prototype.set=function(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e},n.prototype.length=function(){var t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t},n.prototype.isFull=function(){return this.length()===this.capacity},n.prototype.isEmpty=function(){return this.length()===0},n.prototype.push=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)},n.prototype.pushAll=function(t){for(var e=0,i=t;e<i.length;e++){var r=i[e];this.push(r)}},n.prototype.pop=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);var t=this.get(this.end);return this.set(this.end,void 0),t},n.prototype.unshift=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)},n.prototype.shift=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t},n.prototype.shuffleExcise=function(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var e=this.wrap(this.begin+t),i=this.get(e);return this.set(e,this.pop()),i},n}();var gB=function(n){Ve(t,n);function t(){return n.call(this,t.INITIAL_CAPACITY)||this}return t.prototype.isFull=function(){return!1},t.prototype.push=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.push.call(this,e)},t.prototype.unshift=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.unshift.call(this,e)},t.prototype.expand=function(){for(var e=this.capacity*2,i=new Array(e),r=this.length(),a=0;a<r;a++)i[a]=this.get(this.wrap(this.begin+a));this.data=i,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=r},t.INITIAL_CAPACITY=32,t}(Xw);function Jw(n){return new vB(n)}function Hd(n){return new yB(n)}function bB(n,t){return new Zw(n,t)}function SB(n,t){return t===void 0&&(t=Fi.FAIL),new wB(n,t)}var xt=function(){function n(){}return n.prototype.toArray=function(){return se(this,void 0,void 0,function(){var t,e;return oe(this,function(i){switch(i.label){case 0:return t=[],[4,this.next()];case 1:e=i.sent(),i.label=2;case 2:return e.done?[3,4]:(t.push(e.value),[4,this.next()]);case 3:return e=i.sent(),[3,2];case 4:return[2,t]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){var t,e,i;return oe(this,function(r){switch(r.label){case 0:return t=this.prefetch(100),e=[],[4,t.next()];case 1:i=r.sent(),r.label=2;case 2:return i.done?[3,4]:(e.push(i.value),[4,t.next()]);case 3:return i=r.sent(),[3,2];case 4:return[2,e]}})})},n.prototype.resolveFully=function(){return se(this,void 0,void 0,function(){var t;return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:t=e.sent(),e.label=2;case 2:return t.done?[3,4]:[4,this.next()];case 3:return t=e.sent(),[3,2];case 4:return[2]}})})},n.prototype.resolveWhile=function(t){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:return[4,this.next()];case 1:e=r.sent(),i=t(e.value),r.label=2;case 2:return!e.done&&i?[4,this.next()]:[3,4];case 3:return e=r.sent(),i=t(e.value),[3,2];case 4:return[2]}})})},n.prototype.handleErrors=function(t){return new CB(this,t)},n.prototype.filter=function(t){return new NB(this,t)},n.prototype.map=function(t){return new xB(this,t)},n.prototype.mapAsync=function(t){return new Qw(this,t)},n.prototype.serialMapAsync=function(t){return new Qw(this,t).serial()},n.prototype.flatmap=function(t){return new RB(this,t)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.map(t).resolveFully()]})})},n.prototype.serialForEach=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.serialMapAsync(t).resolveWhile(function(i){return i===!0})]})})},n.prototype.rowMajorBatch=function(t,e){return e===void 0&&(e=!0),new TB(this,t,e)},n.prototype.columnMajorBatch=function(t,e,i){e===void 0&&(e=!0),i===void 0&&(i=Kw);var r=this.rowMajorBatch(t,e);return r.map(function(a){return hB(a,i)})},n.prototype.concatenate=function(t,e){return new Zw(Jw([this,t]),e)},n.prototype.take=function(t){return t<0||t==null?this:new AB(this,t)},n.prototype.skip=function(t){return t<0||t==null?this:new IB(this,t)},n.prototype.prefetch=function(t){return new e0(this,t)},n.prototype.shuffle=function(t,e){return new OB(this,t,e)},n.prototype.serial=function(){return new LB(this)},n}(),vB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.items=e,i.trav=0,i}return t.prototype.summary=function(){return"Array of "+this.items.length+" items"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return this.trav>=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:mB(e),done:!1}])})})},t}(xt),yB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(xt),LB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(xt),IB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++<this.maxCount?[4,this.upstream.next()]:[3,2];case 1:return e=i.sent(),e.done?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2,this.upstream.next()]}})})},t}(xt),AB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(xt),TB=function(n){Ve(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length<this.batchSize?[4,this.upstream.next()]:[3,3];case 2:return i=r.sent(),i.done?this.enableSmallLastBatch&&e.length>0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(xt),NB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(xt),xB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(xt),CB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.handler=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(xt),Qw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(xt),Vd=function(n){Ve(t,n);function t(){var e=n.call(this)||this;return e.outputQueue=new gB,e.lastRead=Promise.resolve({value:null,done:!1}),e}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.outputQueue.length()===0?[4,this.pump()]:[3,2];case 1:return e.sent()?[3,0]:[2,{value:null,done:!0}];case 2:return[2,{value:this.outputQueue.shift(),done:!1}]}})})},t}(xt),RB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,!0]}})})},t}(Vd),Zw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.baseErrorHandler=i,r.lastRead=null,r.iterator=null,r.moreIterators=e,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of chained summaries";return e+" -> Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(xt),Fi;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Fi||(Fi={}));var wB=function(n){Ve(t,n);function t(e,i){i===void 0&&(i=Fi.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof xt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,$w(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Fi.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Fi.SHORTEST:return[2,{value:null,done:!0}];case Fi.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(xt),e0=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new Xw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(xt),OB=function(n){Ve(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=Yw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(e0);var hs=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is
+      `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),Zt(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,EB)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,Zt(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,Zt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return Zt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return Zt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return Zt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,Zt(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=Hd(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,bB(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?r=0:r=null,Zt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().skip(t)]}})})},r)},n.prototype.shuffle=function(t,e,i){var r=this;if(i===void 0&&(i=!0),t==null||t<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError("`Dataset.shuffle()` requires bufferSize to be specified.  If your data fits in main memory (for regular JS objects), and/or GPU memory (for `tf.Tensor`s), consider setting "+("bufferSize to the dataset size ("+this.size+" elements)"));var a=this,s=Yw(e||Ne.util.now().toString());return Zt(function(){return se(r,void 0,void 0,function(){var o;return oe(this,function(l){switch(l.label){case 0:return o=s.int32(),i&&(o+=s.int32()),[4,a.iterator()];case 1:return[2,l.sent().shuffle(t,o.toString())]}})})},this.size)},n.prototype.take=function(t){var e=this,i=this,r;return this.size!=null&&this.size>t?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,Zt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function Zt(n,t){return t===void 0&&(t=null),new(function(e){Ve(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(hs))}function DB(n){var t=this;return Zt(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,Jw(n)]})})},n.length)}function kB(n){var t=this;if(!ua(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i<n.length;i++)e=e==null?n[i].size:Math.min(e,n[i].size);else if(n instanceof Object)for(var r in n)e=e==null?n[r].size:Math.min(e,n[r].size);return Zt(function(){return se(t,void 0,void 0,function(){var a;return oe(this,function(s){switch(s.label){case 0:return[4,$w(n,function(o){if(o instanceof hs)return{value:o.iterator(),recurse:!1};if(ua(o))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")})];case 1:return a=s.sent(),[2,SB(a,Fi.SHORTEST)]}})})},e)}function EB(n){if(n===null)return null;var t=n[0];if(pB(t)){var e=FB(n);return{value:e,recurse:!1}}return{value:null,recurse:!0}}function FB(n){if(n.length===0)throw new Error("Can't make a batch of zero elements.");return n[0]instanceof Ne.Tensor?Ne.stack(n):Ne.tensor(n)}var t0=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.input=e,i}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.input.iterator()];case 1:return e=a.sent(),i=e.decodeUTF8(),r=i.split(`
+`).map(function(s){return s.endsWith("\r")&&(s=s.slice(0,-1)),s}),[2,r]}})})},t}(hs);var $o='"',ds=Symbol("out"),n0=Symbol("field"),Xo=Symbol("quote"),qd=Symbol("quoteafterquote"),i0=Symbol("quoteinquote"),r0=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.input=e,r.hasHeader=!0,r.fullColumnNames=null,r.columnNamesValidated=!1,r.columnConfigs=null,r.configuredColumnsOnly=!1,r.delimiter=",",r.delimWhitespace=!1,r.base=new t0(e),i||(i={}),r.hasHeader=!(i.hasHeader===!1),r.fullColumnNames=i.columnNames,r.columnConfigs=i.columnConfigs,r.configuredColumnsOnly=i.configuredColumnsOnly,i.delimWhitespace?(Ne.util.assert(i.delimiter==null,function(){return"Delimiter should not be provided when delimWhitespace is true."}),r.delimWhitespace=!0,r.delimiter=" "):r.delimiter=i.delimiter?i.delimiter:",",r}return t.prototype.columnNames=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:e.sent(),e.label=2;case 2:return[2,this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames]}})})},t.prototype.setColumnNames=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u=this;return oe(this,function(c){switch(c.label){case 0:return[4,this.maybeReadHeaderLine()];case 1:if(e=c.sent(),!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");if(this.fullColumnNames&&e&&Ne.util.assert(e.length===this.fullColumnNames.length,function(){return"The length of provided columnNames ("+u.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."}),this.fullColumnNames||(this.fullColumnNames=e),i=this.fullColumnNames.reduce(function(h,d){return h[d]=h[d]+1||1,h},{}),r=Object.keys(i).filter(function(h){return i[h]>1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a<s.length;a++)if(o=s[a],l=this.fullColumnNames.indexOf(o),l===-1)throw new Error('The key "'+o+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}return this.columnNamesValidated=!0,[2]}})})},t.prototype.maybeReadHeaderLine=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return this.hasHeader?[4,this.base.iterator()]:[3,3];case 1:return e=s.sent(),[4,e.next()];case 2:if(i=s.sent(),i.done)throw new Error("No data was found for CSV parsing.");return r=i.value,a=this.parseRow(r,!1),[2,a];case 3:return[2,null]}})})},t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i=this;return oe(this,function(r){switch(r.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:r.sent(),r.label=2;case 2:return[4,this.base.iterator()];case 3:return e=r.sent(),this.hasHeader&&(e=e.skip(1)),[2,e.map(function(a){return i.makeDataElement(a)})]}})})},t.prototype.makeDataElement=function(e){for(var i=this.parseRow(e),r={},a={},s=0;s<this.fullColumnNames.length;s++){var o=this.fullColumnNames[s],l=this.columnConfigs?this.columnConfigs[o]:null;if(this.configuredColumnsOnly&&!l)continue;var u=i[s],c=null;if(u==="")if(l&&l.default!==void 0)c=l.default;else{if(l&&(l.required||l.isLabel))throw new Error("Required column "+o+" is empty in this line: "+e);c=void 0}else{var h=Number(u);if(isNaN(h))l&&l.dtype==="bool"?c=this.getBoolean(u):c=u;else if(!l||!l.dtype)c=h;else switch(l.dtype){case"float32":c=h;break;case"int32":c=Math.floor(h);break;case"bool":c=this.getBoolean(u);break;default:c=h}}l&&l.isLabel?a[o]=c:r[o]=c}return Object.keys(a).length===0?r:{xs:r,ys:a}},t.prototype.getBoolean=function(e){return e==="1"||e.toLowerCase()==="true"?1:0},t.prototype.parseRow=function(e,i){i===void 0&&(i=!0);for(var r=[],a=0,s=e.length,o=ds,l=0;l<s;l++)switch(o){case ds:switch(e.charAt(l)){case $o:a=l+1,o=Xo;break;case this.delimiter:if(a=l+1,this.delimiter===" "&&this.delimWhitespace)break;r.push(""),o=ds;break;default:o=n0,a=l;break}break;case n0:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l)),o=ds,a=l+1;break}break;case Xo:switch(e.charAt(l)){case $o:o=qd;break}break;case qd:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l-1)),o=ds,a=l+1;break;case $o:o=Xo;break;default:o=i0;break}break;case i0:switch(e.charAt(l)){case $o:o=Xo;break}break}if(o===qd?r.push(e.substring(a,s-1)):r.push(e.substring(a)),i&&r.length!==this.fullColumnNames.length)throw new Error("Invalid row in csv file. Should have "+this.fullColumnNames.length+" elements in a row, but got "+r);return r},t}(hs);var WB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;i.microphoneConfig=e,i.isClosed=!1,i.fftSize=e.fftSize||1024;var r=Math.log2(i.fftSize);if(i.fftSize<0||r<4||r>14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(xt);var UB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(xt);var a0=function(){function n(){}return n}();var s0=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new BB(this,e)},t}(xt),BB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new zB(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(s0),zB=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r<a.length;r++)s=a[r],this.outputQueue.push(s);return this.carryover=i[i.length-1],[2,!0]}})})},t}(Vd);var _B=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.decodeUTF8=function(){return new PB(this)},t}(xt),PB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.impl=new MB(e),i}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(s0),MB=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;if(i.upstream=e,Ne.env().get("IS_BROWSER"))i.decoder=new TextDecoder("utf-8");else{var r=require("string_decoder").StringDecoder;i.decoder=new r("utf8")}return i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(Vd);var o0=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(_B);function VB(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=HB(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new o0(a,t)];case 3:throw new Error(r.statusText)}})})}var HB=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function l0(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var u0=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return l0(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new o0(this.input,this.options)]})})},t}(a0);var c0=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return l0(this.url)?[2,new u0(this.url,this.fileOptions).iterator()]:[2,VB(this.url,this.fileOptions)]})})},t}(a0);function qB(n,t){return t===void 0&&(t={}),new r0(new c0(n),t)}function GB(n){var t=this,e=Hd(n);return Zt(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function YB(n){var t=this;return Zt(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,Hd(function(){return e.next()})]}})})})}function jB(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,UB.create(n,t)]})})}function KB(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,WB.create(n)]})})}var $B="2.6.0";Yt.CSVDataset=r0;Yt.Dataset=hs;Yt.FileDataSource=u0;Yt.TextLineDataset=t0;Yt.URLDataSource=c0;Yt.array=DB;Yt.csv=qB;Yt.func=GB;Yt.generator=YB;Yt.microphone=KB;Yt.version_data=$B;Yt.webcam=jB;Yt.zip=kB});var p0=Le((d0,Gd)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c<u.length;c++){o+=u.charCodeAt(c);var h=.02519603282416938*o;o=h>>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(d0,typeof Gd=="object"&&Gd,typeof define=="function"&&define)});var m0=Le((f0,Yd)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(f0,typeof Yd=="object"&&Yd,typeof define=="function"&&define)});var v0=Le((g0,jd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,u==l.length&&(o.d=o.x<<10^o.x>>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(g0,typeof jd=="object"&&jd,typeof define=="function"&&define)});var b0=Le((y0,Kd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h<c.length;++h)p[h&7]=p[h&7]<<15^c.charCodeAt(h)+p[h+1&7]<<13;for(;p.length<8;)p.push(0);for(h=0;h<8&&p[h]===0;++h);for(h==8?d=p[7]=-1:d=p[h],u.x=p,u.i=0,h=256;h>0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(y0,typeof Kd=="object"&&Kd,typeof define=="function"&&define)});var S0=Le((w0,$d)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f<v;++f)c&&(d^=c.charCodeAt((f+32)%c.length)),f===0&&(m=d),d^=d<<10,d^=d>>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(w0,typeof $d=="object"&&$d,typeof define=="function"&&define)});var I0=Le((L0,Xd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u<l.length+20;u++)o.b^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.a=s.a,o.b=s.b,o.c=s.c,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(L0,typeof Xd=="object"&&Xd,typeof define=="function"&&define)});var A0=Le((vV,Jo)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var N=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),N),R=new p(N),D=function(){for(var k=R.g(r),W=o,F=0;k<l;)k=(k+F)*i,W*=i,F=R.g(1);for(;k>=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,N=this,C=0,R=N.i=N.j=0,D=N.S=[];for(L||(w=[L++]);C<i;)D[C]=C++;for(C=0;C<i;C++)D[C]=D[R=c&R+w[C%L]+(S=D[C])],D[R]=S;(N.g=function(k){for(var W,F=0,P=N.i,H=N.j,_=N.S;k--;)W=_[P=c&P+1],F=F*i+_[c&(_[P]=_[H=c&H+W])+(_[H]=W)];return N.i=P,N.j=H,F})(i)}function f(w,S){return S.i=w.i,S.j=w.j,S.S=w.S.slice(),S}function m(w,S){var L=[],N=typeof w,C;if(S&&N=="object")for(C in w)try{L.push(m(w[C],S-1))}catch(R){}return L.length?L:N=="string"?w:w+"\0"}function g(w,S){for(var L=w+"",N,C=0;C<L.length;)S[c&C]=c&(N^=S[c&C]*19)+L.charCodeAt(C++);return b(S)}function v(){try{var w;return h&&(w=h.randomBytes)?w=w(i):(w=new Uint8Array(i),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(N){var S=e.navigator,L=S&&S.plugins;return[+new Date,e,L,e.screen,b(n)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),n),typeof Jo=="object"&&Jo.exports){Jo.exports=d;try{h=require("crypto")}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return d})})([],Math)});var N0=Le((yV,T0)=>{var XB=p0(),JB=m0(),ZB=v0(),QB=b0(),ez=S0(),tz=I0(),br=A0();br.alea=XB;br.xor128=JB;br.xorwow=ZB;br.xorshift7=QB;br.xor4096=ez;br.tychei=tz;T0.exports=br});var Z0=Le(ps=>{"use strict";Object.defineProperty(ps,"__esModule",{value:!0});var T=lr(),nz=N0();var Jd=function(n,t){return Jd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Jd(n,t)};function iz(n,t){Jd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function x0(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function C0(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}function ae(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&T.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the CPU backend."})})}var rz=T.kernel_impls.nonMaxSuppressionV3Impl,az=T.kernel_impls.split,sz=T.kernel_impls.tile,oz=T.kernel_impls.topkImpl,lz=T.kernel_impls.whereImpl;function Zd(n,t,e,i){if(e==="linear")return n.linear(t);if(e==="relu")return n.relu(t);if(e==="elu")return T.elu(t);if(e==="relu6")return n.relu6(t);if(e==="prelu")return n.prelu(t,i);throw new Error("Activation "+e+" has not been implemented for the CPU backend.")}var R0=function(n){iz(t,n);function t(){var e=n.call(this)||this;return e.blockSize=48,e.firstUse=!0,e.data=new T.DataStorage(e,T.engine()),e}return t.prototype.write=function(e,i,r){this.firstUse&&(this.firstUse=!1,T.env().get("IS_NODE")&&T.backend_util.warn(`
 ============================
 Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed things up dramatically, install our node backend, which binds to TensorFlow C++, by running npm i @tensorflow/tfjs-node, or npm i @tensorflow/tfjs-node-gpu if you have CUDA. Then call require('@tensorflow/tfjs-node'); (-gpu suffix for CUDA) at the start of your program. Visit https://github.com/tensorflow/tfjs-node for more details.
-============================`));var a={};return this.data.set(a,{values:e,dtype:r,refCount:1}),a},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return{dataId:a,shape:e,dtype:i}},t.prototype.incRef=function(e){var i=this.data.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.data.has(e)){var i=this.data.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){this.data.set(e,{values:i,dtype:a,refCount:1})},t.prototype.numDataIds=function(){return this.data.numDataIds()},t.prototype.read=function(e){return x0(this,void 0,void 0,function(){return C0(this,function(i){return[2,this.readSync(e)]})})},t.prototype.readSync=function(e){var i=this.data.get(e),r=i.dtype,a=i.complexTensorInfos;if(r==="complex64"){var s=this.readSync(a.real.dataId),o=this.readSync(a.imag.dataId);return T.backend_util.mergeRealAndImagArrays(s,o)}return this.data.get(e).values},t.prototype.bufferSync=function(e){var i=this.readSync(e.dataId),r=i;if(e.dtype==="string")try{r=i.map(function(a){return T.util.decodeString(a)})}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return T.buffer(e.shape,e.dtype,r)},t.prototype.makeOutput=function(e,i,r){var a=this.write(e,i,r);return T.engine().makeTensorFromDataId(a,i,r,this)},t.prototype.disposeData=function(e){if(this.data.has(e)){var i=this.data.get(e).complexTensorInfos;i!=null&&(this.disposeData(i.real.dataId),this.disposeData(i.imag.dataId)),this.data.delete(e)}},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.data.has(i)){var r=this.data.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.time=function(e){return x0(this,void 0,void 0,function(){var i,r;return C0(this,function(a){return i=T.util.now(),e(),r=T.util.now()-i,[2,{kernelMs:r}]})})},t.prototype.memory=function(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}},t.prototype.stridedSlice=function(e,i,r,a){ae(e,"stridedSlice");var s=T.slice_util.computeOutShape(i,r,a);if(s.some(function(p){return p===0}))return T.tensor([],s);for(var o=T.buffer(s,e.dtype),l=this.bufferSync(e),u=0;u<o.size;u++){for(var c=o.indexToLoc(u),h=new Array(c.length),d=0;d<h.length;d++)h[d]=c[d]*a[d]+i[d];o.set.apply(o,[l.get.apply(l,h)].concat(c))}return o.toTensor()},t.prototype.diag=function(e){for(var i=this.readSync(e.dataId),r=T.buffer([e.size,e.size],e.dtype),a=r.values,s=0;s<i.length;s++)a[s*e.size+s]=i[s];return r.toTensor()},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=T.slice(e,l,u).reshape(a);return c},t.prototype.reverse=function(e,i){ae(e,"reverse");for(var r=T.buffer(e.shape,e.dtype),a=this.bufferSync(e),s=function(l){var u=r.indexToLoc(l),c=u.slice();i.forEach(function(h){return c[h]=e.shape[h]-1-c[h]}),r.set.apply(r,[a.get.apply(a,c)].concat(u))},o=0;o<r.size;o++)s(o);return r.toTensor()},t.prototype.neg=function(e){return ae(e,"neg"),T.mul(T.scalar(-1),e)},t.prototype.addN=function(e){var i=this;ae(e,"addN");for(var r=e.map(function(c){return i.readSync(c.dataId)}),a=T.buffer(e[0].shape,e[0].dtype),s=a.values,o=0;o<e.length;o++)for(var l=r[o],u=0;u<s.length;u++)s[u]+=l[u];return a.toTensor()},t.prototype.softmax=function(e,i){var r=T.util.parseAxisParam([i],e.shape),a=T.max(e,r),s=T.backend_util.expandShapeToKeepDim(a.shape,r),o=T.sub(e,a.reshape(s)),l=T.exp(o),u=this.sum(l,r).reshape(s);return T.div(l,u)},t.prototype.pow=function(e,i){return ae([e,i],"pow"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.pow(r,a)})},t.prototype.batchMatMul=function(e,i,r,a){ae([e,i],"matMul");for(var s=r?e.shape[1]:e.shape[2],o=r?e.shape[2]:e.shape[1],l=a?i.shape[1]:i.shape[2],u=e.shape[0],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=r?[e.strides[0],1,e.strides[1]]:[e.strides[0],e.strides[1],1],p=d[0],f=d[1],m=d[2],g=a?[1,i.strides[1],i.strides[0]]:[i.strides[1],1,i.strides[0]],v=g[0],b=g[1],w=g[2],S=o*l,L=T.buffer([u,o,l],e.dtype),N=L.values,C=this.blockSize,R=0;R<u;R++)for(var D=0;D<o;D+=C)for(var k=0;k<l;k+=C)for(var W=0;W<s;W+=C)for(var F=Math.min(D+C,o),P=Math.min(k+C,l),H=Math.min(W+C,s),_=D;_<F;_++)for(var j=k;j<P;j++){for(var K=0,q=W;q<H;q++)K+=c[R*p+_*f+q*m]*h[q*v+j*b+R*w];N[R*S+(_*l+j)]+=K}return L.toTensor()},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=this.batchMatMul(i,r,a,s);return o&&(c=T.add(c,o)),l&&(c=Jd(this,c,l,u)),c},t.prototype.floorDiv=function(e,i){ae([e,i],"floorDiv");var r=function(s,o){return Math.floor(s/o)},a="int32";return this.broadcastedBinaryOp(e,i,a,r)},t.prototype.sum=function(e,i){ae(e,"sum"),T.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=0,m=0;m<u;++m)f+=h[p+m];c[d]=f}return l},t.prototype.prod=function(e,i){ae(e,"sum");for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=1,m=0;m<u;++m)f*=h[p+m];c[d]=f}return l},t.prototype.unsortedSegmentSum=function(e,i,r){ae(e,"unsortedSegmentSum");for(var a=[],s=e.rank-i.rank,o=0;o<s;++o)i=i.expandDims(o+1);for(var o=0;o<r;++o){var l=T.scalar(o,"int32"),u=T.equal(l,i).asType("float32"),c=u.mul(e).sum(0);a.push(c)}return T.stack(a)},t.prototype.argMin=function(e,i){ae(e,"argMin");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMin",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v<f&&(f=v,m=g)}c[d]=m}return l},t.prototype.argMax=function(e,i){ae(e,"argMax");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMax",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v>f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;d<u.length;d+=c)for(var p=0;p<c;p++){var f=h(d,p);if(p===0)l[f]=r?0:u[f];else{var m=h(d,p-1);l[f]=r?u[m]+l[m]:u[f]+l[m]}}return o},t.prototype.equal=function(e,i){return ae([e,i],"equal"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r===a?1:0})},t.prototype.notEqual=function(e,i){return ae([e,i],"notEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r!==a?1:0})},t.prototype.less=function(e,i){return ae([e,i],"less"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<a?1:0})},t.prototype.lessEqual=function(e,i){return ae([e,i],"lessEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<=a?1:0})},t.prototype.greater=function(e,i){return ae([e,i],"greater"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>a?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d<a.length;d++)for(var p=0;p<h;p++)a[d]===1?u[c++]=s[d]:u[c++]=o[d];return l},t.prototype.where=function(e){ae([e],"where");var i=this.readSync(e.dataId);return lz(e.shape,i)},t.prototype.topk=function(e,i,r){ae(e,"topk");var a=this.readSync(e.dataId);return oz(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){ae(e,"min"),T.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];m<p&&(p=m)}u[h]=p}return o},t.prototype.minimum=function(e,i){return ae([e,i],"minimum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.min(r,a)})},t.prototype.mod=function(e,i){return ae([e,i],"mod"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r%a;return r<0&&a<0||r>=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p&&m}u[h]=p}return o},t.prototype.any=function(e,i){ae(e,"any"),T.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p||m}u[h]=p}return o},t.prototype.squaredDifference=function(e,i){return ae([e,i],"squaredDifference"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r-a;return s*s})},t.prototype.linear=function(e){return e},t.prototype.relu=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.max(0,a[s]);return i},t.prototype.relu6=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.min(Math.max(0,a[s]),6);return i},t.prototype.prelu=function(e,i){return ae([e,i],"prelu"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return r<0?a*r:r})},t.prototype.eluDer=function(e,i){ae([e,i],"eluDer");for(var r=new Float32Array(i.size),a=this.readSync(i.dataId),s=this.readSync(e.dataId),o=0;o<a.length;++o){var l=a[o];l>=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Jd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),N=this.readSync(i.dataId),C=d.values,R=0;R<r.batchSize;++R)for(var D=R*p,k=R*v,W=0;W<r.outHeight;++W)for(var F=k+W*b,P=W*r.strideHeight-c,H=0;H<a;H++){var _=P+H*o;if(_<0||_>=r.inHeight)continue;for(var j=H*i.strides[0],K=D+_*f,q=0;q<r.outWidth;++q)for(var G=F+q*w,Z=q*r.strideWidth-u,X=0;X<s;X++){var ee=Z+X*l;if(ee<0||ee>=r.inWidth)continue;for(var ne=j+X*i.strides[1],ie=K+ee*m,te=ne,re=0;re<r.inChannels;++re){for(var le=L[ie+re*g],he=0;he<r.outChannels;++he)C[G+he*S]+=le*N[te+he];te+=r.outChannels}}}return d.toTensor()},t.prototype.conv3d=function(e,i,r){for(var a=r.filterDepth,s=r.filterHeight,o=r.filterWidth,l=r.dilationDepth,u=r.dilationHeight,c=r.dilationWidth,h=r.padInfo.front,d=r.padInfo.left,p=r.padInfo.top,f=T.buffer(r.outShape,e.dtype),m=this.readSync(e.dataId),g=this.readSync(i.dataId),v=f.values,b=0;b<r.batchSize;++b)for(var w=b*e.strides[0],S=b*f.strides[0],L=0;L<r.outDepth;++L)for(var N=S+L*f.strides[1],C=L*r.strideDepth-h,R=0;R<a;R++){var D=C+R*l;if(D<0||D>=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F<r.outHeight;++F)for(var P=N+F*f.strides[2],H=F*r.strideHeight-p,_=0;_<s;_++){var j=H+_*u;if(j<0||j>=r.inHeight)continue;for(var K=k+_*i.strides[1],q=W+j*e.strides[2],G=0;G<r.outWidth;++G)for(var Z=P+G*r.outChannels,X=G*r.strideWidth-d,ee=0;ee<o;ee++){var ne=X+ee*c;if(ne<0||ne>=r.inWidth)continue;for(var ie=K+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le<r.inChannels;++le){for(var he=m[te+le],ye=0;ye<r.outChannels;++ye)v[Z+ye]+=he*g[re+ye];re+=r.outChannels}}}}return f.toTensor()},t.prototype.conv2dDerInput=function(e,i,r){ae([e,i],"conv2dDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=this.readSync(e.dataId),l=this.readSync(i.dataId),u=i.strides,c=u[0],h=u[1],d=u[2],p=r.batchSize,f=r.filterHeight,m=r.filterWidth,g=r.inChannels,v=r.inHeight,b=r.inWidth,w=r.outChannels,S=r.outHeight,L=r.outWidth,N=r.strideHeight,C=r.strideWidth,R=r.dataFormat,D=f-1-r.padInfo.top,k=m-1-r.padInfo.left,W=R==="channelsLast",F=a.strides[0],P=W?a.strides[1]:a.strides[2],H=W?a.strides[2]:1,_=W?1:a.strides[1],j=e.strides[0],K=W?e.strides[1]:e.strides[2],q=W?e.strides[2]:1,G=W?1:e.strides[1],Z=0;Z<p;++Z)for(var X=0;X<g;++X)for(var ee=0;ee<v;++ee)for(var ne=ee-D,ie=Math.max(0,Math.ceil(ne/N)),te=Math.min(S,(f+ne)/N),re=0;re<b;++re){for(var le=re-k,he=Math.max(0,Math.ceil(le/C)),ye=Math.min(L,(m+le)/C),Oe=0,Fe=ie;Fe<te;++Fe)for(var _e=Fe*N-ne,Me=he;Me<ye;++Me)for(var dt=Me*C-le,Lt=j*Z+K*Fe+q*Me,Ge=c*(f-1-_e)+h*(m-1-dt)+d*X,Ue=0;Ue<w;++Ue){var me=o[Lt+G*Ue],Ft=l[Ge+Ue];Oe+=me*Ft}var lt=F*Z+P*ee+H*re+_*X;s[lt]=Oe}return a.toTensor()},t.prototype.conv3dDerInput=function(e,i,r){for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=o[3],d=this.readSync(e.dataId),p=e.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],N=w[2],C=w[3],R=r.batchSize,D=r.filterDepth,k=r.filterHeight,W=r.filterWidth,F=r.inChannels,P=r.inDepth,H=r.inHeight,_=r.inWidth,j=r.outChannels,K=r.outDepth,q=r.outHeight,G=r.outWidth,Z=r.strideDepth,X=r.strideHeight,ee=r.strideWidth,ne=D-1-r.padInfo.front,ie=k-1-r.padInfo.top,te=W-1-r.padInfo.left,re=0;re<R;++re)for(var le=0;le<F;++le)for(var he=0;he<P;++he)for(var ye=he-ne,Oe=Math.max(0,Math.ceil(ye/Z)),Fe=Math.min(K,(D+ye)/Z),_e=0;_e<H;++_e)for(var Me=_e-ie,dt=Math.max(0,Math.ceil(Me/X)),Lt=Math.min(q,(k+Me)/X),Ge=0;Ge<_;++Ge){for(var Ue=Ge-te,me=Math.max(0,Math.ceil(Ue/ee)),Ft=Math.min(G,(W+Ue)/ee),lt=0,pt=Oe;pt<Fe;++pt)for(var En=pt*Z-ye,Et=dt;Et<Lt;++Et)for(var It=Et*X-Me,Ce=me;Ce<Ft;++Ce)for(var Wt=Ce*ee-Ue,ft=f*re+m*pt+g*Et+v*Ce,it=S*(D-1-En)+L*(k-1-It)+N*(W-1-Wt)+C*le,cn=0;cn<j;++cn){var Os=d[ft+cn],_t=b[it+cn];lt+=Os*_t}s[l*re+u*he+c*_e+h*Ge+le]=lt}return a.toTensor()},t.prototype.conv2dDerFilter=function(e,i,r){ae([e,i],"conv2dDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=r.dataFormat==="channelsLast",c=T.buffer(r.filterShape,"float32"),h=r.padInfo.left,d=r.padInfo.top,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((d-m)/a)),v=Math.min(r.outHeight,(r.inHeight+d-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((h-b)/s)),S=Math.min(r.outWidth,(r.inWidth+h-b)/s),L=0;L<r.inChannels;++L)for(var N=0;N<r.outChannels;++N){for(var C=0,R=0;R<r.batchSize;++R)for(var D=g;D<v;++D)for(var k=m+D*a-d,W=w;W<S;++W){var F=b+W*s-h;u?C+=p.get(R,k,F,L)*f.get(R,D,W,N):C+=p.get(R,L,k,F)*f.get(R,N,D,W)}c.set(C,m,b,L,N)}return c.toTensor()},t.prototype.conv3dDerFilter=function(e,i,r){for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=T.buffer(r.filterShape,"float32"),d=h.values,p=h.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],N=w[2],C=w[3],R=this.readSync(e.dataId),D=e.strides,k=D[0],W=D[1],F=D[2],P=D[3],H=r.padInfo.front,_=r.padInfo.left,j=r.padInfo.top,K=0;K<l;++K)for(var q=Math.max(0,Math.ceil((H-K)/a)),G=Math.min(r.outDepth,(r.inDepth+H-K)/a),Z=K*f,X=0;X<u;++X)for(var ee=Math.max(0,Math.ceil((j-X)/s)),ne=Math.min(r.outHeight,(r.inHeight+j-X)/s),ie=X*m+Z,te=0;te<c;++te)for(var re=Math.max(0,Math.ceil((_-te)/o)),le=Math.min(r.outWidth,(r.inWidth+_-te)/o),he=te*g+ie,ye=0;ye<r.inChannels;++ye)for(var Oe=ye*v+he,Fe=0;Fe<r.outChannels;++Fe){for(var _e=0,Me=0;Me<r.batchSize;++Me)for(var dt=Me*k,Lt=Me*S,Ge=q;Ge<G;++Ge)for(var Ue=K+Ge*a-H,me=Ue*W+dt,Ft=Ge*L+Lt,lt=ee;lt<ne;++lt)for(var pt=X+lt*s-j,En=pt*F+me,Et=lt*N+Ft,It=re;It<le;++It){var Ce=te+It*o-_,Wt=Ce*P+En,ft=It*C+Et;_e+=R[Wt+ye]*b[ft+Fe]}d[Oe+Fe]=_e}return h.toTensor()},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.depthwiseConv2D(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Jd(this,u,o,l)),u},t.prototype.depthwiseConv2D=function(e,i,r){ae([e,i],"depthwiseConv2D");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.outChannels/r.inChannels,d=T.buffer(r.outShape,e.dtype),p=this.readSync(e.dataId),f=this.readSync(i.dataId),m=d.values,g=0;g<r.batchSize;++g)for(var v=g*e.strides[0],b=g*d.strides[0],w=0;w<r.outHeight;++w)for(var S=b+w*d.strides[1],L=w*r.strideHeight-u,N=0;N<a;++N){var C=L+N*o;if(C<0||C>=r.inHeight)continue;for(var R=N*i.strides[0],D=v+C*e.strides[1],k=0;k<r.outWidth;++k)for(var W=S+k*d.strides[2],F=k*r.strideWidth-c,P=0;P<s;++P){var H=F+P*l;if(H<0||H>=r.inWidth)continue;for(var _=R+P*i.strides[1],j=D+H*r.inChannels,K=W,q=_,G=0;G<r.inChannels;++G){for(var Z=p[j+G],X=0;X<h;++X)m[K+X]+=Z*f[q+X];K+=h,q+=h}}}return d.toTensor()},t.prototype.depthwiseConv2DDerInput=function(e,i,r){ae([e,i],"depthwiseConv2DDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=this.readSync(e.dataId),d=e.strides,p=d[0],f=d[1],m=d[2],g=this.readSync(i.dataId),v=i.strides,b=v[0],w=v[1],S=v[2],L=r.batchSize,N=r.filterHeight,C=r.filterWidth,R=r.inChannels,D=r.inHeight,k=r.inWidth,W=r.outChannels,F=r.outHeight,P=r.outWidth,H=r.strideHeight,_=r.strideWidth,j=N-1-r.padInfo.top,K=C-1-r.padInfo.left,q=W/R,G=0;G<L;++G)for(var Z=0;Z<R;++Z)for(var X=0;X<D;++X)for(var ee=X-j,ne=Math.max(0,Math.ceil(ee/H)),ie=Math.min(F,(N+ee)/H),te=0;te<k;++te){for(var re=te-K,le=Math.max(0,Math.ceil(re/_)),he=Math.min(P,(C+re)/_),ye=0,Oe=ne;Oe<ie;++Oe)for(var Fe=Oe*H-ee,_e=le;_e<he;++_e)for(var Me=_e*_-re,dt=p*G+f*Oe+m*_e,Lt=b*(N-1-Fe)+w*(C-1-Me)+S*Z,Ge=0;Ge<q;++Ge){var Ue=Z*q+Ge,me=h[dt+Ue],Ft=g[Lt+Ge];ye+=me*Ft}s[l*G+u*X+c*te+Z]=ye}return a.toTensor()},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){ae([e,i],"depthwiseConv2DDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=T.buffer(r.filterShape,"float32"),c=r.padInfo.left,h=r.padInfo.top,d=r.outChannels/r.inChannels,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((h-m)/a)),v=Math.min(r.outHeight,(r.inHeight+h-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((c-b)/s)),S=Math.min(r.outWidth,(r.inWidth+c-b)/s),L=0;L<r.outChannels;++L){for(var N=Math.trunc(L/d),C=L%d,R=0,D=0;D<r.batchSize;++D)for(var k=g;k<v;++k)for(var W=m+k*a-h,F=w;F<S;++F){var P=b+F*s-c;R+=p.get(D,W,P,N)*f.get(D,k,F,L)}u.set(R,m,b,N,C)}return u.toTensor()},t.prototype.tile=function(e,i){return ae(e,"tile"),sz(this.bufferSync(e),i)},t.prototype.gather=function(e,i,r){ae([e,i],"gather");var a=e.shape.slice(),s=this.readSync(i.dataId);a[r]=s.length;for(var o=T.buffer(a,e.dtype),l=this.bufferSync(e),u=0;u<o.size;++u){var c=o.indexToLoc(u),h=c.slice();h[r]=s[c[r]];var d=l.locToIndex(h);o.values[u]=l.values[d]}return o.toTensor()},t.prototype.batchToSpaceND=function(e,i,r){ae([e],"batchToSpaceND");var a=i.reduce(function(h,d){return h*d}),s=T.backend_util.getReshaped(e.shape,i,a),o=T.backend_util.getPermuted(s.length,i.length),l=T.backend_util.getReshapedPermuted(e.shape,i,a),u=T.backend_util.getSliceBeginCoords(r,i.length),c=T.backend_util.getSliceSize(l,r,i.length);return T.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.pool3d=function(e,i,r){ae(e,"pool3d");for(var a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=r==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,b=this.readSync(e.dataId),w=T.buffer(i.outShape,e.dtype),S=w.values,L=i.outShape[1]*i.outShape[2]*i.outShape[3]*i.outShape[4],N=i.outShape[2]*i.outShape[3]*i.outShape[4],C=i.outShape[3]*i.outShape[4],R=i.outShape[4],D=0;D<i.batchSize;++D)for(var k=D*L,W=D*e.strides[0],F=0;F<i.inChannels;++F)for(var P=0;P<i.outDepth;++P){for(var H=P*a-f,_=H;_<0;)_+=l;for(var j=Math.min(i.inDepth,h+H),K=k+P*N,q=0;q<i.outHeight;++q){for(var G=q*s-m,Z=G;Z<0;)Z+=u;for(var X=Math.min(i.inHeight,d+G),ee=K+q*C,ne=0;ne<i.outWidth;++ne){for(var ie=ne*o-g,te=ie;te<0;)te+=c;for(var re=Math.min(i.inWidth,p+ie),le=ee+ne*R,he=v,ye=0,Oe=0,Fe=_;Fe<j;Fe+=l){for(var _e=W+Fe*e.strides[1],Me=Z;Me<X;Me+=u){for(var dt=_e+Me*e.strides[2],Lt=te;Lt<re;Lt+=c){var Ge=dt+Lt*e.strides[3],Ue=b[Ge+F];if(r==="max"&&Ue>he?he=Ue:r==="avg"&&(ye+=Ue,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),N=this.bufferSync(e),C=0;C<r.batchSize;++C)for(var R=0;R<r.inChannels;++R)for(var D=0;D<r.inDepth;++D)for(var k=0;k<r.inHeight;++k)for(var W=0;W<r.inWidth;++W){for(var F=D-v,P=k-w,H=W-b,_=0,j=0;j<f;j+=h){var K=(F+j)/a;if(K<0||K>=r.outDepth||Math.floor(K)!==K)continue;for(var q=0;q<m;q+=d){var G=(P+q)/s;if(G<0||G>=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z<g;Z+=p){var X=(H+Z)/o;if(X<0||X>=r.outWidth||Math.floor(X)!==X)continue;var ee=N.get(C,K,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b<i.batchSize;++b)for(var w=0;w<i.inChannels;++w)for(var S=0;S<i.outDepth;++S){for(var L=S*a-f,N=L;N<0;)N+=l;for(var C=Math.min(i.inDepth,h+L),R=0;R<i.outHeight;++R){for(var D=R*s-m,k=D;k<0;)k+=u;for(var W=Math.min(i.inHeight,d+D),F=0;F<i.outWidth;++F){for(var P=F*o-g,H=P;H<0;)H+=c;for(var _=Math.min(i.inWidth,p+P),j=Number.NEGATIVE_INFINITY,K=-1,q=N;q<C;q+=l)for(var G=q-L,Z=k;Z<W;Z+=u)for(var X=Z-D,ee=H;ee<_;ee+=c){var ne=ee-P,ie=v.get(b,q,Z,ee,w);ie>=j&&(j=ie,K=G*d*p+X*d+ne)}r.set(K,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),N=0;N<a.batchSize;++N)for(var C=0;C<a.inChannels;++C)for(var R=0;R<a.inDepth;++R)for(var D=0;D<a.inHeight;++D)for(var k=0;k<a.inWidth;++k){for(var W=R-g,F=D-b,P=k-v,H=0,_=0;_<p;_+=c){var j=(W+_)/o;if(j<0||j>=a.outDepth||Math.floor(j)!==j)continue;for(var K=0;K<f;K+=h){var q=(F+K)/l;if(q<0||q>=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<m;G+=d){var Z=(P+G)/u;if(Z<0||Z>=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(N,j,q,Z,C),ee=_*f*m+K*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(N,j,q,Z,C);H+=ie*ne}}}w.set(H,N,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b<o;b++)for(var w=0;w<i;w++)for(var S=g*w,L=Math.floor(S),N=S-L,C=Math.min(l-1,Math.ceil(S)),R=b*e.strides[0]+L*e.strides[1],D=b*e.strides[0]+C*e.strides[1],k=0;k<r;k++)for(var W=v*k,F=Math.floor(W),P=W-F,H=Math.min(u-1,Math.ceil(W)),_=R+F*e.strides[2],j=D+F*e.strides[2],K=R+H*e.strides[2],q=D+H*e.strides[2],G=0;G<c;G++){var Z=h[_+G],X=h[j+G],ee=h[K+G],ne=h[q+G],ie=Z+(ee-Z)*P,te=X+(ne-X)*P,re=ie+(te-ie)*N;d[m++]=re}return T.tensor(d,[o,i,r,c])},t.prototype.resizeBilinearBackprop=function(e,i,r){ae([e,i],"resizeBilinearBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=[r&&h>1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S<s;S++)for(var L=S*i.strides[0],N=0;N<h;N++)for(var C=N*g,R=Math.floor(C),D=Math.min(Math.ceil(C),o-1),k=L+R*i.strides[1],W=L+D*i.strides[1],F=C-R,P=1-F,H=0;H<d;H++)for(var _=H*v,j=Math.floor(_),K=Math.min(Math.ceil(_),l-1),q=_-j,G=1-q,Z=k+j*i.strides[2],X=k+K*i.strides[2],ee=W+j*i.strides[2],ne=W+K*i.strides[2],ie=P*G,te=P*q,re=F*G,le=F*q,he=0;he<u;he++){var ye=b[w++];p[Z+he]+=ye*ie,p[X+he]+=ye*te,p[ee+he]+=ye*re,p[ne+he]+=ye*le}return T.tensor4d(p,[s,l,o,u],i.dtype)},t.prototype.resizeNearestNeighbor=function(e,i,r,a){ae(e,"resizeNearestNeighbor");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(o*i*r*c),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b<o;b++)for(var w=b*e.strides[0],S=0;S<i;S++)for(var L=m*S,N=Math.min(l-1,a?Math.round(L):Math.floor(L)),C=w+N*e.strides[1],R=0;R<r;R++)for(var D=g*R,k=Math.min(u-1,a?Math.round(D):Math.floor(D)),W=C+k*e.strides[2],F=0;F<c;F++){var P=h[W+F];d[v++]=P}return T.tensor(d,[o,i,r,c],e.dtype)},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){ae([e,i],"resizeNearestNeighborBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=this.readSync(e.dataId),m=[r&&h>1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,N=Math.ceil(S)*2+2,C=0;C<s;C++)for(var R=C*i.strides[0],D=0;D<o;D++)for(var k=R+D*i.strides[1],W=Math.floor(D*w),F=Math.floor(W-L/2),P=0;P<l;P++)for(var H=k+P*i.strides[2],_=Math.floor(P*S),j=Math.floor(_-N/2),K=0;K<u;K++){for(var q=0,G=0;G<L;G++){var Z=G+F;if(Z<0||Z>=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie<N;ie++){var te=ie+j;if(te<0||te>=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+K])}}p[H+K]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p<c;p++){var f=d(p),m=u[p]*Math.pow(r+a*f,-s);h[p]=m}return T.tensor4d(h,e.shape)},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){ae(e,"LRNGrad");for(var u=e.shape[3],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=this.readSync(r.dataId),p=new Float32Array(e.size),f=e.size,m=0;m<f;m++){for(var g=m%u,v=m-g+Math.max(0,g-a),b=m-g+Math.min(u,g+a+1),w=0,S=v;S<b;S++)w+=Math.pow(h[S],2);w=o*w+s;for(var S=v;S<b;S++){var L=-2*o*l*h[S]*d[m]/w;m===S&&(L+=Math.pow(w,-l)),L*=c[m],p[S]+=L}}return T.tensor4d(p,e.shape)},t.prototype.multinomial=function(e,i,r,a){ae(e,"multinomial");for(var s=i?e:T.softmax(e),o=s.shape[0],l=s.shape[1],u=T.zeros([o,r],"int32"),c=this.readSync(u.dataId),h=this.readSync(s.dataId),d=0;d<o;++d){var p=d*l,f=new Float32Array(l-1);f[0]=h[p];for(var m=1;m<f.length;++m)f[m]=f[m-1]+h[p+m];for(var g=nz.alea(a.toString()),v=d*r,b=0;b<r;++b){var w=g();c[v+b]=f.length;for(var S=0;S<f.length;S++)if(w<f[S]){c[v+b]=S;break}}}return u},t.prototype.oneHot=function(e,i,r,a){ae(e,"oneHot");var s=new Float32Array(e.size*i);s.fill(a);for(var o=this.readSync(e.dataId),l=0;l<e.size;++l)o[l]>=0&&o[l]<i&&(s[l*i+o[l]]=r);return T.tensor2d(s,[e.size,i],"int32")},t.prototype.nonMaxSuppression=function(e,i,r,a,s){ae(e,"nonMaxSuppression");var o=this.readSync(e.dataId),l=this.readSync(i.dataId);return rz(o,l,r,a,s)},t.prototype.depthToSpace=function(e,i,r){T.util.assert(r==="NHWC",function(){return"Only NHWC dataFormat supported on CPU for depthToSpace. Got "+r}),T.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m<a;++m)for(var g=0;g<u;++g)for(var v=Math.floor(g/i),b=g%i,w=0;w<c;++w)for(var S=Math.floor(w/i),L=w%i,N=(b*i+L)*h,C=0;C<h;++C){var R=C+N,D=R+l*(S+o*(v+s*m));p[f++]=d[D]}return T.tensor4d(p,[a,u,c,h])},t.prototype.broadcastedBinaryOp=function(e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(e.shape,i.shape),o=T.buffer(s,r),l=this.readSync(e.dataId),u=this.readSync(i.dataId),c=T.backend_util.getBroadcastDims(e.shape,s),h=T.backend_util.getBroadcastDims(i.shape,s),d=o.values;if(c.length+h.length===0)for(var p=0;p<d.length;++p)d[p]=a(l[p%l.length],u[p%u.length]);else for(var f=this.bufferSync(e),m=this.bufferSync(i),g=function(v){var b=o.indexToLoc(v),w=b.slice(-e.rank);c.forEach(function(C){return w[C]=0});var S=f.locToIndex(w),L=b.slice(-i.rank);h.forEach(function(C){return L[C]=0});var N=m.locToIndex(L);d[v]=a(l[S],u[N])},p=0;p<d.length;++p)g(p);return o.toTensor()},t.prototype.split=function(e,i,r){return az(e,i,r)},t.prototype.dispose=function(){},t.prototype.floatPrecision=function(){return 32},t.prototype.epsilon=function(){return n.prototype.epsilon.call(this)},t.prototype.cropAndResize=function(e,i,r,a,s,o){for(var l=e.shape,u=l[0],c=l[1],h=l[2],d=l[3],p=i.shape[0],f=a[0],m=a[1],g=T.buffer([p,f,m,d],"float32"),v=this.readSync(i.dataId),b=this.readSync(r.dataId),w=this.readSync(e.dataId),S=e.strides,L=g.strides,N=0;N<p;N++){var C=N*4,R=v[C],D=v[C+1],k=v[C+2],W=v[C+3],F=b[N];if(F>=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_<f;_++){var j=f>1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(j<0||j>c-1){for(var K=0;K<m;K++)for(var q=0;q<d;q++){var G=q+K*L[2]+_*L[1]+N*L[0];g.values[G]=o}continue}if(s==="bilinear")for(var Z=Math.floor(j),X=Math.ceil(j),ee=j-Z,K=0;K<m;K++){var ne=m>1?D*(h-1)+K*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+K*L[2]+_*L[1]+N*L[0];g.values[G]=o}continue}for(var ie=Math.floor(ne),te=Math.ceil(ne),re=ne-ie,q=0;q<d;q++){var G=q+ie*S[2]+Z*S[1]+F*S[0],le=w[G];G=q+te*S[2]+Z*S[1]+F*S[0];var he=w[G];G=q+ie*S[2]+X*S[1]+F*S[0];var ye=w[G];G=q+te*S[2]+X*S[1]+F*S[0];var Oe=w[G],Fe=le+(he-le)*re,_e=ye+(Oe-ye)*re;G=q+K*L[2]+_*L[1]+N*L[0],g.values[G]=Fe+(_e-Fe)*ee}}else for(var K=0;K<m;++K){var ne=m>1?D*(h-1)+K*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+K*L[2]+_*L[1]+N*L[0];g.values[G]=o}continue}for(var Me=Math.round(ne),dt=Math.round(j),q=0;q<d;q++){var Lt=q+Me*S[2]+dt*S[1]+F*S[0],Ge=q+K*L[2]+_*L[1]+N*L[0];g.values[Ge]=w[Lt]}}}}return g.toTensor()},t.prototype.sparseToDense=function(e,i,r,a){var s=T.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.sliceSize,c=s.strides,h=s.outputSize,d=!1;return this.scatter(e,i,r,h,u,l,o,c,a,d)},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=T.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3];if(l===0)return T.tensor([],o,e.dtype);for(var h=new T.TensorBuffer([l,u],e.dtype),d=this.readSync(i.dataId),p=this.readSync(e.dataId),f=0;f<l;f++){for(var m=[],g=0,v=0;v<a;v++){var b=d[f*a+v];g+=b*c[v],m.push(b)}if(g<0||g>=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w<u;w++)h.values[f*u+w]=p[g*u+w]}return h.toTensor().reshape(o)},t.prototype.scatterND=function(e,i,r){var a=T.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=T.scalar(0),d=!0;return this.scatter(e,i,r,c,l,o,s,u,h,d)},t.prototype.fill=function(e,i,r){r=r||T.util.inferDtype(i);var a=T.util.getArrayFromDType(r,T.util.sizeFromShape(e));return a.fill(i),T.engine().makeTensor(a,e,r,this)},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported for string tensors");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){var i=T.util.getArrayFromDType(e.dtype,T.util.sizeFromShape(e.shape));return this.makeOutput(i,e.shape,e.dtype)},t.prototype.linspace=function(e,i,r){return T.backend_util.linspaceImpl(e,i,r)},t.prototype.scatter=function(e,i,r,a,s,o,l,u,c,h){var d=[a/s,s],p=this.readSync(e.dataId),f=this.readSync(i.dataId);if(a===0)return T.tensor([],r,i.dtype);var m=new T.TensorBuffer(d,i.dtype);m.values.fill(this.readSync(c.dataId)[0]);for(var g=0;g<o;g++){for(var v=[],b=0,w=0;w<l;w++){var S=p[g*l+w];v.push(S),b+=S*u[w]}if(b<0||b>=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;L<s;L++)h?m.values[b*s+L]+=f[g*s+L]:m.values[b*s+L]=i.rank===0?f[0]:f[g*s+L]}return m.toTensor().reshape(r)},t}(T.KernelBackend);function O0(n){for(var t=new Float32Array(n.length),e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}var uz=function(n){var t=n.inputs.x,e=n.backend,i=new Float32Array(T.util.sizeFromShape(t.shape));if(t.dtype!=="complex64"){var r=e.data.get(t.dataId).values;i=O0(r)}else for(var a=e.data.get(t.dataId),s=a.complexTensorInfos.real,o=a.complexTensorInfos.imag,l=e.data.get(s.dataId).values,u=e.data.get(o.dataId).values,c=0;c<l.length;c++){var h=l[c],d=u[c];i[c]=Math.hypot(h,d)}return e.makeOutput(i,t.shape,"float32")},cz={kernelName:T.Abs,backendName:"cpu",kernelFunc:uz};function wr(n){return function(t,e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=T.util.computeStrides(s),u=T.util.sizeFromShape(s),c=T.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=T.util.computeStrides(t),f=T.util.computeStrides(e),m=T.backend_util.getBroadcastDims(t,s),g=T.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(var v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(var b=function(w){var S=T.util.indexToLoc(w,o,l),L=S.slice(-h);m.forEach(function(D){return L[D]=0});var N=T.util.locToIndex(L,h,p),C=S.slice(-d);g.forEach(function(D){return C[D]=0});var R=T.util.locToIndex(C,d,f);c[w]=n(i[N],r[R])},v=0;v<c.length;++v)b(v);return[c,s]}}function Cn(n){var t=n.inputs,e=n.backend,i=t.real,r=t.imag,a=e.data.get(i.dataId).values,s=e.data.get(r.dataId).values,o=e.makeTensorInfo(i.shape,"complex64"),l=e.data.get(o.dataId);return l.complexTensorInfos={real:e.makeTensorInfo(i.shape,"float32",a),imag:e.makeTensorInfo(r.shape,"float32",s)},o}var hz={kernelName:T.Complex,backendName:"cpu",kernelFunc:Cn};function ca(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var dz={kernelName:T.Identity,backendName:"cpu",kernelFunc:ca};function fs(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.real,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var pz={kernelName:T.Real,backendName:"cpu",kernelFunc:fs};function ms(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.dtype;if(a==="complex64"){if(r.dtype==="complex64")return ca({inputs:{x:r},backend:e});var s=T.zeros(r.shape),o=ms({inputs:{x:r},backend:e,attrs:{dtype:"float32"}}),l=Cn({inputs:{real:o,imag:s},backend:e});return s.dispose(),e.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){var u=fs({inputs:{input:r},backend:e}),l=ms({inputs:{x:u},backend:e,attrs:{dtype:a}});return e.disposeIntermediateTensorInfo(u),l}if(!T.util.hasEncodingLoss(r.dtype,a)){var l=ca({inputs:{x:r},backend:e});return{dataId:l.dataId,shape:l.shape,dtype:a}}if(a==="int32"){var c=e.data.get(r.dataId).values,h=Int32Array.from(c);return e.makeTensorInfo(r.shape,"int32",h)}if(a==="bool"){var d=e.data.get(r.dataId).values,p=T.util.toTypedArray([0],r.dtype),f=wr(function(v,b){return v!==b?1:0})(r.shape,[],d,p,"bool"),m=f[0],g=f[1];return e.makeTensorInfo(g,"bool",m)}throw new Error("Error in Cast: failed to cast "+r.dtype+" to "+a)}var fz={kernelName:T.Cast,backendName:"cpu",kernelFunc:ms};function ha(n,t,e,i){return e==null?function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;ae([l,u],n);var h=c.data.get(l.dataId).values,d=c.data.get(u.dataId).values,p=i||l.dtype,f=t(l.shape,u.shape,h,d,p),m=f[0],g=f[1];return c.makeTensorInfo(g,p,m)}:function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;if(l.dtype==="complex64"||u.dtype==="complex64"){var h=ms({inputs:{x:l},backend:c,attrs:{dtype:"complex64"}}),d=c.data.get(h.dataId),p=d.complexTensorInfos.real,f=d.complexTensorInfos.imag,m=c.data.get(p.dataId).values,g=c.data.get(f.dataId).values,v=ms({inputs:{x:u},backend:c,attrs:{dtype:"complex64"}}),b=c.data.get(v.dataId),w=b.complexTensorInfos.real,S=b.complexTensorInfos.imag,L=c.data.get(w.dataId).values,N=c.data.get(S.dataId).values,C=e(l.shape,u.shape,m,g,L,N),R=C[0],D=C[1],k=C[2],W=c.makeTensorInfo(k,"float32",R),F=c.makeTensorInfo(k,"float32",D),P=Cn({inputs:{real:W,imag:F},backend:c});return c.disposeIntermediateTensorInfo(h),c.disposeIntermediateTensorInfo(v),c.disposeIntermediateTensorInfo(W),c.disposeIntermediateTensorInfo(F),P}else{var H=c.data.get(l.dataId).values,_=c.data.get(u.dataId).values,j=i||l.dtype,K=t(l.shape,u.shape,H,_,j),q=K[0],k=K[1];return c.makeTensorInfo(k,j,q)}}}function Zd(n){return function(t,e,i,r,a,s){var o=T.backend_util.assertAndGetBroadcastShape(t,e),l=T.util.sizeFromShape(o),u=o.length,c=T.util.computeStrides(o),h=T.util.getTypedArrayFromDType("float32",l),d=T.util.getTypedArrayFromDType("float32",l),p=T.backend_util.getBroadcastDims(t,o),f=T.backend_util.getBroadcastDims(e,o),m=T.backend_util.mergeRealAndImagArrays(i,r),g=T.backend_util.mergeRealAndImagArrays(a,s),v=t.length,b=T.util.computeStrides(t),w=e.length,S=T.util.computeStrides(e);if(p.length+f.length===0)for(var L=0;L<h.length;L++){var N=L%m.length,C=L%g.length,R=n(m[N*2],m[N*2+1],g[C*2],g[C*2+1]);h[L]=R.real,d[L]=R.imag}else for(var D=function(k){var W=T.util.indexToLoc(k,u,c),F=W.slice(-v);p.forEach(function(K){return F[K]=0});var P=T.util.locToIndex(F,v,b),H=W.slice(-w);f.forEach(function(K){return H[K]=0});var _=T.util.locToIndex(H,w,S),j=n(m[P*2],m[P*2+1],g[_*2],g[_*2+1]);h[k]=j.real,d[k]=j.imag},L=0;L<h.length;L++)D(L);return[h,d,o]}}var E0=wr(function(n,t){return n+t}),mz=Zd(function(n,t,e,i){return{real:n+e,imag:t+i}}),D0=ha(T.Add,E0,mz),gz={kernelName:T.Add,backendName:"cpu",kernelFunc:D0};function da(n){return function(t,e,i){for(var r=T.util.getTypedArrayFromDType(e,t.length),a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}function $e(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");for(var l=s,u=l.data.get(o.dataId).values,c=T.util.sizeFromShape(o.shape),h=e||o.dtype,d=T.util.getArrayFromDType(h,c),p=0;p<c;++p)d[p]=t(u[p],a);return l.makeTensorInfo(o.shape,h,d)}}function pa(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");var l=s,u=l.data.get(o.dataId).values,c=e||o.dtype,h=t(u,c,a);return l.makeTensorInfo(o.shape,c,h)}}var k0=da(function(n){return Math.ceil(n)}),vz=pa(T.Ceil,k0),yz={kernelName:T.Ceil,backendName:"cpu",kernelFunc:vz};var F0=da(function(n){return Math.exp(n)}),bz=pa(T.Exp,F0),wz={kernelName:T.Exp,backendName:"cpu",kernelFunc:bz};var W0=da(function(n){return Math.expm1(n)}),Sz=pa(T.Expm1,W0),Lz={kernelName:T.Expm1,backendName:"cpu",kernelFunc:Sz};var U0=da(function(n){return Math.floor(n)}),Iz=pa(T.Floor,U0),Az={kernelName:T.Floor,backendName:"cpu",kernelFunc:Iz};var B0=da(function(n){return Math.log(n)}),Tz=pa(T.Log,B0),Nz={kernelName:T.Log,backendName:"cpu",kernelFunc:Tz};function z0(n,t,e,i){for(var r=T.util.getTypedArrayFromDType(i,T.util.sizeFromShape(e)),a=0;a<r.length;++a){for(var s=a*t,o=n[s],l=0;l<t;++l){var u=n[s+l];u>o&&(o=u)}r[a]=o}return r}var P0=wr(function(n,t){return n*t}),xz=Zd(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),_0=ha(T.Multiply,P0,xz),Cz={kernelName:T.Multiply,backendName:"cpu",kernelFunc:_0};var M0=da(function(n){return 1/Math.sqrt(n)}),Rz=pa(T.Rsqrt,M0),Oz={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:Rz};function H0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;c<s;++c){var h=e.length,d=T.util.computeStrides(e),p=T.util.indexToLoc(c,h,d),f=p.map(function(g,v){return g+t[v]}),m=T.util.locToIndex(f,i.length,o);u[c]=n[m]}return u}function Qd(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.begin,s=i.size;ae(r,"slice");var o=T.slice_util.parseSliceParams(r,a,s),l=o[0],u=o[1];T.slice_util.assertParamsValid(r,l,u);var c=e.data.get(r.dataId).values,h=H0(c,l,u,r.shape,r.dtype);return e.makeTensorInfo(u,r.dtype,h)}var Ez={kernelName:T.Slice,backendName:"cpu",kernelFunc:Qd};var V0=wr(function(n,t){return n-t}),Dz=Zd(function(n,t,e,i){return{real:n-e,imag:t-i}}),q0=ha(T.Sub,V0,Dz),kz={kernelName:T.Sub,backendName:"cpu",kernelFunc:q0};function ep(n,t,e,i,r){for(var a=t.length,s=T.util.sizeFromShape(t),o=T.util.computeStrides(t),l=T.util.computeStrides(r),u=T.util.getTypedArrayFromDType(e,T.util.sizeFromShape(r)),c=0;c<s;++c){for(var h=T.util.indexToLoc(c,a,o),d=new Array(h.length),p=0;p<d.length;p++)d[p]=h[i[p]];var f=T.util.locToIndex(d,a,l);u[f]=n[c]}return u}function G0(n,t,e,i){for(var r=T.util.parseAxisParam(t,e)[0],a=[1,e[0],1],s=0;s<r;s++)a[0]*=e[s];a[1]=e[r];for(var s=r+1;s<e.length;s++)a[2]*=e[s];for(var o={},l=new Int32Array(e[r]),u=new T.TensorBuffer(a,i,n),c=[],h=a[0]===1&&a[2]===1,s=0;s<e[r];s++){var d=void 0;if(h)d=n[s].toString();else{for(var p=[],f=0;f<a[0];f++)for(var m=0;m<a[2];m++)p.push(u.get(f,s,m));d=p.join(",")}if(o[d]!==void 0)l[s]=o[d];else{var g=Object.keys(o).length;o[d]=g,l[s]=g,c.push(s)}}var v=a.slice();v[1]=Object.keys(o).length;var b=new T.TensorBuffer(v,i);c.forEach(function(S,L){for(var N=0;N<a[0];N++)for(var C=0;C<a[2];C++)b.set(u.get(N,S,C),N,L,C)});var w=e.slice();return w[r]=v[1],{outputValues:b.values,outputShape:w,indices:l}}var Fz={__proto__:null,simpleAbsImpl:O0,addImpl:E0,ceilImpl:k0,expImpl:F0,expm1Impl:W0,floorImpl:U0,logImpl:B0,maxImpl:z0,multiplyImpl:P0,rsqrtImpl:M0,sliceImpl:H0,subImpl:V0,transposeImpl:ep,uniqueImpl:G0};var Wz="2.6.0";T.registerBackend("cpu",function(){return new R0},1);var Uz=$e(T.Acos,function(n){return Math.acos(n)}),Bz={kernelName:T.Acos,backendName:"cpu",kernelFunc:Uz};var zz=$e(T.Acosh,function(n){return Math.acosh(n)}),Pz={kernelName:T.Acosh,backendName:"cpu",kernelFunc:zz};var _z=$e(T.Asin,function(n){return Math.asin(n)}),Mz={kernelName:T.Asin,backendName:"cpu",kernelFunc:_z};var Hz=$e(T.Asinh,function(n){return Math.asinh(n)}),Vz={kernelName:T.Asinh,backendName:"cpu",kernelFunc:Hz};var qz=$e(T.Atan,function(n){return Math.atan(n)}),Gz={kernelName:T.Atan,backendName:"cpu",kernelFunc:qz};var Yz=$e(T.Atanh,function(n){return Math.atanh(n)}),jz={kernelName:T.Atanh,backendName:"cpu",kernelFunc:Yz};function tp(n,t,e,i,r,a){for(var s=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,u=r.dilationWidth,c=r.effectiveFilterHeight,h=r.effectiveFilterWidth,d=r.padInfo.top,p=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=T.buffer(r.outShape,e),g=m.values,v=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],w=r.outShape[3],S=0;S<r.batchSize;++S)for(var L=S*v,N=S*i[0],C=0;C<r.inChannels;++C)for(var R=0;R<r.outHeight;++R)for(var D=R*s-d,k=Math.max(0,D),W=Math.min(r.inHeight,c+D),F=L+R*b,P=0;P<r.outWidth;++P){for(var H=P*o-p,_=Math.max(0,H),j=Math.min(r.inWidth,h+H),K=f,q=0,G=0,Z=k;Z<W;Z+=l){for(var X=N+Z*i[1],ee=_;ee<j;ee+=u){var ne=X+ee*i[2],ie=n[ne+C];a==="max"&&ie>K?K=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(K))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:K}return m}function Y0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;g<i.batchSize;++g)for(var v=0;v<i.inChannels;++v)for(var b=0;b<i.outHeight;++b){for(var w=b*o-p,S=w;S<0;)S+=u;for(var L=Math.min(i.inHeight,h+w),N=0;N<i.outWidth;++N){for(var C=N*l-f,R=C;R<0;)R+=c;for(var D=Math.min(i.inWidth,d+C),k=Number.NEGATIVE_INFINITY,W=-1,F=S;F<L;F+=u)for(var P=F-w,H=R;H<D;H+=c){var _=H-C,j=m.get(g,F,H,v);j>k&&(k=j,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,N,v)}}return s}function Kz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=ca({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=tp(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var $z={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:Kz};function Xz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),N=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D<c.batchSize;++D)for(var k=0;k<c.inChannels;++k)for(var W=0;W<c.inHeight;++W)for(var F=0;F<c.inWidth;++F){for(var P=W-S,H=F-w,_=0,j=0;j<v;j+=m){var K=(P+j)/h;if(K<0||K>=c.outHeight||Math.floor(K)!==K)continue;for(var q=0;q<b;q+=g){var G=(H+q)/d;if(G<0||G>=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,K,G,k);_+=Z}}L.set(_*N,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var Jz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:Xz};function Zz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,N=0,C=0,R=0;R<c.length;++R)m[R]=f[S++]+(c[R]-h[L++])*p[N++]/Math.sqrt(d[C++]+u),S>=g&&(S=0),L>=w&&(L=0),N>=v&&(N=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var Qz={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:Zz};var eP=$e(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n<e.clipValueMin?e.clipValueMin:n}),tP={kernelName:T.ClipByValue,backendName:"cpu",kernelFunc:eP};function Zo(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.imag,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var nP={kernelName:T.Imag,backendName:"cpu",kernelFunc:Zo};function Wi(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=T.util.sizeFromShape(r.shape),o=T.util.inferFromImplicitShape(a,s),l=T.util.sizeFromShape(o);T.util.assert(s===l,function(){return"The new shape ("+o+") has "+l+" elements and the old "+("shape ("+r.shape+") has "+s+" elements. The new shape and old ")+"shape must have the same number of elements."}),e.incRef(r.dataId);var u=e.data.get(r.dataId);if(u.complexTensorInfos!=null){var c=u.complexTensorInfos.real,h=u.complexTensorInfos.imag;c.shape=o,h.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var iP={kernelName:T.Reshape,backendName:"cpu",kernelFunc:Wi};function gs(n){var t=n.inputs,e=n.backend,i=n.attrs,r=i.axis,a=T.util.parseAxisParam(r,t[0].shape)[0],s=T.backend_util.computeOutShape(t.map(function(S){return S.shape}),a);if(T.util.sizeFromShape(s)===0)return e.makeTensorInfo(s,t[0].dtype,[]);var o=t.filter(function(S){return T.util.sizeFromShape(S.shape)>0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return fs({inputs:{input:S},backend:e})}),c=o.map(function(S){return Zo({inputs:{input:S},backend:e})}),h=gs({inputs:u,backend:e,attrs:{axis:r}}),d=gs({inputs:c,backend:e,attrs:{axis:r}}),p=Cn({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),N=[-1,L];return Wi({inputs:{x:S},backend:e,attrs:{shape:N}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,N=T.util.sizeFromShape(S.shape);m.set(L,g),g+=N})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,N=0,C=0;C<S.shape[0];++C)for(var R=C*s[1]+v,D=0;D<S.shape[1];++D)m[R+D]=L[N++];v+=S.shape[1]})}var b=T.backend_util.computeOutShape(o.map(function(S){return S.shape}),a),w=e.makeTensorInfo(b,t[0].dtype,m);return f.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),w}var rP={kernelName:T.Concat,backendName:"cpu",kernelFunc:gs};var aP=$e(T.Cos,function(n){return Math.cos(n)}),sP={kernelName:T.Cos,backendName:"cpu",kernelFunc:aP};var oP=$e(T.Cosh,function(n){return Math.cosh(n)}),lP={kernelName:T.Cosh,backendName:"cpu",kernelFunc:oP};var uP={kernelName:T.Dilation2D,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=i,l=o.strides,u=o.pad,c=o.dilations,h=e,d=h.data.get(a.dataId).values,p=a.shape.length,f=h.data.get(s.dataId).values,m=s.shape.length,g=T.backend_util.computeDilation2DInfo(a.shape,s.shape,l,u,"NHWC",c),v=g.batchSize,b=g.inHeight,w=g.inWidth,S=g.inChannels,L=g.outHeight,N=g.outWidth,C=g.padInfo,R=g.strideHeight,D=g.strideWidth,k=g.filterHeight,W=g.filterWidth,F=g.dilationHeight,P=g.dilationWidth,H=g.outShape,_=T.util.sizeFromShape(H),j=H.length,K=T.util.getArrayFromDType(a.dtype,_),q=0;q<v;++q)for(var G=0;G<L;++G)for(var Z=G*R-C.top,X=0;X<N;++X)for(var ee=X*D-C.left,ne=0;ne<S;++ne){for(var ie=Number.MIN_SAFE_INTEGER,te=0;te<k;++te){var re=Z+te*F;if(re>=0&&re<b)for(var le=0;le<W;++le){var he=ee+le*P;if(he>=0&&he<w){var ye=T.util.locToIndex([q,re,he,ne],p,T.util.computeStrides(a.shape)),Oe=T.util.locToIndex([te,le,ne],m,T.util.computeStrides(s.shape)),Fe=d[ye]+f[Oe];Fe>ie&&(ie=Fe)}}}var _e=T.util.locToIndex([q,G,X,ne],j,T.util.computeStrides(H));K[_e]=ie}var Me=h.write(T.util.toTypedArray(K,a.dtype),H,a.dtype);return{dataId:Me,shape:H,dtype:a.dtype}}};var cP={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,N=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),j=0;j<g;++j)for(var K=0;K<S;++K)for(var q=K*C-N.top,G=0;G<L;++G)for(var Z=G*R-N.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=0,ie=0,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[j][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[j][K][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var hP={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,N=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),j=0;j<g;++j)for(var K=0;K<S;++K)for(var q=K*C-N.top,G=0;G<L;++G)for(var Z=G*R-N.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=q<0?0:q,ie=Z<0?0:Z,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[j][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=re,ie=he)}}}_[j][ne][ie][X]+=H[j][K][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var dP=wr(function(n,t){return n/t}),pP=ha(T.Div,dP),np={kernelName:T.Div,backendName:"cpu",kernelFunc:pP};var fP=$e(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),mP={kernelName:T.Elu,backendName:"cpu",kernelFunc:fP};var gP=T.backend_util.ERF_P,vP=T.backend_util.ERF_A1,yP=T.backend_util.ERF_A2,bP=T.backend_util.ERF_A3,wP=T.backend_util.ERF_A4,SP=T.backend_util.ERF_A5,LP=$e(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+gP*e);return t*(1-((((SP*i+wP)*i+bP)*i+yP)*i+vP)*i*Math.exp(-e*e))}),IP={kernelName:T.Erf,backendName:"cpu",kernelFunc:LP};function j0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p<r;p++){for(var f=Qd({inputs:{x:o},backend:e,attrs:{begin:[p,0],size:[1,a]}}),m=Qd({inputs:{x:l},backend:e,attrs:{begin:[p,0],size:[1,a]}}),g=Cn({inputs:{real:f,imag:m},backend:e}),v=AP(g,t,e),b=v.real,w=v.imag,S=T.backend_util.mergeRealAndImagArrays(b,w),L=0;L<a;L++){var N=T.backend_util.getComplexWithIndex(S,L);h[p*a+L]=N.real,d[p*a+L]=N.imag}e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(g)}var C=e.makeTensorInfo(u,"float32",h),R=e.makeTensorInfo(u,"float32",d),D=Cn({inputs:{real:C,imag:R},backend:e});return e.disposeIntermediateTensorInfo(C),e.disposeIntermediateTensorInfo(R),D}function AP(n,t,e){var i=T.util.sizeFromShape(n.shape),r=e.data.get(n.dataId),a=e.data.get(r.complexTensorInfos.real.dataId).values,s=e.data.get(r.complexTensorInfos.imag.dataId).values;if(TP(i)){var o=ip(a,s,i,t,e),l=[n.shape[0],n.shape[1]];if(t){var u=e.makeTensorInfo(l,"float32",o.real),c=e.makeTensorInfo(l,"float32",o.imag),h=e.makeTensorInfo([],"float32",T.util.createScalarValue(i,"float32")),d=ca({inputs:{x:h},backend:e}),p=np.kernelFunc({inputs:{a:u,b:h},backend:e}),f=np.kernelFunc({inputs:{a:c,b:d},backend:e}),m=e.data.get(p.dataId).values,g=e.data.get(f.dataId).values;return e.disposeIntermediateTensorInfo(u),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return o}else{var v=T.backend_util.mergeRealAndImagArrays(a,s),b=NP(v,i,t);return T.backend_util.splitRealAndImagArrays(b)}}function TP(n){return(n&n-1)===0}function ip(n,t,e,i,r){if(e===1)return{real:n,imag:t};var a=T.backend_util.mergeRealAndImagArrays(n,t),s=e/2,o=T.backend_util.complexWithEvenIndex(a),l=o.real,u=o.imag,c=[l.length],h=r.makeTensorInfo(c,"float32",l),d=r.makeTensorInfo(c,"float32",u),p=Cn({inputs:{real:h,imag:d},backend:r}),f=T.backend_util.complexWithOddIndex(a),m=f.real,g=f.imag,v=[m.length],b=r.makeTensorInfo(v,"float32",m),w=r.makeTensorInfo(v,"float32",g),S=Cn({inputs:{real:b,imag:w},backend:r}),L=ip(l,u,s,i,r),N=L.real,C=L.imag,R=[N.length],D=r.makeTensorInfo(R,"float32",N),k=r.makeTensorInfo(R,"float32",C),W=Cn({inputs:{real:D,imag:k},backend:r}),F=ip(m,g,s,i,r),P=F.real,H=F.imag,_=[P.length],j=r.makeTensorInfo(_,"float32",P),K=r.makeTensorInfo(_,"float32",H),q=Cn({inputs:{real:j,imag:K},backend:r}),G=T.backend_util.exponents(e,i),Z=[G.real.length],X=r.makeTensorInfo(Z,"float32",G.real),ee=r.makeTensorInfo(Z,"float32",G.imag),ne=Cn({inputs:{real:X,imag:ee},backend:r}),ie=_0({inputs:{a:ne,b:q},backend:r}),te=D0({inputs:{a:W,b:ie},backend:r}),re=q0({inputs:{a:W,b:ie},backend:r}),le=fs({inputs:{input:te},backend:r}),he=fs({inputs:{input:re},backend:r}),ye=Zo({inputs:{input:te},backend:r}),Oe=Zo({inputs:{input:re},backend:r}),Fe=gs({inputs:[le,he],backend:r,attrs:{axis:0}}),_e=gs({inputs:[ye,Oe],backend:r,attrs:{axis:0}}),Me=r.data.get(Fe.dataId).values,dt=r.data.get(_e.dataId).values;return r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(w),r.disposeIntermediateTensorInfo(S),r.disposeIntermediateTensorInfo(D),r.disposeIntermediateTensorInfo(k),r.disposeIntermediateTensorInfo(W),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(q),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(ie),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(re),r.disposeIntermediateTensorInfo(le),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(he),r.disposeIntermediateTensorInfo(Oe),r.disposeIntermediateTensorInfo(Fe),r.disposeIntermediateTensorInfo(_e),{real:Me,imag:dt}}function NP(n,t,e){for(var i=new Float32Array(t*2),r=0;r<t;r++){for(var a=0,s=0,o=0;o<t;o++){var l=T.backend_util.exponent(r*o,t,e),u=T.backend_util.getComplexWithIndex(n,o);a+=u.real*l.real-u.imag*l.imag,s+=u.real*l.imag+u.imag*l.real}e&&(a/=t,s/=t),T.backend_util.assignToTypedArray(i,a,s,r)}return i}function xP(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Wi({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=j0(o,!1,e),u=Wi({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var CP={kernelName:T.FFT,backendName:"cpu",kernelFunc:xP};var RP={kernelName:T.FlipLeftRight,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=i,s=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),o=r.shape,l=o[0],u=o[1],c=o[2],h=o[3],d=a.data.get(r.dataId).values,p=0;p<l;p++)for(var f=p*c*u*h,m=0;m<u;m++)for(var g=m*(c*h),v=0;v<c;v++)for(var b=v*h,w=0;w<h;w++){var S=[l,m,v,w],L=S[2],N=Math.round(c-L),C=f+g+b+w,R=d[C];if(N>=0&&N<c){var D=N*h,k=f+g+D+w;R=d[k]}s[C]=R}var W=a.write(s,r.shape,r.dtype);return{dataId:W,shape:r.shape,dtype:r.dtype}}};function OP(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Wi({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=j0(o,!0,e),u=Wi({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var EP={kernelName:T.IFFT,backendName:"cpu",kernelFunc:OP};var DP=$e(T.IsFinite,function(n){return Number.isFinite(n)?1:0},"bool"),kP={kernelName:T.IsFinite,backendName:"cpu",kernelFunc:DP};var FP=$e(T.IsInf,function(n){return Math.abs(n)===Infinity?1:0},"bool"),WP={kernelName:T.IsInf,backendName:"cpu",kernelFunc:FP};var UP=$e(T.IsNan,function(n){return Number.isNaN(n)?1:0},"bool"),BP={kernelName:T.IsNan,backendName:"cpu",kernelFunc:UP};var zP=$e(T.Log1p,function(n){return Math.log1p(n)}),PP={kernelName:T.Log1p,backendName:"cpu",kernelFunc:zP};var _P=$e(T.LogicalNot,function(n){return n?0:1},"bool"),MP={kernelName:T.LogicalNot,backendName:"cpu",kernelFunc:_P};var HP={kernelName:T.Max,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape,c=u.length,h=T.util.parseAxisParam(s,u),d=h,p=T.backend_util.getAxesPermutation(d,c),f=l.data.get(r.dataId).values;if(p!=null){for(var m=new Array(c),g=0;g<m.length;g++)m[g]=u[p[g]];f=ep(f,u,r.dtype,p,m),d=T.backend_util.getInnerMostAxes(d.length,c),u=m}ae(r,"max"),T.backend_util.assertAxesAreInnerMostDims("max",d,c);var v=T.backend_util.computeOutAndReduceShapes(u,d),b=v[0],w=v[1],S=T.util.sizeFromShape(w),L=z0(f,S,b,r.dtype),N=l.write(L,b,r.dtype),C=b;if(o){var m=T.backend_util.expandShapeToKeepDim(b,h);C=m}return{dataId:N,shape:C,dtype:r.dtype}}};function VP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=ca({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=tp(d,r.shape,r.dtype,p,c,"max");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var qP={kernelName:T.MaxPool,backendName:"cpu",kernelFunc:VP};function GP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ae([a,s],"maxPoolBackprop");for(var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=T.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=e.data.get(o.dataId).values,f=T.buffer(d.outShape,o.dtype,Y0(p,o.shape,o.dtype,d).values),m=d.strideHeight,g=d.strideWidth,v=d.dilationHeight,b=d.dilationWidth,w=d.effectiveFilterHeight,S=d.effectiveFilterWidth,L=S-1-d.padInfo.left,N=w-1-d.padInfo.top,C=T.buffer(o.shape,"float32"),R=e.data.get(r.dataId).values,D=T.buffer(r.shape,"float32",R),k=0;k<d.batchSize;++k)for(var W=0;W<d.inChannels;++W)for(var F=0;F<d.inHeight;++F)for(var P=0;P<d.inWidth;++P){for(var H=F-N,_=P-L,j=0,K=0;K<w;K+=v){var q=(H+K)/m;if(q<0||q>=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<S;G+=b){var Z=(_+G)/g;if(Z<0||Z>=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=K*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);j+=ie*ne}}C.set(j,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var YP={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:GP};function jP(n,t,e,i,r){var a=T.util.computeStrides(t),s=tp(n,t,e,a,r,"max"),o=Y0(n,t,e,r,!0,i);return[s.values,o.values]}var KP={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=jP(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var $P=T.kernel_impls.nonMaxSuppressionV4Impl,XP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=$P(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var JP=T.kernel_impls.nonMaxSuppressionV5Impl,ZP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=JP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var QP=wr(function(n,t){return n!==t?1:0}),e_=ha(T.NotEqual,QP,null,"bool"),t_={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:e_};function n_(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(N,C){return N[0]+r.shape[C]+N[1]}),l=a.map(function(N){return N[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v<c;v++){var b=T.util.indexToLoc(v,h,d),w=b.map(function(N,C){return N+l[C]}),S=T.util.locToIndex(w,f,m);g[S]=u[v]}var L=e.write(g,o,r.dtype);return{dataId:L,shape:o,dtype:r.dtype}}var K0={kernelName:T.PadV2,backendName:"cpu",kernelFunc:n_};var i_=$e(T.Reciprocal,function(n){return 1/n}),r_={kernelName:T.Reciprocal,backendName:"cpu",kernelFunc:i_};var a_={kernelName:T.RotateWithOffset,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),h=r.shape,d=h[0],p=h[1],f=h[2],m=h[3],g=T.backend_util.getImageCenter(l,p,f),v=g[0],b=g[1],w=255,S=Math.sin(s),L=Math.cos(s),N=u.data.get(r.dataId).values,C=0;C<d;C++)for(var R=C*f*p*m,D=0;D<p;D++)for(var k=D*(f*m),W=0;W<f;W++)for(var F=W*m,P=0;P<m;P++){var H=[d,D,W,P],_=H[2],j=H[1],K=(_-v)*L-(j-b)*S,q=(_-v)*S+(j-b)*L;K=Math.round(K+v),q=Math.round(q+b);var G=o;if(typeof o!="number"&&(P===3?G=w:G=o[P]),K>=0&&K<f&&q>=0&&q<p){var Z=q*(f*m),X=K*m,ee=R+Z+X+P;G=N[ee]}var ne=R+k+F+P;c[ne]=G}var ie=u.write(c,r.shape,r.dtype);return{dataId:ie,shape:r.shape,dtype:r.dtype}}};var s_=$e(T.Round,function(n){var t=Math.floor(n);return n-t<.5?Math.floor(n):n-t>.5?Math.ceil(n):t%2===0?t:t+1}),o_={kernelName:T.Round,backendName:"cpu",kernelFunc:s_};var l_=T.backend_util.SELU_SCALEALPHA,u_=T.backend_util.SELU_SCALE,c_=$e(T.Selu,function(n){return n>=0?u_*n:l_*(Math.exp(n)-1)}),h_={kernelName:T.Selu,backendName:"cpu",kernelFunc:c_};var d_=$e(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),p_={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:d_};var f_=$e(T.Sign,function(n){return n<0?-1:n>0?1:0}),m_={kernelName:T.Sign,backendName:"cpu",kernelFunc:f_};var g_=$e(T.Sin,function(n){return Math.sin(n)}),v_={kernelName:T.Sin,backendName:"cpu",kernelFunc:g_};var y_=$e(T.Sinh,function(n){return Math.sinh(n)}),b_={kernelName:T.Sinh,backendName:"cpu",kernelFunc:y_};var w_=11920928955078125e-23,$0=Math.log(w_)+2,S_=$e(T.Softplus,function(n){var t=n>-$0,e=n<$0,i=Math.exp(n),r;return e?r=i:t?r=n:r=Math.log(1+i),r}),L_={kernelName:T.Softplus,backendName:"cpu",kernelFunc:S_};function X0(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm;ae(r,"transpose");for(var s=r.shape.length,o=new Array(s),l=0;l<o.length;l++)o[l]=r.shape[a[l]];var u=i.data.get(r.dataId).values,c=ep(u,r.shape,r.dtype,a,o),h=i.write(c,o,r.dtype);return{dataId:h,shape:o,dtype:r.dtype}}var I_={kernelName:T.Transpose,backendName:"cpu",kernelFunc:X0};function A_(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.blockShape,s=i.paddings;ae([r],"spaceToBatchND");var o=T.util.sizeFromShape(a),l=[[0,0]];l.push.apply(l,s);for(var u=1+a.length;u<r.shape.length;++u)l.push([0,0]);var c=K0.kernelFunc({inputs:{x:r},backend:e,attrs:{paddings:l,constantValue:0}}),h=T.backend_util.getReshaped(c.shape,a,o,!1),d=T.backend_util.getPermuted(h.length,a.length,!1),p=T.backend_util.getReshapedPermuted(c.shape,a,o,!1),f={x:c},m={shape:h},g=Wi({inputs:f,backend:e,attrs:m}),v={x:g},b={perm:d},w=X0({inputs:v,backend:e,attrs:b}),S={x:w},L={shape:p},N=Wi({inputs:S,backend:e,attrs:L});return e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(w),N}var T_={kernelName:T.SpaceToBatchND,backendName:"cpu",kernelFunc:A_};var N_=$e(T.Sqrt,function(n){return Math.sqrt(n)}),x_={kernelName:T.Sqrt,backendName:"cpu",kernelFunc:N_};var C_={kernelName:T.Square,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.x,r=e;ae(i,"square");for(var a=r.data.get(i.dataId).values,s=new Float32Array(a.length),o=0;o<a.length;++o){var l=a[o];s[o]=l*l}var u=r.write(s,i.shape,i.dtype);return{dataId:u,shape:i.shape,dtype:i.dtype}}};var R_=wr(function(n,t){var e=n-t;return e*e}),O_=ha(T.SquaredDifference,R_),E_={kernelName:T.SquaredDifference,backendName:"cpu",kernelFunc:O_};var D_=$e(T.Step,function(n,t){var e=t;return isNaN(n)?NaN:n>0?1:e.alpha}),k_={kernelName:T.Step,backendName:"cpu",kernelFunc:D_};var F_=$e(T.Tan,function(n){return Math.tan(n)}),W_={kernelName:T.Tan,backendName:"cpu",kernelFunc:F_};var U_=$e(T.Tanh,function(n){return Math.tanh(n)}),B_={kernelName:T.Tanh,backendName:"cpu",kernelFunc:U_};function z_(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=G0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var P_={kernelName:T.Unique,backendName:"cpu",kernelFunc:z_};var __=[cz,Bz,Pz,gz,Mz,Vz,Gz,jz,$z,Jz,Qz,fz,yz,tP,hz,rP,sP,lP,uP,hP,cP,np,mP,IP,wz,Lz,CP,RP,Az,dz,EP,nP,kP,WP,BP,Nz,PP,MP,qP,YP,KP,HP,Cz,XP,ZP,t_,K0,pz,r_,iP,a_,o_,Oz,h_,p_,m_,v_,b_,Ez,L_,T_,x_,C_,E_,k_,kz,W_,B_,I_,P_];for(var rp=0,J0=__;rp<J0.length;rp++){var M_=J0[rp];T.registerKernel(M_)}ps.MathBackendCPU=R0;ps.shared=Fz;ps.version_cpu=Wz});var FL=Le(qn=>{"use strict";Object.defineProperty(qn,"__esModule",{value:!0});var x=lr();var ap=function(n,t){return ap=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},ap(n,t)};function H_(n,t){ap(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Qo(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function el(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Sr={},sp={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function Q0(n,t){Sr[n]=t}function Gn(n){if(!(n in Sr)){var t=V_(n);if(t!==null)Sr[n]=t;else return console.log("Could not get context for WebGL version",n),null}var e=Sr[n];return e.isContextLost()?(delete Sr[n],Gn(n)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),Sr[n])}function q_(n){if(typeof OffscreenCanvas!="undefined"&&n===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function V_(n){if(n!==1&&n!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");var t=q_(n);return t.addEventListener("webglcontextlost",function(e){e.preventDefault(),delete Sr[n]},!1),n===1?t.getContext("webgl",sp)||t.getContext("experimental-webgl",sp):t.getContext("webgl2",sp)}var vs;(function(n){n[n.DENSE=0]="DENSE",n[n.SHARED_BATCH=1]="SHARED_BATCH"})(vs||(vs={}));var on;(function(n){n[n.RENDER=0]="RENDER",n[n.UPLOAD=1]="UPLOAD",n[n.PIXELS=2]="PIXELS",n[n.DOWNLOAD=3]="DOWNLOAD"})(on||(on={}));var Ot;(function(n){n[n.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",n[n.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",n[n.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",n[n.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",n[n.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Ot||(Ot={}));function ys(n,t){return[t,n]}function G_(n,t){return n*t}function bs(n){var t=x.util.sizeFromShape(n),e=Math.ceil(t/4);return x.util.sizeToSquarishShape(e)}function fa(n,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(n/2))]}function Y_(n,t){var e=fa(n,t),i=e[0],r=e[1];return i*r*4}function op(n,t){var e=n,i,r,a,s,o,l,u,c,h,d;return x.env().getNumber("WEBGL_VERSION")===2?(i=e.R32F,r=e.R16F,a=e.RGBA16F,s=e.RGBA32F,o=e.RED,u=4,c=1,h=e.HALF_FLOAT,d=e.FLOAT):(i=n.RGBA,r=n.RGBA,a=n.RGBA,s=e.RGBA,o=n.RGBA,u=4,c=4,h=t!=null?t.HALF_FLOAT_OES:null,d=n.FLOAT),l=n.RGBA,{internalFormatFloat:i,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:s,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:u,defaultNumChannels:c,textureTypeHalfFloat:h,textureTypeFloat:d}}function ce(n,t){var e=t();return x.env().getBool("DEBUG")&&j_(n),e}function j_(n){var t=n.getError();if(t!==n.NO_ERROR)throw new Error("WebGL Error: "+eS(n,t))}var K_=596e-10,$_=65504;function tS(n){return!!(x.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||n===0||K_<Math.abs(n)&&Math.abs(n)<$_)}function eS(n,t){switch(t){case n.NO_ERROR:return"NO_ERROR";case n.INVALID_ENUM:return"INVALID_ENUM";case n.INVALID_VALUE:return"INVALID_VALUE";case n.INVALID_OPERATION:return"INVALID_OPERATION";case n.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case n.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case n.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return"Unknown error code "+t}}function ws(n,t){return li(n,function(){return n.getExtension(t)},'Extension "'+t+'" not supported on this browser.')}function nS(n,t){var e=li(n,function(){return n.createShader(n.VERTEX_SHADER)},"Unable to create vertex WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw console.log(n.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function iS(n,t){var e=li(n,function(){return n.createShader(n.FRAGMENT_SHADER)},"Unable to create fragment WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw X_(t,n.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var J_=/ERROR: [0-9]+:([0-9]+):/g;function X_(n,t){var e=J_.exec(t);if(e==null){console.log("Couldn't parse line number in error: "+t),console.log(n);return}for(var i=+e[1],r=n.split(`
+============================`));var a={};return this.data.set(a,{values:e,dtype:r,refCount:1}),a},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return{dataId:a,shape:e,dtype:i}},t.prototype.incRef=function(e){var i=this.data.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.data.has(e)){var i=this.data.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){this.data.set(e,{values:i,dtype:a,refCount:1})},t.prototype.numDataIds=function(){return this.data.numDataIds()},t.prototype.read=function(e){return x0(this,void 0,void 0,function(){return C0(this,function(i){return[2,this.readSync(e)]})})},t.prototype.readSync=function(e){var i=this.data.get(e),r=i.dtype,a=i.complexTensorInfos;if(r==="complex64"){var s=this.readSync(a.real.dataId),o=this.readSync(a.imag.dataId);return T.backend_util.mergeRealAndImagArrays(s,o)}return this.data.get(e).values},t.prototype.bufferSync=function(e){var i=this.readSync(e.dataId),r=i;if(e.dtype==="string")try{r=i.map(function(a){return T.util.decodeString(a)})}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return T.buffer(e.shape,e.dtype,r)},t.prototype.makeOutput=function(e,i,r){var a=this.write(e,i,r);return T.engine().makeTensorFromDataId(a,i,r,this)},t.prototype.disposeData=function(e){if(this.data.has(e)){var i=this.data.get(e).complexTensorInfos;i!=null&&(this.disposeData(i.real.dataId),this.disposeData(i.imag.dataId)),this.data.delete(e)}},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.data.has(i)){var r=this.data.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.time=function(e){return x0(this,void 0,void 0,function(){var i,r;return C0(this,function(a){return i=T.util.now(),e(),r=T.util.now()-i,[2,{kernelMs:r}]})})},t.prototype.memory=function(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}},t.prototype.stridedSlice=function(e,i,r,a){ae(e,"stridedSlice");var s=T.slice_util.computeOutShape(i,r,a);if(s.some(function(p){return p===0}))return T.tensor([],s);for(var o=T.buffer(s,e.dtype),l=this.bufferSync(e),u=0;u<o.size;u++){for(var c=o.indexToLoc(u),h=new Array(c.length),d=0;d<h.length;d++)h[d]=c[d]*a[d]+i[d];o.set.apply(o,[l.get.apply(l,h)].concat(c))}return o.toTensor()},t.prototype.diag=function(e){for(var i=this.readSync(e.dataId),r=T.buffer([e.size,e.size],e.dtype),a=r.values,s=0;s<i.length;s++)a[s*e.size+s]=i[s];return r.toTensor()},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=T.slice(e,l,u).reshape(a);return c},t.prototype.reverse=function(e,i){ae(e,"reverse");for(var r=T.buffer(e.shape,e.dtype),a=this.bufferSync(e),s=function(l){var u=r.indexToLoc(l),c=u.slice();i.forEach(function(h){return c[h]=e.shape[h]-1-c[h]}),r.set.apply(r,[a.get.apply(a,c)].concat(u))},o=0;o<r.size;o++)s(o);return r.toTensor()},t.prototype.neg=function(e){return ae(e,"neg"),T.mul(T.scalar(-1),e)},t.prototype.addN=function(e){var i=this;ae(e,"addN");for(var r=e.map(function(c){return i.readSync(c.dataId)}),a=T.buffer(e[0].shape,e[0].dtype),s=a.values,o=0;o<e.length;o++)for(var l=r[o],u=0;u<s.length;u++)s[u]+=l[u];return a.toTensor()},t.prototype.softmax=function(e,i){var r=T.util.parseAxisParam([i],e.shape),a=T.max(e,r),s=T.backend_util.expandShapeToKeepDim(a.shape,r),o=T.sub(e,a.reshape(s)),l=T.exp(o),u=this.sum(l,r).reshape(s);return T.div(l,u)},t.prototype.pow=function(e,i){return ae([e,i],"pow"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.pow(r,a)})},t.prototype.batchMatMul=function(e,i,r,a){ae([e,i],"matMul");for(var s=r?e.shape[1]:e.shape[2],o=r?e.shape[2]:e.shape[1],l=a?i.shape[1]:i.shape[2],u=e.shape[0],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=r?[e.strides[0],1,e.strides[1]]:[e.strides[0],e.strides[1],1],p=d[0],f=d[1],m=d[2],g=a?[1,i.strides[1],i.strides[0]]:[i.strides[1],1,i.strides[0]],v=g[0],b=g[1],w=g[2],S=o*l,L=T.buffer([u,o,l],e.dtype),N=L.values,C=this.blockSize,R=0;R<u;R++)for(var D=0;D<o;D+=C)for(var k=0;k<l;k+=C)for(var W=0;W<s;W+=C)for(var F=Math.min(D+C,o),P=Math.min(k+C,l),H=Math.min(W+C,s),_=D;_<F;_++)for(var j=k;j<P;j++){for(var K=0,q=W;q<H;q++)K+=c[R*p+_*f+q*m]*h[q*v+j*b+R*w];N[R*S+(_*l+j)]+=K}return L.toTensor()},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=this.batchMatMul(i,r,a,s);return o&&(c=T.add(c,o)),l&&(c=Zd(this,c,l,u)),c},t.prototype.floorDiv=function(e,i){ae([e,i],"floorDiv");var r=function(s,o){return Math.floor(s/o)},a="int32";return this.broadcastedBinaryOp(e,i,a,r)},t.prototype.sum=function(e,i){ae(e,"sum"),T.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=0,m=0;m<u;++m)f+=h[p+m];c[d]=f}return l},t.prototype.prod=function(e,i){ae(e,"sum");for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=1,m=0;m<u;++m)f*=h[p+m];c[d]=f}return l},t.prototype.unsortedSegmentSum=function(e,i,r){ae(e,"unsortedSegmentSum");for(var a=[],s=e.rank-i.rank,o=0;o<s;++o)i=i.expandDims(o+1);for(var o=0;o<r;++o){var l=T.scalar(o,"int32"),u=T.equal(l,i).asType("float32"),c=u.mul(e).sum(0);a.push(c)}return T.stack(a)},t.prototype.argMin=function(e,i){ae(e,"argMin");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMin",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v<f&&(f=v,m=g)}c[d]=m}return l},t.prototype.argMax=function(e,i){ae(e,"argMax");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMax",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v>f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;d<u.length;d+=c)for(var p=0;p<c;p++){var f=h(d,p);if(p===0)l[f]=r?0:u[f];else{var m=h(d,p-1);l[f]=r?u[m]+l[m]:u[f]+l[m]}}return o},t.prototype.equal=function(e,i){return ae([e,i],"equal"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r===a?1:0})},t.prototype.notEqual=function(e,i){return ae([e,i],"notEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r!==a?1:0})},t.prototype.less=function(e,i){return ae([e,i],"less"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<a?1:0})},t.prototype.lessEqual=function(e,i){return ae([e,i],"lessEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<=a?1:0})},t.prototype.greater=function(e,i){return ae([e,i],"greater"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>a?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d<a.length;d++)for(var p=0;p<h;p++)a[d]===1?u[c++]=s[d]:u[c++]=o[d];return l},t.prototype.where=function(e){ae([e],"where");var i=this.readSync(e.dataId);return lz(e.shape,i)},t.prototype.topk=function(e,i,r){ae(e,"topk");var a=this.readSync(e.dataId);return oz(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){ae(e,"min"),T.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];m<p&&(p=m)}u[h]=p}return o},t.prototype.minimum=function(e,i){return ae([e,i],"minimum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.min(r,a)})},t.prototype.mod=function(e,i){return ae([e,i],"mod"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r%a;return r<0&&a<0||r>=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p&&m}u[h]=p}return o},t.prototype.any=function(e,i){ae(e,"any"),T.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p||m}u[h]=p}return o},t.prototype.squaredDifference=function(e,i){return ae([e,i],"squaredDifference"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r-a;return s*s})},t.prototype.linear=function(e){return e},t.prototype.relu=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.max(0,a[s]);return i},t.prototype.relu6=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.min(Math.max(0,a[s]),6);return i},t.prototype.prelu=function(e,i){return ae([e,i],"prelu"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return r<0?a*r:r})},t.prototype.eluDer=function(e,i){ae([e,i],"eluDer");for(var r=new Float32Array(i.size),a=this.readSync(i.dataId),s=this.readSync(e.dataId),o=0;o<a.length;++o){var l=a[o];l>=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Zd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),N=this.readSync(i.dataId),C=d.values,R=0;R<r.batchSize;++R)for(var D=R*p,k=R*v,W=0;W<r.outHeight;++W)for(var F=k+W*b,P=W*r.strideHeight-c,H=0;H<a;H++){var _=P+H*o;if(_<0||_>=r.inHeight)continue;for(var j=H*i.strides[0],K=D+_*f,q=0;q<r.outWidth;++q)for(var G=F+q*w,Z=q*r.strideWidth-u,X=0;X<s;X++){var ee=Z+X*l;if(ee<0||ee>=r.inWidth)continue;for(var ne=j+X*i.strides[1],ie=K+ee*m,te=ne,re=0;re<r.inChannels;++re){for(var le=L[ie+re*g],he=0;he<r.outChannels;++he)C[G+he*S]+=le*N[te+he];te+=r.outChannels}}}return d.toTensor()},t.prototype.conv3d=function(e,i,r){for(var a=r.filterDepth,s=r.filterHeight,o=r.filterWidth,l=r.dilationDepth,u=r.dilationHeight,c=r.dilationWidth,h=r.padInfo.front,d=r.padInfo.left,p=r.padInfo.top,f=T.buffer(r.outShape,e.dtype),m=this.readSync(e.dataId),g=this.readSync(i.dataId),v=f.values,b=0;b<r.batchSize;++b)for(var w=b*e.strides[0],S=b*f.strides[0],L=0;L<r.outDepth;++L)for(var N=S+L*f.strides[1],C=L*r.strideDepth-h,R=0;R<a;R++){var D=C+R*l;if(D<0||D>=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F<r.outHeight;++F)for(var P=N+F*f.strides[2],H=F*r.strideHeight-p,_=0;_<s;_++){var j=H+_*u;if(j<0||j>=r.inHeight)continue;for(var K=k+_*i.strides[1],q=W+j*e.strides[2],G=0;G<r.outWidth;++G)for(var Z=P+G*r.outChannels,X=G*r.strideWidth-d,ee=0;ee<o;ee++){var ne=X+ee*c;if(ne<0||ne>=r.inWidth)continue;for(var ie=K+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le<r.inChannels;++le){for(var he=m[te+le],ye=0;ye<r.outChannels;++ye)v[Z+ye]+=he*g[re+ye];re+=r.outChannels}}}}return f.toTensor()},t.prototype.conv2dDerInput=function(e,i,r){ae([e,i],"conv2dDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=this.readSync(e.dataId),l=this.readSync(i.dataId),u=i.strides,c=u[0],h=u[1],d=u[2],p=r.batchSize,f=r.filterHeight,m=r.filterWidth,g=r.inChannels,v=r.inHeight,b=r.inWidth,w=r.outChannels,S=r.outHeight,L=r.outWidth,N=r.strideHeight,C=r.strideWidth,R=r.dataFormat,D=f-1-r.padInfo.top,k=m-1-r.padInfo.left,W=R==="channelsLast",F=a.strides[0],P=W?a.strides[1]:a.strides[2],H=W?a.strides[2]:1,_=W?1:a.strides[1],j=e.strides[0],K=W?e.strides[1]:e.strides[2],q=W?e.strides[2]:1,G=W?1:e.strides[1],Z=0;Z<p;++Z)for(var X=0;X<g;++X)for(var ee=0;ee<v;++ee)for(var ne=ee-D,ie=Math.max(0,Math.ceil(ne/N)),te=Math.min(S,(f+ne)/N),re=0;re<b;++re){for(var le=re-k,he=Math.max(0,Math.ceil(le/C)),ye=Math.min(L,(m+le)/C),Oe=0,Fe=ie;Fe<te;++Fe)for(var _e=Fe*N-ne,Me=he;Me<ye;++Me)for(var dt=Me*C-le,Lt=j*Z+K*Fe+q*Me,Ge=c*(f-1-_e)+h*(m-1-dt)+d*X,Ue=0;Ue<w;++Ue){var me=o[Lt+G*Ue],Ft=l[Ge+Ue];Oe+=me*Ft}var lt=F*Z+P*ee+H*re+_*X;s[lt]=Oe}return a.toTensor()},t.prototype.conv3dDerInput=function(e,i,r){for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=o[3],d=this.readSync(e.dataId),p=e.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],N=w[2],C=w[3],R=r.batchSize,D=r.filterDepth,k=r.filterHeight,W=r.filterWidth,F=r.inChannels,P=r.inDepth,H=r.inHeight,_=r.inWidth,j=r.outChannels,K=r.outDepth,q=r.outHeight,G=r.outWidth,Z=r.strideDepth,X=r.strideHeight,ee=r.strideWidth,ne=D-1-r.padInfo.front,ie=k-1-r.padInfo.top,te=W-1-r.padInfo.left,re=0;re<R;++re)for(var le=0;le<F;++le)for(var he=0;he<P;++he)for(var ye=he-ne,Oe=Math.max(0,Math.ceil(ye/Z)),Fe=Math.min(K,(D+ye)/Z),_e=0;_e<H;++_e)for(var Me=_e-ie,dt=Math.max(0,Math.ceil(Me/X)),Lt=Math.min(q,(k+Me)/X),Ge=0;Ge<_;++Ge){for(var Ue=Ge-te,me=Math.max(0,Math.ceil(Ue/ee)),Ft=Math.min(G,(W+Ue)/ee),lt=0,pt=Oe;pt<Fe;++pt)for(var En=pt*Z-ye,Et=dt;Et<Lt;++Et)for(var It=Et*X-Me,Ce=me;Ce<Ft;++Ce)for(var Wt=Ce*ee-Ue,ft=f*re+m*pt+g*Et+v*Ce,it=S*(D-1-En)+L*(k-1-It)+N*(W-1-Wt)+C*le,un=0;un<j;++un){var Os=d[ft+un],_t=b[it+un];lt+=Os*_t}s[l*re+u*he+c*_e+h*Ge+le]=lt}return a.toTensor()},t.prototype.conv2dDerFilter=function(e,i,r){ae([e,i],"conv2dDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=r.dataFormat==="channelsLast",c=T.buffer(r.filterShape,"float32"),h=r.padInfo.left,d=r.padInfo.top,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((d-m)/a)),v=Math.min(r.outHeight,(r.inHeight+d-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((h-b)/s)),S=Math.min(r.outWidth,(r.inWidth+h-b)/s),L=0;L<r.inChannels;++L)for(var N=0;N<r.outChannels;++N){for(var C=0,R=0;R<r.batchSize;++R)for(var D=g;D<v;++D)for(var k=m+D*a-d,W=w;W<S;++W){var F=b+W*s-h;u?C+=p.get(R,k,F,L)*f.get(R,D,W,N):C+=p.get(R,L,k,F)*f.get(R,N,D,W)}c.set(C,m,b,L,N)}return c.toTensor()},t.prototype.conv3dDerFilter=function(e,i,r){for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=T.buffer(r.filterShape,"float32"),d=h.values,p=h.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],N=w[2],C=w[3],R=this.readSync(e.dataId),D=e.strides,k=D[0],W=D[1],F=D[2],P=D[3],H=r.padInfo.front,_=r.padInfo.left,j=r.padInfo.top,K=0;K<l;++K)for(var q=Math.max(0,Math.ceil((H-K)/a)),G=Math.min(r.outDepth,(r.inDepth+H-K)/a),Z=K*f,X=0;X<u;++X)for(var ee=Math.max(0,Math.ceil((j-X)/s)),ne=Math.min(r.outHeight,(r.inHeight+j-X)/s),ie=X*m+Z,te=0;te<c;++te)for(var re=Math.max(0,Math.ceil((_-te)/o)),le=Math.min(r.outWidth,(r.inWidth+_-te)/o),he=te*g+ie,ye=0;ye<r.inChannels;++ye)for(var Oe=ye*v+he,Fe=0;Fe<r.outChannels;++Fe){for(var _e=0,Me=0;Me<r.batchSize;++Me)for(var dt=Me*k,Lt=Me*S,Ge=q;Ge<G;++Ge)for(var Ue=K+Ge*a-H,me=Ue*W+dt,Ft=Ge*L+Lt,lt=ee;lt<ne;++lt)for(var pt=X+lt*s-j,En=pt*F+me,Et=lt*N+Ft,It=re;It<le;++It){var Ce=te+It*o-_,Wt=Ce*P+En,ft=It*C+Et;_e+=R[Wt+ye]*b[ft+Fe]}d[Oe+Fe]=_e}return h.toTensor()},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.depthwiseConv2D(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Zd(this,u,o,l)),u},t.prototype.depthwiseConv2D=function(e,i,r){ae([e,i],"depthwiseConv2D");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.outChannels/r.inChannels,d=T.buffer(r.outShape,e.dtype),p=this.readSync(e.dataId),f=this.readSync(i.dataId),m=d.values,g=0;g<r.batchSize;++g)for(var v=g*e.strides[0],b=g*d.strides[0],w=0;w<r.outHeight;++w)for(var S=b+w*d.strides[1],L=w*r.strideHeight-u,N=0;N<a;++N){var C=L+N*o;if(C<0||C>=r.inHeight)continue;for(var R=N*i.strides[0],D=v+C*e.strides[1],k=0;k<r.outWidth;++k)for(var W=S+k*d.strides[2],F=k*r.strideWidth-c,P=0;P<s;++P){var H=F+P*l;if(H<0||H>=r.inWidth)continue;for(var _=R+P*i.strides[1],j=D+H*r.inChannels,K=W,q=_,G=0;G<r.inChannels;++G){for(var Z=p[j+G],X=0;X<h;++X)m[K+X]+=Z*f[q+X];K+=h,q+=h}}}return d.toTensor()},t.prototype.depthwiseConv2DDerInput=function(e,i,r){ae([e,i],"depthwiseConv2DDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=this.readSync(e.dataId),d=e.strides,p=d[0],f=d[1],m=d[2],g=this.readSync(i.dataId),v=i.strides,b=v[0],w=v[1],S=v[2],L=r.batchSize,N=r.filterHeight,C=r.filterWidth,R=r.inChannels,D=r.inHeight,k=r.inWidth,W=r.outChannels,F=r.outHeight,P=r.outWidth,H=r.strideHeight,_=r.strideWidth,j=N-1-r.padInfo.top,K=C-1-r.padInfo.left,q=W/R,G=0;G<L;++G)for(var Z=0;Z<R;++Z)for(var X=0;X<D;++X)for(var ee=X-j,ne=Math.max(0,Math.ceil(ee/H)),ie=Math.min(F,(N+ee)/H),te=0;te<k;++te){for(var re=te-K,le=Math.max(0,Math.ceil(re/_)),he=Math.min(P,(C+re)/_),ye=0,Oe=ne;Oe<ie;++Oe)for(var Fe=Oe*H-ee,_e=le;_e<he;++_e)for(var Me=_e*_-re,dt=p*G+f*Oe+m*_e,Lt=b*(N-1-Fe)+w*(C-1-Me)+S*Z,Ge=0;Ge<q;++Ge){var Ue=Z*q+Ge,me=h[dt+Ue],Ft=g[Lt+Ge];ye+=me*Ft}s[l*G+u*X+c*te+Z]=ye}return a.toTensor()},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){ae([e,i],"depthwiseConv2DDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=T.buffer(r.filterShape,"float32"),c=r.padInfo.left,h=r.padInfo.top,d=r.outChannels/r.inChannels,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((h-m)/a)),v=Math.min(r.outHeight,(r.inHeight+h-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((c-b)/s)),S=Math.min(r.outWidth,(r.inWidth+c-b)/s),L=0;L<r.outChannels;++L){for(var N=Math.trunc(L/d),C=L%d,R=0,D=0;D<r.batchSize;++D)for(var k=g;k<v;++k)for(var W=m+k*a-h,F=w;F<S;++F){var P=b+F*s-c;R+=p.get(D,W,P,N)*f.get(D,k,F,L)}u.set(R,m,b,N,C)}return u.toTensor()},t.prototype.tile=function(e,i){return ae(e,"tile"),sz(this.bufferSync(e),i)},t.prototype.gather=function(e,i,r){ae([e,i],"gather");var a=e.shape.slice(),s=this.readSync(i.dataId);a[r]=s.length;for(var o=T.buffer(a,e.dtype),l=this.bufferSync(e),u=0;u<o.size;++u){var c=o.indexToLoc(u),h=c.slice();h[r]=s[c[r]];var d=l.locToIndex(h);o.values[u]=l.values[d]}return o.toTensor()},t.prototype.batchToSpaceND=function(e,i,r){ae([e],"batchToSpaceND");var a=i.reduce(function(h,d){return h*d}),s=T.backend_util.getReshaped(e.shape,i,a),o=T.backend_util.getPermuted(s.length,i.length),l=T.backend_util.getReshapedPermuted(e.shape,i,a),u=T.backend_util.getSliceBeginCoords(r,i.length),c=T.backend_util.getSliceSize(l,r,i.length);return T.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.pool3d=function(e,i,r){ae(e,"pool3d");for(var a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=r==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,b=this.readSync(e.dataId),w=T.buffer(i.outShape,e.dtype),S=w.values,L=i.outShape[1]*i.outShape[2]*i.outShape[3]*i.outShape[4],N=i.outShape[2]*i.outShape[3]*i.outShape[4],C=i.outShape[3]*i.outShape[4],R=i.outShape[4],D=0;D<i.batchSize;++D)for(var k=D*L,W=D*e.strides[0],F=0;F<i.inChannels;++F)for(var P=0;P<i.outDepth;++P){for(var H=P*a-f,_=H;_<0;)_+=l;for(var j=Math.min(i.inDepth,h+H),K=k+P*N,q=0;q<i.outHeight;++q){for(var G=q*s-m,Z=G;Z<0;)Z+=u;for(var X=Math.min(i.inHeight,d+G),ee=K+q*C,ne=0;ne<i.outWidth;++ne){for(var ie=ne*o-g,te=ie;te<0;)te+=c;for(var re=Math.min(i.inWidth,p+ie),le=ee+ne*R,he=v,ye=0,Oe=0,Fe=_;Fe<j;Fe+=l){for(var _e=W+Fe*e.strides[1],Me=Z;Me<X;Me+=u){for(var dt=_e+Me*e.strides[2],Lt=te;Lt<re;Lt+=c){var Ge=dt+Lt*e.strides[3],Ue=b[Ge+F];if(r==="max"&&Ue>he?he=Ue:r==="avg"&&(ye+=Ue,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),N=this.bufferSync(e),C=0;C<r.batchSize;++C)for(var R=0;R<r.inChannels;++R)for(var D=0;D<r.inDepth;++D)for(var k=0;k<r.inHeight;++k)for(var W=0;W<r.inWidth;++W){for(var F=D-v,P=k-w,H=W-b,_=0,j=0;j<f;j+=h){var K=(F+j)/a;if(K<0||K>=r.outDepth||Math.floor(K)!==K)continue;for(var q=0;q<m;q+=d){var G=(P+q)/s;if(G<0||G>=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z<g;Z+=p){var X=(H+Z)/o;if(X<0||X>=r.outWidth||Math.floor(X)!==X)continue;var ee=N.get(C,K,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b<i.batchSize;++b)for(var w=0;w<i.inChannels;++w)for(var S=0;S<i.outDepth;++S){for(var L=S*a-f,N=L;N<0;)N+=l;for(var C=Math.min(i.inDepth,h+L),R=0;R<i.outHeight;++R){for(var D=R*s-m,k=D;k<0;)k+=u;for(var W=Math.min(i.inHeight,d+D),F=0;F<i.outWidth;++F){for(var P=F*o-g,H=P;H<0;)H+=c;for(var _=Math.min(i.inWidth,p+P),j=Number.NEGATIVE_INFINITY,K=-1,q=N;q<C;q+=l)for(var G=q-L,Z=k;Z<W;Z+=u)for(var X=Z-D,ee=H;ee<_;ee+=c){var ne=ee-P,ie=v.get(b,q,Z,ee,w);ie>=j&&(j=ie,K=G*d*p+X*d+ne)}r.set(K,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),N=0;N<a.batchSize;++N)for(var C=0;C<a.inChannels;++C)for(var R=0;R<a.inDepth;++R)for(var D=0;D<a.inHeight;++D)for(var k=0;k<a.inWidth;++k){for(var W=R-g,F=D-b,P=k-v,H=0,_=0;_<p;_+=c){var j=(W+_)/o;if(j<0||j>=a.outDepth||Math.floor(j)!==j)continue;for(var K=0;K<f;K+=h){var q=(F+K)/l;if(q<0||q>=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<m;G+=d){var Z=(P+G)/u;if(Z<0||Z>=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(N,j,q,Z,C),ee=_*f*m+K*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(N,j,q,Z,C);H+=ie*ne}}}w.set(H,N,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b<o;b++)for(var w=0;w<i;w++)for(var S=g*w,L=Math.floor(S),N=S-L,C=Math.min(l-1,Math.ceil(S)),R=b*e.strides[0]+L*e.strides[1],D=b*e.strides[0]+C*e.strides[1],k=0;k<r;k++)for(var W=v*k,F=Math.floor(W),P=W-F,H=Math.min(u-1,Math.ceil(W)),_=R+F*e.strides[2],j=D+F*e.strides[2],K=R+H*e.strides[2],q=D+H*e.strides[2],G=0;G<c;G++){var Z=h[_+G],X=h[j+G],ee=h[K+G],ne=h[q+G],ie=Z+(ee-Z)*P,te=X+(ne-X)*P,re=ie+(te-ie)*N;d[m++]=re}return T.tensor(d,[o,i,r,c])},t.prototype.resizeBilinearBackprop=function(e,i,r){ae([e,i],"resizeBilinearBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=[r&&h>1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S<s;S++)for(var L=S*i.strides[0],N=0;N<h;N++)for(var C=N*g,R=Math.floor(C),D=Math.min(Math.ceil(C),o-1),k=L+R*i.strides[1],W=L+D*i.strides[1],F=C-R,P=1-F,H=0;H<d;H++)for(var _=H*v,j=Math.floor(_),K=Math.min(Math.ceil(_),l-1),q=_-j,G=1-q,Z=k+j*i.strides[2],X=k+K*i.strides[2],ee=W+j*i.strides[2],ne=W+K*i.strides[2],ie=P*G,te=P*q,re=F*G,le=F*q,he=0;he<u;he++){var ye=b[w++];p[Z+he]+=ye*ie,p[X+he]+=ye*te,p[ee+he]+=ye*re,p[ne+he]+=ye*le}return T.tensor4d(p,[s,l,o,u],i.dtype)},t.prototype.resizeNearestNeighbor=function(e,i,r,a){ae(e,"resizeNearestNeighbor");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(o*i*r*c),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b<o;b++)for(var w=b*e.strides[0],S=0;S<i;S++)for(var L=m*S,N=Math.min(l-1,a?Math.round(L):Math.floor(L)),C=w+N*e.strides[1],R=0;R<r;R++)for(var D=g*R,k=Math.min(u-1,a?Math.round(D):Math.floor(D)),W=C+k*e.strides[2],F=0;F<c;F++){var P=h[W+F];d[v++]=P}return T.tensor(d,[o,i,r,c],e.dtype)},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){ae([e,i],"resizeNearestNeighborBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=this.readSync(e.dataId),m=[r&&h>1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,N=Math.ceil(S)*2+2,C=0;C<s;C++)for(var R=C*i.strides[0],D=0;D<o;D++)for(var k=R+D*i.strides[1],W=Math.floor(D*w),F=Math.floor(W-L/2),P=0;P<l;P++)for(var H=k+P*i.strides[2],_=Math.floor(P*S),j=Math.floor(_-N/2),K=0;K<u;K++){for(var q=0,G=0;G<L;G++){var Z=G+F;if(Z<0||Z>=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie<N;ie++){var te=ie+j;if(te<0||te>=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+K])}}p[H+K]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p<c;p++){var f=d(p),m=u[p]*Math.pow(r+a*f,-s);h[p]=m}return T.tensor4d(h,e.shape)},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){ae(e,"LRNGrad");for(var u=e.shape[3],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=this.readSync(r.dataId),p=new Float32Array(e.size),f=e.size,m=0;m<f;m++){for(var g=m%u,v=m-g+Math.max(0,g-a),b=m-g+Math.min(u,g+a+1),w=0,S=v;S<b;S++)w+=Math.pow(h[S],2);w=o*w+s;for(var S=v;S<b;S++){var L=-2*o*l*h[S]*d[m]/w;m===S&&(L+=Math.pow(w,-l)),L*=c[m],p[S]+=L}}return T.tensor4d(p,e.shape)},t.prototype.multinomial=function(e,i,r,a){ae(e,"multinomial");for(var s=i?e:T.softmax(e),o=s.shape[0],l=s.shape[1],u=T.zeros([o,r],"int32"),c=this.readSync(u.dataId),h=this.readSync(s.dataId),d=0;d<o;++d){var p=d*l,f=new Float32Array(l-1);f[0]=h[p];for(var m=1;m<f.length;++m)f[m]=f[m-1]+h[p+m];for(var g=nz.alea(a.toString()),v=d*r,b=0;b<r;++b){var w=g();c[v+b]=f.length;for(var S=0;S<f.length;S++)if(w<f[S]){c[v+b]=S;break}}}return u},t.prototype.oneHot=function(e,i,r,a){ae(e,"oneHot");var s=new Float32Array(e.size*i);s.fill(a);for(var o=this.readSync(e.dataId),l=0;l<e.size;++l)o[l]>=0&&o[l]<i&&(s[l*i+o[l]]=r);return T.tensor2d(s,[e.size,i],"int32")},t.prototype.nonMaxSuppression=function(e,i,r,a,s){ae(e,"nonMaxSuppression");var o=this.readSync(e.dataId),l=this.readSync(i.dataId);return rz(o,l,r,a,s)},t.prototype.depthToSpace=function(e,i,r){T.util.assert(r==="NHWC",function(){return"Only NHWC dataFormat supported on CPU for depthToSpace. Got "+r}),T.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m<a;++m)for(var g=0;g<u;++g)for(var v=Math.floor(g/i),b=g%i,w=0;w<c;++w)for(var S=Math.floor(w/i),L=w%i,N=(b*i+L)*h,C=0;C<h;++C){var R=C+N,D=R+l*(S+o*(v+s*m));p[f++]=d[D]}return T.tensor4d(p,[a,u,c,h])},t.prototype.broadcastedBinaryOp=function(e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(e.shape,i.shape),o=T.buffer(s,r),l=this.readSync(e.dataId),u=this.readSync(i.dataId),c=T.backend_util.getBroadcastDims(e.shape,s),h=T.backend_util.getBroadcastDims(i.shape,s),d=o.values;if(c.length+h.length===0)for(var p=0;p<d.length;++p)d[p]=a(l[p%l.length],u[p%u.length]);else for(var f=this.bufferSync(e),m=this.bufferSync(i),g=function(v){var b=o.indexToLoc(v),w=b.slice(-e.rank);c.forEach(function(C){return w[C]=0});var S=f.locToIndex(w),L=b.slice(-i.rank);h.forEach(function(C){return L[C]=0});var N=m.locToIndex(L);d[v]=a(l[S],u[N])},p=0;p<d.length;++p)g(p);return o.toTensor()},t.prototype.split=function(e,i,r){return az(e,i,r)},t.prototype.dispose=function(){},t.prototype.floatPrecision=function(){return 32},t.prototype.epsilon=function(){return n.prototype.epsilon.call(this)},t.prototype.cropAndResize=function(e,i,r,a,s,o){for(var l=e.shape,u=l[0],c=l[1],h=l[2],d=l[3],p=i.shape[0],f=a[0],m=a[1],g=T.buffer([p,f,m,d],"float32"),v=this.readSync(i.dataId),b=this.readSync(r.dataId),w=this.readSync(e.dataId),S=e.strides,L=g.strides,N=0;N<p;N++){var C=N*4,R=v[C],D=v[C+1],k=v[C+2],W=v[C+3],F=b[N];if(F>=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_<f;_++){var j=f>1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(j<0||j>c-1){for(var K=0;K<m;K++)for(var q=0;q<d;q++){var G=q+K*L[2]+_*L[1]+N*L[0];g.values[G]=o}continue}if(s==="bilinear")for(var Z=Math.floor(j),X=Math.ceil(j),ee=j-Z,K=0;K<m;K++){var ne=m>1?D*(h-1)+K*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+K*L[2]+_*L[1]+N*L[0];g.values[G]=o}continue}for(var ie=Math.floor(ne),te=Math.ceil(ne),re=ne-ie,q=0;q<d;q++){var G=q+ie*S[2]+Z*S[1]+F*S[0],le=w[G];G=q+te*S[2]+Z*S[1]+F*S[0];var he=w[G];G=q+ie*S[2]+X*S[1]+F*S[0];var ye=w[G];G=q+te*S[2]+X*S[1]+F*S[0];var Oe=w[G],Fe=le+(he-le)*re,_e=ye+(Oe-ye)*re;G=q+K*L[2]+_*L[1]+N*L[0],g.values[G]=Fe+(_e-Fe)*ee}}else for(var K=0;K<m;++K){var ne=m>1?D*(h-1)+K*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+K*L[2]+_*L[1]+N*L[0];g.values[G]=o}continue}for(var Me=Math.round(ne),dt=Math.round(j),q=0;q<d;q++){var Lt=q+Me*S[2]+dt*S[1]+F*S[0],Ge=q+K*L[2]+_*L[1]+N*L[0];g.values[Ge]=w[Lt]}}}}return g.toTensor()},t.prototype.sparseToDense=function(e,i,r,a){var s=T.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.sliceSize,c=s.strides,h=s.outputSize,d=!1;return this.scatter(e,i,r,h,u,l,o,c,a,d)},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=T.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3];if(l===0)return T.tensor([],o,e.dtype);for(var h=new T.TensorBuffer([l,u],e.dtype),d=this.readSync(i.dataId),p=this.readSync(e.dataId),f=0;f<l;f++){for(var m=[],g=0,v=0;v<a;v++){var b=d[f*a+v];g+=b*c[v],m.push(b)}if(g<0||g>=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w<u;w++)h.values[f*u+w]=p[g*u+w]}return h.toTensor().reshape(o)},t.prototype.scatterND=function(e,i,r){var a=T.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=T.scalar(0),d=!0;return this.scatter(e,i,r,c,l,o,s,u,h,d)},t.prototype.fill=function(e,i,r){r=r||T.util.inferDtype(i);var a=T.util.getArrayFromDType(r,T.util.sizeFromShape(e));return a.fill(i),T.engine().makeTensor(a,e,r,this)},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported for string tensors");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){var i=T.util.getArrayFromDType(e.dtype,T.util.sizeFromShape(e.shape));return this.makeOutput(i,e.shape,e.dtype)},t.prototype.linspace=function(e,i,r){return T.backend_util.linspaceImpl(e,i,r)},t.prototype.scatter=function(e,i,r,a,s,o,l,u,c,h){var d=[a/s,s],p=this.readSync(e.dataId),f=this.readSync(i.dataId);if(a===0)return T.tensor([],r,i.dtype);var m=new T.TensorBuffer(d,i.dtype);m.values.fill(this.readSync(c.dataId)[0]);for(var g=0;g<o;g++){for(var v=[],b=0,w=0;w<l;w++){var S=p[g*l+w];v.push(S),b+=S*u[w]}if(b<0||b>=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;L<s;L++)h?m.values[b*s+L]+=f[g*s+L]:m.values[b*s+L]=i.rank===0?f[0]:f[g*s+L]}return m.toTensor().reshape(r)},t}(T.KernelBackend);function O0(n){for(var t=new Float32Array(n.length),e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}var uz=function(n){var t=n.inputs.x,e=n.backend,i=new Float32Array(T.util.sizeFromShape(t.shape));if(t.dtype!=="complex64"){var r=e.data.get(t.dataId).values;i=O0(r)}else for(var a=e.data.get(t.dataId),s=a.complexTensorInfos.real,o=a.complexTensorInfos.imag,l=e.data.get(s.dataId).values,u=e.data.get(o.dataId).values,c=0;c<l.length;c++){var h=l[c],d=u[c];i[c]=Math.hypot(h,d)}return e.makeOutput(i,t.shape,"float32")},cz={kernelName:T.Abs,backendName:"cpu",kernelFunc:uz};function wr(n){return function(t,e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=T.util.computeStrides(s),u=T.util.sizeFromShape(s),c=T.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=T.util.computeStrides(t),f=T.util.computeStrides(e),m=T.backend_util.getBroadcastDims(t,s),g=T.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(var v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(var b=function(w){var S=T.util.indexToLoc(w,o,l),L=S.slice(-h);m.forEach(function(D){return L[D]=0});var N=T.util.locToIndex(L,h,p),C=S.slice(-d);g.forEach(function(D){return C[D]=0});var R=T.util.locToIndex(C,d,f);c[w]=n(i[N],r[R])},v=0;v<c.length;++v)b(v);return[c,s]}}function xn(n){var t=n.inputs,e=n.backend,i=t.real,r=t.imag,a=e.data.get(i.dataId).values,s=e.data.get(r.dataId).values,o=e.makeTensorInfo(i.shape,"complex64"),l=e.data.get(o.dataId);return l.complexTensorInfos={real:e.makeTensorInfo(i.shape,"float32",a),imag:e.makeTensorInfo(r.shape,"float32",s)},o}var hz={kernelName:T.Complex,backendName:"cpu",kernelFunc:xn};function ca(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var dz={kernelName:T.Identity,backendName:"cpu",kernelFunc:ca};function fs(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.real,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var pz={kernelName:T.Real,backendName:"cpu",kernelFunc:fs};function ms(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.dtype;if(a==="complex64"){if(r.dtype==="complex64")return ca({inputs:{x:r},backend:e});var s=T.zeros(r.shape),o=ms({inputs:{x:r},backend:e,attrs:{dtype:"float32"}}),l=xn({inputs:{real:o,imag:s},backend:e});return s.dispose(),e.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){var u=fs({inputs:{input:r},backend:e}),l=ms({inputs:{x:u},backend:e,attrs:{dtype:a}});return e.disposeIntermediateTensorInfo(u),l}if(!T.util.hasEncodingLoss(r.dtype,a)){var l=ca({inputs:{x:r},backend:e});return{dataId:l.dataId,shape:l.shape,dtype:a}}if(a==="int32"){var c=e.data.get(r.dataId).values,h=Int32Array.from(c);return e.makeTensorInfo(r.shape,"int32",h)}if(a==="bool"){var d=e.data.get(r.dataId).values,p=T.util.toTypedArray([0],r.dtype),f=wr(function(v,b){return v!==b?1:0})(r.shape,[],d,p,"bool"),m=f[0],g=f[1];return e.makeTensorInfo(g,"bool",m)}throw new Error("Error in Cast: failed to cast "+r.dtype+" to "+a)}var fz={kernelName:T.Cast,backendName:"cpu",kernelFunc:ms};function ha(n,t,e,i){return e==null?function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;ae([l,u],n);var h=c.data.get(l.dataId).values,d=c.data.get(u.dataId).values,p=i||l.dtype,f=t(l.shape,u.shape,h,d,p),m=f[0],g=f[1];return c.makeTensorInfo(g,p,m)}:function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;if(l.dtype==="complex64"||u.dtype==="complex64"){var h=ms({inputs:{x:l},backend:c,attrs:{dtype:"complex64"}}),d=c.data.get(h.dataId),p=d.complexTensorInfos.real,f=d.complexTensorInfos.imag,m=c.data.get(p.dataId).values,g=c.data.get(f.dataId).values,v=ms({inputs:{x:u},backend:c,attrs:{dtype:"complex64"}}),b=c.data.get(v.dataId),w=b.complexTensorInfos.real,S=b.complexTensorInfos.imag,L=c.data.get(w.dataId).values,N=c.data.get(S.dataId).values,C=e(l.shape,u.shape,m,g,L,N),R=C[0],D=C[1],k=C[2],W=c.makeTensorInfo(k,"float32",R),F=c.makeTensorInfo(k,"float32",D),P=xn({inputs:{real:W,imag:F},backend:c});return c.disposeIntermediateTensorInfo(h),c.disposeIntermediateTensorInfo(v),c.disposeIntermediateTensorInfo(W),c.disposeIntermediateTensorInfo(F),P}else{var H=c.data.get(l.dataId).values,_=c.data.get(u.dataId).values,j=i||l.dtype,K=t(l.shape,u.shape,H,_,j),q=K[0],k=K[1];return c.makeTensorInfo(k,j,q)}}}function Qd(n){return function(t,e,i,r,a,s){var o=T.backend_util.assertAndGetBroadcastShape(t,e),l=T.util.sizeFromShape(o),u=o.length,c=T.util.computeStrides(o),h=T.util.getTypedArrayFromDType("float32",l),d=T.util.getTypedArrayFromDType("float32",l),p=T.backend_util.getBroadcastDims(t,o),f=T.backend_util.getBroadcastDims(e,o),m=T.backend_util.mergeRealAndImagArrays(i,r),g=T.backend_util.mergeRealAndImagArrays(a,s),v=t.length,b=T.util.computeStrides(t),w=e.length,S=T.util.computeStrides(e);if(p.length+f.length===0)for(var L=0;L<h.length;L++){var N=L%m.length,C=L%g.length,R=n(m[N*2],m[N*2+1],g[C*2],g[C*2+1]);h[L]=R.real,d[L]=R.imag}else for(var D=function(k){var W=T.util.indexToLoc(k,u,c),F=W.slice(-v);p.forEach(function(K){return F[K]=0});var P=T.util.locToIndex(F,v,b),H=W.slice(-w);f.forEach(function(K){return H[K]=0});var _=T.util.locToIndex(H,w,S),j=n(m[P*2],m[P*2+1],g[_*2],g[_*2+1]);h[k]=j.real,d[k]=j.imag},L=0;L<h.length;L++)D(L);return[h,d,o]}}var E0=wr(function(n,t){return n+t}),mz=Qd(function(n,t,e,i){return{real:n+e,imag:t+i}}),D0=ha(T.Add,E0,mz),gz={kernelName:T.Add,backendName:"cpu",kernelFunc:D0};function da(n){return function(t,e,i){for(var r=T.util.getTypedArrayFromDType(e,t.length),a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}function $e(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");for(var l=s,u=l.data.get(o.dataId).values,c=T.util.sizeFromShape(o.shape),h=e||o.dtype,d=T.util.getArrayFromDType(h,c),p=0;p<c;++p)d[p]=t(u[p],a);return l.makeTensorInfo(o.shape,h,d)}}function pa(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");var l=s,u=l.data.get(o.dataId).values,c=e||o.dtype,h=t(u,c,a);return l.makeTensorInfo(o.shape,c,h)}}var k0=da(function(n){return Math.ceil(n)}),vz=pa(T.Ceil,k0),yz={kernelName:T.Ceil,backendName:"cpu",kernelFunc:vz};var F0=da(function(n){return Math.exp(n)}),bz=pa(T.Exp,F0),wz={kernelName:T.Exp,backendName:"cpu",kernelFunc:bz};var W0=da(function(n){return Math.expm1(n)}),Sz=pa(T.Expm1,W0),Lz={kernelName:T.Expm1,backendName:"cpu",kernelFunc:Sz};var U0=da(function(n){return Math.floor(n)}),Iz=pa(T.Floor,U0),Az={kernelName:T.Floor,backendName:"cpu",kernelFunc:Iz};var B0=da(function(n){return Math.log(n)}),Tz=pa(T.Log,B0),Nz={kernelName:T.Log,backendName:"cpu",kernelFunc:Tz};function z0(n,t,e,i){for(var r=T.util.getTypedArrayFromDType(i,T.util.sizeFromShape(e)),a=0;a<r.length;++a){for(var s=a*t,o=n[s],l=0;l<t;++l){var u=n[s+l];u>o&&(o=u)}r[a]=o}return r}var P0=wr(function(n,t){return n*t}),xz=Qd(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),_0=ha(T.Multiply,P0,xz),Cz={kernelName:T.Multiply,backendName:"cpu",kernelFunc:_0};var M0=da(function(n){return 1/Math.sqrt(n)}),Rz=pa(T.Rsqrt,M0),Oz={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:Rz};function H0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;c<s;++c){var h=e.length,d=T.util.computeStrides(e),p=T.util.indexToLoc(c,h,d),f=p.map(function(g,v){return g+t[v]}),m=T.util.locToIndex(f,i.length,o);u[c]=n[m]}return u}function ep(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.begin,s=i.size;ae(r,"slice");var o=T.slice_util.parseSliceParams(r,a,s),l=o[0],u=o[1];T.slice_util.assertParamsValid(r,l,u);var c=e.data.get(r.dataId).values,h=H0(c,l,u,r.shape,r.dtype);return e.makeTensorInfo(u,r.dtype,h)}var Ez={kernelName:T.Slice,backendName:"cpu",kernelFunc:ep};var V0=wr(function(n,t){return n-t}),Dz=Qd(function(n,t,e,i){return{real:n-e,imag:t-i}}),q0=ha(T.Sub,V0,Dz),kz={kernelName:T.Sub,backendName:"cpu",kernelFunc:q0};function tp(n,t,e,i,r){for(var a=t.length,s=T.util.sizeFromShape(t),o=T.util.computeStrides(t),l=T.util.computeStrides(r),u=T.util.getTypedArrayFromDType(e,T.util.sizeFromShape(r)),c=0;c<s;++c){for(var h=T.util.indexToLoc(c,a,o),d=new Array(h.length),p=0;p<d.length;p++)d[p]=h[i[p]];var f=T.util.locToIndex(d,a,l);u[f]=n[c]}return u}function G0(n,t,e,i){for(var r=T.util.parseAxisParam(t,e)[0],a=[1,e[0],1],s=0;s<r;s++)a[0]*=e[s];a[1]=e[r];for(var s=r+1;s<e.length;s++)a[2]*=e[s];for(var o={},l=new Int32Array(e[r]),u=new T.TensorBuffer(a,i,n),c=[],h=a[0]===1&&a[2]===1,s=0;s<e[r];s++){var d=void 0;if(h)d=n[s].toString();else{for(var p=[],f=0;f<a[0];f++)for(var m=0;m<a[2];m++)p.push(u.get(f,s,m));d=p.join(",")}if(o[d]!==void 0)l[s]=o[d];else{var g=Object.keys(o).length;o[d]=g,l[s]=g,c.push(s)}}var v=a.slice();v[1]=Object.keys(o).length;var b=new T.TensorBuffer(v,i);c.forEach(function(S,L){for(var N=0;N<a[0];N++)for(var C=0;C<a[2];C++)b.set(u.get(N,S,C),N,L,C)});var w=e.slice();return w[r]=v[1],{outputValues:b.values,outputShape:w,indices:l}}var Fz={__proto__:null,simpleAbsImpl:O0,addImpl:E0,ceilImpl:k0,expImpl:F0,expm1Impl:W0,floorImpl:U0,logImpl:B0,maxImpl:z0,multiplyImpl:P0,rsqrtImpl:M0,sliceImpl:H0,subImpl:V0,transposeImpl:tp,uniqueImpl:G0};var Wz="2.6.0";T.registerBackend("cpu",function(){return new R0},1);var Uz=$e(T.Acos,function(n){return Math.acos(n)}),Bz={kernelName:T.Acos,backendName:"cpu",kernelFunc:Uz};var zz=$e(T.Acosh,function(n){return Math.acosh(n)}),Pz={kernelName:T.Acosh,backendName:"cpu",kernelFunc:zz};var _z=$e(T.Asin,function(n){return Math.asin(n)}),Mz={kernelName:T.Asin,backendName:"cpu",kernelFunc:_z};var Hz=$e(T.Asinh,function(n){return Math.asinh(n)}),Vz={kernelName:T.Asinh,backendName:"cpu",kernelFunc:Hz};var qz=$e(T.Atan,function(n){return Math.atan(n)}),Gz={kernelName:T.Atan,backendName:"cpu",kernelFunc:qz};var Yz=$e(T.Atanh,function(n){return Math.atanh(n)}),jz={kernelName:T.Atanh,backendName:"cpu",kernelFunc:Yz};function np(n,t,e,i,r,a){for(var s=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,u=r.dilationWidth,c=r.effectiveFilterHeight,h=r.effectiveFilterWidth,d=r.padInfo.top,p=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=T.buffer(r.outShape,e),g=m.values,v=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],w=r.outShape[3],S=0;S<r.batchSize;++S)for(var L=S*v,N=S*i[0],C=0;C<r.inChannels;++C)for(var R=0;R<r.outHeight;++R)for(var D=R*s-d,k=Math.max(0,D),W=Math.min(r.inHeight,c+D),F=L+R*b,P=0;P<r.outWidth;++P){for(var H=P*o-p,_=Math.max(0,H),j=Math.min(r.inWidth,h+H),K=f,q=0,G=0,Z=k;Z<W;Z+=l){for(var X=N+Z*i[1],ee=_;ee<j;ee+=u){var ne=X+ee*i[2],ie=n[ne+C];a==="max"&&ie>K?K=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(K))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:K}return m}function Y0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;g<i.batchSize;++g)for(var v=0;v<i.inChannels;++v)for(var b=0;b<i.outHeight;++b){for(var w=b*o-p,S=w;S<0;)S+=u;for(var L=Math.min(i.inHeight,h+w),N=0;N<i.outWidth;++N){for(var C=N*l-f,R=C;R<0;)R+=c;for(var D=Math.min(i.inWidth,d+C),k=Number.NEGATIVE_INFINITY,W=-1,F=S;F<L;F+=u)for(var P=F-w,H=R;H<D;H+=c){var _=H-C,j=m.get(g,F,H,v);j>k&&(k=j,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,N,v)}}return s}function Kz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=ca({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=np(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var $z={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:Kz};function Xz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),N=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D<c.batchSize;++D)for(var k=0;k<c.inChannels;++k)for(var W=0;W<c.inHeight;++W)for(var F=0;F<c.inWidth;++F){for(var P=W-S,H=F-w,_=0,j=0;j<v;j+=m){var K=(P+j)/h;if(K<0||K>=c.outHeight||Math.floor(K)!==K)continue;for(var q=0;q<b;q+=g){var G=(H+q)/d;if(G<0||G>=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,K,G,k);_+=Z}}L.set(_*N,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var Jz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:Xz};function Zz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,N=0,C=0,R=0;R<c.length;++R)m[R]=f[S++]+(c[R]-h[L++])*p[N++]/Math.sqrt(d[C++]+u),S>=g&&(S=0),L>=w&&(L=0),N>=v&&(N=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var Qz={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:Zz};var eP=$e(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n<e.clipValueMin?e.clipValueMin:n}),tP={kernelName:T.ClipByValue,backendName:"cpu",kernelFunc:eP};function Zo(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.imag,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var nP={kernelName:T.Imag,backendName:"cpu",kernelFunc:Zo};function Wi(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=T.util.sizeFromShape(r.shape),o=T.util.inferFromImplicitShape(a,s),l=T.util.sizeFromShape(o);T.util.assert(s===l,function(){return"The new shape ("+o+") has "+l+" elements and the old "+("shape ("+r.shape+") has "+s+" elements. The new shape and old ")+"shape must have the same number of elements."}),e.incRef(r.dataId);var u=e.data.get(r.dataId);if(u.complexTensorInfos!=null){var c=u.complexTensorInfos.real,h=u.complexTensorInfos.imag;c.shape=o,h.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var iP={kernelName:T.Reshape,backendName:"cpu",kernelFunc:Wi};function gs(n){var t=n.inputs,e=n.backend,i=n.attrs,r=i.axis,a=T.util.parseAxisParam(r,t[0].shape)[0],s=T.backend_util.computeOutShape(t.map(function(S){return S.shape}),a);if(T.util.sizeFromShape(s)===0)return e.makeTensorInfo(s,t[0].dtype,[]);var o=t.filter(function(S){return T.util.sizeFromShape(S.shape)>0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return fs({inputs:{input:S},backend:e})}),c=o.map(function(S){return Zo({inputs:{input:S},backend:e})}),h=gs({inputs:u,backend:e,attrs:{axis:r}}),d=gs({inputs:c,backend:e,attrs:{axis:r}}),p=xn({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),N=[-1,L];return Wi({inputs:{x:S},backend:e,attrs:{shape:N}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,N=T.util.sizeFromShape(S.shape);m.set(L,g),g+=N})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,N=0,C=0;C<S.shape[0];++C)for(var R=C*s[1]+v,D=0;D<S.shape[1];++D)m[R+D]=L[N++];v+=S.shape[1]})}var b=T.backend_util.computeOutShape(o.map(function(S){return S.shape}),a),w=e.makeTensorInfo(b,t[0].dtype,m);return f.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),w}var rP={kernelName:T.Concat,backendName:"cpu",kernelFunc:gs};var aP=$e(T.Cos,function(n){return Math.cos(n)}),sP={kernelName:T.Cos,backendName:"cpu",kernelFunc:aP};var oP=$e(T.Cosh,function(n){return Math.cosh(n)}),lP={kernelName:T.Cosh,backendName:"cpu",kernelFunc:oP};var uP={kernelName:T.Dilation2D,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=i,l=o.strides,u=o.pad,c=o.dilations,h=e,d=h.data.get(a.dataId).values,p=a.shape.length,f=h.data.get(s.dataId).values,m=s.shape.length,g=T.backend_util.computeDilation2DInfo(a.shape,s.shape,l,u,"NHWC",c),v=g.batchSize,b=g.inHeight,w=g.inWidth,S=g.inChannels,L=g.outHeight,N=g.outWidth,C=g.padInfo,R=g.strideHeight,D=g.strideWidth,k=g.filterHeight,W=g.filterWidth,F=g.dilationHeight,P=g.dilationWidth,H=g.outShape,_=T.util.sizeFromShape(H),j=H.length,K=T.util.getArrayFromDType(a.dtype,_),q=0;q<v;++q)for(var G=0;G<L;++G)for(var Z=G*R-C.top,X=0;X<N;++X)for(var ee=X*D-C.left,ne=0;ne<S;++ne){for(var ie=Number.MIN_SAFE_INTEGER,te=0;te<k;++te){var re=Z+te*F;if(re>=0&&re<b)for(var le=0;le<W;++le){var he=ee+le*P;if(he>=0&&he<w){var ye=T.util.locToIndex([q,re,he,ne],p,T.util.computeStrides(a.shape)),Oe=T.util.locToIndex([te,le,ne],m,T.util.computeStrides(s.shape)),Fe=d[ye]+f[Oe];Fe>ie&&(ie=Fe)}}}var _e=T.util.locToIndex([q,G,X,ne],j,T.util.computeStrides(H));K[_e]=ie}var Me=h.write(T.util.toTypedArray(K,a.dtype),H,a.dtype);return{dataId:Me,shape:H,dtype:a.dtype}}};var cP={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,N=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),j=0;j<g;++j)for(var K=0;K<S;++K)for(var q=K*C-N.top,G=0;G<L;++G)for(var Z=G*R-N.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=0,ie=0,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[j][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[j][K][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var hP={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,N=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),j=0;j<g;++j)for(var K=0;K<S;++K)for(var q=K*C-N.top,G=0;G<L;++G)for(var Z=G*R-N.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=q<0?0:q,ie=Z<0?0:Z,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[j][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=re,ie=he)}}}_[j][ne][ie][X]+=H[j][K][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var dP=wr(function(n,t){return n/t}),pP=ha(T.Div,dP),ip={kernelName:T.Div,backendName:"cpu",kernelFunc:pP};var fP=$e(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),mP={kernelName:T.Elu,backendName:"cpu",kernelFunc:fP};var gP=T.backend_util.ERF_P,vP=T.backend_util.ERF_A1,yP=T.backend_util.ERF_A2,bP=T.backend_util.ERF_A3,wP=T.backend_util.ERF_A4,SP=T.backend_util.ERF_A5,LP=$e(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+gP*e);return t*(1-((((SP*i+wP)*i+bP)*i+yP)*i+vP)*i*Math.exp(-e*e))}),IP={kernelName:T.Erf,backendName:"cpu",kernelFunc:LP};function j0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p<r;p++){for(var f=ep({inputs:{x:o},backend:e,attrs:{begin:[p,0],size:[1,a]}}),m=ep({inputs:{x:l},backend:e,attrs:{begin:[p,0],size:[1,a]}}),g=xn({inputs:{real:f,imag:m},backend:e}),v=AP(g,t,e),b=v.real,w=v.imag,S=T.backend_util.mergeRealAndImagArrays(b,w),L=0;L<a;L++){var N=T.backend_util.getComplexWithIndex(S,L);h[p*a+L]=N.real,d[p*a+L]=N.imag}e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(g)}var C=e.makeTensorInfo(u,"float32",h),R=e.makeTensorInfo(u,"float32",d),D=xn({inputs:{real:C,imag:R},backend:e});return e.disposeIntermediateTensorInfo(C),e.disposeIntermediateTensorInfo(R),D}function AP(n,t,e){var i=T.util.sizeFromShape(n.shape),r=e.data.get(n.dataId),a=e.data.get(r.complexTensorInfos.real.dataId).values,s=e.data.get(r.complexTensorInfos.imag.dataId).values;if(TP(i)){var o=rp(a,s,i,t,e),l=[n.shape[0],n.shape[1]];if(t){var u=e.makeTensorInfo(l,"float32",o.real),c=e.makeTensorInfo(l,"float32",o.imag),h=e.makeTensorInfo([],"float32",T.util.createScalarValue(i,"float32")),d=ca({inputs:{x:h},backend:e}),p=ip.kernelFunc({inputs:{a:u,b:h},backend:e}),f=ip.kernelFunc({inputs:{a:c,b:d},backend:e}),m=e.data.get(p.dataId).values,g=e.data.get(f.dataId).values;return e.disposeIntermediateTensorInfo(u),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return o}else{var v=T.backend_util.mergeRealAndImagArrays(a,s),b=NP(v,i,t);return T.backend_util.splitRealAndImagArrays(b)}}function TP(n){return(n&n-1)===0}function rp(n,t,e,i,r){if(e===1)return{real:n,imag:t};var a=T.backend_util.mergeRealAndImagArrays(n,t),s=e/2,o=T.backend_util.complexWithEvenIndex(a),l=o.real,u=o.imag,c=[l.length],h=r.makeTensorInfo(c,"float32",l),d=r.makeTensorInfo(c,"float32",u),p=xn({inputs:{real:h,imag:d},backend:r}),f=T.backend_util.complexWithOddIndex(a),m=f.real,g=f.imag,v=[m.length],b=r.makeTensorInfo(v,"float32",m),w=r.makeTensorInfo(v,"float32",g),S=xn({inputs:{real:b,imag:w},backend:r}),L=rp(l,u,s,i,r),N=L.real,C=L.imag,R=[N.length],D=r.makeTensorInfo(R,"float32",N),k=r.makeTensorInfo(R,"float32",C),W=xn({inputs:{real:D,imag:k},backend:r}),F=rp(m,g,s,i,r),P=F.real,H=F.imag,_=[P.length],j=r.makeTensorInfo(_,"float32",P),K=r.makeTensorInfo(_,"float32",H),q=xn({inputs:{real:j,imag:K},backend:r}),G=T.backend_util.exponents(e,i),Z=[G.real.length],X=r.makeTensorInfo(Z,"float32",G.real),ee=r.makeTensorInfo(Z,"float32",G.imag),ne=xn({inputs:{real:X,imag:ee},backend:r}),ie=_0({inputs:{a:ne,b:q},backend:r}),te=D0({inputs:{a:W,b:ie},backend:r}),re=q0({inputs:{a:W,b:ie},backend:r}),le=fs({inputs:{input:te},backend:r}),he=fs({inputs:{input:re},backend:r}),ye=Zo({inputs:{input:te},backend:r}),Oe=Zo({inputs:{input:re},backend:r}),Fe=gs({inputs:[le,he],backend:r,attrs:{axis:0}}),_e=gs({inputs:[ye,Oe],backend:r,attrs:{axis:0}}),Me=r.data.get(Fe.dataId).values,dt=r.data.get(_e.dataId).values;return r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(w),r.disposeIntermediateTensorInfo(S),r.disposeIntermediateTensorInfo(D),r.disposeIntermediateTensorInfo(k),r.disposeIntermediateTensorInfo(W),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(q),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(ie),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(re),r.disposeIntermediateTensorInfo(le),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(he),r.disposeIntermediateTensorInfo(Oe),r.disposeIntermediateTensorInfo(Fe),r.disposeIntermediateTensorInfo(_e),{real:Me,imag:dt}}function NP(n,t,e){for(var i=new Float32Array(t*2),r=0;r<t;r++){for(var a=0,s=0,o=0;o<t;o++){var l=T.backend_util.exponent(r*o,t,e),u=T.backend_util.getComplexWithIndex(n,o);a+=u.real*l.real-u.imag*l.imag,s+=u.real*l.imag+u.imag*l.real}e&&(a/=t,s/=t),T.backend_util.assignToTypedArray(i,a,s,r)}return i}function xP(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Wi({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=j0(o,!1,e),u=Wi({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var CP={kernelName:T.FFT,backendName:"cpu",kernelFunc:xP};var RP={kernelName:T.FlipLeftRight,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=i,s=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),o=r.shape,l=o[0],u=o[1],c=o[2],h=o[3],d=a.data.get(r.dataId).values,p=0;p<l;p++)for(var f=p*c*u*h,m=0;m<u;m++)for(var g=m*(c*h),v=0;v<c;v++)for(var b=v*h,w=0;w<h;w++){var S=[l,m,v,w],L=S[2],N=Math.round(c-L),C=f+g+b+w,R=d[C];if(N>=0&&N<c){var D=N*h,k=f+g+D+w;R=d[k]}s[C]=R}var W=a.write(s,r.shape,r.dtype);return{dataId:W,shape:r.shape,dtype:r.dtype}}};function OP(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Wi({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=j0(o,!0,e),u=Wi({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var EP={kernelName:T.IFFT,backendName:"cpu",kernelFunc:OP};var DP=$e(T.IsFinite,function(n){return Number.isFinite(n)?1:0},"bool"),kP={kernelName:T.IsFinite,backendName:"cpu",kernelFunc:DP};var FP=$e(T.IsInf,function(n){return Math.abs(n)===Infinity?1:0},"bool"),WP={kernelName:T.IsInf,backendName:"cpu",kernelFunc:FP};var UP=$e(T.IsNan,function(n){return Number.isNaN(n)?1:0},"bool"),BP={kernelName:T.IsNan,backendName:"cpu",kernelFunc:UP};var zP=$e(T.Log1p,function(n){return Math.log1p(n)}),PP={kernelName:T.Log1p,backendName:"cpu",kernelFunc:zP};var _P=$e(T.LogicalNot,function(n){return n?0:1},"bool"),MP={kernelName:T.LogicalNot,backendName:"cpu",kernelFunc:_P};var HP={kernelName:T.Max,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape,c=u.length,h=T.util.parseAxisParam(s,u),d=h,p=T.backend_util.getAxesPermutation(d,c),f=l.data.get(r.dataId).values;if(p!=null){for(var m=new Array(c),g=0;g<m.length;g++)m[g]=u[p[g]];f=tp(f,u,r.dtype,p,m),d=T.backend_util.getInnerMostAxes(d.length,c),u=m}ae(r,"max"),T.backend_util.assertAxesAreInnerMostDims("max",d,c);var v=T.backend_util.computeOutAndReduceShapes(u,d),b=v[0],w=v[1],S=T.util.sizeFromShape(w),L=z0(f,S,b,r.dtype),N=l.write(L,b,r.dtype),C=b;if(o){var m=T.backend_util.expandShapeToKeepDim(b,h);C=m}return{dataId:N,shape:C,dtype:r.dtype}}};function VP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=ca({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=np(d,r.shape,r.dtype,p,c,"max");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var qP={kernelName:T.MaxPool,backendName:"cpu",kernelFunc:VP};function GP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ae([a,s],"maxPoolBackprop");for(var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=T.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=e.data.get(o.dataId).values,f=T.buffer(d.outShape,o.dtype,Y0(p,o.shape,o.dtype,d).values),m=d.strideHeight,g=d.strideWidth,v=d.dilationHeight,b=d.dilationWidth,w=d.effectiveFilterHeight,S=d.effectiveFilterWidth,L=S-1-d.padInfo.left,N=w-1-d.padInfo.top,C=T.buffer(o.shape,"float32"),R=e.data.get(r.dataId).values,D=T.buffer(r.shape,"float32",R),k=0;k<d.batchSize;++k)for(var W=0;W<d.inChannels;++W)for(var F=0;F<d.inHeight;++F)for(var P=0;P<d.inWidth;++P){for(var H=F-N,_=P-L,j=0,K=0;K<w;K+=v){var q=(H+K)/m;if(q<0||q>=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<S;G+=b){var Z=(_+G)/g;if(Z<0||Z>=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=K*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);j+=ie*ne}}C.set(j,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var YP={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:GP};function jP(n,t,e,i,r){var a=T.util.computeStrides(t),s=np(n,t,e,a,r,"max"),o=Y0(n,t,e,r,!0,i);return[s.values,o.values]}var KP={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=jP(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var $P=T.kernel_impls.nonMaxSuppressionV4Impl,XP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=$P(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var JP=T.kernel_impls.nonMaxSuppressionV5Impl,ZP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=JP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var QP=wr(function(n,t){return n!==t?1:0}),e_=ha(T.NotEqual,QP,null,"bool"),t_={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:e_};function n_(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(N,C){return N[0]+r.shape[C]+N[1]}),l=a.map(function(N){return N[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v<c;v++){var b=T.util.indexToLoc(v,h,d),w=b.map(function(N,C){return N+l[C]}),S=T.util.locToIndex(w,f,m);g[S]=u[v]}var L=e.write(g,o,r.dtype);return{dataId:L,shape:o,dtype:r.dtype}}var K0={kernelName:T.PadV2,backendName:"cpu",kernelFunc:n_};var i_=$e(T.Reciprocal,function(n){return 1/n}),r_={kernelName:T.Reciprocal,backendName:"cpu",kernelFunc:i_};var a_={kernelName:T.RotateWithOffset,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),h=r.shape,d=h[0],p=h[1],f=h[2],m=h[3],g=T.backend_util.getImageCenter(l,p,f),v=g[0],b=g[1],w=255,S=Math.sin(s),L=Math.cos(s),N=u.data.get(r.dataId).values,C=0;C<d;C++)for(var R=C*f*p*m,D=0;D<p;D++)for(var k=D*(f*m),W=0;W<f;W++)for(var F=W*m,P=0;P<m;P++){var H=[d,D,W,P],_=H[2],j=H[1],K=(_-v)*L-(j-b)*S,q=(_-v)*S+(j-b)*L;K=Math.round(K+v),q=Math.round(q+b);var G=o;if(typeof o!="number"&&(P===3?G=w:G=o[P]),K>=0&&K<f&&q>=0&&q<p){var Z=q*(f*m),X=K*m,ee=R+Z+X+P;G=N[ee]}var ne=R+k+F+P;c[ne]=G}var ie=u.write(c,r.shape,r.dtype);return{dataId:ie,shape:r.shape,dtype:r.dtype}}};var s_=$e(T.Round,function(n){var t=Math.floor(n);return n-t<.5?Math.floor(n):n-t>.5?Math.ceil(n):t%2===0?t:t+1}),o_={kernelName:T.Round,backendName:"cpu",kernelFunc:s_};var l_=T.backend_util.SELU_SCALEALPHA,u_=T.backend_util.SELU_SCALE,c_=$e(T.Selu,function(n){return n>=0?u_*n:l_*(Math.exp(n)-1)}),h_={kernelName:T.Selu,backendName:"cpu",kernelFunc:c_};var d_=$e(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),p_={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:d_};var f_=$e(T.Sign,function(n){return n<0?-1:n>0?1:0}),m_={kernelName:T.Sign,backendName:"cpu",kernelFunc:f_};var g_=$e(T.Sin,function(n){return Math.sin(n)}),v_={kernelName:T.Sin,backendName:"cpu",kernelFunc:g_};var y_=$e(T.Sinh,function(n){return Math.sinh(n)}),b_={kernelName:T.Sinh,backendName:"cpu",kernelFunc:y_};var w_=11920928955078125e-23,$0=Math.log(w_)+2,S_=$e(T.Softplus,function(n){var t=n>-$0,e=n<$0,i=Math.exp(n),r;return e?r=i:t?r=n:r=Math.log(1+i),r}),L_={kernelName:T.Softplus,backendName:"cpu",kernelFunc:S_};function X0(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm;ae(r,"transpose");for(var s=r.shape.length,o=new Array(s),l=0;l<o.length;l++)o[l]=r.shape[a[l]];var u=i.data.get(r.dataId).values,c=tp(u,r.shape,r.dtype,a,o),h=i.write(c,o,r.dtype);return{dataId:h,shape:o,dtype:r.dtype}}var I_={kernelName:T.Transpose,backendName:"cpu",kernelFunc:X0};function A_(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.blockShape,s=i.paddings;ae([r],"spaceToBatchND");var o=T.util.sizeFromShape(a),l=[[0,0]];l.push.apply(l,s);for(var u=1+a.length;u<r.shape.length;++u)l.push([0,0]);var c=K0.kernelFunc({inputs:{x:r},backend:e,attrs:{paddings:l,constantValue:0}}),h=T.backend_util.getReshaped(c.shape,a,o,!1),d=T.backend_util.getPermuted(h.length,a.length,!1),p=T.backend_util.getReshapedPermuted(c.shape,a,o,!1),f={x:c},m={shape:h},g=Wi({inputs:f,backend:e,attrs:m}),v={x:g},b={perm:d},w=X0({inputs:v,backend:e,attrs:b}),S={x:w},L={shape:p},N=Wi({inputs:S,backend:e,attrs:L});return e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(w),N}var T_={kernelName:T.SpaceToBatchND,backendName:"cpu",kernelFunc:A_};var N_=$e(T.Sqrt,function(n){return Math.sqrt(n)}),x_={kernelName:T.Sqrt,backendName:"cpu",kernelFunc:N_};var C_={kernelName:T.Square,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.x,r=e;ae(i,"square");for(var a=r.data.get(i.dataId).values,s=new Float32Array(a.length),o=0;o<a.length;++o){var l=a[o];s[o]=l*l}var u=r.write(s,i.shape,i.dtype);return{dataId:u,shape:i.shape,dtype:i.dtype}}};var R_=wr(function(n,t){var e=n-t;return e*e}),O_=ha(T.SquaredDifference,R_),E_={kernelName:T.SquaredDifference,backendName:"cpu",kernelFunc:O_};var D_=$e(T.Step,function(n,t){var e=t;return isNaN(n)?NaN:n>0?1:e.alpha}),k_={kernelName:T.Step,backendName:"cpu",kernelFunc:D_};var F_=$e(T.Tan,function(n){return Math.tan(n)}),W_={kernelName:T.Tan,backendName:"cpu",kernelFunc:F_};var U_=$e(T.Tanh,function(n){return Math.tanh(n)}),B_={kernelName:T.Tanh,backendName:"cpu",kernelFunc:U_};function z_(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=G0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var P_={kernelName:T.Unique,backendName:"cpu",kernelFunc:z_};var __=[cz,Bz,Pz,gz,Mz,Vz,Gz,jz,$z,Jz,Qz,fz,yz,tP,hz,rP,sP,lP,uP,hP,cP,ip,mP,IP,wz,Lz,CP,RP,Az,dz,EP,nP,kP,WP,BP,Nz,PP,MP,qP,YP,KP,HP,Cz,XP,ZP,t_,K0,pz,r_,iP,a_,o_,Oz,h_,p_,m_,v_,b_,Ez,L_,T_,x_,C_,E_,k_,kz,W_,B_,I_,P_];for(var ap=0,J0=__;ap<J0.length;ap++){var M_=J0[ap];T.registerKernel(M_)}ps.MathBackendCPU=R0;ps.shared=Fz;ps.version_cpu=Wz});var FL=Le(qn=>{"use strict";Object.defineProperty(qn,"__esModule",{value:!0});var x=lr();var sp=function(n,t){return sp=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},sp(n,t)};function H_(n,t){sp(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Qo(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function el(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Sr={},op={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function Q0(n,t){Sr[n]=t}function Gn(n){if(!(n in Sr)){var t=V_(n);if(t!==null)Sr[n]=t;else return console.log("Could not get context for WebGL version",n),null}var e=Sr[n];return e.isContextLost()?(delete Sr[n],Gn(n)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),Sr[n])}function q_(n){if(typeof OffscreenCanvas!="undefined"&&n===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function V_(n){if(n!==1&&n!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");var t=q_(n);return t.addEventListener("webglcontextlost",function(e){e.preventDefault(),delete Sr[n]},!1),n===1?t.getContext("webgl",op)||t.getContext("experimental-webgl",op):t.getContext("webgl2",op)}var vs;(function(n){n[n.DENSE=0]="DENSE",n[n.SHARED_BATCH=1]="SHARED_BATCH"})(vs||(vs={}));var sn;(function(n){n[n.RENDER=0]="RENDER",n[n.UPLOAD=1]="UPLOAD",n[n.PIXELS=2]="PIXELS",n[n.DOWNLOAD=3]="DOWNLOAD"})(sn||(sn={}));var Ot;(function(n){n[n.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",n[n.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",n[n.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",n[n.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",n[n.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Ot||(Ot={}));function ys(n,t){return[t,n]}function G_(n,t){return n*t}function bs(n){var t=x.util.sizeFromShape(n),e=Math.ceil(t/4);return x.util.sizeToSquarishShape(e)}function fa(n,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(n/2))]}function Y_(n,t){var e=fa(n,t),i=e[0],r=e[1];return i*r*4}function lp(n,t){var e=n,i,r,a,s,o,l,u,c,h,d;return x.env().getNumber("WEBGL_VERSION")===2?(i=e.R32F,r=e.R16F,a=e.RGBA16F,s=e.RGBA32F,o=e.RED,u=4,c=1,h=e.HALF_FLOAT,d=e.FLOAT):(i=n.RGBA,r=n.RGBA,a=n.RGBA,s=e.RGBA,o=n.RGBA,u=4,c=4,h=t!=null?t.HALF_FLOAT_OES:null,d=n.FLOAT),l=n.RGBA,{internalFormatFloat:i,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:s,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:u,defaultNumChannels:c,textureTypeHalfFloat:h,textureTypeFloat:d}}function ce(n,t){var e=t();return x.env().getBool("DEBUG")&&j_(n),e}function j_(n){var t=n.getError();if(t!==n.NO_ERROR)throw new Error("WebGL Error: "+eS(n,t))}var K_=596e-10,$_=65504;function tS(n){return!!(x.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||n===0||K_<Math.abs(n)&&Math.abs(n)<$_)}function eS(n,t){switch(t){case n.NO_ERROR:return"NO_ERROR";case n.INVALID_ENUM:return"INVALID_ENUM";case n.INVALID_VALUE:return"INVALID_VALUE";case n.INVALID_OPERATION:return"INVALID_OPERATION";case n.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case n.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case n.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return"Unknown error code "+t}}function ws(n,t){return li(n,function(){return n.getExtension(t)},'Extension "'+t+'" not supported on this browser.')}function nS(n,t){var e=li(n,function(){return n.createShader(n.VERTEX_SHADER)},"Unable to create vertex WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw console.log(n.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function iS(n,t){var e=li(n,function(){return n.createShader(n.FRAGMENT_SHADER)},"Unable to create fragment WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw X_(t,n.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var J_=/ERROR: [0-9]+:([0-9]+):/g;function X_(n,t){var e=J_.exec(t);if(e==null){console.log("Couldn't parse line number in error: "+t),console.log(n);return}for(var i=+e[1],r=n.split(`
 `),a=r.length.toString().length+2,s=r.map(function(d,p){return x.util.rightPad((p+1).toString(),a)+d}),o=0,l=0;l<s.length;l++)o=Math.max(s[l].length,o);var u=s.slice(0,i-1),c=s.slice(i-1,i),h=s.slice(i);console.log(u.join(`
 `)),console.log(t.split(`
 `)[0]),console.log("%c "+x.util.rightPad(c[0],o),"border:1px solid red; background-color:#e3d2d2; color:#a61717"),console.log(h.join(`
-`))}function rS(n){return li(n,function(){return n.createProgram()},"Unable to create WebGLProgram.")}function aS(n,t){if(ce(n,function(){return n.linkProgram(t)}),n.getProgramParameter(t,n.LINK_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function tl(n,t){if(ce(n,function(){return n.validateProgram(t)}),n.getProgramParameter(t,n.VALIDATE_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function sS(n,t){var e=li(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function oS(n,t){var e=li(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ELEMENT_ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ELEMENT_ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function Z_(){return x.env().getNumber("WEBGL_VERSION")===2?1:4}function lS(n){return li(n,function(){return n.createTexture()},"Unable to create WebGLTexture.")}function uS(n,t){var e=x.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(n<=0||t<=0){var i="["+n+"x"+t+"]";throw new Error("Requested texture size "+i+" is invalid.")}if(n>e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function cS(n){return li(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function lp(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function dS(n,t,e){hS(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function Q_(n,t){hS(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function pS(n,t,e){return li(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function fS(n,t,e){return n.getUniformLocation(t,e)}function mS(n,t,e,i){ce(n,function(){return dS(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function eM(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function nl(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function up(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function Ss(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+gS(n,t))}function gS(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function li(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function hS(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(i<n.TEXTURE0||i>e){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function Lr(n,t){return t===void 0&&(t=2),x.util.sizeFromShape(n.slice(0,n.length-t))}function Ir(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function il(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[Lr(n)].concat(Ir(n))),t}function vS(n,t){var e;t===void 0&&(t=!1);var i=x.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?x.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=x.util.squeezeShape(n);n=r.newShape}var a=x.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=Lr(n),o=2,l=2;return n.length&&(e=Ir(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),x.util.sizeToSquarishShape(a).map(function(u){return u*2})}return x.util.sizeToSquarishShape(a)}function rl(n){return n%2===0}function Ls(n,t){if(n=n.slice(-2),t=t.slice(-2),x.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(rl(e)&&rl(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&rl(n[0])&&rl(t[0])}var al,sl;function yS(n){if(al==null){var t=Gn(n);al=t.getParameter(t.MAX_TEXTURE_SIZE)}return al}function tM(){al=null}function nM(){sl=null}function bS(n){if(sl==null){var t=Gn(n);sl=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,sl)}function wS(n){if(n===0)return 0;var t,e=Gn(n);return ln(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:ln(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function ln(n,t){var e=n.getExtension(t);return e!=null}function cp(n){try{var t=Gn(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function SS(n){if(n===0)return!1;var t=Gn(n);if(n===1){if(!ln(t,"OES_texture_float"))return!1}else if(!ln(t,"EXT_color_buffer_float"))return!1;var e=hp(t);return e}function LS(n){if(n===0)return!1;var t=Gn(n);if(n===1){if(!ln(t,"OES_texture_float"))return!1;if(!ln(t,"WEBGL_color_buffer_float"))return!1}else{if(ln(t,"EXT_color_buffer_float"))return hp(t);var e="EXT_color_buffer_half_float";if(ln(t,e)){var i=t.getExtension(e);return iM(t,i)}return!1}var r=hp(t);return r}function hp(n){var t=op(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function iM(n,t){var e=op(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function IS(n){if(n!==2)return!1;var t=Gn(n),e=t.fenceSync!=null;return e}function ma(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&x.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var rM={__proto__:null,callAndCheck:ce,canBeRepresented:tS,getWebGLErrorMessage:eS,getExtensionOrThrow:ws,createVertexShader:nS,createFragmentShader:iS,createProgram:rS,linkProgram:aS,validateProgram:tl,createStaticVertexBuffer:sS,createStaticIndexBuffer:oS,getNumChannels:Z_,createTexture:lS,validateTextureSize:uS,createFramebuffer:cS,bindVertexBufferToProgramAttribute:lp,bindTextureUnit:dS,unbindTextureUnit:Q_,getProgramUniformLocationOrThrow:pS,getProgramUniformLocation:fS,bindTextureToProgramUniformSampler:mS,bindCanvasToFramebuffer:eM,bindColorTextureToFramebuffer:nl,unbindColorTextureFromFramebuffer:up,validateFramebuffer:Ss,getFramebufferErrorMessage:gS,getBatchDim:Lr,getRowsCols:Ir,getShapeAs3D:il,getTextureShapeFromLogicalShape:vS,isReshapeFree:Ls,getWebGLMaxTextureSize:yS,resetMaxTextureSize:tM,resetMaxTexturesInShader:nM,getMaxTexturesInShader:bS,getWebGLDisjointQueryTimerVersion:wS,hasExtension:ln,isWebGLVersionEnabled:cp,isCapableOfRenderingToFloatTexture:SS,isDownloadFloatTextureEnabled:LS,isWebGLFenceEnabled:IS,assertNotComplex:ma};var we=x.env();we.registerFlag("HAS_WEBGL",function(){return we.getNumber("WEBGL_VERSION")>0});we.registerFlag("WEBGL_VERSION",function(){return cp(2)?2:cp(1)?1:0});we.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});we.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return we.get("WEBGL_VERSION")===2});we.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});we.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});we.registerFlag("WEBGL_PACK",function(){return we.getBool("HAS_WEBGL")});we.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_CLIP",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});we.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_REDUCE",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_LAZILY_UNPACK",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_CONV_IM2COL",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return yS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return bS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=we.getNumber("WEBGL_VERSION");return n===0?0:wS(n)});we.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return we.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!x.device_util.isMobile()});we.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return SS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return we.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:we.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});we.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return LS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return IS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=we.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});we.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function aM(n){const t=new Float32Array(n.length);for(let e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}function dp(n){return(t,e,i,r,a)=>{const s=x.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=x.util.computeStrides(s),u=x.util.sizeFromShape(s),c=x.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=x.util.computeStrides(t),f=x.util.computeStrides(e),m=x.backend_util.getBroadcastDims(t,s),g=x.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(let v=0;v<c.length;++v){const b=x.util.indexToLoc(v,o,l),w=b.slice(-h);m.forEach(C=>w[C]=0);const S=x.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const N=x.util.locToIndex(L,d,f);c[v]=n(i[S],r[N])}return[c,s]}}const sM=dp((n,t)=>n+t);function ga(n){return(t,e,i)=>{const r=x.util.getTypedArrayFromDType(e,t.length);for(let a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}const oM=ga(n=>Math.ceil(n));const lM=ga(n=>Math.exp(n));const uM=ga(n=>Math.expm1(n));const cM=ga(n=>Math.floor(n));const hM=ga(n=>Math.log(n));function dM(n,t,e,i){const r=x.util.getTypedArrayFromDType(i,x.util.sizeFromShape(e));for(let a=0;a<r.length;++a){const s=a*t;let o=n[s];for(let l=0;l<t;++l){const u=n[s+l];u>o&&(o=u)}r[a]=o}return r}const pM=dp((n,t)=>n*t);const fM=ga(n=>1/Math.sqrt(n));function mM(n,t,e,i,r){const a=x.slice_util.isSliceContinous(i,t,e),s=x.util.sizeFromShape(e),o=x.util.computeStrides(i);if(a){const u=x.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=x.util.getTypedArrayFromDType(r,s);for(let u=0;u<s;++u){const c=e.length,h=x.util.computeStrides(e),d=x.util.indexToLoc(u,c,h),p=d.map((m,g)=>m+t[g]),f=x.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const gM=dp((n,t)=>n-t);function vM(n,t,e,i,r){const a=t.length,s=x.util.sizeFromShape(t),o=x.util.computeStrides(t),l=x.util.computeStrides(r),u=x.util.getTypedArrayFromDType(e,x.util.sizeFromShape(r));for(let c=0;c<s;++c){const h=x.util.indexToLoc(c,a,o),d=new Array(h.length);for(let f=0;f<d.length;f++)d[f]=h[i[f]];const p=x.util.locToIndex(d,a,l);u[p]=n[c]}return u}function yM(n,t,e,i){const r=x.util.parseAxisParam(t,e)[0],a=[1,e[0],1];for(let f=0;f<r;f++)a[0]*=e[f];a[1]=e[r];for(let f=r+1;f<e.length;f++)a[2]*=e[f];const s={},o=new Int32Array(e[r]),l=new x.TensorBuffer(a,i,n),u=[],c=a[0]===1&&a[2]===1;for(let f=0;f<e[r];f++){let m;if(c)m=n[f].toString();else{const g=[];for(let v=0;v<a[0];v++)for(let b=0;b<a[2];b++)g.push(l.get(v,f,b));m=g.join(",")}if(s[m]!==void 0)o[f]=s[m];else{const g=Object.keys(s).length;s[m]=g,o[f]=g,u.push(f)}}const h=a.slice();h[1]=Object.keys(s).length;const d=new x.TensorBuffer(h,i);u.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let v=0;v<a[2];v++)d.set(l.get(g,f,v),g,m,v)});const p=e.slice();return p[r]=h[1],{outputValues:d.values,outputShape:p,indices:o}}var bM=aM,wM=sM,SM=oM,LM=lM,IM=uM,AM=cM,TM=hM,NM=dM,xM=pM,CM=fM,RM=mM,OM=gM,AS=vM,EM=yM;var DM=function(){function n(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map(function(a,s){return"T"+s});var i=[];this.variableNames.forEach(function(a){i.push("float v"+a+" = get"+a+"AtOutCoords();")});var r=this.variableNames.map(function(a){return"v"+a}).join(" + ");this.userCode=`
+`))}function rS(n){return li(n,function(){return n.createProgram()},"Unable to create WebGLProgram.")}function aS(n,t){if(ce(n,function(){return n.linkProgram(t)}),n.getProgramParameter(t,n.LINK_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function tl(n,t){if(ce(n,function(){return n.validateProgram(t)}),n.getProgramParameter(t,n.VALIDATE_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function sS(n,t){var e=li(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function oS(n,t){var e=li(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ELEMENT_ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ELEMENT_ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function Z_(){return x.env().getNumber("WEBGL_VERSION")===2?1:4}function lS(n){return li(n,function(){return n.createTexture()},"Unable to create WebGLTexture.")}function uS(n,t){var e=x.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(n<=0||t<=0){var i="["+n+"x"+t+"]";throw new Error("Requested texture size "+i+" is invalid.")}if(n>e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function cS(n){return li(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function up(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function dS(n,t,e){hS(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function Q_(n,t){hS(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function pS(n,t,e){return li(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function fS(n,t,e){return n.getUniformLocation(t,e)}function mS(n,t,e,i){ce(n,function(){return dS(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function eM(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function nl(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function cp(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function Ss(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+gS(n,t))}function gS(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function li(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function hS(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(i<n.TEXTURE0||i>e){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function Lr(n,t){return t===void 0&&(t=2),x.util.sizeFromShape(n.slice(0,n.length-t))}function Ir(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function il(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[Lr(n)].concat(Ir(n))),t}function vS(n,t){var e;t===void 0&&(t=!1);var i=x.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?x.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=x.util.squeezeShape(n);n=r.newShape}var a=x.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=Lr(n),o=2,l=2;return n.length&&(e=Ir(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),x.util.sizeToSquarishShape(a).map(function(u){return u*2})}return x.util.sizeToSquarishShape(a)}function rl(n){return n%2===0}function Ls(n,t){if(n=n.slice(-2),t=t.slice(-2),x.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(rl(e)&&rl(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&rl(n[0])&&rl(t[0])}var al,sl;function yS(n){if(al==null){var t=Gn(n);al=t.getParameter(t.MAX_TEXTURE_SIZE)}return al}function tM(){al=null}function nM(){sl=null}function bS(n){if(sl==null){var t=Gn(n);sl=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,sl)}function wS(n){if(n===0)return 0;var t,e=Gn(n);return on(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:on(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function on(n,t){var e=n.getExtension(t);return e!=null}function hp(n){try{var t=Gn(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function SS(n){if(n===0)return!1;var t=Gn(n);if(n===1){if(!on(t,"OES_texture_float"))return!1}else if(!on(t,"EXT_color_buffer_float"))return!1;var e=dp(t);return e}function LS(n){if(n===0)return!1;var t=Gn(n);if(n===1){if(!on(t,"OES_texture_float"))return!1;if(!on(t,"WEBGL_color_buffer_float"))return!1}else{if(on(t,"EXT_color_buffer_float"))return dp(t);var e="EXT_color_buffer_half_float";if(on(t,e)){var i=t.getExtension(e);return iM(t,i)}return!1}var r=dp(t);return r}function dp(n){var t=lp(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function iM(n,t){var e=lp(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function IS(n){if(n!==2)return!1;var t=Gn(n),e=t.fenceSync!=null;return e}function ma(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&x.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var rM={__proto__:null,callAndCheck:ce,canBeRepresented:tS,getWebGLErrorMessage:eS,getExtensionOrThrow:ws,createVertexShader:nS,createFragmentShader:iS,createProgram:rS,linkProgram:aS,validateProgram:tl,createStaticVertexBuffer:sS,createStaticIndexBuffer:oS,getNumChannels:Z_,createTexture:lS,validateTextureSize:uS,createFramebuffer:cS,bindVertexBufferToProgramAttribute:up,bindTextureUnit:dS,unbindTextureUnit:Q_,getProgramUniformLocationOrThrow:pS,getProgramUniformLocation:fS,bindTextureToProgramUniformSampler:mS,bindCanvasToFramebuffer:eM,bindColorTextureToFramebuffer:nl,unbindColorTextureFromFramebuffer:cp,validateFramebuffer:Ss,getFramebufferErrorMessage:gS,getBatchDim:Lr,getRowsCols:Ir,getShapeAs3D:il,getTextureShapeFromLogicalShape:vS,isReshapeFree:Ls,getWebGLMaxTextureSize:yS,resetMaxTextureSize:tM,resetMaxTexturesInShader:nM,getMaxTexturesInShader:bS,getWebGLDisjointQueryTimerVersion:wS,hasExtension:on,isWebGLVersionEnabled:hp,isCapableOfRenderingToFloatTexture:SS,isDownloadFloatTextureEnabled:LS,isWebGLFenceEnabled:IS,assertNotComplex:ma};var we=x.env();we.registerFlag("HAS_WEBGL",function(){return we.getNumber("WEBGL_VERSION")>0});we.registerFlag("WEBGL_VERSION",function(){return hp(2)?2:hp(1)?1:0});we.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});we.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return we.get("WEBGL_VERSION")===2});we.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});we.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});we.registerFlag("WEBGL_PACK",function(){return we.getBool("HAS_WEBGL")});we.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_CLIP",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});we.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_REDUCE",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_LAZILY_UNPACK",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_CONV_IM2COL",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return yS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return bS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=we.getNumber("WEBGL_VERSION");return n===0?0:wS(n)});we.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return we.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!x.device_util.isMobile()});we.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return SS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return we.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:we.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});we.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return LS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return IS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=we.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});we.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function aM(n){const t=new Float32Array(n.length);for(let e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}function pp(n){return(t,e,i,r,a)=>{const s=x.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=x.util.computeStrides(s),u=x.util.sizeFromShape(s),c=x.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=x.util.computeStrides(t),f=x.util.computeStrides(e),m=x.backend_util.getBroadcastDims(t,s),g=x.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(let v=0;v<c.length;++v){const b=x.util.indexToLoc(v,o,l),w=b.slice(-h);m.forEach(C=>w[C]=0);const S=x.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const N=x.util.locToIndex(L,d,f);c[v]=n(i[S],r[N])}return[c,s]}}const sM=pp((n,t)=>n+t);function ga(n){return(t,e,i)=>{const r=x.util.getTypedArrayFromDType(e,t.length);for(let a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}const oM=ga(n=>Math.ceil(n));const lM=ga(n=>Math.exp(n));const uM=ga(n=>Math.expm1(n));const cM=ga(n=>Math.floor(n));const hM=ga(n=>Math.log(n));function dM(n,t,e,i){const r=x.util.getTypedArrayFromDType(i,x.util.sizeFromShape(e));for(let a=0;a<r.length;++a){const s=a*t;let o=n[s];for(let l=0;l<t;++l){const u=n[s+l];u>o&&(o=u)}r[a]=o}return r}const pM=pp((n,t)=>n*t);const fM=ga(n=>1/Math.sqrt(n));function mM(n,t,e,i,r){const a=x.slice_util.isSliceContinous(i,t,e),s=x.util.sizeFromShape(e),o=x.util.computeStrides(i);if(a){const u=x.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=x.util.getTypedArrayFromDType(r,s);for(let u=0;u<s;++u){const c=e.length,h=x.util.computeStrides(e),d=x.util.indexToLoc(u,c,h),p=d.map((m,g)=>m+t[g]),f=x.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const gM=pp((n,t)=>n-t);function vM(n,t,e,i,r){const a=t.length,s=x.util.sizeFromShape(t),o=x.util.computeStrides(t),l=x.util.computeStrides(r),u=x.util.getTypedArrayFromDType(e,x.util.sizeFromShape(r));for(let c=0;c<s;++c){const h=x.util.indexToLoc(c,a,o),d=new Array(h.length);for(let f=0;f<d.length;f++)d[f]=h[i[f]];const p=x.util.locToIndex(d,a,l);u[p]=n[c]}return u}function yM(n,t,e,i){const r=x.util.parseAxisParam(t,e)[0],a=[1,e[0],1];for(let f=0;f<r;f++)a[0]*=e[f];a[1]=e[r];for(let f=r+1;f<e.length;f++)a[2]*=e[f];const s={},o=new Int32Array(e[r]),l=new x.TensorBuffer(a,i,n),u=[],c=a[0]===1&&a[2]===1;for(let f=0;f<e[r];f++){let m;if(c)m=n[f].toString();else{const g=[];for(let v=0;v<a[0];v++)for(let b=0;b<a[2];b++)g.push(l.get(v,f,b));m=g.join(",")}if(s[m]!==void 0)o[f]=s[m];else{const g=Object.keys(s).length;s[m]=g,o[f]=g,u.push(f)}}const h=a.slice();h[1]=Object.keys(s).length;const d=new x.TensorBuffer(h,i);u.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let v=0;v<a[2];v++)d.set(l.get(g,f,v),g,m,v)});const p=e.slice();return p[r]=h[1],{outputValues:d.values,outputShape:p,indices:o}}var bM=aM,wM=sM,SM=oM,LM=lM,IM=uM,AM=cM,TM=hM,NM=dM,xM=pM,CM=fM,RM=mM,OM=gM,AS=vM,EM=yM;var DM=function(){function n(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map(function(a,s){return"T"+s});var i=[];this.variableNames.forEach(function(a){i.push("float v"+a+" = get"+a+"AtOutCoords();")});var r=this.variableNames.map(function(a){return"v"+a}).join(" + ");this.userCode=`
       void main() {
         `+i.join(`
         `)+`
@@ -69,7 +69,7 @@ Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed thi
         }
         setOutput(float(bestIndex));
       }
-    `}return n}();function TS(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function en(n,t){return t===1?[n]:TS(n,t)}function WM(n,t){if(n===1)return"rc";for(var e="",i=0;i<n;i++)e+=t[i],i<n-1&&(e+=",");return e}function Bt(){var n,t,e,i,r,a,s,o,l,u;return x.env().getNumber("WEBGL_VERSION")===2?(n="#version 300 es",t="in",e="out",i="in",r="texture",a="outputColor",s="out vec4 outputColor;",o=`
+    `}return n}();function TS(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function Qt(n,t){return t===1?[n]:TS(n,t)}function WM(n,t){if(n===1)return"rc";for(var e="",i=0;i<n;i++)e+=t[i],i<n-1&&(e+=",");return e}function Bt(){var n,t,e,i,r,a,s,o,l,u;return x.env().getNumber("WEBGL_VERSION")===2?(n="#version 300 es",t="in",e="out",i="in",r="texture",a="outputColor",s="out vec4 outputColor;",o=`
       bool isnan_custom(float val) {
         return (val > 0.0 || val < 0.0) ? false : val != 0.0;
       }
@@ -114,7 +114,7 @@ Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed thi
       ivec4 round(vec4 value) {
         return ivec4(floor(value + vec4(0.5)));
       }
-    `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function Ar(n,t,e){e===void 0&&(e="index");var i=x.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function pp(n){var t=x.util.computeStrides(n).map(function(e){return e.toString()});return`
+    `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function Ar(n,t,e){e===void 0&&(e="index");var i=x.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function fp(n){var t=x.util.computeStrides(n).map(function(e){return e.toString()});return`
   int getFlatIndex(ivec3 coords) {
     return coords.x * `+t[0]+" + coords.y * "+t[1]+` + coords.z;
   }
@@ -715,7 +715,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
       `+d+`
       return get`+i+"("+f+`);
     }
-  `}function Ze(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function ba(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function wa(n,t){return t.map(function(e){return n[e]}).join(", ")}var y5=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,x.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Ze(l),c=en("coords",l),h,d;if(s===1){d=l+1;var p=Ze(d);h=`
+  `}function Ze(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function ba(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function wa(n,t){return t.map(function(e){return n[e]}).join(", ")}var y5=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,x.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Ze(l),c=Qt("coords",l),h,d;if(s===1){d=l+1;var p=Ze(d);h=`
         `+p+" sourceLocR = "+p+"("+c.join()+`, 0);
         ++`+c[l-1]+`;
         `+p+" sourceLocG = "+p+"("+c.join()+`, 0);
@@ -731,7 +731,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
         `+u+` sourceLocA = coords;
         --`+c[l-1]+`;
         `+u+` sourceLocB = coords;
-        --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=en("sourceLocR",d-1).concat("inIdx.r"),b=en("sourceLocG",d-1).concat("inIdx.g"),w=en("sourceLocB",d-1).concat("inIdx.b"),S=en("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",N=r?"":`
+        --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=Qt("sourceLocR",d-1).concat("inIdx.r"),b=Qt("sourceLocG",d-1).concat("inIdx.g"),w=Qt("sourceLocB",d-1).concat("inIdx.b"),S=Qt("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",N=r?"":`
           inIdx = round(vec4(getBestIndicesAChannel(`+v.join()+`),
                              getBestIndicesAChannel(`+b.join()+`),
                              getBestIndicesAChannel(`+w.join()+`),
@@ -891,7 +891,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
     `}return n}();var DS=`
   if (isnan(a)) return a;
   if (isnan(b)) return b;
-`,fp="return a + b;",mp="return a - b;",kS="return a * b;",S5=`
+`,mp="return a + b;",gp="return a - b;",kS="return a * b;",S5=`
   float s = sign(a) * sign(b);
   int ia = round(a);
   int ib = round(b);
@@ -1019,7 +1019,7 @@ return (round(mod(b, 2.0)) != 1) ?
             result.y = (coords + 1) >= `+this.outputShape[0]+` ? 0. : result.y;
             result.z = 0.;
             result.w = 0.;
-          `;else{var l=en("coords",a);s+=`
+          `;else{var l=Qt("coords",a);s+=`
             bool nextRowOutOfBounds =
               (`+l[a-2]+" + 1) >= "+this.outputShape[a-2]+`;
             bool nextColOutOfBounds =
@@ -1090,7 +1090,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+r.join(`
         `)+`
       }
-    `}return n}();var Q5=function(){function n(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=x.backend_util.computeOutShape(t,e);var i=this.outputShape,r=i.length,a=Ze(r),s=en("coords",r),o=["x","y","z","w","u","v"].slice(0,r);this.variableNames=t.map(function(v,b){return"T"+b});var l=new Array(t.length-1);l[0]=t[0][e];for(var u=1;u<l.length;u++)l[u]=l[u-1]+t[u][e];for(var c=o[e],h=o.slice(-2),d=o.join(),p="if ("+c+" < "+l[0]+`) {
+    `}return n}();var Q5=function(){function n(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=x.backend_util.computeOutShape(t,e);var i=this.outputShape,r=i.length,a=Ze(r),s=Qt("coords",r),o=["x","y","z","w","u","v"].slice(0,r);this.variableNames=t.map(function(v,b){return"T"+b});var l=new Array(t.length-1);l[0]=t[0][e];for(var u=1;u<l.length;u++)l[u]=l[u-1]+t[u][e];for(var c=o[e],h=o.slice(-2),d=o.join(),p="if ("+c+" < "+l[0]+`) {
         return getChannel(
             getT0(`+d+"), vec2("+h.join()+`));
         }`,u=1;u<l.length;u++){var f=l[u-1];p+=`
@@ -1965,14 +1965,14 @@ return (round(mod(b, 2.0)) != 1) ?
           float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;
           setOutput(val);
       }
-    `}return n}();var d9=function(){function n(t){this.variableNames=["A"],this.outTexUsage=on.DOWNLOAD;var e=Bt();this.outputShape=t,this.userCode=`
+    `}return n}();var d9=function(){function n(t){this.variableNames=["A"],this.outTexUsage=sn.DOWNLOAD;var e=Bt();this.outputShape=t,this.userCode=`
       `+NS+`
 
       void main() {
         float x = getAAtOutCoords();
         `+e.output+` = encode_float(x);
       }
-    `}return n}();var p9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=on.DOWNLOAD;var e=Bt();this.outputShape=t,this.userCode=`
+    `}return n}();var p9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=sn.DOWNLOAD;var e=Bt();this.outputShape=t,this.userCode=`
       `+NS+`
 
       void main() {
@@ -1981,7 +1981,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+e.output+` = encode_float(x);
       }
     `}return n}();var f9=function(){function n(t,e,i){i===void 0&&(i=!1),this.variableNames=["A"];var r=Bt(),a=e[0],s=e[1];this.outputShape=t;var o="result";i&&(o="floor(result * 255. + 0.5)"),this.userCode=`
-      `+pp(t)+`
+      `+fp(t)+`
 
       void main() {
         ivec3 coords = getOutputCoords();
@@ -2039,7 +2039,7 @@ return (round(mod(b, 2.0)) != 1) ?
             }
           }
         `}this.userCode=`
-      `+pp(t)+`
+      `+fp(t)+`
 
       void main() {
         ivec3 coords = getOutputCoords();
@@ -2118,7 +2118,7 @@ return (round(mod(b, 2.0)) != 1) ?
     void main() {
       gl_Position = vec4(clipSpacePos, 1);
       resultUV = uv;
-    }`;return nS(n,e)}function GS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return sS(n,t)}function YS(n){var t=new Uint16Array([0,1,2,2,1,3]);return oS(n,t)}function Is(n,t,e,i,r,a){uS(t,e);var s=lS(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function gp(n){return n.internalFormatFloat}function jS(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1];return Is(n,a,s,gp(i),i.textureFormatFloat,n.FLOAT)}function vp(n){return n.internalFormatHalfFloat}function KS(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1];return Is(n,a,s,vp(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function yp(n){return n.downloadTextureFormat}function $S(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1];return Is(n,a,s,yp(i),n.RGBA,n.UNSIGNED_BYTE)}function bp(n){return n.internalFormatPackedFloat}function XS(n,t,e,i){var r=fa(t,e),a=r[0],s=r[1];return Is(n,a,s,bp(i),n.RGBA,n.FLOAT)}function wp(n){return n.internalFormatPackedHalfFloat}function JS(n,t,e,i){var r=fa(t,e),a=r[0],s=r[1];return Is(n,a,s,wp(i),n.RGBA,i.textureTypeHalfFloat)}function ZS(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=lp(n,t,"clipSpacePos",e,3,a,i);return s&&lp(n,t,"uv",e,2,a,r)}function QS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function eL(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function tL(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function nL(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function iL(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array(G_(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function rL(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(Y_(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function aL(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var w9={__proto__:null,createVertexShader:qS,createVertexBuffer:GS,createIndexBuffer:YS,getInternalFormatForFloat32MatrixTexture:gp,createFloat32MatrixTexture:jS,getInternalFormatForFloat16MatrixTexture:vp,createFloat16MatrixTexture:KS,getInternalFormatForUnsignedBytesMatrixTexture:yp,createUnsignedBytesMatrixTexture:$S,getInternalFormatForPackedMatrixTexture:bp,createPackedMatrixTexture:XS,getInternalFormatForFloat16PackedMatrixTexture:wp,createFloat16PackedMatrixTexture:JS,bindVertexProgramAttributeStreams:ZS,uploadDenseMatrixToTexture:QS,uploadPixelDataToTexture:eL,createBufferFromOutputTexture:tL,downloadFloat32MatrixFromBuffer:nL,downloadByteEncodedFloatMatrixFromOutputTexture:iL,downloadPackedMatrixFromBuffer:rL,downloadMatrixFromPackedOutputTexture:aL};var sL=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=x.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,Q0(e,t)):this.gl=Gn(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(x.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ws(this.gl,a),ln(this.gl,s))this.textureHalfFloatExtension=ws(this.gl,s);else if(x.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),ln(this.gl,r))this.colorBufferHalfFloatExtension=ws(this.gl,r);else if(x.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",ln(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(ln(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=GS(this.gl),this.indexBuffer=YS(this.gl),this.framebuffer=cS(this.gl),this.textureConfig=op(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return x.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture.  This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),jS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),KS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),$S(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),eL(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),QS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),JS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),XS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(up(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return iL(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return rL(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return nL(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=tL(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(x.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return aL(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=iS(e,t),r=qS(e),a=rS(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),aS(e,a),this.debug&&tl(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=ZS(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&tl(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?pS(this.gl,t,e):fS(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),mS(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=fa(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&tl(this.gl,this.program),Ss(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ws(this.gl,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Qo(this,void 0,void 0,function(){var e=this;return el(this,function(i){switch(i.label){case 0:return[4,x.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=S9(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;x.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),nl(this.gl,t,this.framebuffer),this.debug&&Ss(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(nl(this.gl,this.outputTexture,this.framebuffer),this.debug&&Ss(this.gl)):up(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;nl(r,t,this.framebuffer),this.debug&&Ss(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function S9(n){for(var t=0;t<n.length;++t){var e=n[t]();if(!e)break}return t-1}function L9(n,t,e,i){var r=t.userCode,a=e.map(function(g,v){var b={logicalShape:g.shape,texShape:g.isUniform?null:g.texData.texShape,isUniform:g.isUniform,isPacked:g.isUniform?!1:g.texData.isPacked,flatOffset:null};return g.texData!=null&&g.texData.slice!=null&&g.texData.slice.flatOffset>0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=qM(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);x.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p<t.variableNames.length;p++){var f=t.variableNames[p],m=!1;d[f]=n.getUniformLocation(u,f,m),d["offset"+f]=n.getUniformLocation(u,"offset"+f,m)}return{program:t,source:l,webGLProgram:u,uniformLocations:d,inShapeInfos:s,outShapeInfo:o,infLoc:c,nanLoc:h}}function oL(n,t){if(n.length!==t.length)throw Error("Binary was compiled with "+n.length+" inputs, but "+("was executed with "+t.length+" inputs"));n.forEach(function(e,i){var r=e.logicalShape,a=t[i],s=a.shape;if(!x.util.arraysEqual(r,s))throw Error("Binary was compiled with different shapes than "+("the current args. Shapes "+r+" and "+s+" must match"));if(e.isUniform&&a.isUniform)return;var o=e.texShape,l=a.isUniform?null:a.texData.texShape;if(!x.util.arraysEqual(o,l))throw Error("Binary was compiled with different texture shapes than the"+(" current args. Shape "+o+" and "+l+" must match"))})}function I9(n,t,e,i,r){oL(t.inShapeInfos,e),oL([t.outShapeInfo],[i]);var a=i.texData.texture,s=i.texData.texShape;i.texData.isPacked?n.setOutputPackedMatrixTexture(a,s[0],s[1]):n.setOutputMatrixTexture(a,s[0],s[1]),n.setProgram(t.webGLProgram),x.env().getNumber("WEBGL_VERSION")===1&&(t.infLoc!==null&&n.gl.uniform1f(t.infLoc,Infinity)),t.nanLoc!==null&&n.gl.uniform1f(t.nanLoc,NaN),e.forEach(function(o,l){var u=t.program.variableNames[l],c=t.uniformLocations[u],h=t.uniformLocations["offset"+u];if(c==null)return;if(o.isUniform){if(x.util.sizeFromShape(o.shape)<2)n.gl.uniform1f(c,o.uniformValues[0]);else{var d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),n.gl.uniform1fv(c,d)}return}o.texData.slice!=null&&h!=null&&n.gl.uniform1i(h,o.texData.slice.flatOffset),n.setInputMatrixTexture(o.texData.texture,c,l)}),r!=null&&r(n,t.webGLProgram),n.executeProgram()}function A9(n,t,e){var i="";t.concat(e).forEach(function(s){var o=s.texData!=null&&s.texData.slice!=null&&s.texData.slice.flatOffset>0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var T9=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Bt(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var N=0;N<=1;N++)S+=`
+    }`;return nS(n,e)}function GS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return sS(n,t)}function YS(n){var t=new Uint16Array([0,1,2,2,1,3]);return oS(n,t)}function Is(n,t,e,i,r,a){uS(t,e);var s=lS(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function vp(n){return n.internalFormatFloat}function jS(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1];return Is(n,a,s,vp(i),i.textureFormatFloat,n.FLOAT)}function yp(n){return n.internalFormatHalfFloat}function KS(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1];return Is(n,a,s,yp(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function bp(n){return n.downloadTextureFormat}function $S(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1];return Is(n,a,s,bp(i),n.RGBA,n.UNSIGNED_BYTE)}function wp(n){return n.internalFormatPackedFloat}function XS(n,t,e,i){var r=fa(t,e),a=r[0],s=r[1];return Is(n,a,s,wp(i),n.RGBA,n.FLOAT)}function Sp(n){return n.internalFormatPackedHalfFloat}function JS(n,t,e,i){var r=fa(t,e),a=r[0],s=r[1];return Is(n,a,s,Sp(i),n.RGBA,i.textureTypeHalfFloat)}function ZS(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=up(n,t,"clipSpacePos",e,3,a,i);return s&&up(n,t,"uv",e,2,a,r)}function QS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function eL(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function tL(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function nL(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function iL(n,t,e,i){var r=ys(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array(G_(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function rL(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(Y_(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function aL(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var w9={__proto__:null,createVertexShader:qS,createVertexBuffer:GS,createIndexBuffer:YS,getInternalFormatForFloat32MatrixTexture:vp,createFloat32MatrixTexture:jS,getInternalFormatForFloat16MatrixTexture:yp,createFloat16MatrixTexture:KS,getInternalFormatForUnsignedBytesMatrixTexture:bp,createUnsignedBytesMatrixTexture:$S,getInternalFormatForPackedMatrixTexture:wp,createPackedMatrixTexture:XS,getInternalFormatForFloat16PackedMatrixTexture:Sp,createFloat16PackedMatrixTexture:JS,bindVertexProgramAttributeStreams:ZS,uploadDenseMatrixToTexture:QS,uploadPixelDataToTexture:eL,createBufferFromOutputTexture:tL,downloadFloat32MatrixFromBuffer:nL,downloadByteEncodedFloatMatrixFromOutputTexture:iL,downloadPackedMatrixFromBuffer:rL,downloadMatrixFromPackedOutputTexture:aL};var sL=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=x.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,Q0(e,t)):this.gl=Gn(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(x.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ws(this.gl,a),on(this.gl,s))this.textureHalfFloatExtension=ws(this.gl,s);else if(x.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),on(this.gl,r))this.colorBufferHalfFloatExtension=ws(this.gl,r);else if(x.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",on(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(on(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=GS(this.gl),this.indexBuffer=YS(this.gl),this.framebuffer=cS(this.gl),this.textureConfig=lp(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return x.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture.  This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),jS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),KS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),$S(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),eL(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),QS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),JS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),XS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(cp(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return iL(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return rL(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return nL(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=tL(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(x.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return aL(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=iS(e,t),r=qS(e),a=rS(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),aS(e,a),this.debug&&tl(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=ZS(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&tl(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?pS(this.gl,t,e):fS(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),mS(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=fa(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&tl(this.gl,this.program),Ss(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ws(this.gl,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Qo(this,void 0,void 0,function(){var e=this;return el(this,function(i){switch(i.label){case 0:return[4,x.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=S9(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;x.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),nl(this.gl,t,this.framebuffer),this.debug&&Ss(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(nl(this.gl,this.outputTexture,this.framebuffer),this.debug&&Ss(this.gl)):cp(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;nl(r,t,this.framebuffer),this.debug&&Ss(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function S9(n){for(var t=0;t<n.length;++t){var e=n[t]();if(!e)break}return t-1}function L9(n,t,e,i){var r=t.userCode,a=e.map(function(g,v){var b={logicalShape:g.shape,texShape:g.isUniform?null:g.texData.texShape,isUniform:g.isUniform,isPacked:g.isUniform?!1:g.texData.isPacked,flatOffset:null};return g.texData!=null&&g.texData.slice!=null&&g.texData.slice.flatOffset>0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=qM(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);x.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p<t.variableNames.length;p++){var f=t.variableNames[p],m=!1;d[f]=n.getUniformLocation(u,f,m),d["offset"+f]=n.getUniformLocation(u,"offset"+f,m)}return{program:t,source:l,webGLProgram:u,uniformLocations:d,inShapeInfos:s,outShapeInfo:o,infLoc:c,nanLoc:h}}function oL(n,t){if(n.length!==t.length)throw Error("Binary was compiled with "+n.length+" inputs, but "+("was executed with "+t.length+" inputs"));n.forEach(function(e,i){var r=e.logicalShape,a=t[i],s=a.shape;if(!x.util.arraysEqual(r,s))throw Error("Binary was compiled with different shapes than "+("the current args. Shapes "+r+" and "+s+" must match"));if(e.isUniform&&a.isUniform)return;var o=e.texShape,l=a.isUniform?null:a.texData.texShape;if(!x.util.arraysEqual(o,l))throw Error("Binary was compiled with different texture shapes than the"+(" current args. Shape "+o+" and "+l+" must match"))})}function I9(n,t,e,i,r){oL(t.inShapeInfos,e),oL([t.outShapeInfo],[i]);var a=i.texData.texture,s=i.texData.texShape;i.texData.isPacked?n.setOutputPackedMatrixTexture(a,s[0],s[1]):n.setOutputMatrixTexture(a,s[0],s[1]),n.setProgram(t.webGLProgram),x.env().getNumber("WEBGL_VERSION")===1&&(t.infLoc!==null&&n.gl.uniform1f(t.infLoc,Infinity)),t.nanLoc!==null&&n.gl.uniform1f(t.nanLoc,NaN),e.forEach(function(o,l){var u=t.program.variableNames[l],c=t.uniformLocations[u],h=t.uniformLocations["offset"+u];if(c==null)return;if(o.isUniform){if(x.util.sizeFromShape(o.shape)<2)n.gl.uniform1f(c,o.uniformValues[0]);else{var d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),n.gl.uniform1fv(c,d)}return}o.texData.slice!=null&&h!=null&&n.gl.uniform1i(h,o.texData.slice.flatOffset),n.setInputMatrixTexture(o.texData.texture,c,l)}),r!=null&&r(n,t.webGLProgram),n.executeProgram()}function A9(n,t,e){var i="";t.concat(e).forEach(function(s){var o=s.texData!=null&&s.texData.slice!=null&&s.texData.slice.flatOffset>0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var T9=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Bt(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var N=0;N<=1;N++)S+=`
           blockIndex = rc.y + `+N+`;
           pos = rc.x + `+L+`;
 
@@ -2408,7 +2408,7 @@ return (round(mod(b, 2.0)) != 1) ?
         }
         setOutput(dotProd);
       }
-    `}return n}();var Sp=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) {
+    `}return n}();var Lp=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) {
           vec4 b = getPreluActivationWeightsAtOutCoords();
           `+s+`
         }`:f=`vec4 activation(vec4 x) {
@@ -2475,7 +2475,7 @@ return (round(mod(b, 2.0)) != 1) ?
         void main() {
           setOutput(vec4(getA(), 0., 0., 0.));
         }
-      `;else{var i=en("rc",e),r=Ze(e),a=k9(e,t,i),s=F9(e,t[t.length-1],t[t.length-2],i),o=W9(t,i);this.userCode=`
+      `;else{var i=Qt("rc",e),r=Ze(e),a=k9(e,t,i),s=F9(e,t[t.length-1],t[t.length-2],i),o=W9(t,i);this.userCode=`
         void main() {
           `+r+` rc = getOutputCoords();
 
@@ -2525,7 +2525,7 @@ return (round(mod(b, 2.0)) != 1) ?
           setOutput(getX(`+l+`));
         }
       }
-    `}return n}();var P9=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Ze(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=en("rc",r),u=en("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1;
+    `}return n}();var P9=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Ze(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=Qt("rc",r),u=Qt("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1;
        if(`+c+`) {
       `,r===1?"":`}
        rc = outputLoc;
@@ -2690,7 +2690,7 @@ return (round(mod(b, 2.0)) != 1) ?
         }
         setOutput(`+L+`);
       }
-    `}return n}(),Lp=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=`
+    `}return n}(),Ip=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=`
         const ivec3 strides =
             ivec3(`+o+", "+l+", "+u+`);
         const ivec3 pads = ivec3(`+g+", "+v+", "+b+`);
@@ -2951,7 +2951,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+(r>0?"}":"")+`
       `}this.userCode=`
       `+_9(e)+`
-      `+pp(t)+`
+      `+fp(t)+`
 
       void main() {
         ivec3 rc = getOutputCoords();
@@ -3265,7 +3265,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+s+` coords = getOutputCoords();
         setOutput(getX(`+a+`));
       }
-    `}return n}();var j9=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=en("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Ze(i);i===1?this.userCode=`
+    `}return n}();var j9=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=Qt("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Ze(i);i===1?this.userCode=`
         void main(){
           int rc = getOutputCoords();
           vec4 result = vec4(0.);
@@ -3429,7 +3429,7 @@ return (round(mod(b, 2.0)) != 1) ?
           setOutput(getB(`+a+`));
         }
       }
-    `}return n}();var J9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i="uniform int start["+this.rank+"];",r=X9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+Ip[l]+" = start["+l+"] + coords."+Ip[l]+";"});a=`
+    `}return n}();var J9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i="uniform int start["+this.rank+"];",r=X9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+Ap[l]+" = start["+l+"] + coords."+Ap[l]+";"});a=`
         `+e+` sourceLoc;
         `+e+` coords = getOutputCoords();
         `+s.join(`
@@ -3440,7 +3440,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+a+`
         setOutput(getSource(`+r+`));
       }
-    `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),Ip=["x","y","z","w","u","v"];function X9(n){if(n===1)return"sourceLoc";if(n<=6)return Ip.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var Z9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i=en("coords",this.rank),r=en("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=`
+    `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),Ap=["x","y","z","w","u","v"];function X9(n){if(n===1)return"sourceLoc";if(n<=6)return Ap.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var Z9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i=Qt("coords",this.rank),r=Qt("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=`
       result.x = `+s+`;
       if (++`+i[this.rank-1]+" < "+t[this.rank-1]+`) {
         ++`+r[this.rank-1]+`;
@@ -3478,7 +3478,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+s+` coords = getOutputCoords();
         setOutput(getX(`+o+`));
       }
-    `}return n}();var e6=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=dL(e,i),a=pL(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=hL(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Ot.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Ot.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Ot.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Ot.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Ot.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=dL(i,r),s=pL(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=hL(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=x.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function t6(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function hL(n,t,e,i,r){var a=n6(t,i),s;if(r){var o=fa(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=ys(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=t6(e,a);return s*p}function n6(n,t){switch(n){case Ot.PACKED_2X2_FLOAT32:return bp(t);case Ot.PACKED_2X2_FLOAT16:return wp(t);case Ot.UNPACKED_FLOAT32:return gp(t);case Ot.UNPACKED_FLOAT16:return vp(t);case Ot.PACKED_4X1_UNSIGNED_BYTE:return yp(t);default:throw new Error("Unknown physical texture type "+n)}}function i6(n){return x.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Ot.PACKED_2X2_FLOAT32:Ot.UNPACKED_FLOAT32:n?Ot.PACKED_2X2_FLOAT16:Ot.UNPACKED_FLOAT16}function dL(n,t){if(n===on.UPLOAD)return Ot.PACKED_2X2_FLOAT32;if(n===on.RENDER||n==null)return i6(t);if(n===on.DOWNLOAD||n===on.PIXELS)return Ot.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function pL(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var a6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[r]*e[r];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=r6(t);this.userCode=`
+    `}return n}();var e6=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=dL(e,i),a=pL(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=hL(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Ot.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Ot.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Ot.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Ot.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Ot.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=dL(i,r),s=pL(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=hL(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=x.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function t6(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function hL(n,t,e,i,r){var a=n6(t,i),s;if(r){var o=fa(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=ys(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=t6(e,a);return s*p}function n6(n,t){switch(n){case Ot.PACKED_2X2_FLOAT32:return wp(t);case Ot.PACKED_2X2_FLOAT16:return Sp(t);case Ot.UNPACKED_FLOAT32:return vp(t);case Ot.UNPACKED_FLOAT16:return yp(t);case Ot.PACKED_4X1_UNSIGNED_BYTE:return bp(t);default:throw new Error("Unknown physical texture type "+n)}}function i6(n){return x.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Ot.PACKED_2X2_FLOAT32:Ot.UNPACKED_FLOAT32:n?Ot.PACKED_2X2_FLOAT16:Ot.UNPACKED_FLOAT16}function dL(n,t){if(n===sn.UPLOAD)return Ot.PACKED_2X2_FLOAT32;if(n===sn.RENDER||n==null)return i6(t);if(n===sn.DOWNLOAD||n===sn.PIXELS)return Ot.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function pL(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var a6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[r]*e[r];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=r6(t);this.userCode=`
       void main() {
         `+a+` resRC = getOutputCoords();
         setOutput(getA(`+s+`));
@@ -3634,14 +3634,14 @@ return (round(mod(b, 2.0)) != 1) ?
 
         setOutput(y);
       }
-    `}return n}();var W6=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=en("rc",e),r=Ze(e),a=WM(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=`
+    `}return n}();var W6=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=Qt("rc",e),r=Ze(e),a=WM(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=`
       void main() {
         `+r+` rc = getOutputCoords();
         vec4 packedInput = getA(`+a+`);
 
         setOutput(getChannel(packedInput, `+o+`));
       }
-    `}return n}();var NL=x.backend_util.segment_util,U6=x.kernel_impls.split,B6=x.kernel_impls.tile,z6=x.kernel_impls.topkImpl,P6=x.kernel_impls.whereImpl,_6=1e-7,M6=1e-4,cl={};function H6(n){return n in cl||(cl[n]={}),cl[n]}function hl(n,t){if(t===void 0&&(t=!1),n==="linear")return t?k6:s6;if(n==="relu")return t?IL:mL;if(n==="elu")return t?TL:vL;if(n==="relu6")return t?AL:gL;if(n==="prelu")return t?WS:FS;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var V6=128,q6=600;function G6(){return x.env().global.screen==null?1024:x.env().global.screen.height*x.env().global.screen.width*window.devicePixelRatio*q6/1024/1024}var xL=1e3,CL=function(n){H_(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!x.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=Gn(x.env().getNumber("WEBGL_VERSION"));i.binaryCache=H6(x.env().getNumber("WEBGL_VERSION")),i.gpgpu=new sL(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new e6(i.gpgpu),i.numMBBeforeWarning=G6(),i.texData=new x.DataStorage(i,x.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((x.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||x.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:on.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(x.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:on.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new Ts(l,ul):c=new Re(l,ul);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=x.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=x.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=x.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Qo(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R;return el(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new Ts(s,ul):h=new Re(s,ul),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!x.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&x.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&x.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(bs(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=x.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=x.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),N=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(N)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,N]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i<e.length;i++){var r=e[i];if(!tS(r))throw x.env().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error("The value "+r+" cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'"):Error("The value "+r+" cannot be represented on this device.")}},t.prototype.getValuesFromTexture=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.isPacked,l=x.util.sizeFromShape(a);if(x.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){var u=this.decode(e),c=this.texData.get(u.dataId),h=(i=this.gpgpu).downloadMatrixFromPackedTexture.apply(i,[c.texture].concat(bs(a))).subarray(0,l);return this.disposeIntermediateTensorInfo(u),h}var d=x.env().getBool("WEBGL_PACK")&&o===!0,p=d?il(a):a,f=d?new p9(p):new d9(p),m=this.runWebGLProgram(f,[{shape:p,dtype:s,dataId:e}],"float32"),g=this.texData.get(m.dataId),v=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(g.texture,g.texShape[0],g.texShape[1]).subarray(0,l);return this.disposeIntermediateTensorInfo(m),v},t.prototype.time=function(e){return Qo(this,void 0,void 0,function(){var i,r,a,s,o,l,u;return el(this,function(c){switch(c.label){case 0:return i=this.activeTimers,r=[],a=!1,this.programTimersStack==null?(this.programTimersStack=r,a=!0):this.activeTimers.push(r),this.activeTimers=r,e(),s=x.util.flatten(this.activeTimers.map(function(h){return h.query})).filter(function(h){return h!=null}),o=x.util.flatten(this.activeTimers.map(function(h){return h.name})).filter(function(h){return h!=null}),this.activeTimers=i,a&&(this.programTimersStack=null),l={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=x.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:x.util.now(),endMs:null}},t.prototype.endTimer=function(e){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=x.util.now(),e)},t.prototype.getQueryTime=function(e){return Qo(this,void 0,void 0,function(){var i;return el(this,function(r){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return x.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=x.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=V6);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&x.util.sizeFromShape(s.shape)<i})},t.prototype.getGPGPUContext=function(){return this.gpgpu},t.prototype.complex=function(e,i){var r=this.makeOutput(e.shape,"complex64"),a=this.texData.get(r.dataId);return a.complexTensors={real:x.engine().keep(e.clone()),imag:x.engine().keep(i.clone())},r},t.prototype.real=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.real.clone()},t.prototype.imag=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.imag.clone()},t.prototype.slice=function(e,i,r){if(this.shouldExecuteOnCPU([e])){var a=RM(this.texData.get(e.dataId).values,i,r,e.shape,e.dtype);return this.makeOutput(r,e.dtype,a)}if(x.util.sizeFromShape(r)===0)return x.tensor([],r,e.dtype);var s=this.texData.get(e.dataId).isPacked,o=x.slice_util.isSliceContinous(e.shape,i,r);if(s||!o){var l=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new Z9(r):new J9(r),u=l.getCustomSetupFunc(i);return this.compileAndRun(l,[e],null,u)}return this.uploadToGPU(e.dataId),this.shallowSlice(e,i,r)},t.prototype.shallowSlice=function(e,i,r){var a=this.texData.get(e.dataId),s=this.makeOutput(r,e.dtype),o=this.texData.get(s.dataId);Object.assign(o,a),o.shape=r,o.dtype=e.dtype;var l=x.slice_util.computeFlatOffset(i,e.strides);a.slice&&(l+=a.slice.flatOffset),o.slice={flatOffset:l,origDataId:a.slice&&a.slice.origDataId||e.dataId};var u=this.dataRefCount.get(o.slice.origDataId)||1;return this.dataRefCount.set(o.slice.origDataId,u+1),s},t.prototype.stridedSlice=function(e,i,r,a){var s=this,o=this.tryRunOnCpuOrThrow([e],function(){return s.cpuBackend.stridedSlice(e,i,r,a)});if(o)return o;var l=x.slice_util.computeOutShape(i,r,a);if(l.some(function(c){return c===0}))return x.tensor([],l);var u=new Q9(i,a,l);return this.compileAndRun(u,[e])},t.prototype.reverse=function(e,i){var r=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new j9(e.shape,i):new Y9(e.shape,i);return this.compileAndRun(r,[e])},t.prototype.concat=function(e,i){if(e[0].dtype==="complex64"){var r=e.map(function(f){return x.real(f)}),a=e.map(function(f){return x.imag(f)});return x.complex(this.concat(r,i),this.concat(a,i))}if(e.length===1)return e[0];if(e.length>x.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new Q5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=x.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,x.util.sizeFromShape(f.shape.slice(i)))}),d=new Z5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,yL,e.dtype);var a=new Re(e.shape,yL);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>xL){r&&(e=x.transpose(e,[0,2,1])),a&&(i=x.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=x.upcastType(e.dtype,i.dtype),m=new Sp(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=x.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?hl(l,!0):null,b=new Sp(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new ES(OS.REAL,e.shape,i.shape),o=new ES(OS.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=xM(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,kS,e.dtype);var g=new St(kS,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=x.env().getBool("WEBGL_PACK_NORMALIZATION")?new C9(e.shape,i,r,a,s):new N9(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new x9(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return x.util.decodeString(l)}),s=x.buffer(e.shape,e.dtype,a);return B6(s,i)}var o=new a6(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new P9(e.shape,i,r):new z9(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new y9(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){x.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=x.backend_util.getReshaped(e.shape,i,a),o=x.backend_util.getPermuted(s.length,i.length),l=x.backend_util.getReshapedPermuted(e.shape,i,a),u=x.backend_util.getSliceBeginCoords(r,i.length),c=x.backend_util.getSliceSize(l,r,i.length);return x.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){x.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;o<e.shape.length;++o)s.push([0,0]);var l=e.pad(s),u=x.backend_util.getReshaped(l.shape,i,a,!1),c=x.backend_util.getPermuted(u.length,i.length,!1),h=x.backend_util.getReshapedPermuted(l.shape,i,a,!1),d=x.transpose(l.reshape(u),c);return x.reshape(d,h)},t.prototype.reduce=function(e,i,r){var a=e.shape[0],s=e.shape[1],o=x.backend_util.computeOptimalWindowSize(s),l=Math.ceil(s/o),u={windowSize:o,inSize:s,batchSize:a,outSize:l},c=new lL(u,i),h=this.compileAndRun(c,[e],r);return h.shape[1]===1?h:this.reduce(h,i,r)},t.prototype.argReduce=function(e,i,r){r===void 0&&(r=null);var a=e.shape[0],s=e.shape[1];r!=null&&(a=r.shape[0],s=r.shape[1]);var o=x.backend_util.computeOptimalWindowSize(s),l={windowSize:o,inSize:s,batchSize:a,outSize:Math.ceil(s/o)},u=new FM(l,i,r==null),c=[e];r!=null&&c.push(r);var h=this.compileAndRun(u,c,"int32");return h.shape[1]===1?h:this.argReduce(e,i,h)},t.prototype.argReducePacked=function(e,i,r){r===void 0&&(r=null);var a=r!=null?r.shape:e.shape,s=a[a.length-1],o=x.backend_util.computeOptimalWindowSize(s),l=new y5(a,o,i,r==null),u=r==null?[e]:[e,r],c=this.compileAndRun(l,u,"int32");return c.rank===e.rank?this.argReducePacked(e,i,c):c},t.prototype.sum=function(e,i){x.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o),u=x.sumOutType(e.dtype);return this.reduce(l,"sum",u).reshape(a)},t.prototype.prod=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e],function(){return r.cpuBackend.prod(e,i)});if(a)return a;var s=x.backend_util.computeOutAndReduceShapes(e.shape,i),o=s[0],l=s[1],u=x.util.sizeFromShape(l),c=e.as2D(-1,u),h=x.sumOutType(e.dtype);return this.reduce(c,"prod",h).reshape(o)},t.prototype.unsortedSegmentSum=function(e,i,r){var a=0,s=x.backend_util.getAxesPermutation([a],e.rank),o=e;s!=null&&(o=x.transpose(e,s),a=x.backend_util.getInnerMostAxes(1,e.rank)[0]);var l=NL.computeOutShape(o.shape,a,r),u=x.util.sizeFromShape([o.shape[a]]),c=o.as2D(-1,u),h=x.sumOutType(e.dtype),d=this.segOpCompute(c,"unsortedSegmentSum",i,h,r).reshape(l);return s!=null&&(d=x.transpose(d,x.backend_util.getUndoAxesPermutation(s))),d},t.prototype.segOpCompute=function(e,i,r,a,s){var o=e.shape[0],l=e.shape[1],u=NL.segOpComputeOptimalWindowSize(l,s),c={windowSize:u,inSize:l,batchSize:o,numSegments:s},h=new K9(c,i),d=this.compileAndRun(h,[e,r],a);return d.shape[1]===s?d:(r=x.range(0,s).tile([l/u]),this.segOpCompute(d,i,r,a,s))},t.prototype.argMinMaxReduce=function(e,i,r){var a=[i];if(x.backend_util.assertAxesAreInnerMostDims("arg"+r.charAt(0).toUpperCase()+r.slice(1),a,e.rank),!x.env().getBool("WEBGL_PACK_REDUCE")||e.rank<=2){var s=x.backend_util.computeOutAndReduceShapes(e.shape,a),o=s[0],l=s[1],u=x.util.sizeFromShape(l),c=e.as2D(-1,u);return this.argReduce(c,r).reshape(o)}return this.argReducePacked(e,r)},t.prototype.argMin=function(e,i){return this.argMinMaxReduce(e,i,"min")},t.prototype.argMax=function(e,i){return this.argMinMaxReduce(e,i,"max")},t.prototype.cumsum=function(e,i,r,a){if(i!==e.rank-1)throw new Error("WebGL cumsum shader expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=e.shape[i],o=e,l=0;l<=Math.ceil(Math.log2(s))-1;l++){var u=new MS(e.shape,!1,a),c=u.getCustomSetupFunc(l),h=o;o=this.compileAndRun(u,[o],o.dtype,c),h.dispose()}if(r){var u=new MS(e.shape,r,a),h=o;o=this.compileAndRun(u,[o]),h.dispose()}return o},t.prototype.equal=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,z5,"bool");var r=new St(I5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.notEqual=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,P5,"bool");var r=new St(A5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.less=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.less(e,i)});if(a)return a;if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,_5,"bool");var s=new St(T5,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.lessEqual=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,M5,"bool");var r=new St(N5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.greater=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.greater(e,i)});if(a)return a;if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,H5,"bool");var s=new St(x5,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.greaterEqual=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,V5,"bool");var r=new St(C5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalNot=function(e){var i=new Re(e.shape,E6);return this.compileAndRun(i,[e])},t.prototype.logicalAnd=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,q5,"bool");var r=new St(R5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalOr=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,G5,"bool");var r=new St(O5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.select=function(e,i,r){var a=new $9(e.rank,i.shape,i.rank);return this.compileAndRun(a,[e,i,r],x.upcastType(i.dtype,r.dtype))},t.prototype.where=function(e){x.backend_util.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");var i=e.dataSync();return P6(e.shape,i)},t.prototype.topk=function(e,i,r){var a=e.dataSync();return z6(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){x.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"min",l.dtype).reshape(a)},t.prototype.minimum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.minimum(e,i)});if(a)return a;var s=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(j5,e.shape,i.shape):new St(D5,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.mod=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(K5,e.shape,i.shape):new St(k5,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.maximum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.maximum(e,i)});if(a)return a;var s=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(Y5,e.shape,i.shape):new St(E5,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.all=function(e,i){x.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"all",l.dtype).reshape(a)},t.prototype.any=function(e,i){x.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"any",l.dtype).reshape(a)},t.prototype.floorDiv=function(e,i){var r=S5,a="int32";if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,W5,a);var s=new St(r,e.shape,i.shape);return this.compileAndRun(s,[e,i],a)},t.prototype.add=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,fp);var r=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=wM(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,fp,r);var c=new St(fp,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.packedUnaryOp=function(e,i,r){var a=new Ts(e.shape,i);return this.compileAndRun(a,[e],r)},t.prototype.packedBinaryOp=function(e,i,r,a,s){s===void 0&&(s=!1);var o=new Ui(r,e.shape,i.shape,s);return this.compileAndRun(o,[e,i],a)},t.prototype.complexSeparableBinaryOp=function(e,i,r){var a=this,s=this.texData.get(e.dataId),o=this.texData.get(i.dataId),l=[[s.complexTensors.real,o.complexTensors.real],[s.complexTensors.imag,o.complexTensors.imag]].map(function(d){var p=d[0],f=d[1],m=a.makeComplexComponentTensorInfo(e,p),g=a.makeComplexComponentTensorInfo(i,f),v=new St(r,e.shape,i.shape);return a.compileAndRun(v,[m,g],x.upcastType(p.dtype,f.dtype))}),u=l[0],c=l[1],h=this.complex(u,c);return u.dispose(),c.dispose(),h},t.prototype.makeComplexComponentTensorInfo=function(e,i){return{dataId:i.dataId,dtype:i.dtype,shape:e.shape}},t.prototype.addN=function(e){if(e.length===1)return e[0];if(e.length>x.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return x.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=x.env().getBool("WEBGL_PACK"),u=l?new kM(e[0].shape,o):new DM(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,mp);var r=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=OM(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,mp,e.dtype);var c=new St(mp,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new Ui(U5,e.shape,i.shape):new St(L5,e.shape,i.shape),s=x.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=SM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,bL,e.dtype);var r=new Re(e.shape,bL);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=AM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,wL,e.dtype);var r=new Re(e.shape,wL);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,u6);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,c6);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,h6);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,d6);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,p6);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=LM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,SL,e.dtype);var r=new Re(e.shape,SL);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=IM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,LL,e.dtype);var r=new Re(e.shape,LL);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=x.util.parseAxisParam([i],e.shape),a=x.max(e,r),s=x.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return x.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=TM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,F6,e.dtype);var r=new Re(e.shape,f6);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,m6);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,g6);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=CM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,v6);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,O6);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return x.env().getBool("WEBGL_PACK")?i=new Ts(e.shape,IL):i=new Re(e.shape,mL),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return x.env().getBool("WEBGL_PACK")?i=new Ts(e.shape,AL):i=new Re(e.shape,gL),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(WS,e.shape,i.shape):new St(FS,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,TL,e.dtype);var i=new Re(e.shape,vL);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(B5,e.shape,i.shape):new St(F5,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,o6);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,D6);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;x.env().getBool("WEBGL_PACK_CLIP")?a=new X5(e.shape):a=new $5(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=bM(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,fL,e.dtype);var r=new Re(e.shape,fL);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new J5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,y6);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,b6);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,w6);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,S6);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,L6);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,I6);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,A6);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,T6);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,N6);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,x6);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,C6);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,R6);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,l6(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>xL,v=l[2]%2!==0&&!!u.isPacked;if(g||!x.env().getBool("WEBGL_LAZILY_UNPACK")||!x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=x.reshape(e,[1,b,r.inChannels]),S=x.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return x.reshape(L,r.outShape)}var N=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,N,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,x.util.assert(Ls(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=x.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return x.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,x.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),N=new T9(v,S.shape,r),C=this.compileAndRun(N,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?hl(s,!0):null,W=new Sp(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(x.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?hl(o,!1):null,d=new US(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(x.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new US(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new t9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new e9(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=x.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?hl(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new zS(a,d,c,p),this.compileAndRun(f,h)):(f=new BS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return x.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new zS(r),this.compileAndRun(a,[e,i])):(a=new BS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new a9(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new r9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new s9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new i9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new n9(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return x.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=this.slice(e,l,u).reshape(a);return c},t.prototype.avgPool3d=function(e,i){var r=new Lp(i,"avg",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.avgPool3dBackprop=function(e,i,r){var a=new w5(r);return this.compileAndRun(a,[e],i.dtype)},t.prototype.maxPool3d=function(e,i){var r=new Lp(i,"max",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.maxPool3dBackprop=function(e,i,r,a){var s=!0,o=new Lp(a,"max",s),l=this.compileAndRun(o,[i]),u=new O9(a),c=this.compileAndRun(u,[e,l],i.dtype);return l.dispose(),c},t.prototype.resizeBilinear=function(e,i,r,a){var s=x.env().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new V9(e.shape,i,r,a):new H9(e.shape,i,r,a);return this.compileAndRun(s,[e],"float32")},t.prototype.resizeBilinearBackprop=function(e,i,r){var a=new M9(e,i,r);return this.compileAndRun(a,[e])},t.prototype.resizeNearestNeighbor=function(e,i,r,a){var s=new G9(e.shape,i,r,a);return this.compileAndRun(s,[e])},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){var a=new q9(e,i,r);return this.compileAndRun(a,[e])},t.prototype.multinomial=function(e,i,r,a){var s=i?e:x.softmax(e),o=s.shape[0],l=s.shape[1],u=new E9(o,l,r),c=u.getCustomSetupFunc(a);return this.compileAndRun(u,[s],"int32",c)},t.prototype.oneHot=function(e,i,r,a){var s=new D9(e.size,i,r,a);return this.compileAndRun(s,[e])},t.prototype.diag=function(e){var i=new h9(e.size);return this.compileAndRun(i,[e])},t.prototype.cropAndResize=function(e,i,r,a,s,o){var l=new o9(e.shape,i.shape,a,s,o);return this.compileAndRun(l,[e,i,r],"float32")},t.prototype.depthToSpace=function(e,i,r){x.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new c9(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return U6(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=x.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return x.backend_util.reshapeTensor(x.tensor([]),r);var f=x.scalar(0),m=new cL(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=x.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new cL(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new VS(HS.REAL,e.shape,i),s=new VS(HS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=x.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new b9(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||x.util.inferDtype(i),r==="string"){var a=x.util.getArrayFromDType(r,x.util.sizeFromShape(e));return a.fill(i),x.engine().makeTensor(a,e,r,this)}else{var s=new g9(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return x.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return x.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new W6(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new U9(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[Lr(e.shape)].concat(Ir(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[Lr(i)].concat(Ir(i)),o=new uL(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=il(a),l;r?l=new u9(o):l=new l9(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===vs.DENSE){var c=bs(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),x.util.sizeFromShape(l.shape)===0)return u.values=x.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&x.util.sizeFromShape(w.shape)<=x.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!Ls(S.shape,w.shape)){var L=w,N=w.shape;w.shape=S.shape,w=o.packedReshape(w,N),h.push(w),S=o.texData.get(w.dataId),L.shape=N}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=A9(e,d,p),m=this.getAndSaveBinary(f,function(){return L9(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),I9(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!x.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return x.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!x.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=x.tidy(function(){if(!x.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=x.env().getBool("DEBUG");x.env().set("DEBUG",!1);var r=e.abs(x.scalar(1e-8)).dataSync()[0];if(x.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?_6:M6},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=x.util.now());var p=r.texShape;if(p==null&&(p=vS(a,c),r.texShape=p),o!=null){var f=il(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=fa(p[0],p[1]),g=i[0],v=i[1],m=new m9(f,[v,g],b)):m=new f9(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=on.PIXELS:this.texData.get(w.dataId).usage=on.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),N=this.texData.get(L.dataId);r.texture=N.texture,r.texShape=N.texShape,r.isPacked=N.isPacked,r.usage=N.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=x.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=Y6(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*x.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(x.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(x.KernelBackend);function Y6(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i<e.length;++i)e[i]=Math.round(n[i]);return e}else throw new Error("Unknown dtype "+t)}var j6="2.6.0";function RL(){x.env().set("WEBGL_FORCE_F16_TEXTURES",!0)}x.device_util.isBrowser()&&x.registerBackend("webgl",function(){return new CL},2);var K6={forceHalfFloat:RL},OL="if (isnan(x)) return x;",$6=`
+    `}return n}();var NL=x.backend_util.segment_util,U6=x.kernel_impls.split,B6=x.kernel_impls.tile,z6=x.kernel_impls.topkImpl,P6=x.kernel_impls.whereImpl,_6=1e-7,M6=1e-4,cl={};function H6(n){return n in cl||(cl[n]={}),cl[n]}function hl(n,t){if(t===void 0&&(t=!1),n==="linear")return t?k6:s6;if(n==="relu")return t?IL:mL;if(n==="elu")return t?TL:vL;if(n==="relu6")return t?AL:gL;if(n==="prelu")return t?WS:FS;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var V6=128,q6=600;function G6(){return x.env().global.screen==null?1024:x.env().global.screen.height*x.env().global.screen.width*window.devicePixelRatio*q6/1024/1024}var xL=1e3,CL=function(n){H_(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!x.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=Gn(x.env().getNumber("WEBGL_VERSION"));i.binaryCache=H6(x.env().getNumber("WEBGL_VERSION")),i.gpgpu=new sL(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new e6(i.gpgpu),i.numMBBeforeWarning=G6(),i.texData=new x.DataStorage(i,x.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((x.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||x.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:sn.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(x.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:sn.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new Ts(l,ul):c=new Re(l,ul);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=x.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=x.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=x.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Qo(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R;return el(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new Ts(s,ul):h=new Re(s,ul),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!x.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&x.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&x.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(bs(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=x.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=x.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),N=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(N)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,N]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i<e.length;i++){var r=e[i];if(!tS(r))throw x.env().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error("The value "+r+" cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'"):Error("The value "+r+" cannot be represented on this device.")}},t.prototype.getValuesFromTexture=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.isPacked,l=x.util.sizeFromShape(a);if(x.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){var u=this.decode(e),c=this.texData.get(u.dataId),h=(i=this.gpgpu).downloadMatrixFromPackedTexture.apply(i,[c.texture].concat(bs(a))).subarray(0,l);return this.disposeIntermediateTensorInfo(u),h}var d=x.env().getBool("WEBGL_PACK")&&o===!0,p=d?il(a):a,f=d?new p9(p):new d9(p),m=this.runWebGLProgram(f,[{shape:p,dtype:s,dataId:e}],"float32"),g=this.texData.get(m.dataId),v=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(g.texture,g.texShape[0],g.texShape[1]).subarray(0,l);return this.disposeIntermediateTensorInfo(m),v},t.prototype.time=function(e){return Qo(this,void 0,void 0,function(){var i,r,a,s,o,l,u;return el(this,function(c){switch(c.label){case 0:return i=this.activeTimers,r=[],a=!1,this.programTimersStack==null?(this.programTimersStack=r,a=!0):this.activeTimers.push(r),this.activeTimers=r,e(),s=x.util.flatten(this.activeTimers.map(function(h){return h.query})).filter(function(h){return h!=null}),o=x.util.flatten(this.activeTimers.map(function(h){return h.name})).filter(function(h){return h!=null}),this.activeTimers=i,a&&(this.programTimersStack=null),l={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=x.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:x.util.now(),endMs:null}},t.prototype.endTimer=function(e){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=x.util.now(),e)},t.prototype.getQueryTime=function(e){return Qo(this,void 0,void 0,function(){var i;return el(this,function(r){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return x.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=x.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=V6);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&x.util.sizeFromShape(s.shape)<i})},t.prototype.getGPGPUContext=function(){return this.gpgpu},t.prototype.complex=function(e,i){var r=this.makeOutput(e.shape,"complex64"),a=this.texData.get(r.dataId);return a.complexTensors={real:x.engine().keep(e.clone()),imag:x.engine().keep(i.clone())},r},t.prototype.real=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.real.clone()},t.prototype.imag=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.imag.clone()},t.prototype.slice=function(e,i,r){if(this.shouldExecuteOnCPU([e])){var a=RM(this.texData.get(e.dataId).values,i,r,e.shape,e.dtype);return this.makeOutput(r,e.dtype,a)}if(x.util.sizeFromShape(r)===0)return x.tensor([],r,e.dtype);var s=this.texData.get(e.dataId).isPacked,o=x.slice_util.isSliceContinous(e.shape,i,r);if(s||!o){var l=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new Z9(r):new J9(r),u=l.getCustomSetupFunc(i);return this.compileAndRun(l,[e],null,u)}return this.uploadToGPU(e.dataId),this.shallowSlice(e,i,r)},t.prototype.shallowSlice=function(e,i,r){var a=this.texData.get(e.dataId),s=this.makeOutput(r,e.dtype),o=this.texData.get(s.dataId);Object.assign(o,a),o.shape=r,o.dtype=e.dtype;var l=x.slice_util.computeFlatOffset(i,e.strides);a.slice&&(l+=a.slice.flatOffset),o.slice={flatOffset:l,origDataId:a.slice&&a.slice.origDataId||e.dataId};var u=this.dataRefCount.get(o.slice.origDataId)||1;return this.dataRefCount.set(o.slice.origDataId,u+1),s},t.prototype.stridedSlice=function(e,i,r,a){var s=this,o=this.tryRunOnCpuOrThrow([e],function(){return s.cpuBackend.stridedSlice(e,i,r,a)});if(o)return o;var l=x.slice_util.computeOutShape(i,r,a);if(l.some(function(c){return c===0}))return x.tensor([],l);var u=new Q9(i,a,l);return this.compileAndRun(u,[e])},t.prototype.reverse=function(e,i){var r=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new j9(e.shape,i):new Y9(e.shape,i);return this.compileAndRun(r,[e])},t.prototype.concat=function(e,i){if(e[0].dtype==="complex64"){var r=e.map(function(f){return x.real(f)}),a=e.map(function(f){return x.imag(f)});return x.complex(this.concat(r,i),this.concat(a,i))}if(e.length===1)return e[0];if(e.length>x.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new Q5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=x.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,x.util.sizeFromShape(f.shape.slice(i)))}),d=new Z5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,yL,e.dtype);var a=new Re(e.shape,yL);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>xL){r&&(e=x.transpose(e,[0,2,1])),a&&(i=x.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=x.upcastType(e.dtype,i.dtype),m=new Lp(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=x.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?hl(l,!0):null,b=new Lp(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new ES(OS.REAL,e.shape,i.shape),o=new ES(OS.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=xM(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,kS,e.dtype);var g=new St(kS,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=x.env().getBool("WEBGL_PACK_NORMALIZATION")?new C9(e.shape,i,r,a,s):new N9(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new x9(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return x.util.decodeString(l)}),s=x.buffer(e.shape,e.dtype,a);return B6(s,i)}var o=new a6(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new P9(e.shape,i,r):new z9(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new y9(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){x.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=x.backend_util.getReshaped(e.shape,i,a),o=x.backend_util.getPermuted(s.length,i.length),l=x.backend_util.getReshapedPermuted(e.shape,i,a),u=x.backend_util.getSliceBeginCoords(r,i.length),c=x.backend_util.getSliceSize(l,r,i.length);return x.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){x.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;o<e.shape.length;++o)s.push([0,0]);var l=e.pad(s),u=x.backend_util.getReshaped(l.shape,i,a,!1),c=x.backend_util.getPermuted(u.length,i.length,!1),h=x.backend_util.getReshapedPermuted(l.shape,i,a,!1),d=x.transpose(l.reshape(u),c);return x.reshape(d,h)},t.prototype.reduce=function(e,i,r){var a=e.shape[0],s=e.shape[1],o=x.backend_util.computeOptimalWindowSize(s),l=Math.ceil(s/o),u={windowSize:o,inSize:s,batchSize:a,outSize:l},c=new lL(u,i),h=this.compileAndRun(c,[e],r);return h.shape[1]===1?h:this.reduce(h,i,r)},t.prototype.argReduce=function(e,i,r){r===void 0&&(r=null);var a=e.shape[0],s=e.shape[1];r!=null&&(a=r.shape[0],s=r.shape[1]);var o=x.backend_util.computeOptimalWindowSize(s),l={windowSize:o,inSize:s,batchSize:a,outSize:Math.ceil(s/o)},u=new FM(l,i,r==null),c=[e];r!=null&&c.push(r);var h=this.compileAndRun(u,c,"int32");return h.shape[1]===1?h:this.argReduce(e,i,h)},t.prototype.argReducePacked=function(e,i,r){r===void 0&&(r=null);var a=r!=null?r.shape:e.shape,s=a[a.length-1],o=x.backend_util.computeOptimalWindowSize(s),l=new y5(a,o,i,r==null),u=r==null?[e]:[e,r],c=this.compileAndRun(l,u,"int32");return c.rank===e.rank?this.argReducePacked(e,i,c):c},t.prototype.sum=function(e,i){x.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o),u=x.sumOutType(e.dtype);return this.reduce(l,"sum",u).reshape(a)},t.prototype.prod=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e],function(){return r.cpuBackend.prod(e,i)});if(a)return a;var s=x.backend_util.computeOutAndReduceShapes(e.shape,i),o=s[0],l=s[1],u=x.util.sizeFromShape(l),c=e.as2D(-1,u),h=x.sumOutType(e.dtype);return this.reduce(c,"prod",h).reshape(o)},t.prototype.unsortedSegmentSum=function(e,i,r){var a=0,s=x.backend_util.getAxesPermutation([a],e.rank),o=e;s!=null&&(o=x.transpose(e,s),a=x.backend_util.getInnerMostAxes(1,e.rank)[0]);var l=NL.computeOutShape(o.shape,a,r),u=x.util.sizeFromShape([o.shape[a]]),c=o.as2D(-1,u),h=x.sumOutType(e.dtype),d=this.segOpCompute(c,"unsortedSegmentSum",i,h,r).reshape(l);return s!=null&&(d=x.transpose(d,x.backend_util.getUndoAxesPermutation(s))),d},t.prototype.segOpCompute=function(e,i,r,a,s){var o=e.shape[0],l=e.shape[1],u=NL.segOpComputeOptimalWindowSize(l,s),c={windowSize:u,inSize:l,batchSize:o,numSegments:s},h=new K9(c,i),d=this.compileAndRun(h,[e,r],a);return d.shape[1]===s?d:(r=x.range(0,s).tile([l/u]),this.segOpCompute(d,i,r,a,s))},t.prototype.argMinMaxReduce=function(e,i,r){var a=[i];if(x.backend_util.assertAxesAreInnerMostDims("arg"+r.charAt(0).toUpperCase()+r.slice(1),a,e.rank),!x.env().getBool("WEBGL_PACK_REDUCE")||e.rank<=2){var s=x.backend_util.computeOutAndReduceShapes(e.shape,a),o=s[0],l=s[1],u=x.util.sizeFromShape(l),c=e.as2D(-1,u);return this.argReduce(c,r).reshape(o)}return this.argReducePacked(e,r)},t.prototype.argMin=function(e,i){return this.argMinMaxReduce(e,i,"min")},t.prototype.argMax=function(e,i){return this.argMinMaxReduce(e,i,"max")},t.prototype.cumsum=function(e,i,r,a){if(i!==e.rank-1)throw new Error("WebGL cumsum shader expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=e.shape[i],o=e,l=0;l<=Math.ceil(Math.log2(s))-1;l++){var u=new MS(e.shape,!1,a),c=u.getCustomSetupFunc(l),h=o;o=this.compileAndRun(u,[o],o.dtype,c),h.dispose()}if(r){var u=new MS(e.shape,r,a),h=o;o=this.compileAndRun(u,[o]),h.dispose()}return o},t.prototype.equal=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,z5,"bool");var r=new St(I5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.notEqual=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,P5,"bool");var r=new St(A5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.less=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.less(e,i)});if(a)return a;if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,_5,"bool");var s=new St(T5,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.lessEqual=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,M5,"bool");var r=new St(N5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.greater=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.greater(e,i)});if(a)return a;if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,H5,"bool");var s=new St(x5,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.greaterEqual=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,V5,"bool");var r=new St(C5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalNot=function(e){var i=new Re(e.shape,E6);return this.compileAndRun(i,[e])},t.prototype.logicalAnd=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,q5,"bool");var r=new St(R5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalOr=function(e,i){if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,G5,"bool");var r=new St(O5,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.select=function(e,i,r){var a=new $9(e.rank,i.shape,i.rank);return this.compileAndRun(a,[e,i,r],x.upcastType(i.dtype,r.dtype))},t.prototype.where=function(e){x.backend_util.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");var i=e.dataSync();return P6(e.shape,i)},t.prototype.topk=function(e,i,r){var a=e.dataSync();return z6(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){x.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"min",l.dtype).reshape(a)},t.prototype.minimum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.minimum(e,i)});if(a)return a;var s=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(j5,e.shape,i.shape):new St(D5,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.mod=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(K5,e.shape,i.shape):new St(k5,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.maximum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.maximum(e,i)});if(a)return a;var s=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(Y5,e.shape,i.shape):new St(E5,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.all=function(e,i){x.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"all",l.dtype).reshape(a)},t.prototype.any=function(e,i){x.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);var r=x.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=x.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"any",l.dtype).reshape(a)},t.prototype.floorDiv=function(e,i){var r=S5,a="int32";if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,W5,a);var s=new St(r,e.shape,i.shape);return this.compileAndRun(s,[e,i],a)},t.prototype.add=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,mp);var r=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=wM(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,mp,r);var c=new St(mp,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.packedUnaryOp=function(e,i,r){var a=new Ts(e.shape,i);return this.compileAndRun(a,[e],r)},t.prototype.packedBinaryOp=function(e,i,r,a,s){s===void 0&&(s=!1);var o=new Ui(r,e.shape,i.shape,s);return this.compileAndRun(o,[e,i],a)},t.prototype.complexSeparableBinaryOp=function(e,i,r){var a=this,s=this.texData.get(e.dataId),o=this.texData.get(i.dataId),l=[[s.complexTensors.real,o.complexTensors.real],[s.complexTensors.imag,o.complexTensors.imag]].map(function(d){var p=d[0],f=d[1],m=a.makeComplexComponentTensorInfo(e,p),g=a.makeComplexComponentTensorInfo(i,f),v=new St(r,e.shape,i.shape);return a.compileAndRun(v,[m,g],x.upcastType(p.dtype,f.dtype))}),u=l[0],c=l[1],h=this.complex(u,c);return u.dispose(),c.dispose(),h},t.prototype.makeComplexComponentTensorInfo=function(e,i){return{dataId:i.dataId,dtype:i.dtype,shape:e.shape}},t.prototype.addN=function(e){if(e.length===1)return e[0];if(e.length>x.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return x.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=x.env().getBool("WEBGL_PACK"),u=l?new kM(e[0].shape,o):new DM(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,gp);var r=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=OM(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,gp,e.dtype);var c=new St(gp,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new Ui(U5,e.shape,i.shape):new St(L5,e.shape,i.shape),s=x.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=SM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,bL,e.dtype);var r=new Re(e.shape,bL);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=AM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,wL,e.dtype);var r=new Re(e.shape,wL);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,u6);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,c6);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,h6);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,d6);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,p6);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=LM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,SL,e.dtype);var r=new Re(e.shape,SL);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=IM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,LL,e.dtype);var r=new Re(e.shape,LL);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=x.util.parseAxisParam([i],e.shape),a=x.max(e,r),s=x.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return x.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=TM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,F6,e.dtype);var r=new Re(e.shape,f6);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,m6);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,g6);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=CM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,v6);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,O6);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return x.env().getBool("WEBGL_PACK")?i=new Ts(e.shape,IL):i=new Re(e.shape,mL),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return x.env().getBool("WEBGL_PACK")?i=new Ts(e.shape,AL):i=new Re(e.shape,gL),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(WS,e.shape,i.shape):new St(FS,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,TL,e.dtype);var i=new Re(e.shape,vL);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(B5,e.shape,i.shape):new St(F5,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,o6);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,D6);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;x.env().getBool("WEBGL_PACK_CLIP")?a=new X5(e.shape):a=new $5(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=bM(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,fL,e.dtype);var r=new Re(e.shape,fL);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new J5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,y6);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,b6);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,w6);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,S6);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,L6);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,I6);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,A6);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,T6);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,N6);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,x6);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,C6);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,R6);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,l6(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>xL,v=l[2]%2!==0&&!!u.isPacked;if(g||!x.env().getBool("WEBGL_LAZILY_UNPACK")||!x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=x.reshape(e,[1,b,r.inChannels]),S=x.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return x.reshape(L,r.outShape)}var N=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,N,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,x.util.assert(Ls(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=x.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return x.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,x.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),N=new T9(v,S.shape,r),C=this.compileAndRun(N,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?hl(s,!0):null,W=new Lp(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(x.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?hl(o,!1):null,d=new US(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(x.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new US(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new t9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new e9(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=x.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?hl(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new zS(a,d,c,p),this.compileAndRun(f,h)):(f=new BS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return x.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new zS(r),this.compileAndRun(a,[e,i])):(a=new BS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new a9(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new r9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new s9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new i9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new n9(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return x.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=this.slice(e,l,u).reshape(a);return c},t.prototype.avgPool3d=function(e,i){var r=new Ip(i,"avg",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.avgPool3dBackprop=function(e,i,r){var a=new w5(r);return this.compileAndRun(a,[e],i.dtype)},t.prototype.maxPool3d=function(e,i){var r=new Ip(i,"max",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.maxPool3dBackprop=function(e,i,r,a){var s=!0,o=new Ip(a,"max",s),l=this.compileAndRun(o,[i]),u=new O9(a),c=this.compileAndRun(u,[e,l],i.dtype);return l.dispose(),c},t.prototype.resizeBilinear=function(e,i,r,a){var s=x.env().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new V9(e.shape,i,r,a):new H9(e.shape,i,r,a);return this.compileAndRun(s,[e],"float32")},t.prototype.resizeBilinearBackprop=function(e,i,r){var a=new M9(e,i,r);return this.compileAndRun(a,[e])},t.prototype.resizeNearestNeighbor=function(e,i,r,a){var s=new G9(e.shape,i,r,a);return this.compileAndRun(s,[e])},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){var a=new q9(e,i,r);return this.compileAndRun(a,[e])},t.prototype.multinomial=function(e,i,r,a){var s=i?e:x.softmax(e),o=s.shape[0],l=s.shape[1],u=new E9(o,l,r),c=u.getCustomSetupFunc(a);return this.compileAndRun(u,[s],"int32",c)},t.prototype.oneHot=function(e,i,r,a){var s=new D9(e.size,i,r,a);return this.compileAndRun(s,[e])},t.prototype.diag=function(e){var i=new h9(e.size);return this.compileAndRun(i,[e])},t.prototype.cropAndResize=function(e,i,r,a,s,o){var l=new o9(e.shape,i.shape,a,s,o);return this.compileAndRun(l,[e,i,r],"float32")},t.prototype.depthToSpace=function(e,i,r){x.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new c9(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return U6(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=x.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return x.backend_util.reshapeTensor(x.tensor([]),r);var f=x.scalar(0),m=new cL(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=x.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new cL(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new VS(HS.REAL,e.shape,i),s=new VS(HS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=x.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new b9(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||x.util.inferDtype(i),r==="string"){var a=x.util.getArrayFromDType(r,x.util.sizeFromShape(e));return a.fill(i),x.engine().makeTensor(a,e,r,this)}else{var s=new g9(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return x.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return x.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new W6(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new U9(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[Lr(e.shape)].concat(Ir(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[Lr(i)].concat(Ir(i)),o=new uL(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=il(a),l;r?l=new u9(o):l=new l9(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===vs.DENSE){var c=bs(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),x.util.sizeFromShape(l.shape)===0)return u.values=x.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&x.util.sizeFromShape(w.shape)<=x.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!Ls(S.shape,w.shape)){var L=w,N=w.shape;w.shape=S.shape,w=o.packedReshape(w,N),h.push(w),S=o.texData.get(w.dataId),L.shape=N}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=A9(e,d,p),m=this.getAndSaveBinary(f,function(){return L9(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),I9(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!x.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return x.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!x.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=x.tidy(function(){if(!x.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=x.env().getBool("DEBUG");x.env().set("DEBUG",!1);var r=e.abs(x.scalar(1e-8)).dataSync()[0];if(x.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?_6:M6},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=x.util.now());var p=r.texShape;if(p==null&&(p=vS(a,c),r.texShape=p),o!=null){var f=il(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=fa(p[0],p[1]),g=i[0],v=i[1],m=new m9(f,[v,g],b)):m=new f9(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=sn.PIXELS:this.texData.get(w.dataId).usage=sn.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),N=this.texData.get(L.dataId);r.texture=N.texture,r.texShape=N.texShape,r.isPacked=N.isPacked,r.usage=N.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=x.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=Y6(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*x.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(x.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(x.KernelBackend);function Y6(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i<e.length;++i)e[i]=Math.round(n[i]);return e}else throw new Error("Unknown dtype "+t)}var j6="2.6.0";function RL(){x.env().set("WEBGL_FORCE_F16_TEXTURES",!0)}x.device_util.isBrowser()&&x.registerBackend("webgl",function(){return new CL},2);var K6={forceHalfFloat:RL},OL="if (isnan(x)) return x;",$6=`
   if (isnan(a)) return a;
   if (isnan(b)) return b;
 `,X6=`
@@ -3649,14 +3649,14 @@ return (round(mod(b, 2.0)) != 1) ?
   result.g = isNaN.g > 0. ? NAN : result.g;
   result.b = isNaN.b > 0. ? NAN : result.b;
   result.a = isNaN.a > 0. ? NAN : result.a;
-`;function dl(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function Ap(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(t,l.shape,u.shape,!!e):new St(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var J6=$6+`
+`;function dl(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function Tp(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ui(t,l.shape,u.shape,!!e):new St(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var J6=$6+`
   return atan(a, b);
 `,Z6=`
   vec4 result = atan(a, b);
   vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
   `+X6+`
   return result;
-`,Q6=Ap(J6,Z6),e8={kernelName:x.Atan2,backendName:"webgl",kernelFunc:Q6};function Tp(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var t8={kernelName:x.Identity,backendName:"webgl",kernelFunc:Tp};function n8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ma(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=x.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&x.util.arraysEqual(c.inShape,c.outShape))return Tp({inputs:{x:r},backend:e});var h=new As(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var i8={kernelName:x.AvgPool,backendName:"webgl",kernelFunc:n8};function r8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ma([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=x.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new b5(c);return e.runWebGLProgram(h,[r],s.dtype)}var a8={kernelName:x.AvgPoolBackprop,backendName:"webgl",kernelFunc:r8};var s8=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],x.backend_util.assertAndGetBroadcastShape(t,e),x.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(x.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(x.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
+`,Q6=Tp(J6,Z6),e8={kernelName:x.Atan2,backendName:"webgl",kernelFunc:Q6};function Np(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var t8={kernelName:x.Identity,backendName:"webgl",kernelFunc:Np};function n8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ma(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=x.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&x.util.arraysEqual(c.inShape,c.outShape))return Np({inputs:{x:r},backend:e});var h=new As(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var i8={kernelName:x.AvgPool,backendName:"webgl",kernelFunc:n8};function r8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ma([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=x.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new b5(c);return e.runWebGLProgram(h,[r],s.dtype)}var a8={kernelName:x.AvgPoolBackprop,backendName:"webgl",kernelFunc:r8};var s8=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],x.backend_util.assertAndGetBroadcastShape(t,e),x.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(x.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(x.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
       void main() {
         float x = getXAtOutCoords();
         float mean = getMeanAtOutCoords();
@@ -3703,7 +3703,7 @@ return a / b;`,f8=`
   }
 
   return result;
-`,m8=Ap(p8,f8,!0),g8={kernelName:x.Div,backendName:"webgl",kernelFunc:m8};var v8=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=`
+`,m8=Tp(p8,f8,!0),g8={kernelName:x.Div,backendName:"webgl",kernelFunc:m8};var v8=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=`
         void main() {
           ivec4 coords = getOutputCoords();
           int x = coords[2];
@@ -3773,7 +3773,7 @@ return a / b;`,f8=`
 
         `+e.output+` = result;
       }
-    `}return n}();var L8={kernelName:x.FromPixels,backendName:"webgl",kernelFunc:S8},Sa;function S8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(Sa==null&&(Sa=document.createElement("canvas").getContext("2d")),Sa.canvas.width=u,Sa.canvas.height=c,Sa.drawImage(r,0,0,u,c),r=Sa.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=on.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=x.env().getBool("WEBGL_PACK")?new w8(d):new b8(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function I8(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=x.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function A8(n,t,e,i){for(var r=I8(n.shape),a=n,s=0;s<r.length;s++){var o=r[s],l=o.inSize,u=o.windowSize,c=o.outSize,h=new lL({windowSize:u,inSize:l,batchSize:n.shape[0],outSize:c},e),d=a;a=i.runWebGLProgram(h,[a],t),d.dataId!==n.dataId&&i.disposeData(d.dataId)}return a}function T8(n,t,e){var i=[Lr(n.shape)].concat(Ir(n.shape)),r={dtype:n.dtype,shape:i,dataId:n.dataId},a=[Lr(t)].concat(Ir(t)),s=new uL(a,i),o=!0,l=e.runWebGLProgram(s,[r],n.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function Np(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=e,o=x.util.sizeFromShape(r.shape),l=x.util.inferFromImplicitShape(a,o),u=x.util.sizeFromShape(l);x.util.assert(o===u,function(){return"The new shape ("+l+") has "+u+" elements and the old "+("shape ("+r.shape+") has "+o+" elements. The new shape and old ")+"shape must have the same number of elements."});var c=s.texData.get(r.dataId);return c.isPacked&&!Ls(r.shape,l)&&!(c.texture!==null&&Ls(c.shape,l))?T8(r,l,s):(s.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var N8={kernelName:x.Reshape,backendName:"webgl",kernelFunc:Np};function x8(n,t,e,i){var r=x.util.sizeFromShape(t),a=x.util.sizeFromShape(n.shape),s=a/r,o=Np({inputs:{x:n},attrs:{shape:[s,r]},backend:i}),l=A8(o,n.dtype,"max",i),u=Np({inputs:{x:l},attrs:{shape:e},backend:i});return i.disposeIntermediateTensorInfo(o),i.disposeIntermediateTensorInfo(l),u}var R8=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[e[r]];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=C8(e);this.userCode=`
+    `}return n}();var L8={kernelName:x.FromPixels,backendName:"webgl",kernelFunc:S8},Sa;function S8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(Sa==null&&(Sa=document.createElement("canvas").getContext("2d")),Sa.canvas.width=u,Sa.canvas.height=c,Sa.drawImage(r,0,0,u,c),r=Sa.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=sn.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=x.env().getBool("WEBGL_PACK")?new w8(d):new b8(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function I8(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=x.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function A8(n,t,e,i){for(var r=I8(n.shape),a=n,s=0;s<r.length;s++){var o=r[s],l=o.inSize,u=o.windowSize,c=o.outSize,h=new lL({windowSize:u,inSize:l,batchSize:n.shape[0],outSize:c},e),d=a;a=i.runWebGLProgram(h,[a],t),d.dataId!==n.dataId&&i.disposeData(d.dataId)}return a}function T8(n,t,e){var i=[Lr(n.shape)].concat(Ir(n.shape)),r={dtype:n.dtype,shape:i,dataId:n.dataId},a=[Lr(t)].concat(Ir(t)),s=new uL(a,i),o=!0,l=e.runWebGLProgram(s,[r],n.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function xp(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=e,o=x.util.sizeFromShape(r.shape),l=x.util.inferFromImplicitShape(a,o),u=x.util.sizeFromShape(l);x.util.assert(o===u,function(){return"The new shape ("+l+") has "+u+" elements and the old "+("shape ("+r.shape+") has "+o+" elements. The new shape and old ")+"shape must have the same number of elements."});var c=s.texData.get(r.dataId);return c.isPacked&&!Ls(r.shape,l)&&!(c.texture!==null&&Ls(c.shape,l))?T8(r,l,s):(s.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var N8={kernelName:x.Reshape,backendName:"webgl",kernelFunc:xp};function x8(n,t,e,i){var r=x.util.sizeFromShape(t),a=x.util.sizeFromShape(n.shape),s=a/r,o=xp({inputs:{x:n},attrs:{shape:[s,r]},backend:i}),l=A8(o,n.dtype,"max",i),u=xp({inputs:{x:l},attrs:{shape:e},backend:i});return i.disposeIntermediateTensorInfo(o),i.disposeIntermediateTensorInfo(l),u}var R8=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[e[r]];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=C8(e);this.userCode=`
     void main() {
       `+a+` resRC = getOutputCoords();
       setOutput(getA(`+s+`));
@@ -3795,7 +3795,7 @@ return a / b;`,f8=`
       }
       setOutput(result);
     }
-    `}return n}();function EL(n,t,e){var i=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new O8(n.shape,t):new R8(n.shape,t);return e.runWebGLProgram(i,[n],n.dtype)}var E8={kernelName:x.Max,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape.length,c=x.util.parseAxisParam(s,r.shape),h=c,d=x.backend_util.getAxesPermutation(h,u),p=d!=null,f=l.shouldExecuteOnCPU([r]),m=r;if(p){if(f){for(var g=l.texData.get(m.dataId),v=g.values,b=new Array(u),w=0;w<b.length;w++)b[w]=r.shape[d[w]];var S=AS(v,r.shape,r.dtype,d,b);m=l.makeTensorInfo(b,r.dtype);var L=l.texData.get(m.dataId);L.values=S}else m=EL(r,d,l);h=x.backend_util.getInnerMostAxes(h.length,u)}x.backend_util.assertAxesAreInnerMostDims("max",h,u);var N=x.backend_util.computeOutAndReduceShapes(m.shape,h),C=N[0],R=N[1],D=C;o&&(D=x.backend_util.expandShapeToKeepDim(C,c));var k;if(f){var g=l.texData.get(m.dataId),v=g.values,W=NM(v,x.util.sizeFromShape(R),D,r.dtype);k=l.makeTensorInfo(D,r.dtype);var F=l.texData.get(k.dataId);F.values=W}else k=x8(m,R,D,l);return p&&l.disposeIntermediateTensorInfo(m),k}};function D8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ma(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=x.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&x.util.arraysEqual(c.inShape,c.outShape))return Tp({inputs:{x:r},backend:e});var h=new As(c,"max",!1);return e.runWebGLProgram(h,[r],r.dtype)}var k8={kernelName:x.MaxPool,backendName:"webgl",kernelFunc:D8};function F8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ma([a,s],"maxPoolBackprop");var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=x.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=!0,f=new As(d,"max",p),m=e.runWebGLProgram(f,[o],o.dtype),g=new R9(d),v=e.runWebGLProgram(g,[r,m],o.dtype);return e.disposeIntermediateTensorInfo(m),v}var W8={kernelName:x.MaxPoolBackprop,backendName:"webgl",kernelFunc:F8};function U8(n,t,e,i){var r=new As(e,"max",!1),a=i.runWebGLProgram(r,[n],"float32");r=new As(e,"max",!0,!0,t);var s=i.runWebGLProgram(r,[n],"float32");return[a,s]}var B8={kernelName:x.MaxPoolWithArgmax,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;x.util.assert(r.shape.length===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+r.shape.length+"."});var h=[1,1];x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(o,h),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+o+" and dilations '"+h+"'")});var d=x.backend_util.computePool2DInfo(r.shape,s,o,h,l),p=U8(r,u,d,c),f=p[0],m=p[1];return[f,m]}};var z8={kernelName:x.NonMaxSuppressionV3,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;x.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=e,d=h.readSync(a.dataId),p=h.readSync(s.dataId),f=l,m=u,g=c;return x.kernel_impls.nonMaxSuppressionV3Impl(d,p,f,m,g)}};var P8=x.kernel_impls.nonMaxSuppressionV4Impl,_8={kernelName:x.NonMaxSuppressionV4,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;x.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=P8(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var M8=x.kernel_impls.nonMaxSuppressionV5Impl,H8={kernelName:x.NonMaxSuppressionV5,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;x.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=l,g=u,v=c,b=h,w=M8(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var V8=function(){function n(t,e,i,r){this.variableNames=["Image"],this.outputShape=[];var a=t[1],s=t[2],o=Math.sin(e).toFixed(3),l=Math.cos(e).toFixed(3);this.outputShape=t;var u=x.backend_util.getImageCenter(r,a,s),c=u[0],h=u[1],d=c.toFixed(3),p=h.toFixed(3),f="";typeof i=="number"?f="float outputValue = "+i.toFixed(2)+";":f=`
+    `}return n}();function EL(n,t,e){var i=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new O8(n.shape,t):new R8(n.shape,t);return e.runWebGLProgram(i,[n],n.dtype)}var E8={kernelName:x.Max,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape.length,c=x.util.parseAxisParam(s,r.shape),h=c,d=x.backend_util.getAxesPermutation(h,u),p=d!=null,f=l.shouldExecuteOnCPU([r]),m=r;if(p){if(f){for(var g=l.texData.get(m.dataId),v=g.values,b=new Array(u),w=0;w<b.length;w++)b[w]=r.shape[d[w]];var S=AS(v,r.shape,r.dtype,d,b);m=l.makeTensorInfo(b,r.dtype);var L=l.texData.get(m.dataId);L.values=S}else m=EL(r,d,l);h=x.backend_util.getInnerMostAxes(h.length,u)}x.backend_util.assertAxesAreInnerMostDims("max",h,u);var N=x.backend_util.computeOutAndReduceShapes(m.shape,h),C=N[0],R=N[1],D=C;o&&(D=x.backend_util.expandShapeToKeepDim(C,c));var k;if(f){var g=l.texData.get(m.dataId),v=g.values,W=NM(v,x.util.sizeFromShape(R),D,r.dtype);k=l.makeTensorInfo(D,r.dtype);var F=l.texData.get(k.dataId);F.values=W}else k=x8(m,R,D,l);return p&&l.disposeIntermediateTensorInfo(m),k}};function D8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ma(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=x.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&x.util.arraysEqual(c.inShape,c.outShape))return Np({inputs:{x:r},backend:e});var h=new As(c,"max",!1);return e.runWebGLProgram(h,[r],r.dtype)}var k8={kernelName:x.MaxPool,backendName:"webgl",kernelFunc:D8};function F8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ma([a,s],"maxPoolBackprop");var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=x.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=!0,f=new As(d,"max",p),m=e.runWebGLProgram(f,[o],o.dtype),g=new R9(d),v=e.runWebGLProgram(g,[r,m],o.dtype);return e.disposeIntermediateTensorInfo(m),v}var W8={kernelName:x.MaxPoolBackprop,backendName:"webgl",kernelFunc:F8};function U8(n,t,e,i){var r=new As(e,"max",!1),a=i.runWebGLProgram(r,[n],"float32");r=new As(e,"max",!0,!0,t);var s=i.runWebGLProgram(r,[n],"float32");return[a,s]}var B8={kernelName:x.MaxPoolWithArgmax,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;x.util.assert(r.shape.length===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+r.shape.length+"."});var h=[1,1];x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(o,h),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+o+" and dilations '"+h+"'")});var d=x.backend_util.computePool2DInfo(r.shape,s,o,h,l),p=U8(r,u,d,c),f=p[0],m=p[1];return[f,m]}};var z8={kernelName:x.NonMaxSuppressionV3,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;x.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=e,d=h.readSync(a.dataId),p=h.readSync(s.dataId),f=l,m=u,g=c;return x.kernel_impls.nonMaxSuppressionV3Impl(d,p,f,m,g)}};var P8=x.kernel_impls.nonMaxSuppressionV4Impl,_8={kernelName:x.NonMaxSuppressionV4,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;x.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=P8(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var M8=x.kernel_impls.nonMaxSuppressionV5Impl,H8={kernelName:x.NonMaxSuppressionV5,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;x.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=l,g=u,v=c,b=h,w=M8(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var V8=function(){function n(t,e,i,r){this.variableNames=["Image"],this.outputShape=[];var a=t[1],s=t[2],o=Math.sin(e).toFixed(3),l=Math.cos(e).toFixed(3);this.outputShape=t;var u=x.backend_util.getImageCenter(r,a,s),c=u[0],h=u[1],d=c.toFixed(3),p=h.toFixed(3),f="";typeof i=="number"?f="float outputValue = "+i.toFixed(2)+";":f=`
         vec3 fill = vec3(`+i.join(",")+`);
         float outputValue = fill[coords[3]];`,this.userCode=`
         void main() {
@@ -3814,7 +3814,7 @@ return a / b;`,f8=`
         }
     `}return n}();var q8={kernelName:x.RotateWithOffset,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=new V8(r.shape,s,o,l),h=u.runWebGLProgram(c,[r],r.dtype);return h}};var G8=OL+`
   return sin(x);
-`,Y8=dl(G8),j8={kernelName:x.Sin,backendName:"webgl",kernelFunc:Y8};var K8="return x * x;",$8=dl(K8),X8={kernelName:x.Square,backendName:"webgl",kernelFunc:$8};var DL="return (a - b) * (a - b);",J8=Ap(DL,DL),Z8={kernelName:x.SquaredDifference,backendName:"webgl",kernelFunc:J8};var Q8="return tan(x);",e7=dl(Q8),t7={kernelName:x.Tan,backendName:"webgl",kernelFunc:e7};var n7={kernelName:x.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u<l.length;u++)l[u]=r.shape[a[u]];var c;if(s.shouldExecuteOnCPU([r])){var h=s.texData.get(r.dataId),d=h.values,p=AS(d,r.shape,r.dtype,a,l);c=s.makeTensorInfo(l,r.dtype);var f=s.texData.get(c.dataId);f.values=p}else c=EL(r,a,s);return c}};function i7(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ma(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");var s=i.readSync(a.dataId),o=EM(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var r7={kernelName:x.Unique,backendName:"webgl",kernelFunc:i7};var a7=[e8,i8,a8,u8,d8,g8,y8,L8,t8,E8,k8,W8,B8,z8,_8,H8,N8,q8,j8,X8,Z8,t7,n7,r7];for(var xp=0,kL=a7;xp<kL.length;xp++){var s7=kL[xp];x.registerKernel(s7)}qn.GPGPUContext=sL;qn.MathBackendWebGL=CL;qn.forceHalfFloat=RL;qn.gpgpu_util=w9;qn.setWebGLContext=Q0;qn.version_webgl=j6;qn.webgl=K6;qn.webgl_util=rM});var zt=Le(Nr=>{"use strict";Object.defineProperty(Nr,"__esModule",{value:!0});var Cp=lr(),Rp=Ow(),Op=Gw(),WL=h0(),o7=Z0(),l7=FL();var u7="2.6.0";var c7={"tfjs-core":Cp.version_core,"tfjs-backend-cpu":o7.version_cpu,"tfjs-backend-webgl":l7.version_webgl,"tfjs-data":WL.version_data,"tfjs-layers":Rp.version_layers,"tfjs-converter":Op.version_converter,tfjs:u7};Object.keys(Cp).forEach(function(n){n!=="default"&&Object.defineProperty(Nr,n,{enumerable:!0,get:function(){return Cp[n]}})});Object.keys(Rp).forEach(function(n){n!=="default"&&Object.defineProperty(Nr,n,{enumerable:!0,get:function(){return Rp[n]}})});Object.keys(Op).forEach(function(n){n!=="default"&&Object.defineProperty(Nr,n,{enumerable:!0,get:function(){return Op[n]}})});Nr.data=WL;Nr.version=c7});var _L=Le(pl=>{const ke=zt(),UL=6;function h7(n){const t={strides:[n/16,n/8],anchors:[2,6]},e=[];for(let i=0;i<t.strides.length;i++){const r=t.strides[i],a=Math.floor((n+r-1)/r),s=Math.floor((n+r-1)/r),o=t.anchors[i];for(let l=0;l<a;l++){const u=r*(l+.5);for(let c=0;c<s;c++){const h=r*(c+.5);for(let d=0;d<o;d++)e.push([h,u])}}}return e}const BL=n=>{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},zL=n=>({startEndTensor:n,startPoint:ke.slice(n,[0,0],[-1,2]),endPoint:ke.slice(n,[0,2],[-1,2])}),d7=(n,t)=>{const e=ke.mul(n.startPoint,t),i=ke.mul(n.endPoint,t),r=ke.concat2d([e,i],1);return zL(r)};function p7(n,t,e){const i=ke.slice(n,[0,1],[-1,2]),r=ke.add(i,t),a=ke.slice(n,[0,3],[-1,2]),s=ke.div(a,e),o=ke.div(r,e),l=ke.div(s,2),u=ke.sub(o,l),c=ke.add(o,l),h=ke.mul(u,e),d=ke.mul(c,e),p=1;return ke.concat2d([h,d],p)}function f7(n,t){return ke.tidy(()=>{const e=n.box?n.box:n;return d7(e,t).startEndTensor.squeeze()})}class PL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=h7(t.detector.inputSize),this.anchors=ke.tensor2d(this.anchorsData),this.inputSize=ke.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){if(!n||n.isDisposedInternal||n.shape.length!==4||n.shape[1]<1||n.shape[2]<1)return null;const[t,e,i]=ke.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=ke.mul(ke.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u);let h;if(Array.isArray(c)){const m=c.sort((w,S)=>w.size-S.size),g=ke.concat([m[0],m[2]],2),v=ke.concat([m[1],m[3]],2),b=ke.concat([v,g],1);h=b.squeeze(0)}else h=c.squeeze();const d=p7(h,this.anchors,this.inputSize),p=ke.slice(h,[0,0],[-1,1]),f=ke.sigmoid(p).squeeze();return[h,d,f]}),r=await ke.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>ke.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l<s.length;l++){const u=s[l],c=ke.tidy(()=>{const h=zL(u),d=a[l],p=this.anchorsData[d],f=ke.slice(t,[d,UL-1],[1,-1]).squeeze().reshape([UL,-1]),m=ke.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.width,n.shape[1]/this.height]}}async estimateFaces(n){const t=ke.tidy(()=>(n instanceof ke.Tensor||(n=ke.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:e,scaleFactor:i}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(e.map(async r=>{const a=f7(r,i),[s,o,l]=await Promise.all([r.landmarks,a,r.probability].map(async f=>f.array())),u=r.anchor,[c,h]=i,d=s.map(f=>[(f[0]+u[0])*c,(f[1]+u[1])*h]),p={topLeft:o.slice(0,2),bottomRight:o.slice(2),landmarks:d,probability:l};return BL(r.box),r.landmarks.dispose(),r.probability.dispose(),a.dispose(),p}))}}async function m7(n){const t=await ke.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new PL(t,n);return e}pl.load=m7;pl.BlazeFaceModel=PL;pl.disposeBox=BL});var Dp=Le(Ep=>{Ep.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};Ep.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var ML=Le(xr=>{const g7=zt();function v7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}xr.scaleBoxCoordinates=v7;function kp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}xr.getBoxSize=kp;function Fp(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}xr.getBoxCenter=Fp;function y7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return g7.image.cropAndResize(t,a,[0],e)}xr.cutBoxFromImageAndResize=y7;function b7(n,t=1.5){const e=Fp(n),i=kp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}xr.enlargeBox=b7;function w7(n){const t=Fp(n),e=kp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}xr.squarifyBox=w7});var YL=Le(Rn=>{Rn.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function HL(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Rn.normalizeRadians=HL;function S7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return HL(e)}Rn.computeRotation=S7;function L7(n){return n*180/Math.PI}Rn.radToDegrees=L7;function VL(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function La(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Rn.dot=La;function qL(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Rn.getColumnFrom2DArr=qL;function GL(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(La(n[r],qL(t,a)))}return e}function I7(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=VL(t[0],t[1]),s=GL(a,r),o=VL(-t[0],-t[1]);return GL(s,o)}Rn.buildRotationMatrix=I7;function A7(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-La(t[0],e),-La(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Rn.invertTransformMatrix=A7;function T7(n,t){return[La(n,t[0]),La(n,t[1])]}Rn.rotatePoint=T7;function N7(n,t){return Math.sqrt((n[0]-t[0])**2+(n[1]-t[1])**2)}Rn.xyDistanceBetweenPoints=N7});var KL=Le(jL=>{const Bi=zt(),Yn=ML(),zi=Dp(),Pi=YL(),x7=468,C7=.25,R7=13,O7=[R7,zi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],E7=3,D7=2,k7=[E7,D7],Wp=zi.MESH_ANNOTATIONS.leftEyeLower0,Up=[Wp[0],Wp[Wp.length-1]],Bp=zi.MESH_ANNOTATIONS.rightEyeLower0,zp=[Bp[0],Bp[Bp.length-1]],F7=3,W7=4,U7=71,Pp=76;function fl(n,t,e,i){for(let r=0;r<zi.MESH_TO_IRIS_INDICES_MAP.length;r++){const{key:a,indices:s}=zi.MESH_TO_IRIS_INDICES_MAP[r],o=zi.MESH_ANNOTATIONS[`${e}${a}`],l=i==null;if(l||i.includes(a))for(let u=0;u<s.length;u++){const c=s[u];n[o[u]]=[t[c][0],t[c][1],(t[c][2]+n[o[u]][2])/2]}}}class B7{constructor(n,t,e,i){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.irisModel=e,this.meshWidth=i.mesh.inputSize,this.meshHeight=i.mesh.inputSize,this.irisSize=i.iris.inputSize,this.irisEnlarge=i.iris.enlargeFactor}transformRawCoords(n,t,e,i){const r=Yn.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Pi.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Pi.rotatePoint(d,o),d[2]]),u=Pi.invertTransformMatrix(i),c=[...Yn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Pi.dot(c,u[0]),Pi.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[Up[0]][2],e=n[zp[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Yn.squarifyBox(Yn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Yn.getBoxSize(a);let o=Bi.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Bi.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a<Pp;a++){const s=n[a*3],o=n[a*3+1],l=n[a*3+2];r.push([(i?1-s/this.irisSize:s/this.irisSize)*e[0]+t.startPoint[0],o/this.irisSize*e[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(U7)}}getAdjustedIrisCoords(n,t,e){const i=n[zi.MESH_ANNOTATIONS[`${e}EyeUpper0`][F7]][2],r=n[zi.MESH_ANNOTATIONS[`${e}EyeLower0`][W7]][2],a=(i+r)/2;return t.map((s,o)=>{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Yn.scaleBoxCoordinates(o,r),u=Yn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Bi.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=x7;let[o,l]=O7;s===!1&&([o,l]=k7),a=Pi.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Yn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Pi.IDENTITY_MATRIX;a!==0&&(h=Bi.image.rotateWithOffset(n,a,0,c),d=Pi.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Yn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Bi.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:N,boxSize:C,crop:R}=this.getEyeBox(b,f,Up[0],Up[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,zp[0],zp[1]),F=this.irisModel.predict(Bi.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,Pp*3),{rawCoords:_,iris:j}=this.getEyeCoords(H,N,C,!0),K=P.slice(Pp*3),{rawCoords:q,iris:G}=this.getEyeCoords(K,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(fl(b,_,"left"),fl(b,q,"right")):Z<1?fl(b,_,"left",["EyeUpper0","EyeLower0"]):fl(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,j,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Bi.dispose(b);const S=Yn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const N=Bi.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:N.arraySync()};const C={coords:N,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t<n.length;t++){const e=n[t],i=this.regionsOfInterest[t];let r=0;if(i&&i.startPoint){const[a,s]=e.startPoint,[o,l]=e.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}r<C7&&(this.regionsOfInterest[t]=e)}this.regionsOfInterest=this.regionsOfInterest.slice(0,n.length)}clearRegionOfInterest(n){this.regionsOfInterest[n]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,n),...this.regionsOfInterest.slice(n+1)])}shouldUpdateRegionsOfInterest(){const n=this.regionsOfInterest.length,t=n===0;return this.maxFaces===1||t?t:n!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}jL.Pipeline=B7});var XL=Le($L=>{$L.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var JL=Le(z7=>{Sl(z7,{default:()=>P7});var P7=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var eI=Le(Ns=>{const Cr=zt(),_7=_L(),ZL=Dp(),M7=KL(),H7=XL(),V7=JL().default;class QL{constructor(n,t,e,i){this.pipeline=new M7.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=Cr.tidy(()=>(n instanceof Cr.Tensor||(n=Cr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);Cr.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in ZL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=ZL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?Cr.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function q7(n){const t=await Promise.all([_7.load(n),Cr.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),Cr.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new QL(t[0],t[1],t[2],n);return e}Ns.load=q7;Ns.MediaPipeFaceMesh=QL;Ns.uv_coords=H7;Ns.triangulation=V7});var nI=Le(ml=>{const un=zt(),_i={};let tI={age:0,gender:""},_p=0;async function G7(n,t){const e=un.browser.fromPixels(n),i=un.image.resizeBilinear(e,[t,t]),r=un.cast(un.expandDims(i,0),"float32");return r}async function Y7(n){return _i.age||(_i.age=await un.loadGraphModel(n.face.age.modelPath)),_i.age}async function j7(n){return _i.gender||(_i.gender=await un.loadGraphModel(n.face.gender.modelPath)),_i.gender}async function K7(n,t){if(_p+=1,_p>=t.face.age.skipFrames)return _p=0,tI;let e;if(n instanceof un.Tensor){const r=un.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=un.mul(r,[255]),un.dispose(r)}else e=await G7(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await _i.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,un.dispose(r)}if(t.face.gender.enabled){const r=await _i.gender.predict(e),a=await r.data();i.gender=Math.trunc(100*a[0])<50?"female":"male",un.dispose(r)}return un.dispose(e),tI=i,i}ml.predict=K7;ml.loadAge=Y7;ml.loadGender=j7});var aI=Le(Mp=>{const Pt=zt(),$7=["angry","discust","fear","happy","sad","surpise","neutral"],gl={};let iI=[],Hp=0;const rI=1.5;function X7(n,t){const e=Pt.tidy(()=>{const i=Pt.browser.fromPixels(n,1),r=Pt.image.resizeBilinear(i,[t,t]),a=Pt.cast(Pt.expandDims(r,0),"float32");return a});return e}async function J7(n){return gl.emotion||(gl.emotion=await Pt.loadGraphModel(n.face.emotion.modelPath)),gl.emotion}async function Z7(n,t){if(Hp+=1,Hp>=t.face.emotion.skipFrames)return Hp=0,iI;const e=Pt.tidy(()=>{if(n instanceof Pt.Tensor){const r=Pt.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Pt.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Pt.mul(a,[.2989]),u=Pt.mul(s,[.587]),c=Pt.mul(o,[.114]),h=Pt.addN([l,u,c]);return h}return s}return X7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await gl.emotion.predict(e),a=await r.data();for(let s=0;s<a.length;s++)rI*a[s]>t.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*rI*a[s])/100),emotion:$7[s]});i.sort((s,o)=>o.score-s.score),Pt.dispose(r)}return Pt.dispose(e),iI=i,i}Mp.predict=Z7;Mp.load=J7});var lI=Le(sI=>{const oI=zt();class Q7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;oI.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return oI.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}sI.BaseModel=Q7});var Vp=Le(uI=>{const cI=zt(),eH=lI();class tH extends eH.BaseModel{preprocessInput(n){return cI.tidy(()=>cI.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}uI.MobileNet=tH});var dI=Le(hI=>{function qp(n){return Math.floor(n/2)}class nH{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(qp(n),n);)this.exchange(n,qp(n)),n=qp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(n,t))break;this.exchange(n,t),n=t}}getValueAt(n){return this.getElementValue(this.priorityQueue[n])}less(n,t){return this.getValueAt(n)<this.getValueAt(t)}exchange(n,t){const e=this.priorityQueue[n];this.priorityQueue[n]=this.priorityQueue[t],this.priorityQueue[t]=e}}hI.MaxHeap=nH});var fI=Le(pI=>{const iH=dI();function rH(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;h<c;++h){const d=Math.max(i-r,0),p=Math.min(i+r+1,o);for(let f=d;f<p;++f)if(a.get(h,f,n)>t){l=!1;break}if(!l)break}return l}function aH(n,t,e){const[i,r,a]=e.shape,s=new iH.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o<i;++o)for(let l=0;l<r;++l)for(let u=0;u<a;++u){const c=e.get(o,l,u);if(c<n)continue;rH(u,c,o,l,t,e)&&s.enqueue({score:c,part:{heatmapY:o,heatmapX:l,id:u}})}return s}pI.buildPartWithScoreQueue=aH});var xs=Le(On=>{On.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];On.NUM_KEYPOINTS=On.partNames.length;On.partIds=On.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const sH=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];On.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];On.connectedPartIndices=sH.map(([n,t])=>[On.partIds[n],On.partIds[t]]);On.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var Yp=Le(Mi=>{const oH=xs();function mI(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+oH.NUM_KEYPOINTS)}}Mi.getOffsetPoint=mI;function lH(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=mI(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Mi.getImageCoords=lH;function uH(n,t){const e=new Array(t);for(let i=0;i<t;i++)e[i]=n;return e}Mi.fillArray=uH;function Gp(n,t,e){return n<t?t:n>e?e:n}Mi.clamp=Gp;function cH(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Mi.squaredDistance=cH;function hH(n,t){return{x:n.x+t.x,y:n.y+t.y}}Mi.addVectors=hH;function dH(n,t,e){return{y:Gp(n.y,t,e),x:Gp(n.x,t,e)}}Mi.clampVector=dH});var wI=Le(gI=>{const Cs=xs(),Ia=Yp(),vI=Cs.poseChain.map(([n,t])=>[Cs.partIds[n],Cs.partIds[t]]),jp=vI.map(([,n])=>n),yI=vI.map(([n])=>n);function pH(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function Kp(n,t,e,i){return{y:Ia.clamp(Math.round(n.y/t),0,e-1),x:Ia.clamp(Math.round(n.x/t),0,i-1)}}function bI(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=Kp(t.position,a,l,u),h=pH(n,c,s),d=Ia.addVectors(t.position,h);let p=d;for(let g=0;g<o;g++){const v=Kp(p,a,l,u),b=Ia.getOffsetPoint(v.y,v.x,e,r);p=Ia.addVectors({x:v.x*a,y:v.y*a},{x:b.x,y:b.y})}const f=Kp(p,a,l,u),m=i.get(f.y,f.x,e);return{position:p,part:Cs.partNames[e],score:m}}function fH(n,t,e,i,r,a){const s=t.shape[2],o=jp.length,l=new Array(s),{part:u,score:c}=n,h=Ia.getImageCoords(u,i,e);l[u.id]={score:c,part:Cs.partNames[u.id],position:h};for(let d=o-1;d>=0;--d){const p=jp[d],f=yI[d];l[p]&&!l[f]&&(l[f]=bI(d,l[p],f,t,e,i,a))}for(let d=0;d<o;++d){const p=yI[d],f=jp[d];l[p]&&!l[f]&&(l[f]=bI(d,l[p],f,t,e,i,r))}return l}gI.decodePose=fH});var $p=Le(SI=>{const mH=fI(),gH=wI(),LI=Yp();function II(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return LI.squaredDistance(i,e,s.y,s.x)<=t})}function vH(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(II(n,t,a,o)||(r+=s),r),0);return i/e.length}const yH=1;function bH(n,t,e,i,r,a,s=.5,o=20){const l=[],u=mH.buildPartWithScoreQueue(s,yH,n),c=o*o;for(;l.length<a&&!u.empty();){const h=u.dequeue(),d=LI.getImageCoords(h.part,r,t);if(II(l,c,d,h.part.id))continue;const p=gH.decodePose(h,n,t,r,e,i),f=vH(l,c,p);l.push({keypoints:p,score:f})}return l}SI.decodeMultiplePoses=bH});var Jp=Le(gn=>{const Aa=zt(),wH=xs();function SH(n,t,e){return n<e||t<e}function LH(n,t){return wH.connectedPartIndices.reduce((e,[i,r])=>(SH(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}gn.getAdjacentKeyPoints=LH;const{NEGATIVE_INFINITY:AI,POSITIVE_INFINITY:TI}=Number;function NI(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:AI,maxY:AI,minX:TI,minY:TI})}gn.getBoundingBox=NI;function IH(n){const{minX:t,minY:e,maxX:i,maxY:r}=NI(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}gn.getBoundingBoxPoints=IH;async function AH(n){return Promise.all(n.map(t=>t.buffer()))}gn.toTensorBuffers3D=AH;function xI(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}gn.scalePose=xI;function CI(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>xI(a,t,e,i,r))}gn.scalePoses=CI;function RI(n){return n instanceof Aa.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}gn.getInputTensorDimensions=RI;function Xp(n){return n instanceof Aa.Tensor?n:Aa.browser.fromPixels(n)}gn.toInputTensor=Xp;function TH(n,t,e){return Aa.tidy(()=>{const i=Xp(n);return i.resizeBilinear([t,e])})}gn.toResizedInputTensor=TH;function NH(n,[t,e]){const[i,r]=RI(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s<a?(o=0,l=0,u=Math.round(.5*(a*i-r)),c=Math.round(.5*(a*i-r))):(o=Math.round(.5*(1/a*r-i)),l=Math.round(.5*(1/a*r-i)),u=0,c=0);const h=Aa.tidy(()=>{let d=Xp(n);return d=Aa.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}gn.padAndResizeTo=NH;function xH(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=CI(n,s,o,-a.top,-a.left);return l}gn.scaleAndFlipPoses=xH});var EI=Le(Zp=>{const CH=zt(),RH=Vp(),OH=$p(),vl=Jp();class OI{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=vl.getInputTensorDimensions(n),{resized:a,padding:s}=vl.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await vl.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await OH.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=vl.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}Zp.PoseNet=OI;async function EH(n){const t=await CH.loadGraphModel(n.modelPath),e=new RH.MobileNet(t,n.outputStride);return new OI(e)}async function DH(n){return EH(n)}Zp.load=DH});var kI=Le(tn=>{const kH=Vp(),DI=EI(),FH=$p(),yl=xs(),Rs=Jp();tn.load=DI.load;tn.PoseNet=DI.PoseNet;tn.MobileNet=kH.MobileNet;tn.decodeMultiplePoses=FH.decodeMultiplePoses;tn.partChannels=yl.partChannels;tn.partIds=yl.partIds;tn.partNames=yl.partNames;tn.poseChain=yl.poseChain;tn.getAdjacentKeyPoints=Rs.getAdjacentKeyPoints;tn.getBoundingBox=Rs.getBoundingBox;tn.getBoundingBoxPoints=Rs.getBoundingBoxPoints;tn.scaleAndFlipPoses=Rs.scaleAndFlipPoses;tn.scalePose=Rs.scalePose});var tf=Le(Hi=>{const WH=zt();function Qp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Hi.getBoxSize=Qp;function ef(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Hi.getBoxCenter=ef;function UH(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return WH.image.cropAndResize(t,a,[0],e)}Hi.cutBoxFromImageAndResize=UH;function BH(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}Hi.scaleBoxCoordinates=BH;function zH(n,t=1.5){const e=ef(n),i=Qp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Hi.enlargeBox=zH;function PH(n){const t=ef(n),e=Qp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Hi.squarifyBox=PH;function _H(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}Hi.shiftBox=_H});var WI=Le(FI=>{const qe=zt(),MH=tf();class HH{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=qe.tensor2d(this.anchors),this.inputSizeTensor=qe.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=qe.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return qe.tidy(()=>{const t=qe.slice(n,[0,0],[-1,2]),e=qe.slice(n,[0,2],[-1,2]),i=qe.add(qe.div(t,this.inputSizeTensor),this.anchorsTensor),r=qe.div(e,this.doubleInputSizeTensor),a=qe.mul(qe.sub(i,r),this.inputSizeTensor),s=qe.mul(qe.add(i,r),this.inputSizeTensor);return qe.concat2d([a,s],1)})}normalizeLandmarks(n,t){return qe.tidy(()=>{const e=qe.add(qe.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return qe.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=qe.tidy(()=>qe.mul(qe.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=qe.tidy(()=>qe.sigmoid(qe.slice(i,[0,0],[-1,1])).squeeze()),a=qe.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await qe.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=qe.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=qe.slice(s,[p,0],[1,-1]),m=qe.slice(i,[p,5],[1,14]),g=qe.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=qe.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(MH.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}FI.HandDetector=HH});var BI=Le(UI=>{UI.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var HI=Le(Vi=>{function zI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Vi.normalizeRadians=zI;function VH(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return zI(e)}Vi.computeRotation=VH;const PI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function Ta(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Vi.dot=Ta;function _I(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Vi.getColumnFrom2DArr=_I;function MI(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(Ta(n[r],_I(t,a)))}return e}function qH(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=PI(t[0],t[1]),s=MI(a,r),o=PI(-t[0],-t[1]);return MI(s,o)}Vi.buildRotationMatrix=qH;function GH(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-Ta(t[0],e),-Ta(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Vi.invertTransformMatrix=GH;function YH(n,t){return[Ta(n,t[0]),Ta(n,t[1])]}Vi.rotatePoint=YH});var YI=Le(VI=>{const qI=zt(),jn=tf(),qi=HI(),jH=.8,KH=[0,-.4],$H=[0,-.1],XH=1.65,GI=[0,5,9,13,17,1,2],JH=0,ZH=2;class QH{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return qi.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return jn.enlargeBox(jn.squarifyBox(jn.shiftBox(i,KH)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=jn.enlargeBox(jn.squarifyBox(jn.shiftBox(t,$H)),XH),i=[];for(let r=0;r<GI.length;r++)i.push(n[GI[r]].slice(0,2));return e.palmLandmarks=i,e}transformRawCoords(n,t,e,i){const r=jn.getBoxSize(t),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=qi.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=qi.rotatePoint(d,o);return[...p,d[2]]}),u=qi.invertTransformMatrix(i),c=[...jn.getBoxCenter(t),1],h=[qi.dot(c,u[0]),qi.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=qi.computeRotation(a.palmLandmarks[JH],a.palmLandmarks[ZH]),o=jn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=qI.image.rotateWithOffset(n,s,0,l),c=qi.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=jn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),v<t.minConfidence)return g.dispose(),this.regionsOfInterest[r]=[],i;const b=qI.reshape(g,[-1,3]),w=await b.array();g.dispose(),b.dispose();const S=this.transformRawCoords(w,h,s,c),L=this.getBoxForHandLandmarks(S);this.updateRegionsOfInterest(L,!1,r);const N={landmarks:S,confidence:v,box:{topLeft:L.startPoint,bottomRight:L.endPoint}};i.push(N)}return i}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>jH?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}VI.HandPipeline=QH});var $I=Le(nf=>{const Rr=zt(),eV=WI(),jI=BI(),tV=YI();class KI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=Rr.tidy(()=>(n instanceof Rr.Tensor||(n=Rr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(jI.MESH_ANNOTATIONS))s[o]=jI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}nf.HandPose=KI;async function nV(n){if(Rr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return Rr.util.fetch(n).then(t=>t.json())}async function iV(n){const[t,e,i]=await Promise.all([nV(n.detector.anchors),Rr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),Rr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new eV.HandDetector(e,t,n),a=new tV.HandPipeline(r,i,n),s=new KI(a);return s}nf.load=iV});var XI=Le(rV=>{Sl(rV,{default:()=>aV});var aV={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:5,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:2,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var ZI=Le((XV,JI)=>{JI.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});const jt=zt(),QI=eI(),bl=nI(),e2=aI(),t2=kI(),n2=$I(),i2=XI().default,sV=ZI();let Pe;const kt={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},Kn=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),wl=(...n)=>{Pe.console&&console.log(...n)};function r2(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=r2(a,s):e[r]=s}),e),{})}function oV(n){if(!n)return"input is not defined";const t=n.naturalWidth||n.videoWidth||n.width||n.shape&&n.shape[1]>0;if(!t||t===0)return"input is empty";if(n.readyState&&n.readyState<=2)return"input is not ready";try{jt.getBackend()}catch{return"backend not loaded"}return null}async function lV(n,t){Pe=r2(i2,t);const e=oV(n);return e?(wl(e,n),{error:e}):new Promise(async i=>{const r=Object.values(kt).filter(h=>h).length;r===0&&wl("Human library starting"),jt.getBackend()!==Pe.backend&&(wl("Human library setting backend:",Pe.backend),await jt.setBackend(Pe.backend),await jt.ready());let a;jt.getBackend()==="webgl"&&(a=jt.env().get("WEBGL_PACK_DEPTHWISECONV"),jt.env().set("WEBGL_PACK_DEPTHWISECONV",!0)),Pe.face.enabled&&!kt.facemesh&&(kt.facemesh=await QI.load(Pe.face)),Pe.body.enabled&&!kt.posenet&&(kt.posenet=await t2.load(Pe.body)),Pe.hand.enabled&&!kt.handpose&&(kt.handpose=await n2.load(Pe.hand)),Pe.face.enabled&&Pe.face.age.enabled&&!kt.age&&(kt.age=await bl.loadAge(Pe)),Pe.face.enabled&&Pe.face.gender.enabled&&!kt.gender&&(kt.gender=await bl.loadGender(Pe)),Pe.face.enabled&&Pe.face.emotion.enabled&&!kt.emotion&&(kt.emotion=await e2.load(Pe));const s={};let o;o=Kn(),jt.engine().startScope();const l=Pe.body.enabled?await kt.posenet.estimatePoses(n,Pe.body):[];jt.engine().endScope(),s.body=Math.trunc(Kn()-o),o=Kn(),jt.engine().startScope();const u=Pe.hand.enabled?await kt.handpose.estimateHands(n,Pe.hand):[];jt.engine().endScope(),s.hand=Math.trunc(Kn()-o);const c=[];if(Pe.face.enabled){o=Kn(),jt.engine().startScope();const h=await kt.facemesh.estimateFaces(n,Pe.face);s.face=Math.trunc(Kn()-o);for(const d of h){if(!d.image||d.image.isDisposedInternal){wl("face object is disposed:",d.image);continue}o=Kn();const p=Pe.face.age.enabled||Pe.face.gender.enabled?await bl.predict(d.image,Pe):{};s.agegender=Math.trunc(Kn()-o),o=Kn();const f=Pe.face.emotion.enabled?await e2.predict(d.image,Pe):{};s.emotion=Math.trunc(Kn()-o),d.image.dispose();const m=d.annotations.leftEyeIris&&d.annotations.rightEyeIris?Math.max(d.annotations.leftEyeIris[3][0]-d.annotations.leftEyeIris[1][0],d.annotations.rightEyeIris[3][0]-d.annotations.rightEyeIris[1][0]):0;c.push({confidence:d.confidence,box:d.box,mesh:d.mesh,annotations:d.annotations,age:p.age,gender:p.gender,emotion:f,iris:m!==0?Math.trunc(100*11.7/m)/100:0})}jt.engine().endScope()}jt.env().set("WEBGL_PACK_DEPTHWISECONV",a),s.total=Object.values(s).reduce((h,d)=>h+d),i({face:c,body:l,hand:u,performance:s})})}exports.detect=lV;exports.defaults=i2;exports.config=Pe;exports.models=kt;exports.facemesh=QI;exports.ssrnet=bl;exports.posenet=t2;exports.handpose=n2;exports.tf=jt;exports.version=sV.version;
+`,Y8=dl(G8),j8={kernelName:x.Sin,backendName:"webgl",kernelFunc:Y8};var K8="return x * x;",$8=dl(K8),X8={kernelName:x.Square,backendName:"webgl",kernelFunc:$8};var DL="return (a - b) * (a - b);",J8=Tp(DL,DL),Z8={kernelName:x.SquaredDifference,backendName:"webgl",kernelFunc:J8};var Q8="return tan(x);",e7=dl(Q8),t7={kernelName:x.Tan,backendName:"webgl",kernelFunc:e7};var n7={kernelName:x.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u<l.length;u++)l[u]=r.shape[a[u]];var c;if(s.shouldExecuteOnCPU([r])){var h=s.texData.get(r.dataId),d=h.values,p=AS(d,r.shape,r.dtype,a,l);c=s.makeTensorInfo(l,r.dtype);var f=s.texData.get(c.dataId);f.values=p}else c=EL(r,a,s);return c}};function i7(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ma(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");var s=i.readSync(a.dataId),o=EM(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var r7={kernelName:x.Unique,backendName:"webgl",kernelFunc:i7};var a7=[e8,i8,a8,u8,d8,g8,y8,L8,t8,E8,k8,W8,B8,z8,_8,H8,N8,q8,j8,X8,Z8,t7,n7,r7];for(var Cp=0,kL=a7;Cp<kL.length;Cp++){var s7=kL[Cp];x.registerKernel(s7)}qn.GPGPUContext=sL;qn.MathBackendWebGL=CL;qn.forceHalfFloat=RL;qn.gpgpu_util=w9;qn.setWebGLContext=Q0;qn.version_webgl=j6;qn.webgl=K6;qn.webgl_util=rM});var zt=Le(Nr=>{"use strict";Object.defineProperty(Nr,"__esModule",{value:!0});var Rp=lr(),Op=Ow(),Ep=Gw(),WL=h0(),o7=Z0(),l7=FL();var u7="2.6.0";var c7={"tfjs-core":Rp.version_core,"tfjs-backend-cpu":o7.version_cpu,"tfjs-backend-webgl":l7.version_webgl,"tfjs-data":WL.version_data,"tfjs-layers":Op.version_layers,"tfjs-converter":Ep.version_converter,tfjs:u7};Object.keys(Rp).forEach(function(n){n!=="default"&&Object.defineProperty(Nr,n,{enumerable:!0,get:function(){return Rp[n]}})});Object.keys(Op).forEach(function(n){n!=="default"&&Object.defineProperty(Nr,n,{enumerable:!0,get:function(){return Op[n]}})});Object.keys(Ep).forEach(function(n){n!=="default"&&Object.defineProperty(Nr,n,{enumerable:!0,get:function(){return Ep[n]}})});Nr.data=WL;Nr.version=c7});var _L=Le(pl=>{const ke=zt(),UL=6;function h7(n){const t={strides:[n/16,n/8],anchors:[2,6]},e=[];for(let i=0;i<t.strides.length;i++){const r=t.strides[i],a=Math.floor((n+r-1)/r),s=Math.floor((n+r-1)/r),o=t.anchors[i];for(let l=0;l<a;l++){const u=r*(l+.5);for(let c=0;c<s;c++){const h=r*(c+.5);for(let d=0;d<o;d++)e.push([h,u])}}}return e}const BL=n=>{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},zL=n=>({startEndTensor:n,startPoint:ke.slice(n,[0,0],[-1,2]),endPoint:ke.slice(n,[0,2],[-1,2])}),d7=(n,t)=>{const e=ke.mul(n.startPoint,t),i=ke.mul(n.endPoint,t),r=ke.concat2d([e,i],1);return zL(r)};function p7(n,t,e){const i=ke.slice(n,[0,1],[-1,2]),r=ke.add(i,t),a=ke.slice(n,[0,3],[-1,2]),s=ke.div(a,e),o=ke.div(r,e),l=ke.div(s,2),u=ke.sub(o,l),c=ke.add(o,l),h=ke.mul(u,e),d=ke.mul(c,e),p=1;return ke.concat2d([h,d],p)}function f7(n,t){return ke.tidy(()=>{const e=n.box?n.box:n;return d7(e,t).startEndTensor.squeeze()})}class PL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=h7(t.detector.inputSize),this.anchors=ke.tensor2d(this.anchorsData),this.inputSize=ke.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){if(!n||n.isDisposedInternal||n.shape.length!==4||n.shape[1]<1||n.shape[2]<1)return null;const[t,e,i]=ke.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=ke.mul(ke.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u);let h;if(Array.isArray(c)){const m=c.sort((w,S)=>w.size-S.size),g=ke.concat([m[0],m[2]],2),v=ke.concat([m[1],m[3]],2),b=ke.concat([v,g],1);h=b.squeeze(0)}else h=c.squeeze();const d=p7(h,this.anchors,this.inputSize),p=ke.slice(h,[0,0],[-1,1]),f=ke.sigmoid(p).squeeze();return[h,d,f]}),r=await ke.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>ke.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l<s.length;l++){const u=s[l],c=ke.tidy(()=>{const h=zL(u),d=a[l],p=this.anchorsData[d],f=ke.slice(t,[d,UL-1],[1,-1]).squeeze().reshape([UL,-1]),m=ke.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.width,n.shape[1]/this.height]}}async estimateFaces(n){const t=ke.tidy(()=>(n instanceof ke.Tensor||(n=ke.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:e,scaleFactor:i}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(e.map(async r=>{const a=f7(r,i),[s,o,l]=await Promise.all([r.landmarks,a,r.probability].map(async f=>f.array())),u=r.anchor,[c,h]=i,d=s.map(f=>[(f[0]+u[0])*c,(f[1]+u[1])*h]),p={topLeft:o.slice(0,2),bottomRight:o.slice(2),landmarks:d,probability:l};return BL(r.box),r.landmarks.dispose(),r.probability.dispose(),a.dispose(),p}))}}async function m7(n){const t=await ke.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new PL(t,n);return e}pl.load=m7;pl.BlazeFaceModel=PL;pl.disposeBox=BL});var kp=Le(Dp=>{Dp.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};Dp.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var ML=Le(xr=>{const g7=zt();function v7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}xr.scaleBoxCoordinates=v7;function Fp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}xr.getBoxSize=Fp;function Wp(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}xr.getBoxCenter=Wp;function y7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return g7.image.cropAndResize(t,a,[0],e)}xr.cutBoxFromImageAndResize=y7;function b7(n,t=1.5){const e=Wp(n),i=Fp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}xr.enlargeBox=b7;function w7(n){const t=Wp(n),e=Fp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}xr.squarifyBox=w7});var YL=Le(Cn=>{Cn.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function HL(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Cn.normalizeRadians=HL;function S7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return HL(e)}Cn.computeRotation=S7;function L7(n){return n*180/Math.PI}Cn.radToDegrees=L7;function VL(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function La(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Cn.dot=La;function qL(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Cn.getColumnFrom2DArr=qL;function GL(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(La(n[r],qL(t,a)))}return e}function I7(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=VL(t[0],t[1]),s=GL(a,r),o=VL(-t[0],-t[1]);return GL(s,o)}Cn.buildRotationMatrix=I7;function A7(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-La(t[0],e),-La(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Cn.invertTransformMatrix=A7;function T7(n,t){return[La(n,t[0]),La(n,t[1])]}Cn.rotatePoint=T7;function N7(n,t){return Math.sqrt((n[0]-t[0])**2+(n[1]-t[1])**2)}Cn.xyDistanceBetweenPoints=N7});var KL=Le(jL=>{const Bi=zt(),Yn=ML(),zi=kp(),Pi=YL(),x7=468,C7=.25,R7=13,O7=[R7,zi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],E7=3,D7=2,k7=[E7,D7],Up=zi.MESH_ANNOTATIONS.leftEyeLower0,Bp=[Up[0],Up[Up.length-1]],zp=zi.MESH_ANNOTATIONS.rightEyeLower0,Pp=[zp[0],zp[zp.length-1]],F7=3,W7=4,U7=71,_p=76;function fl(n,t,e,i){for(let r=0;r<zi.MESH_TO_IRIS_INDICES_MAP.length;r++){const{key:a,indices:s}=zi.MESH_TO_IRIS_INDICES_MAP[r],o=zi.MESH_ANNOTATIONS[`${e}${a}`],l=i==null;if(l||i.includes(a))for(let u=0;u<s.length;u++){const c=s[u];n[o[u]]=[t[c][0],t[c][1],(t[c][2]+n[o[u]][2])/2]}}}class B7{constructor(n,t,e,i){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.irisModel=e,this.meshWidth=i.mesh.inputSize,this.meshHeight=i.mesh.inputSize,this.irisSize=i.iris.inputSize,this.irisEnlarge=i.iris.enlargeFactor}transformRawCoords(n,t,e,i){const r=Yn.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Pi.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Pi.rotatePoint(d,o),d[2]]),u=Pi.invertTransformMatrix(i),c=[...Yn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Pi.dot(c,u[0]),Pi.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[Bp[0]][2],e=n[Pp[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Yn.squarifyBox(Yn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Yn.getBoxSize(a);let o=Bi.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Bi.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a<_p;a++){const s=n[a*3],o=n[a*3+1],l=n[a*3+2];r.push([(i?1-s/this.irisSize:s/this.irisSize)*e[0]+t.startPoint[0],o/this.irisSize*e[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(U7)}}getAdjustedIrisCoords(n,t,e){const i=n[zi.MESH_ANNOTATIONS[`${e}EyeUpper0`][F7]][2],r=n[zi.MESH_ANNOTATIONS[`${e}EyeLower0`][W7]][2],a=(i+r)/2;return t.map((s,o)=>{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Yn.scaleBoxCoordinates(o,r),u=Yn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Bi.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=x7;let[o,l]=O7;s===!1&&([o,l]=k7),a=Pi.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Yn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Pi.IDENTITY_MATRIX;a!==0&&(h=Bi.image.rotateWithOffset(n,a,0,c),d=Pi.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Yn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Bi.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:N,boxSize:C,crop:R}=this.getEyeBox(b,f,Bp[0],Bp[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,Pp[0],Pp[1]),F=this.irisModel.predict(Bi.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,_p*3),{rawCoords:_,iris:j}=this.getEyeCoords(H,N,C,!0),K=P.slice(_p*3),{rawCoords:q,iris:G}=this.getEyeCoords(K,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(fl(b,_,"left"),fl(b,q,"right")):Z<1?fl(b,_,"left",["EyeUpper0","EyeLower0"]):fl(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,j,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Bi.dispose(b);const S=Yn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const N=Bi.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:N.arraySync()};const C={coords:N,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t<n.length;t++){const e=n[t],i=this.regionsOfInterest[t];let r=0;if(i&&i.startPoint){const[a,s]=e.startPoint,[o,l]=e.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}r<C7&&(this.regionsOfInterest[t]=e)}this.regionsOfInterest=this.regionsOfInterest.slice(0,n.length)}clearRegionOfInterest(n){this.regionsOfInterest[n]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,n),...this.regionsOfInterest.slice(n+1)])}shouldUpdateRegionsOfInterest(){const n=this.regionsOfInterest.length,t=n===0;return this.maxFaces===1||t?t:n!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}jL.Pipeline=B7});var XL=Le($L=>{$L.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var JL=Le(z7=>{Ll(z7,{default:()=>P7});var P7=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var eI=Le(Ns=>{const Cr=zt(),_7=_L(),ZL=kp(),M7=KL(),H7=XL(),V7=JL().default;class QL{constructor(n,t,e,i){this.pipeline=new M7.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=Cr.tidy(()=>(n instanceof Cr.Tensor||(n=Cr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);Cr.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in ZL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=ZL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?Cr.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function q7(n){const t=await Promise.all([_7.load(n),Cr.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),Cr.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new QL(t[0],t[1],t[2],n);return e}Ns.load=q7;Ns.MediaPipeFaceMesh=QL;Ns.uv_coords=H7;Ns.triangulation=V7});var nI=Le(ml=>{const ln=zt(),_i={};let tI={age:0,gender:""},gl=0;async function G7(n,t){const e=ln.browser.fromPixels(n),i=ln.image.resizeBilinear(e,[t,t]),r=ln.cast(ln.expandDims(i,0),"float32");return r}async function Y7(n){return _i.age||(_i.age=await ln.loadGraphModel(n.face.age.modelPath)),_i.age}async function j7(n){return _i.gender||(_i.gender=await ln.loadGraphModel(n.face.gender.modelPath)),_i.gender}async function K7(n,t){if(gl>t.face.age.skipFrames?gl=0:gl+=1,gl===0)return tI;let e;if(n instanceof ln.Tensor){const r=ln.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=ln.mul(r,[255]),ln.dispose(r)}else e=await G7(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await _i.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,ln.dispose(r)}if(t.face.gender.enabled){const r=await _i.gender.predict(e),a=await r.data(),s=Math.trunc(Math.abs(1.9*100*(a[0]-.5)))/100;s>t.face.gender.minConfidence&&(i.gender=a[0]<=.5?"female":"male",i.confidence=s),ln.dispose(r)}return ln.dispose(e),tI=i,i}ml.predict=K7;ml.loadAge=Y7;ml.loadGender=j7});var aI=Le(Mp=>{const Pt=zt(),$7=["angry","discust","fear","happy","sad","surpise","neutral"],vl={};let iI=[],Hp=0;const rI=1.5;function X7(n,t){const e=Pt.tidy(()=>{const i=Pt.browser.fromPixels(n,1),r=Pt.image.resizeBilinear(i,[t,t]),a=Pt.cast(Pt.expandDims(r,0),"float32");return a});return e}async function J7(n){return vl.emotion||(vl.emotion=await Pt.loadGraphModel(n.face.emotion.modelPath)),vl.emotion}async function Z7(n,t){if(Hp+=1,Hp>=t.face.emotion.skipFrames)return Hp=0,iI;const e=Pt.tidy(()=>{if(n instanceof Pt.Tensor){const r=Pt.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Pt.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Pt.mul(a,[.2989]),u=Pt.mul(s,[.587]),c=Pt.mul(o,[.114]),h=Pt.addN([l,u,c]);return h}return s}return X7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await vl.emotion.predict(e),a=await r.data();for(let s=0;s<a.length;s++)rI*a[s]>t.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*rI*a[s])/100),emotion:$7[s]});i.sort((s,o)=>o.score-s.score),Pt.dispose(r)}return Pt.dispose(e),iI=i,i}Mp.predict=Z7;Mp.load=J7});var lI=Le(sI=>{const oI=zt();class Q7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;oI.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return oI.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}sI.BaseModel=Q7});var Vp=Le(uI=>{const cI=zt(),eH=lI();class tH extends eH.BaseModel{preprocessInput(n){return cI.tidy(()=>cI.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}uI.MobileNet=tH});var dI=Le(hI=>{function qp(n){return Math.floor(n/2)}class nH{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(qp(n),n);)this.exchange(n,qp(n)),n=qp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(n,t))break;this.exchange(n,t),n=t}}getValueAt(n){return this.getElementValue(this.priorityQueue[n])}less(n,t){return this.getValueAt(n)<this.getValueAt(t)}exchange(n,t){const e=this.priorityQueue[n];this.priorityQueue[n]=this.priorityQueue[t],this.priorityQueue[t]=e}}hI.MaxHeap=nH});var fI=Le(pI=>{const iH=dI();function rH(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;h<c;++h){const d=Math.max(i-r,0),p=Math.min(i+r+1,o);for(let f=d;f<p;++f)if(a.get(h,f,n)>t){l=!1;break}if(!l)break}return l}function aH(n,t,e){const[i,r,a]=e.shape,s=new iH.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o<i;++o)for(let l=0;l<r;++l)for(let u=0;u<a;++u){const c=e.get(o,l,u);if(c<n)continue;rH(u,c,o,l,t,e)&&s.enqueue({score:c,part:{heatmapY:o,heatmapX:l,id:u}})}return s}pI.buildPartWithScoreQueue=aH});var xs=Le(Rn=>{Rn.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];Rn.NUM_KEYPOINTS=Rn.partNames.length;Rn.partIds=Rn.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const sH=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];Rn.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];Rn.connectedPartIndices=sH.map(([n,t])=>[Rn.partIds[n],Rn.partIds[t]]);Rn.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var Yp=Le(Mi=>{const oH=xs();function mI(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+oH.NUM_KEYPOINTS)}}Mi.getOffsetPoint=mI;function lH(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=mI(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Mi.getImageCoords=lH;function uH(n,t){const e=new Array(t);for(let i=0;i<t;i++)e[i]=n;return e}Mi.fillArray=uH;function Gp(n,t,e){return n<t?t:n>e?e:n}Mi.clamp=Gp;function cH(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Mi.squaredDistance=cH;function hH(n,t){return{x:n.x+t.x,y:n.y+t.y}}Mi.addVectors=hH;function dH(n,t,e){return{y:Gp(n.y,t,e),x:Gp(n.x,t,e)}}Mi.clampVector=dH});var wI=Le(gI=>{const Cs=xs(),Ia=Yp(),vI=Cs.poseChain.map(([n,t])=>[Cs.partIds[n],Cs.partIds[t]]),jp=vI.map(([,n])=>n),yI=vI.map(([n])=>n);function pH(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function Kp(n,t,e,i){return{y:Ia.clamp(Math.round(n.y/t),0,e-1),x:Ia.clamp(Math.round(n.x/t),0,i-1)}}function bI(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=Kp(t.position,a,l,u),h=pH(n,c,s),d=Ia.addVectors(t.position,h);let p=d;for(let g=0;g<o;g++){const v=Kp(p,a,l,u),b=Ia.getOffsetPoint(v.y,v.x,e,r);p=Ia.addVectors({x:v.x*a,y:v.y*a},{x:b.x,y:b.y})}const f=Kp(p,a,l,u),m=i.get(f.y,f.x,e);return{position:p,part:Cs.partNames[e],score:m}}function fH(n,t,e,i,r,a){const s=t.shape[2],o=jp.length,l=new Array(s),{part:u,score:c}=n,h=Ia.getImageCoords(u,i,e);l[u.id]={score:c,part:Cs.partNames[u.id],position:h};for(let d=o-1;d>=0;--d){const p=jp[d],f=yI[d];l[p]&&!l[f]&&(l[f]=bI(d,l[p],f,t,e,i,a))}for(let d=0;d<o;++d){const p=yI[d],f=jp[d];l[p]&&!l[f]&&(l[f]=bI(d,l[p],f,t,e,i,r))}return l}gI.decodePose=fH});var $p=Le(SI=>{const mH=fI(),gH=wI(),LI=Yp();function II(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return LI.squaredDistance(i,e,s.y,s.x)<=t})}function vH(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(II(n,t,a,o)||(r+=s),r),0);return i/e.length}const yH=1;function bH(n,t,e,i,r,a,s=.5,o=20){const l=[],u=mH.buildPartWithScoreQueue(s,yH,n),c=o*o;for(;l.length<a&&!u.empty();){const h=u.dequeue(),d=LI.getImageCoords(h.part,r,t);if(II(l,c,d,h.part.id))continue;const p=gH.decodePose(h,n,t,r,e,i),f=vH(l,c,p);l.push({keypoints:p,score:f})}return l}SI.decodeMultiplePoses=bH});var Jp=Le(mn=>{const Aa=zt(),wH=xs();function SH(n,t,e){return n<e||t<e}function LH(n,t){return wH.connectedPartIndices.reduce((e,[i,r])=>(SH(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}mn.getAdjacentKeyPoints=LH;const{NEGATIVE_INFINITY:AI,POSITIVE_INFINITY:TI}=Number;function NI(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:AI,maxY:AI,minX:TI,minY:TI})}mn.getBoundingBox=NI;function IH(n){const{minX:t,minY:e,maxX:i,maxY:r}=NI(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}mn.getBoundingBoxPoints=IH;async function AH(n){return Promise.all(n.map(t=>t.buffer()))}mn.toTensorBuffers3D=AH;function xI(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}mn.scalePose=xI;function CI(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>xI(a,t,e,i,r))}mn.scalePoses=CI;function RI(n){return n instanceof Aa.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}mn.getInputTensorDimensions=RI;function Xp(n){return n instanceof Aa.Tensor?n:Aa.browser.fromPixels(n)}mn.toInputTensor=Xp;function TH(n,t,e){return Aa.tidy(()=>{const i=Xp(n);return i.resizeBilinear([t,e])})}mn.toResizedInputTensor=TH;function NH(n,[t,e]){const[i,r]=RI(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s<a?(o=0,l=0,u=Math.round(.5*(a*i-r)),c=Math.round(.5*(a*i-r))):(o=Math.round(.5*(1/a*r-i)),l=Math.round(.5*(1/a*r-i)),u=0,c=0);const h=Aa.tidy(()=>{let d=Xp(n);return d=Aa.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}mn.padAndResizeTo=NH;function xH(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=CI(n,s,o,-a.top,-a.left);return l}mn.scaleAndFlipPoses=xH});var EI=Le(Zp=>{const CH=zt(),RH=Vp(),OH=$p(),yl=Jp();class OI{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=yl.getInputTensorDimensions(n),{resized:a,padding:s}=yl.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await yl.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await OH.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=yl.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}Zp.PoseNet=OI;async function EH(n){const t=await CH.loadGraphModel(n.modelPath),e=new RH.MobileNet(t,n.outputStride);return new OI(e)}async function DH(n){return EH(n)}Zp.load=DH});var kI=Le(en=>{const kH=Vp(),DI=EI(),FH=$p(),bl=xs(),Rs=Jp();en.load=DI.load;en.PoseNet=DI.PoseNet;en.MobileNet=kH.MobileNet;en.decodeMultiplePoses=FH.decodeMultiplePoses;en.partChannels=bl.partChannels;en.partIds=bl.partIds;en.partNames=bl.partNames;en.poseChain=bl.poseChain;en.getAdjacentKeyPoints=Rs.getAdjacentKeyPoints;en.getBoundingBox=Rs.getBoundingBox;en.getBoundingBoxPoints=Rs.getBoundingBoxPoints;en.scaleAndFlipPoses=Rs.scaleAndFlipPoses;en.scalePose=Rs.scalePose});var tf=Le(Hi=>{const WH=zt();function Qp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Hi.getBoxSize=Qp;function ef(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Hi.getBoxCenter=ef;function UH(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return WH.image.cropAndResize(t,a,[0],e)}Hi.cutBoxFromImageAndResize=UH;function BH(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}Hi.scaleBoxCoordinates=BH;function zH(n,t=1.5){const e=ef(n),i=Qp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Hi.enlargeBox=zH;function PH(n){const t=ef(n),e=Qp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Hi.squarifyBox=PH;function _H(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}Hi.shiftBox=_H});var WI=Le(FI=>{const qe=zt(),MH=tf();class HH{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=qe.tensor2d(this.anchors),this.inputSizeTensor=qe.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=qe.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return qe.tidy(()=>{const t=qe.slice(n,[0,0],[-1,2]),e=qe.slice(n,[0,2],[-1,2]),i=qe.add(qe.div(t,this.inputSizeTensor),this.anchorsTensor),r=qe.div(e,this.doubleInputSizeTensor),a=qe.mul(qe.sub(i,r),this.inputSizeTensor),s=qe.mul(qe.add(i,r),this.inputSizeTensor);return qe.concat2d([a,s],1)})}normalizeLandmarks(n,t){return qe.tidy(()=>{const e=qe.add(qe.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return qe.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=qe.tidy(()=>qe.mul(qe.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=qe.tidy(()=>qe.sigmoid(qe.slice(i,[0,0],[-1,1])).squeeze()),a=qe.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await qe.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=qe.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=qe.slice(s,[p,0],[1,-1]),m=qe.slice(i,[p,5],[1,14]),g=qe.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=qe.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(MH.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}FI.HandDetector=HH});var BI=Le(UI=>{UI.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var HI=Le(Vi=>{function zI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Vi.normalizeRadians=zI;function VH(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return zI(e)}Vi.computeRotation=VH;const PI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function Ta(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Vi.dot=Ta;function _I(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Vi.getColumnFrom2DArr=_I;function MI(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(Ta(n[r],_I(t,a)))}return e}function qH(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=PI(t[0],t[1]),s=MI(a,r),o=PI(-t[0],-t[1]);return MI(s,o)}Vi.buildRotationMatrix=qH;function GH(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-Ta(t[0],e),-Ta(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Vi.invertTransformMatrix=GH;function YH(n,t){return[Ta(n,t[0]),Ta(n,t[1])]}Vi.rotatePoint=YH});var YI=Le(VI=>{const qI=zt(),jn=tf(),qi=HI(),jH=.8,KH=[0,-.4],$H=[0,-.1],XH=1.65,GI=[0,5,9,13,17,1,2],JH=0,ZH=2;class QH{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return qi.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return jn.enlargeBox(jn.squarifyBox(jn.shiftBox(i,KH)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=jn.enlargeBox(jn.squarifyBox(jn.shiftBox(t,$H)),XH),i=[];for(let r=0;r<GI.length;r++)i.push(n[GI[r]].slice(0,2));return e.palmLandmarks=i,e}transformRawCoords(n,t,e,i){const r=jn.getBoxSize(t),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=qi.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=qi.rotatePoint(d,o);return[...p,d[2]]}),u=qi.invertTransformMatrix(i),c=[...jn.getBoxCenter(t),1],h=[qi.dot(c,u[0]),qi.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=qi.computeRotation(a.palmLandmarks[JH],a.palmLandmarks[ZH]),o=jn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=qI.image.rotateWithOffset(n,s,0,l),c=qi.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=jn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),v<t.minConfidence)return g.dispose(),this.regionsOfInterest[r]=[],i;const b=qI.reshape(g,[-1,3]),w=await b.array();g.dispose(),b.dispose();const S=this.transformRawCoords(w,h,s,c),L=this.getBoxForHandLandmarks(S);this.updateRegionsOfInterest(L,!1,r);const N={landmarks:S,confidence:v,box:{topLeft:L.startPoint,bottomRight:L.endPoint}};i.push(N)}return i}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>jH?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}VI.HandPipeline=QH});var $I=Le(nf=>{const Rr=zt(),eV=WI(),jI=BI(),tV=YI();class KI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=Rr.tidy(()=>(n instanceof Rr.Tensor||(n=Rr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(jI.MESH_ANNOTATIONS))s[o]=jI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}nf.HandPose=KI;async function nV(n){if(Rr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return Rr.util.fetch(n).then(t=>t.json())}async function iV(n){const[t,e,i]=await Promise.all([nV(n.detector.anchors),Rr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),Rr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new eV.HandDetector(e,t,n),a=new tV.HandPipeline(r,i,n),s=new KI(a);return s}nf.load=iV});var XI=Le(rV=>{Ll(rV,{default:()=>aV});var aV={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,minConfidence:.8,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:10,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:10,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var ZI=Le((XV,JI)=>{JI.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});const On=zt(),QI=eI(),wl=nI(),e2=aI(),t2=kI(),n2=$I(),i2=XI().default,sV=ZI();let Pe;const kt={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},Kn=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),Sl=(...n)=>{Pe.console&&console.log(...n)};function r2(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=r2(a,s):e[r]=s}),e),{})}function oV(n){if(!n)return"input is not defined";const t=n.naturalWidth||n.videoWidth||n.width||n.shape&&n.shape[1]>0;if(!t||t===0)return"input is empty";if(n.readyState&&n.readyState<=2)return"input is not ready";try{On.getBackend()}catch{return"backend not loaded"}return null}async function lV(n,t){Pe=r2(i2,t);const e=oV(n);return e?(Sl(e,n),{error:e}):new Promise(async i=>{const r=Object.values(kt).filter(c=>c).length;r===0&&Sl("Human library starting"),On.getBackend()!==Pe.backend&&(Sl("Human library setting backend:",Pe.backend),await On.setBackend(Pe.backend),await On.ready()),Pe.face.enabled&&!kt.facemesh&&(kt.facemesh=await QI.load(Pe.face)),Pe.body.enabled&&!kt.posenet&&(kt.posenet=await t2.load(Pe.body)),Pe.hand.enabled&&!kt.handpose&&(kt.handpose=await n2.load(Pe.hand)),Pe.face.enabled&&Pe.face.age.enabled&&!kt.age&&(kt.age=await wl.loadAge(Pe)),Pe.face.enabled&&Pe.face.gender.enabled&&!kt.gender&&(kt.gender=await wl.loadGender(Pe)),Pe.face.enabled&&Pe.face.emotion.enabled&&!kt.emotion&&(kt.emotion=await e2.load(Pe));const a={};let s;s=Kn(),On.engine().startScope();const o=Pe.body.enabled?await kt.posenet.estimatePoses(n,Pe.body):[];On.engine().endScope(),a.body=Math.trunc(Kn()-s),s=Kn(),On.engine().startScope();const l=Pe.hand.enabled?await kt.handpose.estimateHands(n,Pe.hand):[];On.engine().endScope(),a.hand=Math.trunc(Kn()-s);const u=[];if(Pe.face.enabled){s=Kn(),On.engine().startScope();const c=await kt.facemesh.estimateFaces(n,Pe.face);a.face=Math.trunc(Kn()-s);for(const h of c){if(!h.image||h.image.isDisposedInternal){Sl("face object is disposed:",h.image);continue}s=Kn();const d=Pe.face.age.enabled||Pe.face.gender.enabled?await wl.predict(h.image,Pe):{};a.agegender=Math.trunc(Kn()-s),s=Kn();const p=Pe.face.emotion.enabled?await e2.predict(h.image,Pe):{};a.emotion=Math.trunc(Kn()-s),h.image.dispose();const f=h.annotations.leftEyeIris&&h.annotations.rightEyeIris?Math.max(h.annotations.leftEyeIris[3][0]-h.annotations.leftEyeIris[1][0],h.annotations.rightEyeIris[3][0]-h.annotations.rightEyeIris[1][0]):0;u.push({confidence:h.confidence,box:h.box,mesh:h.mesh,annotations:h.annotations,age:d.age,gender:d.gender,agConfidence:d.confidence,emotion:p,iris:f!==0?Math.trunc(100*11.7/f)/100:0})}On.engine().endScope()}a.total=Object.values(a).reduce((c,h)=>c+h),i({face:u,body:o,hand:l,performance:a})})}exports.detect=lV;exports.defaults=i2;exports.config=Pe;exports.models=kt;exports.facemesh=QI;exports.ssrnet=wl;exports.posenet=t2;exports.handpose=n2;exports.tf=On;exports.version=sV.version;
 /*! *****************************************************************************
 Copyright (c) Microsoft Corporation. All rights reserved.
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use
diff --git a/dist/human.cjs.map b/dist/human.cjs.map
index 8e705660..dcdd7cb2 100644
--- a/dist/human.cjs.map
+++ b/dist/human.cjs.map
@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../node_modules/node-fetch/lib/index.mjs", "../node_modules/@tensorflow/tfjs-core/src/backends/backend.ts", "../node_modules/@tensorflow/tfjs-core/src/environment.ts", "../node_modules/@tensorflow/tfjs-core/src/global_util.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_names.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/util.ts", "../node_modules/@tensorflow/tfjs-core/src/profiler.ts", "../node_modules/@tensorflow/tfjs-core/src/tape.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_format.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/types.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util.ts", "../node_modules/@tensorflow/tfjs-core/src/engine.ts", "../node_modules/@tensorflow/tfjs-core/src/device_util.ts", "../node_modules/@tensorflow/tfjs-core/src/flags.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util_env.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/operation.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/complex.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/io/types.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/io/router_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/io/indexed_db.ts", "../node_modules/@tensorflow/tfjs-core/src/io/local_storage.ts", "../node_modules/@tensorflow/tfjs-core/src/io/model_management.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_browser.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_node.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/buffer.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clone.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/print.ts", "../node_modules/@tensorflow/tfjs-core/src/base_side_effects.ts", "../node_modules/@tensorflow/tfjs-core/src/io/browser_files.ts", "../node_modules/@tensorflow/tfjs-core/src/io/progress.ts", "../node_modules/@tensorflow/tfjs-core/src/io/weights_loader.ts", "../node_modules/@tensorflow/tfjs-core/src/io/http.ts", "../node_modules/@tensorflow/tfjs-core/src/io/passthrough.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/confusion_matrix.ts", "../node_modules/@tensorflow/tfjs-core/src/math.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/browser.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice_util.ts", "../node_modules/@tensorflow/tfjs-core/src/serialization.ts", "../node_modules/@tensorflow/tfjs-core/src/test_util.ts", "../node_modules/@tensorflow/tfjs-core/src/version.ts", "../node_modules/@tensorflow/tfjs-core/src/globals.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add_n.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/axis_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/basic_lstm_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/diag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/eye.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fill.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reduce_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/segment_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/imag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linspace.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_with_argmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moments.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multi_rnn_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multinomial.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/real.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/outer_product.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rand.ts", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-core/src/ops/rand_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_gamma.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_uniform.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/range.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/setdiff1d_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor5d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor6d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/truncated_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/variable.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/where_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/boolean_mask.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/compare.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/binary_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moving_average.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/in_top_k.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hamming_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hann_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/frame.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/stft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/crop_and_resize.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/flip_left_right.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/rotate_with_offset.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/nonmax_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/array_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/non_max_suppression_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/band_part.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/gram_schmidt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/qr.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/loss_ops_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/compute_weighted_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/absolute_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/cosine_distance.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/hinge_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/huber_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/log_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/mean_squared_error.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/sigmoid_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/softmax_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ops.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adadelta_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adagrad_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adam_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adamax_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/sgd_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/momentum_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/rmsprop_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer_constructors.ts", "../node_modules/@tensorflow/tfjs-core/src/train.ts", "../node_modules/@tensorflow/tfjs-core/src/browser_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rotate_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/array_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf_util.ts", "../node_modules/@tensorflow/tfjs-core/src/log.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/complex_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/backend_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/split_shared.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/tile_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/topk_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/kernel_impls.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Abs_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Add_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AddN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchMatMul_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchToSpaceND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BroadcastTo_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cast_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Ceil_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ClipByValue_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Concat_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2DBackpropInput_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cumsum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/DepthwiseConv2dNative_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Dilation2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Div_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Elu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Erf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Exp_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Expm1_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Floor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FloorDiv_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FusedBatchNorm_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GatherV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GreaterEqual_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Identity_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsFinite_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsInf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsNan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log1p_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LogSoftmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LRN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/min_max_grad_util.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Max_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Maximum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Min_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Minimum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Mod_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Multiply_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Negate_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OneHot_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OnesLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/PadV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Pow_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Prelu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reciprocal_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu6_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reshape_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeBilinear_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeNearestNeighbor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reverse_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Round_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Rsqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SelectV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Selu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sigmoid_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sign_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Slice_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softplus_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SpaceToBatchND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SplitV_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Square_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SquaredDifference_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Step_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sub_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tile_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Transpose_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Unpack_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/UnsortedSegmentSum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ZerosLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/register_all_gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_type.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as3d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as4d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as5d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2D_deprecated.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/flatten.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape_as.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_bool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_float.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_int.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/errors.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/generic_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/math_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/tfjs_backend.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/initializer_config.ts", "../node_modules/@tensorflow/tfjs-layers/src/initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/state.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/types_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/variable_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/variables.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/topology.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/input_layer.ts", "../node_modules/@tensorflow/tfjs-layers/src/logs.ts", "../node_modules/@tensorflow/tfjs-layers/src/base_callbacks.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/serialization.ts", "../node_modules/@tensorflow/tfjs-layers/src/losses.ts", "../node_modules/@tensorflow/tfjs-layers/src/metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/optimizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/user_defined_metadata.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/layer_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/serialization_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/version.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/executor.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/container.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_dataset.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_tensors.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training.ts", "../node_modules/@tensorflow/tfjs-layers/src/models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports.ts", "../node_modules/@tensorflow/tfjs-layers/src/activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/advanced_activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/conv_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_depthwise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/core.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/embeddings.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/merge.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/noise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/normalization.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/padding.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/pooling.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/wrappers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_layers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/callbacks.ts", "../node_modules/@tensorflow/tfjs-converter/src/data/compiled_api.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/register.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/arithmetic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/basic_math.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/control.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/convolution.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/creation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/dynamic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/evaluation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/graph.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/image.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/logical.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/matrices.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/normalization.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/reduction.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/slice_join.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/spectral.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/transformation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_mapper.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/node_value_impl.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/arithmetic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/basic_math_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_array.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_list.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/control_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/convolution_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/creation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/dynamic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/evaluation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/image_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/logical_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/matrices_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/normalization_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/reduction_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/slice_join_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/spectral_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/transformation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/execution_context.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/model_analysis.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_model.ts", "../node_modules/@tensorflow/tfjs-converter/src/version.ts", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-data/src/util/deep_map.ts", "../node_modules/@tensorflow/tfjs-data/src/util/deep_clone.ts", "../node_modules/@tensorflow/tfjs-data/src/util/ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/util/growing_ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/lazy_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/text_line_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/csv_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/microphone_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/webcam_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/datasource.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/string_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/byte_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/file_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/url_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/util/source_util.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/file_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/url_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/readers.ts", "../node_modules/@tensorflow/tfjs-data/src/version.ts", "../node_modules/seedrandom/lib/alea.js", "../node_modules/seedrandom/lib/xor128.js", "../node_modules/seedrandom/lib/xorwow.js", "../node_modules/seedrandom/lib/xorshift7.js", "../node_modules/seedrandom/lib/xor4096.js", "../node_modules/seedrandom/lib/tychei.js", "../node_modules/seedrandom/seedrandom.js", "../node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-backend-cpu/src/cpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/backend_cpu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Abs.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/binary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Complex.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Real.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cast.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/kernel_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Add.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Ceil.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Exp.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Expm1.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Floor.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Multiply.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Rsqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Slice.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sub.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/shared.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/pool_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Clip.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Imag.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Concat.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2D.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropFilter.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropInput.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Elu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Erf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/fft_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IFFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsFinite.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsInf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsNaN.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log1p.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/LogicalNot.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NotEqual.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/PadV2.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reciprocal.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Round.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Selu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sigmoid.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sign.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Softplus.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SpaceToBatchND.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Step.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/canvas_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tex_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flags_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Abs.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/binary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Add.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/unary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Ceil.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Exp.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Expm1.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Floor.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Log.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Max_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Multiply.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Rsqrt.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Slice.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Sub.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Transpose_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Unique_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/shared.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/packing_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/glsl_version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/avg_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_complex_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/complex_abs_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_packed_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/crop_and_resize_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/cumsum_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/depth_to_space_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/diag_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fft_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fill_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_nd_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_context.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_math.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/im2col_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_grad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/max_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/mulmat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/multinomial_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/onehot_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pool_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reduce_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reshape_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/scatter_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/segment_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/select_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/strided_slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/texture_manager.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tile_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unpack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/backend_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/kernel_funcs_utils.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Atan2.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flip_left_right_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reduce.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV3.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/rotate_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs/src/version.ts", "../node_modules/@tensorflow/tfjs/src/index.ts", "../src/facemesh/blazeface.js", "../src/facemesh/keypoints.js", "../src/facemesh/box.js", "../src/facemesh/util.js", "../src/facemesh/pipeline.js", "../src/facemesh/uvcoords.js", "../src/facemesh/triangulation.js", "../src/facemesh/facemesh.js", "../src/ssrnet/ssrnet.js", "../src/emotion/emotion.js", "../src/posenet/modelBase.js", "../src/posenet/modelMobileNet.js", "../src/posenet/heapSort.js", "../src/posenet/buildParts.js", "../src/posenet/keypoints.js", "../src/posenet/vectors.js", "../src/posenet/decodePose.js", "../src/posenet/decodeMultiple.js", "../src/posenet/util.js", "../src/posenet/modelPoseNet.js", "../src/posenet/posenet.js", "../src/handpose/box.js", "../src/handpose/handdetector.js", "../src/handpose/keypoints.js", "../src/handpose/util.js", "../src/handpose/pipeline.js", "../src/handpose/handpose.js", "../src/config.js", "../src/index.js"],
-  "sourcesContent": ["import Stream from 'stream';\nimport http from 'http';\nimport Url from 'url';\nimport https from 'https';\nimport zlib from 'zlib';\n\n// Based on https://github.com/tmpvar/jsdom/blob/aa85b2abf07766ff7bf5c1f6daafb3726f2f2db5/lib/jsdom/living/blob.js\n\n// fix for \"Readable\" isn't a named export issue\nconst Readable = Stream.Readable;\n\nconst BUFFER = Symbol('buffer');\nconst TYPE = Symbol('type');\n\nclass Blob {\n\tconstructor() {\n\t\tthis[TYPE] = '';\n\n\t\tconst blobParts = arguments[0];\n\t\tconst options = arguments[1];\n\n\t\tconst buffers = [];\n\t\tlet size = 0;\n\n\t\tif (blobParts) {\n\t\t\tconst a = blobParts;\n\t\t\tconst length = Number(a.length);\n\t\t\tfor (let i = 0; i < length; i++) {\n\t\t\t\tconst element = a[i];\n\t\t\t\tlet buffer;\n\t\t\t\tif (element instanceof Buffer) {\n\t\t\t\t\tbuffer = element;\n\t\t\t\t} else if (ArrayBuffer.isView(element)) {\n\t\t\t\t\tbuffer = Buffer.from(element.buffer, element.byteOffset, element.byteLength);\n\t\t\t\t} else if (element instanceof ArrayBuffer) {\n\t\t\t\t\tbuffer = Buffer.from(element);\n\t\t\t\t} else if (element instanceof Blob) {\n\t\t\t\t\tbuffer = element[BUFFER];\n\t\t\t\t} else {\n\t\t\t\t\tbuffer = Buffer.from(typeof element === 'string' ? element : String(element));\n\t\t\t\t}\n\t\t\t\tsize += buffer.length;\n\t\t\t\tbuffers.push(buffer);\n\t\t\t}\n\t\t}\n\n\t\tthis[BUFFER] = Buffer.concat(buffers);\n\n\t\tlet type = options && options.type !== undefined && String(options.type).toLowerCase();\n\t\tif (type && !/[^\\u0020-\\u007E]/.test(type)) {\n\t\t\tthis[TYPE] = type;\n\t\t}\n\t}\n\tget size() {\n\t\treturn this[BUFFER].length;\n\t}\n\tget type() {\n\t\treturn this[TYPE];\n\t}\n\ttext() {\n\t\treturn Promise.resolve(this[BUFFER].toString());\n\t}\n\tarrayBuffer() {\n\t\tconst buf = this[BUFFER];\n\t\tconst ab = buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n\t\treturn Promise.resolve(ab);\n\t}\n\tstream() {\n\t\tconst readable = new Readable();\n\t\treadable._read = function () {};\n\t\treadable.push(this[BUFFER]);\n\t\treadable.push(null);\n\t\treturn readable;\n\t}\n\ttoString() {\n\t\treturn '[object Blob]';\n\t}\n\tslice() {\n\t\tconst size = this.size;\n\n\t\tconst start = arguments[0];\n\t\tconst end = arguments[1];\n\t\tlet relativeStart, relativeEnd;\n\t\tif (start === undefined) {\n\t\t\trelativeStart = 0;\n\t\t} else if (start < 0) {\n\t\t\trelativeStart = Math.max(size + start, 0);\n\t\t} else {\n\t\t\trelativeStart = Math.min(start, size);\n\t\t}\n\t\tif (end === undefined) {\n\t\t\trelativeEnd = size;\n\t\t} else if (end < 0) {\n\t\t\trelativeEnd = Math.max(size + end, 0);\n\t\t} else {\n\t\t\trelativeEnd = Math.min(end, size);\n\t\t}\n\t\tconst span = Math.max(relativeEnd - relativeStart, 0);\n\n\t\tconst buffer = this[BUFFER];\n\t\tconst slicedBuffer = buffer.slice(relativeStart, relativeStart + span);\n\t\tconst blob = new Blob([], { type: arguments[2] });\n\t\tblob[BUFFER] = slicedBuffer;\n\t\treturn blob;\n\t}\n}\n\nObject.defineProperties(Blob.prototype, {\n\tsize: { enumerable: true },\n\ttype: { enumerable: true },\n\tslice: { enumerable: true }\n});\n\nObject.defineProperty(Blob.prototype, Symbol.toStringTag, {\n\tvalue: 'Blob',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\n/**\n * fetch-error.js\n *\n * FetchError interface for operational errors\n */\n\n/**\n * Create FetchError instance\n *\n * @param   String      message      Error message for human\n * @param   String      type         Error type for machine\n * @param   String      systemError  For Node.js system error\n * @return  FetchError\n */\nfunction FetchError(message, type, systemError) {\n  Error.call(this, message);\n\n  this.message = message;\n  this.type = type;\n\n  // when err.type is `system`, err.code contains system error code\n  if (systemError) {\n    this.code = this.errno = systemError.code;\n  }\n\n  // hide custom error implementation details from end-users\n  Error.captureStackTrace(this, this.constructor);\n}\n\nFetchError.prototype = Object.create(Error.prototype);\nFetchError.prototype.constructor = FetchError;\nFetchError.prototype.name = 'FetchError';\n\nlet convert;\ntry {\n\tconvert = require('encoding').convert;\n} catch (e) {}\n\nconst INTERNALS = Symbol('Body internals');\n\n// fix an issue where \"PassThrough\" isn't a named export for node <10\nconst PassThrough = Stream.PassThrough;\n\n/**\n * Body mixin\n *\n * Ref: https://fetch.spec.whatwg.org/#body\n *\n * @param   Stream  body  Readable stream\n * @param   Object  opts  Response options\n * @return  Void\n */\nfunction Body(body) {\n\tvar _this = this;\n\n\tvar _ref = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {},\n\t    _ref$size = _ref.size;\n\n\tlet size = _ref$size === undefined ? 0 : _ref$size;\n\tvar _ref$timeout = _ref.timeout;\n\tlet timeout = _ref$timeout === undefined ? 0 : _ref$timeout;\n\n\tif (body == null) {\n\t\t// body is undefined or null\n\t\tbody = null;\n\t} else if (isURLSearchParams(body)) {\n\t\t// body is a URLSearchParams\n\t\tbody = Buffer.from(body.toString());\n\t} else if (isBlob(body)) ; else if (Buffer.isBuffer(body)) ; else if (Object.prototype.toString.call(body) === '[object ArrayBuffer]') {\n\t\t// body is ArrayBuffer\n\t\tbody = Buffer.from(body);\n\t} else if (ArrayBuffer.isView(body)) {\n\t\t// body is ArrayBufferView\n\t\tbody = Buffer.from(body.buffer, body.byteOffset, body.byteLength);\n\t} else if (body instanceof Stream) ; else {\n\t\t// none of the above\n\t\t// coerce to string then buffer\n\t\tbody = Buffer.from(String(body));\n\t}\n\tthis[INTERNALS] = {\n\t\tbody,\n\t\tdisturbed: false,\n\t\terror: null\n\t};\n\tthis.size = size;\n\tthis.timeout = timeout;\n\n\tif (body instanceof Stream) {\n\t\tbody.on('error', function (err) {\n\t\t\tconst error = err.name === 'AbortError' ? err : new FetchError(`Invalid response body while trying to fetch ${_this.url}: ${err.message}`, 'system', err);\n\t\t\t_this[INTERNALS].error = error;\n\t\t});\n\t}\n}\n\nBody.prototype = {\n\tget body() {\n\t\treturn this[INTERNALS].body;\n\t},\n\n\tget bodyUsed() {\n\t\treturn this[INTERNALS].disturbed;\n\t},\n\n\t/**\n  * Decode response as ArrayBuffer\n  *\n  * @return  Promise\n  */\n\tarrayBuffer() {\n\t\treturn consumeBody.call(this).then(function (buf) {\n\t\t\treturn buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n\t\t});\n\t},\n\n\t/**\n  * Return raw response as Blob\n  *\n  * @return Promise\n  */\n\tblob() {\n\t\tlet ct = this.headers && this.headers.get('content-type') || '';\n\t\treturn consumeBody.call(this).then(function (buf) {\n\t\t\treturn Object.assign(\n\t\t\t// Prevent copying\n\t\t\tnew Blob([], {\n\t\t\t\ttype: ct.toLowerCase()\n\t\t\t}), {\n\t\t\t\t[BUFFER]: buf\n\t\t\t});\n\t\t});\n\t},\n\n\t/**\n  * Decode response as json\n  *\n  * @return  Promise\n  */\n\tjson() {\n\t\tvar _this2 = this;\n\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\ttry {\n\t\t\t\treturn JSON.parse(buffer.toString());\n\t\t\t} catch (err) {\n\t\t\t\treturn Body.Promise.reject(new FetchError(`invalid json response body at ${_this2.url} reason: ${err.message}`, 'invalid-json'));\n\t\t\t}\n\t\t});\n\t},\n\n\t/**\n  * Decode response as text\n  *\n  * @return  Promise\n  */\n\ttext() {\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\treturn buffer.toString();\n\t\t});\n\t},\n\n\t/**\n  * Decode response as buffer (non-spec api)\n  *\n  * @return  Promise\n  */\n\tbuffer() {\n\t\treturn consumeBody.call(this);\n\t},\n\n\t/**\n  * Decode response as text, while automatically detecting the encoding and\n  * trying to decode to UTF-8 (non-spec api)\n  *\n  * @return  Promise\n  */\n\ttextConverted() {\n\t\tvar _this3 = this;\n\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\treturn convertBody(buffer, _this3.headers);\n\t\t});\n\t}\n};\n\n// In browsers, all properties are enumerable.\nObject.defineProperties(Body.prototype, {\n\tbody: { enumerable: true },\n\tbodyUsed: { enumerable: true },\n\tarrayBuffer: { enumerable: true },\n\tblob: { enumerable: true },\n\tjson: { enumerable: true },\n\ttext: { enumerable: true }\n});\n\nBody.mixIn = function (proto) {\n\tfor (const name of Object.getOwnPropertyNames(Body.prototype)) {\n\t\t// istanbul ignore else: future proof\n\t\tif (!(name in proto)) {\n\t\t\tconst desc = Object.getOwnPropertyDescriptor(Body.prototype, name);\n\t\t\tObject.defineProperty(proto, name, desc);\n\t\t}\n\t}\n};\n\n/**\n * Consume and convert an entire Body to a Buffer.\n *\n * Ref: https://fetch.spec.whatwg.org/#concept-body-consume-body\n *\n * @return  Promise\n */\nfunction consumeBody() {\n\tvar _this4 = this;\n\n\tif (this[INTERNALS].disturbed) {\n\t\treturn Body.Promise.reject(new TypeError(`body used already for: ${this.url}`));\n\t}\n\n\tthis[INTERNALS].disturbed = true;\n\n\tif (this[INTERNALS].error) {\n\t\treturn Body.Promise.reject(this[INTERNALS].error);\n\t}\n\n\tlet body = this.body;\n\n\t// body is null\n\tif (body === null) {\n\t\treturn Body.Promise.resolve(Buffer.alloc(0));\n\t}\n\n\t// body is blob\n\tif (isBlob(body)) {\n\t\tbody = body.stream();\n\t}\n\n\t// body is buffer\n\tif (Buffer.isBuffer(body)) {\n\t\treturn Body.Promise.resolve(body);\n\t}\n\n\t// istanbul ignore if: should never happen\n\tif (!(body instanceof Stream)) {\n\t\treturn Body.Promise.resolve(Buffer.alloc(0));\n\t}\n\n\t// body is stream\n\t// get ready to actually consume the body\n\tlet accum = [];\n\tlet accumBytes = 0;\n\tlet abort = false;\n\n\treturn new Body.Promise(function (resolve, reject) {\n\t\tlet resTimeout;\n\n\t\t// allow timeout on slow response body\n\t\tif (_this4.timeout) {\n\t\t\tresTimeout = setTimeout(function () {\n\t\t\t\tabort = true;\n\t\t\t\treject(new FetchError(`Response timeout while trying to fetch ${_this4.url} (over ${_this4.timeout}ms)`, 'body-timeout'));\n\t\t\t}, _this4.timeout);\n\t\t}\n\n\t\t// handle stream errors\n\t\tbody.on('error', function (err) {\n\t\t\tif (err.name === 'AbortError') {\n\t\t\t\t// if the request was aborted, reject with this Error\n\t\t\t\tabort = true;\n\t\t\t\treject(err);\n\t\t\t} else {\n\t\t\t\t// other errors, such as incorrect content-encoding\n\t\t\t\treject(new FetchError(`Invalid response body while trying to fetch ${_this4.url}: ${err.message}`, 'system', err));\n\t\t\t}\n\t\t});\n\n\t\tbody.on('data', function (chunk) {\n\t\t\tif (abort || chunk === null) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (_this4.size && accumBytes + chunk.length > _this4.size) {\n\t\t\t\tabort = true;\n\t\t\t\treject(new FetchError(`content size at ${_this4.url} over limit: ${_this4.size}`, 'max-size'));\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\taccumBytes += chunk.length;\n\t\t\taccum.push(chunk);\n\t\t});\n\n\t\tbody.on('end', function () {\n\t\t\tif (abort) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tclearTimeout(resTimeout);\n\n\t\t\ttry {\n\t\t\t\tresolve(Buffer.concat(accum, accumBytes));\n\t\t\t} catch (err) {\n\t\t\t\t// handle streams that have accumulated too much data (issue #414)\n\t\t\t\treject(new FetchError(`Could not create Buffer from response body for ${_this4.url}: ${err.message}`, 'system', err));\n\t\t\t}\n\t\t});\n\t});\n}\n\n/**\n * Detect buffer encoding and convert to target encoding\n * ref: http://www.w3.org/TR/2011/WD-html5-20110113/parsing.html#determining-the-character-encoding\n *\n * @param   Buffer  buffer    Incoming buffer\n * @param   String  encoding  Target encoding\n * @return  String\n */\nfunction convertBody(buffer, headers) {\n\tif (typeof convert !== 'function') {\n\t\tthrow new Error('The package `encoding` must be installed to use the textConverted() function');\n\t}\n\n\tconst ct = headers.get('content-type');\n\tlet charset = 'utf-8';\n\tlet res, str;\n\n\t// header\n\tif (ct) {\n\t\tres = /charset=([^;]*)/i.exec(ct);\n\t}\n\n\t// no charset in content type, peek at response body for at most 1024 bytes\n\tstr = buffer.slice(0, 1024).toString();\n\n\t// html5\n\tif (!res && str) {\n\t\tres = /<meta.+?charset=(['\"])(.+?)\\1/i.exec(str);\n\t}\n\n\t// html4\n\tif (!res && str) {\n\t\tres = /<meta[\\s]+?http-equiv=(['\"])content-type\\1[\\s]+?content=(['\"])(.+?)\\2/i.exec(str);\n\t\tif (!res) {\n\t\t\tres = /<meta[\\s]+?content=(['\"])(.+?)\\1[\\s]+?http-equiv=(['\"])content-type\\3/i.exec(str);\n\t\t\tif (res) {\n\t\t\t\tres.pop(); // drop last quote\n\t\t\t}\n\t\t}\n\n\t\tif (res) {\n\t\t\tres = /charset=(.*)/i.exec(res.pop());\n\t\t}\n\t}\n\n\t// xml\n\tif (!res && str) {\n\t\tres = /<\\?xml.+?encoding=(['\"])(.+?)\\1/i.exec(str);\n\t}\n\n\t// found charset\n\tif (res) {\n\t\tcharset = res.pop();\n\n\t\t// prevent decode issues when sites use incorrect encoding\n\t\t// ref: https://hsivonen.fi/encoding-menu/\n\t\tif (charset === 'gb2312' || charset === 'gbk') {\n\t\t\tcharset = 'gb18030';\n\t\t}\n\t}\n\n\t// turn raw buffers into a single utf-8 buffer\n\treturn convert(buffer, 'UTF-8', charset).toString();\n}\n\n/**\n * Detect a URLSearchParams object\n * ref: https://github.com/bitinn/node-fetch/issues/296#issuecomment-307598143\n *\n * @param   Object  obj     Object to detect by type or brand\n * @return  String\n */\nfunction isURLSearchParams(obj) {\n\t// Duck-typing as a necessary condition.\n\tif (typeof obj !== 'object' || typeof obj.append !== 'function' || typeof obj.delete !== 'function' || typeof obj.get !== 'function' || typeof obj.getAll !== 'function' || typeof obj.has !== 'function' || typeof obj.set !== 'function') {\n\t\treturn false;\n\t}\n\n\t// Brand-checking and more duck-typing as optional condition.\n\treturn obj.constructor.name === 'URLSearchParams' || Object.prototype.toString.call(obj) === '[object URLSearchParams]' || typeof obj.sort === 'function';\n}\n\n/**\n * Check if `obj` is a W3C `Blob` object (which `File` inherits from)\n * @param  {*} obj\n * @return {boolean}\n */\nfunction isBlob(obj) {\n\treturn typeof obj === 'object' && typeof obj.arrayBuffer === 'function' && typeof obj.type === 'string' && typeof obj.stream === 'function' && typeof obj.constructor === 'function' && typeof obj.constructor.name === 'string' && /^(Blob|File)$/.test(obj.constructor.name) && /^(Blob|File)$/.test(obj[Symbol.toStringTag]);\n}\n\n/**\n * Clone body given Res/Req instance\n *\n * @param   Mixed  instance  Response or Request instance\n * @return  Mixed\n */\nfunction clone(instance) {\n\tlet p1, p2;\n\tlet body = instance.body;\n\n\t// don't allow cloning a used body\n\tif (instance.bodyUsed) {\n\t\tthrow new Error('cannot clone body after it is used');\n\t}\n\n\t// check that body is a stream and not form-data object\n\t// note: we can't clone the form-data object without having it as a dependency\n\tif (body instanceof Stream && typeof body.getBoundary !== 'function') {\n\t\t// tee instance body\n\t\tp1 = new PassThrough();\n\t\tp2 = new PassThrough();\n\t\tbody.pipe(p1);\n\t\tbody.pipe(p2);\n\t\t// set instance body to teed body and return the other teed body\n\t\tinstance[INTERNALS].body = p1;\n\t\tbody = p2;\n\t}\n\n\treturn body;\n}\n\n/**\n * Performs the operation \"extract a `Content-Type` value from |object|\" as\n * specified in the specification:\n * https://fetch.spec.whatwg.org/#concept-bodyinit-extract\n *\n * This function assumes that instance.body is present.\n *\n * @param   Mixed  instance  Any options.body input\n */\nfunction extractContentType(body) {\n\tif (body === null) {\n\t\t// body is null\n\t\treturn null;\n\t} else if (typeof body === 'string') {\n\t\t// body is string\n\t\treturn 'text/plain;charset=UTF-8';\n\t} else if (isURLSearchParams(body)) {\n\t\t// body is a URLSearchParams\n\t\treturn 'application/x-www-form-urlencoded;charset=UTF-8';\n\t} else if (isBlob(body)) {\n\t\t// body is blob\n\t\treturn body.type || null;\n\t} else if (Buffer.isBuffer(body)) {\n\t\t// body is buffer\n\t\treturn null;\n\t} else if (Object.prototype.toString.call(body) === '[object ArrayBuffer]') {\n\t\t// body is ArrayBuffer\n\t\treturn null;\n\t} else if (ArrayBuffer.isView(body)) {\n\t\t// body is ArrayBufferView\n\t\treturn null;\n\t} else if (typeof body.getBoundary === 'function') {\n\t\t// detect form data input from form-data module\n\t\treturn `multipart/form-data;boundary=${body.getBoundary()}`;\n\t} else if (body instanceof Stream) {\n\t\t// body is stream\n\t\t// can't really do much about this\n\t\treturn null;\n\t} else {\n\t\t// Body constructor defaults other things to string\n\t\treturn 'text/plain;charset=UTF-8';\n\t}\n}\n\n/**\n * The Fetch Standard treats this as if \"total bytes\" is a property on the body.\n * For us, we have to explicitly get it with a function.\n *\n * ref: https://fetch.spec.whatwg.org/#concept-body-total-bytes\n *\n * @param   Body    instance   Instance of Body\n * @return  Number?            Number of bytes, or null if not possible\n */\nfunction getTotalBytes(instance) {\n\tconst body = instance.body;\n\n\n\tif (body === null) {\n\t\t// body is null\n\t\treturn 0;\n\t} else if (isBlob(body)) {\n\t\treturn body.size;\n\t} else if (Buffer.isBuffer(body)) {\n\t\t// body is buffer\n\t\treturn body.length;\n\t} else if (body && typeof body.getLengthSync === 'function') {\n\t\t// detect form data input from form-data module\n\t\tif (body._lengthRetrievers && body._lengthRetrievers.length == 0 || // 1.x\n\t\tbody.hasKnownLength && body.hasKnownLength()) {\n\t\t\t// 2.x\n\t\t\treturn body.getLengthSync();\n\t\t}\n\t\treturn null;\n\t} else {\n\t\t// body is stream\n\t\treturn null;\n\t}\n}\n\n/**\n * Write a Body to a Node.js WritableStream (e.g. http.Request) object.\n *\n * @param   Body    instance   Instance of Body\n * @return  Void\n */\nfunction writeToStream(dest, instance) {\n\tconst body = instance.body;\n\n\n\tif (body === null) {\n\t\t// body is null\n\t\tdest.end();\n\t} else if (isBlob(body)) {\n\t\tbody.stream().pipe(dest);\n\t} else if (Buffer.isBuffer(body)) {\n\t\t// body is buffer\n\t\tdest.write(body);\n\t\tdest.end();\n\t} else {\n\t\t// body is stream\n\t\tbody.pipe(dest);\n\t}\n}\n\n// expose Promise\nBody.Promise = global.Promise;\n\n/**\n * headers.js\n *\n * Headers class offers convenient helpers\n */\n\nconst invalidTokenRegex = /[^\\^_`a-zA-Z\\-0-9!#$%&'*+.|~]/;\nconst invalidHeaderCharRegex = /[^\\t\\x20-\\x7e\\x80-\\xff]/;\n\nfunction validateName(name) {\n\tname = `${name}`;\n\tif (invalidTokenRegex.test(name) || name === '') {\n\t\tthrow new TypeError(`${name} is not a legal HTTP header name`);\n\t}\n}\n\nfunction validateValue(value) {\n\tvalue = `${value}`;\n\tif (invalidHeaderCharRegex.test(value)) {\n\t\tthrow new TypeError(`${value} is not a legal HTTP header value`);\n\t}\n}\n\n/**\n * Find the key in the map object given a header name.\n *\n * Returns undefined if not found.\n *\n * @param   String  name  Header name\n * @return  String|Undefined\n */\nfunction find(map, name) {\n\tname = name.toLowerCase();\n\tfor (const key in map) {\n\t\tif (key.toLowerCase() === name) {\n\t\t\treturn key;\n\t\t}\n\t}\n\treturn undefined;\n}\n\nconst MAP = Symbol('map');\nclass Headers {\n\t/**\n  * Headers class\n  *\n  * @param   Object  headers  Response headers\n  * @return  Void\n  */\n\tconstructor() {\n\t\tlet init = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : undefined;\n\n\t\tthis[MAP] = Object.create(null);\n\n\t\tif (init instanceof Headers) {\n\t\t\tconst rawHeaders = init.raw();\n\t\t\tconst headerNames = Object.keys(rawHeaders);\n\n\t\t\tfor (const headerName of headerNames) {\n\t\t\t\tfor (const value of rawHeaders[headerName]) {\n\t\t\t\t\tthis.append(headerName, value);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\treturn;\n\t\t}\n\n\t\t// We don't worry about converting prop to ByteString here as append()\n\t\t// will handle it.\n\t\tif (init == null) ; else if (typeof init === 'object') {\n\t\t\tconst method = init[Symbol.iterator];\n\t\t\tif (method != null) {\n\t\t\t\tif (typeof method !== 'function') {\n\t\t\t\t\tthrow new TypeError('Header pairs must be iterable');\n\t\t\t\t}\n\n\t\t\t\t// sequence<sequence<ByteString>>\n\t\t\t\t// Note: per spec we have to first exhaust the lists then process them\n\t\t\t\tconst pairs = [];\n\t\t\t\tfor (const pair of init) {\n\t\t\t\t\tif (typeof pair !== 'object' || typeof pair[Symbol.iterator] !== 'function') {\n\t\t\t\t\t\tthrow new TypeError('Each header pair must be iterable');\n\t\t\t\t\t}\n\t\t\t\t\tpairs.push(Array.from(pair));\n\t\t\t\t}\n\n\t\t\t\tfor (const pair of pairs) {\n\t\t\t\t\tif (pair.length !== 2) {\n\t\t\t\t\t\tthrow new TypeError('Each header pair must be a name/value tuple');\n\t\t\t\t\t}\n\t\t\t\t\tthis.append(pair[0], pair[1]);\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\t// record<ByteString, ByteString>\n\t\t\t\tfor (const key of Object.keys(init)) {\n\t\t\t\t\tconst value = init[key];\n\t\t\t\t\tthis.append(key, value);\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tthrow new TypeError('Provided initializer must be an object');\n\t\t}\n\t}\n\n\t/**\n  * Return combined header value given name\n  *\n  * @param   String  name  Header name\n  * @return  Mixed\n  */\n\tget(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key === undefined) {\n\t\t\treturn null;\n\t\t}\n\n\t\treturn this[MAP][key].join(', ');\n\t}\n\n\t/**\n  * Iterate over all headers\n  *\n  * @param   Function  callback  Executed for each item with parameters (value, name, thisArg)\n  * @param   Boolean   thisArg   `this` context for callback function\n  * @return  Void\n  */\n\tforEach(callback) {\n\t\tlet thisArg = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : undefined;\n\n\t\tlet pairs = getHeaders(this);\n\t\tlet i = 0;\n\t\twhile (i < pairs.length) {\n\t\t\tvar _pairs$i = pairs[i];\n\t\t\tconst name = _pairs$i[0],\n\t\t\t      value = _pairs$i[1];\n\n\t\t\tcallback.call(thisArg, value, name, this);\n\t\t\tpairs = getHeaders(this);\n\t\t\ti++;\n\t\t}\n\t}\n\n\t/**\n  * Overwrite header values given name\n  *\n  * @param   String  name   Header name\n  * @param   String  value  Header value\n  * @return  Void\n  */\n\tset(name, value) {\n\t\tname = `${name}`;\n\t\tvalue = `${value}`;\n\t\tvalidateName(name);\n\t\tvalidateValue(value);\n\t\tconst key = find(this[MAP], name);\n\t\tthis[MAP][key !== undefined ? key : name] = [value];\n\t}\n\n\t/**\n  * Append a value onto existing header\n  *\n  * @param   String  name   Header name\n  * @param   String  value  Header value\n  * @return  Void\n  */\n\tappend(name, value) {\n\t\tname = `${name}`;\n\t\tvalue = `${value}`;\n\t\tvalidateName(name);\n\t\tvalidateValue(value);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key !== undefined) {\n\t\t\tthis[MAP][key].push(value);\n\t\t} else {\n\t\t\tthis[MAP][name] = [value];\n\t\t}\n\t}\n\n\t/**\n  * Check for header name existence\n  *\n  * @param   String   name  Header name\n  * @return  Boolean\n  */\n\thas(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\treturn find(this[MAP], name) !== undefined;\n\t}\n\n\t/**\n  * Delete all header values given name\n  *\n  * @param   String  name  Header name\n  * @return  Void\n  */\n\tdelete(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key !== undefined) {\n\t\t\tdelete this[MAP][key];\n\t\t}\n\t}\n\n\t/**\n  * Return raw headers (non-spec api)\n  *\n  * @return  Object\n  */\n\traw() {\n\t\treturn this[MAP];\n\t}\n\n\t/**\n  * Get an iterator on keys.\n  *\n  * @return  Iterator\n  */\n\tkeys() {\n\t\treturn createHeadersIterator(this, 'key');\n\t}\n\n\t/**\n  * Get an iterator on values.\n  *\n  * @return  Iterator\n  */\n\tvalues() {\n\t\treturn createHeadersIterator(this, 'value');\n\t}\n\n\t/**\n  * Get an iterator on entries.\n  *\n  * This is the default iterator of the Headers object.\n  *\n  * @return  Iterator\n  */\n\t[Symbol.iterator]() {\n\t\treturn createHeadersIterator(this, 'key+value');\n\t}\n}\nHeaders.prototype.entries = Headers.prototype[Symbol.iterator];\n\nObject.defineProperty(Headers.prototype, Symbol.toStringTag, {\n\tvalue: 'Headers',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nObject.defineProperties(Headers.prototype, {\n\tget: { enumerable: true },\n\tforEach: { enumerable: true },\n\tset: { enumerable: true },\n\tappend: { enumerable: true },\n\thas: { enumerable: true },\n\tdelete: { enumerable: true },\n\tkeys: { enumerable: true },\n\tvalues: { enumerable: true },\n\tentries: { enumerable: true }\n});\n\nfunction getHeaders(headers) {\n\tlet kind = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : 'key+value';\n\n\tconst keys = Object.keys(headers[MAP]).sort();\n\treturn keys.map(kind === 'key' ? function (k) {\n\t\treturn k.toLowerCase();\n\t} : kind === 'value' ? function (k) {\n\t\treturn headers[MAP][k].join(', ');\n\t} : function (k) {\n\t\treturn [k.toLowerCase(), headers[MAP][k].join(', ')];\n\t});\n}\n\nconst INTERNAL = Symbol('internal');\n\nfunction createHeadersIterator(target, kind) {\n\tconst iterator = Object.create(HeadersIteratorPrototype);\n\titerator[INTERNAL] = {\n\t\ttarget,\n\t\tkind,\n\t\tindex: 0\n\t};\n\treturn iterator;\n}\n\nconst HeadersIteratorPrototype = Object.setPrototypeOf({\n\tnext() {\n\t\t// istanbul ignore if\n\t\tif (!this || Object.getPrototypeOf(this) !== HeadersIteratorPrototype) {\n\t\t\tthrow new TypeError('Value of `this` is not a HeadersIterator');\n\t\t}\n\n\t\tvar _INTERNAL = this[INTERNAL];\n\t\tconst target = _INTERNAL.target,\n\t\t      kind = _INTERNAL.kind,\n\t\t      index = _INTERNAL.index;\n\n\t\tconst values = getHeaders(target, kind);\n\t\tconst len = values.length;\n\t\tif (index >= len) {\n\t\t\treturn {\n\t\t\t\tvalue: undefined,\n\t\t\t\tdone: true\n\t\t\t};\n\t\t}\n\n\t\tthis[INTERNAL].index = index + 1;\n\n\t\treturn {\n\t\t\tvalue: values[index],\n\t\t\tdone: false\n\t\t};\n\t}\n}, Object.getPrototypeOf(Object.getPrototypeOf([][Symbol.iterator]())));\n\nObject.defineProperty(HeadersIteratorPrototype, Symbol.toStringTag, {\n\tvalue: 'HeadersIterator',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\n/**\n * Export the Headers object in a form that Node.js can consume.\n *\n * @param   Headers  headers\n * @return  Object\n */\nfunction exportNodeCompatibleHeaders(headers) {\n\tconst obj = Object.assign({ __proto__: null }, headers[MAP]);\n\n\t// http.request() only supports string as Host header. This hack makes\n\t// specifying custom Host header possible.\n\tconst hostHeaderKey = find(headers[MAP], 'Host');\n\tif (hostHeaderKey !== undefined) {\n\t\tobj[hostHeaderKey] = obj[hostHeaderKey][0];\n\t}\n\n\treturn obj;\n}\n\n/**\n * Create a Headers object from an object of headers, ignoring those that do\n * not conform to HTTP grammar productions.\n *\n * @param   Object  obj  Object of headers\n * @return  Headers\n */\nfunction createHeadersLenient(obj) {\n\tconst headers = new Headers();\n\tfor (const name of Object.keys(obj)) {\n\t\tif (invalidTokenRegex.test(name)) {\n\t\t\tcontinue;\n\t\t}\n\t\tif (Array.isArray(obj[name])) {\n\t\t\tfor (const val of obj[name]) {\n\t\t\t\tif (invalidHeaderCharRegex.test(val)) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tif (headers[MAP][name] === undefined) {\n\t\t\t\t\theaders[MAP][name] = [val];\n\t\t\t\t} else {\n\t\t\t\t\theaders[MAP][name].push(val);\n\t\t\t\t}\n\t\t\t}\n\t\t} else if (!invalidHeaderCharRegex.test(obj[name])) {\n\t\t\theaders[MAP][name] = [obj[name]];\n\t\t}\n\t}\n\treturn headers;\n}\n\nconst INTERNALS$1 = Symbol('Response internals');\n\n// fix an issue where \"STATUS_CODES\" aren't a named export for node <10\nconst STATUS_CODES = http.STATUS_CODES;\n\n/**\n * Response class\n *\n * @param   Stream  body  Readable stream\n * @param   Object  opts  Response options\n * @return  Void\n */\nclass Response {\n\tconstructor() {\n\t\tlet body = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : null;\n\t\tlet opts = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\t\tBody.call(this, body, opts);\n\n\t\tconst status = opts.status || 200;\n\t\tconst headers = new Headers(opts.headers);\n\n\t\tif (body != null && !headers.has('Content-Type')) {\n\t\t\tconst contentType = extractContentType(body);\n\t\t\tif (contentType) {\n\t\t\t\theaders.append('Content-Type', contentType);\n\t\t\t}\n\t\t}\n\n\t\tthis[INTERNALS$1] = {\n\t\t\turl: opts.url,\n\t\t\tstatus,\n\t\t\tstatusText: opts.statusText || STATUS_CODES[status],\n\t\t\theaders,\n\t\t\tcounter: opts.counter\n\t\t};\n\t}\n\n\tget url() {\n\t\treturn this[INTERNALS$1].url || '';\n\t}\n\n\tget status() {\n\t\treturn this[INTERNALS$1].status;\n\t}\n\n\t/**\n  * Convenience property representing if the request ended normally\n  */\n\tget ok() {\n\t\treturn this[INTERNALS$1].status >= 200 && this[INTERNALS$1].status < 300;\n\t}\n\n\tget redirected() {\n\t\treturn this[INTERNALS$1].counter > 0;\n\t}\n\n\tget statusText() {\n\t\treturn this[INTERNALS$1].statusText;\n\t}\n\n\tget headers() {\n\t\treturn this[INTERNALS$1].headers;\n\t}\n\n\t/**\n  * Clone this response\n  *\n  * @return  Response\n  */\n\tclone() {\n\t\treturn new Response(clone(this), {\n\t\t\turl: this.url,\n\t\t\tstatus: this.status,\n\t\t\tstatusText: this.statusText,\n\t\t\theaders: this.headers,\n\t\t\tok: this.ok,\n\t\t\tredirected: this.redirected\n\t\t});\n\t}\n}\n\nBody.mixIn(Response.prototype);\n\nObject.defineProperties(Response.prototype, {\n\turl: { enumerable: true },\n\tstatus: { enumerable: true },\n\tok: { enumerable: true },\n\tredirected: { enumerable: true },\n\tstatusText: { enumerable: true },\n\theaders: { enumerable: true },\n\tclone: { enumerable: true }\n});\n\nObject.defineProperty(Response.prototype, Symbol.toStringTag, {\n\tvalue: 'Response',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nconst INTERNALS$2 = Symbol('Request internals');\n\n// fix an issue where \"format\", \"parse\" aren't a named export for node <10\nconst parse_url = Url.parse;\nconst format_url = Url.format;\n\nconst streamDestructionSupported = 'destroy' in Stream.Readable.prototype;\n\n/**\n * Check if a value is an instance of Request.\n *\n * @param   Mixed   input\n * @return  Boolean\n */\nfunction isRequest(input) {\n\treturn typeof input === 'object' && typeof input[INTERNALS$2] === 'object';\n}\n\nfunction isAbortSignal(signal) {\n\tconst proto = signal && typeof signal === 'object' && Object.getPrototypeOf(signal);\n\treturn !!(proto && proto.constructor.name === 'AbortSignal');\n}\n\n/**\n * Request class\n *\n * @param   Mixed   input  Url or Request instance\n * @param   Object  init   Custom options\n * @return  Void\n */\nclass Request {\n\tconstructor(input) {\n\t\tlet init = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\t\tlet parsedURL;\n\n\t\t// normalize input\n\t\tif (!isRequest(input)) {\n\t\t\tif (input && input.href) {\n\t\t\t\t// in order to support Node.js' Url objects; though WHATWG's URL objects\n\t\t\t\t// will fall into this branch also (since their `toString()` will return\n\t\t\t\t// `href` property anyway)\n\t\t\t\tparsedURL = parse_url(input.href);\n\t\t\t} else {\n\t\t\t\t// coerce input to a string before attempting to parse\n\t\t\t\tparsedURL = parse_url(`${input}`);\n\t\t\t}\n\t\t\tinput = {};\n\t\t} else {\n\t\t\tparsedURL = parse_url(input.url);\n\t\t}\n\n\t\tlet method = init.method || input.method || 'GET';\n\t\tmethod = method.toUpperCase();\n\n\t\tif ((init.body != null || isRequest(input) && input.body !== null) && (method === 'GET' || method === 'HEAD')) {\n\t\t\tthrow new TypeError('Request with GET/HEAD method cannot have body');\n\t\t}\n\n\t\tlet inputBody = init.body != null ? init.body : isRequest(input) && input.body !== null ? clone(input) : null;\n\n\t\tBody.call(this, inputBody, {\n\t\t\ttimeout: init.timeout || input.timeout || 0,\n\t\t\tsize: init.size || input.size || 0\n\t\t});\n\n\t\tconst headers = new Headers(init.headers || input.headers || {});\n\n\t\tif (inputBody != null && !headers.has('Content-Type')) {\n\t\t\tconst contentType = extractContentType(inputBody);\n\t\t\tif (contentType) {\n\t\t\t\theaders.append('Content-Type', contentType);\n\t\t\t}\n\t\t}\n\n\t\tlet signal = isRequest(input) ? input.signal : null;\n\t\tif ('signal' in init) signal = init.signal;\n\n\t\tif (signal != null && !isAbortSignal(signal)) {\n\t\t\tthrow new TypeError('Expected signal to be an instanceof AbortSignal');\n\t\t}\n\n\t\tthis[INTERNALS$2] = {\n\t\t\tmethod,\n\t\t\tredirect: init.redirect || input.redirect || 'follow',\n\t\t\theaders,\n\t\t\tparsedURL,\n\t\t\tsignal\n\t\t};\n\n\t\t// node-fetch-only options\n\t\tthis.follow = init.follow !== undefined ? init.follow : input.follow !== undefined ? input.follow : 20;\n\t\tthis.compress = init.compress !== undefined ? init.compress : input.compress !== undefined ? input.compress : true;\n\t\tthis.counter = init.counter || input.counter || 0;\n\t\tthis.agent = init.agent || input.agent;\n\t}\n\n\tget method() {\n\t\treturn this[INTERNALS$2].method;\n\t}\n\n\tget url() {\n\t\treturn format_url(this[INTERNALS$2].parsedURL);\n\t}\n\n\tget headers() {\n\t\treturn this[INTERNALS$2].headers;\n\t}\n\n\tget redirect() {\n\t\treturn this[INTERNALS$2].redirect;\n\t}\n\n\tget signal() {\n\t\treturn this[INTERNALS$2].signal;\n\t}\n\n\t/**\n  * Clone this request\n  *\n  * @return  Request\n  */\n\tclone() {\n\t\treturn new Request(this);\n\t}\n}\n\nBody.mixIn(Request.prototype);\n\nObject.defineProperty(Request.prototype, Symbol.toStringTag, {\n\tvalue: 'Request',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nObject.defineProperties(Request.prototype, {\n\tmethod: { enumerable: true },\n\turl: { enumerable: true },\n\theaders: { enumerable: true },\n\tredirect: { enumerable: true },\n\tclone: { enumerable: true },\n\tsignal: { enumerable: true }\n});\n\n/**\n * Convert a Request to Node.js http request options.\n *\n * @param   Request  A Request instance\n * @return  Object   The options object to be passed to http.request\n */\nfunction getNodeRequestOptions(request) {\n\tconst parsedURL = request[INTERNALS$2].parsedURL;\n\tconst headers = new Headers(request[INTERNALS$2].headers);\n\n\t// fetch step 1.3\n\tif (!headers.has('Accept')) {\n\t\theaders.set('Accept', '*/*');\n\t}\n\n\t// Basic fetch\n\tif (!parsedURL.protocol || !parsedURL.hostname) {\n\t\tthrow new TypeError('Only absolute URLs are supported');\n\t}\n\n\tif (!/^https?:$/.test(parsedURL.protocol)) {\n\t\tthrow new TypeError('Only HTTP(S) protocols are supported');\n\t}\n\n\tif (request.signal && request.body instanceof Stream.Readable && !streamDestructionSupported) {\n\t\tthrow new Error('Cancellation of streamed requests with AbortSignal is not supported in node < 8');\n\t}\n\n\t// HTTP-network-or-cache fetch steps 2.4-2.7\n\tlet contentLengthValue = null;\n\tif (request.body == null && /^(POST|PUT)$/i.test(request.method)) {\n\t\tcontentLengthValue = '0';\n\t}\n\tif (request.body != null) {\n\t\tconst totalBytes = getTotalBytes(request);\n\t\tif (typeof totalBytes === 'number') {\n\t\t\tcontentLengthValue = String(totalBytes);\n\t\t}\n\t}\n\tif (contentLengthValue) {\n\t\theaders.set('Content-Length', contentLengthValue);\n\t}\n\n\t// HTTP-network-or-cache fetch step 2.11\n\tif (!headers.has('User-Agent')) {\n\t\theaders.set('User-Agent', 'node-fetch/1.0 (+https://github.com/bitinn/node-fetch)');\n\t}\n\n\t// HTTP-network-or-cache fetch step 2.15\n\tif (request.compress && !headers.has('Accept-Encoding')) {\n\t\theaders.set('Accept-Encoding', 'gzip,deflate');\n\t}\n\n\tlet agent = request.agent;\n\tif (typeof agent === 'function') {\n\t\tagent = agent(parsedURL);\n\t}\n\n\tif (!headers.has('Connection') && !agent) {\n\t\theaders.set('Connection', 'close');\n\t}\n\n\t// HTTP-network fetch step 4.2\n\t// chunked encoding is handled by Node.js\n\n\treturn Object.assign({}, parsedURL, {\n\t\tmethod: request.method,\n\t\theaders: exportNodeCompatibleHeaders(headers),\n\t\tagent\n\t});\n}\n\n/**\n * abort-error.js\n *\n * AbortError interface for cancelled requests\n */\n\n/**\n * Create AbortError instance\n *\n * @param   String      message      Error message for human\n * @return  AbortError\n */\nfunction AbortError(message) {\n  Error.call(this, message);\n\n  this.type = 'aborted';\n  this.message = message;\n\n  // hide custom error implementation details from end-users\n  Error.captureStackTrace(this, this.constructor);\n}\n\nAbortError.prototype = Object.create(Error.prototype);\nAbortError.prototype.constructor = AbortError;\nAbortError.prototype.name = 'AbortError';\n\n// fix an issue where \"PassThrough\", \"resolve\" aren't a named export for node <10\nconst PassThrough$1 = Stream.PassThrough;\nconst resolve_url = Url.resolve;\n\n/**\n * Fetch function\n *\n * @param   Mixed    url   Absolute url or Request instance\n * @param   Object   opts  Fetch options\n * @return  Promise\n */\nfunction fetch(url, opts) {\n\n\t// allow custom promise\n\tif (!fetch.Promise) {\n\t\tthrow new Error('native promise missing, set fetch.Promise to your favorite alternative');\n\t}\n\n\tBody.Promise = fetch.Promise;\n\n\t// wrap http.request into fetch\n\treturn new fetch.Promise(function (resolve, reject) {\n\t\t// build request object\n\t\tconst request = new Request(url, opts);\n\t\tconst options = getNodeRequestOptions(request);\n\n\t\tconst send = (options.protocol === 'https:' ? https : http).request;\n\t\tconst signal = request.signal;\n\n\t\tlet response = null;\n\n\t\tconst abort = function abort() {\n\t\t\tlet error = new AbortError('The user aborted a request.');\n\t\t\treject(error);\n\t\t\tif (request.body && request.body instanceof Stream.Readable) {\n\t\t\t\trequest.body.destroy(error);\n\t\t\t}\n\t\t\tif (!response || !response.body) return;\n\t\t\tresponse.body.emit('error', error);\n\t\t};\n\n\t\tif (signal && signal.aborted) {\n\t\t\tabort();\n\t\t\treturn;\n\t\t}\n\n\t\tconst abortAndFinalize = function abortAndFinalize() {\n\t\t\tabort();\n\t\t\tfinalize();\n\t\t};\n\n\t\t// send request\n\t\tconst req = send(options);\n\t\tlet reqTimeout;\n\n\t\tif (signal) {\n\t\t\tsignal.addEventListener('abort', abortAndFinalize);\n\t\t}\n\n\t\tfunction finalize() {\n\t\t\treq.abort();\n\t\t\tif (signal) signal.removeEventListener('abort', abortAndFinalize);\n\t\t\tclearTimeout(reqTimeout);\n\t\t}\n\n\t\tif (request.timeout) {\n\t\t\treq.once('socket', function (socket) {\n\t\t\t\treqTimeout = setTimeout(function () {\n\t\t\t\t\treject(new FetchError(`network timeout at: ${request.url}`, 'request-timeout'));\n\t\t\t\t\tfinalize();\n\t\t\t\t}, request.timeout);\n\t\t\t});\n\t\t}\n\n\t\treq.on('error', function (err) {\n\t\t\treject(new FetchError(`request to ${request.url} failed, reason: ${err.message}`, 'system', err));\n\t\t\tfinalize();\n\t\t});\n\n\t\treq.on('response', function (res) {\n\t\t\tclearTimeout(reqTimeout);\n\n\t\t\tconst headers = createHeadersLenient(res.headers);\n\n\t\t\t// HTTP fetch step 5\n\t\t\tif (fetch.isRedirect(res.statusCode)) {\n\t\t\t\t// HTTP fetch step 5.2\n\t\t\t\tconst location = headers.get('Location');\n\n\t\t\t\t// HTTP fetch step 5.3\n\t\t\t\tconst locationURL = location === null ? null : resolve_url(request.url, location);\n\n\t\t\t\t// HTTP fetch step 5.5\n\t\t\t\tswitch (request.redirect) {\n\t\t\t\t\tcase 'error':\n\t\t\t\t\t\treject(new FetchError(`uri requested responds with a redirect, redirect mode is set to error: ${request.url}`, 'no-redirect'));\n\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\treturn;\n\t\t\t\t\tcase 'manual':\n\t\t\t\t\t\t// node-fetch-specific step: make manual redirect a bit easier to use by setting the Location header value to the resolved URL.\n\t\t\t\t\t\tif (locationURL !== null) {\n\t\t\t\t\t\t\t// handle corrupted header\n\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\theaders.set('Location', locationURL);\n\t\t\t\t\t\t\t} catch (err) {\n\t\t\t\t\t\t\t\t// istanbul ignore next: nodejs server prevent invalid response headers, we can't test this through normal request\n\t\t\t\t\t\t\t\treject(err);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\tbreak;\n\t\t\t\t\tcase 'follow':\n\t\t\t\t\t\t// HTTP-redirect fetch step 2\n\t\t\t\t\t\tif (locationURL === null) {\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 5\n\t\t\t\t\t\tif (request.counter >= request.follow) {\n\t\t\t\t\t\t\treject(new FetchError(`maximum redirect reached at: ${request.url}`, 'max-redirect'));\n\t\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 6 (counter increment)\n\t\t\t\t\t\t// Create a new Request object.\n\t\t\t\t\t\tconst requestOpts = {\n\t\t\t\t\t\t\theaders: new Headers(request.headers),\n\t\t\t\t\t\t\tfollow: request.follow,\n\t\t\t\t\t\t\tcounter: request.counter + 1,\n\t\t\t\t\t\t\tagent: request.agent,\n\t\t\t\t\t\t\tcompress: request.compress,\n\t\t\t\t\t\t\tmethod: request.method,\n\t\t\t\t\t\t\tbody: request.body,\n\t\t\t\t\t\t\tsignal: request.signal,\n\t\t\t\t\t\t\ttimeout: request.timeout,\n\t\t\t\t\t\t\tsize: request.size\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 9\n\t\t\t\t\t\tif (res.statusCode !== 303 && request.body && getTotalBytes(request) === null) {\n\t\t\t\t\t\t\treject(new FetchError('Cannot follow redirect with body being a readable stream', 'unsupported-redirect'));\n\t\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 11\n\t\t\t\t\t\tif (res.statusCode === 303 || (res.statusCode === 301 || res.statusCode === 302) && request.method === 'POST') {\n\t\t\t\t\t\t\trequestOpts.method = 'GET';\n\t\t\t\t\t\t\trequestOpts.body = undefined;\n\t\t\t\t\t\t\trequestOpts.headers.delete('content-length');\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 15\n\t\t\t\t\t\tresolve(fetch(new Request(locationURL, requestOpts)));\n\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// prepare response\n\t\t\tres.once('end', function () {\n\t\t\t\tif (signal) signal.removeEventListener('abort', abortAndFinalize);\n\t\t\t});\n\t\t\tlet body = res.pipe(new PassThrough$1());\n\n\t\t\tconst response_options = {\n\t\t\t\turl: request.url,\n\t\t\t\tstatus: res.statusCode,\n\t\t\t\tstatusText: res.statusMessage,\n\t\t\t\theaders: headers,\n\t\t\t\tsize: request.size,\n\t\t\t\ttimeout: request.timeout,\n\t\t\t\tcounter: request.counter\n\t\t\t};\n\n\t\t\t// HTTP-network fetch step 12.1.1.3\n\t\t\tconst codings = headers.get('Content-Encoding');\n\n\t\t\t// HTTP-network fetch step 12.1.1.4: handle content codings\n\n\t\t\t// in following scenarios we ignore compression support\n\t\t\t// 1. compression support is disabled\n\t\t\t// 2. HEAD request\n\t\t\t// 3. no Content-Encoding header\n\t\t\t// 4. no content response (204)\n\t\t\t// 5. content not modified response (304)\n\t\t\tif (!request.compress || request.method === 'HEAD' || codings === null || res.statusCode === 204 || res.statusCode === 304) {\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// For Node v6+\n\t\t\t// Be less strict when decoding compressed responses, since sometimes\n\t\t\t// servers send slightly invalid responses that are still accepted\n\t\t\t// by common browsers.\n\t\t\t// Always using Z_SYNC_FLUSH is what cURL does.\n\t\t\tconst zlibOptions = {\n\t\t\t\tflush: zlib.Z_SYNC_FLUSH,\n\t\t\t\tfinishFlush: zlib.Z_SYNC_FLUSH\n\t\t\t};\n\n\t\t\t// for gzip\n\t\t\tif (codings == 'gzip' || codings == 'x-gzip') {\n\t\t\t\tbody = body.pipe(zlib.createGunzip(zlibOptions));\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// for deflate\n\t\t\tif (codings == 'deflate' || codings == 'x-deflate') {\n\t\t\t\t// handle the infamous raw deflate response from old servers\n\t\t\t\t// a hack for old IIS and Apache servers\n\t\t\t\tconst raw = res.pipe(new PassThrough$1());\n\t\t\t\traw.once('data', function (chunk) {\n\t\t\t\t\t// see http://stackoverflow.com/questions/37519828\n\t\t\t\t\tif ((chunk[0] & 0x0F) === 0x08) {\n\t\t\t\t\t\tbody = body.pipe(zlib.createInflate());\n\t\t\t\t\t} else {\n\t\t\t\t\t\tbody = body.pipe(zlib.createInflateRaw());\n\t\t\t\t\t}\n\t\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\t\tresolve(response);\n\t\t\t\t});\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// for br\n\t\t\tif (codings == 'br' && typeof zlib.createBrotliDecompress === 'function') {\n\t\t\t\tbody = body.pipe(zlib.createBrotliDecompress());\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// otherwise, use response as-is\n\t\t\tresponse = new Response(body, response_options);\n\t\t\tresolve(response);\n\t\t});\n\n\t\twriteToStream(req, request);\n\t});\n}\n/**\n * Redirect code matching\n *\n * @param   Number   code  Status code\n * @return  Boolean\n */\nfetch.isRedirect = function (code) {\n\treturn code === 301 || code === 302 || code === 303 || code === 307 || code === 308;\n};\n\n// expose Promise\nfetch.Promise = global.Promise;\n\nexport default fetch;\nexport { Headers, Request, Response, FetchError };\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n  kernelMs: number|{error: string};\n  getExtraProfileInfo?(): string;  // a field for additional timing information\n                                   // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n  disposeData(dataId: DataId): void;\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void;\n  memory(): {unreliable: boolean;};  // Backend-specific information.\n  /** Returns number of data ids currently in the storage. */\n  numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage<T> {\n  private data = new WeakMap<DataId, T>();\n  private dataIdsCount = 0;\n\n  constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n  get(dataId: DataId) {\n    if (!this.data.has(dataId)) {\n      this.dataMover.moveData(this.backend, dataId);\n    }\n    return this.data.get(dataId);\n  }\n\n  set(dataId: DataId, value: T): void {\n    this.dataIdsCount++;\n    this.data.set(dataId, value);\n  }\n\n  has(dataId: DataId): boolean {\n    return this.data.has(dataId);\n  }\n\n  delete(dataId: DataId): boolean {\n    this.dataIdsCount--;\n    return this.data.delete(dataId);\n  }\n\n  numDataIds(): number {\n    return this.dataIdsCount;\n  }\n}\n\nexport interface DataMover {\n  /**\n   * To be called by backends whenever they see a dataId that they don't own.\n   * Upon calling this method, the mover will fetch the tensor from another\n   * backend and register it with the current active backend.\n   */\n  moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo>;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo> {\n    return notYetImplemented('time');\n  }\n  read(dataId: object): Promise<BackendValues> {\n    return notYetImplemented('read');\n  }\n  readSync(dataId: object): BackendValues {\n    return notYetImplemented('readSync');\n  }\n  numDataIds(): number {\n    return notYetImplemented('numDataIds');\n  }\n  disposeData(dataId: object): void {\n    return notYetImplemented('disposeData');\n  }\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    return notYetImplemented('write');\n  }\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    return notYetImplemented('move');\n  }\n  memory(): {unreliable: boolean; reasons?: string[]} {\n    return notYetImplemented('memory');\n  }\n  /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n  floatPrecision(): 16|32 {\n    return notYetImplemented('floatPrecision');\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    return notYetImplemented('batchMatMul');\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          FusedBatchMatMulConfig): Tensor3D {\n    return notYetImplemented('fusedBatchMatMul');\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    return notYetImplemented('slice');\n  }\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    return notYetImplemented('stridedSlice');\n  }\n  unstack(x: Tensor, axis: number): Tensor[] {\n    return notYetImplemented('unstack');\n  }\n  reverse<T extends Tensor>(a: T, axis: number[]): T {\n    return notYetImplemented('reverse');\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    return notYetImplemented('concat');\n  }\n\n  neg<T extends Tensor>(a: T): T {\n    return notYetImplemented('neg');\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('add');\n  }\n  addN<T extends Tensor>(tensors: T[]): T {\n    return notYetImplemented('addN');\n  }\n  subtract(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('subtract');\n  }\n  multiply(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('multiply');\n  }\n  realDivide(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('realDivide');\n  }\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('floorDiv');\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('sum');\n  }\n  prod(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('prod');\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    return notYetImplemented('unsortedSegmentSum');\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMin');\n  }\n  argMax(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMax');\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('equal');\n  }\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('notEqual');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('less');\n  }\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('lessEqual');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greater');\n  }\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greaterEqual');\n  }\n\n  logicalNot<T extends Tensor>(a: T): T {\n    return notYetImplemented('logicalNot');\n  }\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalAnd');\n  }\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalOr');\n  }\n\n  where(condition: Tensor): Tensor2D {\n    return notYetImplemented('where');\n  }\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('select');\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    return notYetImplemented('topk');\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('min');\n  }\n  minimum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('minimum');\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('mod');\n  }\n\n  max(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('max');\n  }\n  maximum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('maximum');\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('all');\n  }\n  any(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('any');\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('squaredDifference');\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    return notYetImplemented('ceil');\n  }\n  floor<T extends Tensor>(x: T): T {\n    return notYetImplemented('floor');\n  }\n  round<T extends Tensor>(x: T): T {\n    return notYetImplemented('round');\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    return notYetImplemented('sign');\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    return notYetImplemented('isNaN');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    return notYetImplemented('isInf');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    return notYetImplemented('isFinite');\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    return notYetImplemented('pow');\n  }\n  exp<T extends Tensor>(x: T): T {\n    return notYetImplemented('exp');\n  }\n  expm1<T extends Tensor>(x: T): T {\n    return notYetImplemented('expm1');\n  }\n  softmax<T extends Tensor>(x: T, dim: number): T {\n    return notYetImplemented('softmax');\n  }\n  log<T extends Tensor>(x: T): T {\n    return notYetImplemented('log');\n  }\n  log1p<T extends Tensor>(x: T): T {\n    return notYetImplemented('log1p');\n  }\n  sqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('sqrt');\n  }\n  rsqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('rsqrt');\n  }\n  square<T extends Tensor>(x: T): T {\n    return notYetImplemented('square');\n  }\n  reciprocal<T extends Tensor>(x: T): T {\n    return notYetImplemented('reciprocal');\n  }\n  relu<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu');\n  }\n  relu6<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu6');\n  }\n  prelu<T extends Tensor>(x: T, a: T): T {\n    return notYetImplemented('prelu');\n  }\n  elu<T extends Tensor>(x: T): T {\n    return notYetImplemented('elu');\n  }\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    return notYetImplemented('eluDer');\n  }\n  selu<T extends Tensor>(x: T): T {\n    return notYetImplemented('selu');\n  }\n  int<T extends Tensor>(x: T): T {\n    return notYetImplemented('int');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    return notYetImplemented('clip');\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    return notYetImplemented('abs');\n  }\n  complexAbs<T extends Tensor>(x: T): T {\n    return notYetImplemented('complexAbs');\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    return notYetImplemented('sigmoid');\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    return notYetImplemented('softplus');\n  }\n\n  sin<T extends Tensor>(x: T): T {\n    return notYetImplemented('sin');\n  }\n  cos<T extends Tensor>(x: T): T {\n    return notYetImplemented('cos');\n  }\n  tan<T extends Tensor>(x: T): T {\n    return notYetImplemented('tan');\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    return notYetImplemented('asin');\n  }\n  acos<T extends Tensor>(x: T): T {\n    return notYetImplemented('acos');\n  }\n  atan<T extends Tensor>(x: T): T {\n    return notYetImplemented('atan');\n  }\n  atan2<T extends Tensor>(a: T, b: T): T {\n    return notYetImplemented('atan2');\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('sinh');\n  }\n  cosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('cosh');\n  }\n  tanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('tanh');\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('asinh');\n  }\n  acosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('acosh');\n  }\n  atanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('atanh');\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    return notYetImplemented('erf');\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    return notYetImplemented('step');\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedConv2d');\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2d');\n  }\n  conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('conv2dDerInput');\n  }\n  conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2dDerFilter');\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedDepthwiseConv2D');\n  }\n\n  depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2D');\n  }\n  depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerInput');\n  }\n  depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerFilter');\n  }\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3d');\n  }\n  conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n      Tensor5D {\n    return notYetImplemented('conv3dDerInput');\n  }\n  conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3dDerFilter');\n  }\n  maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('maxPool');\n  }\n  maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('maxPoolBackprop');\n  }\n  avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPool');\n  }\n  avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPoolBackprop');\n  }\n  avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3d');\n  }\n  avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3dBackprop');\n  }\n  maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3d');\n  }\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3dBackprop');\n  }\n\n  reshape<T extends Tensor, R extends Rank>(x: T, shape: ShapeMap[R]):\n      Tensor<R> {\n    return notYetImplemented('reshape');\n  }\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return notYetImplemented('cast');\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    return notYetImplemented('tile');\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    return notYetImplemented('pad');\n  }\n\n  transpose<T extends Tensor>(x: T, perm: number[]): T {\n    return notYetImplemented('transpose');\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    return notYetImplemented('gather');\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    return notYetImplemented('gatherND');\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    return notYetImplemented('scatterND');\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    return notYetImplemented('batchToSpaceND');\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: number[][]): T {\n    return notYetImplemented('spaceToBatchND');\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeBilinear');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    return notYetImplemented('resizeBilinearBackprop');\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHEight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighbor');\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighborBackprop');\n  }\n\n  batchNorm(\n      x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n      offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n      varianceEpsilon?: number): Tensor4D {\n    return notYetImplemented('batchNorm');\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    return notYetImplemented('localResponseNormalization4D');\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n      bias: number, alpha: number, beta: number): Tensor4D {\n    return notYetImplemented('LRNGrad');\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    return notYetImplemented('multinomial');\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    return notYetImplemented('oneHot');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    return notYetImplemented('cumsum');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold?: number): Tensor1D {\n    return notYetImplemented('nonMaxSuppression');\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('fft');\n  }\n  ifft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('ifft');\n  }\n  complex<T extends Tensor>(real: T, imag: T): T {\n    return notYetImplemented('complex');\n  }\n  real<T extends Tensor>(input: T): T {\n    return notYetImplemented('real');\n  }\n  imag<T extends Tensor>(input: T): T {\n    return notYetImplemented('imag');\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    return notYetImplemented('cropAndResize');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n    return notYetImplemented('depthToSpace');\n  }\n\n  // Aligns with the \"SplitV\" kernel in TensorFlow.\n  split<T extends Tensor>(value: T, sizeSplits: number[], axis: number): T[] {\n    return notYetImplemented('split');\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    return notYetImplemented('sparseToDense');\n  }\n\n  diag(x: Tensor): Tensor {\n    return notYetImplemented('diag');\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    return notYetImplemented('fill');\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('onesLike');\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('zerosLike');\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return notYetImplemented('linspace');\n  }\n\n  dispose(): void {\n    return notYetImplemented('dispose');\n  }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n  throw new Error(\n      `'${kernelName}' not yet implemented or not found in the registry. ` +\n      `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise<FlagValue>);\nexport type Flags = {\n  [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n  evaluationFn: FlagEvaluationFn;\n  setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n  private flags: Flags = {};\n  private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n  private urlFlags: Flags = {};\n\n  platformName: string;\n  platform: Platform;\n\n  // tslint:disable-next-line: no-any\n  constructor(public global: any) {\n    this.populateURLFlags();\n  }\n\n  setPlatform(platformName: string, platform: Platform) {\n    if (this.platform != null) {\n      console.warn(\n          `Platform ${this.platformName} has already been set. ` +\n          `Overwriting the platform with ${platform}.`);\n    }\n    this.platformName = platformName;\n    this.platform = platform;\n  }\n\n  registerFlag(\n      flagName: string, evaluationFn: FlagEvaluationFn,\n      setHook?: (value: FlagValue) => void) {\n    this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n    // Override the flag value from the URL. This has to happen here because the\n    // environment is initialized before flags get registered.\n    if (this.urlFlags[flagName] != null) {\n      const flagValue = this.urlFlags[flagName];\n      console.warn(\n          `Setting feature override from URL ${flagName}: ${flagValue}.`);\n      this.set(flagName, flagValue);\n    }\n  }\n\n  async getAsync(flagName: string): Promise<FlagValue> {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    this.flags[flagName] = await this.evaluateFlag(flagName);\n    return this.flags[flagName];\n  }\n\n  get(flagName: string): FlagValue {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    const flagValue = this.evaluateFlag(flagName);\n    if (flagValue instanceof Promise) {\n      throw new Error(\n          `Flag ${flagName} cannot be synchronously evaluated. ` +\n          `Please use getAsync() instead.`);\n    }\n\n    this.flags[flagName] = flagValue;\n\n    return this.flags[flagName];\n  }\n\n  getNumber(flagName: string): number {\n    return this.get(flagName) as number;\n  }\n\n  getBool(flagName: string): boolean {\n    return this.get(flagName) as boolean;\n  }\n\n  getFlags(): Flags {\n    return this.flags;\n  }\n  // For backwards compatibility.\n  get features(): Flags {\n    return this.flags;\n  }\n\n  set(flagName: string, value: FlagValue): void {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot set flag ${flagName} as it has not been registered.`);\n    }\n    this.flags[flagName] = value;\n    if (this.flagRegistry[flagName].setHook != null) {\n      this.flagRegistry[flagName].setHook(value);\n    }\n  }\n\n  private evaluateFlag(flagName: string): FlagValue|Promise<FlagValue> {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n    }\n    return this.flagRegistry[flagName].evaluationFn();\n  }\n\n  setFlags(flags: Flags) {\n    this.flags = Object.assign({}, flags);\n  }\n\n  reset() {\n    this.flags = {};\n    this.urlFlags = {};\n    this.populateURLFlags();\n  }\n\n  private populateURLFlags(): void {\n    if (typeof this.global === 'undefined' ||\n        typeof this.global.location === 'undefined' ||\n        typeof this.global.location.search === 'undefined') {\n      return;\n    }\n\n    const urlParams = getQueryParams(this.global.location.search);\n    if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n      const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n      keyValues.forEach(keyValue => {\n        const [key, value] = keyValue.split(':') as [string, string];\n        this.urlFlags[key] = parseValue(key, value);\n      });\n    }\n  }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n  const params = {};\n  queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n    decodeParam(params, t[0], t[1]);\n    return t.join('=');\n  });\n  return params;\n}\n\nfunction decodeParam(\n    params: {[key: string]: string}, name: string, value?: string) {\n  params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n  value = value.toLowerCase();\n  if (value === 'true' || value === 'false') {\n    return value === 'true';\n  } else if (`${+ value}` === value) {\n    return +value;\n  }\n  throw new Error(\n      `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n  return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n  ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map<string, any>};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map<string, any>} {\n  if (globalNameSpace == null) {\n    // tslint:disable-next-line:no-any\n    let ns: any;\n    if (typeof (window) !== 'undefined') {\n      ns = window;\n    } else if (typeof (global) !== 'undefined') {\n      ns = global;\n    } else if (typeof (process) !== 'undefined') {\n      ns = process;\n    } else if (typeof (self) !== 'undefined') {\n      ns = self;\n    } else {\n      throw new Error('Could not find a global object');\n    }\n    globalNameSpace = ns;\n  }\n  return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map<string, any> {\n  const ns = getGlobalNamespace();\n  if (ns._tfGlobals == null) {\n    ns._tfGlobals = new Map();\n  }\n  return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n *             the first time it is fetched.\n */\nexport function getGlobal<T>(key: string, init: () => T): T {\n  const globalMap = getGlobalMap();\n  if (globalMap.has(key)) {\n    return globalMap.get(key);\n  } else {\n    const singleton = init();\n    globalMap.set(key, singleton);\n    return globalMap.get(key);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AllAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AnyAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMaxAttrs {\n  axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMinAttrs {\n  axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\nexport interface BatchMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BatchToSpaceNDAttrs {\n  blockShape: number[];\n  crops: number[][];\n}\n\nexport type BinaryInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BroadCastToAttrs {\n  shape: number[];\n  inputShape: number[];  // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n  dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n  clipValueMin: number;\n  clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick<NamedTensorInfoMap, 'real'|'imag'>;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n  axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'dy'>;\nexport interface Conv2DBackpropFilterAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'|'filter'>;\nexport interface Conv2DBackpropInputAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv3DAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'y'>;\n\nexport interface Conv3DBackpropFilterAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'>;\nexport interface Conv3DBackpropInputAttrs {\n  pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface CumsumAttrs {\n  axis: number;\n  exclusive: boolean;\n  reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n    Pick<NamedTensorInfoMap, 'image'|'boxes'|'boxInd'>;\nexport interface CropAndResizeAttrs {\n  cropSize: [number, number];\n  method: 'bilinear'|'nearest';\n  extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface DepthToSpaceAttrs {\n  blockSize: number;\n  dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface DepthwiseConv2dNativeAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n    'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'dy'>;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n    'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'dy'>;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Dilation2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick<NamedTensorInfoMap, 'dy'|'y'>;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n  shape: number[];\n  value: number|string;\n  dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick<NamedTensorInfoMap, 'image'>;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n    Pick<NamedTensorInfoMap, 'x'|'scale'|'offset'|'mean'|'variance'>;\nexport interface FusedBatchNormAttrs {\n  varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick<NamedTensorInfoMap, 'x'|'indices'>;\nexport interface GatherV2Attrs {\n  axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick<NamedTensorInfoMap, 'params'|'indices'>;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n  start: number;\n  stop: number;\n  num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface LogSoftmaxAttrs {\n  axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface LRNAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick<NamedTensorInfoMap, 'x'|'y'|'dy'>;\nexport interface LRNBackpropAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxAttrs {\n  reductionIndices: number|number[];\n  keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolWithArgmaxAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MeanAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MinAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV3Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV4Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV5Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick<NamedTensorInfoMap, 'indices'>;\nexport interface OneHotAttrs {\n  depth: number;\n  onValue: number;\n  offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface PadV2Attrs {\n  paddings: Array<[number, number]>;\n  constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick<NamedTensorInfoMap, 'x'|'alpha'>;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ProdAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n  start: number;\n  stop: number;\n  step: number;\n  dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReshapeAttrs {\n  shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeNearestNeighborAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n    Pick<NamedTensorInfoMap, 'images'>;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeBilinearAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick<NamedTensorInfoMap, 'images'>;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReverseAttrs {\n  dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick<NamedTensorInfoMap, 'indices'|'updates'>;\nexport interface ScatterNdAttrs {\n  shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick<NamedTensorInfoMap, 'condition'|'t'|'e'>;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SliceAttrs {\n  begin: number|number[];\n  size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SumAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SpaceToBatchNDAttrs {\n  blockShape: number[];\n  paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SplitVAttrs {\n  numOrSizeSplits: number[]|number;\n  axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface SoftmaxAttrs {\n  dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n    Pick<NamedTensorInfoMap, 'sparseIndices'|'sparseValues'|'defaultValue'>;\nexport interface SparseToDenseAttrs {\n  outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface StridedSliceAttrs {\n  begin: number[];\n  end: number[];\n  strides: number[];\n  beginMask: number;\n  endMask: number;\n  ellipsisMask: number;\n  newAxisMask: number;\n  shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TileAttrs {\n  reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TopKAttrs {\n  k: number;\n  sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TransposeAttrs {\n  perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface UniqueAttrs {\n  axis: number;\n}\n\nexport type UnaryInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick<NamedTensorInfoMap, 'value'>;\nexport interface UnpackAttrs {\n  axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n    Pick<NamedTensorInfoMap, 'x'|'segmentIds'>;\nexport interface UnsortedSegmentSumAttrs {\n  numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n  alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n  pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n      HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n  numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick<NamedTensorInfoMap, 'image'>;\nexport interface RotateWithOffsetAttrs {\n  radians: number;\n  fillValue: number|[number, number, number];\n  center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n  a: TensorInfo;\n  b: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n  activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n    getGlobal('kernelRegistry', () => new Map<string, KernelConfig>());\nconst gradRegistry =\n    getGlobal('gradRegistry', () => new Map<string, GradConfig>());\n\nexport type DataId = object;\n\ntype AttributeValue =\n    number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray<AttributeValue>;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n  inputs: NamedTensorInfoMap,\n  backend: {},\n  attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n    (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n        NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n  kernelName: string;\n  backendName: string;\n  kernelFunc: KernelFunc;\n  setupFunc?: KernelSetupFunc;\n  disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n  kernelName: string;\n  inputsToSave?: string[];\n  // When saveAllInputs is true, all inputs will be saved. Only use this flag\n  // if inputs is an array of Tensors.\n  saveAllInputs?: boolean;\n  outputsToSave?: boolean[];\n  gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n  dataId: DataId;\n  shape: number[];\n  dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n  [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n  [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n    kernelName: string, backendName: string): KernelConfig {\n  const key = makeKey(kernelName, backendName);\n  return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n  return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n  const it = kernelRegistry.entries();\n  const result: KernelConfig[] = [];\n\n  while (true) {\n    const {done, value} = it.next();\n    if (done) {\n      break;\n    }\n    const [key, config] = value;\n    const [backend, ] = key.split('_');\n    if (backend === backendName) {\n      result.push(config);\n    }\n  }\n  return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n  const {kernelName, backendName} = config;\n  const key = makeKey(kernelName, backendName);\n  if (kernelRegistry.has(key)) {\n    console.warn(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is already registered`);\n  }\n  kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n  const {kernelName} = config;\n\n  if (gradRegistry.has(kernelName)) {\n    // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n    // to debug mode.\n    if (env().getBool('DEBUG')) {\n      console.warn(`Overriding the gradient for '${kernelName}'`);\n    }\n  }\n  gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n    kernelName: string, backendName: string): void {\n  const key = makeKey(kernelName, backendName);\n  if (!kernelRegistry.has(key)) {\n    throw new Error(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is not registered`);\n  }\n  kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n  if (!gradRegistry.has(kernelName)) {\n    throw new Error(\n        `The gradient '${kernelName}' for backend is not registered`);\n  }\n  gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n    registeredBackendName: string, newBackendName: string): void {\n  const kernels = getKernelsForBackend(registeredBackendName);\n  kernels.forEach(kernelConfig => {\n    const newKernelConfig =\n        Object.assign({}, kernelConfig, {backendName: newBackendName});\n    registerKernel(newKernelConfig);\n  });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n  return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n                        Float32Array): void {\n  let counter = array.length;\n  let temp = 0;\n  let index = 0;\n  // While there are elements in the array\n  while (counter > 0) {\n    // Pick a random index\n    index = (Math.random() * counter) | 0;\n    // Decrease counter by 1\n    counter--;\n    // And swap the last element with it\n    temp = array[counter];\n    array[counter] = array[index];\n    array[index] = temp;\n  }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n  return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n  return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n  let sum = 0;\n  for (let i = 0; i < arr.length; i++) {\n    sum += arr[i];\n  }\n  return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n  const r = Math.random();\n  return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n  let result = 0;\n  for (let i = 0; i < a.length; i++) {\n    const diff = Number(a[i]) - Number(b[i]);\n    result += diff * diff;\n  }\n  return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n *     error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n  if (!expr) {\n    throw new Error(typeof msg === 'string' ? msg : msg());\n  }\n}\n\nexport function assertShapesMatch(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  assert(\n      arraysEqual(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n  assert(\n      a != null,\n      () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten<number>().\n/**\n *  Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n *  @param arr The nested array to flatten.\n *  @param result The destination array which holds the elements.\n *  @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n *      to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten<T extends number|boolean|string|Promise<number>|TypedArray>(\n    arr: T|RecursiveArray<T>, result: T[] = [], skipTypedArray = false): T[] {\n  if (result == null) {\n    result = [];\n  }\n  if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n    for (let i = 0; i < arr.length; ++i) {\n      flatten(arr[i], result, skipTypedArray);\n    }\n  } else {\n    result.push(arr as T);\n  }\n  return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n  if (shape.length === 0) {\n    // Scalar.\n    return 1;\n  }\n  let size = shape[0];\n  for (let i = 1; i < shape.length; i++) {\n    size *= shape[i];\n  }\n  return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n  return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n  if (n1 === n2) {\n    return true;\n  }\n  if (n1 == null || n2 == null) {\n    return false;\n  }\n\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function isInt(a: number): boolean {\n  return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n  // tslint:disable-next-line:no-any\n  if ((Math as any).tanh != null) {\n    // tslint:disable-next-line:no-any\n    return (Math as any).tanh(x);\n  }\n  if (x === Infinity) {\n    return 1;\n  } else if (x === -Infinity) {\n    return -1;\n  } else {\n    const e2x = Math.exp(2 * x);\n    return (e2x - 1) / (e2x + 1);\n  }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n  const width = Math.ceil(Math.sqrt(size));\n  return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n  const shuffledIndices = new Uint32Array(n);\n  for (let i = 0; i < n; ++i) {\n    shuffledIndices[i] = i;\n  }\n  shuffle(shuffledIndices);\n  return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n  if (size <= a.length) {\n    return a;\n  }\n  return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n    checkFn: () => boolean, delayFn = (counter: number) => 0,\n    maxCounter?: number): Promise<void> {\n  return new Promise<void>((resolve, reject) => {\n    let tryCount = 0;\n\n    const tryFn = () => {\n      if (checkFn()) {\n        resolve();\n        return;\n      }\n\n      tryCount++;\n\n      const nextBackoff = delayFn(tryCount);\n\n      if (maxCounter != null && tryCount >= maxCounter) {\n        reject();\n        return;\n      }\n      setTimeout(tryFn, nextBackoff);\n    };\n\n    tryFn();\n  });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n    shape: number[], size: number): number[] {\n  let shapeProd = 1;\n  let implicitIdx = -1;\n\n  for (let i = 0; i < shape.length; ++i) {\n    if (shape[i] >= 0) {\n      shapeProd *= shape[i];\n    } else if (shape[i] === -1) {\n      if (implicitIdx !== -1) {\n        throw Error(\n            `Shapes can only have 1 implicit size. ` +\n            `Found -1 at dim ${implicitIdx} and dim ${i}`);\n      }\n      implicitIdx = i;\n    } else if (shape[i] < 0) {\n      throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n    }\n  }\n\n  if (implicitIdx === -1) {\n    if (size > 0 && size !== shapeProd) {\n      throw Error(`Size(${size}) must match the product of shape ${shape}`);\n    }\n    return shape;\n  }\n\n  if (shapeProd === 0) {\n    throw Error(\n        `Cannot infer the missing size in [${shape}] when ` +\n        `there are 0 elements`);\n  }\n  if (size % shapeProd !== 0) {\n    throw Error(\n        `The implicit shape can't be a fractional number. ` +\n        `Got ${size} / ${shapeProd}`);\n  }\n\n  const newShape = shape.slice();\n  newShape[implicitIdx] = size / shapeProd;\n  return newShape;\n}\n\nexport function parseAxisParam(\n    axis: number|number[], shape: number[]): number[] {\n  const rank = shape.length;\n\n  // Normalize input\n  axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n  // Check for valid range\n  assert(\n      axis.every(ax => ax >= -rank && ax < rank),\n      () =>\n          `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n          `got axis ${axis}`);\n\n  // Check for only integers\n  assert(\n      axis.every(ax => isInt(ax)),\n      () => `All values in axis param must be integers but ` +\n          `got axis ${axis}`);\n\n  // Handle negative axis.\n  return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n    {newShape: number[], keptDims: number[]} {\n  const newShape: number[] = [];\n  const keptDims: number[] = [];\n  const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n  const axes = (axis == null || isEmptyArray) ?\n      null :\n      parseAxisParam(axis, shape).sort();\n  let j = 0;\n  for (let i = 0; i < shape.length; ++i) {\n    if (axes != null) {\n      if (axes[j] === i && shape[i] !== 1) {\n        throw new Error(\n            `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n      }\n      if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n        newShape.push(shape[i]);\n        keptDims.push(i);\n      }\n      if (axes[j] <= i) {\n        j++;\n      }\n    }\n    if (shape[i] !== 1) {\n      newShape.push(shape[i]);\n      keptDims.push(i);\n    }\n  }\n  return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType<D extends NumericDataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType<D extends DataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else if (dtype === 'string') {\n    values = new Array<'string'>(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors<D extends DataType>(\n    vals: DataTypeMap[D]|number[], dtype: D): void {\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n    }\n  }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n  return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n      dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n  if (newType === 'complex64') {\n    return false;\n  }\n  if (newType === 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'bool' && oldType === 'bool') {\n    return false;\n  }\n  return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n  return a instanceof Float32Array || a instanceof Int32Array ||\n      a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n  if (dtype === 'float32' || dtype === 'int32') {\n    return 4;\n  } else if (dtype === 'complex64') {\n    return 8;\n  } else if (dtype === 'bool') {\n    return 1;\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n  if (arr == null) {\n    return 0;\n  }\n  let bytes = 0;\n  arr.forEach(x => bytes += x.length);\n  return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n  return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n  return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n  return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n  if (Array.isArray(values)) {\n    return inferDtype(values[0]);\n  }\n  if (values instanceof Float32Array) {\n    return 'float32';\n  } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n    return 'int32';\n  } else if (isNumber(values)) {\n    return 'float32';\n  } else if (isString(values)) {\n    return 'string';\n  } else if (isBoolean(values)) {\n    return 'bool';\n  }\n  return 'float32';\n}\n\nexport function isFunction(f: Function) {\n  return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n  for (let i = start; i < size; ++i) {\n    if (size % i === 0) {\n      return i;\n    }\n  }\n  return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n  const rank = shape.length;\n  if (rank < 2) {\n    return [];\n  }\n\n  // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n  // strides.\n  const strides = new Array(rank - 1);\n  strides[rank - 2] = shape[rank - 1];\n  for (let i = rank - 3; i >= 0; --i) {\n    strides[i] = strides[i + 1] * shape[i + 1];\n  }\n  return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n    value: DataType, dtype: DataType): BackendValues {\n  if (dtype === 'string') {\n    return encodeString(value);\n  }\n\n  return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n  if (dtype === 'string') {\n    throw new Error('Cannot convert a string[] to a TypedArray');\n  }\n  if (Array.isArray(a)) {\n    a = flatten(a);\n  }\n\n  if (env().getBool('DEBUG')) {\n    checkConversionForErrors(a as number[], dtype);\n  }\n  if (noConversionNeeded(a, dtype)) {\n    return a as TypedArray;\n  }\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(a as number[]);\n  } else if (dtype === 'int32') {\n    return new Int32Array(a as number[]);\n  } else if (dtype === 'bool') {\n    const bool = new Uint8Array((a as number[]).length);\n    for (let i = 0; i < bool.length; ++i) {\n      if (Math.round((a as number[])[i]) !== 0) {\n        bool[i] = 1;\n      }\n    }\n    return bool;\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n  const ret = new Array();\n  if (shape.length === 1) {\n    const d = shape[0];\n    for (let i = 0; i < d; i++) {\n      ret[i] = a[offset + i];\n    }\n  } else {\n    const d = shape[0];\n    const rest = shape.slice(1);\n    const len = rest.reduce((acc, c) => acc * c);\n    for (let i = 0; i < d; i++) {\n      ret[i] = createNestedArray(offset + i * len, rest, a);\n    }\n  }\n  return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n  if (shape.length === 0) {\n    // Scalar type should return a single number.\n    return a[0];\n  }\n  const size = shape.reduce((acc, c) => acc * c);\n  if (size === 0) {\n    // A tensor with shape zero should be turned into empty list.\n    return [];\n  }\n  if (size !== a.length) {\n    throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n  }\n\n  return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n  return (a instanceof Float32Array && dtype === 'float32') ||\n      (a instanceof Int32Array && dtype === 'int32') ||\n      (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  const array = makeZerosTypedArray(size, dtype);\n  for (let i = 0; i < array.length; i++) {\n    array[i] = 1;\n  }\n  return array;\n}\n\nexport function makeZerosTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'int32') {\n    return new Int32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'bool') {\n    return new Uint8Array(size) as DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray<D extends DataType>(\n    shape: number[], dtype: D) {\n  const size = shape.reduce((prev, curr) => prev * curr, 1);\n  if (dtype == null || dtype === 'float32') {\n    return toNestedArray(shape, new Float32Array(size));\n  } else if (dtype === 'int32') {\n    return toNestedArray(shape, new Int32Array(size));\n  } else if (dtype === 'bool') {\n    return toNestedArray(shape, new Uint8Array(size));\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n  return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n  shape.forEach(dimSize => {\n    assert(\n        Number.isInteger(dimSize) && dimSize >= 0,\n        () =>\n            `Tensor must have a shape comprised of positive integers but got ` +\n            `shape [${shape}].`);\n  });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n    path: string, requestInits?: RequestInit): Promise<Response> {\n  return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n  encoding = encoding || 'utf-8';\n  return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n  encoding = encoding || 'utf-8';\n  return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n    locs: number[], rank: number, strides: number[]): number {\n  if (rank === 0) {\n    return 0;\n  } else if (rank === 1) {\n    return locs[0];\n  }\n  let index = locs[locs.length - 1];\n  for (let i = 0; i < locs.length - 1; ++i) {\n    index += strides[i] * locs[i];\n  }\n  return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n    index: number, rank: number, strides: number[]): number[] {\n  if (rank === 0) {\n    return [];\n  } else if (rank === 1) {\n    return [index];\n  }\n  const locs: number[] = new Array(rank);\n  for (let i = 0; i < locs.length - 1; ++i) {\n    locs[i] = Math.floor(index / strides[i]);\n    index -= locs[i] * strides[i];\n  }\n  locs[locs.length - 1] = index;\n  return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n  kernelName: string,\n  outputs: Tensor[],\n  inputs: NamedTensorMap,\n  timeMs: Promise<number|{error: string}>,\n  extraInfo: Promise<string>\n};\n\nexport class Profiler {\n  constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n    if (logger == null) {\n      this.logger = new Logger();\n    }\n  }\n\n  profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n      KernelProfile {\n    let outputs: Tensor[];\n    const holdResultWrapperFn = () => {\n      outputs = f();\n    };\n    const timer = this.backendTimer.time(holdResultWrapperFn);\n\n    for (let i = 0; i < outputs.length; i++) {\n      const output = outputs[i];\n      // Dangling promise here because we don't want to propagate up\n      // asynchronicity.\n      output.data().then(tensorVals => {\n        checkComputationForErrors(tensorVals, output.dtype, kernelName);\n      });\n    }\n\n    const kernelProfile = {\n      kernelName,\n      outputs,\n      inputs,\n      timeMs: timer.then(timing => timing.kernelMs),\n      extraInfo: timer.then(\n          timing => timing.getExtraProfileInfo != null ?\n              timing.getExtraProfileInfo() :\n              '')\n    };\n    return kernelProfile;\n  }\n\n  logKernelProfile(kernelProfile: KernelProfile): void {\n    const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n    outputs.forEach(result => {\n      Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n        this.logger.logKernelProfile(\n            kernelName, result, valueContainer[0], valueContainer[1], inputs,\n            valueContainer[2]);\n      });\n    });\n  }\n}\n\nexport function checkComputationForErrors<D extends DataType>(\n    vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n  if (dtype !== 'float32') {\n    // Only floating point computations will generate NaN values\n    return false;\n  }\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      // Throwing custom exception so behavior is testable.\n      console.warn(`Found ${num} in the result of '${kernelName}'`);\n      return true;\n    }\n  }\n  return false;\n}\n\nexport class Logger {\n  logKernelProfile(\n      name: string, result: Tensor, vals: TypedArray,\n      timeMs: number|{error: string}, inputs: NamedTensorMap,\n      extraInfo?: string) {\n    const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n                                              timeMs['error'];\n    const paddedName = util.rightPad(name, 25);\n    const rank = result.rank;\n    const size = result.size;\n    const shape = util.rightPad(result.shape.toString(), 14);\n    let inputShapesDescription = '';\n\n    for (const name in inputs) {\n      const input = inputs[name];\n      if (input != null) {\n        // The input might be a non-tensor (e.g HTMLImageElement), in which case\n        // we claim the output shape as input shape.\n        const inputShape = input.shape || result.shape;\n        const inputRank = inputShape.length;\n        inputShapesDescription +=\n            `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n      }\n    }\n\n    console.log(\n        `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n            inputShapesDescription}\\t%c${extraInfo}`,\n        'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n        'color: green', 'color: steelblue');\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n  id: number;\n  kernelName: string;\n  outputs: Tensor[];\n  inputs: NamedTensorMap;\n  // Optional params, defined only for ops with gradient impl.\n  gradient?: (dys: Tensor[]) => NamedGradientMap;\n  saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n  [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n    tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n  // Forward pass to compute all the nodes and Tensors that are transitively a\n  // function of x.\n  const tensorsFromX: {[tensorId: number]: boolean} = {};\n  const nodesFromX: {[nodeId: number]: boolean} = {};\n  for (let i = 0; i < xs.length; i++) {\n    tensorsFromX[xs[i].id] = true;\n  }\n\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n    for (const inputName in nodeInputs) {\n      const input = nodeInputs[inputName];\n\n      let anyInputFromX = false;\n      for (let j = 0; j < xs.length; j++) {\n        if (tensorsFromX[input.id]) {\n          node.outputs.forEach(output => tensorsFromX[output.id] = true);\n          anyInputFromX = true;\n          nodesFromX[node.id] = true;\n          break;\n        }\n      }\n\n      if (anyInputFromX) {\n        break;\n      }\n    }\n  }\n\n  // Backward pass to find all of the nodes and Tensors that lead to y.\n  const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n  tensorsLeadToY[y.id] = true;\n  const nodesToY: {[nodeId: number]: boolean} = {};\n\n  for (let i = tape.length - 1; i >= 0; i--) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n\n    // If any of the outputs lead to y, mark all of the inputs as leading to y.\n    for (let j = 0; j < node.outputs.length; j++) {\n      if (tensorsLeadToY[node.outputs[j].id]) {\n        for (const inputName in nodeInputs) {\n          tensorsLeadToY[nodeInputs[inputName].id] = true;\n          nodesToY[node.id] = true;\n        }\n        break;\n      }\n    }\n  }\n\n  // Return the paths that come from x and lead to y.\n  const filteredTape: TapeNode[] = [];\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n\n    if (nodesFromX[node.id] && nodesToY[node.id]) {\n      // Prune the inputs from the node that aren't a function of x.\n      const prunedInputs: {[inputName: string]: Tensor} = {};\n      for (const inputName in node.inputs) {\n        const nodeInput = node.inputs[inputName];\n        if (tensorsFromX[nodeInput.id]) {\n          prunedInputs[inputName] = nodeInput;\n        }\n      }\n\n      // Copy the node and overwrite inputsAndArgs to the pruned version.\n      const prunedNode = Object.assign({}, node);\n      prunedNode.inputs = prunedInputs;\n      prunedNode.outputs = node.outputs;\n\n      filteredTape.push(prunedNode);\n    }\n  }\n\n  return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n    tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n    filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n    add: (a: Tensor, b: Tensor) => Tensor) {\n  // Walk the tape backward and keep a map of Tensor to its gradient.\n  for (let i = filteredTape.length - 1; i >= 0; i--) {\n    const node = filteredTape[i];\n\n    const dys: Tensor[] = [];\n    node.outputs.forEach(o => {\n      const gradTensor = tensorAccumulatedGradientMap[o.id];\n      if (gradTensor != null) {\n        dys.push(gradTensor);\n      } else {\n        // This particular output is not in the back-propagation subgraph, so it\n        // does not affect the final output, thus we put null for its dy.\n        dys.push(null);\n      }\n    });\n\n    if (node.gradient == null) {\n      throw new Error(\n          `Cannot compute gradient: gradient function not found ` +\n          `for ${node.kernelName}.`);\n    }\n\n    // Backprop dy through this node and accumulate gradients over the inputs.\n    const inputGradients = node.gradient(dys);\n\n    for (const inputName in node.inputs) {\n      if (!(inputName in inputGradients)) {\n        throw new Error(\n            `Cannot backprop through input ${inputName}. ` +\n            `Available gradients found: ${Object.keys(inputGradients)}.`);\n      }\n\n      // Call the gradient function.\n      const dx = tidy(() => inputGradients[inputName]());\n      if (dx.dtype !== 'float32') {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n      }\n      const x = node.inputs[inputName];\n      if (!util.arraysEqual(dx.shape, x.shape)) {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `'${inputName}' has shape '${dx.shape}', which does not match ` +\n            `the shape of the input '${x.shape}'`);\n      }\n\n      if (tensorAccumulatedGradientMap[x.id] == null) {\n        tensorAccumulatedGradientMap[x.id] = dx;\n      } else {\n        const curGradient = tensorAccumulatedGradientMap[x.id];\n        tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n        curGradient.dispose();\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    verbose: boolean) {\n  const strides = computeStrides(shape);\n  const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n  const rank = shape.length;\n  const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n  const lines = ['Tensor'];\n  if (verbose) {\n    lines.push(`  dtype: ${dtype}`);\n    lines.push(`  rank: ${rank}`);\n    lines.push(`  shape: [${shape}]`);\n    lines.push(`  values:`);\n  }\n  lines.push(valsLines.map(l => '    ' + l).join('\\n'));\n  return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[]): number[] {\n  const n = sizeFromShape(shape);\n  const numCols = strides[strides.length - 1];\n  const padPerCol = new Array(numCols).fill(0);\n  const rank = shape.length;\n  const valuesOrTuples =\n      dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n  if (rank > 1) {\n    for (let row = 0; row < n / numCols; row++) {\n      const offset = row * numCols;\n      for (let j = 0; j < numCols; j++) {\n        padPerCol[j] = Math.max(\n            padPerCol[j],\n            valToString(valuesOrTuples[offset + j], 0, dtype).length);\n      }\n    }\n  }\n  return padPerCol;\n}\n\nfunction valToString(\n    val: number|string|[number, number], pad: number, dtype: DataType) {\n  let valStr: string;\n  if (Array.isArray(val)) {\n    valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n        `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n  } else if (isString(val)) {\n    valStr = `'${val}'`;\n  } else if (dtype === 'bool') {\n    valStr = boolNumToString(val);\n  } else {\n    valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n  }\n\n  return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n  return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[], padPerCol: number[], isLast = true): string[] {\n  const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n  const size = shape[0];\n  const rank = shape.length;\n  if (rank === 0) {\n    if (dtype === 'complex64') {\n      const complexTuple = createComplexTuples(vals);\n      return [valToString(complexTuple[0], 0, dtype)];\n    }\n    if (dtype === 'bool') {\n      return [boolNumToString(vals[0] as number)];\n    }\n    return [vals[0].toString()];\n  }\n\n  if (rank === 1) {\n    if (size > FORMAT_LIMIT_NUM_VALS) {\n      const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n      let firstVals = Array.from<number|string|[number, number]>(\n          vals.slice(0, firstValsSize));\n      let lastVals = Array.from<number|string|[number, number]>(vals.slice(\n          (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n          size * storagePerElement));\n      if (dtype === 'complex64') {\n        firstVals = createComplexTuples(firstVals);\n        lastVals = createComplexTuples(lastVals);\n      }\n      return [\n        '[' +\n        firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n            .join(', ') +\n        ', ..., ' +\n        lastVals\n            .map(\n                (x, i) => valToString(\n                    x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n            .join(', ') +\n        ']'\n      ];\n    }\n    const displayVals: Array<number|string|[number, number]> =\n        dtype === 'complex64' ? createComplexTuples(vals) :\n                                Array.from<number|string>(vals);\n\n    return [\n      '[' +\n      displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n          .join(', ') +\n      ']'\n    ];\n  }\n\n  // The array is rank 2 or more.\n  const subshape = shape.slice(1);\n  const substrides = strides.slice(1);\n  const stride = strides[0] * storagePerElement;\n  const lines: string[] = [];\n  if (size > FORMAT_LIMIT_NUM_VALS) {\n    for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          false /* isLast */));\n    }\n    lines.push('...');\n    for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  } else {\n    for (let i = 0; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  }\n  const sep = rank === 2 ? ',' : '';\n  lines[0] = '[' + lines[0] + sep;\n  for (let i = 1; i < lines.length - 1; i++) {\n    lines[i] = ' ' + lines[i] + sep;\n  }\n  let newLineSep = ',\\n';\n  for (let i = 2; i < rank; i++) {\n    newLineSep += '\\n';\n  }\n  lines[lines.length - 1] =\n      ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n  return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n                             TypedArray): Array<[number, number]> {\n  const complexTuples: Array<[number, number]> = [];\n  for (let i = 0; i < vals.length; i += 2) {\n    complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n  }\n  return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData<D extends DataType> {\n  dataId?: DataId;\n  values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer<R extends Rank, D extends DataType = 'float32'> {\n  size: number;\n  shape: ShapeMap[R];\n  strides: number[];\n  values: DataTypeMap[D];\n\n  constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.size = util.sizeFromShape(shape);\n\n    if (values != null) {\n      const n = values.length;\n      util.assert(\n          n === this.size,\n          () => `Length of values '${n}' does not match the size ` +\n              `inferred by the shape '${this.size}'.`);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `complex64 dtype TensorBuffers are not supported. Please create ` +\n          `a TensorBuffer for the real and imaginary parts separately and ` +\n          `call tf.complex(real, imag).`);\n    }\n    this.values = values || util.getArrayFromDType(dtype, this.size);\n    this.strides = computeStrides(shape);\n  }\n\n  /**\n   * Sets a value in the buffer at a given location.\n   *\n   * @param value The value to set.\n   * @param locs  The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  set(value: SingleValueMap[D], ...locs: number[]): void {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    util.assert(\n        locs.length === this.rank,\n        () => `The number of provided coordinates (${locs.length}) must ` +\n            `match the rank (${this.rank})`);\n\n    const index = this.locToIndex(locs);\n    this.values[index] = value as number;\n  }\n\n  /**\n   * Returns the value in the buffer at the provided location.\n   *\n   * @param locs The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  get(...locs: number[]): SingleValueMap[D] {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    let i = 0;\n    for (const loc of locs) {\n      if (loc < 0 || loc >= this.shape[i]) {\n        const msg = `Requested out of range element at ${locs}. ` +\n            `  Buffer shape=${this.shape}`;\n        throw new Error(msg);\n      }\n      i++;\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return this.values[index] as SingleValueMap[D];\n  }\n\n  locToIndex(locs: number[]): number {\n    if (this.rank === 0) {\n      return 0;\n    } else if (this.rank === 1) {\n      return locs[0];\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return index;\n  }\n\n  indexToLoc(index: number): number[] {\n    if (this.rank === 0) {\n      return [];\n    } else if (this.rank === 1) {\n      return [index];\n    }\n    const locs: number[] = new Array(this.shape.length);\n    for (let i = 0; i < locs.length - 1; ++i) {\n      locs[i] = Math.floor(index / this.strides[i]);\n      index -= locs[i] * this.strides[i];\n    }\n    locs[locs.length - 1] = index;\n    return locs;\n  }\n\n  get rank() {\n    return this.shape.length;\n  }\n\n  /**\n   * Creates an immutable `tf.Tensor` object from the buffer.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  toTensor(): Tensor<R> {\n    return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n        Tensor<R>;\n  }\n}\n\nexport interface TensorTracker {\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: Backend): Tensor;\n  makeVariable(\n      initialValue: Tensor, trainable?: boolean, name?: string,\n      dtype?: DataType): Variable;\n  incRef(a: Tensor, backend: Backend): void;\n  disposeTensor(t: Tensor): void;\n  disposeVariable(v: Variable): void;\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n  cast<T extends Tensor>(x: T, dtype: DataType): T;\n  buffer<R extends Rank, D extends DataType>(\n      shape: ShapeMap[R], dtype: D,\n      values?: DataTypeMap[D]): TensorBuffer<R, D>;\n  print<T extends Tensor>(x: T, verbose: boolean): void;\n  clone<T extends Tensor>(x: T): T;\n  // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n  trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n  opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n  deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object;  // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor<R extends Rank = Rank> {\n  /** Unique id of this tensor. */\n  readonly id: number;\n  /**\n   * Id of the bucket holding the data for this tensor. Multiple arrays can\n   * point to the same bucket (e.g. when calling array.reshape()).\n   */\n  dataId: DataId;\n  /** The shape of the tensor. */\n  readonly shape: ShapeMap[R];\n  /** Number of elements in the tensor. */\n  readonly size: number;\n  /** The data type for the array. */\n  readonly dtype: DataType;\n  /** The rank type for the array (see `Rank` enum). */\n  readonly rankType: R;\n\n  /** Whether this tensor has been globally kept. */\n  kept = false;\n  /** The id of the scope this tensor is being tracked in. */\n  scopeId: number;\n\n  /**\n   * Number of elements to skip in each dimension when indexing. See\n   * https://docs.scipy.org/doc/numpy/reference/generated/\\\n   * numpy.ndarray.strides.html\n   */\n  readonly strides: number[];\n\n  constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.dtype = dtype || 'float32';\n    this.size = util.sizeFromShape(shape);\n    this.strides = computeStrides(shape);\n    this.dataId = dataId;\n    this.id = id;\n    this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n  }\n\n  get rank(): number {\n    return this.shape.length;\n  }\n\n  /**\n   * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async buffer<D extends DataType = 'float32'>(): Promise<TensorBuffer<R, D>> {\n    const vals = await this.data<D>();\n    return opHandler.buffer(this.shape, this.dtype as D, vals);\n  }\n\n  /**\n   * Returns a `tf.TensorBuffer` that holds the underlying data.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  bufferSync<D extends DataType = 'float32'>(): TensorBuffer<R, D> {\n    return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * asynchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async array(): Promise<ArrayMap[R]> {\n    const vals = await this.data();\n    return toNestedArray(this.shape, vals) as ArrayMap[R];\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * synchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  arraySync(): ArrayMap[R] {\n    return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n  }\n\n  /**\n   * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n   * promise of `TypedArray` that resolves when the computation has finished.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]> {\n    this.throwIfDisposed();\n    const data = trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      const bytes = await data as Uint8Array[];\n      try {\n        return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as Promise<DataTypeMap[D]>;\n  }\n\n  /**\n   * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n   * UI thread until the values are ready, which can cause performance issues.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D] {\n    this.throwIfDisposed();\n    const data = trackerFn().readSync(this.dataId);\n    if (this.dtype === 'string') {\n      try {\n        return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n            DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as DataTypeMap[D];\n  }\n\n  /** Returns the underlying bytes of the tensor's data. */\n  async bytes(): Promise<Uint8Array[]|Uint8Array> {\n    this.throwIfDisposed();\n    const data = await trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      return data as Uint8Array[];\n    } else {\n      return new Uint8Array((data as TypedArray).buffer);\n    }\n  }\n\n  /**\n   * Disposes `tf.Tensor` from memory.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dispose(): void {\n    if (this.isDisposed) {\n      return;\n    }\n    trackerFn().disposeTensor(this);\n    this.isDisposedInternal = true;\n  }\n\n  protected isDisposedInternal = false;\n  get isDisposed(): boolean {\n    return this.isDisposedInternal;\n  }\n\n  throwIfDisposed() {\n    if (this.isDisposed) {\n      throw new Error(`Tensor is disposed.`);\n    }\n  }\n\n  /**\n   * Prints the `tf.Tensor`. See `tf.print` for details.\n   *\n   * @param verbose Whether to print verbose information about the tensor,\n   *    including dtype and size.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  print(verbose = false): void {\n    return opHandler.print(this, verbose);\n  }\n\n  /**\n   * Returns a copy of the tensor. See `tf.clone` for details.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  clone<T extends Tensor>(this: T): T {\n    this.throwIfDisposed();\n    return opHandler.clone(this);\n  }\n\n  /**\n   * Returns a human-readable description of the tensor. Useful for logging.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  toString(verbose = false): string {\n    const vals = this.dataSync();\n    return tensorToString(vals, this.shape, this.dtype, verbose);\n  }\n\n  cast<T extends this>(dtype: DataType): T {\n    this.throwIfDisposed();\n    return opHandler.cast(this as T, dtype);\n  }\n  variable(trainable = true, name?: string, dtype?: DataType): Variable<R> {\n    this.throwIfDisposed();\n    return trackerFn().makeVariable(this, trainable, name, dtype) as\n        Variable<R>;\n  }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n  value: (instance: Tensor) => {\n    // Implementation note: we should use properties of the object that will be\n    // defined before the constructor body has finished executing (methods).\n    // This is because when this code is transpiled by babel, babel will call\n    // classCallCheck before the constructor body is run.\n    // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n    return !!instance && instance.data != null && instance.dataSync != null &&\n        instance.throwIfDisposed != null;\n  }\n});\n\nexport interface NumericTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: NumericDataType;\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D];\n  data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]>;\n}\n\nexport interface StringTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: 'string';\n  dataSync<D extends DataType = 'string'>(): DataTypeMap[D];\n  data<D extends DataType = 'string'>(): Promise<DataTypeMap[D]>;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor<Rank.R0>;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor<Rank.R1>;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor<Rank.R2>;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor<Rank.R3>;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor<Rank.R4>;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor<Rank.R5>;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor<Rank.R6>;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable<R extends Rank = Rank> extends Tensor<R> {\n  name: string;\n\n  constructor(\n      initialValue: Tensor<R>, public trainable: boolean, name: string,\n      tensorId: number) {\n    super(\n        initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n    this.name = name;\n  }\n\n  /**\n   * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n   * the same shape and dtype as the old `tf.Tensor`.\n   *\n   * @param newValue New tensor to be assigned to this variable.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  assign(newValue: Tensor<R>): void {\n    if (newValue.dtype !== this.dtype) {\n      throw new Error(\n          `dtype of the new value (${newValue.dtype}) and ` +\n          `previous value (${this.dtype}) must match`);\n    }\n    if (!util.arraysEqual(newValue.shape, this.shape)) {\n      throw new Error(\n          `shape of the new value (${newValue.shape}) and ` +\n          `previous value (${this.shape}) must match`);\n    }\n    trackerFn().disposeTensor(this);\n    this.dataId = newValue.dataId;\n    trackerFn().incRef(this, null /* backend */);\n  }\n\n  dispose(): void {\n    trackerFn().disposeVariable(this);\n    this.isDisposedInternal = true;\n  }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n  value: (instance: Variable) => {\n    return instance instanceof Tensor && instance.assign != null &&\n        instance.assign instanceof Function;\n  }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n  R0: number[];\n  R1: [number];\n  R2: [number, number];\n  R3: [number, number, number];\n  R4: [number, number, number, number];\n  R5: [number, number, number, number, number];\n  R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n  R0: number;\n  R1: number[];\n  R2: number[][];\n  R3: number[][][];\n  R4: number[][][][];\n  R5: number[][][][][];\n  R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n  float32: Float32Array;\n  int32: Int32Array;\n  bool: Uint8Array;\n  complex64: Float32Array;\n  string: string[];\n}\n\nexport interface SingleValueMap {\n  bool: boolean;\n  int32: number;\n  float32: number;\n  complex64: number;\n  string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n  R0 = 'R0',\n  R1 = 'R1',\n  R2 = 'R2',\n  R3 = 'R3',\n  R4 = 'R4',\n  R5 = 'R5',\n  R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray<T> =\n    T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray<T extends any> {\n  [index: number]: T|RecursiveArray<T>;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'int32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'bool',\n  'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n  'float32' = 'float32',\n  'int32' = 'float32',\n  'bool' = 'float32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n  'float32' = 'complex64',\n  'int32' = 'complex64',\n  'bool' = 'complex64',\n  'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n  'float32': UpcastFloat32AndMap,\n  'int32': UpcastInt32AndMap,\n  'bool': UpcastBoolAndMap,\n  'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n  if (typeA === 'string' || typeB === 'string') {\n    if (typeA === 'string' && typeB === 'string') {\n      return 'string';\n    }\n    throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n  }\n  return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n  return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n    TypedArray|number|boolean|string|RecursiveArray<number|number[]|TypedArray>|\n    RecursiveArray<boolean>|RecursiveArray<string>|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n    string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n    boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n    boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n    TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n    string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n    TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n    string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n  width: number;\n  height: number;\n  data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch<T extends Tensor>(a: T, b: T): [T, T] {\n  if (a.dtype === b.dtype) {\n    return [a, b];\n  }\n  const dtype = upcastType(a.dtype, b.dtype);\n  return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n  assert(\n      a.dtype === b.dtype,\n      () => `The dtypes of the first(${a.dtype}) and` +\n          ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n  return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n *   `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n *   is safe to pass any object here, except that `Promise`s are not\n *   supported.\n * @returns An array of `Tensors` found within the passed object. If the\n *   argument is simply a `Tensor', a list containing that `Tensor` is\n *   returned. If the object is not a `Tensor` or does not\n *   contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n  const list: Tensor[] = [];\n  const seen = new Set<{}|void>();\n  walkTensorContainer(result, list, seen);\n  return list;\n}\n\nfunction walkTensorContainer(\n    container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n  if (container == null) {\n    return;\n  }\n  if (container instanceof Tensor) {\n    list.push(container);\n    return;\n  }\n  if (!isIterable(container)) {\n    return;\n  }\n  // Iteration over keys works also for arrays.\n  const iterable = container as TensorContainerArray;\n  for (const k in iterable) {\n    const val = iterable[k];\n    if (!seen.has(val)) {\n      seen.add(val);\n      walkTensorContainer(val, list, seen);\n    }\n  }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n  return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc<T> = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n *   value: Tensor,\n *   gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc<T extends Tensor> =\n    (...inputs: Array<Tensor|GradSaveFunc>) => {\n      value: T;\n      gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n    };\n\nexport type MemoryInfo = {\n  numTensors: number; numDataBuffers: number; numBytes: number;\n  unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n  name: string; bytesAdded: number; totalBytesSnapshot: number;\n  tensorsAdded: number;\n  totalTensorsSnapshot: number;\n  inputShapes: number[][];\n  outputShapes: number[][];\n  kernelTimeMs: number | {error: string} | Promise<number|{error: string}>;\n  extraInfo: string | Promise<string>;\n};\n\nexport type ProfileInfo = {\n  newBytes: number; newTensors: number; peakBytes: number;\n  kernels: KernelInfo[];\n  result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n  wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn<T extends TensorContainer> = () => T;\n\ninterface ScopeState {\n  track: Tensor[];\n  name: string;\n  id: number;\n}\n\nclass EngineState {\n  // Public since optimizers will use it.\n  registeredVariables: NamedVariableMap = {};\n\n  nextTapeNodeId = 0;\n  numBytes = 0;\n  numTensors = 0;\n  numStringTensors = 0;\n  numDataBuffers = 0;\n\n  activeTape: TapeNode[];\n  // Number of nested tf.grad() statements when computing higher-order\n  // gradients. E.g. `1` for first-order gradients and `2` for second-order\n  // gradients. Used to track if the tape should be removed after a backprop.\n  gradientDepth = 0;\n  // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n  // off the tape.\n  kernelDepth = 0;\n\n  // Keep Tensors that parallel the tapes.\n  activeScope: ScopeState;\n  scopeStack: ScopeState[] = [];\n  /**\n   * Keeps track of the number of data moves during a kernel execution. We\n   * maintain a stack since kernels can call other kernels, recursively.\n   */\n  numDataMovesStack: number[] = [];\n  nextScopeId = 0;\n\n  tensorInfo = new WeakMap<DataId, {\n    backend: KernelBackend,\n    bytes: number,\n    dtype: DataType,\n    shape: number[],\n    refCount: number\n  }>();\n\n  profiling = false;\n  activeProfile: ProfileInfo =\n      {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n  dispose() {\n    for (const variableName in this.registeredVariables) {\n      this.registeredVariables[variableName].dispose();\n    }\n  }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n  state: EngineState;\n  backendName: string;\n  registry: {[id: string]: KernelBackend} = {};\n  registryFactory: {\n    [id: string]: {\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority: number\n    }\n  } = {};\n\n  private profiler: Profiler;\n  private backendInstance: KernelBackend;\n  private pendingBackendInit: Promise<boolean>;\n  private pendingBackendInitId = 0;\n\n  constructor(public ENV: Environment) {\n    this.state = new EngineState();\n  }\n\n  async ready(): Promise<void> {\n    if (this.pendingBackendInit != null) {\n      return this.pendingBackendInit.then(() => {});\n    }\n    if (this.backendInstance != null) {\n      return;\n    }\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const success = await this.initializeBackend(backendName).success;\n      if (success) {\n        await this.setBackend(backendName);\n        return;\n      }\n    }\n\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  get backend(): KernelBackend {\n    if (this.pendingBackendInit != null) {\n      throw new Error(\n          `Backend '${this.backendName}' has not yet been initialized. Make ` +\n          `sure to await tf.ready() or await tf.setBackend() before calling ` +\n          `other methods`);\n    }\n    if (this.backendInstance == null) {\n      const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n      if (asyncInit) {\n        throw new Error(\n            `The highest priority backend '${name}' has not yet been ` +\n            `initialized. Make sure to await tf.ready() or ` +\n            `await tf.setBackend() before calling other methods`);\n      }\n      this.setBackend(name);\n    }\n    return this.backendInstance;\n  }\n\n  backendNames(): string[] {\n    return Object.keys(this.registryFactory);\n  }\n\n  findBackend(backendName: string): KernelBackend {\n    if (!(backendName in this.registry)) {\n      // If the backend hasn't been initialized but we have a registry entry for\n      // it, initialize it and return it.\n      if (backendName in this.registryFactory) {\n        const {asyncInit} = this.initializeBackend(backendName);\n        if (asyncInit) {\n          // Backend is not ready yet.\n          return null;\n        }\n      } else {\n        return null;\n      }\n    }\n    return this.registry[backendName];\n  }\n\n  findBackendFactory(backendName: string):\n      () => KernelBackend | Promise<KernelBackend> {\n    if (!(backendName in this.registryFactory)) {\n      return null;\n    }\n    return this.registryFactory[backendName].factory;\n  }\n\n  registerBackend(\n      backendName: string,\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority = 1): boolean {\n    if (backendName in this.registryFactory) {\n      console.warn(\n          `${backendName} backend was already registered. ` +\n          `Reusing existing backend factory.`);\n      return false;\n    }\n    this.registryFactory[backendName] = {factory, priority};\n    return true;\n  }\n\n  async setBackend(backendName: string): Promise<boolean> {\n    if (this.registryFactory[backendName] == null) {\n      throw new Error(`Backend name '${backendName}' not found in registry`);\n    }\n    this.backendName = backendName;\n    if (this.registry[backendName] == null) {\n      this.backendInstance = null;\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      const result = asyncInit ? await success : success;\n      if (!result) {\n        return false;\n      }\n    }\n    this.backendInstance = this.registry[backendName];\n    this.setupRegisteredKernels();\n    // Reset the profiler.\n    this.profiler = new Profiler(this.backendInstance);\n\n    return true;\n  }\n\n  private setupRegisteredKernels(): void {\n    const kernels = getKernelsForBackend(this.backendName);\n    kernels.forEach(kernel => {\n      if (kernel.setupFunc != null) {\n        kernel.setupFunc(this.backendInstance);\n      }\n    });\n  }\n\n  private disposeRegisteredKernels(backendName: string): void {\n    const kernels = getKernelsForBackend(backendName);\n    kernels.forEach(kernel => {\n      if (kernel.disposeFunc != null) {\n        kernel.disposeFunc(this.registry[backendName]);\n      }\n    });\n  }\n\n  /**\n   * Initializes a backend by looking up the backend name in the factory\n   * registry and calling the factory method. Returns a boolean representing\n   * whether the initialization of the backend suceeded. Throws an error if\n   * there is no backend in the factory registry.\n   */\n  private initializeBackend(backendName: string):\n      {success: boolean|Promise<boolean>, asyncInit: boolean} {\n    const registryFactoryEntry = this.registryFactory[backendName];\n    if (registryFactoryEntry == null) {\n      throw new Error(\n          `Cannot initialize backend ${backendName}, no registration found.`);\n    }\n\n    try {\n      const backend = registryFactoryEntry.factory();\n      /* Test if the factory returns a promise.\n      Done in a more liberal way than\n      previous 'Promise.resolve(backend)===backend'\n      as we needed to account for custom Promise\n      implementations (e.g. Angular) */\n      if (backend && !(backend instanceof KernelBackend)\n          && typeof backend.then === 'function') {\n        const promiseId = ++this.pendingBackendInitId;\n        const success =\n            backend\n                .then(backendInstance => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.registry[backendName] = backendInstance;\n                  this.pendingBackendInit = null;\n                  return true;\n                })\n                .catch(err => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.pendingBackendInit = null;\n                  console.warn(\n                      `Initialization of backend ${backendName} failed`);\n                  console.warn(err.stack || err.message);\n                  return false;\n                });\n        this.pendingBackendInit = success;\n        return {success, asyncInit: true};\n      } else {\n        this.registry[backendName] = backend as KernelBackend;\n        return {success: true, asyncInit: false};\n      }\n    } catch (err) {\n      console.warn(`Initialization of backend ${backendName} failed`);\n      console.warn(err.stack || err.message);\n      return {success: false, asyncInit: false};\n    }\n  }\n\n  removeBackend(backendName: string): void {\n    if (!(backendName in this.registryFactory)) {\n      throw new Error(`${backendName} backend not found in registry`);\n    }\n    if (this.backendName === backendName && this.pendingBackendInit != null) {\n      // There is a pending promise of the backend we want to remove. Make it\n      // obsolete.\n      this.pendingBackendInitId++;\n    }\n\n    if (backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n\n    delete this.registryFactory[backendName];\n\n    // Unset the backend if it is active.\n    if (this.backendName === backendName) {\n      this.pendingBackendInit = null;\n      this.backendName = null;\n      this.backendInstance = null;\n    }\n  }\n\n  private getSortedBackends(): string[] {\n    if (Object.keys(this.registryFactory).length === 0) {\n      throw new Error('No backend found in registry.');\n    }\n    return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n      // Highest priority comes first.\n      return this.registryFactory[b].priority -\n          this.registryFactory[a].priority;\n    });\n  }\n\n  private initializeBackendsAndReturnBest():\n      {name: string, asyncInit: boolean} {\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      if (asyncInit || success) {\n        return {name: backendName, asyncInit};\n      }\n    }\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  moveData(backend: KernelBackend, dataId: DataId) {\n    const info = this.state.tensorInfo.get(dataId);\n    const srcBackend = info.backend;\n    const values = this.readSync(dataId);\n    // Delete the tensor from the old backend and move it to the new\n    // backend.\n    srcBackend.disposeData(dataId);\n    info.backend = backend;\n    backend.move(dataId, values, info.shape, info.dtype);\n    if (this.shouldCheckForMemLeaks()) {\n      // Track the number of moves during a kernel execution to correctly\n      // detect memory leaks.\n      this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n    }\n  }\n\n  tidy<T extends TensorContainer>(nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>):\n      T {\n    let name: string = null;\n    if (fn == null) {\n      // Called with only 1 argument.\n      if (typeof nameOrFn !== 'function') {\n        throw new Error('Please provide a function to tidy()');\n      }\n      fn = nameOrFn;\n    } else {\n      // Called with 2 arguments.\n      if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n        throw new Error(\n            'When calling with two arguments, the first argument ' +\n            'to tidy() must be a string');\n      }\n      if (typeof fn !== 'function') {\n        throw new Error(\n            'When calling with two arguments, the 2nd argument ' +\n            'to tidy() must be a function');\n      }\n      name = nameOrFn as string;\n      // TODO(nsthorat,smilkov): Do operation logging and performance\n      // profiling.\n    }\n    let result: T;\n    return this.scopedRun(\n        () => this.startScope(name), () => this.endScope(result), () => {\n          result = fn();\n          if (result instanceof Promise) {\n            console.error('Cannot return a Promise inside of tidy.');\n          }\n          return result;\n        });\n  }\n\n  private scopedRun<T>(start: () => void, end: () => void, f: () => T): T {\n    start();\n    try {\n      const res = f();\n      end();\n      return res;\n    } catch (ex) {\n      end();\n      throw ex;\n    }\n  }\n\n  private static nextTensorId = 0;\n  private nextTensorId(): number {\n    return Engine.nextTensorId++;\n  }\n\n  private static nextVariableId = 0;\n  private nextVariableId(): number {\n    return Engine.nextVariableId++;\n  }\n\n  /**\n   * This method is called instead of the public-facing tensor.clone() when\n   * saving a tensor for backwards pass. It makes sure to add the clone\n   * operation to the tape regardless of being called inside a kernel\n   * execution.\n   *\n   * This method will go away once all kernels are modularized since we won't\n   * need to turn off the tape inside runKernel().\n   */\n  private clone(x: Tensor): Tensor {\n    const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n    const inputs = {x};\n    const grad = (dy: Tensor) => ({\n      x: () => {\n        const dtype = 'float32';\n        const gradInputs = {x: dy};\n        const attrs = {dtype};\n\n        return ENGINE.runKernelFunc(\n            backend => backend.cast(dy, dtype),\n            gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n            attrs as {} as NamedAttrMap);\n      }\n    });\n    const saved: Tensor[] = [];\n    this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n    return y;\n  }\n\n  /**\n   * Execute a kernel with the given name and return the output tensor.\n   *\n   * @param kernelName The name of the kernel to execute.\n   * @param inputs A map of input names to tensors.\n   * @param attrs A map of attribute names to their values. An attribute is a\n   *     primitive (non-tensor) input to the kernel.\n   * @param inputsToSave A list of tensors, inputs to save for the backprop\n   *     computation.\n   * @param outputsToSave A list of booleans, specifying which output to save\n   *     for the backprop computation. These are booleans since the output\n   * tensors are not visible to the user.\n   */\n  runKernel(\n      kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n      inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n    const forwardFunc: null = null;\n    const backwardsFunc: null = null;\n    // Call runKernel as a stop-gap until we modularize all kernels.\n    // Once we modularize all kernels, we will remove the existing\n    // `runKernelFunc`.\n    return this.runKernelFunc(\n        forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n        outputsToSave);\n  }\n\n  private shouldCheckForMemLeaks(): boolean {\n    return this.ENV.getBool('IS_TEST');\n  }\n\n  private checkKernelForMemLeak(\n      kernelName: string, numDataIdsBefore: number,\n      outInfos: TensorInfo[]): void {\n    const numDataIdsAfter = this.backend.numDataIds();\n\n    // Count the number of data ids associated with the result of the kernel.\n    let numOutputDataIds = 0;\n    outInfos.forEach(info => {\n      // Complex numbers allocate 3 data ids, one for 'real', one for\n      // 'imaginary', and one for the container that holds the former two.\n      numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n    });\n\n    // Account for the number of moves during kernel execution. A \"data move\"\n    // can happen in the middle of a kernel execution, placing a new (key,value)\n    // pair in the data storage. Since data moves have net zero effect (we\n    // always remove the data from the old backend), we have to cancel them out\n    // when detecting memory leaks.\n    const numMoves =\n        this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n    const dataIdsLeaked =\n        numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n    if (dataIdsLeaked > 0) {\n      throw new Error(\n          `Backend '${this.backendName}' has an internal memory leak ` +\n          `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n    }\n  }\n\n  /**\n   * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n   *     only for kernels that are not yet fully modularized.\n   */\n  runKernelFunc<T extends Tensor|Tensor[], I extends NamedTensorMap>(\n      forwardFunc: ForwardFunc<T>, inputs: I,\n      backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n      kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n      outputsToSave?: boolean[]): T {\n    let outputs: Tensor[];\n    let saved: Tensor[] = [];\n    const isTapeOn = this.isTapeOn();\n    if (kernelName == null) {\n      kernelName =\n          this.state.activeScope != null ? this.state.activeScope.name : '';\n    }\n\n    const startingBytecount = this.state.numBytes;\n    const startingNumTensors = this.state.numTensors;\n\n    if (this.shouldCheckForMemLeaks()) {\n      this.state.numDataMovesStack.push(0);\n    }\n\n    let kernelFunc: () => Tensor[];\n    const kernel = getKernel(kernelName, this.backendName);\n    let out: TensorInfo|TensorInfo[];\n    if (kernel != null) {\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n        const outInfos = Array.isArray(out) ? out : [out];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n        }\n        const outTensors = outInfos.map(\n            ({dataId, shape, dtype}) =>\n                this.makeTensorFromDataId(dataId, shape, dtype));\n\n        // Save the inputs and outputs.\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (isTapeOn) {\n          let tensorsToSave =\n              this.getTensorsForGradient(kernelName, inputs, outTensors);\n          if (tensorsToSave == null) {\n            // Fallback for ops that call runKernelFunc and pass in\n            // inputsToSave and outputsToSave. Currently this is the set of ops\n            // with kernel support in the WASM backend. Once those ops and\n            // respective gradients are modularised we can remove this path.\n            if (outputsToSave == null) {\n              outputsToSave = [];\n            }\n            const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n            tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n          }\n          saved = this.saveTensorsForBackwardMode(tensorsToSave);\n        }\n        return outTensors;\n      };\n    } else {\n      const saveFunc: GradSaveFunc = (tensors) => {\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (!isTapeOn) {\n          return;\n        }\n        saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n      };\n\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n        const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n        }\n        return outs;\n      };\n    }\n\n    // Stop recording to a tape when running a kernel.\n    let kernelProfile: KernelProfile;\n    this.scopedRun(\n        () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n          if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n            outputs = kernelFunc();\n          } else {\n            kernelProfile = this.profiler.profileKernel(\n                kernelName, inputs, () => kernelFunc());\n            if (this.ENV.getBool('DEBUG')) {\n              this.profiler.logKernelProfile(kernelProfile);\n            }\n            outputs = kernelProfile.outputs;\n          }\n        });\n\n    if (isTapeOn) {\n      this.addTapeNode(\n          kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n    }\n\n    if (this.state.profiling) {\n      this.state.activeProfile.kernels.push({\n        name: kernelName,\n        bytesAdded: this.state.numBytes - startingBytecount,\n        totalBytesSnapshot: this.state.numBytes,\n        tensorsAdded: this.state.numTensors - startingNumTensors,\n        totalTensorsSnapshot: this.state.numTensors,\n        inputShapes: Object.keys(inputs).map(\n            key => inputs[key] != null ? inputs[key].shape : null),\n        outputShapes: outputs.map(item => item.shape),\n        kernelTimeMs: kernelProfile.timeMs,\n        extraInfo: kernelProfile.extraInfo\n      });\n    }\n    return (Array.isArray(out) ? outputs : outputs[0]) as T;\n  }\n\n  /**\n   * Saves tensors used in forward mode for use in backward mode.\n   *\n   * @param tensors the list of tensors to save.\n   */\n  private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n    const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n    return saved;\n  }\n\n  /**\n   * Returns a list of tensors to save for a given gradient calculation.\n   *\n   * Returns undefined if their is no registered gradient for this kernel in the\n   * gradient registry.\n   *\n   * @param kernelName name of kernel to look up gradient for.\n   * @param inputs a map of input tensors.\n   * @param outputs an array of output tensors from forward mode of kernel.\n   */\n  private getTensorsForGradient(\n      kernelName: string, inputs: NamedTensorMap,\n      outputs: Tensor[]): Tensor[]|null {\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      const inputsToSave: string[] = gradConfig.inputsToSave || [];\n      const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n      // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n      // specified in inputsToSave will be saved.\n      let inputTensorsToSave: Tensor[];\n      if (gradConfig.saveAllInputs) {\n        util.assert(\n            Array.isArray(inputs),\n            () => 'saveAllInputs is true, expected inputs to be an array.');\n\n        inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n      } else {\n        inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n      }\n\n      const outputTensorsToSave: Tensor[] =\n          outputs.filter((_, i) => outputsToSave[i]);\n\n      return inputTensorsToSave.concat(outputTensorsToSave);\n    }\n    // TODO(yassogba) throw exception here once all runkernelFunc calls with\n    // inputsToSave/outputsToSave are removed\n    return null;\n  }\n\n  /**\n   * Internal method used by public APIs for tensor creation. Makes a new\n   * tensor with the provided shape, dtype and values. It always\n   * creates a new data id and writes the values to the underlying backend.\n   */\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    if (values == null) {\n      throw new Error('Values passed to engine.makeTensor() are null');\n    }\n    dtype = dtype || 'float32';\n    backend = backend || this.backend;\n    let backendVals = values as BackendValues;\n    if (dtype === 'string' && util.isString(values[0])) {\n      backendVals = (values as string[]).map(d => util.encodeString(d));\n    }\n    const dataId = backend.write(backendVals, shape, dtype);\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n\n    // Count bytes for string tensors.\n    if (dtype === 'string') {\n      const info = this.state.tensorInfo.get(dataId);\n      const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n      this.state.numBytes += newBytes - info.bytes;\n      info.bytes = newBytes;\n    }\n    return t;\n  }\n\n  /**\n   * Internal method used by backends. Makes a new tensor\n   * that is a wrapper around an existing data id. It doesn't create\n   * a new data id, only increments the ref count used in memory tracking.\n   */\n  makeTensorFromDataId(\n      dataId: DataId, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    dtype = dtype || 'float32';\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n    return t;\n  }\n\n  makeVariable(\n      initialValue: Tensor, trainable = true, name?: string,\n      dtype?: DataType): Variable {\n    name = name || this.nextVariableId().toString();\n    if (dtype != null && dtype !== initialValue.dtype) {\n      initialValue = initialValue.cast(dtype);\n    }\n    const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n    if (this.state.registeredVariables[v.name] != null) {\n      throw new Error(`Variable with name ${v.name} was already registered`);\n    }\n    this.state.registeredVariables[v.name] = v;\n    this.incRef(v, this.backend);\n    return v;\n  }\n\n  incRef(a: Tensor, backend: KernelBackend): void {\n    const refCount = this.state.tensorInfo.has(a.dataId) ?\n        this.state.tensorInfo.get(a.dataId).refCount :\n        0;\n    this.state.numTensors++;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors++;\n    }\n    if (refCount === 0) {\n      this.state.numDataBuffers++;\n\n      // Bytes for complex numbers are counted by their components. Bytes for\n      // string tensors are counted when writing values.\n      let bytes = 0;\n      if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n        bytes = a.size * util.bytesPerElement(a.dtype);\n      }\n      this.state.tensorInfo.set(a.dataId, {\n        backend: backend || this.backend,\n        dtype: a.dtype,\n        shape: a.shape,\n        bytes,\n        refCount: 0\n      });\n      this.state.numBytes += bytes;\n    }\n\n    this.state.tensorInfo.get(a.dataId).refCount++;\n\n    if (!(a instanceof Variable)) {\n      this.track(a);\n    }\n  }\n\n  disposeTensor(a: Tensor): void {\n    if (!this.state.tensorInfo.has(a.dataId)) {\n      return;\n    }\n\n    this.state.numTensors--;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors--;\n    }\n    const info = this.state.tensorInfo.get(a.dataId);\n    const refCount = info.refCount;\n\n    if (refCount <= 1) {\n      // Don't count bytes for complex numbers as they are counted by their\n      // components.\n      if (a.dtype !== 'complex64') {\n        this.state.numBytes -= info.bytes;\n      }\n      this.state.numDataBuffers--;\n\n      info.backend.disposeData(a.dataId);\n      this.state.tensorInfo.delete(a.dataId);\n    } else {\n      this.state.tensorInfo.get(a.dataId).refCount--;\n    }\n    // TODO(nsthorat): Construct an error and save the stack trace for\n    // debugging when in debug mode. Creating a stack trace is too expensive\n    // to do unconditionally.\n  }\n\n  disposeVariables(): void {\n    for (const varName in this.state.registeredVariables) {\n      const v = this.state.registeredVariables[varName];\n      this.disposeVariable(v);\n    }\n  }\n\n  disposeVariable(v: Variable): void {\n    this.disposeTensor(v);\n    if (this.state.registeredVariables[v.name] != null) {\n      delete this.state.registeredVariables[v.name];\n    }\n  }\n\n  memory(): MemoryInfo {\n    const info = this.backend.memory() as MemoryInfo;\n    info.numTensors = this.state.numTensors;\n    info.numDataBuffers = this.state.numDataBuffers;\n    info.numBytes = this.state.numBytes;\n    if (this.state.numStringTensors > 0) {\n      info.unreliable = true;\n      if (info.reasons == null) {\n        info.reasons = [];\n      }\n      info.reasons.push(\n          'Memory usage by string tensors is approximate ' +\n          '(2 bytes per character)');\n    }\n    return info;\n  }\n\n  async profile(query: () => (TensorContainer | Promise<TensorContainer>)):\n      Promise<ProfileInfo> {\n    this.state.profiling = true;\n\n    const startBytes = this.state.numBytes;\n    const startNumTensors = this.state.numTensors;\n\n    this.state.activeProfile.kernels = [];\n    this.state.activeProfile.result = await query();\n\n    this.state.profiling = false;\n\n    this.state.activeProfile.peakBytes = Math.max(\n        ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n    this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n    this.state.activeProfile.newTensors =\n        this.state.numTensors - startNumTensors;\n    for (const kernel of this.state.activeProfile.kernels) {\n      kernel.kernelTimeMs = await kernel.kernelTimeMs;\n      kernel.extraInfo = await kernel.extraInfo;\n    }\n    return this.state.activeProfile;\n  }\n\n  isTapeOn(): boolean {\n    return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n  }\n\n  private addTapeNode(\n      kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n      gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n    const tapeNode: TapeNode =\n        {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      gradientsFunc = gradConfig.gradFunc;\n    }\n    if (gradientsFunc != null) {\n      tapeNode.gradient = (dys: Tensor[]) => {\n        // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n        // the backprop graph to the user as null instead of zeros\n        dys = dys.map((dy, i) => {\n          if (dy == null) {\n            const output = outputs[i];\n            const vals = util.makeZerosTypedArray(output.size, output.dtype);\n            return this.makeTensor(vals, output.shape, output.dtype);\n          }\n          return dy;\n        });\n        // Grad functions of ops with single outputs expect a dy, while ops\n        // with multiple outputs expect dys (array of dy).\n        return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n      };\n    }\n    this.state.activeTape.push(tapeNode);\n  }\n\n  keep<T extends Tensor>(result: T): T {\n    result.kept = true;\n    return result;\n  }\n\n  private startTape() {\n    if (this.state.gradientDepth === 0) {\n      this.state.activeTape = [];\n    }\n    this.state.gradientDepth++;\n  }\n\n  private endTape() {\n    this.state.gradientDepth--;\n  }\n\n  /**\n   * Start a scope. Use this with endScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  startScope(name?: string) {\n    const scopeInfo: ScopeState = {\n      track: [],\n      name: 'unnamed scope',\n      id: this.state.nextScopeId++\n    };\n    if (name) {\n      scopeInfo.name = name;\n    }\n    this.state.scopeStack.push(scopeInfo);\n    this.state.activeScope = scopeInfo;\n  }\n\n  /**\n   * End a scope. Use this with startScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  endScope(result?: TensorContainer) {\n    const tensorsToTrackInParent = getTensorsInContainer(result);\n    const tensorsToTrackInParentSet =\n        new Set(tensorsToTrackInParent.map(t => t.id));\n\n    // Dispose the arrays tracked in this scope.\n    for (let i = 0; i < this.state.activeScope.track.length; i++) {\n      const tensor = this.state.activeScope.track[i];\n      if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n        tensor.dispose();\n      }\n    }\n\n    const oldScope = this.state.scopeStack.pop();\n    this.state.activeScope = this.state.scopeStack.length === 0 ?\n        null :\n        this.state.scopeStack[this.state.scopeStack.length - 1];\n\n    // Track the current result in the parent scope.\n    tensorsToTrackInParent.forEach(tensor => {\n      // Only track the tensor if was allocated in the inner scope and is not\n      // globally kept.\n      if (!tensor.kept && tensor.scopeId === oldScope.id) {\n        this.track(tensor);\n      }\n    });\n  }\n\n  /**\n   * Returns gradients of `f` with respect to each of the `xs`. The gradients\n   * returned are of the same length as `xs`, but some might be null if `f`\n   * was not a function of that `x`. It also takes optional dy to multiply the\n   * gradient, which defaults to `1`.\n   */\n  gradients<T extends Tensor>(\n      f: () => T, xs: Tensor[], dy?: T,\n      allowNoGradients = false): {value: T, grads: Tensor[]} {\n    util.assert(\n        xs.length > 0, () => 'gradients() received an empty list of xs.');\n    if (dy != null && dy.dtype !== 'float32') {\n      throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n    }\n\n    const y = this.scopedRun(\n        () => this.startTape(), () => this.endTape(),\n        () => this.tidy('forward', f));\n\n    util.assert(\n        y instanceof Tensor,\n        () => 'The result y returned by f() must be a tensor.');\n    // Filter out the nodes that don't connect x => y.\n    const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n    if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n      throw new Error(\n          'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n          'that the f you passed encloses all operations that lead from x ' +\n          'to y.');\n    }\n\n    return this.tidy('backward', () => {\n      const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n      accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n      // Backprop gradients through the filtered nodes.\n      backpropagateGradients(\n          accumulatedGradientMap, filteredTape,\n          // Pass the tidy function to avoid circular dep with `tape.ts`.\n          f => this.tidy(f as ScopeFn<Tensor>),\n          // Pass an add function to avoide a circular dep with `tape.ts`.\n          add);\n      const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n      if (this.state.gradientDepth === 0) {\n        // This means that we are not computing higher-order gradients\n        // and can clean up the tape.\n        this.state.activeTape.forEach(node => {\n          for (const tensor of node.saved) {\n            tensor.dispose();\n          }\n        });\n        this.state.activeTape = null;\n      }\n      return {value: y, grads};\n    });\n  }\n\n  customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n      (...args: Array<Tensor|GradSaveFunc>) => T {\n    util.assert(\n        util.isFunction(f),\n        () => 'The f passed in customGrad(f) must be a function.');\n    return (...inputs: Tensor[]): T => {\n      util.assert(\n          inputs.every(t => t instanceof Tensor),\n          () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n              'tensors');\n\n      let res: {\n        value: T,\n        gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n      };\n      const inputMap: NamedTensorMap = {};\n      inputs.forEach((input, i) => {\n        inputMap[i] = input;\n      });\n      return this.runKernelFunc(\n          (_, save) => {\n            res = f(...[...inputs, save]);\n            util.assert(\n                res.value instanceof Tensor,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.value` is a tensor');\n            util.assert(\n                util.isFunction(res.gradFunc),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function.');\n            return res.value;\n          },\n          inputMap,\n          (dy: T, saved: Tensor[]) => {\n            const gradRes = res.gradFunc(dy, saved);\n            const grads: Tensor[] =\n                Array.isArray(gradRes) ? gradRes : [gradRes];\n            util.assert(\n                grads.length === inputs.length,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'the same number of tensors as inputs passed to f(...).');\n            util.assert(\n                grads.every(t => t instanceof Tensor),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'a list of only tensors.');\n            const gradMap: {[key: string]: () => Tensor} = {};\n            grads.forEach((grad, i) => {\n              gradMap[i] = () => grad;\n            });\n            return gradMap;\n          });\n    };\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.readSync(dataId);\n  }\n  read(dataId: DataId): Promise<BackendValues> {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.read(dataId);\n  }\n\n  async time(query: () => void): Promise<TimingInfo> {\n    const start = now();\n    const timingInfo = await this.backend.time(query) as TimingInfo;\n    timingInfo.wallMs = now() - start;\n    return timingInfo;\n  }\n\n  /**\n   * Tracks a Tensor in the current scope to be automatically cleaned up\n   * when the current scope ends, and returns the value.\n   *\n   * @param result The Tensor to track in the current scope.\n   */\n  private track<T extends Tensor>(result: T): T {\n    if (this.state.activeScope != null) {\n      result.scopeId = this.state.activeScope.id;\n      this.state.activeScope.track.push(result);\n    }\n\n    return result;\n  }\n\n  get registeredVariables(): NamedVariableMap {\n    return this.state.registeredVariables;\n  }\n\n  /**\n   * Resets the engine state. Removes all backends but does not remove\n   * registered backend factories.\n   */\n  reset(): void {\n    // Make any pending promise obsolete.\n    this.pendingBackendInitId++;\n\n    this.state.dispose();\n    this.ENV.reset();\n    this.state = new EngineState();\n\n    for (const backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n    this.backendName = null;\n    this.backendInstance = null;\n    this.pendingBackendInit = null;\n  }\n}\n\nfunction ones(shape: number[]): Tensor {\n  const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n  return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n  const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n  if (ns._tfengine == null) {\n    const environment = new Environment(ns);\n    ns._tfengine = new Engine(environment);\n  }\n  setEnvironmentGlobal(ns._tfengine.ENV);\n\n  // Tell the current tensor interface that the global engine is responsible\n  // for tracking.\n  setTensorTracker(() => ns._tfengine);\n  return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n  // We duplicate Add here to avoid a circular dependency with add.ts.\n  const inputs = {a, b};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.add(a, b);\n    save([a, b]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n  return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n  if (_isNavigatorDefined()) {\n    // tslint:disable-next-line:no-any\n    const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n    // tslint:disable-next-line:max-line-length\n    return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n               .test(a) ||\n        // tslint:disable-next-line:max-line-length\n        /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n            .test(a.substr(0, 4));\n  }\n  return false;\n}\n\nexport function isBrowser(): boolean {\n  return (typeof window !== 'undefined' && window.document != null) ||\n      //@ts-ignore\n      (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n  if (debugValue) {\n    console.warn(\n        'Debugging mode is ON. The output of every math call will ' +\n        'be downloaded to CPU and checked for NaNs. ' +\n        'This significantly impacts performance.');\n  }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n    'IS_NODE',\n    () => (typeof process !== 'undefined') &&\n        (typeof process.versions !== 'undefined') &&\n        (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n    'IS_CHROME',\n    () => typeof navigator !== 'undefined' && navigator != null &&\n        navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n        /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n    'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n  let firstElem: typeof val = val;\n\n  if (isTypedArray(val)) {\n    return dtype === 'string' ? [] : [val.length];\n  }\n  if (!Array.isArray(val)) {\n    return [];  // Scalar.\n  }\n  const shape: number[] = [];\n\n  while (Array.isArray(firstElem) ||\n         isTypedArray(firstElem) && dtype !== 'string') {\n    shape.push(firstElem.length);\n    firstElem = firstElem[0];\n  }\n  if (Array.isArray(val) &&\n      env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n    deepAssertShapeConsistency(val, shape, []);\n  }\n\n  return shape;\n}\n\nfunction deepAssertShapeConsistency(\n    val: TensorLike, shape: number[], indices: number[]) {\n  indices = indices || [];\n  if (!(Array.isArray(val)) && !isTypedArray(val)) {\n    assert(\n        shape.length === 0,\n        () => `Element arr[${indices.join('][')}] is a primitive, ` +\n            `but should be an array/TypedArray of ${shape[0]} elements`);\n    return;\n  }\n  assert(\n      shape.length > 0,\n      () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n          `but is an array of ${val.length} elements`);\n  assert(\n      val.length === shape[0],\n      () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n          `elements, but has ${val.length} elements`);\n  const subShape = shape.slice(1);\n  for (let i = 0; i < val.length; ++i) {\n    deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n  }\n}\n\nfunction assertDtype(\n    expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n    functionName: string) {\n  if (expectedDtype == null) {\n    return;\n  }\n  if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n      expectedDtype === 'numeric' && actualDType === 'string') {\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must ` +\n        `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n  }\n}\n\nexport function convertToTensor<T extends Tensor>(\n    x: T|TensorLike, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T {\n  if (x instanceof Tensor) {\n    assertDtype(parseAsDtype, x.dtype, argName, functionName);\n    return x;\n  }\n  let inferredDtype = inferDtype(x);\n  // If the user expects a bool/int/float, use that info to update the\n  // inferredDtype when it is not a string.\n  if (inferredDtype !== 'string' &&\n      ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n    inferredDtype = parseAsDtype as DataType;\n  }\n  assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n  if ((x == null) ||\n      (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n       typeof x !== 'boolean' && typeof x !== 'string')) {\n    const type = x == null ? 'null' : (x as {}).constructor.name;\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must be a ` +\n        `Tensor or TensorLike, but got '${type}'`);\n  }\n  const inferredShape = inferShape(x, inferredDtype);\n  if (!isTypedArray(x) && !Array.isArray(x)) {\n    x = [x] as number[];\n  }\n  const skipTypedArray = true;\n  const values = inferredDtype !== 'string' ?\n      toTypedArray(x, inferredDtype as DataType) :\n      flatten(x as string[], [], skipTypedArray) as string[];\n  return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray<T extends Tensor>(\n    arg: Array<T|TensorLike>, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n  if (!Array.isArray(arg)) {\n    throw new Error(\n        `Argument ${argName} passed to ${functionName} must be a ` +\n        '`Tensor[]` or `TensorLike[]`');\n  }\n  const tensors = arg as T[];\n  return tensors.map(\n      (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n      parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op<T extends Function>(f: {[name: string]: T}): T {\n  const keys = Object.keys(f);\n  if (keys.length !== 1) {\n    throw new Error(\n        `Please provide an object with a single key ` +\n        `(operation name) mapping to a function. Got an object with ` +\n        `${keys.length} keys.`);\n  }\n\n  let opName = keys[0];\n  const fn = f[opName];\n\n  // Strip the underscore from the end of the function name.\n  if (opName.endsWith('_')) {\n    opName = opName.substring(0, opName.length - 1);\n  }\n\n  // add an __op suffix to distinguish ops from kernels in tf.profile\n  opName = opName + OP_SCOPE_SUFFIX;\n\n  // tslint:disable-next-line:no-any\n  const f2 = (...args: any[]) => {\n    ENGINE.startScope(opName);\n    try {\n      const result = fn(...args);\n      if (result instanceof Promise) {\n        console.error('Cannot return a Promise inside of tidy.');\n      }\n      ENGINE.endScope(result);\n      return result;\n    } catch (ex) {\n      ENGINE.endScope(null);\n      throw ex;\n    }\n  };\n  Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n  // tslint:disable-next-line:no-any\n  return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_<T extends Tensor>(real: T|TensorLike, imag: T|TensorLike): T {\n  const $real = convertToTensor(real, 'real', 'complex');\n  const $imag = convertToTensor(imag, 'imag', 'complex');\n  util.assertShapesMatch(\n      $real.shape, $imag.shape,\n      `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n          `must match in call to tf.complex().`);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.complex($real, $imag);\n  };\n  const inputs: ComplexInputs = {real: $real, imag: $imag};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n    values: TensorLike, shape: number[], inferredShape: number[],\n    dtype?: DataType): Tensor {\n  if (dtype == null) {\n    dtype = inferDtype(values);\n  }\n  if (dtype === 'complex64') {\n    throw new Error(\n        `Cannot construct a complex64 tensor directly. ` +\n        `Please use tf.complex(real, imag).`);\n  }\n  if (!isTypedArray(values) && !Array.isArray(values) &&\n      typeof values !== 'number' && typeof values !== 'boolean' &&\n      typeof values !== 'string') {\n    throw new Error(\n        'values passed to tensor(values) must be a number/boolean/string or ' +\n        'an array of numbers/booleans/strings, or a TypedArray');\n  }\n  if (shape != null) {\n    assertNonNegativeIntegerDimensions(shape);\n\n    const providedSize = sizeFromShape(shape);\n    const inferredSize = sizeFromShape(inferredShape);\n    assert(\n        providedSize === inferredSize,\n        () =>\n            `Based on the provided shape, [${shape}], the tensor should have ` +\n            `${providedSize} values but has ${inferredSize}`);\n\n    for (let i = 0; i < inferredShape.length; ++i) {\n      const inferred = inferredShape[i];\n      const flatDimsDontMatch = i === inferredShape.length - 1 ?\n          inferred !== sizeFromShape(shape.slice(i)) :\n          true;\n      assert(\n          inferredShape[i] === shape[i] || !flatDimsDontMatch,\n          () => `Error creating a new Tensor. Inferred shape ` +\n              `(${inferredShape}) does not match the provided ` +\n              `shape (${shape}). `);\n    }\n  }\n\n  if (!isTypedArray(values) && !Array.isArray(values)) {\n    values = [values] as number[];\n  }\n\n  shape = shape || inferredShape;\n  values = dtype !== 'string' ?\n      toTypedArray(values, dtype) :\n      flatten(values as string[], [], true) as string[];\n  return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`. If the values are strings,\n *     they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor<R extends Rank>(\n    values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor<R> {\n  const inferredShape = inferShape(values, dtype);\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n  'float32': 4,\n  'float16': 2,\n  'int32': 4,\n  'uint16': 2,\n  'uint8': 1,\n  'bool': 1,\n  'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n  /**\n   * An ordered list of paths.\n   *\n   * Paths are intentionally abstract in order to be general. For example, they\n   * can be relative URL paths or relative paths on the file system.\n   */\n  paths: string[];\n\n  /**\n   * Specifications of the weights stored in the paths.\n   */\n  weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n  /**\n   * Name of the weight, e.g., 'Dense_1/bias'\n   */\n  name: string;\n\n  /**\n   * Shape of the weight.\n   */\n  shape: number[];\n\n  /**\n   * Data type of the weight.\n   */\n  dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n  /**\n   * Type of the weight.\n   *\n   * Optional.\n   *\n   * The value 'optimizer' indicates the weight belongs to an optimizer\n   * (i.e., used only during model training and not during inference).\n   */\n  group?: WeightGroup;\n\n  /**\n   * Information for dequantization of the weight.\n   */\n  quantization?: {\n    scale?: number,  // The scaling constant to multiply by.\n    min?: number,    // The (possibly nudged) minimum weight to add.\n       dtype: 'uint16'|'uint8'|'float16'  // The dtype of the quantized weights.\n  };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n  /**\n   * Whether to save only the trainable weights of the model, ignoring the\n   * non-trainable ones.\n   */\n  trainableOnly?: boolean;\n\n  /**\n   * Whether the optimizer will be saved (if exists).\n   *\n   * Default: `false`.\n   */\n  includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n  /**\n   * Information about the model artifacts saved.\n   */\n  modelArtifactsInfo: ModelArtifactsInfo;\n\n  /**\n   * HTTP responses from the server that handled the model-saving request (if\n   * any). This is applicable only to server-based saving routes.\n   */\n  responses?: Response[];\n\n  /**\n   * Error messages and related data (if any).\n   */\n  errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n  /**\n   * Timestamp for when the model is saved.\n   */\n  dateSaved: Date;\n\n  /**\n   * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n   * come in a JSON format and none of our IOHandlers support a non json\n   * format. We could conder replacing this with 'Binary' if we want to\n   * allow future handlers to save to non json formats (though they will\n   * probably want more information than 'Binary').\n   * Type of the model topology\n   *\n   * Type of the model topology\n   *\n   * Possible values:\n   *   - JSON: JSON config (human-readable, e.g., Keras JSON).\n   *   - GraphDef: TensorFlow\n   *     [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n   *     protocol buffer (binary).\n   */\n  modelTopologyType: 'JSON'|'GraphDef';\n\n  /**\n   * Size of model topology (Keras JSON or GraphDef), in bytes.\n   */\n  modelTopologyBytes?: number;\n\n  /**\n   * Size of weight specification or manifest, in bytes.\n   */\n  weightSpecsBytes?: number;\n\n  /**\n   * Size of weight value data, in bytes.\n   */\n  weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  // See\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n  /** Optimizer used for the model training. */\n  optimizer_config: {};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Loss function(s) for the model's output(s). */\n  loss: string|string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Metric function(s) for the model's output(s). */\n  metrics?: string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  weighted_metrics?: string[];\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  sample_weight_mode?: string;\n\n  loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers.  For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology?: {}|ArrayBuffer;\n\n  /**\n   * Serialized configuration for the model's training.\n   */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weight specifications.\n   *\n   * This corresponds to the weightsData below.\n   */\n  weightSpecs?: WeightsManifestEntry[];\n\n  /**\n   * Binary buffer for all weight values concatenated in the order specified\n   * by `weightSpecs`.\n   */\n  weightData?: ArrayBuffer;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology: {};\n\n  /** Model training configuration. */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weights manifest.\n   *\n   * The weights manifest consists of an ordered list of weight-manifest\n   * groups. Each weight-manifest group consists of a number of weight values\n   * stored in a number of paths. See the documentation of\n   * `WeightsManifestConfig` for more details.\n   */\n  weightsManifest: WeightsManifestConfig;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise<ModelArtifacts>;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n    Promise<SaveResult>;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n  save?: SaveHandler;\n  load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n  /**\n   * List all models in the model store.\n   *\n   * @returns A dictionary mapping paths of existing models to their\n   *   model artifacts info. Model artifacts info include type of the model's\n   *   topology, byte sizes of the topology, weights, etc.\n   */\n  listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n  /**\n   * Remove a model specified by `path`.\n   *\n   * @param path\n   * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n   *   is successful).\n   * @throws Error if deletion fails, e.g., if no model exists at `path`.\n   */\n  removeModel(path: string): Promise<ModelArtifactsInfo>;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n  /**\n   * RequestInit (options) for HTTP requests.\n   *\n   * For detailed information on the supported fields, see\n   * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n   *     https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n   */\n  requestInit?: RequestInit;\n\n  /**\n   * Progress callback.\n   */\n  onProgress?: OnProgressCallback;\n\n  /**\n   * A function used to override the `window.fetch` function.\n   */\n  fetchFunc?: Function;\n\n  /**\n   * Strict loading model: whether extraneous weights or missing\n   * weights should trigger an `Error`.\n   *\n   * If `true`, require that the provided weights exactly match those\n   * required by the layers. `false` means that both extra weights\n   * and missing weights will be silently ignored.\n   *\n   * Default: `true`.\n   */\n  strict?: boolean;\n\n  /**\n   * Path prefix for weight files, by default this is calculated from the\n   * path of the model JSON file.\n   *\n   * For instance, if the path to the model JSON file is\n   * `http://localhost/foo/model.json`, then the default path prefix will be\n   * `http://localhost/foo/`. If a weight file has the path value\n   * `group1-shard1of2` in the weight manifest, then the weight file will be\n   * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n   * if you provide a `weightPathPrefix` value of\n   * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n   * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n   */\n  weightPathPrefix?: string;\n\n  /**\n   * Whether the module or model is to be loaded from TF Hub.\n   *\n   * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n   * standard model file name and the query parameters.\n   *\n   * Default: `false`.\n   */\n  fromTFHub?: boolean;\n\n  /**\n   * An async function to convert weight file name to URL. The weight file\n   * names are stored in model.json's weightsManifest.paths field. By default we\n   * consider weight files are colocated with the model.json file. For example:\n   *     model.json URL: https://www.google.com/models/1/model.json\n   *     group1-shard1of1.bin url:\n   *        https://www.google.com/models/1/group1-shard1of1.bin\n   *\n   * With this func you can convert the weight file name to any URL.\n   */\n  weightUrlConverter?: (weightFileName: string) => Promise<string>;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n  /**\n   * Is this request for a binary file (as opposed to a json file)\n   */\n  isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n *     concatenated.\n *   - An `Array` of `WeightManifestEntry`s, carrying information including\n *     tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n    tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n    Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n  // TODO(adarob, cais): Support quantization.\n  const specs: WeightsManifestEntry[] = [];\n  const dataPromises: Array<Promise<TypedArray>> = [];\n\n  const names: string[] = Array.isArray(tensors) ?\n      tensors.map(tensor => tensor.name) :\n      Object.keys(tensors);\n\n  for (let i = 0; i < names.length; ++i) {\n    const name = names[i];\n    const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n    if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n        t.dtype !== 'string' && t.dtype !== 'complex64') {\n      throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n    }\n    const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n    if (t.dtype === 'string') {\n      const utf8bytes = new Promise<TypedArray>(async resolve => {\n        const vals = await t.bytes() as Uint8Array[];\n        const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n            NUM_BYTES_STRING_LENGTH * vals.length;\n        const bytes = new Uint8Array(totalNumBytes);\n        let offset = 0;\n        for (let i = 0; i < vals.length; i++) {\n          const val = vals[i];\n          const bytesOfLength =\n              new Uint8Array(new Uint32Array([val.length]).buffer);\n          bytes.set(bytesOfLength, offset);\n          offset += NUM_BYTES_STRING_LENGTH;\n          bytes.set(val, offset);\n          offset += val.length;\n        }\n        resolve(bytes);\n      });\n      dataPromises.push(utf8bytes);\n    } else {\n      dataPromises.push(t.data());\n    }\n    if (group != null) {\n      spec.group = group;\n    }\n    specs.push(spec);\n  }\n\n  const tensorValues = await Promise.all(dataPromises);\n  return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n *   concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n *   whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n *   to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n    buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n  // TODO(adarob, cais): Support quantization.\n  const out: NamedTensorMap = {};\n  let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n  let offset = 0;\n  for (const spec of specs) {\n    const name = spec.name;\n    const dtype = spec.dtype;\n    const shape = spec.shape;\n    const size = sizeFromShape(shape);\n    let values: TypedArray|string[]|Uint8Array[];\n\n    if ('quantization' in spec) {\n      const quantization = spec.quantization;\n      if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n        if (!('min' in quantization && 'scale' in quantization)) {\n          throw new Error(\n              `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n              `doesn't have corresponding metadata min and scale.`);\n        }\n      } else if (quantization.dtype === 'float16') {\n        if (dtype !== 'float32') {\n          throw new Error(\n              `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n              `which only supports weights of type float32 not ${dtype}.`);\n        }\n      } else {\n        throw new Error(\n            `Weight ${spec.name} has unknown ` +\n            `quantization dtype ${quantization.dtype}. ` +\n            `Supported quantization dtypes are: ` +\n            `'uint8', 'uint16', and 'float16'.`);\n      }\n      const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n      const byteBuffer =\n          buffer.slice(offset, offset + size * quantizationSizeFactor);\n      const quantizedArray = (quantization.dtype === 'uint8') ?\n          new Uint8Array(byteBuffer) :\n          new Uint16Array(byteBuffer);\n      if (dtype === 'float32') {\n        if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n          values = new Float32Array(quantizedArray.length);\n          for (let i = 0; i < quantizedArray.length; i++) {\n            const v = quantizedArray[i];\n            values[i] = v * quantization.scale + quantization.min;\n          }\n        } else if (quantization.dtype === 'float16') {\n          if (float16Decode === undefined) {\n            float16Decode = getFloat16Decoder();\n          }\n          values = float16Decode(quantizedArray as Uint16Array);\n        } else {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type float32.`);\n        }\n      } else if (dtype === 'int32') {\n        if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type int32.`);\n        }\n        values = new Int32Array(quantizedArray.length);\n        for (let i = 0; i < quantizedArray.length; i++) {\n          const v = quantizedArray[i];\n          values[i] = Math.round(v * quantization.scale + quantization.min);\n        }\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * quantizationSizeFactor;\n    } else if (dtype === 'string') {\n      const size = sizeFromShape(spec.shape);\n      values = [];\n      for (let i = 0; i < size; i++) {\n        const byteLength = new Uint32Array(\n            buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n        offset += NUM_BYTES_STRING_LENGTH;\n        const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n        (values as Uint8Array[]).push(bytes);\n        offset += byteLength;\n      }\n    } else {\n      const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n      const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n      if (dtype === 'float32') {\n        values = new Float32Array(byteBuffer);\n      } else if (dtype === 'int32') {\n        values = new Int32Array(byteBuffer);\n      } else if (dtype === 'bool') {\n        values = new Uint8Array(byteBuffer);\n      } else if (dtype === 'complex64') {\n        values = new Float32Array(byteBuffer);\n        const real = new Float32Array(values.length / 2);\n        const image = new Float32Array(values.length / 2);\n        for (let i = 0; i < real.length; i++) {\n          real[i] = values[i * 2];\n          image[i] = values[i * 2 + 1];\n        }\n        const realTensor = tensor(real, shape, 'float32');\n        const imageTensor = tensor(image, shape, 'float32');\n        out[name] = complex(realTensor, imageTensor);\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * dtypeFactor;\n    }\n    if (dtype !== 'complex64') {\n      out[name] = tensor(values, shape, dtype);\n    }\n  }\n  return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n  // TODO(adarob, cais): Support quantization.\n  if (xs === null) {\n    throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n  }\n\n  let totalByteLength = 0;\n\n  // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n  // can have a different byte length from that of the `TypedArray` itself,\n  // for example, when the `TypedArray` is created from an offset in an\n  // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n  // the `TypedArray` in byte length. If an element of `xs` does not show\n  // this property, a new `TypedArray` that satisfy this property will be\n  // constructed and pushed into `normalizedXs`.\n  const normalizedXs: TypedArray[] = [];\n  xs.forEach((x: TypedArray) => {\n    totalByteLength += x.byteLength;\n    // tslint:disable:no-any\n    normalizedXs.push(\n        x.byteLength === x.buffer.byteLength ? x :\n                                               new (x.constructor as any)(x));\n    if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n          x as any instanceof Uint8Array)) {\n      throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n    }\n    // tslint:enable:no-any\n  });\n\n  const y = new Uint8Array(totalByteLength);\n  let offset = 0;\n  normalizedXs.forEach((x: TypedArray) => {\n    y.set(new Uint8Array(x.buffer), offset);\n    offset += x.byteLength;\n  });\n\n  return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n    (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n     typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n  if (useNodeBuffer) {\n    return Buffer.byteLength(str);\n  }\n  return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n  if (useNodeBuffer) {\n    return Buffer.from(buffer).toString('base64');\n  }\n  const buf = new Uint8Array(buffer);\n  let s = '';\n  for (let i = 0, l = buf.length; i < l; i++) {\n    s += String.fromCharCode(buf[i]);\n  }\n  return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n  if (useNodeBuffer) {\n    const buf = Buffer.from(str, 'base64');\n    return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n  }\n  const s = atob(str);\n  const buffer = new Uint8Array(s.length);\n  for (let i = 0; i < s.length; ++i) {\n    buffer.set([s.charCodeAt(i)], i);\n  }\n  return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n  if (buffers.length === 1) {\n    return buffers[0];\n  }\n\n  let totalByteLength = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    totalByteLength += buffer.byteLength;\n  });\n\n  const temp = new Uint8Array(totalByteLength);\n  let offset = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    temp.set(new Uint8Array(buffer), offset);\n    offset += buffer.byteLength;\n  });\n  return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n  const SEPARATOR = '/';\n  path = path.trim();\n  while (path.endsWith(SEPARATOR)) {\n    path = path.slice(0, path.length - 1);\n  }\n  const items = path.split(SEPARATOR);\n  return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n    ModelArtifactsInfo {\n  if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n    throw new Error('Expected JSON model topology, received ArrayBuffer.');\n  }\n\n  return {\n    dateSaved: new Date(),\n    modelTopologyType: 'JSON',\n    modelTopologyBytes: modelArtifacts.modelTopology == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n    weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n    weightDataBytes: modelArtifacts.weightData == null ?\n        0 :\n        modelArtifacts.weightData.byteLength,\n  };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n  const convertMantissa = (i: number): number => {\n    let m = i << 13;\n    let e = 0;\n\n    while ((m & 0x00800000) === 0) {\n      e -= 0x00800000;\n      m <<= 1;\n    }\n    m &= ~0x00800000;\n    e += 0x38800000;\n\n    return m | e;\n  };\n\n  const mantisaTable = new Uint32Array(2048);\n\n  mantisaTable[0] = 0;\n  for (let i = 1; i < 1024; i++) {\n    mantisaTable[i] = convertMantissa(i);\n  }\n  for (let i = 1024; i < 2048; i++) {\n    mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n  }\n\n  return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n  const exponentTable = new Uint32Array(64);\n\n  exponentTable[0] = 0;\n  exponentTable[31] = 0x47800000;\n  exponentTable[32] = 0x80000000;\n  exponentTable[63] = 0xc7800000;\n  for (let i = 1; i < 31; i++) {\n    exponentTable[i] = i << 23;\n  }\n  for (let i = 33; i < 63; i++) {\n    exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n  }\n\n  return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n  const offsetTable = new Uint32Array(64);\n\n  for (let i = 0; i < 64; i++) {\n    offsetTable[i] = 1024;\n  }\n  offsetTable[0] = offsetTable[32] = 0;\n\n  return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n *          the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n  // Algorithm is based off of\n  // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n  // Cache lookup tables\n  const mantisaTable = computeFloat16MantisaTable();\n  const exponentTable = computeFloat16ExponentTable();\n  const offsetTable = computeFloat16OffsetTable();\n\n  return (quantizedArray: Uint16Array) => {\n    const buffer = new ArrayBuffer(4 * quantizedArray.length);\n    const bufferUint32View = new Uint32Array(buffer);\n    for (let index = 0; index < quantizedArray.length; index++) {\n      const float16Bits = quantizedArray[index];\n      const float32Bits =\n          mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n          exponentTable[float16Bits >> 10];\n      bufferUint32View[index] = float32Bits;\n    }\n    return new Float32Array(buffer);\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n    IOHandler;\n\nexport class IORouterRegistry {\n  // Singleton instance.\n  private static instance: IORouterRegistry;\n\n  private saveRouters: IORouter[];\n  private loadRouters: IORouter[];\n\n  private constructor() {\n    this.saveRouters = [];\n    this.loadRouters = [];\n  }\n\n  private static getInstance(): IORouterRegistry {\n    if (IORouterRegistry.instance == null) {\n      IORouterRegistry.instance = new IORouterRegistry();\n    }\n    return IORouterRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerSaveRouter(saveRouter: IORouter) {\n    IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n  }\n\n  /**\n   * Register a load-handler router.\n   *\n   * @param loadRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `load` method defined or `null`.\n   */\n  static registerLoadRouter(loadRouter: IORouter) {\n    IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n  }\n\n  /**\n   * Look up IOHandler for saving, given a URL-like string.\n   *\n   * @param url\n   * @returns If only one match is found, an instance of IOHandler with the\n   * `save` method defined. If no match is found, `null`.\n   * @throws Error, if more than one match is found.\n   */\n  static getSaveHandlers(url: string|string[]): IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'save');\n  }\n\n  /**\n   * Look up IOHandler for loading, given a URL-like string.\n   *\n   * @param url\n   * @param loadOptions Optional, custom load options.\n   * @returns All valid handlers for `url`, given the currently registered\n   *   handler routers.\n   */\n  static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n      IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n  }\n\n  private static getHandlers(\n      url: string|string[], handlerType: 'save'|'load',\n      loadOptions?: LoadOptions): IOHandler[] {\n    const validHandlers: IOHandler[] = [];\n    const routers = handlerType === 'load' ?\n        IORouterRegistry.getInstance().loadRouters :\n        IORouterRegistry.getInstance().saveRouters;\n    routers.forEach(router => {\n      const handler = router(url, loadOptions);\n      if (handler !== null) {\n        validHandlers.push(handler);\n      }\n    });\n    return validHandlers;\n  }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n    IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n    (url: string|string[], loadOptions?: LoadOptions) =>\n        IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n//    as the type of topology (JSON vs binary), byte size of the topology, byte\n//    size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise<void> {\n  const idbFactory = getIndexedDBFactory();\n\n  return new Promise<void>((resolve, reject) => {\n    const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n    deleteRequest.onsuccess = () => resolve();\n    deleteRequest.onerror = error => reject(error);\n  });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n  if (!env().getBool('IS_BROWSER')) {\n    // TODO(cais): Add more info about what IOHandler subtypes are available.\n    //   Maybe point to a doc page on the web and/or automatically determine\n    //   the available IOHandlers and print them in the error message.\n    throw new Error(\n        'Failed to obtain IndexedDB factory because the current environment' +\n        'is not a web browser.');\n  }\n  // tslint:disable-next-line:no-any\n  const theWindow: any = typeof window === 'undefined' ? self : window;\n  const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n      theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n      theWindow.shimIndexedDB;\n  if (factory == null) {\n    throw new Error(\n        'The current browser does not appear to support IndexedDB.');\n  }\n  return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n  const db = openRequest.result as IDBDatabase;\n  db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n  db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n  protected readonly indexedDB: IDBFactory;\n  protected readonly modelPath: string;\n\n  static readonly URL_SCHEME = 'indexeddb://';\n\n  constructor(modelPath: string) {\n    this.indexedDB = getIndexedDBFactory();\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For IndexedDB, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    // TODO(cais): Support saving GraphDef models.\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    return this.databaseAction(this.modelPath, modelArtifacts) as\n        Promise<SaveResult>;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    return this.databaseAction(this.modelPath) as Promise<ModelArtifacts>;\n  }\n\n  /**\n   * Perform database action to put model artifacts into or read model artifacts\n   * from IndexedDB object store.\n   *\n   * Whether the action is put or get depends on whether `modelArtifacts` is\n   * specified. If it is specified, the action will be put; otherwise the action\n   * will be get.\n   *\n   * @param modelPath A unique string path for the model.\n   * @param modelArtifacts If specified, it will be the model artifacts to be\n   *   stored in IndexedDB.\n   * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n   *   of `ModelArtifacts`, if the action is get.\n   */\n  private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n      Promise<ModelArtifacts|SaveResult> {\n    return new Promise<ModelArtifacts|SaveResult>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n\n        if (modelArtifacts == null) {\n          // Read model out from object store.\n          const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n          const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n          const getRequest = modelStore.get(this.modelPath);\n          getRequest.onsuccess = () => {\n            if (getRequest.result == null) {\n              db.close();\n              return reject(new Error(\n                  `Cannot find model with path '${this.modelPath}' ` +\n                  `in IndexedDB.`));\n            } else {\n              resolve(getRequest.result.modelArtifacts);\n            }\n          };\n          getRequest.onerror = error => {\n            db.close();\n            return reject(getRequest.error);\n          };\n          modelTx.oncomplete = () => db.close();\n        } else {\n          // Put model into object store.\n          const modelArtifactsInfo: ModelArtifactsInfo =\n              getModelArtifactsInfoForJSON(modelArtifacts);\n          // First, put ModelArtifactsInfo into info store.\n          const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n          let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n          const putInfoRequest =\n              infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n          let modelTx: IDBTransaction;\n          putInfoRequest.onsuccess = () => {\n            // Second, put model data into model store.\n            modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n            const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n            const putModelRequest = modelStore.put({\n              modelPath: this.modelPath,\n              modelArtifacts,\n              modelArtifactsInfo\n            });\n            putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n            putModelRequest.onerror = error => {\n              // If the put-model request fails, roll back the info entry as\n              // well.\n              infoStore = infoTx.objectStore(INFO_STORE_NAME);\n              const deleteInfoRequest = infoStore.delete(this.modelPath);\n              deleteInfoRequest.onsuccess = () => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n              deleteInfoRequest.onerror = error => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n            };\n          };\n          putInfoRequest.onerror = error => {\n            db.close();\n            return reject(putInfoRequest.error);\n          };\n          infoTx.oncomplete = () => {\n            if (modelTx == null) {\n              db.close();\n            } else {\n              modelTx.oncomplete = () => db.close();\n            }\n          };\n        }\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n      return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n *   which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n  return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n      key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n      key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n  private indexedDB: IDBFactory;\n\n  constructor() {\n    this.indexedDB = getIndexedDBFactory();\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    return new Promise<{[path: string]: ModelArtifactsInfo}>(\n        (resolve, reject) => {\n          const openRequest =\n              this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n          openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n          openRequest.onsuccess = () => {\n            const db = openRequest.result;\n            const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n            const store = tx.objectStore(INFO_STORE_NAME);\n            // tslint:disable:max-line-length\n            // Need to cast `store` as `any` here because TypeScript's DOM\n            // library does not have the `getAll()` method even though the\n            // method is supported in the latest version of most mainstream\n            // browsers:\n            // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n            // tslint:enable:max-line-length\n            // tslint:disable-next-line:no-any\n            const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n            getAllInfoRequest.onsuccess = () => {\n              const out: {[path: string]: ModelArtifactsInfo} = {};\n              for (const item of getAllInfoRequest.result) {\n                out[item.modelPath] = item.modelArtifactsInfo;\n              }\n              resolve(out);\n            };\n            getAllInfoRequest.onerror = error => {\n              db.close();\n              return reject(getAllInfoRequest.error);\n            };\n            tx.oncomplete = () => db.close();\n          };\n          openRequest.onerror = error => reject(openRequest.error);\n        });\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    return new Promise<ModelArtifactsInfo>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n        const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n        const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n        const getInfoRequest = infoStore.get(path);\n        let modelTx: IDBTransaction;\n        getInfoRequest.onsuccess = () => {\n          if (getInfoRequest.result == null) {\n            db.close();\n            return reject(new Error(\n                `Cannot find model with path '${path}' ` +\n                `in IndexedDB.`));\n          } else {\n            // First, delete the entry in the info store.\n            const deleteInfoRequest = infoStore.delete(path);\n            const deleteModelData = () => {\n              // Second, delete the entry in the model store.\n              modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n              const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n              const deleteModelRequest = modelStore.delete(path);\n              deleteModelRequest.onsuccess = () =>\n                  resolve(getInfoRequest.result.modelArtifactsInfo);\n              deleteModelRequest.onerror = error =>\n                  reject(getInfoRequest.error);\n            };\n            // Proceed with deleting model data regardless of whether deletion\n            // of info data succeeds or not.\n            deleteInfoRequest.onsuccess = deleteModelData;\n            deleteInfoRequest.onerror = error => {\n              deleteModelData();\n              db.close();\n              return reject(getInfoRequest.error);\n            };\n          }\n        };\n        getInfoRequest.onerror = error => {\n          db.close();\n          return reject(getInfoRequest.error);\n        };\n\n        infoTx.oncomplete = () => {\n          if (modelTx == null) {\n            db.close();\n          } else {\n            modelTx.oncomplete = () => db.close();\n          }\n        };\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n  if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n      typeof window.localStorage === 'undefined') {\n    throw new Error(\n        'purgeLocalStorageModels() cannot proceed because local storage is ' +\n        'unavailable in the current environment.');\n  }\n  const LS = window.localStorage;\n  const purgedModelPaths: string[] = [];\n  for (let i = 0; i < LS.length; ++i) {\n    const key = LS.key(i);\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    if (key.startsWith(prefix) && key.length > prefix.length) {\n      LS.removeItem(key);\n      const modelName = getModelPathFromKey(key);\n      if (purgedModelPaths.indexOf(modelName) === -1) {\n        purgedModelPaths.push(modelName);\n      }\n    }\n  }\n  return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n} {\n  return {\n    info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n    topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n    weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n    weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n    modelMetadata:\n        [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n  };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n  const items = key.split(PATH_SEPARATOR);\n  if (items.length < 3) {\n    throw new Error(`Invalid key format: ${key}`);\n  }\n  return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n      key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n      key;\n}\n\ndeclare type LocalStorageKeys = {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n  protected readonly LS: Storage;\n  protected readonly modelPath: string;\n  protected readonly keys: LocalStorageKeys;\n\n  static readonly URL_SCHEME = 'localstorage://';\n\n  constructor(modelPath: string) {\n    if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n        typeof window.localStorage === 'undefined') {\n      // TODO(cais): Add more info about what IOHandler subtypes are\n      // available.\n      //   Maybe point to a doc page on the web and/or automatically determine\n      //   the available IOHandlers and print them in the error message.\n      throw new Error(\n          'The current environment does not support local storage.');\n    }\n    this.LS = window.localStorage;\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For local storage, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n    this.keys = getModelKeys(this.modelPath);\n  }\n\n  /**\n   * Save model artifacts to browser local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @param modelArtifacts The model artifacts to be stored.\n   * @returns An instance of SaveResult.\n   */\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const topology = JSON.stringify(modelArtifacts.modelTopology);\n      const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n      const modelArtifactsInfo: ModelArtifactsInfo =\n          getModelArtifactsInfoForJSON(modelArtifacts);\n\n      try {\n        this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n        this.LS.setItem(this.keys.topology, topology);\n        this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n        this.LS.setItem(\n            this.keys.weightData,\n            arrayBufferToBase64String(modelArtifacts.weightData));\n        this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n          format: modelArtifacts.format,\n          generatedBy: modelArtifacts.generatedBy,\n          convertedBy: modelArtifacts.convertedBy,\n          userDefinedMetadata: modelArtifacts.userDefinedMetadata\n        }));\n\n        return {modelArtifactsInfo};\n      } catch (err) {\n        // If saving failed, clean up all items saved so far.\n        this.LS.removeItem(this.keys.info);\n        this.LS.removeItem(this.keys.topology);\n        this.LS.removeItem(this.keys.weightSpecs);\n        this.LS.removeItem(this.keys.weightData);\n        this.LS.removeItem(this.keys.modelMetadata);\n\n        throw new Error(\n            `Failed to save model '${this.modelPath}' to local storage: ` +\n            `size quota being exceeded is a possible cause of this failure: ` +\n            `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n            `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n            `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n      }\n    }\n  }\n\n  /**\n   * Load a model from local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const info =\n        JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n    if (info == null) {\n      throw new Error(\n          `In local storage, there is no model with name '${this.modelPath}'`);\n    }\n\n    if (info.modelTopologyType !== 'JSON') {\n      throw new Error(\n          'BrowserLocalStorage does not support loading non-JSON model ' +\n          'topology yet.');\n    }\n\n    const out: ModelArtifacts = {};\n\n    // Load topology.\n    const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n    if (topology == null) {\n      throw new Error(\n          `In local storage, the topology of model '${this.modelPath}' ` +\n          `is missing.`);\n    }\n    out.modelTopology = topology;\n\n    // Load weight specs.\n    const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n    if (weightSpecs == null) {\n      throw new Error(\n          `In local storage, the weight specs of model '${this.modelPath}' ` +\n          `are missing.`);\n    }\n    out.weightSpecs = weightSpecs;\n\n    // Load meta-data fields.\n    const metadataString = this.LS.getItem(this.keys.modelMetadata);\n    if (metadataString != null) {\n      const metadata = JSON.parse(metadataString) as ModelArtifacts;\n      out.format = metadata['format'];\n      out.generatedBy = metadata['generatedBy'];\n      out.convertedBy = metadata['convertedBy'];\n      out.userDefinedMetadata = metadata['userDefinedMetadata'];\n    }\n\n    // Load weight data.\n    const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n    if (weightDataBase64 == null) {\n      throw new Error(\n          `In local storage, the binary weight values of model ` +\n          `'${this.modelPath}' are missing.`);\n    }\n    out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n    return out;\n  }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n      return browserLocalStorage(\n          url.slice(BrowserLocalStorage.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n *   - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n *     model, such as date saved, type of the topology, size in bytes, etc.\n *   - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n *     style models, this is a stringized JSON.\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n *     model, can be used to decode the saved binary weight values (see\n *     item below).\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n *     weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n *   `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n  return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n  private readonly LS: Storage;\n\n  constructor() {\n    assert(\n        env().getBool('IS_BROWSER'),\n        () => 'Current environment is not a web browser');\n    assert(\n        typeof window === 'undefined' ||\n            typeof window.localStorage !== 'undefined',\n        () => 'Current browser does not appear to support localStorage');\n    this.LS = window.localStorage;\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    const out: {[path: string]: ModelArtifactsInfo} = {};\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n    for (let i = 0; i < this.LS.length; ++i) {\n      const key = this.LS.key(i);\n      if (key.startsWith(prefix) && key.endsWith(suffix)) {\n        const modelPath = getModelPathFromKey(key);\n        out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n      }\n    }\n    return out;\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    const keys = getModelKeys(path);\n    if (this.LS.getItem(keys.info) == null) {\n      throw new Error(`Cannot find model at path '${path}'`);\n    }\n    const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n    this.LS.removeItem(keys.info);\n    this.LS.removeItem(keys.topology);\n    this.LS.removeItem(keys.weightSpecs);\n    this.LS.removeItem(keys.weightData);\n    return info;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n *   string.\n * - Moving or copying model from one path to another in the same medium or from\n *   one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n  // Singleton instance.\n  private static instance: ModelStoreManagerRegistry;\n\n  private managers: {[scheme: string]: ModelStoreManager};\n\n  private constructor() {\n    this.managers = {};\n  }\n\n  private static getInstance(): ModelStoreManagerRegistry {\n    if (ModelStoreManagerRegistry.instance == null) {\n      ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n    }\n    return ModelStoreManagerRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerManager(scheme: string, manager: ModelStoreManager) {\n    assert(scheme != null, () => 'scheme must not be undefined or null.');\n    if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n      scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n    }\n    assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n    const registry = ModelStoreManagerRegistry.getInstance();\n    assert(\n        registry.managers[scheme] == null,\n        () => `A model store manager is already registered for scheme '${\n            scheme}'.`);\n    registry.managers[scheme] = manager;\n  }\n\n  static getManager(scheme: string): ModelStoreManager {\n    const manager = this.getInstance().managers[scheme];\n    if (manager == null) {\n      throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n    }\n    return manager;\n  }\n\n  static getSchemes(): string[] {\n    return Object.keys(this.getInstance().managers);\n  }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n *   Scheme: e.g., 'localstorage' in the example above.\n *   Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n  if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n    throw new Error(\n        `The url string provided does not contain a scheme. ` +\n        `Supported schemes are: ` +\n        `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n  }\n  return {\n    scheme: url.split(URL_SCHEME_SUFFIX)[0],\n    path: url.split(URL_SCHEME_SUFFIX)[1],\n  };\n}\n\nasync function cloneModelInternal(\n    sourceURL: string, destURL: string,\n    deleteSource = false): Promise<ModelArtifactsInfo> {\n  assert(\n      sourceURL !== destURL,\n      () => `Old path and new path are the same: '${sourceURL}'`);\n\n  const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n  assert(\n      loadHandlers.length > 0,\n      () => `Copying failed because no load handler is found for source URL ${\n          sourceURL}.`);\n  assert(\n      loadHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `load handlers for source URL ${sourceURL}.`);\n  const loadHandler = loadHandlers[0];\n\n  const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n  assert(\n      saveHandlers.length > 0,\n      () => `Copying failed because no save handler is found for destination ` +\n          `URL ${destURL}.`);\n  assert(\n      saveHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `save handlers for destination URL ${destURL}.`);\n  const saveHandler = saveHandlers[0];\n\n  const sourceScheme = parseURL(sourceURL).scheme;\n  const sourcePath = parseURL(sourceURL).path;\n  const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n  const modelArtifacts = await loadHandler.load();\n\n  // If moving within the same storage medium, remove the old model as soon as\n  // the loading is done. Without doing this, it is possible that the combined\n  // size of the two models will cause the cloning to fail.\n  if (deleteSource && sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  const saveResult = await saveHandler.save(modelArtifacts);\n\n  // If moving between mediums, the deletion is done after the save succeeds.\n  // This guards against the case in which saving to the destination medium\n  // fails.\n  if (deleteSource && !sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n *   'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n  const schemes = ModelStoreManagerRegistry.getSchemes();\n  const out: {[url: string]: ModelArtifactsInfo} = {};\n  for (const scheme of schemes) {\n    const schemeOut =\n        await ModelStoreManagerRegistry.getManager(scheme).listModels();\n    for (const path in schemeOut) {\n      const url = scheme + URL_SCHEME_SUFFIX + path;\n      out[url] = schemeOut[path];\n    }\n  }\n  return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n *   'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n *   is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise<ModelArtifactsInfo> {\n  const schemeAndPath = parseURL(url);\n  const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n  return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function copyModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = false;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function moveModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = true;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n  // According to the spec, the built-in encoder can do only UTF-8 encoding.\n  // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n  private textEncoder: TextEncoder;\n\n  fetch(path: string, init?: RequestInit): Promise<Response> {\n    return fetch(path, init);\n  }\n\n  now(): number {\n    return performance.now();\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Browser's encoder only supports utf-8, but got ${encoding}`);\n    }\n    if (this.textEncoder == null) {\n      this.textEncoder = new TextEncoder();\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    return new TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_BROWSER')) {\n  env().setPlatform('browser', new PlatformBrowser());\n\n  // Register LocalStorage IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n  } catch (err) {\n  }\n\n  // Register IndexedDB IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n  } catch (err) {\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n  // tslint:disable-next-line:no-require-imports\n  importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise<Response>;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n  systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n  systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n  return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n  private textEncoder: TextEncoder;\n  // tslint:disable-next-line:no-any\n  util: any;\n\n  constructor() {\n    // tslint:disable-next-line:no-require-imports\n    this.util = require('util');\n    // According to the spec, the built-in encoder can do only UTF-8 encoding.\n    // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n    this.textEncoder = new this.util.TextEncoder();\n  }\n\n  fetch(path: string, requestInits?: RequestInit): Promise<Response> {\n    if (env().global.fetch != null) {\n      return env().global.fetch(path, requestInits);\n    }\n\n    if (systemFetch == null) {\n      systemFetch = getNodeFetch.importFetch();\n    }\n    return systemFetch(path, requestInits);\n  }\n\n  now(): number {\n    const time = process.hrtime();\n    return time[0] * 1000 + time[1] / 1000000;\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Node built-in encoder only supports utf-8, but got ${encoding}`);\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    if (bytes.length === 0) {\n      return '';\n    }\n    return new this.util.TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_NODE')) {\n  env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer<R extends Rank, D extends DataType = 'float32'>(\n    shape: ShapeMap[R], dtype: D = 'float32' as D,\n    values?: DataTypeMap[D]): TensorBuffer<R, D> {\n  dtype = dtype || 'float32' as D;\n  util.assertNonNegativeIntegerDimensions(shape);\n  return new TensorBuffer<R, D>(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_<T extends Tensor>(x: T|TensorLike, dtype: DataType): T {\n  const $x = convertToTensor(x, 'x', 'cast');\n\n  // Sanity checks.\n  if (!util.isValidDtype(dtype)) {\n    throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n  }\n  if (dtype === 'string' && $x.dtype !== 'string' ||\n      dtype !== 'string' && $x.dtype === 'string') {\n    throw new Error('Only strings can be casted to strings');\n  }\n\n  const inputs: CastInputs = {x: $x};\n  const attrs: CastAttrs = {dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n      null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'clone', null);\n  const forward = () =>\n      ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n  const inputs: IdentityInputs = {x: $x};\n\n  // Note this op is called tf.identity in python. Hence the kernel name used\n  // here.\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print<T extends Tensor>(x: T, verbose = false): void {\n  console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n  buffer,\n  cast,\n  clone,\n  print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer<T>(f: () => T): Promise<T> {\n  return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n  private readonly modelTopologyFileName: string;\n  private readonly weightDataFileName: string;\n  private readonly jsonAnchor: HTMLAnchorElement;\n  private readonly weightDataAnchor: HTMLAnchorElement;\n\n  static readonly URL_SCHEME = 'downloads://';\n\n  constructor(fileNamePrefix?: string) {\n    if (!env().getBool('IS_BROWSER')) {\n      // TODO(cais): Provide info on what IOHandlers are available under the\n      //   current environment.\n      throw new Error(\n          'browserDownloads() cannot proceed because the current environment ' +\n          'is not a browser.');\n    }\n\n    if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n      fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n    }\n    if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n      fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n    }\n\n    this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n    this.weightDataFileName =\n        fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (typeof (document) === 'undefined') {\n      throw new Error(\n          'Browser downloads are not supported in ' +\n          'this environment since `document` is not present');\n    }\n    const weightsURL = window.URL.createObjectURL(new Blob(\n        [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserDownloads.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const weightsManifest: WeightsManifestConfig = [{\n        paths: ['./' + this.weightDataFileName],\n        weights: modelArtifacts.weightSpecs\n      }];\n      const modelTopologyAndWeightManifest: ModelJSON = {\n        modelTopology: modelArtifacts.modelTopology,\n        format: modelArtifacts.format,\n        generatedBy: modelArtifacts.generatedBy,\n        convertedBy: modelArtifacts.convertedBy,\n        weightsManifest\n      };\n      const modelTopologyAndWeightManifestURL =\n          window.URL.createObjectURL(new Blob(\n              [JSON.stringify(modelTopologyAndWeightManifest)],\n              {type: 'application/json'}));\n\n      // If anchor elements are not provided, create them without attaching them\n      // to parents, so that the downloaded file names can be controlled.\n      const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n                                                   this.jsonAnchor;\n      jsonAnchor.download = this.modelTopologyFileName;\n      jsonAnchor.href = modelTopologyAndWeightManifestURL;\n      // Trigger downloads by evoking a click event on the download anchors.\n      // When multiple downloads are started synchronously, Firefox will only\n      // save the last one.\n      await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n      if (modelArtifacts.weightData != null) {\n        const weightDataAnchor = this.weightDataAnchor == null ?\n            document.createElement('a') :\n            this.weightDataAnchor;\n        weightDataAnchor.download = this.weightDataFileName;\n        weightDataAnchor.href = weightsURL;\n        await defer(\n            () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n      }\n\n      return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n    }\n  }\n}\n\nclass BrowserFiles implements IOHandler {\n  private readonly files: File[];\n\n  constructor(files: File[]) {\n    if (files == null || files.length < 1) {\n      throw new Error(\n          `When calling browserFiles, at least 1 file is required, ` +\n          `but received ${files}`);\n    }\n    this.files = files;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    const jsonFile = this.files[0];\n    const weightFiles = this.files.slice(1);\n\n    return new Promise<ModelArtifacts>((resolve, reject) => {\n      const jsonReader = new FileReader();\n      jsonReader.onload = (event: Event) => {\n        // tslint:disable-next-line:no-any\n        const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n        const modelTopology = modelJSON.modelTopology;\n        if (modelTopology == null) {\n          reject(new Error(\n              `modelTopology field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        if (weightFiles.length === 0) {\n          resolve({modelTopology});\n        }\n\n        const weightsManifest = modelJSON.weightsManifest;\n        if (weightsManifest == null) {\n          reject(new Error(\n              `weightManifest field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        let pathToFile: {[path: string]: File};\n        try {\n          pathToFile =\n              this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n        } catch (err) {\n          reject(err);\n          return;\n        }\n\n        const weightSpecs: WeightsManifestEntry[] = [];\n        const paths: string[] = [];\n        const perFileBuffers: ArrayBuffer[] = [];\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            paths.push(path);\n            perFileBuffers.push(null);\n          });\n          weightSpecs.push(...weightsGroup.weights);\n        });\n\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            const weightFileReader = new FileReader();\n            weightFileReader.onload = (event: Event) => {\n              // tslint:disable-next-line:no-any\n              const weightData = (event.target as any).result as ArrayBuffer;\n              const index = paths.indexOf(path);\n              perFileBuffers[index] = weightData;\n              if (perFileBuffers.indexOf(null) === -1) {\n                resolve({\n                  modelTopology,\n                  weightSpecs,\n                  weightData: concatenateArrayBuffers(perFileBuffers),\n                  format: modelJSON.format,\n                  generatedBy: modelJSON.generatedBy,\n                  convertedBy: modelJSON.convertedBy,\n                  userDefinedMetadata: modelJSON.userDefinedMetadata\n                });\n              }\n            };\n            weightFileReader.onerror = error =>\n                reject(`Failed to weights data from file of path '${path}'.`);\n            weightFileReader.readAsArrayBuffer(pathToFile[path]);\n          });\n        });\n      };\n      jsonReader.onerror = error => reject(\n          `Failed to read model topology and weights manifest JSON ` +\n          `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n          `Keras-style tf.Model artifacts only.`);\n      jsonReader.readAsText(jsonFile);\n    });\n  }\n\n  /**\n   * Check the compatibility between weights manifest and weight files.\n   */\n  private checkManifestAndWeightFiles(\n      manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n    const basenames: string[] = [];\n    const fileNames = files.map(file => basename(file.name));\n    const pathToFile: {[path: string]: File} = {};\n    for (const group of manifest) {\n      group.paths.forEach(path => {\n        const pathBasename = basename(path);\n        if (basenames.indexOf(pathBasename) !== -1) {\n          throw new Error(\n              `Duplicate file basename found in weights manifest: ` +\n              `'${pathBasename}'`);\n        }\n        basenames.push(pathBasename);\n        if (fileNames.indexOf(pathBasename) === -1) {\n          throw new Error(\n              `Weight file with basename '${pathBasename}' is not provided.`);\n        } else {\n          pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n        }\n      });\n    }\n\n    if (basenames.length !== files.length) {\n      throw new Error(\n          `Mismatch in the number of files in weights manifest ` +\n          `(${basenames.length}) and the number of weight files provided ` +\n          `(${files.length}).`);\n    }\n    return pathToFile;\n  }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n      return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * //   'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n *   `tf.Model`, `fileNamePrefix` should follow either of the following two\n *   formats:\n *   1. `null` or `undefined`, in which case the default file\n *      names will be used:\n *      - 'model.json' for the JSON file containing the model topology and\n *        weights manifest.\n *      - 'model.weights.bin' for the binary file containing the binary weight\n *        values.\n *   2. A single string or an Array of a single string, as the file name prefix.\n *      For example, if `'foo'` is provided, the downloaded JSON\n *      file and binary weights file will be named 'foo.json' and\n *      'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n  return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * //   elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (`<input type=\"file\" ...>`)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n *   which an `Array` of `File`s is expected (in that order):\n *   - A JSON file containing the model topology and weight manifest.\n *   - Optionally, One or more binary files containing the binary weights.\n *     These files must have names that match the paths in the `weightsManifest`\n *     contained by the aforementioned JSON file, or errors will be thrown\n *     during loading. These weights files have the same format as the ones\n *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n *     Python PIP package. If no weights files are provided, only the model\n *     topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n  return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n    promises: Array<Promise<{}|void>>, onProgress: OnProgressCallback,\n    startFraction?: number, endFraction?: number) {\n  checkPromises(promises);\n  startFraction = startFraction == null ? 0 : startFraction;\n  endFraction = endFraction == null ? 1 : endFraction;\n  checkFraction(startFraction, endFraction);\n  let resolvedPromise = 0;\n\n  const registerMonitor = (promise: Promise<{}>) => {\n    promise.then(value => {\n      const fraction = startFraction +\n          ++resolvedPromise / promises.length * (endFraction - startFraction);\n      // pass fraction as parameter to callback function.\n      onProgress(fraction);\n      return value;\n    });\n    return promise;\n  };\n\n  function checkPromises(promises: Array<Promise<{}|void>>): void {\n    assert(\n        promises != null && Array.isArray(promises) && promises.length > 0,\n        () => 'promises must be a none empty array');\n  }\n\n  function checkFraction(startFraction: number, endFraction: number): void {\n    assert(\n        startFraction >= 0 && startFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got startFraction ${startFraction}`);\n    assert(\n        endFraction >= 0 && endFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got endFraction ${endFraction}`);\n    assert(\n        endFraction >= startFraction,\n        () => `startFraction must be no more than endFraction, but ` +\n            `got startFraction ${startFraction} and endFraction ` +\n            `${endFraction}`);\n  }\n\n  return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n *   before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n *   length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n    fetchURLs: string[], loadOptions?: LoadOptions): Promise<ArrayBuffer[]> {\n  if (loadOptions == null) {\n    loadOptions = {};\n  }\n\n  const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n                                                    loadOptions.fetchFunc;\n\n  // Create the requests for all of the weights in parallel.\n  const requests = fetchURLs.map(\n      fetchURL =>\n          fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n  const fetchStartFraction = 0;\n  const fetchEndFraction = 0.5;\n\n  const responses = loadOptions.onProgress == null ?\n      await Promise.all(requests) :\n      await monitorPromisesProgress(\n          requests, loadOptions.onProgress, fetchStartFraction,\n          fetchEndFraction);\n\n  const bufferPromises = responses.map(response => response.arrayBuffer());\n\n  const bufferStartFraction = 0.5;\n  const bufferEndFraction = 1;\n\n  const buffers = loadOptions.onProgress == null ?\n      await Promise.all(bufferPromises) :\n      await monitorPromisesProgress(\n          bufferPromises, loadOptions.onProgress, bufferStartFraction,\n          bufferEndFraction);\n  return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n *     Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n    manifest: WeightsManifestConfig, filePathPrefix = '',\n    weightNames?: string[],\n    requestInit?: RequestInit): Promise<NamedTensorMap> {\n  // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n  // single weight from a group, the whole group will be fetched. At a future\n  // date, we should support fetching only the individual shards within a\n  // group that are needed to reconstruct the requested weight.\n  // TODO(cais): Use `decodeWeights` for implementation.\n\n  const fetchWeights = (fetchUrls: string[]) =>\n      loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n  const loadWeights = weightsLoaderFactory(fetchWeights);\n\n  return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n *   filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n *   fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n    fetchWeightsFunction: (fetchUrls: string[]) => Promise<ArrayBuffer[]>):\n    (manifest: WeightsManifestConfig, filePathPrefix?: string,\n     weightNames?: string[]) => Promise<NamedTensorMap> {\n  return async(\n             manifest: WeightsManifestConfig, filePathPrefix = '',\n             weightNames?: string[]): Promise<NamedTensorMap> => {\n    // Collect all the groups, weights, and their relative offsets to be\n    // fetched.\n    const groupIndicesToFetchMap = manifest.map(() => false);\n    const groupWeightsToFetch: {\n      [group: number]: Array<{\n        manifestEntry: WeightsManifestEntry; groupOffset: number;\n        sizeBytes: number;\n      }>\n    } = {};\n    const weightsFound =\n        weightNames != null ? weightNames.map(() => false) : [];\n    const allManifestWeightNames: string[] = [];\n    manifest.forEach((manifestGroupConfig, groupIndex) => {\n      let groupOffset = 0;\n      manifestGroupConfig.weights.forEach(weightsEntry => {\n        const rawDtype = ('quantization' in weightsEntry) ?\n            weightsEntry.quantization.dtype :\n            weightsEntry.dtype;\n\n        const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n            util.sizeFromShape(weightsEntry.shape);\n\n        const enqueueWeightsForFetchingFn = () => {\n          groupIndicesToFetchMap[groupIndex] = true;\n          if (groupWeightsToFetch[groupIndex] == null) {\n            groupWeightsToFetch[groupIndex] = [];\n          }\n\n          groupWeightsToFetch[groupIndex].push({\n            manifestEntry: weightsEntry,\n            groupOffset,\n            sizeBytes: weightsBytes\n          });\n        };\n\n        if (weightNames != null) {\n          weightNames.forEach((weightName, weightIndex) => {\n            if (weightName === weightsEntry.name) {\n              enqueueWeightsForFetchingFn();\n              weightsFound[weightIndex] = true;\n            }\n          });\n        } else {\n          enqueueWeightsForFetchingFn();\n        }\n\n        allManifestWeightNames.push(weightsEntry.name);\n        groupOffset += weightsBytes;\n      });\n    });\n\n    if (!weightsFound.every(found => found)) {\n      const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n      throw new Error(\n          `Could not find weights in manifest with names: ` +\n          `${weightsNotFound.join(', ')}. \\n` +\n          `Manifest JSON has weights with names: ` +\n          `${allManifestWeightNames.join(', ')}.`);\n    }\n\n    // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n    // IDs.\n    const groupIndicesToFetch =\n        groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n          if (shouldFetch) {\n            accumulator.push(i);\n          }\n          return accumulator;\n        }, []);\n\n    const fetchUrls: string[] = [];\n    groupIndicesToFetch.forEach(i => {\n      manifest[i].paths.forEach(filepath => {\n        const fetchUrl = filePathPrefix +\n            (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n        fetchUrls.push(fetchUrl);\n      });\n    });\n    const buffers = await fetchWeightsFunction(fetchUrls);\n\n    const weightsTensorMap: NamedTensorMap = {};\n    let bufferIndexOffset = 0;\n    groupIndicesToFetch.forEach(i => {\n      const numBuffers = manifest[i].paths.length;\n\n      let groupBytes = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        groupBytes += buffers[bufferIndexOffset + i].byteLength;\n      }\n\n      // Create a buffer for the whole group.\n      const groupBuffer = new ArrayBuffer(groupBytes);\n      const groupByteBuffer = new Uint8Array(groupBuffer);\n      let groupBufferOffset = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n        groupByteBuffer.set(buffer, groupBufferOffset);\n        groupBufferOffset += buffer.byteLength;\n      }\n\n      const weightsEntries = groupWeightsToFetch[i];\n      weightsEntries.forEach(weightsEntry => {\n        const byteBuffer = groupBuffer.slice(\n            weightsEntry.groupOffset,\n            weightsEntry.groupOffset + weightsEntry.sizeBytes);\n        const nameToTensorMap =\n            decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n        for (const name in nameToTensorMap) {\n          weightsTensorMap[name] = nameToTensorMap[name];\n        }\n      });\n\n      bufferIndexOffset += numBuffers;\n    });\n\n    return weightsTensorMap;\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n  protected readonly path: string;\n  protected readonly requestInit: RequestInit;\n\n  private readonly fetch: Function;\n  private readonly weightUrlConverter: (weightName: string) => Promise<string>;\n\n  readonly DEFAULT_METHOD = 'POST';\n\n  static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n  private readonly weightPathPrefix: string;\n  private readonly onProgress: OnProgressCallback;\n\n  constructor(path: string, loadOptions?: LoadOptions) {\n    if (loadOptions == null) {\n      loadOptions = {};\n    }\n    this.weightPathPrefix = loadOptions.weightPathPrefix;\n    this.onProgress = loadOptions.onProgress;\n    this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n    if (loadOptions.fetchFunc != null) {\n      assert(\n          typeof loadOptions.fetchFunc === 'function',\n          () => 'Must pass a function that matches the signature of ' +\n              '`fetch` (see ' +\n              'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n      this.fetch = loadOptions.fetchFunc;\n    } else {\n      this.fetch = env().platform.fetch;\n    }\n\n    assert(\n        path != null && path.length > 0,\n        () => 'URL path for http must not be null, undefined or ' +\n            'empty.');\n\n    if (Array.isArray(path)) {\n      assert(\n          path.length === 2,\n          () => 'URL paths for http must have a length of 2, ' +\n              `(actual length is ${path.length}).`);\n    }\n    this.path = path;\n\n    if (loadOptions.requestInit != null &&\n        loadOptions.requestInit.body != null) {\n      throw new Error(\n          'requestInit is expected to have no pre-existing body, but has one.');\n    }\n    this.requestInit = loadOptions.requestInit || {};\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserHTTPRequest.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n    init.body = new FormData();\n\n    const weightsManifest: WeightsManifestConfig = [{\n      paths: ['./model.weights.bin'],\n      weights: modelArtifacts.weightSpecs,\n    }];\n    const modelTopologyAndWeightManifest: ModelJSON = {\n      modelTopology: modelArtifacts.modelTopology,\n      format: modelArtifacts.format,\n      generatedBy: modelArtifacts.generatedBy,\n      convertedBy: modelArtifacts.convertedBy,\n      userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n      weightsManifest\n    };\n\n    init.body.append(\n        'model.json',\n        new Blob(\n            [JSON.stringify(modelTopologyAndWeightManifest)],\n            {type: JSON_TYPE}),\n        'model.json');\n\n    if (modelArtifacts.weightData != null) {\n      init.body.append(\n          'model.weights.bin',\n          new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n          'model.weights.bin');\n    }\n\n    const response = await this.fetch(this.path, init);\n\n    if (response.ok) {\n      return {\n        modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n        responses: [response],\n      };\n    } else {\n      throw new Error(\n          `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n          `${response.status}.`);\n    }\n  }\n\n  /**\n   * Load model artifacts via HTTP request(s).\n   *\n   * See the documentation to `tf.io.http` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model artifacts (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n    if (!modelConfigRequest.ok) {\n      throw new Error(\n          `Request to ${this.path} failed with status code ` +\n          `${modelConfigRequest.status}. Please verify this URL points to ` +\n          `the model JSON of the model to load.`);\n    }\n    let modelConfig: ModelJSON;\n    try {\n      modelConfig = await modelConfigRequest.json();\n    } catch (e) {\n      let message = `Failed to parse model JSON of response from ${this.path}.`;\n      // TODO(nsthorat): Remove this after some time when we're comfortable that\n      // .pb files are mostly gone.\n      if (this.path.endsWith('.pb')) {\n        message += ' Your path contains a .pb file extension. ' +\n            'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n            'in favor of .json models. You can re-convert your Python ' +\n            'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n            'or you can convert your.pb models with the \\'pb2json\\'' +\n            'NPM script in the tensorflow/tfjs-converter repository.';\n      } else {\n        message += ' Please make sure the server is serving valid ' +\n            'JSON for this request.';\n      }\n      throw new Error(message);\n    }\n    const modelTopology = modelConfig.modelTopology;\n    const weightsManifest = modelConfig.weightsManifest;\n    const generatedBy = modelConfig.generatedBy;\n    const convertedBy = modelConfig.convertedBy;\n    const format = modelConfig.format;\n    const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n    // We do not allow both modelTopology and weightsManifest to be missing.\n    if (modelTopology == null && weightsManifest == null) {\n      throw new Error(\n          `The JSON from HTTP path ${this.path} contains neither model ` +\n          `topology or manifest for weights.`);\n    }\n\n    let weightSpecs: WeightsManifestEntry[];\n    let weightData: ArrayBuffer;\n    if (weightsManifest != null) {\n      const results = await this.loadWeights(weightsManifest);\n      [weightSpecs, weightData] = results;\n    }\n\n    const artifacts: ModelArtifacts = {\n      modelTopology,\n      weightSpecs,\n      weightData,\n      userDefinedMetadata,\n      generatedBy,\n      convertedBy,\n      format\n    };\n\n    const initializer = modelConfig.modelInitializer;\n    if (initializer) {\n      artifacts.modelInitializer = initializer;\n    }\n\n    return artifacts;\n  }\n\n  private async loadWeights(weightsManifest: WeightsManifestConfig):\n      Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n    const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n    const [prefix, suffix] = parseUrl(weightPath);\n    const pathPrefix = this.weightPathPrefix || prefix;\n\n    const weightSpecs = [];\n    for (const entry of weightsManifest) {\n      weightSpecs.push(...entry.weights);\n    }\n\n    const fetchURLs: string[] = [];\n    const urlPromises: Array<Promise<string>> = [];\n    for (const weightsGroup of weightsManifest) {\n      for (const path of weightsGroup.paths) {\n        if (this.weightUrlConverter != null) {\n          urlPromises.push(this.weightUrlConverter(path));\n        } else {\n          fetchURLs.push(pathPrefix + path + suffix);\n        }\n      }\n    }\n\n    if (this.weightUrlConverter) {\n      fetchURLs.push(...await Promise.all(urlPromises));\n    }\n\n    const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n      requestInit: this.requestInit,\n      fetchFunc: this.fetch,\n      onProgress: this.onProgress\n    });\n    return [weightSpecs, concatenateArrayBuffers(buffers)];\n  }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n  const lastSlash = url.lastIndexOf('/');\n  const lastSearchParam = url.lastIndexOf('?');\n  const prefix = url.substring(0, lastSlash);\n  const suffix =\n      lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n  return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n  return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n    (url: string, loadOptions?: LoadOptions) => {\n      if (typeof fetch === 'undefined' &&\n          (loadOptions == null || loadOptions.fetchFunc == null)) {\n        // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n        // an environment that is not the browser or node they have to setup a\n        // global fetch polyfill.\n        return null;\n      } else {\n        let isHTTP = true;\n        if (Array.isArray(url)) {\n          isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n        } else {\n          isHTTP = isHTTPScheme(url);\n        }\n        if (isHTTP) {\n          return http(url, loadOptions);\n        }\n      }\n      return null;\n    };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n *   - A JSON file consisting of `modelTopology` and `weightsManifest`.\n *   - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n *     'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n *   Can be an absolute HTTP path (e.g.,\n *   'http://localhost:8000/model-upload)') or a relative path (e.g.,\n *   './model-upload').\n * @param requestInit Request configurations to be used when sending\n *    HTTP request to server using `fetch`. It can contain fields such as\n *    `method`, `credentials`, `headers`, `mode`, etc. See\n *    https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n *    for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n *   following fields:\n *   - weightPathPrefix Optional, this specifies the path prefix for weight\n *     files, by default this is calculated from the path param.\n *   - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n *     the `fetch` from node-fetch can be used here.\n *   - onProgress Optional, progress callback function, fired periodically\n *     before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n  return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n    path: string, loadOptions?: LoadOptions): IOHandler {\n  return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n  constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n  async load(): Promise<ModelArtifacts> {\n    return this.modelArtifacts;\n  }\n}\n\nclass PassthroughSaver implements IOHandler {\n  constructor(\n      private readonly saveHandler:\n          (artifacts: ModelArtifacts) => Promise<SaveResult>) {}\n\n  async save(modelArtifacts: ModelArtifacts) {\n    return this.saveHandler(modelArtifacts);\n  }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n *     modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n *   the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n *   names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n *   concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n    modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n    weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n  if (arguments.length === 1) {\n    const isModelArtifacts =\n        (modelArtifacts as ModelArtifacts).modelTopology != null ||\n        (modelArtifacts as ModelArtifacts).weightSpecs != null;\n    if (isModelArtifacts) {\n      return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n    } else {\n      // Legacy support: with only modelTopology.\n      // TODO(cais): Remove this deprecated API.\n      console.warn(\n          'Please call tf.io.fromMemory() with only one argument. ' +\n          'The argument should be of type ModelArtifacts. ' +\n          'The multi-argument signature of tf.io.fromMemory() has been ' +\n          'deprecated and will be removed in a future release.');\n      return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n    }\n  } else {\n    // Legacy support.\n    // TODO(cais): Remove this deprecated API.\n    console.warn(\n        'Please call tf.io.fromMemory() with only one argument. ' +\n        'The argument should be of type ModelArtifacts. ' +\n        'The multi-argument signature of tf.io.fromMemory() has been ' +\n        'deprecated and will be removed in a future release.');\n    return new PassthroughLoader({\n      modelTopology: modelArtifacts as {},\n      weightSpecs,\n      weightData,\n      trainingConfig\n    });\n  }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n *   // ... do something with the artifacts ...\n *   return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n *     `SaveResult`.\n */\nexport function withSaveHandler(\n    saveHandler: (artifacts: ModelArtifacts) =>\n        Promise<SaveResult>): IOHandler {\n  return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n  browserFiles,\n  browserHTTPRequest,\n  concatenateArrayBuffers,\n  decodeWeights,\n  encodeWeights,\n  fromMemory,\n  getLoadHandlers,\n  getModelArtifactsInfoForJSON,\n  getSaveHandlers,\n  http,\n  IOHandler,\n  isHTTPScheme,\n  LoadHandler,\n  LoadOptions,\n  loadWeights,\n  ModelArtifacts,\n  ModelArtifactsInfo,\n  ModelJSON,\n  ModelStoreManager,\n  OnProgressCallback,\n  registerLoadRouter,\n  registerSaveRouter,\n  RequestDetails,\n  SaveConfig,\n  SaveHandler,\n  SaveResult,\n  WeightGroup,\n  weightsLoaderFactory,\n  WeightsManifestConfig,\n  WeightsManifestEntry,\n  withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  const $x = convertToTensor(x, 'x', 'reshape', null);\n\n  const inputs: ReshapeInputs = {x: $x};\n  const attrs: ReshapeAttrs = {shape};\n  const forward: ForwardFunc<\n      Tensor<R>> = (backend: KernelBackend, save: GradSaveFunc) => {\n    shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n    util.assert(\n        $x.size === util.sizeFromShape(shape),\n        () => 'new shape and old shape must have the same number of elements.');\n    save([$x]);\n    return backend.reshape($x, shape);\n  };\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n      attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print();  // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike, transposeA = false,\n    transposeB = false): T {\n  let $a = convertToTensor(a, 'a', 'matMul');\n  let $b = convertToTensor(b, 'b', 'matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n          `got ranks ${$a.rank} and ${$b.rank}.`);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n                           reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n                           reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([a3D, b3D]);\n\n    return backend.batchMatMul(\n        a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n  };\n\n  const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n  const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n      attrs as {} as NamedAttrMap);\n\n  return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n *     not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n    indices: Tensor|TensorLike, depth: number, onValue = 1,\n    offValue = 0): Tensor {\n  if (depth < 2) {\n    throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n  }\n  const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n  const outShape = [...$indices.shape, depth];\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$indices]);\n    return reshape(\n        backend.oneHot(\n            reshape($indices, [$indices.size]), depth, onValue, offValue),\n        outShape);\n  };\n\n  const inputs: OneHotInputs = {indices: $indices};\n  const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print();  // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_<T extends Tensor>(x: T|TensorLike, perm?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'transpose');\n\n  if (perm == null) {\n    perm = $x.shape.map((s, i) => i).reverse();\n  }\n  util.assert(\n      $x.rank === perm.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of perm ${perm}.`);\n  perm.forEach(axis => {\n    util.assert(\n        axis >= 0 && axis < $x.rank,\n        () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n            ` but got ${perm}`);\n  });\n\n  if ($x.rank <= 1) {\n    return $x.clone();\n  }\n\n  const inputs: TransposeInputs = {x: $x};\n  const attrs: TransposeAttrs = {perm};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n      null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * //  [0, 1, 1],\n * //  [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n *   for the classes. The shape is `[numExamples]`, where\n *   `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n *   0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n *   Its value must be larger than the largest element in `labels` and\n *   `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n *   row `r` and column `c` is the number of times examples of actual class\n *   `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n    labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n    numClasses: number): Tensor2D {\n  const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n  util.assert(\n      numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n      () => `If provided, numClasses must be a positive integer, ` +\n          `but got ${numClasses}`);\n  util.assert(\n      $labels.rank === 1,\n      () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n  util.assert(\n      $predictions.rank === 1,\n      () => `Expected the rank of predictions to be 1, ` +\n          `but got ${$predictions.rank}`);\n  util.assert(\n      $labels.shape[0] === $predictions.shape[0],\n      () => `Mismatch in the number of examples: ` +\n          `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n          `Labels and predictions should have the same number of elements.`);\n  util.assert(\n      numClasses > 0 && Number.isInteger(numClasses),\n      () => `numClasses is required to be a positive integer, but got ` +\n          `${numClasses}`);\n  // TODO(cais): In the future, if oneHot supports tensors inputs for\n  //   `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n  const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n  const oneHotPredictions =\n      oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n  const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n  return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided,  it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n    values: TensorLike3D, shape?: [number, number, number],\n    dtype?: DataType): Tensor3D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 3) {\n    throw new Error('tensor3d() requires shape to have three numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor3d() requires values to be number[][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor3d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement,\n    numChannels = 3): Tensor3D {\n  // Sanity checks.\n  if (numChannels > 4) {\n    throw new Error(\n        'Cannot construct Tensor with more than 4 channels from pixels.');\n  }\n  if (pixels == null) {\n    throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n  }\n  let isPixelData = false;\n  let isImageData = false;\n  let isVideo = false;\n  let isImage = false;\n  let isCanvasLike = false;\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    isPixelData = true;\n  } else if (\n      typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n    isImageData = true;\n  } else if (\n      typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement) {\n    isVideo = true;\n  } else if (\n      typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement) {\n    isImage = true;\n    // tslint:disable-next-line: no-any\n  } else if ((pixels as any).getContext != null) {\n    isCanvasLike = true;\n  } else {\n    throw new Error(\n        'pixels passed to tf.browser.fromPixels() must be either an ' +\n        `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n        `in browser, or OffscreenCanvas, ImageData in webworker` +\n        ` or {data: Uint32Array, width: number, height: number}, ` +\n        `but was ${(pixels as {}).constructor.name}`);\n  }\n  if (isVideo) {\n    const HAVE_CURRENT_DATA_READY_STATE = 2;\n    if (isVideo &&\n        (pixels as HTMLVideoElement).readyState <\n            HAVE_CURRENT_DATA_READY_STATE) {\n      throw new Error(\n          'The video element has not loaded data yet. Please wait for ' +\n          '`loadeddata` event on the <video> element.');\n    }\n  }\n  // If the current backend has 'FromPixels' registered, it has a more\n  // efficient way of handling pixel uploads, so we call that.\n  const kernel = getKernel(FromPixels, ENGINE.backendName);\n  if (kernel != null) {\n    const inputs: FromPixelsInputs = {pixels};\n    const attrs: FromPixelsAttrs = {numChannels};\n    return ENGINE.runKernel(\n               FromPixels, inputs as {} as NamedTensorMap,\n               attrs as {} as NamedAttrMap) as Tensor3D;\n  }\n\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n  let vals: Uint8ClampedArray|Uint8Array;\n\n  if (isCanvasLike) {\n    vals =\n        // tslint:disable-next-line:no-any\n        (pixels as any).getContext('2d').getImageData(0, 0, width, height).data;\n  } else if (isImageData || isPixelData) {\n    vals = (pixels as PixelData | ImageData).data;\n  } else if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement, 0, 0, width, height);\n    vals = fromPixels2DContext.getImageData(0, 0, width, height).data;\n  }\n  let values: Int32Array;\n  if (numChannels === 4) {\n    values = new Int32Array(vals);\n  } else {\n    const numPixels = width * height;\n    values = new Int32Array(numPixels * numChannels);\n    for (let i = 0; i < numPixels; i++) {\n      for (let channel = 0; channel < numChannels; ++channel) {\n        values[i * numChannels + channel] = vals[i * 4 + channel];\n      }\n    }\n  }\n  const outShape: [number, number, number] = [height, width, numChannels];\n  return tensor3d(values, outShape, 'int32');\n}\n\n/**\n * Draws a `tf.Tensor` of pixel values to a byte array or optionally a\n * canvas.\n *\n * When the dtype of the input is 'float32', we assume values in the range\n * [0-1]. Otherwise, when input is 'int32', we assume values in the range\n * [0-255].\n *\n * Returns a promise that resolves when the canvas has been drawn to.\n *\n * @param img A rank-2 or rank-3 tensor. If rank-2, draws grayscale. If\n *     rank-3, must have depth of 1, 3 or 4. When depth of 1, draws\n * grayscale. When depth of 3, we draw with the first three components of\n * the depth dimension corresponding to r, g, b and alpha = 1. When depth of\n * 4, all four components of the depth dimension correspond to r, g, b, a.\n * @param canvas The canvas to draw to.\n *\n * @doc {heading: 'Browser', namespace: 'browser'}\n */\nexport async function toPixels(\n    img: Tensor2D|Tensor3D|TensorLike,\n    canvas?: HTMLCanvasElement): Promise<Uint8ClampedArray> {\n  let $img = convertToTensor(img, 'img', 'toPixels');\n  if (!(img instanceof Tensor)) {\n    // Assume int32 if user passed a native array.\n    const originalImgTensor = $img;\n    $img = cast(originalImgTensor, 'int32');\n    originalImgTensor.dispose();\n  }\n  if ($img.rank !== 2 && $img.rank !== 3) {\n    throw new Error(\n        `toPixels only supports rank 2 or 3 tensors, got rank ${$img.rank}.`);\n  }\n  const [height, width] = $img.shape.slice(0, 2);\n  const depth = $img.rank === 2 ? 1 : $img.shape[2];\n\n  if (depth > 4 || depth === 2) {\n    throw new Error(\n        `toPixels only supports depth of size ` +\n        `1, 3 or 4 but got ${depth}`);\n  }\n\n  if ($img.dtype !== 'float32' && $img.dtype !== 'int32') {\n    throw new Error(\n        `Unsupported type for toPixels: ${$img.dtype}.` +\n        ` Please use float32 or int32 tensors.`);\n  }\n\n  const data = await $img.data();\n  const multiplier = $img.dtype === 'float32' ? 255 : 1;\n  const bytes = new Uint8ClampedArray(width * height * 4);\n\n  for (let i = 0; i < height * width; ++i) {\n    const rgba = [0, 0, 0, 255];\n\n    for (let d = 0; d < depth; d++) {\n      const value = data[i * depth + d];\n\n      if ($img.dtype === 'float32') {\n        if (value < 0 || value > 1) {\n          throw new Error(\n              `Tensor values for a float32 Tensor must be in the ` +\n              `range [0 - 1] but encountered ${value}.`);\n        }\n      } else if ($img.dtype === 'int32') {\n        if (value < 0 || value > 255) {\n          throw new Error(\n              `Tensor values for a int32 Tensor must be in the ` +\n              `range [0 - 255] but encountered ${value}.`);\n        }\n      }\n\n      if (depth === 1) {\n        rgba[0] = value * multiplier;\n        rgba[1] = value * multiplier;\n        rgba[2] = value * multiplier;\n      } else {\n        rgba[d] = value * multiplier;\n      }\n    }\n\n    const j = i * 4;\n    bytes[j + 0] = Math.round(rgba[0]);\n    bytes[j + 1] = Math.round(rgba[1]);\n    bytes[j + 2] = Math.round(rgba[2]);\n    bytes[j + 3] = Math.round(rgba[3]);\n  }\n\n  if (canvas != null) {\n    canvas.width = width;\n    canvas.height = height;\n    const ctx = canvas.getContext('2d');\n    const imageData = new ImageData(bytes, width, height);\n    ctx.putImageData(imageData, 0, 0);\n  }\n  if ($img !== img) {\n    $img.dispose();\n  }\n  return bytes;\n}\n\nexport const fromPixels = op({fromPixels_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\nimport {computeStrides} from '../util';\n\n/**\n * Validate gather nd inputs.\n *\n * @param tensor The tensor contains the source values.\n * @param indices The tensor contains the indices to slice the source.\n *\n * @returns [resultShape, numUpdates, sliceSize, strides]\n */\nexport function prepareAndValidate(\n    tensor: Tensor, indices: Tensor): [number[], number, number, number[]] {\n  if (tensor.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the input to be rank 1 or higher,' +\n        ` but the rank was ${tensor.rank}.`);\n  }\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(\n        'tf.gatherND() expects the indices to be int32 type,' +\n        ` but the dtype was ${indices.dtype}.`);\n  }\n  if (indices.shape[indices.rank - 1] > tensor.rank) {\n    throw new Error(\n        'index innermost dimension length must be <= tensor rank; saw: ' +\n        `${indices.shape[indices.rank - 1]} vs. ${tensor.rank}`);\n  }\n\n  if (tensor.size === 0) {\n    throw new Error(\n        'Requested more than 0 entries, but input is empty.' +\n        ` Input shape: ${tensor.shape}.`);\n  }\n\n  const indicesShape = indices.shape;\n  const sliceRank = indicesShape[indicesShape.length - 1];\n\n  // The result shape is\n  //   indices.shape[:-1] + params.shape[indices.shape[-1]:]\n  let nResult = 1;\n  for (let i = 0; i < indicesShape.length - 1; ++i) {\n    nResult *= indicesShape[i];\n  }\n\n  const inputShape = tensor.shape;\n\n  const resultShape = indicesShape.slice();\n  resultShape.pop();\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < tensor.rank; ++i) {\n    sliceSize *= inputShape[i];\n    resultShape.push(inputShape[i]);\n  }\n\n  const strides =\n      [...computeStrides(tensor.shape).map(stride => stride / sliceSize),\n       1].slice(0, sliceRank);\n\n  return [resultShape, nResult, sliceSize, strides];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {computeStrides, sizeFromShape} from '../util';\n\n/**\n * Check whether updates.shape = indices.shape[:batchDim] +\n * shape[sliceDim:]\n *\n * @param x The input tensor.\n */\nexport function validateUpdateShape(\n    shape: number[], indices: Tensor, updates: Tensor) {\n  const sliceDim = (indices.rank > 1) ? indices.shape[indices.rank - 1] : 1;\n  const batchDim = (indices.rank > 1) ? indices.rank - 1 : 1;\n\n  const shapeError = 'Must have updates.shape = indices.shape[:batchDim] + ' +\n      `shape[sliceDim:], got updates.shape: ${updates.shape}` +\n      `, indices.shape: ${indices.shape}, shape: ${shape}` +\n      `, sliceDim: ${sliceDim}, and batchDim: ${batchDim}.`;\n\n  if (updates.rank < batchDim) {\n    throw new Error(shapeError + ` update.rank < ${batchDim}. `);\n  }\n  if (shape.length < sliceDim + (updates.rank - batchDim)) {\n    throw new Error(\n        shapeError +\n        ` Output shape length < ${sliceDim + (updates.rank - batchDim)}`);\n  }\n  if (updates.rank !== batchDim + shape.length - sliceDim) {\n    throw new Error(\n        shapeError + ` update.rank != ${batchDim + shape.length - sliceDim}`);\n  }\n  for (let d = 0; d < batchDim; ++d) {\n    if (updates.shape[d] !== indices.shape[d]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d}] (${updates.shape[d]}) != indices.shape[${d}] (${\n              indices.shape[d]}).`);\n    }\n  }\n  for (let d = 0; d < updates.rank - batchDim; ++d) {\n    if (updates.shape[d + batchDim] !== shape[d + sliceDim]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d + batchDim}] (${\n              updates.shape[d + batchDim]}) != shape[${d + batchDim}] (${\n              shape[d + batchDim]})`);\n    }\n  }\n}\n\nexport interface ScatterShapeInfo {\n  sliceRank: number;\n  numUpdates: number;\n  sliceSize: number;\n  strides: number[];\n  outputSize: number;\n}\n/**\n * Validate scatter nd inputs.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n */\nexport function validateInput(\n    updates: Tensor, indices: Tensor, shape: number[]) {\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (updates.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the updates to be rank 1 or higher,' +\n        ` but the rank was ${updates.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${\n        indices.dtype}`);\n  }\n  if (shape.length < 1) {\n    throw new Error(\n        `Output rank must be greater or equal to 1, but got shape: ${shape}`);\n  }\n\n  if (shape.length === 0) {\n    if (indices.size === 0) {\n      throw new Error(`Indices specified for empty output. indices shape: ${\n          indices.shape}`);\n    }\n    if (updates.size === 0) {\n      throw new Error(`Updates specified for empty output. updates shape: ${\n          updates.shape}`);\n    }\n  }\n\n  validateUpdateShape(shape, indices, updates);\n}\n\n/**\n * Calculate the shape information for the output.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n *\n * @returns ScatterShapeInfo\n */\nexport function calculateShapes(\n    updates: TensorInfo, indices: TensorInfo,\n    shape: number[]): ScatterShapeInfo {\n  // Calculate the number of dimensions in indices\n  const indicesRank = indices.shape.length;\n  const sliceRank = (indicesRank > 1) ? indices.shape[indicesRank - 1] : 1;\n\n  // Calculate the number of elements that make up each slice of our updated\n  // tensor. This allows us to work with flattened tensors and copy over whole\n  // slices at a time.\n  const totalNd = shape.length;\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < totalNd; ++i) {\n    sliceSize *= shape[i];\n  }\n\n  const safeSliceDim = (sliceRank < 1) ? 1 : sliceRank;\n  const numUpdates = sizeFromShape(indices.shape) / safeSliceDim;\n\n  const strides = [...computeStrides(shape.slice(0, sliceRank)), 1];\n  const outputSize = sizeFromShape(shape);\n  return {sliceRank, numUpdates, sliceSize, strides, outputSize};\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '../kernel_registry';\nimport * as util from '../util';\n\nexport function assertParamsValid(\n    input: TensorInfo, begin: number[], size: number[]): void {\n  const inputRank = input.shape.length;\n  util.assert(\n      inputRank === begin.length,\n      () => `Error in slice${inputRank}D: Length of begin ${begin} must ` +\n          `match the rank of the array (${inputRank}).`);\n  util.assert(\n      inputRank === size.length,\n      () => `Error in slice${inputRank}D: Length of size ${size} must ` +\n          `match the rank of the array (${inputRank}).`);\n\n  for (let i = 0; i < inputRank; ++i) {\n    util.assert(\n        begin[i] + size[i] <= input.shape[i],\n        () => `Error in slice${inputRank}D: begin[${i}] + size[${i}] ` +\n            `(${begin[i] + size[i]}) would overflow input.shape[${i}] (${\n                  input.shape[i]})`);\n  }\n}\n\n/** Converts a binary mask to an array of axes. Used in stridedSlice(). */\nexport function maskToAxes(mask: number): number[] {\n  const axes = [];\n  let axis = 0;\n  while (mask > 0) {\n    if (mask & 1) {\n      axes.push(axis);\n    }\n    mask /= 2;\n    axis++;\n  }\n  return axes;\n}\n\n/** Computes the output shape given the strided slice params. */\nexport function computeOutShape(\n    begin: number[], end: number[], strides: number[]): number[] {\n  const size = [];\n  for (let axis = 0; axis < begin.length; axis++) {\n    size[axis] = Math.ceil((end[axis] - begin[axis]) / strides[axis]);\n  }\n  return size;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stride value. Otherwise, insert.\nexport function stridesWithElidedDims(\n    strides: number[], ellipsisInsertionIndex: number, numElidedAxes: number,\n    inputShape: number[]): number[] {\n  const newStrides = [...strides];\n  for (let i = newStrides.length; i < inputShape.length; i++) {\n    newStrides.push(1);\n  }\n  for (let i = 0; i < numElidedAxes; i++) {\n    if (i === 0) {\n      newStrides[ellipsisInsertionIndex] = 1;\n    } else {\n      newStrides.splice(\n          ellipsisInsertionIndex, 0 /* num elements to delete */,\n          1 /* element to add */);\n      newStrides.pop();\n    }\n  }\n  return newStrides;\n}\n\nfunction unnormalizeAxis(\n    ellipsisInsertionIndex: number, numElidedAxes: number,\n    normalizedAxis: number): number {\n  if (normalizedAxis <= ellipsisInsertionIndex) {\n    return normalizedAxis;\n  }\n\n  return normalizedAxis - (numElidedAxes - 1);\n}\n\nfunction getElidedAxes(numElidedAxes: number, ellipsisInsertionIndex: number) {\n  const elidedAxes = [];\n  for (let i = 0; i < numElidedAxes; i++) {\n    elidedAxes.push(ellipsisInsertionIndex + i);\n  }\n  return elidedAxes;\n}\n\n// Normalize the start, end and strides.\nexport function getNormalizedAxes(\n    inputShape: number[], ellipsisAxes: number[], numInterpolatedAxes: number,\n    begin: number[], end: number[], strides: number[], beginMask: number,\n    endMask: number,\n    ellipsisMask: number): {begin: number[], end: number[], strides: number[]} {\n  const inputRank = inputShape.length;\n  let normalizedBegin = new Array(inputRank),\n      normalizedEnd = new Array(inputRank),\n      normalizedStrides = new Array(inputRank);\n  if (ellipsisAxes.length && numInterpolatedAxes > 0) {\n    const fullIndex = ellipsisAxes[0];\n\n    // The ellipsis applies to the masked index as well as any dimensions\n    // that are interpolated.\n    const numElidedAxes = numInterpolatedAxes + 1;\n    normalizedBegin = startIndicesWithElidedDims(\n        beginMask, fullIndex, numElidedAxes, begin, inputShape);\n    normalizedEnd = stopIndicesWithElidedDims(\n        endMask, fullIndex, numElidedAxes, end, inputShape);\n    normalizedStrides =\n        stridesWithElidedDims(strides, fullIndex, numElidedAxes, inputShape);\n  } else {\n    for (let axis = 0; axis < inputRank; axis++) {\n      normalizedBegin[axis] = startForAxis(\n          beginMask, begin, strides, inputShape, axis, ellipsisMask);\n      normalizedEnd[axis] =\n          stopForAxis(endMask, end, strides, inputShape, axis, ellipsisMask);\n      normalizedStrides[axis] = stridesForAxis(strides, axis, ellipsisMask);\n    }\n  }\n\n  return {\n    begin: normalizedBegin,\n    end: normalizedEnd,\n    strides: normalizedStrides\n  };\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current start value. Otherwise, insert.\nexport function startIndicesWithElidedDims(\n    beginMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalBegin: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = 0;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalBegin[originalAxis];\n      if (beginMask & 1 << originalAxis) {\n        originalValue = 0;\n      }\n\n      newIndices[axis] = originalValue;\n    }\n  }\n  return newIndices;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stop value. Otherwise, insert.\nexport function stopIndicesWithElidedDims(\n    endMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalEnd: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = Number.MAX_SAFE_INTEGER;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalEnd[originalAxis];\n      if (endMask & 1 << originalAxis) {\n        originalValue = Number.MAX_SAFE_INTEGER;\n      }\n      newIndices[axis] = originalValue;\n    }\n  }\n\n  for (let i = 0; i < newIndices.length; i++) {\n    // Handle negative indices\n    const axisSize = inputShape[i];\n    if (newIndices[i] < 0) {\n      newIndices[i] += axisSize;\n    }\n    newIndices[i] = util.clamp(0, newIndices[i], inputShape[i]);\n  }\n  return newIndices;\n}\n\nexport function stridesForAxis(\n    strides: number[], axis: number, ellipsisMask: number): number {\n  let stride = strides[axis];\n  if (ellipsisMask & (1 << axis) || stride == null) {\n    stride = 1;\n  }\n\n  return stride;\n}\n\nexport function startForAxis(\n    beginMask: number, startIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let start = startIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or the begin index is not set\n  // for the axis.\n  if (beginMask & 1 << axis || ellipsisMask & 1 << axis || start == null) {\n    if (stride > 0) {\n      // Forward iteration - use the first element. These values will get\n      // clamped below (Note: We could have set them to 0 and axis_size-1, but\n      // use lowest() and max() to maintain symmetry with StopForAxis())\n      start = Number.MIN_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the last element.\n      start = Number.MAX_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (start < 0) {\n    start += axisSize;\n  }\n\n  // Clamping\n  start = util.clamp(0, start, axisSize - 1);\n\n  return start;\n}\n\nexport function stopForAxis(\n    endMask: number, stopIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let stop = stopIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or if the stop index is not\n  // set for this axis.\n  if (endMask & (1 << axis) || ellipsisMask & (1 << axis) || stop == null) {\n    if (stride > 0) {\n      // Forward iteration - use the last element. These values will get\n      // clamped below\n      stop = Number.MAX_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the first element.\n      stop = Number.MIN_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (stop < 0) {\n    stop += axisSize;\n  }\n\n  // Clamping\n  // Because the end index points one past the last element, we need slightly\n  // different clamping ranges depending on the direction.\n  if (stride > 0) {\n    // Forward iteration\n    stop = util.clamp(0, stop, axisSize);\n  } else {\n    // Backward iteration\n    stop = util.clamp(-1, stop, axisSize - 1);\n  }\n\n  return stop;\n}\n\n/**\n * Returns true if the slice occupies a continous set of elements in the\n * 'flat' space.\n */\nexport function isSliceContinous(\n    shape: number[], begin: number[], size: number[]) {\n  // Index of the first axis that has size > 1.\n  let firstNonOneAxis = size.length;\n  for (let i = 0; i < size.length; i++) {\n    if (size[i] > 1) {\n      firstNonOneAxis = i;\n      break;\n    }\n  }\n\n  for (let i = firstNonOneAxis + 1; i < size.length; i++) {\n    if (begin[i] > 0 || size[i] !== shape[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function computeFlatOffset(begin: number[], strides: number[]): number {\n  let flatOffset = begin.length > 0 ? begin[begin.length - 1] : 1;\n  for (let i = 0; i < begin.length - 1; i++) {\n    flatOffset += begin[i] * strides[i];\n  }\n  return flatOffset;\n}\n\nexport function parseSliceParams(\n    x: TensorInfo, begin: number|number[], size?: number|number[]) {\n  // The following logic allows for more ergonomic calls.\n  let begin_: number[];\n  const xRank = x.shape.length;\n  if (typeof begin === 'number') {\n    begin_ = [begin, ...new Array(xRank - 1).fill(0)];\n  } else if (begin.length < xRank) {\n    begin_ = begin.concat(new Array(xRank - begin.length).fill(0));\n  } else {\n    begin_ = begin.slice();\n  }\n  begin_.forEach(d => {\n    util.assert(\n        d !== -1, () => 'slice() does not support negative begin indexing.');\n  });\n  let size_: number[];\n  if (size == null) {\n    size_ = new Array(xRank).fill(-1);\n  } else if (typeof size === 'number') {\n    size_ = [size, ...new Array(xRank - 1).fill(-1)];\n  } else if (size.length < xRank) {\n    size_ = size.concat(new Array(xRank - size.length).fill(-1));\n  } else {\n    size_ = size;\n  }\n  size_ = size_.map((d, i) => {\n    if (d >= 0) {\n      return d;\n    } else {\n      util.assert(\n          d === -1,\n          () => `Negative size values should be exactly -1 but got ` +\n              `${d} for the slice() size at index ${i}.`);\n      return x.shape[i] - begin_[i];\n    }\n  });\n  return [begin_, size_];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from './util';\n\n/**\n * Types to support JSON-esque data structures internally.\n *\n * Internally ConfigDict's use camelCase keys and values where the\n * values are class names to be instantiated.  On the python side, these\n * will be snake_case.  Internally we allow Enums into the values for better\n * type safety, but these need to be converted to raw primitives (usually\n * strings) for round-tripping with python.\n *\n * toConfig returns the TS-friendly representation. model.toJSON() returns\n * the pythonic version as that's the portable format.  If you need to\n * python-ify a non-model level toConfig output, you'll need to use a\n * convertTsToPythonic from serialization_utils in -Layers.\n *\n */\nexport declare type ConfigDictValue =\n    boolean | number | string | null | ConfigDictArray | ConfigDict;\nexport declare interface ConfigDict {\n  [key: string]: ConfigDictValue;\n}\nexport declare interface ConfigDictArray extends Array<ConfigDictValue> {}\n\n/**\n * Type to represent the class-type of Serializable objects.\n *\n * Ie the class prototype with access to the constructor and any\n * static members/methods. Instance methods are not listed here.\n *\n * Source for this idea: https://stackoverflow.com/a/43607255\n */\nexport declare type SerializableConstructor<T extends Serializable> = {\n  // tslint:disable-next-line:no-any\n  new (...args: any[]): T; className: string; fromConfig: FromConfigMethod<T>;\n};\nexport declare type FromConfigMethod<T extends Serializable> =\n    (cls: SerializableConstructor<T>, config: ConfigDict) => T;\n\n/**\n * Serializable defines the serialization contract.\n *\n * TFJS requires serializable classes to return their className when asked\n * to avoid issues with minification.\n */\nexport abstract class Serializable {\n  /**\n   * Return the class name for this class to use in serialization contexts.\n   *\n   * Generally speaking this will be the same thing that constructor.name\n   * would have returned.  However, the class name needs to be robust\n   * against minification for serialization/deserialization to work properly.\n   *\n   * There's also places such as initializers.VarianceScaling, where\n   * implementation details between different languages led to different\n   * class hierarchies and a non-leaf node is used for serialization purposes.\n   */\n  getClassName(): string {\n    return (this.constructor as SerializableConstructor<Serializable>)\n        .className;\n  }\n\n  /**\n   * Return all the non-weight state needed to serialize this object.\n   */\n  abstract getConfig(): ConfigDict;\n\n  /**\n   * Creates an instance of T from a ConfigDict.\n   *\n   * This works for most descendants of serializable.  A few need to\n   * provide special handling.\n   * @param cls A Constructor for the class to instantiate.\n   * @param config The Configuration for the object.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config);\n  }\n}\n\n/**\n * Maps string keys to class constructors.\n *\n * Used during (de)serialization from the cross-language JSON format, which\n * requires the class name in the serialization format matches the class\n * names as used in Python, should it exist.\n */\nexport class SerializationMap {\n  private static instance: SerializationMap;\n  classNameMap: {\n    [className: string]:\n        [SerializableConstructor<Serializable>, FromConfigMethod<Serializable>]\n  };\n\n  private constructor() {\n    this.classNameMap = {};\n  }\n\n  /**\n   * Returns the singleton instance of the map.\n   */\n  static getMap(): SerializationMap {\n    if (SerializationMap.instance == null) {\n      SerializationMap.instance = new SerializationMap();\n    }\n    return SerializationMap.instance;\n  }\n\n  /**\n   * Registers the class as serializable.\n   */\n  static register<T extends Serializable>(cls: SerializableConstructor<T>) {\n    SerializationMap.getMap().classNameMap[cls.className] =\n        [cls, cls.fromConfig];\n  }\n}\n\n/**\n * Register a class with the serialization map of TensorFlow.js.\n *\n * This is often used for registering custom Layers, so they can be\n * serialized and deserialized.\n *\n * Example:\n *\n * ```js\n * class MyCustomLayer extends tf.layers.Layer {\n *   static className = 'MyCustomLayer';\n *\n *   constructor(config) {\n *     super(config);\n *   }\n * }\n * tf.serialization.registerClass(MyCustomLayer);\n * ```\n *\n * @param cls The class to be registered. It must have a public static member\n *   called `className` defined and the value must be a non-empty string.\n *\n * @doc {heading: 'Models', subheading: 'Serialization', ignoreCI: true}\n */\nexport function registerClass<T extends Serializable>(\n    cls: SerializableConstructor<T>) {\n  assert(\n      cls.className != null,\n      () => `Class being registered does not have the static className ` +\n          `property defined.`);\n  assert(\n      typeof cls.className === 'string',\n      () => `className is required to be a string, but got type ` +\n          typeof cls.className);\n  assert(\n      cls.className.length > 0,\n      () => `Class being registered has an empty-string as its className, ` +\n          `which is disallowed.`);\n\n  SerializationMap.register(cls);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {inferShape} from './tensor_util_env';\nimport {RecursiveArray, TensorLike, TypedArray} from './types';\nimport {arraysEqual, flatten, isString, isTypedArray} from './util';\n\nconst TEST_EPSILON_FLOAT32 = 1e-3;\nexport const TEST_EPSILON_FLOAT16 = 1e-1;\n\nexport function expectArraysClose(\n    actual: TypedArray|number|RecursiveArray<number>,\n    expected: TypedArray|number|RecursiveArray<number>, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, epsilon));\n}\n\nexport function testEpsilon() {\n  return ENGINE.backend.floatPrecision() === 32 ? TEST_EPSILON_FLOAT32 :\n                                                  TEST_EPSILON_FLOAT16;\n}\n\nfunction expectArraysPredicate(\n    actual: TensorLike, expected: TensorLike,\n    predicate: (a: {}, b: {}) => boolean) {\n  let checkClassType = true;\n  if (isTypedArray(actual) || isTypedArray(expected)) {\n    checkClassType = false;\n  }\n  if (isTypedArray(actual) && isTypedArray(expected)) {\n    checkClassType = true;\n  }\n  if (checkClassType) {\n    const aType = actual.constructor.name;\n    const bType = expected.constructor.name;\n\n    if (aType !== bType) {\n      throw new Error(\n          `Arrays are of different type. Actual: ${aType}. ` +\n          `Expected: ${bType}`);\n    }\n  }\n\n  if (Array.isArray(actual) && Array.isArray(expected)) {\n    const actualShape = inferShape(actual);\n    const expectedShape = inferShape(expected);\n    if (!arraysEqual(actualShape, expectedShape)) {\n      throw new Error(\n          `Arrays have different shapes. ` +\n          `Actual: [${actualShape}]. Expected: [${expectedShape}]`);\n    }\n  }\n\n  const actualFlat =\n      isTypedArray(actual) ? actual : flatten(actual as RecursiveArray<number>);\n  const expectedFlat = isTypedArray(expected) ?\n      expected :\n      flatten(expected as RecursiveArray<number>);\n\n  if (actualFlat.length !== expectedFlat.length) {\n    throw new Error(\n        `Arrays have different lengths actual: ${actualFlat.length} vs ` +\n        `expected: ${expectedFlat.length}.\\n` +\n        `Actual:   ${actualFlat}.\\n` +\n        `Expected: ${expectedFlat}.`);\n  }\n  for (let i = 0; i < expectedFlat.length; ++i) {\n    const a = actualFlat[i];\n    const e = expectedFlat[i];\n\n    if (!predicate(a, e)) {\n      throw new Error(\n          `Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${e}.\\n` +\n          `Actual:   ${actualFlat}.\\n` +\n          `Expected: ${expectedFlat}.`);\n    }\n  }\n}\n\nexport interface DoneFn {\n  (): void;\n  fail: (message?: Error|string) => void;\n}\n\nexport function expectPromiseToFail(fn: () => Promise<{}>, done: DoneFn): void {\n  fn().then(() => done.fail(), () => done());\n}\n\nexport function expectArraysEqual(actual: TensorLike, expected: TensorLike) {\n  const exp = typeof expected === 'string' || typeof expected === 'number' ||\n          typeof expected === 'boolean' ?\n      [expected] as number[] :\n      expected as number[];\n  if (isString(actual) || isString((actual as string[])[0]) ||\n      isString(expected) || isString((expected as string[])[0])) {\n    // tslint:disable-next-line: triple-equals\n    return expectArraysPredicate(actual, exp, (a, b) => a == b);\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, 0));\n}\n\nexport function expectNumbersClose(a: number, e: number, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  if (!areClose(a, e, epsilon)) {\n    throw new Error(`Numbers differ: actual === ${a}, expected === ${e}`);\n  }\n}\n\nfunction areClose(a: number, e: number, epsilon: number): boolean {\n  if (!isFinite(a) && !isFinite(e)) {\n    return true;\n  }\n  if (isNaN(a) || isNaN(e) || Math.abs(a - e) > epsilon) {\n    return false;\n  }\n  return true;\n}\n\nexport function expectValuesInRange(\n    actual: TypedArray|number[], low: number, high: number) {\n  for (let i = 0; i < actual.length; i++) {\n    if (actual[i] < low || actual[i] > high) {\n      throw new Error(\n          `Value out of range:${actual[i]} low: ${low}, high: ${high}`);\n    }\n  }\n}\n\nexport function expectArrayBuffersEqual(\n    actual: ArrayBuffer, expected: ArrayBuffer) {\n  // Safari & Jasmine don't like comparing ArrayBuffers directly. Wrapping in\n  // a Float32Array solves this issue.\n  expect(new Float32Array(actual)).toEqual(new Float32Array(expected));\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from './backends/backend';\nimport {ENGINE, Engine, MemoryInfo, ProfileInfo, ScopeFn, TimingInfo} from './engine';\nimport {env} from './environment';\n\nimport {Platform} from './platforms/platform';\nimport {setDeprecationWarningFn, Tensor} from './tensor';\nimport {TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\n\n/**\n * Enables production mode which disables correctness checks in favor of\n * performance.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableProdMode(): void {\n  env().set('PROD', true);\n}\n\n/**\n * Enables debug mode which will log information about all executed kernels:\n * the elapsed time of the kernel execution, as well as the rank, shape, and\n * size of the output tensor.\n *\n * Debug mode will significantly slow down your application as it will\n * download the result of every operation to the CPU. This should not be used in\n * production. Debug mode does not affect the timing information of the kernel\n * execution as we do not measure download time in the kernel execution time.\n *\n * See also: `tf.profile`, `tf.memory`.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableDebugMode(): void {\n  env().set('DEBUG', true);\n}\n\n/** Globally disables deprecation warnings */\nexport function disableDeprecationWarnings(): void {\n  env().set('DEPRECATION_WARNINGS_ENABLED', false);\n  console.warn(`TensorFlow.js deprecation warnings have been disabled.`);\n}\n\n/** Warn users about deprecated functionality. */\nexport function deprecationWarn(msg: string) {\n  if (env().getBool('DEPRECATION_WARNINGS_ENABLED')) {\n    console.warn(\n        msg + ' You can disable deprecation warnings with ' +\n        'tf.disableDeprecationWarnings().');\n  }\n}\nsetDeprecationWarningFn(deprecationWarn);\n\n/**\n * Dispose all variables kept in backend engine.\n *\n * @doc {heading: 'Environment'}\n */\nexport function disposeVariables(): void {\n  ENGINE.disposeVariables();\n}\n\n/**\n * It returns the global engine that keeps track of all tensors and backends.\n *\n * @doc {heading: 'Environment'}\n */\nexport function engine(): Engine {\n  return ENGINE;\n}\n\n/**\n * Returns memory info at the current time in the program. The result is an\n * object with the following properties:\n *\n * - `numBytes`: Number of bytes allocated (undisposed) at this time.\n * - `numTensors`: Number of unique tensors allocated.\n * - `numDataBuffers`: Number of unique data buffers allocated\n *   (undisposed) at this time, which is \u2264 the number of tensors\n *   (e.g. `a.reshape(newShape)` makes a new Tensor that shares the same\n *   data buffer with `a`).\n * - `unreliable`: True if the memory usage is unreliable. See `reasons` when\n *    `unreliable` is true.\n * - `reasons`: `string[]`, reasons why the memory is unreliable, present if\n *    `unreliable` is true.\n *\n * WebGL Properties:\n * - `numBytesInGPU`: Number of bytes allocated (undisposed) in the GPU only at\n *     this time.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function memory(): MemoryInfo {\n  return ENGINE.memory();\n}\n\n/**\n * Executes the provided function `f()` and returns a promise that resolves\n * with information about the function's memory use:\n * - `newBytes`: the number of new bytes allocated\n * - `newTensors`: the number of new tensors created\n * - `peakBytes`: the peak number of bytes allocated\n * - `kernels`: an array of objects for each kernel involved that reports\n * their input and output shapes, number of bytes used, and number of new\n * tensors created.\n *\n * ```js\n * const profile = await tf.profile(() => {\n *   const x = tf.tensor1d([1, 2, 3]);\n *   let x2 = x.square();\n *   x2.dispose();\n *   x2 = x.square();\n *   x2.dispose();\n *   return x;\n * });\n *\n * console.log(`newBytes: ${profile.newBytes}`);\n * console.log(`newTensors: ${profile.newTensors}`);\n * console.log(`byte usage over all kernels: ${profile.kernels.map(k =>\n * k.totalBytesSnapshot)}`);\n * ```\n *\n *\n * @doc {heading: 'Performance', subheading: 'Profile'}\n */\nexport function profile(f: () => (TensorContainer | Promise<TensorContainer>)):\n    Promise<ProfileInfo> {\n  return ENGINE.profile(f);\n}\n\n/**\n * Executes the provided function `fn` and after it is executed, cleans up all\n * intermediate tensors allocated by `fn` except those returned by `fn`.\n * `fn` must not return a Promise (async functions not allowed). The returned\n * result can be a complex object.\n *\n * Using this method helps avoid memory leaks. In general, wrap calls to\n * operations in `tf.tidy` for automatic memory cleanup.\n *\n * NOTE: Variables do *not* get cleaned up when inside a tidy(). If you want to\n * dispose variables, please use `tf.disposeVariables` or call dispose()\n * directly on variables.\n *\n * ```js\n * // y = 2 ^ 2 + 1\n * const y = tf.tidy(() => {\n *   // a, b, and one will be cleaned up when the tidy ends.\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *   const b = a.square();\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * y.print();\n * ```\n *\n * @param nameOrFn The name of the closure, or the function to execute.\n *     If a name is provided, the 2nd argument should be the function.\n *     If debug mode is on, the timing and the memory usage of the function\n *     will be tracked and displayed on the console using the provided name.\n * @param fn The function to execute.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function tidy<T extends TensorContainer>(\n    nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>): T {\n  return ENGINE.tidy(nameOrFn, fn);\n}\n\n/**\n * Disposes any `tf.Tensor`s found within the provided object.\n *\n * @param container an object that may be a `tf.Tensor` or may directly\n *     contain `tf.Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. If\n *     the object is not a `tf.Tensor` or does not contain `Tensors`, nothing\n *     happens. In general it is safe to pass any object here, except that\n *     `Promise`s are not supported.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function dispose(container: TensorContainer) {\n  const tensors = getTensorsInContainer(container);\n  tensors.forEach(tensor => tensor.dispose());\n}\n\n/**\n * Keeps a `tf.Tensor` generated inside a `tf.tidy` from being disposed\n * automatically.\n *\n * ```js\n * let b;\n * const y = tf.tidy(() => {\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *\n *   // b will not be cleaned up by the tidy. a and one will be cleaned up\n *   // when the tidy ends.\n *   b = tf.keep(a.square());\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * console.log('y:');\n * y.print();\n * console.log('b:');\n * b.print();\n * ```\n *\n * @param result The tensor to keep from being disposed.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function keep<T extends Tensor>(result: T): T {\n  return ENGINE.keep(result);\n}\n\n/**\n * Executes `f()` and returns a promise that resolves with timing\n * information.\n *\n * The result is an object with the following properties:\n *\n * - `wallMs`: Wall execution time.\n * - `kernelMs`: Kernel execution time, ignoring data transfer. If using the\n * WebGL backend and the query timer extension is not available, this will\n * return an error object.\n * - On `WebGL` The following additional properties exist:\n *   - `uploadWaitMs`: CPU blocking time on texture uploads.\n *   - `downloadWaitMs`: CPU blocking time on texture downloads (readPixels).\n *\n * ```js\n * const x = tf.randomNormal([20, 20]);\n * const time = await tf.time(() => x.matMul(x));\n *\n * console.log(`kernelMs: ${time.kernelMs}, wallTimeMs: ${time.wallMs}`);\n * ```\n *\n * @param f The function to execute and time.\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nexport function time(f: () => void): Promise<TimingInfo> {\n  return ENGINE.time(f);\n}\n\n/**\n * Sets the backend (cpu, webgl, wasm, etc) responsible for creating tensors and\n * executing operations on those tensors. Returns a promise that resolves\n * to a boolean if the backend initialization was successful.\n *\n * Note this disposes the current backend, if any, as well as any tensors\n * associated with it. A new backend is initialized, even if it is of the\n * same type as the previous one.\n *\n * @param backendName The name of the backend. Currently supports\n *     `'webgl'|'cpu'` in the browser, `'tensorflow'` under node.js\n *     (requires tfjs-node), and `'wasm'` (requires tfjs-backend-wasm).\n *\n * @doc {heading: 'Backends'}\n */\nexport function setBackend(backendName: string): Promise<boolean> {\n  return ENGINE.setBackend(backendName);\n}\n\n/**\n * Returns a promise that resolves when the currently selected backend (or the\n * highest priority one) has initialized. Await this promise when you are using\n * a backend that has async initialization.\n *\n * @doc {heading: 'Backends'}\n */\nexport function ready(): Promise<void> {\n  return ENGINE.ready();\n}\n\n/**\n * Returns the current backend name (cpu, webgl, etc). The backend is\n * responsible for creating tensors and executing operations on those tensors.\n *\n * @doc {heading: 'Backends'}\n */\nexport function getBackend(): string {\n  return ENGINE.backendName;\n}\n\n/**\n * Removes a backend and the registered factory.\n *\n * @doc {heading: 'Backends'}\n */\nexport function removeBackend(name: string): void {\n  ENGINE.removeBackend(name);\n}\n\n/**\n * Finds the backend registered under the provided name. Returns null if the\n * name is not in the registry, or the registration hasn't finished yet.\n */\nexport function findBackend(name: string): KernelBackend {\n  return ENGINE.findBackend(name);\n}\n\n/**\n * Finds the backend factory registered under the provided name. Returns a\n * function that produces a new backend when called. Returns null if the name\n * is not in the registry.\n */\nexport function findBackendFactory(name: string): () =>\n    KernelBackend | Promise<KernelBackend> {\n  return ENGINE.findBackendFactory(name);\n}\n\n/**\n * Registers a global backend. The registration should happen when importing\n * a module file (e.g. when importing `backend_webgl.ts`), and is used for\n * modular builds (e.g. custom tfjs bundle with only webgl support).\n *\n * @param factory The backend factory function. When called, it should\n * return a backend instance, or a promise of an instance.\n * @param priority The priority of the backend (higher = more important).\n *     In case multiple backends are registered, the priority is used to find\n *     the best backend. Defaults to 1.\n * @return False if there is already a registered backend under this name, true\n *     if not.\n *\n * @doc {heading: 'Backends'}\n */\nexport function registerBackend(\n    name: string, factory: () => KernelBackend | Promise<KernelBackend>,\n    priority = 1): boolean {\n  return ENGINE.registerBackend(name, factory, priority);\n}\n\n/**\n * Gets the current backend. If no backends have been initialized, this will\n * attempt to initialize the best backend. Will throw an error if the highest\n * priority backend has async initialization, in which case, you should call\n * 'await tf.ready()' before running other code.\n *\n * @doc {heading: 'Backends'}\n */\nexport function backend(): KernelBackend {\n  return ENGINE.backend;\n}\n\n/**\n * Sets the global platform.\n *\n * @param platformName The name of this platform.\n * @param platform A platform implementation.\n */\nexport function setPlatform(platformName: string, platform: Platform) {\n  env().setPlatform(platformName, platform);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Add, AddInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n *\n * ```js\n * // Broadcast add a with b.\n * const a = tf.scalar(5);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n * @param a The first `tf.Tensor` to add.\n * @param b The second `tf.Tensor` to add. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction add_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'add');\n  let $b = convertToTensor(b, 'b', 'add');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.add($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: AddInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Add) as T;\n}\n\nexport const add = op({add_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FloorDiv, FloorDivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n * The result is rounded with floor function.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.floorDiv(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.floorDiv(b).print();  // or tf.floorDiv(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction floorDiv_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'floorDiv');\n  let $b = convertToTensor(b, 'b', 'floorDiv');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.floorDiv($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: FloorDivInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             FloorDiv) as T;\n}\n\nexport const floorDiv = op({floorDiv_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Div, DivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {floorDiv} from './floorDiv';\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction div_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'int32' && $b.dtype === 'int32') {\n    return floorDiv($a, $b);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.realDivide($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: DivInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Div,\n             attrs) as T;\n}\n\nexport const div = op({div_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Multiply, MultiplyInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Multiplies two `tf.Tensor`s element-wise, A * B. Supports broadcasting.\n *\n * We also expose `tf.mulStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([2, 3, 4, 5]);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n *\n * ```js\n * // Broadcast mul a with b.\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.scalar(5);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n * @param a The first tensor to multiply.\n * @param b The second tensor to multiply. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mul_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mul');\n  let $b = convertToTensor(b, 'b', 'mul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.multiply($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MultiplyInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Multiply) as T;\n}\nexport const mul = op({mul_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Abs, AbsInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes absolute value element-wise: `abs(x)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.abs().print();  // or tf.abs(x)\n * ```\n * @param x The input `tf.Tensor`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction abs_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'abs');\n\n  const inputs: AbsInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    if ($x.dtype === 'complex64') {\n      return backend.complexAbs($x);\n    }\n\n    return backend.abs($x);\n  }, inputs as {} as NamedTensorMap, null /* grad */, Abs);\n}\n\nexport const abs = op({abs_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Acos, AcosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes acos of the input `tf.Tensor` element-wise: `acos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.acos().print();  // or tf.acos(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acos');\n  const inputs: AcosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acos);\n}\nexport const acos = op({acos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Acosh, AcoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the inverse hyperbolic cos of the input `tf.Tensor` element-wise:\n * `acosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([10, 1, 3, 5.7]);\n *\n * x.acosh().print();  // or tf.acosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acosh');\n  const inputs: AcoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acosh);\n}\nexport const acosh = op({acosh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AddN, AddNInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Adds a list of `tf.Tensor`s element-wise, each with the same shape and dtype.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n *\n * tf.addN([a, b, c]).print();\n * ```\n * @param tensors A list of tensors with the same shape and dtype.\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction addN_<T extends Tensor>(tensors: Array<T|TensorLike>): T {\n  util.assert(\n      Array.isArray(tensors),\n      () => 'The argument passed to tf.addN() must be a list of tensors');\n  util.assert(\n      tensors.length >= 1,\n      () => `Must pass at least one tensor to tf.addN(), but got ` +\n          `${tensors.length}`);\n\n  const $tensors =\n      tensors.map((t, i) => convertToTensor(t, `tensors${i}`, 'addN'));\n\n  const firstTensor = $tensors[0];\n  $tensors.forEach(t => {\n    if (t.dtype !== firstTensor.dtype) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same dtype');\n    }\n  });\n\n  $tensors.forEach(t => {\n    if (!util.arraysEqual(t.shape, firstTensor.shape)) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same shape');\n    }\n  });\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.addN($tensors);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: AddNInputs = $tensors;\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, AddN) as\n      T;\n}\n\nexport const addN = op({addN_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\n/**\n * Returns true if the axis specifies the inner most dimensions of the\n * array.\n */\nexport function axesAreInnerMostDims(axes: number[], rank: number): boolean {\n  for (let i = 0; i < axes.length; ++i) {\n    if (axes[axes.length - i - 1] !== rank - 1 - i) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function combineLocations(\n    outputLoc: number[], reduceLoc: number[], axes: number[]): number[] {\n  const rank = outputLoc.length + reduceLoc.length;\n  const loc = [];\n  let outIdx = 0;\n  let reduceIdx = 0;\n  \u00A0 for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      loc.push(outputLoc[outIdx++]);\n    } else {\n      loc.push(reduceLoc[reduceIdx++]);\n    }\n  }\n  return loc;\n}\n\nexport function computeOutAndReduceShapes(\n    aShape: number[], axes: number[]): [number[], number[]] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      outShape.push(aShape[dim]);\n    }\n  }\n  const reduceShape = axes.map(dim => aShape[dim]);\n  return [outShape, reduceShape];\n}\n\nexport function expandShapeToKeepDim(\n    shape: number[], axes: number[]): number[] {\n  const reduceSubShape = axes.map(x => 1);\n  return combineLocations(shape, reduceSubShape, axes);\n}\n\nexport function assertAxesAreInnerMostDims(\n    msg: string, axes: number[], rank: number): void {\n  util.assert(\n      axesAreInnerMostDims(axes, rank),\n      () => `${msg} supports only inner-most axes for now. ` +\n          `Got axes ${axes} and rank-${rank} input.`);\n}\n\n/**\n * Returns the axes permutation to be used with `tf.transpose`, if such\n * permutation is necessary. Otherwise it returns null. This method is used by\n * operations that operate only on inner-most axes.\n */\nexport function getAxesPermutation(axes: number[], rank: number): number[]|\n    null {\n  if (axesAreInnerMostDims(axes, rank)) {\n    return null;\n  }\n  const result: number[] = [];\n  for (let i = 0; i < rank; ++i) {\n    if (axes.indexOf(i) === -1) {\n      result.push(i);\n    }\n  }\n  axes.forEach(axis => result.push(axis));\n  return result;\n}\n\n/** Returns the axes permutation that undoes the original permutation. */\nexport function getUndoAxesPermutation(axes: number[]): number[] {\n  return axes.map((axis, i) => [i, axis])\n      .sort((a, b) => a[1] - b[1])\n      .map(x => x[0]);\n}\n\nexport function getInnerMostAxes(numAxes: number, rank: number): number[] {\n  const res: number[] = [];\n  for (let i = rank - numAxes; i < rank; ++i) {\n    res.push(i);\n  }\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {All, AllAttrs, AllInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical and of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.all().print();  // or tf.all(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.all(axis).print();  // or tf.all(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction all_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'all', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.all($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AllInputs = {x: $x};\n  const attrs: AllAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, All,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const all = op({all_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Any, AnyAttrs, AnyInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical or of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.any().print();  // or tf.any(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.any(axis).print();  // or tf.any(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction any_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'any', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.any($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AnyInputs = {x: $x};\n  const attrs: AnyAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Any,\n             attrs as {} as NamedAttrMap) as T;\n}\n\n// tslint:disable-next-line:variable-name\nexport const any = op({any_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMax, ArgMaxAttrs, ArgMaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the maximum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMax().print();  // or tf.argMax(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMax(axis).print();  // or tf.argMax(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMax_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMax');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMax($x, axes[0]);\n  };\n\n  const inputs: ArgMaxInputs = {x: $x};\n  const attrs: ArgMaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMax,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMax = op({argMax_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMin, ArgMinAttrs, ArgMinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the minimum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMin().print();  // or tf.argMin(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMin(axis).print();  // or tf.argMin(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMin_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMin');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if (axis == null) {\n      axis = 0;\n    }\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMin($x, axes[0]);\n  };\n\n  const inputs: ArgMinInputs = {x: $x};\n  const attrs: ArgMinAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMin,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMin = op({argMin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asin, AsinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes asin of the input `tf.Tensor` element-wise: `asin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asin().print();  // or tf.asin(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asin');\n  const inputs: AsinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asin);\n}\nexport const asin = op({asin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asinh, AsinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic sin of the input `tf.Tensor` element-wise:\n * `asinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asinh().print();  // or tf.asinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asinh');\n\n  const inputs: AsinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asinh);\n}\nexport const asinh = op({asinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atan, AtanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes atan of the input `tf.Tensor` element-wise: `atan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.atan().print();  // or tf.atan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atan');\n\n  const inputs: AtanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atan);\n}\nexport const atan = op({atan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Atan2, Atan2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes arctangent of `tf.Tensor`s a / b element-wise: `atan2(a, b)`.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1.0, 1.0, -1.0, .7]);\n * const b = tf.tensor1d([2.0, 13.0, 3.5, .21]);\n *\n * tf.atan2(a, b).print()\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan2_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'atan2');\n  let $b = convertToTensor(b, 'b', 'atan2');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.atan2($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: Atan2Inputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Atan2) as T;\n}\n\nexport const atan2 = op({atan2_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atanh, AtanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic tan of the input `tf.Tensor` element-wise:\n * `atanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.atanh().print();  // or tf.atanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atanh');\n\n  const inputs: AtanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atanh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atanh);\n}\nexport const atanh = op({atanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\ntype PadType = 'SAME'|'VALID'|'NUMBER'|'EXPLICIT';\n\n// For NHWC should be in the following form:\n//  [[0, 0], [pad_top,pad_bottom], [pad_left, pad_right], [0, 0]]\n// For NCHW should be in the following form:\n//  [[0, 0], [0, 0], [pad_top,pad_bottom], [pad_left, pad_right]]\n// Reference: https://www.tensorflow.org/api_docs/python/tf/nn/conv2d\nexport type ExplicitPadding =\n    [[number, number], [number, number], [number, number], [number, number]];\n\nexport type PadInfo = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  type: PadType\n};\n\nexport type PadInfo3D = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  front: number,\n  back: number,\n  type: PadType\n};\n\n/**\n * Information about the forward pass of a convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv2DInfo = {\n  batchSize: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideHeight: number,\n  strideWidth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo,\n  inShape: [number, number, number, number],\n  outShape: [number, number, number, number],\n  filterShape: [number, number, number, number]\n};\n\n/**\n *\n * @param inputShape Input tensor shape is of the following dimensions:\n *     `[batch, height, width, inChannels]`.\n * @param filterShape The filter shape is of the following dimensions:\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat The data format of the input and output data.\n *     Defaults to 'NHWC'.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.\n *     Defaults to `[1, 1]`. If `dilations` is a single number, then\n *     `dilationHeight == dilationWidth`.\n */\nexport function computeDilation2DInfo(\n    inputShape: [number, number, number, number],\n    filterShape: [number, number, number], strides: number|[number, number],\n    pad: 'same'|'valid'|number, dataFormat: 'NHWC' = 'NHWC',\n    dilations: number|[number, number]) {\n  // `computerConv2DInfo` require filterShape to be in the dimension of:\n  // `[filterHeight, filterWidth, depth, outDepth]`, dilation2d doesn't have\n  // outDepth, it should have the same depth as the input.\n  // Input shape: [batch, height, width, inChannels]\n  const inputChannels = inputShape[3];\n  const $filterShape =\n      [...filterShape, inputChannels] as [number, number, number, number];\n  const $dataFormat = convertConv2DDataFormat(dataFormat);\n\n  return computeConv2DInfo(\n      inputShape, $filterShape, strides, dilations, pad,\n      null /* roundingMode */, null /* depthWise */, $dataFormat);\n}\n\nexport function computePool2DInfo(\n    inShape: [number, number, number, number],\n    filterSize: [number, number]|number, strides: number|[number, number],\n    dilations: number|[number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  const [filterHeight, filterWidth] = parseTupleParam(filterSize);\n\n  let filterShape: [number, number, number, number];\n  if (dataFormat === 'channelsLast') {\n    filterShape = [filterHeight, filterWidth, inShape[3], inShape[3]];\n  } else if (dataFormat === 'channelsFirst') {\n    filterShape = [filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv2DInfo(\n      inShape, filterShape, strides, dilations, pad, roundingMode, false,\n      dataFormat);\n}\n\n/**\n * Computes the information for a forward pass of a pooling3D operation.\n */\nexport function computePool3DInfo(\n    inShape: [number, number, number, number, number],\n    filterSize: number|[number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC'): Conv3DInfo {\n  const [filterDepth, filterHeight, filterWidth] = parse3TupleParam(filterSize);\n\n  let filterShape: [number, number, number, number, number];\n  let $dataFormat: 'channelsFirst'|'channelsLast';\n  if (dataFormat === 'NDHWC') {\n    $dataFormat = 'channelsLast';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[4], inShape[4]];\n  } else if (dataFormat === 'NCDHW') {\n    $dataFormat = 'channelsFirst';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv3DInfo(\n      inShape, filterShape, strides, dilations, pad, false, $dataFormat,\n      roundingMode);\n}\n\n/**\n * Computes the information for a forward pass of a convolution/pooling\n * operation.\n */\nexport function computeConv2DInfo(\n    inShape: [number, number, number, number],\n    filterShape: [number, number, number, number],\n    strides: number|[number, number], dilations: number|[number, number],\n    pad: 'same'|'valid'|number|ExplicitPadding,\n    roundingMode?: 'floor'|'round'|'ceil', depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  let [batchSize, inHeight, inWidth, inChannels] = [-1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterHeight, filterWidth, , filterChannels] = filterShape;\n  const [strideHeight, strideWidth] = parseTupleParam(strides);\n  const [dilationHeight, dilationWidth] = parseTupleParam(dilations);\n\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outHeight, outWidth} = getPadAndOutInfo(\n      pad, inHeight, inWidth, strideHeight, strideWidth, effectiveFilterHeight,\n      effectiveFilterWidth, roundingMode, dataFormat);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inHeight,\n    inWidth,\n    inChannels,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideHeight,\n    strideWidth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\n/**\n * Information about the forward pass of a 3D convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv3DInfo = {\n  batchSize: number,\n  inDepth: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideDepth: number,\n  strideHeight: number,\n  strideWidth: number,\n  dilationDepth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterDepth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterDepth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo3D,\n  inShape: [number, number, number, number, number],\n  outShape: [number, number, number, number, number],\n  filterShape: [number, number, number, number, number]\n};\n\n/**\n * Computes the information for a forward pass of a 3D convolution/pooling\n * operation.\n */\nexport function computeConv3DInfo(\n    inShape: [number, number, number, number, number],\n    filterShape: [number, number, number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast',\n    roundingMode?: 'floor'|'round'|'ceil'): Conv3DInfo {\n  let [batchSize, inDepth, inHeight, inWidth, inChannels] =\n      [-1, -1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inDepth, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inDepth, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterDepth, filterHeight, filterWidth, , filterChannels] =\n      filterShape;\n  const [strideDepth, strideHeight, strideWidth] = parse3TupleParam(strides);\n  const [dilationDepth, dilationHeight, dilationWidth] =\n      parse3TupleParam(dilations);\n\n  const effectiveFilterDepth =\n      getEffectiveFilterSize(filterDepth, dilationDepth);\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outDepth, outHeight, outWidth} = get3DPadAndOutInfo(\n      pad, inDepth, inHeight, inWidth, strideDepth, strideHeight, strideWidth,\n      effectiveFilterDepth, effectiveFilterHeight, effectiveFilterWidth,\n      roundingMode);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outDepth, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outDepth, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inDepth,\n    inHeight,\n    inWidth,\n    inChannels,\n    outDepth,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideDepth,\n    strideHeight,\n    strideWidth,\n    filterDepth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterDepth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationDepth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\nfunction computeOutputShape2D(\n    inShape: [number, number], fieldSize: number, stride: number,\n    zeroPad?: number, roundingMode?: 'floor'|'round'|'ceil'): [number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputRows = inShape[0];\n  const inputCols = inShape[1];\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputRows, outputCols];\n}\n\nfunction computeOutputShape4D(\n    inShape: [number, number, number, number], fieldSize: number,\n    outChannels: number, stride: number, zeroPad?: number,\n    roundingMode?: 'floor'|'round'|'ceil'): [number, number, number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputDepth = inShape[0];\n  const inputRows = inShape[1];\n  const inputCols = inShape[2];\n\n  const outputDepths = conditionalRound(\n      (inputDepth - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputDepths),\n      () => `The output # of depths (${outputDepths}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputDepths, outputRows, outputCols, outChannels];\n}\n\nexport function computeDefaultPad(\n    inputShape: [number, number]|[number, number, number, number],\n    fieldSize: number, stride: number, dilation = 1): number {\n  const effectiveFieldSize = getEffectiveFilterSize(fieldSize, dilation);\n  return Math.floor(\n      (inputShape[0] * (stride - 1) - stride + effectiveFieldSize) / 2);\n}\n\nfunction parseTupleParam(param: number|number[]): [number, number, number] {\n  if (typeof param === 'number') {\n    return [param, param, param];\n  }\n  if (param.length === 2) {\n    return [param[0], param[1], 1];\n  }\n  return param as [number, number, number];\n}\n\nfunction parse3TupleParam(param: number|[number, number, number]):\n    [number, number, number] {\n  return typeof param === 'number' ? [param, param, param] : param;\n}\n\n/* See https://www.tensorflow.org/api_docs/python/tf/nn/atrous_conv2d\n * Atrous convolution is equivalent to standard convolution with upsampled\n * filters with effective_filter_height =\n * filter_height + (filter_height - 1) * (dilation - 1)\n * and effective_filter_width =\n * filter_width + (filter_width - 1) * (dilation - 1),\n * produced by inserting dilation - 1 zeros along consecutive elements across\n * the filters' spatial dimensions.\n * When there is a dilation, this converts a filter dimension to the\n * effective filter dimension, so it can be used in a standard convolution.\n */\nfunction getEffectiveFilterSize(filterSize: number, dilation: number) {\n  if (dilation <= 1) {\n    return filterSize;\n  }\n\n  return filterSize + (filterSize - 1) * (dilation - 1);\n}\n\nfunction getPadAndOutInfo(\n    pad: 'same'|'valid'|number|ExplicitPadding, inHeight: number,\n    inWidth: number, strideHeight: number, strideWidth: number,\n    filterHeight: number, filterWidth: number,\n    roundingMode: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|\n    'channelsLast'): {padInfo: PadInfo, outHeight: number, outWidth: number} {\n  let padInfo: PadInfo;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {top: pad, bottom: pad, left: pad, right: pad, type: padType};\n    const outShape = computeOutputShape2D(\n        [inHeight, inWidth], filterHeight, strideHeight, pad, roundingMode);\n    outHeight = outShape[0];\n    outWidth = outShape[1];\n  } else if (pad === 'same') {\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongHeight =\n        Math.max(0, (outHeight - 1) * strideHeight + filterHeight - inHeight);\n    const padAlongWidth =\n        Math.max(0, (outWidth - 1) * strideWidth + filterWidth - inWidth);\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n    padInfo = {top, bottom, left, right, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {top: 0, bottom: 0, left: 0, right: 0, type: 'VALID'};\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else if (typeof pad === 'object') {\n    const top = dataFormat === 'channelsLast' ? pad[1][0] : pad[2][0];\n    const bottom = dataFormat === 'channelsLast' ? pad[1][1] : pad[2][1];\n    const left = dataFormat === 'channelsLast' ? pad[2][0] : pad[3][0];\n    const right = dataFormat === 'channelsLast' ? pad[2][1] : pad[3][1];\n    const padType = (top === 0 && bottom === 0 && left === 0 && right === 0) ?\n        'VALID' :\n        'EXPLICIT';\n    padInfo = {top, bottom, left, right, type: padType};\n    outHeight = conditionalRound(\n        (inHeight - filterHeight + top + bottom) / strideHeight + 1,\n        roundingMode);\n    outWidth = conditionalRound(\n        (inWidth - filterWidth + left + right) / strideWidth + 1, roundingMode);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outHeight, outWidth};\n}\n\nfunction get3DPadAndOutInfo(\n    pad: 'same'|'valid'|number, inDepth: number, inHeight: number,\n    inWidth: number, strideDepth: number, strideHeight: number,\n    strideWidth: number, filterDepth: number, filterHeight: number,\n    filterWidth: number, roundingMode?: 'floor'|'round'|'ceil'): {\n  padInfo: PadInfo3D,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number\n} {\n  let padInfo: PadInfo3D;\n  let outDepth: number;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {\n      top: pad,\n      bottom: pad,\n      left: pad,\n      right: pad,\n      front: pad,\n      back: pad,\n      type: padType\n    };\n    const outShape = computeOutputShape4D(\n        [inDepth, inHeight, inWidth, 1], filterDepth, 1, strideDepth, pad,\n        roundingMode);\n    outDepth = outShape[0];\n    outHeight = outShape[1];\n    outWidth = outShape[2];\n  } else if (pad === 'same') {\n    outDepth = Math.ceil(inDepth / strideDepth);\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongDepth = (outDepth - 1) * strideDepth + filterDepth - inDepth;\n    const padAlongHeight =\n        (outHeight - 1) * strideHeight + filterHeight - inHeight;\n    const padAlongWidth = (outWidth - 1) * strideWidth + filterWidth - inWidth;\n    const front = Math.floor(padAlongDepth / 2);\n    const back = padAlongDepth - front;\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n\n    padInfo = {top, bottom, left, right, front, back, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {\n      top: 0,\n      bottom: 0,\n      left: 0,\n      right: 0,\n      front: 0,\n      back: 0,\n      type: 'VALID'\n    };\n    outDepth = Math.ceil((inDepth - filterDepth + 1) / strideDepth);\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outDepth, outHeight, outWidth};\n}\n\n/**\n * Rounds a value depending on the rounding mode\n * @param value\n * @param roundingMode\n */\nfunction conditionalRound(\n    value: number, roundingMode?: 'floor'|'round'|'ceil') {\n  if (!roundingMode) {\n    return value;\n  }\n  switch (roundingMode) {\n    case 'round':\n      // used for Caffe Conv\n      return Math.round(value);\n    case 'ceil':\n      // used for Caffe Pool\n      return Math.ceil(value);\n    case 'floor':\n      return Math.floor(value);\n    default:\n      throw new Error(`Unknown roundingMode ${roundingMode}`);\n  }\n}\n\nexport function tupleValuesAreOne(param: number|number[]): boolean {\n  const [dimA, dimB, dimC] = parseTupleParam(param);\n  return dimA === 1 && dimB === 1 && dimC === 1;\n}\n\nexport function eitherStridesOrDilationsAreOne(\n    strides: number|number[], dilations: number|number[]): boolean {\n  return tupleValuesAreOne(strides) || tupleValuesAreOne(dilations);\n}\n\n/**\n * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to\n *    'channelsLast'|'channelsFirst'\n * @param dataFormat in 'NHWC'|'NCHW' mode\n * @return dataFormat in 'channelsLast'|'channelsFirst' mode\n * @throws unknown dataFormat\n */\nexport function convertConv2DDataFormat(dataFormat: 'NHWC'|'NCHW'):\n    'channelsLast'|'channelsFirst' {\n  if (dataFormat === 'NHWC') {\n    return 'channelsLast';\n  } else if (dataFormat === 'NCHW') {\n    return 'channelsFirst';\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D average pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction avgPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'avgPool', 'float32');\n  const dilations = 1;\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in avgPool: x must be rank 4 but got rank ${x4D.rank}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    save([x4D]);\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      return x4D.clone();\n    }\n\n    return backend.avgPool(x4D, convInfo);\n  };\n\n  const inputs: AvgPoolInputs = {x: x4D};\n\n  const attrs: AvgPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, $x.dtype);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool = op({avgPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {AvgPool3D, AvgPool3DAttrs, AvgPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D average pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.avgPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates:\n *     `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction avgPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'avgPool3d', 'float32');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in avgPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in avgPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n\n    save([x5D]);\n\n    return backend.avgPool3d(x5D, convInfo);\n  };\n\n  const inputs: AvgPool3DInputs = {x: x5D};\n\n  const attrs: AvgPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool3D,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, x5D.dtype);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3d = op({avgPool3d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\nexport function assertParamsConsistent(shapes: number[][], axis: number) {\n  const rank = shapes[0].length;\n  shapes.forEach((shape, i) => {\n    util.assert(\n        shape.length === rank,\n        () =>\n            `Error in concat${rank}D: rank of tensors[${i}] must be the same ` +\n            `as the rank of the rest (${rank})`);\n  });\n\n  util.assert(\n      axis >= 0 && axis < rank,\n      () => `Error in concat${rank}D: axis must be between 0 and ${rank - 1}.`);\n\n  const firstShape = shapes[0];\n  shapes.forEach((shape, i) => {\n    for (let r = 0; r < rank; r++) {\n      util.assert(\n          (r === axis) || (shape[r] === firstShape[r]),\n          () => `Error in concat${rank}D: Shape of tensors[${i}] (${shape}) ` +\n              `does not match the shape of the rest (${firstShape}) ` +\n              `along the non-concatenated axis ${i}.`);\n    }\n  });\n}\n\nexport function computeOutShape(shapes: number[][], axis: number): number[] {\n  const outputShape = shapes[0].slice();\n  for (let i = 1; i < shapes.length; i++) {\n    outputShape[axis] += shapes[i][axis];\n  }\n  return outputShape;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Concat, ConcatAttrs, ConcatInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, parseAxisParam, sizeFromShape} from '../util';\n\nimport {assertParamsConsistent, computeOutShape} from './concat_util';\nimport {op} from './operation';\nimport {tensor} from './tensor';\n\n/**\n * Concatenates a list of `tf.Tensor`s along a given axis.\n *\n * The tensors ranks and types must match, and their sizes must match in all\n * dimensions except `axis`.\n *\n * Also available are stricter rank-specific methods that assert that\n * `tensors` are of the given rank:\n *   - `tf.concat1d`\n *   - `tf.concat2d`\n *   - `tf.concat3d`\n *   - `tf.concat4d`\n *\n * Except `tf.concat1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * a.concat(b).print();  // or a.concat(b)\n * ```\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.concat([a, b, c]).print();\n * ```\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [10, 20]]);\n * const b = tf.tensor2d([[3, 4], [30, 40]]);\n * const axis = 1;\n * tf.concat([a, b], axis).print();\n * ```\n * @param tensors A list of tensors to concatenate.\n * @param axis The axis to concate along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction concat_<T extends Tensor>(tensors: Array<T|TensorLike>, axis = 0): T {\n  assert(tensors.length >= 1, () => 'Pass at least one tensor to concat');\n\n  let $tensors = convertToTensorArray(tensors, 'tensors', 'concat');\n\n  if ($tensors[0].dtype === 'complex64') {\n    $tensors.forEach(tensor => {\n      if (tensor.dtype !== 'complex64') {\n        throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${tensor.dtype}. `);\n      }\n    });\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $axis = parseAxisParam(axis, $tensors[0].shape)[0];\n    const outShape = computeOutShape($tensors.map(t => t.shape), $axis);\n    if (sizeFromShape(outShape) === 0) {\n      return tensor([], outShape) as T;\n    }\n    // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n    $tensors = $tensors.filter(t => t.size > 0);\n    if ($tensors.length === 1) {\n      return $tensors[0];\n    }\n\n    const shapes = $tensors.map(t => t.shape);\n    assertParamsConsistent(shapes, $axis);\n\n    const res = backend.concat($tensors, $axis);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: ConcatInputs = $tensors;\n  const attr: ConcatAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Concat,\n             attr as {} as NamedAttrMap) as T;\n}\n\nexport const concat = op({concat_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sigmoid, SigmoidInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sigmoid element-wise, `1 / (1 + exp(-x))`\n *\n * ```js\n * const x = tf.tensor1d([0, -1, 2, -3]);\n *\n * x.sigmoid().print();  // or tf.sigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sigmoid');\n\n  const inputs: SigmoidInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sigmoid($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sigmoid);\n}\nexport const sigmoid = op({sigmoid_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Slice, SliceAttrs, SliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as slice_util from './slice_util';\n\n/**\n * Extracts a slice from a `tf.Tensor` starting at coordinates `begin`\n * and is of size `size`.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `x` is of the given rank:\n *   - `tf.slice1d`\n *   - `tf.slice2d`\n *   - `tf.slice3d`\n *   - `tf.slice4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.slice([1], [2]).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * x.slice([1, 0], [1, 2]).print();\n * ```\n * @param x The input `tf.Tensor` to slice from.\n * @param begin The coordinates to start the slice from. The length can be\n *     less than the rank of x - the rest of the axes will have implicit 0 as\n *     start. Can also be a single number, in which case it specifies the\n *     first axis.\n * @param size The size of the slice. The length can be less than the rank of\n *     x - the rest of the axes will have implicit -1. A value of -1 requests\n *     the rest of the dimensions in the axis. Can also be a single number,\n *     in which case it specifies the size of the first axis.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction slice_<R extends Rank, T extends Tensor<R>>(\n    x: T|TensorLike, begin: number|number[], size?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'slice');\n\n  if ($x.rank === 0) {\n    throw new Error('Slicing scalar is not possible');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const [begin_, size_] = slice_util.parseSliceParams($x, begin, size);\n    slice_util.assertParamsValid($x, begin_, size_);\n    save([$x]);\n    return backend.slice($x, begin_, size_);\n  };\n\n  const inputs: SliceInputs = {x: $x};\n  const attrs: SliceAttrs = {begin, size};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Slice,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const slice = op({slice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tanh, TanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic tangent of the input `tf.Tensor` element-wise: `tanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, 70]);\n *\n * x.tanh().print();  // or tf.tanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tanh');\n\n  const inputs: TanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const y = backend.tanh($x);\n    save([y]);\n    return y;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tanh);\n}\nexport const tanh = op({tanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {add} from './add';\nimport {concat} from './concat';\nimport {matMul} from './mat_mul';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {slice} from './slice';\nimport {tanh} from './tanh';\n\n/**\n * Computes the next state and output of a BasicLSTMCell.\n *\n * Returns `[newC, newH]`.\n *\n * Derived from tf.contrib.rnn.BasicLSTMCell.\n *\n * @param forgetBias Forget bias for the cell.\n * @param lstmKernel The weights for the cell.\n * @param lstmBias The bias for the cell.\n * @param data The input to the cell.\n * @param c Previous cell state.\n * @param h Previous cell output.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction basicLSTMCell_(\n    forgetBias: Scalar|TensorLike, lstmKernel: Tensor2D|TensorLike,\n    lstmBias: Tensor1D|TensorLike, data: Tensor2D|TensorLike,\n    c: Tensor2D|TensorLike, h: Tensor2D|TensorLike): [Tensor2D, Tensor2D] {\n  const $forgetBias =\n      convertToTensor(forgetBias, 'forgetBias', 'basicLSTMCell');\n  const $lstmKernel =\n      convertToTensor(lstmKernel, 'lstmKernel', 'basicLSTMCell');\n  const $lstmBias = convertToTensor(lstmBias, 'lstmBias', 'basicLSTMCell');\n  const $data = convertToTensor(data, 'data', 'basicLSTMCell');\n  const $c = convertToTensor(c, 'c', 'basicLSTMCell');\n  const $h = convertToTensor(h, 'h', 'basicLSTMCell');\n\n  const combined = concat([$data, $h], 1);\n  const weighted = matMul(combined, $lstmKernel);\n  const res: Tensor2D = add(weighted, $lstmBias);\n\n  // i = input_gate, j = new_input, f = forget_gate, o = output_gate\n  const batchSize = res.shape[0];\n  const sliceCols = res.shape[1] / 4;\n  const sliceSize: [number, number] = [batchSize, sliceCols];\n  const i = slice(res, [0, 0], sliceSize);\n  const j = slice(res, [0, sliceCols], sliceSize);\n  const f = slice(res, [0, sliceCols * 2], sliceSize);\n  const o = slice(res, [0, sliceCols * 3], sliceSize);\n\n  const newC: Tensor2D =\n      add(mul(sigmoid(i), tanh(j)),\n          mul($c, sigmoid(add($forgetBias, f)) as Tensor2D));\n  const newH: Tensor2D = mul(tanh(newC), sigmoid(o));\n  return [newC, newH];\n}\n\nexport const basicLSTMCell = op({basicLSTMCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchToSpaceND, BatchToSpaceNDAttrs, BatchToSpaceNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation reshapes the \"batch\" dimension 0 into `M + 1` dimensions of\n * shape `blockShape + [batch]`, interleaves these blocks back into the grid\n * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with\n * the same rank as the input. The spatial dimensions of this intermediate\n * result are then optionally cropped according to `crops` to produce the\n * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise\n * description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);\n * const blockShape = [2, 2];\n * const crops = [[0, 0], [0, 0]];\n *\n * x.batchToSpaceND(blockShape, crops).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param crops A 2-D array.  Must have shape `[M, 2]`, all values must be >= 0.\n * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input\n * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required\n * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,\n * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,\n * x.shape[N-1]]`\n *\n * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /\n * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`\n * according to `crops` to produce the output of shape: `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],\n * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -\n * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction batchToSpaceND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], crops: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'batchToSpaceND');\n  const prod = blockShape.reduce((a, b) => a * b);\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank is ${$x.rank} but should be > than blockShape.length ${\n          blockShape.length}`);\n\n  util.assert(\n      crops.length === blockShape.length,\n      () => `crops.length is ${\n          crops.length} but should be equal to blockShape.length  ${\n          blockShape.length}`);\n\n  util.assert(\n      $x.shape[0] % prod === 0,\n      () => `input tensor batch is ${\n                $x.shape[0]} but is not divisible by the product of ` +\n          `the elements of blockShape ${blockShape.join(' * ')} === ${prod}`);\n\n  const forward: ForwardFunc<T> = backend => {\n    return backend.batchToSpaceND($x, blockShape, crops);\n  };\n\n  const inputs: BatchToSpaceNDInputs = {x: $x};\n  const attrs: BatchToSpaceNDAttrs = {blockShape, crops};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      BatchToSpaceND, attrs as {} as NamedAttrMap);\n}\n\nexport const batchToSpaceND = op({batchToSpaceND_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, Tensor4D} from '../tensor';\nimport {Rank} from '../types';\nimport {reshape} from './reshape';\n\nexport function xAs4D<R extends Rank>(x: Tensor<R>) {\n  let x4D: Tensor4D;\n  if (x.rank === 0 || x.rank === 1) {\n    x4D = reshape(x, [1, 1, 1, x.size]);\n  } else if (x.rank === 2) {\n    x4D = reshape(x, [1, 1, x.shape[0], x.shape[1]]);\n  } else if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  } else {\n    x4D = x as Tensor4D;\n  }\n\n  return x4D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {xAs4D} from './batchnorm_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Batch normalization.\n *\n * As described in\n * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).\n *\n * Mean, variance, scale, and offset can be of two shapes:\n *   - The same shape as the input.\n *   - In the common case, the depth dimension is the last dimension of x, so\n *     the values would be an `tf.Tensor1D` of shape [depth].\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that parameters passed are of given rank\n *   - `tf.batchNorm2d`\n *   - `tf.batchNorm3d`\n *   - `tf.batchNorm4d`\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction batchNorm_<R extends Rank>(\n    x: Tensor<R>|TensorLike, mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor<R>|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor<R>|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n\n  util.assert(\n      $mean.rank === $variance.rank,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      $offset == null || $mean.rank === $offset.rank,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      $scale == null || $mean.rank === $scale.rank,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  const x4D: Tensor4D = xAs4D($x);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([x4D, $mean, $variance, $scale]);\n\n    return backend.batchNorm(\n        x4D, as1DOr4D($mean), as1DOr4D($variance), as1DOr4D($offset),\n        as1DOr4D($scale), varianceEpsilon);\n  };\n\n  const inputs: FusedBatchNormInputs = {\n    x: x4D,\n    scale: $scale,\n    offset: $offset,\n    mean: $mean,\n    variance: $variance\n  };\n\n  const attrs: FusedBatchNormAttrs = {varianceEpsilon};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      FusedBatchNorm, attrs as {} as NamedAttrMap);\n\n  return reshape(res, $x.shape);\n}\n\nfunction as1DOr4D(x: Tensor): Tensor4D|Tensor1D {\n  if (x == null) {\n    return null;\n  }\n  if (x.rank === 0) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [x.size]) as Tensor1D;\n  } else if (x.rank === 1) {\n    return x as Tensor1D;\n  } else if (x.rank === 2) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, 1, x.shape[0], x.shape[1]]) as Tensor4D;\n  } else if (x.rank === 3) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]) as Tensor4D;\n  }\n  return x as Tensor4D;\n}\n\nexport const batchNorm = op({batchNorm_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 2D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm2d_(\n    x: Tensor2D|TensorLike, mean: Tensor2D|Tensor1D|TensorLike,\n    variance: Tensor2D|Tensor1D|TensorLike,\n    offset?: Tensor2D|Tensor1D|TensorLike, scale?: Tensor2D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor2D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor2D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 2,\n      () => `Error in batchNorm2D: x must be rank 2 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 2 || $mean.rank === 1,\n      () => `Error in batchNorm2D: mean must be rank 2 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 2 || $variance.rank === 1,\n      () => `Error in batchNorm2D: variance must be rank 2 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 2 || $scale.rank === 1,\n        () => `Error in batchNorm2D: scale must be rank 2 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 2 || $offset.rank === 1,\n        () => `Error in batchNorm2D: offset must be rank 2 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm2d = op({batchNorm2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 3D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm3d_(\n    x: Tensor3D|TensorLike, mean: Tensor3D|Tensor1D|TensorLike,\n    variance: Tensor3D|Tensor1D|TensorLike,\n    offset?: Tensor3D|Tensor1D|TensorLike, scale?: Tensor3D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor3D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor3D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 3,\n      () => `Error in batchNorm3D: x must be rank 3 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 3 || $mean.rank === 1,\n      () => `Error in batchNorm3D: mean must be rank 3 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 3 || $variance.rank === 1,\n      () => `Error in batchNorm3D: variance must be rank 3 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 3 || $scale.rank === 1,\n        () => `Error in batchNorm3D: scale must be rank 3 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 3 || $offset.rank === 1,\n        () => `Error in batchNorm3D: offset must be rank 3 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm3d = op({batchNorm3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 4D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm4d_(\n    x: Tensor4D|TensorLike, mean: Tensor4D|Tensor1D|TensorLike,\n    variance: Tensor4D|Tensor1D|TensorLike,\n    offset?: Tensor4D|Tensor1D|TensorLike, scale?: Tensor4D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor4D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor4D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 4,\n      () => `Error in batchNorm4D: x must be rank 4 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 4 || $mean.rank === 1,\n      () => `Error in batchNorm4D: mean must be rank 4 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 4 || $variance.rank === 1,\n      () => `Error in batchNorm4D: variance must be rank 4 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 4 || $scale.rank === 1,\n        () => `Error in batchNorm4D: scale must be rank 4 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 4 || $offset.rank === 1,\n        () => `Error in batchNorm4D: offset must be rank 4 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm4d = op({batchNorm4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE} from '../engine';\nimport {BroadcastTo, BroadCastToAttrs, BroadcastToInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Broadcast an array to a compatible shape NumPy-style.\n *\n * The tensor's shape is compared to the broadcast shape from end to beginning.\n * Ones are prepended to the tensor's shape until is has the same length as\n * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is\n * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then\n * the input tensor is tiled N times along that axis (using tf.tile).\n *\n * @param input The tensor that is to be broadcasted.\n * @param shape The input is to be broadcast to this shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction broadcastTo_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  let input = convertToTensor(x, 'broadcastTo', 'x');\n  const xShape = input.shape;\n\n  if (shape.some(d => !(d > 0) || d % 1 !== 0)) {\n    throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);\n  }\n\n  if (shape.length < input.rank) {\n    throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${\n        input.rank}.`);\n  }\n\n  if (shape.length > input.rank) {\n    const newShape = input.shape.slice();\n    while (newShape.length < shape.length) {\n      newShape.unshift(1);\n    }\n    input = reshape(input, newShape);\n  }\n\n  const inputShape = input.shape;\n  const reps: number[] = Array.from(shape);\n  for (let i = shape.length - 1; i >= 0; i--) {\n    if (inputShape[i] === shape[i]) {\n      reps[i] = 1;\n    } else if (input.shape[i] !== 1) {\n      throw new Error(\n          `broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);\n    }\n  }\n  const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);\n\n  if (axes.length === 0) {\n    return clone(input) as Tensor<R>;\n  }\n\n  const forward = (backend: KernelBackend) => backend.tile(input, reps);\n\n  const inputs: BroadcastToInputs = {x: input};\n  const attrs: BroadCastToAttrs = {shape, inputShape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             BroadcastTo, attrs as unknown as NamedAttrMap) as Tensor<R>;\n}\n\nexport const broadcastTo = op({broadcastTo_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Ceil, CeilInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes ceiling of input `tf.Tensor` element-wise: `ceil(x)`\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.ceil().print();  // or tf.ceil(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction ceil_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'ceil');\n\n  const inputs: CeilInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.ceil($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Ceil);\n}\nexport const ceil = op({ceil_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {ClipByValue, ClipByValueAttrs, ClipByValueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)\n * ```\n * @param x The input tensor.\n * @param clipValueMin Lower-bound of range to be clipped to.\n * @param clipValueMax Upper-bound of range to be clipped to.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction clipByValue_<T extends Tensor>(\n    x: T|TensorLike, clipValueMin: number, clipValueMax: number): T {\n  const $x = convertToTensor(x, 'x', 'clipByValue');\n  util.assert(\n      (clipValueMin <= clipValueMax),\n      () => `Error in clip: min (${clipValueMin}) must be ` +\n          `less than or equal to max (${clipValueMax}).`);\n\n  const inputs: ClipByValueInputs = {x: $x};\n  const attrs: ClipByValueAttrs = {clipValueMin, clipValueMax};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const res = backend.clip($x, clipValueMin, clipValueMax);\n        save([$x]);\n        return res;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, ClipByValue,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const clipByValue = op({clipByValue_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor1D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(3) = |r1, g1, b1|\n * B: shape(2) = |r2, g2|\n * C = tf.concat1d([A, B]) == |r1, g1, b1, r2, g2|\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @return The concatenated array.\n */\nfunction concat1d_(tensors: Array<Tensor1D|TensorLike>): Tensor1D {\n  return concat(tensors, 0 /* axis */);\n}\n\nexport const concat1d = op({concat1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor2D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *\n * B: shape(2, 3) = | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * C = tf.concat2d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *                  | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * if axis = 1:\n * C = shape(2, 6) = | r1, g1, b1, r3, g3, b3 |\n *                   | r2, g2, b2, r4, g4, b4 |\n *\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concatenate along.\n * @return The concatenated array.\n */\nfunction concat2d_(\n    tensors: Array<Tensor2D|TensorLike>, axis: number): Tensor2D {\n  return concat(tensors, axis);\n}\n\nexport const concat2d = op({concat2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor3D`s along an axis.\n * See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *\n * B: shape(2, 1, 3) = | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * C = tf.concat3d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *                     | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * if axis = 1:\n * C: shape(2, 2, 3) = | r1, g1, b1, r3, g3, b3 |\n *                     | r2, g2, b2, r4, g4, b4 |\n *\n * if axis = 2:\n * C = shape(2, 1, 6) = | r1, g1, b1, r3, g3, b3 |\n *                      | r2, g2, b2, r4, g4, b4 |\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat3d_(\n    tensors: Array<Tensor3D|TensorLike>, axis: number): Tensor3D {\n  return concat(tensors, axis);\n}\n\nexport const concat3d = op({concat3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor4D`s along an axis.\n * See `concat` for details.\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat4d_(\n    tensors: Array<Tensor4D|TensorLike>, axis: number): Tensor4D {\n  return concat(tensors, axis);\n}\n\nexport const concat4d = op({concat4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2D, Conv2DAttrs, Conv2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 2D convolution over the input x.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2d: input must be rank 4, but got rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  util.assert(\n      inDepth === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${inDepth}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        false, $dataFormat);\n    const res = backend.conv2d(x4D, $filter, convInfo);\n\n    save([x4D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DInputs = {x: x4D, filter: $filter};\n  const attrs:\n      Conv2DAttrs = {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv2D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2d = op({conv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D, Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 1D convolution over the input x.\n *\n * @param x The input tensor, of rank 3 or rank 2, of shape\n *     `[batch, width, inChannels]`. If rank 2, batch of 1 is assumed.\n * @param filter The filter, rank 3, of shape\n *     `[filterWidth, inDepth, outDepth]`.\n * @param stride The number of entries by which the filter is moved right at\n *     each step.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from \"NWC\", \"NCW\". Defaults to \"NWC\",\n *     the data is stored in the order of [batch, in_width, in_channels]. Only\n *     \"NWC\" is currently supported.\n * @param dilation The dilation rate in which we sample input values in\n *     atrous convolution. Defaults to `1`. If it is greater than 1, then\n *     stride must be `1`.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv1d_<T extends Tensor2D|Tensor3D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike, stride: number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv1d');\n  const $filter = convertToTensor(filter, 'filter', 'conv1d');\n\n  let x3D = $x as Tensor3D;\n  let reshapedTo3D = false;\n  if ($x.rank === 2) {\n    reshapedTo3D = true;\n    x3D = reshape($x, [1, $x.shape[0], $x.shape[1]]);\n  }\n\n  util.assert(\n      x3D.rank === 3,\n      () => `Error in conv1d: input must be rank 3, but got rank ${x3D.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in conv1d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv1d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x3D.shape[2] === $filter.shape[1],\n      () => `Error in conv1d: depth of input (${x3D.shape[2]}) must match ` +\n          `input depth for filter ${$filter.shape[1]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(stride, dilation),\n      () => 'Error in conv1D: Either stride or dilation must be 1. ' +\n          `Got stride ${stride} and dilation '${dilation}'`);\n  util.assert(\n      dataFormat === 'NWC',\n      () => `Error in conv1d: got dataFormat of ${\n          dataFormat} but only NWC is currently supported.`);\n\n  const filter4D = reshape(\n      $filter, [1, $filter.shape[0], $filter.shape[1], $filter.shape[2]]);\n  const input4D = reshape(x3D, [x3D.shape[0], 1, x3D.shape[1], x3D.shape[2]]);\n  const strides: [number, number] = [1, stride];\n  const dilations: [number, number] = [1, dilation];\n\n  const conv2dDataFormat = 'NHWC';\n\n  const res = conv2d(\n      (input4D as Tensor4D), (filter4D as Tensor4D), strides, pad,\n      conv2dDataFormat, dilations, dimRoundingMode);\n\n  if (reshapedTo3D) {\n    return reshape(res, [res.shape[2], res.shape[3]]) as T;\n  }\n\n  return reshape(res, [res.shape[0], res.shape[2], res.shape[3]]) as T;\n}\n\nexport const conv1d = op({conv1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs, Conv2DBackpropInputInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 2D convolution.\n *\n * @param xShape The shape of the input: [batch, height, width, inDepth].\n * If length of 3, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 4 or rank 3 of shape\n *   `[batch, outHeight, outWidth, outDepth]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction conv2DBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape4D = xShape as [number, number, number, number];\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n    xShape4D = [1, xShape[0], xShape[1], xShape[2]];\n  }\n\n  util.assert(\n      xShape4D.length === 4,\n      () =>\n          `Error in conv2dDerInput: inShape must be length 4, but got length ` +\n          `${xShape4D.length}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerInput: dy must be rank 4, but got ` +\n          `rank ${dy4D.rank}`);\n  util.assert(\n      filter.rank === 4,\n      () => `Error in conv2dDerInput: filter must be rank 4, but got ` +\n          `rank ${filter.rank}`);\n  const inDepth = dataFormat === 'NHWC' ? xShape4D[3] : xShape4D[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filter.shape[2],\n      () => `Error in conv2dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[2]}.`);\n  util.assert(\n      outDepth === filter.shape[3],\n      () => `Error in conv2dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[3]}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerInput: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const dilations = 1;\n\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        xShape4D, filter.shape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    const res = backend.conv2dDerInput(dy4D, filter, convInfo);\n\n    save([dy4D, filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DBackpropInputInputs = {dy: dy4D, filter};\n\n  const attrs: Conv2DBackpropInputAttrs =\n      {strides, pad, dataFormat, dimRoundingMode, inputShape: xShape4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      Conv2DBackpropInput, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2DBackpropInput = op({conv2DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv2DBackpropInput} from './conv2d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 2D convolution of an image, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 4 or rank 3, of shape\n *   `[batch, height, width, inDepth]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 4 or rank 3:\n *     `[batch, height, width, outDepth]`. If rank 3, batch of 1 is assumed.\n * @param strides The strides of the original convolution:\n *     `[strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n * @param dimRoundingMode The rounding mode used when computing output\n *    dimensions if pad is a number. If none is provided, it will not round\n *    and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2dTranspose_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv2dTranspose');\n\n  return conv2DBackpropInput(\n      outputShape, $x, $filter, strides, pad, 'NHWC', dimRoundingMode);\n}\n\nexport const conv2dTranspose = op({conv2dTranspose_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3D, Conv3DAttrs, Conv3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {eitherStridesOrDilationsAreOne} from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 3D convolution over the input x.\n *\n * @param x The input tensor, of rank 5 or rank 4, of shape\n *     `[batch, depth, height, width, channels]`. If rank 4,\n * batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inChannels, outChannels]`.\n *      inChannels must match between input and filter.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationDepth, dilationHeight,\n *     dilationWidth]` in which we sample input values across the height\n *     and width dimensions in atrous convolution. Defaults to `[1, 1, 1]`.\n *     If `dilations` is a single number, then\n *     `dilationDepth == dilationHeight == dilationWidth`. If it is greater\n *     than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    strides: [number, number, number]|number, pad: 'valid'|'same',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations: [number, number, number]|number = [1, 1, 1]): T {\n  const $x = convertToTensor(x, 'x', 'conv3d');\n  const $filter = convertToTensor(filter, 'filter', 'conv3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3d: input must be rank 5, but got rank ${x5D.rank}.`);\n  util.assert(\n      $filter.rank === 5,\n      () => `Error in conv3d: filter must be rank 5, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x5D.shape[4] === $filter.shape[3],\n      () => `Error in conv3d: depth of input (${x5D.shape[4]}) must match ` +\n          `input depth for filter ${$filter.shape[3]}.`);\n  util.assert(\n      eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv3D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in conv3d: got dataFormat of ${\n          dataFormat} but only NDHWC is currently supported.`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, $filter.shape, strides, dilations, pad);\n    const res = backend.conv3d(x5D, $filter, convInfo);\n\n    save([x5D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv3DInputs = {x: x5D, filter: $filter};\n\n  const attrs: Conv3DAttrs = {strides, pad, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3d = op({conv3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropInputAttrs, Conv3DBackpropInputInputs, Conv3DBackpropInputV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 3D convolution.\n *\n * @param xShape The shape of the input: [batch, depth, height, width,\n * in_channels]. If length of 4, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 5 or rank 4 of shape\n *   `[batch, outDepth, outHeight, outWidth, in_channels]`.\n * If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n */\nfunction conv3DBackpropInput_<T extends Tensor4D|Tensor5D>(\n    xShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    dy: T, filter: Tensor5D, strides: [number, number, number]|number,\n    pad: 'valid'|'same'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape5D = xShape as [number, number, number, number, number];\n  let dy5D = dy as Tensor5D;\n  let reshapedTo5D = false;\n  if (dy.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n    xShape5D = [1, xShape[0], xShape[1], xShape[2], xShape[3]];\n  }\n\n  const inDepth = xShape5D[4];\n  const outDepth = dy5D.shape[4];\n  util.assert(\n      xShape5D.length === 5,\n      () =>\n          `Error in conv3dDerInput: inShape must be length 5, but got length ` +\n          `${xShape5D.length}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerInput: dy must be rank 5, but got ` +\n          `rank ${dy5D.rank}`);\n  util.assert(\n      filter.rank === 5,\n      () => `Error in conv3dDerInput: filter must be rank 5, but got ` +\n          `rank ${filter.rank}`);\n  util.assert(\n      inDepth === filter.shape[3],\n      () => `Error in conv3dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[3]}.`);\n  util.assert(\n      outDepth === filter.shape[4],\n      () => `Error in conv3dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[4]}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        xShape5D, filter.shape, strides, dilations, pad);\n\n    return backend.conv3dDerInput(dy5D, filter, convInfo);\n  };\n\n  const inputs: Conv3DBackpropInputInputs = {dy: dy5D};\n\n  const attrs: Conv3DBackpropInputAttrs = {pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, Conv3DBackpropInputV2,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3DBackpropInput = op({conv3DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D, Tensor5D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv3DBackpropInput} from './conv3d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 3D convolution of a volume, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 5 or rank 4, of shape\n *   `[batch, depth, height, width, inDepth]`. If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[depth, filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 5 or rank 4:\n *     `[batch, depth, height, width, outDepth]`. If rank 3, batch of 1 is\n *    assumed.\n * @param strides The strides of the original convolution:\n *     `[strideDepth, strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3dTranspose_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    outputShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): T {\n  const $x = convertToTensor(x, 'x', 'conv3dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv3dTranspose');\n\n  return conv3DBackpropInput(outputShape, $x, $filter, strides, pad);\n}\n\nexport const conv3dTranspose = op({conv3dTranspose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cos, CosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes cos of the input `tf.Tensor` element-wise: `cos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.cos().print();  // or tf.cos(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cos');\n\n  const inputs: CosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cos);\n}\nexport const cos = op({cos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cosh, CoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic cos of the input `tf.Tensor` element-wise: `cosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.cosh().print();  // or tf.cosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cosh');\n  const inputs: CoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cosh);\n}\nexport const cosh = op({cosh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Cumsum, CumsumAttrs, CumsumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {getAxesPermutation, getInnerMostAxes, getUndoAxesPermutation} from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the cumulative sum of a `tf.Tensor` along `axis`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4]);\n * x.cumsum().print();\n * ```\n * ```js\n * const x = tf.tensor([[1, 2], [3, 4]]);\n * x.cumsum().print();\n * ```\n *\n * @param x The input tensor to be summed.\n * @param axis The axis along which to sum. Optional. Defaults to 0.\n * @param exclusive Whether to perform exclusive cumulative sum. Optional.\n *     Defaults to false. If set to true then the sum of each tensor entry\n *     does not include its own value, but only the values previous to it\n *     along the specified axis.\n * @param reverse Whether to sum in the opposite direction. Optional.\n *     Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Scan'}\n */\nfunction cumsum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis = 0, exclusive = false, reverse = false): T {\n  const $x = convertToTensor(x, 'x', 'cumsum');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const permutation = getAxesPermutation([axis], $x.rank);\n        let permutedX = $x;\n        if (permutation != null) {\n          permutedX = transpose($x, permutation);\n        }\n        const permutedAxis = getInnerMostAxes(1, $x.rank)[0];\n        let value = backend.cumsum(permutedX, permutedAxis, exclusive, reverse);\n        save([$x]);\n\n        if (permutation != null) {\n          const reversePermutation = getUndoAxesPermutation(permutation);\n          value = transpose(value, reversePermutation);\n        }\n        return value;\n      };\n\n  const inputs: CumsumInputs = {x: $x};\n  const attrs: CumsumAttrs = {axis, exclusive, reverse};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Cumsum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const cumsum = op({cumsum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthToSpace, DepthToSpaceAttrs, DepthToSpaceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Rearranges data from depth into blocks of spatial data. More specifically,\n * this op outputs a copy of the input tensor where values from the `depth`\n * dimension are moved in spatial blocks to the `height` and `width` dimensions.\n * The attr `blockSize` indicates the input block size and how the data is\n * moved.\n *\n *  - Chunks of data of size `blockSize * blockSize` from depth are rearranged\n * into non-overlapping blocks of size `blockSize x blockSize`\n *\n *  - The width the output tensor is `inputWidth * blockSize`, whereas the\n * height is `inputHeight * blockSize`\n *\n *  - The Y, X coordinates within each block of the output image are determined\n * by the high order component of the input channel index\n *\n *  - The depth of the input tensor must be divisible by `blockSize *\n * blockSize`\n *\n * The `dataFormat` attr specifies the layout of the input and output tensors\n * with the following options: \"NHWC\": [ `batch, height, width, channels` ]\n * \"NCHW\": [ `batch, channels, height, width` ]\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 1, 1, 4]);\n * const blockSize = 2;\n * const dataFormat = \"NHWC\";\n *\n * tf.depthToSpace(x, blockSize, dataFormat).print();\n * ```\n *\n * @param x The input tensor of rank 4\n * @param blockSIze  An `int` that is `>= 2`. The size of the spatial block\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\"\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction depthToSpace_(\n    x: Tensor4D|TensorLike4D, blockSize: number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'depthToSpace') as Tensor4D;\n\n  const inputHeight = (dataFormat === 'NHWC') ? $x.shape[1] : $x.shape[2];\n  const inputWidth = (dataFormat === 'NHWC') ? $x.shape[2] : $x.shape[3];\n  const inputDepth = (dataFormat === 'NHWC') ? $x.shape[3] : $x.shape[1];\n\n  util.assert(\n      inputHeight * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputHeight} and ${blockSize}  for depthToSpace with input shape\n    ${$x.shape}`);\n\n  util.assert(\n      inputWidth * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputWidth} and ${blockSize} for depthToSpace with input shape\n        ${$x.shape}`);\n\n  util.assert(\n      (inputDepth % (blockSize * blockSize) === 0),\n      () => `Dimension size must be evenly divisible by ${\n          blockSize * blockSize} but is ${\n          inputDepth} for depthToSpace with input shape ${$x.shape}`);\n\n  const forward: ForwardFunc<Tensor4D> = backend =>\n      backend.depthToSpace($x, blockSize, dataFormat);\n\n  const inputs: DepthToSpaceInputs = {x: $x};\n  const attrs: DepthToSpaceAttrs = {blockSize, dataFormat};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      DepthToSpace, attrs as {} as NamedAttrMap);\n}\n\nexport const depthToSpace = op({depthToSpace_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs, DepthwiseConv2dNativeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Depthwise 2D convolution.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction depthwiseConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in depthwiseConv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in depthwiseConv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1];\n    }\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in depthwiseConv2d: Either strides or dilations must be ' +\n            `1. Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        true /* depthwise */);\n    const res = backend.depthwiseConv2D(x4D, $filter, convInfo);\n    save([x4D, $filter]);\n    return res;\n  };\n\n  const inputs: DepthwiseConv2dNativeInputs = {x: x4D, filter: $filter};\n  const attrs: DepthwiseConv2dNativeAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      DepthwiseConv2dNative, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2d = op({depthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Diag, DiagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a diagonal tensor with a given diagonal values.\n *\n * Given a diagonal, this operation returns a tensor with the diagonal and\n * everything else padded with zeros.\n *\n * Assume the input has dimensions `[D1,..., Dk]`, then the output is a tensor\n * of rank 2k with dimensions `[D1,..., Dk, D1,..., Dk]`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * tf.diag(x).print()\n * ```\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4, 5, 6, 6, 8], [4, 2])\n *\n * tf.diag(x).print()\n * ```\n * @param x The input tensor.\n */\nfunction diag_(x: Tensor): Tensor {\n  const $x = convertToTensor(x, 'x', 'diag');\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const flat = reshape($x, [$x.size]);\n    const result = backend.diag(flat);\n    const outShape = [...x.shape, ...x.shape];\n\n    return reshape(result, outShape);\n  };\n\n  const inputs: DiagInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Diag);\n}\n\nexport const diag = op({diag_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DAttrs, Dilation2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the grayscale dilation over the input `x`.\n *\n * @param x The input tensor, rank 3 or rank 4 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter tensor, rank 3, of shape\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat Specify the data format of the input and output data.\n *      Defaults to 'NHWC'. Only 'NHWC' is currently supported. With the\n *      default format \"NHWC\", the data is stored in the order of: [batch,\n *      height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     for atrous morphological dilation. Defaults to `[1, 1]`. If `dilations`\n *     is a single number, then `dilationHeight == dilationWidth`. If it is\n *     greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction dilation2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same',\n    dilations: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'dilation2d');\n  const $filter = convertToTensor(filter, 'filter', 'dilation2d');\n\n  util.assert(\n      $x.rank === 3 || $x.rank === 4,\n      () => `Error in dilation2d: input must be rank 3 or 4, but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in dilation2d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in dilation2d: Only NHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n    reshapedTo4D = true;\n  }\n\n  const inputs: Dilation2DInputs = {x: x4D, filter: $filter};\n  const attrs: Dilation2DAttrs = {strides, pad, dilations};\n\n  const res = ENGINE.runKernel(\n                  Dilation2D, inputs as {} as NamedTensorMap,\n                  attrs as {} as NamedAttrMap) as T;\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res;\n}\n\nexport const dilation2d = op({dilation2d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Returns the dimensions in the input shape that are broadcasted to\n * produce the provided output shape.\n *\n * The returned dimensions are 0-indexed and sorted. An example:\n * inShape = [4, 1, 3]\n * outShape = [5, 4, 3, 3]\n * result = [1]. Dimension 1 (2nd dimension of input) gets broadcasted 1 => 3.\n */\nexport function getBroadcastDims(\n    inShape: number[], outShape: number[]): number[] {\n  const inRank = inShape.length;\n  const dims: number[] = [];\n  for (let i = 0; i < inRank; i++) {\n    const dim = inRank - 1 - i;\n    const a = inShape[dim] || 1;\n    const b = outShape[outShape.length - 1 - i] || 1;\n    if (b > 1 && a === 1) {\n      dims.unshift(dim);\n    }\n  }\n  return dims;\n}\n\n/**\n * Returns the axes in the output space that should be reduced to produce\n * the input space.\n */\nexport function getReductionAxes(\n    inShape: number[], outShape: number[]): number[] {\n  const result: number[] = [];\n  for (let i = 0; i < outShape.length; i++) {\n    const inDim = inShape[inShape.length - i - 1];\n    const outAxis = outShape.length - i - 1;\n    const outDim = outShape[outAxis];\n    if (inDim == null || (inDim === 1 && outDim > 1)) {\n      result.unshift(outAxis);\n    }\n  }\n  return result;\n}\n\nexport function assertAndGetBroadcastShape(\n    shapeA: number[], shapeB: number[]): number[] {\n  const result: number[] = [];\n  const l = Math.max(shapeA.length, shapeB.length);\n\n  for (let i = 0; i < l; i++) {\n    let a = shapeA[shapeA.length - i - 1];\n    if (a == null) {\n      a = 1;\n    }\n    let b = shapeB[shapeB.length - i - 1];\n    if (b == null) {\n      b = 1;\n    }\n    if (a === 1) {\n      result.unshift(b);\n    } else if (b === 1) {\n      result.unshift(a);\n    } else if (a !== b) {\n      const errMsg = `Operands could not be broadcast together with shapes ` +\n          `${shapeA} and ${shapeB}.`;\n      throw Error(errMsg);\n    } else {\n      result.unshift(a);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Equal, EqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a == b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.equal(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction equal_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'equal');\n  let $b = convertToTensor(b, 'b', 'equal');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.equal($a, $b);\n\n  const inputs: EqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, Equal) as T;\n}\n\nexport const equal = op({equal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SelectV2, SelectV2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch} from '../util';\n\nimport {broadcastTo} from './broadcast_to';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the elements, either `a` or `b` depending on the `condition`.\n *\n * If the condition is true, select from `a`, otherwise select from `b`.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const a = tf.tensor1d([1 , 2, 3]);\n * const b = tf.tensor1d([-1, -2, -3]);\n *\n * a.where(cond, b).print();\n * ```\n *\n * @param condition The input condition. Must be of dtype bool.\n * @param a If `condition` is rank 1, `a` may have a higher rank but\n *     its first dimension must match the size of `condition`.\n * @param b A tensor with the same dtype as `a` and with shape that is\n *     compatible with `a`.\n * @return A tensor with same dtype as `a` and `b`, and shape that is\n *     broadcastable from `a` and `b`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction where_<T extends Tensor>(\n    condition: Tensor|TensorLike, a: T|TensorLike, b: T|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'where');\n  const $b = convertToTensor(b, 'b', 'where');\n  const $condition = convertToTensor(condition, 'condition', 'where', 'bool');\n  // TODO: move this logic to forward function when the broadcastTo op is\n  // implemented in WASM.\n  // Find the broadcastable shape for $a and $b.\n  const broadcastShape = assertAndGetBroadcastShape($a.shape, $b.shape);\n  const $broadcastedA = broadcastTo($a, broadcastShape);\n  const $broadcastedB = broadcastTo($b, broadcastShape);\n  if ($condition.rank === 1) {\n    // If condition rank is 1, then the first dimension must match the size of\n    // condition.\n    assert(\n        $condition.shape[0] === $a.shape[0],\n        () => 'The first dimension of `a` must match the size of `condition`.');\n  }\n\n  if ($condition.rank !== 1) {\n    // A must have the same shape as condition.\n    assertShapesMatch(\n        $condition.shape, $broadcastedB.shape, 'Error in where: ');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.select($condition, $broadcastedA, $broadcastedB);\n    save([$condition]);\n    return res;\n  };\n  const inputs: SelectV2Inputs = {\n    condition: $condition,\n    t: $broadcastedA,\n    e: $broadcastedB\n  };\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* gradient */,\n             SelectV2) as T;\n}\n\nexport const where = op({where_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {ZerosLike, ZerosLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.zerosLike(x).print();\n * ```\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction zerosLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'zerosLike');\n  const inputs: ZerosLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             backend => backend.zerosLike($x), inputs as {} as NamedTensorMap,\n             null /* grad */, ZerosLike) as T;\n}\nexport const zerosLike = op({zerosLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {div} from './div';\nimport {equal} from './equal';\nimport {op} from './operation';\nimport {where} from './where';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting. Return 0\n * if denominator is 0.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n * const c = tf.tensor1d([0, 0, 0, 0]);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n * const c = tf.scalar(0);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction divNoNan_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  // TODO: Make this into its own kernel.\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const divResult = div($a, $b);\n  const zeros = zerosLike(divResult);\n  const bEqualsZero = equal($b, zeros);\n  return where(bEqualsZero, zeros, divResult) as T;\n}\n\nexport const divNoNan = op({divNoNan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor,} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices and/or vectors, `t1` and `t2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor2d([[1, 2], [3, 4]]);\n * const c = tf.tensor2d([[1, 2, 3], [4, 5, 6]]);\n *\n * a.dot(b).print();  // or tf.dot(a, b)\n * b.dot(a).print();\n * b.dot(c).print();\n * ```\n * @param t1 The first tensor in the dot operation.\n * @param t2 The second tensor in the dot operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction dot_(t1: Tensor|TensorLike, t2: Tensor|TensorLike): Tensor {\n  const $t1 = convertToTensor(t1, 't1', 'dot');\n  const $t2 = convertToTensor(t2, 't2', 'dot');\n\n  util.assert(\n      ($t1.rank === 1 || $t1.rank === 2) && ($t2.rank === 1 || $t2.rank === 2),\n      () => `Error in dot: inputs must all be rank 1 or 2, but got ranks ` +\n          `${$t1.rank} and ${$t2.rank}.`);\n\n  const t1Inner = ($t1.rank === 1 ? $t1.size : $t1.shape[1]);\n  const t2Inner = ($t2.rank === 1 ? $t2.size : $t2.shape[0]);\n\n  util.assert(\n      t1Inner === t2Inner,\n      () => `Error in dot: inner dimensions of inputs must match, but got ` +\n          `${t1Inner} and ${t2Inner}.`);\n\n  if ($t1.rank === 1 && $t2.rank === 1) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, []);\n  } else if ($t1.rank === 1 && $t2.rank === 2) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else if ($t1.rank === 2 && $t2.rank === 1) {\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul($t1, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else {\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul($t1, t22D);\n    return t1t2;\n  }\n}\n\nexport const dot = op({dot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential linear element-wise: `x > 0 ? e ^ x - 1 : 0`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 1, -3, 2]);\n *\n * x.elu().print();  // or tf.elu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction elu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'elu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.elu($x);\n    save([y]);\n    return y;\n  };\n\n  const inputs: EluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Elu) as\n      T;\n}\n\nexport const elu = op({elu_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Erf, ErfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes gause error function of the input `tf.Tensor` element-wise:\n * `erf(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.erf().print(); // or tf.erf(x);\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction erf_<T extends Tensor>(x: T|TensorLike): T {\n  let $x = convertToTensor(x, 'x', 'erf');\n  util.assert(\n      $x.dtype === 'int32' || $x.dtype === 'float32',\n      () => 'Input dtype must be `int32` or `float32`.');\n\n  if ($x.dtype === 'int32') {\n    $x = cast($x, 'float32');\n  }\n\n  const inputs: ErfInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.erf($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Erf);\n}\nexport const erf = op({erf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Exp, ExpInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.exp().print();  // or tf.exp(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction exp_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'exp');\n\n  const inputs: ExpInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.exp($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Exp);\n}\nexport const exp = op({exp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension\n * into the tensor's shape.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const axis = 1;\n * x.expandDims(axis).print();\n * ```\n *\n * @param x The input tensor whose dimensions to be expanded.\n * @param axis The dimension index at which to insert shape of `1`. Defaults\n *     to 0 (the first dimension).\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction expandDims_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'expandDims', parseAs);\n\n  util.assert(axis <= $x.rank, () => 'Axis must be <= rank of the tensor');\n  const newShape = $x.shape.slice();\n  if (axis < 0) {\n    // Negative value is counted from the tail of rank.\n    util.assert(\n        -($x.rank + 1) <= axis,\n        () => `Axis must be in the interval [${- ($x.rank + 1)}, ${$x.rank}]`);\n    axis = $x.rank + axis + 1;\n  }\n  newShape.splice(axis, 0, 1);\n  return reshape($x, newShape) as T;\n}\n\nexport const expandDims = op({expandDims_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Expm1, Expm1Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` minus one element-wise.\n * `e ^ x - 1`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.expm1().print();  // or tf.expm1(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction expm1_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'expm1');\n\n  const inputs: Expm1Inputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.expm1($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Expm1);\n}\nexport const expm1 = op({expm1_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Tile, TileAttrs, TileInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Construct a tensor by repeating it the number of times given by reps.\n *\n * This operation creates a new tensor by replicating `input` `reps`\n * times. The output tensor's i'th dimension has `input.shape[i] *\n * reps[i]` elements, and the values of `input` are replicated\n * `reps[i]` times along the i'th dimension. For example, tiling\n * `[a, b, c, d]` by `[2]` produces `[a, b, c, d, a, b, c, d]`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n *\n * a.tile([2]).print();    // or a.tile([2])\n * ```\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.tile([1, 2]).print();  // or a.tile([1, 2])\n * ```\n * @param x The tensor to tile.\n * @param reps Determines the number of replications per dimension.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction tile_<T extends Tensor>(x: T|TensorLike, reps: number[]): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'tile', parseAs);\n  util.assert(\n      $x.rank === reps.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of reps ${reps}.`);\n\n  const forward: ForwardFunc<T> = (backend, save) => {\n    const res = backend.tile($x, reps);\n    save([$x]);\n    return res;\n  };\n\n  const inputsToSave = [$x];\n  const inputs: TileInputs = {x: $x};\n  const attrs: TileAttrs = {reps};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, Tile,\n      attrs as unknown as NamedAttrMap, inputsToSave);\n}\n\nexport const tile = op({tile_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {DataType} from '../types';\n\nimport {buffer} from './buffer';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {tile} from './tile';\n\n/**\n * Create an identity matrix.\n *\n * @param numRows Number of rows.\n * @param numColumns Number of columns. Defaults to `numRows`.\n * @param batchShape If provided, will add the batch shape to the beginning\n *   of the shape of the returned `tf.Tensor` by repeating the identity\n *   matrix.\n * @param dtype Data type.\n * @returns Identity matrix of the specified size and data type, possibly\n *   with batch repetition if `batchShape` is specified.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction eye_(\n    numRows: number, numColumns?: number,\n    batchShape?:\n        [\n          number\n        ]|[number,\n           number]|[number, number, number]|[number, number, number, number],\n    dtype: DataType = 'float32'): Tensor2D {\n  if (numColumns == null) {\n    numColumns = numRows;\n  }\n  const buff = buffer([numRows, numColumns], dtype);\n  const n = numRows <= numColumns ? numRows : numColumns;\n  for (let i = 0; i < n; ++i) {\n    buff.set(1, i, i);\n  }\n  const out: Tensor2D = reshape(buff.toTensor(), [numRows, numColumns]);\n  if (batchShape == null) {\n    return out;\n  } else {\n    if (batchShape.length === 1) {\n      return tile(expandDims(out, 0), [batchShape[0], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 2) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(\n                 expandDims(expandDims(out, 0), 0),\n                 [batchShape[0], batchShape[1], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 3) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(expandDims(expandDims(expandDims(out, 0), 0), 0), [\n               batchShape[0], batchShape[1], batchShape[2], 1, 1\n             ]) as Tensor2D;\n    } else {\n      throw new Error(\n          `eye() currently supports only 1D and 2D ` +\n          // tslint:disable-next-line:no-any\n          `batchShapes, but received ${(batchShape as any).length}D.`);\n    }\n  }\n}\n\nexport const eye = op({eye_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Fill, FillAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\n/**\n * Creates a `tf.Tensor` filled with a scalar value.\n *\n * ```js\n * tf.fill([2, 2], 4).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param value The scalar value to fill the tensor with.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n * 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction fill<R extends Rank>(\n    shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n  const attrs: FillAttrs = {shape, value, dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.fill(shape, value, dtype), {}, null, Fill,\n      attrs as {} as NamedAttrMap);\n}\n\nexport {fill};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Floor, FloorInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes floor of input `tf.Tensor` element-wise: `floor(x)`.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.floor().print();  // or tf.floor(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction floor_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'floor');\n\n  const inputs: FloorInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.floor($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Floor);\n}\nexport const floor = op({floor_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inputs of size above this threshold will be parallelized by calling multiple\n * shader programs.\n */\nimport {nearestDivisor} from '../util';\n\nexport const PARALLELIZE_THRESHOLD = 30;\n\nexport interface ReduceInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  outSize: number;\n}\n\nexport function computeOptimalWindowSize(inSize: number): number {\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    return inSize;\n  }\n  return nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {nearestDivisor} from '../util';\n\nimport {PARALLELIZE_THRESHOLD} from './reduce_util';\n\nexport interface SegOpInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  numSegments: number;\n}\n\nexport function segOpComputeOptimalWindowSize(\n    inSize: number, numSegments: number): number {\n  let done = false;\n  let res;\n\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    res = inSize;\n    done = true;\n  } else {\n    res = nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n  }\n\n  while (!done) {\n    if (res > numSegments || res === inSize) {\n      done = true;\n    } else {\n      res = nearestDivisor(inSize, res + 1);\n    }\n  }\n  return res;\n}\n\nexport function computeOutShape(\n    aShape: number[], axis: number, numSegments: number): number[] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (dim !== axis) {\n      outShape.push(aShape[dim]);\n    } else {\n      outShape.push(numSegments);\n    }\n  }\n  return outShape;\n}\n\nexport interface GatherOpShapeInfo {\n  batchSize: number;\n  sliceSize: number;\n  dimSize: number;\n  outputShape: number[];\n}\nexport function collectGatherOpShapeInfo(\n    x: Tensor, indices: Tensor, axis: number): GatherOpShapeInfo {\n  const dimSize = x.shape[axis];\n\n  const outputShape: number[] = [];\n  let batchSize = 1;\n  let sliceSize = 1;\n  for (let i = 0; i < axis; i++) {\n    outputShape.push(x.shape[i]);\n    batchSize *= x.shape[i];\n  }\n\n  for (let i = 0; i < indices.rank; i++) {\n    outputShape.push(indices.shape[i]);\n  }\n\n  for (let i = axis + 1; i < x.rank; i++) {\n    outputShape.push(x.shape[i]);\n    sliceSize *= x.shape[i];\n  }\n\n  return {batchSize, sliceSize, dimSize, outputShape};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherV2, GatherV2Attrs, GatherV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {collectGatherOpShapeInfo} from './segment_util';\n\n/**\n * Gather slices from tensor `x`'s axis `axis` according to `indices`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const indices = tf.tensor1d([1, 3, 3], 'int32');\n *\n * x.gather(indices).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const indices = tf.tensor1d([1, 1, 0], 'int32');\n *\n * x.gather(indices).print();\n * ```\n * @param x The input tensor whose slices to be gathered.\n * @param indices The indices of the values to extract.\n * @param axis The axis over which to select values. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction gather_<T extends Tensor>(\n    x: T|TensorLike, indices: Tensor|TensorLike, axis = 0): T {\n  const $x = convertToTensor(x, 'x', 'gather');\n  const $indices = convertToTensor(indices, 'indices', 'gather', 'int32');\n\n  const inputs: GatherV2Inputs = {x: $x, indices: $indices};\n  const attrs: GatherV2Attrs = {axis};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const parsedAxis = parseAxisParam(axis, $x.shape)[0];\n    const shapeInfo = collectGatherOpShapeInfo($x, $indices, parsedAxis);\n\n    const res =\n        backend.gather($x, reshape($indices, [$indices.size]), parsedAxis);\n    save([$x, $indices]);\n\n    return reshape(res, shapeInfo.outputShape);\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, GatherV2,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const gather = op({gather_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Greater, GreaterInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a > b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greater(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greater_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greater');\n  let $b = convertToTensor(b, 'b', 'greater');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.greater($a, $b);\n\n  const inputs: GreaterInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             Greater) as T;\n}\n\nexport const greater = op({greater_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GreaterEqual, GreaterEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a >= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greaterEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greaterEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greaterEqual');\n  let $b = convertToTensor(b, 'b', 'greaterEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.greaterEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: GreaterEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             GreaterEqual) as T;\n}\n\nexport const greaterEqual = op({greaterEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Imag, ImagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n/**\n * Returns the imaginary part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the imaginary part of each element in input considered as a complex number.\n * If input is real, a tensor of all zeros is returned.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.imag(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction imag_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'imag');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.imag($input);\n  };\n\n  const inputs: ImagInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Imag) as T;\n}\n\nexport const imag = op({imag_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsFinite, IsFiniteInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are finite.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isFinite().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isFinite_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isFinite');\n\n  const inputs: IsFiniteInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isFinite($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsFinite);\n}\nexport const isFinite = op({isFinite_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsInf, IsInfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are Infinity or -Infinity.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isInf().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isInf_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isInf');\n\n  const inputs: IsInfInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isInf($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsInf);\n}\nexport const isInf = op({isInf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsNan, IsNanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * RReturns which elements of x are NaN.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isNaN().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isNaN_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isNaN');\n  const inputs: IsNanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.isNaN($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsNan);\n}\nexport const isNaN = op({isNaN_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Maximum, MaximumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the max of a and b (`a > b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `tf.maximumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * ```js\n * // Broadcast maximum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction maximum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'maximum');\n  let $b = convertToTensor(b, 'b', 'maximum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.maximum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: MaximumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Maximum) as T;\n}\n\nexport const maximum = op({maximum_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar} from '../tensor';\nimport {DataType} from '../types';\nimport {isTypedArray} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-0 `tf.Tensor` (scalar) with the provided value and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.scalar` as it makes the code more readable.\n *\n * ```js\n * tf.scalar(3.14).print();\n * ```\n *\n * @param value The value of the scalar.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function scalar(\n    value: number|boolean|string|Uint8Array, dtype?: DataType): Scalar {\n  if (((isTypedArray(value) && dtype !== 'string') || Array.isArray(value)) &&\n      dtype !== 'complex64') {\n    throw new Error(\n        'Error creating a new Scalar: value must be a primitive ' +\n        '(number|boolean|string)');\n  }\n  if (dtype === 'string' && isTypedArray(value) &&\n      !(value instanceof Uint8Array)) {\n    throw new Error(\n        'When making a scalar from encoded string, ' +\n        'the value must be `Uint8Array`.');\n  }\n  const shape: number[] = [];\n  const inferredShape: number[] = [];\n  return makeTensor(value, shape, inferredShape, dtype) as Scalar;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {maximum} from './maximum';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {scalar} from './scalar';\n\n/**\n * Computes leaky rectified linear element-wise.\n *\n * See\n * [http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf](\n *     http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf)\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.leakyRelu(0.1).print();  // or tf.leakyRelu(x, 0.1)\n * ```\n * @param x The input tensor.\n * @param alpha The scaling factor for negative values, defaults to 0.2.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction leakyRelu_<T extends Tensor>(x: T|TensorLike, alpha = 0.2): T {\n  const $x = convertToTensor(x, 'x', 'leakyRelu');\n  return maximum(mul(scalar(alpha), $x), $x);\n}\n\nexport const leakyRelu = op({leakyRelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Less, LessInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a < b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.less(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction less_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'less');\n  let $b = convertToTensor(b, 'b', 'less');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.less($a, $b);\n\n  const inputs: LessInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Less) as\n      T;\n}\n\nexport const less = op({less_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LessEqual, LessEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a <= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.lessEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction lessEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'lessEqual');\n  let $b = convertToTensor(b, 'b', 'lessEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.lessEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: LessEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LessEqual) as T;\n}\n\nexport const lessEqual = op({lessEqual_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LinSpace, LinSpaceAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor1D} from '../tensor';\n\n/**\n * Return an evenly spaced sequence of numbers over the given interval.\n *\n * ```js\n * tf.linspace(0, 9, 10).print();\n * ```\n * @param start The start value of the sequence.\n * @param stop The end value of the sequence.\n * @param num The number of values to generate.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function linspace(start: number, stop: number, num: number): Tensor1D {\n  if (num <= 0) {\n    throw new Error('The number of values should be positive.');\n  }\n\n  const attrs: LinSpaceAttrs = {start, stop, num};\n  return ENGINE.runKernelFunc(\n      backend => backend.linspace(start, stop, num), {} /* inputs */,\n      null /* grad */, LinSpace, attrs as {} as NamedAttrMap);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRN, LRNAttrs, LRNInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Normalizes the activation of a local neighborhood across or within\n * channels.\n *\n * @param x The input tensor. The 4-D input tensor is treated as a 3-D array\n *     of 1D vectors (along the last dimension), and each vector is\n *     normalized independently.\n * @param depthRadius The number of adjacent channels in the 1D normalization\n *     window.\n * @param bias A constant bias term for the basis.\n * @param alpha A scale factor, usually positive.\n * @param beta An exponent.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction localResponseNormalization_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const $x = convertToTensor(x, 'x', 'localResponseNormalization');\n  util.assert(\n      $x.rank === 4 || $x.rank === 3,\n      () => `Error in localResponseNormalization: x must be rank 3 or 4 but got\n               rank ${$x.rank}.`);\n  util.assert(\n      util.isInt(depthRadius),\n      () => `Error in localResponseNormalization: depthRadius must be an ` +\n          `integer but got depthRadius ${depthRadius}.`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.localResponseNormalization4D(\n        x4D, depthRadius, bias, alpha, beta);\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: LRNInputs = {x: x4D};\n\n  const attrs: LRNAttrs = {depthRadius, bias, alpha, beta};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, LRN,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  } else {\n    return res as T;\n  }\n}\n\nexport const localResponseNormalization = op({localResponseNormalization_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log, LogInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` element-wise: `ln(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E]);\n *\n * x.log().print();  // or tf.log(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log');\n\n  const inputs: LogInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log);\n}\nexport const log = op({log_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log1p, Log1pInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` plus one\n * element-wise: `ln(1 + x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E - 1]);\n *\n * x.log1p().print();  // or tf.log1p(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log1p_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log1p');\n\n  const inputs: Log1pInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log1p($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log1p);\n}\nexport const log1p = op({log1p_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {CustomGradientFunc, ENGINE} from './engine';\nimport {Scalar, Tensor, Variable} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {convertToTensor, convertToTensorArray} from './tensor_util_env';\nimport {TensorLike} from './types';\nimport * as util from './util';\n\n/**\n * Provided `f(x)`, returns another function `g(x, dy?)`, which gives the\n * gradient of `f(x)` with respect to `x`.\n *\n * If `dy` is provided, the gradient of `f(x).mul(dy).sum()` with respect to\n * `x` is computed instead. `f(x)` must take a single tensor `x` and return a\n * single tensor `y`. If `f()` takes multiple inputs, use `tf.grads` instead.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.grad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * g(x).print();\n * ```\n *\n * ```js\n * // f(x) = x ^ 3\n * const f = x => x.pow(tf.scalar(3, 'int32'));\n * // f'(x) = 3x ^ 2\n * const g = tf.grad(f);\n * // f''(x) = 6x\n * const gg = tf.grad(g);\n *\n * const x = tf.tensor1d([2, 3]);\n * gg(x).print();\n * ```\n *\n * @param f The function f(x), to compute gradient for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grad(f: (x: Tensor) => Tensor): (\n    x: TensorLike|Tensor, dy?: TensorLike|Tensor) => Tensor {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grad(f) must be a function');\n  return (x: TensorLike|Tensor, dy?: TensorLike|Tensor): Tensor => {\n    // x can be of any dtype, thus null as the last argument.\n    const $x = convertToTensor(x, 'x', 'tf.grad', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grad') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f($x), [$x], $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grad(f)(x, dy) must match the shape ' +\n                'returned by f(x)');\n      }\n      checkGrads(grads);\n      return grads[0];\n    });\n  };\n}\n\n/**\n * Provided `f(x1, x2,...)`, returns another function `g([x1, x2,...], dy?)`,\n * which gives an array of gradients of `f()` with respect to each input\n * [`x1`,`x2`,...].\n *\n * If `dy` is passed when calling `g()`, the gradient of\n * `f(x1,...).mul(dy).sum()` with respect to each input is computed instead.\n * The provided `f` must take one or more tensors and return a single tensor\n * `y`. If `f()` takes a single input, we recommend using `tf.grad` instead.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df / da = b, df / db = a\n * const g = tf.grads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const [da, db] = g([a, b]);\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @param f The function `f(x1, x2,...)` to compute gradients for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grads(f: (...args: Tensor[]) => Tensor): (\n    args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike) => Tensor[] {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grads(f) must be a function');\n  return (args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike): Tensor[] => {\n    util.assert(\n        Array.isArray(args),\n        () => 'The args passed in grads(f)(args) must be an array ' +\n            'of `Tensor`s or `TensorLike`s');\n    // args can be of any dtype, thus null as the last argument.\n    const $args = convertToTensorArray(args, 'args', 'tf.grads', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grads') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f(...$args), $args, $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grads(f)([x1,...], dy) must ' +\n                'match the shape returned by f([x1,...])');\n      }\n      checkGrads(grads);\n      return grads;\n    });\n  };\n}\n\n/**\n * Like `tf.grad`, but also returns the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grad: The gradient of `f(x)` w.r.t `x` (result of `tf.grad`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.valueAndGrad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * const {value, grad} = g(x);\n *\n * console.log('value');\n * value.print();\n * console.log('grad');\n * grad.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrad<I extends Tensor, O extends Tensor>(f: (x: I) => O): (\n    x: I, dy?: O) => {\n  value: O;\n  grad: I;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrad(f) must be a function');\n  return (x: I, dy?: O) => {\n    util.assert(\n        x instanceof Tensor,\n        () => 'The x passed in valueAndGrad(f)(x) must be a tensor');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrad(f)(x, dy) must be a tensor');\n    const {grads, value} = ENGINE.gradients(() => f(x), [x], dy);\n    checkGrads(grads);\n    return {grad: grads[0] as I, value};\n  };\n}\n\n/**\n * Like `tf.grads`, but returns also the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grads: The gradients of `f()` w.r.t each input (result of `tf.grads`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df/da = b, df/db = a\n * const g = tf.valueAndGrads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const {value, grads} = g([a, b]);\n *\n * const [da, db] = grads;\n *\n * console.log('value');\n * value.print();\n *\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrads<O extends Tensor>(f: (...args: Tensor[]) => O): (\n    args: Tensor[], dy?: O) => {\n  grads: Tensor[];\n  value: O;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrads(f) must be a function');\n  return (args: Tensor[], dy?: O) => {\n    util.assert(\n        Array.isArray(args) && args.every(arg => arg instanceof Tensor),\n        () => 'The args passed in valueAndGrads(f)(args) must be array of ' +\n            'tensors');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrads(f)(args, dy) must be a tensor');\n    const res = ENGINE.gradients(() => f(...args), args, dy);\n    if (dy != null) {\n      util.assertShapesMatch(\n          res.value.shape, dy.shape,\n          'The shape of dy passed in valueAndGrads(f)([x1,...], dy) must ' +\n              'match the shape returned by f([x1,...])');\n    }\n    checkGrads(res.grads);\n    return res;\n  };\n}\n\n/**\n * Computes and returns the gradient of f(x) with respect to the list of\n * trainable variables provided by `varList`. If no list is provided, it\n * defaults to all trainable variables.\n *\n * ```js\n * const a = tf.variable(tf.tensor1d([3, 4]));\n * const b = tf.variable(tf.tensor1d([5, 6]));\n * const x = tf.tensor1d([1, 2]);\n *\n * // f(a, b) = a * x ^ 2 + b * x\n * const f = () => a.mul(x.square()).add(b.mul(x)).sum();\n * // df/da = x ^ 2, df/db = x\n * const {value, grads} = tf.variableGrads(f);\n *\n * Object.keys(grads).forEach(varName => grads[varName].print());\n * ```\n *\n * @param f The function to execute. f() should return a scalar.\n * @param varList The list of variables to compute the gradients with respect\n *     to. Defaults to all trainable variables.\n * @returns An object with the following keys and values:\n *   - `value`: The value of the function `f`.\n *   - `grads`: A map from the names of the variables to the gradients.\n *     If the `varList` argument is provided explicitly and contains a subset of\n *     non-trainable variables, this map in the return value will contain keys\n *     that map the names of the non-trainable variables to `null`.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction variableGrads(f: () => Scalar, varList?: Variable[]):\n    {value: Scalar, grads: NamedTensorMap} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in variableGrads(f) must be a function');\n  util.assert(\n      varList == null ||\n          Array.isArray(varList) && varList.every(v => v instanceof Variable),\n      () =>\n          'The varList passed in variableGrads(f, varList) must be an array ' +\n          'of variables');\n\n  const specifiedVarList = varList != null;\n  if (!specifiedVarList) {\n    // Get all of the trainable variables.\n    varList = [];\n    for (const varName in ENGINE.registeredVariables) {\n      varList.push(ENGINE.registeredVariables[varName]);\n    }\n  }\n\n  const specifiedNonTrainable: Variable[] =\n      specifiedVarList ? varList.filter(variable => !variable.trainable) : null;\n\n  // Prune non-trainable variables.\n  const originalVarCount = varList.length;\n  varList = varList.filter(variable => variable.trainable);\n  util.assert(\n      varList.length > 0,\n      () => `variableGrads() expects at least one of the input variables to ` +\n          `be trainable, but none of the ${originalVarCount} variables is ` +\n          `trainable.`);\n\n  const allowNoGradients = true;\n  const {value, grads} = ENGINE.gradients(f, varList, null, allowNoGradients);\n\n  util.assert(\n      grads.some(g => g != null),\n      () => 'Cannot find a connection between any variable and the result of ' +\n          'the loss function y=f(x). Please make sure the operations that ' +\n          'use variables are inside the function f passed to minimize().');\n  util.assert(\n      value.rank === 0,\n      () => `The f passed in variableGrads(f) must return a scalar, but it ` +\n          `returned a rank-${value.rank} tensor`);\n\n  const namedGrads: NamedTensorMap = {};\n  varList.forEach((v, i) => {\n    if (grads[i] != null) {\n      namedGrads[v.name] = grads[i];\n    }\n  });\n  if (specifiedNonTrainable != null) {\n    // If varList is explicitly provided and contains non-trainable values,\n    // add them to the returned gradients with `null` values.\n    specifiedNonTrainable.forEach(v => namedGrads[v.name] = null);\n  }\n  return {value, grads: namedGrads};\n}\n\n/**\n * Overrides the gradient computation of a function `f`.\n *\n * Takes a function\n * `f(...inputs, save) => {value: Tensor, gradFunc: (dy, saved) => Tensor[]}`\n * and returns another function `g(...inputs)` which takes the same inputs as\n * `f`. When called, `g` returns `f().value`. In backward mode, custom gradients\n * with respect to each input of `f` are computed using `f().gradFunc`.\n *\n * The `save` function passsed to `f` should be used for saving tensors needed\n * in the gradient. And the `saved` passed to the `gradFunc` is a\n * `NamedTensorMap`, which contains those saved tensor.\n *\n * ```js\n * const customOp = tf.customGrad((x, save) => {\n *   // Save x to make sure it's available later for the gradient.\n *   save([x]);\n *   // Override gradient of our custom x ^ 2 op to be dy * abs(x);\n *   return {\n *     value: x.square(),\n *     // Note `saved.x` which points to the `x` we saved earlier.\n *     gradFunc: (dy, saved) => [dy.mul(saved[0].abs())]\n *   };\n * });\n *\n * const x = tf.tensor1d([-1, -2, 3]);\n * const dx = tf.grad(x => customOp(x));\n *\n * console.log(`f(x):`);\n * customOp(x).print();\n * console.log(`f'(x):`);\n * dx(x).print();\n * ```\n *\n * @param f The function to evaluate in forward mode, which should return\n *     `{value: Tensor, gradFunc: (dy, saved) => Tensor[]}`, where `gradFunc`\n *     returns the custom gradients of `f` with respect to its inputs.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n    (...args: Tensor[]) => T {\n  return ENGINE.customGrad(f);\n}\n\nfunction checkGrads(grads: Tensor[]) {\n  const numNullGradients = grads.filter(g => g == null).length;\n  if (numNullGradients > 0) {\n    throw new Error(\n        `Cannot compute gradient of y=f(x) with respect to x. Make sure that\n    the f you passed encloses all operations that lead from x to y.`);\n  }\n}\n\nexport {\n  customGrad,\n  variableGrads,\n  valueAndGrad,\n  valueAndGrads,\n  grad,\n  grads,\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Negate, NegateInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes `-1 * x` element-wise.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, -2, 0], [2, 2]);\n *\n * x.neg().print();  // or tf.neg(x)\n * ```\n *\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction neg_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'neg');\n\n  const inputs: NegateInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.neg($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Negate);\n}\nexport const neg = op({neg_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softplus, SoftplusInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes softplus of the input `tf.Tensor` element-wise: `log(exp(x) + 1)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.softplus().print();  // or tf.softplus(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction softplus_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'softplus');\n\n  const inputs: SoftplusInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.softplus($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Softplus);\n}\nexport const softplus = op({softplus_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {mul} from './mul';\nimport {neg} from './neg';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {softplus} from './softplus';\n\n/**\n * Computes log sigmoid of the input `tf.Tensor` element-wise:\n * `logSigmoid(x)`. For numerical stability, we use `-tf.softplus(-x)`.\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.logSigmoid().print();  // or tf.logSigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction logSigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logSigmoid');\n\n  // Use a custom gradient to maintain previous implementation.\n  // There is no LogSigmoid kernel in TF so we can't use engine.runKernel\n  // directly\n  const customOp = customGrad((x: Tensor) => {\n    // TODO(yassogba) we can remove the chained softplus call here only\n    // after backends have modualrized softplus at which point we can call\n    // engine runKernel(..., Sotfplus, ...) directly.\n    const value = neg(softplus(neg(x)));\n\n    const gradFunc = (dy: T) => {\n      const derX = mul(dy, sigmoid(neg(x)));\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\nexport const logSigmoid = op({logSigmoid_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Max, MaxAttrs, MaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the maximum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.max().print();  // or tf.max(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.max(axis).print();  // or tf.max(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction max_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'max');\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = util.parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let maxInput = $x;\n        if (permutedAxes != null) {\n          maxInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, maxInput.rank);\n        }\n\n        const y = backend.max(maxInput, axes);\n        if (permutedAxes != null) {\n          maxInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape = axis_util.expandShapeToKeepDim(\n              res.shape, util.parseAxisParam(axis, $x.shape));\n          res = reshape(res, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MaxInputs = {x: $x};\n  const attrs: MaxAttrs = {reductionIndices: axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Max,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const max = op({max_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sub, SubInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Subtracts two `tf.Tensor`s element-wise, A - B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n *\n * ```js\n * // Broadcast subtract a with b.\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.scalar(5);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n * @param a The first `tf.Tensor` to subtract from.\n * @param b The second `tf.Tensor` to be subtracted. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction sub_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'sub');\n  let $b = convertToTensor(b, 'b', 'sub');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.subtract($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SubInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sub) as\n      T;\n}\n\nexport const sub = op({sub_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sum, SumAttrs, SumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the sum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If axes has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.sum().print();  // or tf.sum(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.sum(axis).print();  // or tf.sum(x, axis)\n * ```\n *\n * @param x The input tensor to compute the sum over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction sum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'sum');\n  if ($x.dtype === 'bool') {\n    $x = cast($x, 'int32');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.sum(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n    return value;\n  };\n\n  const inputs: SumInputs = {x: $x};\n  const attrs: SumAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const sum = op({sum_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LogSoftmax, LogSoftmaxAttrs, LogSoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log softmax.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param axis The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction logSoftmax_<T extends Tensor>(logits: T|TensorLike, axis = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'logSoftmax');\n\n  if (axis === -1) {\n    axis = $logits.rank - 1;\n  }\n  if (axis !== $logits.rank - 1) {\n    throw Error(\n        'Log Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and axis was ${axis}`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const keepDims = true;\n    const xMax = max(logits, axis, true);\n    const shifted = sub(logits, xMax);\n    const value =\n        sub(cast(shifted, 'float32'), log(sum(exp(shifted), axis, keepDims)));\n    save([value]);\n    return value;\n  };\n\n  const inputs: LogSoftmaxInputs = {logits: $logits};\n  const attrs: LogSoftmaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LogSoftmax, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const logSoftmax = op({logSoftmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {add} from './add';\nimport {expandShapeToKeepDim} from './axis_util';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log(sum(exp(elements across the reduction dimensions)).\n *\n * Reduces the input along the dimensions given in `axis`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.logSumExp().print();  // or tf.logSumExp(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.logSumExp(axis).print();  // or tf.logSumExp(a, axis)\n * ```\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. If null (the default),\n *     reduces all dimensions.\n * @param keepDims If true, retains reduced dimensions with length\n *     of 1. Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction logSumExp_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'logSumExp');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const xMax = max($x, axes, true /* keepDims */);\n  const a = sub($x, xMax);\n  const b = exp(a);\n  const c = sum(b, axes);\n  const d = log(c);\n  const res = add(reshape(xMax, d.shape), d);\n\n  if (keepDims) {\n    const newShape = expandShapeToKeepDim(res.shape, axes);\n    return reshape(res, newShape) as T;\n  }\n  return res as T;\n}\n\nexport const logSumExp = op({logSumExp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {LogicalAnd, LogicalAndInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a AND b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalAnd(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalAnd_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalAnd', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalAnd', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalAndInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalAnd($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalAnd) as T;\n}\n\nexport const logicalAnd = op({logicalAnd_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalNot, LogicalNotInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `NOT x` element-wise.\n *\n * ```js\n * const a = tf.tensor1d([false, true], 'bool');\n *\n * a.logicalNot().print();\n * ```\n *\n * @param x The input tensor. Must be of dtype 'bool'.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalNot_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logicalNot', 'bool');\n  const inputs: LogicalNotInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.logicalNot($x), inputs as {} as NamedTensorMap,\n      null /* grad */, LogicalNot);\n}\n\nexport const logicalNot = op({logicalNot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalOr, LogicalOrInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a OR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalOr(b).print();\n * ```\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalOr_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalOr', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalOr', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalOrInputs = {a: $a, b: $b};\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalOr($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalOr) as T;\n}\nexport const logicalOr = op({logicalOr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {logicalAnd} from './logical_and';\nimport {logicalNot} from './logical_not';\nimport {logicalOr} from './logical_or';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a XOR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalXor(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalXor_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalXor', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalXor', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  // x ^ y = (x | y) & ~(x & y)\n  return logicalAnd(logicalOr(a, b), logicalNot(logicalAnd(a, b)));\n}\n\nexport const logicalXor = op({logicalXor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool, MaxPoolAttrs, MaxPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D max pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction maxPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'maxPool');\n  const dilations = 1;\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in maxPool: input must be rank 4 but got rank ${x4D.rank}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    let y;\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      y = x4D.clone();\n    } else {\n      y = backend.maxPool(x4D, convInfo);\n    }\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: MaxPoolInputs = {x: x4D};\n\n  const attrs: MaxPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const maxPool = op({maxPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {MaxPool3D, MaxPool3DAttrs, MaxPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D max pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.maxPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number = [1, 1, 1],\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'maxPool3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in maxPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in maxPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n    const y = backend.maxPool3d(x5D, convInfo);\n    save([x5D, y]);\n    return y;\n  };\n\n  const inputs: MaxPool3DInputs = {x: x5D};\n\n  const attrs: MaxPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3d = op({maxPool3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the 2D max pooling of an image with Argmax index.\n * The indices in argmax are flattened, so that a maximum value at position `[b,\n * y, x, c]` becomes flattened index: `(y * width + x) * channels + c` if\n * include_batch_in_index is False; `((b * height + y) * width + x) * channels\n * +c` if include_batch_in_index is True.\n *\n * The indices returned are always in `[0, height) x [0, width)` before\n * flattening.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param includeBatchIndex Defaults to False. Whether to include batch\n *    dimension in flattened index of argmax.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPoolWithArgmax_<T extends Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    includeBatchInIndex = false): NamedTensorMap {\n  const $x = convertToTensor(x, 'x', 'maxPoolWithArgmax');\n\n  const inputs: MaxPoolWithArgmaxInputs = {x: $x};\n  const attrs:\n      MaxPoolWithArgmaxAttrs = {filterSize, strides, pad, includeBatchInIndex};\n\n  const result = ENGINE.runKernel(\n                     MaxPoolWithArgmax, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {result: result[0], indexes: result[1]};\n}\n\nexport const maxPoolWithArgmax = op({maxPoolWithArgmax_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeZerosTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0.\n *\n * ```js\n * tf.zeros([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Can\n *     be 'float32', 'int32' or 'bool'. Defaults to 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function zeros<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = zeros(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeZerosTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeOnesTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\nimport {zeros} from './zeros';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1.\n *\n * ```js\n * tf.ones([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n *     'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function ones<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = ones(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeOnesTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam, sizeFromShape} from '../util';\n\nimport {computeOutAndReduceShapes} from './axis_util';\nimport {cast} from './cast';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {ones} from './ones';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sum} from './sum';\n\n/**\n * Computes the mean of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces `x` along the dimensions given in `axis`. Unless `keepDims` is\n * true, the rank of the `tf.Tensor` is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and a `tf.Tensor` with\n * a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.mean().print();  // or tf.mean(a)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.mean(axis).print();  // or tf.mean(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction mean_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'mean');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const shapes = computeOutAndReduceShapes($x.shape, axes);\n  const reduceShape = shapes[1];\n  const reduceSize = sizeFromShape(reduceShape);\n\n  // Use a custom gradient to bypass 2 gradient backprops since mean is used\n  // extremely often.\n  const customOp = customGrad((x: Tensor) => {\n    const reduceSizeScalar = scalar(reduceSize);\n    // Cast if needed.\n    const xReduce = reduceSizeScalar.dtype === x.dtype ?\n        x :\n        cast(x, reduceSizeScalar.dtype);\n    const res = div(xReduce, reduceSizeScalar);\n    const value = sum(res, axis, keepDims);\n\n    const gradFunc = (dy: Tensor) => {\n      const expandedDyShape = x.shape.slice();\n      axes.forEach(axis => {\n        expandedDyShape[axis] = 1;\n      });\n      const expandedDy = reshape(dy, expandedDyShape);\n      const derX = div(mul(expandedDy, ones(x.shape, 'float32')), reduceSize);\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\n\nexport const mean = op({mean_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Min, MinAttrs, MinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the minimum value from the input.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axes`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axes` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.min().print();  // or tf.min(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.min(axis).print();  // or tf.min(x, axis)\n * ```\n *\n * @param x The input Tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction min_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'min');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let minInput = $x;\n        if (permutedAxes != null) {\n          minInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n        }\n\n        const y = backend.min(minInput, axes);\n        if (permutedAxes != null) {\n          minInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape =\n              axis_util.expandShapeToKeepDim(res.shape, origAxes);\n          res = reshape(y, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MinInputs = {x: $x};\n  const attrs: MinAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Min,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const min = op({min_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Minimum, MinimumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the min of a and b (`a < b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `minimumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * ```js\n * // Broadcast minimum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction minimum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'minimum');\n  let $b = convertToTensor(b, 'b', 'minimum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.minimum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MinimumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Minimum) as T;\n}\n\nexport const minimum = op({minimum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Mod, ModInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns the mod of a and b element-wise.\n * `floor(x / y) * y + mod(x, y) = x`\n * Supports broadcasting.\n *\n * We also expose `tf.modStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * ```js\n * // Broadcast a mod b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mod_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mod');\n  let $b = convertToTensor(b, 'b', 'mod');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.mod($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: ModInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Mod) as T;\n}\n\nexport const mod = op({mod_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Computes square of `x` element-wise: `x ^ 2`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.sqrt(2), -1]);\n *\n * x.square().print();  // or tf.square(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction square_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'square');\n  const attrs = {};\n  const inputsToSave = [$x];\n  const outputsToSave: boolean[] = [];\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    return backend.square($x);\n  }, {x: $x}, null /* grad */, 'Square', attrs, inputsToSave, outputsToSave);\n}\n\nexport const square = op({square_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim} from './axis_util';\nimport {cast} from './cast';\nimport {mean} from './mean';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {square} from './square';\nimport {sub} from './sub';\n\n/**\n * Calculates the mean and variance of `x`. The mean and variance are\n * calculated by aggregating the contents of `x` across `axes`. If `x` is\n * 1-D and `axes = [0]` this is just the mean and variance of a vector.\n *\n * @param x The input tensor.\n * @param axis The dimension(s) along with to compute mean and\n *     variance. By default it reduces all dimensions.\n * @param keepDims If true, the moments have the same dimensionality as the\n *     input.\n * @return An object with two keys: `mean` and `variance`.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction moments_(\n    x: Tensor|TensorLike, axis: number|number[] = null,\n    keepDims = false): {mean: Tensor, variance: Tensor} {\n  x = convertToTensor(x, 'x', 'moments');\n  const axes = parseAxisParam(axis, x.shape);\n  const xMean = mean(x, axes, keepDims);\n  let keepDimsShape = xMean.shape;\n  if (!keepDims) {\n    keepDimsShape = expandShapeToKeepDim(xMean.shape, axes);\n  }\n  const devSquared =\n      square(sub(cast(x, 'float32'), reshape(xMean, keepDimsShape)));\n  const variance = mean(devSquared, axes, keepDims);\n  return {mean: xMean, variance};\n}\n\nexport const moments = op({moments_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor, convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * @docalias (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D]\n */\nexport type LSTMCellFunc = {\n  (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D];\n};\n\n/**\n * Computes the next states and outputs of a stack of LSTMCells.\n *\n * Each cell output is used as input to the next cell.\n *\n * Returns `[cellState, cellOutput]`.\n *\n * Derived from tf.contrib.rn.MultiRNNCell.\n *\n * @param lstmCells Array of LSTMCell functions.\n * @param data The input to the cell.\n * @param c Array of previous cell states.\n * @param h Array of previous cell outputs.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction multiRNNCell_(\n    lstmCells: LSTMCellFunc[], data: Tensor2D|TensorLike,\n    c: Array<Tensor2D|TensorLike>,\n    h: Array<Tensor2D|TensorLike>): [Tensor2D[], Tensor2D[]] {\n  const $data = convertToTensor(data, 'data', 'multiRNNCell');\n  const $c = convertToTensorArray(c, 'c', 'multiRNNCell');\n  const $h = convertToTensorArray(h, 'h', 'multiRNNCell');\n\n  let input = $data;\n  const newStates = [];\n  for (let i = 0; i < lstmCells.length; i++) {\n    const output = lstmCells[i](input, $c[i], $h[i]);\n    newStates.push(output[0]);\n    newStates.push(output[1]);\n    input = output[1];\n  }\n  const newC: Tensor2D[] = [];\n  const newH: Tensor2D[] = [];\n  for (let i = 0; i < newStates.length; i += 2) {\n    newC.push(newStates[i]);\n    newH.push(newStates[i + 1]);\n  }\n  return [newC, newH];\n}\nexport const multiRNNCell = op({multiRNNCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a `tf.Tensor` with values drawn from a multinomial distribution.\n *\n * ```js\n * const probs = tf.tensor([.75, .25]);\n * tf.multinomial(probs, 3).print();\n * ```\n *\n * @param logits 1D array with unnormalized log-probabilities, or\n *     2D array of shape `[batchSize, numOutcomes]`. See the `normalized`\n *     parameter.\n * @param numSamples Number of samples to draw for each row slice.\n * @param seed The seed number.\n * @param normalized Whether the provided `logits` are normalized true\n *     probabilities (sum to 1). Defaults to false.\n * @return 1D array of shape `[numSamples]`, or 2D array of shape\n *     `[batchSize, numSamples]`, depending on the rank of the input.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction multinomial_(\n    logits: Tensor1D|Tensor2D|TensorLike, numSamples: number, seed?: number,\n    normalized = false): Tensor1D|Tensor2D {\n  const $logits = convertToTensor(logits, 'logits', 'multinomial');\n  const numOutcomes = $logits.size;\n  const origRank = $logits.rank;\n  if (numOutcomes < 2) {\n    throw new Error(\n        `Error in multinomial: you need at least 2 outcomes, but got ` +\n        `${numOutcomes}.`);\n  }\n  if (origRank > 2) {\n    throw new Error(`Rank of probabilities must be 1 or 2, but is ${origRank}`);\n  }\n  seed = seed || Math.random();\n  const logits2D: Tensor2D =\n      origRank === 1 ? reshape($logits, [1, -1]) : $logits as Tensor2D;\n  const res = ENGINE.runKernelFunc(\n      backend => backend.multinomial(logits2D, normalized, numSamples, seed),\n      {logits2D});\n  // tslint:disable-next-line:no-unnecessary-type-assertion\n  return origRank === 1 ? reshape(res, [res.size]) as Tensor1D : res;\n}\n\nexport const multinomial = op({multinomial_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {NotEqual, NotEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a != b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([0, 2, 3]);\n *\n * a.notEqual(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction notEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'notEqual');\n  let $b = convertToTensor(b, 'b', 'notEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => backend.notEqual($a, $b);\n\n  const inputs: NotEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             NotEqual) as T;\n}\n\nexport const notEqual = op({notEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Real, RealInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the real part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the real part of each element in input considered as a complex number.\n *\n * If the input is real, it simply makes a clone.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.real(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction real_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'real');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.real($input);\n  };\n\n  const inputs: RealInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Real) as T;\n}\n\nexport const real = op({real_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OnesLike, OnesLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {complex} from './complex';\nimport {imag} from './imag';\nimport {op} from './operation';\nimport {real} from './real';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.onesLike(x).print();\n * ```\n * @param x A tensor.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction onesLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'onesLike');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if ($x.dtype === 'complex64') {\n      const r = onesLike(real($x));\n      const i = zerosLike(imag($x));\n      return complex(r, i);\n    }\n\n    return backend.onesLike($x);\n  };\n\n  const inputs: OnesLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             OnesLike) as T;\n}\n\nexport const onesLike = op({onesLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the outer product of two vectors, `v1` and `v2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([3, 4, 5]);\n *\n * tf.outerProduct(a, b).print();\n * ```\n * @param v1 The first vector in the outer product operation.\n * @param v2 The second vector in the outer product operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction outerProduct_(\n    v1: Tensor1D|TensorLike, v2: Tensor1D|TensorLike): Tensor2D {\n  const $v1 = convertToTensor(v1, 'v1', 'outerProduct');\n  const $v2 = convertToTensor(v2, 'v2', 'outerProduct');\n\n  util.assert(\n      $v1.rank === 1 && $v2.rank === 1,\n      () => `Error in outerProduct: inputs must be rank 1, but got ranks ` +\n          `${$v1.rank} and ${$v2.rank}.`);\n\n  const v12D = reshape($v1, [-1, 1]);\n  const v22D = reshape($v2, [1, -1]);\n  return matMul(v12D, v22D) as Tensor2D;\n}\n\nexport const outerProduct = op({outerProduct_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {PadV2, PadV2Attrs, PadV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Pads a `tf.Tensor` with a given value and paddings.\n *\n * This operation currently only implements the `CONSTANT` mode.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `paddings` is of given length.\n *   - `tf.pad1d`\n *   - `tf.pad2d`\n *   - `tf.pad3d`\n *   - `tf.pad4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.pad([[1, 2]]).print();\n * ```\n * @param x The tensor to pad.\n * @param paddings An array of length `R` (the rank of the tensor), where\n * each element is a length-2 tuple of ints `[padBefore, padAfter]`,\n * specifying how much to pad along each dimension of the tensor.\n * @param constantValue The pad value to use. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction pad_<T extends Tensor>(\n    x: T|TensorLike, paddings: Array<[number, number]>, constantValue = 0): T {\n  const $x = convertToTensor(x, 'x', 'pad');\n  if ($x.rank === 0) {\n    throw new Error('pad(scalar) is not defined. Pass non-scalar to pad');\n  }\n  const forward: ForwardFunc<T> = (backend, save) => {\n    save([$x]);\n    return backend.pad($x, paddings, constantValue);\n  };\n\n  const attrs: PadV2Attrs = {paddings, constantValue};\n  const inputs: PadV2Inputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, PadV2,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const pad = op({pad_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor1D` with a given value and paddings. See `pad` for details.\n */\nfunction pad1d_(\n    x: Tensor1D|TensorLike, paddings: [number, number],\n    constantValue = 0): Tensor1D {\n  assert(\n      paddings.length === 2,\n      () => 'Invalid number of paddings. Must be length of 2.');\n  return pad(x, [paddings], constantValue);\n}\n\nexport const pad1d = op({pad1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor2D` with a given value and paddings. See `pad` for details.\n */\nfunction pad2d_(\n    x: Tensor2D|TensorLike, paddings: [[number, number], [number, number]],\n    constantValue = 0): Tensor2D {\n  assert(\n      paddings.length === 2 && paddings[0].length === 2 &&\n          paddings[1].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad2d = op({pad2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor3D` with a given value and paddings. See `pad` for details.\n */\nfunction pad3d_(\n    x: Tensor3D|TensorLike,\n    paddings: [[number, number], [number, number], [number, number]],\n    constantValue = 0): Tensor3D {\n  assert(\n      paddings.length === 3 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad3d = op({pad3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor4D` with a given value and paddings. See `pad` for details.\n */\nfunction pad4d_(\n    x: Tensor4D|TensorLike,\n    paddings:\n        [\n          [number, number], [number, number], [number, number], [number, number]\n        ],\n    constantValue = 0): Tensor4D {\n  assert(\n      paddings.length === 4 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2 &&\n          paddings[3].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad4d = op({pad4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation divides \"spatial\" dimensions `[1, ..., M]` of the input into\n * a grid of blocks of shape `blockShape`, and interleaves these blocks with\n * the \"batch\" dimension (0) such that in the output, the spatial\n * dimensions `[1, ..., M]` correspond to the position within the grid,\n * and the batch dimension combines both the position within a spatial block\n * and the original batch position. Prior to division into blocks,\n * the spatial dimensions of the input are optionally zero padded\n * according to `paddings`. See below for a precise description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);\n * const blockShape = [2, 2];\n * const paddings = [[0, 0], [0, 0]];\n *\n * x.spaceToBatchND(blockShape, paddings).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param paddings A 2-D array. Must have shape `[M, 2]`, all values must be >=\n *     0. `paddings[i] = [padStart, padEnd]` specifies the amount to zero-pad\n * from input dimension `i + 1`, which corresponds to spatial dimension `i`. It\n * is required that\n * `(inputShape[i + 1] + padStart + padEnd) % blockShape[i] === 0`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Zero-pad the start and end of dimensions `[1, ..., M]` of the input\n * according to `paddings` to produce `padded` of shape paddedShape.\n *\n * 2. Reshape `padded` to `reshapedPadded` of shape:\n * `[batch] + [paddedShape[1] / blockShape[0], blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1], blockShape[M-1]] + remainingShape`\n *\n * 3. Permute dimensions of `reshapedPadded` to produce `permutedReshapedPadded`\n * of shape: `blockShape + [batch] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * 4. Reshape `permutedReshapedPadded` to flatten `blockShape` into the\n * batch dimension, producing an output tensor of shape:\n * `[batch * prod(blockShape)] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction spaceToBatchND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], paddings: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'spaceToBatchND');\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank ${$x.rank} should be > than [blockShape] ${\n          blockShape.length}`);\n\n  util.assert(\n      paddings.length === blockShape.length,\n      () => `paddings.shape[0] ${\n          paddings.length} must be equal to [blockShape] ${blockShape.length}`);\n\n  util.assert(\n      $x.shape.reduce(\n          (a, b, i) => {\n            if (i > 0 && i <= blockShape.length) {\n              return a &&\n                  ((b + paddings[i - 1][0] + paddings[i - 1][1]) %\n                       blockShape[i - 1] ===\n                   0);\n            }\n            return a;\n          },\n          true),\n      () => `input spatial dimensions ${$x.shape.slice(1)} with paddings ${\n          paddings.toString()} must be divisible by blockShapes ${\n          blockShape.toString()}`);\n\n  const forward: ForwardFunc<T> = backend =>\n      backend.spaceToBatchND($x, blockShape, paddings);\n\n  const inputs: SpaceToBatchNDInputs = {x: $x};\n  const attrs: SpaceToBatchNDAttrs = {blockShape, paddings};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      SpaceToBatchND, attrs as {} as NamedAttrMap);\n}\n\nexport const spaceToBatchND = op({spaceToBatchND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {avgPool} from './avg_pool';\nimport {batchToSpaceND} from './batch_to_space_nd';\nimport * as conv_util from './conv_util';\nimport {maxPool} from './max_pool';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {spaceToBatchND} from './space_to_batch_nd';\n\n/**\n * Performs an N-D pooling operation\n *\n * @param input The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param windowShape The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param poolingType The type of pooling, either 'max' or 'avg'.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilationRate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction pool_<T extends Tensor3D|Tensor4D>(\n    input: T|TensorLike, windowShape: [number, number]|number,\n    poolingType: 'avg'|'max', pad: 'valid'|'same'|number,\n    dilations?: [number, number]|number, strides?: [number, number]|number) {\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  if (strides == null) {\n    strides = 1;\n  }\n  if (pad === 0) {\n    pad = 'valid';\n  }\n\n  const $x = convertToTensor(input, 'x', 'maxPool');\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in pool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = conv_util.computePool2DInfo(\n      x4D.shape, windowShape, strides, dilations, pad);\n  const dilation: [number, number] =\n      [convInfo.dilationHeight, convInfo.dilationWidth];\n\n  // The following implementation does batchToSpace(pool(spaceToBatch(x)))\n  // whenever dilation > 1 since the TF kernels do not support dilation > 1.\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L1037\n\n  let basePadding: number[][];\n  if (pad === 'same') {\n    basePadding = withSpaceToBatchBasePaddings(\n        [convInfo.filterHeight, convInfo.filterWidth], dilation);\n  } else {\n    basePadding = [[0, 0], [0, 0]];\n  }\n\n  const isDilationOne = dilation[0] === 1 && dilation[1] === 1;\n  const [adjustedPadding, adjustedCrops] = requiredSpaceToBatchPaddings(\n      [convInfo.inHeight, convInfo.inWidth], dilation, basePadding);\n  const convertedPad = isDilationOne ? pad : 'valid';\n  const convertedX =\n      isDilationOne ? x4D : spaceToBatchND(x4D, dilation, adjustedPadding);\n\n  const forwardOp = poolingType === 'avg' ?\n      () => avgPool(convertedX, windowShape, strides, convertedPad) :\n      () => maxPool(convertedX, windowShape, strides, convertedPad);\n  const y = forwardOp();\n\n  const res = isDilationOne ? y : batchToSpaceND(y, dilation, adjustedCrops);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\n// Helper function to compute crops and paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/array_ops.py#L2184\nfunction requiredSpaceToBatchPaddings(\n    inputShape: [number, number], blockShape: [number, number],\n    basePadding: number[][]) {\n  const padStart = basePadding.map(b => b[0]);\n  const origPadEnd = basePadding.map(b => b[1]);\n  const fullInputShape = inputShape.concat(padStart, origPadEnd);\n  const padEndExtra = blockShape.map((b, i) => (b - fullInputShape[i] % b) % b);\n  const padEnd = origPadEnd.map((s, i) => s + padEndExtra[i]);\n  const paddings = blockShape.map((_, i) => [padStart[i], padEnd[i]]);\n  const crops = blockShape.map((_, i) => [0, padEndExtra[i]]);\n  return [paddings, crops];\n}\n\n// Helper function to compute base paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L524\nfunction withSpaceToBatchBasePaddings(\n    filterShape: [number, number], dilation: [number, number]) {\n  // Spatial dimensions of the filters and the upsampled filters in which we\n  // introduce (rate - 1) zeros between consecutive filter values.\n  const dilatedFilterShape = filterShape.map((s, i) => {\n    return s + (s - 1) * (dilation[i] - 1);\n  });\n  const padExtraShape = dilatedFilterShape.map(s => s - 1);\n\n  // When padding is odd, we pad more at end, following the same\n  // convention as conv2d.\n  const padExtraStart = padExtraShape.map(s => Math.floor(s / 2));\n  const padExtraEnd = padExtraShape.map((s, i) => s - padExtraStart[i]);\n  return padExtraShape.map((_, i) => {\n    return [padExtraStart[i], padExtraEnd[i]];\n  });\n}\n\nexport const pool = op({pool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Pow, PowInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the power of one `tf.Tensor` to another. Supports broadcasting.\n *\n * Given a `tf.Tensor` x and a `tf.Tensor` y, this operation computes x^y for\n * corresponding elements in x and y. The result's dtype will be the upcasted\n * type of the `base` and `exp` dtypes.\n *\n * ```js\n * const a = tf.tensor([[2, 3], [4, 5]])\n * const b = tf.tensor([[1, 2], [3, 0]]).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n *\n * ```js\n * const a = tf.tensor([[1, 2], [3, 4]])\n * const b = tf.tensor(2).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n * We also expose `powStrict` which has the same signature as this op and\n * asserts that `base` and `exp` are the same shape (does not broadcast).\n *\n * @param base The base `tf.Tensor` to pow element-wise.\n * @param exp The exponent `tf.Tensor` to pow element-wise.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction pow_<T extends Tensor>(\n    base: Tensor|TensorLike, exp: Tensor|TensorLike): T {\n  let $base = convertToTensor(base, 'base', 'pow');\n  let $exp = convertToTensor(exp, 'exp', 'pow');\n  [$base, $exp] = makeTypesMatch($base, $exp);\n\n  const inputs: PowInputs = {a: $base, b: $exp};\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.pow($base, $exp);\n    save([$base, $exp, y]);\n    return y;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Pow) as T;\n}\n\nexport const pow = op({pow_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prelu, PreluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes leaky rectified linear element-wise with parametric alphas.\n *\n * `x < 0 ? alpha * x : f(x) = x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n * const alpha = tf.scalar(0.1);\n *\n * x.prelu(alpha).print();  // or tf.prelu(x, alpha)\n * ```\n * @param x The input tensor.\n * @param alpha Scaling factor for negative values.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction prelu_<T extends Tensor>(x: T|TensorLike, alpha: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'prelu');\n  const $alpha = convertToTensor(alpha, 'alpha', 'prelu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.prelu($x, $alpha);\n    save([$x, $alpha]);\n    return res;\n  };\n\n  const inputs: PreluInputs = {x: $x, alpha: $alpha};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prelu) as\n      T;\n}\n\nexport const prelu = op({prelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prod, ProdAttrs, ProdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the product of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.prod().print();  // or tf.prod(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.prod(axis).print();  // or tf.prod(x, axis)\n * ```\n *\n * @param x The input tensor to compute the product over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction prod_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'prod');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if ($x.dtype === 'bool') {\n      $x = cast($x, 'int32');\n    }\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.prod(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n\n    return value as T;\n  };\n\n  const inputs: ProdInputs = {x: $x};\n  const attrs: ProdAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prod,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const prod = op({prod_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a random number generator\n * function defined by the user.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param randFunction A random number generator function which is called\n * for each element in the output tensor.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n */\nfunction rand_<R extends Rank>(\n    shape: ShapeMap[R], randFunction: () => number,\n    dtype?: DataType): Tensor<R> {\n  const size = sizeFromShape(shape);\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  for (let i = 0; i < size; i++) {\n    values[i] = randFunction();\n  }\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n\nexport const rand = op({rand_});\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as seedrandom from 'seedrandom';\n\nimport {expectNumbersClose, testEpsilon} from '../test_util';\nimport {TypedArray} from '../types';\n\nexport interface RandomBase {\n  nextValue(): number;\n}\n\nexport interface RandomGamma {\n  nextValue(): number;\n}\n\nexport interface RandNormalDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\nexport interface RandGammaDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\n// https://en.wikipedia.org/wiki/Marsaglia_polar_method\nexport class MPRandGauss implements RandomBase {\n  private mean: number;\n  private stdDev: number;\n  private nextVal: number;\n  private dtype?: keyof RandNormalDataTypes;\n  private truncated?: boolean;\n  private upper?: number;\n  private lower?: number;\n  private random: seedrandom.prng;\n\n  constructor(\n      mean: number, stdDeviation: number, dtype?: keyof RandNormalDataTypes,\n      truncated?: boolean, seed?: number) {\n    this.mean = mean;\n    this.stdDev = stdDeviation;\n    this.dtype = dtype;\n    this.nextVal = NaN;\n    this.truncated = truncated;\n    if (this.truncated) {\n      this.upper = this.mean + this.stdDev * 2;\n      this.lower = this.mean - this.stdDev * 2;\n    }\n    const seedValue = seed ? seed : Math.random();\n    this.random = seedrandom.alea(seedValue.toString());\n  }\n\n  /** Returns next sample from a Gaussian distribution. */\n  public nextValue(): number {\n    if (!isNaN(this.nextVal)) {\n      const value = this.nextVal;\n      this.nextVal = NaN;\n      return value;\n    }\n\n    let resultX: number, resultY: number;\n    let isValid = false;\n    while (!isValid) {\n      let v1: number, v2: number, s: number;\n      do {\n        v1 = 2 * this.random() - 1;\n        v2 = 2 * this.random() - 1;\n        s = v1 * v1 + v2 * v2;\n      } while (s >= 1 || s === 0);\n\n      const mul = Math.sqrt(-2.0 * Math.log(s) / s);\n      resultX = this.mean + this.stdDev * v1 * mul;\n      resultY = this.mean + this.stdDev * v2 * mul;\n\n      if (!this.truncated || this.isValidTruncated(resultX)) {\n        isValid = true;\n      }\n    }\n\n    if (!this.truncated || this.isValidTruncated(resultY)) {\n      this.nextVal = this.convertValue(resultY);\n    }\n    return this.convertValue(resultX);\n  }\n\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype == null || this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  /** Returns true if less than 2-standard-deviations from the mean. */\n  private isValidTruncated(value: number): boolean {\n    return value <= this.upper && value >= this.lower;\n  }\n}\n\n// Marsaglia, George, and Wai Wan Tsang. 2000. \"A Simple Method for Generating\n// Gamma Variables.\"\nexport class RandGamma implements RandomGamma {\n  private alpha: number;\n  private beta: number;\n  private d: number;\n  private c: number;\n  private dtype?: keyof RandGammaDataTypes;\n  private randu: seedrandom.prng;\n  private randn: MPRandGauss;\n\n  constructor(\n      alpha: number, beta: number, dtype: keyof RandGammaDataTypes,\n      seed?: number) {\n    this.alpha = alpha;\n    this.beta = 1 / beta;  // convert rate to scale parameter\n    this.dtype = dtype;\n\n    const seedValue = seed ? seed : Math.random();\n    this.randu = seedrandom.alea(seedValue.toString());\n    this.randn = new MPRandGauss(0, 1, dtype, false, this.randu());\n\n    if (alpha < 1) {\n      this.d = alpha + (2 / 3);\n    } else {\n      this.d = alpha - (1 / 3);\n    }\n    this.c = 1 / Math.sqrt(9 * this.d);\n  }\n\n  /** Returns next sample from a gamma distribution. */\n  public nextValue(): number {\n    let x2: number, v0: number, v1: number, x: number, u: number, v: number;\n    while (true) {\n      do {\n        x = this.randn.nextValue();\n        v = 1 + (this.c * x);\n      } while (v <= 0);\n      v *= v * v;\n      x2 = x * x;\n      v0 = 1 - (0.331 * x2 * x2);\n      v1 = (0.5 * x2) + (this.d * (1 - v + Math.log(v)));\n      u = this.randu();\n      if (u < v0 || Math.log(u) < v1) {\n        break;\n      }\n    }\n    v = (1 / this.beta) * this.d * v;\n    if (this.alpha < 1) {\n      v *= Math.pow(this.randu(), 1 / this.alpha);\n    }\n    return this.convertValue(v);\n  }\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n}\n\nexport class UniformRandom implements RandomBase {\n  private min: number;\n  private range: number;\n  private random: seedrandom.prng;\n  private dtype?: keyof RandNormalDataTypes;\n\n  constructor(\n      min = 0, max = 1, dtype?: keyof RandNormalDataTypes,\n      seed?: string|number) {\n    this.min = min;\n    this.range = max - min;\n    this.dtype = dtype;\n    if (seed == null) {\n      seed = Math.random();\n    }\n    if (typeof seed === 'number') {\n      seed = seed.toString();\n    }\n\n    if (!this.canReturnFloat() && this.range <= 1) {\n      throw new Error(\n          `The difference between ${min} - ${max} <= 1 and dtype is not float`);\n    }\n    this.random = seedrandom.alea(seed);\n  }\n\n  /** Handles proper rounding for non floating point numbers. */\n  private canReturnFloat = () =>\n      (this.dtype == null || this.dtype === 'float32');\n\n  private convertValue(value: number): number {\n    if (this.canReturnFloat()) {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  nextValue() {\n    return this.convertValue(this.min + this.range * this.random());\n  }\n}\n\nexport function jarqueBeraNormalityTest(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test\n  const n = values.length;\n  const s = skewness(values);\n  const k = kurtosis(values);\n  const jb = n / 6 * (Math.pow(s, 2) + 0.25 * Math.pow(k - 3, 2));\n  // JB test requires 2-degress of freedom from Chi-Square @ 0.95:\n  // http://www.itl.nist.gov/div898/handbook/eda/section3/eda3674.htm\n  const CHI_SQUARE_2DEG = 5.991;\n  if (jb > CHI_SQUARE_2DEG) {\n    throw new Error(`Invalid p-value for JB: ${jb}`);\n  }\n}\n\nexport function expectArrayInMeanStdRange(\n    actual: TypedArray|number[], expectedMean: number, expectedStdDev: number,\n    epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  const actualMean = mean(actual);\n  expectNumbersClose(actualMean, expectedMean, epsilon);\n  expectNumbersClose(\n      standardDeviation(actual, actualMean), expectedStdDev, epsilon);\n}\n\nfunction mean(values: TypedArray|number[]) {\n  let sum = 0;\n  for (let i = 0; i < values.length; i++) {\n    sum += values[i];\n  }\n  return sum / values.length;\n}\n\nfunction standardDeviation(values: TypedArray|number[], mean: number) {\n  let squareDiffSum = 0;\n  for (let i = 0; i < values.length; i++) {\n    const diff = values[i] - mean;\n    squareDiffSum += diff * diff;\n  }\n  return Math.sqrt(squareDiffSum / values.length);\n}\n\nfunction kurtosis(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Kurtosis\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum4 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum4 += Math.pow(v, 4);\n  }\n  return (1 / n) * sum4 / Math.pow((1 / n) * sum2, 2);\n}\n\nfunction skewness(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Skewness\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum3 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum3 += Math.pow(v, 3);\n  }\n  return (1 / n) * sum3 / Math.pow((1 / (n - 1)) * sum2, 3 / 2);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {RandGamma} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a gamma distribution.\n *\n * ```js\n * tf.randomGamma([2, 2], 1).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param alpha The shape parameter of the gamma distribution.\n * @param beta The inverse scale parameter of the gamma distribution. Defaults\n *     to 1.\n * @param dtype The data type of the output. Defaults to float32.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomGamma_<R extends Rank>(\n    shape: ShapeMap[R], alpha: number, beta = 1,\n    dtype: 'float32'|'int32' = 'float32', seed?: number): Tensor<R> {\n  if (beta == null) {\n    beta = 1;\n  }\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const rgamma = new RandGamma(alpha, beta, dtype, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = rgamma.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomGamma = op({randomGamma_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a normal distribution.\n *\n * ```js\n * tf.randomNormal([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, false /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomNormal = op({randomNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {UniformRandom} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a uniform distribution.\n *\n * The generated values follow a uniform distribution in the range [minval,\n * maxval). The lower bound minval is included in the range, while the upper\n * bound maxval is excluded.\n *\n * ```js\n * tf.randomUniform([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param minval The lower bound on the range of random values to generate.\n *   Defaults to 0.\n * @param maxval The upper bound on the range of random values to generate.\n *   Defaults to 1.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomUniform_<R extends Rank>(\n    shape: ShapeMap[R], minval = 0, maxval = 1, dtype: DataType = 'float32',\n    seed?: number|string): Tensor<R> {\n  const res = buffer(shape, dtype);\n  const random = new UniformRandom(minval, maxval, null, seed);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = random.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomUniform = op({randomUniform_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike1D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-1 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor1d` as it makes the code more readable.\n *\n * ```js\n * tf.tensor1d([1, 2, 3]).print();\n * ```\n *\n * @param values The values of the tensor. Can be array of numbers,\n *     or a `TypedArray`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor1d(values: TensorLike1D, dtype?: DataType): Tensor1D {\n  assertNonNull(values);\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 1) {\n    throw new Error('tensor1d() requires values to be a flat/TypedArray');\n  }\n  const shape: number[] = null;\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Range, RangeAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {makeZerosTypedArray} from '../util';\n\nimport {tensor1d} from './tensor1d';\nimport {zeros} from './zeros';\n\n/**\n * Creates a new `tf.Tensor1D` filled with the numbers in the range provided.\n *\n * The tensor is a is half-open interval meaning it includes start, but\n * excludes stop. Decrementing ranges and negative step values are also\n * supported.sv\n *\n *\n * ```js\n * tf.range(0, 9, 2).print();\n * ```\n *\n * @param start An integer start value\n * @param stop An integer stop value\n * @param step An integer increment (will default to 1 or -1)\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function range(\n    start: number, stop: number, step = 1,\n    dtype: 'float32'|'int32' = 'float32'): Tensor1D {\n  if (step === 0) {\n    throw new Error('Cannot have a step of zero');\n  }\n\n  const forward: ForwardFunc<Tensor> = () => {\n    const sameStartStop = start === stop;\n    const increasingRangeNegativeStep = start < stop && step < 0;\n    const decreasingRangePositiveStep = stop < start && step > 1;\n\n    if (sameStartStop || increasingRangeNegativeStep ||\n        decreasingRangePositiveStep) {\n      return zeros([0], dtype);\n    }\n\n    const numElements = Math.abs(Math.ceil((stop - start) / step));\n    const values = makeZerosTypedArray(numElements, dtype);\n\n    if (stop < start && step === 1) {\n      // Auto adjust the step's sign if it hasn't been set\n      // (or was set to 1)\n      step = -1;\n    }\n\n    values[0] = start;\n    for (let i = 1; i < values.length; i++) {\n      values[i] = values[i - 1] + step;\n    }\n\n    return tensor1d(values, dtype);\n  };\n\n  const attrs: RangeAttrs = {start, stop, step, dtype};\n\n  return ENGINE.runKernelFunc(\n             forward, {} /* inputs */, null /* grad */, Range,\n             attrs as {} as NamedAttrMap) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Reciprocal, ReciprocalInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of x element-wise: `1 / x`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, 2]);\n *\n * x.reciprocal().print();  // or tf.reciprocal(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction reciprocal_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'reciprocal');\n\n  const inputs: ReciprocalInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.reciprocal($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Reciprocal);\n}\nexport const reciprocal = op({reciprocal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu, ReluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport { cast } from './cast';\n\n/**\n * Computes rectified linear element-wise: `max(x, 0)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.relu().print();  // or tf.relu(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu($x);\n  };\n\n  const inputs: ReluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu) as\n      T;\n}\n\nexport const relu = op({relu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu6, Relu6Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes rectified linear 6 element-wise: `min(max(x, 0), 6)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 8]);\n *\n * x.relu6().print();  // or tf.relu6(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu6_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu6');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu6($x);\n  };\n\n  const inputs: Relu6Inputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu6) as\n      T;\n}\n\nexport const relu6 = op({relu6_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reverse, ReverseAttrs, ReverseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Reverses a `tf.Tensor` along a specified axis.\n *\n * Also available are stricter rank-specific methods that assert that `x` is\n * of the given rank:\n *   - `tf.reverse1d`\n *   - `tf.reverse2d`\n *   - `tf.reverse3d`\n *   - `tf.reverse4d`\n *\n * Except `tf.reverse1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.reverse().print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.reverse(axis).print();\n * ```\n * @param x The input tensor to be reversed.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction reverse_<T extends Tensor>(\n    x: T|TensorLike, axis?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'reverse');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const axes = parseAxisParam(axis, $x.shape);\n    if ($x.rank === 0) {\n      return clone($x);\n    }\n    const res = backend.reverse($x, axes);\n    return reshape(res, $x.shape);\n  };\n\n  const inputs: ReverseInputs = {x: $x};\n  const attrs: ReverseAttrs = {dims: axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Reverse, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const reverse = op({reverse_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor1D`.\n *\n * @param x The input tensor.\n */\nfunction reverse1d_(x: Tensor1D|TensorLike): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 1,\n      () => `Error in reverse1D: x must be rank 1 but got rank ${$x.rank}.`);\n  return reverse($x, 0);\n}\n\nexport const reverse1d = op({reverse1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor2D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse2d_(x: Tensor2D|TensorLike, axis?: number|number[]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 2,\n      () => `Error in reverse2D: x must be rank 2 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse2d = op({reverse2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor3D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse3d_(x: Tensor3D|TensorLike, axis?: number|number[]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 3,\n      () => `Error in reverse3D: x must be rank 3 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse3d = op({reverse3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor4D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse4d_(x: Tensor4D|TensorLike, axis?: number|number[]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 4,\n      () => `Error in reverse4D: x must be rank 4 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse4d = op({reverse4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Round, RoundInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes round of input `tf.Tensor` element-wise: `round(x)`.\n * It implements banker's rounding.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.round().print();  // or tf.round(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction round_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'round');\n  const inputs: RoundInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.round($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Round);\n}\n\nexport const round = op({round_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Rsqrt, RsqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of square root of the input `tf.Tensor` element-wise:\n * `y = 1 / sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.rsqrt().print();  // or tf.rsqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction rsqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'rsqrt');\n\n  const inputs: RsqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.rsqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Rsqrt);\n}\nexport const rsqrt = op({rsqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Selu, SeluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes scaled exponential linear element-wise.\n *\n * `x < 0 ? scale * alpha * (exp(x) - 1) : x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.selu().print();  // or tf.selu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction selu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'selu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.selu($x);\n    save([$x]);\n    return res;\n  };\n\n  const inputs: SeluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Selu) as\n      T;\n}\n\nexport const selu = op({selu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport {depthwiseConv2d} from './depthwise_conv2d';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * 2-D convolution with separable filters.\n *\n * Performs a depthwise convolution that acts separately on channels followed\n * by a pointwise convolution that mixes channels. Note that this is\n * separability between dimensions [1, 2] and 3, not spatial separability\n * between dimensions 1 and 2.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param depthwiseFilter The depthwise filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`. This is\n *     the filter used in the first step.\n * @param pointwiseFilter The pointwise filter tensor, rank 4, of shape\n *     `[1, 1, inChannels * channelMultiplier, outChannels]`. This is\n *     the filter used in the second step.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction separableConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthwiseFilter: Tensor4D|TensorLike,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'separableConv2d');\n  const $depthwiseFilter =\n      convertToTensor(depthwiseFilter, 'depthwiseFilter', 'separableConv2d');\n  const $pointwiseFilter =\n      convertToTensor(pointwiseFilter, 'pointwiseFilter', 'separableConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  if (dataFormat === 'NCHW') {\n    throw new Error(\n        'separableConv2d currently does not support dataFormat NCHW; only ' +\n        'NHWC is supported');\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in separableConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $depthwiseFilter.rank === 4,\n      () => `Error in separableConv2d: depthwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.rank === 4,\n      () => `Error in separableConv2d: pointwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.shape[0] === 1,\n      () =>\n          `Error in separableConv2d: the first dimension of pointwise filter ` +\n          ` must be 1, but got ${$pointwiseFilter.shape[0]}.`);\n  util.assert(\n      $pointwiseFilter.shape[1] === 1,\n      () => `Error in separableConv2d: the second dimension of pointwise ` +\n          `filter must be 1, but got ${$pointwiseFilter.shape[1]}.`);\n\n  const inChannels = $depthwiseFilter.shape[2];\n  const channelMultiplier = $depthwiseFilter.shape[3];\n  util.assert(\n      $pointwiseFilter.shape[2] === inChannels * channelMultiplier,\n      () =>\n          `Error in separableConv2d: the third dimension of pointwise filter ` +\n          `must be ${inChannels * channelMultiplier}, ` +\n          `but got ${$pointwiseFilter.shape[2]}.`);\n\n  const depthwise = depthwiseConv2d(\n      x4D, $depthwiseFilter, strides, pad, dataFormat, dilation);\n  const pointwiseStride = 1;\n  const res =\n      conv2d(depthwise, $pointwiseFilter, pointwiseStride, 'valid', dataFormat);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const separableConv2d = op({separableConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\n/**\n * Computes the difference between two lists of numbers.\n *\n * Given a Tensor `x` and a Tensor `y`, this operation returns a Tensor `out`\n * that represents all values that are in `x` but not in `y`. The returned\n * Tensor `out` is sorted in the same order that the numbers appear in `x`\n * (duplicates are preserved). This operation also returns a Tensor indices that\n * represents the position of each out element in `x`. In other words:\n *\n * `out[i] = x[idx[i]] for i in [0, 1, ..., out.length - 1]`\n *\n * ```js\n * const x = [1, 2, 3, 4, 5, 6];\n * const y = [1, 3, 5];\n *\n * const [out, indices] = await tf.setdiff1dAsync(x, y);\n * out.print(); // [2, 4, 6]\n * indices.print(); // [1, 3, 5]\n * ```\n *\n * @param x 1-D Tensor. Values to keep.\n * @param y 1-D Tensor. Must have the same type as x. Values to exclude in the\n *     output.\n * @returns Promise of Tensor tuple [out, indices].\n *  out: Tensor with the same type as x.\n *  indices: A Tensor of type int32.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nasync function setdiff1dAsync_(\n    x: Tensor|TensorLike, y: Tensor|TensorLike): Promise<[Tensor, Tensor]> {\n  const $x = convertToTensor(x, 'x', 'setdiff1d');\n  const $y = convertToTensor(y, 'y', 'setdiff1d');\n\n  util.assert(\n      $x.dtype === $y.dtype,\n      () => `x and y should have the same dtype, but got x (${\n          $x.dtype}) and y (${$y.dtype}).`);\n\n  util.assert(\n      $x.rank === 1, () => `x should be 1D tensor, but got x (${$x.shape}).`);\n\n  util.assert(\n      $y.rank === 1, () => `y should be 1D tensor, but got y (${$y.shape}).`);\n\n  const xVals = await $x.data();\n  const yVals = await $y.data();\n  const ySet = new Set(yVals);\n\n  let outputSize = 0;\n  for (let i = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      outputSize++;\n    }\n  }\n\n  const buffer = new TensorBuffer([outputSize], $x.dtype);\n  const indices = new TensorBuffer([outputSize], 'int32');\n  for (let i = 0, p = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      buffer.values[p] = xVals[i];\n      indices.values[p] = i;\n      p++;\n    }\n  }\n  return [buffer.toTensor(), indices.toTensor()];\n}\nexport const setdiff1dAsync = setdiff1dAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sign, SignInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns an element-wise indication of the sign of a number.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3, NaN, 0]);\n *\n * x.sign().print();  // or tf.sign(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sign_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sign');\n  const inputs: SignInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.sign($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Sign);\n}\nexport const sign = op({sign_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sin, SinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sin of the input Tensor element-wise: `sin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.sin().print();  // or tf.sin(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sin');\n\n  const inputs: SinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sin);\n}\nexport const sin = op({sin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sinh, SinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic sin of the input `tf.Tensor` element-wise: `sinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.sinh().print();  // or tf.sinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sinh');\n  const inputs: SinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sinh);\n}\nexport const sinh = op({sinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 1D slice from 1D array starting at coordinates `begin` and is\n * of length `size`. See `slice` for details.\n */\nfunction slice1d_(\n    x: Tensor1D|TensorLike, begin: number, size: number): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'slice1d');\n  util.assert(\n      $x.rank === 1,\n      () =>\n          `slice1d expects a rank-1 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, [begin], [size]);\n}\nexport const slice1d = op({slice1d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 2D slice from a 2D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice2d_(\n    x: Tensor2D|TensorLike, begin: [number, number],\n    size: [number, number]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'slice2d');\n  util.assert(\n      $x.rank === 2,\n      () =>\n          `slice2d expects a rank-2 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice2d = op({slice2d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 3D slice from a 3D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice3d_(\n    x: Tensor3D|TensorLike, begin: [number, number, number],\n    size: [number, number, number]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'slice3d');\n  util.assert(\n      $x.rank === 3,\n      () =>\n          `slice3d expects a rank-3 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice3d = op({slice3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 4D slice from a 4D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice4d_(\n    x: Tensor4D|TensorLike, begin: [number, number, number, number],\n    size: [number, number, number, number]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'slice4d');\n  util.assert(\n      $x.rank === 4,\n      () =>\n          `slice4d expects a rank-4 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice4d = op({slice4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softmax, SoftmaxAttrs, SoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the softmax normalized vector given the logits.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction softmax_<T extends Tensor>(logits: T|TensorLike, dim = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'softmax', 'float32');\n\n  if (dim === -1) {\n    dim = $logits.rank - 1;\n  }\n  if (dim !== $logits.rank - 1) {\n    throw Error(\n        'Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and dim was ${dim}`);\n  }\n\n  const inputs: SoftmaxInputs = {logits: $logits};\n  const attrs: SoftmaxAttrs = {dim};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const y = backend.softmax($logits, dim);\n        save([y]);\n        return y;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, Softmax,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const softmax = op({softmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FFT, FFTInputs} from '../../kernel_names';\nimport {Tensor} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\n\n/**\n * Fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the inner-most\n * dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.fft().print();  // tf.spectral.fft(x).print();\n * ```\n * @param input The complex input to compute an fft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction fft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.fft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: FFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D = input.as2D(batch, innerDimensionSize);\n    const result = backend.fft(input2D);\n    return result.reshape(input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, FFT);\n}\n\nexport const fft = op({fft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {IFFT, IFFTInputs} from '../../kernel_names';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Inverse fast Fourier transform.\n *\n * Computes the inverse 1-dimensional discrete Fourier transform over the\n * inner-most dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.ifft().print();  // tf.spectral.ifft(x).print();\n * ```\n * @param input The complex input to compute an ifft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction ifft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.ifft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: IFFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D: Tensor2D = reshape(input, [batch, innerDimensionSize]);\n    const result = backend.ifft(input2D);\n    return reshape(result, input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, IFFT);\n}\n\nexport const ifft = op({ifft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {reverse} from '../reverse';\nimport {scalar} from '../scalar';\nimport {slice} from '../slice';\n\nimport {ifft} from './ifft';\n\n/**\n * Inversed real value input fast Fourier transform.\n *\n * Computes the 1-dimensional inversed discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([0, 0, 0]);\n * const x = tf.complex(real, imag);\n *\n * x.irfft().print();\n * ```\n * @param input The real value input to compute an irfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction irfft_(input: Tensor): Tensor {\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n  let ret: Tensor;\n  if (innerDimensionSize <= 2) {\n    const complexInput = reshape(input, [batch, innerDimensionSize]);\n    ret = ifft(complexInput);\n  } else {\n    // The length of unique components of the DFT of a real-valued signal\n    // is 2 * (input_len - 1)\n    const outputShape = [batch, 2 * (innerDimensionSize - 1)];\n    const realInput = reshape(real(input), [batch, innerDimensionSize]);\n    const imagInput = reshape(imag(input), [batch, innerDimensionSize]);\n\n    const realConjugate =\n        reverse(slice(realInput, [0, 1], [batch, innerDimensionSize - 2]), 1);\n    const imagConjugate: Tensor2D = mul(\n        reverse(slice(imagInput, [0, 1], [batch, innerDimensionSize - 2]), 1),\n        scalar(-1));\n\n    const r = concat([realInput, realConjugate], 1);\n    const i = concat([imagInput, imagConjugate], 1);\n    const complexInput =\n        reshape(complex(r, i), [outputShape[0], outputShape[1]]);\n    ret = ifft(complexInput);\n  }\n  ret = real(ret);\n  // reshape the result if the input is 3D tensor.\n  if (input.rank === 3 && input.shape[0] !== 0) {\n    const temp = ret;\n    const batch = input.shape[0];\n    ret = reshape(ret, [batch, ret.shape[0] / batch, ret.shape[1]]);\n    temp.dispose();\n  }\n  return ret;\n}\n\nexport const irfft = op({irfft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {assert} from '../util';\n\n/**\n * Prepare the split size array. When the input is a number, the axis is evenly\n * divided among the split size. When the input contains the negative value, the\n * rest of the axis is allocated toward that.\n */\nexport function prepareSplitSize(\n    x: Tensor|TensorInfo, numOrSizeSplits: number[]|number,\n    axis = 0): number[] {\n  let splitSizes = [];\n  if (typeof (numOrSizeSplits) === 'number') {\n    assert(\n        x.shape[axis] % numOrSizeSplits === 0,\n        () => 'Number of splits must evenly divide the axis.');\n    splitSizes =\n        new Array(numOrSizeSplits).fill(x.shape[axis] / numOrSizeSplits);\n  } else {\n    const numOfNegs = numOrSizeSplits.reduce((count, value) => {\n      if (value === -1) {\n        count += 1;\n      }\n      return count;\n    }, 0);\n    assert(\n        numOfNegs <= 1,\n        () => 'There should be only one negative value in split array.');\n    const negIndex = numOrSizeSplits.indexOf(-1);\n    // Allow the number of split array to be -1, which indicates the rest\n    // of dimension is allocated to that split.\n    if (negIndex !== -1) {\n      const total = numOrSizeSplits.reduce((a, b) => b > 0 ? a + b : a);\n      numOrSizeSplits[negIndex] = x.shape[axis] - total;\n    }\n    assert(\n        x.shape[axis] === numOrSizeSplits.reduce((a, b) => a + b),\n        () => 'The sum of sizes must match the size of the axis dimension.');\n    splitSizes = numOrSizeSplits;\n  }\n\n  return splitSizes;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SplitV, SplitVAttrs, SplitVInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {prepareSplitSize} from './split_util';\n\n/**\n * Splits a `tf.Tensor` into sub tensors.\n *\n * If `numOrSizeSplits` is a number, splits `x` along dimension `axis`\n * into `numOrSizeSplits` smaller tensors.\n * Requires that `numOrSizeSplits` evenly divides `x.shape[axis]`.\n *\n * If `numOrSizeSplits` is a number array, splits `x` into\n * `numOrSizeSplits.length` pieces. The shape of the `i`-th piece has the\n * same size as `x` except along dimension `axis` where the size is\n * `numOrSizeSplits[i]`.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4, 5, 6, 7, 8], [2, 4]);\n * const [a, b] = tf.split(x, 2, 1);\n * a.print();\n * b.print();\n *\n * const [c, d, e] = tf.split(x, [1, 2, 1], 1);\n * c.print();\n * d.print();\n * e.print();\n * ```\n *\n * @param x The input tensor to split.\n * @param numOrSizeSplits Either an integer indicating the number of\n * splits along the axis or an array of integers containing the sizes of\n * each output tensor along the axis. If a number then it must evenly divide\n * `x.shape[axis]`; otherwise the sum of sizes must match `x.shape[axis]`.\n * Can contain one -1 indicating that dimension is to be inferred.\n * @param axis The dimension along which to split. Defaults to 0 (the first\n * dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction split_<T extends Tensor>(\n    x: Tensor|TensorLike, numOrSizeSplits: number[]|number, axis = 0): T[] {\n  const $x = convertToTensor(x, 'x', 'split');\n\n  const forward: ForwardFunc<Tensor> = (backend, _) => {\n    const $axis = parseAxisParam(axis, $x.shape)[0];\n    const splitSizes = prepareSplitSize($x, numOrSizeSplits, $axis);\n    return backend.split($x, splitSizes, $axis) as {} as T;\n  };\n\n  const inputs: SplitVInputs = {x: $x};\n  const attr: SplitVAttrs = {numOrSizeSplits, axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, SplitV,\n             attr as {} as NamedAttrMap) as {} as T[];\n}\n\nexport const split = op({split_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {assert} from '../../util';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {split} from '../split';\nimport {zeros} from '../zeros';\nimport {zerosLike} from '../zeros_like';\n\nimport {fft} from './fft';\n\n/**\n * Real value input fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n *\n * real.rfft().print();\n * ```\n * @param input The real value input to compute an rfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction rfft_(input: Tensor, fftLength?: number): Tensor {\n  assert(\n      input.dtype === 'float32',\n      () => `The dtype for rfft() must be real value but got ${input.dtype}`);\n\n  let innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n\n  let adjustedInput: Tensor;\n  if (fftLength != null && fftLength < innerDimensionSize) {\n    // Need to crop\n    const begin = input.shape.map(v => 0);\n    const size = input.shape.map(v => v);\n    size[input.shape.length - 1] = fftLength;\n    adjustedInput = slice(input, begin, size);\n    innerDimensionSize = fftLength;\n  } else if (fftLength != null && fftLength > innerDimensionSize) {\n    // Need to pad with zeros\n    const zerosShape = input.shape.map(v => v);\n    zerosShape[input.shape.length - 1] = fftLength - innerDimensionSize;\n    adjustedInput = concat([input, zeros(zerosShape)], input.shape.length - 1);\n    innerDimensionSize = fftLength;\n  } else {\n    adjustedInput = input;\n  }\n\n  // Complement the input with zero imaginary numbers.\n  const zerosInput = zerosLike(adjustedInput);\n  const complexInput =\n      reshape(complex(adjustedInput, zerosInput), [batch, innerDimensionSize]);\n\n  const ret = fft(complexInput);\n\n  // Exclude complex conjugations. These conjugations are put symmetrically.\n  const half = Math.floor(innerDimensionSize / 2) + 1;\n  const realValues = real(ret);\n  const imagValues = imag(ret);\n  const realComplexConjugate = split(\n      realValues, [half, innerDimensionSize - half],\n      realValues.shape.length - 1);\n  const imagComplexConjugate = split(\n      imagValues, [half, innerDimensionSize - half],\n      imagValues.shape.length - 1);\n\n  const outputShape = adjustedInput.shape.slice();\n  outputShape[adjustedInput.shape.length - 1] = half;\n\n  return reshape(\n      complex(realComplexConjugate[0], imagComplexConjugate[0]), outputShape);\n}\n\nexport const rfft = op({rfft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sqrt, SqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes square root of the input `tf.Tensor` element-wise: `y = sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.sqrt().print();  // or tf.sqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sqrt');\n\n  const inputs: SqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sqrt);\n}\nexport const sqrt = op({sqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SquaredDifference, SquaredDifferenceInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns (a - b) * (a - b) element-wise.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * ```js\n * // Broadcast squared difference  a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction squaredDifference_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'squaredDifference');\n  let $b = convertToTensor(b, 'b', 'squaredDifference');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.squaredDifference($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SquaredDifferenceInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             SquaredDifference, attrs) as T;\n}\n\nexport const squaredDifference = op({squaredDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {squeezeShape} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Removes dimensions of size 1 from the shape of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4], [1, 1, 4]);\n * x.squeeze().print();\n * ```\n *\n * @param x The input tensor to be squeezed.\n * @param axis An optional list of numbers. If specified, only\n *     squeezes the dimensions listed. The dimension index starts at 0. It\n * is an error to squeeze a dimension that is not 1.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction squeeze_<T extends Tensor>(x: Tensor|TensorLike, axis?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'squeeze');\n  return reshape($x, squeezeShape($x.shape, axis).newShape) as T;\n}\n\nexport const squeeze = op({squeeze_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {concat} from './concat';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\n\n/**\n * Stacks a list of rank-`R` `tf.Tensor`s into one rank-`(R+1)` `tf.Tensor`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.stack([a, b, c]).print();\n * ```\n *\n * @param tensors A list of tensor objects with the same shape and dtype.\n * @param axis The axis to stack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction stack_<T extends Tensor>(\n    tensors: Array<T|TensorLike>, axis = 0): Tensor {\n  const $tensors = convertToTensorArray(tensors, 'tensors', 'stack');\n\n  util.assert(\n      $tensors.length >= 1, () => 'Pass at least one tensor to tf.stack');\n\n  if ($tensors.length === 1) {\n    return expandDims($tensors[0], axis);\n  }\n\n  const rank = $tensors[0].rank;\n  const shape = $tensors[0].shape;\n  const dtype = $tensors[0].dtype;\n\n  util.assert(axis <= rank, () => 'Axis must be <= rank of the tensor');\n\n  $tensors.forEach(t => {\n    util.assertShapesMatch(\n        shape, t.shape,\n        'All tensors passed to stack must have matching shapes');\n    util.assert(\n        dtype === t.dtype,\n        () => 'All tensors passed to stack must have matching dtypes');\n  });\n\n  const expandedTensors = $tensors.map(t => expandDims(t, axis));\n  // Stack exists in the TensorFlow C++ API\n  // (https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/stack) but not\n  // in\n  // https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/core/ops/ops.pbtxt.\n  // Therefore we are treating it like a high-level op rather than\n  // creating a dedicated stack kernel.\n  return concat(expandedTensors, axis);\n}\n\nexport const stack = op({stack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Step, StepAttrs, StepInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes step of the input `tf.Tensor` element-wise: `x > 0 ? 1 : alpha * x`\n *\n * ```js\n * const x = tf.tensor1d([0, 2, -1, -3]);\n *\n * x.step(.5).print();  // or tf.step(x, .5)\n * ```\n * @param x The input tensor.\n * @param alpha The gradient when input is negative.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction step_<T extends Tensor>(x: T|TensorLike, alpha = 0.0): T {\n  const $x = convertToTensor(x, 'x', 'step');\n\n  const inputs: StepInputs = {x: $x};\n  const attrs: StepAttrs = {alpha};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.step($x, alpha), inputs as {} as NamedTensorMap,\n      null /* grad */, Step, attrs as {} as NamedAttrMap);\n}\nexport const step = op({step_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {StridedSlice, StridedSliceAttrs, StridedSliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {slice} from './slice';\nimport {computeOutShape, getNormalizedAxes, maskToAxes} from './slice_util';\n\n/**\n * Extracts a strided slice of a tensor.\n *\n * Roughly speaking, this op extracts a slice of size (end-begin)/stride from\n * the given input tensor (x). Starting at the location specified by begin the\n * slice continues by adding stride to the index until all dimensions are not\n * less than end. Note that a stride can be negative, which causes a reverse\n * slice.\n *\n * ```js\n * const t = tf.tensor3d([1, 1, 1 ,2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6],\n *    [3, 2, 3]);\n * t.stridedSlice([1, 0, 0], [2, 1, 3], [1, 1, 1]).print()  // [[[3, 3, 3]]]\n * t.stridedSlice([1, 0, 0], [2, 2, 3], [1, 1, 1]).print()  // [[[3, 3, 3],\n *                                                     // [4, 4, 4]]]\n * t.stridedSlice([1, -1, 0], [2, -3, 3], [1, -1, 1]).print() // [[[4, 4, 4],\n *                                                     // [3, 3, 3]]]\n * ```\n *\n * @param x The tensor to stride slice.\n * @param begin The coordinates to start the slice from.\n * @param end: The coordinates to end the slice at.\n * @param strides: The size of the slice.\n * @param beginMask: If the ith bit of beginMask is set, begin[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param endMask: If the ith bit of endMask is set, end[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param shrinkAxisMask: a bitmask where bit i implies that\n * the ith specification should shrink the dimensionality. begin and end must\n * imply a slice of size 1 in the dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction stridedSlice_(\n    x: Tensor|TensorLike, begin: number[], end: number[], strides?: number[],\n    beginMask = 0, endMask = 0, ellipsisMask = 0, newAxisMask = 0,\n    shrinkAxisMask = 0): Tensor {\n  let $x = convertToTensor(x, 'x', 'stridedSlice');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if (strides == null) {\n      strides = new Array(begin.length);\n    }\n\n    const ellipsisAxes = maskToAxes(ellipsisMask);\n    if (ellipsisAxes.length > 1) {\n      throw new Error('Multiple ellipses in slice is not allowed.');\n    }\n\n    if (ellipsisMask !== 0 && newAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and newAxisMask is not yet supported.');\n    }\n\n    if (ellipsisMask !== 0 && shrinkAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and shrinkAxisMask is not yet supported.');\n    }\n\n    const numInterpolatedAxes = $x.rank - begin.length;\n\n    // Expand the dims of x based on the newAxisMask.\n    const expandAxes = maskToAxes(newAxisMask);\n    const newShape = $x.shape.slice();\n    expandAxes.forEach(axis => {\n      begin[axis] = 0;\n      end[axis] = 1;\n      newShape.splice(axis, 0, 1);\n    });\n    $x = reshape($x, newShape);\n\n    const {\n      begin: normalizedBegin,\n      end: normalizedEnd,\n      strides: normalizedStrides\n    } =\n        getNormalizedAxes(\n            $x.shape, ellipsisAxes, numInterpolatedAxes, begin, end, strides,\n            beginMask, endMask, ellipsisMask);\n    begin = normalizedBegin;\n    end = normalizedEnd;\n    strides = normalizedStrides;\n\n    const shrinkAxes = maskToAxes(shrinkAxisMask);\n    // Adjust the ends based on the shrink mask.\n    shrinkAxes.forEach(axis => {\n      end[axis] = begin[axis] + 1;\n      strides[axis] = 1;\n    });\n\n    // Figure out the output shape.\n    const size = computeOutShape(begin, end, strides);\n    // Remove the axes based on shrinkMask.\n    const outShape = size.filter((_, axis) => shrinkAxes.indexOf(axis) === -1);\n\n    const nonStrided = strides.every(v => v === 1);\n    if (nonStrided) {\n      return reshape(slice($x, begin, size), outShape);\n    }\n\n    const res = backend.stridedSlice($x, begin, end, strides);\n    return reshape(res, outShape);\n  };\n\n  const inputs: StridedSliceInputs = {x: $x};\n  const attrs: StridedSliceAttrs = {\n    begin,\n    end,\n    strides,\n    beginMask,\n    endMask,\n    ellipsisMask,\n    newAxisMask,\n    shrinkAxisMask\n  };\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, StridedSlice,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const stridedSlice = op({stridedSlice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tan, TanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes tan of the input `tf.Tensor` element-wise, `tan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.tan().print();  // or tf.tan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tan');\n\n  const inputs: TanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.tan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tan);\n}\nexport const tan = op({tan_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike2D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-2 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor2d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor2d([[1, 2], [3, 4]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor2d(\n    values: TensorLike2D, shape?: [number, number],\n    dtype?: DataType): Tensor2D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 2) {\n    throw new Error('tensor2d() requires shape to have two numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 2 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor2d() requires values to be number[][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor2d() requires shape to be provided when `values` ' +\n        'are a flat/TypedArray');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-4 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor4d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor4d([[[[1], [2]], [[3], [4]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor4d(\n    values: TensorLike4D, shape?: [number, number, number, number],\n    dtype?: DataType): Tensor4D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 4) {\n    throw new Error('tensor4d() requires shape to have four numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 4 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor4d() requires values to be number[][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor4d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor4D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor5D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike5D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-5 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor5d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor5d([[[[[1], [2]], [[3], [4]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor5d(\n    values: TensorLike5D, shape?: [number, number, number, number, number],\n    dtype?: DataType): Tensor5D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 5) {\n    throw new Error('tensor5d() requires shape to have five numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 5 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor5d() requires values to be ' +\n        'number[][][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor5d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor5D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor6D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike6D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-6 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor6d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor6d([[[[[[1],[2]],[[3],[4]]],[[[5],[6]],[[7],[8]]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor6d([1, 2, 3, 4, 5, 6, 7, 8], [1, 1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor6d(\n    values: TensorLike6D,\n    shape?: [number, number, number, number, number, number],\n    dtype?: DataType): Tensor6D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 6) {\n    throw new Error('tensor6d() requires shape to have six numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 6 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor6d() requires values to be number[][][][][][] or ' +\n        'flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor6d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  shape = shape ||\n      inferredShape as [number, number, number, number, number, number];\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor6D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {TopK, TopKAttrs, TopKInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {NumericTensor, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Finds the values and indices of the `k` largest entries along the last\n * dimension.\n *\n * If the input is a vector (rank=1), finds the k largest entries in the vector\n * and outputs their values and indices as vectors. Thus values[j] is the j-th\n * largest entry in input, and its index is indices[j].\n * For higher rank inputs, computes the top k entries along the last dimension.\n *\n * If two elements are equal, the lower-index element appears first.\n *\n * ```js\n * const a = tf.tensor2d([[1, 5], [4, 3]]);\n * const {values, indices} = tf.topk(a);\n * values.print();\n * indices.print();\n * ```\n * @param x 1-D or higher `tf.Tensor` with last dimension being at least `k`.\n * @param k Number of top elements to look for along the last dimension.\n * @param sorted If true, the resulting `k` elements will be sorted by the\n *     values in descending order.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction topk_<T extends Tensor>(\n    x: T|TensorLike, k = 1, sorted = true): {values: T, indices: T} {\n  const $x = convertToTensor(x, 'x', 'topk');\n  if ($x.rank === 0) {\n    throw new Error('topk() expects the input to be of rank 1 or higher');\n  }\n  const lastDim = $x.shape[$x.shape.length - 1];\n  if (k > lastDim) {\n    throw new Error(\n        `'k' passed to topk() must be <= the last dimension (${lastDim}) ` +\n        `but got ${k}`);\n  }\n\n  const inputs: TopKInputs = {x: $x};\n  const attrs: TopKAttrs = {k, sorted};\n\n  const [values, indices] = ENGINE.runKernelFunc(\n      b => b.topk($x as NumericTensor, k, sorted),\n      inputs as {} as NamedTensorMap, null /* grad */, TopK,\n      attrs as {} as NamedAttrMap);\n\n  return {values, indices} as {values: T, indices: T};\n}\n\nexport const topk = op({topk_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a truncated normal\n * distribution.\n *\n * ```js\n * tf.truncatedNormal([2, 2]).print();\n * ```\n *\n * The generated values follow a normal distribution with specified mean and\n * standard deviation, except that values whose magnitude is more than 2\n * standard deviations from the mean are dropped and re-picked.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output tensor.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction truncatedNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type $ { dtype }`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, true /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const truncatedNormal = op({truncatedNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Unique, UniqueAttrs, UniqueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Finds unique elements along an axis of a tensor.\n *\n * It returns a tensor `values` containing all of the unique elements along the\n * `axis` of the given tensor `x` in the same order that they occur along the\n * `axis` in `x`; `x` does not need to be sorted. It also returns a tensor\n * `indices` the same size as the number of the elements in `x` along the `axis`\n * dimension. It contains the index in the unique output `values`.\n *\n * ```js\n * // A 1-D tensor\n * const a = tf.tensor1d([1, 1, 2, 4, 4, 4, 7, 8, 8]);\n * const {values, indices} = tf.unique(a);\n * values.print();   // [1, 2, 4, 7, 8,]\n * indices.print();  // [0, 0, 1, 2, 2, 2, 3, 4, 4]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=0\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 0)\n * values.print();   // [[1, 0, 0],\n *                   //  [2, 0, 0]]\n * indices.print();  // [0, 0, 1]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=1\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 1)\n * values.print();   // [[1, 0],\n *                   //  [1, 0],\n *                   //  [2, 0]]\n * indices.print();  // [0, 1, 1]\n * ```\n * @param x A tensor (int32, string, bool).\n * @param axis The axis of the tensor to find the unique elements.\n * @returns [uniqueElements, indices] (see above for details)\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction unique_<T extends Tensor>(\n    x: T|TensorLike, axis = 0): {values: T, indices: Tensor1D} {\n  // x can be of any dtype, thus null as the last argument.\n  const $x = convertToTensor(x, 'x', 'unique', null);\n  assert($x.rank > 0, () => 'The input tensor must be at least 1D');\n\n  const inputs: UniqueInputs = {x: $x};\n  const attrs: UniqueAttrs = {axis};\n  const [values, indices] = ENGINE.runKernel(\n                                Unique, inputs as {} as NamedTensorMap,\n                                attrs as {} as NamedAttrMap) as [T, Tensor1D];\n  return {values, indices};\n}\n\nexport const unique = op({unique_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {UnsortedSegmentSum, UnsortedSegmentSumAttrs, UnsortedSegmentSumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, isInt} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Computes the sum along segments of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const segmentIds = tf.tensor1d([1, 2, 0, 1], 'int32');\n * const numSegments = 3;\n *\n * x.unsortedSegmentSum(segmentIds, numSegments).print()\n * //or tf.unsortedSegmentSum(x, segmentIds, numSegments)\n * ```\n * @param x The `tf.Tensor` that will be summed along its segments.\n * @param segmentIds A `tf.Tensor1D` whose rank is equal to the rank of `x`'s\n * dimension along the `axis`.  Maps each element of `x` to a segment.\n * @param numSegments The number of distinct `segmentIds`.\n *\n * @doc {heading: 'Operations', subheading: 'Segment'}\n */\nfunction unsortedSegmentSum_<T extends Tensor>(\n    x: T|TensorLike, segmentIds: Tensor1D|TensorLike, numSegments: number): T {\n  const $x = convertToTensor(x, 'x', 'unsortedSegmentSum');\n  const $segmentIds =\n      convertToTensor(segmentIds, 'segmentIds', 'unsortedSegmentSum', 'int32');\n  assert(isInt(numSegments), () => 'numSegments must be of dtype int');\n\n  const inputs: UnsortedSegmentSumInputs = {x: $x, segmentIds: $segmentIds};\n  const attrs: UnsortedSegmentSumAttrs = {numSegments};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.unsortedSegmentSum($x, $segmentIds, numSegments);\n    save([$segmentIds]);\n    return res;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             UnsortedSegmentSum, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const unsortedSegmentSum = op({unsortedSegmentSum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Unpack, UnpackAttrs, UnpackInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Unstacks a `tf.Tensor` of rank-`R` into a list of rank-`(R-1)` `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * tf.unstack(a).forEach(tensor => tensor.print());\n * ```\n *\n * @param x A tensor object.\n * @param axis The axis to unstack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction unstack_(x: Tensor|TensorLike, axis = 0): Tensor[] {\n  const $x = convertToTensor(x, 'x', 'unstack');\n  util.assert(\n      axis >= -$x.shape.length && axis < $x.shape.length,\n      () =>\n          `Axis = ${axis} is not in [-${$x.shape.length}, ${$x.shape.length})`);\n  if (axis < 0) {\n    axis += $x.shape.length;\n  }\n  const inputs: UnpackInputs = {value: $x};\n  const attrs: UnpackAttrs = {axis};\n  const forward: ForwardFunc<Tensor[]> = (backend: KernelBackend) =>\n      backend.unstack($x, axis);\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Unpack,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const unstack = op({unstack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor, Variable} from '../tensor';\nimport {DataType, Rank} from '../types';\n\n/**\n * Creates a new variable with the provided initial value.\n * ```js\n * const x = tf.variable(tf.tensor([1, 2, 3]));\n * x.assign(tf.tensor([4, 5, 6]));\n *\n * x.print();\n * ```\n *\n * @param initialValue Initial value for the tensor.\n * @param trainable If true, optimizers are allowed to update it.\n * @param name Name of the variable. Defaults to a unique id.\n * @param dtype If set, initialValue will be converted to the given type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function variable<R extends Rank>(\n    initialValue: Tensor<R>, trainable = true, name?: string,\n    dtype?: DataType): Variable<R> {\n  return ENGINE.makeVariable(initialValue, trainable, name, dtype) as\n      Variable<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the Where kernel shared between cpu and webgl */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor2D} from '../tensor';\nimport {TypedArray} from '../types';\n\nexport function whereImpl(condShape: number[], condVals: TypedArray): Tensor2D {\n  const indices = [];\n  for (let i = 0; i < condVals.length; i++) {\n    if (condVals[i]) {\n      indices.push(i);\n    }\n  }\n\n  const inBuffer = buffer(condShape, 'int32');\n\n  const out = buffer([indices.length, condShape.length], 'int32');\n  for (let i = 0; i < indices.length; i++) {\n    const loc = inBuffer.indexToLoc(indices[i]);\n    const offset = i * condShape.length;\n    out.values.set(loc, offset);\n  }\n  return out.toTensor() as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {whereImpl} from '../backends/where_impl';\nimport {Tensor, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\n/**\n * Returns the coordinates of true elements of condition.\n *\n * The coordinates are returned in a 2-D tensor where the first dimension (rows)\n * represents the number of true elements, and the second dimension (columns)\n * represents the coordinates of the true elements. Keep in mind, the shape of\n * the output tensor can vary depending on how many true values there are in\n * input. Indices are output in row-major order. The resulting tensor has the\n * shape `[numTrueElems, condition.rank]`.\n *\n * This is analogous to calling the python `tf.where(cond)` without an x or y.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const result = await tf.whereAsync(cond);\n * result.print();\n * ```\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nasync function whereAsync_(condition: Tensor|TensorLike): Promise<Tensor2D> {\n  const $condition =\n      convertToTensor(condition, 'condition', 'whereAsync', 'bool');\n  const vals = await $condition.data();\n  const res = whereImpl($condition.shape, vals);\n  if (condition !== $condition) {\n    $condition.dispose();\n  }\n  return res;\n}\n\nexport const whereAsync = whereAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {gather} from './gather';\nimport {reshape} from './reshape';\nimport {squeeze} from './squeeze';\nimport {whereAsync} from './where_async';\n\n/**\n * Apply boolean mask to tensor.\n *\n * ```js\n * const tensor = tf.tensor2d([1, 2, 3, 4, 5, 6], [3, 2]);\n * const mask = tf.tensor1d([1, 0, 1], 'bool');\n * const result = await tf.booleanMaskAsync(tensor, mask);\n * result.print();\n * ```\n *\n * @param tensor N-D tensor.\n * @param mask K-D boolean tensor, K <= N and K must be known statically.\n * @param axis A 0-D int Tensor representing the axis in tensor to mask from.\n *     By default, axis is 0 which will mask from the first dimension.\n *     Otherwise K + axis <= N.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nasync function booleanMaskAsync_(\n    tensor: Tensor|TensorLike, mask: Tensor|TensorLike,\n    axis?: number): Promise<Tensor> {\n  const $tensor = convertToTensor(tensor, 'tensor', 'boolMask');\n  const $mask = convertToTensor(mask, 'mask', 'boolMask', 'bool');\n\n  const axisFrom = axis == null ? 0 : axis;\n  const maskDim = $mask.rank;\n  const tensorShape = $tensor.shape;\n\n  util.assert(maskDim > 0, () => 'mask cannot be scalar');\n  util.assertShapesMatch(\n      tensorShape.slice(axisFrom, axisFrom + maskDim), $mask.shape,\n      `mask's shape must match the first K dimensions of tensor's shape,`);\n\n  let leadingSize = 1;\n  for (let i = axisFrom; i < axisFrom + maskDim; i++) {\n    leadingSize *= tensorShape[i];\n  }\n  const targetTensorShape =\n      tensorShape.slice(0, axisFrom)\n          .concat([leadingSize], tensorShape.slice(axisFrom + maskDim));\n  const reshapedTensor = reshape($tensor, targetTensorShape);\n  const reshapedMask = reshape($mask, [-1]);\n  const positivePositions = await whereAsync(reshapedMask);\n  const indices = squeeze(positivePositions, [1]);\n\n  const res = gather(reshapedTensor, indices, axisFrom);\n\n  // Ensure no memory leak.\n  if (tensor !== $tensor) {\n    $tensor.dispose();\n  }\n  if (mask !== $mask) {\n    $mask.dispose();\n  }\n  indices.dispose();\n  reshapedTensor.dispose();\n  reshapedMask.dispose();\n  positivePositions.dispose();\n\n  return res;\n}\n\nexport const booleanMaskAsync = booleanMaskAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertShapesMatch} from '../util';\n\nimport {equal} from './equal';\nimport {greater} from './greater';\nimport {greaterEqual} from './greater_equal';\nimport {less} from './less';\nimport {lessEqual} from './less_equal';\nimport {notEqual} from './not_equal';\nimport {op} from './operation';\n\n/**\n * @deprecated\n * Strict version of `tf.notEqual` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction notEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'notEqualStrict');\n  const $b = convertToTensor(b, 'b', 'notEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in notEqualStrict: ');\n  return notEqual($a, $b);\n}\n\n/**\n * @deprecated\n * Strict version of `tf.less` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction lessStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessStrict');\n  const $b = convertToTensor(b, 'b', 'lessStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessStrict: ');\n  return less($a, $b);\n}\n\nfunction equalStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'equalStrict');\n  const $b = convertToTensor(b, 'b', 'equalStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in equalStrict: ');\n  return equal($a, $b);\n}\n\nfunction lessEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessEqualStrict');\n  const $b = convertToTensor(b, 'b', 'lessEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessEqualStrict: ');\n  return lessEqual($a, $b);\n}\n\nfunction greaterStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterStrict');\n  const $b = convertToTensor(b, 'b', 'greaterStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterStrict: ');\n  return greater($a, $b);\n}\n\nfunction greaterEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterEqualStrict');\n  const $b = convertToTensor(b, 'b', 'greaterEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterEqualStrict: ');\n  return greaterEqual($a, $b);\n}\n\nexport const equalStrict = op({equalStrict_});\nexport const greaterEqualStrict = op({greaterEqualStrict_});\nexport const greaterStrict = op({greaterStrict_});\nexport const lessEqualStrict = op({lessEqualStrict_});\nexport const lessStrict = op({lessStrict_});\nexport const notEqualStrict = op({notEqualStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {maximum} from './maximum';\nimport {minimum} from './minimum';\nimport {mod} from './mod';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {squaredDifference} from './squared_difference';\nimport {sub} from './sub';\n\n/**\n * @deprecated\n * Adds two `tf.Tensor`s element-wise, A + B.\n *\n * Inputs must be the same shape. For broadcasting support, use add() instead.\n *\n * @param a The first Tensor to add element-wise.\n * @param b The second Tensor to add element-wise.\n */\nfunction addStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'addStrict');\n  const $b = convertToTensor(b, 'b', 'addStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in addStrict: ');\n  return add($a, $b);\n}\n\n/**\n * @deprecated\n * Subtracts two `tf.Tensor`s element-wise, A - B. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.sub` instead.\n *\n * @param a The first Tensor to subtract element-wise.\n * @param b The second Tensor to subtract element-wise.\n */\nfunction subStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'subStrict');\n  const $b = convertToTensor(b, 'b', 'subStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in subStrict: ');\n  return sub($a, $b);\n}\n\n/**\n * @deprecated\n * Computes the power of one `tf.Tensor` to another. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.pow` instead.\n *\n * @param base The base tensor to pow element-wise.\n * @param exp The exponent tensor to pow element-wise.\n */\nfunction powStrict_<T extends Tensor>(base: T, exp: Tensor): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  util.assertShapesMatch(base.shape, exp.shape, 'Error in powStrict: ');\n  return pow(base, exp);\n}\n\n/**\n * @deprecated\n * Multiplies two `tf.Tensor`s element-wise, A * B.\n *\n * Inputs must be the same shape. For broadcasting support, use `tf.mul`.\n *\n * @param a The first tensor to multiply.\n * @param b The first tensor to multiply. Must have the same\n *    dtype as `a`.\n */\nfunction mulStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'mul');\n  const $b = convertToTensor(b, 'b', 'mul');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in multiplyStrict: ');\n  return mul($a, $b);\n}\n\n/**\n * @deprecated\n * Divides two `tf.Tensor`s element-wise, A / B. Inputs must\n * be the same shape.\n *\n * @param a The first tensor as the numerator for element-wise division.\n * @param b The second tensor as the denominator for element-wise division.\n */\nfunction divStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'div');\n  const $b = convertToTensor(b, 'b', 'div');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in divideStrict: ');\n  return div($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the mod of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use mod().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction modStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'modStrict');\n  const $b = convertToTensor(b, 'b', 'modStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in modStrict: ');\n  return mod($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the min of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use minimum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction minimumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'minimumStrict');\n  const $b = convertToTensor(b, 'b', 'minimumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in minimumStrict: ');\n  return minimum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the max of a and b (`a > b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use maximum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction maximumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'maximumStrict');\n  const $b = convertToTensor(b, 'b', 'maximumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in maximumStrict: ');\n  return maximum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns (a - b) * (a - b) element-wise.\n *\n * Inputs must be the same shape. For broadcasting support, use\n * `tf.squaredDifference` instead.\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n */\nfunction squaredDifferenceStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'squaredDifferenceStrict');\n  const $b = convertToTensor(b, 'b', 'squaredDifferenceStrict');\n  util.assertShapesMatch(\n      $a.shape, $b.shape, 'Error in squaredDifferenceStrict: ');\n  return squaredDifference($a, $b);\n}\n\nexport const addStrict = op({addStrict_});\nexport const divStrict = op({divStrict_});\nexport const maximumStrict = op({maximumStrict_});\nexport const minimumStrict = op({minimumStrict_});\nexport const modStrict = op({modStrict_});\nexport const mulStrict = op({mulStrict_});\nexport const powStrict = op({powStrict_});\nexport const squaredDifferenceStrict = op({squaredDifferenceStrict_});\nexport const subStrict = op({subStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {abs} from './abs';\nimport * as axis_util from './axis_util';\nimport {max} from './max';\nimport {min} from './min';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sqrt} from './sqrt';\nimport {square} from './square';\nimport {sum} from './sum';\n\n/**\n * Computes the norm of scalar, vectors, and matrices.\n * This function can compute several different vector norms (the 1-norm, the\n * Euclidean or 2-norm, the inf-norm, and in general the p-norm for p > 0)\n * and matrix norms (Frobenius, 1-norm, and inf-norm).\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.norm().print();  // or tf.norm(x)\n * ```\n *\n * @param x The input array.\n * @param ord Optional. Order of the norm. Supported norm types are\n * following:\n *\n *  | ord        | norm for matrices         | norm for vectors\n *  |------------|---------------------------|---------------------\n *  |'euclidean' |Frobenius norm             |2-norm\n *  |'fro'       |Frobenius norm\t           |\n *  |Infinity    |max(sum(abs(x), axis=1))   |max(abs(x))\n *  |-Infinity   |min(sum(abs(x), axis=1))   |min(abs(x))\n *  |1           |max(sum(abs(x), axis=0))   |sum(abs(x))\n *  |2           |                           |sum(abs(x)^2)^1/2*\n *\n * @param axis Optional. If axis is null (the default), the input is\n * considered a vector and a single vector norm is computed over the entire\n * set of values in the Tensor, i.e. norm(x, ord) is equivalent\n * to norm(x.reshape([-1]), ord). If axis is a integer, the input\n * is considered a batch of vectors, and axis determines the axis in x\n * over which to compute vector norms. If axis is a 2-tuple of integer it is\n * considered a batch of matrices and axis determines the axes in NDArray\n * over which to compute a matrix norm.\n * @param keepDims Optional. If true, the norm have the same dimensionality\n * as the input.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction norm_(\n    x: Tensor|TensorLike, ord: number|'euclidean'|'fro' = 'euclidean',\n    axis: number|number[] = null, keepDims = false): Tensor {\n  x = convertToTensor(x, 'x', 'norm');\n\n  const norm = normImpl(x, ord, axis);\n  let keepDimsShape = norm.shape;\n  if (keepDims) {\n    const axes = parseAxisParam(axis, x.shape);\n    keepDimsShape = axis_util.expandShapeToKeepDim(norm.shape, axes);\n  }\n  return reshape(norm, keepDimsShape);\n}\n\nfunction normImpl(\n    x: Tensor, p: number|string, axis: number|number[] = null): Tensor {\n  if (x.rank === 0) {\n    return abs(x);\n  }\n\n  // consider vector when no axis is specified\n  if (x.rank !== 1 && axis === null) {\n    return normImpl(reshape(x, [-1]), p, axis);\n  }\n\n  // vector\n  if (x.rank === 1 || typeof axis === 'number' ||\n      Array.isArray(axis) && axis.length === 1) {\n    if (p === 1) {\n      return sum(abs(x), axis);\n    }\n    if (p === Infinity) {\n      return max(abs(x), axis);\n    }\n    if (p === -Infinity) {\n      return min(abs(x), axis);\n    }\n    if (p === 'euclidean' || p === 2) {\n      // norm(x, 2) = sum(abs(xi) ^ 2) ^ 1/2\n      return sqrt(sum(pow(abs(x), scalar(2, 'int32')), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  // matrix (assumption axis[0] < axis[1])\n  if (Array.isArray(axis) && axis.length === 2) {\n    if (p === 1) {\n      return max(sum(abs(x), axis[0]), axis[1] - 1);\n    }\n    if (p === Infinity) {\n      return max(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === -Infinity) {\n      return min(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === 'fro' || p === 'euclidean') {\n      // norm(x) = sqrt(sum(pow(x, 2)))\n      return sqrt(sum(square(x), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  throw new Error(`Error in norm: invalid axis: ${axis}`);\n}\n\nexport const norm = op({norm_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor} from '../tensor';\nimport {assertTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {scalar} from './scalar';\nimport {sub} from './sub';\n\n/**\n * Compute the moving average of a variable.\n *\n * Without zeroDebias, the moving average operation is defined by:\n *   `v += delta`\n * where\n *   `delta = (1 - decay) * (x - v)`\n *\n * With zeroDebias (default), the `delta` term is scaled to debias the\n * effect of the (assumed) zero-initialization of `v`.\n *   `delta /= (1 - decay ^ step)`\n *\n * For more details on the zero-debiasing algorithm, see:\n *   https://arxiv.org/abs/1412.6980\n *\n * Note that this function is completely stateless and does not keep track of\n * step count. The step count needs to be maintained by the caller and passed\n * in as `step`.\n *\n * @param v The current moving average value.\n * @param x New input value, must have the same shape and dtype as `v`.\n * @param decay The decay factor. Typical values are 0.95 and 0.99.\n * @param step Step count.\n * @param zeroDebias: Whether zeroDebias is to be performed (default: `true`).\n * @returns The new moving average value.\n *\n * @doc {heading: 'Operations', subheading: 'Moving Average'}\n */\nfunction movingAverage_<T extends Tensor>(\n    v: T|TensorLike, x: T|TensorLike, decay: number|Scalar,\n    step?: number|Scalar, zeroDebias = true): T {\n  const $v = convertToTensor(v, 'v', 'movingAverage');\n  const $x = convertToTensor(x, 'x', 'movingAverage');\n  const $decay = convertToTensor(decay, 'decay', 'movingAverage');\n\n  assertTypesMatch($v, $x);\n  util.assert(\n      util.arraysEqual($v.shape, $x.shape), () => 'Shape mismatch in v and x');\n\n  const one = scalar(1);\n  const oneMinusDecay = sub(one, $decay);\n\n  let update = mul(sub($x, $v), oneMinusDecay);\n  if (zeroDebias) {\n    util.assert(\n        step != null, () => 'When using zeroDebias: true, step is required.');\n    const $step = convertToTensor(step, 'step', 'movingAverage');\n    update = div(update, sub(one, pow($decay, $step)));\n  }\n  return add($v, update);\n}\n\nexport const movingAverage = op({movingAverage_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ScatterNd, ScatterNdAttrs, ScatterNdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as scatter_nd_util from './scatter_nd_util';\n\n/**\n * Creates a new tensor by applying sparse updates to individual\n * values or slices within a zero tensor of the given shape tensor according to\n * indices. This operator is the inverse of the `tf.gatherND` operator which\n * extracts values or slices from a given tensor.\n *\n * ```js\n * const indices = tf.tensor2d([4, 3, 1, 7], [4, 1], 'int32');\n * const updates = tf.tensor1d([9, 10, 11, 12]);\n * const shape = [8];\n * tf.scatterND(indices, updates, shape).print() //[0, 11, 0, 10, 9, 0, 0, 12]\n * ```\n *\n * @param indices The tensor contains the indices into the output tensor.\n * @param updates The tensor contains the value for the indices.\n * @param shape: The shape of the output tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction scatterND_<R extends Rank>(\n    indices: Tensor|TensorLike, updates: Tensor|TensorLike,\n    shape: ShapeMap[R]): Tensor<R> {\n  const $indices = convertToTensor(indices, 'indices', 'scatterND', 'int32');\n  const $updates = convertToTensor(updates, 'updates', 'scatterND');\n  scatter_nd_util.validateInput($updates, $indices, shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.scatterND($indices, $updates, shape);\n  };\n\n  const inputs: ScatterNdInputs = {indices: $indices, updates: $updates};\n  const attrs: ScatterNdAttrs = {shape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             ScatterNd, attrs as {} as NamedAttrMap) as Tensor<R>;\n}\n\nexport const scatterND = op({scatterND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\n\n/**\n * Validate sparseToDense inputs.\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape number[]. Shape of the dense output tensor.\n * @param validateIndices boolean. indice validation is not supported, error\n * will be thrown if it is set.\n */\nexport function validateInput(\n    sparseIndices: Tensor, sparseValues: Tensor, outputShape: number[],\n    defaultValues: Tensor) {\n  if (sparseIndices.dtype !== 'int32') {\n    throw new Error(\n        'tf.sparseToDense() expects the indices to be int32 type,' +\n        ` but the dtype was ${sparseIndices.dtype}.`);\n  }\n  if (sparseIndices.rank > 2) {\n    throw new Error(\n        'sparseIndices should be a scalar, vector, or matrix,' +\n        ` but got shape ${sparseIndices.shape}.`);\n  }\n\n  const numElems = sparseIndices.rank > 0 ? sparseIndices.shape[0] : 1;\n  const numDims = sparseIndices.rank > 1 ? sparseIndices.shape[1] : 1;\n\n  if (outputShape.length !== numDims) {\n    throw new Error(\n        'outputShape has incorrect number of elements:,' +\n        ` ${outputShape.length}, should be: ${numDims}.`);\n  }\n\n  const numValues = sparseValues.size;\n  if (!(sparseValues.rank === 0 ||\n        sparseValues.rank === 1 && numValues === numElems)) {\n    throw new Error(\n        'sparseValues has incorrect shape ' +\n        `${sparseValues.shape}, should be [] or [${numElems}]`);\n  }\n\n  if (sparseValues.dtype !== defaultValues.dtype) {\n    throw new Error('sparseValues.dtype must match defaultValues.dtype');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {SparseToDense, SparseToDenseAttrs, SparseToDenseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport * as sparse_to_dense from '../ops/sparse_to_dense_util';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ScalarLike, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Converts a sparse representation into a dense tensor.\n *\n * Builds an array dense with shape outputShape such that:\n *\n * // If sparseIndices is scalar\n * dense[i] = (i == sparseIndices ? sparseValues : defaultValue)\n *\n * // If sparseIndices is a vector, then for each i\n * dense[sparseIndices[i]] = sparseValues[i]\n *\n * // If sparseIndices is an n by d matrix, then for each i in [0, n)\n * dense[sparseIndices[i][0], ..., sparseIndices[i][d-1]] = sparseValues[i]\n * All other values in dense are set to defaultValue. If sparseValues is a\n * scalar, all sparse indices are set to this single value.\n *\n * If indices are repeated the final value is summed over all values for those\n * indices.\n *\n * ```js\n * const indices = tf.tensor1d([4, 5, 6, 1, 2, 3], 'int32');\n * const values = tf.tensor1d([10, 11, 12, 13, 14, 15], 'float32');\n * const shape = [8];\n * tf.sparseToDense(indices, values, shape).print();\n * ```\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape Shape of the dense output tensor. the type is inferred.\n * @param defaultValue Scalar. Value to set for indices not specified in\n * sparseIndices. Defaults to zero.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction sparseToDense_<R extends Rank>(\n    sparseIndices: Tensor|TensorLike, sparseValues: Tensor|TensorLike,\n    outputShape: ShapeMap[R], defaultValue: Scalar|ScalarLike = 0): Tensor<R> {\n  const $sparseIndices =\n      convertToTensor(sparseIndices, 'sparseIndices', 'sparseToDense', 'int32');\n  const $sparseValues =\n      convertToTensor(sparseValues, 'sparseValues', 'sparseToDense');\n  const $defaultValue = convertToTensor(\n      defaultValue, 'defaultValue', 'sparseToDense', $sparseValues.dtype);\n\n  sparse_to_dense.validateInput(\n      $sparseIndices, $sparseValues, outputShape, $defaultValue);\n\n  const inputs: SparseToDenseInputs = {\n    sparseIndices: $sparseIndices,\n    sparseValues: $sparseValues,\n    defaultValue: $defaultValue\n  };\n\n  const attrs: SparseToDenseAttrs = {outputShape};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.sparseToDense(\n          $sparseIndices, $sparseValues, outputShape, $defaultValue),\n      inputs as {} as NamedTensorMap, null /* grad */, SparseToDense,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const sparseToDense = op({sparseToDense_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherNd, GatherNdInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Gather slices from input tensor into a Tensor with shape specified by\n * `indices`.\n *\n * `indices` is an K-dimensional integer tensor, best thought of as a\n * (K-1)-dimensional tensor of indices into input, where each element defines a\n * slice of input:\n * output[\\\\(i_0, ..., i_{K-2}\\\\)] = input[indices[\\\\(i_0, ..., i_{K-2}\\\\)]]\n *\n * Whereas in `tf.gather`, `indices` defines slices into the first dimension of\n * input, in `tf.gatherND`, `indices` defines slices into the first N dimensions\n * of input, where N = indices.shape[-1].\n *\n * The last dimension of indices can be at most the rank of input:\n * indices.shape[-1] <= input.rank\n *\n * The last dimension of `indices` corresponds to elements\n * (if indices.shape[-1] == input.rank) or slices\n * (if indices.shape[-1] < input.rank) along dimension indices.shape[-1] of\n * input.\n * The output tensor has shape\n * indices.shape[:-1] + input.shape[indices.shape[-1]:]\n *\n * Note that on CPU, if an out of bound index is found, an error is returned. On\n * GPU, if an out of bound index is found, a 0 is stored in the corresponding\n * output value.\n *\n * ```js\n * const indices = tf.tensor2d([0, 1, 1, 0], [2,2], 'int32');\n * const input = tf.tensor2d([9, 10, 11, 12], [2, 2]);\n * tf.gatherND(input, indices).print() // [10, 11]\n * ```\n *\n * @param x The tensor from which to gather values.\n * @param indices Index tensor, must be of type int32.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction gatherND_(x: Tensor|TensorLike, indices: Tensor|TensorLike): Tensor {\n  const $indices = convertToTensor(indices, 'indices', 'gatherND', 'int32');\n  const $x = convertToTensor(x, 'x', 'gatherND');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.gatherND($x, $indices);\n  };\n\n  const inputs: GatherNdInputs = {params: $x, indices: $indices};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */, GatherNd);\n}\n\nexport const gatherND = op({gatherND_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\n/**\n * Normalize noise shape based on provided tensor and noise shape.\n *\n * @param x Tensor.\n * @param noiseShape The shape for the randomly generated keep/drop flags, as\n *   an array of numbers. Optional.\n * @returns Normalized noise shape.\n */\nexport function getNoiseShape(x: Tensor, noiseShape?: number[]): number[] {\n  if (noiseShape == null) {\n    return x.shape.slice();\n  }\n  if (util.arraysEqual(x.shape, noiseShape)) {\n    return noiseShape;\n  }\n  if (x.shape.length === noiseShape.length) {\n    const newDimension: number[] = [];\n    for (let i = 0; i < x.shape.length; i++) {\n      if (noiseShape[i] == null && x.shape[i] != null) {\n        newDimension.push(x.shape[i]);\n      } else {\n        newDimension.push(noiseShape[i]);\n      }\n    }\n    return newDimension;\n  }\n\n  return noiseShape;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {getNoiseShape} from './dropout_util';\nimport {floor} from './floor';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {randomUniform} from './random_uniform';\n\n/**\n * Computes dropout.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 2, 1]);\n * const rate = 0.75;\n * const output = tf.dropout(x, rate);\n * output.print();\n * ```\n *\n * @param x A floating point Tensor or TensorLike.\n * @param rate A float in the range [0, 1). The probability that each element\n *   of x is discarded.\n * @param noiseShape An array of numbers of type int32, representing the\n * shape for randomly generated keep/drop flags. If the noiseShape has null\n * value, it will be automatically replaced with the x's relative dimension\n * size. Optional.\n * @param seed Used to create random seeds. Optional.\n * @returns A Tensor of the same shape of x.\n *\n * @doc {heading: 'Operations', subheading: 'Dropout'}\n */\nfunction dropout_(\n    x: Tensor|TensorLike, rate: number, noiseShape?: number[],\n    seed?: number|string): Tensor {\n  const $x = convertToTensor(x, 'x', 'dropout');\n\n  util.assert(\n      $x.dtype === 'float32',\n      () => `x has to be a floating point tensor since it's going to be ` +\n          `scaled, but got a ${$x.dtype} tensor instead.`);\n  util.assert(\n      rate >= 0 && rate < 1,\n      () => `rate must be a float in the range [0, 1), but got ${rate}.`);\n\n  if (rate === 0) {\n    return x instanceof Tensor ? $x.clone() : $x;\n  }\n\n  const $noiseShape = getNoiseShape($x, noiseShape);\n  const keepProb = 1 - rate;\n  const multiplier = div(\n      floor(add(randomUniform($noiseShape, 0, 1, 'float32', seed), keepProb)),\n      keepProb);\n\n  return mul($x, multiplier);\n}\n\nexport const dropout = op({dropout_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {tensor1d} from './tensor1d';\n\nexport function enclosingPowerOfTwo(value: number) {\n  // Return 2**N for integer N such that 2**N >= value.\n  return Math.floor(Math.pow(2, Math.ceil(Math.log(value) / Math.log(2.0))));\n}\n\nexport function cosineWindow(\n    windowLength: number, a: number, b: number): Tensor1D {\n  const even = 1 - windowLength % 2;\n  const newValues = new Float32Array(windowLength);\n  for (let i = 0; i < windowLength; ++i) {\n    const cosArg = (2.0 * Math.PI * i) / (windowLength + even - 1);\n    newValues[i] = a - b * Math.cos(cosArg);\n  }\n  return tensor1d(newValues, 'float32');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch, getTypedArrayFromDType} from '../util';\nimport {tensor} from './tensor';\n\n/**\n * Returns whether the targets are in the top K predictions.\n *\n * ```js\n * const predictions = tf.tensor2d([[20, 10, 40, 30], [30, 50, -20, 10]]);\n * const targets = tf.tensor1d([2, 0]);\n * const precision = await tf.inTopKAsync(predictions, targets);\n * precision.print();\n * ```\n * @param predictions 2-D or higher `tf.Tensor` with last dimension being\n *     at least `k`.\n * @param targets 1-D or higher `tf.Tensor`.\n * @param k Optional Number of top elements to look at for computing precision,\n *     default to 1.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nasync function inTopKAsync_<T extends Tensor, U extends Tensor>(\n    predictions: T|TensorLike, targets: U|TensorLike, k = 1): Promise<U> {\n  const $predictions = convertToTensor(predictions, 'predictions', 'inTopK');\n  const $targets = convertToTensor(targets, 'targets', 'inTopK');\n\n  assert(\n      $predictions.rank > 1,\n      () => 'inTopK() expects the predictions to be of rank 2 or higher, ' +\n          `but got ${$predictions.rank}`);\n  assert(\n      $predictions.rank - 1 === $targets.rank,\n      () => `predictions rank should be 1 larger than ` +\n          `targets rank, but got predictions rank ` +\n          `${$predictions.rank} and targets rank ${$targets.rank}`);\n  assertShapesMatch(\n      $predictions.shape.slice(0, $predictions.shape.length - 1),\n      $targets.shape,\n      `predictions's shape should be align with the targets' shape, ` +\n          'except the last dimension.');\n  const lastDim = $predictions.shape[$predictions.shape.length - 1];\n  assert(\n      k > 0 && k <= lastDim,\n      () => `'k' passed to inTopK() must be > 0 && <= the predictions last ` +\n          `dimension (${lastDim}), but got ${k}`);\n\n  const predictionsVals = await $predictions.data();\n  const targetsVals = await $targets.data();\n\n  // Reshape predictionsVals into a 2d tensor [batch, lastDim]\n  // and look up topK along lastDim.\n  const [batch, size] = [predictionsVals.length / lastDim, lastDim];\n  const precision = getTypedArrayFromDType('bool', batch);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = predictionsVals.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    precision[b] = 0;\n    for (let i = 0; i < k; i++) {\n      if (valAndInd[i].index === targetsVals[b]) {\n        precision[b] = 1;\n        break;\n      }\n    }\n  }\n\n  if (predictions !== $predictions) {\n    $predictions.dispose();\n  }\n  if (targets !== $targets) {\n    $targets.dispose();\n  }\n\n  // Output precision has the same shape as targets.\n  return tensor(precision, $targets.shape, 'bool') as U;\n}\n\nexport const inTopKAsync = inTopKAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropFilter, Conv2DBackpropFilterAttrs, Conv2DBackpropFilterInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 2D convolution.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     [batch, height, width, inChannels]. If rank 3, batch of 1 is assumed.\n * @param dy The dy image, of rank 4 or rank 3, of shape\n *     [batch, height, width, outDepth]. If rank 3, batch of 1 is assumed.\n * @param filterShape The shape of the filter, length 4,\n *     [filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction conv2DBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2dDerFilter: input must be rank 4, but got shape ` +\n          `${x4D.shape}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerFilter: dy must be rank 4, but got shape ` +\n          `${dy4D.shape}.`);\n  util.assert(\n      filterShape.length === 4,\n      () => `Error in conv2dDerFilter: filterShape must be length 4, but got ` +\n          `${filterShape}.`);\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filterShape[2],\n      () => `Error in conv2dDerFilter: depth of input ${inDepth}) must ` +\n          `match input depth in filter (${filterShape[2]}.`);\n  util.assert(\n      outDepth === filterShape[3],\n      () => `Error in conv2dDerFilter: depth of dy (${outDepth}) must ` +\n          `match output depth for filter (${filterShape[3]}).`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerFilter: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, filterShape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    return backend.conv2dDerFilter(x4D, dy4D, convInfo);\n  };\n\n  const inputs: Conv2DBackpropFilterInputs = {x: x4D, dy: dy4D};\n  const attrs:\n      Conv2DBackpropFilterAttrs = {strides, pad, dataFormat, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv2DBackpropFilter, attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const conv2DBackpropFilter = op({conv2DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\nimport * as broadcast_util from './broadcast_util';\nimport {elu} from './elu';\nimport {Activation} from './fused_types';\nimport {mul} from './mul';\nimport {prelu} from './prelu';\nimport {relu} from './relu';\nimport {relu6} from './relu6';\nimport {reshape} from './reshape';\nimport {step} from './step';\nimport {sum} from './sum';\n\n// Returns gradient for fused activation.\nexport function getFusedDyActivation(\n    dy: Tensor, y: Tensor, activation: Activation): Tensor {\n  if (activation == null || activation === 'linear') {\n    return dy;\n  }\n  if (activation === 'relu') {\n    return mul(dy, step(y));\n  }\n  throw new Error(\n      `Cannot compute gradient for fused activation ${activation}.`);\n}\n\n// Returns gradient for fused bias.\nexport function getFusedBiasGradient(\n    bias: Tensor, dyActivation: Tensor): Tensor {\n  let res = dyActivation;\n  const reduceAxes =\n      broadcast_util.getReductionAxes(bias.shape, dyActivation.shape);\n  if (reduceAxes.length > 0) {\n    res = sum(res, reduceAxes);\n  }\n  return reshape(res, bias.shape);\n}\n\nexport function applyActivation(\n    x: Tensor, activation: Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return x;\n  } else if (activation === 'relu') {\n    return relu(x);\n  } else if (activation === 'elu') {\n    return elu(x);\n  } else if (activation === 'relu6') {\n    return relu6(x);\n  } else if (activation === 'prelu') {\n    return prelu(x, preluActivationWeights);\n  }\n  throw new Error(`Unknown fused activation ${activation}.`);\n}\n\n// Whether we should call fused ops.\nexport const shouldFuse = (gradientDepth: number, activation: Activation) => {\n  const gradientMode = gradientDepth > 0;\n  return !gradientMode || activation === 'linear';\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedConv2D, FusedConv2DAttrs, FusedConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {conv2d as unfusedConv2d} from '../conv2d';\nimport {conv2DBackpropFilter} from '../conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../conv2d_backprop_input';\nimport * as conv_util from '../conv_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes a 2D convolution over the input x, optionally fused with adding a\n * bias and applying an activation.\n *\n * ```js\n * const inputDepth = 2;\n * const inShape = [2, 2, 2, inputDepth];\n * const outputDepth = 2;\n * const fSize = 1;\n * const pad = 0;\n * const strides = 1;\n *\n * const x = tf.tensor4d( [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,\n * 16], inShape);\n * const w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth,\n * outputDepth]);\n *\n * tf.fused.conv2d({ x, filter: w, strides, pad, dataFormat: 'NHWC',\n * dilations: [1, 1], bias: tf.scalar(5), activation: 'relu' }).print();\n * ```\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid` output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`) to be\n *     applied\n *      after biasAdd.\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  activation = activation || 'linear';\n\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused conv2d: input must be rank 4, but got rank ` +\n          `${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${x4D.shape[3]}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in conv2d: got dataFormat of ${\n          dataFormat} but only NHWC is currently supported.`);\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused conv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    const [$filter, x4D, y, $bias] =\n        saved as [Tensor4D, Tensor4D, Tensor4D, Tensor];\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused conv2D: ' +\n            `dilation rates greater than 1 ` +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const xDer =\n        conv2DBackpropInput(x4D.shape, dyActivation, $filter, strides, pad);\n    const filterDer =\n        conv2DBackpropFilter(x4D, dyActivation, $filter.shape, strides, pad);\n    const der: Tensor[] = [xDer, filterDer];\n\n    if ($bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      der.push(biasDer);\n    }\n    return der;\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedConv2d({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n\n  const attrs: FusedConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const conv2d = op({fusedConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropFilterInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    convInfo: conv_util.Conv2DInfo): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerFilter(x4D, dy4D, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropFilterInputs = {x: x4D, dy: dy4D};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             DepthwiseConv2dNativeBackpropFilter) as Tensor4D;\n}\n\nexport const depthwiseConv2dNativeBackpropFilter =\n    op({depthwiseConv2dNativeBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropInput, DepthwiseConv2dNativeBackpropInputInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, convInfo: conv_util.Conv2DInfo): T {\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerInput(dy4D, filter, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropInputInputs = {dy: dy4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null,\n      DepthwiseConv2dNativeBackpropInput);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2dNativeBackpropInput =\n    op({depthwiseConv2dNativeBackpropInput_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedDepthwiseConv2D, FusedDepthwiseConv2DAttrs, FusedDepthwiseConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport * as conv_util from '../conv_util';\nimport {depthwiseConv2d as unfusedDepthwiseConv2d} from '../depthwise_conv2d';\nimport {depthwiseConv2dNativeBackpropFilter} from '../depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../depthwise_conv2d_native_backprop_input';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes depthwise 2D convolution, optionally fused with adding a\n * bias and applying an activation.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`).\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedDepthwiseConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedDepthwiseConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused depthwiseConv2d: filter must be rank 4, ` +\n          `but got rank ${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in fused depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () =>\n          'Error in fused depthwiseConv2d: Either strides or dilations must ' +\n          `be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused depthwiseConv2d: pad must be an integer when ` +\n            `using dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n      true /* depthwise */);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused depthwiseConv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused depthwiseConv2d: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${dilations}'`);\n    const [$filter, x4D, y, bias] = saved;\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    const xDer = depthwiseConv2dNativeBackpropInput(\n        (x4D as Tensor4D).shape, dyActivation, $filter as Tensor4D, convInfo);\n    const filterDer = depthwiseConv2dNativeBackpropFilter(\n        x4D as Tensor4D, dyActivation, ($filter as Tensor4D).shape, convInfo);\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [xDer, filterDer, biasDer];\n    }\n    return [xDer, filterDer];\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedDepthwiseConv2D({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedDepthwiseConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: FusedDepthwiseConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const depthwiseConv2d = op({fusedDepthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {_FusedMatMul, _FusedMatMulAttrs, _FusedMatMulInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {matMul as unfusedMatMul} from '../mat_mul';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes the dot product of two matrices with optional activation and bias.\n *\n * ```js\n * const a = tf.tensor2d([-1, -2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const bias = tf.tensor2d([1, 2], [1, 2]);\n *\n * tf.fused.matMul({a, b, bias, activation: 'relu'}).print();\n * ```\n *\n * @param obj An object with the following properties:\n * - `a` First matrix in dot product operation.\n * - `b` Second matrix in dot product operation.\n * - `transposeA` If true, `a` is transposed before multiplication.\n * - `transposeB` If true, `b` is transposed before multiplication.\n * - `bias` Matrix to be added to the result.\n * - `activation` Name of activation kernel (defaults to `linear`).\n * - `preluActivationWeights` Tensor of prelu weights.\n */\nfunction fusedMatMul_<T extends Tensor>({\n  a,\n  b,\n  transposeA = false,\n  transposeB = false,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  a: T|TensorLike,\n  b: T|TensorLike,\n  transposeA?: boolean,\n  transposeB?: boolean,\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedMatMul(a, b, transposeA, transposeB);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  let $a = convertToTensor(a, 'a', 'fused matMul');\n  let $b = convertToTensor(b, 'b', 'fused matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () =>\n          `Error in fused matMul: inputs must have the same rank of at least ` +\n          `2, got ranks ${$a.rank} and ${$b.rank}.`);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in fused matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in fused matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D: Tensor3D = transposeA ?\n      reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n      reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D: Tensor3D = transposeB ?\n      reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n      reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused matMul');\n    [$bias] = makeTypesMatch($bias, $a);\n\n    broadcast_util.assertAndGetBroadcastShape(outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused matMul');\n  }\n\n  const grad = (dy: Tensor3D, saved: Tensor[]) => {\n    const [a3D, b3D, y, $bias] = saved;\n    // we reshape dy because the result of the forward is not\n    // necessarily going to be a 3d tensor due to a reshape done at the end of\n    // the customOp.\n    const dyActivation =\n        getFusedDyActivation(reshape(dy, y.shape), y, activation);\n    let aDer: Tensor;\n    let bDer: Tensor;\n\n    if (!transposeA && !transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, true, false);\n    } else if (!transposeA && transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, false);\n      bDer = unfusedMatMul(dyActivation, a3D, true, false);\n    } else if (transposeA && !transposeB) {\n      aDer = unfusedMatMul(b3D, dyActivation, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, false, false);\n    } else {\n      aDer = unfusedMatMul(b3D, dyActivation, true, true);\n      bDer = unfusedMatMul(dyActivation, a3D, true, true);\n    }\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [aDer, bDer, biasDer];\n    } else {\n      return [aDer, bDer];\n    }\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const y = backend.fusedBatchMatMul({\n      a: a3D,\n      b: b3D,\n      transposeA,\n      transposeB,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return y;\n  };\n\n  const inputs: _FusedMatMulInputs = {\n    a: a3D,\n    b: b3D,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: _FusedMatMulAttrs = {transposeA, transposeB, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((a3D: Tensor3D, b3D: Tensor3D, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n    return customOp(a3D, b3D) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (a3D: Tensor3D, b3D: Tensor3D, $bias: Tensor, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res, $bias]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n\n    return customOpWithBias(a3D, b3D, $bias) as T;\n  }\n}\n\nexport const matMul = op({fusedMatMul_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {conv2d} from './fused/conv2d';\nimport {depthwiseConv2d} from './fused/depthwise_conv2d';\nimport {matMul} from './fused/mat_mul';\nimport {Activation} from './fused_types';\n\nexport {Activation, conv2d, depthwiseConv2d, matMul};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a hamming window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hammingWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hammingWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.54, 0.46);\n}\nexport const hammingWindow = op({hammingWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a Hann window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hannWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hannWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.5, 0.5);\n}\n\nexport const hannWindow = op({hannWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {fill} from '../fill';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {tensor2d} from '../tensor2d';\n\n/**\n * Expands input into frames of frameLength.\n * Slides a window size with frameStep.\n *\n * ```js\n * tf.signal.frame([1, 2, 3], 2, 1).print();\n * ```\n * @param signal The input tensor to be expanded\n * @param frameLength Length of each frame\n * @param frameStep The frame hop size in samples.\n * @param padEnd Whether to pad the end of signal with padValue.\n * @param padValue An number to use where the input signal does\n *     not exist when padEnd is True.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction frame_(\n    signal: Tensor1D, frameLength: number, frameStep: number, padEnd = false,\n    padValue = 0): Tensor {\n  let start = 0;\n  const output: Tensor[] = [];\n  while (start + frameLength <= signal.size) {\n    output.push(slice(signal, start, frameLength));\n    start += frameStep;\n  }\n\n  if (padEnd) {\n    while (start < signal.size) {\n      const padLen = (start + frameLength) - signal.size;\n      const pad = concat([\n        slice(signal, start, frameLength - padLen), fill([padLen], padValue)\n      ]);\n      output.push(pad);\n      start += frameStep;\n    }\n  }\n\n  if (output.length === 0) {\n    return tensor2d([], [0, frameLength]);\n  }\n\n  return reshape(concat(output), [output.length, frameLength]);\n}\nexport const frame = op({frame_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {enclosingPowerOfTwo} from '../signal_ops_util';\nimport {slice} from '../slice';\nimport {rfft} from '../spectral/rfft';\n\nimport {frame} from './frame';\nimport {hannWindow} from './hann_window';\n\n/**\n * Computes the Short-time Fourier Transform of signals\n * See: https://en.wikipedia.org/wiki/Short-time_Fourier_transform\n *\n * ```js\n * const input = tf.tensor1d([1, 1, 1, 1, 1])\n * tf.signal.stft(input, 3, 1).print();\n * ```\n * @param signal 1-dimensional real value tensor.\n * @param frameLength The window length of samples.\n * @param frameStep The number of samples to step.\n * @param fftLength The size of the FFT to apply.\n * @param windowFn A callable that takes a window length and returns 1-d tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction stft_(\n    signal: Tensor1D, frameLength: number, frameStep: number,\n    fftLength?: number,\n    windowFn: (length: number) => Tensor1D = hannWindow): Tensor {\n  if (fftLength == null) {\n    fftLength = enclosingPowerOfTwo(frameLength);\n  }\n  const framedSignal = frame(signal, frameLength, frameStep);\n  const windowedSignal = mul(framedSignal, windowFn(frameLength));\n  const output: Tensor[] = [];\n  for (let i = 0; i < framedSignal.shape[0]; i++) {\n    output.push(\n        rfft(slice(windowedSignal, [i, 0], [1, frameLength]), fftLength));\n  }\n  return concat(output);\n}\nexport const stft = op({stft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {CropAndResize, CropAndResizeAttrs, CropAndResizeInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor1D, Tensor2D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Extracts crops from the input image tensor and resizes them using bilinear\n * sampling or nearest neighbor sampling (possibly with aspect ratio change)\n * to a common output size specified by crop_size.\n *\n * @param image 4d tensor of shape `[batch,imageHeight,imageWidth, depth]`,\n *     where imageHeight and imageWidth must be positive, specifying the\n *     batch of images from which to take crops\n * @param boxes 2d float32 tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the normalized\n *     coordinates of the box in the boxInd[i]'th image in the batch\n * @param boxInd 1d int32 tensor of shape `[numBoxes]` with values in range\n *     `[0, batch)` that specifies the image that the `i`-th box refers to.\n * @param cropSize 1d int32 tensor of 2 elements `[cropHeigh, cropWidth]`\n *     specifying the size to which all crops are resized to.\n * @param method Optional string from `'bilinear' | 'nearest'`,\n *     defaults to bilinear, which specifies the sampling method for resizing\n * @param extrapolationValue A threshold for deciding when to remove boxes based\n *     on score. Defaults to 0.\n * @return A 4D tensor of the shape `[numBoxes,cropHeight,cropWidth,depth]`\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction cropAndResize_(\n    image: Tensor4D|TensorLike,\n    boxes: Tensor2D|TensorLike,\n    boxInd: Tensor1D|TensorLike,\n    cropSize: [number, number],\n    method?: 'bilinear'|'nearest',\n    extrapolationValue?: number,\n    ): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'cropAndResize');\n  const $boxes = convertToTensor(boxes, 'boxes', 'cropAndResize', 'float32');\n  const $boxInd = convertToTensor(boxInd, 'boxInd', 'cropAndResize', 'int32');\n  method = method || 'bilinear';\n  extrapolationValue = extrapolationValue || 0;\n\n  const numBoxes = $boxes.shape[0];\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in cropAndResize: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n  util.assert(\n      $boxes.rank === 2 && $boxes.shape[1] === 4,\n      () => `Error in cropAndResize: boxes must be have size [${numBoxes},4] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      $boxInd.rank === 1 && $boxInd.shape[0] === numBoxes,\n      () => `Error in cropAndResize: boxInd must be have size [${numBoxes}] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      cropSize.length === 2,\n      () => `Error in cropAndResize: cropSize must be of length 2, but got ` +\n          `length ${cropSize.length}.`);\n  util.assert(\n      cropSize[0] >= 1 && cropSize[1] >= 1,\n      () => `cropSize must be atleast [1,1], but was ${cropSize}`);\n  util.assert(\n      method === 'bilinear' || method === 'nearest',\n      () => `method must be bilinear or nearest, but was ${method}`);\n\n  const forward: ForwardFunc<Tensor4D> = (backend) => backend.cropAndResize(\n      $image, $boxes, $boxInd, cropSize, method, extrapolationValue);\n\n  const inputs:\n      CropAndResizeInputs = {image: $image, boxes: $boxes, boxInd: $boxInd};\n  const attrs: CropAndResizeAttrs = {method, extrapolationValue, cropSize};\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, CropAndResize,\n      attrs as {} as NamedAttrMap);\n  return res;\n}\n\nexport const cropAndResize = op({cropAndResize_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FlipLeftRight, FlipLeftRightInputs} from '../../kernel_names';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {op} from '../operation';\n\n/**\n * Flips the image left to right. Currently available in the CPU, WebGL, and\n * WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n */\n/** @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'} */\nfunction flipLeftRight_(image: Tensor4D|TensorLike): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'flipLeftRight', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in flipLeftRight: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: FlipLeftRightInputs = {image: $image};\n  const res =\n      ENGINE.runKernel(FlipLeftRight, inputs as {} as NamedTensorMap, {});\n  return res as Tensor4D;\n}\n\nexport const flipLeftRight = op({flipLeftRight_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Rotates the input image tensor counter-clockwise with an optional offset\n * center of rotation. Currently available in the CPU, WebGL, and WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n * @param radians The amount of rotation.\n * @param fillValue The value to fill in the empty space leftover\n *     after rotation. Can be either a single grayscale value (0-255), or an\n *     array of three numbers `[red, green, blue]` specifying the red, green,\n *     and blue channels. Defaults to `0` (black).\n * @param center The center of rotation. Can be either a single value (0-1), or\n *     an array of two numbers `[centerX, centerY]`. Defaults to `0.5` (rotates\n *     the image around its center).\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction rotateWithOffset_(\n    image: Tensor4D|TensorLike, radians: number,\n    fillValue: number|[number, number, number] = 0,\n    center: number|[number, number] = 0.5): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'rotateWithOffset', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in rotateWithOffset: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: RotateWithOffsetInputs = {image: $image};\n  const attrs: RotateWithOffsetAttrs = {radians, fillValue, center};\n  const res = ENGINE.runKernel(\n      RotateWithOffset, inputs as {} as NamedTensorMap,\n      attrs as {} as NamedAttrMap);\n  return res as Tensor4D;\n}\n\nexport const rotateWithOffset = op({rotateWithOffset_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport * as util from '../util';\n\nfunction nonMaxSuppSanityCheck(\n    boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma?: number): {\n  maxOutputSize: number,\n  iouThreshold: number,\n  scoreThreshold: number,\n  softNmsSigma: number\n} {\n  if (iouThreshold == null) {\n    iouThreshold = 0.5;\n  }\n  if (scoreThreshold == null) {\n    scoreThreshold = Number.NEGATIVE_INFINITY;\n  }\n  if (softNmsSigma == null) {\n    softNmsSigma = 0.0;\n  }\n\n  const numBoxes = boxes.shape[0];\n  maxOutputSize = Math.min(maxOutputSize, numBoxes);\n\n  util.assert(\n      0 <= iouThreshold && iouThreshold <= 1,\n      () => `iouThreshold must be in [0, 1], but was '${iouThreshold}'`);\n  util.assert(\n      boxes.rank === 2,\n      () => `boxes must be a 2D tensor, but was of rank '${boxes.rank}'`);\n  util.assert(\n      boxes.shape[1] === 4,\n      () =>\n          `boxes must have 4 columns, but 2nd dimension was ${boxes.shape[1]}`);\n  util.assert(scores.rank === 1, () => 'scores must be a 1D tensor');\n  util.assert(\n      scores.shape[0] === numBoxes,\n      () => `scores has incompatible shape with boxes. Expected ${numBoxes}, ` +\n          `but was ${scores.shape[0]}`);\n  util.assert(\n      0 <= softNmsSigma && softNmsSigma <= 1,\n      () => `softNmsSigma must be in [0, 1], but was '${softNmsSigma}'`);\n  return {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n}\n\nexport {nonMaxSuppSanityCheck};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV3} from '../../kernel_names';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\nfunction nonMaxSuppression_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Tensor1D {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const attrs = {maxOutputSize, iouThreshold, scoreThreshold};\n  return ENGINE.runKernelFunc(\n      b => b.nonMaxSuppression(\n          $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold),\n      {boxes: $boxes, scores: $scores}, null /* grad */, NonMaxSuppressionV3,\n      attrs);\n}\n\nexport const nonMaxSuppression = op({nonMaxSuppression_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inserts a value into a sorted array. This method allows duplicate, meaning it\n * allows inserting duplicate value, in which case, the element will be inserted\n * at the lowest index of the value.\n * @param arr The array to modify.\n * @param element The element to insert.\n * @param comparator Optional. If no comparator is specified, elements are\n * compared using array_util.defaultComparator, which is suitable for Strings\n * and Numbers in ascending arrays. If the array contains multiple instances of\n * the target value, the left-most instance will be returned. To provide a\n * comparator, it should take 2 arguments to compare and return a negative,\n * zero, or a positive number.\n */\nexport function binaryInsert<T>(\n    arr: T[], element: T, comparator?: (a: T, b: T) => number) {\n  const index = binarySearch(arr, element, comparator);\n  const insertionPoint = index < 0 ? -(index + 1) : index;\n  arr.splice(insertionPoint, 0, element);\n}\n\n/**\n * Searches the array for the target using binary search, returns the index\n * of the found element, or position to insert if element not found. If no\n * comparator is specified, elements are compared using array_\n * util.defaultComparator, which is suitable for Strings and Numbers in\n * ascending arrays. If the array contains multiple instances of the target\n * value, the left-most instance will be returned.\n * @param arr The array to be searched in.\n * @param target The target to be searched for.\n * @param comparator Should take 2 arguments to compare and return a negative,\n *    zero, or a positive number.\n * @return Lowest index of the target value if found, otherwise the insertion\n *    point where the target should be inserted, in the form of\n *    (-insertionPoint - 1).\n */\nexport function binarySearch<T>(\n    arr: T[], target: T, comparator?: (a: T, b: T) => number) {\n  return binarySearch_(arr, target, comparator || defaultComparator);\n}\n\n/**\n * Compares its two arguments for order.\n * @param a The first element to be compared.\n * @param b The second element to be compared.\n * @return A negative number, zero, or a positive number as the first\n *     argument is less than, equal to, or greater than the second.\n */\nfunction defaultComparator<T>(a: T, b: T): number {\n  return a > b ? 1 : a < b ? -1 : 0;\n}\n\nfunction binarySearch_<T>(\n    arr: T[], target: T, comparator: (a: T, b: T) => number) {\n  let left = 0;\n  let right = arr.length;\n  let middle = 0;\n  let found = false;\n  while (left < right) {\n    middle = left + ((right - left) >>> 1);\n    const compareResult = comparator(target, arr[middle]);\n    if (compareResult > 0) {\n      left = middle + 1;\n    } else {\n      right = middle;\n      // If compareResult is 0, the value is found. We record it is found,\n      // and then keep looking because there may be duplicate.\n      found = !compareResult;\n    }\n  }\n\n  return found ? left : -left - 1;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Implementation of the NonMaxSuppression kernel shared between webgl and cpu.\n */\n\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\n\nimport {binaryInsert} from './array_util';\n\ninterface Candidate {\n  score: number;\n  boxIndex: number;\n  suppressBeginIndex: number;\n}\n\nexport function nonMaxSuppressionV3Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number): Tensor1D {\n  return nonMaxSuppressionImpl_(\n             boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n             0 /* softNmsSigma */)\n             .selectedIndices as Tensor1D;\n}\n\nexport function nonMaxSuppressionV4Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    padToMaxOutputSize: boolean): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n      0 /* softNmsSigma */, false /* returnScoresTensor */,\n      padToMaxOutputSize /* padToMaxOutputSize */, true\n      /* returnValidOutputs */);\n}\n\nexport function nonMaxSuppressionV5Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    softNmsSigma: number): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma,\n      true /* returnScoresTensor */);\n}\n\nfunction nonMaxSuppressionImpl_(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma: number,\n    returnScoresTensor = false, padToMaxOutputSize = false,\n    returnValidOutputs = false): NamedTensorMap {\n  // The list is sorted in ascending order, so that we can always pop the\n  // candidate with the largest score in O(1) time.\n  const candidates = [];\n\n  for (let i = 0; i < scores.length; i++) {\n    if (scores[i] > scoreThreshold) {\n      candidates.push({score: scores[i], boxIndex: i, suppressBeginIndex: 0});\n    }\n  }\n\n  candidates.sort(ascendingComparator);\n\n  // If softNmsSigma is 0, the outcome of this algorithm is exactly same as\n  // before.\n  const scale = softNmsSigma > 0 ? (-0.5 / softNmsSigma) : 0.0;\n\n  const selectedIndices: number[] = [];\n  const selectedScores: number[] = [];\n\n  while (selectedIndices.length < maxOutputSize && candidates.length > 0) {\n    const candidate = candidates.pop();\n    const {score: originalScore, boxIndex, suppressBeginIndex} = candidate;\n\n    if (originalScore < scoreThreshold) {\n      break;\n    }\n\n    // Overlapping boxes are likely to have similar scores, therefore we\n    // iterate through the previously selected boxes backwards in order to\n    // see if candidate's score should be suppressed. We use\n    // suppressBeginIndex to track and ensure a candidate can be suppressed\n    // by a selected box no more than once. Also, if the overlap exceeds\n    // iouThreshold, we simply ignore the candidate.\n    let ignoreCandidate = false;\n    for (let j = selectedIndices.length - 1; j >= suppressBeginIndex; --j) {\n      const iou = intersectionOverUnion(boxes, boxIndex, selectedIndices[j]);\n\n      if (iou >= iouThreshold) {\n        ignoreCandidate = true;\n        break;\n      }\n\n      candidate.score =\n          candidate.score * suppressWeight(iouThreshold, scale, iou);\n\n      if (candidate.score <= scoreThreshold) {\n        break;\n      }\n    }\n\n    // At this point, if `candidate.score` has not dropped below\n    // `scoreThreshold`, then we know that we went through all of the\n    // previous selections and can safely update `suppressBeginIndex` to the\n    // end of the selected array. Then we can re-insert the candidate with\n    // the updated score and suppressBeginIndex back in the candidate list.\n    // If on the other hand, `candidate.score` has dropped below the score\n    // threshold, we will not add it back to the candidates list.\n    candidate.suppressBeginIndex = selectedIndices.length;\n\n    if (!ignoreCandidate) {\n      // Candidate has passed all the tests, and is not suppressed, so\n      // select the candidate.\n      if (candidate.score === originalScore) {\n        selectedIndices.push(boxIndex);\n        selectedScores.push(candidate.score);\n      } else if (candidate.score > scoreThreshold) {\n        // Candidate's score is suppressed but is still high enough to be\n        // considered, so add back to the candidates list.\n        binaryInsert(candidates, candidate, ascendingComparator);\n      }\n    }\n  }\n\n  // NonMaxSuppressionV4 feature: padding output to maxOutputSize.\n  const validOutputs = selectedIndices.length;\n  const elemsToPad = maxOutputSize - validOutputs;\n\n  if (padToMaxOutputSize && elemsToPad > 0) {\n    selectedIndices.push(...new Array(elemsToPad).fill(0));\n    selectedScores.push(...new Array(elemsToPad).fill(0.0));\n  }\n\n  const result:\n      NamedTensorMap = {selectedIndices: tensor1d(selectedIndices, 'int32')};\n\n  if (returnScoresTensor) {\n    result['selectedScores'] = tensor1d(selectedScores, 'float32');\n  }\n\n  if (returnValidOutputs) {\n    result['validOutputs'] = scalar(validOutputs, 'int32');\n  }\n\n  return result;\n}\n\nfunction intersectionOverUnion(boxes: TypedArray, i: number, j: number) {\n  const iCoord = boxes.subarray(i * 4, i * 4 + 4);\n  const jCoord = boxes.subarray(j * 4, j * 4 + 4);\n  const yminI = Math.min(iCoord[0], iCoord[2]);\n  const xminI = Math.min(iCoord[1], iCoord[3]);\n  const ymaxI = Math.max(iCoord[0], iCoord[2]);\n  const xmaxI = Math.max(iCoord[1], iCoord[3]);\n  const yminJ = Math.min(jCoord[0], jCoord[2]);\n  const xminJ = Math.min(jCoord[1], jCoord[3]);\n  const ymaxJ = Math.max(jCoord[0], jCoord[2]);\n  const xmaxJ = Math.max(jCoord[1], jCoord[3]);\n  const areaI = (ymaxI - yminI) * (xmaxI - xminI);\n  const areaJ = (ymaxJ - yminJ) * (xmaxJ - xminJ);\n  if (areaI <= 0 || areaJ <= 0) {\n    return 0.0;\n  }\n  const intersectionYmin = Math.max(yminI, yminJ);\n  const intersectionXmin = Math.max(xminI, xminJ);\n  const intersectionYmax = Math.min(ymaxI, ymaxJ);\n  const intersectionXmax = Math.min(xmaxI, xmaxJ);\n  const intersectionArea = Math.max(intersectionYmax - intersectionYmin, 0.0) *\n      Math.max(intersectionXmax - intersectionXmin, 0.0);\n  return intersectionArea / (areaI + areaJ - intersectionArea);\n}\n\n// A Gaussian penalty function, this method always returns values in [0, 1].\n// The weight is a function of similarity, the more overlap two boxes are, the\n// smaller the weight is, meaning highly overlapping boxe will be significantly\n// penalized. On the other hand, a non-overlapping box will not be penalized.\nfunction suppressWeight(iouThreshold: number, scale: number, iou: number) {\n  const weight = Math.exp(scale * iou * iou);\n  return iou <= iouThreshold ? weight : 0.0;\n}\n\nfunction ascendingComparator(c1: Candidate, c2: Candidate) {\n  // For objects with same scores, we make the object with the larger index go\n  // first. In an array that pops from the end, this means that the object with\n  // the smaller index will be popped first. This ensures the same output as\n  // the TensorFlow python version.\n  return (c1.score - c2.score) ||\n      ((c1.score === c2.score) && (c2.boxIndex - c1.boxIndex));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {nonMaxSuppressionV3Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This is the async version of `nonMaxSuppression`\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @return A 1D tensor with the selected box indices.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Promise<Tensor1D> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV3Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionAsync = nonMaxSuppressionAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionWithScore_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const inputs: NonMaxSuppressionV5Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV5Attrs =\n      {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV5, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], selectedScores: result[1]};\n}\n\nexport const nonMaxSuppressionWithScore = op({nonMaxSuppressionWithScore_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV5Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionWithScoreAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV5Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionWithScoreAsync = nonMaxSuppressionWithScoreAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionPadded_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const inputs: NonMaxSuppressionV4Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV4Attrs = {\n    maxOutputSize: $maxOutputSize,\n    iouThreshold: $iouThreshold,\n    scoreThreshold: $scoreThreshold,\n    padToMaxOutputSize\n  };\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV4, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], validOutputs: result[1]};\n}\n\nexport const nonMaxSuppressionPadded = op({nonMaxSuppressionPadded_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV4Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionPaddedAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const [boxesVals, scoresVals] =\n      await Promise.all([$boxes.data(), $scores.data()]);\n\n  // We call a cpu based impl directly with the typedarray data here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV4Impl(\n      boxesVals, scoresVals, $maxOutputSize, $iouThreshold, $scoreThreshold,\n      padToMaxOutputSize);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionPaddedAsync = nonMaxSuppressionPaddedAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Bilinear resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeBilinear_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeBilinear');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeBilinear: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () => `Error in resizeBilinear: new shape must 2D, but got shape ` +\n          `${size}.`);\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n\n  const [newHeight, newWidth] = size;\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeBilinear(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const inputs: ResizeBilinearInputs = {images: batchImages};\n  const attrs: ResizeBilinearAttrs = {alignCorners, size};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeBilinear, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeBilinear = op({resizeBilinear_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * NearestNeighbor resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeNearestNeighbor_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeNearestNeighbor');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeNearestNeighbor: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () =>\n          `Error in resizeNearestNeighbor: new shape must 2D, but got shape ` +\n          `${size}.`);\n  util.assert(\n      $images.dtype === 'float32' || $images.dtype === 'int32',\n      () => '`images` must have `int32` or `float32` as dtype');\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n  const [newHeight, newWidth] = size;\n\n  const inputs: ResizeNearestNeighborInputs = {images: batchImages};\n  const attrs: ResizeNearestNeighborAttrs = {alignCorners, size};\n\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeNearestNeighbor(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeNearestNeighbor, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeNearestNeighbor = op({resizeNearestNeighbor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assert} from '../../util';\n\nimport {greaterEqual} from '../greater_equal';\nimport {lessEqual} from '../less_equal';\nimport {logicalAnd} from '../logical_and';\nimport {op} from '../operation';\nimport {range} from '../range';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\nimport {zeros} from '../zeros';\n\n/**\n * Copy a tensor setting everything outside a central band in each innermost\n * matrix to zero.\n *\n * The band part is computed as follows: Assume input has `k` dimensions\n * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where\n * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`.\n * The indicator function\n * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))`\n * `&& (num_upper < 0 || (n-m) <= num_upper)`\n *\n * ```js\n * const x = tf.tensor2d([[ 0,  1,  2, 3],\n *                        [-1,  0,  1, 2],\n *                        [-2, -1,  0, 1],\n *                        [-3, -2, -1, 0]]);\n * let y = tf.linalg.bandPart(x, 1, -1);\n * y.print(); // [[ 0,  1,  2, 3],\n *            //  [-1,  0,  1, 2],\n *            //  [ 0, -1,  0, 1],\n *            //  [ 0, 0 , -1, 0]]\n * let z = tf.linalg.bandPart(x, 2, 1);\n * z.print(); // [[ 0,  1,  0, 0],\n *            //  [-1,  0,  1, 0],\n *            //  [-2, -1,  0, 1],\n *            //  [ 0, -2, -1, 0]]\n * ```\n *\n * @param x Rank `k` tensor\n * @param numLower Number of subdiagonals to keep.\n *   If negative, keep entire lower triangle.\n * @param numUpper Number of subdiagonals to keep.\n *   If negative, keep entire upper triangle.\n * @returns Rank `k` tensor of the same shape as input.\n *   The extracted banded tensor.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction bandPart_<T extends Tensor>(\n    a: T|TensorLike, numLower: number, numUpper: number): T {\n  assert(\n      numLower % 1 === 0,\n      () => `bandPart(): numLower must be an integer, got ${numLower}.`);\n  assert(\n      numUpper % 1 === 0,\n      () => `bandPart(): numUpper must be an integer, got ${numUpper}.`);\n\n  const $a = convertToTensor(a, 'a', 'bandPart');\n\n  assert(\n      $a.rank >= 2,\n      () => `bandPart(): Rank must be at least 2, got ${$a.rank}.`);\n\n  const shape = $a.shape;\n  const [M, N] = $a.shape.slice(-2);\n\n  if (!(numLower <= M)) {\n    throw new Error(\n        `bandPart(): numLower (${numLower})` +\n        ` must not be greater than the number of rows (${M}).`);\n  }\n  if (!(numUpper <= N)) {\n    throw new Error(\n        `bandPart(): numUpper (${numUpper})` +\n        ` must not be greater than the number of columns (${N}).`);\n  }\n\n  if (numLower < 0) {\n    numLower = M;\n  }\n  if (numUpper < 0) {\n    numUpper = N;\n  }\n\n  const i = reshape(range(0, M, 1, 'int32'), [-1, 1]);\n  const j = range(0, N, 1, 'int32');\n  const ij = sub(i, j);\n\n  const inBand = logicalAnd(\n      lessEqual(ij, scalar(+numLower, 'int32')),\n      greaterEqual(ij, scalar(-numUpper, 'int32')));\n\n  const zero = zeros([M, N], $a.dtype);\n\n  return reshape(\n             stack(unstack(reshape($a, [-1, M, N]))\n                       .map(mat => where(inBand, mat, zero))),\n             shape) as T;\n}\n\nexport const bandPart = op({bandPart_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {div} from '../div';\nimport {mul} from '../mul';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {split} from '../split';\nimport {squeeze} from '../squeeze';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\n/**\n * Gram-Schmidt orthogonalization.\n *\n * ```js\n * const x = tf.tensor2d([[1, 2], [3, 4]]);\n * let y = tf.linalg.gramSchmidt(x);\n * y.print();\n * console.log('Othogonalized:');\n * y.dot(y.transpose()).print();  // should be nearly the identity matrix.\n * console.log('First row direction maintained:');\n * const data = await y.array();\n * console.log(data[0][1] / data[0][0]);  // should be nearly 2.\n * ```\n *\n * @param xs The vectors to be orthogonalized, in one of the two following\n *   formats:\n *   - An Array of `tf.Tensor1D`.\n *   - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows\n *     of `xs`.\n *   In each case, all the vectors must have the same length and the length\n *   must be greater than or equal to the number of vectors.\n * @returns The orthogonalized and normalized vectors or matrix.\n *   Orthogonalization means that the vectors or the rows of the matrix\n *   are orthogonal (zero inner products). Normalization means that each\n *   vector or each row of the matrix has an L2 norm that equals `1`.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D {\n  let inputIsTensor2D: boolean;\n  if (Array.isArray(xs)) {\n    inputIsTensor2D = false;\n    assert(\n        xs != null && xs.length > 0,\n        () => 'Gram-Schmidt process: input must not be null, undefined, or ' +\n            'empty');\n    const dim = xs[0].shape[0];\n    for (let i = 1; i < xs.length; ++i) {\n      assert(\n          xs[i].shape[0] === dim,\n          () =>\n              'Gram-Schmidt: Non-unique lengths found in the input vectors: ' +\n              `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`);\n    }\n  } else {\n    inputIsTensor2D = true;\n    xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0]));\n  }\n\n  assert(\n      xs.length <= xs[0].shape[0],\n      () => `Gram-Schmidt: Number of vectors (${\n                (xs as Tensor1D[]).length}) exceeds ` +\n          `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`);\n\n  const ys: Tensor1D[] = [];\n  const xs1d = xs;\n  for (let i = 0; i < xs.length; ++i) {\n    ys.push(ENGINE.tidy(() => {\n      let x = xs1d[i];\n      if (i > 0) {\n        for (let j = 0; j < i; ++j) {\n          const proj = mul(sum(mul(ys[j], x)), ys[j]);\n          x = sub(x, proj);\n        }\n      }\n      return div(x, norm(x, 'euclidean'));\n    }));\n  }\n\n  if (inputIsTensor2D) {\n    return stack(ys, 0) as Tensor2D;\n  } else {\n    return ys;\n  }\n}\n\nexport const gramSchmidt = op({gramSchmidt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../../engine';\nimport {dispose} from '../../globals';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {clone} from '../clone';\nimport {concat} from '../concat';\nimport {div} from '../div';\nimport {eye} from '../eye';\nimport {greater} from '../greater';\nimport {matMul} from '../mat_mul';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {tensor2d} from '../tensor2d';\nimport {transpose} from '../transpose';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\n\n/**\n * Compute QR decomposition of m-by-n matrix using Householder transformation.\n *\n * Implementation based on\n *   [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf]\n * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf)\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [3, 4]]);\n * let [q, r] = tf.linalg.qr(a);\n * console.log('Q');\n * q.print();\n * console.log('R');\n * r.print();\n * console.log('Orthogonalized');\n * q.dot(q.transpose()).print()  // should be nearly the identity matrix.\n * console.log('Reconstructed');\n * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]];\n * ```\n *\n * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose\n *   it has the shape `[..., M, N]`.\n * @param fullMatrices An optional boolean parameter. Defaults to `false`.\n *   If `true`, compute full-sized `Q`. If `false` (the default),\n *   compute only the leading N columns of `Q` and `R`.\n * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix,\n *   i.e., its columns all have unit norm and are mutually orthogonal.\n *   If `M >= N`,\n *     If `fullMatrices` is `false` (default),\n *       - `Q` has a shape of `[..., M, N]`,\n *       - `R` has a shape of `[..., N, N]`.\n *     If `fullMatrices` is `true` (default),\n *       - `Q` has a shape of `[..., M, M]`,\n *       - `R` has a shape of `[..., M, N]`.\n *   If `M < N`,\n *     - `Q` has a shape of `[..., M, M]`,\n *     - `R` has a shape of `[..., M, N]`.\n * @throws If the rank of `x` is less than 2.\n *\n * @doc {heading:'Operations',\n *       subheading:'Linear Algebra',\n *       namespace:'linalg'}\n */\nfunction qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] {\n  assert(\n      x.rank >= 2,\n      () => `qr() requires input tensor to have a rank >= 2, but got rank ${\n          x.rank}`);\n\n  if (x.rank === 2) {\n    return qr2d(x as Tensor2D, fullMatrices);\n  } else {\n    // Rank > 2.\n    // TODO(cais): Below we split the input into individual 2D tensors,\n    //   perform QR decomposition on them and then stack the results back\n    //   together. We should explore whether this can be parallelized.\n    const outerDimsProd = x.shape.slice(0, x.shape.length - 2)\n                              .reduce((value, prev) => value * prev);\n    const x2ds = unstack(\n        reshape(\n            x,\n            [\n              outerDimsProd, x.shape[x.shape.length - 2],\n              x.shape[x.shape.length - 1]\n            ]),\n        0);\n    const q2ds: Tensor2D[] = [];\n    const r2ds: Tensor2D[] = [];\n    x2ds.forEach(x2d => {\n      const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices);\n      q2ds.push(q2d);\n      r2ds.push(r2d);\n    });\n    const q = reshape(stack(q2ds, 0), x.shape);\n    const r = reshape(stack(r2ds, 0), x.shape);\n    return [q, r];\n  }\n}\n\nfunction qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] {\n  return ENGINE.tidy(() => {\n    assert(\n        x.shape.length === 2,\n        () => `qr2d() requires a 2D Tensor, but got a ${\n            x.shape.length}D Tensor.`);\n\n    const m = x.shape[0];\n    const n = x.shape[1];\n\n    let q = eye(m);    // Orthogonal transform so far.\n    let r = clone(x);  // Transformed matrix so far.\n\n    const one2D = tensor2d([[1]], [1, 1]);\n    let w: Tensor2D = clone(one2D);\n\n    const iters = m >= n ? n : m;\n    for (let j = 0; j < iters; ++j) {\n      // This tidy within the for-loop ensures we clean up temporary\n      // tensors as soon as they are no longer needed.\n      const rTemp = r;\n      const wTemp = w;\n      const qTemp = q;\n      [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => {\n        // Find H = I - tau * w * w', to put zeros below R(j, j).\n        const rjEnd1 = slice(r, [j, j], [m - j, 1]);\n        const normX = norm(rjEnd1);\n        const rjj = slice(r, [j, j], [1, 1]);\n\n        // The sign() function returns 0 on 0, which causes division by zero.\n        const s = where(greater(rjj, 0), tensor2d([[-1]]), tensor2d([[1]]));\n\n        const u1 = sub(rjj, mul(s, normX));\n        const wPre = div(rjEnd1, u1);\n        if (wPre.shape[0] === 1) {\n          w = clone(one2D);\n        } else {\n          w = concat(\n              [\n                one2D,\n                slice(wPre, [1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as\n                    Tensor2D\n              ],\n              0);\n        }\n        const tau = neg(div(matMul(s, u1), normX)) as Tensor2D;\n\n        // -- R := HR, Q := QH.\n        const rjEndAll = slice(r, [j, 0], [m - j, n]);\n        const tauTimesW: Tensor2D = mul(tau, w);\n        const wT: Tensor2D = transpose(w);\n        if (j === 0) {\n          r = sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n        } else {\n          const rTimesTau: Tensor2D =\n              sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n          r = concat([slice(r, [0, 0], [j, n]), rTimesTau], 0);\n        }\n        const tawTimesWT: Tensor2D = transpose(tauTimesW);\n        const qAllJEnd = slice(q, [0, j], [m, q.shape[1] - j]);\n        if (j === 0) {\n          q = sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n        } else {\n          const qTimesTau: Tensor2D =\n              sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n          q = concat([slice(q, [0, 0], [m, j]), qTimesTau], 1);\n        }\n        return [w, r, q];\n      });\n      dispose([rTemp, wTemp, qTemp]);\n    }\n\n    if (!fullMatrices && m > n) {\n      q = slice(q, [0, 0], [m, n]);\n      r = slice(r, [0, 0], [n, n]);\n    }\n\n    return [q, r];\n  }) as [Tensor2D, Tensor2D];\n}\n\nexport const qr = op({qr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport enum Reduction {\n  NONE,\n  MEAN,\n  SUM,\n  SUM_BY_NONZERO_WEIGHTS\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {Reduction} from '../loss_ops_utils';\nimport {mean} from '../mean';\nimport {mul} from '../mul';\nimport {notEqual} from '../not_equal';\nimport {ones} from '../ones';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sum} from '../sum';\n\n/**\n * Computes the weighted loss between two tensors.\n *\n * @param losses Tensor of shape `[batch_size, d1, ... dN]`.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `losses`, and must be broadcastable to `losses` (i.e., all\n *    dimensions must be either `1`, or the same as the corresponding\n *    `losses` dimension).\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction computeWeightedLoss_<T extends Tensor, O extends Tensor>(\n    losses: T|TensorLike, weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $losses = convertToTensor(losses, 'losses', 'computeWeightedLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'computeWeightedLoss');\n  }\n\n  const weightedLoss = ($weights == null) ? $losses : mul($losses, $weights);\n\n  if (reduction === Reduction.NONE) {\n    return weightedLoss as O;\n  }\n  if (reduction === Reduction.SUM) {\n    return sum(weightedLoss);\n  }\n  if (reduction === Reduction.MEAN) {\n    if ($weights == null) {\n      return mean(weightedLoss);\n    } else {\n      const broadcastFactor = $losses.size / $weights.size;\n      const result = div(sum(weightedLoss), sum($weights));\n      return broadcastFactor > 1 ? div(result, scalar(broadcastFactor)) :\n                                   result as O;\n    }\n  }\n  if (reduction === Reduction.SUM_BY_NONZERO_WEIGHTS) {\n    if ($weights == null) {\n      return div(sum(weightedLoss), scalar($losses.size));\n    } else {\n      const broadcastedWeights = mul($weights, ones($losses.shape));\n\n      const numNonZeros =\n          cast(sum(notEqual(broadcastedWeights, scalar(0))), 'float32');\n      return div(sum(weightedLoss), numNonZeros);\n    }\n  }\n\n  throw Error(`Unknown reduction: ${reduction}`);\n}\nexport const computeWeightedLoss = op({computeWeightedLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the absolute difference loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction absoluteDifference_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'absoluteDifference');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'absoluteDifference');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'absoluteDifference');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in absoluteDifference: ');\n\n  const losses = abs(sub($labels, $predictions));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const absoluteDifference = op({absoluteDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the cosine distance loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param axis The dimension along which the cosine distance is computed.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction cosineDistance_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike, axis: number,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'cosineDistance');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'cosineDistance');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'cosineDistance');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in cosineDistance: ');\n\n  const one = scalar(1);\n  const losses = sub(one, sum(mul($labels, $predictions), axis, true));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const cosineDistance = op({cosineDistance_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the Hinge loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction hingeLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $labels = convertToTensor(labels, 'labels', 'hingeLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'hingeLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'hingeLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in hingeLoss: ');\n\n  const one = scalar(1);\n  // Convert binary labels to (-1, 1)\n  $labels = sub(mul(scalar(2), $labels), one);\n  const losses = relu(sub(one, mul($labels, $predictions)));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const hingeLoss = op({hingeLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {Reduction} from '../loss_ops_utils';\nimport {minimum} from '../minimum';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {square} from '../square';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the huber loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param delta Point where huber loss changes from quadratic to linear.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`.\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction huberLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, delta = 1.0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'huberLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'huberLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'huberLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in huberLoss: ');\n\n  const deltaScalar = scalar(delta);\n  const error = abs(sub($predictions, $labels));\n  const quadratic = minimum(error, deltaScalar);\n  const linear = sub(error, quadratic);\n\n  const losses =\n      add(mul(scalar(0.5), square(quadratic)), mul(deltaScalar, linear));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const huberLoss = op({huberLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {log} from '../log';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the log loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param epsilon A small increment to avoid taking log of zero\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction logLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, epsilon = 1e-7,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'logLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'logLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'logLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in logLoss: ');\n\n  const one = scalar(1);\n  const epsilonScalar = scalar(epsilon);\n\n  const l1 = neg(mul($labels, log(add($predictions, epsilonScalar))));\n  const l2 =\n      mul(sub(one, $labels), log(add(sub(one, $predictions), epsilonScalar)));\n  const losses = sub(l1, l2);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const logLoss = op({logLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {squaredDifference} from '../squared_difference';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the mean squared error between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction meanSquaredError_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'meanSquaredError');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'meanSquaredError');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'meanSquaredError');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in meanSquaredError: ');\n\n  const losses = squaredDifference($labels, $predictions);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const meanSquaredError = op({meanSquaredError_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {exp} from '../exp';\nimport {log1p} from '../log1p';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\nfunction sigmoidCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, logits: T|TensorLike): O {\n  const $labels =\n      convertToTensor(labels, 'labels', 'sigmoidCrossEntropyWithLogits');\n  const $logits =\n      convertToTensor(logits, 'logits', 'sigmoidCrossEntropyWithLogits');\n  assertShapesMatch(\n      $labels.shape, $logits.shape, 'Error in sigmoidCrossEntropyWithLogits: ');\n\n  /**\n   * Implementation Details:\n   *\n   * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n   *     z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n   *   = (1 - z) * x + log(1 + exp(-x))\n   *   = x - x * z + log(1 + exp(-x))\n   *\n   *   For x < 0, to avoid overflow in exp(-x), we reformulate the above\n   *     x - x * z + log(1 + exp(-x))\n   *   = log(exp(x)) - x * z + log(1 + exp(-x))\n   *   = - x * z + log(1 + exp(x))\n   *\n   * Hence, to ensure stability and avoid overflow, the implementation uses\n   * this equivalent formulation:\n   *     max(x, 0) - x * z + log(1 + exp(-abs(x)))\n   */\n  const maxOutput = relu($logits);\n  const outputXTarget = mul($logits, $labels);\n  const sigmoidOutput = log1p(exp(neg(abs($logits))));\n\n  return add(sub(maxOutput, outputXTarget), sigmoidOutput);\n}\n\n/**\n * Computes the sigmoid cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newMulticlassLabels = multiclassLabels * (1 - labelSmoothing)\n *                         + 0.5 * labelSmoothing\n *\n * @param multiClassLabels The ground truth output tensor of shape\n * [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction sigmoidCrossEntropy_<T extends Tensor, O extends Tensor>(\n    multiClassLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $multiClassLabels = convertToTensor(\n      multiClassLabels, 'multiClassLabels', 'sigmoidCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'sigmoidCrossEntropy');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'sigmoidCrossEntropy');\n  }\n  assertShapesMatch(\n      $multiClassLabels.shape, $logits.shape, 'Error in sigmoidCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const half = scalar(0.5);\n\n    $multiClassLabels =\n        add(mul($multiClassLabels, sub(one, labelSmoothingScalar)),\n            mul(half, labelSmoothingScalar));\n  }\n  const losses = sigmoidCrossEntropyWithLogits_($multiClassLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const sigmoidCrossEntropy = op({sigmoidCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../../gradients';\nimport {Tensor} from '../../tensor';\nimport {GradSaveFunc} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {expandShapeToKeepDim} from '../axis_util';\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {exp} from '../exp';\nimport {logSumExp} from '../log_sum_exp';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes softmax cross entropy between logits and labels.\n *\n * Measures the probability error in discrete classification tasks in which\n * the classes are mutually exclusive (each entry is in exactly one class).\n * For example, each CIFAR-10 image is labeled with one and only one label: an\n * image can be a dog or a truck, but not both.\n *\n * `NOTE`: While the classes are mutually exclusive, their probabilities need\n * not be. All that is required is that each row of labels is a valid\n * probability distribution. If they are not, the computation of the gradient\n * will be incorrect.\n *\n * `WARNING`: This op expects unscaled logits, since it performs a softmax on\n * logits internally for efficiency. Do not call this op with the output of\n * softmax, as it will produce incorrect results.\n *\n * logits and labels must have the same shape, e.g. [batch_size, num_classes]\n * and the same dtype.\n * @param labels The labels array.\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n */\nfunction softmaxCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T, logits: T, dim = -1): O {\n  if (dim === -1) {\n    dim = logits.rank - 1;\n  }\n\n  if (dim !== logits.rank - 1) {\n    throw Error(\n        `Softmax cross entropy along a non-last dimension is not yet ` +\n        `supported. Labels / logits was rank ${logits.rank} ` +\n        `and dim was ${dim}`);\n  }\n  // Use a custom gradient for numerical stability.\n  const customOp =\n      customGrad((labels: Tensor, logits: Tensor, save: GradSaveFunc) => {\n        // Reference:\n        //   1. http://cs231n.github.io/linear-classify/#softmax\n        //   2. https://blog.feedly.com/tricks-of-the-trade-logsumexp/\n        const keepDims = true;\n        const lse = logSumExp(logits, [dim], keepDims);\n        const logResult = sub(cast(logits, 'float32'), lse);\n        save([labels, logResult]);\n\n        const costVector = neg(mul(logResult, labels));\n        const value: O = sum(costVector, [dim]);\n\n        const gradFunc = (dy: O, saved: Tensor[]) => {\n          const [labels, logResult] = saved;\n          const dyShape = expandShapeToKeepDim(dy.shape, [dim]);\n          return [\n            mul(reshape(dy, dyShape),\n                sub(cast(labels, 'float32'), exp(logResult))),\n            mul(reshape(dy, dyShape),\n                sub(exp(logResult), cast(labels, 'float32'))),\n          ];\n        };\n        return {value, gradFunc};\n      });\n\n  return customOp(labels, logits);\n}\n\n/**\n * Computes the softmax cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newOnehotLabels = onehotLabels * (1 - labelSmoothing)\n *                         + labelSmoothing / numClasses\n *\n * @param onehotLabels One hot encoded labels\n *    [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or 1, and must be\n *    broadcastable to `loss`  of shape [batch_size]\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction softmaxCrossEntropy_<T extends Tensor, O extends Tensor>(\n    onehotLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $onehotLabels =\n      convertToTensor(onehotLabels, 'onehotLabels', 'softmaxCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'softmaxCrossEntropy');\n  let $weights: Tensor = null;\n\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'softmaxCrossEntropy');\n  }\n\n  assertShapesMatch(\n      $onehotLabels.shape, $logits.shape, 'Error in softmaxCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const numClasses = scalar($onehotLabels.shape[1]);\n\n    $onehotLabels =\n        add(mul($onehotLabels, sub(one, labelSmoothingScalar)),\n            div(labelSmoothingScalar, numClasses));\n  }\n\n  const losses = softmaxCrossEntropyWithLogits_($onehotLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const softmaxCrossEntropy = op({softmaxCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Modularized ops.\nexport {abs} from './abs';\nexport {acos} from './acos';\nexport {acosh} from './acosh';\nexport {add} from './add';\nexport {addN} from './add_n';\nexport {all} from './all';\nexport {any} from './any';\nexport {argMax} from './arg_max';\nexport {argMin} from './arg_min';\nexport {asin} from './asin';\nexport {asinh} from './asinh';\nexport {atan} from './atan';\nexport {atan2} from './atan2';\nexport {atanh} from './atanh';\nexport {avgPool} from './avg_pool';\nexport {avgPool3d} from './avg_pool_3d';\nexport {basicLSTMCell} from './basic_lstm_cell';\nexport {batchToSpaceND} from './batch_to_space_nd';\nexport {batchNorm} from './batchnorm';\nexport {batchNorm2d} from './batchnorm2d';\nexport {batchNorm3d} from './batchnorm3d';\nexport {batchNorm4d} from './batchnorm4d';\nexport {broadcastTo} from './broadcast_to';\nexport {buffer} from './buffer';\nexport {cast} from './cast';\nexport {ceil} from './ceil';\nexport {clipByValue} from './clip_by_value';\nexport {clone} from './clone';\nexport {complex} from './complex';\nexport {concat} from './concat';\nexport {concat1d} from './concat_1d';\nexport {concat2d} from './concat_2d';\nexport {concat3d} from './concat_3d';\nexport {concat4d} from './concat_4d';\nexport {conv1d} from './conv1d';\nexport {conv2d} from './conv2d';\nexport {conv2dTranspose} from './conv2d_transpose';\nexport {conv3d} from './conv3d';\nexport {conv3dTranspose} from './conv3d_transpose';\nexport {cos} from './cos';\nexport {cosh} from './cosh';\nexport {cumsum} from './cumsum';\nexport {depthToSpace} from './depth_to_space';\nexport {depthwiseConv2d} from './depthwise_conv2d';\nexport {diag} from './diag';\nexport {dilation2d} from './dilation2d';\nexport {div} from './div';\nexport {divNoNan} from './div_no_nan';\nexport {dot} from './dot';\nexport {elu} from './elu';\nexport {equal} from './equal';\nexport {erf} from './erf';\nexport {exp} from './exp';\nexport {expandDims} from './expand_dims';\nexport {expm1} from './expm1';\nexport {eye} from './eye';\nexport {fill} from './fill';\nexport {floor} from './floor';\nexport {floorDiv} from './floorDiv';\nexport {gather} from './gather';\nexport {greater} from './greater';\nexport {greaterEqual} from './greater_equal';\nexport {imag} from './imag';\nexport {isFinite} from './is_finite';\nexport {isInf} from './is_inf';\nexport {isNaN} from './is_nan';\nexport {leakyRelu} from './leaky_relu';\nexport {less} from './less';\nexport {lessEqual} from './less_equal';\nexport {linspace} from './linspace';\nexport {localResponseNormalization} from './local_response_normalization';\nexport {log} from './log';\nexport {log1p} from './log1p';\nexport {logSigmoid} from './log_sigmoid';\nexport {logSoftmax} from './log_softmax';\nexport {logSumExp} from './log_sum_exp';\nexport {logicalAnd} from './logical_and';\nexport {logicalNot} from './logical_not';\nexport {logicalOr} from './logical_or';\nexport {logicalXor} from './logical_xor';\nexport {matMul} from './mat_mul';\nexport {max} from './max';\nexport {maxPool} from './max_pool';\nexport {maxPool3d} from './max_pool_3d';\nexport {maxPoolWithArgmax} from './max_pool_with_argmax';\nexport {maximum} from './maximum';\nexport {mean} from './mean';\nexport {min} from './min';\nexport {minimum} from './minimum';\nexport {mod} from './mod';\nexport {moments} from './moments';\nexport {mul} from './mul';\nexport {LSTMCellFunc, multiRNNCell} from './multi_rnn_cell';\nexport {multinomial} from './multinomial';\nexport {neg} from './neg';\nexport {notEqual} from './not_equal';\nexport {oneHot} from './one_hot';\nexport {ones} from './ones';\nexport {onesLike} from './ones_like';\nexport {outerProduct} from './outer_product';\nexport {pad} from './pad';\nexport {pad1d} from './pad1d';\nexport {pad2d} from './pad2d';\nexport {pad3d} from './pad3d';\nexport {pad4d} from './pad4d';\nexport {pool} from './pool';\nexport {pow} from './pow';\nexport {prelu} from './prelu';\nexport {print} from './print';\nexport {prod} from './prod';\nexport {rand} from './rand';\nexport {randomGamma} from './random_gamma';\nexport {randomNormal} from './random_normal';\nexport {randomUniform} from './random_uniform';\nexport {range} from './range';\nexport {real} from './real';\nexport {reciprocal} from './reciprocal';\nexport {relu} from './relu';\nexport {relu6} from './relu6';\nexport {reshape} from './reshape';\nexport {reverse} from './reverse';\nexport {reverse1d} from './reverse_1d';\nexport {reverse2d} from './reverse_2d';\nexport {reverse3d} from './reverse_3d';\nexport {reverse4d} from './reverse_4d';\nexport {round} from './round';\nexport {rsqrt} from './rsqrt';\nexport {scalar} from './scalar';\nexport {selu} from './selu';\nexport {separableConv2d} from './separable_conv2d';\nexport {setdiff1dAsync} from './setdiff1d_async';\nexport {sigmoid} from './sigmoid';\nexport {sign} from './sign';\nexport {sin} from './sin';\nexport {sinh} from './sinh';\nexport {slice} from './slice';\nexport {slice1d} from './slice1d';\nexport {slice2d} from './slice2d';\nexport {slice3d} from './slice3d';\nexport {slice4d} from './slice4d';\nexport {softmax} from './softmax';\nexport {softplus} from './softplus';\nexport {spaceToBatchND} from './space_to_batch_nd';\nexport {fft} from './spectral/fft';\nexport {ifft} from './spectral/ifft';\nexport {irfft} from './spectral/irfft';\nexport {rfft} from './spectral/rfft';\nexport {split} from './split';\nexport {sqrt} from './sqrt';\nexport {square} from './square';\nexport {squaredDifference} from './squared_difference';\nexport {squeeze} from './squeeze';\nexport {stack} from './stack';\nexport {step} from './step';\nexport {stridedSlice} from './strided_slice';\nexport {sub} from './sub';\nexport {sum} from './sum';\nexport {tan} from './tan';\nexport {tanh} from './tanh';\nexport {tensor} from './tensor';\nexport {tensor1d} from './tensor1d';\nexport {tensor2d} from './tensor2d';\nexport {tensor3d} from './tensor3d';\nexport {tensor4d} from './tensor4d';\nexport {tensor5d} from './tensor5d';\nexport {tensor6d} from './tensor6d';\nexport {tile} from './tile';\nexport {topk} from './topk';\nexport {truncatedNormal} from './truncated_normal';\nexport {unique} from './unique';\nexport {unsortedSegmentSum} from './unsorted_segment_sum';\nexport {unstack} from './unstack';\nexport {variable} from './variable';\nexport {where} from './where';\nexport {whereAsync} from './where_async';\nexport {zeros} from './zeros';\nexport {zerosLike} from './zeros_like';\n\nexport * from './boolean_mask';\nexport * from './compare';\nexport * from './binary_ops';\nexport * from './transpose';\nexport * from './norm';\nexport * from './moving_average';\nexport * from './scatter_nd';\nexport * from './sparse_to_dense';\nexport * from './gather_nd';\nexport * from './dropout';\nexport * from './signal_ops_util';\nexport * from './in_top_k';\n\nexport {op, OP_SCOPE_SUFFIX} from './operation';\n\nimport {rfft} from './spectral/rfft';\nimport {fft} from './spectral/fft';\nimport {ifft} from './spectral/ifft';\nimport {irfft} from './spectral/irfft';\nconst spectral = {\n  fft,\n  ifft,\n  rfft,\n  irfft\n};\n\nimport * as fused from './fused_ops';\n\nimport {hammingWindow} from './signal/hamming_window';\nimport {hannWindow} from './signal/hann_window';\nimport {frame} from './signal/frame';\nimport {stft} from './signal/stft';\nconst signal = {\n  hammingWindow,\n  hannWindow,\n  frame,\n  stft,\n};\n\n// Image Ops namespace\nimport {cropAndResize} from './image/crop_and_resize';\nimport {flipLeftRight} from './image/flip_left_right';\nimport {rotateWithOffset} from './image/rotate_with_offset';\nimport {nonMaxSuppression} from './image/non_max_suppression';\nimport {nonMaxSuppressionAsync} from './image/non_max_suppression_async';\nimport {nonMaxSuppressionWithScore} from './image/non_max_suppression_with_score';\nimport {nonMaxSuppressionWithScoreAsync} from './image/non_max_suppression_with_score_async';\nimport {nonMaxSuppressionPadded} from './image/non_max_suppression_padded';\nimport {nonMaxSuppressionPaddedAsync} from './image/non_max_suppression_padded_async';\nimport {resizeBilinear} from './image/resize_bilinear';\nimport {resizeNearestNeighbor} from './image/resize_nearest_neighbor';\nconst image = {\n  flipLeftRight,\n  resizeNearestNeighbor,\n  resizeBilinear,\n  rotateWithOffset,\n  cropAndResize,\n  nonMaxSuppression,\n  nonMaxSuppressionAsync,\n  nonMaxSuppressionWithScore,\n  nonMaxSuppressionWithScoreAsync,\n  nonMaxSuppressionPadded,\n  nonMaxSuppressionPaddedAsync\n};\n\n// linalg namespace\nimport {bandPart} from './linalg/band_part';\nimport {gramSchmidt} from './linalg/gram_schmidt';\nimport {qr} from './linalg/qr';\nconst linalg = {\n  bandPart,\n  gramSchmidt,\n  qr\n};\n\n// losses namespace;\nimport {absoluteDifference} from './losses/absolute_difference';\nimport {computeWeightedLoss} from './losses/compute_weighted_loss';\nimport {cosineDistance} from './losses/cosine_distance';\nimport {hingeLoss} from './losses/hinge_loss';\nimport {huberLoss} from './losses/huber_loss';\nimport {logLoss} from './losses/log_loss';\nimport {meanSquaredError} from './losses/mean_squared_error';\nimport {sigmoidCrossEntropy} from './losses/sigmoid_cross_entropy';\nimport {softmaxCrossEntropy} from './losses/softmax_cross_entropy';\nconst losses = {\n  absoluteDifference,\n  computeWeightedLoss,\n  cosineDistance,\n  hingeLoss,\n  huberLoss,\n  logLoss,\n  meanSquaredError,\n  sigmoidCrossEntropy,\n  softmaxCrossEntropy\n};\n\n// Second level exports.\nexport {image, linalg, losses, spectral, fused, signal};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {dispose} from '../globals';\nimport {variableGrads} from '../gradients';\nimport {scalar} from '../ops/ops';\nimport {Serializable} from '../serialization';\nimport {Scalar, Variable} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\n/**\n * A variable that belongs to an optimizer.\n *\n * The `originalName` field is required for keeping track of the canonical\n * name of the variable, which is usually the name of the model weight that\n * the variable is related to plus a suffix, e.g., 'dense1/kernel/momentum'.\n * The name of the `Variable` object itself cannot be used directly due to\n * possible deduplication: Every `Variable` must have a unique name but more\n * than one optimizer objects of the same type may be created for the same model\n * or the same `Variable`.\n */\nexport interface OptimizerVariable {\n  originalName: string;\n  variable: Variable;\n}\n\n/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */\nexport abstract class Optimizer extends Serializable {\n  protected iterations_: number;\n\n  /**\n   * Executes `f()` and minimizes the scalar output of `f()` by computing\n   * gradients of y with respect to the list of trainable variables provided by\n   * `varList`. If no list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to minimize.\n   * @param returnCost Whether to return the scalar cost value produced by\n   * executing `f()`.\n   * @param varList An optional list of variables to update. If specified, only\n   * the trainable variables in varList will be updated by minimize. Defaults to\n   * all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  minimize(f: () => Scalar, returnCost = false, varList?: Variable[]): Scalar\n      |null {\n    const {value, grads} = this.computeGradients(f, varList);\n\n    if (varList != null) {\n      const gradArray: NamedTensor[] =\n          varList.map(v => ({name: v.name, tensor: grads[v.name]}));\n      this.applyGradients(gradArray);\n    } else {\n      this.applyGradients(grads);\n    }\n\n    // Dispose gradients.\n    dispose(grads);\n\n    if (returnCost) {\n      return value;\n    } else {\n      value.dispose();\n      return null;\n    }\n  }\n\n  /**\n   * The number of iterations that this optimizer instance has been invoked for.\n   */\n  get iterations(): number {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return this.iterations_;\n  }\n\n  protected incrementIterations() {\n    this.iterations_ = this.iterations + 1;\n  }\n\n  /**\n   * Executes f() and computes the gradient of the scalar output of f() with\n   * respect to the list of trainable variables provided by `varList`. If no\n   * list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to use for computing\n   * gradients with respect to variables.\n   * @param varList An optional list of variables to compute gradients with\n   * respect to. If specified, only the trainable variables in varList will have\n   * gradients computed with respect to. Defaults to all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  computeGradients(f: () => Scalar, varList?: Variable[]):\n      {value: Scalar, grads: NamedTensorMap} {\n    return variableGrads(f, varList);\n  }\n\n  /**\n   * Updates variables by using the computed gradients.\n   *\n   * @param variableGradients A mapping of variable name to its gradient value.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  abstract applyGradients(variableGradients: NamedTensorMap|\n                          NamedTensor[]): void;\n\n  /**\n   * Dispose the variables (if any) owned by this optimizer instance.\n   */\n  dispose(): void {\n    if (this.iterations_ != null) {\n      dispose(this.iterations_);\n    }\n  }\n\n  async saveIterations(): Promise<NamedTensor> {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return {\n      name: 'iter',  // Named for Python compatibility.\n      // TODO(cais): Use 'int64' type when available.\n      tensor: scalar(this.iterations_, 'int32')\n    };\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for this optimizer yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error(\n        `setWeights() is not implemented for this optimizer class ` +\n        `${this.getClassName()}`);\n  }\n\n  /**\n   * Extract the first element of the weight values and set it\n   * as the iterations counter variable of this instance of optimizer.\n   *\n   * @param weightValues\n   * @returns Weight values with the first element consumed and excluded.\n   */\n  protected async extractIterations(weightValues: NamedTensor[]):\n      Promise<NamedTensor[]> {\n    this.iterations_ = (await weightValues[0].tensor.data())[0];\n    return weightValues.slice(1);\n  }\n}\n\nObject.defineProperty(Optimizer, Symbol.hasInstance, {\n  value: (instance: Optimizer) => {\n    return instance.minimize != null && instance.computeGradients != null &&\n        instance.applyGradients != null;\n  }\n});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/ops';\nimport {square} from '../ops/square';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdadeltaOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adadelta';  // Name matters for Python compatibility.\n  private accumulatedGrads: OptimizerVariable[] = [];\n  private accumulatedUpdates: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected rho: number,\n      protected epsilon: number = null) {\n    super();\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedGrads[i] == null) {\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accum_grad`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedUpdates[i] == null) {\n        this.accumulatedUpdates[i] = {\n          originalName: `${name}/accum_var`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n      const accumulatedUpdate = this.accumulatedUpdates[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad =\n            add(mul(accumulatedGrad, this.rho),\n                mul(square(gradient), 1 - this.rho));\n\n        const updates =\n            mul(div(sqrt(add(accumulatedUpdate, this.epsilon)),\n                    sqrt(add(accumulatedGrad, this.epsilon))),\n                gradient);\n\n        const newAccumulatedUpdate =\n            add(mul(accumulatedUpdate, this.rho),\n                mul(square(updates), 1 - this.rho));\n\n        accumulatedGrad.assign(newAccumulatedGrad);\n        accumulatedUpdate.assign(newAccumulatedUpdate);\n\n        const newValue = add(mul(updates, -this.learningRate), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedUpdates != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n      dispose(this.accumulatedUpdates.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedGrads, ...this.accumulatedUpdates];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedGrads =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedUpdates =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'rho': this.rho,\n      'epsilon': this.epsilon\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['rho'], config['epsilon']);\n  }\n}\nregisterClass(AdadeltaOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {fill} from '../ops/fill';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdagradOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adagrad';  // Note: Name matters for Python compatibility.\n\n  private accumulatedGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private initialAccumulatorValue = 0.1) {\n    super();\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulatedGrads[i] == null) {\n        const trainable = false;\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accumulator`,\n          variable: tidy(\n              () => fill(value.shape, this.initialAccumulatorValue)\n                        .variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad = add(accumulatedGrad, square(gradient));\n        accumulatedGrad.assign(newAccumulatedGrad);\n\n        const newValue = add(\n            mul(div(gradient,\n                    sqrt(add(newAccumulatedGrad, ENGINE.backend.epsilon()))),\n                -this.learningRate),\n            value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedGrads != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulatedGrads.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulatedGrads = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'initialAccumulatorValue': this.initialAccumulatorValue,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['initialAccumulatorValue']);\n  }\n}\nregisterClass(AdagradOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adam';  // Note: Name matters for Python compatibility.\n  private accBeta1: Variable;\n  private accBeta2: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedSecondMoment: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null) {\n    super();\n    tidy(() => {\n      // accB* will be updated by batch.\n      this.accBeta1 = scalar(beta1).variable();\n      this.accBeta2 = scalar(beta2).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const oneMinusAccBeta2 = sub(1, this.accBeta2);\n\n      varNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n        if (this.accumulatedSecondMoment[i] == null) {\n          this.accumulatedSecondMoment[i] = {\n            originalName: `${name}/v`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const secondMoment = this.accumulatedSecondMoment[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n        const newSecondMoment =\n            add(mul(secondMoment, this.beta2),\n                mul(square(gradient), 1 - this.beta2));\n\n        const biasCorrectedFirstMoment = div(newFirstMoment, oneMinusAccBeta1);\n        const biasCorrectedSecondMoment =\n            div(newSecondMoment, oneMinusAccBeta2);\n\n        firstMoment.assign(newFirstMoment);\n        secondMoment.assign(newSecondMoment);\n\n        const newValue =\n            add(mul(div(biasCorrectedFirstMoment,\n                        add(sqrt(biasCorrectedSecondMoment), this.epsilon)),\n                    -this.learningRate),\n                value);\n        value.assign(newValue);\n      });\n\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n      this.accBeta2.assign(mul(this.accBeta2, this.beta2));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.accBeta2.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedSecondMoment != null) {\n      dispose(this.accumulatedSecondMoment.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedFirstMoment, ...this.accumulatedSecondMoment];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    tidy(() => {\n      this.accBeta1.assign(pow(this.beta1, this.iterations_ + 1));\n      this.accBeta2.assign(pow(this.beta2, this.iterations_ + 1));\n    });\n\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedFirstMoment =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedSecondMoment =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon']);\n  }\n}\nregisterClass(AdamOptimizer);\n", "\uFEFF/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {abs} from '../ops/abs';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {maximum} from '../ops/maximum';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamaxOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adamax';  // Note: Name matters for Python compatbility.\n  private accBeta1: Variable;\n  private iteration: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null,\n      protected decay = 0.0) {\n    super();\n\n    tidy(() => {\n      this.iteration = scalar(0).variable();\n      this.accBeta1 = scalar(beta1).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const lr =\n          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));\n\n      variableNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n        if (this.accumulatedWeightedInfNorm[i] == null) {\n          this.accumulatedWeightedInfNorm[i] = {\n            originalName: `${name}/v`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n\n        const ut0 = mul(weightedInfNorm, this.beta2);\n        const ut1 = abs(gradient);\n\n        const newWeightedInfNorm = maximum(ut0, ut1);\n\n        firstMoment.assign(newFirstMoment);\n        weightedInfNorm.assign(newWeightedInfNorm);\n\n        const newValue =\n            add(mul(div(lr, oneMinusAccBeta1),\n                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),\n                value);\n\n        value.assign(newValue);\n      });\n\n      this.iteration.assign(add(this.iteration, 1));\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.iteration.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedWeightedInfNorm != null) {\n      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for Adamax yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error('setWeights() is not implemented for Adamax yet.');\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n      'decay': this.decay\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon'], config['decay']);\n  }\n}\nregisterClass(AdamaxOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {keep, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer} from './optimizer';\n\n/** @doclink Optimizer */\nexport class SGDOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'SGD';  // Note: Name matters for Python compatibility.\n  protected c: Scalar;\n\n  constructor(protected learningRate: number) {\n    super();\n    this.setLearningRate(learningRate);\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    varNames.forEach((name, i) => {\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n      const value = ENGINE.registeredVariables[name];\n      tidy(() => {\n        const newValue = add(mul(this.c, gradient), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  /**\n   * Sets the learning rate of the optimizer.\n   */\n  setLearningRate(learningRate: number) {\n    this.learningRate = learningRate;\n    if (this.c != null) {\n      this.c.dispose();\n    }\n    this.c = keep(scalar(-learningRate));\n  }\n\n  dispose() {\n    this.c.dispose();\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    return [await this.saveIterations()];\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    if (weightValues.length !== 0) {\n      throw new Error('SGD optimizer does not have settable weights.');\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {'learningRate': this.learningRate};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate']);\n  }\n}\nregisterClass(SGDOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {OptimizerVariable} from './optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\n/** @doclink Optimizer */\nexport class MomentumOptimizer extends SGDOptimizer {\n  /** @nocollapse */\n  static className = 'Momentum';  // Name matters for Python compatibility.\n  private m: Scalar;\n  private accumulations: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private momentum: number,\n      private useNesterov = false) {\n    super(learningRate);\n    this.m = scalar(this.momentum);\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulations[i] == null) {\n        const trainable = false;\n        this.accumulations[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const accumulation = this.accumulations[i].variable;\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      tidy(() => {\n        let newValue: Tensor;\n        const newAccumulation = add(mul(this.m, accumulation), gradient);\n        if (this.useNesterov) {\n          newValue = add(\n              mul(this.c, add(gradient, mul(newAccumulation, this.m))), value);\n        } else {\n          newValue = add(mul(this.c, newAccumulation), value);\n        }\n        accumulation.assign(newAccumulation);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.m.dispose();\n    if (this.accumulations != null) {\n      dispose(this.accumulations.map(v => v.variable));\n    }\n  }\n\n  /**\n   * Sets the momentum of the optimizer.\n   *\n   * @param momentum\n   */\n  setMomentum(momentum: number) {\n    this.momentum = momentum;\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulations.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulations = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'momentum': this.momentum,\n      'useNesterov': this.useNesterov\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['momentum'], config['useNesterov']);\n  }\n}\nregisterClass(MomentumOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class RMSPropOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'RMSProp';  // Note: Name matters for Python compatibility.\n  private centered: boolean;\n\n  private accumulatedMeanSquares: OptimizerVariable[] = [];\n  private accumulatedMoments: OptimizerVariable[] = [];\n  private accumulatedMeanGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected decay = 0.9,\n      protected momentum = 0.0, protected epsilon: number = null,\n      centered = false) {\n    super();\n\n    this.centered = centered;\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n    if (learningRate == null) {\n      throw new Error(`learningRate for RMSPropOptimizer must be defined.`);\n    }\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedMeanSquares[i] == null) {\n        this.accumulatedMeanSquares[i] = {\n          originalName: `${name}/rms`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMoments[i] == null) {\n        this.accumulatedMoments[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMeanGrads[i] == null && this.centered) {\n        this.accumulatedMeanGrads[i] = {\n          originalName: `${name}/mg`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedMeanSquare = this.accumulatedMeanSquares[i].variable;\n      const accumulatedMoments = this.accumulatedMoments[i].variable;\n      tidy(() => {\n        const newAccumulatedMeanSquare =\n            add(mul(accumulatedMeanSquare, this.decay),\n                mul(square(gradient), 1 - this.decay));\n\n        if (this.centered) {\n          const accumulatedMeanGrad = this.accumulatedMeanGrads[i].variable;\n          // Centered gradient\n          const newAccumulatedMeanGrad =\n              add(mul(accumulatedMeanGrad, this.decay),\n                  mul(gradient, 1 - this.decay));\n\n          const gradContribution =\n              div(mul(gradient, this.learningRate),\n                  sqrt(\n                      sub(newAccumulatedMeanSquare,\n                          add(square(newAccumulatedMeanGrad), this.epsilon))));\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum), gradContribution);\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMeanGrad.assign(newAccumulatedMeanGrad);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        } else {\n          // Plain gradient\n          const newAccumulatedMeanSquare =\n              add(mul(accumulatedMeanSquare, this.decay),\n                  mul(square(gradient), 1 - this.decay));\n\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum),\n                  div(mul(gradient, this.learningRate),\n                      sqrt(add(newAccumulatedMeanSquare, this.epsilon))));\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        }\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedMeanSquares != null) {\n      dispose(this.accumulatedMeanSquares.map(v => v.variable));\n    }\n    if (this.accumulatedMeanGrads != null && this.centered) {\n      dispose(this.accumulatedMeanGrads.map(v => v.variable));\n    }\n    if (this.accumulatedMoments != null) {\n      dispose(this.accumulatedMoments.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedMeanSquares, ...this.accumulatedMoments];\n    if (this.centered) {\n      variables.push(...this.accumulatedMeanGrads);\n    }\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount =\n        this.centered ? weightValues.length / 3 : weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedMeanSquares =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedMoments =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n    if (this.centered) {\n      this.accumulatedMeanGrads =\n          weightValues.slice(variableCount * 2, variableCount * 3)\n              .map(v => ({\n                     originalName: v.name,\n                     variable: v.tensor.variable(trainable)\n                   }));\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'decay': this.decay,\n      'momentum': this.momentum,\n      'epsilon': this.epsilon,\n      'centered': this.centered\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['decay'], config['momentum'],\n        config['epsilon'], config['centered']);\n  }\n}\nregisterClass(RMSPropOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AdadeltaOptimizer} from './adadelta_optimizer';\nimport {AdagradOptimizer} from './adagrad_optimizer';\nimport {AdamOptimizer} from './adam_optimizer';\nimport {AdamaxOptimizer} from './adamax_optimizer';\nimport {MomentumOptimizer} from './momentum_optimizer';\nimport {RMSPropOptimizer} from './rmsprop_optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\nexport class OptimizerConstructors {\n  /**\n   * Constructs a `tf.SGDOptimizer` that uses stochastic gradient descent.\n   *\n   * ```js\n   * // Fit a quadratic function by learning the coefficients a, b, c.\n   * const xs = tf.tensor1d([0, 1, 2, 3]);\n   * const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);\n   *\n   * const a = tf.scalar(Math.random()).variable();\n   * const b = tf.scalar(Math.random()).variable();\n   * const c = tf.scalar(Math.random()).variable();\n   *\n   * // y = a * x^2 + b * x + c.\n   * const f = x => a.mul(x.square()).add(b.mul(x)).add(c);\n   * const loss = (pred, label) => pred.sub(label).square().mean();\n   *\n   * const learningRate = 0.01;\n   * const optimizer = tf.train.sgd(learningRate);\n   *\n   * // Train the model.\n   * for (let i = 0; i < 10; i++) {\n   *   optimizer.minimize(() => loss(f(xs), ys));\n   * }\n   *\n   * // Make predictions.\n   * console.log(\n   *     `a: ${a.dataSync()}, b: ${b.dataSync()}, c: ${c.dataSync()}`);\n   * const preds = f(xs).dataSync();\n   * preds.forEach((pred, i) => {\n   *   console.log(`x: ${i}, pred: ${pred}`);\n   * });\n   * ```\n   *\n   * @param learningRate The learning rate to use for the SGD algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static sgd(learningRate: number): SGDOptimizer {\n    return new SGDOptimizer(learningRate);\n  }\n\n  /**\n   * Constructs a `tf.MomentumOptimizer` that uses momentum gradient\n   * descent.\n   *\n   * See\n   * [http://proceedings.mlr.press/v28/sutskever13.pdf](\n   * http://proceedings.mlr.press/v28/sutskever13.pdf)\n   *\n   * @param learningRate The learning rate to use for the Momentum gradient\n   * descent algorithm.\n   * @param momentum The momentum to use for the momentum gradient descent\n   * algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static momentum(learningRate: number, momentum: number, useNesterov = false):\n      MomentumOptimizer {\n    return new MomentumOptimizer(learningRate, momentum, useNesterov);\n  }\n\n  /**\n   * Constructs a `tf.RMSPropOptimizer` that uses RMSProp gradient\n   * descent. This implementation uses plain momentum and is not centered\n   * version of RMSProp.\n   *\n   * See\n   * [http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf](\n   * http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)\n   *\n   * @param learningRate The learning rate to use for the RMSProp gradient\n   * descent algorithm.\n   * @param decay The discounting factor for the history/coming gradient.\n   * @param momentum The momentum to use for the RMSProp gradient descent\n   * algorithm.\n   * @param epsilon Small value to avoid zero denominator.\n   * @param centered If true, gradients are normalized by the estimated\n   * variance of the gradient.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static rmsprop(\n      learningRate: number, decay = .9, momentum = 0.0, epsilon: number = null,\n      centered = false): RMSPropOptimizer {\n    return new RMSPropOptimizer(\n        learningRate, decay, momentum, epsilon, centered);\n  }\n\n  /**\n   * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adam gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adam(\n      learningRate = 0.001, beta1 = 0.9, beta2 = 0.999,\n      epsilon: number = null): AdamOptimizer {\n    return new AdamOptimizer(learningRate, beta1, beta2, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.\n   * See [https://arxiv.org/abs/1212.5701](https://arxiv.org/abs/1212.5701)\n   *\n   * @param learningRate The learning rate to use for the Adadelta gradient\n   * descent algorithm.\n   * @param rho The learning rate decay over each update.\n   * @param epsilon A constant epsilon used to better condition the grad\n   * update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adadelta(learningRate = .001, rho = .95, epsilon: number = null):\n      AdadeltaOptimizer {\n    return new AdadeltaOptimizer(learningRate, rho, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdamaxOptimizer` that uses the Adamax algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adamax gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   * @param decay The learning rate decay over each update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adamax(\n      learningRate = 0.002, beta1 = 0.9, beta2 = 0.999, epsilon: number = null,\n      decay = 0.0): AdamaxOptimizer {\n    return new AdamaxOptimizer(learningRate, beta1, beta2, epsilon, decay);\n  }\n\n  /**\n   * Constructs a `tf.AdagradOptimizer` that uses the Adagrad algorithm.\n   * See\n   * [http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf](\n   * http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\n   * or\n   * [http://ruder.io/optimizing-gradient-descent/index.html#adagrad](\n   * http://ruder.io/optimizing-gradient-descent/index.html#adagrad)\n   *\n   * @param learningRate The learning rate to use for the Adagrad gradient\n   * descent algorithm.\n   * @param initialAccumulatorValue Starting value for the accumulators, must be\n   * positive.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adagrad(learningRate: number, initialAccumulatorValue = 0.1):\n      AdagradOptimizer {\n    return new AdagradOptimizer(learningRate, initialAccumulatorValue);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// So typings can propagate.\nimport {AdadeltaOptimizer} from './optimizers/adadelta_optimizer';\nimport {AdagradOptimizer} from './optimizers/adagrad_optimizer';\nimport {AdamOptimizer} from './optimizers/adam_optimizer';\nimport {AdamaxOptimizer} from './optimizers/adamax_optimizer';\nimport {MomentumOptimizer} from './optimizers/momentum_optimizer';\nimport {OptimizerConstructors} from './optimizers/optimizer_constructors';\nimport {RMSPropOptimizer} from './optimizers/rmsprop_optimizer';\nimport {SGDOptimizer} from './optimizers/sgd_optimizer';\n\n// tslint:disable-next-line:no-unused-expression\n[MomentumOptimizer, SGDOptimizer, AdadeltaOptimizer, AdagradOptimizer,\n RMSPropOptimizer, AdamaxOptimizer, AdamOptimizer];\n\nexport const train = {\n  sgd: OptimizerConstructors.sgd,\n  momentum: OptimizerConstructors.momentum,\n  adadelta: OptimizerConstructors.adadelta,\n  adagrad: OptimizerConstructors.adagrad,\n  rmsprop: OptimizerConstructors.rmsprop,\n  adamax: OptimizerConstructors.adamax,\n  adam: OptimizerConstructors.adam\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst delayCallback: Function = (() => {\n  if (typeof requestAnimationFrame !== 'undefined') {\n    return requestAnimationFrame;\n  } else if (typeof setImmediate !== 'undefined') {\n    return setImmediate;\n  }\n  return (f: Function) => f();  // no delays\n})();\n\n/**\n * Returns a promise that resolve when a requestAnimationFrame has completed.\n *\n * On Node.js this uses setImmediate instead of requestAnimationFrame.\n *\n * This is simply a sugar method so that users can do the following:\n * `await tf.nextFrame();`\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nfunction nextFrame(): Promise<void> {\n  return new Promise<void>(resolve => delayCallback(() => resolve()));\n}\n\nexport {nextFrame};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Returns the image center in pixels.\nexport function getImageCenter(\n    center: number|[number, number], imageHeight: number,\n    imageWidth: number): [number, number] {\n  const centerX =\n      imageWidth * (typeof center === 'number' ? center : center[0]);\n  const centerY =\n      imageHeight * (typeof center === 'number' ? center : center[1]);\n  return [centerX, centerY];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Gets the new shape of the input Tensor after it's been reshaped\n * to:\n * [blockShape[0], ..., blockShape[M-1], batch / prod(blockShape),\n * inputShape[1], ..., inputShape[N-1]]\n *\n * See step 1: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshaped(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  let reshaped: number[] = [];\n  if (batchToSpace) {\n    reshaped = reshaped.concat(blockShape.slice(0));\n    reshaped.push(inputShape[0] / prod);\n    reshaped = reshaped.concat(inputShape.slice(1));\n  } else {\n    reshaped = reshaped.concat(inputShape[0]);\n    const spatialLength = blockShape.length;\n    for (let i = 0; i < spatialLength; ++i) {\n      reshaped =\n          reshaped.concat([inputShape[i + 1] / blockShape[i], blockShape[i]]);\n    }\n    reshaped = reshaped.concat(inputShape.slice(spatialLength + 1));\n  }\n  return reshaped;\n}\n\n/**\n * Gets the permutation that will transpose the dimensions of the\n * reshaped tensor to shape:\n *\n * [batch / prod(block_shape),inputShape[1], blockShape[0], ...,\n * inputShape[M], blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * see step 2: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getPermuted(\n    reshapedRank: number, blockShapeRank: number,\n    batchToSpace = true): number[] {\n  const permuted = [];\n  if (batchToSpace) {\n    permuted.push(blockShapeRank);\n    for (let i = blockShapeRank + 1; i < reshapedRank; ++i) {\n      if (i <= 2 * blockShapeRank) {\n        permuted.push(i);\n        permuted.push(i - (blockShapeRank + 1));\n      } else {\n        permuted.push(i);\n      }\n    }\n  } else {\n    const permutedBeforeBatch = [];\n    const permutedAfterBatch = [];\n    for (let i = 1; i < reshapedRank; ++i) {\n      if (i >= blockShapeRank * 2 + 1 || i % 2 === 1) {\n        permutedAfterBatch.push(i);\n      } else {\n        permutedBeforeBatch.push(i);\n      }\n    }\n    permuted.push(...permutedBeforeBatch);\n    permuted.push(0);\n    permuted.push(...permutedAfterBatch);\n  }\n  return permuted;\n}\n\n/**\n * Gets the shape of the reshaped and permuted input Tensor before any cropping\n * is applied.  The new shape will be:\n *\n * [batch / prod(blockShape),inputShape[1] * blockShape[0], ...,\n * inputShape[M] * blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 3: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshapedPermuted(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  const reshapedPermuted = [];\n\n  if (batchToSpace) {\n    reshapedPermuted.push(inputShape[0] / prod);\n  } else {\n    reshapedPermuted.push(inputShape[0] * prod);\n  }\n\n  for (let i = 1; i < inputShape.length; ++i) {\n    if (i <= blockShape.length) {\n      if (batchToSpace) {\n        reshapedPermuted.push(blockShape[i - 1] * inputShape[i]);\n      } else {\n        reshapedPermuted.push(inputShape[i] / blockShape[i - 1]);\n      }\n    } else {\n      reshapedPermuted.push(inputShape[i]);\n    }\n  }\n\n  return reshapedPermuted;\n}\n\n/**\n * Converts the crops argument into the beginning coordinates of a slice\n * operation.\n */\nexport function getSliceBeginCoords(\n    crops: number[][], blockShape: number): number[] {\n  const sliceBeginCoords = [0];\n  for (let i = 0; i < blockShape; ++i) {\n    sliceBeginCoords.push(crops[i][0]);\n  }\n  return sliceBeginCoords;\n}\n\n/**\n * Converts the crops argument into the size of a slice operation.  When\n * combined with getSliceBeginCoords this function allows the reshaped and\n * permuted Tensor to be cropped to its final output shape of:\n *\n * inputShape[1] * blockShape[0] - crops[0,0] - crops[0,1], ...,\n * inputShape[M] * blockShape[M-1] -crops[M-1,0] -\n * crops[M-1,1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 4: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getSliceSize(\n    uncroppedShape: number[], crops: number[][], blockShape: number): number[] {\n  const sliceSize = uncroppedShape.slice(0, 1);\n  for (let i = 0; i < blockShape; ++i) {\n    sliceSize.push(uncroppedShape[i + 1] - crops[i][0] - crops[i][1]);\n  }\n\n  return sliceSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const SELU_SCALEALPHA = 1.7580993408473768599402175208123;\nexport const SELU_SCALE = 1.0507009873554804934193349852946;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const ERF_P = 0.3275911;\nexport const ERF_A1 = 0.254829592;\nexport const ERF_A2 = -0.284496736;\nexport const ERF_A3 = 1.421413741;\nexport const ERF_A4 = -1.453152027;\nexport const ERF_A5 = 1.061405429;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\n\nexport function warn(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.warn(...msg);\n  }\n}\n\nexport function log(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.log(...msg);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TypedArray} from '../types';\n/**\n * Merges real and imaginary Float32Arrays into a single complex Float32Array.\n *\n * The memory layout is interleaved as follows:\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n * complex: [r0, i0, r1, i1, r2, i2]\n *\n * This is the inverse of splitRealAndImagArrays.\n *\n * @param real The real values of the complex tensor values.\n * @param imag The imag values of the complex tensor values.\n * @returns A complex tensor as a Float32Array with merged values.\n */\nexport function mergeRealAndImagArrays(\n    real: Float32Array, imag: Float32Array): Float32Array {\n  if (real.length !== imag.length) {\n    throw new Error(\n        `Cannot merge real and imag arrays of different lengths. real:` +\n        `${real.length}, imag: ${imag.length}.`);\n  }\n  const result = new Float32Array(real.length * 2);\n  for (let i = 0; i < result.length; i += 2) {\n    result[i] = real[i / 2];\n    result[i + 1] = imag[i / 2];\n  }\n  return result;\n}\n\n/**\n * Splits a complex Float32Array into real and imag parts.\n *\n * The memory layout is interleaved as follows:\n * complex: [r0, i0, r1, i1, r2, i2]\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n *\n * This is the inverse of mergeRealAndImagArrays.\n *\n * @param complex The complex tensor values.\n * @returns An object with real and imag Float32Array components of the complex\n *     tensor.\n */\nexport function splitRealAndImagArrays(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(complex.length / 2);\n  const imag = new Float32Array(complex.length / 2);\n  for (let i = 0; i < complex.length; i += 2) {\n    real[i / 2] = complex[i];\n    imag[i / 2] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts even indexed complex values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithEvenIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.ceil(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 0; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts odd indexed comple values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithOddIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.floor(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 2; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Get the map representing a complex value in the given array.\n * @param complex The complex tensor values.\n * @param index An index of the target complex value.\n */\nexport function getComplexWithIndex(\n    complex: Float32Array, index: number): {real: number, imag: number} {\n  const real = complex[index * 2];\n  const imag = complex[index * 2 + 1];\n  return {real, imag};\n}\n\n/**\n * Insert a given complex value into the TypedArray.\n * @param data The array in which the complex value is inserted.\n * @param c The complex value to be inserted.\n * @param index An index of the target complex value.\n */\nexport function assignToTypedArray(\n    data: TypedArray, real: number, imag: number, index: number) {\n  data[index * 2] = real;\n  data[index * 2 + 1] = imag;\n}\n\n/**\n * Make the list of exponent terms used by FFT.\n */\nexport function exponents(\n    n: number, inverse: boolean): {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(n / 2);\n  const imag = new Float32Array(n / 2);\n  for (let i = 0; i < Math.ceil(n / 2); i++) {\n    const x = (inverse ? 2 : -2) * Math.PI * (i / n);\n    real[i] = Math.cos(x);\n    imag[i] = Math.sin(x);\n  }\n  return {real, imag};\n}\n\n/**\n * Make the exponent term used by FFT.\n */\nexport function exponent(\n    k: number, n: number, inverse: boolean): {real: number, imag: number} {\n  const x = (inverse ? 2 : -2) * Math.PI * (k / n);\n  const real = Math.cos(x);\n  const imag = Math.sin(x);\n  return {real, imag};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {cast} from '../ops/cast';\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\nimport {Rank} from '../types';\nimport {DataType, ShapeMap} from '../types';\nimport {hasEncodingLoss, makeZerosTypedArray} from '../util';\n\nimport {KernelBackend} from './backend';\n\n// Utilities needed by backend consumers of tf-core.\nexport * from '../ops/axis_util';\nexport * from '../ops/broadcast_util';\nexport * from '../ops/concat_util';\nexport * from '../ops/conv_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/fused_types';\nexport * from '../ops/reduce_util';\n\nimport * as slice_util from '../ops/slice_util';\nexport {slice_util};\n\nexport {BackendValues, TypedArray, upcastType, PixelData} from '../types';\nexport {MemoryInfo, TimingInfo} from '../engine';\nexport * from '../ops/rotate_util';\nexport * from '../ops/array_ops_util';\nexport * from '../ops/gather_nd_util';\nexport * from '../ops/scatter_nd_util';\nexport * from '../ops/selu_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/erf_util';\nexport * from '../log';\nexport * from '../backends/complex_util';\nexport * from '../ops/split_util';\n\nimport * as segment_util from '../ops/segment_util';\nexport {segment_util};\n\nexport function castTensor<T extends Tensor>(\n    x: T, dtype: DataType, backend: KernelBackend): T {\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return x.clone();\n    }\n    const zerosTensor = zeros(x.shape);\n    const floatX = cast(x, 'float32');\n    const result = backend.complex(floatX, zerosTensor);\n    zerosTensor.dispose();\n    floatX.dispose();\n    return result as T;\n  }\n\n  if (!hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    return ENGINE.makeTensorFromDataId(x.dataId, x.shape, dtype) as T;\n  }\n  if (x.dtype === 'complex64') {\n    const real = backend.real(x);\n    const result = cast(real, dtype);\n    real.dispose();\n    return result;\n  }\n  if (dtype === 'int32') {\n    return backend.int(x);\n  } else if (dtype === 'bool') {\n    const zero = scalar(0, x.dtype);\n    const result = backend.notEqual(x, zero) as T;\n    zero.dispose();\n    return result;\n  } else {\n    throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n  }\n}\n\nexport function reshapeTensor<T extends Tensor, R extends Rank>(\n    x: T, shape: ShapeMap[R]): Tensor<R> {\n  return ENGINE.makeTensorFromDataId(x.dataId, shape, x.dtype) as Tensor<R>;\n}\n\nexport function linspaceImpl(start: number, stop: number, num: number) {\n  const step = (stop - start) / (num - 1);\n\n  const values = makeZerosTypedArray(num, 'float32');\n  values[0] = start;\n  for (let i = 1; i < values.length; i++) {\n    values[i] = values[i - 1] + step;\n  }\n\n  return tensor1d(values, 'float32');\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\n// TODO(annxingyuan): Use this helper in WASM Split kernel once intermediate\n// kernels have been modularized in WebGL and CPU\n// https://github.com/tensorflow/tfjs/issues/2822.\n/** Shared implementation of the split kernel across WebGL and CPU. */\nexport function split<T extends Tensor>(\n    x: T, sizeSplits: number[], axis: number): T[] {\n  const begin = new Array(x.rank).fill(0);\n  const size = x.shape.slice();\n  return sizeSplits.map(s => {\n    const sliceSize = [...size];\n    sliceSize[axis] = s;\n    const sliceT = slice(x, begin, sliceSize);\n    begin[axis] += s;\n    return sliceT;\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * An implementation of the tile kernel shared between webgl and cpu for string\n * tensors only.\n */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {DataType, Rank} from '../types';\n\nexport function tile<R extends Rank>(\n    xBuf: TensorBuffer<R, DataType>, reps: number[]): Tensor<R> {\n  const newShape: number[] = new Array(xBuf.rank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = xBuf.shape[i] * reps[i];\n  }\n  const result = buffer(newShape, xBuf.dtype);\n  for (let i = 0; i < result.values.length; ++i) {\n    const newLoc = result.indexToLoc(i);\n\n    const originalLoc: number[] = new Array(xBuf.rank);\n    for (let j = 0; j < originalLoc.length; j++) {\n      originalLoc[j] = newLoc[j] % xBuf.shape[j];\n    }\n\n    const originalIndex = xBuf.locToIndex(originalLoc);\n\n    result.values[i] = xBuf.values[originalIndex];\n  }\n  return result.toTensor() as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the TopK kernel shared between webgl and cpu. */\n\nimport {tensor} from '../ops/tensor';\nimport {Tensor} from '../tensor';\nimport {NumericDataType, TypedArray} from '../types';\nimport {getTypedArrayFromDType} from '../util';\n\nexport function topkImpl<T extends Tensor>(\n    x: TypedArray, xShape: number[], xDtype: NumericDataType, k: number,\n    sorted: boolean): [T, T] {\n  // Reshape into a 2d tensor [batch, lastDim] and compute topk along lastDim.\n  const lastDim = xShape[xShape.length - 1];\n  const [batch, size] = [x.length / lastDim, lastDim];\n  const allTopKVals = getTypedArrayFromDType(xDtype, batch * k);\n  const allTopKIndices = getTypedArrayFromDType('int32', batch * k);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = x.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    const outOffset = b * k;\n    const topKVals = allTopKVals.subarray(outOffset, outOffset + k);\n    const topKIndices = allTopKIndices.subarray(outOffset, outOffset + k);\n    for (let i = 0; i < k; i++) {\n      topKVals[i] = valAndInd[i].value;\n      topKIndices[i] = valAndInd[i].index;\n    }\n  }\n  // Reshape back to the original input shape, except that the last\n  // dimension is k.\n  const outputShape = xShape.slice();\n  outputShape[outputShape.length - 1] = k;\n  return [\n    tensor(allTopKVals, outputShape, xDtype) as T,\n    tensor(allTopKIndices, outputShape, 'int32') as T\n  ];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport {nonMaxSuppressionV3Impl, nonMaxSuppressionV4Impl, nonMaxSuppressionV5Impl} from './non_max_suppression_impl';\nexport {split} from './split_shared';\nexport {tile} from './tile_impl';\nexport {topkImpl} from './topk_impl';\nexport {whereImpl} from './where_impl';\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const absGradConfig: GradConfig = {\n  kernelName: Abs,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, step(cast(x, 'float32'), -1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acosGradConfig: GradConfig = {\n  kernelName: Acos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = square(cast(x, 'float32'));\n        const b = sqrt(sub(scalar(1), a));\n        return neg(div(dy, b));\n      }\n\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acoshGradConfig: GradConfig = {\n  kernelName: Acosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(sub(square(cast(x, 'float32')), 1));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Add} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const addGradConfig: GradConfig = {\n  kernelName: Add,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AddN} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const addNGradConfig: GradConfig = {\n  kernelName: AddN,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const ders: {[key: string]: () => Tensor} = {};\n    saved.forEach((_, i) => {\n      ders[i] = () => dy.clone();\n    });\n    return ders;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMax} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMaxGradConfig: GradConfig = {\n  kernelName: ArgMax,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMinGradConfig: GradConfig = {\n  kernelName: ArgMin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const asinGradConfig: GradConfig = {\n  kernelName: Asin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, sqrt(sub(scalar(1), square(cast(x, 'float32')))))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const asinhGradConfig: GradConfig = {\n  kernelName: Asinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(add(scalar(1), square(cast(x, 'float32'))));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const atan2GradConfig: GradConfig = {\n  kernelName: Atan2,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const d = add(square(a), square(b));\n      let res = mul(dy, div(b, d));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      const d = add(square(a), square(b));\n      let res = neg(mul(dy, div(a, d)));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const atanGradConfig: GradConfig = {\n  kernelName: Atan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, add(square(cast(x, 'float32')), 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {scalar} from '../ops/scalar';\nimport {Tensor} from '../tensor';\n\nexport const atanhGradConfig: GradConfig = {\n  kernelName: Atanh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, sub(scalar(1), square(cast(x, 'float32'))))};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool3DBackprop, AvgPool3DBackpropAttrs, AvgPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d avg pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0. The dilation\n *     rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction avgPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in avgPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in avgPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.avgPool3dBackprop(dy5D, input5D, convInfo);\n  };\n\n  const inputs: AvgPool3DBackpropInputs = {dy: dy5D, input: input5D};\n\n  const attrs: AvgPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      AvgPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3dBackprop = op({avgPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool3D, AvgPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPool3dBackprop} from '../ops/avg_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const avgPool3DGradConfig: GradConfig = {\n  kernelName: AvgPool3D,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as AvgPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => avgPool3dBackprop(\n          dy as Tensor5D, x, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of an 2D avg pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The input image, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction avgPoolBackprop_<T extends Tensor3D|Tensor4D>(\n    dy: T|TensorLike, input: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy (${\n          $dy.rank})`);\n\n  let input4D = $input as Tensor4D;\n  let dy4D = $dy as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($input.rank === 3) {\n    reshapedTo4D = true;\n    input4D =\n        reshape($input, [1, $input.shape[0], $input.shape[1], $input.shape[2]]);\n    dy4D = reshape($dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2]]);\n  }\n\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in avgPoolBackprop: dy must be rank 4 but got rank ` +\n          `${dy4D.rank}.`);\n  util.assert(\n      input4D.rank === 4,\n      () => `Error in avgPoolBackprop: input must be rank 4 but got rank ` +\n          `${input4D.rank}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        input4D.shape, filterSize, strides, 1 /* dilations */, pad);\n\n    return backend.avgPoolBackprop(dy4D, input4D, convInfo);\n  };\n\n  const inputs: AvgPoolBackpropInputs = {dy: dy4D, input: input4D};\n\n  const attrs: AvgPoolBackpropAttrs = {filterSize, strides, pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, AvgPoolBackprop,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const avgPoolBackprop = op({avgPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool, AvgPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPoolBackprop} from '../ops/avg_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const avgPoolGradConfig: GradConfig = {\n  kernelName: AvgPool,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as AvgPoolAttrs;\n    return {\n      x: () => avgPoolBackprop(dy as Tensor4D, x, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchMatMul, BatchMatMulAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {matMul} from '../ops/mat_mul';\nimport {Tensor, Tensor3D} from '../tensor';\n\nexport const batchMatMulGradConfig: GradConfig = {\n  kernelName: BatchMatMul,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor3D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [a, b] = saved as Tensor3D[];\n\n    const {transposeA, transposeB} = attrs as {} as BatchMatMulAttrs;\n\n    if (!transposeA && !transposeB) {\n      return {\n        a: () => matMul(dy, b, false, true),\n        b: () => matMul(a, dy, true, false)\n      };\n    } else if (!transposeA && transposeB) {\n      return {\n        a: () => matMul(dy, b, false, false),\n        b: () => matMul(dy, a, true, false)\n      };\n    } else if (transposeA && !transposeB) {\n      return {\n        a: () => matMul(b, dy, false, true),\n        b: () => matMul(a, dy, false, false)\n      };\n    } else {\n      return {\n        a: () => matMul(b, dy, true, true),\n        b: () => matMul(dy, a, true, true)\n      };\n    }\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchToSpaceND, BatchToSpaceNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {spaceToBatchND} from '../ops/space_to_batch_nd';\nimport {Tensor} from '../tensor';\n\nexport const batchToSpaceNDGradConfig: GradConfig = {\n  kernelName: BatchToSpaceND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, crops} = attrs as {} as BatchToSpaceNDAttrs;\n    return {x: () => spaceToBatchND(dy, blockShape, crops)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BroadcastTo, BroadCastToAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const broadcastToGradConfig: GradConfig = {\n  kernelName: BroadcastTo,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const broadCastToAttrs: BroadCastToAttrs =\n        attrs as unknown as BroadCastToAttrs;\n\n    const inputShape = broadCastToAttrs.inputShape;\n    const outputShape = broadCastToAttrs.shape;\n\n    const reps: number[] = Array.from(outputShape);\n    for (let i = inputShape.length - 1; i >= 0; i--) {\n      if (inputShape[i] === outputShape[i]) {\n        reps[i] = 1;\n      } else if (inputShape[i] !== 1) {\n        throw new Error(`broadcastTo(): [${\n            inputShape}] cannot be broadcast to [${outputShape}].`);\n      }\n    }\n    const axes: number[] = [];\n    for (let i = 0; i < reps.length; i++) {\n      if (reps[i] > 1) {\n        axes.push(i);\n      }\n    }\n\n    return {x: () => sum(dy, axes, true /* keepDims */)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cast} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const castGradConfig: GradConfig = {\n  kernelName: Cast,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => dy.clone()};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const ceilGradConfig: GradConfig = {\n  kernelName: Ceil,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {lessEqual} from '../ops/less_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const clipByValueGradConfig: GradConfig = {\n  kernelName: ClipByValue,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {clipValueMin, clipValueMax} = attrs as {} as ClipByValueAttrs;\n    return {\n      x: () => where(\n          logicalAnd(greaterEqual(x, clipValueMin), lessEqual(x, clipValueMax)),\n          dy, zerosLike(dy)),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Concat, ConcatAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {split} from '../ops/split';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const concatGradConfig: GradConfig = {\n  kernelName: Concat,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const shapes = saved.map(t => t.shape);\n    const {axis} = attrs as {} as ConcatAttrs;\n    const $axis = parseAxisParam(axis, saved[0].shape)[0];\n    const sizeSplits = shapes.map(s => s[$axis]);\n    const derTensors = split(dy, sizeSplits, $axis);\n    return derTensors.map(t => () => t) as {};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2D, Conv2DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../ops/conv2d_backprop_input';\nimport * as conv_util from '../ops/conv_util';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv2DGradConfig: GradConfig = {\n  kernelName: Conv2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x4D, $filter] = saved as [Tensor4D, Tensor4D];\n    const {dilations, strides, pad, dataFormat} = attrs as {} as Conv2DAttrs;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of conv2D: dilation rates greater than 1 ' +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    return {\n      x: () =>\n          conv2DBackpropInput(x4D.shape, dy, $filter, strides, pad, dataFormat),\n      filter: () =>\n          conv2DBackpropFilter(x4D, dy, $filter.shape, strides, pad, dataFormat)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2d} from '../ops/conv2d';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const conv2DBackpropInputGradConfig: GradConfig = {\n  kernelName: Conv2DBackpropInput,\n  inputsToSave: ['dy', 'filter'],\n  gradFunc: (ddx: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [dy, filter] = saved as [Tensor4D, Tensor4D];\n\n    const {strides, pad, dataFormat, dimRoundingMode} =\n        attrs as {} as Conv2DBackpropInputAttrs;\n\n    return {\n      dy: () => conv2d(\n          ddx, filter, strides, pad, dataFormat, 1 /* dilations */,\n          dimRoundingMode),\n      filter: () => conv2DBackpropFilter(\n          ddx, dy, filter.shape, strides, pad, dataFormat, dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropFilterAttrs, Conv3DBackpropFilterInputs, Conv3DBackpropFilterV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 3D convolution.\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     [batch, depth, height, width, inChannels]. If rank 4, batch of 1 is\n *     assumed.\n * @param dy The dy image, of rank 5 or rank 4, of shape\n *     [batch, depth, height, width, outDepth]. If rank 4, batch of 1 is\n *     assumed.\n * @param filterShape The shape of the filter, length 5,\n *     [filterDepth, filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideDepth, strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction conv3DBackpropFilter_<T extends Tensor4D|Tensor5D>(\n    x: T, dy: T, filterShape: [number, number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): Tensor5D {\n  let x5D = x as Tensor5D;\n  if (x.rank === 4) {\n    x5D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2], x.shape[3]]);\n  }\n  let dy5D = dy as Tensor5D;\n  if (dy5D.rank === 4) {\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3dDerFilter: input must be rank 5, but got shape ` +\n          `${x5D.shape}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerFilter: dy must be rank 5, but got shape ` +\n          `${dy5D.shape}.`);\n  util.assert(\n      filterShape.length === 5,\n      () => `Error in conv3dDerFilter: filterShape must be length 5, but got ` +\n          `${filterShape}.`);\n  util.assert(\n      x5D.shape[4] === filterShape[3],\n      () => `Error in conv3dDerFilter: depth of input ${x5D.shape[4]}) must ` +\n          `match input depth in filter (${filterShape[3]}.`);\n  util.assert(\n      dy5D.shape[4] === filterShape[4],\n      () => `Error in conv3dDerFilter: depth of dy (${dy5D.shape[4]}) must ` +\n          `match output depth for filter (${filterShape[4]}).`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, filterShape, strides, dilations, pad);\n\n    return backend.conv3dDerFilter(x5D, dy5D, convInfo);\n  };\n\n  const inputs: Conv3DBackpropFilterInputs = {x: x5D, y: dy5D};\n\n  const attrs: Conv3DBackpropFilterAttrs = {strides, pad};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv3DBackpropFilterV2, attrs as {} as NamedAttrMap) as Tensor5D;\n}\n\nexport const conv3DBackpropFilter = op({conv3DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv3D, Conv3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv3DBackpropFilter} from '../ops/conv3d_backprop_filter';\nimport {conv3DBackpropInput} from '../ops/conv3d_backprop_input';\nimport {tupleValuesAreOne} from '../ops/conv_util';\nimport {Tensor, Tensor5D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv3DGradConfig: GradConfig = {\n  kernelName: Conv3D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor5D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad} = attrs as {} as Conv3DAttrs;\n    util.assert(\n        tupleValuesAreOne(dilations),\n        () =>\n            'Error in gradient of conv3D: dilation rates greater than 1 are ' +\n            `not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const [x5D, $filter] = saved;\n\n    return {\n      x: () => conv3DBackpropInput(\n          (x5D as Tensor5D).shape, dy, $filter as Tensor5D, strides, pad),\n      filter: () => conv3DBackpropFilter(\n          x5D as Tensor5D, dy, ($filter as Tensor5D).shape, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {sin} from '../ops/sin';\nimport {Tensor} from '../tensor';\n\nexport const cosGradConfig: GradConfig = {\n  kernelName: Cos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(neg(sin(cast(x, 'float32'))), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {sinh} from '../ops/sinh';\nimport {Tensor} from '../tensor';\n\nexport const coshGradConfig: GradConfig = {\n  kernelName: Cosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(sinh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cumsum, CumsumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getAxesPermutation} from '../ops/axis_util';\nimport {cumsum} from '../ops/cumsum';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const cumsumGradConfig: GradConfig = {\n  kernelName: Cumsum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {axis, exclusive, reverse}: CumsumAttrs = attrs as {} as CumsumAttrs;\n\n    return {\n      x: () => {\n        const permutation = getAxesPermutation([axis], x.rank);\n\n        let out = cumsum(dy, axis, exclusive, !reverse);\n\n        if (permutation != null) {\n          out = transpose(out, permutation);\n        }\n\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as conv_util from '../ops/conv_util';\nimport {depthwiseConv2dNativeBackpropFilter} from '../ops/depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../ops/depthwise_conv2d_native_backprop_input';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const depthwiseConv2dNativeGradConfig: GradConfig = {\n  kernelName: DepthwiseConv2dNative,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad, dimRoundingMode} =\n        attrs as {} as DepthwiseConv2dNativeAttrs;\n\n    const $dilations = dilations == null ? [1, 1] : dilations;\n\n    util.assert(\n        conv_util.tupleValuesAreOne($dilations),\n        () => 'Error in gradient of depthwiseConv2dNative: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${$dilations}'`);\n\n    const [x, filter] = saved as [Tensor4D, Tensor4D];\n\n    util.assert(\n        x.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: input must be ` +\n            `rank 4, but got rank ${x.rank}.`);\n    util.assert(\n        filter.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: filter must be ` +\n            `rank 4, but got rank ${filter.rank}.`);\n    util.assert(\n        x.shape[3] === filter.shape[2],\n        () => `Error in gradient of depthwiseConv2d: number of input ` +\n            `channels (${x.shape[3]}) must match the inChannels dimension ` +\n            `in filter ${filter.shape[2]}.`);\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, $dilations),\n        () => 'Error in gradient of depthwiseConv2d: Either strides or ' +\n            `dilations must be  1. Got strides ${strides} and dilations ` +\n            `'${$dilations}'.`);\n\n    if (dimRoundingMode != null) {\n      util.assert(\n          util.isInt(pad as number),\n          () =>\n              `Error in depthwiseConv2d: pad must be an integer when using, ` +\n              `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n    }\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x.shape, filter.shape, strides, $dilations as number | [number, number],\n        pad, dimRoundingMode, true /* depthwise */);\n\n    return {\n      x: () =>\n          depthwiseConv2dNativeBackpropInput(x.shape, dy, filter, convInfo),\n      filter: () =>\n          depthwiseConv2dNativeBackpropFilter(x, dy, filter.shape, convInfo),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DBackpropFilter, Dilation2DBackpropFilterInputs, Dilation2DBackpropInput, Dilation2DBackpropInputInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const dilation2dGradConfig: GradConfig = {\n  kernelName: Dilation2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, filter] = saved as [Tensor4D, Tensor3D];\n\n    const inputInputs: Dilation2DBackpropInputInputs = {x, filter, dy};\n    const filterInputs: Dilation2DBackpropFilterInputs = {x, filter, dy};\n\n    return {\n      x: () => ENGINE.runKernel(\n                   Dilation2DBackpropInput, inputInputs as {} as NamedTensorMap,\n                   attrs) as Tensor,\n      filter: () => ENGINE.runKernel(\n                        Dilation2DBackpropFilter,\n                        filterInputs as {} as NamedTensorMap, attrs) as Tensor\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const divGradConfig: GradConfig = {\n  kernelName: Div,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluGrad, EluGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const eluGradConfig: GradConfig = {\n  kernelName: Elu,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      return backend.eluDer(dy, y);\n    };\n\n    const inputs: EluGradInputs = {dy, y};\n\n    return {\n      x: () => ENGINE.runKernelFunc(\n          backPropKernelFunc, inputs as {} as NamedTensorMap, null /* grad */,\n          EluGrad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Erf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const erfGradConfig: GradConfig = {\n  kernelName: Erf,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const a = mul(exp(neg(square(x))), 2 / Math.sqrt(Math.PI));\n    return {x: () => mul(dy, a)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expGradConfig: GradConfig = {\n  kernelName: Exp,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n    return {x: () => mul(dy, y)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expm1GradConfig: GradConfig = {\n  kernelName: Expm1,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, exp(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const floorGradConfig: GradConfig = {\n  kernelName: Floor,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FloorDiv} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const floorDivGradConfig: GradConfig = {\n  kernelName: FloorDiv,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {FusedBatchNorm, FusedBatchNormAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {rsqrt} from '../ops/rsqrt';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {tile} from '../ops/tile';\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nexport const fusedBatchNormGradConfig: GradConfig = {\n  kernelName: FusedBatchNorm,\n  inputsToSave: ['x', 'mean', 'variance', 'scale'],\n  gradFunc: <R extends Rank>(\n      dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {varianceEpsilon} = attrs as {} as FusedBatchNormAttrs;\n    const [x, mean, variance, scale] = saved;\n\n    const scaleValue = scale == null ? scalar(1) : scale;\n    const reductionAxes = getReductionAxes(mean.shape, x.shape);\n    const tileShape: number[] = [];\n    if (mean.rank === 1) {\n      for (let i = 0; i < x.shape.length - 1; ++i) {\n        tileShape.push(x.shape[i]);\n      }\n      tileShape.push(1);\n    }\n\n    const xMinusMean = sub(x, mean);\n    const dyTimesScaleValue = mul(dy, scaleValue);\n    const oneOverSqrtVariance = rsqrt(add(variance, scalar(varianceEpsilon)));\n    const minusHalfRCube = mul(\n        mul(mul(oneOverSqrtVariance, oneOverSqrtVariance), oneOverSqrtVariance),\n        scalar(-0.5));\n\n    const derX = () => {\n      if (mean.rank === 1) {\n        return reshape(\n            mul(mul(dy,\n                    tile(\n                        reshape(oneOverSqrtVariance, [1, 1, 1, mean.shape[0]]),\n                        tileShape)),\n                scaleValue),\n            x.shape);\n      } else {\n        return reshape(mul(mul(dy, oneOverSqrtVariance), scaleValue), x.shape);\n      }\n    };\n    const derMean = () => {\n      let meanDer =\n          mul(mul(oneOverSqrtVariance, scalar(-1)), dyTimesScaleValue);\n      if (mean.rank === 1) {\n        meanDer = sum(meanDer, reductionAxes);\n      }\n      return reshape(meanDer, mean.shape as ShapeMap[R]);\n    };\n    const derVariance = () => {\n      let varianceDer = mul(mul(minusHalfRCube, xMinusMean), dyTimesScaleValue);\n\n      if (mean.rank === 1) {\n        varianceDer = sum(varianceDer, reductionAxes);\n      }\n      return reshape(varianceDer, mean.shape as ShapeMap[R]);\n    };\n    const derScale = () => {\n      const xMinusMean2TimesRsqrt = mul(xMinusMean, oneOverSqrtVariance);\n\n      let scaleDer = mul(dy, xMinusMean2TimesRsqrt);\n      if (mean.rank === 1) {\n        scaleDer = sum(scaleDer, reductionAxes);\n      }\n      return reshape(scaleDer, mean.shape as ShapeMap[R]);\n    };\n    const derOffset = () => {\n      let offsetDer = dy;\n      if (mean.rank === 1) {\n        offsetDer = sum(offsetDer, reductionAxes);\n      }\n      return reshape(offsetDer, mean.shape as ShapeMap[R]);\n    };\n\n    return {\n      x: derX,\n      mean: derMean,\n      variance: derVariance,\n      scale: derScale,\n      offset: derOffset\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GatherV2, GatherV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getUndoAxesPermutation} from '../ops/axis_util';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {unsortedSegmentSum} from '../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const gatherGradConfig: GradConfig = {\n  kernelName: GatherV2,\n  inputsToSave: ['x', 'indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, indices] = saved;\n    const {axis} = attrs as {} as GatherV2Attrs;\n\n    const parsedAxis = parseAxisParam(axis, x.shape)[0];\n\n    const derX = () => {\n      const paramsShape = x.shape;\n      const indicesSize = indices.size;\n\n      const outerShape = paramsShape.slice(0, parsedAxis);\n      const outerDims = outerShape.length;\n      const innerShape = paramsShape.slice(axis, paramsShape.length).slice(1);\n      const innerDims = innerShape.length;\n\n      const outerAxesIndices = arrayRange(0, outerDims);\n      const innerAxesIndices =\n          arrayRange(outerDims + 1, outerDims + 1 + innerDims);\n\n      const valuesShape = arrayConcat([outerShape, [indicesSize], innerShape]);\n\n      const values = reshape(dy, valuesShape);\n      const reshapedIndices = reshape(indices, [indicesSize]);\n\n      const transposeDims =\n          arrayConcat([[outerDims], outerAxesIndices, innerAxesIndices]);\n      const valuesTranspose = transpose(values, transposeDims);\n      let paramsGrad = unsortedSegmentSum(\n          valuesTranspose, reshapedIndices as Tensor1D, x.shape[parsedAxis]);\n\n      const invertTransposeDims = getUndoAxesPermutation(transposeDims);\n      paramsGrad = transpose(paramsGrad, invertTransposeDims);\n\n      return paramsGrad;\n    };\n    return {x: derX, indices: () => indices};\n  }\n};\n\nfunction arrayRange(start: number, stop: number): number[] {\n  const result = [];\n  for (let i = start; i < stop; ++i) {\n    result.push(i);\n  }\n  return result;\n}\n\nfunction arrayConcat(arrays: number[][]): number[] {\n  const result = [];\n  for (let i = 0; i < arrays.length; ++i) {\n    for (let j = 0; j < arrays[i].length; ++j) {\n      result.push(arrays[i][j]);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GreaterEqual} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const greaterEqualGradConfig: GradConfig = {\n  kernelName: GreaterEqual,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    return {a: () => zerosLike(a), b: () => zerosLike(b)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {Tensor} from '../tensor';\n\nexport const identityGradConfig: GradConfig = {\n  kernelName: Identity,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => cast(dy, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isFiniteGradConfig: GradConfig = {\n  kernelName: IsFinite,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isInfGradConfig: GradConfig = {\n  kernelName: IsInf,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isNanGradConfig: GradConfig = {\n  kernelName: IsNan,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log1p} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const log1pGradConfig: GradConfig = {\n  kernelName: Log1p,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, add(x, 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const logGradConfig: GradConfig = {\n  kernelName: Log,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, cast(x, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {LogSoftmax, LogSoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const logSoftmaxGradConfig: GradConfig = {\n  kernelName: LogSoftmax,\n  inputsToSave: [],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [value] = saved;\n    const {axis} = attrs as {} as LogSoftmaxAttrs;\n    return {\n      logits: () => {\n        const keepDims = true;\n        const softmax = exp(value);\n        return sub(dy, mul(sum(dy, axis, keepDims), softmax));\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRNBackprop, LRNBackpropAttrs, LRNBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport {op} from './operation';\n\nfunction localResponseNormalizationBackprop_<T extends Tensor4D>(\n    x: T, y: T, dy: T, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.LRNGrad(dy, x, y, depthRadius, bias, alpha, beta);\n\n  const inputs: LRNBackpropInputs = {x, y, dy};\n\n  const attrs: LRNBackpropAttrs = {depthRadius, bias, alpha, beta};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LRNBackprop, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const localResponseNormalizationBackprop =\n    op({localResponseNormalizationBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {LRN, LRNAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {localResponseNormalizationBackprop} from '../ops/local_response_normalization_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const lrnGradConfig: GradConfig = {\n  kernelName: LRN,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {depthRadius, bias, alpha, beta} = attrs as {} as LRNAttrs;\n\n    return {\n      x: () => localResponseNormalizationBackprop(\n          x, y, dy, depthRadius, bias, alpha, beta)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as axis_util from '../ops/axis_util';\nimport {cast} from '../ops/cast';\nimport {equal} from '../ops/equal';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\n/**\n * Gradient helper function for the min and max operations.\n */\nexport function gradForMinAndMax<T extends Tensor>(\n    dy: T, y: T, xOrig: Tensor, origAxes: number[], permutedAxes: number[]) {\n  if (y.rank < xOrig.rank) {\n    y = reshape(y, axis_util.expandShapeToKeepDim(y.shape, origAxes)) as T;\n  }\n  if (dy.rank < xOrig.rank) {\n    dy = reshape(dy, axis_util.expandShapeToKeepDim(dy.shape, origAxes)) as T;\n  }\n  return {\n    x: () => {\n      const dx = mul(dy, cast(equal(xOrig, y), dy.dtype));\n      return permutedAxes == null ? dx : transpose(dx, permutedAxes);\n    }\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const maxGradConfig: GradConfig = {\n  kernelName: Max,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const maxAttrs: MaxAttrs = attrs as {} as MaxAttrs;\n    const {reductionIndices} = maxAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const maxGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = maxGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Maximum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {less} from '../ops/less';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const maximumGradConfig: GradConfig = {\n  kernelName: Maximum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(greaterEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(less(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool3DBackprop, MaxPool3DBackpropAttrs, MaxPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d max pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank 4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param output The original output image, of rank 5 of shape\n *     [batchSize, outDepth, outHeight, outWidth, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike, output: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'maxPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPool3dBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let output5D = $output as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n    output5D = reshape($output, [\n      1, $output.shape[0], $output.shape[1], $output.shape[2], $output.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in maxPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in maxPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n  util.assert(\n      output5D.rank === 5,\n      () => `Error in maxPool3dBackprop: output must be rank 5 but got rank ` +\n          `${output5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.maxPool3dBackprop(dy5D, input5D, output5D, convInfo);\n  };\n\n  const inputs:\n      MaxPool3DBackpropInputs = {dy: dy5D, input: input5D, output: output5D};\n\n  const attrs: MaxPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      MaxPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3dBackprop = op({maxPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool3D, MaxPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPool3dBackprop} from '../ops/max_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const maxPool3DGradConfig: GradConfig = {\n  kernelName: MaxPool3D,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor5D, Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as MaxPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => maxPool3dBackprop(\n          dy as Tensor5D, x, y, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\n\n/**\n * Computes the backprop of a 2D max pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The original input image, of rank 4, of shape\n *     [batchSize, height, width, channels].\n * @param output The original output image, of rank 4, of shape\n *     [batchSize, outHeight, outWidth, channels].\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPoolBackprop_(\n    dy: Tensor4D|TensorLike, input: Tensor4D|TensorLike,\n    output: Tensor4D|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  const $dy = convertToTensor(dy, 'dy', 'maxPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPoolBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy ` +\n          `(${$dy.rank})`);\n\n  util.assert(\n      $dy.rank === 4,\n      () => `Error in maxPoolBackprop: dy must be rank 4 but got rank ` +\n          `${$dy.rank}.`);\n  util.assert(\n      $input.rank === 4,\n      () => `Error in maxPoolBackprop: input must be rank 4 but got rank ` +\n          `${$input.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPoolBackprop: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        $input.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    return backend.maxPoolBackprop($dy, $input, $output, convInfo);\n  };\n\n  const inputs:\n      MaxPoolBackpropInputs = {dy: $dy, input: $input, output: $output};\n\n  const attrs:\n      MaxPoolBackpropAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, MaxPoolBackprop,\n             attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const maxPoolBackprop = op({maxPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool, MaxPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPoolBackprop} from '../ops/max_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const maxPoolGradConfig: GradConfig = {\n  kernelName: MaxPool,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as MaxPoolAttrs;\n\n    return {\n      x: () => maxPoolBackprop(dy as Tensor4D, x, y, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Min, MinAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const minGradConfig: GradConfig = {\n  kernelName: Min,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const minAttrs: MinAttrs = attrs as {} as MinAttrs;\n    const {axis} = minAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(axis, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const minGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = minGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Minimum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const minimumGradConfig: GradConfig = {\n  kernelName: Minimum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(lessEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(greater(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Mod} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {floor} from '../ops/floor';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const modGradConfig: GradConfig = {\n  kernelName: Mod,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(dy, reduceAxes), a.shape);\n      }\n      return dy;\n    };\n    const derB = () => {\n      const res = mul(dy, neg(floor(div(a, b))));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Multiply} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const multiplyGradConfig: GradConfig = {\n  kernelName: Multiply,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = mul(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      const res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Negate} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {neg} from '../ops/neg';\nimport {Tensor} from '../tensor';\n\nexport const negateGradConfig: GradConfig = {\n  kernelName: Negate,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => neg(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OneHot} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\n\nexport const oneHotGradConfig: GradConfig = {\n  kernelName: OneHot,\n  inputsToSave: ['indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const indices = saved[0];\n    return {indices: () => zeros(indices.shape, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OnesLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const onesLikeGradConfig: GradConfig = {\n  kernelName: OnesLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PadV2, PadV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\nexport const padV2GradConfig: GradConfig = {\n  kernelName: PadV2,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    // Pad introduces values around the original tensor, so the gradient\n    // slices the original shape out of the gradient.\n    const x = saved[0];\n    const {paddings} = attrs as unknown as PadV2Attrs;\n    const begin = paddings.map(p => p[0]);\n    return {x: () => slice(dy, begin, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Pow} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {log} from '../ops/log';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {reshape} from '../ops/reshape';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const powGradConfig: GradConfig = {\n  kernelName: Pow,\n  inputsToSave: ['a', 'b'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b, y] = saved;\n    const base = a;\n    const exp = b;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(base.shape, exp.shape);\n\n    const derBase = () => {\n      const expFloat = cast(exp, 'float32');\n      let res = mul(dy, mul(expFloat, pow(base, sub(expFloat, scalar(1)))));\n      const reduceAxes = broadcast_util.getReductionAxes(base.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, base.shape);\n    };\n    const derExp = () => {\n      const condition = greater(base, 0);\n      const logBase = where(condition, log(base), zerosLike(base));\n      let res = mul(dy, mul(y, logBase));\n      const reduceAxes = broadcast_util.getReductionAxes(exp.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, exp.shape);\n    };\n    return {a: derBase, b: derExp};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Prelu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const preluGradConfig: GradConfig = {\n  kernelName: Prelu,\n  inputsToSave: ['x', 'alpha'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x, alpha] = saved;\n    const mask = greater(x, 0);\n\n    return {\n      x: () => where(mask, dy, mul(dy, alpha)),\n      alpha: () => {\n        let res = where(mask, zerosLike(dy), mul(dy, x));\n        const reduceAxes = getReductionAxes(alpha.shape, dy.shape);\n        if (reduceAxes.length > 0) {\n          res = sum(res, reduceAxes);\n        }\n        return reshape(res, alpha.shape);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Reciprocal} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const reciprocalGradConfig: GradConfig = {\n  kernelName: Reciprocal,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, neg(square(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu6} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const relu6GradConfig: GradConfig = {\n  kernelName: Relu6,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const mask = mul(lessEqual(x, 6), step(x));\n\n    return {x: () => mul(dy, cast(mask, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const reluGradConfig: GradConfig = {\n  kernelName: Relu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, cast(step(x), 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reshape} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\n\nexport const reshapeGradConfig: GradConfig = {\n  kernelName: Reshape,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => reshape(dy, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearGrad, ResizeBilinearGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeBilinearGradConfig: GradConfig = {\n  kernelName: ResizeBilinear,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeBilinearAttrs;\n      return backend.resizeBilinearBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeBilinearGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeBilinearGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborGrad, ResizeNearestNeighborGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeNearestNeighborGradConfig: GradConfig = {\n  kernelName: ResizeNearestNeighbor,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeNearestNeighborAttrs;\n      return backend.resizeNearestNeighborBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeNearestNeighborGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeNearestNeighborGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reverse, ReverseAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {reverse} from '../ops/reverse';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const reverseGradConfig: GradConfig = {\n  kernelName: Reverse,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dims} = attrs as {} as ReverseAttrs;\n    const axes = parseAxisParam(dims, dy.shape);\n    return {x: () => reverse(dy, axes)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Round} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const roundGradConfig: GradConfig = {\n  kernelName: Round,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rsqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {pow} from '../ops/pow';\nimport {Tensor} from '../tensor';\n\nexport const rsqrtGradConfig: GradConfig = {\n  kernelName: Rsqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => neg(div(dy, mul(pow(x, 1.5), 2)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SelectV2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {logicalNot} from '../ops/logical_not';\nimport {mul} from '../ops/mul';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const selectV2PoolGradConfig: GradConfig = {\n  kernelName: SelectV2,\n  inputsToSave: ['condition'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [condition] = saved;\n    return {\n      // TODO(julianoks): Return null for condition gradient\n      // when backprop supports it.\n      condition: () => cast(zerosLike(condition), 'float32'),\n      t: () => mul(dy, cast(condition, dy.dtype)),\n      e: () => mul(dy, cast(logicalNot(condition), dy.dtype))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Selu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {exp} from '../ops/exp';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {SELU_SCALE, SELU_SCALEALPHA} from '../ops/selu_util';\nimport {where} from '../ops/where';\nimport {Tensor} from '../tensor';\n\nexport const seluGradConfig: GradConfig = {\n  kernelName: Selu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {\n      x: () => {\n        const mask = greater(x, scalar(0));\n\n        const scaleAlpha = scalar(SELU_SCALEALPHA);\n        const scale = scalar(SELU_SCALE);\n\n        const greaterThanZeroDer = mul(dy, scale);\n        const lessEqualZeroDer =\n            mul(mul(dy, scaleAlpha), exp(cast(x, 'float32')));\n\n        return where(mask, greaterThanZeroDer, lessEqualZeroDer);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sigmoid} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const sigmoidGradConfig: GradConfig = {\n  kernelName: Sigmoid,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(dy, mul(y, sub(scalar(1), y)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sign} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const signGradConfig: GradConfig = {\n  kernelName: Sign,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cos} from '../ops/cos';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinGradConfig: GradConfig = {\n  kernelName: Sin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(cos(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cosh} from '../ops/cosh';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinhGradConfig: GradConfig = {\n  kernelName: Sinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(cosh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Slice, SliceAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {pad} from '../ops/pad';\nimport {parseSliceParams} from '../ops/slice_util';\nimport {Tensor} from '../tensor';\n\nexport const sliceGradConfig: GradConfig = {\n  kernelName: Slice,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {begin, size} = attrs as {} as SliceAttrs;\n\n    const inputShape = x.shape;\n    const [begin_, size_] = parseSliceParams(x, begin, size);\n\n    // Create an Nx2 padding where the first column represents how many\n    // zeros are prepended (at start) for each dimension, and the second\n    // column indicates how many zeros are appended (at end).\n\n    // The number of zeros to append is the shape of the input\n    // elementwise-subtracted by both the begin vector and sizes vector.\n    const paddings: Array<[number, number]> = [];\n    for (let i = 0; i < dy.rank; i++) {\n      paddings.push([begin_[i], inputShape[i] - begin_[i] - size_[i]]);\n    }\n    return {x: () => pad(dy, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softmax, SoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const softmaxGradConfig: GradConfig = {\n  kernelName: Softmax,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [y] = saved;\n    const {dim} = attrs as {} as SoftmaxAttrs;\n    const keepDims = true;\n\n    const dyTimesY = mul(dy, y);\n    return {\n      logits: () => sub(dyTimesY, mul(sum(dyTimesY, [dim], keepDims), y))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softplus} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sigmoid} from '../ops/sigmoid';\nimport {Tensor} from '../tensor';\n\nexport const softplusGradConfig: GradConfig = {\n  kernelName: Softplus,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(dy, sigmoid(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SpaceToBatchND, SpaceToBatchNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {batchToSpaceND} from '../ops/batch_to_space_nd';\nimport {Tensor} from '../tensor';\n\nexport const spaceToBatchNDGradConfig: GradConfig = {\n  kernelName: SpaceToBatchND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, paddings} = attrs as {} as SpaceToBatchNDAttrs;\n    return {x: () => batchToSpaceND(dy, blockShape, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {SplitV, SplitVAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {concat} from '../ops/concat';\nimport {Tensor} from '../tensor';\n\nexport const splitVGradConfig: GradConfig = {\n  kernelName: SplitV,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const {axis} = attrs as {} as SplitVAttrs;\n\n    return {x: () => concat(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {Tensor} from '../tensor';\n\nexport const sqrtGradConfig: GradConfig = {\n  kernelName: Sqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, mul(sqrt(cast(x, 'float32')), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const squareGradConfig: GradConfig = {\n  kernelName: Square,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, mul(cast(x, 'float32'), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SquaredDifference} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const squaredDifferenceGradConfig: GradConfig = {\n  kernelName: SquaredDifference,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const two = scalar(2);\n    const derA = () => mul(dy, mul(two, sub(a, b)));\n    const derB = () => mul(dy, mul(two, sub(b, a)));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Step} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const stepGradConfig: GradConfig = {\n  kernelName: Step,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports\n    // it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Sub} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const subGradConfig: GradConfig = {\n  kernelName: Sub,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(neg(res), b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sum, SumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {ones} from '../ops/ones';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const sumGradConfig: GradConfig = {\n  kernelName: Sum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const expandedDyShape = x.shape.slice();\n    const {axis} = attrs as {} as SumAttrs;\n\n    const axes = parseAxisParam(axis, x.shape);\n    axes.forEach(axis => {\n      expandedDyShape[axis] = 1;\n    });\n    const expandedDy = reshape(dy, expandedDyShape);\n    const derX = mul(expandedDy, ones(x.shape, 'float32'));\n\n    return {x: () => derX};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cos} from '../ops/cos';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const tanGradConfig: GradConfig = {\n  kernelName: Tan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, square(cos(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const tanhGradConfig: GradConfig = {\n  kernelName: Tanh,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(sub(scalar(1), square(y)), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tile, TileAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {slice} from '../ops/slice';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const tileGradConfig: GradConfig = {\n  kernelName: Tile,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {reps} = attrs as unknown as TileAttrs;\n\n    const derX = () => {\n      let xGrad = zerosLike(x);\n      // TODO(cais): Maybe reduce memory footprint by avoiding repeated\n      // slicing.\n      if (x.rank === 1) {\n        for (let i = 0; i < reps[0]; ++i) {\n          xGrad = add(xGrad, slice(dy, [i * x.shape[0]], [x.shape[0]]));\n        }\n      } else if (x.rank === 2) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            xGrad = add(xGrad, slice(dy, [i * x.shape[0], j * x.shape[1]], [\n                          x.shape[0], x.shape[1]\n                        ]));\n          }\n        }\n      } else if (x.rank === 3) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              xGrad =\n                  add(xGrad,\n                      slice(\n                          dy, [i * x.shape[0], j * x.shape[1], k * x.shape[2]],\n                          [x.shape[0], x.shape[1], x.shape[2]]));\n            }\n          }\n        }\n      } else if (x.rank === 4) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              for (let l = 0; l < reps[3]; ++l) {\n                xGrad =\n                    add(xGrad,\n                        slice(\n                            dy,\n                            [\n                              i * x.shape[0], j * x.shape[1], k * x.shape[2],\n                              l * x.shape[3]\n                            ],\n                            [x.shape[0], x.shape[1], x.shape[2], x.shape[3]]));\n              }\n            }\n          }\n        }\n      } else {\n        throw new Error(\n            `Gradient for tile operation is not implemented for rank-` +\n            `${x.rank} tensors yet.`);\n      }\n      return xGrad;\n    };\n    return {x: derX};\n  },\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Transpose, TransposeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const transposeGradConfig: GradConfig = {\n  kernelName: Transpose,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const transposeAttrs: TransposeAttrs = attrs as {} as TransposeAttrs;\n    const {perm} = transposeAttrs;\n    const undoPerm = axis_util.getUndoAxesPermutation(perm);\n    return {x: () => transpose(dy, undoPerm)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Unpack, UnpackAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {stack} from '../ops/stack';\nimport {Tensor} from '../tensor';\n\nexport const unpackGradConfig: GradConfig = {\n  kernelName: Unpack,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const unpackAttrs: UnpackAttrs = attrs as {} as UnpackAttrs;\n    const {axis} = unpackAttrs;\n    return {value: () => stack(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {UnsortedSegmentSum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {expandDims} from '../ops/expand_dims';\nimport {gather} from '../ops/gather';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {maximum} from '../ops/maximum';\nimport {ones} from '../ops/ones';\nimport {scalar} from '../ops/scalar';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor, Tensor1D} from '../tensor';\n\nexport const unsortedSegmentSumGradConfig: GradConfig = {\n  kernelName: UnsortedSegmentSum,\n  inputsToSave: ['segmentIds'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [segmentIds] = saved;\n\n    const derX = () => {\n      return gatherDropNegatives(dy, segmentIds as Tensor1D);\n    };\n    return {x: derX};\n  }\n};\n\nfunction gatherDropNegatives<T extends Tensor>(x: T, indices: Tensor1D) {\n  // Helper function for unsorted segment ops. Gathers params for\n  // positive segment ids and gathers 0 for inputs with negative segment id.\n  // Mirrors _GatherDropNegatives from tensorflow/python/ops/math_grad.py\n  const zeroClippedIndices = maximum(indices, zerosLike(indices));\n  const gathered = gather(x, zeroClippedIndices as Tensor1D);\n  let isPositive = greaterEqual(indices, scalar(0, 'int32'));\n  const numIters = gathered.rank - isPositive.rank;\n  for (let i = 0; i < numIters; ++i) {\n    isPositive = expandDims(isPositive, i + 1);\n  }\n  isPositive = logicalAnd(isPositive, ones(gathered.shape, 'bool'));\n  const zeroSlice = zerosLike(gathered);\n  return where(isPositive, gathered, zeroSlice);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ZerosLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const zerosLikeGradConfig: GradConfig = {\n  kernelName: ZerosLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {absGradConfig} from './gradients/Abs_grad';\nimport {acosGradConfig} from './gradients/Acos_grad';\nimport {acoshGradConfig} from './gradients/Acosh_grad';\nimport {addGradConfig} from './gradients/Add_grad';\nimport {addNGradConfig} from './gradients/AddN_grad';\nimport {argMaxGradConfig} from './gradients/ArgMax_grad';\nimport {argMinGradConfig} from './gradients/ArgMin_grad';\nimport {asinGradConfig} from './gradients/Asin_grad';\nimport {asinhGradConfig} from './gradients/Asinh_grad';\nimport {atan2GradConfig} from './gradients/Atan2_grad';\nimport {atanGradConfig} from './gradients/Atan_grad';\nimport {atanhGradConfig} from './gradients/Atanh_grad';\nimport {avgPool3DGradConfig} from './gradients/AvgPool3D_grad';\nimport {avgPoolGradConfig} from './gradients/AvgPool_grad';\nimport {batchMatMulGradConfig} from './gradients/BatchMatMul_grad';\nimport {batchToSpaceNDGradConfig} from './gradients/BatchToSpaceND_grad';\nimport {broadcastToGradConfig} from './gradients/BroadcastTo_grad';\nimport {castGradConfig} from './gradients/Cast_grad';\nimport {ceilGradConfig} from './gradients/Ceil_grad';\nimport {clipByValueGradConfig} from './gradients/ClipByValue_grad';\nimport {concatGradConfig} from './gradients/Concat_grad';\nimport {conv2DGradConfig} from './gradients/Conv2D_grad';\nimport {conv2DBackpropInputGradConfig} from './gradients/Conv2DBackpropInput_grad';\nimport {conv3DGradConfig} from './gradients/Conv3D_grad';\nimport {cosGradConfig} from './gradients/Cos_grad';\nimport {coshGradConfig} from './gradients/Cosh_grad';\nimport {cumsumGradConfig} from './gradients/Cumsum_grad';\nimport {depthwiseConv2dNativeGradConfig} from './gradients/DepthwiseConv2dNative_grad';\nimport {dilation2dGradConfig} from './gradients/Dilation2D_grad';\nimport {divGradConfig} from './gradients/Div_grad';\nimport {eluGradConfig} from './gradients/Elu_grad';\nimport {erfGradConfig} from './gradients/Erf_grad';\nimport {expGradConfig} from './gradients/Exp_grad';\nimport {expm1GradConfig} from './gradients/Expm1_grad';\nimport {floorGradConfig} from './gradients/Floor_grad';\nimport {floorDivGradConfig} from './gradients/FloorDiv_grad';\nimport {fusedBatchNormGradConfig} from './gradients/FusedBatchNorm_grad';\nimport {gatherGradConfig} from './gradients/GatherV2_grad';\nimport {greaterEqualGradConfig} from './gradients/GreaterEqual_grad';\nimport {identityGradConfig} from './gradients/Identity_grad';\nimport {isFiniteGradConfig} from './gradients/IsFinite_grad';\nimport {isInfGradConfig} from './gradients/IsInf_grad';\nimport {isNanGradConfig} from './gradients/IsNan_grad';\nimport {log1pGradConfig} from './gradients/Log1p_grad';\nimport {logGradConfig} from './gradients/Log_grad';\nimport {logSoftmaxGradConfig} from './gradients/LogSoftmax_grad';\nimport {lrnGradConfig} from './gradients/LRN_grad';\nimport {maxGradConfig} from './gradients/Max_grad';\nimport {maximumGradConfig} from './gradients/Maximum_grad';\nimport {maxPool3DGradConfig} from './gradients/MaxPool3D_grad';\nimport {maxPoolGradConfig} from './gradients/MaxPool_grad';\nimport {minGradConfig} from './gradients/Min_grad';\nimport {minimumGradConfig} from './gradients/Minimum_grad';\nimport {modGradConfig} from './gradients/Mod_grad';\nimport {multiplyGradConfig} from './gradients/Multiply_grad';\nimport {negateGradConfig} from './gradients/Negate_grad';\nimport {oneHotGradConfig} from './gradients/OneHot_grad';\nimport {onesLikeGradConfig} from './gradients/OnesLike_grad';\nimport {padV2GradConfig} from './gradients/PadV2_grad';\nimport {powGradConfig} from './gradients/Pow_grad';\nimport {preluGradConfig} from './gradients/Prelu_grad';\nimport {reciprocalGradConfig} from './gradients/Reciprocal_grad';\nimport {relu6GradConfig} from './gradients/Relu6_grad';\nimport {reluGradConfig} from './gradients/Relu_grad';\nimport {reshapeGradConfig} from './gradients/Reshape_grad';\nimport {resizeBilinearGradConfig} from './gradients/ResizeBilinear_grad';\nimport {resizeNearestNeighborGradConfig} from './gradients/ResizeNearestNeighbor_grad';\nimport {reverseGradConfig} from './gradients/Reverse_grad';\nimport {roundGradConfig} from './gradients/Round_grad';\nimport {rsqrtGradConfig} from './gradients/Rsqrt_grad';\nimport {selectV2PoolGradConfig} from './gradients/SelectV2_grad';\nimport {seluGradConfig} from './gradients/Selu_grad';\nimport {sigmoidGradConfig} from './gradients/Sigmoid_grad';\nimport {signGradConfig} from './gradients/Sign_grad';\nimport {sinGradConfig} from './gradients/Sin_grad';\nimport {sinhGradConfig} from './gradients/Sinh_grad';\nimport {sliceGradConfig} from './gradients/Slice_grad';\nimport {softmaxGradConfig} from './gradients/Softmax_grad';\nimport {softplusGradConfig} from './gradients/Softplus_grad';\nimport {spaceToBatchNDGradConfig} from './gradients/SpaceToBatchND_grad';\nimport {splitVGradConfig} from './gradients/SplitV_grad';\nimport {sqrtGradConfig} from './gradients/Sqrt_grad';\nimport {squareGradConfig} from './gradients/Square_grad';\nimport {squaredDifferenceGradConfig} from './gradients/SquaredDifference_grad';\nimport {stepGradConfig} from './gradients/Step_grad';\nimport {subGradConfig} from './gradients/Sub_grad';\nimport {sumGradConfig} from './gradients/Sum_grad';\nimport {tanGradConfig} from './gradients/Tan_grad';\nimport {tanhGradConfig} from './gradients/Tanh_grad';\nimport {tileGradConfig} from './gradients/Tile_grad';\nimport {transposeGradConfig} from './gradients/Transpose_grad';\nimport {unpackGradConfig} from './gradients/Unpack_grad';\nimport {unsortedSegmentSumGradConfig} from './gradients/UnsortedSegmentSum_grad';\nimport {zerosLikeGradConfig} from './gradients/ZerosLike_grad';\nimport {GradConfig} from './kernel_registry';\nimport {registerGradient} from './kernel_registry';\n\n// Export all kernel configs here so that the package can auto register them\nconst gradConfigs: GradConfig[] = [\n  absGradConfig,\n  acosGradConfig,\n  acoshGradConfig,\n  addGradConfig,\n  addNGradConfig,\n  argMaxGradConfig,\n  argMinGradConfig,\n  asinGradConfig,\n  asinhGradConfig,\n  atan2GradConfig,\n  atanGradConfig,\n  atanhGradConfig,\n  avgPool3DGradConfig,\n  avgPoolGradConfig,\n  batchMatMulGradConfig,\n  batchToSpaceNDGradConfig,\n  broadcastToGradConfig,\n  castGradConfig,\n  ceilGradConfig,\n  clipByValueGradConfig,\n  concatGradConfig,\n  conv2DBackpropInputGradConfig,\n  conv2DGradConfig,\n  conv3DGradConfig,\n  cosGradConfig,\n  coshGradConfig,\n  cumsumGradConfig,\n  depthwiseConv2dNativeGradConfig,\n  dilation2dGradConfig,\n  divGradConfig,\n  eluGradConfig,\n  erfGradConfig,\n  expGradConfig,\n  expm1GradConfig,\n  floorDivGradConfig,\n  floorGradConfig,\n  fusedBatchNormGradConfig,\n  gatherGradConfig,\n  greaterEqualGradConfig,\n  identityGradConfig,\n  isFiniteGradConfig,\n  isInfGradConfig,\n  isNanGradConfig,\n  log1pGradConfig,\n  logGradConfig,\n  logSoftmaxGradConfig,\n  lrnGradConfig,\n  maxGradConfig,\n  maxGradConfig,\n  maximumGradConfig,\n  maxPool3DGradConfig,\n  maxPoolGradConfig,\n  minGradConfig,\n  minimumGradConfig,\n  modGradConfig,\n  multiplyGradConfig,\n  negateGradConfig,\n  oneHotGradConfig,\n  onesLikeGradConfig,\n  padV2GradConfig,\n  padV2GradConfig,\n  powGradConfig,\n  preluGradConfig,\n  reciprocalGradConfig,\n  relu6GradConfig,\n  reluGradConfig,\n  reshapeGradConfig,\n  resizeBilinearGradConfig,\n  resizeNearestNeighborGradConfig,\n  reverseGradConfig,\n  roundGradConfig,\n  rsqrtGradConfig,\n  selectV2PoolGradConfig,\n  seluGradConfig,\n  sigmoidGradConfig,\n  signGradConfig,\n  sinGradConfig,\n  sinhGradConfig,\n  sliceGradConfig,\n  softmaxGradConfig,\n  softplusGradConfig,\n  spaceToBatchNDGradConfig,\n  spaceToBatchNDGradConfig,\n  splitVGradConfig,\n  splitVGradConfig,\n  sqrtGradConfig,\n  squaredDifferenceGradConfig,\n  squareGradConfig,\n  stepGradConfig,\n  subGradConfig,\n  sumGradConfig,\n  tanGradConfig,\n  tanhGradConfig,\n  tileGradConfig,\n  transposeGradConfig,\n  unpackGradConfig,\n  unsortedSegmentSumGradConfig,\n  zerosLikeGradConfig\n];\n\nfor (const gradientConfig of gradConfigs) {\n  registerGradient(gradientConfig);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {abs} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    abs<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.abs = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return abs(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acos = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acosh = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {addStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    addStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.addStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return addStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {add} from '../../ops/add';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    add<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.add = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return add(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {all} from '../../ops/all';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    all<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.all = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return all(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {any} from '../../ops/any';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    any<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.any = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return any(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMax} from '../../ops/arg_max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMax<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMax = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return argMax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMin} from '../../ops/arg_min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMin<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMin = function<T extends Tensor>(axis: number): T {\n  this.throwIfDisposed();\n  return argMin(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Scalar, Tensor} from '../../tensor';\nimport {Rank} from '../../types';\nimport {assert} from '../../util';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asScalar<T extends Tensor>(): Scalar;\n  }\n}\n\n/** Converts a size-1 `tf.Tensor` to a `tf.Scalar`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asScalar = function<T extends Tensor>(this: T): Scalar {\n  this.throwIfDisposed();\n  assert(this.size === 1, () => 'The array must have only 1 element.');\n  return reshape(this, []);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asType<T extends Tensor>(this: T, dtype: DataType): T;\n  }\n}\n\n/**\n * Casts a `tf.Tensor` to a specified dtype.\n *\n * @param dtype Data-type to cast the tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asType = function<T extends Tensor>(\n    this: T, dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast<T>(this, dtype);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as1D<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Converts a `tf.Tensor` to a `tf.Tensor1D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as1D = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as2D<T extends Tensor>(rows: number, columns: number): Tensor2D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor2D`.\n *\n * @param rows Number of rows in `tf.Tensor2D`.\n * @param columns Number of columns in `tf.Tensor2D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as2D = function<T extends Tensor>(\n    rows: number, columns: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as3D<T extends Tensor>(rows: number, columns: number, depth: number):\n        Tensor3D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor3D`.\n *\n * @param rows Number of rows in `tf.Tensor3D`.\n * @param columns Number of columns in `tf.Tensor3D`.\n * @param depth Depth of `tf.Tensor3D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as3D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as4D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number): Tensor4D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor4D`.\n *\n * @param rows Number of rows in `tf.Tensor4D`.\n * @param columns Number of columns in `tf.Tensor4D`.\n * @param depth Depth of `tf.Tensor4D`.\n * @param depth2 4th dimension of `tf.Tensor4D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as4D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as5D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number,\n        depth3: number): Tensor5D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor5D`.\n *\n * @param rows Number of rows in `tf.Tensor5D`.\n * @param columns Number of columns in `tf.Tensor5D`.\n * @param depth Depth of `tf.Tensor5D`.\n * @param depth2 4th dimension of `tf.Tensor5D`.\n * @param depth3 5th dimension of 'tf.Tensor5D'\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as5D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number,\n    depth3: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2, depth3]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {atan2} from '../../ops/atan2';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan2<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.atan2 = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return atan2(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {avgPool} from '../../ops/avg_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    avgPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.avgPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return avgPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchToSpaceND} from '../../ops/batch_to_space_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchToSpaceND<R extends Rank>(blockShape: number[], crops: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.batchToSpaceND = function<R extends Rank>(\n    blockShape: number[], crops: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return batchToSpaceND(this, blockShape, crops);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchNorm} from '../../ops/batchnorm';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchNorm<T extends Tensor>(\n        mean: Tensor<R>|Tensor1D|TensorLike,\n        variance: Tensor<R>|Tensor1D|TensorLike,\n        offset?: Tensor<R>|Tensor1D|TensorLike,\n        scale?: Tensor<R>|Tensor1D|TensorLike,\n        varianceEpsilon?: number): Tensor<R>;\n  }\n}\n\nTensor.prototype.batchNorm = function<R extends Rank>(\n    mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  this.throwIfDisposed();\n  return batchNorm(this, mean, variance, offset, scale, varianceEpsilon);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {broadcastTo} from '../../ops/broadcast_to';\nimport {Tensor} from '../../tensor';\nimport {Rank, ShapeMap} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    broadcastTo<R extends Rank>(shape: ShapeMap[R]): Tensor<R>;\n  }\n}\n\nTensor.prototype.broadcastTo = function<R extends Rank>(shape: ShapeMap[R]):\n    Tensor<R> {\n  this.throwIfDisposed();\n  return broadcastTo(this, shape);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cast<T extends Tensor>(dtype: DataType): T;\n  }\n}\n\nTensor.prototype.cast = function<T extends Tensor>(dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast(this, dtype) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ceil} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ceil<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.ceil = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return ceil(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {clipByValue} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    clipByValue<T extends Tensor>(min: number, max: number): Tensor;\n  }\n}\n\nTensor.prototype.clipByValue = function<T extends Tensor>(\n    min: number, max: number): T {\n  this.throwIfDisposed();\n  return clipByValue(this, min, max) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {concat} from '../../ops/concat';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    concat<T extends Tensor>(tensors: T|Array<T|TensorLike>, axis?: number): T;\n  }\n}\n\nTensor.prototype.concat = function<T extends Tensor>(\n    x: T|Array<T|TensorLike>, axis?: number): T {\n  this.throwIfDisposed();\n  if (x instanceof Tensor) {\n    x = [x];\n  }\n  return concat([this, ...x], axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv1d} from '../../ops/conv1d';\nimport {ExplicitPadding} from '../../ops/conv_util';\nimport {Tensor, Tensor2D, Tensor3D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv1d<T extends Tensor2D|Tensor3D>(\n        filter: Tensor3D|TensorLike3D, stride: number,\n        pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n        dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv1d = function<T extends Tensor2D|Tensor3D>(\n    filter: Tensor3D|TensorLike3D, stride: number,\n    pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n    dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv1d(\n             this, filter, stride, pad, dataFormat, dilation,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2dTranspose} from '../../ops/conv2d_transpose';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2dTranspose<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D,\n        outputShape: [number, number, number, number]|[number, number, number],\n        strides: [number, number]|number, pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2dTranspose = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2dTranspose(\n             this, filter, outputShape, strides, pad, dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2d} from '../../ops/conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cos = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cosh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {cumsum} from '../../ops/cumsum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cumsum<R extends Rank>(\n        axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R>;\n  }\n}\n\nTensor.prototype.cumsum = function<R extends Rank>(\n    axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R> {\n  this.throwIfDisposed();\n  return cumsum(this, axis, exclusive, reverse);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthToSpace} from '../../ops/depth_to_space';\nimport {Tensor, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthToSpace<T extends Tensor4D>(\n        blockSize: number, dataFormat: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.depthToSpace = function<T extends Tensor4D>(\n    blockSize: number, dataFormat: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return depthToSpace(this, blockSize, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../../globals';\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2D<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\n/**\n * @deprecated Use `depthwiseConv2d` instead.\n */\nTensor.prototype.depthwiseConv2D = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  deprecationWarn('depthwiseConv2D is deprecated, use depthwiseConv2d instead');\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.depthwiseConv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dilation2d} from '../../ops/dilation2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dilation2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n        pad: 'valid'|'same', dilations?: [number, number]|number,\n        dataFormat?: 'NHWC'): T;\n  }\n}\n\nTensor.prototype.dilation2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n    pad: 'valid'|'same', dilations?: [number, number]|number,\n    dataFormat?: 'NHWC'): T {\n  this.throwIfDisposed();\n  return dilation2d(this, filter, strides, pad, dilations, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {divNoNan} from '../../ops/div_no_nan';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divNoNan<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.divNoNan = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return divNoNan(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {divStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.divStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return divStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {div} from '../../ops/div';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    div<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.div = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return div(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dot} from '../../ops/dot';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dot<T extends Tensor>(b: Tensor|TensorLike): Tensor;\n  }\n}\n\nTensor.prototype.dot = function<T extends Tensor>(b: T|TensorLike): Tensor {\n  this.throwIfDisposed();\n  return dot(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {elu} from '../../ops/elu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    elu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.elu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return elu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {equalStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equalStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.equalStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return equalStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {equal} from '../../ops/equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equal<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.equal = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return equal(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {erf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    erf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.erf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return erf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {exp} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    exp<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.exp = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return exp(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {expandDims} from '../../ops/expand_dims';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expandDims<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.expandDims = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return expandDims(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {expm1} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expm1<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.expm1 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return expm1(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {fft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    fft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.fft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return fft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    flatten<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Flatten a Tensor to a 1D array.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.flatten = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {floor} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floor<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.floor = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return floor(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {floorDiv} from '../../ops/floorDiv';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floorDiv<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.floorDiv = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return floorDiv(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {gather} from '../../ops/gather';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    gather<T extends Tensor>(\n        this: T, indices: Tensor|TensorLike, axis?: number): T;\n  }\n}\n\nTensor.prototype.gather = function<T extends Tensor>(\n    this: T, indices: Tensor|TensorLike, axis?: number): T {\n  this.throwIfDisposed();\n  return gather(this, indices, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return greaterEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greaterEqual} from '../../ops/greater_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greaterEqual = function<T extends Tensor>(b: Tensor|\n                                                           TensorLike): T {\n  this.throwIfDisposed();\n  return greaterEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return greaterStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greater} from '../../ops/greater';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greater<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greater = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return greater(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ifft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ifft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.ifft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return ifft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {irfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    irfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.irfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return irfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isFinite} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isFinite<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isFinite = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isFinite(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isInf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isInf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isInf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isInf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isNaN} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isNaN<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isNaN = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isNaN(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {leakyRelu} from '../../ops/leaky_relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    leakyRelu<T extends Tensor>(alpha: number): T;\n  }\n}\n\nTensor.prototype.leakyRelu = function<T extends Tensor>(\n    this: T, alpha: number): T {\n  this.throwIfDisposed();\n  return leakyRelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.lessEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {lessEqual} from '../../ops/less_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessEqual = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return lessEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {less} from '../../ops/less';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    less<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.less = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return less(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {localResponseNormalization} from '../../ops/local_response_normalization';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    localResponseNormalization<T extends Tensor>(\n        depthRadius?: number, bias?: number, alpha?: number, beta?: number): T;\n  }\n}\n\nTensor.prototype.localResponseNormalization = function<T extends Tensor>(\n    depthRadius?: number, bias?: number, alpha?: number, beta?: number): T {\n  this.throwIfDisposed();\n  return localResponseNormalization(this, depthRadius, bias, alpha, beta) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.logSigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return logSigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSoftmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSoftmax<T extends Tensor>(this: T, axis?: number): T;\n  }\n}\n\nTensor.prototype.logSoftmax = function<T extends Tensor>(\n    this: T, axis?: number): T {\n  this.throwIfDisposed();\n  return logSoftmax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logSumExp} from '../../ops/log_sum_exp';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSumExp<T extends Tensor>(\n        this: T, axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.logSumExp = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return logSumExp(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log1p} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log1p<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log1p = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log1p(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalAnd} from '../../ops/logical_and';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalAnd<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalAnd = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalAnd(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalNot} from '../../ops/logical_not';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalNot<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.logicalNot = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return logicalNot(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalOr} from '../../ops/logical_or';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalOr<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalOr = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return logicalOr(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalXor} from '../../ops/logical_xor';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalXor<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalXor = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalXor(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {matMul} from '../../ops/mat_mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    matMul<T extends Tensor>(\n        b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T;\n  }\n}\n\nTensor.prototype.matMul = function<T extends Tensor>(\n    this: T, b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T {\n  this.throwIfDisposed();\n  return matMul(this, b, transposeA, transposeB);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maxPool} from '../../ops/max_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maxPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.maxPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return maxPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {max} from '../../ops/max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    max<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.max = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return max(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {maximumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.maximumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return maximumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maximum} from '../../ops/maximum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.maximum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return maximum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mean} from '../../ops/mean';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mean<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.mean = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return mean(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {min} from '../../ops/min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    min<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.min = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return min(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {minimumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.minimumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return minimumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {minimum} from '../../ops/minimum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.minimum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return minimum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {modStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    modStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.modStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return modStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mod} from '../../ops/mod';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mod<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mod = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mod(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {mulStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mulStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.mulStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return mulStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mul} from '../../ops/mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mul<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mul = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mul(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {neg} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    neg<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.neg = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return neg(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {norm} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    norm<T extends Tensor>(\n        ord?: number|'euclidean'|'fro', axis?: number|number[],\n        keepDims?: boolean): Tensor;\n  }\n}\n\nTensor.prototype.norm = function<T extends Tensor>(\n    ord?: number|'euclidean'|'fro', axis?: number|number[],\n    keepDims?: boolean) {\n  this.throwIfDisposed();\n  return norm(this, ord, axis, keepDims) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {notEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.notEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return notEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {notEqual} from '../../ops/not_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.notEqual = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return notEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {oneHot} from '../../ops/one_hot';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    oneHot(depth: number, onValue: number, offValue: number): Tensor;\n  }\n}\n\nTensor.prototype.oneHot = function(\n    depth: number, onValue = 1, offValue = 0): Tensor {\n  this.throwIfDisposed();\n  return oneHot(this, depth, onValue, offValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {onesLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    onesLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.onesLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return onesLike(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {pad} from '../../ops/pad';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pad<T extends Tensor>(\n        paddings: Array<[number, number]>, constantValue?: number): T;\n  }\n}\n\nTensor.prototype.pad = function<T extends Tensor>(\n    this: T, paddings: Array<[number, number]>, constantValue: number): T {\n  this.throwIfDisposed();\n  return pad(this, paddings, constantValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pool} from '../../ops/pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pool<T extends Tensor3D|Tensor4D>(\n        windowShape: [number, number]|number, poolingType: 'avg'|'max',\n        padding: 'valid'|'same'|number, diationRate?: [number, number]|number,\n        strides?: [number, number]|number): T;\n  }\n}\n\nTensor.prototype.pool = function<T extends Tensor3D|Tensor4D>(\n    this: T, windowShape: [number, number]|number, poolingType: 'max'|'avg',\n    padding: 'valid'|'same'|number, dilationRate?: [number, number]|number,\n    strides?: [number, number]|number): T {\n  this.throwIfDisposed();\n  return pool(this, windowShape, poolingType, padding, dilationRate, strides);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {powStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    powStrict<T extends Tensor>(exp: Tensor): Tensor;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.powStrict = function<T extends Tensor>(exp: Tensor): T {\n  this.throwIfDisposed();\n  return powStrict(this, exp) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pow} from '../../ops/pow';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pow<T extends Tensor>(exp: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.pow = function<T extends Tensor>(exp: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return pow(this, exp);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prelu} from '../../ops/prelu';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prelu<T extends Tensor>(alpha: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.prelu = function<T extends Tensor>(\n    this: T, alpha: T|TensorLike): T {\n  this.throwIfDisposed();\n  return prelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prod} from '../../ops/prod';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prod<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.prod = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return prod(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {reciprocal} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reciprocal<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.reciprocal = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return reciprocal(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu} from '../../ops/relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu6} from '../../ops/relu6';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu6<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu6 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu6(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshapeAs<T extends Tensor>(x: T): T;\n  }\n}\n\n/**\n * Reshapes the tensor into the shape of the provided tensor.\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.reshapeAs = function<T extends Tensor>(x: T): T {\n  this.throwIfDisposed();\n  return reshape(this, x.shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshape<T extends Tensor>(shape: number[]): T;\n  }\n}\n\nTensor.prototype.reshape = function<T extends Tensor>(shape: number[]): T {\n  this.throwIfDisposed();\n  return reshape(this, shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeBilinear} from '../../ops/image/resize_bilinear';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeBilinear<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeBilinear = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeBilinear(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeNearestNeighbor} from '../../ops/image/resize_nearest_neighbor';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeNearestNeighbor<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeNearestNeighbor = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeNearestNeighbor(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reverse} from '../../ops/reverse';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reverse<T extends Tensor>(this: T, axis?: number|number[]): T;\n  }\n}\n\nTensor.prototype.reverse = function<T extends Tensor>(\n    this: T, axis?: number|number[]): T {\n  this.throwIfDisposed();\n  return reverse(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.rfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return rfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {round} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    round<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.round = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return round(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rsqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rsqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.rsqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return rsqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {selu} from '../../ops/selu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    selu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.selu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return selu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {separableConv2d} from '../../ops/separable_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    separableConv2d<T extends Tensor3D|Tensor4D>(\n        depthwiseFilter: Tensor4D|TensorLike4D,\n        pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n        pad: 'valid'|'same', dilation?: [number, number]|number,\n        dataFormat?: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.separableConv2d = function<T extends Tensor3D|Tensor4D>(\n    depthwiseFilter: Tensor4D|TensorLike4D,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation?: [number, number]|number,\n    dataFormat?: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return separableConv2d(\n             this, depthwiseFilter, pointwiseFilter, strides, pad, dilation,\n             dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sign} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sign<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sign = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sign(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {slice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    slice<T extends Tensor>(\n        this: T, begin: number|number[], size?: number|number[]): T;\n  }\n}\n\nTensor.prototype.slice = function<T extends Tensor>(\n    this: T, begin: number|number[], size?: number|number[]): T {\n  this.throwIfDisposed();\n  return slice(this, begin, size);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softmax<T extends Tensor>(this: T, dim?: number): T;\n  }\n}\n\nTensor.prototype.softmax = function<T extends Tensor>(this: T, dim: number): T {\n  this.throwIfDisposed();\n  return softmax(this, dim);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softplus} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softplus<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.softplus = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return softplus(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {spaceToBatchND} from '../../ops/space_to_batch_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    spaceToBatchND<R extends Rank>(blockShape: number[], paddings: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.spaceToBatchND = function<R extends Rank>(\n    blockShape: number[], paddings: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return spaceToBatchND(this, blockShape, paddings);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {split} from '../../ops/split';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    split<T extends Tensor>(numOrSizeSplits: number[]|number, axis?: number):\n        T[];\n  }\n}\n\nTensor.prototype.split = function<T extends Tensor>(\n    numOrSizeSplits: number[]|number, axis?: number): T[] {\n  this.throwIfDisposed();\n  return split(this, numOrSizeSplits, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {square} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    square<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.square = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return square(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {squaredDifference} from '../../ops/squared_difference';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifference<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.squaredDifference = function<T extends Tensor>(b: Tensor|\n                                                                TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifference(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {squaredDifferenceStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifferenceStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.squaredDifferenceStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifferenceStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {squeeze} from '../../ops/squeeze';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squeeze<T extends Tensor>(axis?: number[]): T;\n  }\n}\n\nTensor.prototype.squeeze = function<T extends Tensor>(axis?: number[]): T {\n  this.throwIfDisposed();\n  return squeeze(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {stack} from '../../ops/stack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stack<T extends Tensor>(x: Tensor|Tensor[], axis?: number): T;\n  }\n}\n\nTensor.prototype.stack = function<T extends Tensor>(\n    x: Tensor|Tensor[], axis?: number): T {\n  this.throwIfDisposed();\n  const tensorsToBeStacked = x instanceof Tensor ? [this, x] : [this, ...x];\n  return stack(tensorsToBeStacked, axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {step} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    step<T extends Tensor>(this: T, alpha?: number): T;\n  }\n}\n\nTensor.prototype.step = function<T extends Tensor>(this: T, alpha?: number) {\n  this.throwIfDisposed();\n  return step(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {stridedSlice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stridedSlice<T extends Tensor>(\n        this: Tensor, begin: number[], end: number[], strides: number[],\n        beginMask?: number, endMask?: number, ellipsisMask?: number,\n        newAxisMask?: number, shrinkAxisMask?: number): Tensor;\n  }\n}\n\nTensor.prototype.stridedSlice = function<T extends Tensor>(\n    this: Tensor, begin: number[], end: number[], strides: number[],\n    beginMask?: number, endMask?: number, ellipsisMask?: number,\n    newAxisMask?: number, shrinkAxisMask?: number): T {\n  this.throwIfDisposed();\n  return stridedSlice(\n             this, begin, end, strides, beginMask, endMask, ellipsisMask,\n             newAxisMask, shrinkAxisMask) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {subStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    subStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.subStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return subStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sub} from '../../ops/sub';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sub<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.sub = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return sub(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sum} from '../../ops/sum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sum<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.sum = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return sum(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tile} from '../../ops/tile';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tile<T extends Tensor>(b: number[]): T;\n  }\n}\n\nTensor.prototype.tile = function<T extends Tensor>(reps: number[]): T {\n  this.throwIfDisposed();\n  return tile(this, reps) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toBool<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `bool`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toBool = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'bool');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toFloat<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `float32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toFloat = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'float32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toInt<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `int32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toInt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'int32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {topk} from '../../ops/topk';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    topk<T extends Tensor>(this: T, k?: number, sorted?: boolean):\n        {values: T, indices: T};\n  }\n}\n\nTensor.prototype.topk = function<T extends Tensor>(\n    this: T, k?: number, sorted?: boolean): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return topk(this, k, sorted) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {transpose} from '../../ops/transpose';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    transpose<T extends Tensor>(perm?: number[]): T;\n  }\n}\n\nTensor.prototype.transpose = function<T extends Tensor>(\n    this: T, perm?: number[]): T {\n  this.throwIfDisposed();\n  return transpose(this, perm);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unique} from '../../ops/unique';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unique<T extends Tensor>(this: T, axis?: number): {values: T, indices: T};\n  }\n}\n\nTensor.prototype.unique = function<T extends Tensor>(\n    this: T, axis?: number): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return unique(this, axis) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unsortedSegmentSum} from '../../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike1D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unsortedSegmentSum<T extends Tensor>(\n        this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T;\n  }\n}\n\nTensor.prototype.unsortedSegmentSum = function<T extends Tensor>(\n    this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T {\n  this.throwIfDisposed();\n  return unsortedSegmentSum(this, segmentIds, numSegments);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {unstack} from '../../ops/unstack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unstack<T extends Tensor>(axis?: number): T[];\n  }\n}\n\nTensor.prototype.unstack = function<T extends Tensor>(axis?: number): T[] {\n  this.throwIfDisposed();\n  return unstack(this, axis) as T[];\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {where} from '../../ops/where';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    where<T extends Tensor>(condition: Tensor|TensorLike, x: Tensor|TensorLike):\n        T;\n  }\n}\n\nTensor.prototype.where = function<T extends Tensor>(\n    condition: Tensor|TensorLike, x: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return where(condition, this, x) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {zerosLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    zerosLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.zerosLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return zerosLike(this);\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {backend} from '@tensorflow/tfjs-core';\nimport {DataFormat} from '../keras_format/common';\n\nlet _epsilon: number;\n\n/**\n * Returns the value of the fuzz factor used in numeric expressions.\n */\nexport function epsilon() {\n  if (_epsilon == null) {\n    _epsilon = backend().epsilon();\n  }\n  return _epsilon;\n}\n\n/**\n * Sets the value of the fuzz factor used in numeric expressions.\n * @param e New value of epsilon.\n */\nexport function setEpsilon(e: number) {\n  _epsilon = e;\n}\n\n/**\n * Returns the default image data format convention.\n */\nexport function imageDataFormat(): DataFormat {\n  return 'channelsLast';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Explicit error types.\n *\n * See the following link for more information about why the code includes\n * calls to setPrototypeOf:\n *\n * https://github.com/Microsoft/TypeScript-wiki/blob/master/Breaking-Changes.md#extending-built-ins-like-error-array-and-map-may-no-longer-work\n */\n// tslint:enable\n\n/**\n * Equivalent of Python's AttributeError.\n */\nexport class AttributeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AttributeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's RuntimeError.\n */\nexport class RuntimeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, RuntimeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's ValueError.\n */\nexport class ValueError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, ValueError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's NotImplementedError.\n */\nexport class NotImplementedError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, NotImplementedError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's AssertionError.\n */\nexport class AssertionError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AssertionError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's IndexError.\n */\nexport class IndexError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, IndexError.prototype);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {DataType, fused, serialization, util} from '@tensorflow/tfjs-core';\n\nimport {AssertionError, ValueError} from '../errors';\n\n// tslint:enable\n\n/**\n * If `value` is an Array, equivalent to Python's `value * numValues`.\n * If `value` is not an Array, equivalent to Python's `[value] * numValues`\n */\n// tslint:disable-next-line:no-any\nexport function pyListRepeat(value: any, numValues: number): any[] {\n  if (Array.isArray(value)) {\n    // tslint:disable-next-line:no-any\n    let newArray: any[] = [];\n    for (let i = 0; i < numValues; i++) {\n      newArray = newArray.concat(value);\n    }\n    return newArray;\n  } else {\n    const newArray = new Array(numValues);\n    newArray.fill(value);\n    return newArray;\n  }\n}\n\nexport function assert(val: boolean, message?: string): void {\n  if (!val) {\n    throw new AssertionError(message);\n  }\n}\n\n/**\n * Count the number of elements of the `array` that are equal to `reference`.\n */\nexport function count<T>(array: T[], refernce: T) {\n  let counter = 0;\n  for (const item of array) {\n    if (item === refernce) {\n      counter++;\n    }\n  }\n  return counter;\n}\n\n/**\n * If an array is of length 1, just return the first element. Otherwise, return\n * the full array.\n * @param tensors\n */\nexport function singletonOrArray<T>(xs: T[]): T|T[] {\n  if (xs.length === 1) {\n    return xs[0];\n  }\n  return xs;\n}\n\n/**\n * Normalizes a list/tensor into a list.\n *\n * If a tensor is passed, we return\n * a list of size 1 containing the tensor.\n *\n * @param x target object to be normalized.\n */\n// tslint:disable-next-line:no-any\nexport function toList(x: any): any[] {\n  if (Array.isArray(x)) {\n    return x;\n  }\n  return [x];\n}\n\n/**\n * Generate a UID for a list\n */\n// tslint:disable-next-line:no-any\nexport function objectListUid(objs: any|any[]): string {\n  const objectList = toList(objs);\n  let retVal = '';\n  for (const obj of objectList) {\n    if (obj.id == null) {\n      throw new ValueError(\n          `Object ${obj} passed to objectListUid without an id`);\n    }\n    if (retVal !== '') {\n      retVal = retVal + ', ';\n    }\n    retVal = `${retVal}${Math.abs(obj.id)}`;\n  }\n  return retVal;\n}\n/**\n * Converts string to snake-case.\n * @param name\n */\nexport function toSnakeCase(name: string): string {\n  const intermediate = name.replace(/(.)([A-Z][a-z0-9]+)/g, '$1_$2');\n  const insecure =\n      intermediate.replace(/([a-z])([A-Z])/g, '$1_$2').toLowerCase();\n  /*\n   If the class is private the name starts with \"_\" which is not secure\n   for creating scopes. We prefix the name with \"private\" in this case.\n   */\n  if (insecure[0] !== '_') {\n    return insecure;\n  }\n  return 'private' + insecure;\n}\n\nexport function toCamelCase(identifier: string): string {\n  // quick return for empty string or single character strings\n  if (identifier.length <= 1) {\n    return identifier;\n  }\n  // Check for the underscore indicating snake_case\n  if (identifier.indexOf('_') === -1) {\n    return identifier;\n  }\n  return identifier.replace(/[_]+(\\w|$)/g, (m, p1) => p1.toUpperCase());\n}\n\n// tslint:disable-next-line:no-any\nlet _GLOBAL_CUSTOM_OBJECTS = {} as {[objName: string]: any};\n\nexport function serializeKerasObject(instance: serialization.Serializable):\n    serialization.ConfigDictValue {\n  if (instance === null || instance === undefined) {\n    return null;\n  }\n  const dict: serialization.ConfigDictValue = {};\n  dict['className'] = instance.getClassName();\n  dict['config'] = instance.getConfig();\n  return dict;\n}\n\n/**\n * Replace ndarray-style scalar objects in serialization objects with numbers.\n *\n * Background: In some versions of tf.keras, certain scalar values in the HDF5\n * model save file can be serialized as: `{'type': 'ndarray', 'value': num}`,\n * where in `num` is a plain number. This method converts such serialization\n * to a `number`.\n *\n * @param config The keras-format serialization object to be processed\n *   (in place).\n */\nfunction convertNDArrayScalarsInConfig(config: serialization.ConfigDictValue):\n    void {\n  if (config == null || typeof config !== 'object') {\n    return;\n  } else if (Array.isArray(config)) {\n    config.forEach(configItem => convertNDArrayScalarsInConfig(configItem));\n  } else {\n    const fields = Object.keys(config);\n    for (const field of fields) {\n      const value = config[field];\n      if (value != null && typeof value === 'object') {\n        if (!Array.isArray(value) && value['type'] === 'ndarray' &&\n            typeof value['value'] === 'number') {\n          config[field] = value['value'];\n        } else {\n          convertNDArrayScalarsInConfig(value as serialization.ConfigDict);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Deserialize a saved Keras Object\n * @param identifier either a string ID or a saved Keras dictionary\n * @param moduleObjects a list of Python class names to object constructors\n * @param customObjects a list of Python class names to object constructors\n * @param printableModuleName debug text for the object being reconstituted\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns a TensorFlow.js Layers object\n */\n// tslint:disable:no-any\nexport function deserializeKerasObject(\n    identifier: string|serialization.ConfigDict,\n    moduleObjects = {} as {[objName: string]: any},\n    customObjects = {} as {[objName: string]: any},\n    printableModuleName = 'object', fastWeightInit = false): any {\n  // tslint:enable\n  if (typeof identifier === 'string') {\n    const functionName = identifier;\n    let fn;\n    if (functionName in customObjects) {\n      fn = customObjects[functionName];\n    } else if (functionName in _GLOBAL_CUSTOM_OBJECTS) {\n      fn = _GLOBAL_CUSTOM_OBJECTS[functionName];\n    } else {\n      fn = moduleObjects[functionName];\n      if (fn == null) {\n        throw new ValueError(\n            `Unknown ${printableModuleName}: ${identifier}. ` +\n            `This may be due to one of the following reasons:\\n` +\n            `1. The ${printableModuleName} is defined in Python, in which ` +\n            `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n            `code.\\n` +\n            `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n            `but is not registered properly with ` +\n            `tf.serialization.registerClass().`);\n        // TODO(cais): Add link to tutorial page on custom layers.\n      }\n    }\n    return fn;\n  } else {\n    // In this case we are dealing with a Keras config dictionary.\n    const config = identifier;\n    if (config['className'] == null || config['config'] == null) {\n      throw new ValueError(\n          `${printableModuleName}: Improper config format: ` +\n          `${JSON.stringify(config)}.\\n` +\n          `'className' and 'config' must set.`);\n    }\n    const className = config['className'] as string;\n    let cls, fromConfig;\n    if (className in customObjects) {\n      [cls, fromConfig] = customObjects[className];\n    } else if (className in _GLOBAL_CUSTOM_OBJECTS) {\n      [cls, fromConfig] = _GLOBAL_CUSTOM_OBJECTS['className'];\n    } else if (className in moduleObjects) {\n      [cls, fromConfig] = moduleObjects[className];\n    }\n    if (cls == null) {\n      throw new ValueError(\n          `Unknown ${printableModuleName}: ${className}. ` +\n          `This may be due to one of the following reasons:\\n` +\n          `1. The ${printableModuleName} is defined in Python, in which ` +\n          `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n          `code.\\n` +\n          `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n          `but is not registered properly with ` +\n          `tf.serialization.registerClass().`);\n      // TODO(cais): Add link to tutorial page on custom layers.\n    }\n    if (fromConfig != null) {\n      // Porting notes: Instead of checking to see whether fromConfig accepts\n      // customObjects, we create a customObjects dictionary and tack it on to\n      // config['config'] as config['config'].customObjects. Objects can use it,\n      // if they want.\n\n      // tslint:disable-next-line:no-any\n      const customObjectsCombined = {} as {[objName: string]: any};\n      for (const key of Object.keys(_GLOBAL_CUSTOM_OBJECTS)) {\n        customObjectsCombined[key] = _GLOBAL_CUSTOM_OBJECTS[key];\n      }\n      for (const key of Object.keys(customObjects)) {\n        customObjectsCombined[key] = customObjects[key];\n      }\n      // Add the customObjects to config\n      const nestedConfig = config['config'] as serialization.ConfigDict;\n      nestedConfig['customObjects'] = customObjectsCombined;\n\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      convertNDArrayScalarsInConfig(config['config']);\n      const returnObj =\n          fromConfig(cls, config['config'], customObjects, fastWeightInit);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n\n      return returnObj;\n    } else {\n      // Then `cls` may be a function returning a class.\n      // In this case by convention `config` holds\n      // the kwargs of the function.\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      // In python this is **config['config'], for tfjs-layers we require\n      // classes that use this fall-through construction method to take\n      // a config interface that mimics the expansion of named parameters.\n      const returnObj = new cls(config['config']);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n      return returnObj;\n    }\n  }\n}\n\n/**\n * Compares two numbers for sorting.\n * @param a\n * @param b\n */\nexport function numberCompare(a: number, b: number) {\n  return (a < b) ? -1 : ((a > b) ? 1 : 0);\n}\n\n/**\n * Comparison of two numbers for reverse sorting.\n * @param a\n * @param b\n */\nexport function reverseNumberCompare(a: number, b: number) {\n  return -1 * numberCompare(a, b);\n}\n\n/**\n * Convert a string into the corresponding DType.\n * @param dtype\n * @returns An instance of DType.\n */\nexport function stringToDType(dtype: string): DataType {\n  switch (dtype) {\n    case 'float32':\n      return 'float32';\n    default:\n      throw new ValueError(`Invalid dtype: ${dtype}`);\n  }\n}\n\n/**\n * Test the element-by-element equality of two Arrays of strings.\n * @param xs First array of strings.\n * @param ys Second array of strings.\n * @returns Wether the two arrays are all equal, element by element.\n */\nexport function stringsEqual(xs: string[], ys: string[]): boolean {\n  if (xs == null || ys == null) {\n    return xs === ys;\n  }\n  if (xs.length !== ys.length) {\n    return false;\n  }\n  for (let i = 0; i < xs.length; ++i) {\n    if (xs[i] !== ys[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Get the unique elements of an array.\n * @param xs Array.\n * @returns An Array consisting of the unique elements in `xs`.\n */\nexport function unique<T>(xs: T[]): T[] {\n  if (xs == null) {\n    return xs;\n  }\n  const out: T[] = [];\n  // TODO(cais): Maybe improve performance by sorting.\n  for (const x of xs) {\n    if (out.indexOf(x) === -1) {\n      out.push(x);\n    }\n  }\n  return out;\n}\n\n/**\n * Determine if an Object is empty (i.e., does not have own properties).\n * @param obj Object\n * @returns Whether the Object is empty.\n * @throws ValueError: If object is `null` or `undefined`.\n */\nexport function isObjectEmpty(obj: {}): boolean {\n  if (obj == null) {\n    throw new ValueError(`Invalid value in obj: ${JSON.stringify(obj)}`);\n  }\n  for (const key in obj) {\n    if (obj.hasOwnProperty(key)) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Helper function used to build type union/enum run-time checkers.\n * @param values The list of allowed values.\n * @param label A string name for the type\n * @param value The value to test.\n * @throws ValueError: If the value is not in values nor `undefined`/`null`.\n */\nexport function checkStringTypeUnionValue(\n    values: string[], label: string, value: string): void {\n  if (value == null) {\n    return;\n  }\n  if (values.indexOf(value) < 0) {\n    throw new ValueError(`${value} is not a valid ${label}.  Valid values are ${\n        values} or null/undefined.`);\n  }\n}\n\n/**\n * Helper function for verifying the types of inputs.\n *\n * Ensures that the elements of `x` are all of type `expectedType`.\n * Also verifies that the length of `x` is within bounds.\n *\n * @param x Object to test.\n * @param expectedType The string expected type of all of the elements in the\n * Array.\n * @param minLength Return false if x.length is less than this.\n * @param maxLength Return false if x.length is greater than this.\n * @returns true if and only if `x` is an `Array<expectedType>` with\n * length >= `minLength` and <= `maxLength`.\n */\n// tslint:disable:no-any\nexport function checkArrayTypeAndLength(\n    x: any, expectedType: string, minLength = 0,\n    maxLength = Infinity): boolean {\n  assert(minLength >= 0);\n  assert(maxLength >= minLength);\n  return (\n      Array.isArray(x) && x.length >= minLength && x.length <= maxLength &&\n      x.every(e => typeof e === expectedType));\n}\n// tslint:enable:no-any\n\n/**\n * Assert that a value or an array of value are positive integer.\n *\n * @param value The value being asserted on. May be a single number or an array\n *   of numbers.\n * @param name Name of the value, used to make the error message.\n */\nexport function assertPositiveInteger(value: number|number[], name: string) {\n  if (Array.isArray(value)) {\n    util.assert(\n        value.length > 0, () => `${name} is unexpectedly an empty array.`);\n    value.forEach(\n        (v, i) => assertPositiveInteger(v, `element ${i + 1} of ${name}`));\n  } else {\n    util.assert(\n        Number.isInteger(value) && value > 0,\n        () => `Expected ${name} to be a positive integer, but got ` +\n            `${formatAsFriendlyString(value)}.`);\n  }\n}\n\n/**\n * Format a value into a display-friendly, human-readable fashion.\n *\n * - `null` is formatted as `'null'`\n * - Strings are formated with flanking pair of quotes.\n * - Arrays are formatted with flanking pair of square brackets.\n *\n * @param value The value to display.\n * @return Formatted string.\n */\n// tslint:disable-next-line:no-any\nexport function formatAsFriendlyString(value: any): string {\n  if (value === null) {\n    return 'null';\n  } else if (Array.isArray(value)) {\n    return '[' + value.map(v => formatAsFriendlyString(v)).join(',') + ']';\n  } else if (typeof value === 'string') {\n    return `\"${value}\"`;\n  } else {\n    return `${value}`;\n  }\n}\n\n/**\n * Returns a function `f2` (decorator) which wraps the original function\n * `f`. `f2` guarantees that `f` can be called at most once\n * every `waitMs` ms. If `f2` is called more often, it will return\n * the last returned result of `f`.\n *\n * @param f The original function `f` to wrap.\n * @param waitMs The time between two consecutive calls to `f` in ms.\n */\nexport function debounce<T>(\n    f: (...args: Array<{}>) => T, waitMs: number): (...args: Array<{}>) => T {\n  let lastTime = util.now();\n  let lastResult: T;\n  const f2 = (...args: Array<{}>) => {\n    const now = util.now();\n    if (now - lastTime < waitMs) {\n      return lastResult;\n    }\n    lastTime = now;\n    lastResult = f(...args);\n    return lastResult;\n  };\n  return f2;\n}\n\n/**\n * Returns the fusable activation given a layers identifier.\n *\n * @param activationName The layers identifier string.\n * @return The name of the fusable activation.\n */\nexport function mapActivationToFusedKernel(activationName: string):\n    fused.Activation {\n  if (activationName === 'relu') {\n    return 'relu';\n  }\n  if (activationName === 'linear') {\n    return 'linear';\n  }\n  if (activationName === 'elu') {\n    return 'elu';\n  }\n  return null;\n}\n\ntype PossibleValues = Array<Array<boolean|string|number>>;\n\n/**\n * Returns the cartesian product of sets of values.\n * This works the same as itertools.product in Python.\n *\n * Example:\n *\n * filters = [128, 256, 512]\n * paddings = ['same', 'valid']\n *\n * product = [ [128, 'same'], [128, 'valid'], [256, 'same'], [256, 'valid'],\n * [512, 'same'], [512, 'valid']]\n *\n * @param arrayOfValues List/array of values.\n * @return The cartesian product.\n */\nexport function getCartesianProductOfValues(...arrayOfValues: PossibleValues):\n    PossibleValues {\n  assert(arrayOfValues.length > 0, 'arrayOfValues is empty');\n\n  for (const values of arrayOfValues) {\n    assert(Array.isArray(values), 'one of the values is not an array');\n    assert(values.length > 0, 'one of the values is empty');\n  }\n\n  return arrayOfValues.reduce((products, values) => {\n    if (products.length === 0) {\n      return values.map(value => [value]);\n    }\n\n    return values\n        .map(value => {\n          return products.map((prevValue) => [...prevValue, value]);\n        })\n        .reduce((flattenedProduct, unflattenedProduct) => {\n          return flattenedProduct.concat(unflattenedProduct);\n        }, []);\n  }, [] as PossibleValues);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/contraints.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport {epsilon} from './backend/common';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\n/**\n * Helper function used by many of the Constraints to find the L2Norms.\n */\nfunction calcL2Norms(w: Tensor, axis: number): Tensor {\n  return tidy(() => tfc.sqrt(tfc.sum(tfc.mul(w, w), axis, true)));\n}\n\n/**\n * Base class for functions that impose constraints on weight values\n *\n * @doc {\n *   heading: 'Constraints',\n *   subheading: 'Classes',\n *   namespace: 'constraints'\n * }\n */\nexport abstract class Constraint extends serialization.Serializable {\n  /* Porting note: was __call__, apply chosen to match other similar choices */\n  abstract apply(w: Tensor): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport interface MaxNormArgs {\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   *\n   *  For instance, in a `Dense` layer the weight matrix\n   *  has shape `[inputDim, outputDim]`,\n   *  set `axis` to `0` to constrain each weight vector\n   *  of length `[inputDim,]`.\n   *  In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   *  the weight tensor has shape\n   *  `[rows, cols, inputDepth, outputDepth]`,\n   *  set `axis` to `[0, 1, 2]`\n   *  to constrain the weights of each filter tensor of size\n   *  `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class MaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MaxNorm';\n  private maxValue: number;\n  private axis: number;\n  private readonly defaultMaxValue = 2;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MaxNormArgs) {\n    super();\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.clipByValue(norms, 0, this.maxValue);\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {maxValue: this.maxValue, axis: this.axis};\n  }\n}\nserialization.registerClass(MaxNorm);\n\nexport interface UnitNormArgs {\n  /**\n   * Axis along which to calculate norms.\n   *\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * [rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class UnitNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'UnitNorm';\n  private axis: number;\n  private readonly defaultAxis = 0;\n  constructor(args: UnitNormArgs) {\n    super();\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(\n        () => tfc.div(w, tfc.add(epsilon(), calcL2Norms(w, this.axis))));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {axis: this.axis};\n  }\n}\nserialization.registerClass(UnitNorm);\n\nexport class NonNeg extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'NonNeg';\n\n  apply(w: Tensor): Tensor {\n    return tfc.relu(w);\n  }\n}\nserialization.registerClass(NonNeg);\n\nexport interface MinMaxNormArgs {\n  /**\n   * Minimum norm for incoming weights\n   */\n  minValue?: number;\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * `[rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n  /**\n   * Rate for enforcing the constraint: weights will be rescaled to yield:\n   * `(1 - rate) * norm + rate * norm.clip(minValue, maxValue)`.\n   * Effectively, this means that rate=1.0 stands for strict\n   * enforcement of the constraint, while rate<1.0 means that\n   * weights will be rescaled at each step to slowly move\n   * towards a value inside the desired interval.\n   */\n  rate?: number;\n}\n\nexport class MinMaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MinMaxNorm';\n  private minValue: number;\n  private maxValue: number;\n  private rate: number;\n  private axis: number;\n  private readonly defaultMinValue = 0.0;\n  private readonly defaultMaxValue = 1.0;\n  private readonly defaultRate = 1.0;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MinMaxNormArgs) {\n    super();\n    this.minValue =\n        args.minValue != null ? args.minValue : this.defaultMinValue;\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.rate = args.rate != null ? args.rate : this.defaultRate;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.add(\n          tfc.mul(\n              this.rate, tfc.clipByValue(norms, this.minValue, this.maxValue)),\n          tfc.mul(1.0 - this.rate, norms));\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      minValue: this.minValue,\n      maxValue: this.maxValue,\n      rate: this.rate,\n      axis: this.axis\n    };\n  }\n}\nserialization.registerClass(MinMaxNorm);\n\n/** @docinline */\nexport type ConstraintIdentifier =\n    'maxNorm'|'minMaxNorm'|'nonNeg'|'unitNorm'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in ConstraintIdentifier]: string} = {\n      'maxNorm': 'MaxNorm',\n      'minMaxNorm': 'MinMaxNorm',\n      'nonNeg': 'NonNeg',\n      'unitNorm': 'UnitNorm'\n    };\n\nexport function serializeConstraint(constraint: Constraint):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeConstraint(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Constraint {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'constraint');\n}\n\nexport function getConstraint(identifier: ConstraintIdentifier|\n                              serialization.ConfigDict|Constraint): Constraint {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeConstraint(config);\n  } else if (identifier instanceof Constraint) {\n    return identifier;\n  } else {\n    return deserializeConstraint(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constraint, MaxNorm, MaxNormArgs, MinMaxNorm, MinMaxNormArgs, NonNeg, UnitNorm, UnitNormArgs} from './constraints';\n\n/**\n * MaxNorm weight constraint.\n *\n * Constrains the weights incident to each hidden unit\n * to have a norm less than or equal to a desired value.\n *\n * References\n *       - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting\n * Srivastava, Hinton, et al.\n * 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Constraints',namespace: 'constraints'}\n */\nexport function maxNorm(args: MaxNormArgs): Constraint {\n  return new MaxNorm(args);\n}\n\n/**\n * Constrains the weights incident to each hidden unit to have unit norm.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function unitNorm(args: UnitNormArgs): Constraint {\n  return new UnitNorm(args);\n}\n\n/**\n * Constains the weight to be non-negative.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function nonNeg(): Constraint {\n  return new NonNeg();\n}\n\n/** @doc {heading: 'Constraints', namespace: 'constraints'} */\nexport function minMaxNorm(config: MinMaxNormArgs): Constraint {\n  return new MinMaxNorm(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// TODO(huan): add layer-specific input shape types (see: https://github.com/tensorflow/tfjs-layers/pull/492)\n/** @docalias (null | number)[] */\nexport type Shape = Array<null | number>;\n\n// The tfjs-core version of DataType must stay synced with this.\nexport type DataType = 'float32'|'int32'|'bool'|'complex64'|'string';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type DataFormat = 'channelsFirst'|'channelsLast';\nexport const VALID_DATA_FORMAT_VALUES = ['channelsFirst', 'channelsLast'];\n\n// These constants have a snake vs. camel distinction.\nexport type DataFormatSerialization = 'channels_first'|'channels_last';\n\n/** @docinline */\nexport type PaddingMode = 'valid'|'same'|'causal';\nexport const VALID_PADDING_MODE_VALUES = ['valid', 'same', 'causal'];\n\n/** @docinline */\nexport type PoolMode = 'max'|'avg';\nexport const VALID_POOL_MODE_VALUES = ['max', 'avg'];\n\n/** @docinline */\nexport type BidirectionalMergeMode = 'sum'|'mul'|'concat'|'ave';\nexport const VALID_BIDIRECTIONAL_MERGE_MODES = ['sum', 'mul', 'concat', 'ave'];\n\n/** @docinline */\nexport type SampleWeightMode = 'temporal';\nexport const VALID_SAMPLE_WEIGHT_MODES = ['temporal'];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Common functions for TensorFlow.js Layers.\n */\nimport {VALID_DATA_FORMAT_VALUES, VALID_PADDING_MODE_VALUES, VALID_POOL_MODE_VALUES} from './keras_format/common';\nimport {checkStringTypeUnionValue} from './utils/generic_utils';\n\n// A map from the requested scoped name of a Tensor to the number of Tensors\n// wanting that name so far.  This allows enforcing name uniqueness by appending\n// an incrementing index, e.g. scope/name, scope/name_1, scope/name_2, etc.\nconst nameMap: Map<string, number> = new Map<string, number>();\n\nexport function checkDataFormat(value?: string): void {\n  checkStringTypeUnionValue(VALID_DATA_FORMAT_VALUES, 'DataFormat', value);\n}\n\nexport function checkPaddingMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_PADDING_MODE_VALUES, 'PaddingMode', value);\n}\n\nexport function checkPoolMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_POOL_MODE_VALUES, 'PoolMode', value);\n}\n\nconst _nameScopeStack: string[] = [];\nconst _nameScopeDivider = '/';\n\n/**\n * Enter namescope, which can be nested.\n */\nexport function nameScope<T>(name: string, fn: () => T): T {\n  _nameScopeStack.push(name);\n  try {\n    const val: T = fn();\n    _nameScopeStack.pop();\n    return val;\n  } catch (e) {\n    _nameScopeStack.pop();\n    throw e;\n  }\n}\n\n/**\n * Get the current namescope as a flat, concatenated string.\n */\nfunction currentNameScopePrefix(): string {\n  if (_nameScopeStack.length === 0) {\n    return '';\n  } else {\n    return _nameScopeStack.join(_nameScopeDivider) + _nameScopeDivider;\n  }\n}\n\n/**\n * Get the name a Tensor (or Variable) would have if not uniqueified.\n * @param tensorName\n * @return Scoped name string.\n */\nexport function getScopedTensorName(tensorName: string): string {\n  if (!isValidTensorName(tensorName)) {\n    throw new Error('Not a valid tensor name: \\'' + tensorName + '\\'');\n  }\n  return currentNameScopePrefix() + tensorName;\n}\n\n/**\n * Get unique names for Tensors and Variables.\n * @param scopedName The fully-qualified name of the Tensor, i.e. as produced by\n *  `getScopedTensorName()`.\n * @return A unique version of the given fully scoped name.\n *   If this is the first time that the scoped name is seen in this session,\n *   then the given `scopedName` is returned unaltered.  If the same name is\n *   seen again (producing a collision), an incrementing suffix is added to the\n *   end of the name, so it takes the form 'scope/name_1', 'scope/name_2', etc.\n */\nexport function getUniqueTensorName(scopedName: string): string {\n  if (!isValidTensorName(scopedName)) {\n    throw new Error('Not a valid tensor name: \\'' + scopedName + '\\'');\n  }\n  if (!nameMap.has(scopedName)) {\n    nameMap.set(scopedName, 0);\n  }\n  const index = nameMap.get(scopedName);\n  nameMap.set(scopedName, nameMap.get(scopedName) + 1);\n\n  if (index > 0) {\n    const result = `${scopedName}_${index}`;\n    // Mark the composed name as used in case someone wants\n    // to call getUniqueTensorName(\"name_1\").\n    nameMap.set(result, 1);\n    return result;\n  } else {\n    return scopedName;\n  }\n}\n\nconst tensorNameRegex = new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\\._\\/]*$/);\n\n/**\n * Determine whether a string is a valid tensor name.\n * @param name\n * @returns A Boolean indicating whether `name` is a valid tensor name.\n */\nexport function isValidTensorName(name: string): boolean {\n  return !!name.match(tensorNameRegex);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Math utility functions.\n *\n * This file contains some frequently used math function that operates on\n * number[] or Float32Array and return a number. Many of these functions are\n * not-so-thick wrappers around TF.js Core functions. But they offer the\n * convenience of\n * 1) not having to convert the inputs into Tensors,\n * 2) not having to convert the returned Tensors to numbers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar, Tensor1D, tensor1d} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\n\nexport type ArrayTypes = Uint8Array|Int32Array|Float32Array;\n\n/**\n * Determine if a number is an integer.\n */\nexport function isInteger(x: number): boolean {\n  return x === parseInt(x.toString(), 10);\n}\n\n/**\n * Calculate the product of an array of numbers.\n * @param array The array to calculate the product over.\n * @param begin Beginning index, inclusive.\n * @param end Ending index, exclusive.\n * @return The product.\n */\nexport function arrayProd(\n    array: number[]|ArrayTypes, begin?: number, end?: number): number {\n  if (begin == null) {\n    begin = 0;\n  }\n  if (end == null) {\n    end = array.length;\n  }\n\n  let prod = 1;\n  for (let i = begin; i < end; ++i) {\n    prod *= array[i];\n  }\n  return prod;\n}\n\n/**\n * A helper function transforms the two input types to an instance of Tensor1D,\n * so the return value can be fed directly into various TF.js Core functions.\n * @param array\n */\nfunction toArray1D(array: number[]|Float32Array): Tensor1D {\n  array = Array.isArray(array) ? new Float32Array(array) : array;\n  return tensor1d(array);\n}\n\n/**\n * Compute minimum value.\n * @param array\n * @return minimum value.\n */\nexport function min(array: number[]|Float32Array): number {\n  return tfc.min(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute maximum value.\n * @param array\n * @return maximum value\n */\nexport function max(array: number[]|Float32Array): number {\n  return tfc.max(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute sum of array.\n * @param array\n * @return The sum.\n */\nexport function sum(array: number[]|Float32Array): number {\n  return tfc.sum(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute mean of array.\n * @param array\n * @return The mean.\n */\nexport function mean(array: number[]|Float32Array): number {\n  return sum(array) / array.length;\n}\n\n/**\n * Compute variance of array.\n * @param array\n * @return The variance.\n */\nexport function variance(array: number[]|Float32Array): number {\n  const demeaned = tfc.sub(toArray1D(array), scalar(mean(array)));\n  const sumSquare = tfc.sum(tfc.mul(demeaned, demeaned)).dataSync()[0];\n  return sumSquare / array.length;\n}\n\n/**\n * Compute median of array.\n * @param array\n * @return The median value.\n */\nexport function median(array: number[]|Float32Array): number {\n  const arraySorted = array.slice().sort((a, b) => a - b);\n  const lowIdx = Math.floor((arraySorted.length - 1) / 2);\n  const highIdx = Math.ceil((arraySorted.length - 1) / 2);\n  if (lowIdx === highIdx) {\n    return arraySorted[lowIdx];\n  }\n  return (arraySorted[lowIdx] + arraySorted[highIdx]) / 2;\n}\n\n/**\n * Generate an array of integers in [begin, end).\n * @param begin Beginning integer, inclusive.\n * @param end Ending integer, exclusive.\n * @returns Range array.\n * @throws ValueError, iff `end` < `begin`.\n */\nexport function range(begin: number, end: number): number[] {\n  if (end < begin) {\n    throw new ValueError(`end (${end}) < begin (${begin}) is forbidden.`);\n  }\n  const out: number[] = [];\n  for (let i = begin; i < end; ++i) {\n    out.push(i);\n  }\n  return out;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * deeplearn.js backend.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {onesLike as coreOnesLike, scalar, Tensor, Tensor1D, tensor1d, Tensor2D, Tensor3D, Tensor4D, Tensor5D, tidy, where, zerosLike as coreZerosLike} from '@tensorflow/tfjs-core';\nimport {checkDataFormat} from '../common';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {HasShape} from '../types';\nimport * as math_utils from '../utils/math_utils';\n\nimport {imageDataFormat} from './common';\n\n// tslint:enable\n\n/* Setting and getting backend from deeplearn.js. */\n\n// Default deeplearn.js backend is WebGL (GPU).\nlet backend: 'cpu'|'webgl' = 'webgl';\n\nexport function setBackend(requestedBackend: 'cpu'|'webgl') {\n  tfc.setBackend(requestedBackend);\n  backend = requestedBackend;\n}\n\nexport function getBackend(): 'cpu'|'webgl' {\n  return backend;\n}\n\n/**\n * Indicates whether the backend is operating symbolically.\n *\n * This function will be used to determine how to interpret user code. If\n * it returns true, calls to the backend construct a symbolic graph; if\n * it returns false, calls to the backend execute immediately.\n */\nexport function isBackendSymbolic(): boolean {\n  return false;\n}\n\n/**\n * Get the number of elements in a Tensor.\n * @param x The Tensor.\n * @return Number of elements in `x`.\n */\nexport function countParams(x: HasShape): number {\n  const shape = x.shape;\n  if (shape.length > 0) {\n    return shape.reduce((a: number, b: number) => a * b);\n  } else {\n    // Scalar.\n    return 1;\n  }\n}\n\n/**\n * Casts a tensor to a different dtype and returns it.\n * @param x Input tensor.\n * @param dtype String: 'float32'|'int32'|'bool'.\n * @returns Tensor of the specified `dtype`.\n */\nexport function cast(x: Tensor, dtype: tfc.DataType): Tensor {\n  return x.asType(dtype);\n}\n\n/**\n * Adds a 1-sized dimension at index \"axis\".\n * @param x Input tensor.\n * @param axis Position where to add the new axis.\n * @returns Result of the dimension expansion.\n */\nexport function expandDims(x: Tensor, axis = -1): Tensor {\n  const outShape = x.shape.slice();\n  if (axis < 0) {\n    axis = outShape.length + axis + 1;\n  }\n  outShape.splice(axis, 0, 1);\n  return x.reshape(outShape);\n}\n\n/**\n * Repeats a 2D tensor.\n *\n * If `x` has shape `[samples, dim]` and `n` is 2, for example, the output\n * will have shape `[samples, 2, dim]`.\n *\n * @param x Input tensor.\n * @param n Integer, number of times to repeat.\n * @returns The result of the repeat operation.\n * @throws ValueError: If input tensor is not 2D.\n */\nexport function repeat(x: Tensor, n: number): Tensor {\n  return tidy(() => {\n    if (x.shape.length !== 2) {\n      throw new ValueError(\n          `repeat() expects a rank-2 tensor, but received a ` +\n          `rank-${x.shape.length} tensor.`);\n    }\n    const y = expandDims(x, 1);\n    return tile(y, [1, n, 1]);\n  });\n}\n\n/**\n * Flatten a Tensor into 1D.\n * @param x Input tensor.\n * @return The result of the flattening `x`.\n */\nexport function flatten(x: Tensor): Tensor {\n  const newShape = [math_utils.arrayProd(x.shape)];\n  return x.reshape(newShape);\n}\n\n/**\n * Turn a nD tensor into a 2D tensor with same 0th dimension.\n * In other words, it flattens each data samples of a batch.\n *\n * @param x The tensor to flatten. The rank of this tensor is required to be 2\n *   or higher.\n * @return The result of the flattening.\n */\nexport function batchFlatten(x: Tensor): Tensor {\n  if (x.rank <= 1) {\n    throw new ValueError(\n        `batchFlatten requires a minimum rank of 2. Got rank: ${x.rank}.`);\n  }\n  const newShape = [x.shape[0], math_utils.arrayProd(x.shape, 1)];\n  return x.reshape(newShape);\n}\n\n/**\n * Do slicing along the first axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the first axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongFirstAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [start, 0], [size, array.shape[1]]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [start, 0, 0],\n            [size, array.shape[1], array.shape[2]]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [start, 0, 0, 0],\n            [size, array.shape[1], array.shape[2], array.shape[3]]);\n      case 5:\n        return tfc.slice(array as Tensor5D, [start, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4]\n        ]);\n      case 6:\n        return tfc.slice(array, [start, 0, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4],\n          array.shape[5]\n        ]);\n      default:\n        throw new ValueError(\n            `sliceAlongFirstAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the last axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the last axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongLastAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [0, start], [array.shape[0], size]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [0, 0, start],\n            [array.shape[0], array.shape[1], size]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [0, 0, 0, start],\n            [array.shape[0], array.shape[1], array.shape[2], size]);\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the sepcified axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size of the slice along the chosen axis.\n * @param choose an axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongAxis(\n    array: Tensor, start: number, size: number, axis: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 3:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice3d(\n                array as Tensor3D, [0, start, 0],\n                [array.shape[0], size, array.shape[2]]);\n          case 3:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 4:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice4d(\n                array as Tensor4D, [0, start, 0, 0],\n                [array.shape[0], size, array.shape[2], array.shape[3]]);\n          case 3:\n            return tfc.slice4d(\n                array as Tensor4D, [0, 0, start, 0],\n                [array.shape[0], array.shape[1], size, array.shape[3]]);\n          case 4:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Concatenates a list of tensors alongside the specified axis.\n * @param tensors `Array` of tensors to concatenate.\n * @param axis Concatenation axis.\n * @returns The result of the concatenation.\n */\nexport function concatenate(tensors: Tensor[], axis = -1): Tensor {\n  let rank: number;\n  if (axis < 0) {\n    rank = tensors[0].rank;\n    if (rank !== 0) {\n      axis = rank;\n    } else {\n      axis = 0;\n    }\n  }\n  if (axis === tensors[0].rank) {\n    // Porting Note: This is necessary because tfc.concat() requires axis to be\n    //   in the interval [-rank, rank).\n    axis = -1;\n  }\n  // Porting Note: Sparse concat is not supported yet.\n  return tfc.concat(tensors, axis);\n}\n\n/**\n * Concatenate two arrays along the first dimension.\n * @param a The 1st `tf.Tensor` to concatenate.\n * @param b The 2nd `tf.Tensor` to concatenate.\n * @returns Result of the concatenation.\n * @throws ValueError: If `a` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function concatAlongFirstAxis(a: Tensor, b: Tensor): Tensor {\n  switch (a.rank) {\n    case 1:\n      return tfc.concat1d([a as Tensor1D, b as Tensor1D]);\n    case 2:\n      return tfc.concat2d([a as Tensor2D, b as Tensor2D], 0);\n    case 3:\n      return tfc.concat3d([a as Tensor3D, b as Tensor3D], 0);\n    case 4:\n      return tfc.concat4d([a as Tensor4D, b as Tensor4D], 0);\n    default:\n      throw new ValueError(\n          `concatAlongFirstAxis() received an unsupported ` +\n          `tensor rank: ${a.rank}`);\n  }\n}\n\n/**\n * Creates a tensor by tiling `x` by `n`.\n * @param x A tensor.\n * @param n An Array of integers or a single integer. If an Array, the length\n *   must be the same as the number of dimensions in `x`. If a single integer,\n *   it will be treated as an Array of length 1.\n */\nexport function tile(x: Tensor, n: number|number[]): Tensor {\n  if (!Array.isArray(n)) {\n    n = [n];\n  }\n  if (x.rank !== n.length) {\n    throw new ValueError(\n        `The length of input n (${n.length}) does not match ` +\n        `the number of dimensions in input x (${x.rank})`);\n  }\n  return tfc.tile(x, n);\n}\n\n/* Creation of random tensors. */\n\n/**\n * Get a tensor with normal distribution of values.\n *\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @return The normal tensor.\n */\nexport function randomNormal(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: 'float32'|'int32',\n    seed?: number): Tensor {\n  return tfc.randomNormal(shape, mean, stddev, dtype, seed);\n}\n\n/* Linear Algebra */\n\n/**\n * Multiply two tensors and returns the result as a tensor.\n *\n * For 2D tensors, this is equivalent to matrix multiplication (matMul).\n * For tensors of higher ranks, it follows the Theano behavior,\n * (e.g. `(2, 3) * (4, 3, 5) -> (2, 4, 5)`).  From the Theano documentation:\n *\n * For N dimensions it is a sum product over the last axis of x and the\n * second-to-last of y:\n *\n * @param a A tensor of at least rank 2.\n * @param b A tensor of at least rank 2.\n * @param activation (optional) A string identifying the activation\n *   function.\n * @return Result of the dot operation.\n */\nexport function dot(\n    a: Tensor, b: Tensor, activation?: tfc.fused.Activation,\n    bias?: Tensor): Tensor {\n  if ((a.rank < 2) || (b.rank < 2)) {\n    throw new NotImplementedError(\n        `dot requires both inputs to be rank >= 2` +\n        ` but got x shape = ${a.shape} and y shape = ${b.shape}`);\n  }\n  if (b.rank >= 3) {\n    const xLastDim = a.shape.slice(-1)[0];\n    const ySecondLastDim = b.shape.slice(-2)[0];\n    if (xLastDim !== ySecondLastDim) {\n      throw new NotImplementedError(\n          `If rank y >= 3, then the second last dim` +\n          ` of y must equal the last dim of x but got x shape = ${\n              a.shape} and ` +\n          ` y shape = ${b.shape}`);\n    }\n  }\n  // Handle basic 2D x 2D case.\n  if ((a.rank === 2) && (b.rank === 2)) {\n    const transposeA = false;\n    const transposeB = false;\n    // tfc.fused.matMul only fuses certain activation functions. Unsupported\n    // activation functions are treated as 'linear' activations, which is\n    // equivalent to a no-op.\n    return tfc.fused.matMul({\n      a,\n      b: b as Tensor2D,\n      transposeA,\n      transposeB,\n      bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n      activation\n    });\n  } else {\n    // Reshape x into the analogous 2D Tensor.\n    const aFirstDims = a.shape.slice();  // Holds all but the last dim of x.\n    const aLastDim = aFirstDims.pop();\n    a = a.reshape([-1, aLastDim]);\n\n    // Reshape y into the analogous 2D Tensor, and keep track of the\n    // required dimensions to reproduce the output shape.\n    const bShape = b.shape.slice();\n    const bLastDim = bShape.pop();\n    const ySecondLastDim = bShape.pop();\n    const yOtherDims = [...bShape, bLastDim];\n    // permutation should be like [r-2, 0, 1, 2, ... r-4, r-3, r-1]\n    // where r is the rank of y.\n    const perm = Array.from({length: b.rank}, (_, i) => {\n      if (i === 0) {\n        return b.rank - 2;\n      } else if (i <= b.rank - 2) {\n        return i - 1;\n      }\n      return i;\n    });\n    b = b.transpose(perm).reshape([ySecondLastDim, -1]);\n\n    // Multiply x and y as 2D Tensors, and then reshape back to original.\n    const outputShape = [...aFirstDims, ...yOtherDims];\n    const transposeA = false;\n    const transposeB = false;\n    return tfc.fused\n        .matMul({\n          a,\n          b,\n          transposeA,\n          transposeB,\n          bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n          activation\n        })\n        .reshape(outputShape);\n  }\n}\n\n/**\n * Compute the sign Tensor of an input Tensor.\n *\n * Elements of the input `tf.Tensor` that are === 0 are mapped to 0.\n * Elements of the input `tf.Tensor` that are > 0 are mapped to 1.\n * Elements of the input `tf.Tensor` that are < 0 are mapped to -1.\n *\n * @param x Input `tf.Tensor`.\n * @return The sign `tf.Tensor`.\n */\nexport function sign(x: Tensor): Tensor {\n  // TODO(cais): Move to the core.\n  return tidy(() => {\n    const zerosLikeX = coreZerosLike(x);\n    const onesLikeX = coreOnesLike(x);\n    return where(\n        tfc.equal(x, zerosLikeX), zerosLikeX,\n        where(\n            tfc.greater(x, coreZerosLike(x)), onesLikeX,\n            tfc.mul(-1, onesLikeX)));\n  });\n}\n\n/**\n * Computes the one-hot representation of an integer tensor.\n * @param indices nD integer tensor of shape\n *   `(batch_size, dim1, dim2, ... dim(n-1))`\n * @param numClasses Integer, number of classes to consider.\n * @returns (n + 1)D one hot representation of the input\n *   with shape `(batch_size, dim1, dim2, ... dim(n-1), num_classes)`\n */\nexport function oneHot(indices: Tensor, numClasses: number): Tensor {\n  return tidy(() => {\n    if (indices.rank !== 1) {\n      throw new Error(\n          'Only 1D one-hot tensors are supported in the ' +\n          'deeplearn backend, at present.');\n    }\n    indices = indices.toInt();\n    return tfc.oneHot(indices as Tensor1D, numClasses).toFloat();\n  });\n}\n\n/* Elementary math functions. */\n\n/**\n * Retrieves the elements of indices `indices` in the tensor `reference`.\n * @param reference A tensor.\n * @param indices An integer tensor of indices or an `Array` of integers.\n * @param axis Axis along which to perform the gather operation.\n * @returns The result of the gathering as a tensor.\n */\nexport function gather(\n    reference: Tensor, indices: number[]|Tensor1D, axis?: number): Tensor {\n  return tidy(() => {\n    if (Array.isArray(indices)) {\n      indices = tensor1d(indices, 'int32');\n    } else {\n      indices = indices.toInt();\n    }\n    return tfc.gather(reference, indices, axis);\n  });\n}\n\n/**\n * Element-wise square.\n * @param x Input tensor.\n * @return element-wise x^2\n */\nexport function square(x: Tensor): Tensor {\n  return tfc.mul(x, x);\n}\n\n/**\n * Element-wise exponentiation.\n *\n * Porting Note: In PyKeras, `a` (the exponent) is a Python integer, which\n *   takes advatnage of the backend's (e.g., TensorFlow's) automatic\n * conversion to tensor. Here we allow `a` to be either a number or a tensor.\n *\n * @param x The base tensor.\n * @param a The exponent, tensor or number. If a number, it is rounded to the\n *   nearest integer and converted to a tensor.\n * @returns A tensor of the same shape as `x`.\n */\nexport function pow(x: Tensor, a: Tensor|number): Tensor {\n  return tidy(() => {\n    if (typeof (a) === 'number') {\n      a = scalar(Math.round(a), 'int32');\n    }\n    if (a.dtype !== 'int32') {\n      throw new NotImplementedError(\n          `Non-int32 dtype (${a.dtype}) is not supported by pow() yet`);\n    }\n    return tfc.pow(x, a);\n  });\n}\n\n/**\n * Reshapes bias tensor according to rank of x.\n */\nfunction reshapeBias(xRank: number, bias: Tensor, dataFormat: string) {\n  const biasShape = bias.shape;\n\n  if (bias.rank !== 1 && bias.rank !== xRank) {\n    throw new ValueError(\n        `Unexpected bias dimensions: ${bias.rank}` +\n        `; expected it to be 1 or ${xRank}`);\n  }\n\n  if (xRank === 5) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1, 1]);\n      } else {\n        return bias.reshape(\n            [1, biasShape[3], biasShape[0], biasShape[1], biasShape[2]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 4) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1]);\n      } else {\n        return bias.reshape([1, biasShape[2], biasShape[0], biasShape[1]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 3) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1]);\n      } else {\n        return bias.reshape([1, biasShape[1], biasShape[0]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank < 3) {\n    return bias;\n  }\n  throw new ValueError(`Unsupported input rank by biasAdd: ${bias.rank}`);\n}\n\n/* Neural-network operations. */\n\n/**\n * Add a bias to a tensor.\n *\n * @param x The tensor to add the bias to.\n * @param bias The bias to add to `x`. Must be 1D or the same rank as `x`.\n * @return Result of the bias adding.\n * @throws ValueError: If the rank of `bias` is incorrect.\n */\nexport function biasAdd(\n    x: Tensor, bias: Tensor, dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n\n    return x.add(reshapeBias(x.rank, bias, dataFormat));\n  });\n}\n\n/**\n * Exponential linear unit (ELU).\n * @param x A tensor or variable to compute the activation function for.\n * @param alpha: A scalar, a scaling factor for the negative section.\n * @return Output of the ELU operation.\n */\nexport function elu(x: Tensor, alpha = 1): Tensor {\n  // TODO(cais): Add support for alpha values other than 1.\n  if (alpha !== 1) {\n    throw new NotImplementedError(\n        `Support for alpha values other than 1 (${alpha}) is not implemented ` +\n        `yet.`);\n  }\n  return tfc.elu(x);\n}\n\n/**\n * Softsign of a tensor.\n *\n * Defined as x / (abs(x) + 1), element-wise.\n *\n * @param x: Input.\n * @returns Output.\n */\nexport function softsign(x: Tensor): Tensor {\n  return tidy(() => tfc.div(x, tfc.abs(x).add(1)));\n}\n\n/**\n * Sets entries in `x` to zero at random, while scaling the entire tensor.\n *\n * @param x input tensor.\n * @param level fraction of the entries in the tensor that will be set to 0.\n * @param noiseShape shape of randomly generated keep/drop flags, must be\n *   broadcastable to the shape of `x`. Optional.\n * @param seed random seed to ensure determinism. Optional.\n * @returns Result of the dropout operation.\n */\nexport function dropout(\n    x: Tensor, level: number, noiseShape?: number[], seed?: number): Tensor {\n  return tidy(() => tfc.dropout(x, level, noiseShape, seed));\n}\n\n/**\n * Element-wise, segment-wise linear approximation of sigmoid.\n *\n * Returns `0.` if `x < -2.5`, `1.` if `x > 2.5`.\n * In `-2.5 <= x <= 2.5`, returns `0.2 * x + 0.5`.\n *\n * @param x Input tensor.\n * @returns Output tensor.\n */\nexport function hardSigmoid(x: Tensor): Tensor {\n  return tidy(() => {\n    const y = tfc.add(.5, tfc.mul(.2, x));\n    return tfc.clipByValue(y, 0, 1);\n  });\n}\n\n/**\n * Invoke `x` in the training phase, and `alt` otherwise.\n *\n * Porting Note: We do not create placeholder tensors for the `training`\n * boolean flag here, because there is no such thing in the TF.js imperative\n * backend.\n *\n * @param x The function to invoke iff `training` is `true`.\n * @param alt The function to invoke iff `training` is `false`.\n * @param training Boolean flag for whether training phase is active.\n * @returns The return value of `x()` if `training` is `true`, or the return\n *   value of `alt()` if `training` is `false`.\n */\nexport function inTrainPhase<T>(x: () => T, alt: () => T, training = false): T {\n  return training ? x() : alt();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {BaseSerialization} from './types';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type FanMode = 'fanIn'|'fanOut'|'fanAvg';\nexport const VALID_FAN_MODE_VALUES = ['fanIn', 'fanOut', 'fanAvg'];\n\n// These constants have a snake vs. camel distinction.\nexport type FanModeSerialization = 'fan_in'|'fan_out'|'fan_avg';\n\n/** @docinline */\nexport type Distribution = 'normal'|'uniform'|'truncatedNormal';\nexport const VALID_DISTRIBUTION_VALUES =\n    ['normal', 'uniform', 'truncatedNormal'];\n// These constants have a snake vs. camel distinction.\nexport type DistributionSerialization = 'normal'|'uniform'|'truncated_normal';\n\nexport type ZerosSerialization = BaseSerialization<'Zeros', {}>;\n\nexport type OnesSerialization = BaseSerialization<'Ones', {}>;\n\nexport type ConstantConfig = {\n  value: number;\n};\n\nexport type ConstantSerialization =\n    BaseSerialization<'Constant', ConstantConfig>;\n\nexport type RandomNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type RandomNormalSerialization =\n    BaseSerialization<'RandomNormal', RandomNormalConfig>;\n\nexport type RandomUniformConfig = {\n  minval?: number;\n  maxval?: number;\n  seed?: number;\n};\n\nexport type RandomUniformSerialization =\n    BaseSerialization<'RandomUniform', RandomUniformConfig>;\n\nexport type TruncatedNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type TruncatedNormalSerialization =\n    BaseSerialization<'TruncatedNormal', TruncatedNormalConfig>;\n\nexport type VarianceScalingConfig = {\n  scale?: number;\n\n  mode?: FanModeSerialization;\n  distribution?: DistributionSerialization;\n  seed?: number;\n};\n\nexport type VarianceScalingSerialization =\n    BaseSerialization<'VarianceScaling', VarianceScalingConfig>;\n\nexport type OrthogonalConfig = {\n  seed?: number;\n  gain?: number;\n};\n\nexport type OrthogonalSerialization =\n    BaseSerialization<'Orthogonal', OrthogonalConfig>;\n\nexport type IdentityConfig = {\n  gain?: number;\n};\n\nexport type IdentitySerialization =\n    BaseSerialization<'Identity', IdentityConfig>;\n\n// Update initializerClassNames below in concert with this.\nexport type InitializerSerialization = ZerosSerialization|OnesSerialization|\n    ConstantSerialization|RandomUniformSerialization|RandomNormalSerialization|\n    TruncatedNormalSerialization|IdentitySerialization|\n    VarianceScalingSerialization|OrthogonalSerialization;\n\nexport type InitializerClassName = InitializerSerialization['class_name'];\n\n// We can't easily extract a string[] from the string union type, but we can\n// recapitulate the list, enforcing at compile time that the values are valid\n// and that we have the right number of them.\n\n/**\n * A string array of valid Initializer class names.\n *\n * This is guaranteed to match the `InitializerClassName` union type.\n */\nexport const initializerClassNames: InitializerClassName[] = [\n  'Zeros', 'Ones', 'Constant', 'RandomNormal', 'RandomUniform',\n  'TruncatedNormal', 'VarianceScaling', 'Orthogonal', 'Identity'\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, eye, linalg, mul, ones, randomUniform, scalar, serialization, Tensor, Tensor2D, tidy, truncatedNormal, zeros} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {checkDataFormat} from './common';\nimport {NotImplementedError, ValueError} from './errors';\nimport {DataFormat, Shape} from './keras_format/common';\nimport {Distribution, FanMode, VALID_DISTRIBUTION_VALUES, VALID_FAN_MODE_VALUES} from './keras_format/initializer_config';\nimport {checkStringTypeUnionValue, deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\nimport {arrayProd} from './utils/math_utils';\n\nexport function checkFanMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_FAN_MODE_VALUES, 'FanMode', value);\n}\n\nexport function checkDistribution(value?: string): void {\n  checkStringTypeUnionValue(VALID_DISTRIBUTION_VALUES, 'Distribution', value);\n}\n\n/**\n * Initializer base class.\n *\n * @doc {\n *   heading: 'Initializers', subheading: 'Classes', namespace: 'initializers'}\n */\nexport abstract class Initializer extends serialization.Serializable {\n  public fromConfigUsesCustomObjects(): boolean {\n    return false;\n  }\n  /**\n   * Generate an initial value.\n   * @param shape\n   * @param dtype\n   * @return The init value.\n   */\n  abstract apply(shape: Shape, dtype?: DataType): Tensor;\n\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport class Zeros extends Initializer {\n  /** @nocollapse */\n  static className = 'Zeros';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return zeros(shape, dtype);\n  }\n}\nserialization.registerClass(Zeros);\n\nexport class Ones extends Initializer {\n  /** @nocollapse */\n  static className = 'Ones';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return ones(shape, dtype);\n  }\n}\nserialization.registerClass(Ones);\n\nexport interface ConstantArgs {\n  /** The value for each element in the variable. */\n  value: number;\n}\n\nexport class Constant extends Initializer {\n  /** @nocollapse */\n  static className = 'Constant';\n  private value: number;\n  constructor(args: ConstantArgs) {\n    super();\n    if (typeof args !== 'object') {\n      throw new ValueError(\n          `Expected argument of type ConstantConfig but got ${args}`);\n    }\n    if (args.value === undefined) {\n      throw new ValueError(`config must have value set but got ${args}`);\n    }\n    this.value = args.value;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => mul(scalar(this.value), ones(shape, dtype)));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      value: this.value,\n    };\n  }\n}\nserialization.registerClass(Constant);\n\nexport interface RandomUniformArgs {\n  /** Lower bound of the range of random values to generate. */\n  minval?: number;\n  /** Upper bound of the range of random values to generate. */\n  maxval?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomUniform extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomUniform';\n  readonly DEFAULT_MINVAL = -0.05;\n  readonly DEFAULT_MAXVAL = 0.05;\n  private minval: number;\n  private maxval: number;\n  private seed: number;\n\n  constructor(args: RandomUniformArgs) {\n    super();\n    this.minval = args.minval || this.DEFAULT_MINVAL;\n    this.maxval = args.maxval || this.DEFAULT_MAXVAL;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return randomUniform(shape, this.minval, this.maxval, dtype);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {minval: this.minval, maxval: this.maxval, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomUniform);\n\nexport interface RandomNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomNormal';\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: RandomNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `randomNormal does not support dType ${dtype}.`);\n    }\n\n    return K.randomNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomNormal);\n\nexport interface TruncatedNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class TruncatedNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'TruncatedNormal';\n\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: TruncatedNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `truncatedNormal does not support dType ${dtype}.`);\n    }\n    return truncatedNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(TruncatedNormal);\n\nexport interface IdentityArgs {\n  /**\n   * Multiplicative factor to apply to the identity matrix.\n   */\n  gain?: number;\n}\n\nexport class Identity extends Initializer {\n  /** @nocollapse */\n  static className = 'Identity';\n  private gain: number;\n  constructor(args: IdentityArgs) {\n    super();\n    this.gain = args.gain != null ? args.gain : 1.0;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length !== 2 || shape[0] !== shape[1]) {\n        throw new ValueError(\n            'Identity matrix initializer can only be used for' +\n            ' 2D square matrices.');\n      } else {\n        return mul(this.gain, eye(shape[0]));\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {gain: this.gain};\n  }\n}\nserialization.registerClass(Identity);\n\n/**\n * Computes the number of input and output units for a weight shape.\n * @param shape Shape of weight.\n * @param dataFormat data format to use for convolution kernels.\n *   Note that all kernels in Keras are standardized on the\n *   CHANNEL_LAST ordering (even when inputs are set to CHANNEL_FIRST).\n * @return An length-2 array: fanIn, fanOut.\n */\nfunction computeFans(\n    shape: Shape, dataFormat: DataFormat = 'channelsLast'): number[] {\n  let fanIn: number;\n  let fanOut: number;\n  checkDataFormat(dataFormat);\n  if (shape.length === 2) {\n    fanIn = shape[0];\n    fanOut = shape[1];\n  } else if ([3, 4, 5].indexOf(shape.length) !== -1) {\n    if (dataFormat === 'channelsFirst') {\n      const receptiveFieldSize = arrayProd(shape, 2);\n      fanIn = shape[1] * receptiveFieldSize;\n      fanOut = shape[0] * receptiveFieldSize;\n    } else if (dataFormat === 'channelsLast') {\n      const receptiveFieldSize = arrayProd(shape, 0, shape.length - 2);\n      fanIn = shape[shape.length - 2] * receptiveFieldSize;\n      fanOut = shape[shape.length - 1] * receptiveFieldSize;\n    }\n  } else {\n    const shapeProd = arrayProd(shape);\n    fanIn = Math.sqrt(shapeProd);\n    fanOut = Math.sqrt(shapeProd);\n  }\n\n  return [fanIn, fanOut];\n}\n\nexport interface VarianceScalingArgs {\n  /** Scaling factor (positive float). */\n  scale?: number;\n\n  /** Fanning mode for inputs and outputs. */\n  mode?: FanMode;\n\n  /** Probabilistic distribution of the values. */\n  distribution?: Distribution;\n\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class VarianceScaling extends Initializer {\n  /** @nocollapse */\n  static className = 'VarianceScaling';\n  private scale: number;\n  private mode: FanMode;\n  private distribution: Distribution;\n  private seed: number;\n\n  /**\n   * Constructor of VarianceScaling.\n   * @throws ValueError for invalid value in scale.\n   */\n  constructor(args: VarianceScalingArgs) {\n    super();\n    if (args.scale < 0.0) {\n      throw new ValueError(\n          `scale must be a positive float. Got: ${args.scale}`);\n    }\n    this.scale = args.scale == null ? 1.0 : args.scale;\n    this.mode = args.mode == null ? 'fanIn' : args.mode;\n    checkFanMode(this.mode);\n    this.distribution =\n        args.distribution == null ? 'normal' : args.distribution;\n    checkDistribution(this.distribution);\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    const fans = computeFans(shape);\n    const fanIn = fans[0];\n    const fanOut = fans[1];\n    let scale = this.scale;\n    if (this.mode === 'fanIn') {\n      scale /= Math.max(1, fanIn);\n    } else if (this.mode === 'fanOut') {\n      scale /= Math.max(1, fanOut);\n    } else {\n      scale /= Math.max(1, (fanIn + fanOut) / 2);\n    }\n\n    if (this.distribution === 'normal') {\n      const stddev = Math.sqrt(scale);\n      dtype = dtype || 'float32';\n      if (dtype !== 'float32' && dtype !== 'int32') {\n        throw new NotImplementedError(\n            `${this.getClassName()} does not support dType ${dtype}.`);\n      }\n      return truncatedNormal(shape, 0, stddev, dtype, this.seed);\n    } else {\n      const limit = Math.sqrt(3 * scale);\n      return randomUniform(shape, -limit, limit, dtype);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      scale: this.scale,\n      mode: this.mode,\n      distribution: this.distribution,\n      seed: this.seed\n    };\n  }\n}\nserialization.registerClass(VarianceScaling);\n\nexport interface SeedOnlyInitializerArgs {\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class GlorotUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotUniform';\n\n  /**\n   * Constructor of GlorotUniform\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotUniform);\n\nexport class GlorotNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotNormal';\n\n  /**\n   * Constructor of GlorotNormal.\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotNormal);\n\nexport class HeNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeNormal);\n\nexport class HeUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeUniform';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeUniform);\n\nexport class LeCunNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunNormal);\n\nexport class LeCunUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunUniform);\n\nexport interface OrthogonalArgs extends SeedOnlyInitializerArgs {\n  /**\n   * Multiplicative factor to apply to the orthogonal matrix. Defaults to 1.\n   */\n  gain?: number;\n}\n\nexport class Orthogonal extends Initializer {\n  /** @nocollapse */\n  static className = 'Orthogonal';\n  readonly DEFAULT_GAIN = 1;\n  protected readonly gain: number;\n  protected readonly seed: number;\n\n  constructor(args?: OrthogonalArgs) {\n    super();\n    this.gain = args.gain == null ? this.DEFAULT_GAIN : args.gain;\n    this.seed = args.seed;\n\n    if (this.seed != null) {\n      throw new NotImplementedError(\n          'Random seed is not implemented for Orthogonal Initializer yet.');\n    }\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length < 2) {\n        throw new NotImplementedError('Shape must be at least 2D.');\n      }\n      if (shape[0] * shape[1] > 2000) {\n        console.warn(\n            `Orthogonal initializer is being called on a matrix with more ` +\n            `than 2000 (${shape[0] * shape[1]}) elements: ` +\n            `Slowness may result.`);\n      }\n\n      // TODO(cais): Add seed support.\n      const normalizedShape =\n          shape[0] > shape[1] ? [shape[1], shape[0]] : shape;\n      const a = K.randomNormal(normalizedShape, 0, 1, 'float32') as Tensor2D;\n      let q = linalg.gramSchmidt(a) as Tensor2D;\n      if (shape[0] > shape[1]) {\n        q = q.transpose();\n      }\n      return mul(this.gain, q);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      gain: this.gain,\n      seed: this.seed,\n    };\n  }\n}\nserialization.registerClass(Orthogonal);\n\n/** @docinline */\nexport type InitializerIdentifier =\n    'constant'|'glorotNormal'|'glorotUniform'|'heNormal'|'heUniform'|'identity'|\n    'leCunNormal'|'leCunUniform'|'ones'|'orthogonal'|'randomNormal'|\n    'randomUniform'|'truncatedNormal'|'varianceScaling'|'zeros'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in InitializerIdentifier]: string} = {\n      'constant': 'Constant',\n      'glorotNormal': 'GlorotNormal',\n      'glorotUniform': 'GlorotUniform',\n      'heNormal': 'HeNormal',\n      'heUniform': 'HeUniform',\n      'identity': 'Identity',\n      'leCunNormal': 'LeCunNormal',\n      'leCunUniform': 'LeCunUniform',\n      'ones': 'Ones',\n      'orthogonal': 'Orthogonal',\n      'randomNormal': 'RandomNormal',\n      'randomUniform': 'RandomUniform',\n      'truncatedNormal': 'TruncatedNormal',\n      'varianceScaling': 'VarianceScaling',\n      'zeros': 'Zeros'\n    };\n\nfunction deserializeInitializer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Initializer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'initializer');\n}\n\nexport function serializeInitializer(initializer: Initializer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(initializer);\n}\n\nexport function getInitializer(identifier: InitializerIdentifier|Initializer|\n                               serialization.ConfigDict): Initializer {\n  if (typeof identifier === 'string') {\n    const className = identifier in INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    /* We have four 'helper' classes for common initializers that\n    all get serialized as 'VarianceScaling' and shouldn't go through\n    the deserializeInitializer pathway. */\n    if (className === 'GlorotNormal') {\n      return new GlorotNormal();\n    } else if (className === 'GlorotUniform') {\n      return new GlorotUniform();\n    } else if (className === 'HeNormal') {\n      return new HeNormal();\n    } else if (className === 'HeUniform') {\n      return new HeUniform();\n    } else if (className === 'LeCunNormal') {\n      return new LeCunNormal();\n    } else if (className === 'LeCunUniform') {\n      return new LeCunUniform();\n    } else {\n      const config: serialization.ConfigDict = {};\n      config['className'] = className;\n      config['config'] = {};\n      return deserializeInitializer(config);\n    }\n  } else if (identifier instanceof Initializer) {\n    return identifier;\n  } else {\n    return deserializeInitializer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constant, ConstantArgs, GlorotNormal, GlorotUniform, HeNormal, HeUniform, Identity, IdentityArgs, Initializer, LeCunNormal, LeCunUniform, Ones, Orthogonal, OrthogonalArgs, RandomNormal, RandomNormalArgs, RandomUniform, RandomUniformArgs, SeedOnlyInitializerArgs, TruncatedNormal, TruncatedNormalArgs, VarianceScaling, VarianceScalingArgs, Zeros} from './initializers';\n\n/**\n * Initializer that generates tensors initialized to 0.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function zeros(): Zeros {\n  return new Zeros();\n}\n\n/**\n * Initializer that generates tensors initialized to 1.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function ones(): Initializer {\n  return new Ones();\n}\n\n/**\n * Initializer that generates values initialized to some constant.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function constant(args: ConstantArgs): Initializer {\n  return new Constant(args);\n}\n\n/**\n * Initializer that generates random values initialized to a uniform\n * distribution.\n *\n * Values will be distributed uniformly between the configured minval and\n * maxval.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomUniform(args: RandomUniformArgs): Initializer {\n  return new RandomUniform(args);\n}\n\n/**\n * Initializer that generates random values initialized to a normal\n * distribution.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomNormal(args: RandomNormalArgs): Initializer {\n  return new RandomNormal(args);\n}\n\n/**\n * Initializer that generates random values initialized to a truncated normal.\n * distribution.\n *\n * These values are similar to values from a `RandomNormal` except that values\n * more than two standard deviations from the mean are discarded and re-drawn.\n * This is the recommended initializer for neural network weights and filters.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function truncatedNormal(args: TruncatedNormalArgs): Initializer {\n  return new TruncatedNormal(args);\n}\n\n/**\n * Initializer that generates the identity matrix.\n * Only use for square 2D matrices.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function identity(args: IdentityArgs): Initializer {\n  return new Identity(args);\n}\n\n/**\n * Initializer capable of adapting its scale to the shape of weights.\n * With distribution=NORMAL, samples are drawn from a truncated normal\n * distribution centered on zero, with `stddev = sqrt(scale / n)` where n is:\n *   - number of input units in the weight tensor, if mode = FAN_IN.\n *   - number of output units, if mode = FAN_OUT.\n *   - average of the numbers of input and output units, if mode = FAN_AVG.\n * With distribution=UNIFORM,\n * samples are drawn from a uniform distribution\n * within [-limit, limit], with `limit = sqrt(3 * scale / n)`.\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function varianceScaling(config: VarianceScalingArgs): Initializer {\n  return new VarianceScaling(config);\n}\n\n/**\n * Glorot uniform initializer, also called Xavier uniform initializer.\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotUniform(args);\n}\n\n/**\n * Glorot normal initializer, also called Xavier normal initializer.\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor.\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotNormal(args);\n}\n\n/**\n * He normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function heNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeNormal(args);\n}\n\n/**\n * He uniform initializer.\n *\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / fan_in)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function heUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeUniform(args);\n}\n\n/**\n * LeCun normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(1 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * References:\n *   [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *   [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunNormal(args);\n}\n\n/**\n * LeCun uniform initializer.\n *\n * It draws samples from a uniform distribution in the interval\n * `[-limit, limit]` with `limit = sqrt(3 / fanIn)`,\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunUniform(args);\n}\n\n/**\n * Initializer that generates a random orthogonal matrix.\n *\n * Reference:\n * [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function orthogonal(args: OrthogonalArgs): Initializer {\n  return new Orthogonal(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Utilities related to persistent state in the backend.\n */\n\n/**\n * An ID to track `tf.SymbolicTensor`s and derived classes.\n * Required in different places in engine/topology.ts to identify unique\n * tensors.\n */\nlet _nextUniqueTensorId = 0;\n\nexport function getNextUniqueTensorId(): number {\n  return _nextUniqueTensorId++;\n}\n\nconst _uidPrefixes: {[prefix: string]: number} = {};\n\n/**\n * Provides a unique UID given a string prefix.\n *\n * @param prefix\n */\nexport function getUid(prefix = ''): string {\n  if (!(prefix in _uidPrefixes)) {\n    _uidPrefixes[prefix] = 0;\n  }\n  _uidPrefixes[prefix] += 1;\n  return prefix + _uidPrefixes[prefix].toString();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\n// tslint:enable\n\n/**\n * Determine whether the input is an Array of Shapes.\n */\nexport function isArrayOfShapes(x: Shape|Shape[]): boolean {\n  return Array.isArray(x) && Array.isArray(x[0]);\n}\n\n/**\n * Special case of normalizing shapes to lists.\n *\n * @param x A shape or list of shapes to normalize into a list of Shapes.\n * @return A list of Shapes.\n */\nexport function normalizeShapeList(x: Shape|Shape[]): Shape[] {\n  if (x.length === 0) {\n    return [];\n  }\n  if (!Array.isArray(x[0])) {\n    return [x] as Shape[];\n  }\n  return x as Shape[];\n}\n\n/**\n * Helper function to obtain exactly one Tensor.\n * @param xs: A single `tf.Tensor` or an `Array` of `tf.Tensor`s.\n * @return A single `tf.Tensor`. If `xs` is an `Array`, return the first one.\n * @throws ValueError: If `xs` is an `Array` and its length is not 1.\n */\nexport function getExactlyOneTensor(xs: Tensor|Tensor[]): Tensor {\n  let x: Tensor;\n  if (Array.isArray(xs)) {\n    if (xs.length !== 1) {\n      throw new ValueError(`Expected Tensor length to be 1; got ${xs.length}`);\n    }\n    x = xs[0];\n  } else {\n    x = xs;\n  }\n  return x;\n}\n\n/**\n * Helper function to obtain exactly on instance of Shape.\n *\n * @param shapes Input single `Shape` or Array of `Shape`s.\n * @returns If input is a single `Shape`, return it unchanged. If the input is\n *   an `Array` containing exactly one instance of `Shape`, return the instance.\n *   Otherwise, throw a `ValueError`.\n * @throws ValueError: If input is an `Array` of `Shape`s, and its length is not\n *   1.\n */\nexport function getExactlyOneShape(shapes: Shape|Shape[]): Shape {\n  if (Array.isArray(shapes) && Array.isArray(shapes[0])) {\n    if (shapes.length === 1) {\n      shapes = shapes as Shape[];\n      return shapes[0];\n    } else {\n      throw new ValueError(`Expected exactly 1 Shape; got ${shapes.length}`);\n    }\n  } else {\n    return shapes as Shape;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {LayerVariable} from '../variables';\n\n/**\n * Count the elements in an Array of LayerVariables.\n *\n * @param weights: The LayerVariables of which the constituent numbers are to\n *   be counted.\n * @returns A count of the elements in all the LayerVariables\n */\nexport function countParamsInWeights(weights: LayerVariable[]): number {\n  let count = 0;\n  for (const weight of weights) {\n    if (weight.shape.length === 0) {\n      count += 1;\n    } else {\n      count += weight.shape.reduce((a, b) => a * b);\n    }\n  }\n  return count;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, Tensor, variableGrads} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId} from './backend/state';\nimport {getScopedTensorName, getUniqueTensorName} from './common';\nimport {Constraint} from './constraints';\nimport {NotImplementedError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {HasShape} from './types';\n\nconst DEFAULT_VARIABLE_NAME_PREFIX = 'Variable';\n\n/**\n * A `tf.layers.LayerVariable` is similar to a `tf.Tensor` in that it has a\n * dtype and shape, but its value is mutable.  The value is itself represented\n * as a`tf.Tensor`, and can be read with the `read()` method and updated with\n * the `write()` method.\n */\nexport class LayerVariable {\n  readonly dtype: DataType;\n  readonly shape: Shape;\n\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName: string;\n  private trainable_: boolean;\n\n  protected readonly val: tfc.Variable;\n  readonly constraint: Constraint;\n  /**\n   * Construct Variable from a `tf.Tensor`.\n   *\n   * If not explicitly named, the Variable will be given a name with the\n   * prefix 'Variable'. Variable names are unique. In the case of name\n   * collision, suffixies '_<num>' will be added to the name.\n   *\n   * @param val Initial value of the Variable.\n   * @param name Name of the variable. If `null` or `undefined` is provided, it\n   *   will default a name with the prefix 'Variable'.\n   * @param constraint Optional, projection function to be applied to the\n   * variable after optimize updates\n   * @throws ValueError if `name` is `null` or `undefined`.\n   */\n  constructor(\n      val: Tensor, dtype: DataType = 'float32',\n      name = DEFAULT_VARIABLE_NAME_PREFIX, trainable = true,\n      constraint: Constraint = null) {\n    this.dtype = dtype == null ? 'float32' : dtype;\n    this.shape = val.shape;\n    this.id = getNextUniqueTensorId();\n\n    name = name == null ? DEFAULT_VARIABLE_NAME_PREFIX : name;\n    this.originalName = getScopedTensorName(name);\n    this.name = getUniqueTensorName(this.originalName);\n\n    this.trainable_ = trainable;\n    this.constraint = constraint;\n\n    this.val = tfc.variable(val, this.trainable_, this.name, this.dtype);\n  }\n\n  /**\n   * Get a snapshot of the Variable's value.\n   *\n   * The returned value is a snapshot of the Variable's value at the time of\n   * the invocation. Future mutations in the value of the tensor will only\n   * be reflected by future calls to this method.\n   */\n  read(): Tensor {\n    this.assertNotDisposed();\n    return this.val;\n  }\n\n  /**\n   * Update the value of the Variable.\n   *\n   * @param newVal: The new value to update to. Must be consistent with the\n   *   dtype and shape of the Variable.\n   * @return This Variable.\n   */\n  write(newVal: Tensor) {\n    // TODO(cais): Once  TF.js Core supports Tensor.dtype, check dtype match.\n    this.assertNotDisposed();\n    checkShapesMatch(this.val, newVal);\n    // Skip updating if this is the exact same tensor.\n    if (this.val.id !== newVal.id) {\n      this.val.assign(newVal);\n      if (this.constraint != null) {\n        this.val.assign(this.constraint.apply(this.val));\n      }\n    }\n    return this;\n  }\n\n  /**\n   * Dispose this LayersVariable instance from memory.\n   */\n  dispose(): void {\n    this.assertNotDisposed();\n    this.val.dispose();\n  }\n\n  protected assertNotDisposed(): void {\n    if (this.val.isDisposed) {\n      throw new Error(`LayersVariable ${this.name} is already disposed.`);\n    }\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.trainable_ = trainable;\n    this.val.trainable = trainable;\n  }\n}\n\nfunction checkShapesMatch(x: HasShape, y: HasShape): void {\n  if (x.shape.toString() !== y.shape.toString()) {\n    throw new Error(\n        'Shape mismatch: ' + JSON.stringify(x.shape) + ' vs. ' +\n        JSON.stringify(y.shape));\n  }\n}\n\n/**\n * Create a Variable.\n * @param x The initial value of the `Variable`.\n * @param dtype optional, the type of the variable.\n * @param name optional, the name of the variable, default provided by\n * Variable.\n * @param constraint optional, a constraint to be applied after every update.\n * @return The newly instantiated `Variable`.\n */\nexport function variable(\n    x: Tensor, dtype?: DataType, name?: string,\n    constraint?: Constraint): LayerVariable {\n  return new LayerVariable(x, dtype, name, true, constraint);\n}\n\n/**\n * Instantiates an all-zeros Variable and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-zero Variable.\n */\nexport function zerosVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  return new LayerVariable(tfc.zeros(shape), dtype, name);\n}\n\n/**\n * Instantiates an all-zeros tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function zerosLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.zerosLike(x), dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-ones Variable.\n */\nexport function onesVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  const allocated = tfc.ones(shape);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function onesLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  const allocated = tfc.onesLike(x);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiate an identity matrix and returns it, as a Variable\n *\n * @param size Number of rows/columns.\n * @param dtype Data type of returned Variable.\n * @param name Name of returned Variable.\n * @return A Variable, an identity matrix.\n */\nexport function eyeVariable(\n    size: number, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.eye(size), dtype, name);\n}\n\n/**\n * Get a Variable with uniform distribution of values.\n * @param shape Shape of the tensor.\n * @param minval Lower bound of the uniform distribution.\n * @param maxval Upper bound of the uniform distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The uniform-random Variable.\n */\nexport function randomUniformVariable(\n    shape: Shape, minval: number, maxval: number, dtype?: DataType,\n    seed?: number, name = 'randomUniform'): LayerVariable {\n  return new LayerVariable(\n      tfc.randomUniform(shape, minval, maxval, dtype), dtype, name);\n}\n\n/**\n * Get a Variable with truncated-normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function truncatedNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'truncatedNormal'): LayerVariable {\n  // TODO(cais): Implement logic for dtype and seed once they are supported\n  // by deeplearn.js.\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormal does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.truncatedNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n/**\n * Get a Variable with normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function randomNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'randomNormal'): LayerVariable {\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormalVariable does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.randomNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n\n/**\n * Update the value of a Variable.\n * @param x The Variable to be updated.\n * @param xNew The new value to update to.\n * @return The Variable updated.\n */\nexport function update(x: LayerVariable, xNew: Tensor): LayerVariable {\n  return x.write(xNew);\n}\n\n/**\n * Update the value of a Variable by adding an increment.\n * @param x The Variable to be updated.\n * @param increment The incrment to add to `x`.\n * @return The Variable updated.\n */\nexport function updateAdd(x: LayerVariable, increment: Tensor): LayerVariable {\n  return x.write(tfc.add(x.read(), increment));\n}\n\n/**\n * Update the value of a Variable by subtracting a decrement.\n * @param x The Variable to be updated.\n * @param decrement The decrement to subtract from `x`.\n * @return The Variable updated.\n */\nexport function updateSub(x: LayerVariable, decrement: Tensor): LayerVariable {\n  return x.write(tfc.sub(x.read(), decrement));\n}\n\n/**\n * Get the values of an array of Variables.\n *\n * @param tensors An `Array` of `Variable`s to get the values of.\n * @return The values of the inputs, as an `Array` of`tf.Tensor`s.\n */\nexport function batchGetValue(xs: LayerVariable[]): Tensor[] {\n  return xs.map(x => x.read());\n}\n\n/**\n * Update the value of multiple Variables at once.\n *\n * @param variablesAndValues An `Array`, each element is of type\n *   [Variable, Tensor]. The first item is the\n *   `Variable` of which the value is to be updated. The second item\n *   carries the new value.\n */\nexport function batchSetValue(\n    variablesAndValues: Array<[LayerVariable, Tensor]>): void {\n  variablesAndValues.forEach(variableAndValue => {\n    const variable: LayerVariable = variableAndValue[0];\n    variable.write(variableAndValue[1]);\n  });\n}\n\n/**\n * Returns the gradients of `variables` w.r.t. the return value of `lossFn`.\n * @param lossFn A function which returns a Scalar to be used as the function\n *   value (i.e., numerator) for differentiation.\n * @param variables List of variables to be used as the independent variables\n *   (i.e., denominator) for differentiation.\n * @returns An Array of gradients tensors.\n */\nexport function gradients(\n    lossFn: () => tfc.Scalar, variables: LayerVariable[]): Tensor[] {\n  // TODO(cais): The return type signature can be simplified if deeplearn makes\n  //   the corresponding type public.\n  const variableList =\n      variables.map(variable => variable.read() as tfc.Variable);\n  const valudAndGrads = variableGrads(lossFn, variableList);\n  return variables.map(variable => valudAndGrads.grads[variable.name]);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {DataType, Scalar, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId, getUid} from '../backend/state';\nimport {getScopedTensorName, getUniqueTensorName, nameScope} from '../common';\nimport {Constraint} from '../constraints';\nimport {AttributeError, NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {getInitializer, Initializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs, RegularizerFn} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as types_utils from '../utils/types_utils';\nimport * as variable_utils from '../utils/variable_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\n\n// TODO(michaelterry): This is a stub until it's defined.\nexport type Op = (x: LayerVariable) => LayerVariable;\n\n/**\n * Constructor arguments for InputSpec.\n */\nexport interface InputSpecArgs {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n}\n\n/**\n * Specifies the ndim, dtype and shape of every input to a layer.\n *\n * Every layer should expose (if appropriate) an `inputSpec` attribute:\n * a list of instances of InputSpec (one per input tensor).\n *\n * A null entry in a shape is compatible with any dimension,\n * a null shape is compatible with any shape.\n */\nexport class InputSpec {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n\n  constructor(args: InputSpecArgs) {\n    this.dtype = args.dtype;\n    this.shape = args.shape;\n    /*\n      TODO(michaelterry): Could throw error if ndim and shape are both defined\n        (then backport).\n    */\n    if (args.shape != null) {\n      this.ndim = args.shape.length;\n    } else {\n      this.ndim = args.ndim;\n    }\n    this.maxNDim = args.maxNDim;\n    this.minNDim = args.minNDim;\n    this.axes = args.axes || {};\n  }\n}\n\n/**\n * `tf.SymbolicTensor` is a placeholder for a Tensor without any concrete value.\n *\n * They are most often encountered when building a graph of `Layer`s for a\n * a `tf.LayersModel` and the input data's shape, but not values are known.\n *\n * @doc {heading: 'Models', 'subheading': 'Classes'}\n */\nexport class SymbolicTensor {\n  /* A unique ID for the tensor to be able to differentiate tensors. */\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName?: string;\n  /**\n   * Rank/dimensionality of the tensor.\n   */\n  readonly rank: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  nodeIndex: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  tensorIndex: number;\n\n  /**\n   *\n   * @param dtype\n   * @param shape\n   * @param sourceLayer The Layer that produced this symbolic tensor.\n   * @param inputs The inputs passed to sourceLayer's __call__() method.\n   * @param nodeIndex\n   * @param tensorIndex\n   * @param callArgs The keyword arguments passed to the __call__() method.\n   * @param name\n   * @param outputTensorIndex The index of this tensor in the list of outputs\n   *   returned by apply().\n   */\n  constructor(\n      readonly dtype: DataType, readonly shape: Shape,\n      public sourceLayer: Layer, readonly inputs: SymbolicTensor[],\n      readonly callArgs: Kwargs, name?: string,\n      readonly outputTensorIndex?: number) {\n    this.id = getNextUniqueTensorId();\n    if (name != null) {\n      this.originalName = getScopedTensorName(name);\n      this.name = getUniqueTensorName(this.originalName);\n    }\n    this.rank = shape.length;\n  }\n}\n\n/**\n * Constructor arguments for Node.\n */\nexport interface NodeArgs {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`.\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n}\n\n/**\n * The type of the return value of Layer.dispose() and Container.dispose().\n */\nexport interface DisposeResult {\n  /**\n   * Reference count after the dispose call.\n   */\n  refCountAfterDispose: number;\n\n  /**\n   * Number of variables dispose in this dispose call.\n   */\n  numDisposedVariables: number;\n}\n\nlet _nextNodeID = 0;\n\n/**\n * A `Node` describes the connectivity between two layers.\n *\n * Each time a layer is connected to some new input,\n * a node is added to `layer.inboundNodes`.\n *\n * Each time the output of a layer is used by another layer,\n * a node is added to `layer.outboundNodes`.\n *\n * `nodeIndices` and `tensorIndices` are basically fine-grained coordinates\n * describing the origin of the `inputTensors`, verifying the following:\n *\n * `inputTensors[i] ==\n * inboundLayers[i].inboundNodes[nodeIndices[i]].outputTensors[\n *   tensorIndices[i]]`\n *\n * A node from layer A to layer B is added to:\n *     A.outboundNodes\n *     B.inboundNodes\n */\nexport class Node {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n\n  readonly id: number;\n\n  constructor(\n      args: NodeArgs,\n      // TODO(michaelterry): Define actual type for this.\n      public callArgs?: Kwargs) {\n    this.id = _nextNodeID++;\n    /*\n      Layer instance (NOT a list).\n      this is the layer that takes a list of input tensors\n      and turns them into a list of output tensors.\n      the current node will be added to\n      the inboundNodes of outboundLayer.\n    */\n    this.outboundLayer = args.outboundLayer;\n\n    /*\n        The following 3 properties describe where\n        the input tensors come from: which layers,\n        and for each layer, which node and which\n        tensor output of each node.\n    */\n\n    // List of layer instances.\n    this.inboundLayers = args.inboundLayers;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.nodeIndices = args.nodeIndices;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.tensorIndices = args.tensorIndices;\n\n    /*\n        Following 2 properties:\n        tensor inputs and outputs of outboundLayer.\n    */\n\n    // List of tensors. 1:1 mapping with inboundLayers.\n    this.inputTensors = args.inputTensors;\n    // List of tensors, created by outboundLayer.call().\n    this.outputTensors = args.outputTensors;\n\n    /*\n        Following 2 properties: input and output masks.\n        List of tensors, 1:1 mapping with inputTensor.\n    */\n    this.inputMasks = args.inputMasks;\n    // List of tensors, created by outboundLayer.computeMask().\n    this.outputMasks = args.outputMasks;\n\n    // Following 2 properties: input and output shapes.\n\n    // List of shape tuples, shapes of inputTensors.\n    this.inputShapes = args.inputShapes;\n    // List of shape tuples, shapes of outputTensors.\n    this.outputShapes = args.outputShapes;\n\n    // Add nodes to all layers involved.\n    for (const layer of args.inboundLayers) {\n      if (layer != null) {\n        layer.outboundNodes.push(this);\n      }\n    }\n    args.outboundLayer.inboundNodes.push(this);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const inboundNames: string[] = [];\n    for (const layer of this.inboundLayers) {\n      if (layer != null) {\n        inboundNames.push(layer.name);\n      } else {\n        inboundNames.push(null);\n      }\n    }\n    return {\n      outboundLayer: this.outboundLayer ? this.outboundLayer.name : null,\n      inboundLayers: inboundNames,\n      nodeIndices: this.nodeIndices,\n      tensorIndices: this.tensorIndices\n    };\n  }\n}\n\n/** Constructor arguments for Layer. */\nexport declare interface LayerArgs {\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  inputShape?: Shape;\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  batchInputShape?: Shape;\n  /**\n   * If `inputShape` is specified and `batchInputShape` is *not* specified,\n   * `batchSize` is used to construct the `batchInputShape`: `[batchSize,\n   * ...inputShape]`\n   */\n  batchSize?: number;\n  /**\n   * The data-type for this layer. Defaults to 'float32'.\n   * This argument is only applicable to input layers (the first layer of a\n   * model).\n   */\n  dtype?: DataType;\n  /** Name for this layer. */\n  name?: string;\n  /**\n   * Whether the weights of this layer are updatable by `fit`.\n   * Defaults to true.\n   */\n  trainable?: boolean;\n  /**\n   * Initial weight values of the layer.\n   */\n  weights?: Tensor[];\n  /** Legacy support. Do not use for new code. */\n  inputDType?: DataType;\n}\n\n// If necessary, add `output` arguments to the CallHook function.\n// This is currently used for testing only, but may be used for debugger-related\n// purposes in the future.\nexport type CallHook = (inputs: Tensor|Tensor[], kwargs: Kwargs) => void;\n\nlet _nextLayerID = 0;\n\n/**\n * A layer is a grouping of operations and weights that can be composed to\n * create a `tf.LayersModel`.\n *\n * Layers are constructed by using the functions under the\n * [tf.layers](#Layers-Basic) namespace.\n *\n * @doc {heading: 'Layers', subheading: 'Classes', namespace: 'layers'}\n */\nexport abstract class Layer extends serialization.Serializable {\n  /** Name for this layer. Must be unique within a model. */\n  name: string;\n  /**\n   * List of InputSpec class instances.\n   *\n   * Each entry describes one required input:\n   * - ndim\n   * - dtype\n   * A layer with `n` input tensors must have an `inputSpec` of length `n`.\n   */\n  inputSpec: InputSpec[];\n  supportsMasking: boolean;\n  /** Whether the layer weights will be updated during training. */\n  protected trainable_: boolean;\n  batchInputShape: Shape;\n  dtype: DataType;\n  initialWeights: Tensor[];\n\n  inboundNodes: Node[];\n  outboundNodes: Node[];\n\n  activityRegularizer: Regularizer;\n\n  protected _trainableWeights: LayerVariable[];\n  private _nonTrainableWeights: LayerVariable[];\n  private _losses: RegularizerFn[];\n  // TODO(cais): _updates is currently unused.\n  private _updates: Tensor[];\n  private _built: boolean;\n  private _callHook: CallHook = null;\n\n  private _addedWeightNames: string[] = [];\n\n  readonly id: number;\n\n  // Porting Notes: PyKeras does not have this property in this base Layer\n  //   class. Instead lets Layer subclass set it dynamically and checks the\n  //   value with `hasattr`. In tfjs-layers, we let this be a member of this\n  //   base class.\n  protected _stateful = false;\n\n  protected _refCount: number|null;\n\n  // A flag for whether fast (i.e., all-zero) weight initialization is to\n  // be used during `build()` call. This speeds up weight initialization\n  // by saving unnecessary calls to expensive initializers in cases where\n  // the initialized values will be overwritten by loaded weight values\n  // during model loading.\n  private fastWeightInitDuringBuild: boolean;\n\n  constructor(args: LayerArgs = {}) {\n    super();\n    this.id = _nextLayerID++;\n\n    this.activityRegularizer = null;\n\n    this.inputSpec = null;\n    this.supportsMasking = false;\n\n    // These properties will be set upon call of this.build()\n    this._trainableWeights = [];\n    this._nonTrainableWeights = [];\n    this._losses = [];\n    this._updates = [];\n    this._built = false;\n\n    /*\n      These lists will be filled via successive calls\n      to this.addInboundNode().\n     */\n    this.inboundNodes = [];\n    this.outboundNodes = [];\n\n    let name = args.name;\n    if (!name) {\n      const prefix = this.getClassName();\n      name = generic_utils.toSnakeCase(prefix) + '_' + getUid(prefix);\n    }\n    this.name = name;\n\n    this.trainable_ = args.trainable == null ? true : args.trainable;\n\n    if (args.inputShape != null || args.batchInputShape != null) {\n      /*\n        In this case we will later create an input layer\n        to insert before the current layer\n       */\n      let batchInputShape: Shape;\n      if (args.batchInputShape != null) {\n        batchInputShape = args.batchInputShape;\n      } else if (args.inputShape != null) {\n        let batchSize: number = null;\n        if (args.batchSize != null) {\n          batchSize = args.batchSize;\n        }\n        batchInputShape = [batchSize].concat(args.inputShape);\n      }\n      this.batchInputShape = batchInputShape;\n\n      // Set dtype.\n      let dtype = args.dtype;\n      if (dtype == null) {\n        dtype = args.inputDType;\n      }\n      if (dtype == null) {\n        dtype = 'float32';\n      }\n      this.dtype = dtype;\n    }\n\n    if (args.weights != null) {\n      this.initialWeights = args.weights;\n    } else {\n      this.initialWeights = null;\n    }\n\n    // The value of `_refCount` is initialized to null. When the layer is used\n    // in a symbolic way for the first time, it will be set to 1.\n    this._refCount = null;\n\n    this.fastWeightInitDuringBuild = false;\n  }\n\n  /**\n   * Converts a layer and its index to a unique (immutable type) name.\n   * This function is used internally with `this.containerNodes`.\n   * @param layer The layer.\n   * @param nodeIndex The layer's position (e.g. via enumerate) in a list of\n   *   nodes.\n   *\n   * @returns The unique name.\n   */\n  protected static nodeKey(layer: Layer, nodeIndex: number) {\n    return layer.name + '_ib-' + nodeIndex.toString();\n  }\n\n  /**\n   * Returns this.inboundNode at index nodeIndex.\n   *\n   * Porting note: This is a replacement for _get_node_attribute_at_index()\n   * @param nodeIndex\n   * @param attrName The name of the attribute related to request for this node.\n   */\n  private getNodeAtIndex(nodeIndex: number, attrName: string): Node {\n    if (this.inboundNodes.length === 0) {\n      throw new RuntimeError(\n          'The layer has never been called ' +\n          `and thus has no defined ${attrName}.`);\n    }\n    if (this.inboundNodes.length <= nodeIndex) {\n      throw new ValueError(\n          `Asked to get ${attrName} at node ${nodeIndex}, ` +\n          `but the layer has only ${this.inboundNodes.length} inbound nodes.`);\n    }\n    return this.inboundNodes[nodeIndex];\n  }\n\n  /**\n   * Retrieves the input tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple inputs).\n   */\n  getInputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple outputs).\n   */\n  getOutputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'output').outputTensors);\n  }\n\n  // Properties\n\n  /**\n   * Retrieves the input tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Input tensor or list of input tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get input(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer input\" ' +\n          'is ill-defined. ' +\n          'Use `getInputAt(nodeIndex)` instead.');\n    } else if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' is not connected, no input to return.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Output tensor or list of output tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get output(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has no inbound nodes.');\n    }\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer output\" ' +\n          'is ill-defined. ' +\n          'Use `getOutputAt(nodeIndex)` instead.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'output').outputTensors);\n  }\n\n  get losses(): RegularizerFn[] {\n    return this._losses;\n  }\n\n  /**\n   * Retrieves the Layer's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Layer.loss in PyKeras.\n    //   In PyKeras, Layer.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return this.losses.map(lossFn => lossFn());\n  }\n\n  get updates(): Tensor[] {\n    return this._updates;\n  }\n\n  get built(): boolean {\n    return this._built;\n  }\n\n  set built(built: boolean) {\n    this._built = built;\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this._trainableWeights.forEach(w => w.trainable = trainable);\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (this.trainable_) {\n      return this._trainableWeights.filter(w => w.trainable);\n    } else {\n      return [];\n    }\n  }\n\n  set trainableWeights(weights: LayerVariable[]) {\n    this._trainableWeights = weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    if (this.trainable) {\n      return this._trainableWeights.filter(w => !w.trainable)\n          .concat(this._nonTrainableWeights);\n    } else {\n      return this._trainableWeights.concat(this._nonTrainableWeights);\n    }\n  }\n\n  set nonTrainableWeights(weights: LayerVariable[]) {\n    this._nonTrainableWeights = weights;\n  }\n\n  /**\n   * The concatenation of the lists trainableWeights and nonTrainableWeights\n   * (in this order).\n   */\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  get stateful(): boolean {\n    return this._stateful;\n  }\n\n  /**\n   * Reset the states of the layer.\n   *\n   * This method of the base Layer class is essentially a no-op.\n   * Subclasses that are stateful (e.g., stateful RNNs) should override this\n   * method.\n   */\n  resetStates(): void {\n    if (!this.stateful) {\n      throw new Error(\n          'Cannot call the resetStates() method of a non-stateful Layer ' +\n          'object.');\n    }\n  }\n\n  /**\n   * Checks compatibility between the layer and provided inputs.\n   *\n   * This checks that the tensor(s) `input`\n   * verify the input assumptions of the layer\n   * (if any). If not, exceptions are raised.\n   *\n   * @param inputs Input tensor or list of input tensors.\n   *\n   * @exception ValueError in case of mismatch between\n   *   the provided inputs and the expectations of the layer.\n   */\n  protected assertInputCompatibility(inputs: Tensor|Tensor[]|SymbolicTensor|\n                                     SymbolicTensor[]): void {\n    inputs = generic_utils.toList(inputs);\n    if (this.inputSpec == null || this.inputSpec.length === 0) {\n      return;\n    }\n    const inputSpec = generic_utils.toList(this.inputSpec);\n    if (inputs.length !== inputSpec.length) {\n      throw new ValueError(\n          `Layer ${this.name} expects ${inputSpec.length} inputs, ` +\n          `but it received ${inputs.length} input tensors. ` +\n          `Input received: ${inputs}`);\n    }\n    for (let inputIndex = 0; inputIndex < inputs.length; inputIndex++) {\n      const x = inputs[inputIndex];\n      const spec: InputSpec = inputSpec[inputIndex];\n      if (spec == null) {\n        continue;\n      }\n\n      // Check ndim.\n      const ndim = x.rank;\n      if (spec.ndim != null) {\n        if (ndim !== spec.ndim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}: ` +\n              `expected ndim=${spec.ndim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.maxNDim != null) {\n        if (ndim > spec.maxNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected max_ndim=${spec.maxNDim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.minNDim != null) {\n        if (ndim < spec.minNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected min_ndim=${spec.minNDim}, found ndim=${ndim}.`);\n        }\n      }\n\n      // Check dtype.\n      if (spec.dtype != null) {\n        if (x.dtype !== spec.dtype) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name} ` +\n              `: expected dtype=${spec.dtype}, found dtype=${x.dtype}.`);\n        }\n      }\n\n      // Check specific shape axes.\n      if (spec.axes) {\n        const xShape = x.shape;\n        for (const key in spec.axes) {\n          const axis = Number(key);\n          const value = spec.axes[key];\n          // Perform Python-style slicing in case axis < 0;\n          // TODO(cais): Use https://github.com/alvivi/typescript-underscore to\n          // ensure type safety through Underscore calls.\n          const xShapeAtAxis =\n              axis >= 0 ? xShape[axis] : xShape[xShape.length + axis];\n          if (value != null && [value, null].indexOf(xShapeAtAxis) === -1) {\n            throw new ValueError(\n                `Input ${inputIndex} is incompatible with layer ` +\n                `${this.name}: expected axis ${axis} of input shape to ` +\n                `have value ${value} but got shape ${xShape}.`);\n          }\n        }\n      }\n\n      // Check shape.\n      if (spec.shape != null) {\n        for (let i = 0; i < spec.shape.length; ++i) {\n          const specDim = spec.shape[i];\n          const dim = x.shape[i];\n          if (specDim != null && dim != null) {\n            if (specDim !== dim) {\n              throw new ValueError(\n                  `Input ${inputIndex} is incompatible with layer ` +\n                  `${this.name}: expected shape=${spec.shape}, ` +\n                  `found shape=${x.shape}.`);\n            }\n          }\n        }\n      }\n    }\n  }\n\n  /**\n   * This is where the layer's logic lives.\n   *\n   * @param inputs Input tensor, or list/tuple of input tensors.\n   * @param kwargs Additional keyword arguments.\n   *\n   * @return A tensor or list/tuple of tensors.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return inputs;\n  }\n\n  protected invokeCallHook(inputs: Tensor|Tensor[], kwargs: Kwargs) {\n    if (this._callHook != null) {\n      this._callHook(inputs, kwargs);\n    }\n  }\n\n  /**\n   * Set call hook.\n   * This is currently used for testing only.\n   * @param callHook\n   */\n  setCallHook(callHook: CallHook) {\n    this._callHook = callHook;\n  }\n\n  /**\n   * Clear call hook.\n   * This is currently used for testing only.\n   */\n  clearCallHook() {\n    this._callHook = null;\n  }\n\n  /**\n   * Builds or executes a `Layer's logic.\n   *\n   * When called with `tf.Tensor`(s), execute the `Layer`s computation and\n   * return Tensor(s). For example:\n   *\n   * ```js\n   * const denseLayer = tf.layers.dense({\n   *   units: 1,\n   *   kernelInitializer: 'zeros',\n   *   useBias: false\n   * });\n   *\n   * // Invoke the layer's apply() method with a `tf.Tensor` (with concrete\n   * // numeric values).\n   * const input = tf.ones([2, 2]);\n   * const output = denseLayer.apply(input);\n   *\n   * // The output's value is expected to be [[0], [0]], due to the fact that\n   * // the dense layer has a kernel initialized to all-zeros and does not have\n   * // a bias.\n   * output.print();\n   * ```\n   *\n   * When called with `tf.SymbolicTensor`(s), this will prepare the layer for\n   * future execution.  This entails internal book-keeping on shapes of\n   * expected Tensors, wiring layers together, and initializing weights.\n   *\n   * Calling `apply` with `tf.SymbolicTensor`s are typically used during the\n   * building of non-`tf.Sequential` models. For example:\n   *\n   * ```js\n   * const flattenLayer = tf.layers.flatten();\n   * const denseLayer = tf.layers.dense({units: 1});\n   *\n   * // Use tf.layers.input() to obtain a SymbolicTensor as input to apply().\n   * const input = tf.input({shape: [2, 2]});\n   * const output1 = flattenLayer.apply(input);\n   *\n   * // output1.shape is [null, 4]. The first dimension is the undetermined\n   * // batch size. The second dimension comes from flattening the [2, 2]\n   * // shape.\n   * console.log(JSON.stringify(output1.shape));\n   *\n   * // The output SymbolicTensor of the flatten layer can be used to call\n   * // the apply() of the dense layer:\n   * const output2 = denseLayer.apply(output1);\n   *\n   * // output2.shape is [null, 1]. The first dimension is the undetermined\n   * // batch size. The second dimension matches the number of units of the\n   * // dense layer.\n   * console.log(JSON.stringify(output2.shape));\n   *\n   * // The input and output and be used to construct a model that consists\n   * // of the flatten and dense layers.\n   * const model = tf.model({inputs: input, outputs: output2});\n   * ```\n   *\n   * @param inputs a `tf.Tensor` or `tf.SymbolicTensor` or an Array of them.\n   * @param kwargs Additional keyword arguments to be passed to `call()`.\n   *\n   * @return Output of the layer's `call` method.\n   *\n   * @exception ValueError error in case the layer is missing shape information\n   *   for its `build` call.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  // Porting Note: This is a replacement for __call__() in Python.\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    kwargs = kwargs || {};\n\n    this.assertNotDisposed();\n\n    // Ensure inputs are all the same type.\n    const inputsList = generic_utils.toList(inputs);\n\n    let allAreSymbolic = true;\n    for (const input of inputsList) {\n      if (!(input instanceof SymbolicTensor)) {\n        allAreSymbolic = false;\n        break;\n      }\n    }\n    let noneAreSymbolic = true;\n    for (const input of inputsList) {\n      if (input instanceof SymbolicTensor) {\n        noneAreSymbolic = false;\n        break;\n      }\n    }\n\n    if (allAreSymbolic === noneAreSymbolic) {\n      throw new ValueError(\n          'Arguments to apply() must be all ' +\n          'SymbolicTensors or all Tensors');\n    }\n\n    // TODO(michaelterry): nameScope() may not be necessary.\n    return nameScope(this.name, () => {\n      // Handle laying building (weight creating, input spec locking).\n      if (!this.built) {\n        /*\n          Throw exceptions in case the input is not compatible\n          with the inputSpec specified in the layer constructor.\n         */\n        this.assertInputCompatibility(inputs);\n\n        // Collect input shapes to build layer.\n        const inputShapes: Shape[] = [];\n        for (const xElem of generic_utils.toList(inputs)) {\n          inputShapes.push(xElem.shape);\n        }\n        this.build(generic_utils.singletonOrArray(inputShapes));\n        this.built = true;\n\n        // Load weights that were specified at layer instantiation.\n        if (this.initialWeights) {\n          this.setWeights(this.initialWeights);\n        }\n\n        if (this._refCount === null && noneAreSymbolic) {\n          // The first use of this layer is a non-symbolic call, set ref count\n          // to 1 so the Layer can be properly disposed if its dispose() method\n          // is called.\n          this._refCount = 1;\n        }\n      }\n\n      /*\n        Throw exceptions in case the input is not compatible\n        with the inputSpec set at build time.\n      */\n      this.assertInputCompatibility(inputs);\n\n      // Handle mask propagation.\n      // TODO(michaelterry): Mask propagation not currently implemented.\n\n      // Actually call the layer, collecting output(s), mask(s), and shape(s).\n      if (noneAreSymbolic) {\n        let output = this.call(inputs as Tensor | Tensor[], kwargs);\n        // TODO(michaelterry): Compute the outputMask\n\n        // If the layer returns tensors from its inputs, unmodified,\n        // we copy them to avoid loss of tensor metadata.\n        const outputList: Tensor[] = generic_utils.toList(output);\n        const outputListCopy: Tensor[] = [];\n        // TODO(michaelterry): This copying may not be necessary given our eager\n        // backend.\n        for (let x of outputList) {\n          if (inputsList.indexOf(x) !== -1) {\n            x = x.clone();\n          }\n          outputListCopy.push(x);\n        }\n        output = generic_utils.singletonOrArray(outputListCopy);\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        // TODO(michaelterry): Call addInboundNode()?\n        return output;\n      } else {\n        const inputShape = collectInputShape(inputs);\n        const outputShape = this.computeOutputShape(inputShape);\n        let output: SymbolicTensor|SymbolicTensor[];\n        const outputDType = guessOutputDType(inputs);\n        this.warnOnIncompatibleInputShape(\n            Array.isArray(inputs) ? inputShape[0] as Shape :\n                                    inputShape as Shape);\n\n        if (outputShape != null && outputShape.length > 0 &&\n            Array.isArray(outputShape[0])) {\n          // We have multiple output shapes. Create multiple output tensors.\n          output = (outputShape as Shape[])\n                       .map(\n                           (shape, index) => new SymbolicTensor(\n                               outputDType, shape, this,\n                               generic_utils.toList(inputs), kwargs, this.name,\n                               index));\n        } else {\n          output = new SymbolicTensor(\n              outputDType, outputShape as Shape, this,\n              generic_utils.toList(inputs), kwargs, this.name);\n        }\n\n        /*\n          Add an inbound node to the layer, so that it keeps track\n          of the call and of all new variables created during the call.\n          This also updates the layer history of the output tensor(s).\n          If the input tensor(s) had no previous history,\n          this does nothing.\n        */\n        this.addInboundNode(\n            inputs as SymbolicTensor | SymbolicTensor[], output, null, null,\n            inputShape, outputShape, kwargs);\n        this._refCount++;\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        return output;\n      }\n    });\n  }\n\n  /**\n   * Check compatibility between input shape and this layer's batchInputShape.\n   *\n   * Print warning if any incompatibility is found.\n   *\n   * @param inputShape Input shape to be checked.\n   */\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {\n    if (this.batchInputShape == null) {\n      return;\n    } else if (inputShape.length !== this.batchInputShape.length) {\n      console.warn(\n          `The rank of the input tensor provided (shape: ` +\n          `${JSON.stringify(inputShape)}) does not match that of the ` +\n          `batchInputShape (${JSON.stringify(this.batchInputShape)}) ` +\n          `of the layer ${this.name}`);\n    } else {\n      let dimMismatch = false;\n      this.batchInputShape.forEach((dimension, i) => {\n        if (dimension != null && inputShape[i] != null &&\n            inputShape[i] !== dimension) {\n          dimMismatch = true;\n        }\n      });\n      if (dimMismatch) {\n        console.warn(\n            `The shape of the input tensor ` +\n            `(${JSON.stringify(inputShape)}) does not ` +\n            `match the expectation of layer ${this.name}: ` +\n            `${JSON.stringify(this.batchInputShape)}`);\n      }\n    }\n  }\n\n  /**\n   * Retrieves the output shape(s) of a layer.\n   *\n   * Only applicable if the layer has only one inbound node, or if all inbound\n   * nodes have the same output shape.\n   *\n   * @returns Output shape or shapes.\n   * @throws AttributeError: if the layer is connected to more than one incoming\n   *   nodes.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  get outputShape(): Shape|Shape[] {\n    if (this.inboundNodes == null || this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `The layer ${this.name} has never been called and thus has no ` +\n          `defined output shape.`);\n    }\n    const allOutputShapes: string[] = [];\n    for (const node of this.inboundNodes) {\n      const shapeString = JSON.stringify(node.outputShapes);\n      if (allOutputShapes.indexOf(shapeString) === -1) {\n        allOutputShapes.push(shapeString);\n      }\n    }\n    if (allOutputShapes.length === 1) {\n      const outputShapes = this.inboundNodes[0].outputShapes;\n      if (Array.isArray(outputShapes) && Array.isArray(outputShapes[0]) &&\n          outputShapes.length === 1) {\n        return (outputShapes as Shape[])[0];\n      } else {\n        return outputShapes;\n      }\n\n    } else {\n      throw new AttributeError(\n          `The layer ${this.name} has multiple inbound nodes with different ` +\n          `output shapes. Hence the notion of \"output shape\" is ill-defined ` +\n          `for the layer.`);\n      // TODO(cais): Implement getOutputShapeAt().\n    }\n  }\n\n  /**\n   * Counts the total number of numbers (e.g., float32, int32) in the\n   * weights.\n   *\n   * @returns An integer count.\n   * @throws RuntimeError: If the layer is not built yet (in which case its\n   *   weights are not defined yet.)\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  countParams(): number {\n    if (!this.built) {\n      throw new RuntimeError(\n          `You tried to call countParams() on ${this.name}, ` +\n          `but the layer is not built yet. Build it first by calling ` +\n          `build(batchInputShape).`);\n    }\n    return variable_utils.countParamsInWeights(this.weights);\n  }\n\n  /**\n   * Creates the layer weights.\n   *\n   * Must be implemented on all layers that have weights.\n   *\n   * Called when apply() is called to construct the weights.\n   *\n   * @param inputShape A `Shape` or array of `Shape` (unused).\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  build(inputShape: Shape|Shape[]) {\n    this.built = true;\n  }\n\n  /**\n   * Returns the current values of the weights of the layer.\n   *\n   * @param trainableOnly Whether to get the values of only trainable weights.\n   * @returns Weight values as an `Array` of `tf.Tensor`s.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getWeights(trainableOnly = false): Tensor[] {\n    return batchGetValue(trainableOnly ? this.trainableWeights : this.weights);\n  }\n\n  /**\n   * Sets the weights of the layer, from Tensors.\n   *\n   * @param weights a list of Tensors. The number of arrays and their shape\n   *   must match number of the dimensions of the weights of the layer (i.e.\n   *   it should match the output of `getWeights`).\n   *\n   * @exception ValueError If the provided weights list does not match the\n   *   layer's specifications.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  setWeights(weights: Tensor[]): void {\n    tidy(() => {\n      const params = this.weights;\n      if (params.length !== weights.length) {\n        // TODO(cais): Restore the following and use `providedWeights`, instead\n        // of `weights` in the error message, once the deeplearn.js bug is\n        // fixed: https://github.com/PAIR-code/deeplearnjs/issues/498 const\n        // providedWeights = JSON.stringify(weights).substr(0, 50);\n        throw new ValueError(\n            `You called setWeights(weights) on layer \"${this.name}\" ` +\n            `with a weight list of length ${weights.length}, ` +\n            `but the layer was expecting ${params.length} weights. ` +\n            `Provided weights: ${weights}...`);\n      }\n      if (params.length === 0) {\n        return;\n      }\n      const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n      const paramValues = batchGetValue(params);\n      for (let i = 0; i < paramValues.length; ++i) {\n        const pv = paramValues[i];\n        const p = params[i];\n        const w = weights[i];\n        if (!util.arraysEqual(pv.shape, w.shape)) {\n          throw new ValueError(\n              `Layer weight shape ${pv.shape} ` +\n              `not compatible with provided weight shape ${w.shape}`);\n        }\n        weightValueTuples.push([p, w]);\n      }\n      batchSetValue(weightValueTuples);\n    });\n  }\n\n  /**\n   * Adds a weight variable to the layer.\n   *\n   * @param name Name of the new weight variable.\n   * @param shape The shape of the weight.\n   * @param dtype The dtype of the weight.\n   * @param initializer An initializer instance.\n   * @param regularizer A regularizer instance.\n   * @param trainable Whether the weight should be trained via backprop or not\n   *   (assuming that the layer itself is also trainable).\n   * @param constraint An optional trainable.\n   * @return The created weight variable.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  protected addWeight(\n      name: string, shape: Shape, dtype?: DataType, initializer?: Initializer,\n      regularizer?: Regularizer, trainable?: boolean,\n      constraint?: Constraint): LayerVariable {\n    // Reject duplicate weight names.\n    if (this._addedWeightNames.indexOf(name) !== -1) {\n      throw new ValueError(\n          `Duplicate weight name ${name} for layer ${this.name}`);\n    }\n    this._addedWeightNames.push(name);\n\n    if (dtype == null) {\n      dtype = 'float32';\n    }\n\n    if (this.fastWeightInitDuringBuild) {\n      initializer = getInitializer('zeros');\n    }\n    const initValue = initializer.apply(shape, dtype);\n    const weight =\n        new LayerVariable(initValue, dtype, name, trainable, constraint);\n    initValue.dispose();\n    // Request backend not to dispose the weights of the model on scope() exit.\n    if (regularizer != null) {\n      this.addLoss(() => regularizer.apply(weight.read()));\n    }\n    if (trainable == null) {\n      trainable = true;\n    }\n    if (trainable) {\n      this._trainableWeights.push(weight);\n    } else {\n      this._nonTrainableWeights.push(weight);\n    }\n    return weight;\n  }\n\n  /**\n   * Set the fast-weight-initialization flag.\n   *\n   * In cases where the initialized weight values will be immediately\n   * overwritten by loaded weight values during model loading, setting\n   * the flag to `true` saves unnecessary calls to potentially expensive\n   * initializers and speeds up the loading process.\n   *\n   * @param value Target value of the flag.\n   */\n  setFastWeightInitDuringBuild(value: boolean) {\n    this.fastWeightInitDuringBuild = value;\n  }\n\n  /**\n   * Add losses to the layer.\n   *\n   * The loss may potentionally be conditional on some inputs tensors,\n   * for instance activity losses are conditional on the layer's inputs.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  addLoss(losses: RegularizerFn|RegularizerFn[]): void {\n    if (losses == null || Array.isArray(losses) && losses.length === 0) {\n      return;\n    }\n    // Update this.losses\n    losses = generic_utils.toList(losses);\n    if (this._losses !== undefined && this._losses !== null) {\n      this.losses.push(...losses);\n    }\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (!this.supportsMasking) {\n      if (mask != null) {\n        if (Array.isArray(mask)) {\n          mask.forEach(maskElement => {\n            if (maskElement != null) {\n              throw new TypeError(\n                  `Layer ${this.name} does not support masking, ` +\n                  'but was passed an inputMask.');\n            }\n          });\n        } else {\n          throw new TypeError(\n              `Layer ${this.name} does not support masking, ` +\n              'but was passed an inputMask.');\n        }\n      }\n      // masking not explicitly supported: return null as mask\n      return null;\n    }\n    // if masking is explictly supported, by default\n    // carry over the input mask\n    return mask;\n  }\n\n  /**\n   * Internal method to create an inbound node for the layer.\n   *\n   * @param inputTensors List of input tensors.\n   * @param outputTensors List of output tensors.\n   * @param inputMasks List of input masks (a mask can be a tensor, or null).\n   * @param outputMasks List of output masks (a mask can be a tensor, or null).\n   * @param inputShapes List of input shape tuples.\n   * @param outputShapes List of output shape tuples.\n   * @param kwargs Dictionary of keyword arguments that were passed to the\n   *   `call` method of the layer at the call that created the node.\n   */\n  private addInboundNode(\n      inputTensors: SymbolicTensor|SymbolicTensor[],\n      outputTensors: SymbolicTensor|SymbolicTensor[],\n      inputMasks: Tensor|Tensor[], outputMasks: Tensor|Tensor[],\n      inputShapes: Shape|Shape[], outputShapes: Shape|Shape[],\n      kwargs: {} = null): void {\n    const inputTensorList: SymbolicTensor[] =\n        generic_utils.toList(inputTensors);\n    outputTensors = generic_utils.toList(outputTensors);\n    inputMasks = generic_utils.toList(inputMasks);\n    outputMasks = generic_utils.toList(outputMasks);\n    inputShapes = types_utils.normalizeShapeList(inputShapes);\n    outputShapes = types_utils.normalizeShapeList(outputShapes);\n\n    // Collect input tensor(s) coordinates.\n    const inboundLayers: Layer[] = [];\n    const nodeIndices: number[] = [];\n    const tensorIndices: number[] = [];\n    for (const x of inputTensorList) {\n      /*\n       * TODO(michaelterry): Keras adds this value to tensors; it's not\n       * clear whether we'll use this or not.\n       */\n      inboundLayers.push(x.sourceLayer);\n      nodeIndices.push(x.nodeIndex);\n      tensorIndices.push(x.tensorIndex);\n    }\n\n    // Create node, add it to inbound nodes.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node(\n        {\n          outboundLayer: this,\n          inboundLayers,\n          nodeIndices,\n          tensorIndices,\n          inputTensors: inputTensorList,\n          outputTensors,\n          inputMasks,\n          outputMasks,\n          inputShapes,\n          outputShapes\n        },\n        kwargs);\n\n    // Update tensor history\n    for (let i = 0; i < outputTensors.length; i++) {\n      // TODO(michaelterry: _uses_learning_phase not tracked.\n      outputTensors[i].sourceLayer = this;\n      outputTensors[i].nodeIndex = this.inboundNodes.length - 1;\n      outputTensors[i].tensorIndex = i;\n    }\n  }\n\n  /**\n   * Returns the config of the layer.\n   *\n   * A layer config is a TS dictionary (serializable)\n   * containing the configuration of a layer.\n   * The same layer can be reinstantiated later\n   * (without its trained weights) from this configuration.\n   *\n   * The config of a layer does not include connectivity\n   * information, nor the layer class name.  These are handled\n   * by 'Container' (one layer of abstraction above).\n   *\n   * Porting Note: The TS dictionary follows TS naming standrds for\n   * keys, and uses tfjs-layers type-safe Enums.  Serialization methods\n   * should use a helper function to convert to the pythonic storage\n   * standard. (see serialization_utils.convertTsToPythonic)\n   *\n   * @returns TS dictionary of configuration.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getConfig(): serialization.ConfigDict {\n    const config:\n        serialization.ConfigDict = {name: this.name, trainable: this.trainable};\n    if (this.batchInputShape != null) {\n      config['batchInputShape'] = this.batchInputShape;\n    }\n    if (this.dtype != null) {\n      config['dtype'] = this.dtype;\n    }\n    return config;\n  }\n\n  /**\n   * Dispose the weight variables that this Layer instance holds.\n   *\n   * @returns {number} Number of disposed variables.\n   */\n  protected disposeWeights(): number {\n    this.weights.forEach(weight => weight.dispose());\n    return this.weights.length;\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Layer '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose layer's weights.\n   *\n   * This method decrease the reference count of the Layer object by 1.\n   *\n   * A Layer is reference-counted. Its reference count is incremented by 1\n   * the first item its `apply()` method is called and when it becomes a part\n   * of a new `Node` (through calling the `apply()`) method on a\n   * `tf.SymbolicTensor`).\n   *\n   * If the reference count of a Layer becomes 0, all the weights will be\n   * disposed and the underlying memory (e.g., the textures allocated in WebGL)\n   * will be freed.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * weights of the Layer will *not* be disposed.\n   *\n   * After a Layer is disposed, it cannot be used in calls such as `apply()`,\n   * `getWeights()` or `setWeights()` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the Container after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the layer has already\n   *   been disposed.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  dispose(): DisposeResult {\n    if (!this.built) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been ` +\n          `built yet.`);\n    }\n\n    if (this._refCount === null) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been used ` +\n          `yet.`);\n    }\n\n    this.assertNotDisposed();\n\n    let numDisposedVariables = 0;\n    if (--this._refCount === 0) {\n      numDisposedVariables = this.disposeWeights();\n    }\n\n    return {refCountAfterDispose: this._refCount, numDisposedVariables};\n  }\n}\n\n/**\n * Collects the input shape(s) of a list of `tf.Tensor`s or\n * `tf.SymbolicTensor`s.\n *\n * TODO(michaelterry): Update PyKeras docs (backport).\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return List of shape tuples (or single tuple), one tuple per input.\n */\nfunction collectInputShape(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                           Tensor[]): Shape|Shape[] {\n  inputTensors =\n      generic_utils.toList(inputTensors) as SymbolicTensor[] | Tensor[];\n  const shapes: Shape[] = [];\n  for (const x of inputTensors) {\n    shapes.push(x.shape);\n  }\n  return generic_utils.singletonOrArray(shapes);\n}\n\n/**\n * Guesses output dtype based on inputs.\n *\n * At present, just returns 'float32' for any input.\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return The guessed DType. At present, always returns 'float32'.\n */\nfunction guessOutputDType(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                          Tensor[]): DataType {\n  return 'float32';\n}\n\n/**\n * Returns the list of input tensors necessary to compute `tensor`.\n *\n * Output will always be a list of tensors (potentially with 1 element).\n *\n * @param tensor The tensor to start from.\n * @param layer Origin layer of the tensor.\n * @param nodeIndex Origin node index of the tensor.\n *\n * @return Array of input tensors.\n */\nexport function getSourceInputs(\n    tensor: SymbolicTensor, layer?: Layer,\n    nodeIndex?: number): SymbolicTensor[] {\n  if (layer == null || (nodeIndex != null && nodeIndex > 0)) {\n    layer = tensor.sourceLayer;\n    nodeIndex = tensor.nodeIndex;\n  }\n  if (layer.inboundNodes.length === 0) {\n    return [tensor];\n  } else {\n    const node = layer.inboundNodes[nodeIndex];\n    if (node.inboundLayers.length === 0) {\n      return node.inputTensors;\n    } else {\n      const sourceTensors: SymbolicTensor[] = [];\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const x = node.inputTensors[i];\n        const layer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const previousSources = getSourceInputs(x, layer, nodeIndex);\n        // Avoid input redundancy.\n        for (const x of previousSources) {\n          if (sourceTensors.indexOf(x) === -1) {\n            sourceTensors.push(x);\n          }\n        }\n      }\n      return sourceTensors;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\n\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/**\n * Constructor arguments for InputLayer.\n *\n * Note: You should provide only inputShape or batchInputShape (not both).\n * If only inputShape is provided, then the batchInputShape is determined by\n * the batchSize argument and the inputShape: [batchSize].concat(inputShape).\n */\nexport declare interface InputLayerArgs {\n  /** Input shape, not including the batch axis. */\n  inputShape?: Shape;\n  /** Optional input batch size (integer or null). */\n  batchSize?: number;\n  /** Batch input shape, including the batch axis. */\n  batchInputShape?: Shape;\n  /** Datatype of the input.  */\n  dtype?: DataType;\n  /**\n   * Whether the placeholder created is meant to be sparse.\n   */\n  sparse?: boolean;  // TODO(michaelterry): Not clear whether we'll need this.\n\n  /** Name of the layer. */\n  name?: string;\n}\n\nexport class InputLayer extends Layer {\n  /** @nocollapse */\n  static readonly className = 'InputLayer';\n  sparse: boolean;\n  constructor(args: InputLayerArgs) {\n    super({\n      dtype: args.dtype,\n      name: args.name != null ? args.name : getUid('input').toString()\n    });\n    // Normalize config.batchSize and config.sparse\n    if (args.batchSize == null) {\n      args.batchSize = null;\n    }\n    if (args.sparse == null) {\n      args.sparse = false;\n    }\n\n    this.trainable = false;\n    this.built = true;\n    this.sparse = args.sparse;\n\n    if (args.inputShape != null && args.batchInputShape != null) {\n      throw new ValueError(\n          'Only provide the inputShape OR ' +\n          'batchInputShape argument to inputLayer, not both at the same time.');\n    }\n    let batchInputShape = args.batchInputShape;\n    if (batchInputShape == null) {\n      if (args.inputShape == null) {\n        throw new ValueError(\n            'An InputLayer should be passed either a ' +\n            '`batchInputShape` or an `inputShape`.');\n      } else {\n        batchInputShape = [args.batchSize].concat(args.inputShape);\n      }\n    } else {\n      // TODO(michaelterry): Backport to PyKeras\n      if (args.batchSize != null) {\n        throw new ValueError(\n            'Cannot specify batchSize if batchInputShape is ' +\n            'specified when creating an InputLayer.');\n      }\n    }\n\n    const dtype = args.dtype || 'float32';\n\n    this.batchInputShape = batchInputShape;\n    this.dtype = dtype;\n    // TODO(michaelterry): Backport this to PyKeras?\n    this.inputSpec = [{shape: batchInputShape}];\n\n    const inputTensor = new SymbolicTensor(\n        this.dtype, this.batchInputShape, this, [], {}, this.name);\n    inputTensor.nodeIndex = 0;\n    inputTensor.tensorIndex = 0;\n\n    // Create an input node to add to this.outboundNode.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: [inputTensor],\n      outputTensors: [inputTensor],\n      inputMasks: [null],\n      outputMasks: [null],\n      inputShapes: [batchInputShape],\n      outputShapes: [batchInputShape]\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor {\n    throw new ValueError(\n        'Cannot pass any input to an ' +\n        `InputLayer's apply() method. InputLayer name: ${this.name}`);\n  }\n\n  dispose(): DisposeResult {\n    // dispose() for InputLayer is overridden as no-op.\n    return {refCountAfterDispose: this._refCount, numDisposedVariables: 0};\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      batchInputShape: this.batchInputShape,\n      dtype: this.dtype,\n      sparse: this.sparse,\n      name: this.name\n    };\n  }\n}\nserialization.registerClass(InputLayer);\n\n/**\n * Config for the Input function.\n *\n * Note: You should provide only shape or batchShape (not both).\n * If only shape is provided, then the batchShape becomes\n * [null].concat(inputShape).\n */\nexport interface InputConfig {\n  /**\n   * A shape, not including the batch size. For instance, `shape=[32]`\n   * indicates that the expected input will be batches of 32-dimensional\n   * vectors.\n   */\n  shape?: Shape;\n  /**\n   * A shape tuple (integer), including the batch size. For instance,\n   * `batchShape=[10, 32]` indicates that the expected input will be batches of\n   * 10 32-dimensional vectors. `batchShape=[null, 32]` indicates batches of an\n   * arbitrary number of 32-dimensional vectors.\n   */\n  batchShape?: Shape;\n  /**\n   * An optional name string for the layer. Should be unique in a model (do not\n   * reuse the same name twice). It will be autogenerated if it isn't provided.\n   */\n  name?: string;\n  dtype?: DataType;\n  /**\n   * A boolean specifying whether the placeholder to be created is sparse.\n   */\n  sparse?: boolean;\n}\n\nexport function Input(config: InputConfig): SymbolicTensor {\n  if (config.batchShape == null && config.shape == null) {\n    throw new Error(\n        'Please provide to Input either a `shape`' +\n        ' or a `batchShape` argument. Note that ' +\n        '`shape` does not include the batch ' +\n        'dimension.');\n  }\n  if (config.batchShape != null && config.shape != null) {\n    // TODO(michaelterry): Backport to PyKeras.\n    throw new ValueError(\n        'Please provide either a `shape` or `batchShape` ' +\n        'argument to Input, but not both.');\n  }\n  let batchShape = config.batchShape;\n  if (config.shape != null && batchShape == null) {\n    batchShape = [null].concat(config.shape);\n  }\n\n  let dtype = config.dtype;\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n\n  const inputLayer = new InputLayer({\n    batchInputShape: batchShape,\n    name: config.name,\n    dtype,\n    sparse: config.sparse\n  });\n\n  const outputs = inputLayer.inboundNodes[0].outputTensors;\n  return outputs[0];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, Scalar} from '@tensorflow/tfjs-core';\n\n/**\n * Logs in which values can be either numbers or Tensors (Scalars).\n *\n * Used internally.\n */\nexport type UnresolvedLogs = {\n  [key: string]: number|Scalar;\n};\n\n/**\n * Turn any Scalar values in a Logs object into actual number values.\n *\n * @param logs The `Logs` object to be resolved in place.\n */\nexport async function resolveScalarsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n  const keys: string[] = [];\n  const scalarsToDispose: Scalar[] = [];\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      const valueScalar = value;\n      promises.push(valueScalar.data());\n      keys.push(key);\n      scalarsToDispose.push(valueScalar);\n    }\n  }\n  if (promises.length > 0) {\n    const values = await Promise.all(promises);\n    for (let i = 0; i < values.length; ++i) {\n      logs[keys[i]] = values[i][0];\n    }\n    // Dispose the original scalar tensors.\n    dispose(scalarsToDispose);\n  }\n}\n\n/**\n * Dispose all Tensors in an UnresolvedLogs object.\n *\n * @param logs An `UnresolvedLogs` object potentially containing `tf.Tensor`s in\n *   places where the values can be `tf.Tensor` or `number`.\n */\nexport function disposeTensorsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      value.dispose();\n    }\n  }\n}\n\n/**\n * Logs in which values can only be numbers.\n *\n * Used when calling client-provided custom callbacks.\n */\nexport type Logs = {\n  [key: string]: number;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {add, div, keep, mul, nextFrame, Scalar, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Container} from './engine/container';\nimport {ValueError} from './errors';\nimport {Logs, resolveScalarsInLogs, UnresolvedLogs} from './logs';\nimport * as generic_utils from './utils/generic_utils';\n\n/** Verbosity logging level when fitting a model. */\nexport enum ModelLoggingVerbosity {\n  SILENT = 0,\n  VERBOSE = 1\n}\n\n/** How often to yield to the main thread when training (in ms). */\nexport const DEFAULT_YIELD_EVERY_MS = 125;\n\nexport type Params = {\n  [key: string]: number|string|boolean|number[]|string[]|boolean[];\n};\n\nexport type YieldEveryOptions = 'auto'|'batch'|'epoch'|'never'|number;\n\n/**\n * Abstract base class used to build new callbacks.\n *\n * The `logs` dictionary that callback methods take as argument will contain\n * keys for quantities relevant to the current batch or epoch.\n *\n * Currently, the `.fit()` method of the `Sequential` model class\n * will include the following quantities in the `logs` that\n * it passes to its callbacks:\n *\n * onEpochEnd: Logs include `acc` and `loss`, and optionally include `valLoss`\n *   (if validation is enabled in `fit`), and `valAcc` (if validation and\n *   accuracy monitoring are enabled).\n * onBatchBegin: Logs include `size`, the number of samples in the current\n *   batch.\n * onBatchEnd: Logs include `loss`, and optionally `acc` (if accuracy monitoring\n *   is enabled).\n */\nexport abstract class BaseCallback {\n  // TODO(michaelterry): This type is a best guess.\n  validationData: Tensor|Tensor[] = null;\n  /**\n   * Training parameters (eg. verbosity, batch size, number of epochs...).\n   */\n  params: Params;\n\n  setParams(params: Params): void {\n    this.params = params;\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {}\n\n  async onTrainBegin(logs?: UnresolvedLogs) {}\n\n  async onTrainEnd(logs?: UnresolvedLogs) {}\n\n  // LayersModel needs to call Callback.setModel(), but cannot actually depend\n  // on Callback because that creates a cyclic dependency.  Providing this no-op\n  // method on BaseCallback breaks the cycle: this way LayersModel can depend on\n  // BaseCallback but not on Callback.  The argument is typed as `Container`\n  // (the superclass of LayersModel) to avoid recapitulating the cycle. Callback\n  // overrides this method and enforces that the argument is really a\n  // LayersModel.\n  setModel(model: Container): void {\n    // Do nothing. Use Callback instead of BaseCallback to track the model.\n  }\n}\n\n/**\n * Container abstracting a list of callbacks.\n */\nexport class CallbackList {\n  callbacks: BaseCallback[];\n  queueLength: number;\n\n  // TODO(cais): When the need arises, uncomment the following lines and\n  // implement the queue for time values.\n  // private deltaTBatch: number;\n  // private deltaTsBatchBegin: Array<number>;\n  // private deltaTsBatchEnd: Array<number>;\n\n  /**\n   * Constructor of CallbackList.\n   * @param callbacks Array of `Callback` instances.\n   * @param queueLength Queue length for keeping running statistics over\n   *   callback execution time.\n   */\n  constructor(callbacks?: BaseCallback[], queueLength = 10) {\n    // TODO(cais): Make use of queueLength when implementing the queue for time\n    // values.\n    if (callbacks == null) {\n      callbacks = [];\n    }\n    this.callbacks = callbacks;\n    this.queueLength = queueLength;\n  }\n\n  append(callback: BaseCallback): void {\n    this.callbacks.push(callback);\n  }\n\n  setParams(params: Params): void {\n    for (const callback of this.callbacks) {\n      callback.setParams(params);\n    }\n  }\n\n  setModel(model: Container): void {\n    for (const callback of this.callbacks) {\n      callback.setModel(model);\n    }\n  }\n\n  /**\n   * Called at the start of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochBegin(epoch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochEnd(epoch, logs);\n    }\n  }\n\n  /**\n   * Called  right before processing a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchBegin(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchEnd(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the beginning of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainBegin(logs);\n    }\n  }\n\n  /**\n   * Called at the end of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainEnd(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainEnd(logs);\n    }\n  }\n}\n\n/**\n * Callback that accumulates epoch averages of metrics.\n *\n * This callback is automatically applied to every LayersModel.\n */\nexport class BaseLogger extends BaseCallback {\n  private seen: number;\n  private totals: UnresolvedLogs;\n\n  constructor() {\n    super();\n  }\n\n  async onEpochBegin(epoch: number) {\n    this.seen = 0;\n    this.totals = {};\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const batchSize = logs['size'] == null ? 0 : logs['size'] as number;\n    this.seen += batchSize;\n    for (const key in logs) {\n      const value = logs[key];\n      if (typeof value === 'number') {\n        if (!this.totals.hasOwnProperty(key)) {\n          this.totals[key] = 0;\n        }\n        this.totals[key] = this.totals[key] as number + value * batchSize;\n      } else {\n        let oldTotalsToDispose: Scalar;\n        if (key in this.totals) {\n          oldTotalsToDispose = this.totals[key] as Scalar;\n        } else {\n          this.totals[key] = 0;\n        }\n        const total: Scalar =\n            tidy(() => add((this.totals[key]), mul(value, batchSize)));\n        this.totals[key] = total;\n        if (oldTotalsToDispose != null) {\n          oldTotalsToDispose.dispose();\n        }\n      }\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs != null) {\n      for (const key of this.params['metrics'] as string[]) {\n        if (this.totals[key] == null) {\n          continue;\n        }\n        if (typeof this.totals[key] === 'number') {\n          logs[key] = this.totals[key] as number / this.seen;\n        } else {\n          tidy(() => {\n            const log: Scalar = mul(div(1, this.seen), this.totals[key]);\n            logs[key] = log;\n            (this.totals[key] as Tensor).dispose();\n            keep(logs[key] as Scalar);\n          });\n        }\n      }\n    }\n  }\n}\n\n/**\n * Callback that records events into a `History` object. This callback is\n * automatically applied to every TF.js Layers model. The `History` object\n * gets returned by the `fit` method of models.\n */\nexport class History extends BaseCallback {\n  epoch: number[];\n  history: {[key: string]: Array<number|Tensor>};\n\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    this.epoch = [];\n    this.history = {};\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    this.epoch.push(epoch);\n    for (const key in logs) {\n      if (this.history[key] == null) {\n        this.history[key] = [];\n      }\n      this.history[key].push(logs[key]);\n    }\n  }\n\n  /**\n   * Await the values of all losses and metrics.\n   */\n  async syncData() {\n    const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n    const keys: string[] = [];\n    const indices: number[] = [];\n    for (const key in this.history) {\n      const valueArray = this.history[key];\n      for (let i = 0; i < valueArray.length; ++i) {\n        if (typeof valueArray[i] !== 'number') {\n          const valueScalar = valueArray[i] as Tensor;\n          promises.push(valueScalar.data());\n          keys.push(key);\n          indices.push(i);\n        }\n      }\n    }\n    const values = await Promise.all(promises);\n    for (let n = 0; n < values.length; ++n) {\n      const tensorToDispose = this.history[keys[n]][indices[n]] as Tensor;\n      tensorToDispose.dispose();\n      this.history[keys[n]][indices[n]] = values[n][0];\n    }\n  }\n}\n\nexport interface CustomCallbackArgs {\n  onTrainBegin?: (logs?: Logs) => void | Promise<void>;\n  onTrainEnd?: (logs?: Logs) => void | Promise<void>;\n  onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n}\n\n/**\n * Custom callback for training.\n */\nexport class CustomCallback extends BaseCallback {\n  protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;\n  protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;\n  protected readonly epochBegin:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly epochEnd:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchBegin:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchEnd:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly yield:\n      (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n\n  private yieldEvery: YieldEveryOptions;\n  private currentEpoch = 0;\n\n  constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions) {\n    super();\n    this.yieldEvery = yieldEvery || 'auto';\n    if (this.yieldEvery === 'auto') {\n      this.yieldEvery = DEFAULT_YIELD_EVERY_MS;\n    }\n    if (this.yieldEvery === 'never' && args.onYield != null) {\n      throw new Error(\n          'yieldEvery is `never` but you provided an `onYield` callback. ' +\n          'Either change `yieldEvery` or remove the callback');\n    }\n    if (util.isNumber(this.yieldEvery)) {\n      // Decorate `maybeWait` so it will be called at most once every\n      // `yieldEvery` ms.\n      this.maybeWait = generic_utils.debounce(\n          this.maybeWait.bind(this), this.yieldEvery as number);\n    }\n    this.trainBegin = args.onTrainBegin;\n    this.trainEnd = args.onTrainEnd;\n    this.epochBegin = args.onEpochBegin;\n    this.epochEnd = args.onEpochEnd;\n    this.batchBegin = args.onBatchBegin;\n    this.batchEnd = args.onBatchEnd;\n    this.yield = args.onYield;\n  }\n\n  async maybeWait(epoch: number, batch: number, logs: UnresolvedLogs) {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.yield != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.yield(epoch, batch, logs as Logs));\n    }\n    ps.push(nextFrame());\n    await Promise.all(ps);\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    this.currentEpoch = epoch;\n    if (this.epochBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.epochBegin(epoch, logs as Logs);\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.epochEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.epochEnd(epoch, logs as Logs));\n    }\n    if (this.yieldEvery === 'epoch') {\n      ps.push(nextFrame());\n    }\n    await Promise.all(ps);\n  }\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    if (this.batchBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.batchBegin(batch, logs as Logs);\n    }\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.batchEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.batchEnd(batch, logs as Logs));\n    }\n    if (this.yieldEvery === 'batch') {\n      ps.push(nextFrame());\n    } else if (util.isNumber(this.yieldEvery)) {\n      ps.push(this.maybeWait(this.currentEpoch, batch, logs));\n    }\n    await Promise.all(ps);\n  }\n\n  async onTrainBegin(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainBegin(logs as Logs);\n    }\n  }\n\n  async onTrainEnd(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainEnd != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainEnd(logs as Logs);\n    }\n  }\n}\n\n/**\n * Standardize callbacks or configurations of them to an Array of callbacks.\n */\nexport function standardizeCallbacks(\n    callbacks: BaseCallback|BaseCallback[]|CustomCallbackArgs|\n    CustomCallbackArgs[],\n    yieldEvery: YieldEveryOptions): BaseCallback[] {\n  if (callbacks == null) {\n    callbacks = {} as BaseCallback;\n  }\n  if (callbacks instanceof BaseCallback) {\n    return [callbacks];\n  }\n  if (Array.isArray(callbacks) && callbacks[0] instanceof BaseCallback) {\n    return callbacks as BaseCallback[];\n  }\n  // Convert custom callback configs to custom callback objects.\n  const callbackConfigs =\n      generic_utils.toList(callbacks) as CustomCallbackArgs[];\n  return callbackConfigs.map(\n      callbackConfig => new CustomCallback(callbackConfig, yieldEvery));\n}\n\nexport declare type BaseCallbackConstructor = {\n  new (): BaseCallback\n};\n\n/**\n * A global registry for callback constructors to be used during\n * LayersModel.fit().\n */\nexport class CallbackConstructorRegistry {\n  private static constructors:\n      {[verbosityLevel: number]: BaseCallbackConstructor[]} = {};\n\n  /**\n   * Blocks public access to constructor.\n   */\n  private constructor() {}\n\n  /**\n   * Register a tf.LayersModel.fit() callback constructor.\n   *\n   * The registered callback constructor will be used to instantiate\n   * callbacks for every tf.LayersModel.fit() call afterwards.\n   *\n   * @param verbosityLevel Level of verbosity at which the `callbackConstructor`\n   *   is to be reigstered.\n   * @param callbackConstructor A no-arg constructor for `tf.Callback`.\n   * @throws Error, if the same callbackConstructor has been registered before,\n   *   either at the same or a different `verbosityLevel`.\n   */\n  static registerCallbackConstructor(\n      verbosityLevel: number, callbackConstructor: BaseCallbackConstructor) {\n    util.assert(\n        verbosityLevel >= 0 && Number.isInteger(verbosityLevel),\n        () => `Verbosity level is expected to be an integer >= 0, ` +\n            `but got ${verbosityLevel}`);\n    CallbackConstructorRegistry.checkForDuplicate(callbackConstructor);\n    if (CallbackConstructorRegistry.constructors[verbosityLevel] == null) {\n      CallbackConstructorRegistry.constructors[verbosityLevel] = [];\n    }\n    CallbackConstructorRegistry.constructors[verbosityLevel].push(\n        callbackConstructor);\n  }\n\n  private static checkForDuplicate(callbackConstructor:\n                                       BaseCallbackConstructor) {\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const constructors = CallbackConstructorRegistry.constructors[+levelName];\n      constructors.forEach(ctor => {\n        if (ctor === callbackConstructor) {\n          throw new ValueError('Duplicate callback constructor.');\n        }\n      });\n    }\n  }\n\n  /**\n   * Clear all registered callback constructors.\n   */\n  protected static clear() {\n    CallbackConstructorRegistry.constructors = {};\n  }\n\n  /**\n   * Create callbacks using the registered callback constructors.\n   *\n   * Given `verbosityLevel`, all constructors registered at that level or above\n   * will be called and the instantiated callbacks will be used.\n   *\n   * @param verbosityLevel: Level of verbosity.\n   */\n  static createCallbacks(verbosityLevel: number): BaseCallback[] {\n    const constructors: BaseCallbackConstructor[] = [];\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const level = +levelName;\n      if (verbosityLevel >= level) {\n        constructors.push(...CallbackConstructorRegistry.constructors[level]);\n      }\n    }\n    return constructors.map(ctor => new ctor());\n  }\n}\n\nexport function configureCallbacks(\n    callbacks: BaseCallback[], verbose: ModelLoggingVerbosity, epochs: number,\n    initialEpoch: number, numTrainSamples: number, stepsPerEpoch: number,\n    batchSize: number, doValidation: boolean,\n    callbackMetrics: string[]): {callbackList: CallbackList, history: History} {\n  const history = new History();\n  const actualCallbacks: BaseCallback[] = [\n    new BaseLogger(), ...CallbackConstructorRegistry.createCallbacks(verbose)\n  ];\n  if (callbacks != null) {\n    actualCallbacks.push(...callbacks);\n  }\n  actualCallbacks.push(history);\n  const callbackList = new CallbackList(actualCallbacks);\n\n  // TODO(cais): Figure out when this LayersModel instance can have a\n  // dynamically\n  //   set property called 'callback_model' as in PyKeras.\n\n  callbackList.setParams({\n    epochs,\n    initialEpoch,\n    samples: numTrainSamples,\n    steps: stepsPerEpoch,\n    batchSize,\n    verbose,\n    doValidation,\n    metrics: callbackMetrics,\n  });\n  return {callbackList, history};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source layers/__init__.py */\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {deserializeKerasObject} from '../utils/generic_utils';\n\n/**\n * Instantiate a layer from a config dictionary.\n * @param config dict of the form {class_name: str, config: dict}\n * @param customObjects dict mapping class names (or function names)\n *   of custom (non-Keras) objects to class/functions\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns Layer instance (may be LayersModel, Sequential, Layer...)\n */\nexport function deserialize(\n    config: serialization.ConfigDict,\n    customObjects = {} as serialization.ConfigDict,\n    fastWeightInit = false): serialization.Serializable {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'layer', fastWeightInit);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: losses.py */\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\nimport * as K from './backend/tfjs_backend';\nimport {ValueError} from './errors';\nimport {LossOrMetricFn} from './types';\n\n/**\n * Normalizes a tensor wrt the L2 norm alongside the specified axis.\n * @param x\n * @param axis Axis along which to perform normalization.\n */\nexport function l2Normalize(x: Tensor, axis?: number): Tensor {\n  return tidy(() => {\n    if (x.dtype !== 'float32') {\n      x = x.asType('float32');\n    }\n    const squareSum = tfc.sum(K.square(x), axis, true);\n    const epsilonTensor = tfc.fill(squareSum.shape, epsilon());\n    const norm = tfc.sqrt(tfc.maximum(squareSum, epsilonTensor));\n    return tfc.div(x, norm);\n  });\n}\n\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(K.square(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(tfc.abs(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const diff = tfc.sub(yTrue, yPred);\n    const clippedTrue =\n        tfc.clipByValue(tfc.abs(yTrue), epsilon(), Number.MAX_VALUE);\n    const absResult = tfc.abs(tfc.div(diff, clippedTrue));\n    return tfc.mul(100, tfc.mean(absResult, -1));\n  });\n}\n\nexport function meanSquaredLogarithmicError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), Number.MAX_VALUE);\n    const firstLog = tfc.log(tfc.add(1, clippedPred));\n\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), Number.MAX_VALUE);\n    const secondLog = tfc.log(tfc.add(1, clippedTrue));\n\n    return tfc.mean(K.square(tfc.sub(firstLog, secondLog)), -1);\n  });\n}\n\nexport function squaredHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(K.square(maxResult), -1);\n  });\n}\n\nexport function hinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(maxResult, -1);\n  });\n}\n\nexport function categoricalHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const pos = tfc.sum(tfc.mul(yTrue, yPred), -1);\n    const neg = tfc.max(tfc.mul(tfc.sub(1, yTrue), yPred), -1);\n    return tfc.maximum(0, tfc.add(1, tfc.sub(neg, pos)));\n  });\n}\n\n/**\n * Logarithm of the hyperbolic cosine of the prediction error.\n *\n * `log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and\n * to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly\n * like the mean squared error, but will not be so strongly affected by the\n * occasional wildly incorrect prediction.\n */\nexport function logcosh(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const log2 = Math.log(2);\n    const predictionDiff = tfc.sub(yPred, yTrue);\n    const logcoshResult = tfc.sub(\n        tfc.add(predictionDiff, tfc.softplus(tfc.mul(-2, predictionDiff))),\n        log2);\n    return tfc.mean(logcoshResult, -1);\n  });\n}\n\nexport function categoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    if (fromLogits) {\n      output = tfc.softmax(output);\n    } else {\n      // scale preds so that the class probabilities of each sample sum to 1.\n      const outputSum = tfc.sum(output, output.shape.length - 1, true);\n      output = tfc.div(output, outputSum);\n    }\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    return tfc.neg(tfc.sum(\n        tfc.mul(target.toFloat(), tfc.log(output)), output.shape.length - 1));\n  });\n}\n\n/**\n * Categorical crossentropy with integer targets.\n *\n * @param target An integer tensor.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n */\nexport function sparseCategoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    const flatTarget = tfc.floor(K.flatten(target)).toInt() as Tensor1D;\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    const outputShape = output.shape;\n    const oneHotTarget =\n        tfc.oneHot(flatTarget, outputShape[outputShape.length - 1])\n            .reshape(outputShape);\n    return categoricalCrossentropy(oneHotTarget, output, fromLogits);\n  });\n}\n\n/**\n * From TensorFlow's implementation in nn_impl.py:\n *\n * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n *      z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n *    = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n *    = (1 - z) * x + log(1 + exp(-x))\n *    = x - x * z + log(1 + exp(-x))\n * For x < 0, to avoid overflow in exp(-x), we reformulate the above\n *      x - x * z + log(1 + exp(-x))\n *    = log(exp(x)) - x * z + log(1 + exp(-x))\n *    = - x * z + log(1 + exp(x))\n * Hence, to ensure stability and avoid overflow, the implementation uses this\n * equivalent formulation\n *    max(x, 0) - x * z + log(1 + exp(-abs(x)))\n *\n * @param labels The labels.\n * @param logits The logits.\n */\nexport function sigmoidCrossEntropyWithLogits(\n    labels: Tensor, logits: Tensor): Tensor {\n  if (!util.arraysEqual(labels.shape, logits.shape)) {\n    throw new ValueError(\n        `logits and labels must have the same shape, but got shapes ` +\n        `${JSON.stringify(labels.shape)} and ${JSON.stringify(logits.shape)}`);\n  }\n  return tidy(() => {\n    // The logistic loss formula from above is\n    //   x - x * z + log(1 + exp(-x))\n    // For x < 0, a more numerically stable formula is\n    //   -x * z + log(1 + exp(x))\n    // Note that these two expressions can be combined into the following:\n    //   max(x, 0) - x * z + log(1 + exp(-abs(x)))\n    const reluLogits = logits.relu();\n    const negAbsLogits = logits.abs().neg();\n    return reluLogits.sub(logits.mul(labels)).add(negAbsLogits.exp().log1p());\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    let y: Tensor;\n    y = tfc.clipByValue(yPred, epsilon(), 1 - epsilon());\n    y = tfc.log(tfc.div(y, tfc.sub(1, y)));\n    return tfc.mean(sigmoidCrossEntropyWithLogits(yTrue, y), -1);\n  });\n}\n\nexport function kullbackLeiblerDivergence(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), 1);\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), 1);\n    return tfc.sum(\n        tfc.mul(yTrue, tfc.log(tfc.div(clippedTrue, clippedPred))), -1);\n  });\n}\n\nexport function poisson(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const logPred = tfc.log(tfc.add(epsilon(), yPred));\n    return tfc.mean(tfc.sub(yPred, tfc.mul(yTrue, logPred)), -1);\n  });\n}\n\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const trueNormalized = l2Normalize(yTrue, -1);\n    const predNormalized = l2Normalize(yPred, -1);\n    const trueXPred = tfc.mul(trueNormalized, predNormalized);\n    return tfc.neg(tfc.sum(trueXPred, -1));\n  });\n}\n\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const msle = meanSquaredLogarithmicError;\nexport const MSLE = meanSquaredLogarithmicError;\nexport const kld = kullbackLeiblerDivergence;\nexport const KLD = kullbackLeiblerDivergence;\nexport const cosine = cosineProximity;\n\n// TODO(michaelterry): Add deserialize() function.\n\nexport const lossesMap: {[functionName: string]: LossOrMetricFn} = {\n  meanSquaredError,\n  meanAbsoluteError,\n  meanAbsolutePercentageError,\n  meanSquaredLogarithmicError,\n  squaredHinge,\n  hinge,\n  categoricalHinge,\n  logcosh,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  binaryCrossentropy,\n  kullbackLeiblerDivergence,\n  poisson,\n  cosineProximity\n};\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function get(identifierOrFn: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifierOrFn === 'string') {\n    if (identifierOrFn in lossesMap) {\n      return lossesMap[identifierOrFn];\n    }\n    let errMsg = `Unknown loss ${identifierOrFn}`;\n    if (identifierOrFn.toLowerCase().includes('softmaxcrossentropy')) {\n      errMsg = `Unknown loss ${identifierOrFn}. ` +\n          'Use \"categoricalCrossentropy\" as the string name for ' +\n          'tf.losses.softmaxCrossEntropy';\n    }\n    throw new ValueError(errMsg);\n  } else {\n    return identifierOrFn;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Built-in metrics.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {NotImplementedError, ValueError} from './errors';\nimport {categoricalCrossentropy as categoricalCrossentropyLoss, cosineProximity, meanAbsoluteError, meanAbsolutePercentageError, meanSquaredError, sparseCategoricalCrossentropy as sparseCategoricalCrossentropyLoss} from './losses';\nimport {binaryCrossentropy as lossBinaryCrossentropy} from './losses';\nimport {lossesMap} from './losses';\nimport {LossOrMetricFn} from './types';\nimport * as util from './utils/generic_utils';\n\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const threshold = tfc.mul(.5, tfc.onesLike(yPred));\n    const yPredThresholded = K.cast(tfc.greater(yPred, threshold), yTrue.dtype);\n    return tfc.mean(tfc.equal(yTrue, yPredThresholded), -1);\n  });\n}\n\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(\n      () => K.cast(\n          tfc.equal(tfc.argMax(yTrue, -1), tfc.argMax(yPred, -1)), 'float32'));\n}\n\nfunction truePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nfunction falseNegatives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(0)).sum().cast('float32');\n  });\n}\n\nfunction falsePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(0), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fp = falsePositives(yTrue, yPred);\n\n    const denominator = tp.add(fp);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fn = falseNegatives(yTrue, yPred);\n\n    const denominator = tp.add(fn);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return lossBinaryCrossentropy(yTrue, yPred);\n}\n\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  if (yTrue.rank === yPred.rank) {\n    yTrue = yTrue.squeeze([yTrue.rank - 1]);\n  }\n  yPred = yPred.argMax(-1);\n  if (yPred.dtype !== yTrue.dtype) {\n    yPred = yPred.asType(yTrue.dtype);\n  }\n  return tfc.equal(yTrue, yPred).asType('float32');\n}\n\nexport function topKCategoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\nexport function sparseTopKCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\n// Aliases.\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const categoricalCrossentropy = categoricalCrossentropyLoss;\nexport const cosine = cosineProximity;\nexport const sparseCategoricalCrossentropy = sparseCategoricalCrossentropyLoss;\n\n// TODO(cais, nielsene): Add serialize().\n\nexport const metricsMap: {[functionName: string]: LossOrMetricFn} = {\n  binaryAccuracy,\n  categoricalAccuracy,\n  precision,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  mse,\n  MSE,\n  mae,\n  MAE,\n  mape,\n  MAPE,\n  cosine\n};\n\nexport function get(identifier: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifier === 'string' && identifier in metricsMap) {\n    return metricsMap[identifier];\n  } else if (typeof identifier !== 'string' && identifier != null) {\n    return identifier;\n  } else {\n    throw new ValueError(`Unknown metric ${identifier}`);\n  }\n}\n\n/**\n * Get the shortcut function name.\n *\n * If the fn name is a string,\n *   directly return the string name.\n * If the function is included in metricsMap or lossesMap,\n *   return key of the map.\n *   - If the function relative to multiple keys,\n *     return the first found key as the function name.\n *   - If the function exists in both lossesMap and metricsMap,\n *     search lossesMap first.\n * If the function is not included in metricsMap or lossesMap,\n *   return the function name.\n *\n * @param fn loss function, metric function, or short cut name.\n * @returns Loss or Metric name in string.\n */\nexport function getLossOrMetricName(fn: string|LossOrMetricFn): string {\n  util.assert(fn !== null, `Unknown LossOrMetricFn ${fn}`);\n  if (typeof fn === 'string') {\n    return fn;\n  } else {\n    let fnName;\n    for (const key of Object.keys(lossesMap)) {\n      if (lossesMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    for (const key of Object.keys(metricsMap)) {\n      if (metricsMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    return (fn as Function).name;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Optimizers.\n */\n\nimport {Optimizer, train} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\n\nimport {ValueError} from './errors';\n\n// Add (de)serialize()\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function getOptimizer(identifier: string): Optimizer {\n  const optimizerMap: {[optimizerName: string]: () => Optimizer} = {\n    'Adagrad': () => train.adagrad(0.01),\n    'Adadelta': () => train.adadelta(1, 0.95, epsilon()),\n    'Adam': () => train.adam(0.001, 0.9, 0.999, epsilon()),\n    'Adamax': () => train.adamax(0.002, 0.9, 0.999, epsilon(), 0),\n    'RMSProp': () => train.rmsprop(0.001, 0.9, 0, epsilon()),\n    'SGD': () => train.sgd(0.01)\n  };\n  optimizerMap['adagrad'] = optimizerMap['Adagrad'];\n  optimizerMap['adadelta'] = optimizerMap['Adadelta'];\n  optimizerMap['adam'] = optimizerMap['Adam'];\n  optimizerMap['adamax'] = optimizerMap['Adamax'];\n  optimizerMap['rmsprop'] = optimizerMap['RMSProp'];\n  optimizerMap['sgd'] = optimizerMap['SGD'];\n\n  if (identifier in optimizerMap) {\n    return optimizerMap[identifier]();\n  }\n  throw new ValueError(`Unknown Optimizer ${identifier}`);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/** Utility functions related to user-defined metadata. */\n\n// Maximum recommended serialized size for user-defined metadata.\n// Beyond this limit, a warning message will be printed during model loading and\n// saving.\nexport const MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH = 1 * 1024 * 1024;\n\n/**\n * Check validity of user-defined metadata.\n *\n * @param userDefinedMetadata\n * @param modelName Name of the model that the user-defined metadata belongs to.\n *   Used during construction of error messages.\n * @param checkSize Whether to check the size of the metadata is under\n *   recommended limit. Default: `false`. If `true`, will try stringify the\n *   JSON object and print a console warning if the serialzied size is above the\n *   limit.\n * @throws Error if `userDefinedMetadata` is not a plain JSON object.\n */\nexport function checkUserDefinedMetadata(\n    userDefinedMetadata: {}, modelName: string, checkSize = false): void {\n  if (userDefinedMetadata == null ||\n      typeof userDefinedMetadata !== 'object' ||\n      Object.getPrototypeOf(userDefinedMetadata) !== Object.prototype ||\n      !plainObjectCheck(userDefinedMetadata)) {\n    throw new Error(\n        'User-defined metadata is expected to be a JSON object, but is not.');\n  }\n\n  if (checkSize) {\n    const out = JSON.stringify(userDefinedMetadata);\n    if (out.length > MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH) {\n      console.warn(\n          `User-defined metadata of model \"${modelName}\" is too large in ` +\n          `size (length=${out.length} when serialized). It is not ` +\n          `recommended to store such large objects in user-defined metadata. ` +\n          `Please make sure its serialized length is <= ` +\n          `${MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH}.`);\n    }\n  }\n}\n\n/**\n * Check if an input is plain JSON object or any valid subfield of it.\n *\n * @param x The input to be checked.\n * @param assertObject Whether to assert `x` is a JSON object, i.e., reject\n *   cases of arrays and primitives.\n * @return Returns `true` if and only if `x` is a plain JSON object,\n *   a JSON-valid primitive including string, number, boolean and null,\n *   or an array of the said types.\n */\n// tslint:disable-next-line:no-any\nexport function plainObjectCheck(x: any): boolean {\n  if (x === null) {\n    // Note: typeof `null` is 'object', and `null` is valid in JSON.\n    return true;\n  } else if (typeof x === 'object') {\n    if (Object.getPrototypeOf(x) === Object.prototype) {\n      // `x` is a JavaScript object and its prototype is Object.\n      const keys = Object.keys(x);\n      for (const key of keys) {\n        if (typeof key !== 'string') {\n          // JSON keys must be strings.\n          return false;\n        }\n        if (!plainObjectCheck(x[key])) {  // Recursive call.\n          return false;\n        }\n      }\n      return true;\n    } else {\n      // `x` is a JavaScript object but its prototype is not Object.\n      if (Array.isArray(x)) {\n        // `x` is a JavaScript array.\n        for (const item of x) {\n          if (!plainObjectCheck(item)) {  // Recursive call.\n            return false;\n          }\n        }\n        return true;\n      } else {\n        // `x` is a JavaScript object and its prototype is not Object,\n        // and it's not an Array. I.e., it's a complex object such as\n        // `Error` and `Date`.\n        return false;\n      }\n    }\n  } else {\n    // `x` is not a JavaScript object or `null`.\n    const xType = typeof x;\n    return xType === 'string' || xType === 'number' || xType === 'boolean';\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {Container} from '../engine/container';\nimport {Layer, Node} from '../engine/topology';\nimport {countParamsInWeights} from './variable_utils';\n\n/**\n * Print the summary of a LayersModel object.\n *\n * @param model tf.LayersModel instance.\n * @param lineLength Total length of printed lines. Set this to adapt to the\n *   display to different terminal or console sizes.\n * @param positions Relative or absolute positions of log elements in each\n *   line. Each number corresponds to right-most (i.e., ending) position of a\n *   column.\n *   If not provided, defaults to `[0.45, 0.85, 1]` for sequential-like\n *   models and `[0.33, 0.55, 0.67, 1]` for non-sequential like models.\n * @param printFn Print function to use.\n *   It will be called on each line of the summary. You can provide a custom\n *   function in order to capture the string summary. Defaults to `console.log`.\n */\nexport function printSummary(\n    model: Container, lineLength?: number, positions?: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void =\n        console.log): void {\n  const sequentialLike = isModelSequentialLike(model);\n\n  // Header names for different log elements.\n  const toDisplay: string[] = ['Layer (type)', 'Output shape', 'Param #'];\n  if (sequentialLike) {\n    lineLength = lineLength || 65;\n    positions = positions || [0.45, 0.85, 1];\n  } else {\n    lineLength = lineLength || 98;\n    positions = positions || [0.33, 0.55, 0.67, 1];\n    // Header names for different log elements.\n  }\n\n  if (positions[positions.length - 1] <= 1) {\n    // `positions` is relative. Convert it to absolute positioning.\n    positions = positions.map(p => Math.floor(lineLength * p));\n  }\n\n  let relevantNodes: Node[];\n  if (!sequentialLike) {\n    toDisplay.push('Receives inputs');\n    relevantNodes = [];\n    for (const depth in model.nodesByDepth) {\n      relevantNodes.push(...model.nodesByDepth[depth]);\n    }\n  }\n\n  printFn('_'.repeat(lineLength));\n  printRow(toDisplay, positions, printFn);\n  printFn('='.repeat(lineLength));\n\n  const layers = model.layers;\n  for (let i = 0; i < layers.length; ++i) {\n    if (sequentialLike) {\n      printLayerSummary(layers[i], positions, printFn);\n    } else {\n      printLayerSummaryWithConnections(\n          layers[i], positions, relevantNodes, printFn);\n    }\n    printFn((i === layers.length - 1 ? '=' : '_').repeat(lineLength));\n  }\n\n  // tslint:disable-next-line:no-any\n  (model as any).checkTrainableWeightsConsistency();\n\n  const trainableCount = countTrainableParams(model);\n  const nonTrainableCount = countParamsInWeights(model.nonTrainableWeights);\n\n  printFn(`Total params: ${trainableCount + nonTrainableCount}`);\n  printFn(`Trainable params: ${trainableCount}`);\n  printFn(`Non-trainable params: ${nonTrainableCount}`);\n  printFn('_'.repeat(lineLength));\n}\n\nfunction countTrainableParams(model: Container): number {\n  let trainableCount: number;\n  // tslint:disable:no-any\n  if ((model as any).collectedTrainableWeights != null) {\n    trainableCount =\n        countParamsInWeights((model as any).collectedTrainableWeights);\n  } else {\n    trainableCount = countParamsInWeights(model.trainableWeights);\n  }\n  // tslint:enable:no-any\n  return trainableCount;\n}\n\nfunction isModelSequentialLike(model: Container): boolean {\n  let sequentialLike = true;\n  const nodesByDepth: Node[][] = [];\n  const nodes: Node[] = [];\n  for (const depth in model.nodesByDepth) {\n    nodesByDepth.push(model.nodesByDepth[depth]);\n  }\n  for (const depthNodes of nodesByDepth) {\n    if (depthNodes.length > 1 ||\n        depthNodes.length === 1 && depthNodes[0].inboundLayers.length > 1) {\n      sequentialLike = false;\n      break;\n    }\n    nodes.push(...depthNodes);\n  }\n  if (sequentialLike) {\n    // Search for shared layers.\n    for (const layer of model.layers) {\n      let flag = false;\n      for (const node of layer.inboundNodes) {\n        if (nodes.indexOf(node) !== -1) {\n          if (flag) {\n            sequentialLike = false;\n            break;\n          } else {\n            flag = true;\n          }\n        }\n      }\n      if (!sequentialLike) {\n        break;\n      }\n    }\n  }\n  return sequentialLike;\n}\n\nfunction printRow(\n    fields: string[], positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void = console.log) {\n  let line = '';\n  for (let i = 0; i < fields.length; ++i) {\n    if (i > 0) {\n      line = line.slice(0, line.length - 1) + ' ';\n    }\n    line += fields[i];\n    line = line.slice(0, positions[i]);\n    line += ' '.repeat(positions[i] - line.length);\n  }\n  printFn(line);\n}\n\n/**\n * Prints a summary for a single Layer, without connectivity information.\n *\n * @param layer: Layer instance to print.\n */\nfunction printLayerSummary(\n    layer: Layer, positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const name = layer.name;\n  const className = layer.getClassName();\n  const fields: string[] =\n      [`${name} (${className})`, outputShape, layer.countParams().toString()];\n  printRow(fields, positions, printFn);\n}\n\n/**\n * Prints a summary for a single Layer, with connectivity information.\n */\nfunction printLayerSummaryWithConnections(\n    layer: Layer, positions: number[], relevantNodes: Node[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const connections: string[] = [];\n  for (const node of layer.inboundNodes) {\n    if (relevantNodes != null && relevantNodes.length > 0 &&\n        relevantNodes.indexOf(node) === -1) {\n      continue;\n    }\n    for (let i = 0; i < node.inboundLayers.length; ++i) {\n      const inboundLayer = node.inboundLayers[i].name;\n      const inboundLayerIndex = node.nodeIndices[i];\n      const inboundTensorIndex = node.tensorIndices[i];\n      connections.push(\n          `${inboundLayer}[${inboundLayerIndex}][${inboundTensorIndex}]`);\n    }\n  }\n  const name = layer.name;\n  const className = layer.getClassName();\n  const firstConnection = connections.length === 0 ? '' : connections[0];\n  const fields: string[] = [\n    `${name} (${className})`, outputShape, layer.countParams().toString(),\n    firstConnection\n  ];\n\n  printRow(fields, positions, printFn);\n  for (let i = 1; i < connections.length; ++i) {\n    printRow(['', '', '', connections[i]], positions, printFn);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Porting note: This file doesn't exist in PyKeras.\n// Its purpose here is to centralize the boundary layer between\n// tfjs-layers's internal Config TS-Centric format and PyKeras's\n// serialized Python Config format.\n\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {PyJsonValue} from '../keras_format/types';\nimport * as generic_utils from '../utils/generic_utils';\n// tslint:enable\n\n/**\n * Test whether a value in an array is the name of a LayersModel or Layer.\n * @param key The key name that the value is found under. Note that the key\n *   may not be at the level immediately above the value, if the value is in a\n *   nested array.\n * @param index Index of the value in the Array that it is found in.\n * @param value The value object.\n * @returns A boolean indicating whether value is a name.\n */\nfunction isArrayItemInputOrOutputName<T>(\n    key: string, index: number, value: T): boolean {\n  return (key === 'inboundNodes' || key === 'outputLayers' ||\n          key === 'inputLayers') &&\n      index === 0 && typeof value === 'string';\n}\n\n/**\n * Convert a Pythonic config object to TypeScript config object.\n * @param pythonicConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertPythonicToTs(\n    pythonicConfig: PyJsonValue, key?: string): serialization.ConfigDictValue {\n  if (pythonicConfig === null) {\n    return null;\n  } else if (typeof pythonicConfig === 'string') {\n    return generic_utils.toCamelCase(pythonicConfig);\n  } else if (\n      (typeof pythonicConfig === 'number') ||\n      (typeof pythonicConfig === 'boolean')) {\n    return pythonicConfig;\n  } else if (pythonicConfig instanceof Array) {\n    const tsArray = [];\n    const arrayLength = pythonicConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = pythonicConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        tsArray.push(item);\n      } else {\n        tsArray.push(convertPythonicToTs(item, key));\n      }\n    }\n    return tsArray;\n  } else {\n    const tsDict: serialization.ConfigDict = {};\n    for (const pythonicKey of Object.keys(pythonicConfig)) {\n      const pythonicValue = pythonicConfig[pythonicKey];\n      if (pythonicKey === 'name' && typeof pythonicValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // camel-case conversion.\n        tsDict[pythonicKey] = pythonicValue;\n      } else {\n        const tsKey = generic_utils.toCamelCase(pythonicKey);\n        tsDict[tsKey] = convertPythonicToTs(pythonicValue, tsKey);\n      }\n    }\n    return tsDict;\n  }\n}\n\n/**\n * Convert a TypeScript config object to Python config object.\n * @param tsConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertTsToPythonic(\n    tsConfig: serialization.ConfigDictValue, key?: string): PyJsonValue {\n  if (tsConfig === null || tsConfig === undefined) {\n    return null;\n  } else if (typeof tsConfig === 'string') {\n    return generic_utils.toSnakeCase(tsConfig);\n  } else if (\n      (typeof tsConfig === 'number') || (typeof tsConfig === 'boolean')) {\n    return tsConfig;\n  } else if (tsConfig instanceof Array) {\n    const pyArray = [];\n    const arrayLength = tsConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = tsConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        pyArray.push(item);\n      } else {\n        pyArray.push(convertTsToPythonic(item, key));\n      }\n    }\n    return pyArray;\n  } else {\n    const pyDict: serialization.ConfigDict = {};\n    for (const tsKey of Object.keys(tsConfig)) {\n      const tsValue = tsConfig[tsKey];\n      const pyKey = generic_utils.toSnakeCase(tsKey);\n      if ((tsKey === 'name' || tsKey === 'className') &&\n          typeof tsValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // snake-case conversion.\n        pyDict[pyKey] = tsValue;\n      } else {\n        pyDict[pyKey] = convertTsToPythonic(tsValue, tsKey);\n      }\n    }\n    return pyDict;\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Executor: Evaluates SymbolicTensor based on feeds.\n */\n\nimport {cast, dispose, memory, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {ValueError} from '../errors';\nimport {Kwargs} from '../types';\nimport {toList} from '../utils/generic_utils';\n\nimport {InputLayer} from './input_layer';\nimport {SymbolicTensor} from './topology';\n\n/**\n * Helper function to check the dtype and shape compatibility of a feed value.\n */\nfunction assertFeedCompatibility(key: SymbolicTensor, val: Tensor): Tensor {\n  // Check dtype compatibility.\n  if (key.dtype == null || key.dtype === val.dtype) {\n    //  a.  If types match, return val tensor as is.\n    return val;\n  }\n  try {\n    //  b. Attempt to convert to expected type.\n    return cast(val, key.dtype);\n  } catch (err) {\n    //  c. If conversion fails, return helpful error.\n    throw new ValueError(\n        `The dtype of the feed (${val.dtype}) can not be cast to the dtype ` +\n        `of the key '${key.name}' (${key.dtype}).`);\n  }\n}\n\n/**\n * A concrete Tensor value for a symbolic tensor as the key.\n */\nexport interface Feed {\n  key: SymbolicTensor;\n  value: Tensor;\n}\n\n/**\n * FeedDict: A mapping from unique SymbolicTensors to feed values for them.\n * A feed value is a concrete value represented as an `Tensor`.\n */\nexport class FeedDict {\n  private id2Value: {[id: number]: Tensor} = {};\n  private id2Mask: {[id: number]: Tensor} = {};\n  private name2Id: {[name: string]: number} = {};\n\n  /**\n   * Constructor, optionally does copy-construction.\n   * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case\n   *   copy-construction will be performed.\n   */\n  constructor(feeds?: Feed[]|FeedDict) {\n    if (feeds instanceof FeedDict) {\n      for (const id in feeds.id2Value) {\n        this.id2Value[id] = feeds.id2Value[id];\n        if (id in feeds.id2Mask) {\n          this.id2Mask[id] = feeds.id2Mask[id];\n        }\n      }\n    } else {\n      if (feeds == null) {\n        return;\n      }\n      for (const feed of feeds) {\n        this.add(feed.key, feed.value);\n      }\n    }\n  }\n\n  /**\n   * Add a key-value pair to the FeedDict.\n   *\n   * @param key The key of the feed.\n   * @param value The value of the tensor feed.\n   * @param mask The value of the mask feed (optional).\n   * @returns This `FeedDict`.\n   * @throws ValueError: If the key `SymbolicTensor` already exists in the\n   *   `FeedDict`.\n   */\n  add(key: SymbolicTensor, value: Tensor, mask?: Tensor): FeedDict {\n    if (this.id2Value[key.id] == null) {\n      this.id2Value[key.id] = assertFeedCompatibility(key, value);\n      this.name2Id[key.name] = key.id;\n      if (mask != null) {\n        this.id2Mask[key.id] = mask;\n      }\n    } else {\n      throw new ValueError(`Duplicate key: name=${key.name}, id=${key.id}`);\n    }\n    return this;\n  }\n\n  /**\n   * Add a Feed to the FeedDict.\n   * @param feed The new `Feed` to add.\n   * @returns This `FeedDict`.\n   */\n  addFeed(feed: Feed) {\n    this.add(feed.key, feed.value);\n  }\n\n  /**\n   * Probe whether a key already exists in the FeedDict.\n   * @param key\n   */\n  hasKey(key: SymbolicTensor): boolean {\n    return this.id2Value[key.id] != null;\n  }\n\n  /**\n   * Get all the SymbolicTensor available in this FeedDict.\n   */\n  names(): string[] {\n    return Object.keys(this.name2Id);\n  }\n\n  /**\n   * Get the feed value for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed value.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getValue(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Value[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Value[id];\n    }\n  }\n\n  /**\n   * Get the feed mask for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed mask.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getMask(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Mask[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Mask[id];\n    }\n  }\n\n  /** Dispose all mask Tensors held by this object. */\n  disposeMasks() {\n    if (this.id2Mask != null) {\n      dispose(this.id2Mask);\n    }\n  }\n}\n\n// Cache for topologically sorted SymbolicTensors for given execution\n// targets (i.e., fetches).\nconst cachedSorted: {[concatFetchNames: string]: SymbolicTensor[]} = {};\n\n// Cache for recipient count maps for given execution targets (i.e., fetches).\nconst cachedRecipientCounts:\n    {[concatFetchNames: string]: {[fetchName: string]: number}} = {};\n\n/**\n * Interface for the optional object used for probing the memory\n * usage and other statistics during execution.\n */\nexport interface ExecutionProbe {\n  /**\n   * Maximum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  maxNumTensors?: number;\n\n  /**\n   * Minimum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  minNumTensors?: number;\n}\n\n/**\n * Execute a SymbolicTensor by using concrete feed values.\n *\n * A `SymbolicTensor` object is a node in a computation graph of TF.js\n * Layers. The object is backed by a source layer and input\n * `SymbolicTensor`s to the source layer. This method evaluates\n * the `call()` method of the source layer, using concrete values of the\n * inputs obtained from either\n * * `feedDict`, if the input key exists in `feedDict`, or else,\n * * a recursive call to `execute()` itself.\n *\n * @param x: The `SymbolicTensor` to execute.\n * @param feedDict: The feed values, as base condition of the recursion.\n *   execution.\n * @param kwargs: Optional keyword arguments.\n * @param probe: A probe object (of interface `ExecutionProbe`) used for\n *   testing memory footprint of `execute` calls.\n * @returns Result of the execution.\n * @throws ValueError: If any `SymbolicTensor`s from `InputLayer`s\n *   encountered during the execution lacks a feed value in `feedDict`.\n */\nexport function execute(\n    fetches: SymbolicTensor|SymbolicTensor[], feedDict: FeedDict,\n    kwargs?: Kwargs, probe?: ExecutionProbe): Tensor|\n    Tensor[]|[Tensor | Tensor[]] {\n  const training: boolean = kwargs == null ? false : kwargs['training'];\n\n  const arrayFetches = Array.isArray(fetches);\n  const fetchArray: SymbolicTensor[] =\n      arrayFetches ? fetches as SymbolicTensor[] : [fetches as SymbolicTensor];\n\n  const outputNames = fetchArray.map(t => t.name);\n  const finalOutputs: Tensor[] = [];\n  const feedNames = feedDict.names();\n  for (const outputName of outputNames) {\n    if (feedNames.indexOf(outputName) !== -1) {\n      finalOutputs.push(feedDict.getValue(outputName));\n    } else {\n      finalOutputs.push(null);\n    }\n  }\n\n  if (probe != null) {\n    // For optional probing of memory footprint during execution.\n    probe.maxNumTensors = -Infinity;\n    probe.minNumTensors = Infinity;\n  }\n\n  // Check cache.\n  const fetchAndFeedKey =\n      outputNames.join(',') + '|' + feedDict.names().join(',');\n  let sorted: SymbolicTensor[];\n  let recipientCounts: {[fetchName: string]: number};\n  if (cachedSorted[fetchAndFeedKey] == null) {\n    // Cache doesn't contain the desired combination of fetches. Compute\n    // topological sort for the combination for the first time.\n    const out = getTopologicalSortAndRecipientCounts(fetchArray, feedDict);\n    sorted = out.sorted;\n    recipientCounts = out.recipientCounts;\n\n    // Store results in cache for future use.\n    cachedSorted[fetchAndFeedKey] = sorted;\n    cachedRecipientCounts[fetchAndFeedKey] = recipientCounts;\n  }\n  sorted = cachedSorted[fetchAndFeedKey];\n  recipientCounts = {};\n  if (!training) {\n    Object.assign(recipientCounts, cachedRecipientCounts[fetchAndFeedKey]);\n  }\n\n  const internalFeedDict = new FeedDict(feedDict);\n\n  // Start iterative execution on the topologically-sorted SymbolicTensors.\n  for (let i = 0; i < sorted.length; ++i) {\n    if (probe != null) {\n      // For optional probing of memory usage during execution.\n      const numTensors = memory().numTensors;\n      if (numTensors > probe.maxNumTensors) {\n        probe.maxNumTensors = numTensors;\n      }\n      if (numTensors < probe.minNumTensors) {\n        probe.minNumTensors = numTensors;\n      }\n    }\n\n    const symbolic = sorted[i];\n    const srcLayer = symbolic.sourceLayer;\n    if (srcLayer instanceof InputLayer) {\n      continue;\n    }\n    const inputValues: Tensor[] = [];\n    const inputMasks: Tensor[] = [];\n    const tensorsToDispose: Tensor[] = [];\n\n    let maskExists = false;\n    for (const input of symbolic.inputs) {\n      const value = internalFeedDict.getValue(input);\n      const mask = internalFeedDict.getMask(input);\n      inputValues.push(value);\n      inputMasks.push(mask);\n      if (mask != null) {\n        maskExists = true;\n      }\n      if (!training) {\n        recipientCounts[input.name]--;\n        if (recipientCounts[input.name] === 0 && !feedDict.hasKey(input) &&\n            outputNames.indexOf(input.name) === -1 && !value.isDisposed &&\n            input.sourceLayer.stateful !== true) {\n          tensorsToDispose.push(value);\n        }\n      }\n    }\n\n    if (maskExists) {\n      kwargs = kwargs || {};\n      kwargs['mask'] = inputMasks[0];\n    }\n    const outputTensors =\n        toList(srcLayer.apply(inputValues, kwargs)) as Tensor[];\n    let outputMask: Tensor|Tensor[] = null;\n    if (srcLayer.supportsMasking) {\n      outputMask = srcLayer.computeMask(inputValues, inputMasks);\n    }\n    const layerOutputs = getNodeOutputs(symbolic);\n    const outputSymbolicTensors =\n        Array.isArray(layerOutputs) ? layerOutputs : [layerOutputs];\n    for (let i = 0; i < outputSymbolicTensors.length; ++i) {\n      if (!internalFeedDict.hasKey(outputSymbolicTensors[i])) {\n        internalFeedDict.add(\n            outputSymbolicTensors[i], outputTensors[i],\n            Array.isArray(outputMask) ? outputMask[0] : outputMask);\n      }\n      const index = outputNames.indexOf(outputSymbolicTensors[i].name);\n      if (index !== -1) {\n        finalOutputs[index] = outputTensors[i];\n      }\n    }\n\n    if (!training) {\n      // Clean up Tensors that are no longer needed.\n      dispose(tensorsToDispose);\n    }\n  }\n  // NOTE(cais): Unlike intermediate tensors, we don't discard mask\n  // tensors as we go, because these tensors are sometimes passed over a\n  // series of mutliple layers, i.e., not obeying the immediate input\n  // relations in the graph. If this becomes a memory-usage concern,\n  // we can improve this in the future.\n  internalFeedDict.disposeMasks();\n\n  return arrayFetches ? finalOutputs : finalOutputs[0];\n}\n\ntype RecipientCounts = {\n  [fetchName: string]: number\n};\n\nexport type RecipientMap = {\n  [fetchName: string]: Set<string>;\n};\n\n/**\n * Sort the `SymbolicTensor`s topologically, for an array of fetches.\n *\n * This function calls getTopologicalSortAndRecipientCountsForOneFetch and\n * merges their results.\n *\n * @param fetch The array of fetches requested. Must be a non-empty array.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientCounts: Recipient counts for all SymbolicTensors in `sorted`.\n */\nfunction getTopologicalSortAndRecipientCounts(\n    fetches: SymbolicTensor[], feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientCounts: RecipientCounts} {\n  util.assert(\n      fetches != null && fetches.length > 0,\n      () => `Expected at least one fetch, got none`);\n\n  let finalSorted: SymbolicTensor[] = [];\n  let finalRecipientMap: RecipientMap = {};\n  if (fetches.length === 1) {\n    // Special-casing 1 fetch for efficiency.\n    const out =\n        getTopologicalSortAndRecipientCountsForOneFetch(fetches[0], feedDict);\n    finalSorted = out.sorted;\n    finalRecipientMap = out.recipientMap;\n  } else {\n    const visited = new Set<string>();\n    for (const fetch of fetches) {\n      const {sorted, recipientMap} =\n          getTopologicalSortAndRecipientCountsForOneFetch(fetch, feedDict);\n\n      // Merge sorted SymbolicTensor Arrays.\n      for (const symbolicTensor of sorted) {\n        if (!visited.has(symbolicTensor.name)) {\n          finalSorted.push(symbolicTensor);\n          visited.add(symbolicTensor.name);\n        }\n      }\n\n      // Merge recipient maps.\n      for (const name in recipientMap) {\n        if (finalRecipientMap[name] == null) {\n          finalRecipientMap[name] = new Set<string>();\n        }\n        recipientMap[name].forEach(\n            recipient => finalRecipientMap[name].add(recipient));\n      }\n    }\n  }\n  return {\n    sorted: finalSorted,\n    recipientCounts: recipientMap2Counts(finalRecipientMap)\n  };\n}\n\nfunction recipientMap2Counts(recipientMap: RecipientMap): RecipientCounts {\n  const recipientCounts: RecipientCounts = {};\n  for (const name in recipientMap) {\n    recipientCounts[name] = recipientMap[name].size;\n  }\n  return recipientCounts;\n}\n\n/**\n * Sort the `SymbolicTensor`s topologically, for a single fetch.\n *\n * This helper function processes the upstream SymbolicTensors of a single\n * fetch.\n *\n * @param fetch The single fetch requested.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientMap: Recipient names for all SymbolicTensors in `sorted`.\n */\nexport function getTopologicalSortAndRecipientCountsForOneFetch(\n    fetch: SymbolicTensor, feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientMap: RecipientMap} {\n  const visited = new Set<string>();\n  const sorted: SymbolicTensor[] = [];\n  const recipientMap: RecipientMap = {};\n\n  // Put keys of the feedDict into visited first, so they don't have to be\n  // walked. This is needed in case where there are feeds for intermediate\n  // SymbolicTensors of the graph.\n  for (const key of feedDict.names()) {\n    visited.add(key);\n  }\n\n  const stack: SymbolicTensor[] = [];\n  const marks: number[] = [];\n\n  // Initial population of stack and marks.\n  stack.push(fetch);\n\n  while (stack.length > 0) {\n    const top = stack[stack.length - 1];\n    if (visited.has(top.name)) {\n      stack.pop();\n      continue;\n    }\n    const topIsMarked = marks[marks.length - 1] === stack.length - 1;\n    if (top.inputs.length === 0 || topIsMarked) {\n      // Input SymbolicTensor or all children have been visited.\n      stack.pop();\n      sorted.push(top);\n      visited.add(top.name);\n      if (topIsMarked) {\n        marks.pop();\n      }\n    } else {\n      // A non-input SymbolicTensor whose upstream SymbolicTensors haven't\n      // been visited yet. Push them onto the stack.\n      marks.push(stack.length - 1);\n      for (const input of top.inputs) {\n        // Increment the recipient count. Note that this needs to happen\n        // regardless of whether the SymbolicTensor has been visited before.\n        if (recipientMap[input.name] == null) {\n          recipientMap[input.name] = new Set<string>();\n        }\n        recipientMap[input.name].add(top.name);\n\n        if (visited.has(input.name)) {\n          continue;  // Avoid repeated visits to the same SymbolicTensor.\n        }\n        stack.push(input);\n      }\n    }\n  }\n  return {sorted, recipientMap};\n}\n\n/**\n * Get the symbolic output tensors of the node to which a given fetch belongs.\n * @param fetch The fetched symbolic tensor.\n * @returns The Array of symbolic tensors output by the node to which `fetch`\n *   belongs.\n */\nfunction getNodeOutputs(fetch: SymbolicTensor): SymbolicTensor|\n    SymbolicTensor[] {\n  let layerOutputs: SymbolicTensor|SymbolicTensor[];\n  if (fetch.sourceLayer.inboundNodes.length === 1) {\n    layerOutputs = fetch.sourceLayer.output;\n  } else {\n    let nodeIndex: number = null;\n    for (let i = 0; i < fetch.sourceLayer.inboundNodes.length; ++i) {\n      for (const outputTensor of fetch.sourceLayer.inboundNodes[i]\n               .outputTensors) {\n        if (outputTensor.id === fetch.id) {\n          nodeIndex = i;\n          break;\n        }\n      }\n    }\n    layerOutputs = fetch.sourceLayer.getOutputAt(nodeIndex);\n  }\n  return layerOutputs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {NamedTensorMap, Scalar, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {TensorKeyWithArgsArray} from '../keras_format/node_config';\nimport {PyJsonDict} from '../keras_format/types';\nimport {deserialize as deserializeLayer} from '../layers/serialization';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {convertTsToPythonic} from '../utils/serialization_utils';\nimport * as types_utils from '../utils/types_utils';\nimport {batchSetValue, LayerVariable} from '../variables';\nimport {version as layersVersion} from '../version';\n\nimport {execute, FeedDict} from './executor';\nimport {InputLayer} from './input_layer';\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/** Constructor config for Container. */\nexport interface ContainerArgs {\n  inputs: SymbolicTensor|SymbolicTensor[];\n  outputs: SymbolicTensor|SymbolicTensor[];\n  name?: string;\n}\n\n/**\n * A Container is a directed acyclic graph of layers.\n *\n * It is the topological form of a \"model\". A LayersModel\n * is simply a Container with added training routines.\n *\n */\nexport abstract class Container extends Layer {\n  inputs: SymbolicTensor[];\n  outputs: SymbolicTensor[];\n\n  inputLayers: Layer[];\n  inputLayersNodeIndices: number[];\n  inputLayersTensorIndices: number[];\n\n  outputLayers: Layer[];\n  outputLayersNodeIndices: number[];\n  outputLayersTensorIndices: number[];\n\n  layers: Layer[];\n  layersByDepth: {[depth: string]: Layer[]};\n  nodesByDepth: {[depth: string]: Node[]};\n\n  internalContainerRefs: Container[];\n\n  containerNodes = new Set<string>();\n\n  // TODO(michaelterry): Add cache support\n  // private outputMaskCache: any;\n  // private outputTensorCache: any;\n  // private outputShapeCache: any;\n\n  inputNames: string[];\n  outputNames: string[];\n  feedInputShapes: Shape[];\n\n  protected internalInputShapes: Shape[];\n  protected internalOutputShapes: Shape[];\n  // TODO(cais): Maybe 'feed' should not in the names of these variables,\n  //   due to the fact that our backend is not symbolic.\n  protected feedInputNames: string[];\n  protected feedOutputNames: string[];\n\n  constructor(args: ContainerArgs) {\n    // No args passed to super's constructor.\n    super({});\n    this.name = args.name;\n    if (this.name == null) {\n      const prefix = this.getClassName().toLowerCase();\n      this.name = getUid(prefix);\n    }\n\n    this.supportsMasking = false;\n    this.trainable_ = true;\n\n    // TODO(michaelterry): Initialize perInputLosses/Updates here.\n\n    // Container-specific properties.\n    if (Array.isArray(args.inputs)) {\n      this.inputs = args.inputs.slice();\n    } else {\n      this.inputs = [args.inputs];\n    }\n    if (Array.isArray(args.outputs)) {\n      this.outputs = args.outputs.slice();\n    } else {\n      this.outputs = [args.outputs];\n    }\n\n    // Check for redundancy in inputs.\n    if (generic_utils.unique(this.inputs).length !== this.inputs.length) {\n      throw new ValueError(\n          'The list of inputs passed to the model is ' +\n          'redundant. All inputs should only appear once. Found: ' +\n          `${this.inputs.map(x => x.name)}`);\n    }\n\n    // Check for redundancy in outputs.\n    if (generic_utils.unique(this.outputs).length !== this.outputs.length) {\n      console.warn(\n          'The list of outputs passed to the model is redundant. ' +\n          'All outputs should only appear once. Found: ' +\n          `${this.outputs.map(x => x.name)}`);\n    }\n\n    /*\n      List of initial layers (1 to 1 mapping with this.inputs, hence the same\n      layer might appear twice)\n    */\n    this.inputLayers = [];\n    this.inputLayersNodeIndices = [];\n    this.inputLayersTensorIndices = [];\n    /*\n      List of layers (1 to 1 mapping with this.outputs, hence the same layer\n      might appear twice)\n    */\n    this.outputLayers = [];\n    this.outputLayersNodeIndices = [];\n    this.outputLayersTensorIndices = [];\n    /*\n      All layers in order of horizontal graph traversal. Entries are unique.\n      Includes input and output layers.\n    */\n    this.layers = [];\n\n    /*\n      References to container layers that were constructed internally. We need\n      these to properly dispose of tensors from nested containers.\n    */\n    this.internalContainerRefs = [];\n\n    // TODO(michaelterry): Determine if caching still needed with eager\n    // backend.\n    /*\n      This is for performance optimization when calling the Container on new\n      inputs. Every time the Container is called on a set on input tensors,\n      we compute the output tensors, output masks and output shapes in one pass,\n      then cache them here. When one of these outputs is queried later,\n      we retrieve it from there instead of recomputing it.\n    */\n    // this.outputTensorCache = {};\n    // this.outputShapeCache = {};\n\n    // Build this.outputLayers:\n    for (const x of this.outputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      this.outputLayers.push(layer);\n      this.outputLayersNodeIndices.push(nodeIndex);\n      this.outputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // TODO(michaelterry): Add output mask cache code.\n\n    // Build this.inputLayers:\n    for (const x of this.inputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      /*\n        It's supposed to be an input layer, so only one node\n        and one tensor output.\n      */\n      generic_utils.assert(nodeIndex === 0, 'input layer has >1 nodes');\n      generic_utils.assert(tensorIndex === 0, 'input layer has >1 tensors');\n      this.inputLayers.push(layer);\n      this.inputLayersNodeIndices.push(nodeIndex);\n      this.inputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // Build this.inputNames and this.outputNames.\n    this.inputNames = [];\n    this.outputNames = [];\n    this.feedInputShapes = [];\n    this.feedInputNames = [];\n    this.feedOutputNames = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      // Check that layer is an InputLayer.\n      if (!(layer instanceof InputLayer)) {\n        throw new TypeError(\n            'Input layers to a LayersModel must be InputLayer objects. ' +\n            `Received inputs: ${args.inputs}. ` +\n            `Input ${i} (0-based) originates ` +\n            `from layer type ${layer.getClassName()}.`);\n      }\n      this.inputNames.push(layer.name);\n      this.feedInputShapes.push(layer.batchInputShape);\n\n      this.feedInputNames.push(layer.name);\n    }\n    for (const layer of this.outputLayers) {\n      this.outputNames.push(layer.name);\n    }\n\n    this.internalInputShapes = this.inputs.map(x => x.shape);\n    this.internalOutputShapes = this.outputs.map(x => x.shape);\n\n    /*\n      Container_nodes: set of nodes included in the graph (not all nodes\n      included in the layers are relevant to the current graph).\n    */\n    // ids of all nodes relevant to the Container:\n    const nodesDepths: {[nodeID: string]: number} = {};\n    // To recover nodes from their ID.\n    const nodeIDToNode: {[nodeID: string]: Node} = {};\n    const layersDepths: {[layerID: string]: number} = {};\n    // To layers from their ID.\n    const layerIDToLayer: {[layerID: string]: Layer} = {};\n    const layerIndices: {[layerID: string]: number} = {};\n    const nodesInDecreasingDepth: Node[] = [];\n\n    /**\n     * Builds a map of the graph of layers.\n     *\n     * This recursively updates the map `layerIndices`,\n     * the list `nodesInDecreasingDepth` and the set `containerNodes`.\n     *\n     * @param tensor Some tensor in a graph.\n     * @param finishedNodes Set of nodes whose subgraphs have been traversed\n     *         completely. Useful to prevent duplicated work.\n     * @param nodesInProgress Set of nodes that are currently active on the\n     *         recursion stack. Useful to detect cycles.\n     * @param layer Layer from which `tensor` comes from. If not provided,\n     *   will be obtained from tensor.sourceLayer.\n     * @param nodeIndex Node index from which `tensor` comes from.\n     * @param tensorIndex TensorIndex from which `tensor` comes from.\n     *\n     * @exception RuntimeError if a cycle is detected.\n     */\n    const buildMapOfGraph =\n        (tensor: SymbolicTensor, finishedNodes: Node[], nodesInProgress: Node[],\n         layer?: Layer, nodeIndex?: number, tensorIndex?: number) => {\n          if (layer == null || nodeIndex == null || tensorIndex == null) {\n            layer = tensor.sourceLayer;\n            nodeIndex = tensor.nodeIndex;\n            tensorIndex = tensor.tensorIndex;\n          }\n          const node = layer.inboundNodes[nodeIndex];\n\n          // Prevent cycles.\n          if (nodesInProgress.indexOf(node) !== -1) {\n            throw new RuntimeError(\n                `The tensor ${tensor.name} at layer \"${layer.name}\" ` +\n                'is part of a cycle.');\n          }\n\n          // Don't repeat work for shared subgraphs\n          if (finishedNodes.indexOf(node) !== -1) {\n            return;\n          }\n\n          // Update containerNodes.\n          this.containerNodes.add(Container.nodeKey(layer, nodeIndex));\n\n          // Store the traversal order for layer sorting.\n          if (!(layer.id in layerIndices)) {\n            layerIndices[layer.id] = Object.keys(layerIndices).length;\n          }\n\n          if (nodesInProgress.indexOf(node) === -1) {\n            nodesInProgress.push(node);\n          }\n\n          // Propagate to all previous tensors connected to this node.\n          const numInboundLayers = node.inboundLayers.length;\n          for (let i = 0; i < numInboundLayers; i++) {\n            const x = node.inputTensors[i];\n            const layer = node.inboundLayers[i];\n            const nodeIndex = node.nodeIndices[i];\n            const tensorIndex = node.tensorIndices[i];\n            buildMapOfGraph(\n                x, finishedNodes, nodesInProgress, layer, nodeIndex,\n                tensorIndex);\n          }\n          finishedNodes.push(node);\n          while (nodesInProgress.indexOf(node) >= 0) {\n            nodesInProgress.splice(nodesInProgress.indexOf(node), 1);\n          }\n          nodesInDecreasingDepth.push(node);\n        };\n\n    const finishedNodes: Node[] = [];\n    const nodesInProgress: Node[] = [];\n    for (const x of this.outputs) {\n      buildMapOfGraph(x, finishedNodes, nodesInProgress);\n    }\n\n    const reversedNodesInDecreasingDepth =\n        nodesInDecreasingDepth.slice().reverse();\n    for (const node of reversedNodesInDecreasingDepth) {\n      nodeIDToNode[node.id] = node;\n      // If the depth is not set, the node has no outbound nodes (depth 0).\n      if (!(node.id in nodesDepths)) {\n        nodesDepths[node.id] = 0;\n      }\n      let depth = nodesDepths[node.id];\n\n      // Update the depth of the corresponding layer\n      const previousDepth =\n          (layersDepths[node.outboundLayer.id] == null ?\n               0 :\n               layersDepths[node.outboundLayer.id]);\n\n      /*\n        If we've seen this layer before at a higher depth, we should use that\n        depth instead of the node depth.  This is necessary for shared layers\n        that have inputs at different depth levels in the graph.\n      */\n      depth = Math.max(depth, previousDepth);\n      layersDepths[node.outboundLayer.id] = depth;\n      layerIDToLayer[node.outboundLayer.id] = node.outboundLayer;\n      nodesDepths[node.id] = depth;\n\n      // Update the depth of inbound nodes.\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const inboundLayer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const inboundNode = inboundLayer.inboundNodes[nodeIndex];\n        const previousDepth =\n            (nodesDepths[inboundNode.id] == null ? 0 :\n                                                   nodesDepths[inboundNode.id]);\n        nodesDepths[inboundNode.id] = Math.max(depth + 1, previousDepth);\n        nodeIDToNode[inboundNode.id] = inboundNode;\n      }\n    }\n\n    // Build a dict {depth: list of nodes with this depth}\n    const nodesByDepth: {[depth: string]: Node[]} = {};\n    for (const nodeID in nodesDepths) {\n      const depth = nodesDepths[nodeID];\n      if (!(depth in nodesByDepth)) {\n        nodesByDepth[depth] = [];\n      }\n      nodesByDepth[depth].push(nodeIDToNode[nodeID]);\n    }\n\n    // Build a dict {depth: list of layers with this depth}\n    const layersByDepth: {[depth: string]: Layer[]} = {};\n    for (const layerID in layersDepths) {\n      const depth = layersDepths[layerID];\n      if (!(depth in layersByDepth)) {\n        layersByDepth[depth] = [];\n      }\n      layersByDepth[depth].push(layerIDToLayer[layerID]);\n    }\n\n    // Get sorted list of layer depths.\n    let depthKeys = Object.keys(layersByDepth)\n                        .map(x => parseInt(x, 10))\n                        .sort(generic_utils.reverseNumberCompare);\n\n    // Set this.layers and this.layersByDepth.\n    this.layers = [];\n    for (const depth of depthKeys) {\n      const layersForDepth = layersByDepth[depth];\n      // Container.layers needs to have a deterministic order:\n      // here we order them by traversal order.\n      layersForDepth.sort((a, b) => {\n        const aIndex = layerIndices[a.id];\n        const bIndex = layerIndices[b.id];\n        if (aIndex < bIndex) {\n          return -1;\n        }\n        if (aIndex > bIndex) {\n          return 1;\n        }\n        return 0;\n      });\n      for (const layer of layersForDepth) {\n        if (layer instanceof Container) {\n          this.internalContainerRefs.push(layer);\n        }\n        this.layers.push(layer);\n      }\n    }\n    this.layersByDepth = layersByDepth;\n\n    // Get sorted list of node depths;\n    depthKeys = Object.keys(nodesByDepth)\n                    .map(x => parseInt(x, 10))\n                    .sort(generic_utils.reverseNumberCompare);\n\n    // Check that all tensors required are computable.\n    // computable_tensors: all tensors in the graph\n    // that can be computed from the inputs provided.\n    const computableTensors = this.inputs.slice();\n\n    // To provide a better error msg.\n    const layersWithCompleteInput: string[] = [];\n    for (const depth of depthKeys) {\n      for (const node of nodesByDepth[depth]) {\n        const layer = node.outboundLayer;\n        if (layer != null) {\n          for (const x of node.inputTensors) {\n            if (computableTensors.indexOf(x) === -1) {\n              throw new RuntimeError(\n                  `Graph disconnected: cannot obtain value for tensor ${x}` +\n                  ` at layer \"${layer.name}\". ` +\n                  'The following previous layers were accessed without ' +\n                  `issue: ${layersWithCompleteInput}`);\n            }\n          }\n          for (const x of node.outputTensors) {\n            computableTensors.push(x);\n          }\n          layersWithCompleteInput.push(layer.name);\n        }\n      }\n    }\n\n    // Set this.containerNodes and this.nodesByDepth.\n    this.nodesByDepth = nodesByDepth;\n\n    // Ensure name unicity, which will be crucial for serialization\n    // (since serialized nodes refer to layers by their name).\n    const allNames = this.layers.map(x => x.name);\n    for (const name of allNames) {\n      const numOccurrences = allNames.filter(x => x === name).length;\n      if (numOccurrences !== 1) {\n        throw new RuntimeError(\n            `The name \"${name}\" is used ${numOccurrences} times ` +\n            'in the model. All layer names should be unique. Layer names: ' +\n            JSON.stringify(allNames));\n      }\n    }\n\n    // Layer parameters.\n    // The new container starts with a single inbound node\n    // for its inputs, and no outbound nodes.\n    // Will be appended to by future calls to apply().\n    this.outboundNodes = [];\n    // Will be appended to below, and by future calls to apply().\n    this.inboundNodes = [];\n\n    // Create the node linking internal inputs to internal outputs.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: this.inputs,\n      outputTensors: this.outputs,\n      inputMasks: this.inputs.map(x => null),\n      outputMasks: this.outputs.map(x => null),\n      inputShapes: this.inputs.map(x => x.shape),\n      outputShapes: this.outputs.map(x => x.shape)\n    });\n    this.built = true;\n    this._refCount = 1;  // The ref count of a container always start at 1.\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Container '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose a LayersModel's weights.\n   *\n   * This method decrease the reference count of the LayersModel object by 1.\n   *\n   * A LayersModel is reference-counted. Its reference count is incremented by 1\n   * when it is first constructed and when it is used as a Layer of another\n   * LayersModel.\n   *\n   * If the reference count of a LayersModel becomes 0, the `dispose` method of\n   * all its constituent `Layer`s will be called.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * `dispose` method of its constituent `Layer`s will *not* be called.\n   *\n   * After a LayersModel is disposed, it cannot be used in calls such as\n   * 'predict`, `evaluate` or `fit` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the LayersModel after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the LayersModel has\n   *   already been disposed.\n   */\n  dispose(): DisposeResult {\n    this.assertNotDisposed();\n    const result:\n        DisposeResult = {refCountAfterDispose: null, numDisposedVariables: 0};\n    if (--this._refCount === 0) {\n      for (const layer of this.layers) {\n        result.numDisposedVariables += layer.dispose().numDisposedVariables;\n      }\n\n      // Call dispose on each internally created container layer again to ensure\n      // their refCounts hit zero and their tensors are subsequently deleted.\n      for (const container of this.internalContainerRefs) {\n        result.numDisposedVariables += container.dispose().numDisposedVariables;\n      }\n    }\n    result.refCountAfterDispose = this._refCount;\n    return result;\n  }\n\n  get trainable() {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.layers.forEach(layer => {\n      // tslint:disable-next-line:no-any\n      ((layer as any)._trainableWeights as LayerVariable[])\n          .forEach(w => w.trainable = trainable);\n    });\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    // Porting Note: This check below is to prevent errors where the\n    //   _trainableWeights inherited from the parent class (Layer) gets\n    //   inadvertently used.\n    if (this._trainableWeights.length > 0) {\n      throw new ValueError(\n          'Container instance unexpectedly contains _trainableWeights.' +\n          'The trainable weights of a Container are a union of the ' +\n          'trainable weights of its consituent Layers. Its own ' +\n          '_trainableWeights must remain an empty Array.');\n    }\n\n    if (!this.trainable) {\n      return [];\n    }\n    let weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights = weights.concat(layer.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights.push(...layer.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const layer of this.layers) {\n        trainableWeights.push(...layer.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  /**\n   * Loads all layer weights from a JSON object.\n   *\n   * Porting Note: HDF5 weight files cannot be directly loaded in JavaScript /\n   *   TypeScript. The utility script at `scripts/pykeras.py` offers means\n   *   to convert them into JSON strings compatible with this method.\n   * Porting Note: TensorFlow.js Layers supports only loading by name currently.\n   *\n   * @param weights A JSON mapping weight names to weight values as nested\n   *   arrays of numbers, or a `NamedTensorMap`, i.e., a JSON mapping weight\n   *   names to `tf.Tensor` objects.\n   * @param strict Require that the provided weights exactly match those\n   *   required by the container.  Default: `true`.  Passing `false` means that\n   *   extra weights and missing weights will be silently ignored.\n   */\n  loadWeights(weights: NamedTensorMap, strict = true) {\n    const nameToWeight: {[name: string]: LayerVariable} = {};\n    let totalWeightsCount = 0;\n    for (const layer of this.layers) {\n      for (const weight of layer.weights) {\n        if (nameToWeight[weight.originalName] != null) {\n          throw new ValueError(`Duplicate weight name: ${weight.originalName}`);\n        }\n        nameToWeight[weight.originalName] = weight;\n        totalWeightsCount++;\n      }\n    }\n\n    const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n    for (const name in weights) {\n      // TF 2.2.0 added cell name to the weight name in the format of\n      // layer_name/cell_name/weight_name, we need to remove\n      // the inner cell name.\n      let validatedName = name;\n      if (nameToWeight[name] == null) {\n        const tokens = name.split('/');\n        const shortenNameArray =\n            tokens.slice(0, -2).concat([tokens[tokens.length - 1]]);\n        validatedName = shortenNameArray.join('/');\n      }\n      if (nameToWeight[validatedName] != null) {\n        weightValueTuples.push([nameToWeight[validatedName], weights[name]]);\n      } else if (strict) {\n        throw new ValueError(\n            `Provided weight data has no target variable: ${name}`);\n      }\n      delete nameToWeight[validatedName];\n    }\n\n    if (strict) {\n      // Check that all weights are set.\n      const unsetNames: string[] = [];\n      for (const name in nameToWeight) {\n        unsetNames.push(name);\n      }\n      if (unsetNames.length > 0) {\n        throw new ValueError(\n            `${unsetNames.length} of ${\n                totalWeightsCount} weights are not set: ` +\n            `${unsetNames}`);\n      }\n    }\n\n    batchSetValue(weightValueTuples);\n  }\n\n  /**\n   * Util shared between different serialization methods.\n   * @returns LayersModel config with Keras version information added.\n   */\n  protected updatedConfig(): serialization.ConfigDict {\n    const theConfig = this.getConfig();\n    const modelConfig: serialization.ConfigDict = {};\n    modelConfig['className'] = this.getClassName();\n    modelConfig['config'] = theConfig;\n    modelConfig['kerasVersion'] = `tfjs-layers ${layersVersion}`;\n    // TODO(nielsene): Replace something like K.backend() once\n    // possible.\n    modelConfig['backend'] = 'TensorFlow.js';\n    return modelConfig;\n  }\n\n  /**\n   * Returns a JSON string containing the network configuration.\n   *\n   * To load a network from a JSON save file, use\n   * models.modelFromJSON(jsonString);\n   * @param extraJsonArgs Unused in tfjs-layers, maintained for PyKeras\n   * @param returnString Whether the return value should be stringified\n   *    (default: `true`).\n   * @returns a JSON string if `returnString` (default), or a JSON object if\n   *   `!returnString`.\n   */\n  // tslint:disable-next-line:no-any\n  toJSON(unused?: any, returnString = true): string|PyJsonDict {\n    const modelConfig = convertTsToPythonic(this.updatedConfig()) as PyJsonDict;\n    return returnString ? JSON.stringify(modelConfig) : modelConfig;\n  }\n\n  /**\n   * Call the model on new inputs.\n   *\n   * In this case `call` just reapplies all ops in the graph to the new inputs\n   * (e.g. build a new computational graph from the provided inputs).\n   *\n   * @param inputs A tensor or list of tensors.\n   * @param mask A mask or list of masks. A mask can be either a tensor or null\n   *   (no mask).\n   *\n   * @return A tensor if there is a single output, or a list of tensors if there\n   *   are more than one outputs.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      const feedDict = new FeedDict();\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n      return execute(this.outputs, feedDict, kwargs) as Tensor | Tensor[];\n    });\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      let masks: Tensor[];\n      if (mask == null) {\n        masks = generic_utils.pyListRepeat(null, inputs.length);\n      } else {\n        masks = generic_utils.toList(mask);\n      }\n      // TODO(michaelterry): Add support for mask caching.\n      return this.runInternalGraph(inputs, masks)[1];\n    });\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    const inputShapes = types_utils.normalizeShapeList(inputShape);\n    if (inputShapes.length !== this.inputLayers.length) {\n      throw new ValueError(\n          `Invalid inputShape argument ${inputShape}: ` +\n          `model has ${this.inputLayers.length} tensor inputs.`);\n    }\n\n    // TODO(michaelterry): Add caching\n    const layersToOutputShapes: {[shapeKey: string]: Shape} = {};\n    for (let i = 0; i < inputShapes.length; i++) {\n      const layer = this.inputLayers[i];\n      const inputShape = inputShapes[i];\n      // It's an input layer: computeOutputShape is identity,\n      // and there is only one node and one tensor output.\n      const shapeKey = layer.name + '_0_0';\n      layersToOutputShapes[shapeKey] = inputShape;\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    // Iterate over nodes, by depth level.\n    if (depthKeys.length > 1) {\n      for (const depth of depthKeys) {\n        const nodes = this.nodesByDepth[depth];\n        for (const node of nodes) {\n          // This is always a single layer, never a list.\n          const layer = node.outboundLayer;\n          if (this.inputLayers.map(x => x.id).indexOf(layer.id) !== -1) {\n            // We've already covered the input layers a few lines above.\n            continue;\n          }\n          // Potentially redundant list, same size of node.inputTensors.\n          const inputShapes: Shape[] = [];\n          for (let j = 0; j < node.inboundLayers.length; j++) {\n            const inboundLayer = node.inboundLayers[j];\n            const nodeIndex = node.nodeIndices[j];\n            const tensorIndex = node.tensorIndices[j];\n            const shapeKey = `${inboundLayer.name}_${nodeIndex}_${tensorIndex}`;\n            const inputShape = layersToOutputShapes[shapeKey];\n            inputShapes.push(inputShape);\n          }\n\n          const outputShape = layer.computeOutputShape(\n              generic_utils.singletonOrArray(inputShapes));\n\n          const outputShapes = types_utils.normalizeShapeList(outputShape);\n          const nodeIndex = layer.inboundNodes.indexOf(node);\n          for (let j = 0; j < outputShapes.length; j++) {\n            const shapeKey = `${layer.name}_${nodeIndex}_${j}`;\n            layersToOutputShapes[shapeKey] = outputShapes[j];\n          }\n        }\n      }\n    }\n\n    // Read final output shapes from layersToOutputShapes.\n    const outputShapes: Shape[] = [];\n    const outputShapeKeys: string[] = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      const shapeKey = `${layer.name}_${nodeIndex}_${tensorIndex}`;\n      outputShapeKeys.push(shapeKey);\n    }\n\n    for (let i = 0; i < outputShapeKeys.length; i++) {\n      const key = outputShapeKeys[i];\n      generic_utils.assert(key in layersToOutputShapes);\n      outputShapes.push(layersToOutputShapes[key]);\n    }\n\n    // TODO(michaelterry): Update cache\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  /**\n   * Computes output tensors for new inputs.\n   *\n   * Note:\n   *   - Expects `inputs` to be a list (potentially with 1 element).\n   *\n   * @param inputs List of tensors\n   * @param masks List of masks (tensors or null).\n   * @return Three lists: outputTensors, outputMasks, outputShapes\n   */\n  protected runInternalGraph(inputs: Tensor[], masks?: Tensor[]):\n      [Tensor[], Tensor[], Shape[]] {\n    if (masks == null) {\n      masks = generic_utils.pyListRepeat(null, inputs.length);\n    }\n\n    // Dictionary mapping reference tensors to tuples\n    // (computed tensor, compute mask)\n    // we assume a 1:1 mapping from tensor to mask\n    // TODO: raise exception when a `.computeMask()` call\n    // does not return a list the same size as `call`\n    const tensorMap: {[tensorID: string]: [Tensor, Tensor]} = {};\n    for (let i = 0; i < this.inputs.length; ++i) {\n      const x = this.inputs[i];\n      const y = inputs[i];\n      const mask = masks[i];\n      tensorMap[x.id] = [y, mask];\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    for (const depth of depthKeys) {\n      const nodes = this.nodesByDepth[depth];\n      for (const node of nodes) {\n        // This is always a single layer, never a list.\n        const layer = node.outboundLayer;\n        const referenceInputTensors = node.inputTensors;\n        const referenceOutputTensors = node.outputTensors;\n\n        // If all previous input tensors are available in tensorMap,\n        // then call node.inboundLayer on them.\n        // List of tuples [input, mask]:\n        const computedData = new Array<[Tensor, Tensor]>();\n        for (const x of referenceInputTensors) {\n          if (x.id in tensorMap) {\n            computedData.push(tensorMap[x.id]);\n          }\n        }\n        if (computedData.length === referenceInputTensors.length) {\n          // TODO(michaelterry): Add K.name_scope here, if we need it.\n          let kwargs: Kwargs = {};\n          let computedTensors: Tensor[];\n          let computedMasks: Tensor[];\n          let outputTensors: Tensor[];\n          let outputMasks: Tensor[];\n          // call layer\n          if (node.callArgs != null) {\n            kwargs = node.callArgs;\n          }\n          if (computedData.length === 1) {\n            const [computedTensor, computedMask] = computedData[0];\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMask;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensor, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensor, computedMask));\n            computedTensors = [computedTensor];\n            computedMasks = [computedMask];\n          } else {\n            computedTensors = computedData.map(x => x[0]);\n            computedMasks = computedData.map(x => x[1]);\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMasks;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensors, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensors, computedMasks));\n          }\n\n          if (layer.activityRegularizer) {\n            throw new NotImplementedError(\n                'LayersModel invocation with concrete Tensor value(s) in the ' +\n                'presence of activity regularizer(s) is not supported yet.');\n          }\n          // TODO(michaelterry): Add model updates and losses\n\n          // Update tensor map.\n          for (let i = 0; i < referenceOutputTensors.length; ++i) {\n            const x = referenceOutputTensors[i];\n            const y = outputTensors[i];\n            const mask = outputMasks[i];\n            tensorMap[x.id] = [y, mask];\n          }\n        }\n      }\n    }\n\n    const outputTensors: Tensor[] = [];\n    const outputMasks: Tensor[] = [];\n    const outputShapes: Shape[] = [];\n    for (const x of this.outputs) {\n      generic_utils.assert(\n          x.id in tensorMap, `Could not compute output ${x.name} : ${x.id}`);\n      const [tensor, mask] = tensorMap[x.id];\n      outputShapes.push(tensor.shape);\n      outputTensors.push(tensor);\n      outputMasks.push(mask);\n    }\n\n    // TODO(michaelterry): Add support for caches.\n    return [outputTensors, outputMasks, outputShapes];\n  }\n\n  /**\n   * Builds a map of internal node keys to node ordering.\n   * Used in serializaion a node orderings may change as unused nodes are\n   * dropped. Porting Note:  This helper method was pulled out of getConfig to\n   * improve readability.\n   * @param layers An array of Layers in the model.\n   * @returns Map of Node Keys to index order within the layer.\n   */\n  private buildNodeConversionMap(layers: Layer[]): {[nodeKey: string]: number} {\n    const nodeConversionMap: {[nodeKey: string]: number} = {};\n    let keptNodes: number;\n    for (const layer of this.layers) {\n      keptNodes = layer instanceof Container ? 1 : 0;\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        if (this.containerNodes.has(nodeKey)) {\n          // i.e. we mark it to be saved\n          nodeConversionMap[nodeKey] = keptNodes;\n          keptNodes += 1;\n        }\n      }\n    }\n    return nodeConversionMap;\n  }\n\n  /**\n   * Retrieves a layer based on either its name (unique) or index.\n   *\n   * Indices are based on order of horizontal graph traversal (bottom-up).\n   *\n   * If both `name` and `index` are specified, `index` takes precedence.\n   *\n   * @param name Name of layer.\n   * @param index Index of layer.\n   * @returns A Layer instance.\n   * @throws ValueError: In case of invalid layer name or index.\n   *\n   * @doc {\n   *    heading: 'Layers',\n   *    subheading: 'Classes',\n   *    namespace: 'layers',\n   *    subclasses: ['LayersModel']\n   * }\n   */\n  getLayer(name?: string, index?: number): Layer {\n    if (index != null) {\n      if (this.layers.length <= index) {\n        throw new ValueError(\n            `Was asked to retrieve layer at index ${index}, but model only ` +\n            `has ${this.layers.length} layer(s).`);\n      } else {\n        return this.layers[index];\n      }\n    } else {\n      if (name == null) {\n        throw new ValueError('Provide either a layer name or layer index');\n      }\n    }\n\n    for (const layer of this.layers) {\n      if (layer.name === name) {\n        return layer;\n      }\n    }\n    throw new ValueError(`No such layer: ${name}`);\n  }\n\n  /**\n   * Retrieves the Container's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Container.loss in PyKeras.\n    //   In PyKeras, Container.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return tidy(() => {\n      const losses: Scalar[] = [];\n      for (const layer of this.layers) {\n        for (let nodeIndex = 0; nodeIndex < layer.inboundNodes.length;\n             ++nodeIndex) {\n          const nodeKey = Container.nodeKey(layer, nodeIndex);\n          if (this.containerNodes.has(nodeKey)) {\n            losses.push(...layer.calculateLosses());\n          }\n        }\n      }\n      // TODO(cais): Add any unconditional model-level losses?\n      return losses;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {name: this.name};\n\n    // Build a map from layer unique name (self._node_key)\n    // to the index of the nodes that are saved in the config.\n    // Only nodes in container_nodes are saved.\n    const nodeConversionMap: {[nodeKey: string]: number} =\n        this.buildNodeConversionMap(this.layers);\n\n    // Serialize and save the layers in layerConfigs\n    const layerConfigs = [];\n    for (const layer of this.layers) {\n      const layerClassName = layer.getClassName();\n      const layerConfig = layer.getConfig();\n      const filteredInboundNodes = [];\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const node = layer.inboundNodes[originalNodeIndex];\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        let kwargs = {};\n        if (this.containerNodes.has(nodeKey)) {\n          // The node is relevant to the model:\n          // add to filteredInboundNodes.\n          if (node.callArgs) {\n            try {\n              JSON.stringify(node.callArgs);\n              kwargs = node.callArgs;\n            } catch (err) {\n              console.warn(\n                  `Layer ${layer.name} was passed ` +\n                  `non-serializable keyword arguments: ` +\n                  `${node.callArgs}. They will not be included ` +\n                  `in the serialized model (and thus will be ` +\n                  `missing at deserialization time).`);\n              kwargs = {};\n            }\n          }\n          if (node.inboundLayers.length > 0) {\n            const nodeData = [];\n            for (let i = 0; i < node.inboundLayers.length; i++) {\n              const inboundLayer = node.inboundLayers[i];\n              const nodeIndex = node.nodeIndices[i];\n              const tensorIndex = node.tensorIndices[i];\n              const nodeKey = Container.nodeKey(inboundLayer, nodeIndex);\n              let newNodeIndex = nodeConversionMap[nodeKey];\n              if (newNodeIndex == null) {\n                newNodeIndex = 0;\n              }\n              nodeData.push(\n                  [inboundLayer.name, newNodeIndex, tensorIndex, kwargs]);\n            }\n            filteredInboundNodes.push(nodeData);\n          }\n        }\n      }\n      const dict: serialization.ConfigDict = {};\n      dict['name'] = layer.name;\n      dict['className'] = layerClassName;\n      dict['config'] = layerConfig;\n      dict['inboundNodes'] = filteredInboundNodes;\n      layerConfigs.push(dict);\n    }\n    config['layers'] = layerConfigs;\n    // Gather info about inputs and outputs\n    const modelInputs = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      const nodeIndex = this.inputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.inputLayersTensorIndices[i];\n      modelInputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['inputLayers'] = modelInputs;\n\n    const modelOutputs = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      modelOutputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['outputLayers'] = modelOutputs;\n    return config;\n  }\n\n  /**\n   * Instantiates a LayersModel from its config (output of `get_config()`).\n   * @param cls the class to create\n   * @param config LayersModel config dictionary.\n   * @param customObjects An optional dictionary of custom objects.\n   * @param fastWeightInit Optional flag to use fast weight initialization\n   *   during deserialization. This is applicable to cases in which\n   *   the initialization will be immediately overwritten by loaded weight\n   *   values. Default: `false`.\n   * @returns A LayersModel instance.\n   * @throws ValueError: In case of improperly formatted config dict.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    // Layer instances created during\n    // the graph reconstruction process\n    const createdLayers: {[layerName: string]: Layer} = {};\n\n    // Dictionary mapping layer instances to\n    // node data that specifies a layer call.\n    // It acts as a queue that maintains any unprocessed\n    // layer call until it becomes possible to process it\n    // (i.e. until the input tensors to the call all exist).\n    const unprocessedNodes: {[layer: string]: TensorKeyWithArgsArray[][]} = {};\n    function addUnprocessedNode(\n        layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      if (!(layer.name in unprocessedNodes)) {\n        unprocessedNodes[layer.name] = [nodeData];\n      } else {\n        unprocessedNodes[layer.name].push(nodeData);\n      }\n    }\n\n    function processNode(layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      const inputTensors: SymbolicTensor[] = [];\n      let kwargs;\n      for (const inputData of nodeData) {\n        const inboundLayerName = inputData[0];\n        const inboundNodeIndex = inputData[1];\n        const inboundTensorIndex = inputData[2];\n\n        kwargs = inputData[3] == null ?\n            {} :\n            inputData[3] as serialization.ConfigDict;\n        if (!(inboundLayerName in createdLayers)) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundLayer = createdLayers[inboundLayerName];\n        if (inboundLayer.inboundNodes.length <= inboundNodeIndex) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundNode = inboundLayer.inboundNodes[inboundNodeIndex];\n        inputTensors.push(inboundNode.outputTensors[inboundTensorIndex]);\n      }\n      // Call layer on its inputs, thus creating the node\n      // and building the layer if needed.\n      // Note: This has Eager vs Graph Implications.\n      if (inputTensors.length > 0) {\n        layer.apply(\n            generic_utils.singletonOrArray(inputTensors),\n            kwargs);  // was ** kwargs\n      }\n    }\n\n    /**\n     * Deserialize a layer, then call it on appropriate inputs.\n     * @param layerData: layer config dict.\n     * @throws ValueError: In case of improperly formatted `layer_data`\n     * dict.\n     */\n    function processLayer(layerData: serialization.ConfigDict|null) {\n      const layerName = layerData['name'] as string;\n      // Instantiate layer.\n      const layer =\n          deserializeLayer(\n              layerData,\n              config['customObjects'] != null ?\n                  config['customObjects'] as serialization.ConfigDict :\n                  {}) as Layer;\n      layer.setFastWeightInitDuringBuild(fastWeightInit);\n      createdLayers[layerName] = layer;\n      // Gather layer inputs.\n      const inboundNodesData =\n          layerData['inboundNodes'] as TensorKeyWithArgsArray[][];\n      inboundNodesData.forEach(nodeData => {\n        if (!(nodeData instanceof Array)) {\n          throw new ValueError(\n              `Corrupted configuration, expected array for nodeData: ${\n                  nodeData}`);\n        }\n        // We don't process nodes (i.e. make layer calls)\n        // on the fly because the inbound node may not yet exist,\n        // in case of layer shared at different topological depths\n        // (e.g.a model such as A(B(A(B(x)))))\n        addUnprocessedNode(layer, nodeData);\n      });\n    }\n\n    // First, we create all layers and enqueue nodes to be processed.\n    const name = config['name'];\n    const layersFromConfig = config['layers'] as serialization.ConfigDict[];\n    for (const layerData of layersFromConfig) {\n      processLayer(layerData);\n    }\n\n    // Then we process nodes in order of layer depth.\n    // Nodes that cannot yet be processed(if the inbound node\n    // does not yet exist) are re - enqueued, and the process\n    // is repeated until all nodes are processed.\n    while (!generic_utils.isObjectEmpty(unprocessedNodes)) {\n      for (const layerData of layersFromConfig) {\n        const layer = createdLayers[layerData['name'] as string];\n        if (layer.name in unprocessedNodes) {\n          const currentUnprocessedNodesForLayer = unprocessedNodes[layer.name];\n          delete unprocessedNodes[layer.name];\n          for (const nodeData of currentUnprocessedNodesForLayer) {\n            processNode(layer, nodeData);\n          }\n        }\n      }\n    }\n\n    const inputTensors: SymbolicTensor[] = [];\n    const outputTensors: SymbolicTensor[] = [];\n    const inputLayersFromConfig =\n        config['inputLayers'] as serialization.ConfigDict[];\n    for (const layerData of inputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      inputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    const outputLayersFromConfig =\n        config['outputLayers'] as serialization.ConfigDict[];\n    for (const layerData of outputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      outputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    return new cls({inputs: inputTensors, outputs: outputTensors, name});\n  }\n\n  /**\n   * Determine whether the container is stateful.\n   *\n   * Porting Note: this is the equivalent of the stateful @property of\n   *   the Container class in PyKeras.\n   */\n  get stateful(): boolean {\n    // Porting Note: This check is to prevent inadvertent setting of the\n    //   _stateful property of the Container instance.\n    if (this._stateful) {\n      throw new ValueError(\n          'Container instance unexpectedly has _stateful = true. The ' +\n          'statefulness of a Container is determined by the Layers it ' +\n          'contains. Its _stateful property must remain the default false.');\n    }\n    for (const layer of this.layers) {\n      if (layer.stateful) {\n        return true;\n      }\n    }\n    return false;\n  }\n\n  /**\n   * Reset the state of all stateful constituent layers (if any).\n   *\n   * Examples of stateful layers include RNN layers whose `stateful` property\n   * is set as `true`.\n   */\n  resetStates() {\n    tidy(() => {\n      this.layers.forEach(layer => {\n        // tslint:disable:no-any\n        if (layer.stateful) {\n          layer.resetStates();\n        }\n        // tslint:enable:no-any\n      });\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, mul, Tensor, Tensor1D, tensor1d, tidy} from '@tensorflow/tfjs-core';\n\n/**\n * For multi-class classification problems, this object is designed to store a\n * mapping from class index to the \"weight\" of the class, where higher weighted\n * classes have larger impact on loss, accuracy, and other metrics.\n *\n * This is useful for cases in which you want the model to \"pay more attention\"\n * to examples from an under-represented class, e.g., in unbalanced datasets.\n */\nexport type ClassWeight = {\n  [classIndex: number]: number\n};\n\n/**\n * Class weighting for a model with multiple outputs.\n *\n * This object maps each output name to a class-weighting object.\n */\nexport type ClassWeightMap = {\n  [outputName: string]: ClassWeight\n};\n\nfunction standardizeSampleOrClassWeights(\n    xWeight: ClassWeight|ClassWeight[]|ClassWeightMap, outputNames: string[],\n    weightType: 'sampleWeight'|'classWeight'): ClassWeight[] {\n  const numOutputs = outputNames.length;\n  if (xWeight == null || (Array.isArray(xWeight) && xWeight.length === 0)) {\n    return outputNames.map(name => null);\n  }\n  if (numOutputs === 1) {\n    if (Array.isArray(xWeight) && xWeight.length === 1) {\n      return xWeight;\n    } else if (typeof xWeight === 'object' && outputNames[0] in xWeight) {\n      return [(xWeight as ClassWeightMap)[outputNames[0]]];\n    } else {\n      return [xWeight as ClassWeight];\n    }\n  }\n  if (Array.isArray(xWeight)) {\n    if (xWeight.length !== numOutputs) {\n      throw new Error(\n          `Provided ${weightType} is an array of ${xWeight.length} ` +\n          `element(s), but the model has ${numOutputs} outputs. ` +\n          `Make sure a set of weights is provided for each model output.`);\n    }\n    return xWeight;\n  } else if (\n      typeof xWeight === 'object' && Object.keys(xWeight).length > 0 &&\n      typeof (xWeight as ClassWeightMap)[Object.keys(xWeight)[0]] ===\n          'object') {\n    const output: ClassWeight[] = [];\n    outputNames.forEach(outputName => {\n      if (outputName in xWeight) {\n        output.push((xWeight as ClassWeightMap)[outputName]);\n      } else {\n        output.push(null);\n      }\n    });\n    return output;\n  } else {\n    throw new Error(\n        `The model has multiple (${numOutputs}) outputs, ` +\n        `so ${weightType} must be either an array with ` +\n        `${numOutputs} elements or an object with ${outputNames} keys. ` +\n        `Provided ${weightType} not understood: ${JSON.stringify(xWeight)}`);\n  }\n}\n\n/**\n * Standardize class weighting objects.\n *\n * This function takes a single class-weighting object, an array of them,\n * or a map from output name to class-weighting object. It compares it to the\n * output name(s) of the model, base on which it outputs an array of\n * class-weighting objects of which the length matches the number of outputs.\n *\n * @param classWeight Input class-weighting object(s).\n * @param outputNames All output name(s) of the model.\n * @return An array of class-weighting objects. The length of the array matches\n *   the model's number of outputs.\n */\nexport function standardizeClassWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'classWeight');\n}\n\nexport function standardizeSampleWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'sampleWeight');\n}\n\n/**\n * Standardize by-sample and/or by-class weights for training.\n *\n * Note that this function operates on one model output at a time. For a model\n * with multiple outputs, you must call this function multiple times.\n *\n * @param y The target tensor that the by-sample and/or by-class weight is for.\n *     The values of y are assumed to encode the classes, either directly\n *     as an integer index, or as one-hot encoding.\n * @param sampleWeight By-sample weights.\n * @param classWeight By-class weights: an object mapping class indices\n *     (integers) to a weight (float) to apply to the model's loss for the\n *     samples from this class during training. This can be useful to tell the\n *     model to \"pay more attention\" to samples from an under-represented class.\n * @param sampleWeightMode The mode for the sample weights.\n * @return A Promise of weight tensor, of which the size of the first dimension\n *     matches that of `y`.\n */\nexport async function standardizeWeights(\n    y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight,\n    sampleWeightMode?: 'temporal'): Promise<Tensor> {\n  if (sampleWeight != null || sampleWeightMode != null) {\n    // TODO(cais): Once 'temporal' mode is implemented, document it in the doc\n    // string.\n    throw new Error('Support sampleWeight is not implemented yet');\n  }\n\n  if (classWeight != null) {\n    // Apply class weights per sample.\n    const yClasses: Tensor1D = tidy(() => {\n      if (y.shape.length === 1) {\n        // Assume class indices.\n        return y.clone() as Tensor1D;\n      } else if (y.shape.length === 2) {\n        if (y.shape[1] > 1) {\n          // Assume one-hot encoding of classes.\n          const axis = 1;\n          return y.argMax(axis);\n        } else if (y.shape[1] === 1) {\n          // Class index.\n          return y.reshape([y.shape[0]]);\n        } else {\n          throw new Error(\n              `Encountered unexpected last-dimension size (${y.shape[1]}) ` +\n              `during handling of class weights. The size is expected to be ` +\n              `>= 1.`);\n        }\n      } else {\n        throw new Error(\n            `Unexpected rank of target (y) tensor (${y.rank}) during ` +\n            `handling of class weights. The rank is expected to be 1 or 2.`);\n      }\n    });\n\n    const yClassIndices = Array.from(await yClasses.data());\n    dispose(yClasses);\n    const classSampleWeight: number[] = [];\n    yClassIndices.forEach(classIndex => {\n      if (classWeight[classIndex] == null) {\n        throw new Error(\n            `classWeight must contain all classes in the training data. ` +\n            `The class ${classIndex} exists in the data but not in ` +\n            `classWeight`);\n      } else {\n        classSampleWeight.push(classWeight[classIndex]);\n      }\n    });\n\n    return tensor1d(classSampleWeight, 'float32');\n  } else {\n    return null;\n  }\n}\n\n/**\n * Apply per-sample weights on the loss values from a number of samples.\n *\n * @param losses Loss tensor of shape `[batchSize]`.\n * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.\n * @returns Tensor of the same shape as`losses`.\n */\nexport function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor) {\n  return mul(losses, sampleWeights);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using TensorFlow.js datasets.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar} from '@tensorflow/tfjs-core';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {TensorOrArrayOrMap} from '../types';\nimport {singletonOrArray, toList} from '../utils/generic_utils';\n\nimport {Dataset, LazyIterator} from './dataset_stub';\nimport {ClassWeight, ClassWeightMap, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Interface for configuring model training based on a dataset object.\n */\nexport interface ModelFitDatasetArgs<T> {\n  /**\n   * (Optional) Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. It should\n   * typically be equal to the number of samples of your dataset divided by\n   * the batch size, so that `fitDataset`() call can utilize the entire dataset.\n   * If it is not provided, use `done` return value in `iterator.next()` as\n   * signal to finish an epoch.\n   */\n  batchesPerEpoch?: number;\n\n  /**\n   * Integer number of times to iterate over the training dataset.\n   */\n  epochs: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be any of the following:\n   *\n   *   - An array `[xVal, yVal]`, where the two values may be `tf.Tensor`,\n   *     an array of Tensors, or a map of string to Tensor.\n   *   - Similarly, an array ` [xVal, yVal, valSampleWeights]`\n   *     (not implemented yet).\n   *   - a `Dataset` object with elements of the form `{xs: xVal, ys: yVal}`,\n   *     where `xs` and `ys` are the feature and label tensors, respectively.\n   *\n   * If `validationData` is an Array of Tensor objects, each `tf.Tensor` will be\n   * sliced into batches during validation, using the parameter\n   * `validationBatchSize` (which defaults to 32). The entirety of the\n   * `tf.Tensor` objects will be used in the validation.\n   *\n   * If `validationData` is a dataset object, and the `validationBatches`\n   * parameter is specified, the validation will use `validationBatches` batches\n   * drawn from the dataset object. If `validationBatches` parameter is not\n   * specified, the validation will stop when the dataset is exhausted.\n   *\n   * The model will not be trained on this data.\n   */\n  validationData?: [\n    TensorOrArrayOrMap, TensorOrArrayOrMap\n  ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|Dataset<T>;\n\n  /**\n   * Optional batch size for validation.\n   *\n   * Used only if `validationData` is an array of `tf.Tensor` objects, i.e., not\n   * a dataset object.\n   *\n   * If not specified, its value defaults to 32.\n   */\n  validationBatchSize?: number;\n\n  /**\n   * (Optional) Only relevant if `validationData` is specified and is a dataset\n   * object.\n   *\n   * Total number of batches of samples to draw from `validationData` for\n   * validation purpose before stopping at the end of every epoch. If not\n   * specified, `evaluateDataset` will use `iterator.next().done` as signal to\n   * stop validation.\n   */\n  validationBatches?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - a `number`: Will yield every `number` milliseconds.\n   *   - `'never'`: never yield. (But yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n}\n\nexport interface FitDatasetElement {\n  xs: TensorOrArrayOrMap;\n  ys: TensorOrArrayOrMap;\n}\n\n/**\n * Interface for configuring model evaluation based on a dataset object.\n */\nexport interface ModelEvaluateDatasetArgs {\n  /**\n   * Number of batches to draw from the dataset object before ending the\n   * evaluation.\n   */\n  batches?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n}\n\n// Default batch size used during tensor-based validation.\nconst DEFAULT_VALIDATION_BATCH_SIZE = 32;\n\n/**\n * Standardize the output of a dataset iterator for use by\n * LayersModel.fitDataset().\n *\n * @param model: A `tf.LayersModel` object.\n * @param iteratorOut The output of a dataset iterator. It is required to be\n *   an object of the form `{xs: TensorOrArrayOrMap, ys:\n * TensorOrArrayOrMap}`, where `TensorOrArrayOrMap` is a single `tf.Tensor`,\n * a `tf.Tensor[]`, or a flat map from string names to `tf.Tensor`s.\n * @returns A flat array of `tf.Tensor` objects: the input `tf.Tensor`s\n *   followed by the target `tf.Tensor`s.  When `tf.Tensor`s are provided\n *   as a map, the order in the resulting array is taken from the `inputNames`\n *   and `outputNames` of the model.\n */\nfunction standardizeDataIteratorOutput(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, iteratorOut: {}): {xs: tfc.Tensor[], ys: tfc.Tensor[]} {\n  let xs: TensorOrArrayOrMap;\n  let ys: TensorOrArrayOrMap;\n\n  const iteratorOutObj = iteratorOut as FitDatasetElement;\n  xs = iteratorOutObj['xs'];\n  ys = iteratorOutObj['ys'];\n  tfc.util.assert(\n      xs != null && ys != null,\n      () => 'A Dataset iterator for fitDataset() is expected to generate ' +\n          'objects of the form `{xs: xVal, ys: yVal}`, where the two ' +\n          'values may be `tf.Tensor`, an array of Tensors, or a map of ' +\n          'string to Tensor.  The provided Dataset instead generates ' +\n          `${iteratorOut}`);\n\n  const flattenedXs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('input', model.inputNames, xs);\n  const flattenedYs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('output', model.outputNames, ys);\n\n  const batchSize: number = flattenedXs[0].shape[0];\n\n  tfc.util.assert(\n      flattenedXs.length === model.inputs.length,\n      () => `LayersModel has ${model.inputs.length} inputs, but the dataset ` +\n          `provides ${flattenedXs.length} inputs.  (Expected input keys: ` +\n          `${JSON.stringify(model.inputNames)})`);\n\n  tfc.util.assert(\n      flattenedYs.length === model.outputs.length,\n      () =>\n          `LayersModel has ${model.outputs.length} outputs, but the dataset ` +\n          `provides ${flattenedYs.length} outputs.  (Expected output keys: ` +\n          `${JSON.stringify(model.outputNames)})`);\n\n  for (let xIndex = 0; xIndex < flattenedXs.length; xIndex++) {\n    tfc.util.assert(\n        flattenedXs[xIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: input ` +\n            `${model.inputNames[xIndex]} has ${\n                  flattenedXs[xIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  for (let yIndex = 0; yIndex < flattenedYs.length; yIndex++) {\n    tfc.util.assert(\n        flattenedYs[yIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: output ` +\n            `${model.outputNames[yIndex]} has ${\n                  flattenedYs[yIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  return {xs: flattenedXs, ys: flattenedYs};\n}\n\nfunction flattenTensorOrArrayOrMap(\n    inputOrOutput: string, names: string[], values: TensorOrArrayOrMap) {\n  if (values instanceof tfc.Tensor) {\n    return [values];\n  } else if (Array.isArray(values)) {\n    tfc.util.assert(\n        values.length === names.length,\n        () => `Received an array of ${values.length} Tensors, but expected ${\n            names.length} to match the ${inputOrOutput} keys ${names}.`);\n    return values;\n  } else {\n    const result: tfc.Tensor[] = [];\n    // Check that all the required keys are available.\n    for (const name of names) {\n      if (values[name] == null) {\n        throw new ValueError(\n            `The feature data generated by the dataset lacks the required ` +\n            `${inputOrOutput} key '${name}'.`);\n      }\n      result.push(values[name]);\n    }\n    return result;\n  }\n}\n\nfunction standardizeTensorValidationData<T>(\n    data:\n        [\n          tfc.Tensor|tfc.Tensor[], tfc.Tensor|tfc.Tensor[]\n        ]|[tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n           tfc.Tensor | tfc.Tensor[]]):\n    {xs: tfc.Tensor|tfc.Tensor[], ys: tfc.Tensor|tfc.Tensor[]} {\n  if (data.length === 3) {\n    throw new NotImplementedError(\n        'Validation with sample weights is not implemented yet.');\n  }\n  return {xs: data[0], ys: data[1]};\n}\n\nexport async function fitDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>,\n    args: ModelFitDatasetArgs<T>): Promise<History> {\n  const hasBatchesPerEpoch = args.batchesPerEpoch != null;\n  tfc.util.assert(\n      model.optimizer != null,\n      () => 'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileConfig).');\n\n  tfc.util.assert(\n      args != null,\n      () => `For fitDataset(), the 2nd argument (config) is required, ` +\n          `but it is not provided in this call.`);\n  tfc.util.assert(\n      args.epochs != null && args.epochs > 0 && Number.isInteger(args.epochs),\n      () => `For fitDataset(), config.epochs is expected to be a positive ` +\n          `integer, but got ${args.epochs}`);\n  tfc.util.assert(\n      !hasBatchesPerEpoch ||\n          (args.batchesPerEpoch > 0 && Number.isInteger(args.batchesPerEpoch)),\n      () => `For fitDataset(), config.batchesPerEpoch is expected to be a ` +\n          `positive integer if specified, but got ${args.batchesPerEpoch}`);\n  tfc.util.assert(\n      // tslint:disable-next-line:no-any\n      (args as any)['validationSplit'] == null,\n      () => '`validationSplit` is not supported by `fitDataset()`. ' +\n          'Use validationData instead.');\n\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n\n  try {\n    const doValidation = args.validationData != null;\n    let valXs: tfc.Tensor|tfc.Tensor[];\n    let valYs: tfc.Tensor|tfc.Tensor[];\n    if (doValidation) {\n      if (isDatasetObject(args.validationData)) {\n        tfc.util.assert(\n            args.validationBatches == null ||\n                (args.validationBatches > 0 &&\n                 Number.isInteger(args.validationBatches)),\n            () => `For fitDataset() with dataset-based validation, ` +\n                `config.validationBatches is expected not to be provided, ` +\n                `or to be a positive integer, ` +\n                `but got ${args.validationBatches}`);\n      } else {\n        const validationData = standardizeTensorValidationData(\n            args.validationData as\n                    [tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[]] |\n            [\n              tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n              tfc.Tensor | tfc.Tensor[]\n            ]);\n        valXs = validationData.xs;\n        valYs = validationData.ys;\n      }\n    }\n\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let callbackMetrics: string[];\n    if (doValidation) {\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const verbose = args.verbose == null ? 1 : args.verbose;\n    const {callbackList, history} = configureCallbacks(\n        callbacks, verbose, args.epochs, null, null,\n        getStepsPerEpoch(dataset, args),\n        null,  // Batch size determined by the dataset itself.\n        doValidation, callbackMetrics);\n    callbackList.setModel(model);\n    model.history = history;\n\n    await callbackList.onTrainBegin();\n    model.stopTraining_ = false;\n    let epoch = args.initialEpoch == null ? 0 : args.initialEpoch;\n\n    let dataIterator = await dataset.iterator();\n    while (epoch < args.epochs) {\n      const epochLogs: UnresolvedLogs = {};\n      await callbackList.onEpochBegin(epoch);\n      let stepsDone = 0;\n      let batchIndex = 0;\n      if (!hasBatchesPerEpoch) {\n        dataIterator = await dataset.iterator();\n      }\n      while (hasBatchesPerEpoch ? stepsDone < args.batchesPerEpoch : true) {\n        const iteratorOut = await dataIterator.next();\n\n        // If `batchesPerEpoch` is specified, the dataset should not be\n        // exhausted until all epoches are done.\n        if (hasBatchesPerEpoch && iteratorOut.done) {\n          console.warn(\n              'You provided `batchesPerEpoch` as ' +\n              `${args.batchesPerEpoch}, ` +\n              'but your dataset iterator ran out of data after ' +\n              `${stepsDone} batches; ` +\n              'interrupting training. Make sure that your ' +\n              'dataset can generate at least `batchesPerEpoch * epochs` ' +\n              'batches (in this case, ' +\n              `${args.batchesPerEpoch * args.epochs} batches). ` +\n              'You may need to use the repeat() function when building ' +\n              'your dataset.');\n          break;\n        }\n\n        if (iteratorOut.value != null) {\n          const {xs, ys} =\n              standardizeDataIteratorOutput(model, iteratorOut.value);\n          const batchLogs: UnresolvedLogs = {};\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = xs[0].shape[0];\n\n          await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n          const sampleWeights: tfc.Tensor[] = [];\n          if (args.classWeight != null) {\n            const standardClassWeights =\n                standardizeClassWeights(args.classWeight, model.outputNames);\n            for (let i = 0; i < standardClassWeights.length; ++i) {\n              sampleWeights.push(await standardizeWeights(\n                  ys[i], null, standardClassWeights[i]));\n            }\n          }\n\n          // Train on batch.\n          const ins = xs.concat(ys).concat(sampleWeights);\n          const outs = trainFunction(ins);\n          tfc.dispose(ins);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n          }\n\n          await callbackList.onBatchEnd(batchIndex, batchLogs);\n          disposeTensorsInLogs(batchLogs);\n\n          batchIndex++;\n          stepsDone++;\n        }\n\n        if (hasBatchesPerEpoch ? stepsDone >= args.batchesPerEpoch :\n                                 iteratorOut.done) {\n          // Epoch finished. Perform validation.\n          if (doValidation) {\n            let valOuts: tfc.Scalar[];\n            if (isDatasetObject(args.validationData)) {\n              valOuts = toList(await model.evaluateDataset(\n                  args.validationData, {batches: args.validationBatches}));\n            } else {\n              valOuts = toList(model.evaluate(valXs, valYs, {\n                batchSize: args.validationBatchSize == null ?\n                    DEFAULT_VALIDATION_BATCH_SIZE :\n                    args.validationBatchSize,\n                verbose: 0\n              }));\n            }\n            for (let i = 0; i < model.metricsNames.length; ++i) {\n              epochLogs[`val_${model.metricsNames[i]}`] = valOuts[i];\n            }\n          }\n          // Call `break` to exit one epoch lopp after validation is done. If\n          // config.batchesPerEpoch is specified, an epoch while loop will\n          // stop when `stepsDone >= config.batchesPerEpoch`. When\n          // config.batchesPerEpoch is not provided, the following `break` is\n          // required to exit the while lopp after dataset is exhausted.\n          break;\n        }\n\n        if (model.stopTraining_) {\n          break;\n        }\n      }\n      await callbackList.onEpochEnd(epoch, epochLogs);\n      epoch++;\n      if (model.stopTraining_) {\n        break;\n      }\n    }\n    await callbackList.onTrainEnd();\n    await model.history.syncData();\n    return model.history;\n  } finally {\n    model.isTraining = false;\n  }\n}\n\n/** Helper function that determines number of steps (batches) per epoch. */\nfunction getStepsPerEpoch<T>(\n    dataset: Dataset<T>, args: ModelFitDatasetArgs<T>): number {\n  // Attempt to determine # of batches in an epoch.\n  let stepsPerEpoch: number = null;\n  if (args.batchesPerEpoch != null) {\n    stepsPerEpoch = args.batchesPerEpoch;\n  } else if (Number.isFinite(dataset.size)) {\n    stepsPerEpoch = dataset.size;\n  }\n  return stepsPerEpoch;\n}\n\n// Check if provided object is a Dataset object by checking its .iterator\n// element.\nfunction isDatasetObject<T>(\n    dataset:\n        [\n          TensorOrArrayOrMap, TensorOrArrayOrMap\n        ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|\n    Dataset<T>): boolean {\n  return (typeof (dataset as Dataset<T>).iterator === 'function');\n}\n\n// Check if provided object is a LazyIterator object by checking it's .next\n// element.\nfunction isLazyIteratorObject<T>(iterator: Dataset<T>|\n                                 LazyIterator<T>): boolean {\n  return (typeof (iterator as LazyIterator<T>).next === 'function');\n}\n\nexport async function evaluateDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>|LazyIterator<T>,\n    args: ModelEvaluateDatasetArgs): Promise<tfc.Scalar|tfc.Scalar[]> {\n  args = args || {};\n  const hasBatches = args.batches != null;\n  const f = model.testFunction;\n  let outs: tfc.Scalar[] = [];\n  if (args.verbose > 0) {\n    throw new NotImplementedError('Verbose mode is not implemented yet.');\n  }\n\n  tfc.util.assert(\n      !hasBatches || (args.batches > 0 && Number.isInteger(args.batches)),\n      () => 'Test loop expects `batches` to be a positive integer, but ' +\n          `received ${JSON.stringify(args.batches)}`);\n  const dataIterator = isLazyIteratorObject(dataset) ?\n      dataset as LazyIterator<T>:\n      await (dataset as Dataset<T>).iterator();\n  // Keeps track of number of examples used in this evaluation.\n  let numExamples = 0;\n  let batch = 0;\n\n  while (hasBatches ? batch < args.batches : true) {\n    const iteratorOut = await dataIterator.next();\n    outs = tfc.tidy(() => {\n      if (iteratorOut.value) {\n        // TODO(cais): Once real dataset is available, use\n        //   `map(x => standardizeDataIteratorOutput(model, x).map(f)`.\n        const {xs, ys} =\n            standardizeDataIteratorOutput(model, iteratorOut.value);\n        const xsAndYs = xs.concat(ys);\n        const batchOuts = tfc.tidy(() => f(xsAndYs));\n        tfc.dispose(xsAndYs);\n\n        if (batch === 0) {\n          for (let i = 0; i < batchOuts.length; ++i) {\n            outs.push(scalar(0));\n          }\n        }\n\n        const batchSize = xsAndYs[0].shape[0];\n        for (let i = 0; i < batchOuts.length; ++i) {\n          const batchOut = batchOuts[i];\n          const oldScalar = outs[i];\n          outs[i] =\n              tfc.tidy(() => tfc.add(outs[i], tfc.mul(batchSize, batchOut)));\n          if (batch > 0) {\n            tfc.dispose(oldScalar);\n          }\n        }\n        tfc.dispose(batchOuts);\n        numExamples += batchSize;\n\n        ++batch;\n      }\n      return outs;\n    });\n\n    if (iteratorOut.done) {\n      if (hasBatches) {\n        console.warn(\n            'Your dataset iterator ran out of data during evaluateDataset(). ' +\n            'Interrupting evalution. Make sure that your ' +\n            'dataset can generate at least `batches` ' +\n            `batches (in this case, ${args.batches} batches). ` +\n            'You may need to use the repeat() function when building ' +\n            'your dataset.');\n      }\n      break;\n    }\n  }\n\n  for (let i = 0; i < outs.length; ++i) {\n    const oldScalar = outs[i];\n    outs[i] = tfc.div(outs[i], numExamples);\n    tfc.dispose(oldScalar);\n  }\n\n  return singletonOrArray(outs);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using tf.Tensor objects.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Scalar, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport {expandDims, gather, sliceAlongFirstAxis} from '../backend/tfjs_backend';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {range} from '../utils/math_utils';\nimport {ClassWeight, ClassWeightMap} from './training_utils';\n\n/**\n * Interface configuration model training based on data as `tf.Tensor`s.\n */\nexport interface ModelFitArgs {\n  /**\n   * Number of samples per gradient update. If unspecified, it\n   * will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Integer number of times to iterate over the training data arrays.\n   */\n  epochs?: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Float between 0 and 1: fraction of the training data\n   * to be used as validation data. The model will set apart this fraction of\n   * the training data, will not train on it, and will evaluate the loss and\n   * any model metrics on this data at the end of each epoch.\n   * The validation data is selected from the last samples in the `x` and `y`\n   * data provided, before shuffling.\n   */\n  validationSplit?: number;\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal,\n   * valSampleWeights]. The model will not be trained on this data.\n   * `validationData` will override `validationSplit`.\n   */\n  validationData?: [\n    Tensor|Tensor[], Tensor|Tensor[]\n  ]|[Tensor | Tensor[], Tensor|Tensor[], Tensor|Tensor[]];\n\n  /**\n   * Whether to shuffle the training data before each epoch. Has\n   * no effect when `stepsPerEpoch` is not `null`.\n   */\n  shuffle?: boolean;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n\n  /**\n   * Optional array of the same length as x, containing\n   * weights to apply to the model's loss for each sample. In the case of\n   * temporal data, you can pass a 2D array with shape (samples,\n   * sequenceLength), to apply a different weight to every timestep of every\n   * sample. In this case you should make sure to specify\n   * sampleWeightMode=\"temporal\" in compile().\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. When training\n   * with Input Tensors such as TensorFlow data tensors, the default `null` is\n   * equal to the number of unique samples in your dataset divided by the\n   * batch size, or 1 if that cannot be determined.\n   */\n  stepsPerEpoch?: number;\n\n  /**\n   * Only relevant if `stepsPerEpoch` is specified. Total number of steps\n   * (batches of samples) to validate before stopping.\n   */\n  validationSteps?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - any `number`: yield every `number` milliseconds.\n   *   - `'never'`: never yield. (yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n}\n\nexport function checkBatchSize(batchSize: number) {\n  tfc.util.assert(\n      batchSize > 0 && Number.isInteger(batchSize),\n      () => `batchSize is required to be a positive integer, but got ${\n          batchSize}`);\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by start and stop indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArraysByIndices()` together.\n *\n * @param arrays: the input.\n * @param start: the starting index (inclusive).\n * @param stop: the stopping index (exclusive).\n * @returns The result of the slicing. If `arrays` is an `Array` of\n *   `tf.Tensor`s, the slicing will be applied to all elements of the `Array`\n *   in the same way.\n */\nexport function sliceArrays(\n    arrays: Tensor|Tensor[], start: number, stop: number): Tensor|Tensor[] {\n  if (arrays == null) {\n    return [null];\n  } else if (Array.isArray(arrays)) {\n    return arrays.map(array => sliceAlongFirstAxis(array, start, stop - start));\n  } else {  // Tensor.\n    return sliceAlongFirstAxis(arrays, start, stop - start);\n  }\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by random-order indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArrays()` together.\n *\n * @param arrays The input `tf.Tensor` or `Array` of `tf.Tensor`s to slice.\n *   If an `Array` of `tf.Tensor`s, all `tf.Tensor`s will be sliced in the\n *   same fashion.\n * @param indices The indices to use for slicing along the first (batch)\n *   dimension.\n * @returns Result(s) of the slicing.\n */\nexport function sliceArraysByIndices(\n    arrays: Tensor|Tensor[], indices: Tensor1D): Tensor|Tensor[] {\n  return tfc.tidy(() => {\n    if (arrays == null) {\n      return null;\n    } else if (Array.isArray(arrays)) {\n      return arrays.map(\n          array => (sliceArraysByIndices(array, indices) as Tensor));\n    } else {\n      // TODO(cais): indices should be a pre-constructed Tensor1D to avoid\n      //   tensor1d() calls.\n      return gather(\n          arrays, indices.dtype === 'int32' ? indices : indices.toInt());\n    }\n  });\n}\n\n/**\n * Returns a list of batch indices (tuples of indices).\n * @param size: Integer, total size of the data to slice into batches.\n * @param batchSize: Integer, batch size.\n * @returns An Array of [batchStart, batchEnd] tuples. batchStart is\n *   inclusive; batchEnd is exclusive. I.e., each batch consists of indices x\n *   that satisfy batchStart <= x < batchEnd.\n */\nexport function makeBatches(\n    size: number, batchSize: number): Array<[number, number]> {\n  const output: Array<[number, number]> = [];\n  let batchStart = 0;\n  let batchEnd: number = null;\n  while (batchStart < size) {\n    batchEnd = batchStart + batchSize;\n    if (batchEnd >= size) {\n      batchEnd = size;\n    }\n    output.push([batchStart, batchEnd]);\n    batchStart = batchEnd;\n  }\n  return output;\n}\n\n/**\n * Abstract fit function for `f(ins)`.\n * @param f A Function returning a list of tensors. For training, this\n *   function is expected to perform the updates to the variables.\n * @param ins List of tensors to be fed to `f`.\n * @param outLabels List of strings, display names of the outputs of `f`.\n * @param batchSize Integer batch size or `== null` if unknown. Default : 32.\n * @param epochs Number of times to iterate over the data. Default : 1.\n * @param verbose Verbosity mode: 0, 1, or 2. Default: 1.\n * @param callbacks List of callbacks to be called during training.\n * @param valF Function to call for validation.\n * @param valIns List of tensors to be fed to `valF`.\n * @param shuffle Whether to shuffle the data at the beginning of every\n * epoch. Default : true.\n * @param callbackMetrics List of strings, the display names of the metrics\n *   passed to the callbacks. They should be the concatenation of the\n *   display names of the outputs of `f` and the list of display names\n *   of the outputs of `valF`.\n * @param initialEpoch Epoch at which to start training (useful for\n *   resuming a previous training run). Default : 0.\n * @param stepsPerEpoch Total number of steps (batches on samples) before\n *   declaring one epoch finished and starting the next epoch. Ignored with\n *   the default value of `undefined` or `null`.\n * @param validationSteps Number of steps to run validation for (only if\n *   doing validation from data tensors). Not applicable for tfjs-layers.\n * @returns A `History` object.\n */\nasync function fitLoop(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, f: (data: Tensor[]) => Scalar[], ins: Tensor[],\n    outLabels?: string[], batchSize?: number, epochs?: number, verbose?: number,\n    callbacks?: BaseCallback[], valF?: (data: Tensor[]) => Scalar[],\n    valIns?: Tensor[], shuffle?: boolean|string, callbackMetrics?: string[],\n    initialEpoch?: number, stepsPerEpoch?: number,\n    validationSteps?: number): Promise<History> {\n  if (batchSize == null) {\n    batchSize = 32;\n  }\n  if (epochs == null) {\n    epochs = 1;\n  }\n  if (shuffle == null) {\n    shuffle = true;\n  }\n  if (initialEpoch == null) {\n    initialEpoch = 0;\n  }\n\n  // TODO(cais): Change const to let below when implementing validation.\n  let doValidation = false;\n  if (valF != null && valIns != null) {\n    doValidation = true;\n    // TODO(cais): verbose message.\n  }\n  if (validationSteps != null) {\n    doValidation = true;\n    if (stepsPerEpoch == null) {\n      throw new ValueError(\n          'Can only use `validationSteps` when doing step-wise training, ' +\n          'i.e., `stepsPerEpoch` must be set.');\n    }\n  }\n\n  const numTrainSamples =\n      model.checkNumSamples(ins, batchSize, stepsPerEpoch, 'steps_per_epoch');\n  let indexArray: number[];\n  if (numTrainSamples != null) {\n    indexArray = range(0, numTrainSamples);\n  }\n\n  if (verbose == null) {\n    verbose = 1;\n  }\n\n  const {callbackList, history} = configureCallbacks(\n      callbacks, verbose, epochs, initialEpoch, numTrainSamples, stepsPerEpoch,\n      batchSize, doValidation, callbackMetrics);\n  callbackList.setModel(model);\n  model.history = history;\n  await callbackList.onTrainBegin();\n  model.stopTraining_ = false;\n  // TODO(cais): Take care of callbacks.validation_data as in PyKeras.\n  // TODO(cais): Pre-convert feeds for performance as in PyKeras.\n\n  for (let epoch = initialEpoch; epoch < epochs; ++epoch) {\n    await callbackList.onEpochBegin(epoch);\n    const epochLogs: UnresolvedLogs = {};\n    if (stepsPerEpoch != null) {\n      throw new NotImplementedError(\n          'stepsPerEpoch mode is not implemented yet.');\n    } else {\n      if (shuffle === 'batch') {\n        throw new NotImplementedError('batch shuffling is not implemneted yet');\n      } else if (shuffle) {\n        util.shuffle(indexArray);\n      }\n      // Convert the potentially shuffled indices to Tensor1D, to avoid the\n      // cost of repeated creation of Array1Ds later on.\n      const epochIndexArray1D = tensor1d(indexArray);\n\n      const batches = makeBatches(numTrainSamples, batchSize);\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchLogs: UnresolvedLogs = {};\n        await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n        tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds = sliceAlongFirstAxis(\n                               epochIndexArray1D, batchStart,\n                               batchEnd - batchStart) as Tensor1D;\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = batchEnd - batchStart;\n\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Tensor[];\n          const outs = f(insBatch);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n            // TODO(cais): Use scope() to avoid ownership.\n          }\n\n          if (batchIndex === batches.length - 1) {  // Last batch.\n            if (doValidation) {\n              const valOuts = model.testLoop(valF, valIns, batchSize);\n              // Porting Notes: In tfjs-layers, valOuts is always an Array.\n              for (let i = 0; i < outLabels.length; ++i) {\n                const label = outLabels[i];\n                const out = valOuts[i];\n                tfc.keep(out);\n                // TODO(cais): Use scope() to avoid ownership.\n                epochLogs['val_' + label] = out;\n              }\n            }\n          }\n        });\n\n        await callbackList.onBatchEnd(batchIndex, batchLogs);\n        disposeTensorsInLogs(batchLogs);\n\n        if (model.stopTraining_) {\n          break;\n        }\n        // TODO(cais): return outs as list of Tensor.\n      }\n\n      epochIndexArray1D.dispose();\n    }\n    // TODO(cais): Run validation at the end of the epoch.\n    await callbackList.onEpochEnd(epoch, epochLogs);\n    if (model.stopTraining_) {\n      break;\n    }\n  }\n  await callbackList.onTrainEnd();\n\n  await model.history.syncData();\n  return model.history;\n}\n\nexport async function fitTensors(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    args: ModelFitArgs = {}): Promise<History> {\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n  let inputs: Tensor[];\n  let targets: Tensor[];\n  let inputValX: Tensor|Tensor[];\n  let inputValY: Tensor|Tensor[];\n  let valX: Tensor|Tensor[];\n  let valY: Tensor|Tensor[];\n  let sampleWeights: Tensor[];\n  try {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // Validate user data.\n    // TODO(cais): Support sampleWeight.\n    const checkBatchAxis = false;\n    const standardizedOuts =\n        await model.standardizeUserData(\n            x, y, args.sampleWeight, args.classWeight, checkBatchAxis,\n            batchSize) as [Tensor[], Tensor[], Tensor[]];\n    inputs = standardizedOuts[0];\n    targets = standardizedOuts[1];\n    sampleWeights = standardizedOuts[2];\n\n    // Prepare validation data.\n    let doValidation = false;\n    let valIns: Tensor[];\n    if (args.validationData != null && args.validationData.length > 0) {\n      doValidation = true;\n      if (args.validationData.length === 2) {\n        // config.validationData consists of valX and valY.\n        inputValX = args.validationData[0];\n        inputValY = args.validationData[1];\n      } else if (args.validationData.length === 3) {\n        throw new NotImplementedError(\n            'validationData including sample weights is not supported yet.');\n      } else {\n        throw new ValueError(\n            `When passing validation data, it must contain 2 (valX, valY) ` +\n            `or 3 (valX, valY, valSampleWeight) items; ` +\n            `${args.validationData} is invalid.`);\n      }\n\n      const checkBatchAxis = true;\n      const valStandardized =\n          await model.standardizeUserData(\n              inputValX, inputValY, null, /** Unused sample weights. */\n              null,                       /** Unused class weights. */\n              checkBatchAxis, batchSize) as [Tensor[], Tensor[], Tensor[]];\n      valX = valStandardized[0];\n      valY = valStandardized[1];\n      valIns = valX.concat(valY);\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (\n        args.validationSplit != null && args.validationSplit > 0 &&\n        args.validationSplit < 1) {\n      doValidation = true;\n      // Porting Note: In tfjs-layers, inputs[0] is always a Tensor.\n      const splitAt =\n          Math.floor(inputs[0].shape[0] * (1 - args.validationSplit));\n      const originalBatchSize = inputs[0].shape[0];\n      valX = sliceArrays(inputs, splitAt, originalBatchSize) as Tensor[];\n      inputs = sliceArrays(inputs, 0, splitAt) as Tensor[];\n      valY = sliceArrays(targets, splitAt, originalBatchSize) as Tensor[];\n      targets = sliceArrays(targets, 0, splitAt) as Tensor[];\n      // TODO(cais): Once sampleWeights becomes available, slice it to get\n      //   valSampleWeights.\n      valIns = valX.concat(valY);\n\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (args.validationSteps != null) {\n      doValidation = true;\n      // TODO(cais): Add useLearningPhase.\n    }\n\n    const ins = inputs.concat(targets).concat(sampleWeights);\n\n    model.checkTrainableWeightsConsistency();\n\n    // TODO(cais): Handle use_learning_phase and learning_phase?\n\n    // Porting Note: Here we see a key deviation of tfjs-layers from\n    // Keras.\n    //  Due to the imperative nature of tfjs-layers' backend (tfjs-core),\n    //  we do not construct symbolic computation graphs to embody the\n    //  training process. Instead, we define a function that performs the\n    //  training action. In PyKeras, the data (inputs and targets) are fed\n    //  through graph placeholders. In tfjs-layers, the data are fed as\n    //  function arguments. Since the function are defined below in the\n    //  scope, we don't have equivalents of PyKeras's\n    //  `_make_train_funciton`.\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let valFunction: (data: Tensor[]) => Scalar[];\n    let callbackMetrics: string[];\n    if (doValidation) {\n      model.makeTestFunction();\n      valFunction = model.testFunction;\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      valFunction = null;\n      valIns = [];\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const out = await fitLoop(\n        model, trainFunction, ins, outLabels, batchSize, args.epochs,\n        args.verbose, callbacks, valFunction, valIns, args.shuffle,\n        callbackMetrics, args.initialEpoch, null, null);\n    return out;\n  } finally {\n    model.isTraining = false;\n    // Memory clean up.\n    disposeNewTensors(inputs, x);\n    disposeNewTensors(targets, y);\n    disposeNewTensors(valX as Tensor[], inputValX);\n    disposeNewTensors(valY as Tensor[], inputValY);\n    if (sampleWeights != null) {\n      tfc.dispose(sampleWeights);\n    }\n  }\n  // TODO(cais): Add value to outLabels.\n}\n\n/**\n * Ensure tensors all have a rank of at least 2.\n *\n * If a tensor has a rank of 1, it is dimension-expanded to rank 2.\n * If any tensor has a rank of 0 (i.e., is a scalar), an error will be thrown.\n */\nexport function ensureTensorsRank2OrHigher(tensors: Tensor|Tensor[]): Tensor[] {\n  const outs: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    tensors = [tensors];\n  }\n\n  // Make Tensors at least 2D.\n  for (let i = 0; i < tensors.length; ++i) {\n    const tensor = tensors[i];\n    if (tensor.rank === 1) {\n      outs.push(expandDims(tensor, 1));\n    } else if (tensor.rank === 0) {\n      throw new Error(\n          'Expected tensor to be at least 1D, but received a 0D tensor ' +\n          '(scalar).');\n    } else {\n      outs.push(tensor);\n    }\n  }\n  return outs;\n}\n\n/**\n * Compare a set of tensors with a reference (old) set, discard the ones\n * in the new set that are not present in the reference set.\n *\n * This method is used for memory clenaup during calls such as\n * LayersModel.fit().\n *\n * @param tensors New set which may contain Tensors not present in\n *   `refTensors`.\n * @param refTensors Reference Tensor set.\n */\n// TODO(cais, kangyizhang): Deduplicate with tfjs-data.\nexport function disposeNewTensors(\n    tensors: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    refTensors: Tensor|Tensor[]|{[inputName: string]: Tensor}): void {\n  if (tensors == null) {\n    return;\n  }\n  const oldTensorIds: number[] = [];\n  if (refTensors instanceof Tensor) {\n    oldTensorIds.push(refTensors.id);\n  } else if (Array.isArray(refTensors)) {\n    refTensors.forEach(t => oldTensorIds.push(t.id));\n  } else if (refTensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in refTensors) {\n      const oldTensor = refTensors[name];\n      oldTensorIds.push(oldTensor.id);\n    }\n  }\n\n  const tensorsToDispose: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    if (oldTensorIds.indexOf(tensors.id) === -1) {\n      tensorsToDispose.push(tensors);\n    }\n  } else if (Array.isArray(tensors)) {\n    tensors.forEach(t => {\n      if (oldTensorIds.indexOf(t.id) === -1) {\n        tensorsToDispose.push(t);\n      }\n    });\n  } else if (tensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in tensors) {\n      const tensor = tensors[name];\n      if (oldTensorIds.indexOf(tensor.id) === -1) {\n        tensorsToDispose.push(tensor);\n      }\n    }\n  }\n\n  tensorsToDispose.forEach(t => {\n    if (!t.isDisposed) {\n      t.dispose();\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: engine/training.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {io, ModelPredictConfig as ModelPredictArgs, NamedTensorMap, Optimizer, Scalar, scalar, serialization, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {History, ModelLoggingVerbosity} from '../base_callbacks';\nimport {nameScope} from '../common';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {LossIdentifier} from '../keras_format/loss_config';\nimport {OptimizerSerialization} from '../keras_format/optimizer_config';\nimport {MetricsIdentifier, TrainingConfig} from '../keras_format/training_config';\nimport {deserialize} from '../layers/serialization';\nimport * as losses from '../losses';\nimport * as Metrics from '../metrics';\nimport * as optimizers from '../optimizers';\nimport {LossOrMetricFn, NamedTensor} from '../types';\nimport {checkUserDefinedMetadata} from '../user_defined_metadata';\nimport {count, pyListRepeat, singletonOrArray, toCamelCase, toSnakeCase, unique} from '../utils/generic_utils';\nimport {printSummary} from '../utils/layer_utils';\nimport {range} from '../utils/math_utils';\nimport {convertPythonicToTs} from '../utils/serialization_utils';\nimport {LayerVariable} from '../variables';\nimport {version} from '../version';\n\nimport {Container, ContainerArgs} from './container';\nimport {Dataset} from './dataset_stub';\nimport {execute, FeedDict} from './executor';\nimport {DisposeResult, SymbolicTensor} from './topology';\nimport {evaluateDataset, fitDataset, ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './training_dataset';\nimport {checkBatchSize, disposeNewTensors, ensureTensorsRank2OrHigher, fitTensors, makeBatches, ModelFitArgs, sliceArrays, sliceArraysByIndices} from './training_tensors';\nimport {ClassWeight, ClassWeightMap, computeWeightedLoss, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Helper function for polymorphic input data: 1. singleton Tensor.\n */\nexport function isDataTensor(x: Tensor|Tensor[]|{[inputName: string]: Tensor}|\n                             {[inputName: string]: Tensor[]}): boolean {\n  return x instanceof Tensor;\n}\n\n/**\n * Helper function for polymorphic input data: 2. Array of Tensor.\n */\nexport function isDataArray(x: Tensor|Tensor[]|\n                            {[inputName: string]: Tensor}): boolean {\n  return Array.isArray(x);\n}\n\n/**\n * Helper function for polymorphic input data: 3. \"dict\" of Tensor.\n */\nexport function isDataDict(x: Tensor|Tensor[]|\n                           {[inputName: string]: Tensor}): boolean {\n  return !isDataTensor(x) && !isDataArray(x);\n}\n\n/**\n * Normalizes inputs and targets provided by users.\n * @param data User-provided input data (polymorphic).\n * @param names An Array of expected Tensor names.\n * @param shapes Optional Array of expected Tensor shapes.\n * @param checkBatchAxis Whether to check that the batch axis of the arrays\n *   match  the expected value found in `shapes`.\n * @param exceptionPrefix String prefix used for exception formatting.\n * @returns List of standardized input Tensors (one Tensor per model input).\n * @throws ValueError: in case of improperly formatted user data.\n */\nexport function standardizeInputData(\n    data: Tensor|Tensor[]|{[inputName: string]: Tensor}, names: string[],\n    shapes?: Shape[], checkBatchAxis = true, exceptionPrefix = ''): Tensor[] {\n  if (names == null || names.length === 0) {\n    // Check for the case where the model expected no data, but some data got\n    // sent.\n    if (data != null) {\n      let gotUnexpectedData = false;\n      if (isDataArray(data) && (data as Tensor[]).length > 0) {\n        gotUnexpectedData = true;\n      } else if (isDataDict(data)) {\n        for (const key in data) {\n          if (data.hasOwnProperty(key)) {\n            gotUnexpectedData = true;\n            break;\n          }\n        }\n      } else {\n        // `data` is a singleton Tensor in this case.\n        gotUnexpectedData = true;\n      }\n      if (gotUnexpectedData) {\n        throw new ValueError(\n            `Error when checking model ${exceptionPrefix} expected no data, ` +\n            `but got ${data}`);\n      }\n    }\n    return [];\n  }\n  if (data == null) {\n    return names.map(name => null);\n  }\n\n  let arrays: Tensor[];\n  if (isDataDict(data)) {\n    data = data as {[inputName: string]: Tensor};\n    arrays = [];\n    for (const name of names) {\n      if (data[name] == null) {\n        throw new ValueError(\n            `No data provided for \"${name}\". Need data for each key in: ` +\n            `${names}`);\n      }\n      arrays.push(data[name]);\n    }\n  } else if (isDataArray(data)) {\n    data = data as Tensor[];\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `model expected. Expected to see ${names.length} Tensor(s), but ` +\n          `instead got the following list of Tensor(s): ${data}`);\n    }\n    arrays = data;\n  } else {\n    data = data as Tensor;\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model ${exceptionPrefix} expects ${names.length} Tensor(s), ` +\n          `but only received one Tensor. Found: Tensor with shape ${\n              data.shape}`);\n    }\n    arrays = [data];\n  }\n\n  arrays = ensureTensorsRank2OrHigher(arrays);\n\n  // Check shape compatibility.\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s). but got array with ` +\n            `shape ${array.shape}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          // Skip the first (batch) axis.\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null && refDim >= 0 && dim !== refDim) {\n          throw new ValueError(\n              `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n              `to have shape [${shapes[i]}], but got array with shape ` +\n              `[${array.shape}].`);\n        }\n      }\n    }\n  }\n  return arrays;\n}\n\n/**\n * User input validation for Tensors.\n * @param inputs `Array` of `tf.Tensor`s for inputs.\n * @param targets `Array` of `tf.Tensor`s for targets.\n * @param weights Optional `Array` of `tf.Tensor`s for sample weights.\n * @throws ValueError: in case of incorrectly formatted data.\n */\nexport function checkArrayLengths(\n    inputs: Tensor[], targets: Tensor[], weights?: Tensor[]) {\n  const setX = unique(inputs.map(input => input.shape[0]));\n  setX.sort();\n  const setY = unique(targets.map(target => target.shape[0]));\n  setY.sort();\n  // TODO(cais): Check `weights` as well.\n  if (setX.length > 1) {\n    throw new ValueError(\n        `All input Tensors (x) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(inputs.map(input => input.shape))}`);\n  }\n  if (setY.length > 1) {\n    throw new ValueError(\n        `All target Tensors (y) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(targets.map(target => target.shape))}`);\n  }\n  if (setX.length > 0 && setY.length > 0 && !util.arraysEqual(setX, setY)) {\n    throw new ValueError(\n        `Input Tensors should have the same number of samples as target ` +\n        `Tensors. Found ${setX[0]} input sample(s) and ${setY[0]} target ` +\n        `sample(s).`);\n  }\n}\n\n/**\n * Validation on the compatibility of targes and loss functions.\n *\n * This helps prevent users from using loss functions incorrectly.\n *\n * @param targets `Array` of `tf.Tensor`s of targets.\n * @param lossFns `Array` of loss functions.\n * @param outputShapes `Array` of shapes of model outputs.\n */\nfunction checkLossAndTargetCompatibility(\n    targets: Tensor[], lossFns: LossOrMetricFn[], outputShapes: Shape[]) {\n  // TODO(cais): Dedicated test coverage?\n  const keyLosses = [\n    losses.meanSquaredError, losses.binaryCrossentropy,\n    losses.categoricalCrossentropy\n  ];\n  for (let i = 0; i < targets.length; ++i) {\n    const y = targets[i];\n    const loss = lossFns[i];\n    const shape = outputShapes[i];\n    if (loss == null) {\n      continue;\n    }\n    if (loss === losses.categoricalCrossentropy) {\n      if (y.shape[y.shape.length - 1] === 1) {\n        throw new ValueError(\n            `You are passing a target array of shape ${y.shape} while using ` +\n            `a loss 'categorical_crossentropy'. 'categorical_crossentropy'` +\n            `expects targets to be binary matrices (1s and 0s) of shape ` +\n            `[samples, classes].`);\n        // TODO(cais): Example code in error message.\n      }\n    }\n    if (keyLosses.indexOf(loss) !== -1) {\n      const slicedYShape = y.shape.slice(1);\n      const slicedShape = shape.slice(1);\n      for (let j = 0; j < slicedYShape.length; ++j) {\n        const targetDim = slicedYShape[j];\n        const outDim = slicedShape[j];\n        if (outDim != null && targetDim !== outDim) {\n          throw new ValueError(\n              `A target Tensor with shape ${y.shape} was passed for an ` +\n              `output of shape ${shape}, while using a loss function that ` +\n              `expects targets to have the same shape as the output.`);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Check inputs provided by the user.\n *\n * Porting Note: This corresponds to _standardize_input_data() in Python\n *   Keras. Because of the strong typing in TF.js, we do not need to convert\n *   the data. Specifically:\n *   1) in PyKeras, `data` can be `DataFrame` instances from pandas, for\n *      example. We don't need to worry about that here because there is no\n *      widely popular javascript/typesdcript equivalent of pandas (so far).\n *      If one becomes available in the future, we can add support.\n *   2) in PyKeras, inputs can be Python dict. But here we are stipulating\n * that the data is either a single `tf.Tensor` or an Array of `tf.Tensor`s. We\n * may add support for `Object` data inputs in the future when the need\n * arises.\n *\n * Instead, we perform basic checks for number of parameters and shapes.\n *\n * @param data: The input data.\n * @param names: Name for the inputs, from the model.\n * @param shapes: Expected shapes for the input data, from the model.\n * @param checkBatchAxis: Whether the size along the batch axis (i.e., the\n *   first dimension) will be checked for matching.\n * @param exceptionPrefix: Execption prefix message, used in generating error\n *   messages.\n * @throws ValueError: on incorrect number of inputs or mismatches in shapes.\n */\nfunction checkInputData(\n    data: Tensor|Tensor[], names: string[], shapes?: Shape[],\n    checkBatchAxis = true, exceptionPrefix = '') {\n  let arrays: Tensor[];\n  if (Array.isArray(data)) {\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `the model expected. Expected to see ${names.length} Tensor(s),` +\n          ` but instead got ${data.length} Tensors(s).`);\n    }\n    arrays = data;\n  } else {\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model expects ${names.length} ${exceptionPrefix} Tensors, ` +\n          `but only received one Tensor. Found: array with shape ` +\n          `${JSON.stringify(data.shape)}.`);\n    }\n    arrays = [data];\n  }\n\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s), but got array with ` +\n            `shape ${JSON.stringify(array.shape)}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null) {\n          if (refDim !== dim) {\n            throw new ValueError(\n                `Error when checking ${exceptionPrefix}: expected ` +\n                `${names[i]} to have shape ${JSON.stringify(shapes[i])} but ` +\n                `got array with shape ${JSON.stringify(array.shape)}.`);\n          }\n        }\n      }\n    }\n  }\n}\n\n/**\n * Maps metric functions to model outputs.\n * @param metrics An shortcut strings name, metric function, `Array` or dict\n *   (`Object`) of metric functions.\n * @param outputNames An `Array` of the names of model outputs.\n * @returns An `Array` (one entry per model output) of `Array` of metric\n *   functions. For instance, if the model has 2 outputs, and for the first\n *   output we want to compute `binaryAccuracy` and `binaryCrossentropy`,\n *   and just `binaryAccuracy` for the second output, the `Array` would look\n *   like:\n *     `[[binaryAccuracy, binaryCrossentropy],  [binaryAccuracy]]`\n * @throws TypeError: incompatible metrics format.\n */\nexport function collectMetrics(\n    metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n    {[outputName: string]: string | LossOrMetricFn},\n    outputNames: string[]): Array<Array<string|LossOrMetricFn>> {\n  if (metrics == null || Array.isArray(metrics) && metrics.length === 0) {\n    return outputNames.map(name => []);\n  }\n\n  let wrappedMetrics: Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  if (typeof metrics === 'string' || typeof metrics === 'function') {\n    wrappedMetrics = [metrics];\n  } else if (Array.isArray(metrics) || typeof metrics === 'object') {\n    wrappedMetrics = metrics as Array<string|LossOrMetricFn>|\n        {[outputName: string]: string} | {[outputName: string]: LossOrMetricFn};\n  } else {\n    throw new TypeError(\n        'Type of metrics argument not understood. Expected an string,' +\n        `function, Array, or Object, found: ${metrics}`);\n  }\n\n  if (Array.isArray(wrappedMetrics)) {\n    // We then apply all metrics to all outputs.\n    return outputNames.map(\n        name => wrappedMetrics as Array<string|LossOrMetricFn>);\n  } else {\n    // In this case, metrics is a dict.\n    const nestedMetrics: Array<Array<string|LossOrMetricFn>> = [];\n    for (const name of outputNames) {\n      let outputMetrics: string|LossOrMetricFn|Array<string|LossOrMetricFn> =\n          wrappedMetrics.hasOwnProperty(name) ? wrappedMetrics[name] : [];\n      if (!Array.isArray(outputMetrics)) {\n        outputMetrics = [outputMetrics];\n      }\n      nestedMetrics.push(outputMetrics);\n    }\n    return nestedMetrics;\n  }\n}\n\nexport interface ModelEvaluateArgs {\n  /**\n   * Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * Tensor of weights to weight the contribution of different samples to the\n   * loss and metrics.\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * integer: total number of steps (batches of samples)\n   * before declaring the evaluation round finished. Ignored with the default\n   * value of `undefined`.\n   */\n  steps?: number;\n}\n\n/**\n * Configuration for calls to `LayersModel.compile()`.\n */\nexport interface ModelCompileArgs {\n  /**\n   * An instance of `tf.train.Optimizer` or a string name for an Optimizer.\n   */\n  optimizer: string|Optimizer;\n\n  /**\n   * Object function(s) or name(s) of object function(s).\n   * If the model has multiple outputs, you can use a different loss\n   * on each output by passing a dictionary or an Array of losses.\n   * The loss value that will be minimized by the model will then be the sum\n   * of all individual losses.\n   */\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n\n  /**\n   * List of metrics to be evaluated by the model during training and testing.\n   * Typically you will use `metrics=['accuracy']`.\n   * To specify different metrics for different outputs of a multi-output\n   * model, you could also pass a dictionary.\n   */\n  metrics?: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n\n  // TODO(cais): Add lossWeights, sampleWeightMode, weightedMetrics, and\n  //   targetTensors.\n}\n\nconst LAYERS_MODEL_FORMAT_NAME = 'layers-model';\n\n/**\n * A `tf.LayersModel` is a directed, acyclic graph of `tf.Layer`s plus methods\n * for training, evaluation, prediction and saving.\n *\n * `tf.LayersModel` is the basic unit of training, inference and evaluation in\n * TensorFlow.js. To create a `tf.LayersModel`, use `tf.LayersModel`.\n *\n * See also:\n *   `tf.Sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class LayersModel extends Container implements tfc.InferenceModel {\n  // The class name is 'Model' rather than 'LayersModel' for backwards\n  // compatibility since this class name shows up in the serialization format.\n  /** @nocollapse */\n  static className = 'Model';\n  protected optimizer_: Optimizer;\n  // Whether the model instance owns the optimizer: `true` if and only if\n  // `optimizer` is created from a string parameter during `compile()` call.\n  protected isOptimizerOwned: boolean;\n\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n  lossFunctions: LossOrMetricFn[];\n\n  // TODO(cais): These private variables should probably not have the string\n  //   'feed' in their names, because we are not dealing with a symbolic\n  //   backend.\n  private feedOutputShapes: Shape[];\n  private feedLossFns: LossOrMetricFn[];\n  private collectedTrainableWeights: LayerVariable[];\n  private testFunction: (data: Tensor[]) => Scalar[];\n  history: History;\n\n  // A public property that can be set by Callbacks to order early stopping\n  // during `fit()` calls.\n  protected stopTraining_: boolean;\n  protected isTraining: boolean;\n\n  metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  metricsNames: string[];\n  // Porting Note: `metrics_tensors` in PyKeras is a symbolic tensor. But given\n  //   the imperative nature of tfjs-core, `metricsTensors` is a\n  //   TypeScript function here.\n  //   Also note that due to the imperative nature of tfjs-core, `metricsTensor`\n  //   here needs an output index to keep track of which output of the\n  //   LayersModel a metric belongs to. This is unlike `metrics_tensors` in\n  //   PyKeras, which is a `list` of symbolic tensors, each of which has\n  //   implicit \"knowledge\" of the outputs it depends on.\n  metricsTensors: Array<[LossOrMetricFn, number]>;\n\n  // User defind metadata (if any).\n  private userDefinedMetadata: {};\n\n  constructor(args: ContainerArgs) {\n    super(args);\n    this.isTraining = false;\n  }\n\n  /**\n   * Print a text summary of the model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - If the model has non-sequential-like topology, the inputs each layer\n   *   receives\n   * - The total number of trainable and non-trainable parameters of the model.\n   *\n   * ```js\n   * const input1 = tf.input({shape: [10]});\n   * const input2 = tf.input({shape: [20]});\n   * const dense1 = tf.layers.dense({units: 4}).apply(input1);\n   * const dense2 = tf.layers.dense({units: 8}).apply(input2);\n   * const concat = tf.layers.concatenate().apply([dense1, dense2]);\n   * const output =\n   *     tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat);\n   *\n   * const model = tf.model({inputs: [input1, input2], outputs: output});\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      throw new ValueError(\n          `This model has never been called, thus its weights have not been ` +\n          `created yet. So no summary can be displayed. Build the model ` +\n          `first (e.g., by calling it on some test data).`);\n    }\n    printSummary(this, lineLength, positions, printFn);\n  }\n\n  /**\n   * Configures and prepares the model for training and evaluation.  Compiling\n   * outfits the model with an optimizer, loss, and/or metrics.  Calling `fit`\n   * or `evaluate` on an un-compiled model will throw an error.\n   *\n   * @param args a `ModelCompileArgs` specifying the loss, optimizer, and\n   * metrics to be used for fitting and evaluating this model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  compile(args: ModelCompileArgs): void {\n    if (args.loss == null) {\n      args.loss = [];\n    }\n    this.loss = args.loss;\n\n    if (typeof args.optimizer === 'string') {\n      this.optimizer_ = optimizers.getOptimizer(args.optimizer);\n      this.isOptimizerOwned = true;\n    } else {\n      if (!(args.optimizer instanceof Optimizer)) {\n        throw new ValueError(\n            `User-defined optimizer must be an instance of tf.Optimizer.`);\n      }\n      this.optimizer_ = args.optimizer;\n      this.isOptimizerOwned = false;\n    }\n\n    // TODO(cais): Add lossWeights.\n    // TODO(cais): Add sampleWeightMode.\n\n    // Prepare loss functions.\n    let lossFunctions: LossOrMetricFn[] = [];\n    if (!Array.isArray(args.loss) && typeof args.loss !== 'string' &&\n        typeof args.loss !== 'function') {\n      args.loss = args.loss as {[outputName: string]: string};\n      for (const name in args.loss) {\n        if (this.outputNames.indexOf(name) === -1) {\n          throw new ValueError(\n              `Unknown entry in loss dictionary: \"${name}\". ` +\n              `Only expected the following keys: ${this.outputNames}`);\n        }\n      }\n      for (const name of this.outputNames) {\n        if (args.loss[name] == null) {\n          console.warn(\n              `Output \"${name}\" is missing from loss dictionary. We assume ` +\n              `this was done on purpose, and we will not be expecting data ` +\n              `to be passed to ${name} during training`);\n        }\n        lossFunctions.push(losses.get(args.loss[name]));\n      }\n    } else if (Array.isArray(args.loss)) {\n      if (args.loss.length !== this.outputs.length) {\n        throw new ValueError(\n            `When passing an Array as loss, it should have one entry per ` +\n            `model output. The model has ${this.outputs.length} output(s), ` +\n            `but you passed loss=${args.loss}.`);\n      }\n      const theLosses = args.loss as Array<string|LossOrMetricFn>;\n      lossFunctions = theLosses.map(l => losses.get(l));\n    } else {\n      const lossFunction = losses.get(args.loss);\n      this.outputs.forEach(_ => {\n        lossFunctions.push(lossFunction);\n      });\n    }\n\n    this.lossFunctions = lossFunctions;\n\n    this.feedOutputNames = [];\n    this.feedOutputShapes = [];\n    this.feedLossFns = [];\n    for (let i = 0; i < this.outputs.length; ++i) {\n      // TODO(cais): Logic for skipping target(s).\n      const shape = this.internalOutputShapes[i];\n      const name = this.outputNames[i];\n      this.feedOutputNames.push(name);\n      this.feedOutputShapes.push(shape);\n      this.feedLossFns.push(this.lossFunctions[i]);\n    }\n\n    // TODO(cais): Add logic for output masks.\n    // TODO(cais): Add logic for sample weights.\n    const skipTargetIndices: number[] = [];\n\n    // Prepare metrics.\n    this.metrics = args.metrics;\n    // TODO(cais): Add weightedMetrics.\n    this.metricsNames = ['loss'];\n    this.metricsTensors = [];\n\n    // Compute total loss.\n    // Porting Note: In PyKeras, metrics_tensors are symbolic tensor objects.\n    //   Here, metricsTensors are TypeScript functions. This difference is due\n    //   to the difference in symbolic/imperative property of the backends.\n    nameScope('loss', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        // TODO(cais): Add weightedLoss, sampleWeight and mask.\n        //   The following line should be weightedLoss\n        const weightedLoss = this.lossFunctions[i];\n        if (this.outputs.length > 1) {\n          this.metricsTensors.push([weightedLoss, i]);\n          this.metricsNames.push(this.outputNames[i] + '_loss');\n        }\n      }\n\n      // Porting Note: Due to the imperative nature of the backend, we calculate\n      //   the regularizer penalties in the totalLossFunction, instead of here.\n    });\n\n    const nestedMetrics = collectMetrics(args.metrics, this.outputNames);\n    // TODO(cais): Add nestedWeightedMetrics.\n\n    /**\n     * Helper function used in loop below.\n     */\n    const appendMetric =\n        (outputIndex: number, metricName: string,\n         metricTensor: LossOrMetricFn) => {\n          if (this.outputNames.length > 1) {\n            metricName = this.outputNames[outputIndex] + '_' + metricName;\n          }\n          this.metricsNames.push(metricName);\n          this.metricsTensors.push([metricTensor, outputIndex]);\n        };\n\n    nameScope('metric', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        const outputMetrics = nestedMetrics[i];\n        // TODO(cais): Add weights and outputWeightedMetrics.\n\n        // TODO(cais): Add optional arg `weights` to the following function.\n        const handleMetrics = (metrics: Array<string|LossOrMetricFn>) => {\n          const metricNamePrefix = '';\n          let metricName: string;\n          let accFn: LossOrMetricFn;\n          let weightedMetricFn: LossOrMetricFn;\n          //  TODO(cais): Use 'weights_' for weighted metrics.\n\n          for (const metric of metrics) {\n            if (typeof metric === 'string' &&\n                ['accuracy', 'acc', 'crossentropy', 'ce'].indexOf(metric) !==\n                    -1) {\n              const outputShape = this.internalOutputShapes[i];\n\n              if (outputShape[outputShape.length - 1] === 1 ||\n                  this.lossFunctions[i] === losses.binaryCrossentropy) {\n                // case: binary accuracy/crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryCrossentropy;\n                }\n              } else if (\n                  this.lossFunctions[i] ===\n                  losses.sparseCategoricalCrossentropy) {\n                // case: categorical accuracy / crossentropy with sparse\n                // targets.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalCrossentropy;\n                }\n              } else {\n                // case: categorical accuracy / crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalCrossentropy;\n                }\n              }\n              let suffix: string;\n              if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                suffix = 'acc';\n              } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                suffix = 'ce';\n              }\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = accFn;\n              metricName = metricNamePrefix + suffix;\n            } else {\n              const metricFn = Metrics.get(metric);\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = metricFn;\n              metricName =\n                  metricNamePrefix + Metrics.getLossOrMetricName(metric);\n            }\n\n            // TODO(cais): Add weighting and masking to metricResult.\n            let metricResult: LossOrMetricFn;\n            nameScope(metricName, () => {\n              metricResult = weightedMetricFn;\n            });\n            appendMetric(i, metricName, metricResult);\n          }\n        };\n\n        handleMetrics(outputMetrics);\n        // TODO(cais): Call handleMetrics with weights.\n      }\n    });\n\n    // Porting Notes: Given the imperative backend of tfjs-core,\n    //   there is no need for constructing the symbolic graph and placeholders.\n    this.collectedTrainableWeights = this.trainableWeights;\n  }\n\n  /**\n   * Check trainable weights count consistency.\n   *\n   * This will raise a warning if `this.trainableWeights` and\n   * `this.collectedTrainableWeights` are inconsistent (i.e., have different\n   * numbers of parameters).\n   * Inconsistency will typically arise when one modifies `model.trainable`\n   * without calling `model.compile()` again.\n   */\n  protected checkTrainableWeightsConsistency(): void {\n    if (this.collectedTrainableWeights == null) {\n      return;\n    }\n    if (this.trainableWeights.length !==\n        this.collectedTrainableWeights.length) {\n      console.warn(\n          'Discrepancy between trainableweights and collected trainable ' +\n          'weights. Did you set `model.trainable` without calling ' +\n          '`model.compile()` afterwards?');\n    }\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(\n   *     tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4});\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateArgs`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // TODO(cais): Standardize `config.sampleWeights` as well.\n    // Validate user data.\n    const checkBatchAxis = true;\n    const standardizedOuts =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    try {\n      // TODO(cais): If uses `useLearningPhase`, set the corresponding element\n      // of the input to 0.\n      const ins = standardizedOuts[0].concat(standardizedOuts[1]);\n      this.makeTestFunction();\n      const f = this.testFunction;\n      const testOuts =\n          this.testLoop(f, ins, batchSize, args.verbose, args.steps);\n      return singletonOrArray(testOuts);\n    } finally {\n      disposeNewTensors(standardizedOuts[0], x);\n      disposeNewTensors(standardizedOuts[1], y);\n    }\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args?: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    this.makeTestFunction();\n    return evaluateDataset(this, dataset, args);\n  }\n\n  /**\n   * Get number of samples provided for training, evaluation or prediction.\n   *\n   * @param ins Input `tf.Tensor`.\n   * @param batchSize Integer batch size, optional.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring loop finished. Optional.\n   * @param stepsName The public API's parameter name for `steps`.\n   * @returns Number of samples provided.\n   */\n  private checkNumSamples(\n      ins: Tensor|Tensor[], batchSize?: number, steps?: number,\n      stepsName = 'steps'): number {\n    let numSamples: number;\n    if (steps != null) {\n      numSamples = null;\n      if (batchSize != null) {\n        throw new ValueError(\n            `If ${stepsName} is set, batchSize must be null or undefined.` +\n            `Got batchSize = ${batchSize}`);\n      }\n    } else if (ins != null) {\n      if (Array.isArray(ins)) {\n        numSamples = ins[0].shape[0];\n      } else {\n        numSamples = ins.shape[0];\n      }\n    } else {\n      throw new ValueError(\n          `Either the input data should have a defined shape, or ` +\n          `${stepsName} shoud be specified.`);\n    }\n    return numSamples;\n  }\n\n  /**\n   * Execute internal tensors of the model with input data feed.\n   * @param inputs Input data feed. Must match the inputs of the model.\n   * @param outputs Names of the output tensors to be fetched. Must match\n   *   names of the SymbolicTensors that belong to the graph.\n   * @returns Fetched values for `outputs`.\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs: string|string[]):\n      Tensor|Tensor[] {\n    if (Array.isArray(outputs) && outputs.length === 0) {\n      throw new ValueError(\n          '`outputs` is an empty Array, which is not allowed.');\n    }\n\n    const outputsIsArray = Array.isArray(outputs);\n    const outputNames =\n        (outputsIsArray ? outputs as string[] : [outputs as string]);\n    const outputSymbolicTensors = this.retrieveSymbolicTensors(outputNames);\n\n    // Format the input into a FeedDict.\n    const feedDict = new FeedDict();\n    if (inputs instanceof Tensor) {\n      inputs = [inputs];\n    }\n    if (Array.isArray(inputs)) {\n      if (inputs.length !== this.inputs.length) {\n        throw new ValueError(\n            `The number of inputs provided (${inputs.length}) ` +\n            `does not match the number of inputs of this model ` +\n            `(${this.inputs.length}).`);\n      }\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n    } else {\n      for (const input of this.inputs) {\n        const tensorValue = inputs[input.name];\n        if (tensorValue == null) {\n          throw new ValueError(\n              `No value is provided for the model's input ${input.name}`);\n        }\n        feedDict.add(input, tensorValue);\n      }\n    }\n\n    // Run execution.\n    const executeOutputs = execute(outputSymbolicTensors, feedDict) as Tensor[];\n    return outputsIsArray ? executeOutputs : executeOutputs[0];\n  }\n\n  /**\n   * Retrieve the model's internal symbolic tensors from symbolic-tensor names.\n   */\n  private retrieveSymbolicTensors(symbolicTensorNames: string[]):\n      SymbolicTensor[] {\n    const outputSymbolicTensors: SymbolicTensor[] =\n        pyListRepeat(null, symbolicTensorNames.length);\n    let outputsRemaining = symbolicTensorNames.length;\n    for (const layer of this.layers) {\n      const layerOutputs: SymbolicTensor[] =\n          Array.isArray(layer.output) ? layer.output : [layer.output];\n      const layerOutputNames = layerOutputs.map(output => output.name);\n      for (let i = 0; i < symbolicTensorNames.length; ++i) {\n        const index = layerOutputNames.indexOf(symbolicTensorNames[i]);\n        if (index !== -1) {\n          outputSymbolicTensors[i] = layerOutputs[index];\n          outputsRemaining--;\n        }\n        if (outputsRemaining === 0) {\n          break;\n        }\n      }\n      if (outputsRemaining === 0) {\n        break;\n      }\n    }\n\n    if (outputsRemaining > 0) {\n      const remainingNames: string[] = [];\n      outputSymbolicTensors.forEach((tensor, i) => {\n        if (tensor == null) {\n          remainingNames.push(symbolicTensorNames[i]);\n        }\n      });\n      throw new ValueError(\n          `Cannot find SymbolicTensors for output name(s): ` +\n          `${JSON.stringify(remainingNames)}`);\n    }\n    return outputSymbolicTensors;\n  }\n\n  /**\n   * Helper method to loop over some data in batches.\n   *\n   * Porting Note: Not using the functional approach in the Python equivalent\n   *   due to the imperative backend.\n   * Porting Note: Does not support step mode currently.\n   *\n   * @param ins: input data\n   * @param batchSize: integer batch size.\n   * @param verbose: verbosity model\n   * @returns: Predictions as `tf.Tensor` (if a single output) or an `Array` of\n   *   `tf.Tensor` (if multipe outputs).\n   */\n  private predictLoop(ins: Tensor|Tensor[], batchSize = 32, verbose = false):\n      Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins);\n      if (verbose) {\n        throw new NotImplementedError(\n            'Verbose predictLoop() is not implemented yet.');\n      }\n\n      // Sample-based predictions.\n      // Porting Note: Tensor currently does not support sliced assignments as\n      //   in numpy, e.g., x[1:3] = y. Therefore we use concatenation while\n      //   iterating over the batches.\n\n      const batches = makeBatches(numSamples, batchSize);\n      const outsBatches: Tensor[][] = this.outputs.map(output => []);\n\n      // TODO(cais): Can the scope() be pushed down inside the for loop?\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchOuts = tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          // TODO(cais): Take care of the case of the last element is a flag for\n          //   training/test.\n          const insBatch = sliceArrays(ins, batchStart, batchEnd);\n\n          // Construct the feeds for execute();\n          const feeds = [];\n          if (Array.isArray(insBatch)) {\n            for (let i = 0; i < insBatch.length; ++i) {\n              feeds.push({key: this.inputs[i], value: insBatch[i]});\n            }\n          } else {\n            feeds.push({key: this.inputs[0], value: insBatch});\n          }\n          const feedDict = new FeedDict(feeds);\n          return execute(this.outputs, feedDict) as Tensor[];\n        });\n        batchOuts.forEach((batchOut, i) => outsBatches[i].push(batchOut));\n      }\n      return singletonOrArray(\n          outsBatches.map(batches => tfc.concat(batches, 0)));\n    });\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFlow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([8, 10]), {batchSize: 4}).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param args A `ModelPredictArgs` object containing optional fields.\n   *\n   * @return Prediction results as a `tf.Tensor`(s).\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    const xsRank2OrHigher = ensureTensorsRank2OrHigher(x);\n    checkInputData(\n        xsRank2OrHigher, this.inputNames, this.feedInputShapes, false);\n    try {\n      // TODO(cais): Take care of stateful models.\n      //   if (this.stateful) ...\n      // TODO(cais): Take care of the learning_phase boolean flag.\n      //   if (this.useLearningPhase) ...\n      const batchSize = args.batchSize == null ? 32 : args.batchSize;\n      checkBatchSize(batchSize);\n      return this.predictLoop(xsRank2OrHigher, batchSize);\n    } finally {\n      disposeNewTensors(xsRank2OrHigher, x);\n    }\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predictOnBatch(tf.ones([8, 10])).print();\n   * ```\n   * @param x: Input samples, as a Tensor (for models with exactly one\n   *   input) or an array of Tensors (for models with more than one input).\n   * @return Tensor(s) of predictions\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predictOnBatch(x: Tensor|Tensor[]): Tensor|Tensor[] {\n    checkInputData(x, this.inputNames, this.feedInputShapes, true);\n    // TODO(cais): Take care of the learning_phase boolean flag.\n    //   if (this.useLearningPhase) ...\n    const batchSize = (Array.isArray(x) ? x[0] : x).shape[0];\n    return this.predictLoop(x, batchSize);\n  }\n\n  protected standardizeUserDataXY(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor}, checkBatchAxis = true,\n      batchSize?: number): [Tensor[], Tensor[]] {\n    // TODO(cais): Add sampleWeight, classWeight\n    if (this.optimizer_ == null) {\n      throw new RuntimeError(\n          'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileArgs).');\n    }\n    const outputShapes: Shape[] = [];\n    for (let i = 0; i < this.feedOutputShapes.length; ++i) {\n      const outputShape = this.feedOutputShapes[i];\n      const lossFn = this.feedLossFns[i];\n      if (lossFn === losses.sparseCategoricalCrossentropy) {\n        outputShapes.push(\n            outputShape.slice(0, outputShape.length - 1).concat([1]));\n      } else {\n        // Porting Note: Because of strong typing `lossFn` must be a function.\n        outputShapes.push(outputShape);\n      }\n    }\n    x = standardizeInputData(\n        x, this.feedInputNames, this.feedInputShapes, false, 'input');\n    y = standardizeInputData(\n        y, this.feedOutputNames, outputShapes, false, 'target');\n    // TODO(cais): Standardize sampleWeights & classWeights.\n    checkArrayLengths(x, y, null);\n    // TODO(cais): Check sampleWeights as well.\n    checkLossAndTargetCompatibility(y, this.feedLossFns, this.feedOutputShapes);\n    if (this.stateful && batchSize != null && batchSize > 0) {\n      if (x[0].shape[0] % batchSize !== 0) {\n        throw new ValueError(\n            `In a stateful network, you should only pass inputs with a ` +\n            `number of samples that is divisible by the batch size ` +\n            `${batchSize}. Found: ${x[0].shape[0]} sample(s).`);\n      }\n    }\n    return [x, y];\n  }\n\n  protected async standardizeUserData(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      sampleWeight?: Tensor|Tensor[]|{[outputName: string]: Tensor},\n      classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap,\n      checkBatchAxis = true,\n      batchSize?: number): Promise<[Tensor[], Tensor[], Tensor[]]> {\n    const [standardXs, standardYs] =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    // TODO(cais): Handle sampleWeights.\n    if (sampleWeight != null) {\n      throw new Error('sample weight is not supported yet.');\n    }\n\n    let standardSampleWeights: Tensor[] = null;\n    if (classWeight != null) {\n      const classWeights =\n          standardizeClassWeights(classWeight, this.outputNames);\n      standardSampleWeights = [];\n      for (let i = 0; i < classWeights.length; ++i) {\n        standardSampleWeights.push(\n            await standardizeWeights(standardYs[i], null, classWeights[i]));\n      }\n    }\n\n    // TODO(cais): Deal with the case of model.stateful == true.\n    return [standardXs, standardYs, standardSampleWeights];\n  }\n\n  /**\n   * Loop over some test data in batches.\n   * @param f A Function returning a list of tensors.\n   * @param ins Array of tensors to be fed to `f`.\n   * @param batchSize Integer batch size or `null` / `undefined`.\n   * @param verbose verbosity mode.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring test finished. Ignored with the default value of `null` /\n   * `undefined`.\n   * @returns Array of Scalars.\n   */\n  private testLoop(\n      f: (data: Tensor[]) => Scalar[], ins: Tensor[], batchSize?: number,\n      verbose = 0, steps?: number): Scalar[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins, batchSize, steps, 'steps');\n      const outs: Scalar[] = [];\n      if (verbose > 0) {\n        throw new NotImplementedError('Verbose mode is not implemented yet.');\n      }\n      // TODO(cais): Use `indicesForConversionToDense' to prevent slow down.\n      if (steps != null) {\n        throw new NotImplementedError(\n            'steps mode in testLoop() is not implemented yet');\n      } else {\n        const batches = makeBatches(numSamples, batchSize);\n        const indexArray = tensor1d(range(0, numSamples));\n        for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds =\n              K.sliceAlongFirstAxis(\n                  indexArray, batchStart, batchEnd - batchStart) as Tensor1D;\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Scalar[];\n          const batchOuts = f(insBatch);\n          if (batchIndex === 0) {\n            for (let i = 0; i < batchOuts.length; ++i) {\n              outs.push(scalar(0));\n            }\n          }\n          for (let i = 0; i < batchOuts.length; ++i) {\n            const batchOut = batchOuts[i];\n            outs[i] =\n                tfc.add(outs[i], tfc.mul(batchEnd - batchStart, batchOut));\n          }\n        }\n        for (let i = 0; i < outs.length; ++i) {\n          outs[i] = tfc.div(outs[i], numSamples);\n        }\n      }\n      return outs;\n    });\n  }\n\n  protected getDedupedMetricsNames(): string[] {\n    const outLabels = this.metricsNames;\n    // Rename duplicated metrics names (can happen with an output layer\n    // shared among multiple dataflows).\n    const dedupedOutLabels = [];\n    for (let i = 0; i < outLabels.length; ++i) {\n      const label = outLabels[i];\n      let newLabel = label;\n      if (count(outLabels, label) > 1) {\n        const dupIndex = count(outLabels.slice(0, i), label);\n        newLabel += `_${dupIndex}`;\n      }\n      dedupedOutLabels.push(newLabel);\n    }\n    return dedupedOutLabels;\n  }\n\n  /**\n   * Creates a function that performs the following actions:\n   *\n   * 1. computes the losses\n   * 2. sums them to get the total loss\n   * 3. call the optimizer computes the gradients of the LayersModel's\n   *    trainable weights w.r.t. the total loss and update the variables\n   * 4. calculates the metrics\n   * 5. returns the values of the losses and metrics.\n   */\n  protected makeTrainFunction(): (data: Tensor[]) => Scalar[] {\n    return (data: Tensor[]) => {\n      const lossValues: Scalar[] = [];\n\n      const inputs = data.slice(0, this.inputs.length);\n      const targets = data.slice(\n          this.inputs.length, this.inputs.length + this.outputs.length);\n      const sampleWeights = data.slice(\n          this.inputs.length + this.outputs.length,\n          this.inputs.length + this.outputs.length * 2);\n\n      const metricsValues: Scalar[] = [];\n\n      // Create a function that computes the total loss based on the\n      // inputs. This function is used for obtaining gradients through\n      // backprop.\n      const totalLossFunction = () => {\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs =\n            execute(this.outputs, feedDict, {'training': true}) as Tensor[];\n        // TODO(cais): Take care of the case of multiple outputs from a\n        //   single layer?\n\n        let totalLoss: Tensor;\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          let loss = lossFunction(targets[i], outputs[i]);\n          if (sampleWeights[i] != null) {\n            loss = computeWeightedLoss(loss, sampleWeights[i]);\n          }\n\n          // TODO(cais): push Scalar instead.\n          const meanLoss: Scalar = tfc.mean(loss);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          lossValues.push(meanLoss);\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n        }\n\n        // Compute the metrics.\n        // TODO(cais): These should probably be calculated outside\n        //   totalLossFunction to benefit speed?\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          let weightedMetric: Scalar;\n\n          if (this.outputs.length > 1 && i < this.outputs.length) {\n            weightedMetric = lossValues[i];\n          } else {\n            const metric = this.metricsTensors[i][0];\n            const outputIndex = this.metricsTensors[i][1];\n            weightedMetric =\n                tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          }\n\n          tfc.keep(weightedMetric);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          metricsValues.push(weightedMetric);\n        }\n\n        totalLoss = tfc.mean(totalLoss);\n\n        // Add regularizer penalties.\n        this.calculateLosses().forEach(regularizerLoss => {\n          totalLoss = tfc.add(totalLoss, regularizerLoss);\n        });\n\n        return totalLoss as Scalar;\n      };\n\n      const variables = this.collectedTrainableWeights.map(\n          param => param.read() as tfc.Variable);\n      const returnCost = true;\n      const totalLossValue =\n          this.optimizer_.minimize(totalLossFunction, returnCost, variables);\n\n      return [totalLossValue].concat(metricsValues);\n    };\n  }\n\n  /**\n   * Create a function which, when invoked with an array of `tf.Tensor`s as a\n   * batch of inputs, returns the prespecified loss and metrics of the model\n   * under the batch of input data.\n   */\n  private makeTestFunction() {\n    this.testFunction = (data: Tensor[]) => {\n      return tfc.tidy(() => {\n        const valOutputs: Scalar[] = [];\n        let totalLoss: Scalar;\n        const inputs = data.slice(0, this.inputs.length);\n        const targets = data.slice(\n            this.inputs.length, this.inputs.length + this.outputs.length);\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs = execute(this.outputs, feedDict) as Tensor[];\n        // Compute total loss.\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          // TODO(cais): Add sample weighting and replace the simple\n          // averaging.\n          const loss: Scalar = tfc.mean(lossFunction(targets[i], outputs[i]));\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n          valOutputs.push(totalLoss);\n        }\n        // Compute the metrics.\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          const metric = this.metricsTensors[i][0];\n          const outputIndex = this.metricsTensors[i][1];\n          // TODO(cais): Replace K.mean() with a proper weighting function.\n          const meanMetric =\n              tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          valOutputs.push(meanMetric as Scalar);\n        }\n        return valOutputs;\n      });\n    };\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a\n   * dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *     layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * for (let i = 1; i < 5 ; ++i) {\n   *   const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *       batchSize: 4,\n   *       epochs: 3\n   *   });\n   *   console.log(\"Loss after Epoch \" + i + \" : \" + h.history.loss[0]);\n   * }\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you\n   * can also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named,\n   * you can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args A `ModelFitArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input\n   * data and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    return fitTensors(this, x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below when it's possible to instantiate\n  //   actual dataset objects.\n  /**\n   * Trains the model using a dataset object.\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for training. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs.\n   *   Of the two items in the array, the first is the input feature(s) and\n   *   the second is the output target(s).\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    return fitDataset(this, dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    // TODO(cais): Support sampleWeight and classWeight.\n    // TODO(cais): Support Dataset objects.\n    const standardizeOut = await this.standardizeUserData(x, y);\n    const inputs = standardizeOut[0];\n    const targets = standardizeOut[1];\n    const trainFunction = this.makeTrainFunction();\n    const losses = trainFunction(inputs.concat(targets));\n    const lossValues: number[] = [];\n    for (const loss of losses) {\n      const v = await loss.data();\n      lossValues.push(v[0]);\n    }\n    tfc.dispose(losses);\n    return singletonOrArray(lossValues);\n  }\n\n  /**\n   * Extract weight values of the model.\n   *\n   * @param config: An instance of `io.SaveConfig`, which specifies\n   * model-saving options such as whether only trainable weights are to be\n   * saved.\n   * @returns A `NamedTensorMap` mapping original weight names (i.e.,\n   *   non-uniqueified weight names) to their values.\n   */\n  protected getNamedWeights(config?: io.SaveConfig): NamedTensor[] {\n    const namedWeights: NamedTensor[] = [];\n\n    const trainableOnly = config != null && config.trainableOnly;\n    const weights = trainableOnly ? this.trainableWeights : this.weights;\n    const weightValues = this.getWeights(trainableOnly);\n    for (let i = 0; i < weights.length; ++i) {\n      if (trainableOnly && !weights[i].trainable) {\n        // Optionally skip non-trainable weights.\n        continue;\n      }\n      namedWeights.push(\n          {name: weights[i].originalName, tensor: weightValues[i]});\n    }\n    return namedWeights;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const input = tf.input({shape: [10]});\n   * const output = tf.layers.dense({units: 1}).apply(input);\n   * const model = tf.model({inputs: [input], outputs: [output]});\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10\n   * values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    this.stopTraining_ = stop;\n  }\n\n  get stopTraining(): boolean {\n    return this.stopTraining_;\n  }\n\n  get optimizer(): Optimizer {\n    return this.optimizer_;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    if (this.optimizer_ !== optimizer) {\n      this.optimizer_ = optimizer;\n      this.isOptimizerOwned = false;\n    }\n  }\n\n  dispose(): DisposeResult {\n    const result = super.dispose();\n    if (result.refCountAfterDispose === 0 && this.optimizer != null &&\n        this.isOptimizerOwned) {\n      const numTensorsBeforeOptmizerDisposal = tfc.memory().numTensors;\n      this.optimizer_.dispose();\n      result.numDisposedVariables +=\n          numTensorsBeforeOptmizerDisposal - tfc.memory().numTensors;\n    }\n    return result;\n  }\n\n  private getLossIdentifiers(): LossIdentifier|LossIdentifier[]|\n      {[outputName: string]: LossIdentifier} {\n    let lossNames: LossIdentifier|LossIdentifier[]|\n        {[outputName: string]: LossIdentifier};\n    if (typeof this.loss === 'string') {\n      lossNames = toSnakeCase(this.loss) as LossIdentifier;\n    } else if (Array.isArray(this.loss)) {\n      for (const loss of this.loss) {\n        if (typeof loss !== 'string') {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n      lossNames = (this.loss as string[]).map(name => toSnakeCase(name)) as\n          LossIdentifier[];\n    } else {\n      const outputNames = Object.keys(this.loss);\n      lossNames = {} as {[outputName: string]: LossIdentifier};\n      const losses =\n          this.loss as {[outputName: string]: LossOrMetricFn | string};\n      for (const outputName of outputNames) {\n        if (typeof losses[outputName] === 'string') {\n          lossNames[outputName] =\n              toSnakeCase(losses[outputName] as string) as LossIdentifier;\n        } else {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n    }\n    return lossNames;\n  }\n\n  private getMetricIdentifiers(): MetricsIdentifier[]|\n      {[key: string]: MetricsIdentifier} {\n    if (typeof this.metrics === 'string' ||\n        typeof this.metrics === 'function') {\n      return [toSnakeCase(Metrics.getLossOrMetricName(this.metrics))];\n    } else if (Array.isArray(this.metrics)) {\n      return this.metrics.map(\n          metric => toSnakeCase(Metrics.getLossOrMetricName(metric)));\n    } else {\n      const metricsIdentifiers: {[key: string]: MetricsIdentifier} = {};\n      for (const key in this.metrics) {\n        metricsIdentifiers[key] =\n            toSnakeCase(Metrics.getLossOrMetricName(this.metrics[key]));\n      }\n      return metricsIdentifiers;\n    }\n  }\n\n  protected getTrainingConfig(): TrainingConfig {\n    return {\n      loss: this.getLossIdentifiers(),\n      metrics: this.getMetricIdentifiers(),\n      optimizer_config: {\n        class_name: this.optimizer.getClassName(),\n        config: this.optimizer.getConfig()\n      } as OptimizerSerialization\n    };\n    // TODO(cais): Add weight_metrics when they are supported.\n    // TODO(cais): Add sample_weight_mode when it's supported.\n    // TODO(cais): Add loss_weights when it's supported.\n  }\n\n  loadTrainingConfig(trainingConfig: TrainingConfig) {\n    if (trainingConfig.weighted_metrics != null) {\n      throw new Error('Loading weight_metrics is not supported yet.');\n    }\n    if (trainingConfig.loss_weights != null) {\n      throw new Error('Loading loss_weights is not supported yet.');\n    }\n    if (trainingConfig.sample_weight_mode != null) {\n      throw new Error('Loading sample_weight_mode is not supported yet.');\n    }\n\n    const tsConfig = convertPythonicToTs(trainingConfig.optimizer_config) as\n        serialization.ConfigDict;\n    const optimizer = deserialize(tsConfig) as Optimizer;\n\n    let loss;\n    if (typeof trainingConfig.loss === 'string') {\n      loss = toCamelCase(trainingConfig.loss);\n    } else if (Array.isArray(trainingConfig.loss)) {\n      loss = trainingConfig.loss.map(lossEntry => toCamelCase(lossEntry));\n    } else if (trainingConfig.loss != null) {\n      loss = {} as {[outputName: string]: LossIdentifier};\n      for (const key in trainingConfig.loss) {\n        loss[key] = toCamelCase(trainingConfig.loss[key]) as LossIdentifier;\n      }\n    }\n\n    let metrics;\n    if (Array.isArray(trainingConfig.metrics)) {\n      metrics = trainingConfig.metrics.map(metric => toCamelCase(metric));\n    } else if (trainingConfig.metrics != null) {\n      metrics = {} as {[outputName: string]: MetricsIdentifier};\n      for (const key in trainingConfig.metrics) {\n        metrics[key] = toCamelCase(trainingConfig.metrics[key]);\n      }\n    }\n\n    this.compile({loss, metrics, optimizer});\n  }\n\n  /**\n   * Save the configuration and/or weights of the LayersModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 2. Saving `model`'s topology and weights to browser\n   * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('indexeddb://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 3. Saving `model`'s topology and weights as two files\n   * (`my-model-1.json` and `my-model-1.weights.bin`) downloaded from\n   * browser.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('downloads://my-model-1');\n   * ```\n   *\n   * Example 4. Send  `model`'s topology and weights to an HTTP server.\n   * See the documentation of `tf.io.http` for more details\n   * including specifying request parameters and implementation of the\n   * server.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('http://my-server/model/upload');\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new ValueError(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new ValueError(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new ValueError(\n          'LayersModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    const weightDataAndSpecs =\n        await io.encodeWeights(this.getNamedWeights(config));\n\n    const returnString = false;\n    const unusedArg: {} = null;\n    const modelConfig = this.toJSON(unusedArg, returnString);\n    const modelArtifacts: io.ModelArtifacts = {\n      modelTopology: modelConfig,\n      format: LAYERS_MODEL_FORMAT_NAME,\n      generatedBy: `TensorFlow.js tfjs-layers v${version}`,\n      convertedBy: null,\n    };\n\n    const includeOptimizer = config == null ? false : config.includeOptimizer;\n    if (includeOptimizer && this.optimizer != null) {\n      modelArtifacts.trainingConfig = this.getTrainingConfig();\n      const weightType = 'optimizer';\n      const {data: optimizerWeightData, specs: optimizerWeightSpecs} =\n          await io.encodeWeights(await this.optimizer.getWeights(), weightType);\n      weightDataAndSpecs.specs.push(...optimizerWeightSpecs);\n      weightDataAndSpecs.data = io.concatenateArrayBuffers(\n          [weightDataAndSpecs.data, optimizerWeightData]);\n    }\n\n    if (this.userDefinedMetadata != null) {\n      // Check serialized size of user-defined metadata.\n      const checkSize = true;\n      checkUserDefinedMetadata(this.userDefinedMetadata, this.name, checkSize);\n      modelArtifacts.userDefinedMetadata = this.userDefinedMetadata;\n    }\n\n    modelArtifacts.weightData = weightDataAndSpecs.data;\n    modelArtifacts.weightSpecs = weightDataAndSpecs.specs;\n    return handlerOrURL.save(modelArtifacts);\n  }\n\n  /**\n   * Set user-defined metadata.\n   *\n   * The set metadata will be serialized together with the topology\n   * and weights of the model during `save()` calls.\n   *\n   * @param setUserDefinedMetadata\n   */\n  setUserDefinedMetadata(userDefinedMetadata: {}): void {\n    checkUserDefinedMetadata(userDefinedMetadata, this.name);\n    this.userDefinedMetadata = userDefinedMetadata;\n  }\n\n  /**\n   * Get user-defined metadata.\n   *\n   * The metadata is supplied via one of the two routes:\n   *   1. By calling `setUserDefinedMetadata()`.\n   *   2. Loaded during model loading (if the model is constructed\n   *      via `tf.loadLayersModel()`.)\n   *\n   * If no user-defined metadata is available from either of the\n   * two routes, this function will return `undefined`.\n   */\n  getUserDefinedMetadata(): {} {\n    return this.userDefinedMetadata;\n  }\n}\nserialization.registerClass(LayersModel);\n\n/**\n * A `tf.Functional` is an alias to `tf.LayersModel`.\n *\n * See also:\n *   `tf.LayersModel`, `tf.Sequential`, `tf.loadLayersModel`.\n */\n/** @doc {heading: 'Models', subheading: 'Classes'} */\nexport class Functional extends LayersModel {\n  static className = 'Functional';\n}\nserialization.registerClass(Functional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source keras/models.py */\n\nimport {dispose, io, NamedTensorMap, Optimizer, Scalar, serialization, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {getUid} from './backend/state';\nimport {History} from './base_callbacks';\nimport {Dataset} from './engine/dataset_stub';\nimport {Input} from './engine/input_layer';\nimport {getSourceInputs, Layer, Node, SymbolicTensor} from './engine/topology';\nimport {LayersModel, ModelCompileArgs, ModelEvaluateArgs} from './engine/training';\nimport {ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './engine/training_dataset';\nimport {ModelFitArgs} from './engine/training_tensors';\nimport {NotImplementedError, RuntimeError, ValueError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {TrainingConfig} from './keras_format/training_config';\nimport {PyJsonDict} from './keras_format/types';\nimport {deserialize} from './layers/serialization';\nimport {Kwargs, NamedTensor} from './types';\nimport * as generic_utils from './utils/generic_utils';\nimport {convertPythonicToTs} from './utils/serialization_utils';\nimport {getExactlyOneShape} from './utils/types_utils';\n\n/**\n * Parses a JSON model configuration file and returns a model instance.\n *\n * ```js\n * // This example shows how to serialize a model using `toJSON()` and\n * // deserialize it as another model using `tf.models.modelFromJSON()`.\n * // Note: this example serializes and deserializes only the topology\n * // of the model; the weights of the loaded model will be different\n * // from those of the the original model, due to random weight\n * // initialization.\n * // To load the topology and weights of a model, use `tf.loadLayersModel()`.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.repeatVector({inputShape: [2], n: 4}));\n * // Serialize `model1` as a JSON object.\n * const model1JSON = model1.toJSON(null, false);\n * model1.summary();\n *\n * const model2 = await tf.models.modelFromJSON(model1JSON);\n * model2.summary();\n * ```\n *\n *  @param modelAndWeightsConfig JSON object or string encoding a model and\n *       weights configuration. It can also be only the topology JSON of the\n *       model, in which case the weights will not be loaded.\n *  @param custom_objects Optional dictionary mapping names\n *       (strings) to custom classes or functions to be\n *       considered during deserialization.\n * @returns A TensorFlow.js Layers `tf.LayersModel` instance (uncompiled).\n */\nexport async function modelFromJSON(\n    modelAndWeightsConfig: ModelAndWeightsConfig|PyJsonDict,\n    customObjects?: serialization.ConfigDict): Promise<LayersModel> {\n  if (!('modelTopology' in modelAndWeightsConfig)) {\n    modelAndWeightsConfig = {modelTopology: modelAndWeightsConfig};\n  }\n  modelAndWeightsConfig = modelAndWeightsConfig as ModelAndWeightsConfig;\n\n  let modelTopology = modelAndWeightsConfig.modelTopology;\n  if (modelTopology['model_config'] != null) {\n    // If the model-topology JSON contains a 'model_config' field, then it is\n    // a full model JSON (e.g., from `keras.Model.save()`), which contains\n    // not only the model's architecture in its 'model_config' field, but\n    // additional information such as the model's optimizer. We use only the\n    // 'model_config' field currently.\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n  const tsConfig =\n      convertPythonicToTs(modelTopology) as serialization.ConfigDict;\n  const model = deserialize(tsConfig, customObjects) as LayersModel;\n\n  if (modelAndWeightsConfig.weightsManifest != null) {\n    // Load the weight values keyed by the original tensor names in the model\n    // file that was loaded.  These should match the keys of the weight\n    // manifest.\n    const weightValues = await io.loadWeights(\n        modelAndWeightsConfig.weightsManifest, modelAndWeightsConfig.pathPrefix,\n        model.weights.map(weight => weight.originalName));\n\n    // Map the weights to the unique tensor names generated during model loading\n    const uniqueWeightValues: NamedTensorMap = {};\n    for (const weight of model.weights) {\n      uniqueWeightValues[weight.originalName] =\n          weightValues[weight.originalName];\n    }\n\n    model.loadWeights(uniqueWeightValues);\n    // Dispose temporary weight values.\n    dispose(weightValues);\n  }\n  return model;\n}\n\n/**\n * Options for loading a saved mode in TensorFlow.js format.\n */\nexport interface ModelAndWeightsConfig {\n  /**\n   * A JSON object or JSON string containing the model config.\n   *\n   * This can be either of the following two formats:\n   *   - A model archiecture-only config,  i.e., a format consistent with the\n   *     return value of`keras.Model.to_json()`.\n   *   - A full model config, containing not only model architecture, but also\n   *     training options and state, i.e., a format consistent with the return\n   *     value of `keras.models.save_model()`.\n   */\n  modelTopology: PyJsonDict;\n\n  /**\n   * A weights manifest in TensorFlow.js format.\n   */\n  weightsManifest?: io.WeightsManifestConfig;\n\n  /**\n   * Path to prepend to the paths in `weightManifest` before fetching.\n   *\n   * The path may optionally end in a slash ('/').\n   */\n  pathPrefix?: string;\n}\n\n// TODO(nielsene): Remove after: https://github.com/tensorflow/tfjs/issues/400\nexport interface ModelPredictArgs {\n  /**\n   * Optional. Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Optional. Verbosity mode. Defaults to false.\n   */\n  verbose?: boolean;\n}\n\n/**\n * Load a model, including its topology and optionally weights.  See the\n * Tutorial named \"How to import a Keras Model\" for usage examples.\n *\n * Example 1: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 2. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * Example 4. Load a model from an HTTP server.\n *\n * ```js\n * const model = await\n *     tf.loadLayersModel('https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. This path will be\n *      interpreted as a relative HTTP path, to which `fetch` will be used to\n *      request the model topology and weight manifest JSON.\n *      The content of the JSON file is assumed to be a JSON object with the\n *      following fields and values:\n *      - 'modelTopology': A JSON object that can be either of:\n *        1. a model architecture JSON consistent with the format of the return\n *            value of `keras.Model.to_json()`\n *        2. a full model JSON in the format of `keras.models.save_model()`.\n *      - 'weightsManifest': A TensorFlow.js weights manifest.\n *      See the Python converter function `save_model()` for more details.\n *      It is also assumed that model weights can be accessed from relative\n *      paths described by the `paths` fields in weights manifest.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A progress callback of the form:\n *     `(fraction: number) => void`. This callback can be used to monitor the\n *     model-loading process.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n */\nexport async function loadLayersModelInternal(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (typeof pathOrIOHandler === 'string') {\n    const handlers = io.getLoadHandlers(pathOrIOHandler, options);\n    if (handlers.length === 0) {\n      // For backward compatibility: if no load handler can be found,\n      // assume it is a relative http path.\n      // TODO(cais): Reformat the args into a single `LoadOptions` once the core\n      // is refactored.\n      handlers.push(io.browserHTTPRequest(pathOrIOHandler, options));\n    } else if (handlers.length > 1) {\n      throw new ValueError(\n          `Found more than one (${handlers.length}) load handlers for ` +\n          `URL '${pathOrIOHandler}'`);\n    }\n    pathOrIOHandler = handlers[0];\n  }\n  return loadLayersModelFromIOHandler(pathOrIOHandler, undefined, options);\n}\n\n/**\n * Load a model and optionally its weights, using an IOHandler object.\n *\n * @param handler The instance of `IOHandler` to be used during the model\n *   loading.\n * @param customObjects Any optional custom objects to be used during model\n *   loading.\n * @param strict Whether the weight loading will be done in strict mode.\n *   Default: `true`.\n */\nexport async function loadLayersModelFromIOHandler(\n    handler: io.IOHandler, customObjects?: serialization.ConfigDict,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (handler.load == null) {\n    throw new ValueError(\n        'Cannot proceed with model loading because the IOHandler provided ' +\n        'does not have the `load` method implemented.');\n  }\n  const artifacts = await handler.load();\n  let modelTopology = artifacts.modelTopology as PyJsonDict;\n  if (modelTopology['model_config'] != null) {\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n\n  const strict = options.strict == null ? true : options.strict;\n  // If weights are provided and the weight-loading mode is strict, use\n  // fast weight initialization. This skips costly initializers such as\n  // 'orthogonal' and saves unnecessary computation in cases where\n  // the initialized weight values will immediately be overwritten by\n  // loaded weight values.\n  const fastWeightInit =\n      artifacts.weightData != null && artifacts.weightSpecs != null && strict;\n  const model =\n      deserialize(\n          convertPythonicToTs(modelTopology) as serialization.ConfigDict,\n          customObjects, fastWeightInit) as LayersModel;\n\n  const trainingConfig = artifacts.trainingConfig as TrainingConfig;\n  if (trainingConfig != null) {\n    model.loadTrainingConfig(trainingConfig);\n  }\n  if (artifacts.userDefinedMetadata != null) {\n    model.setUserDefinedMetadata(artifacts.userDefinedMetadata);\n  }\n\n  // If weightData is present, load the weights into the model.\n  if (artifacts.weightData != null) {\n    // Loading weights requires weightSpecs.\n    if (artifacts.weightSpecs == null) {\n      throw new ValueError(\n          'LayersModel artifacts contains weight data, but not weight specs. ' +\n          'Therefore loading of weights cannot proceed.');\n    }\n\n    const {modelWeights, optimizerWeights} = decodeModelAndOptimizerWeights(\n        artifacts.weightData, artifacts.weightSpecs);\n    model.loadWeights(modelWeights, strict);\n\n    if (model.optimizer != null && optimizerWeights.length > 0) {\n      await model.optimizer.setWeights(optimizerWeights);\n    }\n\n    // Dispose temporary weight values.\n    dispose(modelWeights);\n    dispose(optimizerWeights.map(w => w.tensor));\n  }\n  return model;\n}\n\nfunction decodeModelAndOptimizerWeights(\n    buffer: ArrayBuffer, specs: io.WeightsManifestEntry[]):\n    {modelWeights: NamedTensorMap, optimizerWeights: NamedTensor[]} {\n  const name2Tensor = io.decodeWeights(buffer, specs);\n  const modelWeights: NamedTensorMap = {};\n  const optimizerWeights: NamedTensor[] = [];\n  specs.forEach(spec => {\n    if (spec.group === 'optimizer') {\n      optimizerWeights.push({name: spec.name, tensor: name2Tensor[spec.name]});\n    } else {\n      modelWeights[spec.name] = name2Tensor[spec.name];\n    }\n  });\n  return {modelWeights, optimizerWeights};\n}\n\n/**\n * Configuration for a Sequential model.\n */\nexport interface SequentialArgs {\n  /** Stack of layers for the model. */\n  layers?: Layer[];\n\n  /** The name of this model. */\n  name?: string;\n}\n\n/**\n * A model with a stack of layers, feeding linearly from one to the next.\n *\n * `tf.sequential` is a factory function that creates an instance of\n * `tf.Sequential`.\n *\n * ```js\n *  // Define a model for linear regression.\n *  const model = tf.sequential();\n *  model.add(tf.layers.dense({units: 1, inputShape: [1]}));\n *\n *  // Prepare the model for training: Specify the loss and the optimizer.\n *  model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n *\n *  // Generate some synthetic data for training.\n *  const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);\n *  const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);\n *\n *  // Train the model using the data then do inference on a data point the\n *  // model hasn't seen:\n *  await model.fit(xs, ys);\n *  model.predict(tf.tensor2d([5], [1, 1])).print();\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class Sequential extends LayersModel {\n  /** @nocollapse */\n  static className = 'Sequential';\n  private model: LayersModel;\n  constructor(args?: SequentialArgs) {\n    super({inputs: [], outputs: []});\n    args = args || {};\n\n    this.trainable = true;\n    this.built = false;\n\n    // Set model name.\n    this.name = (args.name != null) ? args.name : getUid('sequential_');\n\n    // Add to the model any layers passed to the constructor.\n    if (args.layers != null) {\n      for (const layer of args.layers) {\n        this.add(layer);\n      }\n    }\n  }\n\n  // Helper function to Sequential.add  Throws if the new output shape will be\n  // invalid.\n  private checkShape(layer: Layer) {\n    const shape = layer.inboundNodes[0].outputTensors[0].shape;\n    if (shape.some(x => x < 0)) {\n      throw new ValueError(\n          'Negative dimension size caused by adding layer ' +\n          `${layer.name} with input shape [` +\n          `${layer.inboundNodes[0].inputTensors[0].shape}]`);\n    }\n  }\n\n  /**\n   * Adds a layer instance on top of the layer stack.\n   *\n   * ```js\n   *  const model = tf.sequential();\n   *  model.add(tf.layers.dense({units: 8, inputShape: [1]}));\n   *  model.add(tf.layers.dense({units: 4, activation: 'relu6'}));\n   *  model.add(tf.layers.dense({units: 1, activation: 'relu6'}));\n   *  // Note that the untrained model is random at this point.\n   *  model.predict(tf.randomNormal([10, 1])).print();\n   * ```\n   * @param layer Layer instance.\n   *\n   * @exception ValueError In case the `layer` argument does not know its\n   * input shape.\n   * @exception ValueError In case the `layer` argument has multiple output\n   *   tensors, or is already connected somewhere else (forbidden in\n   *   `Sequential` models).\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  add(layer: Layer): void {\n    const isLayerModelInstance =\n        layer instanceof Sequential || layer instanceof LayersModel;\n    let modelLayer: LayersModel;\n    if (isLayerModelInstance) {\n      modelLayer = layer as LayersModel;\n      if (modelLayer.outputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      if (modelLayer.inputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single input tensor. ' +\n            'For multi-input layers, ' +\n            'use the functional API.');\n      }\n    }\n\n    if (this.outputs.length === 0) {\n      // first layer in model: check that it is an input layer\n      if (layer.inboundNodes.length === 0) {\n        // create an input layer\n        if (layer.batchInputShape == null) {\n          throw new ValueError(\n              'The first layer in a Sequential model must ' +\n              'get an `inputShape` or `batchInputShape` argument.');\n        }\n        // Instantiate the input layer.\n        const x = Input({\n          batchShape: layer.batchInputShape,\n          dtype: layer.dtype,\n          name: layer.name + '_input'\n        });\n        // This will build the current layer and create the node connecting\n        // the current layer to the input layer we just created.\n        layer.apply(x);\n      }\n\n      if (isLayerModelInstance) {\n        this.outputs = modelLayer.outputs;\n        this.inputs = modelLayer.inputs;\n      } else {\n        if (layer.inboundNodes.length !== 1) {\n          throw new ValueError(\n              'A layer added to a Sequential model must not already be ' +\n              `connected somewhere else. LayersModel received layer ${\n                  layer.name} ` +\n              `which has ${layer.inboundNodes.length} pre-existing inbound ` +\n              'connections.');\n        }\n\n        if (layer.inboundNodes[0].outputTensors.length !== 1) {\n          throw new ValueError(\n              'All layers in a Sequential model ' +\n              'should have a single output tensor. ' +\n              'For multi-output layers, ' +\n              'use the functional API.');\n        }\n        this.checkShape(layer);\n        this.outputs = [layer.inboundNodes[0].outputTensors[0]];\n        this.inputs = getSourceInputs(this.outputs[0]);\n      }\n\n      this.inboundNodes = [];\n      // We create an input node, which we will keep updated\n      // as we add more layers.\n      // (This call has side effects.)\n      // tslint:disable-next-line:no-unused-expression\n      new Node({\n        outboundLayer: this,\n        inboundLayers: [],\n        nodeIndices: [],\n        tensorIndices: [],\n        inputTensors: this.inputs,\n        outputTensors: this.outputs,\n        // no model-level masking for now\n        inputMasks: generic_utils.pyListRepeat(null, this.inputs.length),\n        outputMasks: [null],\n        inputShapes: this.inputs.map(x => x.shape),\n        outputShapes: this.outputs[0].shape\n      });\n    } else {\n      const outputTensor = layer.apply(this.outputs[0]);\n      if (Array.isArray(outputTensor)) {\n        throw new TypeError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      this.checkShape(layer);\n      this.outputs = [outputTensor as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n\n    this.layers.push(layer);\n    this.built = false;\n  }\n\n  /**\n   * Removes the last layer in the model.\n   *\n   * @exception TypeError if there are no layers in the model.\n   */\n  pop(): void {\n    if (this.layers.length === 0) {\n      throw new TypeError('There are no layers in the model.');\n    }\n\n    this.layers.pop();\n    if (this.layers.length === 0) {\n      this.outputs = [];\n      this.inboundNodes = [];\n      this.outboundNodes = [];\n    } else {\n      const lastLayerIndex = this.layers.length - 1;\n      this.layers[lastLayerIndex].outboundNodes = [];\n      this.outputs = [this.layers[lastLayerIndex].output as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.call(inputs, kwargs);\n  }\n\n  build(inputShape?: Shape|Shape[]) {\n    // Call `getExactlyOneShape` without using its return value,\n    // to verify that exactly one input shape is provided.\n    getExactlyOneShape(inputShape);\n\n    if (this.inputs.length === 0 || this.outputs.length === 0) {\n      throw new TypeError(\n          'Sequential model cannot be built: model is empty.' +\n          ' Add some layers first.');\n    }\n    // actually create the model\n    this.model = new LayersModel({\n      inputs: this.inputs,\n      outputs: this.outputs[0],\n      name: this.name + '_model'\n    });\n    this.model.trainable = this.trainable;\n\n    // mirror model attributes\n    this.supportsMasking = this.model.supportsMasking;\n    // TODO(michaelterry): Add caches\n    this.inputLayers = this.model.inputLayers;\n    this.inputLayersNodeIndices = this.model.inputLayersNodeIndices;\n    this.inputLayersTensorIndices = this.model.inputLayersTensorIndices;\n    this.outputLayers = this.model.outputLayers;\n    this.outputLayersNodeIndices = this.model.outputLayersNodeIndices;\n    this.outputLayersTensorIndices = this.model.outputLayersTensorIndices;\n    this.nodesByDepth = this.model.nodesByDepth;\n    this.containerNodes = this.model.containerNodes;\n    this.outputNames = this.model.outputNames;\n    this.inputNames = this.model.inputNames;\n    // TODO(michaelterry): Add feedInputNames, feedInputs, if needed.\n    // TODO(michaelterry): Add callbackModel if needed.\n    this.built = true;\n  }\n\n  countParams(): number {\n    if (!this.built) {\n      this.build();\n    }\n    return super.countParams();\n  }\n\n  /**\n   * Print a text summary of the Sequential model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - The total number of trainable and non-trainable parameters of the\n   * model.\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(\n   *     tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'}));\n   * model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n   *\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      this.build();\n    }\n    super.summary(lineLength, positions, printFn);\n  }\n\n  /**\n   * Sets the weights of the model.\n   *\n   * @param weights Should be a list of Tensors with shapes and types matching\n   *   the output of `model.getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    if (this.model == null) {\n      this.build();\n    }\n    this.model.setWeights(weights);\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   * });\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateConfig`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluate(x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluateDataset(dataset, args);\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([2, 10])).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param conifg A `ModelPredictConfig` object containing optional fields.\n   *\n   * @return `tf.Tensor`(s) of predictions.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predict(x, args);\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * @param x: Input samples, as a Tensor, or list of Tensors (if the model\n   *   has multiple inputs).\n   * @return Tensor(s) of predictions\n   */\n  predictOnBatch(x: Tensor): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predictOnBatch(x);\n  }\n\n  /**\n   * See `LayersModel.compile`.\n   *\n   * @param args\n   */\n  compile(args: ModelCompileArgs): void {\n    this.build();\n    this.model.compile(args);\n    this.optimizer_ = this.model.optimizer;\n    // tslint:disable-next-line:no-any\n    this.isOptimizerOwned = (this.model as any).isOptimizerOwned;\n    this.loss = this.model.loss;\n    this.metrics = this.model.metrics;\n    // TODO(cais): Add this.lossWeights, this.sampleWeightMode,\n    //   this.weightedMetrics, this.targets.\n    this.metricsTensors = this.model.metricsTensors;\n    this.metricsNames = this.model.metricsNames;\n    // TODO(cais): Add sampleWeights.\n  }\n\n  get optimizer(): Optimizer {\n    return this.model == null ? undefined : this.model.optimizer;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    this.model.optimizer = optimizer;\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   *   epochs: 3\n   * });\n   * console.log(history.history.loss[0]);\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you can\n   * also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named, you\n   *  can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args  A `ModelFitConfig`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fit(x, y, args);\n  }\n\n  /**\n   * Trains the model using a dataset object.\n   *\n   * ```js\n   * const xArray = [\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   * ];\n   * const yArray = [1, 1, 1, 1];\n   * // Create a dataset from the JavaScript array.\n   * const xDataset = tf.data.array(xArray);\n   * const yDataset = tf.data.array(yArray);\n   * // Zip combines the `x` and `y` Datasets into a single Dataset, the\n   * // iterator of which will return an object containing of two tensors,\n   * // corresponding to `x` and `y`.  The call to `batch(4)` will bundle\n   * // four such samples into a single object, with the same keys now pointing\n   * // to tensors that hold 4 examples, organized along the batch dimension.\n   * // The call to `shuffle(4)` causes each iteration through the dataset to\n   * // happen in a different order.  The size of the shuffle window is 4.\n   * const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset})\n   *     .batch(4)\n   *     .shuffle(4);\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [9]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fitDataset(xyDataset, {\n   *   epochs: 4,\n   *   callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)}\n   * });\n   * ```\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected to\n   *   generate a dataset iterator object, the `next()` method of which is\n   *   expected to produce data batches for evaluation. The return value of the\n   *   `next()` call ought to contain a boolean `done` field and a `value`\n   *   field.\n   *\n   *   The `value` field is expected to be an object of with fields\n   *   `xs` and `ys`, which point to the feature tensor and the target tensor,\n   *   respectively. This case is for models with exactly one input and one\n   *   output (e.g.. a sequential model). For example:\n   *   ```js\n   *   {value: {xs: xsTensor, ys: ysTensor}, done: false}\n   *   ```\n   *\n   *   If the model has multiple inputs, the `xs` field of `value` should\n   *   be an object mapping input names to their respective feature tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: {\n   *         input_1: xsTensor1,\n   *         input_2: xsTensor2\n   *       },\n   *       ys: ysTensor\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   *   If the model has multiple outputs, the `ys` field of `value` should\n   *   be an object mapping output names to their respective target tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: xsTensor,\n   *       ys: {\n   *         output_1: ysTensor1,\n   *         output_2: ysTensor2\n   *       },\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fitDataset(dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    return this.model.trainOnBatch(x, y);\n  }\n\n  /* See parent class for JsDoc */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    let configArray: serialization.ConfigDictArray;\n    let extraModelConfig: serialization.ConfigDict = {};\n    if (config instanceof Array) {\n      if (!(config[0].className != null) ||\n          config[0]['className'] === 'Merge') {\n        throw new ValueError('Legacy serialization format not supported yet.');\n      }\n      configArray = config;\n    } else {\n      util.assert(\n          config['layers'] != null,\n          () =>\n              `When the config data for a Sequential model is not an Array, ` +\n              `it must be an Object that contains the 'layers' field.`);\n      configArray = config['layers'] as serialization.ConfigDictArray;\n      delete config['layers'];\n      extraModelConfig = config;\n    }\n\n    const model = new cls(extraModelConfig);\n    if (!(model instanceof Sequential)) {\n      throw new NotImplementedError(\n          `Sequential.fromConfig called on non-Sequential input: ${model}`);\n    }\n    for (const conf of configArray) {\n      const customObjects: serialization.ConfigDict = undefined;\n      const layer = deserialize(\n                        conf as serialization.ConfigDict, customObjects,\n                        fastWeightInit) as Layer;\n      if (fastWeightInit) {\n        layer.setFastWeightInitDuringBuild(true);\n      }\n      model.add(layer);\n    }\n    return model;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(tf.layers.dense({units: 1, inputShape: [10]}));\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10 values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    // TODO(cais): When refactoring to remove the composition pattern happens,\n    // remove this method overriding.\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot set the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    this.model.stopTraining = stop;\n  }\n\n  get stopTraining(): boolean {\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot get the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    return this.model.stopTraining;\n  }\n\n  // TODO(cais): Override get trainableWeights() here\n\n  // tslint:disable-next-line:no-any\n  getConfig(): any {\n    // NOTE(cais): We override the return type of getConfig() to `any` here,\n    //   because the `Sequential` class is a special case among `Container`\n    //   subtypes in that its getConfig() method returns an Array (not a\n    //   dict).\n    const layers: serialization.ConfigDict[] = [];\n    for (const layer of this.layers) {\n      const dict: serialization.ConfigDict = {};\n      dict['className'] = layer.getClassName();\n      dict['config'] = layer.getConfig();\n      layers.push(dict);\n    }\n    return {name: this.name, layers};\n  }\n}\nserialization.registerClass(Sequential);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Exported functions.\n */\n\nimport {io} from '@tensorflow/tfjs-core';\n\nimport {BaseCallbackConstructor, CallbackConstructorRegistry} from './base_callbacks';\nimport {ContainerArgs} from './engine/container';\nimport {Input, InputConfig,} from './engine/input_layer';\nimport {SymbolicTensor} from './engine/topology';\nimport {LayersModel} from './engine/training';\nimport {loadLayersModelInternal, Sequential, SequentialArgs} from './models';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// LayersModel and related factory methods.\n\n/**\n * A model is a data structure that consists of `Layers` and defines inputs\n * and outputs.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.model` is more generic, supporting an arbitrary graph (without\n * cycles) of layers. `tf.sequential` is less generic and supports only a linear\n * stack of layers.\n *\n * When creating a `tf.LayersModel`, specify its input(s) and output(s). Layers\n * are used to wire input(s) to output(s).\n *\n * For example, the following code snippet defines a model consisting of\n * two `dense` layers, with 10 and 4 units, respectively.\n *\n * ```js\n * // Define input, which has a size of 5 (not including batch dimension).\n * const input = tf.input({shape: [5]});\n *\n * // First dense layer uses relu activation.\n * const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});\n * // Second dense layer uses softmax activation.\n * const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});\n *\n * // Obtain the output symbolic tensor by applying the layers on the input.\n * const output = denseLayer2.apply(denseLayer1.apply(input));\n *\n * // Create the model based on the inputs.\n * const model = tf.model({inputs: input, outputs: output});\n *\n * // The model can be used for training, evaluation and prediction.\n * // For example, the following line runs prediction with the model on\n * // some fake data.\n * model.predict(tf.ones([2, 5])).print();\n * ```\n * See also:\n *   `tf.sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function model(args: ContainerArgs): LayersModel {\n  return new LayersModel(args);\n}\n\n/**\n * Creates a `tf.Sequential` model.  A sequential model is any model where the\n * outputs of one layer are the inputs to the next layer, i.e. the model\n * topology is a simple 'stack' of layers, with no branching or skipping.\n *\n * This means that the first layer passed to a `tf.Sequential` model should have\n * a defined input shape. What that means is that it should have received an\n * `inputShape` or `batchInputShape` argument, or for some type of layers\n * (recurrent, Dense...) an `inputDim` argument.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.sequential` is less generic, supporting only a linear stack of layers.\n * `tf.model` is more generic and supports an arbitrary graph (without\n * cycles) of layers.\n *\n * Examples:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have an input shape defined.\n * model.add(tf.layers.dense({units: 32, inputShape: [50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output, which equals\n * // `[null, 4]`. The 1st dimension is the undetermined batch dimension; the\n * // 2nd is the output size of the model's last layer.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * It is also possible to specify a batch size (with potentially undetermined\n * batch dimension, denoted by \"null\") for the first layer using the\n * `batchInputShape` key. The following example is equivalent to the above:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have a defined input shape\n * model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * You can also use an `Array` of already-constructed `Layer`s to create\n * a `tf.Sequential` model:\n *\n * ```js\n * const model = tf.sequential({\n *   layers: [tf.layers.dense({units: 32, inputShape: [50]}),\n *            tf.layers.dense({units: 4})]\n * });\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function sequential(config?: SequentialArgs): Sequential {\n  return new Sequential(config);\n}\n\n/**\n * Load a model composed of Layer objects, including its topology and optionally\n * weights. See the Tutorial named \"How to import a Keras Model\" for usage\n * examples.\n *\n * This method is applicable to:\n *\n * 1. Models created with the `tf.layers.*`, `tf.sequential`, and\n * `tf.model` APIs of TensorFlow.js and later saved with the\n * `tf.LayersModel.save` method.\n * 2. Models converted from Keras or TensorFlow tf.keras using the\n * [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).\n *\n * This mode is *not* applicable to TensorFlow `SavedModel`s or their converted\n * forms. For those models, use `tf.loadGraphModel`.\n *\n * Example 1. Load a model from an HTTP server.\n *\n * ```js\n * const model = await tf.loadLayersModel(\n *     'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * Example 2: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 4. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. For file://\n *      (tfjs-node-only), http:// and https:// schemas, the path can be\n *      either absolute or relative.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A function of the signature `(fraction: number) => void',\n *     that can be used as the progress callback for the model loading.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport function loadLayersModel(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  return loadLayersModelInternal(pathOrIOHandler, options);\n}\n\n/**\n * Used to instantiate an input to a model as a `tf.SymbolicTensor`.\n *\n * Users should call the `input` factory function for\n * consistency with other generator functions.\n *\n * Example:\n *\n * ```js\n * // Defines a simple logistic regression model with 32 dimensional input\n * // and 3 dimensional output.\n * const x = tf.input({shape: [32]});\n * const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x);\n * const model = tf.model({inputs: x, outputs: y});\n * model.predict(tf.ones([2, 32])).print();\n * ```\n *\n * Note: `input` is only necessary when using `model`. When using\n * `sequential`, specify `inputShape` for the first layer or use `inputLayer`\n * as the first layer.\n *\n * @doc {heading: 'Models', subheading: 'Inputs'}\n */\nexport function input(config: InputConfig): SymbolicTensor {\n  return Input(config);\n}\n\nexport function registerCallbackConstructor(\n    verbosityLevel: number,\n    callbackConstructor: BaseCallbackConstructor): void {\n  CallbackConstructorRegistry.registerCallbackConstructor(\n      verbosityLevel, callbackConstructor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Layer activation functions\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {ActivationIdentifier} from './keras_format/activation_config';\nimport {deserializeKerasObject} from './utils/generic_utils';\n\n/**\n * Base class for Activations.\n *\n * Special note: due to cross-language compatibility reasons, the\n * static readonly className field in this family of classes must be set to\n * the initialLowerCamelCase name of the activation.\n */\nexport abstract class Activation extends serialization.Serializable {\n  abstract apply(tensor: Tensor, axis?: number): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\n/**\n * Exponential linear unit (ELU).\n * Reference: https://arxiv.org/abs/1511.07289\n */\nexport class Elu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'elu';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x: Input.\n   * @param alpha: Scaling factor the negative section.\n   * @return Output of the ELU activation.\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return K.elu(x, alpha);\n  }\n}\nserialization.registerClass(Elu);\n\n/**\n * Scaled Exponential Linear Unit. (Klambauer et al., 2017).\n * Reference: Self-Normalizing Neural Networks, https://arxiv.org/abs/1706.02515\n * Notes:\n *   - To be used together with the initialization \"lecunNormal\".\n *   - To be used together with the dropout variant \"AlphaDropout\".\n */\nexport class Selu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'selu';\n  apply(x: Tensor): Tensor {\n    return tfc.selu(x);\n  }\n}\nserialization.registerClass(Selu);\n\n/**\n *  Rectified linear unit\n */\nexport class Relu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu';\n  apply(x: Tensor): Tensor {\n    return tfc.relu(x);\n  }\n}\nserialization.registerClass(Relu);\n\n/**\n * Rectified linear unit activation maxing out at 6.0.\n */\nexport class Relu6 extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu6';\n  apply(x: Tensor): Tensor {\n    return tidy(() => tfc.minimum(6.0, tfc.relu(x)));\n  }\n}\nserialization.registerClass(Relu6);\n\n//* Linear activation (no-op) */\nexport class Linear extends Activation {\n  /** @nocollapse */\n  static readonly className = 'linear';\n  apply(x: Tensor): Tensor {\n    return x;\n  }\n}\nserialization.registerClass(Linear);\n\n/**\n * Sigmoid activation function.\n */\nexport class Sigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'sigmoid';\n  apply(x: Tensor): Tensor {\n    return tfc.sigmoid(x);\n  }\n}\nserialization.registerClass(Sigmoid);\n\n/**\n * Segment-wise linear approximation of sigmoid.\n */\nexport class HardSigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'hardSigmoid';\n  apply(x: Tensor): Tensor {\n    return K.hardSigmoid(x);\n  }\n}\nserialization.registerClass(HardSigmoid);\n\n/**\n * Softplus activation function.\n */\nexport class Softplus extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softplus';\n  apply(x: Tensor): Tensor {\n    return tfc.softplus(x);\n  }\n}\nserialization.registerClass(Softplus);\n\n/**\n * Softsign activation function.\n */\nexport class Softsign extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softsign';\n  apply(x: Tensor): Tensor {\n    return K.softsign(x);\n  }\n}\nserialization.registerClass(Softsign);\n\n/**\n * Hyperbolic tangent function.\n */\nexport class Tanh extends Activation {\n  /** @nocollapse */\n  static readonly className = 'tanh';\n  apply(x: Tensor): Tensor {\n    return tfc.tanh(x);\n  }\n}\nserialization.registerClass(Tanh);\n\n/**\n * Softmax activation function\n */\nexport class Softmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softmax';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.softmax(x, axis);\n  }\n}\nserialization.registerClass(Softmax);\n\n/**\n * Log softmax activation function\n */\nexport class LogSoftmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'logSoftmax';\n  /**\n   * Calculate the activation function of log softmax:\n   * log( exp(x_i) / sum(exp(x)) )\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.logSoftmax(x, axis);\n  }\n}\nserialization.registerClass(LogSoftmax);\n\n/**\n * Swish activation function\n */\nexport class Swish extends Activation {\n  /** @nocollapse */\n  static readonly className = 'swish';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param alpha Scaling factor for the sigmoid function.\n   * @returns a Tensor of the same shape as x\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return tidy(() => tfc.sigmoid(x.mul(alpha)).mul(x));\n  }\n}\nserialization.registerClass(Swish);\n\nexport function serializeActivation(activation: Activation): string {\n  return activation.getClassName();\n}\n\nexport function deserializeActivation(\n   config: serialization.ConfigDict,\n   customObjects: serialization.ConfigDict = {}): Activation {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'activation');\n}\n\nexport function getActivation(identifier: ActivationIdentifier|\n                              serialization.ConfigDict|Activation): Activation {\n  if (identifier == null) {\n    const config: serialization.ConfigDict = {};\n    config['className'] = 'linear';\n    config['config'] = {};\n    return deserializeActivation(config);\n  }\n  if (typeof identifier === 'string') {\n    const config: serialization.ConfigDict = {};\n    config['className'] = identifier;\n    config['config'] = {};\n    return deserializeActivation(config);\n  } else if (identifier instanceof Activation) {\n    return identifier;\n  } else {\n    return deserializeActivation(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* original source: keras/regularizers.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {abs, add, Scalar, serialization, sum, Tensor, tidy, zeros} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\nfunction assertObjectArgs(args: L1Args | L2Args | L1L2Args): void {\n  if (args != null && typeof args !== 'object') {\n    throw new Error(\n        `Argument to L1L2 regularizer's constructor is expected to be an ` +\n        `object, but received: ${args}`);\n  }\n}\n\n/**\n * Regularizer base class.\n */\nexport abstract class Regularizer extends serialization.Serializable {\n  abstract apply(x: Tensor): Scalar;\n}\n\nexport interface L1L2Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1?: number;\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2?: number;\n}\n\nexport interface L1Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1: number;\n}\n\nexport interface L2Args {\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2: number;\n}\n\nexport class L1L2 extends Regularizer {\n  /** @nocollapse */\n  static className = 'L1L2';\n\n  private readonly l1: number;\n  private readonly l2: number;\n  private readonly hasL1: boolean;\n  private readonly hasL2: boolean;\n  constructor(args?: L1L2Args) {\n    super();\n\n    assertObjectArgs(args);\n\n    this.l1 = args == null || args.l1 == null ? 0.01 : args.l1;\n    this.l2 = args == null || args.l2 == null ? 0.01 : args.l2;\n    this.hasL1 = this.l1 !== 0;\n    this.hasL2 = this.l2 !== 0;\n  }\n\n  /**\n   * Porting note: Renamed from __call__.\n   * @param x Variable of which to calculate the regularization score.\n   */\n  apply(x: Tensor): Scalar {\n    return tidy(() => {\n      let regularization: Tensor = zeros([1]);\n      if (this.hasL1) {\n        regularization = add(regularization, sum(tfc.mul(this.l1, abs(x))));\n      }\n      if (this.hasL2) {\n        regularization =\n            add(regularization, sum(tfc.mul(this.l2, K.square(x))));\n      }\n      return regularization.asScalar();\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {'l1': this.l1, 'l2': this.l2};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls({l1: config['l1'] as number, l2: config['l2'] as number});\n  }\n}\nserialization.registerClass(L1L2);\n\nexport function l1(args?: L1Args) {\n  assertObjectArgs(args);\n  return new L1L2({l1: args != null ? args.l1 : null, l2: 0});\n}\n\nexport function l2(args: L2Args) {\n  assertObjectArgs(args);\n  return new L1L2({l2: args != null ? args.l2 : null, l1: 0});\n}\n\n/** @docinline */\nexport type RegularizerIdentifier = 'l1l2'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding keras symbols.\nexport const REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in RegularizerIdentifier]: string} = {\n      'l1l2': 'L1L2'\n    };\n\nexport function serializeRegularizer(constraint: Regularizer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeRegularizer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Regularizer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'regularizer');\n}\n\nexport function getRegularizer(identifier: RegularizerIdentifier|\n                               serialization.ConfigDict|\n                               Regularizer): Regularizer {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeRegularizer(config);\n  } else if (identifier instanceof Regularizer) {\n    return identifier;\n  } else {\n    return deserializeRegularizer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n *  Advanced activation layers.\n */\n\nimport {clipByValue, elu, leakyRelu, prelu, relu, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {Softmax as softmaxActivation} from '../activations';\nimport {cast} from '../backend/tfjs_backend';\nimport {Constraint, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface ReLULayerArgs extends LayerArgs {\n  /**\n   * Float, the maximum output value.\n   */\n  maxValue?: number;\n}\n\nexport class ReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ReLU';\n  maxValue: number;\n\n  constructor(args?: ReLULayerArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maxValue = args.maxValue;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    let output = relu(inputs);\n    if (this.maxValue != null) {\n      output = clipByValue(output, 0, this.maxValue);\n    }\n    return output;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {maxValue: this.maxValue};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ReLU);\n\nexport declare interface LeakyReLULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `0.3`.\n   */\n  alpha?: number;\n}\n\nexport class LeakyReLU extends Layer {\n  /** @nocollapse */\n  static className = 'LeakyReLU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 0.3;\n\n  constructor(args?: LeakyReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return leakyRelu(x, this.alpha);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LeakyReLU);\n\nexport declare interface PReLULayerArgs extends LayerArgs {\n  /**\n   * Initializer for the learnable alpha.\n   */\n  alphaInitializer?: Initializer|InitializerIdentifier;\n\n  /**\n   * Regularizer for the learnable alpha.\n   */\n  alphaRegularizer?: Regularizer;\n\n  /**\n   * Constraint for the learnable alpha.\n   */\n  alphaConstraint?: Constraint;\n\n  /**\n   * The axes along which to share learnable parameters for the activation\n   * function. For example, if the incoming feature maps are from a 2D\n   * convolution with output shape `[numExamples, height, width, channels]`,\n   * and you wish to share parameters across space (height and width) so that\n   * each filter channels has only one set of parameters, set\n   * `shared_axes: [1, 2]`.\n   */\n  sharedAxes?: number|number[];\n}\n\nexport class PReLU extends Layer {\n  /** @nocollapse */\n  static className = 'PReLU';\n  private readonly alphaInitializer: Initializer;\n  private readonly alphaRegularizer: Regularizer;\n  private readonly alphaConstraint: Constraint;\n  private readonly sharedAxes: number[];\n  private alpha: LayerVariable;\n\n  readonly DEFAULT_ALPHA_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args?: PReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.supportsMasking = true;\n    this.alphaInitializer =\n        getInitializer(args.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER);\n    this.alphaRegularizer = getRegularizer(args.alphaRegularizer);\n    this.alphaConstraint = getConstraint(args.alphaConstraint);\n    if (args.sharedAxes == null) {\n      this.sharedAxes = null;\n    } else if (Array.isArray(args.sharedAxes)) {\n      this.sharedAxes = args.sharedAxes;\n    } else if (typeof args.sharedAxes === 'number') {\n      this.sharedAxes = [args.sharedAxes];\n    } else {\n      throw new ValueError(\n          `Expected sharedAxes to be a number or an array of numbers, ` +\n          `but got ${args.sharedAxes}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]) {\n    inputShape = getExactlyOneShape(inputShape);\n    const paramShape: Shape = inputShape.slice(1);\n    if (this.sharedAxes != null) {\n      for (const i of this.sharedAxes) {\n        paramShape[i - 1] = 1;\n      }\n    }\n    this.alpha = this.addWeight(\n        'alpha', paramShape, 'float32', this.alphaInitializer,\n        this.alphaRegularizer, true, this.alphaConstraint);\n    // Set input spec.\n    const axes: {[axis: number]: number} = {};\n    if (this.sharedAxes != null) {\n      for (let i = 1; i < inputShape.length; ++i) {\n        axes[i] = inputShape[i];\n      }\n    }\n    this.inputSpec = [new InputSpec({\n      ndim: inputShape.length,\n      axes,\n    })];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    return prelu(inputs, this.alpha.read());\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      alphaInitializer: serializeInitializer(this.alphaInitializer),\n      alphaRegularizer: serializeRegularizer(this.alphaRegularizer),\n      alphaConstraint: serializeConstraint(this.alphaConstraint),\n      sharedAxes: this.sharedAxes\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(PReLU);\n\nexport declare interface ELULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `1.0`.\n   */\n  alpha?: number;\n}\n\nexport class ELU extends Layer {\n  /** @nocollapse */\n  static className = 'ELU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 1.0;\n\n  constructor(args?: ELULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    if (args.alpha != null && args.alpha !== this.DEFAULT_ALPHA) {\n      throw new NotImplementedError(\n          `Non-default alpha value (${args.alpha}) is not supported by the ` +\n          `ELU layer yet.`);\n    }\n\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return elu(x);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ELU);\n\nexport declare interface ThresholdedReLULayerArgs extends LayerArgs {\n  /**\n   * Float >= 0. Threshold location of activation.\n   */\n  theta?: number;\n}\n\nexport class ThresholdedReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ThresholdedReLU';\n  readonly theta: number;\n\n  readonly DEFAULT_THETA = 1.0;\n\n  constructor(args?: ThresholdedReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.theta = args.theta == null ? this.DEFAULT_THETA : args.theta;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return x.mul(cast(x.greater(this.theta), 'float32'));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {theta: this.theta};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ThresholdedReLU);\n\nexport declare interface SoftmaxLayerArgs extends LayerArgs {\n  /**\n   * Integer, axis along which the softmax normalization is applied.\n   * Defaults to `-1` (i.e., the last axis).\n   */\n  axis?: number;\n}\n\nexport class Softmax extends Layer {\n  /** @nocollapse */\n  static className = 'Softmax';\n  readonly axis: number;\n  readonly softmax: (t: Tensor, a?: number) => Tensor;\n  readonly DEFAULT_AXIS = 1.0;\n\n  constructor(args?: SoftmaxLayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.softmax = new softmaxActivation().apply;\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return this.softmax(x, this.axis);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {axis: this.axis};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Softmax);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {ValueError} from '../errors';\nimport {PaddingMode} from '../keras_format/common';\n\nimport {pyListRepeat} from './generic_utils';\nimport {isInteger, max} from './math_utils';\n\n/**\n * Transforms a single number of array of numbers into an array of numbers.\n * @param value\n * @param n: The size of the tuple to be returned.\n * @param name: Name of the parameter, used for generating error messages.\n * @returns An array of numbers.\n */\nexport function normalizeArray(\n    value: number|number[], n: number, name: string): number[] {\n  if (typeof value === 'number') {\n    return pyListRepeat(value, n);\n  } else {\n    if (value.length !== n) {\n      throw new ValueError(\n          `The ${name} argument must be an integer or tuple of ${n} integers.` +\n          ` Received: ${value.length} elements.`);\n    }\n    for (let i = 0; i < n; ++i) {\n      const singleValue = value[i];\n      if (!isInteger(singleValue)) {\n        throw new ValueError(\n            `The ${name} argument must be an integer or tuple of ${n}` +\n            ` integers. Received: ${JSON.stringify(value)} including a` +\n            ` non-integer number ${singleValue}`);\n      }\n    }\n    return value;\n  }\n}\n\n/**\n * Determines output length of a convolution given input length.\n * @param inputLength\n * @param filterSize\n * @param padding\n * @param stride\n * @param dilation: dilation rate.\n */\nexport function convOutputLength(\n    inputLength: number, filterSize: number, padding: PaddingMode,\n    stride: number, dilation = 1): number {\n  if (inputLength == null) {\n    return inputLength;\n  }\n  const dilatedFilterSize = filterSize + (filterSize - 1) * (dilation - 1);\n  let outputLength: number;\n  if (padding === 'same') {\n    outputLength = inputLength;\n  } else {  // VALID\n    outputLength = inputLength - dilatedFilterSize + 1;\n  }\n  return Math.floor((outputLength + stride - 1) / stride);\n}\n\nexport function deconvLength(\n    dimSize: number, strideSize: number, kernelSize: number,\n    padding: PaddingMode): number {\n  if (dimSize == null) {\n    return null;\n  }\n\n  if (padding === 'valid') {\n    dimSize = dimSize * strideSize + max([kernelSize - strideSize, 0]);\n  } else if (padding === 'same') {\n    dimSize = dimSize * strideSize;\n  } else {\n    throw new ValueError(`Unsupport padding mode: ${padding}.`);\n  }\n  return dimSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {fused, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, PaddingMode, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, deconvLength, normalizeArray} from '../utils/conv_utils';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Transpose and cast the input before the conv2d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv2DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  // TODO(cais): Cast type to float32 if not.\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 1]);  // NCHW -> NHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * Transpose and cast the input before the conv3d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv3DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 4, 1]);  // NCDHW -> NDHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * 1D-convolution with bias added.\n *\n * Porting Note: This function does not exist in the Python Keras backend.\n *   It is exactly the same as `conv2d`, except the added `bias`.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.\n * @param bias Bias, rank-3, of shape `[outDepth]`.\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    // Check the ranks of x, kernel and bias.\n    if (x.shape.length !== 3) {\n      throw new ValueError(\n          `The input of a conv1dWithBias operation should be 3, but is ` +\n          `${x.shape.length} instead.`);\n    }\n    if (kernel.shape.length !== 3) {\n      throw new ValueError(\n          `The kernel for a conv1dWithBias operation should be 3, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    if (bias != null && bias.shape.length !== 1) {\n      throw new ValueError(\n          `The bias for a conv1dWithBias operation should be 1, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    // TODO(cais): Support CAUSAL padding mode.\n    if (dataFormat === 'channelsFirst') {\n      x = tfc.transpose(x, [0, 2, 1]);  // NCW -> NWC.\n    }\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    let y: Tensor = tfc.conv1d(\n        x as Tensor2D | Tensor3D, kernel as Tensor3D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias);\n    }\n    return y;\n  });\n}\n\n/**\n * 1D-convolution.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.s\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1d(\n    x: Tensor, kernel: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv1dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 2D pooling.\n */\nexport function conv2d(\n    x: Tensor, kernel: Tensor, strides = [1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv2dWithBiasActivation(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution with an added bias and optional activation.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv2d`, except the added `bias`.\n */\nexport function conv2dWithBiasActivation(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat, dilationRate?: [number, number],\n    activation: fused.Activation = null): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 3 && x.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects input to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    if (kernel.rank !== 3 && kernel.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects kernel to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.fused.conv2d({\n      x: y as Tensor3D | Tensor4D,\n      filter: kernel as Tensor4D,\n      strides: strides as [number, number],\n      pad: padding === 'same' ? 'same' : 'valid',\n      dilations: dilationRate,\n      dataFormat: 'NHWC',\n      bias,\n      activation\n    });\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\n/**\n * 3D Convolution.\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 3D convolution.\n */\nexport function conv3d(\n    x: Tensor, kernel: Tensor, strides = [1, 1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv3dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 3D Convolution with an added bias.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv3d`, except the added `bias`.\n */\nexport function conv3dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 4 && x.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects input to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    if (kernel.rank !== 4 && kernel.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects kernel to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    let y = preprocessConv3DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv3dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.conv3d(\n        y as Tensor4D | tfc.Tensor<tfc.Rank.R5>,\n        kernel as tfc.Tensor<tfc.Rank.R5>, strides as [number, number, number],\n        padding === 'same' ? 'same' : 'valid', 'NDHWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias as Tensor1D);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);\n    }\n    return y;\n  });\n}\n\n/**\n * Base LayerConfig for depthwise and non-depthwise convolutional layers.\n */\nexport declare interface BaseConvLayerArgs extends LayerArgs {\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number]|[number, number, number];\n\n  /**\n   * Activation function of the layer.\n   *\n   * If you don't specify the activation, none is applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector. Defaults to `true`.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the convolutional kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the convolutional kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\n/**\n * LayerConfig for non-depthwise convolutional layers.\n * Applies to non-depthwise convolution of all ranks (e.g, Conv1D, Conv2D,\n * Conv3D).\n */\nexport declare interface ConvLayerArgs extends BaseConvLayerArgs {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n}\n\n/**\n * Abstract convolution layer.\n */\nexport abstract class BaseConv extends Layer {\n  protected readonly rank: number;\n  protected readonly kernelSize: number[];\n  protected readonly strides: number[];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n  protected readonly activation: Activation;\n  protected readonly useBias: boolean;\n  protected readonly dilationRate: number[];\n\n  // Bias-related members are here because all convolution subclasses use the\n  // same configuration parmeters to control bias.  Kernel-related members\n  // are in subclass `Conv` because some subclasses use different parameters to\n  // control kernel properties, for instance, `DepthwiseConv2D` uses\n  // `depthwiseInitializer` instead of `kernelInitializer`.\n  protected readonly biasInitializer?: Initializer;\n  protected readonly biasConstraint?: Constraint;\n  protected readonly biasRegularizer?: Regularizer;\n\n  protected bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(rank: number, args: BaseConvLayerArgs) {\n    super(args as LayerArgs);\n    BaseConv.verifyArgs(args);\n    this.rank = rank;\n    generic_utils.assertPositiveInteger(this.rank, 'rank');\n    if (this.rank !== 1 && this.rank !== 2 && this.rank !== 3) {\n      throw new NotImplementedError(\n          `Convolution layer for rank other than 1, 2, or 3 (${\n              this.rank}) is ` +\n          `not implemented yet.`);\n    }\n    this.kernelSize = normalizeArray(args.kernelSize, rank, 'kernelSize');\n    this.strides = normalizeArray(\n        args.strides == null ? 1 : args.strides, rank, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.activation = getActivation(args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.dilationRate = normalizeArray(\n        args.dilationRate == null ? 1 : args.dilationRate, rank,\n        'dilationRate');\n    if (this.rank === 1 &&\n        (Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)) {\n      throw new ValueError(\n          `dilationRate must be a number or an array of a single number ` +\n          `for 1D convolution, but received ` +\n          `${JSON.stringify(this.dilationRate)}`);\n    } else if (this.rank === 2) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate = [this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 2) {\n        throw new ValueError(\n            `dilationRate must be a number or array of two numbers for 2D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    } else if (this.rank === 3) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate =\n            [this.dilationRate, this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 3) {\n        throw new ValueError(\n            `dilationRate must be a number or array of three numbers for 3D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    }\n  }\n\n  protected static verifyArgs(args: BaseConvLayerArgs) {\n    // Check config.kernelSize type and shape.\n    generic_utils.assert(\n        'kernelSize' in args, `required key 'kernelSize' not in config`);\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 3)) {\n      throw new ValueError(\n          `BaseConv expects config.kernelSize to be number or number[] with ` +\n          `length 1, 2, or 3, but received ${\n              JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      kernelSize: this.kernelSize,\n      strides: this.strides,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\n/**\n * Abstract nD convolution layer.  Ancestor of convolution layers which reduce\n * across channels, i.e., Conv1D and Conv2D, but not DepthwiseConv2D.\n */\nexport abstract class Conv extends BaseConv {\n  protected readonly filters: number;\n\n  protected kernel: LayerVariable = null;\n\n  // Bias-related properties are stored in the superclass `BaseConv` because all\n  // convolution subclasses use the same configuration parameters to control\n  // bias. Kernel-related properties are defined here rather than in the\n  // superclass because some convolution subclasses use different names and\n  // configuration parameters for their internal kernel state.\n  protected readonly kernelInitializer?: Initializer;\n  protected readonly kernelConstraint?: Constraint;\n  protected readonly kernelRegularizer?: Regularizer;\n\n  constructor(rank: number, args: ConvLayerArgs) {\n    super(rank, args as BaseConvLayerArgs);\n    Conv.verifyArgs(args);\n    this.filters = args.filters;\n    generic_utils.assertPositiveInteger(this.filters, 'filters');\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n    const inputDim = inputShape[channelAxis];\n\n    const kernelShape = this.kernelSize.concat([inputDim, this.filters]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.inputSpec = [{ndim: this.rank + 2, axes: {[channelAxis]: inputDim}}];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs: Tensor;\n      const biasValue = this.bias == null ? null : this.bias.read();\n      const fusedActivationName = generic_utils.mapActivationToFusedKernel(\n                                            this.activation.getClassName());\n\n      if (fusedActivationName != null && this.rank === 2) {\n        outputs = conv2dWithBiasActivation(\n            inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n            this.dataFormat, this.dilationRate as [number, number],\n            fusedActivationName);\n      } else {\n        if (this.rank === 1) {\n          outputs = conv1dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides[0],\n              this.padding, this.dataFormat, this.dilationRate[0]);\n        } else if (this.rank === 2) {\n          // TODO(cais): Move up to constructor.\n          outputs = conv2dWithBiasActivation(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number]);\n        } else if (this.rank === 3) {\n          outputs = conv3dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number, number]);\n        } else {\n          throw new NotImplementedError(\n              'convolutions greater than 3D are not implemented yet.');\n        }\n\n        if (this.activation != null) {\n          outputs = this.activation.apply(outputs);\n        }\n      }\n\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const newSpace: number[] = [];\n    const space = (this.dataFormat === 'channelsLast') ?\n        inputShape.slice(1, inputShape.length - 1) :\n        inputShape.slice(2);\n    for (let i = 0; i < space.length; ++i) {\n      const newDim = convOutputLength(\n          space[i], this.kernelSize[i], this.padding, this.strides[i],\n          typeof this.dilationRate === 'number' ? this.dilationRate :\n                                                  this.dilationRate[i]);\n      newSpace.push(newDim);\n    }\n\n    let outputShape = [inputShape[0]];\n    if (this.dataFormat === 'channelsLast') {\n      outputShape = outputShape.concat(newSpace);\n      outputShape.push(this.filters);\n    } else {\n      outputShape.push(this.filters);\n      outputShape = outputShape.concat(newSpace);\n    }\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      filters: this.filters,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // Check config.filters type, shape, and value.\n    if (!('filters' in args) || typeof args.filters !== 'number' ||\n        args.filters < 1) {\n      throw new ValueError(\n          `Convolution layer expected config.filters to be a 'number' > 0 ` +\n          `but got ${JSON.stringify(args.filters)}`);\n    }\n  }\n}\n\nexport class Conv2D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv2D';\n  constructor(args: ConvLayerArgs) {\n    super(2, args);\n    Conv2D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if ((typeof args.kernelSize !== 'number') &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 2)) {\n      throw new ValueError(\n          `Conv2D expects config.kernelSize to be number or number[] with ` +\n          `length 1 or 2, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv2D);\n\nexport class Conv3D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv3D';\n  constructor(args: ConvLayerArgs) {\n    super(3, args);\n    Conv3D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number') {\n      if (!(Array.isArray(args.kernelSize) &&\n            (args.kernelSize.length === 1 || args.kernelSize.length === 3))) {\n        throw new ValueError(\n            `Conv3D expects config.kernelSize to be number or` +\n            ` [number, number, number], but received ${\n                JSON.stringify(args.kernelSize)}.`);\n      }\n    }\n  }\n}\nserialization.registerClass(Conv3D);\n\nexport class Conv2DTranspose extends Conv2D {\n  /** @nocollapse */\n  static className = 'Conv2DTranspose';\n  inputSpec: InputSpec[];\n\n  constructor(args: ConvLayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n\n    if (this.padding !== 'same' && this.padding !== 'valid') {\n      throw new ValueError(\n          `Conv2DTranspose currently supports only padding modes 'same' ` +\n          `and 'valid', but received padding mode ${this.padding}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    if (inputShape.length !== 4) {\n      throw new ValueError(\n          'Input should have rank 4; Received input shape: ' +\n          JSON.stringify(inputShape));\n    }\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          'The channel dimension of the inputs should be defined. ' +\n          'Found `None`.');\n    }\n    const inputDim = inputShape[channelAxis];\n    const kernelShape = this.kernelSize.concat([this.filters, inputDim]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, 'float32', this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    // Set input spec.\n    this.inputSpec =\n        [new InputSpec({ndim: 4, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs);\n      if (input.shape.length !== 4) {\n        throw new ValueError(\n            `Conv2DTranspose.call() expects input tensor to be rank-4, but ` +\n            `received a tensor of rank-${input.shape.length}`);\n      }\n\n      const inputShape = input.shape;\n      const batchSize = inputShape[0];\n\n      let hAxis: number;\n      let wAxis: number;\n      if (this.dataFormat === 'channelsFirst') {\n        hAxis = 2;\n        wAxis = 3;\n      } else {\n        hAxis = 1;\n        wAxis = 2;\n      }\n\n      const height = inputShape[hAxis];\n      const width = inputShape[wAxis];\n      const kernelH = this.kernelSize[0];\n      const kernelW = this.kernelSize[1];\n      const strideH = this.strides[0];\n      const strideW = this.strides[1];\n\n      // Infer the dynamic output shape.\n      const outHeight = deconvLength(height, strideH, kernelH, this.padding);\n      const outWidth = deconvLength(width, strideW, kernelW, this.padding);\n\n      // Porting Note: We don't branch based on `this.dataFormat` here,\n      // because\n      //   the tjfs-core function `conv2dTranspose` called below always\n      //   assumes channelsLast.\n      const outputShape: [number, number, number, number] =\n          [batchSize, outHeight, outWidth, this.filters];\n\n      if (this.dataFormat !== 'channelsLast') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n      }\n      let outputs = tfc.conv2dTranspose(\n          input as Tensor4D, this.kernel.read() as Tensor4D, outputShape,\n          this.strides as [number, number], this.padding as 'same' | 'valid');\n      if (this.dataFormat !== 'channelsLast') {\n        outputs = tfc.transpose(outputs, [0, 3, 1, 2]);\n      }\n\n      if (this.bias != null) {\n        outputs =\n            K.biasAdd(outputs, this.bias.read(), this.dataFormat) as Tensor4D;\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs) as Tensor4D;\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n\n    let channelAxis: number;\n    let heightAxis: number;\n    let widthAxis: number;\n    if (this.dataFormat === 'channelsFirst') {\n      channelAxis = 1;\n      heightAxis = 2;\n      widthAxis = 3;\n    } else {\n      channelAxis = 3;\n      heightAxis = 1;\n      widthAxis = 2;\n    }\n\n    const kernelH = this.kernelSize[0];\n    const kernelW = this.kernelSize[1];\n    const strideH = this.strides[0];\n    const strideW = this.strides[1];\n\n    outputShape[channelAxis] = this.filters;\n    outputShape[heightAxis] =\n        deconvLength(outputShape[heightAxis], strideH, kernelH, this.padding);\n    outputShape[widthAxis] =\n        deconvLength(outputShape[widthAxis], strideW, kernelW, this.padding);\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['dilationRate'];\n    return config;\n  }\n}\nserialization.registerClass(Conv2DTranspose);\n\nexport declare interface SeparableConvLayerArgs extends ConvLayerArgs {\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal\n   * to `filtersIn * depthMultiplier`. Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the pointwise kernel matrix.\n   */\n  pointwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the pointwise kernel matrix.\n   */\n  pointwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the pointwise kernel matrix.\n   */\n  pointwiseConstraint?: ConstraintIdentifier|Constraint;\n}\n\nexport class SeparableConv extends Conv {\n  /** @nocollapse */\n  static className = 'SeparableConv';\n\n  readonly depthMultiplier: number;\n\n  protected readonly depthwiseInitializer?: Initializer;\n  protected readonly depthwiseRegularizer?: Regularizer;\n  protected readonly depthwiseConstraint?: Constraint;\n  protected readonly pointwiseInitializer?: Initializer;\n  protected readonly pointwiseRegularizer?: Regularizer;\n  protected readonly pointwiseConstraint?: Constraint;\n\n  readonly DEFAULT_DEPTHWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n  readonly DEFAULT_POINTWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n\n  protected depthwiseKernel: LayerVariable = null;\n  protected pointwiseKernel: LayerVariable = null;\n\n  constructor(rank: number, config?: SeparableConvLayerArgs) {\n    super(rank, config);\n\n    if (config.filters == null) {\n      throw new ValueError(\n          'The `filters` configuration field is required by SeparableConv, ' +\n          'but is unspecified.');\n    }\n    if (config.kernelInitializer != null || config.kernelRegularizer != null ||\n        config.kernelConstraint != null) {\n      throw new ValueError(\n          'Fields kernelInitializer, kernelRegularizer and kernelConstraint ' +\n          'are invalid for SeparableConv2D. Use depthwiseInitializer, ' +\n          'depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, ' +\n          'pointwiseRegularizer and pointwiseConstraint instead.');\n    }\n    if (config.padding != null && config.padding !== 'same' &&\n        config.padding !== 'valid') {\n      throw new ValueError(\n          `SeparableConv${this.rank}D supports only padding modes: ` +\n          `'same' and 'valid', but received ${JSON.stringify(config.padding)}`);\n    }\n\n    this.depthMultiplier =\n        config.depthMultiplier == null ? 1 : config.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER);\n    this.depthwiseRegularizer = getRegularizer(config.depthwiseRegularizer);\n    this.depthwiseConstraint = getConstraint(config.depthwiseConstraint);\n    this.pointwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER);\n    this.pointwiseRegularizer = getRegularizer(config.pointwiseRegularizer);\n    this.pointwiseConstraint = getConstraint(config.pointwiseConstraint);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < this.rank + 2) {\n      throw new ValueError(\n          `Inputs to SeparableConv${this.rank}D should have rank ` +\n          `${this.rank + 2}, but received input shape: ` +\n          `${JSON.stringify(inputShape)}`);\n    }\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          `The channel dimension of the inputs should be defined, ` +\n          `but found ${JSON.stringify(inputShape[channelAxis])}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape =\n        this.kernelSize.concat([inputDim, this.depthMultiplier]);\n    const pointwiseKernelShape = [];\n    for (let i = 0; i < this.rank; ++i) {\n      pointwiseKernelShape.push(1);\n    }\n    pointwiseKernelShape.push(inputDim * this.depthMultiplier, this.filters);\n\n    const trainable = true;\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, 'float32',\n        this.depthwiseInitializer, this.depthwiseRegularizer, trainable,\n        this.depthwiseConstraint);\n    this.pointwiseKernel = this.addWeight(\n        'pointwise_kernel', pointwiseKernelShape, 'float32',\n        this.pointwiseInitializer, this.pointwiseRegularizer, trainable,\n        this.pointwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, trainable, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n\n    this.inputSpec =\n        [new InputSpec({ndim: this.rank + 2, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      let output: Tensor;\n      if (this.rank === 1) {\n        throw new NotImplementedError(\n            '1D separable convolution is not implemented yet.');\n      } else if (this.rank === 2) {\n        if (this.dataFormat === 'channelsFirst') {\n          inputs = tfc.transpose(inputs, [0, 2, 3, 1]);  // NCHW -> NHWC.\n        }\n\n        output = tfc.separableConv2d(\n            inputs as Tensor4D, this.depthwiseKernel.read() as Tensor4D,\n            this.pointwiseKernel.read() as Tensor4D,\n            this.strides as [number, number], this.padding as 'same' | 'valid',\n            this.dilationRate as [number, number], 'NHWC');\n      }\n\n      if (this.useBias) {\n        output = K.biasAdd(output, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      if (this.dataFormat === 'channelsFirst') {\n        output = tfc.transpose(output, [0, 3, 1, 2]);  // NHWC -> NCHW.\n      }\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['kernelInitializer'];\n    delete config['kernelRegularizer'];\n    delete config['kernelConstraint'];\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['pointwiseInitializer'] =\n        serializeInitializer(this.pointwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['pointwiseRegularizer'] =\n        serializeRegularizer(this.pointwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseConstraint);\n    config['pointwiseConstraint'] =\n        serializeConstraint(this.pointwiseConstraint);\n    return config;\n  }\n}\n\nexport class SeparableConv2D extends SeparableConv {\n  /** @nocollapse */\n  static className = 'SeparableConv2D';\n  constructor(args?: SeparableConvLayerArgs) {\n    super(2, args);\n  }\n}\nserialization.registerClass(SeparableConv2D);\n\nexport class Conv1D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv1D';\n  constructor(args: ConvLayerArgs) {\n    super(1, args);\n    Conv1D.verifyArgs(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['dataFormat'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 1)) {\n      throw new ValueError(\n          `Conv1D expects config.kernelSize to be number or number[] with ` +\n          `length 1, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv1D);\n\nexport declare interface Cropping2DLayerArgs extends LayerArgs {\n  /**\n   * Dimension of the cropping along the width and the height.\n   * - If integer: the same symmetric cropping\n   *  is applied to width and height.\n   * - If list of 2 integers:\n   *   interpreted as two different\n   *   symmetric cropping values for height and width:\n   *   `[symmetric_height_crop, symmetric_width_crop]`.\n   * - If a list of 2 list of 2 integers:\n   *   interpreted as\n   *   `[[top_crop, bottom_crop], [left_crop, right_crop]]`\n   */\n  cropping: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   * `channels_first` corresponds to inputs with shape\n   *   `(batch, channels, ...)`\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class Cropping2D extends Layer {\n  /** @nocollapse */\n  static className = 'Cropping2D';\n  protected readonly cropping: [[number, number], [number, number]];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Cropping2DLayerArgs) {\n    super(args);\n    if (typeof args.cropping === 'number') {\n      this.cropping =\n          [[args.cropping, args.cropping], [args.cropping, args.cropping]];\n    } else if (typeof args.cropping[0] === 'number') {\n      this.cropping = [\n        [args.cropping[0], args.cropping[0]],\n        [args.cropping[1] as number, args.cropping[1] as number]\n      ];\n    } else {\n      this.cropping = args.cropping as [[number, number], [number, number]];\n    }\n    this.dataFormat =\n        args.dataFormat === undefined ? 'channelsLast' : args.dataFormat;\n    this.inputSpec = [{ndim: 4}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      return [\n        inputShape[0], inputShape[1],\n        inputShape[2] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[3] - this.cropping[1][0] - this.cropping[1][1]\n      ];\n    } else {\n      return [\n        inputShape[0],\n        inputShape[1] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[2] - this.cropping[1][0] - this.cropping[1][1], inputShape[3]\n      ];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      if (this.dataFormat === 'channelsLast') {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[1] - this.cropping[0][0] - this.cropping[0][1], 2);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[2] - this.cropping[1][1] - this.cropping[1][0], 3);\n      } else {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[2] - this.cropping[0][0] - this.cropping[0][1], 3);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[3] - this.cropping[1][1] - this.cropping[1][0], 4);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {cropping: this.cropping, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Cropping2D);\n\nexport declare interface UpSampling2DLayerArgs extends LayerArgs {\n  /**\n   * The upsampling factors for rows and columns.\n   *\n   * Defaults to `[2, 2]`.\n   */\n  size?: number[];\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `\"channelsLast\"` corresponds to inputs with shape\n   *   `[batch, ..., channels]`\n   *\n   *  `\"channelsFirst\"` corresponds to inputs with shape `[batch, channels,\n   * ...]`.\n   *\n   * Defaults to `\"channelsLast\"`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class UpSampling2D extends Layer {\n  /** @nocollapse */\n  static className = 'UpSampling2D';\n  protected readonly DEFAULT_SIZE = [2, 2];\n  protected readonly size: number[];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: UpSampling2DLayerArgs) {\n    super(args);\n    this.inputSpec = [{ndim: 4}];\n    this.size = args.size == null ? this.DEFAULT_SIZE : args.size;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      const height =\n          inputShape[2] == null ? null : this.size[0] * inputShape[2];\n      const width = inputShape[3] == null ? null : this.size[1] * inputShape[3];\n      return [inputShape[0], inputShape[1], height, width];\n    } else {\n      const height =\n          inputShape[1] == null ? null : this.size[0] * inputShape[1];\n      const width = inputShape[2] == null ? null : this.size[1] * inputShape[2];\n      return [inputShape[0], height, width, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs) as Tensor4D;\n      const inputShape = input.shape;\n\n      if (this.dataFormat === 'channelsFirst') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n        const height = this.size[0] * inputShape[2];\n        const width = this.size[1] * inputShape[3];\n        const resized = input.resizeNearestNeighbor([height, width]);\n        return tfc.transpose(resized, [0, 3, 1, 2]);\n      } else {\n        const height = this.size[0] * inputShape[1];\n        const width = this.size[1] * inputShape[2];\n        return input.resizeNearestNeighbor([height, width]);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {size: this.size, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(UpSampling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Depthwise Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {BaseConv, BaseConvLayerArgs, ConvLayerArgs, preprocessConv2DInput} from './convolutional';\n\n/**\n * 2D convolution with separable filters.\n * @param x Input tensor.\n * @param depthwiseKernel Convolution kernel for depthwise convolution.\n * @param strides Strides (Array of two integers).\n * @param padding Padding model.\n * @param dataFormat Data format.\n * @param dilationRate Array of two integers, dilation rates for the separable\n *   convolution.\n * @returns Output tensor.\n * @throws ValueError If depthwiseKernel is not a 4D array.\n */\nexport function depthwiseConv2d(\n    x: Tensor, depthwiseKernel: Tensor, strides: [number, number] = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `Input for depthwiseConv2d is required to be 4-D, but is instead ` +\n          `${x.rank}-D`);\n    }\n    if (depthwiseKernel.rank !== 4) {\n      throw new ValueError(\n          `depthwiseKernel is required to be 4-D, but is instead ` +\n          `${depthwiseKernel.rank}-D`);\n    }\n    y = tfc.depthwiseConv2d(\n        y as Tensor4D, depthwiseKernel as Tensor4D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NHWC', dilationRate);\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\nexport declare interface DepthwiseConv2DLayerArgs extends BaseConvLayerArgs {\n  /**\n   * An integer or Array of 2 integers, specifying the width and height of the\n   * 2D convolution window. Can be a single integer to specify the same value\n   * for all spatial dimensions.\n   */\n  kernelSize: number|[number, number];\n\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal to\n   * `filtersIn * depthMultiplier`.\n   * Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   * Default: GlorotNormal.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regulzarizer function for the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class DepthwiseConv2D extends BaseConv {\n  /** @nocollapse */\n  static className = 'DepthwiseConv2D';\n  private readonly depthMultiplier: number;\n  private readonly depthwiseInitializer: Initializer;\n  private readonly depthwiseConstraint: Constraint;\n  private readonly depthwiseRegularizer: Regularizer;\n\n  private depthwiseKernel: LayerVariable = null;\n\n  constructor(args: DepthwiseConv2DLayerArgs) {\n    super(2, args as ConvLayerArgs);\n    this.depthMultiplier =\n        args.depthMultiplier == null ? 1 : args.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        args.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.depthwiseConstraint = getConstraint(args.depthwiseConstraint);\n    this.depthwiseRegularizer = getRegularizer(args.depthwiseRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 4) {\n      throw new ValueError(\n          `Inputs to DepthwiseConv2D should have rank 4. ` +\n          `Received input shape: ${JSON.stringify(inputShape)}.`);\n    }\n    const channelAxis = this.dataFormat === 'channelsFirst' ? 1 : 3;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          'The channel dimension of the inputs to DepthwiseConv2D should ' +\n          `be defined, but is not (${inputShape[channelAxis]}).`);\n    }\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape: Shape = [\n      this.kernelSize[0], this.kernelSize[1], inputDim, this.depthMultiplier\n    ];\n\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, null,\n        this.depthwiseInitializer, this.depthwiseRegularizer, true,\n        this.depthwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [inputDim * this.depthMultiplier], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs = depthwiseConv2d(\n          inputs, this.depthwiseKernel.read(), this.strides as [number, number],\n          this.padding, this.dataFormat, null);\n      // TODO(cais): Add support for dilation.\n      if (this.useBias) {\n        outputs = K.biasAdd(outputs, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs);\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    const cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    const outFilters = this.dataFormat === 'channelsFirst' ?\n        inputShape[1] * this.depthMultiplier :\n        inputShape[3] * this.depthMultiplier;\n    const outRows = convOutputLength(\n        rows, this.kernelSize[0], this.padding, this.strides[0]);\n    const outCols = convOutputLength(\n        cols, this.kernelSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], outFilters, outRows, outCols];\n    } else {\n      // In this case, assume 'channelsLast'.\n      return [inputShape[0], outRows, outCols, outFilters];\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    config['depthMultiplier'] = this.depthMultiplier;\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseRegularizer);\n    return config;\n  }\n}\nserialization.registerClass(DepthwiseConv2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Recurrent Neural Network Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, SymbolicTensor} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, Ones, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs, RnnStepFunction} from '../types';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor, isArrayOfShapes} from '../utils/types_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\nimport {deserialize} from './serialization';\n\n/**\n * Standardize `apply()` args to a single list of tensor inputs.\n *\n * When running a model loaded from file, the input tensors `initialState` and\n * `constants` are passed to `RNN.apply()` as part of `inputs` instead of the\n * dedicated kwargs fields. `inputs` consists of\n * `[inputs, initialState0, initialState1, ..., constant0, constant1]` in this\n * case.\n * This method makes sure that arguments are\n * separated and that `initialState` and `constants` are `Array`s of tensors\n * (or None).\n *\n * @param inputs Tensor or `Array` of  tensors.\n * @param initialState Tensor or `Array` of tensors or `null`/`undefined`.\n * @param constants Tensor or `Array` of tensors or `null`/`undefined`.\n * @returns An object consisting of\n *   inputs: A tensor.\n *   initialState: `Array` of tensors or `null`.\n *   constants: `Array` of tensors or `null`.\n * @throws ValueError, if `inputs` is an `Array` but either `initialState` or\n *   `constants` is provided.\n */\nexport function standardizeArgs(\n    inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    initialState: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    constants: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    numConstants?: number): {\n  inputs: Tensor|SymbolicTensor,\n  initialState: Tensor[]|SymbolicTensor[],\n  constants: Tensor[]|SymbolicTensor[]\n} {\n  if (Array.isArray(inputs)) {\n    if (initialState != null || constants != null) {\n      throw new ValueError(\n          'When inputs is an array, neither initialState or constants ' +\n          'should be provided');\n    }\n    if (numConstants != null) {\n      constants = inputs.slice(inputs.length - numConstants, inputs.length);\n      inputs = inputs.slice(0, inputs.length - numConstants);\n    }\n    if (inputs.length > 1) {\n      initialState = inputs.slice(1, inputs.length);\n    }\n    inputs = inputs[0];\n  }\n\n  function toListOrNull(x: Tensor|Tensor[]|SymbolicTensor|\n                        SymbolicTensor[]): Tensor[]|SymbolicTensor[] {\n    if (x == null || Array.isArray(x)) {\n      return x as Tensor[] | SymbolicTensor[];\n    } else {\n      return [x] as Tensor[] | SymbolicTensor[];\n    }\n  }\n\n  initialState = toListOrNull(initialState);\n  constants = toListOrNull(constants);\n\n  return {inputs, initialState, constants};\n}\n\n/**\n * Iterates over the time dimension of a tensor.\n *\n * @param stepFunction RNN step function.\n *   Parameters:\n *     inputs: tensor with shape `[samples, ...]` (no time dimension),\n *       representing input for the batch of samples at a certain time step.\n *     states: an Array of tensors.\n *   Returns:\n *     outputs: tensor with shape `[samples, outputDim]` (no time dimension).\n *     newStates: list of tensors, same length and shapes as `states`. The first\n *       state in the list must be the output tensor at the previous timestep.\n * @param inputs Tensor of temporal data of shape `[samples, time, ...]` (at\n *   least 3D).\n * @param initialStates Tensor with shape `[samples, outputDim]` (no time\n *   dimension), containing the initial values of the states used in the step\n *   function.\n * @param goBackwards If `true`, do the iteration over the time dimension in\n *   reverse order and return the reversed sequence.\n * @param mask Binary tensor with shape `[sample, time, 1]`, with a zero for\n *   every element that is masked.\n * @param constants An Array of constant values passed at each step.\n * @param unroll Whether to unroll the RNN or to use a symbolic loop. *Not*\n *   applicable to this imperative deeplearn.js backend. Its value is ignored.\n * @param needPerStepOutputs Whether the per-step outputs are to be\n *   concatenated into a single tensor and returned (as the second return\n *   value). Default: `false`. This arg is included so that the relatively\n *   expensive concatenation of the stepwise outputs can be omitted unless\n *   the stepwise outputs need to be kept (e.g., for an LSTM layer of which\n *   `returnSequence` is `true`.)\n * @returns An Array: `[lastOutput, outputs, newStates]`.\n *   lastOutput: the lastest output of the RNN, of shape `[samples, ...]`.\n *   outputs: tensor with shape `[samples, time, ...]` where each entry\n *     `output[s, t]` is the output of the step function at time `t` for sample\n *     `s`. This return value is provided if and only if the\n *     `needPerStepOutputs` is set as `true`. If it is set as `false`, this\n *     return value will be `undefined`.\n *   newStates: Array of tensors, latest states returned by the step function,\n *      of shape `(samples, ...)`.\n * @throws ValueError If input dimension is less than 3.\n *\n * TODO(nielsene): This needs to be tidy-ed.\n */\nexport function rnn(\n    stepFunction: RnnStepFunction, inputs: Tensor, initialStates: Tensor[],\n    goBackwards = false, mask?: Tensor, constants?: Tensor[], unroll = false,\n    needPerStepOutputs = false): [Tensor, Tensor, Tensor[]] {\n  return tfc.tidy(() => {\n    const ndim = inputs.shape.length;\n    if (ndim < 3) {\n      throw new ValueError(`Input should be at least 3D, but is ${ndim}D.`);\n    }\n\n    // Transpose to time-major, i.e., from [batch, time, ...] to [time, batch,\n    // ...].\n    const axes = [1, 0].concat(math_utils.range(2, ndim));\n    inputs = tfc.transpose(inputs, axes);\n\n    if (constants != null) {\n      throw new NotImplementedError(\n          'The rnn() functoin of the deeplearn.js backend does not support ' +\n          'constants yet.');\n    }\n\n    // Porting Note: the unroll option is ignored by the imperative backend.\n    if (unroll) {\n      console.warn(\n          'Backend rnn(): the unroll = true option is not applicable to the ' +\n          'imperative deeplearn.js backend.');\n    }\n\n    if (mask != null) {\n      mask = mask.asType('bool').asType('float32');\n      if (mask.rank === ndim - 1) {\n        mask = tfc.expandDims(mask, -1);\n      }\n      mask = tfc.transpose(mask, axes);\n    }\n\n    if (goBackwards) {\n      inputs = tfc.reverse(inputs, 0);\n      if (mask != null) {\n        mask = tfc.reverse(mask, 0);\n      }\n    }\n\n    // Porting Note: PyKeras with TensorFlow backend uses a symbolic loop\n    //   (tf.while_loop). But for the imperative deeplearn.js backend, we just\n    //   use the usual TypeScript control flow to iterate over the time steps in\n    //   the inputs.\n    // Porting Note: PyKeras patches a \"_use_learning_phase\" attribute to\n    // outputs.\n    //   This is not idiomatic in TypeScript. The info regarding whether we are\n    //   in a learning (i.e., training) phase for RNN is passed in a different\n    //   way.\n\n    const perStepOutputs: Tensor[] = [];\n    let lastOutput: Tensor;\n    let states = initialStates;\n    const timeSteps = inputs.shape[0];\n    const perStepInputs = tfc.unstack(inputs);\n    let perStepMasks: Tensor[];\n    if (mask != null) {\n      perStepMasks = tfc.unstack(mask);\n    }\n\n    for (let t = 0; t < timeSteps; ++t) {\n      const currentInput = perStepInputs[t];\n      const stepOutputs = tfc.tidy(() => stepFunction(currentInput, states));\n\n      if (mask == null) {\n        lastOutput = stepOutputs[0];\n        states = stepOutputs[1];\n      } else {\n        const maskedOutputs = tfc.tidy(() => {\n          const stepMask = perStepMasks[t];\n          const negStepMask = tfc.onesLike(stepMask).sub(stepMask);\n          // TODO(cais): Would tfc.where() be better for performance?\n          const output =\n              stepOutputs[0].mul(stepMask).add(states[0].mul(negStepMask));\n          const newStates = states.map((state, i) => {\n            return stepOutputs[1][i].mul(stepMask).add(state.mul(negStepMask));\n          });\n          return {output, newStates};\n        });\n        lastOutput = maskedOutputs.output;\n        states = maskedOutputs.newStates;\n      }\n\n      if (needPerStepOutputs) {\n        perStepOutputs.push(lastOutput);\n      }\n    }\n    let outputs: Tensor;\n    if (needPerStepOutputs) {\n      const axis = 1;\n      outputs = tfc.stack(perStepOutputs, axis);\n    }\n    return [lastOutput, outputs, states] as [Tensor, Tensor, Tensor[]];\n  });\n}\n\nexport declare interface BaseRNNLayerArgs extends LayerArgs {\n  /**\n   * A RNN cell instance. A RNN cell is a class that has:\n   *   - a `call()` method, which takes `[Tensor, Tensor]` as the\n   *     first input argument. The first item is the input at time t, and\n   *     second item is the cell state at time t.\n   *     The `call()` method returns `[outputAtT, statesAtTPlus1]`.\n   *     The `call()` method of the cell can also take the argument `constants`,\n   *     see section \"Note on passing external constants\" below.\n   *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,\n   *       which are Layer subtypes, to accept two arguments. tfjs-layers does\n   *       not do such overriding. Instead we preseve the `call()` signature,\n   *       which due to its `Tensor|Tensor[]` argument and return value, is\n   *       flexible enough to handle the inputs and states.\n   *   - a `stateSize` attribute. This can be a single integer (single state)\n   *     in which case it is the size of the recurrent state (which should be\n   *     the same as the size of the cell output). This can also be an Array of\n   *     integers (one size per state). In this case, the first entry\n   *     (`stateSize[0]`) should be the same as the size of the cell output.\n   * It is also possible for `cell` to be a list of RNN cell instances, in which\n   * case the cells get stacked on after the other in the RNN, implementing an\n   * efficient stacked RNN.\n   */\n  cell?: RNNCell|RNNCell[];\n\n  /**\n   * Whether to return the last output in the output sequence, or the full\n   * sequence.\n   */\n  returnSequences?: boolean;\n\n  /**\n   * Whether to return the last state in addition to the output.\n   */\n  returnState?: boolean;\n\n  /**\n   * If `true`, process the input sequence backwards and return the reversed\n   * sequence (default: `false`).\n   */\n  goBackwards?: boolean;\n\n  /**\n   * If `true`, the last state for each sample at index i in a batch will be\n   * used as initial state of the sample of index i in the following batch\n   * (default: `false`).\n   *\n   * You can set RNN layers to be \"stateful\", which means that the states\n   * computed for the samples in one batch will be reused as initial states\n   * for the samples in the next batch. This assumes a one-to-one mapping\n   * between samples in different successive batches.\n   *\n   * To enable \"statefulness\":\n   *   - specify `stateful: true` in the layer constructor.\n   *   - specify a fixed batch size for your model, by passing\n   *     - if sequential model:\n   *       `batchInputShape: [...]` to the first layer in your model.\n   *     - else for functional model with 1 or more Input layers:\n   *       `batchShape: [...]` to all the first layers in your model.\n   *     This is the expected shape of your inputs\n   *     *including the batch size*.\n   *     It should be a tuple of integers, e.g., `[32, 10, 100]`.\n   *   - specify `shuffle: false` when calling `LayersModel.fit()`.\n   *\n   * To reset the state of your model, call `resetStates()` on either the\n   * specific layer or on the entire model.\n   */\n  stateful?: boolean;\n  // TODO(cais): Explore whether we can warn users when they fail to set\n  //   `shuffle: false` when training a model consisting of stateful RNNs\n  //   and any stateful Layers in general.\n\n  /**\n   * If `true`, the network will be unrolled, else a symbolic loop will be\n   * used. Unrolling can speed-up a RNN, although it tends to be more memory-\n   * intensive. Unrolling is only suitable for short sequences (default:\n   * `false`).\n   * Porting Note: tfjs-layers has an imperative backend. RNNs are executed with\n   *   normal TypeScript control flow. Hence this property is inapplicable and\n   *   ignored in tfjs-layers.\n   */\n  unroll?: boolean;\n\n  /**\n   * Dimensionality of the input (integer).\n   *   This option (or alternatively, the option `inputShape`) is required when\n   *   this layer is used as the first layer in a model.\n   */\n  inputDim?: number;\n\n  /**\n   * Length of the input sequences, to be specified when it is constant.\n   * This argument is required if you are going to connect `Flatten` then\n   * `Dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed). Note that if the recurrent layer is not the first layer in\n   * your model, you would need to specify the input length at the level of the\n   * first layer (e.g., via the `inputShape` option).\n   */\n  inputLength?: number;\n}\n\nexport class RNN extends Layer {\n  /** @nocollapse */\n  static className = 'RNN';\n  public readonly cell: RNNCell;\n  public readonly returnSequences: boolean;\n  public readonly returnState: boolean;\n  public readonly goBackwards: boolean;\n  public readonly unroll: boolean;\n\n  public stateSpec: InputSpec[];\n  protected states_: Tensor[];\n\n  // NOTE(cais): For stateful RNNs, the old states cannot be disposed right\n  // away when new states are set, because the old states may need to be used\n  // later for backpropagation through time (BPTT) and other purposes. So we\n  // keep them here for final disposal when the state is reset completely\n  // (i.e., through no-arg call to `resetStates()`).\n  protected keptStates: Tensor[][];\n\n  private numConstants: number;\n\n  constructor(args: RNNLayerArgs) {\n    super(args);\n    let cell: RNNCell;\n    if (args.cell == null) {\n      throw new ValueError(\n          'cell property is missing for the constructor of RNN.');\n    } else if (Array.isArray(args.cell)) {\n      cell = new StackedRNNCells({cells: args.cell});\n    } else {\n      cell = args.cell;\n    }\n    if (cell.stateSize == null) {\n      throw new ValueError(\n          'The RNN cell should have an attribute `stateSize` (tuple of ' +\n          'integers, one integer per RNN state).');\n    }\n    this.cell = cell;\n    this.returnSequences =\n        args.returnSequences == null ? false : args.returnSequences;\n    this.returnState = args.returnState == null ? false : args.returnState;\n    this.goBackwards = args.goBackwards == null ? false : args.goBackwards;\n    this._stateful = args.stateful == null ? false : args.stateful;\n    this.unroll = args.unroll == null ? false : args.unroll;\n\n    this.supportsMasking = true;\n    this.inputSpec = [new InputSpec({ndim: 3})];\n    this.stateSpec = null;\n    this.states_ = null;\n    // TODO(cais): Add constantsSpec and numConstants.\n    this.numConstants = null;\n    // TODO(cais): Look into the use of initial_state in the kwargs of the\n    //   constructor.\n\n    this.keptStates = [];\n  }\n\n  // Porting Note: This is the equivalent of `RNN.states` property getter in\n  //   PyKeras.\n  getStates(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      return math_utils.range(0, numStates).map(x => null);\n    } else {\n      return this.states_;\n    }\n  }\n\n  // Porting Note: This is the equivalent of the `RNN.states` property setter in\n  //   PyKeras.\n  setStates(states: Tensor[]): void {\n    this.states_ = states;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    // TODO(cais): Remove the casting once stacked RNN cells become supported.\n    let stateSize = this.cell.stateSize;\n    if (!Array.isArray(stateSize)) {\n      stateSize = [stateSize];\n    }\n    const outputDim = stateSize[0];\n    let outputShape: Shape|Shape[];\n    if (this.returnSequences) {\n      outputShape = [inputShape[0], inputShape[1], outputDim];\n    } else {\n      outputShape = [inputShape[0], outputDim];\n    }\n\n    if (this.returnState) {\n      const stateShape: Shape[] = [];\n      for (const dim of stateSize) {\n        stateShape.push([inputShape[0], dim]);\n      }\n      return [outputShape].concat(stateShape);\n    } else {\n      return outputShape;\n    }\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tfc.tidy(() => {\n      if (Array.isArray(mask)) {\n        mask = mask[0];\n      }\n      const outputMask = this.returnSequences ? mask : null;\n\n      if (this.returnState) {\n        const stateMask = this.states.map(s => null);\n        return [outputMask].concat(stateMask);\n      } else {\n        return outputMask;\n      }\n    });\n  }\n\n  /**\n   * Get the current state tensors of the RNN.\n   *\n   * If the state hasn't been set, return an array of `null`s of the correct\n   * length.\n   */\n  get states(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      const output: Tensor[] = [];\n      for (let i = 0; i < numStates; ++i) {\n        output.push(null);\n      }\n      return output;\n    } else {\n      return this.states_;\n    }\n  }\n\n  set states(s: Tensor[]) {\n    this.states_ = s;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    // Note inputShape will be an Array of Shapes of initial states and\n    // constants if these are passed in apply().\n    const constantShape: Shape[] = null;\n    if (this.numConstants != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    }\n\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    const batchSize: number = this.stateful ? inputShape[0] : null;\n    const inputDim = inputShape.slice(2);\n    this.inputSpec[0] = new InputSpec({shape: [batchSize, null, ...inputDim]});\n\n    // Allow cell (if RNNCell Layer) to build before we set or validate\n    // stateSpec.\n    const stepInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (constantShape != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    } else {\n      this.cell.build(stepInputShape);\n    }\n\n    // Set or validate stateSpec.\n    let stateSize: number[];\n    if (Array.isArray(this.cell.stateSize)) {\n      stateSize = this.cell.stateSize;\n    } else {\n      stateSize = [this.cell.stateSize];\n    }\n\n    if (this.stateSpec != null) {\n      if (!util.arraysEqual(\n              this.stateSpec.map(spec => spec.shape[spec.shape.length - 1]),\n              stateSize)) {\n        throw new ValueError(\n            `An initialState was passed that is not compatible with ` +\n            `cell.stateSize. Received stateSpec=${this.stateSpec}; ` +\n            `However cell.stateSize is ${this.cell.stateSize}`);\n      }\n    } else {\n      this.stateSpec =\n          stateSize.map(dim => new InputSpec({shape: [null, dim]}));\n    }\n    if (this.stateful) {\n      this.resetStates();\n    }\n  }\n\n  /**\n   * Reset the state tensors of the RNN.\n   *\n   * If the `states` argument is `undefined` or `null`, will set the\n   * state tensor(s) of the RNN to all-zero tensors of the appropriate\n   * shape(s).\n   *\n   * If `states` is provided, will set the state tensors of the RNN to its\n   * value.\n   *\n   * @param states Optional externally-provided initial states.\n   * @param training Whether this call is done during training. For stateful\n   *   RNNs, this affects whether the old states are kept or discarded. In\n   *   particular, if `training` is `true`, the old states will be kept so\n   *   that subsequent backpropgataion through time (BPTT) may work properly.\n   *   Else, the old states will be discarded.\n   */\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n      const batchSize = this.inputSpec[0].shape[0];\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n      // Initialize state if null.\n      if (this.states_ == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_ = [tfc.zeros([batchSize, this.cell.stateSize])];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_[0] = tfc.zeros([batchSize, this.cell.stateSize]);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training === true) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n          const dim = Array.isArray(this.cell.stateSize) ?\n              this.cell.stateSize[index] :\n              this.cell.stateSize;\n          const expectedShape = [batchSize, dim];\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n          this.states_[index] = value;\n        }\n      }\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    // TODO(cais): Figure out whether initialState is in kwargs or inputs.\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    // If any of `initial_state` or `constants` are specified and are\n    // `tf.SymbolicTensor`s, then add them to the inputs and temporarily modify\n    // the input_spec to include them.\n\n    let additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    let additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      kwargs['initialState'] = initialState;\n      additionalInputs = additionalInputs.concat(initialState);\n      this.stateSpec = [];\n      for (const state of initialState) {\n        this.stateSpec.push(new InputSpec({shape: state.shape}));\n      }\n      // TODO(cais): Use the following instead.\n      // this.stateSpec = initialState.map(state => new InputSpec({shape:\n      // state.shape}));\n      additionalSpecs = additionalSpecs.concat(this.stateSpec);\n    }\n    if (constants != null) {\n      kwargs['constants'] = constants;\n      additionalInputs = additionalInputs.concat(constants);\n      // TODO(cais): Add this.constantsSpec.\n      this.numConstants = constants.length;\n    }\n\n    const isTensor = additionalInputs[0] instanceof SymbolicTensor;\n    if (isTensor) {\n      // Compute full input spec, including state and constants.\n      const fullInput =\n          [inputs].concat(additionalInputs) as Tensor[] | SymbolicTensor[];\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call with temporarily replaced inputSpec.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output = super.apply(fullInput, kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  // tslint:disable-next-line:no-any\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    // Input shape: `[samples, time (padded with zeros), input_dim]`.\n    // Note that the .build() method of subclasses **must** define\n    // this.inputSpec and this.stateSpec owith complete input shapes.\n    return tidy(() => {\n      const mask = kwargs == null ? null : kwargs['mask'] as Tensor;\n      const training = kwargs == null ? null : kwargs['training'];\n      let initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      inputs = getExactlyOneTensor(inputs);\n      if (initialState == null) {\n        if (this.stateful) {\n          initialState = this.states_;\n        } else {\n          initialState = this.getInitialState(inputs);\n        }\n      }\n\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      if (initialState.length !== numStates) {\n        throw new ValueError(\n            `RNN Layer has ${numStates} state(s) but was passed ` +\n            `${initialState.length} initial state(s).`);\n      }\n      if (this.unroll) {\n        console.warn(\n            'Ignoring unroll = true for RNN layer, due to imperative backend.');\n      }\n\n      const cellCallKwargs: Kwargs = {training};\n\n      // TODO(cais): Add support for constants.\n      const step = (inputs: Tensor, states: Tensor[]) => {\n        // `inputs` and `states` are concatenated to form a single `Array` of\n        // `tf.Tensor`s as the input to `cell.call()`.\n        const outputs =\n            this.cell.call([inputs].concat(states), cellCallKwargs) as Tensor[];\n        // Marshall the return value into output and new states.\n        return [outputs[0], outputs.slice(1)] as [Tensor, Tensor[]];\n      };\n\n      // TODO(cais): Add support for constants.\n\n      const rnnOutputs =\n          rnn(step, inputs, initialState, this.goBackwards, mask, null,\n              this.unroll, this.returnSequences);\n      const lastOutput = rnnOutputs[0];\n      const outputs = rnnOutputs[1];\n      const states = rnnOutputs[2];\n\n      if (this.stateful) {\n        this.resetStates(states, training);\n      }\n\n      const output = this.returnSequences ? outputs : lastOutput;\n\n      // TODO(cais): Porperty set learning phase flag.\n\n      if (this.returnState) {\n        return [output].concat(states);\n      } else {\n        return output;\n      }\n    });\n  }\n\n  getInitialState(inputs: Tensor): Tensor[] {\n    return tidy(() => {\n      // Build an all-zero tensor of shape [samples, outputDim].\n      // [Samples, timeSteps, inputDim].\n      let initialState = tfc.zeros(inputs.shape);\n      // [Samples].\n      initialState = tfc.sum(initialState, [1, 2]);\n      initialState = K.expandDims(initialState);  // [Samples, 1].\n\n      if (Array.isArray(this.cell.stateSize)) {\n        return this.cell.stateSize.map(\n            dim => dim > 1 ? K.tile(initialState, [1, dim]) : initialState);\n      } else {\n        return this.cell.stateSize > 1 ?\n            [K.tile(initialState, [1, this.cell.stateSize])] :\n            [initialState];\n      }\n    });\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    return this.cell.trainableWeights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    if (!this.trainable) {\n      return this.cell.weights;\n    }\n    return this.cell.nonTrainableWeights;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.cell != null) {\n      this.cell.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      returnSequences: this.returnSequences,\n      returnState: this.returnState,\n      goBackwards: this.goBackwards,\n      stateful: this.stateful,\n      unroll: this.unroll,\n    };\n\n    if (this.numConstants != null) {\n      config['numConstants'] = this.numConstants;\n    }\n\n    const cellConfig = this.cell.getConfig();\n\n    if (this.getClassName() === RNN.className) {\n      config['cell'] = {\n        'className': this.cell.getClassName(),\n        'config': cellConfig,\n      } as serialization.ConfigDictValue;\n    }\n\n    // this order is necessary, to prevent cell name from replacing layer name\n    return {...cellConfig, ...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cellConfig = config['cell'] as serialization.ConfigDict;\n    const cell = deserialize(cellConfig, customObjects) as RNNCell;\n    return new cls(Object.assign(config, {cell}));\n  }\n}\nserialization.registerClass(RNN);\n\n// Porting Note: This is a common parent class for RNN cells. There is no\n// equivalent of this in PyKeras. Having a common parent class forgoes the\n//  need for `has_attr(cell, ...)` checks or its TypeScript equivalent.\n/**\n * An RNNCell layer.\n *\n * @doc {heading: 'Layers', subheading: 'Classes'}\n */\nexport abstract class RNNCell extends Layer {\n  /**\n   * Size(s) of the states.\n   * For RNN cells with only a single state, this is a single integer.\n   */\n  public stateSize: number|number[];\n  public dropoutMask: Tensor|Tensor[];\n  public recurrentDropoutMask: Tensor|Tensor[];\n}\n\nexport declare interface SimpleRNNCellLayerArgs extends LayerArgs {\n  /**\n   * units: Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   * Default: hyperbolic tangent ('tanh').\n   * If you pass `null`,  'linear' activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the `kernel` weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the `recurrent_kernel` weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the `kernel` weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the `recurrentKernel` weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraintfunction applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\nexport class SimpleRNNCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'SimpleRNNCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args: SimpleRNNCellLayerArgs) {\n    super(args);\n    this.units = args.units;\n    assertPositiveInteger(this.units, `units`);\n    this.activation = getActivation(\n        args.activation == null ? this.DEFAULT_ACTIVATION : args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    // TODO(cais): Use regularizer.\n    this.kernel = this.addWeight(\n        'kernel', [inputShape[inputShape.length - 1], this.units], null,\n        this.kernelInitializer, this.kernelRegularizer, true,\n        this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  // Porting Note: PyKeras' equivalent of this method takes two tensor inputs:\n  //   `inputs` and `states`. Here, the two tensors are combined into an\n  //   `Tensor[]` Array as the first input argument.\n  //   Similarly, PyKeras' equivalent of this method returns two values:\n  //    `output` and `[output]`. Here the two are combined into one length-2\n  //    `Tensor[]`, consisting of `output` repeated.\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `SimpleRNNCell expects 2 input Tensors, got ${inputs.length}.`);\n      }\n      let prevOutput = inputs[1];\n      inputs = inputs[0];\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training\n                           }) as Tensor;\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(prevOutput),\n                                      rate: this.recurrentDropout,\n                                      training\n                                    }) as Tensor;\n      }\n      let h: Tensor;\n      const dpMask: Tensor = this.dropoutMask as Tensor;\n      const recDpMask: Tensor = this.recurrentDropoutMask as Tensor;\n      if (dpMask != null) {\n        h = K.dot(tfc.mul(inputs, dpMask), this.kernel.read());\n      } else {\n        h = K.dot(inputs, this.kernel.read());\n      }\n      if (this.bias != null) {\n        h = K.biasAdd(h, this.bias.read());\n      }\n      if (recDpMask != null) {\n        prevOutput = tfc.mul(prevOutput, recDpMask);\n      }\n      let output = tfc.add(h, K.dot(prevOutput, this.recurrentKernel.read()));\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      // TODO(cais): Properly set learning phase on output tensor?\n      return [output, output];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(SimpleRNNCell);\n\nexport declare interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {\n  /**\n   * Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   *\n   * Defaults to  hyperbolic tangent (`tanh`)\n   *\n   * If you pass `null`, no activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the recurrentKernel weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the kernel weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the recurrentKernel weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\n/**\n * RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the\n * `cell` property required. This interface is to be used with constructors\n * of concrete RNN layer subtypes.\n */\nexport declare interface RNNLayerArgs extends BaseRNNLayerArgs {\n  cell: RNNCell|RNNCell[];\n}\n\nexport class SimpleRNN extends RNN {\n  /** @nocollapse */\n  static className = 'SimpleRNN';\n  constructor(args: SimpleRNNLayerArgs) {\n    args.cell = new SimpleRNNCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\nserialization.registerClass(SimpleRNN);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n\n  /**\n   * GRU convention (whether to apply reset gate after or before matrix\n   * multiplication). false = \"before\", true = \"after\" (only false is\n   * supported).\n   */\n  resetAfter?: boolean;\n}\n\nexport class GRUCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'GRUCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION: ActivationIdentifier = 'hardSigmoid';\n\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: GRUCellLayerArgs) {\n    super(args);\n    if (args.resetAfter) {\n      throw new ValueError(\n          `GRUCell does not support reset_after parameter set to true.`);\n    }\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 3], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 3], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units * 3], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `GRUCell expects 2 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      let hTMinus1 = inputs[1];  // Previous memory state.\n      inputs = inputs[0];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2, regardless of the actual value of\n      // config.implementation.\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 3\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 3\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor];\n      const recDpMask = this.recurrentDropoutMask as [Tensor, Tensor, Tensor];\n      let z: Tensor;\n      let r: Tensor;\n      let hh: Tensor;\n\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let matrixX = K.dot(inputs, this.kernel.read());\n      if (this.useBias) {\n        matrixX = K.biasAdd(matrixX, this.bias.read());\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n\n      const recurrentKernelValue = this.recurrentKernel.read();\n      const [rk1, rk2] = tfc.split(\n          recurrentKernelValue, [2 * this.units, this.units],\n          recurrentKernelValue.rank - 1);\n      const matrixInner = K.dot(hTMinus1, rk1);\n\n      const [xZ, xR, xH] = tfc.split(matrixX, 3, matrixX.rank - 1);\n      const [recurrentZ, recurrentR] =\n          tfc.split(matrixInner, 2, matrixInner.rank - 1);\n      z = this.recurrentActivation.apply(tfc.add(xZ, recurrentZ));\n      r = this.recurrentActivation.apply(tfc.add(xR, recurrentR));\n\n      const recurrentH = K.dot(tfc.mul(r, hTMinus1), rk2);\n      hh = this.activation.apply(tfc.add(xH, recurrentH));\n\n      const h =\n          tfc.add(tfc.mul(z, hTMinus1), tfc.mul(tfc.add(1, tfc.neg(z)), hh));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n      resetAfter: false\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(GRUCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface GRULayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   * smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class GRU extends RNN {\n  /** @nocollapse */\n  static className = 'GRU';\n  constructor(args: GRULayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new GRUCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(GRU);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class LSTMCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'LSTMCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n  readonly unitForgetBias: boolean;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number[];\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION = 'hardSigmoid';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n\n  readonly DEFAULT_BIAS_INITIALIZER = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: LSTMCellLayerArgs) {\n    super(args);\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.unitForgetBias = args.unitForgetBias;\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = [this.units, this.units];\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 4], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 4], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    let biasInitializer: Initializer;\n    if (this.useBias) {\n      if (this.unitForgetBias) {\n        const capturedBiasInit = this.biasInitializer;\n        const capturedUnits = this.units;\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): Tensor {\n            // TODO(cais): More informative variable names?\n            const bI = capturedBiasInit.apply([capturedUnits]);\n            const bF = (new Ones()).apply([capturedUnits]);\n            const bCAndH = capturedBiasInit.apply([capturedUnits * 2]);\n            return K.concatAlongFirstAxis(\n                K.concatAlongFirstAxis(bI, bF), bCAndH);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n      this.bias = this.addWeight(\n          'bias', [this.units * 4], null, biasInitializer, this.biasRegularizer,\n          true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `LSTMCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n      let hTMinus1 = inputs[1];    // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n      inputs = inputs[0];\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 4\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 4\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor, Tensor];\n      const recDpMask =\n          this.recurrentDropoutMask as [Tensor, Tensor, Tensor, Tensor];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2 regardless of the actual value of\n      // config.implementation.\n      let i: Tensor;\n      let f: Tensor;\n      let c: Tensor;\n      let o: Tensor;\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let z = K.dot(inputs, this.kernel.read());\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n      z = tfc.add(z, K.dot(hTMinus1, this.recurrentKernel.read()));\n      if (this.useBias) {\n        z = K.biasAdd(z, this.bias.read());\n      }\n\n      const [z0, z1, z2, z3] = tfc.split(z, 4, z.rank - 1);\n\n      i = this.recurrentActivation.apply(z0);\n      f = this.recurrentActivation.apply(z1);\n      c = tfc.add(tfc.mul(f, cTMinus1), tfc.mul(i, this.activation.apply(z2)));\n      o = this.recurrentActivation.apply(z3);\n\n      const h = tfc.mul(o, this.activation.apply(c));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      unitForgetBias: this.unitForgetBias,\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(LSTMCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface LSTMLayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *   Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions, whereas mode 2 will\n   *   batch them into fewer, larger operations. These modes will\n   *   have different performance profiles on different hardware and\n   *   for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this config field.\n   */\n  implementation?: number;\n}\n\nexport class LSTM extends RNN {\n  /** @nocollapse */\n  static className = 'LSTM';\n  constructor(args: LSTMLayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new LSTMCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(LSTM);\n\nexport declare interface StackedRNNCellsArgs extends LayerArgs {\n  /**\n   * A `Array` of `RNNCell` instances.\n   */\n  cells: RNNCell[];\n}\n\nexport class StackedRNNCells extends RNNCell {\n  /** @nocollapse */\n  static className = 'StackedRNNCells';\n  protected cells: RNNCell[];\n\n  constructor(args: StackedRNNCellsArgs) {\n    super(args);\n    this.cells = args.cells;\n  }\n\n  get stateSize(): number[] {\n    // States are a flat list in reverse order of the cell stack.\n    // This allows perserving the requirement `stack.statesize[0] ===\n    // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,\n    // assuming one LSTM has states `[h, c]`.\n    const stateSize: number[] = [];\n    for (const cell of this.cells.slice().reverse()) {\n      if (Array.isArray(cell.stateSize)) {\n        stateSize.push(...cell.stateSize);\n      } else {\n        stateSize.push(cell.stateSize);\n      }\n    }\n    return stateSize;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      let states = inputs.slice(1);\n\n      // Recover per-cell states.\n      const nestedStates: Tensor[][] = [];\n      for (const cell of this.cells.slice().reverse()) {\n        if (Array.isArray(cell.stateSize)) {\n          nestedStates.push(states.splice(0, cell.stateSize.length));\n        } else {\n          nestedStates.push(states.splice(0, 1));\n        }\n      }\n      nestedStates.reverse();\n\n      // Call the cells in order and store the returned states.\n      const newNestedStates: Tensor[][] = [];\n      let callInputs: Tensor[];\n      for (let i = 0; i < this.cells.length; ++i) {\n        const cell = this.cells[i];\n        states = nestedStates[i];\n        // TODO(cais): Take care of constants.\n        if (i === 0) {\n          callInputs = [inputs[0]].concat(states);\n        } else {\n          callInputs = [callInputs[0]].concat(states);\n        }\n        callInputs = cell.call(callInputs, kwargs) as Tensor[];\n        newNestedStates.push(callInputs.slice(1));\n      }\n\n      // Format the new states as a flat list in reverse cell order.\n      states = [];\n      for (const cellStates of newNestedStates.slice().reverse()) {\n        states.push(...cellStates);\n      }\n      return [callInputs[0]].concat(states);\n    });\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    if (isArrayOfShapes(inputShape)) {\n      // TODO(cais): Take care of input constants.\n      // const constantShape = inputShape.slice(1);\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n    let outputDim: number;\n    this.cells.forEach((cell, i) => {\n      nameScope(`RNNCell_${i}`, () => {\n        // TODO(cais): Take care of input constants.\n\n        cell.build(inputShape);\n        if (Array.isArray(cell.stateSize)) {\n          outputDim = cell.stateSize[0];\n        } else {\n          outputDim = cell.stateSize;\n        }\n        inputShape = [inputShape[0], outputDim] as Shape;\n      });\n    });\n    this.built = true;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const getCellConfig = (cell: RNNCell) => {\n      return {\n        'className': cell.getClassName(),\n        'config': cell.getConfig(),\n      };\n    };\n\n    const cellConfigs = this.cells.map(getCellConfig);\n\n    const config = {'cells': cellConfigs};\n\n    return {...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cells: RNNCell[] = [];\n    for (const cellConfig of (config['cells'] as serialization.ConfigDict[])) {\n      cells.push(deserialize(cellConfig, customObjects) as RNNCell);\n    }\n    return new cls({cells});\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const cell of this.cells) {\n        trainableWeights.push(...cell.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  /**\n   * Retrieve the weights of a the model.\n   *\n   * @returns A flat `Array` of `tf.Tensor`s.\n   */\n  getWeights(): Tensor[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.weights);\n    }\n    return batchGetValue(weights);\n  }\n\n  /**\n   * Set the weights of the model.\n   *\n   * @param weights An `Array` of `tf.Tensor`s with shapes and types matching\n   *     the output of `getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    const tuples: Array<[LayerVariable, Tensor]> = [];\n    for (const cell of this.cells) {\n      const numParams = cell.weights.length;\n      const inputWeights = weights.splice(numParams);\n      for (let i = 0; i < cell.weights.length; ++i) {\n        tuples.push([cell.weights[i], inputWeights[i]]);\n      }\n    }\n    batchSetValue(tuples);\n  }\n\n  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.\n}\nserialization.registerClass(StackedRNNCells);\n\nexport function generateDropoutMask(args: {\n  ones: () => tfc.Tensor,\n  rate: number,\n  training?: boolean,\n  count?: number,\n}): tfc.Tensor|tfc.Tensor[] {\n  const {ones, rate, training = false, count = 1} = args;\n\n  const droppedInputs = () => K.dropout(ones(), rate);\n\n  const createMask = () => K.inTrainPhase(droppedInputs, ones, training);\n\n  // just in case count is provided with null or undefined\n  if (!count || count <= 1) {\n    return tfc.keep(createMask().clone());\n  }\n\n  const masks = Array(count).fill(undefined).map(createMask);\n\n  return masks.map(m => tfc.keep(m.clone()));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {Activation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint} from '../constraints';\nimport {InputSpec} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {Initializer} from '../initializers';\nimport {DataFormat, DataType, PaddingMode, Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, normalizeArray} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\nimport {BaseRNNLayerArgs, generateDropoutMask, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNNCellLayerArgs} from './recurrent';\n\ndeclare interface ConvRNN2DCellArgs extends\n    Omit<SimpleRNNCellLayerArgs, 'units'> {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number];\n}\n\nabstract class ConvRNN2DCell extends RNNCell {\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n}\n\ndeclare interface ConvRNN2DLayerArgs extends BaseRNNLayerArgs,\n                                             ConvRNN2DCellArgs {}\n\n/**\n * Base class for convolutional-recurrent layers.\n */\nclass ConvRNN2D extends RNN {\n  /** @nocollapse */\n  static className = 'ConvRNN2D';\n\n  readonly cell: ConvRNN2DCell;\n\n  constructor(args: ConvRNN2DLayerArgs) {\n    if (args.unroll) {\n      throw new NotImplementedError(\n          'Unrolling is not possible with convolutional RNNs.');\n    }\n\n    if (Array.isArray(args.cell)) {\n      throw new NotImplementedError(\n          'It is not possible at the moment to stack convolutional cells.');\n    }\n\n    super(args as RNNLayerArgs);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n\n        this.cell.dropoutMask = null;\n      }\n\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n\n        this.cell.recurrentDropoutMask = null;\n      }\n\n      if (kwargs && kwargs['constants']) {\n        throw new ValueError('ConvRNN2D cell does not support constants');\n      }\n\n      const mask = kwargs == null ? null : kwargs['mask'];\n\n      const training = kwargs == null ? null : kwargs['training'];\n\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  computeOutputShape(inputShape: Shape): Shape|Shape[] {\n    let outShape: Shape = this.computeSingleOutputShape(inputShape);\n\n    if (!this.returnSequences) {\n      outShape = [outShape[0], ...outShape.slice(2)];\n    }\n\n    if (this.returnState) {\n      outShape =\n          [outShape, ...Array(2).fill([inputShape[0], ...outShape.slice(-3)])];\n    }\n\n    return outShape;\n  }\n\n  getInitialState(inputs: tfc.Tensor): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      const {stateSize} = this.cell;\n\n      const inputShape = inputs.shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const initialState = tfc.zeros(stateShape);\n\n      if (Array.isArray(stateSize)) {\n        return Array(stateSize.length).fill(initialState);\n      }\n\n      return [initialState];\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tfc.tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n\n      const inputShape = this.inputSpec[0].shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const batchSize = inputShape[0];\n\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n\n      // Initialize state if null.\n      if (this.getStates() == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_ = [tfc.zeros(stateShape)];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_[0] = tfc.zeros(stateShape);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n\n          const expectedShape = stateShape;\n\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n\n          this.states_[index] = value;\n        }\n      }\n\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  protected computeSingleOutputShape(inputShape: Shape): Shape {\n    const {dataFormat, filters, kernelSize, padding, strides, dilationRate} =\n        this.cell;\n\n    const isChannelsFirst = dataFormat === 'channelsFirst';\n\n    const h = inputShape[isChannelsFirst ? 3 : 2];\n    const w = inputShape[isChannelsFirst ? 4 : 3];\n\n    const hOut = convOutputLength(\n        h, kernelSize[0], padding, strides[0], dilationRate[0]);\n    const wOut = convOutputLength(\n        w, kernelSize[1], padding, strides[1], dilationRate[1]);\n\n    const outShape: Shape = [\n      ...inputShape.slice(0, 2),\n      ...(isChannelsFirst ? [filters, hOut, wOut] : [hOut, wOut, filters])\n    ];\n\n    return outShape;\n  }\n}\n\nexport declare interface ConvLSTM2DCellArgs extends\n    Omit<LSTMCellLayerArgs, 'units'>, ConvRNN2DCellArgs {}\n\nexport class ConvLSTM2DCell extends LSTMCell implements ConvRNN2DCell {\n  /** @nocollapse */\n  static className = 'ConvLSTM2DCell';\n\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  constructor(args: ConvLSTM2DCellArgs) {\n    const {\n      filters,\n      kernelSize,\n      strides,\n      padding,\n      dataFormat,\n      dilationRate,\n    } = args;\n\n    super({...args, units: filters});\n\n    this.filters = filters;\n    assertPositiveInteger(this.filters, 'filters');\n\n    this.kernelSize = normalizeArray(kernelSize, 2, 'kernelSize');\n    this.kernelSize.forEach(size => assertPositiveInteger(size, 'kernelSize'));\n\n    this.strides = normalizeArray(strides || 1, 2, 'strides');\n    this.strides.forEach(stride => assertPositiveInteger(stride, 'strides'));\n\n    this.padding = padding || 'valid';\n    checkPaddingMode(this.padding);\n\n    this.dataFormat = dataFormat || 'channelsLast';\n    checkDataFormat(this.dataFormat);\n\n    this.dilationRate = normalizeArray(dilationRate || 1, 2, 'dilationRate');\n    this.dilationRate.forEach(\n        rate => assertPositiveInteger(rate, 'dilationRate'));\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n\n    const numOfKernels = 4;\n\n    const kernelShape =\n        this.kernelSize.concat([inputDim, this.filters * numOfKernels]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n\n    const recurrentKernelShape =\n        this.kernelSize.concat([this.filters, this.filters * numOfKernels]);\n\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', recurrentKernelShape, null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n\n    if (this.useBias) {\n      let biasInitializer: Initializer;\n\n      if (this.unitForgetBias) {\n        const init = this.biasInitializer;\n\n        const filters = this.filters;\n\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): tfc.Tensor {\n            const biasI = init.apply([filters]);\n            const biasF = tfc.ones([filters]);\n            const biasCAndO = init.apply([filters * 2]);\n            return K.concatenate([biasI, biasF, biasCAndO]);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n\n      this.bias = this.addWeight(\n          'bias', [this.filters * numOfKernels], null, biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: tfc.Tensor[], kwargs: Kwargs): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] || false;\n\n      const x = inputs[0];         // Current input\n      const hTMinus1 = inputs[1];  // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n\n      const numOfKernels = 4;\n\n      type DropoutMasks = [tfc.Tensor, tfc.Tensor, tfc.Tensor, tfc.Tensor];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(x),\n                             rate: this.dropout,\n                             training,\n                             count: numOfKernels\n                           }) as tfc.Tensor[];\n      }\n\n      const dropoutMask = this.dropoutMask as DropoutMasks;\n\n      const applyDropout =\n          (x: tfc.Tensor, mask: tfc.Tensor[], index: number) => {\n            if (!mask || !mask[index]) {\n              return x;\n            }\n\n            return tfc.mul(mask[index], x);\n          };\n\n      let xI = applyDropout(x, dropoutMask, 0);\n      let xF = applyDropout(x, dropoutMask, 1);\n      let xC = applyDropout(x, dropoutMask, 2);\n      let xO = applyDropout(x, dropoutMask, 3);\n\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: numOfKernels\n                                    }) as tfc.Tensor[];\n      }\n\n      const recDropoutMask = this.recurrentDropoutMask as DropoutMasks;\n\n      let hI = applyDropout(hTMinus1, recDropoutMask, 0);\n      let hF = applyDropout(hTMinus1, recDropoutMask, 1);\n      let hC = applyDropout(hTMinus1, recDropoutMask, 2);\n      let hO = applyDropout(hTMinus1, recDropoutMask, 3);\n\n      const kernelChannelAxis = 3;\n\n      const [kernelI, kernelF, kernelC, kernelO]: tfc.Tensor[] =\n          tfc.split(this.kernel.read(), numOfKernels, kernelChannelAxis);\n\n      const [biasI, biasF, biasC, biasO]: tfc.Tensor[] = this.useBias ?\n          tfc.split(this.bias.read(), numOfKernels) :\n          [null, null, null, null];\n\n      xI = this.inputConv(xI, kernelI, biasI, this.padding);\n      xF = this.inputConv(xF, kernelF, biasF, this.padding);\n      xC = this.inputConv(xC, kernelC, biasC, this.padding);\n      xO = this.inputConv(xO, kernelO, biasO, this.padding);\n\n      const [recKernelI, recKernelF, recKernelC, recKernelO]: tfc.Tensor[] =\n          tfc.split(\n              this.recurrentKernel.read(), numOfKernels, kernelChannelAxis);\n\n      hI = this.recurrentConv(hI, recKernelI);\n      hF = this.recurrentConv(hF, recKernelF);\n      hC = this.recurrentConv(hC, recKernelC);\n      hO = this.recurrentConv(hO, recKernelO);\n\n      const i = this.recurrentActivation.apply(tfc.add(xI, hI));\n      const f = this.recurrentActivation.apply(tfc.add(xF, hF));\n      const c = tfc.add(\n          tfc.mul(f, cTMinus1),\n          tfc.mul(i, this.activation.apply(tfc.add(xC, hC))));\n      const h = tfc.mul(\n          this.recurrentActivation.apply(tfc.add(xO, hO)),\n          this.activation.apply(c));\n\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): tfc.serialization.ConfigDict {\n    const {'units': _, ...baseConfig} = super.getConfig();\n\n    const config: tfc.serialization.ConfigDict = {\n      filters: this.filters,\n      kernelSize: this.kernelSize,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      strides: this.strides,\n    };\n\n    return {...baseConfig, ...config};\n  }\n\n  inputConv(x: Tensor, w: Tensor, b?: Tensor, padding?: PaddingMode) {\n    const out = tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, this.strides as [number, number],\n        (padding || 'valid') as 'same' | 'valid',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC',\n        this.dilationRate as [number, number]);\n\n    if (b) {\n      return K.biasAdd(out, b, this.dataFormat) as tfc.Tensor3D;\n    }\n\n    return out;\n  }\n\n  recurrentConv(x: Tensor, w: Tensor) {\n    const strides = 1;\n\n    return tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, strides, 'same',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC');\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2DCell);\n\nexport declare interface ConvLSTM2DArgs extends\n    Omit<LSTMLayerArgs, 'units'|'cell'>, ConvRNN2DLayerArgs {}\n\nexport class ConvLSTM2D extends ConvRNN2D {\n  /** @nocollapse */\n  static className = 'ConvLSTM2D';\n\n  constructor(args: ConvLSTM2DArgs) {\n    const cell = new ConvLSTM2DCell(args);\n\n    super({...args, cell} as ConvRNN2DLayerArgs);\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends tfc.serialization.Serializable>(\n      cls: tfc.serialization.SerializableConstructor<T>,\n      config: tfc.serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Basic Layers.\n */\n\nimport {any, notEqual, serialization, Tensor, tidy, transpose, util} from '@tensorflow/tfjs-core';\n\nimport {Activation as ActivationFn, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {DisposeResult, InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {LayerConfig} from '../keras_format/topology_config';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {assertPositiveInteger, mapActivationToFusedKernel} from '../utils/generic_utils';\nimport {arrayProd, range} from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface DropoutLayerArgs extends LayerArgs {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /**\n   * Integer array representing the shape of the binary dropout mask that will\n   * be multiplied with the input.\n   *\n   * For instance, if your inputs have shape `(batchSize, timesteps, features)`\n   * and you want the dropout mask to be the same for all timesteps, you can use\n   * `noise_shape=(batch_size, 1, features)`.\n   */\n  noiseShape?: number[];\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class Dropout extends Layer {\n  /** @nocollapse */\n  static className = 'Dropout';\n  private readonly rate: number;\n  private readonly noiseShape: number[];\n  private readonly seed: number;\n\n  constructor(args: DropoutLayerArgs) {\n    super(args);\n    this.rate = Math.max(Math.min(args.rate, 1), 0);\n    // So that the scalar doesn't get tidied up between executions.\n    this.noiseShape = args.noiseShape;\n    this.seed = args.seed;\n    this.supportsMasking = true;\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    if (this.noiseShape == null) {\n      return this.noiseShape;\n    }\n    const inputShape = input.shape;\n    const noiseShape: Shape = [];\n    for (let i = 0; i < this.noiseShape.length; ++i) {\n      noiseShape.push(\n          this.noiseShape[i] == null ? inputShape[i] : this.noiseShape[i]);\n    }\n    return noiseShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      if (0 < this.rate && this.rate < 1) {\n        const training =\n            kwargs['training'] == null ? false : kwargs['training'];\n        const noiseShape = this.getNoiseShape(input);\n        const output = K.inTrainPhase(\n            () => K.dropout(input, this.rate, noiseShape, this.seed),\n            () => input, training);\n        return output;\n      }\n      return inputs;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      rate: this.rate,\n      noiseShape: this.noiseShape,\n      seed: this.seed,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  dispose(): DisposeResult {\n    return super.dispose();\n  }\n}\nserialization.registerClass(Dropout);\n\nexport declare interface DenseLayerArgs extends LayerArgs {\n  /** Positive integer, dimensionality of the output space. */\n  units: number;\n  /**\n   * Activation function to use.\n   *\n   * If unspecified, no activation is applied.\n   */\n  activation?: ActivationIdentifier;\n  /** Whether to apply a bias. */\n  useBias?: boolean;\n  /**\n   * Initializer for the dense kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * If specified, defines inputShape as `[inputDim]`.\n   */\n  inputDim?: number;\n\n  /**\n   * Constraint for the kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the dense kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport interface SpatialDropout1DLayerConfig extends LayerConfig {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class SpatialDropout1D extends Dropout {\n  /** @nocollapse */\n  static className = 'SpatialDropout1D';\n\n  constructor(args: SpatialDropout1DLayerConfig) {\n    super(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    const inputShape = input.shape;\n    return [inputShape[0], 1, inputShape[2]];\n  }\n}\nserialization.registerClass(SpatialDropout1D);\n\nexport class Dense extends Layer {\n  /** @nocollapse */\n  static className = 'Dense';\n  private units: number;\n  // Default activation: Linear (none).\n  private activation: ActivationFn = null;\n  private useBias = true;\n  private kernelInitializer: Initializer;\n  private biasInitializer: Initializer;\n  private kernel: LayerVariable = null;\n  private bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n  private readonly kernelConstraint?: Constraint;\n  private readonly biasConstraint?: Constraint;\n  private readonly kernelRegularizer?: Regularizer;\n  private readonly biasRegularizer?: Regularizer;\n\n  constructor(args: DenseLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null &&\n        args.inputDim != null) {\n      // This logic is copied from Layer's constructor, since we can't\n      // do exactly what the Python constructor does for Dense().\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      this.batchInputShape = [batchSize, args.inputDim];\n    }\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(args.activation);\n    if (args.useBias != null) {\n      this.useBias = args.useBias;\n    }\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.supportsMasking = true;\n\n    this.inputSpec = [{minNDim: 2}];\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputLastDim = inputShape[inputShape.length - 1];\n    if (this.kernel == null) {\n      this.kernel = this.addWeight(\n          'kernel', [inputLastDim, this.units], null, this.kernelInitializer,\n          this.kernelRegularizer, true, this.kernelConstraint);\n      if (this.useBias) {\n        this.bias = this.addWeight(\n            'bias', [this.units], null, this.biasInitializer,\n            this.biasRegularizer, true, this.biasConstraint);\n      }\n    }\n\n    this.inputSpec = [{minNDim: 2, axes: {[-1]: inputLastDim}}];\n    this.built = true;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    outputShape[outputShape.length - 1] = this.units;\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Dense layer accepts only a single input.\n      const input = getExactlyOneTensor(inputs);\n      const fusedActivationName =\n          mapActivationToFusedKernel(this.activation.getClassName());\n      let output: Tensor;\n\n      if (fusedActivationName != null) {\n        output = K.dot(\n            input, this.kernel.read(), fusedActivationName,\n            this.bias ? this.bias.read() : null);\n      } else {\n        output = K.dot(input, this.kernel.read());\n        if (this.bias != null) {\n          output = K.biasAdd(output, this.bias.read());\n        }\n        if (this.activation != null) {\n          output = this.activation.apply(output);\n        }\n      }\n\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dense);\n\nexport declare interface FlattenLayerArgs extends LayerArgs {\n  /** Image data format: channeLast (default) or channelFirst. */\n  dataFormat?: DataFormat;\n}\n\nexport class Flatten extends Layer {\n  private dataFormat: DataFormat;\n\n  /** @nocollapse */\n  static className = 'Flatten';\n  constructor(args?: FlattenLayerArgs) {\n    args = args || {};\n    super(args);\n    this.inputSpec = [{minNDim: 3}];\n    this.dataFormat = args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    for (const dim of inputShape.slice(1)) {\n      if (dim == null) {\n        throw new ValueError(\n            `The shape of the input to \"Flatten\" is not fully defined ` +\n            `(got ${inputShape.slice(1)}). Make sure to pass a complete ` +\n            `\"input_shape\" or \"batch_input_shape\" argument to the first ` +\n            `layer in your model.`);\n      }\n    }\n    return [inputShape[0], arrayProd(inputShape, 1)];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n\n      let input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsFirst' && input.rank > 1) {\n        const permutation: number[] = [0];\n        for (let i = 2; i < input.rank; ++i) {\n          permutation.push(i);\n        }\n        permutation.push(1);\n        input = input.transpose(permutation);\n      }\n\n      return K.batchFlatten(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {};\n    if (this.dataFormat != null) {\n      config['dataFormat'] = this.dataFormat;\n    }\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Flatten);\n\nexport declare interface ActivationLayerArgs extends LayerArgs {\n  /**\n   * Name of the activation function to use.\n   */\n  activation: ActivationIdentifier;\n}\n\nexport class Activation extends Layer {\n  /** @nocollapse */\n  static className = 'Activation';\n  activation: ActivationFn;\n\n  constructor(args: ActivationLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n    this.activation = getActivation(args.activation);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      return this.activation.apply(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {activation: serializeActivation(this.activation)};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Activation);\n\nexport declare interface ReshapeLayerArgs extends LayerArgs {\n  /** The target shape. Does not include the batch axis. */\n  targetShape: Shape;\n}\n\nexport declare interface RepeatVectorLayerArgs extends LayerArgs {\n  /**\n   * The integer number of times to repeat the input.\n   */\n  n: number;\n}\n\nexport class RepeatVector extends Layer {\n  /** @nocollapse */\n  static className = 'RepeatVector';\n  readonly n: number;\n\n  constructor(args: RepeatVectorLayerArgs) {\n    super(args);\n    this.n = args.n;\n    this.inputSpec = [{ndim: 2}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], this.n, inputShape[1]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      return K.repeat(inputs, this.n);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      n: this.n,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(RepeatVector);\n\nexport class Reshape extends Layer {\n  /** @nocollapse */\n  static className = 'Reshape';\n  private targetShape: Shape;\n\n  constructor(args: ReshapeLayerArgs) {\n    super(args);\n    this.targetShape = args.targetShape;\n\n    // Make sure that all unknown dimensions are represented as `null`.\n    for (let i = 0; i < this.targetShape.length; ++i) {\n      if (this.isUnknown(this.targetShape[i])) {\n        this.targetShape[i] = null;\n      }\n    }\n  }\n\n  private isUnknown(dim: number): boolean {\n    return dim < 0 || dim == null;\n  }\n\n  /**\n   * Finds and replaces a missing dimension in output shape.\n   *\n   * This is a near direct port of the internal Numpy function\n   * `_fix_unknown_dimension` in `numpy/core/src/multiarray/shape.c`.\n   *\n   * @param inputShape: Original shape of array begin reshape.\n   * @param outputShape: Target shape of the array, with at most a single\n   * `null` or negative number, which indicates an underdetermined dimension\n   * that should be derived from `inputShape` and the known dimensions of\n   *   `outputShape`.\n   * @returns: The output shape with `null` replaced with its computed value.\n   * @throws: ValueError: If `inputShape` and `outputShape` do not match.\n   */\n  private fixUnknownDimension(inputShape: Shape, outputShape: Shape): Shape {\n    const errorMsg = 'Total size of new array must be unchanged.';\n    const finalShape = outputShape.slice();\n    let known = 1;\n    let unknown = null;\n    for (let i = 0; i < finalShape.length; ++i) {\n      const dim = finalShape[i];\n      if (this.isUnknown(dim)) {\n        if (unknown === null) {\n          unknown = i;\n        } else {\n          throw new ValueError('Can only specifiy one unknown dimension.');\n        }\n      } else {\n        known *= dim;\n      }\n    }\n\n    const originalSize = arrayProd(inputShape);\n    if (unknown !== null) {\n      if (known === 0 || originalSize % known !== 0) {\n        throw new ValueError(errorMsg);\n      }\n      finalShape[unknown] = originalSize / known;\n    } else if (originalSize !== known) {\n      throw new ValueError(errorMsg);\n    }\n\n    return finalShape;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    let anyUnknownDims = false;\n    for (let i = 0; i < inputShape.length; ++i) {\n      if (this.isUnknown(inputShape[i])) {\n        anyUnknownDims = true;\n        break;\n      }\n    }\n\n    if (anyUnknownDims) {\n      return inputShape.slice(0, 1).concat(this.targetShape);\n    } else {\n      return inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const outputShape = inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n      return input.reshape(outputShape);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      targetShape: this.targetShape,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Reshape);\n\nexport declare interface PermuteLayerArgs extends LayerArgs {\n  /**\n   * Array of integers. Permutation pattern. Does not include the\n   * sample (batch) dimension. Index starts at 1.\n   * For instance, `[2, 1]` permutes the first and second dimensions\n   * of the input.\n   */\n  dims: number[];\n}\n\nexport class Permute extends Layer {\n  /** @nocollapse */\n  static className = 'Permute';\n  readonly dims: number[];\n  private readonly dimsIncludingBatch: number[];\n\n  constructor(args: PermuteLayerArgs) {\n    super(args);\n    if (args.dims == null) {\n      throw new Error(\n          'Required configuration field `dims` is missing during Permute ' +\n          'constructor call.');\n    }\n    if (!Array.isArray(args.dims)) {\n      throw new Error(\n          'Permute constructor requires `dims` to be an Array, but received ' +\n          `${args.dims} instead.`);\n    }\n\n    // Check the validity of the permutation indices.\n    const expectedSortedIndices = range(1, args.dims.length + 1);\n    if (!util.arraysEqual(args.dims.slice().sort(), expectedSortedIndices)) {\n      throw new Error(\n          'Invalid permutation `dims`: ' + JSON.stringify(args.dims) +\n          ' `dims` must contain consecutive integers starting from 1.');\n    }\n\n    this.dims = args.dims;\n    this.dimsIncludingBatch = [0].concat(this.dims);\n    this.inputSpec = [new InputSpec({ndim: this.dims.length + 1})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    this.dims.forEach((dim: number, i: number) => {\n      outputShape[i + 1] = (inputShape as Shape)[dim];\n    });\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return transpose(getExactlyOneTensor(inputs), this.dimsIncludingBatch);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      dims: this.dims,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Permute);\n\nexport declare interface MaskingArgs extends LayerArgs {\n  /**\n   * Masking Value. Defaults to `0.0`.\n   */\n  maskValue?: number;\n}\n\nexport class Masking extends Layer {\n  /** @nocollapse */\n  static className = 'Masking';\n  maskValue: number;\n\n  constructor(args?: MaskingArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maskValue = args.maskValue == null ? 0 : args.maskValue;\n    } else {\n      this.maskValue = 0;\n    }\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig() {\n    const baseConfig = super.getConfig();\n    const config = {maskValue: this.maskValue};\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    const input = getExactlyOneTensor(inputs);\n    const axis = -1;\n    return any(notEqual(input, this.maskValue), axis);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const axis = -1;\n      const keepDims = true;\n      const booleanMask = any(notEqual(input, this.maskValue), axis, keepDims);\n      const output = input.mul(booleanMask.asType(input.dtype));\n      return output;\n    });\n  }\n}\nserialization.registerClass(Masking);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Embedding Layer.\n *\n * Original source: keras/constraints.py\n */\nimport {notEqual, serialization, Tensor, tidy, zerosLike} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface EmbeddingLayerArgs extends LayerArgs {\n  /**\n   * Integer > 0. Size of the vocabulary, i.e. maximum integer index + 1.\n   */\n  inputDim: number;\n  /**\n   * Integer >= 0. Dimension of the dense embedding.\n   */\n  outputDim: number;\n  /**\n   * Initializer for the `embeddings` matrix.\n   */\n  embeddingsInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Regularizer function applied to the `embeddings` matrix.\n   */\n  embeddingsRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Constraint function applied to the `embeddings` matrix.\n   */\n  embeddingsConstraint?: ConstraintIdentifier|Constraint;\n  /**\n   * Whether the input value 0 is a special \"padding\" value that should be\n   * masked out. This is useful when using recurrent layers which may take\n   * variable length input.\n   *\n   * If this is `True` then all subsequent layers in the model need to support\n   * masking or an exception will be raised. If maskZero is set to `True`, as a\n   * consequence, index 0 cannot be used in the vocabulary (inputDim should\n   * equal size of vocabulary + 1).\n   */\n  maskZero?: boolean;\n  /**\n   * Length of input sequences, when it is constant.\n   *\n   * This argument is required if you are going to connect `flatten` then\n   * `dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed).\n   */\n  inputLength?: number|number[];\n}\n\nexport class Embedding extends Layer {\n  /** @nocollapse */\n  static className = 'Embedding';\n  private inputDim: number;\n  private outputDim: number;\n  private embeddingsInitializer: Initializer;\n  private maskZero: boolean;\n  private inputLength: number|number[];\n\n  private embeddings: LayerVariable = null;\n\n  readonly DEFAULT_EMBEDDINGS_INITIALIZER: InitializerIdentifier =\n      'randomUniform';\n  private readonly embeddingsRegularizer?: Regularizer;\n  private readonly embeddingsConstraint?: Constraint;\n\n  constructor(args: EmbeddingLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null) {\n      // Porting Note: This logic is copied from Layer's constructor, since we\n      // can't do exactly what the Python constructor does for Embedding().\n      // Specifically, the super constructor can not be called after the\n      // mutation of the `config` argument.\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      if (args.inputLength == null) {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (None, )\n        this.batchInputShape = [batchSize, null];\n      } else {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (config.inputLength, )\n        this.batchInputShape =\n            [batchSize].concat(generic_utils.toList(args.inputLength));\n      }\n    }\n    this.inputDim = args.inputDim;\n    generic_utils.assertPositiveInteger(this.inputDim, 'inputDim');\n    this.outputDim = args.outputDim;\n    generic_utils.assertPositiveInteger(this.outputDim, 'outputDim');\n    this.embeddingsInitializer = getInitializer(\n        args.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER);\n    this.embeddingsRegularizer = getRegularizer(args.embeddingsRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.embeddingsConstraint = getConstraint(args.embeddingsConstraint);\n    this.maskZero = args.maskZero;\n    this.supportsMasking = args.maskZero;\n    this.inputLength = args.inputLength;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    this.embeddings = this.addWeight(\n        'embeddings', [this.inputDim, this.outputDim], this.dtype,\n        this.embeddingsInitializer, this.embeddingsRegularizer, true,\n        this.embeddingsConstraint);\n    this.built = true;\n  }\n\n  // Override warnOnIncompatibleInputShape because an embedding layer allows\n  // the input to have varying ranks.\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {}\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tidy(() => {\n      if (!this.maskZero) {\n        return null;\n      } else {\n        inputs = getExactlyOneTensor(inputs);\n        return notEqual(inputs, zerosLike(inputs));\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    if (this.inputLength == null) {\n      return [...inputShape, this.outputDim];\n    }\n    // inputLength can be an array if input is 3D or higher.\n    const inLens: number[] = generic_utils.toList(this.inputLength);\n    if (inLens.length !== inputShape.length - 1) {\n      throw new ValueError(\n          `\"inputLength\" is ${this.inputLength}, but received ` +\n          `input shape has shape ${inputShape}`);\n    } else {\n      let i = 0;\n      for (let k = 0; k < inLens.length; ++k) {\n        const s1 = inLens[k];\n        const s2 = inputShape[k + 1];\n        if ((s1 != null) && (s2 != null) && (s1 !== s2)) {\n          throw new ValueError(\n              `\"inputLength\" is ${this.inputLength}, but received ` +\n              `input shape has shape ${inputShape}`);\n        } else if (s1 == null) {\n          inLens[i] = s2;\n        }\n        i++;\n      }\n    }\n    return [inputShape[0], ...inLens, this.outputDim];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Embedding layer accepts only a single input.\n      let input = getExactlyOneTensor(inputs);\n      if (input.dtype !== 'int32') {\n        input = K.cast(input, 'int32');\n      }\n      const output = K.gather(this.embeddings.read(), input.as1D());\n      return output.reshape(\n          getExactlyOneShape(this.computeOutputShape(input.shape)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      inputDim: this.inputDim,\n      outputDim: this.outputDim,\n      embeddingsInitializer: serializeInitializer(this.embeddingsInitializer),\n      embeddingsRegularizer: serializeRegularizer(this.embeddingsRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      embeddingsConstraint: serializeConstraint(this.embeddingsConstraint),\n      maskZero: this.maskZero,\n      inputLength: this.inputLength\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Embedding);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Merge Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {l2Normalize} from '../losses';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as mathUtils from '../utils/math_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\n/**\n * Generic Merge layer for element-wise merge functions.\n *\n * Used to implement `Sum`, `Average`, `Concatenate`, etc.\n */\nexport abstract class Merge extends Layer {\n  protected reshapeRequired: boolean;\n\n  constructor(args?: LayerArgs) {\n    super(args || {});\n    this.supportsMasking = true;\n  }\n\n  /**\n   * Logic for merging multiple tensors, to be overridden by subclasses.\n   * @param inputs\n   */\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    throw new NotImplementedError();\n  }\n\n  /**\n   * Computes the shape of the result of an elementwise operation.\n   *\n   * @param shape1: Shape of the first tensor.\n   * @param shape2: Shape of the second tensor.\n   * @returns Expected output shape when an elementwise operation is carried\n   *   out on 2 tensors with shapes `shape1` and `shape2`.\n   * @throws ValueError: If `shape1` and `shape2` are not compatible for\n   *   element-wise operations.\n   */\n  private computeElementwiseOpOutputShape(shape1: Shape, shape2: Shape): Shape {\n    if (shape1 == null || shape2 == null) {\n      return null;\n    } else if (shape1.length < shape2.length) {\n      return this.computeElementwiseOpOutputShape(shape2, shape1);\n    } else if (shape2.length === 0) {\n      return shape1;\n    }\n    const outputShape: Shape = shape1.slice(0, shape1.length - shape2.length);\n    for (let k = 0; k < shape2.length; ++k) {\n      const i = shape1[shape1.length - shape2.length + k];\n      const j = shape2[k];\n      if (i == null || j == null || i < 0 || j < 0) {\n        outputShape.push(null);\n      } else if (i === 1) {\n        outputShape.push(j);\n      } else if (j === 1) {\n        outputShape.push(i);\n      } else {\n        if (i !== j) {\n          throw new ValueError(\n              'Operands could not be broadcast together with shapes ' +\n              JSON.stringify(shape1) + ' ' + JSON.stringify(shape2));\n        }\n        outputShape.push(i);\n      }\n    }\n    return outputShape;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.\n    if (Array.isArray(inputShape) && !Array.isArray(inputShape[0])) {\n      // Make sure that inputShape is an Array of shape.\n      inputShape = [getExactlyOneShape(inputShape)];\n    }\n    inputShape = inputShape as Shape[];\n    if (inputShape.length < 2) {\n      throw new ValueError(\n          'A merge layer should be called on an Array of at least 2 inputs.' +\n          ` Got ${inputShape.length} input(s).`);\n    }\n\n    // Make sure that there is at most one unique batch size among the input\n    // shapes.\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length > 1) {\n      throw new ValueError(\n          `Can not merge tensors with different batch sizes. ` +\n          `Got tensors with shapes: ${JSON.stringify(inputShape)}.`);\n    }\n\n    let outputShape: Shape =\n        inputShape[0] == null ? null : inputShape[0].slice(1);\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n    // If the inputs have different ranks, we have to reshape them to make them\n    // broadcastable.\n    const allRanks = inputShape.map(shape => shape.length);\n    if (inputShape.indexOf(null) === -1 &&\n        generic_utils.unique(allRanks).length === 1) {\n      this.reshapeRequired = false;\n    } else {\n      this.reshapeRequired = true;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (this.reshapeRequired) {\n        const reshapedInputs: Tensor[] = [];\n        const inputDims = inputs.map(input => input.rank);\n        if (inputDims.indexOf(null) === -1) {\n          // If ranks of all inputs are available, we simply expand each of them\n          // at axis=1 until all of them have the same rank.\n          const maxNDim = mathUtils.max(inputDims);\n          for (let x of inputs) {\n            const xNDim = x.rank;\n            for (let k = 0; k < maxNDim - xNDim; ++k) {\n              x = K.expandDims(x, 1);\n            }\n            reshapedInputs.push(x);\n          }\n          return this.mergeFunction(reshapedInputs);\n        } else {\n          // Transpose all inputs so that batch size is the last dimension.\n          // [batchSize, dim1, dim2, ...] -> [dim1, dim2, ..., batchSize]\n          let transposed = false;\n          for (const x of inputs) {\n            const xNDim = x.rank;\n            if (xNDim == null) {\n              const xShape = x.shape;\n              const batchSize = xShape[0];\n              const newShape = xShape.slice(1).concat([batchSize]);\n              let xTransposed = x.reshape(\n                  [batchSize].concat(mathUtils.arrayProd(xShape.slice(1))));\n              xTransposed = tfc.transpose(xTransposed, [1, 0]);\n              xTransposed = xTransposed.reshape(newShape);\n              reshapedInputs.push(xTransposed);\n              transposed = true;\n            } else if (xNDim > 1) {\n              const dims = mathUtils.range(1, xNDim).concat([0]);\n              reshapedInputs.push(tfc.transpose(x, dims));\n              transposed = true;\n            } else {\n              // We don't transpose inputs if they are 1D vectors or scalars.\n              reshapedInputs.push(x);\n            }\n          }\n          let y = this.mergeFunction(reshapedInputs);\n          const yNDim = y.rank;\n          if (transposed) {\n            // If inputs have been transposed, we have to transpose the output\n            // too.\n            if (yNDim == null) {\n              const yShape = y.shape;\n              const yNDim = yShape.length;\n              const batchSize = yShape[yNDim - 1];\n              const newShape =\n                  [batchSize].concat(yShape.slice(0, yShape.length - 1));\n              y = tfc.transpose(y.reshape([-1, batchSize]), [1, 0])\n                      .reshape(newShape);\n            } else if (yNDim > 1) {\n              const dims = [yNDim - 1].concat(mathUtils.range(0, yNDim - 1));\n              y = tfc.transpose(y, dims);\n            }\n          }\n          return y;\n        }\n      } else {\n        return this.mergeFunction(inputs);\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape[];\n    let outputShape: Shape;\n    if (inputShape[0] == null) {\n      outputShape = null;\n    } else {\n      outputShape = inputShape[0].slice(1);\n    }\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length === 1) {\n      outputShape = batchSizes.concat(outputShape);\n    } else {\n      outputShape = [null].concat(outputShape);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tfc.tidy(() => {\n      if (mask == null) {\n        return null;\n      }\n      if (!Array.isArray(mask)) {\n        throw new ValueError('`mask` should be an Array');\n      }\n      if (!Array.isArray(inputs)) {\n        throw new ValueError('`inputs` should be an Array');\n      }\n      if (mask.length !== inputs.length) {\n        throw new ValueError(\n            `The Array 'inputs' and 'mask' are expected to have the same ` +\n            `length, but have different lengths ` +\n            `(${inputs.length} vs ${mask.length})`);\n      }\n      if (mask.every(m => m == null)) {\n        return null;\n      }\n      mask = mask.map(m => m == null ? m : tfc.expandDims(m, 0));\n      let output = mask[0];\n      for (let i = 1; i < mask.length - 1; ++i) {\n        output = tfc.logicalAnd(output, mask[i]);\n      }\n      return output;\n    });\n  }\n}\n\nexport class Add extends Merge {\n  /** @nocollapse */\n  static className = 'Add';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Add);\n\n/**\n * Calculate the element-wise sum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Add` layer, by using no input argument\n *    or a single configuration argument. The resultant `Add` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const addLayer = tf.layers.add();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = addLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.add([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.add([input1, input2]).print();\n * // Gives [[11, 22], [33, 44]].\n *\n */\nexport function add(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Add({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Add(config);\n  }\n}\n\nexport class Multiply extends Merge {\n  /** @nocollapse */\n  static className = 'Multiply';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.mul(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Multiply);\n\n/**\n * Calculate the element-wise product of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Multiply` layer, by using no input argument\n *    or a single configuration argument. The resultant `Multiply` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const multiplyLayer = tf.layers.multiply();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = multiplyLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.multiply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.multiply([input1, input2]).print();\n * // Gives [[10, 40], [90, 160]].\n *\n */\nexport function multiply(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Multiply({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Multiply(config);\n  }\n}\n\nexport class Average extends Merge {\n  /** @nocollapse */\n  static className = 'Average';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return tfc.mul(1 / inputs.length, output);\n    });\n  }\n}\nserialization.registerClass(Average);\n\n/**\n * Calculate the element-wise arithmetic mean of inputs, which all have the same\n * shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Average` layer, by using no input argument\n *    or a single configuration argument. The resultant `Average` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const averageLayer = tf.layers.average();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = averageLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.average([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.average([input1, input2]).print();\n * // Gives [[5.5, 11], [16.5, 22]].\n *\n */\nexport function average(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Average({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Average(config);\n  }\n}\n\nexport class Maximum extends Merge {\n  /** @nocollapse */\n  static className = 'Maximum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.maximum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Maximum);\n\n/**\n * Calculate the element-wise maximum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Maximum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Maximum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const maximumLayer = tf.layers.maximum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = maximumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.maximum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.maximum([input1, input2]).print();\n * // Gives [[10, 20], [30, 40]].\n *\n */\nexport function maximum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Maximum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Maximum(config);\n  }\n}\n\nexport class Minimum extends Merge {\n  /** @nocollapse */\n  static className = 'Minimum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.minimum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Minimum);\n\n/**\n * Calculate the element-wise minimum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Minimum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Minimum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const minimumLayer = tf.layers.minimum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = minimumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.minimum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.minimum([input1, input2]).print();\n * // Gives [[1, 2], [3, 4]].\n *\n */\nexport function minimum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Minimum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Minimum(config);\n  }\n}\n\nexport declare interface ConcatenateLayerArgs extends LayerArgs {\n  /**\n   * Axis along which to concatenate.\n   */\n  axis?: number;\n}\n\nexport class Concatenate extends Merge {\n  /** @nocollapse */\n  static className = 'Concatenate';\n  readonly DEFAULT_AXIS = -1;\n  private readonly axis: number;\n\n  constructor(args?: ConcatenateLayerArgs) {\n    super(args);\n    if (args == null) {\n      args = {};\n    }\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.]\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0])) ||\n        inputShape.length === 1) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of at least 2 ' +\n          'inputs');\n    }\n    inputShape = inputShape as Shape[];\n\n    let allNoneShape = true;\n    for (const shape of inputShape) {\n      if (shape != null) {\n        allNoneShape = false;\n        break;\n      }\n    }\n    if (allNoneShape) {\n      return;\n    }\n\n    const shapeSet: Shape[] = [];\n    for (let i = 0; i < inputShape.length; ++i) {\n      const shapeWithoutConcatAxis = inputShape[i].slice();\n      shapeWithoutConcatAxis.splice(this.axis, 1);\n      let exists = false;\n      for (const shape of shapeSet) {\n        if (util.arraysEqual(shape, shapeWithoutConcatAxis)) {\n          exists = true;\n          break;\n        }\n      }\n      if (!exists) {\n        shapeSet.push(shapeWithoutConcatAxis);\n      }\n    }\n    if (shapeSet.length > 1) {\n      throw new ValueError(\n          'A `Concatenate` layer requires inputs with matching shapes ' +\n          'except for the concat axis. Got input shapes: ' +\n          JSON.stringify(inputShape));\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      return K.concatenate(inputs, this.axis);\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0]))) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of inputs.');\n    }\n    const inputShapes = inputShape as Shape[];\n    const outputShape = inputShapes[0].slice();\n    const axis = this.axis < 0 ? outputShape.length + this.axis : this.axis;\n    // Porting Note: the line above is because TypeScript doesn't support\n    //   negative indices.\n    for (const shape of inputShapes.slice(1)) {\n      if (outputShape[axis] == null || shape[axis] == null) {\n        outputShape[axis] = null;\n        break;\n      }\n      outputShape[axis] += shape[axis];\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    if (mask == null) {\n      return null;\n    }\n    if (!Array.isArray(mask)) {\n      throw new ValueError('`mask` should be an array for Concatenate');\n    }\n    if (!Array.isArray(inputs)) {\n      throw new ValueError('`inputs` should be an array for Concatenate');\n    }\n    if (mask.length !== inputs.length) {\n      throw new ValueError(\n          `Mismatch in the length of mask (${mask.length}) ` +\n          `and the legnth of inputs (${inputs.length})`);\n    }\n    return tfc.tidy(() => {\n      let allNullMasks = true;\n      mask.forEach(m => {\n        if (m != null) {\n          allNullMasks = false;\n          return;\n        }\n      });\n      if (allNullMasks) {\n        return null;\n      }\n      const outputMasks: Tensor[] = [];\n      for (let i = 0; i < inputs.length; ++i) {\n        if (mask[i] == null) {\n          // Input is unmasked. Append all 1's to masks.\n          outputMasks.push(tfc.onesLike(inputs[i]).asType('bool'));\n        } else if (mask[i].rank < inputs[i].rank) {\n          // Mask is smaller than the input, expand it.\n          outputMasks.push(tfc.expandDims(mask[i], -1));\n        } else {\n          outputMasks.push(mask[i]);\n        }\n      }\n      const concatenatedMasks = tfc.concat(outputMasks, this.axis);\n      return tfc.all(concatenatedMasks, -1, false);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axis': this.axis,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Concatenate);\n\n/**\n * Concatenate an `Array` of inputs.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Concatenate` layer, by using no input argument\n *    or a single configuration argument. The resultant `Concatenate` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const concatLayer = tf.layers.concatenate();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = concatLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 7], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = tf.layers.concatenate([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([[1, 2], [3, 4]], [2, 2]);\n * const input2 = tf.tensor2d([[10, 20], [30, 40]], [2, 2]);\n * tf.layers.concatenate([input1, input2]).print();\n * // Gives [[1, 2, 10, 20], [3, 4, 30, 40]].\n *\n */\nexport function concatenate(config?: SymbolicTensor[]|Tensor[]|\n                            ConcatenateLayerArgs): Layer|SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Concatenate({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Concatenate(config);\n  }\n}\n\nexport declare interface DotLayerArgs extends LayerArgs {\n  /**\n   * Axis or axes along which the dot product will be taken.\n   *\n   * Integer or an Array of integers.\n   */\n  axes: number|[number, number];\n\n  /**\n   * Whether to L2-normalize samples along the dot product axis\n   * before taking the dot product.\n   *\n   * If set to `true`, the output of the dot product isthe cosine\n   * proximity between the two samples.\n   */\n  normalize?: boolean;\n}\n\n/**\n * Interpretable potentially negative axis index.\n *\n * For example, given axis = -1, and dim = 3, this function will return 2.\n *\n * @param axis The axis index, may be a positive, zero or negative integer.\n * @param dim Total number of dimensions, a positive integer.\n * @returns A non-negative axis index equivalent to the input `axis`.\n */\nfunction interpretAxis(axis: number, dim: number): number {\n  while (axis < 0) {\n    axis += dim;\n  }\n  return axis;\n}\n\nfunction batchDot(x: Tensor, y: Tensor, axes: number|[number, number]): Tensor {\n  if (x.shape.length > 3 || y.shape.length > 3) {\n    throw new NotImplementedError(\n        'batchDot is not implemented for tensors of 4D or higher rank yet');\n  }\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of x to be >= 2, ` +\n          `but got ${x.shape.length}`);\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of y to be >= 2, ` +\n          `but got ${y.shape.length}`);\n\n  if (typeof axes === 'number') {\n    axes = [axes, axes];\n  }\n\n  if (x.dtype === 'complex64' || y.dtype === 'complex64') {\n    throw new NotImplementedError(\n        'batchDot is not implemented for complex64-type Tensors yet.');\n  }\n\n  const xNDim = x.shape.length;\n  const yNDim = y.shape.length;\n  if (axes == null) {\n    // Behave like batchMatmul by default.\n    axes = [xNDim - 1, yNDim - 2];\n  }\n  const axesArray = axes as [number, number];\n\n  return tfc.tidy(() => {\n    let diff: number;\n    if (xNDim > yNDim) {\n      diff = xNDim - yNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      y = y.reshape(y.shape.concat(diffShape));\n    } else if (yNDim > xNDim) {\n      diff = yNDim - xNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      x = x.reshape(x.shape.concat(diffShape));\n    } else {\n      diff = 0;\n    }\n\n    let out: Tensor;\n    if (x.shape.length === 2 && y.shape.length === 2) {\n      if (axesArray[0] === axesArray[1]) {\n        out = x.mul(y).sum(axesArray[0]);\n      } else {\n        out = x.transpose([1, 0]).mul(y).sum(axesArray[1]);\n      }\n    } else {\n      const adjX = axesArray[0] !== x.shape.length - 1;\n      const adjY = axesArray[1] === y.shape.length - 1;\n      out = x.matMul(y, adjX, adjY);\n    }\n\n    if (diff > 0) {\n      let idx: number;\n      if (xNDim > yNDim) {\n        idx = xNDim + yNDim - 3;\n      } else {\n        idx = xNDim - 1;\n      }\n      const squeezeAxes: number[] = [];\n      for (let i = idx; i < idx + diff; ++i) {\n        squeezeAxes.push(i);\n      }\n      out = out.squeeze(squeezeAxes);\n    }\n    if (out.shape.length === 1) {\n      out = out.expandDims(1);\n    }\n    return out;\n  });\n}\n\nexport class Dot extends Merge {\n  /** @nocollapse */\n  static className = 'Dot';\n\n  private axes: number|[number, number];\n  private normalize: boolean;\n\n  constructor(args: DotLayerArgs) {\n    super(args);\n    this.axes = args.axes;\n    this.normalize = args.normalize == null ? false : args.normalize;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = inputShape[0] as Shape;\n    const shape2 = inputShape[1] as Shape;\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    if (shape1[axes[0]] !== shape2[axes[1]]) {\n      throw new ValueError(\n          `Dimension incompatibility: ` +\n          `${shape1[axes[0]]} !== ${shape2[axes[1]]}`);\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    if (inputs.length !== 2) {\n      throw new ValueError(\n          'A `Dot` layer must be called on exactly 2 inputs, ' +\n          `but received ${inputs.length} input(s).`);\n    }\n\n    let x1 = inputs[0];\n    let x2 = inputs[1];\n    let axes: [number, number];\n    if (!Array.isArray(this.axes)) {\n      axes = [\n        interpretAxis(this.axes, x1.shape.length),\n        interpretAxis(this.axes, x2.shape.length)\n      ];\n    } else {\n      axes = this.axes.map(\n                 (axis, i) => interpretAxis(\n                     axis, inputs[i].shape.length)) as [number, number];\n    }\n    if (this.normalize) {\n      x1 = l2Normalize(x1, axes[0]);\n      x2 = l2Normalize(x2, axes[1]);\n    }\n    return batchDot(x1, x2, axes);\n  }\n\n  private interpretAxes(shape1: Shape, shape2: Shape): number[] {\n    let axes: number[];\n    if (!Array.isArray(this.axes)) {\n      // `this.axes` is a single integer.\n      axes = [\n        interpretAxis(this.axes, shape1.length),\n        interpretAxis(this.axes, shape2.length)\n      ];\n    } else {\n      // `this.axes` is an Array of integers.\n      axes = this.axes;\n    }\n    return axes;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = (inputShape[0] as Shape).slice();\n    const shape2 = (inputShape[1] as Shape).slice();\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    shape1.splice(axes[0], 1);\n    shape2.splice(axes[1], 1);\n    shape2.splice(0, 1);\n    const outputShape = shape1.concat(shape2);\n    if (outputShape.length === 1) {\n      outputShape.push(1);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return null;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axes': this.axes,\n      'normalize': this.normalize\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dot);\n\n// TODO(cais): Add functional interfaces for the merge layers.\n", "/**\r\n * @license\r\n * Copyright 2018 Google LLC\r\n *\r\n * Use of this source code is governed by an MIT-style\r\n * license that can be found in the LICENSE file or at\r\n * https://opensource.org/licenses/MIT.\r\n * =============================================================================\r\n */\r\n\r\n/**\r\n * TensorFlow.js Layers: Noise Layers.\r\n */\r\n\r\nimport {greaterEqual, randomUniform, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\r\n\r\nimport * as K from '../backend/tfjs_backend';\r\nimport {Layer, LayerArgs} from '../engine/topology';\r\nimport {Shape} from '../keras_format/common';\r\nimport {Kwargs} from '../types';\r\nimport {getExactlyOneTensor} from '../utils/types_utils';\r\n\r\nexport declare interface GaussianNoiseArgs extends LayerArgs {\r\n  /** Standard Deviation.  */\r\n  stddev: number;\r\n}\r\n\r\nexport class GaussianNoise extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianNoise';\r\n  readonly stddev: number;\r\n\r\n  constructor(args: GaussianNoiseArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.stddev = args.stddev;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {stddev: this.stddev};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      const noised = () =>\r\n          K.randomNormal(input.shape, 0, this.stddev).add(input);\r\n      const output =\r\n          K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      return output;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianNoise);\r\n\r\nexport declare interface GaussianDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n}\r\n\r\nexport class GaussianDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianDropout';\r\n  readonly rate: number;\r\n\r\n  constructor(args: GaussianDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      if (this.rate > 0 && this.rate < 1) {\r\n        const noised = () => {\r\n          const stddev = Math.sqrt(this.rate / (1 - this.rate));\r\n          return input.mul(K.randomNormal(input.shape, 1, stddev));\r\n        };\r\n        return K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      }\r\n      return input;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianDropout);\r\n\r\nexport declare interface AlphaDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n  /**\r\n   * A 1-D `Tensor` of type `int32`, representing the\r\n   * shape for randomly generated keep/drop flags.\r\n   */\r\n  noiseShape?: Shape;\r\n}\r\n\r\n/**\r\n * Applies Alpha Dropout to the input.\r\n *\r\n * As it is a regularization layer, it is only active at training time.\r\n *\r\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\r\n * to their original values, in order to ensure the self-normalizing property\r\n * even after this dropout.\r\n * Alpha Dropout fits well to Scaled Exponential Linear Units\r\n * by randomly setting activations to the negative saturation value.\r\n *\r\n * Arguments:\r\n *   - `rate`: float, drop probability (as with `Dropout`).\r\n *     The multiplicative noise will have\r\n *     standard deviation `sqrt(rate / (1 - rate))`.\r\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\r\n *     shape for randomly generated keep/drop flags.\r\n *\r\n * Input shape:\r\n *   Arbitrary. Use the keyword argument `inputShape`\r\n *   (tuple of integers, does not include the samples axis)\r\n *   when using this layer as the first layer in a model.\r\n *\r\n * Output shape:\r\n *   Same shape as input.\r\n *\r\n * References:\r\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\r\n */\r\nexport class AlphaDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'AlphaDropout';\r\n  readonly rate: number;\r\n  readonly noiseShape: Shape;\r\n\r\n  constructor(args: AlphaDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n    this.noiseShape = args.noiseShape;\r\n  }\r\n\r\n  _getNoiseShape(inputs: Tensor|Tensor[]) {\r\n    return this.noiseShape || getExactlyOneTensor(inputs).shape;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      if (this.rate < 1 && this.rate > 0) {\r\n        const noiseShape = this._getNoiseShape(inputs);\r\n\r\n        const droppedInputs = () => {\r\n          const input = getExactlyOneTensor(inputs);\r\n\r\n          const alpha = 1.6732632423543772848170429916717;\r\n          const scale = 1.0507009873554804934193349852946;\r\n\r\n          const alphaP = -alpha * scale;\r\n\r\n          let keptIdx = greaterEqual(randomUniform(noiseShape), this.rate);\r\n\r\n          keptIdx = K.cast(keptIdx, 'float32');  // get default dtype.\r\n\r\n          // Get affine transformation params.\r\n          const a = ((1 - this.rate) * (1 + this.rate * alphaP ** 2)) ** -0.5;\r\n          const b = -a * alphaP * this.rate;\r\n\r\n          // Apply mask.\r\n          const x = input.mul(keptIdx).add(keptIdx.add(-1).mul(alphaP));\r\n\r\n          return x.mul(a).add(b);\r\n        };\r\n        return K.inTrainPhase(\r\n            droppedInputs, () => getExactlyOneTensor(inputs),\r\n            kwargs['training'] || false);\r\n      }\r\n      return inputs;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(AlphaDropout);\r\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Normalization layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {moments, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Applies batch normalization on x given mean, var, beta and gamma.\n *\n * I.e. returns:\n *   `output = (x - mean) / (sqrt(var) + epsilon) * gamma + beta`\n *\n * @param x Input tensor.\n * @param mean Mean of batch.\n * @param variance Variance of batch.\n * @param beta Tensor with which to center the input.\n * @param gamma Tensor by which to scale the input.\n * @param epsilon Fuzz factor.\n * @returns The result of the batch normalization.\n */\nexport function batchNormalization(\n    x: Tensor, mean: Tensor, variance: Tensor, beta?: Tensor, gamma?: Tensor,\n    epsilon = 1e-3): Tensor {\n  let out: Tensor;\n  if (x.rank === 2) {\n    out = tfc.batchNorm2d(\n        x as Tensor2D, mean as Tensor2D | Tensor1D,\n        variance as Tensor2D | Tensor1D, beta as Tensor2D | Tensor1D,\n        gamma as Tensor2D | Tensor1D, epsilon);\n  } else if (x.rank === 3) {\n    // TODO(cais): Check rank; give proper error message.\n    out = tfc.batchNorm3d(\n        x as Tensor3D, mean as Tensor3D | Tensor1D,\n        variance as Tensor3D | Tensor1D, beta as Tensor3D | Tensor1D,\n        gamma as Tensor3D | Tensor1D, epsilon);\n  } else if (x.rank === 4) {\n    out = tfc.batchNorm4d(\n        x as Tensor4D, mean as Tensor4D | Tensor1D,\n        variance as Tensor4D | Tensor1D, beta as Tensor4D | Tensor1D,\n        gamma as Tensor4D | Tensor1D, epsilon);\n  } else {\n    throw new NotImplementedError(\n        `batchNormalization is not implemented for array of rank ${x.rank} ` +\n        `yet`);\n  }\n  return out;\n}\n\n/**\n * Non-broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction regularNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const normed =\n               batchNormalization(x, mean, variance, beta, gamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction broadcastNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const targetShape: number[] = [];\n           for (const axis of math_utils.range(0, x.rank)) {\n             if (reductionAxes.indexOf(axis) !== -1) {\n               targetShape.push(1);\n             } else {\n               targetShape.push(x.shape[axis]);\n             }\n           }\n           const broadcastMean = mean.reshape(targetShape);\n           const broadcastVariance = variance.reshape(targetShape);\n           const broadcastGamma =\n               gamma == null ? null : gamma.reshape(targetShape);\n           const broadcastBeta =\n               beta == null ? null : beta.reshape(targetShape);\n           const normed = batchNormalization(\n               x, broadcastMean, broadcastVariance, broadcastBeta,\n               broadcastGamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Batch normalization for use in training (not inference).\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nexport function normalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  if (util.arraysEqual(\n          reductionAxes.slice().sort(), math_utils.range(0, x.rank - 1))) {\n    return regularNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  } else {\n    return broadcastNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  }\n}\n\nexport declare interface BatchNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The integer axis that should be normalized (typically the features axis).\n   * Defaults to -1.\n   *\n   * For instance, after a `Conv2D` layer with `data_format=\"channels_first\"`,\n   * set `axis=1` in `batchNormalization`.\n   */\n  axis?: number;\n\n  /**\n   * Momentum of the moving average. Defaults to 0.99.\n   */\n  momentum?: number;\n\n  /**\n   * Small float added to the variance to avoid dividing by zero. Defaults to\n   * 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Defaults to `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear (also e.g. `nn.relu`),\n   * this can be disabled since the scaling will be done by the next layer.\n   * Defaults to `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   *  Defaults to 'zeros'.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   *  Defaults to `ones`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving mean.\n   * Defaults to `zeros`\n   */\n  movingMeanInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving variance.\n   *  Defaults to 'Ones'.\n   */\n  movingVarianceInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the beta weight.\n   */\n  betaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for gamma weight.\n   */\n  gammaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer for the beta weight.\n   */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer for the gamma weight.\n   */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class BatchNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'BatchNormalization';\n  private readonly axis: number;\n  private readonly momentum: number;\n  private readonly epsilon: number;\n  private readonly center: boolean;\n  private readonly scale: boolean;\n  private readonly betaInitializer: Initializer;\n  private readonly gammaInitializer: Initializer;\n  private readonly movingMeanInitializer: Initializer;\n  private readonly movingVarianceInitializer: Initializer;\n  private readonly betaConstraint: Constraint;\n  private readonly gammaConstraint: Constraint;\n  private readonly betaRegularizer: Regularizer;\n  private readonly gammaRegularizer: Regularizer;\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n  private movingMean: LayerVariable;\n  private movingVariance: LayerVariable;\n\n  constructor(args?: BatchNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.supportsMasking = true;\n    this.axis = args.axis == null ? -1 : args.axis;\n    this.momentum = args.momentum == null ? 0.99 : args.momentum;\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.movingMeanInitializer =\n        getInitializer(args.movingMeanInitializer || 'zeros');\n    this.movingVarianceInitializer =\n        getInitializer(args.movingVarianceInitializer || 'ones');\n    this.betaConstraint = getConstraint(args.betaConstraint);\n    this.gammaConstraint = getConstraint(args.gammaConstraint);\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const axis = this.axis >= 0 ? this.axis : (this.axis + inputShape.length);\n    const dim = inputShape[axis];\n    if (dim == null) {\n      throw new ValueError(\n          `Axis ${axis} of input tensor should have a defined dimension but ` +\n          `the layer received an input with shape ` +\n          `${JSON.stringify(inputShape)}.`);\n    }\n    this.inputSpec =\n        [new InputSpec({ndim: inputShape.length, axes: {[axis]: dim}})];\n    const shape = [dim];\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', shape, null, this.gammaInitializer, this.gammaRegularizer,\n          true, this.gammaConstraint);\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', shape, null, this.betaInitializer, this.betaRegularizer, true,\n          this.betaConstraint);\n    }\n    this.movingMean = this.addWeight(\n        'moving_mean', shape, null, this.movingMeanInitializer, null, false);\n    this.movingVariance = this.addWeight(\n        'moving_variance', shape, null, this.movingVarianceInitializer, null,\n        false);\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const ndim = inputShape.length;\n      const reductionAxes = math_utils.range(0, ndim);\n      const axis = this.axis >= 0 ? this.axis : (this.axis + ndim);\n      reductionAxes.splice(axis, 1);\n      const broadcastShape = generic_utils.pyListRepeat(1, ndim);\n      broadcastShape[axis] = inputShape[axis];\n\n      const sortedReductionAxes = reductionAxes.slice();\n      sortedReductionAxes.sort();\n      const needsBroadcasting = !util.arraysEqual(\n          sortedReductionAxes, math_utils.range(0, ndim).slice(0, ndim - 1));\n\n      const normalizeInference: () => Tensor = () => {\n        if (needsBroadcasting) {\n          const broadcastMovingMean =\n              this.movingMean.read().reshape(broadcastShape);\n          const broadcastMovingVariance =\n              this.movingVariance.read().reshape(broadcastShape);\n          const broadcastBeta =\n              this.center ? this.beta.read().reshape(broadcastShape) : null;\n          const broadcastGamma =\n              this.scale ? this.gamma.read().reshape(broadcastShape) : null;\n          return batchNormalization(\n              input, broadcastMovingMean, broadcastMovingVariance,\n              broadcastBeta, broadcastGamma, this.epsilon);\n        } else {\n          return batchNormalization(\n              input, this.movingMean.read(), this.movingVariance.read(),\n              this.beta == null ? null : this.beta.read(),\n              this.gamma == null ? null : this.gamma.read(), this.epsilon);\n        }\n      };\n\n      if (!training) {\n        return normalizeInference();\n      }\n\n      const [normedTraining, mean, variance] = normalizeBatchInTraining(\n          input, this.gamma.read(), this.beta.read(), reductionAxes,\n          this.epsilon);\n\n      const doMovingAverage =\n          (variable: LayerVariable, value: Tensor, momentum: number): void => {\n            tfc.tidy(() => {\n              const decay = 1 - momentum;\n              const origValue = variable.read();\n              const updateDelta = origValue.sub(value).mul(decay);\n              variable.write(origValue.sub(updateDelta));\n            });\n          };\n\n      // Perform updates to moving mean and moving variance for training.\n      // Porting Note: In PyKeras, these updates to `movingMean` and\n      //   `movingAverage` are done as a deferred Graph, added to the `Layer`'s\n      //   `update`s using the `add_update()` method. Here we do it imperatively\n      //   and encapsulate the updates in a function that is invoked\n      //   immediately.\n      const updateMovingMeanAndVariance = () => {\n        doMovingAverage(this.movingMean, mean, this.momentum);\n        doMovingAverage(this.movingVariance, variance, this.momentum);\n      };\n      updateMovingMeanAndVariance();\n\n      return normedTraining;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      momentum: this.momentum,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      movingMeanInitializer: serializeInitializer(this.movingMeanInitializer),\n      movingVarianceInitializer:\n          serializeInitializer(this.movingVarianceInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer),\n      betaConstraint: serializeConstraint(this.betaConstraint),\n      gammaConstraint: serializeConstraint(this.gammaConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(BatchNormalization);\n\nexport interface LayerNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The axis or axes that should be normalized (typically, the feature axis.)\n   * Defaults to -1 (the last axis.)\n   */\n  axis?: number|number[];\n\n  /**\n   * A small positive float added to variance to avoid divison by zero.\n   * Defaults to 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Default: `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply output by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear, this can be disabled since scaling will\n   * be done by the next layer.\n   * Default: `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   * Default: `'zeros'`.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   * Default: `'ones'`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /** Regularizer for the beta weight. */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /** Regularizer for the gamma weight. */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class LayerNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'LayerNormalization';\n\n  private axis: number|number[];\n  readonly epsilon: number;\n  readonly center: boolean;\n  readonly scale: boolean;\n  readonly betaInitializer: Initializer;\n  readonly gammaInitializer: Initializer;\n  readonly betaRegularizer: Regularizer;\n  readonly gammaRegularizer: Regularizer;\n\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n\n  constructor(args?: LayerNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.axis = args.axis == null ? -1 : args.axis;\n    if (typeof this.axis === 'number') {\n      if (!Number.isInteger(this.axis)) {\n        throw new Error(\n            `Expected axis to be an integer, but received ${this.axis}`);\n      }\n    } else if (Array.isArray(this.axis)) {\n      for (const axis of this.axis) {\n        if (!Number.isInteger(axis)) {\n          throw new Error(\n              `Expected axis to be an array of integers, ` +\n              `but received ${JSON.stringify(this.axis)}`);\n        }\n      }\n    } else {\n      throw new Error(\n          `Expected axis to be an integer or an array of integers, ` +\n          `but received ${JSON.stringify(this.axis)}`);\n    }\n\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n\n    this.supportsMasking = true;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const nDims = inputShape.length;\n\n    // Convert axis to array and resolve negatives.\n    if (typeof this.axis === 'number') {\n      this.axis = [this.axis];\n    }\n    for (let i = 0; i < this.axis.length; ++i) {\n      if (this.axis[i] < 0) {\n        this.axis[i] += nDims;\n      }\n    }\n\n    // Further validate axes.\n    for (const axis of this.axis) {\n      if (axis < 0 || axis >= nDims) {\n        throw new Error(`Invalid axis: ${axis}`);\n      }\n    }\n    if (this.axis.length !== generic_utils.unique(this.axis).length) {\n      throw new Error(`Found duplicate axes in: ${this.axis}`);\n    }\n\n    const paramShape = this.axis.map(axis => inputShape[axis]) as number[];\n\n    const trainable = true;\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', paramShape, 'float32', this.gammaInitializer,\n          this.gammaRegularizer, trainable);\n    } else {\n      this.gamma = null;\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', paramShape, 'float32', this.betaInitializer,\n          this.betaRegularizer, trainable);\n    } else {\n      this.beta = null;\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const input = getExactlyOneTensor(inputs);\n    const inputShape = input.shape;\n    const nDims = inputShape.length;\n\n    return tidy(() => {\n      const keepDims = true;\n      let {mean, variance} = moments(input, this.axis, keepDims);\n      const broadcastShape = generic_utils.pyListRepeat(1, nDims);\n      for (const dim of this.axis as number[]) {\n        broadcastShape[dim] = inputShape[dim];\n      }\n\n      const broadcast = (v: Tensor) => {\n        if (v != null && v.shape.length !== nDims &&\n            this.axis !== [nDims - 1]) {\n          return v.reshape(broadcastShape);\n        } else {\n          return v;\n        }\n      };\n\n      let scale = broadcast(this.gamma.read());\n      let offset = broadcast(this.beta.read());\n\n      // TODO(https://github.com/tensorflow/tfjs/issues/2120): The tiling below\n      // is a workaround for the limitation of core's batchNormalization?d don't\n      // support broadcasting in their gradients. In addition, the tiling is\n      // necessary to ensure correctness on the browser CPU backend regardless\n      // of forward or backward computation. Remove this workaround once the\n      // limitation is addressed. See .\n      const momentsTiling: number[] = [];\n      const scaleOffsetTiling: number[] = [];\n      for (let i = 0; i < nDims; ++i) {\n        if ((this.axis as number[]).indexOf(i) !== -1) {\n          momentsTiling.push(inputShape[i]);\n          scaleOffsetTiling.push(1);\n        } else {\n          momentsTiling.push(1);\n          scaleOffsetTiling.push(inputShape[i]);\n        }\n      }\n      mean = mean.tile(momentsTiling);\n      variance = variance.tile(momentsTiling);\n      scale = scale.tile(scaleOffsetTiling);\n      offset = offset.tile(scaleOffsetTiling);\n\n      return batchNormalization(\n          input, mean, variance, offset, scale, this.epsilon);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LayerNormalization);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Padding Layers.\n */\n\n// Porting Note: In Python Keras, the padding layers are in convolutional.py,\n//   but we decided to put them in a separate file (padding.ts) for clarity.\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\n/**\n * Pads the middle dimension of a 3D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of 2 integers, how many zeros to add at the start and\n *   end of the middle dimension (i.e., dimension 1).\n * @return A padded 3D `tf.Tensor`.\n */\nexport function temporalPadding(x: Tensor, padding?: [number, number]): Tensor {\n  return tidy(() => {\n    if (x.rank !== 3) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 3-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [1, 1];\n    }\n    if (padding.length !== 2) {\n      throw new ValueError(\n          `temporalPadding expects input padding pattern to be a length-2 ` +\n          `array, but received a length-${padding.length} array.`);\n    }\n\n    const pattern: Array<[number, number]> = [[0, 0], padding, [0, 0]];\n    return tfc.pad(x, pattern);\n  });\n}\n\n/**\n * Pads the 2nd and 3rd dimensions of a 4D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of two `Array`s, each of which is an `Array` of two\n *   integers. The amount of padding at the beginning and end of the 2nd and 3rd\n *   dimensions, respectively.\n * @param dataFormat 'channelsLast' (default) or 'channelsFirst'.\n * @return Padded 4D `tf.Tensor`.\n */\nexport function spatial2dPadding(\n    x: Tensor, padding?: [[number, number], [number, number]],\n    dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 4-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [[1, 1], [1, 1]];\n    }\n    if (padding.length !== 2 || padding[0].length !== 2 ||\n        padding[1].length !== 2) {\n      throw new ValueError(\n          'spatial2dPadding expects `padding` to be an Array of two Arrays, ' +\n          'each of which is an Array of two integers.');\n    }\n\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (dataFormat !== 'channelsLast' && dataFormat !== 'channelsFirst') {\n      throw new ValueError(\n          `Unknown data format: ${dataFormat}. ` +\n          `Supported data formats are 'channelsLast' and 'channelsFirst.`);\n    }\n\n    let pattern: Array<[number, number]>;\n    if (dataFormat === 'channelsFirst') {\n      pattern = [[0, 0], [0, 0], padding[0], padding[1]];\n    } else {\n      pattern = [[0, 0], padding[0], padding[1], [0, 0]];\n    }\n\n    return tfc.pad(x, pattern);\n  });\n}\n\nexport declare interface ZeroPadding2DLayerArgs extends LayerArgs {\n  /**\n   * Integer, or `Array` of 2 integers, or `Array` of 2 `Array`s, each of\n   * which is an `Array` of 2 integers.\n   * - If integer, the same symmetric padding is applied to width and height.\n   * - If Array` of 2 integers, interpreted as two different symmetric values\n   *   for height and width:\n   *   `[symmetricHeightPad, symmetricWidthPad]`.\n   * - If `Array` of 2 `Array`s, interpreted as:\n   *   `[[topPad, bottomPad], [leftPad, rightPad]]`.\n   */\n  padding?: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * One of `'channelsLast'` (default) and `'channelsFirst'`.\n   *\n   * The ordering of the dimensions in the inputs.\n   * `channelsLast` corresponds to inputs with shape\n   * `[batch, height, width, channels]` while `channelsFirst`\n   * corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class ZeroPadding2D extends Layer {\n  /** @nocollapse */\n  static className = 'ZeroPadding2D';\n  readonly dataFormat: DataFormat;\n  readonly padding: [[number, number], [number, number]];\n\n  constructor(args?: ZeroPadding2DLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.dataFormat =\n        args.dataFormat == null ? imageDataFormat() : args.dataFormat;\n    // TODO(cais): Maybe refactor the following logic surrounding `padding`\n    //   into a helper method.\n    if (args.padding == null) {\n      this.padding = [[1, 1], [1, 1]];\n    } else if (typeof args.padding === 'number') {\n      this.padding =\n          [[args.padding, args.padding], [args.padding, args.padding]];\n    } else {\n      args.padding = args.padding;\n      if (args.padding.length !== 2) {\n        throw new ValueError(\n            `ZeroPadding2D expects padding to be a length-2 array, but ` +\n            `received a length-${args.padding.length} array.`);\n      }\n\n      let heightPadding: [number, number];\n      let widthPadding: [number, number];\n      if (typeof args.padding[0] === 'number') {\n        heightPadding = [args.padding[0], args.padding[0]];\n        widthPadding = [args.padding[1] as number, args.padding[1] as number];\n      } else {\n        args.padding = args.padding as [[number, number], [number, number]];\n\n        if (args.padding[0].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects height padding to be a length-2 array, ` +\n              `but received a length-${args.padding[0].length} array.`);\n        }\n        heightPadding = args.padding[0] as [number, number];\n\n        if (args.padding[1].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects width padding to be a length-2 array, ` +\n              `but received a length-${args.padding[1].length} array.`);\n        }\n        widthPadding = args.padding[1] as [number, number];\n      }\n      this.padding = [heightPadding, widthPadding];\n    }\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n\n    let rows: number;\n    let cols: number;\n    if (this.dataFormat === 'channelsFirst') {\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        rows = inputShape[2] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[3] != null && inputShape[3] >= 0) {\n        cols = inputShape[3] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      if (inputShape[1] != null && inputShape[1] >= 0) {\n        rows = inputShape[1] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        cols = inputShape[2] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(\n        () => spatial2dPadding(\n            getExactlyOneTensor(inputs), this.padding, this.dataFormat));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ZeroPadding2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Pooling Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor3D, Tensor4D, Tensor5D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode, checkPoolMode} from '../common';\nimport {InputSpec} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, PaddingMode, PoolMode, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\nimport {preprocessConv2DInput, preprocessConv3DInput} from './convolutional';\n\n/**\n * 2D pooling.\n * @param x\n * @param poolSize\n * @param stridesdes strides. Defaults to [1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 2D pooling.\n */\nexport function pool2d(\n    x: Tensor, poolSize: [number, number], strides?: [number, number],\n    padding?: PaddingMode, dataFormat?: DataFormat,\n    poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // TODO(cais): Remove the preprocessing step once deeplearn.js supports\n    // dataFormat as an input argument.\n    x = preprocessConv2DInput(x, dataFormat);  // x is NHWC after preprocessing.\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      // TODO(cais): Rank check?\n      y = tfc.maxPool(x as Tensor4D, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      // TODO(cais): Check the dtype and rank of x and give clear error message\n      //   if those are incorrect.\n      y = tfc.avgPool(\n          // TODO(cais): Rank check?\n          x as Tensor3D | Tensor4D, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);  // NHWC -> NCHW.\n    }\n    return y;\n  });\n}\n\n/**\n * 3D pooling.\n * @param x\n * @param poolSize. Default to [1, 1, 1].\n * @param strides strides. Defaults to [1, 1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 3D pooling.\n */\nexport function pool3d(\n    x: Tensor5D, poolSize: [number, number, number],\n    strides?: [number, number, number], padding?: PaddingMode,\n    dataFormat?: DataFormat, poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // x is NDHWC after preprocessing.\n    x = preprocessConv3DInput(x as Tensor, dataFormat) as Tensor5D;\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      y = tfc.maxPool3d(x, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      y = tfc.avgPool3d(x, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);  // NDHWC -> NCDHW.\n    }\n    return y;\n  });\n}\n\nexport declare interface Pooling1DLayerArgs extends LayerArgs {\n  /**\n   * Size of the window to pool over, should be an integer.\n   */\n  poolSize?: number|[number];\n  /**\n   * Period at which to sample the pooled values.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number];\n  /** How to fill in data that's not an integer multiple of poolSize. */\n  padding?: PaddingMode;\n}\n\n/**\n * Abstract class for different pooling 1D layers.\n */\nexport abstract class Pooling1D extends Layer {\n  protected readonly poolSize: [number];\n  protected readonly strides: [number];\n  protected readonly padding: PaddingMode;\n\n  /**\n   *\n   * @param args Parameters for the Pooling layer.\n   *\n   * config.poolSize defaults to 2.\n   */\n  constructor(args: Pooling1DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = 2;\n    }\n    super(args);\n    if (typeof args.poolSize === 'number') {\n      this.poolSize = [args.poolSize];\n    } else if (\n        Array.isArray(args.poolSize) &&\n        (args.poolSize as number[]).length === 1 &&\n        typeof (args.poolSize as number[])[0] === 'number') {\n      this.poolSize = args.poolSize;\n    } else {\n      throw new ValueError(\n          `poolSize for 1D convolutional layer must be a number or an ` +\n          `Array of a single number, but received ` +\n          `${JSON.stringify(args.poolSize)}`);\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else {\n      if (typeof args.strides === 'number') {\n        this.strides = [args.strides];\n      } else if (\n          Array.isArray(args.strides) &&\n          (args.strides as number[]).length === 1 &&\n          typeof (args.strides as number[])[0] === 'number') {\n        this.strides = args.strides;\n      } else {\n        throw new ValueError(\n            `strides for 1D convolutional layer must be a number or an ` +\n            `Array of a single number, but received ` +\n            `${JSON.stringify(args.strides)}`);\n      }\n    }\n    assertPositiveInteger(this.strides, 'strides');\n\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const length = convOutputLength(\n        inputShape[1], this.poolSize[0], this.padding, this.strides[0]);\n    return [inputShape[0], length, inputShape[2]];\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Add dummy last dimension.\n      inputs = K.expandDims(getExactlyOneTensor(inputs), 2);\n      const output = this.poolingFunction(\n          getExactlyOneTensor(inputs), [this.poolSize[0], 1],\n          [this.strides[0], 1], this.padding, 'channelsLast');\n      // Remove dummy last dimension.\n      return tfc.squeeze(output, [2]);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'MaxPooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling1D);\n\nexport class AveragePooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'AveragePooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling1D);\n\nexport declare interface Pooling2DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [vertical, horizontal].\n   * Expects an integer or an array of 2 integers.\n   *\n   * For example, `[2, 2]` will halve the input in both spatial dimension.\n   * If only one integer is specified, the same window length\n   * will be used for both dimensions.\n   */\n  poolSize?: number|[number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 2 integers. Integer, tuple of 2 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 2D layers.\n */\nexport abstract class Pooling2D extends Layer {\n  protected readonly poolSize: [number, number];\n  protected readonly strides: [number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling2DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 2) {\n        throw new ValueError(\n            `If the strides property of a 2D pooling layer is an Array, ` +\n            `it is expected to have a length of 2, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    rows =\n        convOutputLength(rows, this.poolSize[0], this.padding, this.strides[0]);\n    cols =\n        convOutputLength(cols, this.poolSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'MaxPooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling2D);\n\nexport class AveragePooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'AveragePooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling2D);\n\nexport declare interface Pooling3DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [depth, height, width].\n   * Expects an integer or an array of 3 integers.\n   *\n   * For example, `[2, 2, 2]` will halve the input in three dimensions.\n   * If only one integer is specified, the same window length\n   * will be used for all dimensions.\n   */\n  poolSize?: number|[number, number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 3 integers. Integer, tuple of 3 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 3D layers.\n */\nexport abstract class Pooling3D extends Layer {\n  protected readonly poolSize: [number, number, number];\n  protected readonly strides: [number, number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling3DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 3) {\n        throw new ValueError(\n            `If the strides property of a 3D pooling layer is an Array, ` +\n            `it is expected to have a length of 3, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let depths =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[4] : inputShape[3];\n    depths = convOutputLength(\n        depths, this.poolSize[0], this.padding, this.strides[0]);\n    rows =\n        convOutputLength(rows, this.poolSize[1], this.padding, this.strides[1]);\n    cols =\n        convOutputLength(cols, this.poolSize[2], this.padding, this.strides[2]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], depths, rows, cols];\n    } else {\n      return [inputShape[0], depths, rows, cols, inputShape[4]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'MaxPooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling3D);\n\nexport class AveragePooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'AveragePooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling3D);\n\n/**\n * Abstract class for different global pooling 1D layers.\n */\nexport abstract class GlobalPooling1D extends Layer {\n  constructor(args: LayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], inputShape[2]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n}\n\nexport class GlobalAveragePooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling1D';\n  constructor(args?: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.mean(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling1D);\n\nexport class GlobalMaxPooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling1D';\n  constructor(args: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.max(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling1D);\n\nexport declare interface GlobalPooling2DLayerArgs extends LayerArgs {\n  /**\n   * One of `CHANNEL_LAST` (default) or `CHANNEL_FIRST`.\n   *\n   * The ordering of the dimensions in the inputs. `CHANNEL_LAST` corresponds\n   * to inputs with shape `[batch, height, width, channels[` while\n   * `CHANNEL_FIRST` corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different global pooling 2D layers.\n */\nexport abstract class GlobalPooling2D extends Layer {\n  protected dataFormat: DataFormat;\n  constructor(args: GlobalPooling2DLayerArgs) {\n    super(args);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape;\n    if (this.dataFormat === 'channelsLast') {\n      return [inputShape[0], inputShape[3]];\n    } else {\n      return [inputShape[0], inputShape[1]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class GlobalAveragePooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.mean(input, [1, 2]);\n      } else {\n        return tfc.mean(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling2D);\n\nexport class GlobalMaxPooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.max(input, [1, 2]);\n      } else {\n        return tfc.max(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Layers that augment the functionality of a base layer.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {InputSpec, Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {BidirectionalMergeMode, Shape, VALID_BIDIRECTIONAL_MERGE_MODES} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {RegularizerFn, RnnStepFunction} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {rnn, RNN, standardizeArgs} from './recurrent';\nimport {deserialize} from './serialization';\n\nexport declare interface WrapperLayerArgs extends LayerArgs {\n  /**\n   * The layer to be wrapped.\n   */\n  layer: Layer;\n}\n\n/**\n * Abstract wrapper base class.\n *\n * Wrappers take another layer and augment it in various ways.\n * Do not use this class as a layer, it is only an abstract base class.\n * Two usable wrappers are the `TimeDistributed` and `Bidirectional` wrappers.\n */\nexport abstract class Wrapper extends Layer {\n  readonly layer: Layer;\n\n  constructor(args: WrapperLayerArgs) {\n    // Porting Note: In PyKeras, `self.layer` is set prior to the calling\n    //   `super()`. But we can't do that here due to TypeScript's restriction.\n    //   See: https://github.com/Microsoft/TypeScript/issues/8277\n    //   As a result, we have to add checks in `get trainable()` and\n    //   `set trainable()` below in order to prevent using `this.layer` when\n    //   its value is `undefined`. The super constructor does use the getter\n    //   and the setter of `this.layer`.\n    super(args);\n    this.layer = args.layer;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    this.built = true;\n  }\n\n  // TODO(cais): Implement activityRegularizer getter.\n\n  get trainable(): boolean {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      return this.layer.trainable;\n    } else {\n      return false;\n    }\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      this.layer.trainable = value;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.layer.trainableWeights;\n  }\n  // TODO(cais): Implement setter for trainableWeights.\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.layer.nonTrainableWeights;\n  }\n  // TODO(cais): Implement setter for nonTrainableWeights.\n\n  get updates(): Tensor[] {\n    // tslint:disable-next-line:no-any\n    return (this.layer as any)._updates;\n  }\n\n  // TODO(cais): Implement getUpdatesFor().\n\n  get losses(): RegularizerFn[] {\n    return this.layer.losses;\n  }\n\n  // TODO(cais): Implement getLossesFor().\n\n  getWeights(): Tensor[] {\n    return this.layer.getWeights();\n  }\n\n  setWeights(weights: Tensor[]): void {\n    this.layer.setWeights(weights);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'layer': {\n        'className': this.layer.getClassName(),\n        'config': this.layer.getConfig(),\n      }\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.layer != null) {\n      this.layer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const layerConfig = config['layer'] as serialization.ConfigDict;\n    const layer = deserialize(layerConfig, customObjects) as Layer;\n    delete config['layer'];\n    const newConfig = {layer};\n    Object.assign(newConfig, config);\n    return new cls(newConfig);\n  }\n}\n\nexport class TimeDistributed extends Wrapper {\n  /** @nocollapse */\n  static className = 'TimeDistributed';\n  constructor(args: WrapperLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 3) {\n      throw new ValueError(\n          `TimeDistributed layer expects an input shape >= 3D, but received ` +\n          `input shape ${JSON.stringify(inputShape)}`);\n    }\n    this.inputSpec = [{shape: inputShape}];\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (!this.layer.built) {\n      this.layer.build(childInputShape);\n      this.layer.built = true;\n    }\n    super.build(inputShape);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    const childOutputShape =\n        this.layer.computeOutputShape(childInputShape) as Shape;\n    const timesteps = inputShape[1];\n    return [childOutputShape[0], timesteps].concat(childOutputShape.slice(1));\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      // TODO(cais): Add 'training' and 'useLearningPhase' to kwargs.\n      inputs = getExactlyOneTensor(inputs);\n      // Porting Note: In tfjs-layers, `inputs` are always concrete tensor\n      // values. Hence the inputs can't have an undetermined first (batch)\n      // dimension, which is why we always use the K.rnn approach here.\n      const step: RnnStepFunction = (inputs: Tensor, states: Tensor[]) => {\n        // TODO(cais): Add useLearningPhase.\n        // NOTE(cais): `layer.call` may return a length-1 array of Tensor in\n        //   some cases (e.g., `layer` is a `Sequential` instance), which is\n        //   why `getExactlyOneTensor` is used below.\n        const output = getExactlyOneTensor(this.layer.call(inputs, kwargs));\n        return [output, []];\n      };\n      const rnnOutputs =\n          rnn(step, inputs, [], false /* goBackwards */, null /* mask */,\n              null /* constants */, false /* unroll */,\n              true /* needPerStepOutputs */);\n      const y = rnnOutputs[1];\n      // TODO(cais): Add activity regularization.\n      // TODO(cais): Add useLearningPhase.\n      return y;\n    });\n  }\n\n  // TODO(cais): Implement detailed computeMask() logic.\n}\nserialization.registerClass(TimeDistributed);\n\nexport function checkBidirectionalMergeMode(value?: string): void {\n  generic_utils.checkStringTypeUnionValue(\n      VALID_BIDIRECTIONAL_MERGE_MODES, 'BidirectionalMergeMode', value);\n}\n\nexport declare interface BidirectionalLayerArgs extends WrapperLayerArgs {\n  /**\n   * The instance of an `RNN` layer to be wrapped.\n   */\n  layer: RNN;\n\n  /**\n   * Mode by which outputs of the forward and backward RNNs are\n   * combined. If `null` or `undefined`, the output will not be\n   * combined, they will be returned as an `Array`.\n   *\n   * If `undefined` (i.e., not provided), defaults to `'concat'`.\n   */\n  mergeMode?: BidirectionalMergeMode;\n}\n\nconst DEFAULT_BIDIRECTIONAL_MERGE_MODE: BidirectionalMergeMode = 'concat';\n\nexport class Bidirectional extends Wrapper {\n  /** @nocollapse */\n  static className = 'Bidirectional';\n  mergeMode: BidirectionalMergeMode;\n  private forwardLayer: RNN;\n  private backwardLayer: RNN;\n  private returnSequences: boolean;\n  private returnState: boolean;\n  private numConstants?: number;\n  private _trainable: boolean;\n\n  constructor(args: BidirectionalLayerArgs) {\n    super(args);\n\n    // Note: When creating `this.forwardLayer`, the original Layer object\n    //   (`config.layer`) ought to be cloned. This is why we call\n    //   `getConfig()` followed by `deserialize()`. Without this cloning,\n    //   the layer names saved during serialization will incorrectly contain\n    //   the 'forward_' prefix. In Python Keras, this is done using\n    //   `copy.copy` (shallow copy), which does not have a simple equivalent\n    //   in JavaScript. JavaScript's `Object.assign()` does not copy\n    //   methods.\n    const layerConfig = args.layer.getConfig();\n    const forwDict: serialization.ConfigDict = {};\n    forwDict['className'] = args.layer.getClassName();\n    forwDict['config'] = layerConfig;\n    this.forwardLayer = deserialize(forwDict) as RNN;\n    layerConfig['goBackwards'] =\n        layerConfig['goBackwards'] === true ? false : true;\n    const backDict: serialization.ConfigDict = {};\n    backDict['className'] = args.layer.getClassName();\n    backDict['config'] = layerConfig;\n    this.backwardLayer = deserialize(backDict) as RNN;\n    this.forwardLayer.name = 'forward_' + this.forwardLayer.name;\n    this.backwardLayer.name = 'backward_' + this.backwardLayer.name;\n\n    this.mergeMode = args.mergeMode === undefined ?\n        DEFAULT_BIDIRECTIONAL_MERGE_MODE :\n        args.mergeMode;\n    checkBidirectionalMergeMode(this.mergeMode);\n    if (args.weights) {\n      throw new NotImplementedError(\n          'weights support is not implemented for Bidirectional layer yet.');\n    }\n    this._stateful = args.layer.stateful;\n    this.returnSequences = args.layer.returnSequences;\n    this.returnState = args.layer.returnState;\n    this.supportsMasking = true;\n    this._trainable = true;\n    this.inputSpec = args.layer.inputSpec;\n    this.numConstants = null;\n  }\n\n  get trainable(): boolean {\n    return this._trainable;\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    this._trainable = value;\n    if (this.forwardLayer != null) {\n      this.forwardLayer.trainable = value;\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.trainable = value;\n    }\n  }\n\n  getWeights(): Tensor[] {\n    return this.forwardLayer.getWeights().concat(\n        this.backwardLayer.getWeights());\n  }\n\n  setWeights(weights: Tensor[]): void {\n    const numWeights = weights.length;\n    const numeightsOver2 = Math.floor(numWeights / 2);\n    this.forwardLayer.setWeights(weights.slice(0, numeightsOver2));\n    this.backwardLayer.setWeights(weights.slice(numeightsOver2));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    let layerShapes: Shape|Shape[] =\n        this.forwardLayer.computeOutputShape(inputShape);\n    if (!(Array.isArray(layerShapes) && Array.isArray(layerShapes[0]))) {\n      layerShapes = [layerShapes as Shape];\n    }\n    layerShapes = layerShapes as Shape[];\n\n    let outputShape: Shape;\n    let outputShapes: Shape[];\n    let stateShape: Shape[];\n    if (this.returnState) {\n      stateShape = layerShapes.slice(1);\n      outputShape = layerShapes[0];\n    } else {\n      outputShape = layerShapes[0];\n    }\n    outputShape = outputShape;\n    if (this.mergeMode === 'concat') {\n      outputShape[outputShape.length - 1] *= 2;\n      outputShapes = [outputShape];\n    } else if (this.mergeMode == null) {\n      outputShapes = [outputShape, outputShape.slice()];\n    } else {\n      outputShapes = [outputShape];\n    }\n\n    if (this.returnState) {\n      if (this.mergeMode == null) {\n        return outputShapes.concat(stateShape).concat(stateShape.slice());\n      }\n      return [outputShape].concat(stateShape).concat(stateShape.slice());\n    }\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    if (Array.isArray(inputs)) {\n      initialState = (inputs as Tensor[] | SymbolicTensor[]).slice(1);\n      inputs = (inputs as Tensor[] | SymbolicTensor[])[0];\n    }\n\n    if ((initialState == null || initialState.length === 0) &&\n        constants == null) {\n      return super.apply(inputs, kwargs);\n    }\n    const additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    const additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      const numStates = initialState.length;\n      if (numStates % 2 > 0) {\n        throw new ValueError(\n            'When passing `initialState` to a Bidrectional RNN, ' +\n            'the state should be an Array containing the states of ' +\n            'the underlying RNNs.');\n      }\n      kwargs['initialState'] = initialState;\n      additionalInputs.push(...initialState);\n      const stateSpecs = (initialState as Array<Tensor|SymbolicTensor>)\n                             .map(state => new InputSpec({shape: state.shape}));\n      this.forwardLayer.stateSpec = stateSpecs.slice(0, numStates / 2);\n      this.backwardLayer.stateSpec = stateSpecs.slice(numStates / 2);\n      additionalSpecs.push(...stateSpecs);\n    }\n    if (constants != null) {\n      throw new NotImplementedError(\n          'Support for constants in Bidirectional layers is not ' +\n          'implemented yet.');\n    }\n\n    const isSymbolicTensor = additionalInputs[0] instanceof SymbolicTensor;\n    for (const tensor of additionalInputs) {\n      if (tensor instanceof SymbolicTensor !== isSymbolicTensor) {\n        throw new ValueError(\n            'The initial state of a Bidirectional layer cannot be ' +\n            'specified as a mix of symbolic and non-symbolic tensors');\n      }\n    }\n\n    if (isSymbolicTensor) {\n      // Compute the full input and specs, including the states.\n      const fullInput = [inputs].concat(additionalInputs);\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call temporarily and replace inputSpec.\n      // Note: with initial states symbolic calls and non-symbolic calls to\n      // this method differ in how the initial states are passed. For\n      // symbolic calls, the initial states are passed in the first arg, as\n      // an Array of SymbolicTensors; for non-symbolic calls, they are\n      // passed in the second arg as a part of the kwargs. Hence the need to\n      // temporarily modify inputSpec here.\n      // TODO(cais): Make refactoring so that this hacky code below is no\n      // longer needed.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output =\n          super.apply(fullInput as Tensor[] | SymbolicTensor[], kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const initialState = kwargs['initialState'];\n\n      let y: Tensor|Tensor[];\n      let yRev: Tensor|Tensor[];\n      if (initialState == null) {\n        y = this.forwardLayer.call(inputs, kwargs);\n        yRev = this.backwardLayer.call(inputs, kwargs);\n      } else {\n        const forwardState = initialState.slice(0, initialState.length / 2);\n        const backwardState = initialState.slice(initialState.length / 2);\n        y = this.forwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: forwardState}));\n        yRev = this.backwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: backwardState}));\n      }\n\n      let states: Tensor[];\n      if (this.returnState) {\n        if (Array.isArray(y)) {\n          states = y.slice(1).concat((yRev as Tensor[]).slice(1));\n        } else {\n        }\n        y = (y as Tensor[])[0];\n        yRev = (yRev as Tensor[])[0];\n      }\n\n      if (this.returnSequences) {\n        yRev = tfc.reverse(yRev as Tensor, 1);\n      }\n\n      let output: Tensor|Tensor[];\n      if (this.mergeMode === 'concat') {\n        output = K.concatenate([y as Tensor, yRev as Tensor]);\n      } else if (this.mergeMode === 'sum') {\n        output = tfc.add(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode === 'ave') {\n        output = tfc.mul(.5, tfc.add(y as Tensor, yRev as Tensor));\n      } else if (this.mergeMode === 'mul') {\n        output = tfc.mul(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode == null) {\n        output = [y as Tensor, yRev as Tensor];\n      }\n\n      // TODO(cais): Properly set learning phase.\n      if (this.returnState) {\n        if (this.mergeMode == null) {\n          return (output as Tensor[]).concat(states);\n        }\n        return [output as Tensor].concat(states);\n      }\n      return output;\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[]): void {\n    this.forwardLayer.resetStates();\n    this.backwardLayer.resetStates();\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    nameScope(this.forwardLayer.name, () => {\n      this.forwardLayer.build(inputShape);\n    });\n    nameScope(this.backwardLayer.name, () => {\n      this.backwardLayer.build(inputShape);\n    });\n    this.built = true;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (Array.isArray(mask)) {\n      mask = mask[0];\n    }\n    let outputMask: Tensor|Tensor[];\n    if (this.returnSequences) {\n      if (this.mergeMode == null) {\n        outputMask = [mask, mask];\n      } else {\n        outputMask = mask;\n      }\n    } else {\n      if (this.mergeMode == null) {\n        outputMask = [null, null];\n      } else {\n        outputMask = null;\n      }\n    }\n    if (this.returnState) {\n      const states = this.forwardLayer.states;\n      const stateMask: Tensor[] = states.map(state => null);\n      if (Array.isArray(outputMask)) {\n        return outputMask.concat(stateMask).concat(stateMask);\n      } else {\n        return [outputMask].concat(stateMask).concat(stateMask);\n      }\n    } else {\n      return outputMask;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.forwardLayer.trainableWeights.concat(\n        this.backwardLayer.trainableWeights);\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.forwardLayer.nonTrainableWeights.concat(\n        this.backwardLayer.nonTrainableWeights);\n  }\n\n  // TODO(cais): Implement constraints().\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.forwardLayer != null) {\n      this.forwardLayer.setFastWeightInitDuringBuild(value);\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'mergeMode': this.mergeMode,\n    };\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    const rnnLayer =\n        deserialize(config['layer'] as serialization.ConfigDict) as RNN;\n    delete config['layer'];\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    if (config['numConstants'] != null) {\n      throw new NotImplementedError(\n          `Deserialization of a Bidirectional layer with numConstants ` +\n          `present is not supported yet.`);\n    }\n    // tslint:disable-next-line:no-any\n    const newConfig: {[key: string]: any} = config;\n    newConfig['layer'] = rnnLayer;\n    return new cls(newConfig);\n  }\n}\nserialization.registerClass(Bidirectional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {InputLayer, InputLayerArgs} from './engine/input_layer';\nimport {Layer, LayerArgs} from './engine/topology';\nimport {input} from './exports';\nimport {ELU, ELULayerArgs, LeakyReLU, LeakyReLULayerArgs, PReLU, PReLULayerArgs, ReLU, ReLULayerArgs, Softmax, SoftmaxLayerArgs, ThresholdedReLU, ThresholdedReLULayerArgs} from './layers/advanced_activations';\nimport {Conv1D, Conv2D, Conv2DTranspose, Conv3D, ConvLayerArgs, Cropping2D, Cropping2DLayerArgs, SeparableConv2D, SeparableConvLayerArgs, UpSampling2D, UpSampling2DLayerArgs} from './layers/convolutional';\nimport {DepthwiseConv2D, DepthwiseConv2DLayerArgs} from './layers/convolutional_depthwise';\nimport {ConvLSTM2D, ConvLSTM2DArgs, ConvLSTM2DCell, ConvLSTM2DCellArgs} from './layers/convolutional_recurrent';\nimport {Activation, ActivationLayerArgs, Dense, DenseLayerArgs, Dropout, DropoutLayerArgs, Flatten, FlattenLayerArgs, Masking, MaskingArgs, Permute, PermuteLayerArgs, RepeatVector, RepeatVectorLayerArgs, Reshape, ReshapeLayerArgs, SpatialDropout1D, SpatialDropout1DLayerConfig} from './layers/core';\nimport {Embedding, EmbeddingLayerArgs} from './layers/embeddings';\nimport {Add, Average, Concatenate, ConcatenateLayerArgs, Dot, DotLayerArgs, Maximum, Minimum, Multiply} from './layers/merge';\nimport {AlphaDropout, AlphaDropoutArgs, GaussianDropout, GaussianDropoutArgs, GaussianNoise, GaussianNoiseArgs} from './layers/noise';\nimport {BatchNormalization, BatchNormalizationLayerArgs, LayerNormalization, LayerNormalizationLayerArgs} from './layers/normalization';\nimport {ZeroPadding2D, ZeroPadding2DLayerArgs} from './layers/padding';\nimport {AveragePooling1D, AveragePooling2D, AveragePooling3D, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalPooling2DLayerArgs, MaxPooling1D, MaxPooling2D, MaxPooling3D, Pooling1DLayerArgs, Pooling2DLayerArgs, Pooling3DLayerArgs} from './layers/pooling';\nimport {GRU, GRUCell, GRUCellLayerArgs, GRULayerArgs, LSTM, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNN, SimpleRNNCell, SimpleRNNCellLayerArgs, SimpleRNNLayerArgs, StackedRNNCells, StackedRNNCellsArgs} from './layers/recurrent';\nimport {Bidirectional, BidirectionalLayerArgs, TimeDistributed, WrapperLayerArgs} from './layers/wrappers';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// Input Layer.\n/**\n * An input layer is an entry point into a `tf.LayersModel`.\n *\n * `InputLayer` is generated automatically for `tf.Sequential`` models by\n * specifying the `inputshape` or `batchInputShape` for the first layer.  It\n * should not be specified explicitly. However, it can be useful sometimes,\n * e.g., when constructing a sequential model from a subset of another\n * sequential model's layers. Like the code snippet below shows.\n *\n * ```js\n * // Define a model which simply adds two inputs.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));\n * model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n * model1.summary();\n * model1.predict(tf.zeros([1, 4])).print();\n *\n * // Construct another model, reusing the second layer of `model1` while\n * // not using the first layer of `model1`. Note that you cannot add the second\n * // layer of `model` directly as the first layer of the new sequential model,\n * // because doing so will lead to an error related to the fact that the layer\n * // is not an input layer. Instead, you need to create an `inputLayer` and add\n * // it to the new sequential model before adding the reused layer.\n * const model2 = tf.sequential();\n * // Use an inputShape that matches the input shape of `model1`'s second\n * // layer.\n * model2.add(tf.layers.inputLayer({inputShape: [3]}));\n * model2.add(model1.layers[1]);\n * model2.summary();\n * model2.predict(tf.zeros([1, 3])).print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}\n */\nexport function inputLayer(args: InputLayerArgs): Layer {\n  return new InputLayer(args);\n}\n\n// Advanced Activation Layers.\n\n/**\n * Exponetial Linear Unit (ELU).\n *\n * It follows:\n * `f(x) =  alpha * (exp(x) - 1.) for x < 0`,\n * `f(x) = x for x >= 0`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Fast and Accurate Deep Network Learning by Exponential Linear Units\n * (ELUs)](https://arxiv.org/abs/1511.07289v1)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function elu(args?: ELULayerArgs): Layer {\n  return new ELU(args);\n}\n\n/**\n * Rectified Linear Unit activation function.\n *\n * Input shape:\n *   Arbitrary. Use the config field `inputShape` (Array of integers, does\n *   not include the sample axis) when using this layer as the first layer\n *   in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function reLU(args?: ReLULayerArgs): Layer {\n  return new ReLU(args);\n}\n\n/**\n * Leaky version of a rectified linear unit.\n *\n * It allows a small gradient when the unit is not active:\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function leakyReLU(args?: LeakyReLULayerArgs): Layer {\n  return new LeakyReLU(args);\n}\n\n/**\n * Parameterized version of a leaky rectified linear unit.\n *\n * It follows\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n * wherein `alpha` is a trainable weight.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function prelu(args?: PReLULayerArgs): Layer {\n  return new PReLU(args);\n}\n\n/**\n * Softmax activation layer.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function softmax(args?: SoftmaxLayerArgs): Layer {\n  return new Softmax(args);\n}\n\n/**\n * Thresholded Rectified Linear Unit.\n *\n * It follows:\n * `f(x) = x for x > theta`,\n * `f(x) = 0 otherwise`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Zero-Bias Autoencoders and the Benefits of Co-Adapting\n * Features](http://arxiv.org/abs/1402.3337)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function thresholdedReLU(args?: ThresholdedReLULayerArgs): Layer {\n  return new ThresholdedReLU(args);\n}\n\n// Convolutional Layers.\n\n/**\n * 1D convolution layer (e.g., temporal convolution).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input over a single spatial (or temporal) dimension\n * to produce a tensor of outputs.\n *\n * If `use_bias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model, provide an\n * `inputShape` argument `Array` or `null`.\n *\n * For example, `inputShape` would be:\n * - `[10, 128]` for sequences of 10 vectors of 128-dimensional vectors\n * - `[null, 128]` for variable-length sequences of 128-dimensional vectors.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional',  namespace: 'layers'}\n */\nexport function conv1d(args: ConvLayerArgs): Layer {\n  return new Conv1D(args);\n}\n\n/**\n * 2D convolution layer (e.g. spatial convolution over images).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 3]` for 128x128 RGB pictures\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2d(args: ConvLayerArgs): Layer {\n  return new Conv2D(args);\n}\n\n/**\n * Transposed convolutional layer (sometimes called Deconvolution).\n *\n * The need for transposed convolutions generally arises\n * from the desire to use a transformation going in the opposite direction of\n * a normal convolution, i.e., from something that has the shape of the output\n * of some convolution to something that has the shape of its input while\n * maintaining a connectivity pattern that is compatible with said\n * convolution.\n *\n * When using this layer as the first layer in a model, provide the\n * configuration `inputShape` (`Array` of integers, does not include the\n * sample axis), e.g., `inputShape: [128, 128, 3]` for 128x128 RGB pictures in\n * `dataFormat: 'channelsLast'`.\n *\n * Input shape:\n *   4D tensor with shape:\n *   `[batch, channels, rows, cols]` if `dataFormat` is `'channelsFirst'`.\n *   or 4D tensor with shape\n *   `[batch, rows, cols, channels]` if `dataFormat` is `'channelsLast`.\n *\n * Output shape:\n *   4D tensor with shape:\n *   `[batch, filters, newRows, newCols]` if `dataFormat` is\n * `'channelsFirst'`. or 4D tensor with shape:\n *   `[batch, newRows, newCols, filters]` if `dataFormat` is `'channelsLast'`.\n *\n * References:\n *   - [A guide to convolution arithmetic for deep\n * learning](https://arxiv.org/abs/1603.07285v1)\n *   - [Deconvolutional\n * Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2dTranspose(args: ConvLayerArgs): Layer {\n  return new Conv2DTranspose(args);\n}\n\n/**\n * 3D convolution layer (e.g. spatial convolution over volumes).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 128, 1]` for 128x128x128 grayscale volumes\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv3d(args: ConvLayerArgs): Layer {\n  return new Conv3D(args);\n}\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Separable convolution consists of first performing\n * a depthwise spatial convolution\n * (which acts on each input channel separately)\n * followed by a pointwise convolution which mixes together the resulting\n * output channels. The `depthMultiplier` argument controls how many\n * output channels are generated per input channel in the depthwise step.\n *\n * Intuitively, separable convolutions can be understood as\n * a way to factorize a convolution kernel into two smaller kernels,\n * or as an extreme version of an Inception block.\n *\n * Input shape:\n *   4D tensor with shape:\n *     `[batch, channels, rows, cols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, rows, cols, channels]` if data_format='channelsLast'.\n *\n * Output shape:\n *   4D tensor with shape:\n *     `[batch, filters, newRows, newCols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, newRows, newCols, filters]` if data_format='channelsLast'.\n *     `rows` and `cols` values might have changed due to padding.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function separableConv2d(args: SeparableConvLayerArgs): Layer {\n  return new SeparableConv2D(args);\n}\n\n/**\n * Cropping layer for 2D input (e.g., image).\n *\n * This layer can crop an input\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, croppedRows, croppedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, croppedRows, croppedCols]`.\n *\n * Examples\n * ```js\n *\n * const model = tf.sequential();\n * model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]],\n *                                inputShape: [128, 128, 3]}));\n * //now output shape is [batch, 124, 124, 3]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function cropping2D(args: Cropping2DLayerArgs): Layer {\n  return new Cropping2D(args);\n}\n\n/**\n * Upsampling layer for 2D inputs.\n *\n * Repeats the rows and columns of the data\n * by size[0] and size[1] respectively.\n *\n *\n * Input shape:\n *    4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *         `[batch, rows, cols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *        `[batch, channels, rows, cols]`\n *\n * Output shape:\n *     4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *        `[batch, upsampledRows, upsampledCols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *         `[batch, channels, upsampledRows, upsampledCols]`\n *\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function upSampling2d(args: UpSampling2DLayerArgs): Layer {\n  return new UpSampling2D(args);\n}\n\n// Convolutional(depthwise) Layers.\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Depthwise Separable convolutions consists in performing just the first step\n * in a depthwise spatial convolution (which acts on each input channel\n * separately). The `depthMultplier` argument controls how many output channels\n * are generated per input channel in the depthwise step.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function depthwiseConv2d(args: DepthwiseConv2DLayerArgs): Layer {\n  return new DepthwiseConv2D(args);\n}\n\n// Basic Layers.\n\n/**\n * Applies an activation function to an output.\n *\n * This layer applies element-wise activation function.  Other layers, notably\n * `dense` can also apply activation functions.  Use this isolated activation\n * function to extract the values before and after the\n * activation. For instance:\n *\n * ```js\n * const input = tf.input({shape: [5]});\n * const denseLayer = tf.layers.dense({units: 1});\n * const activationLayer = tf.layers.activation({activation: 'relu6'});\n *\n * // Obtain the output symbolic tensors by applying the layers in order.\n * const denseOutput = denseLayer.apply(input);\n * const activationOutput = activationLayer.apply(denseOutput);\n *\n * // Create the model based on the inputs.\n * const model = tf.model({\n *     inputs: input,\n *     outputs: [denseOutput, activationOutput]\n * });\n *\n * // Collect both outputs and print separately.\n * const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));\n * denseOut.print();\n * activationOut.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function activation(args: ActivationLayerArgs): Layer {\n  return new Activation(args);\n}\n\n/**\n * Creates a dense (fully connected) layer.\n *\n * This layer implements the operation:\n *   `output = activation(dot(input, kernel) + bias)`\n *\n * `activation` is the element-wise activation function\n *   passed as the `activation` argument.\n *\n * `kernel` is a weights matrix created by the layer.\n *\n * `bias` is a bias vector created by the layer (only applicable if `useBias`\n * is `true`).\n *\n * **Input shape:**\n *\n *   nD `tf.Tensor` with shape: `(batchSize, ..., inputDim)`.\n *\n *   The most common situation would be\n *   a 2D input with shape `(batchSize, inputDim)`.\n *\n * **Output shape:**\n *\n *   nD tensor with shape: `(batchSize, ..., units)`.\n *\n *   For instance, for a 2D input with shape `(batchSize, inputDim)`,\n *   the output would have shape `(batchSize, units)`.\n *\n * Note: if the input to the layer has a rank greater than 2, then it is\n * flattened prior to the initial dot product with the kernel.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dense(args: DenseLayerArgs): Layer {\n  return new Dense(args);\n}\n\n/**\n * Applies\n * [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to\n * the input.\n *\n * Dropout consists in randomly setting a fraction `rate` of input units to 0 at\n * each update during training time, which helps prevent overfitting.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dropout(args: DropoutLayerArgs): Layer {\n  return new Dropout(args);\n}\n\n/**\n * Spatial 1D version of Dropout.\n *\n * This Layer type performs the same function as the Dropout layer, but it drops\n * entire 1D feature maps instead of individual elements. For example, if an\n * input example consists of 3 timesteps and the feature map for each timestep\n * has a size of 4, a `spatialDropout1d` layer may zero out the feature maps\n * of the 1st timesteps and 2nd timesteps completely while sparing all feature\n * elements of the 3rd timestep.\n *\n * If adjacent frames (timesteps) are strongly correlated (as is normally the\n * case in early convolution layers), regular dropout will not regularize the\n * activation and will otherwise just result in merely an effective learning\n * rate decrease. In this case, `spatialDropout1d` will help promote\n * independence among feature maps and should be used instead.\n *\n * **Arguments:**\n *   rate: A floating-point number >=0 and <=1. Fraction of the input elements\n *     to drop.\n *\n * **Input shape:**\n *   3D tensor with shape `(samples, timesteps, channels)`.\n *\n * **Output shape:**\n *   Same as the input shape.\n *\n * References:\n *   - [Efficient Object Localization Using Convolutional\n *      Networks](https://arxiv.org/abs/1411.4280)\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function spatialDropout1d(args: SpatialDropout1DLayerConfig): Layer {\n  return new SpatialDropout1D(args);\n}\n\n/**\n * Flattens the input. Does not affect the batch size.\n *\n * A `Flatten` layer flattens each batch in its inputs to 1D (making the output\n * 2D).\n *\n * For example:\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const flattenLayer = tf.layers.flatten();\n * // Inspect the inferred output shape of the flatten layer, which\n * // equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the\n * // flattening. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(flattenLayer.apply(input).shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function flatten(args?: FlattenLayerArgs): Layer {\n  return new Flatten(args);\n}\n\n/**\n * Repeats the input n times in a new dimension.\n *\n * ```js\n *  const model = tf.sequential();\n *  model.add(tf.layers.repeatVector({n: 4, inputShape: [2]}));\n *  const x = tf.tensor2d([[10, 20]]);\n *  // Use the model to do inference on a data point the model hasn't see\n *  model.predict(x).print();\n *  // output shape is now [batch, 2, 4]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function repeatVector(args: RepeatVectorLayerArgs): Layer {\n  return new RepeatVector(args);\n}\n\n/**\n * Reshapes an input to a certain shape.\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const reshapeLayer = tf.layers.reshape({targetShape: [2, 6]});\n * // Inspect the inferred output shape of the Reshape layer, which\n * // equals `[null, 2, 6]`. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(reshapeLayer.apply(input).shape));\n * ```\n *\n * Input shape:\n *   Arbitrary, although all dimensions in the input shape must be fixed.\n *   Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n *\n * Output shape:\n *   [batchSize, targetShape[0], targetShape[1], ...,\n *    targetShape[targetShape.length - 1]].\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function reshape(args: ReshapeLayerArgs): Layer {\n  return new Reshape(args);\n}\n\n/**\n * Permutes the dimensions of the input according to a given pattern.\n *\n * Useful for, e.g., connecting RNNs and convnets together.\n *\n * Example:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.permute({\n *   dims: [2, 1],\n *   inputShape: [10, 64]\n * }));\n * console.log(model.outputShape);\n * // Now model's output shape is [null, 64, 10], where null is the\n * // unpermuted sample (batch) dimension.\n * ```\n *\n * Input shape:\n *   Arbitrary. Use the configuration field `inputShape` when using this\n *   layer as the first layer in a model.\n *\n * Output shape:\n *   Same rank as the input shape, but with the dimensions re-ordered (i.e.,\n *   permuted) according to the `dims` configuration of this layer.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function permute(args: PermuteLayerArgs): Layer {\n  return new Permute(args);\n}\n\n/**\n * Maps positive integers (indices) into dense vectors of fixed size.\n * eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]\n *\n * **Input shape:** 2D tensor with shape: `[batchSize, sequenceLength]`.\n *\n * **Output shape:** 3D tensor with shape: `[batchSize, sequenceLength,\n * outputDim]`.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function embedding(args: EmbeddingLayerArgs): Layer {\n  return new Embedding(args);\n}\n\n// Merge Layers.\n\n/**\n * Layer that performs element-wise addition on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). The inputs are specified as an\n * `Array` when the `apply` method of the `Add` layer instance is called. For\n * example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const addLayer = tf.layers.add();\n * const sum = addLayer.apply([input1, input2]);\n * console.log(JSON.stringify(sum.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function add(args?: LayerArgs): Layer {\n  return new Add(args);\n}\n\n/**\n * Layer that performs element-wise averaging on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const averageLayer = tf.layers.average();\n * const average = averageLayer.apply([input1, input2]);\n * console.log(JSON.stringify(average.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function average(args?: LayerArgs): Layer {\n  return new Average(args);\n}\n\n/**\n * Layer that concatenates an `Array` of inputs.\n *\n * It takes a list of tensors, all of the same shape except for the\n * concatenation axis, and returns a single tensor, the concatenation\n * of all inputs. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 3]});\n * const concatLayer = tf.layers.concatenate();\n * const output = concatLayer.apply([input1, input2]);\n * console.log(JSON.stringify(output.shape));\n * // You get [null, 2, 5], with the first dimension as the undetermined batch\n * // dimension. The last dimension (5) is the result of concatenating the\n * // last dimensions of the inputs (2 and 3).\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function concatenate(args?: ConcatenateLayerArgs): Layer {\n  return new Concatenate(args);\n}\n\n/**\n * Layer that computes the element-wise maximum an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const maxLayer = tf.layers.maximum();\n * const max = maxLayer.apply([input1, input2]);\n * console.log(JSON.stringify(max.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function maximum(args?: LayerArgs): Layer {\n  return new Maximum(args);\n}\n\n/**\n * Layer that computes the element-wise minimum of an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const minLayer = tf.layers.minimum();\n * const min = minLayer.apply([input1, input2]);\n * console.log(JSON.stringify(min.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function minimum(args?: LayerArgs): Layer {\n  return new Minimum(args);\n}\n\n/**\n * Layer that multiplies (element-wise) an `Array` of inputs.\n *\n * It takes as input an Array of tensors, all of the same\n * shape, and returns a single tensor (also of the same shape).\n * For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const input3 = tf.input({shape: [2, 2]});\n * const multiplyLayer = tf.layers.multiply();\n * const product = multiplyLayer.apply([input1, input2, input3]);\n * console.log(product.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function multiply(args?: LayerArgs): Layer {\n  return new Multiply(args);\n}\n\n/**\n * Layer that computes a dot product between samples in two tensors.\n *\n * E.g., if applied to a list of two tensors `a` and `b` both of shape\n * `[batchSize, n]`, the output will be a tensor of shape `[batchSize, 1]`,\n * where each entry at index `[i, 0]` will be the dot product between\n * `a[i, :]` and `b[i, :]`.\n *\n * Example:\n *\n * ```js\n * const dotLayer = tf.layers.dot({axes: -1});\n * const x1 = tf.tensor2d([[10, 20], [30, 40]]);\n * const x2 = tf.tensor2d([[-1, -2], [-3, -4]]);\n *\n * // Invoke the layer's apply() method in eager (imperative) mode.\n * const y = dotLayer.apply([x1, x2]);\n * y.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function dot(args: DotLayerArgs): Layer {\n  return new Dot(args);\n}\n\n// Normalization Layers.\n\n/**\n * Batch normalization layer (Ioffe and Szegedy, 2014).\n *\n * Normalize the activations of the previous layer at each batch,\n * i.e. applies a transformation that maintains the mean activation\n * close to 0 and the activation standard deviation close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape` (Array of integers, does\n *   not include the sample axis) when calling the constructor of this class,\n *   if this layer is used as a first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Batch Normalization: Accelerating Deep Network Training by Reducing\n * Internal Covariate Shift](https://arxiv.org/abs/1502.03167)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function batchNormalization(args?: BatchNormalizationLayerArgs): Layer {\n  return new BatchNormalization(args);\n}\n\n/**\n * Layer-normalization layer (Ba et al., 2016).\n *\n * Normalizes the activations of the previous layer for each given example in a\n * batch independently, instead of across a batch like in `batchNormalization`.\n * In other words, this layer applies a transformation that maintanis the mean\n * activation within each example close to0 and activation variance close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the argument `inputShape` when using this layer as the first\n *   layer in a model.\n *\n * Output shape:\n *   Same as input.\n *\n * References:\n *   - [Layer Normalization](https://arxiv.org/abs/1607.06450)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function layerNormalization(args?: LayerNormalizationLayerArgs): Layer {\n  return new LayerNormalization(args);\n}\n\n// Padding Layers.\n\n/**\n * Zero-padding layer for 2D input (e.g., image).\n *\n * This layer can add rows and columns of zeros\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, paddedRows, paddedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, paddedRows, paddedCols]`.\n *\n * @doc {heading: 'Layers', subheading: 'Padding', namespace: 'layers'}\n */\nexport function zeroPadding2d(args?: ZeroPadding2DLayerArgs): Layer {\n  return new ZeroPadding2D(args);\n}\n\n// Pooling Layers.\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape: `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * `tf.avgPool1d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling1d(args: Pooling1DLayerArgs): Layer {\n  return new AveragePooling1D(args);\n}\nexport function avgPool1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape:\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, rows, cols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, pooleRows, pooledCols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, pooleRows, pooledCols]`\n *\n * `tf.avgPool2d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling2d(args: Pooling2DLayerArgs): Layer {\n  return new AveragePooling2D(args);\n}\nexport function avgPool2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n\n/**\n * Average pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      4D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling3d(args: Pooling3DLayerArgs): Layer {\n  return new AveragePooling3D(args);\n}\nexport function avgPool3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n\n/**\n * Global average pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling1d(args?: LayerArgs): Layer {\n  return new GlobalAveragePooling1D(args);\n}\n\n/**\n * Global average pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalAveragePooling2D(args);\n}\n\n/**\n * Global max pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling1d(args?: LayerArgs): Layer {\n  return new GlobalMaxPooling1D(args);\n}\n\n/**\n * Global max pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalMaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for temporal data.\n *\n * Input shape:  `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling1d(args: Pooling1DLayerArgs): Layer {\n  return new MaxPooling1D(args);\n}\n\n/**\n * Max pooling operation for spatial data.\n *\n * Input shape\n *   - If `dataFormat === CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, rows, cols, channels]`\n *   - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *       `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, pooleRows, pooledCols, channels]`\n *   - If `dataFormat=CHANNEL_FIRST`:\n *       4D tensor with shape:\n *       `[batchSize, channels, pooleRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling2d(args: Pooling2DLayerArgs): Layer {\n  return new MaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      5D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling3d(args: Pooling3DLayerArgs): Layer {\n  return new MaxPooling3D(args);\n}\n\n// Recurrent Layers.\n\n/**\n * Gated Recurrent Unit - Cho et al. 2014.\n *\n * This is an `RNN` layer consisting of one `GRUCell`. However, unlike\n * the underlying `GRUCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.gru({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `GRUCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gru(args: GRULayerArgs): Layer {\n  return new GRU(args);\n}\n\n/**\n * Cell class for `GRU`.\n *\n * `GRUCell` is distinct from the `RNN` subclass `GRU` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `GRU` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.gruCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `GRUCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.gruCell({units: 4}),\n *   tf.layers.gruCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `gruCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `GRUCell`, use the\n * `tf.layers.gru`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gruCell(args: GRUCellLayerArgs): RNNCell {\n  return new GRUCell(args);\n}\n\n/**\n * Long-Short Term Memory layer - Hochreiter 1997.\n *\n * This is an `RNN` layer consisting of one `LSTMCell`. However, unlike\n * the underlying `LSTMCell`, the `apply` method of `LSTM` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const lstm = tf.layers.lstm({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = lstm.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `LSTMCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstm(args: LSTMLayerArgs): Layer {\n  return new LSTM(args);\n}\n\n/**\n * Cell class for `LSTM`.\n *\n * `LSTMCell` is distinct from the `RNN` subclass `LSTM` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `LSTM` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.lstmCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `LSTMCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.lstmCell({units: 4}),\n *   tf.layers.lstmCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `lstmCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `LSTMCell`, use the\n * `tf.layers.lstm`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstmCell(args: LSTMCellLayerArgs): RNNCell {\n  return new LSTMCell(args);\n}\n\n/**\n * Fully-connected RNN where the output is to be fed back to input.\n *\n * This is an `RNN` layer consisting of one `SimpleRNNCell`. However, unlike\n * the underlying `SimpleRNNCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.simpleRNN({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `SimpleRNNCell`'s number of units.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNN(args: SimpleRNNLayerArgs): Layer {\n  return new SimpleRNN(args);\n}\n\n/**\n * Cell class for `SimpleRNN`.\n *\n * `SimpleRNNCell` is distinct from the `RNN` subclass `SimpleRNN` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `SimpleRNN` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.simpleRNNCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `SimpleRNNCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.simpleRNNCell({units: 4}),\n *   tf.layers.simpleRNNCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `SimpleRNNCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `SimpleRNNCell`, use the\n * `tf.layers.simpleRNN`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNNCell(args: SimpleRNNCellLayerArgs): RNNCell {\n  return new SimpleRNNCell(args);\n}\n\n/**\n * Convolutional LSTM layer - Xingjian Shi 2015.\n *\n * This is an `ConvRNN2D` layer consisting of one `ConvLSTM2DCell`. However,\n * unlike the underlying `ConvLSTM2DCell`, the `apply` method of `ConvLSTM2D`\n * operates on a sequence of inputs. The shape of the input (not including the\n * first, batch dimension) needs to be 4-D, with the first dimension being time\n * steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const batchSize = 4;\n * const sequenceLength = 2;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [batchSize, sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const layer = tf.layers.convLstm2d({filters, kernelSize});\n *\n * const output = layer.apply(input);\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2d(args: ConvLSTM2DArgs): ConvLSTM2D {\n  return new ConvLSTM2D(args);\n}\n\n/**\n * Cell class for `ConvLSTM2D`.\n *\n * `ConvLSTM2DCell` is distinct from the `ConvRNN2D` subclass `ConvLSTM2D` in\n * that its `call` method takes the input data of only a single time step and\n * returns the cell's output at the time step, while `ConvLSTM2D` takes the\n * input data over a number of time steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const sequenceLength = 1;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const cell = tf.layers.convLstm2dCell({filters, kernelSize});\n *\n * cell.build(input.shape);\n *\n * const outputSize = size - kernelSize + 1;\n * const outShape = [sequenceLength, outputSize, outputSize, filters];\n *\n * const initialH = tf.zeros(outShape);\n * const initialC = tf.zeros(outShape);\n *\n * const [o, h, c] = cell.call([input, initialH, initialC], {});\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2dCell(args: ConvLSTM2DCellArgs): ConvLSTM2DCell {\n  return new ConvLSTM2DCell(args);\n}\n\n/**\n * Base class for recurrent layers.\n *\n * Input shape:\n *   3D tensor with shape `[batchSize, timeSteps, inputDim]`.\n *\n * Output shape:\n *   - if `returnState`, an Array of tensors (i.e., `tf.Tensor`s). The first\n *     tensor is the output. The remaining tensors are the states at the\n *     last time step, each with shape `[batchSize, units]`.\n *   - if `returnSequences`, the output will have shape\n *     `[batchSize, timeSteps, units]`.\n *   - else, the output will have shape `[batchSize, units]`.\n *\n * Masking:\n *   This layer supports masking for input data with a variable number\n *   of timesteps. To introduce masks to your data,\n *   use an embedding layer with the `mask_zero` parameter\n *   set to `True`.\n *\n * Notes on using statefulness in RNNs:\n *   You can set RNN layers to be 'stateful', which means that the states\n *   computed for the samples in one batch will be reused as initial states\n *   for the samples in the next batch. This assumes a one-to-one mapping\n *   between samples in different successive batches.\n *\n *   To enable statefulness:\n *     - specify `stateful: true` in the layer constructor.\n *     - specify a fixed batch size for your model, by passing\n *       if sequential model:\n *         `batchInputShape=[...]` to the first layer in your model.\n *       else for functional model with 1 or more Input layers:\n *         `batchShape=[...]` to all the first layers in your model.\n *       This is the expected shape of your inputs *including the batch size*.\n *       It should be a tuple of integers, e.g. `(32, 10, 100)`.\n *     - specify `shuffle=False` when calling fit().\n *\n *   To reset the states of your model, call `.resetStates()` on either\n *   a specific layer, or on your entire model.\n *\n * Note on specifying the initial state of RNNs\n *   You can specify the initial state of RNN layers symbolically by\n *   calling them with the option `initialState`. The value of\n *   `initialState` should be a tensor or list of tensors representing\n *   the initial state of the RNN layer.\n *\n *   You can specify the initial state of RNN layers numerically by\n *   calling `resetStates` with the keyword argument `states`. The value of\n *   `states` should be a numpy array or list of numpy arrays representing\n *   the initial state of the RNN layer.\n *\n * Note on passing external constants to RNNs\n *   You can pass \"external\" constants to the cell using the `constants`\n *   keyword argument of `RNN.call` method. This requires that the `cell.call`\n *   method accepts the same keyword argument `constants`. Such constants\n *   can be used to conditon the cell transformation on additional static inputs\n *   (not changing over time), a.k.a an attention mechanism.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function rnn(args: RNNLayerArgs): Layer {\n  return new RNN(args);\n}\n\n/**\n * Wrapper allowing a stack of RNN cells to behave as a single cell.\n *\n * Used to implement efficient stacked RNNs.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function stackedRNNCells(args: StackedRNNCellsArgs): RNNCell {\n  return new StackedRNNCells(args);\n}\n\n// Wrapper Layers.\n\n/** @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'} */\nexport function bidirectional(args: BidirectionalLayerArgs): Bidirectional {\n  return new Bidirectional(args);\n}\n\n/**\n * This wrapper applies a layer to every temporal slice of an input.\n *\n * The input should be at least 3D,  and the dimension of the index `1` will be\n * considered to be the temporal dimension.\n *\n * Consider a batch of 32 samples, where each sample is a sequence of 10 vectors\n * of 16 dimensions. The batch input shape of the layer is then `[32,  10,\n * 16]`, and the `inputShape`, not including the sample dimension, is\n * `[10, 16]`.\n *\n * You can then use `TimeDistributed` to apply a `Dense` layer to each of the 10\n * timesteps, independently:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.dense({units: 8}),\n *   inputShape: [10, 16],\n * }));\n *\n * // Now model.outputShape = [null, 10, 8].\n * // The output will then have shape `[32, 10, 8]`.\n *\n * // In subsequent layers, there is no need for `inputShape`:\n * model.add(tf.layers.timeDistributed({layer: tf.layers.dense({units: 32})}));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * // Now model.outputShape = [null, 10, 32].\n * ```\n *\n * The output will then have shape `[32, 10, 32]`.\n *\n * `TimeDistributed` can be used with arbitrary layers, not just `Dense`, for\n * instance a `Conv2D` layer.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.conv2d({filters: 64, kernelSize: [3, 3]}),\n *   inputShape: [10, 299, 299, 3],\n * }));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}\n */\nexport function timeDistributed(args: WrapperLayerArgs): Layer {\n  return new TimeDistributed(args);\n}\n\n// Aliases for pooling.\nexport const globalMaxPool1d = globalMaxPooling1d;\nexport const globalMaxPool2d = globalMaxPooling2d;\nexport const maxPool1d = maxPooling1d;\nexport const maxPool2d = maxPooling2d;\n\nexport {Layer, RNN, RNNCell, input /* alias for tf.input */};\n\n/**\n * Apply additive zero-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * This is useful to mitigate overfitting\n * (you could see it as a form of random data augmentation).\n * Gaussian Noise (GS) is a natural choice as corruption process\n * for real valued inputs.\n *\n * # Arguments\n *     stddev: float, standard deviation of the noise distribution.\n *\n * # Input shape\n *         Arbitrary. Use the keyword argument `input_shape`\n *         (tuple of integers, does not include the samples axis)\n *         when using this layer as the first layer in a model.\n *\n * # Output shape\n *         Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianNoise(args: GaussianNoiseArgs): GaussianNoise {\n  return new GaussianNoise(args);\n}\n\n/**\n * Apply multiplicative 1-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting](\n *      http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianDropout(args: GaussianDropoutArgs): GaussianDropout {\n  return new GaussianDropout(args);\n}\n\n/**\n * Applies Alpha Dropout to the input.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\n * to their original values, in order to ensure the self-normalizing property\n * even after this dropout.\n * Alpha Dropout fits well to Scaled Exponential Linear Units\n * by randomly setting activations to the negative saturation value.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\n *     shape for randomly generated keep/drop flags.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function alphaDropout(args: AlphaDropoutArgs): AlphaDropout {\n  return new AlphaDropout(args);\n}\n\n/**\n * Masks a sequence by using a mask value to skip timesteps.\n *\n * If all features for a given sample timestep are equal to `mask_value`,\n * then the sample timestep will be masked (skipped) in all downstream layers\n * (as long as they support masking).\n *\n * If any downstream layer does not support masking yet receives such\n * an input mask, an exception will be raised.\n *\n * Arguments:\n *   - `maskValue`: Either None or mask value to skip.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Mask', namespace: 'layers'}\n */\nexport function masking(args?: MaskingArgs): Layer {\n  return new Masking(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport * as losses from './losses';\nimport * as metrics from './metrics';\n\n/**\n * Binary accuracy metric function.\n *\n * `yTrue` and `yPred` can have 0-1 values. Example:\n * ```js\n * const x = tf.tensor2d([[1, 1, 1, 1], [0, 0, 0, 0]], [2, 4]);\n * const y = tf.tensor2d([[1, 0, 1, 0], [0, 0, 0, 1]], [2, 4]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * `yTrue` and `yPred` can also have floating-number values between 0 and 1, in\n * which case the values will be thresholded at 0.5 to yield 0-1 values (i.e.,\n * a value >= 0.5 and <= 1.0 is interpreted as 1.\n * )\n * Example:\n * ```js\n * const x = tf.tensor1d([1, 1, 1, 1, 0, 0, 0, 0]);\n * const y = tf.tensor1d([0.2, 0.4, 0.6, 0.8, 0.2, 0.3, 0.4, 0.7]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryAccuracy(yTrue, yPred);\n}\n\n/**\n * Binary crossentropy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0], [1], [1], [1]]);\n * const y = tf.tensor2d([[0], [0], [0.5], [1]]);\n * const crossentropy = tf.metrics.binaryCrossentropy(x, y);\n * crossentropy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction, probabilities for the `1` case.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryCrossentropy(yTrue, yPred);\n}\n\n/**\n * Sparse categorical accuracy metric function.\n *\n * Example:\n * ```js\n *\n * const yTrue = tf.tensor1d([1, 1, 2, 2, 0]);\n * const yPred = tf.tensor2d(\n *      [[0, 1, 0], [1, 0, 0], [0, 0.4, 0.6], [0, 0.6, 0.4], [0.7, 0.3, 0]]);\n * const crossentropy = tf.metrics.sparseCategoricalAccuracy(yTrue, yPred);\n * crossentropy.print();\n * ```\n *\n * @param yTrue True labels: indices.\n * @param yPred Predicted probabilities or logits.\n * @returns Accuracy tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.sparseCategoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical accuracy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0, 0, 0, 1], [0, 0, 0, 1]]);\n * const y = tf.tensor2d([[0.1, 0.8, 0.05, 0.05], [0.1, 0.05, 0.05, 0.8]]);\n * const accuracy = tf.metrics.categoricalAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth: one-hot encoding of categories.\n * @param yPred Binary Tensor of prediction: probabilities or logits for the\n *   same categories as in `yTrue`.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical crossentropy between an output tensor and a target tensor.\n *\n * @param target A tensor of the same shape as `output`.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalCrossentropy(yTrue, yPred);\n}\n\n/**\n * Computes the precision of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const precision = tf.metrics.precision(x, y);\n * precision.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Precision Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.precision(yTrue, yPred);\n}\n\n/**\n * Computes the recall of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const recall = tf.metrics.recall(x, y);\n * recall.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Recall Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.recall(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Cosine proximity.\n *\n * Mathematically, cosine proximity is defined as:\n *   `-sum(l2Normalize(yTrue) * l2Normalize(yPred))`,\n * wherein `l2Normalize()` normalizes the L2 norm of the input to 1 and `*`\n * represents element-wise multiplication.\n *\n * ```js\n * const yTrue = tf.tensor2d([[1, 0], [1, 0]]);\n * const yPred = tf.tensor2d([[1 / Math.sqrt(2), 1 / Math.sqrt(2)], [0, 1]]);\n * const proximity = tf.metrics.cosineProximity(yTrue, yPred);\n * proximity.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Cosine proximity Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.cosineProximity(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute error.\n *\n * Mathematically, mean absolute error is defined as:\n *   `mean(abs(yPred - yTrue))`,\n * wherein the `mean` is applied over feature dimensions.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [0, 0], [2, 3]]);\n * const yPred = tf.tensor2d([[0, 1], [0, 1], [-2, -3]]);\n * const mse = tf.metrics.meanAbsoluteError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsoluteError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute percentage error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [10, 20]]);\n * const yPred = tf.tensor2d([[0, 1], [11, 24]]);\n * const mse = tf.metrics.meanAbsolutePercentageError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MAPE`, `tf.metrics.mape`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute percentage error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function MAPE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function mape(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean squared error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [3, 4]]);\n * const yPred = tf.tensor2d([[0, 1], [-3, -4]]);\n * const mse = tf.metrics.meanSquaredError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MSE`, `tf.metrics.mse`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean squared error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function MSE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function mse(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nexport {modelFromJSON} from './models';\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport * as regularizers from './regularizers';\n// tslint:disable-next-line:max-line-length\nimport {L1Args, L1L2, L1L2Args, L2Args, Regularizer} from './regularizers';\n\n/**\n * Regularizer for L1 and L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x)) + sum(l2 * x^2)\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1l2(config?: L1L2Args): Regularizer {\n  return new L1L2(config);\n}\n\n/**\n * Regularizer for L1 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x))\n * @param args l1 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1(config?: L1Args): Regularizer {\n  return regularizers.l1(config);\n}\n\n/**\n * Regularizer for L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l2 * x^2)\n * @param args l2 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l2(config?: L2Args): Regularizer {\n  return regularizers.l2(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {BaseCallback} from './base_callbacks';\nimport {Container} from './engine/container';\nimport {LayersModel} from './engine/training';\nimport {NotImplementedError} from './errors';\nimport {Logs, resolveScalarsInLogs} from './logs';\n\nexport abstract class Callback extends BaseCallback {\n  /** Instance of `keras.models.Model`. Reference of the model being trained. */\n  model: LayersModel = null;\n\n  setModel(model: Container): void {\n    if (!(model instanceof LayersModel)) {\n      throw new Error('model must be a LayersModel, not some other Container');\n    }\n    this.model = model;\n  }\n}\n\nexport interface EarlyStoppingCallbackArgs {\n  /**\n   * Quantity to be monitored.\n   *\n   * Defaults to 'val_loss'.\n   */\n  monitor?: string;\n\n  /**\n   * Minimum change in the monitored quantity to qualify as improvement,\n   * i.e., an absolute change of less than `minDelta` will count as no\n   * improvement.\n   *\n   * Defaults to 0.\n   */\n  minDelta?: number;\n\n  /**\n   * Number of epochs with no improvement after which training will be stopped.\n   *\n   * Defaults to 0.\n   */\n  patience?: number;\n\n  /** Verbosity mode. */\n  verbose?: number;\n\n  /**\n   * Mode: one of 'min', 'max', and 'auto'.\n   * - In 'min' mode, training will be stopped when the quantity monitored has\n   *   stopped decreasing.\n   * - In 'max' mode, training will be stopped when the quantity monitored has\n   *   stopped increasing.\n   * - In 'auto' mode, the direction is inferred automatically from the name of\n   *   the monitored quantity.\n   *\n   * Defaults to 'auto'.\n   */\n  mode?: 'auto'|'min'|'max';\n\n  /**\n   * Baseline value of the monitored quantity.\n   *\n   * If specified, training will be stopped if the model doesn't show\n   * improvement over the baseline.\n   */\n  baseline?: number;\n\n  /**\n   * Whether to restore model weights from the epoch with the best value\n   * of the monitored quantity. If `False`, the model weights obtained at the\n   * at the last step of training are used.\n   *\n   * **`True` is not supported yet.**\n   */\n  restoreBestWeights?: boolean;\n}\n\nfunction less(currVal: number, prevVal: number) {\n  return currVal < prevVal;\n}\n\nfunction greater(currVal: number, prevVal: number) {\n  return currVal > prevVal;\n}\n\n/**\n * A Callback that stops training when a monitored quantity has stopped\n * improving.\n */\nexport class EarlyStopping extends Callback {\n  protected readonly monitor: string;\n  protected readonly minDelta: number;\n  protected readonly patience: number;\n  protected readonly baseline: number;\n  protected readonly verbose: number;\n  protected readonly mode: 'auto'|'min'|'max';\n\n  protected monitorFunc: (currVal: number, prevVal: number) => boolean;\n\n  private wait: number;\n  private stoppedEpoch: number;\n  private best: number;\n\n  constructor(args?: EarlyStoppingCallbackArgs) {\n    super();\n    if (args == null) {\n      args = {};\n    }\n    if (args.restoreBestWeights) {\n      throw new NotImplementedError(\n          'restoreBestWeights = True is not implemented in EarlyStopping yet.');\n    }\n\n    this.monitor = args.monitor || 'val_loss';\n    this.minDelta = Math.abs(args.minDelta || 0);\n    this.patience = args.patience || 0;\n    this.verbose = args.verbose || 0;\n    this.mode = args.mode || 'auto';\n    this.baseline = args.baseline;\n\n    if (['auto', 'min', 'max'].indexOf(this.mode) === -1) {\n      console.warn(\n          `EarlyStopping mode '${this.mode}' is invalid. ` +\n          `Falling back to mode 'auto'.`);\n      this.mode = 'auto';\n    }\n\n    if (this.mode === 'min') {\n      this.monitorFunc = less;\n    } else if (this.mode === 'max') {\n      this.monitorFunc = greater;\n    } else {\n      // For mode === 'auto'.\n      if (this.monitor.indexOf('acc') !== -1) {\n        this.monitorFunc = greater;\n      } else {\n        this.monitorFunc = less;\n      }\n    }\n\n    if (this.monitorFunc === less) {\n      this.minDelta *= -1;\n    }\n  }\n\n  async onTrainBegin(logs?: Logs) {\n    this.wait = 0;\n    this.stoppedEpoch = 0;\n    if (this.baseline != null) {\n      this.best = this.baseline;\n    } else {\n      this.best = this.monitorFunc === less ? Infinity : -Infinity;\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: Logs) {\n    await resolveScalarsInLogs(logs);\n    const current = this.getMonitorValue(logs);\n    if (current == null) {\n      return;\n    }\n\n    if (this.monitorFunc(current - this.minDelta, this.best)) {\n      this.best = current;\n      this.wait = 0;\n      // TODO(cais): Logic for restoreBestWeights.\n    } else {\n      this.wait++;\n      if (this.wait >= this.patience) {\n        this.stoppedEpoch = epoch;\n        this.model.stopTraining = true;\n      }\n      // TODO(cais): Logic for restoreBestWeights.\n    }\n  }\n\n  async onTrainEnd(logs?: Logs) {\n    if (this.stoppedEpoch > 0 && this.verbose) {\n      console.log(`Epoch ${this.stoppedEpoch}: early stopping.`);\n    }\n  }\n\n  private getMonitorValue(logs: Logs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const monitorValue = logs[this.monitor];\n    if (monitorValue == null) {\n      console.warn(\n          `Metric for EarlyStopping ${this.monitor} is not available. ` +\n          `Available metrics are: ${Object.keys(logs)}`);\n    }\n    return monitorValue;\n  }\n}\n\n/**\n * Factory function for a Callback that stops training when a monitored\n * quantity has stopped improving.\n *\n * Early stopping is a type of regularization, and protects model against\n * overfitting.\n *\n * The following example based on fake data illustrates how this callback\n * can be used during `tf.LayersModel.fit()`:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense({\n *   units: 3,\n *   activation: 'softmax',\n *   kernelInitializer: 'ones',\n *   inputShape: [2]\n * }));\n * const xs = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const ys = tf.tensor2d([[1, 0, 0], [0, 1, 0]], [2, 3]);\n * const xsVal = tf.tensor2d([4, 3, 2, 1], [2, 2]);\n * const ysVal = tf.tensor2d([[0, 0, 1], [0, 1, 0]], [2, 3]);\n * model.compile(\n *     {loss: 'categoricalCrossentropy', optimizer: 'sgd', metrics: ['acc']});\n *\n * // Without the EarlyStopping callback, the val_acc value would be:\n * //   0.5, 0.5, 0.5, 0.5, ...\n * // With val_acc being monitored, training should stop after the 2nd epoch.\n * const history = await model.fit(xs, ys, {\n *   epochs: 10,\n *   validationData: [xsVal, ysVal],\n *   callbacks: tf.callbacks.earlyStopping({monitor: 'val_acc'})\n * });\n *\n * // Expect to see a length-2 array.\n * console.log(history.history.val_acc);\n * ```\n *\n * @doc {\n *   heading: 'Callbacks',\n *   namespace: 'callbacks'\n * }\n */\nexport function earlyStopping(args?: EarlyStoppingCallbackArgs) {\n  return new EarlyStopping(args);\n}\n\nexport const callbacks = {earlyStopping};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/* tslint:disable */\n\n/** Properties of an Any. */\nexport declare interface IAny {\n  /** Any typeUrl */\n  typeUrl?: (string|null);\n\n  /** Any value */\n  value?: (Uint8Array|null);\n}\n\n/** DataType enum. */\nexport enum DataType {\n  'DT_INVALID' = 0,\n  'DT_FLOAT' = 1,\n  'DT_DOUBLE' = 2,\n  'DT_INT32' = 3,\n  'DT_UINT8' = 4,\n  'DT_INT16' = 5,\n  'DT_INT8' = 6,\n  'DT_STRING' = 7,\n  'DT_COMPLEX64' = 8,\n  'DT_INT64' = 9,\n  'DT_BOOL' = 10,\n  'DT_QINT8' = 11,\n  'DT_QUINT8' = 12,\n  'DT_QINT32' = 13,\n  'DT_BFLOAT16' = 14,\n  'DT_FLOAT_REF' = 101,\n  'DT_DOUBLE_REF' = 102,\n  'DT_INT32_REF' = 103,\n  'DT_UINT8_REF' = 104,\n  'DT_INT16_REF' = 105,\n  'DT_INT8_REF' = 106,\n  'DT_STRING_REF' = 107,\n  'DT_COMPLEX64_REF' = 108,\n  'DT_INT64_REF' = 109,\n  'DT_BOOL_REF' = 110,\n  'DT_QINT8_REF' = 111,\n  'DT_QUINT8_REF' = 112,\n  'DT_QINT32_REF' = 113,\n  'DT_BFLOAT16_REF' = 114\n}\n\n/** Properties of a TensorShape. */\nexport declare interface ITensorShape {\n  /** TensorShape dim */\n  dim?: (TensorShape.IDim[]|null);\n\n  /** TensorShape unknownRank */\n  unknownRank?: (boolean|null);\n}\n\nexport namespace TensorShape {\n  /** Properties of a Dim. */\n  export declare interface IDim {\n    /** Dim size */\n    size?: (number|string|null);\n\n    /** Dim name */\n    name?: (string|null);\n  }\n}\n\n/** Properties of a Tensor. */\nexport declare interface ITensor {\n  /** Tensor dtype */\n  dtype?: (DataType|null);\n\n  /** Tensor tensorShape */\n  tensorShape?: (ITensorShape|null);\n\n  /** Tensor versionNumber */\n  versionNumber?: (number|null);\n\n  /** Tensor tensorContent */\n  tensorContent?: (Uint8Array|null);\n\n  /** Tensor floatVal */\n  floatVal?: (number[]|null);\n\n  /** Tensor doubleVal */\n  doubleVal?: (number[]|null);\n\n  /** Tensor intVal */\n  intVal?: (number[]|null);\n\n  /** Tensor stringVal */\n  stringVal?: (Uint8Array[]|null);\n\n  /** Tensor scomplexVal */\n  scomplexVal?: (number[]|null);\n\n  /** Tensor int64Val */\n  int64Val?: ((number | string)[]|null);\n\n  /** Tensor boolVal */\n  boolVal?: (boolean[]|null);\n\n  /** Tensor uint32Val */\n  uint32Val?: (number[]|null);\n\n  /** Tensor uint64Val */\n  uint64Val?: ((number | string)[]|null);\n}\n\n/** Properties of an AttrValue. */\nexport declare interface IAttrValue {\n  /** AttrValue list */\n  list?: (AttrValue.IListValue|null);\n\n  /** AttrValue s */\n  s?: (string|null);\n\n  /** AttrValue i */\n  i?: (number|string|null);\n\n  /** AttrValue f */\n  f?: (number|null);\n\n  /** AttrValue b */\n  b?: (boolean|null);\n\n  /** AttrValue type */\n  type?: (DataType|null);\n\n  /** AttrValue shape */\n  shape?: (ITensorShape|null);\n\n  /** AttrValue tensor */\n  tensor?: (ITensor|null);\n\n  /** AttrValue placeholder */\n  placeholder?: (string|null);\n\n  /** AttrValue func */\n  func?: (INameAttrList|null);\n}\n\nexport namespace AttrValue {\n  /** Properties of a ListValue. */\n  export declare interface IListValue {\n    /** ListValue s */\n    s?: (string[]|null);\n\n    /** ListValue i */\n    i?: ((number | string)[]|null);\n\n    /** ListValue f */\n    f?: (number[]|null);\n\n    /** ListValue b */\n    b?: (boolean[]|null);\n\n    /** ListValue type */\n    type?: (DataType[]|null);\n\n    /** ListValue shape */\n    shape?: (ITensorShape[]|null);\n\n    /** ListValue tensor */\n    tensor?: (ITensor[]|null);\n\n    /** ListValue func */\n    func?: (INameAttrList[]|null);\n  }\n}\n\n/** Properties of a NameAttrList. */\nexport declare interface INameAttrList {\n  /** NameAttrList name */\n  name?: (string|null);\n\n  /** NameAttrList attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a NodeDef. */\nexport declare interface INodeDef {\n  /** NodeDef name */\n  name?: (string|null);\n\n  /** NodeDef op */\n  op?: (string|null);\n\n  /** NodeDef input */\n  input?: (string[]|null);\n\n  /** NodeDef device */\n  device?: (string|null);\n\n  /** NodeDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a VersionDef. */\nexport declare interface IVersionDef {\n  /** VersionDef producer */\n  producer?: (number|null);\n\n  /** VersionDef minConsumer */\n  minConsumer?: (number|null);\n\n  /** VersionDef badConsumers */\n  badConsumers?: (number[]|null);\n}\n\n/** Properties of a GraphDef. */\nexport declare interface IGraphDef {\n  /** GraphDef node */\n  node?: (INodeDef[]|null);\n\n  /** GraphDef versions */\n  versions?: (IVersionDef|null);\n\n  /** GraphDef library */\n  library?: (IFunctionDefLibrary|null);\n}\n\n/** Properties of a CollectionDef. */\nexport declare interface ICollectionDef {\n  /** CollectionDef nodeList */\n  nodeList?: (CollectionDef.INodeList|null);\n\n  /** CollectionDef bytesList */\n  bytesList?: (CollectionDef.IBytesList|null);\n\n  /** CollectionDef int64List */\n  int64List?: (CollectionDef.IInt64List|null);\n\n  /** CollectionDef floatList */\n  floatList?: (CollectionDef.IFloatList|null);\n\n  /** CollectionDef anyList */\n  anyList?: (CollectionDef.IAnyList|null);\n}\n\nexport namespace CollectionDef {\n  /** Properties of a NodeList. */\n  export declare interface INodeList {\n    /** NodeList value */\n    value?: (string[]|null);\n  }\n\n  /** Properties of a BytesList. */\n  export declare interface IBytesList {\n    /** BytesList value */\n    value?: (Uint8Array[]|null);\n  }\n\n  /** Properties of an Int64List. */\n  export declare interface IInt64List {\n    /** Int64List value */\n    value?: ((number | string)[]|null);\n  }\n\n  /** Properties of a FloatList. */\n  export declare interface IFloatList {\n    /** FloatList value */\n    value?: (number[]|null);\n  }\n\n  /** Properties of an AnyList. */\n  export declare interface IAnyList {\n    /** AnyList value */\n    value?: (IAny[]|null);\n  }\n}\n\n/** Properties of a SaverDef. */\nexport declare interface ISaverDef {\n  /** SaverDef filenameTensorName */\n  filenameTensorName?: (string|null);\n\n  /** SaverDef saveTensorName */\n  saveTensorName?: (string|null);\n\n  /** SaverDef restoreOpName */\n  restoreOpName?: (string|null);\n\n  /** SaverDef maxToKeep */\n  maxToKeep?: (number|null);\n\n  /** SaverDef sharded */\n  sharded?: (boolean|null);\n\n  /** SaverDef keepCheckpointEveryNHours */\n  keepCheckpointEveryNHours?: (number|null);\n\n  /** SaverDef version */\n  version?: (SaverDef.CheckpointFormatVersion|null);\n}\n\nexport namespace SaverDef {\n  /** CheckpointFormatVersion enum. */\n  export enum CheckpointFormatVersion {'LEGACY' = 0, 'V1' = 1, 'V2' = 2}\n}\n\n/** Properties of a TensorInfo. */\nexport declare interface ITensorInfo {\n  /** TensorInfo name */\n  name?: (string|null);\n\n  /** TensorInfo cooSparse */\n  cooSparse?: (TensorInfo.ICooSparse|null);\n\n  /** TensorInfo dtype */\n  dtype?: (DataType|null);\n\n  /** TensorInfo tensorShape */\n  tensorShape?: (ITensorShape|null);\n}\n\nexport namespace TensorInfo {\n  /** Properties of a CooSparse. */\n  export declare interface ICooSparse {\n    /** CooSparse valuesTensorName */\n    valuesTensorName?: (string|null);\n\n    /** CooSparse indicesTensorName */\n    indicesTensorName?: (string|null);\n\n    /** CooSparse denseShapeTensorName */\n    denseShapeTensorName?: (string|null);\n  }\n}\n\n/** Properties of a SignatureDef. */\nexport declare interface ISignatureDef {\n  /** SignatureDef inputs */\n  inputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef outputs */\n  outputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef methodName */\n  methodName?: (string|null);\n}\n\n/** Properties of an AssetFileDef. */\nexport declare interface IAssetFileDef {\n  /** AssetFileDef tensorInfo */\n  tensorInfo?: (ITensorInfo|null);\n\n  /** AssetFileDef filename */\n  filename?: (string|null);\n}\n\n/** Properties of an OpDef. */\nexport declare interface IOpDef {\n  /** OpDef name */\n  name?: (string|null);\n\n  /** OpDef inputArg */\n  inputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef outputArg */\n  outputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef attr */\n  attr?: (OpDef.IAttrDef[]|null);\n\n  /** OpDef deprecation */\n  deprecation?: (OpDef.IOpDeprecation|null);\n\n  /** OpDef summary */\n  summary?: (string|null);\n\n  /** OpDef description */\n  description?: (string|null);\n\n  /** OpDef isCommutative */\n  isCommutative?: (boolean|null);\n\n  /** OpDef isAggregate */\n  isAggregate?: (boolean|null);\n\n  /** OpDef isStateful */\n  isStateful?: (boolean|null);\n\n  /** OpDef allowsUninitializedInput */\n  allowsUninitializedInput?: (boolean|null);\n}\n\nexport namespace OpDef {\n  /** Properties of an ArgDef. */\n  export declare interface IArgDef {\n    /** ArgDef name */\n    name?: (string|null);\n\n    /** ArgDef description */\n    description?: (string|null);\n\n    /** ArgDef type */\n    type?: (DataType|null);\n\n    /** ArgDef typeAttr */\n    typeAttr?: (string|null);\n\n    /** ArgDef numberAttr */\n    numberAttr?: (string|null);\n\n    /** ArgDef typeListAttr */\n    typeListAttr?: (string|null);\n\n    /** ArgDef isRef */\n    isRef?: (boolean|null);\n  }\n\n  /** Properties of an AttrDef. */\n  export declare interface IAttrDef {\n    /** AttrDef name */\n    name?: (string|null);\n\n    /** AttrDef type */\n    type?: (string|null);\n\n    /** AttrDef defaultValue */\n    defaultValue?: (IAttrValue|null);\n\n    /** AttrDef description */\n    description?: (string|null);\n\n    /** AttrDef hasMinimum */\n    hasMinimum?: (boolean|null);\n\n    /** AttrDef minimum */\n    minimum?: (number|string|null);\n\n    /** AttrDef allowedValues */\n    allowedValues?: (IAttrValue|null);\n  }\n\n  /** Properties of an OpDeprecation. */\n  export declare interface IOpDeprecation {\n    /** OpDeprecation version */\n    version?: (number|null);\n\n    /** OpDeprecation explanation */\n    explanation?: (string|null);\n  }\n}\n\n/** Properties of an OpList. */\nexport declare interface IOpList {\n  /** OpList op */\n  op?: (IOpDef[]|null);\n}\n\n/** Properties of a MetaGraphDef. */\nexport declare interface IMetaGraphDef {\n  /** MetaGraphDef metaInfoDef */\n  metaInfoDef?: (MetaGraphDef.IMetaInfoDef|null);\n\n  /** MetaGraphDef graphDef */\n  graphDef?: (IGraphDef|null);\n\n  /** MetaGraphDef saverDef */\n  saverDef?: (ISaverDef|null);\n\n  /** MetaGraphDef collectionDef */\n  collectionDef?: ({[k: string]: ICollectionDef}|null);\n\n  /** MetaGraphDef signatureDef */\n  signatureDef?: ({[k: string]: ISignatureDef}|null);\n\n  /** MetaGraphDef assetFileDef */\n  assetFileDef?: (IAssetFileDef[]|null);\n}\n\nexport namespace MetaGraphDef {\n  /** Properties of a MetaInfoDef. */\n  export declare interface IMetaInfoDef {\n    /** MetaInfoDef metaGraphVersion */\n    metaGraphVersion?: (string|null);\n\n    /** MetaInfoDef strippedOpList */\n    strippedOpList?: (IOpList|null);\n\n    /** MetaInfoDef anyInfo */\n    anyInfo?: (IAny|null);\n\n    /** MetaInfoDef tags */\n    tags?: (string[]|null);\n\n    /** MetaInfoDef tensorflowVersion */\n    tensorflowVersion?: (string|null);\n\n    /** MetaInfoDef tensorflowGitVersion */\n    tensorflowGitVersion?: (string|null);\n  }\n}\n\n/** Properties of a SavedModel. */\nexport declare interface ISavedModel {\n  /** SavedModel savedModelSchemaVersion */\n  savedModelSchemaVersion?: (number|string|null);\n\n  /** SavedModel metaGraphs */\n  metaGraphs?: (IMetaGraphDef[]|null);\n}\n\n/** Properties of a FunctionDefLibrary. */\nexport declare interface IFunctionDefLibrary {\n  /** FunctionDefLibrary function */\n  'function'?: (IFunctionDef[]|null);\n\n  /** FunctionDefLibrary gradient */\n  gradient?: (IGradientDef[]|null);\n}\n\n/** Properties of a FunctionDef. */\nexport declare interface IFunctionDef {\n  /** FunctionDef signature */\n  signature?: (IOpDef|null);\n\n  /** FunctionDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n\n  /** FunctionDef nodeDef */\n  nodeDef?: (INodeDef[]|null);\n\n  /** FunctionDef ret */\n  ret?: ({[k: string]: string}|null);\n}\n\n/** Properties of a GradientDef. */\nexport declare interface IGradientDef {\n  /** GradientDef functionName */\n  functionName?: (string|null);\n\n  /** GradientDef gradientFunc */\n  gradientFunc?: (string|null);\n}\n", "\n/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpExecutor, OpMapper} from '../types';\n\nconst CUSTOM_OPS: {[key: string]: OpMapper} = {};\n\n/**\n * Register an Op for graph model executor. This allow you to register\n * TensorFlow custom op or override existing op.\n *\n * Here is an example of registering a new MatMul Op.\n * ```js\n * const customMatmul = (node) =>\n *    tf.matMul(\n *        node.inputs[0], node.inputs[1],\n *        node.attrs['transpose_a'], node.attrs['transpose_b']);\n *\n * tf.registerOp('MatMul', customMatmul);\n * ```\n * The inputs and attrs of the node object is based on the TensorFlow op\n * registry.\n *\n * @param name The Tensorflow Op name.\n * @param opFunc An op function which is called with the current graph node\n * during execution and needs to return a tensor or a list of tensors. The node\n * has the following attributes:\n *    - attr: A map from attribute name to its value\n *    - inputs: A list of input tensors\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function registerOp(name: string, opFunc: OpExecutor) {\n  const opMapper: OpMapper = {\n    tfOpName: name,\n    category: 'custom',\n    inputs: [],\n    attrs: [],\n    customExecutor: opFunc\n  };\n\n  CUSTOM_OPS[name] = opMapper;\n}\n\n/**\n * Retrieve the OpMapper object for the registered op.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function getRegisteredOp(name: string): OpMapper {\n  return CUSTOM_OPS[name];\n}\n\n/**\n * Deregister the Op for graph model executor.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function deregisterOp(name: string) {\n  delete CUSTOM_OPS[name];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {clone, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {Node, ValueType} from '../types';\n\nexport function getParamValue(\n    paramName: string, node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  const inputParam = node.inputParams[paramName];\n  if (inputParam && inputParam.inputIndexStart !== undefined) {\n    const start = inputParam.inputIndexStart;\n    const end = inputParam.inputIndexEnd === 0 ?\n        undefined :\n        (inputParam.inputIndexEnd === undefined ? start + 1 :\n                                                  inputParam.inputIndexEnd);\n    if (inputParam.type === 'tensor') {\n      return getTensor(\n          node.inputNames[inputParam.inputIndexStart], tensorMap, context);\n    }\n    if (inputParam.type === 'tensors') {\n      const inputs = node.inputNames.slice(start, end);\n\n      return inputs.map(name => getTensor(name, tensorMap, context));\n    }\n    const tensor =\n        getTensor(node.inputNames.slice(start)[0], tensorMap, context);\n    const data = tensor.dataSync();\n    return inputParam.type === 'number' ?\n        data[0] :\n        util.toNestedArray(tensor.shape, data);\n  }\n  const attrParam = node.attrParams[paramName];\n  return attrParam && attrParam.value;\n}\n\n/**\n * Retrieve the tensor based on input name by extracting the node name and\n * output index information.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensor(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor {\n  const [nodeName, index] = parseNodeName(name);\n  const contextId = context.currentContextIds.find(contextId => {\n    return !!tensorsMap[getNodeNameWithContextId(nodeName, contextId)];\n  });\n\n  return contextId !== undefined ?\n      tensorsMap[getNodeNameWithContextId(nodeName, contextId)][index] :\n      undefined;\n}\n\n/**\n * Retrieve the tensors based on input name for current context.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensorsForCurrentContenxt(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor[] {\n  return tensorsMap[getNodeNameWithContextId(name, context.currentContextId)];\n}\n\n/**\n * Returns the node name and index from the Node input name.\n * @param inputName The input name of the node, in format of\n * node_name:output_index, i.e. MatMul:0, if the output_index is not set, it is\n * default to 0.\n */\nexport function getNodeNameAndIndex(\n    inputName: string, context?: ExecutionContext): [string, number] {\n  const [nodeName, index] = parseNodeName(inputName);\n\n  return [\n    getNodeNameWithContextId(nodeName, context && context.currentContextId),\n    index\n  ];\n}\n\nfunction getNodeNameWithContextId(name: string, contextId?: string): string {\n  return !!contextId ? `${name}-${contextId}` : name;\n}\n\nexport function parseNodeName(name: string): [string, number] {\n  const parts = name.split(':');\n  if (parts.length === 1) {\n    return [name, 0];\n  }\n\n  const nodeName = parts[0];\n  return [nodeName, Number(parts[parts.length - 1])];\n}\n\nexport function split(arr: number[], size: number) {\n  const res = [];\n  for (let i = 0; i < arr.length; i += size) {\n    res.push(arr.slice(i, i + size));\n  }\n  return res;\n}\nexport function getPadding(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  let pad = getParamValue('pad', node, tensorMap, context);\n  if (pad === 'explicit') {\n    // This is 1d array, we need to convert it to 2d array\n    pad = getParamValue('explicitPaddings', node, tensorMap, context);\n    const explicitPadding: [\n      [number, number], [number, number], [number, number], [number, number]\n    ] = [[0, 0], [0, 0], [0, 0], [0, 0]];\n    for (let i = 0; i < 4; i++) {\n      explicitPadding[i][0] = (pad as number[])[i * 2];\n      explicitPadding[i][1] = (pad as number[])[i * 2 + 1];\n    }\n    return explicitPadding;\n  }\n  return pad;\n}\n\n/**\n *  Reuse the tensor if it is marked as keep, otherwise clone the tensor to\n *  avoid disposal. This is important for TensorArray and TensorList ops, since\n *  internally they use a tensor as the id for TensorArray and TensorList, and\n * to simplify lookup, they also use Tensor.id as the key to the internal map.\n * These id tensors have been marked as kept in the backend, we need avoid clone\n * them in order to create new Tensor.id.\n * @param tensor\n */\nexport function cloneTensor(tensor: Tensor): Tensor {\n  return tensor.kept ? tensor : clone(tensor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Add',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddV2',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddN',\n    'category': 'arithmetic',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'BiasAdd',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sub',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'RealDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Div',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'DivNoNan',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mul',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Maximum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Minimum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Pow',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SquaredDifference',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorMod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Abs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan2',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ceil',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ClipByValue',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'clip_value_min', 'name': 'clipValueMin', 'type': 'number'},\n      {'tfName': 'clip_value_max', 'name': 'clipValueMax', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Complex',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'real', 'type': 'tensor'},\n      {'start': 1, 'name': 'imag', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ComplexAbs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Elu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Exp',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Floor',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Imag',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Neg',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Real',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Prelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'alpha', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu6',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'clipValueMin',\n        'name': 'clipValueMin',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'clipValueMax',\n        'name': 'clipValueMax',\n        'type': 'number',\n        'defaultValue': 6\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Selu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sigmoid',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Rsqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Square',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sign',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Round',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Expm1',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log1p',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Reciprocal',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Softplus',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Erf',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axes', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'keep_dims',\n        'name': 'keepDims',\n        'type': 'bool',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LeakyRelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 0.2\n      },\n      {\n        'tfName': 'T',\n        'name': 'dtype',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'LoopCond',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'pred', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Switch',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'data', 'type': 'tensor'},\n      {'start': 1, 'name': 'pred', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Merge',\n    'category': 'control',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Enter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'frame_name', 'name': 'frameName', 'type': 'string'},\n      {'tfName': 'is_constant', 'name': 'isConstant', 'type': 'bool'}\n    ]\n  },\n  {\n    'tfOpName': 'Exit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NextIteration',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'size', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'dynamic_size', 'name': 'dynamicSize', 'type': 'bool'},\n      {'tfName': 'clear_after_read', 'name': 'clearAfterRead', 'type': 'bool'},\n      {\n        'tfName': 'identical_element_shapes',\n        'name': 'identicalElementShapes',\n        'type': 'bool'\n      },\n      {'tfName': 'tensor_array_name', 'name': 'name', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayWriteV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayReadV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{\n      'tfName': 'dtype',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  },\n  {\n    'tfOpName': 'TensorArrayGatherV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayScatterV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArrayConcatV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}, {\n        'tfName': 'element_shape_except0',\n        'name': 'elementShapeExcept0',\n        'type': 'shape',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'TensorArraySplitV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArraySizeV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayCloseV3',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'StatelessIf',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'If',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'StatelessWhile',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'While',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListScatter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListScatterV2',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 3, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGather',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListSetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListReserve',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 1, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListFromTensor',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListStack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs': [\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'},\n      {'tfName': 'num_elements', 'name': 'numElements', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListSplit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListConcat',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListPopBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListPushBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n];\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'AvgPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPoolWithArgmax',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'}, {\n        'tfName': 'include_batch_in_index',\n        'name': 'includeBatchInIndex',\n        'type': 'bool'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AvgPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Conv1D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'stride', 'name': 'stride', 'type': 'number'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NWC'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'dilation',\n        'name': 'dilation',\n        'type': 'number',\n        'defaultValue': 1\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'useCudnnOnGpu', 'name': 'useCudnnOnGpu', 'type': 'bool'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': '_FusedConv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'use_cudnn_on_gpu',\n        'name': 'useCudnnOnGpu',\n        'type': 'bool',\n        'defaultValue': true\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n    ]\n  },\n  {\n    'tfOpName': 'Conv2DBackpropInput',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 2, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 0, 'name': 'outputShape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2d',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'FusedDepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ],\n  },\n  {\n    'tfOpName': 'Dilation2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'rates', 'name': 'dilations', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Fill',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n      {'start': 1, 'name': 'value', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'LinSpace',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'num', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'OneHot',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'depth', 'type': 'number'},\n      {'start': 2, 'name': 'onValue', 'type': 'number', 'defaultValue': 1},\n      {'start': 3, 'name': 'offValue', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [\n      {\n        'tfName': 'axis',\n        'name': 'axis',\n        'type': 'number',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ones',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'OnesLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'RandomUniform',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'minval',\n        'name': 'minval',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'maxval',\n        'name': 'maxval',\n        'type': 'number',\n        'defaultValue': 1\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Range',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'step', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [{'tfName': 'Tidx', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TruncatedNormal',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'means',\n        'name': 'mean',\n        'type': 'number',\n        'defaultValue': 0.0\n      },\n      {\n        'tfName': 'stddev',\n        'name': 'stdDev',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Zeros',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'ZerosLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'Multinomial',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'logits', 'type': 'tensor'},\n      {'start': 1, 'name': 'numSamples', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'},\n      {'tfName': 'seed2', 'name': 'seed2', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'output_dtype', 'name': 'output_dtype', 'type': 'dtype'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'NonMaxSuppressionV2',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV3',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV4',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'T_threshold',\n        'name': 'threshold',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {\n        'tfName': 'pad_to_max_output_size',\n        'name': 'padToMaxOutputSize',\n        'type': 'bool'\n      }\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV5',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'},\n      {'start': 5, 'name': 'softNmsSigma', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Where',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ListDiff',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'TopKV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'k', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'sorted', 'name': 'sorted', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Unique',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n  },\n  {\n    'tfOpName': 'UniqueV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n  },\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'PlaceholderWithDefault',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'default', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'Placeholder',\n    'category': 'graph',\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {'tfOpName': 'Const', 'category': 'graph'}, {\n    'tfOpName': 'Identity',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IdentityN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Snapshot',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Rank',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Size',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Shape',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'ShapeN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Print',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'data', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'message', 'name': 'message', 'type': 'string'}, {\n        'tfName': 'first_n',\n        'name': 'firstN',\n        'type': 'number',\n        'notSupported': true\n      },\n      {\n        'tfName': 'summarize',\n        'name': 'summarize',\n        'type': 'number',\n        'defaultValue': 3\n      }\n    ]\n  },\n  {'tfOpName': 'NoOp', 'category': 'graph', 'inputs': []}, {\n    'tfOpName': 'StopGradient',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'FakeQuantWithMinMaxVars',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'min', 'name': 'min', 'type': 'number'},\n      {'tfName': 'max', 'name': 'max', 'type': 'number'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ResizeBilinear',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ResizeNearestNeighbor',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'CropAndResize',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'image', 'type': 'tensor'},\n      {'start': 1, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 2, 'name': 'boxInd', 'type': 'tensor'},\n      {'start': 3, 'name': 'cropSize', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'method', 'name': 'method', 'type': 'string'}, {\n        'tfName': 'extrapolation_value',\n        'name': 'extrapolationValue',\n        'type': 'number'\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Equal',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NotEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Greater',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'GreaterEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Less',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LessEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalAnd',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalNot',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalOr',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Select',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SelectV2',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': '_FusedMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'}, {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMulV2',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Transpose',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'perm', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FusedBatchNorm',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV2',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV3',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'LRN',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'depth_radius',\n        'name': 'radius',\n        'type': 'number',\n        'defaultValue': 5\n      },\n      {'tfName': 'bias', 'name': 'bias', 'type': 'number', 'defaultValue': 1.0},\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {\n        'tfName': 'beta',\n        'name': 'beta',\n        'type': 'number',\n        'defaultValue': 0.5\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Softmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'LogSoftmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': true,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Max',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Mean',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Min',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Sum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'All',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Any',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'ArgMax',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'ArgMin',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Cumsum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'exclusive', 'name': 'exclusive', 'type': 'bool'},\n      {'tfName': 'reverse', 'name': 'reverse', 'type': 'bool'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ConcatV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': -1, 'name': 'tensors', 'type': 'tensors'},\n      {'start': -1, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs':\n        [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n  },\n  {\n    'tfOpName': 'Concat',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 1, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n      {'start': 0, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs': [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n\n  },\n  {\n    'tfOpName': 'GatherV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Gather',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'validate_indices',\n        'name': 'validateIndices',\n        'type': 'bool',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reverse',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'dims', 'type': 'bool', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ReverseV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Slice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'size', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'StridedSlice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'end', 'type': 'number[]'},\n      {'start': 3, 'name': 'strides', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'begin_mask',\n        'name': 'beginMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'end_mask',\n        'name': 'endMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'new_axis_mask',\n        'name': 'newAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'ellipsis_mask',\n        'name': 'ellipsisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'shrink_axis_mask',\n        'name': 'shrinkAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Pack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Unpack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'num',\n        'name': 'num',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Tile',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'reps', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Split',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'axis', 'type': 'number', 'defaultValue': 0},\n      {'start': 1, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'num_split',\n      'name': 'numOrSizeSplits',\n      'type': 'number',\n      'defaultValue': 1\n    }]\n  },\n  {\n    'tfOpName': 'SplitV',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'numOrSizeSplits', 'type': 'number[]'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'ScatterNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'values', 'type': 'tensor'},\n      {'start': 2, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'GatherNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': false,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IFFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'RFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'IRFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Cast',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'SrcT',\n        'name': 'sdtype',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {'tfName': 'DstT', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'ExpandDims',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Pad',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'constant_value',\n      'name': 'constantValue',\n      'type': 'number',\n      'defaultValue': 0\n    }]\n  },\n  {\n    'tfOpName': 'PadV2',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'}, {\n        'start': 2,\n        'name': 'constantValue',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reshape',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Squeeze',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'axis',\n      'tfDeprecatedName': 'squeeze_dims',\n      'name': 'axis',\n      'type': 'number[]'\n    }]\n  },\n  {\n    'tfOpName': 'SpaceToBatchND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'paddings', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'BatchToSpaceND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'crops', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthToSpace',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'block_size', 'name': 'blockSize', 'type': 'number'},\n      {'tfName': 'data_format', 'name': 'dataFormat', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'BroadcastTo',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': []\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, env} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\n\nimport {getRegisteredOp} from './custom_op/register';\nimport {getNodeNameAndIndex} from './executors/utils';\nimport * as arithmetic from './op_list/arithmetic';\nimport * as basicMath from './op_list/basic_math';\nimport * as control from './op_list/control';\nimport * as convolution from './op_list/convolution';\nimport * as creation from './op_list/creation';\nimport * as dynamic from './op_list/dynamic';\nimport * as evaluation from './op_list/evaluation';\nimport * as graph from './op_list/graph';\nimport * as image from './op_list/image';\nimport * as logical from './op_list/logical';\nimport * as matrices from './op_list/matrices';\nimport * as normalization from './op_list/normalization';\nimport * as reduction from './op_list/reduction';\nimport * as sliceJoin from './op_list/slice_join';\nimport * as spectral from './op_list/spectral';\nimport * as transformation from './op_list/transformation';\nimport {Graph, InputParamValue, Node, OpMapper, ParamValue} from './types';\n\nexport class OperationMapper {\n  private static _instance: OperationMapper;\n\n  private opMappers: {[key: string]: OpMapper};\n\n  // Singleton instance for the mapper\n  public static get Instance() {\n    return this._instance || (this._instance = new this());\n  }\n\n  // Loads the op mapping from the JSON file.\n  private constructor() {\n    const ops = [\n      arithmetic, basicMath, control, convolution, creation, dynamic,\n      evaluation, logical, image, graph, matrices, normalization, reduction,\n      sliceJoin, spectral, transformation\n    ];\n    const mappersJson: OpMapper[] = [].concat(...ops.map(op => op.json));\n\n    this.opMappers = mappersJson.reduce<{[key: string]: OpMapper}>(\n        (map, mapper: OpMapper) => {\n          map[mapper.tfOpName] = mapper;\n          return map;\n        },\n        {});\n  }\n\n  // Converts the model inference graph from Tensorflow GraphDef to local\n  // representation for TensorFlow.js API\n  transformGraph(\n      graph: tensorflow.IGraphDef,\n      signature: tensorflow.ISignatureDef = {}): Graph {\n    const tfNodes = graph.node;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    const initNodes: Node[] = [];\n    const nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n      map[node.name] = this.mapNode(node);\n      if (node.op.startsWith('Placeholder')) {\n        placeholders.push(map[node.name]);\n      } else if (node.op === 'Const') {\n        weights.push(map[node.name]);\n      } else if (node.input == null || node.input.length === 0) {\n        initNodes.push(map[node.name]);\n      }\n      return map;\n    }, {});\n\n    let inputs: Node[] = [];\n    const outputs: Node[] = [];\n    let inputNodeNameToKey: {[key: string]: string} = {};\n    let outputNodeNameToKey: {[key: string]: string} = {};\n    if (signature != null) {\n      inputNodeNameToKey = this.mapSignatureEntries(signature.inputs);\n      outputNodeNameToKey = this.mapSignatureEntries(signature.outputs);\n    }\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    // if signature has not outputs set, add any node that does not have\n    // outputs.\n    if (Object.keys(outputNodeNameToKey).length === 0) {\n      allNodes.forEach(key => {\n        const node = nodes[key];\n        if (node.children.length === 0) {\n          outputs.push(node);\n        }\n      });\n    } else {\n      Object.keys(outputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node != null) {\n          node.signatureKey = outputNodeNameToKey[name];\n          outputs.push(node);\n        }\n      });\n    }\n\n    if (Object.keys(inputNodeNameToKey).length > 0) {\n      Object.keys(inputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node) {\n          node.signatureKey = inputNodeNameToKey[name];\n          inputs.push(node);\n        }\n      });\n    } else {\n      inputs = placeholders;\n    }\n\n    let functions = {};\n    if (graph.library != null && graph.library.function != null) {\n      functions = graph.library.function.reduce((functions, func) => {\n        functions[func.signature.name] = this.mapFunction(func);\n        return functions;\n      }, {} as {[key: string]: Graph});\n    }\n\n    const result: Graph =\n        {nodes, inputs, outputs, weights, placeholders, signature, functions};\n\n    if (initNodes.length > 0) {\n      result.initNodes = initNodes;\n    }\n\n    return result;\n  }\n\n  private mapSignatureEntries(entries: {[k: string]: tensorflow.ITensorInfo}) {\n    return Object.keys(entries || {})\n        .reduce<{[key: string]: string}>((prev, curr) => {\n          prev[entries[curr].name] = curr;\n          return prev;\n        }, {});\n  }\n\n  private mapNode(node: tensorflow.INodeDef): Node {\n    // Unsupported ops will cause an error at run-time (not parse time), since\n    // they may not be used by the actual execution subgraph.\n    const mapper =\n        getRegisteredOp(node.op) || this.opMappers[node.op] || {} as OpMapper;\n    if (node.attr == null) {\n      node.attr = {};\n    }\n\n    const newNode: Node = {\n      name: node.name,\n      op: node.op,\n      category: mapper.category,\n      inputNames:\n          (node.input ||\n           []).map(input => input.startsWith('^') ? input.substr(1) : input),\n      inputs: [],\n      children: [],\n      inputParams: {},\n      attrParams: {},\n      rawAttrs: node.attr\n    };\n\n    if (mapper.inputs != null) {\n      newNode.inputParams =\n          mapper.inputs.reduce<{[key: string]: InputParamValue}>(\n              (map, param) => {\n                map[param.name] = {\n                  type: param.type,\n                  inputIndexStart: param.start,\n                  inputIndexEnd: param.end\n                };\n                return map;\n              },\n              {});\n    }\n    if (mapper.attrs != null) {\n      newNode.attrParams =\n          mapper.attrs.reduce<{[key: string]: ParamValue}>((map, param) => {\n            const type = param.type;\n            let value = undefined;\n            switch (param.type) {\n              case 'string':\n                value = getStringParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'string[]':\n                value = getStringArrayParam(\n                    node.attr, param.tfName, param.defaultValue as string[]);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string[]);\n                }\n                break;\n              case 'number':\n                value = getNumberParam(\n                    node.attr, param.tfName,\n                    (param.defaultValue || 0) as number);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumberParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number);\n                }\n                break;\n              case 'number[]':\n                value = getNumericArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumericArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'bool':\n                value = getBoolParam(\n                    node.attr, param.tfName, param.defaultValue as boolean);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean);\n                }\n                break;\n              case 'bool[]':\n                value = getBoolArrayParam(\n                    node.attr, param.tfName, param.defaultValue as boolean[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean[]);\n                }\n                break;\n              case 'shape':\n                value = getTensorShapeParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'shape[]':\n                value = getTensorShapeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[][]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[][]);\n                }\n                break;\n              case 'dtype':\n                value = getDtypeParam(\n                    node.attr, param.tfName, param.defaultValue as DataType);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType);\n                }\n                break;\n              case 'dtype[]':\n                value = getDtypeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as DataType[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType[]);\n                }\n                break;\n              case 'func':\n                value = getFuncParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getFuncParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'tensor':\n              case 'tensors':\n                break;\n              default:\n                throw new Error(\n                    `Unsupported param type: ${param.type} for op: ${node.op}`);\n            }\n            map[param.name] = {value, type};\n            return map;\n          }, {});\n    }\n    return newNode;\n  }\n\n  // map the TFunctionDef to TFJS graph object\n  private mapFunction(functionDef: tensorflow.IFunctionDef): Graph {\n    const tfNodes = functionDef.nodeDef;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    let nodes: {[key: string]: Node} = {};\n    if (tfNodes != null) {\n      nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n        map[node.name] = this.mapNode(node);\n        if (node.op === 'Const') {\n          weights.push(map[node.name]);\n        }\n        return map;\n      }, {});\n    }\n    const inputs: Node[] = [];\n    const outputs: Node[] = [];\n\n    functionDef.signature.inputArg.forEach(arg => {\n      const [nodeName, ] = getNodeNameAndIndex(arg.name);\n      const node: Node = {\n        name: nodeName,\n        op: 'Placeholder',\n        inputs: [],\n        inputNames: [],\n        category: 'graph',\n        inputParams: {},\n        attrParams: {dtype: {value: parseDtypeParam(arg.type), type: 'dtype'}},\n        children: []\n      };\n      node.signatureKey = arg.name;\n      inputs.push(node);\n      nodes[nodeName] = node;\n    });\n\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    const returnNodeMap = functionDef.ret;\n\n    functionDef.signature.outputArg.forEach(output => {\n      const [nodeName, index] = getNodeNameAndIndex(returnNodeMap[output.name]);\n      const node = nodes[nodeName];\n      if (node != null) {\n        node.defaultOutput = index;\n        outputs.push(node);\n      }\n    });\n\n    const signature = this.mapArgsToSignature(functionDef);\n    return {nodes, inputs, outputs, weights, placeholders, signature};\n  }\n\n  private mapArgsToSignature(functionDef: tensorflow.IFunctionDef):\n      tensorflow.ISignatureDef {\n    return {\n      methodName: functionDef.signature.name,\n      inputs: functionDef.signature.inputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n      outputs: functionDef.signature.outputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg, functionDef.ret);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n    };\n  }\n\n  private mapArgToTensorInfo(\n      arg: tensorflow.OpDef.IArgDef,\n      nameMap?: {[key: string]: string}): tensorflow.ITensorInfo {\n    let name = arg.name;\n    if (nameMap != null) {\n      name = nameMap[name];\n    }\n    return {name, dtype: arg.type};\n  }\n}\n\nexport function decodeBase64(text: string): string {\n  const global = env().global;\n  if (typeof global.atob !== 'undefined') {\n    return global.atob(text);\n  } else if (typeof Buffer !== 'undefined') {\n    return new Buffer(text, 'base64').toString();\n  } else {\n    throw new Error(\n        'Unable to decode base64 in this environment. ' +\n        'Missing built-in atob() or Buffer()');\n  }\n}\n\nexport function parseStringParam(s: []|string, keepCase: boolean): string {\n  const value =\n      Array.isArray(s) ? String.fromCharCode.apply(null, s) : decodeBase64(s);\n  return keepCase ? value : value.toLowerCase();\n}\n\nexport function getStringParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string,\n    keepCase = false): string {\n  const param = attrs[name];\n  if (param != null) {\n    return parseStringParam(param.s, keepCase);\n  }\n  return def;\n}\n\nexport function getBoolParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean): boolean {\n  const param = attrs[name];\n  return param ? param.b : def;\n}\n\nexport function getNumberParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number): number {\n  const param = attrs[name] || {};\n  const value =\n      param['i'] != null ? param['i'] : (param['f'] != null ? param['f'] : def);\n  return (typeof value === 'number') ? value : parseInt(value, 10);\n}\n\nexport function parseDtypeParam(value: string|tensorflow.DataType): DataType {\n  if (typeof (value) === 'string') {\n    // tslint:disable-next-line:no-any\n    value = tensorflow.DataType[value as any];\n  }\n  switch (value) {\n    case tensorflow.DataType.DT_FLOAT:\n      return 'float32';\n    case tensorflow.DataType.DT_INT32:\n    case tensorflow.DataType.DT_INT64:\n    case tensorflow.DataType.DT_INT8:\n    case tensorflow.DataType.DT_UINT8:\n      return 'int32';\n    case tensorflow.DataType.DT_BOOL:\n      return 'bool';\n    case tensorflow.DataType.DT_DOUBLE:\n      return 'float32';\n    case tensorflow.DataType.DT_STRING:\n      return 'string';\n    default:\n      // Unknown dtype error will happen at runtime (instead of parse time),\n      // since these nodes might not be used by the actual subgraph execution.\n      return null;\n  }\n}\n\nexport function getFuncParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: string): string {\n  const param = attrs[name];\n  if (param && param.func) {\n    return param.func.name;\n  }\n  return def;\n}\n\nexport function getDtypeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType): DataType {\n  const param = attrs[name];\n  if (param && param.type) {\n    return parseDtypeParam(param.type);\n  }\n  return def;\n}\n\nexport function getDtypeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType[]): DataType[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.type) {\n    return param.list.type.map(v => parseDtypeParam(v));\n  }\n  return def;\n}\n\nexport function parseTensorShapeParam(shape: tensorflow.ITensorShape): number[]|\n    undefined {\n  if (shape.unknownRank) {\n    return undefined;\n  }\n  if (shape.dim != null) {\n    return shape.dim.map(\n        dim =>\n            (typeof dim.size === 'number') ? dim.size : parseInt(dim.size, 10));\n  }\n  return [];\n}\n\nexport function getTensorShapeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def?: number[]): number[]|undefined {\n  const param = attrs[name];\n  if (param && param.shape) {\n    return parseTensorShapeParam(param.shape);\n  }\n  return def;\n}\n\nexport function getNumericArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[]): number[] {\n  const param = attrs[name];\n  if (param) {\n    return ((param.list.f && param.list.f.length ? param.list.f :\n                                                   param.list.i) ||\n            [])\n        .map(v => (typeof v === 'number') ? v : parseInt(v, 10));\n  }\n  return def;\n}\n\nexport function getStringArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string[],\n    keepCase = false): string[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.s) {\n    return param.list.s.map((v) => {\n      return parseStringParam(v, keepCase);\n    });\n  }\n  return def;\n}\n\nexport function getTensorShapeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[][]): number[][] {\n  const param = attrs[name];\n  if (param && param.list && param.list.shape) {\n    return param.list.shape.map((v) => {\n      return parseTensorShapeParam(v);\n    });\n  }\n  return def;\n}\n\nexport function getBoolArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean[]): boolean[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.b) {\n    return param.list.b;\n  }\n  return def;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {getTensor} from '../executors/utils';\nimport {getBoolArrayParam, getBoolParam, getDtypeArrayParam, getDtypeParam, getNumberParam, getNumericArrayParam, getStringArrayParam, getStringParam, getTensorShapeArrayParam, getTensorShapeParam} from '../operation_mapper';\nimport {GraphNode, Node, ValueType} from '../types';\n\n/**\n * Helper class for lookup inputs and params for nodes in the model graph.\n */\nexport class NodeValueImpl implements GraphNode {\n  public readonly inputs: Tensor[] = [];\n  public readonly attrs: {[key: string]: ValueType} = {};\n  constructor(\n      private node: Node, private tensorMap: NamedTensorsMap,\n      private context: ExecutionContext) {\n    this.inputs = node.inputNames.map(name => this.getInput(name));\n    if (node.rawAttrs != null) {\n      this.attrs = Object.keys(node.rawAttrs)\n                       .reduce((attrs: {[key: string]: ValueType}, key) => {\n                         attrs[key] = this.getAttr(key);\n                         return attrs;\n                       }, {});\n    }\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getInput(name: string): Tensor {\n    return getTensor(name, this.tensorMap, this.context);\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getAttr(name: string, defaultValue?: ValueType): ValueType {\n    const value = this.node.rawAttrs[name];\n    if (value.tensor != null) {\n      return getTensor(name, this.tensorMap, this.context);\n    }\n    if (value.i != null || value.f != null) {\n      return getNumberParam(this.node.rawAttrs, name, defaultValue as number);\n    }\n    if (value.s != null) {\n      return getStringParam(this.node.rawAttrs, name, defaultValue as string);\n    }\n    if (value.b != null) {\n      return getBoolParam(this.node.rawAttrs, name, defaultValue as boolean);\n    }\n    if (value.shape != null) {\n      return getTensorShapeParam(\n          this.node.rawAttrs, name, defaultValue as number[]);\n    }\n    if (value.type != null) {\n      return getDtypeParam(this.node.rawAttrs, name, defaultValue as DataType);\n    }\n    if (value.list != null) {\n      if (value.list.i != null || value.list.f != null) {\n        return getNumericArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[]);\n      }\n      if (value.list.s != null) {\n        return getStringArrayParam(\n            this.node.rawAttrs, name, defaultValue as string[]);\n      }\n      if (value.list.shape != null) {\n        return getTensorShapeArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[][]);\n      }\n      if (value.list.b != null) {\n        return getBoolArrayParam(\n            this.node.rawAttrs, name, defaultValue as boolean[]);\n      }\n      if (value.list.type != null) {\n        return getDtypeArrayParam(\n            this.node.rawAttrs, name, defaultValue as DataType[]);\n      }\n    }\n\n    return defaultValue;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BiasAdd':\n        case 'AddV2':\n        case 'Add': {\n          return [tfOps.add(\n              (getParamValue('a', node, tensorMap, context) as Tensor),\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'AddN': {\n          return [tfOps.addN((\n              getParamValue('tensors', node, tensorMap, context) as Tensor[]))];\n        }\n        case 'FloorMod':\n        case 'Mod':\n          return [tfOps.mod(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'Mul':\n          return [tfOps.mul(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'RealDiv':\n        case 'Div': {\n          return [tfOps.div(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'DivNoNan': {\n          return [tfOps.divNoNan(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'FloorDiv': {\n          return [tfOps.floorDiv(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sub': {\n          return [tfOps.sub(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Minimum': {\n          return [tfOps.minimum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Maximum': {\n          return [tfOps.maximum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Pow': {\n          return [tfOps.pow(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'SquaredDifference': {\n          return [tfOps.squaredDifference(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'arithmetic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Abs':\n        case 'ComplexAbs':\n          return [tfOps.abs(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acos':\n          return [tfOps.acos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acosh':\n          return [tfOps.acosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asin':\n          return [tfOps.asin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asinh':\n          return [tfOps.asinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan':\n          return [tfOps.atan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan2':\n          return [tfOps.atan2(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('y', node, tensorMap, context) as Tensor)];\n        case 'Atanh':\n          return [tfOps.atanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Ceil':\n          return [tfOps.ceil(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Complex':\n          return [tfOps.complex(\n              getParamValue('real', node, tensorMap, context) as Tensor,\n              getParamValue('imag', node, tensorMap, context) as Tensor)];\n        case 'Cos':\n          return [tfOps.cos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Cosh':\n          return [tfOps.cosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Elu':\n          return [tfOps.elu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Erf':\n          return [tfOps.erf(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Exp':\n          return [tfOps.exp(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Expm1': {\n          return [tfOps.expm1(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Floor':\n          return [tfOps.floor(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log':\n          return [tfOps.log(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log1p': {\n          return [tfOps.log1p(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Imag':\n          return [tfOps.imag(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n\n        case 'Neg':\n          return [tfOps.neg(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Reciprocal': {\n          return [tfOps.reciprocal(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Real':\n          return [tfOps.real(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu':\n          return [tfOps.relu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Round': {\n          return [tfOps.round(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Selu':\n          return [tfOps.selu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sigmoid':\n          return [tfOps.sigmoid(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sin':\n          return [tfOps.sin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sign': {\n          return [tfOps.sign(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sinh': {\n          return [tfOps.sinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Softplus': {\n          return [tfOps.softplus(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sqrt': {\n          return [tfOps.sqrt(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Square': {\n          return [tfOps.square(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tanh': {\n          return [tfOps.tanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tan':\n          return [tfOps.tan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu6':\n        case 'ClipByValue':\n          return [tfOps.clipByValue(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('clipValueMin', node, tensorMap, context) as number,\n              getParamValue('clipValueMax', node, tensorMap, context) as\n                  number)];\n        case 'Rsqrt':\n          return [tfOps.rsqrt(\n              getTensor(node.inputNames[0], tensorMap, context))];\n        case 'Prod':\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('axes', node, tensorMap, context) as number[])];\n        case 'LeakyRelu':\n          return [tfOps.leakyRelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as number)];\n        case 'Prelu':\n          return [tfOps.prelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as Tensor)];\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'basic_math';\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n/**\n * This differs from util.assertShapesMatch in that it allows values of\n * negative one, an undefined size of a dimensinon, in a shape to match\n * anything.\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function assertShapesMatchAllowUndefinedSize(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  util.assert(\n      shapesEqualAllowUndefinedSize(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function shapesEqualAllowUndefinedSize(n1: number[], n2: number[]) {\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== -1 && n2[i] !== -1 && n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\nexport interface TensorWithState {\n  tensor?: Tensor;\n  written?: boolean;\n  read?: boolean;\n  cleared?: boolean;\n}\n/**\n * The TensorArray object keeps an array of Tensors.  It\n * allows reading from the array and writing to the array.\n */\nexport class TensorArray {\n  private tensors: TensorWithState[] = [];\n  private closed_ = false;\n  readonly idTensor: Tensor;\n  constructor(\n      readonly name: string, readonly dtype: DataType, private maxSize: number,\n      private elementShape: number[], readonly identicalElementShapes: boolean,\n      readonly dynamicSize: boolean, readonly clearAfterRead: boolean) {\n    this.idTensor = scalar(0);\n    keep(this.idTensor);\n  }\n\n  get id() {\n    return this.idTensor.id;\n  }\n\n  get closed() {\n    return this.closed_;\n  }\n\n  /**\n   * Dispose the tensors and idTensor and mark the TensoryArray as closed.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.tensor.id)) {\n        tensor.tensor.dispose();\n      }\n    });\n    this.tensors = [];\n    this.closed_ = true;\n    this.idTensor.dispose();\n  }\n\n  size(): number {\n    return this.tensors.length;\n  }\n\n  /**\n   * Read the value at location index in the TensorArray.\n   * @param index Number the index to read from.\n   */\n  read(index: number): Tensor {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || index >= this.size()) {\n      throw new Error(`Tried to read from index ${index}, but array size is: ${\n          this.size()}`);\n    }\n\n    const tensorWithState = this.tensors[index];\n    if (tensorWithState.cleared) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not read index ${\n              index} twice because it was cleared after a previous read ` +\n          `(perhaps try setting clear_after_read = false?).`);\n    }\n\n    if (this.clearAfterRead) {\n      tensorWithState.cleared = true;\n    }\n\n    tensorWithState.read = true;\n    return tensorWithState.tensor;\n  }\n\n  /**\n   * Helper method to read multiple tensors from the specified indices.\n   */\n  readMany(indices: number[]): Tensor[] {\n    return indices.map(index => this.read(index));\n  }\n\n  /**\n   * Write value into the index of the TensorArray.\n   * @param index number the index to write to.\n   * @param tensor\n   */\n  write(index: number, tensor: Tensor) {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || !this.dynamicSize && index >= this.maxSize) {\n      throw new Error(`Tried to write to index ${\n          index}, but array is not resizeable and size is: ${this.maxSize}`);\n    }\n\n    const t = this.tensors[index] || {};\n\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray ${\n          this.name}: Could not write to TensorArray index ${index},\n          because the value dtype is ${\n          tensor.dtype}, but TensorArray dtype is ${this.dtype}.`);\n    }\n\n    // Set the shape for the first time write to unknow shape tensor array\n    if (this.size() === 0 &&\n        (this.elementShape == null || this.elementShape.length === 0)) {\n      this.elementShape = tensor.shape;\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape,\n        `TensorArray ${this.name}: Could not write to TensorArray index ${\n            index}.`);\n\n    if (t.read) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been read.`);\n    }\n\n    if (t.written) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been written.`);\n    }\n\n    t.tensor = tensor;\n    keep(tensor);\n    t.written = true;\n\n    this.tensors[index] = t;\n  }\n\n  /**\n   * Helper method to write multiple tensors to the specified indices.\n   */\n  writeMany(indices: number[], tensors: Tensor[]) {\n    if (indices.length !== tensors.length) {\n      throw new Error(\n          `TensorArray ${this.name}: could not write multiple tensors,` +\n          `because the index size: ${\n              indices.length} is not the same as tensors size: ${\n              tensors.length}.`);\n    }\n\n    indices.forEach((i, index) => this.write(i, tensors[index]));\n  }\n\n  /**\n   * Return selected values in the TensorArray as a packed Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param [indices] number[] Optional. Taking values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size(). If not specified returns\n   *    all tensors in the original order.\n   * @param [dtype]\n   */\n  gather(indices?: number[], dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but gather requested dtype ${dtype}`);\n    }\n\n    if (!indices) {\n      indices = [];\n      for (let i = 0; i < this.size(); i++) {\n        indices.push(i);\n      }\n    } else {\n      indices = indices.slice(0, this.size());\n    }\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    // Read all the PersistentTensors into a vector to keep track of\n    // their memory.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape, 'TensorArray shape mismatch: ');\n\n    return stack(tensors, 0);\n  }\n\n  /**\n   * Return the values in the TensorArray as a concatenated Tensor.\n   */\n  concat(dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but concat requested dtype ${dtype}`);\n    }\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    const indices = [];\n    for (let i = 0; i < this.size(); i++) {\n      indices.push(i);\n    }\n    // Collect all the tensors from the tensors array.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape,\n        `TensorArray shape mismatch: tensor array shape (${\n            this.elementShape}) vs first tensor shape (${tensors[0].shape})`);\n\n    return concat(tensors, 0);\n  }\n\n  /**\n   * Scatter the values of a Tensor in specific indices of a TensorArray.\n   * @param indices nummber[] values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size().\n   * @param tensor Tensor input tensor.\n   */\n  scatter(indices: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n\n    if (indices.length !== tensor.shape[0]) {\n      throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n          indices.length} vs. ${tensor.shape[0]}`);\n    }\n\n    const maxIndex = Math.max(...indices);\n\n    if (!this.dynamicSize && maxIndex >= this.maxSize) {\n      throw new Error(\n          `Max index must be < array size (${maxIndex}  vs. ${this.maxSize})`);\n    }\n\n    this.writeMany(indices, unstack(tensor, 0));\n  }\n\n  /**\n   * Split the values of a Tensor into the TensorArray.\n   * @param length number[] with the lengths to use when splitting value along\n   *    its first dimension.\n   * @param tensor Tensor, the tensor to split.\n   */\n  split(length: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n    let totalLength = 0;\n    const cumulativeLengths = length.map(len => {\n      totalLength += len;\n      return totalLength;\n    });\n\n    if (totalLength !== tensor.shape[0]) {\n      throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n    }\n\n    if (!this.dynamicSize && length.length !== this.maxSize) {\n      throw new Error(\n          `TensorArray's size is not equal to the size of lengths (${\n              this.maxSize} vs. ${length.length}), ` +\n          'and the TensorArray is not marked as dynamically resizeable');\n    }\n\n    const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n    const tensors: Tensor[] = [];\n    tidy(() => {\n      tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n      for (let i = 0; i < length.length; ++i) {\n        const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n        const indices = [0, previousLength, 0];\n        const sizes = [1, length[i], elementPerRow];\n        tensors[i] = reshape(slice(tensor, indices, sizes), this.elementShape);\n      }\n      return tensors;\n    });\n    const indices = [];\n    for (let i = 0; i < length.length; i++) {\n      indices[i] = i;\n    }\n    this.writeMany(indices, tensors);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\n/**\n * TensorList stores a container of `tf.Tensor` objects, which are accessible\n * via tensors field.\n *\n * In order to get a copy of the underlying list, use the copy method:\n * ```\n *    TensorList b = a.copy();\n *    b.tensors().pushBack(t);  // This does not modify a.tensors().\n * ```\n *\n * Note that this is not a deep copy: the memory locations of the underlying\n * tensors will still point to the same locations of the corresponding tensors\n * in the original.\n */\n\nexport class TensorList {\n  readonly idTensor: Tensor;\n  maxNumElements: number;\n\n  get id() {\n    return this.idTensor.id;\n  }\n  /**\n   *\n   * @param tensors list of tensors\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param maxNumElements The maximum allowed size of `tensors`. Defaults to -1\n   *   meaning that the size of `tensors` is unbounded.\n   */\n  constructor(\n      readonly tensors: Tensor[], readonly elementShape: number[],\n      readonly elementDtype: DataType, maxNumElements = -1) {\n    if (tensors != null) {\n      tensors.forEach(tensor => {\n        if (elementDtype !== tensor.dtype) {\n          throw new Error(`Invalid data types; op elements ${\n              elementDtype}, but list elements ${tensor.dtype}`);\n        }\n        assertShapesMatchAllowUndefinedSize(\n            elementShape, tensor.shape, 'TensorList shape mismatch: ');\n\n        keep(tensor);\n      });\n    }\n    this.idTensor = scalar(0);\n    this.maxNumElements = maxNumElements;\n    keep(this.idTensor);\n  }\n\n  /**\n   * Get a new TensorList containing a copy of the underlying tensor container.\n   */\n  copy(): TensorList {\n    return new TensorList(\n        [...this.tensors], this.elementShape, this.elementDtype);\n  }\n\n  /**\n   * Dispose the tensors and idTensor and clear the tensor list.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.id)) {\n        tensor.dispose();\n      }\n    });\n    this.tensors.length = 0;\n    this.idTensor.dispose();\n  }\n  /**\n   * The size of the tensors in the tensor list.\n   */\n  size() {\n    return this.tensors.length;\n  }\n\n  /**\n   * Return a tensor that stacks a list of rank-R tf.Tensors into one rank-(R+1)\n   * tf.Tensor.\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param numElements the number of elements to stack\n   */\n  stack(elementShape: number[], elementDtype: DataType, numElements = -1):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (numElements !== -1 && this.tensors.length !== numElements) {\n      throw new Error(`Operation expected a list with ${\n          numElements} elements but got a list with ${\n          this.tensors.length} elements.`);\n    }\n    assertShapesMatchAllowUndefinedSize(\n        elementShape, this.elementShape, 'TensorList shape mismatch: ');\n    return tidy(() => {\n      const reshapedTensors =\n          this.tensors.map(tensor => reshape(tensor, elementShape));\n      return stack(reshapedTensors, 0);\n    });\n  }\n\n  /**\n   * Pop a tensor from the end of the list.\n   * @param elementShape shape of the tensor\n   * @param elementDtype data type of the tensor\n   */\n  popBack(elementShape: number[], elementDtype: DataType): Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (this.size() === 0) {\n      throw new Error('Trying to pop from an empty list.');\n    }\n\n    const tensor = this.tensors.pop();\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, elementShape, 'TensorList shape mismatch: ');\n    return reshape(tensor, elementShape);\n  }\n\n  /**\n   * Push a tensor to the end of the list.\n   * @param tensor Tensor to be pushed.\n   */\n  pushBack(tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, this.elementShape, 'TensorList shape mismatch: ');\n\n    if (this.maxNumElements === this.size()) {\n      throw new Error(`Trying to push element into a full list.`);\n    }\n    keep(tensor);\n    this.tensors.push(tensor);\n  }\n\n  /**\n   * Update the size of the list.\n   * @param size the new size of the list.\n   */\n  resize(size: number) {\n    if (size < 0) {\n      throw new Error(\n          `TensorListResize expects size to be non-negative. Got: ${size}`);\n    }\n\n    if (this.maxNumElements !== -1 && size > this.maxNumElements) {\n      throw new Error(`TensorListResize input size ${\n          size} is greater maxNumElement ${this.maxNumElements}.`);\n    }\n    this.tensors.length = size;\n  }\n\n  /**\n   * Retrieve the element at the provided index\n   * @param elementShape shape of the tensor\n   * @param elementDtype dtype of the tensor\n   * @param elementIndex index of the tensor\n   */\n  getItem(elementIndex: number, elementShape: number[], elementDtype: DataType):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (elementIndex < 0 || elementIndex > this.tensors.length) {\n      throw new Error(`Trying to access element ${\n          elementIndex} in a list with ${this.tensors.length} elements.`);\n    }\n\n    if (this.tensors[elementIndex] == null) {\n      throw new Error(`element at index ${elementIndex} is null.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.tensors[elementIndex].shape, elementShape,\n        'TensorList shape mismatch: ');\n\n    return this.tensors[elementIndex];\n  }\n\n  /**\n   * Set the tensor at the index\n   * @param elementIndex index of the tensor\n   * @param tensor the tensor to be inserted into the list\n   */\n  setItem(elementIndex: number, tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (elementIndex < 0 ||\n        this.maxNumElements !== -1 && elementIndex >= this.maxNumElements) {\n      throw new Error(`Trying to set element ${\n          elementIndex} in a list with max ${this.maxNumElements} elements.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape, 'TensorList shape mismatch: ');\n    keep(tensor);\n    this.tensors[elementIndex] = tensor;\n  }\n\n  /**\n   * Return selected values in the TensorList as a stacked Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param indices indices of tensors to gather\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  gather(indices: number[], elementDtype: DataType, elementShape: number[]):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    // When indices is greater than the size of the list, indices beyond the\n    // size of the list are ignored.\n    indices = indices.slice(0, this.size());\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = indices.map(i => reshape(this.tensors[i], elementShape));\n      return stack(tensors, 0);\n    });\n  }\n\n  /**\n   * Return the values in the TensorList as a concatenated Tensor.\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  concat(elementDtype: DataType, elementShape: number[]): Tensor {\n    if (!!elementDtype && elementDtype !== this.elementDtype) {\n      throw new Error(`TensorList dtype is ${\n          this.elementDtype} but concat requested dtype ${elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = this.tensors.map(t => reshape(t, elementShape));\n      return concat(tensors, 0);\n    });\n  }\n}\n\n/**\n * Creates a TensorList which, when stacked, has the value of tensor.\n * @param tensor from tensor\n * @param elementShape output tensor element shape\n */\nexport function fromTensor(\n    tensor: Tensor, elementShape: number[], elementDtype: DataType) {\n  const dtype = tensor.dtype;\n  if (tensor.shape.length < 1) {\n    throw new Error(\n        `Tensor must be at least a vector, but saw shape: ${tensor.shape}`);\n  }\n  if (tensor.dtype !== elementDtype) {\n    throw new Error(`Invalid data types; op elements ${\n        tensor.dtype}, but list elements ${elementDtype}`);\n  }\n  const outputShape = tensor.shape.slice(1);\n  assertShapesMatchAllowUndefinedSize(\n      outputShape, elementShape, 'TensorList shape mismatch: ');\n\n  const tensorList: Tensor[] = unstack(tensor);\n  return new TensorList(tensorList, elementShape, dtype);\n}\n\n/**\n * Return a TensorList of the given size with empty elements.\n * @param elementShape the shape of the future elements of the list\n * @param elementDtype the desired type of elements in the list\n * @param numElements the number of elements to reserve\n */\nexport function reserve(\n    elementShape: number[], elementDtype: DataType, numElements: number) {\n  return new TensorList([], elementShape, elementDtype, numElements);\n}\n\n/**\n * Put tensors at specific indices of a stacked tensor into a TensorList.\n * @param indices list of indices on how to scatter the tensor.\n * @param tensor input tensor.\n * @param elementShape the shape of the future elements of the list\n * @param numElements the number of elements to scatter\n */\nexport function scatter(\n    tensor: Tensor, indices: number[], elementShape: number[],\n    numElements?: number): TensorList {\n  if (indices.length !== tensor.shape[0]) {\n    throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n        indices.length} vs. ${tensor.shape[0]}`);\n  }\n\n  const maxIndex = Math.max(...indices);\n\n  if (numElements != null && numElements !== -1 && maxIndex >= numElements) {\n    throw new Error(\n        `Max index must be < array size (${maxIndex}  vs. ${numElements})`);\n  }\n\n  const list = new TensorList([], elementShape, tensor.dtype, numElements);\n  const tensors = unstack(tensor, 0);\n  indices.forEach((value, index) => {\n    list.setItem(value, tensors[index]);\n  });\n  return list;\n}\n\n/**\n * Split the values of a Tensor into a TensorList.\n * @param length the lengths to use when splitting value along\n *    its first dimension.\n * @param tensor the tensor to split.\n * @param elementShape the shape of the future elements of the list\n */\nexport function split(\n    tensor: Tensor, length: number[], elementShape: number[]) {\n  let totalLength = 0;\n  const cumulativeLengths = length.map(len => {\n    totalLength += len;\n    return totalLength;\n  });\n\n  if (totalLength !== tensor.shape[0]) {\n    throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n  }\n\n  const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n  const tensors: Tensor[] = tidy(() => {\n    const tensors = [];\n    tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n    for (let i = 0; i < length.length; ++i) {\n      const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n      const indices = [0, previousLength, 0];\n      const sizes = [1, length[i], elementPerRow];\n      tensors[i] = reshape(slice(tensor, indices, sizes), elementShape);\n    }\n    tensor.dispose();\n    return tensors;\n  });\n\n  const list = new TensorList([], elementShape, tensor.dtype, length.length);\n\n  for (let i = 0; i < tensors.length; i++) {\n    list.setItem(i, tensors[i]);\n  }\n  return list;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, scalar, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {TensorArray} from '../../executor/tensor_array';\nimport {fromTensor, reserve, scatter, split} from '../../executor/tensor_list';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'If':\n    case 'StatelessIf': {\n      const thenFunc =\n          getParamValue('thenBranch', node, tensorMap, context) as string;\n      const elseFunc =\n          getParamValue('elseBranch', node, tensorMap, context) as string;\n      const cond = getParamValue('cond', node, tensorMap, context) as Tensor;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n      const condValue = await cond.data();\n      if (condValue[0]) {\n        return context.functionMap[thenFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      } else {\n        return context.functionMap[elseFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      }\n    }\n    case 'While':\n    case 'StatelessWhile': {\n      const bodyFunc =\n          getParamValue('body', node, tensorMap, context) as string;\n      const condFunc =\n          getParamValue('cond', node, tensorMap, context) as string;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n\n      // Calculate the condition of the loop\n      const condResult =\n          (await context.functionMap[condFunc].executeFunctionAsync(\n              args, context.tensorArrayMap, context.tensorListMap));\n      const argIds = args.map(tensor => tensor.id);\n      let condValue = await condResult[0].data();\n      // Dispose the intermediate tensors for condition function\n      condResult.forEach(tensor => {\n        if (!tensor.kept && argIds.indexOf(tensor.id) === -1) {\n          tensor.dispose();\n        }\n      });\n\n      let result: Tensor[] = args;\n\n      while (condValue[0]) {\n        // Record the previous result for intermediate tensor tracking\n        const origResult = result;\n        // Execution the body of the loop\n        result = await context.functionMap[bodyFunc].executeFunctionAsync(\n            result, context.tensorArrayMap, context.tensorListMap);\n        const resultIds = result.map(tensor => tensor.id);\n\n        // Dispose the intermediate tensor for body function that is not global\n        // kept, not input/output of the body function\n        origResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n\n        // Recalcuate the condition of the loop using the latest results.\n        const condResult =\n            (await context.functionMap[condFunc].executeFunctionAsync(\n                result, context.tensorArrayMap, context.tensorListMap));\n        condValue = await condResult[0].data();\n        // Dispose the intermediate tensors for condition function\n        condResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n      }\n      return result;\n    }\n    case 'LoopCond': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      return [cloneTensor(pred)];\n    }\n    case 'Switch': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      let data = getParamValue('data', node, tensorMap, context) as Tensor;\n      if (!data.kept) {\n        data = cloneTensor(data);\n      }\n      // Outputs nodes :0 => false, :1 => true\n      return (await pred.data())[0] ? [undefined, data] : [data, undefined];\n    }\n    case 'Merge': {\n      const inputName = node.inputNames.find(\n          name => getTensor(name, tensorMap, context) !== undefined);\n      if (inputName) {\n        const data = getTensor(inputName, tensorMap, context);\n        return [cloneTensor(data)];\n      }\n      return undefined;\n    }\n    case 'Enter': {\n      const frameId =\n          getParamValue('frameName', node, tensorMap, context) as string;\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.enterFrame(frameId);\n      return [cloneTensor(data)];\n    }\n    case 'Exit': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.exitFrame();\n      return [cloneTensor(data)];\n    }\n    case 'NextIteration': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.nextIteration();\n      return [cloneTensor(data)];\n    }\n    case 'TensorArrayV3': {\n      const size = getParamValue('size', node, tensorMap, context) as number;\n      const dtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const dynamicSize =\n          getParamValue('dynamicSize', node, tensorMap, context) as boolean;\n      const clearAfterRead =\n          getParamValue('clearAfterRead', node, tensorMap, context) as boolean;\n      const identicalElementShapes =\n          getParamValue('identicalElementShapes', node, tensorMap, context) as\n          boolean;\n      const name = getParamValue('name', node, tensorMap, context) as string;\n      const tensorArray = new TensorArray(\n          name, dtype, size, elementShape, identicalElementShapes, dynamicSize,\n          clearAfterRead);\n      context.addTensorArray(tensorArray);\n      return [tensorArray.idTensor, scalar(1.0)];\n    }\n    case 'TensorArrayWriteV3': {\n      const id =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const writeTensorArray = context.getTensorArray(id.id);\n      writeTensorArray.write(index, writeTensor);\n      return [writeTensorArray.idTensor];\n    }\n    case 'TensorArrayReadV3': {\n      const readId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const readTensorArray = context.getTensorArray(readId.id);\n      return [readTensorArray.read(readIndex)];\n    }\n    case 'TensorArrayGatherV3': {\n      const gatherId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const gatherDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const gatherTensorArray = context.getTensorArray(gatherId.id);\n      return [gatherTensorArray.gather(gatherIndices, gatherDtype)];\n    }\n    case 'TensorArrayScatterV3': {\n      const scatterId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const scatterTensorArray = context.getTensorArray(scatterId.id);\n      scatterTensorArray.scatter(scatterIndices, scatterTensor);\n      return [scatterTensorArray.idTensor];\n    }\n    case 'TensorArrayConcatV3': {\n      const concatId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const concatTensorArray = context.getTensorArray(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      return [concatTensorArray.concat(concatDtype)];\n    }\n    case 'TensorArraySplitV3': {\n      const splitId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n      const splitTensorArray = context.getTensorArray(splitId.id);\n      splitTensorArray.split(lengths, splitTensor);\n      return [splitTensorArray.idTensor];\n    }\n    case 'TensorArraySizeV3': {\n      const sizeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const sizeTensorArray = context.getTensorArray(sizeId.id);\n      return [scalar(sizeTensorArray.size(), 'int32')];\n    }\n    case 'TensorArrayCloseV3': {\n      const closeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const closeTensorArray = context.getTensorArray(closeId.id);\n      closeTensorArray.clearAndClose();\n      return [closeTensorArray.idTensor];\n    }\n    case 'TensorListSetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.setItem(index, writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.getItem(readIndex, elementShape, elementDType)];\n    }\n    case 'TensorListScatterV2':\n    case 'TensorListScatter': {\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList =\n          scatter(scatterTensor, scatterIndices, elementShape, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListReserve': {\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = reserve(elementShape, elementDtype, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGather': {\n      const gatherId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(gatherId.id);\n      return [tensorList.gather(gatherIndices, elementDtype, elementShape)];\n    }\n    case 'TensorListStack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.stack(elementShape, elementDtype, numElements)];\n    }\n    case 'TensorListFromTensor': {\n      const tensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = fromTensor(tensor, elementShape, elementDtype);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListConcat': {\n      const concatId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      return [tensorList.concat(concatDtype, elementShape)];\n    }\n    case 'TensorListPushBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.pushBack(writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListPopBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.popBack(elementShape, elementDType)];\n    }\n    case 'TensorListSplit': {\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n\n      const tensorList = split(splitTensor, lengths, elementShape);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'control';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rank, Tensor, Tensor3D, Tensor4D, Tensor5D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getPadding, getParamValue} from './utils';\n\nfunction fusedConvAndDepthWiseParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const [extraOp, activationFunc] =\n      (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n  const isBiasAdd = extraOp === 'biasadd';\n  const isPrelu = activationFunc === 'prelu';\n  const isBatchNorm = extraOp === 'fusedbatchnorm';\n\n  const numArgs =\n      (getParamValue('numArgs', node, tensorMap, context) as number);\n  if (isBiasAdd) {\n    if (isPrelu && numArgs !== 2) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu ' +\n          'must have two extra arguments: bias and alpha.');\n    }\n    if (!isPrelu && numArgs !== 1) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd must have ' +\n          'one extra argument: bias.');\n    }\n  }\n  if (isBatchNorm) {\n    throw new Error(\n        'FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.');\n  }\n  const stride = getParamValue('strides', node, tensorMap, context) as number[];\n  const pad = getPadding(node, tensorMap, context);\n  const dataFormat =\n      (getParamValue('dataFormat', node, tensorMap, context) as string)\n          .toUpperCase();\n  const dilations =\n      getParamValue('dilations', node, tensorMap, context) as number[];\n  const [biasArg, preluArg] =\n      getParamValue('args', node, tensorMap, context) as Tensor[];\n\n  return {\n    stride,\n    pad,\n    dataFormat,\n    dilations,\n    biasArg,\n    preluArg,\n    activationFunc\n  };\n}\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Conv1D': {\n          const stride =\n              getParamValue('stride', node, tensorMap, context) as number;\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilation =\n              getParamValue('dilation', node, tensorMap, context) as number;\n          return [tfOps.conv1d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              stride, pad as 'valid' | 'same', dataFormat as 'NWC' | 'NCW',\n              dilation)];\n        }\n        case 'Conv2D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case '_FusedConv2D': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.conv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n\n        case 'FusedDepthwiseConv2dNative': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.depthwiseConv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n        case 'Conv2DBackpropInput':\n        case 'Conv2dTranspose': {\n          const shape = getParamValue(\n                            'outputShape', node, tensorMap,\n                            context) as [number, number, number] |\n              [number, number, number, number];\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          return [tfOps.conv2dTranspose(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              shape, [stride[1], stride[2]], pad as 'valid' | 'same')];\n        }\n        case 'DepthwiseConv2dNative':\n        case 'DepthwiseConv2d': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n\n          return [tfOps.depthwiseConv2d(\n              getParamValue('input', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case 'Conv3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv3d(\n              getParamValue('x', node, tensorMap, context) as Tensor4D |\n                  Tensor<Rank.R5>,\n              getParamValue('filter', node, tensorMap, context) as\n                  Tensor<Rank.R5>,\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same',\n              dataFormat as 'NDHWC' | 'NCDHW',\n              [dilations[1], dilations[2], dilations[3]])];\n        }\n        case 'AvgPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPoolWithArgmax': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n          const includeBatchInIndex =\n              getParamValue('includeBatchInIndex', node, tensorMap, context) as\n              boolean;\n          const {result, indexes} = tfOps.maxPoolWithArgmax(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same', includeBatchInIndex);\n          return [result, indexes];\n        }\n        case 'AvgPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'MaxPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'Dilation2D': {\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n\n          // strides: [1, stride_height, stride_width, 1].\n          const strideHeight = strides[1];\n          const strideWidth = strides[2];\n\n          // dilations: [1, dilation_height, dilation_width, 1].\n          const dilationHeight = dilations[1];\n          const dilationWidth = dilations[2];\n\n          return [tfOps.dilation2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              [strideHeight, strideWidth], pad as 'valid' | 'same',\n              [dilationHeight, dilationWidth], 'NHWC' /* dataFormat */)];\n        }\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'convolution';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor, Tensor1D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Fill': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const dtype =\n              getParamValue('dtype', node, tensorMap, context) as DataType;\n          const value =\n              getParamValue('value', node, tensorMap, context) as number;\n          return [tfOps.fill(shape, value, dtype)];\n        }\n        case 'LinSpace': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const num = getParamValue('num', node, tensorMap, context) as number;\n          return [tfOps.linspace(start, stop, num)];\n        }\n        case 'Multinomial': {\n          const logits =\n              getParamValue('logits', node, tensorMap, context) as Tensor1D;\n          const numSamples =\n              getParamValue('numSamples', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.multinomial(logits, numSamples, seed)];\n        }\n        case 'OneHot': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          const depth =\n              getParamValue('depth', node, tensorMap, context) as number;\n          const onValue =\n              getParamValue('onValue', node, tensorMap, context) as number;\n          const offValue =\n              getParamValue('offValue', node, tensorMap, context) as number;\n          return [tfOps.oneHot(indices, depth, onValue, offValue)];\n        }\n        case 'Ones': {\n          return [tfOps.ones(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'OnesLike': {\n          return [tfOps.onesLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'RandomUniform': {\n          return [tfOps.randomUniform(\n              // tslint:disable-next-line:no-any\n              getParamValue('shape', node, tensorMap, context) as any,\n              getParamValue('minval', node, tensorMap, context) as number,\n              getParamValue('maxval', node, tensorMap, context) as number,\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'Range': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const step =\n              getParamValue('step', node, tensorMap, context) as number;\n          return [tfOps.range(\n              start, stop, step,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32')];\n        }\n        case 'TruncatedNormal': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const mean =\n              getParamValue('mean', node, tensorMap, context) as number;\n          const stdDev =\n              getParamValue('stdDev', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.truncatedNormal(\n              shape, mean, stdDev,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32',\n              seed)];\n        }\n        case 'Zeros': {\n          return [tfOps.zeros(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'ZerosLike': {\n          return [tfOps.zerosLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'creation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nfunction nmsParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const boxes = getParamValue('boxes', node, tensorMap, context) as Tensor;\n  const scores = getParamValue('scores', node, tensorMap, context) as Tensor;\n  const maxOutputSize =\n      getParamValue('maxOutputSize', node, tensorMap, context) as number;\n  const iouThreshold =\n      getParamValue('iouThreshold', node, tensorMap, context) as number;\n  const scoreThreshold =\n      getParamValue('scoreThreshold', node, tensorMap, context) as number;\n  const softNmsSigma =\n      getParamValue('softNmsSigma', node, tensorMap, context) as number;\n\n  return {\n    boxes,\n    scores,\n    maxOutputSize,\n    iouThreshold,\n    scoreThreshold,\n    softNmsSigma\n  };\n}\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'NonMaxSuppressionV5': {\n      const {\n        boxes,\n        scores,\n        maxOutputSize,\n        iouThreshold,\n        scoreThreshold,\n        softNmsSigma\n      } = nmsParams(node, tensorMap, context);\n\n      const result = await tfOps.image.nonMaxSuppressionWithScoreAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, softNmsSigma);\n\n      return [result.selectedIndices, result.selectedScores];\n    }\n    case 'NonMaxSuppressionV4': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      const padToMaxOutputSize =\n          getParamValue('padToMaxOutputSize', node, tensorMap, context) as\n          boolean;\n\n      const result = await tfOps.image.nonMaxSuppressionPaddedAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, padToMaxOutputSize);\n\n      return [result.selectedIndices, result.validOutputs];\n    }\n    case 'NonMaxSuppressionV3':\n    case 'NonMaxSuppressionV2': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      return [await tfOps.image.nonMaxSuppressionAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold)];\n    }\n    case 'Where': {\n      const condition = tfOps.cast(\n          (getParamValue('condition', node, tensorMap, context) as Tensor),\n          'bool');\n      const result = [await tfOps.whereAsync(condition)];\n      condition.dispose();\n      return result;\n    }\n    case 'ListDiff': {\n      return tfOps.setdiff1dAsync(\n          getParamValue('x', node, tensorMap, context) as Tensor,\n          getParamValue('y', node, tensorMap, context) as Tensor);\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'dynamic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'TopKV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const k = getParamValue('k', node, tensorMap, context) as number;\n              const sorted =\n                  getParamValue('sorted', node, tensorMap, context) as boolean;\n              const result = tfOps.topk(x, k, sorted);\n              return [result.values, result.indices];\n            }\n            case 'Unique': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const result = tfOps.unique(x);\n              return [result.values, result.indices];\n            }\n            case 'UniqueV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const axis =\n                  getParamValue('axis', node, tensorMap, context) as number;\n              const result = tfOps.unique(x, axis);\n              return [result.values, result.indices];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'evaluation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Const': {\n          return tensorMap[node.name];\n        }\n        case 'PlaceholderWithDefault':\n          const def =\n              getParamValue('default', node, tensorMap, context) as Tensor;\n          return [getTensor(node.name, tensorMap, context) || def];\n        case 'Placeholder':\n          return [getTensor(node.name, tensorMap, context)];\n        case 'Identity':\n        case 'StopGradient':\n        case 'FakeQuantWithMinMaxVars': {  // This op is currently ignored.\n          const data = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [cloneTensor(data)];\n        }\n        case 'IdentityN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => cloneTensor(t));\n        case 'Snapshot':\n          const snapshot =\n              (getParamValue('x', node, tensorMap, context) as Tensor);\n          return [cloneTensor(snapshot)];\n        case 'Shape':\n          return [tfOps.tensor1d(\n              (getParamValue('x', node, tensorMap, context) as Tensor).shape,\n              'int32')];\n        case 'ShapeN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => tfOps.tensor1d(t.shape));\n        case 'Size':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).size,\n              'int32')];\n        case 'Rank':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).rank,\n              'int32')];\n        case 'NoOp':\n          return [tfOps.scalar(1)];\n        case 'Print':\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const data =\n              getParamValue('data', node, tensorMap, context) as Tensor[];\n          const message =\n              getParamValue('message', node, tensorMap, context) as string;\n          const summarize =\n              getParamValue('summarize', node, tensorMap, context) as number;\n          console.warn(\n              'The graph has a tf.print() operation,' +\n              'usually used for debugging, which slows down performance.');\n          console.log(message);\n          for (let i = 0; i < data.length; i++) {\n            console.log(Array.prototype.slice.call(data[i].dataSync())\n                            .slice(0, summarize));\n          }\n          return [input];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'graph';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ResizeBilinear': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeBilinear(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'ResizeNearestNeighbor': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeNearestNeighbor(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'CropAndResize': {\n          const image =\n              getParamValue('image', node, tensorMap, context) as Tensor;\n          const boxes =\n              getParamValue('boxes', node, tensorMap, context) as Tensor;\n          const boxInd =\n              getParamValue('boxInd', node, tensorMap, context) as Tensor;\n          const cropSize =\n              getParamValue('cropSize', node, tensorMap, context) as number[];\n          const method =\n              getParamValue('method', node, tensorMap, context) as string;\n          const extrapolationValue =\n              getParamValue('extrapolationValue', node, tensorMap, context) as\n              number;\n          return [tfOps.image.cropAndResize(\n              image as Tensor4D, boxes as Tensor2D, boxInd as Tensor1D,\n              cropSize as [number, number], method as 'bilinear' | 'nearest',\n              extrapolationValue)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'image';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Equal': {\n          return [tfOps.equal(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'NotEqual': {\n          return [tfOps.notEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Greater': {\n          return [tfOps.greater(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'GreaterEqual': {\n          return [tfOps.greaterEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Less': {\n          return [tfOps.less(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LessEqual': {\n          return [tfOps.lessEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalAnd': {\n          return [tfOps.logicalAnd(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalNot': {\n          return [tfOps.logicalNot(\n              getParamValue('a', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalOr': {\n          return [tfOps.logicalOr(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Select':\n        case 'SelectV2': {\n          return [tfOps.where(\n              getParamValue('condition', node, tensorMap, context) as Tensor,\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'logical';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BatchMatMul':\n        case 'BatchMatMulV2':\n        case 'MatMul':\n          return [tfOps.matMul(\n              getParamValue('a', node, tensorMap, context) as Tensor2D,\n              getParamValue('b', node, tensorMap, context) as Tensor2D,\n              getParamValue('transposeA', node, tensorMap, context) as boolean,\n              getParamValue('transposeB', node, tensorMap, context) as\n                  boolean)];\n\n        case 'Transpose':\n          return [tfOps.transpose(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('perm', node, tensorMap, context) as number[])];\n\n        case '_FusedMatMul':\n          const [extraOp, activationFunc] =\n              (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n          const isBiasAdd = extraOp === 'biasadd';\n          const isPrelu = activationFunc === 'prelu';\n\n          const numArgs =\n              (getParamValue('numArgs', node, tensorMap, context) as number);\n          if (isBiasAdd) {\n            if (isPrelu && numArgs !== 2) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd and Prelu must have two ' +\n                  'extra arguments: bias and alpha.');\n            }\n            if (!isPrelu && numArgs !== 1) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd must have one extra argument: bias.');\n            }\n          }\n          const [biasArg, preluArg] =\n              getParamValue('args', node, tensorMap, context) as Tensor[];\n          return [tfOps.fused.matMul({\n            a: getParamValue('a', node, tensorMap, context) as Tensor2D,\n            b: getParamValue('b', node, tensorMap, context) as Tensor2D,\n            transposeA: getParamValue('transposeA', node, tensorMap, context) as\n                boolean,\n            transposeB: getParamValue('transposeB', node, tensorMap, context) as\n                boolean,\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'matrices';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'FusedBatchNorm':\n        case 'FusedBatchNormV2': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'FusedBatchNormV3': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'LRN': {\n          return [tfOps.localResponseNormalization(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('radius', node, tensorMap, context) as number,\n              getParamValue('bias', node, tensorMap, context) as number,\n              getParamValue('alpha', node, tensorMap, context) as number,\n              getParamValue('beta', node, tensorMap, context) as number)];\n        }\n        case 'Softmax': {\n          return [tfOps.softmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogSoftmax': {\n          return [tfOps.logSoftmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'SparseToDense': {\n          return [tfOps.sparseToDense(\n              getParamValue('sparseIndices', node, tensorMap, context) as\n                  Tensor,\n              getParamValue('outputShape', node, tensorMap, context) as Tensor,\n              getParamValue('sparseValues', node, tensorMap, context) as\n                  number[],\n              getParamValue('defaultValue', node, tensorMap, context) as\n                  Scalar)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'normalization';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Max': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.max(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Mean': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.mean(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Min': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.min(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Sum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.sum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'All': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.all(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Any': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.any(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'ArgMax': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMax(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'ArgMin': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMin(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Prod': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Cumsum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const exclusive =\n              getParamValue('exclusive', node, tensorMap, context) as boolean;\n          const reverse =\n              getParamValue('reverse', node, tensorMap, context) as boolean;\n          return [tfOps.cumsum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              exclusive, reverse)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'reduction';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ConcatV2':\n        case 'Concat': {\n          const n = getParamValue('n', node, tensorMap, context) as number;\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          let inputs =\n              getParamValue('tensors', node, tensorMap, context) as Tensor[];\n          inputs = inputs.slice(0, n);\n          return [tfOps.concat(inputs, axis)];\n        }\n        case 'GatherV2':\n        case 'Gather': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          return [tfOps.gather(input, tfOps.cast(indices, 'int32'), axis)];\n        }\n        case 'ReverseV2':\n        case 'Reverse': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [tfOps.reverse(input, axis)];\n        }\n        case 'Slice': {\n          // tslint:disable-next-line:no-any\n          const begin = getParamValue('begin', node, tensorMap, context) as any;\n          // tslint:disable-next-line:no-any\n          const size = getParamValue('size', node, tensorMap, context) as any;\n          return [tfOps.slice(\n              getParamValue('x', node, tensorMap, context) as Tensor, begin,\n              size)];\n        }\n        case 'StridedSlice': {\n          const begin =\n              getParamValue('begin', node, tensorMap, context) as number[];\n          const end =\n              getParamValue('end', node, tensorMap, context) as number[];\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const beginMask =\n              getParamValue('beginMask', node, tensorMap, context) as number;\n          const endMask =\n              getParamValue('endMask', node, tensorMap, context) as number;\n          const ellipsisMask =\n              getParamValue('ellipsisMask', node, tensorMap, context) as number;\n          const newAxisMask =\n              getParamValue('newAxisMask', node, tensorMap, context) as number;\n          const shrinkAxisMask =\n              getParamValue('shrinkAxisMask', node, tensorMap, context) as\n              number;\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return [tfOps.stridedSlice(\n              tensor, begin, end, strides, beginMask, endMask, ellipsisMask,\n              newAxisMask, shrinkAxisMask)];\n        }\n        case 'Pack': {\n          return tidy(() => {\n            const axis =\n                getParamValue('axis', node, tensorMap, context) as number;\n            const tensors =\n                getParamValue('tensors', node, tensorMap, context) as Tensor[];\n            // Reshape the tensors to the first tensor's shape if they don't\n            // match.\n            const shape = tensors[0].shape;\n            const squeezedShape = tfOps.squeeze(tensors[0]).shape;\n            const mapped = tensors.map(tensor => {\n              const sameShape = util.arraysEqual(tensor.shape, shape);\n              if (!sameShape &&\n                  !util.arraysEqual(\n                      tfOps.squeeze(tensor).shape, squeezedShape)) {\n                throw new Error('the input tensors shape does not match');\n              }\n              return sameShape ? tensor : tfOps.reshape(tensor, shape);\n            });\n            return [tfOps.stack(mapped, axis)];\n          });\n        }\n        case 'Unpack': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const tensor =\n              getParamValue('tensor', node, tensorMap, context) as Tensor;\n          return tfOps.unstack(tensor, axis);\n        }\n        case 'Tile': {\n          const reps =\n              getParamValue('reps', node, tensorMap, context) as number[];\n          return [tfOps.tile(\n              getParamValue('x', node, tensorMap, context) as Tensor, reps)];\n        }\n        case 'Split':\n        case 'SplitV': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const numOrSizeSplits =\n              getParamValue('numOrSizeSplits', node, tensorMap, context) as\n                  number |\n              number[];\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return tfOps.split(tensor, numOrSizeSplits, axis);\n        }\n        case 'ScatterNd': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          const values =\n              getParamValue('values', node, tensorMap, context) as Tensor;\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          return [tfOps.scatterND(indices, values, shape)];\n        }\n        case 'GatherNd': {\n          const x = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          return [tfOps.gatherND(x, indices)];\n        }\n        case 'SparseToDense': {\n          const indices =\n              getParamValue('sparseIndices', node, tensorMap, context) as\n              Tensor;\n          const shape =\n              getParamValue('outputShape', node, tensorMap, context) as\n              number[];\n          const sparseValues =\n              getParamValue('sparseValues', node, tensorMap, context) as Tensor;\n          const defaultValue =\n              getParamValue('defaultValue', node, tensorMap, context) as Scalar;\n          return [tfOps.sparseToDense(\n              indices, sparseValues, shape,\n              sparseValues.dtype === defaultValue.dtype ?\n                  defaultValue :\n                  tfOps.cast(defaultValue, sparseValues.dtype))];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'slice_join';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'FFT': {\n              return [tfOps.fft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IFFT': {\n              return [tfOps.ifft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'RFFT': {\n              return [tfOps.rfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IRFFT': {\n              return [tfOps.irfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'spectral';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Cast': {\n          return [tfOps.cast(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('dtype', node, tensorMap, context) as 'int32' |\n                  'float32' | 'bool')];\n        }\n        case 'ExpandDims': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.expandDims(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Squeeze': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          return [tfOps.squeeze(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n\n        case 'Reshape': {\n          return [tfOps.reshape(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        case 'PadV2':\n        case 'Pad': {\n          return [tfOps.pad(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('padding', node, tensorMap, context) as\n                  Array<[number, number]>,\n              getParamValue('constantValue', node, tensorMap, context) as\n                  number)];\n        }\n        case 'SpaceToBatchND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const paddings =\n              getParamValue('paddings', node, tensorMap, context) as number[][];\n          return [tfOps.spaceToBatchND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, paddings)];\n        }\n        case 'BatchToSpaceND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const crops =\n              getParamValue('crops', node, tensorMap, context) as number[][];\n          return [tfOps.batchToSpaceND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, crops)];\n        }\n        case 'DepthToSpace': {\n          const blockSize =\n              getParamValue('blockSize', node, tensorMap, context) as number;\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as\n               string).toUpperCase() as 'NHWC' |\n              'NCHW';\n          return [tfOps.depthToSpace(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              blockSize, dataFormat)];\n        }\n        case 'BroadcastTo': {\n          return [tfOps.broadcastTo(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'transformation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {ExecutionContext} from '../executor/execution_context';\n\nimport {NodeValueImpl} from './custom_op/node_value_impl';\nimport {getRegisteredOp} from './custom_op/register';\nimport * as arithmetic from './executors/arithmetic_executor';\nimport * as basicMath from './executors/basic_math_executor';\nimport * as control from './executors/control_executor';\nimport * as convolution from './executors/convolution_executor';\nimport * as creation from './executors/creation_executor';\nimport * as dynamic from './executors/dynamic_executor';\nimport * as evaluation from './executors/evaluation_executor';\nimport * as graph from './executors/graph_executor';\nimport * as image from './executors/image_executor';\nimport * as logical from './executors/logical_executor';\nimport * as matrices from './executors/matrices_executor';\nimport * as normalization from './executors/normalization_executor';\nimport * as reduction from './executors/reduction_executor';\nimport * as sliceJoin from './executors/slice_join_executor';\nimport * as spectral from './executors/spectral_executor';\nimport * as transformation from './executors/transformation_executor';\nimport {Node} from './types';\n\n/**\n * Executes the op defined by the node object.\n * @param node\n * @param tensorMap contains tensors for executed nodes and weights\n */\nexport function executeOp(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): tfc.Tensor[]|Promise<tfc.Tensor[]> {\n  const value =\n      ((node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) => {\n        switch (node.category) {\n          case 'arithmetic':\n            return tfc.tidy(\n                () => arithmetic.executeOp(node, tensorMap, context));\n          case 'basic_math':\n            return tfc.tidy(\n                () => basicMath.executeOp(node, tensorMap, context));\n          case 'control':\n            return control.executeOp(node, tensorMap, context);\n          case 'convolution':\n            return tfc.tidy(\n                () => convolution.executeOp(node, tensorMap, context));\n          case 'creation':\n            return tfc.tidy(() => creation.executeOp(node, tensorMap, context));\n          case 'dynamic':\n            return dynamic.executeOp(node, tensorMap, context);\n          case 'evaluation':\n            return tfc.tidy(\n                () => evaluation.executeOp(node, tensorMap, context));\n          case 'image':\n            return tfc.tidy(() => image.executeOp(node, tensorMap, context));\n          case 'graph':\n            return tfc.tidy(() => graph.executeOp(node, tensorMap, context));\n          case 'logical':\n            return tfc.tidy(() => logical.executeOp(node, tensorMap, context));\n          case 'matrices':\n            return tfc.tidy(() => matrices.executeOp(node, tensorMap, context));\n          case 'normalization':\n            return tfc.tidy(\n                () => normalization.executeOp(node, tensorMap, context));\n          case 'reduction':\n            return tfc.tidy(\n                () => reduction.executeOp(node, tensorMap, context));\n          case 'slice_join':\n            return tfc.tidy(\n                () => sliceJoin.executeOp(node, tensorMap, context));\n          case 'spectral':\n            return tfc.tidy(() => spectral.executeOp(node, tensorMap, context));\n          case 'transformation':\n            return tfc.tidy(\n                () => transformation.executeOp(node, tensorMap, context));\n          case 'custom':\n            const opMapper = getRegisteredOp(node.op);\n            if (opMapper && opMapper.customExecutor) {\n              return opMapper.customExecutor(\n                  new NodeValueImpl(node, tensorMap, context));\n            } else {\n              throw TypeError(`Custom op ${node.op} is not registered.`);\n            }\n          default:\n            throw TypeError(\n                `Unknown op '${node.op}'. File an issue at ` +\n                `https://github.com/tensorflow/tfjs/issues so we can add it` +\n                `, or register a custom execution with tf.registerOp()`);\n        }\n      })(node, tensorMap, context);\n  if (value instanceof Promise) {\n    return value.then((data) => [].concat(data));\n  }\n  return [].concat(value);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap, TensorArrayMap, TensorListMap} from '../data/types';\n\nimport {TensorArray} from './tensor_array';\nimport {TensorList} from './tensor_list';\nimport {FunctionExecutor} from './types';\n\nexport interface ExecutionContextInfo {\n  id: number;           // the unique id of the context info\n  frameName: string;    // The frame name of the loop, this comes from\n                        // the TensorFlow NodeDef.\n  iterationId: number;  // The iteration id of the loop\n}\n\n/**\n * ExecutionContext captures the runtime environment of the node. It keeps\n * track of the current frame and iteration for the control flow ops.\n *\n * For example, typical Dynamic RNN model may contain loops, for which\n * TensorFlow will generate graphs with Enter/Exit nodes to control the\n * current execution frame, and NextIteration Nodes for iteration id increment.\n * For model with branch logic, TensorFLow will generate Switch/Merge ops.\n */\nexport class ExecutionContext {\n  private rootContext = {id: 0, frameName: '', iterationId: 0};\n  private contexts: ExecutionContextInfo[] = [this.rootContext];\n  private lastId = 0;\n  private _currentContextIds: string[];\n\n  constructor(\n      readonly weightMap: NamedTensorsMap = {},\n      readonly tensorArrayMap: TensorArrayMap = {},\n      readonly tensorListMap: TensorListMap = {},\n      readonly functionMap: {[key: string]: FunctionExecutor} = {}) {\n    this.generateCurrentContextIds();\n  }\n\n  private newFrame(id: number, frameName: string) {\n    return {id, frameName, iterationId: 0};\n  }\n\n  /**\n   * Set the current context\n   * @param contexts: ExecutionContextInfo[] the current path of execution\n   * frames\n   */\n  set currentContext(contexts: ExecutionContextInfo[]) {\n    if (this.contexts !== contexts) {\n      this.contexts = contexts;\n      this.generateCurrentContextIds();\n    }\n  }\n\n  get currentContext(): ExecutionContextInfo[] {\n    return this.contexts;\n  }\n\n  /**\n   * Returns the current context in string format.\n   */\n  get currentContextId(): string {\n    return this._currentContextIds[0];\n  }\n\n  /**\n   * Returns the current context and all parent contexts in string format.\n   * This allow access to the nodes in the current and parent frames.\n   */\n  get currentContextIds(): string[] {\n    return this._currentContextIds;\n  }\n\n  private generateCurrentContextIds() {\n    const names = [];\n    for (let i = 0; i < this.contexts.length - 1; i++) {\n      const contexts = this.contexts.slice(0, this.contexts.length - i);\n      names.push(this.contextIdforContexts(contexts));\n    }\n    names.push('');\n    this._currentContextIds = names;\n  }\n\n  private contextIdforContexts(contexts: ExecutionContextInfo[]) {\n    return contexts ?\n        contexts\n            .map(\n                context => (context.id === 0 && context.iterationId === 0) ?\n                    '' :\n                    `${context.frameName}-${context.iterationId}`)\n            .join('/') :\n        '';\n  }\n\n  /**\n   * Enter a new frame, a new context is pushed on the current context list.\n   * @param frameId new frame id\n   */\n  enterFrame(frameId: string) {\n    if (this.contexts) {\n      this.lastId++;\n      this.contexts = this.contexts.slice();\n      this.contexts.push(this.newFrame(this.lastId, frameId));\n      this._currentContextIds.unshift(this.contextIdforContexts(this.contexts));\n    }\n  }\n\n  /**\n   * Exit the current frame, the last context is removed from the current\n   * context list.\n   */\n  exitFrame() {\n    if (this.contexts && this.contexts.length > 1) {\n      this.contexts = this.contexts.slice();\n      this.contexts.splice(-1);\n      this.currentContextIds.shift();\n    } else {\n      throw new Error('Cannot exit frame, the context is empty');\n    }\n  }\n\n  /**\n   * Enter the next iteration of a loop, the iteration id of last context is\n   * increased.\n   */\n  nextIteration() {\n    if (this.contexts && this.contexts.length > 0) {\n      this.contexts = this.contexts.slice();\n      this.lastId++;\n      const context =\n          Object.assign({}, this.contexts[this.contexts.length - 1]);\n      context.iterationId += 1;\n      context.id = this.lastId;\n      this.contexts.splice(-1, 1, context);\n      this._currentContextIds.splice(\n          0, 1, this.contextIdforContexts(this.contexts));\n    } else {\n      throw new Error('Cannot increase frame iteration, the context is empty');\n    }\n  }\n\n  getWeight(name: string): Tensor[] {\n    return this.weightMap[name];\n  }\n\n  addTensorArray(tensorArray: TensorArray) {\n    this.tensorArrayMap[tensorArray.id] = tensorArray;\n  }\n\n  getTensorArray(id: number): TensorArray {\n    return this.tensorArrayMap[id];\n  }\n\n  addTensorList(tensorList: TensorList) {\n    this.tensorListMap[tensorList.id] = tensorList;\n  }\n\n  getTensorList(id: number): TensorList {\n    return this.tensorListMap[id];\n  }\n\n  dispose(keepIds: Set<number>) {\n    for (const key in this.tensorArrayMap) {\n      this.tensorArrayMap[key].clearAndClose(keepIds);\n    }\n\n    for (const key in this.tensorListMap) {\n      this.tensorListMap[key].clearAndClose(keepIds);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NamedTensorMap} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {parseNodeName} from '../operations/executors/utils';\nimport {Graph, Node} from '../operations/types';\n\nexport interface ExecutionInfo {\n  inputs: NamedTensorMap;\n  outputs: Node[];\n  usedNodes: Set<string>;\n  missingInputs: string[];\n  dynamicNode: Node;\n  syncInputs: string[];\n}\n\n/**\n * Given graph inputs and desired outputs, find the minimal set of nodes\n * to execute in order to compute the outputs. In addition return other useful\n * info such:\n * - Missing inputs needed to compute the output.\n * - Whether the subgraph contains dynamic ops (control flow, dynamic shape).\n * - Alternative inputs in order to avoid async (dynamic op) execution.\n */\nexport function getExecutionSubgraph(\n    inputs: NamedTensorMap, outputs: Node[], weightMap: NamedTensorsMap,\n    initNodes?: Node[]): ExecutionInfo {\n  const usedNodes = new Set<string>();\n  const missingInputs: string[] = [];\n  let dynamicNode: Node = null;\n  let syncInputs: string[] = null;\n\n  // Start with the outputs, going backwards and find all the nodes that are\n  // needed to compute those outputs.\n  const seen = new Set<string>();\n  const inputNodeNames =\n      Object.keys(inputs).map(name => parseNodeName(name)[0]);\n\n  let initNodeNames: string[] = [];\n  if (initNodes != null) {\n    initNodeNames = initNodes.map(node => parseNodeName(node.name)[0]);\n  }\n\n  const frontier = [...outputs];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    if (isControlFlow(node) || isDynamicShape(node)) {\n      if (dynamicNode == null) {\n        dynamicNode = node;\n        syncInputs = dynamicNode.children.map(child => child.name)\n                         .filter(name => usedNodes.has(name));\n      }\n    }\n    usedNodes.add(node.name);\n\n    // Weights are dead end since we already have their values.\n    if (weightMap[node.name] != null) {\n      continue;\n    }\n    // This node is a dead end since it's one of the user-provided inputs.\n    if (inputNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    // This node is a dead end since it doesn't have any inputs.\n    if (initNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    if (node.inputs.length === 0) {\n      missingInputs.push(node.name);\n      continue;\n    }\n    node.inputs.forEach(input => {\n      // Don't add to the frontier if it is already there.\n      if (seen.has(input.name)) {\n        return;\n      }\n      seen.add(input.name);\n      frontier.push(input);\n    });\n  }\n  return {inputs, outputs, usedNodes, missingInputs, dynamicNode, syncInputs};\n}\n\n/**\n * Given the execution info, return a list of nodes in topological order that\n * need to be executed to compute the output.\n */\nexport function getNodesInTopologicalOrder(\n    graph: Graph, weightMap: NamedTensorsMap,\n    executionInfo: ExecutionInfo): Node[] {\n  const {usedNodes, inputs} = executionInfo;\n  const frontier: Node[] = [];\n  const inputNodes = Object.keys(inputs)\n                         .map(name => parseNodeName(name)[0])\n                         .map(name => graph.nodes[name]);\n  const initNodes = graph.initNodes;\n\n  inputNodes.forEach(input => {\n    if (usedNodes.has(input.name)) {\n      frontier.push(input);\n    }\n  });\n  graph.weights.forEach(weight => {\n    if (usedNodes.has(weight.name)) {\n      frontier.push(weight);\n    }\n  });\n  if (initNodes != null) {\n    initNodes.forEach(node => {\n      if (usedNodes.has(node.name)) {\n        frontier.push(node);\n      }\n    });\n  }\n  const seen = new Set<string>();\n  const orderedNodes: Node[] = [];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    seen.add(node.name);\n    if (!weightMap[node.name]) {\n      orderedNodes.push(node);\n    }\n    node.children.forEach(child => {\n      if (!seen.has(child.name) && usedNodes.has(child.name) &&\n          child.inputs.every(input => seen.has(input.name))) {\n        frontier.push(child);\n      }\n    });\n  }\n  return orderedNodes;\n}\n\nconst CONTROL_FLOW_OPS = [\n  'Switch', 'Merge', 'Enter', 'Exit', 'NextIteration', 'StatelessIf',\n  'StatelessWhile', 'if', 'While'\n];\nconst DYNAMIC_SHAPE_OPS = [\n  'NonMaxSuppressionV2', 'NonMaxSuppressionV3', 'NonMaxSuppressionV5', 'Where'\n];\n\nexport function isControlFlow(node: Node) {\n  return CONTROL_FLOW_OPS.indexOf(node.op) >= 0;\n}\n\nexport function isDynamicShape(node: Node) {\n  return DYNAMIC_SHAPE_OPS.indexOf(node.op) >= 0;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NamedTensorMap, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {ISignatureDef} from '../data/compiled_api';\nimport {NamedTensorsMap, TensorArrayMap, TensorInfo, TensorListMap} from '../data/types';\nimport {getNodeNameAndIndex, getParamValue, getTensor, getTensorsForCurrentContenxt, parseNodeName} from '../operations/executors/utils';\nimport {executeOp} from '../operations/operation_executor';\nimport {Graph, Node} from '../operations/types';\n\nimport {ExecutionContext, ExecutionContextInfo} from './execution_context';\nimport {getExecutionSubgraph, getNodesInTopologicalOrder, isControlFlow} from './model_analysis';\nimport {FunctionExecutor} from './types';\n\ninterface NodeWithContexts {\n  contexts: ExecutionContextInfo[];\n  node: Node;\n}\n\nexport class GraphExecutor implements FunctionExecutor {\n  private compiledMap: Map<string, Node[]> = new Map();\n  private _weightMap: NamedTensorsMap = {};\n  private _weightIds: number[];\n  private _signature: ISignatureDef;\n  private _inputs: Node[];\n  private _outputs: Node[];\n  private _initNodes: Node[];\n  private SEPERATOR = ',';\n  private _functions: {[key: string]: Graph} = {};\n  private _functionExecutorMap: {[key: string]: FunctionExecutor} = {};\n\n  get weightIds(): number[] {\n    return this.parent ? this.parent.weightIds : this._weightIds;\n  }\n\n  get functionExecutorMap(): {[key: string]: FunctionExecutor} {\n    return this.parent ? this.parent.functionExecutorMap :\n                         this._functionExecutorMap;\n  }\n\n  get weightMap(): NamedTensorsMap {\n    return this.parent ? this.parent.weightMap : this._weightMap;\n  }\n\n  set weightMap(weightMap: NamedTensorsMap) {\n    const weightIds = Object.keys(weightMap).map(\n        key => weightMap[key].map(tensor => tensor.id));\n    this._weightIds = [].concat(...weightIds);\n    this._weightMap = weightMap;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this._inputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get outputs(): TensorInfo[] {\n    return this._outputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get inputNodes(): string[] {\n    return this._inputs.map(node => node.signatureKey || node.name);\n  }\n\n  get outputNodes(): string[] {\n    return this._outputs.map((node) => {\n      const name = node.signatureKey || node.name;\n      return node.defaultOutput ? (`${name}:${node.defaultOutput}`) : name;\n    });\n  }\n\n  get functions(): {[key: string]: ISignatureDef} {\n    return Object.keys(this._functions).reduce((map, key) => {\n      map[key] = this._functions[key].signature;\n      return map;\n    }, {} as {[key: string]: ISignatureDef});\n  }\n\n  /**\n   *\n   * @param graph Graph the model or function graph to be executed.\n   * @param parent When building function exector you need to set the parent\n   * executor. Since the weights and function executor maps are set at parant\n   * level, that function executor can access the function maps and weight maps\n   * through the parent.\n   */\n  constructor(private graph: Graph, private parent?: GraphExecutor) {\n    this._outputs = graph.outputs;\n    this._inputs = graph.inputs;\n    this._initNodes = graph.initNodes;\n    this._signature = graph.signature;\n    this._functions = graph.functions;\n    // create sub-graph executors\n    if (graph.functions != null) {\n      Object.keys(graph.functions).forEach(name => {\n        this._functionExecutorMap[name] =\n            new GraphExecutor(graph.functions[name], this);\n      });\n    }\n  }\n\n  private getCompilationKey(inputs: Node[], outputs: Node[]): string {\n    const sortedInputs = inputs.map(node => node.name).sort();\n    const sortedOutputs = outputs.map(node => node.name).sort();\n    return sortedInputs.join(this.SEPERATOR) + '--' +\n        sortedOutputs.join(this.SEPERATOR);\n  }\n\n  /**\n   * Compiles the inference graph and returns the minimal set of nodes that are\n   * required for execution, in the correct execution order.\n   */\n  private compile(inputs: NamedTensorMap, outputs: Node[]): Node[] {\n    const executionInfo =\n        getExecutionSubgraph(inputs, outputs, this.weightMap, this._initNodes);\n    const {missingInputs, dynamicNode, syncInputs} = executionInfo;\n    if (dynamicNode != null) {\n      throw new Error(\n          `This execution contains the node '${dynamicNode.name}', which has ` +\n          `the dynamic op '${dynamicNode.op}'. Please use ` +\n          `model.executeAsync() instead. Alternatively, to avoid the ` +\n          `dynamic ops, specify the inputs [${syncInputs}]`);\n    }\n\n    if (missingInputs.length > 0) {\n      const outNames = outputs.map(n => n.name);\n      const inNames = Object.keys(inputs);\n      throw new Error(\n          `Cannot compute the outputs [${outNames}] from the provided inputs ` +\n          `[${inNames}]. Missing the following inputs: [${missingInputs}]`);\n    }\n\n    return getNodesInTopologicalOrder(\n        this.graph, this.weightMap, executionInfo);\n  }\n\n  /**\n   * Executes the inference for given input tensors.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  execute(inputs: NamedTensorMap, outputs: string[]): Tensor[] {\n    inputs = this.mapInputs(inputs);\n    const names = Object.keys(inputs).sort();\n    this.checkInputs(inputs);\n    this.checkInputShapeAndType(inputs);\n    outputs = this.mapOutputs(outputs);\n    this.checkOutputs(outputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputs.map(name => parseNodeName(name)[0]);\n    let outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n\n    // If no outputs are specified, then use the default outputs of the model.\n    if (outputNodes.length === 0) {\n      outputNodes = this._outputs;\n    }\n\n    const compilationKey = this.getCompilationKey(inputNodes, outputNodes);\n\n    // Do nothing if the compiled graph cache contains the input.\n    let orderedNodes = this.compiledMap.get(compilationKey);\n    if (orderedNodes == null) {\n      orderedNodes = this.compile(inputs, outputNodes);\n      this.compiledMap.set(compilationKey, orderedNodes);\n    }\n\n    const tensorArrayMap: TensorArrayMap = {};\n    const tensorListMap: TensorListMap = {};\n    return tidy(() => {\n      const context = new ExecutionContext(\n          this.weightMap, tensorArrayMap, tensorListMap,\n          this.functionExecutorMap);\n      const tensorsMap: NamedTensorsMap = {...this.weightMap};\n\n      Object.keys(inputs).forEach(name => {\n        const [nodeName, index] = parseNodeName(name);\n        const tensors: Tensor[] = [];\n        tensors[index] = inputs[name];\n        tensorsMap[nodeName] = tensors;\n      });\n\n      const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n      const intermediateTensorConsumerCount: {[key: number]: number} = {};\n      for (let i = 0; i < orderedNodes.length; i++) {\n        const node = orderedNodes[i];\n        if (!tensorsMap[node.name]) {\n          const tensors = executeOp(node, tensorsMap, context) as Tensor[];\n          if (tensors instanceof Promise) {\n            throw new Error(\n                `The execution of the op '${node.op}' returned a promise. ` +\n                `Please use model.executeAsync() instead.`);\n          }\n          tensorsMap[node.name] = tensors;\n          this.checkTensorForDisposal(\n              node.name, node, tensorsMap, context, tensorsToKeep,\n              outputNodeNames, intermediateTensorConsumerCount);\n        }\n      }\n      // dispose the context for the root executor\n      if (this.parent == null) {\n        context.dispose(tensorsToKeep);\n      }\n      return outputs.map(name => getTensor(name, tensorsMap, context));\n    });\n  }\n\n  private getFrozenTensorIds(tensorMap: NamedTensorsMap): Set<number> {\n    const ids = [].concat.apply(\n        [],\n        Object.keys(tensorMap)\n            .map(key => tensorMap[key])\n            .map(tensors => tensors.map(tensor => tensor.id)));\n    return new Set(ids);\n  }\n  private checkTensorForDisposal(\n      nodeName: string, node: Node, tensorMap: NamedTensorsMap,\n      context: ExecutionContext, tensorsToKeep: Set<number>,\n      outputNames: string[],\n      intermediateTensorConsumerCount: {[key: string]: number}) {\n    // Skip output nodes and any control flow nodes, since its dependency is\n    // tricky to track correctly.\n    if (node.category === 'control' || outputNames.indexOf(nodeName) !== -1) {\n      return;\n    }\n\n    tensorMap[nodeName].forEach(tensor => {\n      if (tensor != null) {\n        intermediateTensorConsumerCount[tensor.id] =\n            (intermediateTensorConsumerCount[tensor.id] || 0) +\n            node.children.length;\n      }\n    });\n    node.inputs.forEach(input => {\n      // Skip any control flow nodes, since its dependency is tricky to track\n      // correctly.\n      if (input.category !== 'control') {\n        const tensors =\n            getTensorsForCurrentContenxt(input.name, tensorMap, context);\n        if (tensors != null) {\n          tensors.forEach(tensor => {\n            if (tensor && !tensorsToKeep.has(tensor.id)) {\n              const count = intermediateTensorConsumerCount[tensor.id];\n              if (count === 1) {\n                tensor.dispose();\n                delete intermediateTensorConsumerCount[tensor.id];\n              } else if (count != null) {\n                // only intermediate nodes has count set, inputs and weights are\n                // not.\n                intermediateTensorConsumerCount[tensor.id]--;\n              }\n            }\n          });\n        }\n      }\n    });\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  async executeAsync(inputs: NamedTensorMap, outputs: string[]):\n      Promise<Tensor[]> {\n    return this._executeAsync(inputs, outputs);\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   * @param isFunctionExecution Flag for executing a function.\n   * @param tensorArrayMap Optional, global TensorArray map by id. Used for\n   * function execution.\n   * @param tensorArrayMap Optinal global TensorList map by id. Used for\n   * function execution.\n   */\n  private async _executeAsync(\n      inputs: NamedTensorMap, outputs: string[], isFunctionExecution = false,\n      tensorArrayMap: TensorArrayMap = {},\n      tensorListMap: TensorListMap = {}): Promise<Tensor[]> {\n    if (!isFunctionExecution) {\n      inputs = this.mapInputs(inputs);\n      this.checkInputs(inputs);\n      this.checkInputShapeAndType(inputs);\n      outputs = this.mapOutputs(outputs);\n      this.checkOutputs(outputs);\n    }\n\n    const context = new ExecutionContext(\n        this.weightMap, tensorArrayMap, tensorListMap,\n        this.functionExecutorMap);\n\n    // Graph with control flow op requires runtime evaluation of the execution\n    // order, while without control flow the execution order is pre-determined\n    // in the compile method.\n    const tensorMap = await this.executeWithControlFlow(\n        inputs, context, outputs, isFunctionExecution);\n    const results = outputs.map(name => getTensor(name, tensorMap, context));\n\n    // dispose all the intermediate tensors\n    const outputIds = results.map(t => t.id);\n    const inputIds = Object.keys(inputs).map(name => inputs[name].id);\n    const keepIds =\n        new Set<number>([...outputIds, ...inputIds, ...this.weightIds]);\n    Object.keys(tensorMap).forEach(key => {\n      const tensorArray = tensorMap[key];\n      tensorArray.forEach(tensor => {\n        if (tensor && !tensor.isDisposed && !keepIds.has(tensor.id)) {\n          tensor.dispose();\n        }\n      });\n    });\n    // dispose the context for the root executor\n    if (this.parent == null) {\n      context.dispose(keepIds);\n    }\n\n    return results;\n  }\n\n  async executeFunctionAsync(\n      inputs: Tensor[], tensorArrayMap: TensorArrayMap,\n      tensorListMap: TensorListMap): Promise<Tensor[]> {\n    const mappedInputs = inputs.reduce((map, tensor, index) => {\n      map[this.inputs[index].name] = tensor;\n      return map;\n    }, {} as NamedTensorMap);\n\n    return this._executeAsync(\n        mappedInputs, this.outputNodes, true, tensorArrayMap, tensorListMap);\n  }\n  /**\n   * When there are control flow nodes in the graph, the graph execution use\n   * ExecutionContext to keep track of the frames and loop iterators.\n   * @param inputs placeholder tensors for the graph.\n   * @param context the execution context object for current execution.\n   * @param isFunctionExecution Flag for executing a function.\n   */\n  private async executeWithControlFlow(\n      inputs: NamedTensorMap, context: ExecutionContext, outputNames: string[],\n      isFunctionExecution: boolean): Promise<NamedTensorsMap> {\n    const names = Object.keys(inputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputNames.map(name => parseNodeName(name)[0]);\n    const outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n    const {usedNodes, missingInputs, dynamicNode, syncInputs} =\n        getExecutionSubgraph(inputs, outputNodes, this.weightMap);\n\n    const stack: NodeWithContexts[] =\n        [...inputNodes, ...this.graph.weights].map(node => {\n          return {node, contexts: context.currentContext};\n        });\n    const tensorsMap: NamedTensorsMap = {...this.weightMap};\n    Object.keys(inputs).forEach(name => {\n      const [nodeName, index] = parseNodeName(name);\n      const tensors: Tensor[] = [];\n      tensors[index] = inputs[name];\n      tensorsMap[nodeName] = tensors;\n    });\n    const intermediateTensorConsumerCount: {[key: number]: number} = {};\n    const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n    const added: {[key: string]: boolean} = {};\n    while (stack.length > 0) {\n      const promises = this.processStack(\n          inputNodes, stack, context, tensorsMap, added, tensorsToKeep,\n          outputNodeNames, intermediateTensorConsumerCount, usedNodes);\n      await Promise.all(promises);\n    }\n    if (dynamicNode == null && !isFunctionExecution) {\n      console.warn(\n          `This model execution did not contain any nodes with control flow ` +\n          `or dynamic output shapes. You can use model.execute() instead.`);\n    }\n    const missingOutputs =\n        outputNodes\n            .filter(\n                node => !isControlFlow(node) &&\n                    !getTensor(node.name, tensorsMap, context))\n            .map(node => node.name);\n    if (missingOutputs.length > 0) {\n      let alternativeMsg = '';\n      if (dynamicNode != null) {\n        alternativeMsg =\n            `Alternatively, to avoid the dynamic ops, use model.execute() ` +\n            `and specify the inputs [${syncInputs}]`;\n      }\n      throw new Error(\n          `Cannot compute the outputs [${missingOutputs}] from the provided ` +\n          `inputs [${names}]. Consider providing the following inputs: ` +\n          `[${missingInputs}]. ${alternativeMsg}`);\n    }\n    return tensorsMap;\n  }\n\n  private processStack(\n      inputNodes: Node[], stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      tensorsToKeep: Set<number>, outputNames: string[],\n      intermediateTensorConsumerCount: {[key: number]: number},\n      usedNodes: Set<string>) {\n    const promises: Array<Promise<Tensor[]>> = [];\n    while (stack.length > 0) {\n      const item = stack.pop();\n      context.currentContext = item.contexts;\n      let nodeName = '';\n      // The tensor of the Enter op with isConstant set should be set\n      // in the parent scope, so it will be available as constant for the\n      // whole loop.\n      if (item.node.op === 'Enter' &&\n          getParamValue('isConstant', item.node, tensorMap, context)) {\n        [nodeName] = getNodeNameAndIndex(item.node.name, context);\n      }\n\n      // only process nodes that are not provided as input nodes.\n      if (inputNodes.indexOf(item.node) === -1) {\n        const tensors = executeOp(item.node, tensorMap, context);\n        if (!nodeName) {\n          [nodeName] = getNodeNameAndIndex(item.node.name, context);\n        }\n        const currentContext = context.currentContext;\n        if (tensors instanceof Promise) {\n          promises.push(tensors.then(t => {\n            tensorMap[nodeName] = t;\n            context.currentContext = currentContext;\n            this.checkTensorForDisposal(\n                nodeName, item.node, tensorMap, context, tensorsToKeep,\n                outputNames, intermediateTensorConsumerCount);\n            this.processChildNodes(\n                item.node, stack, context, tensorMap, added, usedNodes);\n            return t;\n          }));\n        } else {\n          tensorMap[nodeName] = tensors;\n          this.checkTensorForDisposal(\n              nodeName, item.node, tensorMap, context, tensorsToKeep,\n              outputNames, intermediateTensorConsumerCount);\n          this.processChildNodes(\n              item.node, stack, context, tensorMap, added, usedNodes);\n        }\n      } else {\n        this.processChildNodes(\n            item.node, stack, context, tensorMap, added, usedNodes);\n      }\n    }\n    return promises;\n  }\n\n  private processChildNodes(\n      node: Node, stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      usedNodes: Set<string>) {\n    node.children.forEach((childNode) => {\n      const [nodeName, ] = getNodeNameAndIndex(childNode.name, context);\n      if (added[nodeName] || !usedNodes.has(childNode.name)) {\n        return;\n      }\n      // Merge op can be pushed if any of its inputs has value.\n      if (childNode.op === 'Merge') {\n        if (childNode.inputNames.some(name => {\n              return !!getTensor(name, tensorMap, context);\n            })) {\n          added[nodeName] = true;\n          stack.push({contexts: context.currentContext, node: childNode});\n        }\n      } else  // Otherwise all inputs must to have value.\n          if (childNode.inputNames.every(name => {\n                return !!getTensor(name, tensorMap, context);\n              })) {\n        added[nodeName] = true;\n        stack.push({contexts: context.currentContext, node: childNode});\n      }\n    });\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   */\n  dispose() {\n    Object.keys(this.weightMap)\n        .forEach(\n            key => this.weightMap[key].forEach(tensor => tensor.dispose()));\n  }\n\n  private checkInputShapeAndType(inputs: NamedTensorMap) {\n    Object.keys(inputs).forEach(name => {\n      const input = inputs[name];\n      const [nodeName, ] = parseNodeName(name);\n      const node = this.graph.nodes[nodeName];\n      if (node.attrParams['shape'] && node.attrParams['shape'].value) {\n        const shape = node.attrParams['shape'].value as number[];\n        const match = shape.length === input.shape.length &&\n            input.shape.every(\n                (dim, index) => shape[index] === -1 || shape[index] === dim);\n        util.assert(\n            match,\n            () => `The shape of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be [${shape}], but was ` +\n                `[${input.shape}]`);\n      }\n      if (node.attrParams['dtype'] && node.attrParams['dtype'].value) {\n        util.assert(\n            input.dtype === node.attrParams['dtype'].value as string,\n            () => `The dtype of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be ` +\n                `${node.attrParams['dtype'].value}, but was ${input.dtype}`);\n      }\n    });\n  }\n\n  private mapInputs(inputs: NamedTensorMap) {\n    const result: NamedTensorMap = {};\n    for (const inputName in inputs) {\n      if (this._signature != null && this._signature.inputs != null &&\n          this._signature.inputs[inputName] != null) {\n        const tensor = this._signature.inputs[inputName];\n        result[tensor.name] = inputs[inputName];\n      } else {\n        result[inputName] = inputs[inputName];\n      }\n    }\n    return result;\n  }\n\n  private checkInputs(inputs: NamedTensorMap) {\n    const notInGraph = Object.keys(inputs).filter(name => {\n      const [nodeName] = parseNodeName(name);\n      return this.graph.nodes[nodeName] == null;\n    });\n    if (notInGraph.length > 0) {\n      throw new Error(\n          `The dict provided in model.execute(dict) has ` +\n          `keys: [${notInGraph}] that are not part of graph`);\n    }\n  }\n\n  private mapOutputs(outputs: string[]) {\n    return outputs.map(name => {\n      if (this._signature != null && this._signature.outputs != null &&\n          this._signature.outputs[name] != null) {\n        const tensor = this._signature.outputs[name];\n        return tensor.name;\n      }\n      return name;\n    }, {});\n  }\n\n  private checkOutputs(outputs: string[]): void {\n    outputs.forEach(name => {\n      const [normalizedName] = parseNodeName(name);\n      if (!this.graph.nodes[normalizedName]) {\n        throw new Error(`The output '${name}' is not found in the graph`);\n      }\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {InferenceModel, io, ModelPredictConfig, NamedTensorMap, Tensor} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\nimport {NamedTensorsMap, TensorInfo} from '../data/types';\nimport {OperationMapper} from '../operations/operation_mapper';\n\nimport {GraphExecutor} from './graph_executor';\n\nexport const TFHUB_SEARCH_PARAM = '?tfjs-format=file';\nexport const DEFAULT_MODEL_NAME = 'model.json';\n/**\n * A `tf.GraphModel` is a directed, acyclic graph built from a\n * SavedModel GraphDef and allows inference execution.\n *\n * A `tf.GraphModel` can only be created by loading from a model converted from\n * a [TensorFlow SavedModel](https://www.tensorflow.org/guide/saved_model) using\n * the command line converter tool and loaded via `tf.loadGraphModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class GraphModel implements InferenceModel {\n  private executor: GraphExecutor;\n  private version = 'n/a';\n  private handler: io.IOHandler;\n  private artifacts: io.ModelArtifacts;\n  private initializer: GraphExecutor;\n\n  // Returns the version information for the tensorflow model GraphDef.\n  get modelVersion(): string {\n    return this.version;\n  }\n\n  get inputNodes(): string[] {\n    return this.executor.inputNodes;\n  }\n\n  get outputNodes(): string[] {\n    return this.executor.outputNodes;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this.executor.inputs;\n  }\n\n  get outputs(): TensorInfo[] {\n    return this.executor.outputs;\n  }\n\n  get weights(): NamedTensorsMap {\n    return this.executor.weightMap;\n  }\n\n  /**\n   * @param modelUrl url for the model, or an `io.IOHandler`.\n   * @param weightManifestUrl url for the weight file generated by\n   * scripts/convert.py script.\n   * @param requestOption options for Request, which allows to send credentials\n   * and custom headers.\n   * @param onProgress Optional, progress callback function, fired periodically\n   * before the load is completed.\n   */\n  constructor(\n      private modelUrl: string|io.IOHandler,\n      private loadOptions: io.LoadOptions = {}) {\n    if (loadOptions == null) {\n      this.loadOptions = {};\n    }\n  }\n\n  private findIOHandler() {\n    const path = this.modelUrl;\n    if ((path as io.IOHandler).load != null) {\n      // Path is an IO Handler.\n      this.handler = path as io.IOHandler;\n    } else if (this.loadOptions.requestInit != null) {\n      this.handler = io.browserHTTPRequest(path as string, this.loadOptions);\n    } else {\n      const handlers = io.getLoadHandlers(path as string, this.loadOptions);\n      if (handlers.length === 0) {\n        // For backward compatibility: if no load handler can be found,\n        // assume it is a relative http path.\n        handlers.push(io.browserHTTPRequest(path as string, this.loadOptions));\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) load handlers for ` +\n            `URL '${[path]}'`);\n      }\n      this.handler = handlers[0];\n    }\n  }\n\n  /**\n   * Loads the model and weight files, construct the in memory weight map and\n   * compile the inference graph.\n   */\n  async load(): Promise<boolean> {\n    this.findIOHandler();\n    if (this.handler.load == null) {\n      throw new Error(\n          'Cannot proceed with model loading because the IOHandler provided ' +\n          'does not have the `load` method implemented.');\n    }\n    const artifacts = await this.handler.load();\n\n    return this.loadSync(artifacts);\n  }\n\n  /**\n   * Synchronously construct the in memory weight map and\n   * compile the inference graph. Also initialize hashtable if any.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  loadSync(artifacts: io.ModelArtifacts) {\n    this.artifacts = artifacts;\n    const graph = this.artifacts.modelTopology as tensorflow.IGraphDef;\n    let signature = {};\n    if (this.artifacts.userDefinedMetadata != null) {\n      signature =  // tslint:disable-next-line:no-any\n          (this.artifacts.userDefinedMetadata as any).signature as\n          tensorflow.ISignatureDef;\n    }\n\n    this.version = `${graph.versions.producer}.${graph.versions.minConsumer}`;\n    const weightMap =\n        io.decodeWeights(this.artifacts.weightData, this.artifacts.weightSpecs);\n    this.executor = new GraphExecutor(\n        OperationMapper.Instance.transformGraph(graph, signature));\n    this.executor.weightMap = this.convertTensorMapToTensorsMap(weightMap);\n\n    if (artifacts.modelInitializer != null) {\n      const initializer =\n          OperationMapper.Instance.transformGraph(artifacts.modelInitializer);\n      this.initializer = new GraphExecutor(initializer);\n      this.initializer.weightMap = this.executor.weightMap;\n      this.initializer.execute({}, []);\n    }\n\n    return true;\n  }\n\n  /**\n   * Save the configuration and/or weights of the GraphModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const modelUrl =\n   *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n   * const model = await tf.loadGraphModel(modelUrl);\n   * const zeros = tf.zeros([1, 224, 224, 3]);\n   * model.predict(zeros).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * model.predict(zeros).print();\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new Error(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new Error(\n          'GraphModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    return handlerOrURL.save(this.artifacts);\n  }\n\n  /**\n   * Execute the inference for the input tensors.\n   *\n   * @param input The input tensors, when there is single input for the model,\n   * inputs param should be a `tf.Tensor`. For models with mutliple inputs,\n   * inputs params should be in either `tf.Tensor`[] if the input order is\n   * fixed, or otherwise NamedTensorMap format.\n   *\n   * For model with multiple inputs, we recommend you use NamedTensorMap as the\n   * input type, if you use `tf.Tensor`[], the order of the array needs to\n   * follow the\n   * order of inputNodes array. @see {@link GraphModel.inputNodes}\n   *\n   * You can also feed any intermediate nodes using the NamedTensorMap as the\n   * input type. For example, given the graph\n   *    InputNode => Intermediate => OutputNode,\n   * you can execute the subgraph Intermediate => OutputNode by calling\n   *    model.execute('IntermediateNode' : tf.tensor(...));\n   *\n   * This is useful for models that uses tf.dynamic_rnn, where the intermediate\n   * state needs to be fed manually.\n   *\n   * For batch inference execution, the tensors for each input need to be\n   * concatenated together. For example with mobilenet, the required input shape\n   * is [1, 244, 244, 3], which represents the [batch, height, width, channel].\n   * If we are provide a batched data of 100 images, the input tensor should be\n   * in the shape of [100, 244, 244, 3].\n   *\n   * @param config Prediction configuration for specifying the batch size and\n   * output node names. Currently the batch size option is ignored for graph\n   * model.\n   *\n   * @returns Inference result tensors. The output would be single `tf.Tensor`\n   * if model has single output node, otherwise Tensor[] or NamedTensorMap[]\n   * will be returned for model with multiple outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(inputs: Tensor|Tensor[]|NamedTensorMap, config?: ModelPredictConfig):\n      Tensor|Tensor[]|NamedTensorMap {\n    return this.execute(inputs, this.outputNodes);\n  }\n\n  private normalizeInputs(inputs: Tensor|Tensor[]|\n                          NamedTensorMap): NamedTensorMap {\n    if (!(inputs instanceof Tensor) && !Array.isArray(inputs)) {\n      // The input is already a NamedTensorMap.\n      return inputs;\n    }\n    inputs = Array.isArray(inputs) ? inputs : [inputs];\n    if (inputs.length !== this.inputNodes.length) {\n      throw new Error(\n          'Input tensor count mismatch,' +\n          `the graph model has ${this.inputNodes.length} placeholders, ` +\n          `while there are ${inputs.length} input tensors.`);\n    }\n    return this.inputNodes.reduce((map, inputName, i) => {\n      map[inputName] = (inputs as Tensor[])[i];\n      return map;\n    }, {} as NamedTensorMap);\n  }\n\n  private normalizeOutputs(outputs: string|string[]): string[] {\n    outputs = outputs || this.outputNodes;\n    return !Array.isArray(outputs) ? [outputs] : outputs;\n  }\n\n  /**\n   * Executes inference for the model for given input tensors.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no\n   * outputs are specified, the default outputs of the model would be used.\n   * You can inspect intermediate nodes of the model by adding them to the\n   * outputs array.\n   *\n   * @returns A single tensor if provided with a single output or no outputs\n   * are provided and there is only one default output, otherwise return a\n   * tensor array. The order of the tensor array is the same as the outputs\n   * if provided, otherwise the order of outputNodes attribute of the model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs?: string|string[]):\n      Tensor|Tensor[] {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = this.executor.execute(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n  /**\n   * Executes inference for the model for given input tensors in async\n   * fashion, use this method when your model contains control flow ops.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   *\n   * @returns A Promise of single tensor if provided with a single output or\n   * no outputs are provided and there is only one default output, otherwise\n   * return a tensor map.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async executeAsync(\n      inputs: Tensor|Tensor[]|NamedTensorMap,\n      outputs?: string|string[]): Promise<Tensor|Tensor[]> {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = await this.executor.executeAsync(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n\n  private convertTensorMapToTensorsMap(map: NamedTensorMap): NamedTensorsMap {\n    return Object.keys(map).reduce((newMap: NamedTensorsMap, key) => {\n      newMap[key] = [map[key]];\n      return newMap;\n    }, {});\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  dispose() {\n    this.executor.dispose();\n\n    if (this.initializer) {\n      this.initializer.dispose();\n    }\n  }\n}\n\n/**\n * Load a graph model given a URL to the model definition.\n *\n * Example of loading MobileNetV2 from a URL and making a prediction with a\n * zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n * const model = await tf.loadGraphModel(modelUrl);\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n *\n * Example of loading MobileNetV2 from a TF Hub URL and making a prediction with\n * a zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/2';\n * const model = await tf.loadGraphModel(modelUrl, {fromTFHub: true});\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n * @param modelUrl The url or an `io.IOHandler` that loads the model.\n * @param options Options for the HTTP request, which allows to send credentials\n *    and custom headers.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport async function loadGraphModel(\n    modelUrl: string|io.IOHandler,\n    options: io.LoadOptions = {}): Promise<GraphModel> {\n  if (modelUrl == null) {\n    throw new Error(\n        'modelUrl in loadGraphModel() cannot be null. Please provide a url ' +\n        'or an IOHandler that loads the model');\n  }\n  if (options == null) {\n    options = {};\n  }\n\n  if (options.fromTFHub) {\n    if ((modelUrl as io.IOHandler).load == null) {\n      if (!(modelUrl as string).endsWith('/')) {\n        modelUrl = (modelUrl as string) + '/';\n      }\n      modelUrl = `${modelUrl}${DEFAULT_MODEL_NAME}${TFHUB_SEARCH_PARAM}`;\n    }\n  }\n  const model = new GraphModel(modelUrl, options);\n  await model.load();\n  return model;\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\n\n// tslint:disable:no-any\n\n/**\n * A return value for a mapping function that can be applied via deepMap.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapResult = {\n  value: any,\n  recurse: boolean\n};\n\n/**\n * Apply a mapping function to a nested structure in a recursive manner.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapResult`.  The `DeepMapResult` either provides a\n *   replacement value for that node (i.e., replacing the subtree), or indicates\n *   that the node should be processed recursively.\n */\nexport function deepMap(input: any, mapFn: (x: any) => DeepMapResult): any|\n    any[] {\n  return deepMapInternal(input, mapFn);\n}\n\n/**\n * @param seen: A Map of known object mappings (i.e., memoized results of\n *   `mapFn()`)\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepMapInternal(\n    input: any, mapFn: (x: any) => DeepMapResult,\n    seen: Map<any, any> = new Map(), containedIn: Set<{}> = new Set()): any|\n    any[] {\n  if (input == null) {\n    return null;\n  }\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  if (seen.has(input)) {\n    return seen.get(input);\n  }\n  const result = mapFn(input);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep map function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    seen.set(input, result.value);\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const child = input[k];\n      const childResult = deepMapInternal(child, mapFn, seen, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// TODO(soergel, kangyizhang) Reconsider naming of deepZip() to avoid confusion\n// with zip()\n\n/**\n * Zip nested structures together in a recursive manner.\n *\n * This has the effect of transposing or pivoting data, e.g. converting it from\n * a row-major representation to a column-major representation.\n *\n * For example, `deepZip([{a: 1, b: 2}, {a: 3, b: 4}])` returns\n * `{a: [1, 3], b: [2, 4]}`.\n *\n * The inputs should all have the same nested structure (i.e., of arrays and\n * dicts).  The result is a single object with the same nested structure, where\n * the leaves are arrays collecting the values of the inputs at that location\n * (or, optionally, the result of a custom function applied to those arrays).\n *\n * @param inputs: An array of the objects to zip together.\n * @param zipFn: (optional) A function that expects an array of elements at a\n *   single node of the object tree, and returns a `DeepMapResult`.  The\n *   `DeepMapResult` either provides a result value for that node (i.e.,\n *   representing the subtree), or indicates that the node should be processed\n *   recursively.  The default zipFn recurses as far as possible and places\n *   arrays at the leaves.\n */\nexport function deepZip(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult = zipToList): any|any[] {\n  return deepZipInternal(inputs, zipFn);\n}\n\n/**\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepZipInternal(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult,\n    containedIn: Set<{}> = new Set()): any|any[] {\n  // The recursion follows the structure of input 0; it's assumed that all the\n  // other inputs have the same structure.\n  const input = inputs[0];\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  const result = zipFn(inputs);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep zip function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const children = inputs.map(x => x[k]);\n      const childResult = deepZipInternal(children, zipFn, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// tslint:disable-next-line:no-any\nexport function zipToList(x: any[]): DeepMapResult {\n  if (x === null) {\n    return null;\n  }\n  // TODO(soergel): validate array type?\n\n  if (isIterable(x[0])) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: x, recurse: false};\n  }\n}\n\n/**\n * A return value for an async map function for use with deepMapAndAwaitAll.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapAsyncResult = {\n  value: Promise<any>,\n  recurse: boolean\n};\n\n/**\n * Apply an async mapping function to a nested structure in a recursive manner.\n *\n * This first creates a nested structure of Promises, and then awaits all of\n * those, resulting in a single Promise for a resolved nested structure.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapAsyncResult`.  The `DeepMapAsyncResult` either provides\n *   a `Promise` for a replacement value for that node (i.e., replacing the\n *   subtree), or indicates that the node should be processed recursively.  Note\n *   that the decision whether or not to recurse must be made immediately; only\n *   the mapped value may be promised.\n */\nexport async function deepMapAndAwaitAll(\n    input: any, mapFn: (x: any) => DeepMapAsyncResult): Promise<any|any[]> {\n  const seen: Map<any, Promise<any>> = new Map();\n\n  // First do a normal deepMap, collecting Promises in 'seen' as a side effect.\n  deepMapInternal(input, mapFn, seen);\n\n  // Replace the Promises in 'seen' in place.\n  // Note TypeScript provides no async map iteration, and regular map iteration\n  // is broken too, so sadly we have to do Array.from() to make it work.\n  // (There's no advantage to Promise.all(), and that would be tricky anyway.)\n  for (const key of Array.from(seen.keys())) {\n    const value = seen.get(key);\n    if (value instanceof Promise) {\n      const mappedValue = await value;\n      seen.set(key, mappedValue);\n    }\n  }\n\n  // Normal deepMap again, this time filling in the resolved values.\n  // It's unfortunate that we have to do two passes.\n  // TODO(soergel): test performance and think harder about a fast solution.\n  const result = deepMapInternal(input, mapFn, seen);\n  return result;\n}\n\n/**\n * Determine whether the argument is iterable.\n *\n * @returns true if the argument is an array or any non-Tensor object.\n */\n// tslint:disable-next-line:no-any\nexport function isIterable(obj: any): boolean {\n  return obj != null && (!ArrayBuffer.isView(obj)) &&\n      (Array.isArray(obj) ||\n       (typeof obj === 'object' && !(obj instanceof tf.Tensor)));\n}\n\n/**\n * Determine whether the argument can be converted to Tensor.\n *\n * Tensors, primitives, arrays, and TypedArrays all qualify; anything else does\n * not.\n *\n * @returns true if the argument can be converted to Tensor.\n */\n// tslint:disable-next-line:no-any\nexport function canTensorify(obj: any): boolean {\n  return obj == null || isPrimitive(obj) || Array.isArray(obj) ||\n      (typeof obj === 'object' && (obj instanceof tf.Tensor)) ||\n      tf.util.isTypedArray(obj);\n}\n\n/**\n * Returns true if the given `value` is a primitive type. Otherwise returns\n * false. This is equivalant to node util.isPrimitive\n */\nfunction isPrimitive(value: any): boolean {\n  return (\n      value === null ||\n      (typeof value !== 'object' && typeof value !== 'function'));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {deepMap, DeepMapResult, isIterable} from './deep_map';\n\nexport function deepClone<T>(container: T): T {\n  return deepMap(container, cloneIfTensor);\n}\n\n// tslint:disable-next-line: no-any\nfunction cloneIfTensor(item: any): DeepMapResult {\n  if (item instanceof tf.Tensor) {\n    return ({value: item.clone(), recurse: false});\n  } else if (isIterable(item)) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: item, recurse: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/**\n * A ring buffer, providing O(1) FIFO, LIFO, and related operations.\n */\nexport class RingBuffer<T> {\n  // Note we store the indices in the range 0 <= index < 2*capacity.\n  // This allows us to distinguish the full from the empty case.\n  // See https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/\n  protected begin = 0;  // inclusive\n  protected end = 0;    // exclusive\n  protected doubledCapacity: number;\n\n  protected data: T[];\n\n  /**\n   * Constructs a `RingBuffer`.\n   * @param capacity The number of items that the buffer can accomodate.\n   */\n  constructor(public capacity: number) {\n    if (capacity == null) {\n      throw new RangeError('Can\\'t create a ring buffer of unknown capacity.');\n    }\n    if (capacity < 1) {\n      throw new RangeError('Can\\'t create ring buffer of capacity < 1.');\n    }\n    this.data = new Array<T>(capacity);\n    this.doubledCapacity = 2 * capacity;\n  }\n\n  /**\n   * Map any index into the range 0 <= index < 2*capacity.\n   */\n  protected wrap(index: number) {\n    // don't trust % on negative numbers\n    while (index < 0) {\n      index += this.doubledCapacity;\n    }\n    return index % this.doubledCapacity;\n  }\n\n  protected get(index: number) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t get item at a negative index.');\n    }\n    return this.data[index % this.capacity];\n  }\n\n  protected set(index: number, value: T) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t set item at a negative index.');\n    }\n    this.data[index % this.capacity] = value;\n  }\n\n  /**\n   * Returns the current number of items in the buffer.\n   */\n  length(): number {\n    let length = this.end - this.begin;\n    if (length < 0) {\n      length = this.doubledCapacity + length;\n    }\n    return length;\n  }\n\n  /**\n   * Reports whether the buffer is full.\n   * @returns true if the number of items in the buffer equals its capacity, and\n   *   false otherwise.\n   */\n  isFull() {\n    return this.length() === this.capacity;\n  }\n\n  /**\n   * Reports whether the buffer is empty.\n   * @returns true if the number of items in the buffer equals zero, and\n   *   false otherwise.\n   */\n  isEmpty() {\n    return this.length() === 0;\n  }\n\n  /**\n   * Adds an item to the end of the buffer.\n   */\n  push(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.set(this.end, value);\n    this.end = this.wrap(this.end + 1);\n  }\n\n  /**\n   * Adds many items to the end of the buffer, in order.\n   */\n  pushAll(values: T[]) {\n    for (const value of values) {\n      this.push(value);\n    }\n  }\n\n  /**\n   * Removes and returns the last item in the buffer.\n   */\n  pop(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    this.end = this.wrap(this.end - 1);\n    const result = this.get(this.end);\n    this.set(this.end, undefined);\n    return result;\n  }\n\n  /**\n   * Adds an item to the beginning of the buffer.\n   */\n  unshift(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.begin = this.wrap(this.begin - 1);\n    this.set(this.begin, value);\n  }\n\n  /**\n   * Removes and returns the first item in the buffer.\n   */\n  shift(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const result = this.get(this.begin);\n    this.set(this.begin, undefined);\n    this.begin = this.wrap(this.begin + 1);\n    return result;\n  }\n\n  /**\n   * Removes and returns a specific item in the buffer, and moves the last item\n   * to the vacated slot.  This is useful for implementing a shuffling stream.\n   * Note that this operation necessarily scrambles the original order.\n   *\n   * @param relativeIndex: the index of the item to remove, relative to the\n   *   first item in the buffer (e.g., hiding the ring nature of the underlying\n   *   storage).\n   */\n  shuffleExcise(relativeIndex: number): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const index = this.wrap(this.begin + relativeIndex);\n    const result = this.get(index);\n    this.set(index, this.pop());\n    return result;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {RingBuffer} from './ring_buffer';\n\nexport class GrowingRingBuffer<T> extends RingBuffer<T> {\n  private static INITIAL_CAPACITY = 32;\n\n  /**\n   * Constructs a `GrowingRingBuffer`.\n   */\n  constructor() {\n    super(GrowingRingBuffer.INITIAL_CAPACITY);\n  }\n\n  isFull() {\n    return false;\n  }\n\n  push(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.push(value);\n  }\n\n  unshift(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.unshift(value);\n  }\n\n  /**\n   * Doubles the capacity of the buffer.\n   */\n  private expand() {\n    const newCapacity = this.capacity * 2;\n    const newData = new Array<T>(newCapacity);\n    const len = this.length();\n\n    // Rotate the buffer to start at index 0 again, since we can't just\n    // allocate more space at the end.\n    for (let i = 0; i < len; i++) {\n      newData[i] = this.get(this.wrap(this.begin + i));\n    }\n\n    this.data = newData;\n    this.capacity = newCapacity;\n    this.doubledCapacity = 2 * this.capacity;\n    this.begin = 0;\n    this.end = len;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {Container} from '../types';\nimport {deepClone} from '../util/deep_clone';\nimport {deepMapAndAwaitAll, DeepMapAsyncResult, DeepMapResult, deepZip, zipToList} from '../util/deep_map';\nimport {GrowingRingBuffer} from '../util/growing_ring_buffer';\nimport {RingBuffer} from '../util/ring_buffer';\n\n/**\n * A nested structure of LazyIterators, used as the input to zip().\n */\nexport type IteratorContainer = Container<LazyIterator<tf.TensorContainer>>;\n\n// Here we implement a simple asynchronous iterator.\n// This lets us avoid using either third-party stream libraries or\n// recent TypeScript language support requiring polyfills.\n\n/**\n * Create a `LazyIterator` from an array of items.\n */\nexport function iteratorFromItems<T>(items: T[]): LazyIterator<T> {\n  return new ArrayIterator(items);\n}\n\n/**\n * Create a `LazyIterator` of incrementing integers.\n */\nexport function iteratorFromIncrementing(start: number): LazyIterator<number> {\n  let i = start;\n  return iteratorFromFunction(() => ({value: i++, done: false}));\n}\n\n/**\n * Create a `LazyIterator` from a function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * await iter.forEachAsync(e => console.log(e));\n * ```\n *\n * @param func A function that produces data on each call.\n */\nexport function iteratorFromFunction<T>(\n    func: () =>\n        IteratorResult<T>| Promise<IteratorResult<T>>): LazyIterator<T> {\n  return new FunctionCallIterator(func);\n}\n\n/**\n * Create a `LazyIterator` by concatenating underlying streams, which are\n * themselves provided as a stream.\n *\n * This can also be thought of as a \"stream flatten\" operation.\n *\n * @param baseIterators A stream of streams to be concatenated.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenated<T>(\n    baseIterators: LazyIterator<LazyIterator<T>>,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return new ChainedIterator(baseIterators, baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by concatenating streams produced by calling a\n * stream-generating function a given number of times.\n *\n * Since a `LazyIterator` is read-once, it cannot be repeated, but this\n * function can be used to achieve a similar effect:\n *\n *   LazyIterator.ofConcatenatedFunction(() => new MyIterator(), 6);\n *\n * @param iteratorFunc: A function that produces a new stream on each call.\n * @param count: The number of times to call the function.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenatedFunction<T>(\n    iteratorFunc: () => IteratorResult<LazyIterator<T>>, count: number,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return iteratorFromConcatenated(\n      iteratorFromFunction(iteratorFunc).take(count), baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by zipping together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nexport function iteratorFromZipped<O extends tf.TensorContainer>(\n    iterators: IteratorContainer,\n    mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL): LazyIterator<O> {\n  return new ZipIterator<O>(iterators, mismatchMode);\n}\n\n/**\n * An asynchronous iterator, providing lazy access to a potentially\n * unbounded stream of elements.\n *\n * Iterator can be obtained from a dataset:\n * `const iter = await dataset.iterator();`\n */\nexport abstract class LazyIterator<T> {\n  // This class implements AsyncIterator<T>, but we have not yet set the\n  // TypeScript --downlevelIteration flag to enable that.\n\n  abstract summary(): string;\n\n  /**\n   * Returns a `Promise` for the next element in the stream.\n   *\n   * When an item can be provided successfully, the return value is\n   * `{value:T, done:false}`.\n   *\n   * Calling next() on a closed stream returns `{value:null, done:true}`.\n   */\n  abstract async next(): Promise<IteratorResult<T>>;\n\n  /**\n   * Collect all remaining elements of a bounded stream into an array.\n   * Obviously this will succeed only for small streams that fit in memory.\n   * Useful for testing.\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArray(): Promise<T[]> {\n    const result: T[] = [];\n    let x = await this.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await this.next();\n    }\n    return result;\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArrayForTest(): Promise<T[]> {\n    const stream = this.prefetch(100);\n    const result: T[] = [];\n    let x = await stream.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await stream.next();\n    }\n    return result;\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveFully(): Promise<void> {\n    let x = await this.next();\n    while (!x.done) {\n      x = await this.next();\n    }\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted, or a predicate fails.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveWhile(predicate: (r: T) => boolean): Promise<void> {\n    let x = await this.next();\n    let shouldContinue = predicate(x.value);\n    while ((!x.done) && shouldContinue) {\n      x = await this.next();\n      shouldContinue = predicate(x.value);\n    }\n  }\n\n  /**\n   * Handles errors thrown on this stream using a provided handler function.\n   *\n   * @param handler A function that handles any `Error` thrown during a `next()`\n   *   call and returns true if the stream should continue (dropping the failed\n   *   call) or false if the stream should quietly terminate.  If the handler\n   *   itself throws (or rethrows) an `Error`, that will be propagated.\n   *\n   * @returns A `LazyIterator` of elements passed through from upstream,\n   *   possibly filtering or terminating on upstream `next()` calls that\n   *   throw an `Error`.\n   */\n  handleErrors(handler: (error: Error) => boolean): LazyIterator<T> {\n    return new ErrorHandlingLazyIterator(this, handler);\n  }\n\n  // TODO(soergel): Implement reduce() etc.\n\n  /**\n   * Filters this stream according to `predicate`.\n   *\n   * @param predicate A function mapping a stream element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `LazyIterator` of elements for which the predicate was true.\n   */\n  filter(predicate: (value: T) => boolean): LazyIterator<T> {\n    return new FilterIterator(this, predicate);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  map<O>(transform: (value: T) => O): LazyIterator<O> {\n    return new MapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through an async 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a `Promise` for a\n   *   transformed stream element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  mapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform, forcing serial execution.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  serialMapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform).serial();\n  }\n\n  /**\n   * Maps this stream through a 1-to-many transform.\n   *\n   * @param transform A function mapping a stream element to an array of\n   *   transformed elements.\n   *\n   * @returns A `DataStream` of transformed elements.\n   */\n  flatmap<O>(transform: (value: T) => O[]): LazyIterator<O> {\n    return new FlatmapIterator(this, transform);\n  }\n\n  /**\n   * Apply a function to every element of the stream.\n   *\n   * @param f A function to apply to each stream element.\n   */\n  async forEachAsync(f: (value: T) => void): Promise<void> {\n    return this.map(f).resolveFully();\n  }\n\n  /**\n   * Apply a function to every element of the stream, forcing serial execution.\n   *\n   * @param f A function to apply to each stream element.  Should return 'true'\n   *   to indicate that the stream should continue, or 'false' to cause it to\n   *   terminate.\n   */\n  async serialForEach(f: (value: T) => Promise<boolean>): Promise<void> {\n    return this.serialMapAsync(f).resolveWhile(x => (x === true));\n  }\n\n  /**\n   * Groups elements into batches, represented as arrays of elements.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"Row-major\" means that the resulting batch is simply a collection of\n   * rows: `[row1, row2, row3, ...]`.  This is contrast to the column-major\n   * form, which is needed for vectorized computation.\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `LazyIterator` of batches of elements, represented as arrays\n   *   of the original element type.\n   */\n  rowMajorBatch(batchSize: number, smallLastBatch = true): LazyIterator<T[]> {\n    return new RowMajorBatchIterator(this, batchSize, smallLastBatch);\n  }\n\n  /**\n   * Groups elements into batches, represented in column-major form.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"column-major\" means that the resulting batch is a (potentially\n   * nested) structure representing the columns.  Each column entry, then,\n   * contains a collection of the values found in that column for a range of\n   * input elements.  This representation allows for vectorized computation, in\n   * contrast to the row-major form.\n   *\n   * The inputs should all have the same nested structure (i.e., of arrays and\n   * dicts).  The result is a single object with the same nested structure,\n   * where the leaves are arrays collecting the values of the inputs at that\n   * location (or, optionally, the result of a custom function applied to those\n   * arrays).\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @param zipFn: (optional) A function that expects an array of elements at a\n   *   single node of the object tree, and returns a `DeepMapResult`.  The\n   *   `DeepMapResult` either provides a result value for that node (i.e.,\n   *   representing the subtree), or indicates that the node should be processed\n   *   recursively.  The default zipFn recurses as far as possible and places\n   *   arrays at the leaves.\n   * @returns A `LazyIterator` of batches of elements, represented as an object\n   *   with collections at the leaves.\n   */\n  columnMajorBatch(\n      batchSize: number, smallLastBatch = true,\n      // tslint:disable-next-line:no-any\n      zipFn: (xs: any[]) => DeepMapResult = zipToList):\n      LazyIterator<tf.TensorContainer> {\n    // First collect the desired number of input elements as a row-major batch.\n    const rowBatches = this.rowMajorBatch(batchSize, smallLastBatch);\n    // Now 'rotate' or 'pivot' the data, collecting all values from each column\n    // in the batch (i.e., for each key within the elements) into an array.\n    return rowBatches.map(x => deepZip(x, zipFn));\n  }\n\n  /**\n   * Concatenate this `LazyIterator` with another.\n   *\n   * @param iterator A `LazyIterator` to be concatenated onto this one.\n   * @param baseErrorHandler An optional function that can intercept `Error`s\n   *   raised during a `next()` call on the base stream.  This function can\n   *   decide whether the error should be propagated, whether the error should\n   *   be ignored, or whether the base stream should be terminated.\n   * @returns A `LazyIterator`.\n   */\n  concatenate(\n      iterator: LazyIterator<T>,\n      baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n    return new ChainedIterator(\n        iteratorFromItems([this, iterator]), baseErrorHandler);\n  }\n\n  /**\n   * Limits this stream to return at most `count` items.\n   *\n   * @param count The maximum number of items to provide from the stream. If\n   * a negative or undefined value is given, the entire stream is returned\n   *   unaltered.\n   */\n  take(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new TakeIterator(this, count);\n  }\n\n  /**\n   * Skips the first `count` items in this stream.\n   *\n   * @param count The number of items to skip.  If a negative or undefined\n   * value is given, the entire stream is returned unaltered.\n   */\n  skip(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new SkipIterator(this, count);\n  }\n\n  /**\n   * Prefetch the first `bufferSize` items in this stream.\n   *\n   * Note this prefetches Promises, but makes no guarantees about when those\n   * Promises resolve.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   */\n  prefetch(bufferSize: number): LazyIterator<T> {\n    return new PrefetchIterator(this, bufferSize);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  /**\n   * Randomly shuffles the elements of this stream.\n   *\n   * @param bufferSize: An integer specifying the number of elements from\n   * this stream from which the new stream will sample.\n   * @param seed: (Optional.) An integer specifying the random seed that\n   * will be used to create the distribution.\n   */\n  shuffle(windowSize: number, seed?: string): LazyIterator<T> {\n    return new ShuffleIterator(this, windowSize, seed);\n  }\n\n  /**\n   * Force an iterator to execute serially: each next() call will await the\n   * prior one, so that they cannot execute concurrently.\n   */\n  serial(): LazyIterator<T> {\n    return new SerialIterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on LazyIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// Iterators that just extend LazyIterator directly\n// ============================================================================\n\nclass ArrayIterator<T> extends LazyIterator<T> {\n  private trav = 0;\n  constructor(protected items: T[]) {\n    super();\n  }\n\n  summary() {\n    return `Array of ${this.items.length} items`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.trav >= this.items.length) {\n      return {value: null, done: true};\n    }\n    const item = this.items[this.trav];\n    this.trav++;\n    return {value: deepClone(item), done: false};\n  }\n}\n\nclass FunctionCallIterator<T> extends LazyIterator<T> {\n  constructor(\n      protected nextFn: () => IteratorResult<T>| Promise<IteratorResult<T>>) {\n    super();\n  }\n\n  summary() {\n    return `Function call`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    try {\n      return this.nextFn();\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message =\n          `Error thrown while iterating through a dataset: ${e.message}`;\n      throw e;\n    }\n  }\n}\n\nclass SerialIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(protected upstream: LazyIterator<T>) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Serial`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    return this.upstream.next();\n  }\n}\n\nclass SkipIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  count = 0;\n\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Skip`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider tradeoffs of reading in parallel, eg.\n    // collecting next() promises in an Array and then waiting for\n    // Promise.all() of those. Benefit: pseudo-parallel execution.  Drawback:\n    // maybe delayed GC.\n    while (this.count++ < this.maxCount) {\n      const skipped = await this.upstream.next();\n      // short-circuit if upstream is already empty\n      if (skipped.done) {\n        return skipped;\n      }\n      tf.dispose(skipped.value as {});\n    }\n    return this.upstream.next();\n  }\n}\n\nclass TakeIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Take`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.count++ >= this.maxCount) {\n      return {value: null, done: true};\n    }\n    return this.upstream.next();\n  }\n}\n\n// Note this batch just groups items into row-wise element arrays.\n// Rotating these to a column-wise representation happens only at the dataset\n// level.\nclass RowMajorBatchIterator<T> extends LazyIterator<T[]> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T[]>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected batchSize: number,\n      protected enableSmallLastBatch = true) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> RowMajorBatch`;\n  }\n\n  async next(): Promise<IteratorResult<T[]>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T[]>> {\n    const batch: T[] = [];\n    while (batch.length < this.batchSize) {\n      const item = await this.upstream.next();\n      if (item.done) {\n        if (this.enableSmallLastBatch && batch.length > 0) {\n          return {value: batch, done: false};\n        }\n        return {value: null, done: true};\n      }\n      batch.push(item.value);\n    }\n    return {value: batch, done: false};\n  }\n}\n\nclass FilterIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected predicate: (value: T) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Filter`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      const item = await this.upstream.next();\n      if (item.done || this.predicate(item.value)) {\n        return item;\n      }\n      tf.dispose(item.value as {});\n    }\n  }\n}\n\nclass MapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Map`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\nclass ErrorHandlingLazyIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected handler: (error: Error) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> handleErrors`;\n  }\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      try {\n        return await this.upstream.next();\n      } catch (e) {\n        if (!this.handler(e)) {\n          return {value: null, done: true};\n        }\n        // If the handler returns true, loop and fetch the next upstream item.\n\n        // If the upstream iterator throws an endless stream of errors, and if\n        // the handler says to ignore them, then we loop forever here.  That is\n        // the correct behavior-- it's up to the handler to decide when to stop.\n      }\n    }\n  }\n}\n\nclass AsyncMapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => Promise<O>) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> AsyncMap`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = await this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\n// Iterators that maintain a queue of pending items\n// ============================================================================\n\n/**\n * A base class for transforming streams that operate by maintaining an\n * output queue of elements that are ready to return via next().  This is\n * commonly required when the transformation is 1-to-many:  A call to next()\n * may trigger a call to the underlying stream, which will produce many\n * mapped elements of this stream-- of which we need to return only one, so\n * we have to queue the rest.\n */\nexport abstract class OneToManyIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  protected outputQueue: RingBuffer<T>;\n\n  constructor() {\n    super();\n    this.outputQueue = new GrowingRingBuffer<T>();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  /**\n   * Read one or more chunks from upstream and process them, possibly\n   * reading or writing a carryover, and adding processed items to the\n   * output queue.  Note it's possible that no items are added to the queue\n   * on a given pump() call, even if the upstream stream is not closed\n   * (e.g., because items are filtered).\n   *\n   * @return `true` if any action was taken, i.e. fetching items from the\n   *   upstream source OR adding items to the output queue.  `false` if the\n   *   upstream source is exhausted AND nothing was added to the queue\n   * (i.e., any remaining carryover).\n   */\n  protected abstract async pump(): Promise<boolean>;\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // Fetch so that the queue contains at least one item if possible.\n    // If the upstream source is exhausted, AND there are no items left in\n    // the output queue, then this stream is also exhausted.\n    while (this.outputQueue.length() === 0) {\n      // TODO(soergel): consider parallel reads.\n      if (!await this.pump()) {\n        return {value: null, done: true};\n      }\n    }\n    return {value: this.outputQueue.shift(), done: false};\n  }\n}\nclass FlatmapIterator<I, O> extends OneToManyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O[]) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Flatmap`;\n  }\n\n  async pump(): Promise<boolean> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return false;\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // that's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying any\n    // intermediate Tensors.  Here we are concerned only about the inputs.\n    const mappedArray = this.transform(item.value);\n    const outputTensors =\n        tf.tensor_util.getTensorsInContainer(mappedArray as {});\n    this.outputQueue.pushAll(mappedArray);\n\n    // TODO(soergel) faster intersection, and deduplicate outputTensors\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n\n    return true;\n  }\n}\n\n/**\n * Provides a `LazyIterator` that concatenates a stream of underlying\n * streams.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n */\nexport class ChainedIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>> = null;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private iterator: LazyIterator<T> = null;\n  private moreIterators: LazyIterator<LazyIterator<T>>;\n\n  constructor(\n      iterators: LazyIterator<LazyIterator<T>>,\n      private readonly baseErrorHandler?: (e: Error) => boolean) {\n    super();\n    this.moreIterators = iterators;\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of chained summaries';\n    return `${upstreamSummaries} -> Chained`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    this.lastRead = this.readFromChain(this.lastRead);\n    return this.lastRead;\n  }\n\n  private async readFromChain(lastRead: Promise<IteratorResult<T>>):\n      Promise<IteratorResult<T>> {\n    // Must await on the previous read since the previous read may have advanced\n    // the stream of streams, from which we need to read.\n    // This is unfortunate since we can't parallelize reads. Which means\n    // prefetching of chained streams is a no-op.\n    // One solution is to prefetch immediately upstream of this.\n    await lastRead;\n    if (this.iterator == null) {\n      const iteratorResult = await this.moreIterators.next();\n      if (iteratorResult.done) {\n        // No more streams to stream from.\n        return {value: null, done: true};\n      }\n      this.iterator = iteratorResult.value;\n      if (this.baseErrorHandler != null) {\n        this.iterator = this.iterator.handleErrors(this.baseErrorHandler);\n      }\n    }\n    const itemResult = await this.iterator.next();\n    if (itemResult.done) {\n      this.iterator = null;\n      return this.readFromChain(lastRead);\n    }\n    return itemResult;\n  }\n}\n\nexport enum ZipMismatchMode {\n  FAIL,      // require zipped streams to have the same length\n  SHORTEST,  // terminate zip when the first stream is exhausted\n  LONGEST    // use nulls for exhausted streams; use up the longest stream.\n}\n\n/**\n * Provides a `LazyIterator` that zips together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nclass ZipIterator<O extends tf.TensorContainer> extends LazyIterator<O> {\n  private count = 0;\n  private currentPromise: Promise<IteratorResult<O>> = null;\n\n  constructor(\n      protected readonly iterators: IteratorContainer,\n      protected readonly mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL) {\n    super();\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of zip summaries';\n    return `{${upstreamSummaries}} -> Zip`;\n  }\n\n  private async nextState(afterState: Promise<IteratorResult<O>>):\n      Promise<IteratorResult<O>> {\n    // This chaining ensures that the underlying next() are not even called\n    // before the previous ones have resolved.\n    await afterState;\n\n    // Collect underlying iterator \"done\" signals as a side effect in\n    // getNext()\n    let numIterators = 0;\n    let iteratorsDone = 0;\n\n    function getNext(container: IteratorContainer): DeepMapAsyncResult {\n      if (container instanceof LazyIterator) {\n        const result = container.next();\n        return {\n          value: result.then(x => {\n            numIterators++;\n            if (x.done) {\n              iteratorsDone++;\n            }\n            return x.value;\n          }),\n          recurse: false\n        };\n      } else {\n        return {value: null, recurse: true};\n      }\n    }\n\n    const mapped: O = await deepMapAndAwaitAll(this.iterators, getNext);\n\n    if (numIterators === iteratorsDone) {\n      // The streams have all ended.\n      return {value: null, done: true};\n    }\n    if (iteratorsDone > 0) {\n      switch (this.mismatchMode) {\n        case ZipMismatchMode.FAIL:\n          throw new Error(\n              'Zipped streams should have the same length. ' +\n              `Mismatched at element ${this.count}.`);\n        case ZipMismatchMode.SHORTEST:\n          return {value: null, done: true};\n        case ZipMismatchMode.LONGEST:\n        default:\n          // Continue.  The exhausted streams already produced value: null.\n      }\n    }\n\n    this.count++;\n    return {value: mapped, done: false};\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    this.currentPromise = this.nextState(this.currentPromise);\n    return this.currentPromise;\n  }\n}\n\n// Iterators that maintain a ring buffer of pending promises\n// ============================================================================\n\n/**\n * A stream that prefetches a given number of items from an upstream source,\n * returning them in FIFO order.\n *\n * Note this prefetches Promises, but makes no guarantees about when those\n * Promises resolve.\n */\nexport class PrefetchIterator<T> extends LazyIterator<T> {\n  protected buffer: RingBuffer<Promise<IteratorResult<T>>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected bufferSize: number) {\n    super();\n    this.buffer = new RingBuffer<Promise<IteratorResult<T>>>(bufferSize);\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Prefetch`;\n  }\n\n  /**\n   * Refill the prefetch buffer.  Returns only after the buffer is full, or\n   * the upstream source is exhausted.\n   */\n  protected refill() {\n    while (!this.buffer.isFull()) {\n      const v = this.upstream.next();\n      this.buffer.push(v);\n    }\n  }\n\n  next(): Promise<IteratorResult<T>> {\n    this.refill();\n    // This shift will never throw an error because the buffer is always\n    // full after a refill. If the stream is exhausted, the buffer will be\n    // full of Promises that will resolve to the end-of-stream signal.\n    return this.buffer.shift();\n  }\n}\n\n/**\n * A stream that performs a sliding-window random shuffle on an upstream\n * source. This is like a `PrefetchIterator` except that the items are\n * returned in randomized order.  Mixing naturally improves as the buffer\n * size increases.\n */\nexport class ShuffleIterator<T> extends PrefetchIterator<T> {\n  private readonly random: seedrandom.prng;\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private upstreamExhausted = false;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected windowSize: number,\n      seed?: string) {\n    super(upstream, windowSize);\n    this.random = seedrandom.alea(seed || tf.util.now().toString());\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private randomInt(max: number) {\n    return Math.floor(this.random() * max);\n  }\n\n  protected chooseIndex(): number {\n    return this.randomInt(this.buffer.length());\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider performance\n    if (!this.upstreamExhausted) {\n      this.refill();\n    }\n    while (!this.buffer.isEmpty()) {\n      const chosenIndex = this.chooseIndex();\n      const result = await this.buffer.shuffleExcise(chosenIndex);\n      if (result.done) {\n        this.upstreamExhausted = true;\n      } else {\n        this.refill();\n        return result;\n      }\n    }\n    return {value: null, done: true};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {TensorContainer, TensorLike} from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {iteratorFromConcatenated, iteratorFromFunction, iteratorFromItems, iteratorFromZipped, LazyIterator, ZipMismatchMode} from './iterators/lazy_iterator';\nimport {Container} from './types';\nimport {canTensorify, deepMapAndAwaitAll, DeepMapResult, isIterable} from './util/deep_map';\n\n/**\n * A nested structure of Datasets, used as the input to zip().\n */\nexport type DatasetContainer = Container<Dataset<TensorContainer>>;\n\n// TODO(soergel): consider vectorized operations within the pipeline.\n\n/**\n * Represents a potentially large list of independent data elements (typically\n * 'samples' or 'examples').\n *\n * A 'data example' may be a primitive, an array, a map from string keys to\n * values, or any nested structure of these.\n *\n * A `Dataset` represents an ordered collection of elements, together with a\n * chain of transformations to be performed on those elements. Each\n * transformation is a method of `Dataset` that returns another `Dataset`, so\n * these may be chained, e.g.\n * `const processedDataset = rawDataset.filter(...).map(...).batch(...)`.\n *\n * Data loading and transformation is done in a lazy, streaming fashion.  The\n * dataset may be iterated over multiple times; each iteration starts the data\n * loading anew and recapitulates the transformations.\n *\n * A `Dataset` is typically processed as a stream of unbatched examples --i.e.,\n * its transformations are applied one example at a time. Batching produces a\n * new `Dataset` where each element is a batch. Batching should usually come\n * last in a pipeline, because data transformations are easier to express on a\n * per-example basis than on a per-batch basis.\n *\n * The following code examples are calling `await dataset.forEachAsync(...)` to\n * iterate once over the entire dataset in order to print out the data.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport abstract class Dataset<T extends tf.TensorContainer> {\n  /*\n   * Provide a new stream of elements.  Note this will also start new streams\n   * from any underlying `Dataset`s.\n   *\n   * CAUTION: Any Tensors contained within the elements returned from\n   * this stream *must* be manually disposed to avoid a GPU memory leak.\n   * The tf.tidy() approach cannot be used in an asynchronous context.\n   */\n  abstract async iterator(): Promise<LazyIterator<T>>;\n\n  readonly size: number = null;\n\n  // TODO(soergel): Make Datasets report whether repeated iterator() calls\n  // produce the same result (e.g., reading from a file) or different results\n  // (e.g., from the webcam).  Currently we don't make this distinction but it\n  // could be important for the user to know.\n  // abstract isDeterministic(): boolean;\n\n  /**\n   * Groups elements into batches.\n   *\n   * It is assumed that each of the incoming dataset elements has the same\n   * structure-- i.e. the same set of keys at each location in an object\n   * hierarchy.  For each key, the resulting `Dataset` provides a batched\n   * element collecting all of the incoming values for that key.\n   *\n   *  * Incoming primitives are grouped into a 1-D Tensor.\n   *  * Incoming Tensors are grouped into a new Tensor where the 0'th axis is\n   *    the batch dimension.\n   *  * Incoming arrays are converted to Tensor and then batched.\n   *  * A nested array is interpreted as an n-D Tensor, so the batched result\n   *    has n+1 dimensions.\n   *  * An array that cannot be converted to Tensor produces an error.\n   *\n   * If an array should not be batched as a unit, it should first be converted\n   * to an object with integer keys.\n   *\n   * Here are a few examples:\n   *\n   * Batch a dataset of numbers:\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8]).batch(4);\n   * await a.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of arrays:\n   * ```js\n   * const b = tf.data.array([[1], [2], [3], [4], [5], [6], [7], [8]]).batch(4);\n   * await b.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of objects:\n   * ```js\n   * const c = tf.data.array([{a: 1, b: 11}, {a: 2, b: 12}, {a: 3, b: 13},\n   *   {a: 4, b: 14}, {a: 5, b: 15}, {a: 6, b: 16}, {a: 7, b: 17},\n   *   {a: 8, b: 18}]).batch(4);\n   * await c.forEachAsync(e => {\n   *   console.log('{');\n   *   for(var key in e) {\n   *     console.log(key+':');\n   *     e[key].print();\n   *   }\n   *   console.log('}');\n   * })\n   * ```\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `Dataset`, from which a stream of batches can be obtained.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  batch(batchSize: number, smallLastBatch = true): Dataset<tf.TensorContainer> {\n    const base = this;\n    tf.util.assert(\n        batchSize > 0, () => `batchSize needs to be positive, but it is\n      ${batchSize}`);\n    let size;\n    if (this.size === Infinity || this.size == null) {\n      // If the size of this dataset is infinity or null, the new size keeps the\n      // same.\n      size = this.size;\n    } else if (smallLastBatch) {\n      // If the size of this dataset is known and include small last batch, the\n      // new size is full batch count plus last batch.\n      size = Math.ceil(this.size / batchSize);\n    } else {\n      // If the size of this dataset is known and not include small last batch,\n      // the new size is full batch count.\n      size = Math.floor(this.size / batchSize);\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator())\n          .columnMajorBatch(batchSize, smallLastBatch, deepBatchConcat);\n    }, size);\n  }\n\n  /**\n   * Concatenates this `Dataset` with another.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * const b = tf.data.array([4, 5, 6]);\n   * const c = a.concatenate(b);\n   * await c.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param dataset A `Dataset` to be concatenated onto this one.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  concatenate(dataset: Dataset<T>): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity || dataset.size === Infinity) {\n      // If the size of any of these two dataset is infinity, new size is\n      // infinity.\n      size = Infinity;\n    } else if (this.size != null && dataset.size != null) {\n      // If the size of both datasets are known and not infinity, new size is\n      // sum the size of these two datasets.\n      size = this.size + dataset.size;\n    } else {\n      // If neither of these two datasets has infinite size and any of these two\n      // datasets' size is null, the new size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () =>\n            (await base.iterator()).concatenate(await dataset.iterator()),\n        size);\n  }\n\n  /**\n   * Filters this dataset according to `predicate`.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])\n   *   .filter(x => x%2 === 0);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param predicate A function mapping a dataset element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `Dataset` of elements for which the predicate was true.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  filter(predicate: (value: T) => boolean): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity) {\n      // If the size of this dataset is infinity, new size is infinity\n      size = Infinity;\n    } else {\n      // If this dataset has limited elements, new size is null because it might\n      // exhausted randomly.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).filter(x => tf.tidy(() => predicate(x)));\n    }, size);\n  }\n\n  /**\n   * Apply a function to every element of the dataset.\n   *\n   * After the function is applied to a dataset element, any Tensors contained\n   * within that element are disposed.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param f A function to apply to each dataset element.\n   * @returns A `Promise` that resolves after all elements have been processed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async forEachAsync(f: (input: T) => void): Promise<void> {\n    return (await this.iterator()).forEachAsync(f);\n  }\n\n  /**\n   * Maps this dataset through a 1-to-1 transform.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).map(x => x*x);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a transformed\n   *   dataset element.\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  map<O extends tf.TensorContainer>(transform: (value: T) => O): Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).map(x => tf.tidy(() => transform(x)));\n    }, this.size);\n  }\n\n  /**\n   * Maps this dataset through an async 1-to-1 transform.\n   *\n   * ```js\n   * const a =\n   *  tf.data.array([1, 2, 3]).mapAsync(x => new Promise(function(resolve){\n   *    setTimeout(() => {\n   *      resolve(x * x);\n   *    }, Math.random()*1000 + 500);\n   *  }));\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a `Promise` for a\n   *   transformed dataset element.  This transform is responsible for disposing\n   *   any intermediate `Tensor`s, i.e. by wrapping its computation in\n   *   `tf.tidy()`; that cannot be automated here (as it is in the synchronous\n   *   `map()` case).\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  mapAsync<O extends tf.TensorContainer>(transform: (value: T) => Promise<O>):\n      Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).mapAsync(transform);\n    }, this.size);\n  }\n\n  /**\n   *  Creates a `Dataset` that prefetches elements from this dataset.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  prefetch(bufferSize: number): Dataset<T> {\n    if (bufferSize == null) {\n      throw new RangeError(\n          '`Dataset.prefetch()` requires bufferSize to be specified.');\n    }\n\n    const base = this;\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).prefetch(bufferSize), this.size);\n  }\n\n  /**\n   * Repeats this dataset `count` times.\n   *\n   * NOTE: If this dataset is a function of global state (e.g. a random number\n   * generator), then different repetitions may produce different elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).repeat(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: (Optional) An integer, representing the number of times\n   *   the dataset should be repeated. The default behavior (if `count` is\n   *   `undefined` or negative) is for the dataset be repeated indefinitely.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  repeat(count?: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count > 0) {\n      // If this dataset has size and count is positive, new size is current\n      // size multiply count. This also covers the case that current size is\n      // infinity.\n      size = this.size * count;\n    } else if (count === 0) {\n      // If count is 0, new size is 0.\n      size = 0;\n    } else if (this.size != null && (count === undefined || count < 0)) {\n      // If this dataset has size and count is undefined or negative, the\n      // dataset will be repeated indefinitely and new size is infinity.\n      size = Infinity;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      const iteratorIterator = iteratorFromFunction(\n          async () => ({value: await base.iterator(), done: false}));\n      return iteratorFromConcatenated(iteratorIterator.take(count));\n    }, size);\n  }\n\n  /**\n   * Creates a `Dataset` that skips `count` initial elements from this dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).skip(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be skipped\n   *   to form the new dataset.  If `count` is greater than the size of this\n   *   dataset, the new dataset will contain no elements.  If `count`\n   *   is `undefined` or negative, skips the entire dataset.\n   *\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  skip(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count >= 0 && this.size >= count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is current size minus skipped size.This also covers the case that\n      // current size is infinity.\n      size = this.size - count;\n    } else if (\n        this.size != null &&\n        (this.size < count || count === undefined || count < 0)) {\n      // If the size of this dataset is smaller than count, or count is\n      // undefined or negative, skips the entire dataset and the new size is 0.\n      size = 0;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).skip(count), size);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  static readonly MAX_BUFFER_SIZE = 10000;\n\n  /**\n   * Pseudorandomly shuffles the elements of this dataset. This is done in a\n   * streaming manner, by sampling from a given number of prefetched elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).shuffle(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param bufferSize: An integer specifying the number of elements from this\n   *   dataset from which the new dataset will sample.\n   * @param seed: (Optional) An integer specifying the random seed that will\n   *   be used to create the distribution.\n   * @param reshuffleEachIteration: (Optional) A boolean, which if true\n   *   indicates that the dataset should be pseudorandomly reshuffled each time\n   *   it is iterated over. If false, elements will be returned in the same\n   *   shuffled order on each iteration. (Defaults to `true`.)\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  shuffle(bufferSize: number, seed?: string, reshuffleEachIteration = true):\n      Dataset<T> {\n    if (bufferSize == null || bufferSize < 0) {\n      if (this.size == null) {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.');\n      } else {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.  ' +\n            'If your data fits in main memory (for regular JS objects), ' +\n            'and/or GPU memory (for `tf.Tensor`s), consider setting ' +\n            `bufferSize to the dataset size (${this.size} elements)`);\n      }\n    }\n    const base = this;\n    const random = seedrandom.alea(seed || tf.util.now().toString());\n    return datasetFromIteratorFn(async () => {\n      let seed2 = random.int32();\n      if (reshuffleEachIteration) {\n        seed2 += random.int32();\n      }\n      return (await base.iterator()).shuffle(bufferSize, seed2.toString());\n    }, this.size);\n  }\n\n  /**\n   * Creates a `Dataset` with at most `count` initial elements from this\n   * dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).take(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be taken\n   *   to form the new dataset.  If `count` is `undefined` or negative, or if\n   *   `count` is greater than the size of this dataset, the new dataset will\n   *   contain all elements of this dataset.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  take(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && this.size > count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is count.\n      size = count;\n    } else if (this.size != null && this.size <= count) {\n      // If the size of this dataset is equal or smaller than count, the new\n      // dataset's size is the size of this dataset.\n      size = this.size;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).take(count), size);\n  }\n\n  /**\n   * Collect all elements of this dataset into an array.\n   *\n   * Obviously this will succeed only for small datasets that fit in memory.\n   * Useful for testing and generally should be avoided if possible.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]);\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async toArray() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArray();\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   */\n  async toArrayForTest() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArrayForTest();\n  }\n}\n\n/**\n * Create a `Dataset` defined by a provided iterator() function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * const ds = tf.data.datasetFromIteratorFn(iter);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n */\nexport function datasetFromIteratorFn<T extends tf.TensorContainer>(\n    iteratorFn: () => Promise<LazyIterator<T>>,\n    size: number = null): Dataset<T> {\n  return new class extends Dataset<T> {\n    size = size;\n\n    /*\n     * Provide a new stream of elements.  Note this will also start new streams\n     * from any underlying `Dataset`s.\n     */\n    async iterator(): Promise<LazyIterator<T>> {\n      return iteratorFn();\n    }\n  }\n  ();\n}\n\n/**\n * Create a `Dataset` from an array of elements.\n *\n * Create a Dataset from an array of objects:\n * ```js\n * const a = tf.data.array([{'item': 1}, {'item': 2}, {'item': 3}]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n *\n * Create a Dataset from an array of numbers:\n * ```js\n * const a = tf.data.array([4, 5, 6]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n * @param items An array of elements that will be parsed as items in a dataset.\n *\n * @doc {heading: 'Data', subheading: 'Creation', namespace: 'data'}\n */\nexport function array<T extends tf.TensorContainer>(items: T[]): Dataset<T> {\n  return datasetFromIteratorFn(\n      async () => iteratorFromItems(items), items.length);\n}\n\n/**\n * Create a `Dataset` by zipping together an array, dict, or nested\n * structure of `Dataset`s (and perhaps additional constants).\n * The underlying datasets must provide elements in a consistent order such that\n * they correspond.\n *\n * The number of elements in the resulting dataset is the same as the size of\n * the smallest dataset in datasets.\n *\n * The nested structure of the `datasets` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Note this means that, given an array of two datasets that produce dict\n * elements, the result is a dataset that produces elements that are arrays\n * of two dicts:\n *\n * Zip an array of datasets:\n * ```js\n * console.log('Zip two datasets of objects:');\n * const ds1 = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const ds2 = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const ds3 = tf.data.zip([ds1, ds2]);\n * await ds3.forEachAsync(e => console.log(JSON.stringify(e)));\n *\n * // If the goal is to merge the dicts in order to produce elements like\n * // {a: ..., b: ...}, this requires a second step such as:\n * console.log('Merge the objects:');\n * const ds4 = ds3.map(x => {return {a: x[0].a, b: x[1].b}});\n * await ds4.forEachAsync(e => console.log(e));\n * ```\n *\n * Zip a dict of datasets:\n * ```js\n * const a = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const b = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const c = tf.data.zip({c: a, d: b});\n * await c.forEachAsync(e => console.log(JSON.stringify(e)));\n * ```\n *\n * @doc {heading: 'Data', subheading: 'Operations', namespace: 'data'}\n */\nexport function zip<O extends tf.TensorContainer>(datasets: DatasetContainer):\n    Dataset<O> {\n  // manually type-check the argument for JS users\n  if (!isIterable(datasets)) {\n    throw new Error('The argument to zip() must be an object or array.');\n  }\n  let size;\n  if (Array.isArray(datasets)) {\n    for (let i = 0; i < datasets.length; i++) {\n      size = size == null ? (datasets[i] as Dataset<O>).size :\n                            Math.min(size, (datasets[i] as Dataset<O>).size);\n    }\n  } else if (datasets instanceof Object) {\n    for (const ds in datasets) {\n      size = size == null ? (datasets[ds] as Dataset<O>).size :\n                            Math.min(size, (datasets[ds] as Dataset<O>).size);\n    }\n  }\n  return datasetFromIteratorFn<O>(async () => {\n    const streams = await deepMapAndAwaitAll(datasets, d => {\n      if (d instanceof Dataset) {\n        return {value: d.iterator(), recurse: false};\n      } else if (isIterable(d)) {\n        return {value: null, recurse: true};\n      } else {\n        throw new Error(\n            'Leaves of the structure passed to zip() must be Datasets, ' +\n            'not primitives.');\n      }\n    });\n    return iteratorFromZipped<O>(streams, ZipMismatchMode.SHORTEST);\n  }, size);\n}\n\n/**\n * A zip function for use with deepZip, passed via the columnMajorBatch call.\n *\n * Accepts an array of identically-structured nested elements and either batches\n * them (if they are primitives, numeric arrays, or Tensors) or requests\n * recursion (if not).\n */\n// tslint:disable-next-line:no-any\nfunction deepBatchConcat(rows: any[]): DeepMapResult {\n  if (rows === null) {\n    return null;\n  }\n\n  // use the first item to decide whether to recurse or batch here.\n  const exampleRow = rows[0];\n\n  if (canTensorify(exampleRow)) {\n    // rows is an array of primitives, Tensors, or arrays.  Batch them.\n    const value = batchConcat(rows);\n    return {value, recurse: false};\n  }\n\n  // the example row is an object, so recurse into it.\n  return {value: null, recurse: true};\n}\n\n/**\n * Assembles a list of same-shaped numbers, number arrays, or Tensors\n * into a single new Tensor where axis 0 is the batch dimension.\n */\nfunction batchConcat<T extends(TensorLike | tf.Tensor)>(arrays: T[]):\n    tf.Tensor {\n  if (arrays.length === 0) {\n    // We can't return an empty Tensor because we don't know the element shape.\n    throw new Error('Can\\'t make a batch of zero elements.');\n  }\n\n  if (arrays[0] instanceof tf.Tensor) {\n    // Input is an array of Tensors\n    return tf.stack(arrays as tf.Tensor[]);\n  } else {\n    // Input is a possibly-nested array of numbers.\n    return tf.tensor(arrays as TensorLike);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\n\n/**\n * Represents a potentially large collection of text lines.\n *\n * The results are not batched.\n */\nexport class TextLineDataset extends Dataset<string> {\n  /**\n   * Create a `TextLineDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   */\n  constructor(protected readonly input: DataSource) {\n    super();\n  }\n\n  async iterator(): Promise<LazyIterator<string>> {\n    const inputIterator = await this.input.iterator();\n    const utf8Iterator = inputIterator.decodeUTF8();\n    const lineIterator = utf8Iterator.split('\\n').map(line => {\n      // Windows/DOS format text file has extra line breaker at the end of line.\n      if (line.endsWith('\\r')) {\n        line = line.slice(0, -1);\n      }\n      return line;\n    });\n    return lineIterator;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\nimport {ColumnConfig, CSVConfig} from '../types';\nimport {TextLineDataset} from './text_line_dataset';\n\nconst CODE_QUOTE = '\"';\nconst STATE_OUT = Symbol('out');\nconst STATE_FIELD = Symbol('field');\nconst STATE_QUOTE = Symbol('quote');\nconst STATE_QUOTE_AFTER_QUOTE = Symbol('quoteafterquote');\nconst STATE_WITHIN_QUOTE_IN_QUOTE = Symbol('quoteinquote');\n\n/**\n * Represents a potentially large collection of delimited text records.\n *\n * The produced `TensorContainer`s each contain one key-value pair for\n * every column of the table.  When a field is empty in the incoming data, the\n * resulting value is `undefined`, or throw error if it is required.  Values\n * that can be parsed as numbers are emitted as type `number`, other values\n * are parsed as `string`.\n *\n * The results are not batched.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport class CSVDataset extends Dataset<TensorContainer> {\n  base: TextLineDataset;\n  private hasHeader = true;\n  private fullColumnNames: string[] = null;\n  private columnNamesValidated = false;\n  private columnConfigs: {[key: string]: ColumnConfig} = null;\n  private configuredColumnsOnly = false;\n  private delimiter = ',';\n  private delimWhitespace = false;\n\n  /**\n   * Returns column names of the csv dataset. If `configuredColumnsOnly` is\n   * true, return column names in `columnConfigs`. If `configuredColumnsOnly` is\n   * false and `columnNames` is provided, `columnNames`. If\n   * `configuredColumnsOnly` is false and `columnNames` is not provided, return\n   * all column names parsed from the csv file. For example usage please go to\n   * `tf.data.csv`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async columnNames() {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    return this.configuredColumnsOnly ? Object.keys(this.columnConfigs) :\n                                        this.fullColumnNames;\n  }\n\n  /* 1) If `columnNames` is provided as string[], use this string[] as output\n   * keys in corresponding order. The length must match the number of inferred\n   * columns if `hasHeader` is true .\n   * 2) If `columnNames` is not provided, parse header line as `columnNames` if\n   * hasHeader is true. If `hasHeader` is false, throw an error.\n   * 3) If `columnConfigs` is provided, all the keys in `columnConfigs` must\n   * exist in parsed `columnNames`.\n   */\n  private async setColumnNames() {\n    const columnNamesFromFile = await this.maybeReadHeaderLine();\n    if (!this.fullColumnNames && !columnNamesFromFile) {\n      // Throw an error if columnNames is not provided and no header line.\n      throw new Error(\n          'Column names must be provided if there is no header line.');\n    } else if (this.fullColumnNames && columnNamesFromFile) {\n      // Check provided columnNames match header line.\n      util.assert(\n          columnNamesFromFile.length === this.fullColumnNames.length,\n          () => 'The length of provided columnNames (' +\n              this.fullColumnNames.length.toString() +\n              ') does not match the length of the header line read from ' +\n              'file (' + columnNamesFromFile.length.toString() + ').');\n    }\n    if (!this.fullColumnNames) {\n      this.fullColumnNames = columnNamesFromFile;\n    }\n    // Check if there are duplicate column names.\n    const counts: {[key: string]: number} = this.fullColumnNames.reduce(\n        (countAcc: {[key: string]: number}, name) => {\n          countAcc[name] = (countAcc[name] + 1) || 1;\n          return countAcc;\n        },\n        {});\n    const duplicateNames =\n        Object.keys(counts).filter((name) => (counts[name] > 1));\n    util.assert(\n        duplicateNames.length === 0,\n        () => 'Duplicate column names found: ' + duplicateNames.toString());\n    // Check if keys in columnConfigs match columnNames.\n    if (this.columnConfigs) {\n      for (const key of Object.keys(this.columnConfigs)) {\n        const index = this.fullColumnNames.indexOf(key);\n        if (index === -1) {\n          throw new Error(\n              'The key \"' + key +\n              '\" provided in columnConfigs does not match any of the column ' +\n              'names (' + this.fullColumnNames.toString() + ').');\n        }\n      }\n    }\n    this.columnNamesValidated = true;\n  }\n\n  private async maybeReadHeaderLine() {\n    if (this.hasHeader) {\n      const iter = await this.base.iterator();\n      const firstElement = await iter.next();\n      if (firstElement.done) {\n        throw new Error('No data was found for CSV parsing.');\n      }\n      const firstLine: string = firstElement.value;\n      const headers = this.parseRow(firstLine, false);\n      return headers;\n    } else {\n      return null;\n    }\n  }\n\n  /**\n   * Create a `CSVDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   * @param csvConfig (Optional) A CSVConfig object that contains configurations\n   *     of reading and decoding from CSV file(s).\n   *\n   *     hasHeader: (Optional) A boolean value that indicates whether the first\n   *     row of provided CSV file is a header line with column names, and should\n   *     not be included in the data. Defaults to `true`.\n   *\n   *     columnNames: (Optional) A list of strings that corresponds to\n   *     the CSV column names, in order. If provided, it ignores the column\n   *     names inferred from the header row. If not provided, infers the column\n   *     names from the first row of the records. If hasHeader is false and\n   *     columnNames is not provided, this method throws an error.\n   *\n   *     columnConfigs: (Optional) A dictionary whose key is column names, value\n   *     is an object stating if this column is required, column's data type,\n   *     default value, and if this column is label. If provided, keys must\n   *     correspond to names provided in columnNames or inferred from the file\n   *     header lines. If isLabel is true any column, returns an array of two\n   *     items: the first item is a dict of features key/value pairs, the second\n   *     item is a dict of labels key/value pairs. If no feature is marked as\n   *     label, returns a dict of features only.\n   *\n   *     configuredColumnsOnly (Optional) If true, only columns provided in\n   *     columnConfigs will be parsed and provided during iteration.\n   *\n   *     delimiter (Optional) The string used to parse each line of the input\n   *     file. Defaults to `,`.\n   */\n  constructor(protected readonly input: DataSource, csvConfig?: CSVConfig) {\n    super();\n    this.base = new TextLineDataset(input);\n    if (!csvConfig) {\n      csvConfig = {};\n    }\n    this.hasHeader = csvConfig.hasHeader === false ? false : true;\n    this.fullColumnNames = csvConfig.columnNames;\n    this.columnConfigs = csvConfig.columnConfigs;\n    this.configuredColumnsOnly = csvConfig.configuredColumnsOnly;\n    if (csvConfig.delimWhitespace) {\n      util.assert(\n          csvConfig.delimiter == null,\n          () =>\n              'Delimiter should not be provided when delimWhitespace is true.');\n      this.delimWhitespace = true;\n      this.delimiter = ' ';\n    } else {\n      this.delimiter = csvConfig.delimiter ? csvConfig.delimiter : ',';\n    }\n  }\n\n  async iterator(): Promise<LazyIterator<TensorContainer>> {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    let lines = await this.base.iterator();\n    if (this.hasHeader) {\n      // We previously read the first line to get the columnNames.\n      // Now that we're providing data, skip it.\n      lines = lines.skip(1);\n    }\n    return lines.map(x => this.makeDataElement(x));\n  }\n\n  makeDataElement(line: string): TensorContainer {\n    const values = this.parseRow(line);\n    const features: {[key: string]: TensorContainer} = {};\n    const labels: {[key: string]: TensorContainer} = {};\n\n    for (let i = 0; i < this.fullColumnNames.length; i++) {\n      const key = this.fullColumnNames[i];\n      const config = this.columnConfigs ? this.columnConfigs[key] : null;\n      if (this.configuredColumnsOnly && !config) {\n        // This column is not selected.\n        continue;\n      } else {\n        const value = values[i];\n        let parsedValue = null;\n        if (value === '') {\n          // If default value is provided, use it. If default value is not\n          // provided, set as undefined.\n          if (config && config.default !== undefined) {\n            parsedValue = config.default;\n          } else if (config && (config.required || config.isLabel)) {\n            throw new Error(\n                `Required column ${key} is empty in this line: ${line}`);\n          } else {\n            parsedValue = undefined;\n          }\n        } else {\n          // A value is present, so parse it based on type\n          const valueAsNum = Number(value);\n          if (isNaN(valueAsNum)) {\n            // The value is a string and this column is declared as boolean\n            // in config, parse it as boolean.\n            if (config && config.dtype === 'bool') {\n              parsedValue = this.getBoolean(value);\n            } else {\n              // Set value as string\n              parsedValue = value;\n            }\n          } else if (!config || !config.dtype) {\n            // If this value is a number and no type config is provided, return\n            // it as number.\n            parsedValue = valueAsNum;\n          } else {\n            // If this value is a number and data type is provided, parse it\n            // according to provided data type.\n            switch (config.dtype) {\n              case 'float32':\n                parsedValue = valueAsNum;\n                break;\n              case 'int32':\n                parsedValue = Math.floor(valueAsNum);\n                break;\n              case 'bool':\n                parsedValue = this.getBoolean(value);\n                break;\n              default:\n                parsedValue = valueAsNum;\n            }\n          }\n        }\n        // Check if this column is label.\n        (config && config.isLabel) ? labels[key] = parsedValue :\n                                     features[key] = parsedValue;\n      }\n    }\n    // If label exists, return an object of features and labels as {xs:features,\n    // ys:labels}, otherwise return features only.\n    if (Object.keys(labels).length === 0) {\n      return features;\n\n    } else {\n      return {xs: features, ys: labels};\n    }\n  }\n\n  private getBoolean(value: string): number {\n    if (value === '1' || value.toLowerCase() === 'true') {\n      return 1;\n    } else {\n      return 0;\n    }\n  }\n\n  // adapted from https://beta.observablehq.com/@mbostock/streaming-csv\n  private parseRow(line: string, validateElementCount = true): string[] {\n    const result: string[] = [];\n    let readOffset = 0;\n    const readLength = line.length;\n    let currentState = STATE_OUT;\n    // Goes through the line to parse quote.\n    for (let i = 0; i < readLength; i++) {\n      switch (currentState) {\n        // Before enter a new field\n        case STATE_OUT:\n          switch (line.charAt(i)) {\n            // Enter a quoted field\n            case CODE_QUOTE:\n              readOffset = i + 1;\n              currentState = STATE_QUOTE;\n              break;\n            // Read an empty field\n            case this.delimiter:\n              readOffset = i + 1;\n              // If delimiter is white space and configured to collapse\n              // multiple white spaces, ignore this white space.\n              if (this.delimiter === ' ' && this.delimWhitespace) {\n                break;\n              }\n              result.push('');\n              currentState = STATE_OUT;\n              break;\n            // Enter an unquoted field\n            default:\n              currentState = STATE_FIELD;\n              readOffset = i;\n              break;\n          }\n          break;\n        // In an unquoted field\n        case STATE_FIELD:\n          switch (line.charAt(i)) {\n            // Exit an unquoted field, add it to result\n            case this.delimiter:\n              result.push(line.substring(readOffset, i));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            default:\n          }\n          break;\n        // In a quoted field\n        case STATE_QUOTE:\n          switch (line.charAt(i)) {\n            // Read a quote after a quote\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE_AFTER_QUOTE;\n              break;\n            default:\n          }\n          break;\n        // This state means it's right after a second quote in a field\n        case STATE_QUOTE_AFTER_QUOTE:\n          switch (line.charAt(i)) {\n            // Finished a quoted field\n            case this.delimiter:\n              result.push(line.substring(readOffset, i - 1));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            // Finished a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            // In a quoted part in a quoted field\n            default:\n              currentState = STATE_WITHIN_QUOTE_IN_QUOTE;\n              break;\n          }\n          break;\n        case STATE_WITHIN_QUOTE_IN_QUOTE:\n          switch (line.charAt(i)) {\n            // Exit a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            default:\n          }\n          break;\n        default:\n      }\n    }\n    // Adds last item based on if it is quoted.\n    if (currentState === STATE_QUOTE_AFTER_QUOTE) {\n      result.push(line.substring(readOffset, readLength - 1));\n    } else {\n      result.push(line.substring(readOffset));\n    }\n    // Check if each row has the same number of elements as column names.\n    if (validateElementCount && result.length !== this.fullColumnNames.length) {\n      throw new Error(`Invalid row in csv file. Should have ${\n          this.fullColumnNames.length} elements in a row, but got ${result}`);\n    }\n    return result;\n  }\n}\n\n// TODO(soergel): add more basic datasets for parity with tf.data\n// tf.data.FixedLengthRecordDataset()\n// tf.data.TFRecordDataset()\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env, Tensor, tensor, Tensor2D, Tensor3D, TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {MicrophoneConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of tensors from microphone audio stream. The tensors are\n * representing audio data as frequency-domain spectrogram generated with\n * browser's native FFT. Tensors representing time-domain waveform is available\n * based on configuration. Only works in browser environment.\n */\nexport class MicrophoneIterator extends LazyIterator<TensorContainer> {\n  private isClosed = false;\n  private stream: MediaStream;\n  private readonly fftSize: number;\n  private readonly columnTruncateLength: number;\n  private freqData: Float32Array;\n  private timeData: Float32Array;\n  private readonly numFrames: number;\n  private analyser: AnalyserNode;\n  private audioContext: AudioContext;\n  private sampleRateHz: number;\n  private readonly audioTrackConstraints: MediaTrackConstraints;\n  private readonly smoothingTimeConstant: number;\n  private readonly includeSpectrogram: boolean;\n  private readonly includeWaveform: boolean;\n\n  private constructor(protected readonly microphoneConfig: MicrophoneConfig) {\n    super();\n    this.fftSize = microphoneConfig.fftSize || 1024;\n    const fftSizeLog2 = Math.log2(this.fftSize);\n    if (this.fftSize < 0 || fftSizeLog2 < 4 || fftSizeLog2 > 14 ||\n        !Number.isInteger(fftSizeLog2)) {\n      throw new Error(\n          `Invalid fftSize: it must be a power of 2 between ` +\n          `2 to 4 and 2 to 14, but got ${this.fftSize}`);\n    }\n\n    this.numFrames = microphoneConfig.numFramesPerSpectrogram || 43;\n    this.sampleRateHz = microphoneConfig.sampleRateHz;\n    this.columnTruncateLength =\n        microphoneConfig.columnTruncateLength || this.fftSize;\n    this.audioTrackConstraints = microphoneConfig.audioTrackConstraints;\n    this.smoothingTimeConstant = microphoneConfig.smoothingTimeConstant || 0;\n\n    this.includeSpectrogram =\n        microphoneConfig.includeSpectrogram === false ? false : true;\n    this.includeWaveform =\n        microphoneConfig.includeWaveform === true ? true : false;\n    if (!this.includeSpectrogram && !this.includeWaveform) {\n      throw new Error(\n          'Both includeSpectrogram and includeWaveform are false. ' +\n          'At least one type of data should be returned.');\n    }\n  }\n\n  summary() {\n    return `microphone`;\n  }\n\n  // Construct a MicrophoneIterator and start the audio stream.\n  static async create(microphoneConfig: MicrophoneConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'microphone API is only supported in browser environment.');\n    }\n\n    const microphoneIterator = new MicrophoneIterator(microphoneConfig);\n\n    // Call async function start() to initialize the audio stream.\n    await microphoneIterator.start();\n\n    return microphoneIterator;\n  }\n\n  // Start the audio stream and FFT.\n  async start(): Promise<void> {\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        audio: this.audioTrackConstraints == null ? true :\n                                                    this.audioTrackConstraints,\n        video: false\n      });\n    } catch (e) {\n      throw new Error(\n          `Error thrown while initializing video stream: ${e.message}`);\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain audio from microphone.');\n    }\n\n    const ctxConstructor =\n        // tslint:disable-next-line:no-any\n        (window as any).AudioContext || (window as any).webkitAudioContext;\n    this.audioContext = new ctxConstructor();\n\n    if (!this.sampleRateHz) {\n      // If sample rate is not provided, use the available sample rate on\n      // device.\n      this.sampleRateHz = this.audioContext.sampleRate;\n    } else if (this.audioContext.sampleRate !== this.sampleRateHz) {\n      throw new Error(\n          `Mismatch in sampling rate: ` +\n          `Expected: ${this.sampleRateHz}; ` +\n          `Actual: ${this.audioContext.sampleRate}`);\n    }\n\n    const streamSource = this.audioContext.createMediaStreamSource(this.stream);\n    this.analyser = this.audioContext.createAnalyser();\n    this.analyser.fftSize = this.fftSize * 2;\n    this.analyser.smoothingTimeConstant = this.smoothingTimeConstant;\n    streamSource.connect(this.analyser);\n    this.freqData = new Float32Array(this.fftSize);\n    this.timeData = new Float32Array(this.fftSize);\n    return;\n  }\n\n  async next(): Promise<IteratorResult<TensorContainer>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let spectrogramTensor: Tensor;\n    let waveformTensor: Tensor;\n\n    const audioDataQueue = await this.getAudioData();\n    if (this.includeSpectrogram) {\n      const freqData = this.flattenQueue(audioDataQueue.freqDataQueue);\n      spectrogramTensor = this.getTensorFromAudioDataArray(\n          freqData, [this.numFrames, this.columnTruncateLength, 1]);\n    }\n    if (this.includeWaveform) {\n      const timeData = this.flattenQueue(audioDataQueue.timeDataQueue);\n      waveformTensor = this.getTensorFromAudioDataArray(\n          timeData, [this.numFrames * this.fftSize, 1]);\n    }\n\n    return {\n      value: {'spectrogram': spectrogramTensor, 'waveform': waveformTensor},\n      done: false\n    };\n  }\n\n  // Capture one result from the audio stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<{spectrogram: Tensor3D, waveform: Tensor2D}> {\n    return (await this.next()).value as\n        {spectrogram: Tensor3D, waveform: Tensor2D};\n  }\n\n  private async getAudioData():\n      Promise<{freqDataQueue: Float32Array[], timeDataQueue: Float32Array[]}> {\n    const freqDataQueue: Float32Array[] = [];\n    const timeDataQueue: Float32Array[] = [];\n    let currentFrames = 0;\n    return new Promise(resolve => {\n      const intervalID = setInterval(() => {\n        if (this.includeSpectrogram) {\n          this.analyser.getFloatFrequencyData(this.freqData);\n          // If the audio stream is initializing, return empty queue.\n          if (this.freqData[0] === -Infinity) {\n            resolve({freqDataQueue, timeDataQueue});\n          }\n          freqDataQueue.push(this.freqData.slice(0, this.columnTruncateLength));\n        }\n        if (this.includeWaveform) {\n          this.analyser.getFloatTimeDomainData(this.timeData);\n          timeDataQueue.push(this.timeData.slice());\n        }\n\n        // Clean interval and return when all frames have been collected\n        if (++currentFrames === this.numFrames) {\n          clearInterval(intervalID);\n          resolve({freqDataQueue, timeDataQueue});\n        }\n      }, this.fftSize / this.sampleRateHz * 1e3);\n    });\n  }\n\n  // Stop the audio stream and pause the iterator.\n  stop(): void {\n    if (!this.isClosed) {\n      this.isClosed = true;\n      this.analyser.disconnect();\n      this.audioContext.close();\n      if (this.stream != null && this.stream.getTracks().length > 0) {\n        this.stream.getTracks()[0].stop();\n      }\n    }\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor[]> {\n    throw new Error('Can not convert infinite audio stream to array.');\n  }\n\n  // Return audio sampling rate in Hz\n  getSampleRate(): number {\n    return this.sampleRateHz;\n  }\n\n  private flattenQueue(queue: Float32Array[]): Float32Array {\n    const frameSize = queue[0].length;\n    const freqData = new Float32Array(queue.length * frameSize);\n    queue.forEach((data, i) => freqData.set(data, i * frameSize));\n    return freqData;\n  }\n\n  private getTensorFromAudioDataArray(freqData: Float32Array, shape: number[]):\n      Tensor {\n    const vals = new Float32Array(util.sizeFromShape(shape));\n    // If the data is less than the output shape, the rest is padded with zeros.\n    vals.set(freqData, vals.length - freqData.length);\n    return tensor(vals, shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {browser, env, image, tensor1d, Tensor1D, tensor2d, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\nimport {WebcamConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of image tensors from webcam video stream. Only works in\n * browser environment.\n */\nexport class WebcamIterator extends LazyIterator<Tensor3D> {\n  private isClosed = true;\n  private stream: MediaStream;\n  private resize = false;\n  private cropSize: [number, number];\n  private cropBox: Tensor2D;\n  private cropBoxInd: Tensor1D;\n\n  private constructor(\n      protected readonly webcamVideoElement: HTMLVideoElement,\n      protected readonly webcamConfig: WebcamConfig) {\n    super();\n    if (this.needToResize()) {\n      this.resize = true;\n      this.cropSize =\n          [this.webcamConfig.resizeHeight, this.webcamConfig.resizeWidth];\n      this.cropBoxInd = tensor1d([0], 'int32');\n      if (this.webcamConfig.centerCrop) {\n        // Calculate the box based on resizing shape.\n        const widthCroppingRatio =\n            this.webcamConfig.resizeWidth * 1.0 / this.webcamVideoElement.width;\n        const heightCroppingRatio = this.webcamConfig.resizeHeight * 1.0 /\n            this.webcamVideoElement.height;\n        const widthCropStart = (1 - widthCroppingRatio) / 2;\n        const heightCropStart = (1 - heightCroppingRatio) / 2;\n        const widthCropEnd = widthCropStart + widthCroppingRatio;\n        const heightCropEnd = heightCroppingRatio + heightCropStart;\n        this.cropBox = tensor2d(\n            [heightCropStart, widthCropStart, heightCropEnd, widthCropEnd],\n            [1, 4]);\n      } else {\n        this.cropBox = tensor2d([0, 0, 1, 1], [1, 4]);\n      }\n    }\n  }\n\n  summary() {\n    return `webcam`;\n  }\n\n  // Construct a WebcamIterator and start it's video stream.\n  static async create(\n      webcamVideoElement?: HTMLVideoElement, webcamConfig: WebcamConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'tf.data.webcam is only supported in browser environment.');\n    }\n\n    if (!webcamVideoElement) {\n      // If webcam video element is not provided, create a hidden video element\n      // with provided width and height.\n      webcamVideoElement = document.createElement('video');\n      if (!webcamConfig.resizeWidth || !webcamConfig.resizeHeight) {\n        throw new Error(\n            'Please provide webcam video element, or resizeWidth and ' +\n            'resizeHeight to create a hidden video element.');\n      }\n      webcamVideoElement.width = webcamConfig.resizeWidth;\n      webcamVideoElement.height = webcamConfig.resizeHeight;\n    }\n    const webcamIterator = new WebcamIterator(webcamVideoElement, webcamConfig);\n\n    // Call async function to initialize the video stream.\n    await webcamIterator.start();\n\n    return webcamIterator;\n  }\n\n  // Async function to start video stream.\n  async start(): Promise<void> {\n    if (this.webcamConfig.facingMode) {\n      util.assert(\n          (this.webcamConfig.facingMode === 'user') ||\n              (this.webcamConfig.facingMode === 'environment'),\n          () =>\n              `Invalid webcam facing mode: ${this.webcamConfig.facingMode}. ` +\n              `Please provide 'user' or 'environment'`);\n    }\n\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        video: {\n          deviceId: this.webcamConfig.deviceId,\n          facingMode: this.webcamConfig.facingMode ?\n              this.webcamConfig.facingMode :\n              'user',\n          width: this.webcamVideoElement.width,\n          height: this.webcamVideoElement.height\n        }\n      });\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message = `Error thrown while initializing video stream: ${e.message}`;\n      throw e;\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain video from webcam.');\n    }\n\n    // Older browsers may not have srcObject\n    try {\n      this.webcamVideoElement.srcObject = this.stream;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = window.URL.createObjectURL(this.stream);\n    }\n    // Start the webcam video stream\n    this.webcamVideoElement.play();\n\n    this.isClosed = false;\n\n    return new Promise<void>(resolve => {\n      // Add event listener to make sure the webcam has been fully initialized.\n      this.webcamVideoElement.onloadedmetadata = () => {\n        resolve();\n      };\n    });\n  }\n\n  async next(): Promise<IteratorResult<Tensor3D>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let img;\n    try {\n      img = browser.fromPixels(this.webcamVideoElement);\n    } catch (e) {\n      throw new Error(\n          `Error thrown converting video to pixels: ${JSON.stringify(e)}`);\n    }\n    if (this.resize) {\n      try {\n        return {value: this.cropAndResizeFrame(img), done: false};\n      } catch (e) {\n        throw new Error(`Error thrown cropping the video: ${e.message}`);\n      } finally {\n        img.dispose();\n      }\n    } else {\n      return {value: img, done: false};\n    }\n  }\n\n  private needToResize() {\n    // If resizeWidth and resizeHeight are provided, and different from the\n    // width and height of original HTMLVideoElement, then resizing and cropping\n    // is required.\n    if (this.webcamConfig.resizeWidth && this.webcamConfig.resizeHeight &&\n        (this.webcamVideoElement.width !== this.webcamConfig.resizeWidth ||\n         this.webcamVideoElement.height !== this.webcamConfig.resizeHeight)) {\n      return true;\n    }\n    return false;\n  }\n\n  // Cropping and resizing each frame based on config\n  cropAndResizeFrame(img: Tensor3D): Tensor3D {\n    return tidy(() => {\n      const expandedImage: Tensor4D = img.toFloat().expandDims(0);\n      let resizedImage;\n      resizedImage = image.cropAndResize(\n          expandedImage, this.cropBox, this.cropBoxInd, this.cropSize,\n          'bilinear');\n      // Extract image from batch cropping.\n      const shape = resizedImage.shape;\n      return resizedImage.reshape(shape.slice(1) as [number, number, number]);\n    });\n  }\n\n  // Capture one frame from the video stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<Tensor3D> {\n    return (await this.next()).value;\n  }\n\n  // Stop the video stream and pause webcam iterator.\n  stop(): void {\n    const tracks = this.stream.getTracks();\n\n    tracks.forEach(track => track.stop());\n\n    try {\n      this.webcamVideoElement.srcObject = null;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = null;\n    }\n    this.isClosed = true;\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor3D[]> {\n    throw new Error('Can not convert infinite video stream to array.');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {ByteChunkIterator} from './iterators/byte_chunk_iterator';\n\n/**\n * Represents a data source readable as a stream of binary data chunks.\n *\n * Because `Dataset`s can be read repeatedly (via `Dataset.iterator()`), this\n * provides a means to repeatedly create streams from the underlying data\n * sources.\n */\nexport abstract class DataSource {\n  /**\n   * Obtain a new stream of binary data chunks.\n   *\n   * Starts the new stream from the beginning of the data source, even if other\n   * streams have been obtained previously.\n   */\n  abstract async iterator(): Promise<ByteChunkIterator>;\n\n  // TODO(soergel): consider chainable Dataset construction here\n}\n\n// TODO(soergel): consider convenience factory functions here\n// in combination with chainable source->dataset above, e.g.:\n// tf.data.url(...).asCsvDataset().shuffle().batch()\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\n\nexport abstract class StringIterator extends LazyIterator<string> {\n  /**\n   * Splits a string stream on a given separator.\n   *\n   * It is assumed that the incoming chunk boundaries have no semantic meaning,\n   * so conceptually the incoming stream is treated simply as the concatenation\n   * of its elements.\n   *\n   * The outgoing stream provides chunks corresponding to the results of the\n   * standard string split() operation (even if such a chunk spanned incoming\n   * chunks).  The separators are not included.\n   *\n   * A typical usage is to split a text file (represented as a stream with\n   * arbitrary chunk boundaries) into lines.\n   *\n   * @param upstream A readable stream of strings that can be treated as\n   *   concatenated.\n   * @param separator A character to split on.\n   */\n  split(separator: string): StringIterator {\n    return new SplitIterator(this, separator);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on StringIterator.  Unfortunately they can't be placed in separate files, due\n// to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class SplitIterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass SplitIterator extends StringIterator {\n  private impl: SplitIteratorImpl;\n\n  constructor(protected upstream: LazyIterator<string>, separator: string) {\n    super();\n    this.impl = new SplitIteratorImpl(upstream, separator);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\nclass SplitIteratorImpl extends OneToManyIterator<string> {\n  // A partial string at the end of an upstream chunk\n  carryover = '';\n\n  constructor(\n      protected upstream: LazyIterator<string>, protected separator: string) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Split('${this.separator}')`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    if (chunkResult.done) {\n      if (this.carryover === '') {\n        return false;\n      }\n\n      // Pretend that the pump succeeded in order to emit the small last batch.\n      // The next pump() call will actually fail.\n      this.outputQueue.push(this.carryover);\n      this.carryover = '';\n      return true;\n    }\n    const lines = chunkResult.value.split(this.separator);\n    // Note the behavior: \" ab \".split(' ') === ['', 'ab', '']\n    // Thus the carryover may be '' if the separator falls on a chunk\n    // boundary; this produces the correct result.\n\n    lines[0] = this.carryover + lines[0];\n    for (const line of lines.slice(0, -1)) {\n      this.outputQueue.push(line);\n    }\n    this.carryover = lines[lines.length - 1];\n\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\nimport {StringIterator} from './string_iterator';\n\nexport abstract class ByteChunkIterator extends LazyIterator<Uint8Array> {\n  /**\n   * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n   *\n   * The byte arrays producetd from the ByteChunkIterator on which this is\n   * called will be interpreted as concatenated.  No assumptions are made about\n   * the boundaries of the incoming chunks, so a multi-byte UTF8 encoding of a\n   * character may span the boundary between chunks.  This naturally happens,\n   * for instance, when reading fixed-size byte arrays from a file.\n   */\n  decodeUTF8(): StringIterator {\n    return new Utf8Iterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on ByteChunkIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class Utf8Iterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass Utf8Iterator extends StringIterator {\n  private impl: Utf8IteratorImpl;\n\n  constructor(protected upstream: LazyIterator<Uint8Array>) {\n    super();\n    this.impl = new Utf8IteratorImpl(upstream);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\n/**\n * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n *\n * This is tricky because the incoming byte array boundaries may disrupt a\n * multi-byte UTF8 character. Thus any incomplete character data at the end of\n * a chunk must be carried over and prepended to the next chunk before\n * decoding. Luckily with native decoder, TextDecoder in browser and\n * string_decoder in node, byte array boundaries are handled automatically.\n *\n * In the context of an input pipeline for machine learning, UTF8 decoding is\n * needed to parse text files containing training examples or prediction\n * requests (e.g., formatted as CSV or JSON). We cannot use the built-in\n * decoding provided by FileReader.readAsText() because here we are in a\n * streaming context, which FileReader does not support.\n *\n * @param upstream A `LazyIterator` of `Uint8Arrays` containing UTF8-encoded\n *   text, which should be interpreted as concatenated.  No assumptions are\n *   made about the boundaries of the incoming chunks, so a multi-byte UTF8\n *   encoding of a character may span the boundary between chunks.  This\n *   naturally happens, for instance, when reading fixed-size byte arrays from a\n *   file.\n */\nclass Utf8IteratorImpl extends OneToManyIterator<string> {\n  // `decoder` as `any` here to dynamically assign value based on the\n  // environment.\n  // tslint:disable-next-line:no-any\n  decoder: any;\n\n  constructor(protected readonly upstream: LazyIterator<Uint8Array>) {\n    super();\n    if (env().get('IS_BROWSER')) {\n      this.decoder = new TextDecoder('utf-8');\n    } else {\n      // tslint:disable-next-line:no-require-imports\n      const {StringDecoder} = require('string_decoder');\n      this.decoder = new StringDecoder('utf8');\n    }\n  }\n  summary() {\n    return `${this.upstream.summary()} -> Utf8`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    let chunk;\n    if (chunkResult.done) {\n      return false;\n    } else {\n      chunk = chunkResult.value;\n    }\n\n    let text: string;\n    if (env().get('IS_BROWSER')) {\n      text = this.decoder.decode(chunk, {stream: true});\n    } else {\n      text = this.decoder.write(Buffer.from(chunk.buffer));\n    }\n    this.outputQueue.push(text);\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// inspired by https://github.com/maxogden/filereader-stream\nimport {env, util} from '@tensorflow/tfjs-core';\nimport {FileElement} from '../types';\nimport {ByteChunkIterator} from './byte_chunk_iterator';\n\nexport interface FileChunkIteratorOptions {\n  /** The byte offset at which to begin reading the File or Blob. Default 0. */\n  offset?: number;\n  /** The number of bytes to read at a time. Default 1MB. */\n  chunkSize?: number;\n}\n\n/**\n * Provide a stream of chunks from a File, Blob, or Uint8Array.\n * @param file The source File, Blob or Uint8Array.\n * @param options Optional settings controlling file reading.\n * @returns a lazy Iterator of Uint8Arrays containing sequential chunks of the\n *   input File, Blob or Uint8Array.\n */\nexport class FileChunkIterator extends ByteChunkIterator {\n  offset: number;\n  chunkSize: number;\n\n  constructor(\n      protected file: FileElement,\n      protected options: FileChunkIteratorOptions = {}) {\n    super();\n    util.assert(\n        (file instanceof Uint8Array) ||\n            (env().get('IS_BROWSER') ?\n                 (file instanceof File || file instanceof Blob) :\n                 false),\n        () => 'FileChunkIterator only supports File, Blob and Uint8Array ' +\n            'right now.');\n    this.offset = options.offset || 0;\n    // default 1MB chunk has tolerable perf on large files\n    this.chunkSize = options.chunkSize || 1024 * 1024;\n  }\n\n  summary() {\n    return `FileChunks ${this.file}`;\n  }\n\n  async next(): Promise<IteratorResult<Uint8Array>> {\n    if (this.offset >= ((this.file instanceof Uint8Array) ?\n                            this.file.byteLength :\n                            this.file.size)) {\n      return {value: null, done: true};\n    }\n    const chunk = new Promise<Uint8Array>((resolve, reject) => {\n      const end = this.offset + this.chunkSize;\n      if (this.file instanceof Uint8Array) {\n        // Note if end > this.uint8Array.byteLength, we just get a small last\n        // chunk.\n        resolve(new Uint8Array(this.file.slice(this.offset, end)));\n      } else {\n        // This branch assumes that this.file type is File or Blob, which\n        // means it is in the browser environment.\n\n        // TODO(soergel): is this a performance issue?\n        const fileReader = new FileReader();\n        fileReader.onload = (event) => {\n          let data: string|ArrayBuffer|Uint8Array = fileReader.result;\n          // Not sure we can trust the return type of\n          // FileReader.readAsArrayBuffer See e.g.\n          // https://github.com/node-file-api/FileReader/issues/2\n          if (data instanceof ArrayBuffer) {\n            data = new Uint8Array(data);\n          }\n          if (!(data instanceof Uint8Array)) {\n            return reject(new TypeError('FileReader returned unknown type.'));\n          }\n          resolve(data);\n        };\n        fileReader.onabort = (event) => {\n          return reject(new Error('Aborted'));\n        };\n        fileReader.onerror = (event) => {\n          return reject(new Error(event.type));\n        };\n        // TODO(soergel): better handle onabort, onerror\n        // Note if end > this.file.size, we just get a small last chunk.\n        const slice = this.file.slice(this.offset, end);\n        // We can't use readAsText here (even if we know the file is text)\n        // because the slice boundary may fall within a multi-byte character.\n        fileReader.readAsArrayBuffer(slice);\n      }\n      this.offset = end;\n    });\n    return {value: (await chunk), done: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\nimport {FileChunkIterator, FileChunkIteratorOptions} from './file_chunk_iterator';\n\n/**\n * Provide a stream of chunks from a URL.\n *\n * Note this class first downloads the entire file into memory before providing\n * the first element from the stream.  This is because the Fetch API does not\n * yet reliably provide a reader stream for the response body.\n */\nexport async function urlChunkIterator(\n    url: RequestInfo, options: FileChunkIteratorOptions = {}) {\n  let urlString;\n  let requestInit;\n  if ((typeof url) === 'string') {\n    urlString = url as string;\n  } else {\n    urlString = (url as Request).url;\n    requestInit = getRequestInitFromRequest(url as Request);\n  }\n  const response = await util.fetch(urlString, requestInit);\n  if (response.ok) {\n    const uint8Array = new Uint8Array(await response.arrayBuffer());\n    return new FileChunkIterator(uint8Array, options);\n  } else {\n    throw new Error(response.statusText);\n  }\n}\n\n// Generate RequestInit from Request to match tf.util.fetch signature.\nconst getRequestInitFromRequest = (request: Request) => {\n  const init = {\n    method: request.method,\n    headers: request.headers,\n    body: request.body,\n    mode: request.mode,\n    credentials: request.credentials,\n    cache: request.cache,\n    redirect: request.redirect,\n    referrer: request.referrer,\n    integrity: request.integrity,\n  };\n  return init;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// Skip tslint any type check cause this method is aiming to check type of\n// input.\n// tslint:disable-next-line:no-any\nexport function isLocalPath(source: any): boolean {\n  return (typeof source === 'string') && source.substr(0, 7) === 'file://';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIterator, FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {FileElement} from '../types';\nimport {isLocalPath} from '../util/source_util';\n\n/**\n * Represents a file, blob, or Uint8Array readable as a stream of binary data\n * chunks.\n */\nexport class FileDataSource extends DataSource {\n  /**\n   * Create a `FileDataSource`.\n   *\n   * @param input Local file path, or `File`/`Blob`/`Uint8Array` object to\n   *     read. Local file only works in node environment.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected input: FileElement|string,\n      protected readonly options: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.input) && env().get('IS_NODE')) {\n      // tslint:disable-next-line:no-require-imports\n      const fs = require('fs');\n      this.input = fs.readFileSync((this.input as string).substr(7));\n    }\n    // TODO(kangyizhang): Add LocalFileChunkIterator to split local streaming\n    // with file in browser.\n    return new FileChunkIterator(this.input as FileElement, this.options);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {urlChunkIterator} from '../iterators/url_chunk_iterator';\nimport {isLocalPath} from '../util/source_util';\nimport {FileDataSource} from './file_data_source';\n\n/*\n * Represents a URL readable as a stream of binary data chunks.\n */\nexport class URLDataSource extends DataSource {\n  /**\n   * Create a `URLDataSource`.\n   *\n   * @param url A source URL string, or a `Request` object.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected readonly url: RequestInfo,\n      protected readonly fileOptions: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  // TODO(soergel): provide appropriate caching options.  Currently this\n  // will download the URL anew for each call to iterator().  Since we have\n  // to treat the downloaded file as a blob/buffer anyway, we may as well retain\n  // it-- but that raises GC issues.  Also we may want a persistent disk cache.\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.url)) {\n      return (new FileDataSource(this.url as string, this.fileOptions))\n          .iterator();\n    } else {\n      return urlChunkIterator(this.url, this.fileOptions);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer} from '@tensorflow/tfjs-core';\nimport {Dataset, datasetFromIteratorFn} from './dataset';\nimport {CSVDataset} from './datasets/csv_dataset';\nimport {iteratorFromFunction} from './iterators/lazy_iterator';\nimport {MicrophoneIterator} from './iterators/microphone_iterator';\nimport {WebcamIterator} from './iterators/webcam_iterator';\nimport {URLDataSource} from './sources/url_data_source';\nimport {CSVConfig, MicrophoneConfig, WebcamConfig} from './types';\n\n/**\n * Create a `CSVDataset` by reading and decoding CSV file(s) from provided URL\n * or local path if it's in Node environment.\n *\n * Note: If isLabel in columnConfigs is `true` for at least one column, the\n * element in returned `CSVDataset` will be an object of\n * `{xs:features, ys:labels}`: xs is a dict of features key/value pairs, ys\n * is a dict of labels key/value pairs. If no column is marked as label,\n * returns a dict of features only.\n *\n * ```js\n * const csvUrl =\n * 'https://storage.googleapis.com/tfjs-examples/multivariate-linear-regression/data/boston-housing-train.csv';\n *\n * async function run() {\n *   // We want to predict the column \"medv\", which represents a median value of\n *   // a home (in $1000s), so we mark it as a label.\n *   const csvDataset = tf.data.csv(\n *     csvUrl, {\n *       columnConfigs: {\n *         medv: {\n *           isLabel: true\n *         }\n *       }\n *     });\n *\n *   // Number of features is the number of column names minus one for the label\n *   // column.\n *   const numOfFeatures = (await csvDataset.columnNames()).length - 1;\n *\n *   // Prepare the Dataset for training.\n *   const flattenedDataset =\n *     csvDataset\n *     .map(({xs, ys}) =>\n *       {\n *         // Convert xs(features) and ys(labels) from object form (keyed by\n *         // column name) to array form.\n *         return {xs:Object.values(xs), ys:Object.values(ys)};\n *       })\n *     .batch(10);\n *\n *   // Define the model.\n *   const model = tf.sequential();\n *   model.add(tf.layers.dense({\n *     inputShape: [numOfFeatures],\n *     units: 1\n *   }));\n *   model.compile({\n *     optimizer: tf.train.sgd(0.000001),\n *     loss: 'meanSquaredError'\n *   });\n *\n *   // Fit the model using the prepared Dataset\n *   return model.fitDataset(flattenedDataset, {\n *     epochs: 10,\n *     callbacks: {\n *       onEpochEnd: async (epoch, logs) => {\n *         console.log(epoch + ':' + logs.loss);\n *       }\n *     }\n *   });\n * }\n *\n * await run();\n * ```\n *\n * @param source URL or local path to get CSV file. If it's a local path, it\n * must have prefix `file://` and it only works in node environment.\n * @param csvConfig (Optional) A CSVConfig object that contains configurations\n *     of reading and decoding from CSV file(s).\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function csv(\n    source: RequestInfo, csvConfig: CSVConfig = {}): CSVDataset {\n  return new CSVDataset(new URLDataSource(source), csvConfig);\n}\n\n/**\n * Create a `Dataset` that produces each element by calling a provided function.\n *\n * Note that repeated iterations over this `Dataset` may produce different\n * results, because the function will be called anew for each element of each\n * iteration.\n *\n * Also, beware that the sequence of calls to this function may be out of order\n * in time with respect to the logical order of the Dataset. This is due to the\n * asynchronous lazy nature of stream processing, and depends on downstream\n * transformations (e.g. .shuffle()). If the provided function is pure, this is\n * no problem, but if it is a closure over a mutable state (e.g., a traversal\n * pointer), then the order of the produced elements may be scrambled.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const ds = tf.data.func(func);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param f A function that produces one data element on each call.\n */\nexport function func<T extends TensorContainer>(\n    f: () => IteratorResult<T>| Promise<IteratorResult<T>>): Dataset<T> {\n  const iter = iteratorFromFunction(f);\n  return datasetFromIteratorFn(async () => iter);\n}\n\n/**\n * Create a `Dataset` that produces each element from provided JavaScript\n * generator, which is a function*\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions),\n * or a function that returns an\n * iterator\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions).\n *\n * The returned iterator should have `.next()` function that returns element in\n * format of `{value: TensorContainer, done:boolean}`.\n *\n * Example of creating a dataset from an iterator factory:\n * ```js\n * function makeIterator() {\n *   const numElements = 10;\n *   let index = 0;\n *\n *   const iterator = {\n *     next: () => {\n *       let result;\n *       if (index < numElements) {\n *         result = {value: index, done: false};\n *         index++;\n *         return result;\n *       }\n *       return {value: index, done: true};\n *     }\n *   };\n *   return iterator;\n * }\n * const ds = tf.data.generator(makeIterator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * Example of creating a dataset from a generator:\n * ```js\n * function* dataGenerator() {\n *   const numElements = 10;\n *   let index = 0;\n *   while (index < numElements) {\n *     const x = index;\n *     index++;\n *     yield x;\n *   }\n * }\n *\n * const ds = tf.data.generator(dataGenerator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param generator A Javascript generator function that returns a JavaScript\n *     iterator.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function generator<T extends TensorContainer>(\n    generator: () => Iterator<T>| Promise<Iterator<T>>): Dataset<T> {\n  return datasetFromIteratorFn(async () => {\n    const gen = await generator();\n    return iteratorFromFunction(() => gen.next());\n  });\n}\n\n/**\n * Create an iterator that generate `Tensor`s from webcam video stream. This API\n * only works in Browser environment when the device has webcam.\n *\n * Note: this code snippet only works when the device has a webcam. It will\n * request permission to open the webcam when running.\n * ```js\n * const videoElement = document.createElement('video');\n * videoElement.width = 100;\n * videoElement.height = 100;\n * const cam = await tf.data.webcam(videoElement);\n * const img = await cam.capture();\n * img.print();\n * cam.stop();\n * ```\n *\n * @param webcamVideoElement A `HTMLVideoElement` used to play video from\n *     webcam. If this element is not provided, a hidden `HTMLVideoElement` will\n *     be created. In that case, `resizeWidth` and `resizeHeight` must be\n *     provided to set the generated tensor shape.\n * @param webcamConfig A `WebcamConfig` object that contains configurations of\n *     reading and manipulating data from webcam video stream.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function webcam(\n    webcamVideoElement?: HTMLVideoElement,\n    webcamConfig?: WebcamConfig): Promise<WebcamIterator> {\n  return WebcamIterator.create(webcamVideoElement, webcamConfig);\n}\n\n/**\n * Create an iterator that generate frequency-domain spectrogram `Tensor`s from\n * microphone audio stream with browser's native FFT. This API only works in\n * browser environment when the device has microphone.\n *\n * Note: this code snippet only works when the device has a microphone. It will\n * request permission to open the microphone when running.\n * ```js\n * const mic = await tf.data.microphone({\n *   fftSize: 1024,\n *   columnTruncateLength: 232,\n *   numFramesPerSpectrogram: 43,\n *   sampleRateHz:44100,\n *   includeSpectrogram: true,\n *   includeWaveform: true\n * });\n * const audioData = await mic.capture();\n * const spectrogramTensor = audioData.spectrogram;\n * spectrogramTensor.print();\n * const waveformTensor = audioData.waveform;\n * waveformTensor.print();\n * mic.stop();\n * ```\n *\n * @param microphoneConfig A `MicrophoneConfig` object that contains\n *     configurations of reading audio data from microphone.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function microphone(microphoneConfig?: MicrophoneConfig):\n    Promise<MicrophoneIterator> {\n  return MicrophoneIterator.create(microphoneConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${\n              opName} does not support complex64 tensors in the CPU backend.`);\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {backend_util, BackendTimingInfo, DataStorage, DataType, DataValues, engine, env, kernel_impls, KernelBackend, max, NumericDataType, Rank, Scalar, ShapeMap, slice_util, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, TensorInfo, TypedArray, upcastType, util} from '@tensorflow/tfjs-core';\n\nconst nonMaxSuppressionV3Impl = kernel_impls.nonMaxSuppressionV3Impl;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\nimport * as seedrandom from 'seedrandom';\nimport {assertNotComplex} from './cpu_util';\n\ninterface DataId {}\n\nfunction mapActivation(\n    backend: MathBackendCPU, x: Tensor, activation: backend_util.Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return backend.linear(x);\n  } else if (activation === 'relu') {\n    return backend.relu(x);\n  } else if (activation === 'elu') {\n    return tf.elu(x);\n  } else if (activation === 'relu6') {\n    return backend.relu6(x);\n  } else if (activation === 'prelu') {\n    return backend.prelu(x, preluActivationWeights);\n  }\n  throw new Error(\n      `Activation ${activation} has not been implemented for the CPU backend.`);\n}\n\nexport interface TensorData<D extends DataType> {\n  values?: backend_util.BackendValues;\n  dtype: D;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensorInfos field.\n  complexTensorInfos?: {real: TensorInfo, imag: TensorInfo};\n  // refCount keeps track of how many tensors reference it. Used for memory\n  // management.\n  refCount: number;\n}\n\nexport class MathBackendCPU extends KernelBackend {\n  public blockSize = 48;\n\n  data: DataStorage<TensorData<DataType>>;\n  private firstUse = true;\n\n  constructor() {\n    super();\n    this.data = new DataStorage(this, engine());\n  }\n\n  write(values: backend_util.BackendValues, shape: number[], dtype: DataType):\n      DataId {\n    if (this.firstUse) {\n      this.firstUse = false;\n      if (env().get('IS_NODE')) {\n        backend_util.warn(\n            '\\n============================\\n' +\n            'Hi there \uD83D\uDC4B. Looks like you are running TensorFlow.js in ' +\n            'Node.js. To speed things up dramatically, install our node ' +\n            'backend, which binds to TensorFlow C++, by running ' +\n            'npm i @tensorflow/tfjs-node, ' +\n            'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +\n            'Then call require(\\'@tensorflow/tfjs-node\\'); (-gpu ' +\n            'suffix for CUDA) at the start of your program. ' +\n            'Visit https://github.com/tensorflow/tfjs-node for more details.' +\n            '\\n============================');\n      }\n    }\n    const dataId = {};\n\n    this.data.set(dataId, {values, dtype, refCount: 1});\n\n    return dataId;\n  }\n\n  /**\n   * Create a data bucket in cpu backend.\n   * @param shape Shape of the `TensorInfo`.\n   * @param dtype DType of the `TensorInfo`.\n   * @param values The value of the `TensorInfo` stored as a flattened array.\n   */\n  makeTensorInfo(\n      shape: number[], dtype: DataType,\n      values?: backend_util.BackendValues): TensorInfo {\n    const outId = this.write(values, shape, dtype);\n\n    return {dataId: outId, shape, dtype};\n  }\n\n  /** Increase refCount of a `TensorData`. */\n  incRef(dataId: DataId): void {\n    const tensorData = this.data.get(dataId);\n    tensorData.refCount++;\n  }\n\n  /** Decrease refCount of a `TensorData`. */\n  decRef(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n      tensorData.refCount--;\n    }\n  }\n\n  move(\n      dataId: DataId, values: backend_util.BackendValues, shape: number[],\n      dtype: DataType): void {\n    this.data.set(dataId, {values, dtype, refCount: 1});\n  }\n\n  numDataIds(): number {\n    return this.data.numDataIds();\n  }\n\n  async read(dataId: DataId): Promise<backend_util.BackendValues> {\n    return this.readSync(dataId);\n  }\n  readSync(dataId: DataId): backend_util.BackendValues {\n    const {dtype, complexTensorInfos} = this.data.get(dataId);\n\n    if (dtype === 'complex64') {\n      const realValues =\n          this.readSync(complexTensorInfos.real.dataId) as Float32Array;\n      const imagValues =\n          this.readSync(complexTensorInfos.imag.dataId) as Float32Array;\n      return backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    }\n\n    return this.data.get(dataId).values;\n  }\n\n  private bufferSync<R extends Rank>(t: Tensor<R>): TensorBuffer<R> {\n    const data = this.readSync(t.dataId);\n    let decodedData = data as DataValues;\n    if (t.dtype === 'string') {\n      try {\n        // Decode the bytes into string.\n        decodedData = (data as Uint8Array[]).map(d => util.decodeString(d));\n      } catch {\n        throw new Error('Failed to decode encoded string bytes into utf-8');\n      }\n    }\n    return tf.buffer(t.shape, t.dtype, decodedData) as TensorBuffer<R>;\n  }\n\n  makeOutput<T extends Tensor>(\n      values: backend_util.BackendValues, shape: number[], dtype: DataType): T {\n    const dataId = this.write(values, shape, dtype);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  disposeData(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const {complexTensorInfos} = this.data.get(dataId);\n\n      if (complexTensorInfos != null) {\n        this.disposeData(complexTensorInfos.real.dataId);\n        this.disposeData(complexTensorInfos.imag.dataId);\n      }\n\n      this.data.delete(dataId);\n    }\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n\n      tensorData.refCount--;\n\n      if (tensorData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  async time(f: () => void): Promise<BackendTimingInfo> {\n    const start = util.now();\n    f();\n    const kernelMs = util.now() - start;\n    return {kernelMs};\n  }\n\n  memory() {\n    return {\n      // Unreliable due to automatic gc. The numbers above are cumulative.\n      unreliable: true,\n      reasons:\n          ['The reported memory is an upper bound. Due to automatic garbage ' +\n           'collection, the true allocated memory may be less.']\n    };\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    assertNotComplex(x, 'stridedSlice');\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tf.tensor([], outShape) as T;\n    }\n\n    const buffer = tf.buffer(outShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n    for (let i = 0; i < buffer.size; i++) {\n      const loc = buffer.indexToLoc(i);\n\n      const newLoc: number[] = new Array(loc.length);\n      for (let j = 0; j < newLoc.length; j++) {\n        newLoc[j] = loc[j] * strides[j] + begin[j];\n      }\n      buffer.set(xBuf.get(...newLoc), ...loc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  diag(x: Tensor): Tensor {\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const buffer = tf.buffer([x.size, x.size], x.dtype);\n    const vals = buffer.values;\n    for (let i = 0; i < xVals.length; i++) {\n      vals[i * x.size + i] = xVals[i];\n    }\n    return buffer.toTensor();\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = tf.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    assertNotComplex(x, 'reverse');\n\n    const buffer = tf.buffer(x.shape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < buffer.size; i++) {\n      const outLoc = buffer.indexToLoc(i);\n      const inLoc = outLoc.slice();\n      axis.forEach(ax => inLoc[ax] = x.shape[ax] - 1 - inLoc[ax]);\n      buffer.set(xBuf.get(...inLoc), ...outLoc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'neg');\n\n    // TODO(lina128): Use mul directly once neg is modularized.\n    return tf.mul(tf.scalar(-1), x);\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    assertNotComplex(tensors, 'addN');\n\n    const vals = tensors.map(t => this.readSync(t.dataId) as TypedArray);\n    const result = tf.buffer(tensors[0].shape, tensors[0].dtype as 'float32');\n    const resultVals = result.values;\n    for (let i = 0; i < tensors.length; i++) {\n      const currVals = vals[i];\n      for (let j = 0; j < resultVals.length; j++) {\n        resultVals[j] += currVals[j];\n      }\n    }\n    return result.toTensor() as T;\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n\n    // TODO(lina128): Use sub directly once softmax is modularized.\n    const a = tf.sub(logits, maxLogit.reshape(expandedShape));\n    const b = tf.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax\n    // kernel modularization.\n    return tf.div(b, sumExp);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    assertNotComplex([a, b], 'pow');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.pow(aValue, bValue)) as\n        T;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    assertNotComplex([a, b], 'matMul');\n\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const leftDim = transposeA ? a.shape[2] : a.shape[1];\n    const rightDim = transposeB ? b.shape[1] : b.shape[2];\n    const batchDim = a.shape[0];\n\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const [aBatch, aOuterStep, aInnerStep] = transposeA ?\n        [a.strides[0], 1, a.strides[1]] :\n        [a.strides[0], a.strides[1], 1];\n    const [bInnerStep, bOuterStep, bBatch] = transposeB ?\n        [1, b.strides[1], b.strides[0]] :\n        [b.strides[1], 1, b.strides[0]];\n\n    const size = leftDim * rightDim;\n    const result = tf.buffer([batchDim, leftDim, rightDim], a.dtype);\n    const resVals = result.values as TypedArray;\n    const blockSize = this.blockSize;\n\n    for (let b = 0; b < batchDim; b++) {\n      for (let i0 = 0; i0 < leftDim; i0 += blockSize) {\n        for (let j0 = 0; j0 < rightDim; j0 += blockSize) {\n          for (let k0 = 0; k0 < sharedDim; k0 += blockSize) {\n            // for when blockSize doesn't evenly divide the input\n            const iBlock = Math.min(i0 + blockSize, leftDim);\n            const jBlock = Math.min(j0 + blockSize, rightDim);\n            const kBlock = Math.min(k0 + blockSize, sharedDim);\n\n            for (let i = i0; i < iBlock; i++) {\n              for (let j = j0; j < jBlock; j++) {\n                let sum = 0.0;\n\n                for (let k = k0; k < kBlock; k++) {\n                  sum += aValues[b * aBatch + i * aOuterStep + k * aInnerStep] *\n                      bValues[k * bInnerStep + j * bOuterStep + b * bBatch];\n                }\n                resVals[b * size + (i * rightDim + j)] += sum;\n              }\n            }\n          }\n        }\n      }\n    }\n    return result.toTensor() as Tensor3D;\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    let result = this.batchMatMul(a, b, transposeA, transposeB);\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedBatchMatMul is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor3D;\n    }\n\n    return result;\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'floorDiv');\n\n    const op = (a: number, b: number) => Math.floor(a / b);\n    const outputDtype = 'int32';\n    return this.broadcastedBinaryOp(a, b, outputDtype, op);\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let sum = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        sum += aVals[offset + j];\n      }\n      vals[i] = sum;\n    }\n    return result;\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let prod = 1;\n      for (let j = 0; j < reduceSize; ++j) {\n        prod *= aVals[offset + j];\n      }\n      vals[i] = prod;\n    }\n    return result;\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    assertNotComplex(x, 'unsortedSegmentSum');\n\n    const res = [];\n\n    // Reshape the segment id's so that they can be broadcast with\n    // x. The new shape should be [segmentIds.shape, 1, ..., 1]\n    const numIters = x.rank - segmentIds.rank;\n    for (let i = 0; i < numIters; ++i) {\n      segmentIds = segmentIds.expandDims(i + 1);\n    }\n\n    for (let i = 0; i < numSegments; ++i) {\n      const segmentId = tf.scalar(i, 'int32');\n      const mask = tf.equal(segmentId, segmentIds).asType('float32');\n      const sum = mask.mul(x).sum(0);\n      res.push(sum);\n    }\n\n    return tf.stack(res);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMin');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMin', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      let minIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n          minIndex = j;\n        }\n      }\n      vals[i] = minIndex;\n    }\n    return result;\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMax');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMax', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let max = aVals[offset];\n      let maxIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value > max) {\n          max = value;\n          maxIndex = j;\n        }\n      }\n      vals[i] = maxIndex;\n    }\n    return result;\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    assertNotComplex(x, 'cumsum');\n\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `backend.cumsum in CPU expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(x.shape, resultDtype);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    const finalDim = x.shape[x.rank - 1];\n    const indexAdjuster = reverse ?\n        (i: number, j: number) => i + finalDim - j - 1 :\n        (i: number, j: number) => i + j;\n    for (let i = 0; i < aVals.length; i += finalDim) {\n      for (let j = 0; j < finalDim; j++) {\n        const idx = indexAdjuster(i, j);\n        if (j === 0) {\n          vals[idx] = exclusive ? 0 : aVals[idx];\n        } else {\n          const prevIdx = indexAdjuster(i, j - 1);\n          vals[idx] = exclusive ? aVals[prevIdx] + vals[prevIdx] :\n                                  aVals[idx] + vals[prevIdx];\n        }\n      }\n    }\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'equal');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal === bVal) ? 1 : 0;\n    });\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'notEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal !== bVal) ? 1 : 0;\n    });\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'less');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal < bVal) ? 1 : 0;\n    });\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'lessEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal <= bVal) ? 1 : 0;\n    });\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greater');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal > bVal) ? 1 : 0;\n    });\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greaterEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal >= bVal) ? 1 : 0;\n    });\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalAnd');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal && bVal;\n    });\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalOr');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal || bVal;\n    });\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([condition, a, b], 'select');\n\n    const values = this.readSync(condition.dataId) as TypedArray;\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const result = tf.zeros(a.shape, upcastType(a.dtype, b.dtype));\n    const newValues = this.readSync(result.dataId) as TypedArray;\n    let index = 0;\n    const offset = condition.rank === 0 || condition.rank > 1 || a.rank === 1 ?\n        1 :\n        util.sizeFromShape(a.shape.slice(1));\n\n    for (let i = 0; i < values.length; i++) {\n      for (let j = 0; j < offset; j++) {\n        if (values[i] === 1) {\n          newValues[index++] = aValues[i];\n        } else {\n          newValues[index++] = bValues[i];\n        }\n      }\n    }\n\n    return result;\n  }\n\n  where(condition: Tensor): Tensor2D {\n    assertNotComplex([condition], 'where');\n\n    const condVals = this.readSync(condition.dataId) as TypedArray;\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    assertNotComplex(x, 'topk');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'min');\n\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n        }\n      }\n      vals[i] = min;\n    }\n    return result;\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'minimum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.min(aVal, bVal));\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'mod');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const rem = aVal % bVal;\n      if ((aVal < 0 && bVal < 0) || (aVal >= 0 && bVal >= 0)) {\n        return rem;\n      } else {\n        return (rem + bVal) % bVal;\n      }\n    });\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'maximum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.max(aVal, bVal));\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'all');\n\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let all = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        all = all && value;\n      }\n      vals[i] = all;\n    }\n    return result;\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'any');\n\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let anyVal = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        anyVal = anyVal || value;\n      }\n      vals[i] = anyVal;\n    }\n    return result;\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'squaredDifference');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const diff = aVal - bVal;\n      return diff * diff;\n    });\n  }\n\n  linear<T extends Tensor>(x: T): T {\n    return x;\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.max(0, inVals[i]);\n    }\n    return res as T;\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.min(Math.max(0, inVals[i]), 6);\n    }\n    return res as T;\n  }\n\n  prelu<T extends Tensor>(x: T, a: T): T {\n    assertNotComplex([x, a], 'prelu');\n\n    return this.broadcastedBinaryOp(\n               x, a, x.dtype,\n               (xValue, aValue) => xValue < 0 ? aValue * xValue : xValue) as T;\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    assertNotComplex([dy, y], 'eluDer');\n\n    const resultValues = new Float32Array(y.size);\n    const values = this.readSync(y.dataId) as TypedArray;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    for (let i = 0; i < values.length; ++i) {\n      const v = values[i];\n      if (v >= 1) {\n        resultValues[i] = dyValues[i];\n      } else {\n        resultValues[i] = dyValues[i] * (v + 1);\n      }\n    }\n    return this.makeOutput(resultValues, y.shape, 'float32');\n  }\n\n  atan2<T extends Tensor>(a: T, b: T): T {\n    assertNotComplex([a, b], 'atan2');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.atan2(aValue, bValue)) as\n        T;\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.conv2d(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedConv2d is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, filter], 'conv2d');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n\n    const xBatchStride = x.strides[0];\n    const xRowStride = isChannelsLast ? x.strides[1] : x.strides[2];\n    const xColStride = isChannelsLast ? x.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : x.strides[1];\n    const yBatchStride = y.strides[0];\n    const yRowStride = isChannelsLast ? y.strides[1] : y.strides[2];\n    const yColStride = isChannelsLast ? y.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : y.strides[1];\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * xBatchStride;\n      const yOffset1 = b * yBatchStride;\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * yRowStride;\n        const xRCorner = yR * convInfo.strideHeight - padTop;\n        for (let wR = 0; wR < filterHeight; wR++) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * xRowStride;\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * yColStride;\n            const xCCorner = yC * convInfo.strideWidth - padLeft;\n            for (let wC = 0; wC < filterWidth; wC++) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * xColStride;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1 * xChannelStride];\n                for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                  yVals[yOffset3 + d2 * yChannelStride] +=\n                      xVal * wVals[wOffset3 + d2];\n                }\n                wOffset3 += convInfo.outChannels;\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor() as Tensor4D;\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padFront = convInfo.padInfo.front;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const y = tf.buffer<Rank.R5>(convInfo.outShape, x.dtype as 'float32');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yF = 0; yF < convInfo.outDepth; ++yF) {\n        const yOffset2 = yOffset1 + yF * y.strides[1];\n        const xFCorner = yF * convInfo.strideDepth - padFront;\n        for (let wF = 0; wF < filterDepth; wF++) {\n          const xF = xFCorner + wF * dilationDepth;\n          if (xF < 0 || xF >= convInfo.inDepth) {\n            continue;\n          }\n          const wOffset1 = wF * filter.strides[0];\n          const xOffset2 = xOffset1 + xF * x.strides[1];\n\n          for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n            const yOffset3 = yOffset2 + yR * y.strides[2];\n            const xRCorner = yR * convInfo.strideHeight - padTop;\n            for (let wR = 0; wR < filterHeight; wR++) {\n              const xR = xRCorner + wR * dilationHeight;\n              if (xR < 0 || xR >= convInfo.inHeight) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wR * filter.strides[1];\n              const xOffset3 = xOffset2 + xR * x.strides[2];\n              for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n                const yOffset4 = yOffset3 + yC * convInfo.outChannels;\n                const xCCorner = yC * convInfo.strideWidth - padLeft;\n                for (let wC = 0; wC < filterWidth; wC++) {\n                  const xC = xCCorner + wC * dilationWidth;\n                  if (xC < 0 || xC >= convInfo.inWidth) {\n                    continue;\n                  }\n                  const wOffset3 = wOffset2 + wC * filter.strides[2];\n                  const xOffset4 = xOffset3 + xC * convInfo.inChannels;\n                  let wOffset4 = wOffset3;\n                  for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                    const xVal = xVals[xOffset4 + d1];\n                    for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                      yVals[yOffset4 + d2] += xVal * wVals[wOffset4 + d2];\n                    }\n                    wOffset4 += convInfo.outChannels;\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor();\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'conv2dDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth,\n      dataFormat\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const xBatchStride = dx.strides[0];\n    const xRowStride = isChannelsLast ? dx.strides[1] : dx.strides[2];\n    const xColStride = isChannelsLast ? dx.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : dx.strides[1];\n    const yBatchStride = dy.strides[0];\n    const yRowStride = isChannelsLast ? dy.strides[1] : dy.strides[2];\n    const yColStride = isChannelsLast ? dy.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : dy.strides[1];\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset =\n                    yBatchStride * b + yRowStride * yR + yColStride * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let d2 = 0; d2 < outChannels; ++d2) {\n                  const pixel = dyValues[dyOffset + yChannelStride * d2];\n                  const weight = fltValues[fltOffset + d2];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            const dxOffset = xBatchStride * b + xRowStride * xR +\n                xColStride * xC + xChannelStride * d1;\n            dxValues[dxOffset] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const dx = tf.buffer<Rank.R5>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2, dxS3] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2, fltS3] = filter.strides;\n    const {\n      batchSize,\n      filterDepth,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inDepth,\n      inHeight,\n      inWidth,\n      outChannels,\n      outDepth,\n      outHeight,\n      outWidth,\n      strideDepth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const frontPad = filterDepth - 1 - convInfo.padInfo.front;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        // Frames of depth\n        for (let xF = 0; xF < inDepth; ++xF) {\n          const xFCorner = xF - frontPad;\n          const xFMin = Math.max(0, Math.ceil(xFCorner / strideDepth));\n          const yFMax =\n              Math.min(outDepth, (filterDepth + xFCorner) / strideDepth);\n\n          // Rows as per standard 2d matrix notation\n          for (let xR = 0; xR < inHeight; ++xR) {\n            const xRCorner = xR - topPad;\n            const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n            const yRMax =\n                Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n            // Columns as per standard 2d matrix notation\n            for (let xC = 0; xC < inWidth; ++xC) {\n              const xCCorner = xC - leftPad;\n              const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n              const yCMax =\n                  Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n              let dotProd = 0;\n              for (let yF = xFMin; yF < yFMax; ++yF) {\n                const wF = yF * strideDepth - xFCorner;\n\n                for (let yR = xRMin; yR < yRMax; ++yR) {\n                  const wR = yR * strideHeight - xRCorner;\n\n                  for (let yC = xCMin; yC < yCMax; ++yC) {\n                    const wC = yC * strideWidth - xCCorner;\n                    const dyOffset =\n                        dyS0 * b + dyS1 * yF + dyS2 * yR + dyS3 * yC;\n                    const fltOffset = fltS0 * (filterDepth - 1 - wF) +\n                        fltS1 * (filterHeight - 1 - wR) +\n                        fltS2 * (filterWidth - 1 - wC) + fltS3 * d1;\n\n                    for (let d2 = 0; d2 < outChannels; ++d2) {\n                      const pixel = dyValues[dyOffset + d2];\n                      const weight = fltValues[fltOffset + d2];\n                      dotProd += pixel * weight;\n                    }\n                  }\n                }\n              }\n              dxValues[dxS0 * b + dxS1 * xF + dxS2 * xR + dxS3 * xC + d1] =\n                  dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, dy], 'conv2dDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n          for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n            // Need to convolve.\n            let dotProd = 0;\n            for (let b = 0; b < convInfo.batchSize; ++b) {\n              for (let yR = yRMin; yR < yRMax; ++yR) {\n                const xR = wR + yR * strideHeight - topPad;\n                for (let yC = yCMin; yC < yCMax; ++yC) {\n                  const xC = wC + yC * strideWidth - leftPad;\n                  if (isChannelsLast) {\n                    dotProd +=\n                        xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n                  } else {\n                    dotProd +=\n                        xBuf.get(b, d1, xR, xC) * dyBuf.get(b, d2, yR, yC);\n                  }\n                }\n              }\n            }\n            dW.set(dotProd, wR, wC, d1, d2);\n          }\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const dw = tf.buffer<Rank.R5>(convInfo.filterShape, 'float32');\n    const dwValues = dw.values;\n    const [dwS0, dwS1, dwS2, dwS3] = dw.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const [xS0, xS1, xS2, xS3] = x.strides;\n\n    const frontPad = convInfo.padInfo.front;\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n\n    for (let wF = 0; wF < filterDepth; ++wF) {\n      const yFMin = Math.max(0, Math.ceil((frontPad - wF) / strideDepth));\n      const yFMax = Math.min(\n          convInfo.outDepth, (convInfo.inDepth + frontPad - wF) / strideDepth);\n      const wOffset1 = wF * dwS0;\n\n      for (let wR = 0; wR < filterHeight; ++wR) {\n        const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n        const yRMax = Math.min(\n            convInfo.outHeight,\n            (convInfo.inHeight + topPad - wR) / strideHeight);\n        const wOffset2 = wR * dwS1 + wOffset1;\n\n        for (let wC = 0; wC < filterWidth; ++wC) {\n          const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n          const yCMax = Math.min(\n              convInfo.outWidth,\n              (convInfo.inWidth + leftPad - wC) / strideWidth);\n          const wOffset3 = wC * dwS2 + wOffset2;\n\n          for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n            const wOffset4 = d1 * dwS3 + wOffset3;\n\n            for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n              let dotProd = 0;\n              for (let b = 0; b < convInfo.batchSize; ++b) {\n                const xOffset1 = b * xS0;\n                const yOffset1 = b * dyS0;\n\n                for (let yF = yFMin; yF < yFMax; ++yF) {\n                  const xF = wF + yF * strideDepth - frontPad;\n                  const xOffset2 = xF * xS1 + xOffset1;\n                  const yOffset2 = yF * dyS1 + yOffset1;\n\n                  for (let yR = yRMin; yR < yRMax; ++yR) {\n                    const xR = wR + yR * strideHeight - topPad;\n                    const xOffset3 = xR * xS2 + xOffset2;\n                    const yOffset3 = yR * dyS2 + yOffset2;\n\n                    for (let yC = yCMin; yC < yCMax; ++yC) {\n                      const xC = wC + yC * strideWidth - leftPad;\n                      const xOffset4 = xC * xS3 + xOffset3;\n                      const yOffset4 = yC * dyS3 + yOffset3;\n\n                      dotProd +=\n                          xValues[xOffset4 + d1] * dyValues[yOffset4 + d2];\n                    }\n                  }\n                }\n              }\n              dwValues[wOffset4 + d2] = dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dw.toTensor();\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.depthwiseConv2D(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedDepthwiseConv2D is\n      // modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, filter], 'depthwiseConv2D');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * y.strides[1];\n        const xRCorner = yR * convInfo.strideHeight - padLeft;\n        for (let wR = 0; wR < filterHeight; ++wR) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * x.strides[1];\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * y.strides[2];\n            const xCCorner = yC * convInfo.strideWidth - padTop;\n            for (let wC = 0; wC < filterWidth; ++wC) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * convInfo.inChannels;\n              let yOffset4 = yOffset3;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1];\n                for (let q = 0; q < chMul; ++q) {\n                  yVals[yOffset4 + q] += xVal * wVals[wOffset3 + q];\n                }\n                yOffset4 += chMul;\n                wOffset3 += chMul;\n              }\n            }\n          }\n        }\n      }\n    }\n\n    return y.toTensor() as Tensor4D;\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'depthwiseConv2DDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n    const chMul = outChannels / inChannels;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset = dyS0 * b + dyS1 * yR + dyS2 * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let dm = 0; dm < chMul; ++dm) {\n                  const d2 = d1 * chMul + dm;\n                  const pixel = dyValues[dyOffset + d2];\n                  const weight = fltValues[fltOffset + dm];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            dxValues[dxS0 * b + dxS1 * xR + dxS2 * xC + d1] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, dy], 'depthwiseConv2DDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n          const d1 = Math.trunc(d2 / chMul);\n          const dm = d2 % chMul;\n\n          let dotProd = 0;\n          for (let b = 0; b < convInfo.batchSize; ++b) {\n            for (let yR = yRMin; yR < yRMax; ++yR) {\n              const xR = wR + yR * strideHeight - topPad;\n              for (let yC = yCMin; yC < yCMax; ++yC) {\n                const xC = wC + yC * strideWidth - leftPad;\n                dotProd += xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n              }\n            }\n          }\n          dW.set(dotProd, wR, wC, d1, dm);\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    assertNotComplex(x, 'tile');\n    return tile(this.bufferSync(x), reps) as T;\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    assertNotComplex([x, indices], 'gather');\n\n    const newShape: number[] = x.shape.slice();\n    const indicesValues = this.readSync(indices.dataId) as TypedArray;\n    newShape[axis] = indicesValues.length;\n    const result = tf.buffer(newShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < result.size; ++i) {\n      const newLoc = result.indexToLoc(i);\n\n      const originalLoc: number[] = newLoc.slice();\n      originalLoc[axis] = indicesValues[newLoc[axis]];\n\n      const originalIndex = xBuf.locToIndex(originalLoc);\n      result.values[i] = xBuf.values[originalIndex];\n    }\n    return result.toTensor() as T;\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    assertNotComplex([x], 'batchToSpaceND');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return tf.transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  private pool3d(\n      x: Tensor5D, convInfo: backend_util.Conv3DInfo,\n      poolType: 'max'|'avg'): Tensor5D {\n    assertNotComplex(x, 'pool3d');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const initialValue =\n        (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                              Number.POSITIVE_INFINITY);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = tf.buffer(convInfo.outShape, x.dtype);\n    const outputVals = output.values;\n\n    const outputBatchStrides = convInfo.outShape[1] * convInfo.outShape[2] *\n        convInfo.outShape[3] * convInfo.outShape[4];\n    const outputDepthStrides =\n        convInfo.outShape[2] * convInfo.outShape[3] * convInfo.outShape[4];\n    const outputRowStrides = convInfo.outShape[3] * convInfo.outShape[4];\n    const outputColStrides = convInfo.outShape[4];\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      const outputBatchOffset = batch * outputBatchStrides;\n      const inputBatchOffset = batch * x.strides[0];\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          const outputDepthOffset =\n              outputBatchOffset + yDepth * outputDepthStrides;\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            const outputRowOffset = outputDepthOffset + yRow * outputRowStrides;\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n              // Shader code begins\n              const outputColOffset = outputRowOffset + yCol * outputColStrides;\n              let minMaxValue = initialValue;\n              let avgValue = 0;\n              let count = 0;\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const xDepthOffset = inputBatchOffset + xDepth * x.strides[1];\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const xRowOffset = xDepthOffset + xRow * x.strides[2];\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const xColOffset = xRowOffset + xCol * x.strides[3];\n                    const pixel = xValues[xColOffset + channel];\n                    if ((poolType === 'max' && pixel > minMaxValue)) {\n                      minMaxValue = pixel;\n                    } else if (poolType === 'avg') {\n                      avgValue += pixel;\n                      count++;\n                    }\n                    if (isNaN(minMaxValue)) {\n                      break;\n                    }\n                  }\n                  if (isNaN(minMaxValue)) {\n                    break;\n                  }\n                }\n                if (isNaN(minMaxValue)) {\n                  break;\n                }\n              }\n              const outputOffset = outputColOffset + channel;\n              outputVals[outputOffset] =\n                  poolType === 'avg' ? avgValue / count : minMaxValue;\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor5D;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'avgPool3d');\n\n    return this.pool3d(x, convInfo, 'avg').toFloat();\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([dy, x], 'avgPool3dBackprop');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins.\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel;\n                  }\n                }\n              }\n              dx.set(\n                  dotProd * avgMultiplier, batch, dxDepth, dxRow, dxCol,\n                  channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'maxPool3d');\n\n    return this.pool3d(x, convInfo, 'max').toFloat();\n  }\n\n  private maxPool3dPositions(x: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const maxPositions = tf.buffer(convInfo.outShape, 'int32');\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const xBuf = this.bufferSync(x);\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n\n              // Shader code begins\n              let maxValue = Number.NEGATIVE_INFINITY;\n              let maxPosition = -1;\n\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const wDepth = xDepth - xDepthCorner;\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const wRow = xRow - xRowCorner;\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const wCol = xCol - xColCorner;\n                    const pixel = xBuf.get(batch, xDepth, xRow, xCol, channel);\n                    if (pixel >= maxValue) {\n                      maxValue = pixel;\n                      maxPosition = wDepth * effectiveFilterHeight *\n                              effectiveFilterWidth +\n                          wRow * effectiveFilterHeight + wCol;\n                    }\n                  }\n                }\n              }\n\n              maxPositions.set(maxPosition, batch, yDepth, yRow, yCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return maxPositions.toTensor() as Tensor5D;\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([x, y], 'maxPool3dBackprop');\n\n    const maxPositions = this.maxPool3dPositions(x, convInfo);\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const maxPosBuf = this.bufferSync(maxPositions);\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const maxPos = effectiveFilterDepth *\n                            effectiveFilterHeight * effectiveFilterWidth -\n                        1 -\n                        maxPosBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    const curPos =\n                        wDepth * effectiveFilterHeight * effectiveFilterWidth +\n                        wRow * effectiveFilterWidth + wCol;\n\n                    const mask = maxPos === curPos ? 1 : 0;\n                    if (mask === 0) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel * mask;\n                  }\n                }\n              }\n              dx.set(dotProd, batch, dxDepth, dxRow, dxCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeBilinear');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result = new Float32Array(\n        util.sizeFromShape([batch, newHeight, newWidth, numChannels]));\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n    let outputIdx = 0;\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n    for (let b = 0; b < batch; b++) {\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceRowFloor = Math.floor(sourceFracRow);\n        const rowFrac = sourceFracRow - sourceRowFloor;\n        const sourceRowCeil = Math.min(oldHeight - 1, Math.ceil(sourceFracRow));\n        const topRowOffset = b * x.strides[0] + sourceRowFloor * x.strides[1];\n        const botRowOffset = b * x.strides[0] + sourceRowCeil * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceColFloor = Math.floor(sourceFracCol);\n          const colFrac = sourceFracCol - sourceColFloor;\n          const sourceColCeil =\n              Math.min(oldWidth - 1, Math.ceil(sourceFracCol));\n          const topLeftOffest = topRowOffset + sourceColFloor * x.strides[2];\n          const botLeftOffset = botRowOffset + sourceColFloor * x.strides[2];\n          const topRightOffset = topRowOffset + sourceColCeil * x.strides[2];\n          const botRightOffest = botRowOffset + sourceColCeil * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n\n            // Compute the fractional index of the source.\n            const topLeft = xValues[topLeftOffest + d];\n            const bottomLeft = xValues[botLeftOffset + d];\n            const topRight = xValues[topRightOffset + d];\n            const bottomRight = xValues[botRightOffest + d];\n\n            const top = topLeft + (topRight - topLeft) * colFrac;\n            const bottom = bottomLeft + (bottomRight - bottomLeft) * colFrac;\n            const newValue = top + (bottom - top) * rowFrac;\n\n            result[outputIdx++] = newValue;\n          }\n        }\n      }\n    }\n    return tf.tensor(result, [batch, newHeight, newWidth, numChannels]);\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeBilinearBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass and add the\n    // corresponding coefficient from dy to the gradient (with some\n    // interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/3039375c86a5bbc9610c7725dcaa95d635f87ba2/tensorflow/core/kernels/resize_bilinear_op.cc#L275\n\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    let offset = 0;\n    for (let b = 0; b < batch; b++) {\n      const bOffset = b * x.strides[0];\n      for (let r = 0; r < yHeight; r++) {\n        const dxR = r * heightScale;\n        const topDxRIndex = Math.floor(dxR);\n        const bottomDxRIndex = Math.min(Math.ceil(dxR), xHeight - 1);\n\n        const topDxROffset = bOffset + topDxRIndex * x.strides[1];\n        const bottomDxROffset = bOffset + bottomDxRIndex * x.strides[1];\n\n        const dxRLerp = dxR - topDxRIndex;\n        const inverseDxRLerp = 1.0 - dxRLerp;\n        for (let c = 0; c < yWidth; c++) {\n          const dxC = c * widthScale;\n          const leftDxCIndex = Math.floor(dxC);\n          const rightDxCIndex = Math.min(Math.ceil(dxC), xWidth - 1);\n          const dxCLerp = dxC - leftDxCIndex;\n          const inverseDxCLerp = 1.0 - dxCLerp;\n\n          const topLeftRCOffset = topDxROffset + leftDxCIndex * x.strides[2];\n          const topRightRCOffset = topDxROffset + rightDxCIndex * x.strides[2];\n          const bottomLeftRCOffset =\n              bottomDxROffset + leftDxCIndex * x.strides[2];\n          const bottomRightRCOffset =\n              bottomDxROffset + rightDxCIndex * x.strides[2];\n\n          const inverseDxRLerpTimesInverseDxCLerp =\n              inverseDxRLerp * inverseDxCLerp;\n          const inverseDxRLerpTimesDxCLerp = inverseDxRLerp * dxCLerp;\n          const dxRLerpTimesInverseDxCLerp = dxRLerp * inverseDxCLerp;\n          const dxRLerpTimesDxCLerp = dxRLerp * dxCLerp;\n          for (let d = 0; d < depth; d++) {\n            const dyVal = dyValues[offset++];\n            output[topLeftRCOffset + d] +=\n                dyVal * inverseDxRLerpTimesInverseDxCLerp;\n            output[topRightRCOffset + d] += dyVal * inverseDxRLerpTimesDxCLerp;\n            output[bottomLeftRCOffset + d] +=\n                dyVal * dxRLerpTimesInverseDxCLerp;\n            output[bottomRightRCOffset + d] += dyVal * dxRLerpTimesDxCLerp;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, [batch, xWidth, xHeight, depth], x.dtype);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeNearestNeighbor');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = new Float32Array(batch * newHeight * newWidth * numChannels);\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n\n    let outputOffset = 0;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceNearestRow = Math.min(\n            oldHeight - 1,\n            alignCorners ? Math.round(sourceFracRow) :\n                           Math.floor(sourceFracRow));\n        const rowOffset = batchOffset + sourceNearestRow * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceNearestCol = Math.min(\n              oldWidth - 1,\n              alignCorners ? Math.round(sourceFracCol) :\n                             Math.floor(sourceFracCol));\n          const colOffset = rowOffset + sourceNearestCol * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n            // Compute the fractional index of the source.\n            const newVal = xValues[colOffset + d];\n            output[outputOffset++] = newVal;\n          }\n        }\n      }\n    }\n    return tf.tensor(\n        output, [batch, newHeight, newWidth, numChannels], x.dtype);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeNearestNeighborBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    // Loop over the output space.\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < xHeight; r++) {\n        const rowOffset = batchOffset + r * x.strides[1];\n\n        // Compute bounds for where in dy we will look\n        const startRLerp = Math.floor(r * invHeightScale);\n        const startDyR = Math.floor(startRLerp - (winHeight / 2));\n        for (let c = 0; c < xWidth; c++) {\n          const colOffset = rowOffset + c * x.strides[2];\n\n          // Compute bounds for where in dy we will look\n          const startCLerp = Math.floor(c * invWidthScale);\n          const startDyC = Math.floor(startCLerp - (winWidth / 2));\n\n          for (let d = 0; d < depth; d++) {\n            let accum = 0;\n            // loop over dy\n\n            for (let dyRIndex = 0; dyRIndex < winHeight; dyRIndex++) {\n              const dyR = dyRIndex + startDyR;\n              // Guard against the window exceeding the bounds of dy\n              if (dyR < 0 || dyR >= yHeight) {\n                continue;\n              }\n\n              const dyROffset = batchOffset + dyR * dy.strides[1];\n              const sourceFracRow = dyR * heightScale;\n              const sourceNearestRow = Math.min(\n                  xHeight - 1,\n                  alignCorners ? Math.round(sourceFracRow) :\n                                 Math.floor(sourceFracRow));\n              if (r !== sourceNearestRow) {\n                continue;\n              }\n              for (let dyCIndex = 0; dyCIndex < winWidth; dyCIndex++) {\n                const dyC = dyCIndex + startDyC;\n                // Guard against the window exceeding the bounds of dy\n                if (dyC < 0 || dyC >= yWidth) {\n                  continue;\n                }\n\n                const dyCOffset = dyROffset + dyC * dy.strides[2];\n                const sourceFracCol = dyC * widthScale;\n                const sourceNearestCol = Math.min(\n                    xWidth - 1,\n                    alignCorners ? Math.round(sourceFracCol) :\n                                   Math.floor(sourceFracCol));\n\n                if (c === sourceNearestCol) {\n                  accum += dyValues[dyCOffset + d];\n                }\n              }\n            }\n            output[colOffset + d] = accum;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, x.shape, x.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(x, 'localResponseNormalization4D');\n\n    const channels = x.shape[3];\n    const maxD = channels - 1;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const size = x.size;\n    const result = new Float32Array(size);\n\n    function sumAcrossChannels(offset: number) {\n      const currentChannel = offset % channels;\n      let beginSumOffset =\n          offset - currentChannel + Math.max(0, currentChannel - depthRadius);\n      const endSumOffset = offset - currentChannel +\n          Math.min(currentChannel + depthRadius, maxD);\n\n      let sum = 0.0;\n      for (; beginSumOffset <= endSumOffset; beginSumOffset++) {\n        const z = xValues[beginSumOffset];\n        sum += z * z;\n      }\n      return sum;\n    }\n\n    for (let offset = 0; offset < size; offset++) {\n      const sum = sumAcrossChannels(offset);\n      const val = xValues[offset] * Math.pow(bias + alpha * sum, -beta);\n      result[offset] = val;\n    }\n\n    return tf.tensor4d(result, x.shape);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(dy, 'LRNGrad');\n    const channels = dy.shape[3];\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const inputImageValues = this.readSync(inputImage.dataId) as TypedArray;\n    const outputImageValues = this.readSync(outputImage.dataId) as TypedArray;\n    const result = new Float32Array(dy.size);\n    const size = dy.size;\n\n    for (let offset = 0; offset < size; offset++) {\n      const currentChannel = offset % channels;\n      const depthBegin =\n          (offset - currentChannel) + Math.max(0, currentChannel - depthRadius);\n      const depthEnd = (offset - currentChannel) +\n          Math.min(channels, currentChannel + depthRadius + 1);\n\n      let norm = 0;\n      for (let k = depthBegin; k < depthEnd; k++) {\n        norm += Math.pow(inputImageValues[k], 2);\n      }\n      norm = alpha * norm + bias;\n\n      for (let k = depthBegin; k < depthEnd; k++) {\n        let dyi = -2 * alpha * beta * inputImageValues[k] *\n            outputImageValues[offset] / norm;\n        if (offset === k) {\n          dyi += Math.pow(norm, -beta);\n        }\n        dyi *= dyValues[offset];\n        result[k] += dyi;\n      }\n    }\n    return tf.tensor4d(result, dy.shape);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    assertNotComplex(logits, 'multinomial');\n\n    const probabilities = normalized ? logits : tf.softmax(logits);\n    const batchSize = probabilities.shape[0];\n    const numEvents = probabilities.shape[1];\n    const res = tf.zeros<Rank.R2>([batchSize, numSamples], 'int32');\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const probVals = this.readSync(probabilities.dataId) as TypedArray;\n\n    for (let b = 0; b < batchSize; ++b) {\n      const offset = b * numEvents;\n      // The cdf won't include the last event. It will be implicit if no other\n      // event happened.\n      const cdf = new Float32Array(numEvents - 1);\n      cdf[0] = probVals[offset];\n      for (let event = 1; event < cdf.length; ++event) {\n        cdf[event] = cdf[event - 1] + probVals[offset + event];\n      }\n\n      const random = seedrandom.alea(seed.toString());\n      const outOffset = b * numSamples;\n      for (let sampleId = 0; sampleId < numSamples; ++sampleId) {\n        const r = random();\n\n        // Assume last event happened by default.\n        resVals[outOffset + sampleId] = cdf.length;\n\n        for (let event = 0; event < cdf.length; event++) {\n          if (r < cdf[event]) {\n            resVals[outOffset + sampleId] = event;\n            break;\n          }\n        }\n      }\n    }\n    return res;\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    assertNotComplex(indices, 'oneHot');\n\n    const res = new Float32Array(indices.size * depth);\n    res.fill(offValue);\n    const indicesVal = this.readSync(indices.dataId) as TypedArray;\n\n    for (let event = 0; event < indices.size; ++event) {\n      if (indicesVal[event] >= 0 && indicesVal[event] < depth) {\n        res[event * depth + indicesVal[event]] = onValue;\n      }\n    }\n    return tf.tensor2d(res, [indices.size, depth], 'int32');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold: number): Tensor1D {\n    assertNotComplex(boxes, 'nonMaxSuppression');\n\n    const boxesVals = this.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = this.readSync(scores.dataId) as TypedArray;\n    return nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        dataFormat === 'NHWC',\n        () => `Only NHWC dataFormat supported on CPU for depthToSpace. Got ${\n            dataFormat}`);\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = x.shape[1];\n    const inputWidth = x.shape[2];\n    const inputDepth = x.shape[3];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result =\n        new Float32Array(batchSize * outputHeight * outputWidth * outputDepth);\n\n    let outputIdx = 0;\n    for (let b = 0; b < batchSize; ++b) {\n      for (let h = 0; h < outputHeight; ++h) {\n        const inH = Math.floor(h / blockSize);\n        const offsetH = (h % blockSize);\n        for (let w = 0; w < outputWidth; ++w) {\n          const inW = Math.floor(w / blockSize);\n          const offsetW = (w % blockSize);\n          const offsetD = (offsetH * blockSize + offsetW) * outputDepth;\n          for (let d = 0; d < outputDepth; ++d) {\n            const inD = d + offsetD;\n            const inputIdx =\n                inD + inputDepth * (inW + inputWidth * (inH + inputHeight * b));\n            result[outputIdx++] = xValues[inputIdx];\n          }\n        }\n      }\n    }\n    return tf.tensor4d(\n        result, [batchSize, outputHeight, outputWidth, outputDepth]);\n  }\n\n  private broadcastedBinaryOp(\n      a: Tensor, b: Tensor, dtype: DataType,\n      op: (a: number, b: number) => number): Tensor {\n    const newShape = backend_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const result = tf.buffer(newShape, dtype);\n    const aVals = this.readSync(a.dataId) as TypedArray;\n    const bVals = this.readSync(b.dataId) as TypedArray;\n    const aBroadcastDims = backend_util.getBroadcastDims(a.shape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(b.shape, newShape);\n\n    const resVals = result.values;\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resVals.length; ++i) {\n        resVals[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      const aBuf = this.bufferSync(a);\n      const bBuf = this.bufferSync(b);\n      for (let i = 0; i < resVals.length; ++i) {\n        const loc = result.indexToLoc(i);\n\n        const aLoc = loc.slice(-a.rank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = aBuf.locToIndex(aLoc);\n\n        const bLoc = loc.slice(-b.rank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = bBuf.locToIndex(bLoc);\n\n        resVals[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n    return result.toTensor();\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  dispose() {}\n\n  floatPrecision(): 16|32 {\n    return 32;\n  }\n\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return super.epsilon();\n  }\n\n  cropAndResize(\n      images: Tensor4D,\n      boxes: Tensor2D,\n      boxIndex: Tensor1D,\n      cropSize: [number, number],\n      method: string,\n      extrapolationValue: number,\n  ) {\n    const [batch, imageHeight, imageWidth, numChannels] = images.shape;\n    const numBoxes = boxes.shape[0];\n\n    const [cropHeight, cropWidth] = cropSize;\n    const output =\n        tf.buffer([numBoxes, cropHeight, cropWidth, numChannels], 'float32');\n\n    const boxVals = this.readSync(boxes.dataId) as TypedArray;\n    const boxIndVals = this.readSync(boxIndex.dataId) as TypedArray;\n    const imageVals = this.readSync(images.dataId) as TypedArray;\n\n    const inStride = images.strides;   // to calculate flat indexes into image\n    const outStride = output.strides;  // to calculate flat indexes into output\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op.cc\n    for (let b = 0; b < numBoxes; b++) {\n      const startInd = b * 4;\n      const y1 = boxVals[startInd];\n      const x1 = boxVals[startInd + 1];\n      const y2 = boxVals[startInd + 2];\n      const x2 = boxVals[startInd + 3];\n\n      const bInd: number = boxIndVals[b];\n      if (bInd >= batch) {\n        continue;\n      }\n\n      const heightScale = (cropHeight > 1) ?\n          (y2 - y1) * (imageHeight - 1) / (cropHeight - 1) :\n          0;\n      const widthScale =\n          (cropWidth > 1) ? (x2 - x1) * (imageWidth - 1) / (cropWidth - 1) : 0;\n\n      for (let y = 0; y < cropHeight; y++) {\n        const yInd: number = (cropHeight > 1) ?\n            y1 * (imageHeight - 1) + y * (heightScale) :\n            0.5 * (y1 + y2) * (imageHeight - 1);\n\n        if (yInd < 0 || yInd > imageHeight - 1) {\n          for (let x = 0; x < cropWidth; x++) {\n            for (let c = 0; c < numChannels; c++) {\n              const ind =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = extrapolationValue;\n            }\n          }\n          continue;\n        }\n\n        if (method === 'bilinear') {\n          const topInd = Math.floor(yInd);\n          const bottomInd = Math.ceil(yInd);\n          const yLerp = yInd - topInd;\n\n          for (let x = 0; x < cropWidth; x++) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const leftInd = Math.floor(xInd);\n            const rightInd = Math.ceil(xInd);\n            const xLerp = xInd - leftInd;\n\n            for (let c = 0; c < numChannels; c++) {\n              let ind = c + leftInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topRight = imageVals[ind];\n\n              ind = c + leftInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomRight = imageVals[ind];\n\n              const top = topLeft + (topRight - topLeft) * xLerp;\n              const bottom = bottomLeft + (bottomRight - bottomLeft) * xLerp;\n\n              ind = c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = top + ((bottom - top) * yLerp);\n            }\n          }\n        } else {  // method == \"nearest\"\n          for (let x = 0; x < cropWidth; ++x) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const closestX = Math.round(xInd);\n            const closestY = Math.round(yInd);\n            for (let c = 0; c < numChannels; c++) {\n              const inInd = c + closestX * inStride[2] +\n                  closestY * inStride[1] + bInd * inStride[0];\n              const outInd =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[outInd] = imageVals[inInd];\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor4D;\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n    const sumDupeIndices = false;\n    return this.scatter(\n        sparseIndices, sparseValues, outputShape, outputSize, sliceSize,\n        numUpdates, sliceRank, strides, defaultValue, sumDupeIndices);\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n    if (numSlices === 0) {\n      return tf.tensor([], resultShape, x.dtype);\n    }\n\n    const buffer = new TensorBuffer([numSlices, sliceSize], x.dtype);\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const xData = this.readSync(x.dataId) as TypedArray;\n\n    for (let i = 0; i < numSlices; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        flattenIndex += dim * strides[j];\n        index.push(dim);\n      }\n      if (flattenIndex < 0 || flattenIndex >= x.size / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${x.shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        buffer.values[i * sliceSize + k] = xData[flattenIndex * sliceSize + k];\n      }\n    }\n    return buffer.toTensor().reshape(resultShape);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n    const defaultValue = tf.scalar(0);\n    const sumDupeIndices = true;\n    return this.scatter(\n        indices, updates, shape, outputSize, sliceSize, numUpdates, sliceRank,\n        strides, defaultValue, sumDupeIndices);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n    const values =\n        util.getArrayFromDType(dtype, util.sizeFromShape(shape)) as TypedArray;\n    values.fill(value as number);\n    return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported for string tensors');\n    } else {\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    const values = util.getArrayFromDType(\n                       x.dtype, util.sizeFromShape(x.shape)) as TypedArray;\n    return this.makeOutput(values, x.shape, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  private scatter<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R], outputSize: number,\n      sliceSize: number, numUpdates: number, sliceRank: number,\n      strides: number[], defaultValue: Scalar,\n      sumDupeIndices: boolean): Tensor<R> {\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const updatesData = this.readSync(updates.dataId) as TypedArray;\n\n    if (outputSize === 0) {\n      return tf.tensor([], shape, updates.dtype);\n    }\n\n    const buffer = new TensorBuffer(flattenShape, updates.dtype as 'float32');\n    buffer.values.fill((this.readSync(defaultValue.dataId) as TypedArray)[0]);\n\n    for (let i = 0; i < numUpdates; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        index.push(dim);\n        flattenIndex += dim * strides[j];\n      }\n\n      if (flattenIndex < 0 || flattenIndex >= outputSize / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        if (sumDupeIndices) {\n          buffer.values[flattenIndex * sliceSize + k] +=\n              updatesData[i * sliceSize + k];\n        } else {\n          buffer.values[flattenIndex * sliceSize + k] = updates.rank === 0 ?\n              updatesData[0] :\n              updatesData[i * sliceSize + k];\n        }\n      }\n    }\n    return buffer.toTensor().reshape(shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs, AbsInputs, KernelConfig, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function simpleAbsImpl(vals: TypedArray): Float32Array {\n  const resultValues = new Float32Array(vals.length);\n  for (let i = 0; i < vals.length; ++i) {\n    resultValues[i] = Math.abs(vals[i]);\n  }\n  return resultValues;\n}\n\nexport const absKernelFunc =\n    (args: {inputs: AbsInputs, backend: MathBackendCPU}) => {\n      const {x} = args.inputs;\n      const cpuBackend = args.backend;\n      let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n      if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n        resultValues = simpleAbsImpl(values);\n      } else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals =\n            cpuBackend.data.get(real.dataId).values as Float32Array;\n        const imagVals =\n            cpuBackend.data.get(imag.dataId).values as Float32Array;\n        for (let i = 0; i < realVals.length; i++) {\n          const real = realVals[i];\n          const imag = imagVals[i];\n          resultValues[i] = Math.hypot(real, imag);\n        }\n      }\n      return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n    };\n\nexport const absConfig: KernelConfig = {\n  kernelName: Abs,\n  backendName: 'cpu',\n  kernelFunc: absKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleBinaryKernelImpl, SimpleBinaryOperation} from './binary_types';\n\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op: SimpleBinaryOperation):\n    SimpleBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aVals: TypedArray,\n          bVals: TypedArray, dtype: DataType): [TypedArray, number[]] => {\n    const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    const resultRank = newShape.length;\n    const resultStrides = util.computeStrides(newShape);\n    const resultSize = util.sizeFromShape(newShape);\n\n    const result =\n        util.getTypedArrayFromDType(dtype as NumericDataType, resultSize);\n\n    const aRank = aShape.length;\n    const bRank = bShape.length;\n\n    const aStrides = util.computeStrides(aShape);\n    const bStrides = util.computeStrides(bShape);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < result.length; ++i) {\n        result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      for (let i = 0; i < result.length; ++i) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        result[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n\n    return [result, newShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Complex, ComplexInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function complex(args: {inputs: ComplexInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {real, imag} = inputs;\n\n  const realVals = backend.data.get(real.dataId).values as TypedArray;\n  const imagVals = backend.data.get(imag.dataId).values as TypedArray;\n\n  const complexInfo = backend.makeTensorInfo(real.shape, 'complex64');\n\n  const complex = backend.data.get(complexInfo.dataId);\n\n  // The complex tensor owns the underlying real and imag tensorInfos, only the\n  // complex tensor tracks refCount, when complexData is disposed the\n  // underlying tensorData will be disposed.\n  complex.complexTensorInfos = {\n    real: backend.makeTensorInfo(real.shape, 'float32', realVals),\n    imag: backend.makeTensorInfo(imag.shape, 'float32', imagVals)\n  };\n\n  return complexInfo;\n}\n\nexport const complexConfig: KernelConfig = {\n  kernelName: Complex,\n  backendName: 'cpu',\n  kernelFunc: complex as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendCPU}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'cpu',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Real, RealInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function real(args: {inputs: RealInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const real = backend.data.get(input.dataId).complexTensorInfos.real;\n  const realVal = backend.data.get(real.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the real value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(real.shape, real.dtype, realVal);\n}\n\nexport const realConfig: KernelConfig = {\n  kernelName: Real,\n  backendName: 'cpu',\n  kernelFunc: real as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport * as tf from '@tensorflow/tfjs-core';\nimport {Cast, CastAttrs, CastInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\n\nimport {complex} from './Complex';\nimport {identity} from './Identity';\nimport {real} from './Real';\n\nexport function cast(\n    args: {inputs: CastInputs, backend: MathBackendCPU, attrs: CastAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {dtype} = attrs;\n\n  // Casting to complex64.\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return identity({inputs: {x}, backend});\n    }\n\n    // TODO(lina128): Import kernel function once zeros is modularized.\n    const zerosTensor = tf.zeros(x.shape);\n    const floatX = cast({inputs: {x}, backend, attrs: {dtype: 'float32'}});\n\n    const result =\n        complex({inputs: {real: floatX, imag: zerosTensor}, backend});\n\n    zerosTensor.dispose();\n    backend.disposeIntermediateTensorInfo(floatX);\n\n    return result;\n  }\n\n  // Casting from complex64\n  if (x.dtype === 'complex64') {\n    const realPart = real({inputs: {input: x}, backend});\n    const result = cast({inputs: {x: realPart}, backend, attrs: {dtype}});\n\n    backend.disposeIntermediateTensorInfo(realPart);\n\n    return result;\n  }\n\n  if (!util.hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    const result = identity({inputs: {x}, backend});\n    return {dataId: result.dataId, shape: result.shape, dtype};\n  }\n\n  if (dtype === 'int32') {\n    const values = backend.data.get(x.dataId).values as TypedArray;\n    const resultValues = Int32Array.from(values);\n    return backend.makeTensorInfo(x.shape, 'int32', resultValues);\n  }\n\n  if (dtype === 'bool') {\n    // This is essentially the result of notEqual(x, 0). We avoid using\n    // kernel notEqual to avoid circular dependency, i.e. binary_utils ->\n    // cast -> notEqual -> binary_utils.\n    const xVals = backend.data.get(x.dataId).values as TypedArray;\n    const zero = util.toTypedArray([0], x.dtype);\n\n    const [resultData, resultShape] = createSimpleBinaryKernelImpl(\n        (a, b) => (a !== b) ? 1 : 0)(x.shape, [], xVals, zero, 'bool');\n\n    return backend.makeTensorInfo(resultShape, 'bool', resultData);\n  }\n\n  throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n}\n\nexport const castConfig: KernelConfig = {\n  kernelName: Cast,\n  backendName: 'cpu',\n  kernelFunc: cast as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, BinaryInputs, DataType, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {cast} from '../kernels/Cast';\nimport {complex} from '../kernels/Complex';\n\nimport {ComplexBinaryKernelImpl, ComplexBinaryOperation, SimpleBinaryKernelImpl} from './binary_types';\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param name Kernel name.\n * @param binaryKernelImpl A `SimpleBinaryKernelImpl` for the kernel.\n * @param binaryKernelComplexImpl Optional. If exists, represents a\n *     `ComplexBinaryKernelImpl` for the kernel, will be used when input dtype\n *     is `complex64`.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    name: string, simpleImpl: SimpleBinaryKernelImpl,\n    complexImpl?: ComplexBinaryKernelImpl, dtype?: DataType): KernelFunc {\n  if (complexImpl == null) {\n    return ({inputs, backend}) => {\n      const {a, b} = inputs as BinaryInputs;\n      const cpuBackend = backend as MathBackendCPU;\n\n      assertNotComplex([a, b], name);\n\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    };\n  }\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    if (a.dtype === 'complex64' || b.dtype === 'complex64') {\n      const $aComplex = cast(\n          {inputs: {x: a}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $aComplexVals = cpuBackend.data.get($aComplex.dataId);\n\n      const aReal = $aComplexVals.complexTensorInfos.real;\n      const aImag = $aComplexVals.complexTensorInfos.imag;\n\n      const aRealVals =\n          cpuBackend.data.get(aReal.dataId).values as Float32Array;\n      const aImagVals =\n          cpuBackend.data.get(aImag.dataId).values as Float32Array;\n\n      const $bComplex = cast(\n          {inputs: {x: b}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $bComplexVals = cpuBackend.data.get($bComplex.dataId);\n\n      const bReal = $bComplexVals.complexTensorInfos.real;\n      const bImag = $bComplexVals.complexTensorInfos.imag;\n\n      const bRealVals =\n          cpuBackend.data.get(bReal.dataId).values as Float32Array;\n      const bImagVals =\n          cpuBackend.data.get(bImag.dataId).values as Float32Array;\n\n      const [resultRealData, resultImagData, resultShape] = complexImpl(\n          a.shape, b.shape, aRealVals, aImagVals, bRealVals, bImagVals);\n\n      const resultReal =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultRealData);\n\n      const resultImag =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultImagData);\n\n      const result = complex(\n          {inputs: {real: resultReal, imag: resultImag}, backend: cpuBackend});\n\n      cpuBackend.disposeIntermediateTensorInfo($aComplex);\n      cpuBackend.disposeIntermediateTensorInfo($bComplex);\n      cpuBackend.disposeIntermediateTensorInfo(resultReal);\n      cpuBackend.disposeIntermediateTensorInfo(resultImag);\n\n      return result;\n    } else {\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    }\n  };\n}\n\n/**\n * Template that creates the complex type implementation for binary ops.\n * Supports broadcast.\n */\nexport function createComplexBinaryKernelImpl(op: ComplexBinaryOperation):\n    ComplexBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aRealVals: Float32Array,\n          aImagVals: Float32Array, bRealVals: Float32Array,\n          bImagVals: Float32Array): [TypedArray, TypedArray, number[]] => {\n    const resultShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const resultSize = util.sizeFromShape(resultShape);\n    const resultRank = resultShape.length;\n    const resultStrides = util.computeStrides(resultShape);\n\n    const resultRealVals = util.getTypedArrayFromDType('float32', resultSize);\n    const resultImagVals = util.getTypedArrayFromDType('float32', resultSize);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, resultShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, resultShape);\n\n    const aVals = backend_util.mergeRealAndImagArrays(aRealVals, aImagVals);\n    const bVals = backend_util.mergeRealAndImagArrays(bRealVals, bImagVals);\n\n    const aRank = aShape.length;\n    const aStrides = util.computeStrides(aShape);\n\n    const bRank = bShape.length;\n    const bStrides = util.computeStrides(bShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const aIdx = i % aVals.length;\n        const bIdx = i % bVals.length;\n\n        const result =\n            op(aVals[aIdx * 2], aVals[aIdx * 2 + 1], bVals[bIdx * 2],\n               bVals[bIdx * 2 + 1]);\n\n        resultRealVals[i] = result.real;\n        resultImagVals[i] = result.imag;\n      }\n    } else {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        const opResult =\n            op(aVals[aIndex * 2], aVals[aIndex * 2 + 1], bVals[bIndex * 2],\n               bVals[bIndex * 2 + 1]);\n\n        resultRealVals[i] = opResult.real;\n        resultImagVals[i] = opResult.imag;\n      }\n    }\n    return [resultRealVals, resultImagVals, resultShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Add, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal + bReal, imag: aImag + bImag};\n    }));\n\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\n\nexport const addConfig: KernelConfig = {\n  kernelName: Add,\n  backendName: 'cpu',\n  kernelFunc: add\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NumericDataType, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op: SimpleUnaryOperation):\n    SimpleUnaryImpl {\n  return (values, dtype, attrs) => {\n    const newValues =\n        util.getTypedArrayFromDType(dtype as NumericDataType, values.length);\n    for (let i = 0; i < values.length; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return newValues;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelFunc, TypedArray, UnaryInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param name Kernel name.\n * @param op A `SimpleUnaryOperation` for the kernel.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFunc(\n    name: string, op: SimpleUnaryOperation, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xSize = util.sizeFromShape(x.shape);\n    const $dtype = dtype || x.dtype;\n    const newValues = util.getArrayFromDType($dtype, xSize);\n    for (let i = 0; i < xSize; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for unary ops from the given\n * `SimpleUnaryImpl`..\n * @param name Kernel name.\n * @param unaryImpl A `SimpleUnaryImpl` that implements the op.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFuncFromImpl(\n    name: string, unaryImpl: SimpleUnaryImpl, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const $dtype = dtype || x.dtype;\n    const newValues = unaryImpl(values, $dtype, attrs);\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\n\nexport const ceilConfig: KernelConfig = {\n  kernelName: Ceil,\n  backendName: 'cpu',\n  kernelFunc: ceilKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\n\nexport const expConfig: KernelConfig = {\n  kernelName: Exp,\n  backendName: 'cpu',\n  kernelFunc: expKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\n\nexport const expm1Config: KernelConfig = {\n  kernelName: Expm1,\n  backendName: 'cpu',\n  kernelFunc: expm1KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\n\nexport const floorConfig: KernelConfig = {\n  kernelName: Floor,\n  backendName: 'cpu',\n  kernelFunc: floorKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\n\nexport const logConfig: KernelConfig = {\n  kernelName: Log,\n  backendName: 'cpu',\n  kernelFunc: logKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function maxImpl(\n    aVals: TypedArray, reduceSize: number, outShape: number[],\n    dtype: DataType): TypedArray {\n  const vals = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(outShape));\n\n  for (let i = 0; i < vals.length; ++i) {\n    const offset = i * reduceSize;\n    let max = aVals[offset];\n    for (let j = 0; j < reduceSize; ++j) {\n      const value = aVals[offset + j];\n      if (value > max) {\n        max = value;\n      }\n    }\n    vals[i] = max;\n  }\n  return vals;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Multiply} from '@tensorflow/tfjs-core';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const multiplyImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n      };\n    }));\n\nexport const multiply =\n    binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\n\nexport const multiplyConfig: KernelConfig = {\n  kernelName: Multiply,\n  backendName: 'cpu',\n  kernelFunc: multiply\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Rsqrt} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\n\nexport const rsqrtConfig: KernelConfig = {\n  kernelName: Rsqrt,\n  backendName: 'cpu',\n  kernelFunc: rsqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelConfig, KernelFunc, NumericDataType, Slice, slice_util, SliceAttrs, SliceInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function sliceImpl(\n    vals: TypedArray, begin: number[], size: number[], shape: number[],\n    dtype: DataType): TypedArray {\n  const isContinous = slice_util.isSliceContinous(shape, begin, size);\n  const length = util.sizeFromShape(size);\n  const xStrides = util.computeStrides(shape);\n\n  if (isContinous) {\n    const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n    return vals.subarray(flatOffset, flatOffset + length);\n  }\n\n  const outVals = util.getTypedArrayFromDType(dtype as NumericDataType, length);\n  for (let i = 0; i < length; ++i) {\n    const rank = size.length;\n    const strides = util.computeStrides(size);\n    const loc = util.indexToLoc(i, rank, strides);\n    const xLoc = loc.map((idx: number, j) => idx + begin[j]);\n    const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n    outVals[i] = vals[xIndex];\n  }\n  return outVals;\n}\n\nexport function slice(\n    args: {inputs: SliceInputs, backend: MathBackendCPU, attrs: SliceAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {begin, size} = attrs;\n\n  assertNotComplex(x, 'slice');\n\n  const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n  slice_util.assertParamsValid(x, $begin, $size);\n\n  const vals = backend.data.get(x.dataId).values as TypedArray;\n  const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n  return backend.makeTensorInfo($size, x.dtype, outVals);\n}\n\nexport const sliceConfig: KernelConfig = {\n  kernelName: Slice,\n  backendName: 'cpu',\n  kernelFunc: slice as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sub} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const subImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal - bReal, imag: aImag - bImag};\n    }));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\n\nexport const subConfig: KernelConfig = {\n  kernelName: Sub,\n  backendName: 'cpu',\n  kernelFunc: sub\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function transposeImpl(\n    xVals: TypedArray, xShape: number[], dtype: DataType, perm: number[],\n    newShape: number[]): TypedArray {\n  const xRank = xShape.length;\n  const xSize = util.sizeFromShape(xShape);\n  const xStrides = util.computeStrides(xShape);\n  const newStrides = util.computeStrides(newShape);\n\n  const result = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(newShape));\n\n  for (let i = 0; i < xSize; ++i) {\n    const loc = util.indexToLoc(i, xRank, xStrides);\n\n    // Permute location.\n    const newLoc: number[] = new Array(loc.length);\n    for (let i = 0; i < newLoc.length; i++) {\n      newLoc[i] = loc[perm[i]];\n    }\n\n    const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n    result[newIndex] = xVals[i];\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendValues, DataType, TensorBuffer, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function uniqueImpl(\n    values: BackendValues, axis: number, shape: number[], dtype: DataType): {\n  outputValues: BackendValues,\n  outputShape: number[],\n  indices: BackendValues\n} {\n  // Normalize and validate axis.\n  const $axis = util.parseAxisParam(axis, shape)[0];\n\n  // Calculate the new shape that is suitable for extracting data along the\n  // given axis.\n  //\n  // The rank is 3.\n  // The size of the 1st dimension is the size of all the axes < the given axis.\n  // The size of the 2nd dimension is the same as the size of the given axis.\n  // The size of the 3rd dimension is the size of all the axes > the given axis.\n  //\n  // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n  // newShape would be: [2*3, 5, 4].\n  //\n  // Note that this is not the final output shape. This will be the shape for an\n  // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n  // values along the given axis. To demonstrate how it works, consider the\n  // following example:\n  //\n  // Input: a 3D tensor, with shape [1, 2, 3]\n  // [\n  //   [\n  //      [1,2,3],\n  //      [4,5,6]\n  //   ]\n  // ]\n  // Axis: 2 (the last axis).\n  // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n  //\n  // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n  // 1*2. The re-shaped data would look like:\n  //\n  // [\n  //   [\n  //     [1], [2], [3]\n  //   ],\n  //   [\n  //     [4], [5], [6]\n  //   ]\n  // ]\n  //\n  // Then, we can construct a 3-level nested loop by the following dimension\n  // order to extract the values along the axis (dimension1):\n  // i: dimension1       // 0,1,2 (newShape[1])\n  //   m: dimension0     // 0,1   (newShape[0])\n  //     n: dimension2   // 0     (newShape[2])\n  //\n  //                       m, i, n\n  //                      ---------\n  // Iteration 0: data at [0, 0, 0] => \"1\"\n  // Iteration 1: data at [1, 0, 0] => \"4\"\n  // We got [1,4].\n  // Iteration 2: data at [0, 1, 0] => \"2\"\n  // Iteration 3: data at [1, 1, 0] => \"5\"\n  // We got [2,5].\n  // Iteration 4: data at [0, 2, 0] => \"3\"\n  // Iteration 5: data at [1, 2, 0] => \"6\"\n  // We got [3,6].\n  const newShape = [1, shape[0], 1];\n  for (let i = 0; i < $axis; i++) {\n    newShape[0] *= shape[i];\n  }\n  newShape[1] = shape[$axis];\n  for (let i = $axis + 1; i < shape.length; i++) {\n    newShape[2] *= shape[i];\n  }\n\n  // A map from unique elements (their string representations) to their values\n  // in \"indices\" (below).\n  const uniqueElements: {[key: string]: number} = {};\n  // The indices of each unique element in the original tensor along the given\n  // axis. It is 1D and has the same size as the given axis.\n  const indices = new Int32Array(shape[$axis]);\n  // Create a buffer so we can easily extract value at a given location.\n  const inputBuffer = new TensorBuffer(newShape, dtype, values as TypedArray);\n  // The indices along the given axis that have unique elements. This is a\n  // de-duped version of \"indices\" above.\n  const uniqueIndices: number[] = [];\n  const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n  for (let i = 0; i < shape[$axis]; i++) {\n    // Extract values along the axis.\n    let element: string;\n    if (is1DTensor) {\n      // Fast path for 1D tensor input.\n      element = values[i].toString();\n    } else {\n      const axisValues = [];\n      for (let m = 0; m < newShape[0]; m++) {\n        for (let n = 0; n < newShape[2]; n++) {\n          axisValues.push(inputBuffer.get(m, i, n));\n        }\n      }\n      element = axisValues.join(',');\n    }\n\n    // Dedup and update various indices.\n    if (uniqueElements[element] !== undefined) {\n      indices[i] = uniqueElements[element];\n    } else {\n      const uniqueIndex = Object.keys(uniqueElements).length;\n      uniqueElements[element] = uniqueIndex;\n      indices[i] = uniqueIndex;\n      uniqueIndices.push(i);\n    }\n  }\n\n  // Now we know where each of the unique elements are located along the axis\n  // (uniqueIndices). Extract them from input buffer and store them in the\n  // output buffer.\n  const outputTmpShape = newShape.slice();\n  outputTmpShape[1] = Object.keys(uniqueElements).length;\n  const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n  uniqueIndices.forEach((uniqueElementIndex, i) => {\n    for (let m = 0; m < newShape[0]; m++) {\n      for (let n = 0; n < newShape[2]; n++) {\n        outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n      }\n    }\n  });\n\n  // The output shape can be calculated from the input shape with the size of\n  // the given axis replaced by the number of unique elements along that axis.\n  const outputShape = shape.slice();\n  outputShape[$axis] = outputTmpShape[1];\n\n  return {\n    outputValues: outputBuffer.values as BackendValues,\n    outputShape,\n    indices,\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Shared kernel impls for use in other backends.\nexport {simpleAbsImpl} from './kernels/Abs';\nexport {addImpl} from './kernels/Add';\nexport {ceilImpl} from './kernels/Ceil';\nexport {expImpl} from './kernels/Exp';\nexport {expm1Impl} from './kernels/Expm1';\nexport {floorImpl} from './kernels/Floor';\nexport {logImpl} from './kernels/Log';\nexport {maxImpl} from './kernels/Max_impl';\nexport {multiplyImpl} from './kernels/Multiply';\nexport {rsqrtImpl} from './kernels/Rsqrt';\nexport {sliceImpl} from './kernels/Slice';\nexport {subImpl} from './kernels/Sub';\nexport {transposeImpl} from './kernels/Transpose_impl';\nexport {uniqueImpl} from './kernels/Unique_impl';\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/*\n * base.ts contains all the exports from tfjs-backend-cpu\n * without auto-kernel registration\n */\nimport {registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from './backend_cpu';\nimport * as shared from './shared';\n\nexport {MathBackendCPU} from './backend_cpu';\nexport {version as version_cpu} from './version';\nexport {shared};\n\n// Side effects for default initialization of MathBackendCPU\nregisterBackend('cpu', () => new MathBackendCPU(), 1 /* priority */);\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acosKernelFunc = unaryKernelFunc(Acos, (xi) => Math.acos(xi));\n\nexport const acosConfig: KernelConfig = {\n  kernelName: Acos,\n  backendName: 'cpu',\n  kernelFunc: acosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acoshKernelFunc = unaryKernelFunc(Acosh, (xi) => Math.acosh(xi));\n\nexport const acoshConfig: KernelConfig = {\n  kernelName: Acosh,\n  backendName: 'cpu',\n  kernelFunc: acoshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinKernelFunc = unaryKernelFunc(Asin, (xi) => Math.asin(xi));\n\nexport const asinConfig: KernelConfig = {\n  kernelName: Asin,\n  backendName: 'cpu',\n  kernelFunc: asinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinhKernelFunc = unaryKernelFunc(Asinh, (xi) => Math.asinh(xi));\n\nexport const asinhConfig: KernelConfig = {\n  kernelName: Asinh,\n  backendName: 'cpu',\n  kernelFunc: asinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanKernelFunc = unaryKernelFunc(Atan, (xi) => Math.atan(xi));\n\nexport const atanConfig: KernelConfig = {\n  kernelName: Atan,\n  backendName: 'cpu',\n  kernelFunc: atanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanhKernelFunc = unaryKernelFunc(Atanh, (xi) => Math.atanh(xi));\n\nexport const atanhConfig: KernelConfig = {\n  kernelName: Atanh,\n  backendName: 'cpu',\n  kernelFunc: atanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, buffer, DataType, Rank, TensorBuffer, TypedArray} from '@tensorflow/tfjs-core';\n\nexport function pool(\n    xValues: TypedArray, xShape: number[], dtype: DataType, strides: number[],\n    convInfo: backend_util.Conv2DInfo,\n    poolType: 'max'|'avg'): TensorBuffer<Rank, DataType> {\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const initialValue =\n      (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                            Number.POSITIVE_INFINITY);\n\n  const output = buffer(convInfo.outShape, dtype);\n  const outputVals = output.values;\n\n  const outputBatchStrides =\n      convInfo.outShape[1] * convInfo.outShape[2] * convInfo.outShape[3];\n  const outputRowStrides = convInfo.outShape[2] * convInfo.outShape[3];\n  const outputColStrides = convInfo.outShape[3];\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    const outputBatchOffset = b * outputBatchStrides;\n    const inputBatchOffset = b * strides[0];\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        const outputRowOffset = outputBatchOffset + yR * outputRowStrides;\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          const xCMin = Math.max(0, xCCorner);\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let minMaxValue = initialValue;\n          let avgValue = 0;\n          let count = 0;\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const xROffset = inputBatchOffset + xR * strides[1];\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const xCOffset = xROffset + xC * strides[2];\n              const pixel = xValues[xCOffset + d];\n              if ((poolType === 'max' && pixel > minMaxValue)) {\n                minMaxValue = pixel;\n              } else if (poolType === 'avg') {\n                avgValue += pixel;\n                count++;\n              }\n            }\n            if (isNaN(minMaxValue)) {\n              break;\n            }\n          }\n          const outputOffset = outputRowOffset + yC * outputColStrides + d;\n          outputVals[outputOffset] =\n              poolType === 'avg' ? avgValue / count : minMaxValue;\n        }\n      }\n    }\n  }\n  return output;\n}\n\nexport function maxPoolPositions(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    convInfo: backend_util.Conv2DInfo, flattenPositions = false,\n    includeBatchInIndex = false): TensorBuffer<Rank, 'int32'> {\n  const maxPositions = buffer(convInfo.outShape, 'int32');\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const xBuf = buffer(xShape, dtype, xValues);\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        let xRMin = xRCorner;\n        while (xRMin < 0) {\n          xRMin += dilationHeight;\n        }\n        // const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          let xCMin = xCCorner;\n          while (xCMin < 0) {\n            xCMin += dilationWidth;\n          }\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let maxValue = Number.NEGATIVE_INFINITY;\n          let maxPosition = -1;\n\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const wR = xR - xRCorner;\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const wC = xC - xCCorner;\n              const pixel = xBuf.get(b, xR, xC, d);\n              if (pixel > maxValue) {\n                maxValue = pixel as number;\n                if (flattenPositions) {\n                  maxPosition = includeBatchInIndex ?\n                      ((b * convInfo.inHeight + xR) * convInfo.inWidth + xC) *\n                              convInfo.inChannels +\n                          d :\n                      (xR * convInfo.inWidth + xC) * convInfo.inChannels + d;\n                } else {\n                  maxPosition = wR * effectiveFilterWidth + wC;\n                }\n              }\n            }\n          }\n          maxPositions.set(maxPosition, b, yR, yC, d);\n        }\n      }\n    }\n  }\n  return maxPositions;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function avgPool(\n    args:\n        {inputs: AvgPoolInputs, backend: MathBackendCPU, attrs: AvgPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'avg');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'cpu',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, buffer, KernelConfig, KernelFunc, Rank, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const filterHeight = convInfo.filterHeight;\n  const filterWidth = convInfo.filterWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel;\n            }\n          }\n          dx.set(dotProd * avgMultiplier, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function batchNormKernelFunc(args: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendCPU,\n  attrs: FusedBatchNormAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x, scale, offset, mean, variance} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  assertNotComplex([x, mean, variance, scale, offset], 'batchNorm');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const mVals = backend.data.get(mean.dataId).values as TypedArray;\n  const varVals = backend.data.get(variance.dataId).values as TypedArray;\n  const sVals = scale ? backend.data.get(scale.dataId).values as TypedArray :\n                        new Float32Array([1]);\n  const offVals = offset ?\n      backend.data.get(offset.dataId).values as TypedArray :\n      new Float32Array([0]);\n  const outVals = new Float32Array(xVals.length);\n\n  const offValsLength = offVals.length;\n  const sValsLength = sVals.length;\n  const varValsLength = varVals.length;\n  const mValsLength = mVals.length;\n\n  let offi = 0;\n  let mi = 0;\n  let si = 0;\n  let vi = 0;\n  for (let i = 0; i < xVals.length; ++i) {\n    outVals[i] = offVals[offi++] +\n        (xVals[i] - mVals[mi++]) * sVals[si++] /\n            Math.sqrt(varVals[vi++] + varianceEpsilon);\n    if (offi >= offValsLength) {\n      offi = 0;\n    }\n    if (mi >= mValsLength) {\n      mi = 0;\n    }\n    if (si >= sValsLength) {\n      si = 0;\n    }\n    if (vi >= varValsLength) {\n      vi = 0;\n    }\n  }\n  return backend.makeTensorInfo(x.shape, x.dtype, outVals);\n}\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'cpu',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const clipKernelFunc = unaryKernelFunc(ClipByValue, (xi, attrs) => {\n  const clipAttrs = attrs as {} as ClipByValueAttrs;\n  if (xi > clipAttrs.clipValueMax) {\n    return clipAttrs.clipValueMax;\n  }\n  return xi < clipAttrs.clipValueMin ? clipAttrs.clipValueMin : xi;\n});\n\nexport const clipConfig: KernelConfig = {\n  kernelName: ClipByValue,\n  backendName: 'cpu',\n  kernelFunc: clipKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Imag, ImagInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function imag(args: {inputs: ImagInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const imag = backend.data.get(input.dataId).complexTensorInfos.imag;\n  const imagVal = backend.data.get(imag.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the imag value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(imag.shape, imag.dtype, imagVal);\n}\n\nexport const imagConfig: KernelConfig = {\n  kernelName: Imag,\n  backendName: 'cpu',\n  kernelFunc: imag as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function reshape(\n    args:\n        {inputs: ReshapeInputs, backend: MathBackendCPU, attrs: ReshapeAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  backend.incRef(x.dataId);\n\n  const xData = backend.data.get(x.dataId);\n\n  if (xData.complexTensorInfos != null) {\n    const real = xData.complexTensorInfos.real;\n    const imag = xData.complexTensorInfos.imag;\n\n    real.shape = $shape;\n    imag.shape = $shape;\n  }\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'cpu',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Concat, ConcatAttrs, ConcatInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nimport {complex} from './Complex';\nimport {imag} from './Imag';\nimport {real} from './Real';\nimport {reshape} from './Reshape';\n\nexport function concat(\n    args: {inputs: ConcatInputs, backend: MathBackendCPU, attrs: ConcatAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {axis} = attrs;\n\n  const $axis = util.parseAxisParam(axis, inputs[0].shape)[0];\n  let outShape = backend_util.computeOutShape(inputs.map(t => t.shape), $axis);\n\n  if (util.sizeFromShape(outShape) === 0) {\n    return backend.makeTensorInfo(outShape, inputs[0].dtype, []);\n  }\n\n  // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n  const $inputs = inputs.filter(t => util.sizeFromShape(t.shape) > 0);\n  if ($inputs.length === 1) {\n    return $inputs[0];\n  }\n\n  const shapes = $inputs.map(t => t.shape);\n  backend_util.assertParamsConsistent(shapes, $axis);\n\n  if ($inputs[0].dtype === 'complex64') {\n    const reals = $inputs.map((t) => real({inputs: {input: t}, backend}));\n    const imags = $inputs.map((t) => imag({inputs: {input: t}, backend}));\n\n    const realConcated = concat({inputs: reals, backend, attrs: {axis}});\n    const imagConcated = concat({inputs: imags, backend, attrs: {axis}});\n\n    const result =\n        complex({inputs: {real: realConcated, imag: imagConcated}, backend});\n\n    reals.forEach(r => backend.disposeIntermediateTensorInfo(r));\n    imags.forEach(i => backend.disposeIntermediateTensorInfo(i));\n    backend.disposeIntermediateTensorInfo(realConcated);\n    backend.disposeIntermediateTensorInfo(imagConcated);\n\n    return result;\n  }\n\n  // Any concat of n-dimensional tensors across any axis can be reduced to\n  // a concatenation of two-dimensional tensors across the axis 1 by first\n  // partitioning the axes of the original tensors into those less than the\n  // axis to be concatenated and the rest. Then reshape the tensors\n  // into a two-dimensional tensor by collapsing these two sets of axes and\n  // concatenate the resulting matrices across the axis 1, finally reshaping\n  // the result to have the proper shape.\n  const inputs2D = $inputs.map(t => {\n    const innerSize = util.sizeFromShape(t.shape.slice($axis));\n    const shape = [-1, innerSize];\n    return reshape({inputs: {x: t}, backend, attrs: {shape}});\n  });\n\n  // Concats 2d tensors along axis=1.\n  outShape =\n      backend_util.computeOutShape(inputs2D.map(t => t.shape), 1 /* axis */);\n\n  const outVals = util.getTypedArrayFromDType(\n      $inputs[0].dtype as 'float32', util.sizeFromShape(outShape));\n\n  if (inputs2D[0].shape[0] === 1) {\n    // Use built-in TypedArray.set() method for speed.\n    let offset = 0;\n    inputs2D.forEach(t => {\n      const val = backend.data.get(t.dataId).values as TypedArray;\n      const size = util.sizeFromShape(t.shape);\n\n      outVals.set(val, offset);\n      offset += size;\n    });\n  } else {\n    let colOffset = 0;\n\n    inputs2D.forEach(t => {\n      const tVals = backend.data.get(t.dataId).values as TypedArray;\n\n      let tIdx = 0;\n\n      for (let row = 0; row < t.shape[0]; ++row) {\n        const resIdx = row * outShape[1] + colOffset;\n        for (let col = 0; col < t.shape[1]; ++col) {\n          outVals[resIdx + col] = tVals[tIdx++];\n        }\n      }\n\n      colOffset += t.shape[1];\n    });\n  }\n\n  const finalOutShape =\n      backend_util.computeOutShape($inputs.map(t => t.shape), $axis);\n\n  const outInfo =\n      backend.makeTensorInfo(finalOutShape, inputs[0].dtype, outVals);\n\n  inputs2D.forEach(t => backend.disposeIntermediateTensorInfo(t));\n\n  return outInfo;\n}\n\nexport const concatConfig: KernelConfig = {\n  kernelName: Concat,\n  backendName: 'cpu',\n  kernelFunc: concat as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const cosKernelFunc = unaryKernelFunc(Cos, (xi) => Math.cos(xi));\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'cpu',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const coshKernelFunc = unaryKernelFunc(Cosh, (xi) => Math.cosh(xi));\n\nexport const coshConfig: KernelConfig = {\n  kernelName: Cosh,\n  backendName: 'cpu',\n  kernelFunc: coshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2D, Dilation2DAttrs, Dilation2DInputs, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dConfig: KernelConfig = {\n  kernelName: Dilation2D,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter} = inputs as Dilation2DInputs;\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xRank = x.shape.length;\n\n    const filterVals = cpuBackend.data.get(filter.dataId).values as TypedArray;\n    const filterRank = filter.shape.length;\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    const outSize = util.sizeFromShape(outShape);\n    const outRank = outShape.length;\n    const outputVals = util.getArrayFromDType(x.dtype, outSize);\n\n    // Upsampling the input by fill in `dilation size - 1` values between each\n    // input value.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const xIndex = util.locToIndex(\n                        [b, hIn, wIn, d], xRank, util.computeStrides(x.shape));\n                    const filterIndex = util.locToIndex(\n                        [h, w, d], filterRank,\n                        util.computeStrides(filter.shape));\n                    const val = xVals[xIndex] + filterVals[filterIndex];\n                    if (val > curVal) {\n                      curVal = val;\n                    }\n                  }\n                }\n              }\n            }\n            const outputIndex = util.locToIndex(\n                [b, hOut, wOut, d], outRank, util.computeStrides(outShape));\n            outputVals[outputIndex] = curVal;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(outputVals, x.dtype), outShape, x.dtype);\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropFilter, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropFilterConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropFilter,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropFilter}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed filter gradients has the same dimensions as the filter:\n    // [filterHeight, filterWidth, depth]\n    const gradients = util.makeZerosNestedTypedArray(\n                          filter.shape, filter.dtype) as number[][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hMax = 0;\n            let wMax = 0;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hMax = h;\n                      wMax = w;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[hMax][wMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), filter.shape, filter.dtype);\n\n    return {dataId, shape: filter.shape, dtype: filter.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropInput, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropInputConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropInput,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropInput}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed gradients has the same dimensions as the input:\n    // [batch, inputHeight, inputCols, inChannel]\n    const gradients =\n        util.makeZerosNestedTypedArray(x.shape, x.dtype) as number[][][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hInMax = (hBeg < 0) ? 0 : hBeg;\n            let wInMax = (wBeg < 0) ? 0 : wBeg;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hInMax = hIn;\n                      wInMax = wIn;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[b][hInMax][wInMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), x.shape, x.dtype);\n\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const divImpl =\n    createSimpleBinaryKernelImpl((a: number, b: number) => a / b);\nexport const div = binaryKernelFunc(Div, divImpl);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'cpu',\n  kernelFunc: div\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Elu, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const eluKernelFunc =\n    unaryKernelFunc(Elu, (xi) => xi >= 0 ? xi : (Math.exp(xi) - 1));\n\nexport const eluConfig: KernelConfig = {\n  kernelName: Elu,\n  backendName: 'cpu',\n  kernelFunc: eluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Erf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst p = backend_util.ERF_P;\nconst a1 = backend_util.ERF_A1;\nconst a2 = backend_util.ERF_A2;\nconst a3 = backend_util.ERF_A3;\nconst a4 = backend_util.ERF_A4;\nconst a5 = backend_util.ERF_A5;\n\nexport const erfKernelFunc = unaryKernelFunc(\n    Erf,\n    (xi) => {\n      const sign = Math.sign(xi);\n      const v = Math.abs(xi);\n      const t = 1.0 / (1.0 + p * v);\n      return sign *\n          (1.0 -\n           (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t *\n               Math.exp(-v * v));\n    },\n);\n\nexport const erfConfig: KernelConfig = {\n  kernelName: Erf,\n  backendName: 'cpu',\n  kernelFunc: erfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Tensor, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {add} from '../kernels/Add';\nimport {complex} from '../kernels/Complex';\nimport {concat} from '../kernels/Concat';\nimport {divConfig} from '../kernels/Div';\nimport {identity} from '../kernels/Identity';\nimport {imag} from '../kernels/Imag';\nimport {multiply} from '../kernels/Multiply';\nimport {real} from '../kernels/Real';\nimport {slice} from '../kernels/Slice';\nimport {sub} from '../kernels/Sub';\n\n/**\n * Calculate FFT of inner most elements of batch tensor.\n */\nexport function fftBatch(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): TensorInfo {\n  const inputShape = input.shape;\n  const batch = inputShape[0];\n  const innerDim = inputShape[1];\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const real2D = inputVals.complexTensorInfos.real;\n  const imag2D = inputVals.complexTensorInfos.imag;\n\n  // Collects real and imaginary values separately.\n  const resultShape = [batch, innerDim];\n  const resultSize = util.sizeFromShape(resultShape);\n  const resultReal = util.getTypedArrayFromDType('float32', resultSize);\n  const resultImag = util.getTypedArrayFromDType('float32', resultSize);\n\n  for (let b = 0; b < batch; b++) {\n    // TODO: Support slice ops for complex type.\n    const r = slice({\n      inputs: {x: real2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n    const i = slice({\n      inputs: {x: imag2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n\n    const input = complex({inputs: {real: r, imag: i}, backend: cpuBackend});\n\n    // Run FFT by batch element.\n    const {real, imag} = fftImpl(input, inverse, cpuBackend);\n    const res = backend_util.mergeRealAndImagArrays(real, imag);\n\n    for (let d = 0; d < innerDim; d++) {\n      const c = backend_util.getComplexWithIndex(res, d);\n      resultReal[b * innerDim + d] = c.real;\n      resultImag[b * innerDim + d] = c.imag;\n    }\n\n    cpuBackend.disposeIntermediateTensorInfo(r);\n    cpuBackend.disposeIntermediateTensorInfo(i);\n    cpuBackend.disposeIntermediateTensorInfo(input);\n  }\n\n  const $realInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultReal);\n  const $imagInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultImag);\n\n  const result = complex(\n      {inputs: {real: $realInfo, imag: $imagInfo}, backend: cpuBackend});\n\n  cpuBackend.disposeIntermediateTensorInfo($realInfo);\n  cpuBackend.disposeIntermediateTensorInfo($imagInfo);\n\n  return result;\n}\n\nexport function fftImpl(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  const inputSize = util.sizeFromShape(input.shape);\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const realVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.real.dataId).values as\n      Float32Array;\n\n  const imagVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.imag.dataId).values as\n      Float32Array;\n\n  if (isExponentOf2(inputSize)) {\n    const result =\n        fftRadix2(realVals, imagVals, inputSize, inverse, cpuBackend);\n\n    const resultShape = [input.shape[0], input.shape[1]];\n\n    if (inverse) {\n      const realInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.real);\n      const imagInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.imag);\n\n      const sizeInfo: TensorInfo = cpuBackend.makeTensorInfo(\n          [], 'float32',\n          util.createScalarValue(inputSize as {} as 'float32', 'float32'));\n      const sizeInfoCopy =\n          identity({inputs: {x: sizeInfo}, backend: cpuBackend});\n\n      const divRealInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: realInfo, b: sizeInfo}, backend: cpuBackend}) as\n          TensorInfo;\n      const divImagInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: imagInfo, b: sizeInfoCopy}, backend: cpuBackend}) as\n          TensorInfo;\n\n      const divRealVals =\n          cpuBackend.data.get(divRealInfo.dataId).values as Float32Array;\n      const divImagVals =\n          cpuBackend.data.get(divImagInfo.dataId).values as Float32Array;\n\n      cpuBackend.disposeIntermediateTensorInfo(realInfo);\n      cpuBackend.disposeIntermediateTensorInfo(imagInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfoCopy);\n      cpuBackend.disposeIntermediateTensorInfo(divRealInfo);\n      cpuBackend.disposeIntermediateTensorInfo(divImagInfo);\n\n      return {real: divRealVals, imag: divImagVals};\n    }\n\n    return result;\n  } else {\n    const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n    const rawOutput =\n        fourierTransformByMatmul(data, inputSize, inverse) as Float32Array;\n\n    return backend_util.splitRealAndImagArrays(rawOutput);\n  }\n}\n\nfunction isExponentOf2(size: number): boolean {\n  return (size & size - 1) === 0;\n}\n\n// FFT using Cooley-Tukey algorithm on radix 2 dimensional input.\nfunction fftRadix2(\n    realVals: Float32Array, imagVals: Float32Array, size: number,\n    inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  if (size === 1) {\n    return {real: realVals, imag: imagVals};\n  }\n\n  const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n  const half = size / 2;\n\n  const evenComplex = backend_util.complexWithEvenIndex(data);\n\n  const evenRealVals = evenComplex.real;\n  const evenImagVals = evenComplex.imag;\n\n  const evenShape = [evenRealVals.length];\n\n  const evenRealInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenRealVals);\n  const evenImagInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenImagVals);\n\n  const evenTensorInfo = complex(\n      {inputs: {real: evenRealInfo, imag: evenImagInfo}, backend: cpuBackend});\n\n  const oddComplex = backend_util.complexWithOddIndex(data);\n\n  const oddRealVals = oddComplex.real;\n  const oddImagVals = oddComplex.imag;\n\n  const oddShape = [oddRealVals.length];\n\n  const oddRealInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddRealVals);\n  const oddImagInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddImagVals);\n\n  const oddTensorInfo = complex(\n      {inputs: {real: oddRealInfo, imag: oddImagInfo}, backend: cpuBackend});\n\n  // Recursive call for half part of original input.\n  const $evenComplex =\n      fftRadix2(evenRealVals, evenImagVals, half, inverse, cpuBackend);\n\n  const $evenRealVals = $evenComplex.real;\n  const $evenImagVals = $evenComplex.imag;\n\n  const $evenShape = [$evenRealVals.length];\n\n  const $evenRealInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenRealVals);\n  const $evenImagInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenImagVals);\n\n  const $evenTensorInfo = complex({\n    inputs: {real: $evenRealInfo, imag: $evenImagInfo},\n    backend: cpuBackend\n  });\n\n  const $oddComplex =\n      fftRadix2(oddRealVals, oddImagVals, half, inverse, cpuBackend);\n\n  const $oddRealVals = $oddComplex.real;\n  const $oddImagVals = $oddComplex.imag;\n\n  const $oddShape = [$oddRealVals.length];\n\n  const $oddRealInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddRealVals);\n  const $oddImagInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddImagVals);\n\n  const $oddTensorInfo = complex(\n      {inputs: {real: $oddRealInfo, imag: $oddImagInfo}, backend: cpuBackend});\n\n  const e = backend_util.exponents(size, inverse);\n  const eShape = [e.real.length];\n\n  const eRealInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.real);\n  const eImagInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.imag);\n\n  const complexInfo = complex(\n      {inputs: {real: eRealInfo, imag: eImagInfo}, backend: cpuBackend});\n\n  const exponentInfo =\n      multiply(\n          {inputs: {a: complexInfo, b: $oddTensorInfo}, backend: cpuBackend}) as\n      TensorInfo;\n\n  const addPart = add({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n  const subPart = sub({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n\n  const addPartReal = real({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartReal = real({inputs: {input: subPart}, backend: cpuBackend});\n\n  const addPartImag = imag({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartImag = imag({inputs: {input: subPart}, backend: cpuBackend});\n\n  const $real = concat({\n    inputs: [addPartReal as Tensor, subPartReal as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n  const $imag = concat({\n    inputs: [addPartImag as Tensor, subPartImag as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n\n  const $realVals = cpuBackend.data.get($real.dataId).values as Float32Array;\n  const $imagVals = cpuBackend.data.get($imag.dataId).values as Float32Array;\n\n  cpuBackend.disposeIntermediateTensorInfo(evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(complexInfo);\n  cpuBackend.disposeIntermediateTensorInfo(exponentInfo);\n  cpuBackend.disposeIntermediateTensorInfo(addPart);\n  cpuBackend.disposeIntermediateTensorInfo(subPart);\n  cpuBackend.disposeIntermediateTensorInfo(addPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(addPartImag);\n  cpuBackend.disposeIntermediateTensorInfo(subPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(subPartImag);\n  cpuBackend.disposeIntermediateTensorInfo($real);\n  cpuBackend.disposeIntermediateTensorInfo($imag);\n\n  return {real: $realVals, imag: $imagVals};\n}\n\n// Calculate fourier transform by multplying sinusoid matrix.\nfunction fourierTransformByMatmul(\n    data: TypedArray, size: number, inverse: boolean): TypedArray {\n  const ret = new Float32Array(size * 2);\n  // TODO: Use matmul instead once it supports complex64 type.\n  for (let r = 0; r < size; r++) {\n    let real = 0.0;\n    let imag = 0.0;\n    for (let c = 0; c < size; c++) {\n      const e = backend_util.exponent(r * c, size, inverse);\n      const term = backend_util.getComplexWithIndex(data as Float32Array, c);\n      real += term.real * e.real - term.imag * e.imag;\n      imag += term.real * e.imag + term.imag * e.real;\n    }\n    if (inverse) {\n      real /= size;\n      imag /= size;\n    }\n    backend_util.assignToTypedArray(ret, real, imag, r);\n  }\n  return ret;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FFT, FFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function fft(args: {inputs: FFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, false, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const fftConfig: KernelConfig = {\n  kernelName: FFT,\n  backendName: 'cpu',\n  kernelFunc: fft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n\n            const coordX = Math.round(imageWidth - x);\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n\n            let outputValue = imageVals[outIdx];\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth) {\n              // set the output to the image value at the coordinate position.\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n            output[outIdx] = outputValue;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IFFT, IFFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function ifft(args: {inputs: IFFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, true, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const ifftConfig: KernelConfig = {\n  kernelName: IFFT,\n  backendName: 'cpu',\n  kernelFunc: ifft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isFiniteKernelFunc =\n    unaryKernelFunc(IsFinite, (xi) => Number.isFinite(xi) ? 1 : 0, 'bool');\n\nexport const isFiniteConfig: KernelConfig = {\n  kernelName: IsFinite,\n  backendName: 'cpu',\n  kernelFunc: isFiniteKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isInfKernelFunc =\n    unaryKernelFunc(IsInf, (xi) => Math.abs(xi) === Infinity ? 1 : 0, 'bool');\n\nexport const isInfConfig: KernelConfig = {\n  kernelName: IsInf,\n  backendName: 'cpu',\n  kernelFunc: isInfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isNaNKernelFunc =\n    unaryKernelFunc(IsNan, (xi) => Number.isNaN(xi) ? 1 : 0, 'bool');\n\nexport const isNaNConfig: KernelConfig = {\n  kernelName: IsNan,\n  backendName: 'cpu',\n  kernelFunc: isNaNKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log1p} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const log1pKernelFunc = unaryKernelFunc(Log1p, (xi) => Math.log1p(xi));\n\nexport const log1pConfig: KernelConfig = {\n  kernelName: Log1p,\n  backendName: 'cpu',\n  kernelFunc: log1pKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, LogicalNot} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const logicalNotKernelFunc =\n    unaryKernelFunc(LogicalNot, (xi) => xi ? 0 : 1, 'bool');\n\nexport const logicalNotConfig: KernelConfig = {\n  kernelName: LogicalNot,\n  backendName: 'cpu',\n  kernelFunc: logicalNotKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig} from '@tensorflow/tfjs-core';\nimport {TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    let xShape = x.shape;\n    const xRank = xShape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, xShape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    let xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    if (permutedAxes != null) {\n      const newShape: number[] = new Array(xRank);\n      for (let i = 0; i < newShape.length; i++) {\n        newShape[i] = xShape[permutedAxes[i]];\n      }\n\n      xVals = transposeImpl(xVals, xShape, x.dtype, permutedAxes, newShape);\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n\n      xShape = newShape;\n    }\n\n    assertNotComplex(x, 'max');\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(xShape, axes);\n\n    const reduceSize = util.sizeFromShape(reduceShape);\n\n    const result = maxImpl(xVals, reduceSize, maxOutShape, x.dtype);\n    const dataId = cpuBackend.write(result, maxOutShape, x.dtype);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // reshape\n      const newShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n      outShape = newShape;\n    }\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function maxPool(\n    args:\n        {inputs: MaxPoolInputs, backend: MathBackendCPU, attrs: MaxPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'max');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'cpu',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, buffer, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, Rank, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {maxPoolPositions} from '../utils/pool_utils';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const xValues = backend.data.get(x.dataId).values as TypedArray;\n  const maxPosBuf = buffer(\n      convInfo.outShape, x.dtype,\n      maxPoolPositions(xValues, x.shape, x.dtype, convInfo).values);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n              const maxPos = effectiveFilterHeight * effectiveFilterWidth - 1 -\n                  (maxPosBuf.get(b, dyR, dyC, d) as number);\n              const curPos = wR * effectiveFilterWidth + wC;\n\n              const mask = maxPos === curPos ? 1 : 0;\n              if (mask === 0) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel * mask;\n            }\n          }\n          dx.set(dotProd, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, DataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {maxPoolPositions, pool} from '../utils/pool_utils';\nexport function maxPoolWithArgmaxImpl(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    includeBatchInIndex: boolean, convInfo: backend_util.Conv2DInfo) {\n  const strides = util.computeStrides(xShape);\n  const maxPools = pool(xValues, xShape, dtype, strides, convInfo, 'max');\n  const maxPositions = maxPoolPositions(\n      xValues, xShape, dtype, convInfo, true, includeBatchInIndex);\n\n  return [maxPools.values, maxPositions.values];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'MaxPoolWithArgmax');\n\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        [1, 1], pad);\n    const [pooled, indexes] = maxPoolWithArgmaxImpl(\n        values, x.shape, x.dtype, includeBatchInIndex, convInfo);\n\n    const pooledDataId =\n        cpuBackend.write(pooled as Float32Array, convInfo.outShape, x.dtype);\n    const indexesDataId =\n        cpuBackend.write(indexes as Int32Array, convInfo.outShape, x.dtype);\n    return [\n      {dataId: pooledDataId, shape: convInfo.outShape, dtype: x.dtype},\n      {dataId: indexesDataId, shape: convInfo.outShape, dtype: 'int32'}\n    ];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionPadded');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionWithScore');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NotEqual} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const notEqualImpl =\n    createSimpleBinaryKernelImpl(((a, b) => (a !== b) ? 1 : 0));\nexport const notEqual =\n    binaryKernelFunc(NotEqual, notEqualImpl, null /* complexOp */, 'bool');\n\nexport const notEqualConfig: KernelConfig = {\n  kernelName: NotEqual,\n  backendName: 'cpu',\n  kernelFunc: notEqual\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, NumericDataType, PadV2, PadV2Attrs, PadV2Inputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function padV2(\n    args: {inputs: PadV2Inputs, backend: MathBackendCPU, attrs: PadV2Attrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {paddings, constantValue} = attrs;\n\n  assertNotComplex(x, 'pad');\n\n  const outShape = paddings.map(\n      (p, i) => p[0] /* beforePad */ + x.shape[i] + p[1] /* afterPad */);\n\n  const start = paddings.map(p => p[0]);\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const xSize = util.sizeFromShape(x.shape);\n  const xRank = x.shape.length;\n  const xStrides = util.computeStrides(x.shape);\n\n  const resultSize = util.sizeFromShape(outShape);\n  const resultRank = outShape.length;\n  const resultStrides = util.computeStrides(outShape);\n  const resVals =\n      util.getTypedArrayFromDType(x.dtype as NumericDataType, resultSize);\n\n  if (constantValue !== 0) {\n    resVals.fill(constantValue);\n  }\n\n  for (let i = 0; i < xSize; i++) {\n    const coords = util.indexToLoc(i, xRank, xStrides);\n    const outCoords = coords.map((c, i) => c + start[i]);\n    const outIndex = util.locToIndex(outCoords, resultRank, resultStrides);\n\n    resVals[outIndex] = xVals[i];\n  }\n\n  const outId = backend.write(resVals, outShape, x.dtype);\n\n  return {dataId: outId, shape: outShape, dtype: x.dtype};\n}\n\nexport const padV2Config: KernelConfig = {\n  kernelName: PadV2,\n  backendName: 'cpu',\n  kernelFunc: padV2 as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Reciprocal} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const reciprocalKernelFunc = unaryKernelFunc(Reciprocal, (xi) => 1 / xi);\n\nexport const reciprocalConfig: KernelConfig = {\n  kernelName: Reciprocal,\n  backendName: 'cpu',\n  kernelFunc: reciprocalKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {backend_util, RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const fullOpacityValue = 255;\n\n    const sinFactor = Math.sin(radians);\n    const cosFactor = Math.cos(radians);\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n            const y = coords[1];\n\n            // coordX/coordY are the result of rotating and translating x/y.\n            let coordX = (x - centerX) * cosFactor - (y - centerY) * sinFactor;\n            let coordY = (x - centerX) * sinFactor + (y - centerY) * cosFactor;\n            coordX = Math.round(coordX + centerX);\n            coordY = Math.round(coordY + centerY);\n\n            let outputValue = fillValue;\n            if (typeof fillValue !== 'number') {\n              if (channel === 3) {\n                outputValue = fullOpacityValue;\n              } else {\n                outputValue = fillValue[channel];\n              }\n            }\n\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth && coordY >= 0 &&\n                coordY < imageHeight) {\n              // set the output to the image value at the coordinate position.\n              const rotatedRowOffset = coordY * (imageWidth * numChannels);\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rotatedRowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n            output[outIdx] = outputValue as number;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Round} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const roundKernelFunc = unaryKernelFunc(Round, (xi) => {\n  // The algorithm is based on banker's rounding.\n  const base = Math.floor(xi);\n  if (xi - base < 0.5) {\n    return Math.floor(xi);\n  } else if (xi - base > 0.5) {\n    return Math.ceil(xi);\n  } else {\n    if (base % 2.0 === 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n});\n\nexport const roundConfig: KernelConfig = {\n  kernelName: Round,\n  backendName: 'cpu',\n  kernelFunc: roundKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, Selu} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst scaleAlpha = backend_util.SELU_SCALEALPHA;\nconst scale = backend_util.SELU_SCALE;\n\nexport const seluKernelFunc = unaryKernelFunc(Selu, (xi) => {\n  if (xi >= 0) {\n    return scale * xi;\n  } else {\n    return scaleAlpha * (Math.exp(xi) - 1);\n  }\n});\n\nexport const seluConfig: KernelConfig = {\n  kernelName: Selu,\n  backendName: 'cpu',\n  kernelFunc: seluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sigmoid} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sigmoidKernelFunc =\n    unaryKernelFunc(Sigmoid, (xi) => 1 / (1 + Math.exp(-xi)));\n\nexport const sigmoidConfig: KernelConfig = {\n  kernelName: Sigmoid,\n  backendName: 'cpu',\n  kernelFunc: sigmoidKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sign} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const signKernelFunc = unaryKernelFunc(Sign, (xi) => {\n  if (xi < 0) {\n    return -1;\n  } else if (xi > 0) {\n    return 1;\n  } else {\n    return 0;\n  }\n});\n\nexport const signConfig: KernelConfig = {\n  kernelName: Sign,\n  backendName: 'cpu',\n  kernelFunc: signKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinKernelFunc = unaryKernelFunc(Sin, (xi) => Math.sin(xi));\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'cpu',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sinh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinhKernelFunc = unaryKernelFunc(Sinh, (xi) => Math.sinh(xi));\n\nexport const sinhConfig: KernelConfig = {\n  kernelName: Sinh,\n  backendName: 'cpu',\n  kernelFunc: sinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Softplus} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\n// mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n\n// epsilon is the difference between 1.0 and the next representable float.\n// For a single precision 32 bit float this should be 2^-23, see:\n// https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\nconst epsilon = 1.1920928955078125e-7;\nconst threshold = Math.log(epsilon) + 2.0;\n\nexport const softplusKernelFunc = unaryKernelFunc(Softplus, (xi) => {\n  // Value above which exp(x) may overflow, but softplus(x) == x\n  // is within machine epsilon.\n  const tooLarge = xi > -threshold;\n\n  // Value below which exp(x) may underflow, but softplus(x) == exp(x)\n  // is within machine epsilon.\n  const tooSmall = xi < threshold;\n\n  const expX = Math.exp(xi);\n  let result;\n\n  if (tooSmall) {\n    result = expX;\n  } else if (tooLarge) {\n    result = xi;\n  } else {\n    result = Math.log(1.0 + expX);\n  }\n  return result;\n});\n\nexport const softplusConfig: KernelConfig = {\n  kernelName: Softplus,\n  backendName: 'cpu',\n  kernelFunc: softplusKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {transposeImpl} from './Transpose_impl';\n\nexport function transpose(args: {\n  inputs: TransposeInputs,\n  attrs: TransposeAttrs,\n  backend: MathBackendCPU\n}): TensorInfo {\n  const {inputs, attrs, backend} = args;\n  const {x} = inputs;\n  const {perm} = attrs;\n\n  assertNotComplex(x, 'transpose');\n\n  const xRank = x.shape.length;\n\n  const newShape: number[] = new Array(xRank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = x.shape[perm[i]];\n  }\n\n  const values = backend.data.get(x.dataId).values as TypedArray;\n  const result = transposeImpl(values, x.shape, x.dtype, perm, newShape);\n\n  const dataId = backend.write(result, newShape, x.dtype);\n  return {dataId, shape: newShape, dtype: x.dtype};\n}\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'cpu',\n  kernelFunc: transpose as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, KernelFunc, ReshapeAttrs, ReshapeInputs, SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs, TensorInfo, TransposeAttrs, TransposeInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {padV2Config} from './PadV2';\nimport {reshape} from './Reshape';\nimport {transpose} from './Transpose';\n\nexport function spaceToBatchND(args: {\n  inputs: SpaceToBatchNDInputs,\n  backend: MathBackendCPU,\n  attrs: SpaceToBatchNDAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {blockShape, paddings} = attrs;\n\n  assertNotComplex([x], 'spaceToBatchND');\n\n  const prod = util.sizeFromShape(blockShape);\n\n  const completePaddings: Array<[number, number]> = [[0, 0]];\n  completePaddings.push(...(paddings as Array<[number, number]>));\n\n  for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n    completePaddings.push([0, 0]);\n  }\n\n  const paddedX = padV2Config.kernelFunc({\n    inputs: {x},\n    backend,\n    attrs: {paddings: completePaddings, constantValue: 0}\n  }) as TensorInfo;\n\n  const reshapedPaddedShape =\n      backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n  const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n      reshapedPaddedShape.length, blockShape.length, false);\n\n  const flattenShape =\n      backend_util.getReshapedPermuted(paddedX.shape, blockShape, prod, false);\n\n  const reshapeInputs: ReshapeInputs = {x: paddedX};\n  const reshapeAttrs: ReshapeAttrs = {shape: reshapedPaddedShape};\n  const paddedXReshaped =\n      reshape({inputs: reshapeInputs, backend, attrs: reshapeAttrs});\n\n  const transposeInputs: TransposeInputs = {x: paddedXReshaped};\n  const transposeAttrs:\n      TransposeAttrs = {perm: permutedReshapedPaddedPermutation};\n  const paddedXT =\n      transpose({inputs: transposeInputs, backend, attrs: transposeAttrs});\n\n  const resultReshapeInputs: ReshapeInputs = {x: paddedXT};\n  const resultReshapeAttrs: ReshapeAttrs = {shape: flattenShape};\n  const result = reshape(\n      {inputs: resultReshapeInputs, backend, attrs: resultReshapeAttrs});\n\n  backend.disposeIntermediateTensorInfo(paddedX);\n  backend.disposeIntermediateTensorInfo(paddedXReshaped);\n  backend.disposeIntermediateTensorInfo(paddedXT);\n\n  return result;\n}\n\nexport const spaceToBatchNDConfig: KernelConfig = {\n  kernelName: SpaceToBatchND,\n  backendName: 'cpu',\n  kernelFunc: spaceToBatchND as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sqrt} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sqrtKernelFunc = unaryKernelFunc(Sqrt, (xi) => Math.sqrt(xi));\n\nexport const sqrtConfig: KernelConfig = {\n  kernelName: Sqrt,\n  backendName: 'cpu',\n  kernelFunc: sqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square, SquareInputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend}) => {\n    const {x} = inputs as SquareInputs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'square');\n\n    const values = cpuBackend.data.get(x.dataId).values as Float32Array;\n    const newValues = new Float32Array(values.length);\n    for (let i = 0; i < values.length; ++i) {\n      const value = values[i];\n      newValues[i] = value * value;\n    }\n    const dataId = cpuBackend.write(newValues, x.shape, x.dtype);\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const squaredDifferenceImpl = createSimpleBinaryKernelImpl(((a, b) => {\n  const diff = a - b;\n  return diff * diff;\n}));\nexport const squaredDifference =\n    binaryKernelFunc(SquaredDifference, squaredDifferenceImpl);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'cpu',\n  kernelFunc: squaredDifference\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Step, StepAttrs} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const stepKernelFunc = unaryKernelFunc(Step, (xi, attrs) => {\n  const stepAttrs = attrs as {} as StepAttrs;\n  if (isNaN(xi)) {\n    return NaN;\n  } else {\n    return xi > 0 ? 1 : stepAttrs.alpha;\n  }\n});\n\nexport const stepConfig: KernelConfig = {\n  kernelName: Step,\n  backendName: 'cpu',\n  kernelFunc: stepKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanKernelFunc = unaryKernelFunc(Tan, (xi) => Math.tan(xi));\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'cpu',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tanh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanhKernelFunc = unaryKernelFunc(Tanh, (xi) => Math.tanh(xi));\n\nexport const tanhConfig: KernelConfig = {\n  kernelName: Tanh,\n  backendName: 'cpu',\n  kernelFunc: tanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {uniqueImpl} from './Unique_impl';\n\nexport function unique(\n    args: {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendCPU}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  const values = backend.data.get(x.dataId).values;\n  const {outputValues, outputShape, indices} =\n      uniqueImpl(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'cpu',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// We explicitly import the modular kernels so they get registered in the\n// global registry when we compile the library. A modular build would replace\n// the contents of this file and import only the kernels that are needed.\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {absConfig} from './kernels/Abs';\nimport {acosConfig} from './kernels/Acos';\nimport {acoshConfig} from './kernels/Acosh';\nimport {addConfig} from './kernels/Add';\nimport {asinConfig} from './kernels/Asin';\nimport {asinhConfig} from './kernels/Asinh';\nimport {atanConfig} from './kernels/Atan';\nimport {atanhConfig} from './kernels/Atanh';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {castConfig} from './kernels/Cast';\nimport {ceilConfig} from './kernels/Ceil';\nimport {clipConfig} from './kernels/Clip';\nimport {complexConfig} from './kernels/Complex';\nimport {concatConfig} from './kernels/Concat';\nimport {cosConfig} from './kernels/Cos';\nimport {coshConfig} from './kernels/Cosh';\nimport {dilation2dConfig} from './kernels/Dilation2D';\nimport {dilation2dBackpropFilterConfig} from './kernels/Dilation2DBackpropFilter';\nimport {dilation2dBackpropInputConfig} from './kernels/Dilation2DBackpropInput';\nimport {divConfig} from './kernels/Div';\nimport {eluConfig} from './kernels/Elu';\nimport {erfConfig} from './kernels/Erf';\nimport {expConfig} from './kernels/Exp';\nimport {expm1Config} from './kernels/Expm1';\nimport {fftConfig} from './kernels/FFT';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {floorConfig} from './kernels/Floor';\nimport {identityConfig} from './kernels/Identity';\nimport {ifftConfig} from './kernels/IFFT';\nimport {imagConfig} from './kernels/Imag';\nimport {isFiniteConfig} from './kernels/IsFinite';\nimport {isInfConfig} from './kernels/IsInf';\nimport {isNaNConfig} from './kernels/IsNaN';\nimport {logConfig} from './kernels/Log';\nimport {log1pConfig} from './kernels/Log1p';\nimport {logicalNotConfig} from './kernels/LogicalNot';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {multiplyConfig} from './kernels/Multiply';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {notEqualConfig} from './kernels/NotEqual';\nimport {padV2Config} from './kernels/PadV2';\nimport {realConfig} from './kernels/Real';\nimport {reciprocalConfig} from './kernels/Reciprocal';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {roundConfig} from './kernels/Round';\nimport {rsqrtConfig} from './kernels/Rsqrt';\nimport {seluConfig} from './kernels/Selu';\nimport {sigmoidConfig} from './kernels/Sigmoid';\nimport {signConfig} from './kernels/Sign';\nimport {sinConfig} from './kernels/Sin';\nimport {sinhConfig} from './kernels/Sinh';\nimport {sliceConfig} from './kernels/Slice';\nimport {softplusConfig} from './kernels/Softplus';\nimport {spaceToBatchNDConfig} from './kernels/SpaceToBatchND';\nimport {sqrtConfig} from './kernels/Sqrt';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {stepConfig} from './kernels/Step';\nimport {subConfig} from './kernels/Sub';\nimport {tanConfig} from './kernels/Tan';\nimport {tanhConfig} from './kernels/Tanh';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  absConfig,\n  acosConfig,\n  acoshConfig,\n  addConfig,\n  asinConfig,\n  asinhConfig,\n  atanConfig,\n  atanhConfig,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  castConfig,\n  ceilConfig,\n  clipConfig,\n  complexConfig,\n  concatConfig,\n  cosConfig,\n  coshConfig,\n  dilation2dConfig,\n  dilation2dBackpropInputConfig,\n  dilation2dBackpropFilterConfig,\n  divConfig,\n  eluConfig,\n  erfConfig,\n  expConfig,\n  expm1Config,\n  fftConfig,\n  flipLeftRightConfig,\n  floorConfig,\n  identityConfig,\n  ifftConfig,\n  imagConfig,\n  isFiniteConfig,\n  isInfConfig,\n  isNaNConfig,\n  logConfig,\n  log1pConfig,\n  logicalNotConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  maxConfig,\n  multiplyConfig,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  notEqualConfig,\n  padV2Config,\n  realConfig,\n  reciprocalConfig,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  roundConfig,\n  rsqrtConfig,\n  seluConfig,\n  sigmoidConfig,\n  signConfig,\n  sinConfig,\n  sinhConfig,\n  sliceConfig,\n  softplusConfig,\n  spaceToBatchNDConfig,\n  sqrtConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  stepConfig,\n  subConfig,\n  tanConfig,\n  tanhConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst contexts: {[key: string]: WebGLRenderingContext} = {};\n\nconst WEBGL_ATTRIBUTES: WebGLContextAttributes = {\n  alpha: false,\n  antialias: false,\n  premultipliedAlpha: false,\n  preserveDrawingBuffer: false,\n  depth: false,\n  stencil: false,\n  failIfMajorPerformanceCaveat: true\n};\n\nexport function setWebGLContext(\n    webGLVersion: number, gl: WebGLRenderingContext) {\n  contexts[webGLVersion] = gl;\n}\n\nexport function getWebGLContext(webGLVersion: number): WebGLRenderingContext {\n  if (!(webGLVersion in contexts)) {\n    const newCtx = getWebGLRenderingContext(webGLVersion);\n    if (newCtx !== null) {\n      contexts[webGLVersion] = newCtx;\n    } else {\n      console.log('Could not get context for WebGL version', webGLVersion);\n      return null;\n    }\n  }\n  const gl = contexts[webGLVersion];\n  if (gl.isContextLost()) {\n    delete contexts[webGLVersion];\n    return getWebGLContext(webGLVersion);\n  }\n\n  gl.disable(gl.DEPTH_TEST);\n  gl.disable(gl.STENCIL_TEST);\n  gl.disable(gl.BLEND);\n  gl.disable(gl.DITHER);\n  gl.disable(gl.POLYGON_OFFSET_FILL);\n  gl.disable(gl.SAMPLE_COVERAGE);\n  gl.enable(gl.SCISSOR_TEST);\n  gl.enable(gl.CULL_FACE);\n  gl.cullFace(gl.BACK);\n\n  return contexts[webGLVersion];\n}\n\nfunction createCanvas(webGLVersion: number) {\n  if (typeof OffscreenCanvas !== 'undefined' && webGLVersion === 2) {\n    return new OffscreenCanvas(300, 150);\n  } else if (typeof document !== 'undefined') {\n    return document.createElement('canvas');\n  } else {\n    throw new Error('Cannot create a canvas in this context');\n  }\n}\n\nfunction getWebGLRenderingContext(webGLVersion: number): WebGLRenderingContext {\n  if (webGLVersion !== 1 && webGLVersion !== 2) {\n    throw new Error('Cannot get WebGL rendering context, WebGL is disabled.');\n  }\n  const canvas = createCanvas(webGLVersion);\n\n  canvas.addEventListener('webglcontextlost', (ev: Event) => {\n    ev.preventDefault();\n    delete contexts[webGLVersion];\n  }, false);\n  if (webGLVersion === 1) {\n    return (canvas.getContext('webgl', WEBGL_ATTRIBUTES) ||\n            canvas.getContext('experimental-webgl', WEBGL_ATTRIBUTES)) as\n        WebGLRenderingContext;\n  }\n  return canvas.getContext('webgl2', WEBGL_ATTRIBUTES) as WebGLRenderingContext;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {backend_util, DataId, DataType, Tensor, util} from '@tensorflow/tfjs-core';\n\nexport enum PackingScheme {\n  /**\n   * All values in a single texel are densely packed without any constraints.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 4]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   010|011   020|021\n   * -------   -------   -------\n   * 002|003   012|013   022|023\n   *\n   * 100|101   110|111   120|121\n   * -------   -------   -------\n   * 102|103   112|113   122|123\n   *\n   */\n  DENSE,\n\n  /**\n   * Single texels contain only values from the same batch, and from adjacent\n   * rows and columns.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 5]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   002|003   004|xxx   020|021   022|023   024|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   * 100|101   102|103   104|xxx   120|121   122|123   124|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   */\n  SHARED_BATCH\n}\n\nexport enum TextureUsage {\n  RENDER,\n  UPLOAD,\n  PIXELS,\n  DOWNLOAD\n}\n\nexport enum PhysicalTextureType {\n  UNPACKED_FLOAT16,\n  UNPACKED_FLOAT32,\n  PACKED_4X1_UNSIGNED_BYTE,\n  PACKED_2X2_FLOAT32,\n  PACKED_2X2_FLOAT16\n}\n\nexport interface TextureData {\n  // Required.\n  shape: number[];\n  dtype: DataType;\n\n  // Optional.\n  values?: backend_util.BackendValues;\n  texture?: WebGLTexture;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensors field. When this is defined, texture will be null.\n  complexTensors?: {real: Tensor, imag: Tensor};\n  /** [rows, columns] shape of the texture. */\n  texShape?: [number, number];\n  usage?: TextureUsage;\n  isPacked?: boolean;\n\n  refCount: number;\n\n  // Available when the tensor has been sliced.\n  slice?: {\n    // Offset in the 'flat index' space.\n    flatOffset: number;\n    // Used for counting how many sliced tensors point to the same texture.\n    origDataId: DataId;\n  };\n}\n\nexport function getUnpackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns, rows];\n}\n\nexport function getUnpackedArraySizeFromMatrixSize(\n    matrixSize: number, channelsPerTexture: number): number {\n  return matrixSize * channelsPerTexture;\n}\n\nexport function getColorMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns * 4, rows];\n}\n\n/**\n * Get shape for densely packed RGBA texture.\n */\nexport function getDenseTexShape(shape: number[]): [number, number] {\n  const size = util.sizeFromShape(shape);\n  const texelsNeeded = Math.ceil(size / 4);\n  return util.sizeToSquarishShape(texelsNeeded);\n}\n\nexport function getMatrixSizeFromUnpackedArraySize(\n    unpackedSize: number, channelsPerTexture: number): number {\n  if (unpackedSize % channelsPerTexture !== 0) {\n    throw new Error(\n        `unpackedSize (${unpackedSize}) must be a multiple of ` +\n        `${channelsPerTexture}`);\n  }\n  return unpackedSize / channelsPerTexture;\n}\n\nexport function decodeMatrixFromUnpackedColorRGBAArray(\n    unpackedArray: Float32Array, matrix: Float32Array, channels: number) {\n  const requiredSize = unpackedArray.length * channels / 4;\n  if (matrix.length < requiredSize) {\n    throw new Error(\n        `matrix length (${matrix.length}) must be >= ${requiredSize}`);\n  }\n  let dst = 0;\n  for (let src = 0; src < unpackedArray.length; src += 4) {\n    for (let c = 0; c < channels; c++) {\n      matrix[dst++] = unpackedArray[src + c];\n    }\n  }\n}\n\nexport function getPackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [\n    Math.max(1, Math.ceil(columns / 2)), Math.max(1, Math.ceil(rows / 2))\n  ];\n}\n\nexport function getPackedRGBAArraySizeFromMatrixShape(\n    rows: number, columns: number): number {\n  const [w, h] = getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return w * h * 4;\n}\n\nexport interface TextureConfig {\n  internalFormatFloat: number;\n  textureFormatFloat: number;\n  internalFormatPackedHalfFloat: number;\n  internalFormatHalfFloat: number;\n  internalFormatPackedFloat: number;\n\n  // The format to use during a gl.readPixels call.\n  downloadTextureFormat: number;\n  // How many channels need to be unpacked after a gl.readPixels call.\n  downloadUnpackNumChannels: number;\n\n  defaultNumChannels: number;\n  textureTypeHalfFloat: number;\n  textureTypeFloat: number;\n}\n\nexport function getTextureConfig(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension?: any): TextureConfig {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n\n  let internalFormatFloat: number;\n  let internalFormatHalfFloat: number;\n  let internalFormatPackedHalfFloat: number;\n  let internalFormatPackedFloat: number;\n  let textureFormatFloat: number;\n\n  let downloadTextureFormat: number;\n  let downloadUnpackNumChannels: number;\n\n  let defaultNumChannels: number;\n  let textureTypeHalfFloat: number;\n  let textureTypeFloat: number;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    internalFormatFloat = glany.R32F;\n    internalFormatHalfFloat = glany.R16F;\n    internalFormatPackedHalfFloat = glany.RGBA16F;\n    internalFormatPackedFloat = glany.RGBA32F;\n    textureFormatFloat = glany.RED;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 1;\n    textureTypeHalfFloat = glany.HALF_FLOAT;\n    textureTypeFloat = glany.FLOAT;\n  } else {\n    internalFormatFloat = gl.RGBA;\n    internalFormatHalfFloat = gl.RGBA;\n    internalFormatPackedHalfFloat = gl.RGBA;\n    internalFormatPackedFloat = glany.RGBA;\n    textureFormatFloat = gl.RGBA;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 4;\n    textureTypeHalfFloat = textureHalfFloatExtension != null ?\n        textureHalfFloatExtension.HALF_FLOAT_OES :\n        null;\n    textureTypeFloat = gl.FLOAT;\n  }\n  downloadTextureFormat = gl.RGBA;\n\n  return {\n    internalFormatFloat,\n    internalFormatHalfFloat,\n    internalFormatPackedHalfFloat,\n    internalFormatPackedFloat,\n    textureFormatFloat,\n    downloadTextureFormat,\n    downloadUnpackNumChannels,\n    defaultNumChannels,\n    textureTypeHalfFloat,\n    textureTypeFloat\n  };\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext} from './canvas_util';\nimport {getTextureConfig} from './tex_util';\n\nexport function callAndCheck<T>(gl: WebGLRenderingContext, func: () => T): T {\n  const returnValue = func();\n  if (env().getBool('DEBUG')) {\n    checkWebGLError(gl);\n  }\n  return returnValue;\n}\n\nfunction checkWebGLError(gl: WebGLRenderingContext) {\n  const error = gl.getError();\n  if (error !== gl.NO_ERROR) {\n    throw new Error('WebGL Error: ' + getWebGLErrorMessage(gl, error));\n  }\n}\n\n// https://en.wikipedia.org/wiki/Half-precision_floating-point_format\nconst MIN_FLOAT16 = 5.96e-8;\nconst MAX_FLOAT16 = 65504;\n\nexport function canBeRepresented(num: number): boolean {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED') || num === 0 ||\n      (MIN_FLOAT16 < Math.abs(num) && Math.abs(num) < MAX_FLOAT16)) {\n    return true;\n  }\n  return false;\n}\n\nexport function getWebGLErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.NO_ERROR:\n      return 'NO_ERROR';\n    case gl.INVALID_ENUM:\n      return 'INVALID_ENUM';\n    case gl.INVALID_VALUE:\n      return 'INVALID_VALUE';\n    case gl.INVALID_OPERATION:\n      return 'INVALID_OPERATION';\n    case gl.INVALID_FRAMEBUFFER_OPERATION:\n      return 'INVALID_FRAMEBUFFER_OPERATION';\n    case gl.OUT_OF_MEMORY:\n      return 'OUT_OF_MEMORY';\n    case gl.CONTEXT_LOST_WEBGL:\n      return 'CONTEXT_LOST_WEBGL';\n    default:\n      return `Unknown error code ${status}`;\n  }\n}\n\nexport function getExtensionOrThrow(\n    gl: WebGLRenderingContext, extensionName: string): {} {\n  return throwIfNull<{}>(\n      gl, () => gl.getExtension(extensionName),\n      'Extension \"' + extensionName + '\" not supported on this browser.');\n}\n\nexport function createVertexShader(\n    gl: WebGLRenderingContext, vertexShaderSource: string): WebGLShader {\n  const vertexShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.VERTEX_SHADER),\n      'Unable to create vertex WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(vertexShader, vertexShaderSource));\n  callAndCheck(gl, () => gl.compileShader(vertexShader));\n  if (gl.getShaderParameter(vertexShader, gl.COMPILE_STATUS) === false) {\n    console.log(gl.getShaderInfoLog(vertexShader));\n    throw new Error('Failed to compile vertex shader.');\n  }\n  return vertexShader;\n}\n\nexport function createFragmentShader(\n    gl: WebGLRenderingContext, fragmentShaderSource: string): WebGLShader {\n  const fragmentShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.FRAGMENT_SHADER),\n      'Unable to create fragment WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(fragmentShader, fragmentShaderSource));\n  callAndCheck(gl, () => gl.compileShader(fragmentShader));\n  if (gl.getShaderParameter(fragmentShader, gl.COMPILE_STATUS) === false) {\n    logShaderSourceAndInfoLog(\n        fragmentShaderSource, gl.getShaderInfoLog(fragmentShader));\n    throw new Error('Failed to compile fragment shader.');\n  }\n  return fragmentShader;\n}\n\nconst lineNumberRegex = /ERROR: [0-9]+:([0-9]+):/g;\nfunction logShaderSourceAndInfoLog(\n    shaderSource: string, shaderInfoLog: string) {\n  const lineNumberRegexResult = lineNumberRegex.exec(shaderInfoLog);\n  if (lineNumberRegexResult == null) {\n    console.log(`Couldn't parse line number in error: ${shaderInfoLog}`);\n    console.log(shaderSource);\n    return;\n  }\n\n  const lineNumber = +lineNumberRegexResult[1];\n\n  const shaderLines = shaderSource.split('\\n');\n  const pad = shaderLines.length.toString().length + 2;\n  const linesWithLineNumbers = shaderLines.map(\n      (line, lineNumber) =>\n          util.rightPad((lineNumber + 1).toString(), pad) + line);\n  let maxLineLength = 0;\n  for (let i = 0; i < linesWithLineNumbers.length; i++) {\n    maxLineLength = Math.max(linesWithLineNumbers[i].length, maxLineLength);\n  }\n\n  const beforeErrorLines = linesWithLineNumbers.slice(0, lineNumber - 1);\n  const errorLine = linesWithLineNumbers.slice(lineNumber - 1, lineNumber);\n  const afterErrorLines = linesWithLineNumbers.slice(lineNumber);\n\n  console.log(beforeErrorLines.join('\\n'));\n  console.log(shaderInfoLog.split('\\n')[0]);\n  console.log(\n      `%c ${util.rightPad(errorLine[0], maxLineLength)}`,\n      'border:1px solid red; background-color:#e3d2d2; color:#a61717');\n  console.log(afterErrorLines.join('\\n'));\n}\n\nexport function createProgram(gl: WebGLRenderingContext): WebGLProgram {\n  return throwIfNull<WebGLProgram>(\n      gl, () => gl.createProgram(), 'Unable to create WebGLProgram.');\n}\n\nexport function linkProgram(gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.linkProgram(program));\n  if (gl.getProgramParameter(program, gl.LINK_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Failed to link vertex and fragment shaders.');\n  }\n}\n\nexport function validateProgram(\n    gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.validateProgram(program));\n  if (gl.getProgramParameter(program, gl.VALIDATE_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Shader program validation failed.');\n  }\n}\n\nexport function createStaticVertexBuffer(\n    gl: WebGLRenderingContext, data: Float32Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(gl, () => gl.bufferData(gl.ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function createStaticIndexBuffer(\n    gl: WebGLRenderingContext, data: Uint16Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl, () => gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function getNumChannels(): number {\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    return 1;\n  }\n  return 4;\n}\n\nexport function createTexture(gl: WebGLRenderingContext): WebGLTexture {\n  return throwIfNull<WebGLTexture>(\n      gl, () => gl.createTexture(), 'Unable to create WebGLTexture.');\n}\n\nexport function validateTextureSize(width: number, height: number) {\n  const maxTextureSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if ((width <= 0) || (height <= 0)) {\n    const requested = `[${width}x${height}]`;\n    throw new Error('Requested texture size ' + requested + ' is invalid.');\n  }\n  if ((width > maxTextureSize) || (height > maxTextureSize)) {\n    const requested = `[${width}x${height}]`;\n    const max = `[${maxTextureSize}x${maxTextureSize}]`;\n    throw new Error(\n        'Requested texture size ' + requested +\n        ' greater than WebGL maximum on this browser / GPU ' + max + '.');\n  }\n}\n\nexport function createFramebuffer(gl: WebGLRenderingContext): WebGLFramebuffer {\n  return throwIfNull<WebGLFramebuffer>(\n      gl, () => gl.createFramebuffer(), 'Unable to create WebGLFramebuffer.');\n}\n\nexport function bindVertexBufferToProgramAttribute(\n    gl: WebGLRenderingContext, program: WebGLProgram, attribute: string,\n    buffer: WebGLBuffer, arrayEntriesPerItem: number, itemStrideInBytes: number,\n    itemOffsetInBytes: number): boolean {\n  const loc = gl.getAttribLocation(program, attribute);\n  if (loc === -1) {\n    // The GPU compiler decided to strip out this attribute because it's unused,\n    // thus no need to bind.\n    return false;\n  }\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl,\n      () => gl.vertexAttribPointer(\n          loc, arrayEntriesPerItem, gl.FLOAT, false, itemStrideInBytes,\n          itemOffsetInBytes));\n  callAndCheck(gl, () => gl.enableVertexAttribArray(loc));\n  return true;\n}\n\nexport function bindTextureUnit(\n    gl: WebGLRenderingContext, texture: WebGLTexture, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n}\n\nexport function unbindTextureUnit(\n    gl: WebGLRenderingContext, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function getProgramUniformLocationOrThrow(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return throwIfNull<WebGLUniformLocation>(\n      gl, () => gl.getUniformLocation(program, uniformName),\n      'uniform \"' + uniformName + '\" not present in program.');\n}\n\nexport function getProgramUniformLocation(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return gl.getUniformLocation(program, uniformName);\n}\n\nexport function bindTextureToProgramUniformSampler(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    uniformSamplerLocation: WebGLUniformLocation, textureUnit: number) {\n  callAndCheck(gl, () => bindTextureUnit(gl, texture, textureUnit));\n  callAndCheck(gl, () => gl.uniform1i(uniformSamplerLocation, textureUnit));\n}\n\nexport function bindCanvasToFramebuffer(gl: WebGLRenderingContext) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n  callAndCheck(gl, () => gl.viewport(0, 0, gl.canvas.width, gl.canvas.height));\n  callAndCheck(gl, () => gl.scissor(0, 0, gl.canvas.width, gl.canvas.height));\n}\n\nexport function bindColorTextureToFramebuffer(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0));\n}\n\nexport function unbindColorTextureFromFramebuffer(\n    gl: WebGLRenderingContext, framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, null, 0));\n}\n\nexport function validateFramebuffer(gl: WebGLRenderingContext) {\n  const status = gl.checkFramebufferStatus(gl.FRAMEBUFFER);\n  if (status !== gl.FRAMEBUFFER_COMPLETE) {\n    throw new Error(\n        'Error binding framebuffer: ' + getFramebufferErrorMessage(gl, status));\n  }\n}\n\nexport function getFramebufferErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:\n      return 'FRAMEBUFFER_INCOMPLETE_DIMENSIONS';\n    case gl.FRAMEBUFFER_UNSUPPORTED:\n      return 'FRAMEBUFFER_UNSUPPORTED';\n    default:\n      return `unknown error ${status}`;\n  }\n}\n\nfunction throwIfNull<T>(\n    gl: WebGLRenderingContext, returnTOrNull: () => T | null,\n    failureMessage: string): T {\n  const tOrNull: T|null = callAndCheck(gl, () => returnTOrNull());\n  if (tOrNull == null) {\n    throw new Error(failureMessage);\n  }\n  return tOrNull;\n}\n\nfunction validateTextureUnit(gl: WebGLRenderingContext, textureUnit: number) {\n  const maxTextureUnit = gl.MAX_COMBINED_TEXTURE_IMAGE_UNITS - 1;\n  const glTextureUnit = textureUnit + gl.TEXTURE0;\n  if (glTextureUnit < gl.TEXTURE0 || glTextureUnit > maxTextureUnit) {\n    const textureUnitRange = `[gl.TEXTURE0, gl.TEXTURE${maxTextureUnit}]`;\n    throw new Error(`textureUnit must be in ${textureUnitRange}.`);\n  }\n}\n\nexport function getBatchDim(shape: number[], dimsToSkip = 2): number {\n  return util.sizeFromShape(shape.slice(0, shape.length - dimsToSkip));\n}\n\nexport function getRowsCols(shape: number[]): [number, number] {\n  if (shape.length === 0) {\n    throw Error('Cannot get rows and columns of an empty shape array.');\n  }\n\n  return [\n    shape.length > 1 ? shape[shape.length - 2] : 1, shape[shape.length - 1]\n  ];\n}\n\nexport function getShapeAs3D(shape: number[]): [number, number, number] {\n  let shapeAs3D: [number, number, number] = [1, 1, 1];\n  const isScalar = shape.length === 0 || (shape.length === 1 && shape[0] === 1);\n  if (!isScalar) {\n    shapeAs3D =\n        [getBatchDim(shape), ...getRowsCols(shape)] as [number, number, number];\n  }\n  return shapeAs3D;\n}\n\nexport function getTextureShapeFromLogicalShape(\n    logShape: number[], isPacked = false): [number, number] {\n  let maxTexSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if (isPacked) {\n    maxTexSize = maxTexSize * 2;\n\n    // This logic ensures we accurately count the number of packed texels needed\n    // to accommodate the tensor. We can only pack values in the same texel if\n    // they are from adjacent pairs of rows/cols within the same batch. So if a\n    // tensor has 3 rows, we pretend it has 4 rows in order to account for the\n    // fact that the texels containing the third row are half empty.\n    logShape = logShape.map(\n        (d, i) => i >= logShape.length - 2 ?\n            util.nearestLargerEven(logShape[i]) :\n            logShape[i]);\n\n    // Packed texture height is at least 2 (the channel height of a single\n    // texel).\n    if (logShape.length === 1) {\n      logShape = [2, logShape[0]];\n    }\n  }\n\n  // If logical shape is 2, we don't squeeze, since we want to match physical.\n  if (logShape.length !== 2) {\n    const squeezeResult = util.squeezeShape(logShape);\n    logShape = squeezeResult.newShape;\n  }\n\n  let size = util.sizeFromShape(logShape);\n  if (logShape.length <= 1 && size <= maxTexSize) {\n    return [1, size];\n  } else if (\n      logShape.length === 2 && logShape[0] <= maxTexSize &&\n      logShape[1] <= maxTexSize) {\n    return logShape as [number, number];\n  } else if (\n      logShape.length === 3 && logShape[0] * logShape[1] <= maxTexSize &&\n      logShape[2] <= maxTexSize) {\n    return [logShape[0] * logShape[1], logShape[2]];\n  } else if (\n      logShape.length === 3 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2]];\n  } else if (\n      logShape.length === 4 &&\n      logShape[0] * logShape[1] * logShape[2] <= maxTexSize &&\n      logShape[3] <= maxTexSize) {\n    return [logShape[0] * logShape[1] * logShape[2], logShape[3]];\n  } else if (\n      logShape.length === 4 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] * logShape[3] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2] * logShape[3]];\n  } else {\n    if (isPacked) {\n      // For packed textures size equals the number of channels required to\n      // accommodate the texture data. However in order to squarify such that\n      // inner dimensions stay even, we rewrite size to equal the number of\n      // texels. Then in the return statement we rehydrate the squarified\n      // dimensions to channel units.\n\n      const batchDim = getBatchDim(logShape);\n      let rows = 2, cols = 2;\n      if (logShape.length) {\n        [rows, cols] = getRowsCols(logShape);\n      }\n      size = batchDim * (rows / 2) * (cols / 2);\n      return util.sizeToSquarishShape(size).map(d => d * 2) as [number, number];\n    }\n    return util.sizeToSquarishShape(size);\n  }\n}\n\nfunction isEven(n: number): boolean {\n  return n % 2 === 0;\n}\n\n/**\n * This determines whether reshaping a packed texture requires rearranging\n * the data within the texture, assuming 2x2 packing.\n */\nexport function isReshapeFree(shape1: number[], shape2: number[]): boolean {\n  shape1 = shape1.slice(-2);\n  shape2 = shape2.slice(-2);\n\n  if (util.arraysEqual(shape1, shape2)) {\n    return true;\n  }\n\n  if (!shape1.length || !shape2.length) {  // One of the shapes is a scalar.\n    return true;\n  }\n\n  if (shape1[0] === 0 || shape1[1] === 0 || shape2[0] === 0 ||\n      shape2[1] === 0) {\n    return true;\n  }\n\n  if (shape1.length !== shape2.length) {  // One of the shapes is a vector.\n    const shape1Cols = shape1.slice(-1)[0];\n    const shape2Cols = shape2.slice(-1)[0];\n    if (shape1Cols === shape2Cols) {\n      return true;\n    }\n\n    if (isEven(shape1Cols) && isEven(shape2Cols) &&\n        (shape1[0] === 1 || shape2[0] === 1)) {\n      return true;\n    }\n  }\n  return shape1[1] === shape2[1] && isEven(shape1[0]) && isEven(shape2[0]);\n}\n\n// We cache webgl params because the environment gets reset between\n// unit tests and we don't want to constantly query the WebGLContext for\n// MAX_TEXTURE_SIZE.\nlet MAX_TEXTURE_SIZE: number;\nlet MAX_TEXTURES_IN_SHADER: number;\n\nexport function getWebGLMaxTextureSize(webGLVersion: number): number {\n  if (MAX_TEXTURE_SIZE == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURE_SIZE = gl.getParameter(gl.MAX_TEXTURE_SIZE);\n  }\n  return MAX_TEXTURE_SIZE;\n}\n\nexport function resetMaxTextureSize() {\n  MAX_TEXTURE_SIZE = null;\n}\nexport function resetMaxTexturesInShader() {\n  MAX_TEXTURES_IN_SHADER = null;\n}\n\nexport function getMaxTexturesInShader(webGLVersion: number): number {\n  if (MAX_TEXTURES_IN_SHADER == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURES_IN_SHADER = gl.getParameter(gl.MAX_TEXTURE_IMAGE_UNITS);\n  }\n  // We cap at 16 to avoid spurious runtime \"memory exhausted\" error.\n  return Math.min(16, MAX_TEXTURES_IN_SHADER);\n}\n\nexport function getWebGLDisjointQueryTimerVersion(webGLVersion: number):\n    number {\n  if (webGLVersion === 0) {\n    return 0;\n  }\n\n  let queryTimerVersion: number;\n  const gl = getWebGLContext(webGLVersion);\n\n  if (hasExtension(gl, 'EXT_disjoint_timer_query_webgl2') &&\n      webGLVersion === 2) {\n    queryTimerVersion = 2;\n  } else if (hasExtension(gl, 'EXT_disjoint_timer_query')) {\n    queryTimerVersion = 1;\n  } else {\n    queryTimerVersion = 0;\n  }\n  return queryTimerVersion;\n}\n\nexport function hasExtension(gl: WebGLRenderingContext, extensionName: string) {\n  const ext = gl.getExtension(extensionName);\n  return ext != null;\n}\n\nexport function isWebGLVersionEnabled(webGLVersion: 1|2) {\n  try {\n    const gl = getWebGLContext(webGLVersion);\n    if (gl != null) {\n      return true;\n    }\n  } catch (e) {\n    console.log('Error when getting WebGL context: ', e);\n    return false;\n  }\n  return false;\n}\n\nexport function isCapableOfRenderingToFloatTexture(webGLVersion: number):\n    boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n  } else {\n    if (!hasExtension(gl, 'EXT_color_buffer_float')) {\n      return false;\n    }\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\n/**\n * Check if we can download values from a float/half-float texture.\n *\n * Note that for performance reasons we use binding a texture to a framebuffer\n * as a proxy for ability to download float values later using readPixels. The\n * texture params of this texture will not match those in readPixels exactly\n * but if we are unable to bind some kind of float texture to the frameBuffer\n * then we definitely will not be able to read float values from it.\n */\nexport function isDownloadFloatTextureEnabled(webGLVersion: number): boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n    if (!hasExtension(gl, 'WEBGL_color_buffer_float')) {\n      return false;\n    }\n  } else {\n    if (hasExtension(gl, 'EXT_color_buffer_float')) {\n      return createFloatTextureAndBindToFramebuffer(gl);\n    }\n\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (hasExtension(gl, COLOR_BUFFER_HALF_FLOAT)) {\n      const textureHalfFloatExtension =\n          gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      return createHalfFloatTextureAndBindToFramebuffer(\n          gl, textureHalfFloatExtension);\n    }\n\n    return false;\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\nfunction createFloatTextureAndBindToFramebuffer(gl: WebGLRenderingContext):\n    boolean {\n  const texConfig = getTextureConfig(gl);\n\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nfunction createHalfFloatTextureAndBindToFramebuffer(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension: any): boolean {\n  const texConfig = getTextureConfig(gl, textureHalfFloatExtension);\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatHalfFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeHalfFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nexport function isWebGLFenceEnabled(webGLVersion: number) {\n  if (webGLVersion !== 2) {\n    return false;\n  }\n  const gl = getWebGLContext(webGLVersion);\n\n  // tslint:disable-next-line:no-any\n  const isEnabled = (gl as any).fenceSync != null;\n  return isEnabled;\n}\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${opName} does not support complex64 tensors ` +\n              'in the WebGL backend.');\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {device_util, env} from '@tensorflow/tfjs-core';\n\nimport {getMaxTexturesInShader, getWebGLDisjointQueryTimerVersion, getWebGLMaxTextureSize, isCapableOfRenderingToFloatTexture, isDownloadFloatTextureEnabled, isWebGLFenceEnabled, isWebGLVersionEnabled} from './webgl_util';\n\nconst ENV = env();\n\n/**\n * This file contains WebGL-specific flag registrations.\n */\n\n/**\n * True if WebGL is supported.\n */\nENV.registerFlag('HAS_WEBGL', () => ENV.getNumber('WEBGL_VERSION') > 0);\n\n/** 0: No WebGL, 1: WebGL 1.0, 2: WebGL 2.0. */\nENV.registerFlag('WEBGL_VERSION', () => {\n  if (isWebGLVersionEnabled(2)) {\n    return 2;\n  } else if (isWebGLVersionEnabled(1)) {\n    return 1;\n  }\n  return 0;\n});\n\n/** Whether to check for numerical representation problems. */\nENV.registerFlag('WEBGL_CHECK_NUMERICAL_PROBLEMS', () => false);\n\nENV.registerFlag(\n    'WEBGL_BUFFER_SUPPORTED', () => ENV.get('WEBGL_VERSION') === 2);\n\n/** Whether the WebGL backend will sometimes forward ops to the CPU. */\nENV.registerFlag('WEBGL_CPU_FORWARD', () => true);\n\n/** Whether the WebGL backend will always use f16 textures for rendering. */\nENV.registerFlag('WEBGL_FORCE_F16_TEXTURES', () => false);\n\n/** Whether to turn all packing related flags on. */\nENV.registerFlag('WEBGL_PACK', () => ENV.getBool('HAS_WEBGL'));\n\n/** Whether we will pack the batchnormalization op. */\nENV.registerFlag('WEBGL_PACK_NORMALIZATION', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the clip op. */\nENV.registerFlag('WEBGL_PACK_CLIP', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the depthwise conv op. */\n// TODO: https://github.com/tensorflow/tfjs/issues/1679\nENV.registerFlag('WEBGL_PACK_DEPTHWISECONV', () => false);\n\n/** Whether we will pack binary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_BINARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack unary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_UNARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack array ops. */\nENV.registerFlag(\n    'WEBGL_PACK_ARRAY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack image ops. */\nENV.registerFlag(\n    'WEBGL_PACK_IMAGE_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack reduce ops. */\nENV.registerFlag('WEBGL_PACK_REDUCE', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether packed WebGL kernels lazily unpack their outputs. */\nENV.registerFlag('WEBGL_LAZILY_UNPACK', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will use the im2col algorithm to speed up convolutions. */\nENV.registerFlag('WEBGL_CONV_IM2COL', () => ENV.getBool('WEBGL_PACK'));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURE_SIZE',\n    () => getWebGLMaxTextureSize(ENV.getNumber('WEBGL_VERSION')));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURES_IN_SHADER',\n    () => getMaxTexturesInShader(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * The disjoint_query_timer extension version.\n * 0: disabled, 1: EXT_disjoint_timer_query, 2:\n * EXT_disjoint_timer_query_webgl2.\n * In Firefox with WebGL 2.0,\n * EXT_disjoint_timer_query_webgl2 is not available, so we must use the\n * WebGL 1.0 extension.\n */\nENV.registerFlag('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION', () => {\n  const webGLVersion = ENV.getNumber('WEBGL_VERSION');\n\n  if (webGLVersion === 0) {\n    return 0;\n  }\n  return getWebGLDisjointQueryTimerVersion(webGLVersion);\n});\n\n/**\n * Whether the timer object from the disjoint_query_timer extension gives\n * timing information that is reliable.\n */\nENV.registerFlag(\n    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE',\n    () => ENV.getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0 &&\n        !device_util.isMobile());\n\n/**\n * Whether the device is physically capable of rendering to float32 textures.\n */\nENV.registerFlag(\n    'WEBGL_RENDER_FLOAT32_CAPABLE',\n    () => isCapableOfRenderingToFloatTexture(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Whether rendering to float32 textures is enabled. If disabled, renders to\n * float16 textures.\n */\nENV.registerFlag('WEBGL_RENDER_FLOAT32_ENABLED', () => {\n  return ENV.getBool('WEBGL_FORCE_F16_TEXTURES') ?\n      false :\n      ENV.getBool('WEBGL_RENDER_FLOAT32_CAPABLE');\n});\n\n/**\n * Whether downloading float textures is enabled (16 or 32 bit). If disabled,\n * uses IEEE 754 encoding of the float32 values to 4 uint8 when downloading.\n */\nENV.registerFlag(\n    'WEBGL_DOWNLOAD_FLOAT_ENABLED',\n    () => isDownloadFloatTextureEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/** Whether the fence API is available. */\nENV.registerFlag(\n    'WEBGL_FENCE_API_ENABLED',\n    () => isWebGLFenceEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Tensors with size <= than this will be uploaded as uniforms, not textures.\n */\nENV.registerFlag('WEBGL_SIZE_UPLOAD_UNIFORM', () => {\n  // Use uniform uploads only when 32bit floats are supported. In\n  // 16bit\n  // environments there are problems with comparing a 16bit texture value\n  // with a 32bit uniform value.\n  const useUniforms = ENV.getBool('WEBGL_RENDER_FLOAT32_ENABLED');\n  return useUniforms ? 4 : 0;\n});\n\n/**\n * If the total number of bytes allocated on the GPU is greater than this\n * number, we will aggressively delete textures upon disposal with\n * gl.deleteMatrixTexture, rather than making them available for reuse.\n *\n * Default value -1 indicates that we will never aggressively delete textures.\n */\nENV.registerFlag(\n    'WEBGL_DELETE_TEXTURE_THRESHOLD',\n    () => {\n      return -1;\n    },\n    threshold => {\n      if (threshold < 0 && threshold !== -1) {\n        throw new Error(\n            `WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never ` +\n            `delete) or at least 0, but got ${threshold}.`);\n      }\n    });\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Abs, util } from '@tensorflow/tfjs-core';\nexport function simpleAbsImpl(vals) {\n    const resultValues = new Float32Array(vals.length);\n    for (let i = 0; i < vals.length; ++i) {\n        resultValues[i] = Math.abs(vals[i]);\n    }\n    return resultValues;\n}\nexport const absKernelFunc = (args) => {\n    const { x } = args.inputs;\n    const cpuBackend = args.backend;\n    let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n    if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values;\n        resultValues = simpleAbsImpl(values);\n    }\n    else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals = cpuBackend.data.get(real.dataId).values;\n        const imagVals = cpuBackend.data.get(imag.dataId).values;\n        for (let i = 0; i < realVals.length; i++) {\n            const real = realVals[i];\n            const imag = imagVals[i];\n            resultValues[i] = Math.hypot(real, imag);\n        }\n    }\n    return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n};\nexport const absConfig = {\n    kernelName: Abs,\n    backendName: 'cpu',\n    kernelFunc: absKernelFunc,\n};\n//# sourceMappingURL=Abs.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { backend_util, util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op) {\n    return (aShape, bShape, aVals, bVals, dtype) => {\n        const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n        const resultRank = newShape.length;\n        const resultStrides = util.computeStrides(newShape);\n        const resultSize = util.sizeFromShape(newShape);\n        const result = util.getTypedArrayFromDType(dtype, resultSize);\n        const aRank = aShape.length;\n        const bRank = bShape.length;\n        const aStrides = util.computeStrides(aShape);\n        const bStrides = util.computeStrides(bShape);\n        const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n        const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n        if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n            for (let i = 0; i < result.length; ++i) {\n                result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n            }\n        }\n        else {\n            for (let i = 0; i < result.length; ++i) {\n                const loc = util.indexToLoc(i, resultRank, resultStrides);\n                const aLoc = loc.slice(-aRank);\n                aBroadcastDims.forEach(d => aLoc[d] = 0);\n                const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n                const bLoc = loc.slice(-bRank);\n                bBroadcastDims.forEach(d => bLoc[d] = 0);\n                const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n                result[i] = op(aVals[aIndex], bVals[bIndex]);\n            }\n        }\n        return [result, newShape];\n    };\n}\n//# sourceMappingURL=binary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Add } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal + bReal, imag: aImag + bImag };\n}));\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\nexport const addConfig = {\n    kernelName: Add,\n    backendName: 'cpu',\n    kernelFunc: add\n};\n//# sourceMappingURL=Add.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op) {\n    return (values, dtype, attrs) => {\n        const newValues = util.getTypedArrayFromDType(dtype, values.length);\n        for (let i = 0; i < values.length; ++i) {\n            newValues[i] = op(values[i], attrs);\n        }\n        return newValues;\n    };\n}\n//# sourceMappingURL=unary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Ceil } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\nexport const ceilConfig = {\n    kernelName: Ceil,\n    backendName: 'cpu',\n    kernelFunc: ceilKernelFunc,\n};\n//# sourceMappingURL=Ceil.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Exp } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\nexport const expConfig = {\n    kernelName: Exp,\n    backendName: 'cpu',\n    kernelFunc: expKernelFunc,\n};\n//# sourceMappingURL=Exp.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Expm1 } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\nexport const expm1Config = {\n    kernelName: Expm1,\n    backendName: 'cpu',\n    kernelFunc: expm1KernelFunc,\n};\n//# sourceMappingURL=Expm1.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Floor } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\nexport const floorConfig = {\n    kernelName: Floor,\n    backendName: 'cpu',\n    kernelFunc: floorKernelFunc,\n};\n//# sourceMappingURL=Floor.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Log } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\nexport const logConfig = {\n    kernelName: Log,\n    backendName: 'cpu',\n    kernelFunc: logKernelFunc,\n};\n//# sourceMappingURL=Log.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function maxImpl(aVals, reduceSize, outShape, dtype) {\n    const vals = util.getTypedArrayFromDType(dtype, util.sizeFromShape(outShape));\n    for (let i = 0; i < vals.length; ++i) {\n        const offset = i * reduceSize;\n        let max = aVals[offset];\n        for (let j = 0; j < reduceSize; ++j) {\n            const value = aVals[offset + j];\n            if (value > max) {\n                max = value;\n            }\n        }\n        vals[i] = max;\n    }\n    return vals;\n}\n//# sourceMappingURL=Max_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Multiply } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const multiplyImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n    };\n}));\nexport const multiply = binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\nexport const multiplyConfig = {\n    kernelName: Multiply,\n    backendName: 'cpu',\n    kernelFunc: multiply\n};\n//# sourceMappingURL=Multiply.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Rsqrt } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\nexport const rsqrtConfig = {\n    kernelName: Rsqrt,\n    backendName: 'cpu',\n    kernelFunc: rsqrtKernelFunc,\n};\n//# sourceMappingURL=Rsqrt.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Slice, slice_util, util } from '@tensorflow/tfjs-core';\nimport { assertNotComplex } from '../cpu_util';\nexport function sliceImpl(vals, begin, size, shape, dtype) {\n    const isContinous = slice_util.isSliceContinous(shape, begin, size);\n    const length = util.sizeFromShape(size);\n    const xStrides = util.computeStrides(shape);\n    if (isContinous) {\n        const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n        return vals.subarray(flatOffset, flatOffset + length);\n    }\n    const outVals = util.getTypedArrayFromDType(dtype, length);\n    for (let i = 0; i < length; ++i) {\n        const rank = size.length;\n        const strides = util.computeStrides(size);\n        const loc = util.indexToLoc(i, rank, strides);\n        const xLoc = loc.map((idx, j) => idx + begin[j]);\n        const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n        outVals[i] = vals[xIndex];\n    }\n    return outVals;\n}\nexport function slice(args) {\n    const { inputs, backend, attrs } = args;\n    const { x } = inputs;\n    const { begin, size } = attrs;\n    assertNotComplex(x, 'slice');\n    const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n    slice_util.assertParamsValid(x, $begin, $size);\n    const vals = backend.data.get(x.dataId).values;\n    const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n    return backend.makeTensorInfo($size, x.dtype, outVals);\n}\nexport const sliceConfig = {\n    kernelName: Slice,\n    backendName: 'cpu',\n    kernelFunc: slice\n};\n//# sourceMappingURL=Slice.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Sub } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const subImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal - bReal, imag: aImag - bImag };\n}));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\nexport const subConfig = {\n    kernelName: Sub,\n    backendName: 'cpu',\n    kernelFunc: sub\n};\n//# sourceMappingURL=Sub.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function transposeImpl(xVals, xShape, dtype, perm, newShape) {\n    const xRank = xShape.length;\n    const xSize = util.sizeFromShape(xShape);\n    const xStrides = util.computeStrides(xShape);\n    const newStrides = util.computeStrides(newShape);\n    const result = util.getTypedArrayFromDType(dtype, util.sizeFromShape(newShape));\n    for (let i = 0; i < xSize; ++i) {\n        const loc = util.indexToLoc(i, xRank, xStrides);\n        // Permute location.\n        const newLoc = new Array(loc.length);\n        for (let i = 0; i < newLoc.length; i++) {\n            newLoc[i] = loc[perm[i]];\n        }\n        const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n        result[newIndex] = xVals[i];\n    }\n    return result;\n}\n//# sourceMappingURL=Transpose_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { TensorBuffer, util } from '@tensorflow/tfjs-core';\nexport function uniqueImpl(values, axis, shape, dtype) {\n    // Normalize and validate axis.\n    const $axis = util.parseAxisParam(axis, shape)[0];\n    // Calculate the new shape that is suitable for extracting data along the\n    // given axis.\n    //\n    // The rank is 3.\n    // The size of the 1st dimension is the size of all the axes < the given axis.\n    // The size of the 2nd dimension is the same as the size of the given axis.\n    // The size of the 3rd dimension is the size of all the axes > the given axis.\n    //\n    // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n    // newShape would be: [2*3, 5, 4].\n    //\n    // Note that this is not the final output shape. This will be the shape for an\n    // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n    // values along the given axis. To demonstrate how it works, consider the\n    // following example:\n    //\n    // Input: a 3D tensor, with shape [1, 2, 3]\n    // [\n    //   [\n    //      [1,2,3],\n    //      [4,5,6]\n    //   ]\n    // ]\n    // Axis: 2 (the last axis).\n    // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n    //\n    // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n    // 1*2. The re-shaped data would look like:\n    //\n    // [\n    //   [\n    //     [1], [2], [3]\n    //   ],\n    //   [\n    //     [4], [5], [6]\n    //   ]\n    // ]\n    //\n    // Then, we can construct a 3-level nested loop by the following dimension\n    // order to extract the values along the axis (dimension1):\n    // i: dimension1       // 0,1,2 (newShape[1])\n    //   m: dimension0     // 0,1   (newShape[0])\n    //     n: dimension2   // 0     (newShape[2])\n    //\n    //                       m, i, n\n    //                      ---------\n    // Iteration 0: data at [0, 0, 0] => \"1\"\n    // Iteration 1: data at [1, 0, 0] => \"4\"\n    // We got [1,4].\n    // Iteration 2: data at [0, 1, 0] => \"2\"\n    // Iteration 3: data at [1, 1, 0] => \"5\"\n    // We got [2,5].\n    // Iteration 4: data at [0, 2, 0] => \"3\"\n    // Iteration 5: data at [1, 2, 0] => \"6\"\n    // We got [3,6].\n    const newShape = [1, shape[0], 1];\n    for (let i = 0; i < $axis; i++) {\n        newShape[0] *= shape[i];\n    }\n    newShape[1] = shape[$axis];\n    for (let i = $axis + 1; i < shape.length; i++) {\n        newShape[2] *= shape[i];\n    }\n    // A map from unique elements (their string representations) to their values\n    // in \"indices\" (below).\n    const uniqueElements = {};\n    // The indices of each unique element in the original tensor along the given\n    // axis. It is 1D and has the same size as the given axis.\n    const indices = new Int32Array(shape[$axis]);\n    // Create a buffer so we can easily extract value at a given location.\n    const inputBuffer = new TensorBuffer(newShape, dtype, values);\n    // The indices along the given axis that have unique elements. This is a\n    // de-duped version of \"indices\" above.\n    const uniqueIndices = [];\n    const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n    for (let i = 0; i < shape[$axis]; i++) {\n        // Extract values along the axis.\n        let element;\n        if (is1DTensor) {\n            // Fast path for 1D tensor input.\n            element = values[i].toString();\n        }\n        else {\n            const axisValues = [];\n            for (let m = 0; m < newShape[0]; m++) {\n                for (let n = 0; n < newShape[2]; n++) {\n                    axisValues.push(inputBuffer.get(m, i, n));\n                }\n            }\n            element = axisValues.join(',');\n        }\n        // Dedup and update various indices.\n        if (uniqueElements[element] !== undefined) {\n            indices[i] = uniqueElements[element];\n        }\n        else {\n            const uniqueIndex = Object.keys(uniqueElements).length;\n            uniqueElements[element] = uniqueIndex;\n            indices[i] = uniqueIndex;\n            uniqueIndices.push(i);\n        }\n    }\n    // Now we know where each of the unique elements are located along the axis\n    // (uniqueIndices). Extract them from input buffer and store them in the\n    // output buffer.\n    const outputTmpShape = newShape.slice();\n    outputTmpShape[1] = Object.keys(uniqueElements).length;\n    const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n    uniqueIndices.forEach((uniqueElementIndex, i) => {\n        for (let m = 0; m < newShape[0]; m++) {\n            for (let n = 0; n < newShape[2]; n++) {\n                outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n            }\n        }\n    });\n    // The output shape can be calculated from the input shape with the size of\n    // the given axis replaced by the number of unique elements along that axis.\n    const outputShape = shape.slice();\n    outputShape[$axis] = outputTmpShape[1];\n    return {\n        outputValues: outputBuffer.values,\n        outputShape,\n        indices,\n    };\n}\n//# sourceMappingURL=Unique_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import shared functionality from tfjs-backend-cpu without triggering\n// side effects.\n// tslint:disable-next-line: no-imports-from-dist\nimport * as shared from '@tensorflow/tfjs-backend-cpu/dist/shared';\n\nconst {\n  simpleAbsImpl: simpleAbsImplCPU,\n  addImpl: addImplCPU,\n  ceilImpl: ceilImplCPU,\n  expImpl: expImplCPU,\n  expm1Impl: expm1ImplCPU,\n  floorImpl: floorImplCPU,\n  logImpl: logImplCPU,\n  maxImpl: maxImplCPU,\n  multiplyImpl: multiplyImplCPU,\n  rsqrtImpl: rsqrtImplCPU,\n  sliceImpl: sliceImplCPU,\n  subImpl: subImplCPU,\n  transposeImpl: transposeImplCPU,\n  uniqueImpl: uniqueImplCPU,\n} = shared;\n\nexport {\n  simpleAbsImplCPU,\n  addImplCPU,\n  ceilImplCPU,\n  expImplCPU,\n  expm1ImplCPU,\n  logImplCPU,\n  multiplyImplCPU,\n  sliceImplCPU,\n  subImplCPU,\n  floorImplCPU,\n  maxImplCPU,\n  rsqrtImplCPU,\n  transposeImplCPU,\n  uniqueImplCPU,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`float v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        float result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`vec4 v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        vec4 result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ArgMinMaxProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo, op: 'max'|'min',\n      firstPass: boolean) {\n    const {windowSize, batchSize, outSize} = reduceInfo;\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    this.outputShape = [batchSize, outSize];\n    const compOp = (op === 'max') ? '>' : '<';\n    const indexSnippet = firstPass ?\n        'inOffset + i;' :\n        'round(getBestIndicesA(batch, inOffset + i));';\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        int bestIndex = inOffset;\n        float bestValue = getA(batch, bestIndex);\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          int inIdx = ${indexSnippet};\n          float candidate = getA(batch, inIdx);\n          if (candidate ${compOp} bestValue) {\n            bestValue = candidate;\n            bestIndex = inIdx;\n          }\n        }\n        setOutput(float(bestIndex));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport function getVecChannels(name: string, rank: number): string[] {\n  return ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank).map(d => `${name}.${d}`);\n}\n\nexport function getChannels(name: string, rank: number): string[] {\n  if (rank === 1) {\n    return [name];\n  }\n  return getVecChannels(name, rank);\n}\n\nexport function getSourceCoords(rank: number, dims: string[]): string {\n  if (rank === 1) {\n    return 'rc';\n  }\n\n  let coords = '';\n  for (let i = 0; i < rank; i++) {\n    coords += dims[i];\n    if (i < rank - 1) {\n      coords += ',';\n    }\n  }\n  return coords;\n}", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '@tensorflow/tfjs-core';\n\nexport type GLSL = {\n  version: string,\n  attribute: string,\n  varyingVs: string,\n  varyingFs: string,\n  texture2D: string,\n  output: string,\n  defineOutput: string,\n  defineSpecialNaN: string,\n  defineSpecialInf: string,\n  defineRound: string\n};\n\nexport function getGlslDifferences(): GLSL {\n  let version: string;\n  let attribute: string;\n  let varyingVs: string;\n  let varyingFs: string;\n  let texture2D: string;\n  let output: string;\n  let defineOutput: string;\n  let defineSpecialNaN: string;\n  let defineSpecialInf: string;\n  let defineRound: string;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    version = '#version 300 es';\n    attribute = 'in';\n    varyingVs = 'out';\n    varyingFs = 'in';\n    texture2D = 'texture';\n    output = 'outputColor';\n    defineOutput = 'out vec4 outputColor;';\n\n    // Use custom isnan definition to work across differences between\n    // implementations on various platforms. While this should happen in ANGLE\n    // we still see differences between android and windows (on chrome) when\n    // using isnan directly.\n    defineSpecialNaN = `\n      bool isnan_custom(float val) {\n        return (val > 0.0 || val < 0.0) ? false : val != 0.0;\n      }\n\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan_custom(val.x),\n          isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));\n      }\n\n      #define isnan(value) isnan_custom(value)\n    `;\n    // In webgl 2 we do not need to specify a custom isinf so there is no\n    // need for a special INFINITY constant.\n    defineSpecialInf = ``;\n    defineRound = `\n      #define round(value) newRound(value)\n      int newRound(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 newRound(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  } else {\n    version = '';\n    attribute = 'attribute';\n    varyingVs = 'varying';\n    varyingFs = 'varying';\n    texture2D = 'texture2D';\n    output = 'gl_FragColor';\n    defineOutput = '';\n    // WebGL1 has no built in isnan so we define one here.\n    defineSpecialNaN = `\n      #define isnan(value) isnan_custom(value)\n      bool isnan_custom(float val) {\n        return (val > 0. || val < 1. || val == 0.) ? false : true;\n      }\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));\n      }\n    `;\n    defineSpecialInf = `\n      uniform float INFINITY;\n\n      bool isinf(float val) {\n        return abs(val) == INFINITY;\n      }\n      bvec4 isinf(vec4 val) {\n        return equal(abs(val), vec4(INFINITY));\n      }\n    `;\n    defineRound = `\n      int round(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 round(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  }\n\n  return {\n    version,\n    attribute,\n    varyingVs,\n    varyingFs,\n    texture2D,\n    output,\n    defineOutput,\n    defineSpecialNaN,\n    defineSpecialInf,\n    defineRound\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\n/**\n * Produces GLSL code that derives logical coordinates from a flat\n * index. The code performs integer division with each stride and decrements\n * the index until the index equals the final dimension coordinate.\n */\nexport function getLogicalCoordinatesFromFlatIndex(\n    coords: string[], shape: number[], index = 'index'): string {\n  const strides = util.computeStrides(shape);\n  return strides\n      .map((stride, i) => {\n        const line1 = `int ${coords[i]} = ${index} / ${stride}`;\n        const line2 = i === strides.length - 1 ?\n            `int ${coords[i + 1]} = ${index} - ${coords[i]} * ${stride}` :\n            `index -= ${coords[i]} * ${stride}`;\n        return `${line1}; ${line2};`;\n      })\n      .join('');\n}\n\nfunction buildVec(x: string[]): string {\n  if (x.length === 1) {\n    return `${x[0]}`;\n  }\n  return `vec${x.length}(${x.join(',')})`;\n}\n\n/**\n * Produces GLSL code that computes the dot product of the input x and y\n * vectors. Handles splitting inputs into increments of vec4s when necessary.\n */\nexport function dotify(x: string[], y: string[]): string {\n  if (x.length !== y.length) {\n    throw new Error(\n        `Vectors to be dotted must be of the same length -` +\n        `got ${x.length} and ${y.length}`);\n  }\n\n  const slices: string[] = [];\n  const nearestVec4 = Math.floor(x.length / 4);\n  const nearestVec4Remainder = x.length % 4;\n\n  for (let i = 0; i < nearestVec4; i++) {\n    const xSlice = x.slice(i * 4, i * 4 + 4);\n    const ySlice = y.slice(i * 4, i * 4 + 4);\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  if (nearestVec4Remainder !== 0) {\n    let xSlice = x.slice(nearestVec4 * 4);\n    let ySlice = y.slice(nearestVec4 * 4);\n    if (xSlice.length === 1) {\n      xSlice = xSlice.map(d => `float(${d})`);\n      ySlice = ySlice.map(d => `float(${d})`);\n    }\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  return slices.map((d, i) => `dot(${d})`).join('+');\n}\n\n/**\n * Produces GLSL that computes the flat index from 3D coordinates.\n */\nexport function getFlatIndexFrom3D(shape: [number, number, number]): string {\n  const strides = util.computeStrides(shape).map(d => d.toString());\n\n  return `\n  int getFlatIndex(ivec3 coords) {\n    return coords.x * ${strides[0]} + coords.y * ${strides[1]} + coords.z;\n  }\n`;\n}\n\nexport const ENCODE_FLOAT_SNIPPET = `\n  const float FLOAT_MAX = 1.70141184e38;\n  const float FLOAT_MIN = 1.17549435e-38;\n\n  lowp vec4 encode_float(highp float v) {\n    if (isnan(v)) {\n      return vec4(255, 255, 255, 255);\n    }\n\n    highp float av = abs(v);\n\n    if(av < FLOAT_MIN) {\n      return vec4(0.0, 0.0, 0.0, 0.0);\n    } else if(v > FLOAT_MAX) {\n      return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;\n    } else if(v < -FLOAT_MAX) {\n      return vec4(0.0, 0.0,  128.0, 255.0) / 255.0;\n    }\n\n    highp vec4 c = vec4(0,0,0,0);\n\n    highp float e = floor(log2(av));\n    highp float m = exp2(fract(log2(av))) - 1.0;\n\n    c[2] = floor(128.0 * m);\n    m -= c[2] / 128.0;\n    c[1] = floor(32768.0 * m);\n    m -= c[1] / 32768.0;\n    c[0] = floor(8388608.0 * m);\n\n    highp float ebias = e + 127.0;\n    c[3] = floor(ebias / 2.0);\n    ebias -= c[3] * 2.0;\n    c[2] += floor(ebias) * 128.0;\n\n    c[3] += 128.0 * step(0.0, -v);\n\n    return c / 255.0;\n  }\n`;\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\nconst {getBroadcastDims} = backend_util;\nimport {getGlslDifferences, GLSL} from './glsl_version';\nimport * as shader_util from './shader_compiler_util';\n\nexport type ShapeInfo = {\n  logicalShape: number[],\n  texShape: [number, number],\n  isUniform: boolean,\n  isPacked: boolean,\n  flatOffset: number\n};\n\nexport type InputInfo = {\n  name: string,\n  shapeInfo: ShapeInfo\n};\n\nexport function makeShader(\n    inputsInfo: InputInfo[], outputShape: ShapeInfo, userCode: string,\n    usesPackedTextures: boolean): string {\n  const prefixSnippets: string[] = [];\n  inputsInfo.forEach(x => {\n    const size = util.sizeFromShape(x.shapeInfo.logicalShape);\n\n    // Snippet when we decided to upload the values as uniform.\n    if (x.shapeInfo.isUniform) {\n      prefixSnippets.push(\n          `uniform float ${x.name}${size > 1 ? `[${size}]` : ''};`);\n    } else {\n      prefixSnippets.push(`uniform sampler2D ${x.name};`);\n      prefixSnippets.push(`uniform int offset${x.name};`);\n    }\n  });\n  const inputPrefixSnippet = prefixSnippets.join('\\n');\n\n  const inputSamplingSnippet =\n      inputsInfo\n          .map(x => getInputSamplingSnippet(x, outputShape, usesPackedTextures))\n          .join('\\n');\n  const outTexShape = outputShape.texShape;\n  const glsl = getGlslDifferences();\n  const floatTextureSampleSnippet = getFloatTextureSampleSnippet(glsl);\n  let outputSamplingSnippet: string;\n  let floatTextureSetOutputSnippet: string;\n  let shaderPrefix = getShaderPrefix(glsl);\n\n  if (outputShape.isPacked) {\n    outputSamplingSnippet =\n        getPackedOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRGBASnippet(glsl);\n  } else {\n    outputSamplingSnippet =\n        getOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRSnippet(glsl);\n  }\n\n  if (usesPackedTextures) {\n    shaderPrefix += SHADER_PACKED_PREFIX;\n  }\n\n  const source = [\n    shaderPrefix, floatTextureSampleSnippet, floatTextureSetOutputSnippet,\n    inputPrefixSnippet, outputSamplingSnippet, inputSamplingSnippet, userCode\n  ].join('\\n');\n  return source;\n}\n\nfunction getSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getSamplerScalar(inInfo);\n    case 1:\n      return getSampler1D(inInfo);\n    case 2:\n      return getSampler2D(inInfo);\n    case 3:\n      return getSampler3D(inInfo);\n    case 4:\n      return getSampler4D(inInfo);\n    case 5:\n      return getSampler5D(inInfo);\n    case 6:\n      return getSampler6D(inInfo);\n    default:\n      throw new Error(\n          `${shape.length}-D input sampling` +\n          ` is not yet supported`);\n  }\n}\n\nfunction getPackedSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getPackedSamplerScalar(inInfo);\n    case 1:\n      return getPackedSampler1D(inInfo);\n    case 2:\n      return getPackedSampler2D(inInfo);\n    case 3:\n      return getPackedSampler3D(inInfo);\n    default:\n      return getPackedSamplerND(inInfo);\n  }\n}\n\nfunction getInputSamplingSnippet(\n    inInfo: InputInfo, outShapeInfo: ShapeInfo,\n    usesPackedTextures = false): string {\n  let res = '';\n  if (usesPackedTextures) {\n    res += getPackedSamplerFromInInfo(inInfo);\n  } else {\n    res += getSamplerFromInInfo(inInfo);\n  }\n\n  const inShape = inInfo.shapeInfo.logicalShape;\n  const outShape = outShapeInfo.logicalShape;\n  if (inShape.length <= outShape.length) {\n    if (usesPackedTextures) {\n      res += getPackedSamplerAtOutputCoords(inInfo, outShapeInfo);\n    } else {\n      res += getSamplerAtOutputCoords(inInfo, outShapeInfo);\n    }\n  }\n  return res;\n}\n\nfunction getPackedOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutputPacked1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutputPacked2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutputPacked3DCoords(\n          outShape as [number, number, number], outTexShape);\n    default:\n      return getOutputPackedNDCoords(outShape, outTexShape);\n  }\n}\n\nfunction getOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutput1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutput2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutput3DCoords(\n          outShape as [number, number, number], outTexShape);\n    case 4:\n      return getOutput4DCoords(\n          outShape as [number, number, number, number], outTexShape);\n    case 5:\n      return getOutput5DCoords(\n          outShape as [number, number, number, number, number], outTexShape);\n    case 6:\n      return getOutput6DCoords(\n          outShape as [number, number, number, number, number, number],\n          outTexShape);\n    default:\n      throw new Error(\n          `${outShape.length}-D output sampling is not yet supported`);\n  }\n}\n\nfunction getFloatTextureSampleSnippet(glsl: GLSL): string {\n  return `\n    float sampleTexture(sampler2D textureSampler, vec2 uv) {\n      return ${glsl.texture2D}(textureSampler, uv).r;\n    }\n  `;\n}\n\nfunction getFloatTextureSetRSnippet(glsl: GLSL): string {\n  return `\n    void setOutput(float val) {\n      ${glsl.output} = vec4(val, 0, 0, 0);\n    }\n  `;\n}\n\nfunction getFloatTextureSetRGBASnippet(glsl: GLSL): string {\n  return `\n    void setOutput(vec4 val) {\n      ${glsl.output} = val;\n    }\n  `;\n}\n\nfunction getShaderPrefix(glsl: GLSL): string {\n  const SHADER_PREFIX = `${glsl.version}\n    precision highp float;\n    precision highp int;\n    precision highp sampler2D;\n    ${glsl.varyingFs} vec2 resultUV;\n    ${glsl.defineOutput}\n    const vec2 halfCR = vec2(0.5, 0.5);\n\n    struct ivec5\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n    };\n\n    struct ivec6\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n      int v;\n    };\n\n    uniform float NAN;\n    ${glsl.defineSpecialNaN}\n    ${glsl.defineSpecialInf}\n    ${glsl.defineRound}\n\n    int imod(int x, int y) {\n      return x - y * (x / y);\n    }\n\n    int idiv(int a, int b, float sign) {\n      int res = a / b;\n      int mod = imod(a, b);\n      if (sign < 0. && mod != 0) {\n        res -= 1;\n      }\n      return res;\n    }\n\n    //Based on the work of Dave Hoskins\n    //https://www.shadertoy.com/view/4djSRW\n    #define HASHSCALE1 443.8975\n    float random(float seed){\n      vec2 p = resultUV * seed;\n      vec3 p3  = fract(vec3(p.xyx) * HASHSCALE1);\n      p3 += dot(p3, p3.yzx + 19.19);\n      return fract((p3.x + p3.y) * p3.z);\n    }\n\n    ${SAMPLE_1D_SNIPPET}\n    ${SAMPLE_2D_SNIPPET}\n    ${SAMPLE_3D_SNIPPET}\n  `;\n\n  return SHADER_PREFIX;\n}\n\nconst SAMPLE_1D_SNIPPET = `\nvec2 uvFromFlat(int texNumR, int texNumC, int index) {\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\nvec2 packedUVfrom1D(int texNumR, int texNumC, int index) {\n  int texelIndex = index / 2;\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_2D_SNIPPET = `\nvec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,\n  int texNumC, int row, int col) {\n  int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_3D_SNIPPET = `\nvec2 packedUVfrom3D(int texNumR, int texNumC,\n    int texelsInBatch, int texelsInLogicalRow, int b,\n    int row, int col) {\n  int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SHADER_PACKED_PREFIX = `\n  float getChannel(vec4 frag, vec2 innerDims) {\n    vec2 modCoord = mod(innerDims, 2.);\n    return modCoord.x == 0. ?\n      (modCoord.y == 0. ? frag.r : frag.g) :\n      (modCoord.y == 0. ? frag.b : frag.a);\n  }\n  float getChannel(vec4 frag, int dim) {\n    float modCoord = mod(float(dim), 2.);\n    return modCoord == 0. ? frag.r : frag.g;\n  }\n`;\n\nfunction getOutputScalarCoords() {\n  return `\n    int getOutputCoords() {\n      return 0;\n    }\n  `;\n}\n\nfunction getOutputPacked1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (packedTexShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.x * ${packedTexShape[1]}.0);\n      }\n    `;\n  }\n\n  if (packedTexShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.y * ${packedTexShape[0]}.0);\n      }\n    `;\n  }\n\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      return 2 * (resTexRC.x * ${packedTexShape[1]} + resTexRC.y);\n    }\n  `;\n}\n\nfunction getOutput1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  if (texShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.x * ${texShape[1]}.0);\n      }\n    `;\n  }\n  if (texShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.y * ${texShape[0]}.0);\n      }\n    `;\n  }\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      return resTexRC.x * ${texShape[1]} + resTexRC.y;\n    }\n  `;\n}\n\nfunction getOutputPacked3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texelsInLogicalRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = texelsInLogicalRow * Math.ceil(shape[1] / 2);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec3(b, r, c);\n    }\n  `;\n}\n\nfunction getOutput3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n\nfunction getOutputPackedNDCoords(\n    shape: number[], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  const texelsInLogicalRow = Math.ceil(shape[shape.length - 1] / 2);\n  const texelsInBatch =\n      texelsInLogicalRow * Math.ceil(shape[shape.length - 2] / 2);\n  let texelsInBatchN = texelsInBatch;\n  let batches = ``;\n  let coords = 'b, r, c';\n\n  for (let b = 2; b < shape.length - 1; b++) {\n    texelsInBatchN *= shape[shape.length - b - 1];\n    batches = `\n      int b${b} = index / ${texelsInBatchN};\n      index -= b${b} * ${texelsInBatchN};\n    ` + batches;\n    coords = `b${b}, ` + coords;\n  }\n\n  return `\n    ivec${shape.length} getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      ${batches}\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec${shape.length}(${coords});\n    }\n  `;\n}\n\nfunction getOutput4DCoords(\n    shape: [number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2'], shape);\n\n  return `\n    ivec4 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec4(r, c, d, d2);\n    }\n  `;\n}\n\nfunction getOutput5DCoords(\n    shape: [number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3'], shape);\n\n  return `\n    ivec5 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx * vec2(${texShape[0]},\n                             ${texShape[1]}));\n\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec5 outShape = ivec5(r, c, d, d2, d3);\n      return outShape;\n    }\n  `;\n}\n\nfunction getOutput6DCoords(\n    shape: [number, number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3', 'd4'], shape);\n\n  return `\n    ivec6 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec6 result = ivec6(r, c, d, d2, d3, d4);\n      return result;\n    }\n  `;\n}\n\nfunction getOutputPacked2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return 2 * ivec2(resultUV.yx * vec2(${packedTexShape[0]}, ${\n        packedTexShape[1]}));\n      }\n    `;\n  }\n\n  // texels needed to accommodate a logical row\n  const texelsInLogicalRow = Math.ceil(shape[1] / 2);\n\n  /**\n   * getOutputCoords\n   *\n   * resTexRC: The rows and columns of the texels. If you move over one\n   * texel to the right in the packed texture, you are moving over one column\n   * (not two).\n   *\n   * index: The texel index\n   */\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getOutput2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return ivec2(resultUV.yx * vec2(${texShape[0]}, ${texShape[1]}));\n      }\n    `;\n  }\n  if (shape[1] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(index, 0);\n      }\n    `;\n  }\n  if (shape[0] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(0, index);\n      }\n    `;\n  }\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      int r = index / ${shape[1]};\n      int c = index - r * ${shape[1]};\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getFlatOffsetUniformName(texName: string): string {\n  return `offset${texName}`;\n}\n\nfunction getPackedSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}() {\n      return ${glsl.texture2D}(${texName}, halfCR);\n    }\n  `;\n}\n\nfunction getSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  if (inputInfo.shapeInfo.isUniform) {\n    return `float ${funcName}() {return ${texName};}`;\n  }\n  const [texNumR, texNumC] = inputInfo.shapeInfo.texShape;\n  if (texNumR === 1 && texNumC === 1) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n\n  const [tNumR, tNumC] = inputInfo.shapeInfo.texShape;\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}() {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int index) {\n      vec2 uv = packedUVfrom1D(\n        ${packedTexShape[0]}, ${packedTexShape[1]}, index);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int index) {\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const tNumR = texShape[0];\n  const tNumC = texShape[1];\n\n  if (tNumC === 1 && tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  if (tNumC === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2(0.5, (float(index + ${offset}) + 0.5) / ${tNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2((float(index + ${offset}) + 0.5) / ${tNumC}.0, 0.5);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  return `\n    float ${funcName}(int index) {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const glsl = getGlslDifferences();\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    return `\n      vec4 ${funcName}(int row, int col) {\n        vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n\n        return ${glsl.texture2D}(${texName}, uv);\n      }\n    `;\n  }\n\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const valuesPerRow = Math.ceil(shape[1] / 2);\n\n  return `\n    vec4 ${funcName}(int row, int col) {\n      vec2 uv = packedUVfrom2D(${valuesPerRow}, ${packedTexShape[0]}, ${\n      packedTexShape[1]}, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    const texNumR = texShape[0];\n    const texNumC = texShape[1];\n    return `\n    float ${funcName}(int row, int col) {\n      vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col) {\n        int index = round(dot(vec2(row, col), vec2(${shape[1]}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const offset = getFlatOffsetUniformName(texName);\n  if (texNumC === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2(0.5, (index + 0.5) / ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n  if (texNumR === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2((index + 0.5) / ${texNumC}.0, 0.5);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  return `\n  float ${funcName}(int row, int col) {\n    // Explicitly use integer operations as dot() only works on floats.\n    int index = row * ${shape[1]} + col + ${offset};\n    vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n    return sampleTexture(${texName}, uv);\n  }\n`;\n}\n\nfunction getPackedSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  if (shape[0] === 1) {\n    const squeezedShape = shape.slice(1);\n    const keptDims = [1, 2];\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['b', 'row', 'col'];\n    return `\n        ${getPackedSamplerFromInInfo(newInputInfo)}\n        vec4 ${funcName}(int b, int row, int col) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = valuesPerRow * Math.ceil(shape[1] / 2);\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int b, int row, int col) {\n      vec2 uv = packedUVfrom3D(\n        ${texNumR}, ${texNumC}, ${texelsInBatch}, ${valuesPerRow}, b, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride0 = shape[1] * shape[2];\n  const stride1 = shape[2];\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col', 'depth'];\n    return `\n        ${getSamplerFromInInfo(newInputInfo)}\n        float ${funcName}(int row, int col, int depth) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth) {\n        int index = round(dot(vec3(row, col, depth),\n                          vec3(${stride0}, ${stride1}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n        float ${funcName}(int row, int col, int depth) {\n          float texR = float(row);\n          float texC = dot(vec2(col, depth), vec2(${stride1}, 1));\n          vec2 uv = (vec2(texC, texR) + halfCR) /\n                     vec2(${texNumC}.0, ${texNumR}.0);\n          return sampleTexture(${texName}, uv);\n        }\n      `;\n  }\n\n  if (texNumC === stride1 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col, int depth) {\n      float texR = dot(vec2(row, col), vec2(${shape[1]}, 1));\n      float texC = float(depth);\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n      float ${funcName}(int row, int col, int depth) {\n        // Explicitly use integer operations as dot() only works on floats.\n        int index = row * ${stride0} + col * ${stride1} + depth + ${offset};\n        vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n        return sampleTexture(${texName}, uv);\n      }\n  `;\n}\n\nfunction getPackedSamplerND(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const rank = shape.length;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[rank - 1] / 2);\n  let texelsInBatch = valuesPerRow * Math.ceil(shape[rank - 2] / 2);\n  let params = `int b, int row, int col`;\n  let index = `b * ${texelsInBatch} + (row / 2) * ${valuesPerRow} + (col / 2)`;\n  for (let b = 2; b < rank - 1; b++) {\n    params = `int b${b}, ` + params;\n    texelsInBatch *= shape[rank - b - 1];\n    index = `b${b} * ${texelsInBatch} + ` + index;\n  }\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}(${params}) {\n      int index = ${index};\n      int texR = index / ${texNumC};\n      int texC = index - texR * ${texNumC};\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}, ${texNumR});\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler4D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride2 = shape[3];\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        int index = round(dot(vec4(row, col, depth, depth2),\n                          vec4(${stride0}, ${stride1}, ${stride2}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = float(row);\n        float texC =\n            dot(vec3(col, depth, depth2),\n                vec3(${stride1}, ${stride2}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride2 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = dot(vec3(row, col, depth),\n                         vec3(${shape[1] * shape[2]}, ${shape[2]}, 1));\n        float texC = float(depth2);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} +\n          depth * ${stride2} + depth2;\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler5D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride3 = shape[4];\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float index = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          depth3;\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n                         vec4(${stride1}, ${stride2}, ${stride3}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  if (texNumC === stride3 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float texR = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3]},\n               ${shape[2] * shape[3]}, ${shape[3]}, 1));\n        int texC = depth3;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler6D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3', 'depth4'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  const stride4 = shape[5];\n  const stride3 = shape[4] * stride4;\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n        int index = round(dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          dot(\n            vec2(depth3, depth4),\n            vec2(${stride4}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n          vec4(${stride1}, ${stride2}, ${stride3}, ${stride4})) +\n               float(depth4);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride4 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        float texR = dot(vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3] * shape[4]},\n               ${shape[2] * shape[3] * shape[4]},\n               ${shape[3] * shape[4]},\n               ${shape[4]})) + float(depth3);\n        int texC = depth4;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 * ${stride4} + depth4 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getUniformSampler(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n\n  if (inSize < 2) {\n    return `return ${texName};`;\n  }\n  return `\n    for (int i = 0; i < ${inSize}; i++) {\n      if (i == index) {\n        return ${texName}[i];\n      }\n    }\n  `;\n}\n\nfunction getPackedSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n\n  const type = getCoordsDataType(outRank);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  let output = `return outputValue;`;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n  const isInputScalar = inSize === 1;\n  const outSize = util.sizeFromShape(outShapeInfo.logicalShape);\n  const isOutputScalar = outSize === 1;\n\n  if (inRank === 1 && !isInputScalar && !isOutputScalar) {\n    output = `\n      return vec4(outputValue.xy, outputValue.xy);\n    `;\n  } else if (isInputScalar && !isOutputScalar) {\n    if (outRank === 1) {\n      output = `\n        return vec4(outputValue.x, outputValue.x, 0., 0.);\n      `;\n    } else {\n      output = `\n        return vec4(outputValue.x);\n      `;\n    }\n  } else if (broadcastDims.length) {\n    const rows = inRank - 2;\n    const cols = inRank - 1;\n\n    if (broadcastDims.indexOf(rows) > -1 && broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.x);`;\n    } else if (broadcastDims.indexOf(rows) > -1) {\n      output = `return vec4(outputValue.x, outputValue.y, ` +\n          `outputValue.x, outputValue.y);`;\n    } else if (broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.xx, outputValue.zz);`;\n    }\n  }\n\n  return `\n    vec4 ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      vec4 outputValue = get${texFuncSnippet}(${unpackedCoordsSnippet});\n      ${output}\n    }\n  `;\n}\n\nfunction getSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const outTexShape = outShapeInfo.texShape;\n  const inTexShape = inputInfo.shapeInfo.texShape;\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  if (!inputInfo.shapeInfo.isUniform && inRank === outRank &&\n      inputInfo.shapeInfo.flatOffset == null &&\n      util.arraysEqual(inTexShape, outTexShape)) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, resultUV);\n      }\n    `;\n  }\n\n  const type = getCoordsDataType(outRank);\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  return `\n    float ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      return get${texFuncSnippet}(${unpackedCoordsSnippet});\n    }\n  `;\n}\n\nexport function getCoordsDataType(rank: number): string {\n  if (rank <= 1) {\n    return 'int';\n  } else if (rank === 2) {\n    return 'ivec2';\n  } else if (rank === 3) {\n    return 'ivec3';\n  } else if (rank === 4) {\n    return 'ivec4';\n  } else if (rank === 5) {\n    return 'ivec5';\n  } else if (rank === 6) {\n    return 'ivec6';\n  } else {\n    throw Error(`GPU for rank ${rank} is not yet supported`);\n  }\n}\n\n/** Returns a new input info (a copy) that has a squeezed logical shape. */\nfunction squeezeInputInfo(\n    inInfo: InputInfo, squeezedShape: number[]): InputInfo {\n  // Deep copy.\n  const newInputInfo: InputInfo = JSON.parse(JSON.stringify(inInfo));\n  newInputInfo.shapeInfo.logicalShape = squeezedShape;\n  return newInputInfo;\n}\n\nfunction getSqueezedParams(params: string[], keptDims: number[]): string {\n  return keptDims.map(d => params[d]).join(', ');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ArgMinMaxPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      shape: number[], windowSize: number, op: 'max'|'min',\n      firstPass: boolean) {\n    util.assert(\n        shape.length > 2,\n        () => `Packed arg${\n            op.charAt(0).toUpperCase() +\n            op.slice(1)} supports only inputs with rank above 2.`);\n    const inSize = shape[shape.length - 1];\n    const outSize = Math.ceil(inSize / windowSize);\n    this.outputShape = shape.slice(0, -1);\n    if (outSize > 1) {\n      this.outputShape.push(outSize);\n    }\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    const outShape = this.outputShape;\n    const rank = outShape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n\n    let sourceLocSetup;\n    let sourceRank;\n    if (outSize === 1) {\n      sourceRank = rank + 1;\n      const sourceLocDType = getCoordsDataType(sourceRank);\n      sourceLocSetup = `\n        ${sourceLocDType} sourceLocR = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 1]};\n        ${sourceLocDType} sourceLocG = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 2]};\n        ${sourceLocDType} sourceLocA = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 1]};\n        ${sourceLocDType} sourceLocB = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 2]};`;\n    } else {\n      sourceRank = rank;\n      sourceLocSetup = `\n        ${dtype} sourceLocR = coords;\n        ++${coords[rank - 1]};\n        ${dtype} sourceLocG = coords;\n        ++${coords[rank - 2]};\n        ${dtype} sourceLocA = coords;\n        --${coords[rank - 1]};\n        ${dtype} sourceLocB = coords;\n        --${coords[rank - 2]};`;\n    }\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, sourceRank);\n    const inChannel = '.' + channels[sourceRank - 1];  // e.g. \".b\" for rank 3.\n    const intChannels = channels.map(x => 'int ' + x);\n    const srcRCoords =\n        getChannels('sourceLocR', sourceRank - 1).concat('inIdx.r');\n    const srcGCoords =\n        getChannels('sourceLocG', sourceRank - 1).concat('inIdx.g');\n    const srcBCoords =\n        getChannels('sourceLocB', sourceRank - 1).concat('inIdx.b');\n    const srcACoords =\n        getChannels('sourceLocA', sourceRank - 1).concat('inIdx.a');\n\n    const compOp = (op === 'max') ? 'greaterThan' : 'lessThan';\n    const fetchCandidateIdx = firstPass ? '' : `\n          inIdx = round(vec4(getBestIndicesAChannel(${srcRCoords.join()}),\n                             getBestIndicesAChannel(${srcGCoords.join()}),\n                             getBestIndicesAChannel(${srcBCoords.join()}),\n                             getBestIndicesAChannel(${srcACoords.join()})));`;\n\n    const fetchValue = `vec4(\n            getAChannel(${srcRCoords.join()}),\n            hasNextCol ? getAChannel(${srcGCoords.join()}) : 0.,\n            hasNextRow ? getAChannel(${srcBCoords.join()}) : 0.,\n            hasNextRow && hasNextCol ? getAChannel(${srcACoords.join()}) : 0.)`;\n\n    const getBestIndicesAChannelSnippet = firstPass ? '' : `\n      float getBestIndicesAChannel(${intChannels.join()}) {\n        return getChannel(getBestIndicesA(${channels.join()}),\n                                          vec2(${channels.slice(-2).join()}));\n      }`;\n\n    this.userCode = `\n      float getAChannel(${intChannels.join()}) {\n        return getChannel(getA(${channels.join()}),\n                               vec2(${channels.slice(-2).join()}));\n      }\n      ${getBestIndicesAChannelSnippet}\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        bool hasNextCol = ${coords[rank - 1]} < ${outShape[rank - 1] - 1};\n        bool hasNextRow = ${coords[rank - 2]} < ${outShape[rank - 2] - 1};\n        ${sourceLocSetup}\n        ivec4 srcIdx = ivec4(sourceLocR${inChannel}, sourceLocG${inChannel},\n          sourceLocB${inChannel}, sourceLocA${inChannel}) * ${windowSize};\n        ivec4 inIdx = srcIdx;\n        vec4 bestIndex = vec4(inIdx);\n        vec4 bestValue = ${fetchValue};\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          inIdx = srcIdx;\n          ${fetchCandidateIdx}\n          vec4 candidate = ${fetchValue};\n          bvec4 nan = isnan(candidate);\n          bvec4 replace = bvec4(\n            vec4(${compOp}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));\n\n          bestValue = vec4(replace.x  ? candidate.x : bestValue.x,\n                           replace.y  ? candidate.y : bestValue.y,\n                           replace.z  ? candidate.z : bestValue.z,\n                           replace.w  ? candidate.w : bestValue.w);\n          bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));\n          srcIdx++;\n        }\n        setOutput(bestIndex);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AvgPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth};\n            wC+= ${dilationWidth}) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n\n            dotProd += dyValue * avgMultiplier;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class AvgPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n\n              dotProd += dyValue * avgMultiplier;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\n// (Ar + Ai)(Br + Bi) =\n// ArBr + ArBi + AiBr + AiBi = ArBr - AB + ArBi + AiBr\n// Yr = ArBr - AB\n// Yi = ArBi + AiBr\nexport const COMPLEX_MULTIPLY = {\n  REAL: 'return areal * breal - aimag * bimag;',\n  IMAG: 'return areal * bimag + aimag * breal;'\n};\n\nexport class BinaryOpComplexProgram implements GPGPUProgram {\n  variableNames = ['AReal', 'AImag', 'BReal', 'BImag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    this.userCode = `\n      float binaryOpComplex(\n          float areal, float aimag, float breal, float bimag) {\n        ${op}\n      }\n\n      void main() {\n        float areal = getARealAtOutCoords();\n        float aimag = getAImagAtOutCoords();\n        float breal = getBRealAtOutCoords();\n        float bimag = getBImagAtOutCoords();\n        setOutput(binaryOpComplex(areal, aimag, breal, bimag));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nconst CHECK_NAN_SNIPPET = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const ADD = 'return a + b;';\nexport const SUB = 'return a - b;';\nexport const MUL = 'return a * b;';\n\n// We use native integer division to deal with floating point imprecision. Since\n// we implement floor division and glsl implements truncated division, we\n// correct for this by subtracting 1 from result when the result is negative and\n// there is a remainder.\nexport const INT_DIV = `\n  float s = sign(a) * sign(b);\n  int ia = round(a);\n  int ib = round(b);\n  if (ib != 0) {\n    // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n    return float(idiv(ia, ib, s));\n  } else {\n    return NAN;\n  }\n`;\n\nexport const POW = `\nif(a < 0.0 && floor(b) < b){\n  return NAN;\n}\nif (b == 0.0) {\n  return 1.0;\n}\nreturn (round(mod(b, 2.0)) != 1) ?\n    pow(abs(a), b) : sign(a) * pow(abs(a), b);\n`;\nexport const SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const EQUAL = `return float(a == b);`;\n\nexport const NOT_EQUAL = `return float(a != b);`;\n\nexport const LESS = `return float(a < b);`;\n\nexport const LESS_EQUAL = `return float(a <= b);`;\n\nexport const GREATER = `return float(a > b);`;\n\nexport const GREATER_EQUAL = `return float(a >= b);`;\n\nexport const LOGICAL_AND = `return float(a >= 1.0 && b >= 1.0);`;\n\nexport const LOGICAL_OR = `return float(a >= 1.0 || b >= 1.0);`;\n\nexport const MAX = CHECK_NAN_SNIPPET + `\n  return max(a, b);\n`;\nexport const MIN = CHECK_NAN_SNIPPET + `\n  return min(a, b);\n`;\nexport const MOD = `if (b == 0.0) return NAN;\n  return mod(a, b);`;\n\nexport const ELU_DER = `return (b >= 1.0) ? a : a * (b + 1.0);`;\n\nexport const PRELU = `return (a < 0.) ? b * a : a;`;\n\nexport class BinaryOpProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    this.userCode = `\n      float binaryOperation(float a, float b) {\n        ${op}\n      }\n\n      void main() {\n        float a = getAAtOutCoords();\n        float b = getBAtOutCoords();\n        setOutput(binaryOperation(a, b));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nconst CHECK_NAN_SNIPPET = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\nexport const INT_DIV = `\n  ivec4 ia = round(a);\n  ivec4 ib = round(b);\n  bvec4 cond = notEqual(ib, ivec4(0));\n  ivec4 result = ivec4(0);\n  vec4 s = sign(a) * sign(b);\n\n  // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n  if (cond[0]) {\n    result[0] = idiv(ia[0], ib[0], s[0]);\n  }\n  if (cond[1]) {\n    result[1] = idiv(ia[1], ib[1], s[1]);\n  }\n  if (cond[2]) {\n    result[2] = idiv(ia[2], ib[2], s[2]);\n  }\n  if (cond[3]) {\n    result[3] = idiv(ia[3], ib[3], s[3]);\n  }\n  return vec4(result);\n`;\n\nexport const POW = `\n  // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.\n  vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));\n  vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);\n  vec4 result = multiplier * pow(abs(a), b);\n\n  // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS\n  bvec4 isExpZero = equal(b, vec4(0.0));\n  result.r = isExpZero.r ? 1.0 : result.r;\n  result.g = isExpZero.g ? 1.0 : result.g;\n  result.b = isExpZero.b ? 1.0 : result.b;\n  result.a = isExpZero.a ? 1.0 : result.a;\n\n  vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const PRELU = `\n  vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));\n  return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);\n`;\n\nexport const ELU_DER = `\n  vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));\n  return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));\n`;\n\nexport const EQUAL = `\n  return vec4(equal(a, b));\n`;\n\nexport const NOT_EQUAL = `\n  return vec4(notEqual(a, b));\n`;\n\nexport const LESS = `\n  return vec4(lessThan(a, b));\n`;\n\nexport const LESS_EQUAL = `\n  return vec4(lessThanEqual(a, b));\n`;\n\nexport const GREATER = `\n  return vec4(greaterThan(a, b));\n`;\n\nexport const GREATER_EQUAL = `\n  return vec4(greaterThanEqual(a, b));\n`;\n\nexport const LOGICAL_AND = `\n  return vec4(\n    vec4(greaterThanEqual(a, vec4(1.0))) *\n    vec4(greaterThanEqual(b, vec4(1.0))));\n`;\n\nexport const LOGICAL_OR = `\n  return min(\n    vec4(greaterThanEqual(a, vec4(1.0))) +\n    vec4(greaterThanEqual(b, vec4(1.0))),\n    vec4(1.0));\n`;\n\nexport const MAX = `\n  vec4 result = vec4(max(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MIN = `\n  vec4 result = vec4(min(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MOD = `\n  vec4 result = mod(a, b);\n  vec4 isNaN = vec4(equal(b, vec4(0.0)));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport class BinaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n  supportsBroadcasting = true;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      op: string, aShape: number[], bShape: number[],\n      checkOutOfBounds = false) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const rank = this.outputShape.length;\n    let checkOutOfBoundsString = '';\n    if (checkOutOfBounds) {\n      if (rank === 0 || util.sizeFromShape(this.outputShape) === 1) {\n        checkOutOfBoundsString = `\n          result.y = 0.;\n          result.z = 0.;\n          result.w = 0.;\n        `;\n      } else {\n        const dtype = getCoordsDataType(rank);\n        checkOutOfBoundsString = `\n          ${dtype} coords = getOutputCoords();\n        `;\n        if (rank === 1) {\n          checkOutOfBoundsString += `\n            result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;\n            result.z = 0.;\n            result.w = 0.;\n          `;\n        } else {\n          const channels = getChannels('coords', rank);\n          checkOutOfBoundsString += `\n            bool nextRowOutOfBounds =\n              (${channels[rank - 2]} + 1) >= ${this.outputShape[rank - 2]};\n            bool nextColOutOfBounds =\n              (${channels[rank - 1]} + 1) >= ${this.outputShape[rank - 1]};\n            result.y = nextColOutOfBounds ? 0. : result.y;\n            result.z = nextRowOutOfBounds ? 0. : result.z;\n            result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;\n          `;\n        }\n      }\n    }\n\n    this.userCode = `\n      vec4 binaryOperation(vec4 a, vec4 b) {\n        ${op}\n      }\n\n      void main() {\n        vec4 a = getAAtOutCoords();\n        vec4 b = getBAtOutCoords();\n\n        vec4 result = binaryOperation(a, b);\n        ${checkOutOfBoundsString}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        float value = getAAtOutCoords();\n        if (isnan(value)) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, minVal, maxVal));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        vec4 value = getAAtOutCoords();\n\n        if (any(isnan(value))) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, vec4(minVal), vec4(maxVal)));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ComplexAbsProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(shape: number[]) {\n    this.outputShape = shape;\n    this.userCode = `\n      void main() {\n        float re = abs(getRealAtOutCoords());\n        float im = abs(getImagAtOutCoords());\n        float mx = max(re, im);\n\n        // sadly the length function in glsl is not underflow-safe\n        // (at least not on Intel GPUs). So the safe solution is\n        // to ensure underflow-safety in all cases.\n        setOutput(\n          mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))\n        );\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ConcatProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  // Concats 2d tensors along axis=1. See comments in MathBackendWebGL.concat().\n  constructor(shapes: Array<[number, number]>) {\n    this.outputShape = backend_util.computeOutShape(shapes, 1 /* axis */);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][1];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][1];\n    }\n\n    const snippets = [`if (yC < ${offsets[0]}) setOutput(getT0(yR, yC));`];\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      snippets.push(\n          `else if (yC < ${offsets[i]}) ` +\n          `setOutput(getT${i}(yR, yC-${shift}));`);\n    }\n    const lastIndex = offsets.length;\n    const lastShift = offsets[offsets.length - 1];\n    snippets.push(`else setOutput(getT${lastIndex}(yR, yC-${lastShift}));`);\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int yR = coords.x;\n        int yC = coords.y;\n\n        ${snippets.join('\\n        ')}\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ConcatPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(shapes: number[][], axis: number) {\n    this.outputShape = backend_util.computeOutShape(shapes, axis);\n    const shape = this.outputShape;\n    const rank = shape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][axis];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][axis];\n    }\n\n    const channel = channels[axis];\n    const lastChannels = channels.slice(-2);\n    const allChannels = channels.join();\n\n    let getValueSnippet = `if (${channel} < ${offsets[0]}) {\n        return getChannel(\n            getT0(${allChannels}), vec2(${lastChannels.join()}));\n        }`;\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      // Note: the >= comparison below may seem unnecessary given the check\n      // above but is needed to workaround branch execution issues on some\n      // devices. It makes all the conditions exclusive without relying on\n      // execution order.\n      getValueSnippet += `\n        if (${channel} < ${offsets[i]}  && ${channel} >= ${offsets[i - 1]}) {\n          return getChannel(\n            getT${i}(${shiftedChannels(channels, channel, shift)}),\n            vec2(${shiftedChannels(lastChannels, channel, shift)}));\n        }`;\n    }\n    const lastIndex = offsets.length;\n    const shift = offsets[offsets.length - 1];\n    getValueSnippet += `\n        return getChannel(\n          getT${lastIndex}(${shiftedChannels(channels, channel, shift)}),\n          vec2(${shiftedChannels(lastChannels, channel, shift)}));`;\n\n    this.userCode = `\n      float getValue(${channels.map(x => 'int ' + x)}) {\n        ${getValueSnippet}\n      }\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        vec4 result = vec4(getValue(${coords}), 0., 0., 0.);\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} + 1;\n        if (${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.g = getValue(${coords});\n        }\n\n        ${coords[rank - 2]} = ${coords[rank - 2]} + 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]}) {\n          result.a = getValue(${coords});\n        }\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} - 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]} &&\n            ${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.b = getValue(${coords});\n        }\n        setOutput(result);\n      }\n    `;\n  }\n}\n\n/**\n * Return an expression for coordinates into a vector where a given channel\n * will be offset by [shift].\n *\n * @param channels the channels to consider\n * @param channel the channel we want shifted\n * @param shift  the amount to subtract from the channel.\n *\n * @returns a string of the form 'x, y-[shift], z' where any one channel can\n * have the shift applied.\n */\nfunction shiftedChannels(channels: string[], channel: string, shift: number) {\n  const channelIdx = channels.indexOf(channel);\n  const res = channels.map((c, idx) => {\n    if (idx === channelIdx) {\n      return `${c} - ${shift}`;\n    } else {\n      return c;\n    }\n  });\n  return res.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int d2 = coords.w;\n\n        // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              if (${isChannelsLast}) {\n                float dyValue = getDy(b, yR, yC, d2);\n                float xValue = getX(b, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              } else {\n                float dyValue = getDy(b, d2, yR, yC);\n                float xValue = getX(b, d1, xR, xC);\n                dotProd += (xValue * dyValue);\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[${channelDim}];\n\n        ivec2 dyCorner = ivec2(coords[${rowDim}], coords[${colDim}]) - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n\n              if (${isChannelsLast}) {\n                float xValue = getDy(batch, idyR, idyC, d2);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              } else {\n                float xValue = getDy(batch, d2, idyR, idyC);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    this.userCode = `\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int wF = coords.x;\n        int wR = coords.y;\n        int wC = coords.z;\n        int d1 = coords.w;\n        int d2 = coords.u;\n\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yF = 0; yF < ${convInfo.outDepth}; yF++) {\n            int xF = wF + yF * ${strideDepth} - ${padFront};\n\n            if (xF < 0 || xF >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n              int xR = wR + yR * ${strideHeight} - ${padTop};\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n                int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float dyValue = getDy(b, yF, yR, yC, d2);\n                float xValue = getX(b, xF, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padFront = filterDepth - 1 - convInfo.padInfo.front;\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d1 = coords.u;\n\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyFCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          float dyF = float(dyFCorner + wF) / ${strideDepth}.0;\n\n          if (dyF < 0.0 || dyF >= ${convInfo.outDepth}.0 || fract(dyF) > 0.0) {\n            continue;\n          }\n          int idyF = int(dyF);\n\n          int wFPerm = ${filterDepth} - 1 - wF;\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n              fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            int wRPerm = ${filterHeight} - 1 - wR;\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              int wCPerm = ${filterWidth} - 1 - wC;\n\n              for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n                float xValue = getDy(batch, idyF, idyR, idyC, d2);\n                float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int dm = coords.w;\n        int d2 = d1 * ${channelMul} + dm;\n\n        float dotProd = 0.0;\n\n        // TO DO: Vec4 over the batch size\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float dyValue = getDy(b, yR, yC, d2);\n              float xValue = getX(b, xR, xC, d1);\n              dotProd += (xValue * dyValue);\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class DepthwiseConv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[3];\n        ivec2 dyCorner = coords.yz - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        float dotProd = 0.0;\n\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            // TO DO: Vec4 over the channelMul\n            for (int dm = 0; dm < ${channelMul}; dm++) {\n              int d2 = d1 * ${channelMul} + dm;\n              float xValue = getDy(batch, idyR, idyC, d2);\n              float wValue = getW(wRPerm, wCPerm, d1, dm);\n              dotProd += xValue * wValue;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivationWeights = false) {\n    this.outputShape = convInfo.outShape;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivationWeights) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivationWeights) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d2 = coords[${channelDim}];\n\n        ivec2 xRCCorner =\n            ivec2(coords[${rowDim}], coords[${colDim}]) * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${convInfo.inWidth}) {\n              continue;\n            }\n\n            for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n              vec4 wValues = vec4(\n                getW(wR, wC, d1, d2),\n                getW(wR, wC, d1 + 1, d2),\n                getW(wR, wC, d1 + 2, d2),\n                getW(wR, wC, d1 + 3, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec4 xValues = vec4(\n                  getX(batch, xR, xC, d1),\n                  getX(batch, xR, xC, d1 + 1),\n                  getX(batch, xR, xC, d1 + 2),\n                  getX(batch, xR, xC, d1 + 3)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec4 xValues = vec4(\n                  getX(batch, d1, xR, xC),\n                  getX(batch, d1 + 1, xR, xC),\n                  getX(batch, d1 + 2, xR, xC),\n                  getX(batch, d1 + 3, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n\n            if (${inputDepthVec4Remainder === 1}) {\n\n              if (${isChannelsLast}) {\n                dotProd +=\n                    getX(batch, xR, xC, ${inputDepthNearestVec4}) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              } else {\n                dotProd +=\n                    getX(batch, ${inputDepthNearestVec4}, xR, xC) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              }\n\n            } else if (${inputDepthVec4Remainder === 2}) {\n              vec2 wValues = vec2(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec2 xValues = vec2(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec2 xValues = vec2(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            } else if (${inputDepthVec4Remainder === 3}) {\n              vec3 wValues = vec3(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 2, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec3 xValues = vec3(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec3 xValues = vec3(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 2, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            }\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nexport class Conv3DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.outShape;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n\n    this.userCode = `\n      const ivec3 strides = ivec3(${strideDepth}, ${strideHeight}, ${\n        strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d2 = coords.u;\n\n        ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xFCorner = xFRCCorner.x;\n        int xRCorner = xFRCCorner.y;\n        int xCCorner = xFRCCorner.z;\n\n        // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get\n        // y(yF, yR, yC, d2). ? = to be determined. : = across all\n        // values in that axis.\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          int xF = xFCorner + wF * ${dilationDepth};\n\n          if (xF < 0 || xF >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            int xR = xRCorner + wR * ${dilationHeight};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n                vec4 xValues = vec4(\n                  getX(batch, xF, xR, xC, d1),\n                  getX(batch, xF, xR, xC, d1 + 1),\n                  getX(batch, xF, xR, xC, d1 + 2),\n                  getX(batch, xF, xR, xC, d1 + 3)\n                );\n                vec4 wValues = vec4(\n                  getW(wF, wR, wC, d1, d2),\n                  getW(wF, wR, wC, d1 + 1, d2),\n                  getW(wF, wR, wC, d1 + 2, d2),\n                  getW(wF, wR, wC, d1 + 3, d2)\n                );\n\n                dotProd += dot(xValues, wValues);\n              }\n\n              if (${inputDepthVec4Remainder === 1}) {\n                dotProd +=\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}) *\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2);\n              } else if (${inputDepthVec4Remainder === 2}) {\n                vec2 xValues = vec2(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                vec2 wValues = vec2(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else if (${inputDepthVec4Remainder === 3}) {\n                vec3 xValues = vec3(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                vec3 wValues = vec3(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 2, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2 / ${channelMul};\n        int q = d2 - d1 * ${channelMul};\n\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${xNumRows}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${xNumCols}) {\n              continue;\n            }\n\n            float xVal = getX(batch, xR, xC, d1);\n            float wVal = getW(wR, wC, d1, q);\n            dotProd += xVal * wVal;\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConvPacked2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const texelsAcross = filterWidth;\n\n    let mainLoop = `int xR; int xC; int xCOffset;`;\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `\n          vec4 xTexelR${r}C${c * 2} = vec4(0.);\n          vec4 wR${r}C${c} = vec4(0.);\n          vec4 xR${r}C${c} = vec4(0.);`;\n      }\n    }\n\n    /**\n     * This vectorized implementation works by gathering the values needed for\n     * each output channel's dot product into vec4's and then multiplying them\n     * all together (this happens in the final double for-loop below). Most of\n     * the main loop consists of constructing these vec4's with the minimum\n     * number of texture2D calls, which means making use of all four returned\n     * values from a texture2D call at once.\n     */\n    for (let r = 0; r < filterHeight; r++) {\n      for (let texelC = 0; texelC < texelsAcross; texelC++) {\n        const c = texelC * 2;\n\n        mainLoop += `\n          xR = xRCorner + ${r * dilationHeight};\n          xC = xCCorner + ${c * dilationWidth};\n        `;\n\n        if (strideWidth === 1) {\n          if (c < filterWidth) {\n            // If padding is odd, the outer texels have to be composed.\n            if (padLeft % 2 === 1) {\n              // TODO: Ensure vec4 previous does not result in redundant sample,\n              // and avoid setting xTexelRC's that exceed the boundary in the\n              // first place rather than resetting them to vec4(0)).\n\n              // To compute xCOffset:\n              // - If padding is odd, we must add 1 to ensure we ask for an\n              // even-numbered row.\n              // - We subtract 2 to access the previous texel.\n\n              mainLoop += `\n                xCOffset = xC + 1;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    xTexelR${r}C${c}.zw = vec2(0.);\n                  }\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + 1 - 2;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  vec4 previous = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    previous.zw = vec2(0.);\n                  }\n\n                  xR${r}C${c} = vec4(previous.zw, xTexelR${r}C${c}.xy);\n                } else {\n                  xR${r}C${c} = vec4(0, 0, xTexelR${r}C${c}.xy);\n                }\n              `;\n            } else {\n              // Padding is even, so xRC corresponds to a single texel.\n              mainLoop += `\n                if(xR >= 0 && xR < ${xNumRows} && xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xR${r}C${c} = xTexelR${r}C${c};\n              `;\n            }\n\n            if (c + 1 < filterWidth) {\n              // If dilation is even, the second entry should match the first\n              // (either both are composed or both are single samples). But if\n              // dilation is odd, then the second entry should be the opposite\n              // of the first (if the first is composed, the second is a single\n              // sample, and vice versa.)\n\n              const nextTexelOffset = padLeft % 2 === 0 ?\n                  util.nearestLargerEven(dilationWidth) :\n                  dilationWidth;\n\n              if ((dilationWidth % 2 === 0 && padLeft % 2 === 1) ||\n                  (dilationWidth % 2 !== 0 && padLeft % 2 !== 1)) {\n                mainLoop += `\n                  xCOffset = xC + ${padLeft % 2} + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n                `;\n\n                // If dilation > 1 then the xRC's will not be able to share any\n                // values, so each xRC will require two unique calls to getX.\n                if (dilationWidth > 1) {\n                  mainLoop += `\n                    xCOffset -= 2;\n                    if(xR >= 0 && xR < ${xNumRows} &&\n                      xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                      xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                    } else {\n                      xTexelR${r}C${c} = vec4(0.);\n                    }\n                  `;\n                }\n\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.xy);\n                `;\n              } else {\n                mainLoop += `\n                  xCOffset = xC + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n\n                  xR${r}C${c + 1} = xTexelR${r}C${c + 2};\n                `;\n              }\n            }\n          }\n        } else {  // stride > 1\n          if (c < filterWidth) {\n            mainLoop += `\n              if(xR >= 0 && xR < ${xNumRows}) {\n            `;\n\n            // Depending on whether padLeft is even or odd, we want either the\n            // xy or zw channels from X texels for xR${r}C${c}. If padLeft is\n            // even, xR${r}C${c + 1} is simply the zw channels of texels we've\n            // already sampled. But if padLeft is odd, xR${r}C{$c + 1}.zw will\n            // need to come from the xy channels of a new texel, hence the `vec4\n            // final` initialized below.\n            if (padLeft % 2 === 1) {\n              mainLoop += `\n                xCOffset = xC + 1 - ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                if(xC + 1 >= 0 && xC + 1 < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xC + 1, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  vec4 final = vec4(0.);\n                  xCOffset = xC + 1 + ${strideWidth};\n                  if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    final = getX(batch, xR, xCOffset, d1);\n                  }\n                  xR${r}C${c + 1} = vec4(xTexelR${r}C${c + 2}.xy, final.xy);\n                `;\n              }\n            } else {\n              mainLoop += `\n                if(xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.xy, xTexelR${r}C${c + 2}.xy);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n                `;\n              }\n            }\n\n            mainLoop += `}`;\n          }\n        }\n\n        if (c < filterWidth) {\n          mainLoop += `\n            vec4 wTexelR${r}C${c} = getW(${r}, ${c}, d1, q);\n            wR${r}C${c} = vec4(wTexelR${r}C${c}.xz, wTexelR${r}C${c}.xz);\n          `;\n\n          if (c + 1 < filterWidth) {\n            mainLoop += `\n              vec4 wTexelR${r}C${c + 1} = getW(${r}, ${c + 1}, d1, q);\n              wR${r}C${c + 1} =\n                vec4(wTexelR${r}C${c + 1}.xz, wTexelR${r}C${c + 1}.xz);`;\n          }\n        }\n      }\n    }\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `dotProd += xR${r}C${c} * wR${r}C${c};`;\n      }\n    }\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2;\n        int q = 0;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        vec4 dotProd = vec4(0.);\n\n        ${mainLoop}\n\n        vec4 result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class CropAndResizeProgram implements GPGPUProgram {\n  variableNames = ['Image', 'Boxes', 'BoxInd'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], boxShape: [number, number],\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number) {\n    const [batch, imageHeight, imageWidth, depth] = imageShape;\n    const [numBoxes, ] = boxShape;\n    const [cropHeight, cropWidth] = cropSize;\n    this.outputShape = [numBoxes, cropHeight, cropWidth, depth];\n    const methodId = method === 'bilinear' ? 1 : 0;\n\n    const [inputHeightFloat, inputWidthFloat] =\n        [`${imageHeight - 1}.0`, `${imageWidth - 1}.0`];\n\n    const [heightRatio, heightScale, inY] = cropHeight > 1 ?\n        [\n          `${(imageHeight - 1) / (cropHeight - 1)}`,\n          '(y2-y1) * height_ratio',\n          `y1*${inputHeightFloat} + float(y)*(height_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (y1+y2) * ${inputHeightFloat}`,\n        ];\n    const [widthRatio, widthScale, inX] = cropWidth > 1 ?\n        [\n          `${(imageWidth - 1) / (cropWidth - 1)}`,\n          '(x2-x1) * width_ratio',\n          `x1*${inputWidthFloat} + float(x)*(width_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (x1+x2) * ${inputWidthFloat}`,\n        ];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc\n    this.userCode = `\n      const float height_ratio = float(${heightRatio});\n      const float width_ratio = float(${widthRatio});\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int y = coords[1];\n        int x = coords[2];\n        int d = coords[3];\n\n        // get box vals\n        float y1 = getBoxes(b,0);\n        float x1 = getBoxes(b,1);\n        float y2 = getBoxes(b,2);\n        float x2 = getBoxes(b,3);\n\n        // get image in batch index\n        int bInd = round(getBoxInd(b));\n        if(bInd < 0 || bInd >= ${batch}) {\n          return;\n        }\n\n        float height_scale = ${heightScale};\n        float width_scale = ${widthScale};\n\n        float in_y = ${inY};\n        if( in_y < 0.0 || in_y > ${inputHeightFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n        float in_x = ${inX};\n        if( in_x < 0.0 || in_x > ${inputWidthFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n\n        vec2 sourceFracIndexCR = vec2(in_x,in_y);\n        if(${methodId} == 1) {\n          // Compute the four integer indices.\n          ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);\n          ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));\n\n          float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);\n          float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);\n          float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);\n          float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);\n\n          vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);\n\n          float top = topLeft + (topRight - topLeft) * fracCR.x;\n          float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;\n          float newValue = top + (bottom - top) * fracCR.y;\n          setOutput(newValue);\n        } else {\n          // Compute the coordinators of nearest neighbor point.\n          ivec2 sourceNearestCR = ivec2(floor(\n            sourceFracIndexCR + vec2(0.5,0.5)));\n          float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);\n          setOutput(newValue);\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class CumSumProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  index: WebGLUniformLocation;\n\n  constructor(shape: number[], exclusive: boolean, reverse: boolean) {\n    this.outputShape = shape;\n    const rank = shape.length;\n    const val = exclusive ? '0.0' : `getX(${getCoords(rank, 'coords')})`;\n    const length = shape[shape.length - 1];\n    let condition = '';\n    let idxString = '';\n    // When exclusive is set, the cumsum op becomes roll op that copies the\n    // value from the previous index based on the direction specified by the\n    // reverse flag.\n    if (exclusive) {\n      condition = reverse ? `end != ${length - 1}` : 'end != 0';\n      idxString = reverse ? 'end + 1' : 'end - 1';\n    } else {\n      condition = reverse ? `end + pow2 < ${length}` : 'end >= pow2';\n      idxString = (reverse ? 'end + pow2' : 'end - pow2');\n    }\n\n    this.userCode = `\n      uniform float index;\n      void main() {\n        ${getCoordsDataType(rank)} coords = getOutputCoords();\n        int end = ${getFinalCoord(rank, 'coords')};\n        float val = ${val};\n        int pow2 = int(pow(2.0, index));\n        if (${condition}) {\n          int idx = ${idxString};\n          ${getFinalCoord(rank, 'coords')} = idx;\n          val += getX(${getCoords(rank, 'coords')});\n        }\n        setOutput(val);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(index: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.index == null) {\n        this.index = gpgpu.getUniformLocation(webGLProgram, 'index');\n      }\n      gpgpu.gl.uniform1f(this.index, index);\n    };\n  }\n}\n\nfunction getCoords(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.x, ${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.x, ${name}.y, ${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.x, ${name}.y, ${name}.z, ${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n\nfunction getFinalCoord(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: [number, number, number];\n  packedInputs = false;\n  packedOutput = true;\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getA(rc.x, rc.y, rc.z);\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: [number, number, number];\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthToSpaceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  blockSize: number;\n  dataFormat: string;\n\n  constructor(\n      outputShape: number[], blockSize: number, dataFormat: 'NHWC'|'NCHW') {\n    this.outputShape = outputShape;\n    this.blockSize = blockSize;\n    this.dataFormat = dataFormat;\n    this.userCode = `\n    void main() {\n      ivec4 coords = getOutputCoords();\n      int b = coords[0];\n      int h = ${this.getHeightCoordString()};\n      int w = ${this.getWidthCoordString()};\n      int d = ${this.getDepthCoordString()};\n\n      int in_h = h / ${blockSize};\n      int offset_h = imod(h, ${blockSize});\n      int in_w = w / ${blockSize};\n      int offset_w = imod(w, ${blockSize});\n      int offset_d = (offset_h * ${blockSize} + offset_w) *\n        ${this.getOutputDepthSize()};\n      int in_d = d + offset_d;\n\n      float result = ${this.getInputSamplingString()};\n      setOutput(result);\n    }\n  `;\n  }\n\n  private getHeightCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[1]`;\n    } else {\n      return `coords[2]`;\n    }\n  }\n\n  private getWidthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[2]`;\n    } else {\n      return `coords[3]`;\n    }\n  }\n\n  private getDepthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[3]`;\n    } else {\n      return `coords[1]`;\n    }\n  }\n\n  private getOutputDepthSize(): number {\n    if (this.dataFormat === 'NHWC') {\n      return this.outputShape[3];\n    } else {\n      return this.outputShape[1];\n    }\n  }\n\n  private getInputSamplingString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `getX(b, in_h, in_w, in_d)`;\n    } else {\n      return `getX(b, in_d, in_h, in_w)`;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DiagProgram implements GPGPUProgram {\n  variableNames = ['X'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(size: number) {\n    this.outputShape = [size, size];\n    this.userCode = `\n      void main() {\n          ivec2 coords = getOutputCoords();\n          float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;\n          setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        float x = getAAtOutCoords();\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = false;\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: [number, number, number]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n        float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class EncodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let output = `result`;\n    if (inputIsUnsignedByte) {\n      output = `floor(result * 255. + 0.5)`;\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        int flatIndex = getFlatIndex(coords);\n        int offset = imod(flatIndex, 4);\n\n        flatIndex = idiv(flatIndex, 4, 1.);\n\n        int r = flatIndex / ${width};\n        int c = imod(flatIndex, ${width});\n        vec2 uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n        vec4 values = ${glsl.texture2D}(A, uv);\n\n        float result;\n\n        if(offset == 0) {\n          result = values[0];\n        } else if(offset == 1) {\n          result = values[1];\n        } else if(offset == 2) {\n          result = values[2];\n        } else {\n          result = values[3];\n        }\n\n        ${glsl.output} = vec4(${output}, 0., 0., 0.);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\n/*\nThis is how the shader encodes a tensor with shape = [2, 3, 5]\n(indices are [batch, row, col]).\n\n000|001   002|003   004|xxx   020|021   022|023   024|xxx\n-------   -------   -------   -------   -------   -------\n010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\n100|101   102|103   104|xxx   120|121   122|123   124|xxx\n-------   -------   -------   -------   -------   -------\n110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\nSingle texels contain only values from the same batch, and from adjacent rows\nand columns.\n */\n\nexport class EncodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let mainLoop = '';\n    let output = 'result';\n    if (inputIsUnsignedByte) {\n      output = 'floor(result * 255. + 0.5)';\n    }\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        const channel = row * 2 + col;\n\n        mainLoop += `\n          localCoords = coords;\n          if(localCoords[2] + ${col} < ${outputShape[2]}) {\n            localCoords[2] += ${col};\n            if(localCoords[1] + ${row} < ${outputShape[1]}) {\n              localCoords[1] += ${row};\n\n              flatIndex = getFlatIndex(localCoords);\n              offset = imod(flatIndex, 4);\n\n              flatIndex = idiv(flatIndex, 4, 1.);\n\n              r = flatIndex / ${width};\n              c = imod(flatIndex, ${width});\n              uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n              values = ${glsl.texture2D}(A, uv);\n\n              if(offset == 0) {\n                result[${channel}] = values[0];\n              } else if(offset == 1) {\n                result[${channel}] = values[1];\n              } else if(offset == 2) {\n                result[${channel}] = values[2];\n              } else {\n                result[${channel}] = values[3];\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        vec4 result = vec4(0.);\n        int flatIndex, r, c, offset;\n        ivec3 localCoords;\n        vec2 uv;\n        vec4 values;\n\n        ${mainLoop}\n\n        ${glsl.output} = ${output};\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const COMPLEX_FFT = {\n  REAL: 'return real * expR - imag * expI;',\n  IMAG: 'return real * expI + imag * expR;'\n};\n\nexport class FFTProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, inputShape: [number, number], inverse: boolean) {\n    const innerDim = inputShape[1];\n    this.outputShape = inputShape;\n\n    const exponentMultiplierSnippet =\n        inverse ? `2.0 * ${Math.PI}` : `-2.0 * ${Math.PI}`;\n    const resultDenominator = inverse ? `${innerDim}.0` : '1.0';\n\n    this.userCode = `\n      const float exponentMultiplier = ${exponentMultiplierSnippet};\n\n      float unaryOpComplex(float real, float expR, float imag, float expI) {\n        ${op}\n      }\n\n      float mulMatDFT(int batch, int index) {\n        float indexRatio = float(index) / float(${innerDim});\n        float exponentMultiplierTimesIndexRatio =\n            exponentMultiplier * indexRatio;\n\n        float result = 0.0;\n\n        for (int i = 0; i < ${innerDim}; i++) {\n          // x = (-2|2 * PI / N) * index * i;\n          float x = exponentMultiplierTimesIndexRatio * float(i);\n          float expR = cos(x);\n          float expI = sin(x);\n          float real = getReal(batch, i);\n          float imag = getImag(batch, i);\n\n          result +=\n              unaryOpComplex(real, expR, imag, expI) / ${resultDenominator};\n        }\n\n        return result;\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        setOutput(mulMatDFT(coords[0], coords[1]));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FillProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  valueLoc: WebGLUniformLocation;\n\n  constructor(shape: number[], value: number) {\n    this.variableNames = ['x'];\n    this.outputShape = shape;\n\n    this.userCode = `\n      uniform float value;\n      void main() {\n        // Input can be obtained from uniform value.\n        setOutput(value);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(value: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.valueLoc == null) {\n        this.valueLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'value');\n      }\n      gpgpu.gl.uniform1f(this.valueLoc, value);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherProgram implements GPGPUProgram {\n  variableNames = ['A', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], indicesLength: number, axis: number) {\n    const outputShape: number[] = aShape.slice();\n    outputShape[axis] = indicesLength;\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape, axis);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[], axis: number): string {\n  const rank = aShape.length;\n  if (rank > 4) {\n    throw Error(`Gather for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `int(getIndices(resRC))`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    if (i === axis) {\n      sourceCoords.push(`int(getIndices(${currentCoords[i]}))`);\n    } else {\n      sourceCoords.push(`${currentCoords[i]}`);\n    }\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherNDProgram implements GPGPUProgram {\n  variableNames = ['x', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  constructor(\n      private sliceDim: number, private strides: number[], shape: number[]) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    const strideString = this.sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${this.strides});\n         void main() {\n          ${dtype} coords = getOutputCoords();\n          int flattenIndex = 0;\n          for (int j = 0; j < ${this.sliceDim}; j++) {\n            int index = round(getIndices(coords[0], j));\n            flattenIndex += index * ${strideString};\n          }\n          setOutput(getX(flattenIndex, coords[1]));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PixelData, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {getGlslDifferences} from './glsl_version';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport * as webgl_util from './webgl_util';\n\nexport function createVertexShader(gl: WebGLRenderingContext): WebGLShader {\n  const glsl = getGlslDifferences();\n  const vertexShaderSource = `${glsl.version}\n    precision highp float;\n    ${glsl.attribute} vec3 clipSpacePos;\n    ${glsl.attribute} vec2 uv;\n    ${glsl.varyingVs} vec2 resultUV;\n\n    void main() {\n      gl_Position = vec4(clipSpacePos, 1);\n      resultUV = uv;\n    }`;\n  return webgl_util.createVertexShader(gl, vertexShaderSource);\n}\n\nexport function createVertexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // [x y z u v] * [upper-left, lower-left, upper-right, lower-right]\n  const vertexArray = new Float32Array(\n      [-1, 1, 0, 0, 1, -1, -1, 0, 0, 0, 1, 1, 0, 1, 1, 1, -1, 0, 1, 0]);\n  return webgl_util.createStaticVertexBuffer(gl, vertexArray);\n}\n\nexport function createIndexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // OpenGL (and WebGL) have \"CCW == front\" winding\n  const triangleVertexIndices = new Uint16Array([0, 1, 2, 2, 1, 3]);\n  return webgl_util.createStaticIndexBuffer(gl, triangleVertexIndices);\n}\n\nfunction createAndConfigureTexture(\n    gl: WebGLRenderingContext, width: number, height: number,\n    internalFormat: number, textureFormat: number,\n    textureType: number): WebGLTexture {\n  webgl_util.validateTextureSize(width, height);\n  const texture = webgl_util.createTexture(gl);\n\n  const tex2d = gl.TEXTURE_2D;\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(tex2d, texture));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MIN_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MAG_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          tex2d, 0, internalFormat, width, height, 0, textureFormat,\n          textureType, null));\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n  return texture;\n}\n\nexport function getInternalFormatForFloat32MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatFloat;\n}\n\nexport function createFloat32MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat32MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatHalfFloat;\n}\n\nexport function createFloat16MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, textureConfig.textureTypeHalfFloat);\n}\n\nexport function getInternalFormatForUnsignedBytesMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.downloadTextureFormat;\n}\n\nexport function createUnsignedBytesMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForUnsignedBytesMatrixTexture(textureConfig), gl.RGBA,\n      gl.UNSIGNED_BYTE);\n}\n\nexport function getInternalFormatForPackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedFloat;\n}\n\nexport function createPackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height, getInternalFormatForPackedMatrixTexture(textureConfig),\n      gl.RGBA, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16PackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedHalfFloat;\n}\n\nexport function createFloat16PackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16PackedMatrixTexture(textureConfig), gl.RGBA,\n      textureConfig.textureTypeHalfFloat);\n}\n\nexport function bindVertexProgramAttributeStreams(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    vertexBuffer: WebGLBuffer): boolean {\n  const posOffset = 0;               // x is the first buffer element\n  const uvOffset = 3 * 4;            // uv comes after [x y z]\n  const stride = (3 * 4) + (2 * 4);  // xyz + uv, each entry is 4-byte float.\n  webgl_util.callAndCheck(\n      gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, vertexBuffer));\n  const success = webgl_util.bindVertexBufferToProgramAttribute(\n      gl, program, 'clipSpacePos', vertexBuffer, 3, stride, posOffset);\n  return success &&\n      webgl_util.bindVertexBufferToProgramAttribute(\n          gl, program, 'uv', vertexBuffer, 2, stride, uvOffset);\n}\n\nexport function uploadDenseMatrixToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture, width: number,\n    height: number, data: TypedArray, textureConfig: TextureConfig) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n\n  let dataForUpload: TypedArray, texelDataType: number, internalFormat: number;\n  if (data instanceof Uint8Array) {\n    dataForUpload = new Uint8Array(width * height * 4);\n    texelDataType = gl.UNSIGNED_BYTE;\n    internalFormat = gl.RGBA;\n  } else {\n    dataForUpload = new Float32Array(width * height * 4);\n    texelDataType = gl.FLOAT;\n    internalFormat = textureConfig.internalFormatPackedFloat;\n  }\n\n  dataForUpload.set(data);\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          gl.TEXTURE_2D, 0, internalFormat, width, height, 0, gl.RGBA,\n          texelDataType, dataForUpload));\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function uploadPixelDataToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, pixels.width, pixels.height, 0, gl.RGBA,\n            gl.UNSIGNED_BYTE, (pixels as PixelData).data));\n  } else {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, gl.RGBA, gl.UNSIGNED_BYTE,\n            pixels as ImageData | HTMLImageElement | HTMLCanvasElement |\n                HTMLVideoElement));\n  }\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function createBufferFromOutputTexture(\n    gl2: WebGL2RenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLBuffer {\n  // Create and bind the buffer.\n  const buffer = gl2.createBuffer();\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer));\n\n  // Initialize the buffer to the size of the texture in bytes.\n  const bytesPerFloat = 4;\n  const valuesPerTexel = 4;\n  const bufferSizeBytes = bytesPerFloat * valuesPerTexel * rows * columns;\n\n  webgl_util.callAndCheck(\n      gl2,\n      () => gl2.bufferData(\n          gl2.PIXEL_PACK_BUFFER, bufferSizeBytes, gl2.STREAM_READ));\n\n  // Enqueue a command on the GPU command queue to copy of texture into the\n  // buffer.\n  webgl_util.callAndCheck(\n      gl2, () => gl2.readPixels(0, 0, columns, rows, gl2.RGBA, gl2.FLOAT, 0));\n\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null));\n\n  return buffer;\n}\n\nexport function downloadFloat32MatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer,\n    size: number): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget = new Float32Array(size);\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadByteEncodedFloatMatrixFromOutputTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig) {\n  const [w, h] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n\n  const numChannels = 4;\n  const downloadTarget = new Uint8Array(\n      tex_util.getUnpackedArraySizeFromMatrixSize(rows * columns, numChannels));\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, w, h, textureConfig.downloadTextureFormat, gl.UNSIGNED_BYTE,\n          downloadTarget));\n\n  // By wrapping the buffer in a Float32Array, we use native browser IEEE 754\n  // decoding of the 4 bytes that back each 32 bit float.\n  return new Float32Array(downloadTarget.buffer);\n}\n\nexport function downloadPackedMatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer, batch: number, rows: number,\n    cols: number, physicalRows: number, physicalCols: number,\n    textureConfig: TextureConfig): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget =\n      new Float32Array(tex_util.getPackedRGBAArraySizeFromMatrixShape(\n          physicalRows, physicalCols));\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadMatrixFromPackedOutputTexture(\n    gl: WebGLRenderingContext, physicalRows: number,\n    physicalCols: number): Float32Array {\n  const packedRGBA = new Float32Array(physicalRows * physicalCols * 4);\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, physicalCols, physicalRows, gl.RGBA, gl.FLOAT, packedRGBA));\n\n  return packedRGBA;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, PixelData, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext, setWebGLContext} from './canvas_util';\nimport * as gpgpu_util from './gpgpu_util';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport {WebGL1DisjointQueryTimerExtension, WebGL2DisjointQueryTimerExtension} from './webgl_types';\nimport * as webgl_util from './webgl_util';\n\nexport interface FenceContext {\n  query: WebGLQuery|WebGLSync;\n  isFencePassed(): boolean;\n}\n\nexport class GPGPUContext {\n  gl: WebGLRenderingContext;\n  textureFloatExtension: {};\n  textureHalfFloatExtension: {};\n  colorBufferFloatExtension: {};\n  colorBufferHalfFloatExtension: {};\n  disjointQueryTimerExtension: WebGL2DisjointQueryTimerExtension|\n      WebGL1DisjointQueryTimerExtension;\n  vertexBuffer: WebGLBuffer;\n  indexBuffer: WebGLBuffer;\n  framebuffer: WebGLFramebuffer;\n  outputTexture: WebGLTexture|null = null;\n  program: WebGLProgram|null = null;\n  private disposed = false;\n  private disjoint: boolean;\n  textureConfig: TextureConfig;\n\n  constructor(gl?: WebGLRenderingContext) {\n    const glVersion = env().getNumber('WEBGL_VERSION');\n    if (gl != null) {\n      this.gl = gl;\n      setWebGLContext(glVersion, gl);\n    } else {\n      this.gl = getWebGLContext(glVersion);\n    }\n    // WebGL 2.0 enables texture floats without an extension.\n    let COLOR_BUFFER_FLOAT = 'WEBGL_color_buffer_float';\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (env().getNumber('WEBGL_VERSION') === 1) {\n      const TEXTURE_FLOAT = 'OES_texture_float';\n      const TEXTURE_HALF_FLOAT = 'OES_texture_half_float';\n\n      this.textureFloatExtension =\n          webgl_util.getExtensionOrThrow(this.gl, TEXTURE_FLOAT);\n      if (webgl_util.hasExtension(this.gl, TEXTURE_HALF_FLOAT)) {\n        this.textureHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, TEXTURE_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support half float textures, yet the ' +\n            'environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n\n      this.colorBufferFloatExtension = this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, COLOR_BUFFER_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support color renderable half floats, yet ' +\n            'the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n    } else {\n      COLOR_BUFFER_FLOAT = 'EXT_color_buffer_float';\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_FLOAT)) {\n        this.colorBufferFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      } else if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      } else {\n        throw new Error('GL context does not support color renderable floats');\n      }\n    }\n\n    this.vertexBuffer = gpgpu_util.createVertexBuffer(this.gl);\n    this.indexBuffer = gpgpu_util.createIndexBuffer(this.gl);\n    this.framebuffer = webgl_util.createFramebuffer(this.gl);\n\n    this.textureConfig =\n        tex_util.getTextureConfig(this.gl, this.textureHalfFloatExtension);\n  }\n\n  private get debug(): boolean {\n    return env().getBool('DEBUG');\n  }\n\n  public dispose() {\n    if (this.disposed) {\n      return;\n    }\n    if (this.program != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound WebGLProgram.' +\n          ' This is probably a resource leak, delete the program with ' +\n          'GPGPUContext.deleteProgram before disposing.');\n    }\n    if (this.outputTexture != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound output matrix ' +\n          'texture.  This is probably a resource leak, delete the output ' +\n          'matrix texture with GPGPUContext.deleteMatrixTexture before ' +\n          'disposing.');\n    }\n    const gl = this.gl;\n    webgl_util.callAndCheck(gl, () => gl.finish());\n    webgl_util.callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteFramebuffer(this.framebuffer));\n    webgl_util.callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(\n        gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteBuffer(this.indexBuffer));\n    this.disposed = true;\n  }\n\n  public createFloat32MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat32MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createFloat16MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createUnsignedBytesMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createUnsignedBytesMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public uploadPixelDataToTexture(\n      texture: WebGLTexture,\n      pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadPixelDataToTexture(this.gl, texture, pixels);\n  }\n\n  public uploadDenseMatrixToTexture(\n      texture: WebGLTexture, width: number, height: number, data: TypedArray) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadDenseMatrixToTexture(\n        this.gl, texture, width, height, data, this.textureConfig);\n  }\n\n  public createFloat16PackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16PackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createPackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createPackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public deleteMatrixTexture(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    if (this.outputTexture === texture) {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n      this.outputTexture = null;\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.deleteTexture(texture));\n  }\n\n  public downloadByteEncodedFloatMatrixFromOutputTexture(\n      texture: WebGLTexture, rows: number, columns: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadByteEncodedFloatMatrixFromOutputTexture(\n            this.gl, rows, columns, this.textureConfig));\n  }\n\n  public downloadPackedMatrixFromBuffer(\n      buffer: WebGLBuffer, batch: number, rows: number, columns: number,\n      physicalRows: number, physicalCols: number): Float32Array {\n    return gpgpu_util.downloadPackedMatrixFromBuffer(\n        this.gl, buffer, batch, rows, columns, physicalRows, physicalCols,\n        this.textureConfig);\n  }\n\n  public downloadFloat32MatrixFromBuffer(buffer: WebGLBuffer, size: number):\n      Float32Array {\n    return gpgpu_util.downloadFloat32MatrixFromBuffer(this.gl, buffer, size);\n  }\n\n  public createBufferFromTexture(\n      texture: WebGLTexture, rows: number, columns: number): WebGLBuffer {\n    this.bindTextureToFrameBuffer(texture);\n    const result = gpgpu_util.createBufferFromOutputTexture(\n        this.gl as WebGL2RenderingContext, rows, columns, this.textureConfig);\n    this.unbindTextureToFrameBuffer();\n    return result;\n  }\n\n  public createAndWaitForFence(): Promise<void> {\n    const fenceContext = this.createFence(this.gl);\n    return this.pollFence(fenceContext);\n  }\n\n  private createFence(gl: WebGLRenderingContext): FenceContext {\n    let query: WebGLQuery|WebGLSync;\n    let isFencePassed: () => boolean;\n\n    if (env().getBool('WEBGL_FENCE_API_ENABLED')) {\n      const gl2 = gl as WebGL2RenderingContext;\n\n      const sync = gl2.fenceSync(gl2.SYNC_GPU_COMMANDS_COMPLETE, 0);\n      gl.flush();\n\n      isFencePassed = () => {\n        const status = gl2.clientWaitSync(sync, 0, 0);\n        return status === gl2.ALREADY_SIGNALED ||\n            status === gl2.CONDITION_SATISFIED;\n      };\n\n      query = sync;\n    } else if (\n        env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0) {\n      query = this.beginQuery();\n      this.endQuery();\n      isFencePassed = () => this.isQueryAvailable(\n          query,\n          env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n    } else {\n      // If we have no way to fence, return true immediately. This will fire in\n      // WebGL 1.0 when there is no disjoint query timer. In this case, because\n      // the fence passes immediately, we'll immediately ask for a download of\n      // the texture, which will cause the UI thread to hang.\n      isFencePassed = () => true;\n    }\n\n    return {query, isFencePassed};\n  }\n\n  public downloadMatrixFromPackedTexture(\n      texture: WebGLTexture, physicalRows: number,\n      physicalCols: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadMatrixFromPackedOutputTexture(\n            this.gl, physicalRows, physicalCols));\n  }\n\n  private vertexAttrsAreBound = false;\n\n  public createProgram(fragmentShaderSource: string): WebGLProgram {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    const fragmentShader: WebGLShader =\n        webgl_util.createFragmentShader(gl, fragmentShaderSource);\n    const vertexShader: WebGLShader = gpgpu_util.createVertexShader(gl);\n    const program: WebGLProgram = webgl_util.createProgram(gl);\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, vertexShader));\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, fragmentShader));\n    webgl_util.linkProgram(gl, program);\n    if (this.debug) {\n      webgl_util.validateProgram(gl, program);\n    }\n    if (!this.vertexAttrsAreBound) {\n      this.setProgram(program);\n      this.vertexAttrsAreBound = gpgpu_util.bindVertexProgramAttributeStreams(\n          gl, this.program, this.vertexBuffer);\n    }\n    return program;\n  }\n\n  public deleteProgram(program: WebGLProgram) {\n    this.throwIfDisposed();\n    if (program === this.program) {\n      this.program = null;\n    }\n    if (program != null) {\n      webgl_util.callAndCheck(this.gl, () => this.gl.deleteProgram(program));\n    }\n  }\n\n  public setProgram(program: WebGLProgram|null) {\n    this.throwIfDisposed();\n    this.program = program;\n    if ((this.program != null) && this.debug) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.useProgram(program));\n  }\n\n  public getUniformLocation(\n      program: WebGLProgram, uniformName: string,\n      shouldThrow = true): WebGLUniformLocation {\n    this.throwIfDisposed();\n    if (shouldThrow) {\n      return webgl_util.getProgramUniformLocationOrThrow(\n          this.gl, program, uniformName);\n    } else {\n      return webgl_util.getProgramUniformLocation(\n          this.gl, program, uniformName);\n    }\n  }\n\n  public getAttributeLocation(program: WebGLProgram, attribute: string):\n      number {\n    this.throwIfDisposed();\n    return webgl_util.callAndCheck(\n        this.gl, () => this.gl.getAttribLocation(program, attribute));\n  }\n\n  public getUniformLocationNoThrow(program: WebGLProgram, uniformName: string):\n      WebGLUniformLocation {\n    this.throwIfDisposed();\n    return this.gl.getUniformLocation(program, uniformName);\n  }\n\n  public setInputMatrixTexture(\n      inputMatrixTexture: WebGLTexture, uniformLocation: WebGLUniformLocation,\n      textureUnit: number) {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    webgl_util.bindTextureToProgramUniformSampler(\n        this.gl, inputMatrixTexture, uniformLocation, textureUnit);\n  }\n\n  public setOutputMatrixTexture(\n      outputMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.setOutputMatrixTextureDriver(outputMatrixTexture, columns, rows);\n  }\n\n  public setOutputPackedMatrixTexture(\n      outputPackedMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.throwIfDisposed();\n    const [width, height] =\n        tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n    this.setOutputMatrixTextureDriver(outputPackedMatrixTexture, width, height);\n  }\n\n  public setOutputMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    this.setOutputMatrixWriteRegionDriver(\n        startColumn, startRow, numColumns, numRows);\n  }\n\n  public setOutputPackedMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    throw new Error('setOutputPackedMatrixWriteRegion not implemented.');\n  }\n\n  public debugValidate() {\n    if (this.program != null) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.validateFramebuffer(this.gl);\n  }\n\n  public executeProgram() {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    const gl = this.gl;\n    if (this.debug) {\n      this.debugValidate();\n    }\n    webgl_util.callAndCheck(\n        gl, () => gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_SHORT, 0));\n  }\n\n  public blockUntilAllProgramsCompleted() {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(this.gl, () => this.gl.finish());\n  }\n\n  private getQueryTimerExtension(): WebGL1DisjointQueryTimerExtension\n      |WebGL2DisjointQueryTimerExtension {\n    if (this.disjointQueryTimerExtension == null) {\n      this.disjointQueryTimerExtension =\n          webgl_util.getExtensionOrThrow(\n              this.gl,\n              env().getNumber(\n                  'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2 ?\n                  'EXT_disjoint_timer_query_webgl2' :\n                  'EXT_disjoint_timer_query') as\n              WebGL1DisjointQueryTimerExtension |\n          WebGL2DisjointQueryTimerExtension;\n    }\n    return this.disjointQueryTimerExtension;\n  }\n\n  private getQueryTimerExtensionWebGL2(): WebGL2DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension();\n  }\n\n  private getQueryTimerExtensionWebGL1(): WebGL1DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension() as WebGL1DisjointQueryTimerExtension;\n  }\n\n  beginQuery(): WebGLQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const query = gl2.createQuery();\n      gl2.beginQuery(ext.TIME_ELAPSED_EXT, query);\n      return query;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    const query = ext.createQueryEXT() as WebGLQuery;\n    ext.beginQueryEXT(ext.TIME_ELAPSED_EXT, query);\n    return query;\n  }\n\n  endQuery() {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n      gl2.endQuery(ext.TIME_ELAPSED_EXT);\n      return;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    ext.endQueryEXT(ext.TIME_ELAPSED_EXT);\n  }\n\n  public async waitForQueryAndGetTime(query: WebGLQuery): Promise<number> {\n    await util.repeatedTry(\n        () => this.disposed ||  // while testing contexts are created / disposed\n                                // in rapid succession, so without this check we\n                                // may poll for the query timer indefinitely\n            this.isQueryAvailable(\n                query,\n                env().getNumber(\n                    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION')));\n    return this.getQueryTime(\n        query, env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n  }\n\n  private getQueryTime(query: WebGLQuery, queryTimerVersion: number): number {\n    if (queryTimerVersion === 0) {\n      return null;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n\n      const timeElapsedNanos = gl2.getQueryParameter(query, gl2.QUERY_RESULT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const timeElapsedNanos =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_EXT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    }\n  }\n\n  private isQueryAvailable(query: WebGLQuery, queryTimerVersion: number):\n      boolean {\n    if (queryTimerVersion === 0) {\n      return true;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const available =\n          gl2.getQueryParameter(query, gl2.QUERY_RESULT_AVAILABLE);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const available =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_AVAILABLE_EXT);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    }\n  }\n\n  pollFence(fenceContext: FenceContext) {\n    return new Promise<void>(resolve => {\n      this.addItemToPoll(() => fenceContext.isFencePassed(), () => resolve());\n    });\n  }\n\n  private itemsToPoll: PollItem[] = [];\n\n  pollItems(): void {\n    // Find the last query that has finished.\n    const index = linearSearchLastTrue(this.itemsToPoll.map(x => x.isDoneFn));\n    for (let i = 0; i <= index; ++i) {\n      const {resolveFn} = this.itemsToPoll[i];\n      resolveFn();\n    }\n    this.itemsToPoll = this.itemsToPoll.slice(index + 1);\n  }\n\n  private addItemToPoll(isDoneFn: () => boolean, resolveFn: () => void) {\n    this.itemsToPoll.push({isDoneFn, resolveFn});\n    if (this.itemsToPoll.length > 1) {\n      // We already have a running loop that polls.\n      return;\n    }\n    // Start a new loop that polls.\n    util.repeatedTry(() => {\n      this.pollItems();\n      // End the loop if no more items to poll.\n      return this.itemsToPoll.length === 0;\n    });\n  }\n\n  private bindTextureToFrameBuffer(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    webgl_util.bindColorTextureToFramebuffer(\n        this.gl, texture, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(this.gl);\n    }\n  }\n\n  private unbindTextureToFrameBuffer() {\n    if (this.outputTexture != null) {\n      webgl_util.bindColorTextureToFramebuffer(\n          this.gl, this.outputTexture, this.framebuffer);\n      if (this.debug) {\n        webgl_util.validateFramebuffer(this.gl);\n      }\n    } else {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n    }\n  }\n\n  private downloadMatrixDriver(\n      texture: WebGLTexture,\n      downloadAndDecode: () => Float32Array): Float32Array {\n    this.bindTextureToFrameBuffer(texture);\n    const result = downloadAndDecode();\n    this.unbindTextureToFrameBuffer();\n\n    return result;\n  }\n\n  private setOutputMatrixTextureDriver(\n      outputMatrixTextureMaybePacked: WebGLTexture, width: number,\n      height: number) {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    webgl_util.bindColorTextureToFramebuffer(\n        gl, outputMatrixTextureMaybePacked, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(gl);\n    }\n    this.outputTexture = outputMatrixTextureMaybePacked;\n    webgl_util.callAndCheck(gl, () => gl.viewport(0, 0, width, height));\n    webgl_util.callAndCheck(gl, () => gl.scissor(0, 0, width, height));\n  }\n\n  private setOutputMatrixWriteRegionDriver(\n      x: number, y: number, width: number, height: number) {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(\n        this.gl, () => this.gl.scissor(x, y, width, height));\n  }\n\n  private throwIfDisposed() {\n    if (this.disposed) {\n      throw new Error('Attempted to use disposed GPGPUContext.');\n    }\n  }\n\n  private throwIfNoProgram() {\n    if (this.program == null) {\n      throw new Error('No GPU program is currently set.');\n    }\n  }\n}\n\ntype PollItem = {\n  isDoneFn: () => boolean,\n  resolveFn: () => void\n};\n\n/**\n * Finds the index of the last true element using linear search.\n * Note: We can't do binary search because Chrome expects us to explicitly\n * test all fences before download:\n * https://github.com/tensorflow/tfjs/issues/1145\n */\nexport function linearSearchLastTrue(arr: Array<() => boolean>): number {\n  let i = 0;\n  for (; i < arr.length; ++i) {\n    const isDone = arr[i]();\n    if (!isDone) {\n      break;\n    }\n  }\n  return i - 1;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, Tensor, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport * as shader_compiler from './shader_compiler';\nimport {InputInfo, ShapeInfo} from './shader_compiler';\nimport {PackingScheme, TextureData, TextureUsage} from './tex_util';\n\nexport interface GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  /** If true, this program expects packed input textures. Defaults to false. */\n  packedInputs?: boolean;\n  /** If true, this program produces a packed texture. Defaults to false. */\n  packedOutput?: boolean;\n  /**\n   * Affects what type of texture we allocate for the output. Defaults to\n   * `TextureUsage.RENDER`.\n   */\n  outTexUsage?: TextureUsage;\n  /**\n   * The type of scheme to use when packing texels for the output values.\n   * See `PackingScheme` for details. Defaults to `PackingScheme.SHARED_BATCH`.\n   */\n  outPackingScheme?: PackingScheme;\n}\n\nexport interface GPGPUBinary {\n  webGLProgram: WebGLProgram;\n  program: GPGPUProgram;\n  uniformLocations: {[name: string]: WebGLUniformLocation};\n  source: string;\n  inShapeInfos: ShapeInfo[];\n  outShapeInfo: ShapeInfo;\n  infLoc: WebGLUniformLocation;\n  nanLoc: WebGLUniformLocation;\n}\n\nexport interface TensorData {\n  shape: number[];\n  texData: TextureData;\n  isUniform: boolean;\n  // Available when we decide to upload as uniform instead of texture.\n  uniformValues?: TypedArray;\n}\n\nexport function compileProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, program: GPGPUProgram, inputs: TensorData[],\n    output: TensorData): GPGPUBinary {\n  const userCode = program.userCode;\n  const inputInfos: InputInfo[] = inputs.map((input, i) => {\n    const shapeInfo: ShapeInfo = {\n      logicalShape: input.shape,\n      texShape: input.isUniform ? null : input.texData.texShape,\n      isUniform: input.isUniform,\n      isPacked: input.isUniform ? false : input.texData.isPacked,\n      flatOffset: null\n    };\n    if (input.texData != null && input.texData.slice != null &&\n        input.texData.slice.flatOffset > 0) {\n      shapeInfo.flatOffset = input.texData.slice.flatOffset;\n    }\n    return {name: program.variableNames[i], shapeInfo};\n  });\n  const inShapeInfos = inputInfos.map(x => x.shapeInfo);\n  const outShapeInfo: ShapeInfo = {\n    logicalShape: output.shape,\n    texShape: output.texData.texShape,\n    isUniform: false,\n    isPacked: output.texData.isPacked,\n    flatOffset: null\n  };\n  const source = shader_compiler.makeShader(\n      inputInfos, outShapeInfo, userCode, program.packedInputs);\n\n  const webGLProgram = gpgpu.createProgram(source);\n\n  // Add special uniforms (NAN, INFINITY)\n  let infLoc: WebGLUniformLocation = null;\n  const nanLoc = gpgpu.getUniformLocation(webGLProgram, 'NAN', false);\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    infLoc = gpgpu.getUniformLocation(webGLProgram, 'INFINITY', false);\n  }\n\n  // Add user-defined uniforms\n  const uniformLocations: {[name: string]: WebGLUniformLocation} = {};\n  for (let i = 0; i < program.variableNames.length; i++) {\n    const varName = program.variableNames[i];\n    const shouldThrow = false;\n    uniformLocations[varName] =\n        gpgpu.getUniformLocation(webGLProgram, varName, shouldThrow);\n    uniformLocations[`offset${varName}`] =\n        gpgpu.getUniformLocation(webGLProgram, `offset${varName}`, shouldThrow);\n  }\n\n  return {\n    program,\n    source,\n    webGLProgram,\n    uniformLocations,\n    inShapeInfos,\n    outShapeInfo,\n    infLoc,\n    nanLoc,\n  };\n}\n\nfunction validateBinaryAndProgram(\n    shapeInfos: ShapeInfo[], inputs: TensorData[]) {\n  if (shapeInfos.length !== inputs.length) {\n    throw Error(\n        `Binary was compiled with ${shapeInfos.length} inputs, but ` +\n        `was executed with ${inputs.length} inputs`);\n  }\n\n  shapeInfos.forEach((s, i) => {\n    const shapeA = s.logicalShape;\n    const input = inputs[i];\n    const shapeB = input.shape;\n\n    if (!util.arraysEqual(shapeA, shapeB)) {\n      throw Error(\n          `Binary was compiled with different shapes than ` +\n          `the current args. Shapes ${shapeA} and ${shapeB} must match`);\n    }\n    // The input is uploaded as uniform.\n    if (s.isUniform && input.isUniform) {\n      return;\n    }\n\n    const texShapeA = s.texShape;\n    const texShapeB = input.isUniform ? null : input.texData.texShape;\n    if (!util.arraysEqual(texShapeA, texShapeB)) {\n      throw Error(\n          `Binary was compiled with different texture shapes than the` +\n          ` current args. Shape ${texShapeA} and ${texShapeB} must match`);\n    }\n  });\n}\n\nexport function runProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, binary: GPGPUBinary, inputs: TensorData[],\n    output: TensorData,\n    customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) =>\n        void): void {\n  validateBinaryAndProgram(binary.inShapeInfos, inputs);\n  validateBinaryAndProgram([binary.outShapeInfo], [output]);\n\n  const outTex = output.texData.texture;\n  const outTexShape = output.texData.texShape;\n  if (output.texData.isPacked) {\n    gpgpu.setOutputPackedMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  } else {\n    gpgpu.setOutputMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  }\n  gpgpu.setProgram(binary.webGLProgram);\n\n  // Set special uniforms (NAN, INFINITY)\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    if (binary.infLoc !== null) {\n      gpgpu.gl.uniform1f(binary.infLoc, Infinity);\n    }\n  }\n  if (binary.nanLoc !== null) {\n    gpgpu.gl.uniform1f(binary.nanLoc, NaN);\n  }\n\n  // Set user-defined inputs\n  inputs.forEach((input, i) => {\n    const varName = binary.program.variableNames[i];\n    const varLoc = binary.uniformLocations[varName];\n    const varOffsetLoc = binary.uniformLocations[`offset${varName}`];\n\n    if (varLoc == null) {\n      // The compiler inferred that this variable is not used in this shader.\n      return;\n    }\n\n    if (input.isUniform) {\n      // Upload the values of the tensor as uniform.\n      if (util.sizeFromShape(input.shape) < 2) {\n        gpgpu.gl.uniform1f(varLoc, input.uniformValues[0]);\n      } else {\n        let vals = input.uniformValues;\n        if (!(vals instanceof Float32Array)) {\n          vals = new Float32Array(vals);\n        }\n        gpgpu.gl.uniform1fv(varLoc, vals);\n      }\n      return;\n    }\n\n    // If the input was sliced, upload the flat offset index.\n    if (input.texData.slice != null && varOffsetLoc != null) {\n      gpgpu.gl.uniform1i(varOffsetLoc, input.texData.slice.flatOffset);\n    }\n\n    gpgpu.setInputMatrixTexture(input.texData.texture, varLoc, i);\n  });\n\n  if (customSetup != null) {\n    customSetup(gpgpu, binary.webGLProgram);\n  }\n  gpgpu.executeProgram();\n}\n\nexport function makeShaderKey(\n    program: GPGPUProgram, inputs: TensorData[], output: TensorData): string {\n  let keyInputs = '';\n  inputs.concat(output).forEach(x => {\n    const hasOffset = x.texData != null && x.texData.slice != null &&\n        x.texData.slice.flatOffset > 0;\n    const texShape = x.isUniform ? 'uniform' : x.texData.texShape;\n    keyInputs += `${x.shape}_${texShape}_${hasOffset}`;\n  });\n  const keyUserCode = program.userCode;\n  let key = program.constructor.name;\n  // Fast string concat. See https://jsperf.com/string-concatenation/14.\n  key += '_' + keyInputs + '_' + keyUserCode;\n  return key;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Im2ColPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      outputShape: number[], inputShape: number[],\n      convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = outputShape;\n\n    const {\n      filterWidth,\n      inChannels,\n      strideWidth,\n      strideHeight,\n      padInfo,\n      outWidth,\n      dilationWidth,\n      dilationHeight,\n      dataFormat\n    } = convInfo;\n    const {left, top} = padInfo;\n    const itemsPerBlockRow = inChannels * filterWidth;\n    const glsl = getGlslDifferences();\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const rowDim = isChannelsLast ? 0 : 1;\n    const colDim = isChannelsLast ? 1 : 2;\n\n    let unrolled = ``;\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        unrolled += `\n          blockIndex = rc.y + ${col};\n          pos = rc.x + ${row};\n\n          if(blockIndex < ${outputShape[1]} && pos < ${outputShape[0]}) {\n            offsetY = int(blockIndex / (${outWidth})) * ${strideHeight} - ${\n            top};\n            d0 = offsetY + ${dilationHeight} * (pos / ${itemsPerBlockRow});\n\n            if(d0 < ${inputShape[rowDim]} && d0 >= 0) {\n\n              offsetX = int(mod(float(blockIndex), ${outWidth}.) * ${\n            strideWidth}. - ${left}.);\n              d1 = offsetX + ${dilationWidth} * (int(mod(float(pos), ${\n            itemsPerBlockRow}.) / ${inChannels}.));\n\n              if(d1 < ${inputShape[colDim]} && d1 >= 0) {\n\n                ch = int(mod(float(pos), ${inChannels}.));\n\n                if (${isChannelsLast}) {\n                  innerDims = vec2(d1, ch);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(d0, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                } else {\n                  innerDims = vec2(d0, d1);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(ch, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                }\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      void main() {\n        ivec2 rc = getOutputCoords();\n\n        vec4 result = vec4(0);\n\n        int blockIndex, pos, offsetY, d0, offsetX, d1, ch;\n        vec2 innerDims;\n\n        ${unrolled}\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n        int d = coords[3];\n        float x = getX(b, r, c, d);\n        float sum = 0.0;\n        for (int j = -${rad}; j <= ${rad}; j++) {\n          int idx = d + j;\n          if (idx >= 0 && idx <=  ${maxD}) {\n            float z = getX(b, r, c, idx);\n            sum += z * z;\n          }\n        }\n        float val = x * ${powOperator};\n        setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNGradProgram implements GPGPUProgram {\n  variableNames = ['inputImage', 'outputImage', 'dy'];\n  outputShape: number[] = [];\n  userCode: string;\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n  depth: number;\n\n  constructor(\n      inputShape: number[], depthRadius: number, bias: number, alpha: number,\n      beta: number) {\n    this.outputShape = inputShape;\n    this.depth = inputShape[3];\n    this.depthRadius = depthRadius;\n    this.bias = bias;\n    this.alpha = alpha;\n    this.beta = beta;\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n\n        float result = 0.0;\n        for (int d = 0; d < ${this.depth}; ++d) {\n          int depthBegin = int(max(0.0, float(d - ${depthRadius})));\n          int depthEnd = int(min(float(${this.depth}),\n              float(d + ${depthRadius} + 1)));\n\n          const int MIN_DEPTH_BEGIN = 0;\n          const int MAX_DEPTH_END = ${this.depth};\n\n          float norm = 0.0;\n          for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd) {\n              norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);\n            }\n            else {\n              break;\n            }\n          }\n\n          norm = float(${alpha}) * norm + float(${bias});\n\n          for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd){\n              float dyi = -2.0 * float(${alpha})\n                * float(${beta})\n                * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)\n                / norm;\n              if (k == d) {\n                dyi += pow(norm, -1.0 * ${beta});\n              }\n              if (k == coords[3]) {\n                dyi *= getDy(b, r, c, d);\n                result += dyi;\n              }\n            }\n            else {\n              break;\n            }\n          }\n      }\n      setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords.x;\n        int r = coords.y;\n        int c = coords.z;\n        int d = coords.w;\n\n        bool hasNextCol = d < ${this.outputShape[3]};\n        bool hasNextRow = c < ${this.outputShape[2]};\n\n        vec4 sum = vec4(0.);\n        vec4 xFragAtOutputCoords = getX(b, r, c, d);\n\n        vec4 xAtOutputCoords = vec4(\n          getChannel(xFragAtOutputCoords, vec2(c, d)),\n          hasNextCol ?\n            getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,\n          hasNextRow ?\n            getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0\n        );\n\n        int firstChannel = d - ${rad};\n        vec2 cache = vec2(0.);\n        if(firstChannel >= 0){\n          vec4 firstChannelFrag = getX(b, r, c, firstChannel);\n          cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));\n            if(hasNextRow){\n              cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));\n            }\n        }\n\n        ivec2 depth = ivec2(d, d + 1);\n        for (int j = - ${rad}; j <= ${rad}; j++) {\n          ivec2 idx = depth + j;\n          bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));\n          bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${maxD}));\n\n          bool depthInRange = aboveLowerBound.x && belowUpperBound.x;\n          bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;\n\n          if(depthInRange || depthPlusOneInRange){\n            vec4 z = vec4(0.);\n            vec4 xFragAtCurrentDepth;\n            z.xz = cache.xy;\n            if(depthPlusOneInRange && hasNextCol){\n              xFragAtCurrentDepth = idx.y != d ?\n                getX(b, r, c, idx.y) : xFragAtOutputCoords;\n              z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));\n              if(hasNextRow){\n                z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));\n              }\n            }\n            cache.xy = z.yw;\n            sum += z * z;\n          }\n        }\n        vec4 result = xAtOutputCoords * ${powOperator};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MaxPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex = effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n          wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n            int maxPosValue = ${lastIndex} - int(getMaxPos(b, idyR, idyC, d));\n\n            // Get the current value, check it against the value from the\n            // position matrix.\n            int curPosValue = wR * ${effectiveFilterWidth} + wC;\n            float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n            dotProd += dyValue * mask;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class MaxPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex =\n        effectiveFilterDepth * effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n           wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n              int maxPosValue = ${lastIndex} -\n                  int(getMaxPos(batch, idyD, idyR, idyC, ch));\n\n              // Get the current value, check it against the value from the\n              // position matrix.\n              int curPosValue =\n                  wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                  wR * ${effectiveFilterWidth} + wC;\n              float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n              dotProd += dyValue * mask;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MatMulPackedProgram implements GPGPUProgram {\n  variableNames = ['matrixA', 'matrixB'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      aShape: [number, number, number], outputShape: [number, number, number],\n      transposeA = false, transposeB = false, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = outputShape;\n\n    const sharedDim = transposeA ? aShape[1] : aShape[2];\n    const sharedDimensionPacked = Math.ceil(sharedDim / 2);\n\n    const aSample = transposeA ? 'i * 2, rc.y' : 'rc.y, i * 2';\n    const bSample = transposeB ? 'rc.z, i * 2' : 'i * 2, rc.z';\n    const aSwizzle = transposeA ? ['a.xxyy', 'a.zzww'] : ['a.xxzz', 'a.yyww'];\n    const bSwizzle = transposeB ? ['b.xzxz', 'b.ywyw'] : ['b.xyxy', 'b.zwzw'];\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const float sharedDimension = ${sharedDimensionPacked}.0;\n\n      vec4 dot2x2ARowBCol(ivec3 rc) {\n        vec4 result = vec4(0);\n        for (int i = 0; i < ${sharedDimensionPacked}; i++) {\n          vec4 a = getMatrixA(rc.x, ${aSample});\n          vec4 b = getMatrixB(rc.x, ${bSample});\n\n          // These swizzled products need to be separately added.\n          // See: https://github.com/tensorflow/tfjs/issues/1735\n          result += (${aSwizzle[0]} * ${bSwizzle[0]});\n          result += (${aSwizzle[1]} * ${bSwizzle[1]});\n        }\n        return result;\n      }\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n        vec4 result = dot2x2ARowBCol(rc);\n\n        ${addBiasSnippet}\n\n        ${applyActivationSnippet}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MultinomialProgram implements GPGPUProgram {\n  variableNames = ['probs'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(batchSize: number, numOutcomes: number, numSamples: number) {\n    this.outputShape = [batchSize, numSamples];\n\n    this.userCode = `\n      uniform float seed;\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n\n        float r = random(seed);\n        float cdf = 0.0;\n\n        for (int i = 0; i < ${numOutcomes - 1}; i++) {\n          cdf += getProbs(batch, i);\n\n          if (r < cdf) {\n            setOutput(float(i));\n            return;\n          }\n        }\n\n        // If no other event happened, last event happened.\n        setOutput(float(${numOutcomes - 1}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(seed: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.seedLoc == null) {\n        this.seedLoc = gpgpu.getUniformLocation(webGLProgram, 'seed');\n      }\n      gpgpu.gl.uniform1f(this.seedLoc, seed);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class OneHotProgram implements GPGPUProgram {\n  variableNames = ['indices'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(\n      numIndices: number, depth: number, onValue: number, offValue: number) {\n    this.outputShape = [numIndices, depth];\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int index = round(getIndices(coords.x));\n        setOutput(mix(float(${offValue}), float(${onValue}),\n                      float(index == coords.y)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape:\n          number[]) {  // TODO(https://github.com/tensorflow/tfjs/issues/893):\n                       // Only input / output 3D tensors.\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    if (rank === 0) {\n      this.userCode = `\n        void main() {\n          setOutput(vec4(getA(), 0., 0., 0.));\n        }\n      `;\n    } else {\n      const channels = getChannels('rc', rank);\n      const dtype = getCoordsDataType(rank);\n      const outOfBoundsCondition =\n          getOutOfBoundsCondition(rank, outputShape, channels);\n      const setup = getSetup(\n          rank, outputShape[outputShape.length - 1],\n          outputShape[outputShape.length - 2], channels);\n      const output = getOutput(outputShape, channels);\n\n      this.userCode = `\n        void main() {\n          ${dtype} rc = getOutputCoords();\n\n          if(${outOfBoundsCondition}) {\n            setOutput(vec4(0));\n          } else {\n            ${setup}\n\n            setOutput(vec4(${output}));\n          }\n        }\n      `;\n    }\n  }\n}\n\nfunction getSourceCoordsArr(rank: number, dims: string[]): string[] {\n  const coords = [];\n\n  for (let row = 0; row <= 1; row++) {\n    for (let col = 0; col <= 1; col++) {\n      let coord = `${row === 0 ? 'r' : 'rp1'}, ${col === 0 ? 'c' : 'cp1'}`;\n\n      for (let d = 2; d < rank; d++) {\n        coord = `${dims[dims.length - 1 - d]},` + coord;\n      }\n\n      coords.push(coord);\n    }\n  }\n  return coords;\n}\n\nfunction getOutOfBoundsCondition(\n    rank: number, shape: number[], dims: string[]): string {\n  if (rank === 1) {\n    return `rc > ${shape[0]}`;\n  }\n\n  let cond = '';\n  for (let i = rank - 2; i < rank; i++) {\n    cond += `${dims[i]} >= ${shape[i]}`;\n    if (i < rank - 1) {\n      cond += '||';\n    }\n  }\n\n  return cond;\n}\n\nfunction getSetup(\n    rank: number, cols: number, rows: number, dims: string[]): string {\n  if (rank === 1) {\n    return '';\n  }\n\n  const innerDims = dims.slice(-2);\n\n  return `\n    int r = ${innerDims[0]};\n    int c = ${innerDims[1]};\n    int rp1 = r + 1;\n    int cp1 = c + 1;\n\n    bool cEdge = cp1 >= ${cols};\n    bool rEdge = rp1 >= ${rows};\n  `;\n}\n\nfunction getOutput(shape: number[], dims: string[]): string {\n  const rank = shape.length;\n  const sourceCoords = getSourceCoordsArr(rank, dims);\n  if (rank === 1) {\n    return `getA(rc),\n            rc + 1 >= ${shape[0]} ? 0. : getA(rc + 1),\n            0, 0`;\n  }\n\n  return `getA(${sourceCoords[0]}),\n          cEdge ? 0. : getA(${sourceCoords[1]}),\n          rEdge ? 0. : getA(${sourceCoords[2]}),\n          rEdge || cEdge ? 0. : getA(${sourceCoords[3]})`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const type = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const unpackedCoords =\n        ['coords[0]', 'coords[1]', 'coords[2]', 'coords[3]'].slice(0, rank);\n\n    if (rank === 1) {\n      this.userCode = `\n        int start = ${start};\n        int end = ${end};\n\n        void main() {\n          int outC = getOutputCoords();\n          if (outC < start || outC >= end) {\n            setOutput(float(${constantValue}));\n          } else {\n            setOutput(getX(outC - start));\n          }\n        }\n      `;\n      return;\n    }\n    this.userCode = `\n      ${type} start = ${type}(${start});\n      ${type} end = ${type}(${end});\n\n      void main() {\n        ${type} outC = getOutputCoords();\n        if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {\n          setOutput(float(${constantValue}));\n        } else {\n          ${type} coords = outC - start;\n          setOutput(getX(${unpackedCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const dtype = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const coords = getChannels('rc', rank);\n    const source = getChannels('source', rank);\n    const cLimit = `${coords[rank - 1]} < ${this.outputShape[rank - 1]}`;\n    const innerDims =\n        rank === 1 ? 'source' : `vec2(${source.slice(-2).join()})`;\n\n    const componentSetup = [\n      `${dtype} rc = outputLoc;`, `${coords[rank - 1]} += 1;\n       if(${cLimit}) {\n      `,\n      rank === 1 ? '' : `}\n       rc = outputLoc;\n       ${coords[rank - 2]} += 1;\n       if(${coords[rank - 2]} < ${this.outputShape[rank - 2]}) {`,\n      rank === 1 ? '' : `  ${coords[rank - 1]} += 1;\n         if(${cLimit}) {`\n    ];\n\n    const paddingArea = rank === 1 ?\n        'rc < start || rc >= end' :\n        'any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))';\n    let mainLoop = '';\n    for (let i = 0, j = rank === 1 ? 2 : 4; i < j; i++) {\n      mainLoop += `\n        ${componentSetup[i]}\n        if (${paddingArea}) {\n          result[${i}] = float(${constantValue});\n        } else {\n          ${dtype} source = rc - start;\n          result[${i}] = getChannel(getX(${source.join()}), ${innerDims});\n        }\n      `;\n    }\n    mainLoop += (rank === 1 ? `} ` : `}}`);\n\n    this.userCode = `\n      const ${dtype} start = ${dtype}(${start});\n      const ${dtype} end = ${dtype}(${end});\n\n      void main() {\n        ${dtype} outputLoc = getOutputCoords();\n        vec4 result = vec4(0.);\n        ${mainLoop}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Pool2DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n    const batchFlattenPositionStr = `((batch  * ${convInfo.inHeight} + xR) * ${\n        convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n    const flattenPositionStr =\n        `(xR * ${convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n        const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int batch = coords[0];\n          int d = coords[3];\n\n          ivec2 xRCCorner = coords.yz * strides - pads;\n          int xRCorner = xRCCorner.x;\n          int xCCorner = xRCCorner.y;\n\n          // max/min x(?, ?, d) to get y(yR, yC, d).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n          float avgValue = 0.0;\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              int xC = xCCorner + wC;\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float value = getX(batch, xR, xC, d);\n\n              // If a min / max value has already been found, use it. If not,\n              // use the current value.\n              float currMinMaxValue = mix(\n                  value, minMaxValue, minMaxValueFound);\n              if (value ${compareOp} currMinMaxValue) {\n                minMaxValue = value;\n                minMaxValueFound = 1.0;\n                minMaxPosition = ${\n          flattenPositions ? (includeBatchInIndex ? batchFlattenPositionStr :\n                                                    flattenPositionStr) :\n                             `wR * ${effectiveFilterWidth} + wC`};\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xR, int xC, int d) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xR, xC, d);\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d = coords[3];\n\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // max/min x(?, ?, d) to get y(yR, yC, d).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          int xR = xRCorner + wR;\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              getValue(batch, xR, xC + 3 * ${dilationWidth}, d)\n            );\n\n            ${updateSnippet}\n          }\n\n          int xC = xCCorner + ${filterWidthNearestVec4};\n          if (${filterWidthVec4Remainder === 1}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              initializationValue,\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 2}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 3}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              initializationValue\n            );\n\n            ${updateSnippet}\n          }\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n\nexport class Pool3DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv3DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec3 strides =\n            ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n        const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n        void main() {\n          ivec5 coords = getOutputCoords();\n          int batch = coords.x;\n          int ch = coords.u;\n\n          ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n          int xDCorner = xCorner.x;\n          int xRCorner = xCorner.y;\n          int xCCorner = xCorner.z;\n\n          // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n\n          for (int wD = 0; wD < ${effectiveFilterDepth};\n              wD += ${dilationDepth}) {\n            int xD = xDCorner + wD;\n\n            if (xD < 0 || xD >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int wR = 0; wR < ${effectiveFilterHeight};\n                wR += ${dilationHeight}) {\n              int xR = xRCorner + wR;\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int wC = 0; wC < ${effectiveFilterWidth};\n                  wC += ${dilationWidth}) {\n                int xC = xCCorner + wC;\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float value = getX(batch, xD, xR, xC, ch);\n\n                // If a min / max value has already been found, use it. If not,\n                // use the current value.\n                float currMinMaxValue = mix(\n                    value, minMaxValue, minMaxValueFound);\n                if (value ${compareOp} currMinMaxValue) {\n                  minMaxValue = value;\n                  minMaxValueFound = 1.0;\n                  minMaxPosition = ${\n          flattenPositions ?\n              (includeBatchInIndex ?\n                   `(((batch * ${convInfo.inDepth} + xD) * ${\n                       convInfo.inHeight} + xR) * ${convInfo.inWidth} + xC) * ${\n                       convInfo.inChannels} + ch` :\n                   `((xD * ${convInfo.inHeight} + xR) * ${\n                       convInfo.inWidth} + xC) * ${convInfo.inChannels} + ch`) :\n              `wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                      wR * ${effectiveFilterWidth} + wC`};\n                }\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec3 strides =\n        ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xD, int xR, int xC, int ch) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xD, xR, xC, ch);\n      }\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xDCorner = xCorner.x;\n        int xRCorner = xCorner.y;\n        int xCCorner = xCorner.z;\n\n        // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          int xD = xDCorner + wD;\n\n          if (xD < 0 || xD >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 3 * ${dilationWidth}, ch)\n              );\n\n              ${updateSnippet}\n            }\n\n            int xC = xCCorner + ${filterWidthNearestVec4};\n            if (${filterWidthVec4Remainder === 1}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                initializationValue,\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 2}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 3}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                initializationValue\n              );\n\n              ${updateSnippet}\n            }\n          }\n          setOutput(${returnValue});\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ReduceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo,\n      reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod') {\n    const {windowSize, batchSize, inSize, outSize} = reduceInfo;\n    this.outputShape = [batchSize, outSize];\n\n    let initializationValue = '0.0';\n    let compareOp = ``;\n\n    if (reduceType === 'prod') {\n      initializationValue = '1.0';\n    } else if (reduceType === 'min') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '1.0 / 1e-20';\n      compareOp = `min`;\n    } else if (reduceType === 'max') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n      compareOp = `max`;\n    }\n\n    let returnValue = `${reduceType}(${reduceType}(${reduceType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n\n    if (reduceType === 'sum') {\n      returnValue = `sumValue`;\n    } else if (reduceType === 'prod') {\n      returnValue = `prodValue`;\n    } else if (reduceType === 'all') {\n      returnValue = `allValue`;\n    } else if (reduceType === 'any') {\n      returnValue = `anyValue`;\n    }\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    let updateSnippet = `\n      if (${reduceType === 'sum'}) {\n        sumValue += dot(values, ones);\n      } else if (${reduceType === 'prod'}) {\n        vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);\n        prodValue *= tmp[0] * tmp[1];\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    let vecType = `vec4`;\n\n    if (reduceType === 'all') {\n      initializationValue = '1.0';\n      updateSnippet = `\n        bool reducedAllValue = all(values);\n        float floatedReducedAllValue = float(reducedAllValue);\n        allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    } else if (reduceType === 'any') {\n      initializationValue = '0.0';\n      updateSnippet = `\n        bool reducedAnyValue = any(values);\n        float floatedReducedAnyValue = float(reducedAnyValue);\n        anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    }\n\n    let checkOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float getValue(int batch, int inIdx) {\n        ${checkOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        vec4 minMaxValue = vec4(${initializationValue});\n        float prodValue = 1.0;\n        float sumValue = 0.0;\n        float allValue = 1.0;\n        float anyValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class ReshapePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: [number, number, number], inputShape: [\n    number, number, number\n  ]) {\n    this.outputShape = outputShape;\n\n    let mainLoop = ``;\n    for (let i = 0; i < 4; i++) {\n      let thisRC = `thisRC = rc;`;\n      if (i % 2 === 1) {\n        thisRC += `thisRC.z += 1;`;\n      }\n      if (i > 1) {\n        thisRC += `thisRC.y += 1;`;\n      }\n\n      mainLoop += `\n        ${thisRC}\n        ${i > 0 ? `if(thisRC.y < rows && thisRC.z < cols){` : ''}\n          int flatIndex = getFlatIndex(thisRC);\n\n          ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);\n          vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));\n\n          result[${i}] =\n            getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);\n        ${i > 0 ? '}' : ''}\n      `;\n    }\n\n    this.userCode = `\n      ${getReshapedInputCoords(inputShape)}\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n\n        vec4 result = vec4(0.);\n\n        ivec3 thisRC;\n        int rows = ${outputShape[1]};\n        int cols = ${outputShape[2]};\n\n        ${mainLoop}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nfunction getReshapedInputCoords(shape: [number, number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 inputCoordsFromReshapedOutCoords(int index) {\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(startRLerp - float(winHeight / 2));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(startCLerp - float(winWidth / 2));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float dxR = float(dyR) * heightScale;\n            int topDxRIndex = int(floor(dxR));\n            int bottomDxRIndex = int(min(ceil(dxR), ${xHeight - 1}.0));\n            float dxRLerp = dxR - float(topDxRIndex);\n            float inverseDxRLerp = 1.0 - dxRLerp;\n\n            float dxC = float(dyC) * widthScale;\n            int leftDxCIndex = int(floor(dxC));\n            int rightDxCIndex = int(min(ceil(dxC), ${xWidth - 1}.0));\n            float dxCLerp = dxC - float(leftDxCIndex);\n            float inverseDxCLerp = 1.0 - dxCLerp;\n\n            if (r == topDxRIndex && c == leftDxCIndex) {\n              // topLeft\n              accumulator +=\n                getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;\n            }\n\n            if (r == topDxRIndex && c == rightDxCIndex) {\n              // topRight\n              accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == leftDxCIndex) {\n              // bottomLeft\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == rightDxCIndex) {\n              // bottomRight\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec2 sourceFloorRC = ivec2(sourceFracIndexRC);\n        ivec2 sourceCeilRC = ivec2(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);\n        float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);\n        float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);\n        float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);\n\n        vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);\n\n        float top = topLeft + (topRight - topLeft) * fracRC.y;\n        float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;\n        float newValue = top + (bottom - top) * fracRC.x;\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec3 effectiveInputOverOutputRatioRC = vec3(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec3 inputShapeRC = vec3(${oldHeight}.0, ${oldWidth}.0,\n                                     ${oldWidth}.0);\n\n      float getAValue(int b, int r, int c, int d) {\n        return getChannel(getA(b, r, c, d), vec2(c, d));\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        // Calculate values for next column in yRC.z.\n        ivec3 yRC = coords.yzz + ivec3(0, 0, 1);\n\n        // Fractional source index.\n        vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec3 sourceFloorRC = ivec3(sourceFracIndexRC);\n        ivec3 sourceCeilRC = ivec3(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        // Should we calculate next column and row elements in 2x2 packed cell.\n        bool hasNextCol = d < ${depth - 1};\n        bool hasNextRow = coords.z < ${newWidth - 1};\n\n        // In parallel, construct four corners for all four components in\n        // packed 2x2 cell.\n        vec4 topLeft = vec4(\n          getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 bottomLeft = vec4(\n          getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 topRight = vec4(\n          getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec4 bottomRight = vec4(\n          getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);\n\n        vec4 top = mix(topLeft, topRight, fracRC.yyzz);\n        vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);\n        vec4 newValue = mix(top, bottom, fracRC.x);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeigborBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(floor(startRLerp - float(winHeight / 2)));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(floor(startCLerp - float(winWidth / 2)));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float sourceFracRow =\n              float(${effectiveXSize[0]}) *\n                (float(dyR) / float(${effectiveYSize[0]}));\n\n            float sourceFracCol =\n                float(${effectiveXSize[1]}) *\n                  (float(dyC) / float(${effectiveYSize[1]}));\n\n            int sourceNearestRow = int(min(\n                float(int(${xHeight}) - 1),\n                ${alignCorners} ? float(round(sourceFracRow)) :\n                                  float(floor(sourceFracRow))));\n\n            int sourceNearestCol = int(min(\n                float(int(${xWidth}) - 1),\n                ${alignCorners} ? float(round(sourceFracCol)) :\n                                  float(floor(sourceFracCol))));\n\n            if (r == sourceNearestRow && c == sourceNearestCol) {\n              accumulator += getDy(b, dyR, dyC, d);\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeighborProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    // When align corners is false, we rounds the value with floor.\n    const roundBase = alignCorners ? '0.5' : '0.0';\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the coordinators of nearest neighbor point.\n        ivec2 sourceNearestRC = ivec2(\n          min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${roundBase})));\n\n        float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReverseProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n\n    if (rank === 1) {\n      this.userCode = `\n        void main() {\n          int coord = getOutputCoords();\n          setOutput(getX(${xShape[0]} - coord - 1));\n        }\n      `;\n      return;\n    }\n    const getInCoord = (i: number) => {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - coords[${i}] - 1`;\n      }\n      return `coords[${i}]`;\n    };\n    const inCoords = xShape.map((_, i) => getInCoord(i)).join(',');\n    const type = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${type} coords = getOutputCoords();\n        setOutput(getX(${inCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReversePackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n    const channels = getChannels('rc', rank);\n    const nextColumn =\n        `${channels[rank - 1]} + 1 < ${this.outputShape[rank - 1]}`;\n    const nextRow = `${channels[rank - 2]} + 1 < ${this.outputShape[rank - 2]}`;\n    const type = getCoordsDataType(rank);\n    if (rank === 1) {\n      this.userCode = `\n        void main(){\n          int rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = getChannel(getX(${xShape[0]} - rc - 1),\n            ${xShape[0]} - rc - 1);\n          if(${nextColumn}){\n              result.g = getChannel(getX(${xShape[0]} - (rc  + 1) - 1),\n                ${xShape[0]} - (rc  + 1) - 1);\n          }\n          setOutput(result);\n        }\n      `;\n    } else {\n      this.userCode = `\n        void main() {\n          ${type} rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = ${getR(channels.slice())};\n          if(${nextColumn}){\n            result.g = ${getG(channels.slice())};\n          }\n          if(${nextRow}) {\n            result.b = ${getB(channels.slice())};\n            if(${nextColumn}) {\n              result.a = ${getA(channels.slice())};\n            }\n          }\n          setOutput(result);\n        }\n    `;\n    }\n\n    function getR(channels: string[]): string {\n      return getChannel(channels);\n    }\n\n    function getG(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getB(channels: string[]): string {\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getA(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getChannel(channels: string[]): string {\n      const inCoordsArray = xShape.map((_, i) => getInCoord(i, channels));\n      const inCoords = inCoordsArray.join(',');\n      const innerDims = inCoordsArray.slice(-2).join(',');\n      return `getChannel(getX(${inCoords}), vec2(${innerDims}))`;\n    }\n\n    function getInCoord(i: number, channels1: string[]): string {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - ${channels1[i]} - 1`;\n      } else {\n        return `${channels1[i]}`;\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ScatterProgram implements GPGPUProgram {\n  variableNames = ['updates', 'indices', 'defaultValue'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      updateSize: number, sliceDim: number, indicesRank: number,\n      updatesRank: number, strides: number[], shape: number[],\n      summingDupeIndex = true) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    let indicesString = '';\n    if (indicesRank === 1) {\n      indicesString = 'i';\n    } else if (indicesRank === 2) {\n      indicesString = 'i, j';\n    }\n    const indicesSnippet = `getIndices(${indicesString})`;\n\n    let updatesString = '';\n    if (updatesRank === 1) {\n      updatesString = 'i';\n    } else if (updatesRank === 2) {\n      updatesString = 'i, coords[1]';\n    }\n    const updatesSnippet = `getUpdates(${updatesString})`;\n\n    const strideString = sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${strides});\n\n        void main() {\n          ${dtype} coords = getOutputCoords();\n          float sum = 0.0;\n          bool found = false;\n          for (int i = 0; i < ${updateSize}; i++) {\n            int flattenedIndex = 0;\n            for (int j = 0; j < ${sliceDim}; j++) {\n              int index = round(${indicesSnippet});\n              flattenedIndex += index * ${strideString};\n            }\n            if (flattenedIndex == coords[0]) {\n              sum += ${updatesSnippet};\n              found = true;\n            }\n          }\n          setOutput(mix(getDefaultValue(), sum, float(found)));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class SegmentOpProgram implements GPGPUProgram {\n  variableNames = ['x', 'segmentIds'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      segOpInfo: backend_util.segment_util.SegOpInfo,\n      segOpType: 'unsortedSegmentSum') {\n    const windowSize = segOpInfo.windowSize;\n    const batchSize = segOpInfo.batchSize;\n    const inSize = segOpInfo.inSize;\n    const numSegments = segOpInfo.numSegments;\n    const outSize = numSegments * Math.ceil(inSize / windowSize);\n    this.outputShape = [batchSize, outSize];\n\n    const initializationValue = '0.0';\n    const returnValue = `sumValue`;\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    const updateSnippet = `\n        sumValue += dot(values, segFilter);\n    `;\n\n    let checkValueOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkValueOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n\n    let checkSegmentIdOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkSegmentIdOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return -1.0;\n        }\n      `;\n    }\n\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n\n      float getValue(int batch, int inIdx) {\n        ${checkValueOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      float getSegmentIdAtIndex(int inIdx) {\n        ${checkSegmentIdOutOfBounds}\n        return getSegmentIds(inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = int(floor(float(outIdx) / float(\n          ${numSegments})) * float(${windowSize}));\n        int currentSeg = int(mod(float(outIdx), float(${numSegments})));\n\n        float sumValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          int inIdxSeg = int(getSegmentIdAtIndex(inIdx));\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            0,\n            0,\n            0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n              0,\n              0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            0\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SelectProgram implements GPGPUProgram {\n  variableNames = ['c', 'a', 'b'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(cRank: number, shape: number[], rank: number) {\n    this.outputShape = shape;\n\n    let cCoords;\n    let abCoords;\n    if (rank > 4) {\n      throw Error(`Where for rank ${rank} is not yet supported`);\n    }\n\n    if (rank === 1) {\n      abCoords = `resRC`;\n      cCoords = `resRC`;\n    } else {\n      const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n      const cCoordVars = [];\n      const abCoordVars = [];\n      for (let i = 0; i < shape.length; i++) {\n        abCoordVars.push(`${currentCoords[i]}`);\n        if (i < cRank) {\n          cCoordVars.push(`${currentCoords[i]}`);\n        }\n      }\n      cCoords = cCoordVars.join();\n      abCoords = abCoordVars.join();\n    }\n\n    const dtype = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        float cVal = getC(${cCoords});\n        if (cVal >= 1.0) {\n          setOutput(getA(${abCoords}));\n        } else {\n          setOutput(getB(${abCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SliceProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const uniformPart = `uniform int start[${this.rank}];`;\n    const sourceCoords = getCoords(this.rank);\n\n    let body: string;\n    const coordSum = destSize.map((_, i) => {\n      return `sourceLoc.${coords[i]} = start[${i}] + coords.${coords[i]};`;\n    });\n    body = `\n        ${dtype} sourceLoc;\n        ${dtype} coords = getOutputCoords();\n        ${coordSum.join('\\n')}\n      `;\n    this.userCode = `\n      ${uniformPart}\n      void main() {\n        ${body}\n        setOutput(getSource(${sourceCoords}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n\nconst coords = ['x', 'y', 'z', 'w', 'u', 'v'];\n\nfunction getCoords(rank: number): string {\n  if (rank === 1) {\n    return 'sourceLoc';\n  } else if (rank <= 6) {\n    return coords.slice(0, rank).map(x => 'sourceLoc.' + x).join(',');\n  } else {\n    throw Error(`Slicing for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SlicePackedProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const coords = getChannels('coords', this.rank);\n    const sourceLoc = getChannels('sourceLoc', this.rank);\n\n    const innerDims =\n        this.rank === 1 ? 'sourceLoc' : `vec2(${sourceLoc.slice(-2).join()})`;\n    const getChannel =\n        `getChannel(getSource(${sourceLoc.join()}), ${innerDims})`;\n    const upperRow = `\n      result.x = ${getChannel};\n      if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n        ++${sourceLoc[this.rank - 1]};\n        result.y = ${getChannel};\n        --${sourceLoc[this.rank - 1]};\n      }\n    `;\n    const lowerRow = this.rank === 1 ? '' : `\n      --${coords[this.rank - 1]};\n      if (++${coords[this.rank - 2]} < ${destSize[this.rank - 2]}) {\n        ++${sourceLoc[this.rank - 2]};\n        result.z = ${getChannel};\n        if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n          ++${sourceLoc[this.rank - 1]};\n          result.w = ${getChannel};\n        }\n      }\n    `;\n\n    const sourceLocSetup = this.rank <= 4 ?\n        `sourceLoc = coords +\n            ${dtype}(${destSize.map((_, i) => `start[${i}]`).join()});` :\n        destSize.map((_, i) => `${sourceLoc[i]} = ${coords[i]} + start[${i}];`)\n            .join('\\n');\n    this.userCode = `\n      uniform int start[${this.rank}];\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        ${dtype} sourceLoc;\n        ${sourceLocSetup}\n        vec4 result = vec4(0.);\n        ${upperRow}\n        ${lowerRow}\n        setOutput(result);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class StridedSliceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(begin: number[], strides: number[], size: number[]) {\n    this.outputShape = size;\n    const rank = size.length;\n    const inputDtype = getCoordsDataType(size.length);\n    const dtype = getCoordsDataType(size.length);\n\n    let newCoords = '';\n    if (rank === 1) {\n      newCoords = 'coords * strides + begin';\n    } else {\n      let outputAxis = 0;\n      newCoords =\n          size.map((_, i) => {\n                outputAxis++;\n                return size.length === 1 ?\n                    `coords * strides[${i}] + begin[${i}]` :\n                    `coords[${outputAxis - 1}] * strides[${i}] + begin[${i}]`;\n              })\n              .join(',');\n    }\n\n    this.userCode = `\n      ${inputDtype} begin = ${inputDtype}(${begin});\n      ${inputDtype} strides = ${inputDtype}(${strides});\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        setOutput(getX(${newCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {getInternalFormatForFloat16MatrixTexture, getInternalFormatForFloat16PackedMatrixTexture, getInternalFormatForFloat32MatrixTexture, getInternalFormatForPackedMatrixTexture, getInternalFormatForUnsignedBytesMatrixTexture} from './gpgpu_util';\nimport {getPackedMatrixTextureShapeWidthHeight, getUnpackedMatrixTextureShapeWidthHeight, PhysicalTextureType, TextureConfig, TextureUsage} from './tex_util';\n\nexport class TextureManager {\n  private numUsedTextures = 0;\n  private numFreeTextures = 0;\n  private _numBytesAllocated = 0;\n  private _numBytesFree = 0;  // How many bytes that have been allocated\n                              // are available for reuse.\n  private freeTextures: {[shape: string]: WebGLTexture[]} = {};\n  private logEnabled = false;\n  private usedTextures: {[shape: string]: WebGLTexture[]} = {};\n\n  constructor(private gpgpu: GPGPUContext) {}\n\n  acquireTexture(\n      shapeRC: [number, number], usage: TextureUsage,\n      isPacked: boolean): WebGLTexture {\n    const physicalTexType = getPhysicalFromLogicalTextureType(usage, isPacked);\n\n    const shapeKey = getKeyFromTextureShape(shapeRC, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n    if (!(shapeKey in this.usedTextures)) {\n      this.usedTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shapeRC, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n\n    if (this.freeTextures[shapeKey].length > 0) {\n      this.numFreeTextures--;\n      this.numUsedTextures++;\n      this._numBytesFree -= texBytes;\n      this.log();\n      const newTexture = this.freeTextures[shapeKey].shift();\n      this.usedTextures[shapeKey].push(newTexture);\n      return newTexture;\n    }\n\n    let newTexture: WebGLTexture;\n    if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT32) {\n      newTexture = this.gpgpu.createPackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16PackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT32) {\n      newTexture =\n          this.gpgpu.createFloat32MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (\n        physicalTexType === PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE) {\n      newTexture =\n          this.gpgpu.createUnsignedBytesMatrixTexture(shapeRC[0], shapeRC[1]);\n    }\n    this.usedTextures[shapeKey].push(newTexture);\n\n    this.numUsedTextures++;\n    this._numBytesAllocated += texBytes;\n    this.log();\n\n    return newTexture;\n  }\n\n  releaseTexture(\n      texture: WebGLTexture, shape: [number, number],\n      logicalTexType: TextureUsage, isPacked: boolean): void {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    const physicalTexType =\n        getPhysicalFromLogicalTextureType(logicalTexType, isPacked);\n    const shapeKey = getKeyFromTextureShape(shape, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shape, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n    const deleteTexThreshold = env().get('WEBGL_DELETE_TEXTURE_THRESHOLD');\n    if (deleteTexThreshold !== -1 &&\n        this._numBytesAllocated > deleteTexThreshold) {\n      this.gpgpu.deleteMatrixTexture(texture);\n      this._numBytesAllocated -= texBytes;\n    } else {\n      this.freeTextures[shapeKey].push(texture);\n      this.numFreeTextures++;\n      this._numBytesFree += texBytes;\n    }\n\n    this.numUsedTextures--;\n\n    const texList = this.usedTextures[shapeKey];\n    const texIndex = texList.indexOf(texture);\n    if (texIndex < 0) {\n      throw new Error(\n          'Cannot release a texture that was never provided by this ' +\n          'texture manager');\n    }\n    texList.splice(texIndex, 1);\n    this.log();\n  }\n\n  private log() {\n    if (!this.logEnabled) {\n      return;\n    }\n    const total = this.numFreeTextures + this.numUsedTextures;\n    console.log(\n        'Free/Used', `${this.numFreeTextures} / ${this.numUsedTextures}`,\n        `(${total})`);\n    const freeRatio = this._numBytesFree / this._numBytesAllocated;\n    console.log(`Bytes allocated: ${this._numBytesAllocated}`);\n    console.log(`Bytes unused: ${this._numBytesFree} (${\n        Math.round(100 * freeRatio)}%)`);\n  }\n\n  get numBytesAllocated(): number {\n    return this._numBytesAllocated;\n  }\n\n  get numBytesFree(): number {\n    return this._numBytesFree;\n  }\n\n  getNumUsedTextures(): number {\n    return this.numUsedTextures;\n  }\n\n  getNumFreeTextures(): number {\n    return this.numFreeTextures;\n  }\n\n  dispose() {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    for (const texShape in this.freeTextures) {\n      this.freeTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    for (const texShape in this.usedTextures) {\n      this.usedTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    this.freeTextures = null;\n    this.usedTextures = null;\n    this.numUsedTextures = 0;\n    this.numFreeTextures = 0;\n    this._numBytesAllocated = 0;\n    this._numBytesFree = 0;\n  }\n}\n\nfunction numBytesForInternalFormat(\n    gl: WebGLRenderingContext, internalFormat: number): number {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n  if (internalFormat === glany.R32F) {\n    return 4;\n  } else if (internalFormat === glany.R16F) {\n    return 2;\n  } else if (internalFormat === glany.RGBA32F) {\n    return 16;\n  } else if (internalFormat === gl.RGBA) {\n    return 16;\n  } else if (internalFormat === glany.RGBA16F) {\n    return 8;\n  }\n  throw new Error(`Unknown internal format ${internalFormat}`);\n}\n\nexport function computeBytes(\n    shape: [number, number], physicalTexType: PhysicalTextureType,\n    gl: WebGLRenderingContext, textureConfig: TextureConfig,\n    isPacked: boolean): number {\n  // It is not possible to infer packed status from the texture type because\n  // depending on the textureConfig, different  texture types may resolve to the\n  // same internal format (e.g. in WebGL1, the internal format for\n  // UNPACKED_FLOAT16 textures is gl.RGBA). Therefore we pass in `isPacked`\n  // explicitly.\n  const internalFormat =\n      internalFormatForPhysicalTexType(physicalTexType, textureConfig);\n\n  let numElements: number;\n  if (isPacked) {\n    const [packedWidth, packedHeight] =\n        getPackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = packedWidth * packedHeight;\n\n  } else {\n    const [width, height] =\n        getUnpackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = width * height;\n  }\n\n  const bytesPerElement = numBytesForInternalFormat(gl, internalFormat);\n  return numElements * bytesPerElement;\n}\n\nfunction internalFormatForPhysicalTexType(\n    physicalTexType: PhysicalTextureType,\n    textureConfig: TextureConfig): number {\n  switch (physicalTexType) {\n    case PhysicalTextureType.PACKED_2X2_FLOAT32:\n      return getInternalFormatForPackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_2X2_FLOAT16:\n      return getInternalFormatForFloat16PackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT32:\n      return getInternalFormatForFloat32MatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT16:\n      return getInternalFormatForFloat16MatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE:\n      return getInternalFormatForUnsignedBytesMatrixTexture(textureConfig);\n    default:\n      throw new Error(`Unknown physical texture type ${physicalTexType}`);\n  }\n}\n\nfunction getPhysicalTextureForRendering(isPacked: boolean):\n    PhysicalTextureType {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED')) {\n    if (isPacked) {\n      return PhysicalTextureType.PACKED_2X2_FLOAT32;\n    }\n    return PhysicalTextureType.UNPACKED_FLOAT32;\n  }\n\n  if (isPacked) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT16;\n  }\n  return PhysicalTextureType.UNPACKED_FLOAT16;\n}\n\nfunction getPhysicalFromLogicalTextureType(\n    logicalTexType: TextureUsage, isPacked: boolean): PhysicalTextureType {\n  if (logicalTexType === TextureUsage.UPLOAD) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT32;\n  } else if (logicalTexType === TextureUsage.RENDER || logicalTexType == null) {\n    return getPhysicalTextureForRendering(isPacked);\n  } else if (\n      logicalTexType === TextureUsage.DOWNLOAD ||\n      logicalTexType === TextureUsage.PIXELS) {\n    return PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE;\n  }\n  throw new Error(`Unknown logical texture type ${logicalTexType}`);\n}\n\nfunction getKeyFromTextureShape(\n    shapeRowsCol: [number, number], physicalTexType: PhysicalTextureType,\n    isPacked: boolean): string {\n  return `${shapeRowsCol[0]}_${shapeRowsCol[1]}_${physicalTexType}_${isPacked}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TileProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], reps: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[i] * reps[i];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[]): string {\n  const rank = aShape.length;\n  if (rank > 5) {\n    throw Error(`Tile for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `imod(resRC, ${aShape[0]})`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    sourceCoords.push(`imod(${currentCoords[i]}, ${aShape[i]})`);\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class UnaryOpProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      float unaryOperation(float x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        float x = getAAtOutCoords();\n        float y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n\nconst CHECK_NAN_SNIPPET = `if (isnan(x)) return x;`;\n\nexport const LINEAR = `return x;`;\n\nexport const ABS = `return abs(x);`;\n\nexport const RELU = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : x;\n`;\n\nexport const RELU6 = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : min(6.0, x);\n`;\n\nexport const ELU = `return (x >= 0.0) ? x : (exp(x) - 1.0);`;\n\nexport const SELU = `\n  // Stable and Attracting Fixed Point (0, 1) for Normalized Weights.\n  // see: https://arxiv.org/abs/1706.02515\n  float scaleAlpha = ${backend_util.SELU_SCALEALPHA};\n  float scale = ${backend_util.SELU_SCALE};\n  return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);\n`;\n\nexport function STEP(alpha = 0.0) {\n  return CHECK_NAN_SNIPPET + `\n    return x > 0.0 ? 1.0 : float(${alpha});\n  `;\n}\n\nexport const NEG = `return -x;`;\n\nexport const CEIL = `return ceil(x);`;\n\nexport const FLOOR = `return floor(x);`;\n\nexport const SIGN = `\n  if (isnan(x)) { return 0.0; }\n  return sign(x);\n`;\n\nexport const IS_NAN = `return float(isnan(x));`;\n\nexport const IS_INF = `return float(isinf(x));`;\n\nexport const IS_FINITE = `return float(!isnan(x) && !isinf(x));`;\n\nexport const ROUND = `\n  // OpenGL ES does not support round function.\n  // The algorithm is based on banker's rounding.\n  float base = floor(x);\n  if ((x - base) < 0.5) {\n    return floor(x);\n  } else if ((x - base) > 0.5) {\n    return ceil(x);\n  } else {\n    if (mod(base, 2.0) == 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n`;\n\nexport const EXP = `return exp(x);`;\n\nexport const EXPM1 = `return exp(x) - 1.0;`;\n\nexport const LOG = `if (x < 0.0) return NAN;\n  return log(x);`;\n\nexport const LOG1P = `return log(1.0 + x);`;\n\nexport const SQRT = `return sqrt(x);`;\n\nexport const RSQRT = `return inversesqrt(x);`;\n\nexport const SIGMOID = `return 1.0 / (1.0 + exp(-1.0 * x));`;\n\n/**\n * mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n *\n * epsilon is the difference between 1.0 and the next representable\n * float. For a single precision 32 bit float this should be 2^-23, see:\n * https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\n *\n * too_large = (x > -threshold) is value above which exp(x) may overflow\n * but softplus(x) == x is within machine epsilon\n *\n * too_small = (x < threshold) is value below which exp(x) may underflow,\n * but softplus(x) == exp(x) is within machine epsilon.\n */\nexport const SOFTPLUS = `\n  float epsilon = 1.1920928955078125e-7;\n  float threshold = log(epsilon) + 2.0;\n\n  bool too_large = x > -threshold;\n  bool too_small = x < threshold;\n\n  float result;\n  float exp_x = exp(x);\n\n  if (too_large){\n    result = x;\n  }\n  else if (too_small){\n    result = exp_x;\n  }\n  else{\n    result = log(exp_x + 1.0);\n  }\n  return result;\n`;\n\nexport const ASIN = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return asin(x);\n`;\n\nexport const ACOS = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return acos(x);\n`;\n\nexport const ATAN = CHECK_NAN_SNIPPET + `\n  return atan(x);\n`;\n\nexport const SINH = `\n  float e2x = exp(x);\n  return (e2x - 1.0 / e2x) / 2.0;\n`;\n\nexport const COSH = `\n  float e2x = exp(-x);\n  return (e2x + 1.0 / e2x) / 2.0;\n`;\n\nexport const TANH = `\n  float e2x = exp(-2.0 * abs(x));\n  return sign(x) * (1.0 - e2x) / (1.0 + e2x);\n`;\n\nexport const ASINH = CHECK_NAN_SNIPPET + `return log(x + sqrt(x * x + 1.0));`;\n\nexport const ACOSH = CHECK_NAN_SNIPPET + `\n  if (x < 1.0) return NAN;\n  return log(x + sqrt(x * x - 1.0));`;\n\nexport const ATANH = CHECK_NAN_SNIPPET + `\n  if ((x < -1.0) || (x > 1.0)) return NAN;\n  return (log(1.0 + x) - log(1.0 - x)) / 2.0;`;\n\nexport const ERF = `\n  // Error function is calculated approximately with elementary function.\n  // See \"Handbook of Mathematical Functions with Formulas,\n  // Graphs, and Mathematical Tables\", Abramowitz and Stegun.\n  float p = ${backend_util.ERF_P};\n  float a1 = ${backend_util.ERF_A1};\n  float a2 = ${backend_util.ERF_A2};\n  float a3 = ${backend_util.ERF_A3};\n  float a4 = ${backend_util.ERF_A4};\n  float a5 = ${backend_util.ERF_A5};\n\n  float sign = sign(x);\n  x = abs(x);\n  float t = 1.0 / (1.0 + p * x);\n  return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));\n`;\n\nexport const RECIPROCAL = `return 1.0 / x;`;\n\nexport const LOGICAL_NOT = `return float(!(x >= 1.0));`;\n\nexport const TO_INT = `return float(int(x));`;\n\nexport const CLONE = 'return x;';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const LINEAR = `return x;`;\n\nexport const LOG = `\n  vec4 result = log(x);\n  vec4 isNaN = vec4(lessThan(x, vec4(0.0)));\n  result.r = isNaN.r == 1.0 ? NAN : result.r;\n  result.g = isNaN.g == 1.0 ? NAN : result.g;\n  result.b = isNaN.b == 1.0 ? NAN : result.b;\n  result.a = isNaN.a == 1.0 ? NAN : result.a;\n\n  return result;\n`;\n\nexport const RELU = `\n  vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const RELU6 = `\n  vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const ELU = `\n  vec4 result;\n\n  result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);\n  result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);\n  result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);\n  result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);\n\n  return result;\n`;\n\nexport class UnaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      vec4 unaryOperation(vec4 x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        vec4 x = getAAtOutCoords();\n        vec4 y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels, getSourceCoords} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class UnpackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = false;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: number[]) {\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    const channels = getChannels('rc', rank);\n    const dtype = getCoordsDataType(rank);\n    const sourceCoords = getSourceCoords(rank, channels);\n    const innerDims = channels.slice(-2);\n    const coords = rank <= 1 ? 'rc' : `vec2(${innerDims.join(',')})`;\n\n    this.userCode = `\n      void main() {\n        ${dtype} rc = getOutputCoords();\n        vec4 packedInput = getA(${sourceCoords});\n\n        setOutput(getChannel(packedInput, ${coords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import webgl flags.\nimport './flags_webgl';\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {complex, DataId, div, engine, env, imag, max, MemoryInfo, range, real, RecursiveArray, reshape, scalar, softmax, tensor, tidy, TimingInfo, transpose} from '@tensorflow/tfjs-core';\nimport {backend_util, buffer, kernel_impls, slice_util, util} from '@tensorflow/tfjs-core';\nimport {DataStorage, DataType, KernelBackend, NumericDataType, Rank, Scalar, ShapeMap, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorInfo, TypedArray, upcastType} from '@tensorflow/tfjs-core';\n\nimport {addImplCPU, ceilImplCPU, expImplCPU, expm1ImplCPU, floorImplCPU, logImplCPU, multiplyImplCPU, rsqrtImplCPU, simpleAbsImplCPU, sliceImplCPU, subImplCPU} from './kernel_utils/shared';\n\nconst {segment_util} = backend_util;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\n\nimport {AddNProgram} from './addn_gpu';\nimport {AddNPackedProgram} from './addn_packed_gpu';\nimport {ArgMinMaxProgram} from './argminmax_gpu';\nimport {ArgMinMaxPackedProgram} from './argminmax_packed_gpu';\nimport {AvgPool3DBackpropProgram} from './avg_pool_backprop_gpu';\nimport * as binaryop_complex_gpu from './binaryop_complex_gpu';\nimport {BinaryOpComplexProgram} from './binaryop_complex_gpu';\nimport * as binaryop_gpu from './binaryop_gpu';\nimport {BinaryOpProgram} from './binaryop_gpu';\nimport * as binaryop_packed_gpu from './binaryop_packed_gpu';\nimport {BinaryOpPackedProgram} from './binaryop_packed_gpu';\nimport {getWebGLContext} from './canvas_util';\nimport {ClipProgram} from './clip_gpu';\nimport {ClipPackedProgram} from './clip_packed_gpu';\nimport {ComplexAbsProgram} from './complex_abs_gpu';\nimport {ConcatProgram} from './concat_gpu';\nimport {ConcatPackedProgram} from './concat_packed_gpu';\nimport {Conv2DDerFilterProgram, Conv2DDerInputProgram, Conv3DDerFilterProgram, Conv3DDerInputProgram} from './conv_backprop_gpu';\nimport {DepthwiseConv2DDerFilterProgram, DepthwiseConv2DDerInputProgram} from './conv_backprop_gpu_depthwise';\nimport {Conv2DProgram, Conv3DProgram} from './conv_gpu';\nimport {DepthwiseConv2DProgram} from './conv_gpu_depthwise';\nimport {DepthwiseConvPacked2DProgram} from './conv_packed_gpu_depthwise';\nimport {CropAndResizeProgram} from './crop_and_resize_gpu';\nimport {CumSumProgram} from './cumsum_gpu';\nimport {DecodeMatrixProgram} from './decode_matrix_gpu';\nimport {DecodeMatrixPackedProgram} from './decode_matrix_packed_gpu';\nimport {DepthToSpaceProgram} from './depth_to_space_gpu';\nimport {DiagProgram} from './diag_gpu';\nimport {EncodeFloatProgram} from './encode_float_gpu';\nimport {EncodeFloatPackedProgram} from './encode_float_packed_gpu';\nimport {EncodeMatrixProgram} from './encode_matrix_gpu';\nimport {EncodeMatrixPackedProgram} from './encode_matrix_packed_gpu';\nimport * as fft_gpu from './fft_gpu';\nimport {FFTProgram} from './fft_gpu';\nimport {FillProgram} from './fill_gpu';\nimport {GatherProgram} from './gather_gpu';\nimport {GatherNDProgram} from './gather_nd_gpu';\nimport {GPGPUContext} from './gpgpu_context';\nimport * as gpgpu_math from './gpgpu_math';\nimport {GPGPUBinary, GPGPUProgram, TensorData} from './gpgpu_math';\nimport {Im2ColPackedProgram} from './im2col_packed_gpu';\nimport {LRNProgram} from './lrn_gpu';\nimport {LRNGradProgram} from './lrn_grad_gpu';\nimport {LRNPackedProgram} from './lrn_packed_gpu';\nimport {MaxPool3DBackpropProgram} from './max_pool_backprop_gpu';\nimport {MatMulPackedProgram} from './mulmat_packed_gpu';\nimport {MultinomialProgram} from './multinomial_gpu';\nimport {OneHotProgram} from './onehot_gpu';\nimport {PackProgram} from './pack_gpu';\nimport {PadProgram} from './pad_gpu';\nimport {PadPackedProgram} from './pad_packed_gpu';\nimport {Pool3DProgram} from './pool_gpu';\nimport {ReduceProgram} from './reduce_gpu';\nimport {ReshapePackedProgram} from './reshape_packed_gpu';\nimport {ResizeBilinearBackpropProgram} from './resize_bilinear_backprop_gpu';\nimport {ResizeBilinearProgram} from './resize_bilinear_gpu';\nimport {ResizeBilinearPackedProgram} from './resize_bilinear_packed_gpu';\nimport {ResizeNearestNeigborBackpropProgram} from './resize_nearest_neighbor_backprop_gpu';\nimport {ResizeNearestNeighborProgram} from './resize_nearest_neighbor_gpu';\nimport {ReverseProgram} from './reverse_gpu';\nimport {ReversePackedProgram} from './reverse_packed_gpu';\nimport {ScatterProgram} from './scatter_gpu';\nimport {SegmentOpProgram} from './segment_gpu';\nimport {SelectProgram} from './select_gpu';\nimport {SliceProgram} from './slice_gpu';\nimport {SlicePackedProgram} from './slice_packed_gpu';\nimport {StridedSliceProgram} from './strided_slice_gpu';\nimport * as tex_util from './tex_util';\nimport {TextureData, TextureUsage} from './tex_util';\nimport {TextureManager} from './texture_manager';\nimport {TileProgram} from './tile_gpu';\nimport * as unary_op from './unaryop_gpu';\nimport {UnaryOpProgram} from './unaryop_gpu';\nimport * as unary_packed_op from './unaryop_packed_gpu';\nimport {UnaryOpPackedProgram} from './unaryop_packed_gpu';\nimport {UnpackProgram} from './unpack_gpu';\nimport * as webgl_util from './webgl_util';\nimport {BackendValues} from '@tensorflow/tfjs-core';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\ntype KernelInfo = {\n  name: string; query: Promise<number>;\n};\n\nexport type TimerNode = RecursiveArray<KernelInfo>|KernelInfo;\nexport interface CPUTimerQuery {\n  startMs: number;\n  endMs?: number;\n}\n\nexport interface WebGLMemoryInfo extends MemoryInfo {\n  numBytesInGPU: number;\n  // Tracks the total number of bytes allocated on the GPU, accounting for the\n  // physical texture type.\n  numBytesInGPUAllocated: number;\n  // Tracks byte size of textures that were created and then made available for\n  // reuse (disposed).\n  numBytesInGPUFree: number;\n  unreliable: boolean;\n}\n\nexport interface WebGLTimingInfo extends TimingInfo {\n  uploadWaitMs: number;\n  downloadWaitMs: number;\n}\n\nconst binaryCaches: {[webGLVersion: string]: {[key: string]: GPGPUBinary}} = {};\n\nexport function getBinaryCache(webGLVersion: number) {\n  if (webGLVersion in binaryCaches) {\n    return binaryCaches[webGLVersion];\n  }\n  binaryCaches[webGLVersion] = {};\n  return binaryCaches[webGLVersion];\n}\n\nfunction mapActivationToShaderProgram(\n    activation: backend_util.Activation, packed = false): string {\n  if (activation === 'linear') {\n    if (packed) {\n      return unary_packed_op.LINEAR;\n    }\n    return unary_op.LINEAR;\n  } else if (activation === 'relu') {\n    if (packed) {\n      return unary_packed_op.RELU;\n    }\n    return unary_op.RELU;\n  } else if (activation === 'elu') {\n    if (packed) {\n      return unary_packed_op.ELU;\n    }\n    return unary_op.ELU;\n  } else if (activation === 'relu6') {\n    if (packed) {\n      return unary_packed_op.RELU6;\n    }\n    return unary_op.RELU6;\n  } else if (activation === 'prelu') {\n    if (packed) {\n      return binaryop_packed_gpu.PRELU;\n    }\n    return binaryop_gpu.PRELU;\n  }\n  throw new Error(`Activation ${\n      activation} has not been implemented for the WebGL backend.`);\n}\n\n// Empirically determined constant used to determine size threshold for handing\n// off execution to the CPU.\nconst CPU_HANDOFF_SIZE_THRESHOLD = 128;\n\n// Empirically determined constant used to decide the number of MB on GPU\n// before we warn about high memory use. The MB are this constant * screen area\n// * dpi / 1024 / 1024.\nconst BEFORE_PAGING_CONSTANT = 600;\nfunction numMBBeforeWarning(): number {\n  if (env().global.screen == null) {\n    return 1024;  // 1 GB.\n  }\n  return (env().global.screen.height * env().global.screen.width *\n          window.devicePixelRatio) *\n      BEFORE_PAGING_CONSTANT / 1024 / 1024;\n}\n\n// Empirically determined minimal shared dimension in matmul before we forward\n// to a.mul(b).sum() in order to take advantage of GPU parallelism. See\n// https://github.com/tensorflow/tfjs-core/pull/1379 for benchmarks.\nexport const MATMUL_SHARED_DIM_THRESHOLD = 1000;\n\nexport class MathBackendWebGL extends KernelBackend {\n  texData: DataStorage<TextureData>;\n  gpgpu: GPGPUContext;\n\n  // Maps data ids that have a pending read operation, to list of subscribers.\n  private pendingRead = new WeakMap<DataId, Array<(arr: TypedArray) => void>>();\n  // List of data ids that are scheduled for disposal, but are waiting on a\n  // pending read operation.\n  private pendingDisposal = new WeakSet<DataId>();\n  // Used to count the number of 'shallow' sliced tensors that point to the\n  // same data id.\n  private dataRefCount = new WeakMap<DataId, number>();\n  private numBytesInGPU = 0;\n\n  private canvas: HTMLCanvasElement|OffscreenCanvas;\n\n  private programTimersStack: TimerNode[];\n  private activeTimers: TimerNode[];\n  // Accumulated time spent (including blocking) in uploading data to webgl.\n  private uploadWaitMs = 0;\n  // Accumulated time spent (including blocking in downloading data from webgl.\n  private downloadWaitMs = 0;\n  private cpuBackend: KernelBackend;\n\n  // Number of bits of precision of this backend.\n  private floatPrecisionValue: 32|16;\n\n  private textureManager: TextureManager;\n  private binaryCache: {[key: string]: GPGPUBinary};\n  private gpgpuCreatedLocally: boolean;\n  private numMBBeforeWarning: number;\n  private warnedAboutMemory = false;\n  private warnedAboutCPUBackend = false;\n\n  constructor(gpgpu?: GPGPUContext) {\n    super();\n    if (!env().getBool('HAS_WEBGL')) {\n      throw new Error('WebGL is not supported on this device');\n    }\n\n    if (gpgpu == null) {\n      const gl = getWebGLContext(env().getNumber('WEBGL_VERSION'));\n      this.binaryCache = getBinaryCache(env().getNumber('WEBGL_VERSION'));\n      this.gpgpu = new GPGPUContext(gl);\n      this.canvas = gl.canvas;\n      this.gpgpuCreatedLocally = true;\n    } else {\n      this.gpgpu = gpgpu;\n      this.binaryCache = {};\n      this.gpgpuCreatedLocally = false;\n      this.canvas = gpgpu.gl.canvas;\n    }\n    this.textureManager = new TextureManager(this.gpgpu);\n    this.numMBBeforeWarning = numMBBeforeWarning();\n\n    this.texData = new DataStorage(this, engine());\n  }\n\n  numDataIds() {\n    return this.texData.numDataIds() +\n        (this.cpuBackend ? this.cpuBackend.numDataIds() : 0) -\n        this.pendingDeletes;\n  }\n\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    if (env().getBool('WEBGL_CHECK_NUMERICAL_PROBLEMS') ||\n        env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64' && values != null) {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    const dataId = {};\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n    return dataId;\n  }\n\n  /** Increase refCount of a `TextureData`. */\n  incRef(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    texData.refCount++;\n  }\n\n  /** Decrease refCount of a `TextureData`. */\n  decRef(dataId: DataId): void {\n    if (this.texData.has(dataId)) {\n      const texData = this.texData.get(dataId);\n      texData.refCount--;\n    }\n  }\n\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    if (env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.texData.has(dataId)) {\n      const textureData = this.texData.get(dataId);\n\n      textureData.refCount--;\n\n      if (textureData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    const texData = this.texData.get(dataId);\n    const {values, dtype, complexTensors, slice, shape, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.readSync(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n    if (dtype === 'string') {\n      return values;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let result: Float32Array;\n    if (dtype === 'complex64') {\n      const realValues = complexTensors.real.dataSync() as Float32Array;\n      const imagValues = complexTensors.imag.dataSync() as Float32Array;\n      result = backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    } else {\n      result = this.getValuesFromTexture(dataId);\n    }\n\n    if (shouldTimeProgram) {\n      this.downloadWaitMs += util.now() - start;\n    }\n    return this.convertAndCacheOnCPU(dataId, result);\n  }\n\n  async read(dataId: DataId): Promise<BackendValues> {\n    if (this.pendingRead.has(dataId)) {\n      const subscribers = this.pendingRead.get(dataId);\n      return new Promise<TypedArray>(resolve => subscribers.push(resolve));\n    }\n    const texData = this.texData.get(dataId);\n    const {values, shape, slice, dtype, complexTensors, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.read(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n\n    if (!env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED') &&\n        env().getNumber('WEBGL_VERSION') === 2) {\n      throw new Error(\n          `tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and ` +\n          `WEBGL_VERSION=2 not yet supported.`);\n    }\n\n    let buffer = null;\n    let tmpDownloadTarget: TensorInfo;\n\n    if (dtype !== 'complex64' && env().get('WEBGL_BUFFER_SUPPORTED')) {\n      // Possibly copy the texture into a buffer before inserting a fence.\n      tmpDownloadTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpDownloadTarget.dataId);\n\n      buffer = this.gpgpu.createBufferFromTexture(\n          tmpData.texture, ...tex_util.getDenseTexShape(shape));\n    }\n\n    this.pendingRead.set(dataId, []);\n\n    if (dtype !== 'complex64') {\n      // Create a fence and wait for it to resolve.\n      await this.gpgpu.createAndWaitForFence();\n    }\n\n    // Download the values from the GPU.\n    let vals: Float32Array;\n    if (dtype === 'complex64') {\n      const ps = await Promise.all(\n          [complexTensors.real.data(), complexTensors.imag.data()]);\n      const realValues = ps[0];\n      const imagValues = ps[1];\n      vals = backend_util.mergeRealAndImagArrays(\n          realValues as Float32Array, imagValues as Float32Array);\n    } else if (buffer == null) {\n      vals = this.getValuesFromTexture(dataId);\n    } else {\n      const size = util.sizeFromShape(shape);\n      vals = this.gpgpu.downloadFloat32MatrixFromBuffer(buffer, size);\n    }\n    if (tmpDownloadTarget != null) {\n      this.disposeIntermediateTensorInfo(tmpDownloadTarget);\n    }\n    const dTypeVals = this.convertAndCacheOnCPU(dataId, vals);\n\n    const subscribers = this.pendingRead.get(dataId);\n    this.pendingRead.delete(dataId);\n\n    // Notify all pending reads.\n    subscribers.forEach(resolve => resolve(dTypeVals));\n    if (this.pendingDisposal.has(dataId)) {\n      this.pendingDisposal.delete(dataId);\n      this.disposeData(dataId);\n      this.pendingDeletes--;\n    }\n    return dTypeVals;\n  }\n\n  private checkNumericalProblems(values: BackendValues): void {\n    if (values == null) {\n      return;\n    }\n    for (let i = 0; i < values.length; i++) {\n      const num = values[i] as number;\n      if (!webgl_util.canBeRepresented(num)) {\n        if (env().getBool('WEBGL_RENDER_FLOAT32_CAPABLE')) {\n          throw Error(\n              `The value ${num} cannot be represented with your ` +\n              `current settings. Consider enabling float32 rendering: ` +\n              `'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`);\n        }\n        throw Error(`The value ${num} cannot be represented on this device.`);\n      }\n    }\n  }\n\n  private getValuesFromTexture(dataId: DataId): Float32Array {\n    const {shape, dtype, isPacked} = this.texData.get(dataId);\n    const size = util.sizeFromShape(shape);\n    if (env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED')) {\n      const tmpTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpTarget.dataId);\n      const vals = this.gpgpu\n                       .downloadMatrixFromPackedTexture(\n                           tmpData.texture, ...tex_util.getDenseTexShape(shape))\n                       .subarray(0, size);\n\n      this.disposeIntermediateTensorInfo(tmpTarget);\n\n      return vals;\n    }\n\n    const shouldUsePackedProgram =\n        env().getBool('WEBGL_PACK') && isPacked === true;\n    const outputShape =\n        shouldUsePackedProgram ? webgl_util.getShapeAs3D(shape) : shape;\n    const program = shouldUsePackedProgram ?\n        new EncodeFloatPackedProgram(outputShape as [number, number, number]) :\n        new EncodeFloatProgram(outputShape);\n    const output = this.runWebGLProgram(\n        program, [{shape: outputShape, dtype, dataId}], 'float32');\n    const tmpData = this.texData.get(output.dataId);\n    const vals =\n        this.gpgpu\n            .downloadByteEncodedFloatMatrixFromOutputTexture(\n                tmpData.texture, tmpData.texShape[0], tmpData.texShape[1])\n            .subarray(0, size);\n    this.disposeIntermediateTensorInfo(output);\n\n    return vals;\n  }\n\n  async time(f: () => void): Promise<WebGLTimingInfo> {\n    const oldActiveTimers = this.activeTimers;\n    const newActiveTimers: TimerNode[] = [];\n\n    let outerMostTime = false;\n    if (this.programTimersStack == null) {\n      this.programTimersStack = newActiveTimers;\n      outerMostTime = true;\n    } else {\n      this.activeTimers.push(newActiveTimers);\n    }\n    this.activeTimers = newActiveTimers;\n\n    f();\n\n    // needing to split these up because util.flatten only accepts certain types\n    const flattenedActiveTimerQueries =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.query))\n            .filter(d => d != null);\n    const flattenedActiveTimerNames =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.name))\n            .filter(d => d != null);\n\n    this.activeTimers = oldActiveTimers;\n\n    if (outerMostTime) {\n      this.programTimersStack = null;\n    }\n\n    const res: WebGLTimingInfo = {\n      uploadWaitMs: this.uploadWaitMs,\n      downloadWaitMs: this.downloadWaitMs,\n      kernelMs: null,\n      wallMs: null  // will be filled by the engine\n    };\n\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      const kernelMs = await Promise.all(flattenedActiveTimerQueries);\n\n      res['kernelMs'] = util.sum(kernelMs);\n      res['getExtraProfileInfo'] = () =>\n          kernelMs.map((d, i) => ({name: flattenedActiveTimerNames[i], ms: d}))\n              .map(d => `${d.name}: ${d.ms}`)\n              .join(', ');\n    } else {\n      res['kernelMs'] = {\n        error: 'WebGL query timers are not supported in this environment.'\n      };\n    }\n\n    this.uploadWaitMs = 0;\n    this.downloadWaitMs = 0;\n    return res;\n  }\n  memory(): WebGLMemoryInfo {\n    return {\n      unreliable: false,\n      numBytesInGPU: this.numBytesInGPU,\n      numBytesInGPUAllocated: this.textureManager.numBytesAllocated,\n      numBytesInGPUFree: this.textureManager.numBytesFree\n    } as WebGLMemoryInfo;\n  }\n\n  private startTimer(): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.beginQuery();\n    }\n    return {startMs: util.now(), endMs: null};\n  }\n\n  private endTimer(query: WebGLQuery|CPUTimerQuery): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      this.gpgpu.endQuery();\n      return query;\n    }\n    (query as CPUTimerQuery).endMs = util.now();\n    return query;\n  }\n\n  private async getQueryTime(query: WebGLQuery|CPUTimerQuery): Promise<number> {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.waitForQueryAndGetTime(query as WebGLQuery);\n    }\n    const timerQuery = query as CPUTimerQuery;\n    return timerQuery.endMs - timerQuery.startMs;\n  }\n\n  private pendingDeletes = 0;\n\n  disposeData(dataId: DataId): void {\n    if (this.pendingDisposal.has(dataId)) {\n      return;\n    }\n    if (this.pendingRead.has(dataId)) {\n      this.pendingDisposal.add(dataId);\n      this.pendingDeletes++;\n      return;\n    }\n    // No-op if already disposed.\n    if (!this.texData.has(dataId)) {\n      return;\n    }\n\n    this.releaseGPUData(dataId);\n    const {complexTensors} = this.texData.get(dataId);\n    if (complexTensors != null) {\n      complexTensors.real.dispose();\n      complexTensors.imag.dispose();\n    }\n    this.texData.delete(dataId);\n  }\n\n  private releaseGPUData(dataId: DataId): void {\n    const {texture, dtype, texShape, usage, isPacked, slice} =\n        this.texData.get(dataId);\n    const key = slice && slice.origDataId || dataId;\n    const refCount = this.dataRefCount.get(key);\n    if (refCount > 1) {\n      this.dataRefCount.set(key, refCount - 1);\n    } else {\n      this.dataRefCount.delete(key);\n      if (texture != null) {\n        this.numBytesInGPU -= this.computeBytes(texShape, dtype);\n        this.textureManager.releaseTexture(texture, texShape, usage, isPacked);\n      }\n    }\n    const texData = this.texData.get(dataId);\n    texData.texture = null;\n    texData.texShape = null;\n    texData.isPacked = false;\n    texData.slice = null;\n  }\n\n  getTexture(dataId: DataId): WebGLTexture {\n    this.uploadToGPU(dataId);\n    return this.texData.get(dataId).texture;\n  }\n\n  /**\n   * Returns internal information for the specific data bucket. Used in unit\n   * tests.\n   */\n  getDataInfo(dataId: DataId): TextureData {\n    return this.texData.get(dataId);\n  }\n\n  private getCPUBackend(): KernelBackend|null {\n    if (!env().getBool('WEBGL_CPU_FORWARD')) {\n      return null;\n    }\n\n    if (this.cpuBackend == null) {\n      this.cpuBackend = engine().findBackend('cpu');\n    }\n\n    return this.cpuBackend;\n  }\n\n  /*\n  Tests whether all the inputs to an op are small and on the CPU. This heuristic\n  determines when it would be faster to execute a kernel on the CPU. WebGL\n  kernels opt into running this check and forwarding when appropriate.\n  TODO(https://github.com/tensorflow/tfjs/issues/872): Develop a more\n  sustainable strategy for optimizing backend execution of ops.\n   */\n  shouldExecuteOnCPU(\n      inputs: TensorInfo[],\n      sizeThreshold = CPU_HANDOFF_SIZE_THRESHOLD): boolean {\n    const cpuBackend = this.getCPUBackend();\n    if (!this.warnedAboutCPUBackend && cpuBackend == null) {\n      console.warn(\n          'Your application contains ops that are small enough to be ' +\n          'executed on the CPU backend, however the CPU backend cannot ' +\n          'be found. Consider importing the CPU backend ' +\n          '(@tensorflow/tfjs-backend-cpu) for better performance.');\n\n      this.warnedAboutCPUBackend = true;\n    }\n\n    return cpuBackend != null &&\n        inputs.every(\n            input => this.texData.get(input.dataId).texture == null &&\n                util.sizeFromShape(input.shape) < sizeThreshold);\n  }\n\n  getGPGPUContext(): GPGPUContext {\n    return this.gpgpu;\n  }\n\n  complex<T extends Tensor>(real: T, imag: T): T {\n    const result = this.makeOutput(real.shape, 'complex64');\n    const resultData = this.texData.get(result.dataId);\n    // The backend owns the reference to the underlying real and imaginary\n    // clones. These will explicitly get disposed when the complex tensor is\n    // disposed.\n    resultData.complexTensors = {\n      real: engine().keep(real.clone()),\n      imag: engine().keep(imag.clone())\n    };\n\n    return result as T;\n  }\n  real<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.real.clone() as T;\n  }\n  imag<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.imag.clone() as T;\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = sliceImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, begin, size, x.shape,\n          x.dtype);\n      return this.makeOutput(size, x.dtype, outValues);\n    }\n    // Short-circuit computation if the slice is zero-sized.\n    if (util.sizeFromShape(size) === 0) {\n      return tensor([], size, x.dtype) as T;\n    }\n    const {isPacked} = this.texData.get(x.dataId);\n    const isContinous = slice_util.isSliceContinous(x.shape, begin, size);\n    if (isPacked || !isContinous) {\n      const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n          new SlicePackedProgram(size) :\n          new SliceProgram(size);\n      const customSetup = program.getCustomSetupFunc(begin);\n      return this.compileAndRun(program, [x], null, customSetup);\n    }\n    this.uploadToGPU(x.dataId);\n    return this.shallowSlice(x, begin, size) as T;\n  }\n\n  private shallowSlice(x: Tensor, begin: number[], size: number[]): Tensor {\n    const xTexData = this.texData.get(x.dataId);\n    const t = this.makeOutput(size, x.dtype);\n    const newTexData = this.texData.get(t.dataId);\n    // Copy texture data from the original tensor.\n    Object.assign(newTexData, xTexData);\n    newTexData.shape = size;\n    newTexData.dtype = x.dtype;\n    let flatOffset = slice_util.computeFlatOffset(begin, x.strides);\n    if (xTexData.slice) {\n      // We are slicing an already sliced tensor, so we have to accumulate\n      // the offset.\n      flatOffset += xTexData.slice.flatOffset;\n    }\n    newTexData.slice = {\n      flatOffset,\n      // Point to the original dataId, which is used to do ref counting.\n      origDataId: xTexData.slice && xTexData.slice.origDataId || x.dataId\n    };\n\n    // Increase the ref count for that data bucket.\n    const refCount = this.dataRefCount.get(newTexData.slice.origDataId) || 1;\n    this.dataRefCount.set(newTexData.slice.origDataId, refCount + 1);\n\n    return t;\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x], () => this.cpuBackend.stridedSlice(x, begin, end, strides));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tensor([], outShape) as T;\n    }\n\n    const program = new StridedSliceProgram(begin, strides, outShape);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new ReversePackedProgram(x.shape, axis) :\n        new ReverseProgram(x.shape, axis);\n    return this.compileAndRun(program, [x]);\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    if (tensors[0].dtype === 'complex64') {\n      const reals = tensors.map((t) => real(t));\n      const imags = tensors.map((t) => imag(t));\n      return complex(this.concat(reals, axis), this.concat(imags, axis));\n    }\n\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n    if (tensors.length > env().getNumber('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.concat(tensors.slice(0, midIndex), axis);\n      const rightSide = this.concat(tensors.slice(midIndex), axis);\n      return this.concat([leftSide, rightSide], axis);\n    }\n    if (env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') && tensors[0].rank > 1) {\n      const program = new ConcatPackedProgram(tensors.map(t => t.shape), axis);\n      return this.compileAndRun(program, tensors);\n    }\n    // Any concat of n-dimensional tensors across any axis can be reduced to\n    // a concatenation of two-dimensional tensors across the axis 1 by first\n    // partitioning the axes of the original tensors into those less than the\n    // axis to be concatenated and the rest. Then reshape the tensors\n    // into a two-dimensional tensor by collapsing these two sets of axes and\n    // concatenate the resulting matrices across the axis 1, finally reshaping\n    // the result to have the proper shape.\n    const outShape =\n        backend_util.computeOutShape(tensors.map(t => t.shape), axis);\n    const tensors2D =\n        tensors.map(t => t.as2D(-1, util.sizeFromShape(t.shape.slice(axis))));\n    const program = new ConcatProgram(tensors2D.map(t => t.shape));\n    const res: Tensor = this.compileAndRun(program, tensors2D);\n    return res.reshape(outShape);\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    const cpuRes = this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.neg(x));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.NEG, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.NEG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const [batch, , ] = a.shape;\n\n    // Since the matrices are vectors, it is faster to call mul().sum()\n    // because sum() is O(sqrt(N)) due to divide-and-conquer.\n    if ((outerShapeA === 1 || outerShapeB === 1) &&\n        sharedDim > MATMUL_SHARED_DIM_THRESHOLD) {\n      if (transposeA) {\n        a = transpose(a, [0, 2, 1]);\n      }\n      if (transposeB) {\n        b = transpose(b, [0, 2, 1]);\n      }\n\n      const a3D = outerShapeB === 1 ? a : a.as3D(batch, sharedDim, 1);\n      const axis = outerShapeB === 1 ? 2 : 1;\n      const b3D = outerShapeB === 1 ? b.as3D(batch, 1, sharedDim) : b;\n      return this.multiply(a3D, b3D).sum(axis, true /* keepDims */);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB);\n    return this.compileAndRun<Tensor3D>(program, [a, b], dtype);\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const [batch, , ] = a.shape;\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB,\n        hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [a, b];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun<Tensor3D>(program, inputs, dtype);\n  }\n\n  multiply(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64') {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n\n      const realProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.REAL, a.shape, b.shape);\n      const imagProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.IMAG, a.shape, b.shape);\n\n      const inputs = [\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.imag),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.imag)\n      ];\n      const real = this.compileAndRun<Tensor>(realProgram, inputs);\n      const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n\n      const complex = this.complex(real, imag);\n      real.dispose();\n      imag.dispose();\n      return complex;\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = multiplyImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.MUL, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.MUL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], a.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n        new LRNPackedProgram(x.shape, radius, bias, alpha, beta) :\n        new LRNProgram(x.shape, radius, bias, alpha, beta);\n    return this.compileAndRun(program, [x]);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program =\n        new LRNGradProgram(inputImage.shape, depthRadius, bias, alpha, beta);\n    return this.compileAndRun(program, [inputImage, outputImage, dy]);\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    if (x.dtype === 'string') {\n      const data = this.readSync(x.dataId) as Uint8Array[];\n      const decodedData = data.map(d => util.decodeString(d));\n      const buf = buffer(x.shape, x.dtype, decodedData);\n      return tile(buf, reps) as T;\n    }\n    const program = new TileProgram(x.shape, reps);\n    return this.compileAndRun(program, [x]);\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new PadPackedProgram(x.shape, paddings, constantValue) :\n        new PadProgram(x.shape, paddings, constantValue);\n    return this.compileAndRun(program, [x]);\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x, indices], () => this.cpuBackend.gather(x, indices, axis));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = new GatherProgram(x.shape, indices.size, axis);\n    return this.compileAndRun(program, [x, indices]);\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'batchToSpaceND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: Array<[number, number]>): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'spaceToBatchND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const completePaddings: Array<[number, number]> = [[0, 0]];\n    completePaddings.push(...paddings);\n    for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n      completePaddings.push([0, 0]);\n    }\n\n    const paddedX = x.pad(completePaddings);\n\n    const reshapedPaddedShape =\n        backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n    const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n        reshapedPaddedShape.length, blockShape.length, false);\n\n    const flattenShape = backend_util.getReshapedPermuted(\n        paddedX.shape, blockShape, prod, false);\n\n    const paddedXT = transpose(\n        paddedX.reshape(reshapedPaddedShape),\n        permutedReshapedPaddedPermutation);\n    return reshape(paddedXT, flattenShape) as T;\n  }\n\n  private reduce(\n      x: Tensor2D, reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod',\n      dtype: DataType): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const outSize = Math.ceil(inSize / windowSize);\n    const reduceInfo = {windowSize, inSize, batchSize, outSize};\n    const program = new ReduceProgram(reduceInfo, reduceType);\n    const output = this.compileAndRun<Tensor2D>(program, [x], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.reduce(output, reduceType, dtype);\n  }\n\n  private argReduce(\n      x: Tensor2D, reduceType: 'max'|'min',\n      bestIndicesA: Tensor2D = null): Tensor2D {\n    let batchSize = x.shape[0];\n    let inSize = x.shape[1];\n    if (bestIndicesA != null) {\n      batchSize = bestIndicesA.shape[0];\n      inSize = bestIndicesA.shape[1];\n    }\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const reduceInfo = {\n      windowSize,\n      inSize,\n      batchSize,\n      outSize: Math.ceil(inSize / windowSize)\n    };\n    const program =\n        new ArgMinMaxProgram(reduceInfo, reduceType, bestIndicesA == null);\n    const inputs = [x];\n    if (bestIndicesA != null) {\n      inputs.push(bestIndicesA);\n    }\n    const output = this.compileAndRun<Tensor2D>(program, inputs, 'int32');\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.argReduce(x, reduceType, output);\n  }\n\n  private argReducePacked(\n      x: Tensor, reduceType: 'max'|'min', bestIndicesA: Tensor = null): Tensor {\n    const inShape = bestIndicesA != null ? bestIndicesA.shape : x.shape;\n    const inSize = inShape[inShape.length - 1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const program = new ArgMinMaxPackedProgram(\n        inShape, windowSize, reduceType, bestIndicesA == null);\n    const inputs = bestIndicesA == null ? [x] : [x, bestIndicesA];\n    const output = this.compileAndRun<Tensor>(program, inputs, 'int32');\n    if (output.rank === x.rank) {\n      return this.argReducePacked(x, reduceType, output);\n    }\n    return output;\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'sum', outputDType).reshape(outShape);\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.prod(x, axes));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'prod', outputDType).reshape(outShape);\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    let axis = 0;\n    const permutation = backend_util.getAxesPermutation([axis], x.rank);\n    let permutedX = x;\n    if (permutation != null) {\n      permutedX = transpose(x, permutation);\n      axis = backend_util.getInnerMostAxes(1, x.rank)[0];\n    }\n\n    const outShape =\n        segment_util.computeOutShape(permutedX.shape, axis, numSegments);\n    const inSize = util.sizeFromShape([permutedX.shape[axis]]);\n    const a2D = permutedX.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    let result =\n        this.segOpCompute(\n                a2D, 'unsortedSegmentSum', segmentIds, outputDType, numSegments)\n            .reshape(outShape);\n    if (permutation != null) {\n      result =\n          transpose(result, backend_util.getUndoAxesPermutation(permutation));\n    }\n    return result;\n  }\n\n  private segOpCompute(\n      x: Tensor2D, segOpType: 'unsortedSegmentSum', segmentIds: Tensor1D,\n      dtype: DataType, numSegments: number): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize =\n        segment_util.segOpComputeOptimalWindowSize(inSize, numSegments);\n    const segOpInfo = {windowSize, inSize, batchSize, numSegments};\n    const program = new SegmentOpProgram(segOpInfo, segOpType);\n    const output =\n        this.compileAndRun<Tensor2D>(program, [x, segmentIds], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === numSegments) {\n      return output;\n    }\n    segmentIds = range(0, numSegments).tile([inSize / windowSize]);\n    return this.segOpCompute(output, segOpType, segmentIds, dtype, numSegments);\n  }\n\n  private argMinMaxReduce(x: Tensor, axis: number, reduceType: 'min'|'max'):\n      Tensor {\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims(\n        'arg' + reduceType.charAt(0).toUpperCase() + reduceType.slice(1), axes,\n        x.rank);\n    if (!env().getBool('WEBGL_PACK_REDUCE') || x.rank <= 2) {\n      const [outShape, reduceShape] =\n          backend_util.computeOutAndReduceShapes(x.shape, axes);\n      const inSize = util.sizeFromShape(reduceShape);\n      const a2D = x.as2D(-1, inSize);\n      return this.argReduce(a2D, reduceType).reshape(outShape);\n    }\n    return this.argReducePacked(x, reduceType);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'min');\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'max');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `WebGL cumsum shader expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const size = x.shape[axis];\n    let result = x;\n    // Use cumsum parallel algorithm, ref:\n    // https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-39-parallel-prefix-sum-scan-cuda\n    for (let i = 0; i <= Math.ceil(Math.log2(size)) - 1; i++) {\n      const program = new CumSumProgram(x.shape, false, reverse);\n      const customSetup = program.getCustomSetupFunc(i);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result], result.dtype, customSetup);\n      prevResult.dispose();\n    }\n    // For exclusive cumsum, shift the end result in the direction of sum and\n    // add 0 to the front index.\n    if (exclusive) {\n      const program = new CumSumProgram(x.shape, exclusive, reverse);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result]);\n      prevResult.dispose();\n    }\n\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.EQUAL, 'bool');\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.NOT_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.NOT_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.less(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.LESS, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LESS_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.greater(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.GREATER, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.GREATER, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.GREATER_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.GREATER_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalNot<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOGICAL_NOT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_AND, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_AND, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_OR, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_OR, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    const program = new SelectProgram(condition.rank, a.shape, a.rank);\n    return this.compileAndRun(\n        program, [condition, a, b], upcastType(a.dtype, b.dtype));\n  }\n\n  where(condition: Tensor): Tensor2D {\n    backend_util.warn(\n        'tf.where() in webgl locks the UI thread. ' +\n        'Call tf.whereAsync() instead');\n    const condVals = condition.dataSync();\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    const xVals = x.dataSync();\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'min', a2D.dtype).reshape(outShape);\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.minimum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MIN, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MIN, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MOD, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MOD, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.maximum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MAX, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MAX, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'all', a2D.dtype).reshape(outShape);\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'any', a2D.dtype).reshape(outShape);\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    const op = binaryop_gpu.INT_DIV;\n    const outputDtype = 'int32';\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.INT_DIV, outputDtype);\n    }\n    const program = new BinaryOpProgram(op, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], outputDtype);\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.ADD);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = addImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.ADD, dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.ADD, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  private packedUnaryOp(x: TensorInfo, op: string, dtype: DataType) {\n    const program = new UnaryOpPackedProgram(x.shape, op);\n    return this.compileAndRun<Tensor>(program, [x], dtype);\n  }\n\n  private packedBinaryOp(\n      a: TensorInfo, b: TensorInfo, op: string, dtype: DataType,\n      checkOutOfBounds = false) {\n    const program =\n        new BinaryOpPackedProgram(op, a.shape, b.shape, checkOutOfBounds);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  /**\n   * Computes a complex binary operation that can be decomposed into a simple\n   * binary operation on both the real and imagary parts.\n   */\n  private complexSeparableBinaryOp(a: Tensor, b: Tensor, op: string): Tensor {\n    const aData = this.texData.get(a.dataId);\n    const bData = this.texData.get(b.dataId);\n\n    const [real, imag] = [\n      [aData.complexTensors.real, bData.complexTensors.real],\n      [aData.complexTensors.imag, bData.complexTensors.imag]\n    ].map(complexParts => {\n      const [aPart, bPart] = complexParts;\n\n      const aHandle = this.makeComplexComponentTensorInfo(a, aPart);\n      const bHandle = this.makeComplexComponentTensorInfo(b, bPart);\n\n      const program = new BinaryOpProgram(op, a.shape, b.shape);\n      return this.compileAndRun<Tensor>(\n          program, [aHandle, bHandle], upcastType(aPart.dtype, bPart.dtype));\n    });\n\n    const complex = this.complex(real, imag);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  // Returns a TensorInfo with the complex shape and the dataId of the\n  // underlying part. We need to do this because a reshaped complex tensor is\n  // not reflected in its parts.\n  private makeComplexComponentTensorInfo(\n      complexTensor: Tensor, complexPart: Tensor): TensorInfo {\n    return {\n      dataId: complexPart.dataId,\n      dtype: complexPart.dtype,\n      shape: complexTensor.shape\n    };\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n\n    // Limit the number of uploaded textures for optimization.\n    if (tensors.length > env().get('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.addN(tensors.slice(0, midIndex));\n      const rightSide = this.addN(tensors.slice(midIndex));\n      return this.addN([leftSide, rightSide]);\n    }\n\n    const dtype =\n        tensors.map(t => t.dtype).reduce((d1, d2) => upcastType(d1, d2));\n    const shapes = tensors.map(t => t.shape);\n    // We can make sure shapes are identical in op level.\n    const usePackedOp = env().getBool('WEBGL_PACK');\n    const program = usePackedOp ?\n        new AddNPackedProgram(tensors[0].shape, shapes) :\n        new AddNProgram(tensors[0].shape, shapes);\n    return this.compileAndRun<T>(program, tensors, dtype);\n  }\n\n  subtract(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.SUB);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = subImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.SUB, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.SUB, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    const usePackedOp = env().getBool('WEBGL_PACK_BINARY_OPERATIONS');\n    const program = usePackedOp ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.POW, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.POW, a.shape, b.shape);\n    const dtype = upcastType(a.dtype, b.dtype);\n    return this.compileAndRun<T>(program, [a, b], dtype);\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          ceilImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.CEIL, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.CEIL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  floor<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = floorImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.FLOOR, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.FLOOR);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_NAN);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_INF);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_FINITE);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n\n  round<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ROUND);\n    return this.compileAndRun(program, [x]);\n  }\n\n  exp<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          expImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXP, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXP);\n    return this.compileAndRun(program, [x]);\n  }\n\n  expm1<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = expm1ImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXPM1, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXPM1);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n    const a = this.subtract(logits, maxLogit.reshape(expandedShape));\n    const b = this.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax kernel\n    // modularization.\n    return div(b, sumExp);\n  }\n\n  log<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          logImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.LOG, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  log1p<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG1P);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sqrt<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  rsqrt<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = rsqrtImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.RSQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reciprocal<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.RECIPROCAL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU6);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU6);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  prelu<T extends Tensor>(x: T, alpha: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.PRELU, x.shape, alpha.shape) :\n        new BinaryOpProgram(binaryop_gpu.PRELU, x.shape, alpha.shape);\n    return this.compileAndRun(program, [x, alpha]);\n  }\n\n  elu<T extends Tensor>(x: T): T {\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.ELU, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.ELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.ELU_DER, dy.shape, y.shape) :\n        new BinaryOpProgram(binaryop_gpu.ELU_DER, dy.shape, y.shape);\n    return this.compileAndRun(program, [dy, y]);\n  }\n\n  selu<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  int<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TO_INT);\n    return this.compileAndRun(program, [x], 'int32');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    let program;\n    if (env().getBool('WEBGL_PACK_CLIP')) {\n      program = new ClipPackedProgram(x.shape);\n    } else {\n      program = new ClipProgram(x.shape);\n    }\n    const customSetup = program.getCustomSetupFunc(min, max);\n    return this.compileAndRun(program, [x], null, customSetup);\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    // TODO: handle cases when x is complex.\n    if (this.shouldExecuteOnCPU([x]) && x.dtype !== 'complex64') {\n      const outValues =\n          simpleAbsImplCPU(this.texData.get(x.dataId).values as TypedArray);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.ABS, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.ABS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  complexAbs<T extends Tensor>(x: T): T {\n    const xData = this.texData.get(x.dataId);\n\n    const program = new ComplexAbsProgram(x.shape);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    return this.compileAndRun<Tensor>(program, inputs) as T;\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGMOID);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SOFTPLUS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASIN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acos<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atan<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATAN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.COSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  tanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ERF);\n    return this.compileAndRun(program, [x]);\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.STEP(alpha));\n    return this.compileAndRun(program, [x]);\n  }\n\n  private conv2dByMatMul(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Reshapes conv2D input to 2D tensors, uses matMul and then reshape the\n    // result from 2D to 4D.\n    const xShape = x.shape;\n    const xTexData = this.texData.get(x.dataId);\n    const sharedMatMulDim = convInfo.inChannels;\n    const outerShapeX = xShape[0] * xShape[1] * xShape[2];\n    const outerShapeFilter = convInfo.outChannels;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const transposeA = false;\n    const transposeB = false;\n\n    // TODO: Once reduction ops are packed, batchMatMul will always be packed\n    // and we can remove this condition.\n    const batchMatMulWillBeUnpacked =\n        (outerShapeX === 1 || outerShapeFilter === 1) &&\n        sharedMatMulDim > MATMUL_SHARED_DIM_THRESHOLD;\n    const reshapeWillBeExpensive = xShape[2] % 2 !== 0 && !!xTexData.isPacked;\n\n    if (batchMatMulWillBeUnpacked || !env().getBool('WEBGL_LAZILY_UNPACK') ||\n        !env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ||\n        !reshapeWillBeExpensive) {\n      const targetShape = isChannelsLast ? xShape[0] * xShape[1] * xShape[2] :\n                                           xShape[0] * xShape[2] * xShape[3];\n      const xReshaped = reshape(x, [1, targetShape, convInfo.inChannels]);\n      const filterReshaped =\n          reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n      const result = this.fusedBatchMatMul({\n        a: xReshaped as Tensor3D,\n        b: filterReshaped as Tensor3D,\n        transposeA,\n        transposeB,\n        bias,\n        activation,\n        preluActivationWeights\n      });\n      return reshape(result, convInfo.outShape);\n    }\n\n    // Following optimization is specific to packed |x| with odd row count\n    // (For example, in channelLast mode, 'row count' refers to x.shape[2]):\n    // we avoid expensive packed 2x2 reshape by padding row count to next,\n    // even number. When x.shape[2] is odd, the result of packed batchMatMul is\n    // the same (has the same texture layout and and values in the texture) as\n    // it is for even x.shape[2] + 1. We make the odd-rows tensor to look like\n    // even-rows tensor before the operation and, after the batchMatMul,\n    // fix the even-rows result to have odd number of rows.\n    const targetShape = isChannelsLast ?\n        xShape[0] * xShape[1] * (xShape[2] + 1) :\n        xShape[0] * xShape[2] * (xShape[3] + 1);\n    const xReshaped: TensorInfo = {\n      dataId: x.dataId,\n      shape: [1, targetShape, convInfo.inChannels],\n      dtype: x.dtype\n    };\n    // xTexData.shape gets referenced from GPGPUBinary.inShapeInfos.\n    // Decrementing row count, after batchMatMul->...->compileProgram leads to\n    // invalid row count within the reference in GPGPUBinary.inShapeInfos.\n    // Alternative fix would be to provide a copy to GPGPUBinary.inShapeInfos\n    // in compileProgram method, but that would affect compilation of all\n    // programs - instead, provide a copy here, with even row count, before\n    // calling batchMatMul->...->compileProgram and after that, the original\n    // xTexData.shape is restored.\n    const originalXTexDataShape = xTexData.shape;\n    xTexData.shape = xTexData.shape.slice();\n    xTexData.shape[xTexData.shape.length - 2]++;\n    util.assert(\n        webgl_util.isReshapeFree(xTexData.shape, xReshaped.shape),\n        () => `packed reshape ${xTexData.shape} to ${\n            xReshaped.shape} isn't free`);\n    const filterReshaped =\n        reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n    const pointwiseConv = this.fusedBatchMatMul({\n      a: xReshaped as Tensor3D,\n      b: filterReshaped as Tensor3D,\n      transposeA,\n      transposeB,\n      bias,\n      activation,\n      preluActivationWeights\n    });\n    const pointwiseConvTexData = this.texData.get(pointwiseConv.dataId);\n    util.assert(\n        pointwiseConvTexData.isPacked,\n        () => 'batchMatMul result is expected to be packed');\n    // Restore the input shape to original.\n    xTexData.shape = originalXTexDataShape;\n    // Set the output shape - there is no need for expensive reshape as data\n    // layout is already correct.\n    pointwiseConvTexData.shape = convInfo.outShape;\n    return engine().makeTensorFromDataId(\n               pointwiseConv.dataId, convInfo.outShape, pointwiseConv.dtype) as\n        Tensor4D;\n  }\n\n  private conv2dWithIm2Row(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Rearranges conv2d input so each block to be convolved over forms the\n    // column of a new matrix with shape [filterWidth * filterHeight *\n    // inChannels, outHeight * outWidth]. The filter is also rearranged so each\n    // output channel forms a row of a new matrix with shape [outChannels,\n    // filterWidth * filterHeight * inChannels]. The convolution is then\n    // computed by multiplying these matrices and reshaping the result.\n    const {\n      filterWidth,\n      filterHeight,\n      inChannels,\n      outWidth,\n      outHeight,\n      dataFormat\n    } = convInfo;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n\n    const sharedDim = filterWidth * filterHeight * inChannels;\n    const numCols = outHeight * outWidth;\n    const x2ColShape = [sharedDim, numCols];\n    const transposeA = true;\n    const transposeB = false;\n\n    const xSqueezed = x.squeeze([0]);\n    const w2Row = filter.reshape([1, sharedDim, -1]);\n\n    const im2ColProgram =\n        new Im2ColPackedProgram(x2ColShape, xSqueezed.shape, convInfo);\n    const im2Col: Tensor3D =\n        this.compileAndRun<Tensor2D>(im2ColProgram, [xSqueezed]).reshape([\n          1, x2ColShape[0], x2ColShape[1]\n        ]);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const matmulProgram = new MatMulPackedProgram(\n        im2Col.shape, [1, numCols, convInfo.outChannels], transposeA,\n        transposeB, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [im2Col, w2Row];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    const product = this.compileAndRun<Tensor4D>(matmulProgram, inputs);\n\n    if (isChannelsLast) {\n      return product.reshape([1, outHeight, outWidth, convInfo.outChannels]);\n    } else {\n      return product.reshape([1, convInfo.outChannels, outHeight, outWidth]);\n    }\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && input.shape[0] === 1) {\n      return this.conv2dWithIm2Row(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, false) : null;\n    const program = new Conv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [input, filter];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun(program, inputs);\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(x, filter, convInfo);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && x.shape[0] === 1) {\n      return this.conv2dWithIm2Row(x, filter, convInfo);\n    }\n    const program = new Conv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new Conv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    const program = new Conv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    const shouldPackDepthwiseConv = env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1;\n    const fusedActivation = activation ?\n        mapActivationToShaderProgram(activation, shouldPackDepthwiseConv) :\n        null;\n    const inputs: Tensor[] = [input, filter];\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    if (hasBias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (shouldPackDepthwiseConv) {\n      program = new DepthwiseConvPacked2DProgram(\n          convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n      return this.compileAndRun(program, inputs);\n    }\n\n    program = new DepthwiseConv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    return this.compileAndRun(program, inputs);\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1) {\n      program = new DepthwiseConvPacked2DProgram(convInfo);\n      return this.compileAndRun(program, [x, filter]);\n    }\n\n    program = new DepthwiseConv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Conv3DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return backend_util.castTensor(x, dtype, this);\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = this.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'avg', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const avgPool3dBackpropProgram = new AvgPool3DBackpropProgram(convInfo);\n    return this.compileAndRun(avgPool3dBackpropProgram, [dy], x.dtype);\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'max', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const getPositions = true;\n    const maxPool3dPositionsProgram =\n        new Pool3DProgram(convInfo, 'max', getPositions);\n    const maxPool3dPositions: Tensor5D =\n        this.compileAndRun(maxPool3dPositionsProgram, [x]);\n    const maxPool3dBackPropProgram = new MaxPool3DBackpropProgram(convInfo);\n    const result = this.compileAndRun(\n        maxPool3dBackPropProgram, [dy, maxPool3dPositions], x.dtype);\n    maxPool3dPositions.dispose();\n    return result as Tensor5D;\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_IMAGE_OPERATIONS') ?\n        new ResizeBilinearPackedProgram(\n            x.shape, newHeight, newWidth, alignCorners) :\n        new ResizeBilinearProgram(x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    const program = new ResizeBilinearBackpropProgram(dy, x, alignCorners);\n\n    return this.compileAndRun(program, [dy]);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = new ResizeNearestNeighborProgram(\n        x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x]);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    const program =\n        new ResizeNearestNeigborBackpropProgram(dy, x, alignCorners);\n    return this.compileAndRun(program, [dy]);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    const probs = normalized ? logits : softmax(logits);\n    const batchSize = probs.shape[0];\n    const numOutcomes = probs.shape[1];\n    const program = new MultinomialProgram(batchSize, numOutcomes, numSamples);\n    const customSetup = program.getCustomSetupFunc(seed);\n    return this.compileAndRun(program, [probs], 'int32', customSetup);\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    const program = new OneHotProgram(indices.size, depth, onValue, offValue);\n    return this.compileAndRun(program, [indices]);\n  }\n\n  diag(x: Tensor): Tensor {\n    const program = new DiagProgram(x.size);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    const program = new CropAndResizeProgram(\n        image.shape, boxes.shape, cropSize, method, extrapolationValue);\n    return this.compileAndRun(program, [image, boxes, boxIndex], 'float32');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = (dataFormat === 'NHWC') ? x.shape[1] : x.shape[2];\n    const inputWidth = (dataFormat === 'NHWC') ? x.shape[2] : x.shape[3];\n    const inputDepth = (dataFormat === 'NHWC') ? x.shape[3] : x.shape[1];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const outputShape = (dataFormat === 'NHWC') ?\n        [batchSize, outputHeight, outputWidth, outputDepth] :\n        [batchSize, outputDepth, outputHeight, outputWidth];\n\n    const program = new DepthToSpaceProgram(outputShape, blockSize, dataFormat);\n    return this.compileAndRun(program, [x]);\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n    const flattenIndices = indices.reshape([numUpdates, sliceRank]);\n    const flattenX = updates.reshape([numUpdates, sliceSize]);\n\n    if (outputSize === 0) {\n      return backend_util.reshapeTensor(tensor([]), shape);\n    }\n    const defaultValue = scalar(0);\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, flattenIndices.rank, flattenX.rank, strides,\n        flattenShape);\n    const res: Tensor =\n        this.compileAndRun(program, [flattenX, flattenIndices, defaultValue]);\n    return res.reshape(shape);\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n\n    const sumDupeIndices = false;\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, sparseIndices.rank, sparseValues.rank, strides,\n        [outputSize, 1], sumDupeIndices);\n    const res: Tensor = this.compileAndRun(\n        program, [sparseValues, sparseIndices, defaultValue]);\n    return res.reshape(outputShape);\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    const inverse = false;\n    return this.fftImpl(x, inverse);\n  }\n\n  ifft(x: Tensor2D): Tensor2D {\n    const inverse = true;\n    return this.fftImpl(x, inverse);\n  }\n\n  private fftImpl(x: Tensor2D, inverse: boolean): Tensor2D {\n    const xData = this.texData.get(x.dataId);\n\n    const realProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.REAL, x.shape, inverse);\n    const imagProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.IMAG, x.shape, inverse);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    const real = this.compileAndRun<Tensor>(realProgram, inputs);\n    const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n    const complex = this.complex(real, imag).as2D(x.shape[0], x.shape[1]);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n\n    const flattenIndices = indices.reshape([numSlices, sliceRank]);\n    const flattenX = x.reshape([x.size / sliceSize, sliceSize]);\n    const program =\n        new GatherNDProgram(sliceRank, strides, [numSlices, sliceSize]);\n    const res: Tensor = this.compileAndRun(program, [flattenX, flattenIndices]);\n    return res.reshape(resultShape);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n\n    if (dtype === 'string') {\n      // String type should be handled in CPU memory.\n      const values = util.getArrayFromDType(dtype, util.sizeFromShape(shape));\n      values.fill(value as string);\n      return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n    } else {\n      const program = new FillProgram(shape, value as number);\n      const customSetup = program.getCustomSetupFunc(value as number);\n      return this.compileAndRun(program, [], dtype, customSetup);\n    }\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported under string dtype');\n    } else {\n      // TODO(cais, smilkov): Add WebGL shader for onesLike:\n      //   https://github.com/tensorflow/tfjs/issues/1293\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return this.fill(x.shape, x.dtype === 'string' ? '' : 0, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    // TODO: Use CPU implementation due to the precision problem in Safari.\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  makeTensorInfo(shape: number[], dtype: DataType, values?: BackendValues):\n      TensorInfo {\n    const dataId = this.write(values, shape, dtype);\n    this.texData.get(dataId).usage = null;\n    return {dataId, shape, dtype};\n  }\n\n  private makeOutput<T extends Tensor>(\n      shape: number[], dtype: DataType, values?: BackendValues): T {\n    const {dataId} = this.makeTensorInfo(shape, dtype, values);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  private unpackTensor(input: TensorInfo): TensorInfo {\n    const program = new UnpackProgram(input.shape);\n    return this.runWebGLProgram(program, [input], input.dtype);\n  }\n\n  private packTensor(input: TensorInfo): TensorInfo {\n    const program = new PackProgram(input.shape);\n    const preventEagerUnpackingOutput = true;\n    return this.runWebGLProgram(\n        program, [input], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOutput);\n  }\n\n  private packedReshape(input: TensorInfo, afterShape: number[]): TensorInfo {\n    const input3DShape = [\n      webgl_util.getBatchDim(input.shape),\n      ...webgl_util.getRowsCols(input.shape)\n    ] as [number, number, number];\n    const input3D: TensorInfo = {\n      dtype: input.dtype,\n      shape: input3DShape,\n      dataId: input.dataId\n    };\n    const afterShapeAs3D = [\n      webgl_util.getBatchDim(afterShape), ...webgl_util.getRowsCols(afterShape)\n    ] as [number, number, number];\n\n    const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n    const preventEagerUnpackingOfOutput = true;\n    const output = this.runWebGLProgram(\n        program, [input3D], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOfOutput);\n    return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n  }\n\n  private decode(dataId: DataId): TensorInfo {\n    const texData = this.texData.get(dataId);\n    const {isPacked, shape, dtype} = texData;\n    const shapeAs3D =\n        webgl_util.getShapeAs3D(shape) as [number, number, number];\n    let program;\n    if (isPacked) {\n      program = new DecodeMatrixPackedProgram(shapeAs3D);\n    } else {\n      program = new DecodeMatrixProgram(shapeAs3D);\n    }\n    const preventEagerUnpackingOfOutput = true;\n    const out = this.runWebGLProgram(\n        program, [{shape: shapeAs3D, dtype, dataId}], dtype,\n        null /* customSetup */, preventEagerUnpackingOfOutput);\n    return {dtype, shape, dataId: out.dataId};\n  }\n\n  runWebGLProgram(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): TensorInfo {\n    const output = this.makeTensorInfo(program.outputShape, outputDtype);\n    const outData = this.texData.get(output.dataId);\n    if (program.packedOutput) {\n      outData.isPacked = true;\n    }\n    if (program.outPackingScheme === tex_util.PackingScheme.DENSE) {\n      const texelShape = tex_util.getDenseTexShape(program.outputShape);\n      // For a densely packed output, we explicitly set texShape\n      // so it doesn't get assigned later according to our typical packing\n      // scheme wherein a single texel can only contain values from adjacent\n      // rows/cols.\n      outData.texShape = texelShape.map(d => d * 2) as [number, number];\n    }\n    if (program.outTexUsage != null) {\n      outData.usage = program.outTexUsage;\n    }\n    if (util.sizeFromShape(output.shape) === 0) {\n      // Short-circuit the computation since the result is empty (has 0 in its\n      // shape).\n      outData.values =\n          util.getTypedArrayFromDType(output.dtype as 'float32', 0);\n      return output;\n    }\n\n    const dataToDispose: TensorInfo[] = [];\n    const inputsData: TensorData[] = inputs.map(input => {\n      if (input.dtype === 'complex64') {\n        throw new Error(\n            `GPGPUProgram does not support complex64 input. For complex64 ` +\n            `dtypes, please separate the program into real and imaginary ` +\n            `parts.`);\n      }\n\n      let texData = this.texData.get(input.dataId);\n\n      if (texData.texture == null) {\n        if (!program.packedInputs &&\n            util.sizeFromShape(input.shape) <=\n                env().getNumber('WEBGL_SIZE_UPLOAD_UNIFORM')) {\n          // Upload small tensors that live on the CPU as uniforms, not as\n          // textures. Do this only when the environment supports 32bit floats\n          // due to problems when comparing 16bit floats with 32bit floats.\n          // TODO(https://github.com/tensorflow/tfjs/issues/821): Make it\n          // possible for packed shaders to sample from uniforms.\n          return {\n            shape: input.shape,\n            texData: null,\n            isUniform: true,\n            uniformValues: texData.values as TypedArray\n          };\n        }\n\n        // This ensures that if a packed program's inputs have not yet been\n        // uploaded to the GPU, they get uploaded as packed right off the bat.\n        if (program.packedInputs) {\n          texData.isPacked = true;\n          texData.shape = input.shape;\n        }\n      } else if (!!texData.isPacked !== !!program.packedInputs) {\n        input = texData.isPacked ? this.unpackTensor(input) :\n                                   this.packTensor(input);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n      } else if (\n          texData.isPacked &&\n          !webgl_util.isReshapeFree(texData.shape, input.shape)) {\n        // This is a special case where a texture exists for a tensor\n        // but the shapes are incompatible (due to packing constraints) because\n        // the tensor did not have a chance to go through the packed reshape\n        // shader. This only happens when we reshape the *same* tensor to form\n        // *distinct* inputs to an op, e.g. dotting a vector with itself. This\n        // case will disappear once packed uploading is the default.\n\n        const savedInput = input;\n        const targetShape = input.shape;\n\n        input.shape = texData.shape;\n        input = this.packedReshape(input as Tensor, targetShape);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n\n        savedInput.shape = targetShape;\n      }\n\n      this.uploadToGPU(input.dataId);\n      return {shape: input.shape, texData, isUniform: false};\n    });\n\n    this.uploadToGPU(output.dataId);\n    const outputData:\n        TensorData = {shape: output.shape, texData: outData, isUniform: false};\n    const key = gpgpu_math.makeShaderKey(program, inputsData, outputData);\n    const binary = this.getAndSaveBinary(key, () => {\n      return gpgpu_math.compileProgram(\n          this.gpgpu, program, inputsData, outputData);\n    });\n    const shouldTimeProgram = this.activeTimers != null;\n    let query: WebGLQuery|CPUTimerQuery;\n    if (shouldTimeProgram) {\n      query = this.startTimer();\n    }\n\n    gpgpu_math.runProgram(\n        this.gpgpu, binary, inputsData, outputData, customSetup);\n\n    dataToDispose.forEach(info => this.disposeIntermediateTensorInfo(info));\n\n    if (shouldTimeProgram) {\n      query = this.endTimer(query);\n      this.activeTimers.push(\n          {name: program.constructor.name, query: this.getQueryTime(query)});\n    }\n\n    if (!env().getBool('WEBGL_LAZILY_UNPACK') && outData.isPacked &&\n        preventEagerUnpackingOfOutput === false) {\n      const unpacked = this.unpackTensor(output);\n      this.disposeIntermediateTensorInfo(output);\n      return unpacked;\n    }\n    return output;\n  }\n\n  compileAndRun<K extends TensorInfo>(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype?: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): K {\n    outputDtype = outputDtype || inputs[0].dtype;\n    const outInfo = this.runWebGLProgram(\n        program, inputs, outputDtype, customSetup,\n        preventEagerUnpackingOfOutput);\n    return engine().makeTensorFromDataId(\n               outInfo.dataId, outInfo.shape, outInfo.dtype) as {} as K;\n  }\n\n  private getAndSaveBinary(key: string, getBinary: () => GPGPUBinary):\n      GPGPUBinary {\n    if (!(key in this.binaryCache)) {\n      this.binaryCache[key] = getBinary();\n    }\n    return this.binaryCache[key];\n  }\n\n  getTextureManager(): TextureManager {\n    return this.textureManager;\n  }\n\n  private disposed = false;\n\n  dispose() {\n    if (this.disposed) {\n      return;\n    }\n    // Avoid disposing the compiled webgl programs during unit testing because\n    // it slows down test execution.\n    if (!env().getBool('IS_TEST')) {\n      const allKeys = Object.keys(this.binaryCache);\n      allKeys.forEach(key => {\n        this.gpgpu.deleteProgram(this.binaryCache[key].webGLProgram);\n        delete this.binaryCache[key];\n      });\n    }\n    this.textureManager.dispose();\n    if (this.canvas != null &&\n        (typeof (HTMLCanvasElement) !== 'undefined' &&\n         this.canvas instanceof HTMLCanvasElement)) {\n      this.canvas.remove();\n    } else {\n      this.canvas = null;\n    }\n    if (this.gpgpuCreatedLocally) {\n      this.gpgpu.program = null;\n      this.gpgpu.dispose();\n    }\n    this.disposed = true;\n  }\n\n  floatPrecision(): 16|32 {\n    if (this.floatPrecisionValue == null) {\n      this.floatPrecisionValue = tidy(() => {\n        if (!env().get('WEBGL_RENDER_FLOAT32_ENABLED')) {\n          // Momentarily switching DEBUG flag to false so we don't throw an\n          // error trying to upload a small value.\n          const debugFlag = env().getBool('DEBUG');\n          env().set('DEBUG', false);\n          const underflowCheckValue = this.abs(scalar(1e-8)).dataSync()[0];\n          env().set('DEBUG', debugFlag);\n\n          if (underflowCheckValue > 0) {\n            return 32;\n          }\n        }\n        return 16;\n      });\n    }\n    return this.floatPrecisionValue;\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  private uploadToGPU(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    const {shape, dtype, values, texture, usage, isPacked} = texData;\n\n    if (texture != null) {\n      // Array is already on GPU. No-op.\n      return;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let texShape = texData.texShape;\n    if (texShape == null) {\n      texShape = webgl_util.getTextureShapeFromLogicalShape(shape, isPacked);\n      texData.texShape = texShape;\n    }\n\n    if (values != null) {\n      const shapeAs3D = webgl_util.getShapeAs3D(shape);\n\n      let program;\n      let width = texShape[1], height = texShape[0];\n      const isByteArray = values instanceof Uint8Array;\n\n      if (isPacked) {\n        [width, height] = tex_util.getPackedMatrixTextureShapeWidthHeight(\n            texShape[0], texShape[1]);\n        program = new EncodeMatrixPackedProgram(\n            shapeAs3D, [height, width], isByteArray);\n      } else {\n        program =\n            new EncodeMatrixProgram(shapeAs3D, [height, width], isByteArray);\n      }\n\n      const tempDenseInputHandle = this.makeTensorInfo([height, width], dtype);\n      if (isByteArray) {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.PIXELS;\n      } else {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.UPLOAD;\n      }\n      this.gpgpu.uploadDenseMatrixToTexture(\n          this.getTexture(tempDenseInputHandle.dataId), width, height,\n          values as TypedArray);\n\n      // We want the output to remain packed regardless of the value of\n      // WEBGL_PACK.\n      const preventEagerUnpacking = true;\n      const encodedOutputTarget = this.runWebGLProgram(\n          program, [tempDenseInputHandle], dtype, null, preventEagerUnpacking);\n\n      // Have the original texture assume the identity of the encoded output.\n      const outputTexData = this.texData.get(encodedOutputTarget.dataId);\n      texData.texture = outputTexData.texture;\n      texData.texShape = outputTexData.texShape;\n      texData.isPacked = outputTexData.isPacked;\n      texData.usage = outputTexData.usage;\n\n      this.disposeIntermediateTensorInfo(tempDenseInputHandle);\n      this.texData.delete(encodedOutputTarget.dataId);\n\n      // Once uploaded, don't store the values on cpu.\n      texData.values = null;\n      if (shouldTimeProgram) {\n        this.uploadWaitMs += util.now() - start;\n      }\n    } else {\n      const newTexture = this.acquireTexture(texShape, usage, dtype, isPacked);\n      texData.texture = newTexture;\n    }\n  }\n\n  private convertAndCacheOnCPU(dataId: DataId, float32Values?: Float32Array):\n      TypedArray {\n    const texData = this.texData.get(dataId);\n    const {dtype} = texData;\n\n    this.releaseGPUData(dataId);\n\n    if (float32Values != null) {\n      texData.values = float32ToTypedArray(float32Values, dtype as 'float32');\n    }\n    return texData.values as TypedArray;\n  }\n\n  private acquireTexture(\n      texShape: [number, number], texType: TextureUsage, dtype: DataType,\n      isPacked: boolean): WebGLTexture {\n    this.numBytesInGPU += this.computeBytes(texShape, dtype);\n    if (!this.warnedAboutMemory &&\n        this.numBytesInGPU > this.numMBBeforeWarning * 1024 * 1024) {\n      const mb = (this.numBytesInGPU / 1024 / 1024).toFixed(2);\n      this.warnedAboutMemory = true;\n      console.warn(\n          `High memory usage in GPU: ${mb} MB, ` +\n          `most likely due to a memory leak`);\n    }\n    return this.textureManager.acquireTexture(texShape, texType, isPacked);\n  }\n\n  private computeBytes(shape: [number, number], dtype: DataType) {\n    return shape[0] * shape[1] * util.bytesPerElement(dtype);\n  }\n\n  private tryRunOnCpuOrThrow<T extends Tensor>(\n      inputs: TensorInfo[], fn: () => T): T|null {\n    if (this.shouldExecuteOnCPU(inputs)) {\n      try {\n        return fn();\n      } catch (e) {\n        if (env().getBool('IS_TEST')) {\n          throw new Error('CPU forwarding failed');\n        }\n      }\n    }\n    return null;\n  }\n}\n\nfunction float32ToTypedArray<D extends NumericDataType>(\n    a: Float32Array, dtype: D): tf.DataTypeMap[D] {\n  if (dtype === 'float32' || dtype === 'complex64') {\n    return a as tf.DataTypeMap[D];\n  } else if (dtype === 'int32' || dtype === 'bool') {\n    const result = (dtype === 'int32') ? new Int32Array(a.length) :\n                                         new Uint8Array(a.length);\n    for (let i = 0; i < result.length; ++i) {\n      result[i] = Math.round(a[i]);\n    }\n    return result as tf.DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport * as gpgpu_util from './gpgpu_util';\nimport * as webgl_util from './webgl_util';\n\nexport {MathBackendWebGL, WebGLMemoryInfo, WebGLTimingInfo} from './backend_webgl';\nexport {setWebGLContext} from './canvas_util';\nexport {GPGPUContext} from './gpgpu_context';\nexport {GPGPUProgram} from './gpgpu_math';\n// WebGL specific utils.\nexport {gpgpu_util, webgl_util};\n\n/**\n * Enforce use of half precision textures if available on the platform.\n *\n * @doc {heading: 'Environment', namespace: 'webgl'}\n */\nexport function forceHalfFloat(): void {\n  env().set('WEBGL_FORCE_F16_TEXTURES', true);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// base.ts is the webgl backend without auto kernel registration.\n\nimport {device_util, registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from './backend_webgl';\nexport {version as version_webgl} from './version';\n\nif (device_util.isBrowser()) {\n  registerBackend('webgl', () => new MathBackendWebGL(), 2 /* priority */);\n}\n\n// Export webgl utilities\nexport * from './webgl';\n\n// Export forceHalfFlost under webgl namespace for the union bundle.\nimport {forceHalfFloat} from './webgl';\nexport const webgl = {forceHalfFloat};\n", "import {BinaryInputs, DataType, env, KernelFunc, UnaryInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BinaryOpProgram} from '../binaryop_gpu';\nimport {BinaryOpPackedProgram} from '../binaryop_packed_gpu';\nimport {UnaryOpProgram} from '../unaryop_gpu';\n\nexport const CHECK_NAN_SNIPPET_UNARY = `if (isnan(x)) return x;`;\n\nexport const CHECK_NAN_SNIPPET_BINARY = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const CHECK_NAN_SNIPPET_BINARY_PACKED = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param opSnippets Op snippet to create `UnaryOpProgram`.\n */\nexport function unaryKernelFunc(opSnippet: string): KernelFunc {\n  return ({inputs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = new UnaryOpProgram(x.shape, opSnippet);\n    return webglBackend.runWebGLProgram(program, [x], x.dtype);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param opSnippet Op snippet to create `BinaryOpProgram`.\n * @param packedOpSnippet Op snippet to create `BinaryOpPackedProgram`.\n * @param checkOutOfBoundsForPackedProgram Whether to set checkOutOfBounds=true\n *     when creating BinaryOpPackedProgram.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    opSnippet: string, packedOpSnippet: string,\n    checkOutOfBoundsForPackedProgram?: boolean, dtype?: DataType): KernelFunc {\n  // TODO(jingjin): handle complex64.\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            packedOpSnippet, a.shape, b.shape,\n            !!checkOutOfBoundsForPackedProgram) :\n        new BinaryOpProgram(opSnippet, a.shape, b.shape);\n    const $dtype = dtype || a.dtype;\n    const output = webglBackend.runWebGLProgram(program, [a, b], $dtype);\n    return output;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc, CHECK_NAN_SNIPPET_BINARY, CHECK_NAN_SNIPPET_BINARY_PACKED} from '../kernel_utils/kernel_funcs_utils';\n\nconst ATAN2 = CHECK_NAN_SNIPPET_BINARY + `\n  return atan(a, b);\n`;\n\nconst ATAN2_PACKED = `\n  vec4 result = atan(a, b);\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET_BINARY_PACKED + `\n  return result;\n`;\n\nexport const atan2KernelFunc = binaryKernelFunc(ATAN2, ATAN2_PACKED);\n\nexport const atan2Config: KernelConfig = {\n  kernelName: Atan2,\n  backendName: 'webgl',\n  kernelFunc: atan2KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendWebGL}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'webgl',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function avgPool(args: {\n  inputs: AvgPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const avgPoolProgram = new Pool2DProgram(convInfo, 'avg', false);\n  return backend.runWebGLProgram(avgPoolProgram, [x], 'float32');\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'webgl',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {AvgPool2DBackpropProgram} from '../avg_pool_backprop_gpu';\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const avgPoolBackpropProgram = new AvgPool2DBackpropProgram(convInfo);\n  return backend.runWebGLProgram(avgPoolBackpropProgram, [dy], x.dtype);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = '0.0';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = '1.0';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        float x = getXAtOutCoords();\n        float mean = getMeanAtOutCoords();\n        float variance = getVarianceAtOutCoords();\n        float offset = ${offsetSnippet};\n        float scale = ${scaleSnippet};\n        float inv = scale * inversesqrt(variance + float(${varianceEpsilon}));\n        setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = 'vec4(0.0)';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = 'vec4(1.0)';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        vec4 offset = ${offsetSnippet};\n        vec4 scale = ${scaleSnippet};\n\n        vec4 x = getXAtOutCoords();\n        vec4 mean = getMeanAtOutCoords();\n        vec4 variance = getVarianceAtOutCoords();\n\n        vec4 inv = scale * inversesqrt(variance + vec4(${varianceEpsilon}));\n\n        setOutput((x - mean) * inv + offset);\n      }\n    `;\n  }\n}\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BatchNormProgram} from '../batchnorm_gpu';\nimport {BatchNormPackedProgram} from '../batchnorm_packed_gpu';\n\nexport const batchNormKernelFunc: (params: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendWebGL,\n  attrs: FusedBatchNormAttrs\n}) => TensorInfo = ({inputs, backend, attrs}) => {\n  const {x, mean, variance, offset, scale} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const finalInputs = [x, mean, variance];\n\n  let offsetShape = null;\n  if (offset != null) {\n    offsetShape = offset.shape;\n    finalInputs.push(offset);\n  }\n\n  let scaleShape = null;\n  if (scale != null) {\n    scaleShape = scale.shape;\n    finalInputs.push(scale);\n  }\n\n  const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n      new BatchNormPackedProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon) :\n      new BatchNormProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon);\n  const output =\n      backend.runWebGLProgram(program, finalInputs, finalInputs[0].dtype);\n\n  return output;\n};\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'webgl',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst COS = CHECK_NAN_SNIPPET_UNARY + `\n  return cos(x);\n`;\n\nexport const cosKernelFunc = unaryKernelFunc(COS);\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'webgl',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\n// Without the equality check div produces 0.9999 for a = b, which when\n// floored can cause errors.\nconst DIV = `\nif (a == b) {\n  return 1.0;\n};\nreturn a / b;`;\n\n// We do the same as in ./binaryop_gpu, with vec4 and ivec4.\n// On Linux, the vectorized implementation produces NaNs when a and b are 0.\nconst DIV_PACKED = `\n  // vec4 one = vec4(equal(a, b));\n  // return one + (vec4(1.0) - one) * a / b;\n  vec4 result = a / b;\n  if(a.x == b.x) {\n    result.x = 1.;\n  }\n  if(a.y == b.y) {\n    result.y = 1.;\n  }\n  if(a.z == b.z) {\n    result.z = 1.;\n  }\n  if(a.w == b.w) {\n    result.w = 1.;\n  }\n\n  return result;\n`;\n\nexport const divKernelFunc = binaryKernelFunc(\n    DIV, DIV_PACKED, true /* checkOutOfBoundsForPackedProgram */);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'webgl',\n  kernelFunc: divKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FlipLeftRightProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(imageShape: [number, number, number, number]) {\n    const imageWidth = imageShape[2];\n    this.outputShape = imageShape;\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n\n          int coordX = ${imageWidth} - x;\n          float outputValue;\n          if(coordX >= 0 && coordX < ${imageWidth}) {\n            outputValue = getImage(coords[0], coords[1], coordX, coords[3]);\n          } else {\n            outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {FlipLeftRightProgram} from '../flip_left_right_gpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new FlipLeftRightProgram((image as Tensor4D).shape);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n        vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${width}.0, ${height}.0);\n\n        vec4 values = ${glsl.texture2D}(A, uv);\n        float value;\n        if (depth == 0) {\n          value = values.r;\n        } else if (depth == 1) {\n          value = values.g;\n        } else if (depth == 2) {\n          value = values.b;\n        } else if (depth == 3) {\n          value = values.a;\n        }\n\n        setOutput(floor(value * 255.0 + 0.5));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n\n        vec4 result = vec4(0.);\n\n        for(int row=0; row<=1; row++) {\n          for(int col=0; col<=1; col++) {\n            texC = coords[1] + row;\n            depth = coords[2] + col;\n\n            vec2 uv = (vec2(texC, texR) + halfCR) /\n                       vec2(${width}.0, ${height}.0);\n            vec4 values = ${glsl.texture2D}(A, uv);\n            float value;\n            if (depth == 0) {\n              value = values.r;\n            } else if (depth == 1) {\n              value = values.g;\n            } else if (depth == 2) {\n              value = values.b;\n            } else if (depth == 3) {\n              value = values.a;\n            }\n\n            result[row * 2 + col] = floor(value * 255.0 + 0.5);\n          }\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {TextureUsage} from '../tex_util';\n\nimport {FromPixelsProgram} from './FromPixels_utils/from_pixels_gpu';\nimport {FromPixelsPackedProgram} from './FromPixels_utils/from_pixels_packed_gpu';\n\nexport const fromPixelsConfig: KernelConfig = {\n  kernelName: FromPixels,\n  backendName: 'webgl',\n  kernelFunc: fromPixels as {} as KernelFunc,\n};\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\nfunction fromPixels(args: {\n  inputs: FromPixelsInputs,\n  backend: MathBackendWebGL,\n  attrs: FromPixelsAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  let {pixels} = inputs;\n  const {numChannels} = attrs;\n\n  const isVideo = typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement;\n  const isImage = typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement;\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n\n  const texShape: [number, number] = [height, width];\n  const outShape = [height, width, numChannels];\n\n  if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement | HTMLImageElement, 0, 0, width, height);\n    pixels = fromPixels2DContext.canvas;\n  }\n\n  const tempPixelHandle = backend.makeTensorInfo(texShape, 'int32');\n  // This is a byte texture with pixels.\n  backend.texData.get(tempPixelHandle.dataId).usage = TextureUsage.PIXELS;\n  backend.gpgpu.uploadPixelDataToTexture(\n      backend.getTexture(tempPixelHandle.dataId), pixels as ImageData);\n  const program = env().getBool('WEBGL_PACK') ?\n      new FromPixelsPackedProgram(outShape) :\n      new FromPixelsProgram(outShape);\n  const res = backend.runWebGLProgram(program, [tempPixelHandle], 'int32');\n  backend.disposeData(tempPixelHandle.dataId);\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReduceProgram} from '../reduce_gpu';\n\ntype ReduceTypes = 'all'|'any'|'max'|'min'|'sum'|'prod';\n\n// Returns an array of configuration objects that describe each stage of the\n// reduction.\nfunction getReductionStages(inShape: number[]):\n    Array<{inSize: number, windowSize: number, outSize: number}> {\n  const stages = [];\n\n  while (stages.length === 0 || stages[stages.length - 1].outSize !== 1) {\n    const outSize: number =\n        stages.length ? stages[stages.length - 1].outSize : inShape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(outSize);\n    stages.push({\n      inSize: outSize,\n      windowSize,\n      outSize: Math.ceil(outSize / windowSize)\n    });\n  }\n\n  return stages;\n}\n\nexport function reduce(\n    x: TensorInfo, dtype: DataType, reductionType: ReduceTypes,\n    backend: MathBackendWebGL): TensorInfo {\n  const reductionStages = getReductionStages(x.shape);\n\n  let result = x;\n  for (let i = 0; i < reductionStages.length; i++) {\n    const {inSize, windowSize, outSize} = reductionStages[i];\n\n    const program = new ReduceProgram(\n        {windowSize, inSize, batchSize: x.shape[0], outSize}, reductionType);\n    const previousResult = result;\n    result = backend.runWebGLProgram(program, [result], dtype);\n\n    if (previousResult.dataId !== x.dataId) {\n      backend.disposeData(previousResult.dataId);\n    }\n  }\n\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReshapePackedProgram} from '../reshape_packed_gpu';\nimport {getBatchDim, getRowsCols} from '../webgl_util';\n\nexport function packedReshape(\n    input: TensorInfo, afterShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const input3DShape =\n      [getBatchDim(input.shape),\n       ...getRowsCols(input.shape)] as [number, number, number];\n  const input3D: TensorInfo = {\n    dtype: input.dtype,\n    shape: input3DShape,\n    dataId: input.dataId\n  };\n  const afterShapeAs3D =\n      [getBatchDim(afterShape),\n       ...getRowsCols(afterShape)] as [number, number, number];\n\n  const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n  const preventEagerUnpackingOfOutput = true;\n  const output = backend.runWebGLProgram(\n      program, [input3D], input.dtype, null /* customSetup */,\n      preventEagerUnpackingOfOutput);\n  return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {packedReshape} from '../kernel_utils/reshape';\nimport {isReshapeFree} from '../webgl_util';\n\nexport function reshape(args: {\n  inputs: ReshapeInputs,\n  backend: MathBackendWebGL,\n  attrs: ReshapeAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n  const webglBackend = backend;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  const xTexData = webglBackend.texData.get(x.dataId);\n  if (xTexData.isPacked && !isReshapeFree(x.shape, $shape) &&\n      !(xTexData.texture !== null && isReshapeFree(xTexData.shape, $shape))) {\n    return packedReshape(x, $shape, webglBackend);\n  }\n\n  webglBackend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'webgl',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {reduce} from '../kernel_utils/reduce';\nimport {reshape} from '../kernels/Reshape';\n\nexport function maxImpl(\n    x: TensorInfo, reduceShape: number[], outShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const inSize = util.sizeFromShape(reduceShape);\n  const xSize = util.sizeFromShape(x.shape);\n  const batchSize = xSize / inSize;\n  const reshapedInput =\n      reshape({inputs: {x}, attrs: {shape: [batchSize, inSize]}, backend});\n\n  const reduced = reduce(reshapedInput, x.dtype, 'max', backend);\n  const reshapedOutput =\n      reshape({inputs: {x: reduced}, attrs: {shape: outShape}, backend});\n\n  backend.disposeIntermediateTensorInfo(reshapedInput);\n  backend.disposeIntermediateTensorInfo(reduced);\n\n  return reshapedOutput;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposeProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const switched = getSwitchedCoords(newDim);\n\n    this.userCode = `\n    void main() {\n      ${dtype} resRC = getOutputCoords();\n      setOutput(getA(${switched}));\n    }\n    `;\n  }\n}\n\nfunction getSwitchedCoords(newDim: number[]): string {\n  const rank = newDim.length;\n  if (rank > 6) {\n    throw Error(`Transpose for rank ${rank} is not yet supported`);\n  }\n  const originalOrder =\n      ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u', 'resRC.v'];\n  const switchedCoords = new Array(rank);\n  for (let i = 0; i < newDim.length; i++) {\n    switchedCoords[newDim[i]] = originalOrder[i];\n  }\n  return switchedCoords.join();\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getVecChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    if (this.rank > 6) {\n      throw Error(\n          `Packed transpose for rank ${this.rank} is not yet supported.`);\n    }\n    const dtype = getCoordsDataType(this.rank);\n\n    const outputOrder = getVecChannels('rc', this.rank);\n    const switchedOrder = new Array(this.rank);\n    for (let i = 0; i < newDim.length; i++) {\n      switchedOrder[newDim[i]] = outputOrder[i];\n    }\n    const innerDims = `vec2(${switchedOrder.slice(-2).join()})`;\n    const nextColumn =\n        `++${outputOrder[this.rank - 1]} < ${outputShape[this.rank - 1]}`;\n    const getc = `getChannel(getA(${switchedOrder.join()}), ${innerDims})`;\n\n    this.userCode = `\n    void main() {\n      ${dtype} rc = getOutputCoords();\n      vec4 result = vec4(0.);\n      result[0] = ${getc};\n      if(${nextColumn}) {\n        result[1] = ${getc};\n      }\n      --${outputOrder[this.rank - 1]};\n      if(++${outputOrder[this.rank - 2]} < ${outputShape[this.rank - 2]}) {\n        result[2] = ${getc};\n        if(${nextColumn}) {\n          result[3] = ${getc};\n        }\n      }\n      setOutput(result);\n    }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {transposeImplCPU} from '../kernel_utils/shared';\nimport {TransposeProgram} from '../transpose_gpu';\nimport {TransposePackedProgram} from '../transpose_packed_gpu';\n\nexport function transposeImpl(\n    x: TensorInfo, perm: number[], backend: MathBackendWebGL): TensorInfo {\n  const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n      new TransposePackedProgram(x.shape, perm) :\n      new TransposeProgram(x.shape, perm);\n  return backend.runWebGLProgram(program, [x], x.dtype);\n}\n\nexport {transposeImplCPU};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {maxImplCPU} from '../kernel_utils/shared';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl, transposeImplCPU} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    const maxInputIsTransposed = permutedAxes != null;\n    const shouldExecuteOnCPU = webglBackend.shouldExecuteOnCPU([x]);\n\n    let maxInput = x;\n    if (maxInputIsTransposed) {\n      if (shouldExecuteOnCPU) {\n        const xTexData = webglBackend.texData.get(maxInput.dataId);\n        const values = xTexData.values as TypedArray;\n\n        const newShape: number[] = new Array(xRank);\n        for (let i = 0; i < newShape.length; i++) {\n          newShape[i] = x.shape[permutedAxes[i]];\n        }\n        const maxInputValues =\n            transposeImplCPU(values, x.shape, x.dtype, permutedAxes, newShape);\n\n        maxInput = webglBackend.makeTensorInfo(newShape, x.dtype);\n        const maxInputData = webglBackend.texData.get(maxInput.dataId);\n        maxInputData.values = maxInputValues;\n      } else {\n        maxInput = transposeImpl(x, permutedAxes, webglBackend);\n      }\n\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n    }\n\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(maxInput.shape, axes);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // rather than reshape at the end, set the target shape here.\n      outShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n    }\n\n    let out;\n    if (shouldExecuteOnCPU) {\n      const xTexData = webglBackend.texData.get(maxInput.dataId);\n      const values = xTexData.values as TypedArray;\n\n      const outValues = maxImplCPU(\n          values, util.sizeFromShape(reduceShape), outShape, x.dtype);\n\n      out = webglBackend.makeTensorInfo(outShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = maxImpl(maxInput, reduceShape, outShape, webglBackend);\n    }\n\n    if (maxInputIsTransposed) {\n      webglBackend.disposeIntermediateTensorInfo(maxInput);\n    }\n\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function maxPool(args: {\n  inputs: MaxPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const maxPoolProgram = new Pool2DProgram(convInfo, 'max', false);\n  return backend.runWebGLProgram(maxPoolProgram, [x], x.dtype);\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'webgl',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {MaxPool2DBackpropProgram} from '../max_pool_backprop_gpu';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const getPositions = true;\n  const maxPoolPositionsProgram =\n      new Pool2DProgram(convInfo, 'max', getPositions);\n  const maxPoolPositions: TensorInfo =\n      backend.runWebGLProgram(maxPoolPositionsProgram, [x], x.dtype);\n\n  const maxPoolBackPropProgram = new MaxPool2DBackpropProgram(convInfo);\n  const result = backend.runWebGLProgram(\n      maxPoolBackPropProgram, [dy, maxPoolPositions], x.dtype);\n  backend.disposeIntermediateTensorInfo(maxPoolPositions);\n  return result;\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\n\nexport function maxPoolWithArgmaxImpl(\n    x: TensorInfo, includeBatchInIndex: boolean,\n    convInfo: backend_util.Conv2DInfo,\n    backend: MathBackendWebGL): TensorInfo[] {\n  let program = new Pool2DProgram(convInfo, 'max', false);\n  const poolOutput = backend.runWebGLProgram(program, [x], 'float32');\n\n  program = new Pool2DProgram(convInfo, 'max', true, true, includeBatchInIndex);\n  const indexOutput = backend.runWebGLProgram(program, [x], 'float32');\n  return [poolOutput, indexOutput];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    util.assert(\n        x.shape.length === 4,\n        () => `Error in maxPool: input must be rank 4 but got rank ${\n            x.shape.length}.`);\n    const dilations: [number, number] = [1, 1];\n    util.assert(\n        backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n            `Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        dilations, pad);\n\n    const [result, indexes] =\n        maxPoolWithArgmaxImpl(x, includeBatchInIndex, convInfo, webglBackend);\n    return [result, indexes];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV3, NonMaxSuppressionV3Attrs, NonMaxSuppressionV3Inputs, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV3Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV3,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV3Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold} =\n        attrs as unknown as NonMaxSuppressionV3Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n\n    return kernel_impls.nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal);\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs, TypedArray} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class RotateProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], radians: number,\n      fillValue: number|[number, number, number],\n      center: number|[number, number]) {\n    const imageHeight = imageShape[1];\n    const imageWidth = imageShape[2];\n    const sinFactor = Math.sin(radians).toFixed(3);\n    const cosFactor = Math.cos(radians).toFixed(3);\n    this.outputShape = imageShape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const centerXString = centerX.toFixed(3);\n    const centerYString = centerY.toFixed(3);\n\n    let fillSnippet = '';\n    if (typeof fillValue === 'number') {\n      fillSnippet = `float outputValue = ${fillValue.toFixed(2)};`;\n    } else {\n      fillSnippet = `\n        vec3 fill = vec3(${fillValue.join(',')});\n        float outputValue = fill[coords[3]];`;\n    }\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n          int y = coords[1];\n          float coordXFloat = (float(x) - ${centerXString}) * ${\n        cosFactor} - (float(y) - ${centerYString}) * ${sinFactor};\n          float coordYFloat = (float(x) - ${centerXString}) * ${\n        sinFactor} + (float(y) - ${centerYString}) * ${cosFactor};\n          int coordX = int(round(coordXFloat + ${centerXString}));\n          int coordY = int(round(coordYFloat + ${centerYString}));\n          ${fillSnippet}\n          if(coordX >= 0 && coordX < ${imageWidth} && coordY >= 0 && coordY < ${\n        imageHeight}) {\n            outputValue = getImage(coords[0], coordY, coordX, coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {RotateProgram} from '../rotate_gpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new RotateProgram(\n        (image as Tensor4D).shape, radians, fillValue, center);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SIN = CHECK_NAN_SNIPPET_UNARY + `\n  return sin(x);\n`;\n\nexport const sinKernelFunc = unaryKernelFunc(SIN);\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'webgl',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Square} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARE = `return x * x;`;\n\nexport const squareKernelFunc = unaryKernelFunc(SQUARE);\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'webgl',\n  kernelFunc: squareKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const squaredDifferenceKernelFunc =\n    binaryKernelFunc(SQUARED_DIFFERENCE, SQUARED_DIFFERENCE);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'webgl',\n  kernelFunc: squaredDifferenceKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst TAN = `return tan(x);`;\n\nexport const tanKernelFunc = unaryKernelFunc(TAN);\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'webgl',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {transposeImpl} from './Transpose_impl';\nimport {transposeImplCPU as cpuTranspose} from './Transpose_impl';\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as TransposeInputs;\n    const {perm} = attrs as {} as TransposeAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const newShape: number[] = new Array(xRank);\n    for (let i = 0; i < newShape.length; i++) {\n      newShape[i] = x.shape[perm[i]];\n    }\n\n    let out: TensorInfo;\n    if (webglBackend.shouldExecuteOnCPU([x])) {\n      const xTexData = webglBackend.texData.get(x.dataId);\n      const values = xTexData.values as TypedArray;\n      const outValues = cpuTranspose(values, x.shape, x.dtype, perm, newShape);\n\n      out = webglBackend.makeTensorInfo(newShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = transposeImpl(x, perm, webglBackend);\n    }\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {uniqueImplCPU} from '../kernel_utils/shared';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function unique(\n    args:\n        {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendWebGL}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  // For now, always forward calculation to the CPU backend.\n  console.warn(\n      'WARNING: ',\n      'UI might be locked temporarily as data is being downloaded');\n  const values = backend.readSync(x.dataId);\n  const {outputValues, outputShape, indices} =\n      uniqueImplCPU(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'webgl',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {atan2Config} from './kernels/Atan2';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {cosConfig} from './kernels/Cos';\nimport {divConfig} from './kernels/Div';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {fromPixelsConfig} from './kernels/FromPixels';\nimport {identityConfig} from './kernels/Identity';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {nonMaxSuppressionV3Config} from './kernels/NonMaxSuppressionV3';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {sinConfig} from './kernels/Sin';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {tanConfig} from './kernels/Tan';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  atan2Config,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  cosConfig,\n  divConfig,\n  flipLeftRightConfig,\n  fromPixelsConfig,\n  identityConfig,\n  maxConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  nonMaxSuppressionV3Config,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  sinConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  tanConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport * from '@tensorflow/tfjs-core';\nexport * from '@tensorflow/tfjs-layers';\nexport * from '@tensorflow/tfjs-converter';\n\n// Export data api as tf.data\nimport * as data from '@tensorflow/tfjs-data';\nexport {data};\n\n// Import and register backends.\nimport '@tensorflow/tfjs-backend-cpu';\nimport '@tensorflow/tfjs-backend-webgl';\n\n// Import versions of all sub-packages.\nimport {version_core} from '@tensorflow/tfjs-core';\nimport {version_cpu} from '@tensorflow/tfjs-backend-cpu';\nimport {version_webgl} from '@tensorflow/tfjs-backend-webgl';\nimport {version_data} from '@tensorflow/tfjs-data';\nimport {version_layers} from '@tensorflow/tfjs-layers';\nimport {version_converter} from '@tensorflow/tfjs-converter';\nimport {version as version_union} from './version';\n\nexport const version = {\n  'tfjs-core': version_core,\n  'tfjs-backend-cpu': version_cpu,\n  'tfjs-backend-webgl': version_webgl,\n  'tfjs-data': version_data,\n  'tfjs-layers': version_layers,\n  'tfjs-converter': version_converter,\n  'tfjs': version_union\n};\n", "const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.age.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    obj.gender = Math.trunc(100 * data[0]) < 50 ? 'female' : 'male';\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "export default {\n  backend: 'webgl',\n  console: true,\n  face: {\n    enabled: true, // refers to detector, but since all other face modules rely on detector, it should be a global\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be blazeface-front or blazeface-back\n      inputSize: 256, // fixed value: 128 for front and tfhub and 256 for back\n      maxFaces: 10, // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10, // how many frames to go without running the bounding box detector\n      minConfidence: 0.5, // threshold for discarding a prediction\n      iouThreshold: 0.3, // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7, // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192, // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3, // empiric tuning\n      inputSize: 64, // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json',\n      inputSize: 64, // fixed value\n      skipFrames: 10,\n    },\n    gender: {\n      enabled: true,\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64, // fixed value\n      minConfidence: 0.5,\n      skipFrames: 10,\n      useGrayscale: true,\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257, // fixed value\n    outputStride: 16, // fixed value\n    maxDetections: 5,\n    scoreThreshold: 0.7,\n    nmsRadius: 20,\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256, // fixed value\n    skipFrames: 10,\n    minConfidence: 0.5,\n    iouThreshold: 0.3,\n    scoreThreshold: 0.7,\n    enlargeFactor: 1.65, // empiric tuning\n    maxHands: 2,\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    let savedWebglPackDepthwiseConvFlag;\n    if (tf.getBackend() === 'webgl') {\n      savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n      tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
-  "mappings": "ucAAA,gHAAmB,sBACnB,GAAiB,oBACjB,GAAgB,mBAChB,GAAkB,qBAClB,GAAiB,oBAKX,GAAW,WAAO,SAElB,GAAS,OAAO,UAChB,GAAO,OAAO,QAEpB,SACC,cACC,KAAK,IAAQ,GAEb,KAAM,GAAY,UAAU,GACtB,EAAU,UAAU,GAEpB,EAAU,GAChB,GAAI,GAAO,EAEX,GAAI,GACH,KAAM,GAAI,EACJ,EAAS,OAAO,EAAE,QACxB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAC3B,KAAM,GAAU,EAAE,GAClB,GAAI,GACJ,AAAI,YAAmB,QACtB,EAAS,EACH,AAAI,YAAY,OAAO,GAC7B,EAAS,OAAO,KAAK,EAAQ,OAAQ,EAAQ,WAAY,EAAQ,YAC3D,AAAI,YAAmB,aAC7B,EAAS,OAAO,KAAK,GACf,AAAI,YAAmB,IAC7B,EAAS,EAAQ,IAEjB,EAAS,OAAO,KAAK,MAAO,IAAY,SAAW,EAAU,OAAO,IAErE,GAAQ,EAAO,OACf,EAAQ,KAAK,IAIf,KAAK,IAAU,OAAO,OAAO,GAE7B,GAAI,GAAO,GAAW,EAAQ,OAAS,QAAa,OAAO,EAAQ,MAAM,cACzE,AAAI,GAAQ,CAAC,mBAAmB,KAAK,IACpC,MAAK,IAAQ,MAGX,QACH,MAAO,MAAK,IAAQ,UAEjB,QACH,MAAO,MAAK,IAEb,OACC,MAAO,SAAQ,QAAQ,KAAK,IAAQ,YAErC,cACC,KAAM,GAAM,KAAK,IACX,EAAK,EAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YACjE,MAAO,SAAQ,QAAQ,GAExB,SACC,KAAM,GAAW,GAAI,IACrB,SAAS,MAAQ,aACjB,EAAS,KAAK,KAAK,KACnB,EAAS,KAAK,MACP,EAER,WACC,MAAO,gBAER,QACC,KAAM,GAAO,KAAK,KAEZ,EAAQ,UAAU,GAClB,EAAM,UAAU,GACtB,GAAI,GAAe,EACnB,AAAI,IAAU,OACb,EAAgB,EACV,AAAI,EAAQ,EAClB,EAAgB,KAAK,IAAI,EAAO,EAAO,GAEvC,EAAgB,KAAK,IAAI,EAAO,GAEjC,AAAI,IAAQ,OACX,EAAc,EACR,AAAI,EAAM,EAChB,EAAc,KAAK,IAAI,EAAO,EAAK,GAEnC,EAAc,KAAK,IAAI,EAAK,GAE7B,KAAM,GAAO,KAAK,IAAI,EAAc,EAAe,GAE7C,EAAS,KAAK,IACd,EAAe,EAAO,MAAM,EAAe,EAAgB,GAC3D,EAAO,GAAI,IAAK,GAAI,CAAE,KAAM,UAAU,KAC5C,SAAK,IAAU,EACR,GAIT,OAAO,iBAAiB,GAAK,UAAW,CACvC,KAAM,CAAE,WAAY,IACpB,KAAM,CAAE,WAAY,IACpB,MAAO,CAAE,WAAY,MAGtB,OAAO,eAAe,GAAK,UAAW,OAAO,YAAa,CACzD,MAAO,OACP,SAAU,GACV,WAAY,GACZ,aAAc,KAiBf,YAAoB,EAAS,EAAM,GACjC,MAAM,KAAK,KAAM,GAEjB,KAAK,QAAU,EACf,KAAK,KAAO,EAGZ,AAAI,GACF,MAAK,KAAO,KAAK,MAAQ,EAAY,MAIvC,MAAM,kBAAkB,KAAM,KAAK,aAGrC,GAAW,UAAY,OAAO,OAAO,MAAM,WAC3C,GAAW,UAAU,YAAc,GACnC,GAAW,UAAU,KAAO,aAE5B,GAAI,IACJ,IACC,GAAU,oBAAoB,cACtB,IAET,KAAM,IAAY,OAAO,kBAGnB,GAAc,WAAO,YAW3B,YAAc,GACb,GAAI,GAAQ,KAER,EAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,GAC3E,EAAY,EAAK,KAErB,GAAI,GAAO,IAAc,OAAY,EAAI,EACzC,GAAI,GAAe,EAAK,QACxB,GAAI,GAAU,IAAiB,OAAY,EAAI,EAE/C,AAAI,GAAQ,KAEX,EAAO,KACD,AAAI,GAAkB,GAE5B,EAAO,OAAO,KAAK,EAAK,YAClB,AAAI,GAAO,IAAc,CAAI,OAAO,SAAS,IAAc,CAAI,OAAO,UAAU,SAAS,KAAK,KAAU,uBAE9G,EAAO,OAAO,KAAK,GACb,AAAI,YAAY,OAAO,GAE7B,EAAO,OAAO,KAAK,EAAK,OAAQ,EAAK,WAAY,EAAK,YAChD,AAAI,YAAgB,aAG1B,GAAO,OAAO,KAAK,OAAO,OAE3B,KAAK,IAAa,CACjB,OACA,UAAW,GACX,MAAO,MAER,KAAK,KAAO,EACZ,KAAK,QAAU,EAEf,AAAI,YAAgB,aACnB,EAAK,GAAG,QAAS,SAAU,GAC1B,KAAM,GAAQ,EAAI,OAAS,aAAe,EAAM,GAAI,IAAW,+CAA+C,EAAM,QAAQ,EAAI,UAAW,SAAU,GACrJ,EAAM,IAAW,MAAQ,IAK5B,GAAK,UAAY,IACZ,QACH,MAAO,MAAK,IAAW,SAGpB,YACH,MAAO,MAAK,IAAW,WAQxB,cACC,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,eAS/D,OACC,GAAI,GAAK,KAAK,SAAW,KAAK,QAAQ,IAAI,iBAAmB,GAC7D,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,QAAO,OAEd,GAAI,IAAK,GAAI,CACZ,KAAM,EAAG,gBACN,EACF,IAAS,OAUb,OACC,GAAI,GAAS,KAEb,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,IACC,MAAO,MAAK,MAAM,EAAO,kBACjB,GACR,MAAO,IAAK,QAAQ,OAAO,GAAI,IAAW,iCAAiC,EAAO,eAAe,EAAI,UAAW,qBAUnH,OACC,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,GAAO,cAShB,SACC,MAAO,IAAY,KAAK,OASzB,gBACC,GAAI,GAAS,KAEb,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,IAAY,EAAQ,EAAO,aAMrC,OAAO,iBAAiB,GAAK,UAAW,CACvC,KAAM,CAAE,WAAY,IACpB,SAAU,CAAE,WAAY,IACxB,YAAa,CAAE,WAAY,IAC3B,KAAM,CAAE,WAAY,IACpB,KAAM,CAAE,WAAY,IACpB,KAAM,CAAE,WAAY,MAGrB,GAAK,MAAQ,SAAU,GACtB,SAAW,KAAQ,QAAO,oBAAoB,GAAK,WAElD,GAAI,CAAE,KAAQ,KACb,KAAM,GAAO,OAAO,yBAAyB,GAAK,UAAW,GAC7D,OAAO,eAAe,EAAO,EAAM,KAYtC,cACC,GAAI,GAAS,KAEb,GAAI,KAAK,IAAW,UACnB,MAAO,IAAK,QAAQ,OAAO,GAAI,WAAU,0BAA0B,KAAK,QAKzE,GAFA,KAAK,IAAW,UAAY,GAExB,KAAK,IAAW,MACnB,MAAO,IAAK,QAAQ,OAAO,KAAK,IAAW,OAG5C,GAAI,GAAO,KAAK,KAGhB,GAAI,IAAS,KACZ,MAAO,IAAK,QAAQ,QAAQ,OAAO,MAAM,IAS1C,GALA,AAAI,GAAO,IACV,GAAO,EAAK,UAIT,OAAO,SAAS,GACnB,MAAO,IAAK,QAAQ,QAAQ,GAI7B,GAAI,CAAE,aAAgB,aACrB,MAAO,IAAK,QAAQ,QAAQ,OAAO,MAAM,IAK1C,GAAI,GAAQ,GACR,EAAa,EACb,EAAQ,GAEZ,MAAO,IAAI,IAAK,QAAQ,SAAU,EAAS,GAC1C,GAAI,GAGJ,AAAI,EAAO,SACV,GAAa,WAAW,WACvB,EAAQ,GACR,EAAO,GAAI,IAAW,0CAA0C,EAAO,aAAa,EAAO,aAAc,kBACvG,EAAO,UAIX,EAAK,GAAG,QAAS,SAAU,GAC1B,AAAI,EAAI,OAAS,aAEhB,GAAQ,GACR,EAAO,IAGP,EAAO,GAAI,IAAW,+CAA+C,EAAO,QAAQ,EAAI,UAAW,SAAU,MAI/G,EAAK,GAAG,OAAQ,SAAU,GACzB,GAAI,GAAS,IAAU,KACtB,OAGD,GAAI,EAAO,MAAQ,EAAa,EAAM,OAAS,EAAO,MACrD,EAAQ,GACR,EAAO,GAAI,IAAW,mBAAmB,EAAO,mBAAmB,EAAO,OAAQ,aAClF,OAGD,GAAc,EAAM,OACpB,EAAM,KAAK,KAGZ,EAAK,GAAG,MAAO,WACd,GAAI,EACH,OAGD,aAAa,GAEb,IACC,EAAQ,OAAO,OAAO,EAAO,UACrB,GAER,EAAO,GAAI,IAAW,kDAAkD,EAAO,QAAQ,EAAI,UAAW,SAAU,SAcpH,YAAqB,EAAQ,GAC5B,GAAI,MAAO,KAAY,WACtB,KAAM,IAAI,OAAM,gFAGjB,KAAM,GAAK,EAAQ,IAAI,gBACvB,GAAI,GAAU,QACV,EAAK,EAGT,MAAI,IACH,GAAM,mBAAmB,KAAK,IAI/B,EAAM,EAAO,MAAM,EAAG,MAAM,WAG5B,AAAI,CAAC,GAAO,GACX,GAAM,iCAAiC,KAAK,IAI7C,AAAI,CAAC,GAAO,GACX,GAAM,yEAAyE,KAAK,GACpF,AAAK,GACJ,GAAM,yEAAyE,KAAK,GACpF,AAAI,GACH,EAAI,OAIN,AAAI,GACH,GAAM,gBAAgB,KAAK,EAAI,SAKjC,AAAI,CAAC,GAAO,GACX,GAAM,mCAAmC,KAAK,IAI/C,AAAI,GACH,GAAU,EAAI,MAId,AAAI,KAAY,UAAY,IAAY,QACvC,GAAU,YAKL,GAAQ,EAAQ,QAAS,GAAS,WAU1C,YAA2B,GAE1B,MAAI,OAAO,IAAQ,UAAY,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,KAAQ,YAAc,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,KAAQ,YAAc,MAAO,GAAI,KAAQ,WACxN,GAID,EAAI,YAAY,OAAS,mBAAqB,OAAO,UAAU,SAAS,KAAK,KAAS,4BAA8B,MAAO,GAAI,MAAS,WAQhJ,YAAgB,GACf,MAAO,OAAO,IAAQ,UAAY,MAAO,GAAI,aAAgB,YAAc,MAAO,GAAI,MAAS,UAAY,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,aAAgB,YAAc,MAAO,GAAI,YAAY,MAAS,UAAY,gBAAgB,KAAK,EAAI,YAAY,OAAS,gBAAgB,KAAK,EAAI,OAAO,cASnT,YAAe,GACd,GAAI,GAAI,EACJ,EAAO,EAAS,KAGpB,GAAI,EAAS,SACZ,KAAM,IAAI,OAAM,sCAKjB,MAAI,aAAgB,aAAU,MAAO,GAAK,aAAgB,YAEzD,GAAK,GAAI,IACT,EAAK,GAAI,IACT,EAAK,KAAK,GACV,EAAK,KAAK,GAEV,EAAS,IAAW,KAAO,EAC3B,EAAO,GAGD,EAYR,YAA4B,GAC3B,MAAI,KAAS,KAEL,KACG,MAAO,IAAS,SAEnB,2BACG,GAAkB,GAErB,kDACG,GAAO,GAEV,EAAK,MAAQ,KACV,OAAO,SAAS,IAGhB,QAAO,UAAU,SAAS,KAAK,KAAU,wBAGzC,YAAY,OAAO,IAJtB,KAOG,MAAO,GAAK,aAAgB,WAE/B,gCAAgC,EAAK,gBAClC,YAAgB,YAGnB,KAGA,2BAaT,YAAuB,GACtB,KAAM,GAAO,EAAS,KAGtB,MAAI,KAAS,KAEL,EACG,GAAO,GACV,EAAK,KACF,OAAO,SAAS,GAEnB,EAAK,OACF,GAAQ,MAAO,GAAK,eAAkB,YAE5C,GAAK,mBAAqB,EAAK,kBAAkB,QAAU,GAC/D,EAAK,gBAAkB,EAAK,kBAEpB,EAAK,gBAKN,KAUT,YAAuB,EAAM,GAC5B,KAAM,GAAO,EAAS,KAGtB,AAAI,IAAS,KAEZ,EAAK,MACC,AAAI,GAAO,GACjB,EAAK,SAAS,KAAK,GACb,AAAI,OAAO,SAAS,GAE1B,GAAK,MAAM,GACX,EAAK,OAGL,EAAK,KAAK,GAKZ,GAAK,QAAU,OAAO,QAQtB,KAAM,IAAoB,gCACpB,GAAyB,0BAE/B,YAAsB,GAErB,GADA,EAAO,GAAG,IACN,GAAkB,KAAK,IAAS,IAAS,GAC5C,KAAM,IAAI,WAAU,GAAG,qCAIzB,YAAuB,GAEtB,GADA,EAAQ,GAAG,IACP,GAAuB,KAAK,GAC/B,KAAM,IAAI,WAAU,GAAG,sCAYzB,YAAc,EAAK,GAClB,EAAO,EAAK,cACZ,SAAW,KAAO,GACjB,GAAI,EAAI,gBAAkB,EACzB,MAAO,GAGT,OAGD,KAAM,IAAM,OAAO,OACnB,SAOC,cACC,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,OAI/E,GAFA,KAAK,IAAO,OAAO,OAAO,MAEtB,YAAgB,KACnB,KAAM,GAAa,EAAK,MAClB,EAAc,OAAO,KAAK,GAEhC,SAAW,KAAc,GACxB,SAAW,KAAS,GAAW,GAC9B,KAAK,OAAO,EAAY,GAI1B,OAKD,GAAI,KAAQ,MAAa,GAAI,MAAO,IAAS,UAC5C,KAAM,GAAS,EAAK,OAAO,UAC3B,GAAI,GAAU,MACb,GAAI,MAAO,IAAW,WACrB,KAAM,IAAI,WAAU,iCAKrB,KAAM,GAAQ,GACd,SAAW,KAAQ,IAClB,GAAI,MAAO,IAAS,UAAY,MAAO,GAAK,OAAO,WAAc,WAChE,KAAM,IAAI,WAAU,qCAErB,EAAM,KAAK,MAAM,KAAK,IAGvB,SAAW,KAAQ,IAClB,GAAI,EAAK,SAAW,EACnB,KAAM,IAAI,WAAU,+CAErB,KAAK,OAAO,EAAK,GAAI,EAAK,SAI3B,UAAW,KAAO,QAAO,KAAK,IAC7B,KAAM,GAAQ,EAAK,GACnB,KAAK,OAAO,EAAK,QAInB,MAAM,IAAI,WAAU,0CAUtB,IAAI,GACH,EAAO,GAAG,IACV,GAAa,GACb,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,MAAI,KAAQ,OACJ,KAGD,KAAK,IAAK,GAAK,KAAK,MAU5B,QAAQ,GACP,GAAI,GAAU,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,OAE9E,EAAQ,GAAW,MACnB,EAAI,EACR,KAAO,EAAI,EAAM,SAChB,GAAI,GAAW,EAAM,GACrB,KAAM,GAAO,EAAS,GAChB,EAAQ,EAAS,GAEvB,EAAS,KAAK,EAAS,EAAO,EAAM,MACpC,EAAQ,GAAW,MACnB,KAWF,IAAI,EAAM,GACT,EAAO,GAAG,IACV,EAAQ,GAAG,IACX,GAAa,GACb,GAAc,GACd,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,KAAK,IAAK,IAAQ,OAAY,EAAM,GAAQ,CAAC,GAU9C,OAAO,EAAM,GACZ,EAAO,GAAG,IACV,EAAQ,GAAG,IACX,GAAa,GACb,GAAc,GACd,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,AAAI,IAAQ,OACX,KAAK,IAAK,GAAK,KAAK,GAEpB,KAAK,IAAK,GAAQ,CAAC,GAUrB,IAAI,GACH,SAAO,GAAG,IACV,GAAa,GACN,GAAK,KAAK,IAAM,KAAU,OASlC,OAAO,GACN,EAAO,GAAG,IACV,GAAa,GACb,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,AAAI,IAAQ,QACX,MAAO,MAAK,IAAK,GASnB,MACC,MAAO,MAAK,IAQb,OACC,MAAO,IAAsB,KAAM,OAQpC,SACC,MAAO,IAAsB,KAAM,UAUnC,OAAO,YACP,MAAO,IAAsB,KAAM,cAGrC,GAAQ,UAAU,QAAU,GAAQ,UAAU,OAAO,UAErD,OAAO,eAAe,GAAQ,UAAW,OAAO,YAAa,CAC5D,MAAO,UACP,SAAU,GACV,WAAY,GACZ,aAAc,KAGf,OAAO,iBAAiB,GAAQ,UAAW,CAC1C,IAAK,CAAE,WAAY,IACnB,QAAS,CAAE,WAAY,IACvB,IAAK,CAAE,WAAY,IACnB,OAAQ,CAAE,WAAY,IACtB,IAAK,CAAE,WAAY,IACnB,OAAQ,CAAE,WAAY,IACtB,KAAM,CAAE,WAAY,IACpB,OAAQ,CAAE,WAAY,IACtB,QAAS,CAAE,WAAY,MAGxB,YAAoB,GACnB,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,YAE/E,KAAM,GAAO,OAAO,KAAK,EAAQ,KAAM,OACvC,MAAO,GAAK,IAAI,IAAS,MAAQ,SAAU,GAC1C,MAAO,GAAE,eACN,IAAS,QAAU,SAAU,GAChC,MAAO,GAAQ,IAAK,GAAG,KAAK,OACzB,SAAU,GACb,MAAO,CAAC,EAAE,cAAe,EAAQ,IAAK,GAAG,KAAK,SAIhD,KAAM,IAAW,OAAO,YAExB,YAA+B,EAAQ,GACtC,KAAM,GAAW,OAAO,OAAO,IAC/B,SAAS,IAAY,CACpB,SACA,OACA,MAAO,GAED,EAGR,KAAM,IAA2B,OAAO,eAAe,CACtD,OAEC,GAAI,CAAC,MAAQ,OAAO,eAAe,QAAU,GAC5C,KAAM,IAAI,WAAU,4CAGrB,GAAI,GAAY,KAAK,IACrB,KAAM,GAAS,EAAU,OACnB,EAAO,EAAU,KACjB,EAAQ,EAAU,MAElB,EAAS,GAAW,EAAQ,GAC5B,EAAM,EAAO,OACnB,MAAI,IAAS,EACL,CACN,MAAO,OACP,KAAM,IAIR,MAAK,IAAU,MAAQ,EAAQ,EAExB,CACN,MAAO,EAAO,GACd,KAAM,OAGN,OAAO,eAAe,OAAO,eAAe,GAAG,OAAO,eAEzD,OAAO,eAAe,GAA0B,OAAO,YAAa,CACnE,MAAO,kBACP,SAAU,GACV,WAAY,GACZ,aAAc,KASf,YAAqC,GACpC,KAAM,GAAM,OAAO,OAAO,CAAE,UAAW,MAAQ,EAAQ,KAIjD,EAAgB,GAAK,EAAQ,IAAM,QACzC,MAAI,KAAkB,QACrB,GAAI,GAAiB,EAAI,GAAe,IAGlC,EAUR,YAA8B,GAC7B,KAAM,GAAU,GAAI,IACpB,SAAW,KAAQ,QAAO,KAAK,IAC9B,GAAI,GAAkB,KAAK,GAC1B,SAED,GAAI,MAAM,QAAQ,EAAI,IACrB,SAAW,KAAO,GAAI,IACrB,GAAI,GAAuB,KAAK,GAC/B,SAED,AAAI,EAAQ,IAAK,KAAU,OAC1B,EAAQ,IAAK,GAAQ,CAAC,GAEtB,EAAQ,IAAK,GAAM,KAAK,OAGpB,AAAK,IAAuB,KAAK,EAAI,KAC3C,GAAQ,IAAK,GAAQ,CAAC,EAAI,KAG5B,MAAO,GAGR,KAAM,IAAc,OAAO,sBAGrB,GAAe,WAAK,aAS1B,SACC,cACC,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,KAC3E,EAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,GAE/E,GAAK,KAAK,KAAM,EAAM,GAEtB,KAAM,GAAS,EAAK,QAAU,IACxB,EAAU,GAAI,IAAQ,EAAK,SAEjC,GAAI,GAAQ,MAAQ,CAAC,EAAQ,IAAI,iBAChC,KAAM,GAAc,GAAmB,GACvC,AAAI,GACH,EAAQ,OAAO,eAAgB,GAIjC,KAAK,IAAe,CACnB,IAAK,EAAK,IACV,SACA,WAAY,EAAK,YAAc,GAAa,GAC5C,UACA,QAAS,EAAK,YAIZ,OACH,MAAO,MAAK,IAAa,KAAO,MAG7B,UACH,MAAO,MAAK,IAAa,UAMtB,MACH,MAAO,MAAK,IAAa,QAAU,KAAO,KAAK,IAAa,OAAS,OAGlE,cACH,MAAO,MAAK,IAAa,QAAU,KAGhC,cACH,MAAO,MAAK,IAAa,cAGtB,WACH,MAAO,MAAK,IAAa,QAQ1B,QACC,MAAO,IAAI,IAAS,GAAM,MAAO,CAChC,IAAK,KAAK,IACV,OAAQ,KAAK,OACb,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,GAAI,KAAK,GACT,WAAY,KAAK,cAKpB,GAAK,MAAM,GAAS,WAEpB,OAAO,iBAAiB,GAAS,UAAW,CAC3C,IAAK,CAAE,WAAY,IACnB,OAAQ,CAAE,WAAY,IACtB,GAAI,CAAE,WAAY,IAClB,WAAY,CAAE,WAAY,IAC1B,WAAY,CAAE,WAAY,IAC1B,QAAS,CAAE,WAAY,IACvB,MAAO,CAAE,WAAY,MAGtB,OAAO,eAAe,GAAS,UAAW,OAAO,YAAa,CAC7D,MAAO,WACP,SAAU,GACV,WAAY,GACZ,aAAc,KAGf,KAAM,IAAc,OAAO,qBAGrB,GAAY,WAAI,MAChB,GAAa,WAAI,OAEjB,GAA6B,WAAa,YAAO,SAAS,UAQhE,YAAmB,GAClB,MAAO,OAAO,IAAU,UAAY,MAAO,GAAM,KAAiB,SAGnE,YAAuB,GACtB,KAAM,GAAQ,GAAU,MAAO,IAAW,UAAY,OAAO,eAAe,GAC5E,MAAO,CAAC,CAAE,IAAS,EAAM,YAAY,OAAS,eAU/C,SACC,YAAY,GACX,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,GAE3E,EAGJ,AAAK,GAAU,GAYd,EAAY,GAAU,EAAM,KAX5B,CAAI,GAAS,EAAM,KAIlB,EAAY,GAAU,EAAM,MAG5B,EAAY,GAAU,GAAG,KAE1B,EAAQ,IAKT,GAAI,GAAS,EAAK,QAAU,EAAM,QAAU,MAG5C,GAFA,EAAS,EAAO,cAEX,GAAK,MAAQ,MAAQ,GAAU,IAAU,EAAM,OAAS,OAAU,KAAW,OAAS,IAAW,QACrG,KAAM,IAAI,WAAU,iDAGrB,GAAI,GAAY,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAU,IAAU,EAAM,OAAS,KAAO,GAAM,GAAS,KAEzG,GAAK,KAAK,KAAM,EAAW,CAC1B,QAAS,EAAK,SAAW,EAAM,SAAW,EAC1C,KAAM,EAAK,MAAQ,EAAM,MAAQ,IAGlC,KAAM,GAAU,GAAI,IAAQ,EAAK,SAAW,EAAM,SAAW,IAE7D,GAAI,GAAa,MAAQ,CAAC,EAAQ,IAAI,iBACrC,KAAM,GAAc,GAAmB,GACvC,AAAI,GACH,EAAQ,OAAO,eAAgB,GAIjC,GAAI,GAAS,GAAU,GAAS,EAAM,OAAS,KAG/C,GAFA,AAAI,UAAY,IAAM,GAAS,EAAK,QAEhC,GAAU,MAAQ,CAAC,GAAc,GACpC,KAAM,IAAI,WAAU,mDAGrB,KAAK,IAAe,CACnB,SACA,SAAU,EAAK,UAAY,EAAM,UAAY,SAC7C,UACA,YACA,UAID,KAAK,OAAS,EAAK,SAAW,OAAY,EAAK,OAAS,EAAM,SAAW,OAAY,EAAM,OAAS,GACpG,KAAK,SAAW,EAAK,WAAa,OAAY,EAAK,SAAW,EAAM,WAAa,OAAY,EAAM,SAAW,GAC9G,KAAK,QAAU,EAAK,SAAW,EAAM,SAAW,EAChD,KAAK,MAAQ,EAAK,OAAS,EAAM,SAG9B,UACH,MAAO,MAAK,IAAa,UAGtB,OACH,MAAO,IAAW,KAAK,IAAa,cAGjC,WACH,MAAO,MAAK,IAAa,WAGtB,YACH,MAAO,MAAK,IAAa,YAGtB,UACH,MAAO,MAAK,IAAa,OAQ1B,QACC,MAAO,IAAI,IAAQ,OAIrB,GAAK,MAAM,GAAQ,WAEnB,OAAO,eAAe,GAAQ,UAAW,OAAO,YAAa,CAC5D,MAAO,UACP,SAAU,GACV,WAAY,GACZ,aAAc,KAGf,OAAO,iBAAiB,GAAQ,UAAW,CAC1C,OAAQ,CAAE,WAAY,IACtB,IAAK,CAAE,WAAY,IACnB,QAAS,CAAE,WAAY,IACvB,SAAU,CAAE,WAAY,IACxB,MAAO,CAAE,WAAY,IACrB,OAAQ,CAAE,WAAY,MASvB,YAA+B,GAC9B,KAAM,GAAY,EAAQ,IAAa,UACjC,EAAU,GAAI,IAAQ,EAAQ,IAAa,SAQjD,GALA,AAAK,EAAQ,IAAI,WAChB,EAAQ,IAAI,SAAU,OAInB,CAAC,EAAU,UAAY,CAAC,EAAU,SACrC,KAAM,IAAI,WAAU,oCAGrB,GAAI,CAAC,YAAY,KAAK,EAAU,UAC/B,KAAM,IAAI,WAAU,wCAGrB,GAAI,EAAQ,QAAU,EAAQ,eAAgB,YAAO,UAAY,CAAC,GACjE,KAAM,IAAI,OAAM,mFAIjB,GAAI,GAAqB,KAIzB,GAHA,AAAI,EAAQ,MAAQ,MAAQ,gBAAgB,KAAK,EAAQ,SACxD,GAAqB,KAElB,EAAQ,MAAQ,MACnB,KAAM,GAAa,GAAc,GACjC,AAAI,MAAO,IAAe,UACzB,GAAqB,OAAO,IAG9B,AAAI,GACH,EAAQ,IAAI,iBAAkB,GAI/B,AAAK,EAAQ,IAAI,eAChB,EAAQ,IAAI,aAAc,0DAI3B,AAAI,EAAQ,UAAY,CAAC,EAAQ,IAAI,oBACpC,EAAQ,IAAI,kBAAmB,gBAGhC,GAAI,GAAQ,EAAQ,MACpB,MAAI,OAAO,IAAU,YACpB,GAAQ,EAAM,IAGf,AAAI,CAAC,EAAQ,IAAI,eAAiB,CAAC,GAClC,EAAQ,IAAI,aAAc,SAMpB,OAAO,OAAO,GAAI,EAAW,CACnC,OAAQ,EAAQ,OAChB,QAAS,GAA4B,GACrC,UAgBF,YAAoB,GAClB,MAAM,KAAK,KAAM,GAEjB,KAAK,KAAO,UACZ,KAAK,QAAU,EAGf,MAAM,kBAAkB,KAAM,KAAK,aAGrC,GAAW,UAAY,OAAO,OAAO,MAAM,WAC3C,GAAW,UAAU,YAAc,GACnC,GAAW,UAAU,KAAO,aAG5B,KAAM,IAAgB,WAAO,YACvB,GAAc,WAAI,QASxB,YAAe,EAAK,GAGnB,GAAI,CAAC,GAAM,QACV,KAAM,IAAI,OAAM,0EAGjB,UAAK,QAAU,GAAM,QAGd,GAAI,IAAM,QAAQ,SAAU,EAAS,GAE3C,KAAM,GAAU,GAAI,IAAQ,EAAK,GAC3B,EAAU,GAAsB,GAEhC,EAAQ,GAAQ,WAAa,SAAW,WAAQ,YAAM,QACtD,EAAS,EAAQ,OAEvB,GAAI,GAAW,KAEf,KAAM,GAAQ,WACb,GAAI,GAAQ,GAAI,IAAW,+BAK3B,GAJA,EAAO,GACP,AAAI,EAAQ,MAAQ,EAAQ,eAAgB,YAAO,UAClD,EAAQ,KAAK,QAAQ,GAElB,CAAC,GAAY,CAAC,EAAS,KAAM,OACjC,EAAS,KAAK,KAAK,QAAS,IAG7B,GAAI,GAAU,EAAO,SACpB,IACA,OAGD,KAAM,GAAmB,WACxB,IACA,KAIK,EAAM,EAAK,GACjB,GAAI,GAEJ,AAAI,GACH,EAAO,iBAAiB,QAAS,GAGlC,aACC,EAAI,QACJ,AAAI,GAAQ,EAAO,oBAAoB,QAAS,GAChD,aAAa,GAGd,AAAI,EAAQ,SACX,EAAI,KAAK,SAAU,SAAU,GAC5B,EAAa,WAAW,WACvB,EAAO,GAAI,IAAW,uBAAuB,EAAQ,MAAO,oBAC5D,KACE,EAAQ,WAIb,EAAI,GAAG,QAAS,SAAU,GACzB,EAAO,GAAI,IAAW,cAAc,EAAQ,uBAAuB,EAAI,UAAW,SAAU,IAC5F,MAGD,EAAI,GAAG,WAAY,SAAU,GAC5B,aAAa,GAEb,KAAM,GAAU,GAAqB,EAAI,SAGzC,GAAI,GAAM,WAAW,EAAI,aAExB,KAAM,GAAW,EAAQ,IAAI,YAGvB,EAAc,IAAa,KAAO,KAAO,GAAY,EAAQ,IAAK,GAGxE,OAAQ,EAAQ,cACV,QACJ,EAAO,GAAI,IAAW,0EAA0E,EAAQ,MAAO,gBAC/G,IACA,WACI,SAEJ,GAAI,IAAgB,KAEnB,IACC,EAAQ,IAAI,WAAY,SAChB,GAER,EAAO,GAGT,UACI,SAEJ,GAAI,IAAgB,KACnB,MAID,GAAI,EAAQ,SAAW,EAAQ,QAC9B,EAAO,GAAI,IAAW,gCAAgC,EAAQ,MAAO,iBACrE,IACA,OAKD,KAAM,GAAc,CACnB,QAAS,GAAI,IAAQ,EAAQ,SAC7B,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QAAU,EAC3B,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,OAAQ,EAAQ,OAChB,KAAM,EAAQ,KACd,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,MAIf,GAAI,EAAI,aAAe,KAAO,EAAQ,MAAQ,GAAc,KAAa,MACxE,EAAO,GAAI,IAAW,2DAA4D,yBAClF,IACA,OAID,AAAI,GAAI,aAAe,KAAQ,GAAI,aAAe,KAAO,EAAI,aAAe,MAAQ,EAAQ,SAAW,SACtG,GAAY,OAAS,MACrB,EAAY,KAAO,OACnB,EAAY,QAAQ,OAAO,mBAI5B,EAAQ,GAAM,GAAI,IAAQ,EAAa,KACvC,IACA,QAKH,EAAI,KAAK,MAAO,WACf,AAAI,GAAQ,EAAO,oBAAoB,QAAS,KAEjD,GAAI,GAAO,EAAI,KAAK,GAAI,KAExB,KAAM,GAAmB,CACxB,IAAK,EAAQ,IACb,OAAQ,EAAI,WACZ,WAAY,EAAI,cAChB,QAAS,EACT,KAAM,EAAQ,KACd,QAAS,EAAQ,QACjB,QAAS,EAAQ,SAIZ,EAAU,EAAQ,IAAI,oBAU5B,GAAI,CAAC,EAAQ,UAAY,EAAQ,SAAW,QAAU,IAAY,MAAQ,EAAI,aAAe,KAAO,EAAI,aAAe,KACtH,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,GACR,OAQD,KAAM,GAAc,CACnB,MAAO,WAAK,aACZ,YAAa,WAAK,cAInB,GAAI,GAAW,QAAU,GAAW,UACnC,EAAO,EAAK,KAAK,WAAK,aAAa,IACnC,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,GACR,OAID,GAAI,GAAW,WAAa,GAAW,aAGtC,KAAM,GAAM,EAAI,KAAK,GAAI,KACzB,EAAI,KAAK,OAAQ,SAAU,GAE1B,AAAK,GAAM,GAAK,MAAU,EACzB,EAAO,EAAK,KAAK,WAAK,iBAEtB,EAAO,EAAK,KAAK,WAAK,oBAEvB,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,KAET,OAID,GAAI,GAAW,MAAQ,MAAO,YAAK,wBAA2B,YAC7D,EAAO,EAAK,KAAK,WAAK,0BACtB,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,GACR,OAID,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,KAGT,GAAc,EAAK,KASrB,GAAM,WAAa,SAAU,GAC5B,MAAO,KAAS,KAAO,IAAS,KAAO,IAAS,KAAO,IAAS,KAAO,IAAS,KAIjF,GAAM,QAAU,OAAO,QAEvB,GAAO,IAAQ,6rDCpmDf,AAsBO,GAAM,IAAkB,KAClB,GAAkB,mBA0B7B,WAAoB,EAAgC,GAAhC,KAAA,QAAA,EAAgC,KAAA,UAAA,EAH5C,KAAA,KAAO,GAAI,SACX,KAAA,aAAe,EAIvB,SAAA,UAAA,IAAA,SAAI,GACF,MAAK,MAAK,KAAK,IAAI,IACjB,KAAK,UAAU,SAAS,KAAK,QAAS,GAEjC,KAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,IAAA,SAAI,EAAgB,GAClB,KAAK,eACL,KAAK,KAAK,IAAI,EAAQ,IAGxB,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,MAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,OAAA,SAAO,GACL,YAAK,eACE,KAAK,KAAK,OAAO,IAG1B,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,cAEhB,mBAqBA,cACE,SAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,WACE,MAAO,GAAkB,eAE3B,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,GAAkB,gBAE3B,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAC5C,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAE3D,MAAO,GAAkB,SAE3B,EAAA,UAAA,OAAA,WACE,MAAO,GAAkB,WAG3B,EAAA,UAAA,eAAA,WACE,MAAO,GAAkB,mBAG3B,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAG1D,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAEnD,MAAO,GAAkB,qBAG3B,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,MAAO,GAAkB,UAE3B,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,MAAO,GAAkB,iBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,GAAkB,WAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAG3B,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,uBAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAE3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAG3B,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,MAAO,GAAkB,iBAG3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,MAAO,GAAkB,WAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,MAAO,GAAkB,sBAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,GAAkB,WAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,MAAO,GAAkB,WAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAG3B,EAAA,UAAA,QAAA,SAA0B,GACxB,MAAO,GAAkB,YAG3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAG3B,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,yBAG3B,EAAA,UAAA,gBAAA,SAAgB,EAAiB,EAAkB,GAEjD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,wBAAA,SAAwB,EAAc,EAAkB,GAEtD,MAAO,GAAkB,4BAE3B,EAAA,UAAA,yBAAA,SAAyB,EAAa,EAAc,GAElD,MAAO,GAAkB,6BAE3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,EAAa,GAEtD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SAAkB,EAAc,EAAa,GAC3C,MAAO,GAAkB,sBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAAa,GAC1C,MAAO,GAAkB,sBAG3B,EAAA,UAAA,QAAA,SAA0C,EAAM,GAE9C,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,MAAO,GAAkB,QAG3B,EAAA,UAAA,UAAA,SAA4B,EAAM,GAChC,MAAO,GAAkB,cAG3B,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,MAAO,GAAkB,WAG3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GACpC,MAAO,GAAkB,cAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,mBAG3B,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,MAAO,GAAkB,2BAG3B,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,0BAG3B,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,MAAO,GAAkB,kCAG3B,EAAA,UAAA,UAAA,SACI,EAAa,EAAyB,EACtC,EAA4B,EAC5B,GACF,MAAO,GAAkB,cAG3B,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,MAAO,GAAkB,iCAG3B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EAAuB,EAC3D,EAAc,EAAe,GAC/B,MAAO,GAAkB,YAG3B,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,MAAO,GAAkB,WAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,MAAO,GAAkB,WAG3B,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,MAAO,GAAkB,sBAG3B,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAC3C,MAAO,GAAkB,iBAI3B,EAAA,UAAA,MAAA,SAAwB,EAAU,EAAsB,GACtD,MAAO,GAAkB,UAG3B,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,MAAO,GAAkB,SAG3B,EAAA,UAAA,SAAA,SAAyB,GACvB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,GAAkB,cAG3B,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAkB,aAG3B,EAAA,UAAA,QAAA,WACE,MAAO,GAAkB,YAE7B,KAEA,WAA2B,GACzB,KAAM,IAAI,OACN,IAAI,EAAU,4HCzpBpB,AAoBA,GAAM,IAA4B,0BA6BhC,WAAmB,GAAA,KAAA,OAAA,EATX,KAAA,MAAe,GACf,KAAA,aAAwD,GAExD,KAAA,SAAkB,GAOxB,KAAK,mBAGP,SAAA,UAAA,YAAA,SAAY,EAAsB,GAChC,AAAI,KAAK,UAAY,MACnB,QAAQ,KACJ,YAAY,KAAK,aAAY,0BAC7B,kCAAiC,EAAQ,MAE/C,KAAK,aAAe,EACpB,KAAK,SAAW,GAGlB,EAAA,UAAA,aAAA,SACI,EAAkB,EAClB,GAKF,GAJA,KAAK,aAAa,GAAY,CAAC,aAAY,EAAE,QAAO,GAIhD,KAAK,SAAS,IAAa,MAC7B,GAAM,GAAY,KAAK,SAAS,GAChC,QAAQ,KACJ,qCAAqC,EAAQ,KAAK,EAAS,KAC/D,KAAK,IAAI,EAAU,KAIjB,EAAA,UAAA,SAAN,SAAe,qGACb,MAAI,KAAY,MAAK,MACnB,CAAA,EAAO,KAAK,MAAM,IAGpB,GAAA,KAAK,MAAM,EAAA,EAAY,CAAA,EAAM,KAAK,aAAa,YAA/C,SAAA,GAAuB,EAAA,OACvB,CAAA,EAAO,KAAK,MAAM,UAGpB,EAAA,UAAA,IAAA,SAAI,GACF,GAAI,IAAY,MAAK,MACnB,MAAO,MAAK,MAAM,GAGpB,GAAM,GAAY,KAAK,aAAa,GACpC,GAAI,YAAqB,SACvB,KAAM,IAAI,OACN,QAAQ,EAAQ,sEAItB,YAAK,MAAM,GAAY,EAEhB,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,QAAA,SAAQ,GACN,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,OAGd,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,uCAGd,EAAA,UAAA,IAAA,SAAI,EAAkB,GACpB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,mBAAmB,EAAQ,mCAEjC,KAAK,MAAM,GAAY,EACvB,AAAI,KAAK,aAAa,GAAU,SAAW,MACzC,KAAK,aAAa,GAAU,QAAQ,IAIhC,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,yBAAyB,EAAQ,oCAEvC,MAAO,MAAK,aAAa,GAAU,gBAGrC,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,MAAQ,OAAO,OAAO,GAAI,IAGjC,EAAA,UAAA,MAAA,WACE,KAAK,MAAQ,GACb,KAAK,SAAW,GAChB,KAAK,oBAGC,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,GAAI,MAAO,MAAK,QAAW,aACvB,MAAO,MAAK,OAAO,UAAa,aAChC,MAAO,MAAK,OAAO,SAAS,QAAW,YACzC,OAGF,GAAM,GAAY,GAAe,KAAK,OAAO,SAAS,QACtD,GAAI,KAA6B,IAC/B,GAAM,GAAY,EAAU,IAA2B,MAAM,KAC7D,EAAU,QAAQ,SAAA,GACV,GAAA,GAAA,EAAA,MAAA,KAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,SAAS,GAAO,GAAW,EAAK,OAI7C,iBAE+B,GAC7B,GAAM,GAAS,GACf,SAAY,QAAQ,8BAA+B,SAAC,UAAG,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GACrD,UAAY,EAAQ,EAAE,GAAI,EAAE,IACrB,EAAE,KAAK,OAET,EAGT,YACI,EAAiC,EAAc,GACjD,EAAO,mBAAmB,IAAS,mBAAmB,GAAS,IAGjE,YAAoB,EAAkB,GAEpC,GADA,EAAQ,EAAM,cACV,IAAU,QAAU,IAAU,QAChC,MAAO,KAAU,OACZ,GAAI,GAAG,EAAE,IAAY,EAC1B,MAAO,CAAC,EAEV,KAAM,IAAI,OACN,oCAAoC,EAAK,aAAa,EAAQ,mBAYlE,MAAO,GAAA,IAGE,EAAA,IAAmB,iBACO,GACnC,EAAA,IAAM,EC9MR,AAqBA,GAAI,kBAGF,GAAI,IAAmB,MAErB,GAAI,GAAE,OACN,GAAI,MAAQ,SAAY,YACtB,EAAK,eACI,MAAQ,SAAY,YAC7B,EAAK,eACI,MAAQ,UAAa,YAC9B,EAAK,gBACI,MAAQ,OAAU,YAC3B,EAAK,SAEL,MAAM,IAAI,OAAM,kCAElB,GAAkB,EAEpB,MAAO,IAIT,cACE,GAAM,GAAK,KACX,MAAI,GAAG,YAAc,MACnB,GAAG,WAAa,GAAI,MAEf,EAAG,uBAUiB,EAAa,GACxC,GAAM,GAAY,KAClB,GAAI,EAAU,IAAI,GAChB,MAAO,GAAU,IAAI,GAErB,GAAM,GAAY,IAClB,SAAU,IAAI,EAAK,GACZ,EAAU,IAAI,MCzCZ,IAAM,MAGN,GAAO,OAGP,GAAQ,QAGR,GAAM,MAGN,GAAO,OAGP,GAAM,MAON,GAAM,MAON,GAAS,SAMT,GAAS,SAMT,GAAO,OAGP,GAAQ,QAGR,GAAO,OAGP,GAAQ,QAGR,GAAQ,QAGR,GAAU,UASV,GAAkB,kBAQlB,GAAY,YAWZ,GAAoB,oBAUpB,GAAc,cAOd,GAAiB,iBASjB,GAAc,cAOd,GAAO,OAMP,GAAO,OAGP,GAAc,cAOd,GAAU,UAGV,GAAS,SAMT,GAAS,SAUT,GAAuB,uBASvB,GAAsB,sBAUtB,GAAS,SAST,GAAyB,yBAQzB,GAAwB,wBAMxB,GAAM,MAGN,GAAO,OAGP,GAAS,SAQT,GAAgB,gBAShB,GAAe,eAOf,GAAwB,wBAWxB,GACT,sCAIS,GACT,qCAIS,GAAO,OAGP,GAAa,aAQb,GAA0B,0BAI1B,GAA2B,2BAI3B,GAAM,MAGN,GAAM,MAGN,GAAU,UAGV,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAO,OAOP,GAAgB,gBAGhB,GAAQ,QAGR,GAAW,WAGX,GAAiB,iBAOjB,GAAW,WAMX,GAAW,WAGX,GAAU,UAGV,GAAe,eAGf,GAAW,WAGX,GAAO,OAGP,GAAO,OAGP,GAAW,WAGX,GAAQ,QAGR,GAAQ,QAGR,GAAO,OAGP,GAAY,YAGZ,GAAW,WAMX,GAAM,MAGN,GAAQ,QAGR,GAAa,aAGb,GAAa,aAGb,GAAY,YAGZ,GAAa,aAMb,GAAM,MASN,GAAc,cASd,GAAM,MAON,GAAU,UAGV,GAAU,UASV,GAAkB,kBAUlB,GAAY,YAWZ,GAAoB,oBAWpB,GAAoB,oBASpB,GAAO,OAOP,GAAM,MAON,GAAU,UAGV,GAAM,MAGN,GAAW,WAGX,GAAS,SAGT,GAAW,WAGX,GAAsB,sBAStB,GAAsB,sBAUtB,GAAsB,sBAUtB,GAAW,WAGX,GAAS,SAQT,GAAQ,QAOR,GAAO,OAGP,GAAM,MAGN,GAAQ,QAGR,GAAO,OAOP,GAAQ,QAQR,GAAO,OAGP,GAAa,aAGb,GAAO,OAGP,GAAU,UAMV,GAAwB,wBAOxB,GAA4B,4BAI5B,GAAiB,iBAOjB,GAAqB,qBAGrB,GAAQ,QAGR,GAAU,UAMV,GAAQ,QAGR,GAAQ,QAGR,GAAY,YAMZ,GAAW,WAGX,GAAO,OAGP,GAAQ,QAMR,GAAM,MAGN,GAAO,OAGP,GAAO,OAGP,GAAU,UAGV,GAAW,WAGX,GAAO,OAGP,GAAM,MAON,GAAiB,iBAOjB,GAAS,SAOT,GAAU,UAMV,GAAoB,oBAGpB,GAAS,SAGT,GAAM,MAGN,GAAgB,gBAOhB,GAAe,eAaf,GAAM,MAGN,GAAO,OAGP,GAAO,OAMP,GAAO,OAOP,GAAY,YAMZ,GAAS,SAQT,GAAS,SAMT,GAAqB,qBAOrB,GAAY,YAMZ,GAAO,OAMP,GAAa,aASb,GAAmB,mBAQnB,GAAe,eAef,GAAc,cAgBd,GAAuB,uBC7zBpC,AAuBA,GAAM,IACF,GAAU,iBAAkB,WAAM,MAAA,IAAI,OACpC,GACF,GAAU,eAAgB,WAAM,MAAA,IAAI,mBAqEpC,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,MAAO,IAAe,IAAI,eAOA,GAC1B,MAAO,IAAa,IAAI,eAGW,GAInC,OAHM,GAAK,GAAe,UACpB,EAAyB,MAGvB,GAAA,GAAA,EAAA,OAAC,EAAA,EAAA,KAAM,EAAA,EAAA,MACb,GAAI,EACF,MAEK,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GACL,EAAA,EAAA,MAAA,KAAA,GACP,AAAI,IAAY,GACd,EAAO,KAAK,GAGhB,MAAO,eAcsB,GACtB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,YACb,EAAM,GAAQ,EAAY,GAChC,AAAI,GAAe,IAAI,IACrB,QAAQ,KACJ,eAAe,EAAU,iBACzB,KAAI,EAAW,4BAErB,GAAe,IAAI,EAAK,eAWO,GACxB,GAAA,GAAA,EAAA,WAEP,AAAI,GAAa,IAAI,IAGnB,CAAI,KAAM,QAAQ,UAChB,QAAQ,KAAK,gCAAgC,EAAU,MAG3D,GAAa,IAAI,EAAY,eAW3B,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,GAAI,CAAC,GAAe,IAAI,GACtB,KAAM,IAAI,OACN,eAAe,EAAU,iBACzB,KAAI,EAAW,wBAErB,GAAe,OAAO,eAIW,GACjC,GAAI,CAAC,GAAa,IAAI,GACpB,KAAM,IAAI,OACN,iBAAiB,EAAU,mCAEjC,GAAa,OAAO,eAUlB,EAA+B,GACjC,GAAM,GAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,GAAM,GACF,OAAO,OAAO,GAAI,EAAc,CAAC,YAAa,IAClD,GAAe,KAInB,YAAiB,EAAoB,GACnC,MAAU,GAAW,IAAI,ECrN3B,YAkCwB,GAMtB,OAJI,GAAU,EAAM,OAChB,EAAO,EACP,EAAQ,EAEL,EAAU,GAEf,EAAS,KAAK,SAAW,EAAW,EAEpC,IAEA,EAAO,EAAM,GACb,EAAM,GAAW,EAAM,GACvB,EAAM,GAAS,cAKG,EAAa,EAAW,GAC5C,MAAO,MAAK,IAAI,EAAK,KAAK,IAAI,EAAG,gBAGD,GAChC,MAAO,GAAM,IAAM,EAAI,EAAM,EAAM,cAGjB,GAElB,OADI,GAAM,EACD,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,GAAO,EAAI,GAEb,MAAO,eAUmB,EAAW,GACrC,GAAM,GAAI,KAAK,SACf,MAAQ,GAAI,EAAM,GAAI,GAAK,cAID,EAAe,GAEzC,OADI,GAAS,EACJ,EAAI,EAAG,EAAI,EAAE,OAAQ,KAC5B,GAAM,GAAO,OAAO,EAAE,IAAM,OAAO,EAAE,IACrC,GAAU,EAAO,EAEnB,MAAO,cAkBc,EAAe,GACpC,GAAI,CAAC,EACH,KAAM,IAAI,OAAM,MAAO,IAAQ,SAAW,EAAM,iBAKhD,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EACI,GAAY,EAAQ,GACpB,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlC,GAC5B,EACI,GAAK,KACL,WAAM,MAAA,8EAwBR,EAA0B,EAAkB,GAI9C,GAJ4B,AAAA,IAAA,QAAA,GAAA,IAAkB,AAAA,IAAA,QAAA,GAAA,IAC9C,AAAI,GAAU,MACZ,GAAS,IAEP,MAAM,QAAQ,IAAQ,GAAa,IAAQ,CAAC,EAC9C,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAAQ,EAAI,GAAI,EAAQ,OAG1B,GAAO,KAAK,GAEd,MAAO,eAcqB,GAC5B,GAAI,EAAM,SAAW,EAEnB,MAAO,GAGT,OADI,GAAO,EAAM,GACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,GAAQ,EAAM,GAEhB,MAAO,eAGqB,GAC5B,MAAO,GAAM,SAAW,cAGE,EAAgB,GAC1C,GAAI,IAAO,EACT,MAAO,GAET,GAAI,GAAM,MAAQ,GAAM,KACtB,MAAO,GAGT,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,EAAG,GACf,MAAO,GAGX,MAAO,eAGa,GACpB,MAAO,GAAI,IAAM,cAGE,GAEnB,GAAK,KAAa,MAAQ,KAExB,MAAQ,MAAa,KAAK,GAE5B,GAAI,IAAM,SACR,MAAO,GACF,GAAI,IAAM,UACf,MAAO,GAEP,GAAM,GAAM,KAAK,IAAI,EAAI,GACzB,MAAQ,GAAM,GAAM,GAAM,eAIM,GAClC,GAAM,GAAQ,KAAK,KAAK,KAAK,KAAK,IAClC,MAAO,CAAC,EAAO,KAAK,KAAK,EAAO,gBAeI,GAEpC,OADM,GAAkB,GAAI,aAAY,GAC/B,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAgB,GAAK,EAEvB,UAAQ,GACD,cAGgB,EAAW,GAClC,MAAI,IAAQ,EAAE,OACL,EAEF,EAAI,IAAI,OAAO,EAAO,EAAE,oBAI7B,EAAwB,EACxB,GADwB,MAAA,KAAA,QAAA,GAAA,SAAW,GAAoB,MAAA,KAElD,GAAI,SAAc,SAAC,EAAS,GACjC,GAAI,GAAW,EAET,EAAQ,WACZ,GAAI,KACF,IACA,OAGF,IAEA,GAAM,GAAc,EAAQ,GAE5B,GAAI,GAAc,MAAQ,GAAY,GACpC,IACA,OAEF,WAAW,EAAO,IAGpB,kBAcA,EAAiB,GAInB,OAHI,GAAY,EACZ,EAAc,GAET,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,GAAI,EAAM,IAAM,EACd,GAAa,EAAM,WACV,EAAM,KAAO,IACtB,GAAI,IAAgB,GAClB,KAAM,OACF,yCACA,oBAAmB,EAAW,YAAY,IAEhD,EAAc,UACL,EAAM,GAAK,EACpB,KAAM,OAAM,gCAAgC,EAAM,GAAE,WAAW,GAInE,GAAI,IAAgB,IAClB,GAAI,EAAO,GAAK,IAAS,EACvB,KAAM,OAAM,QAAQ,EAAI,qCAAqC,GAE/D,MAAO,GAGT,GAAI,IAAc,EAChB,KAAM,OACF,qCAAqC,EAAK,+BAGhD,GAAI,EAAO,IAAc,EACvB,KAAM,OACF,oDACA,QAAO,EAAI,MAAM,IAGvB,GAAM,GAAW,EAAM,QACvB,SAAS,GAAe,EAAO,EACxB,cAIL,EAAuB,GACzB,GAAM,GAAO,EAAM,OAGnB,SAAO,GAAQ,KAAO,EAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAK,GAAG,OAAO,GAGzD,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,CAAC,GAAQ,EAAK,IACrC,WACI,MAAA,+CAA+C,EAAI,KAAK,EAAI,SAC5D,aAAY,KAGpB,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,KACvB,WAAM,MAAA,iDACF,aAAY,KAGb,EAAK,IAAI,SAAA,GAAK,MAAA,GAAI,EAAI,EAAO,EAAI,gBAIb,EAAiB,GAS5C,OAPM,GAAqB,GACrB,EAAqB,GACrB,EAAe,GAAQ,MAAQ,MAAM,QAAQ,IAAS,EAAK,SAAW,EACtE,EAAQ,GAAQ,MAAQ,EAC1B,KACA,GAAe,EAAM,GAAO,OAC5B,EAAI,EACC,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,GAAQ,MACV,GAAI,EAAK,KAAO,GAAK,EAAM,KAAO,EAChC,KAAM,IAAI,OACN,sBAAsB,EAAC,mBAAmB,EAAM,GAAE,cAExD,AAAK,GAAK,IAAM,MAAQ,EAAK,GAAK,IAAM,EAAM,KAAO,GACnD,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAEhB,AAAI,EAAK,IAAM,GACb,IAGJ,AAAI,EAAM,KAAO,GACf,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAGlB,MAAO,CAAC,SAAQ,EAAE,SAAQ,eAIxB,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,SACnB,EAAS,GAAI,OAAgB,OAE7B,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAA+B,GACjC,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAC1B,KAAM,OAAM,oBAAoB,EAAK,4BAA4B,EAAG,kBAM7C,GAC3B,MAAO,KAAU,QAAU,IAAU,aAAe,IAAU,WAC1D,IAAU,SAAW,IAAU,qBAOL,EAAmB,GACjD,MAAI,KAAY,aAGZ,KAAY,WAAa,IAAY,aAGrC,IAAY,SAAW,IAAY,WAAa,IAAY,aALvD,GAQL,MAAY,QAAU,IAAY,oBAMX,GAC3B,MAAO,aAAa,eAAgB,YAAa,aAC7C,YAAa,wBAGa,GAC9B,GAAI,IAAU,WAAa,IAAU,QACnC,MAAO,GACF,GAAI,IAAU,YACnB,MAAO,GACF,GAAI,IAAU,OACnB,MAAO,GAEP,KAAM,IAAI,OAAM,iBAAiB,eAUA,GACnC,GAAI,GAAO,KACT,MAAO,GAET,GAAI,GAAQ,EACZ,SAAI,QAAQ,SAAA,GAAK,MAAA,IAAS,EAAE,SACrB,cAIgB,GACvB,MAAO,OAAO,IAAU,UAAY,YAAiB,oBAG7B,GACxB,MAAO,OAAO,IAAU,sBAGD,GACvB,MAAO,OAAO,IAAU,qBAGC,GACzB,MAAI,OAAM,QAAQ,GACT,GAAW,EAAO,IAEvB,YAAkB,cACb,UACE,YAAkB,aAAc,YAAkB,YACpD,QACE,GAAS,GACX,UACE,GAAS,GACX,SACE,GAAU,GACZ,OAEF,sBAGkB,GACzB,MAAO,CAAC,CAAE,IAAK,EAAE,aAAe,EAAE,MAAQ,EAAE,mBAGf,EAAc,GAC3C,OAAS,GAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,GAAI,EAAO,IAAM,EACf,MAAO,GAGX,MAAO,eAGsB,GAC7B,GAAM,GAAO,EAAM,OACnB,GAAI,EAAO,EACT,MAAO,GAKT,GAAM,GAAU,GAAI,OAAM,EAAO,GACjC,EAAQ,EAAO,GAAK,EAAM,EAAO,GACjC,OAAS,GAAI,EAAO,EAAG,GAAK,EAAG,EAAE,EAC/B,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAM,EAAI,GAE1C,MAAO,eAOL,EAAiB,GACnB,MAAI,KAAU,SACL,GAAa,GAGf,GAAa,CAAC,GAAQ,eAGF,EAAe,GAC1C,GAAI,IAAU,SACZ,KAAM,IAAI,OAAM,6CASlB,GAPA,AAAI,MAAM,QAAQ,IAChB,GAAI,GAAQ,IAGd,AAAI,KAAM,QAAQ,UAChB,GAAyB,EAAe,GAEtC,GAAmB,EAAG,GACxB,MAAO,GAET,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,QAEnB,OADM,GAAO,GAAI,YAAY,EAAe,QACnC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,AAAI,KAAK,MAAO,EAAe,MAAQ,GACrC,GAAK,GAAK,GAGd,MAAO,OAEP,MAAM,IAAI,OAAM,qBAAqB,GAIzC,YAA2B,EAAgB,EAAiB,GAC1D,GAAM,GAAM,GAAI,OAChB,GAAI,EAAM,SAAW,EAEnB,OADM,GAAI,EAAM,GACP,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,EAAE,EAAS,OAMtB,QAHM,GAAI,EAAM,GACV,EAAO,EAAM,MAAM,GACnB,EAAM,EAAK,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IACjC,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,GAAkB,EAAS,EAAI,EAAK,EAAM,GAGvD,MAAO,eAIqB,EAAiB,GAC7C,GAAI,EAAM,SAAW,EAEnB,MAAO,GAAE,GAEX,GAAM,GAAO,EAAM,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IAC5C,GAAI,IAAS,EAEX,MAAO,GAET,GAAI,IAAS,EAAE,OACb,KAAM,IAAI,OAAM,IAAI,EAAK,mCAAmC,EAAE,OAAM,KAGtE,MAAO,IAAkB,EAAG,EAAO,GAGrC,YAA4B,EAAe,GACzC,MAAQ,aAAa,eAAgB,IAAU,WAC1C,YAAa,aAAc,IAAU,SACrC,YAAa,aAAc,IAAU,mBAIxC,EAAc,GAEhB,OADM,GAAQ,GAAoB,EAAM,GAC/B,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAM,GAAK,EAEb,MAAO,eAIL,EAAc,GAChB,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,OACnB,MAAO,IAAI,YAAW,GAEtB,KAAM,IAAI,OAAM,qBAAqB,eAUrC,EAAiB,GACnB,GAAM,GAAO,EAAM,OAAO,SAAC,EAAM,GAAS,MAAA,GAAO,GAAM,GACvD,GAAI,GAAS,MAAQ,IAAU,UAC7B,MAAO,IAAc,EAAO,GAAI,cAAa,IACxC,GAAI,IAAU,QACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IACtC,GAAI,IAAU,OACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IAE3C,KAAM,IAAI,OAAM,qBAAqB,iBAgBvC,MAAO,MAAM,SAAS,kBAG2B,GACjD,EAAM,QAAQ,SAAA,GACZ,EACI,OAAO,UAAU,IAAY,GAAW,EACxC,WACI,MAAA,mEACA,WAAU,EAAK,sBAqBvB,EAAc,GAChB,MAAO,MAAM,SAAS,MAAM,EAAM,eAWP,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,SACtC,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAG,eAWL,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,SAC9C,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAO,eAYlC,EAAgB,EAAc,GAChC,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,EAAQ,GAAK,EAAK,GAE7B,MAAO,eAYL,EAAe,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,EAAQ,IACrC,GAAS,EAAK,GAAK,EAAQ,GAE7B,SAAK,EAAK,OAAS,GAAK,EACjB,22BC3xBT,AA+BA,GAAA,IAAA,WACE,WAAoB,EAAoC,GAApC,KAAA,aAAA,EAAoC,KAAA,OAAA,EACtD,AAAI,GAAU,MACZ,MAAK,OAAS,GAAI,KAItB,SAAA,UAAA,cAAA,SAAc,EAAoB,EAAwB,GAQxD,OANI,GACE,EAAsB,WAC1B,EAAU,KAEN,EAAQ,KAAK,aAAa,KAAK,cAE5B,GACP,GAAM,GAAS,EAAQ,GAGvB,EAAO,OAAO,KAAK,SAAA,GACjB,GAA0B,EAAY,EAAO,MAAO,MAL/C,EAAI,EAAG,EAAI,EAAQ,OAAQ,MAA3B,GAST,GAAM,GAAgB,CACpB,WAAU,EACV,QAAO,EACP,OAAM,EACN,OAAQ,EAAM,KAAK,SAAA,GAAU,MAAA,GAAO,WACpC,UAAW,EAAM,KACb,SAAA,GAAU,MAAA,GAAO,qBAAuB,KACpC,EAAO,sBACP,MAEV,MAAO,IAGT,EAAA,UAAA,iBAAA,SAAiB,GAAjB,GAAA,GAAA,KACS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,OAAQ,EAAA,EAAA,OAAQ,EAAA,EAAA,UAE5C,EAAQ,QAAQ,SAAA,GACd,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,IAAY,KAAK,SAAA,GACnD,EAAK,OAAO,iBACR,EAAY,EAAQ,EAAe,GAAI,EAAe,GAAI,EAC1D,EAAe,SAI3B,iBAGI,EAAsB,EAAU,GAClC,GAAI,IAAU,UAEZ,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAE1B,eAAQ,KAAK,SAAS,EAAG,sBAAsB,EAAU,KAClD,GAGX,MAAO,GAGT,GAAA,IAAA,WAAA,cACE,SAAA,UAAA,iBAAA,SACI,EAAc,EAAgB,EAC9B,EAAgC,EAChC,GACF,GAAM,GAAO,MAAO,IAAW,SAAW,GAAiB,EAAM,KAAM,GAC7B,EAAO,MAC3C,EAAa,GAAc,EAAM,IACjC,EAAO,EAAO,KACd,EAAO,EAAO,KACd,EAAQ,GAAc,EAAO,MAAM,WAAY,IACjD,EAAyB,GAE7B,OAAW,KAAQ,IACjB,GAAM,GAAQ,EAAO,GACrB,GAAI,GAAS,MAGX,GAAM,GAAa,EAAM,OAAS,EAAO,MACnC,EAAY,EAAW,OAC7B,GACO,EAAI,KAAK,EAAS,KAAK,GAAY,EAAI,EAAa,IAAE,KAIjE,QAAQ,IACJ,KAAK,EAAU,MAAO,EAAI,MAAO,EAAI,KAAK,EAAK,MAAO,EAAI,MACtD,EAAsB,MAAO,EACjC,mBAAoB,YAAa,aAAc,gBAC/C,eAAgB,qBAExB,KCjIA,YA4CI,EAAkB,EAAc,GAKlC,OAFM,GAA8C,GAC9C,EAA0C,GACvC,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,EAAa,EAAG,GAAG,IAAM,GAG3B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GACZ,EAAa,EAAK,OACxB,OAAW,KAAa,IAItB,OAHM,GAAQ,EAAW,GAErB,EAAgB,GACX,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAa,EAAM,KACrB,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAa,EAAO,IAAM,KACzD,EAAgB,GAChB,EAAW,EAAK,IAAM,GACtB,MAIJ,GAAI,EACF,OAMN,GAAM,GAAgD,GACtD,EAAe,EAAE,IAAM,GAGvB,OAFM,GAAwC,GAErC,EAAI,EAAK,OAAS,EAAG,GAAK,EAAG,IAKpC,OAJM,GAAO,EAAK,GACZ,EAAa,EAAK,OAGf,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,IACvC,GAAI,EAAe,EAAK,QAAQ,GAAG,KACjC,OAAW,KAAa,GACtB,EAAe,EAAW,GAAW,IAAM,GAC3C,EAAS,EAAK,IAAM,GAEtB,MAON,OADM,GAA2B,GACxB,EAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GAElB,GAAI,EAAW,EAAK,KAAO,EAAS,EAAK,KAEvC,GAAM,GAA8C,GACpD,OAAW,KAAa,GAAK,QAC3B,GAAM,GAAY,EAAK,OAAO,GAC9B,AAAI,EAAa,EAAU,KACzB,GAAa,GAAa,GAK9B,GAAM,GAAa,OAAO,OAAO,GAAI,GACrC,EAAW,OAAS,EACpB,EAAW,QAAU,EAAK,QAE1B,EAAa,KAAK,IAItB,MAAO,eAWL,EACA,EAA0B,EAC1B,GAEF,mBAAS,GACP,GAAM,GAAO,EAAa,GAEpB,EAAgB,GAYtB,GAXA,EAAK,QAAQ,QAAQ,SAAA,GACnB,GAAM,GAAa,EAA6B,EAAE,IAClD,AAAI,GAAc,KAChB,EAAI,KAAK,GAIT,EAAI,KAAK,QAIT,EAAK,UAAY,KACnB,KAAM,IAAI,OACN,wDACA,QAAO,EAAK,WAAU,MAI5B,GAAM,GAAiB,EAAK,SAAS,cAE1B,GACT,GAAI,CAAE,KAAa,IACjB,KAAM,IAAI,OACN,iCAAiC,EAAS,KAC1C,+BAA8B,OAAO,KAAK,GAAe,MAI/D,GAAM,GAAK,EAAK,WAAM,MAAA,GAAe,OACrC,GAAI,EAAG,QAAU,UACf,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BAChB,GAAS,wCAAwC,EAAG,MAAK,MAElE,GAAM,GAAI,EAAK,OAAO,GACtB,GAAI,CAAC,GAAiB,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BACnB,KAAI,EAAS,gBAAgB,EAAG,MAAK,4BACrC,4BAA2B,EAAE,MAAK,MAGxC,GAAI,EAA6B,EAAE,KAAO,KACxC,EAA6B,EAAE,IAAM,OAErC,GAAM,GAAc,EAA6B,EAAE,IACnD,EAA6B,EAAE,IAAM,EAAI,EAAa,GACtD,EAAY,YA7BhB,OAAW,KAAa,GAAK,SAAlB,IAxBJ,EAAI,EAAa,OAAS,EAAG,GAAK,EAAG,MAArC,GCvIX,AAqBA,GAAM,IAAwB,GAExB,GAA6B,EAE7B,GAAwB,cAG1B,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAU,GAAe,GACzB,EAAY,GAAwB,EAAM,EAAO,EAAO,GACxD,EAAO,EAAM,OACb,EAAY,GAAkB,EAAM,EAAO,EAAO,EAAS,GAC3D,EAAQ,CAAC,UACf,MAAI,IACF,GAAM,KAAK,YAAY,GACvB,EAAM,KAAK,WAAW,GACtB,EAAM,KAAK,aAAa,EAAK,KAC7B,EAAM,KAAK,cAEb,EAAM,KAAK,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,IAAG,KAAK;IACxC,EAAM,KAAK;GAGpB,YACI,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAI,GAAc,GAClB,EAAU,EAAQ,EAAQ,OAAS,GACnC,EAAY,GAAI,OAAM,GAAS,KAAK,GACpC,EAAO,EAAM,OACb,EACF,IAAU,YAAc,GAAoB,GAAQ,EAExD,GAAI,EAAO,EACT,OAAS,GAAM,EAAG,EAAM,EAAI,EAAS,IAEnC,OADM,GAAS,EAAM,EACZ,EAAI,EAAG,EAAI,EAAS,IAC3B,EAAU,GAAK,KAAK,IAChB,EAAU,GACV,GAAY,EAAe,EAAS,GAAI,EAAG,GAAO,QAI5D,MAAO,GAGT,YACI,EAAqC,EAAa,GACpD,GAAI,GACJ,MAAI,OAAM,QAAQ,GAChB,EAAY,WAAW,EAAI,GAAG,QAAQ,KAAuB,MACtD,YAAW,EAAI,GAAG,QAAQ,KAAuB,KACnD,AAAI,GAAS,GAClB,EAAS,IAAI,EAAG,IACX,AAAI,IAAU,OACnB,EAAS,GAAgB,GAEzB,EAAS,WAAW,EAAI,QAAQ,KAAwB,WAGnD,GAAS,EAAQ,GAG1B,YAAyB,GACvB,MAAO,KAAM,EAAI,QAAU,OAG7B,YACI,EAA2B,EAAiB,EAC5C,EAAmB,EAAqB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAM,GAAoB,IAAU,YAAc,EAAI,EAEhD,EAAO,EAAM,GACb,EAAO,EAAM,OACnB,GAAI,IAAS,GACX,GAAI,IAAU,aACZ,GAAM,GAAe,GAAoB,GACzC,MAAO,CAAC,GAAY,EAAa,GAAI,EAAG,IAE1C,MAAI,KAAU,OACL,CAAC,GAAgB,EAAK,KAExB,CAAC,EAAK,GAAG,YAGlB,GAAI,IAAS,GACX,GAAI,EAAO,IACT,GAAM,GAAgB,GAA6B,EAE/C,EAAY,MAAM,KAClB,EAAK,MAAM,EAAG,IACd,EAAW,MAAM,KAAqC,EAAK,MAC1D,GAAO,IAA8B,EACtC,EAAO,IACX,MAAI,KAAU,aACZ,GAAY,GAAoB,GAChC,EAAW,GAAoB,IAE1B,CACL,IACA,EAAU,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAChD,KAAK,MACV,UACA,EACK,IACG,SAAC,EAAG,GAAM,MAAA,IACN,EAAG,EAAU,EAAO,GAA6B,GAAI,KAC5D,KAAK,MACV,KAGJ,GAAM,GACF,IAAU,YAAc,GAAoB,GACpB,MAAM,KAAoB,GAEtD,MAAO,CACL,IACA,EAAY,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAClD,KAAK,MACV,KAKJ,GAAM,GAAW,EAAM,MAAM,GACvB,EAAa,EAAQ,MAAM,GAC3B,EAAS,EAAQ,GAAK,EACtB,EAAkB,GACxB,GAAI,EAAO,IACT,OAAS,GAAI,EAAG,EAAI,GAA4B,KAC9C,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,KAEN,EAAM,KAAK,OACX,OAAS,GAAI,EAAO,GAA4B,EAAI,EAAM,KACxD,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,SAGnB,QAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,IAGrB,GAAM,GAAM,IAAS,EAAI,IAAM,GAC/B,EAAM,GAAK,IAAM,EAAM,GAAK,EAC5B,OAAS,GAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,EAAM,GAAK,IAAM,EAAM,GAAK,EAG9B,OADI,GAAa;EACR,EAAI,EAAG,EAAI,EAAM,IACxB,GAAc;EAEhB,SAAM,EAAM,OAAS,GACjB,IAAM,EAAM,EAAM,OAAS,GAAK,IAAO,GAAS,GAAK,GAClD,EAGT,YAA6B,GAG3B,OADM,GAAyC,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,GAAK,EACpC,EAAc,KAAK,CAAC,EAAK,GAAI,EAAK,EAAI,KAExC,MAAO,GCnMT,kBA6CE,WAAY,EAA2B,EAAU,GAAjD,GAAA,GAAA,KAIE,GAJqC,KAAA,MAAA,EACrC,KAAK,MAAQ,EAAM,QACnB,KAAK,KAAO,GAAmB,GAE3B,GAAU,MACZ,GAAM,GAAI,EAAO,OACjB,EACI,IAAM,KAAK,KACX,WAAM,MAAA,qBAAqB,EAAC,6BACxB,2BAA0B,EAAK,KAAI,QAE7C,GAAI,IAAU,YACZ,KAAM,IAAI,OACN,8JAIN,KAAK,OAAS,GAAU,GAAuB,EAAO,KAAK,MAC3D,KAAK,QAAU,GAAe,GAWhC,SAAA,UAAA,IAAA,SAAI,UAAJ,GAAA,KAA8B,EAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GAC5B,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAEV,EACI,EAAK,SAAW,KAAK,KACrB,WAAM,MAAA,uCAAuC,EAAK,OAAM,UACpD,oBAAmB,EAAK,KAAI,OAEpC,GAAM,GAAQ,KAAK,WAAW,GAC9B,KAAK,OAAO,GAAS,GAUvB,EAAA,UAAA,IAAA,kBAAI,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACF,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAGV,OADI,GAAI,EACU,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,EAAM,GAAK,GAAO,KAAK,MAAM,IAC/B,GAAM,GAAM,qCAAqC,EAAI,KACjD,mBAAkB,KAAK,OAC3B,KAAM,IAAI,OAAM,GAElB,IAGF,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,MAAK,OAAO,IAGrB,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,IAGT,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,KAAK,MAAM,QACnC,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,KAAK,QAAQ,IAC1C,GAAS,EAAK,GAAK,KAAK,QAAQ,GAElC,SAAK,EAAK,OAAS,GAAK,EACjB,GAGT,OAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAY,WAAW,KAAK,OAAQ,KAAK,MAAO,KAAK,QAGhE,KA8BI,GAAiC,KAEjC,GAAuB,iBAaM,GAC/B,GAAY,cAQe,GAC3B,GAAY,mBA4DZ,WAAY,EAAoB,EAAiB,EAAgB,GAXjE,KAAA,KAAO,GAoIG,KAAA,mBAAqB,GAxH7B,KAAK,MAAQ,EAAM,QACnB,KAAK,MAAQ,GAAS,UACtB,KAAK,KAAO,GAAmB,GAC/B,KAAK,QAAU,GAAe,GAC9B,KAAK,OAAS,EACd,KAAK,GAAK,EACV,KAAK,SAAY,KAAK,KAAO,EAAI,KAAK,KAAK,WAAa,SAG1D,cAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQd,EAAA,UAAA,OAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAU,OAAO,KAAK,MAAO,KAAK,MAAY,UAOvD,EAAA,UAAA,WAAA,WACE,MAAO,IAAU,OAAO,KAAK,MAAO,KAAK,MAAY,KAAK,aAStD,EAAA,UAAA,MAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAc,KAAK,MAAO,UASnC,EAAA,UAAA,UAAA,WACE,MAAO,IAAc,KAAK,MAAO,KAAK,aASlC,EAAA,UAAA,KAAN,mHACE,MAAK,kBACC,EAAO,KAAY,KAAK,KAAK,QAC/B,KAAK,QAAU,SACH,CAAA,EAAM,GADlB,CAAA,EAAA,UACI,EAAQ,EAAA,OACd,IACE,MAAA,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,IAAkB,YACxC,GACA,KAAM,IAAI,OACN,kHAIR,MAAA,CAAA,EAAO,SAST,EAAA,UAAA,SAAA,WACE,KAAK,kBACL,GAAM,GAAO,KAAY,SAAS,KAAK,QACvC,GAAI,KAAK,QAAU,SACjB,IACE,MAAQ,GAAsB,IAAI,SAAA,GAAK,MAAA,IAAkB,WAEzD,GACA,KAAM,IAAI,OACN,iGAIR,MAAO,IAIH,EAAA,UAAA,MAAN,2GACE,YAAK,kBACQ,CAAA,EAAM,KAAY,KAAK,KAAK,gBACzC,MADM,GAAO,EAAA,OACT,KAAK,QAAU,SACjB,CAAA,EAAO,GAEP,CAAA,EAAO,GAAI,YAAY,EAAoB,eAS/C,EAAA,UAAA,QAAA,WACE,GAAI,KAAK,WACP,OAEF,KAAY,cAAc,MAC1B,KAAK,mBAAqB,IAI5B,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,oDAGd,EAAA,UAAA,gBAAA,WACE,GAAI,KAAK,WACP,KAAM,IAAI,OAAM,wBAYpB,EAAA,UAAA,MAAA,SAAM,GAAA,MAAA,KAAA,QAAA,GAAA,IACG,GAAU,MAAM,KAAM,IAO/B,EAAA,UAAA,MAAA,WACE,YAAK,kBACE,GAAU,MAAM,OAQzB,EAAA,UAAA,SAAA,SAAS,GAAA,AAAA,IAAA,QAAA,GAAA,IACP,GAAM,GAAO,KAAK,WAClB,MAAO,IAAe,EAAM,KAAK,MAAO,KAAK,MAAO,IAGtD,EAAA,UAAA,KAAA,SAAqB,GACnB,YAAK,kBACE,GAAU,KAAK,KAAW,IAEnC,EAAA,UAAA,SAAA,SAAS,EAAkB,EAAe,GAAjC,MAAA,KAAA,QAAA,GAAA,IACP,KAAK,kBACE,KAAY,aAAa,KAAM,EAAW,EAAM,IAG3D,KACA,OAAO,eAAe,EAAQ,OAAO,YAAa,CAChD,MAAO,SAAC,GAMN,MAAO,CAAC,CAAC,GAAY,EAAS,MAAQ,MAAQ,EAAS,UAAY,MAC/D,EAAS,iBAAmB,2BAoCiB,GAAA,EAAA,GAGnD,WACI,EAAgC,EAAoB,EACpD,GAFJ,GAAA,GAGE,EAAA,KAAA,KACI,EAAa,MAAO,EAAa,MAAO,EAAa,OAAQ,IAAS,KAHxC,SAAA,UAAA,EAIlC,EAAK,KAAO,IAWd,SAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAS,QAAU,KAAK,MAC1B,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,GAAI,CAAC,GAAiB,EAAS,MAAO,KAAK,OACzC,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,KAAY,cAAc,MAC1B,KAAK,OAAS,EAAS,OACvB,KAAY,OAAO,KAAM,OAG3B,EAAA,UAAA,QAAA,WACE,KAAY,gBAAgB,MAC5B,KAAK,mBAAqB,IAE9B,GAvCqD,GAyCrD,OAAO,eAAe,GAAU,OAAO,YAAa,CAClD,MAAO,SAAC,GACN,MAAO,aAAoB,IAAU,EAAS,QAAU,MACpD,EAAS,iBAAkB,aCnhBnC,AAgEA,AAAA,UAAY,GACV,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,OAPU,EAAA,MAAA,GAAA,KAAI,KAqBhB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,QACA,EAAA,UAAA,cAJG,IAAA,IAAiB,KAOtB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,OACA,EAAA,UAAA,cAJG,IAAA,IAAgB,KAOrB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,UACA,EAAA,KAAA,UACA,EAAA,UAAA,cAJG,IAAA,IAAmB,KAOxB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,YACA,EAAA,MAAA,YACA,EAAA,KAAA,YACA,EAAA,UAAA,cAJG,IAAA,IAAqB,KAO1B,GAAM,IAAgB,CACpB,QAAW,GACX,MAAS,GACT,KAAQ,GACR,UAAa,gBAGY,EAAiB,GAC1C,GAAI,IAAU,UAAY,IAAU,UAClC,GAAI,IAAU,UAAY,IAAU,SAClC,MAAO,SAET,KAAM,IAAI,OAAM,kBAAkB,EAAK,SAAS,GAElD,MAAO,IAAc,GAAO,eAIH,GACzB,MAAO,IAAW,EAAM,SCpI1B,YAsBiD,EAAM,GACrD,GAAI,EAAE,QAAU,EAAE,MAChB,MAAO,CAAC,EAAG,GAEb,GAAM,GAAQ,GAAW,EAAE,MAAO,EAAE,OACpC,MAAO,CAAC,EAAE,KAAK,GAAQ,EAAE,KAAK,gBAGC,EAAW,GAC1C,EACI,EAAE,QAAU,EAAE,MACd,WAAM,MAAA,2BAA2B,EAAE,MAAK,QACpC,YAAW,EAAE,MAAK,oCAGG,EAAgB,GAC7C,MAAO,GAAW,KAAK,SAAA,GAAK,MAAA,GAAE,KAAO,EAAO,iBAeR,GACpC,GAAM,GAAiB,GACjB,EAAO,GAAI,KACjB,UAAoB,EAAQ,EAAM,GAC3B,EAGT,YACI,EAA4B,EAAgB,GAC9C,GAAI,GAAa,KACf,OAEF,GAAI,YAAqB,IACvB,EAAK,KAAK,GACV,OAEF,GAAI,CAAC,GAAW,GACd,OAGF,GAAM,GAAW,EACjB,OAAW,KAAK,IACd,GAAM,GAAM,EAAS,GACrB,AAAK,EAAK,IAAI,IACZ,GAAK,IAAI,GACT,GAAoB,EAAK,EAAM,KAMrC,YAAoB,GAClB,MAAO,OAAM,QAAQ,IAAQ,MAAO,IAAQ,kHCrF9C,AAmFA,GAAA,IAAA,WAAA,aAEE,KAAA,oBAAwC,GAExC,KAAA,eAAiB,EACjB,KAAA,SAAW,EACX,KAAA,WAAa,EACb,KAAA,iBAAmB,EACnB,KAAA,eAAiB,EAMjB,KAAA,cAAgB,EAGhB,KAAA,YAAc,EAId,KAAA,WAA2B,GAK3B,KAAA,kBAA8B,GAC9B,KAAA,YAAc,EAEd,KAAA,WAAa,GAAI,SAQjB,KAAA,UAAY,GACZ,KAAA,cACI,CAAC,SAAU,EAAG,WAAY,EAAG,UAAW,EAAG,QAAS,GAAI,OAAQ,MAEpE,SAAA,UAAA,QAAA,WACE,OAAW,KAAgB,MAAK,oBAC9B,KAAK,oBAAoB,GAAc,WAG7C,KAEA,GAAA,WAgBE,WAAmB,GAAA,KAAA,IAAA,EAbnB,KAAA,SAA0C,GAC1C,KAAA,gBAKI,GAKI,KAAA,qBAAuB,EAG7B,KAAK,MAAQ,GAAI,IAGb,SAAA,UAAA,MAAN,iHACE,GAAI,KAAK,oBAAsB,KAC7B,MAAA,CAAA,EAAO,KAAK,mBAAmB,KAAK,eAEtC,GAAI,KAAK,iBAAmB,KAC1B,MAAA,CAAA,GAEI,EAAiB,KAAK,oBAEnB,EAAI,yBAAG,GAAI,EAAe,OAC3B,GAAc,EAAe,GACnB,CAAA,EAAM,KAAK,kBAAkB,GAAa,UAFnB,CAAA,EAAA,gBAEjC,GAAU,EAAA,OACZ,EACF,CAAA,EAAM,KAAK,WAAW,IADpB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,UALuC,wBAS3C,KAAM,IAAI,OACN,gFAIN,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,GAAI,KAAK,oBAAsB,KAC7B,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,uHAIlC,GAAI,KAAK,iBAAmB,MACpB,GAAA,GAAA,KAAA,kCAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UACb,GAAI,EACF,KAAM,IAAI,OACN,iCAAiC,EAAI,uHAI3C,KAAK,WAAW,GAElB,MAAO,MAAK,iDAGd,EAAA,UAAA,aAAA,WACE,MAAO,QAAO,KAAK,KAAK,kBAG1B,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,CAAE,KAAe,MAAK,UAGxB,GAAI,IAAe,MAAK,iBACf,GAAA,GAAA,KAAA,kBAAA,GAAA,UACP,GAAI,EAEF,MAAO,UAGT,OAAO,MAGX,MAAO,MAAK,SAAS,IAGvB,EAAA,UAAA,mBAAA,SAAmB,GAEjB,MAAM,KAAe,MAAK,gBAGnB,KAAK,gBAAgB,GAAa,QAFhC,MAKX,EAAA,UAAA,gBAAA,SACI,EACA,EACA,GACF,MADE,AAAA,KAAA,QAAA,GAAA,GACE,IAAe,MAAK,gBACtB,SAAQ,KACD,EAAW,sEAEX,IAET,MAAK,gBAAgB,GAAe,CAAC,QAAO,EAAE,SAAQ,GAC/C,KAGH,EAAA,UAAA,WAAN,SAAiB,2GACf,GAAI,KAAK,gBAAgB,IAAgB,KACvC,KAAM,IAAI,OAAM,iBAAiB,EAAW,iCAE9C,MAAK,YAAc,EACf,KAAK,SAAS,IAAgB,KAChC,MAAK,gBAAkB,KACjB,EAAuB,KAAK,kBAAkB,GAA7C,EAAO,EAAA,QAAE,EAAS,EAAA,UACV,EAAY,CAAA,EAAM,GAAlB,CAAA,EAAA,IAHb,CAAA,EAAA,UAGyB,SAAA,EAAA,oBAAgB,EAAA,mBAC3C,GADM,EAAM,EACR,CAAC,EACH,MAAA,CAAA,EAAO,qBAGX,YAAK,gBAAkB,KAAK,SAAS,GACrC,KAAK,yBAEL,KAAK,SAAW,GAAI,IAAS,KAAK,iBAElC,CAAA,EAAO,UAGD,EAAA,UAAA,uBAAR,WAAA,GAAA,GAAA,KACQ,EAAU,GAAqB,KAAK,aAC1C,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,WAAa,MACtB,EAAO,UAAU,EAAK,oBAKpB,EAAA,UAAA,yBAAR,SAAiC,GAAjC,GAAA,GAAA,KACQ,EAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,aAAe,MACxB,EAAO,YAAY,EAAK,SAAS,OAW/B,EAAA,UAAA,kBAAR,SAA0B,GAA1B,GAAA,GAAA,KAEQ,EAAuB,KAAK,gBAAgB,GAClD,GAAI,GAAwB,KAC1B,KAAM,IAAI,OACN,6BAA6B,EAAW,4BAG9C,IACE,GAAM,GAAU,EAAqB,UAMrC,GAAI,GAAW,CAAE,aAAmB,MAC7B,MAAO,GAAQ,MAAS,YAC7B,GAAM,GAAY,EAAE,KAAK,qBACnB,EACF,EACK,KAAK,SAAA,GAEJ,MAAI,GAAY,EAAK,qBACZ,GAET,GAAK,SAAS,GAAe,EAC7B,EAAK,mBAAqB,KACnB,MAER,MAAM,SAAA,GAEL,MAAI,GAAY,EAAK,sBAGrB,GAAK,mBAAqB,KAC1B,QAAQ,KACJ,6BAA6B,EAAW,WAC5C,QAAQ,KAAK,EAAI,OAAS,EAAI,UACvB,KAEjB,YAAK,mBAAqB,EACnB,CAAC,QAAO,EAAE,UAAW,QAE5B,aAAK,SAAS,GAAe,EACtB,CAAC,QAAS,GAAM,UAAW,UAE7B,GACP,eAAQ,KAAK,6BAA6B,EAAW,WACrD,QAAQ,KAAK,EAAI,OAAS,EAAI,SACvB,CAAC,QAAS,GAAO,UAAW,MAIvC,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAE,KAAe,MAAK,iBACxB,KAAM,IAAI,OAAS,EAAW,kCAEhC,AAAI,KAAK,cAAgB,GAAe,KAAK,oBAAsB,MAGjE,KAAK,uBAGP,AAAI,IAAe,MAAK,UACtB,MAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,IAGvB,MAAO,MAAK,gBAAgB,GAG5B,AAAI,KAAK,cAAgB,GACvB,MAAK,mBAAqB,KAC1B,KAAK,YAAc,KACnB,KAAK,gBAAkB,OAInB,EAAA,UAAA,kBAAR,WAAA,GAAA,GAAA,KACE,GAAI,OAAO,KAAK,KAAK,iBAAiB,SAAW,EAC/C,KAAM,IAAI,OAAM,iCAElB,MAAO,QAAO,KAAK,KAAK,iBAAiB,KAAK,SAAC,EAAW,GAExD,MAAO,GAAK,gBAAgB,GAAG,SAC3B,EAAK,gBAAgB,GAAG,YAIxB,EAAA,UAAA,gCAAR,WAIE,OAFM,GAAiB,KAAK,oBAEnB,EAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAc,EAAe,GAC7B,EAAA,KAAA,kBAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAChB,GAAI,GAAa,EACf,MAAO,CAAC,KAAM,EAAa,UAAS,GAGxC,KAAM,IAAI,OACN,2EAIN,EAAA,UAAA,SAAA,SAAS,EAAwB,GAC/B,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAa,EAAK,QAClB,EAAS,KAAK,SAAS,GAG7B,EAAW,YAAY,GACvB,EAAK,QAAU,EACf,EAAQ,KAAK,EAAQ,EAAQ,EAAK,MAAO,EAAK,OAC9C,AAAI,KAAK,0BAGP,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,MAIvE,EAAA,UAAA,KAAA,SAAgC,EAA6B,GAA7D,GAAA,GAAA,KAEM,EAAe,KACnB,GAAI,GAAM,MAER,GAAI,MAAO,IAAa,WACtB,KAAM,IAAI,OAAM,uCAElB,EAAK,OAGL,GAAI,MAAO,IAAa,UAAY,CAAE,aAAoB,SACxD,KAAM,IAAI,OACN,kFAGN,GAAI,MAAO,IAAO,WAChB,KAAM,IAAI,OACN,kFAGN,EAAO,EAIT,GAAI,GACJ,MAAO,MAAK,UACR,WAAM,MAAA,GAAK,WAAW,IAAO,WAAM,MAAA,GAAK,SAAS,IAAS,WACxD,SAAS,IACT,AAAI,YAAkB,UACpB,QAAQ,MAAM,2CAET,KAIP,EAAA,UAAA,UAAR,SAAqB,EAAmB,EAAiB,GACvD,IACA,IACE,GAAM,GAAM,IACZ,WACO,QACA,GACP,UACM,IAKF,EAAA,UAAA,aAAR,WACE,MAAO,GAAO,gBAIR,EAAA,UAAA,eAAR,WACE,MAAO,GAAO,kBAYR,EAAA,UAAA,MAAR,SAAc,GACZ,GAAM,GAAI,KAAK,qBAAqB,EAAE,OAAQ,EAAE,MAAO,EAAE,OACnD,EAAS,CAAC,EAAC,GACX,EAAO,SAAC,GAAe,MAAC,CAC5B,EAAG,WACD,GAAM,GAAQ,UACR,EAAa,CAAC,EAAG,GACjB,EAAQ,CAAC,MAAK,GAEpB,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAC5B,EAAoC,KAAiB,GACrD,MAGF,EAAkB,GACxB,YAAK,YAAY,KAAK,MAAM,YAAY,KAAM,EAAQ,CAAC,GAAI,EAAM,EAAO,IACjE,GAgBT,EAAA,UAAA,UAAA,SACI,EAAoB,EAAwB,EAC5C,EAAyB,GAC3B,GAAM,GAAoB,KACpB,EAAsB,KAI5B,MAAO,MAAK,cACR,EAAa,EAAQ,EAAe,EAAY,EAAO,EACvD,IAGE,EAAA,UAAA,uBAAR,WACE,MAAO,MAAK,IAAI,QAAQ,YAGlB,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAkB,KAAK,QAAQ,aAGjC,EAAmB,EACvB,EAAS,QAAQ,SAAA,GAGf,GAAqB,EAAK,QAAU,YAAc,EAAI,IAQxD,GAAM,GACF,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,GACjE,EACF,EAAkB,EAAmB,EAAmB,EAC5D,GAAI,EAAgB,EAClB,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,iCAC5B,KAAI,EAAa,6BAA6B,EAAU,OAQhE,EAAA,UAAA,cAAA,SACI,EAA6B,EAC7B,EACA,EAAqB,EAAsB,EAC3C,GAJJ,GAAA,GAAA,KAKM,EACA,EAAkB,GAChB,EAAW,KAAK,WACtB,AAAI,GAAc,MAChB,GACI,KAAK,MAAM,aAAe,KAAO,KAAK,MAAM,YAAY,KAAO,IAGrE,GAAM,GAAoB,KAAK,MAAM,SAC/B,EAAqB,KAAK,MAAM,WAEtC,AAAI,KAAK,0BACP,KAAK,MAAM,kBAAkB,KAAK,GAGpC,GAAI,GACE,EAAS,GAAU,EAAY,KAAK,aACtC,EACJ,GAAI,GAAU,KACZ,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAO,WAAW,CAAC,OAAM,EAAE,MAAK,EAAE,QAAS,EAAK,UACtD,GAAM,GAAW,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC7C,AAAI,EAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAE3D,GAAM,GAAa,EAAS,IACxB,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,MACb,MAAA,GAAK,qBAAqB,EAAQ,EAAO,KAMjD,GAAI,GACF,GAAI,GACA,EAAK,sBAAsB,EAAY,EAAQ,GACnD,GAAI,GAAiB,MAKnB,AAAI,GAAiB,MACnB,GAAgB,IAElB,GAAM,GAAa,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAC7D,EAAiB,IAAgB,IAAI,QAAQ,OAAO,GAEtD,EAAQ,EAAK,2BAA2B,GAE1C,MAAO,SAGT,GAAM,GAAyB,SAAC,GAI9B,GAAI,CAAC,EACH,OAEF,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,OAGrD,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAK,KAAK,WAAM,MAAA,GAAY,EAAK,QAAS,KAChD,GAAM,GAAQ,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC1C,MAAI,GAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAEpD,GAKX,GAAI,GACJ,YAAK,UACD,WAAM,MAAA,GAAK,MAAM,eAAe,WAAM,MAAA,GAAK,MAAM,eAAe,WAC9D,AAAI,CAAC,EAAK,IAAI,QAAQ,UAAY,CAAC,EAAK,MAAM,UAC5C,EAAU,IAEV,GAAgB,EAAK,SAAS,cAC1B,EAAY,EAAQ,WAAM,MAAA,OAC9B,AAAI,EAAK,IAAI,QAAQ,UACnB,EAAK,SAAS,iBAAiB,GAEjC,EAAU,EAAc,WAIhC,AAAI,GACF,KAAK,YACD,EAAY,EAAQ,EAAS,EAAe,EAAO,GAGzD,AAAI,KAAK,MAAM,WACb,KAAK,MAAM,cAAc,QAAQ,KAAK,CACpC,KAAM,EACN,WAAY,KAAK,MAAM,SAAW,EAClC,mBAAoB,KAAK,MAAM,SAC/B,aAAc,KAAK,MAAM,WAAa,EACtC,qBAAsB,KAAK,MAAM,WACjC,YAAa,OAAO,KAAK,GAAQ,IAC7B,SAAA,GAAO,MAAA,GAAO,IAAQ,KAAO,EAAO,GAAK,MAAQ,OACrD,aAAc,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,QACvC,aAAc,EAAc,OAC5B,UAAW,EAAc,YAGrB,MAAM,QAAQ,GAAO,EAAU,EAAQ,IAQzC,EAAA,UAAA,2BAAR,SAAmC,GAAnC,GAAA,GAAA,KACQ,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,MACzD,MAAO,IAaD,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAa,GAAY,GAC/B,GAAI,GAAc,MAChB,GAAM,GAAyB,EAAW,cAAgB,GACpD,EAA2B,EAAW,eAAiB,GAIzD,EAAkB,OACtB,AAAI,EAAW,cACb,GACI,MAAM,QAAQ,GACd,WAAM,MAAA,2DAEV,EAAqB,OAAO,KAAK,GAAQ,IAAI,SAAC,GAAQ,MAAA,GAAO,MAE7D,EAAqB,EAAa,IAAI,SAAC,GAAc,MAAA,GAAO,KAG9D,GAAM,GACF,EAAQ,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAE3C,MAAO,GAAmB,OAAO,GAInC,MAAO,OAQT,EAAA,UAAA,WAAA,SACI,EAAoB,EAAiB,EACrC,GACF,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,iDAElB,EAAQ,GAAS,UACjB,EAAU,GAAW,KAAK,QAC1B,GAAI,GAAc,EAClB,AAAI,IAAU,UAAY,GAAc,EAAO,KAC7C,GAAe,EAAoB,IAAI,SAAA,GAAK,MAAA,IAAkB,MAEhE,GAAM,GAAS,EAAQ,MAAM,EAAa,EAAO,GAC3C,EAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAIhD,GAHA,KAAK,OAAO,EAAG,GAGX,IAAU,UACZ,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAW,GAAqB,GACtC,KAAK,MAAM,UAAY,EAAW,EAAK,MACvC,EAAK,MAAQ,EAEf,MAAO,IAQT,EAAA,UAAA,qBAAA,SACI,EAAgB,EAAiB,EACjC,GACF,EAAQ,GAAS,UACjB,GAAM,GAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAChD,YAAK,OAAO,EAAG,GACR,GAGT,EAAA,UAAA,aAAA,SACI,EAAsB,EAAkB,EACxC,GADsB,AAAA,IAAA,QAAA,GAAA,IAExB,EAAO,GAAQ,KAAK,iBAAiB,WACrC,AAAI,GAAS,MAAQ,IAAU,EAAa,OAC1C,GAAe,EAAa,KAAK,IAEnC,GAAM,GAAI,GAAI,IAAS,EAAc,EAAW,EAAM,KAAK,gBAC3D,GAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,KAC5C,KAAM,IAAI,OAAM,sBAAsB,EAAE,KAAI,2BAE9C,YAAK,MAAM,oBAAoB,EAAE,MAAQ,EACzC,KAAK,OAAO,EAAG,KAAK,SACb,GAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAM,GAAW,KAAK,MAAM,WAAW,IAAI,EAAE,QACzC,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,SACpC,EAKJ,GAJA,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAET,IAAa,GACf,KAAK,MAAM,iBAIX,GAAI,GAAQ,EACZ,AAAI,EAAE,QAAU,aAAe,EAAE,QAAU,UACzC,GAAQ,EAAE,KAAO,GAAqB,EAAE,QAE1C,KAAK,MAAM,WAAW,IAAI,EAAE,OAAQ,CAClC,QAAS,GAAW,KAAK,QACzB,MAAO,EAAE,MACT,MAAO,EAAE,MACT,MAAK,EACL,SAAU,IAEZ,KAAK,MAAM,UAAY,EAGzB,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,WAEpC,AAAM,YAAa,KACjB,KAAK,MAAM,IAIf,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAC,KAAK,MAAM,WAAW,IAAI,EAAE,QAC/B,OAGF,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAEb,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,EAAE,QACnC,EAAW,EAAK,SAEtB,AAAI,GAAY,EAGd,CAAI,EAAE,QAAU,aACd,MAAK,MAAM,UAAY,EAAK,OAE9B,KAAK,MAAM,iBAEX,EAAK,QAAQ,YAAY,EAAE,QAC3B,KAAK,MAAM,WAAW,OAAO,EAAE,SAE/B,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,YAOxC,EAAA,UAAA,iBAAA,WACE,OAAW,KAAW,MAAK,MAAM,qBAC/B,GAAM,GAAI,KAAK,MAAM,oBAAoB,GACzC,KAAK,gBAAgB,KAIzB,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,cAAc,GACnB,AAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,MAC5C,MAAO,MAAK,MAAM,oBAAoB,EAAE,OAI5C,EAAA,UAAA,OAAA,WACE,GAAM,GAAO,KAAK,QAAQ,SAC1B,SAAK,WAAa,KAAK,MAAM,WAC7B,EAAK,eAAiB,KAAK,MAAM,eACjC,EAAK,SAAW,KAAK,MAAM,SAC3B,AAAI,KAAK,MAAM,iBAAmB,GAChC,GAAK,WAAa,GAClB,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,IAEjB,EAAK,QAAQ,KACT,0EAGC,GAGH,EAAA,UAAA,QAAN,SAAc,iHAEZ,YAAK,MAAM,UAAY,GAEjB,EAAa,KAAK,MAAM,SACxB,EAAkB,KAAK,MAAM,WAEnC,KAAK,MAAM,cAAc,QAAU,GACnC,EAAA,KAAK,MAAM,cAAuB,CAAA,EAAM,YAAxC,EAAyB,OAAS,EAAA,OAElC,KAAK,MAAM,UAAY,GAEvB,KAAK,MAAM,cAAc,UAAY,KAAK,IAAG,MAAR,KAC9B,KAAK,MAAM,cAAc,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,sBACnD,KAAK,MAAM,cAAc,SAAW,KAAK,MAAM,SAAW,EAC1D,KAAK,MAAM,cAAc,WACrB,KAAK,MAAM,WAAa,MACP,EAAA,KAAK,MAAM,cAAc,+BAAzB,GAAA,EAAA,OAAV,GAAM,EAAA,GACf,EAAA,EAAsB,CAAA,EAAM,EAAO,eADgB,CAAA,EAAA,UACnD,SAAO,aAAe,EAAA,OACtB,EAAA,EAAmB,CAAA,EAAM,EAAO,kBAAhC,EAAO,UAAY,EAAA,wBAFA,wBAIrB,MAAA,CAAA,EAAO,KAAK,MAAM,qBAGpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,MAAM,cAAgB,GAAK,KAAK,MAAM,cAAgB,GAG5D,EAAA,UAAA,YAAR,SACI,EAAoB,EAAwB,EAC5C,EAAyB,EAAiB,GAF9C,GAAA,GAAA,KAGQ,EACF,CAAC,GAAI,KAAK,MAAM,iBAAkB,WAAU,EAAE,OAAM,EAAE,QAAO,EAAE,MAAK,GAElE,EAAa,GAAY,GAC/B,AAAI,GAAc,MAChB,GAAgB,EAAW,UAE7B,AAAI,GAAiB,MACnB,GAAS,SAAW,SAAC,GAGnB,SAAM,EAAI,IAAI,SAAC,EAAI,GACjB,GAAI,GAAM,MACR,GAAM,GAAS,EAAQ,GACjB,EAAO,GAAyB,EAAO,KAAM,EAAO,OAC1D,MAAO,GAAK,WAAW,EAAM,EAAO,MAAO,EAAO,OAEpD,MAAO,KAIF,EAAc,EAAI,OAAS,EAAI,EAAM,EAAI,GAAI,EAAO,KAG/D,KAAK,MAAM,WAAW,KAAK,IAG7B,EAAA,UAAA,KAAA,SAAuB,GACrB,SAAO,KAAO,GACP,GAGD,EAAA,UAAA,UAAR,WACE,AAAI,KAAK,MAAM,gBAAkB,GAC/B,MAAK,MAAM,WAAa,IAE1B,KAAK,MAAM,iBAGL,EAAA,UAAA,QAAR,WACE,KAAK,MAAM,iBAOb,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAwB,CAC5B,MAAO,GACP,KAAM,gBACN,GAAI,KAAK,MAAM,eAEjB,AAAI,GACF,GAAU,KAAO,GAEnB,KAAK,MAAM,WAAW,KAAK,GAC3B,KAAK,MAAM,YAAc,GAO3B,EAAA,UAAA,SAAA,SAAS,GAMP,OANF,GAAA,KACQ,EAAyB,GAAsB,GAC/C,EACF,GAAI,KAAI,EAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,MAGrC,EAAI,EAAG,EAAI,KAAK,MAAM,YAAY,MAAM,OAAQ,KACvD,GAAM,GAAS,KAAK,MAAM,YAAY,MAAM,GAC5C,AAAI,CAAC,EAAO,MAAQ,CAAC,EAA0B,IAAI,EAAO,KACxD,EAAO,UAIX,GAAM,GAAW,KAAK,MAAM,WAAW,MACvC,KAAK,MAAM,YAAc,KAAK,MAAM,WAAW,SAAW,EACtD,KACA,KAAK,MAAM,WAAW,KAAK,MAAM,WAAW,OAAS,GAGzD,EAAuB,QAAQ,SAAA,GAG7B,AAAI,CAAC,EAAO,MAAQ,EAAO,UAAY,EAAS,IAC9C,EAAK,MAAM,MAWjB,EAAA,UAAA,UAAA,SACI,EAAY,EAAc,EAC1B,GAFJ,GAAA,GAAA,KAKE,GAHE,AAAA,IAAA,QAAA,GAAA,IACF,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,8CACrB,GAAM,MAAQ,EAAG,QAAU,UAC7B,KAAM,IAAI,OAAM,0CAA0C,EAAG,MAAK,KAGpE,GAAM,GAAI,KAAK,UACX,WAAM,MAAA,GAAK,aAAa,WAAM,MAAA,GAAK,WACnC,WAAM,MAAA,GAAK,KAAK,UAAW,KAE/B,EACI,YAAa,GACb,WAAM,MAAA,mDAEV,GAAM,GAAe,GAAqB,KAAK,MAAM,WAAY,EAAI,GACrE,GAAI,CAAC,GAAoB,EAAa,SAAW,GAAK,EAAG,OAAS,EAChE,KAAM,IAAI,OACN,uIAKN,MAAO,MAAK,KAAK,WAAY,WAC3B,GAAM,GAAuD,GAC7D,EAAuB,EAAE,IAAO,GAAc,GAAK,EAAE,OAGrD,GACI,EAAwB,EAExB,SAAA,GAAK,MAAA,GAAK,KAAK,IAEf,IACJ,GAAM,GAAQ,EAAG,IAAI,SAAA,GAAK,MAAA,GAAuB,EAAE,MAEnD,MAAI,GAAK,MAAM,gBAAkB,GAG/B,GAAK,MAAM,WAAW,QAAQ,SAAA,GAC5B,OAAqB,GAAA,EAAA,EAAA,EAAK,MAAL,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,EAAO,aAGX,EAAK,MAAM,WAAa,MAEnB,CAAC,MAAO,EAAG,MAAK,MAI3B,EAAA,UAAA,WAAA,SAA6B,GAA7B,GAAA,GAAA,KAEE,SACI,GAAgB,GAChB,WAAM,MAAA,sDACH,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACN,EACI,EAAO,MAAM,SAAA,GAAK,MAAA,aAAa,KAC/B,WAAM,MAAA,qEAGV,GAAI,GAIE,EAA2B,GACjC,SAAO,QAAQ,SAAC,EAAO,GACrB,EAAS,GAAK,IAET,EAAK,cACR,SAAC,EAAG,GACF,SAAM,EAAC,MAAA,OAAQ,EAAM,OAAA,CAAE,KACvB,EACI,EAAI,gBAAiB,GACrB,WAAM,MAAA,+FAEV,EACI,GAAgB,EAAI,UACpB,WAAM,MAAA,qGAEH,EAAI,OAEb,EACA,SAAC,EAAO,GACN,GAAM,GAAU,EAAI,SAAS,EAAI,GAC3B,EACF,MAAM,QAAQ,GAAW,EAAU,CAAC,GACxC,EACI,EAAM,SAAW,EAAO,OACxB,WAAM,MAAA,wKAGV,EACI,EAAM,MAAM,SAAA,GAAK,MAAA,aAAa,KAC9B,WAAM,MAAA,yIAGV,GAAM,GAAyC,GAC/C,SAAM,QAAQ,SAAC,EAAM,GACnB,EAAQ,GAAK,WAAM,MAAA,MAEd,MAKjB,EAAA,UAAA,SAAA,SAAS,GAEP,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,SAAS,IAE/B,EAAA,UAAA,KAAA,SAAK,GAEH,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,KAAK,IAGrB,EAAA,UAAA,KAAN,SAAW,qGACH,SAAQ,KACK,CAAA,EAAM,KAAK,QAAQ,KAAK,WAArC,SAAa,EAAA,OACnB,EAAW,OAAS,KAAQ,EAC5B,CAAA,EAAO,SASD,EAAA,UAAA,MAAR,SAAgC,GAC9B,MAAI,MAAK,MAAM,aAAe,MAC5B,GAAO,QAAU,KAAK,MAAM,YAAY,GACxC,KAAK,MAAM,YAAY,MAAM,KAAK,IAG7B,GAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAOpB,EAAA,UAAA,MAAA,WAEE,KAAK,uBAEL,KAAK,MAAM,UACX,KAAK,IAAI,QACT,KAAK,MAAQ,GAAI,IAEjB,OAAW,KAAe,MAAK,SAC7B,KAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,GAEvB,KAAK,YAAc,KACnB,KAAK,gBAAkB,KACvB,KAAK,mBAAqB,MA/sBb,EAAA,aAAe,EAKf,EAAA,eAAiB,EA4sBlC,KAEA,YAAc,GACZ,GAAM,GAAS,GAAmB,GAAc,GAAQ,WACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,yBAIxC,GAAM,GAAK,KACX,GAAI,EAAG,WAAa,MAClB,GAAM,GAAc,GAAI,IAAY,GACpC,EAAG,UAAY,GAAI,IAAO,GAE5B,UAAqB,EAAG,UAAU,KAIlC,GAAiB,WAAM,MAAA,GAAG,YACnB,EAAG,UAGL,GAAM,GAAS,iBAQF,EAAW,GAE7B,GAAM,GAAS,CAAC,EAAC,EAAE,EAAC,GACpB,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,EAAG,GAC3B,SAAK,CAAC,EAAG,IACF,GACN,EAAgC,KAAqB,ICzrC1D,AAkBA,cACE,MAAO,OAAO,YAAc,aAAe,WAAa,mBAIxD,GAAI,MAEF,GAAM,GAAI,UAAU,WAAa,UAAU,QAAW,OAAe,MAErE,MAAO,2TACK,KAAK,IAEb,0kDACK,KAAK,EAAE,OAAO,EAAG,IAE5B,MAAO,iBAIP,MAAQ,OAAO,SAAW,aAAe,OAAO,UAAY,MAEvD,MAAO,oBAAsB,6DCvCpC,AAqBA,GAAM,IAAM,KAOZ,GAAI,aAAa,QAAS,WAAM,MAAA,IAAO,SAAA,GACrC,AAAI,GACF,QAAQ,KACJ,iJAOR,GAAI,aAAa,aAAc,WAAM,MAAA,QAGrC,GAAI,aACA,UACA,WAAM,MAAC,OAAO,UAAY,aACrB,MAAO,SAAQ,UAAa,aAC5B,MAAO,SAAQ,SAAS,MAAS,cAG1C,GAAI,aACA,YACA,WAAM,MAAA,OAAO,YAAc,aAAe,WAAa,MACnD,UAAU,WAAa,MAAQ,SAAS,KAAK,UAAU,YACvD,aAAa,KAAK,UAAU,UAMpC,GAAI,aAAa,OAAQ,WAAM,MAAA,KAM/B,GAAI,aACA,qCAAsC,WAAM,MAAA,IAAI,QAAQ,WAG5D,GAAI,aAAa,+BAAgC,WAAM,MAAA,KAGvD,GAAI,aAAa,UAAW,WAAM,MAAA,KCvElC,YAuB2B,EAAiB,GAC1C,GAAI,GAAwB,EAE5B,GAAI,GAAa,GACf,MAAO,KAAU,SAAW,GAAK,CAAC,EAAI,QAExC,GAAI,CAAC,MAAM,QAAQ,GACjB,MAAO,GAIT,OAFM,GAAkB,GAEjB,MAAM,QAAQ,IACd,GAAa,IAAc,IAAU,UAC1C,EAAM,KAAK,EAAU,QACrB,EAAY,EAAU,GAExB,MAAI,OAAM,QAAQ,IACd,KAAM,QAAQ,uCAChB,GAA2B,EAAK,EAAO,IAGlC,EAGT,YACI,EAAiB,EAAiB,GAEpC,GADA,EAAU,GAAW,GACjB,CAAE,MAAM,QAAQ,IAAS,CAAC,GAAa,IACzC,EACI,EAAM,SAAW,EACjB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,qBACnC,yCAAwC,EAAM,GAAE,eACxD,OAEF,EACI,EAAM,OAAS,EACf,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,4BACnC,uBAAsB,EAAI,OAAM,eACxC,EACI,EAAI,SAAW,EAAM,GACrB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,iBAAiB,EAAM,GAAE,IAC5D,sBAAqB,EAAI,OAAM,eAEvC,OADM,GAAW,EAAM,MAAM,GACpB,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAA2B,EAAI,GAAI,EAAU,EAAQ,OAAO,IAIhE,YACI,EAAmC,EAAuB,EAC1D,GACF,GAAI,GAAiB,KACnB,OAEF,GAAI,IAAkB,WAAa,IAAkB,GACjD,IAAkB,WAAa,IAAgB,SACjD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,UAChD,OAAM,EAAa,oBAAoB,EAAW,uBAKtD,EAAiB,EAAiB,EAClC,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,YAAa,GACf,UAAY,EAAc,EAAE,MAAO,EAAS,GACrC,EAET,GAAI,GAAgB,GAAW,GAS/B,GANA,AAAI,IAAkB,UAClB,CAAC,OAAQ,QAAS,WAAW,QAAQ,IAAiB,GACxD,GAAgB,GAElB,GAAY,EAAc,EAAe,EAAS,GAE7C,GAAK,MACL,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IAAM,MAAO,IAAM,UACtD,MAAO,IAAM,WAAa,MAAO,IAAM,UAC1C,GAAM,GAAO,GAAK,KAAO,OAAU,EAAS,YAAY,KACxD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,eAChD,mCAAkC,EAAI,MAE5C,GAAM,GAAgB,GAAW,EAAG,GACpC,AAAI,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IACrC,GAAI,CAAC,IAEP,GAAM,GAAiB,GACjB,EAAS,IAAkB,SAC7B,GAAa,EAAG,GAChB,GAAQ,EAAe,GAAI,GAC/B,MAAO,GAAO,WAAW,EAAQ,EAAe,eAI9C,EAA0B,EAAiB,EAC3C,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,OACN,YAAY,EAAO,cAAc,EAAY,2CAGnD,GAAM,GAAU,EAChB,MAAO,GAAQ,IACX,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAM,EAAO,IAAI,EAAC,IAAK,IACjD,GCnIN,GAkBa,IAAkB,kBAOQ,GACrC,GAAM,GAAO,OAAO,KAAK,GACzB,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,OACN,yGAEG,GAAK,OAAM,WAGpB,GAAI,GAAS,EAAK,GACZ,EAAK,EAAE,GAGb,AAAI,EAAO,SAAS,MAClB,GAAS,EAAO,UAAU,EAAG,EAAO,OAAS,IAI/C,EAAS,EAAS,GAGlB,GAAM,GAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,EAAO,WAAW,GAClB,IACE,GAAM,GAAS,EAAE,MAAA,OAAI,GACrB,MAAI,aAAkB,UACpB,QAAQ,MAAM,2CAEhB,EAAO,SAAS,GACT,QACA,GACP,QAAO,SAAS,MACV,IAGV,cAAO,eAAe,EAAI,OAAQ,CAAC,MAAO,EAAQ,aAAc,KAGzD,EC/DT,AA8CA,YAAoC,EAAoB,GACtD,GAAM,GAAQ,EAAgB,EAAM,OAAQ,WACtC,EAAQ,EAAgB,EAAM,OAAQ,WAC5C,GACI,EAAM,MAAO,EAAM,MACnB,yBAAyB,EAAM,MAAK,QAAQ,EAAM,MAAK,yCAG3D,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,QAAQ,EAAO,IAE1B,EAAwB,CAAC,KAAM,EAAO,KAAM,GAClD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC/DnC,YAyBI,EAAoB,EAAiB,EACrC,GAIF,GAHA,AAAI,GAAS,MACX,GAAQ,GAAW,IAEjB,IAAU,YACZ,KAAM,IAAI,OACN,oFAGN,GAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IACxC,MAAO,IAAW,UAAY,MAAO,IAAW,WAChD,MAAO,IAAW,SACpB,KAAM,IAAI,OACN,4HAGN,GAAI,GAAS,MACX,GAAmC,GAEnC,GAAM,GAAe,GAAc,GAC7B,EAAe,GAAc,GACnC,EACI,IAAiB,EACjB,WACI,MAAA,iCAAiC,EAAK,6BACnC,GAAY,mBAAmB,KAE1C,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,EAAE,GAC1C,GAAM,GAAW,EAAc,GACzB,EAAoB,IAAM,EAAc,OAAS,EACnD,IAAa,GAAc,EAAM,MAAM,IACvC,GACJ,EACI,EAAc,KAAO,EAAM,IAAM,CAAC,EAClC,WAAM,MAAA,+CACF,KAAI,EAAa,kCACjB,WAAU,EAAK,UAI3B,MAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IAC1C,GAAS,CAAC,IAGZ,EAAQ,GAAS,EACjB,EAAS,IAAU,SACf,GAAa,EAAQ,GACrB,GAAQ,EAAoB,GAAI,IAC7B,EAAO,WAAW,EAAsB,EAAO,GC1ExD,YAqDI,EAAoB,EAAqB,GAC3C,GAAM,GAAgB,GAAW,EAAQ,GACzC,MAAO,IAAW,EAAQ,EAAO,EAAe,GCvDlD,AAsBO,GAAM,IAAkD,CAC7D,QAAW,EACX,QAAW,EACX,MAAS,EACT,OAAU,EACV,MAAS,EACT,KAAQ,EACR,UAAa,GC7Bf,AA2BA,GAAM,IAA0B,cAoB5B,EAAuC,oHAUzC,IAPM,EAAgC,GAChC,EAA2C,GAE3C,EAAkB,MAAM,QAAQ,GAClC,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,OAC7B,OAAO,KAAK,cAEP,GACP,GAAM,GAAO,EAAM,GACb,EAAI,MAAM,QAAQ,GAAW,EAAQ,GAAG,OAAS,EAAQ,GAC/D,GAAI,EAAE,QAAU,WAAa,EAAE,QAAU,SAAW,EAAE,QAAU,QAC5D,EAAE,QAAU,UAAY,EAAE,QAAU,YACtC,KAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,EAAE,OAE9D,GAAM,GAA6B,CAAC,KAAI,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,OACnE,GAAI,EAAE,QAAU,UACd,GAAM,GAAY,GAAI,SAAoB,SAAM,GAAO,MAAA,IAAA,EAAA,OAAA,OAAA,+EACxC,MAAA,CAAA,EAAM,EAAE,gBAKrB,IALM,EAAO,EAAA,OACP,EAAgB,EAAK,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAE,QAAQ,GACtD,GAA0B,EAAK,OAC7B,EAAQ,GAAI,YAAW,GACzB,EAAS,EACJ,EAAI,EAAG,EAAI,EAAK,OAAQ,IACzB,EAAM,EAAK,GACX,EACF,GAAI,YAAW,GAAI,aAAY,CAAC,EAAI,SAAS,QACjD,EAAM,IAAI,EAAe,GACzB,GAAU,GACV,EAAM,IAAI,EAAK,GACf,GAAU,EAAI,OAEhB,SAAQ,cAEV,EAAa,KAAK,OAElB,GAAa,KAAK,EAAE,QAEtB,AAAI,GAAS,MACX,GAAK,MAAQ,GAEf,EAAM,KAAK,IAjCJ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,IAA3B,GAoCY,MAAA,CAAA,EAAM,QAAQ,IAAI,WAAjC,SAAe,EAAA,OACrB,CAAA,EAAO,CAAC,KAAM,GAAuB,GAAe,MAAK,qBAmBvD,EAAqB,GAKvB,OAHM,GAAsB,GACxB,EACA,EAAS,EACM,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAO,EAAK,KACZ,EAAQ,EAAK,MACb,EAAQ,EAAK,MACb,EAAO,GAAc,GACvB,EAAM,OAEV,GAAI,gBAAkB,IACpB,GAAM,GAAe,EAAK,aAC1B,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,GAAI,CAAE,QAAS,IAAgB,SAAW,IACxC,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,+DAGxD,EAAa,QAAU,WAChC,GAAI,IAAU,UACZ,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,IAC3D,oDAAmD,EAAK,UAG9D,MAAM,IAAI,OACN,UAAU,EAAK,KAAI,gBACnB,uBAAsB,EAAa,MAAK,MACxC,wEAGN,GAAM,GAAyB,GAAqB,EAAa,OAC3D,EACF,EAAO,MAAM,EAAQ,EAAS,EAAO,GACnC,EAAkB,EAAa,QAAU,QAC3C,GAAI,YAAW,GACf,GAAI,aAAY,GACpB,GAAI,IAAU,UACZ,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,EAAS,GAAI,cAAa,EAAe,QACzC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,EAAI,EAAa,MAAQ,EAAa,aAE3C,EAAa,QAAU,UAChC,AAAI,IAAkB,QACpB,GAAgB,MAElB,EAAS,EAAc,OAEvB,MAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,qCAGhD,IAAU,SACnB,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,SAC3D,KAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,2BAGzD,EAAS,GAAI,YAAW,EAAe,QACvC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,KAAK,MAAM,EAAI,EAAa,MAAQ,EAAa,UAG/D,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,UACR,IAAU,UACnB,GAAM,GAAO,GAAc,EAAK,OAChC,EAAS,GACT,OAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAa,GAAI,aACnB,EAAO,MAAM,EAAQ,EAAS,KAA0B,GAC5D,GAAU,GACV,GAAM,GAAQ,GAAI,YAAW,EAAO,MAAM,EAAQ,EAAS,IAC1D,EAAwB,KAAK,GAC9B,GAAU,QAGZ,GAAM,GAAc,GAAqB,GACnC,EAAa,EAAO,MAAM,EAAQ,EAAS,EAAO,GAExD,GAAI,IAAU,UACZ,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,aACnB,EAAS,GAAI,cAAa,GAG1B,OAFM,GAAO,GAAI,cAAa,EAAO,OAAS,GACxC,EAAQ,GAAI,cAAa,EAAO,OAAS,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAK,GAAK,EAAO,EAAI,GACrB,EAAM,GAAK,EAAO,EAAI,EAAI,GAE5B,GAAM,GAAa,GAAO,EAAM,EAAO,WACjC,EAAc,GAAO,EAAO,EAAO,WACzC,EAAI,GAAQ,GAAQ,EAAY,OAEhC,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,EAEnB,AAAI,IAAU,aACZ,GAAI,GAAQ,GAAO,EAAQ,EAAO,IAGtC,MAAO,eAM8B,GAErC,GAAI,IAAO,KACT,KAAM,IAAI,OAAM,wBAAwB,KAAK,UAAU,IAGzD,GAAI,GAAkB,EAShB,EAA6B,GACnC,EAAG,QAAQ,SAAC,GAMV,GALA,GAAmB,EAAE,WAErB,EAAa,KACT,EAAE,aAAe,EAAE,OAAO,WAAa,EACA,GAAK,GAAE,YAAoB,IAClE,CAAE,aAAoB,eAAgB,YAAoB,aACxD,YAAoB,aACxB,KAAM,IAAI,OAAM,mCAAmC,EAAE,YAAY,QAKrE,GAAM,GAAI,GAAI,YAAW,GACrB,EAAS,EACb,SAAa,QAAQ,SAAC,GACpB,EAAE,IAAI,GAAI,YAAW,EAAE,QAAS,GAChC,GAAU,EAAE,aAGP,EAAE,OAIX,GAAM,IAAgB,MAAO,SAAW,aACnC,OAAO,OAAS,aAAe,MAAO,OAAS,aAC/C,MAAO,OAAS,yBAWY,GAC/B,MAAI,IACK,OAAO,WAAW,GAEpB,GAAI,MAAK,CAAC,IAAM,iBASiB,GACxC,GAAI,GACF,MAAO,QAAO,KAAK,GAAQ,SAAS,UAItC,OAFM,GAAM,GAAI,YAAW,GACvB,EAAI,GACC,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAI,EAAG,IACrC,GAAK,OAAO,aAAa,EAAI,IAE/B,MAAO,MAAK,eAS4B,GACxC,GAAI,IACF,GAAM,GAAM,OAAO,KAAK,EAAK,UAC7B,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YAI/D,OAFM,GAAI,KAAK,GACT,EAAS,GAAI,YAAW,EAAE,QACvB,EAAI,EAAG,EAAI,EAAE,OAAQ,EAAE,EAC9B,EAAO,IAAI,CAAC,EAAE,WAAW,IAAK,GAEhC,MAAO,GAAO,mBASwB,GACtC,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAI,GAAkB,EACtB,EAAQ,QAAQ,SAAC,GACf,GAAmB,EAAO,aAG5B,GAAM,GAAO,GAAI,YAAW,GACxB,EAAS,EACb,SAAQ,QAAQ,SAAC,GACf,EAAK,IAAI,GAAI,YAAW,GAAS,GACjC,GAAU,EAAO,aAEZ,EAAK,mBAUW,GACvB,GAAM,GAAY,IAElB,IADA,EAAO,EAAK,OACL,EAAK,SAAS,IACnB,EAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAErC,GAAM,GAAQ,EAAK,MAAM,GACzB,MAAO,GAAM,EAAM,OAAS,eAQe,GAE3C,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OAAM,uDAGlB,MAAO,CACL,UAAW,GAAI,MACf,kBAAmB,OACnB,mBAAoB,EAAe,eAAiB,KAChD,EACA,GAAiB,KAAK,UAAU,EAAe,gBACnD,iBAAkB,EAAe,aAAe,KAC5C,EACA,GAAiB,KAAK,UAAU,EAAe,cACnD,gBAAiB,EAAe,YAAc,KAC1C,EACA,EAAe,WAAW,YAUlC,cACE,GAAM,GAAkB,SAAC,GAIvB,OAHI,GAAI,GAAK,GACT,EAAI,EAEA,GAAI,WAAgB,GAC1B,GAAK,QACL,IAAM,EAER,UAAK,CAAC,QACN,GAAK,UAEE,EAAI,GAGP,EAAe,GAAI,aAAY,MAErC,EAAa,GAAK,EAClB,OAAS,GAAI,EAAG,EAAI,KAAM,IACxB,EAAa,GAAK,EAAgB,GAEpC,OAAS,GAAI,KAAM,EAAI,KAAM,IAC3B,EAAa,GAAK,UAAe,GAAI,MAAS,IAGhD,MAAO,GAST,cACE,GAAM,GAAgB,GAAI,aAAY,IAEtC,EAAc,GAAK,EACnB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,OAAS,GAAI,EAAG,EAAI,GAAI,IACtB,EAAc,GAAK,GAAK,GAE1B,OAAS,GAAI,GAAI,EAAI,GAAI,IACvB,EAAc,GAAK,WAAe,GAAI,IAAO,IAG/C,MAAO,GAST,cAGE,OAFM,GAAc,GAAI,aAAY,IAE3B,EAAI,EAAG,EAAI,GAAI,IACtB,EAAY,GAAK,KAEnB,SAAY,GAAK,EAAY,IAAM,EAE5B,gBAeP,GAAM,GAAe,KACf,EAAgB,KAChB,EAAc,KAEpB,MAAO,UAAC,GAGN,OAFM,GAAS,GAAI,aAAY,EAAI,EAAe,QAC5C,EAAmB,GAAI,aAAY,GAChC,EAAQ,EAAG,EAAQ,EAAe,OAAQ,KACjD,GAAM,GAAc,EAAe,GAC7B,EACF,EAAa,EAAY,GAAe,IAAO,GAAc,OAC7D,EAAc,GAAe,IACjC,EAAiB,GAAS,EAE5B,MAAO,IAAI,cAAa,IChf5B,AAsBA,GAAA,IAAA,WAOE,aACE,KAAK,YAAc,GACnB,KAAK,YAAc,GAGN,SAAA,YAAf,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UASnB,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAS3C,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAW3C,EAAA,gBAAP,SAAuB,GACrB,MAAO,GAAiB,YAAY,EAAK,SAWpC,EAAA,gBAAP,SAAuB,EAAsB,GAE3C,MAAO,GAAiB,YAAY,EAAK,OAAQ,IAGpC,EAAA,YAAf,SACI,EAAsB,EACtB,GACF,GAAM,GAA6B,GAC7B,EAAU,IAAgB,OAC5B,EAAiB,cAAc,YAC/B,EAAiB,cAAc,YACnC,SAAQ,QAAQ,SAAA,GACd,GAAM,GAAU,EAAO,EAAK,GAC5B,AAAI,IAAY,MACd,EAAc,KAAK,KAGhB,GAEX,KAEa,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAkB,SAAC,GAC5B,MAAA,IAAiB,gBAAgB,IACxB,GACT,SAAC,EAAsB,GACnB,MAAA,IAAiB,gBAAgB,EAAK,IC/G9C,AAyBA,GAAM,IAAgB,eAChB,GAAmB,EAKnB,GAAmB,eAInB,GAAkB,mBAexB,cACE,GAAI,CAAC,KAAM,QAAQ,cAIjB,KAAM,IAAI,OACN,2FAIN,GAAM,GAAiB,MAAO,SAAW,YAAc,KAAO,OACxD,EAAU,EAAU,WAAa,EAAU,cAC7C,EAAU,iBAAmB,EAAU,aACvC,EAAU,cACd,GAAI,GAAW,KACb,KAAM,IAAI,OACN,6DAEN,MAAO,GAGT,YAAuB,GACrB,GAAM,GAAK,EAAY,OACvB,EAAG,kBAAkB,GAAkB,CAAC,QAAS,cACjD,EAAG,kBAAkB,GAAiB,CAAC,QAAS,cAQlD,GAAA,IAAA,WAME,WAAY,GAGV,GAFA,KAAK,UAAY,KAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,kEAEN,KAAK,UAAY,EAGb,SAAA,UAAA,KAAN,SAAW,sEAET,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAIN,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,UAAW,SAIvC,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,iBAiB1B,EAAA,UAAA,eAAR,SAAuB,EAAmB,GAA1C,GAAA,GAAA,KAEE,MAAO,IAAI,SAAmC,SAAC,EAAS,GACtD,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OAEvB,GAAI,GAAkB,MAEpB,GAAM,GAAU,EAAG,YAAY,GAAkB,YAC3C,EAAa,EAAQ,YAAY,IACjC,EAAa,EAAW,IAAI,EAAK,WACvC,EAAW,UAAY,WACrB,GAAI,EAAW,QAAU,KACvB,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAK,UAAS,oBAGlD,EAAQ,EAAW,OAAO,iBAG9B,EAAW,QAAU,SAAA,GACnB,SAAG,QACI,EAAO,EAAW,QAE3B,EAAQ,WAAa,WAAM,MAAA,GAAG,cAG9B,GAAM,GACF,GAA6B,GAE3B,EAAS,EAAG,YAAY,GAAiB,aAC3C,EAAY,EAAO,YAAY,IAC7B,EACF,EAAU,IAAI,CAAC,UAAW,EAAK,UAAW,mBAAkB,IAC5D,EACJ,EAAe,UAAY,WAEzB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAkB,EAAW,IAAI,CACrC,UAAW,EAAK,UAChB,eAAc,EACd,mBAAkB,IAEpB,EAAgB,UAAY,WAAM,MAAA,GAAQ,CAAC,mBAAkB,KAC7D,EAAgB,QAAU,SAAA,GAGxB,EAAY,EAAO,YAAY,IAC/B,GAAM,GAAoB,EAAU,OAAO,EAAK,WAChD,EAAkB,UAAY,WAC5B,SAAG,QACI,EAAO,EAAgB,QAEhC,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAgB,UAIpC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAE/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,YAKtC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,WAvHtC,EAAA,WAAa,eA0H/B,KAEa,GAA4B,SAAC,GACxC,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAmBH,GAC/B,MAAO,IAAI,IAAiB,GAG9B,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAiB,YACnC,EAAI,MAAM,GAAiB,WAAW,QACtC,EAGN,GAAA,IAAA,WAGE,aACE,KAAK,UAAY,KAGb,SAAA,UAAA,WAAN,yFACE,MAAA,CAAA,EAAO,GAAI,SACP,SAAC,EAAS,GACR,GAAM,GACF,EAAK,UAAU,KAAK,GAAe,IACvC,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAK,EAAG,YAAY,GAAiB,YACrC,EAAQ,EAAG,YAAY,IASvB,EAAqB,EAAc,SACzC,EAAkB,UAAY,WAE5B,OADM,GAA4C,GAC/B,EAAA,EAAA,EAAA,EAAkB,OAAlB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAI,EAAK,WAAa,EAAK,mBAE7B,EAAQ,IAEV,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAkB,QAElC,EAAG,WAAa,WAAM,MAAA,GAAG,UAE3B,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAIpD,EAAA,UAAA,YAAN,SAAkB,iFAChB,SAAO,GAAiB,GACxB,CAAA,EAAO,GAAI,SAA4B,SAAC,EAAS,GAC/C,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAS,EAAG,YAAY,GAAiB,aACzC,EAAY,EAAO,YAAY,IAE/B,EAAiB,EAAU,IAAI,GACjC,EACJ,EAAe,UAAY,WACzB,GAAI,EAAe,QAAU,KAC3B,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAI,oBAIxC,GAAM,GAAoB,EAAU,OAAO,GACrC,EAAkB,WAEtB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAqB,EAAW,OAAO,GAC7C,EAAmB,UAAY,WAC3B,MAAA,GAAQ,EAAe,OAAO,qBAClC,EAAmB,QAAU,SAAA,GACzB,MAAA,GAAO,EAAe,SAI5B,EAAkB,UAAY,EAC9B,EAAkB,QAAU,SAAA,GAC1B,WACA,EAAG,QACI,EAAO,EAAe,SAInC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAG/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,WAIpC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAGxD,KCjWA,AAyBA,GAAM,IAAiB,IACjB,GAAc,sBACd,GAAc,OACd,GAAwB,iBACxB,GAAsB,eACtB,GAAqB,cACrB,GAAwB,iBA8B9B,YAAsB,GAOpB,MAAO,CACL,KAAM,CAAC,GAAa,EAAM,IAAa,KAAK,IAC5C,SAAU,CAAC,GAAa,EAAM,IAAuB,KAAK,IAC1D,YAAa,CAAC,GAAa,EAAM,IAAqB,KAAK,IAC3D,WAAY,CAAC,GAAa,EAAM,IAAoB,KAAK,IACzD,cACI,CAAC,GAAa,EAAM,IAAuB,KAAK,KAWxD,YAA6B,GAC3B,GAAM,GAAQ,EAAI,MAAM,IACxB,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OAAM,uBAAuB,GAEzC,MAAO,GAAM,MAAM,EAAG,EAAM,OAAS,GAAG,KAAK,IAG/C,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAoB,YACtC,EAAI,MAAM,GAAoB,WAAW,QACzC,EAgBN,GAAA,IAAA,WAOE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,eAAiB,MAAO,SAAW,aAClD,MAAO,QAAO,cAAiB,YAKjC,KAAM,IAAI,OACN,2DAIN,GAFA,KAAK,GAAK,OAAO,aAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,sEAEN,KAAK,UAAY,EACjB,KAAK,KAAO,GAAa,KAAK,WAY1B,SAAA,UAAA,KAAN,SAAW,gFACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAGE,EAAW,KAAK,UAAU,EAAe,eACzC,EAAc,KAAK,UAAU,EAAe,aAE5C,EACF,GAA6B,GAEjC,IACE,YAAK,GAAG,QAAQ,KAAK,KAAK,KAAM,KAAK,UAAU,IAC/C,KAAK,GAAG,QAAQ,KAAK,KAAK,SAAU,GACpC,KAAK,GAAG,QAAQ,KAAK,KAAK,YAAa,GACvC,KAAK,GAAG,QACJ,KAAK,KAAK,WACV,GAA0B,EAAe,aAC7C,KAAK,GAAG,QAAQ,KAAK,KAAK,cAAe,KAAK,UAAU,CACtD,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,uBAGtC,CAAA,EAAO,CAAC,mBAAkB,UACnB,GAEP,WAAK,GAAG,WAAW,KAAK,KAAK,MAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,UAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,aAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,YAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,eAEvB,GAAI,OACN,yBAAyB,KAAK,UAAS,sFAEvC,uBAAsB,EAAmB,mBAAkB,MAC3D,qBAAoB,EAAmB,iBAAgB,MACvD,oBAAmB,EAAmB,gBAAe,qBAazD,EAAA,UAAA,KAAN,gGAGE,GAFM,EACF,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,OACrC,GAAQ,KACV,KAAM,IAAI,OACN,kDAAkD,KAAK,UAAS,KAGtE,GAAI,EAAK,oBAAsB,OAC7B,KAAM,IAAI,OACN,6EAQN,GAJM,EAAsB,GAGtB,EAAW,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,WAClD,GAAY,KACd,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAS,iBAOhE,GAJA,EAAI,cAAgB,EAGd,EAAc,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,cACrD,GAAe,KACjB,KAAM,IAAI,OACN,gDAAgD,KAAK,UAAS,kBAiBpE,GAdA,EAAI,YAAc,EAGZ,EAAiB,KAAK,GAAG,QAAQ,KAAK,KAAK,eACjD,AAAI,GAAkB,MACd,GAAW,KAAK,MAAM,GAC5B,EAAI,OAAS,EAAS,OACtB,EAAI,YAAc,EAAS,YAC3B,EAAI,YAAc,EAAS,YAC3B,EAAI,oBAAsB,EAAS,qBAI/B,EAAmB,KAAK,GAAG,QAAQ,KAAK,KAAK,YAC/C,GAAoB,KACtB,KAAM,IAAI,OACN,uDACA,KAAI,KAAK,UAAS,mBAExB,SAAI,WAAa,GAA0B,GAE3C,CAAA,EAAO,QAzIO,EAAA,WAAa,kBA2I/B,KAEa,GAA+B,SAAC,GAC3C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAoB,aACrD,GACH,EAAI,MAAM,GAAoB,WAAW,SAJxC,MAUX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBA0BA,GAClC,MAAO,IAAI,IAAoB,GAGjC,GAAA,IAAA,WAGE,aACE,EACI,KAAM,QAAQ,cACd,WAAM,MAAA,6CACV,EACI,MAAO,SAAW,aACd,MAAO,QAAO,cAAiB,YACnC,WAAM,MAAA,4DACV,KAAK,GAAK,OAAO,aAGb,SAAA,UAAA,WAAN,8FAIE,IAHM,EAA4C,GAC5C,EAAS,GAAc,GACvB,EAAS,GAAiB,GACvB,EAAI,EAAG,EAAI,KAAK,GAAG,OAAQ,EAAE,EAC9B,EAAM,KAAK,GAAG,IAAI,GACxB,AAAI,EAAI,WAAW,IAAW,EAAI,SAAS,IACnC,GAAY,GAAoB,GACtC,EAAI,GAAa,KAAK,MAAM,KAAK,GAAG,QAAQ,KAGhD,MAAA,CAAA,EAAO,QAGH,EAAA,UAAA,YAAN,SAAkB,8EAGhB,GAFA,EAAO,GAAiB,GAClB,EAAO,GAAa,GACtB,KAAK,GAAG,QAAQ,EAAK,OAAS,KAChC,KAAM,IAAI,OAAM,8BAA8B,EAAI,KAE9C,SAAO,KAAK,MAAM,KAAK,GAAG,QAAQ,EAAK,OAE7C,KAAK,GAAG,WAAW,EAAK,MACxB,KAAK,GAAG,WAAW,EAAK,UACxB,KAAK,GAAG,WAAW,EAAK,aACxB,KAAK,GAAG,WAAW,EAAK,YACxB,CAAA,EAAO,QAEX,KCvVA,AAiCA,GAAM,IAAoB,MAE1B,GAAA,WAME,aACE,KAAK,SAAW,GAGH,SAAA,YAAf,WACE,MAAI,GAA0B,UAAY,MACxC,GAA0B,SAAW,GAAI,IAEpC,EAA0B,UAS5B,EAAA,gBAAP,SAAuB,EAAgB,GACrC,EAAO,GAAU,KAAM,WAAM,MAAA,0CAC7B,AAAI,EAAO,SAAS,KAClB,GAAS,EAAO,MAAM,EAAG,EAAO,QAAQ,MAE1C,EAAO,EAAO,OAAS,EAAG,WAAM,MAAA,wCAChC,GAAM,GAAW,EAA0B,cAC3C,EACI,EAAS,SAAS,IAAW,KAC7B,WAAM,MAAA,2DACF,EAAM,OACd,EAAS,SAAS,GAAU,GAGvB,EAAA,WAAP,SAAkB,GAChB,GAAM,GAAU,KAAK,cAAc,SAAS,GAC5C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,yCAAyC,EAAM,KAEjE,MAAO,IAGF,EAAA,WAAP,WACE,MAAO,QAAO,KAAK,KAAK,cAAc,WAE1C,KAUA,YAAkB,GAChB,GAAI,EAAI,QAAQ,MAAuB,GACrC,KAAM,IAAI,OACN,6EAEA,IAAG,GAA0B,aAAa,KAAK,OAErD,MAAO,CACL,OAAQ,EAAI,MAAM,IAAmB,GACrC,KAAM,EAAI,MAAM,IAAmB,IAIvC,YACI,EAAmB,EACnB,GAAA,MAAA,KAAA,QAAA,GAAA,6GACF,SACI,IAAc,EACd,WAAM,MAAA,wCAAwC,EAAS,MAErD,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,kEACF,EAAS,MACjB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,iCAAgC,EAAS,OAC3C,EAAc,EAAa,GAE3B,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,mEACF,QAAO,EAAO,OACtB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,sCAAqC,EAAO,OAC9C,EAAc,EAAa,GAE3B,EAAe,GAAS,GAAW,OACnC,EAAa,GAAS,GAAW,KACjC,EAAa,IAAiB,GAAS,GAAW,OAEjC,CAAA,EAAM,EAAY,qBAAnC,GAAiB,EAAA,OAKnB,GAAgB,EAClB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIiB,MAAA,CAAA,EAAM,EAAY,KAAK,iBAApC,GAAa,EAAA,OAKf,GAAgB,CAAC,EACnB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIF,MAAA,CAAA,EAAO,EAAW,yBAsCpB,4HACQ,EAAU,GAA0B,aACpC,EAA2C,OAC5B,EAAA,yBAAA,GAAA,EAAA,OAAV,GAAM,EAAA,GAEX,CAAA,EAAM,GAA0B,WAAW,GAAQ,eAF7B,CAAA,EAAA,UACpB,EACF,EAAA,OACJ,IAAW,IAAQ,GACX,EAAM,EAAS,GAAoB,EACzC,EAAI,GAAO,EAAU,oBALJ,wBAQrB,MAAA,CAAA,EAAO,QAoCT,YAA2B,8EACnB,SAAgB,GAAS,GACzB,EAAU,GAA0B,WAAW,EAAc,QACnE,CAAA,EAAO,EAAQ,YAAY,EAAc,WAkD3C,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QAiDhD,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QChWhD,AA0BA,GAAA,IAAA,WAAA,cAKE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,OAAM,EAAM,IAGrB,EAAA,UAAA,IAAA,WACE,MAAO,aAAY,OAGrB,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,kDAAkD,GAExD,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAAI,cAElB,KAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,IAAI,aAAY,GAAU,OAAO,IAE5C,KAEA,GAAI,KAAM,IAAI,eACZ,KAAM,YAAY,UAAW,GAAI,KAGjC,IACE,GAA0B,gBACtB,GAAoB,WAAY,GAAI,WACjC,IAIT,IACE,GAA0B,gBACtB,GAAiB,WAAY,GAAI,WAC9B,KCpEX,AAqBO,GAAM,IAAe,CAE1B,YAAa,WAAM,MAAA,QAIjB,GAaJ,GAAA,WAKE,aAEE,KAAK,KAAO,gBAGZ,KAAK,YAAc,GAAI,MAAK,KAAK,YAGnC,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAI,MAAM,OAAO,OAAS,KACjB,KAAM,OAAO,MAAM,EAAM,GAGlC,CAAI,IAAe,MACjB,IAAc,GAAa,eAEtB,GAAY,EAAM,KAG3B,EAAA,UAAA,IAAA,WACE,GAAM,GAAO,QAAQ,SACrB,MAAO,GAAK,GAAK,IAAO,EAAK,GAAK,KAGpC,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,sDAAsD,GAE5D,MAAO,MAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAI,GAAM,SAAW,EACZ,GAEF,GAAI,MAAK,KAAK,YAAY,GAAU,OAAO,IAEtD,KAEA,AAAI,KAAM,IAAI,YACZ,KAAM,YAAY,OAAQ,GAAI,KCrFhC,YAgDI,EAAoB,EACpB,GADoB,MAAA,KAAA,QAAA,GAAW,WAEjC,EAAQ,GAAS,UACjB,GAAwC,GACjC,GAAI,IAAmB,EAAO,EAAO,GCpD9C,AAuCA,YAAiC,EAAiB,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAGnC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,mCAAmC,GAErD,GAAI,IAAU,UAAY,EAAG,QAAU,UACnC,IAAU,UAAY,EAAG,QAAU,SACrC,KAAM,IAAI,OAAM,yCAGlB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAGhB,IAAO,EAAG,CAAC,MAAK,KC3D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,QAAS,MACtC,EAAU,WACZ,MAAA,GAAO,qBAAqB,EAAG,OAAQ,EAAG,MAAO,EAAG,QAElD,EAAyB,CAAC,EAAG,GAInC,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAQ,EAAG,CAAC,OAAM,KCrD/B,YAgCwC,EAAM,GAAA,AAAA,IAAA,QAAA,GAAA,IAC5C,QAAQ,IAAI,EAAE,SAAS,ICjCzB,AAqBA,KAcA,GAAM,IAAuB,CAC3B,OAAM,GACN,KAAI,GACJ,MAAK,GACL,MAAK,IAEP,GAAa,ICzCb,AA6BA,GAAM,IAA2B,QAC3B,GAA8B,QAC9B,GAAqC,eAE3C,YAAkB,GAChB,MAAO,IAAI,SAAQ,SAAA,GAAW,MAAA,YAAW,KAAU,KAAK,GAG1D,GAAA,IAAA,WAQE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,cAGjB,KAAM,IAAI,OACN,uFAIN,AAAI,EAAe,WAAW,EAAiB,aAC7C,GAAiB,EAAe,MAAM,EAAiB,WAAW,SAEpE,AAAI,IAAkB,MAAQ,EAAe,SAAW,IACtD,GAAiB,IAGnB,KAAK,sBAAwB,EAAiB,GAC9C,KAAK,mBACD,EAAiB,GAGjB,SAAA,UAAA,KAAN,SAAW,6GACT,GAAI,MAAQ,WAAc,YACxB,KAAM,IAAI,OACN,8FAGA,EAAa,OAAO,IAAI,gBAAgB,GAAI,MAC9C,CAAC,EAAe,YAAa,CAAC,KAAM,+BAEpC,GAAe,wBAAyB,cAAxC,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,OACN,gGAGE,SAAyC,CAAC,CAC9C,MAAO,CAAC,KAAO,KAAK,oBACpB,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,gBAAe,GAEX,EACF,OAAO,IAAI,gBAAgB,GAAI,MAC3B,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,sBAIT,EAAa,KAAK,YAAc,KAAO,SAAS,cAAc,KACvB,KAAK,WAClD,EAAW,SAAW,KAAK,sBAC3B,EAAW,KAAO,EAIlB,CAAA,EAAM,GAAM,WAAM,MAAA,GAAW,cAAc,GAAI,YAAW,0BAA1D,GAAA,OAEI,EAAe,YAAc,KACzB,GAAmB,KAAK,kBAAoB,KAC9C,SAAS,cAAc,KACvB,KAAK,iBACT,EAAiB,SAAW,KAAK,mBACjC,EAAiB,KAAO,EACxB,CAAA,EAAM,GACF,WAAM,MAAA,GAAiB,cAAc,GAAI,YAAW,cAPtD,CAAA,EAAA,UAMF,EAAA,wBAIF,MAAA,CAAA,EAAO,CAAC,mBAAoB,GAA6B,WA1E7C,EAAA,WAAa,eA6E/B,KAEA,GAAA,WAGE,WAAY,GACV,GAAI,GAAS,MAAQ,EAAM,OAAS,EAClC,KAAM,IAAI,OACN,2DACA,iBAAgB,IAEtB,KAAK,MAAQ,EAGT,SAAA,UAAA,KAAN,6FACQ,SAAW,KAAK,MAAM,GACtB,EAAc,KAAK,MAAM,MAAM,GAErC,CAAA,EAAO,GAAI,SAAwB,SAAC,EAAS,GAC3C,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GAEnB,GAAM,GAAY,KAAK,MAAO,EAAM,OAAe,QAC7C,EAAgB,EAAU,cAChC,GAAI,GAAiB,MACnB,EAAO,GAAI,OACP,4CAA4C,EAAS,OACzD,OAGF,AAAI,EAAY,SAAW,GACzB,EAAQ,CAAC,cAAa,IAGxB,GAAM,GAAkB,EAAU,gBAClC,GAAI,GAAmB,MACrB,EAAO,GAAI,OACP,6CAA6C,EAAS,OAC1D,OAGF,GAAI,GACJ,IACE,EACI,EAAK,4BAA4B,EAAiB,SAC/C,GACP,EAAO,GACP,OAGF,GAAM,GAAsC,GACtC,EAAkB,GAClB,EAAgC,GACtC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,EAAM,KAAK,GACX,EAAe,KAAK,QAEtB,EAAY,KAAI,MAAhB,EAAoB,EAAa,WAGnC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,GAAM,GAAmB,GAAI,YAC7B,EAAiB,OAAS,SAAC,GAEzB,GAAM,GAAc,EAAM,OAAe,OACnC,EAAQ,EAAM,QAAQ,GAC5B,EAAe,GAAS,EACxB,AAAI,EAAe,QAAQ,QAAU,IACnC,EAAQ,CACN,cAAa,EACb,YAAW,EACX,WAAY,GAAwB,GACpC,OAAQ,EAAU,OAClB,YAAa,EAAU,YACvB,YAAa,EAAU,YACvB,oBAAqB,EAAU,uBAIrC,EAAiB,QAAU,SAAA,GACvB,MAAA,GAAO,6CAA6C,EAAI,OAC5D,EAAiB,kBAAkB,EAAW,SAIpD,EAAW,QAAU,SAAA,GAAS,MAAA,GAC1B,2DACA,eAAc,EAAS,KAAI,qCAC3B,yCACJ,EAAW,WAAW,WAOlB,EAAA,UAAA,4BAAR,SACI,EAAiC,GAInC,OAHM,GAAsB,GACtB,EAAY,EAAM,IAAI,SAAA,GAAQ,MAAA,IAAS,EAAK,QAC5C,EAAqC,GACvB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAM,MAAM,QAAQ,SAAA,GAClB,GAAM,GAAe,GAAS,GAC9B,GAAI,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,sDACA,KAAI,EAAY,MAGtB,GADA,EAAU,KAAK,GACX,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,8BAA8B,EAAY,sBAE9C,EAAW,GAAQ,EAAM,EAAU,QAAQ,MAKjD,GAAI,EAAU,SAAW,EAAM,OAC7B,KAAM,IAAI,OACN,uDACA,KAAI,EAAU,OAAM,8CACpB,KAAI,EAAM,OAAM,OAEtB,MAAO,IAEX,KAEa,GAAmC,SAAC,GAC/C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,gBAyCH,GAAA,MAAA,KAAA,QAAA,GAAA,SACxB,GAAI,IAAiB,eA2CD,GAC3B,MAAO,IAAI,IAAa,GC5V1B,YA8BI,EAAmC,EACnC,EAAwB,GAC1B,EAAc,GACd,EAAgB,GAAwB,EACxC,EAAc,GAAsB,EACpC,EAAc,EAAe,GAC7B,GAAI,GAAkB,EAEhB,EAAkB,SAAC,GACvB,SAAQ,KAAK,SAAA,GACX,GAAM,GAAW,EACb,GAAE,EAAkB,EAAS,OAAU,GAAc,GAEzD,SAAW,GACJ,IAEF,GAGT,WAAuB,GACrB,EACI,GAAY,MAAQ,MAAM,QAAQ,IAAa,EAAS,OAAS,EACjE,WAAM,MAAA,wCAGZ,WAAuB,EAAuB,GAC5C,EACI,GAAiB,GAAK,GAAiB,EACvC,WAAM,MAAA,kDACF,sBAAqB,KAC7B,EACI,GAAe,GAAK,GAAe,EACnC,WAAM,MAAA,kDACF,oBAAmB,KAC3B,EACI,GAAe,EACf,WAAM,MAAA,uDACF,sBAAqB,EAAa,qBAClC,IAAG,KAGb,MAAO,SAAQ,IAAI,EAAS,IAAI,ICvElC,YAqCI,EAAqB,6HACvB,AAAI,IAAe,MACjB,GAAc,IAGV,EAAY,EAAY,WAAa,KAAO,KAAM,SAAS,MACf,EAAY,UAGxD,EAAW,EAAU,IACvB,SAAA,GACI,MAAA,GAAU,EAAU,EAAY,YAAa,CAAC,SAAU,OAE1D,EAAqB,EACrB,EAAmB,GAEP,EAAY,YAAc,KACxC,CAAA,EAAM,QAAQ,IAAI,IADJ,CAAA,EAAA,UACd,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAU,EAAY,WAAY,EAClC,WAFJ,EAAA,EAAA,8BAFE,GAAS,EAMT,EAAiB,EAAU,IAAI,SAAA,GAAY,MAAA,GAAS,gBAEpD,EAAsB,GACtB,EAAoB,EAEV,EAAY,YAAc,KACtC,CAAA,EAAM,QAAQ,IAAI,IADN,CAAA,EAAA,UACZ,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAgB,EAAY,WAAY,EACxC,WAFJ,EAAA,EAAA,wBAFE,SAAO,EAKb,CAAA,EAAO,oBAaL,EAAiC,EACjC,EACA,GAFiC,MAAA,KAAA,QAAA,GAAA,wEAS7B,SAAe,SAAC,GAClB,MAAA,IAAyB,EAAW,CAAC,YAAW,KAC9C,EAAc,GAAqB,GAEzC,CAAA,EAAO,EAAY,EAAU,EAAgB,oBA4B3C,GADJ,GAAA,GAAA,KAIE,MAAO,UACI,EAAiC,EACjC,GADiC,MAAA,KAAA,QAAA,GAAA,4GAqD1C,GAjDM,EAAyB,EAAS,IAAI,WAAM,MAAA,KAC5C,EAKF,GACE,EACF,GAAe,KAAO,EAAY,IAAI,WAAM,MAAA,KAAS,GACnD,EAAmC,GACzC,EAAS,QAAQ,SAAC,EAAqB,GACrC,GAAI,GAAc,EAClB,EAAoB,QAAQ,QAAQ,SAAA,GAClC,GAAM,GAAY,gBAAkB,GAChC,EAAa,aAAa,MAC1B,EAAa,MAEX,EAAe,GAAqB,GACtC,GAAmB,EAAa,OAE9B,EAA8B,WAClC,EAAuB,GAAc,GACrC,AAAI,EAAoB,IAAe,MACrC,GAAoB,GAAc,IAGpC,EAAoB,GAAY,KAAK,CACnC,cAAe,EACf,YAAW,EACX,UAAW,KAIf,AAAI,GAAe,KACjB,EAAY,QAAQ,SAAC,EAAY,GAC/B,AAAI,IAAe,EAAa,MAC9B,KACA,EAAa,GAAe,MAIhC,IAGF,EAAuB,KAAK,EAAa,MACzC,GAAe,MAIf,CAAC,EAAa,MAAM,SAAA,GAAS,MAAA,KACzB,QAAkB,EAAY,OAAO,SAAC,EAAG,GAAM,MAAA,CAAC,EAAa,KAC7D,GAAI,OACN,kDACG,GAAgB,KAAK,MAAK;GAC7B,yCACG,GAAuB,KAAK,MAAK,MAKpC,SACF,EAAuB,OAAO,SAAC,EAAa,EAAa,GACvD,MAAI,IACF,EAAY,KAAK,GAEZ,GACN,IAED,EAAsB,GAC5B,EAAoB,QAAQ,SAAA,GAC1B,EAAS,GAAG,MAAM,QAAQ,SAAA,GACxB,GAAM,GAAW,EACZ,CAAC,EAAe,SAAS,KAAa,GAAN,KAAY,EACjD,EAAU,KAAK,OAGH,CAAA,EAAM,EAAqB,WAArC,SAAU,EAAA,OAEV,EAAmC,GACrC,EAAoB,EACxB,EAAoB,QAAQ,SAAA,GAI1B,OAHM,GAAa,EAAS,GAAG,MAAM,OAEjC,EAAa,EACR,EAAI,EAAG,EAAI,EAAY,IAC9B,GAAc,EAAQ,EAAoB,GAAG,WAO/C,OAHM,GAAc,GAAI,aAAY,GAC9B,EAAkB,GAAI,YAAW,GACnC,EAAoB,EACf,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAS,GAAI,YAAW,EAAQ,EAAoB,IAC1D,EAAgB,IAAI,EAAQ,GAC5B,GAAqB,EAAO,WAG9B,GAAM,GAAiB,EAAoB,GAC3C,EAAe,QAAQ,SAAA,GACrB,GAAM,GAAa,EAAY,MAC3B,EAAa,YACb,EAAa,YAAc,EAAa,WACtC,EACF,GAAc,EAAY,CAAC,EAAa,gBAC5C,OAAW,KAAQ,GACjB,EAAiB,GAAQ,EAAgB,KAI7C,GAAqB,IAGvB,CAAA,EAAO,SCpPX,AA+BA,GAAM,IAAyB,2BACzB,GAAY,mBAClB,GAAA,WAcE,WAAY,EAAc,GAgCxB,GAvCO,KAAA,eAAiB,OAQxB,AAAI,GAAe,MACjB,GAAc,IAEhB,KAAK,iBAAmB,EAAY,iBACpC,KAAK,WAAa,EAAY,WAC9B,KAAK,mBAAqB,EAAY,mBAEtC,AAAI,EAAY,WAAa,KAC3B,GACI,MAAO,GAAY,WAAc,WACjC,WAAM,MAAA,gIAGV,KAAK,MAAQ,EAAY,WAEzB,KAAK,MAAQ,KAAM,SAAS,MAG9B,EACI,GAAQ,MAAQ,EAAK,OAAS,EAC9B,WAAM,MAAA,4DAGV,AAAI,MAAM,QAAQ,IAChB,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,+CACF,sBAAqB,EAAK,OAAM,QAE1C,KAAK,KAAO,EAER,EAAY,aAAe,MAC3B,EAAY,YAAY,MAAQ,KAClC,KAAM,IAAI,OACN,sEAEN,KAAK,YAAc,EAAY,aAAe,GAG1C,SAAA,UAAA,KAAN,SAAW,yGACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,2FAIA,SAAO,OAAO,OAAO,CAAC,OAAQ,KAAK,gBAAiB,KAAK,aAC/D,EAAK,KAAO,GAAI,UAEV,EAAyC,CAAC,CAC9C,MAAO,CAAC,uBACR,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,oBACpC,gBAAe,GAGjB,EAAK,KAAK,OACN,aACA,GAAI,MACA,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,KACX,cAEJ,AAAI,EAAe,YAAc,MAC/B,EAAK,KAAK,OACN,oBACA,GAAI,MAAK,CAAC,EAAe,YAAa,CAAC,KAAM,KAC7C,qBAGW,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,WAE7C,GAFM,EAAW,EAAA,OAEb,EAAS,GACX,MAAA,CAAA,EAAO,CACL,mBAAoB,GAA6B,GACjD,UAAW,CAAC,KAGd,KAAM,IAAI,OACN,gEACG,GAAS,OAAM,YAYpB,EAAA,UAAA,KAAN,uIAC6B,MAAA,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,KAAK,qBAE5D,GAFM,EAAqB,EAAA,OAEvB,CAAC,EAAmB,GACtB,KAAM,IAAI,OACN,cAAc,KAAK,KAAI,4BACpB,GAAmB,OAAM,uCAC5B,sFAIU,CAAA,EAAM,EAAmB,eAAvC,SAAc,EAAA,qCAEV,EAAU,+CAA+C,KAAK,KAAI,IAGtE,AAAI,KAAK,KAAK,SAAS,OACrB,GAAW,+UAOX,GAAW,uEAGP,GAAI,OAAM,UAUlB,GARM,EAAgB,EAAY,cAC5B,EAAkB,EAAY,gBAC9B,EAAc,EAAY,YAC1B,EAAc,EAAY,YAC1B,EAAS,EAAY,OACrB,EAAsB,EAAY,oBAGpC,GAAiB,MAAQ,GAAmB,KAC9C,KAAM,IAAI,OACN,2BAA2B,KAAK,KAAI,mEAMtC,IAAmB,KACL,CAAA,EAAM,KAAK,YAAY,IADrC,CAAA,EAAA,UACI,EAAU,EAAA,OACf,EAAA,EAAA,GAAa,EAAA,EAAA,oBAGV,SAA4B,CAChC,cAAa,EACb,YAAW,EACX,WAAU,EACV,oBAAmB,EACnB,YAAW,EACX,YAAW,EACX,OAAM,GAGF,EAAc,EAAY,iBAChC,AAAI,GACF,GAAU,iBAAmB,GAG/B,CAAA,EAAO,SAGK,EAAA,UAAA,YAAd,SAA0B,2IAOxB,IALM,EAAa,MAAM,QAAQ,KAAK,MAAQ,KAAK,KAAK,GAAK,KAAK,KAC5D,EAAmB,GAAS,GAA3B,EAAM,EAAA,GAAE,EAAM,EAAA,GACf,EAAa,KAAK,kBAAoB,EAEtC,EAAc,GACpB,EAAA,EAAoB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAT,EAAK,EAAA,GACd,EAAY,KAAI,MAAhB,EAAoB,EAAM,SAK5B,IAFM,EAAsB,GACtB,EAAsC,GAC5C,EAAA,EAA2B,EAAA,EAAA,EAAA,EAAA,OAAA,IACzB,IADS,EAAY,EAAA,GACrB,EAAA,EAAmB,EAAA,EAAa,MAAb,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,AAAI,KAAK,oBAAsB,KAC7B,EAAY,KAAK,KAAK,mBAAmB,IAEzC,EAAU,KAAK,EAAa,EAAO,SAKrC,MAAK,sBACP,GAAA,EAAU,MAAI,SAAd,GAAkB,CAAA,EAAM,QAAQ,IAAI,KADlC,CAAA,EAAA,UACF,EAAA,MAAA,EAAA,EAAA,OAAA,CAAkB,EAAA,2BAGJ,MAAA,CAAA,EAAM,GAAyB,EAAW,CACxD,YAAa,KAAK,YAClB,UAAW,KAAK,MAChB,WAAY,KAAK,qBAHb,SAAU,EAAA,OAKhB,CAAA,EAAO,CAAC,EAAa,GAAwB,WA5M/B,EAAA,iBAAmB,eA8MrC,iBAayB,GACvB,GAAM,GAAY,EAAI,YAAY,KAC5B,EAAkB,EAAI,YAAY,KAClC,EAAS,EAAI,UAAU,EAAG,GAC1B,EACF,EAAkB,EAAY,EAAI,UAAU,GAAmB,GACnE,MAAO,CAAC,EAAS,IAAK,eAGK,GAC3B,MAAO,GAAI,MAAM,GAAY,mBAAqB,KAG7C,GAAM,IACT,SAAC,EAAa,GACZ,GAAI,MAAO,QAAU,aAChB,IAAe,MAAQ,EAAY,WAAa,MAInD,MAAO,MAEP,GAAI,GAAS,GAMb,MALA,AAAI,OAAM,QAAQ,GAChB,EAAS,EAAI,MAAM,SAAA,GAAW,MAAA,IAAa,KAE3C,EAAS,GAAa,GAEpB,EACK,GAAK,EAAK,GAGd,MAEb,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAwEf,EAAc,GACjC,MAAO,IAAI,IAAY,EAAM,eAS3B,EAAc,GAChB,MAAO,IAAK,EAAM,GC3XpB,AAuBA,GAAA,IAAA,WACE,WAA6B,GAAA,KAAA,eAAA,EAEvB,SAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,qBAEhB,KAEA,GAAA,WACE,WACqB,GAAA,KAAA,YAAA,EAGf,SAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,YAAY,SAE5B,iBAwBI,EAAmC,EACnC,EAA0B,GAC5B,GAAI,UAAU,SAAW,GACvB,GAAM,GACD,EAAkC,eAAiB,MACnD,EAAkC,aAAe,KACtD,MAAI,GACK,GAAI,IAAkB,GAI7B,SAAQ,KACJ,yNAIG,GAAI,IAAkB,CAAC,cAAe,SAK/C,gBAAQ,KACJ,yNAIG,GAAI,IAAkB,CAC3B,cAAe,EACf,YAAW,EACX,WAAU,EACV,eAAc,gBAqBhB,GAEF,MAAO,IAAI,IAAiB,GCpH9B,oYCAA,AAuDA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MAExC,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,MAAK,GAC5B,EACW,SAAC,EAAwB,GACxC,SAAQ,GAA4B,EAAO,EAAG,MAC9C,EACI,EAAG,OAAS,GAAmB,GAC/B,WAAM,MAAA,mEACV,EAAK,CAAC,IACC,EAAQ,QAAQ,EAAI,IAE7B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAEO,GAAU,EAAG,CAAC,SAAQ,KC1EnC,AA6CA,YACI,EAAiB,EAAiB,EAClC,SADkC,AAAA,IAAA,QAAA,GAAA,IAClC,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAK,EAAgB,EAAG,IAAK,UACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WAAM,MAAA,kEACF,cAAa,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE3C,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,sCAAsC,EAAU,UAC/C,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,kCAAkC,EAAW,UAC5C,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACxD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAExD,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,IAEJ,EAAQ,YACX,EAAiB,EAAiB,EAAY,IAG9C,EAA4B,CAAC,EAAG,EAAK,EAAG,GAExC,EAA0B,CAAC,WAAU,EAAE,WAAU,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAO,GAAQ,EAAK,MAGT,IAAS,EAAG,CAAC,QAAO,KC9GjC,AA+CA,YACI,EAA4B,EAAe,EAC3C,GACF,GAF6C,AAAA,IAAA,QAAA,GAAA,GAC3C,AAAA,IAAA,QAAA,GAAA,GACE,EAAQ,EACV,KAAM,IAAI,OAAM,iDAAiD,GAEnE,GAAM,GAAW,EAAgB,EAAS,UAAW,SAAU,SACzD,EAAe,EAAS,MAAK,OAAA,CAAE,IAE/B,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,IACC,EACH,EAAQ,OACJ,EAAQ,EAAU,CAAC,EAAS,OAAQ,EAAO,EAAS,GACxD,IAGA,EAAuB,CAAC,QAAS,GACjC,EAAqB,CAAC,MAAK,EAAE,QAAO,EAAE,SAAQ,GAEpD,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAS,EAAG,CAAC,QAAO,KCxEjC,AA+CA,YAAsC,EAAiB,GACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAgBnC,GAdA,AAAI,GAAQ,MACV,GAAO,EAAG,MAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAG,WAEnC,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OACzC,EAAK,QAAQ,SAAA,GACX,EACI,GAAQ,GAAK,EAAO,EAAG,KACvB,WAAM,MAAA,+CAA+C,GAAG,KAAO,GAC3D,aAAY,OAGlB,EAAG,MAAQ,EACb,MAAO,GAAG,QAGZ,GAAM,GAA0B,CAAC,EAAG,GAC9B,EAAwB,CAAC,KAAI,GAEnC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IAAO,EACxC,KAAqB,GAAW,MAGzB,IAAY,EAAG,CAAC,WAAU,KC5EvC,YA0DI,EAA6B,EAC7B,GACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,mBAC5C,EACF,EAAgB,EAAa,cAAe,mBAEhD,EACI,GAAc,MAAQ,EAAa,GAAK,OAAO,UAAU,GACzD,WAAM,MAAA,uDACF,YAAW,KACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,gDAAgD,EAAQ,OAClE,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,6CACF,YAAW,EAAa,QAChC,EACI,EAAQ,MAAM,KAAO,EAAa,MAAM,GACxC,WAAM,MAAA,uCACC,GAAQ,MAAM,GAAE,QAAQ,EAAa,MAAM,GAAE,MAChD,oEACR,EACI,EAAa,GAAK,OAAO,UAAU,GACnC,WAAM,MAAA,4DACF,IAAG,KAIX,GAAM,GAAe,GAAO,GAAK,EAAS,SAAU,GAC9C,EACF,GAAO,GAAK,EAAc,SAAU,GAClC,EAA0B,GAAU,GAC1C,MAAO,IAAK,GAAO,EAAe,GAAoB,SAGjD,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KC9FnD,2CCAA,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,mDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,oEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,AA6BA,GAAI,IAyBJ,YACI,EAEA,GAEF,GAFE,AAAA,IAAA,QAAA,GAAA,GAEE,EAAc,EAChB,KAAM,IAAI,OACN,kEAEN,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,4DAElB,GAAI,GAAc,GACd,EAAc,GACd,EAAU,GACV,EAAU,GACV,EAAe,GACnB,GAAK,EAAqB,eAAgB,YACxC,EAAc,WAEZ,MAAQ,YAAe,aAAe,YAAkB,WAC1D,EAAc,WAEZ,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAER,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAEA,EAAe,YAAc,KACvC,EAAe,OAEf,MAAM,IAAI,OACN,6OAIA,YAAY,EAAc,YAAY,OAE5C,GAAI,GACF,GAAM,GAAgC,EACtC,GAAI,GACC,EAA4B,WACzB,EACN,KAAM,IAAI,OACN,yGAMR,GAAM,GAAS,GAAU,GAAY,EAAO,aAC5C,GAAI,GAAU,MACZ,GAAM,GAA2B,CAAC,OAAM,GAClC,EAAyB,CAAC,YAAW,GAC3C,MAAO,GAAO,UACH,GAAY,EACZ,GAGP,GAAA,GAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAMV,EAEJ,AAAI,EACF,EAEK,EAAe,WAAW,MAAM,aAAa,EAAG,EAAG,EAAO,GAAQ,KAClE,AAAI,GAAe,EACxB,EAAQ,EAAiC,KACpC,AAAI,IAAW,IACpB,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAEpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA4B,EAAG,EAAG,EAAO,GAC7C,EAAO,GAAoB,aAAa,EAAG,EAAG,EAAO,GAAQ,MAE/D,GAAI,GACJ,GAAI,IAAgB,EAClB,EAAS,GAAI,YAAW,QAExB,GAAM,GAAY,EAAQ,EAC1B,EAAS,GAAI,YAAW,EAAY,GACpC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAU,EAAG,EAAU,EAAa,EAAE,EAC7C,EAAO,EAAI,EAAc,GAAW,EAAK,EAAI,EAAI,GAIvD,GAAM,GAAqC,CAAC,EAAQ,EAAO,GAC3D,MAAO,IAAS,EAAQ,EAAU,qBAuBhC,EACA,iIAQF,GAPI,EAAO,EAAgB,EAAK,MAAO,YACvC,AAAM,YAAe,IAEb,GAAoB,EAC1B,EAAO,GAAK,EAAmB,SAC/B,EAAkB,WAEhB,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,OACN,wDAAwD,EAAK,KAAI,KAKvE,GAHM,EAAkB,EAAK,MAAM,MAAM,EAAG,GAArC,EAAM,EAAA,GAAE,EAAK,EAAA,GACd,EAAQ,EAAK,OAAS,EAAI,EAAI,EAAK,MAAM,GAE3C,EAAQ,GAAK,IAAU,EACzB,KAAM,IAAI,OACN,wCACA,sBAAqB,IAG3B,GAAI,EAAK,QAAU,WAAa,EAAK,QAAU,QAC7C,KAAM,IAAI,OACN,kCAAkC,EAAK,MAAK,0CAIrC,MAAA,CAAA,EAAM,EAAK,eAIxB,IAJM,EAAO,EAAA,OACP,EAAa,EAAK,QAAU,UAAY,IAAM,EAC9C,EAAQ,GAAI,mBAAkB,EAAQ,EAAS,GAE5C,EAAI,EAAG,EAAI,EAAS,EAAO,EAAE,GAGpC,IAFM,EAAO,CAAC,EAAG,EAAG,EAAG,KAEd,EAAI,EAAG,EAAI,EAAO,KAGzB,GAFM,EAAQ,EAAK,EAAI,EAAQ,GAE3B,EAAK,QAAU,WACjB,GAAI,EAAQ,GAAK,EAAQ,EACvB,KAAM,IAAI,OACN,qDACA,kCAAiC,EAAK,cAEnC,EAAK,QAAU,SACpB,GAAQ,GAAK,EAAQ,KACvB,KAAM,IAAI,OACN,mDACA,oCAAmC,EAAK,MAIhD,AAAI,IAAU,EACZ,GAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,GAElB,EAAK,GAAK,EAAQ,EAIhB,EAAI,EAAI,EACd,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAGjC,MAAI,IAAU,MACZ,GAAO,MAAQ,EACf,EAAO,OAAS,EACV,EAAM,EAAO,WAAW,MACxB,EAAY,GAAI,WAAU,EAAO,EAAO,GAC9C,EAAI,aAAa,EAAW,EAAG,IAEjC,AAAI,IAAS,GACX,EAAK,UAEP,CAAA,EAAO,QAGF,GAAM,IAAa,EAAG,CAAC,YAAW,+DCrOrC,EAAgB,GAClB,GAAI,EAAO,KAAO,EAChB,KAAM,IAAI,OACN,0DACA,sBAAqB,EAAO,KAAI,MAEtC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,4DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OACN,sDACA,uBAAsB,EAAQ,MAAK,MAEzC,GAAI,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAAO,KAC3C,KAAM,IAAI,OACN,iEACG,GAAQ,MAAM,EAAQ,KAAO,GAAE,QAAQ,EAAO,OAGvD,GAAI,EAAO,OAAS,EAClB,KAAM,IAAI,OACN,qDACA,kBAAiB,EAAO,MAAK,MASnC,OANM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAIjD,EAAU,EACL,EAAI,EAAG,EAAI,EAAa,OAAS,EAAG,EAAE,EAC7C,GAAW,EAAa,GAG1B,GAAM,GAAa,EAAO,MAEpB,EAAc,EAAa,QACjC,EAAY,MAGZ,OADI,GAAY,EACP,EAAI,EAAW,EAAI,EAAO,KAAM,EAAE,EACzC,GAAa,EAAW,GACxB,EAAY,KAAK,EAAW,IAG9B,GAAM,GACE,GAAe,EAAO,OAAO,IAAI,SAAA,GAAU,MAAA,GAAS,IAAU,OAAA,CACjE,IAAG,MAAM,EAAG,GAEjB,MAAO,CAAC,EAAa,EAAS,EAAW,6DCtDvC,EAAiB,EAAiB,GACpC,GAAM,GAAY,EAAQ,KAAO,EAAK,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAClE,EAAY,EAAQ,KAAO,EAAK,EAAQ,KAAO,EAAI,EAEnD,EAAa,wDACf,yCAAwC,EAAQ,OAChD,qBAAoB,EAAQ,MAAK,YAAY,GAC7C,gBAAe,EAAQ,mBAAmB,EAAQ,KAEtD,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OAAM,EAAa,mBAAkB,EAAQ,OAEzD,GAAI,EAAM,OAAS,EAAY,GAAQ,KAAO,GAC5C,KAAM,IAAI,OACN,EACA,2BAA0B,GAAY,GAAQ,KAAO,MAE3D,GAAI,EAAQ,OAAS,EAAW,EAAM,OAAS,EAC7C,KAAM,IAAI,OACN,EAAa,oBAAmB,GAAW,EAAM,OAAS,KAEhE,OAAS,GAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,GAAI,EAAQ,MAAM,KAAO,EAAQ,MAAM,GACrC,KAAM,IAAI,OACN,EACA,mBAAkB,EAAC,MAAM,EAAQ,MAAM,GAAE,sBAAsB,EAAC,MAC5D,EAAQ,MAAM,GAAE,OAG5B,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAO,EAAU,EAAE,EAC7C,GAAI,EAAQ,MAAM,EAAI,KAAc,EAAM,EAAI,GAC5C,KAAM,IAAI,OACN,EACA,mBAAkB,GAAI,GAAQ,MAC1B,EAAQ,MAAM,EAAI,GAAS,cAAc,GAAI,GAAQ,MACrD,EAAM,EAAI,GAAS,kBAoB7B,EAAiB,EAAiB,GACpC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OAAM,0DACZ,EAAQ,OAEd,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OACN,6DAA6D,GAGnE,GAAI,EAAM,SAAW,GACnB,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAEd,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAIhB,GAAoB,EAAO,EAAS,eAalC,EAAqB,EACrB,GAWF,OATM,GAAc,EAAQ,MAAM,OAC5B,EAAa,EAAc,EAAK,EAAQ,MAAM,EAAc,GAAK,EAKjE,EAAU,EAAM,OAElB,EAAY,EACP,EAAI,EAAW,EAAI,EAAS,EAAE,EACrC,GAAa,EAAM,GAGrB,GAAM,GAAgB,EAAY,EAAK,EAAI,EACrC,EAAa,GAAc,EAAQ,OAAS,EAE5C,EAAc,GAAe,EAAM,MAAM,EAAG,IAAW,OAAA,CAAE,IACzD,EAAa,GAAc,GACjC,MAAO,CAAC,UAAS,EAAE,WAAU,EAAE,UAAS,EAAE,QAAO,EAAE,WAAU,sFCnJ/D,YAqBI,EAAmB,EAAiB,GACtC,GAAM,GAAY,EAAM,MAAM,OAC9B,EACI,IAAc,EAAM,OACpB,WAAM,MAAA,iBAAiB,EAAS,sBAAsB,EAAK,SACvD,iCAAgC,EAAS,QACjD,EACI,IAAc,EAAK,OACnB,WAAM,MAAA,iBAAiB,EAAS,qBAAqB,EAAI,SACrD,iCAAgC,EAAS,QAEjD,mBAAS,GACP,EACI,EAAM,GAAK,EAAK,IAAM,EAAM,MAAM,GAClC,WAAM,MAAA,iBAAiB,EAAS,YAAY,EAAC,YAAY,EAAC,KACtD,KAAI,GAAM,GAAK,EAAK,IAAE,gCAAgC,EAAC,MACjD,EAAM,MAAM,GAAE,QALrB,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,eAUgB,GAGzB,OAFM,GAAO,GACT,EAAO,EACJ,EAAO,GACZ,AAAI,EAAO,GACT,EAAK,KAAK,GAEZ,GAAQ,EACR,IAEF,MAAO,eAKL,EAAiB,EAAe,GAElC,OADM,GAAO,GACJ,EAAO,EAAG,EAAO,EAAM,OAAQ,IACtC,EAAK,GAAQ,KAAK,KAAM,GAAI,GAAQ,EAAM,IAAS,EAAQ,IAE7D,MAAO,eAML,EAAmB,EAAgC,EACnD,GAEF,OADM,GAAiB,EAAO,QACrB,EAAI,EAAW,OAAQ,EAAI,EAAW,OAAQ,IACrD,EAAW,KAAK,GAElB,OAAS,GAAI,EAAG,EAAI,EAAe,IACjC,AAAI,IAAM,EACR,EAAW,GAA0B,EAErC,GAAW,OACP,EAAwB,EACxB,GACJ,EAAW,OAGf,MAAO,GAGT,YACI,EAAgC,EAChC,GACF,MAAI,IAAkB,EACb,EAGF,EAAkB,GAAgB,GAG3C,YAAuB,EAAuB,GAE5C,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAe,IACjC,EAAW,KAAK,EAAyB,GAE3C,MAAO,eAKL,EAAsB,EAAwB,EAC9C,EAAiB,EAAe,EAAmB,EACnD,EACA,GACF,GAAM,GAAY,EAAW,OACzB,EAAkB,GAAI,OAAM,GAC5B,EAAgB,GAAI,OAAM,GAC1B,EAAoB,GAAI,OAAM,GAClC,GAAI,EAAa,QAAU,EAAsB,GAC/C,GAAM,GAAY,EAAa,GAIzB,EAAgB,EAAsB,EAC5C,EAAkB,GACd,EAAW,EAAW,EAAe,EAAO,GAChD,EAAgB,GACZ,EAAS,EAAW,EAAe,EAAK,GAC5C,EACI,GAAsB,EAAS,EAAW,EAAe,OAE7D,QAAS,GAAO,EAAG,EAAO,EAAW,IACnC,EAAgB,GAAQ,GACpB,EAAW,EAAO,EAAS,EAAY,EAAM,GACjD,EAAc,GACV,GAAY,EAAS,EAAK,EAAS,EAAY,EAAM,GACzD,EAAkB,GAAQ,GAAe,EAAS,EAAM,GAI5D,MAAO,CACL,MAAO,EACP,IAAK,EACL,QAAS,eAOT,EAAmB,EAAgC,EACnD,EAAyB,GAI3B,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAEnB,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAc,GAClC,AAAI,EAAY,GAAK,GACnB,GAAgB,GAGlB,EAAW,GAAQ,EAGvB,MAAO,eAML,EAAiB,EAAgC,EACjD,EAAuB,GAIzB,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAAO,sBAE1B,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAY,GAChC,AAAI,EAAU,GAAK,GACjB,GAAgB,OAAO,kBAEzB,EAAW,GAAQ,EAIvB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,KAErC,GAAM,GAAW,EAAW,GAC5B,AAAI,EAAW,GAAK,GAClB,GAAW,IAAM,GAEnB,EAAW,GAAK,GAAW,EAAG,EAAW,GAAI,EAAW,IAE1D,MAAO,eAIL,EAAmB,EAAc,GACnC,GAAI,GAAS,EAAQ,GACrB,MAAI,GAAgB,GAAK,GAAS,GAAU,OAC1C,GAAS,GAGJ,cAIL,EAAmB,EAAwB,EAC3C,EAAsB,EAAc,GAEtC,GAAI,GAAQ,EAAa,GACnB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAY,GAAK,GAAQ,EAAe,GAAK,GAAQ,GAAS,OAChE,CAAI,EAAS,EAIX,EAAQ,OAAO,iBAGf,EAAQ,OAAO,kBAKnB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAQ,GACV,IAAS,GAIX,EAAQ,GAAW,EAAG,EAAO,EAAW,GAEjC,cAIL,EAAiB,EAAuB,EACxC,EAAsB,EAAc,GAEtC,GAAI,GAAO,EAAY,GACjB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAW,GAAK,GAAS,EAAgB,GAAK,GAAS,GAAQ,OACjE,CAAI,EAAS,EAGX,EAAO,OAAO,iBAGd,EAAO,OAAO,kBAKlB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAO,GACT,IAAQ,GAMV,AAAI,EAAS,EAEX,EAAO,GAAW,EAAG,EAAM,GAG3B,EAAO,GAAW,GAAI,EAAM,EAAW,GAGlC,cAQL,EAAiB,EAAiB,GAGpC,OADI,GAAkB,EAAK,OAClB,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,GAAI,EAAK,GAAK,GACZ,EAAkB,EAClB,MAIJ,OAAS,GAAI,EAAkB,EAAG,EAAI,EAAK,OAAQ,IACjD,GAAI,EAAM,GAAK,GAAK,EAAK,KAAO,EAAM,GACpC,MAAO,GAGX,MAAO,eAGyB,EAAiB,GAEjD,OADI,GAAa,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EACrD,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAc,EAAM,GAAK,EAAQ,GAEnC,MAAO,eAIL,EAAe,EAAwB,GAEzC,GAAI,GACE,EAAQ,EAAE,MAAM,OACtB,AAAI,MAAO,IAAU,SACnB,EAAM,CAAI,GAAK,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,IACzC,AAAI,EAAM,OAAS,EACxB,EAAS,EAAM,OAAO,GAAI,OAAM,EAAQ,EAAM,QAAQ,KAAK,IAE3D,EAAS,EAAM,QAEjB,EAAO,QAAQ,SAAA,GACb,EACI,IAAM,GAAI,WAAM,MAAA,wDAEtB,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAAI,OAAM,GAAO,KAAK,IACzB,AAAI,MAAO,IAAS,SACzB,EAAK,CAAI,GAAI,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,KACvC,AAAI,EAAK,OAAS,EACvB,EAAQ,EAAK,OAAO,GAAI,OAAM,EAAQ,EAAK,QAAQ,KAAK,KAExD,EAAQ,EAEV,EAAQ,EAAM,IAAI,SAAC,EAAG,GACpB,MAAI,IAAK,EACA,EAEP,GACI,IAAM,GACN,WAAM,MAAA,qDACC,GAAC,kCAAkC,EAAC,OACxC,EAAE,MAAM,GAAK,EAAO,MAGxB,CAAC,EAAQ,wSCjWlB,AA8DA,GAAA,IAAA,WAAA,cAYE,SAAA,UAAA,aAAA,WACE,MAAQ,MAAK,YACR,WAiBA,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,IAEnB,KASA,GAAA,WAOE,aACE,KAAK,aAAe,GAMf,SAAA,OAAP,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UAMnB,EAAA,SAAP,SAAwC,GACtC,EAAiB,SAAS,aAAa,EAAI,WACvC,CAAC,EAAK,EAAI,aAElB,iBA2BI,GACF,EACI,EAAI,WAAa,KACjB,WAAM,MAAA,gFAEV,EACI,MAAO,GAAI,WAAc,SACzB,WAAM,MAAA,sDACF,MAAO,GAAI,YACnB,EACI,EAAI,UAAU,OAAS,EACvB,WAAM,MAAA,sFAGV,GAAiB,SAAS,gFC/K5B,AAsBA,GAAM,IAAuB,KAChB,GAAuB,eAGhC,EACA,EAAoD,GACtD,MAAI,IAAW,MACb,GAAU,MAEL,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,mBAInE,MAAO,GAAO,QAAQ,mBAAqB,GAAK,GACA,GAGlD,YACI,EAAoB,EACpB,GACF,GAAI,GAAiB,GAOrB,GANA,AAAI,IAAa,IAAW,GAAa,KACvC,GAAiB,IAEnB,AAAI,GAAa,IAAW,GAAa,IACvC,GAAiB,IAEf,GACF,GAAM,GAAQ,EAAO,YAAY,KAC3B,EAAQ,EAAS,YAAY,KAEnC,GAAI,IAAU,EACZ,KAAM,IAAI,OACN,yCAAyC,EAAK,KAC9C,cAAa,IAIrB,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,IACzC,GAAM,GAAc,GAAW,GACzB,EAAgB,GAAW,GACjC,GAAI,CAAC,GAAY,EAAa,GAC5B,KAAM,IAAI,OACN,iCACA,aAAY,EAAW,iBAAiB,EAAa,MAI7D,GAAM,GACF,GAAa,GAAU,EAAS,GAAQ,GACtC,EAAe,GAAa,GAC9B,EACA,GAAQ,GAEZ,GAAI,EAAW,SAAW,EAAa,OACrC,KAAM,IAAI,OACN,yCAAyC,EAAW,OAAM,OAC1D,cAAa,EAAa,OAAM;GAChC,cAAa,EAAU;GACvB,cAAa,EAAY,MAE/B,OAAS,GAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAI,EAAW,GACf,EAAI,EAAa,GAEvB,GAAI,CAAC,EAAU,EAAG,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAC,OAAO,EAAC,cAAc,EAAC,OAAO,EAAC;EACzD,cAAa,EAAU;GACvB,cAAa,EAAY,mBAUC,EAAuB,GACzD,IAAK,KAAK,WAAM,MAAA,GAAK,QAAQ,WAAM,MAAA,mBAGH,EAAoB,GACpD,GAAM,GAAM,MAAO,IAAa,UAAY,MAAO,IAAa,UACxD,MAAO,IAAa,UACxB,CAAC,GACD,EACJ,MAAI,IAAS,IAAW,GAAU,EAAoB,KAClD,GAAS,IAAa,GAAU,EAAsB,IAEjD,GAAsB,EAAQ,EAAK,SAAC,EAAG,GAAM,MAAA,IAAK,IAEpD,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,iBAGlC,EAAW,EAAW,GAIvD,GAHA,AAAI,GAAW,MACb,GAAU,MAER,CAAC,GAAS,EAAG,EAAG,GAClB,KAAM,IAAI,OAAM,8BAA8B,EAAC,kBAAkB,GAIrE,YAAkB,EAAW,EAAW,GACtC,MAAI,CAAC,SAAS,IAAM,CAAC,SAAS,GACrB,GAEL,QAAM,IAAM,MAAM,IAAM,KAAK,IAAI,EAAI,GAAK,eAO5C,EAA6B,EAAa,GAC5C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,GAAI,EAAO,GAAK,GAAO,EAAO,GAAK,EACjC,KAAM,IAAI,OACN,sBAAsB,EAAO,GAAE,SAAS,EAAG,WAAW,eAM5D,EAAqB,GAGvB,OAAO,GAAI,cAAa,IAAS,QAAQ,GAAI,cAAa,4MCzJ5D,GAGM,IAAU,QCHhB,cAiCE,KAAM,IAAI,OAAQ,kBAkBlB,KAAM,IAAI,QAAS,kBAKnB,KAAM,IAAI,+BAAgC,IAC1C,QAAQ,KAAK,sEAIiB,GAC9B,AAAI,KAAM,QAAQ,iCAChB,QAAQ,KACJ,EAAM,6FAYZ,EAAO,iCASP,MAAO,iBAyBP,MAAO,GAAO,qBAgCQ,GAEtB,MAAO,GAAO,QAAQ,eA4CpB,EAA6B,GAC/B,MAAO,GAAO,KAAK,EAAU,eAcP,GACtB,GAAM,GAAU,GAAsB,GACtC,EAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,wBAmCI,GACrC,MAAO,GAAO,KAAK,eA4BA,GACnB,MAAO,GAAO,KAAK,eAkBM,GACzB,MAAO,GAAO,WAAW,iBAWzB,MAAO,GAAO,sBAUd,MAAO,GAAO,wBAQc,GAC5B,EAAO,cAAc,eAOK,GAC1B,MAAO,GAAO,YAAY,eAQO,GAEjC,MAAO,GAAO,mBAAmB,eAmB/B,EAAc,EACd,GAAA,MAAA,KAAA,QAAA,GAAA,GACK,EAAO,gBAAgB,EAAM,EAAS,iBAY7C,MAAO,GAAO,oBASY,EAAsB,GAChD,KAAM,YAAY,EAAc,GC5XlC,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCnE3B,AAqDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCvErC,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OAGjC,GAFA,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAED,EAAG,QAAU,SAAW,EAAG,QAAU,QACvC,MAAO,IAAS,EAAI,GAGtB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,WAAW,EAAI,GACnC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAC/B,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KC3E3B,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAEA,GAAM,EAAG,CAAC,KAAI,KCpE3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GAEpC,MADA,GAAK,CAAC,IACF,EAAG,QAAU,YACR,EAAQ,WAAW,GAGrB,EAAQ,IAAI,IAClB,EAAgC,KAAiB,OAGzC,IAAM,EAAG,CAAC,KAAI,KCrD3B,AAoCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAuCA,YAAiC,GAC/B,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,+DACV,EACI,EAAQ,QAAU,EAClB,WAAM,MAAA,uDACF,IAAG,EAAQ,UAEnB,GAAM,GACF,EAAQ,IAAI,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAG,UAAU,EAAK,UAEtD,EAAc,EAAS,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAI,EAAE,QAAU,EAAY,MAC1B,KAAM,IAAI,OACN,8DAIR,EAAS,QAAQ,SAAA,GACf,GAAI,CAAC,GAAiB,EAAE,MAAO,EAAY,OACzC,KAAM,IAAI,OACN,8DAIR,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,GACE,GAGH,EAAqB,EAE3B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC/E7B,YAuBqC,EAAgB,GACnD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,GAAI,EAAK,EAAK,OAAS,EAAI,KAAO,EAAO,EAAI,EAC3C,MAAO,GAGX,MAAO,eAIL,EAAqB,EAAqB,GAK1C,OAJI,GAAO,EAAU,OAAS,EAAU,OACpC,EAAM,GACR,EAAS,EACT,EAAY,EACL,EAAM,EAAG,EAAM,EAAM,IAC9B,AAAI,EAAK,QAAQ,KAAS,GACxB,EAAI,KAAK,EAAU,MAEnB,EAAI,KAAK,EAAU,MAGvB,MAAO,eAIL,EAAkB,GAGpB,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,EAAK,QAAQ,KAAS,IACxB,EAAS,KAAK,EAAO,IAGzB,GAAM,GAAc,EAAK,IAAI,SAAA,GAAO,MAAA,GAAO,KAC3C,MAAO,CAAC,EAAU,eAIhB,EAAiB,GACnB,GAAM,GAAiB,EAAK,IAAI,SAAA,GAAK,MAAA,KACrC,MAAO,IAAiB,EAAO,EAAgB,eAI7C,EAAa,EAAgB,GAC/B,EACI,GAAqB,EAAM,GAC3B,WAAM,MAAG,GAAG,2CACR,aAAY,EAAI,aAAa,EAAI,yBAQR,EAAgB,GAEjD,GAAI,GAAqB,EAAM,GAC7B,MAAO,MAGT,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,AAAI,EAAK,QAAQ,KAAO,IACtB,EAAO,KAAK,GAGhB,SAAK,QAAQ,SAAA,GAAQ,MAAA,GAAO,KAAK,KAC1B,cAI8B,GACrC,MAAO,GAAK,IAAI,SAAC,EAAM,GAAM,MAAA,CAAC,EAAG,KAC5B,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAE,KACxB,IAAI,SAAA,GAAK,MAAA,GAAE,iBAGe,EAAiB,GAEhD,OADM,GAAgB,GACb,EAAI,EAAO,EAAS,EAAI,EAAM,EAAE,EACvC,EAAI,KAAK,GAEX,MAAO,GC3GT,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KCxF3B,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAIA,IAAM,EAAG,CAAC,KAAI,KCzF3B,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC7EjC,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,AAAI,GAAQ,MACV,GAAO,GAET,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KChFjC,AAqCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC/C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA2CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAEH,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAQ,EAAG,CAAC,OAAM,KC7D/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,YAoGI,EACA,EAAuC,EACvC,EAA4B,EAC5B,GAD4B,AAAA,IAAA,QAAA,GAAA,QAM9B,GAAM,GAAgB,EAAW,GAC3B,EACE,EAAW,OAAA,CAAE,IACf,EAAc,GAAwB,GAE5C,MAAO,IACH,EAAY,EAAc,EAAS,EAAW,EAC9C,KAAyB,KAAsB,eAIjD,EACA,EAAqC,EACrC,EAAoC,EACpC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,gBACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GAEjB,EACJ,GAAI,IAAe,eACjB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACrD,IAAe,gBACxB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAE9D,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,EAAc,GAC7D,eAOF,EACA,EACA,EACA,EAA4C,EAC5C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,SACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAE9B,EACA,EACJ,GAAI,IAAe,QACjB,EAAc,eACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACxD,IAAe,QACxB,EAAc,gBACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAEjE,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,GAAO,EACtD,eAQF,EACA,EACA,EAAkC,EAClC,EACA,EAAuC,EACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACvC,AAAA,IAAA,QAAA,GAAA,gBACE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GACnC,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WACtB,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAU,EAAA,EAAA,OAElC,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAC9B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GACf,EAAA,GAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAEjB,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,SAIrB,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAW,GAC1C,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAW,EAAU,IAGvC,CACL,UAAS,EACT,WAAU,EACV,SAAQ,EACR,QAAO,EACP,WAAU,EACV,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,aAAY,EACZ,YAAW,EACX,aAAY,EACZ,YAAW,EACX,sBAAqB,EACrB,qBAAoB,EACpB,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,eA2CX,EACA,EACA,EACA,EAA4C,EAC5C,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,gBAEE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GAE5C,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WAC/B,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,OAE3C,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAE3C,EAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAC5B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAe,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGhC,EACF,GAAuB,EAAa,GAClC,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,SAAU,EAAA,EAAA,UAAW,EAAA,EAAA,SAK/B,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAU,EAAW,GACpD,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAU,EAAW,EAAU,IAGjD,CACL,UAAS,EACT,WAAU,EACV,QAAO,EACP,SAAQ,EACR,QAAO,EACP,WAAU,EACV,SAAQ,EACR,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,YAAW,EACX,aAAY,EACZ,YAAW,EACX,YAAW,EACX,aAAY,EACZ,YAAW,EACX,qBAAoB,EACpB,sBAAqB,EACrB,qBAAoB,EACpB,cAAa,EACb,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,GAIf,YACI,EAA2B,EAAmB,EAC9C,EAAkB,GACpB,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAY,GAGtB,YACI,EAA2C,EAC3C,EAAqB,EAAgB,EACrC,GACF,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAa,EAAQ,GACrB,EAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAe,GAChB,GAAa,EAAY,EAAI,GAAW,EAAS,EAAG,GACzD,EACI,GAAW,GACX,WAAM,MAAA,2BAA2B,EAAY,uEAGjD,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAc,EAAY,EAAY,eAI5C,EACA,EAAmB,EAAgB,GAAA,AAAA,IAAA,QAAA,GAAA,GACrC,GAAM,GAAqB,GAAuB,EAAW,GAC7D,MAAO,MAAK,MACP,GAAW,GAAM,GAAS,GAAK,EAAS,GAAsB,GAGrE,YAAyB,GACvB,MAAI,OAAO,IAAU,SACZ,CAAC,EAAO,EAAO,GAEpB,EAAM,SAAW,EACZ,CAAC,EAAM,GAAI,EAAM,GAAI,GAEvB,EAGT,YAA0B,GAExB,MAAO,OAAO,IAAU,SAAW,CAAC,EAAO,EAAO,GAAS,EAc7D,YAAgC,EAAoB,GAClD,MAAI,IAAY,EACP,EAGF,EAAc,GAAa,GAAM,GAAW,GAGrD,YACI,EAA4C,EAC5C,EAAiB,EAAsB,EACvC,EAAsB,EACtB,EACA,GAEF,GAAI,GACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CAAC,IAAK,EAAK,OAAQ,EAAK,KAAM,EAAK,MAAO,EAAK,KAAM,GAC/D,GAAM,GAAW,GACb,CAAC,EAAU,GAAU,EAAc,EAAc,EAAK,GAC1D,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GACF,KAAK,IAAI,EAAI,GAAY,GAAK,EAAe,EAAe,GAC1D,EACF,KAAK,IAAI,EAAI,GAAW,GAAK,EAAc,EAAc,GACvD,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAC9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,gBAClC,IAAQ,QACjB,EAAU,CAAC,IAAK,EAAG,OAAQ,EAAG,KAAM,EAAG,MAAO,EAAG,KAAM,SACvD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,WAC1C,MAAO,IAAQ,UACxB,GAAM,GAAM,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GACzD,EAAS,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC5D,EAAO,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC1D,EAAQ,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC3D,EAAW,IAAQ,GAAK,IAAW,GAAK,IAAS,GAAK,IAAU,EAClE,QACA,WACJ,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,GAC3C,EAAY,GACP,GAAW,EAAe,EAAM,GAAU,EAAe,EAC1D,GACJ,EAAW,GACN,GAAU,EAAc,EAAO,GAAS,EAAc,EAAG,OAE9D,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,UAAS,EAAE,SAAQ,GAGtC,YACI,EAA4B,EAAiB,EAC7C,EAAiB,EAAqB,EACtC,EAAqB,EAAqB,EAC1C,EAAqB,GAMvB,GAAI,GACA,EACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,GAER,GAAM,GAAW,GACb,CAAC,EAAS,EAAU,EAAS,GAAI,EAAa,EAAG,EAAa,EAC9D,GACJ,EAAW,EAAS,GACpB,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAW,KAAK,KAAK,EAAU,GAC/B,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EACD,GAAY,GAAK,EAAe,EAAe,EAC9C,EAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EAAQ,KAAK,MAAM,EAAgB,GACnC,EAAO,EAAgB,EACvB,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAE9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,MAAK,EAAE,KAAI,EAAE,KAAM,gBAC/C,IAAQ,QACjB,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,SAER,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,GACnD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,OAEnD,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,SAAQ,EAAE,UAAS,EAAE,SAAQ,GAQhD,YACI,EAAe,GACjB,GAAI,CAAC,EACH,MAAO,GAET,OAAQ,OACD,QAEH,MAAO,MAAK,MAAM,OACf,OAEH,MAAO,MAAK,KAAK,OACd,QACH,MAAO,MAAK,MAAM,WAElB,KAAM,IAAI,OAAM,wBAAwB,gBAIZ,GAC1B,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,MAAO,KAAS,GAAK,IAAS,GAAK,IAAS,cAI1C,EAA0B,GAC5B,MAAO,IAAkB,IAAY,GAAkB,eAUjB,GAEtC,GAAI,IAAe,OACjB,MAAO,eACF,GAAI,IAAe,OACxB,MAAO,gBAEP,KAAM,IAAI,OAAM,sBAAsB,GC5mB1C,AAoDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,WACxC,EAAY,EAElB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,mDAAmD,EAAI,KAAI,MAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAIJ,MAFA,GAAK,CAAC,IAEF,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UACvC,EAAI,QAGN,EAAQ,QAAQ,EAAK,IAGxB,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAElE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAG,OAEf,EACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KClHnC,AA6EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,WAE5C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAGrB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GAEJ,SAAK,CAAC,IAEC,EAAQ,UAAU,EAAK,IAG1B,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAEjE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAI,OAEhB,EACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KCzJvC,YAmBuC,EAAoB,GACzD,GAAM,GAAO,EAAO,GAAG,OACvB,EAAO,QAAQ,SAAC,EAAO,GACrB,EACI,EAAM,SAAW,EACjB,WACI,MAAA,kBAAkB,EAAI,sBAAsB,EAAC,sBAC7C,6BAA4B,EAAI,SAG1C,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,kBAAkB,EAAI,iCAAiC,GAAO,GAAC,MAEzE,GAAM,GAAa,EAAO,GAC1B,EAAO,QAAQ,SAAC,EAAO,GACrB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EACK,IAAM,GAAU,EAAM,KAAO,EAAW,GACzC,WAAM,MAAA,kBAAkB,EAAI,uBAAuB,EAAC,MAAM,EAAK,KAC3D,0CAAyC,EAAU,MACnD,oCAAmC,EAAC,qBAKlB,EAAoB,GAElD,OADM,GAAc,EAAO,GAAG,QACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAY,IAAS,EAAO,GAAG,GAEjC,MAAO,GClDT,AAqEA,YAAmC,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/D,EAAO,EAAQ,QAAU,EAAG,WAAM,MAAA,uCAElC,GAAI,GAAW,GAAqB,EAAS,UAAW,UAExD,AAAI,EAAS,GAAG,QAAU,aACxB,EAAS,QAAQ,SAAA,GACf,GAAI,EAAO,QAAU,YACnB,KAAM,IAAI,OAAM;uBACD,EAAO,MAAK,QAKjC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAS,GAAG,OAAO,GAChD,EAAW,GAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAC7D,GAAI,GAAc,KAAc,EAC9B,MAAO,IAAO,GAAI,GAIpB,GADA,EAAW,EAAS,OAAO,SAAA,GAAK,MAAA,GAAE,KAAO,IACrC,EAAS,SAAW,EACtB,MAAO,GAAS,GAGlB,GAAM,GAAS,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QACnC,GAAuB,EAAQ,GAE/B,GAAM,GAAM,EAAQ,OAAO,EAAU,GACrC,SAAK,GACE,GAGH,EAAuB,EACvB,EAAoB,CAAC,KAAI,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC/GjC,AAsCA,YAAoC,GAClC,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAAwB,CAAC,EAAG,GAElC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,QAAQ,GAC5B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAU,EAAG,CAAC,SAAQ,KCjDnC,AA8DA,YACI,EAAiB,EAAwB,GAC3C,GAAM,GAAK,EAAgB,EAAG,IAAK,SAEnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,kCAGlB,GAAM,GAA+B,SAAC,EAAS,GACvC,GAAA,GAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,UAA6B,EAAI,EAAQ,GACzC,EAAK,CAAC,IACC,EAAQ,MAAM,EAAI,EAAQ,IAG7B,EAAsB,CAAC,EAAG,GAC1B,EAAoB,CAAC,MAAK,EAAE,KAAI,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCrF/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAI,EAAQ,KAAK,GACvB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA8CA,YACI,EAA+B,EAC/B,EAA+B,EAC/B,EAAwB,GAC1B,GAAM,GACF,EAAgB,EAAY,aAAc,iBACxC,EACF,EAAgB,EAAY,aAAc,iBACxC,EAAY,EAAgB,EAAU,WAAY,iBAClD,EAAQ,EAAgB,EAAM,OAAQ,iBACtC,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAE7B,EAAW,GAAO,CAAC,EAAO,GAAK,GAC/B,EAAW,GAAO,EAAU,GAC5B,EAAgB,GAAI,EAAU,GAG9B,EAAY,EAAI,MAAM,GACtB,EAAY,EAAI,MAAM,GAAK,EAC3B,EAA8B,CAAC,EAAW,GAC1C,EAAI,GAAM,EAAK,CAAC,EAAG,GAAI,GACvB,EAAI,GAAM,EAAK,CAAC,EAAG,GAAY,GAC/B,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GACnC,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GAEnC,EACF,GAAI,EAAI,GAAQ,GAAI,GAAK,IACrB,EAAI,EAAI,GAAQ,GAAI,EAAa,MACnC,EAAiB,EAAI,GAAK,GAAO,GAAQ,IAC/C,MAAO,CAAC,EAAM,MAGH,IAAgB,EAAG,CAAC,eAAc,KC/E/C,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAE7C,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,iBAAiB,EAAG,KAAI,2CAC1B,EAAW,SAEnB,EACI,EAAM,SAAW,EAAW,OAC5B,WAAM,MAAA,mBACF,EAAM,OAAM,8CACZ,EAAW,SAEnB,EACI,EAAG,MAAM,GAAK,IAAS,EACvB,WAAM,MAAA,yBACI,EAAG,MAAM,GAAE,2CACjB,+BAA8B,EAAW,KAAK,OAAM,QAAQ,KAEpE,GAAM,GAA0B,SAAA,GAC9B,MAAO,GAAQ,eAAe,EAAI,EAAY,IAG1C,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,MAAK,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,iBC1FX,GACpC,GAAI,GACJ,MAAI,GAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAG,EAAE,OACxB,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACvC,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAErD,EAAM,EAGD,EChCT,AAwDA,YACI,EAAyB,EACzB,EACA,EACA,EACA,GACF,AAAI,GAAmB,MACrB,GAAkB,MAEpB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,AAAI,GAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAG9C,EACI,EAAM,OAAS,EAAU,KACzB,WAAM,MAAA,iFAEV,EACI,GAAW,MAAQ,EAAM,OAAS,EAAQ,KAC1C,WAAM,MAAA,+EAEV,EACI,GAAU,MAAQ,EAAM,OAAS,EAAO,KACxC,WAAM,MAAA,8EAGV,GAAM,GAAgB,GAAM,GAEtB,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,EAAO,EAAW,IAEtB,EAAQ,UACX,EAAK,GAAS,GAAQ,GAAS,GAAY,GAAS,GACpD,GAAS,GAAS,IAGlB,EAA+B,CACnC,EAAG,EACH,MAAO,EACP,OAAQ,EACR,KAAM,EACN,SAAU,GAGN,EAA6B,CAAC,gBAAe,GAE7C,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAO,GAAQ,EAAK,EAAG,OAGzB,YAAkB,GAChB,MAAI,IAAK,KACA,KAEL,EAAE,OAAS,EAEN,EAAQ,EAAG,CAAC,EAAE,OACZ,EAAE,OAAS,EACb,EACE,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACpC,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAEjD,KAGI,IAAY,EAAG,CAAC,WAAU,KCrGvC,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAE/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC9E3C,AA4CA,YACI,EAAsB,GACxB,GAAI,GAAQ,EAAgB,EAAG,cAAe,KACxC,EAAS,EAAM,MAErB,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,CAAE,GAAI,IAAM,EAAI,IAAM,IACxC,KAAM,IAAI,OAAM,2CAA2C,EAAK,MAGlE,GAAI,EAAM,OAAS,EAAM,KACvB,KAAM,IAAI,OAAM,+BAA+B,EAAM,OAAM,iBACvD,EAAM,KAAI,KAGhB,GAAI,EAAM,OAAS,EAAM,MAEvB,OADM,GAAW,EAAM,MAAM,QACtB,EAAS,OAAS,EAAM,QAC7B,EAAS,QAAQ,GAEnB,EAAQ,EAAQ,EAAO,GAKzB,OAFM,GAAa,EAAM,MACnB,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAM,OAAS,EAAG,GAAK,EAAG,IACrC,GAAI,EAAW,KAAO,EAAM,GAC1B,EAAK,GAAK,UACD,EAAM,MAAM,KAAO,EAC5B,KAAM,IAAI,OACN,mBAAmB,EAAM,6BAA6B,EAAK,MAGnE,GAAM,GAAO,EAAK,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,KAAI,OAAO,SAAA,GAAK,MAAA,IAAK,IAEjE,GAAI,EAAK,SAAW,EAClB,MAAO,IAAM,GAGf,GAAM,GAAU,SAAC,GAA2B,MAAA,GAAQ,KAAK,EAAO,IAE1D,EAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,MAAK,EAAE,WAAU,GAElD,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAa,MAGb,IAAc,EAAG,CAAC,aAAY,KC5F3C,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAyCA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,eACnC,EACK,GAAgB,EACjB,WAAM,MAAA,uBAAuB,EAAY,aACrC,+BAA8B,EAAY,QAElD,GAAM,GAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,aAAY,EAAE,aAAY,GAE3D,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAM,EAAQ,KAAK,EAAI,EAAc,GAC3C,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAc,EAAG,CAAC,aAAY,KC7B3C,YAAmB,GACjB,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCYrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCDrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KC5BrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCnCrC,AA8DA,YACI,EAAiB,EACjB,EACA,EACA,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GACjE,EACI,IAAY,EAAQ,MAAM,GAC1B,WAAM,MAAA,oCAAoC,EAAO,gBAC7C,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,GAAO,GACL,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GACxC,EACY,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAS,EAAG,CAAC,QAAO,KC3EjC,YACI,EAAiB,EAA6B,EAC9C,EACA,EAAiC,EACjC,GADA,AAAA,IAAA,QAAA,GAAA,OAAiC,AAAA,IAAA,QAAA,GAAA,GAEnC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,MAG9C,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAQ,GACjD,WAAM,MAAA,yDACF,eAAc,EAAM,kBAAkB,EAAQ,OACtD,EACI,IAAe,MACf,WAAM,MAAA,sCACF,EAAU,0CAElB,GAAM,GAAW,EACb,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,KAC7D,EAAU,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAA4B,CAAC,EAAG,GAChC,EAA8B,CAAC,EAAG,GAElC,EAAmB,OAEnB,EAAM,GACP,EAAuB,EAAuB,EAAS,EACxD,EAAkB,EAAW,GAEjC,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAGxC,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,QAGhD,IAAS,EAAG,CAAC,QAAO,KCpHjC,AAoDA,YACI,EAAmE,EACnE,EAAkB,EAClB,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KAC1D,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,KAG9C,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,GAAM,GAAU,IAAe,OAAS,EAAS,GAAK,EAAS,GACzD,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAY,EAEZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEE,EAAM,EAAQ,eAAe,EAAM,EAAQ,GAEjD,SAAK,CAAC,EAAM,IAEL,GAGH,EAAoC,CAAC,GAAI,EAAM,OAAM,GAErD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,EAAE,WAAY,GAEtD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAqB,GAEzB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxF3D,YACI,EAAiB,EACjB,EACA,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IACH,EAAa,EAAI,EAAS,EAAS,EAAK,OAAQ,MAGzC,IAAkB,EAAG,CAAC,iBAAgB,KCxDnD,AA8DA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAExE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAA+B,EAAS,GACxC,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,QACf,WAAM,MAAA,sCACF,EAAU,4CAElB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,GAC5C,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GAExC,EAAqB,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,GAEzD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,KAGI,IAAS,EAAG,CAAC,QAAO,KC3HjC,AA6CA,YACI,EAGA,EAAO,EAAkB,EACzB,GACF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACvE,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,EAAO,KAGzD,GAAM,GAAU,EAAS,GACnB,EAAW,EAAK,MAAM,GAC5B,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,GAEhD,MAAO,GAAQ,eAAe,EAAM,EAAQ,IAGxC,EAAoC,CAAC,GAAI,GAEzC,EAAkC,CAAC,IAAG,GAEtC,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxE3D,YACI,EAAiB,EACjB,EAGA,EAA0C,GAC5C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IAAoB,EAAa,EAAI,EAAS,EAAS,MAGnD,IAAkB,EAAG,CAAC,iBAAgB,KCtDnD,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AAqDA,YACI,EAAsB,EAAU,EAAmB,GAA7B,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,IAAmB,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAG,MAC9C,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,IAE5B,GAAM,GAAe,GAAiB,EAAG,EAAG,MAAM,GAC9C,EAAQ,EAAQ,OAAO,EAAW,EAAc,EAAW,GAG/D,GAFA,EAAK,CAAC,IAEF,GAAe,MACjB,GAAM,GAAqB,GAAuB,GAClD,EAAQ,GAAU,EAAO,GAE3B,MAAO,IAGP,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,EAAE,UAAS,EAAE,QAAO,GAEpD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KCnFjC,AAiEA,YACI,EAA0B,EAC1B,GAAA,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,gBAE7B,EAAe,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC/D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC9D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAEpE,EACI,EAAc,GAAa,EAC3B,WAAM,MAAA;MACN,EAAW,QAAQ,EAAS;MAC5B,EAAG,QAEP,EACI,EAAa,GAAa,EAC1B,WAAM,MAAA;MACN,EAAU,QAAQ,EAAS;UACvB,EAAG,QAEX,EACK,EAAc,GAAY,KAAe,EAC1C,WAAM,MAAA,8CACF,EAAY,EAAS,WACrB,EAAU,sCAAsC,EAAG,QAE3D,GAAM,GAAiC,SAAA,GACnC,MAAA,GAAQ,aAAa,EAAI,EAAW,IAElC,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAAC,UAAS,EAAE,WAAU,GAEvD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAc,MAGP,IAAe,EAAG,CAAC,cAAa,KCvG7C,AA2EA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,sDACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAElC,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAGlB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,iEACF,mBAAkB,EAAO,mBAAmB,EAAS,OAE7D,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IACE,EAAM,EAAQ,gBAAgB,EAAK,EAAS,GAClD,SAAK,CAAC,EAAK,IACJ,GAGH,EAAsC,CAAC,EAAG,EAAK,OAAQ,GACvD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KC/InD,AA+CA,YAAe,GACb,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAA,GACnC,GAAM,GAAO,EAAQ,EAAI,CAAC,EAAG,OACvB,EAAS,EAAQ,KAAK,GACtB,EAAe,EAAE,MAAK,OAAK,EAAE,OAEnC,MAAO,GAAQ,EAAQ,IAGnB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAO,EAAG,CAAC,MAAK,KChE7B,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,GADA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IACzC,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAU,EAAgB,EAAQ,SAAU,cAElD,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA,gEACC,GAAG,KAAI,OAClB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,4DACC,GAAQ,KAAI,OACvB,EACI,IAAe,OACf,WAAM,MAAA,0DACF,0BAAyB,KAEjC,GAAI,GAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACzD,EAAe,IAGjB,GAAM,GAA2B,CAAC,EAAG,EAAK,OAAQ,GAC5C,EAAyB,CAAC,QAAO,EAAE,IAAG,EAAE,UAAS,GAEjD,EAAM,EAAO,UACH,GAAY,EACZ,GAEhB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAa,EAAG,CAAC,YAAW,KCvGzC,YA2BI,EAAmB,GAGrB,OAFM,GAAS,EAAQ,OACjB,EAAiB,GACd,EAAI,EAAG,EAAI,EAAQ,KAC1B,GAAM,GAAM,EAAS,EAAI,EACnB,EAAI,EAAQ,IAAQ,EACpB,EAAI,EAAS,EAAS,OAAS,EAAI,IAAM,EAC/C,AAAI,EAAI,GAAK,IAAM,GACjB,EAAK,QAAQ,GAGjB,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAQ,EAAQ,EAAQ,OAAS,EAAI,GACrC,EAAU,EAAS,OAAS,EAAI,EAChC,EAAS,EAAS,GACxB,AAAI,IAAS,MAAS,IAAU,GAAK,EAAS,IAC5C,EAAO,QAAQ,GAGnB,MAAO,eAIL,EAAkB,GAIpB,OAHM,GAAmB,GACnB,EAAI,KAAK,IAAI,EAAO,OAAQ,EAAO,QAEhC,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GACnC,AAAI,GAAK,MACP,GAAI,GAEN,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GAInC,GAHA,AAAI,GAAK,MACP,GAAI,GAEF,IAAM,EACR,EAAO,QAAQ,WACN,IAAM,EACf,EAAO,QAAQ,WACN,IAAM,GACf,GAAM,GAAS,wDACR,GAAM,QAAQ,EAAM,KAC3B,KAAM,OAAM,OAEZ,GAAO,QAAQ,GAGnB,MAAO,GCrFT,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,MAAM,EAAI,IAE5D,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,OAG/C,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AAoDA,YACI,EAA8B,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAa,EAAgB,EAAW,YAAa,QAAS,QAI9D,EAAiB,GAA2B,EAAG,MAAO,EAAG,OACzD,EAAgB,GAAY,EAAI,GAChC,EAAgB,GAAY,EAAI,GACtC,AAAI,EAAW,OAAS,GAGtB,EACI,EAAW,MAAM,KAAO,EAAG,MAAM,GACjC,WAAM,MAAA,mEAGZ,AAAI,EAAW,OAAS,GAEtB,GACI,EAAW,MAAO,EAAc,MAAO,oBAG7C,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,OAAO,EAAY,EAAe,GACtD,SAAK,CAAC,IACC,GAEH,EAAyB,CAC7B,UAAW,EACX,EAAG,EACH,EAAG,GAEL,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,OAGA,IAAQ,EAAG,CAAC,OAAM,KC5F/B,AAuCA,YAAsC,GACpC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAA0B,CAAC,EAAG,GACpC,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,IAAK,EAClC,KAAiB,OAEjB,IAAY,EAAG,CAAC,WAAU,KC9CvC,AA0DA,YACI,EAAsB,SAEpB,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAAY,GAAI,EAAI,GACpB,EAAQ,GAAU,GAClB,EAAc,GAAM,EAAI,GAC9B,MAAO,IAAM,EAAa,EAAO,MAGtB,IAAW,EAAG,CAAC,UAAS,KCvErC,AA2CA,YAAc,EAAuB,GACnC,GAAM,GAAM,EAAgB,EAAI,KAAM,OAChC,EAAM,EAAgB,EAAI,KAAM,OAEtC,EACK,GAAI,OAAS,GAAK,EAAI,OAAS,IAAO,GAAI,OAAS,GAAK,EAAI,OAAS,GACtE,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GACjD,EAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GAOvD,GALA,EACI,IAAY,EACZ,WAAM,MAAA,gEACC,GAAO,QAAQ,EAAO,OAE7B,EAAI,OAAS,GAAK,EAAI,OAAS,GACjC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,YACZ,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,CAAC,EAAK,eAClB,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAK,GACzB,MAAO,GAAQ,EAAM,CAAC,EAAK,YAE3B,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAK,GACzB,MAAO,OAIE,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,GACtB,SAAK,CAAC,IACC,GAGH,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCtD3B,AAyCA,YAAgC,GAC9B,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,EACI,EAAG,QAAU,SAAW,EAAG,QAAU,UACrC,WAAM,MAAA,8CAEV,AAAI,EAAG,QAAU,SACf,GAAK,GAAK,EAAI,YAGhB,GAAM,GAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KC1D3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAyCA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GAC3D,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,aAAc,GAEjD,EAAY,GAAQ,EAAG,KAAM,WAAM,MAAA,uCACnC,GAAM,GAAW,EAAG,MAAM,QAC1B,MAAI,GAAO,GAET,GACI,CAAE,GAAG,KAAO,IAAM,EAClB,WAAM,MAAA,iCAAiC,CAAG,GAAG,KAAO,GAAE,KAAK,EAAG,KAAI,MACtE,EAAO,EAAG,KAAO,EAAO,GAE1B,EAAS,OAAO,EAAM,EAAG,GAClB,EAAQ,EAAI,MAGR,IAAa,EAAG,CAAC,YAAW,KC1DzC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAqDA,YAAiC,EAAiB,GAChD,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,OAAQ,GAC3C,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OAEzC,GAAM,GAA0B,SAAC,EAAS,GACxC,GAAM,GAAM,EAAQ,KAAK,EAAI,GAC7B,SAAK,CAAC,IACC,GAGH,EAAe,CAAC,GAChB,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,GAE9B,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,EAAkC,MAG3B,IAAO,EAAG,CAAC,MAAK,KC5E7B,AAwCA,YACI,EAAiB,EACjB,EAKA,GAAA,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAc,MAChB,GAAa,GAIf,OAFM,GAAO,GAAO,CAAC,EAAS,GAAa,GACrC,EAAI,GAAW,EAAa,EAAU,EACnC,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAK,IAAI,EAAG,EAAG,GAEjB,GAAM,GAAgB,EAAQ,EAAK,WAAY,CAAC,EAAS,IACzD,GAAI,GAAc,KAChB,MAAO,GAEP,GAAI,EAAW,SAAW,EACxB,MAAO,IAAK,GAAW,EAAK,GAAI,CAAC,EAAW,GAAI,EAAG,IAC9C,GAAI,EAAW,SAAW,EAE/B,MAAO,IACI,GAAW,GAAW,EAAK,GAAI,GAC/B,CAAC,EAAW,GAAI,EAAW,GAAI,EAAG,IACxC,GAAI,EAAW,SAAW,EAE/B,MAAO,IAAK,GAAW,GAAW,GAAW,EAAK,GAAI,GAAI,GAAI,CACrD,EAAW,GAAI,EAAW,GAAI,EAAW,GAAI,EAAG,IAGzD,KAAM,IAAI,OACN,2CAEA,8BAA8B,EAAmB,OAAM,UAKpD,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAqCA,YACI,EAAoB,EAAsB,GAC5C,GAAM,GAAmB,CAAC,MAAK,EAAE,MAAK,EAAE,MAAK,GAE7C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAO,EAAO,IAAQ,GAAI,KAAM,GACxD,GC3CN,AAqCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC7C/B,AAuBO,GAAM,IAAwB,eASI,GACvC,MAAI,IAAU,GACL,EAEF,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KCpCrD,YA8BI,EAAgB,GAClB,GAAI,GAAO,GACP,EASJ,IAPA,AAAI,GAAU,GACZ,GAAM,EACN,EAAO,IAEP,EAAM,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KAG7C,CAAC,GACN,AAAI,EAAM,GAAe,IAAQ,EAC/B,EAAO,GAEP,EAAM,GAAe,EAAQ,EAAM,GAGvC,MAAO,eAIL,EAAkB,EAAc,GAGlC,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,IAAQ,EACV,EAAS,KAAK,EAAO,IAErB,EAAS,KAAK,GAGlB,MAAO,eAUL,EAAW,EAAiB,GAM9B,OALM,GAAU,EAAE,MAAM,GAElB,EAAwB,GAC1B,EAAY,EACZ,EAAY,EACP,EAAI,EAAG,EAAI,EAAM,IACxB,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAM,IAChC,EAAY,KAAK,EAAQ,MAAM,IAGjC,OAAS,GAAI,EAAO,EAAG,EAAI,EAAE,KAAM,IACjC,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,MAAO,CAAC,UAAS,EAAE,UAAS,EAAE,QAAO,EAAE,YAAW,2GC5FpD,AAoDA,YACI,EAAiB,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAW,EAAgB,EAAS,UAAW,SAAU,SAEzD,EAAyB,CAAC,EAAG,EAAI,QAAS,GAC1C,EAAuB,CAAC,KAAI,GAE5B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAa,GAAe,EAAM,EAAG,OAAO,GAC5C,EAAY,GAAyB,EAAI,EAAU,GAEnD,EACF,EAAQ,OAAO,EAAI,EAAQ,EAAU,CAAC,EAAS,OAAQ,GAC3D,SAAK,CAAC,EAAI,IAEH,EAAQ,EAAK,EAAU,cAGhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC5EjC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,QAAQ,EAAI,IAE9D,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC3DnC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,aAAa,EAAI,GACrC,SAAK,CAAC,EAAI,IACH,GAGH,EAA6B,CAAC,EAAG,EAAI,EAAG,GAE9C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAe,EAAG,CAAC,cAAa,KC/D7C,AAsCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCnD7B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GAEnC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,SAAS,IAAK,EACnC,KAAiB,OAEV,IAAW,EAAG,CAAC,UAAS,KC/CrC,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC/C/B,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC9C/B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAEhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAGH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,YAsCI,EAAyC,GAC3C,GAAM,IAAa,IAAU,IAAU,UAAa,MAAM,QAAQ,KAC9D,IAAU,YACZ,KAAM,IAAI,OACN,kFAGN,GAAI,IAAU,UAAY,GAAa,IACnC,CAAE,aAAiB,aACrB,KAAM,IAAI,OACN,6EAGN,GAAM,GAAkB,GAClB,EAA0B,GAChC,MAAO,IAAW,EAAO,EAAO,EAAe,GCrDjD,AA2CA,YAAsC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aACnC,MAAO,IAAQ,EAAI,GAAO,GAAQ,GAAK,MAG5B,IAAY,EAAG,CAAC,WAAU,KChDvC,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,QAC7B,EAAK,EAAgB,EAAG,IAAK,QACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAE3D,EAAqB,CAAC,EAAG,EAAI,EAAG,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC1D7B,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,UAAU,EAAI,GAClC,SAAK,CAAC,EAAI,IACH,GAGH,EAA0B,CAAC,EAAG,EAAI,EAAG,GAE3C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAY,EAAG,CAAC,WAAU,KC/DvC,YAkCyB,EAAe,EAAc,GACpD,GAAI,GAAO,EACT,KAAM,IAAI,OAAM,4CAGlB,GAAM,GAAuB,CAAC,MAAK,EAAE,KAAI,EAAE,IAAG,GAC9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,SAAS,EAAO,EAAM,IAAM,GAC/C,KAAiB,GAAU,GC1CjC,AA4CA,YACI,EAAiB,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IACzD,GAAM,GAAK,EAAgB,EAAG,IAAK,8BACnC,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA;sBACU,EAAG,KAAI,MAC3B,EACI,GAAW,GACX,WAAM,MAAA,+DACF,gCAA+B,EAAW,OAElD,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,6BACd,EAAK,EAAa,EAAM,EAAO,GAEnC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAoB,CAAC,EAAG,GAExB,EAAkB,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEpD,KAIE,IAA6B,EAAG,CAAC,4BAA2B,KCvFzE,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AA0DA,YAAc,GAEZ,SACI,GAAgB,GAAI,WAAM,MAAA,+CACvB,SAAC,EAAsB,GAE5B,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MACxC,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,WAAa,KAC1D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,kFAGN,GAAW,GACJ,EAAM,MAkCnB,YAAe,GAEb,SACI,GAAgB,GAAI,WAAM,MAAA,gDACvB,SAAC,EAAgC,GACtC,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,qFAGV,GAAM,GAAQ,GAAqB,EAAM,OAAQ,WAAY,MACvD,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,YAAc,KAC3D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,MAAA,OAAA,IAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,iGAGN,GAAW,GACJ,KA8Bb,YAA0D,GAKxD,SACI,GAAgB,GAChB,WAAM,MAAA,uDACH,SAAC,EAAM,GACZ,EACI,YAAa,GACb,WAAM,MAAA,wDACV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,6DACJ,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,UAAW,GACJ,CAAC,KAAM,EAAM,GAAS,MAAK,IAmCtC,YAAyC,GAKvC,SACI,GAAgB,GAChB,WAAM,MAAA,wDACH,SAAC,EAAgB,GACtB,EACI,MAAM,QAAQ,IAAS,EAAK,MAAM,SAAA,GAAO,MAAA,aAAe,KACxD,WAAM,MAAA,uEAEV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,iEACV,GAAM,GAAM,EAAO,UAAU,WAAM,MAAA,GAAC,MAAA,OAAI,IAAO,EAAM,GACrD,MAAI,IAAM,MACR,GACI,EAAI,MAAM,MAAO,EAAG,MACpB,yGAGN,GAAW,EAAI,OACR,GAkCX,YAAuB,EAAiB,GAEtC,EACI,GAAgB,GAChB,WAAM,MAAA,wDACV,EACI,GAAW,MACP,MAAM,QAAQ,IAAY,EAAQ,MAAM,SAAA,GAAK,MAAA,aAAa,MAC9D,WACI,MAAA,kFAGR,GAAM,GAAmB,GAAW,KACpC,GAAI,CAAC,GAEH,EAAU,GACV,OAAW,KAAW,GAAO,oBAC3B,EAAQ,KAAK,EAAO,oBAAoB,IAI5C,GAAM,GACF,EAAmB,EAAQ,OAAO,SAAA,GAAY,MAAA,CAAC,EAAS,YAAa,KAGnE,EAAmB,EAAQ,OACjC,EAAU,EAAQ,OAAO,SAAA,GAAY,MAAA,GAAS,YAC9C,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,kEACF,kCAAiC,EAAgB,kBACjD,eAER,GAAM,GAAmB,GACnB,EAAA,EAAA,UAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,EACI,EAAM,KAAK,SAAA,GAAK,MAAA,IAAK,OACrB,WAAM,MAAA,iMAGV,EACI,EAAM,OAAS,EACf,WAAM,MAAA,iEACF,oBAAmB,EAAM,KAAI,aAErC,GAAM,GAA6B,GACnC,SAAQ,QAAQ,SAAC,EAAG,GAClB,AAAI,EAAM,IAAM,MACd,GAAW,EAAE,MAAQ,EAAM,MAG/B,AAAI,GAAyB,MAG3B,EAAsB,QAAQ,SAAA,GAAK,MAAA,GAAW,EAAE,MAAQ,OAEnD,CAAC,MAAK,EAAE,MAAO,GA2CxB,YAAsC,GAEpC,MAAO,GAAO,WAAW,GAG3B,YAAoB,GAClB,GAAM,GAAmB,EAAM,OAAO,SAAA,GAAK,MAAA,IAAK,OAAM,OACtD,GAAI,EAAmB,EACrB,KAAM,IAAI,OACN;sEC7XR,AAuCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAuB,CAAC,EAAG,GACjC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,IAAI,IAAK,EAC5B,KAAiB,OAEV,IAAM,EAAG,CAAC,KAAI,KC/C3B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,SAAS,GAC7B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAW,EAAG,CAAC,UAAS,KChDrC,AAyCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAK7B,EAAW,GAAW,SAAC,GAI3B,GAAM,GAAQ,GAAI,GAAS,GAAI,KAEzB,EAAW,SAAC,GAChB,GAAM,GAAO,EAAI,EAAI,GAAQ,GAAI,KACjC,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAEL,IAAa,EAAG,CAAC,YAAW,KC9DzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAoB,EAAM,EAAG,OAC1C,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAS,OAG1D,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GAAgB,GAClB,EAAI,MAAO,GAAoB,EAAM,EAAG,QAC5C,EAAM,EAAQ,EAAK,GACnB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,iBAAkB,EAAM,SAAQ,GAEzD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCnE3B,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAGhB,GAAM,GAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IACN,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,IAAI,EAAW,GACnC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAEzB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KC/F3B,AAsDA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAAU,EAAgB,EAAQ,SAAU,cAKlD,GAHA,AAAI,IAAS,IACX,GAAO,EAAQ,KAAO,GAEpB,IAAS,EAAQ,KAAO,EAC1B,KAAM,OACF,gEACA,oBAAmB,EAAQ,KAAI,iBAAiB,IAGtD,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACX,EAAO,GAAI,EAAQ,EAAM,IACzB,EAAU,GAAI,EAAQ,GACtB,EACF,GAAI,GAAK,EAAS,WAAY,GAAI,GAAI,GAAI,GAAU,EAAM,KAC9D,SAAK,CAAC,IACC,GAGH,EAA2B,CAAC,OAAQ,GACpC,EAAyB,CAAC,KAAI,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAY,MAGZ,IAAa,EAAG,CAAC,YAAW,KCpFzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAO,GAAI,EAAI,EAAM,IACrB,EAAI,GAAI,EAAI,GACZ,EAAI,GAAI,GACR,EAAI,GAAI,EAAG,GACX,EAAI,GAAI,GACR,EAAM,GAAI,EAAQ,EAAM,EAAE,OAAQ,GAExC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,MAGI,IAAY,EAAG,CAAC,WAAU,KChFvC,AAyCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA2B,CAAC,EAAG,EAAI,EAAG,GAE5C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,WAAW,EAAI,IAClC,EAAgC,KAAiB,OAGjD,IAAa,EAAG,CAAC,YAAW,KCtDzC,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,WAAW,IAAK,EACnC,KAAiB,OAGV,IAAa,EAAG,CAAC,YAAW,KC9CzC,AAwCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,QAC1C,EAAK,EAAgB,EAAG,IAAK,YAAa,QAChD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA0B,CAAC,EAAG,EAAI,EAAG,GAC3C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IACjC,EAAgC,KAAiB,OAEjD,IAAY,EAAG,CAAC,WAAU,KCnDvC,AA0CA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,UAA2B,EAAG,MAAO,EAAG,OAGjC,GAAW,GAAU,EAAG,GAAI,GAAW,GAAW,EAAG,QAGjD,IAAa,EAAG,CAAC,YAAW,KCpDzC,AAwDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAC7B,EAAY,EAEd,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAEA,EAEJ,MAAI,GAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UAC9C,EAAI,EAAI,QAER,EAAI,EAAQ,QAAQ,EAAK,GAG3B,EAAK,CAAC,EAAK,IAEJ,GAGH,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAEhE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KCrHnC,AA2EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GAJiB,AAAA,IAAA,QAAA,GAAA,CAA+C,EAAG,EAAG,IAGtE,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE/B,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAErB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GACE,EAAI,EAAQ,UAAU,EAAK,GACjC,SAAK,CAAC,EAAK,IACJ,GAGH,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAE/D,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KClJvC,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAK,EAAgB,EAAG,IAAK,qBAE7B,EAAkC,CAAC,EAAG,GACtC,EACuB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,oBAAmB,GAErE,EAAS,EAAO,UACH,GAAmB,EACnB,GAEnB,MAAO,CAAC,OAAQ,EAAO,GAAI,QAAS,EAAO,OAGhC,IAAoB,EAAG,CAAC,mBAAkB,KC7EvD,YAsCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAM,EAAO,WACpB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAoB,GAAc,GAAQ,GACzD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC7C1C,YAuCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAK,EAAO,WACnB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAmB,GAAc,GAAQ,GACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC9C1C,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAS,GAA0B,EAAG,MAAO,GAC7C,EAAc,EAAO,GACrB,EAAa,GAAc,GAI3B,EAAW,GAAW,SAAC,GAC3B,GAAM,GAAmB,GAAO,GAE1B,EAAU,EAAiB,QAAU,EAAE,MACzC,EACA,GAAK,EAAG,EAAiB,OACvB,EAAM,GAAI,EAAS,GACnB,EAAQ,GAAI,EAAK,EAAM,GAEvB,EAAW,SAAC,GAChB,GAAM,GAAkB,EAAE,MAAM,QAChC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,GAAI,EAAI,EAAY,GAAK,EAAE,MAAO,YAAa,GAC5D,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAGL,IAAO,EAAG,CAAC,MAAK,KCpC7B,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAGpD,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GACF,GAA+B,EAAI,MAAO,GAC9C,EAAM,EAAQ,EAAG,GACjB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAGhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAEH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,AAuDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAEH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AAmCA,YAAmC,GACjC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAQ,GACR,EAAe,CAAC,GAChB,EAA2B,GACjC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,SAAK,CAAC,IACC,EAAQ,OAAO,IACrB,CAAC,EAAG,GAAK,KAAiB,SAAU,EAAO,EAAc,MAGjD,IAAS,EAAG,CAAC,QAAO,KC9CjC,AA4CA,YACI,EAAsB,EACtB,GADsB,AAAA,IAAA,QAAA,GAAA,MACtB,AAAA,IAAA,QAAA,GAAA,IACF,EAAI,EAAgB,EAAG,IAAK,WAC5B,GAAM,GAAO,GAAe,EAAM,EAAE,OAC9B,EAAQ,GAAK,EAAG,EAAM,GACxB,EAAgB,EAAM,MAC1B,AAAK,GACH,GAAgB,GAAqB,EAAM,MAAO,IAEpD,GAAM,GACF,GAAO,GAAI,GAAK,EAAG,WAAY,EAAQ,EAAO,KAC5C,EAAW,GAAK,EAAY,EAAM,GACxC,MAAO,CAAC,KAAM,EAAO,SAAQ,MAGlB,IAAU,EAAG,CAAC,SAAQ,KChBnC,YACI,EAA2B,EAC3B,EACA,GAOF,OANM,GAAQ,EAAgB,EAAM,OAAQ,gBACtC,EAAK,GAAqB,EAAG,IAAK,gBAClC,EAAK,GAAqB,EAAG,IAAK,gBAEpC,EAAQ,EACN,EAAY,GACT,EAAI,EAAG,EAAI,EAAU,OAAQ,KACpC,GAAM,GAAS,EAAU,GAAG,EAAO,EAAG,GAAI,EAAG,IAC7C,EAAU,KAAK,EAAO,IACtB,EAAU,KAAK,EAAO,IACtB,EAAQ,EAAO,GAIjB,OAFM,GAAmB,GACnB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,GAAK,EACzC,EAAK,KAAK,EAAU,IACpB,EAAK,KAAK,EAAU,EAAI,IAE1B,MAAO,CAAC,EAAM,MAEH,IAAe,EAAG,CAAC,cAAa,KCpE7C,AA4CA,YACI,EAAsC,EAAoB,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,eAC5C,EAAc,EAAQ,KACtB,EAAW,EAAQ,KACzB,GAAI,EAAc,EAChB,KAAM,IAAI,OACN,+DACG,GAAW,MAEpB,GAAI,EAAW,EACb,KAAM,IAAI,OAAM,gDAAgD,GAElE,EAAO,GAAQ,KAAK,SACpB,GAAM,GACF,IAAa,EAAI,EAAQ,EAAS,CAAC,EAAG,KAAO,EAC3C,EAAM,EAAO,cACf,SAAA,GAAW,MAAA,GAAQ,YAAY,EAAU,EAAY,EAAY,IACjE,CAAC,SAAQ,IAEb,MAAO,KAAa,EAAI,EAAQ,EAAK,CAAC,EAAI,OAAqB,KAGpD,IAAc,EAAG,CAAC,aAAY,KCpE3C,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,GAAY,MAAA,GAAQ,SAAS,EAAI,IAEjE,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KC1DrC,AAwCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCrD7B,AA0CA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAI,EAAG,QAAU,aACf,GAAM,GAAI,GAAS,GAAK,IAClB,EAAI,GAAU,GAAK,IACzB,MAAO,IAAQ,EAAG,GAGpB,MAAO,GAAQ,SAAS,IAGpB,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCtBrC,YACI,EAAyB,GAC3B,GAAM,GAAM,EAAgB,EAAI,KAAM,gBAChC,EAAM,EAAgB,EAAI,KAAM,gBAEtC,EACI,EAAI,OAAS,GAAK,EAAI,OAAS,EAC/B,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,EAAQ,EAAK,CAAC,EAAG,KAC9B,MAAO,IAAO,EAAM,MAGT,IAAe,EAAG,CAAC,cAAa,KCtD7C,AAmDA,YACI,EAAiB,EAAmC,GAAA,AAAA,IAAA,QAAA,GAAA,GACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAA0B,SAAC,EAAS,GACxC,SAAK,CAAC,IACC,EAAQ,IAAI,EAAI,EAAU,IAG7B,EAAoB,CAAC,SAAQ,EAAE,cAAa,GAC5C,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAM,EAAG,CAAC,KAAI,KC5C3B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,qDACH,GAAI,EAAG,CAAC,GAAW,MAGf,IAAQ,EAAG,CAAC,OAAM,KCT/B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCV/B,YACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,EACvD,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCX/B,YACI,EACA,EAIA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,GACnD,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCxC/B,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAEnC,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,cAAc,EAAG,KAAI,kCACvB,EAAW,SAEnB,EACI,EAAS,SAAW,EAAW,OAC/B,WAAM,MAAA,qBACF,EAAS,OAAM,kCAAkC,EAAW,SAEpE,EACI,EAAG,MAAM,OACL,SAAC,EAAG,EAAG,GACL,MAAI,GAAI,GAAK,GAAK,EAAW,OACpB,GACD,GAAI,EAAS,EAAI,GAAG,GAAK,EAAS,EAAI,GAAG,IACtC,EAAW,EAAI,KACnB,EAEA,GAET,IACJ,WAAM,MAAA,4BAA4B,EAAG,MAAM,MAAM,GAAE,kBAC/C,EAAS,WAAU,qCACnB,EAAW,aAEnB,GAAM,GAA0B,SAAA,GAC5B,MAAA,GAAQ,eAAe,EAAI,EAAY,IAErC,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,SAAQ,GAExD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,KCrHjD,AAwDA,YACI,EAAqB,EACrB,EAA0B,EAC1B,EAAqC,GACvC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,AAAI,GAAW,MACb,GAAU,GAEZ,AAAI,IAAQ,GACV,GAAM,SAGR,GAAM,GAAK,EAAgB,EAAO,IAAK,WACnC,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,yDACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAC1C,EACF,CAAC,EAAS,eAAgB,EAAS,eAOnC,EACJ,AAAI,IAAQ,OACV,EAAc,GACV,CAAC,EAAS,aAAc,EAAS,aAAc,GAEnD,EAAc,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,IAG7B,GAAM,GAAgB,EAAS,KAAO,GAAK,EAAS,KAAO,EACrD,EAAA,GAAA,CAAA,EAAA,SAAA,EAAA,SAAA,EAAA,GAAC,EAAA,EAAA,GAAiB,EAAA,EAAA,GAElB,EAAe,EAAgB,EAAM,QACrC,EACF,EAAgB,EAAM,GAAe,EAAK,EAAU,GAElD,EAAY,IAAgB,MAC9B,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAChD,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAC9C,EAAI,IAEJ,EAAM,EAAgB,EAAI,GAAe,EAAG,EAAU,GAE5D,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,EAMT,YACI,EAA8B,EAC9B,GACF,GAAM,GAAW,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,EAAa,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KACpC,EAAiB,EAAW,OAAO,EAAU,GAC7C,EAAc,EAAW,IAAI,SAAC,EAAG,GAAM,MAAC,GAAI,EAAe,GAAK,GAAK,IACrE,EAAS,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAY,KAClD,EAAW,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAS,GAAI,EAAO,MACzD,EAAQ,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAG,EAAY,MACvD,MAAO,CAAC,EAAU,GAMpB,YACI,EAA+B,GAGjC,GAAM,GAAqB,EAAY,IAAI,SAAC,EAAG,GAC7C,MAAO,GAAK,GAAI,GAAM,GAAS,GAAK,KAEhC,EAAgB,EAAmB,IAAI,SAAA,GAAK,MAAA,GAAI,IAIhD,EAAgB,EAAc,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAI,KACtD,EAAc,EAAc,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAc,KAClE,MAAO,GAAc,IAAI,SAAC,EAAG,GAC3B,MAAO,CAAC,EAAc,GAAI,EAAY,SAI7B,IAAO,EAAG,CAAC,MAAK,KChK7B,AAsDA,YACI,EAAyB,SACvB,EAAQ,EAAgB,EAAM,OAAQ,OACtC,EAAO,EAAgB,EAAK,MAAO,OACvC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,GAAM,GAAoB,CAAC,EAAG,EAAO,EAAG,GAClC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,EAAO,GAC7B,SAAK,CAAC,EAAO,EAAM,IACZ,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AA0CA,YAAkC,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAS,EAAgB,EAAO,QAAS,SAEzC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAGH,EAAsB,CAAC,EAAG,EAAI,MAAO,GAC3C,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AA8DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,QAE3B,EAA+B,SAAC,GACpC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAEhB,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,KAAK,EAAW,GACpC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAGzB,MAAO,IAGH,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,EAAE,SAAQ,GAExC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAO,EAAG,CAAC,MAAK,KChG7B,AAiCA,YACI,EAAoB,EACpB,GACF,GAAM,GAAO,GAAc,GACvB,EAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,IAEd,MAAO,GAAO,WAAW,EAAQ,EAAO,MAG7B,IAAO,EAAG,CAAC,MAAK,6OC1B7B,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,cAEJ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,wBACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,AAyCA,GAAA,IAAA,WAUE,WACI,EAAc,EAAsB,EACpC,EAAqB,GACvB,KAAK,KAAO,EACZ,KAAK,OAAS,EACd,KAAK,MAAQ,EACb,KAAK,QAAU,IACf,KAAK,UAAY,EACjB,AAAI,KAAK,WACP,MAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,EACvC,KAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,GAEzC,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,OAAS,GAAgB,EAAU,YAInC,SAAA,UAAA,UAAP,WACE,GAAI,CAAC,MAAM,KAAK,UACd,GAAM,GAAQ,KAAK,QACnB,YAAK,QAAU,IACR,EAKT,OAFI,GAAiB,EACjB,EAAU,GACP,CAAC,IACN,GAAI,GAAE,OAAU,EAAE,OAAU,EAAC,OAC7B,EACE,GAAK,EAAI,KAAK,SAAW,EACzB,EAAK,EAAI,KAAK,SAAW,EACzB,EAAI,EAAK,EAAK,EAAK,QACZ,GAAK,GAAK,IAAM,GAEzB,GAAM,GAAM,KAAK,KAAK,GAAO,KAAK,IAAI,GAAK,GAC3C,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EACzC,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EAEzC,AAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,GAAU,IAId,MAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,MAAK,QAAU,KAAK,aAAa,IAE5B,KAAK,aAAa,IAInB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,OAAS,MAAQ,KAAK,QAAU,UAChC,EAEF,KAAK,MAAM,IAIZ,EAAA,UAAA,iBAAR,SAAyB,GACvB,MAAO,IAAS,KAAK,OAAS,GAAS,KAAK,OAEhD,KAIA,GAAA,WASE,WACI,EAAe,EAAc,EAC7B,GACF,KAAK,MAAQ,EACb,KAAK,KAAO,EAAI,EAChB,KAAK,MAAQ,EAEb,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,MAAQ,GAAgB,EAAU,YACvC,KAAK,MAAQ,GAAI,IAAY,EAAG,EAAG,EAAO,GAAO,KAAK,SAEtD,AAAI,EAAQ,EACV,KAAK,EAAI,EAAS,EAAI,EAEtB,KAAK,EAAI,EAAS,EAAI,EAExB,KAAK,EAAI,EAAI,KAAK,KAAK,EAAI,KAAK,GAI3B,SAAA,UAAA,UAAP,WAEE,OADI,GAAY,EAAY,EAAY,EAAW,EAAW,KAE5D,EACE,GAAI,KAAK,MAAM,YACf,EAAI,EAAK,KAAK,EAAI,QACX,GAAK,GAMd,GALA,GAAK,EAAI,EACT,EAAK,EAAI,EACT,EAAK,EAAK,KAAQ,EAAK,EACvB,EAAM,GAAM,EAAO,KAAK,EAAK,GAAI,EAAI,KAAK,IAAI,IAC9C,EAAI,KAAK,QACL,EAAI,GAAM,KAAK,IAAI,GAAK,EAC1B,MAGJ,SAAK,EAAI,KAAK,KAAQ,KAAK,EAAI,EAC/B,AAAI,KAAK,MAAQ,GACf,IAAK,KAAK,IAAI,KAAK,QAAS,EAAI,KAAK,QAEhC,KAAK,aAAa,IAGnB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,QAAU,UACV,EAEF,KAAK,MAAM,IAEtB,KAEA,GAAA,WAME,WACI,EAAS,EAAS,EAClB,GAFJ,GAAA,GAAA,KAaE,GAZE,AAAA,IAAA,QAAA,GAAA,GAAS,AAAA,IAAA,QAAA,GAAA,GAoBL,KAAA,eAAiB,WACrB,MAAC,GAAK,OAAS,MAAQ,EAAK,QAAU,WAnBxC,KAAK,IAAM,EACX,KAAK,MAAQ,EAAM,EACnB,KAAK,MAAQ,EACb,AAAI,GAAQ,MACV,GAAO,KAAK,UAEd,AAAI,MAAO,IAAS,UAClB,GAAO,EAAK,YAGV,CAAC,KAAK,kBAAoB,KAAK,OAAS,EAC1C,KAAM,IAAI,OACN,0BAA0B,EAAG,MAAM,EAAG,gCAE5C,KAAK,OAAS,GAAgB,GAOxB,SAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,iBACA,EAEF,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,WACE,MAAO,MAAK,aAAa,KAAK,IAAM,KAAK,MAAQ,KAAK,WAE1D,KCxNA,AAwCA,YACI,EAAoB,EAAe,EACnC,EAAsC,GAOxC,GARqC,AAAA,IAAA,QAAA,GAAA,GACnC,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAQ,MACV,GAAO,GAET,AAAI,GAAS,MACX,GAAQ,WAEN,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,OAAM,yBAAyB,GAI3C,OAFM,GAAS,GAAI,IAAU,EAAO,EAAM,EAAO,GAC3C,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAc,EAAG,CAAC,aAAY,KC5D3C,AAuCA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,yBAAyB,GAK3C,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAuB,GAC1D,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAe,EAAG,CAAC,cAAa,KCtD7C,AA4CA,YACI,EAAoB,EAAY,EAAY,EAC5C,GADoB,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,WAI9C,OAFM,GAAM,GAAO,EAAO,GACpB,EAAS,GAAI,IAAc,EAAQ,EAAQ,KAAM,GAC9C,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAgB,EAAG,CAAC,eAAc,KCvD/C,YAwCyB,EAAsB,GAC7C,GAAc,GACd,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,EAC3B,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAkB,KACxB,MAAO,IAAW,EAAQ,EAAO,EAAe,GC/ClD,YA8CI,EAAe,EAAc,EAC7B,GACF,GAF+B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAA,WACE,IAAS,EACX,KAAM,IAAI,OAAM,8BAGlB,GAAM,GAA+B,WACnC,GAAM,GAAgB,IAAU,EAC1B,EAA8B,EAAQ,GAAQ,EAAO,EACrD,EAA8B,EAAO,GAAS,EAAO,EAE3D,GAAI,GAAiB,GACjB,EACF,MAAO,IAAM,CAAC,GAAI,GAGpB,GAAM,GAAc,KAAK,IAAI,KAAK,KAAM,GAAO,GAAS,IAClD,EAAS,GAAoB,EAAa,GAEhD,AAAI,EAAO,GAAS,IAAS,GAG3B,GAAO,IAGT,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,IAGpB,EAAoB,CAAC,MAAK,EAAE,KAAI,EAAE,KAAI,EAAE,MAAK,GAEnD,MAAO,GAAO,cACH,EAAS,GAAiB,KAAiB,GAC3C,GCnFb,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAE7B,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,WAAW,GAC/B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAa,EAAG,CAAC,YAAW,KChDzC,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,KAAK,IAGhB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,MAAM,IAGjB,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC5D/B,AA6DA,YACI,EAAiB,GACnB,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAA+B,SAAC,GACpC,GAAM,GAAO,GAAe,EAAM,EAAG,OACrC,GAAI,EAAG,OAAS,EACd,MAAO,IAAM,GAEf,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,MAAO,GAAQ,EAAK,EAAG,QAGnB,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,KAAM,GAEnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAS,MAGT,IAAU,EAAG,CAAC,SAAQ,KClFnC,AA6BA,YAAoB,GAClB,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCrCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAGV,IAAQ,EAAG,CAAC,OAAM,KChD/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GAGH,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KCe7B,YACI,EAAiB,EACjB,EAAsC,EACtC,EAAqB,EACrB,GADqB,AAAA,IAAA,QAAA,GAAA,CAAqC,EAAG,IAC7D,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EACF,EAAgB,EAAiB,kBAAmB,mBAClD,EACF,EAAgB,EAAiB,kBAAmB,mBAEpD,EAAM,EACN,EAAe,GAMnB,GALA,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGvD,IAAe,OACjB,KAAM,IAAI,OACN,sFAIN,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,MAAM,KAAO,EAC9B,WACI,MAAA,qEACA,wBAAuB,EAAiB,MAAM,GAAE,OACxD,EACI,EAAiB,MAAM,KAAO,EAC9B,WAAM,MAAA,+DACF,8BAA6B,EAAiB,MAAM,GAAE,OAE9D,GAAM,GAAa,EAAiB,MAAM,GACpC,EAAoB,EAAiB,MAAM,GACjD,EACI,EAAiB,MAAM,KAAO,EAAa,EAC3C,WACI,MAAA,qEACA,YAAW,EAAa,EAAiB,MACzC,YAAW,EAAiB,MAAM,GAAE,OAE5C,GAAM,GAAY,GACd,EAAK,EAAkB,EAAS,EAAK,EAAY,GAC/C,EAAkB,EAClB,EACF,GAAO,EAAW,EAAkB,EAAiB,QAAS,GAElE,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KCxFnD,YACI,EAAsB,uHAClB,SAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aAEnC,EACI,EAAG,QAAU,EAAG,MAChB,WAAM,MAAA,kDACF,EAAG,MAAK,YAAY,EAAG,MAAK,OAEpC,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAEtE,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAExD,CAAA,EAAM,EAAG,eAAjB,SAAQ,EAAA,OACA,CAAA,EAAM,EAAG,eAIvB,IAJM,EAAQ,EAAA,OACR,EAAO,GAAI,KAAI,GAEjB,EAAa,EACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,AAAK,EAAK,IAAI,EAAM,KAClB,IAMJ,IAFM,EAAS,GAAI,IAAa,CAAC,GAAa,EAAG,OAC3C,EAAU,GAAI,IAAa,CAAC,GAAa,SACtC,EAAI,EAAG,EAAI,EAAG,EAAI,EAAM,OAAQ,IACvC,AAAK,EAAK,IAAI,EAAM,KAClB,GAAO,OAAO,GAAK,EAAM,GACzB,EAAQ,OAAO,GAAK,EACpB,KAGJ,MAAA,CAAA,EAAO,CAAC,EAAO,WAAY,EAAQ,qBAExB,IAAiB,GCxF9B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC7C7B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AA6BA,YACI,EAAwB,EAAe,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,CAAC,GAAQ,CAAC,OAEhB,IAAU,EAAG,CAAC,SAAQ,KCtCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AAgDA,YAAoC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxD,GAAM,GAAU,EAAgB,EAAQ,SAAU,UAAW,WAK7D,GAHA,AAAI,IAAQ,IACV,GAAM,EAAQ,KAAO,GAEnB,IAAQ,EAAQ,KAAO,EACzB,KAAM,OACF,4DACA,oBAAmB,EAAQ,KAAI,gBAAgB,IAGrD,GAAM,GAAwB,CAAC,OAAQ,GACjC,EAAsB,CAAC,IAAG,GAEhC,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAI,EAAQ,QAAQ,EAAS,GACnC,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAU,EAAG,CAAC,SAAQ,KCzEnC,AAyCA,YAAc,GACZ,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,qDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAoB,CAAC,MAAK,GAEhC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAU,EAAM,KAAK,EAAO,GAC5B,EAAS,EAAQ,IAAI,GAC3B,MAAO,GAAO,QAAQ,EAAM,QAC3B,EAAgC,KAAqB,OAG7C,IAAM,EAAG,CAAC,KAAI,KC5D3B,AA0CA,YAAe,GACb,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,sDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAqB,CAAC,MAAK,GAEjC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAoB,EAAQ,EAAO,CAAC,EAAO,IAC3C,EAAS,EAAQ,KAAK,GAC5B,MAAO,GAAQ,EAAQ,EAAM,QAC5B,EAAgC,KAAqB,OAG7C,IAAO,EAAG,CAAC,MAAK,KC7D7B,AAgDA,YAAgB,GACd,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EACvB,EACJ,GAAI,GAAsB,GACxB,GAAM,GAAe,EAAQ,EAAO,CAAC,EAAO,IAC5C,EAAM,GAAK,QAIX,GAAM,GAAc,CAAC,EAAO,EAAK,GAAqB,IAChD,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IACzC,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IAEzC,EACF,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACjE,EAA0B,EAC5B,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACnE,GAAO,KAEL,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EACF,EAAQ,GAAQ,EAAG,GAAI,CAAC,EAAY,GAAI,EAAY,KACxD,EAAM,GAAK,GAIb,GAFA,EAAM,GAAK,GAEP,EAAM,OAAS,GAAK,EAAM,MAAM,KAAO,GACzC,GAAM,GAAO,EACP,EAAQ,EAAM,MAAM,GAC1B,EAAM,EAAQ,EAAK,CAAC,EAAO,EAAI,MAAM,GAAK,EAAO,EAAI,MAAM,KAC3D,EAAK,UAEP,MAAO,MAGI,IAAQ,EAAG,CAAC,OAAM,iBC3D3B,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAa,GACjB,GAAI,MAAQ,IAAqB,SAC/B,EACI,EAAE,MAAM,GAAQ,IAAoB,EACpC,WAAM,MAAA,kDACV,EACI,GAAI,OAAM,GAAiB,KAAK,EAAE,MAAM,GAAQ,QAEpD,GAAM,GAAY,EAAgB,OAAO,SAAC,EAAO,GAC/C,MAAI,KAAU,IACZ,IAAS,GAEJ,GACN,GACH,EACI,GAAa,EACb,WAAM,MAAA,4DACV,GAAM,GAAW,EAAgB,QAAQ,IAGzC,GAAI,IAAa,IACf,GAAM,GAAQ,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,EAAI,IAC/D,EAAgB,GAAY,EAAE,MAAM,GAAQ,EAE9C,EACI,EAAE,MAAM,KAAU,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IACvD,WAAM,MAAA,gEACV,EAAa,EAGf,MAAO,GC1DT,AA+DA,YACI,EAAsB,EAAkC,GAAA,AAAA,IAAA,QAAA,GAAA,GAC1D,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAG,OAAO,GACvC,EAAa,GAAiB,EAAI,EAAiB,GACzD,MAAO,GAAQ,MAAM,EAAI,EAAY,IAGjC,EAAuB,CAAC,EAAG,GAC3B,EAAoB,CAAC,gBAAe,EAAE,KAAI,GAEhD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCjF/B,AA+CA,YAAe,EAAe,GAC5B,EACI,EAAM,QAAU,UAChB,WAAM,MAAA,mDAAmD,EAAM,QAEnE,GAAI,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACpD,EAAQ,EAAM,KAAO,EAEvB,EACJ,GAAI,GAAa,MAAQ,EAAY,GAEnC,GAAM,GAAQ,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAC7B,EAAO,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAClC,EAAK,EAAM,MAAM,OAAS,GAAK,EAC/B,EAAgB,GAAM,EAAO,EAAO,GACpC,EAAqB,UACZ,GAAa,MAAQ,EAAY,GAE1C,GAAM,GAAa,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KACxC,EAAW,EAAM,MAAM,OAAS,GAAK,EAAY,EACjD,EAAgB,GAAO,CAAC,EAAO,GAAM,IAAc,EAAM,MAAM,OAAS,GACxE,EAAqB,MAErB,GAAgB,EAIlB,GAAM,GAAa,GAAU,GACvB,EACF,EAAQ,GAAQ,EAAe,GAAa,CAAC,EAAO,IAElD,EAAM,GAAI,GAGV,EAAO,KAAK,MAAM,EAAqB,GAAK,EAC5C,EAAa,GAAK,GAClB,EAAa,GAAK,GAClB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GACxB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GAExB,EAAc,EAAc,MAAM,QACxC,SAAY,EAAc,MAAM,OAAS,GAAK,EAEvC,EACH,GAAQ,EAAqB,GAAI,EAAqB,IAAK,MAGpD,IAAO,EAAG,CAAC,MAAK,KClG7B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AAoDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,qBAC7B,EAAK,EAAgB,EAAG,IAAK,qBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,kBAAkB,EAAI,GAC1C,SAAK,CAAC,EAAI,IACH,GAGH,EAAkC,CAAC,EAAG,EAAI,EAAG,GAC7C,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAmB,MAGnB,IAAoB,EAAG,CAAC,mBAAkB,KC1EvD,AAwCA,YAAoC,EAAsB,GACxD,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,MAAO,GAAQ,EAAI,GAAa,EAAG,MAAO,GAAM,aAGrC,IAAU,EAAG,CAAC,SAAQ,KC7CnC,AAyCA,YACI,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAChC,GAAM,GAAW,GAAqB,EAAS,UAAW,SAK1D,GAHA,EACI,EAAS,QAAU,EAAG,WAAM,MAAA,yCAE5B,EAAS,SAAW,EACtB,MAAO,IAAW,EAAS,GAAI,GAGjC,GAAM,GAAO,EAAS,GAAG,KACnB,EAAQ,EAAS,GAAG,MACpB,EAAQ,EAAS,GAAG,MAE1B,EAAY,GAAQ,EAAM,WAAM,MAAA,uCAEhC,EAAS,QAAQ,SAAA,GACf,GACI,EAAO,EAAE,MACT,yDACJ,EACI,IAAU,EAAE,MACZ,WAAM,MAAA,4DAGZ,GAAM,GAAkB,EAAS,IAAI,SAAA,GAAK,MAAA,IAAW,EAAG,KAOxD,MAAO,IAAO,EAAiB,MAGpB,IAAQ,EAAG,CAAC,OAAM,KC7E/B,AAwCA,YAAiC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAEhB,IAAO,EAAG,CAAC,MAAK,KClD7B,AA+DA,YACI,EAAsB,EAAiB,EAAe,EACtD,EAAe,EAAa,EAAkB,EAC9C,GADA,AAAA,IAAA,QAAA,GAAA,GAAe,AAAA,IAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAAkB,AAAA,IAAA,QAAA,GAAA,GAC9C,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAE3B,EAA+B,SAAC,GACpC,AAAI,GAAW,MACb,GAAU,GAAI,OAAM,EAAM,SAG5B,GAAM,GAAe,GAAW,GAChC,GAAI,EAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAI,IAAiB,GAAK,IAAgB,EACxC,KAAM,IAAI,OACN,iEAGN,GAAI,IAAiB,GAAK,IAAmB,EAC3C,KAAM,IAAI,OACN,oEAGN,GAAM,GAAsB,EAAG,KAAO,EAAM,OAGtC,EAAa,GAAW,GACxB,EAAW,EAAG,MAAM,QAC1B,EAAW,QAAQ,SAAA,GACjB,EAAM,GAAQ,EACd,EAAI,GAAQ,EACZ,EAAS,OAAO,EAAM,EAAG,KAE3B,EAAK,EAAQ,EAAI,GAEX,GAAA,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GACJ,EAAA,EAAA,MACA,EAAA,EAAA,IACA,EAAA,EAAA,QAKF,EAAQ,EACR,EAAM,EACN,EAAU,EAEV,GAAM,GAAa,GAAW,GAE9B,EAAW,QAAQ,SAAA,GACjB,EAAI,GAAQ,EAAM,GAAQ,EAC1B,EAAQ,GAAQ,IAIlB,GAAM,GAAO,GAAgB,EAAO,EAAK,GAEnC,EAAW,EAAK,OAAO,SAAC,EAAG,GAAS,MAAA,GAAW,QAAQ,KAAU,KAEjE,EAAa,EAAQ,MAAM,SAAA,GAAK,MAAA,KAAM,IAC5C,GAAI,EACF,MAAO,GAAQ,GAAM,EAAI,EAAO,GAAO,GAGzC,GAAM,GAAM,EAAQ,aAAa,EAAI,EAAO,EAAK,GACjD,MAAO,GAAQ,EAAK,IAGhB,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAC/B,MAAK,EACL,IAAG,EACH,QAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,YAAW,EACX,eAAc,GAGhB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAe,EAAG,CAAC,cAAa,KCvJ7C,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,kEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,gFAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,sEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,wEAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GCjElD,YAgDI,EACA,EACA,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,0EAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,SAAQ,GACJ,EACG,GAAW,EAAQ,EAAO,EAAe,GCpElD,AAmDA,YACI,EAAiB,EAAO,GAAP,AAAA,IAAA,QAAA,GAAA,GAAO,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAK,EAAgB,EAAG,IAAK,QACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAU,EAAG,MAAM,EAAG,MAAM,OAAS,GAC3C,GAAI,EAAI,EACN,KAAM,IAAI,OACN,uDAAuD,EAAO,KAC9D,YAAW,IAGjB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,EAAC,EAAE,OAAM,GAE7B,EAAA,EAAA,cAAA,SAAA,GAAA,MAAA,GAAA,KAAA,EAAA,EAAA,IAAA,EAAA,KAAA,GAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAKf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAO,EAAG,CAAC,MAAK,KC3E7B,AA4CA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,qCAKlB,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAsB,GACzD,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAkB,EAAG,CAAC,iBAAgB,KC3DnD,AA6EA,YACI,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAEnB,GAAM,GAAK,EAAgB,EAAG,IAAK,SAAU,MAC7C,EAAO,EAAG,KAAO,EAAG,WAAM,MAAA,yCAE1B,GAAM,GAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAC1B,EAAA,EAAA,UAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAS,EAAG,CAAC,QAAO,KC3FjC,AA8CA,YACI,EAAiB,EAAiC,GACpD,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EACF,EAAgB,EAAY,aAAc,qBAAsB,SACpE,EAAO,GAAM,GAAc,WAAM,MAAA,qCAEjC,GAAM,GAAmC,CAAC,EAAG,EAAI,WAAY,GACvD,EAAiC,CAAC,YAAW,GAE7C,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,mBAAmB,EAAI,EAAa,GACxD,SAAK,CAAC,IACC,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAoB,MAGpB,IAAqB,EAAG,CAAC,oBAAmB,KCnEzD,AA2CA,YAAkB,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACtC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,EACI,GAAQ,CAAC,EAAG,MAAM,QAAU,EAAO,EAAG,MAAM,OAC5C,WACI,MAAA,UAAU,EAAI,gBAAgB,EAAG,MAAM,OAAM,KAAK,EAAG,MAAM,OAAM,MACzE,AAAI,EAAO,GACT,IAAQ,EAAG,MAAM,QAEnB,GAAM,GAAuB,CAAC,MAAO,GAC/B,EAAqB,CAAC,KAAI,GAC1B,EAAiC,SAAC,GACpC,MAAA,GAAQ,QAAQ,EAAI,IACxB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAU,EAAG,CAAC,SAAQ,KC7DnC,YAsCI,EAAyB,EAAkB,EAC3C,GADyB,MAAA,KAAA,QAAA,GAAA,IAEpB,EAAO,aAAa,EAAc,EAAW,EAAM,GCxC5D,YAuB0B,EAAqB,GAE7C,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,AAAI,EAAS,IACX,EAAQ,KAAK,GAOjB,OAHM,GAAW,GAAO,EAAW,SAE7B,EAAM,GAAO,CAAC,EAAQ,OAAQ,EAAU,QAAS,SAC9C,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAM,EAAS,WAAW,EAAQ,IAClC,EAAS,EAAI,EAAU,OAC7B,EAAI,OAAO,IAAI,EAAK,GAEtB,MAAO,GAAI,WCEb,YAA2B,uGACnB,SACF,EAAgB,EAAW,YAAa,aAAc,QAC7C,CAAA,EAAM,EAAW,eAAxB,SAAO,EAAA,OACP,EAAM,GAAU,EAAW,MAAO,GACxC,AAAI,IAAc,GAChB,EAAW,UAEb,CAAA,EAAO,WAGI,IAAa,GCpD1B,AA6CA,YACI,EAA2B,EAC3B,2HAcF,IAbM,EAAU,EAAgB,EAAQ,SAAU,YAC5C,EAAQ,EAAgB,EAAM,OAAQ,WAAY,QAElD,EAAW,GAAe,EAC1B,EAAU,EAAM,KAChB,EAAc,EAAQ,MAE5B,EAAY,EAAU,EAAG,WAAM,MAAA,0BAC/B,GACI,EAAY,MAAM,EAAU,EAAW,GAAU,EAAM,MACvD,qEAEA,EAAc,EACT,EAAI,EAAU,EAAI,EAAW,EAAS,IAC7C,GAAe,EAAY,GAEvB,SACF,EAAY,MAAM,EAAG,GAChB,OAAO,CAAC,GAAc,EAAY,MAAM,EAAW,IACtD,EAAiB,EAAQ,EAAS,GAClC,EAAe,EAAQ,EAAO,CAAC,KACX,CAAA,EAAM,GAAW,WAArC,SAAoB,EAAA,OACpB,EAAU,GAAQ,EAAmB,CAAC,IAEtC,EAAM,GAAO,EAAgB,EAAS,GAG5C,AAAI,IAAW,GACb,EAAQ,UAEV,AAAI,IAAS,GACX,EAAM,UAER,EAAQ,UACR,EAAe,UACf,EAAa,UACb,EAAkB,UAElB,CAAA,EAAO,WAGI,IAAmB,GCzFhC,AAuCA,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAK,EAAgB,EAAG,IAAK,kBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,6BAC/B,GAAS,EAAI,GAYtB,YAAuC,EAAiB,GACtD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAK,EAAgB,EAAG,IAAK,cACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,yBAC/B,GAAK,EAAI,GAGlB,YAAwC,EAAiB,GACvD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,eAC7B,EAAK,EAAgB,EAAG,IAAK,eACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,0BAC/B,GAAM,EAAI,GAGnB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAK,EAAgB,EAAG,IAAK,mBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,8BAC/B,GAAU,EAAI,GAGvB,YAA0C,EAAiB,GACzD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,4BAC/B,GAAQ,EAAI,GAGrB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EAAK,EAAgB,EAAG,IAAK,sBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,iCAC/B,GAAa,EAAI,MAGb,IAAc,EAAG,CAAC,aAAY,KAC9B,GAAqB,EAAG,CAAC,oBAAmB,KAC5C,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAkB,EAAG,CAAC,iBAAgB,KACtC,GAAa,EAAG,CAAC,YAAW,KAC5B,GAAiB,EAAG,CAAC,gBAAe,KCpHjD,AA2CA,YAAsC,EAAiB,GACrD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAS,GAC7C,UACI,6EAGJ,GAAuB,EAAK,MAAO,EAAI,MAAO,wBACvC,GAAI,EAAM,GAanB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,6BACpC,EAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,2BACpC,GAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAWjB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAWrB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAarB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,2BAC7B,EAAK,EAAgB,EAAG,IAAK,2BACnC,UACI,EAAG,MAAO,EAAG,MAAO,sCACjB,GAAkB,EAAI,MAGlB,IAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAA0B,EAAG,CAAC,yBAAwB,KACtD,GAAY,EAAG,CAAC,WAAU,KC5NvC,AAwEA,YACI,EAAsB,EACtB,EAA8B,GADR,AAAA,IAAA,QAAA,GAAA,aACtB,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IAChC,EAAI,EAAgB,EAAG,IAAK,QAE5B,GAAM,GAAO,GAAS,EAAG,EAAK,GAC1B,EAAgB,EAAK,MACzB,GAAI,GACF,GAAM,GAAO,GAAe,EAAM,EAAE,OACpC,EAAgB,GAA+B,EAAK,MAAO,GAE7D,MAAO,GAAQ,EAAM,GAGvB,YACI,EAAW,EAAkB,GAC/B,GAD+B,AAAA,IAAA,QAAA,GAAA,MAC3B,EAAE,OAAS,EACb,MAAO,IAAI,GAIb,GAAI,EAAE,OAAS,GAAK,IAAS,KAC3B,MAAO,IAAS,EAAQ,EAAG,CAAC,KAAM,EAAG,GAIvC,GAAI,EAAE,OAAS,GAAK,MAAO,IAAS,UAChC,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,aAAe,IAAM,EAE7B,MAAO,IAAK,GAAI,GAAI,GAAI,GAAI,GAAO,EAAG,UAAW,IAGnD,KAAM,IAAI,OAAM,qCAAqC,GAIvD,GAAI,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,GAAK,GAE7C,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,OAAS,IAAM,YAEvB,MAAO,IAAK,GAAI,GAAO,GAAI,IAG7B,KAAM,IAAI,OAAM,qCAAqC,GAGvD,KAAM,IAAI,OAAM,gCAAgC,MAGrC,IAAO,EAAG,CAAC,MAAK,KC3I7B,AA2DA,YACI,EAAiB,EAAiB,EAClC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAS,EAAgB,EAAO,QAAS,iBAE/C,GAAiB,EAAI,GACrB,EACI,GAAiB,EAAG,MAAO,EAAG,OAAQ,WAAM,MAAA,8BAEhD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAI,EAAK,GAE3B,EAAS,EAAI,GAAI,EAAI,GAAK,GAC9B,GAAI,GACF,EACI,GAAQ,KAAM,WAAM,MAAA,mDACxB,GAAM,GAAQ,EAAgB,EAAM,OAAQ,iBAC5C,EAAS,GAAI,EAAQ,GAAI,EAAK,GAAI,EAAQ,KAE5C,MAAO,IAAI,EAAI,MAGJ,IAAgB,EAAG,CAAC,eAAc,KCnF/C,AA+CA,YACI,EAA4B,EAC5B,GACF,GAAM,GAAW,EAAgB,EAAS,UAAW,YAAa,SAC5D,EAAW,EAAgB,EAAS,UAAW,aACrD,GAA8B,EAAU,EAAU,GAElD,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,UAAU,EAAU,EAAU,IAGzC,EAA0B,CAAC,QAAS,EAAU,QAAS,GACvD,EAAwB,CAAC,MAAK,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAW,MAGX,IAAY,EAAG,CAAC,WAAU,iBClCnC,EAAuB,EAAsB,EAC7C,GACF,GAAI,EAAc,QAAU,QAC1B,KAAM,IAAI,OACN,2DACA,uBAAsB,EAAc,MAAK,MAE/C,GAAI,EAAc,KAAO,EACvB,KAAM,IAAI,OACN,uDACA,mBAAkB,EAAc,MAAK,MAG3C,GAAM,GAAW,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAC7D,EAAU,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAElE,GAAI,EAAY,SAAW,EACzB,KAAM,IAAI,OACN,iDACA,KAAI,EAAY,OAAM,gBAAgB,EAAO,MAGnD,GAAM,GAAY,EAAa,KAC/B,GAAI,CAAE,GAAa,OAAS,GACtB,EAAa,OAAS,GAAK,IAAc,GAC7C,KAAM,IAAI,OACN,oCACG,GAAa,MAAK,sBAAsB,EAAQ,MAGzD,GAAI,EAAa,QAAU,EAAc,MACvC,KAAM,IAAI,OAAM,qDC/DpB,AAkEA,YACI,EAAkC,EAClC,EAA0B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC5B,GAAM,GACF,EAAgB,EAAe,gBAAiB,gBAAiB,SAC/D,EACF,EAAgB,EAAc,eAAgB,iBAC5C,EAAgB,EAClB,EAAc,eAAgB,gBAAiB,EAAc,OAEjE,GACI,EAAgB,EAAe,EAAa,GAEhD,GAAM,GAA8B,CAClC,cAAe,EACf,aAAc,EACd,aAAc,GAGV,EAA4B,CAAC,YAAW,GAE9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,cACf,EAAgB,EAAe,EAAa,IAChD,EAAgC,KAAiB,GACjD,MAGO,IAAgB,EAAG,CAAC,eAAc,KC9F/C,AA8DA,YAAmB,EAAsB,GACvC,GAAM,GAAW,EAAgB,EAAS,UAAW,WAAY,SAC3D,EAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,GACpC,MAAO,GAAQ,SAAS,EAAI,IAGxB,EAAyB,CAAC,OAAQ,EAAI,QAAS,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KAAqB,OAGvD,IAAW,EAAG,CAAC,UAAS,KC5ErC,YA4B8B,EAAW,GACvC,GAAI,GAAc,KAChB,MAAO,GAAE,MAAM,QAEjB,GAAI,GAAiB,EAAE,MAAO,GAC5B,MAAO,GAET,GAAI,EAAE,MAAM,SAAW,EAAW,QAEhC,OADM,GAAyB,GACtB,EAAI,EAAG,EAAI,EAAE,MAAM,OAAQ,IAClC,AAAI,EAAW,IAAM,MAAQ,EAAE,MAAM,IAAM,KACzC,EAAa,KAAK,EAAE,MAAM,IAE1B,EAAa,KAAK,EAAW,IAGjC,MAAO,GAGT,MAAO,GC/CT,AAoDA,YACI,EAAsB,EAAc,EACpC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAUnC,GARA,EACI,EAAG,QAAU,UACb,WAAM,MAAA,8DACF,sBAAqB,EAAG,MAAK,sBACrC,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,qDAAqD,EAAI,MAE/D,IAAS,EACX,MAAO,aAAa,GAAS,EAAG,QAAU,EAG5C,GAAM,GAAc,GAAc,EAAI,GAChC,EAAW,EAAI,EACf,EAAa,GACf,GAAM,GAAI,GAAc,EAAa,EAAG,EAAG,UAAW,GAAO,IAC7D,GAEJ,MAAO,GAAI,EAAI,MAGJ,IAAU,EAAG,CAAC,SAAQ,KC9EnC,YAoBoC,GAElC,MAAO,MAAK,MAAM,KAAK,IAAI,EAAG,KAAK,KAAK,KAAK,IAAI,GAAS,KAAK,IAAI,kBAIjE,EAAsB,EAAW,GAGnC,OAFM,GAAO,EAAI,EAAe,EAC1B,EAAY,GAAI,cAAa,GAC1B,EAAI,EAAG,EAAI,EAAc,EAAE,GAClC,GAAM,GAAU,EAAM,KAAK,GAAK,EAAM,GAAe,EAAO,GAC5D,EAAU,GAAK,EAAI,EAAI,KAAK,IAAI,GAElC,MAAO,IAAS,EAAW,WCjC7B,AAwCA,YACI,EAA2B,EAAuB,GAAA,MAAA,KAAA,QAAA,GAAA,wHAC9C,SAAe,EAAgB,EAAa,cAAe,UAC3D,EAAW,EAAgB,EAAS,UAAW,UAErD,EACI,EAAa,KAAO,EACpB,WAAM,MAAA,+DACF,YAAW,EAAa,QAChC,EACI,EAAa,KAAO,IAAM,EAAS,KACnC,WAAM,MAAA,mFAEC,GAAa,KAAI,qBAAqB,EAAS,QAC1D,GACI,EAAa,MAAM,MAAM,EAAG,EAAa,MAAM,OAAS,GACxD,EAAS,MACT,2FAEE,EAAU,EAAa,MAAM,EAAa,MAAM,OAAS,GAC/D,EACI,EAAI,GAAK,GAAK,EACd,WAAM,MAAA,iEACF,eAAc,EAAO,cAAc,KAEnB,CAAA,EAAM,EAAa,eAArC,SAAkB,EAAA,OACJ,CAAA,EAAM,EAAS,eAOnC,IAPM,EAAc,EAAA,OAId,EAAgB,CAAC,EAAgB,OAAS,EAAS,GAAlD,EAAK,EAAA,GAAE,EAAI,EAAA,GACZ,EAAY,GAAuB,OAAQ,GAExC,EAAI,EAAG,EAAI,EAAO,KAIzB,IAHM,EAAS,EAAI,EACb,EAAO,EAAgB,SAAS,EAAQ,EAAS,GACjD,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAKzC,IAHA,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAErC,EAAU,GAAK,EACN,EAAI,EAAG,EAAI,EAAG,IACrB,GAAI,EAAU,GAAG,QAAU,EAAY,IACrC,EAAU,GAAK,EACf,OAKN,MAAI,KAAgB,GAClB,EAAa,UAEf,AAAI,IAAY,GACd,EAAS,UAIX,CAAA,EAAO,GAAO,EAAW,EAAS,MAAO,iBAG9B,IAAc,GCtG3B,AAiDA,YACI,EAAM,EAAO,EACb,EACA,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE5D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GAC3D,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAY,GACxB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAY,GAAE,OACtD,EACI,IAAa,EAAY,GACzB,WAAM,MAAA,0CAA0C,EAAQ,UACpD,mCAAkC,EAAY,GAAE,QACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EACZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,GAAI,GAClD,EAC0B,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,GAE1E,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAsB,GAG5B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC/G7D,YAgCI,EAAY,EAAW,GACzB,GAAI,GAAc,MAAQ,IAAe,SACvC,MAAO,GAET,GAAI,IAAe,OACjB,MAAO,GAAI,EAAI,GAAK,IAEtB,KAAM,IAAI,OACN,gDAAgD,EAAU,iBAK5D,EAAc,GAChB,GAAI,GAAM,EACJ,EACF,GAAgC,EAAK,MAAO,EAAa,OAC7D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,mBAIvB,EAAW,EACX,GACF,GAAI,IAAe,SACjB,MAAO,GACF,GAAI,IAAe,OACxB,MAAO,IAAK,GACP,GAAI,IAAe,MACxB,MAAO,IAAI,GACN,GAAI,IAAe,QACxB,MAAO,IAAM,GACR,GAAI,IAAe,QACxB,MAAO,IAAM,EAAG,GAElB,KAAM,IAAI,OAAM,4BAA4B,EAAU,KAIjD,GAAM,IAAa,SAAC,EAAuB,GAChD,GAAM,GAAe,EAAgB,EACrC,MAAO,CAAC,GAAgB,IAAe,UC3EzC,AA8FA,YAAmD,MACjD,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAeA,GAFA,EAAa,GAAc,SAEvB,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,6DACC,GAAI,KAAI,OACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,8DACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,6DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,OACf,WAAM,MAAA,sCACF,EAAU,2CAElB,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,GAEnD,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACpB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAGlB,EAAe,GAAqB,EAAI,EAAG,GAEjD,EACI,GAA4B,GAC5B,WAAM,MAAA,oEAEF,uDAAsD,EAAS,OAEvE,GAAM,GACF,GAAoB,EAAI,MAAO,EAAc,EAAS,EAAS,GAC7D,EACF,GAAqB,EAAK,EAAc,EAAQ,MAAO,EAAS,GAC9D,GAAgB,CAAC,EAAM,GAE7B,GAAI,GAAS,MACX,GAAM,IAAU,GAAqB,EAAO,GAC5C,GAAI,KAAK,IAEX,MAAO,KAGH,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,YAAY,CAC9B,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA4B,CAChC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAGpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAS,EAAG,CAAC,aAAY,KC5QtC,AAyBA,YACI,EAAM,EAAO,EACb,GACF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,yBAAyB,EAAK,EAAM,IAE1C,EAAoD,CAAC,EAAG,EAAK,GAAI,GAEvE,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,IAGN,GAAM,IACT,EAAG,CAAC,qCAAoC,KChD5C,AAyBA,YACI,EAAmE,EACnE,EAAkB,GACpB,GAAI,GAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,wBAAwB,EAAM,EAAQ,IAE5C,EAAmD,CAAC,GAAI,GAExD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,IAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IACT,EAAG,CAAC,oCAAmC,KCnD3C,AAyFA,YAA4D,MAC1D,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAaA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,0DACF,iBAAgB,EAAQ,KAAI,OACpC,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,4DACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAClC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,EACI,GAAyC,EAAS,GAClD,WACI,MAAA,oEACA,sBAAqB,EAAO,mBAAmB,EAAS,OAEhE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,0BAAyB,EAAe,gBAAgB,EAAG,OAGrE,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IAEA,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,0BAG/C,GAAM,GAAO,SAAC,EAAc,GAC1B,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAS,OACd,GAAA,GAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAElB,EAAe,GAAqB,EAAI,EAAG,GAE3C,EAAO,GACR,EAAiB,MAAO,EAAc,EAAqB,GAC1D,EAAY,GACd,EAAiB,EAAe,EAAqB,MAAO,GAEhE,GAAI,GAAQ,MACV,GAAM,GAAU,GAAqB,EAAO,GAC5C,MAAO,CAAC,EAAM,EAAW,GAE3B,MAAO,CAAC,EAAM,IAGV,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,qBAAqB,CACvC,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAAqC,CACzC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAEpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAkB,EAAG,CAAC,sBAAqB,KClQxD,AAwDA,YAAwC,SACtC,EAAA,EAAA,EACA,EAAA,EAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAUA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GAAc,EAAG,EAAG,EAAY,GAC7C,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WACI,MAAA,qEACA,iBAAgB,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE9C,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAU,UACrD,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,wCAAwC,EAAW,UAClD,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACnC,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAErC,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAU,EAAM,QAG5D,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACnB,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAAK,GAAA,EAAA,GAAG,GAAA,EAAA,GAId,GACF,GAAqB,EAAQ,EAAI,GAAE,OAAQ,GAAG,GAC9C,GACA,GAgBJ,GAdA,AAAI,CAAC,GAAc,CAAC,EAClB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAM,KACzC,AAAI,CAAC,GAAc,EACxB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,GAAc,EAAK,GAAM,KACzC,AAAI,GAAc,CAAC,EACxB,IAAO,GAAc,EAAK,GAAc,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAO,KAE/C,IAAO,GAAc,EAAK,GAAc,GAAM,IAC9C,GAAO,GAAc,GAAc,EAAK,GAAM,KAG5C,GAAQ,MACV,GAAM,IAAU,GAAqB,GAAO,IAC5C,MAAO,CAAC,GAAM,GAAM,QAEpB,OAAO,CAAC,GAAM,KAIZ,EAA+B,SAAC,GACpC,GAAM,GAAI,EAAQ,iBAAiB,CACjC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA6B,CACjC,EAAG,EACH,EAAG,EACH,KAAM,EACN,uBAAwB,GAEpB,EAA2B,CAAC,WAAU,EAAE,WAAU,EAAE,WAAU,GAIpE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAe,GACxC,GAAM,GAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,IAET,CAAC,MAAO,EAAQ,EAAK,GAAW,SAAU,KAEvD,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAe,EAAe,GAC5C,GAAM,IAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,GAAK,IAEd,CAAC,MAAO,EAAQ,GAAK,GAAW,SAAU,KAGvD,MAAO,GAAiB,EAAK,EAAK,IAI/B,GAAM,IAAS,EAAG,CAAC,aAAY,KClOtC,+DCAA,AAiCA,YAAwB,GACtB,MAAO,IAAa,EAAc,IAAM,KAEnC,GAAM,IAAgB,EAAG,CAAC,eAAc,KCpC/C,AAiCA,YAAqB,GACnB,MAAO,IAAa,EAAc,GAAK,IAGlC,GAAM,IAAa,EAAG,CAAC,YAAW,KCrCzC,AAyCA,YACI,EAAkB,EAAqB,EAAmB,EAC1D,GAD0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,GAGF,OAFI,GAAQ,EACN,EAAmB,GAClB,EAAQ,GAAe,EAAO,MACnC,EAAO,KAAK,GAAM,EAAQ,EAAO,IACjC,GAAS,EAGX,GAAI,EACF,KAAO,EAAQ,EAAO,OACpB,GAAM,GAAU,EAAQ,EAAe,EAAO,KACxC,EAAM,GAAO,CACjB,GAAM,EAAQ,EAAO,EAAc,GAAS,GAAK,CAAC,GAAS,KAE7D,EAAO,KAAK,GACZ,GAAS,EAIb,MAAI,GAAO,SAAW,EACb,GAAS,GAAI,CAAC,EAAG,IAGnB,EAAQ,GAAO,GAAS,CAAC,EAAO,OAAQ,IAE1C,GAAM,IAAQ,EAAG,CAAC,OAAM,KCpE/B,AA4CA,YACI,EAAkB,EAAqB,EACvC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,AAAI,GAAa,MACf,GAAY,GAAoB,IAKlC,OAHM,GAAe,GAAM,EAAQ,EAAa,GAC1C,EAAiB,EAAI,EAAc,EAAS,IAC5C,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAa,MAAM,GAAI,IACzC,EAAO,KACH,GAAK,GAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,EAAG,IAAe,IAE5D,MAAO,IAAO,GAET,GAAM,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAmDA,YACI,EACA,EACA,EACA,EACA,EACA,GAEF,GAAM,GAAS,EAAgB,EAAO,QAAS,iBACzC,EAAS,EAAgB,EAAO,QAAS,gBAAiB,WAC1D,EAAU,EAAgB,EAAQ,SAAU,gBAAiB,SACnE,EAAS,GAAU,WACnB,EAAqB,GAAsB,EAE3C,GAAM,GAAW,EAAO,MAAM,GAE9B,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OACnC,EACI,EAAO,OAAS,GAAK,EAAO,MAAM,KAAO,EACzC,WAAM,MAAA,oDAAoD,EAAQ,OAC9D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAQ,OAAS,GAAK,EAAQ,MAAM,KAAO,EAC3C,WAAM,MAAA,qDAAqD,EAAQ,KAC/D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,iEACF,WAAU,EAAS,OAAM,OACjC,EACI,EAAS,IAAM,GAAK,EAAS,IAAM,EACnC,WAAM,MAAA,2CAA2C,IACrD,EACI,IAAW,YAAc,IAAW,UACpC,WAAM,MAAA,+CAA+C,IAEzD,GAAM,GAAiC,SAAC,GAAY,MAAA,GAAQ,cACxD,EAAQ,EAAQ,EAAS,EAAU,EAAQ,IAEzC,EACoB,CAAC,MAAO,EAAQ,MAAO,EAAQ,OAAQ,GAC3D,EAA4B,CAAC,OAAM,EAAE,mBAAkB,EAAE,SAAQ,GACjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GACJ,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KCtG/C,AAiCA,YAAwB,GACtB,GAAM,GAAS,EAAgB,EAAO,QAAS,gBAAiB,WAEhE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAA8B,CAAC,MAAO,GACtC,EACF,EAAO,UAAU,GAAe,EAAgC,IACpE,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KC/C/C,AA4CA,YACI,EAA4B,EAC5B,EACA,GADA,AAAA,IAAA,QAAA,GAAA,GACA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,mBAAoB,WAEnE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,mDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAAiC,CAAC,MAAO,GACzC,EAA+B,CAAC,QAAO,EAAE,UAAS,EAAE,OAAM,GAC1D,EAAM,EAAO,UACf,GAAkB,EAClB,GACJ,MAAO,GAGF,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC/DrD,AAoBA,YACI,EAAiB,EAAkB,EACnC,EAAsB,EAAwB,GAMhD,AAAI,GAAgB,MAClB,GAAe,IAEjB,AAAI,GAAkB,MACpB,GAAiB,OAAO,mBAE1B,AAAI,GAAgB,MAClB,GAAe,GAGjB,GAAM,GAAW,EAAM,MAAM,GAC7B,SAAgB,KAAK,IAAI,EAAe,GAExC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAClE,EACI,EAAM,OAAS,EACf,WAAM,MAAA,+CAA+C,EAAM,KAAI,MACnE,EACI,EAAM,MAAM,KAAO,EACnB,WACI,MAAA,oDAAoD,EAAM,MAAM,KACxE,EAAY,EAAO,OAAS,EAAG,WAAM,MAAA,+BACrC,EACI,EAAO,MAAM,KAAO,EACpB,WAAM,MAAA,sDAAsD,EAAQ,KAChE,YAAW,EAAO,MAAM,MAChC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAC3D,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GC3DnE,AA0BA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBAC1B,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAExB,GAAM,GAAQ,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,GAC1D,MAAO,GAAO,cACV,SAAA,GAAK,MAAA,GAAE,kBACH,EAAQ,EAAS,EAAe,EAAc,IAClD,CAAC,MAAO,EAAQ,OAAQ,GAAU,KAAiB,GACnD,GAGC,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC/CvD,YA+BI,EAAU,EAAY,GACxB,GAAM,GAAQ,GAAa,EAAK,EAAS,GACnC,EAAiB,EAAQ,EAAI,CAAE,GAAQ,GAAK,EAClD,EAAI,OAAO,EAAgB,EAAG,eAmB5B,EAAU,EAAW,GACvB,MAAO,IAAc,EAAK,EAAQ,GAAc,IAUlD,YAA8B,EAAM,GAClC,MAAO,GAAI,EAAI,EAAI,EAAI,EAAI,GAAK,EAGlC,YACI,EAAU,EAAW,GAKvB,OAJI,GAAO,EACP,EAAQ,EAAI,OACZ,EAAS,EACT,EAAQ,GACL,EAAO,IACZ,EAAS,EAAS,GAAQ,IAAU,GACpC,GAAM,GAAgB,EAAW,EAAQ,EAAI,IAC7C,AAAI,EAAgB,EAClB,EAAO,EAAS,EAEhB,GAAQ,EAGR,EAAQ,CAAC,GAIb,MAAO,GAAQ,EAAO,CAAC,EAAO,ECvFhC,YAoCI,EAAmB,EAAoB,EACvC,EAAsB,GACxB,MAAO,IACI,EAAO,EAAQ,EAAe,EAAc,EAC5C,GACC,4BAIV,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAC5C,EAAsB,GACtB,EAA6C,gBAK/C,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAAgB,EAC5D,IAGN,YACI,EAAmB,EAAoB,EACvC,EAAsB,EAAwB,EAC9C,EAA4B,EAC5B,GADA,AAAA,IAAA,QAAA,GAAA,IAA4B,AAAA,IAAA,QAAA,GAAA,IAC5B,AAAA,IAAA,QAAA,GAAA,IAKF,OAFM,GAAa,GAEV,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,EAAO,GAAK,GACd,EAAW,KAAK,CAAC,MAAO,EAAO,GAAI,SAAU,EAAG,mBAAoB,IAIxE,EAAW,KAAK,IAShB,OALM,GAAQ,EAAe,EAAK,IAAO,EAAgB,EAEnD,EAA4B,GAC5B,EAA2B,GAE1B,EAAgB,OAAS,GAAiB,EAAW,OAAS,IACnE,GAAM,GAAY,EAAW,MACtB,EAAA,EAAA,MAAsB,EAAA,EAAA,SAAU,EAAA,EAAA,mBAEvC,GAAI,EAAgB,EAClB,MAUF,OADI,GAAkB,GACb,EAAI,EAAgB,OAAS,EAAG,GAAK,EAAoB,EAAE,GAClE,GAAM,GAAM,GAAsB,EAAO,EAAU,EAAgB,IAEnE,GAAI,GAAO,GACT,EAAkB,GAClB,MAMF,GAHA,EAAU,MACN,EAAU,MAAQ,GAAe,EAAc,EAAO,GAEtD,EAAU,OAAS,EACrB,MAWJ,EAAU,mBAAqB,EAAgB,OAE/C,AAAK,GAGH,CAAI,EAAU,QAAU,EACtB,GAAgB,KAAK,GACrB,EAAe,KAAK,EAAU,QACzB,AAAI,EAAU,MAAQ,GAG3B,GAAa,EAAY,EAAW,KAM1C,GAAM,GAAe,EAAgB,OAC/B,EAAa,EAAgB,EAEnC,AAAI,GAAsB,EAAa,GACrC,GAAgB,KAAI,MAApB,EAAwB,GAAI,OAAM,GAAY,KAAK,IACnD,EAAe,KAAI,MAAnB,EAAuB,GAAI,OAAM,GAAY,KAAK,KAGpD,GAAM,GACe,CAAC,gBAAiB,GAAS,EAAiB,UAEjE,MAAI,IACF,GAAO,eAAoB,GAAS,EAAgB,YAGtD,AAAI,GACF,GAAO,aAAkB,GAAO,EAAc,UAGzC,EAGT,YAA+B,EAAmB,EAAW,GAC3D,GAAM,GAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAS,GAAQ,GAAU,GAAQ,GACnC,EAAS,GAAQ,GAAU,GAAQ,GACzC,GAAI,GAAS,GAAK,GAAS,EACzB,MAAO,GAET,GAAM,GAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAmB,EAAkB,GACnE,KAAK,IAAI,EAAmB,EAAkB,GAClD,MAAO,GAAoB,GAAQ,EAAQ,GAO7C,YAAwB,EAAsB,EAAe,GAC3D,GAAM,GAAS,KAAK,IAAI,EAAQ,EAAM,GACtC,MAAO,IAAO,EAAe,EAAS,EAGxC,YAA6B,EAAe,GAK1C,MAAQ,GAAG,MAAQ,EAAG,OAChB,EAAG,QAAU,EAAG,OAAW,EAAG,SAAW,EAAG,SC7MpD,AA2CA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,wHACpB,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAED,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,GACxD,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAyB,GC1EtC,AAyDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,GACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEtB,GAAM,GAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EACF,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GAExD,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,eAAgB,EAAO,IAGtD,GAAM,IAA6B,EAAG,CAAC,4BAA2B,KChCzE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,wGACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEC,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,EACpD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAkC,GCxF/C,AAmDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAEzB,EAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EAAkC,CACtC,cAAe,EACf,aAAc,EACd,eAAgB,EAChB,mBAAkB,GAGd,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,aAAc,EAAO,IAGpD,GAAM,IAA0B,EAAG,CAAC,yBAAwB,KCnCnE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,+GACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAG3B,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBADxC,SACF,EAAA,OADG,EAAS,EAAA,GAAE,EAAU,EAAA,GAMtB,EAAM,GACR,EAAW,EAAY,EAAgB,EAAe,EACtD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAA+B,GChF5C,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,2DACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,6DACC,GAAI,OAEf,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAG9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GACZ,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,eACX,EAAa,EAAW,EAAU,IAGlC,EAA+B,CAAC,OAAQ,GACxC,EAA6B,CAAC,aAAY,EAAE,KAAI,GAEhD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCpFjD,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,yBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,kEACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WACI,MAAA,oEACG,GAAI,OACf,EACI,EAAQ,QAAU,WAAa,EAAQ,QAAU,QACjD,WAAM,MAAA,qDAEV,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAE9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GAEZ,EAAsC,CAAC,OAAQ,GAC/C,EAAoC,CAAC,aAAY,EAAE,KAAI,GAEvD,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,sBACX,EAAa,EAAW,EAAU,IAGlC,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAwB,EAAG,CAAC,uBAAsB,KCxF/D,AAyEA,YACI,EAAiB,EAAkB,GACrC,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAClE,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAElE,GAAM,GAAK,EAAgB,EAAG,IAAK,YAEnC,EACI,EAAG,MAAQ,EACX,WAAM,MAAA,4CAA4C,EAAG,KAAI,MAE7D,GAAM,GAAQ,EAAG,MACX,EAAA,EAAA,MAAA,MAAA,IAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,kDAAiD,EAAC,OAExD,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,qDAAoD,EAAC,OAG3D,AAAI,EAAW,GACb,GAAW,GAEb,AAAI,EAAW,GACb,GAAW,GAGb,GAAM,GAAI,EAAQ,GAAM,EAAG,EAAG,EAAG,SAAU,CAAC,GAAI,IAC1C,EAAI,GAAM,EAAG,EAAG,EAAG,SACnB,EAAK,GAAI,EAAG,GAEZ,EAAS,GACX,GAAU,EAAI,GAAO,CAAC,EAAU,UAChC,GAAa,EAAI,GAAO,CAAC,EAAU,WAEjC,EAAO,GAAM,CAAC,EAAG,GAAI,EAAG,OAE9B,MAAO,GACI,GAAM,GAAQ,EAAQ,EAAI,CAAC,GAAI,EAAG,KACvB,IAAI,SAAA,GAAO,MAAA,IAAM,EAAQ,EAAK,MACzC,GAGN,GAAM,IAAW,EAAG,CAAC,UAAS,KC7HrC,AA2DA,YAAsB,GACpB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,EAAkB,GAClB,EACI,GAAM,MAAQ,EAAG,OAAS,EAC1B,WAAM,MAAA,sEAGV,OADM,GAAM,EAAG,GAAG,MAAM,cACf,GACP,EACI,EAAG,GAAG,MAAM,KAAO,EACnB,WACI,MAAA,gEACA,KAAK,EAAkB,GAAG,MAAM,GAAE,QAAQ,EAAG,QAL9C,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,OAQT,GAAkB,GAClB,EAAK,GAAM,EAAI,EAAG,MAAM,GAAI,GAAG,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,CAAC,MAGtD,EACI,EAAG,QAAU,EAAG,GAAG,MAAM,GACzB,WAAM,MAAA,oCACK,EAAkB,OAAM,aAC/B,0BAA0B,EAAkB,GAAG,MAAM,GAAE,QAI/D,OAFM,GAAiB,GACjB,EAAO,aACJ,GACP,EAAG,KAAK,EAAO,KAAK,WAClB,GAAI,GAAI,EAAK,GACb,GAAI,EAAI,EACN,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAO,EAAI,GAAI,EAAI,EAAG,GAAI,IAAK,EAAG,IACxC,EAAI,GAAI,EAAG,GAGf,MAAO,IAAI,EAAG,GAAK,EAAG,kBATjB,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,GAaT,MAAI,GACK,GAAM,EAAI,GAEV,EAIJ,GAAM,IAAc,EAAG,CAAC,aAAY,KC5G3C,AAmFA,YAAa,EAAW,GAMtB,GANsB,AAAA,IAAA,QAAA,GAAA,IACtB,EACI,EAAE,MAAQ,EACV,WAAM,MAAA,gEACF,EAAE,OAEN,EAAE,OAAS,EACb,MAAO,IAAK,EAAe,GAM3B,GAAM,GAAgB,EAAE,MAAM,MAAM,EAAG,EAAE,MAAM,OAAS,GAC7B,OAAO,SAAC,EAAO,GAAS,MAAA,GAAQ,IACrD,EAAO,GACT,EACI,EACA,CACE,EAAe,EAAE,MAAM,EAAE,MAAM,OAAS,GACxC,EAAE,MAAM,EAAE,MAAM,OAAS,KAE/B,GACE,EAAmB,GACnB,EAAmB,GACzB,EAAK,QAAQ,SAAA,GACL,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,KAAK,GACV,EAAK,KAAK,KAEZ,GAAM,GAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OAC9B,EAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OACpC,MAAO,CAAC,EAAG,GAIf,YAAc,EAAa,GAAA,MAAA,KAAA,QAAA,GAAA,IAClB,EAAO,KAAK,WACjB,EACI,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,0CACF,EAAE,MAAM,OAAM,cAYtB,OAVM,GAAI,EAAE,MAAM,GACZ,EAAI,EAAE,MAAM,GAEd,EAAI,GAAI,GACR,EAAI,GAAM,GAER,EAAQ,GAAS,CAAC,CAAC,IAAK,CAAC,EAAG,IAC9B,EAAc,GAAM,GAElB,EAAQ,GAAK,EAAI,EAAI,aAClB,SAGD,EAAQ,EACR,EAAQ,EACR,EAAQ,EACd,EAAA,EAAA,KAAA,wfAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GA8CP,GAAQ,CAAC,EAAO,EAAO,KApDhB,EAAI,EAAG,EAAI,EAAO,EAAE,IAApB,GAuDT,MAAI,CAAC,GAAgB,EAAI,GACvB,GAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,IACzB,EAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,KAGpB,CAAC,EAAG,KAIR,GAAM,IAAK,EAAG,CAAC,IAAG,KCxMzB,AAiBA,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,IAAA,GAAA,MACA,EAAA,EAAA,uBAAA,GAAA,2BAJU,EAAA,WAAA,GAAA,UAAS,KCyBrB,YACI,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GAAM,GAAgB,GAAY,KAAQ,EAAU,EAAI,EAAS,GAEjE,GAAI,IAAc,EAAA,UAAU,KAC1B,MAAO,GAET,GAAI,IAAc,EAAA,UAAU,IAC1B,MAAO,IAAI,GAEb,GAAI,IAAc,EAAA,UAAU,MAC1B,GAAI,GAAY,KACd,MAAO,IAAK,GAEZ,GAAM,GAAkB,EAAQ,KAAO,EAAS,KAC1C,EAAS,GAAI,GAAI,GAAe,GAAI,IAC1C,MAAO,GAAkB,EAAI,GAAI,EAAQ,GAAO,IACnB,EAGjC,GAAI,IAAc,EAAA,UAAU,wBAC1B,GAAI,GAAY,KACd,MAAO,IAAI,GAAI,GAAe,GAAO,EAAQ,OAE7C,GAAM,GAAqB,EAAI,EAAU,GAAK,EAAQ,QAEhD,EACF,GAAK,GAAI,GAAS,EAAoB,GAAO,KAAM,WACvD,MAAO,IAAI,GAAI,GAAe,GAIlC,KAAM,OAAM,sBAAsB,GAE7B,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCnF3D,AA2CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,sBAC5C,EACF,EAAgB,EAAa,cAAe,sBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,uBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,iCAEvC,GAAM,GAAS,GAAI,GAAI,EAAS,IAChC,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAqB,EAAG,CAAC,oBAAmB,KChBzD,YACI,EAAsB,EAA2B,EACjD,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAC5C,EACF,EAAgB,EAAa,cAAe,kBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,mBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,6BAEvC,GAAM,GAAM,GAAO,GACb,EAAS,GAAI,EAAK,GAAI,EAAI,EAAS,GAAe,EAAM,KAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCnBjD,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAU,EAAgB,EAAQ,SAAU,aAC1C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAM,GAAO,GAEnB,EAAU,GAAI,EAAI,GAAO,GAAI,GAAU,GACvC,GAAM,GAAS,GAAK,GAAI,EAAK,EAAI,EAAS,KAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KC9DvC,AAiDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,aAC5C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAc,GAAO,GACrB,EAAQ,GAAI,GAAI,EAAc,IAC9B,EAAY,GAAQ,EAAO,GAC3B,EAAS,GAAI,EAAO,GAEpB,EACF,GAAI,EAAI,GAAO,IAAM,GAAO,IAAa,EAAI,EAAa,IAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KCtEvC,AAgDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,MAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,WAC5C,EAAe,EAAgB,EAAa,cAAe,WAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,YAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,sBAErD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAO,GAEvB,EAAK,GAAI,EAAI,EAAS,GAAI,GAAI,EAAc,MAC5C,EACF,EAAI,GAAI,EAAK,GAAU,GAAI,GAAI,GAAI,EAAK,GAAe,KACrD,EAAS,GAAI,EAAI,GACvB,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAU,EAAG,CAAC,SAAQ,KCrEnC,AA0CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,oBAC5C,EACF,EAAgB,EAAa,cAAe,oBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,qBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,+BAEvC,GAAM,GAAS,GAAkB,EAAS,GAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC3DrD,AAmCA,YACI,EAAsB,GACxB,GAAM,GACF,EAAgB,EAAQ,SAAU,iCAChC,EACF,EAAgB,EAAQ,SAAU,iCACtC,GACI,EAAQ,MAAO,EAAQ,MAAO,4CAsBlC,GAAM,GAAY,GAAK,GACjB,EAAgB,EAAI,EAAS,GAC7B,EAAgB,GAAM,GAAI,GAAI,GAAI,MAExC,MAAO,IAAI,GAAI,EAAW,GAAgB,GAwB5C,YACI,EAAgC,EAChC,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAoB,EACpB,EAAkB,mBAAoB,uBACpC,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KAOvB,GANA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAEjD,GACI,EAAkB,MAAO,EAAQ,MAAO,kCAExC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAO,GAAO,IAEpB,EACI,GAAI,EAAI,EAAmB,GAAI,EAAK,IAChC,EAAI,EAAM,IAEpB,GAAM,GAAS,GAA+B,EAAmB,GAEjE,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxH3D,AA+DA,YACI,EAAW,EAAW,GAKxB,GALwB,AAAA,IAAA,QAAA,GAAA,IACxB,AAAI,IAAQ,IACV,GAAM,EAAO,KAAO,GAGlB,IAAQ,EAAO,KAAO,EACxB,KAAM,OACF,+DACA,wCAAuC,EAAO,KAAI,KAClD,gBAAe,IAGrB,GAAM,GACF,GAAW,SAAC,EAAgB,EAAgB,GAI1C,GAAM,GAAW,GACX,EAAM,GAAU,EAAQ,CAAC,GAAM,GAC/B,EAAY,GAAI,GAAK,EAAQ,WAAY,GAC/C,EAAK,CAAC,EAAQ,IAEd,GAAM,GAAa,GAAI,EAAI,EAAW,IAChC,EAAW,GAAI,EAAY,CAAC,IAE5B,EAAW,SAAC,EAAO,GAChB,GAAA,GAAA,EAAA,GAAQ,EAAA,EAAA,GACT,EAAU,GAAqB,EAAG,MAAO,CAAC,IAChD,MAAO,CACL,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAK,EAAQ,WAAY,GAAI,KACrC,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAI,GAAY,GAAK,EAAQ,eAGzC,MAAO,CAAC,MAAK,EAAE,SAAQ,KAG7B,MAAO,GAAS,EAAQ,GAsB1B,YACI,EAA4B,EAC5B,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GACA,EAAgB,EAAc,eAAgB,uBAC5C,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KASvB,GAPA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GACI,EAAc,MAAO,EAAQ,MAAO,kCAEpC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAa,GAAO,EAAc,MAAM,IAE9C,EACI,GAAI,EAAI,EAAe,GAAI,EAAK,IAC5B,GAAI,EAAsB,IAGpC,GAAM,GAAS,GAA+B,EAAe,GAE7D,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KC3J3D,GAuNM,IAAW,CACf,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,MAAK,IASD,GAAS,CACb,cAAa,GACb,WAAU,GACV,MAAK,GACL,KAAI,IAeA,GAAQ,CACZ,cAAa,GACb,sBAAqB,GACrB,eAAc,GACd,iBAAgB,GAChB,cAAa,GACb,kBAAiB,GACjB,uBAAsB,GACtB,2BAA0B,GAC1B,gCAA+B,GAC/B,wBAAuB,GACvB,6BAA4B,IAOxB,GAAS,CACb,SAAQ,GACR,YAAW,GACX,GAAE,IAaE,GAAS,CACb,mBAAkB,GAClB,oBAAmB,GACnB,eAAc,GACd,UAAS,GACT,UAAS,GACT,QAAO,GACP,iBAAgB,GAChB,oBAAmB,GACnB,oBAAmB,IClSrB,mBAyCwC,GAAA,EAAA,GAAxC,4DAiBE,SAAA,UAAA,SAAA,SAAS,EAAiB,EAAoB,GAApB,AAAA,IAAA,QAAA,GAAA,IAElB,GAAA,GAAA,KAAA,iBAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,GAAI,GAAW,MACb,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,KAAM,OAAQ,EAAM,EAAE,SACrD,KAAK,eAAe,OAEpB,MAAK,eAAe,GAMtB,MAFA,IAAQ,GAEJ,EACK,EAEP,GAAM,UACC,OAOX,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAEd,KAAK,6CAGJ,EAAA,UAAA,oBAAV,WACE,KAAK,YAAc,KAAK,WAAa,GAgBvC,EAAA,UAAA,iBAAA,SAAiB,EAAiB,GAEhC,MAAO,IAAc,EAAG,IAgB1B,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,aAAe,MACtB,GAAQ,KAAK,cAIX,EAAA,UAAA,eAAN,8EACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAErB,CAAA,EAAO,CACL,KAAM,OAEN,OAAQ,GAAO,KAAK,YAAa,gBAI/B,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,gEAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OACN,4DACA,IAAG,KAAK,sBAUE,EAAA,UAAA,kBAAhB,SAAkC,mGAEhC,SAAA,KAAoB,CAAA,EAAM,EAAa,GAAG,OAAO,eAAjD,SAAK,YAAe,EAAA,OAAqC,GACzD,CAAA,EAAO,EAAa,MAAM,UAE9B,GA5HwC,IA8HxC,OAAO,eAAe,GAAW,OAAO,YAAa,CACnD,MAAO,SAAC,GACN,MAAO,GAAS,UAAY,MAAQ,EAAS,kBAAoB,MAC7D,EAAS,gBAAkB,QC1KnC,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAAgC,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,MAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,IAAA,EAChC,EAAA,QAAA,EALN,EAAA,iBAAwC,GACxC,EAAA,mBAA0C,GAOhD,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,cACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,aACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAC3C,EAAoB,EAAK,mBAAmB,GAAG,SAErD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAiB,EAAK,KAC1B,EAAI,GAAO,GAAW,EAAI,EAAK,MAEjC,EACF,EAAI,GAAI,GAAK,GAAI,EAAmB,EAAK,UACjC,GAAK,GAAI,EAAiB,EAAK,WACnC,GAEF,EACF,GAAI,EAAI,EAAmB,EAAK,KAC5B,EAAI,GAAO,GAAU,EAAI,EAAK,MAEtC,EAAgB,OAAO,GACvB,EAAkB,OAAO,GAEzB,GAAM,GAAW,GAAI,EAAI,EAAS,CAAC,EAAK,cAAe,GACvD,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,oBAAsB,MAC7B,IAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,YACzC,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,cAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,iBAAgB,OAAK,KAAK,oBAC/B,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,iBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,IAAO,KAAK,IACZ,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,IAAQ,EAAO,UAjHxD,EAAA,UAAY,WAmHrB,GArHuC,IAsHvC,GAAc,ICrJd,mBA+BsC,GAAA,EAAA,GAMpC,WACc,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,IAD5C,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,aAAA,EAA8B,EAAA,wBAAA,EAHpC,EAAA,iBAAwC,KAOhD,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAM,GAAY,GAClB,EAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,eACrB,SAAU,GACN,WAAM,MAAA,IAAK,EAAM,MAAO,EAAK,yBAClB,SAAS,MAI5B,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAEjD,GAAK,WACH,GAAM,GAAqB,GAAI,EAAiB,GAAO,IACvD,EAAgB,OAAO,GAEvB,GAAM,GAAW,GACb,EAAI,GAAI,EACA,GAAK,GAAI,EAAoB,EAAO,QAAQ,aAChD,CAAC,EAAK,cACV,GACJ,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,kBAAoB,MAC3B,GAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIvC,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,iBAAiB,IAC9D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,iBAAmB,EAAa,IACjC,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,wBAA2B,KAAK,0BAK7B,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,0BA/EzC,EAAA,UAAY,UAiFrB,GAnFsC,IAoFtC,GAAc,ICnHd,mBAkCmC,GAAA,EAAA,GASjC,WACc,EAAgC,EAChC,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MAFvC,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EAL/B,EAAA,uBAA8C,GAC9C,EAAA,wBAA+C,GAMrD,GAAK,WAEH,EAAK,SAAW,GAAO,GAAO,WAC9B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EAAmB,GAAI,EAAG,EAAK,UAErC,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,wBAAwB,IAAM,MACrC,GAAK,wBAAwB,GAAK,CAChC,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAe,EAAK,wBAAwB,GAAG,SAE/C,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QACvD,EACF,GAAI,EAAI,EAAc,EAAK,OACvB,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EAA2B,GAAI,EAAgB,GAC/C,EACF,GAAI,EAAiB,GAEzB,EAAY,OAAO,GACnB,EAAa,OAAO,GAEpB,GAAM,GACF,GAAI,EAAI,GAAI,EACA,GAAI,GAAK,GAA4B,EAAK,UAC9C,CAAC,EAAK,cACV,GACR,EAAM,OAAO,KAGf,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,QAC7C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,SAAS,UAEd,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,yBAA2B,MAClC,GAAQ,KAAK,wBAAwB,IAAI,SAAA,GAAK,MAAA,GAAE,aAI9C,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,yBACrC,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,4GACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACf,GAAK,WACH,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,IACxD,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,MAGpD,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,wBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,UA5IN,EAAA,UAAY,OA8IrB,GAhJmC,IAiJnC,GAAc,ICnLb,mBAiCoC,GAAA,EAAA,GASnC,WACc,EAAgC,EAChC,EAAyB,EACzB,GADyB,AAAA,IAAA,QAAA,GAAA,MACzB,AAAA,IAAA,QAAA,GAAA,GAHd,GAAA,GAIE,EAAA,KAAA,OAAO,KAHK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EACzB,EAAA,MAAA,EANN,EAAA,uBAA8C,GAC9C,EAAA,2BAAkD,GAQxD,GAAK,WACH,EAAK,UAAY,GAAO,GAAG,WAC3B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EACF,GAAI,CAAC,EAAK,aAAc,GAAI,EAAI,EAAK,UAAW,EAAK,OAAQ,IAEjE,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAGxC,AAAI,EAAK,2BAA2B,IAAM,MACxC,GAAK,2BAA2B,GAAK,CACnC,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAIxC,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAkB,EAAK,2BAA2B,GAAG,SAErD,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QAEvD,EAAM,EAAI,EAAiB,EAAK,OAChC,EAAM,GAAI,GAEV,EAAqB,GAAQ,EAAK,GAExC,EAAY,OAAO,GACnB,EAAgB,OAAO,GAEvB,GAAM,GACF,GAAI,EAAI,GAAI,EAAI,GACR,GAAI,EAAgB,GAAI,EAAoB,EAAK,WACrD,GAER,EAAM,OAAO,KAGf,EAAK,UAAU,OAAO,GAAI,EAAK,UAAW,IAC1C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,UAAU,UAEf,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,4BAA8B,MACrC,GAAQ,KAAK,2BAA2B,IAAI,SAAA,GAAK,MAAA,GAAE,aAIjD,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,wDAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OAAM,wDAGlB,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,QAChB,MAAS,KAAK,QAKX,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,QAAY,EAAO,QAvHzB,EAAA,UAAY,SAyHrB,GA3HqC,IA4HrC,GAAc,IC7Jd,mBA6BkC,GAAA,EAAA,GAKhC,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,aAAA,EAEpB,EAAK,gBAAgB,KAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAEF,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAK,WACH,GAAM,GAAW,GAAI,EAAI,EAAK,EAAG,GAAW,GAC5C,EAAM,OAAO,OAGjB,KAAK,uBAMP,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,aAAe,EACpB,AAAI,KAAK,GAAK,MACZ,KAAK,EAAE,UAET,KAAK,EAAI,GAAK,GAAO,CAAC,KAGxB,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,WAGH,EAAA,UAAA,WAAN,qGACU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAA,CAAQ,EAAA,eAGJ,EAAA,UAAA,WAAN,SAAiB,6FACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAC5C,GADA,EAAe,EAAA,OACX,EAAa,SAAW,EAC1B,KAAM,IAAI,OAAM,iEAIpB,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,aAAgB,KAAK,eAIxB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,eA7DjB,EAAA,UAAY,MA+DrB,GAjEkC,IAkElC,GAAc,IC/Fd,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAA8B,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFZ,GAAA,GAGE,EAAA,KAAA,KAAM,IAAa,KAFP,SAAA,aAAA,EAA8B,EAAA,SAAA,EAChC,EAAA,YAAA,EAJJ,EAAA,cAAqC,GAM3C,EAAK,EAAI,GAAO,EAAK,YAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,cAAc,IAAM,MAC3B,GAAM,GAAY,GAClB,EAAK,cAAc,GAAK,CACtB,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,MAInD,GAAM,GAAe,EAAK,cAAc,GAAG,SACrC,EAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAK,WACH,GAAI,GACE,EAAkB,GAAI,EAAI,EAAK,EAAG,GAAe,GACvD,AAAI,EAAK,YACP,EAAW,GACP,EAAI,EAAK,EAAG,GAAI,EAAU,EAAI,EAAiB,EAAK,KAAM,GAE9D,EAAW,GAAI,EAAI,EAAK,EAAG,GAAkB,GAE/C,EAAa,OAAO,GACpB,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,UACP,AAAI,KAAK,eAAiB,MACxB,GAAQ,KAAK,cAAc,IAAI,SAAA,GAAK,MAAA,GAAE,aAS1C,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,SAAW,GAGZ,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,cAAc,IAC3D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,cAAgB,EAAa,IAC9B,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,SAAY,KAAK,SACjB,YAAe,KAAK,cAKjB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,SAAa,EAAO,cA3FlD,EAAA,UAAY,WA6FrB,GA/FuC,IAgGvC,GAAc,IC/Hd,mBAgCsC,GAAA,EAAA,GASpC,WACc,EAAgC,EAChC,EAA0B,EACpC,GAF0C,AAAA,IAAA,QAAA,GAAA,IAChC,AAAA,IAAA,QAAA,GAAA,GAA0B,AAAA,IAAA,QAAA,GAAA,MACpC,AAAA,IAAA,QAAA,GAAA,IAHJ,GAAA,GAIE,EAAA,KAAA,OAAO,KAOP,GAVY,EAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,SAAA,EAA0B,EAAA,QAAA,EANhC,EAAA,uBAA8C,GAC9C,EAAA,mBAA0C,GAC1C,EAAA,qBAA4C,GAQlD,EAAK,SAAW,EAEhB,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,WAE5B,GAAgB,KAClB,KAAM,IAAI,OAAM,+DAIpB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,OACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,qBAAqB,IAAM,MAAQ,EAAK,UAC/C,GAAK,qBAAqB,GAAK,CAC7B,aAAiB,EAAI,MACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAwB,EAAK,uBAAuB,GAAG,SACvD,EAAqB,EAAK,mBAAmB,GAAG,SACtD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEvC,GAAI,EAAK,UACP,GAAM,GAAsB,EAAK,qBAAqB,GAAG,SAEnD,EACF,GAAI,EAAI,EAAqB,EAAK,OAC9B,EAAI,EAAU,EAAI,EAAK,QAEzB,EACF,GAAI,EAAI,EAAU,EAAK,cACnB,GACI,GAAI,EACA,GAAI,GAAO,GAAyB,EAAK,YACnD,EACF,GAAI,EAAI,EAAoB,EAAK,UAAW,GAEhD,EAAsB,OAAO,GAC7B,EAAoB,OAAO,GAC3B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAGb,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EACF,GAAI,EAAI,EAAoB,EAAK,UAC7B,GAAI,EAAI,EAAU,EAAK,cACnB,GAAK,GAAI,EAA0B,EAAK,YAEpD,EAAsB,OAAO,GAC7B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAInB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,sBAAwB,MAAQ,KAAK,UAC5C,GAAQ,KAAK,qBAAqB,IAAI,SAAA,GAAK,MAAA,GAAE,YAE/C,AAAI,KAAK,oBAAsB,MAC7B,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,oBAC7C,AAAI,KAAK,UACP,EAAU,KAAI,MAAd,EAAkB,KAAK,sBAEjB,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EACF,KAAK,SAAW,EAAa,OAAS,EAAI,EAAa,OAAS,EAC9D,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,MAE3C,AAAI,KAAK,UACP,MAAK,qBACD,EAAa,MAAM,EAAgB,EAAG,EAAgB,GACjD,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,iBAK/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,SAAY,KAAK,SACjB,QAAW,KAAK,QAChB,SAAY,KAAK,WAKd,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,SAChD,EAAO,QAAY,EAAO,WA5KzB,EAAA,UAAY,UA8KrB,GAhLsC,IAiLtC,GAAc,ICjNd,AAyBA,GAAA,IAAA,WAAA,cAsCS,SAAA,IAAP,SAAW,GACT,MAAO,IAAI,IAAa,IAkBnB,EAAA,SAAP,SAAgB,EAAsB,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE/C,GAAI,IAAkB,EAAc,EAAU,IAuBhD,EAAA,QAAP,SACI,EAAsB,EAAY,EAAgB,EAClD,GADsB,MAAA,KAAA,QAAA,GAAA,IAAY,AAAA,IAAA,QAAA,GAAA,GAAgB,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,IACK,GAAI,IACP,EAAc,EAAO,EAAU,EAAS,IAevC,EAAA,KAAP,SACI,EAAsB,EAAa,EACnC,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MACnC,AAAA,IAAA,QAAA,GAAA,MACK,GAAI,IAAc,EAAc,EAAO,EAAO,IAehD,EAAA,SAAP,SAAgB,EAAqB,EAAW,GAAhC,MAAA,KAAA,QAAA,GAAA,MAAqB,AAAA,IAAA,QAAA,GAAA,KAAW,AAAA,IAAA,QAAA,GAAA,MAEvC,GAAI,IAAkB,EAAc,EAAK,IAgB3C,EAAA,OAAP,SACI,EAAsB,EAAa,EAAe,EAClD,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MAAe,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,GACK,GAAI,IAAgB,EAAc,EAAO,EAAO,EAAS,IAmB3D,EAAA,QAAP,SAAe,EAAsB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE5B,GAAI,IAAiB,EAAc,IAE9C,KC5LA,GA+Ba,IAAQ,CACnB,IAAK,GAAsB,IAC3B,SAAU,GAAsB,SAChC,SAAU,GAAsB,SAChC,QAAS,GAAsB,QAC/B,QAAS,GAAsB,QAC/B,OAAQ,GAAsB,OAC9B,KAAM,GAAsB,MCtC9B,AAiBA,GAAM,IAA2B,WAC/B,MAAI,OAAO,wBAA0B,YAC5B,sBACE,MAAO,eAAiB,YAC1B,aAEF,SAAC,GAAgB,MAAA,SAa1B,cACE,MAAO,IAAI,SAAc,SAAA,GAAW,MAAA,IAAc,WAAM,MAAA,SCrC1D,YAmBI,EAAiC,EACjC,GACF,GAAM,GACF,EAAc,OAAO,IAAW,SAAW,EAAS,EAAO,IACzD,EACF,EAAe,OAAO,IAAW,SAAW,EAAS,EAAO,IAChE,MAAO,CAAC,EAAS,GCzBnB,YA0BI,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAqB,GACzB,GAAI,EACF,EAAW,EAAS,OAAO,EAAW,MAAM,IAC5C,EAAS,KAAK,EAAW,GAAK,GAC9B,EAAW,EAAS,OAAO,EAAW,MAAM,SAE5C,EAAW,EAAS,OAAO,EAAW,IAEtC,OADM,GAAgB,EAAW,OACxB,EAAI,EAAG,EAAI,EAAe,EAAE,EACnC,EACI,EAAS,OAAO,CAAC,EAAW,EAAI,GAAK,EAAW,GAAI,EAAW,KAErE,EAAW,EAAS,OAAO,EAAW,MAAM,EAAgB,IAE9D,MAAO,eAaL,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAW,GACjB,GAAI,GACF,EAAS,KAAK,GACd,OAAS,GAAI,EAAiB,EAAG,EAAI,EAAc,EAAE,EACnD,AAAI,GAAK,EAAI,EACX,GAAS,KAAK,GACd,EAAS,KAAK,EAAK,GAAiB,KAEpC,EAAS,KAAK,QAMlB,OAFM,GAAsB,GACtB,EAAqB,GAClB,EAAI,EAAG,EAAI,EAAc,EAAE,EAClC,AAAI,GAAK,EAAiB,EAAI,GAAK,EAAI,IAAM,EAC3C,EAAmB,KAAK,GAExB,EAAoB,KAAK,GAG7B,EAAS,KAAI,MAAb,EAAiB,GACjB,EAAS,KAAK,GACd,EAAS,KAAI,MAAb,EAAiB,GAEnB,MAAO,eAaL,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAmB,GAEzB,AAAI,EACF,EAAiB,KAAK,EAAW,GAAK,GAEtC,EAAiB,KAAK,EAAW,GAAK,GAGxC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,GAAK,EAAW,OAClB,AAAI,EACF,EAAiB,KAAK,EAAW,EAAI,GAAK,EAAW,IAErD,EAAiB,KAAK,EAAW,GAAK,EAAW,EAAI,IAGvD,EAAiB,KAAK,EAAW,IAIrC,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,CAAC,GACjB,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAiB,KAAK,EAAM,GAAG,IAEjC,MAAO,eAeL,EAA0B,EAAmB,GAE/C,OADM,GAAY,EAAe,MAAM,EAAG,GACjC,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAU,KAAK,EAAe,EAAI,GAAK,EAAM,GAAG,GAAK,EAAM,GAAG,IAGhE,MAAO,GCvJT,AAiBO,GAAM,IAAkB,mBAClB,GAAa,mBClB1B,AAiBO,GAAM,IAAQ,SACR,GAAS,WACT,GAAS,YACT,GAAS,YACT,GAAS,aACT,GAAS,YCtBtB,qBAmBqB,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACnB,AAAK,KAAM,QAAQ,YACjB,QAAQ,KAAI,MAAZ,QAAgB,wBAIA,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GAClB,AAAK,KAAM,QAAQ,YACjB,QAAQ,IAAG,MAAX,QAAe,GC3BnB,YAiCI,EAAoB,GACtB,GAAI,EAAK,SAAW,EAAK,OACvB,KAAM,IAAI,OACN,gEACG,GAAK,OAAM,WAAW,EAAK,OAAM,MAG1C,OADM,GAAS,GAAI,cAAa,EAAK,OAAS,GACrC,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EACtC,EAAO,GAAK,EAAK,EAAI,GACrB,EAAO,EAAI,GAAK,EAAK,EAAI,GAE3B,MAAO,eAiB8B,GAIrC,OAFM,GAAO,GAAI,cAAa,EAAQ,OAAS,GACzC,EAAO,GAAI,cAAa,EAAQ,OAAS,GACtC,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,EAAI,GAAK,EAAQ,GACtB,EAAK,EAAI,GAAK,EAAQ,EAAI,GAE5B,MAAO,CAAC,KAAI,EAAE,KAAI,eAOiB,GAKnC,OAHM,GAAM,KAAK,KAAK,EAAQ,OAAS,GACjC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAOgB,GAKlC,OAHM,GAAM,KAAK,MAAM,EAAQ,OAAS,GAClC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAShB,EAAuB,GACzB,GAAM,GAAO,EAAQ,EAAQ,GACvB,EAAO,EAAQ,EAAQ,EAAI,GACjC,MAAO,CAAC,KAAI,EAAE,KAAI,eAUhB,EAAkB,EAAc,EAAc,GAChD,EAAK,EAAQ,GAAK,EAClB,EAAK,EAAQ,EAAI,GAAK,cAOpB,EAAW,GAGb,OAFM,GAAO,GAAI,cAAa,EAAI,GAC5B,EAAO,GAAI,cAAa,EAAI,GACzB,EAAI,EAAG,EAAI,KAAK,KAAK,EAAI,GAAI,KACpC,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GAC9C,EAAK,GAAK,KAAK,IAAI,GACnB,EAAK,GAAK,KAAK,IAAI,GAErB,MAAO,CAAC,KAAI,EAAE,KAAI,eAOhB,EAAW,EAAW,GACxB,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GACxC,EAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GACtB,MAAO,CAAC,KAAI,EAAE,KAAI,GCvJpB,YAyDI,EAAM,EAAiB,GACzB,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,GAAE,QAEX,GAAM,GAAc,GAAM,EAAE,OACtB,EAAS,GAAK,EAAG,WACjB,EAAS,EAAQ,QAAQ,EAAQ,GACvC,SAAY,UACZ,EAAO,UACA,EAGT,GAAI,CAAC,GAAgB,EAAE,MAAO,GAG5B,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAE,MAAO,GAExD,GAAI,EAAE,QAAU,aACd,GAAM,GAAO,EAAQ,KAAK,GACpB,EAAS,GAAK,EAAM,GAC1B,SAAK,UACE,EAET,GAAI,IAAU,QACZ,MAAO,GAAQ,IAAI,GACd,GAAI,IAAU,QACnB,GAAM,GAAO,GAAO,EAAG,EAAE,OACnB,EAAS,EAAQ,SAAS,EAAG,GACnC,SAAK,UACE,MAEP,MAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,eAK/D,EAAM,GACR,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAO,EAAE,mBAG3B,EAAe,EAAc,GACxD,GAAM,GAAQ,GAAO,GAAU,GAAM,GAE/B,EAAS,GAAoB,EAAK,WACxC,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,qvCC3G1B,YAyBI,EAAM,EAAsB,GAC9B,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,MAAO,GAAW,IAAI,SAAA,GACpB,GAAM,GAAgB,EAAI,QAC1B,EAAU,GAAQ,EAClB,GAAM,GAAS,GAAM,EAAG,EAAO,GAC/B,SAAM,IAAS,EACR,ICjCX,YA2BI,EAAiC,GAEnC,OADM,GAAqB,GAAI,OAAM,EAAK,MACjC,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAK,MAAM,GAAK,EAAK,GAGrC,OADM,GAAS,GAAO,EAAU,EAAK,OAC5B,EAAI,EAAG,EAAI,EAAO,OAAO,OAAQ,EAAE,GAI1C,OAHM,GAAS,EAAO,WAAW,GAE3B,EAAwB,GAAI,OAAM,EAAK,MACpC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,MAAM,GAG1C,GAAM,GAAgB,EAAK,WAAW,GAEtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,WC7ChB,YAyBI,EAAe,EAAkB,EAAyB,EAC1D,GAOF,OALM,GAAU,EAAO,EAAO,OAAS,GACjC,EAAA,CAAA,EAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAc,GAAuB,EAAQ,EAAQ,GACrD,EAAiB,GAAuB,QAAS,EAAQ,GAEtD,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHM,GAAS,EAAI,EACb,EAAO,EAAE,SAAS,EAAQ,EAAS,GACnC,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAEzC,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAKrC,OAHM,GAAY,EAAI,EAChB,EAAW,EAAY,SAAS,EAAW,EAAY,GACvD,EAAc,EAAe,SAAS,EAAW,EAAY,GAC1D,EAAI,EAAG,EAAI,EAAG,IACrB,EAAS,GAAK,EAAU,GAAG,MAC3B,EAAY,GAAK,EAAU,GAAG,MAKlC,GAAM,GAAc,EAAO,QAC3B,SAAY,EAAY,OAAS,GAAK,EAC/B,CACL,GAAO,EAAa,EAAa,GACjC,GAAO,EAAgB,EAAa,UCxDxC,mJCAA,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,WAAY,SC7BtD,AA4BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAO,GAAK,EAAG,YACnB,EAAI,GAAK,GAAI,GAAO,GAAI,IAC9B,MAAO,IAAI,GAAI,EAAI,QCtC3B,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAK,EAAG,YAAa,IAC/C,MAAO,IAAI,EAAI,OCnCvB,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAGxB,MAAO,CAAC,EAAS,EAAG,KChDxB,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAsC,GAC5C,SAAM,QAAQ,SAAC,EAAG,GAChB,EAAK,GAAK,WAAM,MAAA,GAAG,WAEd,IC7BX,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,mBChChE,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,cAC7C,MAAO,IAAI,EAAI,OCpCvB,AA6BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,EAAI,EAAI,GAAI,EAAG,IACnB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,GAAI,EAAI,EAAI,GAAI,EAAG,KACvB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAExB,MAAO,CAAC,EAAS,EAAG,KCtDxB,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAK,EAAG,YAAa,QC/B7D,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,kBC/B3D,AA2DA,YACI,EAAkB,EAClB,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBAE3C,EAAO,EACP,EAAU,EACV,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,MAIvE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OAEvB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,IAG5C,EAAkC,CAAC,GAAI,EAAM,MAAO,GAEpD,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KChIvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAY,EAAS,EAAY,EACpD,OCpCV,AA8CA,YACI,EAAkB,EAAqB,EACvC,EAAkC,GACpC,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBAE/C,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,gCAC/B,EAAI,KAAI,MAEhB,GAAI,GAAU,EACV,EAAO,EACP,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EACI,EAAQ,EAAQ,CAAC,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KACvE,EAAO,EAAQ,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGhE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,4DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,+DACC,GAAQ,KAAI,OAEvB,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAmB,GAE3D,MAAO,GAAQ,gBAAgB,EAAM,EAAS,IAG1C,EAAgC,CAAC,GAAI,EAAM,MAAO,GAElD,EAA8B,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,GAEvD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAC5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAY,EAAS,OC7BvE,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,WAEnB,MAAI,CAAC,GAAc,CAAC,EACX,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,MAEtB,CAAC,GAAc,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,MAEtB,GAAc,CAAC,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,MAGzB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,KAC7B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,QChDrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,MACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAQtC,OAPM,GACF,EAEE,EAAa,EAAiB,WAC9B,EAAc,EAAiB,MAE/B,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAW,OAAS,EAAG,GAAK,EAAG,IAC1C,GAAI,EAAW,KAAO,EAAY,GAChC,EAAK,GAAK,UACD,EAAW,KAAO,EAC3B,KAAM,IAAI,OAAM,mBACZ,EAAU,6BAA6B,EAAW,MAI1D,OADM,GAAiB,GACd,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,EAAK,GAAK,GACZ,EAAK,KAAK,GAId,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAM,QC/CnC,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,GAAG,YCxBxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAET,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AA0BO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,aAAc,EAAA,EAAA,aACrB,MAAO,CACL,EAAG,WAAM,MAAA,IACL,GAAW,GAAa,EAAG,GAAe,GAAU,EAAG,IACvD,EAAI,GAAU,QCnCxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAS,EAAM,IAAI,SAAA,GAAK,MAAA,GAAE,QACzB,EAAA,EAAA,KACD,EAAQ,GAAe,EAAM,EAAM,GAAG,OAAO,GAC7C,EAAa,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAa,GAAM,EAAI,EAAY,GACzC,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,YAAM,MAAA,QC/BrC,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACN,EAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAEhC,SACI,GAA4B,GAC5B,WAAM,MAAA,8DACF,uDAAsD,EAAS,OAEhE,CACL,EAAG,WACC,MAAA,IAAoB,EAAI,MAAO,EAAI,EAAS,EAAS,EAAK,IAC9D,OAAQ,WACJ,MAAA,IAAqB,EAAK,EAAI,EAAQ,MAAO,EAAS,EAAK,OCxCrE,AAsBO,GAAM,IAA4C,CACvD,WAAY,GACZ,aAAc,CAAC,KAAM,UACrB,SAAU,SAAC,EAAe,EAAiB,GACnC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAAY,EAAA,EAAA,gBAGjC,MAAO,CACL,GAAI,WAAM,MAAA,IACN,EAAK,EAAQ,EAAS,EAAK,EAAY,EACvC,IACJ,OAAQ,WAAM,MAAA,IACV,EAAK,EAAI,EAAO,MAAO,EAAS,EAAK,EAAY,OCpC3D,AA2CA,YACI,EAAM,EAAO,EACb,EAA0C,GAC5C,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEnE,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAEzE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,EACI,EAAI,MAAM,KAAO,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAI,MAAM,GAAE,UAC1D,iCAAgC,EAAY,GAAE,OACtD,EACI,EAAK,MAAM,KAAO,EAAY,GAC9B,WAAM,MAAA,0CAA0C,EAAK,MAAM,GAAE,UACzD,mCAAkC,EAAY,GAAE,QAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAEhD,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,EAAG,GAEjD,EAAmC,CAAC,QAAO,EAAE,IAAG,GAEtD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAwB,GAG9B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC7F7D,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAC3B,EACI,GAAkB,GAClB,WACI,MAAA,kEACA,mDAAkD,EAAS,OAE5D,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAEZ,MAAO,CACL,EAAG,WAAM,MAAA,IACJ,EAAiB,MAAO,EAAI,EAAqB,EAAS,IAC/D,OAAQ,WAAM,MAAA,IACV,EAAiB,EAAK,EAAqB,MAAO,EAAS,OCzCrE,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAI,GAAK,EAAG,aAAc,OC/BvD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UAAW,EAAA,EAAA,QAExB,MAAO,CACL,EAAG,WACD,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAE,MAE7C,EAAM,GAAO,EAAI,EAAM,EAAW,CAAC,GAEvC,MAAI,IAAe,MACjB,GAAM,GAAU,EAAK,IAGhB,MCzCf,AAwBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAG1B,EAAa,GAAoB,CAAC,EAAG,GAE3C,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAU,OAEhB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,EACI,EAAE,OAAS,EACX,WAAM,MAAA,6DACF,yBAAwB,EAAE,KAAI,OACtC,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,8DACF,yBAAwB,EAAO,KAAI,OAC3C,EACI,EAAE,MAAM,KAAO,EAAO,MAAM,GAC5B,WAAM,MAAA,yDACF,cAAa,EAAE,MAAM,GAAE,0CACvB,cAAa,EAAO,MAAM,GAAE,OAEpC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,sCAAqC,EAAO,mBAC5C,KAAI,EAAU,QAEtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WACI,MAAA,gEACA,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAW,GACb,EAAE,MAAO,EAAO,MAAO,EAAS,EAChC,EAAK,EAAiB,IAE1B,MAAO,CACL,EAAG,WACC,MAAA,IAAmC,EAAE,MAAO,EAAI,EAAQ,IAC5D,OAAQ,WACJ,MAAA,IAAoC,EAAG,EAAI,EAAO,MAAO,OC7EnE,AAuBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAA6C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAC3D,EAA+C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAEnE,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,UACH,GAAyB,EACzB,IACb,OAAQ,WAAM,MAAA,GAAO,UACH,GACA,EAAsC,OCtC9D,AA6BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OACnD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAgC,EAAE,MAAO,GAC5D,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GAC/C,MAAO,GAAQ,OAAO,EAAI,IAGtB,EAAwB,CAAC,GAAE,EAAE,EAAC,GAEpC,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,cACZ,EAAoB,EAAgC,KACpD,QCrCV,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAI,EAAI,GAAI,GAAI,GAAO,KAAM,EAAI,KAAK,KAAK,KAAK,KACtD,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC/B7B,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC3B7B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAI,QC5BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AA6BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAiB,EAAE,MAAO,GAC7C,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AA8BO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,IAAK,OAAQ,WAAY,SACxC,SAAU,SACN,EAAY,EAAiB,GACxB,GAAA,GAAA,EAAA,gBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GAAM,EAAA,EAAA,GAAU,EAAA,EAAA,GAEpB,EAAa,GAAgB,GAAO,GACpC,EAAgB,GAAiB,EAAK,MAAO,EAAE,OAC/C,EAAsB,GAC5B,GAAI,EAAK,OAAS,GAChB,OAAS,GAAI,EAAG,EAAI,EAAE,MAAM,OAAS,EAAG,EAAE,EACxC,EAAU,KAAK,EAAE,MAAM,IAEzB,EAAU,KAAK,GAGjB,GAAM,GAAa,GAAI,EAAG,GACpB,EAAoB,EAAI,EAAI,GAC5B,EAAsB,GAAM,GAAI,EAAU,GAAO,KACjD,EAAiB,EACnB,EAAI,EAAI,EAAqB,GAAsB,GACnD,GAAO,MAEL,EAAO,WACX,MAAI,GAAK,OAAS,EACT,EACH,EAAI,EAAI,EACA,GACI,EAAQ,EAAqB,CAAC,EAAG,EAAG,EAAG,EAAK,MAAM,KAClD,IACR,GACJ,EAAE,OAEC,EAAQ,EAAI,EAAI,EAAI,GAAsB,GAAa,EAAE,QAG9D,EAAU,WACd,GAAI,GACA,EAAI,EAAI,EAAqB,GAAO,KAAM,GAC9C,MAAI,GAAK,OAAS,GAChB,GAAU,GAAI,EAAS,IAElB,EAAQ,EAAS,EAAK,QAEzB,EAAc,WAClB,GAAI,GAAc,EAAI,EAAI,EAAgB,GAAa,GAEvD,MAAI,GAAK,OAAS,GAChB,GAAc,GAAI,EAAa,IAE1B,EAAQ,EAAa,EAAK,QAE7B,EAAW,WACf,GAAM,GAAwB,EAAI,EAAY,GAE1C,EAAW,EAAI,EAAI,GACvB,MAAI,GAAK,OAAS,GAChB,GAAW,GAAI,EAAU,IAEpB,EAAQ,EAAU,EAAK,QAE1B,EAAY,WAChB,GAAI,GAAY,EAChB,MAAI,GAAK,OAAS,GAChB,GAAY,GAAI,EAAW,IAEtB,EAAQ,EAAW,EAAK,QAGjC,MAAO,CACL,EAAG,EACH,KAAM,EACN,SAAU,EACV,MAAO,EACP,OAAQ,KC1Gd,AA0BO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,WACpB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACH,EAAA,EAAA,KAED,EAAa,GAAe,EAAM,EAAE,OAAO,GAE3C,EAAO,WACX,GAAM,GAAc,EAAE,MAChB,EAAc,EAAQ,KAEtB,EAAa,EAAY,MAAM,EAAG,GAClC,EAAY,EAAW,OACvB,EAAa,EAAY,MAAM,EAAM,EAAY,QAAQ,MAAM,GAC/D,EAAY,EAAW,OAEvB,EAAmB,GAAW,EAAG,GACjC,EACF,GAAW,EAAY,EAAG,EAAY,EAAI,GAExC,EAAc,GAAY,CAAC,EAAY,CAAC,GAAc,IAEtD,EAAS,EAAQ,EAAI,GACrB,EAAkB,EAAQ,EAAS,CAAC,IAEpC,EACF,GAAY,CAAC,CAAC,GAAY,EAAkB,IAC1C,EAAkB,GAAU,EAAQ,GACtC,EAAa,GACb,EAAiB,EAA6B,EAAE,MAAM,IAEpD,EAAsB,GAAuB,GACnD,SAAa,GAAU,EAAY,GAE5B,GAET,MAAO,CAAC,EAAG,EAAM,QAAS,WAAM,MAAA,OAIpC,YAAoB,EAAe,GAEjC,OADM,GAAS,GACN,EAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,EAAO,KAAK,GAEd,MAAO,GAGT,YAAqB,GAEnB,OADM,GAAS,GACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,EACtC,EAAO,KAAK,EAAO,GAAG,IAG1B,MAAO,GCnFT,AAqBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,IAAI,EAAG,WAAM,MAAA,IAAU,OC1BtD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAK,EAAI,eCzB9B,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC5B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,EAAG,QC5BpC,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,EAAG,gBC5BrC,AAyBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,GACd,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KACP,MAAO,CACL,OAAQ,WACN,GAAM,GAAW,GACX,EAAU,GAAI,GACpB,MAAO,IAAI,EAAI,EAAI,GAAI,EAAI,EAAM,GAAW,QCpCpD,AAyBA,YACI,EAAM,EAAM,EAAO,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,QAAQ,EAAI,EAAG,EAAG,EAAa,EAAM,EAAO,IAElD,EAA4B,CAAC,EAAC,EAAE,EAAC,EAAE,GAAE,GAErC,EAA0B,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAE/D,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAa,GAGnB,GAAM,IACT,EAAG,CAAC,oCAAmC,KCxC3C,AAqBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,YAAa,EAAA,EAAA,KAAM,EAAA,EAAA,MAAO,EAAA,EAAA,KAEjC,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAG,EAAG,EAAI,EAAa,EAAM,EAAO,OC/B9C,YA6BI,EAAO,EAAM,EAAe,EAAoB,GAClD,MAAI,GAAE,KAAO,EAAM,MACjB,GAAI,EAAQ,EAAG,GAA+B,EAAE,MAAO,KAEzD,AAAI,EAAG,KAAO,EAAM,MAClB,GAAK,EAAQ,EAAI,GAA+B,EAAG,MAAO,KAErD,CACL,EAAG,WACD,GAAM,GAAK,EAAI,EAAI,GAAK,GAAM,EAAO,GAAI,EAAG,QAC5C,MAAO,IAAgB,KAAO,EAAK,GAAU,EAAI,KCvCvD,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,iBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAkB,EAAE,OACnD,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAa,EAAG,GAAI,aAC9C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,GAAI,aAC5C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA6DA,YACI,EAAkB,EAAqB,EACvC,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE9C,EAAO,EACP,EAAU,EACV,EAAW,EACX,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KAErE,EAAW,EAAQ,EAAS,CAC1B,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAI3E,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAS,OAAS,EAClB,WAAM,MAAA,kEACC,GAAS,KAAI,OAExB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OACrE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,EAAU,IAGtD,EACwB,CAAC,GAAI,EAAM,MAAO,EAAS,OAAQ,GAE3D,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC1IvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAG,EAAY,EAAS,EAAY,EACvD,OCrCV,AAkDA,YACI,EAAyB,EACzB,EAA6B,EAC7B,EAAkC,EAClC,GACF,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBACzC,EAAU,EAAgB,EAAQ,SAAU,mBAElD,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,+BAC/B,KAAI,EAAI,KAAI,OAEpB,EACI,EAAI,OAAS,EACb,WAAM,MAAA,4DACC,GAAI,KAAI,OACnB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,+DACC,GAAO,KAAI,OACtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAO,MAAO,EAAY,EAAS,EAAmB,EACtD,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAQ,EAAS,IAGjD,EACsB,CAAC,GAAI,EAAK,MAAO,EAAQ,OAAQ,GAEvD,EACqB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAErE,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,GAC/C,GAGN,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAE5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAG,EAAY,EAAS,OC/B1E,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,KACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAM,EAAE,OACvC,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAU,EAAG,GAAI,aAC3C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAQ,EAAG,GAAI,aAC/C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA4BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAI,GAAa,EAAE,OAEjC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAI,GAAM,GAAI,EAAG,MAC/B,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KClDxB,AA0BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,OCzBzB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,WACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAU,EAAM,GACtB,MAAO,CAAC,QAAS,WAAM,MAAA,IAAM,EAAQ,MAAO,eC3BhD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAGtC,GAAM,GAAI,EAAM,GACT,EAAA,EAAA,SACD,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAM,EAAI,EAAO,EAAE,WC/BxC,AAgCO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GACP,EAAO,EACP,EAAM,EACN,EACF,GAA0C,EAAK,MAAO,EAAI,OAExD,EAAU,WACd,GAAM,GAAW,GAAK,EAAK,WACvB,EAAM,EAAI,EAAI,EAAI,EAAU,GAAI,EAAM,GAAI,EAAU,GAAO,OACzD,EAAa,GAAgC,EAAK,MAAO,GAC/D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,QAErB,EAAS,WACb,GAAM,GAAY,GAAQ,EAAM,GAC1B,EAAU,GAAM,EAAW,GAAI,GAAO,GAAU,IAClD,EAAM,EAAI,EAAI,EAAI,EAAG,IACnB,EAAa,GAAgC,EAAI,MAAO,GAC9D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAI,QAE1B,MAAO,CAAC,EAAG,EAAS,EAAG,KC9D3B,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,SACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,GAAQ,EAAG,GAExB,MAAO,CACL,EAAG,WAAM,MAAA,IAAM,EAAM,EAAI,EAAI,EAAI,KACjC,MAAO,WACL,GAAI,GAAM,GAAM,EAAM,GAAU,GAAK,EAAI,EAAI,IACvC,EAAa,GAAiB,EAAM,MAAO,EAAG,OACpD,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAM,WC1ClC,AAwBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,SC7BxC,AAwBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAO,EAAI,GAAU,EAAG,GAAI,GAAK,IAEvC,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAM,gBC/BxC,AAuBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,GAAI,gBC5B3C,AAqBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAQ,EAAI,EAAE,WC1BnC,AAuBO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,uBACX,EAAI,EAAoB,IAGxB,EAAmC,CAAC,OAAM,GAC1C,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAAoB,IAExB,MAAO,CAAC,OAAQ,KCxCpB,AAuBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,8BACX,EAAI,EAAoB,IAGxB,EAA0C,CAAC,OAAM,GACjD,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAA2B,IAE/B,MAAO,CAAC,OAAQ,KCxCpB,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,KACD,EAAO,GAAe,EAAM,EAAG,OACrC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAQ,EAAI,OC3BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAyBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,GAAI,EAAI,EAAI,GAAI,EAAG,KAAM,SC9BlD,AAyBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,aACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAGL,UAAW,WAAM,MAAA,IAAK,GAAU,GAAY,YAC5C,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAW,EAAG,SACpC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAW,GAAY,EAAG,YCnCtD,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CACL,EAAG,WACD,GAAM,GAAO,GAAQ,EAAG,GAAO,IAEzB,EAAa,GAAO,IACpB,EAAQ,GAAO,IAEf,EAAqB,EAAI,EAAI,GAC7B,EACF,EAAI,EAAI,EAAI,GAAa,GAAI,GAAK,EAAG,aAEzC,MAAO,IAAM,EAAM,EAAoB,OC3C/C,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,EAAG,GAAI,GAAO,GAAI,SC9BnD,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAK,EAAG,YAAa,OC7BlD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GActC,OAbO,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,KAER,EAAa,EAAE,MACf,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAQT,EAAoC,GACjC,EAAI,EAAG,EAAI,EAAG,KAAM,IAC3B,EAAS,KAAK,CAAC,EAAO,GAAI,EAAW,GAAK,EAAO,GAAK,EAAM,KAE9D,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,OC3C7B,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,IACD,EAAW,GAEX,EAAW,EAAI,EAAI,GACzB,MAAO,CACL,OAAQ,WAAM,MAAA,IAAI,EAAU,EAAI,GAAI,EAAU,CAAC,GAAM,GAAW,QClCtE,AAuBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAQ,QC7BrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,SACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,KAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAO,EAAI,OC1BhC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAI,GAAK,GAAK,EAAG,YAAa,QC/B3D,AAuBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,GAAK,EAAG,WAAY,QC5BrD,AAwBO,GAAM,IAA0C,CACrD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAM,GAAO,GACb,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MACrC,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MAC3C,MAAO,CAAC,EAAS,EAAG,KChCxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,GAAI,GAAM,EAAE,QAG7B,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAkB,EAAE,MAAM,QACzB,EAAA,EAAA,KAED,EAAO,GAAe,EAAM,EAAE,OACpC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,EAAI,EAAY,GAAK,EAAE,MAAO,YAE3C,MAAO,CAAC,EAAG,WAAM,MAAA,OCxCrB,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAO,GAAI,SC9BxC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAO,GAAI,GAAO,IAAK,OC/BpD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KAED,EAAO,WACX,GAAI,GAAQ,GAAU,GAGtB,GAAI,EAAE,OAAS,EACb,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,IAAK,CAAC,EAAE,MAAM,cAEjD,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IAAK,CACjD,EAAE,MAAM,GAAI,EAAE,MAAM,cAI7B,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IACjD,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,cAI1C,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EACA,CACE,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAC5C,EAAI,EAAE,MAAM,IAEd,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,UAMjE,MAAM,IAAI,OACN,2DACG,GAAE,KAAI,kBAEf,MAAO,IAET,MAAO,CAAC,EAAG,KCpFf,AAuBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAiC,EAChC,EAAA,EAAA,KACD,EAAW,GAAiC,GAClD,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,EAAI,OC7BnC,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACxC,GAAM,GAA2B,EAC1B,EAAA,EAAA,KACP,MAAO,CAAC,MAAO,WAAM,MAAA,IAAM,EAAI,OC1BnC,AA8BO,GAAM,IAA2C,CACtD,WAAY,GACZ,aAAc,CAAC,cACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAAO,WACX,MAAO,IAAoB,EAAI,IAEjC,MAAO,CAAC,EAAG,KAIf,YAA+C,EAAM,GAQnD,OAJM,GAAqB,GAAQ,EAAS,GAAU,IAChD,EAAW,GAAO,EAAG,GACvB,EAAa,GAAa,EAAS,GAAO,EAAG,UAC3C,EAAW,EAAS,KAAO,EAAW,KACnC,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,GAAW,EAAY,EAAI,GAE1C,EAAa,GAAW,EAAY,GAAK,EAAS,MAAO,SACzD,GAAM,GAAY,GAAU,GAC5B,MAAO,IAAM,EAAY,EAAU,GCxDrC,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAkHA,GAAM,IAA4B,CAChC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA6B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAxB,GAAM,IAAc,GAAA,IACvB,GAAiB,ICxNnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA+BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACL,EAAO,KAAK,OAAS,EAAG,WAAM,MAAA,wCACvB,EAAQ,KAAM,KClCvB,AAmCA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAQ,KAAM,ICtCvB,AA8BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AAkCA,EAAO,UAAU,KAAO,SACpB,EAAc,GAChB,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,KCrC9B,AAoCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,GACjC,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,KCvCvC,AAqCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,GAChD,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,KCxC9C,AAwCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,EAC9C,GACF,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,EAAQ,KC5CtD,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA+BA,EAAO,UAAU,UAAY,SACzB,EACA,EACA,EACA,EACA,GACF,YAAK,kBACE,GAAU,KAAM,EAAM,EAAU,EAAQ,EAAO,ICtCxD,AA2BA,EAAO,UAAU,YAAc,SAAyB,GAEtD,YAAK,kBACE,GAAY,KAAM,IC9B3B,AA4BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,YAAc,SAC3B,EAAa,GACf,YAAK,kBACE,GAAY,KAAM,EAAK,IC/BhC,AA0BA,EAAO,UAAU,OAAS,SACtB,EAA0B,GAC5B,YAAK,kBACL,AAAI,YAAa,IACf,GAAI,CAAC,IAEA,GAAM,CAAE,MAAI,OAAK,GAAI,IChC9B,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4C,EAC5C,EAAmB,GACrB,YAAK,kBACE,GACI,KAAM,EAAQ,EAAQ,EAAK,EAAY,EACvC,ICrCb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EACA,EAAkC,EAClC,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAa,EAAS,EAAK,ICrCtD,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAqB,GACtC,YAAK,kBACE,GAAO,KAAM,EAAM,EAAW,IC9BvC,AA2BA,EAAO,UAAU,aAAe,SAC5B,EAAmB,GACrB,YAAK,kBACE,GAAa,KAAM,EAAW,IC9BvC,AAkCA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,UAAgB,8DAChB,KAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,IC3Cb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA6BA,EAAO,UAAU,WAAa,SAC1B,EAA+B,EAC/B,EAAqB,EACrB,GACF,YAAK,kBACE,GAAW,KAAM,EAAQ,EAAS,EAAK,EAAW,IClC3D,AA2BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC9BxB,AA4BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC/BzB,AA2BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC7BnB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC5Bb,AA+BA,EAAO,UAAU,YAAc,SAClB,GACX,YAAK,kBACE,GAAY,KAAM,IClC3B,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA0BA,EAAO,UAAU,WAAa,SAA2B,GACvD,YAAK,kBACE,GAAW,KAAM,IC5B1B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA4BA,EAAO,UAAU,OAAS,SACb,EAA4B,GACvC,YAAK,kBACE,GAAO,KAAM,EAAS,IC/B/B,AA+BA,EAAO,UAAU,mBAAqB,SACzB,GACX,YAAK,kBACE,GAAmB,KAAM,IClClC,AA0BA,EAAO,UAAU,aAAe,SAA2B,GAEzD,YAAK,kBACE,GAAa,KAAM,IC7B5B,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA+BA,EAAO,UAAU,gBAAkB,SACtB,GACX,YAAK,kBACE,GAAgB,KAAM,IClC/B,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA0BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC5BpB,AA2BA,EAAO,UAAU,2BAA6B,SAC1C,EAAsB,EAAe,EAAgB,GACvD,YAAK,kBACE,GAA2B,KAAM,EAAa,EAAM,EAAO,IC9BpE,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA2BA,EAAO,UAAU,UAAY,SAChB,EAAwB,GACnC,YAAK,kBACE,GAAU,KAAM,EAAM,IC9B/B,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA0BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC5BpB,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA2BA,EAAO,UAAU,OAAS,SACb,EAAiB,EAAsB,GAClD,YAAK,kBACE,GAAO,KAAM,EAAG,EAAY,IC9BrC,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,KAAO,SACpB,EAAwB,GAC1B,YAAK,kBACE,GAAK,KAAM,EAAM,IC7B1B,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,EAAI,KAAM,IC5BnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,KAAO,SACpB,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAK,EAAM,IClC/B,AA+BA,EAAO,UAAU,eAAiB,SACrB,GACX,YAAK,kBACE,GAAe,KAAM,IClC9B,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAa,GAAb,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAC9B,KAAK,kBACE,GAAO,KAAM,EAAO,EAAS,IC9BtC,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,IAAM,SACV,EAAmC,GAC9C,YAAK,kBACE,GAAI,KAAM,EAAU,IC/B7B,AA6BA,EAAO,UAAU,KAAO,SACX,EAAsC,EAC/C,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAa,EAAa,EAAS,EAAc,IClCrE,AA+BA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,GAAU,KAAM,ICjCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,MAAQ,SACZ,GACX,YAAK,kBACE,GAAM,KAAM,IC7BrB,AA2BA,EAAO,UAAU,KAAO,SACX,EAAwB,GACnC,YAAK,kBACE,GAAK,KAAM,EAAM,IC9B1B,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA0BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC5Bf,AAkCA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,EAAQ,KAAM,EAAE,QCpCzB,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,EAAQ,KAAM,IC5BvB,AA2BA,EAAO,UAAU,eAAiB,SACrB,EAA8B,GACzC,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,sBAAwB,SAC5B,EAA8B,GACzC,YAAK,kBACE,GAAsB,KAAM,EAAY,IC9BjD,AA0BA,EAAO,UAAU,QAAU,SACd,GACX,YAAK,kBACE,GAAQ,KAAM,IC7BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EAAsC,EACtC,EAAqB,EACrB,GACF,YAAK,kBACE,GACI,KAAM,EAAiB,EAAiB,EAAS,EAAK,EACtD,ICtCb,AA4BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,OC9BjB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA6BA,EAAO,UAAU,MAAQ,SACZ,EAAwB,GACnC,YAAK,kBACE,GAAM,KAAM,EAAO,IChC5B,AA4BA,EAAO,UAAU,QAAU,SAAoC,GAC7D,YAAK,kBACE,GAAQ,KAAM,IC9BvB,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,MAAQ,SACrB,EAAkC,GACpC,YAAK,kBACE,GAAM,KAAM,EAAiB,IC9BtC,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAO,OC9BhB,AA2BA,EAAO,UAAU,kBAAoB,SAA2B,GAE9D,YAAK,kBACE,GAAkB,KAAM,IC9BjC,AA+BA,EAAO,UAAU,wBAA0B,SAC9B,GACX,YAAK,kBACE,GAAwB,KAAM,IClCvC,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,MAAQ,SACrB,EAAoB,GACtB,KAAK,kBACL,GAAM,GAAqB,YAAa,GAAS,CAAC,KAAM,GAAE,CAAI,MAAI,OAAK,GACvE,MAAO,IAAM,EAAoB,IC9BnC,AA4BA,EAAO,UAAU,KAAO,SAAoC,GAC1D,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA+BA,EAAO,UAAU,aAAe,SACd,EAAiB,EAAe,EAC9C,EAAoB,EAAkB,EACtC,EAAsB,GACxB,YAAK,kBACE,GACI,KAAM,EAAO,EAAK,EAAS,EAAW,EAAS,EAC/C,EAAa,ICtC1B,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC7BzB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC7BpB,AAgCA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAQ,KAAM,SClCvB,AAgCA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,KAAM,YClCvB,AAgCA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAQ,KAAM,UClCvB,AA4BA,EAAO,UAAU,KAAO,SACX,EAAY,GACvB,YAAK,kBACE,GAAK,KAAM,EAAG,IC/BvB,AA2BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC9BzB,AA2BA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAO,KAAM,IC9BtB,AA4BA,EAAO,UAAU,mBAAqB,SACzB,EAAmC,GAC9C,YAAK,kBACE,GAAmB,KAAM,EAAY,IC/B9C,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,MAAQ,SACrB,EAA8B,GAChC,YAAK,kBACE,GAAM,EAAW,KAAM,IC/BhC,AA4BA,EAAO,UAAU,UAAY,WAC3B,YAAK,kBACE,GAAU,4vQC9BnB,AAaA,GAAI,kBAMF,MAAI,KAAY,MACd,IAAW,EAAA,UAAU,WAEhB,iBAeP,MAAO,eCrCT,AAuBA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAChC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAa,aAE7C,MAAA,IANkC,OAWlC,EAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OAWhC,GAAA,SAAA,GAAyC,EAAA,EAAA,GACvC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAoB,aAEpD,MAAA,IANyC,OAWzC,GAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OC9EhC,YAuB6B,EAAY,GACvC,GAAI,MAAM,QAAQ,IAGhB,OADI,GAAkB,GACb,EAAI,EAAG,EAAI,EAAW,IAC7B,EAAW,EAAS,OAAO,GAE7B,MAAO,QAEP,GAAM,GAAW,GAAI,OAAM,GAC3B,SAAS,KAAK,GACP,eAIY,EAAc,GACnC,GAAI,CAAC,EACH,KAAM,IAAI,IAAe,eAOJ,EAAY,GAEnC,OADI,GAAU,EACK,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,IAAS,GACX,IAGJ,MAAO,eAQ2B,GAClC,MAAI,GAAG,SAAW,EACT,EAAG,GAEL,cAYc,GACrB,MAAI,OAAM,QAAQ,GACT,EAEF,CAAC,eA0BkB,GAC1B,GAAM,GAAe,EAAK,QAAQ,uBAAwB,SACpD,EACF,EAAa,QAAQ,kBAAmB,SAAS,cAKrD,MAAI,GAAS,KAAO,IACX,EAEF,UAAY,cAGO,GAE1B,MAAI,GAAW,QAAU,GAIrB,EAAW,QAAQ,OAAS,GAHvB,EAMF,EAAW,QAAQ,cAAe,SAAC,EAAG,GAAO,MAAA,GAAG,gBAIzD,GAAI,IAAyB,eAEQ,GAEnC,GAAI,GAAa,KACf,MAAO,MAET,GAAM,GAAsC,GAC5C,SAAK,UAAe,EAAS,eAC7B,EAAK,OAAY,EAAS,YACnB,EAcT,YAAuC,GAErC,GAAI,GAAU,MAAQ,MAAO,IAAW,SACtC,OACK,GAAI,MAAM,QAAQ,GACvB,EAAO,QAAQ,SAAA,GAAc,MAAA,IAA8B,SAG3D,QADM,GAAS,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAO,GACrB,AAAI,GAAS,MAAQ,MAAO,IAAU,UACpC,CAAI,CAAC,MAAM,QAAQ,IAAU,EAAM,OAAY,WAC3C,MAAO,GAAM,OAAa,SAC5B,EAAO,GAAS,EAAM,MAEtB,GAA8B,iBAqBpC,EACA,EACA,EACA,EAAgC,aAElC,GAJE,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,UAAgC,AAAA,IAAA,QAAA,GAAA,IAE9B,MAAO,IAAe,UACxB,GAAM,GAAe,EACjB,EAAE,OACN,GAAI,IAAgB,GAClB,EAAK,EAAc,WACV,IAAgB,IACzB,EAAK,GAAuB,WAE5B,EAAK,EAAc,GACf,GAAM,KACR,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAU;EAE7C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAKR,MAAO,QAGP,GAAM,GAAS,EACf,GAAI,EAAO,WAAgB,MAAQ,EAAO,QAAa,KACrD,KAAM,IAAI,GACH,EAAmB,6BACnB,MAAK,UAAU,GAAO;GACzB,sCAEN,GAAM,GAAY,EAAO,UACrB,EAAG,OAAE,EAAU,OAQnB,GAPA,AAAI,IAAa,GACf,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,GAAA,UAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IAEJ,GAAO,KACT,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAS;EAE5C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAIN,GAAI,GAAc,MAQhB,OADM,GAAwB,GACZ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,GAAuB,GAEtD,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,EAAc,GAG7C,GAAM,GAAe,EAAO,OAC5B,EAAa,cAAmB,EAGhC,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAE9C,GAA8B,EAAO,QACrC,GAAM,GACF,EAAW,EAAK,EAAO,OAAW,EAAe,GACrD,UAAsB,GAAA,GAAO,GAEtB,OAMP,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAK9C,GAAM,GAAY,GAAI,GAAI,EAAO,QACjC,UAAsB,GAAA,GAAO,GACtB,gBAUiB,EAAW,GACvC,MAAQ,GAAI,EAAK,GAAO,EAAI,EAAK,EAAI,cAQF,EAAW,GAC9C,MAAO,GAAK,GAAc,EAAG,eA2CL,GACxB,GAAI,GAAM,KACR,MAAO,GAIT,OAFM,GAAW,GAED,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAI,QAAQ,KAAO,IACrB,EAAI,KAAK,GAGb,MAAO,eASqB,GAC5B,GAAI,GAAO,KACT,KAAM,IAAI,GAAW,yBAAyB,KAAK,UAAU,IAE/D,OAAW,KAAO,GAChB,GAAI,EAAI,eAAe,GACrB,MAAO,GAGX,MAAO,eAWL,EAAkB,EAAe,GACnC,GAAI,GAAS,KACX,OAEF,GAAI,EAAO,QAAQ,GAAS,EAC1B,KAAM,IAAI,GAAc,EAAK,mBAAmB,EAAK,uBACjD,EAAM,mCAoBV,EAAQ,EAAsB,EAC9B,GAD8B,MAAA,KAAA,QAAA,GAAA,GAC9B,AAAA,IAAA,QAAA,GAAA,UACF,GAAO,GAAa,GACpB,GAAO,GAAa,GAEhB,MAAM,QAAQ,IAAM,EAAE,QAAU,GAAa,EAAE,QAAU,GACzD,EAAE,MAAM,SAAA,GAAK,MAAA,OAAO,KAAM,gBAWM,EAAwB,GAC5D,AAAI,MAAM,QAAQ,GAChB,GAAA,KAAK,OACD,EAAM,OAAS,EAAG,WAAM,MAAG,GAAI,qCACnC,EAAM,QACF,SAAC,EAAG,GAAM,MAAA,IAAsB,EAAG,WAAW,GAAI,GAAC,OAAO,MAE9D,EAAA,KAAK,OACD,OAAO,UAAU,IAAU,EAAQ,EACnC,WAAM,MAAA,YAAY,EAAI,sCACf,IAAuB,GAAM,mBAeL,GACrC,MAAI,KAAU,KACL,OACE,MAAM,QAAQ,GAChB,IAAM,EAAM,IAAI,SAAA,GAAK,MAAA,IAAuB,KAAI,KAAK,KAAO,IAC1D,MAAO,IAAU,SACnB,IAAI,EAAK,IAET,GAAG,cAcV,EAA8B,GAChC,GAAI,GAAW,EAAA,KAAK,MAChB,EACE,EAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,GAAM,GAAM,EAAA,KAAK,MACjB,MAAI,GAAM,EAAW,GAGrB,GAAW,EACX,EAAa,EAAC,MAAA,OAAI,IACX,GAET,MAAO,eASkC,GAEzC,MAAI,KAAmB,OACd,OAEL,IAAmB,SACd,SAEL,IAAmB,MACd,MAEF,KCvgBT,AAoBA,YAAqB,EAAW,GAC9B,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAI,EAAM,OAY1D,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAGE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GANyC,EAAA,cAAc,cA8BvD,GAAA,SAAA,GAA6B,EAAA,EAAA,GAQ3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAJQ,SAAA,gBAAkB,EAClB,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,YAAgB,EAAO,EAAG,EAAK,UAC/C,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,SAAU,KAAK,SAAU,KAAM,KAAK,OAtB9B,EAAA,UAAY,UAwB9B,GA1B6B,IA2B7B,EAAA,cAAc,cAAc,IAoB5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAK5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAFQ,SAAA,YAAc,EAG7B,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,KAAW,GAAY,EAAG,EAAK,WAG9D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OAdL,EAAA,UAAY,WAgB9B,GAlB8B,IAmB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAIE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAHF,EAAA,UAAY,SAK9B,GAP4B,IAQ5B,EAAA,cAAc,cAAc,IAoC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAY9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KANQ,SAAA,gBAAkB,EAClB,EAAA,gBAAkB,EAClB,EAAA,YAAc,EACd,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,YACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,IACZ,EAAA,IACI,EAAK,KAAM,EAAA,YAAgB,EAAO,EAAK,SAAU,EAAK,WAC1D,EAAA,IAAQ,EAAM,EAAK,KAAM,IAC7B,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,SAAU,KAAK,SACf,SAAU,KAAK,SACf,KAAM,KAAK,KACX,KAAM,KAAK,OApCC,EAAA,UAAY,aAuC9B,GAzCgC,IA0ChC,EAAA,cAAc,cAAc,IAQrB,GAAM,IACwC,CAC/C,QAAW,UACX,WAAc,aACd,OAAU,SACV,SAAY,wBAGkB,GAElC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA0C,GAC1C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCpQjC,YAyBwB,GACtB,MAAO,IAAI,IAAQ,eAQI,GACvB,MAAO,IAAI,IAAS,iBASpB,MAAO,IAAI,gBAIc,GACzB,MAAO,IAAI,IAAW,0ECjDxB,AAqBO,GAAM,IAA2B,CAAC,gBAAiB,gBAO7C,GAA4B,CAAC,QAAS,OAAQ,UAI9C,GAAyB,CAAC,MAAO,OAIjC,GAAkC,CAAC,MAAO,MAAO,SAAU,OCpCxE,AAmBA,GAAM,IAA+B,GAAI,iBAET,GAC9B,GAA0B,GAA0B,aAAc,eAGnC,GAC/B,GAA0B,GAA2B,cAAe,eAGxC,GAC5B,GAA0B,GAAwB,WAAY,GAGhE,GAAM,IAA4B,GAC5B,GAAoB,gBAKG,EAAc,GACzC,GAAgB,KAAK,GACrB,IACE,GAAM,GAAS,IACf,UAAgB,MACT,QACA,GACP,SAAgB,MACV,GAOV,cACE,MAAI,IAAgB,SAAW,EACtB,GAEA,GAAgB,KAAK,IAAqB,eASjB,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,MAAO,MAA2B,cAaA,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,AAAK,GAAQ,IAAI,IACf,GAAQ,IAAI,EAAY,GAE1B,GAAM,GAAQ,GAAQ,IAAI,GAG1B,GAFA,GAAQ,IAAI,EAAY,GAAQ,IAAI,GAAc,GAE9C,EAAQ,GACV,GAAM,GAAY,EAAU,IAAI,EAGhC,UAAQ,IAAI,EAAQ,GACb,MAEP,OAAO,GAIX,GAAM,IAAkB,GAAI,QAAO,+CAOD,GAChC,MAAO,CAAC,CAAC,EAAK,MAAM,ICjHtB,YA8B0B,GACxB,MAAO,KAAM,SAAS,EAAE,WAAY,gBAWlC,EAA4B,EAAgB,GAC9C,AAAI,GAAS,MACX,GAAQ,GAEV,AAAI,GAAO,MACT,GAAM,EAAM,QAId,OADI,GAAO,EACF,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,GAAQ,EAAM,GAEhB,MAAO,GAQT,YAAmB,GACjB,SAAQ,MAAM,QAAQ,GAAS,GAAI,cAAa,GAAS,EAClD,EAAA,SAAS,eAQE,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAQ1B,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAsDxB,EAAe,GACnC,GAAI,EAAM,EACR,KAAM,IAAI,GAAW,QAAQ,EAAG,cAAc,EAAK,mBAGrD,OADM,GAAgB,GACb,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,EAAI,KAAK,GAEX,MAAO,GChJT,YAwEqB,EAAW,GAC9B,MAAO,GAAE,OAAO,eASS,EAAW,GAAA,AAAA,IAAA,QAAA,GAAA,IACpC,GAAM,GAAW,EAAE,MAAM,QACzB,MAAI,GAAO,GACT,GAAO,EAAS,OAAS,EAAO,GAElC,EAAS,OAAO,EAAM,EAAG,GAClB,EAAE,QAAQ,eAcI,EAAW,GAChC,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,oDACA,SAAQ,EAAE,MAAM,OAAM,aAE5B,GAAM,GAAI,GAAW,EAAG,GACxB,MAAO,IAAK,EAAG,CAAC,EAAG,EAAG,kBASF,GACtB,GAAM,GAAW,CAAC,GAAqB,EAAE,QACzC,MAAO,GAAE,QAAQ,eAWU,GAC3B,GAAI,EAAE,MAAQ,EACZ,KAAM,IAAI,GACN,wDAAwD,EAAE,KAAI,KAEpE,GAAM,GAAW,CAAC,EAAE,MAAM,GAAI,GAAqB,EAAE,MAAO,IAC5D,MAAO,GAAE,QAAQ,eAYf,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,GAAI,CAAC,EAAM,EAAM,MAAM,SACnD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,GAC9B,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,EAAG,GACjC,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,MAAU,EAAmB,CAAC,EAAO,EAAG,EAAG,EAAG,GAAI,CACvD,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SAEjE,GACH,MAAO,GAAA,MAAU,EAAO,CAAC,EAAO,EAAG,EAAG,EAAG,EAAG,GAAI,CAC9C,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAClE,EAAM,MAAM,aAGd,KAAM,IAAI,GACN,8DACA,IAAG,EAAM,sBAcjB,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,GAAQ,CAAC,EAAM,MAAM,GAAI,QACjD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,GAC1B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,QAClC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAG,GAC7B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,YAErD,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAejB,EAAe,EAAe,EAAc,GAC9C,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,GAC9B,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,EAAG,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAO,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpD,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,YAGX,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAWO,EAAmB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC7C,GAAI,GACJ,MAAI,GAAO,GACT,GAAO,EAAQ,GAAG,KAClB,AAAI,IAAS,EACX,EAAO,EAEP,EAAO,GAGX,AAAI,IAAS,EAAQ,GAAG,MAGtB,GAAO,IAGF,EAAA,OAAW,EAAS,eAUQ,EAAW,GAC9C,OAAQ,EAAE,UACH,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,QACjC,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,WAEpD,KAAM,IAAI,GACN,kDACA,iBAAgB,EAAE,oBAWP,EAAW,GAI9B,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAI,CAAC,IAEH,EAAE,OAAS,EAAE,OACf,KAAM,IAAI,GACN,0BAA0B,EAAE,OAAM,oBAClC,yCAAwC,EAAE,KAAI,MAEpD,MAAO,GAAA,KAAS,EAAG,eAgBjB,EAAc,EAAY,EAAc,EACxC,GADc,MAAA,KAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAErB,EAAA,aAAiB,EAAO,EAAM,EAAQ,EAAO,eAsBlD,EAAW,EAAW,EACtB,GACF,GAAK,EAAE,KAAO,GAAO,EAAE,KAAO,EAC5B,KAAM,IAAI,IACN,2CACA,uBAAsB,EAAE,MAAK,kBAAkB,EAAE,QAEvD,GAAI,EAAE,MAAQ,GACZ,GAAM,GAAW,EAAE,MAAM,MAAM,IAAI,GAC7B,EAAiB,EAAE,MAAM,MAAM,IAAI,GACzC,GAAI,IAAa,EACf,KAAM,IAAI,IACN,2CACA,yDACI,EAAE,MAAK,SACX,eAAc,EAAE,QAIxB,GAAK,EAAE,OAAS,GAAO,EAAE,OAAS,GAChC,GAAM,GAAa,GACb,EAAa,GAInB,MAAO,GAAA,MAAU,OAAO,CACtB,EAAC,EACD,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,SAIZ,GAAM,GAAa,EAAE,MAAM,QACrB,EAAW,EAAW,MAC5B,EAAI,EAAE,QAAQ,CAAC,GAAI,IAInB,GAAM,GAAS,EAAE,MAAM,QACjB,EAAW,EAAO,MAClB,EAAiB,EAAO,MACxB,EAAiB,EAAM,OAAA,CAAE,IAGzB,EAAO,MAAM,KAAK,CAAC,OAAQ,EAAE,MAAO,SAAC,EAAG,GAC5C,MAAI,KAAM,EACD,EAAE,KAAO,EACP,GAAK,EAAE,KAAO,EAChB,EAAI,EAEN,IAET,EAAI,EAAE,UAAU,GAAM,QAAQ,CAAC,EAAgB,KAG/C,GAAM,GAAkB,EAAU,OAAK,GACjC,EAAa,GACb,EAAa,GACnB,MAAO,GAAA,MACF,OAAO,CACN,EAAC,EACD,EAAC,EACD,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,IAEX,QAAQ,gBAyDb,EAAmB,EAA4B,GACjD,MAAO,GAAA,KAAK,WACV,MAAI,OAAM,QAAQ,GAChB,EAAU,EAAA,SAAS,EAAS,SAE5B,EAAU,EAAQ,QAEb,EAAA,OAAW,EAAW,EAAS,iBASnB,GACrB,MAAO,GAAA,IAAQ,EAAG,GA+BpB,YAAqB,EAAe,EAAc,GAChD,GAAM,GAAY,EAAK,MAEvB,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,GACN,+BAA+B,EAAK,KACpC,6BAA4B,IAGlC,GAAI,IAAU,GACZ,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,EAAG,IAErC,EAAK,QACR,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,GAAI,EAAU,KAEzD,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAG,EAAU,KAEpC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,IAElC,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,KAE3D,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAU,KAEjC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,IAE/B,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,KAE7C,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAU,KAE9B,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,EAAQ,EACjB,MAAO,GAET,KAAM,IAAI,GAAW,sCAAsC,EAAK,kBAc9D,EAAW,EAAc,GAC3B,MAAO,GAAA,KAAK,WACV,MAAI,IAAc,MAChB,GAAa,MAEf,GAAgB,GAET,EAAE,IAAI,GAAY,EAAE,KAAM,EAAM,kBAUvB,EAAW,GAE7B,GAF6B,AAAA,IAAA,QAAA,GAAA,GAEzB,IAAU,EACZ,KAAM,IAAI,IACN,0CAA0C,EAAK,6BAGrD,MAAO,GAAA,IAAQ,eAWQ,GACvB,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,GAAG,IAAI,kBAc1C,EAAW,EAAe,EAAuB,GACnD,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAG,EAAO,EAAY,iBAY1B,GAC1B,MAAO,GAAA,KAAK,WACV,GAAM,GAAI,EAAA,IAAQ,GAAI,EAAA,IAAQ,GAAI,IAClC,MAAO,GAAA,YAAgB,EAAG,EAAG,iBAiBD,EAAY,EAAc,GAAA,MAAA,KAAA,QAAA,GAAA,IACjD,EAAW,IAAM,IC7sB1B,AAgBO,GAAM,IAAwB,CAAC,QAAS,SAAU,UAO5C,GACT,CAAC,SAAU,UAAW,mBCxB1B,YAoB6B,GAC3B,GAA0B,GAAuB,UAAW,eAG5B,GAChC,GAA0B,GAA2B,eAAgB,GASvE,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DACS,SAAA,UAAA,4BAAP,WACE,MAAO,IAUT,EAAA,UAAA,UAAA,WACE,MAAO,IAEX,GAf0C,EAAA,cAAc,cAiBxD,GAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,MAAM,EAAO,IAHf,EAAA,UAAY,QAKrB,GAP2B,IAQ3B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,KAAK,EAAO,IAHd,EAAA,UAAY,OAKrB,GAP0B,IAQ1B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,GACN,oDAAoD,GAE1D,GAAI,EAAK,QAAU,OACjB,KAAM,IAAI,GAAW,sCAAsC,GAE7D,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAI,EAAA,OAAO,EAAK,OAAQ,EAAA,KAAK,EAAO,OAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,QApBT,EAAA,UAAY,WAuBrB,GAzB8B,IA0B9B,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GASjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,eAAiB,KACjB,EAAA,eAAiB,IAOxB,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,cAAc,EAAO,KAAK,OAAQ,KAAK,OAAQ,IAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,OAAQ,KAAK,OAAQ,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAnBxD,EAAA,UAAY,gBAqBrB,GAvBmC,IAwBnC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAShC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,uCAAuC,EAAK,KAGlD,MAAO,IAAe,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGnE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,eA2BrB,GA7BkC,IA8BlC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,0CAA0C,EAAK,KAErD,MAAO,GAAA,gBAAgB,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGpE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,kBA2BrB,GA7BqC,IA8BrC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,IAG9C,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,SAAW,GAAK,EAAM,KAAO,EAAM,GAC3C,KAAM,IAAI,GACN,wEAGJ,MAAO,GAAA,IAAI,EAAK,KAAM,EAAA,IAAI,EAAM,QAKtC,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OApBd,EAAA,UAAY,WAsBrB,GAxB8B,IAyB9B,EAAA,cAAc,cAAc,IAU5B,YACI,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,gBAChB,GAAI,GACA,EAEJ,GADA,GAAgB,GACZ,EAAM,SAAW,EACnB,EAAQ,EAAM,GACd,EAAS,EAAM,WACN,CAAC,EAAG,EAAG,GAAG,QAAQ,EAAM,UAAY,IAC7C,GAAI,IAAe,iBACjB,GAAM,GAAqB,GAAU,EAAO,GAC5C,EAAQ,EAAM,GAAK,EACnB,EAAS,EAAM,GAAK,UACX,IAAe,gBACxB,GAAM,GAAqB,GAAU,EAAO,EAAG,EAAM,OAAS,GAC9D,EAAQ,EAAM,EAAM,OAAS,GAAK,EAClC,EAAS,EAAM,EAAM,OAAS,GAAK,QAGrC,GAAM,GAAY,GAAU,GAC5B,EAAQ,KAAK,KAAK,GAClB,EAAS,KAAK,KAAK,GAGrB,MAAO,CAAC,EAAO,GAiBjB,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAYnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,EAAK,MAAQ,EACf,KAAM,IAAI,GACN,wCAAwC,EAAK,OAEnD,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAM,EAAK,MAC7C,EAAK,KAAO,EAAK,MAAQ,KAAO,QAAU,EAAK,KAC/C,GAAa,EAAK,MAClB,EAAK,aACD,EAAK,cAAgB,KAAO,SAAW,EAAK,aAChD,GAAkB,EAAK,cACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAO,GAAY,GACnB,EAAQ,EAAK,GACb,EAAS,EAAK,GAChB,EAAQ,KAAK,MASjB,GARA,AAAI,KAAK,OAAS,QAChB,GAAS,KAAK,IAAI,EAAG,GAChB,AAAI,KAAK,OAAS,SACvB,GAAS,KAAK,IAAI,EAAG,GAErB,GAAS,KAAK,IAAI,EAAI,GAAQ,GAAU,GAGtC,KAAK,eAAiB,UACxB,GAAM,GAAS,KAAK,KAAK,GAEzB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACH,KAAK,eAAc,2BAA2B,EAAK,KAE5D,MAAO,GAAA,gBAAgB,EAAO,EAAG,EAAQ,EAAO,KAAK,WAErD,GAAM,GAAQ,KAAK,KAAK,EAAI,GAC5B,MAAO,GAAA,cAAc,EAAO,CAAC,EAAO,EAAO,KAI/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,MACZ,KAAM,KAAK,KACX,aAAc,KAAK,aACnB,KAAM,KAAK,OAzDR,EAAA,UAAY,kBA4DrB,GA9DqC,IA+DrC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,gBAwBrB,GA1BmC,IA2BnC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAWhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,eAwBrB,GA1BkC,IA2BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,WAiBrB,GAnB8B,IAoB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAI7B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,YAiBrB,GAnB+B,IAoB/B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAI/B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBiC,IAoBjC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAIhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBkC,IAoBlC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAO9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GATO,EAAA,aAAe,EAMtB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,KAAO,EAAK,KAEb,EAAK,MAAQ,KACf,KAAM,IAAI,IACN,2EAIR,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,IAAoB,8BAEhC,AAAI,EAAM,GAAK,EAAM,GAAK,KACxB,QAAQ,KACJ,gEACA,eAAc,EAAM,GAAK,EAAM,GAAE,gBACjC,wBAIN,GAAM,GACF,EAAM,GAAK,EAAM,GAAK,CAAC,EAAM,GAAI,EAAM,IAAM,EAC3C,EAAI,GAAe,EAAiB,EAAG,EAAG,WAC5C,EAAI,EAAA,OAAO,YAAY,GAC3B,MAAI,GAAM,GAAK,EAAM,IACnB,GAAI,EAAE,aAED,EAAA,IAAI,EAAK,KAAM,MAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CACL,KAAM,KAAK,KACX,KAAM,KAAK,OA3CR,EAAA,UAAY,aA8CrB,GAhDgC,IAiDhC,EAAA,cAAc,cAAc,IAUrB,GAAM,IACyC,CAChD,SAAY,WACZ,aAAgB,eAChB,cAAiB,gBACjB,SAAY,WACZ,UAAa,YACb,SAAY,WACZ,YAAe,cACf,aAAgB,eAChB,KAAQ,OACR,WAAc,aACd,aAAgB,eAChB,cAAiB,gBACjB,gBAAmB,kBACnB,gBAAmB,kBACnB,MAAS,SAGf,YACI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGgB,GAEnC,MAAO,IAAqB,eAGC,GAE7B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EAIJ,GAAI,IAAc,eAChB,MAAO,IAAI,IACN,GAAI,IAAc,gBACvB,MAAO,IAAI,IACN,GAAI,IAAc,WACvB,MAAO,IAAI,IACN,GAAI,IAAc,YACvB,MAAO,IAAI,IACN,GAAI,IAAc,cACvB,MAAO,IAAI,IACN,GAAI,IAAc,eACvB,MAAO,IAAI,IAEX,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAuB,OAE3B,OAAI,aAAsB,IACxB,EAEA,GAAuB,GCxoBlC,cAkBE,MAAO,IAAI,kBASX,MAAO,IAAI,gBAQY,GACvB,MAAO,IAAI,IAAS,eAYQ,GAC5B,MAAO,IAAI,IAAc,eASE,GAC3B,MAAO,IAAI,IAAa,eAaM,GAC9B,MAAO,IAAI,IAAgB,eASJ,GACvB,MAAO,IAAI,IAAS,eAgBU,GAC9B,MAAO,IAAI,IAAgB,eAgBC,GAC5B,MAAO,IAAI,IAAc,eAgBE,GAC3B,MAAO,IAAI,IAAa,eAeD,GACvB,MAAO,IAAI,IAAS,eAeI,GACxB,MAAO,IAAI,IAAU,eAgBK,GAC1B,MAAO,IAAI,IAAY,eAYI,GAC3B,MAAO,IAAI,IAAa,eAWC,GACzB,MAAO,IAAI,IAAW,kPChNxB,AAmBA,GAAI,IAAsB,gBAGxB,MAAO,MAGT,GAAM,IAA2C,eAO1B,GAAA,MAAA,KAAA,QAAA,GAAA,IACrB,AAAM,IAAU,KACd,IAAa,GAAU,GAEzB,GAAa,IAAW,EACjB,EAAS,GAAa,GAAQ,WCrCvC,YAoBgC,GAC9B,MAAO,OAAM,QAAQ,IAAM,MAAM,QAAQ,EAAE,gBASV,GACjC,MAAI,GAAE,SAAW,EACR,GAEJ,MAAM,QAAQ,EAAE,IAGd,EAFE,CAAC,eAWwB,GAClC,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAG,SAAW,EAChB,KAAM,IAAI,GAAW,uCAAuC,EAAG,QAEjE,EAAI,EAAG,OAEP,GAAI,EAEN,MAAO,eAa0B,GACjC,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,EAAO,KAChD,GAAI,EAAO,SAAW,EACpB,SAAS,EACF,EAAO,GAEd,KAAM,IAAI,GAAW,iCAAiC,EAAO,YAG/D,OAAO,GC9EX,YAmBqC,GAEnC,OADI,GAAQ,EACS,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,AAAI,EAAO,MAAM,SAAW,EAC1B,GAAS,EAET,GAAS,EAAO,MAAM,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAG/C,MAAO,GC5BT,AAoBA,GAAM,IAA+B,yBAqCnC,WACI,EAAa,EACb,EAAqC,EACrC,GAFa,AAAA,IAAA,QAAA,GAAA,WACb,AAAA,IAAA,QAAA,GAAA,IAAqC,AAAA,IAAA,QAAA,GAAA,IACrC,AAAA,IAAA,QAAA,GAAA,MACF,KAAK,MAAQ,GAAgB,UAC7B,KAAK,MAAQ,EAAI,MACjB,KAAK,GAAK,KAEV,EAAO,GAAe,GACtB,KAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,cAErC,KAAK,WAAa,EAClB,KAAK,WAAa,EAElB,KAAK,IAAM,EAAA,SAAa,EAAK,KAAK,WAAY,KAAK,KAAM,KAAK,OAUhE,SAAA,UAAA,KAAA,WACE,YAAK,oBACE,KAAK,KAUd,EAAA,UAAA,MAAA,SAAM,GAEJ,YAAK,oBACL,GAAiB,KAAK,IAAK,GAE3B,AAAI,KAAK,IAAI,KAAO,EAAO,IACzB,MAAK,IAAI,OAAO,GAChB,AAAI,KAAK,YAAc,MACrB,KAAK,IAAI,OAAO,KAAK,WAAW,MAAM,KAAK,OAGxC,MAMT,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,KAAK,IAAI,WAGD,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,IAAI,WACX,KAAM,IAAI,OAAM,kBAAkB,KAAK,KAAI,0BAI/C,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,WAAa,EAClB,KAAK,IAAI,UAAY,mCAEzB,KAEA,YAA0B,EAAa,GACrC,GAAI,EAAE,MAAM,aAAe,EAAE,MAAM,WACjC,KAAM,IAAI,OACN,mBAAqB,KAAK,UAAU,EAAE,OAAS,QAC/C,KAAK,UAAU,EAAE,oBA0LK,GAC5B,MAAO,GAAG,IAAI,SAAA,GAAK,MAAA,GAAE,qBAYnB,GACF,EAAmB,QAAQ,SAAA,GACzB,GAAM,GAA0B,EAAiB,GACjD,EAAS,MAAM,EAAiB,MClVpC,kBAuEE,WAAY,GACV,KAAK,MAAQ,EAAK,MAClB,KAAK,MAAQ,EAAK,MAKlB,AAAI,EAAK,OAAS,KAChB,KAAK,KAAO,EAAK,MAAM,OAEvB,KAAK,KAAO,EAAK,KAEnB,KAAK,QAAU,EAAK,QACpB,KAAK,QAAU,EAAK,QACpB,KAAK,KAAO,EAAK,MAAQ,GAE7B,MAAA,oBA6CE,WACa,EAA0B,EAC5B,EAA6B,EAC3B,EAAkB,EAClB,GAHA,KAAA,MAAA,EAA0B,KAAA,MAAA,EAC5B,KAAA,YAAA,EAA6B,KAAA,OAAA,EAC3B,KAAA,SAAA,EACA,KAAA,kBAAA,EACX,KAAK,GAAK,KACV,AAAI,GAAQ,MACV,MAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,eAEvC,KAAK,KAAO,EAAM,OAEtB,MAAA,MA2DI,GAAc,EAsBlB,GAAA,WAwCE,WACI,EAEO,GAAA,KAAA,SAAA,EACT,KAAK,GAAK,KAQV,KAAK,cAAgB,EAAK,cAU1B,KAAK,cAAgB,EAAK,cAE1B,KAAK,YAAc,EAAK,YAExB,KAAK,cAAgB,EAAK,cAQ1B,KAAK,aAAe,EAAK,aAEzB,KAAK,cAAgB,EAAK,cAM1B,KAAK,WAAa,EAAK,WAEvB,KAAK,YAAc,EAAK,YAKxB,KAAK,YAAc,EAAK,YAExB,KAAK,aAAe,EAAK,aAGzB,OAAoB,GAAA,EAAA,EAAA,EAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MACX,EAAM,cAAc,KAAK,MAG7B,EAAK,cAAc,aAAa,KAAK,MAGvC,SAAA,UAAA,UAAA,WAEE,OADM,GAAyB,GACX,EAAA,EAAA,EAAA,KAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,KACX,EAAa,KAAK,EAAM,MAExB,EAAa,KAAK,MAGtB,MAAO,CACL,cAAe,KAAK,cAAgB,KAAK,cAAc,KAAO,KAC9D,cAAe,EACf,YAAa,KAAK,YAClB,cAAe,KAAK,gBAG1B,KAkDI,GAAe,EAWnB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAmDlC,WAAY,GAAA,AAAA,IAAA,QAAA,GAAA,IAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAtBD,EAAA,UAAsB,KAEtB,EAAA,kBAA8B,GAQ5B,EAAA,UAAY,GAapB,EAAK,GAAK,KAEV,EAAK,oBAAsB,KAE3B,EAAK,UAAY,KACjB,EAAK,gBAAkB,GAGvB,EAAK,kBAAoB,GACzB,EAAK,qBAAuB,GAC5B,EAAK,QAAU,GACf,EAAK,SAAW,GAChB,EAAK,OAAS,GAMd,EAAK,aAAe,GACpB,EAAK,cAAgB,GAErB,GAAI,GAAO,EAAK,KAChB,GAAI,CAAC,GACH,GAAM,GAAS,EAAK,eACpB,EAAO,GAA0B,GAAU,IAAM,GAAO,GAM1D,GAJA,EAAK,KAAO,EAEZ,EAAK,WAAa,EAAK,WAAa,KAAO,GAAO,EAAK,UAEnD,EAAK,YAAc,MAAQ,EAAK,iBAAmB,MAKrD,GAAI,GAAe,OACnB,GAAI,EAAK,iBAAmB,KAC1B,EAAkB,EAAK,wBACd,EAAK,YAAc,MAC5B,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAkB,CAAC,GAAW,OAAO,EAAK,YAE5C,EAAK,gBAAkB,EAGvB,GAAI,GAAQ,EAAK,MACjB,AAAI,GAAS,MACX,GAAQ,EAAK,YAEf,AAAI,GAAS,MACX,GAAQ,WAEV,EAAK,MAAQ,EAGf,MAAI,GAAK,SAAW,KAClB,EAAK,eAAiB,EAAK,QAE3B,EAAK,eAAiB,KAKxB,EAAK,UAAY,KAEjB,EAAK,0BAA4B,KAYlB,SAAA,QAAjB,SAAyB,EAAc,GACrC,MAAO,GAAM,KAAO,OAAS,EAAU,YAUjC,EAAA,UAAA,eAAR,SAAuB,EAAmB,GACxC,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,mCACA,4BAA2B,EAAQ,MAEzC,GAAI,KAAK,aAAa,QAAU,EAC9B,KAAM,IAAI,GACN,gBAAgB,EAAQ,YAAY,EAAS,KAC7C,2BAA0B,KAAK,aAAa,OAAM,oBAExD,MAAO,MAAK,aAAa,IAY3B,EAAA,UAAA,WAAA,SAAW,GACT,MAAO,IACH,KAAK,eAAe,EAAW,SAAS,eAY9C,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,IACH,KAAK,eAAe,EAAW,UAAU,gBAgB/C,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,uHAIC,GAAI,KAAK,aAAa,SAAW,EACtC,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0CAEN,MAAO,IACH,KAAK,eAAe,EAAG,SAAS,+CActC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0BAEN,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,yHAKN,MAAO,IACH,KAAK,eAAe,EAAG,UAAU,gDAGvC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAQd,EAAA,UAAA,gBAAA,WAKE,MAAO,MAAK,OAAO,IAAI,SAAA,GAAU,MAAA,QAGnC,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,0CAGd,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,MAAO,MAAK,YAGd,SAAU,GACR,KAAK,OAAS,mCAGhB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,kBAAkB,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,IAClD,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAI,MAAK,WACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,GAAE,YAErC,QAIX,SAAqB,GACnB,KAAK,kBAAoB,mCAG3B,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAI,MAAK,UACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,CAAC,EAAE,YACxC,OAAO,KAAK,sBAEV,KAAK,kBAAkB,OAAO,KAAK,2BAI9C,SAAwB,GACtB,KAAK,qBAAuB,mCAO9B,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAG3C,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,2CAUd,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,SACR,KAAM,IAAI,OACN,yEAiBE,EAAA,UAAA,yBAAV,SAAmC,GAGjC,GADA,EAAS,GAAqB,GAC1B,KAAK,WAAa,MAAQ,KAAK,UAAU,SAAW,EACtD,OAEF,GAAM,GAAY,GAAqB,KAAK,WAC5C,GAAI,EAAO,SAAW,EAAU,OAC9B,KAAM,IAAI,GACN,SAAS,KAAK,KAAI,YAAY,EAAU,OAAM,YAC9C,oBAAmB,EAAO,OAAM,oBAChC,oBAAmB,IAEzB,OAAS,GAAa,EAAG,EAAa,EAAO,OAAQ,KACnD,GAAM,GAAI,EAAO,GACX,EAAkB,EAAU,GAClC,GAAI,GAAQ,KACV,SAIF,GAAM,GAAO,EAAE,KACf,GAAI,EAAK,MAAQ,MACX,IAAS,EAAK,KAChB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,KAC3D,kBAAiB,EAAK,KAAI,gBAAgB,IAGlD,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,IAG3D,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,EAAI,MAK/D,GAAI,EAAK,OAAS,MACZ,EAAE,QAAU,EAAK,MACnB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,IAC3D,qBAAoB,EAAK,MAAK,iBAAiB,EAAE,MAAK,MAK9D,GAAI,EAAK,MACP,GAAM,GAAS,EAAE,MACjB,OAAW,KAAO,GAAK,MACrB,GAAM,GAAO,OAAO,GACd,EAAQ,EAAK,KAAK,GAIlB,EACF,GAAQ,EAAI,EAAO,GAAQ,EAAO,EAAO,OAAS,GACtD,GAAI,GAAS,MAAQ,CAAC,EAAO,MAAM,QAAQ,KAAkB,GAC3D,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,mBAAmB,EAAI,uBACnC,eAAc,EAAK,kBAAkB,EAAM,OAMrD,GAAI,EAAK,OAAS,KAChB,OAAS,GAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAU,EAAK,MAAM,GACrB,EAAM,EAAE,MAAM,GACpB,GAAI,GAAW,MAAQ,GAAO,MACxB,IAAY,EACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,oBAAoB,EAAK,MAAK,MAC1C,gBAAe,EAAE,MAAK,SAgBtC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,IAGC,EAAA,UAAA,eAAV,SAAyB,EAAyB,GAChD,AAAI,KAAK,WAAa,MACpB,KAAK,UAAU,EAAQ,IAS3B,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,UAAY,GAOnB,EAAA,UAAA,cAAA,WACE,KAAK,UAAY,MAwEnB,EAAA,UAAA,MAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAGE,EAAS,GAAU,GAEnB,KAAK,oBAML,OAHM,GAAa,GAAqB,GAEpC,EAAiB,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,CAAE,aAAiB,MACrB,EAAiB,GACjB,OAIJ,OADI,GAAkB,GACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,YAAiB,KACnB,EAAkB,GAClB,OAIJ,GAAI,IAAmB,EACrB,KAAM,IAAI,GACN,mEAKN,MAAO,IAAU,KAAK,KAAM,WAE1B,GAAI,CAAC,EAAK,OAKR,EAAK,yBAAyB,GAI9B,OADM,GAAuB,GACT,EAAA,EAAA,EAAA,GAAqB,GAArB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,KAAK,EAAM,OAEzB,EAAK,MAAM,GAA+B,IAC1C,EAAK,MAAQ,GAGb,AAAI,EAAK,gBACP,EAAK,WAAW,EAAK,gBAGvB,AAAI,EAAK,YAAc,MAAQ,GAI7B,GAAK,UAAY,GAcrB,GANA,EAAK,yBAAyB,GAM1B,GAUF,OATI,GAAS,EAAK,KAAK,EAA6B,GAK9C,EAAuB,GAAqB,GAC5C,EAA2B,GAGnB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAT,GAAI,GAAC,EAAA,GACR,AAAI,EAAW,QAAQ,KAAO,IAC5B,GAAI,EAAE,SAER,EAAe,KAAK,GAItB,GAFA,EAAS,GAA+B,GAEpC,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAKN,MAAO,QAEP,GAAM,GAAa,GAAkB,GAC/B,EAAc,EAAK,mBAAmB,GACxC,EAAM,OACJ,EAAc,GAAiB,GAgCrC,GA/BA,EAAK,6BACD,MAAM,QAAQ,GAAU,EAAW,GACX,GAE5B,AAAI,GAAe,MAAQ,EAAY,OAAS,GAC5C,MAAM,QAAQ,EAAY,IAE5B,EAAU,EACI,IACG,SAAC,EAAO,GAAU,MAAA,IAAI,IAClB,EAAa,EAAO,EACpB,GAAqB,GAAS,EAAQ,EAAK,KAC3C,KAErB,EAAS,GAAI,IACT,EAAa,EAAsB,EACnC,GAAqB,GAAS,EAAQ,EAAK,MAUjD,EAAK,eACD,EAA6C,EAAQ,KAAM,KAC3D,EAAY,EAAa,GAC7B,EAAK,YAED,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAIN,MAAO,OAYH,EAAA,UAAA,6BAAV,SAAuC,GACrC,GAAI,KAAK,iBAAmB,KAC1B,OACK,GAAI,EAAW,SAAW,KAAK,gBAAgB,OACpD,QAAQ,KACJ,iDACG,MAAK,UAAU,GAAW,iCAC7B,qBAAoB,KAAK,UAAU,KAAK,iBAAgB,MACxD,iBAAgB,KAAK,YAEzB,GAAI,GAAc,GAClB,KAAK,gBAAgB,QAAQ,SAAC,EAAW,GACvC,AAAI,GAAa,MAAQ,EAAW,IAAM,MACtC,EAAW,KAAO,GACpB,GAAc,MAGlB,AAAI,GACF,QAAQ,KACJ,iCACA,KAAI,KAAK,UAAU,GAAW,eAC9B,mCAAkC,KAAK,KAAI,MAC3C,IAAG,KAAK,UAAU,KAAK,qBAiBjC,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,GAAI,KAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,EAC5D,KAAM,IAAI,IACN,aAAa,KAAK,KAAI,gEAI5B,OADM,GAA4B,GACf,EAAA,EAAA,EAAA,KAAK,aAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAc,KAAK,UAAU,EAAK,cACxC,AAAI,EAAgB,QAAQ,KAAiB,IAC3C,EAAgB,KAAK,GAGzB,GAAI,EAAgB,SAAW,GAC7B,GAAM,GAAe,KAAK,aAAa,GAAG,aAC1C,MAAI,OAAM,QAAQ,IAAiB,MAAM,QAAQ,EAAa,KAC1D,EAAa,SAAW,EAClB,EAAyB,GAE1B,MAIT,MAAM,IAAI,IACN,aAAa,KAAK,KAAI,+JAiB9B,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,sCAAsC,KAAK,KAAI,uFAIrD,MAAO,IAAoC,KAAK,UAclD,EAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAWf,EAAA,UAAA,WAAA,SAAW,GAAA,MAAA,KAAA,QAAA,GAAA,IACF,GAAc,EAAgB,KAAK,iBAAmB,KAAK,UAepE,EAAA,UAAA,WAAA,SAAW,GAAX,GAAA,GAAA,KACE,EAAA,KAAK,WACH,GAAM,GAAS,EAAK,QACpB,GAAI,EAAO,SAAW,EAAQ,OAK5B,KAAM,IAAI,GACN,4CAA4C,EAAK,KAAI,KACrD,iCAAgC,EAAQ,OAAM,MAC9C,gCAA+B,EAAO,OAAM,cAC5C,sBAAqB,EAAO,QAElC,GAAI,EAAO,SAAW,EACpB,OAIF,OAFM,GAAoD,GACpD,EAAc,GAAc,GACzB,EAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,GACxC,GAAM,GAAK,EAAY,GACjB,EAAI,EAAO,GACX,EAAI,EAAQ,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,GACN,sBAAsB,EAAG,MAAK,IAC9B,8CAA6C,EAAE,QAErD,EAAkB,KAAK,CAAC,EAAG,IAE7B,GAAc,MAmBR,EAAA,UAAA,UAAV,SACI,EAAc,EAAc,EAAkB,EAC9C,EAA2B,EAC3B,GAEF,GAAI,KAAK,kBAAkB,QAAQ,KAAU,GAC3C,KAAM,IAAI,GACN,yBAAyB,EAAI,cAAc,KAAK,MAEtD,KAAK,kBAAkB,KAAK,GAE5B,AAAI,GAAS,MACX,GAAQ,WAGV,AAAI,KAAK,2BACP,GAAc,GAAe,UAE/B,GAAM,GAAY,EAAY,MAAM,EAAO,GACrC,EACF,GAAI,IAAc,EAAW,EAAO,EAAM,EAAW,GACzD,SAAU,UAEV,AAAI,GAAe,MACjB,KAAK,QAAQ,WAAM,MAAA,GAAY,MAAM,EAAO,UAE9C,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,EACF,KAAK,kBAAkB,KAAK,GAE5B,KAAK,qBAAqB,KAAK,GAE1B,GAaT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,KAAK,0BAA4B,GAWnC,EAAA,UAAA,QAAA,SAAQ,SACN,GAAI,GAAU,MAAQ,MAAM,QAAQ,IAAW,EAAO,SAAW,EAC/D,OAGF,EAAS,GAAqB,GAC9B,AAAI,KAAK,UAAY,QAAa,KAAK,UAAY,MACjD,GAAA,KAAK,QAAO,KAAI,MAAA,EAAI,IAexB,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAYT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,GAAI,CAAC,KAAK,iBACR,GAAI,GAAQ,KACV,GAAI,MAAM,QAAQ,GAChB,EAAK,QAAQ,SAAA,GACX,GAAI,GAAe,KACjB,KAAM,IAAI,WACN,SAAS,EAAK,KAAI,iEAK1B,MAAM,IAAI,WACN,SAAS,KAAK,KAAI,2DAK1B,MAAO,MAIT,MAAO,IAeD,EAAA,UAAA,eAAR,SACI,EACA,EACA,EAA6B,EAC7B,EAA4B,EAC5B,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAM,GACF,GAAqB,GACzB,EAAgB,GAAqB,GACrC,EAAa,GAAqB,GAClC,EAAc,GAAqB,GACnC,EAAc,GAA+B,GAC7C,EAAe,GAA+B,GAM9C,OAHM,GAAyB,GACzB,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GAKV,EAAc,KAAK,EAAE,aACrB,EAAY,KAAK,EAAE,WACnB,EAAc,KAAK,EAAE,aAMvB,GAAI,IACA,CACE,cAAe,KACf,cAAa,EACb,YAAW,EACX,cAAa,EACb,aAAc,EACd,cAAa,EACb,WAAU,EACV,YAAW,EACX,YAAW,EACX,aAAY,GAEd,GAGJ,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,IAExC,EAAc,GAAG,YAAc,KAC/B,EAAc,GAAG,UAAY,KAAK,aAAa,OAAS,EACxD,EAAc,GAAG,YAAc,GAyBnC,EAAA,UAAA,UAAA,WACE,GAAM,GACyB,CAAC,KAAM,KAAK,KAAM,UAAW,KAAK,WACjE,MAAI,MAAK,iBAAmB,MAC1B,GAAO,gBAAqB,KAAK,iBAEnC,AAAI,KAAK,OAAS,MAChB,GAAO,MAAW,KAAK,OAElB,GAQC,EAAA,UAAA,eAAV,WACE,YAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,YAC/B,KAAK,QAAQ,QAGZ,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,UAAU,KAAK,KAAI,2BAkCvC,EAAA,UAAA,QAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,uCAIvC,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,sCAIvC,KAAK,oBAEL,GAAI,GAAuB,EAC3B,MAAI,EAAE,KAAK,YAAc,GACvB,GAAuB,KAAK,kBAGvB,CAAC,qBAAsB,KAAK,UAAW,qBAAoB,IAEtE,GAxoCoC,EAAA,cAAc,cAopClD,YAA2B,GAEzB,EACI,GAAqB,GAEzB,OADM,GAAkB,GACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAO,KAAK,EAAE,OAEhB,MAAO,IAA+B,GAYxC,YAA0B,GAExB,MAAO,sBAeL,EAAwB,EACxB,GAKF,GAJA,AAAI,IAAS,MAAS,GAAa,MAAQ,EAAY,IACrD,GAAQ,EAAO,YACf,EAAY,EAAO,WAEjB,EAAM,aAAa,SAAW,EAChC,MAAO,CAAC,GAER,GAAM,GAAO,EAAM,aAAa,GAChC,GAAI,EAAK,cAAc,SAAW,EAChC,MAAO,GAAK,aAGZ,OADM,GAAkC,GAC/B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,IAM7C,OALM,GAAI,EAAK,aAAa,GACtB,EAAQ,EAAK,cAAc,GAC3B,EAAY,EAAK,YAAY,GAC7B,EAAkB,GAAgB,EAAG,EAAO,GAElC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAc,QAAQ,KAAO,IAC/B,EAAc,KAAK,GAIzB,MAAO,GCvmDb,AA4CA,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CACJ,MAAO,EAAK,MACZ,KAAM,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAO,SAAS,cACtD,KAaF,GAXA,AAAI,EAAK,WAAa,MACpB,GAAK,UAAY,MAEnB,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,IAGhB,EAAK,UAAY,GACjB,EAAK,MAAQ,GACb,EAAK,OAAS,EAAK,OAEf,EAAK,YAAc,MAAQ,EAAK,iBAAmB,KACrD,KAAM,IAAI,GACN,qGAGN,GAAI,GAAkB,EAAK,gBAC3B,GAAI,GAAmB,MACrB,GAAI,EAAK,YAAc,KACrB,KAAM,IAAI,GACN,iFAGJ,EAAkB,CAAC,EAAK,WAAW,OAAO,EAAK,oBAI7C,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,yFAKR,GAAM,GAAQ,EAAK,OAAS,UAE5B,EAAK,gBAAkB,EACvB,EAAK,MAAQ,EAEb,EAAK,UAAY,CAAC,CAAC,MAAO,IAE1B,GAAM,GAAc,GAAI,IACpB,EAAK,MAAO,EAAK,gBAAiB,EAAM,GAAI,GAAI,EAAK,MACzD,SAAY,UAAY,EACxB,EAAY,YAAc,EAK1B,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,CAAC,GACf,cAAe,CAAC,GAChB,WAAY,CAAC,MACb,YAAa,CAAC,MACd,YAAa,CAAC,GACd,aAAc,CAAC,OAInB,SAAA,UAAA,MAAA,SACI,EACA,GACF,KAAM,IAAI,GACN,+BACA,kDAAiD,KAAK,QAG5D,EAAA,UAAA,QAAA,WAEE,MAAO,CAAC,qBAAsB,KAAK,UAAW,qBAAsB,IAGtE,EAAA,UAAA,UAAA,WACE,MAAO,CACL,gBAAiB,KAAK,gBACtB,MAAO,KAAK,MACZ,OAAQ,KAAK,OACb,KAAM,KAAK,OAzFC,EAAA,UAAY,aA4F9B,GA9FgC,IA+FhC,EAAA,cAAc,cAAc,gBAmCN,GACpB,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAC/C,KAAM,IAAI,OACN,gIAKN,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAE/C,KAAM,IAAI,GACN,oFAGN,GAAI,GAAa,EAAO,WACxB,AAAI,EAAO,OAAS,MAAQ,GAAc,MACxC,GAAa,CAAC,MAAM,OAAO,EAAO,QAGpC,GAAI,GAAQ,EAAO,MACnB,AAAI,GAAS,MACX,GAAQ,WAGV,GAAM,GAAa,GAAI,IAAW,CAChC,gBAAiB,EACjB,KAAM,EAAO,KACb,MAAK,EACL,OAAQ,EAAO,SAGX,EAAU,EAAW,aAAa,GAAG,cAC3C,MAAO,GAAQ,GC9MjB,YA0B2C,iHACzC,GAAI,GAAQ,KACV,MAAA,CAAA,GAEI,EAA+D,GAC/D,EAAiB,GACjB,EAA6B,GACnC,IAAW,IAAO,GACV,EAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACb,GAAc,EACpB,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAiB,KAAK,UAGtB,GAAS,OAAS,EACL,CAAA,EAAM,QAAQ,IAAI,IAD/B,CAAA,EAAA,UAEF,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAK,EAAK,IAAM,EAAO,GAAG,GAG5B,EAAA,QAAQ,+CAUyB,GACnC,GAAI,GAAQ,KACV,OAEF,OAAW,KAAO,IAChB,GAAM,GAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACnB,EAAM,WCjEZ,AAoBA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,QAAA,GAAA,YAFU,IAAA,IAAqB,KAM1B,GAAM,IAAyB,IA0BtC,GAAA,WAAA,aAEE,KAAA,eAAkC,KAMlC,SAAA,UAAA,UAAA,SAAU,GACR,KAAK,OAAS,GAGV,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,qFAEb,EAAA,UAAA,WAAN,SAAiB,qFASjB,EAAA,UAAA,SAAA,SAAS,KAGX,mBAqBE,WAAY,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,IAGtC,AAAI,GAAa,MACf,GAAY,IAEd,KAAK,UAAY,EACjB,KAAK,YAAc,EAGrB,SAAA,UAAA,OAAA,SAAO,GACL,KAAK,UAAU,KAAK,IAGtB,EAAA,UAAA,UAAA,SAAU,GACR,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,UAAU,KAIvB,EAAA,UAAA,SAAA,SAAS,GACP,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,SAAS,KAShB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,aAAN,SAAmB,uGACjB,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,KADO,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,WAAN,SAAiB,uGACf,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,KADS,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAI3B,KAOA,GAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,mBACE,GAAA,KAAA,OAAO,KAGH,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,OAAS,YAGV,EAAA,UAAA,WAAN,SAAiB,EAAe,yFAC9B,AAAI,GAAQ,MACV,GAAO,IAEH,EAAY,EAAK,MAAW,KAAO,EAAI,EAAK,KAClD,KAAK,MAAQ,aACF,GACT,GAAM,GAAQ,EAAK,GACnB,GAAI,MAAO,IAAU,SACnB,AAAK,EAAK,OAAO,eAAe,IAC9B,GAAK,OAAO,GAAO,GAErB,EAAK,OAAO,GAAO,EAAK,OAAO,GAAiB,EAAQ,OAExD,GAAI,GAAkB,OACtB,AAAI,IAAO,GAAK,OACd,EAAqB,EAAK,OAAO,GAEjC,EAAK,OAAO,GAAO,EAErB,GAAM,GACF,EAAA,KAAK,WAAM,MAAA,GAAA,IAAK,EAAK,OAAO,GAAO,EAAA,IAAI,EAAO,MAClD,EAAK,OAAO,GAAO,EACnB,AAAI,GAAsB,MACxB,EAAmB,mBAlBzB,IAAW,IAAO,KAAP,kBAwBP,EAAA,UAAA,WAAN,SAAiB,EAAe,2FAC9B,GAAI,GAAQ,KACV,eAAW,GACT,GAAI,EAAK,OAAO,IAAQ,sBAGxB,AAAI,MAAO,GAAK,OAAO,IAAS,SAC9B,EAAK,GAAO,EAAK,OAAO,GAAiB,EAAK,KAE9C,EAAA,KAAK,WACH,GAAM,GAAc,EAAA,IAAI,EAAA,IAAI,EAAG,EAAK,MAAO,EAAK,OAAO,IACvD,EAAK,GAAO,EACX,EAAK,OAAO,GAAgB,UAC7B,EAAA,KAAK,EAAK,cAXhB,EAAA,EAAkB,EAAA,KAAK,OAAO,QAAZ,EAAA,EAAA,OAAA,IAAP,EAAG,EAAA,KAAH,kBAiBjB,GA9DgC,mBAqEH,EAAA,EAAA,GAA7B,4DAIQ,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,MAAQ,GACb,KAAK,QAAU,YAGX,EAAA,UAAA,WAAN,SAAiB,EAAe,4EAC9B,AAAI,GAAQ,MACV,GAAO,IAET,KAAK,MAAM,KAAK,GAChB,IAAW,IAAO,GAChB,AAAI,KAAK,QAAQ,IAAQ,MACvB,MAAK,QAAQ,GAAO,IAEtB,KAAK,QAAQ,GAAK,KAAK,EAAK,mBAO1B,EAAA,UAAA,SAAN,6HACQ,EAA+D,GAC/D,EAAiB,GACjB,EAAoB,GAC1B,IAAW,IAAO,MAAK,QAErB,IADM,EAAa,KAAK,QAAQ,GACvB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,MAAO,GAAW,IAAO,UACrB,GAAc,EAAW,GAC/B,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAQ,KAAK,IAIJ,MAAA,CAAA,EAAM,QAAQ,IAAI,WACjC,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EAC7B,EAAkB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IACtD,EAAgB,UAChB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IAAM,EAAO,GAAG,mBAGpD,GA/C6B,mBA8DO,EAAA,EAAA,GAiBlC,WAAY,EAA0B,GAAtC,GAAA,GACE,EAAA,KAAA,OAAO,KAKP,GARM,EAAA,aAAe,EAIrB,EAAK,WAAa,GAAc,OAChC,AAAI,EAAK,aAAe,QACtB,GAAK,WAAa,IAEhB,EAAK,aAAe,SAAW,EAAK,SAAW,KACjD,KAAM,IAAI,OACN,mHAGN,MAAI,GAAA,KAAK,SAAS,EAAK,aAGrB,GAAK,UAAY,GACb,EAAK,UAAU,KAAK,GAAO,EAAK,aAEtC,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,MAAQ,EAAK,UAGd,SAAA,UAAA,UAAN,SAAgB,EAAe,EAAe,yGACtC,GAAgC,GAClC,KAAK,OAAS,KAChB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,MAAM,EAAO,EAAO,qBAEnC,SAAG,KAAK,EAAA,aACR,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAChC,MAAK,aAAe,EAChB,KAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,SACtB,EAAG,KAAK,EAAA,aAEV,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAC5B,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,QACtB,EAAG,KAAK,EAAA,aACH,AAAI,EAAA,KAAK,SAAS,KAAK,aAC5B,EAAG,KAAK,KAAK,UAAU,KAAK,aAAc,EAAO,IAEnD,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,mGACb,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,WAAtB,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,mGACX,MAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,SAAS,WAApB,EAAA,wCAGN,GA3GoC,gBAiHhC,EAEA,GAIF,GAHA,AAAI,GAAa,MACf,GAAY,IAEV,YAAqB,IACvB,MAAO,CAAC,GAEV,GAAI,MAAM,QAAQ,IAAc,EAAU,YAAc,IACtD,MAAO,GAGT,GAAM,GACF,GAAqB,GACzB,MAAO,GAAgB,IACnB,SAAA,GAAkB,MAAA,IAAI,IAAe,EAAgB,KAW3D,GAAA,IAAA,WAOE,cAcO,SAAA,4BAAP,SACI,EAAwB,GAC1B,EAAA,KAAK,OACD,GAAkB,GAAK,OAAO,UAAU,GACxC,WAAM,MAAA,sDACF,YAAW,KACnB,EAA4B,kBAAkB,GAC9C,AAAI,EAA4B,aAAa,IAAmB,MAC9D,GAA4B,aAAa,GAAkB,IAE7D,EAA4B,aAAa,GAAgB,KACrD,IAGS,EAAA,kBAAf,SAAiC,GAE/B,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAe,EAA4B,aAAa,CAAC,GAC/D,EAAa,QAAQ,SAAA,GACnB,GAAI,IAAS,EACX,KAAM,IAAI,GAAW,uCASZ,EAAA,MAAjB,WACE,EAA4B,aAAe,IAWtC,EAAA,gBAAP,SAAuB,GACrB,GAAM,GAA0C,GAChD,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAQ,CAAC,EACf,AAAI,GAAkB,GACpB,EAAa,KAAI,MAAjB,EAAqB,EAA4B,aAAa,IAGlE,MAAO,GAAa,IAAI,SAAA,GAAQ,MAAA,IAAI,MArEvB,EAAA,aAC6C,GAsE9D,iBAGI,EAA2B,EAAgC,EAC3D,EAAsB,EAAyB,EAC/C,EAAmB,EACnB,GACF,GAAM,GAAU,GAAI,IACd,EAAe,CACnB,GAAI,YAAiB,GAA4B,gBAAgB,IAEnE,AAAI,GAAa,MACf,EAAgB,KAAI,MAApB,EAAwB,GAE1B,EAAgB,KAAK,GACrB,GAAM,GAAe,GAAI,IAAa,GAMtC,SAAa,UAAU,CACrB,OAAM,EACN,aAAY,EACZ,QAAS,EACT,MAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,QAAS,IAEJ,CAAC,aAAY,EAAE,QAAO,GCnlB/B,YA2BI,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,QAAS,GChC9B,YAwB4B,EAAW,GACrC,MAAO,GAAA,KAAK,WACV,AAAI,EAAE,QAAU,WACd,GAAI,EAAE,OAAO,YAEf,GAAM,GAAY,EAAA,IAAQ,GAAS,GAAI,EAAM,IACvC,EAAgB,EAAA,KAAS,EAAU,MAAO,MAC1C,EAAO,EAAA,KAAS,EAAA,QAAY,EAAW,IAC7C,MAAO,GAAA,IAAQ,EAAG,iBAIW,EAAe,GAC9C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAO,IAAS,kBAG5B,EAAe,GAC/C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAO,IAAS,kBAIzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,EAAA,IAAQ,EAAO,GACtB,EACF,EAAA,YAAgB,EAAA,IAAQ,GAAQ,KAAW,OAAO,WAChD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAM,IACxC,MAAO,GAAA,IAAQ,IAAK,EAAA,KAAS,EAAW,mBAKxC,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAE9B,EAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAErC,MAAO,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAU,IAAa,kBAI/B,EAAe,GAC1C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,GAAS,GAAY,kBAInB,EAAe,GACnC,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,EAAW,kBAIE,EAAe,GAC9C,MAAO,GAAA,KAAK,WACV,GAAM,GAAM,EAAA,IAAQ,EAAA,IAAQ,EAAO,GAAQ,IACrC,EAAM,EAAA,IAAQ,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAQ,GAAQ,IACvD,MAAO,GAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAK,mBAY1B,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,KAAK,IAAI,GAChB,EAAiB,EAAA,IAAQ,EAAO,GAChC,EAAgB,EAAA,IAClB,EAAA,IAAQ,EAAgB,EAAA,SAAa,EAAA,IAAQ,GAAI,KACjD,GACJ,MAAO,GAAA,KAAS,EAAe,kBAK/B,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAI,EACF,EAAS,EAAA,QAAY,QAGrB,GAAM,GAAY,EAAA,IAAQ,EAAQ,EAAO,MAAM,OAAS,EAAG,IAC3D,EAAS,EAAA,IAAQ,EAAQ,GAE3B,SAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MACzC,EAAA,IAAQ,EAAA,IACX,EAAA,IAAQ,EAAO,UAAW,EAAA,IAAQ,IAAU,EAAO,MAAM,OAAS,kBActE,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAM,GAAa,EAAA,MAAU,GAAU,IAAS,QAChD,EAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MAChD,GAAM,GAAc,EAAO,MACrB,EACF,EAAA,OAAW,EAAY,EAAY,EAAY,OAAS,IACnD,QAAQ,GACjB,MAAO,IAAwB,EAAc,EAAQ,iBA0BrD,EAAgB,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAO,MAAO,EAAO,OACzC,KAAM,IAAI,GACN,8DACG,MAAK,UAAU,EAAO,OAAM,QAAQ,KAAK,UAAU,EAAO,SAEnE,MAAO,GAAA,KAAK,WAOV,GAAM,GAAa,EAAO,OACpB,EAAe,EAAO,MAAM,MAClC,MAAO,GAAW,IAAI,EAAO,IAAI,IAAS,IAAI,EAAa,MAAM,uBAIlC,EAAe,GAChD,MAAO,GAAA,KAAK,WACV,GAAI,GACJ,SAAI,EAAA,YAAgB,EAAO,KAAW,EAAI,MAC1C,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAG,KAC3B,EAAA,KAAS,GAA8B,EAAO,GAAI,kBAKzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,GAChD,EAAc,EAAA,YAAgB,EAAO,KAAW,GACtD,MAAO,GAAA,IACH,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAA,IAAQ,EAAa,KAAgB,kBAI5C,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,EAAA,IAAQ,EAAA,IAAQ,KAAW,IAC3C,MAAO,GAAA,KAAS,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAO,IAAW,kBAI7B,EAAe,GAC7C,MAAO,GAAA,KAAK,WACV,GAAM,GAAiB,GAAY,EAAO,IACpC,EAAiB,GAAY,EAAO,IACpC,EAAY,EAAA,IAAQ,EAAgB,GAC1C,MAAO,GAAA,IAAQ,EAAA,IAAQ,EAAW,OAkB/B,GAAM,IAAsD,CACjE,iBAAgB,GAChB,kBAAiB,GACjB,4BAA2B,GAC3B,4BAA2B,GAC3B,aAAY,GACZ,MAAK,GACL,iBAAgB,GAChB,QAAO,GACP,wBAAuB,GACvB,8BAA6B,GAC7B,mBAAkB,GAClB,0BAAyB,GACzB,QAAO,GACP,gBAAe,gBAKG,GAClB,GAAI,MAAO,IAAmB,UAC5B,GAAI,IAAkB,IACpB,MAAO,IAAU,GAEnB,GAAI,GAAS,gBAAgB,EAC7B,KAAI,GAAe,cAAc,SAAS,wBACxC,GAAS,gBAAgB,EAAc,wFAInC,GAAI,GAAW,OAErB,OAAO,GC7QX,YAyB+B,EAAe,GAC5C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,IAAQ,GAAI,EAAA,SAAa,IACrC,EAAmB,GAAO,EAAA,QAAY,EAAO,GAAY,EAAM,OACrE,MAAO,GAAA,KAAS,EAAA,MAAU,EAAO,GAAmB,kBAIpB,EAAe,GACjD,MAAO,GAAA,KACH,WAAM,MAAA,IACF,EAAA,MAAU,EAAA,OAAW,EAAO,IAAK,EAAA,OAAW,EAAO,KAAM,aAGnE,YAAuB,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,yBAI3C,EAAe,GACvC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIS,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIqB,EAAe,GAChD,MAAO,IAAuB,EAAO,eAInC,EAAe,GACjB,MAAI,GAAM,OAAS,EAAM,MACvB,GAAQ,EAAM,QAAQ,CAAC,EAAM,KAAO,KAEtC,EAAQ,EAAM,OAAO,IACrB,AAAI,EAAM,QAAU,EAAM,OACxB,GAAQ,EAAM,OAAO,EAAM,QAEtB,EAAA,MAAU,EAAO,GAAO,OAAO,WAajC,GAAM,IAAM,GACN,GAAM,GACN,GAAM,GACN,GAAM,GACN,GAAO,GACP,GAAO,GACP,GAA0B,GAC1B,GAAS,GACT,GAAgC,GAIhC,GAAuD,CAClE,eAAc,GACd,oBAAmB,GACnB,UAAS,GACT,wBAAuB,GACvB,8BAA6B,GAC7B,IAAG,GACH,IAAG,GACH,IAAG,GACH,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,OAAM,gBAGY,GAClB,GAAI,MAAO,IAAe,UAAY,IAAc,IAClD,MAAO,IAAW,GACb,GAAI,MAAO,IAAe,UAAY,GAAc,KACzD,MAAO,GAEP,KAAM,IAAI,GAAW,kBAAkB,eAqBP,GAElC,GADA,GAAY,IAAO,KAAM,0BAA0B,GAC/C,MAAO,IAAO,SAChB,MAAO,GAGP,OADI,GAAM,OACQ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAU,KAAS,GACrB,EAAS,EACT,OAGJ,GAAI,IAAW,OACb,MAAO,GAET,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAW,KAAS,GACtB,EAAS,EACT,OAGJ,MAAI,KAAW,OACN,EAED,EAAgB,KCzL5B,YAwB6B,GAC3B,GAAM,GAA2D,CAC/D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,MAC/B,SAAY,WAAM,MAAA,GAAA,MAAM,SAAS,EAAG,IAAM,OAC1C,KAAQ,WAAM,MAAA,GAAA,MAAM,KAAK,KAAO,GAAK,KAAO,OAC5C,OAAU,WAAM,MAAA,GAAA,MAAM,OAAO,KAAO,GAAK,KAAO,KAAW,IAC3D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,KAAO,GAAK,EAAG,OAC9C,IAAO,WAAM,MAAA,GAAA,MAAM,IAAI,OASzB,GAPA,EAAa,QAAa,EAAa,QACvC,EAAa,SAAc,EAAa,SACxC,EAAa,KAAU,EAAa,KACpC,EAAa,OAAY,EAAa,OACtC,EAAa,QAAa,EAAa,QACvC,EAAa,IAAS,EAAa,IAE/B,IAAc,GAChB,MAAO,GAAa,KAEtB,KAAM,IAAI,GAAW,qBAAqB,GC3C5C,AAeO,GAAM,IAA8C,EAAI,KAAO,iBAelE,EAAyB,EAAmB,GAC9C,GAD8C,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAuB,MACvB,MAAO,IAAwB,UAC/B,OAAO,eAAe,KAAyB,OAAO,WACtD,CAAC,GAAiB,GACpB,KAAM,IAAI,OACN,sEAGN,GAAI,GACF,GAAM,GAAM,KAAK,UAAU,GAC3B,AAAI,EAAI,OAAS,IACf,QAAQ,KACJ,mCAAmC,EAAS,qBAC5C,iBAAgB,EAAI,OAAM,iCAC1B,kHAEG,IAA2C,mBAgBvB,GAC/B,GAAI,IAAM,KAER,MAAO,GACF,GAAI,MAAO,IAAM,SACtB,GAAI,OAAO,eAAe,KAAO,OAAO,WAGtC,OADM,GAAO,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,MAAO,IAAQ,SAEjB,MAAO,GAET,GAAI,CAAC,GAAiB,EAAE,IACtB,MAAO,GAGX,MAAO,WAGH,MAAM,QAAQ,IAEhB,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,GAAiB,GACpB,MAAO,GAGX,MAAO,OAKP,OAAO,QAKX,GAAM,GAAQ,MAAO,GACrB,MAAO,KAAU,UAAY,IAAU,UAAY,IAAU,WCrGjE,YA8BI,EAAkB,EAAqB,EAEvC,GAAA,AAAA,IAAA,QAAA,GACI,QAAQ,KACd,GAAM,GAAiB,GAAsB,GAGvC,EAAsB,CAAC,eAAgB,eAAgB,WAC7D,AAAI,EACF,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAEtC,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAAM,IAI9C,AAAI,EAAU,EAAU,OAAS,IAAM,GAErC,GAAY,EAAU,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAa,MAGzD,GAAI,GACJ,GAAI,CAAC,GACH,EAAU,KAAK,mBACf,EAAgB,GAChB,OAAW,KAAS,GAAM,aACxB,EAAc,KAAI,MAAlB,EAAsB,EAAM,aAAa,IAI7C,EAAQ,IAAI,OAAO,IACnB,GAAS,EAAW,EAAW,GAC/B,EAAQ,IAAI,OAAO,IAGnB,OADM,GAAS,EAAM,OACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EACF,GAAkB,EAAO,GAAI,EAAW,GAExC,GACI,EAAO,GAAI,EAAW,EAAe,GAE3C,EAAS,KAAM,EAAO,OAAS,EAAI,IAAM,KAAK,OAAO,IAItD,EAAc,mCAEf,GAAM,GAAiB,GAAqB,GACtC,EAAoB,GAAqB,EAAM,qBAErD,EAAQ,iBAAiB,GAAiB,IAC1C,EAAQ,qBAAqB,GAC7B,EAAQ,yBAAyB,GACjC,EAAQ,IAAI,OAAO,IAGrB,YAA8B,GAC5B,GAAI,GAEJ,MAAK,GAAc,2BAA6B,KAC9C,EACI,GAAsB,EAAc,2BAExC,EAAiB,GAAqB,EAAM,kBAGvC,EAGT,YAA+B,GAC7B,GAAI,GAAiB,GACf,EAAyB,GACzB,EAAgB,GACtB,OAAW,KAAS,GAAM,aACxB,EAAa,KAAK,EAAM,aAAa,IAEvC,OAAyB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,EAAW,OAAS,GACpB,EAAW,SAAW,GAAK,EAAW,GAAG,cAAc,OAAS,GAClE,EAAiB,GACjB,MAEF,EAAM,KAAI,MAAV,EAAc,GAEhB,GAAI,EAEF,OAAoB,GAAA,EAAA,EAAA,EAAM,OAAN,EAAA,EAAA,OAAA,KAElB,OAFS,GAAK,EAAA,GACV,EAAO,GACQ,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAM,QAAQ,KAAU,GAC1B,GAAI,GACF,EAAiB,GACjB,UAEA,GAAO,GAIb,GAAI,CAAC,EACH,MAIN,MAAO,GAGT,YACI,EAAkB,EAElB,GAAA,AAAA,IAAA,QAAA,GAA6D,QAAQ,KAEvE,OADI,GAAO,GACF,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAI,GACN,GAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAAK,KAE1C,GAAQ,EAAO,GACf,EAAO,EAAK,MAAM,EAAG,EAAU,IAC/B,GAAQ,IAAI,OAAO,EAAU,GAAK,EAAK,QAEzC,EAAQ,GAQV,YACI,EAAc,EAEd,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAGhB,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EACF,CAAI,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,YAChE,GAAS,EAAQ,EAAW,GAM9B,YACI,EAAc,EAAqB,EAEnC,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAIhB,OADM,GAAwB,GACX,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,GAAiB,MAAQ,EAAc,OAAS,GAChD,EAAc,QAAQ,KAAU,GAClC,SAEF,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAAG,KACrC,EAAoB,EAAK,YAAY,GACrC,EAAqB,EAAK,cAAc,GAC9C,EAAY,KACL,EAAY,IAAI,EAAiB,KAAK,EAAkB,MAGnE,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EAAkB,EAAY,SAAW,EAAI,GAAK,EAAY,GAC9D,EAAmB,CACpB,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,WAC3D,GAGF,GAAS,EAAQ,EAAW,GAC5B,OAAS,GAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,EACxC,GAAS,CAAC,GAAI,GAAI,GAAI,EAAY,IAAK,EAAW,GCvNtD,AA8BA,YACI,EAAa,EAAe,GAC9B,MAAQ,KAAQ,gBAAkB,IAAQ,gBAClC,IAAQ,gBACZ,IAAU,GAAK,MAAO,IAAU,qBAUlC,EAA6B,GAC/B,GAAI,IAAmB,KACrB,MAAO,MACF,GAAI,MAAO,IAAmB,SACnC,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAmB,UAC1B,MAAO,IAAmB,UAC7B,MAAO,GACF,GAAI,YAA0B,QAGnC,OAFM,GAAU,GACV,EAAc,EAAe,OAC1B,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAe,GAC5B,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACf,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAArB,GAAM,GAAW,EAAA,GACd,EAAgB,EAAe,GACrC,GAAI,IAAgB,QAAU,MAAO,IAAkB,SAIrD,EAAO,GAAe,OAEtB,GAAM,GAAQ,GAA0B,GACxC,EAAO,GAAS,GAAoB,EAAe,IAGvD,MAAO,gBAWP,EAAyC,GAC3C,GAAI,GAAa,KACf,MAAO,MACF,GAAI,MAAO,IAAa,SAC7B,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAa,UAAc,MAAO,IAAa,UACzD,MAAO,GACF,GAAI,YAAoB,QAG7B,OAFM,GAAU,GACV,EAAc,EAAS,OACpB,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAS,GACtB,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACrB,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAU,EAAS,GACnB,EAAQ,GAA0B,GACxC,AAAK,KAAU,QAAU,IAAU,cAC/B,MAAO,IAAY,SAIrB,EAAO,GAAS,EAEhB,EAAO,GAAS,GAAoB,EAAS,GAGjD,MAAO,IC7HX,GAGM,IAAU,QCHhB,AA0BA,YAAiC,EAAqB,GAEpD,GAAI,EAAI,OAAS,MAAQ,EAAI,QAAU,EAAI,MAEzC,MAAO,GAET,IAEE,MAAO,GAAA,KAAK,EAAK,EAAI,aACd,GAEP,KAAM,IAAI,GACN,0BAA0B,EAAI,MAAK,kCACnC,gBAAe,EAAI,KAAI,MAAM,EAAI,MAAK,QAgB9C,GAAA,IAAA,WAUE,WAAY,GACV,GAVM,KAAA,SAAmC,GACnC,KAAA,QAAkC,GAClC,KAAA,QAAoC,GAQtC,YAAiB,GACnB,OAAW,KAAM,GAAM,SACrB,KAAK,SAAS,GAAM,EAAM,SAAS,GACnC,AAAI,IAAM,GAAM,SACd,MAAK,QAAQ,GAAM,EAAM,QAAQ,SAIrC,GAAI,GAAS,KACX,OAEF,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,KAAK,IAAI,EAAK,IAAK,EAAK,SAe9B,SAAA,UAAA,IAAA,SAAI,EAAqB,EAAe,GACtC,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAK,SAAS,EAAI,IAAM,GAAwB,EAAK,GACrD,KAAK,QAAQ,EAAI,MAAQ,EAAI,GAC7B,AAAI,GAAQ,MACV,MAAK,QAAQ,EAAI,IAAM,OAGzB,MAAM,IAAI,GAAW,uBAAuB,EAAI,KAAI,QAAQ,EAAI,IAElE,MAAO,OAQT,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,IAAI,EAAK,IAAK,EAAK,QAO1B,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,MAAK,SAAS,EAAI,KAAO,MAMlC,EAAA,UAAA,MAAA,WACE,MAAO,QAAO,KAAK,KAAK,UAU1B,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,SAAS,EAAI,SAG3B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,SAAS,KAWzB,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,QAAQ,EAAI,SAG1B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,QAAQ,KAKxB,EAAA,UAAA,aAAA,WACE,AAAI,KAAK,SAAW,MAClB,EAAA,QAAQ,KAAK,UAGnB,KAIM,GAA+D,GAG/D,GAC4D,eA4C9D,EAA0C,EAC1C,EAAiB,GAWnB,OATM,GAAoB,GAAU,KAAO,GAAQ,EAAO,SAEpD,EAAe,MAAM,QAAQ,GAC7B,EACF,EAAe,EAA8B,CAAC,GAE5C,EAAc,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,OACpC,EAAyB,GACzB,EAAY,EAAS,QACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,AAAI,EAAU,QAAQ,KAAgB,GACpC,EAAa,KAAK,EAAS,SAAS,IAEpC,EAAa,KAAK,MAItB,AAAI,GAAS,MAEX,GAAM,cAAgB,UACtB,EAAM,cAAgB,UAIxB,GAAM,GACF,EAAY,KAAK,KAAO,IAAM,EAAS,QAAQ,KAAK,KACpD,EACA,EACJ,GAAI,GAAa,IAAoB,MAGnC,GAAM,GAAM,GAAqC,EAAY,GAC7D,EAAS,EAAI,OACb,EAAkB,EAAI,gBAGtB,GAAa,GAAmB,EAChC,GAAsB,GAAmB,EAE3C,EAAS,GAAa,GACtB,EAAkB,GAClB,AAAK,GACH,OAAO,OAAO,EAAiB,GAAsB,IAMvD,OAHM,GAAmB,GAAI,IAAS,GAG7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAI,GAAS,MAEX,GAAM,GAAa,EAAA,SAAS,WAC5B,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAExB,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAI1B,GAAM,GAAW,EAAO,GAClB,EAAW,EAAS,YAC1B,GAAI,YAAoB,IACtB,SAOF,OALM,GAAwB,GACxB,EAAuB,GACvB,EAA6B,GAE/B,EAAa,GACG,EAAA,EAAA,EAAA,EAAS,OAAT,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAiB,SAAS,GAClC,EAAO,EAAiB,QAAQ,GACtC,EAAY,KAAK,GACjB,EAAW,KAAK,GAChB,AAAI,GAAQ,MACV,GAAa,IAEf,AAAK,GACH,GAAgB,EAAM,QACtB,AAAI,EAAgB,EAAM,QAAU,GAAK,CAAC,EAAS,OAAO,IACtD,EAAY,QAAQ,EAAM,QAAU,IAAM,CAAC,EAAM,YACjD,EAAM,YAAY,WAAa,IACjC,EAAiB,KAAK,IAK5B,AAAI,GACF,GAAS,GAAU,GACnB,EAAO,KAAU,EAAW,IAE9B,GAAM,GACF,GAAO,EAAS,MAAM,EAAa,IACnC,EAA8B,KAClC,AAAI,EAAS,iBACX,GAAa,EAAS,YAAY,EAAa,IAKjD,OAHM,GAAe,GAAe,GAC9B,EACF,MAAM,QAAQ,GAAgB,EAAe,CAAC,GACzC,EAAI,EAAG,EAAI,EAAsB,OAAQ,EAAE,GAClD,AAAK,EAAiB,OAAO,EAAsB,KACjD,EAAiB,IACb,EAAsB,GAAI,EAAc,GACxC,MAAM,QAAQ,GAAc,EAAW,GAAK,GAElD,GAAM,GAAQ,EAAY,QAAQ,EAAsB,GAAG,MAC3D,AAAI,IAAU,IACZ,GAAa,GAAS,EAAc,IAIxC,AAAK,GAEH,EAAA,QAAQ,GAQZ,SAAiB,eAEV,EAAe,EAAe,EAAa,GAsBpD,YACI,EAA2B,GAE7B,EAAA,KAAK,OACD,GAAW,MAAQ,EAAQ,OAAS,EACpC,WAAM,MAAA,0CAEV,GAAI,GAAgC,GAChC,EAAkC,GACtC,GAAI,EAAQ,SAAW,GAErB,GAAM,GACF,GAAgD,EAAQ,GAAI,GAChE,EAAc,EAAI,OAClB,EAAoB,EAAI,iBAGxB,QADM,GAAU,GAAI,KACA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAKlB,OALS,GAAK,EAAA,GACR,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,aAIc,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAxB,GAAM,GAAc,EAAA,GACvB,AAAK,EAAQ,IAAI,EAAe,OAC9B,GAAY,KAAK,GACjB,EAAQ,IAAI,EAAe,sBAKpB,GACT,AAAI,EAAkB,IAAS,MAC7B,GAAkB,GAAQ,GAAI,MAEhC,EAAa,GAAM,QACf,SAAA,GAAa,MAAA,GAAkB,GAAM,IAAI,MAL/C,OAAW,KAAQ,KAAR,GASf,MAAO,CACL,OAAQ,EACR,gBAAiB,GAAoB,IAIzC,YAA6B,GAC3B,GAAM,GAAmC,GACzC,OAAW,KAAQ,GACjB,EAAgB,GAAQ,EAAa,GAAM,KAE7C,MAAO,eAeL,EAAuB,GASzB,OAPM,GAAU,GAAI,KACd,EAA2B,GAC3B,EAA6B,GAKjB,EAAA,EAAA,EAAA,EAAS,QAAT,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAQ,IAAI,GAGd,GAAM,GAA0B,GAC1B,EAAkB,GAKxB,IAFA,EAAM,KAAK,GAEJ,EAAM,OAAS,IACpB,GAAM,GAAM,EAAM,EAAM,OAAS,GACjC,GAAI,EAAQ,IAAI,EAAI,OAClB,EAAM,MACN,SAEF,GAAM,GAAc,EAAM,EAAM,OAAS,KAAO,EAAM,OAAS,EAC/D,GAAI,EAAI,OAAO,SAAW,GAAK,EAE7B,EAAM,MACN,EAAO,KAAK,GACZ,EAAQ,IAAI,EAAI,MAChB,AAAI,GACF,EAAM,WAKR,EAAM,KAAK,EAAM,OAAS,GAC1B,OAAoB,GAAA,EAAA,EAAA,EAAI,OAAJ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GAQd,GALA,AAAI,EAAa,EAAM,OAAS,MAC9B,GAAa,EAAM,MAAQ,GAAI,MAEjC,EAAa,EAAM,MAAM,IAAI,EAAI,MAE7B,EAAQ,IAAI,EAAM,MACpB,SAEF,EAAM,KAAK,KAIjB,MAAO,CAAC,OAAM,EAAE,aAAY,GAS9B,YAAwB,GAEtB,GAAI,GACJ,GAAI,EAAM,YAAY,aAAa,SAAW,EAC5C,EAAe,EAAM,YAAY,YAGjC,OADI,GAAoB,KACf,EAAI,EAAG,EAAI,EAAM,YAAY,aAAa,OAAQ,EAAE,EAC3D,OAA2B,GAAA,EAAA,EAAA,EAAM,YAAY,aAAa,GAChD,cADiB,EAAA,EAAA,OAAA,KAAtB,GAAM,GAAY,EAAA,GAErB,GAAI,EAAa,KAAO,EAAM,IAC5B,EAAY,EACZ,OAIN,EAAe,EAAM,YAAY,YAAY,GAE/C,MAAO,GCjhBT,AA6CA,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAoCtC,WAAY,GAAZ,GAAA,GAEE,EAAA,KAAA,KAAM,KAAG,KAET,GAtBF,EAAA,eAAiB,GAAI,KAqBnB,EAAK,KAAO,EAAK,KACb,EAAK,MAAQ,MACf,GAAM,GAAS,EAAK,eAAe,cACnC,EAAK,KAAO,GAAO,GAqBrB,GAlBA,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAKlB,AAAI,MAAM,QAAQ,EAAK,QACrB,EAAK,OAAS,EAAK,OAAO,QAE1B,EAAK,OAAS,CAAC,EAAK,QAEtB,AAAI,MAAM,QAAQ,EAAK,SACrB,EAAK,QAAU,EAAK,QAAQ,QAE5B,EAAK,QAAU,CAAC,EAAK,SAInB,GAAqB,EAAK,QAAQ,SAAW,EAAK,OAAO,OAC3D,KAAM,IAAI,GACN,mGAEA,IAAG,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,SAIhC,AAAI,GAAqB,EAAK,SAAS,SAAW,EAAK,QAAQ,QAC7D,QAAQ,KACJ,qGAEA,IAAG,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,SAOjC,EAAK,YAAc,GACnB,EAAK,uBAAyB,GAC9B,EAAK,yBAA2B,GAKhC,EAAK,aAAe,GACpB,EAAK,wBAA0B,GAC/B,EAAK,0BAA4B,GAKjC,EAAK,OAAS,GAMd,EAAK,sBAAwB,GAe7B,OAAgB,GAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YACtB,EAAK,aAAa,KAAK,GACvB,EAAK,wBAAwB,KAAK,GAClC,EAAK,0BAA0B,KAAK,GAMtC,OAAgB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YAKtB,GAAqB,IAAc,EAAG,4BACtC,GAAqB,IAAgB,EAAG,8BACxC,EAAK,YAAY,KAAK,GACtB,EAAK,uBAAuB,KAAK,GACjC,EAAK,yBAAyB,KAAK,GAIrC,EAAK,WAAa,GAClB,EAAK,YAAc,GACnB,EAAK,gBAAkB,GACvB,EAAK,eAAiB,GACtB,EAAK,gBAAkB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,EAAK,YAAY,GAE/B,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,WACN,6DACA,qBAAoB,EAAK,OAAM,MAC/B,UAAS,EAAC,0BACV,oBAAmB,EAAM,eAAc,MAE7C,EAAK,WAAW,KAAK,EAAM,MAC3B,EAAK,gBAAgB,KAAK,EAAM,iBAEhC,EAAK,eAAe,KAAK,EAAM,MAEjC,OAAoB,GAAA,EAAA,EAAA,EAAK,aAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,YAAY,KAAK,EAAM,MAG9B,EAAK,oBAAsB,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QAClD,EAAK,qBAAuB,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,QAwFpD,OAjFM,GAA0C,GAE1C,EAAyC,GACzC,EAA4C,GAE5C,EAA6C,GAC7C,EAA4C,GAC5C,EAAiC,GAoBjC,EACF,SAAC,GAAwB,GAAuB,GAC/C,GAAe,GAAoB,IAClC,AAAI,KAAS,MAAQ,IAAa,MAAQ,IAAe,OACvD,IAAQ,GAAO,YACf,GAAY,GAAO,UACnB,GAAc,GAAO,aAEvB,GAAM,IAAO,GAAM,aAAa,IAGhC,GAAI,GAAgB,QAAQ,MAAU,GACpC,KAAM,IAAI,IACN,cAAc,GAAO,KAAI,cAAc,GAAM,KAAI,yBAKvD,GAAI,GAAc,QAAQ,MAAU,GAClC,OAIF,EAAK,eAAe,IAAI,EAAU,QAAQ,GAAO,KAGjD,AAAM,GAAM,KAAM,IAChB,GAAa,GAAM,IAAM,OAAO,KAAK,GAAc,QAGrD,AAAI,GAAgB,QAAQ,MAAU,IACpC,GAAgB,KAAK,IAKvB,OADM,IAAmB,GAAK,cAAc,OACnC,GAAI,EAAG,GAAI,GAAkB,MACpC,GAAM,IAAI,GAAK,aAAa,IACtB,GAAQ,GAAK,cAAc,IAC3B,GAAY,GAAK,YAAY,IAC7B,GAAc,GAAK,cAAc,IACvC,EACI,GAAG,GAAe,GAAiB,GAAO,GAC1C,IAGN,IADA,GAAc,KAAK,IACZ,GAAgB,QAAQ,KAAS,GACtC,GAAgB,OAAO,GAAgB,QAAQ,IAAO,GAExD,EAAuB,KAAK,KAG5B,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAgB,EAAG,EAAe,GAKpC,OAFM,GACF,EAAuB,QAAQ,UAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAa,EAAK,IAAM,EAExB,AAAM,EAAK,KAAM,IACf,GAAY,EAAK,IAAM,GAEzB,GAAI,GAAQ,EAAY,EAAK,IAGvB,EACD,EAAa,EAAK,cAAc,KAAO,KACnC,EACA,EAAa,EAAK,cAAc,IAOzC,EAAQ,KAAK,IAAI,EAAO,GACxB,EAAa,EAAK,cAAc,IAAM,EACtC,EAAe,EAAK,cAAc,IAAM,EAAK,cAC7C,EAAY,EAAK,IAAM,EAGvB,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAa,aAAa,GACxC,EACD,EAAY,EAAY,KAAO,KAAO,EACA,EAAY,EAAY,IACnE,EAAY,EAAY,IAAM,KAAK,IAAI,EAAQ,EAAG,GAClD,EAAa,EAAY,IAAM,GAKnC,GAAM,GAA0C,GAChD,OAAW,KAAU,IACnB,GAAM,GAAQ,EAAY,GAC1B,AAAM,IAAS,IACb,GAAa,GAAS,IAExB,EAAa,GAAO,KAAK,EAAa,IAIxC,GAAM,IAA4C,GAClD,OAAW,MAAW,IACpB,GAAM,GAAQ,EAAa,IAC3B,AAAM,IAAS,KACb,IAAc,GAAS,IAEzB,GAAc,GAAO,KAAK,EAAe,KAI3C,GAAI,IAAY,OAAO,KAAK,IACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAG1B,EAAK,OAAS,GACd,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACR,GAAiB,GAAc,GAGrC,GAAe,KAAK,SAAC,GAAG,IACtB,GAAM,IAAS,EAAa,GAAE,IACxB,GAAS,EAAa,GAAE,IAC9B,MAAI,IAAS,GACJ,GAEL,GAAS,GACJ,EAEF,IAET,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACd,AAAI,YAAiB,IACnB,EAAK,sBAAsB,KAAK,GAElC,EAAK,OAAO,KAAK,IAGrB,EAAK,cAAgB,GAGrB,GAAY,OAAO,KAAK,GACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAStB,OAJM,IAAoB,EAAK,OAAO,QAGhC,GAAoC,GACtB,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,KAClB,OADS,GAAK,GAAA,IACK,GAAA,EAAA,GAAA,EAAa,GAAb,GAAA,GAAA,OAAA,MAAd,GAAM,GAAI,GAAA,IACP,EAAQ,EAAK,cACnB,GAAI,GAAS,MACX,OAAgB,IAAA,EAAA,GAAA,EAAK,aAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAI,GAAkB,QAAQ,KAAO,GACnC,KAAM,IAAI,IACN,sDAAsD,EACtD,eAAc,EAAM,KAAI,OACxB,uDACA,WAAU,KAGlB,OAAgB,IAAA,EAAA,GAAA,EAAK,cAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAkB,KAAK,GAEzB,GAAwB,KAAK,EAAM,OAMzC,EAAK,aAAe,EAKpB,OADM,IAAW,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,mBAC7B,IACT,GAAM,IAAiB,GAAS,OAAO,SAAA,IAAK,MAAA,MAAM,KAAM,OACxD,GAAI,KAAmB,EACrB,KAAM,IAAI,IACN,aAAa,GAAI,aAAa,GAAc,uEAE5C,KAAK,UAAU,MANJ,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAd,GAAM,IAAI,GAAA,OAAJ,IAcX,SAAK,cAAgB,GAErB,EAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,EAAK,OACnB,cAAe,EAAK,QACpB,WAAY,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,QACjC,YAAa,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,QACnC,YAAa,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QACpC,aAAc,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,UAExC,EAAK,MAAQ,GACb,EAAK,UAAY,IAGT,SAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,cAAc,KAAK,KAAI,2BA8B3C,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,GAAM,GACc,CAAC,qBAAsB,KAAM,qBAAsB,GACvE,GAAI,EAAE,KAAK,YAAc,GACvB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAO,sBAAwB,EAAM,UAAU,qBAKjD,OAAwB,GAAA,EAAA,EAAA,KAAK,sBAAL,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAO,sBAAwB,EAAU,UAAU,sBAGvD,SAAO,qBAAuB,KAAK,UAC5B,GAGT,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,OAAO,QAAQ,SAAA,GAEhB,EAAc,kBACX,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,MAElC,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WAIE,GAAI,KAAK,kBAAkB,OAAS,EAClC,KAAM,IAAI,GACN,wNAMN,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADI,GAA2B,GACX,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAU,EAAQ,OAAO,EAAM,kBAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACb,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAQ,KAAI,MAAZ,EAAgB,EAAM,qBAExB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACtB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAiB,KAAI,MAArB,EAAyB,EAAM,kBAEjC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAkB3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,IAGnC,OAFM,GAAgD,GAClD,EAAoB,EACJ,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACO,EAAA,EAAA,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,EAAa,EAAO,eAAiB,KACvC,KAAM,IAAI,GAAW,0BAA0B,EAAO,cAExD,EAAa,EAAO,cAAgB,EACpC,IAIJ,GAAM,GAAoD,GAC1D,OAAW,KAAQ,IAIjB,GAAI,GAAgB,EACpB,GAAI,EAAa,IAAS,MACxB,GAAM,GAAS,EAAK,MAAM,KACpB,EACF,EAAO,MAAM,EAAG,IAAI,OAAO,CAAC,EAAO,EAAO,OAAS,KACvD,EAAgB,EAAiB,KAAK,KAExC,GAAI,EAAa,IAAkB,KACjC,EAAkB,KAAK,CAAC,EAAa,GAAgB,EAAQ,aACpD,EACT,KAAM,IAAI,GACN,gDAAgD,GAEtD,MAAO,GAAa,GAGtB,GAAI,GAEF,GAAM,GAAuB,GAC7B,OAAW,KAAQ,GACjB,EAAW,KAAK,GAElB,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,GACH,EAAW,OAAM,OAChB,EAAiB,yBACrB,IAAG,IAIX,GAAc,IAON,EAAA,UAAA,cAAV,WACE,GAAM,GAAY,KAAK,YACjB,EAAwC,GAC9C,SAAY,UAAe,KAAK,eAChC,EAAY,OAAY,EACxB,EAAY,aAAkB,eAAe,GAG7C,EAAY,QAAa,gBAClB,GAeT,EAAA,UAAA,OAAA,SAAO,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IACnB,GAAM,GAAc,GAAoB,KAAK,iBAC7C,MAAO,GAAe,KAAK,UAAU,GAAe,GAgBtD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAE9B,OADM,GAAW,GAAI,IACZ,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,EAAK,OAAO,GAAI,EAAO,IAEtC,MAAO,IAAQ,EAAK,QAAS,EAAU,MAa3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAC9B,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAA2B,KAAM,EAAO,QAEhD,EAAQ,GAAqB,GAGxB,EAAK,iBAAiB,EAAQ,GAAO,MAahD,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAM,GAAc,GAA+B,GACnD,GAAI,EAAY,SAAW,KAAK,YAAY,OAC1C,KAAM,IAAI,GACN,+BAA+B,EAAU,KACzC,cAAa,KAAK,YAAY,OAAM,oBAK1C,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,EAAY,OAAQ,KACtC,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAa,EAAY,GAGzB,EAAW,EAAM,KAAO,OAC9B,EAAqB,GAAY,EAGnC,GAAM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,GAAK,MAAA,UAAS,EAAG,MACrB,KAAK,IAE5B,GAAI,EAAU,OAAS,EACrB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GAEP,EAAQ,EAAK,cACnB,GAAI,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,QAAQ,EAAM,MAAQ,GAExD,SAIF,OADM,GAAuB,GACpB,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAc,EAAa,KAAI,IAAI,EAAS,IAAI,EAChD,EAAa,EAAqB,GACxC,EAAY,KAAK,GAQnB,OALM,GAAc,EAAM,mBACtB,GAA+B,IAE7B,EAAe,GAA+B,GAC9C,EAAY,EAAM,aAAa,QAAQ,GACpC,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAqB,GAAY,EAAa,IAStD,OAFM,GAAwB,GACxB,EAA4B,GACzB,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GACzC,EAAc,KAAK,0BAA0B,GAC7C,EAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAgB,KAAK,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAgB,OAAQ,KAC1C,GAAM,GAAM,EAAgB,GAC5B,GAAqB,IAAO,IAC5B,EAAa,KAAK,EAAqB,IAIzC,MAAO,IAA+B,IAa9B,EAAA,UAAA,iBAAV,SAA2B,EAAkB,GAE3C,AAAI,GAAS,MACX,GAAQ,GAA2B,KAAM,EAAO,SASlD,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,GACxC,GAAM,GAAI,KAAK,OAAO,GAChB,EAAI,EAAO,GACX,EAAO,EAAM,GACnB,EAAU,EAAE,IAAM,CAAC,EAAG,GAMxB,OAHM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAUjB,OAVS,GAAI,EAAA,GAEP,EAAQ,EAAK,cACb,EAAwB,EAAK,aAC7B,EAAyB,EAAK,cAK9B,EAAe,GAAI,OACT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAE,KAAM,IACV,EAAa,KAAK,EAAU,EAAE,KAGlC,GAAI,EAAa,SAAW,EAAsB,QAEhD,GAAI,GAAiB,GACjB,EAAe,OACf,EAAa,OACb,EAAa,OACb,EAAW,OAKf,GAHA,AAAI,EAAK,UAAY,MACnB,GAAS,EAAK,UAEZ,EAAa,SAAW,GACpB,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GACvB,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAgB,IACpD,EAAc,GACV,EAAM,YAAY,EAAgB,IACtC,EAAkB,CAAC,GACnB,EAAgB,CAAC,OAEjB,GAAkB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KAC1C,EAAgB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KACxC,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAiB,IACrD,EAAc,GACV,EAAM,YAAY,EAAiB,IAGzC,GAAI,EAAM,oBACR,KAAM,IAAI,IACN,yHAMN,OAAS,GAAI,EAAG,EAAI,EAAuB,OAAQ,EAAE,GACnD,GAAM,GAAI,EAAuB,GAC3B,EAAI,EAAc,GAClB,EAAO,EAAY,GACzB,EAAU,EAAE,IAAM,CAAC,EAAG,KAS9B,OAHM,GAA0B,GAC1B,EAAwB,GACxB,EAAwB,GACd,EAAA,EAAA,EAAA,KAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,GACI,EAAE,KAAM,GAAW,4BAA4B,EAAE,KAAI,MAAM,EAAE,IAC3D,GAAA,GAAA,EAAA,EAAA,IAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAa,KAAK,EAAO,OACzB,EAAc,KAAK,GACnB,EAAY,KAAK,GAInB,MAAO,CAAC,EAAe,EAAa,IAW9B,EAAA,UAAA,uBAAR,SAA+B,GAG7B,OAFM,GAAiD,GACnD,EACgB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,YAAiB,GAAY,EAAI,EAC7C,OAAS,GAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,KAAK,eAAe,IAAI,IAE1B,GAAkB,GAAW,EAC7B,GAAa,IAInB,MAAO,IAsBT,EAAA,UAAA,SAAA,SAAS,EAAe,GACtB,GAAI,GAAS,MACX,GAAI,KAAK,OAAO,QAAU,EACxB,KAAM,IAAI,GACN,wCAAwC,EAAK,oBAC7C,QAAO,KAAK,OAAO,OAAM,eAE7B,MAAO,MAAK,OAAO,WAGjB,GAAQ,KACV,KAAM,IAAI,GAAW,8CAIzB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,OAAS,EACjB,MAAO,GAGX,KAAM,IAAI,GAAW,kBAAkB,IAQzC,EAAA,UAAA,gBAAA,WAAA,GAAA,GAAA,KAKE,MAAO,GAAA,KAAK,WAEV,OADM,GAAmB,GACL,EAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACL,EAAY,EAAG,EAAY,EAAM,aAAa,OAClD,EAAE,GACL,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,EAAK,eAAe,IAAI,IAC1B,EAAO,KAAI,MAAX,EAAe,EAAM,mBAK3B,MAAO,MAIX,EAAA,UAAA,UAAA,WAWE,OAVM,GAAmC,CAAC,KAAM,KAAK,MAK/C,EACF,KAAK,uBAAuB,KAAK,QAG/B,EAAe,GACD,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EAAiB,EAAM,eACvB,EAAc,EAAM,YACpB,EAAuB,GACpB,EAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAO,EAAM,aAAa,GAC1B,EAAU,EAAU,QAAQ,EAAO,GACrC,EAAS,GACb,GAAI,KAAK,eAAe,IAAI,IAG1B,GAAI,EAAK,SACP,IACE,KAAK,UAAU,EAAK,UACpB,EAAS,EAAK,eACP,GACP,QAAQ,KACJ,SAAS,EAAM,KAAI,mDAEhB,GAAK,SAAQ,gCAChB,+EAEJ,EAAS,GAGb,GAAI,EAAK,cAAc,OAAS,GAE9B,OADM,GAAW,GACR,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAU,EAAU,QAAQ,EAAc,GAC5C,EAAe,EAAkB,GACrC,AAAI,GAAgB,MAClB,GAAe,GAEjB,EAAS,KACL,CAAC,EAAa,KAAM,EAAc,EAAa,IAErD,EAAqB,KAAK,KAIhC,GAAM,GAAiC,GACvC,EAAK,KAAU,EAAM,KACrB,EAAK,UAAe,EACpB,EAAK,OAAY,EACjB,EAAK,aAAkB,EACvB,EAAa,KAAK,GAEpB,EAAO,OAAY,EAGnB,OADM,GAAc,GACX,EAAI,EAAG,EAAI,KAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAY,KAAK,uBAAuB,GAExC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,yBAAyB,GAClD,EAAY,KAAK,CAAC,EAAM,KAAM,EAAc,IAE9C,EAAO,YAAiB,EAGxB,OADM,GAAe,GACZ,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GAEzC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,0BAA0B,GACnD,EAAa,KAAK,CAAC,EAAM,KAAM,EAAc,IAE/C,SAAO,aAAkB,EAClB,GAgBF,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAA8C,GAO9C,EAAkE,GACxE,WACI,EAAc,GAChB,AAAM,EAAM,OAAQ,GAGlB,EAAiB,EAAM,MAAM,KAAK,GAFlC,EAAiB,EAAM,MAAQ,CAAC,GAMpC,WAAqB,EAAc,GAGjC,OAFM,GAAiC,GACnC,GACoB,GAAA,EAAA,GAAA,EAAA,GAAA,GAAA,OAAA,MAAnB,GAAM,IAAS,GAAA,IACZ,GAAmB,GAAU,GAC7B,GAAmB,GAAU,GAC7B,GAAqB,GAAU,GAKrC,GAHA,GAAS,GAAU,IAAM,KACrB,GACA,GAAU,GACV,CAAE,MAAoB,KACxB,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAe,EAAc,IACnC,GAAI,GAAa,aAAa,QAAU,IACtC,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAc,GAAa,aAAa,IAC9C,EAAa,KAAK,GAAY,cAAc,KAK9C,AAAI,EAAa,OAAS,GACxB,EAAM,MACF,GAA+B,GAC/B,IAUR,WAAsB,GACpB,GAAM,GAAY,EAAU,KAEtB,EACF,GACI,EACA,EAAO,eAAoB,KACvB,EAAO,cACP,IACZ,EAAM,6BAA6B,GACnC,EAAc,GAAa,EAE3B,GAAM,IACF,EAAU,aACd,GAAiB,QAAQ,SAAA,IACvB,GAAI,CAAE,cAAoB,QACxB,KAAM,IAAI,GACN,yDACI,IAMV,EAAmB,EAAO,MAO9B,OAFM,GAAO,EAAO,KACd,EAAmB,EAAO,OACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAa,GAOf,KAAO,CAAC,GAA4B,IAClC,OAAwB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAQ,EAAc,EAAU,MACtC,GAAI,EAAM,OAAQ,IAChB,GAAM,GAAkC,EAAiB,EAAM,MAC/D,MAAO,GAAiB,EAAM,MAC9B,OAAuB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAY,EAAO,KAU3B,OAJM,GAAiC,GACjC,EAAkC,GAClC,EACF,EAAO,YACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAa,KAAK,EAAmB,IAIvC,OAFM,GACF,EAAO,aACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAc,KAAK,EAAmB,IAExC,MAAO,IAAI,GAAI,CAAC,OAAQ,EAAc,QAAS,EAAe,KAAI,KASpE,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WAGE,GAAI,KAAK,UACP,KAAM,IAAI,GACN,wLAIN,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,SACR,MAAO,GAGX,MAAO,oCAST,EAAA,UAAA,YAAA,WAAA,GAAA,GAAA,KACE,EAAA,KAAK,WACH,EAAK,OAAO,QAAQ,SAAA,GAElB,AAAI,EAAM,UACR,EAAM,mBAMhB,GA3vCwC,IC7CxC,AAiCA,YACI,EAAmD,EACnD,GACF,GAAM,GAAa,EAAY,OAC/B,GAAI,GAAW,MAAS,MAAM,QAAQ,IAAY,EAAQ,SAAW,EACnE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,QAEjC,GAAI,IAAe,EACjB,MAAI,OAAM,QAAQ,IAAY,EAAQ,SAAW,EACxC,EACE,MAAO,IAAY,UAAY,EAAY,IAAM,GACnD,CAAE,EAA2B,EAAY,KAEzC,CAAC,GAGZ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAQ,SAAW,EACrB,KAAM,IAAI,OACN,YAAY,EAAU,mBAAmB,EAAQ,OAAM,IACvD,kCAAiC,EAAU,cAC3C,iEAEN,MAAO,WAEL,MAAO,IAAY,UAAY,OAAO,KAAK,GAAS,OAAS,GAC7D,MAAQ,GAA2B,OAAO,KAAK,GAAS,KACpD,UACN,GAAM,GAAwB,GAC9B,SAAY,QAAQ,SAAA,GAClB,AAAI,IAAc,GAChB,EAAO,KAAM,EAA2B,IAExC,EAAO,KAAK,QAGT,MAEP,MAAM,IAAI,OACN,2BAA2B,EAAU,cACrC,OAAM,EAAU,kCACb,GAAU,+BAA+B,EAAW,WACvD,aAAY,EAAU,oBAAoB,KAAK,UAAU,iBAkB7D,EACA,GACF,MAAO,IACH,EAAa,EAAa,2BA6B5B,EAAW,EAAuB,EAClC,2GACF,GAAI,GAAgB,MAAQ,GAAoB,KAG9C,KAAM,IAAI,OAAM,qDAGd,IAAe,KAEX,GAAqB,EAAA,KAAK,WAC9B,GAAI,EAAE,MAAM,SAAW,EAErB,MAAO,GAAE,QACJ,GAAI,EAAE,MAAM,SAAW,EAC5B,GAAI,EAAE,MAAM,GAAK,GAEf,GAAM,GAAO,EACb,MAAO,GAAE,OAAO,QACX,GAAI,EAAE,MAAM,KAAO,EAExB,MAAO,GAAE,QAAQ,CAAC,EAAE,MAAM,KAE1B,KAAM,IAAI,OACN,+CAA+C,EAAE,MAAM,GAAE,4EAK/D,MAAM,IAAI,OACN,yCAAyC,EAAE,KAAI,4EAKjC,EAAA,GAAA,OAAM,KAAK,CAAA,EAAM,EAAS,SA3B9C,CAAA,EAAA,UA2BI,SAAgB,EAAA,MAAA,EAAA,CAAW,EAAA,SACjC,EAAA,QAAQ,GACF,EAA8B,GACpC,EAAc,QAAQ,SAAA,GACpB,GAAI,EAAY,IAAe,KAC7B,KAAM,IAAI,OACN,8DACA,cAAa,EAAU,mCACvB,eAEJ,EAAkB,KAAK,EAAY,MAIvC,CAAA,EAAO,EAAA,SAAS,EAAmB,mBAEnC,MAAA,CAAA,EAAO,uBAWyB,EAAgB,GAClD,MAAO,GAAA,IAAI,EAAQ,GC5LrB,AAyLA,GAAM,IAAgC,GAgBtC,YAII,EAAY,GACd,GAAI,GACA,EAEE,EAAiB,EACvB,EAAK,EAAe,GACpB,EAAK,EAAe,GACpB,EAAA,KAAS,OACL,GAAM,MAAQ,GAAM,KACpB,WAAM,MAAA,+OAIF,IAAG,KAEX,GAAM,GACF,GAA0B,QAAS,EAAM,WAAY,GACnD,EACF,GAA0B,SAAU,EAAM,YAAa,GAErD,EAAoB,EAAY,GAAG,MAAM,GAE/C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,OAAO,OACpC,WAAM,MAAA,mBAAmB,EAAM,OAAO,OAAM,4BACxC,aAAY,EAAY,OAAM,oCAC3B,MAAK,UAAU,EAAM,YAAW,OAE3C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,QAAQ,OACrC,WACI,MAAA,mBAAmB,EAAM,QAAQ,OAAM,6BACvC,aAAY,EAAY,OAAM,sCAC3B,MAAK,UAAU,EAAM,aAAY,OAE5C,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,8BACC,GAAM,WAAW,GAAO,QACrB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,+BACC,GAAM,YAAY,GAAO,QACtB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,MAAO,CAAC,GAAI,EAAa,GAAI,GAG/B,YACI,EAAuB,EAAiB,GAC1C,GAAI,YAAkB,GAAA,OACpB,MAAO,CAAC,GACH,GAAI,MAAM,QAAQ,GACvB,SAAA,KAAS,OACL,EAAO,SAAW,EAAM,OACxB,WAAM,MAAA,wBAAwB,EAAO,OAAM,0BACvC,EAAM,OAAM,iBAAiB,EAAa,SAAS,EAAK,MACzD,EAIP,OAFM,GAAuB,GAEV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,IAAS,KAClB,KAAM,IAAI,GACN,gEACG,GAAa,SAAS,EAAI,OAEnC,EAAO,KAAK,EAAO,IAErB,MAAO,GAIX,YACI,GAMF,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,IACN,0DAEN,MAAO,CAAC,GAAI,EAAK,GAAI,GAAI,EAAK,gBAO5B,EAAY,EACZ,yKA0BF,GAzBM,EAAqB,EAAK,iBAAmB,KACnD,EAAA,KAAS,OACL,EAAM,WAAa,KACnB,WAAM,MAAA,mGAGV,EAAA,KAAS,OACL,GAAQ,KACR,WAAM,MAAA,kGAEV,EAAA,KAAS,OACL,EAAK,QAAU,MAAQ,EAAK,OAAS,GAAK,OAAO,UAAU,EAAK,QAChE,WAAM,MAAA,gEACF,qBAAoB,EAAK,UACjC,EAAA,KAAS,OACL,CAAC,GACI,EAAK,gBAAkB,GAAK,OAAO,UAAU,EAAK,iBACvD,WAAM,MAAA,gEACF,2CAA0C,EAAK,mBACvD,EAAA,KAAS,OAEJ,EAAa,iBAAsB,KACpC,WAAM,MAAA,sFAGN,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,mDAGX,EAAe,EAAK,gBAAkB,KACxC,EAAK,OACL,EAAK,OACT,AAAI,GACF,CAAI,GAAgB,EAAK,gBACvB,EAAA,KAAS,OACL,EAAK,mBAAqB,MACrB,EAAK,kBAAoB,GACzB,OAAO,UAAU,EAAK,mBAC3B,WAAM,MAAA,yIAGF,YAAW,EAAK,qBAElB,GAAiB,GACnB,EAAK,gBAMT,EAAQ,EAAe,GACvB,EAAQ,EAAe,KAIrB,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAe,OACnB,AAAI,EACF,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,IAAK,MAAA,OAAS,MAEzD,EAAkB,EAAU,QAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YACtD,EAAU,EAAK,SAAW,KAAO,EAAI,EAAK,QAC1C,EAA0B,GAC5B,EAAW,EAAS,EAAK,OAAQ,KAAM,KACvC,GAAiB,EAAS,GAC1B,KACA,EAAc,GAJX,EAAY,EAAA,aAAE,EAAA,EAAA,QAKrB,EAAa,SAAS,GACtB,EAAM,QAAU,EAEhB,CAAA,EAAM,EAAa,uBAAnB,SAAA,OACA,EAAM,cAAgB,GAClB,EAAQ,EAAK,cAAgB,KAAO,EAAI,EAAK,aAE9B,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BACZ,GAAQ,EAAK,OACZ,GAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,KAFR,CAAA,EAAA,iBAExB,GAAA,OACI,EAAY,EACZ,EAAa,EACZ,EAAD,CAAA,EAAA,GACa,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BAEV,GAAqB,EAAY,EAAK,gBAAkB,IACzC,CAAA,EAAM,EAAa,QAD0B,CAAA,EAAA,WAKjE,MAJM,GAAc,EAAA,OAIhB,GAAsB,EAAY,KACpC,SAAQ,KACJ,qCACG,GAAK,gBAAe,MACvB,mDACG,GAAS,cACZ,8HAGG,GAAK,gBAAkB,EAAK,OAAM,eACrC,yEAEJ,CAAA,EAAA,KAGE,EAAY,OAAS,KACjB,GACF,GAA8B,EAAO,EAAY,OAD9C,EAAE,EAAA,GAAE,EAAE,EAAA,GAEP,EAA4B,GAClC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAG,GAAG,MAAM,GAEhC,CAAA,EAAM,EAAa,aAAa,EAAY,KAP1C,CAAA,EAAA,cAOF,EAAA,OAEM,EAA8B,IAChC,GAAK,aAAe,MAApB,MAAA,CAAA,EAAA,IACI,EACF,GAAwB,EAAK,YAAa,EAAM,aAC3C,EAAI,2BAAG,GAAI,EAAqB,OACvC,GAAA,GAAA,GAAc,KAAK,CAAA,EAAM,GACrB,EAAG,GAAI,KAAM,EAAqB,MAFO,CAAA,EAAA,YAC7C,EAAA,MAAA,EAAA,CAAmB,EAAA,4BAD4B,QAAE,iBAUrD,IAHM,EAAM,EAAG,OAAO,GAAI,OAAO,GAC3B,EAAO,EAAc,GAC3B,EAAA,QAAY,GACH,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EAChC,EAAQ,EAAU,GAClB,EAAM,EAAK,GACjB,EAAU,GAAS,EACnB,EAAA,KAAS,GAGX,MAAA,CAAA,EAAM,EAAa,WAAW,EAAY,YAA1C,EAAA,OACA,GAAqB,GAErB,IACA,6BAGE,GAAqB,GAAa,EAAK,gBAClB,EAAY,MAE/B,EACE,GAAO,OACP,GAAgB,EAAK,gBACb,GAAA,GAAO,CAAA,EAAM,EAAM,gBACzB,EAAK,eAAgB,CAAC,QAAS,EAAK,sBAFtC,CAAA,EAAA,KAFF,CAAA,EAAA,IAHF,CAAA,EAAA,YAME,SAAU,EAAA,MAAA,OAAA,CAAO,EAAA,wBAGjB,EAAU,GAAO,EAAM,SAAS,EAAO,EAAO,CAC5C,UAAW,EAAK,qBAAuB,KACnC,GACA,EAAK,oBACT,QAAS,wBAGb,IAAS,EAAI,EAAG,EAAI,EAAM,aAAa,OAAQ,EAAE,EAC/C,EAAU,OAAO,EAAM,aAAa,IAAQ,EAAQ,sBAQxD,MAAA,CAAA,EAAA,YAGF,MAAI,GAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,YAErC,MAFA,GAAA,OACA,IACI,EAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,sBAAnB,SAAA,OACA,CAAA,EAAM,EAAM,QAAQ,oBAApB,SAAA,OACA,CAAA,EAAO,EAAM,iBAEb,SAAM,WAAa,8BAKvB,YACI,EAAqB,GAEvB,GAAI,GAAwB,KAC5B,MAAI,GAAK,iBAAmB,KAC1B,EAAgB,EAAK,gBAChB,AAAI,OAAO,SAAS,EAAQ,OACjC,GAAgB,EAAQ,MAEnB,EAKT,YACI,GAKF,MAAQ,OAAQ,GAAuB,UAAa,WAKtD,YAAiC,GAE/B,MAAQ,OAAQ,GAA6B,MAAS,uBAOpD,EAAY,EACZ,uHAKF,GAJA,EAAO,GAAQ,GACT,EAAa,EAAK,SAAW,KAC7B,EAAI,EAAM,aACZ,EAAqB,GACrB,EAAK,QAAU,EACjB,KAAM,IAAI,IAAoB,8CAGhC,GAAA,KAAS,OACL,CAAC,GAAe,EAAK,QAAU,GAAK,OAAO,UAAU,EAAK,SAC1D,WAAM,MAAA,6DACF,aAAY,KAAK,UAAU,EAAK,YACnB,GAAqB,GACtC,GAAA,SADiB,CAAA,EAAA,UAEjB,MAAA,CAAA,EAAO,EAAuB,mBAA9B,EAAA,EAAA,wBAFE,EAAY,EAId,EAAc,EACd,EAAQ,uEAGU,MAAA,CAAA,EAAM,EAAa,eAmCvC,MAnCM,GAAc,EAAA,OACpB,EAAO,EAAA,KAAS,WACd,GAAI,EAAY,OAGR,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAU,EAAG,OAAO,GACpB,EAAY,EAAA,KAAS,WAAM,MAAA,GAAE,KAGnC,GAFA,EAAA,QAAY,GAER,IAAU,EACZ,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAKrB,OADM,GAAY,EAAQ,GAAG,MAAM,cAC1B,GACP,GAAM,GAAW,EAAU,GACrB,EAAY,EAAK,GACvB,EAAK,GACD,EAAA,KAAS,WAAM,MAAA,GAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,MACvD,AAAI,EAAQ,GACV,EAAA,QAAY,IANP,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,IAA/B,GAST,EAAA,QAAY,GACZ,GAAe,EAEf,EAAE,EAEJ,MAAO,KAGL,EAAY,KACd,CAAI,GACF,QAAQ,KACJ,uJAGA,2BAA0B,EAAK,QAAO,eACtC,qHA3CH,GAAa,EAAQ,EAAK,QAAU,YAAI,CAAA,EAAA,2DAkD/C,IAAS,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC3B,EAAY,EAAK,GACvB,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAC3B,EAAA,QAAY,GAGd,MAAA,CAAA,EAAO,GAAiB,SCrmB1B,YAiK+B,GAC7B,EAAA,KAAS,OACL,EAAY,GAAK,OAAO,UAAU,GAClC,WAAM,MAAA,2DACF,gBAiBN,EAAyB,EAAe,GAC1C,MAAI,IAAU,KACL,CAAC,MACC,MAAM,QAAQ,GAChB,EAAO,IAAI,SAAA,GAAS,MAAA,IAAoB,EAAO,EAAO,EAAO,KAE7D,GAAoB,EAAQ,EAAO,EAAO,eAkBjD,EAAyB,GAC3B,MAAO,GAAA,KAAS,WACd,MAAI,IAAU,KACL,KACE,MAAM,QAAQ,GAChB,EAAO,IACV,SAAA,GAAS,MAAC,IAAqB,EAAO,KAInC,GACH,EAAQ,EAAQ,QAAU,QAAU,EAAU,EAAQ,uBAc5D,EAAc,GAIhB,OAHM,GAAkC,GACpC,EAAa,EACb,EAAmB,KAChB,EAAa,GAClB,EAAW,EAAa,EACxB,AAAI,GAAY,GACd,GAAW,GAEb,EAAO,KAAK,CAAC,EAAY,IACzB,EAAa,EAEf,MAAO,GA8BT,YAGI,EAAY,EAAiC,EAC7C,EAAsB,EAAoB,EAAiB,EAC3D,EAA4B,EAC5B,EAAmB,EAA0B,EAC7C,EAAuB,EACvB,mHAoBF,GAnBA,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,GAAU,MACZ,GAAS,GAEX,AAAI,GAAW,MACb,GAAU,IAEZ,AAAI,GAAgB,MAClB,GAAe,GAIb,EAAe,GACnB,AAAI,GAAQ,MAAQ,GAAU,MAC5B,GAAe,IAGb,GAAmB,MACrB,GAAe,GACX,GAAiB,MACnB,KAAM,IAAI,GACN,oGAKF,SACF,EAAM,gBAAgB,EAAK,EAAW,EAAe,mBAEzD,AAAI,GAAmB,MACrB,GAAa,GAAM,EAAG,IAGxB,AAAI,GAAW,MACb,GAAU,GAGN,EAA0B,GAC5B,EAAW,EAAS,EAAQ,EAAc,EAAiB,EAC3D,EAAW,EAAc,GAFtB,EAAY,EAAA,aAAE,EAAO,EAAA,QAG5B,EAAa,SAAS,GACtB,EAAM,QAAU,EAChB,CAAA,EAAM,EAAa,uBAAnB,EAAA,OACA,EAAM,cAAgB,cAIb,qEACP,MAAA,CAAA,EAAM,EAAa,aAAa,cAAhC,EAAA,OACM,EAA4B,IAC9B,IAAiB,MAAjB,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,IACN,qDAEJ,GAAI,IAAY,QACd,KAAM,IAAI,IAAoB,0CACzB,AAAI,GACT,EAAA,KAAK,QAAQ,GAIT,EAAoB,EAAA,SAAS,GAE7B,EAAU,GAAY,EAAiB,cACpC,2DACD,SAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,EAAY,WAA5C,SAAA,OAEA,EAAA,KAAS,WACP,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,GAAW,GACI,EAAmB,EACnB,EAAW,GAChC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAW,EAM/B,OAFM,IAAW,GAAqB,EAAK,IACrC,GAAO,EAAE,IACN,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAK,IACjB,EAAU,IAAS,GACnB,EAAA,KAAS,IAIX,GAAI,IAAe,EAAQ,OAAS,GAC9B,EAGF,OAFM,IAAU,EAAM,SAAS,EAAM,EAAQ,GAEpC,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAQ,IACpB,EAAA,KAAS,IAET,EAAU,OAAS,IAAS,MAMpC,CAAA,EAAM,EAAa,WAAW,EAAY,WAG1C,MAHA,GAAA,OACA,GAAqB,GAEjB,EAAM,kCA3CH,EAAa,yBAAG,GAAa,EAAQ,YAArC,IAA2C,CAAA,EAAA,iEAAE,QAAE,eAiDxD,EAAkB,2BAGpB,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,WACrC,MADA,GAAA,OACI,EAAM,kCAtEH,EAAQ,yBAAc,GAAQ,OAA9B,IAAoC,CAAA,EAAA,iEAAE,QAAE,eA0EjD,MAAA,CAAA,EAAM,EAAa,qBAAnB,SAAA,OAEA,CAAA,EAAM,EAAM,QAAQ,mBAApB,SAAA,OACA,CAAA,EAAO,EAAM,0BAMX,EAAY,EACZ,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,yIACF,GAAI,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,iDASX,EAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIT,EAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAG,EAAG,EAAK,aAAc,EAAK,YAAa,EAC3C,cAHF,EACF,EAAA,OAGJ,EAAS,EAAiB,GAC1B,EAAU,EAAiB,GAC3B,EAAgB,EAAiB,GAG7B,EAAe,GACf,EAAM,QACN,GAAK,gBAAkB,MAAQ,EAAK,eAAe,OAAS,GAA5D,MAAA,CAAA,EAAA,GAEF,GADA,EAAe,GACX,EAAK,eAAe,SAAW,EAEjC,EAAY,EAAK,eAAe,GAChC,EAAY,EAAK,eAAe,OAC3B,MAAI,GAAK,eAAe,SAAW,EAClC,GAAI,IACN,iEAEE,GAAI,GACN,0GAEG,GAAK,eAAc,iBAGtB,SAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAW,EAAW,KACtB,KACA,EAAgB,WAJlB,SACF,EAAA,OAIJ,EAAO,EAAgB,GACvB,EAAO,EAAgB,GACvB,EAAS,EAAK,OAAO,gBAEhB,AACH,EAAK,iBAAmB,MAAQ,EAAK,gBAAkB,GACvD,EAAK,gBAAkB,EACzB,GAAe,GAET,EACF,KAAK,MAAM,EAAO,GAAG,MAAM,GAAM,GAAI,EAAK,kBACxC,EAAoB,EAAO,GAAG,MAAM,GAC1C,EAAO,GAAY,EAAQ,EAAS,GACpC,EAAS,GAAY,EAAQ,EAAG,GAChC,EAAO,GAAY,EAAS,EAAS,GACrC,EAAU,GAAY,EAAS,EAAG,GAGlC,EAAS,EAAK,OAAO,IAGhB,AAAI,EAAK,iBAAmB,MACjC,GAAe,qBAIX,SAAM,EAAO,OAAO,GAAS,OAAO,GAE1C,EAAM,mCAcA,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAW,OACX,EAAe,OACnB,AAAI,EACF,GAAM,mBACN,EAAc,EAAM,aACpB,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,MAEzD,GAAc,KACd,EAAS,GACT,EAAkB,EAAU,SAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YAChD,CAAA,EAAM,GACd,EAAO,EAAe,EAAK,EAAW,EAAW,EAAK,OACtD,EAAK,QAAS,EAAW,EAAa,EAAQ,EAAK,QACnD,EAAiB,EAAK,aAAc,KAAM,cAHxC,SAAM,EAAA,OAIZ,CAAA,EAAO,UAEP,SAAM,WAAa,GAEnB,GAAkB,EAAQ,GAC1B,GAAkB,EAAS,GAC3B,GAAkB,EAAkB,GACpC,GAAkB,EAAkB,GACpC,AAAI,GAAiB,MACnB,EAAA,QAAY,yCAYyB,GACzC,GAAM,GAAiB,GACvB,AAAI,YAAmB,GAAA,QACrB,GAAU,CAAC,IAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAS,EAAQ,GACvB,GAAI,EAAO,OAAS,EAClB,EAAK,KAAK,GAAW,EAAQ,SACxB,GAAI,EAAO,OAAS,EACzB,KAAM,IAAI,OACN,yEAGJ,EAAK,KAAK,IAGd,MAAO,eAgBL,EACA,GACF,GAAI,GAAW,KACb,OAEF,GAAM,GAAyB,GAC/B,GAAI,YAAsB,GAAA,OACxB,EAAa,KAAK,EAAW,YACpB,MAAM,QAAQ,GACvB,EAAW,QAAQ,SAAA,GAAK,MAAA,GAAa,KAAK,EAAE,cACnC,GAAc,KAEvB,OAAW,KAAQ,IACjB,GAAM,GAAY,EAAW,GAC7B,EAAa,KAAK,EAAU,IAIhC,GAAM,GAA6B,GACnC,GAAI,YAAmB,GAAA,OACrB,AAAI,EAAa,QAAQ,EAAQ,MAAQ,IACvC,EAAiB,KAAK,WAEf,MAAM,QAAQ,GACvB,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAa,QAAQ,EAAE,MAAQ,IACjC,EAAiB,KAAK,aAGjB,GAAW,KAEpB,OAAW,KAAQ,IACjB,GAAM,GAAS,EAAQ,GACvB,AAAI,EAAa,QAAQ,EAAO,MAAQ,IACtC,EAAiB,KAAK,GAK5B,EAAiB,QAAQ,SAAA,GACvB,AAAK,EAAE,YACL,EAAE,YCrnBR,YA+C6B,GAE3B,MAAO,aAAa,GAAA,mBAMM,GAE1B,MAAO,OAAM,QAAQ,eAMI,GAEzB,MAAO,CAAC,GAAa,IAAM,CAAC,GAAY,eAetC,EAAqD,EACrD,EAAkB,EAAuB,GAC3C,GADoB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACvC,GAAS,MAAQ,EAAM,SAAW,GAGpC,GAAI,GAAQ,MACV,GAAI,GAAoB,GACxB,GAAI,GAAY,IAAU,EAAkB,OAAS,EACnD,EAAoB,WACX,GAAW,IACpB,OAAW,KAAO,GAChB,GAAI,EAAK,eAAe,IACtB,EAAoB,GACpB,WAKJ,GAAoB,GAEtB,GAAI,EACF,KAAM,IAAI,GACN,6BAA6B,EAAe,sBAC5C,YAAW,IAGnB,MAAO,GAET,GAAI,GAAQ,KACV,MAAO,GAAM,IAAI,SAAA,GAAQ,MAAA,QAG3B,GAAI,GACJ,GAAI,GAAW,IACb,EAAO,EACP,EAAS,GACT,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAK,IAAS,KAChB,KAAM,IAAI,GACN,yBAAyB,EAAI,iCAC7B,IAAG,IAET,EAAO,KAAK,EAAK,aAEV,GAAY,IAErB,GADA,EAAO,EACH,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,oCAAmC,EAAM,OAAM,oBAC/C,iDAAgD,IAEtD,EAAS,OAGT,GADA,EAAO,EACH,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,aAAa,EAAe,YAAY,EAAM,OAAM,eACpD,2DACI,EAAK,QAEf,EAAS,CAAC,GAMZ,GAHA,EAAS,GAA2B,GAGhC,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,EAAM,QAErB,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EAEd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MAAQ,GAAU,GAAK,IAAQ,EAC3C,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,mBAAkB,EAAO,GAAE,gCAC3B,KAAI,EAAM,MAAK,QAK3B,MAAO,eAWL,EAAkB,EAAmB,GACvC,GAAM,GAAO,GAAO,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,MAAM,MACpD,EAAK,OACL,GAAM,GAAO,GAAO,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,MAAM,MAGvD,GAFA,EAAK,OAED,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,mFAEA,IAAG,KAAK,UAAU,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,WAElD,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,oFAEA,IAAG,KAAK,UAAU,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,WAErD,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,CAAC,EAAA,KAAK,YAAY,EAAM,GAChE,KAAM,IAAI,GACN,kEACA,mBAAkB,EAAK,GAAE,wBAAwB,EAAK,GAAE,YACxD,cAaR,YACI,EAAmB,EAA2B,GAMhD,OAJM,GAAY,CAChB,GAAyB,GACzB,IAEO,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAI,EAAQ,GACZ,EAAO,EAAQ,GACf,EAAQ,EAAa,GAC3B,GAAI,GAAQ,KACV,SAEF,GAAI,IAAS,IACP,EAAE,MAAM,EAAE,MAAM,OAAS,KAAO,EAClC,KAAM,IAAI,GACN,2CAA2C,EAAE,MAAK,4JAO1D,GAAI,EAAU,QAAQ,KAAU,GAG9B,OAFM,GAAe,EAAE,MAAM,MAAM,GAC7B,EAAc,EAAM,MAAM,GACvB,EAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAY,EAAa,GACzB,EAAS,EAAY,GAC3B,GAAI,GAAU,MAAQ,IAAc,EAClC,KAAM,IAAI,GACN,8BAA8B,EAAE,MAAK,sBACrC,oBAAmB,EAAK,uCACxB,2DAiCd,YACI,EAAuB,EAAiB,EACxC,EAAuB,GAAvB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACzB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,wCAAuC,EAAM,OAAM,eACnD,qBAAoB,EAAK,OAAM,iBAErC,EAAS,OAET,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,qBAAqB,EAAM,OAAM,IAAI,EAAe,mEAEjD,MAAK,UAAU,EAAK,OAAM,MAEnC,EAAS,CAAC,GAGZ,GAAI,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,KAAK,UAAU,EAAM,SAEpC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EACd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MACR,IAAW,EACb,KAAM,IAAI,GACN,uBAAuB,EAAe,cACnC,GAAM,GAAE,kBAAkB,KAAK,UAAU,EAAO,IAAG,SACtD,yBAAwB,KAAK,UAAU,EAAM,OAAM,oBAsB/D,EAEA,GACF,GAAI,GAAW,MAAQ,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAClE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,KAGjC,GAAI,GAEJ,GAAI,MAAO,IAAY,UAAY,MAAO,IAAY,WACpD,EAAiB,CAAC,WACT,MAAM,QAAQ,IAAY,MAAO,IAAY,SACtD,EAAiB,MAGjB,MAAM,IAAI,WACN,+DACA,uCAAsC,IAG5C,GAAI,MAAM,QAAQ,GAEhB,MAAO,GAAY,IACf,SAAA,GAAQ,MAAA,KAIZ,OADM,GAAqD,GACxC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACT,EACA,EAAe,eAAe,GAAQ,EAAe,GAAQ,GACjE,AAAK,MAAM,QAAQ,IACjB,GAAgB,CAAC,IAEnB,EAAc,KAAK,GAErB,MAAO,GA6DX,GAAM,IAA2B,8BAcA,EAAA,EAAA,GA4C/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WAAa,KAsCpB,SAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAGF,GAHE,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC1D,CAAC,KAAK,MACR,KAAM,IAAI,GACN,gLAIN,GAAa,KAAM,EAAY,EAAW,IAa5C,EAAA,UAAA,QAAA,SAAQ,GAAR,GAAA,GAAA,KAME,GALA,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAEd,KAAK,KAAO,EAAK,KAEb,MAAO,GAAK,WAAc,SAC5B,KAAK,WAAa,GAAwB,EAAK,WAC/C,KAAK,iBAAmB,QAExB,GAAI,CAAE,GAAK,oBAAqB,GAAA,WAC9B,KAAM,IAAI,GACN,+DAEN,KAAK,WAAa,EAAK,UACvB,KAAK,iBAAmB,GAO1B,GAAI,GAAkC,GACtC,GAAI,CAAC,MAAM,QAAQ,EAAK,OAAS,MAAO,GAAK,MAAS,UAClD,MAAO,GAAK,MAAS,YACvB,EAAK,KAAO,EAAK,KACjB,OAAW,KAAQ,GAAK,KACtB,GAAI,KAAK,YAAY,QAAQ,KAAU,GACrC,KAAM,IAAI,GACN,sCAAsC,EAAI,MAC1C,sCAAqC,KAAK,cAGlD,OAAmB,GAAA,EAAA,EAAA,KAAK,YAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAK,KAAK,IAAS,MACrB,QAAQ,KACJ,WAAW,EAAI,4GAEf,oBAAmB,EAAI,qBAE7B,EAAc,KAAK,GAAW,EAAK,KAAK,cAEjC,MAAM,QAAQ,EAAK,OAC5B,GAAI,EAAK,KAAK,SAAW,KAAK,QAAQ,OACpC,KAAM,IAAI,GACN,+DACA,gCAA+B,KAAK,QAAQ,OAAM,gBAClD,wBAAuB,EAAK,KAAI,MAEtC,GAAM,GAAY,EAAK,KACvB,EAAgB,EAAU,IAAI,SAAA,GAAK,MAAA,IAAW,UAE9C,GAAM,GAAe,GAAW,EAAK,MACrC,KAAK,QAAQ,QAAQ,SAAA,GACnB,EAAc,KAAK,KAIvB,KAAK,cAAgB,EAErB,KAAK,gBAAkB,GACvB,KAAK,iBAAmB,GACxB,KAAK,YAAc,GACnB,OAAS,GAAI,EAAG,EAAI,KAAK,QAAQ,OAAQ,EAAE,GAEzC,GAAM,GAAQ,KAAK,qBAAqB,GAClC,EAAO,KAAK,YAAY,GAC9B,KAAK,gBAAgB,KAAK,GAC1B,KAAK,iBAAiB,KAAK,GAC3B,KAAK,YAAY,KAAK,KAAK,cAAc,IAK3C,GAAM,GAA8B,GAGpC,KAAK,QAAU,EAAK,QAEpB,KAAK,aAAe,CAAC,QACrB,KAAK,eAAiB,GAMtB,GAAU,OAAQ,WAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,GACzC,GAAI,EAAkB,QAAQ,KAAO,GACnC,SAIF,GAAM,GAAe,EAAK,cAAc,GACxC,AAAI,EAAK,QAAQ,OAAS,GACxB,GAAK,eAAe,KAAK,CAAC,EAAc,IACxC,EAAK,aAAa,KAAK,EAAK,YAAY,GAAK,aAQnD,GAAM,GAAgB,GAAe,EAAK,QAAS,KAAK,aAMlD,EACF,SAAC,EAAqB,EACrB,GACC,AAAI,EAAK,YAAY,OAAS,GAC5B,GAAa,EAAK,YAAY,GAAe,IAAM,GAErD,EAAK,aAAa,KAAK,GACvB,EAAK,eAAe,KAAK,CAAC,EAAc,KAG9C,GAAU,SAAU,WAClB,mBAAS,GACP,GAAI,EAAkB,QAAQ,KAAO,oBAGrC,GAAM,GAAgB,EAAc,GAI9B,EAAgB,SAAC,GAOrB,OANM,GAAmB,GACrB,EACA,EACA,aAGO,GACT,GAAI,MAAO,IAAW,UAClB,CAAC,WAAY,MAAO,eAAgB,MAAM,QAAQ,KAC9C,IACN,GAAM,GAAc,EAAK,qBAAqB,GAE9C,AAAI,EAAY,EAAY,OAAS,KAAO,GACxC,EAAK,cAAc,KAAO,GAE5B,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAEL,AACH,EAAK,cAAc,KACnB,GAGF,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAIV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAGZ,GAAI,GAAM,OACV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAS,MACJ,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAS,MAGX,EAAmB,EACnB,EAAa,EAAmB,OAEhC,GAAM,GAAW,GAAY,GAE7B,EAAmB,EACnB,EACI,EAAmB,GAA4B,GAIrD,GAAI,GACJ,GAAU,EAAY,WACpB,EAAe,IAEjB,EAAa,EAAG,EAAY,IAtDT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,KAAN,KA0Db,EAAc,IAzEP,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,IAAlC,KAgFX,KAAK,0BAA4B,KAAK,kBAY9B,EAAA,UAAA,iCAAV,WACE,GAAI,KAAK,2BAA6B,KACpC,OAEF,AAAI,KAAK,iBAAiB,SACtB,KAAK,0BAA0B,QACjC,QAAQ,KACJ,sJAqCR,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIf,GAAM,GAAiB,GACjB,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GACrD,IAGE,GAAM,GAAM,EAAiB,GAAG,OAAO,EAAiB,IACxD,KAAK,mBACL,GAAM,GAAI,KAAK,aACT,EACF,KAAK,SAAS,EAAG,EAAK,EAAW,EAAK,QAAS,EAAK,OACxD,MAAO,IAAiB,WAExB,GAAkB,EAAiB,GAAI,GACvC,GAAkB,EAAiB,GAAI,KA0BrC,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,YAAK,mBACL,CAAA,EAAO,GAAgB,KAAM,EAAS,SAahC,EAAA,UAAA,gBAAR,SACI,EAAsB,EAAoB,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,SACF,GAAI,GACJ,GAAI,GAAS,MAEX,GADA,EAAa,KACT,GAAa,KACf,KAAM,IAAI,GACN,MAAM,EAAS,gDACf,oBAAmB,YAEhB,GAAO,KAChB,AAAI,MAAM,QAAQ,GAChB,EAAa,EAAI,GAAG,MAAM,GAE1B,EAAa,EAAI,MAAM,OAGzB,MAAM,IAAI,GACN,yDACG,GAAS,yBAElB,MAAO,IAUT,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,GAAI,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAC/C,KAAM,IAAI,GACN,sDAGN,GAAM,GAAiB,MAAM,QAAQ,GAC/B,EACD,EAAiB,EAAsB,CAAC,GACvC,EAAwB,KAAK,wBAAwB,GAGrD,EAAW,GAAI,IAIrB,GAHA,AAAI,YAAkB,GAAA,QACpB,GAAS,CAAC,IAER,MAAM,QAAQ,IAChB,GAAI,EAAO,SAAW,KAAK,OAAO,OAChC,KAAM,IAAI,GACN,kCAAkC,EAAO,OAAM,uDAE/C,KAAI,KAAK,OAAO,OAAM,OAE5B,OAAS,GAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,KAAK,OAAO,GAAI,EAAO,QAGtC,QAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAc,EAAO,EAAM,MACjC,GAAI,GAAe,KACjB,KAAM,IAAI,GACN,8CAA8C,EAAM,MAE1D,EAAS,IAAI,EAAO,GAKxB,GAAM,GAAiB,GAAQ,EAAuB,GACtD,MAAO,GAAiB,EAAiB,EAAe,IAMlD,EAAA,UAAA,wBAAR,SAAgC,GAK9B,OAHM,GACF,GAAa,KAAM,EAAoB,QACvC,EAAmB,EAAoB,OACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EACF,MAAM,QAAQ,EAAM,QAAU,EAAM,OAAS,CAAC,EAAM,QAClD,EAAmB,EAAa,IAAI,SAAA,GAAU,MAAA,GAAO,OAClD,EAAI,EAAG,EAAI,EAAoB,OAAQ,EAAE,GAChD,GAAM,GAAQ,EAAiB,QAAQ,EAAoB,IAK3D,GAJA,AAAI,IAAU,IACZ,GAAsB,GAAK,EAAa,GACxC,KAEE,IAAqB,EACvB,MAGJ,GAAI,IAAqB,EACvB,MAIJ,GAAI,EAAmB,GACrB,GAAM,GAA2B,GACjC,QAAsB,QAAQ,SAAC,EAAQ,GACrC,AAAI,GAAU,MACZ,EAAe,KAAK,EAAoB,MAGtC,GAAI,GACN,mDACA,IAAG,KAAK,UAAU,KAExB,MAAO,IAgBD,EAAA,UAAA,YAAR,SAAoB,EAAsB,EAAgB,GAA1D,GAAA,GAAA,KAA0C,MAAA,KAAA,QAAA,GAAA,IAAgB,AAAA,IAAA,QAAA,GAAA,IAEjD,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,GACxC,GAAI,EACF,KAAM,IAAI,IACN,iDAYN,OAJM,GAAU,GAAY,EAAY,GAClC,EAA0B,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,gBAGlD,GACP,GAAM,GAAY,EAAA,KAAS,WACzB,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAG/B,EAAW,GAAY,EAAK,EAAY,GAGxC,EAAQ,GACd,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,EAAE,EACrC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAS,SAGnD,GAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,IAE1C,GAAM,GAAW,GAAI,IAAS,GAC9B,MAAO,IAAQ,EAAK,QAAS,KAE/B,EAAU,QAAQ,SAAC,EAAU,GAAM,MAAA,GAAY,GAAG,KAAK,MApBhD,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,IAA/C,GAsBT,MAAO,IACH,EAAY,IAAI,SAAA,GAAW,MAAA,GAAA,OAAW,EAAS,SA+BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAkB,GAA2B,GACnD,GACI,EAAiB,KAAK,WAAY,KAAK,gBAAiB,IAC5D,IAKE,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,UAAe,GACR,KAAK,YAAY,EAAiB,WAEzC,GAAkB,EAAiB,KAmBvC,EAAA,UAAA,eAAA,SAAe,GACb,GAAe,EAAG,KAAK,WAAY,KAAK,gBAAiB,IAGzD,GAAM,GAAa,OAAM,QAAQ,GAAK,EAAE,GAAK,GAAG,MAAM,GACtD,MAAO,MAAK,YAAY,EAAG,IAGnB,EAAA,UAAA,sBAAV,SACI,EACA,EAAkD,EAClD,GAEF,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,IACN,gGAIN,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,KAAK,iBAAiB,OAAQ,EAAE,GAClD,GAAM,GAAc,KAAK,iBAAiB,GACpC,EAAS,KAAK,YAAY,GAChC,AAAI,IAAW,GACb,EAAa,KACT,EAAY,MAAM,EAAG,EAAY,OAAS,GAAG,OAAO,CAAC,KAGzD,EAAa,KAAK,GAWtB,GARA,EAAI,GACA,EAAG,KAAK,eAAgB,KAAK,gBAAiB,GAAO,SACzD,EAAI,GACA,EAAG,KAAK,gBAAiB,EAAc,GAAO,UAElD,GAAkB,EAAG,GAErB,GAAgC,EAAG,KAAK,YAAa,KAAK,kBACtD,KAAK,UAAY,GAAa,MAAQ,EAAY,GAChD,EAAE,GAAG,MAAM,GAAK,IAAc,EAChC,KAAM,IAAI,GACN,mHAEG,GAAS,YAAY,EAAE,GAAG,MAAM,GAAE,gBAG7C,MAAO,CAAC,EAAG,IAGG,EAAA,UAAA,oBAAhB,SACI,EACA,EACA,EACA,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAA,2GAKF,GAHM,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GAD9C,EAAU,EAAA,GAAE,EAAU,EAAA,GAGzB,GAAgB,KAClB,KAAM,IAAI,OAAM,0CAGd,EAAkC,MAClC,IAAe,MAAf,MAAA,CAAA,EAAA,GACI,EACF,GAAwB,EAAa,KAAK,aAC9C,EAAwB,GACf,EAAI,yBAAG,GAAI,EAAa,OAC/B,GAAA,GAAA,GAAsB,KAClB,CAAA,EAAM,GAAmB,EAAW,GAAI,KAAM,EAAa,MAF1B,CAAA,EAAA,UACrC,EAAA,MAAA,EAAA,CACI,EAAA,0BAFmC,QAAE,eAO7C,MAAA,CAAA,EAAO,CAAC,EAAY,EAAY,UAc1B,EAAA,UAAA,SAAR,SACI,EAAiC,EAAe,EAChD,EAAa,GAFjB,GAAA,GAAA,KAEI,MAAA,KAAA,QAAA,GAAA,GACK,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,EAAK,EAAW,EAAO,SACzD,EAAiB,GACvB,GAAI,EAAU,EACZ,KAAM,IAAI,IAAoB,wCAGhC,GAAI,GAAS,KACX,KAAM,IAAI,IACN,mDAIJ,OAFM,GAAU,GAAY,EAAY,GAClC,EAAa,EAAA,SAAS,GAAM,EAAG,IAC5B,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,GACtD,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,EACF,GACI,EAAY,EAAY,EAAW,GAGrC,EAAW,GAAqB,EAAK,GACrC,EAAY,EAAE,GACpB,GAAI,IAAe,EACjB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAGrB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAW,EAAU,GAC3B,EAAK,GACD,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,EAAY,KAGxD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAG/B,MAAO,MAID,EAAA,UAAA,uBAAV,WAKE,OAJM,GAAY,KAAK,aAGjB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAQ,EAAU,GACpB,EAAW,EACf,GAAI,GAAM,EAAW,GAAS,GAC5B,GAAM,GAAW,GAAM,EAAU,MAAM,EAAG,GAAI,GAC9C,GAAY,IAAI,EAElB,EAAiB,KAAK,GAExB,MAAO,IAaC,EAAA,UAAA,kBAAV,WAAA,GAAA,GAAA,KACE,MAAO,UAAC,GACN,GAAM,GAAuB,GAEvB,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAgB,EAAK,MACvB,EAAK,OAAO,OAAS,EAAK,QAAQ,OAClC,EAAK,OAAO,OAAS,EAAK,QAAQ,OAAS,GAEzC,EAA0B,GAK1B,EAAoB,WAExB,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KASjD,OAPM,GAAW,GAAI,IAAS,GACxB,EACF,GAAQ,EAAK,QAAS,EAAU,CAAC,SAAY,KAI7C,EACK,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GACpC,EAAO,EAAa,EAAQ,GAAI,EAAQ,IAC5C,AAAI,EAAc,IAAM,MACtB,GAAO,GAAoB,EAAM,EAAc,KAIjD,GAAM,GAAmB,EAAA,KAAS,GAElC,EAAW,KAAK,GAChB,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAOnC,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAI,GAAc,OAElB,GAAI,EAAK,QAAQ,OAAS,GAAK,EAAI,EAAK,QAAQ,OAC9C,EAAiB,EAAW,QAE5B,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAC3C,EACI,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAGpD,EAAA,KAAS,GAET,EAAc,KAAK,GAGrB,SAAY,EAAA,KAAS,GAGrB,EAAK,kBAAkB,QAAQ,SAAA,GAC7B,EAAY,EAAA,IAAQ,EAAW,KAG1B,GAGH,EAAY,EAAK,0BAA0B,IAC7C,SAAA,GAAS,MAAA,GAAM,SACb,EAAa,GACb,EACF,EAAK,WAAW,SAAS,EAAmB,EAAY,GAE5D,MAAO,CAAC,GAAgB,OAAO,KAS3B,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,KAAK,aAAe,SAAC,GACnB,MAAO,GAAA,KAAS,WAOd,OANM,GAAuB,GACzB,EACE,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KAKjD,OAHM,GAAW,GAAI,IAAS,GACxB,EAAU,GAAQ,EAAK,QAAS,GAE7B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAGlC,EAAe,EAAA,KAAS,EAAa,EAAQ,GAAI,EAAQ,KAC/D,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAEjC,EAAW,KAAK,GAGlB,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAErC,EACF,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAClD,EAAW,KAAK,GAElB,MAAO,OAuCP,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,MAAA,CAAA,EAAO,GAAW,KAAM,EAAG,EAAG,SA0B1B,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,MAAA,CAAA,EAAO,GAAW,KAAM,EAAS,SA0B7B,EAAA,UAAA,aAAN,SACI,EACA,qHAIqB,MAAA,CAAA,EAAM,KAAK,oBAAoB,EAAG,WAAnD,EAAiB,EAAA,OACjB,EAAS,EAAe,GACxB,EAAU,EAAe,GACzB,EAAgB,KAAK,oBACrB,EAAS,EAAc,EAAO,OAAO,IACrC,EAAuB,OACV,EAAA,yBAAA,GAAA,EAAA,OAAR,GAAI,EAAA,GACH,CAAA,EAAM,EAAK,SADE,CAAA,EAAA,UACjB,EAAI,EAAA,OACV,EAAW,KAAK,EAAE,qBAFD,wBAInB,SAAA,QAAY,GACZ,CAAA,EAAO,GAAiB,UAYhB,EAAA,UAAA,gBAAV,SAA0B,GAMxB,OALM,GAA8B,GAE9B,EAAgB,GAAU,MAAQ,EAAO,cACzC,EAAU,EAAgB,KAAK,iBAAmB,KAAK,QACvD,EAAe,KAAK,WAAW,GAC5B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAI,GAAiB,CAAC,EAAQ,GAAG,UAE/B,SAEF,EAAa,KACT,CAAC,KAAM,EAAQ,GAAG,aAAc,OAAQ,EAAa,KAE3D,MAAO,IAiCT,OAAA,eAAI,EAAA,UAAA,eAAY,KAIhB,WACE,MAAO,MAAK,mBALd,SAAiB,GACf,KAAK,cAAgB,mCAOvB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,AAAI,KAAK,aAAe,GACtB,MAAK,WAAa,EAClB,KAAK,iBAAmB,qCAI5B,EAAA,UAAA,QAAA,WACE,GAAM,GAAS,EAAA,UAAM,QAAO,KAAA,MAC5B,GAAI,EAAO,uBAAyB,GAAK,KAAK,WAAa,MACvD,KAAK,kBACP,GAAM,GAAmC,EAAA,SAAa,WACtD,KAAK,WAAW,UAChB,EAAO,sBACH,EAAmC,EAAA,SAAa,WAEtD,MAAO,IAGD,EAAA,UAAA,mBAAR,WAEE,GAAI,GAEJ,GAAI,MAAO,MAAK,MAAS,SACvB,EAAY,GAAY,KAAK,cACpB,MAAM,QAAQ,KAAK,OAC5B,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,OAAM,sDAGpB,EAAa,KAAK,KAAkB,IAAI,SAAA,GAAQ,MAAA,IAAY,UAG5D,GAAM,GAAc,OAAO,KAAK,KAAK,MACrC,EAAY,GAGZ,OAFM,GACF,KAAK,KACgB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,MAAO,GAAO,IAAgB,SAChC,EAAU,GACN,GAAY,EAAO,QAEvB,MAAM,IAAI,OAAM,uDAItB,MAAO,IAGD,EAAA,UAAA,qBAAR,WAEE,GAAI,MAAO,MAAK,SAAY,UACxB,MAAO,MAAK,SAAY,WAC1B,MAAO,CAAC,GAAY,GAA4B,KAAK,WAChD,GAAI,MAAM,QAAQ,KAAK,SAC5B,MAAO,MAAK,QAAQ,IAChB,SAAA,GAAU,MAAA,IAAY,GAA4B,MAEtD,GAAM,GAAyD,GAC/D,OAAW,KAAO,MAAK,QACrB,EAAmB,GACf,GAAY,GAA4B,KAAK,QAAQ,KAE3D,MAAO,IAID,EAAA,UAAA,kBAAV,WACE,MAAO,CACL,KAAM,KAAK,qBACX,QAAS,KAAK,uBACd,iBAAkB,CAChB,WAAY,KAAK,UAAU,eAC3B,OAAQ,KAAK,UAAU,eAQ7B,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,EAAe,kBAAoB,KACrC,KAAM,IAAI,OAAM,gDAElB,GAAI,EAAe,cAAgB,KACjC,KAAM,IAAI,OAAM,8CAElB,GAAI,EAAe,oBAAsB,KACvC,KAAM,IAAI,OAAM,oDAGlB,GAAM,GAAW,GAAoB,EAAe,kBAE9C,EAAY,GAAY,GAE1B,EACJ,GAAI,MAAO,GAAe,MAAS,SACjC,EAAO,GAAY,EAAe,cACzB,MAAM,QAAQ,EAAe,MACtC,EAAO,EAAe,KAAK,IAAI,SAAA,GAAa,MAAA,IAAY,aAC/C,EAAe,MAAQ,MAChC,EAAO,GACP,OAAW,KAAO,GAAe,KAC/B,EAAK,GAAO,GAAY,EAAe,KAAK,IAIhD,GAAI,GACJ,GAAI,MAAM,QAAQ,EAAe,SAC/B,EAAU,EAAe,QAAQ,IAAI,SAAA,GAAU,MAAA,IAAY,aAClD,EAAe,SAAW,MACnC,EAAU,GACV,OAAW,KAAO,GAAe,QAC/B,EAAQ,GAAO,GAAY,EAAe,QAAQ,IAItD,KAAK,QAAQ,CAAC,KAAI,EAAE,QAAO,EAAE,UAAS,KAoFlC,EAAA,UAAA,KAAN,SAAW,EAAmC,+HAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,GACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,GACN,gHAKF,MAAA,CAAA,EAAM,EAAA,GAAG,cAAc,KAAK,gBAAgB,kBAD1C,GACF,EAAA,OAEE,EAAe,GACf,EAAgB,KAChB,EAAc,KAAK,OAAO,EAAW,GACrC,EAAoC,CACxC,cAAe,EACf,OAAQ,GACR,YAAa,8BAA8B,GAC3C,YAAa,MAGT,EAAmB,GAAU,KAAO,GAAQ,EAAO,iBACrD,GAAoB,KAAK,WAAa,KACxC,GAAe,eAAiB,KAAK,oBAC/B,EAAa,YAET,EAAA,GAAA,EAAA,IAAG,cAAc,CAAA,EAAM,KAAK,UAAU,eAJ9C,CAAA,EAAA,UAIE,MAAA,CAAA,EAAM,EAAA,MAAA,EAAA,CAAiB,EAAA,OAAmC,YADxD,EACF,EAAA,OADS,EAAmB,EAAA,KAAS,EAAoB,EAAA,MAE7D,GAAA,EAAmB,OAAM,KAAI,MAAA,EAAI,GACjC,EAAmB,KAAO,EAAA,GAAG,wBACzB,CAAC,EAAmB,KAAM,qBAGhC,MAAI,MAAK,qBAAuB,MAExB,GAAY,GAClB,GAAyB,KAAK,oBAAqB,KAAK,KAAM,GAC9D,EAAe,oBAAsB,KAAK,qBAG5C,EAAe,WAAa,EAAmB,KAC/C,EAAe,YAAc,EAAmB,MAChD,CAAA,EAAO,EAAa,KAAK,UAW3B,EAAA,UAAA,uBAAA,SAAuB,GACrB,GAAyB,EAAqB,KAAK,MACnD,KAAK,oBAAsB,GAc7B,EAAA,UAAA,uBAAA,WACE,MAAO,MAAK,qBAv4CP,EAAA,UAAY,QAy4CrB,GA74CiC,IA84CjC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DACS,SAAA,UAAY,aACrB,GAFgC,IAGhC,EAAA,cAAc,cAAc,IC72D5B,YA8DI,EACA,uHACF,AAAM,iBAAmB,IACvB,GAAwB,CAAC,cAAe,IAE1C,EAAwB,EAEpB,EAAgB,EAAsB,cAC1C,AAAI,EAAc,cAAmB,MAMnC,GAAgB,EAAc,cAE1B,EACF,GAAoB,GAClB,EAAQ,GAAY,EAAU,GAEhC,EAAsB,iBAAmB,KAItB,CAAA,EAAM,EAAA,GAAG,YAC1B,EAAsB,gBAAiB,EAAsB,WAC7D,EAAM,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,iBANrC,CAAA,EAAA,UAUF,IANM,EAAe,EAAA,OAKf,EAAqC,GAC3C,EAAA,EAAqB,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,IAAV,EAAM,EAAA,GACf,EAAmB,EAAO,cACtB,EAAa,EAAO,cAG1B,EAAM,YAAY,GAElB,EAAA,QAAQ,oBAEV,MAAA,CAAA,EAAO,oBAsIL,EACA,4EAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,MAAO,IAAoB,UAE7B,GADM,EAAW,EAAA,GAAG,gBAAgB,EAAiB,GACjD,EAAS,SAAW,EAKtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAiB,YAC5C,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAe,MAE7B,EAAkB,EAAS,GAE7B,MAAA,CAAA,EAAO,GAA6B,EAAiB,OAAW,oBAc9D,EAAuB,EACvB,mHAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,EAAQ,MAAQ,KAClB,KAAM,IAAI,GACN,iHAGY,MAAA,CAAA,EAAM,EAAQ,kBAA1B,EAAY,EAAA,OACd,EAAgB,EAAU,cAC9B,AAAI,EAAc,cAAmB,MACnC,GAAgB,EAAc,cAG1B,EAAS,EAAQ,QAAU,KAAO,GAAO,EAAQ,OAMjD,EACF,EAAU,YAAc,MAAQ,EAAU,aAAe,MAAQ,EAC/D,EACF,GACI,GAAoB,GACpB,EAAe,GAEjB,EAAiB,EAAU,eACjC,AAAI,GAAkB,MACpB,EAAM,mBAAmB,GAE3B,AAAI,EAAU,qBAAuB,MACnC,EAAM,uBAAuB,EAAU,sBAIrC,GAAU,YAAc,MAAxB,MAAA,CAAA,EAAA,GAEF,GAAI,EAAU,aAAe,KAC3B,KAAM,IAAI,GACN,wHAIA,GAAmC,GACrC,EAAU,WAAY,EAAU,aAD7B,EAAY,EAAA,aAAE,EAAgB,EAAA,iBAErC,EAAM,YAAY,EAAc,GAE5B,EAAM,WAAa,MAAQ,EAAiB,OAAS,EACvD,CAAA,EAAM,EAAM,UAAU,WAAW,IAD/B,CAAA,EAAA,UACF,EAAA,wBAIF,EAAA,QAAQ,GACR,EAAA,QAAQ,EAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,2BAEtC,MAAA,CAAA,EAAO,QAGT,YACI,EAAqB,GAEvB,GAAM,GAAc,EAAA,GAAG,cAAc,EAAQ,GACvC,EAA+B,GAC/B,EAAkC,GACxC,SAAM,QAAQ,SAAA,GACZ,AAAI,EAAK,QAAU,YACjB,EAAiB,KAAK,CAAC,KAAM,EAAK,KAAM,OAAQ,EAAY,EAAK,QAEjE,EAAa,EAAK,MAAQ,EAAY,EAAK,QAGxC,CAAC,aAAY,EAAE,iBAAgB,sBAwCR,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CAAC,OAAQ,GAAI,QAAS,MAAI,KAUhC,GATA,EAAO,GAAQ,GAEf,EAAK,UAAY,GACjB,EAAK,MAAQ,GAGb,EAAK,KAAQ,EAAK,MAAQ,KAAQ,EAAK,KAAO,GAAO,eAGjD,EAAK,QAAU,KACjB,OAAoB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,IAAI,YAOP,SAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAQ,EAAM,aAAa,GAAG,cAAc,GAAG,MACrD,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,GAAI,IACtB,KAAM,IAAI,GACN,kDACG,GAAM,KAAI,uBACV,GAAM,aAAa,GAAG,aAAa,GAAG,MAAK,OAyBtD,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GACF,YAAiB,IAAc,YAAiB,IAChD,EACJ,GAAI,GAEF,GADA,EAAa,EACT,EAAW,QAAQ,SAAW,EAChC,KAAM,IAAI,GACN,yHAKN,GAAI,EAAW,OAAO,SAAW,EAC/B,KAAM,IAAI,GACN,uHAOR,GAAI,KAAK,QAAQ,SAAW,GAE1B,GAAI,EAAM,aAAa,SAAW,GAEhC,GAAI,EAAM,iBAAmB,KAC3B,KAAM,IAAI,GACN,iGAIN,GAAM,GAAI,GAAM,CACd,WAAY,EAAM,gBAClB,MAAO,EAAM,MACb,KAAM,EAAM,KAAO,WAIrB,EAAM,MAAM,GAGd,GAAI,EACF,KAAK,QAAU,EAAW,QAC1B,KAAK,OAAS,EAAW,YAEzB,GAAI,EAAM,aAAa,SAAW,EAChC,KAAM,IAAI,GACN,2DACA,yDACI,EAAM,KAAI,KACd,cAAa,EAAM,aAAa,OAAM,0BACtC,gBAGN,GAAI,EAAM,aAAa,GAAG,cAAc,SAAW,EACjD,KAAM,IAAI,GACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,EAAM,aAAa,GAAG,cAAc,IACpD,KAAK,OAAS,GAAgB,KAAK,QAAQ,IAG7C,KAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,KACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,KAAK,OACnB,cAAe,KAAK,QAEpB,WAAY,GAA2B,KAAM,KAAK,OAAO,QACzD,YAAa,CAAC,MACd,YAAa,KAAK,OAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QACpC,aAAc,KAAK,QAAQ,GAAG,aAGhC,GAAM,GAAe,EAAM,MAAM,KAAK,QAAQ,IAC9C,GAAI,MAAM,QAAQ,GAChB,KAAM,IAAI,WACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,GAEhB,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,OAGvD,KAAK,OAAO,KAAK,GACjB,KAAK,MAAQ,IAQf,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,OAAO,SAAW,EACzB,KAAM,IAAI,WAAU,qCAItB,GADA,KAAK,OAAO,MACR,KAAK,OAAO,SAAW,EACzB,KAAK,QAAU,GACf,KAAK,aAAe,GACpB,KAAK,cAAgB,QAErB,GAAM,GAAiB,KAAK,OAAO,OAAS,EAC5C,KAAK,OAAO,GAAgB,cAAgB,GAC5C,KAAK,QAAU,CAAC,KAAK,OAAO,GAAgB,QAE5C,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,SAIzD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,KAAK,EAAQ,IAGjC,EAAA,UAAA,MAAA,SAAM,GAKJ,GAFA,GAAmB,GAEf,KAAK,OAAO,SAAW,GAAK,KAAK,QAAQ,SAAW,EACtD,KAAM,IAAI,WACN,4EAIN,KAAK,MAAQ,GAAI,IAAY,CAC3B,OAAQ,KAAK,OACb,QAAS,KAAK,QAAQ,GACtB,KAAM,KAAK,KAAO,WAEpB,KAAK,MAAM,UAAY,KAAK,UAG5B,KAAK,gBAAkB,KAAK,MAAM,gBAElC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,uBAAyB,KAAK,MAAM,uBACzC,KAAK,yBAA2B,KAAK,MAAM,yBAC3C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,wBAA0B,KAAK,MAAM,wBAC1C,KAAK,0BAA4B,KAAK,MAAM,0BAC5C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,WAAa,KAAK,MAAM,WAG7B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,WACE,MAAK,MAAK,OACR,KAAK,QAEA,EAAA,UAAM,YAAW,KAAA,OAiC1B,EAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAAA,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC9D,AAAK,KAAK,OACR,KAAK,QAEP,EAAA,UAAM,QAAO,KAAA,KAAC,EAAY,EAAW,IASvC,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,OAAS,MAChB,KAAK,QAEP,KAAK,MAAM,WAAW,IAmCxB,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GACF,GADE,AAAA,IAAA,QAAA,GAAA,IACE,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAO,MAAK,MAAM,SAAS,EAAG,EAAG,IAyB7B,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAA,CAAA,EAAO,KAAK,MAAM,gBAAgB,EAAS,SA8B7C,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC1B,AAAI,KAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,QAAQ,EAAG,IAU/B,EAAA,UAAA,eAAA,SAAe,GACb,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,eAAe,IAQnC,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,QACL,KAAK,MAAM,QAAQ,GACnB,KAAK,WAAa,KAAK,MAAM,UAE7B,KAAK,iBAAoB,KAAK,MAAc,iBAC5C,KAAK,KAAO,KAAK,MAAM,KACvB,KAAK,QAAU,KAAK,MAAM,QAG1B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,aAAe,KAAK,MAAM,cAIjC,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAO,OAAY,KAAK,MAAM,eAGrD,SAAc,GACZ,KAAK,MAAM,UAAY,mCAkCnB,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,IAAI,EAAG,EAAG,SAwFxB,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,WAAW,EAAS,SA0BlC,EAAA,UAAA,aAAN,SACI,EACA,sEAEF,MAAA,CAAA,EAAO,KAAK,MAAM,aAAa,EAAG,SAK7B,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GACA,EAA6C,GACjD,GAAI,YAAkB,QACpB,GAAI,CAAE,GAAO,GAAG,WAAa,OACzB,EAAO,GAAG,YAAiB,QAC7B,KAAM,IAAI,GAAW,kDAEvB,EAAc,MAEd,GAAA,KAAK,OACD,EAAO,QAAa,KACpB,WACI,MAAA,wHAER,EAAc,EAAO,OACrB,MAAO,GAAO,OACd,EAAmB,EAGrB,GAAM,GAAQ,GAAI,GAAI,GACtB,GAAI,CAAE,aAAiB,IACrB,KAAM,IAAI,IACN,yDAAyD,GAE/D,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAA0C,OAC1C,EAAQ,GACI,EAAkC,EAClC,GAClB,AAAI,GACF,EAAM,6BAA6B,IAErC,EAAM,IAAI,GAEZ,MAAO,IA+BT,OAAA,eAAI,EAAA,UAAA,eAAY,KAWhB,WACE,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,MAAO,MAAK,MAAM,kBAjBpB,SAAiB,GAGf,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,KAAK,MAAM,aAAe,mCAe5B,EAAA,UAAA,UAAA,WAME,OADM,GAAqC,GACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAiC,GACvC,EAAK,UAAe,EAAM,eAC1B,EAAK,OAAY,EAAM,YACvB,EAAO,KAAK,GAEd,MAAO,CAAC,KAAM,KAAK,KAAM,OAAM,IAvsB1B,EAAA,UAAY,aAysBrB,GA3sBgC,IA4sBhC,EAAA,cAAc,cAAc,IC3kC5B,YAqEsB,GACpB,MAAO,IAAI,IAAY,eA+DE,GACzB,MAAO,IAAI,IAAW,eA+FpB,EACA,GACF,MAAI,IAAW,MACb,GAAU,IAEL,GAAwB,EAAiB,eA0B5B,GACpB,MAAO,IAAM,eAIX,EACA,GACF,GAA4B,4BACxB,EAAgB,GC5QtB,AAwBA,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAEE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GALyC,EAAA,cAAc,cAWvD,GAAA,SAAA,GAAyB,EAAA,EAAA,GAAzB,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,GAAM,EAAG,IATF,EAAA,UAAY,MAW9B,GAbyB,IAczB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAK,EAAA,KAAS,OAF9B,EAAA,UAAY,QAI9B,GAN2B,IAO3B,EAAA,cAAc,cAAc,IAG5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAFO,EAAA,UAAY,SAI9B,GAN4B,IAO5B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,QAAY,IAFL,EAAA,UAAY,UAI9B,GAN6B,IAO7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAAjC,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAc,IAFP,EAAA,UAAY,cAI9B,GANiC,IAOjC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,SAAa,IAFN,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAW,IAFJ,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAeE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,QAAY,EAAG,IAdR,EAAA,UAAY,UAgB9B,GAlB6B,IAmB7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DAgBE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,WAAe,EAAG,IAfX,EAAA,UAAY,aAiB9B,GAnBgC,IAoBhC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,EAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAE,IAAI,IAAQ,IAAI,MATlC,EAAA,UAAY,QAW9B,GAb2B,IAc3B,EAAA,cAAc,cAAc,gBAEQ,GAClC,MAAO,GAAW,2BAIjB,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACM,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,MAChB,GAAM,GAAmC,GACzC,SAAO,UAAe,SACtB,EAAO,OAAY,GACZ,GAAsB,GAE/B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCjQjC,AAiBA,YAA0B,GACxB,GAAI,GAAQ,MAAQ,MAAO,IAAS,SAClC,KAAM,IAAI,OACN,mEACA,0BAAyB,IAOjC,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DAEA,MAAA,IAF0C,EAAA,cAAc,cAqBxD,GAAA,SAAA,GAA0B,EAAA,EAAA,GAQxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAEP,UAAiB,GAEjB,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,MAAQ,EAAK,KAAO,EACzB,EAAK,MAAQ,EAAK,KAAO,IAO3B,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,GAAyB,EAAA,MAAM,CAAC,IACpC,MAAI,GAAK,OACP,GAAiB,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAI,OAEhE,AAAI,EAAK,OACP,GACI,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,GAAS,OAEjD,EAAe,cAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,GAAM,KAAK,GAAI,GAAM,KAAK,KAI7B,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,CAAC,GAAI,EAAO,GAAiB,GAAI,EAAO,MA3ClD,EAAA,UAAY,OA6CrB,GA/C0B,IAgD1B,EAAA,cAAc,cAAc,gBAET,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,gBAGvC,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,IAOnD,GAAM,IACyC,CAChD,KAAQ,oBAGuB,GAEnC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGU,GAG7B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAuB,OACzB,OAAI,aAAsB,IACxB,EAEA,GAAuB,GClJlC,AAmCA,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAKxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,MACV,GAAK,SAAW,EAAK,YAIzB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,EAAS,GAAoB,GAC7B,GAAI,GAAS,EAAA,KAAK,GAClB,MAAI,MAAK,UAAY,MACnB,GAAS,EAAA,YAAY,EAAQ,EAAG,KAAK,WAEhC,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,SAAU,KAAK,UACnD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA5BF,EAAA,UAAY,OA8BrB,GAhC0B,IAiC1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAO7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,GAIvB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,UAAU,EAAG,KAAK,QAG3B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,YA4BrB,GA9B+B,IA+B/B,EAAA,cAAc,cAAc,IA6B5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAWzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAU/B,GAbO,EAAA,0BAAmD,QAI1D,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,gBAAkB,GACvB,EAAK,iBACD,GAAe,EAAK,kBAAoB,EAAK,2BACjD,EAAK,iBAAmB,GAAe,EAAK,kBAC5C,EAAK,gBAAkB,GAAc,EAAK,iBACtC,EAAK,YAAc,KACrB,EAAK,WAAa,aACT,MAAM,QAAQ,EAAK,YAC5B,EAAK,WAAa,EAAK,mBACd,MAAO,GAAK,YAAe,SACpC,EAAK,WAAa,CAAC,EAAK,gBAExB,MAAM,IAAI,GACN,8DACA,YAAW,EAAK,sBAIxB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAChC,GAAM,GAAoB,EAAW,MAAM,GAC3C,GAAI,KAAK,YAAc,KACrB,OAAgB,GAAA,EAAA,EAAA,KAAK,WAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAW,EAAI,GAAK,EAGxB,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAAM,KAAK,iBAEtC,GAAM,GAAiC,GACvC,GAAI,KAAK,YAAc,KACrB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,EAAK,GAAK,EAAW,GAGzB,KAAK,UAAY,CAAC,GAAI,IAAU,CAC9B,KAAM,EAAW,OACjB,KAAI,KAEN,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,SAAS,GAAoB,GACtB,EAAA,MAAM,EAAQ,KAAK,MAAM,SAGlC,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,iBAAkB,GAAqB,KAAK,kBAC5C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAoB,KAAK,iBAC1C,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAxEF,EAAA,UAAY,QA0ErB,GA5E2B,IA6E3B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAK/B,GARO,EAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGL,EAAK,OAAS,MAAQ,EAAK,QAAU,EAAK,cAC5C,KAAM,IAAI,IACN,4BAA4B,EAAK,MAAK,4CAI5C,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,IAAI,IAGb,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjCF,EAAA,UAAY,MAmCrB,GArCyB,IAsCzB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAOnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAE,IAAI,GAAK,EAAE,QAAQ,KAAK,OAAQ,aAG3C,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,kBA6BrB,GA/BqC,IAgCrC,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,aAAe,EAItB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,QAAU,GAAI,MAAoB,MACvC,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,OAG3D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,MAAK,QAAQ,EAAG,KAAK,OAG9B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,KAAM,KAAK,MAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,UA6BrB,GA/B6B,IAgC7B,EAAA,cAAc,cAAc,ICxV5B,YAwBI,EAAwB,EAAW,GACrC,GAAI,MAAO,IAAU,SACnB,MAAO,IAAa,EAAO,GAE3B,GAAI,EAAM,SAAW,EACnB,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EAAC,aACxD,eAAc,EAAM,OAAM,eAEhC,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAc,EAAM,GAC1B,GAAI,CAAC,GAAU,GACb,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EACvD,yBAAwB,KAAK,UAAU,GAAM,gBAC7C,wBAAuB,IAG/B,MAAO,eAaP,EAAqB,EAAoB,EACzC,EAAgB,GAClB,GADkB,AAAA,IAAA,QAAA,GAAA,GACd,GAAe,KACjB,MAAO,GAET,GAAM,GAAoB,EAAc,GAAa,GAAM,GAAW,GAClE,EACJ,MAAI,KAAY,OACd,EAAe,EAEf,EAAe,EAAc,EAAoB,EAE5C,KAAK,MAAO,GAAe,EAAS,GAAK,eAI9C,EAAiB,EAAoB,EACrC,GACF,GAAI,GAAW,KACb,MAAO,MAGT,GAAI,IAAY,QACd,EAAU,EAAU,EAAa,GAAI,CAAC,EAAa,EAAY,YACtD,IAAY,OACrB,EAAU,EAAU,MAEpB,MAAM,IAAI,GAAW,2BAA2B,EAAO,KAEzD,MAAO,GCpFT,YAwCI,EAAW,GAEb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,IAE3B,gBAWT,EAAW,GACb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,IAE9B,gBAsBT,EAAW,EAAgB,EAAc,EAAa,EACtD,EAAyB,GADgB,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,SAC7B,AAAA,IAAA,QAAA,GAAA,GACpB,EAAA,KAAK,WAMV,GALA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAEZ,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,+DACG,GAAE,MAAM,OAAM,cAEvB,GAAI,EAAO,MAAM,SAAW,EAC1B,KAAM,IAAI,GACN,iEACG,GAAO,MAAM,OAAM,aAE5B,GAAI,GAAQ,MAAQ,EAAK,MAAM,SAAW,EACxC,KAAM,IAAI,GACN,+DACG,GAAO,MAAM,OAAM,aAM5B,GAHA,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,KAE1B,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,GAAI,GAAY,EAAA,OACZ,EAA0B,EAAoB,EAC9C,IAAY,OAAS,OAAS,QAAS,MAAO,GAClD,MAAI,IAAQ,MACV,GAAI,GAAU,EAAG,IAEZ,gBAoDP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EAAyB,EAC5C,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,IACvD,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,gEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,iEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,MAAU,OAAO,CACnB,EAAG,EACH,OAAQ,EACR,QAAS,EACT,IAAK,IAAY,OAAS,OAAS,QACnC,UAAW,EACX,WAAY,OACZ,KAAI,EACJ,WAAU,IAEZ,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBA8BP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EACnB,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,EAAG,IAC1D,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,MAEf,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,oEACG,GAAE,KAAI,MAEf,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,OACA,EACA,EAAmC,EACnC,IAAY,OAAS,OAAS,QAAS,QAAS,GACpD,AAAI,GAAQ,MACV,GAAI,GAAU,EAAG,IAEnB,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAmHX,GAAA,IAAA,SAAA,GAAuC,EAAA,EAAA,GAwBrC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,IAAkB,KAIxB,GAVQ,EAAA,KAAsB,KAEvB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAIzD,EAAS,WAAW,GACpB,EAAK,KAAO,EACZ,GAAoC,EAAK,KAAM,QAC3C,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,EAAK,OAAS,EACtD,KAAM,IAAI,IACN,qDACI,EAAK,KAAI,6BAqBnB,GAlBA,EAAK,WAAa,GAAe,EAAK,WAAY,EAAM,cACxD,EAAK,QAAU,GACX,EAAK,SAAW,KAAO,EAAI,EAAK,QAAS,EAAM,WACnD,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,WAAa,GAAc,EAAK,YACrC,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAClD,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,aAAe,GAChB,EAAK,cAAgB,KAAO,EAAI,EAAK,aAAc,EACnD,gBACA,EAAK,OAAS,GACb,MAAM,QAAQ,EAAK,eAAiB,EAAK,aAAa,SAAW,EACpE,KAAM,IAAI,GACN,iGAEA,IAAG,KAAK,UAAU,EAAK,gBACtB,GAAI,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aAAe,CAAC,EAAK,aAAc,EAAK,sBACpC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,gEACA,8BAA6B,KAAK,UAAU,EAAK,wBAE9C,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aACD,CAAC,EAAK,aAAc,EAAK,aAAc,EAAK,sBACvC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,kEACA,8BAA6B,KAAK,UAAU,EAAK,yBAK1C,SAAA,WAAjB,SAA4B,GAI1B,GAFA,GACI,cAAgB,GAAM,2CACtB,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,oEACA,oCACI,KAAK,UAAU,EAAK,YAAW,OAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,gBAAiB,GAAqB,KAAK,iBAC3C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA/GuC,IAqHvC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAcjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAA0B,KAZ9B,SAAA,OAAwB,KAahC,EAAK,WAAW,GAChB,EAAK,QAAU,EAAK,QACpB,GAAoC,EAAK,QAAS,WAClD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,kBAAoB,GAAe,EAAK,qBAG/C,SAAA,UAAA,MAAA,SAAM,SACJ,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAE1B,GAAM,GAAW,EAAW,GAEtB,EAAc,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,UAE3D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,KAAM,KAAK,gBACnC,KAAK,gBAAiB,GAAM,KAAK,iBAGvC,KAAK,UAAY,CAAC,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,KACtE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GACE,EAAY,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACjD,EAAsB,GACU,EAAK,WAAW,gBAEtD,GAAI,GAAuB,MAAQ,EAAK,OAAS,EAC/C,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,aACtB,QAEJ,GAAI,EAAK,OAAS,EAChB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAQ,GACpD,EAAK,QAAS,EAAK,WAAY,EAAK,aAAa,YAC5C,EAAK,OAAS,EAEvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,sBACjB,EAAK,OAAS,EACvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,kBAE1B,MAAM,IAAI,IACN,yDAGN,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAIpC,MAAO,MAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAKhC,OAJM,GAAqB,GACrB,EAAS,KAAK,aAAe,eAC/B,EAAW,MAAM,EAAG,EAAW,OAAS,GACxC,EAAW,MAAM,GACZ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAM,GAAS,GACX,EAAM,GAAI,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,GACzD,MAAO,MAAK,cAAiB,SAAW,KAAK,aACL,KAAK,aAAa,IAC9D,EAAS,KAAK,GAGhB,GAAI,GAAc,CAAC,EAAW,IAC9B,MAAI,MAAK,aAAe,eACtB,GAAc,EAAY,OAAO,GACjC,EAAY,KAAK,KAAK,UAEtB,GAAY,KAAK,KAAK,SACtB,EAAc,EAAY,OAAO,IAE5B,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,kBAAmB,GAAqB,KAAK,mBAC7C,iBAAkB,GAAoB,KAAK,mBAEvC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,CAAE,YAAa,KAAS,MAAO,GAAK,SAAY,UAChD,EAAK,QAAU,EACjB,KAAM,IAAI,GACN,kEACA,YAAW,KAAK,UAAU,EAAK,YAGzC,GA1ImC,IA4InC,GAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAK,MAAO,GAAK,YAAe,UAC5B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,gCAA+B,KAAK,UAAU,EAAK,YAAW,OAnB/D,EAAA,UAAY,SAsBrB,GAxB4B,IAyB5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UACzB,CAAE,OAAM,QAAQ,EAAK,aAClB,GAAK,WAAW,SAAW,GAAK,EAAK,WAAW,SAAW,IAChE,KAAM,IAAI,GACN,mDACA,4CACI,KAAK,UAAU,EAAK,YAAW,OApBtC,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAGX,GAFA,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KAEnC,EAAK,UAAY,QAAU,EAAK,UAAY,QAC9C,KAAM,IAAI,GACN,gEACA,2CAA0C,EAAK,mBAIvD,SAAA,UAAA,MAAA,SAAM,SAGJ,GAFA,EAAa,GAAmB,GAE5B,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,mDACA,KAAK,UAAU,IAGrB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,wEAGN,GAAM,GAAW,EAAW,GACtB,EAAc,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,IAE1D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,UAAW,KAAK,kBACvC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,GAAM,KAAK,iBAIvC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MAC3D,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAM,MAAM,SAAW,EACzB,KAAM,IAAI,GACN,iEACA,8BAA6B,EAAM,MAAM,SAG/C,GAAM,GAAa,EAAM,MACnB,EAAY,EAAW,GAEzB,EACA,EACJ,AAAI,EAAK,aAAe,gBACtB,GAAQ,EACR,EAAQ,GAER,GAAQ,EACR,EAAQ,GAGV,GAAM,GAAS,EAAW,GACpB,EAAQ,EAAW,GACnB,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,QAAQ,GACvB,EAAU,EAAK,QAAQ,GAGvB,EAAY,GAAa,EAAQ,EAAS,EAAS,EAAK,SACxD,EAAW,GAAa,EAAO,EAAS,EAAS,EAAK,SAMtD,EACF,CAAC,EAAW,EAAW,EAAU,EAAK,SAE1C,AAAI,EAAK,aAAe,gBACtB,GAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,KAEzC,GAAI,GAAU,EAAA,gBACV,EAAmB,EAAK,OAAO,OAAoB,EACnD,EAAK,QAA6B,EAAK,SAC3C,MAAI,GAAK,aAAe,gBACtB,GAAU,EAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,KAG7C,AAAI,EAAK,MAAQ,MACf,GACI,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEhD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAE3B,EACA,EACA,EACJ,AAAI,KAAK,aAAe,gBACtB,GAAc,EACd,EAAa,EACb,EAAY,GAEZ,GAAc,EACd,EAAa,EACb,EAAY,GAGd,GAAM,GAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,QAAQ,GACvB,EAAU,KAAK,QAAQ,GAE7B,SAAY,GAAe,KAAK,QAChC,EAAY,GACR,GAAa,EAAY,GAAa,EAAS,EAAS,KAAK,SACjE,EAAY,GACR,GAAa,EAAY,GAAY,EAAS,EAAS,KAAK,SACzD,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,aACP,GA9IF,EAAA,UAAY,kBAgJrB,GAlJqC,IAmJrC,EAAA,cAAc,cAAc,IA0C5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAqBjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAAO,KAEnB,GAXO,EAAA,8BACL,gBACK,EAAA,8BACL,gBAEM,EAAA,gBAAiC,KACjC,EAAA,gBAAiC,KAKrC,EAAO,SAAW,KACpB,KAAM,IAAI,GACN,uFAGN,GAAI,EAAO,mBAAqB,MAAQ,EAAO,mBAAqB,MAChE,EAAO,kBAAoB,KAC7B,KAAM,IAAI,GACN,sPAKN,GAAI,EAAO,SAAW,MAAQ,EAAO,UAAY,QAC7C,EAAO,UAAY,QACrB,KAAM,IAAI,GACN,gBAAgB,EAAK,KAAI,kCACzB,qCAAoC,KAAK,UAAU,EAAO,WAGhE,SAAK,gBACD,EAAO,iBAAmB,KAAO,EAAI,EAAO,gBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,qBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,uBAGlD,SAAA,UAAA,MAAA,SAAM,SAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,KAAK,KAAO,EAClC,KAAM,IAAI,GACN,0BAA0B,KAAK,KAAI,sBAChC,MAAK,KAAO,EAAC,gCAChB,IAAG,KAAK,UAAU,KAExB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,0DACA,cAAa,KAAK,UAAU,EAAW,MAO7C,OAJM,GAAW,EAAW,GACtB,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,kBACrC,EAAuB,GACpB,EAAI,EAAG,EAAI,KAAK,KAAM,EAAE,EAC/B,EAAqB,KAAK,GAE5B,EAAqB,KAAK,EAAW,KAAK,gBAAiB,KAAK,SAEhE,GAAM,GAAY,GAClB,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,EAAW,KAAK,gBAE1C,KAAK,KAAO,KAGd,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MACvE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAE7B,GAAI,GACJ,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,IACN,oDACC,MAAI,GAAK,OAAS,GACvB,CAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAG3C,EAAS,EAAA,gBACL,EAAoB,EAAK,gBAAgB,OACzC,EAAK,gBAAgB,OACrB,EAAK,QAA6B,EAAK,QACvC,EAAK,aAAkC,SAG7C,AAAI,EAAK,SACP,GAAS,GAAU,EAAQ,EAAK,KAAK,OAAQ,EAAK,aAEpD,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAGjC,AAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAEpC,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,kBACd,MAAO,GAAO,kBACd,MAAO,GAAO,iBACd,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,qBAC7B,EAAO,oBACH,GAAoB,KAAK,qBACtB,GAzJF,EAAA,UAAY,gBA2JrB,GA7JmC,IA+JnC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,SACV,GAAA,KAAA,KAAM,EAAG,IAAK,KAFT,SAAA,UAAY,kBAIrB,GANqC,IAOrC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,GAClB,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,WACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,2BAA0B,KAAK,UAAU,EAAK,YAAW,OArB1D,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAgC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAM9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,MAAI,OAAO,GAAK,UAAa,SAC3B,EAAK,SACD,CAAC,CAAC,EAAK,SAAU,EAAK,UAAW,CAAC,EAAK,SAAU,EAAK,WACrD,AAAI,MAAO,GAAK,SAAS,IAAO,SACrC,EAAK,SAAW,CACd,CAAC,EAAK,SAAS,GAAI,EAAK,SAAS,IACjC,CAAC,EAAK,SAAS,GAAc,EAAK,SAAS,KAG7C,EAAK,SAAW,EAAK,SAEvB,EAAK,WACD,EAAK,aAAe,OAAY,eAAiB,EAAK,WAC1D,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAI,MAAK,aAAe,gBACf,CACL,EAAW,GAAI,EAAW,GAC1B,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,IAGlD,CACL,EAAW,GACX,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GAAI,EAAW,KAK5E,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAFA,EAAS,GAAoB,GAEzB,EAAK,aAAe,gBACtB,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,QAEjE,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,OAKvE,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,SAAU,KAAK,SAAU,WAAY,KAAK,YACpD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAhEF,EAAA,UAAY,aAkErB,GApEgC,IAqEhC,EAAA,cAAc,cAAc,IAwB5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAOhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KALM,SAAA,aAAe,CAAC,EAAG,GAMpC,EAAK,UAAY,CAAC,CAAC,KAAM,IACzB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,aAGtD,SAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,KAAK,aAAe,iBACtB,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,QAE9C,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAO,EAAW,MAIrD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAC1B,EAAa,EAAM,MAEzB,GAAI,EAAK,aAAe,iBACtB,EAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,IACvC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GAClC,EAAU,EAAM,sBAAsB,CAAC,EAAQ,IACrD,MAAO,GAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,SAExC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GACxC,MAAO,GAAM,sBAAsB,CAAC,EAAQ,QAKlD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,KAAM,KAAK,KAAM,WAAY,KAAK,YAC5C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlDF,EAAA,UAAY,eAoDrB,GAtDkC,IAuDlC,EAAA,cAAc,cAAc,ICrvC5B,YA6CI,EAAW,EAAyB,EACpC,EAAmB,EACnB,GAFoC,MAAA,KAAA,QAAA,GAAA,CAA6B,EAAG,IACpE,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WACV,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAChB,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,OAEf,GAAI,EAAgB,OAAS,EAC3B,KAAM,IAAI,GACN,yDACG,GAAgB,KAAI,OAE7B,SAAI,EAAA,gBACA,EAAe,EAA6B,EAC5C,IAAY,OAAS,OAAS,QAAS,OAAQ,GACnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,IAsCX,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAsB,KAHzB,SAAA,gBAAiC,KAIvC,EAAK,gBACD,EAAK,iBAAmB,KAAO,EAAI,EAAK,gBAC5C,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,4BACtC,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,qBAAuB,GAAe,EAAK,wBAGlD,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,iDACA,0BAAyB,KAAK,UAAU,GAAW,MAEzD,GAAM,GAAc,KAAK,aAAe,gBAAkB,EAAI,EAC9D,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,iEACA,4BAA2B,EAAW,GAAY,OAExD,GAAM,GAAW,EAAW,GACtB,EAA8B,CAClC,KAAK,WAAW,GAAI,KAAK,WAAW,GAAI,EAAU,KAAK,iBAGzD,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,EAAW,KAAK,iBAAkB,KAAM,KAAK,gBACtD,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GAAU,GACV,EAAQ,EAAK,gBAAgB,OAAQ,EAAK,QAC1C,EAAK,QAAS,EAAK,WAAY,MAEnC,MAAI,GAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEtD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EAAa,KAAK,aAAe,gBACnC,EAAW,GAAK,KAAK,gBACrB,EAAW,GAAK,KAAK,gBACnB,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACnD,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,MAAI,MAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAY,EAAS,GAGrC,CAAC,EAAW,GAAI,EAAS,EAAS,IAI7C,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,SAAO,gBAAqB,KAAK,gBACjC,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,sBACtB,GAjGF,EAAA,UAAY,kBAmGrB,GArGqC,IAsGrC,EAAA,cAAc,cAAc,IClN5B,YA0DI,EACA,EACA,EACA,GAKF,GAAI,MAAM,QAAQ,IAChB,GAAI,GAAgB,MAAQ,GAAa,KACvC,KAAM,IAAI,GACN,iFAGN,AAAI,GAAgB,MAClB,GAAY,EAAO,MAAM,EAAO,OAAS,EAAc,EAAO,QAC9D,EAAS,EAAO,MAAM,EAAG,EAAO,OAAS,IAE3C,AAAI,EAAO,OAAS,GAClB,GAAe,EAAO,MAAM,EAAG,EAAO,SAExC,EAAS,EAAO,GAGlB,WAAsB,GAEpB,MAAI,IAAK,MAAQ,MAAM,QAAQ,GACtB,EAEA,CAAC,GAIZ,SAAe,EAAa,GAC5B,EAAY,EAAa,GAElB,CAAC,OAAM,EAAE,aAAY,EAAE,UAAS,eA+CrC,EAA+B,EAAgB,EAC/C,EAAqB,EAAe,EAAsB,EAC1D,GADA,MAAA,KAAA,QAAA,GAAA,IAA0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,IACK,EAAA,KAAS,WACd,GAAM,GAAO,EAAO,MAAM,OAC1B,GAAI,EAAO,EACT,KAAM,IAAI,GAAW,uCAAuC,EAAI,MAKlE,GAAM,GAAO,CAAC,EAAG,GAAG,OAAO,GAAiB,EAAG,IAG/C,GAFA,EAAS,EAAA,UAAc,EAAQ,GAE3B,GAAa,KACf,KAAM,IAAI,IACN,kFAKN,AAAI,GACF,QAAQ,KACJ,qGAIN,AAAI,GAAQ,MACV,GAAO,EAAK,OAAO,QAAQ,OAAO,WAClC,AAAI,EAAK,OAAS,EAAO,GACvB,GAAO,EAAA,WAAe,EAAM,KAE9B,EAAO,EAAA,UAAc,EAAM,IAG7B,AAAI,GACF,GAAS,EAAA,QAAY,EAAQ,GAC7B,AAAI,GAAQ,MACV,GAAO,EAAA,QAAY,EAAM,KAc7B,GAAM,GAA2B,GAC7B,EACA,EAAS,EACP,EAAY,EAAO,MAAM,GACzB,EAAgB,EAAA,QAAY,GAC9B,EACJ,AAAI,GAAQ,MACV,GAAe,EAAA,QAAY,IAG7B,mBAAS,GACP,GAAM,GAAe,EAAc,GAC7B,EAAc,EAAA,KAAS,WAAM,MAAA,GAAa,EAAc,KAE9D,GAAI,GAAQ,KACV,EAAa,EAAY,GACzB,EAAS,EAAY,QAErB,GAAM,GAAgB,EAAA,KAAS,WAC7B,GAAM,GAAW,EAAa,GACxB,EAAc,EAAA,SAAa,GAAU,IAAI,GAEzC,EACF,EAAY,GAAG,IAAI,GAAU,IAAI,EAAO,GAAG,IAAI,IAC7C,EAAY,EAAO,IAAI,SAAC,EAAO,GACnC,MAAO,GAAY,GAAG,GAAG,IAAI,GAAU,IAAI,EAAM,IAAI,MAEvD,MAAO,CAAC,OAAM,EAAE,UAAS,KAE3B,EAAa,EAAc,OAC3B,EAAS,EAAc,UAGzB,AAAI,GACF,EAAe,KAAK,IAxBf,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,GA2BT,GAAI,GACJ,GAAI,GACF,GAAM,GAAO,EACb,EAAU,EAAA,MAAU,EAAgB,GAEtC,MAAO,CAAC,EAAY,EAAS,wBAyGR,EAAA,EAAA,GAqBvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACP,EACJ,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,GACN,wDAMN,GALO,AAAI,MAAM,QAAQ,EAAK,MAC5B,EAAO,GAAI,IAAgB,CAAC,MAAO,EAAK,OAExC,EAAO,EAAK,KAEV,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,qGAGN,SAAK,KAAO,EACZ,EAAK,gBACD,EAAK,iBAAmB,KAAO,GAAQ,EAAK,gBAChD,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,UAAY,EAAK,UAAY,KAAO,GAAQ,EAAK,SACtD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAQ,EAAK,OAEjD,EAAK,gBAAkB,GACvB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KACvC,EAAK,UAAY,KACjB,EAAK,QAAU,KAEf,EAAK,aAAe,KAIpB,EAAK,WAAa,KAKpB,SAAA,UAAA,UAAA,WACE,GAAI,KAAK,SAAW,MAClB,GAAM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EACtE,MAAO,IAAiB,EAAG,GAAW,IAAI,SAAA,GAAK,MAAA,YAE/C,OAAO,MAAK,SAMhB,EAAA,UAAA,UAAA,SAAU,GACR,KAAK,QAAU,GAGjB,EAAA,UAAA,mBAAA,SAAmB,GACjB,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAGb,GAAI,GAAY,KAAK,KAAK,UAC1B,AAAK,MAAM,QAAQ,IACjB,GAAY,CAAC,IAEf,GAAM,GAAY,EAAU,GACxB,EAOJ,GANA,AAAI,KAAK,gBACP,EAAc,CAAC,EAAW,GAAI,EAAW,GAAI,GAE7C,EAAc,CAAC,EAAW,GAAI,GAG5B,KAAK,aAEP,OADM,GAAsB,GACV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAW,KAAK,CAAC,EAAW,GAAI,IAElC,MAAO,CAAC,GAAa,OAAO,OAE5B,OAAO,IAIX,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAS,WACd,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAM,GAAa,EAAK,gBAAkB,EAAO,KAEjD,GAAI,EAAK,aACP,GAAM,GAAY,EAAK,OAAO,IAAI,SAAA,GAAK,MAAA,QACvC,MAAO,CAAC,GAAY,OAAO,OAE3B,OAAO,MAWb,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,SAAW,MAIlB,OAHM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EAChE,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,EAAO,KAAK,MAEd,MAAO,OAEP,OAAO,MAAK,aAIhB,SAAW,GACT,KAAK,QAAU,mCAGV,EAAA,UAAA,MAAP,SAAa,GAIX,GAAI,KAAK,cAAgB,KACvB,KAAM,IAAI,IACN,oDAGN,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAEb,GAAM,GAAoB,KAAK,SAAW,EAAW,GAAK,KACpD,EAAW,EAAW,MAAM,GAClC,KAAK,UAAU,GAAK,GAAI,IAAU,CAAC,MAAK,CAAG,EAAW,MAAI,OAAK,KAI/D,GAAM,GAAiB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAK7D,KAAK,KAAK,MAAM,GAIlB,GAAI,GAOJ,GANA,AAAI,MAAM,QAAQ,KAAK,KAAK,WAC1B,EAAY,KAAK,KAAK,UAEtB,EAAY,CAAC,KAAK,KAAK,WAGrB,KAAK,WAAa,MACpB,GAAI,CAAC,EAAA,KAAK,YACF,KAAK,UAAU,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,EAAK,MAAM,OAAS,KAC1D,GACN,KAAM,IAAI,GACN,0DACA,uCAAsC,KAAK,UAAS,MACpD,8BAA6B,KAAK,KAAK,gBAG7C,MAAK,UACD,EAAU,IAAI,SAAA,GAAO,MAAA,IAAI,IAAU,CAAC,MAAO,CAAC,KAAM,OAExD,AAAI,KAAK,UACP,KAAK,eAqBT,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAK,WACH,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAEN,GAAM,GAAY,EAAK,UAAU,GAAG,MAAM,GAC1C,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUAQN,GAAI,EAAK,SAAW,KAClB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAU,CAAC,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,qBAEzC,GAAU,KAEnB,EAAA,QAAY,EAAK,SAEjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAQ,GAAK,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,iBAMpD,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAER,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,IAAa,GAKf,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GACf,EAAM,MAAM,QAAQ,EAAK,KAAK,WAChC,EAAK,KAAK,UAAU,GACpB,EAAK,KAAK,UACR,EAAgB,CAAC,EAAW,GAClC,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAEhB,EAAK,QAAQ,GAAS,GAG1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAI5D,EAAA,UAAA,MAAA,SACI,EACA,GAEF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAGX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAC1D,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAMzB,GAAI,GAAiD,GACjD,EAA+B,GACnC,GAAI,GAAgB,MAClB,EAAO,aAAkB,EACzB,EAAmB,EAAiB,OAAO,GAC3C,KAAK,UAAY,GACjB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,UAAU,KAAK,GAAI,IAAU,CAAC,MAAO,EAAM,SAKlD,EAAkB,EAAgB,OAAO,KAAK,WAEhD,AAAI,GAAa,MACf,GAAO,UAAe,EACtB,EAAmB,EAAiB,OAAO,GAE3C,KAAK,aAAe,EAAU,QAGhC,GAAM,GAAW,EAAiB,YAAc,IAChD,GAAI,GAEF,GAAM,GACF,CAAC,GAAQ,OAAO,GACd,EAAgB,KAAK,UAAU,OAAO,GAEtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GAAS,EAAA,UAAM,MAAK,KAAA,KAAC,EAAW,GACtC,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAK/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KAIE,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC5C,EACA,GAAU,KAAO,KAAO,EAAO,aAEnC,EAAS,GAAoB,GAC7B,AAAI,GAAgB,MAClB,CAAI,EAAK,SACP,EAAe,EAAK,QAEpB,EAAe,EAAK,gBAAgB,IAIxC,GAAM,GACF,MAAM,QAAQ,EAAK,KAAK,WAAa,EAAK,KAAK,UAAU,OAAS,EACtE,GAAI,EAAa,SAAW,EAC1B,KAAM,IAAI,GACN,iBAAiB,EAAS,4BACvB,GAAa,OAAM,uBAE5B,AAAI,EAAK,QACP,QAAQ,KACJ,oEAGN,GAAM,GAAyB,CAAC,SAAQ,GAGlC,EAAO,SAAC,EAAgB,GAG5B,GAAM,GACF,EAAK,KAAK,KAAK,CAAC,GAAQ,OAAO,GAAS,GAE5C,MAAO,CAAC,EAAQ,GAAI,EAAQ,MAAM,KAK9B,EACF,GAAI,EAAM,EAAQ,EAAc,EAAK,YAAa,EAAM,KACpD,EAAK,OAAQ,EAAK,iBACpB,EAAa,EAAW,GACxB,EAAU,EAAW,GACrB,EAAS,EAAW,GAE1B,AAAI,EAAK,UACP,EAAK,YAAY,EAAQ,GAG3B,GAAM,GAAS,EAAK,gBAAkB,EAAU,EAIhD,MAAI,GAAK,YACA,CAAC,GAAQ,OAAO,GAEhB,KAKb,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAAI,GAAe,EAAA,MAAU,EAAO,OAKpC,MAHA,GAAe,EAAA,IAAQ,EAAc,CAAC,EAAG,IACzC,EAAe,GAAa,GAExB,MAAM,QAAQ,EAAK,KAAK,WACnB,EAAK,KAAK,UAAU,IACvB,SAAA,GAAO,MAAA,GAAM,EAAI,GAAO,EAAc,CAAC,EAAG,IAAQ,IAE/C,EAAK,KAAK,UAAY,EACzB,CAAC,GAAO,EAAc,CAAC,EAAG,EAAK,KAAK,aACpC,CAAC,MAKX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAK,MAAK,UAIH,KAAK,KAAK,iBAHR,oCAMX,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,MAAK,MAAK,UAGH,KAAK,KAAK,oBAFR,KAAK,KAAK,yCAKrB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,MAAQ,MACf,KAAK,KAAK,6BAA6B,IAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,gBAAiB,KAAK,gBACtB,YAAa,KAAK,YAClB,YAAa,KAAK,YAClB,SAAU,KAAK,SACf,OAAQ,KAAK,QAGf,AAAI,KAAK,cAAgB,MACvB,GAAO,aAAkB,KAAK,cAGhC,GAAM,GAAa,KAAK,KAAK,YAE7B,MAAI,MAAK,iBAAmB,EAAI,WAC9B,GAAO,KAAU,CACf,UAAa,KAAK,KAAK,eACvB,OAAU,IAKd,GAAA,GAAW,EAAe,EAAe,IAIpC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAa,EAAO,KACpB,EAAO,GAAY,EAAY,GACrC,MAAO,IAAI,GAAI,OAAO,OAAO,EAAQ,CAAC,KAAI,MArfrC,EAAA,UAAY,MAufrB,GAzfyB,IA0fzB,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAAtC,4DAQA,MAAA,IARsC,IAwFtC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAiCjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KANJ,SAAA,mBAAqB,OACrB,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAIzD,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,YAAc,KAAO,EAAK,mBAAqB,EAAK,YAC7D,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAG9B,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAEhC,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAW,EAAW,OAAS,GAAI,KAAK,OAAQ,KAC3D,KAAK,kBAAmB,KAAK,kBAAmB,GAChD,KAAK,kBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,OAAQ,KAC9C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IASf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8CAA8C,EAAO,OAAM,KAEjE,GAAI,GAAa,EAAO,GACxB,EAAS,EAAO,GAChB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,KAG/B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,KAGxC,GAAI,GACE,EAAiB,EAAK,YACtB,EAAoB,EAAK,qBAC/B,AAAI,GAAU,KACZ,EAAI,GAAM,EAAA,IAAQ,EAAQ,GAAS,EAAK,OAAO,QAE/C,EAAI,GAAM,EAAQ,EAAK,OAAO,QAEhC,AAAI,EAAK,MAAQ,MACf,GAAI,GAAU,EAAG,EAAK,KAAK,SAE7B,AAAI,GAAa,MACf,GAAa,EAAA,IAAQ,EAAY,IAEnC,GAAI,GAAS,EAAA,IAAQ,EAAG,GAAM,EAAY,EAAK,gBAAgB,SAC/D,MAAI,GAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAI1B,CAAC,EAAQ,MAIpB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,kBAGzB,MAAA,IAAA,GAAW,EAAe,IArKrB,EAAA,UAAY,gBAuKrB,GAzKmC,IA0KnC,EAAA,cAAc,cAAc,IA2F5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAG7B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAK,KAAO,GAAI,IAAc,GAC9B,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IA7BV,EAAA,UAAY,YA+BrB,GAjC+B,IAkC/B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAqC3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAbO,EAAA,mBAAqB,OACrB,EAAA,6BAAqD,cAErD,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAQrD,EAAK,WACP,KAAM,IAAI,GACN,+DAEN,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,KAAK,gBACrC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,uDACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACzD,EAAW,EAAO,GACtB,EAAS,EAAO,GAKhB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EAAY,EAAK,qBACnB,EACA,EACA,EAEJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAU,GAAM,EAAQ,EAAK,OAAO,QACxC,AAAI,EAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,SAEzC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAGzC,GAAM,GAAuB,EAAK,gBAAgB,OAC5C,EAAA,EAAA,MAAA,EAAA,CAAA,EAAA,EAAA,MAAA,EAAA,OAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAGN,EAAc,GAAM,EAAU,GAE9B,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GACT,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAE/C,GAAM,GAAa,GAAM,EAAA,IAAQ,EAAG,GAAW,GAC/C,EAAK,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,IAEvC,GAAM,GACF,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,IAAK,IAElE,MAAO,CAAC,EAAG,MAIf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,eACrB,WAAY,IAGd,MAAA,IAAA,GAAW,EAAe,IAvMrB,EAAA,UAAY,UAyMrB,GA3M6B,IA4M7B,EAAA,cAAc,cAAc,IA8B5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAQ,GACxB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,MAuCrB,GAzCyB,IA0CzB,EAAA,cAAc,cAAc,IAuC5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAsC5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAZJ,SAAA,mBAAqB,OACrB,EAAA,6BAA+B,cAC/B,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAEhC,EAAA,yBAA2B,QASlC,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,EAAK,eAE3B,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,CAAC,EAAK,MAAO,EAAK,OACnC,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,GAAI,GACJ,GAAI,KAAK,SACP,GAAI,KAAK,gBACP,GAAM,GAAmB,KAAK,gBACxB,EAAgB,KAAK,MAC3B,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GAAM,GAAK,EAAiB,MAAM,CAAC,IAC7B,EAAM,GAAI,MAAQ,MAAM,CAAC,IACzB,EAAS,EAAiB,MAAM,CAAC,EAAgB,IACvD,MAAO,IACH,GAAuB,EAAI,GAAK,IAExC,GAZgD,IAEvC,EAAA,UAAY,oBAYrB,GAAkB,KAAK,gBAEzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,EAAiB,KAAK,gBACtD,GAAM,KAAK,oBAEf,MAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,wDACG,GAAO,OAAM,MAEtB,GAAI,GAAW,EAAO,GAChB,EAAW,EAAO,GACxB,EAAS,EAAO,GAChB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EACF,EAAK,qBAKL,EACA,EACA,EACA,EACJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAI,GAAM,EAAQ,EAAK,OAAO,QAClC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAEzC,EAAI,EAAA,IAAQ,EAAG,GAAM,EAAU,EAAK,gBAAgB,SACpD,AAAI,EAAK,SACP,GAAI,GAAU,EAAG,EAAK,KAAK,SAGvB,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,KACnE,EAAI,EAAK,oBAAoB,MAAM,GAEnC,GAAM,GAAI,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,IAE3C,MAAO,CAAC,EAAG,EAAG,MAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,eAAgB,KAAK,eACrB,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,gBAGvB,MAAA,IAAA,GAAW,EAAe,IAnNrB,EAAA,UAAY,WAqNrB,GAvN8B,IAwN9B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAGxB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAS,GACzB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,OAuCrB,GAzC0B,IA0C1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAME,OADM,GAAsB,GACT,EAAA,EAAA,EAAA,KAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAU,KAAI,MAAd,EAAkB,EAAK,WAEvB,EAAU,KAAK,EAAK,WAGxB,MAAO,oCAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,EAKT,OAJI,GAAS,EAAO,MAAM,GAGpB,EAA2B,GACd,EAAA,EAAA,EAAA,EAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAa,KAAK,EAAO,OAAO,EAAG,EAAK,UAAU,SAElD,EAAa,KAAK,EAAO,OAAO,EAAG,IAGvC,EAAa,UAKb,OAFM,GAA8B,GAChC,EACK,EAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAO,EAAK,MAAM,GACxB,EAAS,EAAa,GAEtB,AAAI,IAAM,EACR,EAAa,CAAC,EAAO,IAAI,OAAO,GAEhC,EAAa,CAAC,EAAW,IAAI,OAAO,GAEtC,EAAa,EAAK,KAAK,EAAY,GACnC,EAAgB,KAAK,EAAW,MAAM,IAIxC,EAAS,GACT,OAAyB,GAAA,EAAA,EAAA,EAAgB,QAAQ,UAAxB,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAO,KAAI,MAAX,EAAe,GAEjB,MAAO,CAAC,EAAW,IAAI,OAAO,MAI3B,EAAA,UAAA,MAAP,SAAa,GACX,AAAI,GAAgB,IAGlB,GAAc,EAAuB,IAEvC,EAAa,EACb,GAAI,GACJ,KAAK,MAAM,QAAQ,SAAC,EAAM,GACxB,GAAU,WAAW,EAAK,WAGxB,EAAK,MAAM,GACX,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAY,EAAK,UAAU,GAE3B,EAAY,EAAK,UAEnB,EAAa,CAAC,EAAW,GAAI,OAGjC,KAAK,MAAQ,IAGf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAgB,SAAC,GACrB,MAAO,CACL,UAAa,EAAK,eAClB,OAAU,EAAK,cAIb,EAAc,KAAK,MAAM,IAAI,GAE7B,EAAS,CAAC,MAAS,GAEzB,MAAA,IAAA,GAAW,EAAe,IAIrB,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAElB,OADM,GAAmB,GACA,EAAA,EAAA,EAAC,EAAO,MAAR,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAM,KAAK,GAAY,EAAY,IAErC,MAAO,IAAI,GAAI,CAAC,MAAK,KAGvB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,kBAEvB,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,qBAEvB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACvB,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAiB,KAAI,MAArB,EAAyB,EAAK,kBAEhC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAQT,EAAA,UAAA,WAAA,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,SAEvB,MAAO,IAAc,IASvB,EAAA,UAAA,WAAA,SAAW,GAET,OADM,GAAyC,GAC5B,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,IAGjB,OAHS,GAAI,EAAA,GACP,EAAY,EAAK,QAAQ,OACzB,EAAe,EAAQ,OAAO,GAC3B,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,EACzC,EAAO,KAAK,CAAC,EAAK,QAAQ,GAAI,EAAa,KAG/C,GAAc,IA5KT,EAAA,UAAY,kBAgLrB,GAlLqC,IAmLrC,EAAA,cAAc,cAAc,gBAEQ,GAM3B,GAAA,GAAA,EAAA,KAAM,EAAA,EAAA,KAAM,EAAA,EAAA,SAAA,EAAA,IAAA,OAAA,GAAA,EAAkB,EAAA,EAAA,MAAA,EAAA,IAAA,OAAA,EAAA,EAE/B,EAAgB,WAAM,MAAA,IAAU,IAAQ,IAExC,EAAa,WAAM,MAAA,IAAe,EAAe,EAAM,IAG7D,GAAI,CAAC,GAAS,GAAS,EACrB,MAAO,GAAA,KAAS,IAAa,SAG/B,GAAM,GAAQ,MAAM,GAAO,KAAK,QAAW,IAAI,GAE/C,MAAO,GAAM,IAAI,SAAA,GAAK,MAAA,GAAA,KAAS,EAAE,WC/iEnC,AAiFA,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAArC,4DAyBA,MAAA,IAzBqC,IAiCrC,GAAA,SAAA,GAAwB,EAAA,EAAA,GAMtB,WAAY,GAAZ,GAAA,GAAA,KACE,GAAI,EAAK,OACP,KAAM,IAAI,IACN,sDAGN,GAAI,MAAM,QAAQ,EAAK,MACrB,KAAM,IAAI,IACN,kEAGN,SAAA,EAAA,KAAA,KAAM,IAAqB,KAE3B,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WAad,GAZA,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aAEtB,EAAK,KAAK,YAAc,MAG1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBAEtB,EAAK,KAAK,qBAAuB,MAG/B,GAAU,EAAO,UACnB,KAAM,IAAI,GAAW,6CAGvB,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KAEtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAE1C,EACF,GAAU,KAAO,KAAO,EAAO,aAEnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAI3D,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GAAkB,KAAK,yBAAyB,GAEpD,MAAK,MAAK,iBACR,GAAQ,CAAI,EAAS,IAAE,OAAK,EAAS,MAAM,KAG7C,AAAI,KAAK,aACP,GAAQ,CACH,GAAQ,OAAK,MAAM,GAAG,KAAI,CAAE,EAAW,IAAE,OAAK,EAAS,MAAM,QAG7D,GAGT,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACP,GAAA,GAAA,EAAA,KAAA,UAED,EAAa,EAAO,MAEpB,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAe,EAAA,MAAU,GAE/B,MAAI,OAAM,QAAQ,GACT,MAAM,EAAU,QAAQ,KAAK,GAG/B,CAAC,MAIZ,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAS,WACP,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAGN,GAAM,GAAa,EAAK,UAAU,GAAG,MAE/B,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAY,EAAW,GAE7B,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUASN,GAAI,EAAK,aAAe,KACtB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAU,CAAC,EAAA,MAAU,YAEnB,GAAU,KAEnB,EAAA,QAAY,EAAK,SAGjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAQ,GAAK,EAAA,MAAU,QAO9B,GAJA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAGR,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,EAKF,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GAEf,EAAgB,EAEtB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAGhB,EAAK,QAAQ,GAAS,GAI1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAIlD,EAAA,UAAA,yBAAV,SAAmC,GAC3B,GAAA,GAAA,KAAA,KAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,QAAS,EAAA,EAAA,aAGpD,EAAkB,IAAe,gBAEjC,EAAI,EAAW,EAAkB,EAAI,GACrC,EAAI,EAAW,EAAkB,EAAI,GAErC,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAClD,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAElD,EACD,EAAW,MAAM,EAAG,GAAE,OACrB,EAAkB,CAAC,EAAS,EAAM,GAAQ,CAAC,EAAM,EAAM,IAG7D,MAAO,IAhMF,EAAA,UAAY,YAkMrB,GApMwB,IAyMxB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAWlC,WAAY,GAAZ,GAAA,GAAA,KAEI,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,aAGF,SAAA,EAAA,KAAA,KAAA,GAAA,GAAU,EAAI,CAAE,MAAO,MAAS,KAEhC,EAAK,QAAU,EACf,GAAsB,EAAK,QAAS,WAEpC,EAAK,WAAa,GAAe,EAAY,EAAG,cAChD,EAAK,WAAW,QAAQ,SAAA,GAAQ,MAAA,IAAsB,EAAM,gBAE5D,EAAK,QAAU,GAAe,GAAW,EAAG,EAAG,WAC/C,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,IAAsB,EAAQ,aAE7D,EAAK,QAAU,GAAW,QAC1B,GAAiB,EAAK,SAEtB,EAAK,WAAa,GAAc,eAChC,GAAgB,EAAK,YAErB,EAAK,aAAe,GAAe,GAAgB,EAAG,EAAG,gBACzD,EAAK,aAAa,QACd,SAAA,GAAQ,MAAA,IAAsB,EAAM,oBAGnC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAEhC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAElE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAG1B,GAAM,GAAW,EAAW,GAEtB,EAAe,EAEf,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,QAAU,IAErD,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBAEvC,GAAM,GACF,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,KAAK,QAAU,IAOzD,GALA,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBAEL,KAAK,SACP,GAAI,GAAe,OAEnB,GAAI,KAAK,gBACP,GAAM,GAAO,KAAK,gBAEZ,EAAU,KAAK,QAErB,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAQ,EAAK,MAAM,CAAC,IACpB,EAAQ,EAAA,KAAS,CAAC,IAClB,EAAY,EAAK,MAAM,CAAC,EAAU,IACxC,MAAO,IAAc,CAAC,EAAO,EAAO,KAExC,GAVgD,IAEvC,EAAA,UAAY,oBAUrB,GAAkB,KAAK,gBAGzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,QAAU,GAAe,KAAM,EAC7C,KAAK,gBAAiB,GAAM,KAAK,gBAGvC,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAsB,GAA3B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8DACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,GAEjC,EAAI,EAAO,GACX,EAAW,EAAO,GAClB,EAAW,EAAO,GAElB,EAAe,EAIrB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAI9B,GAAM,GAAc,EAAK,YAEnB,EACF,SAAC,GAAe,GAAoB,IAClC,MAAI,CAAC,IAAQ,CAAC,GAAK,IACV,GAGF,EAAA,IAAQ,GAAK,IAAQ,KAG9B,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAEtC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAIvC,GAAM,GAAiB,EAAK,qBAExB,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAE1C,EAAoB,EAEpB,EAAA,EAAA,MAAA,EAAA,OAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAG5B,EAAA,EAAA,uDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAI5B,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAEvC,GAAA,GAAA,EAAA,MAAA,EAAA,gBAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAI3C,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAE5B,GAAM,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAA,IACN,EAAA,IAAQ,GAAG,GACX,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,MAC3C,GAAI,EAAA,IACN,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC3C,EAAK,WAAW,MAAM,KAE1B,MAAO,CAAC,GAAG,GAAG,OAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAA,EAAA,UAAA,UAAA,KAAA,MAAC,EAAA,EAAA,MAAY,EAAA,GAAA,EAAA,CAAA,UAEb,EAAuC,CAC3C,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,QAAS,KAAK,SAGhB,MAAA,IAAA,GAAW,EAAe,IAG5B,EAAA,UAAA,UAAA,SAAU,EAAW,EAAW,EAAY,GAC1C,GAAM,GAAM,EAAA,OACR,EAAmB,EAAmB,KAAK,QAC1C,GAAW,QACZ,KAAK,aAAe,gBAAkB,OAAS,OAC/C,KAAK,cAET,MAAI,GACK,GAAU,EAAK,EAAG,KAAK,YAGzB,GAGT,EAAA,UAAA,cAAA,SAAc,EAAW,GACvB,GAAM,GAAU,EAEhB,MAAO,GAAA,OACH,EAAmB,EAAmB,EAAS,OAC/C,KAAK,aAAe,gBAAkB,OAAS,SAzO9C,EAAA,UAAY,iBA2OrB,GA7OoC,IA+OpC,EAAA,cAAkB,cAAc,IAKhC,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,GAAI,IAAe,GAEhC,SAAA,EAAA,KAAA,KAAM,GAAA,GAAI,EAAI,CAAE,KAAI,MAAwB,OAIvC,SAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IAZV,EAAA,UAAY,aAcrB,GAhBgC,IAkBhC,EAAA,cAAkB,cAAc,ICjkBhC,AAkDA,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,KAAK,IAAI,KAAK,IAAI,EAAK,KAAM,GAAI,GAE7C,EAAK,WAAa,EAAK,WACvB,EAAK,KAAO,EAAK,KACjB,EAAK,gBAAkB,KAGf,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,KAAK,YAAc,KACrB,MAAO,MAAK,WAId,OAFM,GAAa,EAAM,MACnB,EAAoB,GACjB,EAAI,EAAG,EAAI,KAAK,WAAW,OAAQ,EAAE,EAC5C,EAAW,KACP,KAAK,WAAW,IAAM,KAAO,EAAW,GAAK,KAAK,WAAW,IAEnE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAI,EAAK,MAAQ,EAAK,KAAO,GAC/B,GAAM,GACF,EAAO,UAAe,KAAO,GAAQ,EAAO,SAC1C,EAAa,EAAK,cAAc,GAChC,EAAS,GACX,WAAM,MAAA,IAAU,EAAO,EAAK,KAAM,EAAY,EAAK,OACnD,WAAM,MAAA,IAAO,GACjB,MAAO,GAET,MAAO,MAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,KACX,WAAY,KAAK,WACjB,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAxDf,EAAA,UAAY,UA0DrB,GA5D6B,IA6D7B,EAAA,cAAc,cAAc,IA4D5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAIpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,CAAC,KAAM,MAGjB,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAM,GAAa,EAAM,MACzB,MAAO,CAAC,EAAW,GAAI,EAAG,EAAW,KAThC,EAAA,UAAY,mBAWrB,GAbsC,IActC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAmBzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAhBM,EAAA,WAA2B,KAC3B,EAAA,QAAU,GAGV,EAAA,OAAwB,KACxB,EAAA,KAAsB,KAErB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAQrD,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MACnD,EAAK,UAAY,MAGnB,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAK,gBAAkB,CAAC,EAAW,EAAK,UAG1C,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GAAc,EAAK,YACrC,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAK,SAEtB,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,gBAAkB,GAEvB,EAAK,UAAY,CAAC,CAAC,QAAS,MAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAe,EAAW,EAAW,OAAS,GACpD,AAAI,KAAK,QAAU,MACjB,MAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAc,KAAK,OAAQ,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,kBAIzC,KAAK,UAAY,CAAC,CAAC,QAAS,EAAG,KAAI,GAAA,GAAG,EAAC,IAAK,EAAY,KACxD,KAAK,MAAQ,IAGf,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,SAAY,EAAY,OAAS,GAAK,KAAK,MACpC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAM,GAAQ,GAAoB,GAC5B,EACF,GAA2B,EAAK,WAAW,gBAC3C,EAEJ,MAAI,IAAuB,KACzB,EAAS,GACL,EAAO,EAAK,OAAO,OAAQ,EAC3B,EAAK,KAAO,EAAK,KAAK,OAAS,MAEnC,GAAS,GAAM,EAAO,EAAK,OAAO,QAClC,AAAI,EAAK,MAAQ,MACf,GAAS,GAAU,EAAQ,EAAK,KAAK,SAEvC,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,KAI5B,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArHF,EAAA,UAAY,QAuHrB,GAzH2B,IA0H3B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAO,GAAQ,GACf,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,UAAY,CAAC,CAAC,QAAS,IAC5B,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,OAAkB,GAAA,EAAA,EAAA,EAAW,MAAM,GAAjB,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAO,KACT,KAAM,IAAI,GACN,4DACA,SAAQ,EAAW,MAAM,GAAE,oCAC3B,mFAIR,MAAO,CAAC,EAAW,GAAI,GAAU,EAAY,KAG/C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAK,aAAe,iBAAmB,EAAM,KAAO,GAEtD,OADM,GAAwB,CAAC,GACtB,EAAI,EAAG,EAAI,EAAM,KAAM,EAAE,EAChC,EAAY,KAAK,GAEnB,EAAY,KAAK,GACjB,EAAQ,EAAM,UAAU,GAG1B,MAAO,IAAe,MAI1B,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,GACzC,AAAI,KAAK,YAAc,MACrB,GAAO,WAAgB,KAAK,YAE9B,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA/CF,EAAA,UAAY,UAiDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAK9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,WAAa,GAAc,EAAK,cAGvC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAK,WAAW,MAAM,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,GAAoB,KAAK,aAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArBF,EAAA,UAAY,aAuBrB,GAzBgC,IA0BhC,EAAA,cAAc,cAAc,IAc5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAKhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,EAAI,EAAK,EACd,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,KAAK,EAAG,EAAW,KAG5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAS,GAAoB,GACtB,GAAS,EAAQ,EAAK,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,EAAG,KAAK,GAEJ,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,eA4BrB,GA9BkC,IA+BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,YAAc,EAAK,YAGxB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,EAAE,EAC7C,AAAI,EAAK,UAAU,EAAK,YAAY,KAClC,GAAK,YAAY,GAAK,eAKpB,SAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,GAAM,GAAK,GAAO,MAiBnB,EAAA,UAAA,oBAAR,SAA4B,EAAmB,GAK7C,OAJM,GAAW,6CACX,EAAa,EAAY,QAC3B,EAAQ,EACR,EAAU,KACL,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAM,EAAW,GACvB,GAAI,KAAK,UAAU,GACjB,GAAI,IAAY,KACd,EAAU,MAEV,MAAM,IAAI,GAAW,gDAGvB,IAAS,EAIb,GAAM,GAAe,GAAU,GAC/B,GAAI,IAAY,MACd,GAAI,IAAU,GAAK,EAAe,IAAU,EAC1C,KAAM,IAAI,GAAW,GAEvB,EAAW,GAAW,EAAe,UAC5B,IAAiB,EAC1B,KAAM,IAAI,GAAW,GAGvB,MAAO,IAGT,EAAA,UAAA,mBAAA,SAAmB,GAEjB,OADI,GAAiB,GACZ,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,GAAI,KAAK,UAAU,EAAW,KAC5B,EAAiB,GACjB,MAIJ,MAAI,GACK,EAAW,MAAM,EAAG,GAAG,OAAO,KAAK,aAEnC,EAAW,MAAM,EAAG,GAAG,OAC1B,KAAK,oBAAoB,EAAW,MAAM,GAAI,KAAK,eAI3D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAc,EAAW,MAAM,EAAG,GAAG,OACvC,EAAK,oBAAoB,EAAW,MAAM,GAAI,EAAK,cACvD,MAAO,GAAM,QAAQ,MAIzB,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlGF,EAAA,UAAY,UAoGrB,GAtG6B,IAuG7B,EAAA,cAAc,cAAc,IAY5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAM3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,OACN,mFAGN,GAAI,CAAC,MAAM,QAAQ,EAAK,MACtB,KAAM,IAAI,OACN,oEACG,GAAK,KAAI,cAIlB,GAAM,GAAwB,GAAM,EAAG,EAAK,KAAK,OAAS,GAC1D,GAAI,CAAC,EAAA,KAAK,YAAY,EAAK,KAAK,QAAQ,OAAQ,GAC9C,KAAM,IAAI,OACN,+BAAiC,KAAK,UAAU,EAAK,MACrD,8DAGN,SAAK,KAAO,EAAK,KACjB,EAAK,mBAAqB,CAAC,GAAG,OAAO,EAAK,MAC1C,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,EAAK,KAAK,OAAS,OAG5D,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,YAAK,KAAK,QAAQ,SAAC,EAAa,GAC9B,EAAY,EAAI,GAAM,EAAqB,KAEtC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,UAAU,GAAoB,GAAS,KAAK,qBAGrD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjDF,EAAA,UAAY,UAmDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,KACV,EAAK,UAAY,EAAK,WAAa,KAAO,EAAI,EAAK,UAEnD,EAAK,UAAY,IAIrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,UAAW,KAAK,WAChC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACb,MAAO,GAAA,IAAI,EAAA,SAAS,EAAO,KAAK,WAAY,IAG9C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACP,EAAW,GACX,EAAc,EAAA,IAAI,EAAA,SAAS,EAAO,EAAK,WAAY,EAAM,GACzD,EAAS,EAAM,IAAI,EAAY,OAAO,EAAM,QAClD,MAAO,MAtCJ,EAAA,UAAY,UAyCrB,GA3C6B,IA4C7B,EAAA,cAAc,cAAc,IClqB5B,AA2EA,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAgB7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GATM,EAAA,WAA4B,KAE3B,EAAA,+BACL,gBAME,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MAKrD,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,AAAI,EAAK,aAAe,KAGtB,EAAK,gBAAkB,CAAC,EAAW,MAInC,EAAK,gBACD,CAAC,GAAW,OAAO,GAAqB,EAAK,cAGrD,SAAK,SAAW,EAAK,SACrB,GAAoC,EAAK,SAAU,YACnD,EAAK,UAAY,EAAK,UACtB,GAAoC,EAAK,UAAW,aACpD,EAAK,sBAAwB,GACzB,EAAK,uBAAyB,EAAK,gCACvC,EAAK,sBAAwB,GAAe,EAAK,uBACjD,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,qBAAuB,GAAc,EAAK,sBAC/C,EAAK,SAAW,EAAK,SACrB,EAAK,gBAAkB,EAAK,SAC5B,EAAK,YAAc,EAAK,cAGnB,SAAA,UAAA,MAAP,SAAa,GACX,KAAK,WAAa,KAAK,UACnB,aAAc,CAAC,KAAK,SAAU,KAAK,WAAY,KAAK,MACpD,KAAK,sBAAuB,KAAK,sBAAuB,GACxD,KAAK,sBACT,KAAK,MAAQ,IAKL,EAAA,UAAA,6BAAV,SAAuC,KAEvC,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAK,GAAK,SAGR,GAAS,GAAoB,GACtB,EAAA,SAAS,EAAQ,EAAA,UAAU,KAH3B,QAQb,EAAA,UAAA,mBAAA,SAAmB,GAEjB,GADA,EAAa,GAAmB,GAC5B,KAAK,aAAe,KACtB,MAAW,GAAU,OAAA,CAAE,KAAK,YAG9B,GAAM,GAAmB,GAAqB,KAAK,aACnD,GAAI,EAAO,SAAW,EAAW,OAAS,EACxC,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IAG7B,OADI,GAAI,EACC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAK,EAAO,GACZ,EAAK,EAAW,EAAI,GAC1B,GAAK,GAAM,MAAU,GAAM,MAAU,IAAO,EAC1C,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IACxB,AAAI,GAAM,MACf,GAAO,GAAK,GAEd,IAGJ,MAAA,CAAQ,EAAW,IAAE,OAAK,EAAM,CAAE,KAAK,aAGzC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,AAAI,EAAM,QAAU,SAClB,GAAQ,GAAO,EAAO,UAExB,GAAM,GAAS,GAAS,EAAK,WAAW,OAAQ,EAAM,QACtD,MAAO,GAAO,QACV,GAAmB,EAAK,mBAAmB,EAAM,YAIzD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,UAAW,KAAK,UAChB,sBAAuB,GAAqB,KAAK,uBACjD,sBAAuB,GAAqB,KAAK,uBACjD,oBAAqB,GAAqB,KAAK,qBAC/C,qBAAsB,GAAoB,KAAK,sBAC/C,SAAU,KAAK,SACf,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjIF,EAAA,UAAY,YAmIrB,GArI+B,IAsI/B,EAAA,cAAc,cAAc,ICjN5B,AA+BA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAGlC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAQ,KAAG,KACjB,SAAK,gBAAkB,KAOf,SAAA,UAAA,cAAV,SAAwB,GACtB,KAAM,IAAI,KAaJ,EAAA,UAAA,gCAAR,SAAwC,EAAe,GACrD,GAAI,GAAU,MAAQ,GAAU,KAC9B,MAAO,MACF,GAAI,EAAO,OAAS,EAAO,OAChC,MAAO,MAAK,gCAAgC,EAAQ,GAC/C,GAAI,EAAO,SAAW,EAC3B,MAAO,GAGT,OADM,GAAqB,EAAO,MAAM,EAAG,EAAO,OAAS,EAAO,QACzD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,EAAO,OAAS,EAAO,OAAS,GAC3C,EAAI,EAAO,GACjB,GAAI,GAAK,MAAQ,GAAK,MAAQ,EAAI,GAAK,EAAI,EACzC,EAAY,KAAK,cACR,IAAM,EACf,EAAY,KAAK,WACR,IAAM,EACf,EAAY,KAAK,QAEjB,GAAI,IAAM,EACR,KAAM,IAAI,GACN,wDACA,KAAK,UAAU,GAAU,IAAM,KAAK,UAAU,IAEpD,EAAY,KAAK,IAGrB,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GAOJ,GALA,AAAI,MAAM,QAAQ,IAAe,CAAC,MAAM,QAAQ,EAAW,KAEzD,GAAa,CAAC,GAAmB,KAEnC,EAAa,EACT,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,mEACA,SAAQ,EAAW,OAAM,eAM/B,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAI1B,GADA,EAAa,GAAqB,GAC9B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,qDACA,6BAA4B,KAAK,UAAU,GAAW,MAK5D,OAFI,GACA,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GAC9C,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,GAAM,GAAW,EAAW,IAAI,SAAA,GAAS,MAAA,GAAM,SAC/C,AAAI,EAAW,QAAQ,QAAU,IAC7B,GAAqB,GAAU,SAAW,EAC5C,KAAK,gBAAkB,GAEvB,KAAK,gBAAkB,IAI3B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAK,iBACP,GAAM,GAA2B,GAC3B,EAAY,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,OAC5C,GAAI,EAAU,QAAQ,QAAU,IAI9B,OADM,GAAU,GAAc,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAEZ,OAFO,GAAC,EAAA,GACF,EAAQ,EAAE,KACP,EAAI,EAAG,EAAI,EAAU,EAAO,EAAE,EACrC,EAAI,GAAa,EAAG,GAEtB,EAAe,KAAK,GAEtB,MAAO,GAAK,cAAc,QAK1B,OADI,GAAa,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,KAChB,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAY,EAAO,GACnB,EAAW,EAAO,MAAM,GAAG,OAAO,CAAC,IACrC,EAAc,EAAE,QAChB,CAAC,GAAW,OAAO,GAAoB,EAAO,MAAM,MACxD,EAAc,EAAA,UAAc,EAAa,CAAC,EAAG,IAC7C,EAAc,EAAY,QAAQ,GAClC,EAAe,KAAK,GACpB,EAAa,WACJ,EAAQ,GACjB,GAAM,GAAO,GAAgB,EAAG,GAAO,OAAO,CAAC,IAC/C,EAAe,KAAK,EAAA,UAAc,EAAG,IACrC,EAAa,OAGb,GAAe,KAAK,GAGxB,GAAI,GAAI,EAAK,cAAc,GACrB,EAAQ,EAAE,KAChB,GAAI,GAGF,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAQ,EAAO,OACf,EAAY,EAAO,EAAQ,GAC3B,EACF,CAAC,GAAW,OAAO,EAAO,MAAM,EAAG,EAAO,OAAS,IACvD,EAAI,EAAA,UAAc,EAAE,QAAQ,CAAC,GAAI,IAAa,CAAC,EAAG,IACzC,QAAQ,WACR,EAAQ,GACjB,GAAM,GAAO,CAAC,EAAQ,GAAG,OAAO,GAAgB,EAAG,EAAQ,IAC3D,EAAI,EAAA,UAAc,EAAG,IAGzB,MAAO,QAGT,OAAO,GAAK,cAAc,MAKhC,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,EACb,GAAI,GACJ,AAAI,EAAW,IAAM,KACnB,EAAc,KAEd,EAAc,EAAW,GAAG,MAAM,GAEpC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAG1B,SAAa,GAAqB,GAClC,AAAI,EAAW,SAAW,EACxB,EAAc,EAAW,OAAO,GAEhC,EAAc,CAAC,MAAM,OAAO,GAEvB,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6BAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+BAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,kGAEA,KAAI,EAAO,OAAM,OAAO,EAAK,OAAM,MAEzC,GAAI,EAAK,MAAM,SAAA,GAAK,MAAA,IAAK,OACvB,MAAO,MAET,EAAO,EAAK,IAAI,SAAA,GAAK,MAAA,IAAK,KAAO,EAAI,EAAA,WAAe,EAAG,KAEvD,OADI,GAAS,EAAK,GACT,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAS,EAAA,WAAe,EAAQ,EAAK,IAEvC,MAAO,MAGb,GAlOoC,IAoOpC,GAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,MAcrB,GAhByB,IAiBzB,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAG5B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,WAcrB,GAhB8B,IAiB9B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,GAAA,IAAQ,EAAI,EAAO,OAAQ,MAX/B,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA2D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IAiE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAM/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAJJ,SAAA,aAAe,GAKtB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,MACxD,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,yEAGN,EAAa,EAGb,OADI,GAAe,GACC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,GAAS,MACX,EAAe,GACf,OAGJ,GAAI,EACF,OAIF,OADM,GAAoB,GACjB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAyB,EAAW,GAAG,QAC7C,EAAuB,OAAO,KAAK,KAAM,GAEzC,OADI,GAAS,GACO,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAA,KAAK,YAAY,EAAO,IAC1B,EAAS,GACT,OAGJ,AAAK,GACH,EAAS,KAAK,GAGlB,GAAI,EAAS,OAAS,EACpB,KAAM,IAAI,GACN,4GAEA,KAAK,UAAU,KAIb,EAAA,UAAA,cAAV,SAAwB,GAAxB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAO,IAAc,EAAQ,EAAK,SAItC,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,KAC1D,KAAM,IAAI,GACN,+DAON,OALM,GAAc,EACd,EAAc,EAAY,GAAG,QAC7B,EAAO,KAAK,KAAO,EAAI,EAAY,OAAS,KAAK,KAAO,KAAK,KAG/C,EAAA,EAAA,EAAA,EAAY,MAAM,GAAlB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAY,IAAS,MAAQ,EAAM,IAAS,MAC9C,EAAY,GAAQ,KACpB,MAEF,EAAY,IAAS,EAAM,GAE7B,MAAO,IAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6CAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+CAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,mCAAmC,EAAK,OAAM,KAC9C,8BAA6B,EAAO,OAAM,MAEhD,MAAO,GAAA,KAAS,WACd,GAAI,GAAe,GAOnB,GANA,EAAK,QAAQ,SAAA,GACX,GAAI,GAAK,MACP,EAAe,GACf,UAGA,EACF,MAAO,MAGT,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAK,IAAM,KAEb,EAAY,KAAK,EAAA,SAAa,EAAO,IAAI,OAAO,SAC3C,AAAI,EAAK,GAAG,KAAO,EAAO,GAAG,KAElC,EAAY,KAAK,EAAA,WAAe,EAAK,GAAI,KAEzC,EAAY,KAAK,EAAK,IAG1B,GAAM,GAAoB,EAAA,OAAW,EAAa,EAAK,MACvD,MAAO,GAAA,IAAQ,EAAmB,GAAI,OAI1C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,MAET,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArIF,EAAA,UAAY,cAuIrB,GAzIiC,IA0IjC,EAAA,cAAc,cAAc,IAuF5B,YAAuB,EAAc,GACnC,KAAO,EAAO,GACZ,GAAQ,EAEV,MAAO,GAGT,YAAkB,EAAW,EAAW,GACtC,GAAI,EAAE,MAAM,OAAS,GAAK,EAAE,MAAM,OAAS,EACzC,KAAM,IAAI,IACN,oEAeN,GAbA,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAC3B,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAE3B,AAAI,MAAO,IAAS,UAClB,GAAO,CAAC,EAAM,IAGZ,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,KAAM,IAAI,IACN,+DAGN,GAAM,GAAQ,EAAE,MAAM,OAChB,EAAQ,EAAE,MAAM,OACtB,AAAI,GAAQ,MAEV,GAAO,CAAC,EAAQ,EAAG,EAAQ,IAE7B,GAAM,GAAY,EAElB,MAAO,GAAA,KAAS,WACd,GAAI,GACJ,GAAI,EAAQ,GACV,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,YACpB,EAAQ,GACjB,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,QAE7B,GAAO,EAGT,GAAI,GACJ,GAAI,EAAE,MAAM,SAAW,GAAK,EAAE,MAAM,SAAW,EAC7C,AAAI,EAAU,KAAO,EAAU,GAC7B,EAAM,EAAE,IAAI,GAAG,IAAI,EAAU,IAE7B,EAAM,EAAE,UAAU,CAAC,EAAG,IAAI,IAAI,GAAG,IAAI,EAAU,SAGjD,GAAM,GAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EACzC,EAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EAC/C,EAAM,EAAE,OAAO,EAAG,EAAM,GAG1B,GAAI,EAAO,GACT,GAAI,GAAG,OACP,AAAI,EAAQ,EACV,EAAM,EAAQ,EAAQ,EAEtB,EAAM,EAAQ,EAGhB,OADM,GAAwB,GACrB,EAAI,EAAK,EAAI,EAAM,EAAM,EAAE,EAClC,EAAY,KAAK,GAEnB,EAAM,EAAI,QAAQ,GAEpB,MAAI,GAAI,MAAM,SAAW,GACvB,GAAM,EAAI,WAAW,IAEhB,IAIX,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,EAAK,KACjB,EAAK,UAAY,EAAK,WAAa,KAAO,GAAQ,EAAK,UACvD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAS,EAAW,GACpB,EAAS,EAAW,GAC1B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,GAAI,EAAO,EAAK,MAAQ,EAAO,EAAK,IAClC,KAAM,IAAI,GACN,8BACG,GAAO,EAAK,IAAG,QAAQ,EAAO,EAAK,OAIpC,EAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,qDACA,iBAAgB,EAAO,OAAM,eAGnC,GAAI,GAAK,EAAO,GACZ,EAAK,EAAO,GACZ,EACJ,MAAK,OAAM,QAAQ,KAAK,MAMtB,EAAO,KAAK,KAAK,IACN,SAAC,EAAM,GAAM,MAAA,IACT,EAAM,EAAO,GAAG,MAAM,UAPrC,EAAO,CACL,GAAc,KAAK,KAAM,EAAG,MAAM,QAClC,GAAc,KAAK,KAAM,EAAG,MAAM,SAOtC,AAAI,KAAK,WACP,GAAK,GAAY,EAAI,EAAK,IAC1B,EAAK,GAAY,EAAI,EAAK,KAErB,GAAS,EAAI,EAAI,IAGlB,EAAA,UAAA,cAAR,SAAsB,EAAe,GACnC,GAAI,GACJ,MAAK,OAAM,QAAQ,KAAK,MAQtB,EAAO,KAAK,KANZ,EAAO,CACL,GAAc,KAAK,KAAM,EAAO,QAChC,GAAc,KAAK,KAAM,EAAO,SAM7B,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAU,EAAW,GAAa,QAClC,EAAU,EAAW,GAAa,QACxC,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAG,GACjB,GAAM,GAAc,EAAO,OAAO,GAClC,MAAI,GAAY,SAAW,GACzB,EAAY,KAAK,GAEZ,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,OAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,KACb,UAAa,KAAK,WAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA7GF,EAAA,UAAY,MA+GrB,GAjHyB,IAkHzB,EAAA,cAAc,cAAc,IChjC5B,AA2BA,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAKjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,OAAS,EAAK,SAGrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,OAAQ,KAAK,QAC7B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAS,WACX,MAAA,IAAe,EAAM,MAAO,EAAG,EAAK,QAAQ,IAAI,IAC9C,EACF,GAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAC9D,MAAO,MA5BJ,EAAA,UAAY,gBA+BrB,GAjCmC,IAkCnC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAS,WACb,GAAM,GAAS,KAAK,KAAK,EAAK,KAAQ,GAAI,EAAK,OAC/C,MAAO,GAAM,IAAI,GAAe,EAAM,MAAO,EAAG,KAElD,MAAO,IAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAEnE,MAAO,MA/BJ,EAAA,UAAY,kBAkCrB,GApCqC,IAqCrC,EAAA,cAAc,cAAc,IAyC5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAMhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,KACjB,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,YAAc,GAAoB,GAAQ,OAGxD,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAa,EAAK,eAAe,GAEjC,EAAgB,WACpB,GAAM,GAAQ,GAAoB,GAE5B,EAAQ,mBACR,EAAQ,mBAER,EAAS,CAAC,EAAQ,EAEpB,EAAU,EAAA,aAAa,EAAA,cAAc,GAAa,EAAK,MAE3D,EAAU,GAAO,EAAS,WAG1B,GAAM,GAAI,KAAA,IAAE,GAAI,EAAK,MAAS,GAAI,EAAK,KAAO,KAAA,IAAA,EAAU,IAAO,KACzD,EAAI,CAAC,EAAI,EAAS,EAAK,KAGvB,EAAI,EAAM,IAAI,GAAS,IAAI,EAAQ,IAAI,IAAI,IAAI,IAErD,MAAO,GAAE,IAAI,GAAG,IAAI,IAEtB,MAAO,IACH,EAAe,WAAM,MAAA,IAAoB,IACzC,EAAO,UAAe,IAE5B,MAAO,MAxDJ,EAAA,UAAY,eA2DrB,GA7DkC,IA8DlC,EAAA,cAAc,cAAc,IChN5B,YA4CI,EAAW,EAAc,EAAkB,EAAe,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GACJ,GAAI,EAAE,OAAS,EACb,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EAEpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EACpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,OAElC,MAAM,IAAI,IACN,2DAA2D,EAAE,KAAI,QAGvE,MAAO,GAoBT,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WACH,GAAM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EACF,GAAmB,EAAG,EAAM,EAAU,EAAM,EAAO,GACvD,MAAO,CAAC,EAAQ,EAAM,KAqBjC,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKH,OAJM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EAAwB,GACX,EAAA,EAAA,EAAA,GAAiB,EAAG,EAAE,MAAtB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAc,QAAQ,KAAU,GAClC,EAAY,KAAK,GAEjB,EAAY,KAAK,EAAE,MAAM,IAG7B,GAAM,GAAgB,EAAK,QAAQ,GAC7B,EAAoB,EAAS,QAAQ,GACrC,EACF,GAAS,KAAO,KAAO,EAAM,QAAQ,GACnC,EACF,GAAQ,KAAO,KAAO,EAAK,QAAQ,GACjC,EAAS,GACX,EAAG,EAAe,EAAmB,EACrC,EAAgB,GACpB,MAAO,CAAC,EAAQ,EAAM,iBAgB7B,EAAW,EAAe,EAAc,EACxC,GACF,MADE,AAAA,KAAA,QAAA,GAAA,MACE,EAAA,KAAK,YACD,EAAc,QAAQ,OAAQ,GAAiB,EAAG,EAAE,KAAO,IAC1D,GACH,EAAG,EAAO,EAAM,EAAe,GAE5B,GACH,EAAG,EAAO,EAAM,EAAe,GAsFvC,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAqBtC,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,IAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KAC1C,EAAK,SAAW,EAAK,UAAY,KAAO,IAAO,EAAK,SACpD,EAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,sBACD,GAAe,EAAK,uBAAyB,SACjD,EAAK,0BACD,GAAe,EAAK,2BAA6B,QACrD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAc,EAAK,iBAC1C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,oBAGvC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAO,KAAK,MAAQ,EAAI,KAAK,KAAQ,KAAK,KAAO,EAAW,OAC5D,EAAM,EAAW,GACvB,GAAI,GAAO,KACT,KAAM,IAAI,GACN,QAAQ,EAAI,+FAET,MAAK,UAAU,GAAW,MAEnC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAW,OAAQ,KAAI,GAAA,GAAG,EAAC,GAAO,EAAG,MAC/D,GAAM,GAAQ,CAAC,GACf,AAAI,KAAK,OACP,MAAK,MAAQ,KAAK,UACd,QAAS,EAAO,KAAM,KAAK,iBAAkB,KAAK,iBAClD,GAAM,KAAK,kBAEjB,AAAI,KAAK,QACP,MAAK,KAAO,KAAK,UACb,OAAQ,EAAO,KAAM,KAAK,gBAAiB,KAAK,gBAAiB,GACjE,KAAK,iBAEX,KAAK,WAAa,KAAK,UACnB,cAAe,EAAO,KAAM,KAAK,sBAAuB,KAAM,IAClE,KAAK,eAAiB,KAAK,UACvB,kBAAmB,EAAO,KAAM,KAAK,0BAA2B,KAChE,IACJ,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACvD,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAO,EAAW,OAClB,EAAgB,GAAiB,EAAG,GACpC,EAAO,EAAK,MAAQ,EAAI,EAAK,KAAQ,EAAK,KAAO,EACvD,EAAc,OAAO,EAAM,GAC3B,GAAM,GAAiB,GAA2B,EAAG,GACrD,EAAe,GAAQ,EAAW,GAElC,GAAM,GAAsB,EAAc,QAC1C,EAAoB,OACpB,GAAM,GAAoB,CAAC,EAAA,KAAK,YAC5B,EAAqB,GAAiB,EAAG,GAAM,MAAM,EAAG,EAAO,IAE7D,EAAmC,WACvC,GAAI,GACF,GAAM,GACF,EAAK,WAAW,OAAO,QAAQ,GAC7B,EACF,EAAK,eAAe,OAAO,QAAQ,GACjC,EACF,EAAK,OAAS,EAAK,KAAK,OAAO,QAAQ,GAAkB,KACvD,EACF,EAAK,MAAQ,EAAK,MAAM,OAAO,QAAQ,GAAkB,KAC7D,MAAO,IACH,EAAO,EAAqB,EAC5B,EAAe,EAAgB,EAAK,aAExC,OAAO,IACH,EAAO,EAAK,WAAW,OAAQ,EAAK,eAAe,OACnD,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACrC,EAAK,OAAS,KAAO,KAAO,EAAK,MAAM,OAAQ,EAAK,UAI5D,GAAI,CAAC,EACH,MAAO,KAGH,GAAA,GAAA,GAAA,EAAA,EAAA,MAAA,OAAA,EAAA,KAAA,OAAA,EAAA,EAAA,SAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAM,EAAA,EAAA,GAIvB,EACF,SAAC,EAAyB,EAAe,GACvC,EAAA,KAAS,WACP,GAAM,GAAQ,EAAI,EACZ,EAAY,EAAS,OACrB,EAAc,EAAU,IAAI,GAAO,IAAI,GAC7C,EAAS,MAAM,EAAU,IAAI,OAU/B,EAA8B,WAClC,EAAgB,EAAK,WAAY,EAAM,EAAK,UAC5C,EAAgB,EAAK,eAAgB,EAAU,EAAK,WAEtD,WAEO,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,SAAU,KAAK,SACf,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,sBAAuB,GAAqB,KAAK,uBACjD,0BACI,GAAqB,KAAK,2BAC9B,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,eAAgB,GAAoB,KAAK,gBACzC,gBAAiB,GAAoB,KAAK,kBAEtC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArKF,EAAA,UAAY,qBAuKrB,GAzKwC,IA0KxC,EAAA,cAAc,cAAc,IAkD5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAgBtC,WAAY,GAAZ,GAAA,GAAA,KAOE,GANA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KACtC,MAAO,GAAK,MAAS,UACvB,GAAI,CAAC,OAAO,UAAU,EAAK,MACzB,KAAM,IAAI,OACN,gDAAgD,EAAK,cAElD,MAAM,QAAQ,EAAK,MAC5B,OAAmB,GAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,6CACA,iBAAgB,KAAK,UAAU,EAAK,YAI5C,MAAM,IAAI,OACN,2DACA,iBAAgB,KAAK,UAAU,EAAK,QAG1C,SAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,kBAE5C,EAAK,gBAAkB,KAGlB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAQ,EAAW,OAGzB,AAAI,MAAO,MAAK,MAAS,UACvB,MAAK,KAAO,CAAC,KAAK,OAEpB,OAAS,GAAI,EAAG,EAAI,KAAK,KAAK,OAAQ,EAAE,EACtC,AAAI,KAAK,KAAK,GAAK,GACjB,MAAK,KAAK,IAAM,GAKpB,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,GAAK,GAAQ,EACtB,KAAM,IAAI,OAAM,iBAAiB,GAGrC,GAAI,KAAK,KAAK,SAAW,GAAqB,KAAK,MAAM,OACvD,KAAM,IAAI,OAAM,4BAA4B,KAAK,MAGnD,GAAM,GAAa,KAAK,KAAK,IAAI,SAAA,GAAQ,MAAA,GAAW,KAE9C,EAAY,GAClB,AAAI,KAAK,MACP,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAE3B,KAAK,MAAQ,KAEf,AAAI,KAAK,OACP,KAAK,KAAO,KAAK,UACb,OAAQ,EAAY,UAAW,KAAK,gBACpC,KAAK,gBAAiB,GAE1B,KAAK,KAAO,KAGd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACQ,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAQ,EAAW,OAEzB,MAAO,GAAA,KAAK,WAIV,OAHM,GAAW,GACb,EAAA,EAAA,QAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,SACL,EAAiB,GAA2B,EAAG,GACnC,EAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAe,GAAO,EAAW,GAuBnC,OApBM,GAAY,SAAC,GACjB,MAAI,IAAK,MAAQ,EAAE,MAAM,SAAW,GAChC,EAAK,OAAS,CAAC,EAAQ,GAClB,EAAE,QAAQ,GAEV,GAIP,EAAQ,EAAU,EAAK,MAAM,QAC7B,EAAS,EAAU,EAAK,KAAK,QAQ3B,EAA0B,GAC1B,EAA8B,GAC3B,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,AAAK,EAAK,KAAkB,QAAQ,KAAO,GACzC,GAAc,KAAK,EAAW,IAC9B,EAAkB,KAAK,IAEvB,GAAc,KAAK,GACnB,EAAkB,KAAK,EAAW,KAGtC,SAAO,EAAK,KAAK,GACjB,EAAW,EAAS,KAAK,GACzB,EAAQ,EAAM,KAAK,GACnB,EAAS,EAAO,KAAK,GAEd,GACH,EAAO,EAAM,EAAU,EAAQ,EAAO,EAAK,YAInD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,mBAExC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjKF,EAAA,UAAY,qBAmKrB,GArKwC,IAsKxC,EAAA,cAAc,cAAc,IC9nB5B,YAoEI,EAAW,EACX,GACF,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,kEACG,GAAE,KAAI,eAMf,GAHA,AAAI,GAAW,MACb,GAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,KAErB,EAAQ,SAAW,GAAK,EAAQ,GAAG,SAAW,GAC9C,EAAQ,GAAG,SAAW,EACxB,KAAM,IAAI,GACN,+GAON,GAHA,AAAI,GAAc,MAChB,GAAa,MAEX,IAAe,gBAAkB,IAAe,gBAClD,KAAM,IAAI,GACN,wBAAwB,EAAU,mEAIxC,GAAI,GACJ,MAAI,KAAe,gBACjB,EAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,IAE/C,EAAU,CAAC,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,GAAI,CAAC,EAAG,IAG1C,EAAA,IAAQ,EAAG,KA6BtB,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAMjC,WAAY,GAAZ,GAAA,GAAA,KAUE,GATA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,WACD,EAAK,YAAc,KAAO,KAAoB,EAAK,WAGnD,EAAK,SAAW,KAClB,EAAK,QAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,YACnB,MAAO,GAAK,SAAY,SACjC,EAAK,QACD,CAAC,CAAC,EAAK,QAAS,EAAK,SAAU,CAAC,EAAK,QAAS,EAAK,eAGvD,GADA,EAAK,QAAU,EAAK,QAChB,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,6DACA,sBAAqB,EAAK,QAAQ,OAAM,YAG9C,GAAI,GAAa,OACb,EAAY,OAChB,GAAI,MAAO,GAAK,QAAQ,IAAO,SAC7B,EAAgB,CAAC,EAAK,QAAQ,GAAI,EAAK,QAAQ,IAC/C,EAAe,CAAC,EAAK,QAAQ,GAAc,EAAK,QAAQ,SAIxD,GAFA,EAAK,QAAU,EAAK,QAEhB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,gEACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAIrD,GAFA,EAAgB,EAAK,QAAQ,GAEzB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,+DACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAErD,EAAe,EAAK,QAAQ,GAE9B,EAAK,QAAU,CAAC,EAAe,GAEjC,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAEhC,GAAI,GACA,EACJ,MAAI,MAAK,aAAe,gBACtB,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,IAE5C,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,MAIlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,IACF,GAAoB,GAAS,EAAK,QAAS,EAAK,eAG1D,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAnGF,EAAA,UAAY,gBAqGrB,GAvGmC,IAwGnC,EAAA,cAAc,cAAc,IC5O5B,YA0CI,EAAW,EAA4B,EACvC,EAAuB,EACvB,GACF,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,IAEhB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAKb,EAAI,GAAsB,EAAG,GAC7B,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MAEf,EAAI,EAAA,QAAY,EAAe,EAAU,EAAS,GAIlD,EAAI,EAAA,QAEA,EAA0B,EAAU,EAAS,GAEnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBAeP,EAAa,EACb,EAAoC,EACpC,EAAyB,GAC3B,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,EAAG,IAEnB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAIb,EAAI,GAAsB,EAAa,GACvC,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MACf,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAExC,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAE1C,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAsBX,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAWtC,WAAY,GAAZ,GAAA,GAAA,KAKE,GAJA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,GAElB,EAAA,EAAA,KAAA,KAAM,IAAK,KACP,MAAO,GAAK,UAAa,SAC3B,EAAK,SAAW,CAAC,EAAK,kBAEpB,MAAM,QAAQ,EAAK,WAClB,EAAK,SAAsB,SAAW,GACvC,MAAQ,GAAK,SAAsB,IAAO,SAC5C,EAAK,SAAW,EAAK,aAErB,MAAM,IAAI,GACN,qGAEA,IAAG,KAAK,UAAU,EAAK,YAG7B,GADA,GAAsB,EAAK,SAAU,YACjC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBAEhB,MAAO,GAAK,SAAY,SAC1B,EAAK,QAAU,CAAC,EAAK,iBAEnB,MAAM,QAAQ,EAAK,UAClB,EAAK,QAAqB,SAAW,GACtC,MAAQ,GAAK,QAAqB,IAAO,SAC3C,EAAK,QAAU,EAAK,YAEpB,MAAM,IAAI,GACN,oGAEA,IAAG,KAAK,UAAU,EAAK,WAG/B,UAAsB,EAAK,QAAS,WAEpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAS,GACX,EAAW,GAAI,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IAChE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAW,KAO5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,EAAS,GAAa,GAAoB,GAAS,GACnD,GAAM,GAAS,EAAK,gBAChB,GAAoB,GAAS,CAAC,EAAK,SAAS,GAAI,GAChD,CAAC,EAAK,QAAQ,GAAI,GAAI,EAAK,QAAS,gBAExC,MAAO,GAAA,QAAY,EAAQ,CAAC,OAIhC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,SAEV,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAxFwC,IA0FxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,IAEtB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,UACrB,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,SAErC,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAKrE,MAJA,GACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,GAErC,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,KAQlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAhFwC,IAkFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,EAAG,IAEzB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,SAAU,EAAK,UACpC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,QAAS,EAAK,SAEnD,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAOrE,MANA,GAAS,GACL,EAAQ,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,EAAM,GAE7C,CAAC,EAAW,GAAI,EAAQ,EAAM,EAAM,EAAW,KAS1D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GArFwC,IAuFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,eAcrB,GAhBkC,IAiBlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,mBAcrB,GAhBsC,IAiBtC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAC5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,EAAW,KAGpC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAEd,GAb8C,IAe9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAG1C,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,KAAS,EAAO,MARpB,EAAA,UAAY,yBAWrB,GAb4C,IAc5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAGtC,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,IAAQ,EAAO,MARnB,EAAA,UAAY,qBAWrB,GAbwC,IAcxC,EAAA,cAAc,cAAc,IAiB5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAE5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GAEjB,MADA,GAAa,EACT,KAAK,aAAe,eACf,CAAC,EAAW,GAAI,EAAW,IAE3B,CAAC,EAAW,GAAI,EAAW,KAItC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAGZ,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,KAAK,YAC3B,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA7B8C,IA+B9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAA5C,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,KAAS,EAAO,CAAC,EAAG,IAEpB,EAAA,KAAS,EAAO,CAAC,EAAG,OAR1B,EAAA,UAAY,yBAYrB,GAd4C,IAe5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAAxC,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,IAAQ,EAAO,CAAC,EAAG,IAEnB,EAAA,IAAQ,EAAO,CAAC,EAAG,OARzB,EAAA,UAAY,qBAYrB,GAdwC,IAexC,EAAA,cAAc,cAAc,IC5rB5B,AA4CA,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,GAAZ,GAAA,GAQE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAKf,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAIE,MAAI,MAAK,OAAS,KACT,KAAK,MAAM,UAEX,QAIX,SAAc,GAIZ,AAAI,KAAK,OAAS,MAChB,MAAK,MAAM,UAAY,oCAI3B,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,MAAM,kDAIpB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAIpB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WAEE,MAAQ,MAAK,MAAc,0CAK7B,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,MAAM,wCAKpB,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,MAAM,cAGpB,EAAA,UAAA,WAAA,SAAW,GACT,KAAK,MAAM,WAAW,IAGxB,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAS,CACP,UAAa,KAAK,MAAM,eACxB,OAAU,KAAK,MAAM,cAGnB,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,OAAS,MAChB,KAAK,MAAM,6BAA6B,IAKrC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAc,EAAO,MACrB,EAAQ,GAAY,EAAa,GACvC,MAAO,GAAO,MACd,GAAM,GAAY,CAAC,MAAK,GACxB,cAAO,OAAO,EAAW,GAClB,GAAI,GAAI,IAEnB,GAvGsC,IAyGtC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,oEACA,gBAAe,KAAK,UAAU,KAEpC,KAAK,UAAY,CAAC,CAAC,MAAO,IAC1B,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAChE,AAAK,KAAK,MAAM,OACd,MAAK,MAAM,MAAM,GACjB,KAAK,MAAM,MAAQ,IAErB,EAAA,UAAM,MAAK,KAAA,KAAC,IAGd,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAC1D,EACF,KAAK,MAAM,mBAAmB,GAC5B,EAAY,EAAW,GAC7B,MAAO,CAAC,EAAiB,GAAI,GAAW,OAAO,EAAiB,MAAM,KAGxE,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,EAAS,GAAoB,GAI7B,GAAM,GAAwB,SAAC,EAAgB,GAK7C,GAAM,GAAS,GAAoB,EAAK,MAAM,KAAK,EAAQ,IAC3D,MAAO,CAAC,EAAQ,KAEZ,EACF,GAAI,EAAM,EAAQ,GAAI,GAAyB,KAC3C,KAAsB,GACtB,IACF,EAAI,EAAW,GAGrB,MAAO,MArDJ,EAAA,UAAY,kBA0DrB,GA5DqC,IA6DrC,EAAA,cAAc,cAAc,gBAEgB,GAC1C,GACI,GAAiC,yBAA0B,GAmBjE,GAAM,IAA2D,SAEjE,GAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAUL,EAAc,EAAK,MAAM,YACzB,EAAqC,GAC3C,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,aAAe,GAAY,GAChC,EAAY,YACR,IAAY,cAAmB,IACnC,GAAM,GAAqC,GAW3C,GAVA,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,cAAgB,GAAY,GACjC,EAAK,aAAa,KAAO,WAAa,EAAK,aAAa,KACxD,EAAK,cAAc,KAAO,YAAc,EAAK,cAAc,KAE3D,EAAK,UAAY,EAAK,YAAc,OAChC,GACA,EAAK,UACT,GAA4B,EAAK,WAC7B,EAAK,QACP,KAAM,IAAI,IACN,mEAEN,SAAK,UAAY,EAAK,MAAM,SAC5B,EAAK,gBAAkB,EAAK,MAAM,gBAClC,EAAK,YAAc,EAAK,MAAM,YAC9B,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAClB,EAAK,UAAY,EAAK,MAAM,UAC5B,EAAK,aAAe,OAGtB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GAIZ,KAAK,WAAa,EAClB,AAAI,KAAK,cAAgB,MACvB,MAAK,aAAa,UAAY,GAEhC,AAAI,KAAK,eAAiB,MACxB,MAAK,cAAc,UAAY,oCAInC,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,aAAa,aAAa,OAClC,KAAK,cAAc,eAGzB,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAa,EAAQ,OACrB,EAAiB,KAAK,MAAM,EAAa,GAC/C,KAAK,aAAa,WAAW,EAAQ,MAAM,EAAG,IAC9C,KAAK,cAAc,WAAW,EAAQ,MAAM,KAG9C,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GACA,KAAK,aAAa,mBAAmB,GACzC,AAAM,MAAM,QAAQ,IAAgB,MAAM,QAAQ,EAAY,KAC5D,GAAc,CAAC,IAEjB,EAAc,EAEd,GAAI,GACA,EACA,EAiBJ,MAhBA,AAAI,MAAK,aACP,GAAa,EAAY,MAAM,IAC/B,EAAc,EAAY,GAI5B,EAAc,EACd,AAAI,KAAK,YAAc,SACrB,GAAY,EAAY,OAAS,IAAM,EACvC,EAAe,CAAC,IACX,AAAI,KAAK,WAAa,KAC3B,EAAe,CAAC,EAAa,EAAY,SAEzC,EAAe,CAAC,GAGd,KAAK,YACH,KAAK,WAAa,KACb,EAAa,OAAO,GAAY,OAAO,EAAW,SAEpD,CAAC,GAAa,OAAO,GAAY,OAAO,EAAW,SAErD,GAA+B,IAGxC,EAAA,UAAA,MAAA,SACI,EACA,GACF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAEX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAU1D,GATA,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAEzB,AAAI,MAAM,QAAQ,IAChB,GAAgB,EAAuC,MAAM,GAC7D,EAAU,EAAuC,IAG9C,IAAgB,MAAQ,EAAa,SAAW,IACjD,GAAa,KACf,MAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,GAE7B,GAAM,GAAiD,GACjD,EAA+B,GACrC,GAAI,GAAgB,MAClB,GAAM,GAAY,EAAa,OAC/B,GAAI,EAAY,EAAI,EAClB,KAAM,IAAI,GACN,iIAIN,EAAO,aAAkB,EACzB,EAAiB,KAAI,MAArB,EAAyB,GACzB,GAAM,GAAc,EACI,IAAI,SAAA,GAAS,MAAA,IAAI,IAAU,CAAC,MAAO,EAAM,UACjE,KAAK,aAAa,UAAY,EAAW,MAAM,EAAG,EAAY,GAC9D,KAAK,cAAc,UAAY,EAAW,MAAM,EAAY,GAC5D,EAAgB,KAAI,MAApB,EAAwB,GAE1B,GAAI,GAAa,KACf,KAAM,IAAI,IACN,yEAKN,OADM,GAAmB,EAAiB,YAAc,IACnC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,YAAkB,MAAmB,EACvC,KAAM,IAAI,GACN,gHAKR,GAAI,GAEF,GAAM,GAAY,CAAC,GAAQ,OAAO,GAC5B,EAAgB,KAAK,UAAU,OAAO,GAUtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GACF,EAAA,UAAM,MAAK,KAAA,KAAC,EAA0C,GAC1D,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAI/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAe,EAAO,aAExB,EACA,EACJ,GAAI,GAAgB,KAClB,EAAI,EAAK,aAAa,KAAK,EAAQ,GACnC,EAAO,EAAK,cAAc,KAAK,EAAQ,QAEvC,GAAM,GAAe,EAAa,MAAM,EAAG,EAAa,OAAS,GAC3D,EAAgB,EAAa,MAAM,EAAa,OAAS,GAC/D,EAAI,EAAK,aAAa,KAClB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KACjD,EAAO,EAAK,cAAc,KACtB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KAGnD,GAAI,GACJ,AAAI,EAAK,aACP,CAAI,MAAM,QAAQ,IAChB,GAAS,EAAE,MAAM,GAAG,OAAQ,EAAkB,MAAM,KAGtD,EAAK,EAAe,GACpB,EAAQ,EAAkB,IAG5B,AAAI,EAAK,iBACP,GAAO,EAAA,QAAY,EAAgB,IAGrC,GAAI,GAcJ,MAbA,AAAI,GAAK,YAAc,SACrB,EAAS,GAAc,CAAC,EAAa,IAChC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,GAAI,EAAA,IAAQ,EAAa,IACrC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,WAAa,MAC3B,GAAS,CAAC,EAAa,IAIrB,EAAK,YACH,EAAK,WAAa,KACZ,EAAoB,OAAO,GAE9B,CAAC,GAAkB,OAAO,GAE5B,KAIX,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,aAAa,cAClB,KAAK,cAAc,eAGrB,EAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,GAAU,KAAK,aAAa,KAAM,WAChC,EAAK,aAAa,MAAM,KAE1B,GAAU,KAAK,cAAc,KAAM,WACjC,EAAK,cAAc,MAAM,KAE3B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,SAAY,EAAyB,GAEnC,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAI,GAcJ,GAbA,AAAI,KAAK,gBACP,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,EAAM,GAEpB,EAAa,EAGf,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,KAAM,MAEpB,EAAa,KAGb,KAAK,aACP,GAAM,GAAS,KAAK,aAAa,OAC3B,EAAsB,EAAO,IAAI,SAAA,GAAS,MAAA,QAChD,MAAI,OAAM,QAAQ,GACT,EAAW,OAAO,GAAW,OAAO,GAEpC,CAAC,GAAY,OAAO,GAAW,OAAO,OAG/C,OAAO,IAIX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,aAAa,iBAAiB,OACtC,KAAK,cAAc,mDAGzB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,aAAa,oBAAoB,OACzC,KAAK,cAAc,sDAKzB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,cAAgB,MACvB,KAAK,aAAa,6BAA6B,GAEjD,AAAI,KAAK,eAAiB,MACxB,KAAK,cAAc,6BAA6B,IAIpD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,UAAa,KAAK,WAGd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAIF,EAAA,WAAP,SACI,EACA,GACF,GAAM,GACF,GAAY,EAAO,OAGvB,GAFA,MAAO,GAAO,MAEV,EAAO,cAAmB,KAC5B,KAAM,IAAI,IACN,4FAIN,GAAM,GAAkC,EACxC,SAAU,MAAW,EACd,GAAI,GAAI,IA7VV,EAAA,UAAY,gBA+VrB,GAjWmC,IAkWnC,EAAA,cAAc,cAAc,IC7kB5B,YAkE2B,GACzB,MAAO,IAAI,IAAW,eA6BJ,GAClB,MAAO,IAAI,IAAI,eAoBI,GACnB,MAAO,IAAI,IAAK,eAuBQ,GACxB,MAAO,IAAI,IAAU,eAwBD,GACpB,MAAO,IAAI,IAAM,eAmBK,GACtB,MAAO,IAAI,IAAQ,eA2BW,GAC9B,MAAO,IAAI,IAAgB,eAyBN,GACrB,MAAO,IAAI,IAAO,eAqBG,GACrB,MAAO,IAAI,IAAO,eAsCY,GAC9B,MAAO,IAAI,IAAgB,eAqBN,GACrB,MAAO,IAAI,IAAO,eAgCY,GAC9B,MAAO,IAAI,IAAgB,eAkCF,GACzB,MAAO,IAAI,IAAW,eA2BK,GAC3B,MAAO,IAAI,IAAa,eAeM,GAC9B,MAAO,IAAI,IAAgB,eAoCF,GACzB,MAAO,IAAI,IAAW,eAoCF,GACpB,MAAO,IAAI,IAAM,eAaK,GACtB,MAAO,IAAI,IAAQ,eAmCY,GAC/B,MAAO,IAAI,IAAiB,eAsBN,GACtB,MAAO,IAAI,IAAQ,eAiBQ,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,eA+BG,GACtB,MAAO,IAAI,IAAQ,eAcK,GACxB,MAAO,IAAI,IAAU,eAyBH,GAClB,MAAO,IAAI,IAAI,eAqBO,GACtB,MAAO,IAAI,IAAQ,eAuBO,GAC1B,MAAO,IAAI,IAAY,eAqBD,GACtB,MAAO,IAAI,IAAQ,eAqBG,GACtB,MAAO,IAAI,IAAQ,eAsBI,GACvB,MAAO,IAAI,IAAS,eAyBF,GAClB,MAAO,IAAI,IAAI,eA0BkB,GACjC,MAAO,IAAI,IAAmB,eAuBG,GACjC,MAAO,IAAI,IAAmB,eA2BF,GAC5B,MAAO,IAAI,IAAc,eAgBM,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eA0BO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAwBO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAYa,GACrC,MAAO,IAAI,IAAuB,eAiBG,GACrC,MAAO,IAAI,IAAuB,eAYD,GACjC,MAAO,IAAI,IAAmB,eAiBG,GACjC,MAAO,IAAI,IAAmB,eAYH,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eA4BN,GAClB,MAAO,IAAI,IAAI,eAgDO,GACtB,MAAO,IAAI,IAAQ,eA0BA,GACnB,MAAO,IAAI,IAAK,eAgDO,GACvB,MAAO,IAAI,IAAS,eA2BI,GACxB,MAAO,IAAI,IAAU,eAgDO,GAC5B,MAAO,IAAI,IAAc,eA8BA,GACzB,MAAO,IAAI,IAAW,eAoCO,GAC7B,MAAO,IAAI,IAAe,eA+DR,GAClB,MAAO,IAAI,IAAI,eAUe,GAC9B,MAAO,IAAI,IAAgB,eAMC,GAC5B,MAAO,IAAI,IAAc,eAiDK,GAC9B,MAAO,IAAI,IAAgB,GAItB,GAAM,IAAkB,GAClB,GAAkB,GAClB,GAAY,GACZ,GAAY,eA2BK,GAC5B,MAAO,IAAI,IAAc,eA2BK,GAC9B,MAAO,IAAI,IAAgB,eAkCA,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,glCClnDU,EAAe,GAC5C,MAAO,IAAuB,EAAO,eAoBJ,EAAe,GAChD,MAAO,IAA2B,EAAO,eAuBvC,EAAe,GACjB,MAAO,IAAkC,EAAO,eAqBd,EAAe,GACjD,MAAO,IAA4B,EAAO,eAcJ,EAAe,GACrD,MAAO,IAAgC,EAAO,eAsCtB,EAAe,GACvC,MAAO,IAAkB,EAAO,eAsCX,EAAe,GACpC,MAAO,IAAe,EAAO,eAwBC,EAAe,GAC7C,MAAO,IAAuB,EAAO,eAuBL,EAAe,GAC/C,MAAO,IAAyB,EAAO,eAsBrC,EAAe,GACjB,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAqBlB,EAAe,GAC9C,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,kSC1TxC,yCCAA,YAqBqB,GACnB,MAAO,IAAI,IAAK,eAYC,GACjB,MAAO,IAAgB,eAYN,GACjB,MAAO,IAAgB,+CChDzB,mBAkBuC,EAAA,EAAA,GAAvC,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KAEE,SAAA,MAAqB,OAErB,SAAA,UAAA,SAAA,SAAS,GACP,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,OAAM,yDAElB,KAAK,MAAQ,GAEjB,GAVuC,IAsEvC,YAAc,EAAiB,GAC7B,MAAO,GAAU,EAGnB,YAAiB,EAAiB,GAChC,MAAO,GAAU,qBAOgB,EAAA,EAAA,GAcjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GAHA,AAAI,GAAQ,MACV,GAAO,IAEL,EAAK,mBACP,KAAM,IAAI,IACN,sEAGN,SAAK,QAAU,EAAK,SAAW,WAC/B,EAAK,SAAW,KAAK,IAAI,EAAK,UAAY,GAC1C,EAAK,SAAW,EAAK,UAAY,EACjC,EAAK,QAAU,EAAK,SAAW,EAC/B,EAAK,KAAO,EAAK,MAAQ,OACzB,EAAK,SAAW,EAAK,SAErB,AAAI,CAAC,OAAQ,MAAO,OAAO,QAAQ,EAAK,QAAU,IAChD,SAAQ,KACJ,uBAAuB,EAAK,KAAI,8CAEpC,EAAK,KAAO,QAGd,AAAI,EAAK,OAAS,MAChB,EAAK,YAAc,GACd,AAAI,EAAK,OAAS,OAInB,EAAK,QAAQ,QAAQ,SAAW,GAHpC,EAAK,YAAc,GAMjB,EAAK,YAAc,GAIvB,AAAI,EAAK,cAAgB,IACvB,GAAK,UAAY,MAIf,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,aAAe,EACpB,AAAI,KAAK,UAAY,KACnB,KAAK,KAAO,KAAK,SAEjB,KAAK,KAAO,KAAK,cAAgB,GAAO,SAAW,mBAIjD,EAAA,UAAA,WAAN,SAAiB,EAAe,mGAC9B,MAAA,CAAA,EAAM,GAAqB,WAE3B,MAFA,GAAA,OACM,EAAU,KAAK,gBAAgB,GACjC,GAAW,KACb,CAAA,GAGF,CAAI,KAAK,YAAY,EAAU,KAAK,SAAU,KAAK,MACjD,MAAK,KAAO,EACZ,KAAK,KAAO,GAGZ,MAAK,OACL,AAAI,KAAK,MAAQ,KAAK,UACpB,MAAK,aAAe,EACpB,KAAK,MAAM,aAAe,gBAM1B,EAAA,UAAA,WAAN,SAAiB,sEACf,MAAI,MAAK,aAAe,GAAK,KAAK,SAChC,QAAQ,IAAI,SAAS,KAAK,aAAY,8BAIlC,EAAA,UAAA,gBAAR,SAAwB,GACtB,AAAI,GAAQ,MACV,GAAO,IAET,GAAM,GAAe,EAAK,KAAK,SAC/B,MAAI,IAAgB,MAClB,QAAQ,KACJ,4BAA4B,KAAK,QAAO,sBACxC,2BAA0B,OAAO,KAAK,KAErC,GAEX,GAzGmC,gBAsJL,GAC5B,MAAO,IAAI,IAAc,MAGd,IAAY,CAAC,cAAa,i8DC9PvC,AA8BA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,WAAA,GAAA,aACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,UACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,aAAA,GAAA,eACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,IAAA,UACA,EAAA,EAAA,SAAA,IAAA,WACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,YAAA,IAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,iBAAA,KAAA,mBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,gBAAA,KAAA,oBA7BU,IAAA,IAAQ,QAyRH,IAAjB,AAAA,UAAiB,GAEf,GAAY,GAAZ,AAAA,UAAY,GAAyB,EAAA,EAAA,OAAA,GAAA,SAAc,EAAA,EAAA,GAAA,GAAA,KAAU,EAAA,EAAA,GAAA,GAAA,OAAjD,EAAA,EAAA,yBAAA,GAAA,wBAAuB,OAFpB,IAAA,IAAQ,KCtTzB,AAmBA,GAAM,IAAwC,eA2BnB,EAAc,GACvC,GAAM,GAAqB,CACzB,SAAU,EACV,SAAU,SACV,OAAQ,GACR,MAAO,GACP,eAAgB,GAGlB,GAAW,GAAQ,cAUW,GAC9B,MAAO,IAAW,eAUS,GAC3B,MAAO,IAAW,GC9EpB,WAwBI,EAAmB,EAAY,EAC/B,GACF,GAAM,GAAa,EAAK,YAAY,GACpC,GAAI,GAAc,EAAW,kBAAoB,QAC/C,GAAM,GAAQ,EAAW,gBACnB,EAAM,EAAW,gBAAkB,EACrC,OACC,EAAW,gBAAkB,OAAY,EAAQ,EACR,EAAW,cACzD,GAAI,EAAW,OAAS,SACtB,MAAO,IACH,EAAK,WAAW,EAAW,iBAAkB,EAAW,GAE9D,GAAI,EAAW,OAAS,WACtB,GAAM,GAAS,EAAK,WAAW,MAAM,EAAO,GAE5C,MAAO,GAAO,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAEvD,GAAM,GACF,GAAU,EAAK,WAAW,MAAM,GAAO,GAAI,EAAW,GACpD,EAAO,EAAO,WACpB,MAAO,GAAW,OAAS,SACvB,EAAK,GACL,EAAA,KAAK,cAAc,EAAO,MAAO,GAEvC,GAAM,GAAY,EAAK,WAAW,GAClC,MAAO,IAAa,EAAU,kBAU5B,EAAc,EACd,GACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAY,EAAQ,kBAAkB,KAAK,SAAA,GAC/C,MAAO,CAAC,CAAC,EAAW,GAAyB,EAAU,MAGzD,MAAO,KAAc,OACjB,EAAW,GAAyB,EAAU,IAAY,GAC1D,mBASF,EAAc,EACd,GACF,MAAO,GAAW,GAAyB,EAAM,EAAQ,+BAUvD,EAAmB,GACf,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEjB,MAAO,CACL,GAAyB,EAAU,GAAW,EAAQ,kBACtD,GAIJ,YAAkC,EAAc,GAC9C,MAAO,AAAE,GAAe,EAAI,IAAI,EAAc,cAGlB,GAC5B,GAAM,GAAQ,EAAK,MAAM,KACzB,GAAI,EAAM,SAAW,EACnB,MAAO,CAAC,EAAM,GAGhB,GAAM,GAAW,EAAM,GACvB,MAAO,CAAC,EAAU,OAAO,EAAM,EAAM,OAAS,iBAW5C,EAAY,EACZ,GACF,GAAI,GAAM,EAAc,MAAO,EAAM,EAAW,GAChD,GAAI,IAAQ,YAEV,EAAM,EAAc,mBAAoB,EAAM,EAAW,GAIzD,OAHM,GAEF,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,IACxB,EAAI,EAAG,EAAI,EAAG,IACrB,EAAgB,GAAG,GAAM,EAAiB,EAAI,GAC9C,EAAgB,GAAG,GAAM,EAAiB,EAAI,EAAI,GAEpD,MAAO,GAET,MAAO,eAYmB,GAC1B,MAAO,GAAO,KAAO,EAAS,EAAA,MAAM,GCrJtC,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,mCCnLxE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,UAC7D,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,YAGjE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC7dxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAElD,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,QAC1D,CAAC,OAAU,mBAAoB,KAAQ,iBAAkB,KAAQ,QACjE,CACE,OAAU,2BACV,KAAQ,yBACR,KAAQ,QAEV,CAAC,OAAU,oBAAqB,KAAQ,OAAQ,KAAQ,YAG5D,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CACR,OAAU,QACV,KAAQ,QACR,KAAQ,QACR,aAAgB,MAGpB,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAAU,CACrD,OAAU,wBACV,KAAQ,sBACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,YAG3C,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,YAE3D,CACE,SAAY,cACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,KACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,iBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,WAG9D,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,wCCnWrE,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAAa,CAC7D,OAAU,yBACV,KAAQ,sBACR,KAAQ,QAEV,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAC/C,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,OAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,WACV,KAAQ,WACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,gBAAiB,KAAQ,gBAAiB,KAAQ,QAAS,CACpE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,eACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,mBACV,KAAQ,gBACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,QAItB,CACE,SAAY,sBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,aAE9C,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,kBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,wBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,6BACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,aACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,QAAS,KAAQ,YAAa,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,yCCpVnD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,WAEtC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,SAAU,aAAgB,GAClE,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,SAAU,aAAgB,IAErE,MAAS,CACP,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAEzD,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,IAEjE,MAAS,CAAC,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAExD,CACE,SAAY,kBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,OACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAAW,CACpD,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,WAE7C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAC3C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,UAC7C,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,SACzC,CAAC,OAAU,eAAgB,KAAQ,eAAgB,KAAQ,wCC3KjE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,YAGnD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,WAEjD,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,cACV,KAAQ,YACR,KAAQ,QACR,aAAgB,IAElB,CACE,OAAU,yBACV,KAAQ,qBACR,KAAQ,UAId,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,UAC/C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,WAE5C,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCChGtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAE3D,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,yCCvC3C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,yBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CACE,SAAY,cACZ,SAAY,QACZ,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CAAC,SAAY,QAAS,SAAY,SAAU,CAC1C,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,YACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,SACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAEvC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,UAAW,CAC1D,OAAU,UACV,KAAQ,SACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,SACR,aAAgB,KAItB,CAAC,SAAY,OAAQ,SAAY,QAAS,OAAU,IAAK,CACvD,SAAY,eACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,0BACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,UACzC,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,yCC3G/C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,wBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,aAE3C,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAAW,CACxD,OAAU,sBACV,KAAQ,qBACR,KAAQ,yCCvDhB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,eACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC9ItB,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,eACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAAW,CAC3D,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCCtItB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,eACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GACrE,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,aACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,gBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCCjJtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,QACrD,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,uCC1GvD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,GAAI,KAAQ,UAAW,KAAQ,WACnD,CAAC,MAAS,GAAI,KAAQ,OAAQ,KAAQ,WAExC,MACI,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAEtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,WAClD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAG3E,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,OAAQ,aAAgB,MAGjE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,eACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,YACpC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CACP,CACE,OAAU,aACV,KAAQ,YACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,WACV,KAAQ,UACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,cACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,mBACV,KAAQ,iBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,YAEpD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGzE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,MACV,KAAQ,MACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,MAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAC/D,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,YACV,KAAQ,kBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,kBAAmB,KAAQ,YAChD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAG5C,CACE,SAAY,gBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCC1NtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,mCChDxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,OACV,KAAQ,SACR,KAAQ,QACR,aAAgB,IAElB,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAGhD,CACE,SAAY,aACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,MACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CAAC,CACR,OAAU,iBACV,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,QACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAAa,CACnD,MAAS,EACT,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,OACV,iBAAoB,eACpB,KAAQ,OACR,KAAQ,cAGZ,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,cAG7C,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,eACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,YAG5D,CACE,SAAY,cACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,iCChIb,AAyCA,GAAA,IAAA,WAWE,aACE,GAAM,GAAM,CACV,GAAY,GAAW,GAAS,GAAa,GAAU,GACvD,GAAY,GAAS,GAAO,GAAO,GAAU,GAAe,GAC5D,GAAW,GAAU,IAEjB,EAA0B,GAAG,OAAM,MAAT,GAAa,EAAI,IAAI,SAAA,GAAM,MAAA,GAAG,QAE9D,KAAK,UAAY,EAAY,OACzB,SAAC,EAAK,GACJ,SAAI,EAAO,UAAY,EAChB,GAET,IAlBN,cAAA,eAAkB,EAAA,WAAQ,KAA1B,WACE,MAAO,MAAK,WAAc,MAAK,UAAY,GAAI,wCAsBjD,EAAA,UAAA,eAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAM,KAChB,EAAuB,GACvB,EAAkB,GAClB,EAAoB,GACpB,EAAQ,EAAQ,OAA8B,SAAC,EAAK,GACxD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,GAAG,WAAW,eACrB,EAAa,KAAK,EAAI,EAAK,OACtB,AAAI,EAAK,KAAO,QACrB,EAAQ,KAAK,EAAI,EAAK,OACjB,AAAI,GAAK,OAAS,MAAQ,EAAK,MAAM,SAAW,IACrD,EAAU,KAAK,EAAI,EAAK,OAEnB,GACN,IAEC,EAAiB,GACf,EAAkB,GACpB,EAA8C,GAC9C,EAA+C,GACnD,AAAI,GAAa,MACf,GAAqB,KAAK,oBAAoB,EAAU,QACxD,EAAsB,KAAK,oBAAoB,EAAU,UAE3D,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAMlC,AAAI,OAAO,KAAK,GAAqB,SAAW,EAC9C,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,AAAI,EAAK,SAAS,SAAW,GAC3B,EAAQ,KAAK,KAIjB,OAAO,KAAK,GAAqB,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,aAAe,EAAoB,GACxC,EAAQ,KAAK,MAKnB,AAAI,OAAO,KAAK,GAAoB,OAAS,EAC3C,OAAO,KAAK,GAAoB,QAAQ,SAAA,GAC/B,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GACF,GAAK,aAAe,EAAmB,GACvC,EAAO,KAAK,MAIhB,EAAS,EAGX,GAAI,GAAY,GAChB,AAAI,EAAM,SAAW,MAAQ,EAAM,QAAQ,UAAY,MACrD,GAAY,EAAM,QAAQ,SAAS,OAAO,SAAC,EAAW,GACpD,SAAU,EAAK,UAAU,MAAQ,EAAK,YAAY,GAC3C,GACN,KAGL,GAAM,GACF,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,EAAE,UAAS,GAExE,MAAI,GAAU,OAAS,GACrB,GAAO,UAAY,GAGd,GAGD,EAAA,UAAA,oBAAR,SAA4B,GAC1B,MAAO,QAAO,KAAK,GAAW,IACzB,OAAgC,SAAC,EAAM,GACtC,SAAK,EAAQ,GAAM,MAAQ,EACpB,GACN,KAGD,EAAA,UAAA,QAAR,SAAgB,GAGd,GAAM,GACF,GAAgB,EAAK,KAAO,KAAK,UAAU,EAAK,KAAO,GAC3D,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAGd,GAAM,GAAgB,CACpB,KAAM,EAAK,KACX,GAAI,EAAK,GACT,SAAU,EAAO,SACjB,WACK,GAAK,OACL,IAAI,IAAI,SAAA,GAAS,MAAA,GAAM,WAAW,KAAO,EAAM,OAAO,GAAK,IAChE,OAAQ,GACR,SAAU,GACV,YAAa,GACb,WAAY,GACZ,SAAU,EAAK,MAGjB,MAAI,GAAO,QAAU,MACnB,GAAQ,YACJ,EAAO,OAAO,OACV,SAAC,EAAK,GACJ,SAAI,EAAM,MAAQ,CAChB,KAAM,EAAM,KACZ,gBAAiB,EAAM,MACvB,cAAe,EAAM,KAEhB,GAET,KAEV,AAAI,EAAO,OAAS,MAClB,GAAQ,WACJ,EAAO,MAAM,OAAoC,SAAC,EAAK,GACrD,GAAM,GAAO,EAAM,KACf,EAAQ,OACZ,OAAQ,EAAM,UACP,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAChB,EAAM,cAAgB,GAC3B,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,aACA,UACH,cAEA,KAAM,IAAI,OACN,2BAA2B,EAAM,KAAI,YAAY,EAAK,IAE9D,SAAI,EAAM,MAAQ,CAAC,MAAK,EAAE,KAAI,GACvB,GACN,KAEF,GAID,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAU,EAAY,QACtB,EAAuB,GACvB,EAAkB,GACpB,EAA+B,GACnC,AAAI,GAAW,MACb,GAAQ,EAAQ,OAA8B,SAAC,EAAK,GAClD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,KAAO,SACd,EAAQ,KAAK,EAAI,EAAK,OAEjB,GACN,KAEL,GAAM,GAAiB,GACjB,EAAkB,GAExB,EAAY,UAAU,SAAS,QAAQ,SAAA,GAC9B,GAAA,GAAA,GAAA,EAAA,MAAA,GACD,EAAa,CACjB,KAAM,EACN,GAAI,cACJ,OAAQ,GACR,WAAY,GACZ,SAAU,QACV,YAAa,GACb,WAAY,CAAC,MAAO,CAAC,MAAO,GAAgB,EAAI,MAAO,KAAM,UAC7D,SAAU,IAEZ,EAAK,aAAe,EAAI,KACxB,EAAO,KAAK,GACZ,EAAM,GAAY,IAGpB,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAIlC,GAAM,GAAgB,EAAY,IAElC,EAAY,UAAU,UAAU,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,cAAgB,EACrB,EAAQ,KAAK,MAIjB,GAAM,GAAY,KAAK,mBAAmB,GAC1C,MAAO,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,IAG1D,EAAA,UAAA,mBAAR,SAA2B,GAA3B,GAAA,GAAA,KAEE,MAAO,CACL,WAAY,EAAY,UAAU,KAClC,OAAQ,EAAY,UAAU,SAAS,OACnC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,GACjC,GAET,IACJ,QAAS,EAAY,UAAU,UAAU,OACrC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,EAAK,EAAY,KAClD,GAET,MAIA,EAAA,UAAA,mBAAR,SACI,EACA,GACF,GAAI,GAAO,EAAI,KACf,MAAI,IAAW,MACb,GAAO,EAAQ,IAEV,CAAC,KAAI,EAAE,MAAO,EAAI,OAE7B,iBAE6B,GAC3B,GAAM,GAAS,EAAA,MAAM,OACrB,GAAI,MAAO,GAAO,MAAS,YACzB,MAAO,GAAO,KAAK,GACd,GAAI,MAAO,SAAW,YAC3B,MAAO,IAAI,QAAO,EAAM,UAAU,WAElC,KAAM,IAAI,OACN,gGAKyB,EAAc,GAC7C,GAAM,GACF,MAAM,QAAQ,GAAK,OAAO,aAAa,MAAM,KAAM,GAAK,GAAa,GACzE,MAAO,GAAW,EAAQ,EAAM,0BAI9B,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,KACJ,GAAiB,EAAM,EAAG,GAE5B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAO,GAAQ,EAAM,EAAI,cAIvB,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,IAAS,GACvB,EACF,EAAM,GAAQ,KAAO,EAAM,EAAQ,EAAM,GAAQ,KAAO,EAAM,EAAO,EACzE,MAAQ,OAAO,IAAU,SAAY,EAAQ,SAAS,EAAO,gBAG/B,GAC9B,AAAI,MAAQ,IAAW,UAErB,GAAQ,GAAoB,IAE9B,OAAQ,OACD,IAAoB,SACvB,MAAO,cACJ,IAAoB,aACpB,IAAoB,aACpB,IAAoB,YACpB,IAAoB,SACvB,MAAO,YACJ,IAAoB,QACvB,MAAO,WACJ,IAAoB,UACvB,MAAO,cACJ,IAAoB,UACvB,MAAO,iBAIP,MAAO,mBAKT,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,EAAM,KAAK,KAEb,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,GAAgB,EAAM,MAExB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,KAC7B,EAAM,KAAK,KAAK,IAAI,SAAA,GAAK,MAAA,IAAgB,KAE3C,cAG6B,GAEpC,MAAI,GAAM,YACR,OAEE,EAAM,KAAO,KACR,EAAM,IAAI,IACb,SAAA,GACI,MAAC,OAAO,GAAI,MAAS,SAAY,EAAI,KAAO,SAAS,EAAI,KAAM,MAElE,eAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MACV,GAAsB,EAAM,OAE9B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,GACO,IAAM,KAAK,GAAK,EAAM,KAAK,EAAE,OAAS,EAAM,KAAK,EACX,EAAM,KAAK,IAClD,IACH,IAAI,SAAA,GAAK,MAAC,OAAO,IAAM,SAAY,EAAI,SAAS,EAAG,MAEnD,cAIL,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAAE,IAAI,SAAC,GACvB,MAAO,IAAiB,EAAG,KAGxB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,MAC7B,EAAM,KAAK,MAAM,IAAI,SAAC,GAC3B,MAAO,IAAsB,KAG1B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAEb,ECtkBT,AA4BA,GAAA,IAAA,WAGE,WACY,EAAoB,EACpB,GAFZ,GAAA,GAAA,KACY,KAAA,KAAA,EAAoB,KAAA,UAAA,EACpB,KAAA,QAAA,EAJI,KAAA,OAAmB,GACnB,KAAA,MAAoC,GAIlD,KAAK,OAAS,EAAK,WAAW,IAAI,SAAA,GAAQ,MAAA,GAAK,SAAS,KACxD,AAAI,EAAK,UAAY,MACnB,MAAK,MAAQ,OAAO,KAAK,EAAK,UACZ,OAAO,SAAC,EAAmC,GAC1C,SAAM,GAAO,EAAK,QAAQ,GACnB,GACN,KAQhB,SAAA,UAAA,SAAR,SAAiB,GACf,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,UAOtC,EAAA,UAAA,QAAR,SAAgB,EAAc,GAC5B,GAAM,GAAQ,KAAK,KAAK,SAAS,GACjC,GAAI,EAAM,QAAU,KAClB,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,SAE9C,GAAI,EAAM,GAAK,MAAQ,EAAM,GAAK,KAChC,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAa,KAAK,KAAK,SAAU,EAAM,GAEhD,GAAI,EAAM,OAAS,KACjB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,MAAQ,KAChB,MAAO,IAAc,KAAK,KAAK,SAAU,EAAM,GAEjD,GAAI,EAAM,MAAQ,MAChB,GAAI,EAAM,KAAK,GAAK,MAAQ,EAAM,KAAK,GAAK,KAC1C,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,OAAS,KACtB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,MAAQ,KACrB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAIlC,MAAO,IAEX,KCtGA,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,cACA,YACA,MACH,MAAO,CAAC,EAAA,IACH,EAAc,IAAK,EAAM,EAAW,GACrC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,UAAW,EAAM,EAAW,SAE3C,eACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,cACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,oBACH,MAAO,CAAC,EAAA,kBACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC7F9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACA,aACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,OAAQ,EAAM,EAAW,SACxC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,YACA,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,eAAgB,EAAM,EAAW,GAC/C,EAAc,eAAgB,EAAM,EAAW,SAEhD,QACH,MAAO,CAAC,EAAA,MACJ,GAAU,EAAK,WAAW,GAAI,EAAW,SAC1C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SACxC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SACzC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAE5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC5K9C,YAyBI,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EAAA,KAAK,OACD,GAA8B,EAAQ,GACtC,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlB,EAAc,GAC1D,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,IAAM,EAAG,KAAO,IAAM,EAAG,KAAO,EAAG,GAC/C,MAAO,GAGX,MAAO,GCzCT,AA+BA,GAAA,IAAA,WAIE,WACa,EAAuB,EAAyB,EACjD,EAAiC,EAChC,EAA+B,GAF/B,KAAA,KAAA,EAAuB,KAAA,MAAA,EAAyB,KAAA,QAAA,EACjD,KAAA,aAAA,EAAiC,KAAA,uBAAA,EAChC,KAAA,YAAA,EAA+B,KAAA,eAAA,EANpC,KAAA,QAA6B,GAC7B,KAAA,QAAU,GAMhB,KAAK,SAAW,EAAA,OAAO,GACvB,EAAA,KAAK,KAAK,UAGZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAMd,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,OAAO,MAChD,EAAO,OAAO,YAGlB,KAAK,QAAU,GACf,KAAK,QAAU,GACf,KAAK,SAAS,WAGhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAOtB,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,GAAS,KAAK,OAC7B,KAAM,IAAI,OAAM,4BAA4B,EAAK,wBAC7C,KAAK,QAGX,GAAM,GAAkB,KAAK,QAAQ,GACrC,GAAI,EAAgB,QAClB,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0BACpB,EAAK,wGAIf,MAAI,MAAK,gBACP,GAAgB,QAAU,IAG5B,EAAgB,KAAO,GAChB,EAAgB,QAMzB,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GAAS,MAAA,GAAK,KAAK,MAQxC,EAAA,UAAA,MAAA,SAAM,EAAe,GACnB,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,CAAC,KAAK,aAAe,GAAS,KAAK,QAClD,KAAM,IAAI,OAAM,2BACZ,EAAK,8CAA8C,KAAK,SAG9D,GAAM,GAAI,KAAK,QAAQ,IAAU,GAEjC,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,eACZ,KAAK,KAAI,0CAA0C,EAAK;uCAExD,EAAO,MAAK,8BAA8B,KAAK,MAAK,KAc1D,GAVA,AAAI,KAAK,SAAW,GACf,MAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,IAC7D,MAAK,aAAe,EAAO,OAG7B,GACI,KAAK,aAAc,EAAO,MAC1B,eAAe,KAAK,KAAI,0CACpB,EAAK,KAET,EAAE,KACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,uCAGf,GAAI,EAAE,QACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,0CAGf,EAAE,OAAS,EACX,EAAA,KAAK,GACL,EAAE,QAAU,GAEZ,KAAK,QAAQ,GAAS,GAMxB,EAAA,UAAA,UAAA,SAAU,EAAmB,GAA7B,GAAA,GAAA,KACE,GAAI,EAAQ,SAAW,EAAQ,OAC7B,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,sCACxB,4BACI,EAAQ,OAAM,qCACd,EAAQ,OAAM,MAGxB,EAAQ,QAAQ,SAAC,EAAG,GAAU,MAAA,GAAK,MAAM,EAAG,EAAQ,OAWtD,EAAA,UAAA,OAAA,SAAO,EAAoB,GACzB,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAK,EAMH,EAAU,EAAQ,MAAM,EAAG,KAAK,aALhC,EAAU,GACV,OAAS,GAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAMjB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAKpC,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAAO,gCAElC,EAAA,MAAM,EAAS,IAMxB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAI,KAAK,SAAW,EAClB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAIpC,OADM,GAAU,GACP,EAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAGf,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAC9B,mDACI,KAAK,aAAY,4BAA4B,EAAQ,GAAG,MAAK,KAE9D,EAAA,OAAO,EAAS,IASzB,EAAA,UAAA,QAAA,SAAQ,EAAmB,GACzB,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAGhD,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,CAAC,KAAK,aAAe,GAAY,KAAK,QACxC,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,KAAK,QAAO,KAGtE,KAAK,UAAU,EAAS,EAAA,QAAQ,EAAQ,KAS1C,EAAA,UAAA,MAAA,SAAM,EAAkB,GAAxB,GAAA,GAAA,KACE,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAEhD,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEZ,EAAW,4BAA4B,EAAO,OAGpD,GAAI,CAAC,KAAK,aAAe,EAAO,SAAW,KAAK,QAC9C,KAAM,IAAI,OACN,2DACI,KAAK,QAAO,QAAQ,EAAO,OAAM,kEAI3C,GAAM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,GAC1B,EAAA,KAAK,WACH,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,EAAK,cAE3D,MAAO,KAGT,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAQ,GAAK,EAEf,KAAK,UAAU,EAAS,IAE5B,KC3TA,AAoCA,GAAA,IAAA,WAeE,WACa,EAA4B,EAC5B,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IADxB,KAAA,QAAA,EAA4B,KAAA,aAAA,EAC5B,KAAA,aAAA,EACX,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,IAAiB,EAAO,MAC1B,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,EAAO,OAEhD,GACI,EAAc,EAAO,MAAO,+BAEhC,EAAA,KAAK,KAGT,KAAK,SAAW,EAAA,OAAO,GACvB,KAAK,eAAiB,EACtB,EAAA,KAAK,KAAK,UA5BZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAiCvB,EAAA,UAAA,KAAA,WACE,MAAO,IAAI,GACH,KAAK,QAAO,QAAG,KAAK,aAAc,KAAK,eAMjD,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,MACzC,EAAO,YAGX,KAAK,QAAQ,OAAS,EACtB,KAAK,SAAS,WAKhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAUtB,EAAA,UAAA,MAAA,SAAM,EAAwB,EAAwB,GAAtD,GAAA,GAAA,KAEE,GAFoD,AAAA,IAAA,QAAA,GAAA,IAEhD,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,IAAgB,IAAM,KAAK,QAAQ,SAAW,EAChD,KAAM,IAAI,OAAM,kCACZ,EAAW,iCACX,KAAK,QAAQ,OAAM,cAEzB,UACI,EAAc,KAAK,aAAc,+BAC9B,EAAA,KAAK,WACV,GAAM,GACF,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAA,QAAQ,EAAQ,KAC/C,MAAO,GAAA,MAAM,EAAiB,MASlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAC9B,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAG9C,GAAI,KAAK,SAAW,EAClB,KAAM,IAAI,OAAM,qCAGlB,GAAM,GAAS,KAAK,QAAQ,MAC5B,UACI,EAAO,MAAO,EAAc,+BACzB,EAAA,QAAQ,EAAQ,IAOzB,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAM9C,GAHA,GACI,EAAO,MAAO,KAAK,aAAc,+BAEjC,KAAK,iBAAmB,KAAK,OAC/B,KAAM,IAAI,OAAM,4CAElB,EAAA,KAAK,GACL,KAAK,QAAQ,KAAK,IAOpB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAO,EACT,KAAM,IAAI,OACN,0DAA0D,GAGhE,GAAI,KAAK,iBAAmB,IAAM,EAAO,KAAK,eAC5C,KAAM,IAAI,OAAM,+BACZ,EAAI,6BAA6B,KAAK,eAAc,KAE1D,KAAK,QAAQ,OAAS,GASxB,EAAA,UAAA,QAAA,SAAQ,EAAsB,EAAwB,GAEpD,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,EAAe,GAAK,EAAe,KAAK,QAAQ,OAClD,KAAM,IAAI,OAAM,4BACZ,EAAY,mBAAmB,KAAK,QAAQ,OAAM,cAGxD,GAAI,KAAK,QAAQ,IAAiB,KAChC,KAAM,IAAI,OAAM,oBAAoB,EAAY,aAGlD,UACI,KAAK,QAAQ,GAAc,MAAO,EAClC,+BAEG,KAAK,QAAQ,IAQtB,EAAA,UAAA,QAAA,SAAQ,EAAsB,GAC5B,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAG9C,GAAI,EAAe,GACf,KAAK,iBAAmB,IAAM,GAAgB,KAAK,eACrD,KAAM,IAAI,OAAM,yBACZ,EAAY,uBAAuB,KAAK,eAAc,cAG5D,GACI,KAAK,aAAc,EAAO,MAAO,+BACrC,EAAA,KAAK,GACL,KAAK,QAAQ,GAAgB,GAU/B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAwB,GAAlD,GAAA,GAAA,KAEE,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAU9C,MAPA,IACI,KAAK,aAAc,EAAc,+BAIrC,EAAU,EAAQ,MAAM,EAAG,KAAK,QAE5B,EAAQ,SAAW,EACd,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAK,QAAQ,GAAI,KAC1D,MAAO,GAAA,MAAM,EAAS,MAS1B,EAAA,UAAA,OAAA,SAAO,EAAwB,GAA/B,GAAA,GAAA,KACE,GAAI,CAAC,CAAC,GAAgB,IAAiB,KAAK,aAC1C,KAAM,IAAI,OAAM,uBACZ,KAAK,aAAY,+BAA+B,GAMtD,MAHA,IACI,KAAK,aAAc,EAAc,+BAEjC,KAAK,SAAW,EACX,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAK,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAG,KACjD,MAAO,GAAA,OAAO,EAAS,MAG7B,iBAQI,EAAgB,EAAwB,GAC1C,GAAM,GAAQ,EAAO,MACrB,GAAI,EAAO,MAAM,OAAS,EACxB,KAAM,IAAI,OACN,oDAAoD,EAAO,OAEjE,GAAI,EAAO,QAAU,EACnB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,GAEzC,GAAM,GAAc,EAAO,MAAM,MAAM,GACvC,GACI,EAAa,EAAc,+BAE/B,GAAM,GAAuB,EAAA,QAAQ,GACrC,MAAO,IAAI,IAAW,EAAY,EAAc,eAU9C,EAAwB,EAAwB,GAClD,MAAO,IAAI,IAAW,GAAI,EAAc,EAAc,eAWpD,EAAgB,EAAmB,EACnC,GACF,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,GAAe,MAAQ,IAAgB,IAAM,GAAY,EAC3D,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,EAAW,KAGrE,GAAM,GAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,GACtD,EAAU,EAAA,QAAQ,EAAQ,GAChC,SAAQ,QAAQ,SAAC,EAAO,GACtB,EAAK,QAAQ,EAAO,EAAQ,MAEvB,cAWL,EAAgB,EAAkB,GACpC,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEV,EAAW,4BAA4B,EAAO,OAmBtD,OAhBM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,EAAA,KAAK,WAC7B,GAAM,GAAU,GAChB,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,GAEtD,SAAO,UACA,IAGH,EAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,EAAO,QAE1D,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAK,QAAQ,EAAG,EAAQ,IAE1B,MAAO,GC3YT,AAiBA,GAAA,IAAA,OAUa,GAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,kUACnB,EAAA,EAAK,iBACN,KAAA,MAAA,CAAA,EAAA,OACA,cAAA,MAAA,CAAA,EAAA,OAgBA,QAAA,MAAA,CAAA,EAAA,OACA,iBAAA,MAAA,CAAA,EAAA,OAsDA,WAAA,MAAA,CAAA,EAAA,OAIA,SAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QAOA,OAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAoBA,qBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAQA,sBAAA,MAAA,CAAA,EAAA,QAUA,uBAAA,MAAA,CAAA,EAAA,QAWA,sBAAA,MAAA,CAAA,EAAA,QAQA,qBAAA,MAAA,CAAA,EAAA,QAWA,oBAAA,MAAA,CAAA,EAAA,QAMA,qBAAA,MAAA,CAAA,EAAA,QAOA,oBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAaA,sBAAA,MAAA,CAAA,EAAA,QACA,oBAAA,MAAA,CAAA,EAAA,QAcA,oBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAYA,kBAAA,MAAA,CAAA,EAAA,QAYA,uBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAUA,qBAAA,MAAA,CAAA,EAAA,QASA,oBAAA,MAAA,CAAA,EAAA,QAUA,kBAAA,MAAA,CAAA,EAAA,wBAzTG,SACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClC,CAAA,EAAM,EAAK,eAC7B,MADM,GAAY,GAAA,OACd,EAAU,GACZ,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,gBAE1C,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAKtC,SACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAO,EAAc,OAAQ,EAAM,EAAW,GAI/C,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAFxC,SACD,GAAA,OAEC,EAAS,EAAK,IAAI,SAAA,IAAU,MAAA,IAAO,KACzB,CAAA,EAAM,EAAW,GAAG,eAAhC,EAAY,GAAA,OAEhB,EAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAChD,GAAO,YAIP,EAAmB,gFAIf,UAAa,EAEV,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACzC,EAAQ,EAAQ,eAAgB,EAAQ,uBAD5C,SAAS,GAAA,OAEH,GAAY,EAAO,IAAI,SAAA,IAAU,MAAA,IAAO,KAI9C,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,YAMN,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAQ,EAAQ,eAAgB,EAAQ,uBAF1C,UACD,GAAA,OAEO,CAAA,EAAM,GAAW,GAAG,eAAhC,SAAY,GAAA,OAEZ,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,4CA1BN,GAAU,WAAE,CAAA,EAAA,wCA8BnB,MAAA,CAAA,EAAO,UAGD,SAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAChD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClD,AAAK,EAAK,MACR,GAAO,GAAY,IAGb,CAAA,EAAM,EAAK,gBAAnB,MAAA,CAAA,EAAQ,GAAA,OAAmB,GAAK,CAAC,OAAW,GAAQ,CAAC,EAAM,iBAK3D,MAFM,GAAY,EAAK,WAAW,KAC9B,SAAA,IAAQ,MAAA,IAAU,GAAM,EAAW,KAAa,SAChD,EACI,GAAO,GAAU,EAAW,EAAW,GAC7C,CAAA,EAAO,CAAC,GAAY,MAEtB,CAAA,EAAO,gBAGD,SACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,WAAW,GACnB,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,YACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,gBACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,yBAA0B,EAAM,EAAW,GAEvD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAc,GAAI,IACpB,EAAM,EAAO,EAAM,EAAc,EAAwB,EACzD,GACJ,EAAQ,eAAe,GACvB,CAAA,EAAO,CAAC,EAAY,SAAU,EAAA,OAAO,aAG/B,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAmB,EAAQ,eAAe,EAAG,IACnD,EAAiB,MAAM,EAAO,GAC9B,CAAA,EAAO,CAAC,EAAiB,mBAGnB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAkB,EAAQ,eAAe,EAAO,IACtD,CAAA,EAAO,CAAC,EAAgB,KAAK,aAGvB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAoB,EAAQ,eAAe,EAAS,IAC1D,CAAA,EAAO,CAAC,EAAkB,OAAO,EAAe,aAG1C,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAqB,EAAQ,eAAe,GAAU,IAC5D,GAAmB,QAAQ,GAAgB,IAC3C,CAAA,EAAO,CAAC,GAAmB,mBAGrB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAoB,EAAQ,eAAe,GAAS,IACpD,GACF,EAAc,QAAS,EAAM,EAAW,GAC5C,CAAA,EAAO,CAAC,GAAkB,OAAO,cAG3B,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,MAAM,GAAS,IAChC,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAkB,EAAQ,eAAe,GAAO,IACtD,CAAA,EAAO,CAAC,EAAA,OAAO,GAAgB,OAAQ,mBAGjC,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,gBACjB,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,QAAQ,EAAO,GAC1B,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAE7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAW,EAAc,cAI9C,UACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GACF,GAAQ,GAAe,GAAgB,EAAc,IACzD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,GAAQ,EAAc,GAAc,IACvD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,EAAS,IAClD,CAAA,EAAO,CAAC,GAAW,OAAO,EAAe,GAAc,aAGjD,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,MAAM,EAAc,GAAc,cAG/C,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,GAAW,GAAQ,EAAc,IACpD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAC5C,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,CAAA,EAAO,CAAC,GAAW,OAAO,GAAa,aAGjC,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,SAAS,GACpB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAc,cAGnC,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,UAAW,EAAM,EAAW,GAExC,GAAa,GAAM,GAAa,GAAS,GAC/C,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGnB,KAAM,WAAU,aAAa,EAAK,GAAE,6BCvW1C,AA2BA,YACI,EAAY,EAA4B,GACpC,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAC7B,EAAc,IAAY,iBAE1B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,yGAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,oFAIR,GAAI,EACF,KAAM,IAAI,OACN,yEAEN,GAAM,GAAS,EAAc,UAAW,EAAM,EAAW,GACnD,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGhB,MAAO,CACL,OAAM,EACN,IAAG,EACH,WAAU,EACV,UAAS,EACT,QAAO,EACP,SAAQ,EACR,eAAc,GAIX,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAQ,EAAyB,EACjC,QAED,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,gBACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAIvB,8BACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,gBAAgB,CAClC,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAGvB,0BACA,mBACH,GAAM,GAAQ,EACI,cAAe,EAAM,EACrB,GAEZ,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GACxC,MAAO,CAAC,EAAA,gBACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAO,CAAC,EAAO,GAAI,EAAO,IAAK,QAEhC,4BACA,mBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cAET,MAAO,CAAC,EAAA,gBACJ,EAAc,QAAS,EAAM,EAAW,GAExC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GAEzC,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,EACnC,EACA,CAAC,EAAU,GAAI,EAAU,GAAI,EAAU,UAExC,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,qBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,sBAAuB,EAAM,EAAW,GAEpD,EAAA,EAAA,kBAAA,EAAA,IAAA,EAAA,EAAA,GAAA,CAAA,EAAA,GAAA,EAAA,IAAA,CAAA,EAAA,GAAA,EAAA,IAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,QAIf,MAAO,CAAC,EAAQ,OAEb,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,cACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,YAAa,EAAM,EAAW,GAG1C,EAAe,EAAQ,GACvB,EAAc,EAAQ,GAGtB,EAAiB,EAAU,GAC3B,EAAgB,EAAU,GAEhC,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAc,GAAc,EAC7B,CAAC,EAAgB,GAAgB,iBAIrC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC/S9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,KAAW,EAAO,EAAO,QAE9B,YACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAM,EAAc,MAAO,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,SAAe,EAAO,EAAM,QAEjC,eACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,YAAkB,EAAQ,EAAY,QAE3C,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OAAa,EAAS,EAAO,EAAS,QAE3C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cAEJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,SAEzC,SACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,MACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,SAGzC,mBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,gBACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,GAExC,QAED,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1H9C,AAiBA,GAAA,IAAA,OAUA,YACI,EAAY,EAA4B,GAC1C,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EAAS,EAAc,SAAU,EAAM,EAAW,GAClD,EACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CACL,MAAK,EACL,OAAM,EACN,cAAa,EACb,aAAY,EACZ,eAAc,EACd,aAAY,GAIT,GAAM,IAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,mHACnB,EAAA,EAAK,iBACN,sBAAA,MAAA,CAAA,EAAA,OAgBA,sBAAA,MAAA,CAAA,EAAA,OAcA,sBAAA,MAAA,CAAA,EAAA,OACA,sBAAA,MAAA,CAAA,EAAA,OAQA,QAAA,MAAA,CAAA,EAAA,OAQA,WAAA,MAAA,CAAA,EAAA,uBA9CG,SAOF,GAAU,EAAM,EAAW,GAN7B,EAAK,EAAA,MACL,EAAM,EAAA,OACN,EAAa,EAAA,cACb,EAAY,EAAA,aACZ,EAAc,EAAA,eACd,EAAY,EAAA,aAGC,CAAA,EAAM,EAAA,MAAY,gCAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,wBAGjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAG3D,EACF,EAAc,qBAAsB,EAAM,EAAW,GAG1C,CAAA,EAAM,EAAA,MAAY,6BAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,sBAIjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAGzD,CAAA,EAAM,EAAA,MAAY,uBACtB,EAAmB,EAAoB,EAAe,EACtD,WAFJ,MAAA,CAAA,EAAA,CAAQ,EAAA,gBAKF,SAAY,EAAA,KACb,EAAc,YAAa,EAAM,EAAW,GAC7C,QACY,CAAA,EAAM,EAAA,WAAiB,WAAjC,SAAM,CAAI,EAAA,QAChB,EAAU,UACV,CAAA,EAAO,UAGP,MAAA,CAAA,EAAO,EAAA,eACH,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,6BC3G1C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAS,EAAA,KAAW,EAAG,EAAG,GAChC,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAS,EAAA,OAAa,GAC5B,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAS,EAAA,OAAa,EAAG,GAC/B,MAAO,CAAC,EAAO,OAAQ,EAAO,iBAG9B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCpDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,GAAU,EAAK,UAEnB,yBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,IAAY,OACjD,cACH,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,QACrC,eACA,mBACA,2BACH,GAAM,GAAO,EAAc,IAAK,EAAM,EAAW,GACjD,MAAO,CAAC,GAAY,QAEjB,YACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,IAAY,SACjC,WACH,GAAM,GACD,EAAc,IAAK,EAAM,EAAW,GACzC,MAAO,CAAC,GAAY,QACjB,QACH,MAAO,CAAC,EAAA,SACH,EAAc,IAAK,EAAM,EAAW,GAAoB,MACzD,cACD,SACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,GAAA,SAAe,EAAE,aACtC,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OAAa,QAClB,QACH,GAAM,GAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,QAAQ,KACJ,kGAEJ,QAAQ,IAAI,GACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,QAAQ,IAAI,MAAM,UAAU,MAAM,KAAK,EAAK,GAAG,YAC9B,MAAM,EAAG,IAE5B,MAAO,CAAC,WAGR,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,eAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,yBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,sBAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,iBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,WAAY,EAAM,EAAW,GACzC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,qBAAsB,EAAM,EAAW,GAEzD,MAAO,CAAC,EAAA,MAAY,cAChB,EAAmB,EAAmB,EACtC,EAA8B,EAC9B,YAGJ,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzE9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,eACH,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACA,WACH,MAAO,CAAC,EAAA,MACJ,EAAc,YAAa,EAAM,EAAW,GAC5C,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACA,oBACA,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,aAAc,EAAM,EAAW,GAC7C,EAAc,aAAc,EAAM,EAAW,SAG9C,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SAExC,eACG,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAE7B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,sFAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,iEAGF,GAAA,GAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEhB,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,KAAM,EACN,WAAY,EACZ,uBAAwB,aAI1B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCjF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,qBACA,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,MACH,MAAO,CAAC,EAAA,2BACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,OAAQ,EAAM,EAAW,SAExC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cACJ,EAAc,gBAAiB,EAAM,EAAW,GAEhD,EAAc,cAAe,EAAM,EAAW,GAC9C,EAAc,eAAgB,EAAM,EAAW,GAE/C,EAAc,eAAgB,EAAM,EAAW,aAInD,KAAM,WAAU,aAAa,EAAK,GAAE,yBC9E9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,EAAW,YAGf,KAAM,WAAU,aAAa,EAAK,GAAE,yBCtH9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,eACA,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACvC,EACA,EAAc,UAAW,EAAM,EAAW,GAC9C,SAAS,EAAO,MAAM,EAAG,GAClB,CAAC,EAAA,OAAa,EAAQ,QAE1B,eACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OAAa,EAAO,EAAA,KAAW,EAAS,SAAU,QAEvD,gBACA,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,QAAc,EAAO,QAE1B,SAEH,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAEhD,EAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,gBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,MAAO,EAAM,EAAW,GACpC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAE/C,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,aACJ,EAAQ,EAAO,EAAK,EAAS,EAAW,EAAS,EACjD,EAAa,QAEd,OACH,MAAO,GAAA,KAAK,WACV,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GAGxC,EAAQ,EAAQ,GAAG,MACnB,EAAgB,EAAA,QAAc,EAAQ,IAAI,MAC1C,EAAS,EAAQ,IAAI,SAAA,GACzB,GAAM,GAAY,EAAA,KAAK,YAAY,EAAO,MAAO,GACjD,GAAI,CAAC,GACD,CAAC,EAAA,KAAK,YACF,EAAA,QAAc,GAAQ,MAAO,GACnC,KAAM,IAAI,OAAM,0CAElB,MAAO,GAAY,EAAS,EAAA,QAAc,EAAQ,KAEpD,MAAO,CAAC,EAAA,MAAY,EAAQ,UAG3B,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GAC7C,MAAO,GAAA,QAAc,EAAQ,OAE1B,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,YACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,kBAAmB,EAAM,EAAW,GAGhD,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,GAAA,MAAY,EAAQ,EAAiB,OAEzC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,UAAgB,EAAS,EAAQ,QAEtC,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,SAAe,EAAG,QAEvB,iBACH,GAAM,GACF,EAAc,gBAAiB,EAAM,EAAW,GAE9C,EACF,EAAc,cAAe,EAAM,EAAW,GAE5C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,MAAO,CAAC,EAAA,cACJ,EAAS,EAAc,EACvB,EAAa,QAAU,EAAa,MAChC,EACA,EAAA,KAAW,EAAc,EAAa,iBAG9C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1K9C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBChDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAGzC,cACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAGzD,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,UAAW,EAAM,EAAW,GAE1C,EAAc,gBAAiB,EAAM,EAAW,SAGjD,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,gBACH,GAAM,GACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACrC,cAEb,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAW,QAEZ,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAG5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnG9C,YAgDI,EAAY,EACZ,GACF,GAAM,GACD,SAAC,EAAY,EAA4B,GACxC,OAAQ,EAAK,cACN,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,cACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAsB,EAAM,EAAW,SAC9C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,UACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAkB,EAAM,EAAW,SACtD,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,gBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAwB,EAAM,EAAW,SAChD,YACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,iBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAyB,EAAM,EAAW,SACjD,SACH,GAAM,GAAW,GAAgB,EAAK,IACtC,GAAI,GAAY,EAAS,eACvB,MAAO,GAAS,eACZ,GAAI,IAAc,EAAM,EAAW,IAEvC,KAAM,WAAU,aAAa,EAAK,GAAE,+BAGtC,KAAM,WACF,eAAe,EAAK,GAAE,yIAI7B,EAAM,EAAW,GACxB,MAAI,aAAiB,SACZ,EAAM,KAAK,SAAC,GAAS,MAAA,GAAG,OAAO,KAEjC,GAAG,OAAO,GCvEnB,GAAA,IAAA,WAME,WACa,EACA,EACA,EACA,GAHA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IAHA,KAAA,UAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,YAAA,EATL,KAAA,YAAc,CAAC,GAAI,EAAG,UAAW,GAAI,YAAa,GAClD,KAAA,SAAmC,CAAC,KAAK,aACzC,KAAA,OAAS,EAQf,KAAK,4BAGC,SAAA,UAAA,SAAR,SAAiB,EAAY,GAC3B,MAAO,CAAC,GAAE,EAAE,UAAS,EAAE,YAAa,IAQtC,OAAA,eAAI,EAAA,UAAA,iBAAc,KAOlB,WACE,MAAO,MAAK,cARd,SAAmB,GACjB,AAAI,KAAK,WAAa,GACpB,MAAK,SAAW,EAChB,KAAK,8DAWT,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,mBAAmB,oCAOjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGN,EAAA,UAAA,0BAAR,WAEE,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,KAAK,SAAS,OAAS,EAAG,KAC5C,GAAM,GAAW,KAAK,SAAS,MAAM,EAAG,KAAK,SAAS,OAAS,GAC/D,EAAM,KAAK,KAAK,qBAAqB,IAEvC,EAAM,KAAK,IACX,KAAK,mBAAqB,GAGpB,EAAA,UAAA,qBAAR,SAA6B,GAC3B,MAAO,GACH,EACK,IACG,SAAA,GAAW,MAAC,GAAQ,KAAO,GAAK,EAAQ,cAAgB,EACpD,GACG,EAAQ,UAAS,IAAI,EAAQ,cACvC,KAAK,KACV,IAON,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,UACP,MAAK,SACL,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,KAAK,KAAK,SAAS,KAAK,OAAQ,IAC9C,KAAK,mBAAmB,QAAQ,KAAK,qBAAqB,KAAK,aAQnE,EAAA,UAAA,UAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,EAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,OAAO,IACrB,KAAK,kBAAkB,YAEvB,MAAM,IAAI,OAAM,4CAQpB,EAAA,UAAA,cAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,GAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SACL,GAAM,GACF,OAAO,OAAO,GAAI,KAAK,SAAS,KAAK,SAAS,OAAS,IAC3D,EAAQ,aAAe,EACvB,EAAQ,GAAK,KAAK,OAClB,KAAK,SAAS,OAAO,GAAI,EAAG,GAC5B,KAAK,mBAAmB,OACpB,EAAG,EAAG,KAAK,qBAAqB,KAAK,eAEzC,MAAM,IAAI,OAAM,0DAIpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,UAAU,IAGxB,EAAA,UAAA,eAAA,SAAe,GACb,KAAK,eAAe,EAAY,IAAM,GAGxC,EAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,eAAe,IAG7B,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,cAAc,EAAW,IAAM,GAGtC,EAAA,UAAA,cAAA,SAAc,GACZ,MAAO,MAAK,cAAc,IAG5B,EAAA,UAAA,QAAA,SAAQ,GACN,OAAW,KAAO,MAAK,eACrB,KAAK,eAAe,GAAK,cAAc,GAGzC,OAAW,KAAO,MAAK,cACrB,KAAK,cAAc,GAAK,cAAc,IAG5C,KC1LA,YAyCI,EAAwB,EAAiB,EACzC,GACF,GAAM,GAAY,GAAI,KAChB,EAA0B,GAC5B,EAAoB,KACpB,EAAuB,KAIrB,EAAO,GAAI,KACX,EACF,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAEpD,EAA0B,GAC9B,AAAI,GAAa,MACf,GAAgB,EAAU,IAAI,SAAA,GAAQ,MAAA,IAAc,EAAK,MAAM,MAIjE,OADM,GAAe,EAAO,QACrB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MAWtB,GAVA,AAAI,IAAc,IAAS,GAAe,KACxC,CAAI,GAAe,MACjB,GAAc,EACd,EAAa,EAAY,SAAS,IAAI,SAAA,GAAS,MAAA,GAAM,OACnC,OAAO,SAAA,GAAQ,MAAA,GAAU,IAAI,OAGnD,EAAU,IAAI,EAAK,MAGf,EAAU,EAAK,OAAS,KAC1B,SAGF,GAAI,EAAe,QAAQ,EAAK,QAAU,GACxC,SAGF,GAAI,EAAc,QAAQ,EAAK,QAAU,GACvC,SAEF,GAAI,EAAK,OAAO,SAAW,GACzB,EAAc,KAAK,EAAK,MACxB,SAEF,EAAK,OAAO,QAAQ,SAAA,GAElB,GAAI,EAAK,IAAI,EAAM,MACjB,OAEF,EAAK,IAAI,EAAM,MACf,EAAS,KAAK,KAGlB,MAAO,CAAC,OAAM,EAAE,QAAO,EAAE,UAAS,EAAE,cAAa,EAAE,YAAW,EAAE,WAAU,eAQxE,EAAc,EACd,GACK,GAAA,GAAA,EAAA,UAAW,EAAA,EAAA,OACZ,EAAmB,GACnB,EAAa,OAAO,KAAK,GACP,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAChC,IAAI,SAAA,GAAQ,MAAA,GAAM,MAAM,KAC1C,EAAY,EAAM,UAExB,EAAW,QAAQ,SAAA,GACjB,AAAI,EAAU,IAAI,EAAM,OACtB,EAAS,KAAK,KAGlB,EAAM,QAAQ,QAAQ,SAAA,GACpB,AAAI,EAAU,IAAI,EAAO,OACvB,EAAS,KAAK,KAGlB,AAAI,GAAa,MACf,EAAU,QAAQ,SAAA,GAChB,AAAI,EAAU,IAAI,EAAK,OACrB,EAAS,KAAK,KAMpB,OAFM,GAAO,GAAI,KACX,EAAuB,GACtB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MACtB,EAAK,IAAI,EAAK,MACd,AAAK,EAAU,EAAK,OAClB,EAAa,KAAK,GAEpB,EAAK,SAAS,QAAQ,SAAA,GACpB,AAAI,CAAC,EAAK,IAAI,EAAM,OAAS,EAAU,IAAI,EAAM,OAC7C,EAAM,OAAO,MAAM,SAAA,GAAS,MAAA,GAAK,IAAI,EAAM,SAC7C,EAAS,KAAK,KAIpB,MAAO,GAGT,GAAM,IAAmB,CACvB,SAAU,QAAS,QAAS,OAAQ,gBAAiB,cACrD,iBAAkB,KAAM,SAEpB,GAAoB,CACxB,sBAAuB,sBAAuB,sBAAuB,qBAGzC,GAC5B,MAAO,IAAiB,QAAQ,EAAK,KAAO,cAGf,GAC7B,MAAO,IAAkB,QAAQ,EAAK,KAAO,ECjK/C,AAkCA,GAAA,IAAA,WAsFE,WAAoB,EAAsB,GAA1C,GAAA,GAAA,KAAoB,KAAA,MAAA,EAAsB,KAAA,OAAA,EArFlC,KAAA,YAAmC,GAAI,KACvC,KAAA,WAA8B,GAM9B,KAAA,UAAY,IACZ,KAAA,WAAqC,GACrC,KAAA,qBAA0D,GA6EhE,KAAK,SAAW,EAAM,QACtB,KAAK,QAAU,EAAM,OACrB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UAExB,AAAI,EAAM,WAAa,MACrB,OAAO,KAAK,EAAM,WAAW,QAAQ,SAAA,GACnC,EAAK,qBAAqB,GACtB,GAAI,GAAc,EAAM,UAAU,GAAO,KApFnD,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,4CAGpD,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,oBACZ,KAAK,sDAG5B,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,gBAGpD,SAAc,GACZ,GAAM,GAAY,OAAO,KAAK,GAAW,IACrC,SAAA,GAAO,MAAA,GAAU,GAAK,IAAI,SAAA,GAAU,MAAA,GAAO,OAC/C,KAAK,WAAa,GAAG,OAAM,MAAT,GAAa,GAC/B,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GACtB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,IAAI,SAAA,GACvB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,cAAgB,EAAK,wCAG5D,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,IAAI,SAAC,GACxB,GAAM,GAAO,EAAK,cAAgB,EAAK,KACvC,MAAO,GAAK,cAAoB,EAAI,IAAI,EAAK,cAAmB,qCAIpE,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAAA,GAAA,GAAA,KACE,MAAO,QAAO,KAAK,KAAK,YAAY,OAAO,SAAC,EAAK,GAC/C,SAAI,GAAO,EAAK,WAAW,GAAK,UACzB,GACN,qCA0BG,EAAA,UAAA,kBAAR,SAA0B,EAAgB,GACxC,GAAM,GAAe,EAAO,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OAC7C,EAAgB,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OACrD,MAAO,GAAa,KAAK,KAAK,WAAa,KACvC,EAAc,KAAK,KAAK,YAOtB,EAAA,UAAA,QAAR,SAAgB,EAAwB,GACtC,GAAM,GACF,GAAqB,EAAQ,EAAS,KAAK,UAAW,KAAK,YACxD,EAAA,EAAA,cAAe,EAAA,EAAA,YAAa,EAAA,EAAA,WACnC,GAAI,GAAe,KACjB,KAAM,IAAI,OACN,qCAAqC,EAAY,KAAI,gBACrD,oBAAmB,EAAY,GAAE,kBACjC,6DACA,qCAAoC,EAAU,MAGpD,GAAI,EAAc,OAAS,GACzB,GAAM,GAAW,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,OAC9B,EAAU,OAAO,KAAK,GAC5B,KAAM,IAAI,OACN,+BAA+B,EAAQ,8BACvC,KAAI,EAAO,qCAAqC,EAAa,MAGnE,MAAO,IACH,KAAK,MAAO,KAAK,UAAW,IAYlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAAhC,GAAA,GAAA,KACE,EAAS,KAAK,UAAU,GACxB,GAAM,GAAQ,OAAO,KAAK,GAAQ,OAClC,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,GAClB,GAAM,GACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC5D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAG/D,AAAI,EAAY,SAAW,GACzB,GAAc,KAAK,UAGrB,GAAM,GAAiB,KAAK,kBAAkB,EAAY,GAGtD,EAAe,KAAK,YAAY,IAAI,GACxC,AAAI,GAAgB,MAClB,GAAe,KAAK,QAAQ,EAAQ,GACpC,KAAK,YAAY,IAAI,EAAgB,IAGvC,GAAM,GAAiC,GACjC,EAA+B,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,GAAI,IAChB,EAAK,UAAW,EAAgB,EAChC,EAAK,qBACH,EAAU,GAAA,GAAwB,EAAK,WAE7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAKzB,OAFM,GAAgB,EAAK,mBAAmB,GACxC,EAA2D,GACxD,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAO,EAAa,GAC1B,GAAI,CAAC,EAAW,EAAK,OACnB,GAAM,GAAU,GAAU,EAAM,EAAY,GAC5C,GAAI,YAAmB,SACrB,KAAM,IAAI,OACN,4BAA4B,EAAK,GAAE,kEAGzC,EAAW,EAAK,MAAQ,EACxB,EAAK,uBACD,EAAK,KAAM,EAAM,EAAY,EAAS,EACtC,EAAiB,IAIzB,MAAI,GAAK,QAAU,MACjB,EAAQ,QAAQ,GAEX,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAY,QAInD,EAAA,UAAA,mBAAR,SAA2B,GACzB,GAAM,GAAM,GAAG,OAAO,MAClB,GACA,OAAO,KAAK,GACP,IAAI,SAAA,GAAO,MAAA,GAAU,KACrB,IAAI,SAAA,GAAW,MAAA,GAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,QACrD,MAAO,IAAI,KAAI,IAET,EAAA,UAAA,uBAAR,SACI,EAAkB,EAAY,EAC9B,EAA2B,EAC3B,EACA,GAGF,GAAI,EAAK,WAAa,WAAa,EAAY,QAAQ,KAAc,GACnE,OAGF,EAAU,GAAU,QAAQ,SAAA,GAC1B,AAAI,GAAU,MACZ,GAAgC,EAAO,IAClC,GAAgC,EAAO,KAAO,GAC/C,EAAK,SAAS,UAGtB,EAAK,OAAO,QAAQ,SAAA,GAGlB,GAAI,EAAM,WAAa,WACrB,GAAM,GACF,GAA6B,EAAM,KAAM,EAAW,GACxD,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,GAAU,CAAC,EAAc,IAAI,EAAO,KACtC,GAAM,GAAQ,EAAgC,EAAO,IACrD,AAAI,IAAU,EACZ,GAAO,UACP,MAAO,GAAgC,EAAO,KACzC,AAAI,GAAS,MAGlB,EAAgC,EAAO,aAkB/C,EAAA,UAAA,aAAN,SAAmB,EAAwB,sEAEzC,MAAA,CAAA,EAAO,KAAK,cAAc,EAAQ,SAiBtB,EAAA,UAAA,cAAd,SACI,EAAwB,EAAmB,EAC3C,EACA,GAF2C,MAAA,KAAA,QAAA,GAAA,IAC3C,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,uGACF,MAAK,IACH,GAAS,KAAK,UAAU,GACxB,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,IAGd,EAAU,GAAI,IAChB,KAAK,UAAW,EAAgB,EAChC,KAAK,qBAKS,CAAA,EAAM,KAAK,uBACzB,EAAQ,EAAS,EAAS,WADxB,SAAY,EAAA,OAEZ,EAAU,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAGzD,EAAY,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAW,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,GAAO,GAAM,KACxD,EACF,GAAI,KAAgB,EAAS,OAAK,EAAa,KAAK,YACxD,OAAO,KAAK,GAAW,QAAQ,SAAA,GAC7B,GAAM,GAAc,EAAU,GAC9B,EAAY,QAAQ,SAAA,GAClB,AAAI,GAAU,CAAC,EAAO,YAAc,CAAC,EAAQ,IAAI,EAAO,KACtD,EAAO,cAKb,AAAI,KAAK,QAAU,MACjB,EAAQ,QAAQ,GAGlB,CAAA,EAAO,SAGH,EAAA,UAAA,qBAAN,SACI,EAAkB,EAClB,mFACI,SAAe,EAAO,OAAO,SAAC,EAAK,EAAQ,GAC/C,SAAI,EAAK,OAAO,GAAO,MAAQ,EACxB,GACN,IAEH,CAAA,EAAO,KAAK,cACR,EAAc,KAAK,YAAa,GAAM,EAAgB,SAS9C,EAAA,UAAA,uBAAd,SACI,EAAwB,EAA2B,EACnD,0IACI,EAAQ,OAAO,KAAK,GACpB,EACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAY,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC9D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAC3D,EACF,GAAqB,EAAQ,EAAa,KAAK,WAD5C,EAAS,EAAA,UAAE,EAAa,EAAA,cAAE,EAAW,EAAA,YAAE,EAAU,EAAA,WAGlD,EACE,EAAU,OAAK,KAAK,MAAM,SAAS,IAAI,SAAA,GACzC,MAAO,CAAC,KAAI,EAAE,SAAU,EAAQ,kBAEhC,EAAU,GAAA,GAAwB,KAAK,WAC7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAEnB,EAA2D,GAC3D,EAAgB,KAAK,mBAAmB,GACxC,EAAkC,0BACjC,GAAM,OAAS,EACd,GAAW,KAAK,aAClB,EAAY,EAAO,EAAS,EAAY,EAAO,EAC/C,EAAiB,EAAiC,GACtD,CAAA,EAAM,QAAQ,IAAI,KAJG,CAAA,EAAA,UAIrB,SAAA,oBAaF,GAXA,AAAI,GAAe,MAAQ,CAAC,GAC1B,QAAQ,KACJ,mIAGA,EACF,EACK,OACG,SAAA,GAAQ,MAAA,CAAC,GAAc,IACnB,CAAC,GAAU,EAAK,KAAM,EAAY,KACzC,IAAI,SAAA,GAAQ,MAAA,GAAK,OACtB,EAAe,OAAS,EACtB,QAAiB,GACrB,AAAI,GAAe,MACjB,GACI,gEACA,4BAA2B,EAAU,MAErC,GAAI,OACN,+BAA+B,EAAc,uBAC7C,YAAW,EAAK,gDAChB,KAAI,EAAa,MAAM,IAE7B,MAAA,CAAA,EAAO,SAGD,EAAA,UAAA,aAAR,SACI,EAAoB,EAA2B,EAC/C,EAA4B,EAC5B,EAA4B,EAC5B,EACA,GAEF,OAPF,GAAA,KAMQ,EAAqC,gBAEzC,GAAM,GAAO,EAAM,MACnB,EAAQ,eAAiB,EAAK,SAC9B,GAAI,GAAW,GAUf,GANA,AAAI,EAAK,KAAK,KAAO,SACjB,EAAc,aAAc,EAAK,KAAM,EAAW,IACnD,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAIC,EAAW,QAAQ,EAAK,QAAU,IACpC,GAAM,GAAU,GAAU,EAAK,KAAM,EAAW,GAChD,AAAK,GACF,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAEH,GAAM,GAAiB,EAAQ,eAC/B,AAAI,YAAmB,SACrB,EAAS,KAAK,EAAQ,KAAK,SAAA,GACzB,SAAU,GAAY,EACtB,EAAQ,eAAiB,EACzB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,GAC1C,KAGT,GAAU,GAAY,EACtB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,QAGnD,GAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,WAxC9C,EAAM,OAAS,OA2CtB,MAAO,IAGD,EAAA,UAAA,kBAAR,SACI,EAAY,EAA2B,EACvC,EAA4B,EAC5B,GACF,EAAK,SAAS,QAAQ,SAAC,GACd,GAAA,GAAA,GAAA,EAAA,KAAA,GAAA,GACP,GAAI,EAAM,IAAa,CAAC,EAAU,IAAI,EAAU,MAC9C,OAGF,AAAI,EAAU,KAAO,QACnB,AAAI,EAAU,WAAW,KAAK,SAAA,GACxB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAExC,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,KAGpD,AAAI,EAAU,WAAW,MAAM,SAAA,GACzB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAE5C,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,QAQ1D,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,OAAO,KAAK,KAAK,WACZ,QACG,SAAA,GAAO,MAAA,GAAK,UAAU,GAAK,QAAQ,SAAA,GAAU,MAAA,GAAO,eAGtD,EAAA,UAAA,uBAAR,SAA+B,GAA/B,GAAA,GAAA,KACE,OAAO,KAAK,GAAQ,QAAQ,SAAA,GAC1B,GAAM,GAAQ,EAAO,GACd,EAAA,GAAA,GAAA,GACD,EAAO,EAAK,MAAM,MAAM,GAC9B,GAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,GAAM,GAAQ,EAAK,WAAW,MAAS,MACjC,EAAQ,EAAM,SAAW,EAAM,MAAM,QACvC,EAAM,MAAM,MACR,SAAC,EAAK,GAAU,MAAA,GAAM,KAAW,IAAM,EAAM,KAAW,IAChE,EAAA,KAAK,OACD,EACA,WAAM,MAAA,sBAAsB,EAAK,KAAI,kBACjC,iCAAgC,EAAK,eACrC,KAAI,EAAM,MAAK,OAEzB,AAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,EAAA,KAAK,OACD,EAAM,QAAU,EAAK,WAAW,MAAS,MACzC,WAAM,MAAA,sBAAsB,EAAK,KAAI,8CAE9B,GAAK,WAAW,MAAS,MAAK,aAAa,EAAM,YAK1D,EAAA,UAAA,UAAR,SAAkB,GAChB,GAAM,GAAyB,GAC/B,OAAW,KAAa,GACtB,GAAI,KAAK,YAAc,MAAQ,KAAK,WAAW,QAAU,MACrD,KAAK,WAAW,OAAO,IAAc,MACvC,GAAM,GAAS,KAAK,WAAW,OAAO,GACtC,EAAO,EAAO,MAAQ,EAAO,OAE7B,GAAO,GAAa,EAAO,GAG/B,MAAO,IAGD,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAa,OAAO,KAAK,GAAQ,OAAO,SAAA,GACrC,GAAA,GAAA,GAAA,GAAA,GACP,MAAO,GAAK,MAAM,MAAM,IAAa,OAEvC,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,OACN,gDACA,WAAU,EAAU,kCAIpB,EAAA,UAAA,WAAR,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GACjB,GAAI,EAAK,YAAc,MAAQ,EAAK,WAAW,SAAW,MACtD,EAAK,WAAW,QAAQ,IAAS,MACnC,GAAM,GAAS,EAAK,WAAW,QAAQ,GACvC,MAAO,GAAO,KAEhB,MAAO,IACN,KAGG,EAAA,UAAA,aAAR,SAAqB,GAArB,GAAA,GAAA,KACE,EAAQ,QAAQ,SAAA,GACP,GAAA,GAAA,GAAA,GAAA,GACP,GAAI,CAAC,EAAK,MAAM,MAAM,GACpB,KAAM,IAAI,OAAM,eAAe,EAAI,kCAI3C,KC3lBA,AAyBO,GAAM,IAAqB,oBACrB,GAAqB,2BAoDhC,WACY,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IADA,KAAA,SAAA,EACA,KAAA,YAAA,EAzCJ,KAAA,QAAU,MA0ChB,AAAI,GAAe,MACjB,MAAK,YAAc,IArCvB,cAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,yCAGd,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,SAAS,4CAGvB,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,6CAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,SAAS,wCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,yCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,2CAoBf,EAAA,UAAA,cAAR,WACE,GAAM,GAAO,KAAK,SAClB,GAAK,EAAsB,MAAQ,KAEjC,KAAK,QAAU,UACN,KAAK,YAAY,aAAe,KACzC,KAAK,QAAU,EAAA,GAAG,mBAAmB,EAAgB,KAAK,kBAE1D,GAAM,GAAW,EAAA,GAAG,gBAAgB,EAAgB,KAAK,aACzD,GAAI,EAAS,SAAW,EAGtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAgB,KAAK,sBAChD,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,CAAC,GAAK,MAEpB,KAAK,QAAU,EAAS,KAQtB,EAAA,UAAA,KAAN,2GAEE,GADA,KAAK,gBACD,KAAK,QAAQ,MAAQ,KACvB,KAAM,IAAI,OACN,iHAGY,MAAA,CAAA,EAAM,KAAK,QAAQ,eAA/B,SAAY,EAAA,OAElB,CAAA,EAAO,KAAK,SAAS,UASvB,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,UAAY,EACjB,GAAM,GAAQ,KAAK,UAAU,cACzB,EAAY,GAChB,AAAI,KAAK,UAAU,qBAAuB,MACxC,GACK,KAAK,UAAU,oBAA4B,WAIlD,KAAK,QAAa,EAAM,SAAS,SAAQ,IAAI,EAAM,SAAS,YAC5D,GAAM,GACF,EAAA,GAAG,cAAc,KAAK,UAAU,WAAY,KAAK,UAAU,aAK/D,GAJA,KAAK,SAAW,GAAI,IAChB,GAAgB,SAAS,eAAe,EAAO,IACnD,KAAK,SAAS,UAAY,KAAK,6BAA6B,GAExD,EAAU,kBAAoB,MAChC,GAAM,GACF,GAAgB,SAAS,eAAe,EAAU,kBACtD,KAAK,YAAc,GAAI,IAAc,GACrC,KAAK,YAAY,UAAY,KAAK,SAAS,UAC3C,KAAK,YAAY,QAAQ,GAAI,IAG/B,MAAO,IA+CH,EAAA,UAAA,KAAN,SAAW,EAAmC,4EAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,OACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,OACN,+GAIN,MAAA,CAAA,EAAO,EAAa,KAAK,KAAK,iBAyChC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,MAAO,MAAK,QAAQ,EAAQ,KAAK,cAG3B,EAAA,UAAA,gBAAR,SAAwB,GAEtB,GAAI,CAAE,aAAkB,GAAA,SAAW,CAAC,MAAM,QAAQ,GAEhD,MAAO,GAGT,GADA,EAAS,MAAM,QAAQ,GAAU,EAAS,CAAC,GACvC,EAAO,SAAW,KAAK,WAAW,OACpC,KAAM,IAAI,OACN,+BACA,wBAAuB,KAAK,WAAW,OAAM,mBAC7C,oBAAmB,EAAO,OAAM,oBAEtC,MAAO,MAAK,WAAW,OAAO,SAAC,EAAK,EAAW,GAC7C,SAAI,GAAc,EAAoB,GAC/B,GACN,KAGG,EAAA,UAAA,iBAAR,SAAyB,GACvB,SAAU,GAAW,KAAK,YACnB,AAAC,MAAM,QAAQ,GAAuB,EAAZ,CAAC,IAmBpC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,EAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GAChC,GAAM,GAAS,KAAK,SAAS,QAAQ,EAAQ,GAC7C,MAAO,GAAO,OAAS,EAAI,EAAS,EAAO,IAkBvC,EAAA,UAAA,aAAN,SACI,EACA,mGACF,SAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GACjB,CAAA,EAAM,KAAK,SAAS,aAAa,EAAQ,WAAlD,SAAS,EAAA,OACf,CAAA,EAAO,EAAO,OAAS,EAAI,EAAS,EAAO,UAGrC,EAAA,UAAA,6BAAR,SAAqC,GACnC,MAAO,QAAO,KAAK,GAAK,OAAO,SAAC,EAAyB,GACvD,SAAO,GAAO,CAAC,EAAI,IACZ,GACN,KAQL,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UAEd,AAAI,KAAK,aACP,KAAK,YAAY,WAGvB,iBAiCI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,6FACF,GAAI,GAAY,KACd,KAAM,IAAI,OACN,0GAGN,MAAI,IAAW,MACb,GAAU,IAGZ,AAAI,EAAQ,WACV,CAAK,EAA0B,MAAQ,MACrC,CAAM,EAAoB,SAAS,MACjC,GAAY,EAAsB,KAEpC,EAAW,GAAG,EAAW,GAAqB,KAG5C,EAAQ,GAAI,IAAW,EAAU,GACvC,CAAA,EAAM,EAAM,eAAZ,SAAA,OACA,CAAA,EAAO,QC7ZT,GAGM,IAAU,8hECwBhB,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAEZ,GAAQ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,wBACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,YAkDwB,EAAY,GAElC,MAAO,IAAgB,EAAO,GAShC,YACI,EAAY,EACZ,EAAiC,GAEnC,GAFE,AAAA,IAAA,QAAA,GAAA,GAA0B,MAAO,AAAA,IAAA,QAAA,GAAA,GAA2B,MAE1D,GAAS,KACX,MAAO,MAET,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAI,EAAK,IAAI,GACX,MAAO,GAAK,IAAI,GAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAGL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,GAChB,OAAW,KAAK,IACd,GAAM,GAAQ,EAAM,GACd,EAAc,GAAgB,EAAO,EAAO,EAAM,GACxD,EAAe,GAAK,EAEtB,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAdzD,UAAK,IAAI,EAAO,EAAO,OAChB,EAAO,kBA2Cd,EAAe,GAAA,MAAA,KAAA,QAAA,GAAA,IACV,GAAgB,EAAQ,GAOjC,YACI,EAAe,EACf,GAAA,AAAA,IAAA,QAAA,GAAA,GAA2B,MAG7B,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAEL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,kBACL,GACT,GAAM,GAAW,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC7B,EAAc,GAAgB,EAAU,EAAO,GACrD,EAAe,GAAK,GAHtB,OAAW,KAAK,KAAL,GAKX,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAbzD,OAAO,GAAO,kBAkBQ,GACxB,MAAI,KAAM,KACD,KAIL,GAAW,EAAE,IACR,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAG,QAAS,gBAsC3B,EAAY,+GACR,EAA+B,GAAI,KAGzC,GAAgB,EAAO,EAAO,OAMZ,EAAA,MAAM,KAAK,EAAK,+BAAhB,GAAA,EAAA,OAAP,GAAG,EAAA,GACN,EAAQ,EAAK,IAAI,GACnB,YAAiB,SACC,CAAA,EAAM,GADxB,CAAA,EAAA,IAFmC,CAAA,EAAA,UAG/B,EAAc,EAAA,OACpB,EAAK,IAAI,EAAK,oBAJA,wBAWZ,SAAS,GAAgB,EAAO,EAAO,GAC7C,CAAA,EAAO,oBASkB,GACzB,MAAO,IAAO,MAAS,CAAC,YAAY,OAAO,IACtC,OAAM,QAAQ,IACb,MAAO,IAAQ,UAAY,CAAE,aAAe,IAAA,qBAYvB,GAC3B,MAAO,IAAO,MAAQ,GAAY,IAAQ,MAAM,QAAQ,IACnD,MAAO,IAAQ,UAAa,YAAe,IAAA,QAC5C,GAAA,KAAQ,aAAa,GAO3B,YAAqB,GACnB,MACI,KAAU,MACT,MAAO,IAAU,UAAY,MAAO,IAAU,WCnRrD,YAqB6B,GAC3B,MAAO,IAAQ,EAAW,IAI5B,YAAuB,GACrB,MAAI,aAAgB,IAAA,OACV,CAAC,MAAO,EAAK,QAAS,QAAS,IAC9B,GAAW,GACb,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAM,QAAS,IChClC,AAqBA,GAAA,IAAA,WAcE,WAAmB,GACjB,GADiB,KAAA,SAAA,EAVT,KAAA,MAAQ,EACR,KAAA,IAAM,EAUV,GAAY,KACd,KAAM,IAAI,YAAW,mDAEvB,GAAI,EAAW,EACb,KAAM,IAAI,YAAW,6CAEvB,KAAK,KAAO,GAAI,OAAS,GACzB,KAAK,gBAAkB,EAAI,EAMnB,SAAA,UAAA,KAAV,SAAe,GAEb,KAAO,EAAQ,GACb,GAAS,KAAK,gBAEhB,MAAO,GAAQ,KAAK,iBAGZ,EAAA,UAAA,IAAV,SAAc,GACZ,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,MAAO,MAAK,KAAK,EAAQ,KAAK,WAGtB,EAAA,UAAA,IAAV,SAAc,EAAe,GAC3B,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,KAAK,KAAK,EAAQ,KAAK,UAAY,GAMrC,EAAA,UAAA,OAAA,WACE,GAAI,GAAS,KAAK,IAAM,KAAK,MAC7B,MAAI,GAAS,GACX,GAAS,KAAK,gBAAkB,GAE3B,GAQT,EAAA,UAAA,OAAA,WACE,MAAO,MAAK,WAAa,KAAK,UAQhC,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,WAAa,GAM3B,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,IAAI,KAAK,IAAK,GACnB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,IAMlC,EAAA,UAAA,QAAA,SAAQ,GACN,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,KAAK,KAOd,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,GAChC,GAAM,GAAS,KAAK,IAAI,KAAK,KAC7B,YAAK,IAAI,KAAK,IAAK,QACZ,GAMT,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GACpC,KAAK,IAAI,KAAK,MAAO,IAMvB,EAAA,UAAA,MAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAS,KAAK,IAAI,KAAK,OAC7B,YAAK,IAAI,KAAK,MAAO,QACrB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GAC7B,GAYT,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAQ,KAAK,KAAK,KAAK,MAAQ,GAC/B,EAAS,KAAK,IAAI,GACxB,YAAK,IAAI,EAAO,KAAK,OACd,GAEX,KC/KA,AAoBA,GAAA,IAAA,SAAA,GAA0C,GAAA,EAAA,GAMxC,mBACE,GAAA,KAAA,KAAM,EAAkB,mBAAiB,KAG3C,SAAA,UAAA,OAAA,WACE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,GACH,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,KAAI,KAAA,KAAC,IAGb,EAAA,UAAA,QAAA,SAAQ,GACN,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,QAAO,KAAA,KAAC,IAMR,EAAA,UAAA,OAAR,WAOE,OANM,GAAc,KAAK,SAAW,EAC9B,EAAU,GAAI,OAAS,GACvB,EAAM,KAAK,SAIR,EAAI,EAAG,EAAI,EAAK,IACvB,EAAQ,GAAK,KAAK,IAAI,KAAK,KAAK,KAAK,MAAQ,IAG/C,KAAK,KAAO,EACZ,KAAK,SAAW,EAChB,KAAK,gBAAkB,EAAI,KAAK,SAChC,KAAK,MAAQ,EACb,KAAK,IAAM,GA7CE,EAAA,iBAAmB,GA+CpC,GAhD0C,ICpB1C,YAuCqC,GACnC,MAAO,IAAI,IAAc,eAyBvB,GAEF,MAAO,IAAI,IAAqB,eAgB9B,EACA,GACF,MAAO,IAAI,IAAgB,EAAe,eAmDxC,EACA,GAAA,MAAA,KAAA,QAAA,GAAgC,GAAgB,MAC3C,GAAI,IAAe,EAAW,GAUvC,GAAA,IAAA,WAAA,cAwBQ,SAAA,UAAA,QAAN,6GACQ,SAAc,GACZ,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,KAAK,gBAAf,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAcH,EAAA,UAAA,eAAN,+GACQ,SAAS,KAAK,SAAS,KACvB,EAAc,GACZ,CAAA,EAAM,EAAO,eAAjB,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,EAAO,gBAAjB,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAUH,EAAA,UAAA,aAAN,2GACU,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACR,CAAA,EAAM,KAAK,eAAf,SAAI,EAAA,oCAWF,EAAA,UAAA,aAAN,SAAmB,qGACT,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,8BACzB,CAAC,EAAE,MAAS,EACd,CAAA,EAAM,KAAK,QADiB,CAAA,EAAA,UAChC,SAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,oCAgBjC,EAAA,UAAA,aAAA,SAAa,GACX,MAAO,IAAI,IAA0B,KAAM,IAa7C,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,IAAI,IAAe,KAAM,IAWlC,EAAA,UAAA,IAAA,SAAO,GACL,MAAO,IAAI,IAAY,KAAM,IAW/B,EAAA,UAAA,SAAA,SAAY,GACV,MAAO,IAAI,IAAiB,KAAM,IAWpC,EAAA,UAAA,eAAA,SAAkB,GAChB,MAAO,IAAI,IAAiB,KAAM,GAAW,UAW/C,EAAA,UAAA,QAAA,SAAW,GACT,MAAO,IAAI,IAAgB,KAAM,IAQ7B,EAAA,UAAA,aAAN,SAAmB,sEACjB,MAAA,CAAA,EAAO,KAAK,IAAI,GAAG,qBAUf,EAAA,UAAA,cAAN,SAAoB,sEAClB,MAAA,CAAA,EAAO,KAAK,eAAe,GAAG,aAAa,SAAA,GAAK,MAAC,KAAM,WAqBzD,EAAA,UAAA,cAAA,SAAc,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,IACxB,GAAI,IAAsB,KAAM,EAAW,IAmCpD,EAAA,UAAA,iBAAA,SACI,EAAmB,EAEnB,GAFmB,AAAA,IAAA,QAAA,GAAA,IAEnB,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAAa,KAAK,cAAc,EAAW,GAGjD,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,MAaxC,EAAA,UAAA,YAAA,SACI,EACA,GACF,MAAO,IAAI,IACP,GAAkB,CAAC,KAAM,IAAY,IAU3C,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAShC,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAYhC,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,IAAI,IAAiB,KAAM,IAapC,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAC1B,MAAO,IAAI,IAAgB,KAAM,EAAY,IAO/C,EAAA,UAAA,OAAA,WACE,MAAO,IAAI,IAAe,OAE9B,KAWA,GAAA,SAAA,GAA+B,GAAA,EAAA,GAE7B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,MAAA,EADd,EAAA,KAAO,IAKf,SAAA,UAAA,QAAA,WACE,MAAO,YAAY,KAAK,MAAM,OAAM,UAGhC,EAAA,UAAA,KAAN,oFACE,MAAI,MAAK,MAAQ,KAAK,MAAM,OAC1B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAO,KAAK,MAAM,KAAK,MAC7B,KAAK,OACL,CAAA,EAAO,CAAC,MAAO,GAAU,GAAO,KAAM,WAE1C,GAlB+B,IAoB/B,GAAA,SAAA,GAAsC,GAAA,EAAA,GACpC,WACc,GADd,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,OAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAO,iBAGH,EAAA,UAAA,KAAN,8EACE,IACE,MAAA,CAAA,EAAO,KAAK,gBACL,GAEP,QAAE,QACE,mDAAmD,EAAE,QACnD,iBAGZ,GApBsC,IAsBtC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,8EACE,MAAA,CAAA,EAAO,KAAK,SAAS,aAEzB,GA1BgC,IA4BhC,GAAA,SAAA,GAA8B,GAAA,EAAA,GAQ5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAF3D,EAAA,MAAQ,EAIN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,iHAKS,MAAK,QAAU,KAAK,SACT,CAAA,EAAM,KAAK,SAAS,QADH,CAAA,EAAA,UAGjC,MAFM,GAAU,EAAA,OAEZ,EAAQ,KACV,CAAA,EAAO,GAET,IAAA,QAAW,EAAQ,qBAErB,MAAA,CAAA,EAAO,KAAK,SAAS,cAEzB,GAzC8B,IA2C9B,GAAA,SAAA,GAA8B,GAAA,EAAA,GAE5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAD3D,EAAA,MAAQ,IAKR,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,8EACE,MAAI,MAAK,SAAW,KAAK,SACvB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,CAAA,EAAO,KAAK,SAAS,aAEzB,GAhB8B,IAqB9B,GAAA,SAAA,GAAuC,GAAA,EAAA,GAKrC,WACc,EAAqC,EACrC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EAAqC,EAAA,UAAA,EACrC,EAAA,qBAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,qBAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,6GACQ,EAAa,0BACZ,GAAM,OAAS,KAAK,UACZ,CAAA,EAAM,KAAK,SAAS,QADC,CAAA,EAAA,UAElC,MADM,GAAO,EAAA,OACT,EAAK,KACH,KAAK,sBAAwB,EAAM,OAAS,EAC9C,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,KAE9B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAM,KAAK,EAAK,qBAElB,MAAA,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,WAEhC,GAvCuC,IAyCvC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,2GAEiB,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,MAAQ,KAAK,UAAU,EAAK,OACnC,CAAA,EAAO,GAET,IAAA,QAAW,EAAK,qCAGtB,GAlCgC,IAoChC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAC9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,WAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAc7B,IAZM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAOzD,EAAS,KAAK,UAAU,EAAK,OAC7B,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCgC,IAqChC,GAAA,SAAA,GAA2C,GAAA,EAAA,GAEzC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,QAAA,EAHd,EAAA,MAAQ,EAKN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,oBAO7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGR,EAAA,UAAA,WAAN,yJAGa,CAAA,EAAM,KAAK,SAAS,eAA3B,MAAA,CAAA,EAAO,EAAA,eAEP,kBAAI,AAAC,KAAK,QAAQ,SAChB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,+CAUrC,GA1C2C,IA4C3C,GAAA,SAAA,GAAqC,GAAA,EAAA,GACnC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,KACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAe,GAAA,YAAe,sBAAsB,EAAK,OAOhD,CAAA,EAAM,KAAK,UAAU,EAAK,gBAKzC,IALM,EAAS,EAAA,OACT,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCqC,IAgDrC,GAAA,SAAA,GAAmD,GAAA,EAAA,GAQjD,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,YAAc,GAAI,IACvB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAiBR,EAAA,UAAA,WAAN,2GAIS,MAAK,YAAY,WAAa,EAE9B,CAAA,EAAM,KAAK,QAFoB,CAAA,EAAA,UAEpC,MAAK,GAAA,aACH,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,YAG/B,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,YAAY,QAAS,KAAM,WAEnD,GAjDmD,IAkDnD,GAAA,SAAA,GAAoC,GAAA,EAAA,GAClC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,eAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,IAeT,IAbM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAMzD,EAAc,KAAK,UAAU,EAAK,OAClC,EACF,GAAA,YAAe,sBAAsB,GACzC,KAAK,YAAY,QAAQ,GAIzB,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAIN,MAAA,CAAA,EAAO,UAEX,GArCoC,IAgDpC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAStC,WACI,EACiB,GAFrB,GAAA,GAGE,EAAA,KAAA,OAAO,KADY,SAAA,iBAAA,EARb,EAAA,SAAuC,KAGvC,EAAA,SAA4B,KAOlC,EAAK,cAAgB,IAGvB,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,8CAC1B,MAAU,GAAiB,eAGvB,EAAA,UAAA,KAAN,8EACE,YAAK,SAAW,KAAK,cAAc,KAAK,UACxC,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,cAAd,SAA4B,qGAO1B,MAAA,CAAA,EAAM,gBAAN,GAAA,OACI,KAAK,UAAY,KACI,CAAA,EAAM,KAAK,cAAc,QAD9C,CAAA,EAAA,UAEF,GADM,EAAiB,EAAA,OACnB,EAAe,KAEjB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,KAAK,SAAW,EAAe,MAC/B,AAAI,KAAK,kBAAoB,MAC3B,MAAK,SAAW,KAAK,SAAS,aAAa,KAAK,oCAGjC,MAAA,CAAA,EAAM,KAAK,SAAS,eACvC,MADM,GAAa,EAAA,OACf,EAAW,KACb,MAAK,SAAW,KAChB,CAAA,EAAO,KAAK,cAAc,KAE5B,CAAA,EAAO,SAEX,GApDwC,IAsD5B,GAAZ,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,YAHU,IAAA,IAAe,KAmC3B,GAAA,IAAA,SAAA,GAAwD,GAAA,EAAA,GAItD,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAgC,GAAgB,MAFvE,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,UAAA,EACA,EAAA,aAAA,EALf,EAAA,MAAQ,EACR,EAAA,eAA6C,OAQrD,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,0CAC1B,MAAO,IAAI,EAAiB,YAGhB,EAAA,UAAA,UAAd,SAAwB,2CAWtB,WAAiB,GACf,GAAI,YAAqB,KACvB,GAAM,GAAS,EAAU,OACzB,MAAO,CACL,MAAO,EAAO,KAAK,SAAA,GACjB,WACA,AAAI,EAAE,MACJ,IAEK,EAAE,QAEX,QAAS,QAGX,OAAO,CAAC,MAAO,KAAM,QAAS,gEArBlC,MAAA,CAAA,EAAM,UAAN,SAAA,OAII,EAAe,EACf,EAAgB,EAoBF,CAAA,EAAM,GAAmB,KAAK,UAAW,WAE3D,GAFM,EAAY,EAAA,OAEd,IAAiB,EAEnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAI,EAAgB,EAClB,OAAQ,KAAK,kBACN,IAAgB,KACnB,KAAM,IAAI,OACN,+CACA,0BAAyB,KAAK,MAAK,UACpC,IAAgB,SACnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,SACxB,IAAgB,SAMzB,YAAK,QACL,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAGzB,EAAA,UAAA,KAAN,8EACE,YAAK,eAAiB,KAAK,UAAU,KAAK,gBAC1C,CAAA,EAAO,KAAK,qBAEhB,GAxEwD,IAoFxD,GAAA,SAAA,GAAyC,GAAA,EAAA,GAGvC,WACc,EAAqC,GADnD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAAqC,EAAA,WAAA,EAEjD,EAAK,OAAS,GAAI,IAAuC,KAG3D,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAOzB,EAAA,UAAA,OAAV,WACE,KAAO,CAAC,KAAK,OAAO,WAClB,GAAM,GAAI,KAAK,SAAS,OACxB,KAAK,OAAO,KAAK,KAIrB,EAAA,UAAA,KAAA,WACE,YAAK,SAIE,KAAK,OAAO,SAEvB,GA/ByC,IAuCzC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAUtC,WACc,EAAqC,EAC/C,GAFJ,GAAA,GAGE,EAAA,KAAA,KAAM,EAAU,IAAW,KAFf,SAAA,SAAA,EAAqC,EAAA,WAAA,EAH3C,EAAA,kBAAoB,GAM1B,EAAK,OAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACpD,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGN,EAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,MAAK,MAAM,KAAK,SAAW,IAG1B,EAAA,UAAA,YAAV,WACE,MAAO,MAAK,UAAU,KAAK,OAAO,WAG9B,EAAA,UAAA,WAAN,6GAEE,AAAK,KAAK,mBACR,KAAK,gCAEC,MAAK,OAAO,UAAS,CAAA,EAAA,GACrB,GAAc,KAAK,cACV,CAAA,EAAM,KAAK,OAAO,cAAc,YAC/C,GADM,EAAS,EAAA,OACX,EAAO,KACT,KAAK,kBAAoB,OAEzB,aAAK,SACL,CAAA,EAAO,sBAGX,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,WAE/B,GApDwC,ICjnCxC,kBA6DA,aAWW,KAAA,KAAe,KA+DxB,SAAA,UAAA,MAAA,SAAM,EAAmB,GAAzB,GAAA,GAAA,KAAyB,AAAA,IAAA,QAAA,GAAA,IACvB,GAAM,GAAO,KACb,GAAA,KAAQ,OACJ,EAAY,EAAG,WAAM,MAAA;QACrB,IACJ,GAAI,GACJ,MAAI,MAAK,OAAS,UAAY,KAAK,MAAQ,KAGzC,EAAO,KAAK,KACP,AAAI,EAGT,EAAO,KAAK,KAAK,KAAK,KAAO,GAI7B,EAAO,KAAK,MAAM,KAAK,KAAO,GAEzB,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OACH,iBAAiB,EAAW,EAAgB,WAChD,IAkBL,EAAA,UAAA,YAAA,SAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,UAAY,EAAQ,OAAS,SAG7C,EAAO,SACF,AAAI,KAAK,MAAQ,MAAQ,EAAQ,MAAQ,KAG9C,EAAO,KAAK,KAAO,EAAQ,KAI3B,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,EAAA,MAAA,IAAA,KAAA,SAAA,0BACK,MAAA,CAAA,EAAM,EAAK,mBAAZ,SAAA,GAAC,EAAA,QAAuB,YAAY,CAAA,EAAM,EAAQ,mBAAlD,MAAA,CAAA,EAAA,EAAA,MAAA,EAAA,CAAoC,EAAA,gBACxC,IAmBN,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,SAEhB,EAAO,SAIP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,OAAO,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAClE,IAmBC,EAAA,UAAA,aAAN,SAAmB,6FACT,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,aAAa,UAkB9C,EAAA,UAAA,IAAA,SAAkC,GAAlC,GAAA,GAAA,KACQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,IAAI,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAC/D,KAAK,OA0BV,EAAA,UAAA,SAAA,SAAuC,GAAvC,GAAA,GAAA,KAEQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,SAAS,UACvC,KAAK,OAYV,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,GAAI,GAAc,KAChB,KAAM,IAAI,YACN,6DAGN,GAAM,GAAO,KACb,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,SAAS,UAAa,KAAK,OAqBrE,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,EAAQ,EAI/B,EAAO,KAAK,KAAO,EACd,AAAI,IAAU,EAEnB,EAAO,EACF,AAAI,KAAK,MAAQ,MAAS,KAAU,QAAa,EAAQ,GAG9D,EAAO,SAGP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mDACrB,SAAmB,GACrB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,MAAA,IAAA,KAAA,SAAA,sCAAqB,CAAA,EAAM,EAAK,mBAApB,MAAA,CAAA,EAAE,GAAA,MAAO,EAAA,OAAuB,EAAA,KAAM,GAAK,WAC3D,CAAA,EAAO,GAAyB,EAAiB,KAAK,UACrD,IAoBL,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,GAAS,GAAK,KAAK,MAAQ,EAIlD,EAAO,KAAK,KAAO,EACd,AACH,KAAK,MAAQ,MACZ,MAAK,KAAO,GAAS,IAAU,QAAa,EAAQ,GAGvD,EAAO,EAGP,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IA4BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,EAAe,GAA3C,GAAA,GAAA,KAEE,GAFyC,AAAA,IAAA,QAAA,GAAA,IAErC,GAAc,MAAQ,EAAa,EACrC,KAAI,MAAK,MAAQ,KACT,GAAI,YACN,4DAEE,GAAI,YACN,+KAGA,oCAAmC,KAAK,KAAI,eAGpD,GAAM,GAAO,KACP,EAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACrD,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACvB,SAAQ,EAAO,QACnB,AAAI,GACF,IAAS,EAAO,SAEV,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,QAAQ,EAAY,EAAM,mBACxD,KAAK,OAoBV,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,KAAK,KAAO,EAGnC,EAAO,EACF,AAAI,KAAK,MAAQ,MAAQ,KAAK,MAAQ,EAG3C,EAAO,KAAK,KAGZ,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IAmBjD,EAAA,UAAA,QAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,iBAc3B,EAAA,UAAA,eAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,wBA1HjB,EAAA,gBAAkB,IA4HpC,iBAeI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,GAAA,UAAA,GAAkB,GAAA,EAAA,GAAd,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KACT,SAAA,KAAO,IAMD,SAAA,UAAA,SAAN,8EACE,MAAA,CAAA,EAAO,UAEX,GAVyB,iBAgCyB,GAApD,GAAA,GAAA,KACE,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,GAAkB,SAAQ,EAAM,oBA4CA,GAAlD,GAAA,GAAA,KAGE,GAAI,CAAC,GAAW,GACd,KAAM,IAAI,OAAM,qDAElB,GAAI,GACJ,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAO,GAAQ,KAAQ,EAAS,GAAkB,KAC5B,KAAK,IAAI,EAAO,EAAS,GAAkB,cAE1D,YAAoB,QAC7B,OAAW,KAAM,GACf,EAAO,GAAQ,KAAQ,EAAS,GAAmB,KAC7B,KAAK,IAAI,EAAO,EAAS,GAAmB,MAGtE,MAAO,IAAyB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACd,MAAA,CAAA,EAAM,GAAmB,EAAU,SAAA,GACjD,GAAI,YAAa,IACf,MAAO,CAAC,MAAO,EAAE,WAAY,QAAS,IACjC,GAAI,GAAW,GACpB,MAAO,CAAC,MAAO,KAAM,QAAS,IAE9B,KAAM,IAAI,OACN,uFAPF,SAAU,EAAA,OAWhB,CAAA,EAAO,GAAsB,EAAS,GAAgB,iBACrD,GAWL,YAAyB,GACvB,GAAI,IAAS,KACX,MAAO,MAIT,GAAM,GAAa,EAAK,GAExB,GAAI,GAAa,IAEf,GAAM,GAAQ,GAAY,GAC1B,MAAO,CAAC,MAAK,EAAE,QAAS,IAI1B,MAAO,CAAC,MAAO,KAAM,QAAS,IAOhC,YAAwD,GAEtD,GAAI,EAAO,SAAW,EAEpB,KAAM,IAAI,OAAM,wCAGlB,MAAI,GAAO,YAAc,IAAA,OAEhB,GAAA,MAAS,GAGT,GAAA,OAAU,GC9rBrB,mBA2BqC,GAAA,EAAA,GAMnC,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,IAIzB,SAAA,UAAA,SAAN,+GACwB,MAAA,CAAA,EAAM,KAAK,MAAM,mBAAjC,SAAgB,EAAA,OAChB,EAAe,EAAc,aAC7B,EAAe,EAAa,MAAM;GAAM,IAAI,SAAA,GAEhD,MAAI,GAAK,SAAS,OAChB,GAAO,EAAK,MAAM,EAAG,KAEhB,IAET,CAAA,EAAO,SAEX,GAtBqC,IC3BrC,AAyBA,GAAM,IAAa,IACb,GAAY,OAAO,OACnB,GAAc,OAAO,SACrB,GAAc,OAAO,SACrB,GAA0B,OAAO,mBACjC,GAA8B,OAAO,+BAeX,GAAA,EAAA,GAgI9B,WAA+B,EAAmB,GAAlD,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,EA9HvB,EAAA,UAAY,GACZ,EAAA,gBAA4B,KAC5B,EAAA,qBAAuB,GACvB,EAAA,cAA+C,KAC/C,EAAA,sBAAwB,GACxB,EAAA,UAAY,IACZ,EAAA,gBAAkB,GA0HxB,EAAK,KAAO,GAAI,IAAgB,GAChC,AAAK,GACH,GAAY,IAEd,EAAK,UAAY,IAAU,YAAc,IACzC,EAAK,gBAAkB,EAAU,YACjC,EAAK,cAAgB,EAAU,cAC/B,EAAK,sBAAwB,EAAU,sBACvC,AAAI,EAAU,gBACZ,IAAA,KAAK,OACD,EAAU,WAAa,KACvB,WACI,MAAA,mEACR,EAAK,gBAAkB,GACvB,EAAK,UAAY,KAEjB,EAAK,UAAY,EAAU,UAAY,EAAU,UAAY,MA9H3D,SAAA,UAAA,YAAN,2GACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEF,MAAA,CAAA,EAAO,KAAK,sBAAwB,OAAO,KAAK,KAAK,eACjB,KAAK,uBAW7B,EAAA,UAAA,eAAd,8HAC8B,MAAA,CAAA,EAAM,KAAK,8BACvC,GADM,EAAsB,EAAA,OACxB,CAAC,KAAK,iBAAmB,CAAC,EAE5B,KAAM,IAAI,OACN,6DA0BN,GAzBO,AAAI,KAAK,iBAAmB,GAEjC,GAAA,KAAK,OACD,EAAoB,SAAW,KAAK,gBAAgB,OACpD,WAAM,MAAA,uCACF,EAAK,gBAAgB,OAAO,WAC5B,kEACW,EAAoB,OAAO,WAAa,OAE7D,AAAK,KAAK,iBACR,MAAK,gBAAkB,GAGnB,EAAkC,KAAK,gBAAgB,OACzD,SAAC,EAAmC,GAClC,SAAS,GAAS,EAAS,GAAQ,GAAM,EAClC,GAET,IACE,EACF,OAAO,KAAK,GAAQ,OAAO,SAAC,GAAS,MAAC,GAAO,GAAQ,IACzD,GAAA,KAAK,OACD,EAAe,SAAW,EAC1B,WAAM,MAAA,iCAAmC,EAAe,aAExD,KAAK,eACP,IAAA,EAAA,EAAkB,EAAA,OAAO,KAAK,KAAK,eAAjB,EAAA,EAAA,OAAA,IAEhB,GAFS,EAAG,EAAA,GACN,EAAQ,KAAK,gBAAgB,QAAQ,GACvC,IAAU,GACZ,KAAM,IAAI,OACN,YAAc,EACd,uEACY,KAAK,gBAAgB,WAAa,MAIxD,YAAK,qBAAuB,aAGhB,EAAA,UAAA,oBAAd,uHACM,MAAK,UACM,CAAA,EAAM,KAAK,KAAK,YAD3B,CAAA,EAAA,UACI,SAAO,EAAA,OACQ,CAAA,EAAM,EAAK,eAChC,GADM,EAAe,EAAA,OACjB,EAAa,KACf,KAAM,IAAI,OAAM,sCAEZ,SAAoB,EAAa,MACjC,EAAU,KAAK,SAAS,EAAW,IACzC,CAAA,EAAO,UAEP,MAAA,CAAA,EAAO,YA0DL,EAAA,UAAA,SAAN,wHACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEU,MAAA,CAAA,EAAM,KAAK,KAAK,mBAAxB,SAAQ,EAAA,OACZ,AAAI,KAAK,WAGP,GAAQ,EAAM,KAAK,IAErB,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,GAAK,gBAAgB,YAG7C,EAAA,UAAA,gBAAA,SAAgB,GAKd,OAJM,GAAS,KAAK,SAAS,GACvB,EAA6C,GAC7C,EAA2C,GAExC,EAAI,EAAG,EAAI,KAAK,gBAAgB,OAAQ,KAC/C,GAAM,GAAM,KAAK,gBAAgB,GAC3B,EAAS,KAAK,cAAgB,KAAK,cAAc,GAAO,KAC9D,GAAI,KAAK,uBAAyB,CAAC,EAEjC,SAEA,GAAM,GAAQ,EAAO,GACjB,EAAc,KAClB,GAAI,IAAU,GAGZ,GAAI,GAAU,EAAO,UAAY,OAC/B,EAAc,EAAO,aAChB,GAAI,GAAW,GAAO,UAAY,EAAO,SAC9C,KAAM,IAAI,OACN,mBAAmB,EAAG,2BAA2B,GAErD,EAAc,YAIhB,GAAM,GAAa,OAAO,GAC1B,GAAI,MAAM,GAGR,AAAI,GAAU,EAAO,QAAU,OAC7B,EAAc,KAAK,WAAW,GAG9B,EAAc,UAEP,CAAC,GAAU,CAAC,EAAO,MAG5B,EAAc,MAId,QAAQ,EAAO,WACR,UACH,EAAc,EACd,UACG,QACH,EAAc,KAAK,MAAM,GACzB,UACG,OACH,EAAc,KAAK,WAAW,GAC9B,cAEA,EAAc,GAKtB,AAAC,GAAU,EAAO,QAAW,EAAO,GAAO,EACd,EAAS,GAAO,EAKjD,MAAI,QAAO,KAAK,GAAQ,SAAW,EAC1B,EAGA,CAAC,GAAI,EAAU,GAAI,IAItB,EAAA,UAAA,WAAR,SAAmB,GACjB,MAAI,KAAU,KAAO,EAAM,gBAAkB,OACpC,EAEA,GAKH,EAAA,UAAA,SAAR,SAAiB,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IAM7B,OALM,GAAmB,GACrB,EAAa,EACX,EAAa,EAAK,OACpB,EAAe,GAEV,EAAI,EAAG,EAAI,EAAY,IAC9B,OAAQ,OAED,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAa,EAAI,EACjB,EAAe,GACf,UAEG,MAAK,UAIR,GAHA,EAAa,EAAI,EAGb,KAAK,YAAc,KAAO,KAAK,gBACjC,MAEF,EAAO,KAAK,IACZ,EAAe,GACf,cAGA,EAAe,GACf,EAAa,EACb,MAEJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,IACvC,EAAe,GACf,EAAa,EAAI,EACjB,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,EAAI,IAC3C,EAAe,GACf,EAAa,EAAI,EACjB,UAEG,IACH,EAAe,GACf,cAGA,EAAe,GACf,MAEJ,UACG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,MAWN,GANA,AAAI,IAAiB,GACnB,EAAO,KAAK,EAAK,UAAU,EAAY,EAAa,IAEpD,EAAO,KAAK,EAAK,UAAU,IAGzB,GAAwB,EAAO,SAAW,KAAK,gBAAgB,OACjE,KAAM,IAAI,OAAM,wCACZ,KAAK,gBAAgB,OAAM,+BAA+B,GAEhE,MAAO,IAEX,GA1VgC,IC7ChC,AA4BA,GAAA,IAAA,SAAA,GAAwC,GAAA,EAAA,GAgBtC,WAAuC,GAAvC,GAAA,GACE,EAAA,KAAA,OAAO,KAD8B,EAAA,iBAAA,EAf/B,EAAA,SAAW,GAiBjB,EAAK,QAAU,EAAiB,SAAW,KAC3C,GAAM,GAAc,KAAK,KAAK,EAAK,SACnC,GAAI,EAAK,QAAU,GAAK,EAAc,GAAK,EAAc,IACrD,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,oDACA,gCAA+B,EAAK,UAc1C,GAXA,EAAK,UAAY,EAAiB,yBAA2B,GAC7D,EAAK,aAAe,EAAiB,aACrC,EAAK,qBACD,EAAiB,sBAAwB,EAAK,QAClD,EAAK,sBAAwB,EAAiB,sBAC9C,EAAK,sBAAwB,EAAiB,uBAAyB,EAEvE,EAAK,mBACD,IAAiB,qBAAuB,IAC5C,EAAK,gBACD,EAAiB,kBAAoB,GACrC,CAAC,EAAK,oBAAsB,CAAC,EAAK,gBACpC,KAAM,IAAI,OACN,iHAKR,SAAA,UAAA,QAAA,WACE,MAAO,cAII,EAAA,OAAb,SAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,6FAClB,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGA,SAAqB,GAAI,GAAmB,GAGlD,CAAA,EAAM,EAAmB,gBAAzB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,8IAEI,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,KAAK,uBAAyB,KAAO,GACA,KAAK,sBACjD,MAAO,aAHT,SAAK,OAAS,EAAA,qCAMR,GAAI,OACN,iDAAiD,EAAE,gBAGzD,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,2CAQlB,GALM,EAED,OAAe,cAAiB,OAAe,mBACpD,KAAK,aAAe,GAAI,GAEpB,CAAC,KAAK,aAGR,KAAK,aAAe,KAAK,aAAa,mBAC7B,KAAK,aAAa,aAAe,KAAK,aAC/C,KAAM,IAAI,OACN,8BACA,cAAa,KAAK,aAAY,MAC9B,YAAW,KAAK,aAAa,aAG7B,SAAe,KAAK,aAAa,wBAAwB,KAAK,QACpE,KAAK,SAAW,KAAK,aAAa,iBAClC,KAAK,SAAS,QAAU,KAAK,QAAU,EACvC,KAAK,SAAS,sBAAwB,KAAK,sBAC3C,EAAa,QAAQ,KAAK,UAC1B,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,CAAA,SAGI,EAAA,UAAA,KAAN,mHACE,MAAI,MAAK,SACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAMN,CAAA,EAAM,KAAK,uBAA5B,SAAiB,EAAA,OACvB,AAAI,KAAK,oBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAoB,KAAK,4BACrB,EAAU,CAAC,KAAK,UAAW,KAAK,qBAAsB,KAE5D,AAAI,KAAK,iBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAiB,KAAK,4BAClB,EAAU,CAAC,KAAK,UAAY,KAAK,QAAS,KAGhD,CAAA,EAAO,CACL,MAAO,CAAC,YAAe,EAAmB,SAAY,GACtD,KAAM,WAMJ,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAIf,EAAA,UAAA,aAAd,+FAEQ,SAAgC,GAChC,EAAgC,GAClC,EAAgB,EACpB,CAAA,EAAO,GAAI,SAAQ,SAAA,GACjB,GAAM,GAAa,YAAY,WAC7B,AAAI,EAAK,oBACP,GAAK,SAAS,sBAAsB,EAAK,UAEzC,AAAI,EAAK,SAAS,KAAO,WACvB,EAAQ,CAAC,cAAa,EAAE,cAAa,IAEvC,EAAc,KAAK,EAAK,SAAS,MAAM,EAAG,EAAK,wBAEjD,AAAI,EAAK,iBACP,GAAK,SAAS,uBAAuB,EAAK,UAC1C,EAAc,KAAK,EAAK,SAAS,UAInC,AAAI,EAAE,IAAkB,EAAK,WAC3B,eAAc,GACd,EAAQ,CAAC,cAAa,EAAE,cAAa,MAEtC,EAAK,QAAU,EAAK,aAAe,aAK1C,EAAA,UAAA,KAAA,WACE,AAAK,KAAK,UACR,MAAK,SAAW,GAChB,KAAK,SAAS,aACd,KAAK,aAAa,QAClB,AAAI,KAAK,QAAU,MAAQ,KAAK,OAAO,YAAY,OAAS,GAC1D,KAAK,OAAO,YAAY,GAAG,SAMjC,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAIlB,EAAA,UAAA,cAAA,WACE,MAAO,MAAK,cAGN,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAY,EAAM,GAAG,OACrB,EAAW,GAAI,cAAa,EAAM,OAAS,GACjD,SAAM,QAAQ,SAAC,EAAM,GAAM,MAAA,GAAS,IAAI,EAAM,EAAI,KAC3C,GAGD,EAAA,UAAA,4BAAR,SAAoC,EAAwB,GAE1D,GAAM,GAAO,GAAI,cAAa,GAAA,KAAK,cAAc,IAEjD,SAAK,IAAI,EAAU,EAAK,OAAS,EAAS,QACnC,GAAA,OAAO,EAAM,IAExB,GA7MwC,IC5BxC,AA0BA,GAAA,IAAA,SAAA,GAAoC,GAAA,EAAA,GAQlC,WACuB,EACA,GAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KACP,GAHqB,EAAA,mBAAA,EACA,EAAA,aAAA,EATf,EAAA,SAAW,GAEX,EAAA,OAAS,GASX,EAAK,eAKP,GAJA,EAAK,OAAS,GACd,EAAK,SACD,CAAC,EAAK,aAAa,aAAc,EAAK,aAAa,aACvD,EAAK,WAAa,GAAA,SAAS,CAAC,GAAI,SAC5B,EAAK,aAAa,YAEpB,GAAM,GACF,EAAK,aAAa,YAAc,EAAM,EAAK,mBAAmB,MAC5D,EAAsB,EAAK,aAAa,aAAe,EACzD,EAAK,mBAAmB,OACtB,EAAkB,GAAI,GAAsB,EAC5C,EAAmB,GAAI,GAAuB,EAC9C,EAAe,EAAiB,EAChC,EAAgB,EAAsB,EAC5C,EAAK,QAAU,GAAA,SACX,CAAC,EAAiB,EAAgB,EAAe,GACjD,CAAC,EAAG,QAER,GAAK,QAAU,GAAA,SAAS,CAAC,EAAG,EAAG,EAAG,GAAI,CAAC,EAAG,aAKhD,SAAA,UAAA,QAAA,WACE,MAAO,UAII,EAAA,OAAb,SACI,EAAuC,GAAA,MAAA,KAAA,QAAA,GAAA,6FACzC,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGN,GAAI,CAAC,GAIH,GADA,EAAqB,SAAS,cAAc,SACxC,CAAC,EAAa,aAAe,CAAC,EAAa,aAC7C,KAAM,IAAI,OACN,0GAGN,EAAmB,MAAQ,EAAa,YACxC,EAAmB,OAAS,EAAa,aAErC,SAAiB,GAAI,GAAe,EAAoB,GAG9D,CAAA,EAAM,EAAe,gBAArB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,oHACE,AAAI,KAAK,aAAa,YACpB,GAAA,KAAK,OACA,KAAK,aAAa,aAAe,QAC7B,KAAK,aAAa,aAAe,cACtC,WACI,MAAA,+BAA+B,EAAK,aAAa,WAAU,2FAKnE,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,CACL,SAAU,KAAK,aAAa,SAC5B,WAAY,KAAK,aAAa,WAC1B,KAAK,aAAa,WAClB,OACJ,MAAO,KAAK,mBAAmB,MAC/B,OAAQ,KAAK,mBAAmB,kBAPpC,SAAK,OAAS,EAAA,qCAYd,EAAE,QAAU,iDAAiD,EAAE,QACzD,SAGR,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,uCAIlB,IACE,KAAK,mBAAmB,UAAY,KAAK,aAClC,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,OAAO,IAAI,gBAAgB,KAAK,QAGhE,YAAK,mBAAmB,OAExB,KAAK,SAAW,GAEhB,CAAA,EAAO,GAAI,SAAc,SAAA,GAEvB,EAAK,mBAAmB,iBAAmB,WACzC,cAKA,EAAA,UAAA,KAAN,oFACE,GAAI,KAAK,SACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAI7B,IACE,EAAM,GAAA,QAAQ,WAAW,KAAK,0BACvB,GACP,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAU,IAEjE,GAAI,KAAK,OACP,IACE,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,mBAAmB,GAAM,KAAM,WAC5C,GACP,KAAM,IAAI,OAAM,oCAAoC,EAAE,iBAEtD,EAAI,cAGN,OAAA,CAAA,EAAO,CAAC,MAAO,EAAK,KAAM,oBAItB,EAAA,UAAA,aAAR,WAIE,MAAI,QAAK,aAAa,aAAe,KAAK,aAAa,cAClD,MAAK,mBAAmB,QAAU,KAAK,aAAa,aACpD,KAAK,mBAAmB,SAAW,KAAK,aAAa,gBAO5D,EAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,IAAA,KAAK,WACV,GAAM,GAA0B,EAAI,UAAU,WAAW,GACrD,EACJ,EAAe,GAAA,MAAM,cACjB,EAAe,EAAK,QAAS,EAAK,WAAY,EAAK,SACnD,YAEJ,GAAM,GAAQ,EAAa,MAC3B,MAAO,GAAa,QAAQ,EAAM,MAAM,OAMtC,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAI7B,EAAA,UAAA,KAAA,WACE,GAAM,GAAS,KAAK,OAAO,YAE3B,EAAO,QAAQ,SAAA,GAAS,MAAA,GAAM,SAE9B,IACE,KAAK,mBAAmB,UAAY,WAC7B,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,KAEhC,KAAK,SAAW,IAIlB,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAEpB,GApMoC,IC1BpC,AA2BA,GAAA,IAAA,WAAA,cAUA,MAAA,MCrCA,AAoBA,GAAA,IAAA,SAAA,GAA6C,GAAA,EAAA,GAA7C,4DAmBE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAI,IAAc,KAAM,IAEnC,GAtB6C,IAmC7C,GAAA,SAAA,GAA4B,GAAA,EAAA,GAG1B,WAAsB,EAAgC,GAAtD,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAkB,EAAU,KAG9C,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf4B,IAiB5B,GAAA,SAAA,GAAgC,GAAA,EAAA,GAI9B,WACc,EAA0C,GADxD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAA0C,EAAA,UAAA,EAHxD,EAAA,UAAY,KAOZ,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAAc,KAAK,UAAS,MAGzD,EAAA,UAAA,KAAN,mHACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eACxC,GADM,EAAc,EAAA,OAChB,EAAY,KACd,MAAI,MAAK,YAAc,GACrB,CAAA,EAAO,IAKT,MAAK,YAAY,KAAK,KAAK,WAC3B,KAAK,UAAY,GACjB,CAAA,EAAO,KAQT,IANM,EAAQ,EAAY,MAAM,MAAM,KAAK,WAK3C,EAAM,GAAK,KAAK,UAAY,EAAM,GAClC,EAAA,EAAmB,EAAA,EAAM,MAAM,EAAG,IAAf,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,KAAK,YAAY,KAAK,GAExB,YAAK,UAAY,EAAM,EAAM,OAAS,GAEtC,CAAA,EAAO,UAEX,GAvCgC,ICxEhC,AAsBA,GAAA,IAAA,SAAA,GAAgD,GAAA,EAAA,GAAhD,4DAUE,SAAA,UAAA,WAAA,WACE,MAAO,IAAI,IAAa,OAE5B,GAbgD,IA0BhD,GAAA,SAAA,GAA2B,GAAA,EAAA,GAGzB,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAiB,KAGnC,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf2B,IAuC3B,GAAA,SAAA,GAA+B,GAAA,EAAA,GAM7B,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAF6B,EAAA,SAAA,EAEzB,GAAA,MAAM,IAAI,cACZ,EAAK,QAAU,GAAI,aAAY,cAGxB,GAAA,GAAA,0BAAA,cACP,EAAK,QAAU,GAAI,GAAc,iBAGrC,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,+GACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eAExC,MAFM,GAAc,EAAA,OAEhB,EAAY,KACd,CAAA,EAAO,IAEP,GAAQ,EAAY,MAItB,AAAI,GAAA,MAAM,IAAI,cACZ,EAAO,KAAK,QAAQ,OAAO,EAAO,CAAC,OAAQ,KAE3C,EAAO,KAAK,QAAQ,MAAM,OAAO,KAAK,EAAM,SAE9C,KAAK,YAAY,KAAK,GACtB,CAAA,EAAO,WAEX,GAtC+B,ICvF/B,AAqCA,GAAA,IAAA,SAAA,GAAuC,GAAA,EAAA,GAIrC,WACc,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,KAAA,EACA,EAAA,QAAA,EAEZ,GAAA,KAAK,OACA,YAAgB,aACZ,IAAA,MAAM,IAAI,cACL,YAAgB,OAAQ,YAAgB,MACzC,IACT,WAAM,MAAA,yEAEV,EAAK,OAAS,EAAQ,QAAU,EAEhC,EAAK,UAAY,EAAQ,WAAa,KAAO,OAG/C,SAAA,UAAA,QAAA,WACE,MAAO,cAAc,KAAK,MAGtB,EAAA,UAAA,KAAN,oHACE,MAAI,MAAK,QAAY,MAAK,eAAgB,YAClB,KAAK,KAAK,WACV,KAAK,KAAK,MAChC,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAQ,GAAI,SAAoB,SAAC,EAAS,GAC9C,GAAM,GAAM,EAAK,OAAS,EAAK,UAC/B,GAAI,EAAK,eAAgB,YAGvB,EAAQ,GAAI,YAAW,EAAK,KAAK,MAAM,EAAK,OAAQ,UAMpD,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GACnB,GAAI,GAAsC,EAAW,OAOrD,GAHA,AAAI,YAAgB,cAClB,GAAO,GAAI,YAAW,IAEpB,CAAE,aAAgB,aACpB,MAAO,GAAO,GAAI,WAAU,sCAE9B,EAAQ,IAEV,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,aAE1B,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,EAAM,QAIhC,GAAM,GAAQ,EAAK,KAAK,MAAM,EAAK,OAAQ,GAG3C,EAAW,kBAAkB,GAE/B,EAAK,OAAS,SAEA,CAAA,EAAM,WAAtB,MAAA,CAAA,EAAQ,GAAA,MAAQ,EAAA,OAAc,EAAA,KAAM,GAAK,UAE7C,GAxEuC,ICrCvC,YA6BI,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,qGAGpB,MAAK,OAAO,IAAS,SACnB,EAAY,EAEZ,GAAa,EAAgB,IAC7B,EAAc,GAA0B,IAEzB,CAAA,EAAM,GAAA,KAAK,MAAM,EAAW,iBAAvC,GAAW,EAAA,OACb,EAAS,MACY,WAAU,KAAC,CAAA,EAAM,EAAS,gBAD/C,CAAA,EAAA,UACI,SAAa,GAAA,GAAA,MAAI,WAAU,CAAA,OAAC,EAAA,UAClC,CAAA,EAAO,GAAI,IAAkB,EAAY,WAEzC,KAAM,IAAI,OAAM,EAAS,iBAK7B,GAAM,IAA4B,SAAC,GACjC,GAAM,GAAO,CACX,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,KACd,KAAM,EAAQ,KACd,YAAa,EAAQ,YACrB,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,SAAU,EAAQ,SAClB,UAAW,EAAQ,WAErB,MAAO,IC5DT,YAqB4B,GAC1B,MAAQ,OAAO,IAAW,UAAa,EAAO,OAAO,EAAG,KAAO,UCtBjE,mBA6BoC,GAAA,EAAA,GASlC,WACc,EACS,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,MAAA,EACS,EAAA,QAAA,IAIjB,SAAA,UAAA,SAAN,oFACE,MAAI,IAAY,KAAK,QAAU,GAAA,MAAM,IAAI,YAEjC,GAAK,cACX,KAAK,MAAQ,EAAG,aAAc,KAAK,MAAiB,OAAO,KAI7D,CAAA,EAAO,GAAI,IAAkB,KAAK,MAAsB,KAAK,eAEjE,GAzBoC,IC7BpC,mBA4BmC,GAAA,EAAA,GAQjC,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,IAAA,EACA,EAAA,YAAA,IAQjB,SAAA,UAAA,SAAN,8EACE,MAAI,IAAY,KAAK,KACnB,CAAA,EAAQ,GAAI,IAAe,KAAK,IAAe,KAAK,aAC/C,YAEL,CAAA,EAAO,GAAiB,KAAK,IAAK,KAAK,mBAG7C,GA1BmC,IC5BnC,YA0GI,EAAqB,GAAA,MAAA,KAAA,QAAA,GAAA,IAChB,GAAI,IAAW,GAAI,IAAc,GAAS,eA4B/C,GADJ,GAAA,GAAA,KAEQ,EAAO,GAAqB,GAClC,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,qBAgEvC,GADJ,GAAA,GAAA,KAEE,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACf,MAAA,CAAA,EAAM,YAAZ,SAAM,EAAA,OACZ,CAAA,EAAO,GAAqB,WAAM,MAAA,GAAI,6BAmCtC,EACA,sEACF,MAAA,CAAA,EAAO,GAAe,OAAO,EAAoB,oBAqClB,sEAE/B,MAAA,CAAA,EAAO,GAAmB,OAAO,QCxRnC,GAGM,IAAU,oNCHhB,oBA2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9GnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC7EnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UClFnC,oBAKA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAER,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAG,EAC9B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9FnC,oBAyBA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UChJnC,oBAIA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UCnGnC,oBAwBA,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMjB,MAAO,KAAW,UAAY,GAAO,SACxC,GAAO,QAAU,EAEjB,IACE,EAAa,wBACN,SACJ,AAAK,OAAO,SAAW,YAAc,OAAO,KACjD,OAAO,WAAa,MAAO,OAK3B,GACA,QCrPF,oBAYA,GAAI,IAAO,KAKP,GAAS,KAKT,GAAS,KAQT,GAAY,KASZ,GAAU,KAOV,GAAS,KAIT,GAAK,KAET,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAO,QAAU,6vDC3DjB,YAoBI,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MACF,GAAM,8DC7BpB,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBACvC,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAM/B,YACI,EAAyB,EAAW,EACpC,GACF,GAAI,IAAe,SACjB,MAAO,GAAQ,OAAO,GACjB,GAAI,IAAe,OACxB,MAAO,GAAQ,KAAK,GACf,GAAI,IAAe,MACxB,MAAO,GAAA,IAAO,GACT,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,GAChB,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,EAAG,GAE1B,KAAM,IAAI,OACN,cAAc,EAAU,qEAeM,GAAA,EAAA,GAMlC,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KANF,SAAA,UAAY,GAGX,EAAA,SAAW,GAIjB,EAAK,KAAO,GAAI,GAAA,YAAY,EAAM,EAAA,YAGpC,SAAA,UAAA,MAAA,SAAM,EAAoC,EAAiB,GAEzD,AAAI,KAAK,UACP,MAAK,SAAW,GAChB,AAAI,EAAA,MAAM,IAAI,YACZ,EAAA,aAAa,KACT;;;gCAYR,GAAM,GAAS,GAEf,YAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,IAEzC,GAST,EAAA,UAAA,eAAA,SACI,EAAiB,EACjB,GACF,GAAM,GAAQ,KAAK,MAAM,EAAQ,EAAO,GAExC,MAAO,CAAC,OAAQ,EAAO,MAAK,EAAE,MAAK,IAIrC,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,YAIb,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,aAIf,EAAA,UAAA,KAAA,SACI,EAAgB,EAAoC,EACpD,GACF,KAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,KAGlD,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,KAAK,cAGb,EAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,SAAS,SAEvB,EAAA,UAAA,SAAA,SAAS,GACD,GAAA,GAAA,KAAA,KAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,mBAEd,GAAI,IAAU,aACZ,GAAM,GACF,KAAK,SAAS,EAAmB,KAAK,QACpC,EACF,KAAK,SAAS,EAAmB,KAAK,QAC1C,MAAO,GAAA,aAAa,uBAAuB,EAAY,GAGzD,MAAO,MAAK,KAAK,IAAI,GAAQ,QAGvB,EAAA,UAAA,WAAR,SAAmC,GACjC,GAAM,GAAO,KAAK,SAAS,EAAE,QACzB,EAAc,EAClB,GAAI,EAAE,QAAU,SACd,IAEE,EAAe,EAAsB,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,WAChE,GACA,KAAM,IAAI,OAAM,oDAGpB,MAAO,GAAA,OAAU,EAAE,MAAO,EAAE,MAAO,IAGrC,EAAA,UAAA,WAAA,SACI,EAAoC,EAAiB,GACvD,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAG7D,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,KAAK,IAAI,IACT,GAAA,GAAA,KAAA,KAAA,IAAA,GAAA,mBAEP,AAAI,GAAsB,MACxB,MAAK,YAAY,EAAmB,KAAK,QACzC,KAAK,YAAY,EAAmB,KAAK,SAG3C,KAAK,KAAK,OAAO,KAIrB,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GAEjC,EAAW,WAEX,AAAI,EAAW,SAAW,GACxB,KAAK,YAAY,KAKjB,EAAA,UAAA,KAAN,SAAW,8EACH,SAAQ,EAAA,KAAK,MACnB,IACM,EAAW,EAAA,KAAK,MAAQ,EAC9B,CAAA,EAAO,CAAC,SAAQ,SAGlB,EAAA,UAAA,OAAA,WACE,MAAO,CAEL,WAAY,GACZ,QACI,CAAC,wHAKT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,GAAiB,EAAG,gBAEpB,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAU,GAAI,GAKvB,OAFM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GACpB,EAAI,EAAG,EAAI,EAAO,KAAM,KAI/B,OAHM,GAAM,EAAO,WAAW,GAExB,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,GAAK,EAAQ,GAAK,EAAM,GAE1C,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAM,OAAM,IAGrC,MAAO,GAAO,YAGhB,EAAA,UAAA,KAAA,SAAK,GAIH,OAHM,GAAQ,KAAK,SAAS,EAAE,QACxB,EAAS,EAAA,OAAU,CAAC,EAAE,KAAM,EAAE,MAAO,EAAE,OACvC,EAAO,EAAO,OACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAK,EAAI,EAAE,KAAO,GAAK,EAAM,GAE/B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,EAAA,MAAS,EAAG,EAAO,GAAM,QAAQ,GAE5C,MAAO,IAGT,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAiB,EAAG,WAKpB,OAHM,GAAS,EAAA,OAAU,EAAE,MAAO,EAAE,OAC9B,EAAO,KAAK,WAAW,cAEpB,GACP,GAAM,GAAS,EAAO,WAAW,GAC3B,EAAQ,EAAO,QACrB,EAAK,QAAQ,SAAA,GAAM,MAAA,GAAM,GAAM,EAAE,MAAM,GAAM,EAAI,EAAM,KACvD,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAK,OAAM,KAJ3B,EAAI,EAAG,EAAI,EAAO,KAAM,MAAxB,GAOT,MAAO,GAAO,YAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,UAAiB,EAAG,OAGb,EAAA,IAAO,EAAA,OAAU,IAAK,IAG/B,EAAA,UAAA,KAAA,SAAuB,GAAvB,GAAA,GAAA,KACE,GAAiB,EAAS,QAK1B,OAHM,GAAO,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAK,SAAS,EAAE,UACxC,EAAS,EAAA,OAAU,EAAQ,GAAG,MAAO,EAAQ,GAAG,OAChD,EAAa,EAAO,OACjB,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAElC,OADM,GAAW,EAAK,GACb,EAAI,EAAG,EAAI,EAAW,OAAQ,IACrC,EAAW,IAAM,EAAS,GAG9B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAGhD,EAAI,EAAA,IAAO,EAAQ,EAAS,QAAQ,IACpC,EAAI,EAAA,IAAO,GACX,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAO,EAAG,IAGnB,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,IAAI,EAAQ,MAIjE,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAiB,CAAC,EAAG,GAAI,UAqBzB,OAnBM,GAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAU,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5C,EAAW,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7C,EAAW,EAAE,MAAM,GAEnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,8DAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAAY,EAAA,EAAA,GAGrB,EAAA,8DAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAIzB,EAAO,EAAU,EACjB,EAAS,EAAA,OAAU,CAAC,EAAU,EAAS,GAAW,EAAE,OACpD,EAAU,EAAO,OACjB,EAAY,KAAK,UAEd,EAAI,EAAG,EAAI,EAAU,IAC5B,OAAS,GAAK,EAAG,EAAK,EAAS,GAAM,EACnC,OAAS,GAAK,EAAG,EAAK,EAAU,GAAM,EACpC,OAAS,GAAK,EAAG,EAAK,EAAW,GAAM,EAMrC,OAJM,GAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAE/B,EAAI,EAAI,EAAI,EAAQ,IAC3B,OAAS,GAAI,EAAI,EAAI,EAAQ,KAG3B,OAFI,GAAM,EAED,EAAI,EAAI,EAAI,EAAQ,IAC3B,GAAO,EAAQ,EAAI,EAAS,EAAI,EAAa,EAAI,GAC7C,EAAQ,EAAI,EAAa,EAAI,EAAa,EAAI,GAEpD,EAAQ,EAAI,EAAQ,GAAI,EAAW,KAAO,EAOtD,MAAO,GAAO,YAGhB,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE/C,EAAS,KAAK,YAAY,EAAG,EAAG,EAAY,GAChD,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAIvC,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAiB,CAAC,EAAG,GAAI,YAEzB,GAAM,GAAK,SAAC,EAAW,GAAc,MAAA,MAAK,MAAM,EAAI,IAC9C,EAAc,QACpB,MAAO,MAAK,oBAAoB,EAAG,EAAG,EAAa,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MASvD,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EACD,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAO,EAAM,EAAS,GAExB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,GAAiB,EAAG,OAUpB,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAO,EACF,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAQ,EAAM,EAAS,GAEzB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,EAAG,sBAOpB,OALM,GAAM,GAIN,EAAW,EAAE,KAAO,EAAW,KAC5B,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,EAAW,WAAW,EAAI,GAGzC,OAAS,GAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAY,EAAA,OAAU,EAAG,SACzB,EAAO,EAAA,MAAS,EAAW,GAAY,OAAO,WAC9C,EAAM,EAAK,IAAI,GAAG,IAAI,GAC5B,EAAI,KAAK,GAGX,MAAO,GAAA,MAAS,IAGlB,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAIlD,GAFA,GAAiB,EAAG,UAEhB,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,oDAAoD,GAAE,KAAO,GAAC,IAC9D,iBAAgB,IAWtB,OATM,GAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAE,MAAO,GAC3B,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAW,EAAE,MAAM,EAAE,KAAO,GAC5B,EAAgB,EAClB,SAAC,EAAW,GAAc,MAAA,GAAI,EAAW,EAAI,GAC7C,SAAC,EAAW,GAAc,MAAA,GAAI,GACzB,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACrC,OAAS,GAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAM,EAAc,EAAG,GAC7B,GAAI,IAAM,EACR,EAAK,GAAO,EAAY,EAAI,EAAM,QAElC,GAAM,GAAU,EAAc,EAAG,EAAI,GACrC,EAAK,GAAO,EAAY,EAAM,GAAW,EAAK,GACtB,EAAM,GAAO,EAAK,IAIhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,UAAiB,CAAC,EAAG,GAAI,YAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,UAAiB,CAAC,EAAG,GAAI,QAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,UAAiB,CAAC,EAAG,GAAI,gBAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,UAAiB,CAAC,EAAG,GAAI,cAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAiB,CAAC,EAAW,EAAG,GAAI,UAYpC,OAVM,GAAS,KAAK,SAAS,EAAU,QACjC,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,MAAS,EAAE,MAAO,EAAA,WAAW,EAAE,MAAO,EAAE,QACjD,EAAY,KAAK,SAAS,EAAO,QACnC,EAAQ,EACN,EAAS,EAAU,OAAS,GAAK,EAAU,KAAO,GAAK,EAAE,OAAS,EACpE,EACA,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAE5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,OAAS,GAAI,EAAG,EAAI,EAAQ,IAC1B,AAAI,EAAO,KAAO,EAChB,EAAU,KAAW,EAAQ,GAE7B,EAAU,KAAW,EAAQ,GAKnC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GACJ,GAAiB,CAAC,GAAY,SAE9B,GAAM,GAAW,KAAK,SAAS,EAAU,QACzC,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAiB,EAAG,QAEpB,GAAM,GAAQ,KAAK,SAAS,EAAE,QAC9B,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAM,EAAO,EACnB,MAAK,GAAO,GAAK,EAAO,GAAO,GAAQ,GAAK,GAAQ,EAC3C,EAEC,GAAM,GAAQ,KAK5B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAM,GAAO,EAEf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAS,EAAM,GACV,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAS,GAAU,EAErB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,UAAiB,CAAC,EAAG,GAAI,qBAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAO,EAAO,EACpB,MAAO,GAAO,KAIlB,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,IAGT,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,EAAG,EAAO,IAElC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,KAAK,IAAI,EAAG,EAAO,IAAK,GAEhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MACR,SAAC,EAAQ,GAAW,MAAA,GAAS,EAAI,EAAS,EAAS,KAGhE,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAiB,CAAC,EAAI,GAAI,UAK1B,OAHM,GAAe,GAAI,cAAa,EAAE,MAClC,EAAS,KAAK,SAAS,EAAE,QACzB,EAAW,KAAK,SAAS,EAAG,QACzB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,GACjB,AAAI,GAAK,EACP,EAAa,GAAK,EAAS,GAE3B,EAAa,GAAK,EAAS,GAAM,GAAI,GAGzC,MAAO,MAAK,WAAW,EAAc,EAAE,MAAO,YAGhD,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,MAAM,EAAQ,MAInE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,OAAO,EAAO,EAAQ,GAExC,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAiB,CAAC,EAAG,GAAS,UAyB9B,OAvBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAiB,EAAS,aAAe,eAEzC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OAEnC,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAChD,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAEhD,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EACf,EAAW,EAAI,EACZ,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EACxB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,IAAK,EAAW,EAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAC7B,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,GAAK,GAC1B,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,GAAK,IAClB,GAAO,EAAM,GAAW,IAE9B,IAAY,EAAS,cAOjC,MAAO,GAAE,YAGX,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAiBpC,OAfM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAI,EAAA,OAAmB,EAAS,SAAU,EAAE,OAE5C,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAKF,OAHM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAElC,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAS,YACpC,EAAW,EAAK,EAAS,YAAc,EACpC,GAAK,EAAG,GAAK,EAAa,MACjC,GAAM,IAAK,EAAW,GAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,GAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAAS,WACtC,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,IACrB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,KAAO,GAAO,EAAM,GAAW,IAElD,IAAY,EAAS,eASrC,MAAO,GAAE,YAGX,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,kBAkC/B,OAhCM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAiB,IAAe,eAChC,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GACjD,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GAE9C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,IAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAOjC,OANM,IAAK,GAAK,EAAc,GACxB,GACF,EAAe,EAAI,EAAa,GAAK,EAAa,GAChD,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,EAAiB,IAC7C,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,GAAM,IAAW,EAAe,EAAI,EAAa,GAC7C,EAAa,GAAK,EAAiB,EACvC,EAAS,IAAY,GAK7B,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GA6BF,OA5BM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAE1B,EAAA,EAAA,UACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,GAAA,EAAA,YAEI,GAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,GAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,GAAU,EAAc,EAAI,EAAS,QAAQ,KAE1C,GAAI,EAAG,GAAI,EAAW,EAAE,GAC/B,OAAS,IAAK,EAAG,GAAK,EAAY,EAAE,GAElC,OAAS,IAAK,EAAG,GAAK,EAAS,EAAE,GAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAGzC,GAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,KACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,IAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAc,GAErB,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAQjC,OAPM,IAAK,GAAK,GAAc,GACxB,GACF,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GACxC,GAAY,EAAS,GAAc,EAAI,IACzC,EAAS,GAAe,EAAI,IAC5B,EAAS,GAAc,EAAI,IAAM,EAAQ,GAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAK3B,EAAS,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GAAK,IACpD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAiB,CAAC,EAAG,GAAK,mBAa1B,OAXM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eACzC,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,EAC3C,OAAS,GAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAG5C,OADI,GAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,AAAI,EACF,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAEnD,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAK3D,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAKpC,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAqBzC,OAnBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAC9C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAEhB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAEvB,EAAK,EAAG,EAAK,EAAa,EAAE,EAMnC,OALM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAW,GAAM,IAChD,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAW,GAAM,GACtD,EAAW,EAAK,EAEb,EAAK,EAAG,EAAK,EAAc,EAAE,EAOpC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,GAAQ,KAAK,IACf,EAAS,UACR,GAAS,SAAW,EAAS,GAAM,GAClC,GAAW,EAAK,EAAO,EAEpB,GAAK,EAAG,GAAK,EAAa,EAAE,GAOnC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,IAAM,IAC/C,GAAQ,KAAK,IACf,EAAS,SACR,GAAS,QAAU,EAAU,IAAM,GAClC,GAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,GAG3C,OAFM,IAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,IAE5C,OADI,IAAU,EACL,GAAI,EAAG,GAAI,EAAS,UAAW,EAAE,GAIxC,OAHM,IAAW,GAAI,EACf,GAAW,GAAI,EAEZ,GAAK,EAAO,GAAK,EAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAe,EAC9B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,IACjC,GAAM,IAAK,GAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAE7B,IACI,EAAQ,GAAW,IAAM,EAAS,GAAW,IAKzD,EAAS,GAAW,IAAM,GAMpC,MAAO,GAAG,YAGZ,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,gBAAgB,EAAO,EAAQ,GAEjD,MAAI,IAGF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAiB,CAAC,EAAG,GAAS,mBAc9B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WACxC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OACnC,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,EAAE,GACpC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,EAAE,GACnC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAMF,OAJM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAS,WACtC,EAAW,EACX,EAAW,EACN,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,GAE3C,OADM,GAAO,EAAM,EAAW,GACrB,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAM,EAAW,IAAM,EAAO,EAAM,EAAW,GAEjD,GAAY,EACZ,GAAY,IAQxB,MAAO,GAAE,YAGX,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,2BA0B/B,OAxBM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAQ,EAAc,EAEnB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,GAAK,EAAG,EAAK,EAAU,EAAE,EAMhC,OALM,IAAW,EAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAMjC,OALM,IAAK,GAAK,EAAc,GACxB,GAAW,EAAO,EAAI,EAAO,GAAK,EAAO,GACzC,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAO,EAAE,IAC7B,GAAM,IAAK,EAAK,EAAQ,GAClB,GAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,EAAS,EAAO,EAAI,EAAO,EAAK,EAAO,GAAK,GAAM,GAK1D,MAAO,GAAG,YAGZ,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAiB,CAAC,EAAG,GAAK,4BAc1B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WAExC,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAK5C,OAJM,GAAK,KAAK,MAAM,EAAK,GACrB,EAAK,EAAK,EAEZ,EAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,GAAW,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAIhE,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAIlC,MAAO,GAAG,YAGZ,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,UAAiB,EAAG,QACb,GAAK,KAAK,WAAW,GAAI,IAGlC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,GAAiB,CAAC,EAAG,GAAU,UAE/B,GAAM,GAAqB,EAAE,MAAM,QAC7B,EAAgB,KAAK,SAAS,EAAQ,QAC5C,EAAS,GAAQ,EAAc,OAI/B,OAHM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GAEpB,EAAI,EAAG,EAAI,EAAO,KAAM,EAAE,GACjC,GAAM,GAAS,EAAO,WAAW,GAE3B,EAAwB,EAAO,QACrC,EAAY,GAAQ,EAAc,EAAO,IAEzC,GAAM,GAAgB,EAAK,WAAW,GACtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,YAGhB,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAa,EAAE,QAAQ,GAAW,GAC7B,QAAQ,GACR,MAAM,EAAkB,IAG9B,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAiB,EAAG,UA8BpB,OA5BM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,OAAU,EAAS,SAAU,EAAE,OACxC,EAAa,EAAO,OAEpB,EAAqB,EAAS,SAAS,GAAK,EAAS,SAAS,GAChE,EAAS,SAAS,GAAK,EAAS,SAAS,GACvC,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAGhD,OAFM,GAAoB,EAAQ,EAC5B,EAAmB,EAAQ,EAAE,QAAQ,GAClC,EAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAMf,OAJM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAChD,EACF,EAAoB,EAAS,EACxB,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAKb,OAHM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,GAAkB,EAAoB,EAAO,EAC1C,GAAO,EAAG,GAAO,EAAS,SAAU,EAAE,IAG7C,OAFM,IAAa,GAAO,EAAc,EACpC,GAAU,GACP,GAAU,GACf,IAAW,EASb,OAPM,IACF,KAAK,IAAI,EAAS,QAAS,EAAuB,IAEhD,GAAkB,GAAkB,GAAO,EAC7C,GAAc,EACd,GAAW,EACX,GAAQ,EACH,GAAS,EAAW,GAAS,EACjC,IAAU,GAEb,OADM,IAAe,EAAmB,GAAS,EAAE,QAAQ,GAClD,GAAO,EAAS,GAAO,EAC3B,IAAQ,GAEX,OADM,IAAa,GAAe,GAAO,EAAE,QAAQ,GAC1C,GAAO,GAAS,GAAO,GAC3B,IAAQ,GACX,GAAM,IAAa,GAAa,GAAO,EAAE,QAAQ,GAC3C,GAAQ,EAAQ,GAAa,GAOnC,GANA,AAAK,IAAa,OAAS,GAAQ,GACjC,GAAc,GACT,AAAI,IAAa,OACtB,KAAY,GACZ,MAEE,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAM,IAAe,GAAkB,EACvC,EAAW,IACP,IAAa,MAAQ,GAAW,GAAQ,KAMtD,MAAO,GAAO,YAGhB,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGzC,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,qBAuB1B,OArBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAgB,EAAK,GAAc,EAAe,GAElD,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,KAIjB,EAAG,IACC,EAAU,EAAe,EAAO,EAAS,EAAO,EAChD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGjC,EAAA,UAAA,mBAAR,SAA2B,EAAa,GAiBtC,OAfM,GAAe,EAAA,OAAU,EAAS,SAAU,SAC5C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,KAAK,WAAW,GACpB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAIf,OAFM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAC7C,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAIb,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAO,EAAG,EAAO,EAAS,SAAU,EAAE,GAG7C,OAFM,GAAa,EAAO,EAAc,EACpC,EAAU,EACP,EAAU,GACf,GAAW,EASb,OAPM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAGlD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAS,EAAW,EAAS,EACjC,GAAU,EAEb,OADM,GAAS,EAAS,EACf,EAAO,EAAS,EAAO,EAC3B,GAAQ,EAEX,OADM,GAAO,EAAO,EACX,GAAO,EAAS,GAAO,EAC3B,IAAQ,GACX,GAAM,IAAO,GAAO,EACd,GAAQ,EAAK,IAAI,EAAO,EAAQ,EAAM,GAAM,GAClD,AAAI,IAAS,GACX,GAAW,GACX,EAAc,EAAS,EACf,EACJ,EAAO,EAAwB,IAM3C,EAAa,IAAI,EAAa,EAAO,EAAQ,EAAM,EAAM,KAMnE,MAAO,GAAa,YAGtB,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAiB,CAAC,EAAG,GAAI,qBAoBzB,OAlBM,GAAe,KAAK,mBAAmB,EAAG,GAC1C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAY,KAAK,WAAW,GAC5B,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,GAAS,EACP,EAAwB,EAC5B,EACA,EAAU,IAAI,EAAO,EAAS,EAAO,EAAO,GAC1C,GACF,EAAS,EAAwB,EACjC,EAAO,EAAuB,EAE5B,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,GAAQ,KAIzB,EAAG,IAAI,EAAS,EAAO,EAAS,EAAO,EAAO,GAMxD,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,kBAqBpB,OAnBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cACf,EAAA,KAAK,cAAc,CAAC,EAAO,EAAW,EAAU,KAE9C,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAE9C,EAAY,EACV,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GACvC,EAAI,EAAG,EAAI,EAAO,IACzB,OAAS,GAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EAAgB,KAAK,IAAI,EAAY,EAAG,KAAK,KAAK,IAClD,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAiB,EAAE,QAAQ,GAC7D,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAgB,EAAE,QAAQ,GACzD,EAAI,EAAG,EAAI,EAAU,IAU5B,OATM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EACF,KAAK,IAAI,EAAW,EAAG,KAAK,KAAK,IAC/B,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GACvD,EAAI,EAAG,EAAI,EAAa,KAI/B,GAAM,GAAU,EAAQ,EAAgB,GAClC,EAAa,EAAQ,EAAgB,GACrC,GAAW,EAAQ,EAAiB,GACpC,GAAc,EAAQ,EAAiB,GAEvC,GAAM,EAAW,IAAW,GAAW,EACvC,GAAS,EAAc,IAAc,GAAc,EACnD,GAAW,GAAO,IAAS,IAAO,EAExC,EAAO,KAAe,GAK9B,MAAO,GAAA,OAAU,EAAQ,CAAC,EAAO,EAAW,EAAU,KAGxD,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAChD,GAAiB,CAAC,EAAI,GAAI,0BA+B1B,OA7BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GAOrD,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAMhD,EAAW,KAAK,SAAS,EAAG,QAC9B,EAAS,EACJ,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAU,EAAI,EAAE,QAAQ,GACrB,EAAI,EAAG,EAAI,EAAS,IAU3B,OATM,GAAM,EAAI,EACV,EAAc,KAAK,MAAM,GACzB,EAAiB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAU,GAEpD,EAAe,EAAU,EAAc,EAAE,QAAQ,GACjD,EAAkB,EAAU,EAAiB,EAAE,QAAQ,GAEvD,EAAU,EAAM,EAChB,EAAiB,EAAM,EACpB,EAAI,EAAG,EAAI,EAAQ,IAmB1B,OAlBM,GAAM,EAAI,EACV,EAAe,KAAK,MAAM,GAC1B,EAAgB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAS,GAClD,EAAU,EAAM,EAChB,EAAiB,EAAM,EAEvB,EAAkB,EAAe,EAAe,EAAE,QAAQ,GAC1D,EAAmB,EAAe,EAAgB,EAAE,QAAQ,GAC5D,GACF,EAAkB,EAAe,EAAE,QAAQ,GACzC,GACF,EAAkB,EAAgB,EAAE,QAAQ,GAE1C,GACF,EAAiB,EACf,GAA6B,EAAiB,EAC9C,GAA6B,EAAU,EACvC,GAAsB,EAAU,EAC7B,GAAI,EAAG,GAAI,EAAO,MACzB,GAAM,IAAQ,EAAS,KACvB,EAAO,EAAkB,KACrB,GAAQ,GACZ,EAAO,EAAmB,KAAM,GAAQ,GACxC,EAAO,GAAqB,KACxB,GAAQ,GACZ,EAAO,GAAsB,KAAM,GAAQ,GAKnD,MAAO,GAAA,SAAY,EAAQ,CAAC,EAAO,EAAQ,EAAS,GAAQ,EAAE,QAGhE,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,yBAsBpB,OApBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cAAa,EAAQ,EAAY,EAAW,GAEzD,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GAE5C,EAAe,EACV,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAY,EACZ,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAc,EAAmB,EAAE,QAAQ,GACpD,EAAI,EAAG,EAAI,EAAU,IAO5B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAW,EACX,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAY,EAAmB,EAAE,QAAQ,GAClD,EAAI,EAAG,EAAI,EAAa,KAG/B,GAAM,GAAS,EAAQ,EAAY,GACnC,EAAO,KAAkB,EAKjC,MAAO,GAAA,OACH,EAAQ,CAAC,EAAO,EAAW,EAAU,GAAc,EAAE,QAG3D,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,iCAiC1B,OA/BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GACrD,EAAW,KAAK,SAAS,EAAG,QAK5B,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAGzC,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAS,IAM3B,OALM,GAAY,EAAc,EAAI,EAAE,QAAQ,GAGxC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAY,GAC7C,EAAI,EAAG,EAAI,EAAQ,IAO1B,OANM,GAAY,EAAY,EAAI,EAAE,QAAQ,GAGtC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAW,GAE5C,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHI,GAAQ,EAGH,EAAW,EAAG,EAAW,EAAW,KAC3C,GAAM,GAAM,EAAW,EAEvB,GAAI,EAAM,GAAK,GAAO,EACpB,SAGF,GAAM,GAAY,EAAc,EAAM,EAAG,QAAQ,GAC3C,GAAgB,EAAM,EACtB,GAAmB,KAAK,IAC1B,EAAU,EACV,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAC9B,GAAI,IAAM,GACR,SAEF,OAAS,IAAW,EAAG,GAAW,EAAU,MAC1C,GAAM,IAAM,GAAW,EAEvB,GAAI,GAAM,GAAK,IAAO,EACpB,SAGF,GAAM,IAAY,EAAY,GAAM,EAAG,QAAQ,GACzC,GAAgB,GAAM,EACtB,GAAmB,KAAK,IAC1B,EAAS,EACT,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAE9B,AAAI,IAAM,IACR,IAAS,EAAS,GAAY,KAIpC,EAAO,EAAY,GAAK,EAKhC,MAAO,GAAA,SAAY,EAAQ,EAAE,MAAO,EAAE,QAGxC,EAAA,UAAA,6BAAA,SACI,EAAa,EAAqB,EAAc,EAChD,GACF,GAAiB,EAAG,gCAEpB,GAAM,GAAW,EAAE,MAAM,GACnB,EAAO,EAAW,EAClB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAO,EAAE,KACT,EAAS,GAAI,cAAa,GAEhC,WAA2B,GAQzB,OAPM,GAAiB,EAAS,EAC5B,EACA,EAAS,EAAiB,KAAK,IAAI,EAAG,EAAiB,GACrD,EAAe,EAAS,EAC1B,KAAK,IAAI,EAAiB,EAAa,GAEvC,EAAM,EACH,GAAkB,EAAc,KACrC,GAAM,GAAI,EAAQ,GAClB,GAAO,EAAI,EAEb,MAAO,GAGT,OAAS,GAAS,EAAG,EAAS,EAAM,KAClC,GAAM,GAAM,EAAkB,GACxB,EAAM,EAAQ,GAAU,KAAK,IAAI,EAAO,EAAQ,EAAK,CAAC,GAC5D,EAAO,GAAU,EAGnB,MAAO,GAAA,SAAY,EAAQ,EAAE,QAG/B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAiB,EAAI,WAQrB,OAPM,GAAW,EAAG,MAAM,GACpB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAmB,KAAK,SAAS,EAAW,QAC5C,EAAoB,KAAK,SAAS,EAAY,QAC9C,EAAS,GAAI,cAAa,EAAG,MAC7B,EAAO,EAAG,KAEP,EAAS,EAAG,EAAS,EAAM,KAQlC,OAPM,GAAiB,EAAS,EAC1B,EACD,EAAS,EAAkB,KAAK,IAAI,EAAG,EAAiB,GACvD,EAAY,EAAS,EACvB,KAAK,IAAI,EAAU,EAAiB,EAAc,GAElD,EAAO,EACF,EAAI,EAAY,EAAI,EAAU,IACrC,GAAQ,KAAK,IAAI,EAAiB,GAAI,GAExC,EAAO,EAAQ,EAAO,EAEtB,OAAS,GAAI,EAAY,EAAI,EAAU,KACrC,GAAI,GAAM,GAAK,EAAQ,EAAO,EAAiB,GAC3C,EAAkB,GAAU,EAChC,AAAI,IAAW,GACb,IAAO,KAAK,IAAI,EAAM,CAAC,IAEzB,GAAO,EAAS,GAChB,EAAO,IAAM,GAGjB,MAAO,GAAA,SAAY,EAAQ,EAAG,QAGhC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAiB,EAAQ,eASzB,OAPM,GAAgB,EAAa,EAAS,EAAA,QAAW,GACjD,EAAY,EAAc,MAAM,GAChC,EAAY,EAAc,MAAM,GAChC,EAAM,EAAA,MAAkB,CAAC,EAAW,GAAa,SACjD,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAW,KAAK,SAAS,EAAc,QAEpC,EAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,GAAS,EAAI,EAGb,EAAM,GAAI,cAAa,EAAY,GACzC,EAAI,GAAK,EAAS,GAClB,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,EAAE,EACxC,EAAI,GAAS,EAAI,EAAQ,GAAK,EAAS,EAAS,GAKlD,OAFM,GAAS,GAAA,KAAgB,EAAK,YAC9B,EAAY,EAAI,EACb,EAAW,EAAG,EAAW,EAAY,EAAE,GAC9C,GAAM,GAAI,IAGV,EAAQ,EAAY,GAAY,EAAI,OAEpC,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,IACtC,GAAI,EAAI,EAAI,IACV,EAAQ,EAAY,GAAY,EAChC,QAKR,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAiB,EAAS,UAE1B,GAAM,GAAM,GAAI,cAAa,EAAQ,KAAO,GAC5C,EAAI,KAAK,GAGT,OAFM,GAAa,KAAK,SAAS,EAAQ,QAEhC,EAAQ,EAAG,EAAQ,EAAQ,KAAM,EAAE,EAC1C,AAAI,EAAW,IAAU,GAAK,EAAW,GAAS,GAChD,GAAI,EAAQ,EAAQ,EAAW,IAAU,GAG7C,MAAO,GAAA,SAAY,EAAK,CAAC,EAAQ,KAAM,GAAQ,UAGjD,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,GAAiB,EAAO,qBAExB,GAAM,GAAY,KAAK,SAAS,EAAM,QAChC,EAAa,KAAK,SAAS,EAAO,QACxC,MAAO,IACH,EAAW,EAAY,EAAe,EAAc,IAG1D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,IAAe,OACf,WAAM,MAAA,+DACF,IACR,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAgB9D,OAdM,GAAY,EAAE,MAAM,GACpB,EAAc,EAAE,MAAM,GACtB,EAAa,EAAE,MAAM,GACrB,EAAa,EAAE,MAAM,GAErB,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAU,KAAK,SAAS,EAAE,QAC1B,EACF,GAAI,cAAa,EAAY,EAAe,EAAc,GAE1D,EAAY,EACP,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAI,EAAG,EAAI,EAAc,EAAE,EAGlC,OAFM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACZ,EAAI,EAAG,EAAI,EAAa,EAAE,EAIjC,OAHM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACf,EAAW,GAAU,EAAY,GAAW,EACzC,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAM,EAAI,EACV,EACF,EAAM,EAAc,GAAM,EAAc,GAAM,EAAc,IAChE,EAAO,KAAe,EAAQ,GAKtC,MAAO,GAAA,SACH,EAAQ,CAAC,EAAW,EAAc,EAAa,KAG7C,EAAA,UAAA,oBAAR,SACI,EAAW,EAAW,EACtB,GACF,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAE,MAAO,EAAE,OAC9D,EAAS,EAAA,OAAU,EAAU,GAC7B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GACxD,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GAExD,EAAU,EAAO,OACvB,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,EACpC,EAAQ,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAK3D,QAFM,GAAO,KAAK,WAAW,GACvB,EAAO,KAAK,WAAW,cACpB,GACP,GAAM,GAAM,EAAO,WAAW,GAExB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAEzB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAE/B,EAAQ,GAAK,EAAG,EAAM,GAAS,EAAM,KAX9B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,IAA7B,GAcX,MAAO,GAAO,YAGhB,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,QAAA,aAEA,EAAA,UAAA,eAAA,WACE,MAAO,KAIT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAGtB,EAAA,UAAA,cAAA,SACI,EACA,EACA,EACA,EACA,EACA,GAmBF,OAjBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GACjC,EAAW,EAAM,MAAM,GAEtB,EAAA,EAAA,GAAY,EAAA,EAAA,GACb,EACF,EAAA,OAAU,CAAC,EAAU,EAAY,EAAW,GAAc,WAExD,EAAU,KAAK,SAAS,EAAM,QAC9B,EAAa,KAAK,SAAS,EAAS,QACpC,EAAY,KAAK,SAAS,EAAO,QAEjC,EAAW,EAAO,QAClB,EAAY,EAAO,QAKhB,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAW,EAAI,EACf,EAAK,EAAQ,GACb,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GAExB,EAAe,EAAW,GAChC,GAAI,GAAQ,EACV,SASF,OANM,GAAe,EAAa,EAC7B,GAAK,GAAO,GAAc,GAAM,GAAa,GAC9C,EACE,EACD,EAAY,EAAM,GAAK,GAAO,GAAa,GAAM,GAAY,GAAK,EAE9D,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAgB,EAAa,EAC/B,EAAM,GAAc,GAAK,EAAK,EAC9B,GAAO,GAAK,GAAO,GAAc,GAErC,GAAI,EAAO,GAAK,EAAO,EAAc,GACnC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAGzB,SAGF,GAAI,IAAW,WAKb,OAJM,GAAS,KAAK,MAAM,GACpB,EAAY,KAAK,KAAK,GACtB,GAAQ,EAAO,EAEZ,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAOF,OAJM,IAAU,KAAK,MAAM,IACrB,GAAW,KAAK,KAAK,IACrB,GAAQ,GAAO,GAEZ,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAI,GAAM,EAAI,GAAU,EAAS,GAAK,EAAS,EAAS,GACpD,EAAO,EAAS,GACd,GAAU,EAAU,GAE1B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAS,EAAS,GACjD,EAAO,EAAS,GACpB,GAAM,IAAW,EAAU,GAE3B,EAAM,EAAI,GAAU,EAAS,GAAK,EAAY,EAAS,GACnD,EAAO,EAAS,GACpB,GAAM,IAAa,EAAU,GAE7B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAY,EAAS,GACpD,EAAO,EAAS,GACpB,GAAM,IAAc,EAAU,GAExB,GAAM,GAAW,IAAW,IAAW,GACvC,GAAS,GAAc,IAAc,IAAc,GAEzD,EAAM,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC9D,EAAO,OAAO,GAAO,GAAQ,IAAS,IAAO,QAIjD,QAAS,GAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAKF,OAFM,IAAW,KAAK,MAAM,IACtB,GAAW,KAAK,MAAM,GACnB,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,IAAQ,EAAI,GAAW,EAAS,GAClC,GAAW,EAAS,GAAK,EAAO,EAAS,GACvC,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,IAAU,EAAU,OAM5C,MAAO,GAAO,YAGhB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAiB,GACvB,MAAO,MAAK,QACR,EAAe,EAAc,EAAa,EAAY,EACtD,EAAY,EAAW,EAAS,EAAc,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAE1C,GAAI,IAAc,EAChB,MAAO,GAAA,OAAU,GAAI,EAAa,EAAE,OAOtC,OAJM,GAAS,GAAI,GAAA,aAAa,CAAC,EAAW,GAAY,EAAE,OACpD,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAQ,KAAK,SAAS,EAAE,QAErB,EAAI,EAAG,EAAI,EAAW,KAG7B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,GAAgB,EAAM,EAAQ,GAC9B,EAAM,KAAK,GAEb,GAAI,EAAe,GAAK,GAAgB,EAAE,KAAO,EAC/C,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,EAAE,OAGzD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,EAAO,OAAO,EAAI,EAAY,GAAK,EAAM,EAAe,EAAY,GAGxE,MAAO,GAAO,WAAW,QAAQ,IAGnC,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAe,EAAA,OAAU,GACzB,EAAiB,GACvB,MAAO,MAAK,QACR,EAAS,EAAS,EAAO,EAAY,EAAW,EAAY,EAC5D,EAAS,EAAc,IAG7B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,EAAQ,GAAS,EAAA,KAAK,WAAW,GACjC,GAAM,GACF,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IACrD,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,OAGnD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAEhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,GAAM,GAAS,EAAA,KAAK,kBACD,EAAE,MAAO,EAAA,KAAK,cAAc,EAAE,QACjD,MAAO,MAAK,WAAW,EAAQ,EAAE,MAAO,EAAE,QAG5C,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGxC,EAAA,UAAA,QAAR,SACI,EAAiB,EAAiB,EAAoB,EACtD,EAAmB,EAAoB,EACvC,EAAmB,EACnB,GACF,GAAM,GAAe,CAAC,EAAa,EAAW,GAExC,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAc,KAAK,SAAS,EAAQ,QAE1C,GAAI,IAAe,EACjB,MAAO,GAAA,OAAU,GAAI,EAAO,EAAQ,OAGtC,GAAM,GAAS,GAAI,GAAA,aAAa,EAAc,EAAQ,OACtD,EAAO,OAAO,KAAM,KAAK,SAAS,EAAa,QAAuB,IAEtE,OAAS,GAAI,EAAG,EAAI,EAAY,KAG9B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,EAAM,KAAK,GACX,GAAgB,EAAM,EAAQ,GAGhC,GAAI,EAAe,GAAK,GAAgB,EAAa,EACnD,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,GAGvD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,AAAI,EACF,EAAO,OAAO,EAAe,EAAY,IACrC,EAAY,EAAI,EAAY,GAEhC,EAAO,OAAO,EAAe,EAAY,GAAK,EAAQ,OAAS,EAC3D,EAAY,GACZ,EAAY,EAAI,EAAY,GAItC,MAAO,GAAO,WAAW,QAAQ,IAErC,GAzjFoC,EAAA,eC5DpC,YAqB8B,GAE5B,OADM,GAAe,GAAI,cAAa,EAAK,QAClC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAa,GAAK,KAAK,IAAI,EAAK,IAElC,MAAO,GAGF,GAAM,IACT,SAAC,GACQ,GAAA,GAAA,EAAA,OAAA,EACD,EAAa,EAAK,QACpB,EAAe,GAAI,cAAa,EAAA,KAAK,cAAc,EAAE,QACzD,GAAI,EAAE,QAAU,aACd,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC7C,EAAe,GAAc,OAS7B,QAPM,GAAc,EAAW,KAAK,IAAI,EAAE,QACpC,EAAO,EAAY,mBAAmB,KACtC,EAAO,EAAY,mBAAmB,KACtC,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC/B,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAO,EAAS,GAChB,EAAO,EAAS,GACtB,EAAa,GAAK,KAAK,MAAM,EAAM,GAGvC,MAAO,GAAW,WAAW,EAAc,EAAE,MAAO,YAG7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICzDd,YAwB6C,GAE3C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAmB,GACzB,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAE3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAEhC,EACF,EAAA,KAAK,uBAAuB,EAA0B,GAEpD,EAAQ,EAAO,OACf,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAE7D,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAG1D,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAE5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,KAX7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,IAA5B,GAeX,MAAO,CAAC,EAAQ,IClEpB,YAqBwB,GAEf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,KAAM,EAAA,EAAA,KAEP,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACzC,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAEzC,EAAc,EAAQ,eAAe,EAAK,MAAO,aAEjD,EAAU,EAAQ,KAAK,IAAI,EAAY,QAK7C,SAAQ,mBAAqB,CAC3B,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,GACpD,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,IAG/C,EAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC/Cd,YAsBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IClCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAGP,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAIvC,GAAM,GAAc,EAAA,MAAS,EAAE,OACzB,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,EAAE,MAAO,CAAC,MAAO,aAEpD,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAQ,KAAM,GAAc,QAAO,IAE/D,SAAY,UACZ,EAAQ,8BAA8B,GAE/B,EAIT,GAAI,EAAE,QAAU,aACd,GAAM,GAAW,GAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,IAC5C,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAO,EAAE,MAAO,CAAC,MAAK,KAElE,SAAQ,8BAA8B,GAE/B,EAGT,GAAI,CAAC,EAAA,KAAK,gBAAgB,EAAE,MAAO,IAGjC,GAAM,GAAS,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAC7C,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAO,MAAO,MAAK,GAG3D,GAAI,IAAU,SACZ,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAe,WAAW,KAAK,GACrC,MAAO,GAAQ,eAAe,EAAE,MAAO,QAAS,GAGlD,GAAI,IAAU,QAIZ,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAO,EAAA,KAAK,aAAa,CAAC,GAAI,EAAE,OAEhC,EAAA,GAAA,SAAA,EAAA,GAAA,MAAA,KAAA,EAAA,EAAA,IAAA,EAAA,MAAA,GAAA,EAAA,EAAA,QAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAQ,eAAe,EAAa,OAAQ,GAGrD,KAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,GAG1D,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC9Fd,YAsCI,EAAc,EACd,EAAuC,GACzC,MAAI,IAAe,KACV,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAiB,CAAC,EAAG,GAAI,GAEzB,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,IAInD,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,aACzC,GAAM,GAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGjC,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAY,KAAM,GAAa,QAAS,IAE5D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,OAEP,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,iBASd,GAE5C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAyB,EACzB,GACN,GAAM,GAAc,EAAA,aAAa,2BAA2B,EAAQ,GAC9D,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAY,OACzB,EAAgB,EAAA,KAAK,eAAe,GAEpC,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GACxD,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GAExD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAEvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GACvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GAEvD,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAO,EAAI,EAAM,OACjB,EAAO,EAAI,EAAM,OAEjB,EACF,EAAG,EAAM,EAAO,GAAI,EAAM,EAAO,EAAI,GAAI,EAAM,EAAO,GACnD,EAAM,EAAO,EAAI,IAExB,EAAe,GAAK,EAAO,KAC3B,EAAe,GAAK,EAAO,SAG7B,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EACF,EAAG,EAAM,EAAS,GAAI,EAAM,EAAS,EAAI,GAAI,EAAM,EAAS,GACzD,EAAM,EAAS,EAAI,IAE1B,EAAe,GAAK,EAAS,KAC7B,EAAe,GAAK,EAAS,MAhBtB,EAAI,EAAG,EAAI,EAAe,OAAQ,MAAlC,GAmBX,MAAO,CAAC,EAAgB,EAAgB,ICvL5C,AAsBO,GAAM,IAAU,GAA8B,SAAC,EAAG,GAAM,MAAA,GAAI,IACtD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAGpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAwBsC,GAEpC,MAAO,UAAC,EAAQ,EAAO,GAGrB,OAFM,GACF,EAAA,KAAK,uBAAuB,EAA0B,EAAO,QACxD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,IChCX,YAiCI,EAAc,EAA0B,GAC1C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAQlB,OALM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,GAAS,EAAE,MACpB,EAAY,EAAA,KAAK,kBAAkB,EAAQ,GACxC,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,gBAclD,EAAc,EAA4B,GAC5C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAGlB,GAAM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAS,GAAS,EAAE,MACpB,EAAY,EAAU,EAAQ,EAAQ,GAC5C,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,IC3EtD,AAsBO,GAAM,IAAW,GAAsB,SAAC,GAAO,MAAA,MAAK,KAAK,KACnD,GAAiB,GAAwB,EAAA,KAAM,IAE/C,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,YAoBI,EAAmB,EAAoB,EACvC,GAIF,OAHM,GAAO,EAAA,KAAK,uBACd,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,GCpCT,AAqBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CACL,KAAM,EAAQ,EAAQ,EAAQ,EAC9B,KAAM,EAAQ,EAAQ,EAAQ,KAIzB,GACT,GAAiB,EAAA,SAAU,GAAc,IAEhC,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICrCd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,GAAI,KAAK,KAAK,KACxD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,YAuBI,EAAkB,EAAiB,EAAgB,EACnD,GACF,GAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,GACF,GAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAIhD,OADM,GAAU,EAAA,KAAK,uBAAuB,EAA0B,GAC7D,EAAI,EAAG,EAAI,EAAQ,EAAE,GAC5B,GAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,SAAC,EAAa,GAAM,MAAA,GAAM,EAAM,KAC/C,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEpB,MAAO,eAIL,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAAO,EAAA,EAAA,KAEd,GAAiB,EAAG,SAEd,GAAA,GAAA,EAAA,WAAA,iBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAA,WAAW,kBAAkB,EAAG,EAAQ,GAExC,GAAM,GAAO,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAClC,EAAU,GAAU,EAAM,EAAQ,EAAO,EAAE,MAAO,EAAE,OAC1D,MAAO,GAAQ,eAAe,EAAO,EAAE,MAAO,GAGzC,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IClEd,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAEpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAqBI,EAAmB,EAAkB,EAAiB,EACtD,GASF,OARM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GAEjC,EAAS,EAAA,KAAK,uBAChB,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAO,EAAE,GAK3B,OAJM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAGhC,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,EAAK,IAGvB,GAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE3B,MAAO,GC3CT,YAoBI,EAAuB,EAAc,EAAiB,GAgExD,OA1DM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAyDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GACtB,EAAI,EAAG,EAAI,EAAO,IACzB,EAAS,IAAM,EAAM,GAEvB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACxC,EAAS,IAAM,EAAM,GAevB,OAVM,GAA0C,GAG1C,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAA0B,GAC1B,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EAC/C,EAAI,EAAG,EAAI,EAAM,GAAQ,KAEhC,GAAI,GAAO,OACX,GAAI,EAEF,EAAU,EAAO,GAAG,gBAGpB,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG1C,EAAU,EAAW,KAAK,KAI5B,GAAI,EAAe,KAAa,OAC9B,EAAQ,GAAK,EAAe,QAE5B,GAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAOvB,GAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,GAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,SAAC,EAAoB,GACzC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAOxE,GAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAE7B,CACL,aAAc,EAAa,OAC3B,YAAW,EACX,QAAO,GCzJX,+MCAA,GAGM,IAAU,QCHhB,AA8BA,EAAA,gBAAgB,MAAO,WAAM,MAAA,IAAI,KAAkB,GC9BnD,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,YAoBI,EAAqB,EAAkB,EAAiB,EACxD,EACA,GAsBF,OArBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAS,EAAA,OAAO,EAAS,SAAU,GACnC,EAAa,EAAO,OAEpB,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAoB,EAAI,EACxB,EAAmB,EAAI,EAAQ,GAC5B,EAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAM1C,OALM,GAAW,EAAK,EAAe,EAC/B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,EAAkB,EAAoB,EAAK,EACxC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAQzC,OAPM,GAAW,EAAK,EAAc,EAC9B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAc,EACd,EAAW,EACX,EAAQ,EACH,EAAK,EAAO,EAAK,EAAO,GAAM,GAErC,OADM,GAAW,EAAmB,EAAK,EAAQ,GACxC,GAAK,EAAO,GAAK,EAAO,IAAM,GACrC,GAAM,IAAW,EAAW,GAAK,EAAQ,GACnC,GAAQ,EAAQ,GAAW,GACjC,AAAK,IAAa,OAAS,GAAQ,EACjC,EAAc,GACT,AAAI,IAAa,OACtB,IAAY,GACZ,KAGJ,GAAI,MAAM,GACR,MAGJ,GAAM,IAAe,EAAkB,EAAK,EAAmB,EAC/D,EAAW,IACP,IAAa,MAAQ,EAAW,EAAQ,EAKpD,MAAO,eAIL,EAAqB,EAAkB,EACvC,EAAmC,EACnC,GADmC,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,IAYF,OAXM,GAAe,EAAA,OAAO,EAAS,SAAU,SACzC,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,EAAA,OAAO,EAAQ,EAAO,GAC1B,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,GAG1C,OAFM,GAAW,EAAK,EAAe,EACjC,EAAQ,EACL,EAAQ,GACb,GAAS,EAKX,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAGzC,OAFM,GAAW,EAAK,EAAc,EAChC,EAAQ,EACL,EAAQ,GACb,GAAS,EAOX,OALM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAK,EAAO,EAAK,EAAO,GAAM,EAErC,OADM,GAAK,EAAK,EACP,EAAK,EAAO,EAAK,EAAO,GAAM,GACrC,GAAM,GAAK,EAAK,EACV,EAAQ,EAAK,IAAI,EAAG,EAAI,EAAI,GAClC,AAAI,EAAQ,GACV,GAAW,EACX,AAAI,EACF,EAAc,EACR,IAAI,EAAS,SAAW,GAAM,EAAS,QAAU,GAC3C,EAAS,WACb,EACH,GAAK,EAAS,QAAU,GAAM,EAAS,WAAa,EAEzD,EAAc,EAAK,EAAuB,GAKlD,EAAa,IAAI,EAAa,EAAG,EAAI,EAAI,IAKjD,MAAO,GCrJT,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAqBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBAyB9B,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAgB,EAAK,GAAe,GAEpC,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAGF,GAAM,GAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAGf,EAAG,IAAI,EAAU,EAAe,EAAG,EAAK,EAAK,GAKrD,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,IC1Fd,YAsBoC,GAK3B,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EAAG,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,KAAM,EAAA,EAAA,SAE/B,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGV,GAAiB,CAAC,EAAG,EAAM,EAAU,EAAO,GAAS,aAEhD,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAsBpB,OAnBM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACtC,EAAU,EAAQ,KAAK,IAAI,EAAS,QAAQ,OAC5C,EAAQ,EAAQ,EAAQ,KAAK,IAAI,EAAM,QAAQ,OAC/B,GAAI,cAAa,CAAC,IAClC,EAAU,EACZ,EAAQ,KAAK,IAAI,EAAO,QAAQ,OAChC,GAAI,cAAa,CAAC,IAChB,EAAU,GAAI,cAAa,EAAM,QAEjC,EAAgB,EAAQ,OACxB,EAAc,EAAM,OACpB,EAAgB,EAAQ,OACxB,EAAc,EAAM,OAEtB,EAAO,EACP,EAAK,EACL,EAAK,EACL,EAAK,EACA,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,EAAQ,GAAK,EAAQ,KAChB,GAAM,GAAK,EAAM,MAAS,EAAM,KAC7B,KAAK,KAAK,EAAQ,KAAQ,GAClC,AAAI,GAAQ,GACV,GAAO,GAET,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAGT,MAAO,GAAQ,eAAe,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAM,IAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,IC5Fd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,YAAa,SAAC,EAAI,GAC9D,GAAM,GAAY,EAClB,MAAI,GAAK,EAAU,aACV,EAAU,aAEZ,EAAK,EAAU,aAAe,EAAU,aAAe,IAGnD,GAA2B,CACtC,WAAY,EAAA,YACZ,YAAa,MACb,WAAY,IChCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YAsBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAED,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,EAAQ,OAAO,EAAE,QAEjB,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAEjC,GAAI,EAAM,oBAAsB,MAC9B,GAAM,GAAO,EAAM,mBAAmB,KAChC,EAAO,EAAM,mBAAmB,KAEtC,EAAK,MAAQ,EACb,EAAK,MAAQ,EAGf,MAAO,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,OAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,ICzDd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,KAED,EAAQ,EAAA,KAAK,eAAe,EAAM,EAAO,GAAG,OAAO,GACrD,EAAW,EAAA,aAAa,gBAAgB,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtE,GAAI,EAAA,KAAK,cAAc,KAAc,EACnC,MAAO,GAAQ,eAAe,EAAU,EAAO,GAAG,MAAO,IAI3D,GAAM,GAAU,EAAO,OAAO,SAAA,GAAK,MAAA,GAAA,KAAK,cAAc,EAAE,OAAS,IACjE,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAM,GAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAGlC,GAFA,EAAA,aAAa,uBAAuB,EAAQ,GAExC,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAC5D,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAE5D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAC3D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAE3D,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAO,IAEtE,SAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAUT,GAAM,GAAW,EAAQ,IAAI,SAAA,GAC3B,GAAM,GAAY,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAC7C,EAAQ,CAAC,GAAI,GACnB,MAAO,IAAQ,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAO,EAAE,MAAO,CAAC,MAAK,OAIxD,EACI,EAAA,aAAa,gBAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAE7D,GAAM,GAAU,EAAA,KAAK,uBACjB,EAAQ,GAAG,MAAoB,EAAA,KAAK,cAAc,IAEtD,GAAI,EAAS,GAAG,MAAM,KAAO,GAE3B,GAAI,GAAS,EACb,EAAS,QAAQ,SAAA,GACf,GAAM,GAAM,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACjC,EAAO,EAAA,KAAK,cAAc,EAAE,OAElC,EAAQ,IAAI,EAAK,GACjB,GAAU,SAGZ,GAAI,GAAY,EAEhB,EAAS,QAAQ,SAAA,GAKf,OAJM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAErC,EAAO,EAEF,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EAEpC,OADM,GAAS,EAAM,EAAS,GAAK,EAC1B,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EACpC,EAAQ,EAAS,GAAO,EAAM,KAIlC,GAAa,EAAE,MAAM,KAIzB,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtD,EACF,EAAQ,eAAe,EAAe,EAAO,GAAG,MAAO,GAE3D,SAAS,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KAErD,EAGF,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,ICjId,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,SAAC,GAwCX,OAxCY,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OACJ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAE,MAAM,OAEhB,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAChD,EAAa,EAAO,MAAM,OAE1B,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOI,EAAU,EAAA,KAAK,cAAc,GAC7B,EAAU,EAAS,OACnB,EAAa,EAAA,KAAK,kBAAkB,EAAE,MAAO,GAM1C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,IAAO,EAAO,EAAc,EAAQ,KACjC,GAAI,EAAG,GAAI,EAAY,EAAE,IAEhC,OADI,IAAS,OAAO,iBACX,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,GAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAS,EAAA,KAAK,WAChB,CAAC,EAAG,GAAK,GAAK,IAAI,EAAO,EAAA,KAAK,eAAe,EAAE,QAC7C,GAAc,EAAA,KAAK,WACrB,CAAC,GAAG,GAAG,IAAI,EACX,EAAA,KAAK,eAAe,EAAO,QACzB,GAAM,EAAM,IAAU,EAAW,IACvC,AAAI,GAAM,IACR,IAAS,MAMnB,GAAM,IAAc,EAAA,KAAK,WACrB,CAAC,EAAG,EAAM,EAAM,IAAI,EAAS,EAAA,KAAK,eAAe,IACrD,EAAW,IAAe,GAMlC,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAY,EAAE,OAAQ,EAAU,EAAE,OAExD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAU,MAAO,EAAE,SCrG9C,AAsBO,GAAM,IAA+C,CAC1D,WAAY,EAAA,yBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,yBAAwB,QACtC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EAAY,EAAA,KAAK,0BACD,EAAO,MAAO,EAAO,OAOlC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAO,EACP,GAAO,EACF,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAO,GACP,GAAO,MAMjB,EAAU,IAAM,IAAM,IAAM,EAAI,GAAG,GAAM,GAAM,GAMvD,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAO,MAAO,EAAO,OAEhE,MAAO,CAAC,OAAM,GAAE,MAAO,EAAO,MAAO,MAAO,EAAO,SCrHvD,AAsBO,GAAM,IAA8C,CACzD,WAAY,EAAA,wBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,wBAAuB,QACrC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EACF,EAAA,KAAK,0BAA0B,EAAE,MAAO,EAAE,OAOrC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAU,EAAO,EAAK,EAAI,EAC1B,GAAU,EAAO,EAAK,EAAI,EACrB,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAS,GACT,GAAS,MAMnB,EAAU,GAAG,IAAQ,IAAQ,IAAM,EAAI,GAAG,GAAM,GAAM,GAM9D,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAE,MAAO,EAAE,OAEtD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrH7C,AAsBO,GAAM,IACT,GAA6B,SAAC,EAAW,GAAc,MAAA,GAAI,IAClD,GAAM,GAAiB,EAAA,IAAK,IAE5B,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC7Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,IAAM,EAAI,EAAM,KAAK,IAAI,GAAM,IAEnD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC3Bd,AAqBA,GAAM,IAAI,EAAA,aAAa,MACjB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAEX,GAAgB,GACzB,EAAA,IACA,SAAC,GACC,GAAM,GAAO,KAAK,KAAK,GACjB,EAAI,KAAK,IAAI,GACb,EAAI,EAAO,GAAM,GAAI,GAC3B,MAAO,GACF,GACK,OAAK,EAAI,IAAM,EAAK,IAAM,EAAI,IAAM,EAAI,IAAM,EAC/C,KAAK,IAAI,CAAC,EAAI,MAIhB,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Cd,YAmCI,EAAmB,EACnB,GAgBF,OAfM,GAAa,EAAM,MACnB,EAAQ,EAAW,GACnB,EAAW,EAAW,GAEtB,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EAAS,EAAU,mBAAmB,KACtC,EAAS,EAAU,mBAAmB,KAGtC,EAAc,CAAC,EAAO,GACtB,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAA,KAAK,uBAAuB,UAAW,GACpD,EAAa,EAAA,KAAK,uBAAuB,UAAW,GAEjD,EAAI,EAAG,EAAI,EAAO,KAmBzB,OAjBM,GAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAE7B,EAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAG7B,EAAQ,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAG,KAAM,GAAI,QAAS,IAGtD,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,KACP,EAAM,EAAA,aAAa,uBAAuB,EAAM,GAE7C,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAI,EAAA,aAAa,oBAAoB,EAAK,GAChD,EAAW,EAAI,EAAW,GAAK,EAAE,KACjC,EAAW,EAAI,EAAW,GAAK,EAAE,KAGnC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAG3C,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,GAChD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,GAAY,QAAS,IAE1D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,cAIL,EAAmB,EACnB,GACF,GAAM,GAAY,EAAA,KAAK,cAAc,EAAM,OAErC,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAG5D,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAGlE,GAAI,GAAc,IAChB,GAAM,GACF,GAAU,EAAU,EAAU,EAAW,EAAS,GAEhD,EAAc,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAEjD,GAAI,GACF,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,EAAO,MACvD,EACF,EAAW,eAAe,EAAa,UAAW,EAAO,MAEvD,EAAuB,EAAW,eACpC,GAAI,UACJ,EAAA,KAAK,kBAAkB,EAA8B,YACnD,EACF,GAAS,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAS,IAExC,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAW,QAAS,IAEhD,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAe,QAAS,IAGpD,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OACtC,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OAE5C,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,CAAC,KAAM,EAAa,KAAM,GAGnC,MAAO,QAEP,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EACF,GAAyB,EAAM,EAAW,GAE9C,MAAO,GAAA,aAAa,uBAAuB,IAI/C,YAAuB,GACrB,MAAQ,GAAO,EAAO,KAAO,EAI/B,YACI,EAAwB,EAAwB,EAChD,EACA,GACF,GAAI,IAAS,EACX,MAAO,CAAC,KAAM,EAAU,KAAM,GAGhC,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EAAO,EAAO,EAEd,EAAc,EAAA,aAAa,qBAAqB,GAEhD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAa,EAAA,aAAa,oBAAoB,GAE9C,EAAc,EAAW,KACzB,EAAc,EAAW,KAEzB,EAAW,CAAC,EAAY,QAExB,EACF,EAAW,eAAe,EAAU,UAAW,GAC7C,EACF,EAAW,eAAe,EAAU,UAAW,GAE7C,EAAgB,GAClB,CAAC,OAAQ,CAAC,KAAM,EAAa,KAAM,GAAc,QAAS,IAGxD,EACF,GAAU,EAAc,EAAc,EAAM,EAAS,GAEnD,EAAgB,EAAa,KAC7B,EAAgB,EAAa,KAE7B,EAAa,CAAC,EAAc,QAE5B,EACF,EAAW,eAAe,EAAY,UAAW,GAC/C,EACF,EAAW,eAAe,EAAY,UAAW,GAE/C,EAAkB,GAAQ,CAC9B,OAAQ,CAAC,KAAM,EAAe,KAAM,GACpC,QAAS,IAGL,EACF,GAAU,EAAa,EAAa,EAAM,EAAS,GAEjD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAI,EAAA,aAAa,UAAU,EAAM,GACjC,EAAS,CAAC,EAAE,KAAK,QAEjB,EAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAC3D,GAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAE3D,GAAc,GAChB,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,IAAY,QAAS,IAEpD,GACF,GACI,CAAC,OAAQ,CAAC,EAAG,GAAa,EAAG,GAAiB,QAAS,IAGzD,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAErB,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAGrB,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAEV,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAGV,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAC9C,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAEpD,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IAElC,CAAC,KAAM,GAAW,KAAM,IAIjC,YACI,EAAkB,EAAc,GAGlC,OAFM,GAAM,GAAI,cAAa,EAAO,GAE3B,EAAI,EAAG,EAAI,EAAM,KAGxB,OAFI,GAAO,EACP,EAAO,EACF,EAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAI,EAAA,aAAa,SAAS,EAAI,EAAG,EAAM,GACvC,EAAO,EAAA,aAAa,oBAAoB,EAAsB,GACpE,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAC3C,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAE7C,AAAI,GACF,IAAQ,EACR,GAAQ,GAEV,EAAA,aAAa,mBAAmB,EAAK,EAAM,EAAM,GAEnD,MAAO,GCjVT,YAuBoB,GAEX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAO,GAElC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICtDd,AAsBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,MACb,WAAY,SAAC,GAUX,OAVY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GAEX,EAAS,KAAK,MAAM,EAAa,GACjC,EAAS,EAAc,EAAY,EAAY,EAEjD,EAAc,EAAU,GAE5B,GAAI,GAAU,GAAK,EAAS,GAE1B,GAAM,GAAmB,EAAS,EAC5B,EACF,EAAc,EAAY,EAAmB,EACjD,EAAc,EAAU,GAE1B,EAAO,GAAU,EAMzB,GAAM,GAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,EAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SCpErD,YAuBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAM,GAEjC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtDd,AAqBO,GAAM,IACT,GAAgB,EAAA,SAAU,SAAC,GAAO,MAAA,QAAO,SAAS,GAAM,EAAI,GAAG,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,IAAI,KAAQ,SAAW,EAAI,GAAG,QAEzD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,QAAO,MAAM,GAAM,EAAI,GAAG,QAEhD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAK,EAAI,GAAG,QAEvC,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC3Bd,AA2BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAa,EACf,EAAS,EAAE,MACT,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,EAAkB,GACnD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACvD,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC1C,GAAI,GAAgB,MAElB,OADM,GAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAO,EAAa,IAGpC,EAAQ,GAAc,EAAO,EAAQ,EAAE,MAAO,EAAc,GAC5D,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAElD,EAAS,EAGX,GAAiB,EAAG,OACpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGd,EAAa,EAAA,KAAK,cAAc,GAEhC,EAAS,GAAQ,EAAO,EAAY,EAAa,EAAE,OACnD,EAAS,EAAW,MAAM,EAAQ,EAAa,EAAE,OAEnD,EAAW,EACf,GAAI,GAEF,GAAM,GAAW,EAAA,aAAa,qBAAqB,EAAa,GAChE,EAAW,EAGb,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,SCtE9C,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAyBlC,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAY,EAAA,OACd,EAAS,SAAU,EAAE,MACrB,GAAiB,EAAS,EAAE,MAAO,EAAE,MAAO,GAAU,QACpD,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,GAAM,GAAS,EAAwB,EAAuB,EACzD,EAAU,IAAI,EAAG,EAAK,EAAK,GAC1B,GAAS,EAAK,EAAuB,EAErC,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAAQ,IAGvB,EAAG,IAAI,EAAS,EAAG,EAAK,EAAK,GAKrC,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,ICnGd,YAoBI,EAAqB,EAAkB,EACvC,EAA8B,GAChC,GAAM,GAAU,EAAA,KAAK,eAAe,GAC9B,EAAW,GAAK,EAAS,EAAQ,EAAO,EAAS,EAAU,OAC3D,EAAe,GACjB,EAAS,EAAQ,EAAO,EAAU,GAAM,GAE5C,MAAO,CAAC,EAAS,OAAQ,EAAa,QC3BxC,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAa,EACnB,GAAiB,EAAG,qBAEpB,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,CAAC,EAAG,GAAI,GACN,EAAA,GAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGT,EACF,EAAW,MAAM,EAAwB,EAAS,SAAU,EAAE,OAC5D,EACF,EAAW,MAAM,EAAuB,EAAS,SAAU,EAAE,OACjE,MAAO,CACL,CAAC,OAAQ,EAAc,MAAO,EAAS,SAAU,MAAO,EAAE,OAC1D,CAAC,OAAQ,EAAe,MAAO,EAAS,SAAU,MAAO,YC/C/D,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEnB,GAAiB,EAAO,2BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC3C7B,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEnB,GAAiB,EAAO,8BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAG,GAAM,MAAC,KAAM,EAAK,EAAI,IAC/C,GACT,GAAiB,EAAA,SAAU,GAAc,KAAsB,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC9Bd,YAuBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,SAAU,EAAA,EAAA,cAEjB,GAAiB,EAAG,OAEpB,GAAM,GAAW,EAAS,IACtB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAE,MAAM,GAAK,EAAE,KAE9C,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAE5B,EAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAQ,EAAE,MAAM,OAChB,EAAW,EAAA,KAAK,eAAe,EAAE,OAEjC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EACF,EAAA,KAAK,uBAAuB,EAAE,MAA0B,GAE5D,AAAI,IAAkB,GACpB,EAAQ,KAAK,GAGf,OAAS,GAAI,EAAG,EAAI,EAAO,KACzB,GAAM,GAAS,EAAA,KAAK,WAAW,EAAG,EAAO,GACnC,EAAY,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAM,KAC3C,EAAW,EAAA,KAAK,WAAW,EAAW,EAAY,GAExD,EAAQ,GAAY,EAAM,GAG5B,GAAM,GAAQ,EAAQ,MAAM,EAAS,EAAU,EAAE,OAEjD,MAAO,CAAC,OAAQ,EAAO,MAAO,EAAU,MAAO,EAAE,OAG5C,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICnEd,AAqBO,GAAM,IAAuB,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAI,IAE/D,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,MACb,WAAY,SAAC,GAiBX,OAjBY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAmB,IAEnB,EAAY,KAAK,IAAI,GACrB,EAAY,KAAK,IAAI,GACrB,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GACX,EAAI,EAAO,GAGb,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACrD,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACzD,EAAS,KAAK,MAAM,EAAS,GAC7B,EAAS,KAAK,MAAM,EAAS,GAE7B,GAAI,GAAc,EAUlB,GATA,AAAI,MAAO,IAAc,UACvB,CAAI,IAAY,EACd,EAAc,EAEd,EAAc,EAAU,IAKxB,GAAU,GAAK,EAAS,GAAc,GAAU,GAChD,EAAS,GAEX,GAAM,GAAmB,EAAU,GAAa,GAC1C,EAAmB,EAAS,EAC5B,GACF,EAAc,EAAmB,EAAmB,EACxD,EAAc,EAAU,IAG1B,GAAM,IAAS,EAAc,EAAY,EAAY,EACrD,EAAO,IAAU,EAMzB,GAAM,IAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,GAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SC3FrD,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAErD,GAAM,GAAO,KAAK,MAAM,GACxB,MAAI,GAAK,EAAO,GACP,KAAK,MAAM,GACT,EAAK,EAAO,GACd,KAAK,KAAK,GAEb,EAAO,IAAQ,EACV,EAEA,EAAO,IAKP,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICxCd,AAqBA,GAAM,IAAa,EAAA,aAAa,gBAC1B,GAAQ,EAAA,aAAa,WAEd,GAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,IAAM,EACD,GAAQ,EAER,GAAc,MAAK,IAAI,GAAM,KAI3B,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICnCd,AAqBO,GAAM,IACT,GAAgB,EAAA,QAAS,SAAC,GAAO,MAAA,GAAK,GAAI,KAAK,IAAI,CAAC,MAE3C,GAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,GAAK,EACA,GACE,EAAK,EACP,EAEA,IAIE,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IClCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AA0BA,GAAM,IAAU,sBACV,GAAY,KAAK,IAAI,IAAW,EAEzB,GAAqB,GAAgB,EAAA,SAAU,SAAC,GAG3D,GAAM,GAAW,EAAK,CAAC,GAIjB,EAAW,EAAK,GAEhB,EAAO,KAAK,IAAI,GAClB,EAEJ,MAAI,GACF,EAAS,EACJ,AAAI,EACT,EAAS,EAET,EAAS,KAAK,IAAI,EAAM,GAEnB,IAGI,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICtDd,YAwB0B,GAKjB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EACA,EAAA,EAAA,KAEP,GAAiB,EAAG,aAKpB,OAHM,GAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAS,GAAc,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAEvD,EAAS,EAAQ,MAAM,EAAQ,EAAU,EAAE,OACjD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,OAGrC,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,MACb,WAAY,ICpDd,YA0B+B,GAKtB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,WAAY,EAAA,EAAA,SAEnB,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAA,KAAK,cAAc,GAE1B,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAA0B,GAE1B,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,GAAY,WAAW,CACrC,OAAQ,CAAC,EAAC,GACV,QAAO,EACP,MAAO,CAAC,SAAU,EAAkB,cAAe,KAG/C,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EACF,EAAA,aAAa,oBAAoB,EAAQ,MAAO,EAAY,EAAM,IAEhE,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,MAAO,GACrC,EACF,GAAQ,CAAC,OAAQ,EAAe,QAAO,EAAE,MAAO,IAE9C,EAAmC,CAAC,EAAG,GACvC,EACe,CAAC,KAAM,GACtB,EACF,GAAU,CAAC,OAAQ,EAAiB,QAAO,EAAE,MAAO,IAElD,EAAqC,CAAC,EAAG,GACzC,EAAmC,CAAC,MAAO,GAC3C,EAAS,GACX,CAAC,OAAQ,EAAqB,QAAO,EAAE,MAAO,IAElD,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAAqC,CAChD,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,ICvFd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,EACD,EAAa,EACnB,GAAiB,EAAG,UAIpB,OAFM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAY,GAAI,cAAa,EAAO,QACjC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAQ,EAAO,GACrB,EAAU,GAAK,EAAQ,EAEzB,GAAM,GAAS,EAAW,MAAM,EAAW,EAAE,MAAO,EAAE,OACtD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrC7C,AAsBO,GAAM,IAAwB,GAA8B,SAAC,EAAG,GACrE,GAAM,GAAO,EAAI,EACjB,MAAO,GAAO,IAEH,GACT,GAAiB,EAAA,kBAAmB,IAE3B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,IChCd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,EAAI,GACvD,GAAM,GAAY,EAClB,MAAI,OAAM,GACD,IAEA,EAAK,EAAI,EAAI,EAAU,QAIrB,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICjCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,YAyBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAEpB,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,IC5Cd,AA6FA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,0yDCvKjB,AAiBA,GAAM,IAAmD,GAEnD,GAA2C,CAC/C,MAAO,GACP,UAAW,GACX,mBAAoB,GACpB,sBAAuB,GACvB,MAAO,GACP,QAAS,GACT,6BAA8B,gBAI5B,EAAsB,GACxB,GAAS,GAAgB,cAGK,GAC9B,GAAI,CAAE,KAAgB,MACpB,GAAM,GAAS,GAAyB,GACxC,GAAI,IAAW,KACb,GAAS,GAAgB,MAEzB,gBAAQ,IAAI,0CAA2C,GAChD,KAGX,GAAM,GAAK,GAAS,GACpB,MAAI,GAAG,gBACL,OAAO,IAAS,GACT,GAAgB,IAGzB,GAAG,QAAQ,EAAG,YACd,EAAG,QAAQ,EAAG,cACd,EAAG,QAAQ,EAAG,OACd,EAAG,QAAQ,EAAG,QACd,EAAG,QAAQ,EAAG,qBACd,EAAG,QAAQ,EAAG,iBACd,EAAG,OAAO,EAAG,cACb,EAAG,OAAO,EAAG,WACb,EAAG,SAAS,EAAG,MAER,GAAS,IAGlB,YAAsB,GACpB,GAAI,MAAO,kBAAoB,aAAe,IAAiB,EAC7D,MAAO,IAAI,iBAAgB,IAAK,KAC3B,GAAI,MAAO,WAAa,YAC7B,MAAO,UAAS,cAAc,UAE9B,KAAM,IAAI,OAAM,0CAIpB,YAAkC,GAChC,GAAI,IAAiB,GAAK,IAAiB,EACzC,KAAM,IAAI,OAAM,0DAElB,GAAM,GAAS,GAAa,GAM5B,MAJA,GAAO,iBAAiB,mBAAoB,SAAC,GAC3C,EAAG,iBACH,MAAO,IAAS,IACf,IACC,IAAiB,EACX,EAAO,WAAW,QAAS,KAC3B,EAAO,WAAW,qBAAsB,IAG3C,EAAO,WAAW,SAAU,ICxFrC,AAoBA,GAAY,IAAZ,AAAA,UAAY,GAgBV,EAAA,EAAA,MAAA,GAAA,QAkBA,EAAA,EAAA,aAAA,GAAA,iBAlCU,IAAA,IAAa,KAqCzB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,SAAA,GAAA,aAJU,IAAA,IAAY,KAOxB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,yBAAA,GAAA,2BACA,EAAA,EAAA,mBAAA,GAAA,qBACA,EAAA,EAAA,mBAAA,GAAA,uBALU,IAAA,IAAmB,iBAqC3B,EAAc,GAChB,MAAO,CAAC,EAAS,eAIf,EAAoB,GACtB,MAAO,GAAa,cAWW,GAC/B,GAAM,GAAO,EAAA,KAAK,cAAc,GAC1B,EAAe,KAAK,KAAK,EAAO,GACtC,MAAO,GAAA,KAAK,oBAAoB,eA6B9B,EAAc,GAChB,MAAO,CACL,KAAK,IAAI,EAAG,KAAK,KAAK,EAAU,IAAK,KAAK,IAAI,EAAG,KAAK,KAAK,EAAO,iBAKlE,EAAc,GACV,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,GAAI,EAAI,cAsBb,EAA2B,GAE7B,GAAM,GAAQ,EAEV,EACA,EACA,EACA,EACA,EAEA,EACA,EAEA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAsB,EAAM,KAC5B,EAA0B,EAAM,KAChC,EAAgC,EAAM,QACtC,EAA4B,EAAM,QAClC,EAAqB,EAAM,IAC3B,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,EAAM,WAC7B,EAAmB,EAAM,OAEzB,GAAsB,EAAG,KACzB,EAA0B,EAAG,KAC7B,EAAgC,EAAG,KACnC,EAA4B,EAAM,KAClC,EAAqB,EAAG,KACxB,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,GAA6B,KAChD,EAA0B,eAC1B,KACJ,EAAmB,EAAG,OAExB,EAAwB,EAAG,KAEpB,CACL,oBAAmB,EACnB,wBAAuB,EACvB,8BAA6B,EAC7B,0BAAyB,EACzB,mBAAkB,EAClB,sBAAqB,EACrB,0BAAyB,EACzB,mBAAkB,EAClB,qBAAoB,EACpB,iBAAgB,GCzOpB,YAsBgC,EAA2B,GACzD,GAAM,GAAc,IACpB,MAAI,GAAA,MAAM,QAAQ,UAChB,GAAgB,GAEX,EAGT,YAAyB,GACvB,GAAM,GAAQ,EAAG,WACjB,GAAI,IAAU,EAAG,SACf,KAAM,IAAI,OAAM,gBAAkB,GAAqB,EAAI,IAK/D,GAAM,IAAc,QACd,GAAc,kBAEa,GAC/B,MAAI,KAAA,MAAM,QAAQ,iCAAmC,IAAQ,GACxD,GAAc,KAAK,IAAI,IAAQ,KAAK,IAAI,GAAO,gBAOlD,EAA2B,GAC7B,OAAQ,OACD,GAAG,SACN,MAAO,eACJ,GAAG,aACN,MAAO,mBACJ,GAAG,cACN,MAAO,oBACJ,GAAG,kBACN,MAAO,wBACJ,GAAG,8BACN,MAAO,oCACJ,GAAG,cACN,MAAO,oBACJ,GAAG,mBACN,MAAO,6BAEP,MAAO,sBAAsB,eAK/B,EAA2B,GAC7B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,aAAa,IAC1B,cAAgB,EAAgB,gDAIlC,EAA2B,GAC7B,GAAM,GAA4B,GAC9B,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,gBAC7B,wCAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAc,KACrD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAc,EAAG,kBAAoB,GAC7D,cAAQ,IAAI,EAAG,iBAAiB,IAC1B,GAAI,OAAM,oCAElB,MAAO,eAIL,EAA2B,GAC7B,GAAM,GAA8B,GAChC,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,kBAC7B,0CAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAgB,KACvD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAgB,EAAG,kBAAoB,GAC/D,SACI,EAAsB,EAAG,iBAAiB,IACxC,GAAI,OAAM,sCAElB,MAAO,GAGT,GAAM,IAAkB,2BACxB,YACI,EAAsB,GACxB,GAAM,GAAwB,GAAgB,KAAK,GACnD,GAAI,GAAyB,MAC3B,QAAQ,IAAI,wCAAwC,GACpD,QAAQ,IAAI,GACZ,OAWF,OARM,GAAa,CAAC,EAAsB,GAEpC,EAAc,EAAa,MAAM;GACjC,EAAM,EAAY,OAAO,WAAW,OAAS,EAC7C,EAAuB,EAAY,IACrC,SAAC,EAAM,GACH,MAAA,GAAA,KAAK,SAAU,GAAa,GAAG,WAAY,GAAO,IACtD,EAAgB,EACX,EAAI,EAAG,EAAI,EAAqB,OAAQ,IAC/C,EAAgB,KAAK,IAAI,EAAqB,GAAG,OAAQ,GAG3D,GAAM,GAAmB,EAAqB,MAAM,EAAG,EAAa,GAC9D,EAAY,EAAqB,MAAM,EAAa,EAAG,GACvD,EAAkB,EAAqB,MAAM,GAEnD,QAAQ,IAAI,EAAiB,KAAK;IAClC,QAAQ,IAAI,EAAc,MAAM;GAAM,IACtC,QAAQ,IACJ,MAAM,EAAA,KAAK,SAAS,EAAU,GAAI,GAClC,iEACJ,QAAQ,IAAI,EAAgB,KAAK;gBAGL,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGR,EAA2B,GAErD,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,KAClC,EAAG,oBAAoB,EAAS,EAAG,eAAiB,GACtD,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,2DAKhB,EAA2B,GAE7B,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,KACtC,EAAG,oBAAoB,EAAS,EAAG,mBAAqB,GAC1D,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,iDAKhB,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,EAAM,EAAG,eACxD,cAIL,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,KAC9D,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,EAAM,EAAG,eACvD,gBAIP,MAAI,GAAA,MAAM,UAAU,mBAAqB,EAChC,EAEF,cAGqB,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGA,EAAe,GACjD,GAAM,GAAiB,EAAA,MAAM,UAAU,0BACvC,GAAK,GAAS,GAAO,GAAU,GAC7B,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IACrC,KAAM,IAAI,OAAM,0BAA4B,EAAY,gBAE1D,GAAK,EAAQ,GAAoB,EAAS,GACxC,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IAC/B,EAAM,IAAI,EAAc,IAAI,EAAc,IAChD,KAAM,IAAI,OACN,0BAA4B,EAC5B,qDAAuD,EAAM,kBAInC,GAChC,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,qBAAqB,kDAIpC,EAA2B,EAAuB,EAClD,EAAqB,EAA6B,EAClD,GACF,GAAM,GAAM,EAAG,kBAAkB,EAAS,GAC1C,MAAI,KAAQ,GAGH,GAET,IAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GACI,EACA,WAAM,MAAA,GAAG,oBACL,EAAK,EAAqB,EAAG,MAAO,GAAO,EAC3C,KACR,GAAa,EAAI,WAAM,MAAA,GAAG,wBAAwB,KAC3C,gBAIL,EAA2B,EAAuB,GACpD,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,iBAInD,EAA2B,GAC7B,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAInD,EAA2B,EAC3B,GACF,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,mBAAmB,EAAS,IACzC,YAAc,EAAc,yCAI9B,EAA2B,EAC3B,GACF,MAAO,GAAG,mBAAmB,EAAS,eAIpC,EAA2B,EAC3B,EAA8C,GAChD,GAAa,EAAI,WAAM,MAAA,IAAgB,EAAI,EAAS,KACpD,GAAa,EAAI,WAAM,MAAA,GAAG,UAAU,EAAwB,iBAGtB,GACtC,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QAC1D,GAAa,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,UACpE,GAAa,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,sBAIjE,EAA2B,EAC3B,GACF,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,iBAIpE,EAA2B,GAC7B,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,KAAM,iBAGjC,GAClC,GAAM,GAAS,EAAG,uBAAuB,EAAG,aAC5C,GAAI,IAAW,EAAG,qBAChB,KAAM,IAAI,OACN,8BAAgC,GAA2B,EAAI,gBAKnE,EAA2B,GAC7B,OAAQ,OACD,GAAG,kCACN,MAAO,wCACJ,GAAG,0CACN,MAAO,gDACJ,GAAG,kCACN,MAAO,wCACJ,GAAG,wBACN,MAAO,kCAEP,MAAO,iBAAiB,GAI9B,YACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAa,EAAI,WAAM,MAAA,OAC/C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,GAElB,MAAO,GAGT,YAA6B,EAA2B,GACtD,GAAM,GAAiB,EAAG,iCAAmC,EACvD,EAAgB,EAAc,EAAG,SACvC,GAAI,EAAgB,EAAG,UAAY,EAAgB,GACjD,GAAM,GAAmB,2BAA2B,EAAc,IAClE,KAAM,IAAI,OAAM,0BAA0B,EAAgB,kBAIlC,EAAiB,GAAA,MAAA,KAAA,QAAA,GAAA,GACpC,EAAA,KAAK,cAAc,EAAM,MAAM,EAAG,EAAM,OAAS,gBAG9B,GAC1B,GAAI,EAAM,SAAW,EACnB,KAAM,OAAM,wDAGd,MAAO,CACL,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EAAG,EAAM,EAAM,OAAS,gBAI5C,GAC3B,GAAI,GAAsC,CAAC,EAAG,EAAG,GAC3C,EAAW,EAAM,SAAW,GAAM,EAAM,SAAW,GAAK,EAAM,KAAO,EAC3E,MAAK,IACH,GACI,CAAC,GAAY,IAAM,OAAK,GAAY,KAEnC,cAIL,EAAoB,SAAA,AAAA,IAAA,QAAA,GAAA,IACtB,GAAI,GAAa,EAAA,MAAM,UAAU,0BAsBjC,GArBA,AAAI,GACF,GAAa,EAAa,EAO1B,EAAW,EAAS,IAChB,SAAC,EAAG,GAAM,MAAA,IAAK,EAAS,OAAS,EAC7B,EAAA,KAAK,kBAAkB,EAAS,IAChC,EAAS,KAIjB,AAAI,EAAS,SAAW,GACtB,GAAW,CAAC,EAAG,EAAS,MAKxB,EAAS,SAAW,GACtB,GAAM,GAAgB,EAAA,KAAK,aAAa,GACxC,EAAW,EAAc,SAG3B,GAAI,GAAO,EAAA,KAAK,cAAc,GAC9B,GAAI,EAAS,QAAU,GAAK,GAAQ,EAClC,MAAO,CAAC,EAAG,GACN,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,IAAM,EACjB,MAAO,GACF,GACH,EAAS,SAAW,GAAK,EAAS,GAAK,EAAS,IAAM,GACtD,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAI,EAAS,IACvC,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,IAAM,EAC/B,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,IACvC,GACH,EAAS,SAAW,GACpB,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,GAC3C,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAK,EAAS,GAAI,EAAS,IACrD,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,EAC7C,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,GAAK,EAAS,IAE1D,GAAI,GAOF,GAAM,GAAW,GAAY,GACzB,EAAO,EAAG,EAAO,EACrB,MAAI,GAAS,QACX,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,IAET,EAAO,EAAY,GAAO,GAAM,GAAO,GAChC,EAAA,KAAK,oBAAoB,GAAM,IAAI,SAAA,GAAK,MAAA,GAAI,IAErD,MAAO,GAAA,KAAK,oBAAoB,GAIpC,YAAgB,GACd,MAAO,GAAI,IAAM,cAOW,EAAkB,GAI9C,GAHA,EAAS,EAAO,MAAM,IACtB,EAAS,EAAO,MAAM,IAElB,EAAA,KAAK,YAAY,EAAQ,GAC3B,MAAO,GAGT,GAAI,CAAC,EAAO,QAAU,CAAC,EAAO,OAC5B,MAAO,GAGT,GAAI,EAAO,KAAO,GAAK,EAAO,KAAO,GAAK,EAAO,KAAO,GACpD,EAAO,KAAO,EAChB,MAAO,GAGT,GAAI,EAAO,SAAW,EAAO,QAC3B,GAAM,GAAa,EAAO,MAAM,IAAI,GAC9B,EAAa,EAAO,MAAM,IAAI,GACpC,GAAI,IAAe,EACjB,MAAO,GAGT,GAAI,GAAO,IAAe,GAAO,IAC5B,GAAO,KAAO,GAAK,EAAO,KAAO,GACpC,MAAO,GAGX,MAAO,GAAO,KAAO,EAAO,IAAM,GAAO,EAAO,KAAO,GAAO,EAAO,IAMvE,GAAI,IACA,eAEmC,GACrC,GAAI,IAAoB,MACtB,GAAM,GAAK,GAAgB,GAC3B,GAAmB,EAAG,aAAa,EAAG,kBAExC,MAAO,kBAIP,GAAmB,mBAGnB,GAAyB,iBAGY,GACrC,GAAI,IAA0B,MAC5B,GAAM,GAAK,GAAgB,GAC3B,GAAyB,EAAG,aAAa,EAAG,yBAG9C,MAAO,MAAK,IAAI,GAAI,gBAG4B,GAEhD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAI,GACE,EAAK,GAAgB,GAE3B,MAAI,IAAa,EAAI,oCACjB,IAAiB,EACnB,EAAoB,EACf,AAAI,GAAa,EAAI,4BAC1B,EAAoB,EAEpB,EAAoB,EAEf,cAGoB,EAA2B,GACtD,GAAM,GAAM,EAAG,aAAa,GAC5B,MAAO,IAAO,iBAGsB,GACpC,IACE,GAAM,GAAK,GAAgB,GAC3B,GAAI,GAAM,KACR,MAAO,SAEF,GACP,eAAQ,IAAI,qCAAsC,GAC3C,GAET,MAAO,eAG0C,GAEjD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,WAGL,CAAC,GAAa,EAAI,0BACpB,MAAO,GAIX,GAAM,GAAwB,GAAuC,GACrE,MAAO,eAYqC,GAC5C,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,GAET,GAAI,CAAC,GAAa,EAAI,4BACpB,MAAO,QAGT,GAAI,GAAa,EAAI,0BACnB,MAAO,IAAuC,GAGhD,GAAM,GAA0B,8BAChC,GAAI,GAAa,EAAI,IACnB,GAAM,GACF,EAAG,aAAa,GACpB,MAAO,IACH,EAAI,GAGV,MAAO,GAGT,GAAM,GAAwB,GAAuC,GACrE,MAAO,GAGT,YAAgD,GAE9C,GAAM,GAAY,GAAiB,GAE7B,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,oBAAqB,EAAO,EAAQ,EAChE,EAAU,mBAAoB,EAAU,iBAAkB,MAE9D,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,EAGT,YAEI,EAA2B,GAC7B,GAAM,GAAY,GAAiB,EAAI,GACjC,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,wBAAyB,EAAO,EAAQ,EACpE,EAAU,mBAAoB,EAAU,qBAAsB,MAElE,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,cAG2B,GAClC,GAAI,IAAiB,EACnB,MAAO,GAET,GAAM,GAAK,GAAgB,GAGrB,EAAa,EAAW,WAAa,KAC3C,MAAO,eAIL,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MAAG,GAAM,smCC7qBzB,AAqBA,GAAM,IAAM,EAAA,MASZ,GAAI,aAAa,YAAa,WAAM,MAAA,IAAI,UAAU,iBAAmB,IAGrE,GAAI,aAAa,gBAAiB,WAChC,MAAI,IAAsB,GACjB,EACE,GAAsB,GACxB,EAEF,IAIT,GAAI,aAAa,iCAAkC,WAAM,MAAA,KAEzD,GAAI,aACA,yBAA0B,WAAM,MAAA,IAAI,IAAI,mBAAqB,IAGjE,GAAI,aAAa,oBAAqB,WAAM,MAAA,KAG5C,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aAAa,aAAc,WAAM,MAAA,IAAI,QAAQ,eAGjD,GAAI,aAAa,2BAA4B,WAAM,MAAA,IAAI,QAAQ,gBAG/D,GAAI,aAAa,kBAAmB,WAAM,MAAA,IAAI,QAAQ,gBAItD,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aACA,+BAAgC,WAAM,MAAA,IAAI,QAAQ,gBAGtD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aAAa,sBAAuB,WAAM,MAAA,IAAI,QAAQ,gBAG1D,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aACA,yBACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAG/C,GAAI,aACA,+BACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAU/C,GAAI,aAAa,+CAAgD,WAC/D,GAAM,GAAe,GAAI,UAAU,iBAEnC,MAAI,KAAiB,EACZ,EAEF,GAAkC,KAO3C,GAAI,aACA,gDACA,WAAM,MAAA,IAAI,UAAU,gDAAkD,GAClE,CAAC,EAAA,YAAY,aAKrB,GAAI,aACA,+BACA,WAAM,MAAA,IAAmC,GAAI,UAAU,oBAM3D,GAAI,aAAa,+BAAgC,WAC/C,MAAO,IAAI,QAAQ,4BACf,GACA,GAAI,QAAQ,kCAOlB,GAAI,aACA,+BACA,WAAM,MAAA,IAA8B,GAAI,UAAU,oBAGtD,GAAI,aACA,0BACA,WAAM,MAAA,IAAoB,GAAI,UAAU,oBAK5C,GAAI,aAAa,4BAA6B,WAK5C,GAAM,GAAc,GAAI,QAAQ,gCAChC,MAAO,GAAc,EAAI,IAU3B,GAAI,aACA,iCACA,WACE,MAAO,IAET,SAAA,GACE,GAAI,EAAY,GAAK,IAAc,GACjC,KAAM,IAAI,OACN,+DACA,mCAAkC,EAAS,QC1LvD,AAiBO,YAAuB,GAC1B,KAAM,GAAe,GAAI,cAAa,EAAK,QAC3C,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC/B,EAAa,GAAK,KAAK,IAAI,EAAK,IAEpC,MAAO,GCtBX,AAoBO,YAAsC,GACzC,MAAO,CAAC,EAAQ,EAAQ,EAAO,EAAO,KAClC,KAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAC3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAQ,EAAO,OACf,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAC7D,GAAI,EAAe,OAAS,EAAe,SAAW,EAClD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAI5D,QAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACjC,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GACrC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GACtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAC5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,IAG5C,MAAO,CAAC,EAAQ,IClDxB,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAG,IAAM,EAAI,GCnBnE,AAoBO,YAA+B,GAClC,MAAO,CAAC,EAAQ,EAAO,KACnB,KAAM,GAAY,EAAA,KAAK,uBAAuB,EAAO,EAAO,QAC5D,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAU,GAAK,EAAG,EAAO,GAAI,GAEjC,MAAO,IC1Bf,AAmBO,KAAM,IAAW,GAAsB,AAAC,GAAO,KAAK,KAAK,ICnBhE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAiBO,YAAiB,EAAO,EAAY,EAAU,GACjD,KAAM,GAAO,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACnE,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAC/B,KAAM,GAAS,EAAI,EACnB,GAAI,GAAM,EAAM,GAChB,OAAS,GAAI,EAAG,EAAI,EAAY,EAAE,GAC9B,KAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACR,GAAM,GAGd,EAAK,GAAK,EAEd,MAAO,GC9BX,AAmBO,KAAM,IAAe,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBvF,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,EAAI,KAAK,KAAK,ICnBrE,AAkBO,YAAmB,EAAM,EAAO,EAAM,EAAO,GAChD,KAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GACrC,GAAI,GACA,KAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAElD,KAAM,GAAU,EAAA,KAAK,uBAAuB,EAAO,GACnD,OAAS,GAAI,EAAG,EAAI,EAAQ,EAAE,GAC1B,KAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,CAAC,EAAK,IAAM,EAAM,EAAM,IACvC,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEtB,MAAO,GCnCX,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBlF,AAiBO,YAAuB,EAAO,EAAQ,EAAO,EAAM,GACtD,KAAM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GACjC,EAAS,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACrE,OAAS,GAAI,EAAG,EAAI,EAAO,EAAE,GACzB,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAEhC,EAAS,GAAI,OAAM,EAAI,QAC7B,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAO,GAAK,EAAI,EAAK,IAEzB,KAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE7B,MAAO,GCjCX,AAiBO,YAAoB,EAAQ,EAAM,EAAO,GAE5C,KAAM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAwDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GAC/B,OAAS,GAAI,EAAG,EAAI,EAAO,IACvB,EAAS,IAAM,EAAM,GAEzB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACtC,EAAS,IAAM,EAAM,GAIzB,KAAM,GAAiB,GAGjB,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAAgB,GAChB,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EACxD,OAAS,GAAI,EAAG,EAAI,EAAM,GAAQ,KAE9B,GAAI,GACJ,GAAI,EAEA,EAAU,EAAO,GAAG,gBAGpB,KAAM,GAAa,GACnB,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG9C,EAAU,EAAW,KAAK,KAG9B,GAAI,EAAe,KAAa,OAC5B,EAAQ,GAAK,EAAe,QAG5B,KAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAM3B,KAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,KAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,CAAC,EAAoB,KACvC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAM9E,KAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAC7B,CACH,aAAc,EAAa,OAC3B,cACA,WC9IR,AAuBE,GAAA,IAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GCpCF,AAmBA,GAAA,IAAA,WAKE,WAAY,EAAuB,GAHnC,KAAA,YAAwB,GAItB,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,UAAU,EAAQ,SAAS,EAAQ,oBAInD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;yBAEZ,EAAS;;;MAKlC,MAAA,MClDA,AAmBA,GAAA,IAAA,WAOE,WAAY,EAAuB,GALnC,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,SAAS,EAAQ,SAAS,EAAQ,oBAIlD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;wBAEb,EAAS;;;MAKjC,MAAA,MCpDA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAqC,EACrC,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAC9B,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,KAAK,YAAc,CAAC,EAAW,GAC/B,GAAM,GAAU,IAAO,MAAS,IAAM,IAChC,EAAe,EACjB,gBACA,+CAEJ,KAAK,SAAW;;;;;kCAKc,EAAU;;;;;8BAKd,EAAU;wBAChB,EAAY;;0BAEV,EAAM;;;;;;;MAShC,MAAA,MC5DA,YAiB+B,EAAc,GAC3C,MAAO,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAG,GAAI,IAAI,gBAG/C,EAAc,GACxC,MAAI,KAAS,EACJ,CAAC,GAEH,GAAe,EAAM,eAGE,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,KAIT,OADI,GAAS,GACJ,EAAI,EAAG,EAAI,EAAM,IACxB,GAAU,EAAK,GACf,AAAI,EAAI,EAAO,GACb,IAAU,KAGd,MAAO,GCxCT,cAgCE,GAAI,GACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAU,kBACV,EAAY,KACZ,EAAY,MACZ,EAAY,KACZ,EAAY,UACZ,EAAS,cACT,EAAe,wBAMf,EAAmB;;;;;;;;;;;MAcnB,EAAmB,GACnB,EAAc;;;;;;;;;OAWd,GAAU,GACV,EAAY,YACZ,EAAY,UACZ,EAAY,UACZ,EAAY,YACZ,EAAS,eACT,EAAe,GAEf,EAAmB;;;;;;;;MASnB,EAAmB;;;;;;;;;MAUnB,EAAc;;;;;;;;OAWT,CACL,QAAO,EACP,UAAS,EACT,UAAS,EACT,UAAS,EACT,UAAS,EACT,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,iBAAgB,EAChB,YAAW,GClIf,YAyBI,EAAkB,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,SACrC,GAAM,GAAU,EAAA,KAAK,eAAe,GACpC,MAAO,GACF,IAAI,SAAC,EAAQ,GACZ,GAAM,GAAQ,OAAO,EAAO,GAAE,MAAM,EAAK,MAAM,EACzC,EAAQ,IAAM,EAAQ,OAAS,EACjC,OAAO,EAAO,EAAI,GAAE,MAAM,EAAK,MAAM,EAAO,GAAE,MAAM,EACpD,YAAY,EAAO,GAAE,MAAM,EAC/B,MAAU,GAAK,KAAK,EAAK,MAE1B,KAAK,gBA+CuB,GACjC,GAAM,GAAU,EAAA,KAAK,eAAe,GAAO,IAAI,SAAA,GAAK,MAAA,GAAE,aAEtD,MAAO;;wBAEe,EAAQ,GAAE,iBAAiB,EAAQ,GAAE;;EAKtD,GAAM,IAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EC5FpC,AAkBO,GAAA,IAAA,EAAA,aAAA,6BAkBH,EAAyB,EAAwB,EACjD,GACF,GAAM,GAA2B,GACjC,EAAW,QAAQ,SAAA,GACjB,GAAM,GAAO,EAAA,KAAK,cAAc,EAAE,UAAU,cAG5C,AAAI,EAAE,UAAU,UACd,EAAe,KACX,iBAAiB,EAAE,KAAO,GAAO,EAAI,IAAI,EAAI,IAAM,IAAE,KAEzD,GAAe,KAAK,qBAAqB,EAAE,KAAI,KAC/C,EAAe,KAAK,qBAAqB,EAAE,KAAI,QAGnD,GAAM,GAAqB,EAAe,KAAK;GAEzC,EACF,EACK,IAAI,SAAA,GAAK,MAAA,IAAwB,EAAG,EAAa,KACjD,KAAK;GACR,EAAc,EAAY,SAC1B,EAAO,KACP,EAA4B,GAA6B,GAC3D,EACA,EACA,EAAe,GAAgB,GAEnC,AAAI,EAAY,SACd,GACI,GAA+B,EAAY,aAAc,GAC7D,EAA+B,GAA8B,IAE7D,GACI,GAAyB,EAAY,aAAc,GACvD,EAA+B,GAA2B,IAG5D,AAAI,GACF,IAAgB,IAGlB,GAAM,GAAS,CACb,EAAc,EAA2B,EACzC,EAAoB,EAAuB,EAAsB,GACjE,KAAK;GACP,MAAO,GAGT,YAA8B,GAC5B,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAiB,OACrB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,WAEpB,KAAM,IAAI,OACH,EAAM,OAAM,2CAKzB,YAAoC,GAClC,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAuB,OAC3B,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,WAE1B,MAAO,IAAmB,IAIhC,YACI,EAAmB,EACnB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAM,GACV,AAAI,EACF,GAAO,GAA2B,GAElC,GAAO,GAAqB,GAG9B,GAAM,GAAU,EAAO,UAAU,aAC3B,EAAW,EAAa,aAC9B,MAAI,GAAQ,QAAU,EAAS,QAC7B,CAAI,EACF,GAAO,GAA+B,EAAQ,GAE9C,GAAO,GAAyB,EAAQ,IAGrC,EAGT,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAwB,EAAsB,OAClD,GACH,MAAO,IAAwB,EAA8B,OAC1D,GACH,MAAO,IACH,EAAsC,WAE1C,MAAO,IAAwB,EAAU,IAI/C,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAkB,EAAsB,OAC5C,GACH,MAAO,IAAkB,EAA8B,OACpD,GACH,MAAO,IACH,EAAsC,OACvC,GACH,MAAO,IACH,EAA8C,OAC/C,GACH,MAAO,IACH,EAAsD,OACvD,GACH,MAAO,IACH,EACA,WAEJ,KAAM,IAAI,OACH,EAAS,OAAM,4CAI5B,YAAsC,GACpC,MAAO;;eAEM,EAAK,UAAS;;IAK7B,YAAoC,GAClC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAuC,GACrC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAyB,GACvB,GAAM,GAAmB,EAAK,QAAO;;;;MAIjC,EAAK,UAAS;MACd,EAAK,aAAY;;;;;;;;;;;;;;;;;;;;;;;MAuBjB,EAAK,iBAAgB;MACrB,EAAK,iBAAgB;MACrB,EAAK,YAAW;;;;;;;;;;;;;;;;;;;;;;;;;MAyBhB,GAAiB;MACjB,GAAiB;MACjB,GAAiB;IAGrB,MAAO,GAGT,GAAM,IAAoB;;;;;;;;;;;;EAcpB,GAAoB;;;;;;;;EAUpB,GAAoB;;;;;;;;;EAWpB,GAAuB;;;;;;;;;;;EAa7B,cACE,MAAO;;;;IAOT,YACI,EAAiB,GACnB,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,MAAI,GAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAKjD,EAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAK9C;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;IAKlD,YACI,EAAiB,GACnB,MAAI,GAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIvC,EAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIpC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;4BACnC,EAAS,GAAE;;IAKvC,YACI,EAAiC,GACnC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAqB,KAAK,KAAK,EAAM,GAAK,GAC1C,EAAgB,EAAqB,KAAK,KAAK,EAAM,GAAK,GAEhE,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;wBAE1B,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAiC,GACnC,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EAAiB,GAWnB,OAVM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEnD,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EACF,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EAAiB,EACjB,EAAU,GACV,EAAS,UAEJ,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAkB,EAAM,EAAM,OAAS,EAAI,GAC3C,EAAU;aACD,EAAC,cAAc,EAAc;kBACxB,EAAC,MAAM,EAAc;MAC/B,EACJ,EAAS,IAAI,EAAC,KAAO,EAGvB,MAAO;UACC,EAAM,OAAM;;oCAEc,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;QAE1C,EAAO;;wBAES,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;mBAE3B,EAAM,OAAM,IAAI,EAAM;;IAKzC,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,MAAO,GAE3B,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,MAAO,GAEjC,MAAO;;kDAEyC,EAAS,GAAE;+BAC9B,EAAS,GAAE;;iCAET,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,KAAM,MAAO,GAEvC,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EAAyB,GAC3B,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,GAAI,EAAA,KAAK,YAAY,EAAO,GAC1B,MAAO;;8CAEmC,EAAe,GAAE,KACvD,EAAe,GAAE;;MAMvB,GAAM,GAAqB,KAAK,KAAK,EAAM,GAAK,GAWhD,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;;iCAE1C,EAAe,GAAE;6BACrB,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAyB,GAC3B,MAAI,GAAA,KAAK,YAAY,EAAO,GACnB;;0CAE+B,EAAS,GAAE,KAAK,EAAS,GAAE;;MAI/D,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKxC,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKrC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;wBACpB,EAAM,GAAE;4BACJ,EAAM,GAAE;;;IAMpC,YAAkC,GAChC,MAAO,SAAS,EAGlB,YAAgC,GAC9B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAO,KACb,MAAO;WACE,EAAQ;eACJ,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAA0B,GACxB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACzE,GAAI,EAAU,UAAU,UACtB,MAAO,SAAS,EAAQ,cAAc,EAAO,KAEzC,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAChB,GAAI,IAAY,GAAK,IAAY,EAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAK9B,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,KAAK,EAAM;6BAC1B,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAe,GAAE,KAAK,EAAe,GAAE;eAClC,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEzE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;UACZ,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAQ,EAAS,GACjB,EAAQ,EAAS,GAEvB,GAAI,IAAU,GAAK,IAAU,EAC3B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAI,KAAU,EACL;cACG,EAAQ;6CACuB,EAAM,cAAc,EAAK;+BACvC,EAAO;;MAIhC,IAAU,EACL;cACG,EAAQ;wCACkB,EAAM,cAAc,EAAK;+BAClC,EAAO;;MAI7B;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,aAAa,EAAM;6BAClC,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAE/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAO,KACb,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,GAC9C,MAAO;aACE,EAAQ;qDACgC,EAAO,OAAO,EAAO;;iBAEzD,EAAK,UAAS,IAAI,EAAO;;MAKxC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAe,KAAK,KAAK,EAAM,GAAK,GAE1C,MAAO;WACE,EAAQ;iCACc,EAAY,KAAK,EAAe,GAAE,KAC7D,EAAe,GAAE;eACR,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAErC,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,IAC9C,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACzB,MAAO;YACC,EAAQ;mDAC+B,EAAO,OAAO,EAAO;6BAC3C,EAAO;;IAK5B,GAAA,GAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,OACvB,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;qDAC+B,EAAM,GAAE;UACnD,GAAkB,GAAU;;MAKpC,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAS,GAAyB,GACxC,MAAI,KAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;4CACtB,EAAO;6BACtB,EAAO;;IAI9B,IAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;uCAC3B,EAAO;6BACjB,EAAO;;IAK3B;UACC,EAAQ;;wBAEM,EAAM,GAAE,YAAY,EAAM;2BACvB,EAAO,KAAK,EAAO;2BACnB,EAAO;;EAKlC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEzD,GAAI,EAAM,KAAO,GACf,GAAM,GAAgB,EAAM,MAAM,GAC5B,EAAW,CAAC,EAAG,GACf,EAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,IAAK,MAAO,OAC5B,MAAO;UACD,GAA2B,GAAa;eACnC,EAAQ;mBACJ,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAM,GAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpC,EAAgB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAO,KAAK,EAAO,KAAK,EAAa,KAAK,EAAY;eACjD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAAK,EAAM,GAC3B,EAAU,EAAM,GAEhB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,SAC9B,MAAO;UACD,GAAqB,GAAa;gBAC5B,EAAQ;mBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO;UAC1C,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAa,EAAU,UAAU,WACvC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;gBACK,EAAQ;;oDAE4B,EAAO;;4BAE/B,EAAO,OAAO,EAAO;iCAChB,EAAO;;QAKtC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;YACC,EAAQ;8CAC0B,EAAM,GAAE;;qDAED,EAAO,OAAO,EAAO;6BAC7C,EAAO;;IAKlC,GAAM,GAAS,GAAyB,GACxC,MAAO;cACK,EAAQ;;4BAEM,EAAO,YAAY,EAAO,cAAc,EAAM;+BAC3C,EAAO,KAAK,EAAO;+BACnB,EAAO;;IAKtC,YAA4B,GAe1B,OAdM,GAAQ,EAAU,UAAU,aAC5B,EAAO,EAAM,OACb,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC7C,EAAgB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC3D,EAAS,0BACT,EAAQ,OAAO,EAAa,kBAAkB,EAAY,eACrD,EAAI,EAAG,EAAI,EAAO,EAAG,IAC5B,EAAS,QAAQ,EAAC,KAAO,EACzB,GAAiB,EAAM,EAAO,EAAI,GAClC,EAAQ,IAAI,EAAC,MAAM,EAAa,MAAQ,EAE1C,GAAM,GAAO,KACb,MAAO;WACE,EAAQ,IAAI,EAAM;oBACT,EAAK;2BACE,EAAO;kCACA,EAAO;qDACY,EAAO,KAAK,EAAO;eACzD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,UACvC,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO,KAAK,EAAO;UACtD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;uBAIC,EAAO,KAAK,EAAO;;0BAEhB,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;gCAEU,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGvC,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO;oBAChC,EAAO;6BACE,EAAO,KAAK,EAAO,aAAa,EAAM;6BACtC,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,UACjD,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;iBAGL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;UAElD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;gCAGU,EAAO,KAAK,EAAO,KAAK,EAAO;;0BAErC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAKpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGxB,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAM;6BACpB,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEnE,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,SAAU,UAC3D,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;;iBAEL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAM,GAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAE3B,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;mBAGzC,EAAO;UAChB,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;0BAGlC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBACzC,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE;iBACnB,EAAM,GAAE;;;yBAGA,EAAO,OAAO,EAAO;+BACf,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;;0BAGM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAO,eAAe,EAAM;6BAC1C,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAA2B,GACzB,GAAM,GAAU,EAAU,KACpB,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAEtD,MAAI,GAAS,EACJ,UAAU,EAAO,IAEnB;0BACiB,EAAM;;iBAEf,EAAO;;;IAMxB,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAEpC,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAE7C,EAAO,GAAkB,GACzB,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,AAAI,EAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAGpC,GAAI,GAAS,sBACP,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAChD,EAAgB,IAAW,EAC3B,EAAU,EAAA,KAAK,cAAc,EAAa,cAC1C,EAAiB,IAAY,EAEnC,GAAI,IAAW,GAAK,CAAC,GAAiB,CAAC,EACrC,EAAS;;cAGA,GAAiB,CAAC,EAC3B,AAAI,IAAY,EACd,EAAS;;QAIT,EAAS;;gBAIF,EAAc,QACvB,GAAM,GAAO,EAAS,EAChB,EAAO,EAAS,EAEtB,AAAI,EAAc,QAAQ,GAAQ,IAAM,EAAc,QAAQ,GAAQ,GACpE,EAAS,8BACJ,AAAI,EAAc,QAAQ,GAAQ,GACvC,EAAS,2EAEJ,AAAI,EAAc,QAAQ,GAAQ,IACvC,GAAS,gDAIb,MAAO;WACE,EAAQ;QACX,EAAI;QACJ,EAAa;8BACS,EAAc,IAAI,EAAqB;QAC7D,EAAM;;IAKd,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAc,EAAa,SAC3B,EAAa,EAAU,UAAU,SACjC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAE1C,GAAI,CAAC,EAAU,UAAU,WAAa,IAAW,GAC7C,EAAU,UAAU,YAAc,MAClC,EAAA,KAAK,YAAY,EAAY,GAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAKpC,GAAM,GAAO,GAAkB,GACzB,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAC7C,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,MAAI,GAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAG7B;YACG,EAAQ;QACZ,EAAI;QACJ,EAAa;kBACH,EAAc,IAAI,EAAqB;;gBAKvB,GAChC,GAAI,GAAQ,EACV,MAAO,MACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QAEP,KAAM,OAAM,gBAAgB,EAAI,yBAKpC,YACI,EAAmB,GAErB,GAAM,GAA0B,KAAK,MAAM,KAAK,UAAU,IAC1D,SAAa,UAAU,aAAe,EAC/B,EAGT,YAA2B,EAAkB,GAC3C,MAAO,GAAS,IAAI,SAAA,GAAK,MAAA,GAAO,KAAI,KAAK,MCt2C3C,AAuBA,GAAA,IAAA,WAOE,WACI,EAAiB,EAAoB,EACrC,GARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,EAAA,KAAK,OACD,EAAM,OAAS,EACf,WAAM,MAAA,aACF,GAAG,OAAO,GAAG,cACb,EAAG,MAAM,IAAE,6CACnB,GAAM,GAAS,EAAM,EAAM,OAAS,GAC9B,EAAU,KAAK,KAAK,EAAS,GACnC,KAAK,YAAc,EAAM,MAAM,EAAG,IAClC,AAAI,EAAU,GACZ,KAAK,YAAY,KAAK,GAExB,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,GAAM,GAAW,KAAK,YAChB,EAAO,EAAS,OAChB,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAEjC,EACA,EACJ,GAAI,IAAY,GACd,EAAa,EAAO,EACpB,GAAM,GAAiB,GAAkB,GACzC,EAAiB;UACb,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE,QAEtB,GAAa,EACb,EAAiB;UACb,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE,IAExB,GAAM,GAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACnD,EAAY,IAAM,EAAS,EAAa,GACxC,EAAc,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IACzC,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAE/C,EAAU,IAAO,MAAS,cAAgB,WAC1C,EAAoB,EAAY,GAAK;sDACO,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM,OAE7D,EAAa;0BACG,EAAW,OAAM;uCACJ,EAAW,OAAM;uCACjB,EAAW,OAAM;qDACH,EAAW,OAAM,UAE5D,EAAgC,EAAY,GAAK;qCACtB,EAAY,OAAM;4CACX,EAAS,OAAM;iDACV,EAAS,MAAM,IAAI,OAAM;SAGtE,KAAK,SAAW;0BACM,EAAY,OAAM;iCACX,EAAS,OAAM;sCACV,EAAS,MAAM,IAAI,OAAM;;QAEvD,EAA6B;;UAE3B,EAAK;4BACa,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;4BAC5C,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;UAC9D,EAAc;yCACiB,EAAS,eAAe,EAAS;sBACpD,EAAS,eAAe,EAAS,OAAO,EAAU;;;2BAG7C,EAAU;;8BAEP,EAAU;;YAE5B,EAAiB;6BACA,EAAU;;;mBAGpB,EAAM;;;;;;;;;;;MAazB,MAAA,MCjJA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAe,GAE1C,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;0CACT,EAAa;;;;;;;;;;;;;;gCAcvB,EAAqB;oBACjC,EAAc;gDACc,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;mBACnC,EAAa;kDACkB,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;;;;;;;MAevD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAc,EAAe,GAExD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAa;;;;;;;;;;;;;;;;;gCAiBvB,EAAoB;oBAChC,EAAa;gDACe,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;;;;;;;;MAgBzD,MAAA,MC3KA,AAwBO,GAAM,IAAmB,CAC9B,KAAM,wCACN,KAAM,yCAGR,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,QAAS,QAAS,QAAS,SAK1C,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GAEnE,KAAK,SAAW;;;UAGV,EAAE;;;;;;;;;;MAYZ,MAAA,MCpDA,AAqBA,GAAM,IAAoB;;;EAKb,GAAM,gBACN,GAAM,gBACN,GAAM,gBAMN,GAAU;;;;;;;;;;EAYV,GAAM;;;;;;;;;EAYN,GAAQ,wBAER,GAAY,wBAEZ,GAAO,uBAEP,GAAa,wBAEb,GAAU,uBAEV,GAAgB,wBAEhB,GAAc,sCAEd,GAAa,sCAEb,GAAM,GAAoB;;EAG1B,GAAM,GAAoB;;EAG1B,GAAM;qBAGN,GAAU,yCAEV,GAAQ,+BAErB,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;MAUZ,MAAA,MC1GA,AAuBA,GAAM,IAAoB;;;;;EAOb,GAAU;;;;;;;;;;;;;;;;;;;;;EAuBV,GAAM;;;;;;;;;;;;;;IAef,GAAoB;;EAIX,GAAQ;;;EAKR,GAAU;;;EAKV,GAAQ;;EAIR,GAAY;;EAIZ,GAAO;;EAIP,GAAa;;EAIb,GAAU;;EAIV,GAAgB;;EAIhB,GAAc;;;;EAMd,GAAa;;;;;EAOb,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIxB,GAAA,WAQE,WACI,EAAY,EAAkB,EAC9B,GAAA,AAAA,IAAA,QAAA,GAAA,IATJ,KAAA,cAAgB,CAAC,IAAK,KAGtB,KAAA,qBAAuB,GACvB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,GAAM,GAAO,KAAK,YAAY,OAC1B,EAAyB,GAC7B,GAAI,EACF,GAAI,IAAS,GAAK,EAAA,KAAK,cAAc,KAAK,eAAiB,EACzD,EAAyB;;;;eAMzB,GAAM,GAAQ,GAAkB,GAIhC,GAHA,EAAyB;YACrB,EAAK;UAEL,IAAS,EACX,GAA0B;yCACK,KAAK,YAAY,GAAE;;;iBAKlD,GAAM,GAAW,GAAY,SAAU,GACvC,GAA0B;;iBAEnB,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;iBAExD,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;;;aASrE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;UAQF,EAAsB;;;;MAMhC,MAAA,MC9MA,AAoBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,KASf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAgBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KCzDA,AAoBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;MAiBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KC5DA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,OAAQ,QAKvB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAepB,MAAA,MCzCA,AAoBA,GAAA,IAAA,WAME,WAAY,GAJZ,KAAA,YAAwB,GAKtB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAI1C,OADM,GAAW,CAAC,YAAY,EAAQ,GAAE,+BAC/B,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAC1B,EAAS,KACL,iBAAiB,EAAQ,GAAE,KAC3B,kBAAiB,EAAC,WAAW,EAAK,QAExC,GAAM,GAAY,EAAQ,OACpB,EAAY,EAAQ,EAAQ,OAAS,GAC3C,EAAS,KAAK,sBAAsB,EAAS,WAAW,EAAS,OAEjE,KAAK,SAAW;;;;;;UAMV,EAAS,KAAK;WAAa;;MAIrC,MAAA,MCzDA,AAuBA,GAAA,IAAA,WAOE,WAAY,EAAoB,GALhC,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,GAAM,GAAQ,KAAK,YACb,EAAO,EAAM,OACb,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAC/B,EAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACzD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAW1C,OARM,GAAU,EAAS,GACnB,EAAe,EAAS,MAAM,IAC9B,EAAc,EAAS,OAEzB,EAAkB,OAAO,EAAO,MAAM,EAAQ,GAAE;;oBAEpC,EAAW,WAAW,EAAa,OAAM;WAEhD,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAK1B,GAAmB;cACX,EAAO,MAAM,EAAQ,GAAE,QAAQ,EAAO,OAAO,EAAQ,EAAI,GAAE;;kBAEvD,EAAC,IAAI,GAAgB,EAAU,EAAS,GAAM;mBAC7C,GAAgB,EAAc,EAAS,GAAM;WAG5D,GAAM,GAAY,EAAQ,OACpB,EAAQ,EAAQ,EAAQ,OAAS,GACvC,GAAmB;;gBAEP,EAAS,IAAI,GAAgB,EAAU,EAAS,GAAM;iBACrD,GAAgB,EAAc,EAAS,GAAM,MAE1D,KAAK,SAAW;uBACG,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IAAE;UAC1C,EAAe;;;;UAIf,EAAK;sCACuB,EAAM;;UAElC,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;cACrC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;;MAMtC,MAAA,MAaA,YAAyB,EAAoB,EAAiB,GAC5D,GAAM,GAAa,EAAS,QAAQ,GAC9B,EAAM,EAAS,IAAI,SAAC,EAAG,GAC3B,MAAI,KAAQ,EACA,EAAC,MAAM,EAEV,IAGX,MAAO,GAAI,OC3Hb,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAiB,EAAS,aAAe,eAE/C,KAAK,SAAW;;;;;;;;;;;;8BAYU,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;oBAIhC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAExC,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;wCAEI,EAAM,aAAa,EAAM;;;;;;;gCAOjC,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;oCAEF,EAAS,YAAW;;oBAEpC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAEjC,KAAK,SAAW;;;;;;;;;;;8BAWU,EAAS,UAAS;kCACd,EAAS,SAAQ;iCAClB,EAAW,MAAM,EAAQ;;kCAExB,EAAS,QAAO;;;;oCAId,EAAS,UAAS;mCACnB,EAAY,MAAM,EAAM;;oCAEvB,EAAS,SAAQ;;;;sCAIf,EAAS,SAAQ;qCAClB,EAAW,MAAM,EAAO;;sCAEvB,EAAS,QAAO;;;;;;;;;;;;;MAetD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAEnD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAchC,EAAW;gDACK,EAAW;;oCAEvB,EAAS,SAAQ;;;;;yBAK5B,EAAW;;kCAEF,EAAY;kDACI,EAAY;;sCAExB,EAAS,UAAS;;;;;;2BAM7B,EAAY;;oCAEH,EAAW;oDACK,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;6BAM5B,EAAW;;sCAEF,EAAS,YAAW;;;;;;;;;;MAY1D,MAAA,MCzSA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;;;;;;;wBAOI,EAAU;;;;;8BAKJ,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;;;;;;;;;MAcpD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;gCAYnB,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;;oCAGF,EAAU;8BAChB,EAAU;;;;;;;;;MAWxC,MAAA,MCzIA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAChD,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAEpC,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;;2BAGT,EAAM,aAAa,EAAM;;;;;;;gCAOpB,EAAY;qCACP,EAAc;;gCAEnB,EAAS,SAAQ;;;;kCAIf,EAAW;uCACN,EAAa;;kCAElB,EAAS,QAAO;;;;oCAId,EAAqB;;;;;;;;oBAQrC,EAAc;;;;;;;;;;;;;;;;;;;kBAmBhB,KAA4B,GAAC;;oBAE3B,EAAc;;0CAEQ,EAAqB;mCAC5B,EAAqB;;;kCAGtB,EAAqB;mCACpB,EAAqB;;;yBAG/B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;;;;;yBAK5B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;gCACrB,EAAqB;;;;;;;;;;UAU3C,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAEtD,KAAK,SAAW;oCACgB,EAAW,KAAK,EAAY,KACxD,EAAW;iCACc,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;gCAgBhC,EAAW;qCACN,EAAa;;gCAElB,EAAS,QAAO;;;;kCAId,EAAY;uCACP,EAAc;;kCAEnB,EAAS,SAAQ;;;;oCAIf,EAAW;yCACN,EAAa;;oCAElB,EAAS,QAAO;;;;sCAId,EAAqB;;;;;;;;;;;;;;;;;oBAiBvC,KAA4B,GAAC;;4CAEL,EAAqB;qCAC5B,EAAqB;2BAC/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;;;2BAG/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;qCACrB,EAAqB;;;;;;;;;MAW1D,MAAA,MC3TA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAa,EAAS,YAAc,EAAS,WAE/C,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;wBAO3B,EAAU;4BACN,EAAU;;;;;;;;;gCASN,EAAY;qCACP,EAAc;;gCAEnB,EAAQ;;;;kCAIN,EAAW;uCACN,EAAa;;kCAElB,EAAQ;;;;;;;;;;;UAWhC,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MCrHA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAR/B,KAAA,cAAgB,CAAC,IAAK,KACtB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,SAgB5B,OAdM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAEjB,EAAW,gCAEN,EAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY;wBACI,EAAC,IAAI,EAAI,EAAC;mBACf,EAAC,IAAI,EAAC;mBACN,EAAC,IAAI,EAAC,eAYrB,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAS,EAAG,EAAS,EAAc,KAC1C,GAAM,GAAI,EAAS,EAOnB,GALA,GAAY;4BACQ,EAAI,EAAc;4BAClB,EAAI,EAAa;UAGjC,IAAgB,GAClB,GAAI,EAAI,GAEN,CAAI,EAAU,IAAM,EAUlB,GAAY;;qCAEW,EAAQ,mCAC3B,EAAQ;2BACC,EAAC,IAAI,EAAC;;;;uCAIM,EAAQ;6BAClB,EAAC,IAAI,EAAC;;;2BAGR,EAAC,IAAI,EAAC;;;;qCAII,EAAQ,mCAC3B,EAAQ;;;;;uCAKa,EAAQ;;;;sBAIzB,EAAC,IAAI,EAAC,+BAA+B,EAAC,IAAI,EAAC;;sBAE3C,EAAC,IAAI,EAAC,wBAAwB,EAAC,IAAI,EAAC;;gBAK5C,GAAY;qCACW,EAAQ,uBAAuB,EAAQ;2BACjD,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;oBAGb,EAAC,IAAI,EAAC,aAAa,EAAC,IAAI,EAAC;gBAI7B,EAAI,EAAI,IAOV,GAAM,GAAkB,EAAU,IAAM,EACpC,EAAA,KAAK,kBAAkB,GACvB,EAEJ,AAAK,EAAgB,IAAM,GAAK,EAAU,IAAM,GAC3C,EAAgB,IAAM,GAAK,EAAU,IAAM,EAC9C,IAAY;oCACQ,EAAU,EAAC,MAAM,EAAe;;uCAE7B,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;kBAMvB,AAAI,EAAgB,GAClB,IAAY;;yCAEW,EAAQ;oDACG,EAAQ;+BAC7B,EAAC,IAAI,EAAC;;+BAEN,EAAC,IAAI,EAAC;;qBAKrB,GAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;mBAG5C,GAAY;oCACQ,EAAe;;uCAEZ,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;;sBAGjB,EAAC,IAAI,GAAI,GAAC,aAAa,EAAC,IAAI,GAAI,GAAC;uBAM7C,AAAI,GAAI,GACN,IAAY;mCACW,EAAQ;cAS/B,AAAI,EAAU,IAAM,EAClB,IAAY;sCACY,EAAW;iDACA,EAAQ;2BAC9B,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;6CAGY,EAAQ;2BAC1B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;;wCAEY,EAAW;mDACA,EAAQ;;;sBAGrC,EAAC,IAAI,GAAI,GAAC,kBAAkB,EAAC,IAAI,GAAI,GAAC;oBAI9C,IAAY;qCACW,EAAQ;2BAClB,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;kCAGC,EAAW;iDACI,EAAQ;2BAC9B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;oBAKhD,GAAY,KAIhB,AAAI,EAAI,GACN,IAAY;0BACI,EAAC,IAAI,EAAC,WAAW,EAAC,KAAK,EAAC;gBAClC,EAAC,IAAI,EAAC,kBAAkB,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,EAAC;YAGzD,AAAI,EAAI,EAAI,GACV,IAAY;4BACI,EAAC,IAAI,GAAI,GAAC,WAAW,EAAC,KAAK,GAAI,GAAC;kBAC1C,EAAC,IAAI,GAAI,GAAC;8BACE,EAAC,IAAI,GAAI,GAAC,eAAe,EAAC,IAAI,GAAI,GAAC,UAM7D,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY,gBAAgB,EAAC,IAAI,EAAC,QAAQ,EAAC,IAAI,EAAC,IAIpD,GAAI,GAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;UAezC,EAAQ;;;UAGR,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MC3UA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAA4B,EAC5B,GAPJ,KAAA,cAAgB,CAAC,QAAS,QAAS,UACnC,KAAA,YAAwB,GAOf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAChC,EAAA,EAAA,GACA,EAAA,EAAA,GAAY,EAAA,EAAA,GACnB,KAAK,YAAc,CAAC,EAAU,EAAY,EAAW,GACrD,GAAM,GAAW,IAAW,WAAa,EAAI,EAEvC,EAAA,CAAA,EAAA,EAAA,KAAA,EAAA,EAAA,MAAC,EAAA,EAAA,GAAkB,EAAA,EAAA,GAGnB,EAAA,EAAA,oHAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAa,EAAA,EAAA,GAW3B,EAAA,EAAA,kHAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAe/B,KAAK,SAAW;yCACqB,EAAW;wCACZ,EAAU;;;;;;;;;;;;;;;;iCAgBjB,EAAK;;;;+BAIP,EAAW;8BACZ,EAAU;;uBAEjB,EAAG;mCACS,EAAgB;4BACvB,EAAkB;;;uBAGvB,EAAG;mCACS,EAAe;4BACtB,EAAkB;;;;;aAKjC,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;MA0BrB,MAAA,MC1GA,GAAA,WAQE,WAAY,EAAiB,EAAoB,GAPjD,KAAA,cAAgB,CAAC,KAQf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAM,OACb,EAAM,EAAY,MAAQ,QAAQ,GAAU,EAAM,UAAS,IAC3D,EAAS,EAAM,EAAM,OAAS,GAChC,EAAY,GACZ,EAAY,GAIhB,AAAI,EACF,GAAY,EAAU,UAAU,GAAS,GAAM,WAC/C,EAAY,EAAU,UAAY,WAElC,GAAY,EAAU,gBAAgB,EAAW,cACjD,EAAa,EAAU,aAAe,cAGxC,KAAK,SAAW;;;UAGV,GAAkB,GAAK;oBACb,GAAc,EAAM,UAAS;sBAC3B,EAAG;;cAEX,EAAS;sBACD,EAAS;YACnB,GAAc,EAAM,UAAS;wBACjB,GAAU,EAAM,UAAS;;;;MAO/C,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,OAAS,MAChB,GAAK,MAAQ,EAAM,mBAAmB,EAAc,UAEtD,EAAM,GAAG,UAAU,EAAK,MAAO,KAGrC,KAEA,YAAmB,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,KACpB,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,KAC/B,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,OAAO,EAAI,KAE/C,KAAM,OAAM,2BAA2B,EAAI,yBAI/C,YAAuB,EAAc,GACnC,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KAEd,KAAM,OAAM,2BAA2B,EAAI,yBCjG/C,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAEjB,KAAA,aAAe,GACf,KAAA,aAAe,GAEf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAuB,EAAmB,GAP9C,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAOtB,KAAK,YAAc,EACnB,KAAK,UAAY,EACjB,KAAK,WAAa,EAClB,KAAK,SAAW;;;;gBAIJ,KAAK,uBAAsB;gBAC3B,KAAK,sBAAqB;gBAC1B,KAAK,sBAAqB;;uBAEnB,EAAS;+BACD,EAAS;uBACjB,EAAS;+BACD,EAAS;mCACL,EAAS;UAClC,KAAK,qBAAoB;;;uBAGZ,KAAK,yBAAwB;;;IAM1C,SAAA,UAAA,qBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,mBAAR,WACE,MAAI,MAAK,aAAe,OACf,KAAK,YAAY,GAEjB,KAAK,YAAY,IAIpB,EAAA,UAAA,uBAAR,WACE,MAAI,MAAK,aAAe,OACf,4BAEA,6BAGb,KC5FA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,CAAC,EAAM,GAC1B,KAAK,SAAW;;;;;;MAQpB,MAAA,MClCA,AAsBA,GAAA,IAAA,WAME,WAAY,GALZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;UAIlB,EAAK,OAAM;;MAIrB,MAAA,MCxCA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;;UAKlB,EAAK,OAAM;;MAIrB,MAAA,MC3CA,AAqBA,GAAA,IAAA,WAKE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IANJ,KAAA,cAAgB,CAAC,KAOf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAS,SACb,AAAI,GACF,GAAS,8BAGX,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;8BAUrB,EAAK;kCACD,EAAK;iDACU,EAAK,OAAO,EAAM;wBAC3C,EAAK,UAAS;;;;;;;;;;;;;;UAc5B,EAAK,OAAM,WAAW,EAAM;;MAItC,MAAA,MCtEA,AAqCA,GAAA,IAAA,WAOE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAW,GACX,EAAS,SACb,AAAI,GACF,GAAS,8BAGX,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAC1B,GAAM,GAAU,EAAM,EAAI,EAE1B,GAAY;;gCAEY,EAAG,MAAM,EAAY,GAAE;gCACvB,EAAG;kCACD,EAAG,MAAM,EAAY,GAAE;kCACvB,EAAG;;;;;;;gCAOL,EAAK;oCACD,EAAK;kDACS,EAAK,OAAO,EAAM;yBAC3C,EAAK,UAAS;;;yBAGd,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;;;UAQ5B,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;;UAWzC,EAAQ;;UAER,EAAK,OAAM,MAAM,EAAM;;MAIjC,MAAA,MC/GA,AAmBO,GAAM,IAAc,CACzB,KAAM,oCACN,KAAM,qCAGR,GAAA,WAKE,WAAY,EAAY,EAA8B,GAJtD,KAAA,cAAgB,CAAC,OAAQ,QAKvB,GAAM,GAAW,EAAW,GAC5B,KAAK,YAAc,EAEnB,GAAM,GACF,EAAU,SAAS,KAAK,GAAO,UAAU,KAAK,GAC5C,EAAoB,EAAa,EAAQ,KAAO,MAEtD,KAAK,SAAW;yCACqB,EAAyB;;;UAGxD,EAAE;;;;kDAIsC,EAAQ;;;;;;8BAM5B,EAAQ;;;;;;;;;yDASmB,EAAiB;;;;;;;;;;MAY1E,MAAA,MCxEA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAiB,GAL7B,KAAA,YAAwB,GAMtB,KAAK,cAAgB,CAAC,KACtB,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;;MASlB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,UAEhE,EAAM,GAAG,UAAU,EAAK,SAAU,KAGxC,KChDA,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,EAAuB,GALrD,KAAA,cAAgB,CAAC,IAAK,WAMpB,GAAM,GAAwB,EAAO,QACrC,EAAY,GAAQ,EACpB,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,EAAQ,GAE7C,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,EAAkB,GACzC,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,mBAAmB,EAAI,yBAErC,GAAI,IAAS,EACX,MAAO,yBAMT,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAElD,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,IAAM,EACR,EAAa,KAAK,kBAAkB,EAAc,GAAE,MAEpD,EAAa,KAAK,GAAG,EAAc,IAGvC,MAAO,GAAa,OC3CtB,GAAA,IAAA,WAIE,WACY,EAA0B,EAAmB,GAA7C,KAAA,SAAA,EAA0B,KAAA,QAAA,EAJtC,KAAA,cAAgB,CAAC,IAAK,WAKpB,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAChC,EAAe,KAAK,SAAW,EAAI,aAAe,UACxD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,KAAK,QAAO;;YAElD,EAAK;;gCAEe,KAAK,SAAQ;;sCAEP,EAAY;;;;QAMlD,MAAA,MC1CA,YAwBmC,GACjC,GAAM,GAAO,KACP,EAAwB,EAAK,QAAO;;MAEtC,EAAK,UAAS;MACd,EAAK,UAAS;MACd,EAAK,UAAS;;;;;OAMlB,MAAO,IAA8B,EAAI,eAGR,GAEjC,GAAM,GAAc,GAAI,cACpB,CAAC,GAAI,EAAG,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,GAAI,EAAG,EAAG,IAClE,MAAO,IAAoC,EAAI,eAGf,GAEhC,GAAM,GAAwB,GAAI,aAAY,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,IAC9D,MAAO,IAAmC,EAAI,GAGhD,YACI,EAA2B,EAAe,EAC1C,EAAwB,EACxB,GACF,GAA+B,EAAO,GACtC,GAAM,GAAU,GAAyB,GAEnC,EAAQ,EAAG,WACjB,UAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAO,KACxD,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAO,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAC5C,EAAa,QACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,QACzD,cAIL,GACF,MAAO,GAAc,gCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAG,mBAIvC,GACF,MAAO,GAAc,oCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAc,kCAIlD,GACF,MAAO,GAAc,kCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAG,2BAIL,GACF,MAAO,GAAc,sCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EAAQ,GAAwC,GAC3D,EAAG,KAAM,EAAG,mBAId,GACF,MAAO,GAAc,0CAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAc,kCAIhB,EAA2B,EAC3B,GACF,GAAM,GAAY,EACZ,EAAW,EAAI,EACf,EAAU,EAAI,EAAM,EAAI,EAC9B,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KAC7C,GAAM,GAAU,GACZ,EAAI,EAAS,eAAgB,EAAc,EAAG,EAAQ,GAC1D,MAAO,IACH,GACI,EAAI,EAAS,KAAM,EAAc,EAAG,EAAQ,eAIlD,EAA2B,EAAuB,EAClD,EAAgB,EAAkB,GACpC,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAEhE,GAAI,GAA2B,EAAuB,EACtD,AAAI,YAAgB,YAClB,GAAgB,GAAI,YAAW,EAAQ,EAAS,GAChD,EAAgB,EAAG,cACnB,EAAiB,EAAG,MAEpB,GAAgB,GAAI,cAAa,EAAQ,EAAS,GAClD,EAAgB,EAAG,MACnB,EAAiB,EAAc,2BAGjC,EAAc,IAAI,GAElB,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAAG,KACvD,EAAe,KAEvB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA2B,EAC3B,GAEF,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAChE,AAAK,EAAqB,eAAgB,YACxC,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAO,MAAO,EAAO,OAAQ,EAAG,EAAG,KAC9D,EAAG,cAAgB,EAAqB,QAEhD,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAG,KAAM,EAAG,cACvC,KAIV,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA6B,EAAc,EAC3C,GAEF,GAAM,GAAS,EAAI,eACnB,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,KAGrD,GAAM,GAAgB,EAChB,EAAiB,EACjB,EAAkB,EAAgB,EAAiB,EAAO,EAEhE,UACI,EACA,WAAM,MAAA,GAAI,WACN,EAAI,kBAAmB,EAAiB,EAAI,eAIpD,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAG,EAAG,EAAS,EAAM,EAAI,KAAM,EAAI,MAAO,KAExE,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,QAE9C,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAM,EAEN,EAAiB,GAAI,cAAa,GAExC,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAGJ,EAAc,EACd,EAAiB,GAAI,YACvB,GAA4C,EAAO,EAAS,IAEhE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAG,EAAG,EAAc,sBAAuB,EAAG,cACpD,KAID,GAAI,cAAa,EAAe,oBAIrC,EAA2B,EAAqB,EAAe,EAC/D,EAAc,EAAsB,EACpC,GACF,GAAM,GAAM,EAEN,EACF,GAAI,cAAa,GACb,EAAc,IAEtB,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAa,GAAI,cAAa,EAAe,EAAe,GAClE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAc,EAAc,EAAG,KAAM,EAAG,MAAO,KAEtD,owBCxTT,kBAgDE,WAAY,GANZ,KAAA,cAAmC,KACnC,KAAA,QAA6B,KACrB,KAAA,SAAW,GAqOX,KAAA,oBAAsB,GAsPtB,KAAA,YAA0B,GAtdhC,GAAM,GAAY,EAAA,MAAM,UAAU,iBAClC,AAAI,GAAM,KACR,MAAK,GAAK,EACV,GAAgB,EAAW,IAE3B,KAAK,GAAK,GAAgB,GAG5B,GAAI,GAAqB,2BACnB,EAA0B,8BAChC,GAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAM,GAAgB,oBAChB,EAAqB,yBAI3B,GAFA,KAAK,sBACD,GAA+B,KAAK,GAAI,GACxC,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,sHAKN,GADA,KAAK,0BAA4B,KAAK,GAAG,aAAa,GAClD,GAAwB,KAAK,GAAI,GACnC,KAAK,8BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,uIAIN,EAAqB,yBACjB,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,KAAK,GAAG,aAAa,WAChB,GAAwB,KAAK,GAAI,GAC1C,KAAK,8BACD,KAAK,GAAG,aAAa,OAEzB,MAAM,IAAI,OAAM,uDAIpB,KAAK,aAAe,GAA8B,KAAK,IACvD,KAAK,YAAc,GAA6B,KAAK,IACrD,KAAK,YAAc,GAA6B,KAAK,IAErD,KAAK,cACD,GAA0B,KAAK,GAAI,KAAK,2BAG9C,cAAA,eAAY,EAAA,UAAA,QAAK,KAAjB,WACE,MAAO,GAAA,MAAM,QAAQ,0CAGhB,EAAA,UAAA,QAAP,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAEF,AAAI,KAAK,SAAW,MAClB,QAAQ,KACJ,wKAIN,AAAI,KAAK,eAAiB,MACxB,QAAQ,KACJ,sMAKN,GAAM,GAAK,KAAK,GAChB,GAAwB,EAAI,WAAM,MAAA,GAAG,WACrC,GAAwB,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QACrE,GAAwB,EAAI,WAAM,MAAA,GAAG,kBAAkB,EAAK,eAC5D,GAAwB,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,QACjE,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,QACrD,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAK,eACvD,KAAK,SAAW,IAGX,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,yBAAP,SACI,EACA,GACF,KAAK,kBACL,GAAoC,KAAK,GAAI,EAAS,IAGjD,EAAA,UAAA,2BAAP,SACI,EAAuB,EAAe,EAAgB,GACxD,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,EAAO,EAAQ,EAAM,KAAK,gBAG3C,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,0BAAP,SAAiC,EAAc,GAE7C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,oBAAP,SAA2B,GAA3B,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,KAAK,gBAAkB,GACzB,IAA6C,KAAK,GAAI,KAAK,aAC3D,KAAK,cAAgB,MAEvB,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAGxD,EAAA,UAAA,gDAAP,SACI,EAAuB,EAAc,GADzC,GAAA,GAAA,KAEE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAM,EAAS,EAAK,kBAGhC,EAAA,UAAA,+BAAP,SACI,EAAqB,EAAe,EAAc,EAClD,EAAsB,GACxB,MAAO,IACH,KAAK,GAAI,EAAQ,EAAO,EAAM,EAAS,EAAc,EACrD,KAAK,gBAGJ,EAAA,UAAA,gCAAP,SAAuC,EAAqB,GAE1D,MAAO,IAA2C,KAAK,GAAI,EAAQ,IAG9D,EAAA,UAAA,wBAAP,SACI,EAAuB,EAAc,GACvC,KAAK,yBAAyB,GAC9B,GAAM,GAAS,GACX,KAAK,GAA8B,EAAM,EAAS,KAAK,eAC3D,YAAK,6BACE,GAGF,EAAA,UAAA,sBAAP,WACE,GAAM,GAAe,KAAK,YAAY,KAAK,IAC3C,MAAO,MAAK,UAAU,IAGhB,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACM,EACA,EAEJ,GAAI,EAAA,MAAM,QAAQ,4BAChB,GAAM,GAAM,EAEN,EAAO,EAAI,UAAU,EAAI,2BAA4B,GAC3D,EAAG,QAEH,EAAgB,WACd,GAAM,GAAS,EAAI,eAAe,EAAM,EAAG,GAC3C,MAAO,KAAW,EAAI,kBAClB,IAAW,EAAI,qBAGrB,EAAQ,MACH,AACH,GAAA,MAAM,UAAU,gDAAkD,EACpE,GAAQ,KAAK,aACb,KAAK,WACL,EAAgB,WAAM,MAAA,GAAK,iBACvB,EACA,EAAA,MAAM,UAAU,mDAMpB,EAAgB,WAAM,MAAA,IAGxB,MAAO,CAAC,MAAK,EAAE,cAAa,IAGvB,EAAA,UAAA,gCAAP,SACI,EAAuB,EACvB,GAFJ,GAAA,GAAA,KAGE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAc,MAK1B,EAAA,UAAA,cAAP,SAAqB,GACnB,KAAK,kBACL,GAAM,GAAK,KAAK,GACV,EACF,GAAgC,EAAI,GAClC,EAA4B,GAA8B,GAC1D,EAAwB,GAAyB,GACvD,UAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAuB,EAAI,GAC3B,AAAI,KAAK,OACP,GAA2B,EAAI,GAEjC,AAAK,KAAK,qBACR,MAAK,WAAW,GAChB,KAAK,oBAAsB,GACvB,EAAI,KAAK,QAAS,KAAK,eAEtB,GAGF,EAAA,UAAA,cAAP,SAAqB,GAArB,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,IAAY,KAAK,SACnB,MAAK,QAAU,MAEjB,AAAI,GAAW,MACb,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAI1D,EAAA,UAAA,WAAP,SAAkB,GAAlB,GAAA,GAAA,KACE,KAAK,kBACL,KAAK,QAAU,EACf,AAAK,KAAK,SAAW,MAAS,KAAK,OACjC,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,WAAW,MAGrD,EAAA,UAAA,mBAAP,SACI,EAAuB,EACvB,GAEF,MAFE,AAAA,KAAA,QAAA,GAAA,IACF,KAAK,kBACD,EACK,GACH,KAAK,GAAI,EAAS,GAEf,GACH,KAAK,GAAI,EAAS,IAInB,EAAA,UAAA,qBAAP,SAA4B,EAAuB,GAAnD,GAAA,GAAA,KAEE,YAAK,kBACE,GACH,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,kBAAkB,EAAS,MAGjD,EAAA,UAAA,0BAAP,SAAiC,EAAuB,GAEtD,YAAK,kBACE,KAAK,GAAG,mBAAmB,EAAS,IAGtC,EAAA,UAAA,sBAAP,SACI,EAAkC,EAClC,GACF,KAAK,kBACL,KAAK,mBACL,GACI,KAAK,GAAI,EAAoB,EAAiB,IAG7C,EAAA,UAAA,uBAAP,SACI,EAAmC,EAAc,GACnD,KAAK,6BAA6B,EAAqB,EAAS,IAG3D,EAAA,UAAA,6BAAP,SACI,EAAyC,EAAc,GACzD,KAAK,kBACC,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,KAAK,6BAA6B,EAA2B,EAAO,IAG/D,EAAA,UAAA,2BAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAK,iCACD,EAAa,EAAU,EAAY,IAGlC,EAAA,UAAA,iCAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAM,IAAI,OAAM,sDAGX,EAAA,UAAA,cAAP,WACE,AAAI,KAAK,SAAW,MAClB,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAA+B,KAAK,KAG/B,EAAA,UAAA,eAAP,WACE,KAAK,kBACL,KAAK,mBACL,GAAM,GAAK,KAAK,GAChB,AAAI,KAAK,OACP,KAAK,gBAEP,GACI,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,UAAW,EAAG,EAAG,eAAgB,MAG7D,EAAA,UAAA,+BAAP,WAAA,GAAA,GAAA,KACE,KAAK,kBACL,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,YAGzC,EAAA,UAAA,uBAAR,WAEE,MAAI,MAAK,6BAA+B,MACtC,MAAK,4BACD,GACI,KAAK,GACL,EAAA,MAAM,UACF,kDAAoD,EACpD,kCACA,6BAIP,KAAK,6BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGd,EAAA,UAAA,WAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EAAQ,EAAI,cAClB,SAAI,WAAW,EAAI,iBAAkB,GAC9B,EAET,GAAM,GAAM,KAAK,+BACX,EAAQ,EAAI,iBAClB,SAAI,cAAc,EAAI,iBAAkB,GACjC,GAGT,EAAA,UAAA,SAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BACjB,EAAI,SAAS,EAAI,kBACjB,OAEF,GAAM,GAAM,KAAK,+BACjB,EAAI,YAAY,EAAI,mBAGT,EAAA,UAAA,uBAAb,SAAoC,wGAClC,MAAA,CAAA,EAAM,EAAA,KAAK,YACP,WAAM,MAAA,GAAK,UAGP,EAAK,iBACD,EACA,EAAA,MAAM,UACF,2DAPhB,SAAA,OAQA,CAAA,EAAO,KAAK,aACR,EAAO,EAAA,MAAM,UAAU,wDAGrB,EAAA,UAAA,aAAR,SAAqB,EAAmB,GACtC,GAAI,IAAsB,EACxB,MAAO,MAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GAEX,EAAmB,EAAI,kBAAkB,EAAO,EAAI,cAE1D,MAAO,GAAmB,SAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,kBAErC,MAAO,GAAmB,MAItB,EAAA,UAAA,iBAAR,SAAyB,EAAmB,GAE1C,GAAI,IAAsB,EACxB,MAAO,GAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,wBACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,cAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,4BACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,WAI9B,EAAA,UAAA,UAAA,SAAU,GAAV,GAAA,GAAA,KACE,MAAO,IAAI,SAAc,SAAA,GACvB,EAAK,cAAc,WAAM,MAAA,GAAa,iBAAiB,WAAM,MAAA,UAMjE,EAAA,UAAA,UAAA,WAGE,OADM,GAAQ,GAAqB,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,YACtD,EAAI,EAAG,GAAK,EAAO,EAAE,GACrB,GAAA,GAAA,KAAA,YAAA,GAAA,UACP,IAEF,KAAK,YAAc,KAAK,YAAY,MAAM,EAAQ,IAG5C,EAAA,UAAA,cAAR,SAAsB,EAAyB,GAA/C,GAAA,GAAA,KAEE,GADA,KAAK,YAAY,KAAK,CAAC,SAAQ,EAAE,UAAS,IACtC,KAAK,YAAY,OAAS,EAE5B,OAGF,EAAA,KAAK,YAAY,WACf,SAAK,YAEE,EAAK,YAAY,SAAW,KAI/B,EAAA,UAAA,yBAAR,SAAiC,GAC/B,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,KAAK,aAC3B,AAAI,KAAK,OACP,GAA+B,KAAK,KAIhC,EAAA,UAAA,2BAAR,WACE,AAAI,KAAK,eAAiB,KACxB,IACI,KAAK,GAAI,KAAK,cAAe,KAAK,aACtC,AAAI,KAAK,OACP,GAA+B,KAAK,KAGtC,GAA6C,KAAK,GAAI,KAAK,cAIvD,EAAA,UAAA,qBAAR,SACI,EACA,GACF,KAAK,yBAAyB,GAC9B,GAAM,GAAS,IACf,YAAK,6BAEE,GAGD,EAAA,UAAA,6BAAR,SACI,EAA8C,EAC9C,GACF,KAAK,kBACL,GAAM,GAAK,KAAK,GAChB,GACI,EAAI,EAAgC,KAAK,aAC7C,AAAI,KAAK,OACP,GAA+B,GAEjC,KAAK,cAAgB,EACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAO,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAO,MAGpD,EAAA,UAAA,iCAAR,SACI,EAAW,EAAW,EAAe,GADzC,GAAA,GAAA,KAEE,KAAK,kBACL,GACI,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,QAAQ,EAAG,EAAG,EAAO,MAG1C,EAAA,UAAA,gBAAR,WACE,GAAI,KAAK,SACP,KAAM,IAAI,OAAM,4CAIZ,EAAA,UAAA,iBAAR,WACE,GAAI,KAAK,SAAW,KAClB,KAAM,IAAI,OAAM,qCAGtB,iBAaqC,GAEnC,OADI,GAAI,EACD,EAAI,EAAI,OAAQ,EAAE,GACvB,GAAM,GAAS,EAAI,KACnB,GAAI,CAAC,EACH,MAGJ,MAAO,GAAI,ECtnBb,YAgEI,EAAqB,EAAuB,EAC5C,GACF,GAAM,GAAW,EAAQ,SACnB,EAA0B,EAAO,IAAI,SAAC,EAAO,GACjD,GAAM,GAAuB,CAC3B,aAAc,EAAM,MACpB,SAAU,EAAM,UAAY,KAAO,EAAM,QAAQ,SACjD,UAAW,EAAM,UACjB,SAAU,EAAM,UAAY,GAAQ,EAAM,QAAQ,SAClD,WAAY,MAEd,MAAI,GAAM,SAAW,MAAQ,EAAM,QAAQ,OAAS,MAChD,EAAM,QAAQ,MAAM,WAAa,GACnC,GAAU,WAAa,EAAM,QAAQ,MAAM,YAEtC,CAAC,KAAM,EAAQ,cAAc,GAAI,UAAS,KAE7C,EAAe,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,YACrC,EAA0B,CAC9B,aAAc,EAAO,MACrB,SAAU,EAAO,QAAQ,SACzB,UAAW,GACX,SAAU,EAAO,QAAQ,SACzB,WAAY,MAER,EAAS,GACX,EAAY,EAAc,EAAU,EAAQ,cAE1C,EAAe,EAAM,cAAc,GAGrC,EAA+B,KAC7B,EAAS,EAAM,mBAAmB,EAAc,MAAO,IAC7D,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAS,EAAM,mBAAmB,EAAc,WAAY,KAK9D,OADM,GAA2D,GACxD,EAAI,EAAG,EAAI,EAAQ,cAAc,OAAQ,KAChD,GAAM,GAAU,EAAQ,cAAc,GAChC,EAAc,GACpB,EAAiB,GACb,EAAM,mBAAmB,EAAc,EAAS,GACpD,EAAiB,SAAS,GACtB,EAAM,mBAAmB,EAAc,SAAS,EAAW,GAGjE,MAAO,CACL,QAAO,EACP,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,aAAY,EACZ,aAAY,EACZ,OAAM,EACN,OAAM,GAIV,YACI,EAAyB,GAC3B,GAAI,EAAW,SAAW,EAAO,OAC/B,KAAM,OACF,4BAA4B,EAAW,OAAM,gBAC7C,sBAAqB,EAAO,OAAM,YAGxC,EAAW,QAAQ,SAAC,EAAG,GACrB,GAAM,GAAS,EAAE,aACX,EAAQ,EAAO,GACf,EAAS,EAAM,MAErB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAQ,GAC5B,KAAM,OACF,kDACA,6BAA4B,EAAM,QAAQ,EAAM,gBAGtD,GAAI,EAAE,WAAa,EAAM,UACvB,OAGF,GAAM,GAAY,EAAE,SACd,EAAY,EAAM,UAAY,KAAO,EAAM,QAAQ,SACzD,GAAI,CAAC,EAAA,KAAK,YAAY,EAAW,GAC/B,KAAM,OACF,6DACA,yBAAwB,EAAS,QAAQ,EAAS,8BAMxD,EAAqB,EAAqB,EAC1C,EACA,GAEF,GAAyB,EAAO,aAAc,GAC9C,GAAyB,CAAC,EAAO,cAAe,CAAC,IAEjD,GAAM,GAAS,EAAO,QAAQ,QACxB,EAAc,EAAO,QAAQ,SACnC,AAAI,EAAO,QAAQ,SACjB,EAAM,6BAA6B,EAAQ,EAAY,GAAI,EAAY,IAEvE,EAAM,uBAAuB,EAAQ,EAAY,GAAI,EAAY,IAEnE,EAAM,WAAW,EAAO,cAGxB,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,CAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,WAGtC,AAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,KAIpC,EAAO,QAAQ,SAAC,EAAO,GACrB,GAAM,GAAU,EAAO,QAAQ,cAAc,GACvC,EAAS,EAAO,iBAAiB,GACjC,EAAe,EAAO,iBAAiB,SAAS,GAEtD,GAAI,GAAU,KAEZ,OAGF,GAAI,EAAM,WAER,GAAI,EAAA,KAAK,cAAc,EAAM,OAAS,EACpC,EAAM,GAAG,UAAU,EAAQ,EAAM,cAAc,SAE/C,GAAI,GAAO,EAAM,cACjB,AAAM,YAAgB,eACpB,GAAO,GAAI,cAAa,IAE1B,EAAM,GAAG,WAAW,EAAQ,GAE9B,OAIF,AAAI,EAAM,QAAQ,OAAS,MAAQ,GAAgB,MACjD,EAAM,GAAG,UAAU,EAAc,EAAM,QAAQ,MAAM,YAGvD,EAAM,sBAAsB,EAAM,QAAQ,QAAS,EAAQ,KAG7D,AAAI,GAAe,MACjB,EAAY,EAAO,EAAO,cAE5B,EAAM,6BAIJ,EAAuB,EAAsB,GAC/C,GAAI,GAAY,GAChB,EAAO,OAAO,GAAQ,QAAQ,SAAA,GAC5B,GAAM,GAAY,EAAE,SAAW,MAAQ,EAAE,QAAQ,OAAS,MACtD,EAAE,QAAQ,MAAM,WAAa,EAC3B,EAAW,EAAE,UAAY,UAAY,EAAE,QAAQ,SACrD,GAAgB,EAAE,MAAK,IAAI,EAAQ,IAAI,IAEzC,GAAM,GAAc,EAAQ,SACxB,EAAM,EAAQ,YAAY,KAE9B,UAAO,IAAM,EAAY,IAAM,EACxB,EC5OT,AAqBA,GAAA,IAAA,WAOE,WACI,EAAuB,EACvB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAsBnB,OAnBE,GAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,cACA,EAAA,EAAA,eACA,EAAA,EAAA,WAEK,EAAA,EAAA,KAAM,EAAA,EAAA,IACP,EAAmB,EAAa,EAChC,EAAO,KACP,EAAiB,IAAe,eAChC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAEhC,EAAW,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,GAAY;gCACY,EAAG;yBACV,EAAG;;4BAEA,EAAY,GAAE,aAAa,EAAY,GAAE;0CAC3B,EAAQ,QAAQ,EAAY,MAC1D,EAAG;6BACc,EAAc,aAAa,EAAgB;;sBAElD,EAAW,GAAO;;qDAEa,EAAQ,QACjD,EAAW,OAAO,EAAI;+BACH,EAAa,2BAChC,EAAgB,QAAQ,EAAU;;wBAEtB,EAAW,GAAO;;2CAEC,EAAU;;sBAE/B,EAAc;;2BAET,GAAM,EAAI,GAAG;;;;;2BAKb,GAAM,EAAI,GAAG;;;;;;;UAWpC,KAAK,SAAW;;;;;;;;;UASV,EAAQ;;UAER,EAAK,OAAM;;MAIrB,MAAA,MC5GA,AAmBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAgB,EAAc,EAChD,GANJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMtB,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;;wBASI,EAAG,UAAU,EAAG;;oCAEJ,EAAI;;;;;0BAKd,EAAW;;;MAKrC,MAAA,MClEA,AAmBA,GAAA,IAAA,WAUE,WACI,EAAsB,EAAqB,EAAc,EACzD,GAXJ,KAAA,cAAgB,CAAC,aAAc,cAAe,MAC9C,KAAA,YAAwB,GAWtB,KAAK,YAAc,EACnB,KAAK,MAAQ,EAAW,GACxB,KAAK,YAAc,EACnB,KAAK,KAAO,EACZ,KAAK,MAAQ,EACb,KAAK,KAAO,EACZ,KAAK,SAAW;;;;;;;;8BAQU,KAAK,MAAK;oDACY,EAAW;yCACtB,KAAK,MAAK;0BACzB,EAAW;;;sCAGC,KAAK,MAAK;;;;;;;;;;;;;;;yBAevB,EAAK,oBAAoB,EAAI;;;;;;;yCAOb,EAAK;0BACpB,EAAI;;;;0CAIY,EAAI;;;;;;;;;;;;;;MAgB9C,MAAA,MC/FA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAgB,EAAc,EAChD,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;gCAQY,KAAK,YAAY,GAAE;gCACnB,KAAK,YAAY,GAAE;;;;;;;;;;;;;;;iCAelB,EAAG;;;;;;;;;;;yBAWX,EAAG,UAAU,EAAG;;;6DAGoB,EAAI;;;;;;;;;;;;;;;;;;;;;0CAqBvB,EAAW;;;MAKrD,MAAA,MC/GA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAY,EAAwB,EAAuB,EACjE,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAcnB,EAAqB;kBACnC,EAAc;gDACgB,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;kDACJ,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;gCAOvB,EAAS;;;;qCAIJ,EAAoB;;;;;;;;MAUzD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EACF,EAAuB,EAAwB,EAAuB,EAC1E,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;gCAiBhC,EAAoB;mBACjC,EAAa;gDACgB,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;kCAOvB,EAAS;;;;;;yBAMlB,EAAqB,MAAM,EAAoB;yBAC/C,EAAoB;;;;;;;;;MAW7C,MAAA,MChLA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkC,EAClC,EAAoB,EAAoB,EACxC,EAA2B,GAD3B,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IACxC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAT/B,KAAA,cAAgB,CAAC,UAAW,WAC5B,KAAA,aAAe,GACf,KAAA,aAAe,GAQb,KAAK,YAAc,EAEnB,GAAM,GAAY,EAAa,EAAO,GAAK,EAAO,GAC5C,EAAwB,KAAK,KAAK,EAAY,GAE9C,EAAU,EAAa,cAAgB,cACvC,EAAU,EAAa,cAAgB,cACvC,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAC1D,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAE5D,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;sCAEa,EAAqB;;;;8BAI7B,EAAqB;sCACb,EAAO;sCACP,EAAO;;;;uBAItB,EAAS,GAAE,MAAM,EAAS,GAAE;uBAC5B,EAAS,GAAE,MAAM,EAAS,GAAE;;;;;;;;;UASzC,EAAc;;UAEd,EAAsB;;;;MAMhC,MAAA,MChGA,AAoBA,GAAA,IAAA,WAQE,WAAY,EAAmB,EAAqB,GAPpD,KAAA,cAAgB,CAAC,SAQf,KAAK,YAAc,CAAC,EAAW,GAE/B,KAAK,SAAW;;;;;;;;;;8BAUU,GAAc,GAAC;;;;;;;;;;0BAUnB,GAAc,GAAC;;MAKvC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAM,mBAAmB,EAAc,SAExD,EAAM,GAAG,UAAU,EAAK,QAAS,KAGvC,KChEA,AAmBA,GAAA,IAAA,WAQE,WACI,EAAoB,EAAe,EAAiB,GARxD,KAAA,cAAgB,CAAC,WASf,KAAK,YAAc,CAAC,EAAY,GAEhC,KAAK,SAAW;;;;8BAIU,EAAQ,YAAY,EAAO;;;MAKzD,MAAA,MCxCA,AAqBA,GAAA,IAAA,WAOE,WACI,GAPJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEzB,GAAI,IAAS,EACX,KAAK,SAAW;;;;aAMhB,GAAM,GAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EACF,GAAwB,EAAM,EAAa,GACzC,EAAQ,GACV,EAAM,EAAY,EAAY,OAAS,GACvC,EAAY,EAAY,OAAS,GAAI,GACnC,EAAS,GAAU,EAAa,GAEtC,KAAK,SAAW;;YAEV,EAAK;;eAEF,EAAoB;;;cAGrB,EAAK;;6BAEU,EAAM;;;SAMnC,MAAA,MAEA,YAA4B,EAAc,GAGxC,OAFM,GAAS,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAG1B,OAFI,GAAW,KAAQ,EAAI,IAAM,OAAK,KAAK,KAAQ,EAAI,IAAM,OAEpD,EAAI,EAAG,EAAI,EAAM,IACxB,EAAW,EAAK,EAAK,OAAS,EAAI,GAAE,IAAM,EAG5C,EAAO,KAAK,GAGhB,MAAO,GAGT,YACI,EAAc,EAAiB,GACjC,GAAI,IAAS,EACX,MAAO,QAAQ,EAAM,GAIvB,OADI,GAAO,GACF,EAAI,EAAO,EAAG,EAAI,EAAM,IAC/B,GAAW,EAAK,GAAE,OAAO,EAAM,GAC/B,AAAI,EAAI,EAAO,GACb,IAAQ,MAIZ,MAAO,GAGT,YACI,EAAc,EAAc,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,GAGT,GAAM,GAAY,EAAK,MAAM,IAE7B,MAAO;cACK,EAAU,GAAE;cACZ,EAAU,GAAE;;;;0BAIA,EAAI;0BACJ,EAAI;IAI9B,YAAmB,EAAiB,GAClC,GAAM,GAAO,EAAM,OACb,EAAe,GAAmB,EAAM,GAC9C,MAAI,KAAS,EACJ;wBACa,EAAM,GAAE;kBAIvB,QAAQ,EAAa,GAAE;8BACF,EAAa,GAAE;8BACf,EAAa,GAAE;uCACN,EAAa,GAAE,ICrItD,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAClB,GANJ,KAAA,cAAgB,CAAC,KAOf,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KACnD,GAAM,GAAO,EAAO,OACd,EAAO,GAAkB,GAEzB,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EACF,CAAC,YAAa,YAAa,YAAa,aAAa,MAAM,EAAG,GAElE,GAAI,IAAS,GACX,KAAK,SAAW;sBACA,EAAK;oBACP,EAAG;;;;;8BAKO,EAAa;;;;;QAMrC,OAEF,KAAK,SAAW;QACZ,EAAI,YAAY,EAAI,IAAI,EAAK;QAC7B,EAAI,UAAU,EAAI,IAAI,EAAG;;;UAGvB,EAAI;;4BAEc,EAAa;;YAE7B,EAAI;2BACW,EAAc;;;MAKzC,MAAA,MCrEA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAClB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KA4BnD,OA3BM,GAAO,EAAO,OACd,EAAQ,GAAkB,GAE1B,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EAAS,GAAY,KAAM,GAC3B,EAAS,GAAY,SAAU,GAC/B,EAAY,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAC1D,EACF,IAAS,EAAI,SAAW,QAAQ,EAAO,MAAM,IAAI,OAAM,IAErD,EAAiB,CAClB,EAAK,mBAAuB,EAAO,EAAO,GAAE;YACzC,EAAM;QAEZ,IAAS,EAAI,GAAK;;SAEf,EAAO,EAAO,GAAE;YACb,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAAE,MACtD,IAAS,EAAI,GAAK,KAAK,EAAO,EAAO,GAAE;cAC/B,EAAM,OAGV,EAAc,IAAS,EACzB,0BACA,6DACA,EAAW,GACN,EAAI,EAAG,EAAI,IAAS,EAAI,EAAI,EAAG,EAAI,EAAG,IAC7C,GAAY;UACR,EAAe,GAAE;cACb,EAAW;mBACN,EAAC,aAAa,EAAa;;YAElC,EAAK;mBACE,EAAC,uBAAuB,EAAO,OAAM,MAAM,EAAS;;QAInE,GAAa,IAAS,EAAI,KAAO,KAEjC,KAAK,SAAW;cACN,EAAK,YAAY,EAAK,IAAI,EAAK;cAC/B,EAAK,UAAU,EAAK,IAAI,EAAG;;;UAG/B,EAAK;;UAEL,EAAQ;;;MAKlB,MAAA,MCrFA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MACzB,EAA0B,cAAc,EAAS,SAAQ,YAC3D,EAAS,QAAO,YAAY,EAAS,WAAU,OAC7C,EACF,SAAS,EAAS,QAAO,YAAY,EAAS,WAAU,OAExD,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;sCACgB,EAAY,KAAK,EAAW;mCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBnB,EAAqB;sBACjC,EAAc;;;kCAGF,EAAS,SAAQ;;;;oCAIf,EAAoB;wBAChC,EAAa;;;oCAGD,EAAS,QAAO;;;;;;;;;;0BAU1B,EAAS;;;mCAIzB,GAAoB,EAAsB,EACA,EACvB,QAAQ,EAAoB,SAAO;;;;;;QAO1D,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;oCACgB,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;0CACT,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;kCAkBZ,EAAmB;;;;gCAIrB,EAAqB;oBACjC,EAAc;;;gCAGF,EAAS,SAAQ;;;;kCAIf,EAAsB;uCACjB,EAAa;;;;yCAIX,EAAa;6CACT,EAAa;6CACb,EAAa;;;cAG5C,EAAa;;;gCAGK,EAAsB;gBACtC,KAA6B,GAAC;;;;;;;;cAQhC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;;;;;cAKxC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;6CACT,EAAa;;;;cAI5C,EAAa;;;oBAGP,EAAW;;MAI/B,MAAA,MAEA,GAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MAE3B,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;;oBAEF,EAAW,KAAK,EAAY,KAAK,EAAW;mCAC7B,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBhC,EAAoB;sBAChC,EAAa;;;kCAGD,EAAS,QAAO;;;;oCAId,EAAqB;wBACjC,EAAc;;;oCAGF,EAAS,SAAQ;;;;sCAIf,EAAoB;0BAChC,EAAa;;;sCAGD,EAAS,QAAO;;;;;;;;;;4BAU1B,EAAS;;;qCAI3B,GACK,EACI,cAAc,EAAS,QAAO,YAC1B,EAAS,SAAQ,YAAY,EAAS,QAAO,YAC7C,EAAS,WAAU,QACvB,UAAU,EAAS,SAAQ,YACvB,EAAS,QAAO,YAAY,EAAS,WAAU,QACxD,QAAQ,EAAqB,MAAM,EAAoB;6BACxC,EAAoB,SAAO;;;;;;;QAQlD,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;;gBAEJ,EAAW,KAAK,EAAY,KAAK,EAAW;iCAC3B,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;;kCAmBZ,EAAmB;;;;gCAIrB,EAAoB;oBAChC,EAAa;;;gCAGD,EAAS,QAAO;;;;kCAId,EAAqB;oBACnC,EAAc;;;kCAGA,EAAS,SAAQ;;;;oCAIf,EAAsB;yCACjB,EAAa;;;;+CAIP,EAAa;mDACT,EAAa;mDACb,EAAa;;;gBAGhD,EAAa;;;kCAGK,EAAsB;kBACtC,KAA6B,GAAC;;;;;;;;gBAQhC,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;;;;;gBAK5C,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;mDACT,EAAa;;;;gBAIhD,EAAa;;;sBAGP,EAAW;;;MAKjC,MAAA,MCvcA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,OAAQ,EAAA,EAAA,QACtC,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAI,GAAsB,MACtB,EAAY,GAEhB,AAAI,IAAe,OACjB,EAAsB,MACjB,AAAI,IAAe,MAExB,GAAsB,cACtB,EAAY,OACP,AAAI,IAAe,OAExB,GAAsB,eACtB,EAAY,OAGd,GAAI,GAAiB,EAAU,IAAI,EAAU,IAAI,EAAU,qEAG3D,AAAI,IAAe,MACjB,EAAc,WACT,AAAI,IAAe,OACxB,EAAc,YACT,AAAI,IAAe,MACxB,EAAc,WACT,AAAI,IAAe,OACxB,GAAc,YAGhB,GAAM,GAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEzC,EAAgB;YACZ,KAAe,OAAK;;mBAEb,KAAe,QAAM;;;;wBAIhB,EAAS;;MAIzB,EAAU,OAEd,AAAI,IAAe,MACjB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SACL,AAAI,IAAe,OACxB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SAGZ,GAAI,GAAmB,GACvB,AAAI,EAAS,EAAa,GACxB,GAAmB;oCACW,EAAM;;;SAKtC,KAAK,SAAW;0CACsB,EAAmB;;;;UAInD,EAAgB;;;;;;;;kCAQQ,EAAU;;kCAEV,EAAmB;;;;;;8BAMvB,EAAqB;;YAEvC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;YAC/B,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCvKA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAuC,GANnD,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAGnB,OADI,GAAW,GACN,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAS,eACb,AAAI,EAAI,IAAM,GACZ,IAAU,kBAEZ,AAAI,EAAI,GACN,IAAU,kBAGZ,GAAY;UACR,EAAM;UACN,GAAI,EAAI,0CAA4C,IAAE;;;;;;mBAM7C,EAAC;;UAEV,GAAI,EAAI,IAAM,IAAE;QAItB,KAAK,SAAW;QACZ,GAAuB,GAAW;QAClC,GAA+B,GAAY;;;;;;;;qBAQ9B,EAAY,GAAE;qBACd,EAAY,GAAE;;UAEzB,EAAQ;;;;MAMlB,MAAA,MAEA,YAAgC,GAC9B,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;QAED,EAAsB;;;ICnF9B,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;;sDAMY,GAAU,GAAC;;;;;;qDAMZ,GAAS,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAgC/D,MAAA,MC1IA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MA+B/D,MAAA,MC3EA,AAmBA,GAAA,IAAA,WAOE,WACI,EAA8C,EAC9C,EAAkB,GARtB,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;uCACxB,EAAQ;;;;;;;;;;;;;;;;;;;;;;gCAsBf,GAAQ,GAAC;uCACF,GAAW,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAkDnD,MAAA,MCzHA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;sBAKpB,EAAe,GAAE;sCACD,EAAe,GAAE;;;wBAG/B,EAAe,GAAE;wCACD,EAAe,GAAE;;;4BAG7B,EAAO;kBACjB,EAAY;;;;4BAIF,EAAM;kBAChB,EAAY;;;;;;;;;;;;MAc9B,MAAA,MC/HA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAI5C,EAAY,EAAe,MAAQ,MAEzC,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;8DAaD,EAAS;;;;;;MAQvE,MAAA,MCpEA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAI5C,GAFA,KAAK,YAAc,EAEf,IAAS,GACX,KAAK,SAAW;;;2BAGK,EAAO,GAAE;;QAG9B,OAEF,GAAM,GAAa,SAAC,GAClB,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,aAAa,EAAC,QAE5B,UAAU,EAAC,KAEd,EAAW,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,KAAI,KAAK,KACpD,EAAO,GAAkB,GAE/B,KAAK,SAAW;;UAEV,EAAI;yBACW,EAAQ;;MAIjC,MAAA,MC1DA,AAqBA,GAAA,IAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAE5C,KAAK,YAAc,EACnB,GAAM,GAAW,GAAY,KAAM,GAC7B,EACC,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACrD,EAAa,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACjE,EAAO,GAAkB,GAC/B,AAAI,IAAS,EACX,KAAK,SAAW;;;;uCAIiB,EAAO,GAAE;cAClC,EAAO,GAAE;eACR,EAAU;2CACkB,EAAO,GAAE;kBAClC,EAAO,GAAE;;;;QAMrB,KAAK,SAAW;;YAEV,EAAI;;uBAEO,EAAK,EAAS,SAAQ;eAC9B,EAAU;yBACA,EAAK,EAAS,SAAQ;;eAEhC,EAAO;yBACG,EAAK,EAAS,SAAQ;iBAC9B,EAAU;2BACA,EAAK,EAAS,SAAQ;;;;;MAQ7C,WAAc,GACZ,MAAO,GAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QAChD,EAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAoB,GAClB,GAAM,GAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,EAAG,KACnD,EAAW,EAAc,KAAK,KAC9B,EAAY,EAAc,MAAM,IAAI,KAAK,KAC/C,MAAO,mBAAmB,EAAQ,WAAW,EAAS,KAGxD,WAAoB,EAAW,GAC7B,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,MAAM,EAAU,GAAE,OAE9B,GAAG,EAAU,IAI5B,MAAA,MC7GA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAoB,EAAkB,EACtC,EAAqB,EAAmB,EACxC,GAPJ,KAAA,cAAgB,CAAC,UAAW,UAAW,gBAQrC,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAClC,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,QAElB,GAAM,GAAiB,cAAc,EAAa,IAE9C,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,gBAElB,GAAM,GAAiB,cAAc,EAAa,IAE5C,EAAe,EAAW,EAAI,aAAe,UACnD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,EAAO;;;YAG7C,EAAK;;;gCAGe,EAAU;;kCAER,EAAQ;kCACR,EAAc;0CACN,EAAY;;;uBAG/B,EAAc;;;;;;QAQrC,MAAA,MCvEA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,IAAK,cAOpB,GAAM,GAAa,EAAU,WACvB,EAAY,EAAU,UACtB,EAAS,EAAU,OACnB,EAAc,EAAU,YACxB,EAAU,EAAc,KAAK,KAAK,EAAS,GACjD,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAM,GAAsB,MACtB,EAAc,WAEd,EAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEvC,EAAgB;;MAIlB,EAAwB,GAC5B,AAAI,EAAS,EAAa,GACxB,GAAwB;oCACM,EAAM;;;SAMtC,GAAI,GAA4B,GAChC,AAAI,EAAS,EAAa,GACxB,GAA4B;oCACE,EAAM;;;SAMtC,KAAK,SAAW;0CACsB,EAAmB;;;UAGnD,EAAqB;;;;;UAKrB,EAAyB;;;;;;;;;YASvB,EAAW,cAAc,EAAU;wDACS,EAAW;;;;8BAIrC,EAAqB;;;;;;;;;;;;;;;;YAgBvC,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;;;;;;;;;;;;;;;;;YAiB/B,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCjKA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAe,EAAiB,GAJ5C,KAAA,cAAgB,CAAC,IAAK,IAAK,KAKzB,KAAK,YAAc,EAEnB,GAAI,GACA,EACJ,GAAI,EAAO,EACT,KAAM,OAAM,kBAAkB,EAAI,yBAGpC,GAAI,IAAS,EACX,EAAW,QACX,EAAU,aAKV,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAClD,EAAa,GACb,EAAc,GACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAY,KAAK,GAAG,EAAc,IAClC,AAAI,EAAI,GACN,EAAW,KAAK,GAAG,EAAc,IAGrC,EAAU,EAAW,OACrB,EAAW,EAAY,OAGzB,GAAM,GAAQ,GAAkB,GAEhC,KAAK,SAAW;;UAEV,EAAK;4BACa,EAAO;;2BAER,EAAQ;;2BAER,EAAQ;;;MAKnC,MAAA,MCjEA,AAqBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,UASf,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAc,qBAAqB,KAAK,KAAI,KAC5C,EAAe,GAAU,KAAK,MAEhC,EACE,EAAW,EAAS,IAAI,SAAC,EAAG,GAChC,MAAO,aAAa,GAAO,GAAE,YAAY,EAAC,cAAc,GAAO,GAAE,MAEnE,EAAO;UACD,EAAK;UACL,EAAK;UACL,EAAS,KAAK;GAAK;QAEzB,KAAK,SAAW;QACZ,EAAW;;UAET,EAAI;8BACgB,EAAY;;MAKxC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KAEM,GAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,YAAmB,GACjB,GAAI,IAAS,EACX,MAAO,YACF,GAAI,GAAQ,EACjB,MAAO,IAAO,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAA,aAAe,IAAG,KAAK,KAE7D,KAAM,OAAM,oBAAoB,EAAI,yBCpFxC,AAsBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,UACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAS,GAAY,SAAU,KAAK,MACpC,EAAY,GAAY,YAAa,KAAK,MAE1C,EACF,KAAK,OAAS,EAAI,YAAc,QAAQ,EAAU,MAAM,IAAI,OAAM,IAChE,EACF,wBAAwB,EAAU,OAAM,MAAM,EAAS,IACrD,EAAW;mBACF,EAAU;cACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;YACnB,EAAU,KAAK,KAAO,GAAE;;MAG1B,EAAW,KAAK,OAAS,EAAI,GAAK;UAClC,EAAO,KAAK,KAAO,GAAE;cACjB,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;gBACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;cACpD,EAAU,KAAK,KAAO,GAAE;uBACf,EAAU;;;MAKvB,EAAiB,KAAK,MAAQ,EAChC;cACM,EAAK,IAAI,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,SAAS,EAAC,MAAK,OAAM,KAC3D,EAAS,IAAI,SAAC,EAAG,GAAM,MAAG,GAAU,GAAE,MAAM,EAAO,GAAE,YAAY,EAAC,OAC7D,KAAK;GACd,KAAK,SAAW;0BACM,KAAK,KAAI;;UAEzB,EAAK;UACL,EAAK;UACL,EAAc;;UAEd,EAAQ;UACR,EAAQ;;;MAMhB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KCtGA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAiB,EAAmB,GAJhD,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAK,OACZ,EAAa,GAAkB,EAAK,QACpC,EAAQ,GAAkB,EAAK,QAEjC,EAAY,GAChB,GAAI,IAAS,EACX,EAAY,gCAEZ,GAAI,GAAa,EACjB,EACI,EAAK,IAAI,SAAC,EAAG,GACP,WACO,EAAK,SAAW,EACnB,oBAAoB,EAAC,aAAa,EAAC,IACnC,UAAU,GAAa,GAAC,eAAe,EAAC,aAAa,EAAC,MAE3D,KAAK,KAGhB,KAAK,SAAW;QACZ,EAAU,YAAY,EAAU,IAAI,EAAK;QACzC,EAAU,cAAc,EAAU,IAAI,EAAO;;;UAG3C,EAAK;yBACU,EAAS;;MAIlC,MAAA,MCxDA,AAuBA,GAAA,IAAA,WAUE,WAAoB,GAAA,KAAA,MAAA,EATZ,KAAA,gBAAkB,EAClB,KAAA,gBAAkB,EAClB,KAAA,mBAAqB,EACrB,KAAA,cAAgB,EAEhB,KAAA,aAAkD,GAClD,KAAA,WAAa,GACb,KAAA,aAAkD,GAI1D,SAAA,UAAA,eAAA,SACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAkC,EAAO,GAE3D,EAAW,GAAuB,EAAS,EAAiB,GAClE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAEhC,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAS,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cACpD,GAEJ,GAAI,KAAK,aAAa,GAAU,OAAS,GACvC,KAAK,kBACL,KAAK,kBACL,KAAK,eAAiB,EACtB,KAAK,MACL,GAAM,GAAa,KAAK,aAAa,GAAU,QAC/C,YAAK,aAAa,GAAU,KAAK,GAC1B,EAGT,GAAI,GACJ,MAAI,KAAoB,GAAoB,mBAC1C,EAAa,KAAK,MAAM,0BAA0B,EAAQ,GAAI,EAAQ,IACjE,AAAI,IAAoB,GAAoB,mBACjD,EACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,IAC/D,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AACH,IAAoB,GAAoB,0BAC1C,GACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,KAEtE,KAAK,aAAa,GAAU,KAAK,GAEjC,KAAK,kBACL,KAAK,oBAAsB,EAC3B,KAAK,MAEE,GAGT,EAAA,UAAA,eAAA,SACI,EAAuB,EACvB,EAA8B,GAChC,GAAI,KAAK,cAAgB,KAEvB,OAEF,GAAM,GACF,GAAkC,EAAgB,GAChD,EAAW,GAAuB,EAAO,EAAiB,GAChE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAO,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cAClD,GACE,EAAqB,EAAA,MAAM,IAAI,kCACrC,AAAI,IAAuB,IACvB,KAAK,mBAAqB,EAC5B,MAAK,MAAM,oBAAoB,GAC/B,KAAK,oBAAsB,GAE3B,MAAK,aAAa,GAAU,KAAK,GACjC,KAAK,kBACL,KAAK,eAAiB,GAGxB,KAAK,kBAEL,GAAM,GAAU,KAAK,aAAa,GAC5B,EAAW,EAAQ,QAAQ,GACjC,GAAI,EAAW,EACb,KAAM,IAAI,OACN,4EAGN,EAAQ,OAAO,EAAU,GACzB,KAAK,OAGC,EAAA,UAAA,IAAR,WACE,GAAI,CAAC,KAAK,WACR,OAEF,GAAM,GAAQ,KAAK,gBAAkB,KAAK,gBAC1C,QAAQ,IACJ,YAAgB,KAAK,gBAAe,MAAM,KAAK,gBAC/C,IAAI,EAAK,KACb,GAAM,GAAY,KAAK,cAAgB,KAAK,mBAC5C,QAAQ,IAAI,oBAAoB,KAAK,oBACrC,QAAQ,IAAI,iBAAiB,KAAK,cAAa,KAC3C,KAAK,MAAM,IAAM,GAAU,OAGjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGd,OAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,+CAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,cAAgB,KAEvB,OAEF,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,KAAK,aAAe,KACpB,KAAK,aAAe,KACpB,KAAK,gBAAkB,EACvB,KAAK,gBAAkB,EACvB,KAAK,mBAAqB,EAC1B,KAAK,cAAgB,GAEzB,KAEA,YACI,EAA2B,GAE7B,GAAM,GAAQ,EACd,GAAI,IAAmB,EAAM,KAC3B,MAAO,GACF,GAAI,IAAmB,EAAM,KAClC,MAAO,GACF,GAAI,IAAmB,EAAM,QAClC,MAAO,IACF,GAAI,IAAmB,EAAG,KAC/B,MAAO,IACF,GAAI,IAAmB,EAAM,QAClC,MAAO,GAET,KAAM,IAAI,OAAM,2BAA2B,eAIzC,EAAyB,EACzB,EAA2B,EAC3B,GAMF,GAAM,GACF,GAAiC,EAAiB,GAElD,EACJ,GAAI,GACI,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAEpB,EAAc,EAAc,OAGtB,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,EAAc,EAAQ,EAGxB,GAAM,GAAkB,GAA0B,EAAI,GACtD,MAAO,GAAc,EAGvB,YACI,EACA,GACF,OAAQ,OACD,IAAoB,mBACvB,MAAO,IAAwC,OAC5C,IAAoB,mBACvB,MAAO,IAA+C,OACnD,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,yBACvB,MAAO,IAA+C,WAEtD,KAAM,IAAI,OAAM,iCAAiC,IAIvD,YAAwC,GAEtC,MAAI,GAAA,MAAM,QAAQ,gCACZ,EACK,GAAoB,mBAEtB,GAAoB,iBAGzB,EACK,GAAoB,mBAEtB,GAAoB,iBAG7B,YACI,EAA8B,GAChC,GAAI,IAAmB,GAAa,OAClC,MAAO,IAAoB,mBACtB,GAAI,IAAmB,GAAa,QAAU,GAAkB,KACrE,MAAO,IAA+B,GACjC,GACH,IAAmB,GAAa,UAChC,IAAmB,GAAa,OAClC,MAAO,IAAoB,yBAE7B,KAAM,IAAI,OAAM,gCAAgC,GAGlD,YACI,EAAgC,EAChC,GACF,MAAU,GAAa,GAAE,IAAI,EAAa,GAAE,IAAI,EAAe,IAAI,ECxRrE,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,GAEpC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,GAErC,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,GACvB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,iBAAiB,EAAI,yBAEnC,GAAI,IAAS,EACX,MAAO,eAAe,EAAO,GAAE,IAMjC,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,UAAW,WAE7D,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAa,KAAK,QAAQ,EAAc,GAAE,KAAK,EAAO,GAAE,KAE1D,MAAO,GAAa,OC5DtB,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MAEM,GAAoB,0BAEb,GAAS,YAET,GAAM,iBAEN,GAAO,GAAoB;;EAI3B,GAAQ,GAAoB;;EAI5B,GAAM,0CAEN,GAAO;;;uBAGG,EAAA,aAAa,gBAAe;kBACjC,EAAA,aAAa,WAAU;;cAIpB,GAAA,MAAA,KAAA,QAAA,GAAA,GACZ,GAAoB;mCACM,EAAK;KAIjC,GAAM,IAAM,aAEN,GAAO,kBAEP,GAAQ,mBAER,GAAO;;;EAKP,GAAS,0BAET,GAAS,0BAET,GAAY,wCAEZ,GAAQ;;;;;;;;;;;;;;;EAiBR,GAAM,iBAEN,GAAQ,uBAER,GAAM;kBAGN,GAAQ,uBAER,GAAO,kBAEP,GAAQ,yBAER,GAAU,sCAeV,GAAW;;;;;;;;;;;;;;;;;;;;EAsBX,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;EAI3B,GAAO;;;EAKP,GAAO;;;EAKP,GAAO;;;EAKP,GAAQ,GAAoB,qCAE5B,GAAQ,GAAoB;;sCAI5B,GAAQ,GAAoB;;+CAI5B,GAAM;;;;cAIL,EAAA,aAAa,MAAK;eACjB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;;;;;;EAQrB,GAAa,kBAEb,GAAc,6BAEd,GAAS,wBAET,GAAQ,YC9NrB,AAmBO,GAAM,IAAS,YAET,GAAM;;;;;;;;;EAWN,GAAO;;;;;;;;;;EAYP,GAAQ;;;;;;;;;;EAYR,GAAM;;;;;;;;;EAWnB,GAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MCzFA,AAqBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEnB,EAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EAAe,GAAgB,EAAM,GACrC,EAAY,EAAS,MAAM,IAC3B,EAAS,GAAQ,EAAI,KAAO,QAAQ,EAAU,KAAK,KAAI,IAE7D,KAAK,SAAW;;UAEV,EAAK;kCACmB,EAAY;;4CAEF,EAAM;;MAIlD,MAAA,MC/CA,AA2BO,GAAA,IAAA,EAAA,aAAA,aACD,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAiFlB,GAAkB,KAClB,GAAkB,KA4BzB,GAAuE,eAE9C,GAC7B,MAAI,KAAgB,KAGpB,IAAa,GAAgB,IACtB,GAAa,GAGtB,YACI,EAAqC,GACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACnC,IAAe,SACjB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,OACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,MACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GAET,KAAM,IAAI,OAAM,cACZ,EAAU,oDAKhB,GAAM,IAA6B,IAK7B,GAAyB,IAC/B,cACE,MAAI,GAAA,MAAM,OAAO,QAAU,KAClB,KAED,EAAA,MAAM,OAAO,OAAO,OAAS,EAAA,MAAM,OAAO,OAAO,MACjD,OAAO,iBACX,GAAyB,KAAO,KAM/B,GAAM,IAA8B,mBAEL,GAAA,EAAA,GAkCpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GA/BM,EAAA,YAAc,GAAI,SAGlB,EAAA,gBAAkB,GAAI,SAGtB,EAAA,aAAe,GAAI,SACnB,EAAA,cAAgB,EAOhB,EAAA,aAAe,EAEf,EAAA,eAAiB,EAUjB,EAAA,kBAAoB,GACpB,EAAA,sBAAwB,GAgXxB,EAAA,eAAiB,EAq+DjB,EAAA,SAAW,GAj1Eb,CAAC,EAAA,MAAM,QAAQ,aACjB,KAAM,IAAI,OAAM,yCAGlB,GAAI,GAAS,MACX,GAAM,GAAK,GAAgB,EAAA,MAAM,UAAU,kBAC3C,EAAK,YAAc,GAAe,EAAA,MAAM,UAAU,kBAClD,EAAK,MAAQ,GAAI,IAAa,GAC9B,EAAK,OAAS,EAAG,OACjB,EAAK,oBAAsB,OAE3B,GAAK,MAAQ,EACb,EAAK,YAAc,GACnB,EAAK,oBAAsB,GAC3B,EAAK,OAAS,EAAM,GAAG,OAEzB,SAAK,eAAiB,GAAI,IAAe,EAAK,OAC9C,EAAK,mBAAqB,KAE1B,EAAK,QAAU,GAAI,GAAA,YAAY,EAAM,EAAA,YAGvC,SAAA,UAAA,WAAA,WACE,MAAO,MAAK,QAAQ,aACf,MAAK,WAAa,KAAK,WAAW,aAAe,GAClD,KAAK,gBAGX,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAK5C,GAJA,AAAI,GAAA,MAAM,QAAQ,mCACd,EAAA,MAAM,QAAQ,WAChB,KAAK,uBAAuB,GAE1B,IAAU,aAAe,GAAU,KACrC,KAAM,IAAI,OACN,yEAGN,GAAM,GAAS,GACf,YAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,IAC1D,GAIT,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,YAIV,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,aAIZ,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAK3D,GAHA,AAAI,EAAA,MAAM,QAAQ,UAChB,KAAK,uBAAuB,GAE1B,IAAU,YACZ,KAAM,IAAI,OACN,yEAGN,KAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,KAGnE,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAc,KAAK,QAAQ,IAAI,GAErC,EAAY,WAEZ,AAAI,EAAY,SAAW,GACzB,KAAK,YAAY,KAKvB,EAAA,UAAA,SAAA,SAAS,GACP,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,eAAgB,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SAKpD,GAAI,GAAS,MACX,GAAI,GAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEtC,GAAM,GACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,SAAS,EAAI,QAC/B,YAAK,8BAA8B,GAC5B,EAET,GAAI,GAAU,KACZ,MAAO,MAAK,qBAAqB,GAEnC,GAAI,IAAU,SACZ,MAAO,GAET,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GACJ,GAAI,IAAU,aACZ,GAAM,GAAa,EAAe,KAAK,WACjC,EAAa,EAAe,KAAK,WACvC,EAAS,EAAA,aAAa,uBAAuB,EAAY,OAEzD,GAAS,KAAK,qBAAqB,GAGrC,MAAI,IACF,MAAK,gBAAkB,EAAA,KAAK,MAAQ,GAE/B,KAAK,qBAAqB,EAAQ,IAGrC,EAAA,UAAA,KAAN,SAAW,6IACT,GAAI,KAAK,YAAY,IAAI,GACjB,SAAc,KAAK,YAAY,IAAI,GACzC,CAAA,EAAO,GAAI,SAAoB,SAAA,GAAW,MAAA,GAAY,KAAK,MAQ7D,GANM,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAyD,EAAO,OAAxD,EAAiD,EAAO,MAAjD,EAA0C,EAAO,MAA1C,EAAmC,EAAO,MAAnC,EAA4B,EAAO,eAAnB,EAAY,EAAO,SAKnE,GAAS,KACP,SAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEhC,EACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,KAAK,EAAI,QAC3B,KAAK,8BAA8B,GACnC,CAAA,EAAO,GAGT,GAAI,GAAU,KACZ,MAAA,CAAA,EAAO,KAAK,qBAAqB,IAGnC,GAAI,CAAC,EAAA,MAAM,QAAQ,iCACf,EAAA,MAAM,UAAU,mBAAqB,EACvC,KAAM,IAAI,OACN,sGAIF,GAAS,KAGb,AAAI,IAAU,aAAe,EAAA,MAAM,IAAI,2BAErC,GAAoB,KAAK,OAAO,GAC1B,EAAU,KAAK,QAAQ,IAAI,EAAkB,QAEnD,EAAS,GAAA,KAAK,OAAM,wBAAuB,MAAA,EAAA,CACvC,EAAQ,SAAO,OAAK,GAA0B,MAGpD,KAAK,YAAY,IAAI,EAAQ,IAEzB,IAAU,YAEZ,CAAA,EAAM,KAAK,MAAM,yBAFf,CAAA,EAAA,UAEF,EAAA,8BAKE,KAAU,YACD,CAAA,EAAM,QAAQ,IACrB,CAAC,EAAe,KAAK,OAAQ,EAAe,KAAK,UAFnD,CAAA,EAAA,UACI,SAAK,EAAA,OAEL,EAAa,EAAG,GAChB,EAAa,EAAG,GACtB,EAAO,EAAA,aAAa,uBAChB,EAA4B,gBAC3B,AAAI,GAAU,KACnB,EAAO,KAAK,qBAAqB,GAE3B,GAAO,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,MAAM,gCAAgC,EAAQ,qBAE5D,MAAI,IAAqB,MACvB,KAAK,8BAA8B,GAE/B,EAAY,KAAK,qBAAqB,EAAQ,GAE9C,EAAc,KAAK,YAAY,IAAI,GACzC,KAAK,YAAY,OAAO,GAGxB,EAAY,QAAQ,SAAA,GAAW,MAAA,GAAQ,KACvC,AAAI,KAAK,gBAAgB,IAAI,IAC3B,MAAK,gBAAgB,OAAO,GAC5B,KAAK,YAAY,GACjB,KAAK,kBAEP,CAAA,EAAO,SAGD,EAAA,UAAA,uBAAR,SAA+B,GAC7B,GAAI,GAAU,KACZ,OAEF,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,KACjC,GAAM,GAAM,EAAO,GACnB,GAAI,CAAC,GAA4B,GAC/B,KAAI,GAAA,MAAM,QAAQ,gCACV,MACF,aAAa,EAAG,iJAIhB,MAAM,aAAa,EAAG,4CAK1B,EAAA,UAAA,qBAAR,SAA6B,SACrB,EAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SACf,EAAO,EAAA,KAAK,cAAc,GAChC,GAAI,EAAA,MAAM,QAAQ,iCAChB,GAAM,GAAY,KAAK,OAAO,GACxB,EAAU,KAAK,QAAQ,IAAI,EAAU,QACrC,EAAO,GAAA,KAAK,OACA,gCAA+B,MAAA,EAAA,CAC5B,EAAQ,SAAO,OAAK,GAA0B,KACjD,SAAS,EAAG,GAE9B,YAAK,8BAA8B,GAE5B,EAGT,GAAM,GACF,EAAA,MAAM,QAAQ,eAAiB,IAAa,GAC1C,EACF,EAAyB,GAAwB,GAAS,EACxD,EAAU,EACZ,GAAI,IAAyB,GAC7B,GAAI,IAAmB,GACrB,EAAS,KAAK,gBAChB,EAAS,CAAC,CAAC,MAAO,EAAa,MAAK,EAAE,OAAM,IAAI,WAC9C,EAAU,KAAK,QAAQ,IAAI,EAAO,QAClC,EACF,KAAK,MACA,gDACG,EAAQ,QAAS,EAAQ,SAAS,GAAI,EAAQ,SAAS,IAC1D,SAAS,EAAG,GACrB,YAAK,8BAA8B,GAE5B,GAGH,EAAA,UAAA,KAAN,SAAW,qHACH,GAAkB,KAAK,aACvB,EAA+B,GAEjC,EAAgB,GACpB,AAAI,KAAK,oBAAsB,KAC7B,MAAK,mBAAqB,EAC1B,EAAgB,IAEhB,KAAK,aAAa,KAAK,GAEzB,KAAK,aAAe,EAEpB,IAGM,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,SACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OACpB,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,QACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OAE1B,KAAK,aAAe,EAEpB,AAAI,GACF,MAAK,mBAAqB,MAGtB,EAAuB,CAC3B,aAAc,KAAK,aACnB,eAAgB,KAAK,eACrB,SAAU,KACV,OAAQ,MAGN,EAAA,MAAM,UAAU,iDAAmD,EACpD,CAAA,EAAM,QAAQ,IAAI,IADjC,CAAA,EAAA,UACI,SAAW,EAAA,OAEjB,EAAI,SAAc,EAAA,KAAK,IAAI,GAC3B,EAAI,oBAAyB,WACzB,MAAA,GAAS,IAAI,SAAC,EAAG,GAAM,MAAC,CAAC,KAAM,EAA0B,GAAI,GAAI,KAC5D,IAAI,SAAA,GAAK,MAAG,GAAE,KAAI,KAAK,EAAE,KACzB,KAAK,oBAEd,EAAI,SAAc,CAChB,MAAO,8EAIX,YAAK,aAAe,EACpB,KAAK,eAAiB,EACtB,CAAA,EAAO,SAET,EAAA,UAAA,OAAA,WACE,MAAO,CACL,WAAY,GACZ,cAAe,KAAK,cACpB,uBAAwB,KAAK,eAAe,kBAC5C,kBAAmB,KAAK,eAAe,eAInC,EAAA,UAAA,WAAR,WACE,MAAI,GAAA,MAAM,UAAU,iDAAmD,EAC9D,KAAK,MAAM,aAEb,CAAC,QAAS,EAAA,KAAK,MAAO,MAAO,OAG9B,EAAA,UAAA,SAAR,SAAiB,GACf,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,MAAK,MAAM,WACJ,GAER,GAAwB,MAAQ,EAAA,KAAK,MAC/B,IAGK,EAAA,UAAA,aAAd,SAA2B,4EACzB,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,CAAA,EAAO,KAAK,MAAM,uBAAuB,IAErC,GAAa,EACnB,CAAA,EAAO,EAAW,MAAQ,EAAW,eAKvC,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,gBAAgB,IAAI,GAC3B,OAEF,GAAI,KAAK,YAAY,IAAI,IACvB,KAAK,gBAAgB,IAAI,GACzB,KAAK,iBACL,OAGF,GAAI,CAAC,KAAK,QAAQ,IAAI,GACpB,OAGF,KAAK,eAAe,GACb,GAAA,GAAA,KAAA,QAAA,IAAA,GAAA,eACP,AAAI,GAAkB,MACpB,GAAe,KAAK,UACpB,EAAe,KAAK,WAEtB,KAAK,QAAQ,OAAO,IAGd,EAAA,UAAA,eAAR,SAAuB,GACf,GAAA,GAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAE5C,EAAM,GAAS,EAAM,YAAc,EACnC,EAAW,KAAK,aAAa,IAAI,GACvC,AAAI,EAAW,EACb,KAAK,aAAa,IAAI,EAAK,EAAW,GAEtC,MAAK,aAAa,OAAO,GACzB,AAAI,GAAW,MACb,MAAK,eAAiB,KAAK,aAAa,EAAU,GAClD,KAAK,eAAe,eAAe,EAAS,EAAU,EAAO,KAGjE,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,QAAU,KAClB,EAAQ,SAAW,KACnB,EAAQ,SAAW,GACnB,EAAQ,MAAQ,MAGlB,EAAA,UAAA,WAAA,SAAW,GACT,YAAK,YAAY,GACV,KAAK,QAAQ,IAAI,GAAQ,SAOlC,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,MAAK,QAAQ,IAAI,IAGlB,EAAA,UAAA,cAAR,WACE,MAAK,GAAA,MAAM,QAAQ,qBAInB,CAAI,KAAK,YAAc,MACrB,MAAK,WAAa,EAAA,SAAS,YAAY,QAGlC,KAAK,YAPH,MAiBX,EAAA,UAAA,mBAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAa,KAAK,gBACxB,MAAI,CAAC,KAAK,uBAAyB,GAAc,MAC/C,SAAQ,KACJ,6NAKJ,KAAK,sBAAwB,IAGxB,GAAc,MACjB,EAAO,MACH,SAAA,GAAS,MAAA,GAAK,QAAQ,IAAI,EAAM,QAAQ,SAAW,MAC/C,EAAA,KAAK,cAAc,EAAM,OAAS,KAGhD,EAAA,UAAA,gBAAA,WACE,MAAO,MAAK,OAGd,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,GAAM,GAAS,KAAK,WAAW,EAAK,MAAO,aACrC,EAAa,KAAK,QAAQ,IAAI,EAAO,QAI3C,SAAW,eAAiB,CAC1B,KAAM,EAAA,SAAS,KAAK,EAAK,SACzB,KAAM,EAAA,SAAS,KAAK,EAAK,UAGpB,GAET,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAExC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAGxC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAO,EAAM,EAAE,MAChE,EAAE,OACN,MAAO,MAAK,WAAW,EAAM,EAAE,MAAO,GAGxC,GAAI,EAAA,KAAK,cAAc,KAAU,EAC/B,MAAO,GAAA,OAAO,GAAI,EAAM,EAAE,OAErB,GAAA,GAAA,KAAA,QAAA,IAAA,EAAA,QAAA,SACD,EAAc,EAAA,WAAW,iBAAiB,EAAE,MAAO,EAAO,GAChE,GAAI,GAAY,CAAC,GACf,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAmB,GACvB,GAAI,IAAa,GACf,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,GAEhD,YAAK,YAAY,EAAE,QACZ,KAAK,aAAa,EAAG,EAAO,IAG7B,EAAA,UAAA,aAAR,SAAqB,EAAW,EAAiB,GAC/C,GAAM,GAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAI,KAAK,WAAW,EAAM,EAAE,OAC5B,EAAa,KAAK,QAAQ,IAAI,EAAE,QAEtC,OAAO,OAAO,EAAY,GAC1B,EAAW,MAAQ,EACnB,EAAW,MAAQ,EAAE,MACrB,GAAI,GAAa,EAAA,WAAW,kBAAkB,EAAO,EAAE,SACvD,AAAI,EAAS,OAGX,IAAc,EAAS,MAAM,YAE/B,EAAW,MAAQ,CACjB,WAAU,EAEV,WAAY,EAAS,OAAS,EAAS,MAAM,YAAc,EAAE,QAI/D,GAAM,GAAW,KAAK,aAAa,IAAI,EAAW,MAAM,aAAe,EACvE,YAAK,aAAa,IAAI,EAAW,MAAM,WAAY,EAAW,GAEvD,GAGT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GAD1C,GAAA,GAAA,KAEQ,EAAS,KAAK,mBAChB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,aAAa,EAAG,EAAO,EAAK,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAO,GAAI,GAGpB,GAAM,GAAU,GAAI,IAAoB,EAAO,EAAS,GACxD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAqB,EAAE,MAAO,GAClC,GAAI,IAAe,EAAE,MAAO,GAChC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,GAAI,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KAChC,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KACtC,MAAO,GAAA,QAAQ,KAAK,OAAO,EAAO,GAAO,KAAK,OAAO,EAAO,IAG9D,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAEjB,GAAI,EAAQ,OAAS,EAAA,MAAM,UAAU,iCACnC,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,OAAO,EAAQ,MAAM,EAAG,GAAW,GACnD,EAAY,KAAK,OAAO,EAAQ,MAAM,GAAW,GACvD,MAAO,MAAK,OAAO,CAAC,EAAU,GAAY,GAE5C,GAAI,EAAA,MAAM,QAAQ,gCAAkC,EAAQ,GAAG,KAAO,GACpE,GAAM,GAAU,GAAI,IAAoB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACnE,MAAO,MAAK,cAAc,EAAS,GASrC,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACtD,EACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAAK,GAAI,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,OAC3D,EAAU,GAAI,IAAc,EAAU,IAAI,SAAA,GAAK,MAAA,GAAE,SACjD,EAAc,KAAK,cAAc,EAAS,GAChD,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAsB,GAAtB,GAAA,GAAA,KACQ,EAAS,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,IAAI,KACtE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAE/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAM,GAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAA,EAAA,MAAC,EAAA,EAAA,GAIP,GAAK,KAAgB,GAAK,IAAgB,IACtC,EAAY,IACd,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAE1B,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAG1B,GAAM,GAAM,IAAgB,EAAI,EAAI,EAAE,KAAK,EAAO,EAAW,GACvD,EAAO,IAAgB,EAAI,EAAI,EAC/B,EAAM,IAAgB,EAAI,EAAE,KAAK,EAAO,EAAG,GAAa,EAC9D,MAAO,MAAK,SAAS,EAAK,GAAK,IAAI,EAAM,IAG3C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,GAC5D,MAAO,MAAK,cAAwB,EAAS,CAAC,EAAG,GAAI,IAGvD,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE7C,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAA,EAAA,MAAC,EAAA,EAAA,GAED,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,EACxD,EAAS,EAAiB,GACxB,EAAuB,CAAC,EAAG,GACjC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAwB,EAAS,EAAQ,IAGvD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aACd,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OACrD,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OAErD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAExD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAE/C,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,EAGT,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,EAAE,QAG/C,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAQ,EAAM,EAAO,GACnD,GAAI,IAAW,EAAE,MAAO,EAAQ,EAAM,EAAO,GACjD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAM,GACF,GAAI,IAAe,EAAW,MAAO,EAAa,EAAM,EAAO,GACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAY,EAAa,KAG/D,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAI,EAAE,QAAU,UACd,GAAM,GAAO,KAAK,SAAS,EAAE,QACvB,EAAc,EAAK,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,KAC9C,EAAM,EAAA,OAAO,EAAE,MAAO,EAAE,MAAO,GACrC,MAAO,IAAK,EAAK,GAEnB,GAAM,GAAU,GAAI,IAAY,EAAE,MAAO,GACzC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAU,GACxC,GAAI,IAAW,EAAE,MAAO,EAAU,GACtC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAAlD,GAAA,GAAA,KACQ,EAAS,KAAK,mBAChB,CAAC,EAAG,GAAU,WAAM,MAAA,GAAK,WAAW,OAAO,EAAG,EAAS,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAQ,KAAM,GACzD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAEV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAU,EAAE,QAAQ,GAAW,GAC1B,QAAQ,GACR,MAAM,EAAkB,IAGtC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAGV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAAyB,GACzB,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,EAAE,IAAI,GAEhB,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EAAe,EAAA,aAAa,oBAC9B,EAAQ,MAAO,EAAY,EAAM,IAE/B,EAAW,EAAA,UACb,EAAQ,QAAQ,GAChB,GACJ,MAAO,GAAA,QAAQ,EAAU,IAGnB,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,KAAK,KAAK,EAAS,GAC7B,EAAa,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,QAAO,GACpD,EAAU,GAAI,IAAc,EAAY,GACxC,EAAS,KAAK,cAAwB,EAAS,CAAC,GAAI,GAE1D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,OAAO,EAAQ,EAAY,IAGjC,EAAA,UAAA,UAAR,SACI,EAAa,EACb,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACrB,AAAI,GAAgB,MAClB,GAAY,EAAa,MAAM,GAC/B,EAAS,EAAa,MAAM,IAE9B,GAAM,GAAa,EAAA,aAAa,yBAAyB,GACnD,EAAa,CACjB,WAAU,EACV,OAAM,EACN,UAAS,EACT,QAAS,KAAK,KAAK,EAAS,IAExB,EACF,GAAI,IAAiB,EAAY,EAAY,GAAgB,MAC3D,EAAS,CAAC,GAChB,AAAI,GAAgB,MAClB,EAAO,KAAK,GAEd,GAAM,GAAS,KAAK,cAAwB,EAAS,EAAQ,SAE7D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,UAAU,EAAG,EAAY,IAG/B,EAAA,UAAA,gBAAR,SACI,EAAW,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MACtC,GAAM,GAAU,GAAgB,KAAO,EAAa,MAAQ,EAAE,MACxD,EAAS,EAAQ,EAAQ,OAAS,GAClC,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,GAAI,IAChB,EAAS,EAAY,EAAY,GAAgB,MAC/C,EAAS,GAAgB,KAAO,CAAC,GAAK,CAAC,EAAG,GAC1C,EAAS,KAAK,cAAsB,EAAS,EAAQ,SAC3D,MAAI,GAAO,OAAS,EAAE,KACb,KAAK,gBAAgB,EAAG,EAAY,GAEtC,GAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,MAAO,GAAa,QAAQ,IAGtD,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAC/D,GAAI,EACF,MAAO,GAGH,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,OAAQ,GAAa,QAAQ,IAGvD,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAI,GAAO,EACL,EAAc,EAAA,aAAa,mBAAmB,CAAC,GAAO,EAAE,MAC1D,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,EAAA,UAAU,EAAG,GACzB,EAAO,EAAA,aAAa,iBAAiB,EAAG,EAAE,MAAM,IAGlD,GAAM,GACF,GAAa,gBAAgB,EAAU,MAAO,EAAM,GAClD,EAAS,EAAA,KAAK,cAAc,CAAC,EAAU,MAAM,KAC7C,EAAM,EAAU,KAAK,GAAI,GACzB,EAAc,EAAA,WAAc,EAAE,OAChC,EACA,KAAK,aACG,EAAK,qBAAsB,EAAY,EAAa,GACvD,QAAQ,GACjB,MAAI,IAAe,MACjB,GACI,EAAA,UAAU,EAAQ,EAAA,aAAa,uBAAuB,KAErD,GAGD,EAAA,UAAA,aAAR,SACI,EAAa,EAAiC,EAC9C,EAAiB,GACnB,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EACF,GAAa,8BAA8B,EAAQ,GACjD,EAAY,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,YAAW,GACvD,EAAU,GAAI,IAAiB,EAAW,GAC1C,EACF,KAAK,cAAwB,EAAS,CAAC,EAAG,GAAa,GAE3D,MAAI,GAAO,MAAM,KAAO,EACf,EAET,GAAa,EAAA,MAAM,EAAG,GAAa,KAAK,CAAC,EAAS,IAC3C,KAAK,aAAa,EAAQ,EAAW,EAAY,EAAO,KAGzD,EAAA,UAAA,gBAAR,SAAwB,EAAW,EAAc,GAE/C,GAAM,GAAO,CAAC,GAId,GAHA,EAAA,aAAa,2BACT,MAAQ,EAAW,OAAO,GAAG,cAAgB,EAAW,MAAM,GAAI,EAClE,EAAE,MACF,CAAC,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAQ,GAC7C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,UAAU,EAAK,GAAY,QAAQ,GAEjD,MAAO,MAAK,gBAAgB,EAAG,IAGjC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,GAAI,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,kDAAkD,GAAE,KAAO,GAAC,IAC5D,iBAAgB,IAMtB,OAJM,GAAO,EAAE,MAAM,GACjB,EAAS,EAGJ,EAAI,EAAG,GAAK,KAAK,KAAK,KAAK,KAAK,IAAS,EAAG,KACnD,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,GAAO,GAC5C,EAAc,EAAQ,mBAAmB,GACzC,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,GAAS,EAAO,MAAO,GAC7D,EAAW,UAIb,GAAI,GACF,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAW,GAChD,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,IACtC,EAAW,UAGb,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA2B,QAE9D,GAAM,GAAU,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAE,OACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA+B,QAElE,GAAM,GACF,GAAI,IAAgB,GAAwB,EAAE,MAAO,EAAE,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAClE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA0B,QAG7D,GAAM,GAAU,GAAI,IAAgB,GAAmB,EAAE,MAAO,EAAE,OAClE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA6B,QAGhE,GAAM,GAAU,GAAI,IAAgB,GAAsB,EAAE,MAAO,EAAE,OACrE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAAmC,QAE/C,GAAM,GACF,GAAI,IAAgB,GAA4B,EAAE,MAAO,EAAE,OAC/D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAiC,QAEpE,GAAM,GACF,GAAI,IAAgB,GAA0B,EAAE,MAAO,EAAE,OAC7D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAM,GAAU,GAAI,IAAc,EAAU,KAAM,EAAE,MAAO,EAAE,MAC7D,MAAO,MAAK,cACR,EAAS,CAAC,EAAW,EAAG,GAAI,EAAA,WAAW,EAAE,MAAO,EAAE,SAGxD,EAAA,UAAA,MAAA,SAAM,GACJ,EAAA,aAAa,KACT,yEAEJ,GAAM,GAAW,EAAU,WAC3B,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAM,GAAQ,EAAE,WAChB,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAK,GACL,EAAc,QACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAA6B,GAEzC,GAAM,GAAU,GAAI,IAAgB,EAAI,EAAE,MAAO,EAAE,OACnD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAG1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,GAErD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAG7C,EAAA,UAAA,cAAR,SAAsB,EAAe,EAAY,GAC/C,GAAM,GAAU,GAAI,IAAqB,EAAE,MAAO,GAClD,MAAO,MAAK,cAAsB,EAAS,CAAC,GAAI,IAG1C,EAAA,UAAA,eAAR,SACI,EAAe,EAAe,EAAY,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GACF,GAAI,IAAsB,EAAI,EAAE,MAAO,EAAE,MAAO,GACpD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAO7C,EAAA,UAAA,yBAAR,SAAiC,EAAW,EAAW,GAAvD,GAAA,GAAA,KACQ,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAA,4SAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAcP,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,GAMD,EAAA,UAAA,+BAAR,SACI,EAAuB,GACzB,MAAO,CACL,OAAQ,EAAY,OACpB,MAAO,EAAY,MACnB,MAAO,EAAc,QAIzB,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAIjB,GAAI,EAAQ,OAAS,EAAA,MAAM,IAAI,iCAC7B,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,KAAK,EAAQ,MAAM,EAAG,IACtC,EAAY,KAAK,KAAK,EAAQ,MAAM,IAC1C,MAAO,MAAK,KAAK,CAAC,EAAU,IAG9B,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAO,OAAO,SAAC,EAAI,GAAO,MAAA,GAAA,WAAW,EAAI,KAC1D,EAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAE5B,EAAc,EAAA,MAAM,QAAQ,cAC5B,EAAU,EACZ,GAAI,IAAkB,EAAQ,GAAG,MAAO,GACxC,GAAI,IAAY,EAAQ,GAAG,MAAO,GACtC,MAAO,MAAK,cAAiB,EAAS,EAAS,IAGjD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,GAAM,GAAc,EAAA,MAAM,QAAQ,gCAC5B,EAAU,EACZ,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OAC/C,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,MAAO,MAAK,cAAiB,EAAS,CAAC,EAAG,GAAI,IAGhD,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAY,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACnE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAe,EAAE,OAGhD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAG1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAChD,EAAI,KAAK,SAAS,EAAQ,EAAS,QAAQ,IAC3C,EAAI,KAAK,IAAI,GACb,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAI,EAAG,IAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAGtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAE3C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA2B,EAAE,MAAO,EAAM,OAC9C,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAM,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAEtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA6B,EAAG,MAAO,EAAE,OAC7C,GAAI,IAAgB,GAAsB,EAAG,MAAO,EAAE,OAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,UAG1C,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,GAAI,GACJ,AAAI,EAAA,MAAM,QAAQ,mBAChB,EAAU,GAAI,IAAkB,EAAE,OAElC,EAAU,GAAI,IAAY,EAAE,OAE9B,GAAM,GAAc,EAAQ,mBAAmB,EAAK,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,IAGhD,EAAA,UAAA,IAAA,SAAsB,GAEpB,GAAI,KAAK,mBAAmB,CAAC,KAAO,EAAE,QAAU,aAC9C,GAAM,GACF,GAAiB,KAAK,QAAQ,IAAI,EAAE,QAAQ,QAChD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAU,GAAI,IAAkB,EAAE,OAClC,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAG9D,MAAO,MAAK,cAAsB,EAAS,IAG7C,EAAA,UAAA,QAAA,SAA0B,GACxB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,GAAc,IAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,KAG9B,EAAA,UAAA,eAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAGF,GAAM,GAAS,EAAE,MACX,EAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAkB,EAAS,WAC3B,EAAc,EAAO,GAAK,EAAO,GAAK,EAAO,GAC7C,EAAmB,EAAS,YAC5B,EAAiB,EAAS,aAAe,eACzC,EAAa,GACb,EAAa,GAIb,EACD,KAAgB,GAAK,IAAqB,IAC3C,EAAkB,GAChB,EAAyB,EAAO,GAAK,IAAM,GAAK,CAAC,CAAC,EAAS,SAEjE,GAAI,GAA6B,CAAC,EAAA,MAAM,QAAQ,wBAC5C,CAAC,EAAA,MAAM,QAAQ,iCACf,CAAC,GACH,GAAM,GAAc,EAAiB,EAAO,GAAK,EAAO,GAAK,EAAO,GAC/B,EAAO,GAAK,EAAO,GAAK,EAAO,GAC9D,EAAY,EAAA,QAAQ,EAAG,CAAC,EAAG,EAAa,EAAS,aACjD,EACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAS,KAAK,iBAAiB,CACnC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAExB,MAAO,GAAA,QAAQ,EAAQ,EAAS,UAWlC,GAAM,GAAc,EAChB,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACrC,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACnC,EAAwB,CAC5B,OAAQ,EAAE,OACV,MAAO,CAAC,EAAG,EAAa,EAAS,YACjC,MAAO,EAAE,OAUL,EAAwB,EAAS,MACvC,EAAS,MAAQ,EAAS,MAAM,QAChC,EAAS,MAAM,EAAS,MAAM,OAAS,KACvC,EAAA,KAAK,OACD,GAAyB,EAAS,MAAO,EAAU,OACnD,WAAM,MAAA,kBAAkB,EAAS,MAAK,OAClC,EAAU,MAAK,gBACvB,GAAM,GACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAgB,KAAK,iBAAiB,CAC1C,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAElB,EAAuB,KAAK,QAAQ,IAAI,EAAc,QAC5D,SAAA,KAAK,OACD,EAAqB,SACrB,WAAM,MAAA,gDAEV,EAAS,MAAQ,EAGjB,EAAqB,MAAQ,EAAS,SAC/B,EAAA,SAAS,qBACL,EAAc,OAAQ,EAAS,SAAU,EAAc,QAI5D,EAAA,UAAA,iBAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAQA,GAAA,GAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,WAGI,EAAiB,IAAe,eAEhC,EAAY,EAAc,EAAe,EACzC,EAAU,EAAY,EACtB,EAAa,CAAC,EAAW,GACzB,EAAa,GACb,EAAa,GAEb,EAAY,EAAE,QAAQ,CAAC,IACvB,EAAQ,EAAO,QAAQ,CAAC,EAAG,EAAW,KAEtC,EACF,GAAI,IAAoB,EAAY,EAAU,MAAO,GACnD,EACF,KAAK,cAAwB,EAAe,CAAC,IAAY,QAAQ,CAC/D,EAAG,EAAW,GAAI,EAAW,KAG7B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAgB,GAAI,IACtB,EAAO,MAAO,CAAC,EAAG,EAAS,EAAS,aAAc,EAClD,EAAY,EAAS,EAAiB,GACpC,EAAuB,CAAC,EAAQ,GACtC,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEd,GAAM,GAAU,KAAK,cAAwB,EAAe,GAE5D,MAAI,GACK,EAAQ,QAAQ,CAAC,EAAG,EAAW,EAAU,EAAS,cAElD,EAAQ,QAAQ,CAAC,EAAG,EAAS,YAAa,EAAW,KAIhE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAEjD,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAM,MAAM,KAAO,EAC3D,MAAO,MAAK,iBACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAGjD,GAAM,GAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAS,KAC7D,EAAU,GAAI,IAChB,EAAU,EAAS,EAAiB,GAClC,EAAuB,CAAC,EAAO,GACrC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAc,EAAS,IAGrC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eAAe,EAAG,EAAQ,GAExC,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAM,KAAO,EACvD,MAAO,MAAK,iBAAiB,EAAG,EAAQ,GAE1C,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAExC,EAA0B,EAAA,MAAM,QAAQ,6BAC1C,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EAC7C,EAAkB,EACpB,GAA6B,EAAY,GACzC,KACE,EAAmB,CAAC,EAAO,GAE3B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KAC5D,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAGd,GAAI,GACJ,MAAI,GACF,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,IAGrC,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,KAGrC,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,6BACd,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EACjD,GAAU,GAAI,IAA6B,GACpC,KAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,GAAU,GAAI,IAAuB,GAC9B,KAAK,cAAc,EAAS,CAAC,EAAG,MAGzC,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAA+B,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAM,GAAU,GAAI,IAAgC,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAA,aAAa,WAAW,EAAG,EAAO,OAG3C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,KAAK,MAAM,EAAG,EAAO,GAAM,QAAQ,GAE9C,MAAO,IAGT,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GAA2B,GAAI,IAAyB,GAC9D,MAAO,MAAK,cAAc,EAA0B,CAAC,GAAK,EAAE,QAG9D,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAM,GAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,KAAK,cAAc,EAA2B,CAAC,IAC7C,EAA2B,GAAI,IAAyB,GACxD,EAAS,KAAK,cAChB,EAA0B,CAAC,EAAI,GAAqB,EAAE,OAC1D,SAAmB,UACZ,GAGT,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IACA,EAAE,MAAO,EAAW,EAAU,GAClC,GAAI,IAAsB,EAAE,MAAO,EAAW,EAAU,GAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,GAAM,GAAU,GAAI,IAA8B,EAAI,EAAG,GAEzD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,GAAI,IAChB,EAAE,MAAO,EAAW,EAAU,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GACF,GAAI,IAAoC,EAAI,EAAG,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAM,GAAQ,EAAa,EAAS,EAAA,QAAQ,GACtC,EAAY,EAAM,MAAM,GACxB,EAAc,EAAM,MAAM,GAC1B,EAAU,GAAI,IAAmB,EAAW,EAAa,GACzD,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAQ,QAAS,IAGvD,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAM,GAAU,GAAI,IAAc,EAAQ,KAAM,EAAO,EAAS,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAAI,IAAY,EAAE,MAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,GAAM,GAAU,GAAI,IAChB,EAAM,MAAO,EAAM,MAAO,EAAU,EAAQ,GAChD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAO,EAAO,GAAW,YAG/D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAE9D,GAAM,GAAY,EAAE,MAAM,GACpB,EAAe,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAE5D,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAe,IAAe,OAChC,CAAC,EAAW,EAAc,EAAa,GACvC,CAAC,EAAW,EAAa,EAAc,GAErC,EAAU,GAAI,IAAoB,EAAa,EAAW,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAG5C,EAAe,CAAC,EAAa,EAAW,GACxC,EAAiB,EAAQ,QAAQ,CAAC,EAAY,IAC9C,EAAW,EAAQ,QAAQ,CAAC,EAAY,IAE9C,GAAI,IAAe,EACjB,MAAO,GAAA,aAAa,cAAc,EAAA,OAAO,IAAK,GAEhD,GAAM,GAAe,EAAA,OAAO,GACtB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAe,KAAM,EAAS,KAAM,EAC3D,GACE,EACF,KAAK,cAAc,EAAS,CAAC,EAAU,EAAgB,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAGjC,EAAiB,GACjB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAc,KAAM,EAAa,KAAM,EAC9D,CAAC,EAAY,GAAI,GACf,EAAc,KAAK,cACrB,EAAS,CAAC,EAAc,EAAe,IAC3C,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGzB,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGjB,EAAA,UAAA,QAAR,SAAgB,EAAa,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAGxD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAU,KAAK,QAAQ,EAAM,GAAM,KAAK,EAAE,MAAM,GAAI,EAAE,MAAM,IAClE,SAAK,UACL,EAAK,UACE,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAGpC,EAAiB,EAAQ,QAAQ,CAAC,EAAW,IAC7C,EAAW,EAAE,QAAQ,CAAC,EAAE,KAAO,EAAW,IAC1C,EACF,GAAI,IAAgB,EAAW,EAAS,CAAC,EAAW,IAClD,EAAc,KAAK,cAAc,EAAS,CAAC,EAAU,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAG5C,GAFA,EAAQ,GAAS,EAAA,KAAK,WAAW,GAE7B,IAAU,UAEZ,GAAM,GAAS,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IAChE,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,WAEjD,GAAM,GAAU,GAAI,IAAY,EAAO,GACjC,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,GAAI,EAAO,KAIlD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAIhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAE,QAAU,SAAW,GAAK,EAAG,EAAE,QAG7D,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GAEpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGhD,EAAA,UAAA,eAAA,SAAe,EAAiB,EAAiB,GAE/C,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,YAAK,QAAQ,IAAI,GAAQ,MAAQ,KAC1B,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAGtB,EAAA,UAAA,WAAR,SACI,EAAiB,EAAiB,GAC7B,GAAA,GAAA,KAAA,eAAA,EAAA,EAAA,GAAA,OACP,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAGrD,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAU,GAAI,IAAc,EAAM,OACxC,MAAO,MAAK,gBAAgB,EAAS,CAAC,GAAQ,EAAM,QAG9C,EAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAU,GAAI,IAAY,EAAM,OAChC,EAA8B,GACpC,MAAO,MAAK,gBACR,EAAS,CAAC,GAAQ,EAAM,MAAO,KAC/B,IAGE,EAAA,UAAA,cAAR,SAAsB,EAAmB,GACvC,GAAM,GAAe,CACnB,GAAuB,EAAM,eAC1B,GAAuB,EAAM,QAE5B,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EAAiB,CACrB,GAAuB,WAAgB,GAAuB,IAG1D,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,KAAK,gBAChB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,QAG1D,EAAA,UAAA,OAAR,SAAe,GACb,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,MAClB,EACF,GAAwB,GACxB,EACJ,AAAI,EACF,EAAU,GAAI,IAA0B,GAExC,EAAU,GAAI,IAAoB,GAEpC,GAAM,GAAgC,GAChC,EAAM,KAAK,gBACb,EAAS,CAAC,CAAC,MAAO,EAAW,MAAK,EAAE,OAAM,IAAI,EAC9C,KAAwB,GAC5B,MAAO,CAAC,MAAK,EAAE,MAAK,EAAE,OAAQ,EAAI,SAGpC,EAAA,UAAA,gBAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAHJ,GAAA,GAAA,KAGI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,KAAK,eAAe,EAAQ,YAAa,GAClD,EAAU,KAAK,QAAQ,IAAI,EAAO,QAIxC,GAHA,AAAI,EAAQ,cACV,GAAQ,SAAW,IAEjB,EAAQ,mBAAqB,GAAuB,OACtD,GAAM,GAAa,GAA0B,EAAQ,aAKrD,EAAQ,SAAW,EAAW,IAAI,SAAA,GAAK,MAAA,GAAI,IAK7C,GAHA,AAAI,EAAQ,aAAe,MACzB,GAAQ,MAAQ,EAAQ,aAEtB,EAAA,KAAK,cAAc,EAAO,SAAW,EAGvC,SAAQ,OACJ,EAAA,KAAK,uBAAuB,EAAO,MAAoB,GACpD,EAGT,GAAM,GAA8B,GAC9B,EAA2B,EAAO,IAAI,SAAA,GAC1C,GAAI,EAAM,QAAU,YAClB,KAAM,IAAI,OACN,mIAKN,GAAI,GAAU,EAAK,QAAQ,IAAI,EAAM,QAErC,GAAI,EAAQ,SAAW,MACrB,GAAI,CAAC,EAAQ,cACT,EAAA,KAAK,cAAc,EAAM,QACrB,EAAA,MAAM,UAAU,6BAMtB,MAAO,CACL,MAAO,EAAM,MACb,QAAS,KACT,UAAW,GACX,cAAe,EAAQ,QAM3B,AAAI,EAAQ,cACV,GAAQ,SAAW,GACnB,EAAQ,MAAQ,EAAM,eAEf,CAAC,CAAC,EAAQ,WAAa,CAAC,CAAC,EAAQ,aAC1C,EAAQ,EAAQ,SAAW,EAAK,aAAa,GAClB,EAAK,WAAW,GAC3C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,gBAE/B,EAAQ,UACR,CAAC,GAAyB,EAAQ,MAAO,EAAM,QAQjD,GAAM,GAAa,EACb,EAAc,EAAM,MAE1B,EAAM,MAAQ,EAAQ,MACtB,EAAQ,EAAK,cAAc,EAAiB,GAC5C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,QAEjC,EAAW,MAAQ,EAGrB,SAAK,YAAY,EAAM,QAChB,CAAC,MAAO,EAAM,MAAO,QAAO,EAAE,UAAW,MAGlD,KAAK,YAAY,EAAO,QACxB,GAAM,GACW,CAAC,MAAO,EAAO,MAAO,QAAS,EAAS,UAAW,IAC9D,EAAM,GAAyB,EAAS,EAAY,GACpD,EAAS,KAAK,iBAAiB,EAAK,WACxC,MAAO,IACH,EAAK,MAAO,EAAS,EAAY,KAEjC,EAAoB,KAAK,cAAgB,KAC3C,EAgBJ,GAfA,AAAI,GACF,GAAQ,KAAK,cAGf,GACI,KAAK,MAAO,EAAQ,EAAY,EAAY,GAEhD,EAAc,QAAQ,SAAA,GAAQ,MAAA,GAAK,8BAA8B,KAEjE,AAAI,GACF,GAAQ,KAAK,SAAS,GACtB,KAAK,aAAa,KACd,CAAC,KAAM,EAAQ,YAAY,KAAM,MAAO,KAAK,aAAa,MAG5D,CAAC,EAAA,MAAM,QAAQ,wBAA0B,EAAQ,UACjD,IAAkC,IACpC,GAAM,GAAW,KAAK,aAAa,GACnC,YAAK,8BAA8B,GAC5B,EAET,MAAO,IAGT,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,EAAc,GAAe,EAAO,GAAG,MACvC,GAAM,GAAU,KAAK,gBACjB,EAAS,EAAQ,EAAa,EAC9B,GACJ,MAAO,GAAA,SAAS,qBACL,EAAQ,OAAQ,EAAQ,MAAO,EAAQ,QAG5C,EAAA,UAAA,iBAAR,SAAyB,EAAa,GAEpC,MAAM,KAAO,MAAK,aAChB,MAAK,YAAY,GAAO,KAEnB,KAAK,YAAY,IAG1B,EAAA,UAAA,kBAAA,WACE,MAAO,MAAK,gBAKd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAIF,GAAI,CAAC,EAAA,MAAM,QAAQ,YACjB,GAAM,GAAU,OAAO,KAAK,KAAK,aACjC,EAAQ,QAAQ,SAAA,GACd,EAAK,MAAM,cAAc,EAAK,YAAY,GAAK,cAC/C,MAAO,GAAK,YAAY,KAG5B,KAAK,eAAe,UACpB,AAAI,KAAK,QAAU,MACd,MAAQ,oBAAuB,aAC/B,KAAK,iBAAkB,mBAC1B,KAAK,OAAO,SAEZ,KAAK,OAAS,KAEhB,AAAI,KAAK,qBACP,MAAK,MAAM,QAAU,KACrB,KAAK,MAAM,WAEb,KAAK,SAAW,IAGlB,EAAA,UAAA,eAAA,WAAA,GAAA,GAAA,KACE,MAAI,MAAK,qBAAuB,MAC9B,MAAK,oBAAsB,EAAA,KAAK,WAC9B,GAAI,CAAC,EAAA,MAAM,IAAI,iCAGb,GAAM,GAAY,EAAA,MAAM,QAAQ,SAChC,EAAA,MAAM,IAAI,QAAS,IACnB,GAAM,GAAsB,EAAK,IAAI,EAAA,OAAO,OAAO,WAAW,GAG9D,GAFA,EAAA,MAAM,IAAI,QAAS,GAEf,EAAsB,EACxB,MAAO,IAGX,MAAO,OAGJ,KAAK,qBAGd,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAGlD,EAAA,UAAA,YAAR,SAAoB,SACZ,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAE7C,GAAI,GAAW,KAEb,OAEF,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GAAW,EAAQ,SAMvB,GALA,AAAI,GAAY,MACd,GAAW,GAA2C,EAAO,GAC7D,EAAQ,SAAW,GAGjB,GAAU,MACZ,GAAM,GAAY,GAAwB,GAEtC,EAAO,OACP,EAAQ,EAAS,GAAI,EAAS,EAAS,GACrC,EAAc,YAAkB,YAEtC,AAAI,EACF,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,EAAU,GAAI,IACV,EAAW,CAAC,EAAQ,GAAQ,IAEhC,EACI,GAAI,IAAoB,EAAW,CAAC,EAAQ,GAAQ,GAG1D,GAAM,GAAuB,KAAK,eAAe,CAAC,EAAQ,GAAQ,GAClE,AAAI,EACF,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEjB,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEnB,KAAK,MAAM,2BACP,KAAK,WAAW,EAAqB,QAAS,EAAO,EACrD,GAIJ,GAAM,GAAwB,GACxB,EAAsB,KAAK,gBAC7B,EAAS,CAAC,GAAuB,EAAO,KAAM,GAG5C,EAAgB,KAAK,QAAQ,IAAI,EAAoB,QAC3D,EAAQ,QAAU,EAAc,QAChC,EAAQ,SAAW,EAAc,SACjC,EAAQ,SAAW,EAAc,SACjC,EAAQ,MAAQ,EAAc,MAE9B,KAAK,8BAA8B,GACnC,KAAK,QAAQ,OAAO,EAAoB,QAGxC,EAAQ,OAAS,KACjB,AAAI,GACF,MAAK,cAAgB,EAAA,KAAK,MAAQ,QAGpC,GAAM,GAAa,KAAK,eAAe,EAAU,EAAO,EAAO,GAC/D,EAAQ,QAAU,IAId,EAAA,UAAA,qBAAR,SAA6B,EAAgB,GAE3C,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAEP,YAAK,eAAe,GAEpB,AAAI,GAAiB,MACnB,GAAQ,OAAS,GAAoB,EAAe,IAE/C,EAAQ,QAGT,EAAA,UAAA,eAAR,SACI,EAA4B,EAAuB,EACnD,GAEF,GADA,KAAK,eAAiB,KAAK,aAAa,EAAU,GAC9C,CAAC,KAAK,mBACN,KAAK,cAAgB,KAAK,mBAAqB,KAAO,MACxD,GAAM,GAAM,MAAK,cAAgB,KAAO,MAAM,QAAQ,GACtD,KAAK,kBAAoB,GACzB,QAAQ,KACJ,6BAA6B,EAAE,yCAGrC,MAAO,MAAK,eAAe,eAAe,EAAU,EAAS,IAGvD,EAAA,UAAA,aAAR,SAAqB,EAAyB,GAC5C,MAAO,GAAM,GAAK,EAAM,GAAK,EAAA,KAAK,gBAAgB,IAG5C,EAAA,UAAA,mBAAR,SACI,EAAsB,GACxB,GAAI,KAAK,mBAAmB,GAC1B,IACE,MAAO,WACA,GACP,GAAI,EAAA,MAAM,QAAQ,WAChB,KAAM,IAAI,OAAM,yBAItB,MAAO,OAEX,GAriFsC,EAAA,eAuiFtC,YACI,EAAiB,GACnB,GAAI,IAAU,WAAa,IAAU,YACnC,MAAO,GACF,GAAI,IAAU,SAAW,IAAU,QAGxC,OAFM,GAAU,IAAU,QAAW,GAAI,YAAW,EAAE,QACjB,GAAI,YAAW,EAAE,QAC7C,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,KAAK,MAAM,EAAE,IAE3B,MAAO,OAEP,MAAM,IAAI,OAAM,iBAAiB,GChwFrC,GAGM,IAAU,QCHhB,cAmCE,EAAA,MAAM,IAAI,2BAA4B,ICnCxC,AAuBA,AAAI,EAAA,YAAY,aACd,EAAA,gBAAgB,QAAS,WAAM,MAAA,IAAI,KAAoB,MAQ5C,IAAQ,CAAC,eAAc,ICzBvB,GAA0B,0BAE1B,GAA2B;;;EAK3B,GAAkC;;;;;cAWf,GAC9B,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EACD,EAAe,EACf,EAAU,GAAI,IAAe,EAAE,MAAO,GAC5C,MAAO,GAAa,gBAAgB,EAAS,CAAC,GAAI,EAAE,oBAepD,EAAmB,EACnB,EAA4C,GAG9C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAe,EACf,EAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,EAAiB,EAAE,MAAO,EAAE,MAC5B,CAAC,CAAC,GACN,GAAI,IAAgB,EAAW,EAAE,MAAO,EAAE,OACxC,EAAS,GAAS,EAAE,MACpB,EAAS,EAAa,gBAAgB,EAAS,CAAC,EAAG,GAAI,GAC7D,MAAO,IC3DX,AAsBA,GAAM,IAAQ,GAA2B;;EAInC,GAAe;;;IAIjB,GAAkC;;EAIzB,GAAkB,GAAiB,GAAO,IAE1C,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,QACb,WAAY,ICvCd,YAqBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,QACb,WAAY,ICjCd,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,WAG/C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBACvB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAyB,GAAI,IAAyB,GAC5D,MAAO,GAAQ,gBAAgB,EAAwB,CAAC,GAAK,EAAE,OAG1D,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,IC3Cd,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,YAAwB,GAOtB,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,MACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,MACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;yBAKK,EAAa;wBACd,EAAY;2DACuB,EAAe;;;MAK1E,MAAA,MC5DA,AAoBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,YACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,YACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;wBAEI,EAAa;uBACd,EAAY;;;;;;yDAMsB,EAAe;;;;MAMxE,MAAA,MChEA,AAuBO,GAAM,IAIM,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,EAAA,EAAG,EAAA,EAAA,KAAM,EAAA,EAAA,SAAU,EAAA,EAAA,OAAQ,EAAA,EAAA,MAElC,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGL,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAGpB,GAAM,GAAc,CAAC,EAAG,EAAM,GAE1B,EAAc,KAClB,AAAI,GAAU,MACZ,GAAc,EAAO,MACrB,EAAY,KAAK,IAGnB,GAAI,GAAa,KACjB,AAAI,GAAS,MACX,GAAa,EAAM,MACnB,EAAY,KAAK,IAGnB,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACJ,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACF,EACF,EAAQ,gBAAgB,EAAS,EAAa,EAAY,GAAG,OAEjE,MAAO,IAGI,GAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,QACb,WAAY,IC/Ed,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAuBA,GAAM,IAAM;;;;eAQN,GAAa;;;;;;;;;;;;;;;;;;EAoBN,GAAgB,GACzB,GAAK,GAAY,IAER,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,ICzDd,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAItB,GAAM,GAAa,EAAW,GAC9B,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;yBAKK,EAAU;;uCAEI,EAAU;;;;;;;MASjD,MAAA,MC5CA,AAuBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,MACD,EAAe,EAEf,EAAU,GAAI,IAAsB,EAAmB,OACvD,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KChCX,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;uDAMmC,EAAK,OAAO,EAAM;;wBAEjD,EAAK,UAAS;;;;;;;;;;;;;;MAgBtC,MAAA,MCrDA,AAoBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;;8BAeU,EAAK,OAAO,EAAM;4BACpB,EAAK,UAAS;;;;;;;;;;;;;;;;UAgBhC,EAAK,OAAM;;MAIrB,MAAA,MCnEA,AA0BO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,QACb,WAAY,IAGV,GAEJ,YAAoB,GAKX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACnB,EAAA,EAAA,OACE,EAAA,EAAA,YAED,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAOR,EAA6B,CAAC,EAAQ,GACtC,EAAW,CAAC,EAAQ,EAAO,GAEjC,AAAI,IAAW,IACb,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAGpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA+C,EAAG,EAAG,EAAO,GAChE,EAAS,GAAoB,QAG/B,GAAM,GAAkB,EAAQ,eAAe,EAAU,SAEzD,EAAQ,QAAQ,IAAI,EAAgB,QAAQ,MAAQ,GAAa,OACjE,EAAQ,MAAM,yBACV,EAAQ,WAAW,EAAgB,QAAS,GAChD,GAAM,GAAU,EAAA,MAAM,QAAQ,cAC1B,GAAI,IAAwB,GAC5B,GAAI,IAAkB,GACpB,EAAM,EAAQ,gBAAgB,EAAS,CAAC,GAAkB,SAChE,SAAQ,YAAY,EAAgB,QAC7B,EC/ET,AA0BA,YAA4B,GAI1B,OAFM,GAAS,GAER,EAAO,SAAW,GAAK,EAAO,EAAO,OAAS,GAAG,UAAY,IAClE,GAAM,GACF,EAAO,OAAS,EAAO,EAAO,OAAS,GAAG,QAAU,EAAQ,GAC1D,EAAa,EAAA,aAAa,yBAAyB,GACzD,EAAO,KAAK,CACV,OAAQ,EACR,WAAU,EACV,QAAS,KAAK,KAAK,EAAU,KAIjC,MAAO,eAIL,EAAe,EAAiB,EAChC,GAIF,OAHM,GAAkB,GAAmB,EAAE,OAEzC,EAAS,EACJ,EAAI,EAAG,EAAI,EAAgB,OAAQ,KACpC,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,WAAY,EAAA,EAAA,QAErB,EAAU,GAAI,IAChB,CAAC,WAAU,EAAE,OAAM,EAAE,UAAW,EAAE,MAAM,GAAI,QAAO,GAAG,GACpD,EAAiB,EACvB,EAAS,EAAQ,gBAAgB,EAAS,CAAC,GAAS,GAEpD,AAAI,EAAe,SAAW,EAAE,QAC9B,EAAQ,YAAY,EAAe,QAIvC,MAAO,GC/DT,YAwBI,EAAmB,EACnB,GACF,GAAM,GACF,CAAC,GAAY,EAAM,QAAM,OACrB,GAAY,EAAM,QACpB,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EACF,CAAC,GAAY,IAAW,OACpB,GAAY,IAEd,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,EAAQ,gBACnB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,OC3ClE,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MACD,EAAe,EAEf,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QAC5C,MAAI,GAAS,UAAY,CAAC,GAAc,EAAE,MAAO,IAC7C,CAAE,GAAS,UAAY,MAAQ,GAAc,EAAS,MAAO,IACxD,GAAc,EAAG,EAAQ,GAGlC,GAAa,OAAO,EAAE,QAEf,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,QAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICzDd,YAwBI,EAAe,EAAuB,EACtC,GACF,GAAM,GAAS,EAAA,KAAK,cAAc,GAC5B,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAY,EAAQ,EACpB,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAC,GAAG,MAAO,CAAC,MAAO,CAAC,EAAW,IAAU,QAAO,IAEhE,EAAU,GAAO,EAAe,EAAE,MAAO,MAAO,GAChD,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAU,MAAO,CAAC,MAAO,GAAW,QAAO,IAEpE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,ECvCT,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAEjC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAW,GAAkB,GAEnC,KAAK,SAAW;;QAEZ,EAAK;uBACU,EAAQ;;MAI/B,MAAA,MAEA,YAA2B,GACzB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,sBAAsB,EAAI,yBAKxC,OAHM,GACF,CAAC,UAAW,UAAW,UAAW,UAAW,UAAW,WACtD,EAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAe,EAAO,IAAM,EAAc,GAE5C,MAAO,GAAe,OCxDxB,AAqBA,GAAA,IAAA,WAQE,WAAY,EAAkB,GAP9B,KAAA,cAAgB,CAAC,KAIjB,KAAA,aAAe,GACf,KAAA,aAAe,GAIb,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAIjC,GAFA,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACpB,KAAK,KAAO,EACd,KAAM,OACF,6BAA6B,KAAK,KAAI,0BAM5C,OAJM,GAAQ,GAAkB,KAAK,MAE/B,EAAc,GAAe,KAAM,KAAK,MACxC,EAAgB,GAAI,OAAM,KAAK,MAC5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAc,EAAO,IAAM,EAAY,GAEzC,GAAM,GAAY,QAAQ,EAAc,MAAM,IAAI,OAAM,IAClD,EACF,KAAK,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAC3D,EAAO,mBAAmB,EAAc,OAAM,MAAM,EAAS,IAEnE,KAAK,SAAW;;QAEZ,EAAK;;oBAEO,EAAI;WACb,EAAU;sBACC,EAAI;;UAEhB,EAAY,KAAK,KAAO,GAAE;aACvB,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAAE;sBACjD,EAAI;aACb,EAAU;wBACC,EAAI;;;;;MAO5B,MAAA,MCvEA,YAyBI,EAAe,EAAgB,GACjC,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAuB,EAAE,MAAO,GACpC,GAAI,IAAiB,EAAE,MAAO,GAClC,MAAO,GAAQ,gBAAgB,EAAS,CAAC,GAAI,EAAE,OC7BjD,AA0BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAW,EAAA,KAAK,eAAe,EAAkB,EAAE,OACrD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACrD,EAAuB,GAAgB,KACvC,EAAqB,EAAa,mBAAmB,CAAC,IAExD,EAAW,EACf,GAAI,GACF,GAAI,GAKF,OAJM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAa,IAErC,GAAM,GACF,GAAiB,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAc,GAE7D,EAAW,EAAa,eAAe,EAAU,EAAE,OACnD,GAAM,GAAe,EAAa,QAAQ,IAAI,EAAS,QACvD,EAAa,OAAS,MAEtB,GAAW,GAAc,EAAG,EAAc,GAG5C,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAGpD,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGhB,EAAW,EACf,AAAI,GAEF,GAAW,EAAA,aAAa,qBAAqB,EAAa,IAG5D,GAAI,GACJ,GAAI,GACF,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAY,GACd,EAAQ,EAAA,KAAK,cAAc,GAAc,EAAU,EAAE,OAEzD,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAQ,EAAU,EAAa,EAAU,GAGjD,MAAI,IACF,EAAa,8BAA8B,GAGtC,IC9FX,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,EAAE,OAGjD,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAuBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAC3B,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,EAAQ,gBAAgB,EAAyB,CAAC,GAAI,EAAE,OAEtD,EAAyB,GAAI,IAAyB,GACtD,EAAS,EAAQ,gBACnB,EAAwB,CAAC,EAAI,GAAmB,EAAE,OACtD,SAAQ,8BAA8B,GAC/B,EAGF,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,ICrDd,YAuBI,EAAe,EACf,EACA,GACF,GAAI,GAAU,GAAI,IAAc,EAAU,MAAO,IAC3C,EAAa,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAEzD,EAAU,GAAI,IAAc,EAAU,MAAO,GAAM,GAAM,GACzD,GAAM,GAAc,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAC1D,MAAO,CAAC,EAAY,GC/BtB,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAe,EAErB,EAAA,KAAK,OACD,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,uDACF,EAAE,MAAM,OAAM,MACtB,GAAM,GAA8B,CAAC,EAAG,GACxC,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,GAET,EAAA,GAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAEf,MAAO,CAAC,EAAQ,KCjDpB,AAoBO,GAAM,IAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAG9B,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EAE1B,MAAO,GAAA,aAAa,wBAChB,EAAW,EAAY,EAAkB,EACzC,KC3CR,AAkBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC1C7B,AAmBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAqBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EACA,GAPJ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAOtB,GAAM,GAAc,EAAW,GACzB,EAAa,EAAW,GACxB,EAAY,KAAK,IAAI,GAAS,QAAQ,GACtC,EAAY,KAAK,IAAI,GAAS,QAAQ,GAC5C,KAAK,YAAc,EAEb,GAAA,GAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAgB,EAAQ,QAAQ,GAChC,EAAgB,EAAQ,QAAQ,GAElC,EAAc,GAClB,AAAI,MAAO,IAAc,SACvB,EAAc,uBAAuB,EAAU,QAAQ,GAAE,IAEzD,EAAc;2BACO,EAAU,KAAK,KAAI;8CAI1C,KAAK,SAAW;;;;;4CAKwB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;4CACpB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;iDACf,EAAa;iDACb,EAAa;YAClD,EAAW;uCACgB,EAAU,+BACzC,EAAW;;;;;MAOnB,MAAA,MCtEA,AAuBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAe,EAEf,EAAU,GAAI,IACf,EAAmB,MAAO,EAAS,EAAW,GAC7C,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KClCX,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAqBA,GAAM,IAAS,gBAEF,GAAmB,GAAgB,IAEnC,GAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IC5Bd,AAqBA,GAAM,IAAqB,4BAEd,GACT,GAAiB,GAAoB,IAE5B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,IC7Bd,AAqBA,GAAM,IAAM,iBAEC,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC5Bd,AAwBO,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,QACb,WAAY,SAAC,GAQX,OARY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACA,EAAA,EAAA,KACD,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAI,GACJ,GAAI,EAAa,mBAAmB,CAAC,KACnC,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QACtC,EAAS,EAAS,OAClB,EAAY,GAAa,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAE/D,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAc,EAAG,EAAM,GAE/B,MAAO,KCnDX,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAGpB,QAAQ,KACJ,YACA,8DACJ,GAAM,GAAS,EAAQ,SAAS,EAAE,QAC5B,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IChDd,AA4CA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,+RCxEjB,AAGA,GAAM,IAAU,QCHhB,GAsCa,IAAU,CACrB,YAAa,GAAA,aACb,mBAAoB,GAAA,YACpB,qBAAsB,GAAA,cACtB,YAAa,GAAA,aACb,cAAe,GAAA,eACf,iBAAkB,GAAA,kBAClB,KAAQ,yZC7CV,oBAAM,IAAK,KAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,GAAG,IAAI,EAAI,WAAY,GAChC,EAAO,GAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,GAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,GAAG,IAAI,EAAW,GAC5B,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,GAAG,IAAI,EAAU,GACtC,EAAoB,GAAG,IAAI,EAAS,GACpC,EAAc,GAAG,IAAI,EAAoB,GACzC,EAAS,GAAG,IAAI,EAAmB,GACnC,EAAO,GAAG,IAAI,EAAmB,GACjC,EAAkB,GAAG,IAAI,EAAQ,GACjC,EAAgB,GAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,IAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,IAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,GAAG,SAAS,KAAK,aAChC,KAAK,UAAY,GAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,GAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,GAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,GAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,GAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,GACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,GAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,IAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,kBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,oBAAM,IAAK,KAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,GAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,YAAc,KClDtB,kBAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,GAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,GAAQ,wBAA0B,KCvFlC,eACA,KAAM,IAAK,KACL,GAAW,KACX,GAAY,KACZ,GAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,GAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,GAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,GAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,GAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,GAAU,yBAAyB,GACtD,EAAkB,GAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,GAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,GAAS,YAAY,GAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,GAAS,WAAW,GACpC,GAAI,GAAO,GAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,GAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,GAAS,oBAAoB,EAAe,GACxD,EAAc,GAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,GAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,GAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,GAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,GAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,GAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,EAAgB,KAAM,GAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,EAAa,QAAS,EAAiB,KAAM,GAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,EAAkB,KAAK,UAAU,QAAQ,GAAG,OAAO,CAAC,EAAa,KACjE,EAAqB,EAAe,WAC1C,EAAe,UACf,KAAM,GAAc,EAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,EAAkB,KAAM,GAAsB,KAAK,aAAa,EAAa,EAAY,EAAgB,IACtH,EAAe,EAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,EAAmB,KAAM,GAAuB,KAAK,aAAa,EAAc,EAAa,GAC1G,EAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,GAAiC,GAC5C,IAAsB,EAAW,EAAkB,QACnD,GAAsB,EAAW,EAAmB,UAE/C,AAAI,EAAgC,EACzC,GAAsB,EAAW,EAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,EAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,GAAyB,KAAK,sBAAsB,EAAW,EAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,EAAoB,SAC1F,EAAY,EAAU,OAAO,GAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,GAAG,QAAQ,GACX,KAAM,GAAe,GAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,GAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,GAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,GAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,kBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,0CAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,oBAAM,IAAK,KACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QAAS,GAAQ,GAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,GAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,GAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,oBAAM,IAAK,KAEL,GAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,GAAG,QAAQ,WAAW,GAC/B,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,IAAO,KAAK,IAAO,IAAM,KAAM,IAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,GAAO,IAGhB,kBAA0B,GACxB,MAAK,IAAO,QAAQ,IAAO,OAAS,KAAM,IAAG,eAAe,EAAO,KAAK,OAAO,YACxE,GAAO,OAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,IAAI,WAC3B,UAAQ,EACD,GAET,GAAI,GACJ,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,GAAG,IAAI,EAAQ,CAAC,MAC1B,GAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,IAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,GAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,IAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OAC3B,EAAI,OAAS,KAAK,MAAM,IAAM,EAAK,IAAM,GAAK,SAAW,OACzD,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KCzDrB,oBAAM,IAAK,KAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,GAAG,KAAK,KACrB,KAAM,GAAS,GAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,IAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,GAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,GAAG,IAAI,EAAG,CAAC,QAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAY,GAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,oBAAM,IAAK,KAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,oBAAM,IAAK,KACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,eACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,oBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,kBAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,GAAQ,cAAgB,GAAQ,UAAU,OAC1C,GAAQ,QAAU,GAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,GAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,GAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,GAAQ,QAAQ,GAAa,GAAQ,QAAQ,KACnI,GAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,oBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,GAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,GAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,GAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,GAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,GAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,GAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,GAAQ,YAAc,KCnDtB,oBAAM,IAAY,KACZ,GAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,GAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,GAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,GAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,GAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,oBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,oBAAM,IAAK,KACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,GAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,GAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,GAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,GAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,GAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,GAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,IAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,GAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,IAAG,OAAS,EAAQ,GAAG,QAAQ,WAAW,GAEpE,GAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,IAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,GAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,GAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,GAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,GAAQ,kBAAoB,KCrH5B,oBAAM,IAAK,KACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,oBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,GAAQ,KAAO,GAAa,KAC5B,GAAQ,QAAU,GAAa,QAE/B,GAAQ,UAAY,GAAe,UACnC,GAAQ,oBAAsB,GAAe,oBAC7C,GAAQ,aAAe,GAAU,aACjC,GAAQ,QAAU,GAAU,QAC5B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,eAAiB,GAAK,eAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,kBAAoB,GAAK,kBACjC,GAAQ,UAAY,GAAK,YCnBzB,oBAAM,IAAK,KAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,GAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,SAAW,KCtEnB,oBAAM,IAAK,KACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,GAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,GAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,GAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,IAAG,KAAK,KACb,KAAM,GAAa,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,GAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,IAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,IAAG,KAAK,KACb,KAAM,GAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,IAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,GAAG,KAAK,IAAM,GAAG,IAAI,GAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,GAAG,KAAK,IAAM,GAAG,QAAQ,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,GAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,GAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,GAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,kBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,2BAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,KCzEtB,oBAAM,IAAK,KACL,GAAW,KACX,GAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,IAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,IAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,GAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,GAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,GAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,GAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,GAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,oBAAM,IAAK,KACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,GAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,IAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,0CAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GACT,SAAU,CACR,UAAW,sCACX,UAAW,IACX,SAAU,GACV,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCACX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,EACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,EACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,6vECnEjB,KAAM,IAAK,KACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,IAGJ,KAAM,IAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,GAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,GAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,GAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,GAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,IAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,GAAG,eAAiB,GAAO,SAC7B,IAAI,iCAAkC,GAAO,SAC7C,KAAM,IAAG,WAAW,GAAO,SAC3B,KAAM,IAAG,SAGX,GAAI,GACJ,AAAI,GAAG,eAAiB,SACtB,GAAkC,GAAG,MAAM,IAAI,4BAC/C,GAAG,MAAM,IAAI,2BAA4B,KAI3C,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,GAAO,OACvF,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,IAAI,SAAW,CAAC,GAAO,KAAK,IAAO,IAAM,KAAM,IAAO,QAAQ,KACrG,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,OAAO,SAAW,CAAC,GAAO,QAAQ,IAAO,OAAS,KAAM,IAAO,WAAW,KACjH,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,QAAQ,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,KAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,QAAQ,cAAc,EAAO,GAAO,MAAQ,GAC/F,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAAQ,GAChG,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,GAAO,KAAK,SACd,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAChE,EAAK,KAAO,KAAK,MAAM,KAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,KACZ,KAAM,GAAW,GAAO,KAAK,IAAI,SAAW,GAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,IAAU,GACrH,EAAK,UAAY,KAAK,MAAM,KAAQ,GAEpC,EAAY,KACZ,KAAM,GAAc,GAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,IAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,KAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,GAAG,SAAS,WAId,GAAG,MAAM,IAAI,2BAA4B,GAGzC,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,QAAQ,OAAS,GACjB,QAAQ,SAAW,GACnB,QAAQ,OAAS,GACjB,QAAQ,OAAS,GACjB,QAAQ,SAAW,GACnB,QAAQ,OAAS,GACjB,QAAQ,QAAU,GAClB,QAAQ,SAAW,GACnB,QAAQ,GAAK,GACb,QAAQ,QAAU,GAAI",
+  "sourcesContent": ["import Stream from 'stream';\nimport http from 'http';\nimport Url from 'url';\nimport https from 'https';\nimport zlib from 'zlib';\n\n// Based on https://github.com/tmpvar/jsdom/blob/aa85b2abf07766ff7bf5c1f6daafb3726f2f2db5/lib/jsdom/living/blob.js\n\n// fix for \"Readable\" isn't a named export issue\nconst Readable = Stream.Readable;\n\nconst BUFFER = Symbol('buffer');\nconst TYPE = Symbol('type');\n\nclass Blob {\n\tconstructor() {\n\t\tthis[TYPE] = '';\n\n\t\tconst blobParts = arguments[0];\n\t\tconst options = arguments[1];\n\n\t\tconst buffers = [];\n\t\tlet size = 0;\n\n\t\tif (blobParts) {\n\t\t\tconst a = blobParts;\n\t\t\tconst length = Number(a.length);\n\t\t\tfor (let i = 0; i < length; i++) {\n\t\t\t\tconst element = a[i];\n\t\t\t\tlet buffer;\n\t\t\t\tif (element instanceof Buffer) {\n\t\t\t\t\tbuffer = element;\n\t\t\t\t} else if (ArrayBuffer.isView(element)) {\n\t\t\t\t\tbuffer = Buffer.from(element.buffer, element.byteOffset, element.byteLength);\n\t\t\t\t} else if (element instanceof ArrayBuffer) {\n\t\t\t\t\tbuffer = Buffer.from(element);\n\t\t\t\t} else if (element instanceof Blob) {\n\t\t\t\t\tbuffer = element[BUFFER];\n\t\t\t\t} else {\n\t\t\t\t\tbuffer = Buffer.from(typeof element === 'string' ? element : String(element));\n\t\t\t\t}\n\t\t\t\tsize += buffer.length;\n\t\t\t\tbuffers.push(buffer);\n\t\t\t}\n\t\t}\n\n\t\tthis[BUFFER] = Buffer.concat(buffers);\n\n\t\tlet type = options && options.type !== undefined && String(options.type).toLowerCase();\n\t\tif (type && !/[^\\u0020-\\u007E]/.test(type)) {\n\t\t\tthis[TYPE] = type;\n\t\t}\n\t}\n\tget size() {\n\t\treturn this[BUFFER].length;\n\t}\n\tget type() {\n\t\treturn this[TYPE];\n\t}\n\ttext() {\n\t\treturn Promise.resolve(this[BUFFER].toString());\n\t}\n\tarrayBuffer() {\n\t\tconst buf = this[BUFFER];\n\t\tconst ab = buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n\t\treturn Promise.resolve(ab);\n\t}\n\tstream() {\n\t\tconst readable = new Readable();\n\t\treadable._read = function () {};\n\t\treadable.push(this[BUFFER]);\n\t\treadable.push(null);\n\t\treturn readable;\n\t}\n\ttoString() {\n\t\treturn '[object Blob]';\n\t}\n\tslice() {\n\t\tconst size = this.size;\n\n\t\tconst start = arguments[0];\n\t\tconst end = arguments[1];\n\t\tlet relativeStart, relativeEnd;\n\t\tif (start === undefined) {\n\t\t\trelativeStart = 0;\n\t\t} else if (start < 0) {\n\t\t\trelativeStart = Math.max(size + start, 0);\n\t\t} else {\n\t\t\trelativeStart = Math.min(start, size);\n\t\t}\n\t\tif (end === undefined) {\n\t\t\trelativeEnd = size;\n\t\t} else if (end < 0) {\n\t\t\trelativeEnd = Math.max(size + end, 0);\n\t\t} else {\n\t\t\trelativeEnd = Math.min(end, size);\n\t\t}\n\t\tconst span = Math.max(relativeEnd - relativeStart, 0);\n\n\t\tconst buffer = this[BUFFER];\n\t\tconst slicedBuffer = buffer.slice(relativeStart, relativeStart + span);\n\t\tconst blob = new Blob([], { type: arguments[2] });\n\t\tblob[BUFFER] = slicedBuffer;\n\t\treturn blob;\n\t}\n}\n\nObject.defineProperties(Blob.prototype, {\n\tsize: { enumerable: true },\n\ttype: { enumerable: true },\n\tslice: { enumerable: true }\n});\n\nObject.defineProperty(Blob.prototype, Symbol.toStringTag, {\n\tvalue: 'Blob',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\n/**\n * fetch-error.js\n *\n * FetchError interface for operational errors\n */\n\n/**\n * Create FetchError instance\n *\n * @param   String      message      Error message for human\n * @param   String      type         Error type for machine\n * @param   String      systemError  For Node.js system error\n * @return  FetchError\n */\nfunction FetchError(message, type, systemError) {\n  Error.call(this, message);\n\n  this.message = message;\n  this.type = type;\n\n  // when err.type is `system`, err.code contains system error code\n  if (systemError) {\n    this.code = this.errno = systemError.code;\n  }\n\n  // hide custom error implementation details from end-users\n  Error.captureStackTrace(this, this.constructor);\n}\n\nFetchError.prototype = Object.create(Error.prototype);\nFetchError.prototype.constructor = FetchError;\nFetchError.prototype.name = 'FetchError';\n\nlet convert;\ntry {\n\tconvert = require('encoding').convert;\n} catch (e) {}\n\nconst INTERNALS = Symbol('Body internals');\n\n// fix an issue where \"PassThrough\" isn't a named export for node <10\nconst PassThrough = Stream.PassThrough;\n\n/**\n * Body mixin\n *\n * Ref: https://fetch.spec.whatwg.org/#body\n *\n * @param   Stream  body  Readable stream\n * @param   Object  opts  Response options\n * @return  Void\n */\nfunction Body(body) {\n\tvar _this = this;\n\n\tvar _ref = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {},\n\t    _ref$size = _ref.size;\n\n\tlet size = _ref$size === undefined ? 0 : _ref$size;\n\tvar _ref$timeout = _ref.timeout;\n\tlet timeout = _ref$timeout === undefined ? 0 : _ref$timeout;\n\n\tif (body == null) {\n\t\t// body is undefined or null\n\t\tbody = null;\n\t} else if (isURLSearchParams(body)) {\n\t\t// body is a URLSearchParams\n\t\tbody = Buffer.from(body.toString());\n\t} else if (isBlob(body)) ; else if (Buffer.isBuffer(body)) ; else if (Object.prototype.toString.call(body) === '[object ArrayBuffer]') {\n\t\t// body is ArrayBuffer\n\t\tbody = Buffer.from(body);\n\t} else if (ArrayBuffer.isView(body)) {\n\t\t// body is ArrayBufferView\n\t\tbody = Buffer.from(body.buffer, body.byteOffset, body.byteLength);\n\t} else if (body instanceof Stream) ; else {\n\t\t// none of the above\n\t\t// coerce to string then buffer\n\t\tbody = Buffer.from(String(body));\n\t}\n\tthis[INTERNALS] = {\n\t\tbody,\n\t\tdisturbed: false,\n\t\terror: null\n\t};\n\tthis.size = size;\n\tthis.timeout = timeout;\n\n\tif (body instanceof Stream) {\n\t\tbody.on('error', function (err) {\n\t\t\tconst error = err.name === 'AbortError' ? err : new FetchError(`Invalid response body while trying to fetch ${_this.url}: ${err.message}`, 'system', err);\n\t\t\t_this[INTERNALS].error = error;\n\t\t});\n\t}\n}\n\nBody.prototype = {\n\tget body() {\n\t\treturn this[INTERNALS].body;\n\t},\n\n\tget bodyUsed() {\n\t\treturn this[INTERNALS].disturbed;\n\t},\n\n\t/**\n  * Decode response as ArrayBuffer\n  *\n  * @return  Promise\n  */\n\tarrayBuffer() {\n\t\treturn consumeBody.call(this).then(function (buf) {\n\t\t\treturn buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n\t\t});\n\t},\n\n\t/**\n  * Return raw response as Blob\n  *\n  * @return Promise\n  */\n\tblob() {\n\t\tlet ct = this.headers && this.headers.get('content-type') || '';\n\t\treturn consumeBody.call(this).then(function (buf) {\n\t\t\treturn Object.assign(\n\t\t\t// Prevent copying\n\t\t\tnew Blob([], {\n\t\t\t\ttype: ct.toLowerCase()\n\t\t\t}), {\n\t\t\t\t[BUFFER]: buf\n\t\t\t});\n\t\t});\n\t},\n\n\t/**\n  * Decode response as json\n  *\n  * @return  Promise\n  */\n\tjson() {\n\t\tvar _this2 = this;\n\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\ttry {\n\t\t\t\treturn JSON.parse(buffer.toString());\n\t\t\t} catch (err) {\n\t\t\t\treturn Body.Promise.reject(new FetchError(`invalid json response body at ${_this2.url} reason: ${err.message}`, 'invalid-json'));\n\t\t\t}\n\t\t});\n\t},\n\n\t/**\n  * Decode response as text\n  *\n  * @return  Promise\n  */\n\ttext() {\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\treturn buffer.toString();\n\t\t});\n\t},\n\n\t/**\n  * Decode response as buffer (non-spec api)\n  *\n  * @return  Promise\n  */\n\tbuffer() {\n\t\treturn consumeBody.call(this);\n\t},\n\n\t/**\n  * Decode response as text, while automatically detecting the encoding and\n  * trying to decode to UTF-8 (non-spec api)\n  *\n  * @return  Promise\n  */\n\ttextConverted() {\n\t\tvar _this3 = this;\n\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\treturn convertBody(buffer, _this3.headers);\n\t\t});\n\t}\n};\n\n// In browsers, all properties are enumerable.\nObject.defineProperties(Body.prototype, {\n\tbody: { enumerable: true },\n\tbodyUsed: { enumerable: true },\n\tarrayBuffer: { enumerable: true },\n\tblob: { enumerable: true },\n\tjson: { enumerable: true },\n\ttext: { enumerable: true }\n});\n\nBody.mixIn = function (proto) {\n\tfor (const name of Object.getOwnPropertyNames(Body.prototype)) {\n\t\t// istanbul ignore else: future proof\n\t\tif (!(name in proto)) {\n\t\t\tconst desc = Object.getOwnPropertyDescriptor(Body.prototype, name);\n\t\t\tObject.defineProperty(proto, name, desc);\n\t\t}\n\t}\n};\n\n/**\n * Consume and convert an entire Body to a Buffer.\n *\n * Ref: https://fetch.spec.whatwg.org/#concept-body-consume-body\n *\n * @return  Promise\n */\nfunction consumeBody() {\n\tvar _this4 = this;\n\n\tif (this[INTERNALS].disturbed) {\n\t\treturn Body.Promise.reject(new TypeError(`body used already for: ${this.url}`));\n\t}\n\n\tthis[INTERNALS].disturbed = true;\n\n\tif (this[INTERNALS].error) {\n\t\treturn Body.Promise.reject(this[INTERNALS].error);\n\t}\n\n\tlet body = this.body;\n\n\t// body is null\n\tif (body === null) {\n\t\treturn Body.Promise.resolve(Buffer.alloc(0));\n\t}\n\n\t// body is blob\n\tif (isBlob(body)) {\n\t\tbody = body.stream();\n\t}\n\n\t// body is buffer\n\tif (Buffer.isBuffer(body)) {\n\t\treturn Body.Promise.resolve(body);\n\t}\n\n\t// istanbul ignore if: should never happen\n\tif (!(body instanceof Stream)) {\n\t\treturn Body.Promise.resolve(Buffer.alloc(0));\n\t}\n\n\t// body is stream\n\t// get ready to actually consume the body\n\tlet accum = [];\n\tlet accumBytes = 0;\n\tlet abort = false;\n\n\treturn new Body.Promise(function (resolve, reject) {\n\t\tlet resTimeout;\n\n\t\t// allow timeout on slow response body\n\t\tif (_this4.timeout) {\n\t\t\tresTimeout = setTimeout(function () {\n\t\t\t\tabort = true;\n\t\t\t\treject(new FetchError(`Response timeout while trying to fetch ${_this4.url} (over ${_this4.timeout}ms)`, 'body-timeout'));\n\t\t\t}, _this4.timeout);\n\t\t}\n\n\t\t// handle stream errors\n\t\tbody.on('error', function (err) {\n\t\t\tif (err.name === 'AbortError') {\n\t\t\t\t// if the request was aborted, reject with this Error\n\t\t\t\tabort = true;\n\t\t\t\treject(err);\n\t\t\t} else {\n\t\t\t\t// other errors, such as incorrect content-encoding\n\t\t\t\treject(new FetchError(`Invalid response body while trying to fetch ${_this4.url}: ${err.message}`, 'system', err));\n\t\t\t}\n\t\t});\n\n\t\tbody.on('data', function (chunk) {\n\t\t\tif (abort || chunk === null) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (_this4.size && accumBytes + chunk.length > _this4.size) {\n\t\t\t\tabort = true;\n\t\t\t\treject(new FetchError(`content size at ${_this4.url} over limit: ${_this4.size}`, 'max-size'));\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\taccumBytes += chunk.length;\n\t\t\taccum.push(chunk);\n\t\t});\n\n\t\tbody.on('end', function () {\n\t\t\tif (abort) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tclearTimeout(resTimeout);\n\n\t\t\ttry {\n\t\t\t\tresolve(Buffer.concat(accum, accumBytes));\n\t\t\t} catch (err) {\n\t\t\t\t// handle streams that have accumulated too much data (issue #414)\n\t\t\t\treject(new FetchError(`Could not create Buffer from response body for ${_this4.url}: ${err.message}`, 'system', err));\n\t\t\t}\n\t\t});\n\t});\n}\n\n/**\n * Detect buffer encoding and convert to target encoding\n * ref: http://www.w3.org/TR/2011/WD-html5-20110113/parsing.html#determining-the-character-encoding\n *\n * @param   Buffer  buffer    Incoming buffer\n * @param   String  encoding  Target encoding\n * @return  String\n */\nfunction convertBody(buffer, headers) {\n\tif (typeof convert !== 'function') {\n\t\tthrow new Error('The package `encoding` must be installed to use the textConverted() function');\n\t}\n\n\tconst ct = headers.get('content-type');\n\tlet charset = 'utf-8';\n\tlet res, str;\n\n\t// header\n\tif (ct) {\n\t\tres = /charset=([^;]*)/i.exec(ct);\n\t}\n\n\t// no charset in content type, peek at response body for at most 1024 bytes\n\tstr = buffer.slice(0, 1024).toString();\n\n\t// html5\n\tif (!res && str) {\n\t\tres = /<meta.+?charset=(['\"])(.+?)\\1/i.exec(str);\n\t}\n\n\t// html4\n\tif (!res && str) {\n\t\tres = /<meta[\\s]+?http-equiv=(['\"])content-type\\1[\\s]+?content=(['\"])(.+?)\\2/i.exec(str);\n\t\tif (!res) {\n\t\t\tres = /<meta[\\s]+?content=(['\"])(.+?)\\1[\\s]+?http-equiv=(['\"])content-type\\3/i.exec(str);\n\t\t\tif (res) {\n\t\t\t\tres.pop(); // drop last quote\n\t\t\t}\n\t\t}\n\n\t\tif (res) {\n\t\t\tres = /charset=(.*)/i.exec(res.pop());\n\t\t}\n\t}\n\n\t// xml\n\tif (!res && str) {\n\t\tres = /<\\?xml.+?encoding=(['\"])(.+?)\\1/i.exec(str);\n\t}\n\n\t// found charset\n\tif (res) {\n\t\tcharset = res.pop();\n\n\t\t// prevent decode issues when sites use incorrect encoding\n\t\t// ref: https://hsivonen.fi/encoding-menu/\n\t\tif (charset === 'gb2312' || charset === 'gbk') {\n\t\t\tcharset = 'gb18030';\n\t\t}\n\t}\n\n\t// turn raw buffers into a single utf-8 buffer\n\treturn convert(buffer, 'UTF-8', charset).toString();\n}\n\n/**\n * Detect a URLSearchParams object\n * ref: https://github.com/bitinn/node-fetch/issues/296#issuecomment-307598143\n *\n * @param   Object  obj     Object to detect by type or brand\n * @return  String\n */\nfunction isURLSearchParams(obj) {\n\t// Duck-typing as a necessary condition.\n\tif (typeof obj !== 'object' || typeof obj.append !== 'function' || typeof obj.delete !== 'function' || typeof obj.get !== 'function' || typeof obj.getAll !== 'function' || typeof obj.has !== 'function' || typeof obj.set !== 'function') {\n\t\treturn false;\n\t}\n\n\t// Brand-checking and more duck-typing as optional condition.\n\treturn obj.constructor.name === 'URLSearchParams' || Object.prototype.toString.call(obj) === '[object URLSearchParams]' || typeof obj.sort === 'function';\n}\n\n/**\n * Check if `obj` is a W3C `Blob` object (which `File` inherits from)\n * @param  {*} obj\n * @return {boolean}\n */\nfunction isBlob(obj) {\n\treturn typeof obj === 'object' && typeof obj.arrayBuffer === 'function' && typeof obj.type === 'string' && typeof obj.stream === 'function' && typeof obj.constructor === 'function' && typeof obj.constructor.name === 'string' && /^(Blob|File)$/.test(obj.constructor.name) && /^(Blob|File)$/.test(obj[Symbol.toStringTag]);\n}\n\n/**\n * Clone body given Res/Req instance\n *\n * @param   Mixed  instance  Response or Request instance\n * @return  Mixed\n */\nfunction clone(instance) {\n\tlet p1, p2;\n\tlet body = instance.body;\n\n\t// don't allow cloning a used body\n\tif (instance.bodyUsed) {\n\t\tthrow new Error('cannot clone body after it is used');\n\t}\n\n\t// check that body is a stream and not form-data object\n\t// note: we can't clone the form-data object without having it as a dependency\n\tif (body instanceof Stream && typeof body.getBoundary !== 'function') {\n\t\t// tee instance body\n\t\tp1 = new PassThrough();\n\t\tp2 = new PassThrough();\n\t\tbody.pipe(p1);\n\t\tbody.pipe(p2);\n\t\t// set instance body to teed body and return the other teed body\n\t\tinstance[INTERNALS].body = p1;\n\t\tbody = p2;\n\t}\n\n\treturn body;\n}\n\n/**\n * Performs the operation \"extract a `Content-Type` value from |object|\" as\n * specified in the specification:\n * https://fetch.spec.whatwg.org/#concept-bodyinit-extract\n *\n * This function assumes that instance.body is present.\n *\n * @param   Mixed  instance  Any options.body input\n */\nfunction extractContentType(body) {\n\tif (body === null) {\n\t\t// body is null\n\t\treturn null;\n\t} else if (typeof body === 'string') {\n\t\t// body is string\n\t\treturn 'text/plain;charset=UTF-8';\n\t} else if (isURLSearchParams(body)) {\n\t\t// body is a URLSearchParams\n\t\treturn 'application/x-www-form-urlencoded;charset=UTF-8';\n\t} else if (isBlob(body)) {\n\t\t// body is blob\n\t\treturn body.type || null;\n\t} else if (Buffer.isBuffer(body)) {\n\t\t// body is buffer\n\t\treturn null;\n\t} else if (Object.prototype.toString.call(body) === '[object ArrayBuffer]') {\n\t\t// body is ArrayBuffer\n\t\treturn null;\n\t} else if (ArrayBuffer.isView(body)) {\n\t\t// body is ArrayBufferView\n\t\treturn null;\n\t} else if (typeof body.getBoundary === 'function') {\n\t\t// detect form data input from form-data module\n\t\treturn `multipart/form-data;boundary=${body.getBoundary()}`;\n\t} else if (body instanceof Stream) {\n\t\t// body is stream\n\t\t// can't really do much about this\n\t\treturn null;\n\t} else {\n\t\t// Body constructor defaults other things to string\n\t\treturn 'text/plain;charset=UTF-8';\n\t}\n}\n\n/**\n * The Fetch Standard treats this as if \"total bytes\" is a property on the body.\n * For us, we have to explicitly get it with a function.\n *\n * ref: https://fetch.spec.whatwg.org/#concept-body-total-bytes\n *\n * @param   Body    instance   Instance of Body\n * @return  Number?            Number of bytes, or null if not possible\n */\nfunction getTotalBytes(instance) {\n\tconst body = instance.body;\n\n\n\tif (body === null) {\n\t\t// body is null\n\t\treturn 0;\n\t} else if (isBlob(body)) {\n\t\treturn body.size;\n\t} else if (Buffer.isBuffer(body)) {\n\t\t// body is buffer\n\t\treturn body.length;\n\t} else if (body && typeof body.getLengthSync === 'function') {\n\t\t// detect form data input from form-data module\n\t\tif (body._lengthRetrievers && body._lengthRetrievers.length == 0 || // 1.x\n\t\tbody.hasKnownLength && body.hasKnownLength()) {\n\t\t\t// 2.x\n\t\t\treturn body.getLengthSync();\n\t\t}\n\t\treturn null;\n\t} else {\n\t\t// body is stream\n\t\treturn null;\n\t}\n}\n\n/**\n * Write a Body to a Node.js WritableStream (e.g. http.Request) object.\n *\n * @param   Body    instance   Instance of Body\n * @return  Void\n */\nfunction writeToStream(dest, instance) {\n\tconst body = instance.body;\n\n\n\tif (body === null) {\n\t\t// body is null\n\t\tdest.end();\n\t} else if (isBlob(body)) {\n\t\tbody.stream().pipe(dest);\n\t} else if (Buffer.isBuffer(body)) {\n\t\t// body is buffer\n\t\tdest.write(body);\n\t\tdest.end();\n\t} else {\n\t\t// body is stream\n\t\tbody.pipe(dest);\n\t}\n}\n\n// expose Promise\nBody.Promise = global.Promise;\n\n/**\n * headers.js\n *\n * Headers class offers convenient helpers\n */\n\nconst invalidTokenRegex = /[^\\^_`a-zA-Z\\-0-9!#$%&'*+.|~]/;\nconst invalidHeaderCharRegex = /[^\\t\\x20-\\x7e\\x80-\\xff]/;\n\nfunction validateName(name) {\n\tname = `${name}`;\n\tif (invalidTokenRegex.test(name) || name === '') {\n\t\tthrow new TypeError(`${name} is not a legal HTTP header name`);\n\t}\n}\n\nfunction validateValue(value) {\n\tvalue = `${value}`;\n\tif (invalidHeaderCharRegex.test(value)) {\n\t\tthrow new TypeError(`${value} is not a legal HTTP header value`);\n\t}\n}\n\n/**\n * Find the key in the map object given a header name.\n *\n * Returns undefined if not found.\n *\n * @param   String  name  Header name\n * @return  String|Undefined\n */\nfunction find(map, name) {\n\tname = name.toLowerCase();\n\tfor (const key in map) {\n\t\tif (key.toLowerCase() === name) {\n\t\t\treturn key;\n\t\t}\n\t}\n\treturn undefined;\n}\n\nconst MAP = Symbol('map');\nclass Headers {\n\t/**\n  * Headers class\n  *\n  * @param   Object  headers  Response headers\n  * @return  Void\n  */\n\tconstructor() {\n\t\tlet init = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : undefined;\n\n\t\tthis[MAP] = Object.create(null);\n\n\t\tif (init instanceof Headers) {\n\t\t\tconst rawHeaders = init.raw();\n\t\t\tconst headerNames = Object.keys(rawHeaders);\n\n\t\t\tfor (const headerName of headerNames) {\n\t\t\t\tfor (const value of rawHeaders[headerName]) {\n\t\t\t\t\tthis.append(headerName, value);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\treturn;\n\t\t}\n\n\t\t// We don't worry about converting prop to ByteString here as append()\n\t\t// will handle it.\n\t\tif (init == null) ; else if (typeof init === 'object') {\n\t\t\tconst method = init[Symbol.iterator];\n\t\t\tif (method != null) {\n\t\t\t\tif (typeof method !== 'function') {\n\t\t\t\t\tthrow new TypeError('Header pairs must be iterable');\n\t\t\t\t}\n\n\t\t\t\t// sequence<sequence<ByteString>>\n\t\t\t\t// Note: per spec we have to first exhaust the lists then process them\n\t\t\t\tconst pairs = [];\n\t\t\t\tfor (const pair of init) {\n\t\t\t\t\tif (typeof pair !== 'object' || typeof pair[Symbol.iterator] !== 'function') {\n\t\t\t\t\t\tthrow new TypeError('Each header pair must be iterable');\n\t\t\t\t\t}\n\t\t\t\t\tpairs.push(Array.from(pair));\n\t\t\t\t}\n\n\t\t\t\tfor (const pair of pairs) {\n\t\t\t\t\tif (pair.length !== 2) {\n\t\t\t\t\t\tthrow new TypeError('Each header pair must be a name/value tuple');\n\t\t\t\t\t}\n\t\t\t\t\tthis.append(pair[0], pair[1]);\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\t// record<ByteString, ByteString>\n\t\t\t\tfor (const key of Object.keys(init)) {\n\t\t\t\t\tconst value = init[key];\n\t\t\t\t\tthis.append(key, value);\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tthrow new TypeError('Provided initializer must be an object');\n\t\t}\n\t}\n\n\t/**\n  * Return combined header value given name\n  *\n  * @param   String  name  Header name\n  * @return  Mixed\n  */\n\tget(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key === undefined) {\n\t\t\treturn null;\n\t\t}\n\n\t\treturn this[MAP][key].join(', ');\n\t}\n\n\t/**\n  * Iterate over all headers\n  *\n  * @param   Function  callback  Executed for each item with parameters (value, name, thisArg)\n  * @param   Boolean   thisArg   `this` context for callback function\n  * @return  Void\n  */\n\tforEach(callback) {\n\t\tlet thisArg = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : undefined;\n\n\t\tlet pairs = getHeaders(this);\n\t\tlet i = 0;\n\t\twhile (i < pairs.length) {\n\t\t\tvar _pairs$i = pairs[i];\n\t\t\tconst name = _pairs$i[0],\n\t\t\t      value = _pairs$i[1];\n\n\t\t\tcallback.call(thisArg, value, name, this);\n\t\t\tpairs = getHeaders(this);\n\t\t\ti++;\n\t\t}\n\t}\n\n\t/**\n  * Overwrite header values given name\n  *\n  * @param   String  name   Header name\n  * @param   String  value  Header value\n  * @return  Void\n  */\n\tset(name, value) {\n\t\tname = `${name}`;\n\t\tvalue = `${value}`;\n\t\tvalidateName(name);\n\t\tvalidateValue(value);\n\t\tconst key = find(this[MAP], name);\n\t\tthis[MAP][key !== undefined ? key : name] = [value];\n\t}\n\n\t/**\n  * Append a value onto existing header\n  *\n  * @param   String  name   Header name\n  * @param   String  value  Header value\n  * @return  Void\n  */\n\tappend(name, value) {\n\t\tname = `${name}`;\n\t\tvalue = `${value}`;\n\t\tvalidateName(name);\n\t\tvalidateValue(value);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key !== undefined) {\n\t\t\tthis[MAP][key].push(value);\n\t\t} else {\n\t\t\tthis[MAP][name] = [value];\n\t\t}\n\t}\n\n\t/**\n  * Check for header name existence\n  *\n  * @param   String   name  Header name\n  * @return  Boolean\n  */\n\thas(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\treturn find(this[MAP], name) !== undefined;\n\t}\n\n\t/**\n  * Delete all header values given name\n  *\n  * @param   String  name  Header name\n  * @return  Void\n  */\n\tdelete(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key !== undefined) {\n\t\t\tdelete this[MAP][key];\n\t\t}\n\t}\n\n\t/**\n  * Return raw headers (non-spec api)\n  *\n  * @return  Object\n  */\n\traw() {\n\t\treturn this[MAP];\n\t}\n\n\t/**\n  * Get an iterator on keys.\n  *\n  * @return  Iterator\n  */\n\tkeys() {\n\t\treturn createHeadersIterator(this, 'key');\n\t}\n\n\t/**\n  * Get an iterator on values.\n  *\n  * @return  Iterator\n  */\n\tvalues() {\n\t\treturn createHeadersIterator(this, 'value');\n\t}\n\n\t/**\n  * Get an iterator on entries.\n  *\n  * This is the default iterator of the Headers object.\n  *\n  * @return  Iterator\n  */\n\t[Symbol.iterator]() {\n\t\treturn createHeadersIterator(this, 'key+value');\n\t}\n}\nHeaders.prototype.entries = Headers.prototype[Symbol.iterator];\n\nObject.defineProperty(Headers.prototype, Symbol.toStringTag, {\n\tvalue: 'Headers',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nObject.defineProperties(Headers.prototype, {\n\tget: { enumerable: true },\n\tforEach: { enumerable: true },\n\tset: { enumerable: true },\n\tappend: { enumerable: true },\n\thas: { enumerable: true },\n\tdelete: { enumerable: true },\n\tkeys: { enumerable: true },\n\tvalues: { enumerable: true },\n\tentries: { enumerable: true }\n});\n\nfunction getHeaders(headers) {\n\tlet kind = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : 'key+value';\n\n\tconst keys = Object.keys(headers[MAP]).sort();\n\treturn keys.map(kind === 'key' ? function (k) {\n\t\treturn k.toLowerCase();\n\t} : kind === 'value' ? function (k) {\n\t\treturn headers[MAP][k].join(', ');\n\t} : function (k) {\n\t\treturn [k.toLowerCase(), headers[MAP][k].join(', ')];\n\t});\n}\n\nconst INTERNAL = Symbol('internal');\n\nfunction createHeadersIterator(target, kind) {\n\tconst iterator = Object.create(HeadersIteratorPrototype);\n\titerator[INTERNAL] = {\n\t\ttarget,\n\t\tkind,\n\t\tindex: 0\n\t};\n\treturn iterator;\n}\n\nconst HeadersIteratorPrototype = Object.setPrototypeOf({\n\tnext() {\n\t\t// istanbul ignore if\n\t\tif (!this || Object.getPrototypeOf(this) !== HeadersIteratorPrototype) {\n\t\t\tthrow new TypeError('Value of `this` is not a HeadersIterator');\n\t\t}\n\n\t\tvar _INTERNAL = this[INTERNAL];\n\t\tconst target = _INTERNAL.target,\n\t\t      kind = _INTERNAL.kind,\n\t\t      index = _INTERNAL.index;\n\n\t\tconst values = getHeaders(target, kind);\n\t\tconst len = values.length;\n\t\tif (index >= len) {\n\t\t\treturn {\n\t\t\t\tvalue: undefined,\n\t\t\t\tdone: true\n\t\t\t};\n\t\t}\n\n\t\tthis[INTERNAL].index = index + 1;\n\n\t\treturn {\n\t\t\tvalue: values[index],\n\t\t\tdone: false\n\t\t};\n\t}\n}, Object.getPrototypeOf(Object.getPrototypeOf([][Symbol.iterator]())));\n\nObject.defineProperty(HeadersIteratorPrototype, Symbol.toStringTag, {\n\tvalue: 'HeadersIterator',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\n/**\n * Export the Headers object in a form that Node.js can consume.\n *\n * @param   Headers  headers\n * @return  Object\n */\nfunction exportNodeCompatibleHeaders(headers) {\n\tconst obj = Object.assign({ __proto__: null }, headers[MAP]);\n\n\t// http.request() only supports string as Host header. This hack makes\n\t// specifying custom Host header possible.\n\tconst hostHeaderKey = find(headers[MAP], 'Host');\n\tif (hostHeaderKey !== undefined) {\n\t\tobj[hostHeaderKey] = obj[hostHeaderKey][0];\n\t}\n\n\treturn obj;\n}\n\n/**\n * Create a Headers object from an object of headers, ignoring those that do\n * not conform to HTTP grammar productions.\n *\n * @param   Object  obj  Object of headers\n * @return  Headers\n */\nfunction createHeadersLenient(obj) {\n\tconst headers = new Headers();\n\tfor (const name of Object.keys(obj)) {\n\t\tif (invalidTokenRegex.test(name)) {\n\t\t\tcontinue;\n\t\t}\n\t\tif (Array.isArray(obj[name])) {\n\t\t\tfor (const val of obj[name]) {\n\t\t\t\tif (invalidHeaderCharRegex.test(val)) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tif (headers[MAP][name] === undefined) {\n\t\t\t\t\theaders[MAP][name] = [val];\n\t\t\t\t} else {\n\t\t\t\t\theaders[MAP][name].push(val);\n\t\t\t\t}\n\t\t\t}\n\t\t} else if (!invalidHeaderCharRegex.test(obj[name])) {\n\t\t\theaders[MAP][name] = [obj[name]];\n\t\t}\n\t}\n\treturn headers;\n}\n\nconst INTERNALS$1 = Symbol('Response internals');\n\n// fix an issue where \"STATUS_CODES\" aren't a named export for node <10\nconst STATUS_CODES = http.STATUS_CODES;\n\n/**\n * Response class\n *\n * @param   Stream  body  Readable stream\n * @param   Object  opts  Response options\n * @return  Void\n */\nclass Response {\n\tconstructor() {\n\t\tlet body = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : null;\n\t\tlet opts = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\t\tBody.call(this, body, opts);\n\n\t\tconst status = opts.status || 200;\n\t\tconst headers = new Headers(opts.headers);\n\n\t\tif (body != null && !headers.has('Content-Type')) {\n\t\t\tconst contentType = extractContentType(body);\n\t\t\tif (contentType) {\n\t\t\t\theaders.append('Content-Type', contentType);\n\t\t\t}\n\t\t}\n\n\t\tthis[INTERNALS$1] = {\n\t\t\turl: opts.url,\n\t\t\tstatus,\n\t\t\tstatusText: opts.statusText || STATUS_CODES[status],\n\t\t\theaders,\n\t\t\tcounter: opts.counter\n\t\t};\n\t}\n\n\tget url() {\n\t\treturn this[INTERNALS$1].url || '';\n\t}\n\n\tget status() {\n\t\treturn this[INTERNALS$1].status;\n\t}\n\n\t/**\n  * Convenience property representing if the request ended normally\n  */\n\tget ok() {\n\t\treturn this[INTERNALS$1].status >= 200 && this[INTERNALS$1].status < 300;\n\t}\n\n\tget redirected() {\n\t\treturn this[INTERNALS$1].counter > 0;\n\t}\n\n\tget statusText() {\n\t\treturn this[INTERNALS$1].statusText;\n\t}\n\n\tget headers() {\n\t\treturn this[INTERNALS$1].headers;\n\t}\n\n\t/**\n  * Clone this response\n  *\n  * @return  Response\n  */\n\tclone() {\n\t\treturn new Response(clone(this), {\n\t\t\turl: this.url,\n\t\t\tstatus: this.status,\n\t\t\tstatusText: this.statusText,\n\t\t\theaders: this.headers,\n\t\t\tok: this.ok,\n\t\t\tredirected: this.redirected\n\t\t});\n\t}\n}\n\nBody.mixIn(Response.prototype);\n\nObject.defineProperties(Response.prototype, {\n\turl: { enumerable: true },\n\tstatus: { enumerable: true },\n\tok: { enumerable: true },\n\tredirected: { enumerable: true },\n\tstatusText: { enumerable: true },\n\theaders: { enumerable: true },\n\tclone: { enumerable: true }\n});\n\nObject.defineProperty(Response.prototype, Symbol.toStringTag, {\n\tvalue: 'Response',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nconst INTERNALS$2 = Symbol('Request internals');\n\n// fix an issue where \"format\", \"parse\" aren't a named export for node <10\nconst parse_url = Url.parse;\nconst format_url = Url.format;\n\nconst streamDestructionSupported = 'destroy' in Stream.Readable.prototype;\n\n/**\n * Check if a value is an instance of Request.\n *\n * @param   Mixed   input\n * @return  Boolean\n */\nfunction isRequest(input) {\n\treturn typeof input === 'object' && typeof input[INTERNALS$2] === 'object';\n}\n\nfunction isAbortSignal(signal) {\n\tconst proto = signal && typeof signal === 'object' && Object.getPrototypeOf(signal);\n\treturn !!(proto && proto.constructor.name === 'AbortSignal');\n}\n\n/**\n * Request class\n *\n * @param   Mixed   input  Url or Request instance\n * @param   Object  init   Custom options\n * @return  Void\n */\nclass Request {\n\tconstructor(input) {\n\t\tlet init = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\t\tlet parsedURL;\n\n\t\t// normalize input\n\t\tif (!isRequest(input)) {\n\t\t\tif (input && input.href) {\n\t\t\t\t// in order to support Node.js' Url objects; though WHATWG's URL objects\n\t\t\t\t// will fall into this branch also (since their `toString()` will return\n\t\t\t\t// `href` property anyway)\n\t\t\t\tparsedURL = parse_url(input.href);\n\t\t\t} else {\n\t\t\t\t// coerce input to a string before attempting to parse\n\t\t\t\tparsedURL = parse_url(`${input}`);\n\t\t\t}\n\t\t\tinput = {};\n\t\t} else {\n\t\t\tparsedURL = parse_url(input.url);\n\t\t}\n\n\t\tlet method = init.method || input.method || 'GET';\n\t\tmethod = method.toUpperCase();\n\n\t\tif ((init.body != null || isRequest(input) && input.body !== null) && (method === 'GET' || method === 'HEAD')) {\n\t\t\tthrow new TypeError('Request with GET/HEAD method cannot have body');\n\t\t}\n\n\t\tlet inputBody = init.body != null ? init.body : isRequest(input) && input.body !== null ? clone(input) : null;\n\n\t\tBody.call(this, inputBody, {\n\t\t\ttimeout: init.timeout || input.timeout || 0,\n\t\t\tsize: init.size || input.size || 0\n\t\t});\n\n\t\tconst headers = new Headers(init.headers || input.headers || {});\n\n\t\tif (inputBody != null && !headers.has('Content-Type')) {\n\t\t\tconst contentType = extractContentType(inputBody);\n\t\t\tif (contentType) {\n\t\t\t\theaders.append('Content-Type', contentType);\n\t\t\t}\n\t\t}\n\n\t\tlet signal = isRequest(input) ? input.signal : null;\n\t\tif ('signal' in init) signal = init.signal;\n\n\t\tif (signal != null && !isAbortSignal(signal)) {\n\t\t\tthrow new TypeError('Expected signal to be an instanceof AbortSignal');\n\t\t}\n\n\t\tthis[INTERNALS$2] = {\n\t\t\tmethod,\n\t\t\tredirect: init.redirect || input.redirect || 'follow',\n\t\t\theaders,\n\t\t\tparsedURL,\n\t\t\tsignal\n\t\t};\n\n\t\t// node-fetch-only options\n\t\tthis.follow = init.follow !== undefined ? init.follow : input.follow !== undefined ? input.follow : 20;\n\t\tthis.compress = init.compress !== undefined ? init.compress : input.compress !== undefined ? input.compress : true;\n\t\tthis.counter = init.counter || input.counter || 0;\n\t\tthis.agent = init.agent || input.agent;\n\t}\n\n\tget method() {\n\t\treturn this[INTERNALS$2].method;\n\t}\n\n\tget url() {\n\t\treturn format_url(this[INTERNALS$2].parsedURL);\n\t}\n\n\tget headers() {\n\t\treturn this[INTERNALS$2].headers;\n\t}\n\n\tget redirect() {\n\t\treturn this[INTERNALS$2].redirect;\n\t}\n\n\tget signal() {\n\t\treturn this[INTERNALS$2].signal;\n\t}\n\n\t/**\n  * Clone this request\n  *\n  * @return  Request\n  */\n\tclone() {\n\t\treturn new Request(this);\n\t}\n}\n\nBody.mixIn(Request.prototype);\n\nObject.defineProperty(Request.prototype, Symbol.toStringTag, {\n\tvalue: 'Request',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nObject.defineProperties(Request.prototype, {\n\tmethod: { enumerable: true },\n\turl: { enumerable: true },\n\theaders: { enumerable: true },\n\tredirect: { enumerable: true },\n\tclone: { enumerable: true },\n\tsignal: { enumerable: true }\n});\n\n/**\n * Convert a Request to Node.js http request options.\n *\n * @param   Request  A Request instance\n * @return  Object   The options object to be passed to http.request\n */\nfunction getNodeRequestOptions(request) {\n\tconst parsedURL = request[INTERNALS$2].parsedURL;\n\tconst headers = new Headers(request[INTERNALS$2].headers);\n\n\t// fetch step 1.3\n\tif (!headers.has('Accept')) {\n\t\theaders.set('Accept', '*/*');\n\t}\n\n\t// Basic fetch\n\tif (!parsedURL.protocol || !parsedURL.hostname) {\n\t\tthrow new TypeError('Only absolute URLs are supported');\n\t}\n\n\tif (!/^https?:$/.test(parsedURL.protocol)) {\n\t\tthrow new TypeError('Only HTTP(S) protocols are supported');\n\t}\n\n\tif (request.signal && request.body instanceof Stream.Readable && !streamDestructionSupported) {\n\t\tthrow new Error('Cancellation of streamed requests with AbortSignal is not supported in node < 8');\n\t}\n\n\t// HTTP-network-or-cache fetch steps 2.4-2.7\n\tlet contentLengthValue = null;\n\tif (request.body == null && /^(POST|PUT)$/i.test(request.method)) {\n\t\tcontentLengthValue = '0';\n\t}\n\tif (request.body != null) {\n\t\tconst totalBytes = getTotalBytes(request);\n\t\tif (typeof totalBytes === 'number') {\n\t\t\tcontentLengthValue = String(totalBytes);\n\t\t}\n\t}\n\tif (contentLengthValue) {\n\t\theaders.set('Content-Length', contentLengthValue);\n\t}\n\n\t// HTTP-network-or-cache fetch step 2.11\n\tif (!headers.has('User-Agent')) {\n\t\theaders.set('User-Agent', 'node-fetch/1.0 (+https://github.com/bitinn/node-fetch)');\n\t}\n\n\t// HTTP-network-or-cache fetch step 2.15\n\tif (request.compress && !headers.has('Accept-Encoding')) {\n\t\theaders.set('Accept-Encoding', 'gzip,deflate');\n\t}\n\n\tlet agent = request.agent;\n\tif (typeof agent === 'function') {\n\t\tagent = agent(parsedURL);\n\t}\n\n\tif (!headers.has('Connection') && !agent) {\n\t\theaders.set('Connection', 'close');\n\t}\n\n\t// HTTP-network fetch step 4.2\n\t// chunked encoding is handled by Node.js\n\n\treturn Object.assign({}, parsedURL, {\n\t\tmethod: request.method,\n\t\theaders: exportNodeCompatibleHeaders(headers),\n\t\tagent\n\t});\n}\n\n/**\n * abort-error.js\n *\n * AbortError interface for cancelled requests\n */\n\n/**\n * Create AbortError instance\n *\n * @param   String      message      Error message for human\n * @return  AbortError\n */\nfunction AbortError(message) {\n  Error.call(this, message);\n\n  this.type = 'aborted';\n  this.message = message;\n\n  // hide custom error implementation details from end-users\n  Error.captureStackTrace(this, this.constructor);\n}\n\nAbortError.prototype = Object.create(Error.prototype);\nAbortError.prototype.constructor = AbortError;\nAbortError.prototype.name = 'AbortError';\n\n// fix an issue where \"PassThrough\", \"resolve\" aren't a named export for node <10\nconst PassThrough$1 = Stream.PassThrough;\nconst resolve_url = Url.resolve;\n\n/**\n * Fetch function\n *\n * @param   Mixed    url   Absolute url or Request instance\n * @param   Object   opts  Fetch options\n * @return  Promise\n */\nfunction fetch(url, opts) {\n\n\t// allow custom promise\n\tif (!fetch.Promise) {\n\t\tthrow new Error('native promise missing, set fetch.Promise to your favorite alternative');\n\t}\n\n\tBody.Promise = fetch.Promise;\n\n\t// wrap http.request into fetch\n\treturn new fetch.Promise(function (resolve, reject) {\n\t\t// build request object\n\t\tconst request = new Request(url, opts);\n\t\tconst options = getNodeRequestOptions(request);\n\n\t\tconst send = (options.protocol === 'https:' ? https : http).request;\n\t\tconst signal = request.signal;\n\n\t\tlet response = null;\n\n\t\tconst abort = function abort() {\n\t\t\tlet error = new AbortError('The user aborted a request.');\n\t\t\treject(error);\n\t\t\tif (request.body && request.body instanceof Stream.Readable) {\n\t\t\t\trequest.body.destroy(error);\n\t\t\t}\n\t\t\tif (!response || !response.body) return;\n\t\t\tresponse.body.emit('error', error);\n\t\t};\n\n\t\tif (signal && signal.aborted) {\n\t\t\tabort();\n\t\t\treturn;\n\t\t}\n\n\t\tconst abortAndFinalize = function abortAndFinalize() {\n\t\t\tabort();\n\t\t\tfinalize();\n\t\t};\n\n\t\t// send request\n\t\tconst req = send(options);\n\t\tlet reqTimeout;\n\n\t\tif (signal) {\n\t\t\tsignal.addEventListener('abort', abortAndFinalize);\n\t\t}\n\n\t\tfunction finalize() {\n\t\t\treq.abort();\n\t\t\tif (signal) signal.removeEventListener('abort', abortAndFinalize);\n\t\t\tclearTimeout(reqTimeout);\n\t\t}\n\n\t\tif (request.timeout) {\n\t\t\treq.once('socket', function (socket) {\n\t\t\t\treqTimeout = setTimeout(function () {\n\t\t\t\t\treject(new FetchError(`network timeout at: ${request.url}`, 'request-timeout'));\n\t\t\t\t\tfinalize();\n\t\t\t\t}, request.timeout);\n\t\t\t});\n\t\t}\n\n\t\treq.on('error', function (err) {\n\t\t\treject(new FetchError(`request to ${request.url} failed, reason: ${err.message}`, 'system', err));\n\t\t\tfinalize();\n\t\t});\n\n\t\treq.on('response', function (res) {\n\t\t\tclearTimeout(reqTimeout);\n\n\t\t\tconst headers = createHeadersLenient(res.headers);\n\n\t\t\t// HTTP fetch step 5\n\t\t\tif (fetch.isRedirect(res.statusCode)) {\n\t\t\t\t// HTTP fetch step 5.2\n\t\t\t\tconst location = headers.get('Location');\n\n\t\t\t\t// HTTP fetch step 5.3\n\t\t\t\tconst locationURL = location === null ? null : resolve_url(request.url, location);\n\n\t\t\t\t// HTTP fetch step 5.5\n\t\t\t\tswitch (request.redirect) {\n\t\t\t\t\tcase 'error':\n\t\t\t\t\t\treject(new FetchError(`uri requested responds with a redirect, redirect mode is set to error: ${request.url}`, 'no-redirect'));\n\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\treturn;\n\t\t\t\t\tcase 'manual':\n\t\t\t\t\t\t// node-fetch-specific step: make manual redirect a bit easier to use by setting the Location header value to the resolved URL.\n\t\t\t\t\t\tif (locationURL !== null) {\n\t\t\t\t\t\t\t// handle corrupted header\n\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\theaders.set('Location', locationURL);\n\t\t\t\t\t\t\t} catch (err) {\n\t\t\t\t\t\t\t\t// istanbul ignore next: nodejs server prevent invalid response headers, we can't test this through normal request\n\t\t\t\t\t\t\t\treject(err);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\tbreak;\n\t\t\t\t\tcase 'follow':\n\t\t\t\t\t\t// HTTP-redirect fetch step 2\n\t\t\t\t\t\tif (locationURL === null) {\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 5\n\t\t\t\t\t\tif (request.counter >= request.follow) {\n\t\t\t\t\t\t\treject(new FetchError(`maximum redirect reached at: ${request.url}`, 'max-redirect'));\n\t\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 6 (counter increment)\n\t\t\t\t\t\t// Create a new Request object.\n\t\t\t\t\t\tconst requestOpts = {\n\t\t\t\t\t\t\theaders: new Headers(request.headers),\n\t\t\t\t\t\t\tfollow: request.follow,\n\t\t\t\t\t\t\tcounter: request.counter + 1,\n\t\t\t\t\t\t\tagent: request.agent,\n\t\t\t\t\t\t\tcompress: request.compress,\n\t\t\t\t\t\t\tmethod: request.method,\n\t\t\t\t\t\t\tbody: request.body,\n\t\t\t\t\t\t\tsignal: request.signal,\n\t\t\t\t\t\t\ttimeout: request.timeout,\n\t\t\t\t\t\t\tsize: request.size\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 9\n\t\t\t\t\t\tif (res.statusCode !== 303 && request.body && getTotalBytes(request) === null) {\n\t\t\t\t\t\t\treject(new FetchError('Cannot follow redirect with body being a readable stream', 'unsupported-redirect'));\n\t\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 11\n\t\t\t\t\t\tif (res.statusCode === 303 || (res.statusCode === 301 || res.statusCode === 302) && request.method === 'POST') {\n\t\t\t\t\t\t\trequestOpts.method = 'GET';\n\t\t\t\t\t\t\trequestOpts.body = undefined;\n\t\t\t\t\t\t\trequestOpts.headers.delete('content-length');\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 15\n\t\t\t\t\t\tresolve(fetch(new Request(locationURL, requestOpts)));\n\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// prepare response\n\t\t\tres.once('end', function () {\n\t\t\t\tif (signal) signal.removeEventListener('abort', abortAndFinalize);\n\t\t\t});\n\t\t\tlet body = res.pipe(new PassThrough$1());\n\n\t\t\tconst response_options = {\n\t\t\t\turl: request.url,\n\t\t\t\tstatus: res.statusCode,\n\t\t\t\tstatusText: res.statusMessage,\n\t\t\t\theaders: headers,\n\t\t\t\tsize: request.size,\n\t\t\t\ttimeout: request.timeout,\n\t\t\t\tcounter: request.counter\n\t\t\t};\n\n\t\t\t// HTTP-network fetch step 12.1.1.3\n\t\t\tconst codings = headers.get('Content-Encoding');\n\n\t\t\t// HTTP-network fetch step 12.1.1.4: handle content codings\n\n\t\t\t// in following scenarios we ignore compression support\n\t\t\t// 1. compression support is disabled\n\t\t\t// 2. HEAD request\n\t\t\t// 3. no Content-Encoding header\n\t\t\t// 4. no content response (204)\n\t\t\t// 5. content not modified response (304)\n\t\t\tif (!request.compress || request.method === 'HEAD' || codings === null || res.statusCode === 204 || res.statusCode === 304) {\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// For Node v6+\n\t\t\t// Be less strict when decoding compressed responses, since sometimes\n\t\t\t// servers send slightly invalid responses that are still accepted\n\t\t\t// by common browsers.\n\t\t\t// Always using Z_SYNC_FLUSH is what cURL does.\n\t\t\tconst zlibOptions = {\n\t\t\t\tflush: zlib.Z_SYNC_FLUSH,\n\t\t\t\tfinishFlush: zlib.Z_SYNC_FLUSH\n\t\t\t};\n\n\t\t\t// for gzip\n\t\t\tif (codings == 'gzip' || codings == 'x-gzip') {\n\t\t\t\tbody = body.pipe(zlib.createGunzip(zlibOptions));\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// for deflate\n\t\t\tif (codings == 'deflate' || codings == 'x-deflate') {\n\t\t\t\t// handle the infamous raw deflate response from old servers\n\t\t\t\t// a hack for old IIS and Apache servers\n\t\t\t\tconst raw = res.pipe(new PassThrough$1());\n\t\t\t\traw.once('data', function (chunk) {\n\t\t\t\t\t// see http://stackoverflow.com/questions/37519828\n\t\t\t\t\tif ((chunk[0] & 0x0F) === 0x08) {\n\t\t\t\t\t\tbody = body.pipe(zlib.createInflate());\n\t\t\t\t\t} else {\n\t\t\t\t\t\tbody = body.pipe(zlib.createInflateRaw());\n\t\t\t\t\t}\n\t\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\t\tresolve(response);\n\t\t\t\t});\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// for br\n\t\t\tif (codings == 'br' && typeof zlib.createBrotliDecompress === 'function') {\n\t\t\t\tbody = body.pipe(zlib.createBrotliDecompress());\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// otherwise, use response as-is\n\t\t\tresponse = new Response(body, response_options);\n\t\t\tresolve(response);\n\t\t});\n\n\t\twriteToStream(req, request);\n\t});\n}\n/**\n * Redirect code matching\n *\n * @param   Number   code  Status code\n * @return  Boolean\n */\nfetch.isRedirect = function (code) {\n\treturn code === 301 || code === 302 || code === 303 || code === 307 || code === 308;\n};\n\n// expose Promise\nfetch.Promise = global.Promise;\n\nexport default fetch;\nexport { Headers, Request, Response, FetchError };\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n  kernelMs: number|{error: string};\n  getExtraProfileInfo?(): string;  // a field for additional timing information\n                                   // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n  disposeData(dataId: DataId): void;\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void;\n  memory(): {unreliable: boolean;};  // Backend-specific information.\n  /** Returns number of data ids currently in the storage. */\n  numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage<T> {\n  private data = new WeakMap<DataId, T>();\n  private dataIdsCount = 0;\n\n  constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n  get(dataId: DataId) {\n    if (!this.data.has(dataId)) {\n      this.dataMover.moveData(this.backend, dataId);\n    }\n    return this.data.get(dataId);\n  }\n\n  set(dataId: DataId, value: T): void {\n    this.dataIdsCount++;\n    this.data.set(dataId, value);\n  }\n\n  has(dataId: DataId): boolean {\n    return this.data.has(dataId);\n  }\n\n  delete(dataId: DataId): boolean {\n    this.dataIdsCount--;\n    return this.data.delete(dataId);\n  }\n\n  numDataIds(): number {\n    return this.dataIdsCount;\n  }\n}\n\nexport interface DataMover {\n  /**\n   * To be called by backends whenever they see a dataId that they don't own.\n   * Upon calling this method, the mover will fetch the tensor from another\n   * backend and register it with the current active backend.\n   */\n  moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo>;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo> {\n    return notYetImplemented('time');\n  }\n  read(dataId: object): Promise<BackendValues> {\n    return notYetImplemented('read');\n  }\n  readSync(dataId: object): BackendValues {\n    return notYetImplemented('readSync');\n  }\n  numDataIds(): number {\n    return notYetImplemented('numDataIds');\n  }\n  disposeData(dataId: object): void {\n    return notYetImplemented('disposeData');\n  }\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    return notYetImplemented('write');\n  }\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    return notYetImplemented('move');\n  }\n  memory(): {unreliable: boolean; reasons?: string[]} {\n    return notYetImplemented('memory');\n  }\n  /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n  floatPrecision(): 16|32 {\n    return notYetImplemented('floatPrecision');\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    return notYetImplemented('batchMatMul');\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          FusedBatchMatMulConfig): Tensor3D {\n    return notYetImplemented('fusedBatchMatMul');\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    return notYetImplemented('slice');\n  }\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    return notYetImplemented('stridedSlice');\n  }\n  unstack(x: Tensor, axis: number): Tensor[] {\n    return notYetImplemented('unstack');\n  }\n  reverse<T extends Tensor>(a: T, axis: number[]): T {\n    return notYetImplemented('reverse');\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    return notYetImplemented('concat');\n  }\n\n  neg<T extends Tensor>(a: T): T {\n    return notYetImplemented('neg');\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('add');\n  }\n  addN<T extends Tensor>(tensors: T[]): T {\n    return notYetImplemented('addN');\n  }\n  subtract(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('subtract');\n  }\n  multiply(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('multiply');\n  }\n  realDivide(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('realDivide');\n  }\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('floorDiv');\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('sum');\n  }\n  prod(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('prod');\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    return notYetImplemented('unsortedSegmentSum');\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMin');\n  }\n  argMax(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMax');\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('equal');\n  }\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('notEqual');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('less');\n  }\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('lessEqual');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greater');\n  }\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greaterEqual');\n  }\n\n  logicalNot<T extends Tensor>(a: T): T {\n    return notYetImplemented('logicalNot');\n  }\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalAnd');\n  }\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalOr');\n  }\n\n  where(condition: Tensor): Tensor2D {\n    return notYetImplemented('where');\n  }\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('select');\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    return notYetImplemented('topk');\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('min');\n  }\n  minimum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('minimum');\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('mod');\n  }\n\n  max(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('max');\n  }\n  maximum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('maximum');\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('all');\n  }\n  any(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('any');\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('squaredDifference');\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    return notYetImplemented('ceil');\n  }\n  floor<T extends Tensor>(x: T): T {\n    return notYetImplemented('floor');\n  }\n  round<T extends Tensor>(x: T): T {\n    return notYetImplemented('round');\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    return notYetImplemented('sign');\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    return notYetImplemented('isNaN');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    return notYetImplemented('isInf');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    return notYetImplemented('isFinite');\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    return notYetImplemented('pow');\n  }\n  exp<T extends Tensor>(x: T): T {\n    return notYetImplemented('exp');\n  }\n  expm1<T extends Tensor>(x: T): T {\n    return notYetImplemented('expm1');\n  }\n  softmax<T extends Tensor>(x: T, dim: number): T {\n    return notYetImplemented('softmax');\n  }\n  log<T extends Tensor>(x: T): T {\n    return notYetImplemented('log');\n  }\n  log1p<T extends Tensor>(x: T): T {\n    return notYetImplemented('log1p');\n  }\n  sqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('sqrt');\n  }\n  rsqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('rsqrt');\n  }\n  square<T extends Tensor>(x: T): T {\n    return notYetImplemented('square');\n  }\n  reciprocal<T extends Tensor>(x: T): T {\n    return notYetImplemented('reciprocal');\n  }\n  relu<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu');\n  }\n  relu6<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu6');\n  }\n  prelu<T extends Tensor>(x: T, a: T): T {\n    return notYetImplemented('prelu');\n  }\n  elu<T extends Tensor>(x: T): T {\n    return notYetImplemented('elu');\n  }\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    return notYetImplemented('eluDer');\n  }\n  selu<T extends Tensor>(x: T): T {\n    return notYetImplemented('selu');\n  }\n  int<T extends Tensor>(x: T): T {\n    return notYetImplemented('int');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    return notYetImplemented('clip');\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    return notYetImplemented('abs');\n  }\n  complexAbs<T extends Tensor>(x: T): T {\n    return notYetImplemented('complexAbs');\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    return notYetImplemented('sigmoid');\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    return notYetImplemented('softplus');\n  }\n\n  sin<T extends Tensor>(x: T): T {\n    return notYetImplemented('sin');\n  }\n  cos<T extends Tensor>(x: T): T {\n    return notYetImplemented('cos');\n  }\n  tan<T extends Tensor>(x: T): T {\n    return notYetImplemented('tan');\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    return notYetImplemented('asin');\n  }\n  acos<T extends Tensor>(x: T): T {\n    return notYetImplemented('acos');\n  }\n  atan<T extends Tensor>(x: T): T {\n    return notYetImplemented('atan');\n  }\n  atan2<T extends Tensor>(a: T, b: T): T {\n    return notYetImplemented('atan2');\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('sinh');\n  }\n  cosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('cosh');\n  }\n  tanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('tanh');\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('asinh');\n  }\n  acosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('acosh');\n  }\n  atanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('atanh');\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    return notYetImplemented('erf');\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    return notYetImplemented('step');\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedConv2d');\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2d');\n  }\n  conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('conv2dDerInput');\n  }\n  conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2dDerFilter');\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedDepthwiseConv2D');\n  }\n\n  depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2D');\n  }\n  depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerInput');\n  }\n  depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerFilter');\n  }\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3d');\n  }\n  conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n      Tensor5D {\n    return notYetImplemented('conv3dDerInput');\n  }\n  conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3dDerFilter');\n  }\n  maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('maxPool');\n  }\n  maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('maxPoolBackprop');\n  }\n  avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPool');\n  }\n  avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPoolBackprop');\n  }\n  avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3d');\n  }\n  avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3dBackprop');\n  }\n  maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3d');\n  }\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3dBackprop');\n  }\n\n  reshape<T extends Tensor, R extends Rank>(x: T, shape: ShapeMap[R]):\n      Tensor<R> {\n    return notYetImplemented('reshape');\n  }\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return notYetImplemented('cast');\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    return notYetImplemented('tile');\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    return notYetImplemented('pad');\n  }\n\n  transpose<T extends Tensor>(x: T, perm: number[]): T {\n    return notYetImplemented('transpose');\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    return notYetImplemented('gather');\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    return notYetImplemented('gatherND');\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    return notYetImplemented('scatterND');\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    return notYetImplemented('batchToSpaceND');\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: number[][]): T {\n    return notYetImplemented('spaceToBatchND');\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeBilinear');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    return notYetImplemented('resizeBilinearBackprop');\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHEight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighbor');\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighborBackprop');\n  }\n\n  batchNorm(\n      x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n      offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n      varianceEpsilon?: number): Tensor4D {\n    return notYetImplemented('batchNorm');\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    return notYetImplemented('localResponseNormalization4D');\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n      bias: number, alpha: number, beta: number): Tensor4D {\n    return notYetImplemented('LRNGrad');\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    return notYetImplemented('multinomial');\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    return notYetImplemented('oneHot');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    return notYetImplemented('cumsum');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold?: number): Tensor1D {\n    return notYetImplemented('nonMaxSuppression');\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('fft');\n  }\n  ifft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('ifft');\n  }\n  complex<T extends Tensor>(real: T, imag: T): T {\n    return notYetImplemented('complex');\n  }\n  real<T extends Tensor>(input: T): T {\n    return notYetImplemented('real');\n  }\n  imag<T extends Tensor>(input: T): T {\n    return notYetImplemented('imag');\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    return notYetImplemented('cropAndResize');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n    return notYetImplemented('depthToSpace');\n  }\n\n  // Aligns with the \"SplitV\" kernel in TensorFlow.\n  split<T extends Tensor>(value: T, sizeSplits: number[], axis: number): T[] {\n    return notYetImplemented('split');\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    return notYetImplemented('sparseToDense');\n  }\n\n  diag(x: Tensor): Tensor {\n    return notYetImplemented('diag');\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    return notYetImplemented('fill');\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('onesLike');\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('zerosLike');\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return notYetImplemented('linspace');\n  }\n\n  dispose(): void {\n    return notYetImplemented('dispose');\n  }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n  throw new Error(\n      `'${kernelName}' not yet implemented or not found in the registry. ` +\n      `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise<FlagValue>);\nexport type Flags = {\n  [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n  evaluationFn: FlagEvaluationFn;\n  setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n  private flags: Flags = {};\n  private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n  private urlFlags: Flags = {};\n\n  platformName: string;\n  platform: Platform;\n\n  // tslint:disable-next-line: no-any\n  constructor(public global: any) {\n    this.populateURLFlags();\n  }\n\n  setPlatform(platformName: string, platform: Platform) {\n    if (this.platform != null) {\n      console.warn(\n          `Platform ${this.platformName} has already been set. ` +\n          `Overwriting the platform with ${platform}.`);\n    }\n    this.platformName = platformName;\n    this.platform = platform;\n  }\n\n  registerFlag(\n      flagName: string, evaluationFn: FlagEvaluationFn,\n      setHook?: (value: FlagValue) => void) {\n    this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n    // Override the flag value from the URL. This has to happen here because the\n    // environment is initialized before flags get registered.\n    if (this.urlFlags[flagName] != null) {\n      const flagValue = this.urlFlags[flagName];\n      console.warn(\n          `Setting feature override from URL ${flagName}: ${flagValue}.`);\n      this.set(flagName, flagValue);\n    }\n  }\n\n  async getAsync(flagName: string): Promise<FlagValue> {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    this.flags[flagName] = await this.evaluateFlag(flagName);\n    return this.flags[flagName];\n  }\n\n  get(flagName: string): FlagValue {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    const flagValue = this.evaluateFlag(flagName);\n    if (flagValue instanceof Promise) {\n      throw new Error(\n          `Flag ${flagName} cannot be synchronously evaluated. ` +\n          `Please use getAsync() instead.`);\n    }\n\n    this.flags[flagName] = flagValue;\n\n    return this.flags[flagName];\n  }\n\n  getNumber(flagName: string): number {\n    return this.get(flagName) as number;\n  }\n\n  getBool(flagName: string): boolean {\n    return this.get(flagName) as boolean;\n  }\n\n  getFlags(): Flags {\n    return this.flags;\n  }\n  // For backwards compatibility.\n  get features(): Flags {\n    return this.flags;\n  }\n\n  set(flagName: string, value: FlagValue): void {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot set flag ${flagName} as it has not been registered.`);\n    }\n    this.flags[flagName] = value;\n    if (this.flagRegistry[flagName].setHook != null) {\n      this.flagRegistry[flagName].setHook(value);\n    }\n  }\n\n  private evaluateFlag(flagName: string): FlagValue|Promise<FlagValue> {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n    }\n    return this.flagRegistry[flagName].evaluationFn();\n  }\n\n  setFlags(flags: Flags) {\n    this.flags = Object.assign({}, flags);\n  }\n\n  reset() {\n    this.flags = {};\n    this.urlFlags = {};\n    this.populateURLFlags();\n  }\n\n  private populateURLFlags(): void {\n    if (typeof this.global === 'undefined' ||\n        typeof this.global.location === 'undefined' ||\n        typeof this.global.location.search === 'undefined') {\n      return;\n    }\n\n    const urlParams = getQueryParams(this.global.location.search);\n    if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n      const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n      keyValues.forEach(keyValue => {\n        const [key, value] = keyValue.split(':') as [string, string];\n        this.urlFlags[key] = parseValue(key, value);\n      });\n    }\n  }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n  const params = {};\n  queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n    decodeParam(params, t[0], t[1]);\n    return t.join('=');\n  });\n  return params;\n}\n\nfunction decodeParam(\n    params: {[key: string]: string}, name: string, value?: string) {\n  params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n  value = value.toLowerCase();\n  if (value === 'true' || value === 'false') {\n    return value === 'true';\n  } else if (`${+ value}` === value) {\n    return +value;\n  }\n  throw new Error(\n      `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n  return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n  ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map<string, any>};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map<string, any>} {\n  if (globalNameSpace == null) {\n    // tslint:disable-next-line:no-any\n    let ns: any;\n    if (typeof (window) !== 'undefined') {\n      ns = window;\n    } else if (typeof (global) !== 'undefined') {\n      ns = global;\n    } else if (typeof (process) !== 'undefined') {\n      ns = process;\n    } else if (typeof (self) !== 'undefined') {\n      ns = self;\n    } else {\n      throw new Error('Could not find a global object');\n    }\n    globalNameSpace = ns;\n  }\n  return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map<string, any> {\n  const ns = getGlobalNamespace();\n  if (ns._tfGlobals == null) {\n    ns._tfGlobals = new Map();\n  }\n  return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n *             the first time it is fetched.\n */\nexport function getGlobal<T>(key: string, init: () => T): T {\n  const globalMap = getGlobalMap();\n  if (globalMap.has(key)) {\n    return globalMap.get(key);\n  } else {\n    const singleton = init();\n    globalMap.set(key, singleton);\n    return globalMap.get(key);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AllAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AnyAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMaxAttrs {\n  axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMinAttrs {\n  axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\nexport interface BatchMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BatchToSpaceNDAttrs {\n  blockShape: number[];\n  crops: number[][];\n}\n\nexport type BinaryInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BroadCastToAttrs {\n  shape: number[];\n  inputShape: number[];  // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n  dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n  clipValueMin: number;\n  clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick<NamedTensorInfoMap, 'real'|'imag'>;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n  axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'dy'>;\nexport interface Conv2DBackpropFilterAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'|'filter'>;\nexport interface Conv2DBackpropInputAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv3DAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'y'>;\n\nexport interface Conv3DBackpropFilterAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'>;\nexport interface Conv3DBackpropInputAttrs {\n  pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface CumsumAttrs {\n  axis: number;\n  exclusive: boolean;\n  reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n    Pick<NamedTensorInfoMap, 'image'|'boxes'|'boxInd'>;\nexport interface CropAndResizeAttrs {\n  cropSize: [number, number];\n  method: 'bilinear'|'nearest';\n  extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface DepthToSpaceAttrs {\n  blockSize: number;\n  dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface DepthwiseConv2dNativeAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n    'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'dy'>;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n    'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'dy'>;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Dilation2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick<NamedTensorInfoMap, 'dy'|'y'>;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n  shape: number[];\n  value: number|string;\n  dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick<NamedTensorInfoMap, 'image'>;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n    Pick<NamedTensorInfoMap, 'x'|'scale'|'offset'|'mean'|'variance'>;\nexport interface FusedBatchNormAttrs {\n  varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick<NamedTensorInfoMap, 'x'|'indices'>;\nexport interface GatherV2Attrs {\n  axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick<NamedTensorInfoMap, 'params'|'indices'>;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n  start: number;\n  stop: number;\n  num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface LogSoftmaxAttrs {\n  axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface LRNAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick<NamedTensorInfoMap, 'x'|'y'|'dy'>;\nexport interface LRNBackpropAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxAttrs {\n  reductionIndices: number|number[];\n  keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolWithArgmaxAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MeanAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MinAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV3Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV4Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV5Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick<NamedTensorInfoMap, 'indices'>;\nexport interface OneHotAttrs {\n  depth: number;\n  onValue: number;\n  offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface PadV2Attrs {\n  paddings: Array<[number, number]>;\n  constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick<NamedTensorInfoMap, 'x'|'alpha'>;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ProdAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n  start: number;\n  stop: number;\n  step: number;\n  dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReshapeAttrs {\n  shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeNearestNeighborAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n    Pick<NamedTensorInfoMap, 'images'>;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeBilinearAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick<NamedTensorInfoMap, 'images'>;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReverseAttrs {\n  dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick<NamedTensorInfoMap, 'indices'|'updates'>;\nexport interface ScatterNdAttrs {\n  shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick<NamedTensorInfoMap, 'condition'|'t'|'e'>;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SliceAttrs {\n  begin: number|number[];\n  size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SumAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SpaceToBatchNDAttrs {\n  blockShape: number[];\n  paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SplitVAttrs {\n  numOrSizeSplits: number[]|number;\n  axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface SoftmaxAttrs {\n  dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n    Pick<NamedTensorInfoMap, 'sparseIndices'|'sparseValues'|'defaultValue'>;\nexport interface SparseToDenseAttrs {\n  outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface StridedSliceAttrs {\n  begin: number[];\n  end: number[];\n  strides: number[];\n  beginMask: number;\n  endMask: number;\n  ellipsisMask: number;\n  newAxisMask: number;\n  shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TileAttrs {\n  reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TopKAttrs {\n  k: number;\n  sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TransposeAttrs {\n  perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface UniqueAttrs {\n  axis: number;\n}\n\nexport type UnaryInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick<NamedTensorInfoMap, 'value'>;\nexport interface UnpackAttrs {\n  axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n    Pick<NamedTensorInfoMap, 'x'|'segmentIds'>;\nexport interface UnsortedSegmentSumAttrs {\n  numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n  alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n  pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n      HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n  numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick<NamedTensorInfoMap, 'image'>;\nexport interface RotateWithOffsetAttrs {\n  radians: number;\n  fillValue: number|[number, number, number];\n  center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n  a: TensorInfo;\n  b: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n  activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n    getGlobal('kernelRegistry', () => new Map<string, KernelConfig>());\nconst gradRegistry =\n    getGlobal('gradRegistry', () => new Map<string, GradConfig>());\n\nexport type DataId = object;\n\ntype AttributeValue =\n    number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray<AttributeValue>;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n  inputs: NamedTensorInfoMap,\n  backend: {},\n  attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n    (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n        NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n  kernelName: string;\n  backendName: string;\n  kernelFunc: KernelFunc;\n  setupFunc?: KernelSetupFunc;\n  disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n  kernelName: string;\n  inputsToSave?: string[];\n  // When saveAllInputs is true, all inputs will be saved. Only use this flag\n  // if inputs is an array of Tensors.\n  saveAllInputs?: boolean;\n  outputsToSave?: boolean[];\n  gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n  dataId: DataId;\n  shape: number[];\n  dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n  [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n  [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n    kernelName: string, backendName: string): KernelConfig {\n  const key = makeKey(kernelName, backendName);\n  return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n  return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n  const it = kernelRegistry.entries();\n  const result: KernelConfig[] = [];\n\n  while (true) {\n    const {done, value} = it.next();\n    if (done) {\n      break;\n    }\n    const [key, config] = value;\n    const [backend, ] = key.split('_');\n    if (backend === backendName) {\n      result.push(config);\n    }\n  }\n  return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n  const {kernelName, backendName} = config;\n  const key = makeKey(kernelName, backendName);\n  if (kernelRegistry.has(key)) {\n    console.warn(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is already registered`);\n  }\n  kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n  const {kernelName} = config;\n\n  if (gradRegistry.has(kernelName)) {\n    // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n    // to debug mode.\n    if (env().getBool('DEBUG')) {\n      console.warn(`Overriding the gradient for '${kernelName}'`);\n    }\n  }\n  gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n    kernelName: string, backendName: string): void {\n  const key = makeKey(kernelName, backendName);\n  if (!kernelRegistry.has(key)) {\n    throw new Error(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is not registered`);\n  }\n  kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n  if (!gradRegistry.has(kernelName)) {\n    throw new Error(\n        `The gradient '${kernelName}' for backend is not registered`);\n  }\n  gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n    registeredBackendName: string, newBackendName: string): void {\n  const kernels = getKernelsForBackend(registeredBackendName);\n  kernels.forEach(kernelConfig => {\n    const newKernelConfig =\n        Object.assign({}, kernelConfig, {backendName: newBackendName});\n    registerKernel(newKernelConfig);\n  });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n  return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n                        Float32Array): void {\n  let counter = array.length;\n  let temp = 0;\n  let index = 0;\n  // While there are elements in the array\n  while (counter > 0) {\n    // Pick a random index\n    index = (Math.random() * counter) | 0;\n    // Decrease counter by 1\n    counter--;\n    // And swap the last element with it\n    temp = array[counter];\n    array[counter] = array[index];\n    array[index] = temp;\n  }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n  return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n  return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n  let sum = 0;\n  for (let i = 0; i < arr.length; i++) {\n    sum += arr[i];\n  }\n  return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n  const r = Math.random();\n  return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n  let result = 0;\n  for (let i = 0; i < a.length; i++) {\n    const diff = Number(a[i]) - Number(b[i]);\n    result += diff * diff;\n  }\n  return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n *     error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n  if (!expr) {\n    throw new Error(typeof msg === 'string' ? msg : msg());\n  }\n}\n\nexport function assertShapesMatch(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  assert(\n      arraysEqual(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n  assert(\n      a != null,\n      () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten<number>().\n/**\n *  Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n *  @param arr The nested array to flatten.\n *  @param result The destination array which holds the elements.\n *  @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n *      to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten<T extends number|boolean|string|Promise<number>|TypedArray>(\n    arr: T|RecursiveArray<T>, result: T[] = [], skipTypedArray = false): T[] {\n  if (result == null) {\n    result = [];\n  }\n  if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n    for (let i = 0; i < arr.length; ++i) {\n      flatten(arr[i], result, skipTypedArray);\n    }\n  } else {\n    result.push(arr as T);\n  }\n  return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n  if (shape.length === 0) {\n    // Scalar.\n    return 1;\n  }\n  let size = shape[0];\n  for (let i = 1; i < shape.length; i++) {\n    size *= shape[i];\n  }\n  return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n  return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n  if (n1 === n2) {\n    return true;\n  }\n  if (n1 == null || n2 == null) {\n    return false;\n  }\n\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function isInt(a: number): boolean {\n  return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n  // tslint:disable-next-line:no-any\n  if ((Math as any).tanh != null) {\n    // tslint:disable-next-line:no-any\n    return (Math as any).tanh(x);\n  }\n  if (x === Infinity) {\n    return 1;\n  } else if (x === -Infinity) {\n    return -1;\n  } else {\n    const e2x = Math.exp(2 * x);\n    return (e2x - 1) / (e2x + 1);\n  }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n  const width = Math.ceil(Math.sqrt(size));\n  return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n  const shuffledIndices = new Uint32Array(n);\n  for (let i = 0; i < n; ++i) {\n    shuffledIndices[i] = i;\n  }\n  shuffle(shuffledIndices);\n  return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n  if (size <= a.length) {\n    return a;\n  }\n  return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n    checkFn: () => boolean, delayFn = (counter: number) => 0,\n    maxCounter?: number): Promise<void> {\n  return new Promise<void>((resolve, reject) => {\n    let tryCount = 0;\n\n    const tryFn = () => {\n      if (checkFn()) {\n        resolve();\n        return;\n      }\n\n      tryCount++;\n\n      const nextBackoff = delayFn(tryCount);\n\n      if (maxCounter != null && tryCount >= maxCounter) {\n        reject();\n        return;\n      }\n      setTimeout(tryFn, nextBackoff);\n    };\n\n    tryFn();\n  });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n    shape: number[], size: number): number[] {\n  let shapeProd = 1;\n  let implicitIdx = -1;\n\n  for (let i = 0; i < shape.length; ++i) {\n    if (shape[i] >= 0) {\n      shapeProd *= shape[i];\n    } else if (shape[i] === -1) {\n      if (implicitIdx !== -1) {\n        throw Error(\n            `Shapes can only have 1 implicit size. ` +\n            `Found -1 at dim ${implicitIdx} and dim ${i}`);\n      }\n      implicitIdx = i;\n    } else if (shape[i] < 0) {\n      throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n    }\n  }\n\n  if (implicitIdx === -1) {\n    if (size > 0 && size !== shapeProd) {\n      throw Error(`Size(${size}) must match the product of shape ${shape}`);\n    }\n    return shape;\n  }\n\n  if (shapeProd === 0) {\n    throw Error(\n        `Cannot infer the missing size in [${shape}] when ` +\n        `there are 0 elements`);\n  }\n  if (size % shapeProd !== 0) {\n    throw Error(\n        `The implicit shape can't be a fractional number. ` +\n        `Got ${size} / ${shapeProd}`);\n  }\n\n  const newShape = shape.slice();\n  newShape[implicitIdx] = size / shapeProd;\n  return newShape;\n}\n\nexport function parseAxisParam(\n    axis: number|number[], shape: number[]): number[] {\n  const rank = shape.length;\n\n  // Normalize input\n  axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n  // Check for valid range\n  assert(\n      axis.every(ax => ax >= -rank && ax < rank),\n      () =>\n          `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n          `got axis ${axis}`);\n\n  // Check for only integers\n  assert(\n      axis.every(ax => isInt(ax)),\n      () => `All values in axis param must be integers but ` +\n          `got axis ${axis}`);\n\n  // Handle negative axis.\n  return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n    {newShape: number[], keptDims: number[]} {\n  const newShape: number[] = [];\n  const keptDims: number[] = [];\n  const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n  const axes = (axis == null || isEmptyArray) ?\n      null :\n      parseAxisParam(axis, shape).sort();\n  let j = 0;\n  for (let i = 0; i < shape.length; ++i) {\n    if (axes != null) {\n      if (axes[j] === i && shape[i] !== 1) {\n        throw new Error(\n            `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n      }\n      if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n        newShape.push(shape[i]);\n        keptDims.push(i);\n      }\n      if (axes[j] <= i) {\n        j++;\n      }\n    }\n    if (shape[i] !== 1) {\n      newShape.push(shape[i]);\n      keptDims.push(i);\n    }\n  }\n  return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType<D extends NumericDataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType<D extends DataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else if (dtype === 'string') {\n    values = new Array<'string'>(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors<D extends DataType>(\n    vals: DataTypeMap[D]|number[], dtype: D): void {\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n    }\n  }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n  return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n      dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n  if (newType === 'complex64') {\n    return false;\n  }\n  if (newType === 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'bool' && oldType === 'bool') {\n    return false;\n  }\n  return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n  return a instanceof Float32Array || a instanceof Int32Array ||\n      a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n  if (dtype === 'float32' || dtype === 'int32') {\n    return 4;\n  } else if (dtype === 'complex64') {\n    return 8;\n  } else if (dtype === 'bool') {\n    return 1;\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n  if (arr == null) {\n    return 0;\n  }\n  let bytes = 0;\n  arr.forEach(x => bytes += x.length);\n  return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n  return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n  return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n  return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n  if (Array.isArray(values)) {\n    return inferDtype(values[0]);\n  }\n  if (values instanceof Float32Array) {\n    return 'float32';\n  } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n    return 'int32';\n  } else if (isNumber(values)) {\n    return 'float32';\n  } else if (isString(values)) {\n    return 'string';\n  } else if (isBoolean(values)) {\n    return 'bool';\n  }\n  return 'float32';\n}\n\nexport function isFunction(f: Function) {\n  return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n  for (let i = start; i < size; ++i) {\n    if (size % i === 0) {\n      return i;\n    }\n  }\n  return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n  const rank = shape.length;\n  if (rank < 2) {\n    return [];\n  }\n\n  // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n  // strides.\n  const strides = new Array(rank - 1);\n  strides[rank - 2] = shape[rank - 1];\n  for (let i = rank - 3; i >= 0; --i) {\n    strides[i] = strides[i + 1] * shape[i + 1];\n  }\n  return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n    value: DataType, dtype: DataType): BackendValues {\n  if (dtype === 'string') {\n    return encodeString(value);\n  }\n\n  return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n  if (dtype === 'string') {\n    throw new Error('Cannot convert a string[] to a TypedArray');\n  }\n  if (Array.isArray(a)) {\n    a = flatten(a);\n  }\n\n  if (env().getBool('DEBUG')) {\n    checkConversionForErrors(a as number[], dtype);\n  }\n  if (noConversionNeeded(a, dtype)) {\n    return a as TypedArray;\n  }\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(a as number[]);\n  } else if (dtype === 'int32') {\n    return new Int32Array(a as number[]);\n  } else if (dtype === 'bool') {\n    const bool = new Uint8Array((a as number[]).length);\n    for (let i = 0; i < bool.length; ++i) {\n      if (Math.round((a as number[])[i]) !== 0) {\n        bool[i] = 1;\n      }\n    }\n    return bool;\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n  const ret = new Array();\n  if (shape.length === 1) {\n    const d = shape[0];\n    for (let i = 0; i < d; i++) {\n      ret[i] = a[offset + i];\n    }\n  } else {\n    const d = shape[0];\n    const rest = shape.slice(1);\n    const len = rest.reduce((acc, c) => acc * c);\n    for (let i = 0; i < d; i++) {\n      ret[i] = createNestedArray(offset + i * len, rest, a);\n    }\n  }\n  return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n  if (shape.length === 0) {\n    // Scalar type should return a single number.\n    return a[0];\n  }\n  const size = shape.reduce((acc, c) => acc * c);\n  if (size === 0) {\n    // A tensor with shape zero should be turned into empty list.\n    return [];\n  }\n  if (size !== a.length) {\n    throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n  }\n\n  return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n  return (a instanceof Float32Array && dtype === 'float32') ||\n      (a instanceof Int32Array && dtype === 'int32') ||\n      (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  const array = makeZerosTypedArray(size, dtype);\n  for (let i = 0; i < array.length; i++) {\n    array[i] = 1;\n  }\n  return array;\n}\n\nexport function makeZerosTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'int32') {\n    return new Int32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'bool') {\n    return new Uint8Array(size) as DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray<D extends DataType>(\n    shape: number[], dtype: D) {\n  const size = shape.reduce((prev, curr) => prev * curr, 1);\n  if (dtype == null || dtype === 'float32') {\n    return toNestedArray(shape, new Float32Array(size));\n  } else if (dtype === 'int32') {\n    return toNestedArray(shape, new Int32Array(size));\n  } else if (dtype === 'bool') {\n    return toNestedArray(shape, new Uint8Array(size));\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n  return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n  shape.forEach(dimSize => {\n    assert(\n        Number.isInteger(dimSize) && dimSize >= 0,\n        () =>\n            `Tensor must have a shape comprised of positive integers but got ` +\n            `shape [${shape}].`);\n  });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n    path: string, requestInits?: RequestInit): Promise<Response> {\n  return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n  encoding = encoding || 'utf-8';\n  return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n  encoding = encoding || 'utf-8';\n  return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n    locs: number[], rank: number, strides: number[]): number {\n  if (rank === 0) {\n    return 0;\n  } else if (rank === 1) {\n    return locs[0];\n  }\n  let index = locs[locs.length - 1];\n  for (let i = 0; i < locs.length - 1; ++i) {\n    index += strides[i] * locs[i];\n  }\n  return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n    index: number, rank: number, strides: number[]): number[] {\n  if (rank === 0) {\n    return [];\n  } else if (rank === 1) {\n    return [index];\n  }\n  const locs: number[] = new Array(rank);\n  for (let i = 0; i < locs.length - 1; ++i) {\n    locs[i] = Math.floor(index / strides[i]);\n    index -= locs[i] * strides[i];\n  }\n  locs[locs.length - 1] = index;\n  return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n  kernelName: string,\n  outputs: Tensor[],\n  inputs: NamedTensorMap,\n  timeMs: Promise<number|{error: string}>,\n  extraInfo: Promise<string>\n};\n\nexport class Profiler {\n  constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n    if (logger == null) {\n      this.logger = new Logger();\n    }\n  }\n\n  profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n      KernelProfile {\n    let outputs: Tensor[];\n    const holdResultWrapperFn = () => {\n      outputs = f();\n    };\n    const timer = this.backendTimer.time(holdResultWrapperFn);\n\n    for (let i = 0; i < outputs.length; i++) {\n      const output = outputs[i];\n      // Dangling promise here because we don't want to propagate up\n      // asynchronicity.\n      output.data().then(tensorVals => {\n        checkComputationForErrors(tensorVals, output.dtype, kernelName);\n      });\n    }\n\n    const kernelProfile = {\n      kernelName,\n      outputs,\n      inputs,\n      timeMs: timer.then(timing => timing.kernelMs),\n      extraInfo: timer.then(\n          timing => timing.getExtraProfileInfo != null ?\n              timing.getExtraProfileInfo() :\n              '')\n    };\n    return kernelProfile;\n  }\n\n  logKernelProfile(kernelProfile: KernelProfile): void {\n    const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n    outputs.forEach(result => {\n      Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n        this.logger.logKernelProfile(\n            kernelName, result, valueContainer[0], valueContainer[1], inputs,\n            valueContainer[2]);\n      });\n    });\n  }\n}\n\nexport function checkComputationForErrors<D extends DataType>(\n    vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n  if (dtype !== 'float32') {\n    // Only floating point computations will generate NaN values\n    return false;\n  }\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      // Throwing custom exception so behavior is testable.\n      console.warn(`Found ${num} in the result of '${kernelName}'`);\n      return true;\n    }\n  }\n  return false;\n}\n\nexport class Logger {\n  logKernelProfile(\n      name: string, result: Tensor, vals: TypedArray,\n      timeMs: number|{error: string}, inputs: NamedTensorMap,\n      extraInfo?: string) {\n    const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n                                              timeMs['error'];\n    const paddedName = util.rightPad(name, 25);\n    const rank = result.rank;\n    const size = result.size;\n    const shape = util.rightPad(result.shape.toString(), 14);\n    let inputShapesDescription = '';\n\n    for (const name in inputs) {\n      const input = inputs[name];\n      if (input != null) {\n        // The input might be a non-tensor (e.g HTMLImageElement), in which case\n        // we claim the output shape as input shape.\n        const inputShape = input.shape || result.shape;\n        const inputRank = inputShape.length;\n        inputShapesDescription +=\n            `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n      }\n    }\n\n    console.log(\n        `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n            inputShapesDescription}\\t%c${extraInfo}`,\n        'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n        'color: green', 'color: steelblue');\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n  id: number;\n  kernelName: string;\n  outputs: Tensor[];\n  inputs: NamedTensorMap;\n  // Optional params, defined only for ops with gradient impl.\n  gradient?: (dys: Tensor[]) => NamedGradientMap;\n  saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n  [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n    tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n  // Forward pass to compute all the nodes and Tensors that are transitively a\n  // function of x.\n  const tensorsFromX: {[tensorId: number]: boolean} = {};\n  const nodesFromX: {[nodeId: number]: boolean} = {};\n  for (let i = 0; i < xs.length; i++) {\n    tensorsFromX[xs[i].id] = true;\n  }\n\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n    for (const inputName in nodeInputs) {\n      const input = nodeInputs[inputName];\n\n      let anyInputFromX = false;\n      for (let j = 0; j < xs.length; j++) {\n        if (tensorsFromX[input.id]) {\n          node.outputs.forEach(output => tensorsFromX[output.id] = true);\n          anyInputFromX = true;\n          nodesFromX[node.id] = true;\n          break;\n        }\n      }\n\n      if (anyInputFromX) {\n        break;\n      }\n    }\n  }\n\n  // Backward pass to find all of the nodes and Tensors that lead to y.\n  const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n  tensorsLeadToY[y.id] = true;\n  const nodesToY: {[nodeId: number]: boolean} = {};\n\n  for (let i = tape.length - 1; i >= 0; i--) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n\n    // If any of the outputs lead to y, mark all of the inputs as leading to y.\n    for (let j = 0; j < node.outputs.length; j++) {\n      if (tensorsLeadToY[node.outputs[j].id]) {\n        for (const inputName in nodeInputs) {\n          tensorsLeadToY[nodeInputs[inputName].id] = true;\n          nodesToY[node.id] = true;\n        }\n        break;\n      }\n    }\n  }\n\n  // Return the paths that come from x and lead to y.\n  const filteredTape: TapeNode[] = [];\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n\n    if (nodesFromX[node.id] && nodesToY[node.id]) {\n      // Prune the inputs from the node that aren't a function of x.\n      const prunedInputs: {[inputName: string]: Tensor} = {};\n      for (const inputName in node.inputs) {\n        const nodeInput = node.inputs[inputName];\n        if (tensorsFromX[nodeInput.id]) {\n          prunedInputs[inputName] = nodeInput;\n        }\n      }\n\n      // Copy the node and overwrite inputsAndArgs to the pruned version.\n      const prunedNode = Object.assign({}, node);\n      prunedNode.inputs = prunedInputs;\n      prunedNode.outputs = node.outputs;\n\n      filteredTape.push(prunedNode);\n    }\n  }\n\n  return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n    tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n    filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n    add: (a: Tensor, b: Tensor) => Tensor) {\n  // Walk the tape backward and keep a map of Tensor to its gradient.\n  for (let i = filteredTape.length - 1; i >= 0; i--) {\n    const node = filteredTape[i];\n\n    const dys: Tensor[] = [];\n    node.outputs.forEach(o => {\n      const gradTensor = tensorAccumulatedGradientMap[o.id];\n      if (gradTensor != null) {\n        dys.push(gradTensor);\n      } else {\n        // This particular output is not in the back-propagation subgraph, so it\n        // does not affect the final output, thus we put null for its dy.\n        dys.push(null);\n      }\n    });\n\n    if (node.gradient == null) {\n      throw new Error(\n          `Cannot compute gradient: gradient function not found ` +\n          `for ${node.kernelName}.`);\n    }\n\n    // Backprop dy through this node and accumulate gradients over the inputs.\n    const inputGradients = node.gradient(dys);\n\n    for (const inputName in node.inputs) {\n      if (!(inputName in inputGradients)) {\n        throw new Error(\n            `Cannot backprop through input ${inputName}. ` +\n            `Available gradients found: ${Object.keys(inputGradients)}.`);\n      }\n\n      // Call the gradient function.\n      const dx = tidy(() => inputGradients[inputName]());\n      if (dx.dtype !== 'float32') {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n      }\n      const x = node.inputs[inputName];\n      if (!util.arraysEqual(dx.shape, x.shape)) {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `'${inputName}' has shape '${dx.shape}', which does not match ` +\n            `the shape of the input '${x.shape}'`);\n      }\n\n      if (tensorAccumulatedGradientMap[x.id] == null) {\n        tensorAccumulatedGradientMap[x.id] = dx;\n      } else {\n        const curGradient = tensorAccumulatedGradientMap[x.id];\n        tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n        curGradient.dispose();\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    verbose: boolean) {\n  const strides = computeStrides(shape);\n  const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n  const rank = shape.length;\n  const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n  const lines = ['Tensor'];\n  if (verbose) {\n    lines.push(`  dtype: ${dtype}`);\n    lines.push(`  rank: ${rank}`);\n    lines.push(`  shape: [${shape}]`);\n    lines.push(`  values:`);\n  }\n  lines.push(valsLines.map(l => '    ' + l).join('\\n'));\n  return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[]): number[] {\n  const n = sizeFromShape(shape);\n  const numCols = strides[strides.length - 1];\n  const padPerCol = new Array(numCols).fill(0);\n  const rank = shape.length;\n  const valuesOrTuples =\n      dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n  if (rank > 1) {\n    for (let row = 0; row < n / numCols; row++) {\n      const offset = row * numCols;\n      for (let j = 0; j < numCols; j++) {\n        padPerCol[j] = Math.max(\n            padPerCol[j],\n            valToString(valuesOrTuples[offset + j], 0, dtype).length);\n      }\n    }\n  }\n  return padPerCol;\n}\n\nfunction valToString(\n    val: number|string|[number, number], pad: number, dtype: DataType) {\n  let valStr: string;\n  if (Array.isArray(val)) {\n    valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n        `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n  } else if (isString(val)) {\n    valStr = `'${val}'`;\n  } else if (dtype === 'bool') {\n    valStr = boolNumToString(val);\n  } else {\n    valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n  }\n\n  return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n  return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[], padPerCol: number[], isLast = true): string[] {\n  const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n  const size = shape[0];\n  const rank = shape.length;\n  if (rank === 0) {\n    if (dtype === 'complex64') {\n      const complexTuple = createComplexTuples(vals);\n      return [valToString(complexTuple[0], 0, dtype)];\n    }\n    if (dtype === 'bool') {\n      return [boolNumToString(vals[0] as number)];\n    }\n    return [vals[0].toString()];\n  }\n\n  if (rank === 1) {\n    if (size > FORMAT_LIMIT_NUM_VALS) {\n      const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n      let firstVals = Array.from<number|string|[number, number]>(\n          vals.slice(0, firstValsSize));\n      let lastVals = Array.from<number|string|[number, number]>(vals.slice(\n          (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n          size * storagePerElement));\n      if (dtype === 'complex64') {\n        firstVals = createComplexTuples(firstVals);\n        lastVals = createComplexTuples(lastVals);\n      }\n      return [\n        '[' +\n        firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n            .join(', ') +\n        ', ..., ' +\n        lastVals\n            .map(\n                (x, i) => valToString(\n                    x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n            .join(', ') +\n        ']'\n      ];\n    }\n    const displayVals: Array<number|string|[number, number]> =\n        dtype === 'complex64' ? createComplexTuples(vals) :\n                                Array.from<number|string>(vals);\n\n    return [\n      '[' +\n      displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n          .join(', ') +\n      ']'\n    ];\n  }\n\n  // The array is rank 2 or more.\n  const subshape = shape.slice(1);\n  const substrides = strides.slice(1);\n  const stride = strides[0] * storagePerElement;\n  const lines: string[] = [];\n  if (size > FORMAT_LIMIT_NUM_VALS) {\n    for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          false /* isLast */));\n    }\n    lines.push('...');\n    for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  } else {\n    for (let i = 0; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  }\n  const sep = rank === 2 ? ',' : '';\n  lines[0] = '[' + lines[0] + sep;\n  for (let i = 1; i < lines.length - 1; i++) {\n    lines[i] = ' ' + lines[i] + sep;\n  }\n  let newLineSep = ',\\n';\n  for (let i = 2; i < rank; i++) {\n    newLineSep += '\\n';\n  }\n  lines[lines.length - 1] =\n      ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n  return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n                             TypedArray): Array<[number, number]> {\n  const complexTuples: Array<[number, number]> = [];\n  for (let i = 0; i < vals.length; i += 2) {\n    complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n  }\n  return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData<D extends DataType> {\n  dataId?: DataId;\n  values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer<R extends Rank, D extends DataType = 'float32'> {\n  size: number;\n  shape: ShapeMap[R];\n  strides: number[];\n  values: DataTypeMap[D];\n\n  constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.size = util.sizeFromShape(shape);\n\n    if (values != null) {\n      const n = values.length;\n      util.assert(\n          n === this.size,\n          () => `Length of values '${n}' does not match the size ` +\n              `inferred by the shape '${this.size}'.`);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `complex64 dtype TensorBuffers are not supported. Please create ` +\n          `a TensorBuffer for the real and imaginary parts separately and ` +\n          `call tf.complex(real, imag).`);\n    }\n    this.values = values || util.getArrayFromDType(dtype, this.size);\n    this.strides = computeStrides(shape);\n  }\n\n  /**\n   * Sets a value in the buffer at a given location.\n   *\n   * @param value The value to set.\n   * @param locs  The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  set(value: SingleValueMap[D], ...locs: number[]): void {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    util.assert(\n        locs.length === this.rank,\n        () => `The number of provided coordinates (${locs.length}) must ` +\n            `match the rank (${this.rank})`);\n\n    const index = this.locToIndex(locs);\n    this.values[index] = value as number;\n  }\n\n  /**\n   * Returns the value in the buffer at the provided location.\n   *\n   * @param locs The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  get(...locs: number[]): SingleValueMap[D] {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    let i = 0;\n    for (const loc of locs) {\n      if (loc < 0 || loc >= this.shape[i]) {\n        const msg = `Requested out of range element at ${locs}. ` +\n            `  Buffer shape=${this.shape}`;\n        throw new Error(msg);\n      }\n      i++;\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return this.values[index] as SingleValueMap[D];\n  }\n\n  locToIndex(locs: number[]): number {\n    if (this.rank === 0) {\n      return 0;\n    } else if (this.rank === 1) {\n      return locs[0];\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return index;\n  }\n\n  indexToLoc(index: number): number[] {\n    if (this.rank === 0) {\n      return [];\n    } else if (this.rank === 1) {\n      return [index];\n    }\n    const locs: number[] = new Array(this.shape.length);\n    for (let i = 0; i < locs.length - 1; ++i) {\n      locs[i] = Math.floor(index / this.strides[i]);\n      index -= locs[i] * this.strides[i];\n    }\n    locs[locs.length - 1] = index;\n    return locs;\n  }\n\n  get rank() {\n    return this.shape.length;\n  }\n\n  /**\n   * Creates an immutable `tf.Tensor` object from the buffer.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  toTensor(): Tensor<R> {\n    return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n        Tensor<R>;\n  }\n}\n\nexport interface TensorTracker {\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: Backend): Tensor;\n  makeVariable(\n      initialValue: Tensor, trainable?: boolean, name?: string,\n      dtype?: DataType): Variable;\n  incRef(a: Tensor, backend: Backend): void;\n  disposeTensor(t: Tensor): void;\n  disposeVariable(v: Variable): void;\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n  cast<T extends Tensor>(x: T, dtype: DataType): T;\n  buffer<R extends Rank, D extends DataType>(\n      shape: ShapeMap[R], dtype: D,\n      values?: DataTypeMap[D]): TensorBuffer<R, D>;\n  print<T extends Tensor>(x: T, verbose: boolean): void;\n  clone<T extends Tensor>(x: T): T;\n  // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n  trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n  opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n  deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object;  // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor<R extends Rank = Rank> {\n  /** Unique id of this tensor. */\n  readonly id: number;\n  /**\n   * Id of the bucket holding the data for this tensor. Multiple arrays can\n   * point to the same bucket (e.g. when calling array.reshape()).\n   */\n  dataId: DataId;\n  /** The shape of the tensor. */\n  readonly shape: ShapeMap[R];\n  /** Number of elements in the tensor. */\n  readonly size: number;\n  /** The data type for the array. */\n  readonly dtype: DataType;\n  /** The rank type for the array (see `Rank` enum). */\n  readonly rankType: R;\n\n  /** Whether this tensor has been globally kept. */\n  kept = false;\n  /** The id of the scope this tensor is being tracked in. */\n  scopeId: number;\n\n  /**\n   * Number of elements to skip in each dimension when indexing. See\n   * https://docs.scipy.org/doc/numpy/reference/generated/\\\n   * numpy.ndarray.strides.html\n   */\n  readonly strides: number[];\n\n  constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.dtype = dtype || 'float32';\n    this.size = util.sizeFromShape(shape);\n    this.strides = computeStrides(shape);\n    this.dataId = dataId;\n    this.id = id;\n    this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n  }\n\n  get rank(): number {\n    return this.shape.length;\n  }\n\n  /**\n   * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async buffer<D extends DataType = 'float32'>(): Promise<TensorBuffer<R, D>> {\n    const vals = await this.data<D>();\n    return opHandler.buffer(this.shape, this.dtype as D, vals);\n  }\n\n  /**\n   * Returns a `tf.TensorBuffer` that holds the underlying data.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  bufferSync<D extends DataType = 'float32'>(): TensorBuffer<R, D> {\n    return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * asynchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async array(): Promise<ArrayMap[R]> {\n    const vals = await this.data();\n    return toNestedArray(this.shape, vals) as ArrayMap[R];\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * synchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  arraySync(): ArrayMap[R] {\n    return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n  }\n\n  /**\n   * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n   * promise of `TypedArray` that resolves when the computation has finished.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]> {\n    this.throwIfDisposed();\n    const data = trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      const bytes = await data as Uint8Array[];\n      try {\n        return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as Promise<DataTypeMap[D]>;\n  }\n\n  /**\n   * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n   * UI thread until the values are ready, which can cause performance issues.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D] {\n    this.throwIfDisposed();\n    const data = trackerFn().readSync(this.dataId);\n    if (this.dtype === 'string') {\n      try {\n        return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n            DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as DataTypeMap[D];\n  }\n\n  /** Returns the underlying bytes of the tensor's data. */\n  async bytes(): Promise<Uint8Array[]|Uint8Array> {\n    this.throwIfDisposed();\n    const data = await trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      return data as Uint8Array[];\n    } else {\n      return new Uint8Array((data as TypedArray).buffer);\n    }\n  }\n\n  /**\n   * Disposes `tf.Tensor` from memory.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dispose(): void {\n    if (this.isDisposed) {\n      return;\n    }\n    trackerFn().disposeTensor(this);\n    this.isDisposedInternal = true;\n  }\n\n  protected isDisposedInternal = false;\n  get isDisposed(): boolean {\n    return this.isDisposedInternal;\n  }\n\n  throwIfDisposed() {\n    if (this.isDisposed) {\n      throw new Error(`Tensor is disposed.`);\n    }\n  }\n\n  /**\n   * Prints the `tf.Tensor`. See `tf.print` for details.\n   *\n   * @param verbose Whether to print verbose information about the tensor,\n   *    including dtype and size.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  print(verbose = false): void {\n    return opHandler.print(this, verbose);\n  }\n\n  /**\n   * Returns a copy of the tensor. See `tf.clone` for details.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  clone<T extends Tensor>(this: T): T {\n    this.throwIfDisposed();\n    return opHandler.clone(this);\n  }\n\n  /**\n   * Returns a human-readable description of the tensor. Useful for logging.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  toString(verbose = false): string {\n    const vals = this.dataSync();\n    return tensorToString(vals, this.shape, this.dtype, verbose);\n  }\n\n  cast<T extends this>(dtype: DataType): T {\n    this.throwIfDisposed();\n    return opHandler.cast(this as T, dtype);\n  }\n  variable(trainable = true, name?: string, dtype?: DataType): Variable<R> {\n    this.throwIfDisposed();\n    return trackerFn().makeVariable(this, trainable, name, dtype) as\n        Variable<R>;\n  }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n  value: (instance: Tensor) => {\n    // Implementation note: we should use properties of the object that will be\n    // defined before the constructor body has finished executing (methods).\n    // This is because when this code is transpiled by babel, babel will call\n    // classCallCheck before the constructor body is run.\n    // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n    return !!instance && instance.data != null && instance.dataSync != null &&\n        instance.throwIfDisposed != null;\n  }\n});\n\nexport interface NumericTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: NumericDataType;\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D];\n  data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]>;\n}\n\nexport interface StringTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: 'string';\n  dataSync<D extends DataType = 'string'>(): DataTypeMap[D];\n  data<D extends DataType = 'string'>(): Promise<DataTypeMap[D]>;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor<Rank.R0>;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor<Rank.R1>;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor<Rank.R2>;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor<Rank.R3>;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor<Rank.R4>;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor<Rank.R5>;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor<Rank.R6>;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable<R extends Rank = Rank> extends Tensor<R> {\n  name: string;\n\n  constructor(\n      initialValue: Tensor<R>, public trainable: boolean, name: string,\n      tensorId: number) {\n    super(\n        initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n    this.name = name;\n  }\n\n  /**\n   * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n   * the same shape and dtype as the old `tf.Tensor`.\n   *\n   * @param newValue New tensor to be assigned to this variable.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  assign(newValue: Tensor<R>): void {\n    if (newValue.dtype !== this.dtype) {\n      throw new Error(\n          `dtype of the new value (${newValue.dtype}) and ` +\n          `previous value (${this.dtype}) must match`);\n    }\n    if (!util.arraysEqual(newValue.shape, this.shape)) {\n      throw new Error(\n          `shape of the new value (${newValue.shape}) and ` +\n          `previous value (${this.shape}) must match`);\n    }\n    trackerFn().disposeTensor(this);\n    this.dataId = newValue.dataId;\n    trackerFn().incRef(this, null /* backend */);\n  }\n\n  dispose(): void {\n    trackerFn().disposeVariable(this);\n    this.isDisposedInternal = true;\n  }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n  value: (instance: Variable) => {\n    return instance instanceof Tensor && instance.assign != null &&\n        instance.assign instanceof Function;\n  }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n  R0: number[];\n  R1: [number];\n  R2: [number, number];\n  R3: [number, number, number];\n  R4: [number, number, number, number];\n  R5: [number, number, number, number, number];\n  R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n  R0: number;\n  R1: number[];\n  R2: number[][];\n  R3: number[][][];\n  R4: number[][][][];\n  R5: number[][][][][];\n  R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n  float32: Float32Array;\n  int32: Int32Array;\n  bool: Uint8Array;\n  complex64: Float32Array;\n  string: string[];\n}\n\nexport interface SingleValueMap {\n  bool: boolean;\n  int32: number;\n  float32: number;\n  complex64: number;\n  string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n  R0 = 'R0',\n  R1 = 'R1',\n  R2 = 'R2',\n  R3 = 'R3',\n  R4 = 'R4',\n  R5 = 'R5',\n  R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray<T> =\n    T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray<T extends any> {\n  [index: number]: T|RecursiveArray<T>;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'int32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'bool',\n  'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n  'float32' = 'float32',\n  'int32' = 'float32',\n  'bool' = 'float32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n  'float32' = 'complex64',\n  'int32' = 'complex64',\n  'bool' = 'complex64',\n  'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n  'float32': UpcastFloat32AndMap,\n  'int32': UpcastInt32AndMap,\n  'bool': UpcastBoolAndMap,\n  'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n  if (typeA === 'string' || typeB === 'string') {\n    if (typeA === 'string' && typeB === 'string') {\n      return 'string';\n    }\n    throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n  }\n  return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n  return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n    TypedArray|number|boolean|string|RecursiveArray<number|number[]|TypedArray>|\n    RecursiveArray<boolean>|RecursiveArray<string>|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n    string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n    boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n    boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n    TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n    string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n    TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n    string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n  width: number;\n  height: number;\n  data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch<T extends Tensor>(a: T, b: T): [T, T] {\n  if (a.dtype === b.dtype) {\n    return [a, b];\n  }\n  const dtype = upcastType(a.dtype, b.dtype);\n  return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n  assert(\n      a.dtype === b.dtype,\n      () => `The dtypes of the first(${a.dtype}) and` +\n          ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n  return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n *   `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n *   is safe to pass any object here, except that `Promise`s are not\n *   supported.\n * @returns An array of `Tensors` found within the passed object. If the\n *   argument is simply a `Tensor', a list containing that `Tensor` is\n *   returned. If the object is not a `Tensor` or does not\n *   contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n  const list: Tensor[] = [];\n  const seen = new Set<{}|void>();\n  walkTensorContainer(result, list, seen);\n  return list;\n}\n\nfunction walkTensorContainer(\n    container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n  if (container == null) {\n    return;\n  }\n  if (container instanceof Tensor) {\n    list.push(container);\n    return;\n  }\n  if (!isIterable(container)) {\n    return;\n  }\n  // Iteration over keys works also for arrays.\n  const iterable = container as TensorContainerArray;\n  for (const k in iterable) {\n    const val = iterable[k];\n    if (!seen.has(val)) {\n      seen.add(val);\n      walkTensorContainer(val, list, seen);\n    }\n  }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n  return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc<T> = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n *   value: Tensor,\n *   gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc<T extends Tensor> =\n    (...inputs: Array<Tensor|GradSaveFunc>) => {\n      value: T;\n      gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n    };\n\nexport type MemoryInfo = {\n  numTensors: number; numDataBuffers: number; numBytes: number;\n  unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n  name: string; bytesAdded: number; totalBytesSnapshot: number;\n  tensorsAdded: number;\n  totalTensorsSnapshot: number;\n  inputShapes: number[][];\n  outputShapes: number[][];\n  kernelTimeMs: number | {error: string} | Promise<number|{error: string}>;\n  extraInfo: string | Promise<string>;\n};\n\nexport type ProfileInfo = {\n  newBytes: number; newTensors: number; peakBytes: number;\n  kernels: KernelInfo[];\n  result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n  wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn<T extends TensorContainer> = () => T;\n\ninterface ScopeState {\n  track: Tensor[];\n  name: string;\n  id: number;\n}\n\nclass EngineState {\n  // Public since optimizers will use it.\n  registeredVariables: NamedVariableMap = {};\n\n  nextTapeNodeId = 0;\n  numBytes = 0;\n  numTensors = 0;\n  numStringTensors = 0;\n  numDataBuffers = 0;\n\n  activeTape: TapeNode[];\n  // Number of nested tf.grad() statements when computing higher-order\n  // gradients. E.g. `1` for first-order gradients and `2` for second-order\n  // gradients. Used to track if the tape should be removed after a backprop.\n  gradientDepth = 0;\n  // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n  // off the tape.\n  kernelDepth = 0;\n\n  // Keep Tensors that parallel the tapes.\n  activeScope: ScopeState;\n  scopeStack: ScopeState[] = [];\n  /**\n   * Keeps track of the number of data moves during a kernel execution. We\n   * maintain a stack since kernels can call other kernels, recursively.\n   */\n  numDataMovesStack: number[] = [];\n  nextScopeId = 0;\n\n  tensorInfo = new WeakMap<DataId, {\n    backend: KernelBackend,\n    bytes: number,\n    dtype: DataType,\n    shape: number[],\n    refCount: number\n  }>();\n\n  profiling = false;\n  activeProfile: ProfileInfo =\n      {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n  dispose() {\n    for (const variableName in this.registeredVariables) {\n      this.registeredVariables[variableName].dispose();\n    }\n  }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n  state: EngineState;\n  backendName: string;\n  registry: {[id: string]: KernelBackend} = {};\n  registryFactory: {\n    [id: string]: {\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority: number\n    }\n  } = {};\n\n  private profiler: Profiler;\n  private backendInstance: KernelBackend;\n  private pendingBackendInit: Promise<boolean>;\n  private pendingBackendInitId = 0;\n\n  constructor(public ENV: Environment) {\n    this.state = new EngineState();\n  }\n\n  async ready(): Promise<void> {\n    if (this.pendingBackendInit != null) {\n      return this.pendingBackendInit.then(() => {});\n    }\n    if (this.backendInstance != null) {\n      return;\n    }\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const success = await this.initializeBackend(backendName).success;\n      if (success) {\n        await this.setBackend(backendName);\n        return;\n      }\n    }\n\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  get backend(): KernelBackend {\n    if (this.pendingBackendInit != null) {\n      throw new Error(\n          `Backend '${this.backendName}' has not yet been initialized. Make ` +\n          `sure to await tf.ready() or await tf.setBackend() before calling ` +\n          `other methods`);\n    }\n    if (this.backendInstance == null) {\n      const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n      if (asyncInit) {\n        throw new Error(\n            `The highest priority backend '${name}' has not yet been ` +\n            `initialized. Make sure to await tf.ready() or ` +\n            `await tf.setBackend() before calling other methods`);\n      }\n      this.setBackend(name);\n    }\n    return this.backendInstance;\n  }\n\n  backendNames(): string[] {\n    return Object.keys(this.registryFactory);\n  }\n\n  findBackend(backendName: string): KernelBackend {\n    if (!(backendName in this.registry)) {\n      // If the backend hasn't been initialized but we have a registry entry for\n      // it, initialize it and return it.\n      if (backendName in this.registryFactory) {\n        const {asyncInit} = this.initializeBackend(backendName);\n        if (asyncInit) {\n          // Backend is not ready yet.\n          return null;\n        }\n      } else {\n        return null;\n      }\n    }\n    return this.registry[backendName];\n  }\n\n  findBackendFactory(backendName: string):\n      () => KernelBackend | Promise<KernelBackend> {\n    if (!(backendName in this.registryFactory)) {\n      return null;\n    }\n    return this.registryFactory[backendName].factory;\n  }\n\n  registerBackend(\n      backendName: string,\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority = 1): boolean {\n    if (backendName in this.registryFactory) {\n      console.warn(\n          `${backendName} backend was already registered. ` +\n          `Reusing existing backend factory.`);\n      return false;\n    }\n    this.registryFactory[backendName] = {factory, priority};\n    return true;\n  }\n\n  async setBackend(backendName: string): Promise<boolean> {\n    if (this.registryFactory[backendName] == null) {\n      throw new Error(`Backend name '${backendName}' not found in registry`);\n    }\n    this.backendName = backendName;\n    if (this.registry[backendName] == null) {\n      this.backendInstance = null;\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      const result = asyncInit ? await success : success;\n      if (!result) {\n        return false;\n      }\n    }\n    this.backendInstance = this.registry[backendName];\n    this.setupRegisteredKernels();\n    // Reset the profiler.\n    this.profiler = new Profiler(this.backendInstance);\n\n    return true;\n  }\n\n  private setupRegisteredKernels(): void {\n    const kernels = getKernelsForBackend(this.backendName);\n    kernels.forEach(kernel => {\n      if (kernel.setupFunc != null) {\n        kernel.setupFunc(this.backendInstance);\n      }\n    });\n  }\n\n  private disposeRegisteredKernels(backendName: string): void {\n    const kernels = getKernelsForBackend(backendName);\n    kernels.forEach(kernel => {\n      if (kernel.disposeFunc != null) {\n        kernel.disposeFunc(this.registry[backendName]);\n      }\n    });\n  }\n\n  /**\n   * Initializes a backend by looking up the backend name in the factory\n   * registry and calling the factory method. Returns a boolean representing\n   * whether the initialization of the backend suceeded. Throws an error if\n   * there is no backend in the factory registry.\n   */\n  private initializeBackend(backendName: string):\n      {success: boolean|Promise<boolean>, asyncInit: boolean} {\n    const registryFactoryEntry = this.registryFactory[backendName];\n    if (registryFactoryEntry == null) {\n      throw new Error(\n          `Cannot initialize backend ${backendName}, no registration found.`);\n    }\n\n    try {\n      const backend = registryFactoryEntry.factory();\n      /* Test if the factory returns a promise.\n      Done in a more liberal way than\n      previous 'Promise.resolve(backend)===backend'\n      as we needed to account for custom Promise\n      implementations (e.g. Angular) */\n      if (backend && !(backend instanceof KernelBackend)\n          && typeof backend.then === 'function') {\n        const promiseId = ++this.pendingBackendInitId;\n        const success =\n            backend\n                .then(backendInstance => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.registry[backendName] = backendInstance;\n                  this.pendingBackendInit = null;\n                  return true;\n                })\n                .catch(err => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.pendingBackendInit = null;\n                  console.warn(\n                      `Initialization of backend ${backendName} failed`);\n                  console.warn(err.stack || err.message);\n                  return false;\n                });\n        this.pendingBackendInit = success;\n        return {success, asyncInit: true};\n      } else {\n        this.registry[backendName] = backend as KernelBackend;\n        return {success: true, asyncInit: false};\n      }\n    } catch (err) {\n      console.warn(`Initialization of backend ${backendName} failed`);\n      console.warn(err.stack || err.message);\n      return {success: false, asyncInit: false};\n    }\n  }\n\n  removeBackend(backendName: string): void {\n    if (!(backendName in this.registryFactory)) {\n      throw new Error(`${backendName} backend not found in registry`);\n    }\n    if (this.backendName === backendName && this.pendingBackendInit != null) {\n      // There is a pending promise of the backend we want to remove. Make it\n      // obsolete.\n      this.pendingBackendInitId++;\n    }\n\n    if (backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n\n    delete this.registryFactory[backendName];\n\n    // Unset the backend if it is active.\n    if (this.backendName === backendName) {\n      this.pendingBackendInit = null;\n      this.backendName = null;\n      this.backendInstance = null;\n    }\n  }\n\n  private getSortedBackends(): string[] {\n    if (Object.keys(this.registryFactory).length === 0) {\n      throw new Error('No backend found in registry.');\n    }\n    return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n      // Highest priority comes first.\n      return this.registryFactory[b].priority -\n          this.registryFactory[a].priority;\n    });\n  }\n\n  private initializeBackendsAndReturnBest():\n      {name: string, asyncInit: boolean} {\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      if (asyncInit || success) {\n        return {name: backendName, asyncInit};\n      }\n    }\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  moveData(backend: KernelBackend, dataId: DataId) {\n    const info = this.state.tensorInfo.get(dataId);\n    const srcBackend = info.backend;\n    const values = this.readSync(dataId);\n    // Delete the tensor from the old backend and move it to the new\n    // backend.\n    srcBackend.disposeData(dataId);\n    info.backend = backend;\n    backend.move(dataId, values, info.shape, info.dtype);\n    if (this.shouldCheckForMemLeaks()) {\n      // Track the number of moves during a kernel execution to correctly\n      // detect memory leaks.\n      this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n    }\n  }\n\n  tidy<T extends TensorContainer>(nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>):\n      T {\n    let name: string = null;\n    if (fn == null) {\n      // Called with only 1 argument.\n      if (typeof nameOrFn !== 'function') {\n        throw new Error('Please provide a function to tidy()');\n      }\n      fn = nameOrFn;\n    } else {\n      // Called with 2 arguments.\n      if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n        throw new Error(\n            'When calling with two arguments, the first argument ' +\n            'to tidy() must be a string');\n      }\n      if (typeof fn !== 'function') {\n        throw new Error(\n            'When calling with two arguments, the 2nd argument ' +\n            'to tidy() must be a function');\n      }\n      name = nameOrFn as string;\n      // TODO(nsthorat,smilkov): Do operation logging and performance\n      // profiling.\n    }\n    let result: T;\n    return this.scopedRun(\n        () => this.startScope(name), () => this.endScope(result), () => {\n          result = fn();\n          if (result instanceof Promise) {\n            console.error('Cannot return a Promise inside of tidy.');\n          }\n          return result;\n        });\n  }\n\n  private scopedRun<T>(start: () => void, end: () => void, f: () => T): T {\n    start();\n    try {\n      const res = f();\n      end();\n      return res;\n    } catch (ex) {\n      end();\n      throw ex;\n    }\n  }\n\n  private static nextTensorId = 0;\n  private nextTensorId(): number {\n    return Engine.nextTensorId++;\n  }\n\n  private static nextVariableId = 0;\n  private nextVariableId(): number {\n    return Engine.nextVariableId++;\n  }\n\n  /**\n   * This method is called instead of the public-facing tensor.clone() when\n   * saving a tensor for backwards pass. It makes sure to add the clone\n   * operation to the tape regardless of being called inside a kernel\n   * execution.\n   *\n   * This method will go away once all kernels are modularized since we won't\n   * need to turn off the tape inside runKernel().\n   */\n  private clone(x: Tensor): Tensor {\n    const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n    const inputs = {x};\n    const grad = (dy: Tensor) => ({\n      x: () => {\n        const dtype = 'float32';\n        const gradInputs = {x: dy};\n        const attrs = {dtype};\n\n        return ENGINE.runKernelFunc(\n            backend => backend.cast(dy, dtype),\n            gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n            attrs as {} as NamedAttrMap);\n      }\n    });\n    const saved: Tensor[] = [];\n    this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n    return y;\n  }\n\n  /**\n   * Execute a kernel with the given name and return the output tensor.\n   *\n   * @param kernelName The name of the kernel to execute.\n   * @param inputs A map of input names to tensors.\n   * @param attrs A map of attribute names to their values. An attribute is a\n   *     primitive (non-tensor) input to the kernel.\n   * @param inputsToSave A list of tensors, inputs to save for the backprop\n   *     computation.\n   * @param outputsToSave A list of booleans, specifying which output to save\n   *     for the backprop computation. These are booleans since the output\n   * tensors are not visible to the user.\n   */\n  runKernel(\n      kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n      inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n    const forwardFunc: null = null;\n    const backwardsFunc: null = null;\n    // Call runKernel as a stop-gap until we modularize all kernels.\n    // Once we modularize all kernels, we will remove the existing\n    // `runKernelFunc`.\n    return this.runKernelFunc(\n        forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n        outputsToSave);\n  }\n\n  private shouldCheckForMemLeaks(): boolean {\n    return this.ENV.getBool('IS_TEST');\n  }\n\n  private checkKernelForMemLeak(\n      kernelName: string, numDataIdsBefore: number,\n      outInfos: TensorInfo[]): void {\n    const numDataIdsAfter = this.backend.numDataIds();\n\n    // Count the number of data ids associated with the result of the kernel.\n    let numOutputDataIds = 0;\n    outInfos.forEach(info => {\n      // Complex numbers allocate 3 data ids, one for 'real', one for\n      // 'imaginary', and one for the container that holds the former two.\n      numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n    });\n\n    // Account for the number of moves during kernel execution. A \"data move\"\n    // can happen in the middle of a kernel execution, placing a new (key,value)\n    // pair in the data storage. Since data moves have net zero effect (we\n    // always remove the data from the old backend), we have to cancel them out\n    // when detecting memory leaks.\n    const numMoves =\n        this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n    const dataIdsLeaked =\n        numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n    if (dataIdsLeaked > 0) {\n      throw new Error(\n          `Backend '${this.backendName}' has an internal memory leak ` +\n          `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n    }\n  }\n\n  /**\n   * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n   *     only for kernels that are not yet fully modularized.\n   */\n  runKernelFunc<T extends Tensor|Tensor[], I extends NamedTensorMap>(\n      forwardFunc: ForwardFunc<T>, inputs: I,\n      backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n      kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n      outputsToSave?: boolean[]): T {\n    let outputs: Tensor[];\n    let saved: Tensor[] = [];\n    const isTapeOn = this.isTapeOn();\n    if (kernelName == null) {\n      kernelName =\n          this.state.activeScope != null ? this.state.activeScope.name : '';\n    }\n\n    const startingBytecount = this.state.numBytes;\n    const startingNumTensors = this.state.numTensors;\n\n    if (this.shouldCheckForMemLeaks()) {\n      this.state.numDataMovesStack.push(0);\n    }\n\n    let kernelFunc: () => Tensor[];\n    const kernel = getKernel(kernelName, this.backendName);\n    let out: TensorInfo|TensorInfo[];\n    if (kernel != null) {\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n        const outInfos = Array.isArray(out) ? out : [out];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n        }\n        const outTensors = outInfos.map(\n            ({dataId, shape, dtype}) =>\n                this.makeTensorFromDataId(dataId, shape, dtype));\n\n        // Save the inputs and outputs.\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (isTapeOn) {\n          let tensorsToSave =\n              this.getTensorsForGradient(kernelName, inputs, outTensors);\n          if (tensorsToSave == null) {\n            // Fallback for ops that call runKernelFunc and pass in\n            // inputsToSave and outputsToSave. Currently this is the set of ops\n            // with kernel support in the WASM backend. Once those ops and\n            // respective gradients are modularised we can remove this path.\n            if (outputsToSave == null) {\n              outputsToSave = [];\n            }\n            const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n            tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n          }\n          saved = this.saveTensorsForBackwardMode(tensorsToSave);\n        }\n        return outTensors;\n      };\n    } else {\n      const saveFunc: GradSaveFunc = (tensors) => {\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (!isTapeOn) {\n          return;\n        }\n        saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n      };\n\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n        const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n        }\n        return outs;\n      };\n    }\n\n    // Stop recording to a tape when running a kernel.\n    let kernelProfile: KernelProfile;\n    this.scopedRun(\n        () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n          if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n            outputs = kernelFunc();\n          } else {\n            kernelProfile = this.profiler.profileKernel(\n                kernelName, inputs, () => kernelFunc());\n            if (this.ENV.getBool('DEBUG')) {\n              this.profiler.logKernelProfile(kernelProfile);\n            }\n            outputs = kernelProfile.outputs;\n          }\n        });\n\n    if (isTapeOn) {\n      this.addTapeNode(\n          kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n    }\n\n    if (this.state.profiling) {\n      this.state.activeProfile.kernels.push({\n        name: kernelName,\n        bytesAdded: this.state.numBytes - startingBytecount,\n        totalBytesSnapshot: this.state.numBytes,\n        tensorsAdded: this.state.numTensors - startingNumTensors,\n        totalTensorsSnapshot: this.state.numTensors,\n        inputShapes: Object.keys(inputs).map(\n            key => inputs[key] != null ? inputs[key].shape : null),\n        outputShapes: outputs.map(item => item.shape),\n        kernelTimeMs: kernelProfile.timeMs,\n        extraInfo: kernelProfile.extraInfo\n      });\n    }\n    return (Array.isArray(out) ? outputs : outputs[0]) as T;\n  }\n\n  /**\n   * Saves tensors used in forward mode for use in backward mode.\n   *\n   * @param tensors the list of tensors to save.\n   */\n  private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n    const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n    return saved;\n  }\n\n  /**\n   * Returns a list of tensors to save for a given gradient calculation.\n   *\n   * Returns undefined if their is no registered gradient for this kernel in the\n   * gradient registry.\n   *\n   * @param kernelName name of kernel to look up gradient for.\n   * @param inputs a map of input tensors.\n   * @param outputs an array of output tensors from forward mode of kernel.\n   */\n  private getTensorsForGradient(\n      kernelName: string, inputs: NamedTensorMap,\n      outputs: Tensor[]): Tensor[]|null {\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      const inputsToSave: string[] = gradConfig.inputsToSave || [];\n      const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n      // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n      // specified in inputsToSave will be saved.\n      let inputTensorsToSave: Tensor[];\n      if (gradConfig.saveAllInputs) {\n        util.assert(\n            Array.isArray(inputs),\n            () => 'saveAllInputs is true, expected inputs to be an array.');\n\n        inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n      } else {\n        inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n      }\n\n      const outputTensorsToSave: Tensor[] =\n          outputs.filter((_, i) => outputsToSave[i]);\n\n      return inputTensorsToSave.concat(outputTensorsToSave);\n    }\n    // TODO(yassogba) throw exception here once all runkernelFunc calls with\n    // inputsToSave/outputsToSave are removed\n    return null;\n  }\n\n  /**\n   * Internal method used by public APIs for tensor creation. Makes a new\n   * tensor with the provided shape, dtype and values. It always\n   * creates a new data id and writes the values to the underlying backend.\n   */\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    if (values == null) {\n      throw new Error('Values passed to engine.makeTensor() are null');\n    }\n    dtype = dtype || 'float32';\n    backend = backend || this.backend;\n    let backendVals = values as BackendValues;\n    if (dtype === 'string' && util.isString(values[0])) {\n      backendVals = (values as string[]).map(d => util.encodeString(d));\n    }\n    const dataId = backend.write(backendVals, shape, dtype);\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n\n    // Count bytes for string tensors.\n    if (dtype === 'string') {\n      const info = this.state.tensorInfo.get(dataId);\n      const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n      this.state.numBytes += newBytes - info.bytes;\n      info.bytes = newBytes;\n    }\n    return t;\n  }\n\n  /**\n   * Internal method used by backends. Makes a new tensor\n   * that is a wrapper around an existing data id. It doesn't create\n   * a new data id, only increments the ref count used in memory tracking.\n   */\n  makeTensorFromDataId(\n      dataId: DataId, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    dtype = dtype || 'float32';\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n    return t;\n  }\n\n  makeVariable(\n      initialValue: Tensor, trainable = true, name?: string,\n      dtype?: DataType): Variable {\n    name = name || this.nextVariableId().toString();\n    if (dtype != null && dtype !== initialValue.dtype) {\n      initialValue = initialValue.cast(dtype);\n    }\n    const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n    if (this.state.registeredVariables[v.name] != null) {\n      throw new Error(`Variable with name ${v.name} was already registered`);\n    }\n    this.state.registeredVariables[v.name] = v;\n    this.incRef(v, this.backend);\n    return v;\n  }\n\n  incRef(a: Tensor, backend: KernelBackend): void {\n    const refCount = this.state.tensorInfo.has(a.dataId) ?\n        this.state.tensorInfo.get(a.dataId).refCount :\n        0;\n    this.state.numTensors++;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors++;\n    }\n    if (refCount === 0) {\n      this.state.numDataBuffers++;\n\n      // Bytes for complex numbers are counted by their components. Bytes for\n      // string tensors are counted when writing values.\n      let bytes = 0;\n      if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n        bytes = a.size * util.bytesPerElement(a.dtype);\n      }\n      this.state.tensorInfo.set(a.dataId, {\n        backend: backend || this.backend,\n        dtype: a.dtype,\n        shape: a.shape,\n        bytes,\n        refCount: 0\n      });\n      this.state.numBytes += bytes;\n    }\n\n    this.state.tensorInfo.get(a.dataId).refCount++;\n\n    if (!(a instanceof Variable)) {\n      this.track(a);\n    }\n  }\n\n  disposeTensor(a: Tensor): void {\n    if (!this.state.tensorInfo.has(a.dataId)) {\n      return;\n    }\n\n    this.state.numTensors--;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors--;\n    }\n    const info = this.state.tensorInfo.get(a.dataId);\n    const refCount = info.refCount;\n\n    if (refCount <= 1) {\n      // Don't count bytes for complex numbers as they are counted by their\n      // components.\n      if (a.dtype !== 'complex64') {\n        this.state.numBytes -= info.bytes;\n      }\n      this.state.numDataBuffers--;\n\n      info.backend.disposeData(a.dataId);\n      this.state.tensorInfo.delete(a.dataId);\n    } else {\n      this.state.tensorInfo.get(a.dataId).refCount--;\n    }\n    // TODO(nsthorat): Construct an error and save the stack trace for\n    // debugging when in debug mode. Creating a stack trace is too expensive\n    // to do unconditionally.\n  }\n\n  disposeVariables(): void {\n    for (const varName in this.state.registeredVariables) {\n      const v = this.state.registeredVariables[varName];\n      this.disposeVariable(v);\n    }\n  }\n\n  disposeVariable(v: Variable): void {\n    this.disposeTensor(v);\n    if (this.state.registeredVariables[v.name] != null) {\n      delete this.state.registeredVariables[v.name];\n    }\n  }\n\n  memory(): MemoryInfo {\n    const info = this.backend.memory() as MemoryInfo;\n    info.numTensors = this.state.numTensors;\n    info.numDataBuffers = this.state.numDataBuffers;\n    info.numBytes = this.state.numBytes;\n    if (this.state.numStringTensors > 0) {\n      info.unreliable = true;\n      if (info.reasons == null) {\n        info.reasons = [];\n      }\n      info.reasons.push(\n          'Memory usage by string tensors is approximate ' +\n          '(2 bytes per character)');\n    }\n    return info;\n  }\n\n  async profile(query: () => (TensorContainer | Promise<TensorContainer>)):\n      Promise<ProfileInfo> {\n    this.state.profiling = true;\n\n    const startBytes = this.state.numBytes;\n    const startNumTensors = this.state.numTensors;\n\n    this.state.activeProfile.kernels = [];\n    this.state.activeProfile.result = await query();\n\n    this.state.profiling = false;\n\n    this.state.activeProfile.peakBytes = Math.max(\n        ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n    this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n    this.state.activeProfile.newTensors =\n        this.state.numTensors - startNumTensors;\n    for (const kernel of this.state.activeProfile.kernels) {\n      kernel.kernelTimeMs = await kernel.kernelTimeMs;\n      kernel.extraInfo = await kernel.extraInfo;\n    }\n    return this.state.activeProfile;\n  }\n\n  isTapeOn(): boolean {\n    return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n  }\n\n  private addTapeNode(\n      kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n      gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n    const tapeNode: TapeNode =\n        {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      gradientsFunc = gradConfig.gradFunc;\n    }\n    if (gradientsFunc != null) {\n      tapeNode.gradient = (dys: Tensor[]) => {\n        // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n        // the backprop graph to the user as null instead of zeros\n        dys = dys.map((dy, i) => {\n          if (dy == null) {\n            const output = outputs[i];\n            const vals = util.makeZerosTypedArray(output.size, output.dtype);\n            return this.makeTensor(vals, output.shape, output.dtype);\n          }\n          return dy;\n        });\n        // Grad functions of ops with single outputs expect a dy, while ops\n        // with multiple outputs expect dys (array of dy).\n        return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n      };\n    }\n    this.state.activeTape.push(tapeNode);\n  }\n\n  keep<T extends Tensor>(result: T): T {\n    result.kept = true;\n    return result;\n  }\n\n  private startTape() {\n    if (this.state.gradientDepth === 0) {\n      this.state.activeTape = [];\n    }\n    this.state.gradientDepth++;\n  }\n\n  private endTape() {\n    this.state.gradientDepth--;\n  }\n\n  /**\n   * Start a scope. Use this with endScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  startScope(name?: string) {\n    const scopeInfo: ScopeState = {\n      track: [],\n      name: 'unnamed scope',\n      id: this.state.nextScopeId++\n    };\n    if (name) {\n      scopeInfo.name = name;\n    }\n    this.state.scopeStack.push(scopeInfo);\n    this.state.activeScope = scopeInfo;\n  }\n\n  /**\n   * End a scope. Use this with startScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  endScope(result?: TensorContainer) {\n    const tensorsToTrackInParent = getTensorsInContainer(result);\n    const tensorsToTrackInParentSet =\n        new Set(tensorsToTrackInParent.map(t => t.id));\n\n    // Dispose the arrays tracked in this scope.\n    for (let i = 0; i < this.state.activeScope.track.length; i++) {\n      const tensor = this.state.activeScope.track[i];\n      if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n        tensor.dispose();\n      }\n    }\n\n    const oldScope = this.state.scopeStack.pop();\n    this.state.activeScope = this.state.scopeStack.length === 0 ?\n        null :\n        this.state.scopeStack[this.state.scopeStack.length - 1];\n\n    // Track the current result in the parent scope.\n    tensorsToTrackInParent.forEach(tensor => {\n      // Only track the tensor if was allocated in the inner scope and is not\n      // globally kept.\n      if (!tensor.kept && tensor.scopeId === oldScope.id) {\n        this.track(tensor);\n      }\n    });\n  }\n\n  /**\n   * Returns gradients of `f` with respect to each of the `xs`. The gradients\n   * returned are of the same length as `xs`, but some might be null if `f`\n   * was not a function of that `x`. It also takes optional dy to multiply the\n   * gradient, which defaults to `1`.\n   */\n  gradients<T extends Tensor>(\n      f: () => T, xs: Tensor[], dy?: T,\n      allowNoGradients = false): {value: T, grads: Tensor[]} {\n    util.assert(\n        xs.length > 0, () => 'gradients() received an empty list of xs.');\n    if (dy != null && dy.dtype !== 'float32') {\n      throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n    }\n\n    const y = this.scopedRun(\n        () => this.startTape(), () => this.endTape(),\n        () => this.tidy('forward', f));\n\n    util.assert(\n        y instanceof Tensor,\n        () => 'The result y returned by f() must be a tensor.');\n    // Filter out the nodes that don't connect x => y.\n    const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n    if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n      throw new Error(\n          'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n          'that the f you passed encloses all operations that lead from x ' +\n          'to y.');\n    }\n\n    return this.tidy('backward', () => {\n      const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n      accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n      // Backprop gradients through the filtered nodes.\n      backpropagateGradients(\n          accumulatedGradientMap, filteredTape,\n          // Pass the tidy function to avoid circular dep with `tape.ts`.\n          f => this.tidy(f as ScopeFn<Tensor>),\n          // Pass an add function to avoide a circular dep with `tape.ts`.\n          add);\n      const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n      if (this.state.gradientDepth === 0) {\n        // This means that we are not computing higher-order gradients\n        // and can clean up the tape.\n        this.state.activeTape.forEach(node => {\n          for (const tensor of node.saved) {\n            tensor.dispose();\n          }\n        });\n        this.state.activeTape = null;\n      }\n      return {value: y, grads};\n    });\n  }\n\n  customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n      (...args: Array<Tensor|GradSaveFunc>) => T {\n    util.assert(\n        util.isFunction(f),\n        () => 'The f passed in customGrad(f) must be a function.');\n    return (...inputs: Tensor[]): T => {\n      util.assert(\n          inputs.every(t => t instanceof Tensor),\n          () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n              'tensors');\n\n      let res: {\n        value: T,\n        gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n      };\n      const inputMap: NamedTensorMap = {};\n      inputs.forEach((input, i) => {\n        inputMap[i] = input;\n      });\n      return this.runKernelFunc(\n          (_, save) => {\n            res = f(...[...inputs, save]);\n            util.assert(\n                res.value instanceof Tensor,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.value` is a tensor');\n            util.assert(\n                util.isFunction(res.gradFunc),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function.');\n            return res.value;\n          },\n          inputMap,\n          (dy: T, saved: Tensor[]) => {\n            const gradRes = res.gradFunc(dy, saved);\n            const grads: Tensor[] =\n                Array.isArray(gradRes) ? gradRes : [gradRes];\n            util.assert(\n                grads.length === inputs.length,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'the same number of tensors as inputs passed to f(...).');\n            util.assert(\n                grads.every(t => t instanceof Tensor),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'a list of only tensors.');\n            const gradMap: {[key: string]: () => Tensor} = {};\n            grads.forEach((grad, i) => {\n              gradMap[i] = () => grad;\n            });\n            return gradMap;\n          });\n    };\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.readSync(dataId);\n  }\n  read(dataId: DataId): Promise<BackendValues> {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.read(dataId);\n  }\n\n  async time(query: () => void): Promise<TimingInfo> {\n    const start = now();\n    const timingInfo = await this.backend.time(query) as TimingInfo;\n    timingInfo.wallMs = now() - start;\n    return timingInfo;\n  }\n\n  /**\n   * Tracks a Tensor in the current scope to be automatically cleaned up\n   * when the current scope ends, and returns the value.\n   *\n   * @param result The Tensor to track in the current scope.\n   */\n  private track<T extends Tensor>(result: T): T {\n    if (this.state.activeScope != null) {\n      result.scopeId = this.state.activeScope.id;\n      this.state.activeScope.track.push(result);\n    }\n\n    return result;\n  }\n\n  get registeredVariables(): NamedVariableMap {\n    return this.state.registeredVariables;\n  }\n\n  /**\n   * Resets the engine state. Removes all backends but does not remove\n   * registered backend factories.\n   */\n  reset(): void {\n    // Make any pending promise obsolete.\n    this.pendingBackendInitId++;\n\n    this.state.dispose();\n    this.ENV.reset();\n    this.state = new EngineState();\n\n    for (const backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n    this.backendName = null;\n    this.backendInstance = null;\n    this.pendingBackendInit = null;\n  }\n}\n\nfunction ones(shape: number[]): Tensor {\n  const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n  return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n  const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n  if (ns._tfengine == null) {\n    const environment = new Environment(ns);\n    ns._tfengine = new Engine(environment);\n  }\n  setEnvironmentGlobal(ns._tfengine.ENV);\n\n  // Tell the current tensor interface that the global engine is responsible\n  // for tracking.\n  setTensorTracker(() => ns._tfengine);\n  return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n  // We duplicate Add here to avoid a circular dependency with add.ts.\n  const inputs = {a, b};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.add(a, b);\n    save([a, b]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n  return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n  if (_isNavigatorDefined()) {\n    // tslint:disable-next-line:no-any\n    const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n    // tslint:disable-next-line:max-line-length\n    return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n               .test(a) ||\n        // tslint:disable-next-line:max-line-length\n        /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n            .test(a.substr(0, 4));\n  }\n  return false;\n}\n\nexport function isBrowser(): boolean {\n  return (typeof window !== 'undefined' && window.document != null) ||\n      //@ts-ignore\n      (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n  if (debugValue) {\n    console.warn(\n        'Debugging mode is ON. The output of every math call will ' +\n        'be downloaded to CPU and checked for NaNs. ' +\n        'This significantly impacts performance.');\n  }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n    'IS_NODE',\n    () => (typeof process !== 'undefined') &&\n        (typeof process.versions !== 'undefined') &&\n        (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n    'IS_CHROME',\n    () => typeof navigator !== 'undefined' && navigator != null &&\n        navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n        /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n    'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n  let firstElem: typeof val = val;\n\n  if (isTypedArray(val)) {\n    return dtype === 'string' ? [] : [val.length];\n  }\n  if (!Array.isArray(val)) {\n    return [];  // Scalar.\n  }\n  const shape: number[] = [];\n\n  while (Array.isArray(firstElem) ||\n         isTypedArray(firstElem) && dtype !== 'string') {\n    shape.push(firstElem.length);\n    firstElem = firstElem[0];\n  }\n  if (Array.isArray(val) &&\n      env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n    deepAssertShapeConsistency(val, shape, []);\n  }\n\n  return shape;\n}\n\nfunction deepAssertShapeConsistency(\n    val: TensorLike, shape: number[], indices: number[]) {\n  indices = indices || [];\n  if (!(Array.isArray(val)) && !isTypedArray(val)) {\n    assert(\n        shape.length === 0,\n        () => `Element arr[${indices.join('][')}] is a primitive, ` +\n            `but should be an array/TypedArray of ${shape[0]} elements`);\n    return;\n  }\n  assert(\n      shape.length > 0,\n      () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n          `but is an array of ${val.length} elements`);\n  assert(\n      val.length === shape[0],\n      () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n          `elements, but has ${val.length} elements`);\n  const subShape = shape.slice(1);\n  for (let i = 0; i < val.length; ++i) {\n    deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n  }\n}\n\nfunction assertDtype(\n    expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n    functionName: string) {\n  if (expectedDtype == null) {\n    return;\n  }\n  if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n      expectedDtype === 'numeric' && actualDType === 'string') {\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must ` +\n        `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n  }\n}\n\nexport function convertToTensor<T extends Tensor>(\n    x: T|TensorLike, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T {\n  if (x instanceof Tensor) {\n    assertDtype(parseAsDtype, x.dtype, argName, functionName);\n    return x;\n  }\n  let inferredDtype = inferDtype(x);\n  // If the user expects a bool/int/float, use that info to update the\n  // inferredDtype when it is not a string.\n  if (inferredDtype !== 'string' &&\n      ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n    inferredDtype = parseAsDtype as DataType;\n  }\n  assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n  if ((x == null) ||\n      (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n       typeof x !== 'boolean' && typeof x !== 'string')) {\n    const type = x == null ? 'null' : (x as {}).constructor.name;\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must be a ` +\n        `Tensor or TensorLike, but got '${type}'`);\n  }\n  const inferredShape = inferShape(x, inferredDtype);\n  if (!isTypedArray(x) && !Array.isArray(x)) {\n    x = [x] as number[];\n  }\n  const skipTypedArray = true;\n  const values = inferredDtype !== 'string' ?\n      toTypedArray(x, inferredDtype as DataType) :\n      flatten(x as string[], [], skipTypedArray) as string[];\n  return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray<T extends Tensor>(\n    arg: Array<T|TensorLike>, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n  if (!Array.isArray(arg)) {\n    throw new Error(\n        `Argument ${argName} passed to ${functionName} must be a ` +\n        '`Tensor[]` or `TensorLike[]`');\n  }\n  const tensors = arg as T[];\n  return tensors.map(\n      (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n      parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op<T extends Function>(f: {[name: string]: T}): T {\n  const keys = Object.keys(f);\n  if (keys.length !== 1) {\n    throw new Error(\n        `Please provide an object with a single key ` +\n        `(operation name) mapping to a function. Got an object with ` +\n        `${keys.length} keys.`);\n  }\n\n  let opName = keys[0];\n  const fn = f[opName];\n\n  // Strip the underscore from the end of the function name.\n  if (opName.endsWith('_')) {\n    opName = opName.substring(0, opName.length - 1);\n  }\n\n  // add an __op suffix to distinguish ops from kernels in tf.profile\n  opName = opName + OP_SCOPE_SUFFIX;\n\n  // tslint:disable-next-line:no-any\n  const f2 = (...args: any[]) => {\n    ENGINE.startScope(opName);\n    try {\n      const result = fn(...args);\n      if (result instanceof Promise) {\n        console.error('Cannot return a Promise inside of tidy.');\n      }\n      ENGINE.endScope(result);\n      return result;\n    } catch (ex) {\n      ENGINE.endScope(null);\n      throw ex;\n    }\n  };\n  Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n  // tslint:disable-next-line:no-any\n  return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_<T extends Tensor>(real: T|TensorLike, imag: T|TensorLike): T {\n  const $real = convertToTensor(real, 'real', 'complex');\n  const $imag = convertToTensor(imag, 'imag', 'complex');\n  util.assertShapesMatch(\n      $real.shape, $imag.shape,\n      `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n          `must match in call to tf.complex().`);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.complex($real, $imag);\n  };\n  const inputs: ComplexInputs = {real: $real, imag: $imag};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n    values: TensorLike, shape: number[], inferredShape: number[],\n    dtype?: DataType): Tensor {\n  if (dtype == null) {\n    dtype = inferDtype(values);\n  }\n  if (dtype === 'complex64') {\n    throw new Error(\n        `Cannot construct a complex64 tensor directly. ` +\n        `Please use tf.complex(real, imag).`);\n  }\n  if (!isTypedArray(values) && !Array.isArray(values) &&\n      typeof values !== 'number' && typeof values !== 'boolean' &&\n      typeof values !== 'string') {\n    throw new Error(\n        'values passed to tensor(values) must be a number/boolean/string or ' +\n        'an array of numbers/booleans/strings, or a TypedArray');\n  }\n  if (shape != null) {\n    assertNonNegativeIntegerDimensions(shape);\n\n    const providedSize = sizeFromShape(shape);\n    const inferredSize = sizeFromShape(inferredShape);\n    assert(\n        providedSize === inferredSize,\n        () =>\n            `Based on the provided shape, [${shape}], the tensor should have ` +\n            `${providedSize} values but has ${inferredSize}`);\n\n    for (let i = 0; i < inferredShape.length; ++i) {\n      const inferred = inferredShape[i];\n      const flatDimsDontMatch = i === inferredShape.length - 1 ?\n          inferred !== sizeFromShape(shape.slice(i)) :\n          true;\n      assert(\n          inferredShape[i] === shape[i] || !flatDimsDontMatch,\n          () => `Error creating a new Tensor. Inferred shape ` +\n              `(${inferredShape}) does not match the provided ` +\n              `shape (${shape}). `);\n    }\n  }\n\n  if (!isTypedArray(values) && !Array.isArray(values)) {\n    values = [values] as number[];\n  }\n\n  shape = shape || inferredShape;\n  values = dtype !== 'string' ?\n      toTypedArray(values, dtype) :\n      flatten(values as string[], [], true) as string[];\n  return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`. If the values are strings,\n *     they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor<R extends Rank>(\n    values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor<R> {\n  const inferredShape = inferShape(values, dtype);\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n  'float32': 4,\n  'float16': 2,\n  'int32': 4,\n  'uint16': 2,\n  'uint8': 1,\n  'bool': 1,\n  'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n  /**\n   * An ordered list of paths.\n   *\n   * Paths are intentionally abstract in order to be general. For example, they\n   * can be relative URL paths or relative paths on the file system.\n   */\n  paths: string[];\n\n  /**\n   * Specifications of the weights stored in the paths.\n   */\n  weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n  /**\n   * Name of the weight, e.g., 'Dense_1/bias'\n   */\n  name: string;\n\n  /**\n   * Shape of the weight.\n   */\n  shape: number[];\n\n  /**\n   * Data type of the weight.\n   */\n  dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n  /**\n   * Type of the weight.\n   *\n   * Optional.\n   *\n   * The value 'optimizer' indicates the weight belongs to an optimizer\n   * (i.e., used only during model training and not during inference).\n   */\n  group?: WeightGroup;\n\n  /**\n   * Information for dequantization of the weight.\n   */\n  quantization?: {\n    scale?: number,  // The scaling constant to multiply by.\n    min?: number,    // The (possibly nudged) minimum weight to add.\n       dtype: 'uint16'|'uint8'|'float16'  // The dtype of the quantized weights.\n  };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n  /**\n   * Whether to save only the trainable weights of the model, ignoring the\n   * non-trainable ones.\n   */\n  trainableOnly?: boolean;\n\n  /**\n   * Whether the optimizer will be saved (if exists).\n   *\n   * Default: `false`.\n   */\n  includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n  /**\n   * Information about the model artifacts saved.\n   */\n  modelArtifactsInfo: ModelArtifactsInfo;\n\n  /**\n   * HTTP responses from the server that handled the model-saving request (if\n   * any). This is applicable only to server-based saving routes.\n   */\n  responses?: Response[];\n\n  /**\n   * Error messages and related data (if any).\n   */\n  errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n  /**\n   * Timestamp for when the model is saved.\n   */\n  dateSaved: Date;\n\n  /**\n   * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n   * come in a JSON format and none of our IOHandlers support a non json\n   * format. We could conder replacing this with 'Binary' if we want to\n   * allow future handlers to save to non json formats (though they will\n   * probably want more information than 'Binary').\n   * Type of the model topology\n   *\n   * Type of the model topology\n   *\n   * Possible values:\n   *   - JSON: JSON config (human-readable, e.g., Keras JSON).\n   *   - GraphDef: TensorFlow\n   *     [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n   *     protocol buffer (binary).\n   */\n  modelTopologyType: 'JSON'|'GraphDef';\n\n  /**\n   * Size of model topology (Keras JSON or GraphDef), in bytes.\n   */\n  modelTopologyBytes?: number;\n\n  /**\n   * Size of weight specification or manifest, in bytes.\n   */\n  weightSpecsBytes?: number;\n\n  /**\n   * Size of weight value data, in bytes.\n   */\n  weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  // See\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n  /** Optimizer used for the model training. */\n  optimizer_config: {};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Loss function(s) for the model's output(s). */\n  loss: string|string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Metric function(s) for the model's output(s). */\n  metrics?: string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  weighted_metrics?: string[];\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  sample_weight_mode?: string;\n\n  loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers.  For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology?: {}|ArrayBuffer;\n\n  /**\n   * Serialized configuration for the model's training.\n   */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weight specifications.\n   *\n   * This corresponds to the weightsData below.\n   */\n  weightSpecs?: WeightsManifestEntry[];\n\n  /**\n   * Binary buffer for all weight values concatenated in the order specified\n   * by `weightSpecs`.\n   */\n  weightData?: ArrayBuffer;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology: {};\n\n  /** Model training configuration. */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weights manifest.\n   *\n   * The weights manifest consists of an ordered list of weight-manifest\n   * groups. Each weight-manifest group consists of a number of weight values\n   * stored in a number of paths. See the documentation of\n   * `WeightsManifestConfig` for more details.\n   */\n  weightsManifest: WeightsManifestConfig;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise<ModelArtifacts>;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n    Promise<SaveResult>;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n  save?: SaveHandler;\n  load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n  /**\n   * List all models in the model store.\n   *\n   * @returns A dictionary mapping paths of existing models to their\n   *   model artifacts info. Model artifacts info include type of the model's\n   *   topology, byte sizes of the topology, weights, etc.\n   */\n  listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n  /**\n   * Remove a model specified by `path`.\n   *\n   * @param path\n   * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n   *   is successful).\n   * @throws Error if deletion fails, e.g., if no model exists at `path`.\n   */\n  removeModel(path: string): Promise<ModelArtifactsInfo>;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n  /**\n   * RequestInit (options) for HTTP requests.\n   *\n   * For detailed information on the supported fields, see\n   * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n   *     https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n   */\n  requestInit?: RequestInit;\n\n  /**\n   * Progress callback.\n   */\n  onProgress?: OnProgressCallback;\n\n  /**\n   * A function used to override the `window.fetch` function.\n   */\n  fetchFunc?: Function;\n\n  /**\n   * Strict loading model: whether extraneous weights or missing\n   * weights should trigger an `Error`.\n   *\n   * If `true`, require that the provided weights exactly match those\n   * required by the layers. `false` means that both extra weights\n   * and missing weights will be silently ignored.\n   *\n   * Default: `true`.\n   */\n  strict?: boolean;\n\n  /**\n   * Path prefix for weight files, by default this is calculated from the\n   * path of the model JSON file.\n   *\n   * For instance, if the path to the model JSON file is\n   * `http://localhost/foo/model.json`, then the default path prefix will be\n   * `http://localhost/foo/`. If a weight file has the path value\n   * `group1-shard1of2` in the weight manifest, then the weight file will be\n   * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n   * if you provide a `weightPathPrefix` value of\n   * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n   * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n   */\n  weightPathPrefix?: string;\n\n  /**\n   * Whether the module or model is to be loaded from TF Hub.\n   *\n   * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n   * standard model file name and the query parameters.\n   *\n   * Default: `false`.\n   */\n  fromTFHub?: boolean;\n\n  /**\n   * An async function to convert weight file name to URL. The weight file\n   * names are stored in model.json's weightsManifest.paths field. By default we\n   * consider weight files are colocated with the model.json file. For example:\n   *     model.json URL: https://www.google.com/models/1/model.json\n   *     group1-shard1of1.bin url:\n   *        https://www.google.com/models/1/group1-shard1of1.bin\n   *\n   * With this func you can convert the weight file name to any URL.\n   */\n  weightUrlConverter?: (weightFileName: string) => Promise<string>;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n  /**\n   * Is this request for a binary file (as opposed to a json file)\n   */\n  isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n *     concatenated.\n *   - An `Array` of `WeightManifestEntry`s, carrying information including\n *     tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n    tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n    Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n  // TODO(adarob, cais): Support quantization.\n  const specs: WeightsManifestEntry[] = [];\n  const dataPromises: Array<Promise<TypedArray>> = [];\n\n  const names: string[] = Array.isArray(tensors) ?\n      tensors.map(tensor => tensor.name) :\n      Object.keys(tensors);\n\n  for (let i = 0; i < names.length; ++i) {\n    const name = names[i];\n    const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n    if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n        t.dtype !== 'string' && t.dtype !== 'complex64') {\n      throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n    }\n    const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n    if (t.dtype === 'string') {\n      const utf8bytes = new Promise<TypedArray>(async resolve => {\n        const vals = await t.bytes() as Uint8Array[];\n        const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n            NUM_BYTES_STRING_LENGTH * vals.length;\n        const bytes = new Uint8Array(totalNumBytes);\n        let offset = 0;\n        for (let i = 0; i < vals.length; i++) {\n          const val = vals[i];\n          const bytesOfLength =\n              new Uint8Array(new Uint32Array([val.length]).buffer);\n          bytes.set(bytesOfLength, offset);\n          offset += NUM_BYTES_STRING_LENGTH;\n          bytes.set(val, offset);\n          offset += val.length;\n        }\n        resolve(bytes);\n      });\n      dataPromises.push(utf8bytes);\n    } else {\n      dataPromises.push(t.data());\n    }\n    if (group != null) {\n      spec.group = group;\n    }\n    specs.push(spec);\n  }\n\n  const tensorValues = await Promise.all(dataPromises);\n  return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n *   concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n *   whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n *   to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n    buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n  // TODO(adarob, cais): Support quantization.\n  const out: NamedTensorMap = {};\n  let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n  let offset = 0;\n  for (const spec of specs) {\n    const name = spec.name;\n    const dtype = spec.dtype;\n    const shape = spec.shape;\n    const size = sizeFromShape(shape);\n    let values: TypedArray|string[]|Uint8Array[];\n\n    if ('quantization' in spec) {\n      const quantization = spec.quantization;\n      if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n        if (!('min' in quantization && 'scale' in quantization)) {\n          throw new Error(\n              `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n              `doesn't have corresponding metadata min and scale.`);\n        }\n      } else if (quantization.dtype === 'float16') {\n        if (dtype !== 'float32') {\n          throw new Error(\n              `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n              `which only supports weights of type float32 not ${dtype}.`);\n        }\n      } else {\n        throw new Error(\n            `Weight ${spec.name} has unknown ` +\n            `quantization dtype ${quantization.dtype}. ` +\n            `Supported quantization dtypes are: ` +\n            `'uint8', 'uint16', and 'float16'.`);\n      }\n      const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n      const byteBuffer =\n          buffer.slice(offset, offset + size * quantizationSizeFactor);\n      const quantizedArray = (quantization.dtype === 'uint8') ?\n          new Uint8Array(byteBuffer) :\n          new Uint16Array(byteBuffer);\n      if (dtype === 'float32') {\n        if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n          values = new Float32Array(quantizedArray.length);\n          for (let i = 0; i < quantizedArray.length; i++) {\n            const v = quantizedArray[i];\n            values[i] = v * quantization.scale + quantization.min;\n          }\n        } else if (quantization.dtype === 'float16') {\n          if (float16Decode === undefined) {\n            float16Decode = getFloat16Decoder();\n          }\n          values = float16Decode(quantizedArray as Uint16Array);\n        } else {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type float32.`);\n        }\n      } else if (dtype === 'int32') {\n        if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type int32.`);\n        }\n        values = new Int32Array(quantizedArray.length);\n        for (let i = 0; i < quantizedArray.length; i++) {\n          const v = quantizedArray[i];\n          values[i] = Math.round(v * quantization.scale + quantization.min);\n        }\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * quantizationSizeFactor;\n    } else if (dtype === 'string') {\n      const size = sizeFromShape(spec.shape);\n      values = [];\n      for (let i = 0; i < size; i++) {\n        const byteLength = new Uint32Array(\n            buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n        offset += NUM_BYTES_STRING_LENGTH;\n        const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n        (values as Uint8Array[]).push(bytes);\n        offset += byteLength;\n      }\n    } else {\n      const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n      const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n      if (dtype === 'float32') {\n        values = new Float32Array(byteBuffer);\n      } else if (dtype === 'int32') {\n        values = new Int32Array(byteBuffer);\n      } else if (dtype === 'bool') {\n        values = new Uint8Array(byteBuffer);\n      } else if (dtype === 'complex64') {\n        values = new Float32Array(byteBuffer);\n        const real = new Float32Array(values.length / 2);\n        const image = new Float32Array(values.length / 2);\n        for (let i = 0; i < real.length; i++) {\n          real[i] = values[i * 2];\n          image[i] = values[i * 2 + 1];\n        }\n        const realTensor = tensor(real, shape, 'float32');\n        const imageTensor = tensor(image, shape, 'float32');\n        out[name] = complex(realTensor, imageTensor);\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * dtypeFactor;\n    }\n    if (dtype !== 'complex64') {\n      out[name] = tensor(values, shape, dtype);\n    }\n  }\n  return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n  // TODO(adarob, cais): Support quantization.\n  if (xs === null) {\n    throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n  }\n\n  let totalByteLength = 0;\n\n  // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n  // can have a different byte length from that of the `TypedArray` itself,\n  // for example, when the `TypedArray` is created from an offset in an\n  // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n  // the `TypedArray` in byte length. If an element of `xs` does not show\n  // this property, a new `TypedArray` that satisfy this property will be\n  // constructed and pushed into `normalizedXs`.\n  const normalizedXs: TypedArray[] = [];\n  xs.forEach((x: TypedArray) => {\n    totalByteLength += x.byteLength;\n    // tslint:disable:no-any\n    normalizedXs.push(\n        x.byteLength === x.buffer.byteLength ? x :\n                                               new (x.constructor as any)(x));\n    if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n          x as any instanceof Uint8Array)) {\n      throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n    }\n    // tslint:enable:no-any\n  });\n\n  const y = new Uint8Array(totalByteLength);\n  let offset = 0;\n  normalizedXs.forEach((x: TypedArray) => {\n    y.set(new Uint8Array(x.buffer), offset);\n    offset += x.byteLength;\n  });\n\n  return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n    (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n     typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n  if (useNodeBuffer) {\n    return Buffer.byteLength(str);\n  }\n  return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n  if (useNodeBuffer) {\n    return Buffer.from(buffer).toString('base64');\n  }\n  const buf = new Uint8Array(buffer);\n  let s = '';\n  for (let i = 0, l = buf.length; i < l; i++) {\n    s += String.fromCharCode(buf[i]);\n  }\n  return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n  if (useNodeBuffer) {\n    const buf = Buffer.from(str, 'base64');\n    return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n  }\n  const s = atob(str);\n  const buffer = new Uint8Array(s.length);\n  for (let i = 0; i < s.length; ++i) {\n    buffer.set([s.charCodeAt(i)], i);\n  }\n  return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n  if (buffers.length === 1) {\n    return buffers[0];\n  }\n\n  let totalByteLength = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    totalByteLength += buffer.byteLength;\n  });\n\n  const temp = new Uint8Array(totalByteLength);\n  let offset = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    temp.set(new Uint8Array(buffer), offset);\n    offset += buffer.byteLength;\n  });\n  return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n  const SEPARATOR = '/';\n  path = path.trim();\n  while (path.endsWith(SEPARATOR)) {\n    path = path.slice(0, path.length - 1);\n  }\n  const items = path.split(SEPARATOR);\n  return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n    ModelArtifactsInfo {\n  if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n    throw new Error('Expected JSON model topology, received ArrayBuffer.');\n  }\n\n  return {\n    dateSaved: new Date(),\n    modelTopologyType: 'JSON',\n    modelTopologyBytes: modelArtifacts.modelTopology == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n    weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n    weightDataBytes: modelArtifacts.weightData == null ?\n        0 :\n        modelArtifacts.weightData.byteLength,\n  };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n  const convertMantissa = (i: number): number => {\n    let m = i << 13;\n    let e = 0;\n\n    while ((m & 0x00800000) === 0) {\n      e -= 0x00800000;\n      m <<= 1;\n    }\n    m &= ~0x00800000;\n    e += 0x38800000;\n\n    return m | e;\n  };\n\n  const mantisaTable = new Uint32Array(2048);\n\n  mantisaTable[0] = 0;\n  for (let i = 1; i < 1024; i++) {\n    mantisaTable[i] = convertMantissa(i);\n  }\n  for (let i = 1024; i < 2048; i++) {\n    mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n  }\n\n  return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n  const exponentTable = new Uint32Array(64);\n\n  exponentTable[0] = 0;\n  exponentTable[31] = 0x47800000;\n  exponentTable[32] = 0x80000000;\n  exponentTable[63] = 0xc7800000;\n  for (let i = 1; i < 31; i++) {\n    exponentTable[i] = i << 23;\n  }\n  for (let i = 33; i < 63; i++) {\n    exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n  }\n\n  return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n  const offsetTable = new Uint32Array(64);\n\n  for (let i = 0; i < 64; i++) {\n    offsetTable[i] = 1024;\n  }\n  offsetTable[0] = offsetTable[32] = 0;\n\n  return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n *          the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n  // Algorithm is based off of\n  // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n  // Cache lookup tables\n  const mantisaTable = computeFloat16MantisaTable();\n  const exponentTable = computeFloat16ExponentTable();\n  const offsetTable = computeFloat16OffsetTable();\n\n  return (quantizedArray: Uint16Array) => {\n    const buffer = new ArrayBuffer(4 * quantizedArray.length);\n    const bufferUint32View = new Uint32Array(buffer);\n    for (let index = 0; index < quantizedArray.length; index++) {\n      const float16Bits = quantizedArray[index];\n      const float32Bits =\n          mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n          exponentTable[float16Bits >> 10];\n      bufferUint32View[index] = float32Bits;\n    }\n    return new Float32Array(buffer);\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n    IOHandler;\n\nexport class IORouterRegistry {\n  // Singleton instance.\n  private static instance: IORouterRegistry;\n\n  private saveRouters: IORouter[];\n  private loadRouters: IORouter[];\n\n  private constructor() {\n    this.saveRouters = [];\n    this.loadRouters = [];\n  }\n\n  private static getInstance(): IORouterRegistry {\n    if (IORouterRegistry.instance == null) {\n      IORouterRegistry.instance = new IORouterRegistry();\n    }\n    return IORouterRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerSaveRouter(saveRouter: IORouter) {\n    IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n  }\n\n  /**\n   * Register a load-handler router.\n   *\n   * @param loadRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `load` method defined or `null`.\n   */\n  static registerLoadRouter(loadRouter: IORouter) {\n    IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n  }\n\n  /**\n   * Look up IOHandler for saving, given a URL-like string.\n   *\n   * @param url\n   * @returns If only one match is found, an instance of IOHandler with the\n   * `save` method defined. If no match is found, `null`.\n   * @throws Error, if more than one match is found.\n   */\n  static getSaveHandlers(url: string|string[]): IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'save');\n  }\n\n  /**\n   * Look up IOHandler for loading, given a URL-like string.\n   *\n   * @param url\n   * @param loadOptions Optional, custom load options.\n   * @returns All valid handlers for `url`, given the currently registered\n   *   handler routers.\n   */\n  static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n      IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n  }\n\n  private static getHandlers(\n      url: string|string[], handlerType: 'save'|'load',\n      loadOptions?: LoadOptions): IOHandler[] {\n    const validHandlers: IOHandler[] = [];\n    const routers = handlerType === 'load' ?\n        IORouterRegistry.getInstance().loadRouters :\n        IORouterRegistry.getInstance().saveRouters;\n    routers.forEach(router => {\n      const handler = router(url, loadOptions);\n      if (handler !== null) {\n        validHandlers.push(handler);\n      }\n    });\n    return validHandlers;\n  }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n    IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n    (url: string|string[], loadOptions?: LoadOptions) =>\n        IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n//    as the type of topology (JSON vs binary), byte size of the topology, byte\n//    size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise<void> {\n  const idbFactory = getIndexedDBFactory();\n\n  return new Promise<void>((resolve, reject) => {\n    const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n    deleteRequest.onsuccess = () => resolve();\n    deleteRequest.onerror = error => reject(error);\n  });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n  if (!env().getBool('IS_BROWSER')) {\n    // TODO(cais): Add more info about what IOHandler subtypes are available.\n    //   Maybe point to a doc page on the web and/or automatically determine\n    //   the available IOHandlers and print them in the error message.\n    throw new Error(\n        'Failed to obtain IndexedDB factory because the current environment' +\n        'is not a web browser.');\n  }\n  // tslint:disable-next-line:no-any\n  const theWindow: any = typeof window === 'undefined' ? self : window;\n  const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n      theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n      theWindow.shimIndexedDB;\n  if (factory == null) {\n    throw new Error(\n        'The current browser does not appear to support IndexedDB.');\n  }\n  return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n  const db = openRequest.result as IDBDatabase;\n  db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n  db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n  protected readonly indexedDB: IDBFactory;\n  protected readonly modelPath: string;\n\n  static readonly URL_SCHEME = 'indexeddb://';\n\n  constructor(modelPath: string) {\n    this.indexedDB = getIndexedDBFactory();\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For IndexedDB, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    // TODO(cais): Support saving GraphDef models.\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    return this.databaseAction(this.modelPath, modelArtifacts) as\n        Promise<SaveResult>;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    return this.databaseAction(this.modelPath) as Promise<ModelArtifacts>;\n  }\n\n  /**\n   * Perform database action to put model artifacts into or read model artifacts\n   * from IndexedDB object store.\n   *\n   * Whether the action is put or get depends on whether `modelArtifacts` is\n   * specified. If it is specified, the action will be put; otherwise the action\n   * will be get.\n   *\n   * @param modelPath A unique string path for the model.\n   * @param modelArtifacts If specified, it will be the model artifacts to be\n   *   stored in IndexedDB.\n   * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n   *   of `ModelArtifacts`, if the action is get.\n   */\n  private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n      Promise<ModelArtifacts|SaveResult> {\n    return new Promise<ModelArtifacts|SaveResult>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n\n        if (modelArtifacts == null) {\n          // Read model out from object store.\n          const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n          const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n          const getRequest = modelStore.get(this.modelPath);\n          getRequest.onsuccess = () => {\n            if (getRequest.result == null) {\n              db.close();\n              return reject(new Error(\n                  `Cannot find model with path '${this.modelPath}' ` +\n                  `in IndexedDB.`));\n            } else {\n              resolve(getRequest.result.modelArtifacts);\n            }\n          };\n          getRequest.onerror = error => {\n            db.close();\n            return reject(getRequest.error);\n          };\n          modelTx.oncomplete = () => db.close();\n        } else {\n          // Put model into object store.\n          const modelArtifactsInfo: ModelArtifactsInfo =\n              getModelArtifactsInfoForJSON(modelArtifacts);\n          // First, put ModelArtifactsInfo into info store.\n          const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n          let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n          const putInfoRequest =\n              infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n          let modelTx: IDBTransaction;\n          putInfoRequest.onsuccess = () => {\n            // Second, put model data into model store.\n            modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n            const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n            const putModelRequest = modelStore.put({\n              modelPath: this.modelPath,\n              modelArtifacts,\n              modelArtifactsInfo\n            });\n            putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n            putModelRequest.onerror = error => {\n              // If the put-model request fails, roll back the info entry as\n              // well.\n              infoStore = infoTx.objectStore(INFO_STORE_NAME);\n              const deleteInfoRequest = infoStore.delete(this.modelPath);\n              deleteInfoRequest.onsuccess = () => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n              deleteInfoRequest.onerror = error => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n            };\n          };\n          putInfoRequest.onerror = error => {\n            db.close();\n            return reject(putInfoRequest.error);\n          };\n          infoTx.oncomplete = () => {\n            if (modelTx == null) {\n              db.close();\n            } else {\n              modelTx.oncomplete = () => db.close();\n            }\n          };\n        }\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n      return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n *   which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n  return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n      key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n      key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n  private indexedDB: IDBFactory;\n\n  constructor() {\n    this.indexedDB = getIndexedDBFactory();\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    return new Promise<{[path: string]: ModelArtifactsInfo}>(\n        (resolve, reject) => {\n          const openRequest =\n              this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n          openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n          openRequest.onsuccess = () => {\n            const db = openRequest.result;\n            const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n            const store = tx.objectStore(INFO_STORE_NAME);\n            // tslint:disable:max-line-length\n            // Need to cast `store` as `any` here because TypeScript's DOM\n            // library does not have the `getAll()` method even though the\n            // method is supported in the latest version of most mainstream\n            // browsers:\n            // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n            // tslint:enable:max-line-length\n            // tslint:disable-next-line:no-any\n            const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n            getAllInfoRequest.onsuccess = () => {\n              const out: {[path: string]: ModelArtifactsInfo} = {};\n              for (const item of getAllInfoRequest.result) {\n                out[item.modelPath] = item.modelArtifactsInfo;\n              }\n              resolve(out);\n            };\n            getAllInfoRequest.onerror = error => {\n              db.close();\n              return reject(getAllInfoRequest.error);\n            };\n            tx.oncomplete = () => db.close();\n          };\n          openRequest.onerror = error => reject(openRequest.error);\n        });\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    return new Promise<ModelArtifactsInfo>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n        const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n        const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n        const getInfoRequest = infoStore.get(path);\n        let modelTx: IDBTransaction;\n        getInfoRequest.onsuccess = () => {\n          if (getInfoRequest.result == null) {\n            db.close();\n            return reject(new Error(\n                `Cannot find model with path '${path}' ` +\n                `in IndexedDB.`));\n          } else {\n            // First, delete the entry in the info store.\n            const deleteInfoRequest = infoStore.delete(path);\n            const deleteModelData = () => {\n              // Second, delete the entry in the model store.\n              modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n              const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n              const deleteModelRequest = modelStore.delete(path);\n              deleteModelRequest.onsuccess = () =>\n                  resolve(getInfoRequest.result.modelArtifactsInfo);\n              deleteModelRequest.onerror = error =>\n                  reject(getInfoRequest.error);\n            };\n            // Proceed with deleting model data regardless of whether deletion\n            // of info data succeeds or not.\n            deleteInfoRequest.onsuccess = deleteModelData;\n            deleteInfoRequest.onerror = error => {\n              deleteModelData();\n              db.close();\n              return reject(getInfoRequest.error);\n            };\n          }\n        };\n        getInfoRequest.onerror = error => {\n          db.close();\n          return reject(getInfoRequest.error);\n        };\n\n        infoTx.oncomplete = () => {\n          if (modelTx == null) {\n            db.close();\n          } else {\n            modelTx.oncomplete = () => db.close();\n          }\n        };\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n  if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n      typeof window.localStorage === 'undefined') {\n    throw new Error(\n        'purgeLocalStorageModels() cannot proceed because local storage is ' +\n        'unavailable in the current environment.');\n  }\n  const LS = window.localStorage;\n  const purgedModelPaths: string[] = [];\n  for (let i = 0; i < LS.length; ++i) {\n    const key = LS.key(i);\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    if (key.startsWith(prefix) && key.length > prefix.length) {\n      LS.removeItem(key);\n      const modelName = getModelPathFromKey(key);\n      if (purgedModelPaths.indexOf(modelName) === -1) {\n        purgedModelPaths.push(modelName);\n      }\n    }\n  }\n  return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n} {\n  return {\n    info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n    topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n    weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n    weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n    modelMetadata:\n        [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n  };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n  const items = key.split(PATH_SEPARATOR);\n  if (items.length < 3) {\n    throw new Error(`Invalid key format: ${key}`);\n  }\n  return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n      key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n      key;\n}\n\ndeclare type LocalStorageKeys = {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n  protected readonly LS: Storage;\n  protected readonly modelPath: string;\n  protected readonly keys: LocalStorageKeys;\n\n  static readonly URL_SCHEME = 'localstorage://';\n\n  constructor(modelPath: string) {\n    if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n        typeof window.localStorage === 'undefined') {\n      // TODO(cais): Add more info about what IOHandler subtypes are\n      // available.\n      //   Maybe point to a doc page on the web and/or automatically determine\n      //   the available IOHandlers and print them in the error message.\n      throw new Error(\n          'The current environment does not support local storage.');\n    }\n    this.LS = window.localStorage;\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For local storage, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n    this.keys = getModelKeys(this.modelPath);\n  }\n\n  /**\n   * Save model artifacts to browser local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @param modelArtifacts The model artifacts to be stored.\n   * @returns An instance of SaveResult.\n   */\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const topology = JSON.stringify(modelArtifacts.modelTopology);\n      const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n      const modelArtifactsInfo: ModelArtifactsInfo =\n          getModelArtifactsInfoForJSON(modelArtifacts);\n\n      try {\n        this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n        this.LS.setItem(this.keys.topology, topology);\n        this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n        this.LS.setItem(\n            this.keys.weightData,\n            arrayBufferToBase64String(modelArtifacts.weightData));\n        this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n          format: modelArtifacts.format,\n          generatedBy: modelArtifacts.generatedBy,\n          convertedBy: modelArtifacts.convertedBy,\n          userDefinedMetadata: modelArtifacts.userDefinedMetadata\n        }));\n\n        return {modelArtifactsInfo};\n      } catch (err) {\n        // If saving failed, clean up all items saved so far.\n        this.LS.removeItem(this.keys.info);\n        this.LS.removeItem(this.keys.topology);\n        this.LS.removeItem(this.keys.weightSpecs);\n        this.LS.removeItem(this.keys.weightData);\n        this.LS.removeItem(this.keys.modelMetadata);\n\n        throw new Error(\n            `Failed to save model '${this.modelPath}' to local storage: ` +\n            `size quota being exceeded is a possible cause of this failure: ` +\n            `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n            `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n            `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n      }\n    }\n  }\n\n  /**\n   * Load a model from local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const info =\n        JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n    if (info == null) {\n      throw new Error(\n          `In local storage, there is no model with name '${this.modelPath}'`);\n    }\n\n    if (info.modelTopologyType !== 'JSON') {\n      throw new Error(\n          'BrowserLocalStorage does not support loading non-JSON model ' +\n          'topology yet.');\n    }\n\n    const out: ModelArtifacts = {};\n\n    // Load topology.\n    const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n    if (topology == null) {\n      throw new Error(\n          `In local storage, the topology of model '${this.modelPath}' ` +\n          `is missing.`);\n    }\n    out.modelTopology = topology;\n\n    // Load weight specs.\n    const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n    if (weightSpecs == null) {\n      throw new Error(\n          `In local storage, the weight specs of model '${this.modelPath}' ` +\n          `are missing.`);\n    }\n    out.weightSpecs = weightSpecs;\n\n    // Load meta-data fields.\n    const metadataString = this.LS.getItem(this.keys.modelMetadata);\n    if (metadataString != null) {\n      const metadata = JSON.parse(metadataString) as ModelArtifacts;\n      out.format = metadata['format'];\n      out.generatedBy = metadata['generatedBy'];\n      out.convertedBy = metadata['convertedBy'];\n      out.userDefinedMetadata = metadata['userDefinedMetadata'];\n    }\n\n    // Load weight data.\n    const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n    if (weightDataBase64 == null) {\n      throw new Error(\n          `In local storage, the binary weight values of model ` +\n          `'${this.modelPath}' are missing.`);\n    }\n    out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n    return out;\n  }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n      return browserLocalStorage(\n          url.slice(BrowserLocalStorage.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n *   - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n *     model, such as date saved, type of the topology, size in bytes, etc.\n *   - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n *     style models, this is a stringized JSON.\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n *     model, can be used to decode the saved binary weight values (see\n *     item below).\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n *     weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n *   `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n  return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n  private readonly LS: Storage;\n\n  constructor() {\n    assert(\n        env().getBool('IS_BROWSER'),\n        () => 'Current environment is not a web browser');\n    assert(\n        typeof window === 'undefined' ||\n            typeof window.localStorage !== 'undefined',\n        () => 'Current browser does not appear to support localStorage');\n    this.LS = window.localStorage;\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    const out: {[path: string]: ModelArtifactsInfo} = {};\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n    for (let i = 0; i < this.LS.length; ++i) {\n      const key = this.LS.key(i);\n      if (key.startsWith(prefix) && key.endsWith(suffix)) {\n        const modelPath = getModelPathFromKey(key);\n        out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n      }\n    }\n    return out;\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    const keys = getModelKeys(path);\n    if (this.LS.getItem(keys.info) == null) {\n      throw new Error(`Cannot find model at path '${path}'`);\n    }\n    const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n    this.LS.removeItem(keys.info);\n    this.LS.removeItem(keys.topology);\n    this.LS.removeItem(keys.weightSpecs);\n    this.LS.removeItem(keys.weightData);\n    return info;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n *   string.\n * - Moving or copying model from one path to another in the same medium or from\n *   one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n  // Singleton instance.\n  private static instance: ModelStoreManagerRegistry;\n\n  private managers: {[scheme: string]: ModelStoreManager};\n\n  private constructor() {\n    this.managers = {};\n  }\n\n  private static getInstance(): ModelStoreManagerRegistry {\n    if (ModelStoreManagerRegistry.instance == null) {\n      ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n    }\n    return ModelStoreManagerRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerManager(scheme: string, manager: ModelStoreManager) {\n    assert(scheme != null, () => 'scheme must not be undefined or null.');\n    if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n      scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n    }\n    assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n    const registry = ModelStoreManagerRegistry.getInstance();\n    assert(\n        registry.managers[scheme] == null,\n        () => `A model store manager is already registered for scheme '${\n            scheme}'.`);\n    registry.managers[scheme] = manager;\n  }\n\n  static getManager(scheme: string): ModelStoreManager {\n    const manager = this.getInstance().managers[scheme];\n    if (manager == null) {\n      throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n    }\n    return manager;\n  }\n\n  static getSchemes(): string[] {\n    return Object.keys(this.getInstance().managers);\n  }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n *   Scheme: e.g., 'localstorage' in the example above.\n *   Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n  if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n    throw new Error(\n        `The url string provided does not contain a scheme. ` +\n        `Supported schemes are: ` +\n        `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n  }\n  return {\n    scheme: url.split(URL_SCHEME_SUFFIX)[0],\n    path: url.split(URL_SCHEME_SUFFIX)[1],\n  };\n}\n\nasync function cloneModelInternal(\n    sourceURL: string, destURL: string,\n    deleteSource = false): Promise<ModelArtifactsInfo> {\n  assert(\n      sourceURL !== destURL,\n      () => `Old path and new path are the same: '${sourceURL}'`);\n\n  const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n  assert(\n      loadHandlers.length > 0,\n      () => `Copying failed because no load handler is found for source URL ${\n          sourceURL}.`);\n  assert(\n      loadHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `load handlers for source URL ${sourceURL}.`);\n  const loadHandler = loadHandlers[0];\n\n  const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n  assert(\n      saveHandlers.length > 0,\n      () => `Copying failed because no save handler is found for destination ` +\n          `URL ${destURL}.`);\n  assert(\n      saveHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `save handlers for destination URL ${destURL}.`);\n  const saveHandler = saveHandlers[0];\n\n  const sourceScheme = parseURL(sourceURL).scheme;\n  const sourcePath = parseURL(sourceURL).path;\n  const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n  const modelArtifacts = await loadHandler.load();\n\n  // If moving within the same storage medium, remove the old model as soon as\n  // the loading is done. Without doing this, it is possible that the combined\n  // size of the two models will cause the cloning to fail.\n  if (deleteSource && sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  const saveResult = await saveHandler.save(modelArtifacts);\n\n  // If moving between mediums, the deletion is done after the save succeeds.\n  // This guards against the case in which saving to the destination medium\n  // fails.\n  if (deleteSource && !sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n *   'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n  const schemes = ModelStoreManagerRegistry.getSchemes();\n  const out: {[url: string]: ModelArtifactsInfo} = {};\n  for (const scheme of schemes) {\n    const schemeOut =\n        await ModelStoreManagerRegistry.getManager(scheme).listModels();\n    for (const path in schemeOut) {\n      const url = scheme + URL_SCHEME_SUFFIX + path;\n      out[url] = schemeOut[path];\n    }\n  }\n  return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n *   'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n *   is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise<ModelArtifactsInfo> {\n  const schemeAndPath = parseURL(url);\n  const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n  return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function copyModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = false;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function moveModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = true;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n  // According to the spec, the built-in encoder can do only UTF-8 encoding.\n  // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n  private textEncoder: TextEncoder;\n\n  fetch(path: string, init?: RequestInit): Promise<Response> {\n    return fetch(path, init);\n  }\n\n  now(): number {\n    return performance.now();\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Browser's encoder only supports utf-8, but got ${encoding}`);\n    }\n    if (this.textEncoder == null) {\n      this.textEncoder = new TextEncoder();\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    return new TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_BROWSER')) {\n  env().setPlatform('browser', new PlatformBrowser());\n\n  // Register LocalStorage IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n  } catch (err) {\n  }\n\n  // Register IndexedDB IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n  } catch (err) {\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n  // tslint:disable-next-line:no-require-imports\n  importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise<Response>;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n  systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n  systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n  return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n  private textEncoder: TextEncoder;\n  // tslint:disable-next-line:no-any\n  util: any;\n\n  constructor() {\n    // tslint:disable-next-line:no-require-imports\n    this.util = require('util');\n    // According to the spec, the built-in encoder can do only UTF-8 encoding.\n    // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n    this.textEncoder = new this.util.TextEncoder();\n  }\n\n  fetch(path: string, requestInits?: RequestInit): Promise<Response> {\n    if (env().global.fetch != null) {\n      return env().global.fetch(path, requestInits);\n    }\n\n    if (systemFetch == null) {\n      systemFetch = getNodeFetch.importFetch();\n    }\n    return systemFetch(path, requestInits);\n  }\n\n  now(): number {\n    const time = process.hrtime();\n    return time[0] * 1000 + time[1] / 1000000;\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Node built-in encoder only supports utf-8, but got ${encoding}`);\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    if (bytes.length === 0) {\n      return '';\n    }\n    return new this.util.TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_NODE')) {\n  env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer<R extends Rank, D extends DataType = 'float32'>(\n    shape: ShapeMap[R], dtype: D = 'float32' as D,\n    values?: DataTypeMap[D]): TensorBuffer<R, D> {\n  dtype = dtype || 'float32' as D;\n  util.assertNonNegativeIntegerDimensions(shape);\n  return new TensorBuffer<R, D>(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_<T extends Tensor>(x: T|TensorLike, dtype: DataType): T {\n  const $x = convertToTensor(x, 'x', 'cast');\n\n  // Sanity checks.\n  if (!util.isValidDtype(dtype)) {\n    throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n  }\n  if (dtype === 'string' && $x.dtype !== 'string' ||\n      dtype !== 'string' && $x.dtype === 'string') {\n    throw new Error('Only strings can be casted to strings');\n  }\n\n  const inputs: CastInputs = {x: $x};\n  const attrs: CastAttrs = {dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n      null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'clone', null);\n  const forward = () =>\n      ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n  const inputs: IdentityInputs = {x: $x};\n\n  // Note this op is called tf.identity in python. Hence the kernel name used\n  // here.\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print<T extends Tensor>(x: T, verbose = false): void {\n  console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n  buffer,\n  cast,\n  clone,\n  print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer<T>(f: () => T): Promise<T> {\n  return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n  private readonly modelTopologyFileName: string;\n  private readonly weightDataFileName: string;\n  private readonly jsonAnchor: HTMLAnchorElement;\n  private readonly weightDataAnchor: HTMLAnchorElement;\n\n  static readonly URL_SCHEME = 'downloads://';\n\n  constructor(fileNamePrefix?: string) {\n    if (!env().getBool('IS_BROWSER')) {\n      // TODO(cais): Provide info on what IOHandlers are available under the\n      //   current environment.\n      throw new Error(\n          'browserDownloads() cannot proceed because the current environment ' +\n          'is not a browser.');\n    }\n\n    if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n      fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n    }\n    if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n      fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n    }\n\n    this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n    this.weightDataFileName =\n        fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (typeof (document) === 'undefined') {\n      throw new Error(\n          'Browser downloads are not supported in ' +\n          'this environment since `document` is not present');\n    }\n    const weightsURL = window.URL.createObjectURL(new Blob(\n        [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserDownloads.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const weightsManifest: WeightsManifestConfig = [{\n        paths: ['./' + this.weightDataFileName],\n        weights: modelArtifacts.weightSpecs\n      }];\n      const modelTopologyAndWeightManifest: ModelJSON = {\n        modelTopology: modelArtifacts.modelTopology,\n        format: modelArtifacts.format,\n        generatedBy: modelArtifacts.generatedBy,\n        convertedBy: modelArtifacts.convertedBy,\n        weightsManifest\n      };\n      const modelTopologyAndWeightManifestURL =\n          window.URL.createObjectURL(new Blob(\n              [JSON.stringify(modelTopologyAndWeightManifest)],\n              {type: 'application/json'}));\n\n      // If anchor elements are not provided, create them without attaching them\n      // to parents, so that the downloaded file names can be controlled.\n      const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n                                                   this.jsonAnchor;\n      jsonAnchor.download = this.modelTopologyFileName;\n      jsonAnchor.href = modelTopologyAndWeightManifestURL;\n      // Trigger downloads by evoking a click event on the download anchors.\n      // When multiple downloads are started synchronously, Firefox will only\n      // save the last one.\n      await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n      if (modelArtifacts.weightData != null) {\n        const weightDataAnchor = this.weightDataAnchor == null ?\n            document.createElement('a') :\n            this.weightDataAnchor;\n        weightDataAnchor.download = this.weightDataFileName;\n        weightDataAnchor.href = weightsURL;\n        await defer(\n            () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n      }\n\n      return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n    }\n  }\n}\n\nclass BrowserFiles implements IOHandler {\n  private readonly files: File[];\n\n  constructor(files: File[]) {\n    if (files == null || files.length < 1) {\n      throw new Error(\n          `When calling browserFiles, at least 1 file is required, ` +\n          `but received ${files}`);\n    }\n    this.files = files;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    const jsonFile = this.files[0];\n    const weightFiles = this.files.slice(1);\n\n    return new Promise<ModelArtifacts>((resolve, reject) => {\n      const jsonReader = new FileReader();\n      jsonReader.onload = (event: Event) => {\n        // tslint:disable-next-line:no-any\n        const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n        const modelTopology = modelJSON.modelTopology;\n        if (modelTopology == null) {\n          reject(new Error(\n              `modelTopology field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        if (weightFiles.length === 0) {\n          resolve({modelTopology});\n        }\n\n        const weightsManifest = modelJSON.weightsManifest;\n        if (weightsManifest == null) {\n          reject(new Error(\n              `weightManifest field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        let pathToFile: {[path: string]: File};\n        try {\n          pathToFile =\n              this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n        } catch (err) {\n          reject(err);\n          return;\n        }\n\n        const weightSpecs: WeightsManifestEntry[] = [];\n        const paths: string[] = [];\n        const perFileBuffers: ArrayBuffer[] = [];\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            paths.push(path);\n            perFileBuffers.push(null);\n          });\n          weightSpecs.push(...weightsGroup.weights);\n        });\n\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            const weightFileReader = new FileReader();\n            weightFileReader.onload = (event: Event) => {\n              // tslint:disable-next-line:no-any\n              const weightData = (event.target as any).result as ArrayBuffer;\n              const index = paths.indexOf(path);\n              perFileBuffers[index] = weightData;\n              if (perFileBuffers.indexOf(null) === -1) {\n                resolve({\n                  modelTopology,\n                  weightSpecs,\n                  weightData: concatenateArrayBuffers(perFileBuffers),\n                  format: modelJSON.format,\n                  generatedBy: modelJSON.generatedBy,\n                  convertedBy: modelJSON.convertedBy,\n                  userDefinedMetadata: modelJSON.userDefinedMetadata\n                });\n              }\n            };\n            weightFileReader.onerror = error =>\n                reject(`Failed to weights data from file of path '${path}'.`);\n            weightFileReader.readAsArrayBuffer(pathToFile[path]);\n          });\n        });\n      };\n      jsonReader.onerror = error => reject(\n          `Failed to read model topology and weights manifest JSON ` +\n          `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n          `Keras-style tf.Model artifacts only.`);\n      jsonReader.readAsText(jsonFile);\n    });\n  }\n\n  /**\n   * Check the compatibility between weights manifest and weight files.\n   */\n  private checkManifestAndWeightFiles(\n      manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n    const basenames: string[] = [];\n    const fileNames = files.map(file => basename(file.name));\n    const pathToFile: {[path: string]: File} = {};\n    for (const group of manifest) {\n      group.paths.forEach(path => {\n        const pathBasename = basename(path);\n        if (basenames.indexOf(pathBasename) !== -1) {\n          throw new Error(\n              `Duplicate file basename found in weights manifest: ` +\n              `'${pathBasename}'`);\n        }\n        basenames.push(pathBasename);\n        if (fileNames.indexOf(pathBasename) === -1) {\n          throw new Error(\n              `Weight file with basename '${pathBasename}' is not provided.`);\n        } else {\n          pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n        }\n      });\n    }\n\n    if (basenames.length !== files.length) {\n      throw new Error(\n          `Mismatch in the number of files in weights manifest ` +\n          `(${basenames.length}) and the number of weight files provided ` +\n          `(${files.length}).`);\n    }\n    return pathToFile;\n  }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n      return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * //   'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n *   `tf.Model`, `fileNamePrefix` should follow either of the following two\n *   formats:\n *   1. `null` or `undefined`, in which case the default file\n *      names will be used:\n *      - 'model.json' for the JSON file containing the model topology and\n *        weights manifest.\n *      - 'model.weights.bin' for the binary file containing the binary weight\n *        values.\n *   2. A single string or an Array of a single string, as the file name prefix.\n *      For example, if `'foo'` is provided, the downloaded JSON\n *      file and binary weights file will be named 'foo.json' and\n *      'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n  return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * //   elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (`<input type=\"file\" ...>`)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n *   which an `Array` of `File`s is expected (in that order):\n *   - A JSON file containing the model topology and weight manifest.\n *   - Optionally, One or more binary files containing the binary weights.\n *     These files must have names that match the paths in the `weightsManifest`\n *     contained by the aforementioned JSON file, or errors will be thrown\n *     during loading. These weights files have the same format as the ones\n *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n *     Python PIP package. If no weights files are provided, only the model\n *     topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n  return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n    promises: Array<Promise<{}|void>>, onProgress: OnProgressCallback,\n    startFraction?: number, endFraction?: number) {\n  checkPromises(promises);\n  startFraction = startFraction == null ? 0 : startFraction;\n  endFraction = endFraction == null ? 1 : endFraction;\n  checkFraction(startFraction, endFraction);\n  let resolvedPromise = 0;\n\n  const registerMonitor = (promise: Promise<{}>) => {\n    promise.then(value => {\n      const fraction = startFraction +\n          ++resolvedPromise / promises.length * (endFraction - startFraction);\n      // pass fraction as parameter to callback function.\n      onProgress(fraction);\n      return value;\n    });\n    return promise;\n  };\n\n  function checkPromises(promises: Array<Promise<{}|void>>): void {\n    assert(\n        promises != null && Array.isArray(promises) && promises.length > 0,\n        () => 'promises must be a none empty array');\n  }\n\n  function checkFraction(startFraction: number, endFraction: number): void {\n    assert(\n        startFraction >= 0 && startFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got startFraction ${startFraction}`);\n    assert(\n        endFraction >= 0 && endFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got endFraction ${endFraction}`);\n    assert(\n        endFraction >= startFraction,\n        () => `startFraction must be no more than endFraction, but ` +\n            `got startFraction ${startFraction} and endFraction ` +\n            `${endFraction}`);\n  }\n\n  return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n *   before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n *   length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n    fetchURLs: string[], loadOptions?: LoadOptions): Promise<ArrayBuffer[]> {\n  if (loadOptions == null) {\n    loadOptions = {};\n  }\n\n  const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n                                                    loadOptions.fetchFunc;\n\n  // Create the requests for all of the weights in parallel.\n  const requests = fetchURLs.map(\n      fetchURL =>\n          fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n  const fetchStartFraction = 0;\n  const fetchEndFraction = 0.5;\n\n  const responses = loadOptions.onProgress == null ?\n      await Promise.all(requests) :\n      await monitorPromisesProgress(\n          requests, loadOptions.onProgress, fetchStartFraction,\n          fetchEndFraction);\n\n  const bufferPromises = responses.map(response => response.arrayBuffer());\n\n  const bufferStartFraction = 0.5;\n  const bufferEndFraction = 1;\n\n  const buffers = loadOptions.onProgress == null ?\n      await Promise.all(bufferPromises) :\n      await monitorPromisesProgress(\n          bufferPromises, loadOptions.onProgress, bufferStartFraction,\n          bufferEndFraction);\n  return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n *     Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n    manifest: WeightsManifestConfig, filePathPrefix = '',\n    weightNames?: string[],\n    requestInit?: RequestInit): Promise<NamedTensorMap> {\n  // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n  // single weight from a group, the whole group will be fetched. At a future\n  // date, we should support fetching only the individual shards within a\n  // group that are needed to reconstruct the requested weight.\n  // TODO(cais): Use `decodeWeights` for implementation.\n\n  const fetchWeights = (fetchUrls: string[]) =>\n      loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n  const loadWeights = weightsLoaderFactory(fetchWeights);\n\n  return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n *   filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n *   fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n    fetchWeightsFunction: (fetchUrls: string[]) => Promise<ArrayBuffer[]>):\n    (manifest: WeightsManifestConfig, filePathPrefix?: string,\n     weightNames?: string[]) => Promise<NamedTensorMap> {\n  return async(\n             manifest: WeightsManifestConfig, filePathPrefix = '',\n             weightNames?: string[]): Promise<NamedTensorMap> => {\n    // Collect all the groups, weights, and their relative offsets to be\n    // fetched.\n    const groupIndicesToFetchMap = manifest.map(() => false);\n    const groupWeightsToFetch: {\n      [group: number]: Array<{\n        manifestEntry: WeightsManifestEntry; groupOffset: number;\n        sizeBytes: number;\n      }>\n    } = {};\n    const weightsFound =\n        weightNames != null ? weightNames.map(() => false) : [];\n    const allManifestWeightNames: string[] = [];\n    manifest.forEach((manifestGroupConfig, groupIndex) => {\n      let groupOffset = 0;\n      manifestGroupConfig.weights.forEach(weightsEntry => {\n        const rawDtype = ('quantization' in weightsEntry) ?\n            weightsEntry.quantization.dtype :\n            weightsEntry.dtype;\n\n        const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n            util.sizeFromShape(weightsEntry.shape);\n\n        const enqueueWeightsForFetchingFn = () => {\n          groupIndicesToFetchMap[groupIndex] = true;\n          if (groupWeightsToFetch[groupIndex] == null) {\n            groupWeightsToFetch[groupIndex] = [];\n          }\n\n          groupWeightsToFetch[groupIndex].push({\n            manifestEntry: weightsEntry,\n            groupOffset,\n            sizeBytes: weightsBytes\n          });\n        };\n\n        if (weightNames != null) {\n          weightNames.forEach((weightName, weightIndex) => {\n            if (weightName === weightsEntry.name) {\n              enqueueWeightsForFetchingFn();\n              weightsFound[weightIndex] = true;\n            }\n          });\n        } else {\n          enqueueWeightsForFetchingFn();\n        }\n\n        allManifestWeightNames.push(weightsEntry.name);\n        groupOffset += weightsBytes;\n      });\n    });\n\n    if (!weightsFound.every(found => found)) {\n      const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n      throw new Error(\n          `Could not find weights in manifest with names: ` +\n          `${weightsNotFound.join(', ')}. \\n` +\n          `Manifest JSON has weights with names: ` +\n          `${allManifestWeightNames.join(', ')}.`);\n    }\n\n    // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n    // IDs.\n    const groupIndicesToFetch =\n        groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n          if (shouldFetch) {\n            accumulator.push(i);\n          }\n          return accumulator;\n        }, []);\n\n    const fetchUrls: string[] = [];\n    groupIndicesToFetch.forEach(i => {\n      manifest[i].paths.forEach(filepath => {\n        const fetchUrl = filePathPrefix +\n            (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n        fetchUrls.push(fetchUrl);\n      });\n    });\n    const buffers = await fetchWeightsFunction(fetchUrls);\n\n    const weightsTensorMap: NamedTensorMap = {};\n    let bufferIndexOffset = 0;\n    groupIndicesToFetch.forEach(i => {\n      const numBuffers = manifest[i].paths.length;\n\n      let groupBytes = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        groupBytes += buffers[bufferIndexOffset + i].byteLength;\n      }\n\n      // Create a buffer for the whole group.\n      const groupBuffer = new ArrayBuffer(groupBytes);\n      const groupByteBuffer = new Uint8Array(groupBuffer);\n      let groupBufferOffset = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n        groupByteBuffer.set(buffer, groupBufferOffset);\n        groupBufferOffset += buffer.byteLength;\n      }\n\n      const weightsEntries = groupWeightsToFetch[i];\n      weightsEntries.forEach(weightsEntry => {\n        const byteBuffer = groupBuffer.slice(\n            weightsEntry.groupOffset,\n            weightsEntry.groupOffset + weightsEntry.sizeBytes);\n        const nameToTensorMap =\n            decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n        for (const name in nameToTensorMap) {\n          weightsTensorMap[name] = nameToTensorMap[name];\n        }\n      });\n\n      bufferIndexOffset += numBuffers;\n    });\n\n    return weightsTensorMap;\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n  protected readonly path: string;\n  protected readonly requestInit: RequestInit;\n\n  private readonly fetch: Function;\n  private readonly weightUrlConverter: (weightName: string) => Promise<string>;\n\n  readonly DEFAULT_METHOD = 'POST';\n\n  static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n  private readonly weightPathPrefix: string;\n  private readonly onProgress: OnProgressCallback;\n\n  constructor(path: string, loadOptions?: LoadOptions) {\n    if (loadOptions == null) {\n      loadOptions = {};\n    }\n    this.weightPathPrefix = loadOptions.weightPathPrefix;\n    this.onProgress = loadOptions.onProgress;\n    this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n    if (loadOptions.fetchFunc != null) {\n      assert(\n          typeof loadOptions.fetchFunc === 'function',\n          () => 'Must pass a function that matches the signature of ' +\n              '`fetch` (see ' +\n              'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n      this.fetch = loadOptions.fetchFunc;\n    } else {\n      this.fetch = env().platform.fetch;\n    }\n\n    assert(\n        path != null && path.length > 0,\n        () => 'URL path for http must not be null, undefined or ' +\n            'empty.');\n\n    if (Array.isArray(path)) {\n      assert(\n          path.length === 2,\n          () => 'URL paths for http must have a length of 2, ' +\n              `(actual length is ${path.length}).`);\n    }\n    this.path = path;\n\n    if (loadOptions.requestInit != null &&\n        loadOptions.requestInit.body != null) {\n      throw new Error(\n          'requestInit is expected to have no pre-existing body, but has one.');\n    }\n    this.requestInit = loadOptions.requestInit || {};\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserHTTPRequest.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n    init.body = new FormData();\n\n    const weightsManifest: WeightsManifestConfig = [{\n      paths: ['./model.weights.bin'],\n      weights: modelArtifacts.weightSpecs,\n    }];\n    const modelTopologyAndWeightManifest: ModelJSON = {\n      modelTopology: modelArtifacts.modelTopology,\n      format: modelArtifacts.format,\n      generatedBy: modelArtifacts.generatedBy,\n      convertedBy: modelArtifacts.convertedBy,\n      userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n      weightsManifest\n    };\n\n    init.body.append(\n        'model.json',\n        new Blob(\n            [JSON.stringify(modelTopologyAndWeightManifest)],\n            {type: JSON_TYPE}),\n        'model.json');\n\n    if (modelArtifacts.weightData != null) {\n      init.body.append(\n          'model.weights.bin',\n          new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n          'model.weights.bin');\n    }\n\n    const response = await this.fetch(this.path, init);\n\n    if (response.ok) {\n      return {\n        modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n        responses: [response],\n      };\n    } else {\n      throw new Error(\n          `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n          `${response.status}.`);\n    }\n  }\n\n  /**\n   * Load model artifacts via HTTP request(s).\n   *\n   * See the documentation to `tf.io.http` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model artifacts (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n    if (!modelConfigRequest.ok) {\n      throw new Error(\n          `Request to ${this.path} failed with status code ` +\n          `${modelConfigRequest.status}. Please verify this URL points to ` +\n          `the model JSON of the model to load.`);\n    }\n    let modelConfig: ModelJSON;\n    try {\n      modelConfig = await modelConfigRequest.json();\n    } catch (e) {\n      let message = `Failed to parse model JSON of response from ${this.path}.`;\n      // TODO(nsthorat): Remove this after some time when we're comfortable that\n      // .pb files are mostly gone.\n      if (this.path.endsWith('.pb')) {\n        message += ' Your path contains a .pb file extension. ' +\n            'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n            'in favor of .json models. You can re-convert your Python ' +\n            'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n            'or you can convert your.pb models with the \\'pb2json\\'' +\n            'NPM script in the tensorflow/tfjs-converter repository.';\n      } else {\n        message += ' Please make sure the server is serving valid ' +\n            'JSON for this request.';\n      }\n      throw new Error(message);\n    }\n    const modelTopology = modelConfig.modelTopology;\n    const weightsManifest = modelConfig.weightsManifest;\n    const generatedBy = modelConfig.generatedBy;\n    const convertedBy = modelConfig.convertedBy;\n    const format = modelConfig.format;\n    const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n    // We do not allow both modelTopology and weightsManifest to be missing.\n    if (modelTopology == null && weightsManifest == null) {\n      throw new Error(\n          `The JSON from HTTP path ${this.path} contains neither model ` +\n          `topology or manifest for weights.`);\n    }\n\n    let weightSpecs: WeightsManifestEntry[];\n    let weightData: ArrayBuffer;\n    if (weightsManifest != null) {\n      const results = await this.loadWeights(weightsManifest);\n      [weightSpecs, weightData] = results;\n    }\n\n    const artifacts: ModelArtifacts = {\n      modelTopology,\n      weightSpecs,\n      weightData,\n      userDefinedMetadata,\n      generatedBy,\n      convertedBy,\n      format\n    };\n\n    const initializer = modelConfig.modelInitializer;\n    if (initializer) {\n      artifacts.modelInitializer = initializer;\n    }\n\n    return artifacts;\n  }\n\n  private async loadWeights(weightsManifest: WeightsManifestConfig):\n      Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n    const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n    const [prefix, suffix] = parseUrl(weightPath);\n    const pathPrefix = this.weightPathPrefix || prefix;\n\n    const weightSpecs = [];\n    for (const entry of weightsManifest) {\n      weightSpecs.push(...entry.weights);\n    }\n\n    const fetchURLs: string[] = [];\n    const urlPromises: Array<Promise<string>> = [];\n    for (const weightsGroup of weightsManifest) {\n      for (const path of weightsGroup.paths) {\n        if (this.weightUrlConverter != null) {\n          urlPromises.push(this.weightUrlConverter(path));\n        } else {\n          fetchURLs.push(pathPrefix + path + suffix);\n        }\n      }\n    }\n\n    if (this.weightUrlConverter) {\n      fetchURLs.push(...await Promise.all(urlPromises));\n    }\n\n    const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n      requestInit: this.requestInit,\n      fetchFunc: this.fetch,\n      onProgress: this.onProgress\n    });\n    return [weightSpecs, concatenateArrayBuffers(buffers)];\n  }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n  const lastSlash = url.lastIndexOf('/');\n  const lastSearchParam = url.lastIndexOf('?');\n  const prefix = url.substring(0, lastSlash);\n  const suffix =\n      lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n  return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n  return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n    (url: string, loadOptions?: LoadOptions) => {\n      if (typeof fetch === 'undefined' &&\n          (loadOptions == null || loadOptions.fetchFunc == null)) {\n        // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n        // an environment that is not the browser or node they have to setup a\n        // global fetch polyfill.\n        return null;\n      } else {\n        let isHTTP = true;\n        if (Array.isArray(url)) {\n          isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n        } else {\n          isHTTP = isHTTPScheme(url);\n        }\n        if (isHTTP) {\n          return http(url, loadOptions);\n        }\n      }\n      return null;\n    };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n *   - A JSON file consisting of `modelTopology` and `weightsManifest`.\n *   - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n *     'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n *   Can be an absolute HTTP path (e.g.,\n *   'http://localhost:8000/model-upload)') or a relative path (e.g.,\n *   './model-upload').\n * @param requestInit Request configurations to be used when sending\n *    HTTP request to server using `fetch`. It can contain fields such as\n *    `method`, `credentials`, `headers`, `mode`, etc. See\n *    https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n *    for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n *   following fields:\n *   - weightPathPrefix Optional, this specifies the path prefix for weight\n *     files, by default this is calculated from the path param.\n *   - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n *     the `fetch` from node-fetch can be used here.\n *   - onProgress Optional, progress callback function, fired periodically\n *     before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n  return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n    path: string, loadOptions?: LoadOptions): IOHandler {\n  return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n  constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n  async load(): Promise<ModelArtifacts> {\n    return this.modelArtifacts;\n  }\n}\n\nclass PassthroughSaver implements IOHandler {\n  constructor(\n      private readonly saveHandler:\n          (artifacts: ModelArtifacts) => Promise<SaveResult>) {}\n\n  async save(modelArtifacts: ModelArtifacts) {\n    return this.saveHandler(modelArtifacts);\n  }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n *     modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n *   the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n *   names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n *   concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n    modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n    weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n  if (arguments.length === 1) {\n    const isModelArtifacts =\n        (modelArtifacts as ModelArtifacts).modelTopology != null ||\n        (modelArtifacts as ModelArtifacts).weightSpecs != null;\n    if (isModelArtifacts) {\n      return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n    } else {\n      // Legacy support: with only modelTopology.\n      // TODO(cais): Remove this deprecated API.\n      console.warn(\n          'Please call tf.io.fromMemory() with only one argument. ' +\n          'The argument should be of type ModelArtifacts. ' +\n          'The multi-argument signature of tf.io.fromMemory() has been ' +\n          'deprecated and will be removed in a future release.');\n      return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n    }\n  } else {\n    // Legacy support.\n    // TODO(cais): Remove this deprecated API.\n    console.warn(\n        'Please call tf.io.fromMemory() with only one argument. ' +\n        'The argument should be of type ModelArtifacts. ' +\n        'The multi-argument signature of tf.io.fromMemory() has been ' +\n        'deprecated and will be removed in a future release.');\n    return new PassthroughLoader({\n      modelTopology: modelArtifacts as {},\n      weightSpecs,\n      weightData,\n      trainingConfig\n    });\n  }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n *   // ... do something with the artifacts ...\n *   return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n *     `SaveResult`.\n */\nexport function withSaveHandler(\n    saveHandler: (artifacts: ModelArtifacts) =>\n        Promise<SaveResult>): IOHandler {\n  return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n  browserFiles,\n  browserHTTPRequest,\n  concatenateArrayBuffers,\n  decodeWeights,\n  encodeWeights,\n  fromMemory,\n  getLoadHandlers,\n  getModelArtifactsInfoForJSON,\n  getSaveHandlers,\n  http,\n  IOHandler,\n  isHTTPScheme,\n  LoadHandler,\n  LoadOptions,\n  loadWeights,\n  ModelArtifacts,\n  ModelArtifactsInfo,\n  ModelJSON,\n  ModelStoreManager,\n  OnProgressCallback,\n  registerLoadRouter,\n  registerSaveRouter,\n  RequestDetails,\n  SaveConfig,\n  SaveHandler,\n  SaveResult,\n  WeightGroup,\n  weightsLoaderFactory,\n  WeightsManifestConfig,\n  WeightsManifestEntry,\n  withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  const $x = convertToTensor(x, 'x', 'reshape', null);\n\n  const inputs: ReshapeInputs = {x: $x};\n  const attrs: ReshapeAttrs = {shape};\n  const forward: ForwardFunc<\n      Tensor<R>> = (backend: KernelBackend, save: GradSaveFunc) => {\n    shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n    util.assert(\n        $x.size === util.sizeFromShape(shape),\n        () => 'new shape and old shape must have the same number of elements.');\n    save([$x]);\n    return backend.reshape($x, shape);\n  };\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n      attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print();  // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike, transposeA = false,\n    transposeB = false): T {\n  let $a = convertToTensor(a, 'a', 'matMul');\n  let $b = convertToTensor(b, 'b', 'matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n          `got ranks ${$a.rank} and ${$b.rank}.`);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n                           reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n                           reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([a3D, b3D]);\n\n    return backend.batchMatMul(\n        a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n  };\n\n  const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n  const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n      attrs as {} as NamedAttrMap);\n\n  return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n *     not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n    indices: Tensor|TensorLike, depth: number, onValue = 1,\n    offValue = 0): Tensor {\n  if (depth < 2) {\n    throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n  }\n  const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n  const outShape = [...$indices.shape, depth];\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$indices]);\n    return reshape(\n        backend.oneHot(\n            reshape($indices, [$indices.size]), depth, onValue, offValue),\n        outShape);\n  };\n\n  const inputs: OneHotInputs = {indices: $indices};\n  const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print();  // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_<T extends Tensor>(x: T|TensorLike, perm?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'transpose');\n\n  if (perm == null) {\n    perm = $x.shape.map((s, i) => i).reverse();\n  }\n  util.assert(\n      $x.rank === perm.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of perm ${perm}.`);\n  perm.forEach(axis => {\n    util.assert(\n        axis >= 0 && axis < $x.rank,\n        () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n            ` but got ${perm}`);\n  });\n\n  if ($x.rank <= 1) {\n    return $x.clone();\n  }\n\n  const inputs: TransposeInputs = {x: $x};\n  const attrs: TransposeAttrs = {perm};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n      null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * //  [0, 1, 1],\n * //  [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n *   for the classes. The shape is `[numExamples]`, where\n *   `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n *   0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n *   Its value must be larger than the largest element in `labels` and\n *   `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n *   row `r` and column `c` is the number of times examples of actual class\n *   `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n    labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n    numClasses: number): Tensor2D {\n  const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n  util.assert(\n      numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n      () => `If provided, numClasses must be a positive integer, ` +\n          `but got ${numClasses}`);\n  util.assert(\n      $labels.rank === 1,\n      () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n  util.assert(\n      $predictions.rank === 1,\n      () => `Expected the rank of predictions to be 1, ` +\n          `but got ${$predictions.rank}`);\n  util.assert(\n      $labels.shape[0] === $predictions.shape[0],\n      () => `Mismatch in the number of examples: ` +\n          `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n          `Labels and predictions should have the same number of elements.`);\n  util.assert(\n      numClasses > 0 && Number.isInteger(numClasses),\n      () => `numClasses is required to be a positive integer, but got ` +\n          `${numClasses}`);\n  // TODO(cais): In the future, if oneHot supports tensors inputs for\n  //   `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n  const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n  const oneHotPredictions =\n      oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n  const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n  return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided,  it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n    values: TensorLike3D, shape?: [number, number, number],\n    dtype?: DataType): Tensor3D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 3) {\n    throw new Error('tensor3d() requires shape to have three numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor3d() requires values to be number[][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor3d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement,\n    numChannels = 3): Tensor3D {\n  // Sanity checks.\n  if (numChannels > 4) {\n    throw new Error(\n        'Cannot construct Tensor with more than 4 channels from pixels.');\n  }\n  if (pixels == null) {\n    throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n  }\n  let isPixelData = false;\n  let isImageData = false;\n  let isVideo = false;\n  let isImage = false;\n  let isCanvasLike = false;\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    isPixelData = true;\n  } else if (\n      typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n    isImageData = true;\n  } else if (\n      typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement) {\n    isVideo = true;\n  } else if (\n      typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement) {\n    isImage = true;\n    // tslint:disable-next-line: no-any\n  } else if ((pixels as any).getContext != null) {\n    isCanvasLike = true;\n  } else {\n    throw new Error(\n        'pixels passed to tf.browser.fromPixels() must be either an ' +\n        `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n        `in browser, or OffscreenCanvas, ImageData in webworker` +\n        ` or {data: Uint32Array, width: number, height: number}, ` +\n        `but was ${(pixels as {}).constructor.name}`);\n  }\n  if (isVideo) {\n    const HAVE_CURRENT_DATA_READY_STATE = 2;\n    if (isVideo &&\n        (pixels as HTMLVideoElement).readyState <\n            HAVE_CURRENT_DATA_READY_STATE) {\n      throw new Error(\n          'The video element has not loaded data yet. Please wait for ' +\n          '`loadeddata` event on the <video> element.');\n    }\n  }\n  // If the current backend has 'FromPixels' registered, it has a more\n  // efficient way of handling pixel uploads, so we call that.\n  const kernel = getKernel(FromPixels, ENGINE.backendName);\n  if (kernel != null) {\n    const inputs: FromPixelsInputs = {pixels};\n    const attrs: FromPixelsAttrs = {numChannels};\n    return ENGINE.runKernel(\n               FromPixels, inputs as {} as NamedTensorMap,\n               attrs as {} as NamedAttrMap) as Tensor3D;\n  }\n\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n  let vals: Uint8ClampedArray|Uint8Array;\n\n  if (isCanvasLike) {\n    vals =\n        // tslint:disable-next-line:no-any\n        (pixels as any).getContext('2d').getImageData(0, 0, width, height).data;\n  } else if (isImageData || isPixelData) {\n    vals = (pixels as PixelData | ImageData).data;\n  } else if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement, 0, 0, width, height);\n    vals = fromPixels2DContext.getImageData(0, 0, width, height).data;\n  }\n  let values: Int32Array;\n  if (numChannels === 4) {\n    values = new Int32Array(vals);\n  } else {\n    const numPixels = width * height;\n    values = new Int32Array(numPixels * numChannels);\n    for (let i = 0; i < numPixels; i++) {\n      for (let channel = 0; channel < numChannels; ++channel) {\n        values[i * numChannels + channel] = vals[i * 4 + channel];\n      }\n    }\n  }\n  const outShape: [number, number, number] = [height, width, numChannels];\n  return tensor3d(values, outShape, 'int32');\n}\n\n/**\n * Draws a `tf.Tensor` of pixel values to a byte array or optionally a\n * canvas.\n *\n * When the dtype of the input is 'float32', we assume values in the range\n * [0-1]. Otherwise, when input is 'int32', we assume values in the range\n * [0-255].\n *\n * Returns a promise that resolves when the canvas has been drawn to.\n *\n * @param img A rank-2 or rank-3 tensor. If rank-2, draws grayscale. If\n *     rank-3, must have depth of 1, 3 or 4. When depth of 1, draws\n * grayscale. When depth of 3, we draw with the first three components of\n * the depth dimension corresponding to r, g, b and alpha = 1. When depth of\n * 4, all four components of the depth dimension correspond to r, g, b, a.\n * @param canvas The canvas to draw to.\n *\n * @doc {heading: 'Browser', namespace: 'browser'}\n */\nexport async function toPixels(\n    img: Tensor2D|Tensor3D|TensorLike,\n    canvas?: HTMLCanvasElement): Promise<Uint8ClampedArray> {\n  let $img = convertToTensor(img, 'img', 'toPixels');\n  if (!(img instanceof Tensor)) {\n    // Assume int32 if user passed a native array.\n    const originalImgTensor = $img;\n    $img = cast(originalImgTensor, 'int32');\n    originalImgTensor.dispose();\n  }\n  if ($img.rank !== 2 && $img.rank !== 3) {\n    throw new Error(\n        `toPixels only supports rank 2 or 3 tensors, got rank ${$img.rank}.`);\n  }\n  const [height, width] = $img.shape.slice(0, 2);\n  const depth = $img.rank === 2 ? 1 : $img.shape[2];\n\n  if (depth > 4 || depth === 2) {\n    throw new Error(\n        `toPixels only supports depth of size ` +\n        `1, 3 or 4 but got ${depth}`);\n  }\n\n  if ($img.dtype !== 'float32' && $img.dtype !== 'int32') {\n    throw new Error(\n        `Unsupported type for toPixels: ${$img.dtype}.` +\n        ` Please use float32 or int32 tensors.`);\n  }\n\n  const data = await $img.data();\n  const multiplier = $img.dtype === 'float32' ? 255 : 1;\n  const bytes = new Uint8ClampedArray(width * height * 4);\n\n  for (let i = 0; i < height * width; ++i) {\n    const rgba = [0, 0, 0, 255];\n\n    for (let d = 0; d < depth; d++) {\n      const value = data[i * depth + d];\n\n      if ($img.dtype === 'float32') {\n        if (value < 0 || value > 1) {\n          throw new Error(\n              `Tensor values for a float32 Tensor must be in the ` +\n              `range [0 - 1] but encountered ${value}.`);\n        }\n      } else if ($img.dtype === 'int32') {\n        if (value < 0 || value > 255) {\n          throw new Error(\n              `Tensor values for a int32 Tensor must be in the ` +\n              `range [0 - 255] but encountered ${value}.`);\n        }\n      }\n\n      if (depth === 1) {\n        rgba[0] = value * multiplier;\n        rgba[1] = value * multiplier;\n        rgba[2] = value * multiplier;\n      } else {\n        rgba[d] = value * multiplier;\n      }\n    }\n\n    const j = i * 4;\n    bytes[j + 0] = Math.round(rgba[0]);\n    bytes[j + 1] = Math.round(rgba[1]);\n    bytes[j + 2] = Math.round(rgba[2]);\n    bytes[j + 3] = Math.round(rgba[3]);\n  }\n\n  if (canvas != null) {\n    canvas.width = width;\n    canvas.height = height;\n    const ctx = canvas.getContext('2d');\n    const imageData = new ImageData(bytes, width, height);\n    ctx.putImageData(imageData, 0, 0);\n  }\n  if ($img !== img) {\n    $img.dispose();\n  }\n  return bytes;\n}\n\nexport const fromPixels = op({fromPixels_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\nimport {computeStrides} from '../util';\n\n/**\n * Validate gather nd inputs.\n *\n * @param tensor The tensor contains the source values.\n * @param indices The tensor contains the indices to slice the source.\n *\n * @returns [resultShape, numUpdates, sliceSize, strides]\n */\nexport function prepareAndValidate(\n    tensor: Tensor, indices: Tensor): [number[], number, number, number[]] {\n  if (tensor.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the input to be rank 1 or higher,' +\n        ` but the rank was ${tensor.rank}.`);\n  }\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(\n        'tf.gatherND() expects the indices to be int32 type,' +\n        ` but the dtype was ${indices.dtype}.`);\n  }\n  if (indices.shape[indices.rank - 1] > tensor.rank) {\n    throw new Error(\n        'index innermost dimension length must be <= tensor rank; saw: ' +\n        `${indices.shape[indices.rank - 1]} vs. ${tensor.rank}`);\n  }\n\n  if (tensor.size === 0) {\n    throw new Error(\n        'Requested more than 0 entries, but input is empty.' +\n        ` Input shape: ${tensor.shape}.`);\n  }\n\n  const indicesShape = indices.shape;\n  const sliceRank = indicesShape[indicesShape.length - 1];\n\n  // The result shape is\n  //   indices.shape[:-1] + params.shape[indices.shape[-1]:]\n  let nResult = 1;\n  for (let i = 0; i < indicesShape.length - 1; ++i) {\n    nResult *= indicesShape[i];\n  }\n\n  const inputShape = tensor.shape;\n\n  const resultShape = indicesShape.slice();\n  resultShape.pop();\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < tensor.rank; ++i) {\n    sliceSize *= inputShape[i];\n    resultShape.push(inputShape[i]);\n  }\n\n  const strides =\n      [...computeStrides(tensor.shape).map(stride => stride / sliceSize),\n       1].slice(0, sliceRank);\n\n  return [resultShape, nResult, sliceSize, strides];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {computeStrides, sizeFromShape} from '../util';\n\n/**\n * Check whether updates.shape = indices.shape[:batchDim] +\n * shape[sliceDim:]\n *\n * @param x The input tensor.\n */\nexport function validateUpdateShape(\n    shape: number[], indices: Tensor, updates: Tensor) {\n  const sliceDim = (indices.rank > 1) ? indices.shape[indices.rank - 1] : 1;\n  const batchDim = (indices.rank > 1) ? indices.rank - 1 : 1;\n\n  const shapeError = 'Must have updates.shape = indices.shape[:batchDim] + ' +\n      `shape[sliceDim:], got updates.shape: ${updates.shape}` +\n      `, indices.shape: ${indices.shape}, shape: ${shape}` +\n      `, sliceDim: ${sliceDim}, and batchDim: ${batchDim}.`;\n\n  if (updates.rank < batchDim) {\n    throw new Error(shapeError + ` update.rank < ${batchDim}. `);\n  }\n  if (shape.length < sliceDim + (updates.rank - batchDim)) {\n    throw new Error(\n        shapeError +\n        ` Output shape length < ${sliceDim + (updates.rank - batchDim)}`);\n  }\n  if (updates.rank !== batchDim + shape.length - sliceDim) {\n    throw new Error(\n        shapeError + ` update.rank != ${batchDim + shape.length - sliceDim}`);\n  }\n  for (let d = 0; d < batchDim; ++d) {\n    if (updates.shape[d] !== indices.shape[d]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d}] (${updates.shape[d]}) != indices.shape[${d}] (${\n              indices.shape[d]}).`);\n    }\n  }\n  for (let d = 0; d < updates.rank - batchDim; ++d) {\n    if (updates.shape[d + batchDim] !== shape[d + sliceDim]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d + batchDim}] (${\n              updates.shape[d + batchDim]}) != shape[${d + batchDim}] (${\n              shape[d + batchDim]})`);\n    }\n  }\n}\n\nexport interface ScatterShapeInfo {\n  sliceRank: number;\n  numUpdates: number;\n  sliceSize: number;\n  strides: number[];\n  outputSize: number;\n}\n/**\n * Validate scatter nd inputs.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n */\nexport function validateInput(\n    updates: Tensor, indices: Tensor, shape: number[]) {\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (updates.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the updates to be rank 1 or higher,' +\n        ` but the rank was ${updates.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${\n        indices.dtype}`);\n  }\n  if (shape.length < 1) {\n    throw new Error(\n        `Output rank must be greater or equal to 1, but got shape: ${shape}`);\n  }\n\n  if (shape.length === 0) {\n    if (indices.size === 0) {\n      throw new Error(`Indices specified for empty output. indices shape: ${\n          indices.shape}`);\n    }\n    if (updates.size === 0) {\n      throw new Error(`Updates specified for empty output. updates shape: ${\n          updates.shape}`);\n    }\n  }\n\n  validateUpdateShape(shape, indices, updates);\n}\n\n/**\n * Calculate the shape information for the output.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n *\n * @returns ScatterShapeInfo\n */\nexport function calculateShapes(\n    updates: TensorInfo, indices: TensorInfo,\n    shape: number[]): ScatterShapeInfo {\n  // Calculate the number of dimensions in indices\n  const indicesRank = indices.shape.length;\n  const sliceRank = (indicesRank > 1) ? indices.shape[indicesRank - 1] : 1;\n\n  // Calculate the number of elements that make up each slice of our updated\n  // tensor. This allows us to work with flattened tensors and copy over whole\n  // slices at a time.\n  const totalNd = shape.length;\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < totalNd; ++i) {\n    sliceSize *= shape[i];\n  }\n\n  const safeSliceDim = (sliceRank < 1) ? 1 : sliceRank;\n  const numUpdates = sizeFromShape(indices.shape) / safeSliceDim;\n\n  const strides = [...computeStrides(shape.slice(0, sliceRank)), 1];\n  const outputSize = sizeFromShape(shape);\n  return {sliceRank, numUpdates, sliceSize, strides, outputSize};\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '../kernel_registry';\nimport * as util from '../util';\n\nexport function assertParamsValid(\n    input: TensorInfo, begin: number[], size: number[]): void {\n  const inputRank = input.shape.length;\n  util.assert(\n      inputRank === begin.length,\n      () => `Error in slice${inputRank}D: Length of begin ${begin} must ` +\n          `match the rank of the array (${inputRank}).`);\n  util.assert(\n      inputRank === size.length,\n      () => `Error in slice${inputRank}D: Length of size ${size} must ` +\n          `match the rank of the array (${inputRank}).`);\n\n  for (let i = 0; i < inputRank; ++i) {\n    util.assert(\n        begin[i] + size[i] <= input.shape[i],\n        () => `Error in slice${inputRank}D: begin[${i}] + size[${i}] ` +\n            `(${begin[i] + size[i]}) would overflow input.shape[${i}] (${\n                  input.shape[i]})`);\n  }\n}\n\n/** Converts a binary mask to an array of axes. Used in stridedSlice(). */\nexport function maskToAxes(mask: number): number[] {\n  const axes = [];\n  let axis = 0;\n  while (mask > 0) {\n    if (mask & 1) {\n      axes.push(axis);\n    }\n    mask /= 2;\n    axis++;\n  }\n  return axes;\n}\n\n/** Computes the output shape given the strided slice params. */\nexport function computeOutShape(\n    begin: number[], end: number[], strides: number[]): number[] {\n  const size = [];\n  for (let axis = 0; axis < begin.length; axis++) {\n    size[axis] = Math.ceil((end[axis] - begin[axis]) / strides[axis]);\n  }\n  return size;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stride value. Otherwise, insert.\nexport function stridesWithElidedDims(\n    strides: number[], ellipsisInsertionIndex: number, numElidedAxes: number,\n    inputShape: number[]): number[] {\n  const newStrides = [...strides];\n  for (let i = newStrides.length; i < inputShape.length; i++) {\n    newStrides.push(1);\n  }\n  for (let i = 0; i < numElidedAxes; i++) {\n    if (i === 0) {\n      newStrides[ellipsisInsertionIndex] = 1;\n    } else {\n      newStrides.splice(\n          ellipsisInsertionIndex, 0 /* num elements to delete */,\n          1 /* element to add */);\n      newStrides.pop();\n    }\n  }\n  return newStrides;\n}\n\nfunction unnormalizeAxis(\n    ellipsisInsertionIndex: number, numElidedAxes: number,\n    normalizedAxis: number): number {\n  if (normalizedAxis <= ellipsisInsertionIndex) {\n    return normalizedAxis;\n  }\n\n  return normalizedAxis - (numElidedAxes - 1);\n}\n\nfunction getElidedAxes(numElidedAxes: number, ellipsisInsertionIndex: number) {\n  const elidedAxes = [];\n  for (let i = 0; i < numElidedAxes; i++) {\n    elidedAxes.push(ellipsisInsertionIndex + i);\n  }\n  return elidedAxes;\n}\n\n// Normalize the start, end and strides.\nexport function getNormalizedAxes(\n    inputShape: number[], ellipsisAxes: number[], numInterpolatedAxes: number,\n    begin: number[], end: number[], strides: number[], beginMask: number,\n    endMask: number,\n    ellipsisMask: number): {begin: number[], end: number[], strides: number[]} {\n  const inputRank = inputShape.length;\n  let normalizedBegin = new Array(inputRank),\n      normalizedEnd = new Array(inputRank),\n      normalizedStrides = new Array(inputRank);\n  if (ellipsisAxes.length && numInterpolatedAxes > 0) {\n    const fullIndex = ellipsisAxes[0];\n\n    // The ellipsis applies to the masked index as well as any dimensions\n    // that are interpolated.\n    const numElidedAxes = numInterpolatedAxes + 1;\n    normalizedBegin = startIndicesWithElidedDims(\n        beginMask, fullIndex, numElidedAxes, begin, inputShape);\n    normalizedEnd = stopIndicesWithElidedDims(\n        endMask, fullIndex, numElidedAxes, end, inputShape);\n    normalizedStrides =\n        stridesWithElidedDims(strides, fullIndex, numElidedAxes, inputShape);\n  } else {\n    for (let axis = 0; axis < inputRank; axis++) {\n      normalizedBegin[axis] = startForAxis(\n          beginMask, begin, strides, inputShape, axis, ellipsisMask);\n      normalizedEnd[axis] =\n          stopForAxis(endMask, end, strides, inputShape, axis, ellipsisMask);\n      normalizedStrides[axis] = stridesForAxis(strides, axis, ellipsisMask);\n    }\n  }\n\n  return {\n    begin: normalizedBegin,\n    end: normalizedEnd,\n    strides: normalizedStrides\n  };\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current start value. Otherwise, insert.\nexport function startIndicesWithElidedDims(\n    beginMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalBegin: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = 0;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalBegin[originalAxis];\n      if (beginMask & 1 << originalAxis) {\n        originalValue = 0;\n      }\n\n      newIndices[axis] = originalValue;\n    }\n  }\n  return newIndices;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stop value. Otherwise, insert.\nexport function stopIndicesWithElidedDims(\n    endMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalEnd: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = Number.MAX_SAFE_INTEGER;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalEnd[originalAxis];\n      if (endMask & 1 << originalAxis) {\n        originalValue = Number.MAX_SAFE_INTEGER;\n      }\n      newIndices[axis] = originalValue;\n    }\n  }\n\n  for (let i = 0; i < newIndices.length; i++) {\n    // Handle negative indices\n    const axisSize = inputShape[i];\n    if (newIndices[i] < 0) {\n      newIndices[i] += axisSize;\n    }\n    newIndices[i] = util.clamp(0, newIndices[i], inputShape[i]);\n  }\n  return newIndices;\n}\n\nexport function stridesForAxis(\n    strides: number[], axis: number, ellipsisMask: number): number {\n  let stride = strides[axis];\n  if (ellipsisMask & (1 << axis) || stride == null) {\n    stride = 1;\n  }\n\n  return stride;\n}\n\nexport function startForAxis(\n    beginMask: number, startIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let start = startIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or the begin index is not set\n  // for the axis.\n  if (beginMask & 1 << axis || ellipsisMask & 1 << axis || start == null) {\n    if (stride > 0) {\n      // Forward iteration - use the first element. These values will get\n      // clamped below (Note: We could have set them to 0 and axis_size-1, but\n      // use lowest() and max() to maintain symmetry with StopForAxis())\n      start = Number.MIN_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the last element.\n      start = Number.MAX_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (start < 0) {\n    start += axisSize;\n  }\n\n  // Clamping\n  start = util.clamp(0, start, axisSize - 1);\n\n  return start;\n}\n\nexport function stopForAxis(\n    endMask: number, stopIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let stop = stopIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or if the stop index is not\n  // set for this axis.\n  if (endMask & (1 << axis) || ellipsisMask & (1 << axis) || stop == null) {\n    if (stride > 0) {\n      // Forward iteration - use the last element. These values will get\n      // clamped below\n      stop = Number.MAX_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the first element.\n      stop = Number.MIN_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (stop < 0) {\n    stop += axisSize;\n  }\n\n  // Clamping\n  // Because the end index points one past the last element, we need slightly\n  // different clamping ranges depending on the direction.\n  if (stride > 0) {\n    // Forward iteration\n    stop = util.clamp(0, stop, axisSize);\n  } else {\n    // Backward iteration\n    stop = util.clamp(-1, stop, axisSize - 1);\n  }\n\n  return stop;\n}\n\n/**\n * Returns true if the slice occupies a continous set of elements in the\n * 'flat' space.\n */\nexport function isSliceContinous(\n    shape: number[], begin: number[], size: number[]) {\n  // Index of the first axis that has size > 1.\n  let firstNonOneAxis = size.length;\n  for (let i = 0; i < size.length; i++) {\n    if (size[i] > 1) {\n      firstNonOneAxis = i;\n      break;\n    }\n  }\n\n  for (let i = firstNonOneAxis + 1; i < size.length; i++) {\n    if (begin[i] > 0 || size[i] !== shape[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function computeFlatOffset(begin: number[], strides: number[]): number {\n  let flatOffset = begin.length > 0 ? begin[begin.length - 1] : 1;\n  for (let i = 0; i < begin.length - 1; i++) {\n    flatOffset += begin[i] * strides[i];\n  }\n  return flatOffset;\n}\n\nexport function parseSliceParams(\n    x: TensorInfo, begin: number|number[], size?: number|number[]) {\n  // The following logic allows for more ergonomic calls.\n  let begin_: number[];\n  const xRank = x.shape.length;\n  if (typeof begin === 'number') {\n    begin_ = [begin, ...new Array(xRank - 1).fill(0)];\n  } else if (begin.length < xRank) {\n    begin_ = begin.concat(new Array(xRank - begin.length).fill(0));\n  } else {\n    begin_ = begin.slice();\n  }\n  begin_.forEach(d => {\n    util.assert(\n        d !== -1, () => 'slice() does not support negative begin indexing.');\n  });\n  let size_: number[];\n  if (size == null) {\n    size_ = new Array(xRank).fill(-1);\n  } else if (typeof size === 'number') {\n    size_ = [size, ...new Array(xRank - 1).fill(-1)];\n  } else if (size.length < xRank) {\n    size_ = size.concat(new Array(xRank - size.length).fill(-1));\n  } else {\n    size_ = size;\n  }\n  size_ = size_.map((d, i) => {\n    if (d >= 0) {\n      return d;\n    } else {\n      util.assert(\n          d === -1,\n          () => `Negative size values should be exactly -1 but got ` +\n              `${d} for the slice() size at index ${i}.`);\n      return x.shape[i] - begin_[i];\n    }\n  });\n  return [begin_, size_];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from './util';\n\n/**\n * Types to support JSON-esque data structures internally.\n *\n * Internally ConfigDict's use camelCase keys and values where the\n * values are class names to be instantiated.  On the python side, these\n * will be snake_case.  Internally we allow Enums into the values for better\n * type safety, but these need to be converted to raw primitives (usually\n * strings) for round-tripping with python.\n *\n * toConfig returns the TS-friendly representation. model.toJSON() returns\n * the pythonic version as that's the portable format.  If you need to\n * python-ify a non-model level toConfig output, you'll need to use a\n * convertTsToPythonic from serialization_utils in -Layers.\n *\n */\nexport declare type ConfigDictValue =\n    boolean | number | string | null | ConfigDictArray | ConfigDict;\nexport declare interface ConfigDict {\n  [key: string]: ConfigDictValue;\n}\nexport declare interface ConfigDictArray extends Array<ConfigDictValue> {}\n\n/**\n * Type to represent the class-type of Serializable objects.\n *\n * Ie the class prototype with access to the constructor and any\n * static members/methods. Instance methods are not listed here.\n *\n * Source for this idea: https://stackoverflow.com/a/43607255\n */\nexport declare type SerializableConstructor<T extends Serializable> = {\n  // tslint:disable-next-line:no-any\n  new (...args: any[]): T; className: string; fromConfig: FromConfigMethod<T>;\n};\nexport declare type FromConfigMethod<T extends Serializable> =\n    (cls: SerializableConstructor<T>, config: ConfigDict) => T;\n\n/**\n * Serializable defines the serialization contract.\n *\n * TFJS requires serializable classes to return their className when asked\n * to avoid issues with minification.\n */\nexport abstract class Serializable {\n  /**\n   * Return the class name for this class to use in serialization contexts.\n   *\n   * Generally speaking this will be the same thing that constructor.name\n   * would have returned.  However, the class name needs to be robust\n   * against minification for serialization/deserialization to work properly.\n   *\n   * There's also places such as initializers.VarianceScaling, where\n   * implementation details between different languages led to different\n   * class hierarchies and a non-leaf node is used for serialization purposes.\n   */\n  getClassName(): string {\n    return (this.constructor as SerializableConstructor<Serializable>)\n        .className;\n  }\n\n  /**\n   * Return all the non-weight state needed to serialize this object.\n   */\n  abstract getConfig(): ConfigDict;\n\n  /**\n   * Creates an instance of T from a ConfigDict.\n   *\n   * This works for most descendants of serializable.  A few need to\n   * provide special handling.\n   * @param cls A Constructor for the class to instantiate.\n   * @param config The Configuration for the object.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config);\n  }\n}\n\n/**\n * Maps string keys to class constructors.\n *\n * Used during (de)serialization from the cross-language JSON format, which\n * requires the class name in the serialization format matches the class\n * names as used in Python, should it exist.\n */\nexport class SerializationMap {\n  private static instance: SerializationMap;\n  classNameMap: {\n    [className: string]:\n        [SerializableConstructor<Serializable>, FromConfigMethod<Serializable>]\n  };\n\n  private constructor() {\n    this.classNameMap = {};\n  }\n\n  /**\n   * Returns the singleton instance of the map.\n   */\n  static getMap(): SerializationMap {\n    if (SerializationMap.instance == null) {\n      SerializationMap.instance = new SerializationMap();\n    }\n    return SerializationMap.instance;\n  }\n\n  /**\n   * Registers the class as serializable.\n   */\n  static register<T extends Serializable>(cls: SerializableConstructor<T>) {\n    SerializationMap.getMap().classNameMap[cls.className] =\n        [cls, cls.fromConfig];\n  }\n}\n\n/**\n * Register a class with the serialization map of TensorFlow.js.\n *\n * This is often used for registering custom Layers, so they can be\n * serialized and deserialized.\n *\n * Example:\n *\n * ```js\n * class MyCustomLayer extends tf.layers.Layer {\n *   static className = 'MyCustomLayer';\n *\n *   constructor(config) {\n *     super(config);\n *   }\n * }\n * tf.serialization.registerClass(MyCustomLayer);\n * ```\n *\n * @param cls The class to be registered. It must have a public static member\n *   called `className` defined and the value must be a non-empty string.\n *\n * @doc {heading: 'Models', subheading: 'Serialization', ignoreCI: true}\n */\nexport function registerClass<T extends Serializable>(\n    cls: SerializableConstructor<T>) {\n  assert(\n      cls.className != null,\n      () => `Class being registered does not have the static className ` +\n          `property defined.`);\n  assert(\n      typeof cls.className === 'string',\n      () => `className is required to be a string, but got type ` +\n          typeof cls.className);\n  assert(\n      cls.className.length > 0,\n      () => `Class being registered has an empty-string as its className, ` +\n          `which is disallowed.`);\n\n  SerializationMap.register(cls);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {inferShape} from './tensor_util_env';\nimport {RecursiveArray, TensorLike, TypedArray} from './types';\nimport {arraysEqual, flatten, isString, isTypedArray} from './util';\n\nconst TEST_EPSILON_FLOAT32 = 1e-3;\nexport const TEST_EPSILON_FLOAT16 = 1e-1;\n\nexport function expectArraysClose(\n    actual: TypedArray|number|RecursiveArray<number>,\n    expected: TypedArray|number|RecursiveArray<number>, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, epsilon));\n}\n\nexport function testEpsilon() {\n  return ENGINE.backend.floatPrecision() === 32 ? TEST_EPSILON_FLOAT32 :\n                                                  TEST_EPSILON_FLOAT16;\n}\n\nfunction expectArraysPredicate(\n    actual: TensorLike, expected: TensorLike,\n    predicate: (a: {}, b: {}) => boolean) {\n  let checkClassType = true;\n  if (isTypedArray(actual) || isTypedArray(expected)) {\n    checkClassType = false;\n  }\n  if (isTypedArray(actual) && isTypedArray(expected)) {\n    checkClassType = true;\n  }\n  if (checkClassType) {\n    const aType = actual.constructor.name;\n    const bType = expected.constructor.name;\n\n    if (aType !== bType) {\n      throw new Error(\n          `Arrays are of different type. Actual: ${aType}. ` +\n          `Expected: ${bType}`);\n    }\n  }\n\n  if (Array.isArray(actual) && Array.isArray(expected)) {\n    const actualShape = inferShape(actual);\n    const expectedShape = inferShape(expected);\n    if (!arraysEqual(actualShape, expectedShape)) {\n      throw new Error(\n          `Arrays have different shapes. ` +\n          `Actual: [${actualShape}]. Expected: [${expectedShape}]`);\n    }\n  }\n\n  const actualFlat =\n      isTypedArray(actual) ? actual : flatten(actual as RecursiveArray<number>);\n  const expectedFlat = isTypedArray(expected) ?\n      expected :\n      flatten(expected as RecursiveArray<number>);\n\n  if (actualFlat.length !== expectedFlat.length) {\n    throw new Error(\n        `Arrays have different lengths actual: ${actualFlat.length} vs ` +\n        `expected: ${expectedFlat.length}.\\n` +\n        `Actual:   ${actualFlat}.\\n` +\n        `Expected: ${expectedFlat}.`);\n  }\n  for (let i = 0; i < expectedFlat.length; ++i) {\n    const a = actualFlat[i];\n    const e = expectedFlat[i];\n\n    if (!predicate(a, e)) {\n      throw new Error(\n          `Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${e}.\\n` +\n          `Actual:   ${actualFlat}.\\n` +\n          `Expected: ${expectedFlat}.`);\n    }\n  }\n}\n\nexport interface DoneFn {\n  (): void;\n  fail: (message?: Error|string) => void;\n}\n\nexport function expectPromiseToFail(fn: () => Promise<{}>, done: DoneFn): void {\n  fn().then(() => done.fail(), () => done());\n}\n\nexport function expectArraysEqual(actual: TensorLike, expected: TensorLike) {\n  const exp = typeof expected === 'string' || typeof expected === 'number' ||\n          typeof expected === 'boolean' ?\n      [expected] as number[] :\n      expected as number[];\n  if (isString(actual) || isString((actual as string[])[0]) ||\n      isString(expected) || isString((expected as string[])[0])) {\n    // tslint:disable-next-line: triple-equals\n    return expectArraysPredicate(actual, exp, (a, b) => a == b);\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, 0));\n}\n\nexport function expectNumbersClose(a: number, e: number, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  if (!areClose(a, e, epsilon)) {\n    throw new Error(`Numbers differ: actual === ${a}, expected === ${e}`);\n  }\n}\n\nfunction areClose(a: number, e: number, epsilon: number): boolean {\n  if (!isFinite(a) && !isFinite(e)) {\n    return true;\n  }\n  if (isNaN(a) || isNaN(e) || Math.abs(a - e) > epsilon) {\n    return false;\n  }\n  return true;\n}\n\nexport function expectValuesInRange(\n    actual: TypedArray|number[], low: number, high: number) {\n  for (let i = 0; i < actual.length; i++) {\n    if (actual[i] < low || actual[i] > high) {\n      throw new Error(\n          `Value out of range:${actual[i]} low: ${low}, high: ${high}`);\n    }\n  }\n}\n\nexport function expectArrayBuffersEqual(\n    actual: ArrayBuffer, expected: ArrayBuffer) {\n  // Safari & Jasmine don't like comparing ArrayBuffers directly. Wrapping in\n  // a Float32Array solves this issue.\n  expect(new Float32Array(actual)).toEqual(new Float32Array(expected));\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from './backends/backend';\nimport {ENGINE, Engine, MemoryInfo, ProfileInfo, ScopeFn, TimingInfo} from './engine';\nimport {env} from './environment';\n\nimport {Platform} from './platforms/platform';\nimport {setDeprecationWarningFn, Tensor} from './tensor';\nimport {TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\n\n/**\n * Enables production mode which disables correctness checks in favor of\n * performance.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableProdMode(): void {\n  env().set('PROD', true);\n}\n\n/**\n * Enables debug mode which will log information about all executed kernels:\n * the elapsed time of the kernel execution, as well as the rank, shape, and\n * size of the output tensor.\n *\n * Debug mode will significantly slow down your application as it will\n * download the result of every operation to the CPU. This should not be used in\n * production. Debug mode does not affect the timing information of the kernel\n * execution as we do not measure download time in the kernel execution time.\n *\n * See also: `tf.profile`, `tf.memory`.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableDebugMode(): void {\n  env().set('DEBUG', true);\n}\n\n/** Globally disables deprecation warnings */\nexport function disableDeprecationWarnings(): void {\n  env().set('DEPRECATION_WARNINGS_ENABLED', false);\n  console.warn(`TensorFlow.js deprecation warnings have been disabled.`);\n}\n\n/** Warn users about deprecated functionality. */\nexport function deprecationWarn(msg: string) {\n  if (env().getBool('DEPRECATION_WARNINGS_ENABLED')) {\n    console.warn(\n        msg + ' You can disable deprecation warnings with ' +\n        'tf.disableDeprecationWarnings().');\n  }\n}\nsetDeprecationWarningFn(deprecationWarn);\n\n/**\n * Dispose all variables kept in backend engine.\n *\n * @doc {heading: 'Environment'}\n */\nexport function disposeVariables(): void {\n  ENGINE.disposeVariables();\n}\n\n/**\n * It returns the global engine that keeps track of all tensors and backends.\n *\n * @doc {heading: 'Environment'}\n */\nexport function engine(): Engine {\n  return ENGINE;\n}\n\n/**\n * Returns memory info at the current time in the program. The result is an\n * object with the following properties:\n *\n * - `numBytes`: Number of bytes allocated (undisposed) at this time.\n * - `numTensors`: Number of unique tensors allocated.\n * - `numDataBuffers`: Number of unique data buffers allocated\n *   (undisposed) at this time, which is \u2264 the number of tensors\n *   (e.g. `a.reshape(newShape)` makes a new Tensor that shares the same\n *   data buffer with `a`).\n * - `unreliable`: True if the memory usage is unreliable. See `reasons` when\n *    `unreliable` is true.\n * - `reasons`: `string[]`, reasons why the memory is unreliable, present if\n *    `unreliable` is true.\n *\n * WebGL Properties:\n * - `numBytesInGPU`: Number of bytes allocated (undisposed) in the GPU only at\n *     this time.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function memory(): MemoryInfo {\n  return ENGINE.memory();\n}\n\n/**\n * Executes the provided function `f()` and returns a promise that resolves\n * with information about the function's memory use:\n * - `newBytes`: the number of new bytes allocated\n * - `newTensors`: the number of new tensors created\n * - `peakBytes`: the peak number of bytes allocated\n * - `kernels`: an array of objects for each kernel involved that reports\n * their input and output shapes, number of bytes used, and number of new\n * tensors created.\n *\n * ```js\n * const profile = await tf.profile(() => {\n *   const x = tf.tensor1d([1, 2, 3]);\n *   let x2 = x.square();\n *   x2.dispose();\n *   x2 = x.square();\n *   x2.dispose();\n *   return x;\n * });\n *\n * console.log(`newBytes: ${profile.newBytes}`);\n * console.log(`newTensors: ${profile.newTensors}`);\n * console.log(`byte usage over all kernels: ${profile.kernels.map(k =>\n * k.totalBytesSnapshot)}`);\n * ```\n *\n *\n * @doc {heading: 'Performance', subheading: 'Profile'}\n */\nexport function profile(f: () => (TensorContainer | Promise<TensorContainer>)):\n    Promise<ProfileInfo> {\n  return ENGINE.profile(f);\n}\n\n/**\n * Executes the provided function `fn` and after it is executed, cleans up all\n * intermediate tensors allocated by `fn` except those returned by `fn`.\n * `fn` must not return a Promise (async functions not allowed). The returned\n * result can be a complex object.\n *\n * Using this method helps avoid memory leaks. In general, wrap calls to\n * operations in `tf.tidy` for automatic memory cleanup.\n *\n * NOTE: Variables do *not* get cleaned up when inside a tidy(). If you want to\n * dispose variables, please use `tf.disposeVariables` or call dispose()\n * directly on variables.\n *\n * ```js\n * // y = 2 ^ 2 + 1\n * const y = tf.tidy(() => {\n *   // a, b, and one will be cleaned up when the tidy ends.\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *   const b = a.square();\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * y.print();\n * ```\n *\n * @param nameOrFn The name of the closure, or the function to execute.\n *     If a name is provided, the 2nd argument should be the function.\n *     If debug mode is on, the timing and the memory usage of the function\n *     will be tracked and displayed on the console using the provided name.\n * @param fn The function to execute.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function tidy<T extends TensorContainer>(\n    nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>): T {\n  return ENGINE.tidy(nameOrFn, fn);\n}\n\n/**\n * Disposes any `tf.Tensor`s found within the provided object.\n *\n * @param container an object that may be a `tf.Tensor` or may directly\n *     contain `tf.Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. If\n *     the object is not a `tf.Tensor` or does not contain `Tensors`, nothing\n *     happens. In general it is safe to pass any object here, except that\n *     `Promise`s are not supported.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function dispose(container: TensorContainer) {\n  const tensors = getTensorsInContainer(container);\n  tensors.forEach(tensor => tensor.dispose());\n}\n\n/**\n * Keeps a `tf.Tensor` generated inside a `tf.tidy` from being disposed\n * automatically.\n *\n * ```js\n * let b;\n * const y = tf.tidy(() => {\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *\n *   // b will not be cleaned up by the tidy. a and one will be cleaned up\n *   // when the tidy ends.\n *   b = tf.keep(a.square());\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * console.log('y:');\n * y.print();\n * console.log('b:');\n * b.print();\n * ```\n *\n * @param result The tensor to keep from being disposed.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function keep<T extends Tensor>(result: T): T {\n  return ENGINE.keep(result);\n}\n\n/**\n * Executes `f()` and returns a promise that resolves with timing\n * information.\n *\n * The result is an object with the following properties:\n *\n * - `wallMs`: Wall execution time.\n * - `kernelMs`: Kernel execution time, ignoring data transfer. If using the\n * WebGL backend and the query timer extension is not available, this will\n * return an error object.\n * - On `WebGL` The following additional properties exist:\n *   - `uploadWaitMs`: CPU blocking time on texture uploads.\n *   - `downloadWaitMs`: CPU blocking time on texture downloads (readPixels).\n *\n * ```js\n * const x = tf.randomNormal([20, 20]);\n * const time = await tf.time(() => x.matMul(x));\n *\n * console.log(`kernelMs: ${time.kernelMs}, wallTimeMs: ${time.wallMs}`);\n * ```\n *\n * @param f The function to execute and time.\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nexport function time(f: () => void): Promise<TimingInfo> {\n  return ENGINE.time(f);\n}\n\n/**\n * Sets the backend (cpu, webgl, wasm, etc) responsible for creating tensors and\n * executing operations on those tensors. Returns a promise that resolves\n * to a boolean if the backend initialization was successful.\n *\n * Note this disposes the current backend, if any, as well as any tensors\n * associated with it. A new backend is initialized, even if it is of the\n * same type as the previous one.\n *\n * @param backendName The name of the backend. Currently supports\n *     `'webgl'|'cpu'` in the browser, `'tensorflow'` under node.js\n *     (requires tfjs-node), and `'wasm'` (requires tfjs-backend-wasm).\n *\n * @doc {heading: 'Backends'}\n */\nexport function setBackend(backendName: string): Promise<boolean> {\n  return ENGINE.setBackend(backendName);\n}\n\n/**\n * Returns a promise that resolves when the currently selected backend (or the\n * highest priority one) has initialized. Await this promise when you are using\n * a backend that has async initialization.\n *\n * @doc {heading: 'Backends'}\n */\nexport function ready(): Promise<void> {\n  return ENGINE.ready();\n}\n\n/**\n * Returns the current backend name (cpu, webgl, etc). The backend is\n * responsible for creating tensors and executing operations on those tensors.\n *\n * @doc {heading: 'Backends'}\n */\nexport function getBackend(): string {\n  return ENGINE.backendName;\n}\n\n/**\n * Removes a backend and the registered factory.\n *\n * @doc {heading: 'Backends'}\n */\nexport function removeBackend(name: string): void {\n  ENGINE.removeBackend(name);\n}\n\n/**\n * Finds the backend registered under the provided name. Returns null if the\n * name is not in the registry, or the registration hasn't finished yet.\n */\nexport function findBackend(name: string): KernelBackend {\n  return ENGINE.findBackend(name);\n}\n\n/**\n * Finds the backend factory registered under the provided name. Returns a\n * function that produces a new backend when called. Returns null if the name\n * is not in the registry.\n */\nexport function findBackendFactory(name: string): () =>\n    KernelBackend | Promise<KernelBackend> {\n  return ENGINE.findBackendFactory(name);\n}\n\n/**\n * Registers a global backend. The registration should happen when importing\n * a module file (e.g. when importing `backend_webgl.ts`), and is used for\n * modular builds (e.g. custom tfjs bundle with only webgl support).\n *\n * @param factory The backend factory function. When called, it should\n * return a backend instance, or a promise of an instance.\n * @param priority The priority of the backend (higher = more important).\n *     In case multiple backends are registered, the priority is used to find\n *     the best backend. Defaults to 1.\n * @return False if there is already a registered backend under this name, true\n *     if not.\n *\n * @doc {heading: 'Backends'}\n */\nexport function registerBackend(\n    name: string, factory: () => KernelBackend | Promise<KernelBackend>,\n    priority = 1): boolean {\n  return ENGINE.registerBackend(name, factory, priority);\n}\n\n/**\n * Gets the current backend. If no backends have been initialized, this will\n * attempt to initialize the best backend. Will throw an error if the highest\n * priority backend has async initialization, in which case, you should call\n * 'await tf.ready()' before running other code.\n *\n * @doc {heading: 'Backends'}\n */\nexport function backend(): KernelBackend {\n  return ENGINE.backend;\n}\n\n/**\n * Sets the global platform.\n *\n * @param platformName The name of this platform.\n * @param platform A platform implementation.\n */\nexport function setPlatform(platformName: string, platform: Platform) {\n  env().setPlatform(platformName, platform);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Add, AddInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n *\n * ```js\n * // Broadcast add a with b.\n * const a = tf.scalar(5);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n * @param a The first `tf.Tensor` to add.\n * @param b The second `tf.Tensor` to add. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction add_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'add');\n  let $b = convertToTensor(b, 'b', 'add');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.add($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: AddInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Add) as T;\n}\n\nexport const add = op({add_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FloorDiv, FloorDivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n * The result is rounded with floor function.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.floorDiv(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.floorDiv(b).print();  // or tf.floorDiv(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction floorDiv_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'floorDiv');\n  let $b = convertToTensor(b, 'b', 'floorDiv');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.floorDiv($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: FloorDivInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             FloorDiv) as T;\n}\n\nexport const floorDiv = op({floorDiv_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Div, DivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {floorDiv} from './floorDiv';\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction div_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'int32' && $b.dtype === 'int32') {\n    return floorDiv($a, $b);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.realDivide($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: DivInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Div,\n             attrs) as T;\n}\n\nexport const div = op({div_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Multiply, MultiplyInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Multiplies two `tf.Tensor`s element-wise, A * B. Supports broadcasting.\n *\n * We also expose `tf.mulStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([2, 3, 4, 5]);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n *\n * ```js\n * // Broadcast mul a with b.\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.scalar(5);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n * @param a The first tensor to multiply.\n * @param b The second tensor to multiply. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mul_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mul');\n  let $b = convertToTensor(b, 'b', 'mul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.multiply($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MultiplyInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Multiply) as T;\n}\nexport const mul = op({mul_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Abs, AbsInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes absolute value element-wise: `abs(x)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.abs().print();  // or tf.abs(x)\n * ```\n * @param x The input `tf.Tensor`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction abs_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'abs');\n\n  const inputs: AbsInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    if ($x.dtype === 'complex64') {\n      return backend.complexAbs($x);\n    }\n\n    return backend.abs($x);\n  }, inputs as {} as NamedTensorMap, null /* grad */, Abs);\n}\n\nexport const abs = op({abs_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Acos, AcosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes acos of the input `tf.Tensor` element-wise: `acos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.acos().print();  // or tf.acos(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acos');\n  const inputs: AcosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acos);\n}\nexport const acos = op({acos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Acosh, AcoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the inverse hyperbolic cos of the input `tf.Tensor` element-wise:\n * `acosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([10, 1, 3, 5.7]);\n *\n * x.acosh().print();  // or tf.acosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acosh');\n  const inputs: AcoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acosh);\n}\nexport const acosh = op({acosh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AddN, AddNInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Adds a list of `tf.Tensor`s element-wise, each with the same shape and dtype.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n *\n * tf.addN([a, b, c]).print();\n * ```\n * @param tensors A list of tensors with the same shape and dtype.\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction addN_<T extends Tensor>(tensors: Array<T|TensorLike>): T {\n  util.assert(\n      Array.isArray(tensors),\n      () => 'The argument passed to tf.addN() must be a list of tensors');\n  util.assert(\n      tensors.length >= 1,\n      () => `Must pass at least one tensor to tf.addN(), but got ` +\n          `${tensors.length}`);\n\n  const $tensors =\n      tensors.map((t, i) => convertToTensor(t, `tensors${i}`, 'addN'));\n\n  const firstTensor = $tensors[0];\n  $tensors.forEach(t => {\n    if (t.dtype !== firstTensor.dtype) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same dtype');\n    }\n  });\n\n  $tensors.forEach(t => {\n    if (!util.arraysEqual(t.shape, firstTensor.shape)) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same shape');\n    }\n  });\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.addN($tensors);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: AddNInputs = $tensors;\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, AddN) as\n      T;\n}\n\nexport const addN = op({addN_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\n/**\n * Returns true if the axis specifies the inner most dimensions of the\n * array.\n */\nexport function axesAreInnerMostDims(axes: number[], rank: number): boolean {\n  for (let i = 0; i < axes.length; ++i) {\n    if (axes[axes.length - i - 1] !== rank - 1 - i) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function combineLocations(\n    outputLoc: number[], reduceLoc: number[], axes: number[]): number[] {\n  const rank = outputLoc.length + reduceLoc.length;\n  const loc = [];\n  let outIdx = 0;\n  let reduceIdx = 0;\n  \u00A0 for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      loc.push(outputLoc[outIdx++]);\n    } else {\n      loc.push(reduceLoc[reduceIdx++]);\n    }\n  }\n  return loc;\n}\n\nexport function computeOutAndReduceShapes(\n    aShape: number[], axes: number[]): [number[], number[]] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      outShape.push(aShape[dim]);\n    }\n  }\n  const reduceShape = axes.map(dim => aShape[dim]);\n  return [outShape, reduceShape];\n}\n\nexport function expandShapeToKeepDim(\n    shape: number[], axes: number[]): number[] {\n  const reduceSubShape = axes.map(x => 1);\n  return combineLocations(shape, reduceSubShape, axes);\n}\n\nexport function assertAxesAreInnerMostDims(\n    msg: string, axes: number[], rank: number): void {\n  util.assert(\n      axesAreInnerMostDims(axes, rank),\n      () => `${msg} supports only inner-most axes for now. ` +\n          `Got axes ${axes} and rank-${rank} input.`);\n}\n\n/**\n * Returns the axes permutation to be used with `tf.transpose`, if such\n * permutation is necessary. Otherwise it returns null. This method is used by\n * operations that operate only on inner-most axes.\n */\nexport function getAxesPermutation(axes: number[], rank: number): number[]|\n    null {\n  if (axesAreInnerMostDims(axes, rank)) {\n    return null;\n  }\n  const result: number[] = [];\n  for (let i = 0; i < rank; ++i) {\n    if (axes.indexOf(i) === -1) {\n      result.push(i);\n    }\n  }\n  axes.forEach(axis => result.push(axis));\n  return result;\n}\n\n/** Returns the axes permutation that undoes the original permutation. */\nexport function getUndoAxesPermutation(axes: number[]): number[] {\n  return axes.map((axis, i) => [i, axis])\n      .sort((a, b) => a[1] - b[1])\n      .map(x => x[0]);\n}\n\nexport function getInnerMostAxes(numAxes: number, rank: number): number[] {\n  const res: number[] = [];\n  for (let i = rank - numAxes; i < rank; ++i) {\n    res.push(i);\n  }\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {All, AllAttrs, AllInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical and of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.all().print();  // or tf.all(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.all(axis).print();  // or tf.all(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction all_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'all', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.all($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AllInputs = {x: $x};\n  const attrs: AllAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, All,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const all = op({all_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Any, AnyAttrs, AnyInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical or of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.any().print();  // or tf.any(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.any(axis).print();  // or tf.any(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction any_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'any', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.any($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AnyInputs = {x: $x};\n  const attrs: AnyAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Any,\n             attrs as {} as NamedAttrMap) as T;\n}\n\n// tslint:disable-next-line:variable-name\nexport const any = op({any_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMax, ArgMaxAttrs, ArgMaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the maximum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMax().print();  // or tf.argMax(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMax(axis).print();  // or tf.argMax(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMax_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMax');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMax($x, axes[0]);\n  };\n\n  const inputs: ArgMaxInputs = {x: $x};\n  const attrs: ArgMaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMax,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMax = op({argMax_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMin, ArgMinAttrs, ArgMinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the minimum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMin().print();  // or tf.argMin(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMin(axis).print();  // or tf.argMin(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMin_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMin');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if (axis == null) {\n      axis = 0;\n    }\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMin($x, axes[0]);\n  };\n\n  const inputs: ArgMinInputs = {x: $x};\n  const attrs: ArgMinAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMin,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMin = op({argMin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asin, AsinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes asin of the input `tf.Tensor` element-wise: `asin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asin().print();  // or tf.asin(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asin');\n  const inputs: AsinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asin);\n}\nexport const asin = op({asin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asinh, AsinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic sin of the input `tf.Tensor` element-wise:\n * `asinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asinh().print();  // or tf.asinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asinh');\n\n  const inputs: AsinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asinh);\n}\nexport const asinh = op({asinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atan, AtanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes atan of the input `tf.Tensor` element-wise: `atan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.atan().print();  // or tf.atan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atan');\n\n  const inputs: AtanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atan);\n}\nexport const atan = op({atan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Atan2, Atan2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes arctangent of `tf.Tensor`s a / b element-wise: `atan2(a, b)`.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1.0, 1.0, -1.0, .7]);\n * const b = tf.tensor1d([2.0, 13.0, 3.5, .21]);\n *\n * tf.atan2(a, b).print()\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan2_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'atan2');\n  let $b = convertToTensor(b, 'b', 'atan2');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.atan2($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: Atan2Inputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Atan2) as T;\n}\n\nexport const atan2 = op({atan2_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atanh, AtanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic tan of the input `tf.Tensor` element-wise:\n * `atanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.atanh().print();  // or tf.atanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atanh');\n\n  const inputs: AtanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atanh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atanh);\n}\nexport const atanh = op({atanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\ntype PadType = 'SAME'|'VALID'|'NUMBER'|'EXPLICIT';\n\n// For NHWC should be in the following form:\n//  [[0, 0], [pad_top,pad_bottom], [pad_left, pad_right], [0, 0]]\n// For NCHW should be in the following form:\n//  [[0, 0], [0, 0], [pad_top,pad_bottom], [pad_left, pad_right]]\n// Reference: https://www.tensorflow.org/api_docs/python/tf/nn/conv2d\nexport type ExplicitPadding =\n    [[number, number], [number, number], [number, number], [number, number]];\n\nexport type PadInfo = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  type: PadType\n};\n\nexport type PadInfo3D = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  front: number,\n  back: number,\n  type: PadType\n};\n\n/**\n * Information about the forward pass of a convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv2DInfo = {\n  batchSize: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideHeight: number,\n  strideWidth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo,\n  inShape: [number, number, number, number],\n  outShape: [number, number, number, number],\n  filterShape: [number, number, number, number]\n};\n\n/**\n *\n * @param inputShape Input tensor shape is of the following dimensions:\n *     `[batch, height, width, inChannels]`.\n * @param filterShape The filter shape is of the following dimensions:\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat The data format of the input and output data.\n *     Defaults to 'NHWC'.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.\n *     Defaults to `[1, 1]`. If `dilations` is a single number, then\n *     `dilationHeight == dilationWidth`.\n */\nexport function computeDilation2DInfo(\n    inputShape: [number, number, number, number],\n    filterShape: [number, number, number], strides: number|[number, number],\n    pad: 'same'|'valid'|number, dataFormat: 'NHWC' = 'NHWC',\n    dilations: number|[number, number]) {\n  // `computerConv2DInfo` require filterShape to be in the dimension of:\n  // `[filterHeight, filterWidth, depth, outDepth]`, dilation2d doesn't have\n  // outDepth, it should have the same depth as the input.\n  // Input shape: [batch, height, width, inChannels]\n  const inputChannels = inputShape[3];\n  const $filterShape =\n      [...filterShape, inputChannels] as [number, number, number, number];\n  const $dataFormat = convertConv2DDataFormat(dataFormat);\n\n  return computeConv2DInfo(\n      inputShape, $filterShape, strides, dilations, pad,\n      null /* roundingMode */, null /* depthWise */, $dataFormat);\n}\n\nexport function computePool2DInfo(\n    inShape: [number, number, number, number],\n    filterSize: [number, number]|number, strides: number|[number, number],\n    dilations: number|[number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  const [filterHeight, filterWidth] = parseTupleParam(filterSize);\n\n  let filterShape: [number, number, number, number];\n  if (dataFormat === 'channelsLast') {\n    filterShape = [filterHeight, filterWidth, inShape[3], inShape[3]];\n  } else if (dataFormat === 'channelsFirst') {\n    filterShape = [filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv2DInfo(\n      inShape, filterShape, strides, dilations, pad, roundingMode, false,\n      dataFormat);\n}\n\n/**\n * Computes the information for a forward pass of a pooling3D operation.\n */\nexport function computePool3DInfo(\n    inShape: [number, number, number, number, number],\n    filterSize: number|[number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC'): Conv3DInfo {\n  const [filterDepth, filterHeight, filterWidth] = parse3TupleParam(filterSize);\n\n  let filterShape: [number, number, number, number, number];\n  let $dataFormat: 'channelsFirst'|'channelsLast';\n  if (dataFormat === 'NDHWC') {\n    $dataFormat = 'channelsLast';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[4], inShape[4]];\n  } else if (dataFormat === 'NCDHW') {\n    $dataFormat = 'channelsFirst';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv3DInfo(\n      inShape, filterShape, strides, dilations, pad, false, $dataFormat,\n      roundingMode);\n}\n\n/**\n * Computes the information for a forward pass of a convolution/pooling\n * operation.\n */\nexport function computeConv2DInfo(\n    inShape: [number, number, number, number],\n    filterShape: [number, number, number, number],\n    strides: number|[number, number], dilations: number|[number, number],\n    pad: 'same'|'valid'|number|ExplicitPadding,\n    roundingMode?: 'floor'|'round'|'ceil', depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  let [batchSize, inHeight, inWidth, inChannels] = [-1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterHeight, filterWidth, , filterChannels] = filterShape;\n  const [strideHeight, strideWidth] = parseTupleParam(strides);\n  const [dilationHeight, dilationWidth] = parseTupleParam(dilations);\n\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outHeight, outWidth} = getPadAndOutInfo(\n      pad, inHeight, inWidth, strideHeight, strideWidth, effectiveFilterHeight,\n      effectiveFilterWidth, roundingMode, dataFormat);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inHeight,\n    inWidth,\n    inChannels,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideHeight,\n    strideWidth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\n/**\n * Information about the forward pass of a 3D convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv3DInfo = {\n  batchSize: number,\n  inDepth: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideDepth: number,\n  strideHeight: number,\n  strideWidth: number,\n  dilationDepth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterDepth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterDepth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo3D,\n  inShape: [number, number, number, number, number],\n  outShape: [number, number, number, number, number],\n  filterShape: [number, number, number, number, number]\n};\n\n/**\n * Computes the information for a forward pass of a 3D convolution/pooling\n * operation.\n */\nexport function computeConv3DInfo(\n    inShape: [number, number, number, number, number],\n    filterShape: [number, number, number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast',\n    roundingMode?: 'floor'|'round'|'ceil'): Conv3DInfo {\n  let [batchSize, inDepth, inHeight, inWidth, inChannels] =\n      [-1, -1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inDepth, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inDepth, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterDepth, filterHeight, filterWidth, , filterChannels] =\n      filterShape;\n  const [strideDepth, strideHeight, strideWidth] = parse3TupleParam(strides);\n  const [dilationDepth, dilationHeight, dilationWidth] =\n      parse3TupleParam(dilations);\n\n  const effectiveFilterDepth =\n      getEffectiveFilterSize(filterDepth, dilationDepth);\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outDepth, outHeight, outWidth} = get3DPadAndOutInfo(\n      pad, inDepth, inHeight, inWidth, strideDepth, strideHeight, strideWidth,\n      effectiveFilterDepth, effectiveFilterHeight, effectiveFilterWidth,\n      roundingMode);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outDepth, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outDepth, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inDepth,\n    inHeight,\n    inWidth,\n    inChannels,\n    outDepth,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideDepth,\n    strideHeight,\n    strideWidth,\n    filterDepth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterDepth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationDepth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\nfunction computeOutputShape2D(\n    inShape: [number, number], fieldSize: number, stride: number,\n    zeroPad?: number, roundingMode?: 'floor'|'round'|'ceil'): [number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputRows = inShape[0];\n  const inputCols = inShape[1];\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputRows, outputCols];\n}\n\nfunction computeOutputShape4D(\n    inShape: [number, number, number, number], fieldSize: number,\n    outChannels: number, stride: number, zeroPad?: number,\n    roundingMode?: 'floor'|'round'|'ceil'): [number, number, number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputDepth = inShape[0];\n  const inputRows = inShape[1];\n  const inputCols = inShape[2];\n\n  const outputDepths = conditionalRound(\n      (inputDepth - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputDepths),\n      () => `The output # of depths (${outputDepths}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputDepths, outputRows, outputCols, outChannels];\n}\n\nexport function computeDefaultPad(\n    inputShape: [number, number]|[number, number, number, number],\n    fieldSize: number, stride: number, dilation = 1): number {\n  const effectiveFieldSize = getEffectiveFilterSize(fieldSize, dilation);\n  return Math.floor(\n      (inputShape[0] * (stride - 1) - stride + effectiveFieldSize) / 2);\n}\n\nfunction parseTupleParam(param: number|number[]): [number, number, number] {\n  if (typeof param === 'number') {\n    return [param, param, param];\n  }\n  if (param.length === 2) {\n    return [param[0], param[1], 1];\n  }\n  return param as [number, number, number];\n}\n\nfunction parse3TupleParam(param: number|[number, number, number]):\n    [number, number, number] {\n  return typeof param === 'number' ? [param, param, param] : param;\n}\n\n/* See https://www.tensorflow.org/api_docs/python/tf/nn/atrous_conv2d\n * Atrous convolution is equivalent to standard convolution with upsampled\n * filters with effective_filter_height =\n * filter_height + (filter_height - 1) * (dilation - 1)\n * and effective_filter_width =\n * filter_width + (filter_width - 1) * (dilation - 1),\n * produced by inserting dilation - 1 zeros along consecutive elements across\n * the filters' spatial dimensions.\n * When there is a dilation, this converts a filter dimension to the\n * effective filter dimension, so it can be used in a standard convolution.\n */\nfunction getEffectiveFilterSize(filterSize: number, dilation: number) {\n  if (dilation <= 1) {\n    return filterSize;\n  }\n\n  return filterSize + (filterSize - 1) * (dilation - 1);\n}\n\nfunction getPadAndOutInfo(\n    pad: 'same'|'valid'|number|ExplicitPadding, inHeight: number,\n    inWidth: number, strideHeight: number, strideWidth: number,\n    filterHeight: number, filterWidth: number,\n    roundingMode: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|\n    'channelsLast'): {padInfo: PadInfo, outHeight: number, outWidth: number} {\n  let padInfo: PadInfo;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {top: pad, bottom: pad, left: pad, right: pad, type: padType};\n    const outShape = computeOutputShape2D(\n        [inHeight, inWidth], filterHeight, strideHeight, pad, roundingMode);\n    outHeight = outShape[0];\n    outWidth = outShape[1];\n  } else if (pad === 'same') {\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongHeight =\n        Math.max(0, (outHeight - 1) * strideHeight + filterHeight - inHeight);\n    const padAlongWidth =\n        Math.max(0, (outWidth - 1) * strideWidth + filterWidth - inWidth);\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n    padInfo = {top, bottom, left, right, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {top: 0, bottom: 0, left: 0, right: 0, type: 'VALID'};\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else if (typeof pad === 'object') {\n    const top = dataFormat === 'channelsLast' ? pad[1][0] : pad[2][0];\n    const bottom = dataFormat === 'channelsLast' ? pad[1][1] : pad[2][1];\n    const left = dataFormat === 'channelsLast' ? pad[2][0] : pad[3][0];\n    const right = dataFormat === 'channelsLast' ? pad[2][1] : pad[3][1];\n    const padType = (top === 0 && bottom === 0 && left === 0 && right === 0) ?\n        'VALID' :\n        'EXPLICIT';\n    padInfo = {top, bottom, left, right, type: padType};\n    outHeight = conditionalRound(\n        (inHeight - filterHeight + top + bottom) / strideHeight + 1,\n        roundingMode);\n    outWidth = conditionalRound(\n        (inWidth - filterWidth + left + right) / strideWidth + 1, roundingMode);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outHeight, outWidth};\n}\n\nfunction get3DPadAndOutInfo(\n    pad: 'same'|'valid'|number, inDepth: number, inHeight: number,\n    inWidth: number, strideDepth: number, strideHeight: number,\n    strideWidth: number, filterDepth: number, filterHeight: number,\n    filterWidth: number, roundingMode?: 'floor'|'round'|'ceil'): {\n  padInfo: PadInfo3D,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number\n} {\n  let padInfo: PadInfo3D;\n  let outDepth: number;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {\n      top: pad,\n      bottom: pad,\n      left: pad,\n      right: pad,\n      front: pad,\n      back: pad,\n      type: padType\n    };\n    const outShape = computeOutputShape4D(\n        [inDepth, inHeight, inWidth, 1], filterDepth, 1, strideDepth, pad,\n        roundingMode);\n    outDepth = outShape[0];\n    outHeight = outShape[1];\n    outWidth = outShape[2];\n  } else if (pad === 'same') {\n    outDepth = Math.ceil(inDepth / strideDepth);\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongDepth = (outDepth - 1) * strideDepth + filterDepth - inDepth;\n    const padAlongHeight =\n        (outHeight - 1) * strideHeight + filterHeight - inHeight;\n    const padAlongWidth = (outWidth - 1) * strideWidth + filterWidth - inWidth;\n    const front = Math.floor(padAlongDepth / 2);\n    const back = padAlongDepth - front;\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n\n    padInfo = {top, bottom, left, right, front, back, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {\n      top: 0,\n      bottom: 0,\n      left: 0,\n      right: 0,\n      front: 0,\n      back: 0,\n      type: 'VALID'\n    };\n    outDepth = Math.ceil((inDepth - filterDepth + 1) / strideDepth);\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outDepth, outHeight, outWidth};\n}\n\n/**\n * Rounds a value depending on the rounding mode\n * @param value\n * @param roundingMode\n */\nfunction conditionalRound(\n    value: number, roundingMode?: 'floor'|'round'|'ceil') {\n  if (!roundingMode) {\n    return value;\n  }\n  switch (roundingMode) {\n    case 'round':\n      // used for Caffe Conv\n      return Math.round(value);\n    case 'ceil':\n      // used for Caffe Pool\n      return Math.ceil(value);\n    case 'floor':\n      return Math.floor(value);\n    default:\n      throw new Error(`Unknown roundingMode ${roundingMode}`);\n  }\n}\n\nexport function tupleValuesAreOne(param: number|number[]): boolean {\n  const [dimA, dimB, dimC] = parseTupleParam(param);\n  return dimA === 1 && dimB === 1 && dimC === 1;\n}\n\nexport function eitherStridesOrDilationsAreOne(\n    strides: number|number[], dilations: number|number[]): boolean {\n  return tupleValuesAreOne(strides) || tupleValuesAreOne(dilations);\n}\n\n/**\n * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to\n *    'channelsLast'|'channelsFirst'\n * @param dataFormat in 'NHWC'|'NCHW' mode\n * @return dataFormat in 'channelsLast'|'channelsFirst' mode\n * @throws unknown dataFormat\n */\nexport function convertConv2DDataFormat(dataFormat: 'NHWC'|'NCHW'):\n    'channelsLast'|'channelsFirst' {\n  if (dataFormat === 'NHWC') {\n    return 'channelsLast';\n  } else if (dataFormat === 'NCHW') {\n    return 'channelsFirst';\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D average pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction avgPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'avgPool', 'float32');\n  const dilations = 1;\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in avgPool: x must be rank 4 but got rank ${x4D.rank}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    save([x4D]);\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      return x4D.clone();\n    }\n\n    return backend.avgPool(x4D, convInfo);\n  };\n\n  const inputs: AvgPoolInputs = {x: x4D};\n\n  const attrs: AvgPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, $x.dtype);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool = op({avgPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {AvgPool3D, AvgPool3DAttrs, AvgPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D average pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.avgPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates:\n *     `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction avgPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'avgPool3d', 'float32');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in avgPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in avgPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n\n    save([x5D]);\n\n    return backend.avgPool3d(x5D, convInfo);\n  };\n\n  const inputs: AvgPool3DInputs = {x: x5D};\n\n  const attrs: AvgPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool3D,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, x5D.dtype);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3d = op({avgPool3d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\nexport function assertParamsConsistent(shapes: number[][], axis: number) {\n  const rank = shapes[0].length;\n  shapes.forEach((shape, i) => {\n    util.assert(\n        shape.length === rank,\n        () =>\n            `Error in concat${rank}D: rank of tensors[${i}] must be the same ` +\n            `as the rank of the rest (${rank})`);\n  });\n\n  util.assert(\n      axis >= 0 && axis < rank,\n      () => `Error in concat${rank}D: axis must be between 0 and ${rank - 1}.`);\n\n  const firstShape = shapes[0];\n  shapes.forEach((shape, i) => {\n    for (let r = 0; r < rank; r++) {\n      util.assert(\n          (r === axis) || (shape[r] === firstShape[r]),\n          () => `Error in concat${rank}D: Shape of tensors[${i}] (${shape}) ` +\n              `does not match the shape of the rest (${firstShape}) ` +\n              `along the non-concatenated axis ${i}.`);\n    }\n  });\n}\n\nexport function computeOutShape(shapes: number[][], axis: number): number[] {\n  const outputShape = shapes[0].slice();\n  for (let i = 1; i < shapes.length; i++) {\n    outputShape[axis] += shapes[i][axis];\n  }\n  return outputShape;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Concat, ConcatAttrs, ConcatInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, parseAxisParam, sizeFromShape} from '../util';\n\nimport {assertParamsConsistent, computeOutShape} from './concat_util';\nimport {op} from './operation';\nimport {tensor} from './tensor';\n\n/**\n * Concatenates a list of `tf.Tensor`s along a given axis.\n *\n * The tensors ranks and types must match, and their sizes must match in all\n * dimensions except `axis`.\n *\n * Also available are stricter rank-specific methods that assert that\n * `tensors` are of the given rank:\n *   - `tf.concat1d`\n *   - `tf.concat2d`\n *   - `tf.concat3d`\n *   - `tf.concat4d`\n *\n * Except `tf.concat1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * a.concat(b).print();  // or a.concat(b)\n * ```\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.concat([a, b, c]).print();\n * ```\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [10, 20]]);\n * const b = tf.tensor2d([[3, 4], [30, 40]]);\n * const axis = 1;\n * tf.concat([a, b], axis).print();\n * ```\n * @param tensors A list of tensors to concatenate.\n * @param axis The axis to concate along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction concat_<T extends Tensor>(tensors: Array<T|TensorLike>, axis = 0): T {\n  assert(tensors.length >= 1, () => 'Pass at least one tensor to concat');\n\n  let $tensors = convertToTensorArray(tensors, 'tensors', 'concat');\n\n  if ($tensors[0].dtype === 'complex64') {\n    $tensors.forEach(tensor => {\n      if (tensor.dtype !== 'complex64') {\n        throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${tensor.dtype}. `);\n      }\n    });\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $axis = parseAxisParam(axis, $tensors[0].shape)[0];\n    const outShape = computeOutShape($tensors.map(t => t.shape), $axis);\n    if (sizeFromShape(outShape) === 0) {\n      return tensor([], outShape) as T;\n    }\n    // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n    $tensors = $tensors.filter(t => t.size > 0);\n    if ($tensors.length === 1) {\n      return $tensors[0];\n    }\n\n    const shapes = $tensors.map(t => t.shape);\n    assertParamsConsistent(shapes, $axis);\n\n    const res = backend.concat($tensors, $axis);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: ConcatInputs = $tensors;\n  const attr: ConcatAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Concat,\n             attr as {} as NamedAttrMap) as T;\n}\n\nexport const concat = op({concat_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sigmoid, SigmoidInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sigmoid element-wise, `1 / (1 + exp(-x))`\n *\n * ```js\n * const x = tf.tensor1d([0, -1, 2, -3]);\n *\n * x.sigmoid().print();  // or tf.sigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sigmoid');\n\n  const inputs: SigmoidInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sigmoid($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sigmoid);\n}\nexport const sigmoid = op({sigmoid_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Slice, SliceAttrs, SliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as slice_util from './slice_util';\n\n/**\n * Extracts a slice from a `tf.Tensor` starting at coordinates `begin`\n * and is of size `size`.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `x` is of the given rank:\n *   - `tf.slice1d`\n *   - `tf.slice2d`\n *   - `tf.slice3d`\n *   - `tf.slice4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.slice([1], [2]).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * x.slice([1, 0], [1, 2]).print();\n * ```\n * @param x The input `tf.Tensor` to slice from.\n * @param begin The coordinates to start the slice from. The length can be\n *     less than the rank of x - the rest of the axes will have implicit 0 as\n *     start. Can also be a single number, in which case it specifies the\n *     first axis.\n * @param size The size of the slice. The length can be less than the rank of\n *     x - the rest of the axes will have implicit -1. A value of -1 requests\n *     the rest of the dimensions in the axis. Can also be a single number,\n *     in which case it specifies the size of the first axis.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction slice_<R extends Rank, T extends Tensor<R>>(\n    x: T|TensorLike, begin: number|number[], size?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'slice');\n\n  if ($x.rank === 0) {\n    throw new Error('Slicing scalar is not possible');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const [begin_, size_] = slice_util.parseSliceParams($x, begin, size);\n    slice_util.assertParamsValid($x, begin_, size_);\n    save([$x]);\n    return backend.slice($x, begin_, size_);\n  };\n\n  const inputs: SliceInputs = {x: $x};\n  const attrs: SliceAttrs = {begin, size};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Slice,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const slice = op({slice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tanh, TanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic tangent of the input `tf.Tensor` element-wise: `tanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, 70]);\n *\n * x.tanh().print();  // or tf.tanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tanh');\n\n  const inputs: TanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const y = backend.tanh($x);\n    save([y]);\n    return y;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tanh);\n}\nexport const tanh = op({tanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {add} from './add';\nimport {concat} from './concat';\nimport {matMul} from './mat_mul';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {slice} from './slice';\nimport {tanh} from './tanh';\n\n/**\n * Computes the next state and output of a BasicLSTMCell.\n *\n * Returns `[newC, newH]`.\n *\n * Derived from tf.contrib.rnn.BasicLSTMCell.\n *\n * @param forgetBias Forget bias for the cell.\n * @param lstmKernel The weights for the cell.\n * @param lstmBias The bias for the cell.\n * @param data The input to the cell.\n * @param c Previous cell state.\n * @param h Previous cell output.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction basicLSTMCell_(\n    forgetBias: Scalar|TensorLike, lstmKernel: Tensor2D|TensorLike,\n    lstmBias: Tensor1D|TensorLike, data: Tensor2D|TensorLike,\n    c: Tensor2D|TensorLike, h: Tensor2D|TensorLike): [Tensor2D, Tensor2D] {\n  const $forgetBias =\n      convertToTensor(forgetBias, 'forgetBias', 'basicLSTMCell');\n  const $lstmKernel =\n      convertToTensor(lstmKernel, 'lstmKernel', 'basicLSTMCell');\n  const $lstmBias = convertToTensor(lstmBias, 'lstmBias', 'basicLSTMCell');\n  const $data = convertToTensor(data, 'data', 'basicLSTMCell');\n  const $c = convertToTensor(c, 'c', 'basicLSTMCell');\n  const $h = convertToTensor(h, 'h', 'basicLSTMCell');\n\n  const combined = concat([$data, $h], 1);\n  const weighted = matMul(combined, $lstmKernel);\n  const res: Tensor2D = add(weighted, $lstmBias);\n\n  // i = input_gate, j = new_input, f = forget_gate, o = output_gate\n  const batchSize = res.shape[0];\n  const sliceCols = res.shape[1] / 4;\n  const sliceSize: [number, number] = [batchSize, sliceCols];\n  const i = slice(res, [0, 0], sliceSize);\n  const j = slice(res, [0, sliceCols], sliceSize);\n  const f = slice(res, [0, sliceCols * 2], sliceSize);\n  const o = slice(res, [0, sliceCols * 3], sliceSize);\n\n  const newC: Tensor2D =\n      add(mul(sigmoid(i), tanh(j)),\n          mul($c, sigmoid(add($forgetBias, f)) as Tensor2D));\n  const newH: Tensor2D = mul(tanh(newC), sigmoid(o));\n  return [newC, newH];\n}\n\nexport const basicLSTMCell = op({basicLSTMCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchToSpaceND, BatchToSpaceNDAttrs, BatchToSpaceNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation reshapes the \"batch\" dimension 0 into `M + 1` dimensions of\n * shape `blockShape + [batch]`, interleaves these blocks back into the grid\n * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with\n * the same rank as the input. The spatial dimensions of this intermediate\n * result are then optionally cropped according to `crops` to produce the\n * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise\n * description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);\n * const blockShape = [2, 2];\n * const crops = [[0, 0], [0, 0]];\n *\n * x.batchToSpaceND(blockShape, crops).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param crops A 2-D array.  Must have shape `[M, 2]`, all values must be >= 0.\n * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input\n * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required\n * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,\n * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,\n * x.shape[N-1]]`\n *\n * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /\n * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`\n * according to `crops` to produce the output of shape: `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],\n * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -\n * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction batchToSpaceND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], crops: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'batchToSpaceND');\n  const prod = blockShape.reduce((a, b) => a * b);\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank is ${$x.rank} but should be > than blockShape.length ${\n          blockShape.length}`);\n\n  util.assert(\n      crops.length === blockShape.length,\n      () => `crops.length is ${\n          crops.length} but should be equal to blockShape.length  ${\n          blockShape.length}`);\n\n  util.assert(\n      $x.shape[0] % prod === 0,\n      () => `input tensor batch is ${\n                $x.shape[0]} but is not divisible by the product of ` +\n          `the elements of blockShape ${blockShape.join(' * ')} === ${prod}`);\n\n  const forward: ForwardFunc<T> = backend => {\n    return backend.batchToSpaceND($x, blockShape, crops);\n  };\n\n  const inputs: BatchToSpaceNDInputs = {x: $x};\n  const attrs: BatchToSpaceNDAttrs = {blockShape, crops};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      BatchToSpaceND, attrs as {} as NamedAttrMap);\n}\n\nexport const batchToSpaceND = op({batchToSpaceND_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, Tensor4D} from '../tensor';\nimport {Rank} from '../types';\nimport {reshape} from './reshape';\n\nexport function xAs4D<R extends Rank>(x: Tensor<R>) {\n  let x4D: Tensor4D;\n  if (x.rank === 0 || x.rank === 1) {\n    x4D = reshape(x, [1, 1, 1, x.size]);\n  } else if (x.rank === 2) {\n    x4D = reshape(x, [1, 1, x.shape[0], x.shape[1]]);\n  } else if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  } else {\n    x4D = x as Tensor4D;\n  }\n\n  return x4D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {xAs4D} from './batchnorm_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Batch normalization.\n *\n * As described in\n * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).\n *\n * Mean, variance, scale, and offset can be of two shapes:\n *   - The same shape as the input.\n *   - In the common case, the depth dimension is the last dimension of x, so\n *     the values would be an `tf.Tensor1D` of shape [depth].\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that parameters passed are of given rank\n *   - `tf.batchNorm2d`\n *   - `tf.batchNorm3d`\n *   - `tf.batchNorm4d`\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction batchNorm_<R extends Rank>(\n    x: Tensor<R>|TensorLike, mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor<R>|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor<R>|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n\n  util.assert(\n      $mean.rank === $variance.rank,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      $offset == null || $mean.rank === $offset.rank,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      $scale == null || $mean.rank === $scale.rank,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  const x4D: Tensor4D = xAs4D($x);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([x4D, $mean, $variance, $scale]);\n\n    return backend.batchNorm(\n        x4D, as1DOr4D($mean), as1DOr4D($variance), as1DOr4D($offset),\n        as1DOr4D($scale), varianceEpsilon);\n  };\n\n  const inputs: FusedBatchNormInputs = {\n    x: x4D,\n    scale: $scale,\n    offset: $offset,\n    mean: $mean,\n    variance: $variance\n  };\n\n  const attrs: FusedBatchNormAttrs = {varianceEpsilon};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      FusedBatchNorm, attrs as {} as NamedAttrMap);\n\n  return reshape(res, $x.shape);\n}\n\nfunction as1DOr4D(x: Tensor): Tensor4D|Tensor1D {\n  if (x == null) {\n    return null;\n  }\n  if (x.rank === 0) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [x.size]) as Tensor1D;\n  } else if (x.rank === 1) {\n    return x as Tensor1D;\n  } else if (x.rank === 2) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, 1, x.shape[0], x.shape[1]]) as Tensor4D;\n  } else if (x.rank === 3) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]) as Tensor4D;\n  }\n  return x as Tensor4D;\n}\n\nexport const batchNorm = op({batchNorm_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 2D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm2d_(\n    x: Tensor2D|TensorLike, mean: Tensor2D|Tensor1D|TensorLike,\n    variance: Tensor2D|Tensor1D|TensorLike,\n    offset?: Tensor2D|Tensor1D|TensorLike, scale?: Tensor2D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor2D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor2D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 2,\n      () => `Error in batchNorm2D: x must be rank 2 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 2 || $mean.rank === 1,\n      () => `Error in batchNorm2D: mean must be rank 2 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 2 || $variance.rank === 1,\n      () => `Error in batchNorm2D: variance must be rank 2 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 2 || $scale.rank === 1,\n        () => `Error in batchNorm2D: scale must be rank 2 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 2 || $offset.rank === 1,\n        () => `Error in batchNorm2D: offset must be rank 2 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm2d = op({batchNorm2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 3D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm3d_(\n    x: Tensor3D|TensorLike, mean: Tensor3D|Tensor1D|TensorLike,\n    variance: Tensor3D|Tensor1D|TensorLike,\n    offset?: Tensor3D|Tensor1D|TensorLike, scale?: Tensor3D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor3D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor3D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 3,\n      () => `Error in batchNorm3D: x must be rank 3 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 3 || $mean.rank === 1,\n      () => `Error in batchNorm3D: mean must be rank 3 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 3 || $variance.rank === 1,\n      () => `Error in batchNorm3D: variance must be rank 3 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 3 || $scale.rank === 1,\n        () => `Error in batchNorm3D: scale must be rank 3 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 3 || $offset.rank === 1,\n        () => `Error in batchNorm3D: offset must be rank 3 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm3d = op({batchNorm3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 4D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm4d_(\n    x: Tensor4D|TensorLike, mean: Tensor4D|Tensor1D|TensorLike,\n    variance: Tensor4D|Tensor1D|TensorLike,\n    offset?: Tensor4D|Tensor1D|TensorLike, scale?: Tensor4D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor4D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor4D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 4,\n      () => `Error in batchNorm4D: x must be rank 4 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 4 || $mean.rank === 1,\n      () => `Error in batchNorm4D: mean must be rank 4 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 4 || $variance.rank === 1,\n      () => `Error in batchNorm4D: variance must be rank 4 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 4 || $scale.rank === 1,\n        () => `Error in batchNorm4D: scale must be rank 4 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 4 || $offset.rank === 1,\n        () => `Error in batchNorm4D: offset must be rank 4 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm4d = op({batchNorm4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE} from '../engine';\nimport {BroadcastTo, BroadCastToAttrs, BroadcastToInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Broadcast an array to a compatible shape NumPy-style.\n *\n * The tensor's shape is compared to the broadcast shape from end to beginning.\n * Ones are prepended to the tensor's shape until is has the same length as\n * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is\n * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then\n * the input tensor is tiled N times along that axis (using tf.tile).\n *\n * @param input The tensor that is to be broadcasted.\n * @param shape The input is to be broadcast to this shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction broadcastTo_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  let input = convertToTensor(x, 'broadcastTo', 'x');\n  const xShape = input.shape;\n\n  if (shape.some(d => !(d > 0) || d % 1 !== 0)) {\n    throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);\n  }\n\n  if (shape.length < input.rank) {\n    throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${\n        input.rank}.`);\n  }\n\n  if (shape.length > input.rank) {\n    const newShape = input.shape.slice();\n    while (newShape.length < shape.length) {\n      newShape.unshift(1);\n    }\n    input = reshape(input, newShape);\n  }\n\n  const inputShape = input.shape;\n  const reps: number[] = Array.from(shape);\n  for (let i = shape.length - 1; i >= 0; i--) {\n    if (inputShape[i] === shape[i]) {\n      reps[i] = 1;\n    } else if (input.shape[i] !== 1) {\n      throw new Error(\n          `broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);\n    }\n  }\n  const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);\n\n  if (axes.length === 0) {\n    return clone(input) as Tensor<R>;\n  }\n\n  const forward = (backend: KernelBackend) => backend.tile(input, reps);\n\n  const inputs: BroadcastToInputs = {x: input};\n  const attrs: BroadCastToAttrs = {shape, inputShape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             BroadcastTo, attrs as unknown as NamedAttrMap) as Tensor<R>;\n}\n\nexport const broadcastTo = op({broadcastTo_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Ceil, CeilInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes ceiling of input `tf.Tensor` element-wise: `ceil(x)`\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.ceil().print();  // or tf.ceil(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction ceil_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'ceil');\n\n  const inputs: CeilInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.ceil($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Ceil);\n}\nexport const ceil = op({ceil_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {ClipByValue, ClipByValueAttrs, ClipByValueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)\n * ```\n * @param x The input tensor.\n * @param clipValueMin Lower-bound of range to be clipped to.\n * @param clipValueMax Upper-bound of range to be clipped to.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction clipByValue_<T extends Tensor>(\n    x: T|TensorLike, clipValueMin: number, clipValueMax: number): T {\n  const $x = convertToTensor(x, 'x', 'clipByValue');\n  util.assert(\n      (clipValueMin <= clipValueMax),\n      () => `Error in clip: min (${clipValueMin}) must be ` +\n          `less than or equal to max (${clipValueMax}).`);\n\n  const inputs: ClipByValueInputs = {x: $x};\n  const attrs: ClipByValueAttrs = {clipValueMin, clipValueMax};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const res = backend.clip($x, clipValueMin, clipValueMax);\n        save([$x]);\n        return res;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, ClipByValue,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const clipByValue = op({clipByValue_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor1D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(3) = |r1, g1, b1|\n * B: shape(2) = |r2, g2|\n * C = tf.concat1d([A, B]) == |r1, g1, b1, r2, g2|\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @return The concatenated array.\n */\nfunction concat1d_(tensors: Array<Tensor1D|TensorLike>): Tensor1D {\n  return concat(tensors, 0 /* axis */);\n}\n\nexport const concat1d = op({concat1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor2D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *\n * B: shape(2, 3) = | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * C = tf.concat2d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *                  | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * if axis = 1:\n * C = shape(2, 6) = | r1, g1, b1, r3, g3, b3 |\n *                   | r2, g2, b2, r4, g4, b4 |\n *\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concatenate along.\n * @return The concatenated array.\n */\nfunction concat2d_(\n    tensors: Array<Tensor2D|TensorLike>, axis: number): Tensor2D {\n  return concat(tensors, axis);\n}\n\nexport const concat2d = op({concat2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor3D`s along an axis.\n * See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *\n * B: shape(2, 1, 3) = | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * C = tf.concat3d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *                     | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * if axis = 1:\n * C: shape(2, 2, 3) = | r1, g1, b1, r3, g3, b3 |\n *                     | r2, g2, b2, r4, g4, b4 |\n *\n * if axis = 2:\n * C = shape(2, 1, 6) = | r1, g1, b1, r3, g3, b3 |\n *                      | r2, g2, b2, r4, g4, b4 |\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat3d_(\n    tensors: Array<Tensor3D|TensorLike>, axis: number): Tensor3D {\n  return concat(tensors, axis);\n}\n\nexport const concat3d = op({concat3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor4D`s along an axis.\n * See `concat` for details.\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat4d_(\n    tensors: Array<Tensor4D|TensorLike>, axis: number): Tensor4D {\n  return concat(tensors, axis);\n}\n\nexport const concat4d = op({concat4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2D, Conv2DAttrs, Conv2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 2D convolution over the input x.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2d: input must be rank 4, but got rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  util.assert(\n      inDepth === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${inDepth}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        false, $dataFormat);\n    const res = backend.conv2d(x4D, $filter, convInfo);\n\n    save([x4D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DInputs = {x: x4D, filter: $filter};\n  const attrs:\n      Conv2DAttrs = {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv2D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2d = op({conv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D, Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 1D convolution over the input x.\n *\n * @param x The input tensor, of rank 3 or rank 2, of shape\n *     `[batch, width, inChannels]`. If rank 2, batch of 1 is assumed.\n * @param filter The filter, rank 3, of shape\n *     `[filterWidth, inDepth, outDepth]`.\n * @param stride The number of entries by which the filter is moved right at\n *     each step.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from \"NWC\", \"NCW\". Defaults to \"NWC\",\n *     the data is stored in the order of [batch, in_width, in_channels]. Only\n *     \"NWC\" is currently supported.\n * @param dilation The dilation rate in which we sample input values in\n *     atrous convolution. Defaults to `1`. If it is greater than 1, then\n *     stride must be `1`.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv1d_<T extends Tensor2D|Tensor3D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike, stride: number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv1d');\n  const $filter = convertToTensor(filter, 'filter', 'conv1d');\n\n  let x3D = $x as Tensor3D;\n  let reshapedTo3D = false;\n  if ($x.rank === 2) {\n    reshapedTo3D = true;\n    x3D = reshape($x, [1, $x.shape[0], $x.shape[1]]);\n  }\n\n  util.assert(\n      x3D.rank === 3,\n      () => `Error in conv1d: input must be rank 3, but got rank ${x3D.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in conv1d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv1d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x3D.shape[2] === $filter.shape[1],\n      () => `Error in conv1d: depth of input (${x3D.shape[2]}) must match ` +\n          `input depth for filter ${$filter.shape[1]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(stride, dilation),\n      () => 'Error in conv1D: Either stride or dilation must be 1. ' +\n          `Got stride ${stride} and dilation '${dilation}'`);\n  util.assert(\n      dataFormat === 'NWC',\n      () => `Error in conv1d: got dataFormat of ${\n          dataFormat} but only NWC is currently supported.`);\n\n  const filter4D = reshape(\n      $filter, [1, $filter.shape[0], $filter.shape[1], $filter.shape[2]]);\n  const input4D = reshape(x3D, [x3D.shape[0], 1, x3D.shape[1], x3D.shape[2]]);\n  const strides: [number, number] = [1, stride];\n  const dilations: [number, number] = [1, dilation];\n\n  const conv2dDataFormat = 'NHWC';\n\n  const res = conv2d(\n      (input4D as Tensor4D), (filter4D as Tensor4D), strides, pad,\n      conv2dDataFormat, dilations, dimRoundingMode);\n\n  if (reshapedTo3D) {\n    return reshape(res, [res.shape[2], res.shape[3]]) as T;\n  }\n\n  return reshape(res, [res.shape[0], res.shape[2], res.shape[3]]) as T;\n}\n\nexport const conv1d = op({conv1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs, Conv2DBackpropInputInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 2D convolution.\n *\n * @param xShape The shape of the input: [batch, height, width, inDepth].\n * If length of 3, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 4 or rank 3 of shape\n *   `[batch, outHeight, outWidth, outDepth]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction conv2DBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape4D = xShape as [number, number, number, number];\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n    xShape4D = [1, xShape[0], xShape[1], xShape[2]];\n  }\n\n  util.assert(\n      xShape4D.length === 4,\n      () =>\n          `Error in conv2dDerInput: inShape must be length 4, but got length ` +\n          `${xShape4D.length}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerInput: dy must be rank 4, but got ` +\n          `rank ${dy4D.rank}`);\n  util.assert(\n      filter.rank === 4,\n      () => `Error in conv2dDerInput: filter must be rank 4, but got ` +\n          `rank ${filter.rank}`);\n  const inDepth = dataFormat === 'NHWC' ? xShape4D[3] : xShape4D[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filter.shape[2],\n      () => `Error in conv2dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[2]}.`);\n  util.assert(\n      outDepth === filter.shape[3],\n      () => `Error in conv2dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[3]}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerInput: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const dilations = 1;\n\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        xShape4D, filter.shape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    const res = backend.conv2dDerInput(dy4D, filter, convInfo);\n\n    save([dy4D, filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DBackpropInputInputs = {dy: dy4D, filter};\n\n  const attrs: Conv2DBackpropInputAttrs =\n      {strides, pad, dataFormat, dimRoundingMode, inputShape: xShape4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      Conv2DBackpropInput, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2DBackpropInput = op({conv2DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv2DBackpropInput} from './conv2d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 2D convolution of an image, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 4 or rank 3, of shape\n *   `[batch, height, width, inDepth]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 4 or rank 3:\n *     `[batch, height, width, outDepth]`. If rank 3, batch of 1 is assumed.\n * @param strides The strides of the original convolution:\n *     `[strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n * @param dimRoundingMode The rounding mode used when computing output\n *    dimensions if pad is a number. If none is provided, it will not round\n *    and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2dTranspose_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv2dTranspose');\n\n  return conv2DBackpropInput(\n      outputShape, $x, $filter, strides, pad, 'NHWC', dimRoundingMode);\n}\n\nexport const conv2dTranspose = op({conv2dTranspose_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3D, Conv3DAttrs, Conv3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {eitherStridesOrDilationsAreOne} from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 3D convolution over the input x.\n *\n * @param x The input tensor, of rank 5 or rank 4, of shape\n *     `[batch, depth, height, width, channels]`. If rank 4,\n * batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inChannels, outChannels]`.\n *      inChannels must match between input and filter.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationDepth, dilationHeight,\n *     dilationWidth]` in which we sample input values across the height\n *     and width dimensions in atrous convolution. Defaults to `[1, 1, 1]`.\n *     If `dilations` is a single number, then\n *     `dilationDepth == dilationHeight == dilationWidth`. If it is greater\n *     than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    strides: [number, number, number]|number, pad: 'valid'|'same',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations: [number, number, number]|number = [1, 1, 1]): T {\n  const $x = convertToTensor(x, 'x', 'conv3d');\n  const $filter = convertToTensor(filter, 'filter', 'conv3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3d: input must be rank 5, but got rank ${x5D.rank}.`);\n  util.assert(\n      $filter.rank === 5,\n      () => `Error in conv3d: filter must be rank 5, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x5D.shape[4] === $filter.shape[3],\n      () => `Error in conv3d: depth of input (${x5D.shape[4]}) must match ` +\n          `input depth for filter ${$filter.shape[3]}.`);\n  util.assert(\n      eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv3D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in conv3d: got dataFormat of ${\n          dataFormat} but only NDHWC is currently supported.`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, $filter.shape, strides, dilations, pad);\n    const res = backend.conv3d(x5D, $filter, convInfo);\n\n    save([x5D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv3DInputs = {x: x5D, filter: $filter};\n\n  const attrs: Conv3DAttrs = {strides, pad, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3d = op({conv3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropInputAttrs, Conv3DBackpropInputInputs, Conv3DBackpropInputV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 3D convolution.\n *\n * @param xShape The shape of the input: [batch, depth, height, width,\n * in_channels]. If length of 4, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 5 or rank 4 of shape\n *   `[batch, outDepth, outHeight, outWidth, in_channels]`.\n * If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n */\nfunction conv3DBackpropInput_<T extends Tensor4D|Tensor5D>(\n    xShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    dy: T, filter: Tensor5D, strides: [number, number, number]|number,\n    pad: 'valid'|'same'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape5D = xShape as [number, number, number, number, number];\n  let dy5D = dy as Tensor5D;\n  let reshapedTo5D = false;\n  if (dy.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n    xShape5D = [1, xShape[0], xShape[1], xShape[2], xShape[3]];\n  }\n\n  const inDepth = xShape5D[4];\n  const outDepth = dy5D.shape[4];\n  util.assert(\n      xShape5D.length === 5,\n      () =>\n          `Error in conv3dDerInput: inShape must be length 5, but got length ` +\n          `${xShape5D.length}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerInput: dy must be rank 5, but got ` +\n          `rank ${dy5D.rank}`);\n  util.assert(\n      filter.rank === 5,\n      () => `Error in conv3dDerInput: filter must be rank 5, but got ` +\n          `rank ${filter.rank}`);\n  util.assert(\n      inDepth === filter.shape[3],\n      () => `Error in conv3dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[3]}.`);\n  util.assert(\n      outDepth === filter.shape[4],\n      () => `Error in conv3dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[4]}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        xShape5D, filter.shape, strides, dilations, pad);\n\n    return backend.conv3dDerInput(dy5D, filter, convInfo);\n  };\n\n  const inputs: Conv3DBackpropInputInputs = {dy: dy5D};\n\n  const attrs: Conv3DBackpropInputAttrs = {pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, Conv3DBackpropInputV2,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3DBackpropInput = op({conv3DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D, Tensor5D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv3DBackpropInput} from './conv3d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 3D convolution of a volume, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 5 or rank 4, of shape\n *   `[batch, depth, height, width, inDepth]`. If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[depth, filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 5 or rank 4:\n *     `[batch, depth, height, width, outDepth]`. If rank 3, batch of 1 is\n *    assumed.\n * @param strides The strides of the original convolution:\n *     `[strideDepth, strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3dTranspose_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    outputShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): T {\n  const $x = convertToTensor(x, 'x', 'conv3dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv3dTranspose');\n\n  return conv3DBackpropInput(outputShape, $x, $filter, strides, pad);\n}\n\nexport const conv3dTranspose = op({conv3dTranspose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cos, CosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes cos of the input `tf.Tensor` element-wise: `cos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.cos().print();  // or tf.cos(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cos');\n\n  const inputs: CosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cos);\n}\nexport const cos = op({cos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cosh, CoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic cos of the input `tf.Tensor` element-wise: `cosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.cosh().print();  // or tf.cosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cosh');\n  const inputs: CoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cosh);\n}\nexport const cosh = op({cosh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Cumsum, CumsumAttrs, CumsumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {getAxesPermutation, getInnerMostAxes, getUndoAxesPermutation} from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the cumulative sum of a `tf.Tensor` along `axis`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4]);\n * x.cumsum().print();\n * ```\n * ```js\n * const x = tf.tensor([[1, 2], [3, 4]]);\n * x.cumsum().print();\n * ```\n *\n * @param x The input tensor to be summed.\n * @param axis The axis along which to sum. Optional. Defaults to 0.\n * @param exclusive Whether to perform exclusive cumulative sum. Optional.\n *     Defaults to false. If set to true then the sum of each tensor entry\n *     does not include its own value, but only the values previous to it\n *     along the specified axis.\n * @param reverse Whether to sum in the opposite direction. Optional.\n *     Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Scan'}\n */\nfunction cumsum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis = 0, exclusive = false, reverse = false): T {\n  const $x = convertToTensor(x, 'x', 'cumsum');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const permutation = getAxesPermutation([axis], $x.rank);\n        let permutedX = $x;\n        if (permutation != null) {\n          permutedX = transpose($x, permutation);\n        }\n        const permutedAxis = getInnerMostAxes(1, $x.rank)[0];\n        let value = backend.cumsum(permutedX, permutedAxis, exclusive, reverse);\n        save([$x]);\n\n        if (permutation != null) {\n          const reversePermutation = getUndoAxesPermutation(permutation);\n          value = transpose(value, reversePermutation);\n        }\n        return value;\n      };\n\n  const inputs: CumsumInputs = {x: $x};\n  const attrs: CumsumAttrs = {axis, exclusive, reverse};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Cumsum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const cumsum = op({cumsum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthToSpace, DepthToSpaceAttrs, DepthToSpaceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Rearranges data from depth into blocks of spatial data. More specifically,\n * this op outputs a copy of the input tensor where values from the `depth`\n * dimension are moved in spatial blocks to the `height` and `width` dimensions.\n * The attr `blockSize` indicates the input block size and how the data is\n * moved.\n *\n *  - Chunks of data of size `blockSize * blockSize` from depth are rearranged\n * into non-overlapping blocks of size `blockSize x blockSize`\n *\n *  - The width the output tensor is `inputWidth * blockSize`, whereas the\n * height is `inputHeight * blockSize`\n *\n *  - The Y, X coordinates within each block of the output image are determined\n * by the high order component of the input channel index\n *\n *  - The depth of the input tensor must be divisible by `blockSize *\n * blockSize`\n *\n * The `dataFormat` attr specifies the layout of the input and output tensors\n * with the following options: \"NHWC\": [ `batch, height, width, channels` ]\n * \"NCHW\": [ `batch, channels, height, width` ]\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 1, 1, 4]);\n * const blockSize = 2;\n * const dataFormat = \"NHWC\";\n *\n * tf.depthToSpace(x, blockSize, dataFormat).print();\n * ```\n *\n * @param x The input tensor of rank 4\n * @param blockSIze  An `int` that is `>= 2`. The size of the spatial block\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\"\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction depthToSpace_(\n    x: Tensor4D|TensorLike4D, blockSize: number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'depthToSpace') as Tensor4D;\n\n  const inputHeight = (dataFormat === 'NHWC') ? $x.shape[1] : $x.shape[2];\n  const inputWidth = (dataFormat === 'NHWC') ? $x.shape[2] : $x.shape[3];\n  const inputDepth = (dataFormat === 'NHWC') ? $x.shape[3] : $x.shape[1];\n\n  util.assert(\n      inputHeight * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputHeight} and ${blockSize}  for depthToSpace with input shape\n    ${$x.shape}`);\n\n  util.assert(\n      inputWidth * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputWidth} and ${blockSize} for depthToSpace with input shape\n        ${$x.shape}`);\n\n  util.assert(\n      (inputDepth % (blockSize * blockSize) === 0),\n      () => `Dimension size must be evenly divisible by ${\n          blockSize * blockSize} but is ${\n          inputDepth} for depthToSpace with input shape ${$x.shape}`);\n\n  const forward: ForwardFunc<Tensor4D> = backend =>\n      backend.depthToSpace($x, blockSize, dataFormat);\n\n  const inputs: DepthToSpaceInputs = {x: $x};\n  const attrs: DepthToSpaceAttrs = {blockSize, dataFormat};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      DepthToSpace, attrs as {} as NamedAttrMap);\n}\n\nexport const depthToSpace = op({depthToSpace_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs, DepthwiseConv2dNativeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Depthwise 2D convolution.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction depthwiseConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in depthwiseConv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in depthwiseConv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1];\n    }\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in depthwiseConv2d: Either strides or dilations must be ' +\n            `1. Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        true /* depthwise */);\n    const res = backend.depthwiseConv2D(x4D, $filter, convInfo);\n    save([x4D, $filter]);\n    return res;\n  };\n\n  const inputs: DepthwiseConv2dNativeInputs = {x: x4D, filter: $filter};\n  const attrs: DepthwiseConv2dNativeAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      DepthwiseConv2dNative, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2d = op({depthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Diag, DiagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a diagonal tensor with a given diagonal values.\n *\n * Given a diagonal, this operation returns a tensor with the diagonal and\n * everything else padded with zeros.\n *\n * Assume the input has dimensions `[D1,..., Dk]`, then the output is a tensor\n * of rank 2k with dimensions `[D1,..., Dk, D1,..., Dk]`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * tf.diag(x).print()\n * ```\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4, 5, 6, 6, 8], [4, 2])\n *\n * tf.diag(x).print()\n * ```\n * @param x The input tensor.\n */\nfunction diag_(x: Tensor): Tensor {\n  const $x = convertToTensor(x, 'x', 'diag');\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const flat = reshape($x, [$x.size]);\n    const result = backend.diag(flat);\n    const outShape = [...x.shape, ...x.shape];\n\n    return reshape(result, outShape);\n  };\n\n  const inputs: DiagInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Diag);\n}\n\nexport const diag = op({diag_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DAttrs, Dilation2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the grayscale dilation over the input `x`.\n *\n * @param x The input tensor, rank 3 or rank 4 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter tensor, rank 3, of shape\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat Specify the data format of the input and output data.\n *      Defaults to 'NHWC'. Only 'NHWC' is currently supported. With the\n *      default format \"NHWC\", the data is stored in the order of: [batch,\n *      height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     for atrous morphological dilation. Defaults to `[1, 1]`. If `dilations`\n *     is a single number, then `dilationHeight == dilationWidth`. If it is\n *     greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction dilation2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same',\n    dilations: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'dilation2d');\n  const $filter = convertToTensor(filter, 'filter', 'dilation2d');\n\n  util.assert(\n      $x.rank === 3 || $x.rank === 4,\n      () => `Error in dilation2d: input must be rank 3 or 4, but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in dilation2d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in dilation2d: Only NHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n    reshapedTo4D = true;\n  }\n\n  const inputs: Dilation2DInputs = {x: x4D, filter: $filter};\n  const attrs: Dilation2DAttrs = {strides, pad, dilations};\n\n  const res = ENGINE.runKernel(\n                  Dilation2D, inputs as {} as NamedTensorMap,\n                  attrs as {} as NamedAttrMap) as T;\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res;\n}\n\nexport const dilation2d = op({dilation2d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Returns the dimensions in the input shape that are broadcasted to\n * produce the provided output shape.\n *\n * The returned dimensions are 0-indexed and sorted. An example:\n * inShape = [4, 1, 3]\n * outShape = [5, 4, 3, 3]\n * result = [1]. Dimension 1 (2nd dimension of input) gets broadcasted 1 => 3.\n */\nexport function getBroadcastDims(\n    inShape: number[], outShape: number[]): number[] {\n  const inRank = inShape.length;\n  const dims: number[] = [];\n  for (let i = 0; i < inRank; i++) {\n    const dim = inRank - 1 - i;\n    const a = inShape[dim] || 1;\n    const b = outShape[outShape.length - 1 - i] || 1;\n    if (b > 1 && a === 1) {\n      dims.unshift(dim);\n    }\n  }\n  return dims;\n}\n\n/**\n * Returns the axes in the output space that should be reduced to produce\n * the input space.\n */\nexport function getReductionAxes(\n    inShape: number[], outShape: number[]): number[] {\n  const result: number[] = [];\n  for (let i = 0; i < outShape.length; i++) {\n    const inDim = inShape[inShape.length - i - 1];\n    const outAxis = outShape.length - i - 1;\n    const outDim = outShape[outAxis];\n    if (inDim == null || (inDim === 1 && outDim > 1)) {\n      result.unshift(outAxis);\n    }\n  }\n  return result;\n}\n\nexport function assertAndGetBroadcastShape(\n    shapeA: number[], shapeB: number[]): number[] {\n  const result: number[] = [];\n  const l = Math.max(shapeA.length, shapeB.length);\n\n  for (let i = 0; i < l; i++) {\n    let a = shapeA[shapeA.length - i - 1];\n    if (a == null) {\n      a = 1;\n    }\n    let b = shapeB[shapeB.length - i - 1];\n    if (b == null) {\n      b = 1;\n    }\n    if (a === 1) {\n      result.unshift(b);\n    } else if (b === 1) {\n      result.unshift(a);\n    } else if (a !== b) {\n      const errMsg = `Operands could not be broadcast together with shapes ` +\n          `${shapeA} and ${shapeB}.`;\n      throw Error(errMsg);\n    } else {\n      result.unshift(a);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Equal, EqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a == b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.equal(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction equal_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'equal');\n  let $b = convertToTensor(b, 'b', 'equal');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.equal($a, $b);\n\n  const inputs: EqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, Equal) as T;\n}\n\nexport const equal = op({equal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SelectV2, SelectV2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch} from '../util';\n\nimport {broadcastTo} from './broadcast_to';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the elements, either `a` or `b` depending on the `condition`.\n *\n * If the condition is true, select from `a`, otherwise select from `b`.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const a = tf.tensor1d([1 , 2, 3]);\n * const b = tf.tensor1d([-1, -2, -3]);\n *\n * a.where(cond, b).print();\n * ```\n *\n * @param condition The input condition. Must be of dtype bool.\n * @param a If `condition` is rank 1, `a` may have a higher rank but\n *     its first dimension must match the size of `condition`.\n * @param b A tensor with the same dtype as `a` and with shape that is\n *     compatible with `a`.\n * @return A tensor with same dtype as `a` and `b`, and shape that is\n *     broadcastable from `a` and `b`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction where_<T extends Tensor>(\n    condition: Tensor|TensorLike, a: T|TensorLike, b: T|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'where');\n  const $b = convertToTensor(b, 'b', 'where');\n  const $condition = convertToTensor(condition, 'condition', 'where', 'bool');\n  // TODO: move this logic to forward function when the broadcastTo op is\n  // implemented in WASM.\n  // Find the broadcastable shape for $a and $b.\n  const broadcastShape = assertAndGetBroadcastShape($a.shape, $b.shape);\n  const $broadcastedA = broadcastTo($a, broadcastShape);\n  const $broadcastedB = broadcastTo($b, broadcastShape);\n  if ($condition.rank === 1) {\n    // If condition rank is 1, then the first dimension must match the size of\n    // condition.\n    assert(\n        $condition.shape[0] === $a.shape[0],\n        () => 'The first dimension of `a` must match the size of `condition`.');\n  }\n\n  if ($condition.rank !== 1) {\n    // A must have the same shape as condition.\n    assertShapesMatch(\n        $condition.shape, $broadcastedB.shape, 'Error in where: ');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.select($condition, $broadcastedA, $broadcastedB);\n    save([$condition]);\n    return res;\n  };\n  const inputs: SelectV2Inputs = {\n    condition: $condition,\n    t: $broadcastedA,\n    e: $broadcastedB\n  };\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* gradient */,\n             SelectV2) as T;\n}\n\nexport const where = op({where_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {ZerosLike, ZerosLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.zerosLike(x).print();\n * ```\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction zerosLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'zerosLike');\n  const inputs: ZerosLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             backend => backend.zerosLike($x), inputs as {} as NamedTensorMap,\n             null /* grad */, ZerosLike) as T;\n}\nexport const zerosLike = op({zerosLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {div} from './div';\nimport {equal} from './equal';\nimport {op} from './operation';\nimport {where} from './where';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting. Return 0\n * if denominator is 0.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n * const c = tf.tensor1d([0, 0, 0, 0]);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n * const c = tf.scalar(0);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction divNoNan_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  // TODO: Make this into its own kernel.\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const divResult = div($a, $b);\n  const zeros = zerosLike(divResult);\n  const bEqualsZero = equal($b, zeros);\n  return where(bEqualsZero, zeros, divResult) as T;\n}\n\nexport const divNoNan = op({divNoNan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor,} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices and/or vectors, `t1` and `t2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor2d([[1, 2], [3, 4]]);\n * const c = tf.tensor2d([[1, 2, 3], [4, 5, 6]]);\n *\n * a.dot(b).print();  // or tf.dot(a, b)\n * b.dot(a).print();\n * b.dot(c).print();\n * ```\n * @param t1 The first tensor in the dot operation.\n * @param t2 The second tensor in the dot operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction dot_(t1: Tensor|TensorLike, t2: Tensor|TensorLike): Tensor {\n  const $t1 = convertToTensor(t1, 't1', 'dot');\n  const $t2 = convertToTensor(t2, 't2', 'dot');\n\n  util.assert(\n      ($t1.rank === 1 || $t1.rank === 2) && ($t2.rank === 1 || $t2.rank === 2),\n      () => `Error in dot: inputs must all be rank 1 or 2, but got ranks ` +\n          `${$t1.rank} and ${$t2.rank}.`);\n\n  const t1Inner = ($t1.rank === 1 ? $t1.size : $t1.shape[1]);\n  const t2Inner = ($t2.rank === 1 ? $t2.size : $t2.shape[0]);\n\n  util.assert(\n      t1Inner === t2Inner,\n      () => `Error in dot: inner dimensions of inputs must match, but got ` +\n          `${t1Inner} and ${t2Inner}.`);\n\n  if ($t1.rank === 1 && $t2.rank === 1) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, []);\n  } else if ($t1.rank === 1 && $t2.rank === 2) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else if ($t1.rank === 2 && $t2.rank === 1) {\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul($t1, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else {\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul($t1, t22D);\n    return t1t2;\n  }\n}\n\nexport const dot = op({dot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential linear element-wise: `x > 0 ? e ^ x - 1 : 0`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 1, -3, 2]);\n *\n * x.elu().print();  // or tf.elu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction elu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'elu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.elu($x);\n    save([y]);\n    return y;\n  };\n\n  const inputs: EluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Elu) as\n      T;\n}\n\nexport const elu = op({elu_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Erf, ErfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes gause error function of the input `tf.Tensor` element-wise:\n * `erf(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.erf().print(); // or tf.erf(x);\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction erf_<T extends Tensor>(x: T|TensorLike): T {\n  let $x = convertToTensor(x, 'x', 'erf');\n  util.assert(\n      $x.dtype === 'int32' || $x.dtype === 'float32',\n      () => 'Input dtype must be `int32` or `float32`.');\n\n  if ($x.dtype === 'int32') {\n    $x = cast($x, 'float32');\n  }\n\n  const inputs: ErfInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.erf($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Erf);\n}\nexport const erf = op({erf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Exp, ExpInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.exp().print();  // or tf.exp(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction exp_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'exp');\n\n  const inputs: ExpInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.exp($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Exp);\n}\nexport const exp = op({exp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension\n * into the tensor's shape.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const axis = 1;\n * x.expandDims(axis).print();\n * ```\n *\n * @param x The input tensor whose dimensions to be expanded.\n * @param axis The dimension index at which to insert shape of `1`. Defaults\n *     to 0 (the first dimension).\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction expandDims_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'expandDims', parseAs);\n\n  util.assert(axis <= $x.rank, () => 'Axis must be <= rank of the tensor');\n  const newShape = $x.shape.slice();\n  if (axis < 0) {\n    // Negative value is counted from the tail of rank.\n    util.assert(\n        -($x.rank + 1) <= axis,\n        () => `Axis must be in the interval [${- ($x.rank + 1)}, ${$x.rank}]`);\n    axis = $x.rank + axis + 1;\n  }\n  newShape.splice(axis, 0, 1);\n  return reshape($x, newShape) as T;\n}\n\nexport const expandDims = op({expandDims_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Expm1, Expm1Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` minus one element-wise.\n * `e ^ x - 1`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.expm1().print();  // or tf.expm1(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction expm1_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'expm1');\n\n  const inputs: Expm1Inputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.expm1($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Expm1);\n}\nexport const expm1 = op({expm1_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Tile, TileAttrs, TileInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Construct a tensor by repeating it the number of times given by reps.\n *\n * This operation creates a new tensor by replicating `input` `reps`\n * times. The output tensor's i'th dimension has `input.shape[i] *\n * reps[i]` elements, and the values of `input` are replicated\n * `reps[i]` times along the i'th dimension. For example, tiling\n * `[a, b, c, d]` by `[2]` produces `[a, b, c, d, a, b, c, d]`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n *\n * a.tile([2]).print();    // or a.tile([2])\n * ```\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.tile([1, 2]).print();  // or a.tile([1, 2])\n * ```\n * @param x The tensor to tile.\n * @param reps Determines the number of replications per dimension.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction tile_<T extends Tensor>(x: T|TensorLike, reps: number[]): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'tile', parseAs);\n  util.assert(\n      $x.rank === reps.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of reps ${reps}.`);\n\n  const forward: ForwardFunc<T> = (backend, save) => {\n    const res = backend.tile($x, reps);\n    save([$x]);\n    return res;\n  };\n\n  const inputsToSave = [$x];\n  const inputs: TileInputs = {x: $x};\n  const attrs: TileAttrs = {reps};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, Tile,\n      attrs as unknown as NamedAttrMap, inputsToSave);\n}\n\nexport const tile = op({tile_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {DataType} from '../types';\n\nimport {buffer} from './buffer';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {tile} from './tile';\n\n/**\n * Create an identity matrix.\n *\n * @param numRows Number of rows.\n * @param numColumns Number of columns. Defaults to `numRows`.\n * @param batchShape If provided, will add the batch shape to the beginning\n *   of the shape of the returned `tf.Tensor` by repeating the identity\n *   matrix.\n * @param dtype Data type.\n * @returns Identity matrix of the specified size and data type, possibly\n *   with batch repetition if `batchShape` is specified.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction eye_(\n    numRows: number, numColumns?: number,\n    batchShape?:\n        [\n          number\n        ]|[number,\n           number]|[number, number, number]|[number, number, number, number],\n    dtype: DataType = 'float32'): Tensor2D {\n  if (numColumns == null) {\n    numColumns = numRows;\n  }\n  const buff = buffer([numRows, numColumns], dtype);\n  const n = numRows <= numColumns ? numRows : numColumns;\n  for (let i = 0; i < n; ++i) {\n    buff.set(1, i, i);\n  }\n  const out: Tensor2D = reshape(buff.toTensor(), [numRows, numColumns]);\n  if (batchShape == null) {\n    return out;\n  } else {\n    if (batchShape.length === 1) {\n      return tile(expandDims(out, 0), [batchShape[0], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 2) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(\n                 expandDims(expandDims(out, 0), 0),\n                 [batchShape[0], batchShape[1], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 3) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(expandDims(expandDims(expandDims(out, 0), 0), 0), [\n               batchShape[0], batchShape[1], batchShape[2], 1, 1\n             ]) as Tensor2D;\n    } else {\n      throw new Error(\n          `eye() currently supports only 1D and 2D ` +\n          // tslint:disable-next-line:no-any\n          `batchShapes, but received ${(batchShape as any).length}D.`);\n    }\n  }\n}\n\nexport const eye = op({eye_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Fill, FillAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\n/**\n * Creates a `tf.Tensor` filled with a scalar value.\n *\n * ```js\n * tf.fill([2, 2], 4).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param value The scalar value to fill the tensor with.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n * 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction fill<R extends Rank>(\n    shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n  const attrs: FillAttrs = {shape, value, dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.fill(shape, value, dtype), {}, null, Fill,\n      attrs as {} as NamedAttrMap);\n}\n\nexport {fill};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Floor, FloorInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes floor of input `tf.Tensor` element-wise: `floor(x)`.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.floor().print();  // or tf.floor(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction floor_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'floor');\n\n  const inputs: FloorInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.floor($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Floor);\n}\nexport const floor = op({floor_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inputs of size above this threshold will be parallelized by calling multiple\n * shader programs.\n */\nimport {nearestDivisor} from '../util';\n\nexport const PARALLELIZE_THRESHOLD = 30;\n\nexport interface ReduceInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  outSize: number;\n}\n\nexport function computeOptimalWindowSize(inSize: number): number {\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    return inSize;\n  }\n  return nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {nearestDivisor} from '../util';\n\nimport {PARALLELIZE_THRESHOLD} from './reduce_util';\n\nexport interface SegOpInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  numSegments: number;\n}\n\nexport function segOpComputeOptimalWindowSize(\n    inSize: number, numSegments: number): number {\n  let done = false;\n  let res;\n\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    res = inSize;\n    done = true;\n  } else {\n    res = nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n  }\n\n  while (!done) {\n    if (res > numSegments || res === inSize) {\n      done = true;\n    } else {\n      res = nearestDivisor(inSize, res + 1);\n    }\n  }\n  return res;\n}\n\nexport function computeOutShape(\n    aShape: number[], axis: number, numSegments: number): number[] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (dim !== axis) {\n      outShape.push(aShape[dim]);\n    } else {\n      outShape.push(numSegments);\n    }\n  }\n  return outShape;\n}\n\nexport interface GatherOpShapeInfo {\n  batchSize: number;\n  sliceSize: number;\n  dimSize: number;\n  outputShape: number[];\n}\nexport function collectGatherOpShapeInfo(\n    x: Tensor, indices: Tensor, axis: number): GatherOpShapeInfo {\n  const dimSize = x.shape[axis];\n\n  const outputShape: number[] = [];\n  let batchSize = 1;\n  let sliceSize = 1;\n  for (let i = 0; i < axis; i++) {\n    outputShape.push(x.shape[i]);\n    batchSize *= x.shape[i];\n  }\n\n  for (let i = 0; i < indices.rank; i++) {\n    outputShape.push(indices.shape[i]);\n  }\n\n  for (let i = axis + 1; i < x.rank; i++) {\n    outputShape.push(x.shape[i]);\n    sliceSize *= x.shape[i];\n  }\n\n  return {batchSize, sliceSize, dimSize, outputShape};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherV2, GatherV2Attrs, GatherV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {collectGatherOpShapeInfo} from './segment_util';\n\n/**\n * Gather slices from tensor `x`'s axis `axis` according to `indices`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const indices = tf.tensor1d([1, 3, 3], 'int32');\n *\n * x.gather(indices).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const indices = tf.tensor1d([1, 1, 0], 'int32');\n *\n * x.gather(indices).print();\n * ```\n * @param x The input tensor whose slices to be gathered.\n * @param indices The indices of the values to extract.\n * @param axis The axis over which to select values. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction gather_<T extends Tensor>(\n    x: T|TensorLike, indices: Tensor|TensorLike, axis = 0): T {\n  const $x = convertToTensor(x, 'x', 'gather');\n  const $indices = convertToTensor(indices, 'indices', 'gather', 'int32');\n\n  const inputs: GatherV2Inputs = {x: $x, indices: $indices};\n  const attrs: GatherV2Attrs = {axis};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const parsedAxis = parseAxisParam(axis, $x.shape)[0];\n    const shapeInfo = collectGatherOpShapeInfo($x, $indices, parsedAxis);\n\n    const res =\n        backend.gather($x, reshape($indices, [$indices.size]), parsedAxis);\n    save([$x, $indices]);\n\n    return reshape(res, shapeInfo.outputShape);\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, GatherV2,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const gather = op({gather_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Greater, GreaterInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a > b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greater(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greater_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greater');\n  let $b = convertToTensor(b, 'b', 'greater');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.greater($a, $b);\n\n  const inputs: GreaterInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             Greater) as T;\n}\n\nexport const greater = op({greater_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GreaterEqual, GreaterEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a >= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greaterEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greaterEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greaterEqual');\n  let $b = convertToTensor(b, 'b', 'greaterEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.greaterEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: GreaterEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             GreaterEqual) as T;\n}\n\nexport const greaterEqual = op({greaterEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Imag, ImagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n/**\n * Returns the imaginary part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the imaginary part of each element in input considered as a complex number.\n * If input is real, a tensor of all zeros is returned.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.imag(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction imag_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'imag');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.imag($input);\n  };\n\n  const inputs: ImagInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Imag) as T;\n}\n\nexport const imag = op({imag_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsFinite, IsFiniteInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are finite.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isFinite().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isFinite_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isFinite');\n\n  const inputs: IsFiniteInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isFinite($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsFinite);\n}\nexport const isFinite = op({isFinite_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsInf, IsInfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are Infinity or -Infinity.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isInf().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isInf_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isInf');\n\n  const inputs: IsInfInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isInf($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsInf);\n}\nexport const isInf = op({isInf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsNan, IsNanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * RReturns which elements of x are NaN.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isNaN().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isNaN_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isNaN');\n  const inputs: IsNanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.isNaN($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsNan);\n}\nexport const isNaN = op({isNaN_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Maximum, MaximumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the max of a and b (`a > b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `tf.maximumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * ```js\n * // Broadcast maximum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction maximum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'maximum');\n  let $b = convertToTensor(b, 'b', 'maximum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.maximum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: MaximumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Maximum) as T;\n}\n\nexport const maximum = op({maximum_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar} from '../tensor';\nimport {DataType} from '../types';\nimport {isTypedArray} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-0 `tf.Tensor` (scalar) with the provided value and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.scalar` as it makes the code more readable.\n *\n * ```js\n * tf.scalar(3.14).print();\n * ```\n *\n * @param value The value of the scalar.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function scalar(\n    value: number|boolean|string|Uint8Array, dtype?: DataType): Scalar {\n  if (((isTypedArray(value) && dtype !== 'string') || Array.isArray(value)) &&\n      dtype !== 'complex64') {\n    throw new Error(\n        'Error creating a new Scalar: value must be a primitive ' +\n        '(number|boolean|string)');\n  }\n  if (dtype === 'string' && isTypedArray(value) &&\n      !(value instanceof Uint8Array)) {\n    throw new Error(\n        'When making a scalar from encoded string, ' +\n        'the value must be `Uint8Array`.');\n  }\n  const shape: number[] = [];\n  const inferredShape: number[] = [];\n  return makeTensor(value, shape, inferredShape, dtype) as Scalar;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {maximum} from './maximum';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {scalar} from './scalar';\n\n/**\n * Computes leaky rectified linear element-wise.\n *\n * See\n * [http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf](\n *     http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf)\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.leakyRelu(0.1).print();  // or tf.leakyRelu(x, 0.1)\n * ```\n * @param x The input tensor.\n * @param alpha The scaling factor for negative values, defaults to 0.2.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction leakyRelu_<T extends Tensor>(x: T|TensorLike, alpha = 0.2): T {\n  const $x = convertToTensor(x, 'x', 'leakyRelu');\n  return maximum(mul(scalar(alpha), $x), $x);\n}\n\nexport const leakyRelu = op({leakyRelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Less, LessInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a < b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.less(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction less_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'less');\n  let $b = convertToTensor(b, 'b', 'less');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.less($a, $b);\n\n  const inputs: LessInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Less) as\n      T;\n}\n\nexport const less = op({less_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LessEqual, LessEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a <= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.lessEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction lessEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'lessEqual');\n  let $b = convertToTensor(b, 'b', 'lessEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.lessEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: LessEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LessEqual) as T;\n}\n\nexport const lessEqual = op({lessEqual_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LinSpace, LinSpaceAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor1D} from '../tensor';\n\n/**\n * Return an evenly spaced sequence of numbers over the given interval.\n *\n * ```js\n * tf.linspace(0, 9, 10).print();\n * ```\n * @param start The start value of the sequence.\n * @param stop The end value of the sequence.\n * @param num The number of values to generate.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function linspace(start: number, stop: number, num: number): Tensor1D {\n  if (num <= 0) {\n    throw new Error('The number of values should be positive.');\n  }\n\n  const attrs: LinSpaceAttrs = {start, stop, num};\n  return ENGINE.runKernelFunc(\n      backend => backend.linspace(start, stop, num), {} /* inputs */,\n      null /* grad */, LinSpace, attrs as {} as NamedAttrMap);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRN, LRNAttrs, LRNInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Normalizes the activation of a local neighborhood across or within\n * channels.\n *\n * @param x The input tensor. The 4-D input tensor is treated as a 3-D array\n *     of 1D vectors (along the last dimension), and each vector is\n *     normalized independently.\n * @param depthRadius The number of adjacent channels in the 1D normalization\n *     window.\n * @param bias A constant bias term for the basis.\n * @param alpha A scale factor, usually positive.\n * @param beta An exponent.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction localResponseNormalization_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const $x = convertToTensor(x, 'x', 'localResponseNormalization');\n  util.assert(\n      $x.rank === 4 || $x.rank === 3,\n      () => `Error in localResponseNormalization: x must be rank 3 or 4 but got\n               rank ${$x.rank}.`);\n  util.assert(\n      util.isInt(depthRadius),\n      () => `Error in localResponseNormalization: depthRadius must be an ` +\n          `integer but got depthRadius ${depthRadius}.`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.localResponseNormalization4D(\n        x4D, depthRadius, bias, alpha, beta);\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: LRNInputs = {x: x4D};\n\n  const attrs: LRNAttrs = {depthRadius, bias, alpha, beta};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, LRN,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  } else {\n    return res as T;\n  }\n}\n\nexport const localResponseNormalization = op({localResponseNormalization_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log, LogInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` element-wise: `ln(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E]);\n *\n * x.log().print();  // or tf.log(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log');\n\n  const inputs: LogInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log);\n}\nexport const log = op({log_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log1p, Log1pInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` plus one\n * element-wise: `ln(1 + x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E - 1]);\n *\n * x.log1p().print();  // or tf.log1p(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log1p_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log1p');\n\n  const inputs: Log1pInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log1p($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log1p);\n}\nexport const log1p = op({log1p_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {CustomGradientFunc, ENGINE} from './engine';\nimport {Scalar, Tensor, Variable} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {convertToTensor, convertToTensorArray} from './tensor_util_env';\nimport {TensorLike} from './types';\nimport * as util from './util';\n\n/**\n * Provided `f(x)`, returns another function `g(x, dy?)`, which gives the\n * gradient of `f(x)` with respect to `x`.\n *\n * If `dy` is provided, the gradient of `f(x).mul(dy).sum()` with respect to\n * `x` is computed instead. `f(x)` must take a single tensor `x` and return a\n * single tensor `y`. If `f()` takes multiple inputs, use `tf.grads` instead.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.grad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * g(x).print();\n * ```\n *\n * ```js\n * // f(x) = x ^ 3\n * const f = x => x.pow(tf.scalar(3, 'int32'));\n * // f'(x) = 3x ^ 2\n * const g = tf.grad(f);\n * // f''(x) = 6x\n * const gg = tf.grad(g);\n *\n * const x = tf.tensor1d([2, 3]);\n * gg(x).print();\n * ```\n *\n * @param f The function f(x), to compute gradient for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grad(f: (x: Tensor) => Tensor): (\n    x: TensorLike|Tensor, dy?: TensorLike|Tensor) => Tensor {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grad(f) must be a function');\n  return (x: TensorLike|Tensor, dy?: TensorLike|Tensor): Tensor => {\n    // x can be of any dtype, thus null as the last argument.\n    const $x = convertToTensor(x, 'x', 'tf.grad', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grad') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f($x), [$x], $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grad(f)(x, dy) must match the shape ' +\n                'returned by f(x)');\n      }\n      checkGrads(grads);\n      return grads[0];\n    });\n  };\n}\n\n/**\n * Provided `f(x1, x2,...)`, returns another function `g([x1, x2,...], dy?)`,\n * which gives an array of gradients of `f()` with respect to each input\n * [`x1`,`x2`,...].\n *\n * If `dy` is passed when calling `g()`, the gradient of\n * `f(x1,...).mul(dy).sum()` with respect to each input is computed instead.\n * The provided `f` must take one or more tensors and return a single tensor\n * `y`. If `f()` takes a single input, we recommend using `tf.grad` instead.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df / da = b, df / db = a\n * const g = tf.grads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const [da, db] = g([a, b]);\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @param f The function `f(x1, x2,...)` to compute gradients for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grads(f: (...args: Tensor[]) => Tensor): (\n    args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike) => Tensor[] {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grads(f) must be a function');\n  return (args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike): Tensor[] => {\n    util.assert(\n        Array.isArray(args),\n        () => 'The args passed in grads(f)(args) must be an array ' +\n            'of `Tensor`s or `TensorLike`s');\n    // args can be of any dtype, thus null as the last argument.\n    const $args = convertToTensorArray(args, 'args', 'tf.grads', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grads') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f(...$args), $args, $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grads(f)([x1,...], dy) must ' +\n                'match the shape returned by f([x1,...])');\n      }\n      checkGrads(grads);\n      return grads;\n    });\n  };\n}\n\n/**\n * Like `tf.grad`, but also returns the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grad: The gradient of `f(x)` w.r.t `x` (result of `tf.grad`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.valueAndGrad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * const {value, grad} = g(x);\n *\n * console.log('value');\n * value.print();\n * console.log('grad');\n * grad.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrad<I extends Tensor, O extends Tensor>(f: (x: I) => O): (\n    x: I, dy?: O) => {\n  value: O;\n  grad: I;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrad(f) must be a function');\n  return (x: I, dy?: O) => {\n    util.assert(\n        x instanceof Tensor,\n        () => 'The x passed in valueAndGrad(f)(x) must be a tensor');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrad(f)(x, dy) must be a tensor');\n    const {grads, value} = ENGINE.gradients(() => f(x), [x], dy);\n    checkGrads(grads);\n    return {grad: grads[0] as I, value};\n  };\n}\n\n/**\n * Like `tf.grads`, but returns also the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grads: The gradients of `f()` w.r.t each input (result of `tf.grads`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df/da = b, df/db = a\n * const g = tf.valueAndGrads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const {value, grads} = g([a, b]);\n *\n * const [da, db] = grads;\n *\n * console.log('value');\n * value.print();\n *\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrads<O extends Tensor>(f: (...args: Tensor[]) => O): (\n    args: Tensor[], dy?: O) => {\n  grads: Tensor[];\n  value: O;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrads(f) must be a function');\n  return (args: Tensor[], dy?: O) => {\n    util.assert(\n        Array.isArray(args) && args.every(arg => arg instanceof Tensor),\n        () => 'The args passed in valueAndGrads(f)(args) must be array of ' +\n            'tensors');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrads(f)(args, dy) must be a tensor');\n    const res = ENGINE.gradients(() => f(...args), args, dy);\n    if (dy != null) {\n      util.assertShapesMatch(\n          res.value.shape, dy.shape,\n          'The shape of dy passed in valueAndGrads(f)([x1,...], dy) must ' +\n              'match the shape returned by f([x1,...])');\n    }\n    checkGrads(res.grads);\n    return res;\n  };\n}\n\n/**\n * Computes and returns the gradient of f(x) with respect to the list of\n * trainable variables provided by `varList`. If no list is provided, it\n * defaults to all trainable variables.\n *\n * ```js\n * const a = tf.variable(tf.tensor1d([3, 4]));\n * const b = tf.variable(tf.tensor1d([5, 6]));\n * const x = tf.tensor1d([1, 2]);\n *\n * // f(a, b) = a * x ^ 2 + b * x\n * const f = () => a.mul(x.square()).add(b.mul(x)).sum();\n * // df/da = x ^ 2, df/db = x\n * const {value, grads} = tf.variableGrads(f);\n *\n * Object.keys(grads).forEach(varName => grads[varName].print());\n * ```\n *\n * @param f The function to execute. f() should return a scalar.\n * @param varList The list of variables to compute the gradients with respect\n *     to. Defaults to all trainable variables.\n * @returns An object with the following keys and values:\n *   - `value`: The value of the function `f`.\n *   - `grads`: A map from the names of the variables to the gradients.\n *     If the `varList` argument is provided explicitly and contains a subset of\n *     non-trainable variables, this map in the return value will contain keys\n *     that map the names of the non-trainable variables to `null`.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction variableGrads(f: () => Scalar, varList?: Variable[]):\n    {value: Scalar, grads: NamedTensorMap} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in variableGrads(f) must be a function');\n  util.assert(\n      varList == null ||\n          Array.isArray(varList) && varList.every(v => v instanceof Variable),\n      () =>\n          'The varList passed in variableGrads(f, varList) must be an array ' +\n          'of variables');\n\n  const specifiedVarList = varList != null;\n  if (!specifiedVarList) {\n    // Get all of the trainable variables.\n    varList = [];\n    for (const varName in ENGINE.registeredVariables) {\n      varList.push(ENGINE.registeredVariables[varName]);\n    }\n  }\n\n  const specifiedNonTrainable: Variable[] =\n      specifiedVarList ? varList.filter(variable => !variable.trainable) : null;\n\n  // Prune non-trainable variables.\n  const originalVarCount = varList.length;\n  varList = varList.filter(variable => variable.trainable);\n  util.assert(\n      varList.length > 0,\n      () => `variableGrads() expects at least one of the input variables to ` +\n          `be trainable, but none of the ${originalVarCount} variables is ` +\n          `trainable.`);\n\n  const allowNoGradients = true;\n  const {value, grads} = ENGINE.gradients(f, varList, null, allowNoGradients);\n\n  util.assert(\n      grads.some(g => g != null),\n      () => 'Cannot find a connection between any variable and the result of ' +\n          'the loss function y=f(x). Please make sure the operations that ' +\n          'use variables are inside the function f passed to minimize().');\n  util.assert(\n      value.rank === 0,\n      () => `The f passed in variableGrads(f) must return a scalar, but it ` +\n          `returned a rank-${value.rank} tensor`);\n\n  const namedGrads: NamedTensorMap = {};\n  varList.forEach((v, i) => {\n    if (grads[i] != null) {\n      namedGrads[v.name] = grads[i];\n    }\n  });\n  if (specifiedNonTrainable != null) {\n    // If varList is explicitly provided and contains non-trainable values,\n    // add them to the returned gradients with `null` values.\n    specifiedNonTrainable.forEach(v => namedGrads[v.name] = null);\n  }\n  return {value, grads: namedGrads};\n}\n\n/**\n * Overrides the gradient computation of a function `f`.\n *\n * Takes a function\n * `f(...inputs, save) => {value: Tensor, gradFunc: (dy, saved) => Tensor[]}`\n * and returns another function `g(...inputs)` which takes the same inputs as\n * `f`. When called, `g` returns `f().value`. In backward mode, custom gradients\n * with respect to each input of `f` are computed using `f().gradFunc`.\n *\n * The `save` function passsed to `f` should be used for saving tensors needed\n * in the gradient. And the `saved` passed to the `gradFunc` is a\n * `NamedTensorMap`, which contains those saved tensor.\n *\n * ```js\n * const customOp = tf.customGrad((x, save) => {\n *   // Save x to make sure it's available later for the gradient.\n *   save([x]);\n *   // Override gradient of our custom x ^ 2 op to be dy * abs(x);\n *   return {\n *     value: x.square(),\n *     // Note `saved.x` which points to the `x` we saved earlier.\n *     gradFunc: (dy, saved) => [dy.mul(saved[0].abs())]\n *   };\n * });\n *\n * const x = tf.tensor1d([-1, -2, 3]);\n * const dx = tf.grad(x => customOp(x));\n *\n * console.log(`f(x):`);\n * customOp(x).print();\n * console.log(`f'(x):`);\n * dx(x).print();\n * ```\n *\n * @param f The function to evaluate in forward mode, which should return\n *     `{value: Tensor, gradFunc: (dy, saved) => Tensor[]}`, where `gradFunc`\n *     returns the custom gradients of `f` with respect to its inputs.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n    (...args: Tensor[]) => T {\n  return ENGINE.customGrad(f);\n}\n\nfunction checkGrads(grads: Tensor[]) {\n  const numNullGradients = grads.filter(g => g == null).length;\n  if (numNullGradients > 0) {\n    throw new Error(\n        `Cannot compute gradient of y=f(x) with respect to x. Make sure that\n    the f you passed encloses all operations that lead from x to y.`);\n  }\n}\n\nexport {\n  customGrad,\n  variableGrads,\n  valueAndGrad,\n  valueAndGrads,\n  grad,\n  grads,\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Negate, NegateInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes `-1 * x` element-wise.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, -2, 0], [2, 2]);\n *\n * x.neg().print();  // or tf.neg(x)\n * ```\n *\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction neg_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'neg');\n\n  const inputs: NegateInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.neg($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Negate);\n}\nexport const neg = op({neg_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softplus, SoftplusInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes softplus of the input `tf.Tensor` element-wise: `log(exp(x) + 1)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.softplus().print();  // or tf.softplus(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction softplus_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'softplus');\n\n  const inputs: SoftplusInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.softplus($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Softplus);\n}\nexport const softplus = op({softplus_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {mul} from './mul';\nimport {neg} from './neg';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {softplus} from './softplus';\n\n/**\n * Computes log sigmoid of the input `tf.Tensor` element-wise:\n * `logSigmoid(x)`. For numerical stability, we use `-tf.softplus(-x)`.\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.logSigmoid().print();  // or tf.logSigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction logSigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logSigmoid');\n\n  // Use a custom gradient to maintain previous implementation.\n  // There is no LogSigmoid kernel in TF so we can't use engine.runKernel\n  // directly\n  const customOp = customGrad((x: Tensor) => {\n    // TODO(yassogba) we can remove the chained softplus call here only\n    // after backends have modualrized softplus at which point we can call\n    // engine runKernel(..., Sotfplus, ...) directly.\n    const value = neg(softplus(neg(x)));\n\n    const gradFunc = (dy: T) => {\n      const derX = mul(dy, sigmoid(neg(x)));\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\nexport const logSigmoid = op({logSigmoid_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Max, MaxAttrs, MaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the maximum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.max().print();  // or tf.max(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.max(axis).print();  // or tf.max(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction max_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'max');\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = util.parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let maxInput = $x;\n        if (permutedAxes != null) {\n          maxInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, maxInput.rank);\n        }\n\n        const y = backend.max(maxInput, axes);\n        if (permutedAxes != null) {\n          maxInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape = axis_util.expandShapeToKeepDim(\n              res.shape, util.parseAxisParam(axis, $x.shape));\n          res = reshape(res, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MaxInputs = {x: $x};\n  const attrs: MaxAttrs = {reductionIndices: axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Max,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const max = op({max_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sub, SubInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Subtracts two `tf.Tensor`s element-wise, A - B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n *\n * ```js\n * // Broadcast subtract a with b.\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.scalar(5);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n * @param a The first `tf.Tensor` to subtract from.\n * @param b The second `tf.Tensor` to be subtracted. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction sub_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'sub');\n  let $b = convertToTensor(b, 'b', 'sub');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.subtract($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SubInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sub) as\n      T;\n}\n\nexport const sub = op({sub_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sum, SumAttrs, SumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the sum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If axes has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.sum().print();  // or tf.sum(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.sum(axis).print();  // or tf.sum(x, axis)\n * ```\n *\n * @param x The input tensor to compute the sum over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction sum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'sum');\n  if ($x.dtype === 'bool') {\n    $x = cast($x, 'int32');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.sum(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n    return value;\n  };\n\n  const inputs: SumInputs = {x: $x};\n  const attrs: SumAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const sum = op({sum_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LogSoftmax, LogSoftmaxAttrs, LogSoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log softmax.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param axis The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction logSoftmax_<T extends Tensor>(logits: T|TensorLike, axis = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'logSoftmax');\n\n  if (axis === -1) {\n    axis = $logits.rank - 1;\n  }\n  if (axis !== $logits.rank - 1) {\n    throw Error(\n        'Log Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and axis was ${axis}`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const keepDims = true;\n    const xMax = max(logits, axis, true);\n    const shifted = sub(logits, xMax);\n    const value =\n        sub(cast(shifted, 'float32'), log(sum(exp(shifted), axis, keepDims)));\n    save([value]);\n    return value;\n  };\n\n  const inputs: LogSoftmaxInputs = {logits: $logits};\n  const attrs: LogSoftmaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LogSoftmax, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const logSoftmax = op({logSoftmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {add} from './add';\nimport {expandShapeToKeepDim} from './axis_util';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log(sum(exp(elements across the reduction dimensions)).\n *\n * Reduces the input along the dimensions given in `axis`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.logSumExp().print();  // or tf.logSumExp(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.logSumExp(axis).print();  // or tf.logSumExp(a, axis)\n * ```\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. If null (the default),\n *     reduces all dimensions.\n * @param keepDims If true, retains reduced dimensions with length\n *     of 1. Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction logSumExp_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'logSumExp');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const xMax = max($x, axes, true /* keepDims */);\n  const a = sub($x, xMax);\n  const b = exp(a);\n  const c = sum(b, axes);\n  const d = log(c);\n  const res = add(reshape(xMax, d.shape), d);\n\n  if (keepDims) {\n    const newShape = expandShapeToKeepDim(res.shape, axes);\n    return reshape(res, newShape) as T;\n  }\n  return res as T;\n}\n\nexport const logSumExp = op({logSumExp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {LogicalAnd, LogicalAndInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a AND b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalAnd(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalAnd_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalAnd', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalAnd', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalAndInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalAnd($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalAnd) as T;\n}\n\nexport const logicalAnd = op({logicalAnd_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalNot, LogicalNotInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `NOT x` element-wise.\n *\n * ```js\n * const a = tf.tensor1d([false, true], 'bool');\n *\n * a.logicalNot().print();\n * ```\n *\n * @param x The input tensor. Must be of dtype 'bool'.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalNot_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logicalNot', 'bool');\n  const inputs: LogicalNotInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.logicalNot($x), inputs as {} as NamedTensorMap,\n      null /* grad */, LogicalNot);\n}\n\nexport const logicalNot = op({logicalNot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalOr, LogicalOrInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a OR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalOr(b).print();\n * ```\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalOr_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalOr', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalOr', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalOrInputs = {a: $a, b: $b};\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalOr($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalOr) as T;\n}\nexport const logicalOr = op({logicalOr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {logicalAnd} from './logical_and';\nimport {logicalNot} from './logical_not';\nimport {logicalOr} from './logical_or';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a XOR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalXor(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalXor_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalXor', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalXor', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  // x ^ y = (x | y) & ~(x & y)\n  return logicalAnd(logicalOr(a, b), logicalNot(logicalAnd(a, b)));\n}\n\nexport const logicalXor = op({logicalXor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool, MaxPoolAttrs, MaxPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D max pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction maxPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'maxPool');\n  const dilations = 1;\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in maxPool: input must be rank 4 but got rank ${x4D.rank}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    let y;\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      y = x4D.clone();\n    } else {\n      y = backend.maxPool(x4D, convInfo);\n    }\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: MaxPoolInputs = {x: x4D};\n\n  const attrs: MaxPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const maxPool = op({maxPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {MaxPool3D, MaxPool3DAttrs, MaxPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D max pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.maxPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number = [1, 1, 1],\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'maxPool3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in maxPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in maxPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n    const y = backend.maxPool3d(x5D, convInfo);\n    save([x5D, y]);\n    return y;\n  };\n\n  const inputs: MaxPool3DInputs = {x: x5D};\n\n  const attrs: MaxPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3d = op({maxPool3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the 2D max pooling of an image with Argmax index.\n * The indices in argmax are flattened, so that a maximum value at position `[b,\n * y, x, c]` becomes flattened index: `(y * width + x) * channels + c` if\n * include_batch_in_index is False; `((b * height + y) * width + x) * channels\n * +c` if include_batch_in_index is True.\n *\n * The indices returned are always in `[0, height) x [0, width)` before\n * flattening.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param includeBatchIndex Defaults to False. Whether to include batch\n *    dimension in flattened index of argmax.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPoolWithArgmax_<T extends Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    includeBatchInIndex = false): NamedTensorMap {\n  const $x = convertToTensor(x, 'x', 'maxPoolWithArgmax');\n\n  const inputs: MaxPoolWithArgmaxInputs = {x: $x};\n  const attrs:\n      MaxPoolWithArgmaxAttrs = {filterSize, strides, pad, includeBatchInIndex};\n\n  const result = ENGINE.runKernel(\n                     MaxPoolWithArgmax, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {result: result[0], indexes: result[1]};\n}\n\nexport const maxPoolWithArgmax = op({maxPoolWithArgmax_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeZerosTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0.\n *\n * ```js\n * tf.zeros([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Can\n *     be 'float32', 'int32' or 'bool'. Defaults to 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function zeros<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = zeros(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeZerosTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeOnesTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\nimport {zeros} from './zeros';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1.\n *\n * ```js\n * tf.ones([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n *     'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function ones<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = ones(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeOnesTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam, sizeFromShape} from '../util';\n\nimport {computeOutAndReduceShapes} from './axis_util';\nimport {cast} from './cast';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {ones} from './ones';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sum} from './sum';\n\n/**\n * Computes the mean of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces `x` along the dimensions given in `axis`. Unless `keepDims` is\n * true, the rank of the `tf.Tensor` is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and a `tf.Tensor` with\n * a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.mean().print();  // or tf.mean(a)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.mean(axis).print();  // or tf.mean(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction mean_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'mean');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const shapes = computeOutAndReduceShapes($x.shape, axes);\n  const reduceShape = shapes[1];\n  const reduceSize = sizeFromShape(reduceShape);\n\n  // Use a custom gradient to bypass 2 gradient backprops since mean is used\n  // extremely often.\n  const customOp = customGrad((x: Tensor) => {\n    const reduceSizeScalar = scalar(reduceSize);\n    // Cast if needed.\n    const xReduce = reduceSizeScalar.dtype === x.dtype ?\n        x :\n        cast(x, reduceSizeScalar.dtype);\n    const res = div(xReduce, reduceSizeScalar);\n    const value = sum(res, axis, keepDims);\n\n    const gradFunc = (dy: Tensor) => {\n      const expandedDyShape = x.shape.slice();\n      axes.forEach(axis => {\n        expandedDyShape[axis] = 1;\n      });\n      const expandedDy = reshape(dy, expandedDyShape);\n      const derX = div(mul(expandedDy, ones(x.shape, 'float32')), reduceSize);\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\n\nexport const mean = op({mean_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Min, MinAttrs, MinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the minimum value from the input.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axes`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axes` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.min().print();  // or tf.min(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.min(axis).print();  // or tf.min(x, axis)\n * ```\n *\n * @param x The input Tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction min_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'min');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let minInput = $x;\n        if (permutedAxes != null) {\n          minInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n        }\n\n        const y = backend.min(minInput, axes);\n        if (permutedAxes != null) {\n          minInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape =\n              axis_util.expandShapeToKeepDim(res.shape, origAxes);\n          res = reshape(y, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MinInputs = {x: $x};\n  const attrs: MinAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Min,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const min = op({min_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Minimum, MinimumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the min of a and b (`a < b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `minimumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * ```js\n * // Broadcast minimum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction minimum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'minimum');\n  let $b = convertToTensor(b, 'b', 'minimum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.minimum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MinimumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Minimum) as T;\n}\n\nexport const minimum = op({minimum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Mod, ModInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns the mod of a and b element-wise.\n * `floor(x / y) * y + mod(x, y) = x`\n * Supports broadcasting.\n *\n * We also expose `tf.modStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * ```js\n * // Broadcast a mod b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mod_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mod');\n  let $b = convertToTensor(b, 'b', 'mod');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.mod($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: ModInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Mod) as T;\n}\n\nexport const mod = op({mod_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Computes square of `x` element-wise: `x ^ 2`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.sqrt(2), -1]);\n *\n * x.square().print();  // or tf.square(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction square_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'square');\n  const attrs = {};\n  const inputsToSave = [$x];\n  const outputsToSave: boolean[] = [];\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    return backend.square($x);\n  }, {x: $x}, null /* grad */, 'Square', attrs, inputsToSave, outputsToSave);\n}\n\nexport const square = op({square_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim} from './axis_util';\nimport {cast} from './cast';\nimport {mean} from './mean';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {square} from './square';\nimport {sub} from './sub';\n\n/**\n * Calculates the mean and variance of `x`. The mean and variance are\n * calculated by aggregating the contents of `x` across `axes`. If `x` is\n * 1-D and `axes = [0]` this is just the mean and variance of a vector.\n *\n * @param x The input tensor.\n * @param axis The dimension(s) along with to compute mean and\n *     variance. By default it reduces all dimensions.\n * @param keepDims If true, the moments have the same dimensionality as the\n *     input.\n * @return An object with two keys: `mean` and `variance`.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction moments_(\n    x: Tensor|TensorLike, axis: number|number[] = null,\n    keepDims = false): {mean: Tensor, variance: Tensor} {\n  x = convertToTensor(x, 'x', 'moments');\n  const axes = parseAxisParam(axis, x.shape);\n  const xMean = mean(x, axes, keepDims);\n  let keepDimsShape = xMean.shape;\n  if (!keepDims) {\n    keepDimsShape = expandShapeToKeepDim(xMean.shape, axes);\n  }\n  const devSquared =\n      square(sub(cast(x, 'float32'), reshape(xMean, keepDimsShape)));\n  const variance = mean(devSquared, axes, keepDims);\n  return {mean: xMean, variance};\n}\n\nexport const moments = op({moments_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor, convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * @docalias (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D]\n */\nexport type LSTMCellFunc = {\n  (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D];\n};\n\n/**\n * Computes the next states and outputs of a stack of LSTMCells.\n *\n * Each cell output is used as input to the next cell.\n *\n * Returns `[cellState, cellOutput]`.\n *\n * Derived from tf.contrib.rn.MultiRNNCell.\n *\n * @param lstmCells Array of LSTMCell functions.\n * @param data The input to the cell.\n * @param c Array of previous cell states.\n * @param h Array of previous cell outputs.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction multiRNNCell_(\n    lstmCells: LSTMCellFunc[], data: Tensor2D|TensorLike,\n    c: Array<Tensor2D|TensorLike>,\n    h: Array<Tensor2D|TensorLike>): [Tensor2D[], Tensor2D[]] {\n  const $data = convertToTensor(data, 'data', 'multiRNNCell');\n  const $c = convertToTensorArray(c, 'c', 'multiRNNCell');\n  const $h = convertToTensorArray(h, 'h', 'multiRNNCell');\n\n  let input = $data;\n  const newStates = [];\n  for (let i = 0; i < lstmCells.length; i++) {\n    const output = lstmCells[i](input, $c[i], $h[i]);\n    newStates.push(output[0]);\n    newStates.push(output[1]);\n    input = output[1];\n  }\n  const newC: Tensor2D[] = [];\n  const newH: Tensor2D[] = [];\n  for (let i = 0; i < newStates.length; i += 2) {\n    newC.push(newStates[i]);\n    newH.push(newStates[i + 1]);\n  }\n  return [newC, newH];\n}\nexport const multiRNNCell = op({multiRNNCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a `tf.Tensor` with values drawn from a multinomial distribution.\n *\n * ```js\n * const probs = tf.tensor([.75, .25]);\n * tf.multinomial(probs, 3).print();\n * ```\n *\n * @param logits 1D array with unnormalized log-probabilities, or\n *     2D array of shape `[batchSize, numOutcomes]`. See the `normalized`\n *     parameter.\n * @param numSamples Number of samples to draw for each row slice.\n * @param seed The seed number.\n * @param normalized Whether the provided `logits` are normalized true\n *     probabilities (sum to 1). Defaults to false.\n * @return 1D array of shape `[numSamples]`, or 2D array of shape\n *     `[batchSize, numSamples]`, depending on the rank of the input.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction multinomial_(\n    logits: Tensor1D|Tensor2D|TensorLike, numSamples: number, seed?: number,\n    normalized = false): Tensor1D|Tensor2D {\n  const $logits = convertToTensor(logits, 'logits', 'multinomial');\n  const numOutcomes = $logits.size;\n  const origRank = $logits.rank;\n  if (numOutcomes < 2) {\n    throw new Error(\n        `Error in multinomial: you need at least 2 outcomes, but got ` +\n        `${numOutcomes}.`);\n  }\n  if (origRank > 2) {\n    throw new Error(`Rank of probabilities must be 1 or 2, but is ${origRank}`);\n  }\n  seed = seed || Math.random();\n  const logits2D: Tensor2D =\n      origRank === 1 ? reshape($logits, [1, -1]) : $logits as Tensor2D;\n  const res = ENGINE.runKernelFunc(\n      backend => backend.multinomial(logits2D, normalized, numSamples, seed),\n      {logits2D});\n  // tslint:disable-next-line:no-unnecessary-type-assertion\n  return origRank === 1 ? reshape(res, [res.size]) as Tensor1D : res;\n}\n\nexport const multinomial = op({multinomial_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {NotEqual, NotEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a != b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([0, 2, 3]);\n *\n * a.notEqual(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction notEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'notEqual');\n  let $b = convertToTensor(b, 'b', 'notEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => backend.notEqual($a, $b);\n\n  const inputs: NotEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             NotEqual) as T;\n}\n\nexport const notEqual = op({notEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Real, RealInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the real part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the real part of each element in input considered as a complex number.\n *\n * If the input is real, it simply makes a clone.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.real(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction real_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'real');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.real($input);\n  };\n\n  const inputs: RealInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Real) as T;\n}\n\nexport const real = op({real_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OnesLike, OnesLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {complex} from './complex';\nimport {imag} from './imag';\nimport {op} from './operation';\nimport {real} from './real';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.onesLike(x).print();\n * ```\n * @param x A tensor.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction onesLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'onesLike');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if ($x.dtype === 'complex64') {\n      const r = onesLike(real($x));\n      const i = zerosLike(imag($x));\n      return complex(r, i);\n    }\n\n    return backend.onesLike($x);\n  };\n\n  const inputs: OnesLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             OnesLike) as T;\n}\n\nexport const onesLike = op({onesLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the outer product of two vectors, `v1` and `v2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([3, 4, 5]);\n *\n * tf.outerProduct(a, b).print();\n * ```\n * @param v1 The first vector in the outer product operation.\n * @param v2 The second vector in the outer product operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction outerProduct_(\n    v1: Tensor1D|TensorLike, v2: Tensor1D|TensorLike): Tensor2D {\n  const $v1 = convertToTensor(v1, 'v1', 'outerProduct');\n  const $v2 = convertToTensor(v2, 'v2', 'outerProduct');\n\n  util.assert(\n      $v1.rank === 1 && $v2.rank === 1,\n      () => `Error in outerProduct: inputs must be rank 1, but got ranks ` +\n          `${$v1.rank} and ${$v2.rank}.`);\n\n  const v12D = reshape($v1, [-1, 1]);\n  const v22D = reshape($v2, [1, -1]);\n  return matMul(v12D, v22D) as Tensor2D;\n}\n\nexport const outerProduct = op({outerProduct_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {PadV2, PadV2Attrs, PadV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Pads a `tf.Tensor` with a given value and paddings.\n *\n * This operation currently only implements the `CONSTANT` mode.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `paddings` is of given length.\n *   - `tf.pad1d`\n *   - `tf.pad2d`\n *   - `tf.pad3d`\n *   - `tf.pad4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.pad([[1, 2]]).print();\n * ```\n * @param x The tensor to pad.\n * @param paddings An array of length `R` (the rank of the tensor), where\n * each element is a length-2 tuple of ints `[padBefore, padAfter]`,\n * specifying how much to pad along each dimension of the tensor.\n * @param constantValue The pad value to use. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction pad_<T extends Tensor>(\n    x: T|TensorLike, paddings: Array<[number, number]>, constantValue = 0): T {\n  const $x = convertToTensor(x, 'x', 'pad');\n  if ($x.rank === 0) {\n    throw new Error('pad(scalar) is not defined. Pass non-scalar to pad');\n  }\n  const forward: ForwardFunc<T> = (backend, save) => {\n    save([$x]);\n    return backend.pad($x, paddings, constantValue);\n  };\n\n  const attrs: PadV2Attrs = {paddings, constantValue};\n  const inputs: PadV2Inputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, PadV2,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const pad = op({pad_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor1D` with a given value and paddings. See `pad` for details.\n */\nfunction pad1d_(\n    x: Tensor1D|TensorLike, paddings: [number, number],\n    constantValue = 0): Tensor1D {\n  assert(\n      paddings.length === 2,\n      () => 'Invalid number of paddings. Must be length of 2.');\n  return pad(x, [paddings], constantValue);\n}\n\nexport const pad1d = op({pad1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor2D` with a given value and paddings. See `pad` for details.\n */\nfunction pad2d_(\n    x: Tensor2D|TensorLike, paddings: [[number, number], [number, number]],\n    constantValue = 0): Tensor2D {\n  assert(\n      paddings.length === 2 && paddings[0].length === 2 &&\n          paddings[1].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad2d = op({pad2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor3D` with a given value and paddings. See `pad` for details.\n */\nfunction pad3d_(\n    x: Tensor3D|TensorLike,\n    paddings: [[number, number], [number, number], [number, number]],\n    constantValue = 0): Tensor3D {\n  assert(\n      paddings.length === 3 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad3d = op({pad3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor4D` with a given value and paddings. See `pad` for details.\n */\nfunction pad4d_(\n    x: Tensor4D|TensorLike,\n    paddings:\n        [\n          [number, number], [number, number], [number, number], [number, number]\n        ],\n    constantValue = 0): Tensor4D {\n  assert(\n      paddings.length === 4 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2 &&\n          paddings[3].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad4d = op({pad4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation divides \"spatial\" dimensions `[1, ..., M]` of the input into\n * a grid of blocks of shape `blockShape`, and interleaves these blocks with\n * the \"batch\" dimension (0) such that in the output, the spatial\n * dimensions `[1, ..., M]` correspond to the position within the grid,\n * and the batch dimension combines both the position within a spatial block\n * and the original batch position. Prior to division into blocks,\n * the spatial dimensions of the input are optionally zero padded\n * according to `paddings`. See below for a precise description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);\n * const blockShape = [2, 2];\n * const paddings = [[0, 0], [0, 0]];\n *\n * x.spaceToBatchND(blockShape, paddings).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param paddings A 2-D array. Must have shape `[M, 2]`, all values must be >=\n *     0. `paddings[i] = [padStart, padEnd]` specifies the amount to zero-pad\n * from input dimension `i + 1`, which corresponds to spatial dimension `i`. It\n * is required that\n * `(inputShape[i + 1] + padStart + padEnd) % blockShape[i] === 0`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Zero-pad the start and end of dimensions `[1, ..., M]` of the input\n * according to `paddings` to produce `padded` of shape paddedShape.\n *\n * 2. Reshape `padded` to `reshapedPadded` of shape:\n * `[batch] + [paddedShape[1] / blockShape[0], blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1], blockShape[M-1]] + remainingShape`\n *\n * 3. Permute dimensions of `reshapedPadded` to produce `permutedReshapedPadded`\n * of shape: `blockShape + [batch] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * 4. Reshape `permutedReshapedPadded` to flatten `blockShape` into the\n * batch dimension, producing an output tensor of shape:\n * `[batch * prod(blockShape)] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction spaceToBatchND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], paddings: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'spaceToBatchND');\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank ${$x.rank} should be > than [blockShape] ${\n          blockShape.length}`);\n\n  util.assert(\n      paddings.length === blockShape.length,\n      () => `paddings.shape[0] ${\n          paddings.length} must be equal to [blockShape] ${blockShape.length}`);\n\n  util.assert(\n      $x.shape.reduce(\n          (a, b, i) => {\n            if (i > 0 && i <= blockShape.length) {\n              return a &&\n                  ((b + paddings[i - 1][0] + paddings[i - 1][1]) %\n                       blockShape[i - 1] ===\n                   0);\n            }\n            return a;\n          },\n          true),\n      () => `input spatial dimensions ${$x.shape.slice(1)} with paddings ${\n          paddings.toString()} must be divisible by blockShapes ${\n          blockShape.toString()}`);\n\n  const forward: ForwardFunc<T> = backend =>\n      backend.spaceToBatchND($x, blockShape, paddings);\n\n  const inputs: SpaceToBatchNDInputs = {x: $x};\n  const attrs: SpaceToBatchNDAttrs = {blockShape, paddings};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      SpaceToBatchND, attrs as {} as NamedAttrMap);\n}\n\nexport const spaceToBatchND = op({spaceToBatchND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {avgPool} from './avg_pool';\nimport {batchToSpaceND} from './batch_to_space_nd';\nimport * as conv_util from './conv_util';\nimport {maxPool} from './max_pool';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {spaceToBatchND} from './space_to_batch_nd';\n\n/**\n * Performs an N-D pooling operation\n *\n * @param input The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param windowShape The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param poolingType The type of pooling, either 'max' or 'avg'.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilationRate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction pool_<T extends Tensor3D|Tensor4D>(\n    input: T|TensorLike, windowShape: [number, number]|number,\n    poolingType: 'avg'|'max', pad: 'valid'|'same'|number,\n    dilations?: [number, number]|number, strides?: [number, number]|number) {\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  if (strides == null) {\n    strides = 1;\n  }\n  if (pad === 0) {\n    pad = 'valid';\n  }\n\n  const $x = convertToTensor(input, 'x', 'maxPool');\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in pool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = conv_util.computePool2DInfo(\n      x4D.shape, windowShape, strides, dilations, pad);\n  const dilation: [number, number] =\n      [convInfo.dilationHeight, convInfo.dilationWidth];\n\n  // The following implementation does batchToSpace(pool(spaceToBatch(x)))\n  // whenever dilation > 1 since the TF kernels do not support dilation > 1.\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L1037\n\n  let basePadding: number[][];\n  if (pad === 'same') {\n    basePadding = withSpaceToBatchBasePaddings(\n        [convInfo.filterHeight, convInfo.filterWidth], dilation);\n  } else {\n    basePadding = [[0, 0], [0, 0]];\n  }\n\n  const isDilationOne = dilation[0] === 1 && dilation[1] === 1;\n  const [adjustedPadding, adjustedCrops] = requiredSpaceToBatchPaddings(\n      [convInfo.inHeight, convInfo.inWidth], dilation, basePadding);\n  const convertedPad = isDilationOne ? pad : 'valid';\n  const convertedX =\n      isDilationOne ? x4D : spaceToBatchND(x4D, dilation, adjustedPadding);\n\n  const forwardOp = poolingType === 'avg' ?\n      () => avgPool(convertedX, windowShape, strides, convertedPad) :\n      () => maxPool(convertedX, windowShape, strides, convertedPad);\n  const y = forwardOp();\n\n  const res = isDilationOne ? y : batchToSpaceND(y, dilation, adjustedCrops);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\n// Helper function to compute crops and paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/array_ops.py#L2184\nfunction requiredSpaceToBatchPaddings(\n    inputShape: [number, number], blockShape: [number, number],\n    basePadding: number[][]) {\n  const padStart = basePadding.map(b => b[0]);\n  const origPadEnd = basePadding.map(b => b[1]);\n  const fullInputShape = inputShape.concat(padStart, origPadEnd);\n  const padEndExtra = blockShape.map((b, i) => (b - fullInputShape[i] % b) % b);\n  const padEnd = origPadEnd.map((s, i) => s + padEndExtra[i]);\n  const paddings = blockShape.map((_, i) => [padStart[i], padEnd[i]]);\n  const crops = blockShape.map((_, i) => [0, padEndExtra[i]]);\n  return [paddings, crops];\n}\n\n// Helper function to compute base paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L524\nfunction withSpaceToBatchBasePaddings(\n    filterShape: [number, number], dilation: [number, number]) {\n  // Spatial dimensions of the filters and the upsampled filters in which we\n  // introduce (rate - 1) zeros between consecutive filter values.\n  const dilatedFilterShape = filterShape.map((s, i) => {\n    return s + (s - 1) * (dilation[i] - 1);\n  });\n  const padExtraShape = dilatedFilterShape.map(s => s - 1);\n\n  // When padding is odd, we pad more at end, following the same\n  // convention as conv2d.\n  const padExtraStart = padExtraShape.map(s => Math.floor(s / 2));\n  const padExtraEnd = padExtraShape.map((s, i) => s - padExtraStart[i]);\n  return padExtraShape.map((_, i) => {\n    return [padExtraStart[i], padExtraEnd[i]];\n  });\n}\n\nexport const pool = op({pool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Pow, PowInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the power of one `tf.Tensor` to another. Supports broadcasting.\n *\n * Given a `tf.Tensor` x and a `tf.Tensor` y, this operation computes x^y for\n * corresponding elements in x and y. The result's dtype will be the upcasted\n * type of the `base` and `exp` dtypes.\n *\n * ```js\n * const a = tf.tensor([[2, 3], [4, 5]])\n * const b = tf.tensor([[1, 2], [3, 0]]).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n *\n * ```js\n * const a = tf.tensor([[1, 2], [3, 4]])\n * const b = tf.tensor(2).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n * We also expose `powStrict` which has the same signature as this op and\n * asserts that `base` and `exp` are the same shape (does not broadcast).\n *\n * @param base The base `tf.Tensor` to pow element-wise.\n * @param exp The exponent `tf.Tensor` to pow element-wise.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction pow_<T extends Tensor>(\n    base: Tensor|TensorLike, exp: Tensor|TensorLike): T {\n  let $base = convertToTensor(base, 'base', 'pow');\n  let $exp = convertToTensor(exp, 'exp', 'pow');\n  [$base, $exp] = makeTypesMatch($base, $exp);\n\n  const inputs: PowInputs = {a: $base, b: $exp};\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.pow($base, $exp);\n    save([$base, $exp, y]);\n    return y;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Pow) as T;\n}\n\nexport const pow = op({pow_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prelu, PreluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes leaky rectified linear element-wise with parametric alphas.\n *\n * `x < 0 ? alpha * x : f(x) = x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n * const alpha = tf.scalar(0.1);\n *\n * x.prelu(alpha).print();  // or tf.prelu(x, alpha)\n * ```\n * @param x The input tensor.\n * @param alpha Scaling factor for negative values.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction prelu_<T extends Tensor>(x: T|TensorLike, alpha: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'prelu');\n  const $alpha = convertToTensor(alpha, 'alpha', 'prelu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.prelu($x, $alpha);\n    save([$x, $alpha]);\n    return res;\n  };\n\n  const inputs: PreluInputs = {x: $x, alpha: $alpha};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prelu) as\n      T;\n}\n\nexport const prelu = op({prelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prod, ProdAttrs, ProdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the product of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.prod().print();  // or tf.prod(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.prod(axis).print();  // or tf.prod(x, axis)\n * ```\n *\n * @param x The input tensor to compute the product over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction prod_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'prod');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if ($x.dtype === 'bool') {\n      $x = cast($x, 'int32');\n    }\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.prod(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n\n    return value as T;\n  };\n\n  const inputs: ProdInputs = {x: $x};\n  const attrs: ProdAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prod,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const prod = op({prod_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a random number generator\n * function defined by the user.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param randFunction A random number generator function which is called\n * for each element in the output tensor.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n */\nfunction rand_<R extends Rank>(\n    shape: ShapeMap[R], randFunction: () => number,\n    dtype?: DataType): Tensor<R> {\n  const size = sizeFromShape(shape);\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  for (let i = 0; i < size; i++) {\n    values[i] = randFunction();\n  }\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n\nexport const rand = op({rand_});\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as seedrandom from 'seedrandom';\n\nimport {expectNumbersClose, testEpsilon} from '../test_util';\nimport {TypedArray} from '../types';\n\nexport interface RandomBase {\n  nextValue(): number;\n}\n\nexport interface RandomGamma {\n  nextValue(): number;\n}\n\nexport interface RandNormalDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\nexport interface RandGammaDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\n// https://en.wikipedia.org/wiki/Marsaglia_polar_method\nexport class MPRandGauss implements RandomBase {\n  private mean: number;\n  private stdDev: number;\n  private nextVal: number;\n  private dtype?: keyof RandNormalDataTypes;\n  private truncated?: boolean;\n  private upper?: number;\n  private lower?: number;\n  private random: seedrandom.prng;\n\n  constructor(\n      mean: number, stdDeviation: number, dtype?: keyof RandNormalDataTypes,\n      truncated?: boolean, seed?: number) {\n    this.mean = mean;\n    this.stdDev = stdDeviation;\n    this.dtype = dtype;\n    this.nextVal = NaN;\n    this.truncated = truncated;\n    if (this.truncated) {\n      this.upper = this.mean + this.stdDev * 2;\n      this.lower = this.mean - this.stdDev * 2;\n    }\n    const seedValue = seed ? seed : Math.random();\n    this.random = seedrandom.alea(seedValue.toString());\n  }\n\n  /** Returns next sample from a Gaussian distribution. */\n  public nextValue(): number {\n    if (!isNaN(this.nextVal)) {\n      const value = this.nextVal;\n      this.nextVal = NaN;\n      return value;\n    }\n\n    let resultX: number, resultY: number;\n    let isValid = false;\n    while (!isValid) {\n      let v1: number, v2: number, s: number;\n      do {\n        v1 = 2 * this.random() - 1;\n        v2 = 2 * this.random() - 1;\n        s = v1 * v1 + v2 * v2;\n      } while (s >= 1 || s === 0);\n\n      const mul = Math.sqrt(-2.0 * Math.log(s) / s);\n      resultX = this.mean + this.stdDev * v1 * mul;\n      resultY = this.mean + this.stdDev * v2 * mul;\n\n      if (!this.truncated || this.isValidTruncated(resultX)) {\n        isValid = true;\n      }\n    }\n\n    if (!this.truncated || this.isValidTruncated(resultY)) {\n      this.nextVal = this.convertValue(resultY);\n    }\n    return this.convertValue(resultX);\n  }\n\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype == null || this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  /** Returns true if less than 2-standard-deviations from the mean. */\n  private isValidTruncated(value: number): boolean {\n    return value <= this.upper && value >= this.lower;\n  }\n}\n\n// Marsaglia, George, and Wai Wan Tsang. 2000. \"A Simple Method for Generating\n// Gamma Variables.\"\nexport class RandGamma implements RandomGamma {\n  private alpha: number;\n  private beta: number;\n  private d: number;\n  private c: number;\n  private dtype?: keyof RandGammaDataTypes;\n  private randu: seedrandom.prng;\n  private randn: MPRandGauss;\n\n  constructor(\n      alpha: number, beta: number, dtype: keyof RandGammaDataTypes,\n      seed?: number) {\n    this.alpha = alpha;\n    this.beta = 1 / beta;  // convert rate to scale parameter\n    this.dtype = dtype;\n\n    const seedValue = seed ? seed : Math.random();\n    this.randu = seedrandom.alea(seedValue.toString());\n    this.randn = new MPRandGauss(0, 1, dtype, false, this.randu());\n\n    if (alpha < 1) {\n      this.d = alpha + (2 / 3);\n    } else {\n      this.d = alpha - (1 / 3);\n    }\n    this.c = 1 / Math.sqrt(9 * this.d);\n  }\n\n  /** Returns next sample from a gamma distribution. */\n  public nextValue(): number {\n    let x2: number, v0: number, v1: number, x: number, u: number, v: number;\n    while (true) {\n      do {\n        x = this.randn.nextValue();\n        v = 1 + (this.c * x);\n      } while (v <= 0);\n      v *= v * v;\n      x2 = x * x;\n      v0 = 1 - (0.331 * x2 * x2);\n      v1 = (0.5 * x2) + (this.d * (1 - v + Math.log(v)));\n      u = this.randu();\n      if (u < v0 || Math.log(u) < v1) {\n        break;\n      }\n    }\n    v = (1 / this.beta) * this.d * v;\n    if (this.alpha < 1) {\n      v *= Math.pow(this.randu(), 1 / this.alpha);\n    }\n    return this.convertValue(v);\n  }\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n}\n\nexport class UniformRandom implements RandomBase {\n  private min: number;\n  private range: number;\n  private random: seedrandom.prng;\n  private dtype?: keyof RandNormalDataTypes;\n\n  constructor(\n      min = 0, max = 1, dtype?: keyof RandNormalDataTypes,\n      seed?: string|number) {\n    this.min = min;\n    this.range = max - min;\n    this.dtype = dtype;\n    if (seed == null) {\n      seed = Math.random();\n    }\n    if (typeof seed === 'number') {\n      seed = seed.toString();\n    }\n\n    if (!this.canReturnFloat() && this.range <= 1) {\n      throw new Error(\n          `The difference between ${min} - ${max} <= 1 and dtype is not float`);\n    }\n    this.random = seedrandom.alea(seed);\n  }\n\n  /** Handles proper rounding for non floating point numbers. */\n  private canReturnFloat = () =>\n      (this.dtype == null || this.dtype === 'float32');\n\n  private convertValue(value: number): number {\n    if (this.canReturnFloat()) {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  nextValue() {\n    return this.convertValue(this.min + this.range * this.random());\n  }\n}\n\nexport function jarqueBeraNormalityTest(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test\n  const n = values.length;\n  const s = skewness(values);\n  const k = kurtosis(values);\n  const jb = n / 6 * (Math.pow(s, 2) + 0.25 * Math.pow(k - 3, 2));\n  // JB test requires 2-degress of freedom from Chi-Square @ 0.95:\n  // http://www.itl.nist.gov/div898/handbook/eda/section3/eda3674.htm\n  const CHI_SQUARE_2DEG = 5.991;\n  if (jb > CHI_SQUARE_2DEG) {\n    throw new Error(`Invalid p-value for JB: ${jb}`);\n  }\n}\n\nexport function expectArrayInMeanStdRange(\n    actual: TypedArray|number[], expectedMean: number, expectedStdDev: number,\n    epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  const actualMean = mean(actual);\n  expectNumbersClose(actualMean, expectedMean, epsilon);\n  expectNumbersClose(\n      standardDeviation(actual, actualMean), expectedStdDev, epsilon);\n}\n\nfunction mean(values: TypedArray|number[]) {\n  let sum = 0;\n  for (let i = 0; i < values.length; i++) {\n    sum += values[i];\n  }\n  return sum / values.length;\n}\n\nfunction standardDeviation(values: TypedArray|number[], mean: number) {\n  let squareDiffSum = 0;\n  for (let i = 0; i < values.length; i++) {\n    const diff = values[i] - mean;\n    squareDiffSum += diff * diff;\n  }\n  return Math.sqrt(squareDiffSum / values.length);\n}\n\nfunction kurtosis(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Kurtosis\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum4 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum4 += Math.pow(v, 4);\n  }\n  return (1 / n) * sum4 / Math.pow((1 / n) * sum2, 2);\n}\n\nfunction skewness(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Skewness\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum3 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum3 += Math.pow(v, 3);\n  }\n  return (1 / n) * sum3 / Math.pow((1 / (n - 1)) * sum2, 3 / 2);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {RandGamma} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a gamma distribution.\n *\n * ```js\n * tf.randomGamma([2, 2], 1).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param alpha The shape parameter of the gamma distribution.\n * @param beta The inverse scale parameter of the gamma distribution. Defaults\n *     to 1.\n * @param dtype The data type of the output. Defaults to float32.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomGamma_<R extends Rank>(\n    shape: ShapeMap[R], alpha: number, beta = 1,\n    dtype: 'float32'|'int32' = 'float32', seed?: number): Tensor<R> {\n  if (beta == null) {\n    beta = 1;\n  }\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const rgamma = new RandGamma(alpha, beta, dtype, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = rgamma.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomGamma = op({randomGamma_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a normal distribution.\n *\n * ```js\n * tf.randomNormal([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, false /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomNormal = op({randomNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {UniformRandom} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a uniform distribution.\n *\n * The generated values follow a uniform distribution in the range [minval,\n * maxval). The lower bound minval is included in the range, while the upper\n * bound maxval is excluded.\n *\n * ```js\n * tf.randomUniform([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param minval The lower bound on the range of random values to generate.\n *   Defaults to 0.\n * @param maxval The upper bound on the range of random values to generate.\n *   Defaults to 1.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomUniform_<R extends Rank>(\n    shape: ShapeMap[R], minval = 0, maxval = 1, dtype: DataType = 'float32',\n    seed?: number|string): Tensor<R> {\n  const res = buffer(shape, dtype);\n  const random = new UniformRandom(minval, maxval, null, seed);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = random.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomUniform = op({randomUniform_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike1D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-1 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor1d` as it makes the code more readable.\n *\n * ```js\n * tf.tensor1d([1, 2, 3]).print();\n * ```\n *\n * @param values The values of the tensor. Can be array of numbers,\n *     or a `TypedArray`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor1d(values: TensorLike1D, dtype?: DataType): Tensor1D {\n  assertNonNull(values);\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 1) {\n    throw new Error('tensor1d() requires values to be a flat/TypedArray');\n  }\n  const shape: number[] = null;\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Range, RangeAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {makeZerosTypedArray} from '../util';\n\nimport {tensor1d} from './tensor1d';\nimport {zeros} from './zeros';\n\n/**\n * Creates a new `tf.Tensor1D` filled with the numbers in the range provided.\n *\n * The tensor is a is half-open interval meaning it includes start, but\n * excludes stop. Decrementing ranges and negative step values are also\n * supported.sv\n *\n *\n * ```js\n * tf.range(0, 9, 2).print();\n * ```\n *\n * @param start An integer start value\n * @param stop An integer stop value\n * @param step An integer increment (will default to 1 or -1)\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function range(\n    start: number, stop: number, step = 1,\n    dtype: 'float32'|'int32' = 'float32'): Tensor1D {\n  if (step === 0) {\n    throw new Error('Cannot have a step of zero');\n  }\n\n  const forward: ForwardFunc<Tensor> = () => {\n    const sameStartStop = start === stop;\n    const increasingRangeNegativeStep = start < stop && step < 0;\n    const decreasingRangePositiveStep = stop < start && step > 1;\n\n    if (sameStartStop || increasingRangeNegativeStep ||\n        decreasingRangePositiveStep) {\n      return zeros([0], dtype);\n    }\n\n    const numElements = Math.abs(Math.ceil((stop - start) / step));\n    const values = makeZerosTypedArray(numElements, dtype);\n\n    if (stop < start && step === 1) {\n      // Auto adjust the step's sign if it hasn't been set\n      // (or was set to 1)\n      step = -1;\n    }\n\n    values[0] = start;\n    for (let i = 1; i < values.length; i++) {\n      values[i] = values[i - 1] + step;\n    }\n\n    return tensor1d(values, dtype);\n  };\n\n  const attrs: RangeAttrs = {start, stop, step, dtype};\n\n  return ENGINE.runKernelFunc(\n             forward, {} /* inputs */, null /* grad */, Range,\n             attrs as {} as NamedAttrMap) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Reciprocal, ReciprocalInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of x element-wise: `1 / x`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, 2]);\n *\n * x.reciprocal().print();  // or tf.reciprocal(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction reciprocal_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'reciprocal');\n\n  const inputs: ReciprocalInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.reciprocal($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Reciprocal);\n}\nexport const reciprocal = op({reciprocal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu, ReluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport { cast } from './cast';\n\n/**\n * Computes rectified linear element-wise: `max(x, 0)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.relu().print();  // or tf.relu(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu($x);\n  };\n\n  const inputs: ReluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu) as\n      T;\n}\n\nexport const relu = op({relu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu6, Relu6Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes rectified linear 6 element-wise: `min(max(x, 0), 6)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 8]);\n *\n * x.relu6().print();  // or tf.relu6(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu6_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu6');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu6($x);\n  };\n\n  const inputs: Relu6Inputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu6) as\n      T;\n}\n\nexport const relu6 = op({relu6_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reverse, ReverseAttrs, ReverseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Reverses a `tf.Tensor` along a specified axis.\n *\n * Also available are stricter rank-specific methods that assert that `x` is\n * of the given rank:\n *   - `tf.reverse1d`\n *   - `tf.reverse2d`\n *   - `tf.reverse3d`\n *   - `tf.reverse4d`\n *\n * Except `tf.reverse1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.reverse().print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.reverse(axis).print();\n * ```\n * @param x The input tensor to be reversed.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction reverse_<T extends Tensor>(\n    x: T|TensorLike, axis?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'reverse');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const axes = parseAxisParam(axis, $x.shape);\n    if ($x.rank === 0) {\n      return clone($x);\n    }\n    const res = backend.reverse($x, axes);\n    return reshape(res, $x.shape);\n  };\n\n  const inputs: ReverseInputs = {x: $x};\n  const attrs: ReverseAttrs = {dims: axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Reverse, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const reverse = op({reverse_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor1D`.\n *\n * @param x The input tensor.\n */\nfunction reverse1d_(x: Tensor1D|TensorLike): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 1,\n      () => `Error in reverse1D: x must be rank 1 but got rank ${$x.rank}.`);\n  return reverse($x, 0);\n}\n\nexport const reverse1d = op({reverse1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor2D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse2d_(x: Tensor2D|TensorLike, axis?: number|number[]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 2,\n      () => `Error in reverse2D: x must be rank 2 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse2d = op({reverse2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor3D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse3d_(x: Tensor3D|TensorLike, axis?: number|number[]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 3,\n      () => `Error in reverse3D: x must be rank 3 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse3d = op({reverse3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor4D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse4d_(x: Tensor4D|TensorLike, axis?: number|number[]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 4,\n      () => `Error in reverse4D: x must be rank 4 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse4d = op({reverse4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Round, RoundInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes round of input `tf.Tensor` element-wise: `round(x)`.\n * It implements banker's rounding.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.round().print();  // or tf.round(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction round_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'round');\n  const inputs: RoundInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.round($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Round);\n}\n\nexport const round = op({round_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Rsqrt, RsqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of square root of the input `tf.Tensor` element-wise:\n * `y = 1 / sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.rsqrt().print();  // or tf.rsqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction rsqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'rsqrt');\n\n  const inputs: RsqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.rsqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Rsqrt);\n}\nexport const rsqrt = op({rsqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Selu, SeluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes scaled exponential linear element-wise.\n *\n * `x < 0 ? scale * alpha * (exp(x) - 1) : x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.selu().print();  // or tf.selu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction selu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'selu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.selu($x);\n    save([$x]);\n    return res;\n  };\n\n  const inputs: SeluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Selu) as\n      T;\n}\n\nexport const selu = op({selu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport {depthwiseConv2d} from './depthwise_conv2d';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * 2-D convolution with separable filters.\n *\n * Performs a depthwise convolution that acts separately on channels followed\n * by a pointwise convolution that mixes channels. Note that this is\n * separability between dimensions [1, 2] and 3, not spatial separability\n * between dimensions 1 and 2.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param depthwiseFilter The depthwise filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`. This is\n *     the filter used in the first step.\n * @param pointwiseFilter The pointwise filter tensor, rank 4, of shape\n *     `[1, 1, inChannels * channelMultiplier, outChannels]`. This is\n *     the filter used in the second step.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction separableConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthwiseFilter: Tensor4D|TensorLike,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'separableConv2d');\n  const $depthwiseFilter =\n      convertToTensor(depthwiseFilter, 'depthwiseFilter', 'separableConv2d');\n  const $pointwiseFilter =\n      convertToTensor(pointwiseFilter, 'pointwiseFilter', 'separableConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  if (dataFormat === 'NCHW') {\n    throw new Error(\n        'separableConv2d currently does not support dataFormat NCHW; only ' +\n        'NHWC is supported');\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in separableConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $depthwiseFilter.rank === 4,\n      () => `Error in separableConv2d: depthwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.rank === 4,\n      () => `Error in separableConv2d: pointwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.shape[0] === 1,\n      () =>\n          `Error in separableConv2d: the first dimension of pointwise filter ` +\n          ` must be 1, but got ${$pointwiseFilter.shape[0]}.`);\n  util.assert(\n      $pointwiseFilter.shape[1] === 1,\n      () => `Error in separableConv2d: the second dimension of pointwise ` +\n          `filter must be 1, but got ${$pointwiseFilter.shape[1]}.`);\n\n  const inChannels = $depthwiseFilter.shape[2];\n  const channelMultiplier = $depthwiseFilter.shape[3];\n  util.assert(\n      $pointwiseFilter.shape[2] === inChannels * channelMultiplier,\n      () =>\n          `Error in separableConv2d: the third dimension of pointwise filter ` +\n          `must be ${inChannels * channelMultiplier}, ` +\n          `but got ${$pointwiseFilter.shape[2]}.`);\n\n  const depthwise = depthwiseConv2d(\n      x4D, $depthwiseFilter, strides, pad, dataFormat, dilation);\n  const pointwiseStride = 1;\n  const res =\n      conv2d(depthwise, $pointwiseFilter, pointwiseStride, 'valid', dataFormat);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const separableConv2d = op({separableConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\n/**\n * Computes the difference between two lists of numbers.\n *\n * Given a Tensor `x` and a Tensor `y`, this operation returns a Tensor `out`\n * that represents all values that are in `x` but not in `y`. The returned\n * Tensor `out` is sorted in the same order that the numbers appear in `x`\n * (duplicates are preserved). This operation also returns a Tensor indices that\n * represents the position of each out element in `x`. In other words:\n *\n * `out[i] = x[idx[i]] for i in [0, 1, ..., out.length - 1]`\n *\n * ```js\n * const x = [1, 2, 3, 4, 5, 6];\n * const y = [1, 3, 5];\n *\n * const [out, indices] = await tf.setdiff1dAsync(x, y);\n * out.print(); // [2, 4, 6]\n * indices.print(); // [1, 3, 5]\n * ```\n *\n * @param x 1-D Tensor. Values to keep.\n * @param y 1-D Tensor. Must have the same type as x. Values to exclude in the\n *     output.\n * @returns Promise of Tensor tuple [out, indices].\n *  out: Tensor with the same type as x.\n *  indices: A Tensor of type int32.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nasync function setdiff1dAsync_(\n    x: Tensor|TensorLike, y: Tensor|TensorLike): Promise<[Tensor, Tensor]> {\n  const $x = convertToTensor(x, 'x', 'setdiff1d');\n  const $y = convertToTensor(y, 'y', 'setdiff1d');\n\n  util.assert(\n      $x.dtype === $y.dtype,\n      () => `x and y should have the same dtype, but got x (${\n          $x.dtype}) and y (${$y.dtype}).`);\n\n  util.assert(\n      $x.rank === 1, () => `x should be 1D tensor, but got x (${$x.shape}).`);\n\n  util.assert(\n      $y.rank === 1, () => `y should be 1D tensor, but got y (${$y.shape}).`);\n\n  const xVals = await $x.data();\n  const yVals = await $y.data();\n  const ySet = new Set(yVals);\n\n  let outputSize = 0;\n  for (let i = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      outputSize++;\n    }\n  }\n\n  const buffer = new TensorBuffer([outputSize], $x.dtype);\n  const indices = new TensorBuffer([outputSize], 'int32');\n  for (let i = 0, p = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      buffer.values[p] = xVals[i];\n      indices.values[p] = i;\n      p++;\n    }\n  }\n  return [buffer.toTensor(), indices.toTensor()];\n}\nexport const setdiff1dAsync = setdiff1dAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sign, SignInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns an element-wise indication of the sign of a number.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3, NaN, 0]);\n *\n * x.sign().print();  // or tf.sign(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sign_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sign');\n  const inputs: SignInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.sign($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Sign);\n}\nexport const sign = op({sign_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sin, SinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sin of the input Tensor element-wise: `sin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.sin().print();  // or tf.sin(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sin');\n\n  const inputs: SinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sin);\n}\nexport const sin = op({sin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sinh, SinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic sin of the input `tf.Tensor` element-wise: `sinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.sinh().print();  // or tf.sinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sinh');\n  const inputs: SinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sinh);\n}\nexport const sinh = op({sinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 1D slice from 1D array starting at coordinates `begin` and is\n * of length `size`. See `slice` for details.\n */\nfunction slice1d_(\n    x: Tensor1D|TensorLike, begin: number, size: number): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'slice1d');\n  util.assert(\n      $x.rank === 1,\n      () =>\n          `slice1d expects a rank-1 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, [begin], [size]);\n}\nexport const slice1d = op({slice1d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 2D slice from a 2D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice2d_(\n    x: Tensor2D|TensorLike, begin: [number, number],\n    size: [number, number]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'slice2d');\n  util.assert(\n      $x.rank === 2,\n      () =>\n          `slice2d expects a rank-2 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice2d = op({slice2d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 3D slice from a 3D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice3d_(\n    x: Tensor3D|TensorLike, begin: [number, number, number],\n    size: [number, number, number]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'slice3d');\n  util.assert(\n      $x.rank === 3,\n      () =>\n          `slice3d expects a rank-3 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice3d = op({slice3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 4D slice from a 4D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice4d_(\n    x: Tensor4D|TensorLike, begin: [number, number, number, number],\n    size: [number, number, number, number]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'slice4d');\n  util.assert(\n      $x.rank === 4,\n      () =>\n          `slice4d expects a rank-4 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice4d = op({slice4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softmax, SoftmaxAttrs, SoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the softmax normalized vector given the logits.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction softmax_<T extends Tensor>(logits: T|TensorLike, dim = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'softmax', 'float32');\n\n  if (dim === -1) {\n    dim = $logits.rank - 1;\n  }\n  if (dim !== $logits.rank - 1) {\n    throw Error(\n        'Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and dim was ${dim}`);\n  }\n\n  const inputs: SoftmaxInputs = {logits: $logits};\n  const attrs: SoftmaxAttrs = {dim};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const y = backend.softmax($logits, dim);\n        save([y]);\n        return y;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, Softmax,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const softmax = op({softmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FFT, FFTInputs} from '../../kernel_names';\nimport {Tensor} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\n\n/**\n * Fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the inner-most\n * dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.fft().print();  // tf.spectral.fft(x).print();\n * ```\n * @param input The complex input to compute an fft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction fft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.fft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: FFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D = input.as2D(batch, innerDimensionSize);\n    const result = backend.fft(input2D);\n    return result.reshape(input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, FFT);\n}\n\nexport const fft = op({fft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {IFFT, IFFTInputs} from '../../kernel_names';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Inverse fast Fourier transform.\n *\n * Computes the inverse 1-dimensional discrete Fourier transform over the\n * inner-most dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.ifft().print();  // tf.spectral.ifft(x).print();\n * ```\n * @param input The complex input to compute an ifft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction ifft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.ifft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: IFFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D: Tensor2D = reshape(input, [batch, innerDimensionSize]);\n    const result = backend.ifft(input2D);\n    return reshape(result, input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, IFFT);\n}\n\nexport const ifft = op({ifft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {reverse} from '../reverse';\nimport {scalar} from '../scalar';\nimport {slice} from '../slice';\n\nimport {ifft} from './ifft';\n\n/**\n * Inversed real value input fast Fourier transform.\n *\n * Computes the 1-dimensional inversed discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([0, 0, 0]);\n * const x = tf.complex(real, imag);\n *\n * x.irfft().print();\n * ```\n * @param input The real value input to compute an irfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction irfft_(input: Tensor): Tensor {\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n  let ret: Tensor;\n  if (innerDimensionSize <= 2) {\n    const complexInput = reshape(input, [batch, innerDimensionSize]);\n    ret = ifft(complexInput);\n  } else {\n    // The length of unique components of the DFT of a real-valued signal\n    // is 2 * (input_len - 1)\n    const outputShape = [batch, 2 * (innerDimensionSize - 1)];\n    const realInput = reshape(real(input), [batch, innerDimensionSize]);\n    const imagInput = reshape(imag(input), [batch, innerDimensionSize]);\n\n    const realConjugate =\n        reverse(slice(realInput, [0, 1], [batch, innerDimensionSize - 2]), 1);\n    const imagConjugate: Tensor2D = mul(\n        reverse(slice(imagInput, [0, 1], [batch, innerDimensionSize - 2]), 1),\n        scalar(-1));\n\n    const r = concat([realInput, realConjugate], 1);\n    const i = concat([imagInput, imagConjugate], 1);\n    const complexInput =\n        reshape(complex(r, i), [outputShape[0], outputShape[1]]);\n    ret = ifft(complexInput);\n  }\n  ret = real(ret);\n  // reshape the result if the input is 3D tensor.\n  if (input.rank === 3 && input.shape[0] !== 0) {\n    const temp = ret;\n    const batch = input.shape[0];\n    ret = reshape(ret, [batch, ret.shape[0] / batch, ret.shape[1]]);\n    temp.dispose();\n  }\n  return ret;\n}\n\nexport const irfft = op({irfft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {assert} from '../util';\n\n/**\n * Prepare the split size array. When the input is a number, the axis is evenly\n * divided among the split size. When the input contains the negative value, the\n * rest of the axis is allocated toward that.\n */\nexport function prepareSplitSize(\n    x: Tensor|TensorInfo, numOrSizeSplits: number[]|number,\n    axis = 0): number[] {\n  let splitSizes = [];\n  if (typeof (numOrSizeSplits) === 'number') {\n    assert(\n        x.shape[axis] % numOrSizeSplits === 0,\n        () => 'Number of splits must evenly divide the axis.');\n    splitSizes =\n        new Array(numOrSizeSplits).fill(x.shape[axis] / numOrSizeSplits);\n  } else {\n    const numOfNegs = numOrSizeSplits.reduce((count, value) => {\n      if (value === -1) {\n        count += 1;\n      }\n      return count;\n    }, 0);\n    assert(\n        numOfNegs <= 1,\n        () => 'There should be only one negative value in split array.');\n    const negIndex = numOrSizeSplits.indexOf(-1);\n    // Allow the number of split array to be -1, which indicates the rest\n    // of dimension is allocated to that split.\n    if (negIndex !== -1) {\n      const total = numOrSizeSplits.reduce((a, b) => b > 0 ? a + b : a);\n      numOrSizeSplits[negIndex] = x.shape[axis] - total;\n    }\n    assert(\n        x.shape[axis] === numOrSizeSplits.reduce((a, b) => a + b),\n        () => 'The sum of sizes must match the size of the axis dimension.');\n    splitSizes = numOrSizeSplits;\n  }\n\n  return splitSizes;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SplitV, SplitVAttrs, SplitVInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {prepareSplitSize} from './split_util';\n\n/**\n * Splits a `tf.Tensor` into sub tensors.\n *\n * If `numOrSizeSplits` is a number, splits `x` along dimension `axis`\n * into `numOrSizeSplits` smaller tensors.\n * Requires that `numOrSizeSplits` evenly divides `x.shape[axis]`.\n *\n * If `numOrSizeSplits` is a number array, splits `x` into\n * `numOrSizeSplits.length` pieces. The shape of the `i`-th piece has the\n * same size as `x` except along dimension `axis` where the size is\n * `numOrSizeSplits[i]`.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4, 5, 6, 7, 8], [2, 4]);\n * const [a, b] = tf.split(x, 2, 1);\n * a.print();\n * b.print();\n *\n * const [c, d, e] = tf.split(x, [1, 2, 1], 1);\n * c.print();\n * d.print();\n * e.print();\n * ```\n *\n * @param x The input tensor to split.\n * @param numOrSizeSplits Either an integer indicating the number of\n * splits along the axis or an array of integers containing the sizes of\n * each output tensor along the axis. If a number then it must evenly divide\n * `x.shape[axis]`; otherwise the sum of sizes must match `x.shape[axis]`.\n * Can contain one -1 indicating that dimension is to be inferred.\n * @param axis The dimension along which to split. Defaults to 0 (the first\n * dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction split_<T extends Tensor>(\n    x: Tensor|TensorLike, numOrSizeSplits: number[]|number, axis = 0): T[] {\n  const $x = convertToTensor(x, 'x', 'split');\n\n  const forward: ForwardFunc<Tensor> = (backend, _) => {\n    const $axis = parseAxisParam(axis, $x.shape)[0];\n    const splitSizes = prepareSplitSize($x, numOrSizeSplits, $axis);\n    return backend.split($x, splitSizes, $axis) as {} as T;\n  };\n\n  const inputs: SplitVInputs = {x: $x};\n  const attr: SplitVAttrs = {numOrSizeSplits, axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, SplitV,\n             attr as {} as NamedAttrMap) as {} as T[];\n}\n\nexport const split = op({split_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {assert} from '../../util';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {split} from '../split';\nimport {zeros} from '../zeros';\nimport {zerosLike} from '../zeros_like';\n\nimport {fft} from './fft';\n\n/**\n * Real value input fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n *\n * real.rfft().print();\n * ```\n * @param input The real value input to compute an rfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction rfft_(input: Tensor, fftLength?: number): Tensor {\n  assert(\n      input.dtype === 'float32',\n      () => `The dtype for rfft() must be real value but got ${input.dtype}`);\n\n  let innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n\n  let adjustedInput: Tensor;\n  if (fftLength != null && fftLength < innerDimensionSize) {\n    // Need to crop\n    const begin = input.shape.map(v => 0);\n    const size = input.shape.map(v => v);\n    size[input.shape.length - 1] = fftLength;\n    adjustedInput = slice(input, begin, size);\n    innerDimensionSize = fftLength;\n  } else if (fftLength != null && fftLength > innerDimensionSize) {\n    // Need to pad with zeros\n    const zerosShape = input.shape.map(v => v);\n    zerosShape[input.shape.length - 1] = fftLength - innerDimensionSize;\n    adjustedInput = concat([input, zeros(zerosShape)], input.shape.length - 1);\n    innerDimensionSize = fftLength;\n  } else {\n    adjustedInput = input;\n  }\n\n  // Complement the input with zero imaginary numbers.\n  const zerosInput = zerosLike(adjustedInput);\n  const complexInput =\n      reshape(complex(adjustedInput, zerosInput), [batch, innerDimensionSize]);\n\n  const ret = fft(complexInput);\n\n  // Exclude complex conjugations. These conjugations are put symmetrically.\n  const half = Math.floor(innerDimensionSize / 2) + 1;\n  const realValues = real(ret);\n  const imagValues = imag(ret);\n  const realComplexConjugate = split(\n      realValues, [half, innerDimensionSize - half],\n      realValues.shape.length - 1);\n  const imagComplexConjugate = split(\n      imagValues, [half, innerDimensionSize - half],\n      imagValues.shape.length - 1);\n\n  const outputShape = adjustedInput.shape.slice();\n  outputShape[adjustedInput.shape.length - 1] = half;\n\n  return reshape(\n      complex(realComplexConjugate[0], imagComplexConjugate[0]), outputShape);\n}\n\nexport const rfft = op({rfft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sqrt, SqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes square root of the input `tf.Tensor` element-wise: `y = sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.sqrt().print();  // or tf.sqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sqrt');\n\n  const inputs: SqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sqrt);\n}\nexport const sqrt = op({sqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SquaredDifference, SquaredDifferenceInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns (a - b) * (a - b) element-wise.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * ```js\n * // Broadcast squared difference  a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction squaredDifference_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'squaredDifference');\n  let $b = convertToTensor(b, 'b', 'squaredDifference');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.squaredDifference($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SquaredDifferenceInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             SquaredDifference, attrs) as T;\n}\n\nexport const squaredDifference = op({squaredDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {squeezeShape} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Removes dimensions of size 1 from the shape of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4], [1, 1, 4]);\n * x.squeeze().print();\n * ```\n *\n * @param x The input tensor to be squeezed.\n * @param axis An optional list of numbers. If specified, only\n *     squeezes the dimensions listed. The dimension index starts at 0. It\n * is an error to squeeze a dimension that is not 1.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction squeeze_<T extends Tensor>(x: Tensor|TensorLike, axis?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'squeeze');\n  return reshape($x, squeezeShape($x.shape, axis).newShape) as T;\n}\n\nexport const squeeze = op({squeeze_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {concat} from './concat';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\n\n/**\n * Stacks a list of rank-`R` `tf.Tensor`s into one rank-`(R+1)` `tf.Tensor`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.stack([a, b, c]).print();\n * ```\n *\n * @param tensors A list of tensor objects with the same shape and dtype.\n * @param axis The axis to stack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction stack_<T extends Tensor>(\n    tensors: Array<T|TensorLike>, axis = 0): Tensor {\n  const $tensors = convertToTensorArray(tensors, 'tensors', 'stack');\n\n  util.assert(\n      $tensors.length >= 1, () => 'Pass at least one tensor to tf.stack');\n\n  if ($tensors.length === 1) {\n    return expandDims($tensors[0], axis);\n  }\n\n  const rank = $tensors[0].rank;\n  const shape = $tensors[0].shape;\n  const dtype = $tensors[0].dtype;\n\n  util.assert(axis <= rank, () => 'Axis must be <= rank of the tensor');\n\n  $tensors.forEach(t => {\n    util.assertShapesMatch(\n        shape, t.shape,\n        'All tensors passed to stack must have matching shapes');\n    util.assert(\n        dtype === t.dtype,\n        () => 'All tensors passed to stack must have matching dtypes');\n  });\n\n  const expandedTensors = $tensors.map(t => expandDims(t, axis));\n  // Stack exists in the TensorFlow C++ API\n  // (https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/stack) but not\n  // in\n  // https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/core/ops/ops.pbtxt.\n  // Therefore we are treating it like a high-level op rather than\n  // creating a dedicated stack kernel.\n  return concat(expandedTensors, axis);\n}\n\nexport const stack = op({stack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Step, StepAttrs, StepInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes step of the input `tf.Tensor` element-wise: `x > 0 ? 1 : alpha * x`\n *\n * ```js\n * const x = tf.tensor1d([0, 2, -1, -3]);\n *\n * x.step(.5).print();  // or tf.step(x, .5)\n * ```\n * @param x The input tensor.\n * @param alpha The gradient when input is negative.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction step_<T extends Tensor>(x: T|TensorLike, alpha = 0.0): T {\n  const $x = convertToTensor(x, 'x', 'step');\n\n  const inputs: StepInputs = {x: $x};\n  const attrs: StepAttrs = {alpha};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.step($x, alpha), inputs as {} as NamedTensorMap,\n      null /* grad */, Step, attrs as {} as NamedAttrMap);\n}\nexport const step = op({step_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {StridedSlice, StridedSliceAttrs, StridedSliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {slice} from './slice';\nimport {computeOutShape, getNormalizedAxes, maskToAxes} from './slice_util';\n\n/**\n * Extracts a strided slice of a tensor.\n *\n * Roughly speaking, this op extracts a slice of size (end-begin)/stride from\n * the given input tensor (x). Starting at the location specified by begin the\n * slice continues by adding stride to the index until all dimensions are not\n * less than end. Note that a stride can be negative, which causes a reverse\n * slice.\n *\n * ```js\n * const t = tf.tensor3d([1, 1, 1 ,2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6],\n *    [3, 2, 3]);\n * t.stridedSlice([1, 0, 0], [2, 1, 3], [1, 1, 1]).print()  // [[[3, 3, 3]]]\n * t.stridedSlice([1, 0, 0], [2, 2, 3], [1, 1, 1]).print()  // [[[3, 3, 3],\n *                                                     // [4, 4, 4]]]\n * t.stridedSlice([1, -1, 0], [2, -3, 3], [1, -1, 1]).print() // [[[4, 4, 4],\n *                                                     // [3, 3, 3]]]\n * ```\n *\n * @param x The tensor to stride slice.\n * @param begin The coordinates to start the slice from.\n * @param end: The coordinates to end the slice at.\n * @param strides: The size of the slice.\n * @param beginMask: If the ith bit of beginMask is set, begin[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param endMask: If the ith bit of endMask is set, end[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param shrinkAxisMask: a bitmask where bit i implies that\n * the ith specification should shrink the dimensionality. begin and end must\n * imply a slice of size 1 in the dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction stridedSlice_(\n    x: Tensor|TensorLike, begin: number[], end: number[], strides?: number[],\n    beginMask = 0, endMask = 0, ellipsisMask = 0, newAxisMask = 0,\n    shrinkAxisMask = 0): Tensor {\n  let $x = convertToTensor(x, 'x', 'stridedSlice');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if (strides == null) {\n      strides = new Array(begin.length);\n    }\n\n    const ellipsisAxes = maskToAxes(ellipsisMask);\n    if (ellipsisAxes.length > 1) {\n      throw new Error('Multiple ellipses in slice is not allowed.');\n    }\n\n    if (ellipsisMask !== 0 && newAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and newAxisMask is not yet supported.');\n    }\n\n    if (ellipsisMask !== 0 && shrinkAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and shrinkAxisMask is not yet supported.');\n    }\n\n    const numInterpolatedAxes = $x.rank - begin.length;\n\n    // Expand the dims of x based on the newAxisMask.\n    const expandAxes = maskToAxes(newAxisMask);\n    const newShape = $x.shape.slice();\n    expandAxes.forEach(axis => {\n      begin[axis] = 0;\n      end[axis] = 1;\n      newShape.splice(axis, 0, 1);\n    });\n    $x = reshape($x, newShape);\n\n    const {\n      begin: normalizedBegin,\n      end: normalizedEnd,\n      strides: normalizedStrides\n    } =\n        getNormalizedAxes(\n            $x.shape, ellipsisAxes, numInterpolatedAxes, begin, end, strides,\n            beginMask, endMask, ellipsisMask);\n    begin = normalizedBegin;\n    end = normalizedEnd;\n    strides = normalizedStrides;\n\n    const shrinkAxes = maskToAxes(shrinkAxisMask);\n    // Adjust the ends based on the shrink mask.\n    shrinkAxes.forEach(axis => {\n      end[axis] = begin[axis] + 1;\n      strides[axis] = 1;\n    });\n\n    // Figure out the output shape.\n    const size = computeOutShape(begin, end, strides);\n    // Remove the axes based on shrinkMask.\n    const outShape = size.filter((_, axis) => shrinkAxes.indexOf(axis) === -1);\n\n    const nonStrided = strides.every(v => v === 1);\n    if (nonStrided) {\n      return reshape(slice($x, begin, size), outShape);\n    }\n\n    const res = backend.stridedSlice($x, begin, end, strides);\n    return reshape(res, outShape);\n  };\n\n  const inputs: StridedSliceInputs = {x: $x};\n  const attrs: StridedSliceAttrs = {\n    begin,\n    end,\n    strides,\n    beginMask,\n    endMask,\n    ellipsisMask,\n    newAxisMask,\n    shrinkAxisMask\n  };\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, StridedSlice,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const stridedSlice = op({stridedSlice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tan, TanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes tan of the input `tf.Tensor` element-wise, `tan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.tan().print();  // or tf.tan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tan');\n\n  const inputs: TanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.tan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tan);\n}\nexport const tan = op({tan_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike2D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-2 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor2d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor2d([[1, 2], [3, 4]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor2d(\n    values: TensorLike2D, shape?: [number, number],\n    dtype?: DataType): Tensor2D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 2) {\n    throw new Error('tensor2d() requires shape to have two numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 2 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor2d() requires values to be number[][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor2d() requires shape to be provided when `values` ' +\n        'are a flat/TypedArray');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-4 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor4d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor4d([[[[1], [2]], [[3], [4]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor4d(\n    values: TensorLike4D, shape?: [number, number, number, number],\n    dtype?: DataType): Tensor4D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 4) {\n    throw new Error('tensor4d() requires shape to have four numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 4 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor4d() requires values to be number[][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor4d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor4D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor5D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike5D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-5 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor5d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor5d([[[[[1], [2]], [[3], [4]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor5d(\n    values: TensorLike5D, shape?: [number, number, number, number, number],\n    dtype?: DataType): Tensor5D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 5) {\n    throw new Error('tensor5d() requires shape to have five numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 5 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor5d() requires values to be ' +\n        'number[][][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor5d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor5D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor6D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike6D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-6 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor6d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor6d([[[[[[1],[2]],[[3],[4]]],[[[5],[6]],[[7],[8]]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor6d([1, 2, 3, 4, 5, 6, 7, 8], [1, 1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor6d(\n    values: TensorLike6D,\n    shape?: [number, number, number, number, number, number],\n    dtype?: DataType): Tensor6D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 6) {\n    throw new Error('tensor6d() requires shape to have six numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 6 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor6d() requires values to be number[][][][][][] or ' +\n        'flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor6d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  shape = shape ||\n      inferredShape as [number, number, number, number, number, number];\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor6D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {TopK, TopKAttrs, TopKInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {NumericTensor, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Finds the values and indices of the `k` largest entries along the last\n * dimension.\n *\n * If the input is a vector (rank=1), finds the k largest entries in the vector\n * and outputs their values and indices as vectors. Thus values[j] is the j-th\n * largest entry in input, and its index is indices[j].\n * For higher rank inputs, computes the top k entries along the last dimension.\n *\n * If two elements are equal, the lower-index element appears first.\n *\n * ```js\n * const a = tf.tensor2d([[1, 5], [4, 3]]);\n * const {values, indices} = tf.topk(a);\n * values.print();\n * indices.print();\n * ```\n * @param x 1-D or higher `tf.Tensor` with last dimension being at least `k`.\n * @param k Number of top elements to look for along the last dimension.\n * @param sorted If true, the resulting `k` elements will be sorted by the\n *     values in descending order.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction topk_<T extends Tensor>(\n    x: T|TensorLike, k = 1, sorted = true): {values: T, indices: T} {\n  const $x = convertToTensor(x, 'x', 'topk');\n  if ($x.rank === 0) {\n    throw new Error('topk() expects the input to be of rank 1 or higher');\n  }\n  const lastDim = $x.shape[$x.shape.length - 1];\n  if (k > lastDim) {\n    throw new Error(\n        `'k' passed to topk() must be <= the last dimension (${lastDim}) ` +\n        `but got ${k}`);\n  }\n\n  const inputs: TopKInputs = {x: $x};\n  const attrs: TopKAttrs = {k, sorted};\n\n  const [values, indices] = ENGINE.runKernelFunc(\n      b => b.topk($x as NumericTensor, k, sorted),\n      inputs as {} as NamedTensorMap, null /* grad */, TopK,\n      attrs as {} as NamedAttrMap);\n\n  return {values, indices} as {values: T, indices: T};\n}\n\nexport const topk = op({topk_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a truncated normal\n * distribution.\n *\n * ```js\n * tf.truncatedNormal([2, 2]).print();\n * ```\n *\n * The generated values follow a normal distribution with specified mean and\n * standard deviation, except that values whose magnitude is more than 2\n * standard deviations from the mean are dropped and re-picked.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output tensor.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction truncatedNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type $ { dtype }`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, true /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const truncatedNormal = op({truncatedNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Unique, UniqueAttrs, UniqueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Finds unique elements along an axis of a tensor.\n *\n * It returns a tensor `values` containing all of the unique elements along the\n * `axis` of the given tensor `x` in the same order that they occur along the\n * `axis` in `x`; `x` does not need to be sorted. It also returns a tensor\n * `indices` the same size as the number of the elements in `x` along the `axis`\n * dimension. It contains the index in the unique output `values`.\n *\n * ```js\n * // A 1-D tensor\n * const a = tf.tensor1d([1, 1, 2, 4, 4, 4, 7, 8, 8]);\n * const {values, indices} = tf.unique(a);\n * values.print();   // [1, 2, 4, 7, 8,]\n * indices.print();  // [0, 0, 1, 2, 2, 2, 3, 4, 4]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=0\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 0)\n * values.print();   // [[1, 0, 0],\n *                   //  [2, 0, 0]]\n * indices.print();  // [0, 0, 1]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=1\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 1)\n * values.print();   // [[1, 0],\n *                   //  [1, 0],\n *                   //  [2, 0]]\n * indices.print();  // [0, 1, 1]\n * ```\n * @param x A tensor (int32, string, bool).\n * @param axis The axis of the tensor to find the unique elements.\n * @returns [uniqueElements, indices] (see above for details)\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction unique_<T extends Tensor>(\n    x: T|TensorLike, axis = 0): {values: T, indices: Tensor1D} {\n  // x can be of any dtype, thus null as the last argument.\n  const $x = convertToTensor(x, 'x', 'unique', null);\n  assert($x.rank > 0, () => 'The input tensor must be at least 1D');\n\n  const inputs: UniqueInputs = {x: $x};\n  const attrs: UniqueAttrs = {axis};\n  const [values, indices] = ENGINE.runKernel(\n                                Unique, inputs as {} as NamedTensorMap,\n                                attrs as {} as NamedAttrMap) as [T, Tensor1D];\n  return {values, indices};\n}\n\nexport const unique = op({unique_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {UnsortedSegmentSum, UnsortedSegmentSumAttrs, UnsortedSegmentSumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, isInt} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Computes the sum along segments of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const segmentIds = tf.tensor1d([1, 2, 0, 1], 'int32');\n * const numSegments = 3;\n *\n * x.unsortedSegmentSum(segmentIds, numSegments).print()\n * //or tf.unsortedSegmentSum(x, segmentIds, numSegments)\n * ```\n * @param x The `tf.Tensor` that will be summed along its segments.\n * @param segmentIds A `tf.Tensor1D` whose rank is equal to the rank of `x`'s\n * dimension along the `axis`.  Maps each element of `x` to a segment.\n * @param numSegments The number of distinct `segmentIds`.\n *\n * @doc {heading: 'Operations', subheading: 'Segment'}\n */\nfunction unsortedSegmentSum_<T extends Tensor>(\n    x: T|TensorLike, segmentIds: Tensor1D|TensorLike, numSegments: number): T {\n  const $x = convertToTensor(x, 'x', 'unsortedSegmentSum');\n  const $segmentIds =\n      convertToTensor(segmentIds, 'segmentIds', 'unsortedSegmentSum', 'int32');\n  assert(isInt(numSegments), () => 'numSegments must be of dtype int');\n\n  const inputs: UnsortedSegmentSumInputs = {x: $x, segmentIds: $segmentIds};\n  const attrs: UnsortedSegmentSumAttrs = {numSegments};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.unsortedSegmentSum($x, $segmentIds, numSegments);\n    save([$segmentIds]);\n    return res;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             UnsortedSegmentSum, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const unsortedSegmentSum = op({unsortedSegmentSum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Unpack, UnpackAttrs, UnpackInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Unstacks a `tf.Tensor` of rank-`R` into a list of rank-`(R-1)` `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * tf.unstack(a).forEach(tensor => tensor.print());\n * ```\n *\n * @param x A tensor object.\n * @param axis The axis to unstack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction unstack_(x: Tensor|TensorLike, axis = 0): Tensor[] {\n  const $x = convertToTensor(x, 'x', 'unstack');\n  util.assert(\n      axis >= -$x.shape.length && axis < $x.shape.length,\n      () =>\n          `Axis = ${axis} is not in [-${$x.shape.length}, ${$x.shape.length})`);\n  if (axis < 0) {\n    axis += $x.shape.length;\n  }\n  const inputs: UnpackInputs = {value: $x};\n  const attrs: UnpackAttrs = {axis};\n  const forward: ForwardFunc<Tensor[]> = (backend: KernelBackend) =>\n      backend.unstack($x, axis);\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Unpack,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const unstack = op({unstack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor, Variable} from '../tensor';\nimport {DataType, Rank} from '../types';\n\n/**\n * Creates a new variable with the provided initial value.\n * ```js\n * const x = tf.variable(tf.tensor([1, 2, 3]));\n * x.assign(tf.tensor([4, 5, 6]));\n *\n * x.print();\n * ```\n *\n * @param initialValue Initial value for the tensor.\n * @param trainable If true, optimizers are allowed to update it.\n * @param name Name of the variable. Defaults to a unique id.\n * @param dtype If set, initialValue will be converted to the given type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function variable<R extends Rank>(\n    initialValue: Tensor<R>, trainable = true, name?: string,\n    dtype?: DataType): Variable<R> {\n  return ENGINE.makeVariable(initialValue, trainable, name, dtype) as\n      Variable<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the Where kernel shared between cpu and webgl */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor2D} from '../tensor';\nimport {TypedArray} from '../types';\n\nexport function whereImpl(condShape: number[], condVals: TypedArray): Tensor2D {\n  const indices = [];\n  for (let i = 0; i < condVals.length; i++) {\n    if (condVals[i]) {\n      indices.push(i);\n    }\n  }\n\n  const inBuffer = buffer(condShape, 'int32');\n\n  const out = buffer([indices.length, condShape.length], 'int32');\n  for (let i = 0; i < indices.length; i++) {\n    const loc = inBuffer.indexToLoc(indices[i]);\n    const offset = i * condShape.length;\n    out.values.set(loc, offset);\n  }\n  return out.toTensor() as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {whereImpl} from '../backends/where_impl';\nimport {Tensor, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\n/**\n * Returns the coordinates of true elements of condition.\n *\n * The coordinates are returned in a 2-D tensor where the first dimension (rows)\n * represents the number of true elements, and the second dimension (columns)\n * represents the coordinates of the true elements. Keep in mind, the shape of\n * the output tensor can vary depending on how many true values there are in\n * input. Indices are output in row-major order. The resulting tensor has the\n * shape `[numTrueElems, condition.rank]`.\n *\n * This is analogous to calling the python `tf.where(cond)` without an x or y.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const result = await tf.whereAsync(cond);\n * result.print();\n * ```\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nasync function whereAsync_(condition: Tensor|TensorLike): Promise<Tensor2D> {\n  const $condition =\n      convertToTensor(condition, 'condition', 'whereAsync', 'bool');\n  const vals = await $condition.data();\n  const res = whereImpl($condition.shape, vals);\n  if (condition !== $condition) {\n    $condition.dispose();\n  }\n  return res;\n}\n\nexport const whereAsync = whereAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {gather} from './gather';\nimport {reshape} from './reshape';\nimport {squeeze} from './squeeze';\nimport {whereAsync} from './where_async';\n\n/**\n * Apply boolean mask to tensor.\n *\n * ```js\n * const tensor = tf.tensor2d([1, 2, 3, 4, 5, 6], [3, 2]);\n * const mask = tf.tensor1d([1, 0, 1], 'bool');\n * const result = await tf.booleanMaskAsync(tensor, mask);\n * result.print();\n * ```\n *\n * @param tensor N-D tensor.\n * @param mask K-D boolean tensor, K <= N and K must be known statically.\n * @param axis A 0-D int Tensor representing the axis in tensor to mask from.\n *     By default, axis is 0 which will mask from the first dimension.\n *     Otherwise K + axis <= N.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nasync function booleanMaskAsync_(\n    tensor: Tensor|TensorLike, mask: Tensor|TensorLike,\n    axis?: number): Promise<Tensor> {\n  const $tensor = convertToTensor(tensor, 'tensor', 'boolMask');\n  const $mask = convertToTensor(mask, 'mask', 'boolMask', 'bool');\n\n  const axisFrom = axis == null ? 0 : axis;\n  const maskDim = $mask.rank;\n  const tensorShape = $tensor.shape;\n\n  util.assert(maskDim > 0, () => 'mask cannot be scalar');\n  util.assertShapesMatch(\n      tensorShape.slice(axisFrom, axisFrom + maskDim), $mask.shape,\n      `mask's shape must match the first K dimensions of tensor's shape,`);\n\n  let leadingSize = 1;\n  for (let i = axisFrom; i < axisFrom + maskDim; i++) {\n    leadingSize *= tensorShape[i];\n  }\n  const targetTensorShape =\n      tensorShape.slice(0, axisFrom)\n          .concat([leadingSize], tensorShape.slice(axisFrom + maskDim));\n  const reshapedTensor = reshape($tensor, targetTensorShape);\n  const reshapedMask = reshape($mask, [-1]);\n  const positivePositions = await whereAsync(reshapedMask);\n  const indices = squeeze(positivePositions, [1]);\n\n  const res = gather(reshapedTensor, indices, axisFrom);\n\n  // Ensure no memory leak.\n  if (tensor !== $tensor) {\n    $tensor.dispose();\n  }\n  if (mask !== $mask) {\n    $mask.dispose();\n  }\n  indices.dispose();\n  reshapedTensor.dispose();\n  reshapedMask.dispose();\n  positivePositions.dispose();\n\n  return res;\n}\n\nexport const booleanMaskAsync = booleanMaskAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertShapesMatch} from '../util';\n\nimport {equal} from './equal';\nimport {greater} from './greater';\nimport {greaterEqual} from './greater_equal';\nimport {less} from './less';\nimport {lessEqual} from './less_equal';\nimport {notEqual} from './not_equal';\nimport {op} from './operation';\n\n/**\n * @deprecated\n * Strict version of `tf.notEqual` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction notEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'notEqualStrict');\n  const $b = convertToTensor(b, 'b', 'notEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in notEqualStrict: ');\n  return notEqual($a, $b);\n}\n\n/**\n * @deprecated\n * Strict version of `tf.less` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction lessStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessStrict');\n  const $b = convertToTensor(b, 'b', 'lessStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessStrict: ');\n  return less($a, $b);\n}\n\nfunction equalStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'equalStrict');\n  const $b = convertToTensor(b, 'b', 'equalStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in equalStrict: ');\n  return equal($a, $b);\n}\n\nfunction lessEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessEqualStrict');\n  const $b = convertToTensor(b, 'b', 'lessEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessEqualStrict: ');\n  return lessEqual($a, $b);\n}\n\nfunction greaterStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterStrict');\n  const $b = convertToTensor(b, 'b', 'greaterStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterStrict: ');\n  return greater($a, $b);\n}\n\nfunction greaterEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterEqualStrict');\n  const $b = convertToTensor(b, 'b', 'greaterEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterEqualStrict: ');\n  return greaterEqual($a, $b);\n}\n\nexport const equalStrict = op({equalStrict_});\nexport const greaterEqualStrict = op({greaterEqualStrict_});\nexport const greaterStrict = op({greaterStrict_});\nexport const lessEqualStrict = op({lessEqualStrict_});\nexport const lessStrict = op({lessStrict_});\nexport const notEqualStrict = op({notEqualStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {maximum} from './maximum';\nimport {minimum} from './minimum';\nimport {mod} from './mod';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {squaredDifference} from './squared_difference';\nimport {sub} from './sub';\n\n/**\n * @deprecated\n * Adds two `tf.Tensor`s element-wise, A + B.\n *\n * Inputs must be the same shape. For broadcasting support, use add() instead.\n *\n * @param a The first Tensor to add element-wise.\n * @param b The second Tensor to add element-wise.\n */\nfunction addStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'addStrict');\n  const $b = convertToTensor(b, 'b', 'addStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in addStrict: ');\n  return add($a, $b);\n}\n\n/**\n * @deprecated\n * Subtracts two `tf.Tensor`s element-wise, A - B. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.sub` instead.\n *\n * @param a The first Tensor to subtract element-wise.\n * @param b The second Tensor to subtract element-wise.\n */\nfunction subStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'subStrict');\n  const $b = convertToTensor(b, 'b', 'subStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in subStrict: ');\n  return sub($a, $b);\n}\n\n/**\n * @deprecated\n * Computes the power of one `tf.Tensor` to another. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.pow` instead.\n *\n * @param base The base tensor to pow element-wise.\n * @param exp The exponent tensor to pow element-wise.\n */\nfunction powStrict_<T extends Tensor>(base: T, exp: Tensor): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  util.assertShapesMatch(base.shape, exp.shape, 'Error in powStrict: ');\n  return pow(base, exp);\n}\n\n/**\n * @deprecated\n * Multiplies two `tf.Tensor`s element-wise, A * B.\n *\n * Inputs must be the same shape. For broadcasting support, use `tf.mul`.\n *\n * @param a The first tensor to multiply.\n * @param b The first tensor to multiply. Must have the same\n *    dtype as `a`.\n */\nfunction mulStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'mul');\n  const $b = convertToTensor(b, 'b', 'mul');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in multiplyStrict: ');\n  return mul($a, $b);\n}\n\n/**\n * @deprecated\n * Divides two `tf.Tensor`s element-wise, A / B. Inputs must\n * be the same shape.\n *\n * @param a The first tensor as the numerator for element-wise division.\n * @param b The second tensor as the denominator for element-wise division.\n */\nfunction divStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'div');\n  const $b = convertToTensor(b, 'b', 'div');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in divideStrict: ');\n  return div($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the mod of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use mod().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction modStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'modStrict');\n  const $b = convertToTensor(b, 'b', 'modStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in modStrict: ');\n  return mod($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the min of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use minimum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction minimumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'minimumStrict');\n  const $b = convertToTensor(b, 'b', 'minimumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in minimumStrict: ');\n  return minimum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the max of a and b (`a > b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use maximum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction maximumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'maximumStrict');\n  const $b = convertToTensor(b, 'b', 'maximumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in maximumStrict: ');\n  return maximum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns (a - b) * (a - b) element-wise.\n *\n * Inputs must be the same shape. For broadcasting support, use\n * `tf.squaredDifference` instead.\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n */\nfunction squaredDifferenceStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'squaredDifferenceStrict');\n  const $b = convertToTensor(b, 'b', 'squaredDifferenceStrict');\n  util.assertShapesMatch(\n      $a.shape, $b.shape, 'Error in squaredDifferenceStrict: ');\n  return squaredDifference($a, $b);\n}\n\nexport const addStrict = op({addStrict_});\nexport const divStrict = op({divStrict_});\nexport const maximumStrict = op({maximumStrict_});\nexport const minimumStrict = op({minimumStrict_});\nexport const modStrict = op({modStrict_});\nexport const mulStrict = op({mulStrict_});\nexport const powStrict = op({powStrict_});\nexport const squaredDifferenceStrict = op({squaredDifferenceStrict_});\nexport const subStrict = op({subStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {abs} from './abs';\nimport * as axis_util from './axis_util';\nimport {max} from './max';\nimport {min} from './min';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sqrt} from './sqrt';\nimport {square} from './square';\nimport {sum} from './sum';\n\n/**\n * Computes the norm of scalar, vectors, and matrices.\n * This function can compute several different vector norms (the 1-norm, the\n * Euclidean or 2-norm, the inf-norm, and in general the p-norm for p > 0)\n * and matrix norms (Frobenius, 1-norm, and inf-norm).\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.norm().print();  // or tf.norm(x)\n * ```\n *\n * @param x The input array.\n * @param ord Optional. Order of the norm. Supported norm types are\n * following:\n *\n *  | ord        | norm for matrices         | norm for vectors\n *  |------------|---------------------------|---------------------\n *  |'euclidean' |Frobenius norm             |2-norm\n *  |'fro'       |Frobenius norm\t           |\n *  |Infinity    |max(sum(abs(x), axis=1))   |max(abs(x))\n *  |-Infinity   |min(sum(abs(x), axis=1))   |min(abs(x))\n *  |1           |max(sum(abs(x), axis=0))   |sum(abs(x))\n *  |2           |                           |sum(abs(x)^2)^1/2*\n *\n * @param axis Optional. If axis is null (the default), the input is\n * considered a vector and a single vector norm is computed over the entire\n * set of values in the Tensor, i.e. norm(x, ord) is equivalent\n * to norm(x.reshape([-1]), ord). If axis is a integer, the input\n * is considered a batch of vectors, and axis determines the axis in x\n * over which to compute vector norms. If axis is a 2-tuple of integer it is\n * considered a batch of matrices and axis determines the axes in NDArray\n * over which to compute a matrix norm.\n * @param keepDims Optional. If true, the norm have the same dimensionality\n * as the input.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction norm_(\n    x: Tensor|TensorLike, ord: number|'euclidean'|'fro' = 'euclidean',\n    axis: number|number[] = null, keepDims = false): Tensor {\n  x = convertToTensor(x, 'x', 'norm');\n\n  const norm = normImpl(x, ord, axis);\n  let keepDimsShape = norm.shape;\n  if (keepDims) {\n    const axes = parseAxisParam(axis, x.shape);\n    keepDimsShape = axis_util.expandShapeToKeepDim(norm.shape, axes);\n  }\n  return reshape(norm, keepDimsShape);\n}\n\nfunction normImpl(\n    x: Tensor, p: number|string, axis: number|number[] = null): Tensor {\n  if (x.rank === 0) {\n    return abs(x);\n  }\n\n  // consider vector when no axis is specified\n  if (x.rank !== 1 && axis === null) {\n    return normImpl(reshape(x, [-1]), p, axis);\n  }\n\n  // vector\n  if (x.rank === 1 || typeof axis === 'number' ||\n      Array.isArray(axis) && axis.length === 1) {\n    if (p === 1) {\n      return sum(abs(x), axis);\n    }\n    if (p === Infinity) {\n      return max(abs(x), axis);\n    }\n    if (p === -Infinity) {\n      return min(abs(x), axis);\n    }\n    if (p === 'euclidean' || p === 2) {\n      // norm(x, 2) = sum(abs(xi) ^ 2) ^ 1/2\n      return sqrt(sum(pow(abs(x), scalar(2, 'int32')), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  // matrix (assumption axis[0] < axis[1])\n  if (Array.isArray(axis) && axis.length === 2) {\n    if (p === 1) {\n      return max(sum(abs(x), axis[0]), axis[1] - 1);\n    }\n    if (p === Infinity) {\n      return max(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === -Infinity) {\n      return min(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === 'fro' || p === 'euclidean') {\n      // norm(x) = sqrt(sum(pow(x, 2)))\n      return sqrt(sum(square(x), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  throw new Error(`Error in norm: invalid axis: ${axis}`);\n}\n\nexport const norm = op({norm_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor} from '../tensor';\nimport {assertTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {scalar} from './scalar';\nimport {sub} from './sub';\n\n/**\n * Compute the moving average of a variable.\n *\n * Without zeroDebias, the moving average operation is defined by:\n *   `v += delta`\n * where\n *   `delta = (1 - decay) * (x - v)`\n *\n * With zeroDebias (default), the `delta` term is scaled to debias the\n * effect of the (assumed) zero-initialization of `v`.\n *   `delta /= (1 - decay ^ step)`\n *\n * For more details on the zero-debiasing algorithm, see:\n *   https://arxiv.org/abs/1412.6980\n *\n * Note that this function is completely stateless and does not keep track of\n * step count. The step count needs to be maintained by the caller and passed\n * in as `step`.\n *\n * @param v The current moving average value.\n * @param x New input value, must have the same shape and dtype as `v`.\n * @param decay The decay factor. Typical values are 0.95 and 0.99.\n * @param step Step count.\n * @param zeroDebias: Whether zeroDebias is to be performed (default: `true`).\n * @returns The new moving average value.\n *\n * @doc {heading: 'Operations', subheading: 'Moving Average'}\n */\nfunction movingAverage_<T extends Tensor>(\n    v: T|TensorLike, x: T|TensorLike, decay: number|Scalar,\n    step?: number|Scalar, zeroDebias = true): T {\n  const $v = convertToTensor(v, 'v', 'movingAverage');\n  const $x = convertToTensor(x, 'x', 'movingAverage');\n  const $decay = convertToTensor(decay, 'decay', 'movingAverage');\n\n  assertTypesMatch($v, $x);\n  util.assert(\n      util.arraysEqual($v.shape, $x.shape), () => 'Shape mismatch in v and x');\n\n  const one = scalar(1);\n  const oneMinusDecay = sub(one, $decay);\n\n  let update = mul(sub($x, $v), oneMinusDecay);\n  if (zeroDebias) {\n    util.assert(\n        step != null, () => 'When using zeroDebias: true, step is required.');\n    const $step = convertToTensor(step, 'step', 'movingAverage');\n    update = div(update, sub(one, pow($decay, $step)));\n  }\n  return add($v, update);\n}\n\nexport const movingAverage = op({movingAverage_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ScatterNd, ScatterNdAttrs, ScatterNdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as scatter_nd_util from './scatter_nd_util';\n\n/**\n * Creates a new tensor by applying sparse updates to individual\n * values or slices within a zero tensor of the given shape tensor according to\n * indices. This operator is the inverse of the `tf.gatherND` operator which\n * extracts values or slices from a given tensor.\n *\n * ```js\n * const indices = tf.tensor2d([4, 3, 1, 7], [4, 1], 'int32');\n * const updates = tf.tensor1d([9, 10, 11, 12]);\n * const shape = [8];\n * tf.scatterND(indices, updates, shape).print() //[0, 11, 0, 10, 9, 0, 0, 12]\n * ```\n *\n * @param indices The tensor contains the indices into the output tensor.\n * @param updates The tensor contains the value for the indices.\n * @param shape: The shape of the output tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction scatterND_<R extends Rank>(\n    indices: Tensor|TensorLike, updates: Tensor|TensorLike,\n    shape: ShapeMap[R]): Tensor<R> {\n  const $indices = convertToTensor(indices, 'indices', 'scatterND', 'int32');\n  const $updates = convertToTensor(updates, 'updates', 'scatterND');\n  scatter_nd_util.validateInput($updates, $indices, shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.scatterND($indices, $updates, shape);\n  };\n\n  const inputs: ScatterNdInputs = {indices: $indices, updates: $updates};\n  const attrs: ScatterNdAttrs = {shape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             ScatterNd, attrs as {} as NamedAttrMap) as Tensor<R>;\n}\n\nexport const scatterND = op({scatterND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\n\n/**\n * Validate sparseToDense inputs.\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape number[]. Shape of the dense output tensor.\n * @param validateIndices boolean. indice validation is not supported, error\n * will be thrown if it is set.\n */\nexport function validateInput(\n    sparseIndices: Tensor, sparseValues: Tensor, outputShape: number[],\n    defaultValues: Tensor) {\n  if (sparseIndices.dtype !== 'int32') {\n    throw new Error(\n        'tf.sparseToDense() expects the indices to be int32 type,' +\n        ` but the dtype was ${sparseIndices.dtype}.`);\n  }\n  if (sparseIndices.rank > 2) {\n    throw new Error(\n        'sparseIndices should be a scalar, vector, or matrix,' +\n        ` but got shape ${sparseIndices.shape}.`);\n  }\n\n  const numElems = sparseIndices.rank > 0 ? sparseIndices.shape[0] : 1;\n  const numDims = sparseIndices.rank > 1 ? sparseIndices.shape[1] : 1;\n\n  if (outputShape.length !== numDims) {\n    throw new Error(\n        'outputShape has incorrect number of elements:,' +\n        ` ${outputShape.length}, should be: ${numDims}.`);\n  }\n\n  const numValues = sparseValues.size;\n  if (!(sparseValues.rank === 0 ||\n        sparseValues.rank === 1 && numValues === numElems)) {\n    throw new Error(\n        'sparseValues has incorrect shape ' +\n        `${sparseValues.shape}, should be [] or [${numElems}]`);\n  }\n\n  if (sparseValues.dtype !== defaultValues.dtype) {\n    throw new Error('sparseValues.dtype must match defaultValues.dtype');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {SparseToDense, SparseToDenseAttrs, SparseToDenseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport * as sparse_to_dense from '../ops/sparse_to_dense_util';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ScalarLike, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Converts a sparse representation into a dense tensor.\n *\n * Builds an array dense with shape outputShape such that:\n *\n * // If sparseIndices is scalar\n * dense[i] = (i == sparseIndices ? sparseValues : defaultValue)\n *\n * // If sparseIndices is a vector, then for each i\n * dense[sparseIndices[i]] = sparseValues[i]\n *\n * // If sparseIndices is an n by d matrix, then for each i in [0, n)\n * dense[sparseIndices[i][0], ..., sparseIndices[i][d-1]] = sparseValues[i]\n * All other values in dense are set to defaultValue. If sparseValues is a\n * scalar, all sparse indices are set to this single value.\n *\n * If indices are repeated the final value is summed over all values for those\n * indices.\n *\n * ```js\n * const indices = tf.tensor1d([4, 5, 6, 1, 2, 3], 'int32');\n * const values = tf.tensor1d([10, 11, 12, 13, 14, 15], 'float32');\n * const shape = [8];\n * tf.sparseToDense(indices, values, shape).print();\n * ```\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape Shape of the dense output tensor. the type is inferred.\n * @param defaultValue Scalar. Value to set for indices not specified in\n * sparseIndices. Defaults to zero.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction sparseToDense_<R extends Rank>(\n    sparseIndices: Tensor|TensorLike, sparseValues: Tensor|TensorLike,\n    outputShape: ShapeMap[R], defaultValue: Scalar|ScalarLike = 0): Tensor<R> {\n  const $sparseIndices =\n      convertToTensor(sparseIndices, 'sparseIndices', 'sparseToDense', 'int32');\n  const $sparseValues =\n      convertToTensor(sparseValues, 'sparseValues', 'sparseToDense');\n  const $defaultValue = convertToTensor(\n      defaultValue, 'defaultValue', 'sparseToDense', $sparseValues.dtype);\n\n  sparse_to_dense.validateInput(\n      $sparseIndices, $sparseValues, outputShape, $defaultValue);\n\n  const inputs: SparseToDenseInputs = {\n    sparseIndices: $sparseIndices,\n    sparseValues: $sparseValues,\n    defaultValue: $defaultValue\n  };\n\n  const attrs: SparseToDenseAttrs = {outputShape};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.sparseToDense(\n          $sparseIndices, $sparseValues, outputShape, $defaultValue),\n      inputs as {} as NamedTensorMap, null /* grad */, SparseToDense,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const sparseToDense = op({sparseToDense_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherNd, GatherNdInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Gather slices from input tensor into a Tensor with shape specified by\n * `indices`.\n *\n * `indices` is an K-dimensional integer tensor, best thought of as a\n * (K-1)-dimensional tensor of indices into input, where each element defines a\n * slice of input:\n * output[\\\\(i_0, ..., i_{K-2}\\\\)] = input[indices[\\\\(i_0, ..., i_{K-2}\\\\)]]\n *\n * Whereas in `tf.gather`, `indices` defines slices into the first dimension of\n * input, in `tf.gatherND`, `indices` defines slices into the first N dimensions\n * of input, where N = indices.shape[-1].\n *\n * The last dimension of indices can be at most the rank of input:\n * indices.shape[-1] <= input.rank\n *\n * The last dimension of `indices` corresponds to elements\n * (if indices.shape[-1] == input.rank) or slices\n * (if indices.shape[-1] < input.rank) along dimension indices.shape[-1] of\n * input.\n * The output tensor has shape\n * indices.shape[:-1] + input.shape[indices.shape[-1]:]\n *\n * Note that on CPU, if an out of bound index is found, an error is returned. On\n * GPU, if an out of bound index is found, a 0 is stored in the corresponding\n * output value.\n *\n * ```js\n * const indices = tf.tensor2d([0, 1, 1, 0], [2,2], 'int32');\n * const input = tf.tensor2d([9, 10, 11, 12], [2, 2]);\n * tf.gatherND(input, indices).print() // [10, 11]\n * ```\n *\n * @param x The tensor from which to gather values.\n * @param indices Index tensor, must be of type int32.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction gatherND_(x: Tensor|TensorLike, indices: Tensor|TensorLike): Tensor {\n  const $indices = convertToTensor(indices, 'indices', 'gatherND', 'int32');\n  const $x = convertToTensor(x, 'x', 'gatherND');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.gatherND($x, $indices);\n  };\n\n  const inputs: GatherNdInputs = {params: $x, indices: $indices};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */, GatherNd);\n}\n\nexport const gatherND = op({gatherND_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\n/**\n * Normalize noise shape based on provided tensor and noise shape.\n *\n * @param x Tensor.\n * @param noiseShape The shape for the randomly generated keep/drop flags, as\n *   an array of numbers. Optional.\n * @returns Normalized noise shape.\n */\nexport function getNoiseShape(x: Tensor, noiseShape?: number[]): number[] {\n  if (noiseShape == null) {\n    return x.shape.slice();\n  }\n  if (util.arraysEqual(x.shape, noiseShape)) {\n    return noiseShape;\n  }\n  if (x.shape.length === noiseShape.length) {\n    const newDimension: number[] = [];\n    for (let i = 0; i < x.shape.length; i++) {\n      if (noiseShape[i] == null && x.shape[i] != null) {\n        newDimension.push(x.shape[i]);\n      } else {\n        newDimension.push(noiseShape[i]);\n      }\n    }\n    return newDimension;\n  }\n\n  return noiseShape;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {getNoiseShape} from './dropout_util';\nimport {floor} from './floor';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {randomUniform} from './random_uniform';\n\n/**\n * Computes dropout.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 2, 1]);\n * const rate = 0.75;\n * const output = tf.dropout(x, rate);\n * output.print();\n * ```\n *\n * @param x A floating point Tensor or TensorLike.\n * @param rate A float in the range [0, 1). The probability that each element\n *   of x is discarded.\n * @param noiseShape An array of numbers of type int32, representing the\n * shape for randomly generated keep/drop flags. If the noiseShape has null\n * value, it will be automatically replaced with the x's relative dimension\n * size. Optional.\n * @param seed Used to create random seeds. Optional.\n * @returns A Tensor of the same shape of x.\n *\n * @doc {heading: 'Operations', subheading: 'Dropout'}\n */\nfunction dropout_(\n    x: Tensor|TensorLike, rate: number, noiseShape?: number[],\n    seed?: number|string): Tensor {\n  const $x = convertToTensor(x, 'x', 'dropout');\n\n  util.assert(\n      $x.dtype === 'float32',\n      () => `x has to be a floating point tensor since it's going to be ` +\n          `scaled, but got a ${$x.dtype} tensor instead.`);\n  util.assert(\n      rate >= 0 && rate < 1,\n      () => `rate must be a float in the range [0, 1), but got ${rate}.`);\n\n  if (rate === 0) {\n    return x instanceof Tensor ? $x.clone() : $x;\n  }\n\n  const $noiseShape = getNoiseShape($x, noiseShape);\n  const keepProb = 1 - rate;\n  const multiplier = div(\n      floor(add(randomUniform($noiseShape, 0, 1, 'float32', seed), keepProb)),\n      keepProb);\n\n  return mul($x, multiplier);\n}\n\nexport const dropout = op({dropout_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {tensor1d} from './tensor1d';\n\nexport function enclosingPowerOfTwo(value: number) {\n  // Return 2**N for integer N such that 2**N >= value.\n  return Math.floor(Math.pow(2, Math.ceil(Math.log(value) / Math.log(2.0))));\n}\n\nexport function cosineWindow(\n    windowLength: number, a: number, b: number): Tensor1D {\n  const even = 1 - windowLength % 2;\n  const newValues = new Float32Array(windowLength);\n  for (let i = 0; i < windowLength; ++i) {\n    const cosArg = (2.0 * Math.PI * i) / (windowLength + even - 1);\n    newValues[i] = a - b * Math.cos(cosArg);\n  }\n  return tensor1d(newValues, 'float32');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch, getTypedArrayFromDType} from '../util';\nimport {tensor} from './tensor';\n\n/**\n * Returns whether the targets are in the top K predictions.\n *\n * ```js\n * const predictions = tf.tensor2d([[20, 10, 40, 30], [30, 50, -20, 10]]);\n * const targets = tf.tensor1d([2, 0]);\n * const precision = await tf.inTopKAsync(predictions, targets);\n * precision.print();\n * ```\n * @param predictions 2-D or higher `tf.Tensor` with last dimension being\n *     at least `k`.\n * @param targets 1-D or higher `tf.Tensor`.\n * @param k Optional Number of top elements to look at for computing precision,\n *     default to 1.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nasync function inTopKAsync_<T extends Tensor, U extends Tensor>(\n    predictions: T|TensorLike, targets: U|TensorLike, k = 1): Promise<U> {\n  const $predictions = convertToTensor(predictions, 'predictions', 'inTopK');\n  const $targets = convertToTensor(targets, 'targets', 'inTopK');\n\n  assert(\n      $predictions.rank > 1,\n      () => 'inTopK() expects the predictions to be of rank 2 or higher, ' +\n          `but got ${$predictions.rank}`);\n  assert(\n      $predictions.rank - 1 === $targets.rank,\n      () => `predictions rank should be 1 larger than ` +\n          `targets rank, but got predictions rank ` +\n          `${$predictions.rank} and targets rank ${$targets.rank}`);\n  assertShapesMatch(\n      $predictions.shape.slice(0, $predictions.shape.length - 1),\n      $targets.shape,\n      `predictions's shape should be align with the targets' shape, ` +\n          'except the last dimension.');\n  const lastDim = $predictions.shape[$predictions.shape.length - 1];\n  assert(\n      k > 0 && k <= lastDim,\n      () => `'k' passed to inTopK() must be > 0 && <= the predictions last ` +\n          `dimension (${lastDim}), but got ${k}`);\n\n  const predictionsVals = await $predictions.data();\n  const targetsVals = await $targets.data();\n\n  // Reshape predictionsVals into a 2d tensor [batch, lastDim]\n  // and look up topK along lastDim.\n  const [batch, size] = [predictionsVals.length / lastDim, lastDim];\n  const precision = getTypedArrayFromDType('bool', batch);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = predictionsVals.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    precision[b] = 0;\n    for (let i = 0; i < k; i++) {\n      if (valAndInd[i].index === targetsVals[b]) {\n        precision[b] = 1;\n        break;\n      }\n    }\n  }\n\n  if (predictions !== $predictions) {\n    $predictions.dispose();\n  }\n  if (targets !== $targets) {\n    $targets.dispose();\n  }\n\n  // Output precision has the same shape as targets.\n  return tensor(precision, $targets.shape, 'bool') as U;\n}\n\nexport const inTopKAsync = inTopKAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropFilter, Conv2DBackpropFilterAttrs, Conv2DBackpropFilterInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 2D convolution.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     [batch, height, width, inChannels]. If rank 3, batch of 1 is assumed.\n * @param dy The dy image, of rank 4 or rank 3, of shape\n *     [batch, height, width, outDepth]. If rank 3, batch of 1 is assumed.\n * @param filterShape The shape of the filter, length 4,\n *     [filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction conv2DBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2dDerFilter: input must be rank 4, but got shape ` +\n          `${x4D.shape}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerFilter: dy must be rank 4, but got shape ` +\n          `${dy4D.shape}.`);\n  util.assert(\n      filterShape.length === 4,\n      () => `Error in conv2dDerFilter: filterShape must be length 4, but got ` +\n          `${filterShape}.`);\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filterShape[2],\n      () => `Error in conv2dDerFilter: depth of input ${inDepth}) must ` +\n          `match input depth in filter (${filterShape[2]}.`);\n  util.assert(\n      outDepth === filterShape[3],\n      () => `Error in conv2dDerFilter: depth of dy (${outDepth}) must ` +\n          `match output depth for filter (${filterShape[3]}).`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerFilter: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, filterShape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    return backend.conv2dDerFilter(x4D, dy4D, convInfo);\n  };\n\n  const inputs: Conv2DBackpropFilterInputs = {x: x4D, dy: dy4D};\n  const attrs:\n      Conv2DBackpropFilterAttrs = {strides, pad, dataFormat, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv2DBackpropFilter, attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const conv2DBackpropFilter = op({conv2DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\nimport * as broadcast_util from './broadcast_util';\nimport {elu} from './elu';\nimport {Activation} from './fused_types';\nimport {mul} from './mul';\nimport {prelu} from './prelu';\nimport {relu} from './relu';\nimport {relu6} from './relu6';\nimport {reshape} from './reshape';\nimport {step} from './step';\nimport {sum} from './sum';\n\n// Returns gradient for fused activation.\nexport function getFusedDyActivation(\n    dy: Tensor, y: Tensor, activation: Activation): Tensor {\n  if (activation == null || activation === 'linear') {\n    return dy;\n  }\n  if (activation === 'relu') {\n    return mul(dy, step(y));\n  }\n  throw new Error(\n      `Cannot compute gradient for fused activation ${activation}.`);\n}\n\n// Returns gradient for fused bias.\nexport function getFusedBiasGradient(\n    bias: Tensor, dyActivation: Tensor): Tensor {\n  let res = dyActivation;\n  const reduceAxes =\n      broadcast_util.getReductionAxes(bias.shape, dyActivation.shape);\n  if (reduceAxes.length > 0) {\n    res = sum(res, reduceAxes);\n  }\n  return reshape(res, bias.shape);\n}\n\nexport function applyActivation(\n    x: Tensor, activation: Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return x;\n  } else if (activation === 'relu') {\n    return relu(x);\n  } else if (activation === 'elu') {\n    return elu(x);\n  } else if (activation === 'relu6') {\n    return relu6(x);\n  } else if (activation === 'prelu') {\n    return prelu(x, preluActivationWeights);\n  }\n  throw new Error(`Unknown fused activation ${activation}.`);\n}\n\n// Whether we should call fused ops.\nexport const shouldFuse = (gradientDepth: number, activation: Activation) => {\n  const gradientMode = gradientDepth > 0;\n  return !gradientMode || activation === 'linear';\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedConv2D, FusedConv2DAttrs, FusedConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {conv2d as unfusedConv2d} from '../conv2d';\nimport {conv2DBackpropFilter} from '../conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../conv2d_backprop_input';\nimport * as conv_util from '../conv_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes a 2D convolution over the input x, optionally fused with adding a\n * bias and applying an activation.\n *\n * ```js\n * const inputDepth = 2;\n * const inShape = [2, 2, 2, inputDepth];\n * const outputDepth = 2;\n * const fSize = 1;\n * const pad = 0;\n * const strides = 1;\n *\n * const x = tf.tensor4d( [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,\n * 16], inShape);\n * const w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth,\n * outputDepth]);\n *\n * tf.fused.conv2d({ x, filter: w, strides, pad, dataFormat: 'NHWC',\n * dilations: [1, 1], bias: tf.scalar(5), activation: 'relu' }).print();\n * ```\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid` output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`) to be\n *     applied\n *      after biasAdd.\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  activation = activation || 'linear';\n\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused conv2d: input must be rank 4, but got rank ` +\n          `${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${x4D.shape[3]}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in conv2d: got dataFormat of ${\n          dataFormat} but only NHWC is currently supported.`);\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused conv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    const [$filter, x4D, y, $bias] =\n        saved as [Tensor4D, Tensor4D, Tensor4D, Tensor];\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused conv2D: ' +\n            `dilation rates greater than 1 ` +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const xDer =\n        conv2DBackpropInput(x4D.shape, dyActivation, $filter, strides, pad);\n    const filterDer =\n        conv2DBackpropFilter(x4D, dyActivation, $filter.shape, strides, pad);\n    const der: Tensor[] = [xDer, filterDer];\n\n    if ($bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      der.push(biasDer);\n    }\n    return der;\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedConv2d({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n\n  const attrs: FusedConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const conv2d = op({fusedConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropFilterInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    convInfo: conv_util.Conv2DInfo): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerFilter(x4D, dy4D, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropFilterInputs = {x: x4D, dy: dy4D};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             DepthwiseConv2dNativeBackpropFilter) as Tensor4D;\n}\n\nexport const depthwiseConv2dNativeBackpropFilter =\n    op({depthwiseConv2dNativeBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropInput, DepthwiseConv2dNativeBackpropInputInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, convInfo: conv_util.Conv2DInfo): T {\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerInput(dy4D, filter, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropInputInputs = {dy: dy4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null,\n      DepthwiseConv2dNativeBackpropInput);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2dNativeBackpropInput =\n    op({depthwiseConv2dNativeBackpropInput_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedDepthwiseConv2D, FusedDepthwiseConv2DAttrs, FusedDepthwiseConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport * as conv_util from '../conv_util';\nimport {depthwiseConv2d as unfusedDepthwiseConv2d} from '../depthwise_conv2d';\nimport {depthwiseConv2dNativeBackpropFilter} from '../depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../depthwise_conv2d_native_backprop_input';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes depthwise 2D convolution, optionally fused with adding a\n * bias and applying an activation.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`).\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedDepthwiseConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedDepthwiseConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused depthwiseConv2d: filter must be rank 4, ` +\n          `but got rank ${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in fused depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () =>\n          'Error in fused depthwiseConv2d: Either strides or dilations must ' +\n          `be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused depthwiseConv2d: pad must be an integer when ` +\n            `using dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n      true /* depthwise */);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused depthwiseConv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused depthwiseConv2d: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${dilations}'`);\n    const [$filter, x4D, y, bias] = saved;\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    const xDer = depthwiseConv2dNativeBackpropInput(\n        (x4D as Tensor4D).shape, dyActivation, $filter as Tensor4D, convInfo);\n    const filterDer = depthwiseConv2dNativeBackpropFilter(\n        x4D as Tensor4D, dyActivation, ($filter as Tensor4D).shape, convInfo);\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [xDer, filterDer, biasDer];\n    }\n    return [xDer, filterDer];\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedDepthwiseConv2D({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedDepthwiseConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: FusedDepthwiseConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const depthwiseConv2d = op({fusedDepthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {_FusedMatMul, _FusedMatMulAttrs, _FusedMatMulInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {matMul as unfusedMatMul} from '../mat_mul';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes the dot product of two matrices with optional activation and bias.\n *\n * ```js\n * const a = tf.tensor2d([-1, -2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const bias = tf.tensor2d([1, 2], [1, 2]);\n *\n * tf.fused.matMul({a, b, bias, activation: 'relu'}).print();\n * ```\n *\n * @param obj An object with the following properties:\n * - `a` First matrix in dot product operation.\n * - `b` Second matrix in dot product operation.\n * - `transposeA` If true, `a` is transposed before multiplication.\n * - `transposeB` If true, `b` is transposed before multiplication.\n * - `bias` Matrix to be added to the result.\n * - `activation` Name of activation kernel (defaults to `linear`).\n * - `preluActivationWeights` Tensor of prelu weights.\n */\nfunction fusedMatMul_<T extends Tensor>({\n  a,\n  b,\n  transposeA = false,\n  transposeB = false,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  a: T|TensorLike,\n  b: T|TensorLike,\n  transposeA?: boolean,\n  transposeB?: boolean,\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedMatMul(a, b, transposeA, transposeB);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  let $a = convertToTensor(a, 'a', 'fused matMul');\n  let $b = convertToTensor(b, 'b', 'fused matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () =>\n          `Error in fused matMul: inputs must have the same rank of at least ` +\n          `2, got ranks ${$a.rank} and ${$b.rank}.`);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in fused matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in fused matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D: Tensor3D = transposeA ?\n      reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n      reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D: Tensor3D = transposeB ?\n      reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n      reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused matMul');\n    [$bias] = makeTypesMatch($bias, $a);\n\n    broadcast_util.assertAndGetBroadcastShape(outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused matMul');\n  }\n\n  const grad = (dy: Tensor3D, saved: Tensor[]) => {\n    const [a3D, b3D, y, $bias] = saved;\n    // we reshape dy because the result of the forward is not\n    // necessarily going to be a 3d tensor due to a reshape done at the end of\n    // the customOp.\n    const dyActivation =\n        getFusedDyActivation(reshape(dy, y.shape), y, activation);\n    let aDer: Tensor;\n    let bDer: Tensor;\n\n    if (!transposeA && !transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, true, false);\n    } else if (!transposeA && transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, false);\n      bDer = unfusedMatMul(dyActivation, a3D, true, false);\n    } else if (transposeA && !transposeB) {\n      aDer = unfusedMatMul(b3D, dyActivation, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, false, false);\n    } else {\n      aDer = unfusedMatMul(b3D, dyActivation, true, true);\n      bDer = unfusedMatMul(dyActivation, a3D, true, true);\n    }\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [aDer, bDer, biasDer];\n    } else {\n      return [aDer, bDer];\n    }\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const y = backend.fusedBatchMatMul({\n      a: a3D,\n      b: b3D,\n      transposeA,\n      transposeB,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return y;\n  };\n\n  const inputs: _FusedMatMulInputs = {\n    a: a3D,\n    b: b3D,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: _FusedMatMulAttrs = {transposeA, transposeB, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((a3D: Tensor3D, b3D: Tensor3D, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n    return customOp(a3D, b3D) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (a3D: Tensor3D, b3D: Tensor3D, $bias: Tensor, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res, $bias]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n\n    return customOpWithBias(a3D, b3D, $bias) as T;\n  }\n}\n\nexport const matMul = op({fusedMatMul_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {conv2d} from './fused/conv2d';\nimport {depthwiseConv2d} from './fused/depthwise_conv2d';\nimport {matMul} from './fused/mat_mul';\nimport {Activation} from './fused_types';\n\nexport {Activation, conv2d, depthwiseConv2d, matMul};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a hamming window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hammingWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hammingWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.54, 0.46);\n}\nexport const hammingWindow = op({hammingWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a Hann window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hannWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hannWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.5, 0.5);\n}\n\nexport const hannWindow = op({hannWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {fill} from '../fill';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {tensor2d} from '../tensor2d';\n\n/**\n * Expands input into frames of frameLength.\n * Slides a window size with frameStep.\n *\n * ```js\n * tf.signal.frame([1, 2, 3], 2, 1).print();\n * ```\n * @param signal The input tensor to be expanded\n * @param frameLength Length of each frame\n * @param frameStep The frame hop size in samples.\n * @param padEnd Whether to pad the end of signal with padValue.\n * @param padValue An number to use where the input signal does\n *     not exist when padEnd is True.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction frame_(\n    signal: Tensor1D, frameLength: number, frameStep: number, padEnd = false,\n    padValue = 0): Tensor {\n  let start = 0;\n  const output: Tensor[] = [];\n  while (start + frameLength <= signal.size) {\n    output.push(slice(signal, start, frameLength));\n    start += frameStep;\n  }\n\n  if (padEnd) {\n    while (start < signal.size) {\n      const padLen = (start + frameLength) - signal.size;\n      const pad = concat([\n        slice(signal, start, frameLength - padLen), fill([padLen], padValue)\n      ]);\n      output.push(pad);\n      start += frameStep;\n    }\n  }\n\n  if (output.length === 0) {\n    return tensor2d([], [0, frameLength]);\n  }\n\n  return reshape(concat(output), [output.length, frameLength]);\n}\nexport const frame = op({frame_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {enclosingPowerOfTwo} from '../signal_ops_util';\nimport {slice} from '../slice';\nimport {rfft} from '../spectral/rfft';\n\nimport {frame} from './frame';\nimport {hannWindow} from './hann_window';\n\n/**\n * Computes the Short-time Fourier Transform of signals\n * See: https://en.wikipedia.org/wiki/Short-time_Fourier_transform\n *\n * ```js\n * const input = tf.tensor1d([1, 1, 1, 1, 1])\n * tf.signal.stft(input, 3, 1).print();\n * ```\n * @param signal 1-dimensional real value tensor.\n * @param frameLength The window length of samples.\n * @param frameStep The number of samples to step.\n * @param fftLength The size of the FFT to apply.\n * @param windowFn A callable that takes a window length and returns 1-d tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction stft_(\n    signal: Tensor1D, frameLength: number, frameStep: number,\n    fftLength?: number,\n    windowFn: (length: number) => Tensor1D = hannWindow): Tensor {\n  if (fftLength == null) {\n    fftLength = enclosingPowerOfTwo(frameLength);\n  }\n  const framedSignal = frame(signal, frameLength, frameStep);\n  const windowedSignal = mul(framedSignal, windowFn(frameLength));\n  const output: Tensor[] = [];\n  for (let i = 0; i < framedSignal.shape[0]; i++) {\n    output.push(\n        rfft(slice(windowedSignal, [i, 0], [1, frameLength]), fftLength));\n  }\n  return concat(output);\n}\nexport const stft = op({stft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {CropAndResize, CropAndResizeAttrs, CropAndResizeInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor1D, Tensor2D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Extracts crops from the input image tensor and resizes them using bilinear\n * sampling or nearest neighbor sampling (possibly with aspect ratio change)\n * to a common output size specified by crop_size.\n *\n * @param image 4d tensor of shape `[batch,imageHeight,imageWidth, depth]`,\n *     where imageHeight and imageWidth must be positive, specifying the\n *     batch of images from which to take crops\n * @param boxes 2d float32 tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the normalized\n *     coordinates of the box in the boxInd[i]'th image in the batch\n * @param boxInd 1d int32 tensor of shape `[numBoxes]` with values in range\n *     `[0, batch)` that specifies the image that the `i`-th box refers to.\n * @param cropSize 1d int32 tensor of 2 elements `[cropHeigh, cropWidth]`\n *     specifying the size to which all crops are resized to.\n * @param method Optional string from `'bilinear' | 'nearest'`,\n *     defaults to bilinear, which specifies the sampling method for resizing\n * @param extrapolationValue A threshold for deciding when to remove boxes based\n *     on score. Defaults to 0.\n * @return A 4D tensor of the shape `[numBoxes,cropHeight,cropWidth,depth]`\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction cropAndResize_(\n    image: Tensor4D|TensorLike,\n    boxes: Tensor2D|TensorLike,\n    boxInd: Tensor1D|TensorLike,\n    cropSize: [number, number],\n    method?: 'bilinear'|'nearest',\n    extrapolationValue?: number,\n    ): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'cropAndResize');\n  const $boxes = convertToTensor(boxes, 'boxes', 'cropAndResize', 'float32');\n  const $boxInd = convertToTensor(boxInd, 'boxInd', 'cropAndResize', 'int32');\n  method = method || 'bilinear';\n  extrapolationValue = extrapolationValue || 0;\n\n  const numBoxes = $boxes.shape[0];\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in cropAndResize: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n  util.assert(\n      $boxes.rank === 2 && $boxes.shape[1] === 4,\n      () => `Error in cropAndResize: boxes must be have size [${numBoxes},4] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      $boxInd.rank === 1 && $boxInd.shape[0] === numBoxes,\n      () => `Error in cropAndResize: boxInd must be have size [${numBoxes}] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      cropSize.length === 2,\n      () => `Error in cropAndResize: cropSize must be of length 2, but got ` +\n          `length ${cropSize.length}.`);\n  util.assert(\n      cropSize[0] >= 1 && cropSize[1] >= 1,\n      () => `cropSize must be atleast [1,1], but was ${cropSize}`);\n  util.assert(\n      method === 'bilinear' || method === 'nearest',\n      () => `method must be bilinear or nearest, but was ${method}`);\n\n  const forward: ForwardFunc<Tensor4D> = (backend) => backend.cropAndResize(\n      $image, $boxes, $boxInd, cropSize, method, extrapolationValue);\n\n  const inputs:\n      CropAndResizeInputs = {image: $image, boxes: $boxes, boxInd: $boxInd};\n  const attrs: CropAndResizeAttrs = {method, extrapolationValue, cropSize};\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, CropAndResize,\n      attrs as {} as NamedAttrMap);\n  return res;\n}\n\nexport const cropAndResize = op({cropAndResize_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FlipLeftRight, FlipLeftRightInputs} from '../../kernel_names';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {op} from '../operation';\n\n/**\n * Flips the image left to right. Currently available in the CPU, WebGL, and\n * WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n */\n/** @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'} */\nfunction flipLeftRight_(image: Tensor4D|TensorLike): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'flipLeftRight', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in flipLeftRight: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: FlipLeftRightInputs = {image: $image};\n  const res =\n      ENGINE.runKernel(FlipLeftRight, inputs as {} as NamedTensorMap, {});\n  return res as Tensor4D;\n}\n\nexport const flipLeftRight = op({flipLeftRight_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Rotates the input image tensor counter-clockwise with an optional offset\n * center of rotation. Currently available in the CPU, WebGL, and WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n * @param radians The amount of rotation.\n * @param fillValue The value to fill in the empty space leftover\n *     after rotation. Can be either a single grayscale value (0-255), or an\n *     array of three numbers `[red, green, blue]` specifying the red, green,\n *     and blue channels. Defaults to `0` (black).\n * @param center The center of rotation. Can be either a single value (0-1), or\n *     an array of two numbers `[centerX, centerY]`. Defaults to `0.5` (rotates\n *     the image around its center).\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction rotateWithOffset_(\n    image: Tensor4D|TensorLike, radians: number,\n    fillValue: number|[number, number, number] = 0,\n    center: number|[number, number] = 0.5): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'rotateWithOffset', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in rotateWithOffset: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: RotateWithOffsetInputs = {image: $image};\n  const attrs: RotateWithOffsetAttrs = {radians, fillValue, center};\n  const res = ENGINE.runKernel(\n      RotateWithOffset, inputs as {} as NamedTensorMap,\n      attrs as {} as NamedAttrMap);\n  return res as Tensor4D;\n}\n\nexport const rotateWithOffset = op({rotateWithOffset_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport * as util from '../util';\n\nfunction nonMaxSuppSanityCheck(\n    boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma?: number): {\n  maxOutputSize: number,\n  iouThreshold: number,\n  scoreThreshold: number,\n  softNmsSigma: number\n} {\n  if (iouThreshold == null) {\n    iouThreshold = 0.5;\n  }\n  if (scoreThreshold == null) {\n    scoreThreshold = Number.NEGATIVE_INFINITY;\n  }\n  if (softNmsSigma == null) {\n    softNmsSigma = 0.0;\n  }\n\n  const numBoxes = boxes.shape[0];\n  maxOutputSize = Math.min(maxOutputSize, numBoxes);\n\n  util.assert(\n      0 <= iouThreshold && iouThreshold <= 1,\n      () => `iouThreshold must be in [0, 1], but was '${iouThreshold}'`);\n  util.assert(\n      boxes.rank === 2,\n      () => `boxes must be a 2D tensor, but was of rank '${boxes.rank}'`);\n  util.assert(\n      boxes.shape[1] === 4,\n      () =>\n          `boxes must have 4 columns, but 2nd dimension was ${boxes.shape[1]}`);\n  util.assert(scores.rank === 1, () => 'scores must be a 1D tensor');\n  util.assert(\n      scores.shape[0] === numBoxes,\n      () => `scores has incompatible shape with boxes. Expected ${numBoxes}, ` +\n          `but was ${scores.shape[0]}`);\n  util.assert(\n      0 <= softNmsSigma && softNmsSigma <= 1,\n      () => `softNmsSigma must be in [0, 1], but was '${softNmsSigma}'`);\n  return {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n}\n\nexport {nonMaxSuppSanityCheck};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV3} from '../../kernel_names';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\nfunction nonMaxSuppression_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Tensor1D {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const attrs = {maxOutputSize, iouThreshold, scoreThreshold};\n  return ENGINE.runKernelFunc(\n      b => b.nonMaxSuppression(\n          $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold),\n      {boxes: $boxes, scores: $scores}, null /* grad */, NonMaxSuppressionV3,\n      attrs);\n}\n\nexport const nonMaxSuppression = op({nonMaxSuppression_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inserts a value into a sorted array. This method allows duplicate, meaning it\n * allows inserting duplicate value, in which case, the element will be inserted\n * at the lowest index of the value.\n * @param arr The array to modify.\n * @param element The element to insert.\n * @param comparator Optional. If no comparator is specified, elements are\n * compared using array_util.defaultComparator, which is suitable for Strings\n * and Numbers in ascending arrays. If the array contains multiple instances of\n * the target value, the left-most instance will be returned. To provide a\n * comparator, it should take 2 arguments to compare and return a negative,\n * zero, or a positive number.\n */\nexport function binaryInsert<T>(\n    arr: T[], element: T, comparator?: (a: T, b: T) => number) {\n  const index = binarySearch(arr, element, comparator);\n  const insertionPoint = index < 0 ? -(index + 1) : index;\n  arr.splice(insertionPoint, 0, element);\n}\n\n/**\n * Searches the array for the target using binary search, returns the index\n * of the found element, or position to insert if element not found. If no\n * comparator is specified, elements are compared using array_\n * util.defaultComparator, which is suitable for Strings and Numbers in\n * ascending arrays. If the array contains multiple instances of the target\n * value, the left-most instance will be returned.\n * @param arr The array to be searched in.\n * @param target The target to be searched for.\n * @param comparator Should take 2 arguments to compare and return a negative,\n *    zero, or a positive number.\n * @return Lowest index of the target value if found, otherwise the insertion\n *    point where the target should be inserted, in the form of\n *    (-insertionPoint - 1).\n */\nexport function binarySearch<T>(\n    arr: T[], target: T, comparator?: (a: T, b: T) => number) {\n  return binarySearch_(arr, target, comparator || defaultComparator);\n}\n\n/**\n * Compares its two arguments for order.\n * @param a The first element to be compared.\n * @param b The second element to be compared.\n * @return A negative number, zero, or a positive number as the first\n *     argument is less than, equal to, or greater than the second.\n */\nfunction defaultComparator<T>(a: T, b: T): number {\n  return a > b ? 1 : a < b ? -1 : 0;\n}\n\nfunction binarySearch_<T>(\n    arr: T[], target: T, comparator: (a: T, b: T) => number) {\n  let left = 0;\n  let right = arr.length;\n  let middle = 0;\n  let found = false;\n  while (left < right) {\n    middle = left + ((right - left) >>> 1);\n    const compareResult = comparator(target, arr[middle]);\n    if (compareResult > 0) {\n      left = middle + 1;\n    } else {\n      right = middle;\n      // If compareResult is 0, the value is found. We record it is found,\n      // and then keep looking because there may be duplicate.\n      found = !compareResult;\n    }\n  }\n\n  return found ? left : -left - 1;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Implementation of the NonMaxSuppression kernel shared between webgl and cpu.\n */\n\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\n\nimport {binaryInsert} from './array_util';\n\ninterface Candidate {\n  score: number;\n  boxIndex: number;\n  suppressBeginIndex: number;\n}\n\nexport function nonMaxSuppressionV3Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number): Tensor1D {\n  return nonMaxSuppressionImpl_(\n             boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n             0 /* softNmsSigma */)\n             .selectedIndices as Tensor1D;\n}\n\nexport function nonMaxSuppressionV4Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    padToMaxOutputSize: boolean): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n      0 /* softNmsSigma */, false /* returnScoresTensor */,\n      padToMaxOutputSize /* padToMaxOutputSize */, true\n      /* returnValidOutputs */);\n}\n\nexport function nonMaxSuppressionV5Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    softNmsSigma: number): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma,\n      true /* returnScoresTensor */);\n}\n\nfunction nonMaxSuppressionImpl_(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma: number,\n    returnScoresTensor = false, padToMaxOutputSize = false,\n    returnValidOutputs = false): NamedTensorMap {\n  // The list is sorted in ascending order, so that we can always pop the\n  // candidate with the largest score in O(1) time.\n  const candidates = [];\n\n  for (let i = 0; i < scores.length; i++) {\n    if (scores[i] > scoreThreshold) {\n      candidates.push({score: scores[i], boxIndex: i, suppressBeginIndex: 0});\n    }\n  }\n\n  candidates.sort(ascendingComparator);\n\n  // If softNmsSigma is 0, the outcome of this algorithm is exactly same as\n  // before.\n  const scale = softNmsSigma > 0 ? (-0.5 / softNmsSigma) : 0.0;\n\n  const selectedIndices: number[] = [];\n  const selectedScores: number[] = [];\n\n  while (selectedIndices.length < maxOutputSize && candidates.length > 0) {\n    const candidate = candidates.pop();\n    const {score: originalScore, boxIndex, suppressBeginIndex} = candidate;\n\n    if (originalScore < scoreThreshold) {\n      break;\n    }\n\n    // Overlapping boxes are likely to have similar scores, therefore we\n    // iterate through the previously selected boxes backwards in order to\n    // see if candidate's score should be suppressed. We use\n    // suppressBeginIndex to track and ensure a candidate can be suppressed\n    // by a selected box no more than once. Also, if the overlap exceeds\n    // iouThreshold, we simply ignore the candidate.\n    let ignoreCandidate = false;\n    for (let j = selectedIndices.length - 1; j >= suppressBeginIndex; --j) {\n      const iou = intersectionOverUnion(boxes, boxIndex, selectedIndices[j]);\n\n      if (iou >= iouThreshold) {\n        ignoreCandidate = true;\n        break;\n      }\n\n      candidate.score =\n          candidate.score * suppressWeight(iouThreshold, scale, iou);\n\n      if (candidate.score <= scoreThreshold) {\n        break;\n      }\n    }\n\n    // At this point, if `candidate.score` has not dropped below\n    // `scoreThreshold`, then we know that we went through all of the\n    // previous selections and can safely update `suppressBeginIndex` to the\n    // end of the selected array. Then we can re-insert the candidate with\n    // the updated score and suppressBeginIndex back in the candidate list.\n    // If on the other hand, `candidate.score` has dropped below the score\n    // threshold, we will not add it back to the candidates list.\n    candidate.suppressBeginIndex = selectedIndices.length;\n\n    if (!ignoreCandidate) {\n      // Candidate has passed all the tests, and is not suppressed, so\n      // select the candidate.\n      if (candidate.score === originalScore) {\n        selectedIndices.push(boxIndex);\n        selectedScores.push(candidate.score);\n      } else if (candidate.score > scoreThreshold) {\n        // Candidate's score is suppressed but is still high enough to be\n        // considered, so add back to the candidates list.\n        binaryInsert(candidates, candidate, ascendingComparator);\n      }\n    }\n  }\n\n  // NonMaxSuppressionV4 feature: padding output to maxOutputSize.\n  const validOutputs = selectedIndices.length;\n  const elemsToPad = maxOutputSize - validOutputs;\n\n  if (padToMaxOutputSize && elemsToPad > 0) {\n    selectedIndices.push(...new Array(elemsToPad).fill(0));\n    selectedScores.push(...new Array(elemsToPad).fill(0.0));\n  }\n\n  const result:\n      NamedTensorMap = {selectedIndices: tensor1d(selectedIndices, 'int32')};\n\n  if (returnScoresTensor) {\n    result['selectedScores'] = tensor1d(selectedScores, 'float32');\n  }\n\n  if (returnValidOutputs) {\n    result['validOutputs'] = scalar(validOutputs, 'int32');\n  }\n\n  return result;\n}\n\nfunction intersectionOverUnion(boxes: TypedArray, i: number, j: number) {\n  const iCoord = boxes.subarray(i * 4, i * 4 + 4);\n  const jCoord = boxes.subarray(j * 4, j * 4 + 4);\n  const yminI = Math.min(iCoord[0], iCoord[2]);\n  const xminI = Math.min(iCoord[1], iCoord[3]);\n  const ymaxI = Math.max(iCoord[0], iCoord[2]);\n  const xmaxI = Math.max(iCoord[1], iCoord[3]);\n  const yminJ = Math.min(jCoord[0], jCoord[2]);\n  const xminJ = Math.min(jCoord[1], jCoord[3]);\n  const ymaxJ = Math.max(jCoord[0], jCoord[2]);\n  const xmaxJ = Math.max(jCoord[1], jCoord[3]);\n  const areaI = (ymaxI - yminI) * (xmaxI - xminI);\n  const areaJ = (ymaxJ - yminJ) * (xmaxJ - xminJ);\n  if (areaI <= 0 || areaJ <= 0) {\n    return 0.0;\n  }\n  const intersectionYmin = Math.max(yminI, yminJ);\n  const intersectionXmin = Math.max(xminI, xminJ);\n  const intersectionYmax = Math.min(ymaxI, ymaxJ);\n  const intersectionXmax = Math.min(xmaxI, xmaxJ);\n  const intersectionArea = Math.max(intersectionYmax - intersectionYmin, 0.0) *\n      Math.max(intersectionXmax - intersectionXmin, 0.0);\n  return intersectionArea / (areaI + areaJ - intersectionArea);\n}\n\n// A Gaussian penalty function, this method always returns values in [0, 1].\n// The weight is a function of similarity, the more overlap two boxes are, the\n// smaller the weight is, meaning highly overlapping boxe will be significantly\n// penalized. On the other hand, a non-overlapping box will not be penalized.\nfunction suppressWeight(iouThreshold: number, scale: number, iou: number) {\n  const weight = Math.exp(scale * iou * iou);\n  return iou <= iouThreshold ? weight : 0.0;\n}\n\nfunction ascendingComparator(c1: Candidate, c2: Candidate) {\n  // For objects with same scores, we make the object with the larger index go\n  // first. In an array that pops from the end, this means that the object with\n  // the smaller index will be popped first. This ensures the same output as\n  // the TensorFlow python version.\n  return (c1.score - c2.score) ||\n      ((c1.score === c2.score) && (c2.boxIndex - c1.boxIndex));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {nonMaxSuppressionV3Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This is the async version of `nonMaxSuppression`\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @return A 1D tensor with the selected box indices.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Promise<Tensor1D> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV3Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionAsync = nonMaxSuppressionAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionWithScore_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const inputs: NonMaxSuppressionV5Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV5Attrs =\n      {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV5, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], selectedScores: result[1]};\n}\n\nexport const nonMaxSuppressionWithScore = op({nonMaxSuppressionWithScore_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV5Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionWithScoreAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV5Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionWithScoreAsync = nonMaxSuppressionWithScoreAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionPadded_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const inputs: NonMaxSuppressionV4Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV4Attrs = {\n    maxOutputSize: $maxOutputSize,\n    iouThreshold: $iouThreshold,\n    scoreThreshold: $scoreThreshold,\n    padToMaxOutputSize\n  };\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV4, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], validOutputs: result[1]};\n}\n\nexport const nonMaxSuppressionPadded = op({nonMaxSuppressionPadded_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV4Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionPaddedAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const [boxesVals, scoresVals] =\n      await Promise.all([$boxes.data(), $scores.data()]);\n\n  // We call a cpu based impl directly with the typedarray data here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV4Impl(\n      boxesVals, scoresVals, $maxOutputSize, $iouThreshold, $scoreThreshold,\n      padToMaxOutputSize);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionPaddedAsync = nonMaxSuppressionPaddedAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Bilinear resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeBilinear_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeBilinear');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeBilinear: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () => `Error in resizeBilinear: new shape must 2D, but got shape ` +\n          `${size}.`);\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n\n  const [newHeight, newWidth] = size;\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeBilinear(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const inputs: ResizeBilinearInputs = {images: batchImages};\n  const attrs: ResizeBilinearAttrs = {alignCorners, size};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeBilinear, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeBilinear = op({resizeBilinear_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * NearestNeighbor resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeNearestNeighbor_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeNearestNeighbor');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeNearestNeighbor: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () =>\n          `Error in resizeNearestNeighbor: new shape must 2D, but got shape ` +\n          `${size}.`);\n  util.assert(\n      $images.dtype === 'float32' || $images.dtype === 'int32',\n      () => '`images` must have `int32` or `float32` as dtype');\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n  const [newHeight, newWidth] = size;\n\n  const inputs: ResizeNearestNeighborInputs = {images: batchImages};\n  const attrs: ResizeNearestNeighborAttrs = {alignCorners, size};\n\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeNearestNeighbor(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeNearestNeighbor, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeNearestNeighbor = op({resizeNearestNeighbor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assert} from '../../util';\n\nimport {greaterEqual} from '../greater_equal';\nimport {lessEqual} from '../less_equal';\nimport {logicalAnd} from '../logical_and';\nimport {op} from '../operation';\nimport {range} from '../range';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\nimport {zeros} from '../zeros';\n\n/**\n * Copy a tensor setting everything outside a central band in each innermost\n * matrix to zero.\n *\n * The band part is computed as follows: Assume input has `k` dimensions\n * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where\n * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`.\n * The indicator function\n * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))`\n * `&& (num_upper < 0 || (n-m) <= num_upper)`\n *\n * ```js\n * const x = tf.tensor2d([[ 0,  1,  2, 3],\n *                        [-1,  0,  1, 2],\n *                        [-2, -1,  0, 1],\n *                        [-3, -2, -1, 0]]);\n * let y = tf.linalg.bandPart(x, 1, -1);\n * y.print(); // [[ 0,  1,  2, 3],\n *            //  [-1,  0,  1, 2],\n *            //  [ 0, -1,  0, 1],\n *            //  [ 0, 0 , -1, 0]]\n * let z = tf.linalg.bandPart(x, 2, 1);\n * z.print(); // [[ 0,  1,  0, 0],\n *            //  [-1,  0,  1, 0],\n *            //  [-2, -1,  0, 1],\n *            //  [ 0, -2, -1, 0]]\n * ```\n *\n * @param x Rank `k` tensor\n * @param numLower Number of subdiagonals to keep.\n *   If negative, keep entire lower triangle.\n * @param numUpper Number of subdiagonals to keep.\n *   If negative, keep entire upper triangle.\n * @returns Rank `k` tensor of the same shape as input.\n *   The extracted banded tensor.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction bandPart_<T extends Tensor>(\n    a: T|TensorLike, numLower: number, numUpper: number): T {\n  assert(\n      numLower % 1 === 0,\n      () => `bandPart(): numLower must be an integer, got ${numLower}.`);\n  assert(\n      numUpper % 1 === 0,\n      () => `bandPart(): numUpper must be an integer, got ${numUpper}.`);\n\n  const $a = convertToTensor(a, 'a', 'bandPart');\n\n  assert(\n      $a.rank >= 2,\n      () => `bandPart(): Rank must be at least 2, got ${$a.rank}.`);\n\n  const shape = $a.shape;\n  const [M, N] = $a.shape.slice(-2);\n\n  if (!(numLower <= M)) {\n    throw new Error(\n        `bandPart(): numLower (${numLower})` +\n        ` must not be greater than the number of rows (${M}).`);\n  }\n  if (!(numUpper <= N)) {\n    throw new Error(\n        `bandPart(): numUpper (${numUpper})` +\n        ` must not be greater than the number of columns (${N}).`);\n  }\n\n  if (numLower < 0) {\n    numLower = M;\n  }\n  if (numUpper < 0) {\n    numUpper = N;\n  }\n\n  const i = reshape(range(0, M, 1, 'int32'), [-1, 1]);\n  const j = range(0, N, 1, 'int32');\n  const ij = sub(i, j);\n\n  const inBand = logicalAnd(\n      lessEqual(ij, scalar(+numLower, 'int32')),\n      greaterEqual(ij, scalar(-numUpper, 'int32')));\n\n  const zero = zeros([M, N], $a.dtype);\n\n  return reshape(\n             stack(unstack(reshape($a, [-1, M, N]))\n                       .map(mat => where(inBand, mat, zero))),\n             shape) as T;\n}\n\nexport const bandPart = op({bandPart_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {div} from '../div';\nimport {mul} from '../mul';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {split} from '../split';\nimport {squeeze} from '../squeeze';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\n/**\n * Gram-Schmidt orthogonalization.\n *\n * ```js\n * const x = tf.tensor2d([[1, 2], [3, 4]]);\n * let y = tf.linalg.gramSchmidt(x);\n * y.print();\n * console.log('Othogonalized:');\n * y.dot(y.transpose()).print();  // should be nearly the identity matrix.\n * console.log('First row direction maintained:');\n * const data = await y.array();\n * console.log(data[0][1] / data[0][0]);  // should be nearly 2.\n * ```\n *\n * @param xs The vectors to be orthogonalized, in one of the two following\n *   formats:\n *   - An Array of `tf.Tensor1D`.\n *   - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows\n *     of `xs`.\n *   In each case, all the vectors must have the same length and the length\n *   must be greater than or equal to the number of vectors.\n * @returns The orthogonalized and normalized vectors or matrix.\n *   Orthogonalization means that the vectors or the rows of the matrix\n *   are orthogonal (zero inner products). Normalization means that each\n *   vector or each row of the matrix has an L2 norm that equals `1`.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D {\n  let inputIsTensor2D: boolean;\n  if (Array.isArray(xs)) {\n    inputIsTensor2D = false;\n    assert(\n        xs != null && xs.length > 0,\n        () => 'Gram-Schmidt process: input must not be null, undefined, or ' +\n            'empty');\n    const dim = xs[0].shape[0];\n    for (let i = 1; i < xs.length; ++i) {\n      assert(\n          xs[i].shape[0] === dim,\n          () =>\n              'Gram-Schmidt: Non-unique lengths found in the input vectors: ' +\n              `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`);\n    }\n  } else {\n    inputIsTensor2D = true;\n    xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0]));\n  }\n\n  assert(\n      xs.length <= xs[0].shape[0],\n      () => `Gram-Schmidt: Number of vectors (${\n                (xs as Tensor1D[]).length}) exceeds ` +\n          `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`);\n\n  const ys: Tensor1D[] = [];\n  const xs1d = xs;\n  for (let i = 0; i < xs.length; ++i) {\n    ys.push(ENGINE.tidy(() => {\n      let x = xs1d[i];\n      if (i > 0) {\n        for (let j = 0; j < i; ++j) {\n          const proj = mul(sum(mul(ys[j], x)), ys[j]);\n          x = sub(x, proj);\n        }\n      }\n      return div(x, norm(x, 'euclidean'));\n    }));\n  }\n\n  if (inputIsTensor2D) {\n    return stack(ys, 0) as Tensor2D;\n  } else {\n    return ys;\n  }\n}\n\nexport const gramSchmidt = op({gramSchmidt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../../engine';\nimport {dispose} from '../../globals';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {clone} from '../clone';\nimport {concat} from '../concat';\nimport {div} from '../div';\nimport {eye} from '../eye';\nimport {greater} from '../greater';\nimport {matMul} from '../mat_mul';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {tensor2d} from '../tensor2d';\nimport {transpose} from '../transpose';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\n\n/**\n * Compute QR decomposition of m-by-n matrix using Householder transformation.\n *\n * Implementation based on\n *   [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf]\n * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf)\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [3, 4]]);\n * let [q, r] = tf.linalg.qr(a);\n * console.log('Q');\n * q.print();\n * console.log('R');\n * r.print();\n * console.log('Orthogonalized');\n * q.dot(q.transpose()).print()  // should be nearly the identity matrix.\n * console.log('Reconstructed');\n * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]];\n * ```\n *\n * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose\n *   it has the shape `[..., M, N]`.\n * @param fullMatrices An optional boolean parameter. Defaults to `false`.\n *   If `true`, compute full-sized `Q`. If `false` (the default),\n *   compute only the leading N columns of `Q` and `R`.\n * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix,\n *   i.e., its columns all have unit norm and are mutually orthogonal.\n *   If `M >= N`,\n *     If `fullMatrices` is `false` (default),\n *       - `Q` has a shape of `[..., M, N]`,\n *       - `R` has a shape of `[..., N, N]`.\n *     If `fullMatrices` is `true` (default),\n *       - `Q` has a shape of `[..., M, M]`,\n *       - `R` has a shape of `[..., M, N]`.\n *   If `M < N`,\n *     - `Q` has a shape of `[..., M, M]`,\n *     - `R` has a shape of `[..., M, N]`.\n * @throws If the rank of `x` is less than 2.\n *\n * @doc {heading:'Operations',\n *       subheading:'Linear Algebra',\n *       namespace:'linalg'}\n */\nfunction qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] {\n  assert(\n      x.rank >= 2,\n      () => `qr() requires input tensor to have a rank >= 2, but got rank ${\n          x.rank}`);\n\n  if (x.rank === 2) {\n    return qr2d(x as Tensor2D, fullMatrices);\n  } else {\n    // Rank > 2.\n    // TODO(cais): Below we split the input into individual 2D tensors,\n    //   perform QR decomposition on them and then stack the results back\n    //   together. We should explore whether this can be parallelized.\n    const outerDimsProd = x.shape.slice(0, x.shape.length - 2)\n                              .reduce((value, prev) => value * prev);\n    const x2ds = unstack(\n        reshape(\n            x,\n            [\n              outerDimsProd, x.shape[x.shape.length - 2],\n              x.shape[x.shape.length - 1]\n            ]),\n        0);\n    const q2ds: Tensor2D[] = [];\n    const r2ds: Tensor2D[] = [];\n    x2ds.forEach(x2d => {\n      const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices);\n      q2ds.push(q2d);\n      r2ds.push(r2d);\n    });\n    const q = reshape(stack(q2ds, 0), x.shape);\n    const r = reshape(stack(r2ds, 0), x.shape);\n    return [q, r];\n  }\n}\n\nfunction qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] {\n  return ENGINE.tidy(() => {\n    assert(\n        x.shape.length === 2,\n        () => `qr2d() requires a 2D Tensor, but got a ${\n            x.shape.length}D Tensor.`);\n\n    const m = x.shape[0];\n    const n = x.shape[1];\n\n    let q = eye(m);    // Orthogonal transform so far.\n    let r = clone(x);  // Transformed matrix so far.\n\n    const one2D = tensor2d([[1]], [1, 1]);\n    let w: Tensor2D = clone(one2D);\n\n    const iters = m >= n ? n : m;\n    for (let j = 0; j < iters; ++j) {\n      // This tidy within the for-loop ensures we clean up temporary\n      // tensors as soon as they are no longer needed.\n      const rTemp = r;\n      const wTemp = w;\n      const qTemp = q;\n      [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => {\n        // Find H = I - tau * w * w', to put zeros below R(j, j).\n        const rjEnd1 = slice(r, [j, j], [m - j, 1]);\n        const normX = norm(rjEnd1);\n        const rjj = slice(r, [j, j], [1, 1]);\n\n        // The sign() function returns 0 on 0, which causes division by zero.\n        const s = where(greater(rjj, 0), tensor2d([[-1]]), tensor2d([[1]]));\n\n        const u1 = sub(rjj, mul(s, normX));\n        const wPre = div(rjEnd1, u1);\n        if (wPre.shape[0] === 1) {\n          w = clone(one2D);\n        } else {\n          w = concat(\n              [\n                one2D,\n                slice(wPre, [1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as\n                    Tensor2D\n              ],\n              0);\n        }\n        const tau = neg(div(matMul(s, u1), normX)) as Tensor2D;\n\n        // -- R := HR, Q := QH.\n        const rjEndAll = slice(r, [j, 0], [m - j, n]);\n        const tauTimesW: Tensor2D = mul(tau, w);\n        const wT: Tensor2D = transpose(w);\n        if (j === 0) {\n          r = sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n        } else {\n          const rTimesTau: Tensor2D =\n              sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n          r = concat([slice(r, [0, 0], [j, n]), rTimesTau], 0);\n        }\n        const tawTimesWT: Tensor2D = transpose(tauTimesW);\n        const qAllJEnd = slice(q, [0, j], [m, q.shape[1] - j]);\n        if (j === 0) {\n          q = sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n        } else {\n          const qTimesTau: Tensor2D =\n              sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n          q = concat([slice(q, [0, 0], [m, j]), qTimesTau], 1);\n        }\n        return [w, r, q];\n      });\n      dispose([rTemp, wTemp, qTemp]);\n    }\n\n    if (!fullMatrices && m > n) {\n      q = slice(q, [0, 0], [m, n]);\n      r = slice(r, [0, 0], [n, n]);\n    }\n\n    return [q, r];\n  }) as [Tensor2D, Tensor2D];\n}\n\nexport const qr = op({qr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport enum Reduction {\n  NONE,\n  MEAN,\n  SUM,\n  SUM_BY_NONZERO_WEIGHTS\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {Reduction} from '../loss_ops_utils';\nimport {mean} from '../mean';\nimport {mul} from '../mul';\nimport {notEqual} from '../not_equal';\nimport {ones} from '../ones';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sum} from '../sum';\n\n/**\n * Computes the weighted loss between two tensors.\n *\n * @param losses Tensor of shape `[batch_size, d1, ... dN]`.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `losses`, and must be broadcastable to `losses` (i.e., all\n *    dimensions must be either `1`, or the same as the corresponding\n *    `losses` dimension).\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction computeWeightedLoss_<T extends Tensor, O extends Tensor>(\n    losses: T|TensorLike, weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $losses = convertToTensor(losses, 'losses', 'computeWeightedLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'computeWeightedLoss');\n  }\n\n  const weightedLoss = ($weights == null) ? $losses : mul($losses, $weights);\n\n  if (reduction === Reduction.NONE) {\n    return weightedLoss as O;\n  }\n  if (reduction === Reduction.SUM) {\n    return sum(weightedLoss);\n  }\n  if (reduction === Reduction.MEAN) {\n    if ($weights == null) {\n      return mean(weightedLoss);\n    } else {\n      const broadcastFactor = $losses.size / $weights.size;\n      const result = div(sum(weightedLoss), sum($weights));\n      return broadcastFactor > 1 ? div(result, scalar(broadcastFactor)) :\n                                   result as O;\n    }\n  }\n  if (reduction === Reduction.SUM_BY_NONZERO_WEIGHTS) {\n    if ($weights == null) {\n      return div(sum(weightedLoss), scalar($losses.size));\n    } else {\n      const broadcastedWeights = mul($weights, ones($losses.shape));\n\n      const numNonZeros =\n          cast(sum(notEqual(broadcastedWeights, scalar(0))), 'float32');\n      return div(sum(weightedLoss), numNonZeros);\n    }\n  }\n\n  throw Error(`Unknown reduction: ${reduction}`);\n}\nexport const computeWeightedLoss = op({computeWeightedLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the absolute difference loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction absoluteDifference_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'absoluteDifference');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'absoluteDifference');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'absoluteDifference');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in absoluteDifference: ');\n\n  const losses = abs(sub($labels, $predictions));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const absoluteDifference = op({absoluteDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the cosine distance loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param axis The dimension along which the cosine distance is computed.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction cosineDistance_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike, axis: number,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'cosineDistance');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'cosineDistance');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'cosineDistance');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in cosineDistance: ');\n\n  const one = scalar(1);\n  const losses = sub(one, sum(mul($labels, $predictions), axis, true));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const cosineDistance = op({cosineDistance_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the Hinge loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction hingeLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $labels = convertToTensor(labels, 'labels', 'hingeLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'hingeLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'hingeLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in hingeLoss: ');\n\n  const one = scalar(1);\n  // Convert binary labels to (-1, 1)\n  $labels = sub(mul(scalar(2), $labels), one);\n  const losses = relu(sub(one, mul($labels, $predictions)));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const hingeLoss = op({hingeLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {Reduction} from '../loss_ops_utils';\nimport {minimum} from '../minimum';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {square} from '../square';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the huber loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param delta Point where huber loss changes from quadratic to linear.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`.\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction huberLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, delta = 1.0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'huberLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'huberLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'huberLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in huberLoss: ');\n\n  const deltaScalar = scalar(delta);\n  const error = abs(sub($predictions, $labels));\n  const quadratic = minimum(error, deltaScalar);\n  const linear = sub(error, quadratic);\n\n  const losses =\n      add(mul(scalar(0.5), square(quadratic)), mul(deltaScalar, linear));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const huberLoss = op({huberLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {log} from '../log';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the log loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param epsilon A small increment to avoid taking log of zero\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction logLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, epsilon = 1e-7,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'logLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'logLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'logLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in logLoss: ');\n\n  const one = scalar(1);\n  const epsilonScalar = scalar(epsilon);\n\n  const l1 = neg(mul($labels, log(add($predictions, epsilonScalar))));\n  const l2 =\n      mul(sub(one, $labels), log(add(sub(one, $predictions), epsilonScalar)));\n  const losses = sub(l1, l2);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const logLoss = op({logLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {squaredDifference} from '../squared_difference';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the mean squared error between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction meanSquaredError_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'meanSquaredError');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'meanSquaredError');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'meanSquaredError');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in meanSquaredError: ');\n\n  const losses = squaredDifference($labels, $predictions);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const meanSquaredError = op({meanSquaredError_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {exp} from '../exp';\nimport {log1p} from '../log1p';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\nfunction sigmoidCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, logits: T|TensorLike): O {\n  const $labels =\n      convertToTensor(labels, 'labels', 'sigmoidCrossEntropyWithLogits');\n  const $logits =\n      convertToTensor(logits, 'logits', 'sigmoidCrossEntropyWithLogits');\n  assertShapesMatch(\n      $labels.shape, $logits.shape, 'Error in sigmoidCrossEntropyWithLogits: ');\n\n  /**\n   * Implementation Details:\n   *\n   * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n   *     z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n   *   = (1 - z) * x + log(1 + exp(-x))\n   *   = x - x * z + log(1 + exp(-x))\n   *\n   *   For x < 0, to avoid overflow in exp(-x), we reformulate the above\n   *     x - x * z + log(1 + exp(-x))\n   *   = log(exp(x)) - x * z + log(1 + exp(-x))\n   *   = - x * z + log(1 + exp(x))\n   *\n   * Hence, to ensure stability and avoid overflow, the implementation uses\n   * this equivalent formulation:\n   *     max(x, 0) - x * z + log(1 + exp(-abs(x)))\n   */\n  const maxOutput = relu($logits);\n  const outputXTarget = mul($logits, $labels);\n  const sigmoidOutput = log1p(exp(neg(abs($logits))));\n\n  return add(sub(maxOutput, outputXTarget), sigmoidOutput);\n}\n\n/**\n * Computes the sigmoid cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newMulticlassLabels = multiclassLabels * (1 - labelSmoothing)\n *                         + 0.5 * labelSmoothing\n *\n * @param multiClassLabels The ground truth output tensor of shape\n * [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction sigmoidCrossEntropy_<T extends Tensor, O extends Tensor>(\n    multiClassLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $multiClassLabels = convertToTensor(\n      multiClassLabels, 'multiClassLabels', 'sigmoidCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'sigmoidCrossEntropy');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'sigmoidCrossEntropy');\n  }\n  assertShapesMatch(\n      $multiClassLabels.shape, $logits.shape, 'Error in sigmoidCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const half = scalar(0.5);\n\n    $multiClassLabels =\n        add(mul($multiClassLabels, sub(one, labelSmoothingScalar)),\n            mul(half, labelSmoothingScalar));\n  }\n  const losses = sigmoidCrossEntropyWithLogits_($multiClassLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const sigmoidCrossEntropy = op({sigmoidCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../../gradients';\nimport {Tensor} from '../../tensor';\nimport {GradSaveFunc} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {expandShapeToKeepDim} from '../axis_util';\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {exp} from '../exp';\nimport {logSumExp} from '../log_sum_exp';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes softmax cross entropy between logits and labels.\n *\n * Measures the probability error in discrete classification tasks in which\n * the classes are mutually exclusive (each entry is in exactly one class).\n * For example, each CIFAR-10 image is labeled with one and only one label: an\n * image can be a dog or a truck, but not both.\n *\n * `NOTE`: While the classes are mutually exclusive, their probabilities need\n * not be. All that is required is that each row of labels is a valid\n * probability distribution. If they are not, the computation of the gradient\n * will be incorrect.\n *\n * `WARNING`: This op expects unscaled logits, since it performs a softmax on\n * logits internally for efficiency. Do not call this op with the output of\n * softmax, as it will produce incorrect results.\n *\n * logits and labels must have the same shape, e.g. [batch_size, num_classes]\n * and the same dtype.\n * @param labels The labels array.\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n */\nfunction softmaxCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T, logits: T, dim = -1): O {\n  if (dim === -1) {\n    dim = logits.rank - 1;\n  }\n\n  if (dim !== logits.rank - 1) {\n    throw Error(\n        `Softmax cross entropy along a non-last dimension is not yet ` +\n        `supported. Labels / logits was rank ${logits.rank} ` +\n        `and dim was ${dim}`);\n  }\n  // Use a custom gradient for numerical stability.\n  const customOp =\n      customGrad((labels: Tensor, logits: Tensor, save: GradSaveFunc) => {\n        // Reference:\n        //   1. http://cs231n.github.io/linear-classify/#softmax\n        //   2. https://blog.feedly.com/tricks-of-the-trade-logsumexp/\n        const keepDims = true;\n        const lse = logSumExp(logits, [dim], keepDims);\n        const logResult = sub(cast(logits, 'float32'), lse);\n        save([labels, logResult]);\n\n        const costVector = neg(mul(logResult, labels));\n        const value: O = sum(costVector, [dim]);\n\n        const gradFunc = (dy: O, saved: Tensor[]) => {\n          const [labels, logResult] = saved;\n          const dyShape = expandShapeToKeepDim(dy.shape, [dim]);\n          return [\n            mul(reshape(dy, dyShape),\n                sub(cast(labels, 'float32'), exp(logResult))),\n            mul(reshape(dy, dyShape),\n                sub(exp(logResult), cast(labels, 'float32'))),\n          ];\n        };\n        return {value, gradFunc};\n      });\n\n  return customOp(labels, logits);\n}\n\n/**\n * Computes the softmax cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newOnehotLabels = onehotLabels * (1 - labelSmoothing)\n *                         + labelSmoothing / numClasses\n *\n * @param onehotLabels One hot encoded labels\n *    [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or 1, and must be\n *    broadcastable to `loss`  of shape [batch_size]\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction softmaxCrossEntropy_<T extends Tensor, O extends Tensor>(\n    onehotLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $onehotLabels =\n      convertToTensor(onehotLabels, 'onehotLabels', 'softmaxCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'softmaxCrossEntropy');\n  let $weights: Tensor = null;\n\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'softmaxCrossEntropy');\n  }\n\n  assertShapesMatch(\n      $onehotLabels.shape, $logits.shape, 'Error in softmaxCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const numClasses = scalar($onehotLabels.shape[1]);\n\n    $onehotLabels =\n        add(mul($onehotLabels, sub(one, labelSmoothingScalar)),\n            div(labelSmoothingScalar, numClasses));\n  }\n\n  const losses = softmaxCrossEntropyWithLogits_($onehotLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const softmaxCrossEntropy = op({softmaxCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Modularized ops.\nexport {abs} from './abs';\nexport {acos} from './acos';\nexport {acosh} from './acosh';\nexport {add} from './add';\nexport {addN} from './add_n';\nexport {all} from './all';\nexport {any} from './any';\nexport {argMax} from './arg_max';\nexport {argMin} from './arg_min';\nexport {asin} from './asin';\nexport {asinh} from './asinh';\nexport {atan} from './atan';\nexport {atan2} from './atan2';\nexport {atanh} from './atanh';\nexport {avgPool} from './avg_pool';\nexport {avgPool3d} from './avg_pool_3d';\nexport {basicLSTMCell} from './basic_lstm_cell';\nexport {batchToSpaceND} from './batch_to_space_nd';\nexport {batchNorm} from './batchnorm';\nexport {batchNorm2d} from './batchnorm2d';\nexport {batchNorm3d} from './batchnorm3d';\nexport {batchNorm4d} from './batchnorm4d';\nexport {broadcastTo} from './broadcast_to';\nexport {buffer} from './buffer';\nexport {cast} from './cast';\nexport {ceil} from './ceil';\nexport {clipByValue} from './clip_by_value';\nexport {clone} from './clone';\nexport {complex} from './complex';\nexport {concat} from './concat';\nexport {concat1d} from './concat_1d';\nexport {concat2d} from './concat_2d';\nexport {concat3d} from './concat_3d';\nexport {concat4d} from './concat_4d';\nexport {conv1d} from './conv1d';\nexport {conv2d} from './conv2d';\nexport {conv2dTranspose} from './conv2d_transpose';\nexport {conv3d} from './conv3d';\nexport {conv3dTranspose} from './conv3d_transpose';\nexport {cos} from './cos';\nexport {cosh} from './cosh';\nexport {cumsum} from './cumsum';\nexport {depthToSpace} from './depth_to_space';\nexport {depthwiseConv2d} from './depthwise_conv2d';\nexport {diag} from './diag';\nexport {dilation2d} from './dilation2d';\nexport {div} from './div';\nexport {divNoNan} from './div_no_nan';\nexport {dot} from './dot';\nexport {elu} from './elu';\nexport {equal} from './equal';\nexport {erf} from './erf';\nexport {exp} from './exp';\nexport {expandDims} from './expand_dims';\nexport {expm1} from './expm1';\nexport {eye} from './eye';\nexport {fill} from './fill';\nexport {floor} from './floor';\nexport {floorDiv} from './floorDiv';\nexport {gather} from './gather';\nexport {greater} from './greater';\nexport {greaterEqual} from './greater_equal';\nexport {imag} from './imag';\nexport {isFinite} from './is_finite';\nexport {isInf} from './is_inf';\nexport {isNaN} from './is_nan';\nexport {leakyRelu} from './leaky_relu';\nexport {less} from './less';\nexport {lessEqual} from './less_equal';\nexport {linspace} from './linspace';\nexport {localResponseNormalization} from './local_response_normalization';\nexport {log} from './log';\nexport {log1p} from './log1p';\nexport {logSigmoid} from './log_sigmoid';\nexport {logSoftmax} from './log_softmax';\nexport {logSumExp} from './log_sum_exp';\nexport {logicalAnd} from './logical_and';\nexport {logicalNot} from './logical_not';\nexport {logicalOr} from './logical_or';\nexport {logicalXor} from './logical_xor';\nexport {matMul} from './mat_mul';\nexport {max} from './max';\nexport {maxPool} from './max_pool';\nexport {maxPool3d} from './max_pool_3d';\nexport {maxPoolWithArgmax} from './max_pool_with_argmax';\nexport {maximum} from './maximum';\nexport {mean} from './mean';\nexport {min} from './min';\nexport {minimum} from './minimum';\nexport {mod} from './mod';\nexport {moments} from './moments';\nexport {mul} from './mul';\nexport {LSTMCellFunc, multiRNNCell} from './multi_rnn_cell';\nexport {multinomial} from './multinomial';\nexport {neg} from './neg';\nexport {notEqual} from './not_equal';\nexport {oneHot} from './one_hot';\nexport {ones} from './ones';\nexport {onesLike} from './ones_like';\nexport {outerProduct} from './outer_product';\nexport {pad} from './pad';\nexport {pad1d} from './pad1d';\nexport {pad2d} from './pad2d';\nexport {pad3d} from './pad3d';\nexport {pad4d} from './pad4d';\nexport {pool} from './pool';\nexport {pow} from './pow';\nexport {prelu} from './prelu';\nexport {print} from './print';\nexport {prod} from './prod';\nexport {rand} from './rand';\nexport {randomGamma} from './random_gamma';\nexport {randomNormal} from './random_normal';\nexport {randomUniform} from './random_uniform';\nexport {range} from './range';\nexport {real} from './real';\nexport {reciprocal} from './reciprocal';\nexport {relu} from './relu';\nexport {relu6} from './relu6';\nexport {reshape} from './reshape';\nexport {reverse} from './reverse';\nexport {reverse1d} from './reverse_1d';\nexport {reverse2d} from './reverse_2d';\nexport {reverse3d} from './reverse_3d';\nexport {reverse4d} from './reverse_4d';\nexport {round} from './round';\nexport {rsqrt} from './rsqrt';\nexport {scalar} from './scalar';\nexport {selu} from './selu';\nexport {separableConv2d} from './separable_conv2d';\nexport {setdiff1dAsync} from './setdiff1d_async';\nexport {sigmoid} from './sigmoid';\nexport {sign} from './sign';\nexport {sin} from './sin';\nexport {sinh} from './sinh';\nexport {slice} from './slice';\nexport {slice1d} from './slice1d';\nexport {slice2d} from './slice2d';\nexport {slice3d} from './slice3d';\nexport {slice4d} from './slice4d';\nexport {softmax} from './softmax';\nexport {softplus} from './softplus';\nexport {spaceToBatchND} from './space_to_batch_nd';\nexport {fft} from './spectral/fft';\nexport {ifft} from './spectral/ifft';\nexport {irfft} from './spectral/irfft';\nexport {rfft} from './spectral/rfft';\nexport {split} from './split';\nexport {sqrt} from './sqrt';\nexport {square} from './square';\nexport {squaredDifference} from './squared_difference';\nexport {squeeze} from './squeeze';\nexport {stack} from './stack';\nexport {step} from './step';\nexport {stridedSlice} from './strided_slice';\nexport {sub} from './sub';\nexport {sum} from './sum';\nexport {tan} from './tan';\nexport {tanh} from './tanh';\nexport {tensor} from './tensor';\nexport {tensor1d} from './tensor1d';\nexport {tensor2d} from './tensor2d';\nexport {tensor3d} from './tensor3d';\nexport {tensor4d} from './tensor4d';\nexport {tensor5d} from './tensor5d';\nexport {tensor6d} from './tensor6d';\nexport {tile} from './tile';\nexport {topk} from './topk';\nexport {truncatedNormal} from './truncated_normal';\nexport {unique} from './unique';\nexport {unsortedSegmentSum} from './unsorted_segment_sum';\nexport {unstack} from './unstack';\nexport {variable} from './variable';\nexport {where} from './where';\nexport {whereAsync} from './where_async';\nexport {zeros} from './zeros';\nexport {zerosLike} from './zeros_like';\n\nexport * from './boolean_mask';\nexport * from './compare';\nexport * from './binary_ops';\nexport * from './transpose';\nexport * from './norm';\nexport * from './moving_average';\nexport * from './scatter_nd';\nexport * from './sparse_to_dense';\nexport * from './gather_nd';\nexport * from './dropout';\nexport * from './signal_ops_util';\nexport * from './in_top_k';\n\nexport {op, OP_SCOPE_SUFFIX} from './operation';\n\nimport {rfft} from './spectral/rfft';\nimport {fft} from './spectral/fft';\nimport {ifft} from './spectral/ifft';\nimport {irfft} from './spectral/irfft';\nconst spectral = {\n  fft,\n  ifft,\n  rfft,\n  irfft\n};\n\nimport * as fused from './fused_ops';\n\nimport {hammingWindow} from './signal/hamming_window';\nimport {hannWindow} from './signal/hann_window';\nimport {frame} from './signal/frame';\nimport {stft} from './signal/stft';\nconst signal = {\n  hammingWindow,\n  hannWindow,\n  frame,\n  stft,\n};\n\n// Image Ops namespace\nimport {cropAndResize} from './image/crop_and_resize';\nimport {flipLeftRight} from './image/flip_left_right';\nimport {rotateWithOffset} from './image/rotate_with_offset';\nimport {nonMaxSuppression} from './image/non_max_suppression';\nimport {nonMaxSuppressionAsync} from './image/non_max_suppression_async';\nimport {nonMaxSuppressionWithScore} from './image/non_max_suppression_with_score';\nimport {nonMaxSuppressionWithScoreAsync} from './image/non_max_suppression_with_score_async';\nimport {nonMaxSuppressionPadded} from './image/non_max_suppression_padded';\nimport {nonMaxSuppressionPaddedAsync} from './image/non_max_suppression_padded_async';\nimport {resizeBilinear} from './image/resize_bilinear';\nimport {resizeNearestNeighbor} from './image/resize_nearest_neighbor';\nconst image = {\n  flipLeftRight,\n  resizeNearestNeighbor,\n  resizeBilinear,\n  rotateWithOffset,\n  cropAndResize,\n  nonMaxSuppression,\n  nonMaxSuppressionAsync,\n  nonMaxSuppressionWithScore,\n  nonMaxSuppressionWithScoreAsync,\n  nonMaxSuppressionPadded,\n  nonMaxSuppressionPaddedAsync\n};\n\n// linalg namespace\nimport {bandPart} from './linalg/band_part';\nimport {gramSchmidt} from './linalg/gram_schmidt';\nimport {qr} from './linalg/qr';\nconst linalg = {\n  bandPart,\n  gramSchmidt,\n  qr\n};\n\n// losses namespace;\nimport {absoluteDifference} from './losses/absolute_difference';\nimport {computeWeightedLoss} from './losses/compute_weighted_loss';\nimport {cosineDistance} from './losses/cosine_distance';\nimport {hingeLoss} from './losses/hinge_loss';\nimport {huberLoss} from './losses/huber_loss';\nimport {logLoss} from './losses/log_loss';\nimport {meanSquaredError} from './losses/mean_squared_error';\nimport {sigmoidCrossEntropy} from './losses/sigmoid_cross_entropy';\nimport {softmaxCrossEntropy} from './losses/softmax_cross_entropy';\nconst losses = {\n  absoluteDifference,\n  computeWeightedLoss,\n  cosineDistance,\n  hingeLoss,\n  huberLoss,\n  logLoss,\n  meanSquaredError,\n  sigmoidCrossEntropy,\n  softmaxCrossEntropy\n};\n\n// Second level exports.\nexport {image, linalg, losses, spectral, fused, signal};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {dispose} from '../globals';\nimport {variableGrads} from '../gradients';\nimport {scalar} from '../ops/ops';\nimport {Serializable} from '../serialization';\nimport {Scalar, Variable} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\n/**\n * A variable that belongs to an optimizer.\n *\n * The `originalName` field is required for keeping track of the canonical\n * name of the variable, which is usually the name of the model weight that\n * the variable is related to plus a suffix, e.g., 'dense1/kernel/momentum'.\n * The name of the `Variable` object itself cannot be used directly due to\n * possible deduplication: Every `Variable` must have a unique name but more\n * than one optimizer objects of the same type may be created for the same model\n * or the same `Variable`.\n */\nexport interface OptimizerVariable {\n  originalName: string;\n  variable: Variable;\n}\n\n/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */\nexport abstract class Optimizer extends Serializable {\n  protected iterations_: number;\n\n  /**\n   * Executes `f()` and minimizes the scalar output of `f()` by computing\n   * gradients of y with respect to the list of trainable variables provided by\n   * `varList`. If no list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to minimize.\n   * @param returnCost Whether to return the scalar cost value produced by\n   * executing `f()`.\n   * @param varList An optional list of variables to update. If specified, only\n   * the trainable variables in varList will be updated by minimize. Defaults to\n   * all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  minimize(f: () => Scalar, returnCost = false, varList?: Variable[]): Scalar\n      |null {\n    const {value, grads} = this.computeGradients(f, varList);\n\n    if (varList != null) {\n      const gradArray: NamedTensor[] =\n          varList.map(v => ({name: v.name, tensor: grads[v.name]}));\n      this.applyGradients(gradArray);\n    } else {\n      this.applyGradients(grads);\n    }\n\n    // Dispose gradients.\n    dispose(grads);\n\n    if (returnCost) {\n      return value;\n    } else {\n      value.dispose();\n      return null;\n    }\n  }\n\n  /**\n   * The number of iterations that this optimizer instance has been invoked for.\n   */\n  get iterations(): number {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return this.iterations_;\n  }\n\n  protected incrementIterations() {\n    this.iterations_ = this.iterations + 1;\n  }\n\n  /**\n   * Executes f() and computes the gradient of the scalar output of f() with\n   * respect to the list of trainable variables provided by `varList`. If no\n   * list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to use for computing\n   * gradients with respect to variables.\n   * @param varList An optional list of variables to compute gradients with\n   * respect to. If specified, only the trainable variables in varList will have\n   * gradients computed with respect to. Defaults to all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  computeGradients(f: () => Scalar, varList?: Variable[]):\n      {value: Scalar, grads: NamedTensorMap} {\n    return variableGrads(f, varList);\n  }\n\n  /**\n   * Updates variables by using the computed gradients.\n   *\n   * @param variableGradients A mapping of variable name to its gradient value.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  abstract applyGradients(variableGradients: NamedTensorMap|\n                          NamedTensor[]): void;\n\n  /**\n   * Dispose the variables (if any) owned by this optimizer instance.\n   */\n  dispose(): void {\n    if (this.iterations_ != null) {\n      dispose(this.iterations_);\n    }\n  }\n\n  async saveIterations(): Promise<NamedTensor> {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return {\n      name: 'iter',  // Named for Python compatibility.\n      // TODO(cais): Use 'int64' type when available.\n      tensor: scalar(this.iterations_, 'int32')\n    };\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for this optimizer yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error(\n        `setWeights() is not implemented for this optimizer class ` +\n        `${this.getClassName()}`);\n  }\n\n  /**\n   * Extract the first element of the weight values and set it\n   * as the iterations counter variable of this instance of optimizer.\n   *\n   * @param weightValues\n   * @returns Weight values with the first element consumed and excluded.\n   */\n  protected async extractIterations(weightValues: NamedTensor[]):\n      Promise<NamedTensor[]> {\n    this.iterations_ = (await weightValues[0].tensor.data())[0];\n    return weightValues.slice(1);\n  }\n}\n\nObject.defineProperty(Optimizer, Symbol.hasInstance, {\n  value: (instance: Optimizer) => {\n    return instance.minimize != null && instance.computeGradients != null &&\n        instance.applyGradients != null;\n  }\n});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/ops';\nimport {square} from '../ops/square';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdadeltaOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adadelta';  // Name matters for Python compatibility.\n  private accumulatedGrads: OptimizerVariable[] = [];\n  private accumulatedUpdates: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected rho: number,\n      protected epsilon: number = null) {\n    super();\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedGrads[i] == null) {\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accum_grad`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedUpdates[i] == null) {\n        this.accumulatedUpdates[i] = {\n          originalName: `${name}/accum_var`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n      const accumulatedUpdate = this.accumulatedUpdates[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad =\n            add(mul(accumulatedGrad, this.rho),\n                mul(square(gradient), 1 - this.rho));\n\n        const updates =\n            mul(div(sqrt(add(accumulatedUpdate, this.epsilon)),\n                    sqrt(add(accumulatedGrad, this.epsilon))),\n                gradient);\n\n        const newAccumulatedUpdate =\n            add(mul(accumulatedUpdate, this.rho),\n                mul(square(updates), 1 - this.rho));\n\n        accumulatedGrad.assign(newAccumulatedGrad);\n        accumulatedUpdate.assign(newAccumulatedUpdate);\n\n        const newValue = add(mul(updates, -this.learningRate), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedUpdates != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n      dispose(this.accumulatedUpdates.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedGrads, ...this.accumulatedUpdates];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedGrads =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedUpdates =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'rho': this.rho,\n      'epsilon': this.epsilon\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['rho'], config['epsilon']);\n  }\n}\nregisterClass(AdadeltaOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {fill} from '../ops/fill';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdagradOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adagrad';  // Note: Name matters for Python compatibility.\n\n  private accumulatedGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private initialAccumulatorValue = 0.1) {\n    super();\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulatedGrads[i] == null) {\n        const trainable = false;\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accumulator`,\n          variable: tidy(\n              () => fill(value.shape, this.initialAccumulatorValue)\n                        .variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad = add(accumulatedGrad, square(gradient));\n        accumulatedGrad.assign(newAccumulatedGrad);\n\n        const newValue = add(\n            mul(div(gradient,\n                    sqrt(add(newAccumulatedGrad, ENGINE.backend.epsilon()))),\n                -this.learningRate),\n            value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedGrads != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulatedGrads.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulatedGrads = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'initialAccumulatorValue': this.initialAccumulatorValue,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['initialAccumulatorValue']);\n  }\n}\nregisterClass(AdagradOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adam';  // Note: Name matters for Python compatibility.\n  private accBeta1: Variable;\n  private accBeta2: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedSecondMoment: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null) {\n    super();\n    tidy(() => {\n      // accB* will be updated by batch.\n      this.accBeta1 = scalar(beta1).variable();\n      this.accBeta2 = scalar(beta2).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const oneMinusAccBeta2 = sub(1, this.accBeta2);\n\n      varNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n        if (this.accumulatedSecondMoment[i] == null) {\n          this.accumulatedSecondMoment[i] = {\n            originalName: `${name}/v`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const secondMoment = this.accumulatedSecondMoment[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n        const newSecondMoment =\n            add(mul(secondMoment, this.beta2),\n                mul(square(gradient), 1 - this.beta2));\n\n        const biasCorrectedFirstMoment = div(newFirstMoment, oneMinusAccBeta1);\n        const biasCorrectedSecondMoment =\n            div(newSecondMoment, oneMinusAccBeta2);\n\n        firstMoment.assign(newFirstMoment);\n        secondMoment.assign(newSecondMoment);\n\n        const newValue =\n            add(mul(div(biasCorrectedFirstMoment,\n                        add(sqrt(biasCorrectedSecondMoment), this.epsilon)),\n                    -this.learningRate),\n                value);\n        value.assign(newValue);\n      });\n\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n      this.accBeta2.assign(mul(this.accBeta2, this.beta2));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.accBeta2.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedSecondMoment != null) {\n      dispose(this.accumulatedSecondMoment.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedFirstMoment, ...this.accumulatedSecondMoment];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    tidy(() => {\n      this.accBeta1.assign(pow(this.beta1, this.iterations_ + 1));\n      this.accBeta2.assign(pow(this.beta2, this.iterations_ + 1));\n    });\n\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedFirstMoment =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedSecondMoment =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon']);\n  }\n}\nregisterClass(AdamOptimizer);\n", "\uFEFF/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {abs} from '../ops/abs';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {maximum} from '../ops/maximum';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamaxOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adamax';  // Note: Name matters for Python compatbility.\n  private accBeta1: Variable;\n  private iteration: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null,\n      protected decay = 0.0) {\n    super();\n\n    tidy(() => {\n      this.iteration = scalar(0).variable();\n      this.accBeta1 = scalar(beta1).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const lr =\n          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));\n\n      variableNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n        if (this.accumulatedWeightedInfNorm[i] == null) {\n          this.accumulatedWeightedInfNorm[i] = {\n            originalName: `${name}/v`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n\n        const ut0 = mul(weightedInfNorm, this.beta2);\n        const ut1 = abs(gradient);\n\n        const newWeightedInfNorm = maximum(ut0, ut1);\n\n        firstMoment.assign(newFirstMoment);\n        weightedInfNorm.assign(newWeightedInfNorm);\n\n        const newValue =\n            add(mul(div(lr, oneMinusAccBeta1),\n                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),\n                value);\n\n        value.assign(newValue);\n      });\n\n      this.iteration.assign(add(this.iteration, 1));\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.iteration.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedWeightedInfNorm != null) {\n      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for Adamax yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error('setWeights() is not implemented for Adamax yet.');\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n      'decay': this.decay\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon'], config['decay']);\n  }\n}\nregisterClass(AdamaxOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {keep, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer} from './optimizer';\n\n/** @doclink Optimizer */\nexport class SGDOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'SGD';  // Note: Name matters for Python compatibility.\n  protected c: Scalar;\n\n  constructor(protected learningRate: number) {\n    super();\n    this.setLearningRate(learningRate);\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    varNames.forEach((name, i) => {\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n      const value = ENGINE.registeredVariables[name];\n      tidy(() => {\n        const newValue = add(mul(this.c, gradient), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  /**\n   * Sets the learning rate of the optimizer.\n   */\n  setLearningRate(learningRate: number) {\n    this.learningRate = learningRate;\n    if (this.c != null) {\n      this.c.dispose();\n    }\n    this.c = keep(scalar(-learningRate));\n  }\n\n  dispose() {\n    this.c.dispose();\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    return [await this.saveIterations()];\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    if (weightValues.length !== 0) {\n      throw new Error('SGD optimizer does not have settable weights.');\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {'learningRate': this.learningRate};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate']);\n  }\n}\nregisterClass(SGDOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {OptimizerVariable} from './optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\n/** @doclink Optimizer */\nexport class MomentumOptimizer extends SGDOptimizer {\n  /** @nocollapse */\n  static className = 'Momentum';  // Name matters for Python compatibility.\n  private m: Scalar;\n  private accumulations: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private momentum: number,\n      private useNesterov = false) {\n    super(learningRate);\n    this.m = scalar(this.momentum);\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulations[i] == null) {\n        const trainable = false;\n        this.accumulations[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const accumulation = this.accumulations[i].variable;\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      tidy(() => {\n        let newValue: Tensor;\n        const newAccumulation = add(mul(this.m, accumulation), gradient);\n        if (this.useNesterov) {\n          newValue = add(\n              mul(this.c, add(gradient, mul(newAccumulation, this.m))), value);\n        } else {\n          newValue = add(mul(this.c, newAccumulation), value);\n        }\n        accumulation.assign(newAccumulation);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.m.dispose();\n    if (this.accumulations != null) {\n      dispose(this.accumulations.map(v => v.variable));\n    }\n  }\n\n  /**\n   * Sets the momentum of the optimizer.\n   *\n   * @param momentum\n   */\n  setMomentum(momentum: number) {\n    this.momentum = momentum;\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulations.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulations = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'momentum': this.momentum,\n      'useNesterov': this.useNesterov\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['momentum'], config['useNesterov']);\n  }\n}\nregisterClass(MomentumOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class RMSPropOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'RMSProp';  // Note: Name matters for Python compatibility.\n  private centered: boolean;\n\n  private accumulatedMeanSquares: OptimizerVariable[] = [];\n  private accumulatedMoments: OptimizerVariable[] = [];\n  private accumulatedMeanGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected decay = 0.9,\n      protected momentum = 0.0, protected epsilon: number = null,\n      centered = false) {\n    super();\n\n    this.centered = centered;\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n    if (learningRate == null) {\n      throw new Error(`learningRate for RMSPropOptimizer must be defined.`);\n    }\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedMeanSquares[i] == null) {\n        this.accumulatedMeanSquares[i] = {\n          originalName: `${name}/rms`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMoments[i] == null) {\n        this.accumulatedMoments[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMeanGrads[i] == null && this.centered) {\n        this.accumulatedMeanGrads[i] = {\n          originalName: `${name}/mg`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedMeanSquare = this.accumulatedMeanSquares[i].variable;\n      const accumulatedMoments = this.accumulatedMoments[i].variable;\n      tidy(() => {\n        const newAccumulatedMeanSquare =\n            add(mul(accumulatedMeanSquare, this.decay),\n                mul(square(gradient), 1 - this.decay));\n\n        if (this.centered) {\n          const accumulatedMeanGrad = this.accumulatedMeanGrads[i].variable;\n          // Centered gradient\n          const newAccumulatedMeanGrad =\n              add(mul(accumulatedMeanGrad, this.decay),\n                  mul(gradient, 1 - this.decay));\n\n          const gradContribution =\n              div(mul(gradient, this.learningRate),\n                  sqrt(\n                      sub(newAccumulatedMeanSquare,\n                          add(square(newAccumulatedMeanGrad), this.epsilon))));\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum), gradContribution);\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMeanGrad.assign(newAccumulatedMeanGrad);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        } else {\n          // Plain gradient\n          const newAccumulatedMeanSquare =\n              add(mul(accumulatedMeanSquare, this.decay),\n                  mul(square(gradient), 1 - this.decay));\n\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum),\n                  div(mul(gradient, this.learningRate),\n                      sqrt(add(newAccumulatedMeanSquare, this.epsilon))));\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        }\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedMeanSquares != null) {\n      dispose(this.accumulatedMeanSquares.map(v => v.variable));\n    }\n    if (this.accumulatedMeanGrads != null && this.centered) {\n      dispose(this.accumulatedMeanGrads.map(v => v.variable));\n    }\n    if (this.accumulatedMoments != null) {\n      dispose(this.accumulatedMoments.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedMeanSquares, ...this.accumulatedMoments];\n    if (this.centered) {\n      variables.push(...this.accumulatedMeanGrads);\n    }\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount =\n        this.centered ? weightValues.length / 3 : weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedMeanSquares =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedMoments =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n    if (this.centered) {\n      this.accumulatedMeanGrads =\n          weightValues.slice(variableCount * 2, variableCount * 3)\n              .map(v => ({\n                     originalName: v.name,\n                     variable: v.tensor.variable(trainable)\n                   }));\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'decay': this.decay,\n      'momentum': this.momentum,\n      'epsilon': this.epsilon,\n      'centered': this.centered\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['decay'], config['momentum'],\n        config['epsilon'], config['centered']);\n  }\n}\nregisterClass(RMSPropOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AdadeltaOptimizer} from './adadelta_optimizer';\nimport {AdagradOptimizer} from './adagrad_optimizer';\nimport {AdamOptimizer} from './adam_optimizer';\nimport {AdamaxOptimizer} from './adamax_optimizer';\nimport {MomentumOptimizer} from './momentum_optimizer';\nimport {RMSPropOptimizer} from './rmsprop_optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\nexport class OptimizerConstructors {\n  /**\n   * Constructs a `tf.SGDOptimizer` that uses stochastic gradient descent.\n   *\n   * ```js\n   * // Fit a quadratic function by learning the coefficients a, b, c.\n   * const xs = tf.tensor1d([0, 1, 2, 3]);\n   * const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);\n   *\n   * const a = tf.scalar(Math.random()).variable();\n   * const b = tf.scalar(Math.random()).variable();\n   * const c = tf.scalar(Math.random()).variable();\n   *\n   * // y = a * x^2 + b * x + c.\n   * const f = x => a.mul(x.square()).add(b.mul(x)).add(c);\n   * const loss = (pred, label) => pred.sub(label).square().mean();\n   *\n   * const learningRate = 0.01;\n   * const optimizer = tf.train.sgd(learningRate);\n   *\n   * // Train the model.\n   * for (let i = 0; i < 10; i++) {\n   *   optimizer.minimize(() => loss(f(xs), ys));\n   * }\n   *\n   * // Make predictions.\n   * console.log(\n   *     `a: ${a.dataSync()}, b: ${b.dataSync()}, c: ${c.dataSync()}`);\n   * const preds = f(xs).dataSync();\n   * preds.forEach((pred, i) => {\n   *   console.log(`x: ${i}, pred: ${pred}`);\n   * });\n   * ```\n   *\n   * @param learningRate The learning rate to use for the SGD algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static sgd(learningRate: number): SGDOptimizer {\n    return new SGDOptimizer(learningRate);\n  }\n\n  /**\n   * Constructs a `tf.MomentumOptimizer` that uses momentum gradient\n   * descent.\n   *\n   * See\n   * [http://proceedings.mlr.press/v28/sutskever13.pdf](\n   * http://proceedings.mlr.press/v28/sutskever13.pdf)\n   *\n   * @param learningRate The learning rate to use for the Momentum gradient\n   * descent algorithm.\n   * @param momentum The momentum to use for the momentum gradient descent\n   * algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static momentum(learningRate: number, momentum: number, useNesterov = false):\n      MomentumOptimizer {\n    return new MomentumOptimizer(learningRate, momentum, useNesterov);\n  }\n\n  /**\n   * Constructs a `tf.RMSPropOptimizer` that uses RMSProp gradient\n   * descent. This implementation uses plain momentum and is not centered\n   * version of RMSProp.\n   *\n   * See\n   * [http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf](\n   * http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)\n   *\n   * @param learningRate The learning rate to use for the RMSProp gradient\n   * descent algorithm.\n   * @param decay The discounting factor for the history/coming gradient.\n   * @param momentum The momentum to use for the RMSProp gradient descent\n   * algorithm.\n   * @param epsilon Small value to avoid zero denominator.\n   * @param centered If true, gradients are normalized by the estimated\n   * variance of the gradient.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static rmsprop(\n      learningRate: number, decay = .9, momentum = 0.0, epsilon: number = null,\n      centered = false): RMSPropOptimizer {\n    return new RMSPropOptimizer(\n        learningRate, decay, momentum, epsilon, centered);\n  }\n\n  /**\n   * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adam gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adam(\n      learningRate = 0.001, beta1 = 0.9, beta2 = 0.999,\n      epsilon: number = null): AdamOptimizer {\n    return new AdamOptimizer(learningRate, beta1, beta2, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.\n   * See [https://arxiv.org/abs/1212.5701](https://arxiv.org/abs/1212.5701)\n   *\n   * @param learningRate The learning rate to use for the Adadelta gradient\n   * descent algorithm.\n   * @param rho The learning rate decay over each update.\n   * @param epsilon A constant epsilon used to better condition the grad\n   * update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adadelta(learningRate = .001, rho = .95, epsilon: number = null):\n      AdadeltaOptimizer {\n    return new AdadeltaOptimizer(learningRate, rho, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdamaxOptimizer` that uses the Adamax algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adamax gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   * @param decay The learning rate decay over each update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adamax(\n      learningRate = 0.002, beta1 = 0.9, beta2 = 0.999, epsilon: number = null,\n      decay = 0.0): AdamaxOptimizer {\n    return new AdamaxOptimizer(learningRate, beta1, beta2, epsilon, decay);\n  }\n\n  /**\n   * Constructs a `tf.AdagradOptimizer` that uses the Adagrad algorithm.\n   * See\n   * [http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf](\n   * http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\n   * or\n   * [http://ruder.io/optimizing-gradient-descent/index.html#adagrad](\n   * http://ruder.io/optimizing-gradient-descent/index.html#adagrad)\n   *\n   * @param learningRate The learning rate to use for the Adagrad gradient\n   * descent algorithm.\n   * @param initialAccumulatorValue Starting value for the accumulators, must be\n   * positive.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adagrad(learningRate: number, initialAccumulatorValue = 0.1):\n      AdagradOptimizer {\n    return new AdagradOptimizer(learningRate, initialAccumulatorValue);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// So typings can propagate.\nimport {AdadeltaOptimizer} from './optimizers/adadelta_optimizer';\nimport {AdagradOptimizer} from './optimizers/adagrad_optimizer';\nimport {AdamOptimizer} from './optimizers/adam_optimizer';\nimport {AdamaxOptimizer} from './optimizers/adamax_optimizer';\nimport {MomentumOptimizer} from './optimizers/momentum_optimizer';\nimport {OptimizerConstructors} from './optimizers/optimizer_constructors';\nimport {RMSPropOptimizer} from './optimizers/rmsprop_optimizer';\nimport {SGDOptimizer} from './optimizers/sgd_optimizer';\n\n// tslint:disable-next-line:no-unused-expression\n[MomentumOptimizer, SGDOptimizer, AdadeltaOptimizer, AdagradOptimizer,\n RMSPropOptimizer, AdamaxOptimizer, AdamOptimizer];\n\nexport const train = {\n  sgd: OptimizerConstructors.sgd,\n  momentum: OptimizerConstructors.momentum,\n  adadelta: OptimizerConstructors.adadelta,\n  adagrad: OptimizerConstructors.adagrad,\n  rmsprop: OptimizerConstructors.rmsprop,\n  adamax: OptimizerConstructors.adamax,\n  adam: OptimizerConstructors.adam\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst delayCallback: Function = (() => {\n  if (typeof requestAnimationFrame !== 'undefined') {\n    return requestAnimationFrame;\n  } else if (typeof setImmediate !== 'undefined') {\n    return setImmediate;\n  }\n  return (f: Function) => f();  // no delays\n})();\n\n/**\n * Returns a promise that resolve when a requestAnimationFrame has completed.\n *\n * On Node.js this uses setImmediate instead of requestAnimationFrame.\n *\n * This is simply a sugar method so that users can do the following:\n * `await tf.nextFrame();`\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nfunction nextFrame(): Promise<void> {\n  return new Promise<void>(resolve => delayCallback(() => resolve()));\n}\n\nexport {nextFrame};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Returns the image center in pixels.\nexport function getImageCenter(\n    center: number|[number, number], imageHeight: number,\n    imageWidth: number): [number, number] {\n  const centerX =\n      imageWidth * (typeof center === 'number' ? center : center[0]);\n  const centerY =\n      imageHeight * (typeof center === 'number' ? center : center[1]);\n  return [centerX, centerY];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Gets the new shape of the input Tensor after it's been reshaped\n * to:\n * [blockShape[0], ..., blockShape[M-1], batch / prod(blockShape),\n * inputShape[1], ..., inputShape[N-1]]\n *\n * See step 1: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshaped(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  let reshaped: number[] = [];\n  if (batchToSpace) {\n    reshaped = reshaped.concat(blockShape.slice(0));\n    reshaped.push(inputShape[0] / prod);\n    reshaped = reshaped.concat(inputShape.slice(1));\n  } else {\n    reshaped = reshaped.concat(inputShape[0]);\n    const spatialLength = blockShape.length;\n    for (let i = 0; i < spatialLength; ++i) {\n      reshaped =\n          reshaped.concat([inputShape[i + 1] / blockShape[i], blockShape[i]]);\n    }\n    reshaped = reshaped.concat(inputShape.slice(spatialLength + 1));\n  }\n  return reshaped;\n}\n\n/**\n * Gets the permutation that will transpose the dimensions of the\n * reshaped tensor to shape:\n *\n * [batch / prod(block_shape),inputShape[1], blockShape[0], ...,\n * inputShape[M], blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * see step 2: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getPermuted(\n    reshapedRank: number, blockShapeRank: number,\n    batchToSpace = true): number[] {\n  const permuted = [];\n  if (batchToSpace) {\n    permuted.push(blockShapeRank);\n    for (let i = blockShapeRank + 1; i < reshapedRank; ++i) {\n      if (i <= 2 * blockShapeRank) {\n        permuted.push(i);\n        permuted.push(i - (blockShapeRank + 1));\n      } else {\n        permuted.push(i);\n      }\n    }\n  } else {\n    const permutedBeforeBatch = [];\n    const permutedAfterBatch = [];\n    for (let i = 1; i < reshapedRank; ++i) {\n      if (i >= blockShapeRank * 2 + 1 || i % 2 === 1) {\n        permutedAfterBatch.push(i);\n      } else {\n        permutedBeforeBatch.push(i);\n      }\n    }\n    permuted.push(...permutedBeforeBatch);\n    permuted.push(0);\n    permuted.push(...permutedAfterBatch);\n  }\n  return permuted;\n}\n\n/**\n * Gets the shape of the reshaped and permuted input Tensor before any cropping\n * is applied.  The new shape will be:\n *\n * [batch / prod(blockShape),inputShape[1] * blockShape[0], ...,\n * inputShape[M] * blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 3: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshapedPermuted(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  const reshapedPermuted = [];\n\n  if (batchToSpace) {\n    reshapedPermuted.push(inputShape[0] / prod);\n  } else {\n    reshapedPermuted.push(inputShape[0] * prod);\n  }\n\n  for (let i = 1; i < inputShape.length; ++i) {\n    if (i <= blockShape.length) {\n      if (batchToSpace) {\n        reshapedPermuted.push(blockShape[i - 1] * inputShape[i]);\n      } else {\n        reshapedPermuted.push(inputShape[i] / blockShape[i - 1]);\n      }\n    } else {\n      reshapedPermuted.push(inputShape[i]);\n    }\n  }\n\n  return reshapedPermuted;\n}\n\n/**\n * Converts the crops argument into the beginning coordinates of a slice\n * operation.\n */\nexport function getSliceBeginCoords(\n    crops: number[][], blockShape: number): number[] {\n  const sliceBeginCoords = [0];\n  for (let i = 0; i < blockShape; ++i) {\n    sliceBeginCoords.push(crops[i][0]);\n  }\n  return sliceBeginCoords;\n}\n\n/**\n * Converts the crops argument into the size of a slice operation.  When\n * combined with getSliceBeginCoords this function allows the reshaped and\n * permuted Tensor to be cropped to its final output shape of:\n *\n * inputShape[1] * blockShape[0] - crops[0,0] - crops[0,1], ...,\n * inputShape[M] * blockShape[M-1] -crops[M-1,0] -\n * crops[M-1,1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 4: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getSliceSize(\n    uncroppedShape: number[], crops: number[][], blockShape: number): number[] {\n  const sliceSize = uncroppedShape.slice(0, 1);\n  for (let i = 0; i < blockShape; ++i) {\n    sliceSize.push(uncroppedShape[i + 1] - crops[i][0] - crops[i][1]);\n  }\n\n  return sliceSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const SELU_SCALEALPHA = 1.7580993408473768599402175208123;\nexport const SELU_SCALE = 1.0507009873554804934193349852946;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const ERF_P = 0.3275911;\nexport const ERF_A1 = 0.254829592;\nexport const ERF_A2 = -0.284496736;\nexport const ERF_A3 = 1.421413741;\nexport const ERF_A4 = -1.453152027;\nexport const ERF_A5 = 1.061405429;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\n\nexport function warn(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.warn(...msg);\n  }\n}\n\nexport function log(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.log(...msg);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TypedArray} from '../types';\n/**\n * Merges real and imaginary Float32Arrays into a single complex Float32Array.\n *\n * The memory layout is interleaved as follows:\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n * complex: [r0, i0, r1, i1, r2, i2]\n *\n * This is the inverse of splitRealAndImagArrays.\n *\n * @param real The real values of the complex tensor values.\n * @param imag The imag values of the complex tensor values.\n * @returns A complex tensor as a Float32Array with merged values.\n */\nexport function mergeRealAndImagArrays(\n    real: Float32Array, imag: Float32Array): Float32Array {\n  if (real.length !== imag.length) {\n    throw new Error(\n        `Cannot merge real and imag arrays of different lengths. real:` +\n        `${real.length}, imag: ${imag.length}.`);\n  }\n  const result = new Float32Array(real.length * 2);\n  for (let i = 0; i < result.length; i += 2) {\n    result[i] = real[i / 2];\n    result[i + 1] = imag[i / 2];\n  }\n  return result;\n}\n\n/**\n * Splits a complex Float32Array into real and imag parts.\n *\n * The memory layout is interleaved as follows:\n * complex: [r0, i0, r1, i1, r2, i2]\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n *\n * This is the inverse of mergeRealAndImagArrays.\n *\n * @param complex The complex tensor values.\n * @returns An object with real and imag Float32Array components of the complex\n *     tensor.\n */\nexport function splitRealAndImagArrays(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(complex.length / 2);\n  const imag = new Float32Array(complex.length / 2);\n  for (let i = 0; i < complex.length; i += 2) {\n    real[i / 2] = complex[i];\n    imag[i / 2] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts even indexed complex values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithEvenIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.ceil(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 0; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts odd indexed comple values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithOddIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.floor(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 2; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Get the map representing a complex value in the given array.\n * @param complex The complex tensor values.\n * @param index An index of the target complex value.\n */\nexport function getComplexWithIndex(\n    complex: Float32Array, index: number): {real: number, imag: number} {\n  const real = complex[index * 2];\n  const imag = complex[index * 2 + 1];\n  return {real, imag};\n}\n\n/**\n * Insert a given complex value into the TypedArray.\n * @param data The array in which the complex value is inserted.\n * @param c The complex value to be inserted.\n * @param index An index of the target complex value.\n */\nexport function assignToTypedArray(\n    data: TypedArray, real: number, imag: number, index: number) {\n  data[index * 2] = real;\n  data[index * 2 + 1] = imag;\n}\n\n/**\n * Make the list of exponent terms used by FFT.\n */\nexport function exponents(\n    n: number, inverse: boolean): {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(n / 2);\n  const imag = new Float32Array(n / 2);\n  for (let i = 0; i < Math.ceil(n / 2); i++) {\n    const x = (inverse ? 2 : -2) * Math.PI * (i / n);\n    real[i] = Math.cos(x);\n    imag[i] = Math.sin(x);\n  }\n  return {real, imag};\n}\n\n/**\n * Make the exponent term used by FFT.\n */\nexport function exponent(\n    k: number, n: number, inverse: boolean): {real: number, imag: number} {\n  const x = (inverse ? 2 : -2) * Math.PI * (k / n);\n  const real = Math.cos(x);\n  const imag = Math.sin(x);\n  return {real, imag};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {cast} from '../ops/cast';\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\nimport {Rank} from '../types';\nimport {DataType, ShapeMap} from '../types';\nimport {hasEncodingLoss, makeZerosTypedArray} from '../util';\n\nimport {KernelBackend} from './backend';\n\n// Utilities needed by backend consumers of tf-core.\nexport * from '../ops/axis_util';\nexport * from '../ops/broadcast_util';\nexport * from '../ops/concat_util';\nexport * from '../ops/conv_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/fused_types';\nexport * from '../ops/reduce_util';\n\nimport * as slice_util from '../ops/slice_util';\nexport {slice_util};\n\nexport {BackendValues, TypedArray, upcastType, PixelData} from '../types';\nexport {MemoryInfo, TimingInfo} from '../engine';\nexport * from '../ops/rotate_util';\nexport * from '../ops/array_ops_util';\nexport * from '../ops/gather_nd_util';\nexport * from '../ops/scatter_nd_util';\nexport * from '../ops/selu_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/erf_util';\nexport * from '../log';\nexport * from '../backends/complex_util';\nexport * from '../ops/split_util';\n\nimport * as segment_util from '../ops/segment_util';\nexport {segment_util};\n\nexport function castTensor<T extends Tensor>(\n    x: T, dtype: DataType, backend: KernelBackend): T {\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return x.clone();\n    }\n    const zerosTensor = zeros(x.shape);\n    const floatX = cast(x, 'float32');\n    const result = backend.complex(floatX, zerosTensor);\n    zerosTensor.dispose();\n    floatX.dispose();\n    return result as T;\n  }\n\n  if (!hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    return ENGINE.makeTensorFromDataId(x.dataId, x.shape, dtype) as T;\n  }\n  if (x.dtype === 'complex64') {\n    const real = backend.real(x);\n    const result = cast(real, dtype);\n    real.dispose();\n    return result;\n  }\n  if (dtype === 'int32') {\n    return backend.int(x);\n  } else if (dtype === 'bool') {\n    const zero = scalar(0, x.dtype);\n    const result = backend.notEqual(x, zero) as T;\n    zero.dispose();\n    return result;\n  } else {\n    throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n  }\n}\n\nexport function reshapeTensor<T extends Tensor, R extends Rank>(\n    x: T, shape: ShapeMap[R]): Tensor<R> {\n  return ENGINE.makeTensorFromDataId(x.dataId, shape, x.dtype) as Tensor<R>;\n}\n\nexport function linspaceImpl(start: number, stop: number, num: number) {\n  const step = (stop - start) / (num - 1);\n\n  const values = makeZerosTypedArray(num, 'float32');\n  values[0] = start;\n  for (let i = 1; i < values.length; i++) {\n    values[i] = values[i - 1] + step;\n  }\n\n  return tensor1d(values, 'float32');\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\n// TODO(annxingyuan): Use this helper in WASM Split kernel once intermediate\n// kernels have been modularized in WebGL and CPU\n// https://github.com/tensorflow/tfjs/issues/2822.\n/** Shared implementation of the split kernel across WebGL and CPU. */\nexport function split<T extends Tensor>(\n    x: T, sizeSplits: number[], axis: number): T[] {\n  const begin = new Array(x.rank).fill(0);\n  const size = x.shape.slice();\n  return sizeSplits.map(s => {\n    const sliceSize = [...size];\n    sliceSize[axis] = s;\n    const sliceT = slice(x, begin, sliceSize);\n    begin[axis] += s;\n    return sliceT;\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * An implementation of the tile kernel shared between webgl and cpu for string\n * tensors only.\n */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {DataType, Rank} from '../types';\n\nexport function tile<R extends Rank>(\n    xBuf: TensorBuffer<R, DataType>, reps: number[]): Tensor<R> {\n  const newShape: number[] = new Array(xBuf.rank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = xBuf.shape[i] * reps[i];\n  }\n  const result = buffer(newShape, xBuf.dtype);\n  for (let i = 0; i < result.values.length; ++i) {\n    const newLoc = result.indexToLoc(i);\n\n    const originalLoc: number[] = new Array(xBuf.rank);\n    for (let j = 0; j < originalLoc.length; j++) {\n      originalLoc[j] = newLoc[j] % xBuf.shape[j];\n    }\n\n    const originalIndex = xBuf.locToIndex(originalLoc);\n\n    result.values[i] = xBuf.values[originalIndex];\n  }\n  return result.toTensor() as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the TopK kernel shared between webgl and cpu. */\n\nimport {tensor} from '../ops/tensor';\nimport {Tensor} from '../tensor';\nimport {NumericDataType, TypedArray} from '../types';\nimport {getTypedArrayFromDType} from '../util';\n\nexport function topkImpl<T extends Tensor>(\n    x: TypedArray, xShape: number[], xDtype: NumericDataType, k: number,\n    sorted: boolean): [T, T] {\n  // Reshape into a 2d tensor [batch, lastDim] and compute topk along lastDim.\n  const lastDim = xShape[xShape.length - 1];\n  const [batch, size] = [x.length / lastDim, lastDim];\n  const allTopKVals = getTypedArrayFromDType(xDtype, batch * k);\n  const allTopKIndices = getTypedArrayFromDType('int32', batch * k);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = x.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    const outOffset = b * k;\n    const topKVals = allTopKVals.subarray(outOffset, outOffset + k);\n    const topKIndices = allTopKIndices.subarray(outOffset, outOffset + k);\n    for (let i = 0; i < k; i++) {\n      topKVals[i] = valAndInd[i].value;\n      topKIndices[i] = valAndInd[i].index;\n    }\n  }\n  // Reshape back to the original input shape, except that the last\n  // dimension is k.\n  const outputShape = xShape.slice();\n  outputShape[outputShape.length - 1] = k;\n  return [\n    tensor(allTopKVals, outputShape, xDtype) as T,\n    tensor(allTopKIndices, outputShape, 'int32') as T\n  ];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport {nonMaxSuppressionV3Impl, nonMaxSuppressionV4Impl, nonMaxSuppressionV5Impl} from './non_max_suppression_impl';\nexport {split} from './split_shared';\nexport {tile} from './tile_impl';\nexport {topkImpl} from './topk_impl';\nexport {whereImpl} from './where_impl';\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const absGradConfig: GradConfig = {\n  kernelName: Abs,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, step(cast(x, 'float32'), -1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acosGradConfig: GradConfig = {\n  kernelName: Acos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = square(cast(x, 'float32'));\n        const b = sqrt(sub(scalar(1), a));\n        return neg(div(dy, b));\n      }\n\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acoshGradConfig: GradConfig = {\n  kernelName: Acosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(sub(square(cast(x, 'float32')), 1));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Add} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const addGradConfig: GradConfig = {\n  kernelName: Add,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AddN} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const addNGradConfig: GradConfig = {\n  kernelName: AddN,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const ders: {[key: string]: () => Tensor} = {};\n    saved.forEach((_, i) => {\n      ders[i] = () => dy.clone();\n    });\n    return ders;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMax} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMaxGradConfig: GradConfig = {\n  kernelName: ArgMax,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMinGradConfig: GradConfig = {\n  kernelName: ArgMin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const asinGradConfig: GradConfig = {\n  kernelName: Asin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, sqrt(sub(scalar(1), square(cast(x, 'float32')))))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const asinhGradConfig: GradConfig = {\n  kernelName: Asinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(add(scalar(1), square(cast(x, 'float32'))));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const atan2GradConfig: GradConfig = {\n  kernelName: Atan2,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const d = add(square(a), square(b));\n      let res = mul(dy, div(b, d));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      const d = add(square(a), square(b));\n      let res = neg(mul(dy, div(a, d)));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const atanGradConfig: GradConfig = {\n  kernelName: Atan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, add(square(cast(x, 'float32')), 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {scalar} from '../ops/scalar';\nimport {Tensor} from '../tensor';\n\nexport const atanhGradConfig: GradConfig = {\n  kernelName: Atanh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, sub(scalar(1), square(cast(x, 'float32'))))};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool3DBackprop, AvgPool3DBackpropAttrs, AvgPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d avg pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0. The dilation\n *     rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction avgPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in avgPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in avgPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.avgPool3dBackprop(dy5D, input5D, convInfo);\n  };\n\n  const inputs: AvgPool3DBackpropInputs = {dy: dy5D, input: input5D};\n\n  const attrs: AvgPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      AvgPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3dBackprop = op({avgPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool3D, AvgPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPool3dBackprop} from '../ops/avg_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const avgPool3DGradConfig: GradConfig = {\n  kernelName: AvgPool3D,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as AvgPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => avgPool3dBackprop(\n          dy as Tensor5D, x, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of an 2D avg pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The input image, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction avgPoolBackprop_<T extends Tensor3D|Tensor4D>(\n    dy: T|TensorLike, input: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy (${\n          $dy.rank})`);\n\n  let input4D = $input as Tensor4D;\n  let dy4D = $dy as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($input.rank === 3) {\n    reshapedTo4D = true;\n    input4D =\n        reshape($input, [1, $input.shape[0], $input.shape[1], $input.shape[2]]);\n    dy4D = reshape($dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2]]);\n  }\n\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in avgPoolBackprop: dy must be rank 4 but got rank ` +\n          `${dy4D.rank}.`);\n  util.assert(\n      input4D.rank === 4,\n      () => `Error in avgPoolBackprop: input must be rank 4 but got rank ` +\n          `${input4D.rank}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        input4D.shape, filterSize, strides, 1 /* dilations */, pad);\n\n    return backend.avgPoolBackprop(dy4D, input4D, convInfo);\n  };\n\n  const inputs: AvgPoolBackpropInputs = {dy: dy4D, input: input4D};\n\n  const attrs: AvgPoolBackpropAttrs = {filterSize, strides, pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, AvgPoolBackprop,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const avgPoolBackprop = op({avgPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool, AvgPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPoolBackprop} from '../ops/avg_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const avgPoolGradConfig: GradConfig = {\n  kernelName: AvgPool,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as AvgPoolAttrs;\n    return {\n      x: () => avgPoolBackprop(dy as Tensor4D, x, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchMatMul, BatchMatMulAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {matMul} from '../ops/mat_mul';\nimport {Tensor, Tensor3D} from '../tensor';\n\nexport const batchMatMulGradConfig: GradConfig = {\n  kernelName: BatchMatMul,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor3D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [a, b] = saved as Tensor3D[];\n\n    const {transposeA, transposeB} = attrs as {} as BatchMatMulAttrs;\n\n    if (!transposeA && !transposeB) {\n      return {\n        a: () => matMul(dy, b, false, true),\n        b: () => matMul(a, dy, true, false)\n      };\n    } else if (!transposeA && transposeB) {\n      return {\n        a: () => matMul(dy, b, false, false),\n        b: () => matMul(dy, a, true, false)\n      };\n    } else if (transposeA && !transposeB) {\n      return {\n        a: () => matMul(b, dy, false, true),\n        b: () => matMul(a, dy, false, false)\n      };\n    } else {\n      return {\n        a: () => matMul(b, dy, true, true),\n        b: () => matMul(dy, a, true, true)\n      };\n    }\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchToSpaceND, BatchToSpaceNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {spaceToBatchND} from '../ops/space_to_batch_nd';\nimport {Tensor} from '../tensor';\n\nexport const batchToSpaceNDGradConfig: GradConfig = {\n  kernelName: BatchToSpaceND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, crops} = attrs as {} as BatchToSpaceNDAttrs;\n    return {x: () => spaceToBatchND(dy, blockShape, crops)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BroadcastTo, BroadCastToAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const broadcastToGradConfig: GradConfig = {\n  kernelName: BroadcastTo,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const broadCastToAttrs: BroadCastToAttrs =\n        attrs as unknown as BroadCastToAttrs;\n\n    const inputShape = broadCastToAttrs.inputShape;\n    const outputShape = broadCastToAttrs.shape;\n\n    const reps: number[] = Array.from(outputShape);\n    for (let i = inputShape.length - 1; i >= 0; i--) {\n      if (inputShape[i] === outputShape[i]) {\n        reps[i] = 1;\n      } else if (inputShape[i] !== 1) {\n        throw new Error(`broadcastTo(): [${\n            inputShape}] cannot be broadcast to [${outputShape}].`);\n      }\n    }\n    const axes: number[] = [];\n    for (let i = 0; i < reps.length; i++) {\n      if (reps[i] > 1) {\n        axes.push(i);\n      }\n    }\n\n    return {x: () => sum(dy, axes, true /* keepDims */)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cast} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const castGradConfig: GradConfig = {\n  kernelName: Cast,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => dy.clone()};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const ceilGradConfig: GradConfig = {\n  kernelName: Ceil,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {lessEqual} from '../ops/less_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const clipByValueGradConfig: GradConfig = {\n  kernelName: ClipByValue,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {clipValueMin, clipValueMax} = attrs as {} as ClipByValueAttrs;\n    return {\n      x: () => where(\n          logicalAnd(greaterEqual(x, clipValueMin), lessEqual(x, clipValueMax)),\n          dy, zerosLike(dy)),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Concat, ConcatAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {split} from '../ops/split';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const concatGradConfig: GradConfig = {\n  kernelName: Concat,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const shapes = saved.map(t => t.shape);\n    const {axis} = attrs as {} as ConcatAttrs;\n    const $axis = parseAxisParam(axis, saved[0].shape)[0];\n    const sizeSplits = shapes.map(s => s[$axis]);\n    const derTensors = split(dy, sizeSplits, $axis);\n    return derTensors.map(t => () => t) as {};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2D, Conv2DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../ops/conv2d_backprop_input';\nimport * as conv_util from '../ops/conv_util';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv2DGradConfig: GradConfig = {\n  kernelName: Conv2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x4D, $filter] = saved as [Tensor4D, Tensor4D];\n    const {dilations, strides, pad, dataFormat} = attrs as {} as Conv2DAttrs;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of conv2D: dilation rates greater than 1 ' +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    return {\n      x: () =>\n          conv2DBackpropInput(x4D.shape, dy, $filter, strides, pad, dataFormat),\n      filter: () =>\n          conv2DBackpropFilter(x4D, dy, $filter.shape, strides, pad, dataFormat)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2d} from '../ops/conv2d';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const conv2DBackpropInputGradConfig: GradConfig = {\n  kernelName: Conv2DBackpropInput,\n  inputsToSave: ['dy', 'filter'],\n  gradFunc: (ddx: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [dy, filter] = saved as [Tensor4D, Tensor4D];\n\n    const {strides, pad, dataFormat, dimRoundingMode} =\n        attrs as {} as Conv2DBackpropInputAttrs;\n\n    return {\n      dy: () => conv2d(\n          ddx, filter, strides, pad, dataFormat, 1 /* dilations */,\n          dimRoundingMode),\n      filter: () => conv2DBackpropFilter(\n          ddx, dy, filter.shape, strides, pad, dataFormat, dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropFilterAttrs, Conv3DBackpropFilterInputs, Conv3DBackpropFilterV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 3D convolution.\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     [batch, depth, height, width, inChannels]. If rank 4, batch of 1 is\n *     assumed.\n * @param dy The dy image, of rank 5 or rank 4, of shape\n *     [batch, depth, height, width, outDepth]. If rank 4, batch of 1 is\n *     assumed.\n * @param filterShape The shape of the filter, length 5,\n *     [filterDepth, filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideDepth, strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction conv3DBackpropFilter_<T extends Tensor4D|Tensor5D>(\n    x: T, dy: T, filterShape: [number, number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): Tensor5D {\n  let x5D = x as Tensor5D;\n  if (x.rank === 4) {\n    x5D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2], x.shape[3]]);\n  }\n  let dy5D = dy as Tensor5D;\n  if (dy5D.rank === 4) {\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3dDerFilter: input must be rank 5, but got shape ` +\n          `${x5D.shape}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerFilter: dy must be rank 5, but got shape ` +\n          `${dy5D.shape}.`);\n  util.assert(\n      filterShape.length === 5,\n      () => `Error in conv3dDerFilter: filterShape must be length 5, but got ` +\n          `${filterShape}.`);\n  util.assert(\n      x5D.shape[4] === filterShape[3],\n      () => `Error in conv3dDerFilter: depth of input ${x5D.shape[4]}) must ` +\n          `match input depth in filter (${filterShape[3]}.`);\n  util.assert(\n      dy5D.shape[4] === filterShape[4],\n      () => `Error in conv3dDerFilter: depth of dy (${dy5D.shape[4]}) must ` +\n          `match output depth for filter (${filterShape[4]}).`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, filterShape, strides, dilations, pad);\n\n    return backend.conv3dDerFilter(x5D, dy5D, convInfo);\n  };\n\n  const inputs: Conv3DBackpropFilterInputs = {x: x5D, y: dy5D};\n\n  const attrs: Conv3DBackpropFilterAttrs = {strides, pad};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv3DBackpropFilterV2, attrs as {} as NamedAttrMap) as Tensor5D;\n}\n\nexport const conv3DBackpropFilter = op({conv3DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv3D, Conv3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv3DBackpropFilter} from '../ops/conv3d_backprop_filter';\nimport {conv3DBackpropInput} from '../ops/conv3d_backprop_input';\nimport {tupleValuesAreOne} from '../ops/conv_util';\nimport {Tensor, Tensor5D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv3DGradConfig: GradConfig = {\n  kernelName: Conv3D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor5D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad} = attrs as {} as Conv3DAttrs;\n    util.assert(\n        tupleValuesAreOne(dilations),\n        () =>\n            'Error in gradient of conv3D: dilation rates greater than 1 are ' +\n            `not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const [x5D, $filter] = saved;\n\n    return {\n      x: () => conv3DBackpropInput(\n          (x5D as Tensor5D).shape, dy, $filter as Tensor5D, strides, pad),\n      filter: () => conv3DBackpropFilter(\n          x5D as Tensor5D, dy, ($filter as Tensor5D).shape, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {sin} from '../ops/sin';\nimport {Tensor} from '../tensor';\n\nexport const cosGradConfig: GradConfig = {\n  kernelName: Cos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(neg(sin(cast(x, 'float32'))), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {sinh} from '../ops/sinh';\nimport {Tensor} from '../tensor';\n\nexport const coshGradConfig: GradConfig = {\n  kernelName: Cosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(sinh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cumsum, CumsumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getAxesPermutation} from '../ops/axis_util';\nimport {cumsum} from '../ops/cumsum';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const cumsumGradConfig: GradConfig = {\n  kernelName: Cumsum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {axis, exclusive, reverse}: CumsumAttrs = attrs as {} as CumsumAttrs;\n\n    return {\n      x: () => {\n        const permutation = getAxesPermutation([axis], x.rank);\n\n        let out = cumsum(dy, axis, exclusive, !reverse);\n\n        if (permutation != null) {\n          out = transpose(out, permutation);\n        }\n\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as conv_util from '../ops/conv_util';\nimport {depthwiseConv2dNativeBackpropFilter} from '../ops/depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../ops/depthwise_conv2d_native_backprop_input';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const depthwiseConv2dNativeGradConfig: GradConfig = {\n  kernelName: DepthwiseConv2dNative,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad, dimRoundingMode} =\n        attrs as {} as DepthwiseConv2dNativeAttrs;\n\n    const $dilations = dilations == null ? [1, 1] : dilations;\n\n    util.assert(\n        conv_util.tupleValuesAreOne($dilations),\n        () => 'Error in gradient of depthwiseConv2dNative: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${$dilations}'`);\n\n    const [x, filter] = saved as [Tensor4D, Tensor4D];\n\n    util.assert(\n        x.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: input must be ` +\n            `rank 4, but got rank ${x.rank}.`);\n    util.assert(\n        filter.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: filter must be ` +\n            `rank 4, but got rank ${filter.rank}.`);\n    util.assert(\n        x.shape[3] === filter.shape[2],\n        () => `Error in gradient of depthwiseConv2d: number of input ` +\n            `channels (${x.shape[3]}) must match the inChannels dimension ` +\n            `in filter ${filter.shape[2]}.`);\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, $dilations),\n        () => 'Error in gradient of depthwiseConv2d: Either strides or ' +\n            `dilations must be  1. Got strides ${strides} and dilations ` +\n            `'${$dilations}'.`);\n\n    if (dimRoundingMode != null) {\n      util.assert(\n          util.isInt(pad as number),\n          () =>\n              `Error in depthwiseConv2d: pad must be an integer when using, ` +\n              `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n    }\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x.shape, filter.shape, strides, $dilations as number | [number, number],\n        pad, dimRoundingMode, true /* depthwise */);\n\n    return {\n      x: () =>\n          depthwiseConv2dNativeBackpropInput(x.shape, dy, filter, convInfo),\n      filter: () =>\n          depthwiseConv2dNativeBackpropFilter(x, dy, filter.shape, convInfo),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DBackpropFilter, Dilation2DBackpropFilterInputs, Dilation2DBackpropInput, Dilation2DBackpropInputInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const dilation2dGradConfig: GradConfig = {\n  kernelName: Dilation2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, filter] = saved as [Tensor4D, Tensor3D];\n\n    const inputInputs: Dilation2DBackpropInputInputs = {x, filter, dy};\n    const filterInputs: Dilation2DBackpropFilterInputs = {x, filter, dy};\n\n    return {\n      x: () => ENGINE.runKernel(\n                   Dilation2DBackpropInput, inputInputs as {} as NamedTensorMap,\n                   attrs) as Tensor,\n      filter: () => ENGINE.runKernel(\n                        Dilation2DBackpropFilter,\n                        filterInputs as {} as NamedTensorMap, attrs) as Tensor\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const divGradConfig: GradConfig = {\n  kernelName: Div,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluGrad, EluGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const eluGradConfig: GradConfig = {\n  kernelName: Elu,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      return backend.eluDer(dy, y);\n    };\n\n    const inputs: EluGradInputs = {dy, y};\n\n    return {\n      x: () => ENGINE.runKernelFunc(\n          backPropKernelFunc, inputs as {} as NamedTensorMap, null /* grad */,\n          EluGrad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Erf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const erfGradConfig: GradConfig = {\n  kernelName: Erf,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const a = mul(exp(neg(square(x))), 2 / Math.sqrt(Math.PI));\n    return {x: () => mul(dy, a)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expGradConfig: GradConfig = {\n  kernelName: Exp,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n    return {x: () => mul(dy, y)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expm1GradConfig: GradConfig = {\n  kernelName: Expm1,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, exp(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const floorGradConfig: GradConfig = {\n  kernelName: Floor,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FloorDiv} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const floorDivGradConfig: GradConfig = {\n  kernelName: FloorDiv,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {FusedBatchNorm, FusedBatchNormAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {rsqrt} from '../ops/rsqrt';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {tile} from '../ops/tile';\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nexport const fusedBatchNormGradConfig: GradConfig = {\n  kernelName: FusedBatchNorm,\n  inputsToSave: ['x', 'mean', 'variance', 'scale'],\n  gradFunc: <R extends Rank>(\n      dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {varianceEpsilon} = attrs as {} as FusedBatchNormAttrs;\n    const [x, mean, variance, scale] = saved;\n\n    const scaleValue = scale == null ? scalar(1) : scale;\n    const reductionAxes = getReductionAxes(mean.shape, x.shape);\n    const tileShape: number[] = [];\n    if (mean.rank === 1) {\n      for (let i = 0; i < x.shape.length - 1; ++i) {\n        tileShape.push(x.shape[i]);\n      }\n      tileShape.push(1);\n    }\n\n    const xMinusMean = sub(x, mean);\n    const dyTimesScaleValue = mul(dy, scaleValue);\n    const oneOverSqrtVariance = rsqrt(add(variance, scalar(varianceEpsilon)));\n    const minusHalfRCube = mul(\n        mul(mul(oneOverSqrtVariance, oneOverSqrtVariance), oneOverSqrtVariance),\n        scalar(-0.5));\n\n    const derX = () => {\n      if (mean.rank === 1) {\n        return reshape(\n            mul(mul(dy,\n                    tile(\n                        reshape(oneOverSqrtVariance, [1, 1, 1, mean.shape[0]]),\n                        tileShape)),\n                scaleValue),\n            x.shape);\n      } else {\n        return reshape(mul(mul(dy, oneOverSqrtVariance), scaleValue), x.shape);\n      }\n    };\n    const derMean = () => {\n      let meanDer =\n          mul(mul(oneOverSqrtVariance, scalar(-1)), dyTimesScaleValue);\n      if (mean.rank === 1) {\n        meanDer = sum(meanDer, reductionAxes);\n      }\n      return reshape(meanDer, mean.shape as ShapeMap[R]);\n    };\n    const derVariance = () => {\n      let varianceDer = mul(mul(minusHalfRCube, xMinusMean), dyTimesScaleValue);\n\n      if (mean.rank === 1) {\n        varianceDer = sum(varianceDer, reductionAxes);\n      }\n      return reshape(varianceDer, mean.shape as ShapeMap[R]);\n    };\n    const derScale = () => {\n      const xMinusMean2TimesRsqrt = mul(xMinusMean, oneOverSqrtVariance);\n\n      let scaleDer = mul(dy, xMinusMean2TimesRsqrt);\n      if (mean.rank === 1) {\n        scaleDer = sum(scaleDer, reductionAxes);\n      }\n      return reshape(scaleDer, mean.shape as ShapeMap[R]);\n    };\n    const derOffset = () => {\n      let offsetDer = dy;\n      if (mean.rank === 1) {\n        offsetDer = sum(offsetDer, reductionAxes);\n      }\n      return reshape(offsetDer, mean.shape as ShapeMap[R]);\n    };\n\n    return {\n      x: derX,\n      mean: derMean,\n      variance: derVariance,\n      scale: derScale,\n      offset: derOffset\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GatherV2, GatherV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getUndoAxesPermutation} from '../ops/axis_util';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {unsortedSegmentSum} from '../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const gatherGradConfig: GradConfig = {\n  kernelName: GatherV2,\n  inputsToSave: ['x', 'indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, indices] = saved;\n    const {axis} = attrs as {} as GatherV2Attrs;\n\n    const parsedAxis = parseAxisParam(axis, x.shape)[0];\n\n    const derX = () => {\n      const paramsShape = x.shape;\n      const indicesSize = indices.size;\n\n      const outerShape = paramsShape.slice(0, parsedAxis);\n      const outerDims = outerShape.length;\n      const innerShape = paramsShape.slice(axis, paramsShape.length).slice(1);\n      const innerDims = innerShape.length;\n\n      const outerAxesIndices = arrayRange(0, outerDims);\n      const innerAxesIndices =\n          arrayRange(outerDims + 1, outerDims + 1 + innerDims);\n\n      const valuesShape = arrayConcat([outerShape, [indicesSize], innerShape]);\n\n      const values = reshape(dy, valuesShape);\n      const reshapedIndices = reshape(indices, [indicesSize]);\n\n      const transposeDims =\n          arrayConcat([[outerDims], outerAxesIndices, innerAxesIndices]);\n      const valuesTranspose = transpose(values, transposeDims);\n      let paramsGrad = unsortedSegmentSum(\n          valuesTranspose, reshapedIndices as Tensor1D, x.shape[parsedAxis]);\n\n      const invertTransposeDims = getUndoAxesPermutation(transposeDims);\n      paramsGrad = transpose(paramsGrad, invertTransposeDims);\n\n      return paramsGrad;\n    };\n    return {x: derX, indices: () => indices};\n  }\n};\n\nfunction arrayRange(start: number, stop: number): number[] {\n  const result = [];\n  for (let i = start; i < stop; ++i) {\n    result.push(i);\n  }\n  return result;\n}\n\nfunction arrayConcat(arrays: number[][]): number[] {\n  const result = [];\n  for (let i = 0; i < arrays.length; ++i) {\n    for (let j = 0; j < arrays[i].length; ++j) {\n      result.push(arrays[i][j]);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GreaterEqual} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const greaterEqualGradConfig: GradConfig = {\n  kernelName: GreaterEqual,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    return {a: () => zerosLike(a), b: () => zerosLike(b)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {Tensor} from '../tensor';\n\nexport const identityGradConfig: GradConfig = {\n  kernelName: Identity,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => cast(dy, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isFiniteGradConfig: GradConfig = {\n  kernelName: IsFinite,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isInfGradConfig: GradConfig = {\n  kernelName: IsInf,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isNanGradConfig: GradConfig = {\n  kernelName: IsNan,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log1p} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const log1pGradConfig: GradConfig = {\n  kernelName: Log1p,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, add(x, 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const logGradConfig: GradConfig = {\n  kernelName: Log,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, cast(x, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {LogSoftmax, LogSoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const logSoftmaxGradConfig: GradConfig = {\n  kernelName: LogSoftmax,\n  inputsToSave: [],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [value] = saved;\n    const {axis} = attrs as {} as LogSoftmaxAttrs;\n    return {\n      logits: () => {\n        const keepDims = true;\n        const softmax = exp(value);\n        return sub(dy, mul(sum(dy, axis, keepDims), softmax));\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRNBackprop, LRNBackpropAttrs, LRNBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport {op} from './operation';\n\nfunction localResponseNormalizationBackprop_<T extends Tensor4D>(\n    x: T, y: T, dy: T, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.LRNGrad(dy, x, y, depthRadius, bias, alpha, beta);\n\n  const inputs: LRNBackpropInputs = {x, y, dy};\n\n  const attrs: LRNBackpropAttrs = {depthRadius, bias, alpha, beta};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LRNBackprop, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const localResponseNormalizationBackprop =\n    op({localResponseNormalizationBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {LRN, LRNAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {localResponseNormalizationBackprop} from '../ops/local_response_normalization_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const lrnGradConfig: GradConfig = {\n  kernelName: LRN,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {depthRadius, bias, alpha, beta} = attrs as {} as LRNAttrs;\n\n    return {\n      x: () => localResponseNormalizationBackprop(\n          x, y, dy, depthRadius, bias, alpha, beta)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as axis_util from '../ops/axis_util';\nimport {cast} from '../ops/cast';\nimport {equal} from '../ops/equal';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\n/**\n * Gradient helper function for the min and max operations.\n */\nexport function gradForMinAndMax<T extends Tensor>(\n    dy: T, y: T, xOrig: Tensor, origAxes: number[], permutedAxes: number[]) {\n  if (y.rank < xOrig.rank) {\n    y = reshape(y, axis_util.expandShapeToKeepDim(y.shape, origAxes)) as T;\n  }\n  if (dy.rank < xOrig.rank) {\n    dy = reshape(dy, axis_util.expandShapeToKeepDim(dy.shape, origAxes)) as T;\n  }\n  return {\n    x: () => {\n      const dx = mul(dy, cast(equal(xOrig, y), dy.dtype));\n      return permutedAxes == null ? dx : transpose(dx, permutedAxes);\n    }\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const maxGradConfig: GradConfig = {\n  kernelName: Max,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const maxAttrs: MaxAttrs = attrs as {} as MaxAttrs;\n    const {reductionIndices} = maxAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const maxGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = maxGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Maximum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {less} from '../ops/less';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const maximumGradConfig: GradConfig = {\n  kernelName: Maximum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(greaterEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(less(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool3DBackprop, MaxPool3DBackpropAttrs, MaxPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d max pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank 4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param output The original output image, of rank 5 of shape\n *     [batchSize, outDepth, outHeight, outWidth, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike, output: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'maxPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPool3dBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let output5D = $output as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n    output5D = reshape($output, [\n      1, $output.shape[0], $output.shape[1], $output.shape[2], $output.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in maxPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in maxPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n  util.assert(\n      output5D.rank === 5,\n      () => `Error in maxPool3dBackprop: output must be rank 5 but got rank ` +\n          `${output5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.maxPool3dBackprop(dy5D, input5D, output5D, convInfo);\n  };\n\n  const inputs:\n      MaxPool3DBackpropInputs = {dy: dy5D, input: input5D, output: output5D};\n\n  const attrs: MaxPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      MaxPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3dBackprop = op({maxPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool3D, MaxPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPool3dBackprop} from '../ops/max_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const maxPool3DGradConfig: GradConfig = {\n  kernelName: MaxPool3D,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor5D, Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as MaxPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => maxPool3dBackprop(\n          dy as Tensor5D, x, y, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\n\n/**\n * Computes the backprop of a 2D max pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The original input image, of rank 4, of shape\n *     [batchSize, height, width, channels].\n * @param output The original output image, of rank 4, of shape\n *     [batchSize, outHeight, outWidth, channels].\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPoolBackprop_(\n    dy: Tensor4D|TensorLike, input: Tensor4D|TensorLike,\n    output: Tensor4D|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  const $dy = convertToTensor(dy, 'dy', 'maxPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPoolBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy ` +\n          `(${$dy.rank})`);\n\n  util.assert(\n      $dy.rank === 4,\n      () => `Error in maxPoolBackprop: dy must be rank 4 but got rank ` +\n          `${$dy.rank}.`);\n  util.assert(\n      $input.rank === 4,\n      () => `Error in maxPoolBackprop: input must be rank 4 but got rank ` +\n          `${$input.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPoolBackprop: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        $input.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    return backend.maxPoolBackprop($dy, $input, $output, convInfo);\n  };\n\n  const inputs:\n      MaxPoolBackpropInputs = {dy: $dy, input: $input, output: $output};\n\n  const attrs:\n      MaxPoolBackpropAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, MaxPoolBackprop,\n             attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const maxPoolBackprop = op({maxPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool, MaxPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPoolBackprop} from '../ops/max_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const maxPoolGradConfig: GradConfig = {\n  kernelName: MaxPool,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as MaxPoolAttrs;\n\n    return {\n      x: () => maxPoolBackprop(dy as Tensor4D, x, y, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Min, MinAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const minGradConfig: GradConfig = {\n  kernelName: Min,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const minAttrs: MinAttrs = attrs as {} as MinAttrs;\n    const {axis} = minAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(axis, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const minGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = minGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Minimum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const minimumGradConfig: GradConfig = {\n  kernelName: Minimum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(lessEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(greater(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Mod} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {floor} from '../ops/floor';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const modGradConfig: GradConfig = {\n  kernelName: Mod,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(dy, reduceAxes), a.shape);\n      }\n      return dy;\n    };\n    const derB = () => {\n      const res = mul(dy, neg(floor(div(a, b))));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Multiply} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const multiplyGradConfig: GradConfig = {\n  kernelName: Multiply,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = mul(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      const res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Negate} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {neg} from '../ops/neg';\nimport {Tensor} from '../tensor';\n\nexport const negateGradConfig: GradConfig = {\n  kernelName: Negate,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => neg(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OneHot} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\n\nexport const oneHotGradConfig: GradConfig = {\n  kernelName: OneHot,\n  inputsToSave: ['indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const indices = saved[0];\n    return {indices: () => zeros(indices.shape, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OnesLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const onesLikeGradConfig: GradConfig = {\n  kernelName: OnesLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PadV2, PadV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\nexport const padV2GradConfig: GradConfig = {\n  kernelName: PadV2,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    // Pad introduces values around the original tensor, so the gradient\n    // slices the original shape out of the gradient.\n    const x = saved[0];\n    const {paddings} = attrs as unknown as PadV2Attrs;\n    const begin = paddings.map(p => p[0]);\n    return {x: () => slice(dy, begin, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Pow} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {log} from '../ops/log';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {reshape} from '../ops/reshape';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const powGradConfig: GradConfig = {\n  kernelName: Pow,\n  inputsToSave: ['a', 'b'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b, y] = saved;\n    const base = a;\n    const exp = b;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(base.shape, exp.shape);\n\n    const derBase = () => {\n      const expFloat = cast(exp, 'float32');\n      let res = mul(dy, mul(expFloat, pow(base, sub(expFloat, scalar(1)))));\n      const reduceAxes = broadcast_util.getReductionAxes(base.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, base.shape);\n    };\n    const derExp = () => {\n      const condition = greater(base, 0);\n      const logBase = where(condition, log(base), zerosLike(base));\n      let res = mul(dy, mul(y, logBase));\n      const reduceAxes = broadcast_util.getReductionAxes(exp.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, exp.shape);\n    };\n    return {a: derBase, b: derExp};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Prelu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const preluGradConfig: GradConfig = {\n  kernelName: Prelu,\n  inputsToSave: ['x', 'alpha'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x, alpha] = saved;\n    const mask = greater(x, 0);\n\n    return {\n      x: () => where(mask, dy, mul(dy, alpha)),\n      alpha: () => {\n        let res = where(mask, zerosLike(dy), mul(dy, x));\n        const reduceAxes = getReductionAxes(alpha.shape, dy.shape);\n        if (reduceAxes.length > 0) {\n          res = sum(res, reduceAxes);\n        }\n        return reshape(res, alpha.shape);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Reciprocal} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const reciprocalGradConfig: GradConfig = {\n  kernelName: Reciprocal,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, neg(square(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu6} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const relu6GradConfig: GradConfig = {\n  kernelName: Relu6,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const mask = mul(lessEqual(x, 6), step(x));\n\n    return {x: () => mul(dy, cast(mask, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const reluGradConfig: GradConfig = {\n  kernelName: Relu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, cast(step(x), 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reshape} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\n\nexport const reshapeGradConfig: GradConfig = {\n  kernelName: Reshape,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => reshape(dy, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearGrad, ResizeBilinearGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeBilinearGradConfig: GradConfig = {\n  kernelName: ResizeBilinear,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeBilinearAttrs;\n      return backend.resizeBilinearBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeBilinearGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeBilinearGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborGrad, ResizeNearestNeighborGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeNearestNeighborGradConfig: GradConfig = {\n  kernelName: ResizeNearestNeighbor,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeNearestNeighborAttrs;\n      return backend.resizeNearestNeighborBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeNearestNeighborGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeNearestNeighborGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reverse, ReverseAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {reverse} from '../ops/reverse';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const reverseGradConfig: GradConfig = {\n  kernelName: Reverse,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dims} = attrs as {} as ReverseAttrs;\n    const axes = parseAxisParam(dims, dy.shape);\n    return {x: () => reverse(dy, axes)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Round} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const roundGradConfig: GradConfig = {\n  kernelName: Round,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rsqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {pow} from '../ops/pow';\nimport {Tensor} from '../tensor';\n\nexport const rsqrtGradConfig: GradConfig = {\n  kernelName: Rsqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => neg(div(dy, mul(pow(x, 1.5), 2)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SelectV2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {logicalNot} from '../ops/logical_not';\nimport {mul} from '../ops/mul';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const selectV2PoolGradConfig: GradConfig = {\n  kernelName: SelectV2,\n  inputsToSave: ['condition'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [condition] = saved;\n    return {\n      // TODO(julianoks): Return null for condition gradient\n      // when backprop supports it.\n      condition: () => cast(zerosLike(condition), 'float32'),\n      t: () => mul(dy, cast(condition, dy.dtype)),\n      e: () => mul(dy, cast(logicalNot(condition), dy.dtype))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Selu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {exp} from '../ops/exp';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {SELU_SCALE, SELU_SCALEALPHA} from '../ops/selu_util';\nimport {where} from '../ops/where';\nimport {Tensor} from '../tensor';\n\nexport const seluGradConfig: GradConfig = {\n  kernelName: Selu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {\n      x: () => {\n        const mask = greater(x, scalar(0));\n\n        const scaleAlpha = scalar(SELU_SCALEALPHA);\n        const scale = scalar(SELU_SCALE);\n\n        const greaterThanZeroDer = mul(dy, scale);\n        const lessEqualZeroDer =\n            mul(mul(dy, scaleAlpha), exp(cast(x, 'float32')));\n\n        return where(mask, greaterThanZeroDer, lessEqualZeroDer);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sigmoid} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const sigmoidGradConfig: GradConfig = {\n  kernelName: Sigmoid,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(dy, mul(y, sub(scalar(1), y)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sign} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const signGradConfig: GradConfig = {\n  kernelName: Sign,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cos} from '../ops/cos';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinGradConfig: GradConfig = {\n  kernelName: Sin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(cos(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cosh} from '../ops/cosh';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinhGradConfig: GradConfig = {\n  kernelName: Sinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(cosh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Slice, SliceAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {pad} from '../ops/pad';\nimport {parseSliceParams} from '../ops/slice_util';\nimport {Tensor} from '../tensor';\n\nexport const sliceGradConfig: GradConfig = {\n  kernelName: Slice,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {begin, size} = attrs as {} as SliceAttrs;\n\n    const inputShape = x.shape;\n    const [begin_, size_] = parseSliceParams(x, begin, size);\n\n    // Create an Nx2 padding where the first column represents how many\n    // zeros are prepended (at start) for each dimension, and the second\n    // column indicates how many zeros are appended (at end).\n\n    // The number of zeros to append is the shape of the input\n    // elementwise-subtracted by both the begin vector and sizes vector.\n    const paddings: Array<[number, number]> = [];\n    for (let i = 0; i < dy.rank; i++) {\n      paddings.push([begin_[i], inputShape[i] - begin_[i] - size_[i]]);\n    }\n    return {x: () => pad(dy, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softmax, SoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const softmaxGradConfig: GradConfig = {\n  kernelName: Softmax,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [y] = saved;\n    const {dim} = attrs as {} as SoftmaxAttrs;\n    const keepDims = true;\n\n    const dyTimesY = mul(dy, y);\n    return {\n      logits: () => sub(dyTimesY, mul(sum(dyTimesY, [dim], keepDims), y))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softplus} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sigmoid} from '../ops/sigmoid';\nimport {Tensor} from '../tensor';\n\nexport const softplusGradConfig: GradConfig = {\n  kernelName: Softplus,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(dy, sigmoid(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SpaceToBatchND, SpaceToBatchNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {batchToSpaceND} from '../ops/batch_to_space_nd';\nimport {Tensor} from '../tensor';\n\nexport const spaceToBatchNDGradConfig: GradConfig = {\n  kernelName: SpaceToBatchND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, paddings} = attrs as {} as SpaceToBatchNDAttrs;\n    return {x: () => batchToSpaceND(dy, blockShape, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {SplitV, SplitVAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {concat} from '../ops/concat';\nimport {Tensor} from '../tensor';\n\nexport const splitVGradConfig: GradConfig = {\n  kernelName: SplitV,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const {axis} = attrs as {} as SplitVAttrs;\n\n    return {x: () => concat(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {Tensor} from '../tensor';\n\nexport const sqrtGradConfig: GradConfig = {\n  kernelName: Sqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, mul(sqrt(cast(x, 'float32')), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const squareGradConfig: GradConfig = {\n  kernelName: Square,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, mul(cast(x, 'float32'), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SquaredDifference} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const squaredDifferenceGradConfig: GradConfig = {\n  kernelName: SquaredDifference,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const two = scalar(2);\n    const derA = () => mul(dy, mul(two, sub(a, b)));\n    const derB = () => mul(dy, mul(two, sub(b, a)));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Step} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const stepGradConfig: GradConfig = {\n  kernelName: Step,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports\n    // it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Sub} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const subGradConfig: GradConfig = {\n  kernelName: Sub,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(neg(res), b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sum, SumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {ones} from '../ops/ones';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const sumGradConfig: GradConfig = {\n  kernelName: Sum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const expandedDyShape = x.shape.slice();\n    const {axis} = attrs as {} as SumAttrs;\n\n    const axes = parseAxisParam(axis, x.shape);\n    axes.forEach(axis => {\n      expandedDyShape[axis] = 1;\n    });\n    const expandedDy = reshape(dy, expandedDyShape);\n    const derX = mul(expandedDy, ones(x.shape, 'float32'));\n\n    return {x: () => derX};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cos} from '../ops/cos';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const tanGradConfig: GradConfig = {\n  kernelName: Tan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, square(cos(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const tanhGradConfig: GradConfig = {\n  kernelName: Tanh,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(sub(scalar(1), square(y)), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tile, TileAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {slice} from '../ops/slice';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const tileGradConfig: GradConfig = {\n  kernelName: Tile,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {reps} = attrs as unknown as TileAttrs;\n\n    const derX = () => {\n      let xGrad = zerosLike(x);\n      // TODO(cais): Maybe reduce memory footprint by avoiding repeated\n      // slicing.\n      if (x.rank === 1) {\n        for (let i = 0; i < reps[0]; ++i) {\n          xGrad = add(xGrad, slice(dy, [i * x.shape[0]], [x.shape[0]]));\n        }\n      } else if (x.rank === 2) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            xGrad = add(xGrad, slice(dy, [i * x.shape[0], j * x.shape[1]], [\n                          x.shape[0], x.shape[1]\n                        ]));\n          }\n        }\n      } else if (x.rank === 3) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              xGrad =\n                  add(xGrad,\n                      slice(\n                          dy, [i * x.shape[0], j * x.shape[1], k * x.shape[2]],\n                          [x.shape[0], x.shape[1], x.shape[2]]));\n            }\n          }\n        }\n      } else if (x.rank === 4) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              for (let l = 0; l < reps[3]; ++l) {\n                xGrad =\n                    add(xGrad,\n                        slice(\n                            dy,\n                            [\n                              i * x.shape[0], j * x.shape[1], k * x.shape[2],\n                              l * x.shape[3]\n                            ],\n                            [x.shape[0], x.shape[1], x.shape[2], x.shape[3]]));\n              }\n            }\n          }\n        }\n      } else {\n        throw new Error(\n            `Gradient for tile operation is not implemented for rank-` +\n            `${x.rank} tensors yet.`);\n      }\n      return xGrad;\n    };\n    return {x: derX};\n  },\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Transpose, TransposeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const transposeGradConfig: GradConfig = {\n  kernelName: Transpose,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const transposeAttrs: TransposeAttrs = attrs as {} as TransposeAttrs;\n    const {perm} = transposeAttrs;\n    const undoPerm = axis_util.getUndoAxesPermutation(perm);\n    return {x: () => transpose(dy, undoPerm)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Unpack, UnpackAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {stack} from '../ops/stack';\nimport {Tensor} from '../tensor';\n\nexport const unpackGradConfig: GradConfig = {\n  kernelName: Unpack,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const unpackAttrs: UnpackAttrs = attrs as {} as UnpackAttrs;\n    const {axis} = unpackAttrs;\n    return {value: () => stack(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {UnsortedSegmentSum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {expandDims} from '../ops/expand_dims';\nimport {gather} from '../ops/gather';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {maximum} from '../ops/maximum';\nimport {ones} from '../ops/ones';\nimport {scalar} from '../ops/scalar';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor, Tensor1D} from '../tensor';\n\nexport const unsortedSegmentSumGradConfig: GradConfig = {\n  kernelName: UnsortedSegmentSum,\n  inputsToSave: ['segmentIds'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [segmentIds] = saved;\n\n    const derX = () => {\n      return gatherDropNegatives(dy, segmentIds as Tensor1D);\n    };\n    return {x: derX};\n  }\n};\n\nfunction gatherDropNegatives<T extends Tensor>(x: T, indices: Tensor1D) {\n  // Helper function for unsorted segment ops. Gathers params for\n  // positive segment ids and gathers 0 for inputs with negative segment id.\n  // Mirrors _GatherDropNegatives from tensorflow/python/ops/math_grad.py\n  const zeroClippedIndices = maximum(indices, zerosLike(indices));\n  const gathered = gather(x, zeroClippedIndices as Tensor1D);\n  let isPositive = greaterEqual(indices, scalar(0, 'int32'));\n  const numIters = gathered.rank - isPositive.rank;\n  for (let i = 0; i < numIters; ++i) {\n    isPositive = expandDims(isPositive, i + 1);\n  }\n  isPositive = logicalAnd(isPositive, ones(gathered.shape, 'bool'));\n  const zeroSlice = zerosLike(gathered);\n  return where(isPositive, gathered, zeroSlice);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ZerosLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const zerosLikeGradConfig: GradConfig = {\n  kernelName: ZerosLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {absGradConfig} from './gradients/Abs_grad';\nimport {acosGradConfig} from './gradients/Acos_grad';\nimport {acoshGradConfig} from './gradients/Acosh_grad';\nimport {addGradConfig} from './gradients/Add_grad';\nimport {addNGradConfig} from './gradients/AddN_grad';\nimport {argMaxGradConfig} from './gradients/ArgMax_grad';\nimport {argMinGradConfig} from './gradients/ArgMin_grad';\nimport {asinGradConfig} from './gradients/Asin_grad';\nimport {asinhGradConfig} from './gradients/Asinh_grad';\nimport {atan2GradConfig} from './gradients/Atan2_grad';\nimport {atanGradConfig} from './gradients/Atan_grad';\nimport {atanhGradConfig} from './gradients/Atanh_grad';\nimport {avgPool3DGradConfig} from './gradients/AvgPool3D_grad';\nimport {avgPoolGradConfig} from './gradients/AvgPool_grad';\nimport {batchMatMulGradConfig} from './gradients/BatchMatMul_grad';\nimport {batchToSpaceNDGradConfig} from './gradients/BatchToSpaceND_grad';\nimport {broadcastToGradConfig} from './gradients/BroadcastTo_grad';\nimport {castGradConfig} from './gradients/Cast_grad';\nimport {ceilGradConfig} from './gradients/Ceil_grad';\nimport {clipByValueGradConfig} from './gradients/ClipByValue_grad';\nimport {concatGradConfig} from './gradients/Concat_grad';\nimport {conv2DGradConfig} from './gradients/Conv2D_grad';\nimport {conv2DBackpropInputGradConfig} from './gradients/Conv2DBackpropInput_grad';\nimport {conv3DGradConfig} from './gradients/Conv3D_grad';\nimport {cosGradConfig} from './gradients/Cos_grad';\nimport {coshGradConfig} from './gradients/Cosh_grad';\nimport {cumsumGradConfig} from './gradients/Cumsum_grad';\nimport {depthwiseConv2dNativeGradConfig} from './gradients/DepthwiseConv2dNative_grad';\nimport {dilation2dGradConfig} from './gradients/Dilation2D_grad';\nimport {divGradConfig} from './gradients/Div_grad';\nimport {eluGradConfig} from './gradients/Elu_grad';\nimport {erfGradConfig} from './gradients/Erf_grad';\nimport {expGradConfig} from './gradients/Exp_grad';\nimport {expm1GradConfig} from './gradients/Expm1_grad';\nimport {floorGradConfig} from './gradients/Floor_grad';\nimport {floorDivGradConfig} from './gradients/FloorDiv_grad';\nimport {fusedBatchNormGradConfig} from './gradients/FusedBatchNorm_grad';\nimport {gatherGradConfig} from './gradients/GatherV2_grad';\nimport {greaterEqualGradConfig} from './gradients/GreaterEqual_grad';\nimport {identityGradConfig} from './gradients/Identity_grad';\nimport {isFiniteGradConfig} from './gradients/IsFinite_grad';\nimport {isInfGradConfig} from './gradients/IsInf_grad';\nimport {isNanGradConfig} from './gradients/IsNan_grad';\nimport {log1pGradConfig} from './gradients/Log1p_grad';\nimport {logGradConfig} from './gradients/Log_grad';\nimport {logSoftmaxGradConfig} from './gradients/LogSoftmax_grad';\nimport {lrnGradConfig} from './gradients/LRN_grad';\nimport {maxGradConfig} from './gradients/Max_grad';\nimport {maximumGradConfig} from './gradients/Maximum_grad';\nimport {maxPool3DGradConfig} from './gradients/MaxPool3D_grad';\nimport {maxPoolGradConfig} from './gradients/MaxPool_grad';\nimport {minGradConfig} from './gradients/Min_grad';\nimport {minimumGradConfig} from './gradients/Minimum_grad';\nimport {modGradConfig} from './gradients/Mod_grad';\nimport {multiplyGradConfig} from './gradients/Multiply_grad';\nimport {negateGradConfig} from './gradients/Negate_grad';\nimport {oneHotGradConfig} from './gradients/OneHot_grad';\nimport {onesLikeGradConfig} from './gradients/OnesLike_grad';\nimport {padV2GradConfig} from './gradients/PadV2_grad';\nimport {powGradConfig} from './gradients/Pow_grad';\nimport {preluGradConfig} from './gradients/Prelu_grad';\nimport {reciprocalGradConfig} from './gradients/Reciprocal_grad';\nimport {relu6GradConfig} from './gradients/Relu6_grad';\nimport {reluGradConfig} from './gradients/Relu_grad';\nimport {reshapeGradConfig} from './gradients/Reshape_grad';\nimport {resizeBilinearGradConfig} from './gradients/ResizeBilinear_grad';\nimport {resizeNearestNeighborGradConfig} from './gradients/ResizeNearestNeighbor_grad';\nimport {reverseGradConfig} from './gradients/Reverse_grad';\nimport {roundGradConfig} from './gradients/Round_grad';\nimport {rsqrtGradConfig} from './gradients/Rsqrt_grad';\nimport {selectV2PoolGradConfig} from './gradients/SelectV2_grad';\nimport {seluGradConfig} from './gradients/Selu_grad';\nimport {sigmoidGradConfig} from './gradients/Sigmoid_grad';\nimport {signGradConfig} from './gradients/Sign_grad';\nimport {sinGradConfig} from './gradients/Sin_grad';\nimport {sinhGradConfig} from './gradients/Sinh_grad';\nimport {sliceGradConfig} from './gradients/Slice_grad';\nimport {softmaxGradConfig} from './gradients/Softmax_grad';\nimport {softplusGradConfig} from './gradients/Softplus_grad';\nimport {spaceToBatchNDGradConfig} from './gradients/SpaceToBatchND_grad';\nimport {splitVGradConfig} from './gradients/SplitV_grad';\nimport {sqrtGradConfig} from './gradients/Sqrt_grad';\nimport {squareGradConfig} from './gradients/Square_grad';\nimport {squaredDifferenceGradConfig} from './gradients/SquaredDifference_grad';\nimport {stepGradConfig} from './gradients/Step_grad';\nimport {subGradConfig} from './gradients/Sub_grad';\nimport {sumGradConfig} from './gradients/Sum_grad';\nimport {tanGradConfig} from './gradients/Tan_grad';\nimport {tanhGradConfig} from './gradients/Tanh_grad';\nimport {tileGradConfig} from './gradients/Tile_grad';\nimport {transposeGradConfig} from './gradients/Transpose_grad';\nimport {unpackGradConfig} from './gradients/Unpack_grad';\nimport {unsortedSegmentSumGradConfig} from './gradients/UnsortedSegmentSum_grad';\nimport {zerosLikeGradConfig} from './gradients/ZerosLike_grad';\nimport {GradConfig} from './kernel_registry';\nimport {registerGradient} from './kernel_registry';\n\n// Export all kernel configs here so that the package can auto register them\nconst gradConfigs: GradConfig[] = [\n  absGradConfig,\n  acosGradConfig,\n  acoshGradConfig,\n  addGradConfig,\n  addNGradConfig,\n  argMaxGradConfig,\n  argMinGradConfig,\n  asinGradConfig,\n  asinhGradConfig,\n  atan2GradConfig,\n  atanGradConfig,\n  atanhGradConfig,\n  avgPool3DGradConfig,\n  avgPoolGradConfig,\n  batchMatMulGradConfig,\n  batchToSpaceNDGradConfig,\n  broadcastToGradConfig,\n  castGradConfig,\n  ceilGradConfig,\n  clipByValueGradConfig,\n  concatGradConfig,\n  conv2DBackpropInputGradConfig,\n  conv2DGradConfig,\n  conv3DGradConfig,\n  cosGradConfig,\n  coshGradConfig,\n  cumsumGradConfig,\n  depthwiseConv2dNativeGradConfig,\n  dilation2dGradConfig,\n  divGradConfig,\n  eluGradConfig,\n  erfGradConfig,\n  expGradConfig,\n  expm1GradConfig,\n  floorDivGradConfig,\n  floorGradConfig,\n  fusedBatchNormGradConfig,\n  gatherGradConfig,\n  greaterEqualGradConfig,\n  identityGradConfig,\n  isFiniteGradConfig,\n  isInfGradConfig,\n  isNanGradConfig,\n  log1pGradConfig,\n  logGradConfig,\n  logSoftmaxGradConfig,\n  lrnGradConfig,\n  maxGradConfig,\n  maxGradConfig,\n  maximumGradConfig,\n  maxPool3DGradConfig,\n  maxPoolGradConfig,\n  minGradConfig,\n  minimumGradConfig,\n  modGradConfig,\n  multiplyGradConfig,\n  negateGradConfig,\n  oneHotGradConfig,\n  onesLikeGradConfig,\n  padV2GradConfig,\n  padV2GradConfig,\n  powGradConfig,\n  preluGradConfig,\n  reciprocalGradConfig,\n  relu6GradConfig,\n  reluGradConfig,\n  reshapeGradConfig,\n  resizeBilinearGradConfig,\n  resizeNearestNeighborGradConfig,\n  reverseGradConfig,\n  roundGradConfig,\n  rsqrtGradConfig,\n  selectV2PoolGradConfig,\n  seluGradConfig,\n  sigmoidGradConfig,\n  signGradConfig,\n  sinGradConfig,\n  sinhGradConfig,\n  sliceGradConfig,\n  softmaxGradConfig,\n  softplusGradConfig,\n  spaceToBatchNDGradConfig,\n  spaceToBatchNDGradConfig,\n  splitVGradConfig,\n  splitVGradConfig,\n  sqrtGradConfig,\n  squaredDifferenceGradConfig,\n  squareGradConfig,\n  stepGradConfig,\n  subGradConfig,\n  sumGradConfig,\n  tanGradConfig,\n  tanhGradConfig,\n  tileGradConfig,\n  transposeGradConfig,\n  unpackGradConfig,\n  unsortedSegmentSumGradConfig,\n  zerosLikeGradConfig\n];\n\nfor (const gradientConfig of gradConfigs) {\n  registerGradient(gradientConfig);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {abs} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    abs<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.abs = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return abs(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acos = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acosh = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {addStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    addStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.addStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return addStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {add} from '../../ops/add';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    add<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.add = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return add(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {all} from '../../ops/all';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    all<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.all = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return all(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {any} from '../../ops/any';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    any<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.any = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return any(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMax} from '../../ops/arg_max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMax<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMax = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return argMax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMin} from '../../ops/arg_min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMin<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMin = function<T extends Tensor>(axis: number): T {\n  this.throwIfDisposed();\n  return argMin(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Scalar, Tensor} from '../../tensor';\nimport {Rank} from '../../types';\nimport {assert} from '../../util';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asScalar<T extends Tensor>(): Scalar;\n  }\n}\n\n/** Converts a size-1 `tf.Tensor` to a `tf.Scalar`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asScalar = function<T extends Tensor>(this: T): Scalar {\n  this.throwIfDisposed();\n  assert(this.size === 1, () => 'The array must have only 1 element.');\n  return reshape(this, []);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asType<T extends Tensor>(this: T, dtype: DataType): T;\n  }\n}\n\n/**\n * Casts a `tf.Tensor` to a specified dtype.\n *\n * @param dtype Data-type to cast the tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asType = function<T extends Tensor>(\n    this: T, dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast<T>(this, dtype);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as1D<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Converts a `tf.Tensor` to a `tf.Tensor1D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as1D = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as2D<T extends Tensor>(rows: number, columns: number): Tensor2D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor2D`.\n *\n * @param rows Number of rows in `tf.Tensor2D`.\n * @param columns Number of columns in `tf.Tensor2D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as2D = function<T extends Tensor>(\n    rows: number, columns: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as3D<T extends Tensor>(rows: number, columns: number, depth: number):\n        Tensor3D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor3D`.\n *\n * @param rows Number of rows in `tf.Tensor3D`.\n * @param columns Number of columns in `tf.Tensor3D`.\n * @param depth Depth of `tf.Tensor3D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as3D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as4D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number): Tensor4D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor4D`.\n *\n * @param rows Number of rows in `tf.Tensor4D`.\n * @param columns Number of columns in `tf.Tensor4D`.\n * @param depth Depth of `tf.Tensor4D`.\n * @param depth2 4th dimension of `tf.Tensor4D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as4D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as5D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number,\n        depth3: number): Tensor5D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor5D`.\n *\n * @param rows Number of rows in `tf.Tensor5D`.\n * @param columns Number of columns in `tf.Tensor5D`.\n * @param depth Depth of `tf.Tensor5D`.\n * @param depth2 4th dimension of `tf.Tensor5D`.\n * @param depth3 5th dimension of 'tf.Tensor5D'\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as5D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number,\n    depth3: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2, depth3]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {atan2} from '../../ops/atan2';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan2<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.atan2 = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return atan2(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {avgPool} from '../../ops/avg_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    avgPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.avgPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return avgPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchToSpaceND} from '../../ops/batch_to_space_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchToSpaceND<R extends Rank>(blockShape: number[], crops: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.batchToSpaceND = function<R extends Rank>(\n    blockShape: number[], crops: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return batchToSpaceND(this, blockShape, crops);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchNorm} from '../../ops/batchnorm';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchNorm<T extends Tensor>(\n        mean: Tensor<R>|Tensor1D|TensorLike,\n        variance: Tensor<R>|Tensor1D|TensorLike,\n        offset?: Tensor<R>|Tensor1D|TensorLike,\n        scale?: Tensor<R>|Tensor1D|TensorLike,\n        varianceEpsilon?: number): Tensor<R>;\n  }\n}\n\nTensor.prototype.batchNorm = function<R extends Rank>(\n    mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  this.throwIfDisposed();\n  return batchNorm(this, mean, variance, offset, scale, varianceEpsilon);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {broadcastTo} from '../../ops/broadcast_to';\nimport {Tensor} from '../../tensor';\nimport {Rank, ShapeMap} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    broadcastTo<R extends Rank>(shape: ShapeMap[R]): Tensor<R>;\n  }\n}\n\nTensor.prototype.broadcastTo = function<R extends Rank>(shape: ShapeMap[R]):\n    Tensor<R> {\n  this.throwIfDisposed();\n  return broadcastTo(this, shape);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cast<T extends Tensor>(dtype: DataType): T;\n  }\n}\n\nTensor.prototype.cast = function<T extends Tensor>(dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast(this, dtype) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ceil} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ceil<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.ceil = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return ceil(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {clipByValue} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    clipByValue<T extends Tensor>(min: number, max: number): Tensor;\n  }\n}\n\nTensor.prototype.clipByValue = function<T extends Tensor>(\n    min: number, max: number): T {\n  this.throwIfDisposed();\n  return clipByValue(this, min, max) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {concat} from '../../ops/concat';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    concat<T extends Tensor>(tensors: T|Array<T|TensorLike>, axis?: number): T;\n  }\n}\n\nTensor.prototype.concat = function<T extends Tensor>(\n    x: T|Array<T|TensorLike>, axis?: number): T {\n  this.throwIfDisposed();\n  if (x instanceof Tensor) {\n    x = [x];\n  }\n  return concat([this, ...x], axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv1d} from '../../ops/conv1d';\nimport {ExplicitPadding} from '../../ops/conv_util';\nimport {Tensor, Tensor2D, Tensor3D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv1d<T extends Tensor2D|Tensor3D>(\n        filter: Tensor3D|TensorLike3D, stride: number,\n        pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n        dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv1d = function<T extends Tensor2D|Tensor3D>(\n    filter: Tensor3D|TensorLike3D, stride: number,\n    pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n    dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv1d(\n             this, filter, stride, pad, dataFormat, dilation,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2dTranspose} from '../../ops/conv2d_transpose';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2dTranspose<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D,\n        outputShape: [number, number, number, number]|[number, number, number],\n        strides: [number, number]|number, pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2dTranspose = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2dTranspose(\n             this, filter, outputShape, strides, pad, dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2d} from '../../ops/conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cos = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cosh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {cumsum} from '../../ops/cumsum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cumsum<R extends Rank>(\n        axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R>;\n  }\n}\n\nTensor.prototype.cumsum = function<R extends Rank>(\n    axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R> {\n  this.throwIfDisposed();\n  return cumsum(this, axis, exclusive, reverse);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthToSpace} from '../../ops/depth_to_space';\nimport {Tensor, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthToSpace<T extends Tensor4D>(\n        blockSize: number, dataFormat: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.depthToSpace = function<T extends Tensor4D>(\n    blockSize: number, dataFormat: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return depthToSpace(this, blockSize, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../../globals';\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2D<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\n/**\n * @deprecated Use `depthwiseConv2d` instead.\n */\nTensor.prototype.depthwiseConv2D = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  deprecationWarn('depthwiseConv2D is deprecated, use depthwiseConv2d instead');\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.depthwiseConv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dilation2d} from '../../ops/dilation2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dilation2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n        pad: 'valid'|'same', dilations?: [number, number]|number,\n        dataFormat?: 'NHWC'): T;\n  }\n}\n\nTensor.prototype.dilation2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n    pad: 'valid'|'same', dilations?: [number, number]|number,\n    dataFormat?: 'NHWC'): T {\n  this.throwIfDisposed();\n  return dilation2d(this, filter, strides, pad, dilations, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {divNoNan} from '../../ops/div_no_nan';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divNoNan<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.divNoNan = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return divNoNan(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {divStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.divStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return divStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {div} from '../../ops/div';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    div<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.div = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return div(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dot} from '../../ops/dot';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dot<T extends Tensor>(b: Tensor|TensorLike): Tensor;\n  }\n}\n\nTensor.prototype.dot = function<T extends Tensor>(b: T|TensorLike): Tensor {\n  this.throwIfDisposed();\n  return dot(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {elu} from '../../ops/elu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    elu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.elu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return elu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {equalStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equalStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.equalStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return equalStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {equal} from '../../ops/equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equal<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.equal = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return equal(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {erf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    erf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.erf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return erf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {exp} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    exp<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.exp = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return exp(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {expandDims} from '../../ops/expand_dims';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expandDims<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.expandDims = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return expandDims(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {expm1} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expm1<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.expm1 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return expm1(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {fft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    fft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.fft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return fft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    flatten<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Flatten a Tensor to a 1D array.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.flatten = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {floor} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floor<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.floor = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return floor(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {floorDiv} from '../../ops/floorDiv';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floorDiv<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.floorDiv = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return floorDiv(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {gather} from '../../ops/gather';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    gather<T extends Tensor>(\n        this: T, indices: Tensor|TensorLike, axis?: number): T;\n  }\n}\n\nTensor.prototype.gather = function<T extends Tensor>(\n    this: T, indices: Tensor|TensorLike, axis?: number): T {\n  this.throwIfDisposed();\n  return gather(this, indices, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return greaterEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greaterEqual} from '../../ops/greater_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greaterEqual = function<T extends Tensor>(b: Tensor|\n                                                           TensorLike): T {\n  this.throwIfDisposed();\n  return greaterEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return greaterStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greater} from '../../ops/greater';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greater<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greater = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return greater(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ifft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ifft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.ifft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return ifft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {irfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    irfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.irfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return irfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isFinite} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isFinite<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isFinite = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isFinite(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isInf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isInf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isInf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isInf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isNaN} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isNaN<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isNaN = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isNaN(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {leakyRelu} from '../../ops/leaky_relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    leakyRelu<T extends Tensor>(alpha: number): T;\n  }\n}\n\nTensor.prototype.leakyRelu = function<T extends Tensor>(\n    this: T, alpha: number): T {\n  this.throwIfDisposed();\n  return leakyRelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.lessEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {lessEqual} from '../../ops/less_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessEqual = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return lessEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {less} from '../../ops/less';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    less<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.less = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return less(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {localResponseNormalization} from '../../ops/local_response_normalization';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    localResponseNormalization<T extends Tensor>(\n        depthRadius?: number, bias?: number, alpha?: number, beta?: number): T;\n  }\n}\n\nTensor.prototype.localResponseNormalization = function<T extends Tensor>(\n    depthRadius?: number, bias?: number, alpha?: number, beta?: number): T {\n  this.throwIfDisposed();\n  return localResponseNormalization(this, depthRadius, bias, alpha, beta) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.logSigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return logSigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSoftmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSoftmax<T extends Tensor>(this: T, axis?: number): T;\n  }\n}\n\nTensor.prototype.logSoftmax = function<T extends Tensor>(\n    this: T, axis?: number): T {\n  this.throwIfDisposed();\n  return logSoftmax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logSumExp} from '../../ops/log_sum_exp';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSumExp<T extends Tensor>(\n        this: T, axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.logSumExp = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return logSumExp(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log1p} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log1p<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log1p = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log1p(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalAnd} from '../../ops/logical_and';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalAnd<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalAnd = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalAnd(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalNot} from '../../ops/logical_not';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalNot<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.logicalNot = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return logicalNot(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalOr} from '../../ops/logical_or';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalOr<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalOr = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return logicalOr(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalXor} from '../../ops/logical_xor';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalXor<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalXor = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalXor(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {matMul} from '../../ops/mat_mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    matMul<T extends Tensor>(\n        b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T;\n  }\n}\n\nTensor.prototype.matMul = function<T extends Tensor>(\n    this: T, b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T {\n  this.throwIfDisposed();\n  return matMul(this, b, transposeA, transposeB);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maxPool} from '../../ops/max_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maxPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.maxPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return maxPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {max} from '../../ops/max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    max<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.max = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return max(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {maximumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.maximumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return maximumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maximum} from '../../ops/maximum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.maximum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return maximum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mean} from '../../ops/mean';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mean<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.mean = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return mean(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {min} from '../../ops/min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    min<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.min = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return min(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {minimumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.minimumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return minimumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {minimum} from '../../ops/minimum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.minimum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return minimum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {modStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    modStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.modStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return modStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mod} from '../../ops/mod';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mod<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mod = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mod(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {mulStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mulStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.mulStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return mulStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mul} from '../../ops/mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mul<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mul = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mul(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {neg} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    neg<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.neg = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return neg(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {norm} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    norm<T extends Tensor>(\n        ord?: number|'euclidean'|'fro', axis?: number|number[],\n        keepDims?: boolean): Tensor;\n  }\n}\n\nTensor.prototype.norm = function<T extends Tensor>(\n    ord?: number|'euclidean'|'fro', axis?: number|number[],\n    keepDims?: boolean) {\n  this.throwIfDisposed();\n  return norm(this, ord, axis, keepDims) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {notEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.notEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return notEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {notEqual} from '../../ops/not_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.notEqual = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return notEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {oneHot} from '../../ops/one_hot';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    oneHot(depth: number, onValue: number, offValue: number): Tensor;\n  }\n}\n\nTensor.prototype.oneHot = function(\n    depth: number, onValue = 1, offValue = 0): Tensor {\n  this.throwIfDisposed();\n  return oneHot(this, depth, onValue, offValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {onesLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    onesLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.onesLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return onesLike(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {pad} from '../../ops/pad';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pad<T extends Tensor>(\n        paddings: Array<[number, number]>, constantValue?: number): T;\n  }\n}\n\nTensor.prototype.pad = function<T extends Tensor>(\n    this: T, paddings: Array<[number, number]>, constantValue: number): T {\n  this.throwIfDisposed();\n  return pad(this, paddings, constantValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pool} from '../../ops/pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pool<T extends Tensor3D|Tensor4D>(\n        windowShape: [number, number]|number, poolingType: 'avg'|'max',\n        padding: 'valid'|'same'|number, diationRate?: [number, number]|number,\n        strides?: [number, number]|number): T;\n  }\n}\n\nTensor.prototype.pool = function<T extends Tensor3D|Tensor4D>(\n    this: T, windowShape: [number, number]|number, poolingType: 'max'|'avg',\n    padding: 'valid'|'same'|number, dilationRate?: [number, number]|number,\n    strides?: [number, number]|number): T {\n  this.throwIfDisposed();\n  return pool(this, windowShape, poolingType, padding, dilationRate, strides);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {powStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    powStrict<T extends Tensor>(exp: Tensor): Tensor;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.powStrict = function<T extends Tensor>(exp: Tensor): T {\n  this.throwIfDisposed();\n  return powStrict(this, exp) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pow} from '../../ops/pow';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pow<T extends Tensor>(exp: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.pow = function<T extends Tensor>(exp: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return pow(this, exp);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prelu} from '../../ops/prelu';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prelu<T extends Tensor>(alpha: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.prelu = function<T extends Tensor>(\n    this: T, alpha: T|TensorLike): T {\n  this.throwIfDisposed();\n  return prelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prod} from '../../ops/prod';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prod<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.prod = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return prod(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {reciprocal} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reciprocal<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.reciprocal = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return reciprocal(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu} from '../../ops/relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu6} from '../../ops/relu6';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu6<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu6 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu6(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshapeAs<T extends Tensor>(x: T): T;\n  }\n}\n\n/**\n * Reshapes the tensor into the shape of the provided tensor.\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.reshapeAs = function<T extends Tensor>(x: T): T {\n  this.throwIfDisposed();\n  return reshape(this, x.shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshape<T extends Tensor>(shape: number[]): T;\n  }\n}\n\nTensor.prototype.reshape = function<T extends Tensor>(shape: number[]): T {\n  this.throwIfDisposed();\n  return reshape(this, shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeBilinear} from '../../ops/image/resize_bilinear';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeBilinear<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeBilinear = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeBilinear(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeNearestNeighbor} from '../../ops/image/resize_nearest_neighbor';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeNearestNeighbor<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeNearestNeighbor = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeNearestNeighbor(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reverse} from '../../ops/reverse';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reverse<T extends Tensor>(this: T, axis?: number|number[]): T;\n  }\n}\n\nTensor.prototype.reverse = function<T extends Tensor>(\n    this: T, axis?: number|number[]): T {\n  this.throwIfDisposed();\n  return reverse(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.rfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return rfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {round} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    round<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.round = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return round(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rsqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rsqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.rsqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return rsqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {selu} from '../../ops/selu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    selu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.selu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return selu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {separableConv2d} from '../../ops/separable_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    separableConv2d<T extends Tensor3D|Tensor4D>(\n        depthwiseFilter: Tensor4D|TensorLike4D,\n        pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n        pad: 'valid'|'same', dilation?: [number, number]|number,\n        dataFormat?: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.separableConv2d = function<T extends Tensor3D|Tensor4D>(\n    depthwiseFilter: Tensor4D|TensorLike4D,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation?: [number, number]|number,\n    dataFormat?: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return separableConv2d(\n             this, depthwiseFilter, pointwiseFilter, strides, pad, dilation,\n             dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sign} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sign<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sign = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sign(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {slice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    slice<T extends Tensor>(\n        this: T, begin: number|number[], size?: number|number[]): T;\n  }\n}\n\nTensor.prototype.slice = function<T extends Tensor>(\n    this: T, begin: number|number[], size?: number|number[]): T {\n  this.throwIfDisposed();\n  return slice(this, begin, size);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softmax<T extends Tensor>(this: T, dim?: number): T;\n  }\n}\n\nTensor.prototype.softmax = function<T extends Tensor>(this: T, dim: number): T {\n  this.throwIfDisposed();\n  return softmax(this, dim);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softplus} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softplus<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.softplus = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return softplus(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {spaceToBatchND} from '../../ops/space_to_batch_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    spaceToBatchND<R extends Rank>(blockShape: number[], paddings: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.spaceToBatchND = function<R extends Rank>(\n    blockShape: number[], paddings: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return spaceToBatchND(this, blockShape, paddings);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {split} from '../../ops/split';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    split<T extends Tensor>(numOrSizeSplits: number[]|number, axis?: number):\n        T[];\n  }\n}\n\nTensor.prototype.split = function<T extends Tensor>(\n    numOrSizeSplits: number[]|number, axis?: number): T[] {\n  this.throwIfDisposed();\n  return split(this, numOrSizeSplits, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {square} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    square<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.square = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return square(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {squaredDifference} from '../../ops/squared_difference';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifference<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.squaredDifference = function<T extends Tensor>(b: Tensor|\n                                                                TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifference(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {squaredDifferenceStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifferenceStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.squaredDifferenceStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifferenceStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {squeeze} from '../../ops/squeeze';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squeeze<T extends Tensor>(axis?: number[]): T;\n  }\n}\n\nTensor.prototype.squeeze = function<T extends Tensor>(axis?: number[]): T {\n  this.throwIfDisposed();\n  return squeeze(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {stack} from '../../ops/stack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stack<T extends Tensor>(x: Tensor|Tensor[], axis?: number): T;\n  }\n}\n\nTensor.prototype.stack = function<T extends Tensor>(\n    x: Tensor|Tensor[], axis?: number): T {\n  this.throwIfDisposed();\n  const tensorsToBeStacked = x instanceof Tensor ? [this, x] : [this, ...x];\n  return stack(tensorsToBeStacked, axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {step} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    step<T extends Tensor>(this: T, alpha?: number): T;\n  }\n}\n\nTensor.prototype.step = function<T extends Tensor>(this: T, alpha?: number) {\n  this.throwIfDisposed();\n  return step(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {stridedSlice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stridedSlice<T extends Tensor>(\n        this: Tensor, begin: number[], end: number[], strides: number[],\n        beginMask?: number, endMask?: number, ellipsisMask?: number,\n        newAxisMask?: number, shrinkAxisMask?: number): Tensor;\n  }\n}\n\nTensor.prototype.stridedSlice = function<T extends Tensor>(\n    this: Tensor, begin: number[], end: number[], strides: number[],\n    beginMask?: number, endMask?: number, ellipsisMask?: number,\n    newAxisMask?: number, shrinkAxisMask?: number): T {\n  this.throwIfDisposed();\n  return stridedSlice(\n             this, begin, end, strides, beginMask, endMask, ellipsisMask,\n             newAxisMask, shrinkAxisMask) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {subStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    subStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.subStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return subStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sub} from '../../ops/sub';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sub<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.sub = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return sub(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sum} from '../../ops/sum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sum<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.sum = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return sum(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tile} from '../../ops/tile';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tile<T extends Tensor>(b: number[]): T;\n  }\n}\n\nTensor.prototype.tile = function<T extends Tensor>(reps: number[]): T {\n  this.throwIfDisposed();\n  return tile(this, reps) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toBool<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `bool`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toBool = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'bool');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toFloat<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `float32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toFloat = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'float32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toInt<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `int32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toInt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'int32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {topk} from '../../ops/topk';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    topk<T extends Tensor>(this: T, k?: number, sorted?: boolean):\n        {values: T, indices: T};\n  }\n}\n\nTensor.prototype.topk = function<T extends Tensor>(\n    this: T, k?: number, sorted?: boolean): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return topk(this, k, sorted) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {transpose} from '../../ops/transpose';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    transpose<T extends Tensor>(perm?: number[]): T;\n  }\n}\n\nTensor.prototype.transpose = function<T extends Tensor>(\n    this: T, perm?: number[]): T {\n  this.throwIfDisposed();\n  return transpose(this, perm);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unique} from '../../ops/unique';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unique<T extends Tensor>(this: T, axis?: number): {values: T, indices: T};\n  }\n}\n\nTensor.prototype.unique = function<T extends Tensor>(\n    this: T, axis?: number): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return unique(this, axis) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unsortedSegmentSum} from '../../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike1D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unsortedSegmentSum<T extends Tensor>(\n        this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T;\n  }\n}\n\nTensor.prototype.unsortedSegmentSum = function<T extends Tensor>(\n    this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T {\n  this.throwIfDisposed();\n  return unsortedSegmentSum(this, segmentIds, numSegments);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {unstack} from '../../ops/unstack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unstack<T extends Tensor>(axis?: number): T[];\n  }\n}\n\nTensor.prototype.unstack = function<T extends Tensor>(axis?: number): T[] {\n  this.throwIfDisposed();\n  return unstack(this, axis) as T[];\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {where} from '../../ops/where';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    where<T extends Tensor>(condition: Tensor|TensorLike, x: Tensor|TensorLike):\n        T;\n  }\n}\n\nTensor.prototype.where = function<T extends Tensor>(\n    condition: Tensor|TensorLike, x: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return where(condition, this, x) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {zerosLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    zerosLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.zerosLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return zerosLike(this);\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {backend} from '@tensorflow/tfjs-core';\nimport {DataFormat} from '../keras_format/common';\n\nlet _epsilon: number;\n\n/**\n * Returns the value of the fuzz factor used in numeric expressions.\n */\nexport function epsilon() {\n  if (_epsilon == null) {\n    _epsilon = backend().epsilon();\n  }\n  return _epsilon;\n}\n\n/**\n * Sets the value of the fuzz factor used in numeric expressions.\n * @param e New value of epsilon.\n */\nexport function setEpsilon(e: number) {\n  _epsilon = e;\n}\n\n/**\n * Returns the default image data format convention.\n */\nexport function imageDataFormat(): DataFormat {\n  return 'channelsLast';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Explicit error types.\n *\n * See the following link for more information about why the code includes\n * calls to setPrototypeOf:\n *\n * https://github.com/Microsoft/TypeScript-wiki/blob/master/Breaking-Changes.md#extending-built-ins-like-error-array-and-map-may-no-longer-work\n */\n// tslint:enable\n\n/**\n * Equivalent of Python's AttributeError.\n */\nexport class AttributeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AttributeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's RuntimeError.\n */\nexport class RuntimeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, RuntimeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's ValueError.\n */\nexport class ValueError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, ValueError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's NotImplementedError.\n */\nexport class NotImplementedError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, NotImplementedError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's AssertionError.\n */\nexport class AssertionError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AssertionError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's IndexError.\n */\nexport class IndexError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, IndexError.prototype);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {DataType, fused, serialization, util} from '@tensorflow/tfjs-core';\n\nimport {AssertionError, ValueError} from '../errors';\n\n// tslint:enable\n\n/**\n * If `value` is an Array, equivalent to Python's `value * numValues`.\n * If `value` is not an Array, equivalent to Python's `[value] * numValues`\n */\n// tslint:disable-next-line:no-any\nexport function pyListRepeat(value: any, numValues: number): any[] {\n  if (Array.isArray(value)) {\n    // tslint:disable-next-line:no-any\n    let newArray: any[] = [];\n    for (let i = 0; i < numValues; i++) {\n      newArray = newArray.concat(value);\n    }\n    return newArray;\n  } else {\n    const newArray = new Array(numValues);\n    newArray.fill(value);\n    return newArray;\n  }\n}\n\nexport function assert(val: boolean, message?: string): void {\n  if (!val) {\n    throw new AssertionError(message);\n  }\n}\n\n/**\n * Count the number of elements of the `array` that are equal to `reference`.\n */\nexport function count<T>(array: T[], refernce: T) {\n  let counter = 0;\n  for (const item of array) {\n    if (item === refernce) {\n      counter++;\n    }\n  }\n  return counter;\n}\n\n/**\n * If an array is of length 1, just return the first element. Otherwise, return\n * the full array.\n * @param tensors\n */\nexport function singletonOrArray<T>(xs: T[]): T|T[] {\n  if (xs.length === 1) {\n    return xs[0];\n  }\n  return xs;\n}\n\n/**\n * Normalizes a list/tensor into a list.\n *\n * If a tensor is passed, we return\n * a list of size 1 containing the tensor.\n *\n * @param x target object to be normalized.\n */\n// tslint:disable-next-line:no-any\nexport function toList(x: any): any[] {\n  if (Array.isArray(x)) {\n    return x;\n  }\n  return [x];\n}\n\n/**\n * Generate a UID for a list\n */\n// tslint:disable-next-line:no-any\nexport function objectListUid(objs: any|any[]): string {\n  const objectList = toList(objs);\n  let retVal = '';\n  for (const obj of objectList) {\n    if (obj.id == null) {\n      throw new ValueError(\n          `Object ${obj} passed to objectListUid without an id`);\n    }\n    if (retVal !== '') {\n      retVal = retVal + ', ';\n    }\n    retVal = `${retVal}${Math.abs(obj.id)}`;\n  }\n  return retVal;\n}\n/**\n * Converts string to snake-case.\n * @param name\n */\nexport function toSnakeCase(name: string): string {\n  const intermediate = name.replace(/(.)([A-Z][a-z0-9]+)/g, '$1_$2');\n  const insecure =\n      intermediate.replace(/([a-z])([A-Z])/g, '$1_$2').toLowerCase();\n  /*\n   If the class is private the name starts with \"_\" which is not secure\n   for creating scopes. We prefix the name with \"private\" in this case.\n   */\n  if (insecure[0] !== '_') {\n    return insecure;\n  }\n  return 'private' + insecure;\n}\n\nexport function toCamelCase(identifier: string): string {\n  // quick return for empty string or single character strings\n  if (identifier.length <= 1) {\n    return identifier;\n  }\n  // Check for the underscore indicating snake_case\n  if (identifier.indexOf('_') === -1) {\n    return identifier;\n  }\n  return identifier.replace(/[_]+(\\w|$)/g, (m, p1) => p1.toUpperCase());\n}\n\n// tslint:disable-next-line:no-any\nlet _GLOBAL_CUSTOM_OBJECTS = {} as {[objName: string]: any};\n\nexport function serializeKerasObject(instance: serialization.Serializable):\n    serialization.ConfigDictValue {\n  if (instance === null || instance === undefined) {\n    return null;\n  }\n  const dict: serialization.ConfigDictValue = {};\n  dict['className'] = instance.getClassName();\n  dict['config'] = instance.getConfig();\n  return dict;\n}\n\n/**\n * Replace ndarray-style scalar objects in serialization objects with numbers.\n *\n * Background: In some versions of tf.keras, certain scalar values in the HDF5\n * model save file can be serialized as: `{'type': 'ndarray', 'value': num}`,\n * where in `num` is a plain number. This method converts such serialization\n * to a `number`.\n *\n * @param config The keras-format serialization object to be processed\n *   (in place).\n */\nfunction convertNDArrayScalarsInConfig(config: serialization.ConfigDictValue):\n    void {\n  if (config == null || typeof config !== 'object') {\n    return;\n  } else if (Array.isArray(config)) {\n    config.forEach(configItem => convertNDArrayScalarsInConfig(configItem));\n  } else {\n    const fields = Object.keys(config);\n    for (const field of fields) {\n      const value = config[field];\n      if (value != null && typeof value === 'object') {\n        if (!Array.isArray(value) && value['type'] === 'ndarray' &&\n            typeof value['value'] === 'number') {\n          config[field] = value['value'];\n        } else {\n          convertNDArrayScalarsInConfig(value as serialization.ConfigDict);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Deserialize a saved Keras Object\n * @param identifier either a string ID or a saved Keras dictionary\n * @param moduleObjects a list of Python class names to object constructors\n * @param customObjects a list of Python class names to object constructors\n * @param printableModuleName debug text for the object being reconstituted\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns a TensorFlow.js Layers object\n */\n// tslint:disable:no-any\nexport function deserializeKerasObject(\n    identifier: string|serialization.ConfigDict,\n    moduleObjects = {} as {[objName: string]: any},\n    customObjects = {} as {[objName: string]: any},\n    printableModuleName = 'object', fastWeightInit = false): any {\n  // tslint:enable\n  if (typeof identifier === 'string') {\n    const functionName = identifier;\n    let fn;\n    if (functionName in customObjects) {\n      fn = customObjects[functionName];\n    } else if (functionName in _GLOBAL_CUSTOM_OBJECTS) {\n      fn = _GLOBAL_CUSTOM_OBJECTS[functionName];\n    } else {\n      fn = moduleObjects[functionName];\n      if (fn == null) {\n        throw new ValueError(\n            `Unknown ${printableModuleName}: ${identifier}. ` +\n            `This may be due to one of the following reasons:\\n` +\n            `1. The ${printableModuleName} is defined in Python, in which ` +\n            `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n            `code.\\n` +\n            `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n            `but is not registered properly with ` +\n            `tf.serialization.registerClass().`);\n        // TODO(cais): Add link to tutorial page on custom layers.\n      }\n    }\n    return fn;\n  } else {\n    // In this case we are dealing with a Keras config dictionary.\n    const config = identifier;\n    if (config['className'] == null || config['config'] == null) {\n      throw new ValueError(\n          `${printableModuleName}: Improper config format: ` +\n          `${JSON.stringify(config)}.\\n` +\n          `'className' and 'config' must set.`);\n    }\n    const className = config['className'] as string;\n    let cls, fromConfig;\n    if (className in customObjects) {\n      [cls, fromConfig] = customObjects[className];\n    } else if (className in _GLOBAL_CUSTOM_OBJECTS) {\n      [cls, fromConfig] = _GLOBAL_CUSTOM_OBJECTS['className'];\n    } else if (className in moduleObjects) {\n      [cls, fromConfig] = moduleObjects[className];\n    }\n    if (cls == null) {\n      throw new ValueError(\n          `Unknown ${printableModuleName}: ${className}. ` +\n          `This may be due to one of the following reasons:\\n` +\n          `1. The ${printableModuleName} is defined in Python, in which ` +\n          `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n          `code.\\n` +\n          `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n          `but is not registered properly with ` +\n          `tf.serialization.registerClass().`);\n      // TODO(cais): Add link to tutorial page on custom layers.\n    }\n    if (fromConfig != null) {\n      // Porting notes: Instead of checking to see whether fromConfig accepts\n      // customObjects, we create a customObjects dictionary and tack it on to\n      // config['config'] as config['config'].customObjects. Objects can use it,\n      // if they want.\n\n      // tslint:disable-next-line:no-any\n      const customObjectsCombined = {} as {[objName: string]: any};\n      for (const key of Object.keys(_GLOBAL_CUSTOM_OBJECTS)) {\n        customObjectsCombined[key] = _GLOBAL_CUSTOM_OBJECTS[key];\n      }\n      for (const key of Object.keys(customObjects)) {\n        customObjectsCombined[key] = customObjects[key];\n      }\n      // Add the customObjects to config\n      const nestedConfig = config['config'] as serialization.ConfigDict;\n      nestedConfig['customObjects'] = customObjectsCombined;\n\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      convertNDArrayScalarsInConfig(config['config']);\n      const returnObj =\n          fromConfig(cls, config['config'], customObjects, fastWeightInit);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n\n      return returnObj;\n    } else {\n      // Then `cls` may be a function returning a class.\n      // In this case by convention `config` holds\n      // the kwargs of the function.\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      // In python this is **config['config'], for tfjs-layers we require\n      // classes that use this fall-through construction method to take\n      // a config interface that mimics the expansion of named parameters.\n      const returnObj = new cls(config['config']);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n      return returnObj;\n    }\n  }\n}\n\n/**\n * Compares two numbers for sorting.\n * @param a\n * @param b\n */\nexport function numberCompare(a: number, b: number) {\n  return (a < b) ? -1 : ((a > b) ? 1 : 0);\n}\n\n/**\n * Comparison of two numbers for reverse sorting.\n * @param a\n * @param b\n */\nexport function reverseNumberCompare(a: number, b: number) {\n  return -1 * numberCompare(a, b);\n}\n\n/**\n * Convert a string into the corresponding DType.\n * @param dtype\n * @returns An instance of DType.\n */\nexport function stringToDType(dtype: string): DataType {\n  switch (dtype) {\n    case 'float32':\n      return 'float32';\n    default:\n      throw new ValueError(`Invalid dtype: ${dtype}`);\n  }\n}\n\n/**\n * Test the element-by-element equality of two Arrays of strings.\n * @param xs First array of strings.\n * @param ys Second array of strings.\n * @returns Wether the two arrays are all equal, element by element.\n */\nexport function stringsEqual(xs: string[], ys: string[]): boolean {\n  if (xs == null || ys == null) {\n    return xs === ys;\n  }\n  if (xs.length !== ys.length) {\n    return false;\n  }\n  for (let i = 0; i < xs.length; ++i) {\n    if (xs[i] !== ys[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Get the unique elements of an array.\n * @param xs Array.\n * @returns An Array consisting of the unique elements in `xs`.\n */\nexport function unique<T>(xs: T[]): T[] {\n  if (xs == null) {\n    return xs;\n  }\n  const out: T[] = [];\n  // TODO(cais): Maybe improve performance by sorting.\n  for (const x of xs) {\n    if (out.indexOf(x) === -1) {\n      out.push(x);\n    }\n  }\n  return out;\n}\n\n/**\n * Determine if an Object is empty (i.e., does not have own properties).\n * @param obj Object\n * @returns Whether the Object is empty.\n * @throws ValueError: If object is `null` or `undefined`.\n */\nexport function isObjectEmpty(obj: {}): boolean {\n  if (obj == null) {\n    throw new ValueError(`Invalid value in obj: ${JSON.stringify(obj)}`);\n  }\n  for (const key in obj) {\n    if (obj.hasOwnProperty(key)) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Helper function used to build type union/enum run-time checkers.\n * @param values The list of allowed values.\n * @param label A string name for the type\n * @param value The value to test.\n * @throws ValueError: If the value is not in values nor `undefined`/`null`.\n */\nexport function checkStringTypeUnionValue(\n    values: string[], label: string, value: string): void {\n  if (value == null) {\n    return;\n  }\n  if (values.indexOf(value) < 0) {\n    throw new ValueError(`${value} is not a valid ${label}.  Valid values are ${\n        values} or null/undefined.`);\n  }\n}\n\n/**\n * Helper function for verifying the types of inputs.\n *\n * Ensures that the elements of `x` are all of type `expectedType`.\n * Also verifies that the length of `x` is within bounds.\n *\n * @param x Object to test.\n * @param expectedType The string expected type of all of the elements in the\n * Array.\n * @param minLength Return false if x.length is less than this.\n * @param maxLength Return false if x.length is greater than this.\n * @returns true if and only if `x` is an `Array<expectedType>` with\n * length >= `minLength` and <= `maxLength`.\n */\n// tslint:disable:no-any\nexport function checkArrayTypeAndLength(\n    x: any, expectedType: string, minLength = 0,\n    maxLength = Infinity): boolean {\n  assert(minLength >= 0);\n  assert(maxLength >= minLength);\n  return (\n      Array.isArray(x) && x.length >= minLength && x.length <= maxLength &&\n      x.every(e => typeof e === expectedType));\n}\n// tslint:enable:no-any\n\n/**\n * Assert that a value or an array of value are positive integer.\n *\n * @param value The value being asserted on. May be a single number or an array\n *   of numbers.\n * @param name Name of the value, used to make the error message.\n */\nexport function assertPositiveInteger(value: number|number[], name: string) {\n  if (Array.isArray(value)) {\n    util.assert(\n        value.length > 0, () => `${name} is unexpectedly an empty array.`);\n    value.forEach(\n        (v, i) => assertPositiveInteger(v, `element ${i + 1} of ${name}`));\n  } else {\n    util.assert(\n        Number.isInteger(value) && value > 0,\n        () => `Expected ${name} to be a positive integer, but got ` +\n            `${formatAsFriendlyString(value)}.`);\n  }\n}\n\n/**\n * Format a value into a display-friendly, human-readable fashion.\n *\n * - `null` is formatted as `'null'`\n * - Strings are formated with flanking pair of quotes.\n * - Arrays are formatted with flanking pair of square brackets.\n *\n * @param value The value to display.\n * @return Formatted string.\n */\n// tslint:disable-next-line:no-any\nexport function formatAsFriendlyString(value: any): string {\n  if (value === null) {\n    return 'null';\n  } else if (Array.isArray(value)) {\n    return '[' + value.map(v => formatAsFriendlyString(v)).join(',') + ']';\n  } else if (typeof value === 'string') {\n    return `\"${value}\"`;\n  } else {\n    return `${value}`;\n  }\n}\n\n/**\n * Returns a function `f2` (decorator) which wraps the original function\n * `f`. `f2` guarantees that `f` can be called at most once\n * every `waitMs` ms. If `f2` is called more often, it will return\n * the last returned result of `f`.\n *\n * @param f The original function `f` to wrap.\n * @param waitMs The time between two consecutive calls to `f` in ms.\n */\nexport function debounce<T>(\n    f: (...args: Array<{}>) => T, waitMs: number): (...args: Array<{}>) => T {\n  let lastTime = util.now();\n  let lastResult: T;\n  const f2 = (...args: Array<{}>) => {\n    const now = util.now();\n    if (now - lastTime < waitMs) {\n      return lastResult;\n    }\n    lastTime = now;\n    lastResult = f(...args);\n    return lastResult;\n  };\n  return f2;\n}\n\n/**\n * Returns the fusable activation given a layers identifier.\n *\n * @param activationName The layers identifier string.\n * @return The name of the fusable activation.\n */\nexport function mapActivationToFusedKernel(activationName: string):\n    fused.Activation {\n  if (activationName === 'relu') {\n    return 'relu';\n  }\n  if (activationName === 'linear') {\n    return 'linear';\n  }\n  if (activationName === 'elu') {\n    return 'elu';\n  }\n  return null;\n}\n\ntype PossibleValues = Array<Array<boolean|string|number>>;\n\n/**\n * Returns the cartesian product of sets of values.\n * This works the same as itertools.product in Python.\n *\n * Example:\n *\n * filters = [128, 256, 512]\n * paddings = ['same', 'valid']\n *\n * product = [ [128, 'same'], [128, 'valid'], [256, 'same'], [256, 'valid'],\n * [512, 'same'], [512, 'valid']]\n *\n * @param arrayOfValues List/array of values.\n * @return The cartesian product.\n */\nexport function getCartesianProductOfValues(...arrayOfValues: PossibleValues):\n    PossibleValues {\n  assert(arrayOfValues.length > 0, 'arrayOfValues is empty');\n\n  for (const values of arrayOfValues) {\n    assert(Array.isArray(values), 'one of the values is not an array');\n    assert(values.length > 0, 'one of the values is empty');\n  }\n\n  return arrayOfValues.reduce((products, values) => {\n    if (products.length === 0) {\n      return values.map(value => [value]);\n    }\n\n    return values\n        .map(value => {\n          return products.map((prevValue) => [...prevValue, value]);\n        })\n        .reduce((flattenedProduct, unflattenedProduct) => {\n          return flattenedProduct.concat(unflattenedProduct);\n        }, []);\n  }, [] as PossibleValues);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/contraints.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport {epsilon} from './backend/common';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\n/**\n * Helper function used by many of the Constraints to find the L2Norms.\n */\nfunction calcL2Norms(w: Tensor, axis: number): Tensor {\n  return tidy(() => tfc.sqrt(tfc.sum(tfc.mul(w, w), axis, true)));\n}\n\n/**\n * Base class for functions that impose constraints on weight values\n *\n * @doc {\n *   heading: 'Constraints',\n *   subheading: 'Classes',\n *   namespace: 'constraints'\n * }\n */\nexport abstract class Constraint extends serialization.Serializable {\n  /* Porting note: was __call__, apply chosen to match other similar choices */\n  abstract apply(w: Tensor): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport interface MaxNormArgs {\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   *\n   *  For instance, in a `Dense` layer the weight matrix\n   *  has shape `[inputDim, outputDim]`,\n   *  set `axis` to `0` to constrain each weight vector\n   *  of length `[inputDim,]`.\n   *  In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   *  the weight tensor has shape\n   *  `[rows, cols, inputDepth, outputDepth]`,\n   *  set `axis` to `[0, 1, 2]`\n   *  to constrain the weights of each filter tensor of size\n   *  `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class MaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MaxNorm';\n  private maxValue: number;\n  private axis: number;\n  private readonly defaultMaxValue = 2;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MaxNormArgs) {\n    super();\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.clipByValue(norms, 0, this.maxValue);\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {maxValue: this.maxValue, axis: this.axis};\n  }\n}\nserialization.registerClass(MaxNorm);\n\nexport interface UnitNormArgs {\n  /**\n   * Axis along which to calculate norms.\n   *\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * [rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class UnitNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'UnitNorm';\n  private axis: number;\n  private readonly defaultAxis = 0;\n  constructor(args: UnitNormArgs) {\n    super();\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(\n        () => tfc.div(w, tfc.add(epsilon(), calcL2Norms(w, this.axis))));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {axis: this.axis};\n  }\n}\nserialization.registerClass(UnitNorm);\n\nexport class NonNeg extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'NonNeg';\n\n  apply(w: Tensor): Tensor {\n    return tfc.relu(w);\n  }\n}\nserialization.registerClass(NonNeg);\n\nexport interface MinMaxNormArgs {\n  /**\n   * Minimum norm for incoming weights\n   */\n  minValue?: number;\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * `[rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n  /**\n   * Rate for enforcing the constraint: weights will be rescaled to yield:\n   * `(1 - rate) * norm + rate * norm.clip(minValue, maxValue)`.\n   * Effectively, this means that rate=1.0 stands for strict\n   * enforcement of the constraint, while rate<1.0 means that\n   * weights will be rescaled at each step to slowly move\n   * towards a value inside the desired interval.\n   */\n  rate?: number;\n}\n\nexport class MinMaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MinMaxNorm';\n  private minValue: number;\n  private maxValue: number;\n  private rate: number;\n  private axis: number;\n  private readonly defaultMinValue = 0.0;\n  private readonly defaultMaxValue = 1.0;\n  private readonly defaultRate = 1.0;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MinMaxNormArgs) {\n    super();\n    this.minValue =\n        args.minValue != null ? args.minValue : this.defaultMinValue;\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.rate = args.rate != null ? args.rate : this.defaultRate;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.add(\n          tfc.mul(\n              this.rate, tfc.clipByValue(norms, this.minValue, this.maxValue)),\n          tfc.mul(1.0 - this.rate, norms));\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      minValue: this.minValue,\n      maxValue: this.maxValue,\n      rate: this.rate,\n      axis: this.axis\n    };\n  }\n}\nserialization.registerClass(MinMaxNorm);\n\n/** @docinline */\nexport type ConstraintIdentifier =\n    'maxNorm'|'minMaxNorm'|'nonNeg'|'unitNorm'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in ConstraintIdentifier]: string} = {\n      'maxNorm': 'MaxNorm',\n      'minMaxNorm': 'MinMaxNorm',\n      'nonNeg': 'NonNeg',\n      'unitNorm': 'UnitNorm'\n    };\n\nexport function serializeConstraint(constraint: Constraint):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeConstraint(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Constraint {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'constraint');\n}\n\nexport function getConstraint(identifier: ConstraintIdentifier|\n                              serialization.ConfigDict|Constraint): Constraint {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeConstraint(config);\n  } else if (identifier instanceof Constraint) {\n    return identifier;\n  } else {\n    return deserializeConstraint(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constraint, MaxNorm, MaxNormArgs, MinMaxNorm, MinMaxNormArgs, NonNeg, UnitNorm, UnitNormArgs} from './constraints';\n\n/**\n * MaxNorm weight constraint.\n *\n * Constrains the weights incident to each hidden unit\n * to have a norm less than or equal to a desired value.\n *\n * References\n *       - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting\n * Srivastava, Hinton, et al.\n * 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Constraints',namespace: 'constraints'}\n */\nexport function maxNorm(args: MaxNormArgs): Constraint {\n  return new MaxNorm(args);\n}\n\n/**\n * Constrains the weights incident to each hidden unit to have unit norm.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function unitNorm(args: UnitNormArgs): Constraint {\n  return new UnitNorm(args);\n}\n\n/**\n * Constains the weight to be non-negative.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function nonNeg(): Constraint {\n  return new NonNeg();\n}\n\n/** @doc {heading: 'Constraints', namespace: 'constraints'} */\nexport function minMaxNorm(config: MinMaxNormArgs): Constraint {\n  return new MinMaxNorm(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// TODO(huan): add layer-specific input shape types (see: https://github.com/tensorflow/tfjs-layers/pull/492)\n/** @docalias (null | number)[] */\nexport type Shape = Array<null | number>;\n\n// The tfjs-core version of DataType must stay synced with this.\nexport type DataType = 'float32'|'int32'|'bool'|'complex64'|'string';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type DataFormat = 'channelsFirst'|'channelsLast';\nexport const VALID_DATA_FORMAT_VALUES = ['channelsFirst', 'channelsLast'];\n\n// These constants have a snake vs. camel distinction.\nexport type DataFormatSerialization = 'channels_first'|'channels_last';\n\n/** @docinline */\nexport type PaddingMode = 'valid'|'same'|'causal';\nexport const VALID_PADDING_MODE_VALUES = ['valid', 'same', 'causal'];\n\n/** @docinline */\nexport type PoolMode = 'max'|'avg';\nexport const VALID_POOL_MODE_VALUES = ['max', 'avg'];\n\n/** @docinline */\nexport type BidirectionalMergeMode = 'sum'|'mul'|'concat'|'ave';\nexport const VALID_BIDIRECTIONAL_MERGE_MODES = ['sum', 'mul', 'concat', 'ave'];\n\n/** @docinline */\nexport type SampleWeightMode = 'temporal';\nexport const VALID_SAMPLE_WEIGHT_MODES = ['temporal'];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Common functions for TensorFlow.js Layers.\n */\nimport {VALID_DATA_FORMAT_VALUES, VALID_PADDING_MODE_VALUES, VALID_POOL_MODE_VALUES} from './keras_format/common';\nimport {checkStringTypeUnionValue} from './utils/generic_utils';\n\n// A map from the requested scoped name of a Tensor to the number of Tensors\n// wanting that name so far.  This allows enforcing name uniqueness by appending\n// an incrementing index, e.g. scope/name, scope/name_1, scope/name_2, etc.\nconst nameMap: Map<string, number> = new Map<string, number>();\n\nexport function checkDataFormat(value?: string): void {\n  checkStringTypeUnionValue(VALID_DATA_FORMAT_VALUES, 'DataFormat', value);\n}\n\nexport function checkPaddingMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_PADDING_MODE_VALUES, 'PaddingMode', value);\n}\n\nexport function checkPoolMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_POOL_MODE_VALUES, 'PoolMode', value);\n}\n\nconst _nameScopeStack: string[] = [];\nconst _nameScopeDivider = '/';\n\n/**\n * Enter namescope, which can be nested.\n */\nexport function nameScope<T>(name: string, fn: () => T): T {\n  _nameScopeStack.push(name);\n  try {\n    const val: T = fn();\n    _nameScopeStack.pop();\n    return val;\n  } catch (e) {\n    _nameScopeStack.pop();\n    throw e;\n  }\n}\n\n/**\n * Get the current namescope as a flat, concatenated string.\n */\nfunction currentNameScopePrefix(): string {\n  if (_nameScopeStack.length === 0) {\n    return '';\n  } else {\n    return _nameScopeStack.join(_nameScopeDivider) + _nameScopeDivider;\n  }\n}\n\n/**\n * Get the name a Tensor (or Variable) would have if not uniqueified.\n * @param tensorName\n * @return Scoped name string.\n */\nexport function getScopedTensorName(tensorName: string): string {\n  if (!isValidTensorName(tensorName)) {\n    throw new Error('Not a valid tensor name: \\'' + tensorName + '\\'');\n  }\n  return currentNameScopePrefix() + tensorName;\n}\n\n/**\n * Get unique names for Tensors and Variables.\n * @param scopedName The fully-qualified name of the Tensor, i.e. as produced by\n *  `getScopedTensorName()`.\n * @return A unique version of the given fully scoped name.\n *   If this is the first time that the scoped name is seen in this session,\n *   then the given `scopedName` is returned unaltered.  If the same name is\n *   seen again (producing a collision), an incrementing suffix is added to the\n *   end of the name, so it takes the form 'scope/name_1', 'scope/name_2', etc.\n */\nexport function getUniqueTensorName(scopedName: string): string {\n  if (!isValidTensorName(scopedName)) {\n    throw new Error('Not a valid tensor name: \\'' + scopedName + '\\'');\n  }\n  if (!nameMap.has(scopedName)) {\n    nameMap.set(scopedName, 0);\n  }\n  const index = nameMap.get(scopedName);\n  nameMap.set(scopedName, nameMap.get(scopedName) + 1);\n\n  if (index > 0) {\n    const result = `${scopedName}_${index}`;\n    // Mark the composed name as used in case someone wants\n    // to call getUniqueTensorName(\"name_1\").\n    nameMap.set(result, 1);\n    return result;\n  } else {\n    return scopedName;\n  }\n}\n\nconst tensorNameRegex = new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\\._\\/]*$/);\n\n/**\n * Determine whether a string is a valid tensor name.\n * @param name\n * @returns A Boolean indicating whether `name` is a valid tensor name.\n */\nexport function isValidTensorName(name: string): boolean {\n  return !!name.match(tensorNameRegex);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Math utility functions.\n *\n * This file contains some frequently used math function that operates on\n * number[] or Float32Array and return a number. Many of these functions are\n * not-so-thick wrappers around TF.js Core functions. But they offer the\n * convenience of\n * 1) not having to convert the inputs into Tensors,\n * 2) not having to convert the returned Tensors to numbers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar, Tensor1D, tensor1d} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\n\nexport type ArrayTypes = Uint8Array|Int32Array|Float32Array;\n\n/**\n * Determine if a number is an integer.\n */\nexport function isInteger(x: number): boolean {\n  return x === parseInt(x.toString(), 10);\n}\n\n/**\n * Calculate the product of an array of numbers.\n * @param array The array to calculate the product over.\n * @param begin Beginning index, inclusive.\n * @param end Ending index, exclusive.\n * @return The product.\n */\nexport function arrayProd(\n    array: number[]|ArrayTypes, begin?: number, end?: number): number {\n  if (begin == null) {\n    begin = 0;\n  }\n  if (end == null) {\n    end = array.length;\n  }\n\n  let prod = 1;\n  for (let i = begin; i < end; ++i) {\n    prod *= array[i];\n  }\n  return prod;\n}\n\n/**\n * A helper function transforms the two input types to an instance of Tensor1D,\n * so the return value can be fed directly into various TF.js Core functions.\n * @param array\n */\nfunction toArray1D(array: number[]|Float32Array): Tensor1D {\n  array = Array.isArray(array) ? new Float32Array(array) : array;\n  return tensor1d(array);\n}\n\n/**\n * Compute minimum value.\n * @param array\n * @return minimum value.\n */\nexport function min(array: number[]|Float32Array): number {\n  return tfc.min(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute maximum value.\n * @param array\n * @return maximum value\n */\nexport function max(array: number[]|Float32Array): number {\n  return tfc.max(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute sum of array.\n * @param array\n * @return The sum.\n */\nexport function sum(array: number[]|Float32Array): number {\n  return tfc.sum(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute mean of array.\n * @param array\n * @return The mean.\n */\nexport function mean(array: number[]|Float32Array): number {\n  return sum(array) / array.length;\n}\n\n/**\n * Compute variance of array.\n * @param array\n * @return The variance.\n */\nexport function variance(array: number[]|Float32Array): number {\n  const demeaned = tfc.sub(toArray1D(array), scalar(mean(array)));\n  const sumSquare = tfc.sum(tfc.mul(demeaned, demeaned)).dataSync()[0];\n  return sumSquare / array.length;\n}\n\n/**\n * Compute median of array.\n * @param array\n * @return The median value.\n */\nexport function median(array: number[]|Float32Array): number {\n  const arraySorted = array.slice().sort((a, b) => a - b);\n  const lowIdx = Math.floor((arraySorted.length - 1) / 2);\n  const highIdx = Math.ceil((arraySorted.length - 1) / 2);\n  if (lowIdx === highIdx) {\n    return arraySorted[lowIdx];\n  }\n  return (arraySorted[lowIdx] + arraySorted[highIdx]) / 2;\n}\n\n/**\n * Generate an array of integers in [begin, end).\n * @param begin Beginning integer, inclusive.\n * @param end Ending integer, exclusive.\n * @returns Range array.\n * @throws ValueError, iff `end` < `begin`.\n */\nexport function range(begin: number, end: number): number[] {\n  if (end < begin) {\n    throw new ValueError(`end (${end}) < begin (${begin}) is forbidden.`);\n  }\n  const out: number[] = [];\n  for (let i = begin; i < end; ++i) {\n    out.push(i);\n  }\n  return out;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * deeplearn.js backend.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {onesLike as coreOnesLike, scalar, Tensor, Tensor1D, tensor1d, Tensor2D, Tensor3D, Tensor4D, Tensor5D, tidy, where, zerosLike as coreZerosLike} from '@tensorflow/tfjs-core';\nimport {checkDataFormat} from '../common';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {HasShape} from '../types';\nimport * as math_utils from '../utils/math_utils';\n\nimport {imageDataFormat} from './common';\n\n// tslint:enable\n\n/* Setting and getting backend from deeplearn.js. */\n\n// Default deeplearn.js backend is WebGL (GPU).\nlet backend: 'cpu'|'webgl' = 'webgl';\n\nexport function setBackend(requestedBackend: 'cpu'|'webgl') {\n  tfc.setBackend(requestedBackend);\n  backend = requestedBackend;\n}\n\nexport function getBackend(): 'cpu'|'webgl' {\n  return backend;\n}\n\n/**\n * Indicates whether the backend is operating symbolically.\n *\n * This function will be used to determine how to interpret user code. If\n * it returns true, calls to the backend construct a symbolic graph; if\n * it returns false, calls to the backend execute immediately.\n */\nexport function isBackendSymbolic(): boolean {\n  return false;\n}\n\n/**\n * Get the number of elements in a Tensor.\n * @param x The Tensor.\n * @return Number of elements in `x`.\n */\nexport function countParams(x: HasShape): number {\n  const shape = x.shape;\n  if (shape.length > 0) {\n    return shape.reduce((a: number, b: number) => a * b);\n  } else {\n    // Scalar.\n    return 1;\n  }\n}\n\n/**\n * Casts a tensor to a different dtype and returns it.\n * @param x Input tensor.\n * @param dtype String: 'float32'|'int32'|'bool'.\n * @returns Tensor of the specified `dtype`.\n */\nexport function cast(x: Tensor, dtype: tfc.DataType): Tensor {\n  return x.asType(dtype);\n}\n\n/**\n * Adds a 1-sized dimension at index \"axis\".\n * @param x Input tensor.\n * @param axis Position where to add the new axis.\n * @returns Result of the dimension expansion.\n */\nexport function expandDims(x: Tensor, axis = -1): Tensor {\n  const outShape = x.shape.slice();\n  if (axis < 0) {\n    axis = outShape.length + axis + 1;\n  }\n  outShape.splice(axis, 0, 1);\n  return x.reshape(outShape);\n}\n\n/**\n * Repeats a 2D tensor.\n *\n * If `x` has shape `[samples, dim]` and `n` is 2, for example, the output\n * will have shape `[samples, 2, dim]`.\n *\n * @param x Input tensor.\n * @param n Integer, number of times to repeat.\n * @returns The result of the repeat operation.\n * @throws ValueError: If input tensor is not 2D.\n */\nexport function repeat(x: Tensor, n: number): Tensor {\n  return tidy(() => {\n    if (x.shape.length !== 2) {\n      throw new ValueError(\n          `repeat() expects a rank-2 tensor, but received a ` +\n          `rank-${x.shape.length} tensor.`);\n    }\n    const y = expandDims(x, 1);\n    return tile(y, [1, n, 1]);\n  });\n}\n\n/**\n * Flatten a Tensor into 1D.\n * @param x Input tensor.\n * @return The result of the flattening `x`.\n */\nexport function flatten(x: Tensor): Tensor {\n  const newShape = [math_utils.arrayProd(x.shape)];\n  return x.reshape(newShape);\n}\n\n/**\n * Turn a nD tensor into a 2D tensor with same 0th dimension.\n * In other words, it flattens each data samples of a batch.\n *\n * @param x The tensor to flatten. The rank of this tensor is required to be 2\n *   or higher.\n * @return The result of the flattening.\n */\nexport function batchFlatten(x: Tensor): Tensor {\n  if (x.rank <= 1) {\n    throw new ValueError(\n        `batchFlatten requires a minimum rank of 2. Got rank: ${x.rank}.`);\n  }\n  const newShape = [x.shape[0], math_utils.arrayProd(x.shape, 1)];\n  return x.reshape(newShape);\n}\n\n/**\n * Do slicing along the first axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the first axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongFirstAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [start, 0], [size, array.shape[1]]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [start, 0, 0],\n            [size, array.shape[1], array.shape[2]]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [start, 0, 0, 0],\n            [size, array.shape[1], array.shape[2], array.shape[3]]);\n      case 5:\n        return tfc.slice(array as Tensor5D, [start, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4]\n        ]);\n      case 6:\n        return tfc.slice(array, [start, 0, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4],\n          array.shape[5]\n        ]);\n      default:\n        throw new ValueError(\n            `sliceAlongFirstAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the last axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the last axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongLastAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [0, start], [array.shape[0], size]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [0, 0, start],\n            [array.shape[0], array.shape[1], size]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [0, 0, 0, start],\n            [array.shape[0], array.shape[1], array.shape[2], size]);\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the sepcified axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size of the slice along the chosen axis.\n * @param choose an axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongAxis(\n    array: Tensor, start: number, size: number, axis: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 3:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice3d(\n                array as Tensor3D, [0, start, 0],\n                [array.shape[0], size, array.shape[2]]);\n          case 3:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 4:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice4d(\n                array as Tensor4D, [0, start, 0, 0],\n                [array.shape[0], size, array.shape[2], array.shape[3]]);\n          case 3:\n            return tfc.slice4d(\n                array as Tensor4D, [0, 0, start, 0],\n                [array.shape[0], array.shape[1], size, array.shape[3]]);\n          case 4:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Concatenates a list of tensors alongside the specified axis.\n * @param tensors `Array` of tensors to concatenate.\n * @param axis Concatenation axis.\n * @returns The result of the concatenation.\n */\nexport function concatenate(tensors: Tensor[], axis = -1): Tensor {\n  let rank: number;\n  if (axis < 0) {\n    rank = tensors[0].rank;\n    if (rank !== 0) {\n      axis = rank;\n    } else {\n      axis = 0;\n    }\n  }\n  if (axis === tensors[0].rank) {\n    // Porting Note: This is necessary because tfc.concat() requires axis to be\n    //   in the interval [-rank, rank).\n    axis = -1;\n  }\n  // Porting Note: Sparse concat is not supported yet.\n  return tfc.concat(tensors, axis);\n}\n\n/**\n * Concatenate two arrays along the first dimension.\n * @param a The 1st `tf.Tensor` to concatenate.\n * @param b The 2nd `tf.Tensor` to concatenate.\n * @returns Result of the concatenation.\n * @throws ValueError: If `a` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function concatAlongFirstAxis(a: Tensor, b: Tensor): Tensor {\n  switch (a.rank) {\n    case 1:\n      return tfc.concat1d([a as Tensor1D, b as Tensor1D]);\n    case 2:\n      return tfc.concat2d([a as Tensor2D, b as Tensor2D], 0);\n    case 3:\n      return tfc.concat3d([a as Tensor3D, b as Tensor3D], 0);\n    case 4:\n      return tfc.concat4d([a as Tensor4D, b as Tensor4D], 0);\n    default:\n      throw new ValueError(\n          `concatAlongFirstAxis() received an unsupported ` +\n          `tensor rank: ${a.rank}`);\n  }\n}\n\n/**\n * Creates a tensor by tiling `x` by `n`.\n * @param x A tensor.\n * @param n An Array of integers or a single integer. If an Array, the length\n *   must be the same as the number of dimensions in `x`. If a single integer,\n *   it will be treated as an Array of length 1.\n */\nexport function tile(x: Tensor, n: number|number[]): Tensor {\n  if (!Array.isArray(n)) {\n    n = [n];\n  }\n  if (x.rank !== n.length) {\n    throw new ValueError(\n        `The length of input n (${n.length}) does not match ` +\n        `the number of dimensions in input x (${x.rank})`);\n  }\n  return tfc.tile(x, n);\n}\n\n/* Creation of random tensors. */\n\n/**\n * Get a tensor with normal distribution of values.\n *\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @return The normal tensor.\n */\nexport function randomNormal(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: 'float32'|'int32',\n    seed?: number): Tensor {\n  return tfc.randomNormal(shape, mean, stddev, dtype, seed);\n}\n\n/* Linear Algebra */\n\n/**\n * Multiply two tensors and returns the result as a tensor.\n *\n * For 2D tensors, this is equivalent to matrix multiplication (matMul).\n * For tensors of higher ranks, it follows the Theano behavior,\n * (e.g. `(2, 3) * (4, 3, 5) -> (2, 4, 5)`).  From the Theano documentation:\n *\n * For N dimensions it is a sum product over the last axis of x and the\n * second-to-last of y:\n *\n * @param a A tensor of at least rank 2.\n * @param b A tensor of at least rank 2.\n * @param activation (optional) A string identifying the activation\n *   function.\n * @return Result of the dot operation.\n */\nexport function dot(\n    a: Tensor, b: Tensor, activation?: tfc.fused.Activation,\n    bias?: Tensor): Tensor {\n  if ((a.rank < 2) || (b.rank < 2)) {\n    throw new NotImplementedError(\n        `dot requires both inputs to be rank >= 2` +\n        ` but got x shape = ${a.shape} and y shape = ${b.shape}`);\n  }\n  if (b.rank >= 3) {\n    const xLastDim = a.shape.slice(-1)[0];\n    const ySecondLastDim = b.shape.slice(-2)[0];\n    if (xLastDim !== ySecondLastDim) {\n      throw new NotImplementedError(\n          `If rank y >= 3, then the second last dim` +\n          ` of y must equal the last dim of x but got x shape = ${\n              a.shape} and ` +\n          ` y shape = ${b.shape}`);\n    }\n  }\n  // Handle basic 2D x 2D case.\n  if ((a.rank === 2) && (b.rank === 2)) {\n    const transposeA = false;\n    const transposeB = false;\n    // tfc.fused.matMul only fuses certain activation functions. Unsupported\n    // activation functions are treated as 'linear' activations, which is\n    // equivalent to a no-op.\n    return tfc.fused.matMul({\n      a,\n      b: b as Tensor2D,\n      transposeA,\n      transposeB,\n      bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n      activation\n    });\n  } else {\n    // Reshape x into the analogous 2D Tensor.\n    const aFirstDims = a.shape.slice();  // Holds all but the last dim of x.\n    const aLastDim = aFirstDims.pop();\n    a = a.reshape([-1, aLastDim]);\n\n    // Reshape y into the analogous 2D Tensor, and keep track of the\n    // required dimensions to reproduce the output shape.\n    const bShape = b.shape.slice();\n    const bLastDim = bShape.pop();\n    const ySecondLastDim = bShape.pop();\n    const yOtherDims = [...bShape, bLastDim];\n    // permutation should be like [r-2, 0, 1, 2, ... r-4, r-3, r-1]\n    // where r is the rank of y.\n    const perm = Array.from({length: b.rank}, (_, i) => {\n      if (i === 0) {\n        return b.rank - 2;\n      } else if (i <= b.rank - 2) {\n        return i - 1;\n      }\n      return i;\n    });\n    b = b.transpose(perm).reshape([ySecondLastDim, -1]);\n\n    // Multiply x and y as 2D Tensors, and then reshape back to original.\n    const outputShape = [...aFirstDims, ...yOtherDims];\n    const transposeA = false;\n    const transposeB = false;\n    return tfc.fused\n        .matMul({\n          a,\n          b,\n          transposeA,\n          transposeB,\n          bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n          activation\n        })\n        .reshape(outputShape);\n  }\n}\n\n/**\n * Compute the sign Tensor of an input Tensor.\n *\n * Elements of the input `tf.Tensor` that are === 0 are mapped to 0.\n * Elements of the input `tf.Tensor` that are > 0 are mapped to 1.\n * Elements of the input `tf.Tensor` that are < 0 are mapped to -1.\n *\n * @param x Input `tf.Tensor`.\n * @return The sign `tf.Tensor`.\n */\nexport function sign(x: Tensor): Tensor {\n  // TODO(cais): Move to the core.\n  return tidy(() => {\n    const zerosLikeX = coreZerosLike(x);\n    const onesLikeX = coreOnesLike(x);\n    return where(\n        tfc.equal(x, zerosLikeX), zerosLikeX,\n        where(\n            tfc.greater(x, coreZerosLike(x)), onesLikeX,\n            tfc.mul(-1, onesLikeX)));\n  });\n}\n\n/**\n * Computes the one-hot representation of an integer tensor.\n * @param indices nD integer tensor of shape\n *   `(batch_size, dim1, dim2, ... dim(n-1))`\n * @param numClasses Integer, number of classes to consider.\n * @returns (n + 1)D one hot representation of the input\n *   with shape `(batch_size, dim1, dim2, ... dim(n-1), num_classes)`\n */\nexport function oneHot(indices: Tensor, numClasses: number): Tensor {\n  return tidy(() => {\n    if (indices.rank !== 1) {\n      throw new Error(\n          'Only 1D one-hot tensors are supported in the ' +\n          'deeplearn backend, at present.');\n    }\n    indices = indices.toInt();\n    return tfc.oneHot(indices as Tensor1D, numClasses).toFloat();\n  });\n}\n\n/* Elementary math functions. */\n\n/**\n * Retrieves the elements of indices `indices` in the tensor `reference`.\n * @param reference A tensor.\n * @param indices An integer tensor of indices or an `Array` of integers.\n * @param axis Axis along which to perform the gather operation.\n * @returns The result of the gathering as a tensor.\n */\nexport function gather(\n    reference: Tensor, indices: number[]|Tensor1D, axis?: number): Tensor {\n  return tidy(() => {\n    if (Array.isArray(indices)) {\n      indices = tensor1d(indices, 'int32');\n    } else {\n      indices = indices.toInt();\n    }\n    return tfc.gather(reference, indices, axis);\n  });\n}\n\n/**\n * Element-wise square.\n * @param x Input tensor.\n * @return element-wise x^2\n */\nexport function square(x: Tensor): Tensor {\n  return tfc.mul(x, x);\n}\n\n/**\n * Element-wise exponentiation.\n *\n * Porting Note: In PyKeras, `a` (the exponent) is a Python integer, which\n *   takes advatnage of the backend's (e.g., TensorFlow's) automatic\n * conversion to tensor. Here we allow `a` to be either a number or a tensor.\n *\n * @param x The base tensor.\n * @param a The exponent, tensor or number. If a number, it is rounded to the\n *   nearest integer and converted to a tensor.\n * @returns A tensor of the same shape as `x`.\n */\nexport function pow(x: Tensor, a: Tensor|number): Tensor {\n  return tidy(() => {\n    if (typeof (a) === 'number') {\n      a = scalar(Math.round(a), 'int32');\n    }\n    if (a.dtype !== 'int32') {\n      throw new NotImplementedError(\n          `Non-int32 dtype (${a.dtype}) is not supported by pow() yet`);\n    }\n    return tfc.pow(x, a);\n  });\n}\n\n/**\n * Reshapes bias tensor according to rank of x.\n */\nfunction reshapeBias(xRank: number, bias: Tensor, dataFormat: string) {\n  const biasShape = bias.shape;\n\n  if (bias.rank !== 1 && bias.rank !== xRank) {\n    throw new ValueError(\n        `Unexpected bias dimensions: ${bias.rank}` +\n        `; expected it to be 1 or ${xRank}`);\n  }\n\n  if (xRank === 5) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1, 1]);\n      } else {\n        return bias.reshape(\n            [1, biasShape[3], biasShape[0], biasShape[1], biasShape[2]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 4) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1]);\n      } else {\n        return bias.reshape([1, biasShape[2], biasShape[0], biasShape[1]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 3) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1]);\n      } else {\n        return bias.reshape([1, biasShape[1], biasShape[0]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank < 3) {\n    return bias;\n  }\n  throw new ValueError(`Unsupported input rank by biasAdd: ${bias.rank}`);\n}\n\n/* Neural-network operations. */\n\n/**\n * Add a bias to a tensor.\n *\n * @param x The tensor to add the bias to.\n * @param bias The bias to add to `x`. Must be 1D or the same rank as `x`.\n * @return Result of the bias adding.\n * @throws ValueError: If the rank of `bias` is incorrect.\n */\nexport function biasAdd(\n    x: Tensor, bias: Tensor, dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n\n    return x.add(reshapeBias(x.rank, bias, dataFormat));\n  });\n}\n\n/**\n * Exponential linear unit (ELU).\n * @param x A tensor or variable to compute the activation function for.\n * @param alpha: A scalar, a scaling factor for the negative section.\n * @return Output of the ELU operation.\n */\nexport function elu(x: Tensor, alpha = 1): Tensor {\n  // TODO(cais): Add support for alpha values other than 1.\n  if (alpha !== 1) {\n    throw new NotImplementedError(\n        `Support for alpha values other than 1 (${alpha}) is not implemented ` +\n        `yet.`);\n  }\n  return tfc.elu(x);\n}\n\n/**\n * Softsign of a tensor.\n *\n * Defined as x / (abs(x) + 1), element-wise.\n *\n * @param x: Input.\n * @returns Output.\n */\nexport function softsign(x: Tensor): Tensor {\n  return tidy(() => tfc.div(x, tfc.abs(x).add(1)));\n}\n\n/**\n * Sets entries in `x` to zero at random, while scaling the entire tensor.\n *\n * @param x input tensor.\n * @param level fraction of the entries in the tensor that will be set to 0.\n * @param noiseShape shape of randomly generated keep/drop flags, must be\n *   broadcastable to the shape of `x`. Optional.\n * @param seed random seed to ensure determinism. Optional.\n * @returns Result of the dropout operation.\n */\nexport function dropout(\n    x: Tensor, level: number, noiseShape?: number[], seed?: number): Tensor {\n  return tidy(() => tfc.dropout(x, level, noiseShape, seed));\n}\n\n/**\n * Element-wise, segment-wise linear approximation of sigmoid.\n *\n * Returns `0.` if `x < -2.5`, `1.` if `x > 2.5`.\n * In `-2.5 <= x <= 2.5`, returns `0.2 * x + 0.5`.\n *\n * @param x Input tensor.\n * @returns Output tensor.\n */\nexport function hardSigmoid(x: Tensor): Tensor {\n  return tidy(() => {\n    const y = tfc.add(.5, tfc.mul(.2, x));\n    return tfc.clipByValue(y, 0, 1);\n  });\n}\n\n/**\n * Invoke `x` in the training phase, and `alt` otherwise.\n *\n * Porting Note: We do not create placeholder tensors for the `training`\n * boolean flag here, because there is no such thing in the TF.js imperative\n * backend.\n *\n * @param x The function to invoke iff `training` is `true`.\n * @param alt The function to invoke iff `training` is `false`.\n * @param training Boolean flag for whether training phase is active.\n * @returns The return value of `x()` if `training` is `true`, or the return\n *   value of `alt()` if `training` is `false`.\n */\nexport function inTrainPhase<T>(x: () => T, alt: () => T, training = false): T {\n  return training ? x() : alt();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {BaseSerialization} from './types';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type FanMode = 'fanIn'|'fanOut'|'fanAvg';\nexport const VALID_FAN_MODE_VALUES = ['fanIn', 'fanOut', 'fanAvg'];\n\n// These constants have a snake vs. camel distinction.\nexport type FanModeSerialization = 'fan_in'|'fan_out'|'fan_avg';\n\n/** @docinline */\nexport type Distribution = 'normal'|'uniform'|'truncatedNormal';\nexport const VALID_DISTRIBUTION_VALUES =\n    ['normal', 'uniform', 'truncatedNormal'];\n// These constants have a snake vs. camel distinction.\nexport type DistributionSerialization = 'normal'|'uniform'|'truncated_normal';\n\nexport type ZerosSerialization = BaseSerialization<'Zeros', {}>;\n\nexport type OnesSerialization = BaseSerialization<'Ones', {}>;\n\nexport type ConstantConfig = {\n  value: number;\n};\n\nexport type ConstantSerialization =\n    BaseSerialization<'Constant', ConstantConfig>;\n\nexport type RandomNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type RandomNormalSerialization =\n    BaseSerialization<'RandomNormal', RandomNormalConfig>;\n\nexport type RandomUniformConfig = {\n  minval?: number;\n  maxval?: number;\n  seed?: number;\n};\n\nexport type RandomUniformSerialization =\n    BaseSerialization<'RandomUniform', RandomUniformConfig>;\n\nexport type TruncatedNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type TruncatedNormalSerialization =\n    BaseSerialization<'TruncatedNormal', TruncatedNormalConfig>;\n\nexport type VarianceScalingConfig = {\n  scale?: number;\n\n  mode?: FanModeSerialization;\n  distribution?: DistributionSerialization;\n  seed?: number;\n};\n\nexport type VarianceScalingSerialization =\n    BaseSerialization<'VarianceScaling', VarianceScalingConfig>;\n\nexport type OrthogonalConfig = {\n  seed?: number;\n  gain?: number;\n};\n\nexport type OrthogonalSerialization =\n    BaseSerialization<'Orthogonal', OrthogonalConfig>;\n\nexport type IdentityConfig = {\n  gain?: number;\n};\n\nexport type IdentitySerialization =\n    BaseSerialization<'Identity', IdentityConfig>;\n\n// Update initializerClassNames below in concert with this.\nexport type InitializerSerialization = ZerosSerialization|OnesSerialization|\n    ConstantSerialization|RandomUniformSerialization|RandomNormalSerialization|\n    TruncatedNormalSerialization|IdentitySerialization|\n    VarianceScalingSerialization|OrthogonalSerialization;\n\nexport type InitializerClassName = InitializerSerialization['class_name'];\n\n// We can't easily extract a string[] from the string union type, but we can\n// recapitulate the list, enforcing at compile time that the values are valid\n// and that we have the right number of them.\n\n/**\n * A string array of valid Initializer class names.\n *\n * This is guaranteed to match the `InitializerClassName` union type.\n */\nexport const initializerClassNames: InitializerClassName[] = [\n  'Zeros', 'Ones', 'Constant', 'RandomNormal', 'RandomUniform',\n  'TruncatedNormal', 'VarianceScaling', 'Orthogonal', 'Identity'\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, eye, linalg, mul, ones, randomUniform, scalar, serialization, Tensor, Tensor2D, tidy, truncatedNormal, zeros} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {checkDataFormat} from './common';\nimport {NotImplementedError, ValueError} from './errors';\nimport {DataFormat, Shape} from './keras_format/common';\nimport {Distribution, FanMode, VALID_DISTRIBUTION_VALUES, VALID_FAN_MODE_VALUES} from './keras_format/initializer_config';\nimport {checkStringTypeUnionValue, deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\nimport {arrayProd} from './utils/math_utils';\n\nexport function checkFanMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_FAN_MODE_VALUES, 'FanMode', value);\n}\n\nexport function checkDistribution(value?: string): void {\n  checkStringTypeUnionValue(VALID_DISTRIBUTION_VALUES, 'Distribution', value);\n}\n\n/**\n * Initializer base class.\n *\n * @doc {\n *   heading: 'Initializers', subheading: 'Classes', namespace: 'initializers'}\n */\nexport abstract class Initializer extends serialization.Serializable {\n  public fromConfigUsesCustomObjects(): boolean {\n    return false;\n  }\n  /**\n   * Generate an initial value.\n   * @param shape\n   * @param dtype\n   * @return The init value.\n   */\n  abstract apply(shape: Shape, dtype?: DataType): Tensor;\n\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport class Zeros extends Initializer {\n  /** @nocollapse */\n  static className = 'Zeros';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return zeros(shape, dtype);\n  }\n}\nserialization.registerClass(Zeros);\n\nexport class Ones extends Initializer {\n  /** @nocollapse */\n  static className = 'Ones';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return ones(shape, dtype);\n  }\n}\nserialization.registerClass(Ones);\n\nexport interface ConstantArgs {\n  /** The value for each element in the variable. */\n  value: number;\n}\n\nexport class Constant extends Initializer {\n  /** @nocollapse */\n  static className = 'Constant';\n  private value: number;\n  constructor(args: ConstantArgs) {\n    super();\n    if (typeof args !== 'object') {\n      throw new ValueError(\n          `Expected argument of type ConstantConfig but got ${args}`);\n    }\n    if (args.value === undefined) {\n      throw new ValueError(`config must have value set but got ${args}`);\n    }\n    this.value = args.value;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => mul(scalar(this.value), ones(shape, dtype)));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      value: this.value,\n    };\n  }\n}\nserialization.registerClass(Constant);\n\nexport interface RandomUniformArgs {\n  /** Lower bound of the range of random values to generate. */\n  minval?: number;\n  /** Upper bound of the range of random values to generate. */\n  maxval?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomUniform extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomUniform';\n  readonly DEFAULT_MINVAL = -0.05;\n  readonly DEFAULT_MAXVAL = 0.05;\n  private minval: number;\n  private maxval: number;\n  private seed: number;\n\n  constructor(args: RandomUniformArgs) {\n    super();\n    this.minval = args.minval || this.DEFAULT_MINVAL;\n    this.maxval = args.maxval || this.DEFAULT_MAXVAL;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return randomUniform(shape, this.minval, this.maxval, dtype);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {minval: this.minval, maxval: this.maxval, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomUniform);\n\nexport interface RandomNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomNormal';\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: RandomNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `randomNormal does not support dType ${dtype}.`);\n    }\n\n    return K.randomNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomNormal);\n\nexport interface TruncatedNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class TruncatedNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'TruncatedNormal';\n\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: TruncatedNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `truncatedNormal does not support dType ${dtype}.`);\n    }\n    return truncatedNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(TruncatedNormal);\n\nexport interface IdentityArgs {\n  /**\n   * Multiplicative factor to apply to the identity matrix.\n   */\n  gain?: number;\n}\n\nexport class Identity extends Initializer {\n  /** @nocollapse */\n  static className = 'Identity';\n  private gain: number;\n  constructor(args: IdentityArgs) {\n    super();\n    this.gain = args.gain != null ? args.gain : 1.0;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length !== 2 || shape[0] !== shape[1]) {\n        throw new ValueError(\n            'Identity matrix initializer can only be used for' +\n            ' 2D square matrices.');\n      } else {\n        return mul(this.gain, eye(shape[0]));\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {gain: this.gain};\n  }\n}\nserialization.registerClass(Identity);\n\n/**\n * Computes the number of input and output units for a weight shape.\n * @param shape Shape of weight.\n * @param dataFormat data format to use for convolution kernels.\n *   Note that all kernels in Keras are standardized on the\n *   CHANNEL_LAST ordering (even when inputs are set to CHANNEL_FIRST).\n * @return An length-2 array: fanIn, fanOut.\n */\nfunction computeFans(\n    shape: Shape, dataFormat: DataFormat = 'channelsLast'): number[] {\n  let fanIn: number;\n  let fanOut: number;\n  checkDataFormat(dataFormat);\n  if (shape.length === 2) {\n    fanIn = shape[0];\n    fanOut = shape[1];\n  } else if ([3, 4, 5].indexOf(shape.length) !== -1) {\n    if (dataFormat === 'channelsFirst') {\n      const receptiveFieldSize = arrayProd(shape, 2);\n      fanIn = shape[1] * receptiveFieldSize;\n      fanOut = shape[0] * receptiveFieldSize;\n    } else if (dataFormat === 'channelsLast') {\n      const receptiveFieldSize = arrayProd(shape, 0, shape.length - 2);\n      fanIn = shape[shape.length - 2] * receptiveFieldSize;\n      fanOut = shape[shape.length - 1] * receptiveFieldSize;\n    }\n  } else {\n    const shapeProd = arrayProd(shape);\n    fanIn = Math.sqrt(shapeProd);\n    fanOut = Math.sqrt(shapeProd);\n  }\n\n  return [fanIn, fanOut];\n}\n\nexport interface VarianceScalingArgs {\n  /** Scaling factor (positive float). */\n  scale?: number;\n\n  /** Fanning mode for inputs and outputs. */\n  mode?: FanMode;\n\n  /** Probabilistic distribution of the values. */\n  distribution?: Distribution;\n\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class VarianceScaling extends Initializer {\n  /** @nocollapse */\n  static className = 'VarianceScaling';\n  private scale: number;\n  private mode: FanMode;\n  private distribution: Distribution;\n  private seed: number;\n\n  /**\n   * Constructor of VarianceScaling.\n   * @throws ValueError for invalid value in scale.\n   */\n  constructor(args: VarianceScalingArgs) {\n    super();\n    if (args.scale < 0.0) {\n      throw new ValueError(\n          `scale must be a positive float. Got: ${args.scale}`);\n    }\n    this.scale = args.scale == null ? 1.0 : args.scale;\n    this.mode = args.mode == null ? 'fanIn' : args.mode;\n    checkFanMode(this.mode);\n    this.distribution =\n        args.distribution == null ? 'normal' : args.distribution;\n    checkDistribution(this.distribution);\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    const fans = computeFans(shape);\n    const fanIn = fans[0];\n    const fanOut = fans[1];\n    let scale = this.scale;\n    if (this.mode === 'fanIn') {\n      scale /= Math.max(1, fanIn);\n    } else if (this.mode === 'fanOut') {\n      scale /= Math.max(1, fanOut);\n    } else {\n      scale /= Math.max(1, (fanIn + fanOut) / 2);\n    }\n\n    if (this.distribution === 'normal') {\n      const stddev = Math.sqrt(scale);\n      dtype = dtype || 'float32';\n      if (dtype !== 'float32' && dtype !== 'int32') {\n        throw new NotImplementedError(\n            `${this.getClassName()} does not support dType ${dtype}.`);\n      }\n      return truncatedNormal(shape, 0, stddev, dtype, this.seed);\n    } else {\n      const limit = Math.sqrt(3 * scale);\n      return randomUniform(shape, -limit, limit, dtype);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      scale: this.scale,\n      mode: this.mode,\n      distribution: this.distribution,\n      seed: this.seed\n    };\n  }\n}\nserialization.registerClass(VarianceScaling);\n\nexport interface SeedOnlyInitializerArgs {\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class GlorotUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotUniform';\n\n  /**\n   * Constructor of GlorotUniform\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotUniform);\n\nexport class GlorotNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotNormal';\n\n  /**\n   * Constructor of GlorotNormal.\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotNormal);\n\nexport class HeNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeNormal);\n\nexport class HeUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeUniform';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeUniform);\n\nexport class LeCunNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunNormal);\n\nexport class LeCunUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunUniform);\n\nexport interface OrthogonalArgs extends SeedOnlyInitializerArgs {\n  /**\n   * Multiplicative factor to apply to the orthogonal matrix. Defaults to 1.\n   */\n  gain?: number;\n}\n\nexport class Orthogonal extends Initializer {\n  /** @nocollapse */\n  static className = 'Orthogonal';\n  readonly DEFAULT_GAIN = 1;\n  protected readonly gain: number;\n  protected readonly seed: number;\n\n  constructor(args?: OrthogonalArgs) {\n    super();\n    this.gain = args.gain == null ? this.DEFAULT_GAIN : args.gain;\n    this.seed = args.seed;\n\n    if (this.seed != null) {\n      throw new NotImplementedError(\n          'Random seed is not implemented for Orthogonal Initializer yet.');\n    }\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length < 2) {\n        throw new NotImplementedError('Shape must be at least 2D.');\n      }\n      if (shape[0] * shape[1] > 2000) {\n        console.warn(\n            `Orthogonal initializer is being called on a matrix with more ` +\n            `than 2000 (${shape[0] * shape[1]}) elements: ` +\n            `Slowness may result.`);\n      }\n\n      // TODO(cais): Add seed support.\n      const normalizedShape =\n          shape[0] > shape[1] ? [shape[1], shape[0]] : shape;\n      const a = K.randomNormal(normalizedShape, 0, 1, 'float32') as Tensor2D;\n      let q = linalg.gramSchmidt(a) as Tensor2D;\n      if (shape[0] > shape[1]) {\n        q = q.transpose();\n      }\n      return mul(this.gain, q);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      gain: this.gain,\n      seed: this.seed,\n    };\n  }\n}\nserialization.registerClass(Orthogonal);\n\n/** @docinline */\nexport type InitializerIdentifier =\n    'constant'|'glorotNormal'|'glorotUniform'|'heNormal'|'heUniform'|'identity'|\n    'leCunNormal'|'leCunUniform'|'ones'|'orthogonal'|'randomNormal'|\n    'randomUniform'|'truncatedNormal'|'varianceScaling'|'zeros'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in InitializerIdentifier]: string} = {\n      'constant': 'Constant',\n      'glorotNormal': 'GlorotNormal',\n      'glorotUniform': 'GlorotUniform',\n      'heNormal': 'HeNormal',\n      'heUniform': 'HeUniform',\n      'identity': 'Identity',\n      'leCunNormal': 'LeCunNormal',\n      'leCunUniform': 'LeCunUniform',\n      'ones': 'Ones',\n      'orthogonal': 'Orthogonal',\n      'randomNormal': 'RandomNormal',\n      'randomUniform': 'RandomUniform',\n      'truncatedNormal': 'TruncatedNormal',\n      'varianceScaling': 'VarianceScaling',\n      'zeros': 'Zeros'\n    };\n\nfunction deserializeInitializer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Initializer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'initializer');\n}\n\nexport function serializeInitializer(initializer: Initializer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(initializer);\n}\n\nexport function getInitializer(identifier: InitializerIdentifier|Initializer|\n                               serialization.ConfigDict): Initializer {\n  if (typeof identifier === 'string') {\n    const className = identifier in INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    /* We have four 'helper' classes for common initializers that\n    all get serialized as 'VarianceScaling' and shouldn't go through\n    the deserializeInitializer pathway. */\n    if (className === 'GlorotNormal') {\n      return new GlorotNormal();\n    } else if (className === 'GlorotUniform') {\n      return new GlorotUniform();\n    } else if (className === 'HeNormal') {\n      return new HeNormal();\n    } else if (className === 'HeUniform') {\n      return new HeUniform();\n    } else if (className === 'LeCunNormal') {\n      return new LeCunNormal();\n    } else if (className === 'LeCunUniform') {\n      return new LeCunUniform();\n    } else {\n      const config: serialization.ConfigDict = {};\n      config['className'] = className;\n      config['config'] = {};\n      return deserializeInitializer(config);\n    }\n  } else if (identifier instanceof Initializer) {\n    return identifier;\n  } else {\n    return deserializeInitializer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constant, ConstantArgs, GlorotNormal, GlorotUniform, HeNormal, HeUniform, Identity, IdentityArgs, Initializer, LeCunNormal, LeCunUniform, Ones, Orthogonal, OrthogonalArgs, RandomNormal, RandomNormalArgs, RandomUniform, RandomUniformArgs, SeedOnlyInitializerArgs, TruncatedNormal, TruncatedNormalArgs, VarianceScaling, VarianceScalingArgs, Zeros} from './initializers';\n\n/**\n * Initializer that generates tensors initialized to 0.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function zeros(): Zeros {\n  return new Zeros();\n}\n\n/**\n * Initializer that generates tensors initialized to 1.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function ones(): Initializer {\n  return new Ones();\n}\n\n/**\n * Initializer that generates values initialized to some constant.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function constant(args: ConstantArgs): Initializer {\n  return new Constant(args);\n}\n\n/**\n * Initializer that generates random values initialized to a uniform\n * distribution.\n *\n * Values will be distributed uniformly between the configured minval and\n * maxval.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomUniform(args: RandomUniformArgs): Initializer {\n  return new RandomUniform(args);\n}\n\n/**\n * Initializer that generates random values initialized to a normal\n * distribution.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomNormal(args: RandomNormalArgs): Initializer {\n  return new RandomNormal(args);\n}\n\n/**\n * Initializer that generates random values initialized to a truncated normal.\n * distribution.\n *\n * These values are similar to values from a `RandomNormal` except that values\n * more than two standard deviations from the mean are discarded and re-drawn.\n * This is the recommended initializer for neural network weights and filters.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function truncatedNormal(args: TruncatedNormalArgs): Initializer {\n  return new TruncatedNormal(args);\n}\n\n/**\n * Initializer that generates the identity matrix.\n * Only use for square 2D matrices.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function identity(args: IdentityArgs): Initializer {\n  return new Identity(args);\n}\n\n/**\n * Initializer capable of adapting its scale to the shape of weights.\n * With distribution=NORMAL, samples are drawn from a truncated normal\n * distribution centered on zero, with `stddev = sqrt(scale / n)` where n is:\n *   - number of input units in the weight tensor, if mode = FAN_IN.\n *   - number of output units, if mode = FAN_OUT.\n *   - average of the numbers of input and output units, if mode = FAN_AVG.\n * With distribution=UNIFORM,\n * samples are drawn from a uniform distribution\n * within [-limit, limit], with `limit = sqrt(3 * scale / n)`.\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function varianceScaling(config: VarianceScalingArgs): Initializer {\n  return new VarianceScaling(config);\n}\n\n/**\n * Glorot uniform initializer, also called Xavier uniform initializer.\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotUniform(args);\n}\n\n/**\n * Glorot normal initializer, also called Xavier normal initializer.\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor.\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotNormal(args);\n}\n\n/**\n * He normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function heNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeNormal(args);\n}\n\n/**\n * He uniform initializer.\n *\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / fan_in)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function heUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeUniform(args);\n}\n\n/**\n * LeCun normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(1 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * References:\n *   [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *   [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunNormal(args);\n}\n\n/**\n * LeCun uniform initializer.\n *\n * It draws samples from a uniform distribution in the interval\n * `[-limit, limit]` with `limit = sqrt(3 / fanIn)`,\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunUniform(args);\n}\n\n/**\n * Initializer that generates a random orthogonal matrix.\n *\n * Reference:\n * [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function orthogonal(args: OrthogonalArgs): Initializer {\n  return new Orthogonal(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Utilities related to persistent state in the backend.\n */\n\n/**\n * An ID to track `tf.SymbolicTensor`s and derived classes.\n * Required in different places in engine/topology.ts to identify unique\n * tensors.\n */\nlet _nextUniqueTensorId = 0;\n\nexport function getNextUniqueTensorId(): number {\n  return _nextUniqueTensorId++;\n}\n\nconst _uidPrefixes: {[prefix: string]: number} = {};\n\n/**\n * Provides a unique UID given a string prefix.\n *\n * @param prefix\n */\nexport function getUid(prefix = ''): string {\n  if (!(prefix in _uidPrefixes)) {\n    _uidPrefixes[prefix] = 0;\n  }\n  _uidPrefixes[prefix] += 1;\n  return prefix + _uidPrefixes[prefix].toString();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\n// tslint:enable\n\n/**\n * Determine whether the input is an Array of Shapes.\n */\nexport function isArrayOfShapes(x: Shape|Shape[]): boolean {\n  return Array.isArray(x) && Array.isArray(x[0]);\n}\n\n/**\n * Special case of normalizing shapes to lists.\n *\n * @param x A shape or list of shapes to normalize into a list of Shapes.\n * @return A list of Shapes.\n */\nexport function normalizeShapeList(x: Shape|Shape[]): Shape[] {\n  if (x.length === 0) {\n    return [];\n  }\n  if (!Array.isArray(x[0])) {\n    return [x] as Shape[];\n  }\n  return x as Shape[];\n}\n\n/**\n * Helper function to obtain exactly one Tensor.\n * @param xs: A single `tf.Tensor` or an `Array` of `tf.Tensor`s.\n * @return A single `tf.Tensor`. If `xs` is an `Array`, return the first one.\n * @throws ValueError: If `xs` is an `Array` and its length is not 1.\n */\nexport function getExactlyOneTensor(xs: Tensor|Tensor[]): Tensor {\n  let x: Tensor;\n  if (Array.isArray(xs)) {\n    if (xs.length !== 1) {\n      throw new ValueError(`Expected Tensor length to be 1; got ${xs.length}`);\n    }\n    x = xs[0];\n  } else {\n    x = xs;\n  }\n  return x;\n}\n\n/**\n * Helper function to obtain exactly on instance of Shape.\n *\n * @param shapes Input single `Shape` or Array of `Shape`s.\n * @returns If input is a single `Shape`, return it unchanged. If the input is\n *   an `Array` containing exactly one instance of `Shape`, return the instance.\n *   Otherwise, throw a `ValueError`.\n * @throws ValueError: If input is an `Array` of `Shape`s, and its length is not\n *   1.\n */\nexport function getExactlyOneShape(shapes: Shape|Shape[]): Shape {\n  if (Array.isArray(shapes) && Array.isArray(shapes[0])) {\n    if (shapes.length === 1) {\n      shapes = shapes as Shape[];\n      return shapes[0];\n    } else {\n      throw new ValueError(`Expected exactly 1 Shape; got ${shapes.length}`);\n    }\n  } else {\n    return shapes as Shape;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {LayerVariable} from '../variables';\n\n/**\n * Count the elements in an Array of LayerVariables.\n *\n * @param weights: The LayerVariables of which the constituent numbers are to\n *   be counted.\n * @returns A count of the elements in all the LayerVariables\n */\nexport function countParamsInWeights(weights: LayerVariable[]): number {\n  let count = 0;\n  for (const weight of weights) {\n    if (weight.shape.length === 0) {\n      count += 1;\n    } else {\n      count += weight.shape.reduce((a, b) => a * b);\n    }\n  }\n  return count;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, Tensor, variableGrads} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId} from './backend/state';\nimport {getScopedTensorName, getUniqueTensorName} from './common';\nimport {Constraint} from './constraints';\nimport {NotImplementedError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {HasShape} from './types';\n\nconst DEFAULT_VARIABLE_NAME_PREFIX = 'Variable';\n\n/**\n * A `tf.layers.LayerVariable` is similar to a `tf.Tensor` in that it has a\n * dtype and shape, but its value is mutable.  The value is itself represented\n * as a`tf.Tensor`, and can be read with the `read()` method and updated with\n * the `write()` method.\n */\nexport class LayerVariable {\n  readonly dtype: DataType;\n  readonly shape: Shape;\n\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName: string;\n  private trainable_: boolean;\n\n  protected readonly val: tfc.Variable;\n  readonly constraint: Constraint;\n  /**\n   * Construct Variable from a `tf.Tensor`.\n   *\n   * If not explicitly named, the Variable will be given a name with the\n   * prefix 'Variable'. Variable names are unique. In the case of name\n   * collision, suffixies '_<num>' will be added to the name.\n   *\n   * @param val Initial value of the Variable.\n   * @param name Name of the variable. If `null` or `undefined` is provided, it\n   *   will default a name with the prefix 'Variable'.\n   * @param constraint Optional, projection function to be applied to the\n   * variable after optimize updates\n   * @throws ValueError if `name` is `null` or `undefined`.\n   */\n  constructor(\n      val: Tensor, dtype: DataType = 'float32',\n      name = DEFAULT_VARIABLE_NAME_PREFIX, trainable = true,\n      constraint: Constraint = null) {\n    this.dtype = dtype == null ? 'float32' : dtype;\n    this.shape = val.shape;\n    this.id = getNextUniqueTensorId();\n\n    name = name == null ? DEFAULT_VARIABLE_NAME_PREFIX : name;\n    this.originalName = getScopedTensorName(name);\n    this.name = getUniqueTensorName(this.originalName);\n\n    this.trainable_ = trainable;\n    this.constraint = constraint;\n\n    this.val = tfc.variable(val, this.trainable_, this.name, this.dtype);\n  }\n\n  /**\n   * Get a snapshot of the Variable's value.\n   *\n   * The returned value is a snapshot of the Variable's value at the time of\n   * the invocation. Future mutations in the value of the tensor will only\n   * be reflected by future calls to this method.\n   */\n  read(): Tensor {\n    this.assertNotDisposed();\n    return this.val;\n  }\n\n  /**\n   * Update the value of the Variable.\n   *\n   * @param newVal: The new value to update to. Must be consistent with the\n   *   dtype and shape of the Variable.\n   * @return This Variable.\n   */\n  write(newVal: Tensor) {\n    // TODO(cais): Once  TF.js Core supports Tensor.dtype, check dtype match.\n    this.assertNotDisposed();\n    checkShapesMatch(this.val, newVal);\n    // Skip updating if this is the exact same tensor.\n    if (this.val.id !== newVal.id) {\n      this.val.assign(newVal);\n      if (this.constraint != null) {\n        this.val.assign(this.constraint.apply(this.val));\n      }\n    }\n    return this;\n  }\n\n  /**\n   * Dispose this LayersVariable instance from memory.\n   */\n  dispose(): void {\n    this.assertNotDisposed();\n    this.val.dispose();\n  }\n\n  protected assertNotDisposed(): void {\n    if (this.val.isDisposed) {\n      throw new Error(`LayersVariable ${this.name} is already disposed.`);\n    }\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.trainable_ = trainable;\n    this.val.trainable = trainable;\n  }\n}\n\nfunction checkShapesMatch(x: HasShape, y: HasShape): void {\n  if (x.shape.toString() !== y.shape.toString()) {\n    throw new Error(\n        'Shape mismatch: ' + JSON.stringify(x.shape) + ' vs. ' +\n        JSON.stringify(y.shape));\n  }\n}\n\n/**\n * Create a Variable.\n * @param x The initial value of the `Variable`.\n * @param dtype optional, the type of the variable.\n * @param name optional, the name of the variable, default provided by\n * Variable.\n * @param constraint optional, a constraint to be applied after every update.\n * @return The newly instantiated `Variable`.\n */\nexport function variable(\n    x: Tensor, dtype?: DataType, name?: string,\n    constraint?: Constraint): LayerVariable {\n  return new LayerVariable(x, dtype, name, true, constraint);\n}\n\n/**\n * Instantiates an all-zeros Variable and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-zero Variable.\n */\nexport function zerosVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  return new LayerVariable(tfc.zeros(shape), dtype, name);\n}\n\n/**\n * Instantiates an all-zeros tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function zerosLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.zerosLike(x), dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-ones Variable.\n */\nexport function onesVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  const allocated = tfc.ones(shape);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function onesLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  const allocated = tfc.onesLike(x);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiate an identity matrix and returns it, as a Variable\n *\n * @param size Number of rows/columns.\n * @param dtype Data type of returned Variable.\n * @param name Name of returned Variable.\n * @return A Variable, an identity matrix.\n */\nexport function eyeVariable(\n    size: number, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.eye(size), dtype, name);\n}\n\n/**\n * Get a Variable with uniform distribution of values.\n * @param shape Shape of the tensor.\n * @param minval Lower bound of the uniform distribution.\n * @param maxval Upper bound of the uniform distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The uniform-random Variable.\n */\nexport function randomUniformVariable(\n    shape: Shape, minval: number, maxval: number, dtype?: DataType,\n    seed?: number, name = 'randomUniform'): LayerVariable {\n  return new LayerVariable(\n      tfc.randomUniform(shape, minval, maxval, dtype), dtype, name);\n}\n\n/**\n * Get a Variable with truncated-normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function truncatedNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'truncatedNormal'): LayerVariable {\n  // TODO(cais): Implement logic for dtype and seed once they are supported\n  // by deeplearn.js.\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormal does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.truncatedNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n/**\n * Get a Variable with normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function randomNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'randomNormal'): LayerVariable {\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormalVariable does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.randomNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n\n/**\n * Update the value of a Variable.\n * @param x The Variable to be updated.\n * @param xNew The new value to update to.\n * @return The Variable updated.\n */\nexport function update(x: LayerVariable, xNew: Tensor): LayerVariable {\n  return x.write(xNew);\n}\n\n/**\n * Update the value of a Variable by adding an increment.\n * @param x The Variable to be updated.\n * @param increment The incrment to add to `x`.\n * @return The Variable updated.\n */\nexport function updateAdd(x: LayerVariable, increment: Tensor): LayerVariable {\n  return x.write(tfc.add(x.read(), increment));\n}\n\n/**\n * Update the value of a Variable by subtracting a decrement.\n * @param x The Variable to be updated.\n * @param decrement The decrement to subtract from `x`.\n * @return The Variable updated.\n */\nexport function updateSub(x: LayerVariable, decrement: Tensor): LayerVariable {\n  return x.write(tfc.sub(x.read(), decrement));\n}\n\n/**\n * Get the values of an array of Variables.\n *\n * @param tensors An `Array` of `Variable`s to get the values of.\n * @return The values of the inputs, as an `Array` of`tf.Tensor`s.\n */\nexport function batchGetValue(xs: LayerVariable[]): Tensor[] {\n  return xs.map(x => x.read());\n}\n\n/**\n * Update the value of multiple Variables at once.\n *\n * @param variablesAndValues An `Array`, each element is of type\n *   [Variable, Tensor]. The first item is the\n *   `Variable` of which the value is to be updated. The second item\n *   carries the new value.\n */\nexport function batchSetValue(\n    variablesAndValues: Array<[LayerVariable, Tensor]>): void {\n  variablesAndValues.forEach(variableAndValue => {\n    const variable: LayerVariable = variableAndValue[0];\n    variable.write(variableAndValue[1]);\n  });\n}\n\n/**\n * Returns the gradients of `variables` w.r.t. the return value of `lossFn`.\n * @param lossFn A function which returns a Scalar to be used as the function\n *   value (i.e., numerator) for differentiation.\n * @param variables List of variables to be used as the independent variables\n *   (i.e., denominator) for differentiation.\n * @returns An Array of gradients tensors.\n */\nexport function gradients(\n    lossFn: () => tfc.Scalar, variables: LayerVariable[]): Tensor[] {\n  // TODO(cais): The return type signature can be simplified if deeplearn makes\n  //   the corresponding type public.\n  const variableList =\n      variables.map(variable => variable.read() as tfc.Variable);\n  const valudAndGrads = variableGrads(lossFn, variableList);\n  return variables.map(variable => valudAndGrads.grads[variable.name]);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {DataType, Scalar, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId, getUid} from '../backend/state';\nimport {getScopedTensorName, getUniqueTensorName, nameScope} from '../common';\nimport {Constraint} from '../constraints';\nimport {AttributeError, NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {getInitializer, Initializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs, RegularizerFn} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as types_utils from '../utils/types_utils';\nimport * as variable_utils from '../utils/variable_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\n\n// TODO(michaelterry): This is a stub until it's defined.\nexport type Op = (x: LayerVariable) => LayerVariable;\n\n/**\n * Constructor arguments for InputSpec.\n */\nexport interface InputSpecArgs {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n}\n\n/**\n * Specifies the ndim, dtype and shape of every input to a layer.\n *\n * Every layer should expose (if appropriate) an `inputSpec` attribute:\n * a list of instances of InputSpec (one per input tensor).\n *\n * A null entry in a shape is compatible with any dimension,\n * a null shape is compatible with any shape.\n */\nexport class InputSpec {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n\n  constructor(args: InputSpecArgs) {\n    this.dtype = args.dtype;\n    this.shape = args.shape;\n    /*\n      TODO(michaelterry): Could throw error if ndim and shape are both defined\n        (then backport).\n    */\n    if (args.shape != null) {\n      this.ndim = args.shape.length;\n    } else {\n      this.ndim = args.ndim;\n    }\n    this.maxNDim = args.maxNDim;\n    this.minNDim = args.minNDim;\n    this.axes = args.axes || {};\n  }\n}\n\n/**\n * `tf.SymbolicTensor` is a placeholder for a Tensor without any concrete value.\n *\n * They are most often encountered when building a graph of `Layer`s for a\n * a `tf.LayersModel` and the input data's shape, but not values are known.\n *\n * @doc {heading: 'Models', 'subheading': 'Classes'}\n */\nexport class SymbolicTensor {\n  /* A unique ID for the tensor to be able to differentiate tensors. */\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName?: string;\n  /**\n   * Rank/dimensionality of the tensor.\n   */\n  readonly rank: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  nodeIndex: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  tensorIndex: number;\n\n  /**\n   *\n   * @param dtype\n   * @param shape\n   * @param sourceLayer The Layer that produced this symbolic tensor.\n   * @param inputs The inputs passed to sourceLayer's __call__() method.\n   * @param nodeIndex\n   * @param tensorIndex\n   * @param callArgs The keyword arguments passed to the __call__() method.\n   * @param name\n   * @param outputTensorIndex The index of this tensor in the list of outputs\n   *   returned by apply().\n   */\n  constructor(\n      readonly dtype: DataType, readonly shape: Shape,\n      public sourceLayer: Layer, readonly inputs: SymbolicTensor[],\n      readonly callArgs: Kwargs, name?: string,\n      readonly outputTensorIndex?: number) {\n    this.id = getNextUniqueTensorId();\n    if (name != null) {\n      this.originalName = getScopedTensorName(name);\n      this.name = getUniqueTensorName(this.originalName);\n    }\n    this.rank = shape.length;\n  }\n}\n\n/**\n * Constructor arguments for Node.\n */\nexport interface NodeArgs {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`.\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n}\n\n/**\n * The type of the return value of Layer.dispose() and Container.dispose().\n */\nexport interface DisposeResult {\n  /**\n   * Reference count after the dispose call.\n   */\n  refCountAfterDispose: number;\n\n  /**\n   * Number of variables dispose in this dispose call.\n   */\n  numDisposedVariables: number;\n}\n\nlet _nextNodeID = 0;\n\n/**\n * A `Node` describes the connectivity between two layers.\n *\n * Each time a layer is connected to some new input,\n * a node is added to `layer.inboundNodes`.\n *\n * Each time the output of a layer is used by another layer,\n * a node is added to `layer.outboundNodes`.\n *\n * `nodeIndices` and `tensorIndices` are basically fine-grained coordinates\n * describing the origin of the `inputTensors`, verifying the following:\n *\n * `inputTensors[i] ==\n * inboundLayers[i].inboundNodes[nodeIndices[i]].outputTensors[\n *   tensorIndices[i]]`\n *\n * A node from layer A to layer B is added to:\n *     A.outboundNodes\n *     B.inboundNodes\n */\nexport class Node {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n\n  readonly id: number;\n\n  constructor(\n      args: NodeArgs,\n      // TODO(michaelterry): Define actual type for this.\n      public callArgs?: Kwargs) {\n    this.id = _nextNodeID++;\n    /*\n      Layer instance (NOT a list).\n      this is the layer that takes a list of input tensors\n      and turns them into a list of output tensors.\n      the current node will be added to\n      the inboundNodes of outboundLayer.\n    */\n    this.outboundLayer = args.outboundLayer;\n\n    /*\n        The following 3 properties describe where\n        the input tensors come from: which layers,\n        and for each layer, which node and which\n        tensor output of each node.\n    */\n\n    // List of layer instances.\n    this.inboundLayers = args.inboundLayers;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.nodeIndices = args.nodeIndices;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.tensorIndices = args.tensorIndices;\n\n    /*\n        Following 2 properties:\n        tensor inputs and outputs of outboundLayer.\n    */\n\n    // List of tensors. 1:1 mapping with inboundLayers.\n    this.inputTensors = args.inputTensors;\n    // List of tensors, created by outboundLayer.call().\n    this.outputTensors = args.outputTensors;\n\n    /*\n        Following 2 properties: input and output masks.\n        List of tensors, 1:1 mapping with inputTensor.\n    */\n    this.inputMasks = args.inputMasks;\n    // List of tensors, created by outboundLayer.computeMask().\n    this.outputMasks = args.outputMasks;\n\n    // Following 2 properties: input and output shapes.\n\n    // List of shape tuples, shapes of inputTensors.\n    this.inputShapes = args.inputShapes;\n    // List of shape tuples, shapes of outputTensors.\n    this.outputShapes = args.outputShapes;\n\n    // Add nodes to all layers involved.\n    for (const layer of args.inboundLayers) {\n      if (layer != null) {\n        layer.outboundNodes.push(this);\n      }\n    }\n    args.outboundLayer.inboundNodes.push(this);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const inboundNames: string[] = [];\n    for (const layer of this.inboundLayers) {\n      if (layer != null) {\n        inboundNames.push(layer.name);\n      } else {\n        inboundNames.push(null);\n      }\n    }\n    return {\n      outboundLayer: this.outboundLayer ? this.outboundLayer.name : null,\n      inboundLayers: inboundNames,\n      nodeIndices: this.nodeIndices,\n      tensorIndices: this.tensorIndices\n    };\n  }\n}\n\n/** Constructor arguments for Layer. */\nexport declare interface LayerArgs {\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  inputShape?: Shape;\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  batchInputShape?: Shape;\n  /**\n   * If `inputShape` is specified and `batchInputShape` is *not* specified,\n   * `batchSize` is used to construct the `batchInputShape`: `[batchSize,\n   * ...inputShape]`\n   */\n  batchSize?: number;\n  /**\n   * The data-type for this layer. Defaults to 'float32'.\n   * This argument is only applicable to input layers (the first layer of a\n   * model).\n   */\n  dtype?: DataType;\n  /** Name for this layer. */\n  name?: string;\n  /**\n   * Whether the weights of this layer are updatable by `fit`.\n   * Defaults to true.\n   */\n  trainable?: boolean;\n  /**\n   * Initial weight values of the layer.\n   */\n  weights?: Tensor[];\n  /** Legacy support. Do not use for new code. */\n  inputDType?: DataType;\n}\n\n// If necessary, add `output` arguments to the CallHook function.\n// This is currently used for testing only, but may be used for debugger-related\n// purposes in the future.\nexport type CallHook = (inputs: Tensor|Tensor[], kwargs: Kwargs) => void;\n\nlet _nextLayerID = 0;\n\n/**\n * A layer is a grouping of operations and weights that can be composed to\n * create a `tf.LayersModel`.\n *\n * Layers are constructed by using the functions under the\n * [tf.layers](#Layers-Basic) namespace.\n *\n * @doc {heading: 'Layers', subheading: 'Classes', namespace: 'layers'}\n */\nexport abstract class Layer extends serialization.Serializable {\n  /** Name for this layer. Must be unique within a model. */\n  name: string;\n  /**\n   * List of InputSpec class instances.\n   *\n   * Each entry describes one required input:\n   * - ndim\n   * - dtype\n   * A layer with `n` input tensors must have an `inputSpec` of length `n`.\n   */\n  inputSpec: InputSpec[];\n  supportsMasking: boolean;\n  /** Whether the layer weights will be updated during training. */\n  protected trainable_: boolean;\n  batchInputShape: Shape;\n  dtype: DataType;\n  initialWeights: Tensor[];\n\n  inboundNodes: Node[];\n  outboundNodes: Node[];\n\n  activityRegularizer: Regularizer;\n\n  protected _trainableWeights: LayerVariable[];\n  private _nonTrainableWeights: LayerVariable[];\n  private _losses: RegularizerFn[];\n  // TODO(cais): _updates is currently unused.\n  private _updates: Tensor[];\n  private _built: boolean;\n  private _callHook: CallHook = null;\n\n  private _addedWeightNames: string[] = [];\n\n  readonly id: number;\n\n  // Porting Notes: PyKeras does not have this property in this base Layer\n  //   class. Instead lets Layer subclass set it dynamically and checks the\n  //   value with `hasattr`. In tfjs-layers, we let this be a member of this\n  //   base class.\n  protected _stateful = false;\n\n  protected _refCount: number|null;\n\n  // A flag for whether fast (i.e., all-zero) weight initialization is to\n  // be used during `build()` call. This speeds up weight initialization\n  // by saving unnecessary calls to expensive initializers in cases where\n  // the initialized values will be overwritten by loaded weight values\n  // during model loading.\n  private fastWeightInitDuringBuild: boolean;\n\n  constructor(args: LayerArgs = {}) {\n    super();\n    this.id = _nextLayerID++;\n\n    this.activityRegularizer = null;\n\n    this.inputSpec = null;\n    this.supportsMasking = false;\n\n    // These properties will be set upon call of this.build()\n    this._trainableWeights = [];\n    this._nonTrainableWeights = [];\n    this._losses = [];\n    this._updates = [];\n    this._built = false;\n\n    /*\n      These lists will be filled via successive calls\n      to this.addInboundNode().\n     */\n    this.inboundNodes = [];\n    this.outboundNodes = [];\n\n    let name = args.name;\n    if (!name) {\n      const prefix = this.getClassName();\n      name = generic_utils.toSnakeCase(prefix) + '_' + getUid(prefix);\n    }\n    this.name = name;\n\n    this.trainable_ = args.trainable == null ? true : args.trainable;\n\n    if (args.inputShape != null || args.batchInputShape != null) {\n      /*\n        In this case we will later create an input layer\n        to insert before the current layer\n       */\n      let batchInputShape: Shape;\n      if (args.batchInputShape != null) {\n        batchInputShape = args.batchInputShape;\n      } else if (args.inputShape != null) {\n        let batchSize: number = null;\n        if (args.batchSize != null) {\n          batchSize = args.batchSize;\n        }\n        batchInputShape = [batchSize].concat(args.inputShape);\n      }\n      this.batchInputShape = batchInputShape;\n\n      // Set dtype.\n      let dtype = args.dtype;\n      if (dtype == null) {\n        dtype = args.inputDType;\n      }\n      if (dtype == null) {\n        dtype = 'float32';\n      }\n      this.dtype = dtype;\n    }\n\n    if (args.weights != null) {\n      this.initialWeights = args.weights;\n    } else {\n      this.initialWeights = null;\n    }\n\n    // The value of `_refCount` is initialized to null. When the layer is used\n    // in a symbolic way for the first time, it will be set to 1.\n    this._refCount = null;\n\n    this.fastWeightInitDuringBuild = false;\n  }\n\n  /**\n   * Converts a layer and its index to a unique (immutable type) name.\n   * This function is used internally with `this.containerNodes`.\n   * @param layer The layer.\n   * @param nodeIndex The layer's position (e.g. via enumerate) in a list of\n   *   nodes.\n   *\n   * @returns The unique name.\n   */\n  protected static nodeKey(layer: Layer, nodeIndex: number) {\n    return layer.name + '_ib-' + nodeIndex.toString();\n  }\n\n  /**\n   * Returns this.inboundNode at index nodeIndex.\n   *\n   * Porting note: This is a replacement for _get_node_attribute_at_index()\n   * @param nodeIndex\n   * @param attrName The name of the attribute related to request for this node.\n   */\n  private getNodeAtIndex(nodeIndex: number, attrName: string): Node {\n    if (this.inboundNodes.length === 0) {\n      throw new RuntimeError(\n          'The layer has never been called ' +\n          `and thus has no defined ${attrName}.`);\n    }\n    if (this.inboundNodes.length <= nodeIndex) {\n      throw new ValueError(\n          `Asked to get ${attrName} at node ${nodeIndex}, ` +\n          `but the layer has only ${this.inboundNodes.length} inbound nodes.`);\n    }\n    return this.inboundNodes[nodeIndex];\n  }\n\n  /**\n   * Retrieves the input tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple inputs).\n   */\n  getInputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple outputs).\n   */\n  getOutputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'output').outputTensors);\n  }\n\n  // Properties\n\n  /**\n   * Retrieves the input tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Input tensor or list of input tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get input(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer input\" ' +\n          'is ill-defined. ' +\n          'Use `getInputAt(nodeIndex)` instead.');\n    } else if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' is not connected, no input to return.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Output tensor or list of output tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get output(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has no inbound nodes.');\n    }\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer output\" ' +\n          'is ill-defined. ' +\n          'Use `getOutputAt(nodeIndex)` instead.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'output').outputTensors);\n  }\n\n  get losses(): RegularizerFn[] {\n    return this._losses;\n  }\n\n  /**\n   * Retrieves the Layer's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Layer.loss in PyKeras.\n    //   In PyKeras, Layer.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return this.losses.map(lossFn => lossFn());\n  }\n\n  get updates(): Tensor[] {\n    return this._updates;\n  }\n\n  get built(): boolean {\n    return this._built;\n  }\n\n  set built(built: boolean) {\n    this._built = built;\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this._trainableWeights.forEach(w => w.trainable = trainable);\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (this.trainable_) {\n      return this._trainableWeights.filter(w => w.trainable);\n    } else {\n      return [];\n    }\n  }\n\n  set trainableWeights(weights: LayerVariable[]) {\n    this._trainableWeights = weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    if (this.trainable) {\n      return this._trainableWeights.filter(w => !w.trainable)\n          .concat(this._nonTrainableWeights);\n    } else {\n      return this._trainableWeights.concat(this._nonTrainableWeights);\n    }\n  }\n\n  set nonTrainableWeights(weights: LayerVariable[]) {\n    this._nonTrainableWeights = weights;\n  }\n\n  /**\n   * The concatenation of the lists trainableWeights and nonTrainableWeights\n   * (in this order).\n   */\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  get stateful(): boolean {\n    return this._stateful;\n  }\n\n  /**\n   * Reset the states of the layer.\n   *\n   * This method of the base Layer class is essentially a no-op.\n   * Subclasses that are stateful (e.g., stateful RNNs) should override this\n   * method.\n   */\n  resetStates(): void {\n    if (!this.stateful) {\n      throw new Error(\n          'Cannot call the resetStates() method of a non-stateful Layer ' +\n          'object.');\n    }\n  }\n\n  /**\n   * Checks compatibility between the layer and provided inputs.\n   *\n   * This checks that the tensor(s) `input`\n   * verify the input assumptions of the layer\n   * (if any). If not, exceptions are raised.\n   *\n   * @param inputs Input tensor or list of input tensors.\n   *\n   * @exception ValueError in case of mismatch between\n   *   the provided inputs and the expectations of the layer.\n   */\n  protected assertInputCompatibility(inputs: Tensor|Tensor[]|SymbolicTensor|\n                                     SymbolicTensor[]): void {\n    inputs = generic_utils.toList(inputs);\n    if (this.inputSpec == null || this.inputSpec.length === 0) {\n      return;\n    }\n    const inputSpec = generic_utils.toList(this.inputSpec);\n    if (inputs.length !== inputSpec.length) {\n      throw new ValueError(\n          `Layer ${this.name} expects ${inputSpec.length} inputs, ` +\n          `but it received ${inputs.length} input tensors. ` +\n          `Input received: ${inputs}`);\n    }\n    for (let inputIndex = 0; inputIndex < inputs.length; inputIndex++) {\n      const x = inputs[inputIndex];\n      const spec: InputSpec = inputSpec[inputIndex];\n      if (spec == null) {\n        continue;\n      }\n\n      // Check ndim.\n      const ndim = x.rank;\n      if (spec.ndim != null) {\n        if (ndim !== spec.ndim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}: ` +\n              `expected ndim=${spec.ndim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.maxNDim != null) {\n        if (ndim > spec.maxNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected max_ndim=${spec.maxNDim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.minNDim != null) {\n        if (ndim < spec.minNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected min_ndim=${spec.minNDim}, found ndim=${ndim}.`);\n        }\n      }\n\n      // Check dtype.\n      if (spec.dtype != null) {\n        if (x.dtype !== spec.dtype) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name} ` +\n              `: expected dtype=${spec.dtype}, found dtype=${x.dtype}.`);\n        }\n      }\n\n      // Check specific shape axes.\n      if (spec.axes) {\n        const xShape = x.shape;\n        for (const key in spec.axes) {\n          const axis = Number(key);\n          const value = spec.axes[key];\n          // Perform Python-style slicing in case axis < 0;\n          // TODO(cais): Use https://github.com/alvivi/typescript-underscore to\n          // ensure type safety through Underscore calls.\n          const xShapeAtAxis =\n              axis >= 0 ? xShape[axis] : xShape[xShape.length + axis];\n          if (value != null && [value, null].indexOf(xShapeAtAxis) === -1) {\n            throw new ValueError(\n                `Input ${inputIndex} is incompatible with layer ` +\n                `${this.name}: expected axis ${axis} of input shape to ` +\n                `have value ${value} but got shape ${xShape}.`);\n          }\n        }\n      }\n\n      // Check shape.\n      if (spec.shape != null) {\n        for (let i = 0; i < spec.shape.length; ++i) {\n          const specDim = spec.shape[i];\n          const dim = x.shape[i];\n          if (specDim != null && dim != null) {\n            if (specDim !== dim) {\n              throw new ValueError(\n                  `Input ${inputIndex} is incompatible with layer ` +\n                  `${this.name}: expected shape=${spec.shape}, ` +\n                  `found shape=${x.shape}.`);\n            }\n          }\n        }\n      }\n    }\n  }\n\n  /**\n   * This is where the layer's logic lives.\n   *\n   * @param inputs Input tensor, or list/tuple of input tensors.\n   * @param kwargs Additional keyword arguments.\n   *\n   * @return A tensor or list/tuple of tensors.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return inputs;\n  }\n\n  protected invokeCallHook(inputs: Tensor|Tensor[], kwargs: Kwargs) {\n    if (this._callHook != null) {\n      this._callHook(inputs, kwargs);\n    }\n  }\n\n  /**\n   * Set call hook.\n   * This is currently used for testing only.\n   * @param callHook\n   */\n  setCallHook(callHook: CallHook) {\n    this._callHook = callHook;\n  }\n\n  /**\n   * Clear call hook.\n   * This is currently used for testing only.\n   */\n  clearCallHook() {\n    this._callHook = null;\n  }\n\n  /**\n   * Builds or executes a `Layer's logic.\n   *\n   * When called with `tf.Tensor`(s), execute the `Layer`s computation and\n   * return Tensor(s). For example:\n   *\n   * ```js\n   * const denseLayer = tf.layers.dense({\n   *   units: 1,\n   *   kernelInitializer: 'zeros',\n   *   useBias: false\n   * });\n   *\n   * // Invoke the layer's apply() method with a `tf.Tensor` (with concrete\n   * // numeric values).\n   * const input = tf.ones([2, 2]);\n   * const output = denseLayer.apply(input);\n   *\n   * // The output's value is expected to be [[0], [0]], due to the fact that\n   * // the dense layer has a kernel initialized to all-zeros and does not have\n   * // a bias.\n   * output.print();\n   * ```\n   *\n   * When called with `tf.SymbolicTensor`(s), this will prepare the layer for\n   * future execution.  This entails internal book-keeping on shapes of\n   * expected Tensors, wiring layers together, and initializing weights.\n   *\n   * Calling `apply` with `tf.SymbolicTensor`s are typically used during the\n   * building of non-`tf.Sequential` models. For example:\n   *\n   * ```js\n   * const flattenLayer = tf.layers.flatten();\n   * const denseLayer = tf.layers.dense({units: 1});\n   *\n   * // Use tf.layers.input() to obtain a SymbolicTensor as input to apply().\n   * const input = tf.input({shape: [2, 2]});\n   * const output1 = flattenLayer.apply(input);\n   *\n   * // output1.shape is [null, 4]. The first dimension is the undetermined\n   * // batch size. The second dimension comes from flattening the [2, 2]\n   * // shape.\n   * console.log(JSON.stringify(output1.shape));\n   *\n   * // The output SymbolicTensor of the flatten layer can be used to call\n   * // the apply() of the dense layer:\n   * const output2 = denseLayer.apply(output1);\n   *\n   * // output2.shape is [null, 1]. The first dimension is the undetermined\n   * // batch size. The second dimension matches the number of units of the\n   * // dense layer.\n   * console.log(JSON.stringify(output2.shape));\n   *\n   * // The input and output and be used to construct a model that consists\n   * // of the flatten and dense layers.\n   * const model = tf.model({inputs: input, outputs: output2});\n   * ```\n   *\n   * @param inputs a `tf.Tensor` or `tf.SymbolicTensor` or an Array of them.\n   * @param kwargs Additional keyword arguments to be passed to `call()`.\n   *\n   * @return Output of the layer's `call` method.\n   *\n   * @exception ValueError error in case the layer is missing shape information\n   *   for its `build` call.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  // Porting Note: This is a replacement for __call__() in Python.\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    kwargs = kwargs || {};\n\n    this.assertNotDisposed();\n\n    // Ensure inputs are all the same type.\n    const inputsList = generic_utils.toList(inputs);\n\n    let allAreSymbolic = true;\n    for (const input of inputsList) {\n      if (!(input instanceof SymbolicTensor)) {\n        allAreSymbolic = false;\n        break;\n      }\n    }\n    let noneAreSymbolic = true;\n    for (const input of inputsList) {\n      if (input instanceof SymbolicTensor) {\n        noneAreSymbolic = false;\n        break;\n      }\n    }\n\n    if (allAreSymbolic === noneAreSymbolic) {\n      throw new ValueError(\n          'Arguments to apply() must be all ' +\n          'SymbolicTensors or all Tensors');\n    }\n\n    // TODO(michaelterry): nameScope() may not be necessary.\n    return nameScope(this.name, () => {\n      // Handle laying building (weight creating, input spec locking).\n      if (!this.built) {\n        /*\n          Throw exceptions in case the input is not compatible\n          with the inputSpec specified in the layer constructor.\n         */\n        this.assertInputCompatibility(inputs);\n\n        // Collect input shapes to build layer.\n        const inputShapes: Shape[] = [];\n        for (const xElem of generic_utils.toList(inputs)) {\n          inputShapes.push(xElem.shape);\n        }\n        this.build(generic_utils.singletonOrArray(inputShapes));\n        this.built = true;\n\n        // Load weights that were specified at layer instantiation.\n        if (this.initialWeights) {\n          this.setWeights(this.initialWeights);\n        }\n\n        if (this._refCount === null && noneAreSymbolic) {\n          // The first use of this layer is a non-symbolic call, set ref count\n          // to 1 so the Layer can be properly disposed if its dispose() method\n          // is called.\n          this._refCount = 1;\n        }\n      }\n\n      /*\n        Throw exceptions in case the input is not compatible\n        with the inputSpec set at build time.\n      */\n      this.assertInputCompatibility(inputs);\n\n      // Handle mask propagation.\n      // TODO(michaelterry): Mask propagation not currently implemented.\n\n      // Actually call the layer, collecting output(s), mask(s), and shape(s).\n      if (noneAreSymbolic) {\n        let output = this.call(inputs as Tensor | Tensor[], kwargs);\n        // TODO(michaelterry): Compute the outputMask\n\n        // If the layer returns tensors from its inputs, unmodified,\n        // we copy them to avoid loss of tensor metadata.\n        const outputList: Tensor[] = generic_utils.toList(output);\n        const outputListCopy: Tensor[] = [];\n        // TODO(michaelterry): This copying may not be necessary given our eager\n        // backend.\n        for (let x of outputList) {\n          if (inputsList.indexOf(x) !== -1) {\n            x = x.clone();\n          }\n          outputListCopy.push(x);\n        }\n        output = generic_utils.singletonOrArray(outputListCopy);\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        // TODO(michaelterry): Call addInboundNode()?\n        return output;\n      } else {\n        const inputShape = collectInputShape(inputs);\n        const outputShape = this.computeOutputShape(inputShape);\n        let output: SymbolicTensor|SymbolicTensor[];\n        const outputDType = guessOutputDType(inputs);\n        this.warnOnIncompatibleInputShape(\n            Array.isArray(inputs) ? inputShape[0] as Shape :\n                                    inputShape as Shape);\n\n        if (outputShape != null && outputShape.length > 0 &&\n            Array.isArray(outputShape[0])) {\n          // We have multiple output shapes. Create multiple output tensors.\n          output = (outputShape as Shape[])\n                       .map(\n                           (shape, index) => new SymbolicTensor(\n                               outputDType, shape, this,\n                               generic_utils.toList(inputs), kwargs, this.name,\n                               index));\n        } else {\n          output = new SymbolicTensor(\n              outputDType, outputShape as Shape, this,\n              generic_utils.toList(inputs), kwargs, this.name);\n        }\n\n        /*\n          Add an inbound node to the layer, so that it keeps track\n          of the call and of all new variables created during the call.\n          This also updates the layer history of the output tensor(s).\n          If the input tensor(s) had no previous history,\n          this does nothing.\n        */\n        this.addInboundNode(\n            inputs as SymbolicTensor | SymbolicTensor[], output, null, null,\n            inputShape, outputShape, kwargs);\n        this._refCount++;\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        return output;\n      }\n    });\n  }\n\n  /**\n   * Check compatibility between input shape and this layer's batchInputShape.\n   *\n   * Print warning if any incompatibility is found.\n   *\n   * @param inputShape Input shape to be checked.\n   */\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {\n    if (this.batchInputShape == null) {\n      return;\n    } else if (inputShape.length !== this.batchInputShape.length) {\n      console.warn(\n          `The rank of the input tensor provided (shape: ` +\n          `${JSON.stringify(inputShape)}) does not match that of the ` +\n          `batchInputShape (${JSON.stringify(this.batchInputShape)}) ` +\n          `of the layer ${this.name}`);\n    } else {\n      let dimMismatch = false;\n      this.batchInputShape.forEach((dimension, i) => {\n        if (dimension != null && inputShape[i] != null &&\n            inputShape[i] !== dimension) {\n          dimMismatch = true;\n        }\n      });\n      if (dimMismatch) {\n        console.warn(\n            `The shape of the input tensor ` +\n            `(${JSON.stringify(inputShape)}) does not ` +\n            `match the expectation of layer ${this.name}: ` +\n            `${JSON.stringify(this.batchInputShape)}`);\n      }\n    }\n  }\n\n  /**\n   * Retrieves the output shape(s) of a layer.\n   *\n   * Only applicable if the layer has only one inbound node, or if all inbound\n   * nodes have the same output shape.\n   *\n   * @returns Output shape or shapes.\n   * @throws AttributeError: if the layer is connected to more than one incoming\n   *   nodes.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  get outputShape(): Shape|Shape[] {\n    if (this.inboundNodes == null || this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `The layer ${this.name} has never been called and thus has no ` +\n          `defined output shape.`);\n    }\n    const allOutputShapes: string[] = [];\n    for (const node of this.inboundNodes) {\n      const shapeString = JSON.stringify(node.outputShapes);\n      if (allOutputShapes.indexOf(shapeString) === -1) {\n        allOutputShapes.push(shapeString);\n      }\n    }\n    if (allOutputShapes.length === 1) {\n      const outputShapes = this.inboundNodes[0].outputShapes;\n      if (Array.isArray(outputShapes) && Array.isArray(outputShapes[0]) &&\n          outputShapes.length === 1) {\n        return (outputShapes as Shape[])[0];\n      } else {\n        return outputShapes;\n      }\n\n    } else {\n      throw new AttributeError(\n          `The layer ${this.name} has multiple inbound nodes with different ` +\n          `output shapes. Hence the notion of \"output shape\" is ill-defined ` +\n          `for the layer.`);\n      // TODO(cais): Implement getOutputShapeAt().\n    }\n  }\n\n  /**\n   * Counts the total number of numbers (e.g., float32, int32) in the\n   * weights.\n   *\n   * @returns An integer count.\n   * @throws RuntimeError: If the layer is not built yet (in which case its\n   *   weights are not defined yet.)\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  countParams(): number {\n    if (!this.built) {\n      throw new RuntimeError(\n          `You tried to call countParams() on ${this.name}, ` +\n          `but the layer is not built yet. Build it first by calling ` +\n          `build(batchInputShape).`);\n    }\n    return variable_utils.countParamsInWeights(this.weights);\n  }\n\n  /**\n   * Creates the layer weights.\n   *\n   * Must be implemented on all layers that have weights.\n   *\n   * Called when apply() is called to construct the weights.\n   *\n   * @param inputShape A `Shape` or array of `Shape` (unused).\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  build(inputShape: Shape|Shape[]) {\n    this.built = true;\n  }\n\n  /**\n   * Returns the current values of the weights of the layer.\n   *\n   * @param trainableOnly Whether to get the values of only trainable weights.\n   * @returns Weight values as an `Array` of `tf.Tensor`s.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getWeights(trainableOnly = false): Tensor[] {\n    return batchGetValue(trainableOnly ? this.trainableWeights : this.weights);\n  }\n\n  /**\n   * Sets the weights of the layer, from Tensors.\n   *\n   * @param weights a list of Tensors. The number of arrays and their shape\n   *   must match number of the dimensions of the weights of the layer (i.e.\n   *   it should match the output of `getWeights`).\n   *\n   * @exception ValueError If the provided weights list does not match the\n   *   layer's specifications.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  setWeights(weights: Tensor[]): void {\n    tidy(() => {\n      const params = this.weights;\n      if (params.length !== weights.length) {\n        // TODO(cais): Restore the following and use `providedWeights`, instead\n        // of `weights` in the error message, once the deeplearn.js bug is\n        // fixed: https://github.com/PAIR-code/deeplearnjs/issues/498 const\n        // providedWeights = JSON.stringify(weights).substr(0, 50);\n        throw new ValueError(\n            `You called setWeights(weights) on layer \"${this.name}\" ` +\n            `with a weight list of length ${weights.length}, ` +\n            `but the layer was expecting ${params.length} weights. ` +\n            `Provided weights: ${weights}...`);\n      }\n      if (params.length === 0) {\n        return;\n      }\n      const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n      const paramValues = batchGetValue(params);\n      for (let i = 0; i < paramValues.length; ++i) {\n        const pv = paramValues[i];\n        const p = params[i];\n        const w = weights[i];\n        if (!util.arraysEqual(pv.shape, w.shape)) {\n          throw new ValueError(\n              `Layer weight shape ${pv.shape} ` +\n              `not compatible with provided weight shape ${w.shape}`);\n        }\n        weightValueTuples.push([p, w]);\n      }\n      batchSetValue(weightValueTuples);\n    });\n  }\n\n  /**\n   * Adds a weight variable to the layer.\n   *\n   * @param name Name of the new weight variable.\n   * @param shape The shape of the weight.\n   * @param dtype The dtype of the weight.\n   * @param initializer An initializer instance.\n   * @param regularizer A regularizer instance.\n   * @param trainable Whether the weight should be trained via backprop or not\n   *   (assuming that the layer itself is also trainable).\n   * @param constraint An optional trainable.\n   * @return The created weight variable.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  protected addWeight(\n      name: string, shape: Shape, dtype?: DataType, initializer?: Initializer,\n      regularizer?: Regularizer, trainable?: boolean,\n      constraint?: Constraint): LayerVariable {\n    // Reject duplicate weight names.\n    if (this._addedWeightNames.indexOf(name) !== -1) {\n      throw new ValueError(\n          `Duplicate weight name ${name} for layer ${this.name}`);\n    }\n    this._addedWeightNames.push(name);\n\n    if (dtype == null) {\n      dtype = 'float32';\n    }\n\n    if (this.fastWeightInitDuringBuild) {\n      initializer = getInitializer('zeros');\n    }\n    const initValue = initializer.apply(shape, dtype);\n    const weight =\n        new LayerVariable(initValue, dtype, name, trainable, constraint);\n    initValue.dispose();\n    // Request backend not to dispose the weights of the model on scope() exit.\n    if (regularizer != null) {\n      this.addLoss(() => regularizer.apply(weight.read()));\n    }\n    if (trainable == null) {\n      trainable = true;\n    }\n    if (trainable) {\n      this._trainableWeights.push(weight);\n    } else {\n      this._nonTrainableWeights.push(weight);\n    }\n    return weight;\n  }\n\n  /**\n   * Set the fast-weight-initialization flag.\n   *\n   * In cases where the initialized weight values will be immediately\n   * overwritten by loaded weight values during model loading, setting\n   * the flag to `true` saves unnecessary calls to potentially expensive\n   * initializers and speeds up the loading process.\n   *\n   * @param value Target value of the flag.\n   */\n  setFastWeightInitDuringBuild(value: boolean) {\n    this.fastWeightInitDuringBuild = value;\n  }\n\n  /**\n   * Add losses to the layer.\n   *\n   * The loss may potentionally be conditional on some inputs tensors,\n   * for instance activity losses are conditional on the layer's inputs.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  addLoss(losses: RegularizerFn|RegularizerFn[]): void {\n    if (losses == null || Array.isArray(losses) && losses.length === 0) {\n      return;\n    }\n    // Update this.losses\n    losses = generic_utils.toList(losses);\n    if (this._losses !== undefined && this._losses !== null) {\n      this.losses.push(...losses);\n    }\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (!this.supportsMasking) {\n      if (mask != null) {\n        if (Array.isArray(mask)) {\n          mask.forEach(maskElement => {\n            if (maskElement != null) {\n              throw new TypeError(\n                  `Layer ${this.name} does not support masking, ` +\n                  'but was passed an inputMask.');\n            }\n          });\n        } else {\n          throw new TypeError(\n              `Layer ${this.name} does not support masking, ` +\n              'but was passed an inputMask.');\n        }\n      }\n      // masking not explicitly supported: return null as mask\n      return null;\n    }\n    // if masking is explictly supported, by default\n    // carry over the input mask\n    return mask;\n  }\n\n  /**\n   * Internal method to create an inbound node for the layer.\n   *\n   * @param inputTensors List of input tensors.\n   * @param outputTensors List of output tensors.\n   * @param inputMasks List of input masks (a mask can be a tensor, or null).\n   * @param outputMasks List of output masks (a mask can be a tensor, or null).\n   * @param inputShapes List of input shape tuples.\n   * @param outputShapes List of output shape tuples.\n   * @param kwargs Dictionary of keyword arguments that were passed to the\n   *   `call` method of the layer at the call that created the node.\n   */\n  private addInboundNode(\n      inputTensors: SymbolicTensor|SymbolicTensor[],\n      outputTensors: SymbolicTensor|SymbolicTensor[],\n      inputMasks: Tensor|Tensor[], outputMasks: Tensor|Tensor[],\n      inputShapes: Shape|Shape[], outputShapes: Shape|Shape[],\n      kwargs: {} = null): void {\n    const inputTensorList: SymbolicTensor[] =\n        generic_utils.toList(inputTensors);\n    outputTensors = generic_utils.toList(outputTensors);\n    inputMasks = generic_utils.toList(inputMasks);\n    outputMasks = generic_utils.toList(outputMasks);\n    inputShapes = types_utils.normalizeShapeList(inputShapes);\n    outputShapes = types_utils.normalizeShapeList(outputShapes);\n\n    // Collect input tensor(s) coordinates.\n    const inboundLayers: Layer[] = [];\n    const nodeIndices: number[] = [];\n    const tensorIndices: number[] = [];\n    for (const x of inputTensorList) {\n      /*\n       * TODO(michaelterry): Keras adds this value to tensors; it's not\n       * clear whether we'll use this or not.\n       */\n      inboundLayers.push(x.sourceLayer);\n      nodeIndices.push(x.nodeIndex);\n      tensorIndices.push(x.tensorIndex);\n    }\n\n    // Create node, add it to inbound nodes.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node(\n        {\n          outboundLayer: this,\n          inboundLayers,\n          nodeIndices,\n          tensorIndices,\n          inputTensors: inputTensorList,\n          outputTensors,\n          inputMasks,\n          outputMasks,\n          inputShapes,\n          outputShapes\n        },\n        kwargs);\n\n    // Update tensor history\n    for (let i = 0; i < outputTensors.length; i++) {\n      // TODO(michaelterry: _uses_learning_phase not tracked.\n      outputTensors[i].sourceLayer = this;\n      outputTensors[i].nodeIndex = this.inboundNodes.length - 1;\n      outputTensors[i].tensorIndex = i;\n    }\n  }\n\n  /**\n   * Returns the config of the layer.\n   *\n   * A layer config is a TS dictionary (serializable)\n   * containing the configuration of a layer.\n   * The same layer can be reinstantiated later\n   * (without its trained weights) from this configuration.\n   *\n   * The config of a layer does not include connectivity\n   * information, nor the layer class name.  These are handled\n   * by 'Container' (one layer of abstraction above).\n   *\n   * Porting Note: The TS dictionary follows TS naming standrds for\n   * keys, and uses tfjs-layers type-safe Enums.  Serialization methods\n   * should use a helper function to convert to the pythonic storage\n   * standard. (see serialization_utils.convertTsToPythonic)\n   *\n   * @returns TS dictionary of configuration.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getConfig(): serialization.ConfigDict {\n    const config:\n        serialization.ConfigDict = {name: this.name, trainable: this.trainable};\n    if (this.batchInputShape != null) {\n      config['batchInputShape'] = this.batchInputShape;\n    }\n    if (this.dtype != null) {\n      config['dtype'] = this.dtype;\n    }\n    return config;\n  }\n\n  /**\n   * Dispose the weight variables that this Layer instance holds.\n   *\n   * @returns {number} Number of disposed variables.\n   */\n  protected disposeWeights(): number {\n    this.weights.forEach(weight => weight.dispose());\n    return this.weights.length;\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Layer '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose layer's weights.\n   *\n   * This method decrease the reference count of the Layer object by 1.\n   *\n   * A Layer is reference-counted. Its reference count is incremented by 1\n   * the first item its `apply()` method is called and when it becomes a part\n   * of a new `Node` (through calling the `apply()`) method on a\n   * `tf.SymbolicTensor`).\n   *\n   * If the reference count of a Layer becomes 0, all the weights will be\n   * disposed and the underlying memory (e.g., the textures allocated in WebGL)\n   * will be freed.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * weights of the Layer will *not* be disposed.\n   *\n   * After a Layer is disposed, it cannot be used in calls such as `apply()`,\n   * `getWeights()` or `setWeights()` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the Container after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the layer has already\n   *   been disposed.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  dispose(): DisposeResult {\n    if (!this.built) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been ` +\n          `built yet.`);\n    }\n\n    if (this._refCount === null) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been used ` +\n          `yet.`);\n    }\n\n    this.assertNotDisposed();\n\n    let numDisposedVariables = 0;\n    if (--this._refCount === 0) {\n      numDisposedVariables = this.disposeWeights();\n    }\n\n    return {refCountAfterDispose: this._refCount, numDisposedVariables};\n  }\n}\n\n/**\n * Collects the input shape(s) of a list of `tf.Tensor`s or\n * `tf.SymbolicTensor`s.\n *\n * TODO(michaelterry): Update PyKeras docs (backport).\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return List of shape tuples (or single tuple), one tuple per input.\n */\nfunction collectInputShape(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                           Tensor[]): Shape|Shape[] {\n  inputTensors =\n      generic_utils.toList(inputTensors) as SymbolicTensor[] | Tensor[];\n  const shapes: Shape[] = [];\n  for (const x of inputTensors) {\n    shapes.push(x.shape);\n  }\n  return generic_utils.singletonOrArray(shapes);\n}\n\n/**\n * Guesses output dtype based on inputs.\n *\n * At present, just returns 'float32' for any input.\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return The guessed DType. At present, always returns 'float32'.\n */\nfunction guessOutputDType(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                          Tensor[]): DataType {\n  return 'float32';\n}\n\n/**\n * Returns the list of input tensors necessary to compute `tensor`.\n *\n * Output will always be a list of tensors (potentially with 1 element).\n *\n * @param tensor The tensor to start from.\n * @param layer Origin layer of the tensor.\n * @param nodeIndex Origin node index of the tensor.\n *\n * @return Array of input tensors.\n */\nexport function getSourceInputs(\n    tensor: SymbolicTensor, layer?: Layer,\n    nodeIndex?: number): SymbolicTensor[] {\n  if (layer == null || (nodeIndex != null && nodeIndex > 0)) {\n    layer = tensor.sourceLayer;\n    nodeIndex = tensor.nodeIndex;\n  }\n  if (layer.inboundNodes.length === 0) {\n    return [tensor];\n  } else {\n    const node = layer.inboundNodes[nodeIndex];\n    if (node.inboundLayers.length === 0) {\n      return node.inputTensors;\n    } else {\n      const sourceTensors: SymbolicTensor[] = [];\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const x = node.inputTensors[i];\n        const layer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const previousSources = getSourceInputs(x, layer, nodeIndex);\n        // Avoid input redundancy.\n        for (const x of previousSources) {\n          if (sourceTensors.indexOf(x) === -1) {\n            sourceTensors.push(x);\n          }\n        }\n      }\n      return sourceTensors;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\n\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/**\n * Constructor arguments for InputLayer.\n *\n * Note: You should provide only inputShape or batchInputShape (not both).\n * If only inputShape is provided, then the batchInputShape is determined by\n * the batchSize argument and the inputShape: [batchSize].concat(inputShape).\n */\nexport declare interface InputLayerArgs {\n  /** Input shape, not including the batch axis. */\n  inputShape?: Shape;\n  /** Optional input batch size (integer or null). */\n  batchSize?: number;\n  /** Batch input shape, including the batch axis. */\n  batchInputShape?: Shape;\n  /** Datatype of the input.  */\n  dtype?: DataType;\n  /**\n   * Whether the placeholder created is meant to be sparse.\n   */\n  sparse?: boolean;  // TODO(michaelterry): Not clear whether we'll need this.\n\n  /** Name of the layer. */\n  name?: string;\n}\n\nexport class InputLayer extends Layer {\n  /** @nocollapse */\n  static readonly className = 'InputLayer';\n  sparse: boolean;\n  constructor(args: InputLayerArgs) {\n    super({\n      dtype: args.dtype,\n      name: args.name != null ? args.name : getUid('input').toString()\n    });\n    // Normalize config.batchSize and config.sparse\n    if (args.batchSize == null) {\n      args.batchSize = null;\n    }\n    if (args.sparse == null) {\n      args.sparse = false;\n    }\n\n    this.trainable = false;\n    this.built = true;\n    this.sparse = args.sparse;\n\n    if (args.inputShape != null && args.batchInputShape != null) {\n      throw new ValueError(\n          'Only provide the inputShape OR ' +\n          'batchInputShape argument to inputLayer, not both at the same time.');\n    }\n    let batchInputShape = args.batchInputShape;\n    if (batchInputShape == null) {\n      if (args.inputShape == null) {\n        throw new ValueError(\n            'An InputLayer should be passed either a ' +\n            '`batchInputShape` or an `inputShape`.');\n      } else {\n        batchInputShape = [args.batchSize].concat(args.inputShape);\n      }\n    } else {\n      // TODO(michaelterry): Backport to PyKeras\n      if (args.batchSize != null) {\n        throw new ValueError(\n            'Cannot specify batchSize if batchInputShape is ' +\n            'specified when creating an InputLayer.');\n      }\n    }\n\n    const dtype = args.dtype || 'float32';\n\n    this.batchInputShape = batchInputShape;\n    this.dtype = dtype;\n    // TODO(michaelterry): Backport this to PyKeras?\n    this.inputSpec = [{shape: batchInputShape}];\n\n    const inputTensor = new SymbolicTensor(\n        this.dtype, this.batchInputShape, this, [], {}, this.name);\n    inputTensor.nodeIndex = 0;\n    inputTensor.tensorIndex = 0;\n\n    // Create an input node to add to this.outboundNode.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: [inputTensor],\n      outputTensors: [inputTensor],\n      inputMasks: [null],\n      outputMasks: [null],\n      inputShapes: [batchInputShape],\n      outputShapes: [batchInputShape]\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor {\n    throw new ValueError(\n        'Cannot pass any input to an ' +\n        `InputLayer's apply() method. InputLayer name: ${this.name}`);\n  }\n\n  dispose(): DisposeResult {\n    // dispose() for InputLayer is overridden as no-op.\n    return {refCountAfterDispose: this._refCount, numDisposedVariables: 0};\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      batchInputShape: this.batchInputShape,\n      dtype: this.dtype,\n      sparse: this.sparse,\n      name: this.name\n    };\n  }\n}\nserialization.registerClass(InputLayer);\n\n/**\n * Config for the Input function.\n *\n * Note: You should provide only shape or batchShape (not both).\n * If only shape is provided, then the batchShape becomes\n * [null].concat(inputShape).\n */\nexport interface InputConfig {\n  /**\n   * A shape, not including the batch size. For instance, `shape=[32]`\n   * indicates that the expected input will be batches of 32-dimensional\n   * vectors.\n   */\n  shape?: Shape;\n  /**\n   * A shape tuple (integer), including the batch size. For instance,\n   * `batchShape=[10, 32]` indicates that the expected input will be batches of\n   * 10 32-dimensional vectors. `batchShape=[null, 32]` indicates batches of an\n   * arbitrary number of 32-dimensional vectors.\n   */\n  batchShape?: Shape;\n  /**\n   * An optional name string for the layer. Should be unique in a model (do not\n   * reuse the same name twice). It will be autogenerated if it isn't provided.\n   */\n  name?: string;\n  dtype?: DataType;\n  /**\n   * A boolean specifying whether the placeholder to be created is sparse.\n   */\n  sparse?: boolean;\n}\n\nexport function Input(config: InputConfig): SymbolicTensor {\n  if (config.batchShape == null && config.shape == null) {\n    throw new Error(\n        'Please provide to Input either a `shape`' +\n        ' or a `batchShape` argument. Note that ' +\n        '`shape` does not include the batch ' +\n        'dimension.');\n  }\n  if (config.batchShape != null && config.shape != null) {\n    // TODO(michaelterry): Backport to PyKeras.\n    throw new ValueError(\n        'Please provide either a `shape` or `batchShape` ' +\n        'argument to Input, but not both.');\n  }\n  let batchShape = config.batchShape;\n  if (config.shape != null && batchShape == null) {\n    batchShape = [null].concat(config.shape);\n  }\n\n  let dtype = config.dtype;\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n\n  const inputLayer = new InputLayer({\n    batchInputShape: batchShape,\n    name: config.name,\n    dtype,\n    sparse: config.sparse\n  });\n\n  const outputs = inputLayer.inboundNodes[0].outputTensors;\n  return outputs[0];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, Scalar} from '@tensorflow/tfjs-core';\n\n/**\n * Logs in which values can be either numbers or Tensors (Scalars).\n *\n * Used internally.\n */\nexport type UnresolvedLogs = {\n  [key: string]: number|Scalar;\n};\n\n/**\n * Turn any Scalar values in a Logs object into actual number values.\n *\n * @param logs The `Logs` object to be resolved in place.\n */\nexport async function resolveScalarsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n  const keys: string[] = [];\n  const scalarsToDispose: Scalar[] = [];\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      const valueScalar = value;\n      promises.push(valueScalar.data());\n      keys.push(key);\n      scalarsToDispose.push(valueScalar);\n    }\n  }\n  if (promises.length > 0) {\n    const values = await Promise.all(promises);\n    for (let i = 0; i < values.length; ++i) {\n      logs[keys[i]] = values[i][0];\n    }\n    // Dispose the original scalar tensors.\n    dispose(scalarsToDispose);\n  }\n}\n\n/**\n * Dispose all Tensors in an UnresolvedLogs object.\n *\n * @param logs An `UnresolvedLogs` object potentially containing `tf.Tensor`s in\n *   places where the values can be `tf.Tensor` or `number`.\n */\nexport function disposeTensorsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      value.dispose();\n    }\n  }\n}\n\n/**\n * Logs in which values can only be numbers.\n *\n * Used when calling client-provided custom callbacks.\n */\nexport type Logs = {\n  [key: string]: number;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {add, div, keep, mul, nextFrame, Scalar, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Container} from './engine/container';\nimport {ValueError} from './errors';\nimport {Logs, resolveScalarsInLogs, UnresolvedLogs} from './logs';\nimport * as generic_utils from './utils/generic_utils';\n\n/** Verbosity logging level when fitting a model. */\nexport enum ModelLoggingVerbosity {\n  SILENT = 0,\n  VERBOSE = 1\n}\n\n/** How often to yield to the main thread when training (in ms). */\nexport const DEFAULT_YIELD_EVERY_MS = 125;\n\nexport type Params = {\n  [key: string]: number|string|boolean|number[]|string[]|boolean[];\n};\n\nexport type YieldEveryOptions = 'auto'|'batch'|'epoch'|'never'|number;\n\n/**\n * Abstract base class used to build new callbacks.\n *\n * The `logs` dictionary that callback methods take as argument will contain\n * keys for quantities relevant to the current batch or epoch.\n *\n * Currently, the `.fit()` method of the `Sequential` model class\n * will include the following quantities in the `logs` that\n * it passes to its callbacks:\n *\n * onEpochEnd: Logs include `acc` and `loss`, and optionally include `valLoss`\n *   (if validation is enabled in `fit`), and `valAcc` (if validation and\n *   accuracy monitoring are enabled).\n * onBatchBegin: Logs include `size`, the number of samples in the current\n *   batch.\n * onBatchEnd: Logs include `loss`, and optionally `acc` (if accuracy monitoring\n *   is enabled).\n */\nexport abstract class BaseCallback {\n  // TODO(michaelterry): This type is a best guess.\n  validationData: Tensor|Tensor[] = null;\n  /**\n   * Training parameters (eg. verbosity, batch size, number of epochs...).\n   */\n  params: Params;\n\n  setParams(params: Params): void {\n    this.params = params;\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {}\n\n  async onTrainBegin(logs?: UnresolvedLogs) {}\n\n  async onTrainEnd(logs?: UnresolvedLogs) {}\n\n  // LayersModel needs to call Callback.setModel(), but cannot actually depend\n  // on Callback because that creates a cyclic dependency.  Providing this no-op\n  // method on BaseCallback breaks the cycle: this way LayersModel can depend on\n  // BaseCallback but not on Callback.  The argument is typed as `Container`\n  // (the superclass of LayersModel) to avoid recapitulating the cycle. Callback\n  // overrides this method and enforces that the argument is really a\n  // LayersModel.\n  setModel(model: Container): void {\n    // Do nothing. Use Callback instead of BaseCallback to track the model.\n  }\n}\n\n/**\n * Container abstracting a list of callbacks.\n */\nexport class CallbackList {\n  callbacks: BaseCallback[];\n  queueLength: number;\n\n  // TODO(cais): When the need arises, uncomment the following lines and\n  // implement the queue for time values.\n  // private deltaTBatch: number;\n  // private deltaTsBatchBegin: Array<number>;\n  // private deltaTsBatchEnd: Array<number>;\n\n  /**\n   * Constructor of CallbackList.\n   * @param callbacks Array of `Callback` instances.\n   * @param queueLength Queue length for keeping running statistics over\n   *   callback execution time.\n   */\n  constructor(callbacks?: BaseCallback[], queueLength = 10) {\n    // TODO(cais): Make use of queueLength when implementing the queue for time\n    // values.\n    if (callbacks == null) {\n      callbacks = [];\n    }\n    this.callbacks = callbacks;\n    this.queueLength = queueLength;\n  }\n\n  append(callback: BaseCallback): void {\n    this.callbacks.push(callback);\n  }\n\n  setParams(params: Params): void {\n    for (const callback of this.callbacks) {\n      callback.setParams(params);\n    }\n  }\n\n  setModel(model: Container): void {\n    for (const callback of this.callbacks) {\n      callback.setModel(model);\n    }\n  }\n\n  /**\n   * Called at the start of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochBegin(epoch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochEnd(epoch, logs);\n    }\n  }\n\n  /**\n   * Called  right before processing a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchBegin(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchEnd(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the beginning of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainBegin(logs);\n    }\n  }\n\n  /**\n   * Called at the end of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainEnd(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainEnd(logs);\n    }\n  }\n}\n\n/**\n * Callback that accumulates epoch averages of metrics.\n *\n * This callback is automatically applied to every LayersModel.\n */\nexport class BaseLogger extends BaseCallback {\n  private seen: number;\n  private totals: UnresolvedLogs;\n\n  constructor() {\n    super();\n  }\n\n  async onEpochBegin(epoch: number) {\n    this.seen = 0;\n    this.totals = {};\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const batchSize = logs['size'] == null ? 0 : logs['size'] as number;\n    this.seen += batchSize;\n    for (const key in logs) {\n      const value = logs[key];\n      if (typeof value === 'number') {\n        if (!this.totals.hasOwnProperty(key)) {\n          this.totals[key] = 0;\n        }\n        this.totals[key] = this.totals[key] as number + value * batchSize;\n      } else {\n        let oldTotalsToDispose: Scalar;\n        if (key in this.totals) {\n          oldTotalsToDispose = this.totals[key] as Scalar;\n        } else {\n          this.totals[key] = 0;\n        }\n        const total: Scalar =\n            tidy(() => add((this.totals[key]), mul(value, batchSize)));\n        this.totals[key] = total;\n        if (oldTotalsToDispose != null) {\n          oldTotalsToDispose.dispose();\n        }\n      }\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs != null) {\n      for (const key of this.params['metrics'] as string[]) {\n        if (this.totals[key] == null) {\n          continue;\n        }\n        if (typeof this.totals[key] === 'number') {\n          logs[key] = this.totals[key] as number / this.seen;\n        } else {\n          tidy(() => {\n            const log: Scalar = mul(div(1, this.seen), this.totals[key]);\n            logs[key] = log;\n            (this.totals[key] as Tensor).dispose();\n            keep(logs[key] as Scalar);\n          });\n        }\n      }\n    }\n  }\n}\n\n/**\n * Callback that records events into a `History` object. This callback is\n * automatically applied to every TF.js Layers model. The `History` object\n * gets returned by the `fit` method of models.\n */\nexport class History extends BaseCallback {\n  epoch: number[];\n  history: {[key: string]: Array<number|Tensor>};\n\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    this.epoch = [];\n    this.history = {};\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    this.epoch.push(epoch);\n    for (const key in logs) {\n      if (this.history[key] == null) {\n        this.history[key] = [];\n      }\n      this.history[key].push(logs[key]);\n    }\n  }\n\n  /**\n   * Await the values of all losses and metrics.\n   */\n  async syncData() {\n    const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n    const keys: string[] = [];\n    const indices: number[] = [];\n    for (const key in this.history) {\n      const valueArray = this.history[key];\n      for (let i = 0; i < valueArray.length; ++i) {\n        if (typeof valueArray[i] !== 'number') {\n          const valueScalar = valueArray[i] as Tensor;\n          promises.push(valueScalar.data());\n          keys.push(key);\n          indices.push(i);\n        }\n      }\n    }\n    const values = await Promise.all(promises);\n    for (let n = 0; n < values.length; ++n) {\n      const tensorToDispose = this.history[keys[n]][indices[n]] as Tensor;\n      tensorToDispose.dispose();\n      this.history[keys[n]][indices[n]] = values[n][0];\n    }\n  }\n}\n\nexport interface CustomCallbackArgs {\n  onTrainBegin?: (logs?: Logs) => void | Promise<void>;\n  onTrainEnd?: (logs?: Logs) => void | Promise<void>;\n  onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n}\n\n/**\n * Custom callback for training.\n */\nexport class CustomCallback extends BaseCallback {\n  protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;\n  protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;\n  protected readonly epochBegin:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly epochEnd:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchBegin:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchEnd:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly yield:\n      (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n\n  private yieldEvery: YieldEveryOptions;\n  private currentEpoch = 0;\n\n  constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions) {\n    super();\n    this.yieldEvery = yieldEvery || 'auto';\n    if (this.yieldEvery === 'auto') {\n      this.yieldEvery = DEFAULT_YIELD_EVERY_MS;\n    }\n    if (this.yieldEvery === 'never' && args.onYield != null) {\n      throw new Error(\n          'yieldEvery is `never` but you provided an `onYield` callback. ' +\n          'Either change `yieldEvery` or remove the callback');\n    }\n    if (util.isNumber(this.yieldEvery)) {\n      // Decorate `maybeWait` so it will be called at most once every\n      // `yieldEvery` ms.\n      this.maybeWait = generic_utils.debounce(\n          this.maybeWait.bind(this), this.yieldEvery as number);\n    }\n    this.trainBegin = args.onTrainBegin;\n    this.trainEnd = args.onTrainEnd;\n    this.epochBegin = args.onEpochBegin;\n    this.epochEnd = args.onEpochEnd;\n    this.batchBegin = args.onBatchBegin;\n    this.batchEnd = args.onBatchEnd;\n    this.yield = args.onYield;\n  }\n\n  async maybeWait(epoch: number, batch: number, logs: UnresolvedLogs) {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.yield != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.yield(epoch, batch, logs as Logs));\n    }\n    ps.push(nextFrame());\n    await Promise.all(ps);\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    this.currentEpoch = epoch;\n    if (this.epochBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.epochBegin(epoch, logs as Logs);\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.epochEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.epochEnd(epoch, logs as Logs));\n    }\n    if (this.yieldEvery === 'epoch') {\n      ps.push(nextFrame());\n    }\n    await Promise.all(ps);\n  }\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    if (this.batchBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.batchBegin(batch, logs as Logs);\n    }\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.batchEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.batchEnd(batch, logs as Logs));\n    }\n    if (this.yieldEvery === 'batch') {\n      ps.push(nextFrame());\n    } else if (util.isNumber(this.yieldEvery)) {\n      ps.push(this.maybeWait(this.currentEpoch, batch, logs));\n    }\n    await Promise.all(ps);\n  }\n\n  async onTrainBegin(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainBegin(logs as Logs);\n    }\n  }\n\n  async onTrainEnd(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainEnd != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainEnd(logs as Logs);\n    }\n  }\n}\n\n/**\n * Standardize callbacks or configurations of them to an Array of callbacks.\n */\nexport function standardizeCallbacks(\n    callbacks: BaseCallback|BaseCallback[]|CustomCallbackArgs|\n    CustomCallbackArgs[],\n    yieldEvery: YieldEveryOptions): BaseCallback[] {\n  if (callbacks == null) {\n    callbacks = {} as BaseCallback;\n  }\n  if (callbacks instanceof BaseCallback) {\n    return [callbacks];\n  }\n  if (Array.isArray(callbacks) && callbacks[0] instanceof BaseCallback) {\n    return callbacks as BaseCallback[];\n  }\n  // Convert custom callback configs to custom callback objects.\n  const callbackConfigs =\n      generic_utils.toList(callbacks) as CustomCallbackArgs[];\n  return callbackConfigs.map(\n      callbackConfig => new CustomCallback(callbackConfig, yieldEvery));\n}\n\nexport declare type BaseCallbackConstructor = {\n  new (): BaseCallback\n};\n\n/**\n * A global registry for callback constructors to be used during\n * LayersModel.fit().\n */\nexport class CallbackConstructorRegistry {\n  private static constructors:\n      {[verbosityLevel: number]: BaseCallbackConstructor[]} = {};\n\n  /**\n   * Blocks public access to constructor.\n   */\n  private constructor() {}\n\n  /**\n   * Register a tf.LayersModel.fit() callback constructor.\n   *\n   * The registered callback constructor will be used to instantiate\n   * callbacks for every tf.LayersModel.fit() call afterwards.\n   *\n   * @param verbosityLevel Level of verbosity at which the `callbackConstructor`\n   *   is to be reigstered.\n   * @param callbackConstructor A no-arg constructor for `tf.Callback`.\n   * @throws Error, if the same callbackConstructor has been registered before,\n   *   either at the same or a different `verbosityLevel`.\n   */\n  static registerCallbackConstructor(\n      verbosityLevel: number, callbackConstructor: BaseCallbackConstructor) {\n    util.assert(\n        verbosityLevel >= 0 && Number.isInteger(verbosityLevel),\n        () => `Verbosity level is expected to be an integer >= 0, ` +\n            `but got ${verbosityLevel}`);\n    CallbackConstructorRegistry.checkForDuplicate(callbackConstructor);\n    if (CallbackConstructorRegistry.constructors[verbosityLevel] == null) {\n      CallbackConstructorRegistry.constructors[verbosityLevel] = [];\n    }\n    CallbackConstructorRegistry.constructors[verbosityLevel].push(\n        callbackConstructor);\n  }\n\n  private static checkForDuplicate(callbackConstructor:\n                                       BaseCallbackConstructor) {\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const constructors = CallbackConstructorRegistry.constructors[+levelName];\n      constructors.forEach(ctor => {\n        if (ctor === callbackConstructor) {\n          throw new ValueError('Duplicate callback constructor.');\n        }\n      });\n    }\n  }\n\n  /**\n   * Clear all registered callback constructors.\n   */\n  protected static clear() {\n    CallbackConstructorRegistry.constructors = {};\n  }\n\n  /**\n   * Create callbacks using the registered callback constructors.\n   *\n   * Given `verbosityLevel`, all constructors registered at that level or above\n   * will be called and the instantiated callbacks will be used.\n   *\n   * @param verbosityLevel: Level of verbosity.\n   */\n  static createCallbacks(verbosityLevel: number): BaseCallback[] {\n    const constructors: BaseCallbackConstructor[] = [];\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const level = +levelName;\n      if (verbosityLevel >= level) {\n        constructors.push(...CallbackConstructorRegistry.constructors[level]);\n      }\n    }\n    return constructors.map(ctor => new ctor());\n  }\n}\n\nexport function configureCallbacks(\n    callbacks: BaseCallback[], verbose: ModelLoggingVerbosity, epochs: number,\n    initialEpoch: number, numTrainSamples: number, stepsPerEpoch: number,\n    batchSize: number, doValidation: boolean,\n    callbackMetrics: string[]): {callbackList: CallbackList, history: History} {\n  const history = new History();\n  const actualCallbacks: BaseCallback[] = [\n    new BaseLogger(), ...CallbackConstructorRegistry.createCallbacks(verbose)\n  ];\n  if (callbacks != null) {\n    actualCallbacks.push(...callbacks);\n  }\n  actualCallbacks.push(history);\n  const callbackList = new CallbackList(actualCallbacks);\n\n  // TODO(cais): Figure out when this LayersModel instance can have a\n  // dynamically\n  //   set property called 'callback_model' as in PyKeras.\n\n  callbackList.setParams({\n    epochs,\n    initialEpoch,\n    samples: numTrainSamples,\n    steps: stepsPerEpoch,\n    batchSize,\n    verbose,\n    doValidation,\n    metrics: callbackMetrics,\n  });\n  return {callbackList, history};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source layers/__init__.py */\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {deserializeKerasObject} from '../utils/generic_utils';\n\n/**\n * Instantiate a layer from a config dictionary.\n * @param config dict of the form {class_name: str, config: dict}\n * @param customObjects dict mapping class names (or function names)\n *   of custom (non-Keras) objects to class/functions\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns Layer instance (may be LayersModel, Sequential, Layer...)\n */\nexport function deserialize(\n    config: serialization.ConfigDict,\n    customObjects = {} as serialization.ConfigDict,\n    fastWeightInit = false): serialization.Serializable {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'layer', fastWeightInit);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: losses.py */\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\nimport * as K from './backend/tfjs_backend';\nimport {ValueError} from './errors';\nimport {LossOrMetricFn} from './types';\n\n/**\n * Normalizes a tensor wrt the L2 norm alongside the specified axis.\n * @param x\n * @param axis Axis along which to perform normalization.\n */\nexport function l2Normalize(x: Tensor, axis?: number): Tensor {\n  return tidy(() => {\n    if (x.dtype !== 'float32') {\n      x = x.asType('float32');\n    }\n    const squareSum = tfc.sum(K.square(x), axis, true);\n    const epsilonTensor = tfc.fill(squareSum.shape, epsilon());\n    const norm = tfc.sqrt(tfc.maximum(squareSum, epsilonTensor));\n    return tfc.div(x, norm);\n  });\n}\n\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(K.square(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(tfc.abs(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const diff = tfc.sub(yTrue, yPred);\n    const clippedTrue =\n        tfc.clipByValue(tfc.abs(yTrue), epsilon(), Number.MAX_VALUE);\n    const absResult = tfc.abs(tfc.div(diff, clippedTrue));\n    return tfc.mul(100, tfc.mean(absResult, -1));\n  });\n}\n\nexport function meanSquaredLogarithmicError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), Number.MAX_VALUE);\n    const firstLog = tfc.log(tfc.add(1, clippedPred));\n\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), Number.MAX_VALUE);\n    const secondLog = tfc.log(tfc.add(1, clippedTrue));\n\n    return tfc.mean(K.square(tfc.sub(firstLog, secondLog)), -1);\n  });\n}\n\nexport function squaredHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(K.square(maxResult), -1);\n  });\n}\n\nexport function hinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(maxResult, -1);\n  });\n}\n\nexport function categoricalHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const pos = tfc.sum(tfc.mul(yTrue, yPred), -1);\n    const neg = tfc.max(tfc.mul(tfc.sub(1, yTrue), yPred), -1);\n    return tfc.maximum(0, tfc.add(1, tfc.sub(neg, pos)));\n  });\n}\n\n/**\n * Logarithm of the hyperbolic cosine of the prediction error.\n *\n * `log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and\n * to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly\n * like the mean squared error, but will not be so strongly affected by the\n * occasional wildly incorrect prediction.\n */\nexport function logcosh(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const log2 = Math.log(2);\n    const predictionDiff = tfc.sub(yPred, yTrue);\n    const logcoshResult = tfc.sub(\n        tfc.add(predictionDiff, tfc.softplus(tfc.mul(-2, predictionDiff))),\n        log2);\n    return tfc.mean(logcoshResult, -1);\n  });\n}\n\nexport function categoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    if (fromLogits) {\n      output = tfc.softmax(output);\n    } else {\n      // scale preds so that the class probabilities of each sample sum to 1.\n      const outputSum = tfc.sum(output, output.shape.length - 1, true);\n      output = tfc.div(output, outputSum);\n    }\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    return tfc.neg(tfc.sum(\n        tfc.mul(target.toFloat(), tfc.log(output)), output.shape.length - 1));\n  });\n}\n\n/**\n * Categorical crossentropy with integer targets.\n *\n * @param target An integer tensor.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n */\nexport function sparseCategoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    const flatTarget = tfc.floor(K.flatten(target)).toInt() as Tensor1D;\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    const outputShape = output.shape;\n    const oneHotTarget =\n        tfc.oneHot(flatTarget, outputShape[outputShape.length - 1])\n            .reshape(outputShape);\n    return categoricalCrossentropy(oneHotTarget, output, fromLogits);\n  });\n}\n\n/**\n * From TensorFlow's implementation in nn_impl.py:\n *\n * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n *      z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n *    = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n *    = (1 - z) * x + log(1 + exp(-x))\n *    = x - x * z + log(1 + exp(-x))\n * For x < 0, to avoid overflow in exp(-x), we reformulate the above\n *      x - x * z + log(1 + exp(-x))\n *    = log(exp(x)) - x * z + log(1 + exp(-x))\n *    = - x * z + log(1 + exp(x))\n * Hence, to ensure stability and avoid overflow, the implementation uses this\n * equivalent formulation\n *    max(x, 0) - x * z + log(1 + exp(-abs(x)))\n *\n * @param labels The labels.\n * @param logits The logits.\n */\nexport function sigmoidCrossEntropyWithLogits(\n    labels: Tensor, logits: Tensor): Tensor {\n  if (!util.arraysEqual(labels.shape, logits.shape)) {\n    throw new ValueError(\n        `logits and labels must have the same shape, but got shapes ` +\n        `${JSON.stringify(labels.shape)} and ${JSON.stringify(logits.shape)}`);\n  }\n  return tidy(() => {\n    // The logistic loss formula from above is\n    //   x - x * z + log(1 + exp(-x))\n    // For x < 0, a more numerically stable formula is\n    //   -x * z + log(1 + exp(x))\n    // Note that these two expressions can be combined into the following:\n    //   max(x, 0) - x * z + log(1 + exp(-abs(x)))\n    const reluLogits = logits.relu();\n    const negAbsLogits = logits.abs().neg();\n    return reluLogits.sub(logits.mul(labels)).add(negAbsLogits.exp().log1p());\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    let y: Tensor;\n    y = tfc.clipByValue(yPred, epsilon(), 1 - epsilon());\n    y = tfc.log(tfc.div(y, tfc.sub(1, y)));\n    return tfc.mean(sigmoidCrossEntropyWithLogits(yTrue, y), -1);\n  });\n}\n\nexport function kullbackLeiblerDivergence(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), 1);\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), 1);\n    return tfc.sum(\n        tfc.mul(yTrue, tfc.log(tfc.div(clippedTrue, clippedPred))), -1);\n  });\n}\n\nexport function poisson(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const logPred = tfc.log(tfc.add(epsilon(), yPred));\n    return tfc.mean(tfc.sub(yPred, tfc.mul(yTrue, logPred)), -1);\n  });\n}\n\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const trueNormalized = l2Normalize(yTrue, -1);\n    const predNormalized = l2Normalize(yPred, -1);\n    const trueXPred = tfc.mul(trueNormalized, predNormalized);\n    return tfc.neg(tfc.sum(trueXPred, -1));\n  });\n}\n\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const msle = meanSquaredLogarithmicError;\nexport const MSLE = meanSquaredLogarithmicError;\nexport const kld = kullbackLeiblerDivergence;\nexport const KLD = kullbackLeiblerDivergence;\nexport const cosine = cosineProximity;\n\n// TODO(michaelterry): Add deserialize() function.\n\nexport const lossesMap: {[functionName: string]: LossOrMetricFn} = {\n  meanSquaredError,\n  meanAbsoluteError,\n  meanAbsolutePercentageError,\n  meanSquaredLogarithmicError,\n  squaredHinge,\n  hinge,\n  categoricalHinge,\n  logcosh,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  binaryCrossentropy,\n  kullbackLeiblerDivergence,\n  poisson,\n  cosineProximity\n};\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function get(identifierOrFn: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifierOrFn === 'string') {\n    if (identifierOrFn in lossesMap) {\n      return lossesMap[identifierOrFn];\n    }\n    let errMsg = `Unknown loss ${identifierOrFn}`;\n    if (identifierOrFn.toLowerCase().includes('softmaxcrossentropy')) {\n      errMsg = `Unknown loss ${identifierOrFn}. ` +\n          'Use \"categoricalCrossentropy\" as the string name for ' +\n          'tf.losses.softmaxCrossEntropy';\n    }\n    throw new ValueError(errMsg);\n  } else {\n    return identifierOrFn;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Built-in metrics.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {NotImplementedError, ValueError} from './errors';\nimport {categoricalCrossentropy as categoricalCrossentropyLoss, cosineProximity, meanAbsoluteError, meanAbsolutePercentageError, meanSquaredError, sparseCategoricalCrossentropy as sparseCategoricalCrossentropyLoss} from './losses';\nimport {binaryCrossentropy as lossBinaryCrossentropy} from './losses';\nimport {lossesMap} from './losses';\nimport {LossOrMetricFn} from './types';\nimport * as util from './utils/generic_utils';\n\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const threshold = tfc.mul(.5, tfc.onesLike(yPred));\n    const yPredThresholded = K.cast(tfc.greater(yPred, threshold), yTrue.dtype);\n    return tfc.mean(tfc.equal(yTrue, yPredThresholded), -1);\n  });\n}\n\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(\n      () => K.cast(\n          tfc.equal(tfc.argMax(yTrue, -1), tfc.argMax(yPred, -1)), 'float32'));\n}\n\nfunction truePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nfunction falseNegatives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(0)).sum().cast('float32');\n  });\n}\n\nfunction falsePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(0), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fp = falsePositives(yTrue, yPred);\n\n    const denominator = tp.add(fp);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fn = falseNegatives(yTrue, yPred);\n\n    const denominator = tp.add(fn);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return lossBinaryCrossentropy(yTrue, yPred);\n}\n\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  if (yTrue.rank === yPred.rank) {\n    yTrue = yTrue.squeeze([yTrue.rank - 1]);\n  }\n  yPred = yPred.argMax(-1);\n  if (yPred.dtype !== yTrue.dtype) {\n    yPred = yPred.asType(yTrue.dtype);\n  }\n  return tfc.equal(yTrue, yPred).asType('float32');\n}\n\nexport function topKCategoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\nexport function sparseTopKCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\n// Aliases.\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const categoricalCrossentropy = categoricalCrossentropyLoss;\nexport const cosine = cosineProximity;\nexport const sparseCategoricalCrossentropy = sparseCategoricalCrossentropyLoss;\n\n// TODO(cais, nielsene): Add serialize().\n\nexport const metricsMap: {[functionName: string]: LossOrMetricFn} = {\n  binaryAccuracy,\n  categoricalAccuracy,\n  precision,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  mse,\n  MSE,\n  mae,\n  MAE,\n  mape,\n  MAPE,\n  cosine\n};\n\nexport function get(identifier: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifier === 'string' && identifier in metricsMap) {\n    return metricsMap[identifier];\n  } else if (typeof identifier !== 'string' && identifier != null) {\n    return identifier;\n  } else {\n    throw new ValueError(`Unknown metric ${identifier}`);\n  }\n}\n\n/**\n * Get the shortcut function name.\n *\n * If the fn name is a string,\n *   directly return the string name.\n * If the function is included in metricsMap or lossesMap,\n *   return key of the map.\n *   - If the function relative to multiple keys,\n *     return the first found key as the function name.\n *   - If the function exists in both lossesMap and metricsMap,\n *     search lossesMap first.\n * If the function is not included in metricsMap or lossesMap,\n *   return the function name.\n *\n * @param fn loss function, metric function, or short cut name.\n * @returns Loss or Metric name in string.\n */\nexport function getLossOrMetricName(fn: string|LossOrMetricFn): string {\n  util.assert(fn !== null, `Unknown LossOrMetricFn ${fn}`);\n  if (typeof fn === 'string') {\n    return fn;\n  } else {\n    let fnName;\n    for (const key of Object.keys(lossesMap)) {\n      if (lossesMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    for (const key of Object.keys(metricsMap)) {\n      if (metricsMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    return (fn as Function).name;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Optimizers.\n */\n\nimport {Optimizer, train} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\n\nimport {ValueError} from './errors';\n\n// Add (de)serialize()\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function getOptimizer(identifier: string): Optimizer {\n  const optimizerMap: {[optimizerName: string]: () => Optimizer} = {\n    'Adagrad': () => train.adagrad(0.01),\n    'Adadelta': () => train.adadelta(1, 0.95, epsilon()),\n    'Adam': () => train.adam(0.001, 0.9, 0.999, epsilon()),\n    'Adamax': () => train.adamax(0.002, 0.9, 0.999, epsilon(), 0),\n    'RMSProp': () => train.rmsprop(0.001, 0.9, 0, epsilon()),\n    'SGD': () => train.sgd(0.01)\n  };\n  optimizerMap['adagrad'] = optimizerMap['Adagrad'];\n  optimizerMap['adadelta'] = optimizerMap['Adadelta'];\n  optimizerMap['adam'] = optimizerMap['Adam'];\n  optimizerMap['adamax'] = optimizerMap['Adamax'];\n  optimizerMap['rmsprop'] = optimizerMap['RMSProp'];\n  optimizerMap['sgd'] = optimizerMap['SGD'];\n\n  if (identifier in optimizerMap) {\n    return optimizerMap[identifier]();\n  }\n  throw new ValueError(`Unknown Optimizer ${identifier}`);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/** Utility functions related to user-defined metadata. */\n\n// Maximum recommended serialized size for user-defined metadata.\n// Beyond this limit, a warning message will be printed during model loading and\n// saving.\nexport const MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH = 1 * 1024 * 1024;\n\n/**\n * Check validity of user-defined metadata.\n *\n * @param userDefinedMetadata\n * @param modelName Name of the model that the user-defined metadata belongs to.\n *   Used during construction of error messages.\n * @param checkSize Whether to check the size of the metadata is under\n *   recommended limit. Default: `false`. If `true`, will try stringify the\n *   JSON object and print a console warning if the serialzied size is above the\n *   limit.\n * @throws Error if `userDefinedMetadata` is not a plain JSON object.\n */\nexport function checkUserDefinedMetadata(\n    userDefinedMetadata: {}, modelName: string, checkSize = false): void {\n  if (userDefinedMetadata == null ||\n      typeof userDefinedMetadata !== 'object' ||\n      Object.getPrototypeOf(userDefinedMetadata) !== Object.prototype ||\n      !plainObjectCheck(userDefinedMetadata)) {\n    throw new Error(\n        'User-defined metadata is expected to be a JSON object, but is not.');\n  }\n\n  if (checkSize) {\n    const out = JSON.stringify(userDefinedMetadata);\n    if (out.length > MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH) {\n      console.warn(\n          `User-defined metadata of model \"${modelName}\" is too large in ` +\n          `size (length=${out.length} when serialized). It is not ` +\n          `recommended to store such large objects in user-defined metadata. ` +\n          `Please make sure its serialized length is <= ` +\n          `${MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH}.`);\n    }\n  }\n}\n\n/**\n * Check if an input is plain JSON object or any valid subfield of it.\n *\n * @param x The input to be checked.\n * @param assertObject Whether to assert `x` is a JSON object, i.e., reject\n *   cases of arrays and primitives.\n * @return Returns `true` if and only if `x` is a plain JSON object,\n *   a JSON-valid primitive including string, number, boolean and null,\n *   or an array of the said types.\n */\n// tslint:disable-next-line:no-any\nexport function plainObjectCheck(x: any): boolean {\n  if (x === null) {\n    // Note: typeof `null` is 'object', and `null` is valid in JSON.\n    return true;\n  } else if (typeof x === 'object') {\n    if (Object.getPrototypeOf(x) === Object.prototype) {\n      // `x` is a JavaScript object and its prototype is Object.\n      const keys = Object.keys(x);\n      for (const key of keys) {\n        if (typeof key !== 'string') {\n          // JSON keys must be strings.\n          return false;\n        }\n        if (!plainObjectCheck(x[key])) {  // Recursive call.\n          return false;\n        }\n      }\n      return true;\n    } else {\n      // `x` is a JavaScript object but its prototype is not Object.\n      if (Array.isArray(x)) {\n        // `x` is a JavaScript array.\n        for (const item of x) {\n          if (!plainObjectCheck(item)) {  // Recursive call.\n            return false;\n          }\n        }\n        return true;\n      } else {\n        // `x` is a JavaScript object and its prototype is not Object,\n        // and it's not an Array. I.e., it's a complex object such as\n        // `Error` and `Date`.\n        return false;\n      }\n    }\n  } else {\n    // `x` is not a JavaScript object or `null`.\n    const xType = typeof x;\n    return xType === 'string' || xType === 'number' || xType === 'boolean';\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {Container} from '../engine/container';\nimport {Layer, Node} from '../engine/topology';\nimport {countParamsInWeights} from './variable_utils';\n\n/**\n * Print the summary of a LayersModel object.\n *\n * @param model tf.LayersModel instance.\n * @param lineLength Total length of printed lines. Set this to adapt to the\n *   display to different terminal or console sizes.\n * @param positions Relative or absolute positions of log elements in each\n *   line. Each number corresponds to right-most (i.e., ending) position of a\n *   column.\n *   If not provided, defaults to `[0.45, 0.85, 1]` for sequential-like\n *   models and `[0.33, 0.55, 0.67, 1]` for non-sequential like models.\n * @param printFn Print function to use.\n *   It will be called on each line of the summary. You can provide a custom\n *   function in order to capture the string summary. Defaults to `console.log`.\n */\nexport function printSummary(\n    model: Container, lineLength?: number, positions?: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void =\n        console.log): void {\n  const sequentialLike = isModelSequentialLike(model);\n\n  // Header names for different log elements.\n  const toDisplay: string[] = ['Layer (type)', 'Output shape', 'Param #'];\n  if (sequentialLike) {\n    lineLength = lineLength || 65;\n    positions = positions || [0.45, 0.85, 1];\n  } else {\n    lineLength = lineLength || 98;\n    positions = positions || [0.33, 0.55, 0.67, 1];\n    // Header names for different log elements.\n  }\n\n  if (positions[positions.length - 1] <= 1) {\n    // `positions` is relative. Convert it to absolute positioning.\n    positions = positions.map(p => Math.floor(lineLength * p));\n  }\n\n  let relevantNodes: Node[];\n  if (!sequentialLike) {\n    toDisplay.push('Receives inputs');\n    relevantNodes = [];\n    for (const depth in model.nodesByDepth) {\n      relevantNodes.push(...model.nodesByDepth[depth]);\n    }\n  }\n\n  printFn('_'.repeat(lineLength));\n  printRow(toDisplay, positions, printFn);\n  printFn('='.repeat(lineLength));\n\n  const layers = model.layers;\n  for (let i = 0; i < layers.length; ++i) {\n    if (sequentialLike) {\n      printLayerSummary(layers[i], positions, printFn);\n    } else {\n      printLayerSummaryWithConnections(\n          layers[i], positions, relevantNodes, printFn);\n    }\n    printFn((i === layers.length - 1 ? '=' : '_').repeat(lineLength));\n  }\n\n  // tslint:disable-next-line:no-any\n  (model as any).checkTrainableWeightsConsistency();\n\n  const trainableCount = countTrainableParams(model);\n  const nonTrainableCount = countParamsInWeights(model.nonTrainableWeights);\n\n  printFn(`Total params: ${trainableCount + nonTrainableCount}`);\n  printFn(`Trainable params: ${trainableCount}`);\n  printFn(`Non-trainable params: ${nonTrainableCount}`);\n  printFn('_'.repeat(lineLength));\n}\n\nfunction countTrainableParams(model: Container): number {\n  let trainableCount: number;\n  // tslint:disable:no-any\n  if ((model as any).collectedTrainableWeights != null) {\n    trainableCount =\n        countParamsInWeights((model as any).collectedTrainableWeights);\n  } else {\n    trainableCount = countParamsInWeights(model.trainableWeights);\n  }\n  // tslint:enable:no-any\n  return trainableCount;\n}\n\nfunction isModelSequentialLike(model: Container): boolean {\n  let sequentialLike = true;\n  const nodesByDepth: Node[][] = [];\n  const nodes: Node[] = [];\n  for (const depth in model.nodesByDepth) {\n    nodesByDepth.push(model.nodesByDepth[depth]);\n  }\n  for (const depthNodes of nodesByDepth) {\n    if (depthNodes.length > 1 ||\n        depthNodes.length === 1 && depthNodes[0].inboundLayers.length > 1) {\n      sequentialLike = false;\n      break;\n    }\n    nodes.push(...depthNodes);\n  }\n  if (sequentialLike) {\n    // Search for shared layers.\n    for (const layer of model.layers) {\n      let flag = false;\n      for (const node of layer.inboundNodes) {\n        if (nodes.indexOf(node) !== -1) {\n          if (flag) {\n            sequentialLike = false;\n            break;\n          } else {\n            flag = true;\n          }\n        }\n      }\n      if (!sequentialLike) {\n        break;\n      }\n    }\n  }\n  return sequentialLike;\n}\n\nfunction printRow(\n    fields: string[], positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void = console.log) {\n  let line = '';\n  for (let i = 0; i < fields.length; ++i) {\n    if (i > 0) {\n      line = line.slice(0, line.length - 1) + ' ';\n    }\n    line += fields[i];\n    line = line.slice(0, positions[i]);\n    line += ' '.repeat(positions[i] - line.length);\n  }\n  printFn(line);\n}\n\n/**\n * Prints a summary for a single Layer, without connectivity information.\n *\n * @param layer: Layer instance to print.\n */\nfunction printLayerSummary(\n    layer: Layer, positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const name = layer.name;\n  const className = layer.getClassName();\n  const fields: string[] =\n      [`${name} (${className})`, outputShape, layer.countParams().toString()];\n  printRow(fields, positions, printFn);\n}\n\n/**\n * Prints a summary for a single Layer, with connectivity information.\n */\nfunction printLayerSummaryWithConnections(\n    layer: Layer, positions: number[], relevantNodes: Node[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const connections: string[] = [];\n  for (const node of layer.inboundNodes) {\n    if (relevantNodes != null && relevantNodes.length > 0 &&\n        relevantNodes.indexOf(node) === -1) {\n      continue;\n    }\n    for (let i = 0; i < node.inboundLayers.length; ++i) {\n      const inboundLayer = node.inboundLayers[i].name;\n      const inboundLayerIndex = node.nodeIndices[i];\n      const inboundTensorIndex = node.tensorIndices[i];\n      connections.push(\n          `${inboundLayer}[${inboundLayerIndex}][${inboundTensorIndex}]`);\n    }\n  }\n  const name = layer.name;\n  const className = layer.getClassName();\n  const firstConnection = connections.length === 0 ? '' : connections[0];\n  const fields: string[] = [\n    `${name} (${className})`, outputShape, layer.countParams().toString(),\n    firstConnection\n  ];\n\n  printRow(fields, positions, printFn);\n  for (let i = 1; i < connections.length; ++i) {\n    printRow(['', '', '', connections[i]], positions, printFn);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Porting note: This file doesn't exist in PyKeras.\n// Its purpose here is to centralize the boundary layer between\n// tfjs-layers's internal Config TS-Centric format and PyKeras's\n// serialized Python Config format.\n\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {PyJsonValue} from '../keras_format/types';\nimport * as generic_utils from '../utils/generic_utils';\n// tslint:enable\n\n/**\n * Test whether a value in an array is the name of a LayersModel or Layer.\n * @param key The key name that the value is found under. Note that the key\n *   may not be at the level immediately above the value, if the value is in a\n *   nested array.\n * @param index Index of the value in the Array that it is found in.\n * @param value The value object.\n * @returns A boolean indicating whether value is a name.\n */\nfunction isArrayItemInputOrOutputName<T>(\n    key: string, index: number, value: T): boolean {\n  return (key === 'inboundNodes' || key === 'outputLayers' ||\n          key === 'inputLayers') &&\n      index === 0 && typeof value === 'string';\n}\n\n/**\n * Convert a Pythonic config object to TypeScript config object.\n * @param pythonicConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertPythonicToTs(\n    pythonicConfig: PyJsonValue, key?: string): serialization.ConfigDictValue {\n  if (pythonicConfig === null) {\n    return null;\n  } else if (typeof pythonicConfig === 'string') {\n    return generic_utils.toCamelCase(pythonicConfig);\n  } else if (\n      (typeof pythonicConfig === 'number') ||\n      (typeof pythonicConfig === 'boolean')) {\n    return pythonicConfig;\n  } else if (pythonicConfig instanceof Array) {\n    const tsArray = [];\n    const arrayLength = pythonicConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = pythonicConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        tsArray.push(item);\n      } else {\n        tsArray.push(convertPythonicToTs(item, key));\n      }\n    }\n    return tsArray;\n  } else {\n    const tsDict: serialization.ConfigDict = {};\n    for (const pythonicKey of Object.keys(pythonicConfig)) {\n      const pythonicValue = pythonicConfig[pythonicKey];\n      if (pythonicKey === 'name' && typeof pythonicValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // camel-case conversion.\n        tsDict[pythonicKey] = pythonicValue;\n      } else {\n        const tsKey = generic_utils.toCamelCase(pythonicKey);\n        tsDict[tsKey] = convertPythonicToTs(pythonicValue, tsKey);\n      }\n    }\n    return tsDict;\n  }\n}\n\n/**\n * Convert a TypeScript config object to Python config object.\n * @param tsConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertTsToPythonic(\n    tsConfig: serialization.ConfigDictValue, key?: string): PyJsonValue {\n  if (tsConfig === null || tsConfig === undefined) {\n    return null;\n  } else if (typeof tsConfig === 'string') {\n    return generic_utils.toSnakeCase(tsConfig);\n  } else if (\n      (typeof tsConfig === 'number') || (typeof tsConfig === 'boolean')) {\n    return tsConfig;\n  } else if (tsConfig instanceof Array) {\n    const pyArray = [];\n    const arrayLength = tsConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = tsConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        pyArray.push(item);\n      } else {\n        pyArray.push(convertTsToPythonic(item, key));\n      }\n    }\n    return pyArray;\n  } else {\n    const pyDict: serialization.ConfigDict = {};\n    for (const tsKey of Object.keys(tsConfig)) {\n      const tsValue = tsConfig[tsKey];\n      const pyKey = generic_utils.toSnakeCase(tsKey);\n      if ((tsKey === 'name' || tsKey === 'className') &&\n          typeof tsValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // snake-case conversion.\n        pyDict[pyKey] = tsValue;\n      } else {\n        pyDict[pyKey] = convertTsToPythonic(tsValue, tsKey);\n      }\n    }\n    return pyDict;\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Executor: Evaluates SymbolicTensor based on feeds.\n */\n\nimport {cast, dispose, memory, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {ValueError} from '../errors';\nimport {Kwargs} from '../types';\nimport {toList} from '../utils/generic_utils';\n\nimport {InputLayer} from './input_layer';\nimport {SymbolicTensor} from './topology';\n\n/**\n * Helper function to check the dtype and shape compatibility of a feed value.\n */\nfunction assertFeedCompatibility(key: SymbolicTensor, val: Tensor): Tensor {\n  // Check dtype compatibility.\n  if (key.dtype == null || key.dtype === val.dtype) {\n    //  a.  If types match, return val tensor as is.\n    return val;\n  }\n  try {\n    //  b. Attempt to convert to expected type.\n    return cast(val, key.dtype);\n  } catch (err) {\n    //  c. If conversion fails, return helpful error.\n    throw new ValueError(\n        `The dtype of the feed (${val.dtype}) can not be cast to the dtype ` +\n        `of the key '${key.name}' (${key.dtype}).`);\n  }\n}\n\n/**\n * A concrete Tensor value for a symbolic tensor as the key.\n */\nexport interface Feed {\n  key: SymbolicTensor;\n  value: Tensor;\n}\n\n/**\n * FeedDict: A mapping from unique SymbolicTensors to feed values for them.\n * A feed value is a concrete value represented as an `Tensor`.\n */\nexport class FeedDict {\n  private id2Value: {[id: number]: Tensor} = {};\n  private id2Mask: {[id: number]: Tensor} = {};\n  private name2Id: {[name: string]: number} = {};\n\n  /**\n   * Constructor, optionally does copy-construction.\n   * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case\n   *   copy-construction will be performed.\n   */\n  constructor(feeds?: Feed[]|FeedDict) {\n    if (feeds instanceof FeedDict) {\n      for (const id in feeds.id2Value) {\n        this.id2Value[id] = feeds.id2Value[id];\n        if (id in feeds.id2Mask) {\n          this.id2Mask[id] = feeds.id2Mask[id];\n        }\n      }\n    } else {\n      if (feeds == null) {\n        return;\n      }\n      for (const feed of feeds) {\n        this.add(feed.key, feed.value);\n      }\n    }\n  }\n\n  /**\n   * Add a key-value pair to the FeedDict.\n   *\n   * @param key The key of the feed.\n   * @param value The value of the tensor feed.\n   * @param mask The value of the mask feed (optional).\n   * @returns This `FeedDict`.\n   * @throws ValueError: If the key `SymbolicTensor` already exists in the\n   *   `FeedDict`.\n   */\n  add(key: SymbolicTensor, value: Tensor, mask?: Tensor): FeedDict {\n    if (this.id2Value[key.id] == null) {\n      this.id2Value[key.id] = assertFeedCompatibility(key, value);\n      this.name2Id[key.name] = key.id;\n      if (mask != null) {\n        this.id2Mask[key.id] = mask;\n      }\n    } else {\n      throw new ValueError(`Duplicate key: name=${key.name}, id=${key.id}`);\n    }\n    return this;\n  }\n\n  /**\n   * Add a Feed to the FeedDict.\n   * @param feed The new `Feed` to add.\n   * @returns This `FeedDict`.\n   */\n  addFeed(feed: Feed) {\n    this.add(feed.key, feed.value);\n  }\n\n  /**\n   * Probe whether a key already exists in the FeedDict.\n   * @param key\n   */\n  hasKey(key: SymbolicTensor): boolean {\n    return this.id2Value[key.id] != null;\n  }\n\n  /**\n   * Get all the SymbolicTensor available in this FeedDict.\n   */\n  names(): string[] {\n    return Object.keys(this.name2Id);\n  }\n\n  /**\n   * Get the feed value for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed value.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getValue(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Value[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Value[id];\n    }\n  }\n\n  /**\n   * Get the feed mask for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed mask.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getMask(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Mask[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Mask[id];\n    }\n  }\n\n  /** Dispose all mask Tensors held by this object. */\n  disposeMasks() {\n    if (this.id2Mask != null) {\n      dispose(this.id2Mask);\n    }\n  }\n}\n\n// Cache for topologically sorted SymbolicTensors for given execution\n// targets (i.e., fetches).\nconst cachedSorted: {[concatFetchNames: string]: SymbolicTensor[]} = {};\n\n// Cache for recipient count maps for given execution targets (i.e., fetches).\nconst cachedRecipientCounts:\n    {[concatFetchNames: string]: {[fetchName: string]: number}} = {};\n\n/**\n * Interface for the optional object used for probing the memory\n * usage and other statistics during execution.\n */\nexport interface ExecutionProbe {\n  /**\n   * Maximum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  maxNumTensors?: number;\n\n  /**\n   * Minimum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  minNumTensors?: number;\n}\n\n/**\n * Execute a SymbolicTensor by using concrete feed values.\n *\n * A `SymbolicTensor` object is a node in a computation graph of TF.js\n * Layers. The object is backed by a source layer and input\n * `SymbolicTensor`s to the source layer. This method evaluates\n * the `call()` method of the source layer, using concrete values of the\n * inputs obtained from either\n * * `feedDict`, if the input key exists in `feedDict`, or else,\n * * a recursive call to `execute()` itself.\n *\n * @param x: The `SymbolicTensor` to execute.\n * @param feedDict: The feed values, as base condition of the recursion.\n *   execution.\n * @param kwargs: Optional keyword arguments.\n * @param probe: A probe object (of interface `ExecutionProbe`) used for\n *   testing memory footprint of `execute` calls.\n * @returns Result of the execution.\n * @throws ValueError: If any `SymbolicTensor`s from `InputLayer`s\n *   encountered during the execution lacks a feed value in `feedDict`.\n */\nexport function execute(\n    fetches: SymbolicTensor|SymbolicTensor[], feedDict: FeedDict,\n    kwargs?: Kwargs, probe?: ExecutionProbe): Tensor|\n    Tensor[]|[Tensor | Tensor[]] {\n  const training: boolean = kwargs == null ? false : kwargs['training'];\n\n  const arrayFetches = Array.isArray(fetches);\n  const fetchArray: SymbolicTensor[] =\n      arrayFetches ? fetches as SymbolicTensor[] : [fetches as SymbolicTensor];\n\n  const outputNames = fetchArray.map(t => t.name);\n  const finalOutputs: Tensor[] = [];\n  const feedNames = feedDict.names();\n  for (const outputName of outputNames) {\n    if (feedNames.indexOf(outputName) !== -1) {\n      finalOutputs.push(feedDict.getValue(outputName));\n    } else {\n      finalOutputs.push(null);\n    }\n  }\n\n  if (probe != null) {\n    // For optional probing of memory footprint during execution.\n    probe.maxNumTensors = -Infinity;\n    probe.minNumTensors = Infinity;\n  }\n\n  // Check cache.\n  const fetchAndFeedKey =\n      outputNames.join(',') + '|' + feedDict.names().join(',');\n  let sorted: SymbolicTensor[];\n  let recipientCounts: {[fetchName: string]: number};\n  if (cachedSorted[fetchAndFeedKey] == null) {\n    // Cache doesn't contain the desired combination of fetches. Compute\n    // topological sort for the combination for the first time.\n    const out = getTopologicalSortAndRecipientCounts(fetchArray, feedDict);\n    sorted = out.sorted;\n    recipientCounts = out.recipientCounts;\n\n    // Store results in cache for future use.\n    cachedSorted[fetchAndFeedKey] = sorted;\n    cachedRecipientCounts[fetchAndFeedKey] = recipientCounts;\n  }\n  sorted = cachedSorted[fetchAndFeedKey];\n  recipientCounts = {};\n  if (!training) {\n    Object.assign(recipientCounts, cachedRecipientCounts[fetchAndFeedKey]);\n  }\n\n  const internalFeedDict = new FeedDict(feedDict);\n\n  // Start iterative execution on the topologically-sorted SymbolicTensors.\n  for (let i = 0; i < sorted.length; ++i) {\n    if (probe != null) {\n      // For optional probing of memory usage during execution.\n      const numTensors = memory().numTensors;\n      if (numTensors > probe.maxNumTensors) {\n        probe.maxNumTensors = numTensors;\n      }\n      if (numTensors < probe.minNumTensors) {\n        probe.minNumTensors = numTensors;\n      }\n    }\n\n    const symbolic = sorted[i];\n    const srcLayer = symbolic.sourceLayer;\n    if (srcLayer instanceof InputLayer) {\n      continue;\n    }\n    const inputValues: Tensor[] = [];\n    const inputMasks: Tensor[] = [];\n    const tensorsToDispose: Tensor[] = [];\n\n    let maskExists = false;\n    for (const input of symbolic.inputs) {\n      const value = internalFeedDict.getValue(input);\n      const mask = internalFeedDict.getMask(input);\n      inputValues.push(value);\n      inputMasks.push(mask);\n      if (mask != null) {\n        maskExists = true;\n      }\n      if (!training) {\n        recipientCounts[input.name]--;\n        if (recipientCounts[input.name] === 0 && !feedDict.hasKey(input) &&\n            outputNames.indexOf(input.name) === -1 && !value.isDisposed &&\n            input.sourceLayer.stateful !== true) {\n          tensorsToDispose.push(value);\n        }\n      }\n    }\n\n    if (maskExists) {\n      kwargs = kwargs || {};\n      kwargs['mask'] = inputMasks[0];\n    }\n    const outputTensors =\n        toList(srcLayer.apply(inputValues, kwargs)) as Tensor[];\n    let outputMask: Tensor|Tensor[] = null;\n    if (srcLayer.supportsMasking) {\n      outputMask = srcLayer.computeMask(inputValues, inputMasks);\n    }\n    const layerOutputs = getNodeOutputs(symbolic);\n    const outputSymbolicTensors =\n        Array.isArray(layerOutputs) ? layerOutputs : [layerOutputs];\n    for (let i = 0; i < outputSymbolicTensors.length; ++i) {\n      if (!internalFeedDict.hasKey(outputSymbolicTensors[i])) {\n        internalFeedDict.add(\n            outputSymbolicTensors[i], outputTensors[i],\n            Array.isArray(outputMask) ? outputMask[0] : outputMask);\n      }\n      const index = outputNames.indexOf(outputSymbolicTensors[i].name);\n      if (index !== -1) {\n        finalOutputs[index] = outputTensors[i];\n      }\n    }\n\n    if (!training) {\n      // Clean up Tensors that are no longer needed.\n      dispose(tensorsToDispose);\n    }\n  }\n  // NOTE(cais): Unlike intermediate tensors, we don't discard mask\n  // tensors as we go, because these tensors are sometimes passed over a\n  // series of mutliple layers, i.e., not obeying the immediate input\n  // relations in the graph. If this becomes a memory-usage concern,\n  // we can improve this in the future.\n  internalFeedDict.disposeMasks();\n\n  return arrayFetches ? finalOutputs : finalOutputs[0];\n}\n\ntype RecipientCounts = {\n  [fetchName: string]: number\n};\n\nexport type RecipientMap = {\n  [fetchName: string]: Set<string>;\n};\n\n/**\n * Sort the `SymbolicTensor`s topologically, for an array of fetches.\n *\n * This function calls getTopologicalSortAndRecipientCountsForOneFetch and\n * merges their results.\n *\n * @param fetch The array of fetches requested. Must be a non-empty array.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientCounts: Recipient counts for all SymbolicTensors in `sorted`.\n */\nfunction getTopologicalSortAndRecipientCounts(\n    fetches: SymbolicTensor[], feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientCounts: RecipientCounts} {\n  util.assert(\n      fetches != null && fetches.length > 0,\n      () => `Expected at least one fetch, got none`);\n\n  let finalSorted: SymbolicTensor[] = [];\n  let finalRecipientMap: RecipientMap = {};\n  if (fetches.length === 1) {\n    // Special-casing 1 fetch for efficiency.\n    const out =\n        getTopologicalSortAndRecipientCountsForOneFetch(fetches[0], feedDict);\n    finalSorted = out.sorted;\n    finalRecipientMap = out.recipientMap;\n  } else {\n    const visited = new Set<string>();\n    for (const fetch of fetches) {\n      const {sorted, recipientMap} =\n          getTopologicalSortAndRecipientCountsForOneFetch(fetch, feedDict);\n\n      // Merge sorted SymbolicTensor Arrays.\n      for (const symbolicTensor of sorted) {\n        if (!visited.has(symbolicTensor.name)) {\n          finalSorted.push(symbolicTensor);\n          visited.add(symbolicTensor.name);\n        }\n      }\n\n      // Merge recipient maps.\n      for (const name in recipientMap) {\n        if (finalRecipientMap[name] == null) {\n          finalRecipientMap[name] = new Set<string>();\n        }\n        recipientMap[name].forEach(\n            recipient => finalRecipientMap[name].add(recipient));\n      }\n    }\n  }\n  return {\n    sorted: finalSorted,\n    recipientCounts: recipientMap2Counts(finalRecipientMap)\n  };\n}\n\nfunction recipientMap2Counts(recipientMap: RecipientMap): RecipientCounts {\n  const recipientCounts: RecipientCounts = {};\n  for (const name in recipientMap) {\n    recipientCounts[name] = recipientMap[name].size;\n  }\n  return recipientCounts;\n}\n\n/**\n * Sort the `SymbolicTensor`s topologically, for a single fetch.\n *\n * This helper function processes the upstream SymbolicTensors of a single\n * fetch.\n *\n * @param fetch The single fetch requested.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientMap: Recipient names for all SymbolicTensors in `sorted`.\n */\nexport function getTopologicalSortAndRecipientCountsForOneFetch(\n    fetch: SymbolicTensor, feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientMap: RecipientMap} {\n  const visited = new Set<string>();\n  const sorted: SymbolicTensor[] = [];\n  const recipientMap: RecipientMap = {};\n\n  // Put keys of the feedDict into visited first, so they don't have to be\n  // walked. This is needed in case where there are feeds for intermediate\n  // SymbolicTensors of the graph.\n  for (const key of feedDict.names()) {\n    visited.add(key);\n  }\n\n  const stack: SymbolicTensor[] = [];\n  const marks: number[] = [];\n\n  // Initial population of stack and marks.\n  stack.push(fetch);\n\n  while (stack.length > 0) {\n    const top = stack[stack.length - 1];\n    if (visited.has(top.name)) {\n      stack.pop();\n      continue;\n    }\n    const topIsMarked = marks[marks.length - 1] === stack.length - 1;\n    if (top.inputs.length === 0 || topIsMarked) {\n      // Input SymbolicTensor or all children have been visited.\n      stack.pop();\n      sorted.push(top);\n      visited.add(top.name);\n      if (topIsMarked) {\n        marks.pop();\n      }\n    } else {\n      // A non-input SymbolicTensor whose upstream SymbolicTensors haven't\n      // been visited yet. Push them onto the stack.\n      marks.push(stack.length - 1);\n      for (const input of top.inputs) {\n        // Increment the recipient count. Note that this needs to happen\n        // regardless of whether the SymbolicTensor has been visited before.\n        if (recipientMap[input.name] == null) {\n          recipientMap[input.name] = new Set<string>();\n        }\n        recipientMap[input.name].add(top.name);\n\n        if (visited.has(input.name)) {\n          continue;  // Avoid repeated visits to the same SymbolicTensor.\n        }\n        stack.push(input);\n      }\n    }\n  }\n  return {sorted, recipientMap};\n}\n\n/**\n * Get the symbolic output tensors of the node to which a given fetch belongs.\n * @param fetch The fetched symbolic tensor.\n * @returns The Array of symbolic tensors output by the node to which `fetch`\n *   belongs.\n */\nfunction getNodeOutputs(fetch: SymbolicTensor): SymbolicTensor|\n    SymbolicTensor[] {\n  let layerOutputs: SymbolicTensor|SymbolicTensor[];\n  if (fetch.sourceLayer.inboundNodes.length === 1) {\n    layerOutputs = fetch.sourceLayer.output;\n  } else {\n    let nodeIndex: number = null;\n    for (let i = 0; i < fetch.sourceLayer.inboundNodes.length; ++i) {\n      for (const outputTensor of fetch.sourceLayer.inboundNodes[i]\n               .outputTensors) {\n        if (outputTensor.id === fetch.id) {\n          nodeIndex = i;\n          break;\n        }\n      }\n    }\n    layerOutputs = fetch.sourceLayer.getOutputAt(nodeIndex);\n  }\n  return layerOutputs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {NamedTensorMap, Scalar, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {TensorKeyWithArgsArray} from '../keras_format/node_config';\nimport {PyJsonDict} from '../keras_format/types';\nimport {deserialize as deserializeLayer} from '../layers/serialization';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {convertTsToPythonic} from '../utils/serialization_utils';\nimport * as types_utils from '../utils/types_utils';\nimport {batchSetValue, LayerVariable} from '../variables';\nimport {version as layersVersion} from '../version';\n\nimport {execute, FeedDict} from './executor';\nimport {InputLayer} from './input_layer';\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/** Constructor config for Container. */\nexport interface ContainerArgs {\n  inputs: SymbolicTensor|SymbolicTensor[];\n  outputs: SymbolicTensor|SymbolicTensor[];\n  name?: string;\n}\n\n/**\n * A Container is a directed acyclic graph of layers.\n *\n * It is the topological form of a \"model\". A LayersModel\n * is simply a Container with added training routines.\n *\n */\nexport abstract class Container extends Layer {\n  inputs: SymbolicTensor[];\n  outputs: SymbolicTensor[];\n\n  inputLayers: Layer[];\n  inputLayersNodeIndices: number[];\n  inputLayersTensorIndices: number[];\n\n  outputLayers: Layer[];\n  outputLayersNodeIndices: number[];\n  outputLayersTensorIndices: number[];\n\n  layers: Layer[];\n  layersByDepth: {[depth: string]: Layer[]};\n  nodesByDepth: {[depth: string]: Node[]};\n\n  internalContainerRefs: Container[];\n\n  containerNodes = new Set<string>();\n\n  // TODO(michaelterry): Add cache support\n  // private outputMaskCache: any;\n  // private outputTensorCache: any;\n  // private outputShapeCache: any;\n\n  inputNames: string[];\n  outputNames: string[];\n  feedInputShapes: Shape[];\n\n  protected internalInputShapes: Shape[];\n  protected internalOutputShapes: Shape[];\n  // TODO(cais): Maybe 'feed' should not in the names of these variables,\n  //   due to the fact that our backend is not symbolic.\n  protected feedInputNames: string[];\n  protected feedOutputNames: string[];\n\n  constructor(args: ContainerArgs) {\n    // No args passed to super's constructor.\n    super({});\n    this.name = args.name;\n    if (this.name == null) {\n      const prefix = this.getClassName().toLowerCase();\n      this.name = getUid(prefix);\n    }\n\n    this.supportsMasking = false;\n    this.trainable_ = true;\n\n    // TODO(michaelterry): Initialize perInputLosses/Updates here.\n\n    // Container-specific properties.\n    if (Array.isArray(args.inputs)) {\n      this.inputs = args.inputs.slice();\n    } else {\n      this.inputs = [args.inputs];\n    }\n    if (Array.isArray(args.outputs)) {\n      this.outputs = args.outputs.slice();\n    } else {\n      this.outputs = [args.outputs];\n    }\n\n    // Check for redundancy in inputs.\n    if (generic_utils.unique(this.inputs).length !== this.inputs.length) {\n      throw new ValueError(\n          'The list of inputs passed to the model is ' +\n          'redundant. All inputs should only appear once. Found: ' +\n          `${this.inputs.map(x => x.name)}`);\n    }\n\n    // Check for redundancy in outputs.\n    if (generic_utils.unique(this.outputs).length !== this.outputs.length) {\n      console.warn(\n          'The list of outputs passed to the model is redundant. ' +\n          'All outputs should only appear once. Found: ' +\n          `${this.outputs.map(x => x.name)}`);\n    }\n\n    /*\n      List of initial layers (1 to 1 mapping with this.inputs, hence the same\n      layer might appear twice)\n    */\n    this.inputLayers = [];\n    this.inputLayersNodeIndices = [];\n    this.inputLayersTensorIndices = [];\n    /*\n      List of layers (1 to 1 mapping with this.outputs, hence the same layer\n      might appear twice)\n    */\n    this.outputLayers = [];\n    this.outputLayersNodeIndices = [];\n    this.outputLayersTensorIndices = [];\n    /*\n      All layers in order of horizontal graph traversal. Entries are unique.\n      Includes input and output layers.\n    */\n    this.layers = [];\n\n    /*\n      References to container layers that were constructed internally. We need\n      these to properly dispose of tensors from nested containers.\n    */\n    this.internalContainerRefs = [];\n\n    // TODO(michaelterry): Determine if caching still needed with eager\n    // backend.\n    /*\n      This is for performance optimization when calling the Container on new\n      inputs. Every time the Container is called on a set on input tensors,\n      we compute the output tensors, output masks and output shapes in one pass,\n      then cache them here. When one of these outputs is queried later,\n      we retrieve it from there instead of recomputing it.\n    */\n    // this.outputTensorCache = {};\n    // this.outputShapeCache = {};\n\n    // Build this.outputLayers:\n    for (const x of this.outputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      this.outputLayers.push(layer);\n      this.outputLayersNodeIndices.push(nodeIndex);\n      this.outputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // TODO(michaelterry): Add output mask cache code.\n\n    // Build this.inputLayers:\n    for (const x of this.inputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      /*\n        It's supposed to be an input layer, so only one node\n        and one tensor output.\n      */\n      generic_utils.assert(nodeIndex === 0, 'input layer has >1 nodes');\n      generic_utils.assert(tensorIndex === 0, 'input layer has >1 tensors');\n      this.inputLayers.push(layer);\n      this.inputLayersNodeIndices.push(nodeIndex);\n      this.inputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // Build this.inputNames and this.outputNames.\n    this.inputNames = [];\n    this.outputNames = [];\n    this.feedInputShapes = [];\n    this.feedInputNames = [];\n    this.feedOutputNames = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      // Check that layer is an InputLayer.\n      if (!(layer instanceof InputLayer)) {\n        throw new TypeError(\n            'Input layers to a LayersModel must be InputLayer objects. ' +\n            `Received inputs: ${args.inputs}. ` +\n            `Input ${i} (0-based) originates ` +\n            `from layer type ${layer.getClassName()}.`);\n      }\n      this.inputNames.push(layer.name);\n      this.feedInputShapes.push(layer.batchInputShape);\n\n      this.feedInputNames.push(layer.name);\n    }\n    for (const layer of this.outputLayers) {\n      this.outputNames.push(layer.name);\n    }\n\n    this.internalInputShapes = this.inputs.map(x => x.shape);\n    this.internalOutputShapes = this.outputs.map(x => x.shape);\n\n    /*\n      Container_nodes: set of nodes included in the graph (not all nodes\n      included in the layers are relevant to the current graph).\n    */\n    // ids of all nodes relevant to the Container:\n    const nodesDepths: {[nodeID: string]: number} = {};\n    // To recover nodes from their ID.\n    const nodeIDToNode: {[nodeID: string]: Node} = {};\n    const layersDepths: {[layerID: string]: number} = {};\n    // To layers from their ID.\n    const layerIDToLayer: {[layerID: string]: Layer} = {};\n    const layerIndices: {[layerID: string]: number} = {};\n    const nodesInDecreasingDepth: Node[] = [];\n\n    /**\n     * Builds a map of the graph of layers.\n     *\n     * This recursively updates the map `layerIndices`,\n     * the list `nodesInDecreasingDepth` and the set `containerNodes`.\n     *\n     * @param tensor Some tensor in a graph.\n     * @param finishedNodes Set of nodes whose subgraphs have been traversed\n     *         completely. Useful to prevent duplicated work.\n     * @param nodesInProgress Set of nodes that are currently active on the\n     *         recursion stack. Useful to detect cycles.\n     * @param layer Layer from which `tensor` comes from. If not provided,\n     *   will be obtained from tensor.sourceLayer.\n     * @param nodeIndex Node index from which `tensor` comes from.\n     * @param tensorIndex TensorIndex from which `tensor` comes from.\n     *\n     * @exception RuntimeError if a cycle is detected.\n     */\n    const buildMapOfGraph =\n        (tensor: SymbolicTensor, finishedNodes: Node[], nodesInProgress: Node[],\n         layer?: Layer, nodeIndex?: number, tensorIndex?: number) => {\n          if (layer == null || nodeIndex == null || tensorIndex == null) {\n            layer = tensor.sourceLayer;\n            nodeIndex = tensor.nodeIndex;\n            tensorIndex = tensor.tensorIndex;\n          }\n          const node = layer.inboundNodes[nodeIndex];\n\n          // Prevent cycles.\n          if (nodesInProgress.indexOf(node) !== -1) {\n            throw new RuntimeError(\n                `The tensor ${tensor.name} at layer \"${layer.name}\" ` +\n                'is part of a cycle.');\n          }\n\n          // Don't repeat work for shared subgraphs\n          if (finishedNodes.indexOf(node) !== -1) {\n            return;\n          }\n\n          // Update containerNodes.\n          this.containerNodes.add(Container.nodeKey(layer, nodeIndex));\n\n          // Store the traversal order for layer sorting.\n          if (!(layer.id in layerIndices)) {\n            layerIndices[layer.id] = Object.keys(layerIndices).length;\n          }\n\n          if (nodesInProgress.indexOf(node) === -1) {\n            nodesInProgress.push(node);\n          }\n\n          // Propagate to all previous tensors connected to this node.\n          const numInboundLayers = node.inboundLayers.length;\n          for (let i = 0; i < numInboundLayers; i++) {\n            const x = node.inputTensors[i];\n            const layer = node.inboundLayers[i];\n            const nodeIndex = node.nodeIndices[i];\n            const tensorIndex = node.tensorIndices[i];\n            buildMapOfGraph(\n                x, finishedNodes, nodesInProgress, layer, nodeIndex,\n                tensorIndex);\n          }\n          finishedNodes.push(node);\n          while (nodesInProgress.indexOf(node) >= 0) {\n            nodesInProgress.splice(nodesInProgress.indexOf(node), 1);\n          }\n          nodesInDecreasingDepth.push(node);\n        };\n\n    const finishedNodes: Node[] = [];\n    const nodesInProgress: Node[] = [];\n    for (const x of this.outputs) {\n      buildMapOfGraph(x, finishedNodes, nodesInProgress);\n    }\n\n    const reversedNodesInDecreasingDepth =\n        nodesInDecreasingDepth.slice().reverse();\n    for (const node of reversedNodesInDecreasingDepth) {\n      nodeIDToNode[node.id] = node;\n      // If the depth is not set, the node has no outbound nodes (depth 0).\n      if (!(node.id in nodesDepths)) {\n        nodesDepths[node.id] = 0;\n      }\n      let depth = nodesDepths[node.id];\n\n      // Update the depth of the corresponding layer\n      const previousDepth =\n          (layersDepths[node.outboundLayer.id] == null ?\n               0 :\n               layersDepths[node.outboundLayer.id]);\n\n      /*\n        If we've seen this layer before at a higher depth, we should use that\n        depth instead of the node depth.  This is necessary for shared layers\n        that have inputs at different depth levels in the graph.\n      */\n      depth = Math.max(depth, previousDepth);\n      layersDepths[node.outboundLayer.id] = depth;\n      layerIDToLayer[node.outboundLayer.id] = node.outboundLayer;\n      nodesDepths[node.id] = depth;\n\n      // Update the depth of inbound nodes.\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const inboundLayer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const inboundNode = inboundLayer.inboundNodes[nodeIndex];\n        const previousDepth =\n            (nodesDepths[inboundNode.id] == null ? 0 :\n                                                   nodesDepths[inboundNode.id]);\n        nodesDepths[inboundNode.id] = Math.max(depth + 1, previousDepth);\n        nodeIDToNode[inboundNode.id] = inboundNode;\n      }\n    }\n\n    // Build a dict {depth: list of nodes with this depth}\n    const nodesByDepth: {[depth: string]: Node[]} = {};\n    for (const nodeID in nodesDepths) {\n      const depth = nodesDepths[nodeID];\n      if (!(depth in nodesByDepth)) {\n        nodesByDepth[depth] = [];\n      }\n      nodesByDepth[depth].push(nodeIDToNode[nodeID]);\n    }\n\n    // Build a dict {depth: list of layers with this depth}\n    const layersByDepth: {[depth: string]: Layer[]} = {};\n    for (const layerID in layersDepths) {\n      const depth = layersDepths[layerID];\n      if (!(depth in layersByDepth)) {\n        layersByDepth[depth] = [];\n      }\n      layersByDepth[depth].push(layerIDToLayer[layerID]);\n    }\n\n    // Get sorted list of layer depths.\n    let depthKeys = Object.keys(layersByDepth)\n                        .map(x => parseInt(x, 10))\n                        .sort(generic_utils.reverseNumberCompare);\n\n    // Set this.layers and this.layersByDepth.\n    this.layers = [];\n    for (const depth of depthKeys) {\n      const layersForDepth = layersByDepth[depth];\n      // Container.layers needs to have a deterministic order:\n      // here we order them by traversal order.\n      layersForDepth.sort((a, b) => {\n        const aIndex = layerIndices[a.id];\n        const bIndex = layerIndices[b.id];\n        if (aIndex < bIndex) {\n          return -1;\n        }\n        if (aIndex > bIndex) {\n          return 1;\n        }\n        return 0;\n      });\n      for (const layer of layersForDepth) {\n        if (layer instanceof Container) {\n          this.internalContainerRefs.push(layer);\n        }\n        this.layers.push(layer);\n      }\n    }\n    this.layersByDepth = layersByDepth;\n\n    // Get sorted list of node depths;\n    depthKeys = Object.keys(nodesByDepth)\n                    .map(x => parseInt(x, 10))\n                    .sort(generic_utils.reverseNumberCompare);\n\n    // Check that all tensors required are computable.\n    // computable_tensors: all tensors in the graph\n    // that can be computed from the inputs provided.\n    const computableTensors = this.inputs.slice();\n\n    // To provide a better error msg.\n    const layersWithCompleteInput: string[] = [];\n    for (const depth of depthKeys) {\n      for (const node of nodesByDepth[depth]) {\n        const layer = node.outboundLayer;\n        if (layer != null) {\n          for (const x of node.inputTensors) {\n            if (computableTensors.indexOf(x) === -1) {\n              throw new RuntimeError(\n                  `Graph disconnected: cannot obtain value for tensor ${x}` +\n                  ` at layer \"${layer.name}\". ` +\n                  'The following previous layers were accessed without ' +\n                  `issue: ${layersWithCompleteInput}`);\n            }\n          }\n          for (const x of node.outputTensors) {\n            computableTensors.push(x);\n          }\n          layersWithCompleteInput.push(layer.name);\n        }\n      }\n    }\n\n    // Set this.containerNodes and this.nodesByDepth.\n    this.nodesByDepth = nodesByDepth;\n\n    // Ensure name unicity, which will be crucial for serialization\n    // (since serialized nodes refer to layers by their name).\n    const allNames = this.layers.map(x => x.name);\n    for (const name of allNames) {\n      const numOccurrences = allNames.filter(x => x === name).length;\n      if (numOccurrences !== 1) {\n        throw new RuntimeError(\n            `The name \"${name}\" is used ${numOccurrences} times ` +\n            'in the model. All layer names should be unique. Layer names: ' +\n            JSON.stringify(allNames));\n      }\n    }\n\n    // Layer parameters.\n    // The new container starts with a single inbound node\n    // for its inputs, and no outbound nodes.\n    // Will be appended to by future calls to apply().\n    this.outboundNodes = [];\n    // Will be appended to below, and by future calls to apply().\n    this.inboundNodes = [];\n\n    // Create the node linking internal inputs to internal outputs.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: this.inputs,\n      outputTensors: this.outputs,\n      inputMasks: this.inputs.map(x => null),\n      outputMasks: this.outputs.map(x => null),\n      inputShapes: this.inputs.map(x => x.shape),\n      outputShapes: this.outputs.map(x => x.shape)\n    });\n    this.built = true;\n    this._refCount = 1;  // The ref count of a container always start at 1.\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Container '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose a LayersModel's weights.\n   *\n   * This method decrease the reference count of the LayersModel object by 1.\n   *\n   * A LayersModel is reference-counted. Its reference count is incremented by 1\n   * when it is first constructed and when it is used as a Layer of another\n   * LayersModel.\n   *\n   * If the reference count of a LayersModel becomes 0, the `dispose` method of\n   * all its constituent `Layer`s will be called.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * `dispose` method of its constituent `Layer`s will *not* be called.\n   *\n   * After a LayersModel is disposed, it cannot be used in calls such as\n   * 'predict`, `evaluate` or `fit` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the LayersModel after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the LayersModel has\n   *   already been disposed.\n   */\n  dispose(): DisposeResult {\n    this.assertNotDisposed();\n    const result:\n        DisposeResult = {refCountAfterDispose: null, numDisposedVariables: 0};\n    if (--this._refCount === 0) {\n      for (const layer of this.layers) {\n        result.numDisposedVariables += layer.dispose().numDisposedVariables;\n      }\n\n      // Call dispose on each internally created container layer again to ensure\n      // their refCounts hit zero and their tensors are subsequently deleted.\n      for (const container of this.internalContainerRefs) {\n        result.numDisposedVariables += container.dispose().numDisposedVariables;\n      }\n    }\n    result.refCountAfterDispose = this._refCount;\n    return result;\n  }\n\n  get trainable() {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.layers.forEach(layer => {\n      // tslint:disable-next-line:no-any\n      ((layer as any)._trainableWeights as LayerVariable[])\n          .forEach(w => w.trainable = trainable);\n    });\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    // Porting Note: This check below is to prevent errors where the\n    //   _trainableWeights inherited from the parent class (Layer) gets\n    //   inadvertently used.\n    if (this._trainableWeights.length > 0) {\n      throw new ValueError(\n          'Container instance unexpectedly contains _trainableWeights.' +\n          'The trainable weights of a Container are a union of the ' +\n          'trainable weights of its consituent Layers. Its own ' +\n          '_trainableWeights must remain an empty Array.');\n    }\n\n    if (!this.trainable) {\n      return [];\n    }\n    let weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights = weights.concat(layer.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights.push(...layer.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const layer of this.layers) {\n        trainableWeights.push(...layer.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  /**\n   * Loads all layer weights from a JSON object.\n   *\n   * Porting Note: HDF5 weight files cannot be directly loaded in JavaScript /\n   *   TypeScript. The utility script at `scripts/pykeras.py` offers means\n   *   to convert them into JSON strings compatible with this method.\n   * Porting Note: TensorFlow.js Layers supports only loading by name currently.\n   *\n   * @param weights A JSON mapping weight names to weight values as nested\n   *   arrays of numbers, or a `NamedTensorMap`, i.e., a JSON mapping weight\n   *   names to `tf.Tensor` objects.\n   * @param strict Require that the provided weights exactly match those\n   *   required by the container.  Default: `true`.  Passing `false` means that\n   *   extra weights and missing weights will be silently ignored.\n   */\n  loadWeights(weights: NamedTensorMap, strict = true) {\n    const nameToWeight: {[name: string]: LayerVariable} = {};\n    let totalWeightsCount = 0;\n    for (const layer of this.layers) {\n      for (const weight of layer.weights) {\n        if (nameToWeight[weight.originalName] != null) {\n          throw new ValueError(`Duplicate weight name: ${weight.originalName}`);\n        }\n        nameToWeight[weight.originalName] = weight;\n        totalWeightsCount++;\n      }\n    }\n\n    const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n    for (const name in weights) {\n      // TF 2.2.0 added cell name to the weight name in the format of\n      // layer_name/cell_name/weight_name, we need to remove\n      // the inner cell name.\n      let validatedName = name;\n      if (nameToWeight[name] == null) {\n        const tokens = name.split('/');\n        const shortenNameArray =\n            tokens.slice(0, -2).concat([tokens[tokens.length - 1]]);\n        validatedName = shortenNameArray.join('/');\n      }\n      if (nameToWeight[validatedName] != null) {\n        weightValueTuples.push([nameToWeight[validatedName], weights[name]]);\n      } else if (strict) {\n        throw new ValueError(\n            `Provided weight data has no target variable: ${name}`);\n      }\n      delete nameToWeight[validatedName];\n    }\n\n    if (strict) {\n      // Check that all weights are set.\n      const unsetNames: string[] = [];\n      for (const name in nameToWeight) {\n        unsetNames.push(name);\n      }\n      if (unsetNames.length > 0) {\n        throw new ValueError(\n            `${unsetNames.length} of ${\n                totalWeightsCount} weights are not set: ` +\n            `${unsetNames}`);\n      }\n    }\n\n    batchSetValue(weightValueTuples);\n  }\n\n  /**\n   * Util shared between different serialization methods.\n   * @returns LayersModel config with Keras version information added.\n   */\n  protected updatedConfig(): serialization.ConfigDict {\n    const theConfig = this.getConfig();\n    const modelConfig: serialization.ConfigDict = {};\n    modelConfig['className'] = this.getClassName();\n    modelConfig['config'] = theConfig;\n    modelConfig['kerasVersion'] = `tfjs-layers ${layersVersion}`;\n    // TODO(nielsene): Replace something like K.backend() once\n    // possible.\n    modelConfig['backend'] = 'TensorFlow.js';\n    return modelConfig;\n  }\n\n  /**\n   * Returns a JSON string containing the network configuration.\n   *\n   * To load a network from a JSON save file, use\n   * models.modelFromJSON(jsonString);\n   * @param extraJsonArgs Unused in tfjs-layers, maintained for PyKeras\n   * @param returnString Whether the return value should be stringified\n   *    (default: `true`).\n   * @returns a JSON string if `returnString` (default), or a JSON object if\n   *   `!returnString`.\n   */\n  // tslint:disable-next-line:no-any\n  toJSON(unused?: any, returnString = true): string|PyJsonDict {\n    const modelConfig = convertTsToPythonic(this.updatedConfig()) as PyJsonDict;\n    return returnString ? JSON.stringify(modelConfig) : modelConfig;\n  }\n\n  /**\n   * Call the model on new inputs.\n   *\n   * In this case `call` just reapplies all ops in the graph to the new inputs\n   * (e.g. build a new computational graph from the provided inputs).\n   *\n   * @param inputs A tensor or list of tensors.\n   * @param mask A mask or list of masks. A mask can be either a tensor or null\n   *   (no mask).\n   *\n   * @return A tensor if there is a single output, or a list of tensors if there\n   *   are more than one outputs.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      const feedDict = new FeedDict();\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n      return execute(this.outputs, feedDict, kwargs) as Tensor | Tensor[];\n    });\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      let masks: Tensor[];\n      if (mask == null) {\n        masks = generic_utils.pyListRepeat(null, inputs.length);\n      } else {\n        masks = generic_utils.toList(mask);\n      }\n      // TODO(michaelterry): Add support for mask caching.\n      return this.runInternalGraph(inputs, masks)[1];\n    });\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    const inputShapes = types_utils.normalizeShapeList(inputShape);\n    if (inputShapes.length !== this.inputLayers.length) {\n      throw new ValueError(\n          `Invalid inputShape argument ${inputShape}: ` +\n          `model has ${this.inputLayers.length} tensor inputs.`);\n    }\n\n    // TODO(michaelterry): Add caching\n    const layersToOutputShapes: {[shapeKey: string]: Shape} = {};\n    for (let i = 0; i < inputShapes.length; i++) {\n      const layer = this.inputLayers[i];\n      const inputShape = inputShapes[i];\n      // It's an input layer: computeOutputShape is identity,\n      // and there is only one node and one tensor output.\n      const shapeKey = layer.name + '_0_0';\n      layersToOutputShapes[shapeKey] = inputShape;\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    // Iterate over nodes, by depth level.\n    if (depthKeys.length > 1) {\n      for (const depth of depthKeys) {\n        const nodes = this.nodesByDepth[depth];\n        for (const node of nodes) {\n          // This is always a single layer, never a list.\n          const layer = node.outboundLayer;\n          if (this.inputLayers.map(x => x.id).indexOf(layer.id) !== -1) {\n            // We've already covered the input layers a few lines above.\n            continue;\n          }\n          // Potentially redundant list, same size of node.inputTensors.\n          const inputShapes: Shape[] = [];\n          for (let j = 0; j < node.inboundLayers.length; j++) {\n            const inboundLayer = node.inboundLayers[j];\n            const nodeIndex = node.nodeIndices[j];\n            const tensorIndex = node.tensorIndices[j];\n            const shapeKey = `${inboundLayer.name}_${nodeIndex}_${tensorIndex}`;\n            const inputShape = layersToOutputShapes[shapeKey];\n            inputShapes.push(inputShape);\n          }\n\n          const outputShape = layer.computeOutputShape(\n              generic_utils.singletonOrArray(inputShapes));\n\n          const outputShapes = types_utils.normalizeShapeList(outputShape);\n          const nodeIndex = layer.inboundNodes.indexOf(node);\n          for (let j = 0; j < outputShapes.length; j++) {\n            const shapeKey = `${layer.name}_${nodeIndex}_${j}`;\n            layersToOutputShapes[shapeKey] = outputShapes[j];\n          }\n        }\n      }\n    }\n\n    // Read final output shapes from layersToOutputShapes.\n    const outputShapes: Shape[] = [];\n    const outputShapeKeys: string[] = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      const shapeKey = `${layer.name}_${nodeIndex}_${tensorIndex}`;\n      outputShapeKeys.push(shapeKey);\n    }\n\n    for (let i = 0; i < outputShapeKeys.length; i++) {\n      const key = outputShapeKeys[i];\n      generic_utils.assert(key in layersToOutputShapes);\n      outputShapes.push(layersToOutputShapes[key]);\n    }\n\n    // TODO(michaelterry): Update cache\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  /**\n   * Computes output tensors for new inputs.\n   *\n   * Note:\n   *   - Expects `inputs` to be a list (potentially with 1 element).\n   *\n   * @param inputs List of tensors\n   * @param masks List of masks (tensors or null).\n   * @return Three lists: outputTensors, outputMasks, outputShapes\n   */\n  protected runInternalGraph(inputs: Tensor[], masks?: Tensor[]):\n      [Tensor[], Tensor[], Shape[]] {\n    if (masks == null) {\n      masks = generic_utils.pyListRepeat(null, inputs.length);\n    }\n\n    // Dictionary mapping reference tensors to tuples\n    // (computed tensor, compute mask)\n    // we assume a 1:1 mapping from tensor to mask\n    // TODO: raise exception when a `.computeMask()` call\n    // does not return a list the same size as `call`\n    const tensorMap: {[tensorID: string]: [Tensor, Tensor]} = {};\n    for (let i = 0; i < this.inputs.length; ++i) {\n      const x = this.inputs[i];\n      const y = inputs[i];\n      const mask = masks[i];\n      tensorMap[x.id] = [y, mask];\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    for (const depth of depthKeys) {\n      const nodes = this.nodesByDepth[depth];\n      for (const node of nodes) {\n        // This is always a single layer, never a list.\n        const layer = node.outboundLayer;\n        const referenceInputTensors = node.inputTensors;\n        const referenceOutputTensors = node.outputTensors;\n\n        // If all previous input tensors are available in tensorMap,\n        // then call node.inboundLayer on them.\n        // List of tuples [input, mask]:\n        const computedData = new Array<[Tensor, Tensor]>();\n        for (const x of referenceInputTensors) {\n          if (x.id in tensorMap) {\n            computedData.push(tensorMap[x.id]);\n          }\n        }\n        if (computedData.length === referenceInputTensors.length) {\n          // TODO(michaelterry): Add K.name_scope here, if we need it.\n          let kwargs: Kwargs = {};\n          let computedTensors: Tensor[];\n          let computedMasks: Tensor[];\n          let outputTensors: Tensor[];\n          let outputMasks: Tensor[];\n          // call layer\n          if (node.callArgs != null) {\n            kwargs = node.callArgs;\n          }\n          if (computedData.length === 1) {\n            const [computedTensor, computedMask] = computedData[0];\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMask;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensor, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensor, computedMask));\n            computedTensors = [computedTensor];\n            computedMasks = [computedMask];\n          } else {\n            computedTensors = computedData.map(x => x[0]);\n            computedMasks = computedData.map(x => x[1]);\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMasks;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensors, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensors, computedMasks));\n          }\n\n          if (layer.activityRegularizer) {\n            throw new NotImplementedError(\n                'LayersModel invocation with concrete Tensor value(s) in the ' +\n                'presence of activity regularizer(s) is not supported yet.');\n          }\n          // TODO(michaelterry): Add model updates and losses\n\n          // Update tensor map.\n          for (let i = 0; i < referenceOutputTensors.length; ++i) {\n            const x = referenceOutputTensors[i];\n            const y = outputTensors[i];\n            const mask = outputMasks[i];\n            tensorMap[x.id] = [y, mask];\n          }\n        }\n      }\n    }\n\n    const outputTensors: Tensor[] = [];\n    const outputMasks: Tensor[] = [];\n    const outputShapes: Shape[] = [];\n    for (const x of this.outputs) {\n      generic_utils.assert(\n          x.id in tensorMap, `Could not compute output ${x.name} : ${x.id}`);\n      const [tensor, mask] = tensorMap[x.id];\n      outputShapes.push(tensor.shape);\n      outputTensors.push(tensor);\n      outputMasks.push(mask);\n    }\n\n    // TODO(michaelterry): Add support for caches.\n    return [outputTensors, outputMasks, outputShapes];\n  }\n\n  /**\n   * Builds a map of internal node keys to node ordering.\n   * Used in serializaion a node orderings may change as unused nodes are\n   * dropped. Porting Note:  This helper method was pulled out of getConfig to\n   * improve readability.\n   * @param layers An array of Layers in the model.\n   * @returns Map of Node Keys to index order within the layer.\n   */\n  private buildNodeConversionMap(layers: Layer[]): {[nodeKey: string]: number} {\n    const nodeConversionMap: {[nodeKey: string]: number} = {};\n    let keptNodes: number;\n    for (const layer of this.layers) {\n      keptNodes = layer instanceof Container ? 1 : 0;\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        if (this.containerNodes.has(nodeKey)) {\n          // i.e. we mark it to be saved\n          nodeConversionMap[nodeKey] = keptNodes;\n          keptNodes += 1;\n        }\n      }\n    }\n    return nodeConversionMap;\n  }\n\n  /**\n   * Retrieves a layer based on either its name (unique) or index.\n   *\n   * Indices are based on order of horizontal graph traversal (bottom-up).\n   *\n   * If both `name` and `index` are specified, `index` takes precedence.\n   *\n   * @param name Name of layer.\n   * @param index Index of layer.\n   * @returns A Layer instance.\n   * @throws ValueError: In case of invalid layer name or index.\n   *\n   * @doc {\n   *    heading: 'Layers',\n   *    subheading: 'Classes',\n   *    namespace: 'layers',\n   *    subclasses: ['LayersModel']\n   * }\n   */\n  getLayer(name?: string, index?: number): Layer {\n    if (index != null) {\n      if (this.layers.length <= index) {\n        throw new ValueError(\n            `Was asked to retrieve layer at index ${index}, but model only ` +\n            `has ${this.layers.length} layer(s).`);\n      } else {\n        return this.layers[index];\n      }\n    } else {\n      if (name == null) {\n        throw new ValueError('Provide either a layer name or layer index');\n      }\n    }\n\n    for (const layer of this.layers) {\n      if (layer.name === name) {\n        return layer;\n      }\n    }\n    throw new ValueError(`No such layer: ${name}`);\n  }\n\n  /**\n   * Retrieves the Container's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Container.loss in PyKeras.\n    //   In PyKeras, Container.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return tidy(() => {\n      const losses: Scalar[] = [];\n      for (const layer of this.layers) {\n        for (let nodeIndex = 0; nodeIndex < layer.inboundNodes.length;\n             ++nodeIndex) {\n          const nodeKey = Container.nodeKey(layer, nodeIndex);\n          if (this.containerNodes.has(nodeKey)) {\n            losses.push(...layer.calculateLosses());\n          }\n        }\n      }\n      // TODO(cais): Add any unconditional model-level losses?\n      return losses;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {name: this.name};\n\n    // Build a map from layer unique name (self._node_key)\n    // to the index of the nodes that are saved in the config.\n    // Only nodes in container_nodes are saved.\n    const nodeConversionMap: {[nodeKey: string]: number} =\n        this.buildNodeConversionMap(this.layers);\n\n    // Serialize and save the layers in layerConfigs\n    const layerConfigs = [];\n    for (const layer of this.layers) {\n      const layerClassName = layer.getClassName();\n      const layerConfig = layer.getConfig();\n      const filteredInboundNodes = [];\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const node = layer.inboundNodes[originalNodeIndex];\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        let kwargs = {};\n        if (this.containerNodes.has(nodeKey)) {\n          // The node is relevant to the model:\n          // add to filteredInboundNodes.\n          if (node.callArgs) {\n            try {\n              JSON.stringify(node.callArgs);\n              kwargs = node.callArgs;\n            } catch (err) {\n              console.warn(\n                  `Layer ${layer.name} was passed ` +\n                  `non-serializable keyword arguments: ` +\n                  `${node.callArgs}. They will not be included ` +\n                  `in the serialized model (and thus will be ` +\n                  `missing at deserialization time).`);\n              kwargs = {};\n            }\n          }\n          if (node.inboundLayers.length > 0) {\n            const nodeData = [];\n            for (let i = 0; i < node.inboundLayers.length; i++) {\n              const inboundLayer = node.inboundLayers[i];\n              const nodeIndex = node.nodeIndices[i];\n              const tensorIndex = node.tensorIndices[i];\n              const nodeKey = Container.nodeKey(inboundLayer, nodeIndex);\n              let newNodeIndex = nodeConversionMap[nodeKey];\n              if (newNodeIndex == null) {\n                newNodeIndex = 0;\n              }\n              nodeData.push(\n                  [inboundLayer.name, newNodeIndex, tensorIndex, kwargs]);\n            }\n            filteredInboundNodes.push(nodeData);\n          }\n        }\n      }\n      const dict: serialization.ConfigDict = {};\n      dict['name'] = layer.name;\n      dict['className'] = layerClassName;\n      dict['config'] = layerConfig;\n      dict['inboundNodes'] = filteredInboundNodes;\n      layerConfigs.push(dict);\n    }\n    config['layers'] = layerConfigs;\n    // Gather info about inputs and outputs\n    const modelInputs = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      const nodeIndex = this.inputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.inputLayersTensorIndices[i];\n      modelInputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['inputLayers'] = modelInputs;\n\n    const modelOutputs = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      modelOutputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['outputLayers'] = modelOutputs;\n    return config;\n  }\n\n  /**\n   * Instantiates a LayersModel from its config (output of `get_config()`).\n   * @param cls the class to create\n   * @param config LayersModel config dictionary.\n   * @param customObjects An optional dictionary of custom objects.\n   * @param fastWeightInit Optional flag to use fast weight initialization\n   *   during deserialization. This is applicable to cases in which\n   *   the initialization will be immediately overwritten by loaded weight\n   *   values. Default: `false`.\n   * @returns A LayersModel instance.\n   * @throws ValueError: In case of improperly formatted config dict.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    // Layer instances created during\n    // the graph reconstruction process\n    const createdLayers: {[layerName: string]: Layer} = {};\n\n    // Dictionary mapping layer instances to\n    // node data that specifies a layer call.\n    // It acts as a queue that maintains any unprocessed\n    // layer call until it becomes possible to process it\n    // (i.e. until the input tensors to the call all exist).\n    const unprocessedNodes: {[layer: string]: TensorKeyWithArgsArray[][]} = {};\n    function addUnprocessedNode(\n        layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      if (!(layer.name in unprocessedNodes)) {\n        unprocessedNodes[layer.name] = [nodeData];\n      } else {\n        unprocessedNodes[layer.name].push(nodeData);\n      }\n    }\n\n    function processNode(layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      const inputTensors: SymbolicTensor[] = [];\n      let kwargs;\n      for (const inputData of nodeData) {\n        const inboundLayerName = inputData[0];\n        const inboundNodeIndex = inputData[1];\n        const inboundTensorIndex = inputData[2];\n\n        kwargs = inputData[3] == null ?\n            {} :\n            inputData[3] as serialization.ConfigDict;\n        if (!(inboundLayerName in createdLayers)) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundLayer = createdLayers[inboundLayerName];\n        if (inboundLayer.inboundNodes.length <= inboundNodeIndex) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundNode = inboundLayer.inboundNodes[inboundNodeIndex];\n        inputTensors.push(inboundNode.outputTensors[inboundTensorIndex]);\n      }\n      // Call layer on its inputs, thus creating the node\n      // and building the layer if needed.\n      // Note: This has Eager vs Graph Implications.\n      if (inputTensors.length > 0) {\n        layer.apply(\n            generic_utils.singletonOrArray(inputTensors),\n            kwargs);  // was ** kwargs\n      }\n    }\n\n    /**\n     * Deserialize a layer, then call it on appropriate inputs.\n     * @param layerData: layer config dict.\n     * @throws ValueError: In case of improperly formatted `layer_data`\n     * dict.\n     */\n    function processLayer(layerData: serialization.ConfigDict|null) {\n      const layerName = layerData['name'] as string;\n      // Instantiate layer.\n      const layer =\n          deserializeLayer(\n              layerData,\n              config['customObjects'] != null ?\n                  config['customObjects'] as serialization.ConfigDict :\n                  {}) as Layer;\n      layer.setFastWeightInitDuringBuild(fastWeightInit);\n      createdLayers[layerName] = layer;\n      // Gather layer inputs.\n      const inboundNodesData =\n          layerData['inboundNodes'] as TensorKeyWithArgsArray[][];\n      inboundNodesData.forEach(nodeData => {\n        if (!(nodeData instanceof Array)) {\n          throw new ValueError(\n              `Corrupted configuration, expected array for nodeData: ${\n                  nodeData}`);\n        }\n        // We don't process nodes (i.e. make layer calls)\n        // on the fly because the inbound node may not yet exist,\n        // in case of layer shared at different topological depths\n        // (e.g.a model such as A(B(A(B(x)))))\n        addUnprocessedNode(layer, nodeData);\n      });\n    }\n\n    // First, we create all layers and enqueue nodes to be processed.\n    const name = config['name'];\n    const layersFromConfig = config['layers'] as serialization.ConfigDict[];\n    for (const layerData of layersFromConfig) {\n      processLayer(layerData);\n    }\n\n    // Then we process nodes in order of layer depth.\n    // Nodes that cannot yet be processed(if the inbound node\n    // does not yet exist) are re - enqueued, and the process\n    // is repeated until all nodes are processed.\n    while (!generic_utils.isObjectEmpty(unprocessedNodes)) {\n      for (const layerData of layersFromConfig) {\n        const layer = createdLayers[layerData['name'] as string];\n        if (layer.name in unprocessedNodes) {\n          const currentUnprocessedNodesForLayer = unprocessedNodes[layer.name];\n          delete unprocessedNodes[layer.name];\n          for (const nodeData of currentUnprocessedNodesForLayer) {\n            processNode(layer, nodeData);\n          }\n        }\n      }\n    }\n\n    const inputTensors: SymbolicTensor[] = [];\n    const outputTensors: SymbolicTensor[] = [];\n    const inputLayersFromConfig =\n        config['inputLayers'] as serialization.ConfigDict[];\n    for (const layerData of inputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      inputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    const outputLayersFromConfig =\n        config['outputLayers'] as serialization.ConfigDict[];\n    for (const layerData of outputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      outputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    return new cls({inputs: inputTensors, outputs: outputTensors, name});\n  }\n\n  /**\n   * Determine whether the container is stateful.\n   *\n   * Porting Note: this is the equivalent of the stateful @property of\n   *   the Container class in PyKeras.\n   */\n  get stateful(): boolean {\n    // Porting Note: This check is to prevent inadvertent setting of the\n    //   _stateful property of the Container instance.\n    if (this._stateful) {\n      throw new ValueError(\n          'Container instance unexpectedly has _stateful = true. The ' +\n          'statefulness of a Container is determined by the Layers it ' +\n          'contains. Its _stateful property must remain the default false.');\n    }\n    for (const layer of this.layers) {\n      if (layer.stateful) {\n        return true;\n      }\n    }\n    return false;\n  }\n\n  /**\n   * Reset the state of all stateful constituent layers (if any).\n   *\n   * Examples of stateful layers include RNN layers whose `stateful` property\n   * is set as `true`.\n   */\n  resetStates() {\n    tidy(() => {\n      this.layers.forEach(layer => {\n        // tslint:disable:no-any\n        if (layer.stateful) {\n          layer.resetStates();\n        }\n        // tslint:enable:no-any\n      });\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, mul, Tensor, Tensor1D, tensor1d, tidy} from '@tensorflow/tfjs-core';\n\n/**\n * For multi-class classification problems, this object is designed to store a\n * mapping from class index to the \"weight\" of the class, where higher weighted\n * classes have larger impact on loss, accuracy, and other metrics.\n *\n * This is useful for cases in which you want the model to \"pay more attention\"\n * to examples from an under-represented class, e.g., in unbalanced datasets.\n */\nexport type ClassWeight = {\n  [classIndex: number]: number\n};\n\n/**\n * Class weighting for a model with multiple outputs.\n *\n * This object maps each output name to a class-weighting object.\n */\nexport type ClassWeightMap = {\n  [outputName: string]: ClassWeight\n};\n\nfunction standardizeSampleOrClassWeights(\n    xWeight: ClassWeight|ClassWeight[]|ClassWeightMap, outputNames: string[],\n    weightType: 'sampleWeight'|'classWeight'): ClassWeight[] {\n  const numOutputs = outputNames.length;\n  if (xWeight == null || (Array.isArray(xWeight) && xWeight.length === 0)) {\n    return outputNames.map(name => null);\n  }\n  if (numOutputs === 1) {\n    if (Array.isArray(xWeight) && xWeight.length === 1) {\n      return xWeight;\n    } else if (typeof xWeight === 'object' && outputNames[0] in xWeight) {\n      return [(xWeight as ClassWeightMap)[outputNames[0]]];\n    } else {\n      return [xWeight as ClassWeight];\n    }\n  }\n  if (Array.isArray(xWeight)) {\n    if (xWeight.length !== numOutputs) {\n      throw new Error(\n          `Provided ${weightType} is an array of ${xWeight.length} ` +\n          `element(s), but the model has ${numOutputs} outputs. ` +\n          `Make sure a set of weights is provided for each model output.`);\n    }\n    return xWeight;\n  } else if (\n      typeof xWeight === 'object' && Object.keys(xWeight).length > 0 &&\n      typeof (xWeight as ClassWeightMap)[Object.keys(xWeight)[0]] ===\n          'object') {\n    const output: ClassWeight[] = [];\n    outputNames.forEach(outputName => {\n      if (outputName in xWeight) {\n        output.push((xWeight as ClassWeightMap)[outputName]);\n      } else {\n        output.push(null);\n      }\n    });\n    return output;\n  } else {\n    throw new Error(\n        `The model has multiple (${numOutputs}) outputs, ` +\n        `so ${weightType} must be either an array with ` +\n        `${numOutputs} elements or an object with ${outputNames} keys. ` +\n        `Provided ${weightType} not understood: ${JSON.stringify(xWeight)}`);\n  }\n}\n\n/**\n * Standardize class weighting objects.\n *\n * This function takes a single class-weighting object, an array of them,\n * or a map from output name to class-weighting object. It compares it to the\n * output name(s) of the model, base on which it outputs an array of\n * class-weighting objects of which the length matches the number of outputs.\n *\n * @param classWeight Input class-weighting object(s).\n * @param outputNames All output name(s) of the model.\n * @return An array of class-weighting objects. The length of the array matches\n *   the model's number of outputs.\n */\nexport function standardizeClassWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'classWeight');\n}\n\nexport function standardizeSampleWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'sampleWeight');\n}\n\n/**\n * Standardize by-sample and/or by-class weights for training.\n *\n * Note that this function operates on one model output at a time. For a model\n * with multiple outputs, you must call this function multiple times.\n *\n * @param y The target tensor that the by-sample and/or by-class weight is for.\n *     The values of y are assumed to encode the classes, either directly\n *     as an integer index, or as one-hot encoding.\n * @param sampleWeight By-sample weights.\n * @param classWeight By-class weights: an object mapping class indices\n *     (integers) to a weight (float) to apply to the model's loss for the\n *     samples from this class during training. This can be useful to tell the\n *     model to \"pay more attention\" to samples from an under-represented class.\n * @param sampleWeightMode The mode for the sample weights.\n * @return A Promise of weight tensor, of which the size of the first dimension\n *     matches that of `y`.\n */\nexport async function standardizeWeights(\n    y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight,\n    sampleWeightMode?: 'temporal'): Promise<Tensor> {\n  if (sampleWeight != null || sampleWeightMode != null) {\n    // TODO(cais): Once 'temporal' mode is implemented, document it in the doc\n    // string.\n    throw new Error('Support sampleWeight is not implemented yet');\n  }\n\n  if (classWeight != null) {\n    // Apply class weights per sample.\n    const yClasses: Tensor1D = tidy(() => {\n      if (y.shape.length === 1) {\n        // Assume class indices.\n        return y.clone() as Tensor1D;\n      } else if (y.shape.length === 2) {\n        if (y.shape[1] > 1) {\n          // Assume one-hot encoding of classes.\n          const axis = 1;\n          return y.argMax(axis);\n        } else if (y.shape[1] === 1) {\n          // Class index.\n          return y.reshape([y.shape[0]]);\n        } else {\n          throw new Error(\n              `Encountered unexpected last-dimension size (${y.shape[1]}) ` +\n              `during handling of class weights. The size is expected to be ` +\n              `>= 1.`);\n        }\n      } else {\n        throw new Error(\n            `Unexpected rank of target (y) tensor (${y.rank}) during ` +\n            `handling of class weights. The rank is expected to be 1 or 2.`);\n      }\n    });\n\n    const yClassIndices = Array.from(await yClasses.data());\n    dispose(yClasses);\n    const classSampleWeight: number[] = [];\n    yClassIndices.forEach(classIndex => {\n      if (classWeight[classIndex] == null) {\n        throw new Error(\n            `classWeight must contain all classes in the training data. ` +\n            `The class ${classIndex} exists in the data but not in ` +\n            `classWeight`);\n      } else {\n        classSampleWeight.push(classWeight[classIndex]);\n      }\n    });\n\n    return tensor1d(classSampleWeight, 'float32');\n  } else {\n    return null;\n  }\n}\n\n/**\n * Apply per-sample weights on the loss values from a number of samples.\n *\n * @param losses Loss tensor of shape `[batchSize]`.\n * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.\n * @returns Tensor of the same shape as`losses`.\n */\nexport function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor) {\n  return mul(losses, sampleWeights);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using TensorFlow.js datasets.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar} from '@tensorflow/tfjs-core';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {TensorOrArrayOrMap} from '../types';\nimport {singletonOrArray, toList} from '../utils/generic_utils';\n\nimport {Dataset, LazyIterator} from './dataset_stub';\nimport {ClassWeight, ClassWeightMap, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Interface for configuring model training based on a dataset object.\n */\nexport interface ModelFitDatasetArgs<T> {\n  /**\n   * (Optional) Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. It should\n   * typically be equal to the number of samples of your dataset divided by\n   * the batch size, so that `fitDataset`() call can utilize the entire dataset.\n   * If it is not provided, use `done` return value in `iterator.next()` as\n   * signal to finish an epoch.\n   */\n  batchesPerEpoch?: number;\n\n  /**\n   * Integer number of times to iterate over the training dataset.\n   */\n  epochs: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be any of the following:\n   *\n   *   - An array `[xVal, yVal]`, where the two values may be `tf.Tensor`,\n   *     an array of Tensors, or a map of string to Tensor.\n   *   - Similarly, an array ` [xVal, yVal, valSampleWeights]`\n   *     (not implemented yet).\n   *   - a `Dataset` object with elements of the form `{xs: xVal, ys: yVal}`,\n   *     where `xs` and `ys` are the feature and label tensors, respectively.\n   *\n   * If `validationData` is an Array of Tensor objects, each `tf.Tensor` will be\n   * sliced into batches during validation, using the parameter\n   * `validationBatchSize` (which defaults to 32). The entirety of the\n   * `tf.Tensor` objects will be used in the validation.\n   *\n   * If `validationData` is a dataset object, and the `validationBatches`\n   * parameter is specified, the validation will use `validationBatches` batches\n   * drawn from the dataset object. If `validationBatches` parameter is not\n   * specified, the validation will stop when the dataset is exhausted.\n   *\n   * The model will not be trained on this data.\n   */\n  validationData?: [\n    TensorOrArrayOrMap, TensorOrArrayOrMap\n  ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|Dataset<T>;\n\n  /**\n   * Optional batch size for validation.\n   *\n   * Used only if `validationData` is an array of `tf.Tensor` objects, i.e., not\n   * a dataset object.\n   *\n   * If not specified, its value defaults to 32.\n   */\n  validationBatchSize?: number;\n\n  /**\n   * (Optional) Only relevant if `validationData` is specified and is a dataset\n   * object.\n   *\n   * Total number of batches of samples to draw from `validationData` for\n   * validation purpose before stopping at the end of every epoch. If not\n   * specified, `evaluateDataset` will use `iterator.next().done` as signal to\n   * stop validation.\n   */\n  validationBatches?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - a `number`: Will yield every `number` milliseconds.\n   *   - `'never'`: never yield. (But yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n}\n\nexport interface FitDatasetElement {\n  xs: TensorOrArrayOrMap;\n  ys: TensorOrArrayOrMap;\n}\n\n/**\n * Interface for configuring model evaluation based on a dataset object.\n */\nexport interface ModelEvaluateDatasetArgs {\n  /**\n   * Number of batches to draw from the dataset object before ending the\n   * evaluation.\n   */\n  batches?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n}\n\n// Default batch size used during tensor-based validation.\nconst DEFAULT_VALIDATION_BATCH_SIZE = 32;\n\n/**\n * Standardize the output of a dataset iterator for use by\n * LayersModel.fitDataset().\n *\n * @param model: A `tf.LayersModel` object.\n * @param iteratorOut The output of a dataset iterator. It is required to be\n *   an object of the form `{xs: TensorOrArrayOrMap, ys:\n * TensorOrArrayOrMap}`, where `TensorOrArrayOrMap` is a single `tf.Tensor`,\n * a `tf.Tensor[]`, or a flat map from string names to `tf.Tensor`s.\n * @returns A flat array of `tf.Tensor` objects: the input `tf.Tensor`s\n *   followed by the target `tf.Tensor`s.  When `tf.Tensor`s are provided\n *   as a map, the order in the resulting array is taken from the `inputNames`\n *   and `outputNames` of the model.\n */\nfunction standardizeDataIteratorOutput(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, iteratorOut: {}): {xs: tfc.Tensor[], ys: tfc.Tensor[]} {\n  let xs: TensorOrArrayOrMap;\n  let ys: TensorOrArrayOrMap;\n\n  const iteratorOutObj = iteratorOut as FitDatasetElement;\n  xs = iteratorOutObj['xs'];\n  ys = iteratorOutObj['ys'];\n  tfc.util.assert(\n      xs != null && ys != null,\n      () => 'A Dataset iterator for fitDataset() is expected to generate ' +\n          'objects of the form `{xs: xVal, ys: yVal}`, where the two ' +\n          'values may be `tf.Tensor`, an array of Tensors, or a map of ' +\n          'string to Tensor.  The provided Dataset instead generates ' +\n          `${iteratorOut}`);\n\n  const flattenedXs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('input', model.inputNames, xs);\n  const flattenedYs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('output', model.outputNames, ys);\n\n  const batchSize: number = flattenedXs[0].shape[0];\n\n  tfc.util.assert(\n      flattenedXs.length === model.inputs.length,\n      () => `LayersModel has ${model.inputs.length} inputs, but the dataset ` +\n          `provides ${flattenedXs.length} inputs.  (Expected input keys: ` +\n          `${JSON.stringify(model.inputNames)})`);\n\n  tfc.util.assert(\n      flattenedYs.length === model.outputs.length,\n      () =>\n          `LayersModel has ${model.outputs.length} outputs, but the dataset ` +\n          `provides ${flattenedYs.length} outputs.  (Expected output keys: ` +\n          `${JSON.stringify(model.outputNames)})`);\n\n  for (let xIndex = 0; xIndex < flattenedXs.length; xIndex++) {\n    tfc.util.assert(\n        flattenedXs[xIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: input ` +\n            `${model.inputNames[xIndex]} has ${\n                  flattenedXs[xIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  for (let yIndex = 0; yIndex < flattenedYs.length; yIndex++) {\n    tfc.util.assert(\n        flattenedYs[yIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: output ` +\n            `${model.outputNames[yIndex]} has ${\n                  flattenedYs[yIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  return {xs: flattenedXs, ys: flattenedYs};\n}\n\nfunction flattenTensorOrArrayOrMap(\n    inputOrOutput: string, names: string[], values: TensorOrArrayOrMap) {\n  if (values instanceof tfc.Tensor) {\n    return [values];\n  } else if (Array.isArray(values)) {\n    tfc.util.assert(\n        values.length === names.length,\n        () => `Received an array of ${values.length} Tensors, but expected ${\n            names.length} to match the ${inputOrOutput} keys ${names}.`);\n    return values;\n  } else {\n    const result: tfc.Tensor[] = [];\n    // Check that all the required keys are available.\n    for (const name of names) {\n      if (values[name] == null) {\n        throw new ValueError(\n            `The feature data generated by the dataset lacks the required ` +\n            `${inputOrOutput} key '${name}'.`);\n      }\n      result.push(values[name]);\n    }\n    return result;\n  }\n}\n\nfunction standardizeTensorValidationData<T>(\n    data:\n        [\n          tfc.Tensor|tfc.Tensor[], tfc.Tensor|tfc.Tensor[]\n        ]|[tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n           tfc.Tensor | tfc.Tensor[]]):\n    {xs: tfc.Tensor|tfc.Tensor[], ys: tfc.Tensor|tfc.Tensor[]} {\n  if (data.length === 3) {\n    throw new NotImplementedError(\n        'Validation with sample weights is not implemented yet.');\n  }\n  return {xs: data[0], ys: data[1]};\n}\n\nexport async function fitDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>,\n    args: ModelFitDatasetArgs<T>): Promise<History> {\n  const hasBatchesPerEpoch = args.batchesPerEpoch != null;\n  tfc.util.assert(\n      model.optimizer != null,\n      () => 'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileConfig).');\n\n  tfc.util.assert(\n      args != null,\n      () => `For fitDataset(), the 2nd argument (config) is required, ` +\n          `but it is not provided in this call.`);\n  tfc.util.assert(\n      args.epochs != null && args.epochs > 0 && Number.isInteger(args.epochs),\n      () => `For fitDataset(), config.epochs is expected to be a positive ` +\n          `integer, but got ${args.epochs}`);\n  tfc.util.assert(\n      !hasBatchesPerEpoch ||\n          (args.batchesPerEpoch > 0 && Number.isInteger(args.batchesPerEpoch)),\n      () => `For fitDataset(), config.batchesPerEpoch is expected to be a ` +\n          `positive integer if specified, but got ${args.batchesPerEpoch}`);\n  tfc.util.assert(\n      // tslint:disable-next-line:no-any\n      (args as any)['validationSplit'] == null,\n      () => '`validationSplit` is not supported by `fitDataset()`. ' +\n          'Use validationData instead.');\n\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n\n  try {\n    const doValidation = args.validationData != null;\n    let valXs: tfc.Tensor|tfc.Tensor[];\n    let valYs: tfc.Tensor|tfc.Tensor[];\n    if (doValidation) {\n      if (isDatasetObject(args.validationData)) {\n        tfc.util.assert(\n            args.validationBatches == null ||\n                (args.validationBatches > 0 &&\n                 Number.isInteger(args.validationBatches)),\n            () => `For fitDataset() with dataset-based validation, ` +\n                `config.validationBatches is expected not to be provided, ` +\n                `or to be a positive integer, ` +\n                `but got ${args.validationBatches}`);\n      } else {\n        const validationData = standardizeTensorValidationData(\n            args.validationData as\n                    [tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[]] |\n            [\n              tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n              tfc.Tensor | tfc.Tensor[]\n            ]);\n        valXs = validationData.xs;\n        valYs = validationData.ys;\n      }\n    }\n\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let callbackMetrics: string[];\n    if (doValidation) {\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const verbose = args.verbose == null ? 1 : args.verbose;\n    const {callbackList, history} = configureCallbacks(\n        callbacks, verbose, args.epochs, null, null,\n        getStepsPerEpoch(dataset, args),\n        null,  // Batch size determined by the dataset itself.\n        doValidation, callbackMetrics);\n    callbackList.setModel(model);\n    model.history = history;\n\n    await callbackList.onTrainBegin();\n    model.stopTraining_ = false;\n    let epoch = args.initialEpoch == null ? 0 : args.initialEpoch;\n\n    let dataIterator = await dataset.iterator();\n    while (epoch < args.epochs) {\n      const epochLogs: UnresolvedLogs = {};\n      await callbackList.onEpochBegin(epoch);\n      let stepsDone = 0;\n      let batchIndex = 0;\n      if (!hasBatchesPerEpoch) {\n        dataIterator = await dataset.iterator();\n      }\n      while (hasBatchesPerEpoch ? stepsDone < args.batchesPerEpoch : true) {\n        const iteratorOut = await dataIterator.next();\n\n        // If `batchesPerEpoch` is specified, the dataset should not be\n        // exhausted until all epoches are done.\n        if (hasBatchesPerEpoch && iteratorOut.done) {\n          console.warn(\n              'You provided `batchesPerEpoch` as ' +\n              `${args.batchesPerEpoch}, ` +\n              'but your dataset iterator ran out of data after ' +\n              `${stepsDone} batches; ` +\n              'interrupting training. Make sure that your ' +\n              'dataset can generate at least `batchesPerEpoch * epochs` ' +\n              'batches (in this case, ' +\n              `${args.batchesPerEpoch * args.epochs} batches). ` +\n              'You may need to use the repeat() function when building ' +\n              'your dataset.');\n          break;\n        }\n\n        if (iteratorOut.value != null) {\n          const {xs, ys} =\n              standardizeDataIteratorOutput(model, iteratorOut.value);\n          const batchLogs: UnresolvedLogs = {};\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = xs[0].shape[0];\n\n          await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n          const sampleWeights: tfc.Tensor[] = [];\n          if (args.classWeight != null) {\n            const standardClassWeights =\n                standardizeClassWeights(args.classWeight, model.outputNames);\n            for (let i = 0; i < standardClassWeights.length; ++i) {\n              sampleWeights.push(await standardizeWeights(\n                  ys[i], null, standardClassWeights[i]));\n            }\n          }\n\n          // Train on batch.\n          const ins = xs.concat(ys).concat(sampleWeights);\n          const outs = trainFunction(ins);\n          tfc.dispose(ins);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n          }\n\n          await callbackList.onBatchEnd(batchIndex, batchLogs);\n          disposeTensorsInLogs(batchLogs);\n\n          batchIndex++;\n          stepsDone++;\n        }\n\n        if (hasBatchesPerEpoch ? stepsDone >= args.batchesPerEpoch :\n                                 iteratorOut.done) {\n          // Epoch finished. Perform validation.\n          if (doValidation) {\n            let valOuts: tfc.Scalar[];\n            if (isDatasetObject(args.validationData)) {\n              valOuts = toList(await model.evaluateDataset(\n                  args.validationData, {batches: args.validationBatches}));\n            } else {\n              valOuts = toList(model.evaluate(valXs, valYs, {\n                batchSize: args.validationBatchSize == null ?\n                    DEFAULT_VALIDATION_BATCH_SIZE :\n                    args.validationBatchSize,\n                verbose: 0\n              }));\n            }\n            for (let i = 0; i < model.metricsNames.length; ++i) {\n              epochLogs[`val_${model.metricsNames[i]}`] = valOuts[i];\n            }\n          }\n          // Call `break` to exit one epoch lopp after validation is done. If\n          // config.batchesPerEpoch is specified, an epoch while loop will\n          // stop when `stepsDone >= config.batchesPerEpoch`. When\n          // config.batchesPerEpoch is not provided, the following `break` is\n          // required to exit the while lopp after dataset is exhausted.\n          break;\n        }\n\n        if (model.stopTraining_) {\n          break;\n        }\n      }\n      await callbackList.onEpochEnd(epoch, epochLogs);\n      epoch++;\n      if (model.stopTraining_) {\n        break;\n      }\n    }\n    await callbackList.onTrainEnd();\n    await model.history.syncData();\n    return model.history;\n  } finally {\n    model.isTraining = false;\n  }\n}\n\n/** Helper function that determines number of steps (batches) per epoch. */\nfunction getStepsPerEpoch<T>(\n    dataset: Dataset<T>, args: ModelFitDatasetArgs<T>): number {\n  // Attempt to determine # of batches in an epoch.\n  let stepsPerEpoch: number = null;\n  if (args.batchesPerEpoch != null) {\n    stepsPerEpoch = args.batchesPerEpoch;\n  } else if (Number.isFinite(dataset.size)) {\n    stepsPerEpoch = dataset.size;\n  }\n  return stepsPerEpoch;\n}\n\n// Check if provided object is a Dataset object by checking its .iterator\n// element.\nfunction isDatasetObject<T>(\n    dataset:\n        [\n          TensorOrArrayOrMap, TensorOrArrayOrMap\n        ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|\n    Dataset<T>): boolean {\n  return (typeof (dataset as Dataset<T>).iterator === 'function');\n}\n\n// Check if provided object is a LazyIterator object by checking it's .next\n// element.\nfunction isLazyIteratorObject<T>(iterator: Dataset<T>|\n                                 LazyIterator<T>): boolean {\n  return (typeof (iterator as LazyIterator<T>).next === 'function');\n}\n\nexport async function evaluateDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>|LazyIterator<T>,\n    args: ModelEvaluateDatasetArgs): Promise<tfc.Scalar|tfc.Scalar[]> {\n  args = args || {};\n  const hasBatches = args.batches != null;\n  const f = model.testFunction;\n  let outs: tfc.Scalar[] = [];\n  if (args.verbose > 0) {\n    throw new NotImplementedError('Verbose mode is not implemented yet.');\n  }\n\n  tfc.util.assert(\n      !hasBatches || (args.batches > 0 && Number.isInteger(args.batches)),\n      () => 'Test loop expects `batches` to be a positive integer, but ' +\n          `received ${JSON.stringify(args.batches)}`);\n  const dataIterator = isLazyIteratorObject(dataset) ?\n      dataset as LazyIterator<T>:\n      await (dataset as Dataset<T>).iterator();\n  // Keeps track of number of examples used in this evaluation.\n  let numExamples = 0;\n  let batch = 0;\n\n  while (hasBatches ? batch < args.batches : true) {\n    const iteratorOut = await dataIterator.next();\n    outs = tfc.tidy(() => {\n      if (iteratorOut.value) {\n        // TODO(cais): Once real dataset is available, use\n        //   `map(x => standardizeDataIteratorOutput(model, x).map(f)`.\n        const {xs, ys} =\n            standardizeDataIteratorOutput(model, iteratorOut.value);\n        const xsAndYs = xs.concat(ys);\n        const batchOuts = tfc.tidy(() => f(xsAndYs));\n        tfc.dispose(xsAndYs);\n\n        if (batch === 0) {\n          for (let i = 0; i < batchOuts.length; ++i) {\n            outs.push(scalar(0));\n          }\n        }\n\n        const batchSize = xsAndYs[0].shape[0];\n        for (let i = 0; i < batchOuts.length; ++i) {\n          const batchOut = batchOuts[i];\n          const oldScalar = outs[i];\n          outs[i] =\n              tfc.tidy(() => tfc.add(outs[i], tfc.mul(batchSize, batchOut)));\n          if (batch > 0) {\n            tfc.dispose(oldScalar);\n          }\n        }\n        tfc.dispose(batchOuts);\n        numExamples += batchSize;\n\n        ++batch;\n      }\n      return outs;\n    });\n\n    if (iteratorOut.done) {\n      if (hasBatches) {\n        console.warn(\n            'Your dataset iterator ran out of data during evaluateDataset(). ' +\n            'Interrupting evalution. Make sure that your ' +\n            'dataset can generate at least `batches` ' +\n            `batches (in this case, ${args.batches} batches). ` +\n            'You may need to use the repeat() function when building ' +\n            'your dataset.');\n      }\n      break;\n    }\n  }\n\n  for (let i = 0; i < outs.length; ++i) {\n    const oldScalar = outs[i];\n    outs[i] = tfc.div(outs[i], numExamples);\n    tfc.dispose(oldScalar);\n  }\n\n  return singletonOrArray(outs);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using tf.Tensor objects.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Scalar, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport {expandDims, gather, sliceAlongFirstAxis} from '../backend/tfjs_backend';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {range} from '../utils/math_utils';\nimport {ClassWeight, ClassWeightMap} from './training_utils';\n\n/**\n * Interface configuration model training based on data as `tf.Tensor`s.\n */\nexport interface ModelFitArgs {\n  /**\n   * Number of samples per gradient update. If unspecified, it\n   * will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Integer number of times to iterate over the training data arrays.\n   */\n  epochs?: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Float between 0 and 1: fraction of the training data\n   * to be used as validation data. The model will set apart this fraction of\n   * the training data, will not train on it, and will evaluate the loss and\n   * any model metrics on this data at the end of each epoch.\n   * The validation data is selected from the last samples in the `x` and `y`\n   * data provided, before shuffling.\n   */\n  validationSplit?: number;\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal,\n   * valSampleWeights]. The model will not be trained on this data.\n   * `validationData` will override `validationSplit`.\n   */\n  validationData?: [\n    Tensor|Tensor[], Tensor|Tensor[]\n  ]|[Tensor | Tensor[], Tensor|Tensor[], Tensor|Tensor[]];\n\n  /**\n   * Whether to shuffle the training data before each epoch. Has\n   * no effect when `stepsPerEpoch` is not `null`.\n   */\n  shuffle?: boolean;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n\n  /**\n   * Optional array of the same length as x, containing\n   * weights to apply to the model's loss for each sample. In the case of\n   * temporal data, you can pass a 2D array with shape (samples,\n   * sequenceLength), to apply a different weight to every timestep of every\n   * sample. In this case you should make sure to specify\n   * sampleWeightMode=\"temporal\" in compile().\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. When training\n   * with Input Tensors such as TensorFlow data tensors, the default `null` is\n   * equal to the number of unique samples in your dataset divided by the\n   * batch size, or 1 if that cannot be determined.\n   */\n  stepsPerEpoch?: number;\n\n  /**\n   * Only relevant if `stepsPerEpoch` is specified. Total number of steps\n   * (batches of samples) to validate before stopping.\n   */\n  validationSteps?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - any `number`: yield every `number` milliseconds.\n   *   - `'never'`: never yield. (yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n}\n\nexport function checkBatchSize(batchSize: number) {\n  tfc.util.assert(\n      batchSize > 0 && Number.isInteger(batchSize),\n      () => `batchSize is required to be a positive integer, but got ${\n          batchSize}`);\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by start and stop indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArraysByIndices()` together.\n *\n * @param arrays: the input.\n * @param start: the starting index (inclusive).\n * @param stop: the stopping index (exclusive).\n * @returns The result of the slicing. If `arrays` is an `Array` of\n *   `tf.Tensor`s, the slicing will be applied to all elements of the `Array`\n *   in the same way.\n */\nexport function sliceArrays(\n    arrays: Tensor|Tensor[], start: number, stop: number): Tensor|Tensor[] {\n  if (arrays == null) {\n    return [null];\n  } else if (Array.isArray(arrays)) {\n    return arrays.map(array => sliceAlongFirstAxis(array, start, stop - start));\n  } else {  // Tensor.\n    return sliceAlongFirstAxis(arrays, start, stop - start);\n  }\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by random-order indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArrays()` together.\n *\n * @param arrays The input `tf.Tensor` or `Array` of `tf.Tensor`s to slice.\n *   If an `Array` of `tf.Tensor`s, all `tf.Tensor`s will be sliced in the\n *   same fashion.\n * @param indices The indices to use for slicing along the first (batch)\n *   dimension.\n * @returns Result(s) of the slicing.\n */\nexport function sliceArraysByIndices(\n    arrays: Tensor|Tensor[], indices: Tensor1D): Tensor|Tensor[] {\n  return tfc.tidy(() => {\n    if (arrays == null) {\n      return null;\n    } else if (Array.isArray(arrays)) {\n      return arrays.map(\n          array => (sliceArraysByIndices(array, indices) as Tensor));\n    } else {\n      // TODO(cais): indices should be a pre-constructed Tensor1D to avoid\n      //   tensor1d() calls.\n      return gather(\n          arrays, indices.dtype === 'int32' ? indices : indices.toInt());\n    }\n  });\n}\n\n/**\n * Returns a list of batch indices (tuples of indices).\n * @param size: Integer, total size of the data to slice into batches.\n * @param batchSize: Integer, batch size.\n * @returns An Array of [batchStart, batchEnd] tuples. batchStart is\n *   inclusive; batchEnd is exclusive. I.e., each batch consists of indices x\n *   that satisfy batchStart <= x < batchEnd.\n */\nexport function makeBatches(\n    size: number, batchSize: number): Array<[number, number]> {\n  const output: Array<[number, number]> = [];\n  let batchStart = 0;\n  let batchEnd: number = null;\n  while (batchStart < size) {\n    batchEnd = batchStart + batchSize;\n    if (batchEnd >= size) {\n      batchEnd = size;\n    }\n    output.push([batchStart, batchEnd]);\n    batchStart = batchEnd;\n  }\n  return output;\n}\n\n/**\n * Abstract fit function for `f(ins)`.\n * @param f A Function returning a list of tensors. For training, this\n *   function is expected to perform the updates to the variables.\n * @param ins List of tensors to be fed to `f`.\n * @param outLabels List of strings, display names of the outputs of `f`.\n * @param batchSize Integer batch size or `== null` if unknown. Default : 32.\n * @param epochs Number of times to iterate over the data. Default : 1.\n * @param verbose Verbosity mode: 0, 1, or 2. Default: 1.\n * @param callbacks List of callbacks to be called during training.\n * @param valF Function to call for validation.\n * @param valIns List of tensors to be fed to `valF`.\n * @param shuffle Whether to shuffle the data at the beginning of every\n * epoch. Default : true.\n * @param callbackMetrics List of strings, the display names of the metrics\n *   passed to the callbacks. They should be the concatenation of the\n *   display names of the outputs of `f` and the list of display names\n *   of the outputs of `valF`.\n * @param initialEpoch Epoch at which to start training (useful for\n *   resuming a previous training run). Default : 0.\n * @param stepsPerEpoch Total number of steps (batches on samples) before\n *   declaring one epoch finished and starting the next epoch. Ignored with\n *   the default value of `undefined` or `null`.\n * @param validationSteps Number of steps to run validation for (only if\n *   doing validation from data tensors). Not applicable for tfjs-layers.\n * @returns A `History` object.\n */\nasync function fitLoop(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, f: (data: Tensor[]) => Scalar[], ins: Tensor[],\n    outLabels?: string[], batchSize?: number, epochs?: number, verbose?: number,\n    callbacks?: BaseCallback[], valF?: (data: Tensor[]) => Scalar[],\n    valIns?: Tensor[], shuffle?: boolean|string, callbackMetrics?: string[],\n    initialEpoch?: number, stepsPerEpoch?: number,\n    validationSteps?: number): Promise<History> {\n  if (batchSize == null) {\n    batchSize = 32;\n  }\n  if (epochs == null) {\n    epochs = 1;\n  }\n  if (shuffle == null) {\n    shuffle = true;\n  }\n  if (initialEpoch == null) {\n    initialEpoch = 0;\n  }\n\n  // TODO(cais): Change const to let below when implementing validation.\n  let doValidation = false;\n  if (valF != null && valIns != null) {\n    doValidation = true;\n    // TODO(cais): verbose message.\n  }\n  if (validationSteps != null) {\n    doValidation = true;\n    if (stepsPerEpoch == null) {\n      throw new ValueError(\n          'Can only use `validationSteps` when doing step-wise training, ' +\n          'i.e., `stepsPerEpoch` must be set.');\n    }\n  }\n\n  const numTrainSamples =\n      model.checkNumSamples(ins, batchSize, stepsPerEpoch, 'steps_per_epoch');\n  let indexArray: number[];\n  if (numTrainSamples != null) {\n    indexArray = range(0, numTrainSamples);\n  }\n\n  if (verbose == null) {\n    verbose = 1;\n  }\n\n  const {callbackList, history} = configureCallbacks(\n      callbacks, verbose, epochs, initialEpoch, numTrainSamples, stepsPerEpoch,\n      batchSize, doValidation, callbackMetrics);\n  callbackList.setModel(model);\n  model.history = history;\n  await callbackList.onTrainBegin();\n  model.stopTraining_ = false;\n  // TODO(cais): Take care of callbacks.validation_data as in PyKeras.\n  // TODO(cais): Pre-convert feeds for performance as in PyKeras.\n\n  for (let epoch = initialEpoch; epoch < epochs; ++epoch) {\n    await callbackList.onEpochBegin(epoch);\n    const epochLogs: UnresolvedLogs = {};\n    if (stepsPerEpoch != null) {\n      throw new NotImplementedError(\n          'stepsPerEpoch mode is not implemented yet.');\n    } else {\n      if (shuffle === 'batch') {\n        throw new NotImplementedError('batch shuffling is not implemneted yet');\n      } else if (shuffle) {\n        util.shuffle(indexArray);\n      }\n      // Convert the potentially shuffled indices to Tensor1D, to avoid the\n      // cost of repeated creation of Array1Ds later on.\n      const epochIndexArray1D = tensor1d(indexArray);\n\n      const batches = makeBatches(numTrainSamples, batchSize);\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchLogs: UnresolvedLogs = {};\n        await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n        tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds = sliceAlongFirstAxis(\n                               epochIndexArray1D, batchStart,\n                               batchEnd - batchStart) as Tensor1D;\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = batchEnd - batchStart;\n\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Tensor[];\n          const outs = f(insBatch);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n            // TODO(cais): Use scope() to avoid ownership.\n          }\n\n          if (batchIndex === batches.length - 1) {  // Last batch.\n            if (doValidation) {\n              const valOuts = model.testLoop(valF, valIns, batchSize);\n              // Porting Notes: In tfjs-layers, valOuts is always an Array.\n              for (let i = 0; i < outLabels.length; ++i) {\n                const label = outLabels[i];\n                const out = valOuts[i];\n                tfc.keep(out);\n                // TODO(cais): Use scope() to avoid ownership.\n                epochLogs['val_' + label] = out;\n              }\n            }\n          }\n        });\n\n        await callbackList.onBatchEnd(batchIndex, batchLogs);\n        disposeTensorsInLogs(batchLogs);\n\n        if (model.stopTraining_) {\n          break;\n        }\n        // TODO(cais): return outs as list of Tensor.\n      }\n\n      epochIndexArray1D.dispose();\n    }\n    // TODO(cais): Run validation at the end of the epoch.\n    await callbackList.onEpochEnd(epoch, epochLogs);\n    if (model.stopTraining_) {\n      break;\n    }\n  }\n  await callbackList.onTrainEnd();\n\n  await model.history.syncData();\n  return model.history;\n}\n\nexport async function fitTensors(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    args: ModelFitArgs = {}): Promise<History> {\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n  let inputs: Tensor[];\n  let targets: Tensor[];\n  let inputValX: Tensor|Tensor[];\n  let inputValY: Tensor|Tensor[];\n  let valX: Tensor|Tensor[];\n  let valY: Tensor|Tensor[];\n  let sampleWeights: Tensor[];\n  try {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // Validate user data.\n    // TODO(cais): Support sampleWeight.\n    const checkBatchAxis = false;\n    const standardizedOuts =\n        await model.standardizeUserData(\n            x, y, args.sampleWeight, args.classWeight, checkBatchAxis,\n            batchSize) as [Tensor[], Tensor[], Tensor[]];\n    inputs = standardizedOuts[0];\n    targets = standardizedOuts[1];\n    sampleWeights = standardizedOuts[2];\n\n    // Prepare validation data.\n    let doValidation = false;\n    let valIns: Tensor[];\n    if (args.validationData != null && args.validationData.length > 0) {\n      doValidation = true;\n      if (args.validationData.length === 2) {\n        // config.validationData consists of valX and valY.\n        inputValX = args.validationData[0];\n        inputValY = args.validationData[1];\n      } else if (args.validationData.length === 3) {\n        throw new NotImplementedError(\n            'validationData including sample weights is not supported yet.');\n      } else {\n        throw new ValueError(\n            `When passing validation data, it must contain 2 (valX, valY) ` +\n            `or 3 (valX, valY, valSampleWeight) items; ` +\n            `${args.validationData} is invalid.`);\n      }\n\n      const checkBatchAxis = true;\n      const valStandardized =\n          await model.standardizeUserData(\n              inputValX, inputValY, null, /** Unused sample weights. */\n              null,                       /** Unused class weights. */\n              checkBatchAxis, batchSize) as [Tensor[], Tensor[], Tensor[]];\n      valX = valStandardized[0];\n      valY = valStandardized[1];\n      valIns = valX.concat(valY);\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (\n        args.validationSplit != null && args.validationSplit > 0 &&\n        args.validationSplit < 1) {\n      doValidation = true;\n      // Porting Note: In tfjs-layers, inputs[0] is always a Tensor.\n      const splitAt =\n          Math.floor(inputs[0].shape[0] * (1 - args.validationSplit));\n      const originalBatchSize = inputs[0].shape[0];\n      valX = sliceArrays(inputs, splitAt, originalBatchSize) as Tensor[];\n      inputs = sliceArrays(inputs, 0, splitAt) as Tensor[];\n      valY = sliceArrays(targets, splitAt, originalBatchSize) as Tensor[];\n      targets = sliceArrays(targets, 0, splitAt) as Tensor[];\n      // TODO(cais): Once sampleWeights becomes available, slice it to get\n      //   valSampleWeights.\n      valIns = valX.concat(valY);\n\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (args.validationSteps != null) {\n      doValidation = true;\n      // TODO(cais): Add useLearningPhase.\n    }\n\n    const ins = inputs.concat(targets).concat(sampleWeights);\n\n    model.checkTrainableWeightsConsistency();\n\n    // TODO(cais): Handle use_learning_phase and learning_phase?\n\n    // Porting Note: Here we see a key deviation of tfjs-layers from\n    // Keras.\n    //  Due to the imperative nature of tfjs-layers' backend (tfjs-core),\n    //  we do not construct symbolic computation graphs to embody the\n    //  training process. Instead, we define a function that performs the\n    //  training action. In PyKeras, the data (inputs and targets) are fed\n    //  through graph placeholders. In tfjs-layers, the data are fed as\n    //  function arguments. Since the function are defined below in the\n    //  scope, we don't have equivalents of PyKeras's\n    //  `_make_train_funciton`.\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let valFunction: (data: Tensor[]) => Scalar[];\n    let callbackMetrics: string[];\n    if (doValidation) {\n      model.makeTestFunction();\n      valFunction = model.testFunction;\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      valFunction = null;\n      valIns = [];\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const out = await fitLoop(\n        model, trainFunction, ins, outLabels, batchSize, args.epochs,\n        args.verbose, callbacks, valFunction, valIns, args.shuffle,\n        callbackMetrics, args.initialEpoch, null, null);\n    return out;\n  } finally {\n    model.isTraining = false;\n    // Memory clean up.\n    disposeNewTensors(inputs, x);\n    disposeNewTensors(targets, y);\n    disposeNewTensors(valX as Tensor[], inputValX);\n    disposeNewTensors(valY as Tensor[], inputValY);\n    if (sampleWeights != null) {\n      tfc.dispose(sampleWeights);\n    }\n  }\n  // TODO(cais): Add value to outLabels.\n}\n\n/**\n * Ensure tensors all have a rank of at least 2.\n *\n * If a tensor has a rank of 1, it is dimension-expanded to rank 2.\n * If any tensor has a rank of 0 (i.e., is a scalar), an error will be thrown.\n */\nexport function ensureTensorsRank2OrHigher(tensors: Tensor|Tensor[]): Tensor[] {\n  const outs: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    tensors = [tensors];\n  }\n\n  // Make Tensors at least 2D.\n  for (let i = 0; i < tensors.length; ++i) {\n    const tensor = tensors[i];\n    if (tensor.rank === 1) {\n      outs.push(expandDims(tensor, 1));\n    } else if (tensor.rank === 0) {\n      throw new Error(\n          'Expected tensor to be at least 1D, but received a 0D tensor ' +\n          '(scalar).');\n    } else {\n      outs.push(tensor);\n    }\n  }\n  return outs;\n}\n\n/**\n * Compare a set of tensors with a reference (old) set, discard the ones\n * in the new set that are not present in the reference set.\n *\n * This method is used for memory clenaup during calls such as\n * LayersModel.fit().\n *\n * @param tensors New set which may contain Tensors not present in\n *   `refTensors`.\n * @param refTensors Reference Tensor set.\n */\n// TODO(cais, kangyizhang): Deduplicate with tfjs-data.\nexport function disposeNewTensors(\n    tensors: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    refTensors: Tensor|Tensor[]|{[inputName: string]: Tensor}): void {\n  if (tensors == null) {\n    return;\n  }\n  const oldTensorIds: number[] = [];\n  if (refTensors instanceof Tensor) {\n    oldTensorIds.push(refTensors.id);\n  } else if (Array.isArray(refTensors)) {\n    refTensors.forEach(t => oldTensorIds.push(t.id));\n  } else if (refTensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in refTensors) {\n      const oldTensor = refTensors[name];\n      oldTensorIds.push(oldTensor.id);\n    }\n  }\n\n  const tensorsToDispose: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    if (oldTensorIds.indexOf(tensors.id) === -1) {\n      tensorsToDispose.push(tensors);\n    }\n  } else if (Array.isArray(tensors)) {\n    tensors.forEach(t => {\n      if (oldTensorIds.indexOf(t.id) === -1) {\n        tensorsToDispose.push(t);\n      }\n    });\n  } else if (tensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in tensors) {\n      const tensor = tensors[name];\n      if (oldTensorIds.indexOf(tensor.id) === -1) {\n        tensorsToDispose.push(tensor);\n      }\n    }\n  }\n\n  tensorsToDispose.forEach(t => {\n    if (!t.isDisposed) {\n      t.dispose();\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: engine/training.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {io, ModelPredictConfig as ModelPredictArgs, NamedTensorMap, Optimizer, Scalar, scalar, serialization, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {History, ModelLoggingVerbosity} from '../base_callbacks';\nimport {nameScope} from '../common';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {LossIdentifier} from '../keras_format/loss_config';\nimport {OptimizerSerialization} from '../keras_format/optimizer_config';\nimport {MetricsIdentifier, TrainingConfig} from '../keras_format/training_config';\nimport {deserialize} from '../layers/serialization';\nimport * as losses from '../losses';\nimport * as Metrics from '../metrics';\nimport * as optimizers from '../optimizers';\nimport {LossOrMetricFn, NamedTensor} from '../types';\nimport {checkUserDefinedMetadata} from '../user_defined_metadata';\nimport {count, pyListRepeat, singletonOrArray, toCamelCase, toSnakeCase, unique} from '../utils/generic_utils';\nimport {printSummary} from '../utils/layer_utils';\nimport {range} from '../utils/math_utils';\nimport {convertPythonicToTs} from '../utils/serialization_utils';\nimport {LayerVariable} from '../variables';\nimport {version} from '../version';\n\nimport {Container, ContainerArgs} from './container';\nimport {Dataset} from './dataset_stub';\nimport {execute, FeedDict} from './executor';\nimport {DisposeResult, SymbolicTensor} from './topology';\nimport {evaluateDataset, fitDataset, ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './training_dataset';\nimport {checkBatchSize, disposeNewTensors, ensureTensorsRank2OrHigher, fitTensors, makeBatches, ModelFitArgs, sliceArrays, sliceArraysByIndices} from './training_tensors';\nimport {ClassWeight, ClassWeightMap, computeWeightedLoss, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Helper function for polymorphic input data: 1. singleton Tensor.\n */\nexport function isDataTensor(x: Tensor|Tensor[]|{[inputName: string]: Tensor}|\n                             {[inputName: string]: Tensor[]}): boolean {\n  return x instanceof Tensor;\n}\n\n/**\n * Helper function for polymorphic input data: 2. Array of Tensor.\n */\nexport function isDataArray(x: Tensor|Tensor[]|\n                            {[inputName: string]: Tensor}): boolean {\n  return Array.isArray(x);\n}\n\n/**\n * Helper function for polymorphic input data: 3. \"dict\" of Tensor.\n */\nexport function isDataDict(x: Tensor|Tensor[]|\n                           {[inputName: string]: Tensor}): boolean {\n  return !isDataTensor(x) && !isDataArray(x);\n}\n\n/**\n * Normalizes inputs and targets provided by users.\n * @param data User-provided input data (polymorphic).\n * @param names An Array of expected Tensor names.\n * @param shapes Optional Array of expected Tensor shapes.\n * @param checkBatchAxis Whether to check that the batch axis of the arrays\n *   match  the expected value found in `shapes`.\n * @param exceptionPrefix String prefix used for exception formatting.\n * @returns List of standardized input Tensors (one Tensor per model input).\n * @throws ValueError: in case of improperly formatted user data.\n */\nexport function standardizeInputData(\n    data: Tensor|Tensor[]|{[inputName: string]: Tensor}, names: string[],\n    shapes?: Shape[], checkBatchAxis = true, exceptionPrefix = ''): Tensor[] {\n  if (names == null || names.length === 0) {\n    // Check for the case where the model expected no data, but some data got\n    // sent.\n    if (data != null) {\n      let gotUnexpectedData = false;\n      if (isDataArray(data) && (data as Tensor[]).length > 0) {\n        gotUnexpectedData = true;\n      } else if (isDataDict(data)) {\n        for (const key in data) {\n          if (data.hasOwnProperty(key)) {\n            gotUnexpectedData = true;\n            break;\n          }\n        }\n      } else {\n        // `data` is a singleton Tensor in this case.\n        gotUnexpectedData = true;\n      }\n      if (gotUnexpectedData) {\n        throw new ValueError(\n            `Error when checking model ${exceptionPrefix} expected no data, ` +\n            `but got ${data}`);\n      }\n    }\n    return [];\n  }\n  if (data == null) {\n    return names.map(name => null);\n  }\n\n  let arrays: Tensor[];\n  if (isDataDict(data)) {\n    data = data as {[inputName: string]: Tensor};\n    arrays = [];\n    for (const name of names) {\n      if (data[name] == null) {\n        throw new ValueError(\n            `No data provided for \"${name}\". Need data for each key in: ` +\n            `${names}`);\n      }\n      arrays.push(data[name]);\n    }\n  } else if (isDataArray(data)) {\n    data = data as Tensor[];\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `model expected. Expected to see ${names.length} Tensor(s), but ` +\n          `instead got the following list of Tensor(s): ${data}`);\n    }\n    arrays = data;\n  } else {\n    data = data as Tensor;\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model ${exceptionPrefix} expects ${names.length} Tensor(s), ` +\n          `but only received one Tensor. Found: Tensor with shape ${\n              data.shape}`);\n    }\n    arrays = [data];\n  }\n\n  arrays = ensureTensorsRank2OrHigher(arrays);\n\n  // Check shape compatibility.\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s). but got array with ` +\n            `shape ${array.shape}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          // Skip the first (batch) axis.\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null && refDim >= 0 && dim !== refDim) {\n          throw new ValueError(\n              `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n              `to have shape [${shapes[i]}], but got array with shape ` +\n              `[${array.shape}].`);\n        }\n      }\n    }\n  }\n  return arrays;\n}\n\n/**\n * User input validation for Tensors.\n * @param inputs `Array` of `tf.Tensor`s for inputs.\n * @param targets `Array` of `tf.Tensor`s for targets.\n * @param weights Optional `Array` of `tf.Tensor`s for sample weights.\n * @throws ValueError: in case of incorrectly formatted data.\n */\nexport function checkArrayLengths(\n    inputs: Tensor[], targets: Tensor[], weights?: Tensor[]) {\n  const setX = unique(inputs.map(input => input.shape[0]));\n  setX.sort();\n  const setY = unique(targets.map(target => target.shape[0]));\n  setY.sort();\n  // TODO(cais): Check `weights` as well.\n  if (setX.length > 1) {\n    throw new ValueError(\n        `All input Tensors (x) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(inputs.map(input => input.shape))}`);\n  }\n  if (setY.length > 1) {\n    throw new ValueError(\n        `All target Tensors (y) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(targets.map(target => target.shape))}`);\n  }\n  if (setX.length > 0 && setY.length > 0 && !util.arraysEqual(setX, setY)) {\n    throw new ValueError(\n        `Input Tensors should have the same number of samples as target ` +\n        `Tensors. Found ${setX[0]} input sample(s) and ${setY[0]} target ` +\n        `sample(s).`);\n  }\n}\n\n/**\n * Validation on the compatibility of targes and loss functions.\n *\n * This helps prevent users from using loss functions incorrectly.\n *\n * @param targets `Array` of `tf.Tensor`s of targets.\n * @param lossFns `Array` of loss functions.\n * @param outputShapes `Array` of shapes of model outputs.\n */\nfunction checkLossAndTargetCompatibility(\n    targets: Tensor[], lossFns: LossOrMetricFn[], outputShapes: Shape[]) {\n  // TODO(cais): Dedicated test coverage?\n  const keyLosses = [\n    losses.meanSquaredError, losses.binaryCrossentropy,\n    losses.categoricalCrossentropy\n  ];\n  for (let i = 0; i < targets.length; ++i) {\n    const y = targets[i];\n    const loss = lossFns[i];\n    const shape = outputShapes[i];\n    if (loss == null) {\n      continue;\n    }\n    if (loss === losses.categoricalCrossentropy) {\n      if (y.shape[y.shape.length - 1] === 1) {\n        throw new ValueError(\n            `You are passing a target array of shape ${y.shape} while using ` +\n            `a loss 'categorical_crossentropy'. 'categorical_crossentropy'` +\n            `expects targets to be binary matrices (1s and 0s) of shape ` +\n            `[samples, classes].`);\n        // TODO(cais): Example code in error message.\n      }\n    }\n    if (keyLosses.indexOf(loss) !== -1) {\n      const slicedYShape = y.shape.slice(1);\n      const slicedShape = shape.slice(1);\n      for (let j = 0; j < slicedYShape.length; ++j) {\n        const targetDim = slicedYShape[j];\n        const outDim = slicedShape[j];\n        if (outDim != null && targetDim !== outDim) {\n          throw new ValueError(\n              `A target Tensor with shape ${y.shape} was passed for an ` +\n              `output of shape ${shape}, while using a loss function that ` +\n              `expects targets to have the same shape as the output.`);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Check inputs provided by the user.\n *\n * Porting Note: This corresponds to _standardize_input_data() in Python\n *   Keras. Because of the strong typing in TF.js, we do not need to convert\n *   the data. Specifically:\n *   1) in PyKeras, `data` can be `DataFrame` instances from pandas, for\n *      example. We don't need to worry about that here because there is no\n *      widely popular javascript/typesdcript equivalent of pandas (so far).\n *      If one becomes available in the future, we can add support.\n *   2) in PyKeras, inputs can be Python dict. But here we are stipulating\n * that the data is either a single `tf.Tensor` or an Array of `tf.Tensor`s. We\n * may add support for `Object` data inputs in the future when the need\n * arises.\n *\n * Instead, we perform basic checks for number of parameters and shapes.\n *\n * @param data: The input data.\n * @param names: Name for the inputs, from the model.\n * @param shapes: Expected shapes for the input data, from the model.\n * @param checkBatchAxis: Whether the size along the batch axis (i.e., the\n *   first dimension) will be checked for matching.\n * @param exceptionPrefix: Execption prefix message, used in generating error\n *   messages.\n * @throws ValueError: on incorrect number of inputs or mismatches in shapes.\n */\nfunction checkInputData(\n    data: Tensor|Tensor[], names: string[], shapes?: Shape[],\n    checkBatchAxis = true, exceptionPrefix = '') {\n  let arrays: Tensor[];\n  if (Array.isArray(data)) {\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `the model expected. Expected to see ${names.length} Tensor(s),` +\n          ` but instead got ${data.length} Tensors(s).`);\n    }\n    arrays = data;\n  } else {\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model expects ${names.length} ${exceptionPrefix} Tensors, ` +\n          `but only received one Tensor. Found: array with shape ` +\n          `${JSON.stringify(data.shape)}.`);\n    }\n    arrays = [data];\n  }\n\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s), but got array with ` +\n            `shape ${JSON.stringify(array.shape)}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null) {\n          if (refDim !== dim) {\n            throw new ValueError(\n                `Error when checking ${exceptionPrefix}: expected ` +\n                `${names[i]} to have shape ${JSON.stringify(shapes[i])} but ` +\n                `got array with shape ${JSON.stringify(array.shape)}.`);\n          }\n        }\n      }\n    }\n  }\n}\n\n/**\n * Maps metric functions to model outputs.\n * @param metrics An shortcut strings name, metric function, `Array` or dict\n *   (`Object`) of metric functions.\n * @param outputNames An `Array` of the names of model outputs.\n * @returns An `Array` (one entry per model output) of `Array` of metric\n *   functions. For instance, if the model has 2 outputs, and for the first\n *   output we want to compute `binaryAccuracy` and `binaryCrossentropy`,\n *   and just `binaryAccuracy` for the second output, the `Array` would look\n *   like:\n *     `[[binaryAccuracy, binaryCrossentropy],  [binaryAccuracy]]`\n * @throws TypeError: incompatible metrics format.\n */\nexport function collectMetrics(\n    metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n    {[outputName: string]: string | LossOrMetricFn},\n    outputNames: string[]): Array<Array<string|LossOrMetricFn>> {\n  if (metrics == null || Array.isArray(metrics) && metrics.length === 0) {\n    return outputNames.map(name => []);\n  }\n\n  let wrappedMetrics: Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  if (typeof metrics === 'string' || typeof metrics === 'function') {\n    wrappedMetrics = [metrics];\n  } else if (Array.isArray(metrics) || typeof metrics === 'object') {\n    wrappedMetrics = metrics as Array<string|LossOrMetricFn>|\n        {[outputName: string]: string} | {[outputName: string]: LossOrMetricFn};\n  } else {\n    throw new TypeError(\n        'Type of metrics argument not understood. Expected an string,' +\n        `function, Array, or Object, found: ${metrics}`);\n  }\n\n  if (Array.isArray(wrappedMetrics)) {\n    // We then apply all metrics to all outputs.\n    return outputNames.map(\n        name => wrappedMetrics as Array<string|LossOrMetricFn>);\n  } else {\n    // In this case, metrics is a dict.\n    const nestedMetrics: Array<Array<string|LossOrMetricFn>> = [];\n    for (const name of outputNames) {\n      let outputMetrics: string|LossOrMetricFn|Array<string|LossOrMetricFn> =\n          wrappedMetrics.hasOwnProperty(name) ? wrappedMetrics[name] : [];\n      if (!Array.isArray(outputMetrics)) {\n        outputMetrics = [outputMetrics];\n      }\n      nestedMetrics.push(outputMetrics);\n    }\n    return nestedMetrics;\n  }\n}\n\nexport interface ModelEvaluateArgs {\n  /**\n   * Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * Tensor of weights to weight the contribution of different samples to the\n   * loss and metrics.\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * integer: total number of steps (batches of samples)\n   * before declaring the evaluation round finished. Ignored with the default\n   * value of `undefined`.\n   */\n  steps?: number;\n}\n\n/**\n * Configuration for calls to `LayersModel.compile()`.\n */\nexport interface ModelCompileArgs {\n  /**\n   * An instance of `tf.train.Optimizer` or a string name for an Optimizer.\n   */\n  optimizer: string|Optimizer;\n\n  /**\n   * Object function(s) or name(s) of object function(s).\n   * If the model has multiple outputs, you can use a different loss\n   * on each output by passing a dictionary or an Array of losses.\n   * The loss value that will be minimized by the model will then be the sum\n   * of all individual losses.\n   */\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n\n  /**\n   * List of metrics to be evaluated by the model during training and testing.\n   * Typically you will use `metrics=['accuracy']`.\n   * To specify different metrics for different outputs of a multi-output\n   * model, you could also pass a dictionary.\n   */\n  metrics?: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n\n  // TODO(cais): Add lossWeights, sampleWeightMode, weightedMetrics, and\n  //   targetTensors.\n}\n\nconst LAYERS_MODEL_FORMAT_NAME = 'layers-model';\n\n/**\n * A `tf.LayersModel` is a directed, acyclic graph of `tf.Layer`s plus methods\n * for training, evaluation, prediction and saving.\n *\n * `tf.LayersModel` is the basic unit of training, inference and evaluation in\n * TensorFlow.js. To create a `tf.LayersModel`, use `tf.LayersModel`.\n *\n * See also:\n *   `tf.Sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class LayersModel extends Container implements tfc.InferenceModel {\n  // The class name is 'Model' rather than 'LayersModel' for backwards\n  // compatibility since this class name shows up in the serialization format.\n  /** @nocollapse */\n  static className = 'Model';\n  protected optimizer_: Optimizer;\n  // Whether the model instance owns the optimizer: `true` if and only if\n  // `optimizer` is created from a string parameter during `compile()` call.\n  protected isOptimizerOwned: boolean;\n\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n  lossFunctions: LossOrMetricFn[];\n\n  // TODO(cais): These private variables should probably not have the string\n  //   'feed' in their names, because we are not dealing with a symbolic\n  //   backend.\n  private feedOutputShapes: Shape[];\n  private feedLossFns: LossOrMetricFn[];\n  private collectedTrainableWeights: LayerVariable[];\n  private testFunction: (data: Tensor[]) => Scalar[];\n  history: History;\n\n  // A public property that can be set by Callbacks to order early stopping\n  // during `fit()` calls.\n  protected stopTraining_: boolean;\n  protected isTraining: boolean;\n\n  metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  metricsNames: string[];\n  // Porting Note: `metrics_tensors` in PyKeras is a symbolic tensor. But given\n  //   the imperative nature of tfjs-core, `metricsTensors` is a\n  //   TypeScript function here.\n  //   Also note that due to the imperative nature of tfjs-core, `metricsTensor`\n  //   here needs an output index to keep track of which output of the\n  //   LayersModel a metric belongs to. This is unlike `metrics_tensors` in\n  //   PyKeras, which is a `list` of symbolic tensors, each of which has\n  //   implicit \"knowledge\" of the outputs it depends on.\n  metricsTensors: Array<[LossOrMetricFn, number]>;\n\n  // User defind metadata (if any).\n  private userDefinedMetadata: {};\n\n  constructor(args: ContainerArgs) {\n    super(args);\n    this.isTraining = false;\n  }\n\n  /**\n   * Print a text summary of the model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - If the model has non-sequential-like topology, the inputs each layer\n   *   receives\n   * - The total number of trainable and non-trainable parameters of the model.\n   *\n   * ```js\n   * const input1 = tf.input({shape: [10]});\n   * const input2 = tf.input({shape: [20]});\n   * const dense1 = tf.layers.dense({units: 4}).apply(input1);\n   * const dense2 = tf.layers.dense({units: 8}).apply(input2);\n   * const concat = tf.layers.concatenate().apply([dense1, dense2]);\n   * const output =\n   *     tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat);\n   *\n   * const model = tf.model({inputs: [input1, input2], outputs: output});\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      throw new ValueError(\n          `This model has never been called, thus its weights have not been ` +\n          `created yet. So no summary can be displayed. Build the model ` +\n          `first (e.g., by calling it on some test data).`);\n    }\n    printSummary(this, lineLength, positions, printFn);\n  }\n\n  /**\n   * Configures and prepares the model for training and evaluation.  Compiling\n   * outfits the model with an optimizer, loss, and/or metrics.  Calling `fit`\n   * or `evaluate` on an un-compiled model will throw an error.\n   *\n   * @param args a `ModelCompileArgs` specifying the loss, optimizer, and\n   * metrics to be used for fitting and evaluating this model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  compile(args: ModelCompileArgs): void {\n    if (args.loss == null) {\n      args.loss = [];\n    }\n    this.loss = args.loss;\n\n    if (typeof args.optimizer === 'string') {\n      this.optimizer_ = optimizers.getOptimizer(args.optimizer);\n      this.isOptimizerOwned = true;\n    } else {\n      if (!(args.optimizer instanceof Optimizer)) {\n        throw new ValueError(\n            `User-defined optimizer must be an instance of tf.Optimizer.`);\n      }\n      this.optimizer_ = args.optimizer;\n      this.isOptimizerOwned = false;\n    }\n\n    // TODO(cais): Add lossWeights.\n    // TODO(cais): Add sampleWeightMode.\n\n    // Prepare loss functions.\n    let lossFunctions: LossOrMetricFn[] = [];\n    if (!Array.isArray(args.loss) && typeof args.loss !== 'string' &&\n        typeof args.loss !== 'function') {\n      args.loss = args.loss as {[outputName: string]: string};\n      for (const name in args.loss) {\n        if (this.outputNames.indexOf(name) === -1) {\n          throw new ValueError(\n              `Unknown entry in loss dictionary: \"${name}\". ` +\n              `Only expected the following keys: ${this.outputNames}`);\n        }\n      }\n      for (const name of this.outputNames) {\n        if (args.loss[name] == null) {\n          console.warn(\n              `Output \"${name}\" is missing from loss dictionary. We assume ` +\n              `this was done on purpose, and we will not be expecting data ` +\n              `to be passed to ${name} during training`);\n        }\n        lossFunctions.push(losses.get(args.loss[name]));\n      }\n    } else if (Array.isArray(args.loss)) {\n      if (args.loss.length !== this.outputs.length) {\n        throw new ValueError(\n            `When passing an Array as loss, it should have one entry per ` +\n            `model output. The model has ${this.outputs.length} output(s), ` +\n            `but you passed loss=${args.loss}.`);\n      }\n      const theLosses = args.loss as Array<string|LossOrMetricFn>;\n      lossFunctions = theLosses.map(l => losses.get(l));\n    } else {\n      const lossFunction = losses.get(args.loss);\n      this.outputs.forEach(_ => {\n        lossFunctions.push(lossFunction);\n      });\n    }\n\n    this.lossFunctions = lossFunctions;\n\n    this.feedOutputNames = [];\n    this.feedOutputShapes = [];\n    this.feedLossFns = [];\n    for (let i = 0; i < this.outputs.length; ++i) {\n      // TODO(cais): Logic for skipping target(s).\n      const shape = this.internalOutputShapes[i];\n      const name = this.outputNames[i];\n      this.feedOutputNames.push(name);\n      this.feedOutputShapes.push(shape);\n      this.feedLossFns.push(this.lossFunctions[i]);\n    }\n\n    // TODO(cais): Add logic for output masks.\n    // TODO(cais): Add logic for sample weights.\n    const skipTargetIndices: number[] = [];\n\n    // Prepare metrics.\n    this.metrics = args.metrics;\n    // TODO(cais): Add weightedMetrics.\n    this.metricsNames = ['loss'];\n    this.metricsTensors = [];\n\n    // Compute total loss.\n    // Porting Note: In PyKeras, metrics_tensors are symbolic tensor objects.\n    //   Here, metricsTensors are TypeScript functions. This difference is due\n    //   to the difference in symbolic/imperative property of the backends.\n    nameScope('loss', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        // TODO(cais): Add weightedLoss, sampleWeight and mask.\n        //   The following line should be weightedLoss\n        const weightedLoss = this.lossFunctions[i];\n        if (this.outputs.length > 1) {\n          this.metricsTensors.push([weightedLoss, i]);\n          this.metricsNames.push(this.outputNames[i] + '_loss');\n        }\n      }\n\n      // Porting Note: Due to the imperative nature of the backend, we calculate\n      //   the regularizer penalties in the totalLossFunction, instead of here.\n    });\n\n    const nestedMetrics = collectMetrics(args.metrics, this.outputNames);\n    // TODO(cais): Add nestedWeightedMetrics.\n\n    /**\n     * Helper function used in loop below.\n     */\n    const appendMetric =\n        (outputIndex: number, metricName: string,\n         metricTensor: LossOrMetricFn) => {\n          if (this.outputNames.length > 1) {\n            metricName = this.outputNames[outputIndex] + '_' + metricName;\n          }\n          this.metricsNames.push(metricName);\n          this.metricsTensors.push([metricTensor, outputIndex]);\n        };\n\n    nameScope('metric', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        const outputMetrics = nestedMetrics[i];\n        // TODO(cais): Add weights and outputWeightedMetrics.\n\n        // TODO(cais): Add optional arg `weights` to the following function.\n        const handleMetrics = (metrics: Array<string|LossOrMetricFn>) => {\n          const metricNamePrefix = '';\n          let metricName: string;\n          let accFn: LossOrMetricFn;\n          let weightedMetricFn: LossOrMetricFn;\n          //  TODO(cais): Use 'weights_' for weighted metrics.\n\n          for (const metric of metrics) {\n            if (typeof metric === 'string' &&\n                ['accuracy', 'acc', 'crossentropy', 'ce'].indexOf(metric) !==\n                    -1) {\n              const outputShape = this.internalOutputShapes[i];\n\n              if (outputShape[outputShape.length - 1] === 1 ||\n                  this.lossFunctions[i] === losses.binaryCrossentropy) {\n                // case: binary accuracy/crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryCrossentropy;\n                }\n              } else if (\n                  this.lossFunctions[i] ===\n                  losses.sparseCategoricalCrossentropy) {\n                // case: categorical accuracy / crossentropy with sparse\n                // targets.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalCrossentropy;\n                }\n              } else {\n                // case: categorical accuracy / crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalCrossentropy;\n                }\n              }\n              let suffix: string;\n              if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                suffix = 'acc';\n              } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                suffix = 'ce';\n              }\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = accFn;\n              metricName = metricNamePrefix + suffix;\n            } else {\n              const metricFn = Metrics.get(metric);\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = metricFn;\n              metricName =\n                  metricNamePrefix + Metrics.getLossOrMetricName(metric);\n            }\n\n            // TODO(cais): Add weighting and masking to metricResult.\n            let metricResult: LossOrMetricFn;\n            nameScope(metricName, () => {\n              metricResult = weightedMetricFn;\n            });\n            appendMetric(i, metricName, metricResult);\n          }\n        };\n\n        handleMetrics(outputMetrics);\n        // TODO(cais): Call handleMetrics with weights.\n      }\n    });\n\n    // Porting Notes: Given the imperative backend of tfjs-core,\n    //   there is no need for constructing the symbolic graph and placeholders.\n    this.collectedTrainableWeights = this.trainableWeights;\n  }\n\n  /**\n   * Check trainable weights count consistency.\n   *\n   * This will raise a warning if `this.trainableWeights` and\n   * `this.collectedTrainableWeights` are inconsistent (i.e., have different\n   * numbers of parameters).\n   * Inconsistency will typically arise when one modifies `model.trainable`\n   * without calling `model.compile()` again.\n   */\n  protected checkTrainableWeightsConsistency(): void {\n    if (this.collectedTrainableWeights == null) {\n      return;\n    }\n    if (this.trainableWeights.length !==\n        this.collectedTrainableWeights.length) {\n      console.warn(\n          'Discrepancy between trainableweights and collected trainable ' +\n          'weights. Did you set `model.trainable` without calling ' +\n          '`model.compile()` afterwards?');\n    }\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(\n   *     tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4});\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateArgs`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // TODO(cais): Standardize `config.sampleWeights` as well.\n    // Validate user data.\n    const checkBatchAxis = true;\n    const standardizedOuts =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    try {\n      // TODO(cais): If uses `useLearningPhase`, set the corresponding element\n      // of the input to 0.\n      const ins = standardizedOuts[0].concat(standardizedOuts[1]);\n      this.makeTestFunction();\n      const f = this.testFunction;\n      const testOuts =\n          this.testLoop(f, ins, batchSize, args.verbose, args.steps);\n      return singletonOrArray(testOuts);\n    } finally {\n      disposeNewTensors(standardizedOuts[0], x);\n      disposeNewTensors(standardizedOuts[1], y);\n    }\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args?: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    this.makeTestFunction();\n    return evaluateDataset(this, dataset, args);\n  }\n\n  /**\n   * Get number of samples provided for training, evaluation or prediction.\n   *\n   * @param ins Input `tf.Tensor`.\n   * @param batchSize Integer batch size, optional.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring loop finished. Optional.\n   * @param stepsName The public API's parameter name for `steps`.\n   * @returns Number of samples provided.\n   */\n  private checkNumSamples(\n      ins: Tensor|Tensor[], batchSize?: number, steps?: number,\n      stepsName = 'steps'): number {\n    let numSamples: number;\n    if (steps != null) {\n      numSamples = null;\n      if (batchSize != null) {\n        throw new ValueError(\n            `If ${stepsName} is set, batchSize must be null or undefined.` +\n            `Got batchSize = ${batchSize}`);\n      }\n    } else if (ins != null) {\n      if (Array.isArray(ins)) {\n        numSamples = ins[0].shape[0];\n      } else {\n        numSamples = ins.shape[0];\n      }\n    } else {\n      throw new ValueError(\n          `Either the input data should have a defined shape, or ` +\n          `${stepsName} shoud be specified.`);\n    }\n    return numSamples;\n  }\n\n  /**\n   * Execute internal tensors of the model with input data feed.\n   * @param inputs Input data feed. Must match the inputs of the model.\n   * @param outputs Names of the output tensors to be fetched. Must match\n   *   names of the SymbolicTensors that belong to the graph.\n   * @returns Fetched values for `outputs`.\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs: string|string[]):\n      Tensor|Tensor[] {\n    if (Array.isArray(outputs) && outputs.length === 0) {\n      throw new ValueError(\n          '`outputs` is an empty Array, which is not allowed.');\n    }\n\n    const outputsIsArray = Array.isArray(outputs);\n    const outputNames =\n        (outputsIsArray ? outputs as string[] : [outputs as string]);\n    const outputSymbolicTensors = this.retrieveSymbolicTensors(outputNames);\n\n    // Format the input into a FeedDict.\n    const feedDict = new FeedDict();\n    if (inputs instanceof Tensor) {\n      inputs = [inputs];\n    }\n    if (Array.isArray(inputs)) {\n      if (inputs.length !== this.inputs.length) {\n        throw new ValueError(\n            `The number of inputs provided (${inputs.length}) ` +\n            `does not match the number of inputs of this model ` +\n            `(${this.inputs.length}).`);\n      }\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n    } else {\n      for (const input of this.inputs) {\n        const tensorValue = inputs[input.name];\n        if (tensorValue == null) {\n          throw new ValueError(\n              `No value is provided for the model's input ${input.name}`);\n        }\n        feedDict.add(input, tensorValue);\n      }\n    }\n\n    // Run execution.\n    const executeOutputs = execute(outputSymbolicTensors, feedDict) as Tensor[];\n    return outputsIsArray ? executeOutputs : executeOutputs[0];\n  }\n\n  /**\n   * Retrieve the model's internal symbolic tensors from symbolic-tensor names.\n   */\n  private retrieveSymbolicTensors(symbolicTensorNames: string[]):\n      SymbolicTensor[] {\n    const outputSymbolicTensors: SymbolicTensor[] =\n        pyListRepeat(null, symbolicTensorNames.length);\n    let outputsRemaining = symbolicTensorNames.length;\n    for (const layer of this.layers) {\n      const layerOutputs: SymbolicTensor[] =\n          Array.isArray(layer.output) ? layer.output : [layer.output];\n      const layerOutputNames = layerOutputs.map(output => output.name);\n      for (let i = 0; i < symbolicTensorNames.length; ++i) {\n        const index = layerOutputNames.indexOf(symbolicTensorNames[i]);\n        if (index !== -1) {\n          outputSymbolicTensors[i] = layerOutputs[index];\n          outputsRemaining--;\n        }\n        if (outputsRemaining === 0) {\n          break;\n        }\n      }\n      if (outputsRemaining === 0) {\n        break;\n      }\n    }\n\n    if (outputsRemaining > 0) {\n      const remainingNames: string[] = [];\n      outputSymbolicTensors.forEach((tensor, i) => {\n        if (tensor == null) {\n          remainingNames.push(symbolicTensorNames[i]);\n        }\n      });\n      throw new ValueError(\n          `Cannot find SymbolicTensors for output name(s): ` +\n          `${JSON.stringify(remainingNames)}`);\n    }\n    return outputSymbolicTensors;\n  }\n\n  /**\n   * Helper method to loop over some data in batches.\n   *\n   * Porting Note: Not using the functional approach in the Python equivalent\n   *   due to the imperative backend.\n   * Porting Note: Does not support step mode currently.\n   *\n   * @param ins: input data\n   * @param batchSize: integer batch size.\n   * @param verbose: verbosity model\n   * @returns: Predictions as `tf.Tensor` (if a single output) or an `Array` of\n   *   `tf.Tensor` (if multipe outputs).\n   */\n  private predictLoop(ins: Tensor|Tensor[], batchSize = 32, verbose = false):\n      Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins);\n      if (verbose) {\n        throw new NotImplementedError(\n            'Verbose predictLoop() is not implemented yet.');\n      }\n\n      // Sample-based predictions.\n      // Porting Note: Tensor currently does not support sliced assignments as\n      //   in numpy, e.g., x[1:3] = y. Therefore we use concatenation while\n      //   iterating over the batches.\n\n      const batches = makeBatches(numSamples, batchSize);\n      const outsBatches: Tensor[][] = this.outputs.map(output => []);\n\n      // TODO(cais): Can the scope() be pushed down inside the for loop?\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchOuts = tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          // TODO(cais): Take care of the case of the last element is a flag for\n          //   training/test.\n          const insBatch = sliceArrays(ins, batchStart, batchEnd);\n\n          // Construct the feeds for execute();\n          const feeds = [];\n          if (Array.isArray(insBatch)) {\n            for (let i = 0; i < insBatch.length; ++i) {\n              feeds.push({key: this.inputs[i], value: insBatch[i]});\n            }\n          } else {\n            feeds.push({key: this.inputs[0], value: insBatch});\n          }\n          const feedDict = new FeedDict(feeds);\n          return execute(this.outputs, feedDict) as Tensor[];\n        });\n        batchOuts.forEach((batchOut, i) => outsBatches[i].push(batchOut));\n      }\n      return singletonOrArray(\n          outsBatches.map(batches => tfc.concat(batches, 0)));\n    });\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFlow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([8, 10]), {batchSize: 4}).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param args A `ModelPredictArgs` object containing optional fields.\n   *\n   * @return Prediction results as a `tf.Tensor`(s).\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    const xsRank2OrHigher = ensureTensorsRank2OrHigher(x);\n    checkInputData(\n        xsRank2OrHigher, this.inputNames, this.feedInputShapes, false);\n    try {\n      // TODO(cais): Take care of stateful models.\n      //   if (this.stateful) ...\n      // TODO(cais): Take care of the learning_phase boolean flag.\n      //   if (this.useLearningPhase) ...\n      const batchSize = args.batchSize == null ? 32 : args.batchSize;\n      checkBatchSize(batchSize);\n      return this.predictLoop(xsRank2OrHigher, batchSize);\n    } finally {\n      disposeNewTensors(xsRank2OrHigher, x);\n    }\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predictOnBatch(tf.ones([8, 10])).print();\n   * ```\n   * @param x: Input samples, as a Tensor (for models with exactly one\n   *   input) or an array of Tensors (for models with more than one input).\n   * @return Tensor(s) of predictions\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predictOnBatch(x: Tensor|Tensor[]): Tensor|Tensor[] {\n    checkInputData(x, this.inputNames, this.feedInputShapes, true);\n    // TODO(cais): Take care of the learning_phase boolean flag.\n    //   if (this.useLearningPhase) ...\n    const batchSize = (Array.isArray(x) ? x[0] : x).shape[0];\n    return this.predictLoop(x, batchSize);\n  }\n\n  protected standardizeUserDataXY(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor}, checkBatchAxis = true,\n      batchSize?: number): [Tensor[], Tensor[]] {\n    // TODO(cais): Add sampleWeight, classWeight\n    if (this.optimizer_ == null) {\n      throw new RuntimeError(\n          'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileArgs).');\n    }\n    const outputShapes: Shape[] = [];\n    for (let i = 0; i < this.feedOutputShapes.length; ++i) {\n      const outputShape = this.feedOutputShapes[i];\n      const lossFn = this.feedLossFns[i];\n      if (lossFn === losses.sparseCategoricalCrossentropy) {\n        outputShapes.push(\n            outputShape.slice(0, outputShape.length - 1).concat([1]));\n      } else {\n        // Porting Note: Because of strong typing `lossFn` must be a function.\n        outputShapes.push(outputShape);\n      }\n    }\n    x = standardizeInputData(\n        x, this.feedInputNames, this.feedInputShapes, false, 'input');\n    y = standardizeInputData(\n        y, this.feedOutputNames, outputShapes, false, 'target');\n    // TODO(cais): Standardize sampleWeights & classWeights.\n    checkArrayLengths(x, y, null);\n    // TODO(cais): Check sampleWeights as well.\n    checkLossAndTargetCompatibility(y, this.feedLossFns, this.feedOutputShapes);\n    if (this.stateful && batchSize != null && batchSize > 0) {\n      if (x[0].shape[0] % batchSize !== 0) {\n        throw new ValueError(\n            `In a stateful network, you should only pass inputs with a ` +\n            `number of samples that is divisible by the batch size ` +\n            `${batchSize}. Found: ${x[0].shape[0]} sample(s).`);\n      }\n    }\n    return [x, y];\n  }\n\n  protected async standardizeUserData(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      sampleWeight?: Tensor|Tensor[]|{[outputName: string]: Tensor},\n      classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap,\n      checkBatchAxis = true,\n      batchSize?: number): Promise<[Tensor[], Tensor[], Tensor[]]> {\n    const [standardXs, standardYs] =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    // TODO(cais): Handle sampleWeights.\n    if (sampleWeight != null) {\n      throw new Error('sample weight is not supported yet.');\n    }\n\n    let standardSampleWeights: Tensor[] = null;\n    if (classWeight != null) {\n      const classWeights =\n          standardizeClassWeights(classWeight, this.outputNames);\n      standardSampleWeights = [];\n      for (let i = 0; i < classWeights.length; ++i) {\n        standardSampleWeights.push(\n            await standardizeWeights(standardYs[i], null, classWeights[i]));\n      }\n    }\n\n    // TODO(cais): Deal with the case of model.stateful == true.\n    return [standardXs, standardYs, standardSampleWeights];\n  }\n\n  /**\n   * Loop over some test data in batches.\n   * @param f A Function returning a list of tensors.\n   * @param ins Array of tensors to be fed to `f`.\n   * @param batchSize Integer batch size or `null` / `undefined`.\n   * @param verbose verbosity mode.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring test finished. Ignored with the default value of `null` /\n   * `undefined`.\n   * @returns Array of Scalars.\n   */\n  private testLoop(\n      f: (data: Tensor[]) => Scalar[], ins: Tensor[], batchSize?: number,\n      verbose = 0, steps?: number): Scalar[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins, batchSize, steps, 'steps');\n      const outs: Scalar[] = [];\n      if (verbose > 0) {\n        throw new NotImplementedError('Verbose mode is not implemented yet.');\n      }\n      // TODO(cais): Use `indicesForConversionToDense' to prevent slow down.\n      if (steps != null) {\n        throw new NotImplementedError(\n            'steps mode in testLoop() is not implemented yet');\n      } else {\n        const batches = makeBatches(numSamples, batchSize);\n        const indexArray = tensor1d(range(0, numSamples));\n        for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds =\n              K.sliceAlongFirstAxis(\n                  indexArray, batchStart, batchEnd - batchStart) as Tensor1D;\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Scalar[];\n          const batchOuts = f(insBatch);\n          if (batchIndex === 0) {\n            for (let i = 0; i < batchOuts.length; ++i) {\n              outs.push(scalar(0));\n            }\n          }\n          for (let i = 0; i < batchOuts.length; ++i) {\n            const batchOut = batchOuts[i];\n            outs[i] =\n                tfc.add(outs[i], tfc.mul(batchEnd - batchStart, batchOut));\n          }\n        }\n        for (let i = 0; i < outs.length; ++i) {\n          outs[i] = tfc.div(outs[i], numSamples);\n        }\n      }\n      return outs;\n    });\n  }\n\n  protected getDedupedMetricsNames(): string[] {\n    const outLabels = this.metricsNames;\n    // Rename duplicated metrics names (can happen with an output layer\n    // shared among multiple dataflows).\n    const dedupedOutLabels = [];\n    for (let i = 0; i < outLabels.length; ++i) {\n      const label = outLabels[i];\n      let newLabel = label;\n      if (count(outLabels, label) > 1) {\n        const dupIndex = count(outLabels.slice(0, i), label);\n        newLabel += `_${dupIndex}`;\n      }\n      dedupedOutLabels.push(newLabel);\n    }\n    return dedupedOutLabels;\n  }\n\n  /**\n   * Creates a function that performs the following actions:\n   *\n   * 1. computes the losses\n   * 2. sums them to get the total loss\n   * 3. call the optimizer computes the gradients of the LayersModel's\n   *    trainable weights w.r.t. the total loss and update the variables\n   * 4. calculates the metrics\n   * 5. returns the values of the losses and metrics.\n   */\n  protected makeTrainFunction(): (data: Tensor[]) => Scalar[] {\n    return (data: Tensor[]) => {\n      const lossValues: Scalar[] = [];\n\n      const inputs = data.slice(0, this.inputs.length);\n      const targets = data.slice(\n          this.inputs.length, this.inputs.length + this.outputs.length);\n      const sampleWeights = data.slice(\n          this.inputs.length + this.outputs.length,\n          this.inputs.length + this.outputs.length * 2);\n\n      const metricsValues: Scalar[] = [];\n\n      // Create a function that computes the total loss based on the\n      // inputs. This function is used for obtaining gradients through\n      // backprop.\n      const totalLossFunction = () => {\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs =\n            execute(this.outputs, feedDict, {'training': true}) as Tensor[];\n        // TODO(cais): Take care of the case of multiple outputs from a\n        //   single layer?\n\n        let totalLoss: Tensor;\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          let loss = lossFunction(targets[i], outputs[i]);\n          if (sampleWeights[i] != null) {\n            loss = computeWeightedLoss(loss, sampleWeights[i]);\n          }\n\n          // TODO(cais): push Scalar instead.\n          const meanLoss: Scalar = tfc.mean(loss);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          lossValues.push(meanLoss);\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n        }\n\n        // Compute the metrics.\n        // TODO(cais): These should probably be calculated outside\n        //   totalLossFunction to benefit speed?\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          let weightedMetric: Scalar;\n\n          if (this.outputs.length > 1 && i < this.outputs.length) {\n            weightedMetric = lossValues[i];\n          } else {\n            const metric = this.metricsTensors[i][0];\n            const outputIndex = this.metricsTensors[i][1];\n            weightedMetric =\n                tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          }\n\n          tfc.keep(weightedMetric);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          metricsValues.push(weightedMetric);\n        }\n\n        totalLoss = tfc.mean(totalLoss);\n\n        // Add regularizer penalties.\n        this.calculateLosses().forEach(regularizerLoss => {\n          totalLoss = tfc.add(totalLoss, regularizerLoss);\n        });\n\n        return totalLoss as Scalar;\n      };\n\n      const variables = this.collectedTrainableWeights.map(\n          param => param.read() as tfc.Variable);\n      const returnCost = true;\n      const totalLossValue =\n          this.optimizer_.minimize(totalLossFunction, returnCost, variables);\n\n      return [totalLossValue].concat(metricsValues);\n    };\n  }\n\n  /**\n   * Create a function which, when invoked with an array of `tf.Tensor`s as a\n   * batch of inputs, returns the prespecified loss and metrics of the model\n   * under the batch of input data.\n   */\n  private makeTestFunction() {\n    this.testFunction = (data: Tensor[]) => {\n      return tfc.tidy(() => {\n        const valOutputs: Scalar[] = [];\n        let totalLoss: Scalar;\n        const inputs = data.slice(0, this.inputs.length);\n        const targets = data.slice(\n            this.inputs.length, this.inputs.length + this.outputs.length);\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs = execute(this.outputs, feedDict) as Tensor[];\n        // Compute total loss.\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          // TODO(cais): Add sample weighting and replace the simple\n          // averaging.\n          const loss: Scalar = tfc.mean(lossFunction(targets[i], outputs[i]));\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n          valOutputs.push(totalLoss);\n        }\n        // Compute the metrics.\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          const metric = this.metricsTensors[i][0];\n          const outputIndex = this.metricsTensors[i][1];\n          // TODO(cais): Replace K.mean() with a proper weighting function.\n          const meanMetric =\n              tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          valOutputs.push(meanMetric as Scalar);\n        }\n        return valOutputs;\n      });\n    };\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a\n   * dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *     layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * for (let i = 1; i < 5 ; ++i) {\n   *   const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *       batchSize: 4,\n   *       epochs: 3\n   *   });\n   *   console.log(\"Loss after Epoch \" + i + \" : \" + h.history.loss[0]);\n   * }\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you\n   * can also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named,\n   * you can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args A `ModelFitArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input\n   * data and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    return fitTensors(this, x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below when it's possible to instantiate\n  //   actual dataset objects.\n  /**\n   * Trains the model using a dataset object.\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for training. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs.\n   *   Of the two items in the array, the first is the input feature(s) and\n   *   the second is the output target(s).\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    return fitDataset(this, dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    // TODO(cais): Support sampleWeight and classWeight.\n    // TODO(cais): Support Dataset objects.\n    const standardizeOut = await this.standardizeUserData(x, y);\n    const inputs = standardizeOut[0];\n    const targets = standardizeOut[1];\n    const trainFunction = this.makeTrainFunction();\n    const losses = trainFunction(inputs.concat(targets));\n    const lossValues: number[] = [];\n    for (const loss of losses) {\n      const v = await loss.data();\n      lossValues.push(v[0]);\n    }\n    tfc.dispose(losses);\n    return singletonOrArray(lossValues);\n  }\n\n  /**\n   * Extract weight values of the model.\n   *\n   * @param config: An instance of `io.SaveConfig`, which specifies\n   * model-saving options such as whether only trainable weights are to be\n   * saved.\n   * @returns A `NamedTensorMap` mapping original weight names (i.e.,\n   *   non-uniqueified weight names) to their values.\n   */\n  protected getNamedWeights(config?: io.SaveConfig): NamedTensor[] {\n    const namedWeights: NamedTensor[] = [];\n\n    const trainableOnly = config != null && config.trainableOnly;\n    const weights = trainableOnly ? this.trainableWeights : this.weights;\n    const weightValues = this.getWeights(trainableOnly);\n    for (let i = 0; i < weights.length; ++i) {\n      if (trainableOnly && !weights[i].trainable) {\n        // Optionally skip non-trainable weights.\n        continue;\n      }\n      namedWeights.push(\n          {name: weights[i].originalName, tensor: weightValues[i]});\n    }\n    return namedWeights;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const input = tf.input({shape: [10]});\n   * const output = tf.layers.dense({units: 1}).apply(input);\n   * const model = tf.model({inputs: [input], outputs: [output]});\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10\n   * values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    this.stopTraining_ = stop;\n  }\n\n  get stopTraining(): boolean {\n    return this.stopTraining_;\n  }\n\n  get optimizer(): Optimizer {\n    return this.optimizer_;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    if (this.optimizer_ !== optimizer) {\n      this.optimizer_ = optimizer;\n      this.isOptimizerOwned = false;\n    }\n  }\n\n  dispose(): DisposeResult {\n    const result = super.dispose();\n    if (result.refCountAfterDispose === 0 && this.optimizer != null &&\n        this.isOptimizerOwned) {\n      const numTensorsBeforeOptmizerDisposal = tfc.memory().numTensors;\n      this.optimizer_.dispose();\n      result.numDisposedVariables +=\n          numTensorsBeforeOptmizerDisposal - tfc.memory().numTensors;\n    }\n    return result;\n  }\n\n  private getLossIdentifiers(): LossIdentifier|LossIdentifier[]|\n      {[outputName: string]: LossIdentifier} {\n    let lossNames: LossIdentifier|LossIdentifier[]|\n        {[outputName: string]: LossIdentifier};\n    if (typeof this.loss === 'string') {\n      lossNames = toSnakeCase(this.loss) as LossIdentifier;\n    } else if (Array.isArray(this.loss)) {\n      for (const loss of this.loss) {\n        if (typeof loss !== 'string') {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n      lossNames = (this.loss as string[]).map(name => toSnakeCase(name)) as\n          LossIdentifier[];\n    } else {\n      const outputNames = Object.keys(this.loss);\n      lossNames = {} as {[outputName: string]: LossIdentifier};\n      const losses =\n          this.loss as {[outputName: string]: LossOrMetricFn | string};\n      for (const outputName of outputNames) {\n        if (typeof losses[outputName] === 'string') {\n          lossNames[outputName] =\n              toSnakeCase(losses[outputName] as string) as LossIdentifier;\n        } else {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n    }\n    return lossNames;\n  }\n\n  private getMetricIdentifiers(): MetricsIdentifier[]|\n      {[key: string]: MetricsIdentifier} {\n    if (typeof this.metrics === 'string' ||\n        typeof this.metrics === 'function') {\n      return [toSnakeCase(Metrics.getLossOrMetricName(this.metrics))];\n    } else if (Array.isArray(this.metrics)) {\n      return this.metrics.map(\n          metric => toSnakeCase(Metrics.getLossOrMetricName(metric)));\n    } else {\n      const metricsIdentifiers: {[key: string]: MetricsIdentifier} = {};\n      for (const key in this.metrics) {\n        metricsIdentifiers[key] =\n            toSnakeCase(Metrics.getLossOrMetricName(this.metrics[key]));\n      }\n      return metricsIdentifiers;\n    }\n  }\n\n  protected getTrainingConfig(): TrainingConfig {\n    return {\n      loss: this.getLossIdentifiers(),\n      metrics: this.getMetricIdentifiers(),\n      optimizer_config: {\n        class_name: this.optimizer.getClassName(),\n        config: this.optimizer.getConfig()\n      } as OptimizerSerialization\n    };\n    // TODO(cais): Add weight_metrics when they are supported.\n    // TODO(cais): Add sample_weight_mode when it's supported.\n    // TODO(cais): Add loss_weights when it's supported.\n  }\n\n  loadTrainingConfig(trainingConfig: TrainingConfig) {\n    if (trainingConfig.weighted_metrics != null) {\n      throw new Error('Loading weight_metrics is not supported yet.');\n    }\n    if (trainingConfig.loss_weights != null) {\n      throw new Error('Loading loss_weights is not supported yet.');\n    }\n    if (trainingConfig.sample_weight_mode != null) {\n      throw new Error('Loading sample_weight_mode is not supported yet.');\n    }\n\n    const tsConfig = convertPythonicToTs(trainingConfig.optimizer_config) as\n        serialization.ConfigDict;\n    const optimizer = deserialize(tsConfig) as Optimizer;\n\n    let loss;\n    if (typeof trainingConfig.loss === 'string') {\n      loss = toCamelCase(trainingConfig.loss);\n    } else if (Array.isArray(trainingConfig.loss)) {\n      loss = trainingConfig.loss.map(lossEntry => toCamelCase(lossEntry));\n    } else if (trainingConfig.loss != null) {\n      loss = {} as {[outputName: string]: LossIdentifier};\n      for (const key in trainingConfig.loss) {\n        loss[key] = toCamelCase(trainingConfig.loss[key]) as LossIdentifier;\n      }\n    }\n\n    let metrics;\n    if (Array.isArray(trainingConfig.metrics)) {\n      metrics = trainingConfig.metrics.map(metric => toCamelCase(metric));\n    } else if (trainingConfig.metrics != null) {\n      metrics = {} as {[outputName: string]: MetricsIdentifier};\n      for (const key in trainingConfig.metrics) {\n        metrics[key] = toCamelCase(trainingConfig.metrics[key]);\n      }\n    }\n\n    this.compile({loss, metrics, optimizer});\n  }\n\n  /**\n   * Save the configuration and/or weights of the LayersModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 2. Saving `model`'s topology and weights to browser\n   * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('indexeddb://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 3. Saving `model`'s topology and weights as two files\n   * (`my-model-1.json` and `my-model-1.weights.bin`) downloaded from\n   * browser.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('downloads://my-model-1');\n   * ```\n   *\n   * Example 4. Send  `model`'s topology and weights to an HTTP server.\n   * See the documentation of `tf.io.http` for more details\n   * including specifying request parameters and implementation of the\n   * server.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('http://my-server/model/upload');\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new ValueError(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new ValueError(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new ValueError(\n          'LayersModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    const weightDataAndSpecs =\n        await io.encodeWeights(this.getNamedWeights(config));\n\n    const returnString = false;\n    const unusedArg: {} = null;\n    const modelConfig = this.toJSON(unusedArg, returnString);\n    const modelArtifacts: io.ModelArtifacts = {\n      modelTopology: modelConfig,\n      format: LAYERS_MODEL_FORMAT_NAME,\n      generatedBy: `TensorFlow.js tfjs-layers v${version}`,\n      convertedBy: null,\n    };\n\n    const includeOptimizer = config == null ? false : config.includeOptimizer;\n    if (includeOptimizer && this.optimizer != null) {\n      modelArtifacts.trainingConfig = this.getTrainingConfig();\n      const weightType = 'optimizer';\n      const {data: optimizerWeightData, specs: optimizerWeightSpecs} =\n          await io.encodeWeights(await this.optimizer.getWeights(), weightType);\n      weightDataAndSpecs.specs.push(...optimizerWeightSpecs);\n      weightDataAndSpecs.data = io.concatenateArrayBuffers(\n          [weightDataAndSpecs.data, optimizerWeightData]);\n    }\n\n    if (this.userDefinedMetadata != null) {\n      // Check serialized size of user-defined metadata.\n      const checkSize = true;\n      checkUserDefinedMetadata(this.userDefinedMetadata, this.name, checkSize);\n      modelArtifacts.userDefinedMetadata = this.userDefinedMetadata;\n    }\n\n    modelArtifacts.weightData = weightDataAndSpecs.data;\n    modelArtifacts.weightSpecs = weightDataAndSpecs.specs;\n    return handlerOrURL.save(modelArtifacts);\n  }\n\n  /**\n   * Set user-defined metadata.\n   *\n   * The set metadata will be serialized together with the topology\n   * and weights of the model during `save()` calls.\n   *\n   * @param setUserDefinedMetadata\n   */\n  setUserDefinedMetadata(userDefinedMetadata: {}): void {\n    checkUserDefinedMetadata(userDefinedMetadata, this.name);\n    this.userDefinedMetadata = userDefinedMetadata;\n  }\n\n  /**\n   * Get user-defined metadata.\n   *\n   * The metadata is supplied via one of the two routes:\n   *   1. By calling `setUserDefinedMetadata()`.\n   *   2. Loaded during model loading (if the model is constructed\n   *      via `tf.loadLayersModel()`.)\n   *\n   * If no user-defined metadata is available from either of the\n   * two routes, this function will return `undefined`.\n   */\n  getUserDefinedMetadata(): {} {\n    return this.userDefinedMetadata;\n  }\n}\nserialization.registerClass(LayersModel);\n\n/**\n * A `tf.Functional` is an alias to `tf.LayersModel`.\n *\n * See also:\n *   `tf.LayersModel`, `tf.Sequential`, `tf.loadLayersModel`.\n */\n/** @doc {heading: 'Models', subheading: 'Classes'} */\nexport class Functional extends LayersModel {\n  static className = 'Functional';\n}\nserialization.registerClass(Functional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source keras/models.py */\n\nimport {dispose, io, NamedTensorMap, Optimizer, Scalar, serialization, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {getUid} from './backend/state';\nimport {History} from './base_callbacks';\nimport {Dataset} from './engine/dataset_stub';\nimport {Input} from './engine/input_layer';\nimport {getSourceInputs, Layer, Node, SymbolicTensor} from './engine/topology';\nimport {LayersModel, ModelCompileArgs, ModelEvaluateArgs} from './engine/training';\nimport {ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './engine/training_dataset';\nimport {ModelFitArgs} from './engine/training_tensors';\nimport {NotImplementedError, RuntimeError, ValueError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {TrainingConfig} from './keras_format/training_config';\nimport {PyJsonDict} from './keras_format/types';\nimport {deserialize} from './layers/serialization';\nimport {Kwargs, NamedTensor} from './types';\nimport * as generic_utils from './utils/generic_utils';\nimport {convertPythonicToTs} from './utils/serialization_utils';\nimport {getExactlyOneShape} from './utils/types_utils';\n\n/**\n * Parses a JSON model configuration file and returns a model instance.\n *\n * ```js\n * // This example shows how to serialize a model using `toJSON()` and\n * // deserialize it as another model using `tf.models.modelFromJSON()`.\n * // Note: this example serializes and deserializes only the topology\n * // of the model; the weights of the loaded model will be different\n * // from those of the the original model, due to random weight\n * // initialization.\n * // To load the topology and weights of a model, use `tf.loadLayersModel()`.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.repeatVector({inputShape: [2], n: 4}));\n * // Serialize `model1` as a JSON object.\n * const model1JSON = model1.toJSON(null, false);\n * model1.summary();\n *\n * const model2 = await tf.models.modelFromJSON(model1JSON);\n * model2.summary();\n * ```\n *\n *  @param modelAndWeightsConfig JSON object or string encoding a model and\n *       weights configuration. It can also be only the topology JSON of the\n *       model, in which case the weights will not be loaded.\n *  @param custom_objects Optional dictionary mapping names\n *       (strings) to custom classes or functions to be\n *       considered during deserialization.\n * @returns A TensorFlow.js Layers `tf.LayersModel` instance (uncompiled).\n */\nexport async function modelFromJSON(\n    modelAndWeightsConfig: ModelAndWeightsConfig|PyJsonDict,\n    customObjects?: serialization.ConfigDict): Promise<LayersModel> {\n  if (!('modelTopology' in modelAndWeightsConfig)) {\n    modelAndWeightsConfig = {modelTopology: modelAndWeightsConfig};\n  }\n  modelAndWeightsConfig = modelAndWeightsConfig as ModelAndWeightsConfig;\n\n  let modelTopology = modelAndWeightsConfig.modelTopology;\n  if (modelTopology['model_config'] != null) {\n    // If the model-topology JSON contains a 'model_config' field, then it is\n    // a full model JSON (e.g., from `keras.Model.save()`), which contains\n    // not only the model's architecture in its 'model_config' field, but\n    // additional information such as the model's optimizer. We use only the\n    // 'model_config' field currently.\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n  const tsConfig =\n      convertPythonicToTs(modelTopology) as serialization.ConfigDict;\n  const model = deserialize(tsConfig, customObjects) as LayersModel;\n\n  if (modelAndWeightsConfig.weightsManifest != null) {\n    // Load the weight values keyed by the original tensor names in the model\n    // file that was loaded.  These should match the keys of the weight\n    // manifest.\n    const weightValues = await io.loadWeights(\n        modelAndWeightsConfig.weightsManifest, modelAndWeightsConfig.pathPrefix,\n        model.weights.map(weight => weight.originalName));\n\n    // Map the weights to the unique tensor names generated during model loading\n    const uniqueWeightValues: NamedTensorMap = {};\n    for (const weight of model.weights) {\n      uniqueWeightValues[weight.originalName] =\n          weightValues[weight.originalName];\n    }\n\n    model.loadWeights(uniqueWeightValues);\n    // Dispose temporary weight values.\n    dispose(weightValues);\n  }\n  return model;\n}\n\n/**\n * Options for loading a saved mode in TensorFlow.js format.\n */\nexport interface ModelAndWeightsConfig {\n  /**\n   * A JSON object or JSON string containing the model config.\n   *\n   * This can be either of the following two formats:\n   *   - A model archiecture-only config,  i.e., a format consistent with the\n   *     return value of`keras.Model.to_json()`.\n   *   - A full model config, containing not only model architecture, but also\n   *     training options and state, i.e., a format consistent with the return\n   *     value of `keras.models.save_model()`.\n   */\n  modelTopology: PyJsonDict;\n\n  /**\n   * A weights manifest in TensorFlow.js format.\n   */\n  weightsManifest?: io.WeightsManifestConfig;\n\n  /**\n   * Path to prepend to the paths in `weightManifest` before fetching.\n   *\n   * The path may optionally end in a slash ('/').\n   */\n  pathPrefix?: string;\n}\n\n// TODO(nielsene): Remove after: https://github.com/tensorflow/tfjs/issues/400\nexport interface ModelPredictArgs {\n  /**\n   * Optional. Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Optional. Verbosity mode. Defaults to false.\n   */\n  verbose?: boolean;\n}\n\n/**\n * Load a model, including its topology and optionally weights.  See the\n * Tutorial named \"How to import a Keras Model\" for usage examples.\n *\n * Example 1: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 2. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * Example 4. Load a model from an HTTP server.\n *\n * ```js\n * const model = await\n *     tf.loadLayersModel('https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. This path will be\n *      interpreted as a relative HTTP path, to which `fetch` will be used to\n *      request the model topology and weight manifest JSON.\n *      The content of the JSON file is assumed to be a JSON object with the\n *      following fields and values:\n *      - 'modelTopology': A JSON object that can be either of:\n *        1. a model architecture JSON consistent with the format of the return\n *            value of `keras.Model.to_json()`\n *        2. a full model JSON in the format of `keras.models.save_model()`.\n *      - 'weightsManifest': A TensorFlow.js weights manifest.\n *      See the Python converter function `save_model()` for more details.\n *      It is also assumed that model weights can be accessed from relative\n *      paths described by the `paths` fields in weights manifest.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A progress callback of the form:\n *     `(fraction: number) => void`. This callback can be used to monitor the\n *     model-loading process.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n */\nexport async function loadLayersModelInternal(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (typeof pathOrIOHandler === 'string') {\n    const handlers = io.getLoadHandlers(pathOrIOHandler, options);\n    if (handlers.length === 0) {\n      // For backward compatibility: if no load handler can be found,\n      // assume it is a relative http path.\n      // TODO(cais): Reformat the args into a single `LoadOptions` once the core\n      // is refactored.\n      handlers.push(io.browserHTTPRequest(pathOrIOHandler, options));\n    } else if (handlers.length > 1) {\n      throw new ValueError(\n          `Found more than one (${handlers.length}) load handlers for ` +\n          `URL '${pathOrIOHandler}'`);\n    }\n    pathOrIOHandler = handlers[0];\n  }\n  return loadLayersModelFromIOHandler(pathOrIOHandler, undefined, options);\n}\n\n/**\n * Load a model and optionally its weights, using an IOHandler object.\n *\n * @param handler The instance of `IOHandler` to be used during the model\n *   loading.\n * @param customObjects Any optional custom objects to be used during model\n *   loading.\n * @param strict Whether the weight loading will be done in strict mode.\n *   Default: `true`.\n */\nexport async function loadLayersModelFromIOHandler(\n    handler: io.IOHandler, customObjects?: serialization.ConfigDict,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (handler.load == null) {\n    throw new ValueError(\n        'Cannot proceed with model loading because the IOHandler provided ' +\n        'does not have the `load` method implemented.');\n  }\n  const artifacts = await handler.load();\n  let modelTopology = artifacts.modelTopology as PyJsonDict;\n  if (modelTopology['model_config'] != null) {\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n\n  const strict = options.strict == null ? true : options.strict;\n  // If weights are provided and the weight-loading mode is strict, use\n  // fast weight initialization. This skips costly initializers such as\n  // 'orthogonal' and saves unnecessary computation in cases where\n  // the initialized weight values will immediately be overwritten by\n  // loaded weight values.\n  const fastWeightInit =\n      artifacts.weightData != null && artifacts.weightSpecs != null && strict;\n  const model =\n      deserialize(\n          convertPythonicToTs(modelTopology) as serialization.ConfigDict,\n          customObjects, fastWeightInit) as LayersModel;\n\n  const trainingConfig = artifacts.trainingConfig as TrainingConfig;\n  if (trainingConfig != null) {\n    model.loadTrainingConfig(trainingConfig);\n  }\n  if (artifacts.userDefinedMetadata != null) {\n    model.setUserDefinedMetadata(artifacts.userDefinedMetadata);\n  }\n\n  // If weightData is present, load the weights into the model.\n  if (artifacts.weightData != null) {\n    // Loading weights requires weightSpecs.\n    if (artifacts.weightSpecs == null) {\n      throw new ValueError(\n          'LayersModel artifacts contains weight data, but not weight specs. ' +\n          'Therefore loading of weights cannot proceed.');\n    }\n\n    const {modelWeights, optimizerWeights} = decodeModelAndOptimizerWeights(\n        artifacts.weightData, artifacts.weightSpecs);\n    model.loadWeights(modelWeights, strict);\n\n    if (model.optimizer != null && optimizerWeights.length > 0) {\n      await model.optimizer.setWeights(optimizerWeights);\n    }\n\n    // Dispose temporary weight values.\n    dispose(modelWeights);\n    dispose(optimizerWeights.map(w => w.tensor));\n  }\n  return model;\n}\n\nfunction decodeModelAndOptimizerWeights(\n    buffer: ArrayBuffer, specs: io.WeightsManifestEntry[]):\n    {modelWeights: NamedTensorMap, optimizerWeights: NamedTensor[]} {\n  const name2Tensor = io.decodeWeights(buffer, specs);\n  const modelWeights: NamedTensorMap = {};\n  const optimizerWeights: NamedTensor[] = [];\n  specs.forEach(spec => {\n    if (spec.group === 'optimizer') {\n      optimizerWeights.push({name: spec.name, tensor: name2Tensor[spec.name]});\n    } else {\n      modelWeights[spec.name] = name2Tensor[spec.name];\n    }\n  });\n  return {modelWeights, optimizerWeights};\n}\n\n/**\n * Configuration for a Sequential model.\n */\nexport interface SequentialArgs {\n  /** Stack of layers for the model. */\n  layers?: Layer[];\n\n  /** The name of this model. */\n  name?: string;\n}\n\n/**\n * A model with a stack of layers, feeding linearly from one to the next.\n *\n * `tf.sequential` is a factory function that creates an instance of\n * `tf.Sequential`.\n *\n * ```js\n *  // Define a model for linear regression.\n *  const model = tf.sequential();\n *  model.add(tf.layers.dense({units: 1, inputShape: [1]}));\n *\n *  // Prepare the model for training: Specify the loss and the optimizer.\n *  model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n *\n *  // Generate some synthetic data for training.\n *  const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);\n *  const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);\n *\n *  // Train the model using the data then do inference on a data point the\n *  // model hasn't seen:\n *  await model.fit(xs, ys);\n *  model.predict(tf.tensor2d([5], [1, 1])).print();\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class Sequential extends LayersModel {\n  /** @nocollapse */\n  static className = 'Sequential';\n  private model: LayersModel;\n  constructor(args?: SequentialArgs) {\n    super({inputs: [], outputs: []});\n    args = args || {};\n\n    this.trainable = true;\n    this.built = false;\n\n    // Set model name.\n    this.name = (args.name != null) ? args.name : getUid('sequential_');\n\n    // Add to the model any layers passed to the constructor.\n    if (args.layers != null) {\n      for (const layer of args.layers) {\n        this.add(layer);\n      }\n    }\n  }\n\n  // Helper function to Sequential.add  Throws if the new output shape will be\n  // invalid.\n  private checkShape(layer: Layer) {\n    const shape = layer.inboundNodes[0].outputTensors[0].shape;\n    if (shape.some(x => x < 0)) {\n      throw new ValueError(\n          'Negative dimension size caused by adding layer ' +\n          `${layer.name} with input shape [` +\n          `${layer.inboundNodes[0].inputTensors[0].shape}]`);\n    }\n  }\n\n  /**\n   * Adds a layer instance on top of the layer stack.\n   *\n   * ```js\n   *  const model = tf.sequential();\n   *  model.add(tf.layers.dense({units: 8, inputShape: [1]}));\n   *  model.add(tf.layers.dense({units: 4, activation: 'relu6'}));\n   *  model.add(tf.layers.dense({units: 1, activation: 'relu6'}));\n   *  // Note that the untrained model is random at this point.\n   *  model.predict(tf.randomNormal([10, 1])).print();\n   * ```\n   * @param layer Layer instance.\n   *\n   * @exception ValueError In case the `layer` argument does not know its\n   * input shape.\n   * @exception ValueError In case the `layer` argument has multiple output\n   *   tensors, or is already connected somewhere else (forbidden in\n   *   `Sequential` models).\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  add(layer: Layer): void {\n    const isLayerModelInstance =\n        layer instanceof Sequential || layer instanceof LayersModel;\n    let modelLayer: LayersModel;\n    if (isLayerModelInstance) {\n      modelLayer = layer as LayersModel;\n      if (modelLayer.outputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      if (modelLayer.inputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single input tensor. ' +\n            'For multi-input layers, ' +\n            'use the functional API.');\n      }\n    }\n\n    if (this.outputs.length === 0) {\n      // first layer in model: check that it is an input layer\n      if (layer.inboundNodes.length === 0) {\n        // create an input layer\n        if (layer.batchInputShape == null) {\n          throw new ValueError(\n              'The first layer in a Sequential model must ' +\n              'get an `inputShape` or `batchInputShape` argument.');\n        }\n        // Instantiate the input layer.\n        const x = Input({\n          batchShape: layer.batchInputShape,\n          dtype: layer.dtype,\n          name: layer.name + '_input'\n        });\n        // This will build the current layer and create the node connecting\n        // the current layer to the input layer we just created.\n        layer.apply(x);\n      }\n\n      if (isLayerModelInstance) {\n        this.outputs = modelLayer.outputs;\n        this.inputs = modelLayer.inputs;\n      } else {\n        if (layer.inboundNodes.length !== 1) {\n          throw new ValueError(\n              'A layer added to a Sequential model must not already be ' +\n              `connected somewhere else. LayersModel received layer ${\n                  layer.name} ` +\n              `which has ${layer.inboundNodes.length} pre-existing inbound ` +\n              'connections.');\n        }\n\n        if (layer.inboundNodes[0].outputTensors.length !== 1) {\n          throw new ValueError(\n              'All layers in a Sequential model ' +\n              'should have a single output tensor. ' +\n              'For multi-output layers, ' +\n              'use the functional API.');\n        }\n        this.checkShape(layer);\n        this.outputs = [layer.inboundNodes[0].outputTensors[0]];\n        this.inputs = getSourceInputs(this.outputs[0]);\n      }\n\n      this.inboundNodes = [];\n      // We create an input node, which we will keep updated\n      // as we add more layers.\n      // (This call has side effects.)\n      // tslint:disable-next-line:no-unused-expression\n      new Node({\n        outboundLayer: this,\n        inboundLayers: [],\n        nodeIndices: [],\n        tensorIndices: [],\n        inputTensors: this.inputs,\n        outputTensors: this.outputs,\n        // no model-level masking for now\n        inputMasks: generic_utils.pyListRepeat(null, this.inputs.length),\n        outputMasks: [null],\n        inputShapes: this.inputs.map(x => x.shape),\n        outputShapes: this.outputs[0].shape\n      });\n    } else {\n      const outputTensor = layer.apply(this.outputs[0]);\n      if (Array.isArray(outputTensor)) {\n        throw new TypeError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      this.checkShape(layer);\n      this.outputs = [outputTensor as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n\n    this.layers.push(layer);\n    this.built = false;\n  }\n\n  /**\n   * Removes the last layer in the model.\n   *\n   * @exception TypeError if there are no layers in the model.\n   */\n  pop(): void {\n    if (this.layers.length === 0) {\n      throw new TypeError('There are no layers in the model.');\n    }\n\n    this.layers.pop();\n    if (this.layers.length === 0) {\n      this.outputs = [];\n      this.inboundNodes = [];\n      this.outboundNodes = [];\n    } else {\n      const lastLayerIndex = this.layers.length - 1;\n      this.layers[lastLayerIndex].outboundNodes = [];\n      this.outputs = [this.layers[lastLayerIndex].output as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.call(inputs, kwargs);\n  }\n\n  build(inputShape?: Shape|Shape[]) {\n    // Call `getExactlyOneShape` without using its return value,\n    // to verify that exactly one input shape is provided.\n    getExactlyOneShape(inputShape);\n\n    if (this.inputs.length === 0 || this.outputs.length === 0) {\n      throw new TypeError(\n          'Sequential model cannot be built: model is empty.' +\n          ' Add some layers first.');\n    }\n    // actually create the model\n    this.model = new LayersModel({\n      inputs: this.inputs,\n      outputs: this.outputs[0],\n      name: this.name + '_model'\n    });\n    this.model.trainable = this.trainable;\n\n    // mirror model attributes\n    this.supportsMasking = this.model.supportsMasking;\n    // TODO(michaelterry): Add caches\n    this.inputLayers = this.model.inputLayers;\n    this.inputLayersNodeIndices = this.model.inputLayersNodeIndices;\n    this.inputLayersTensorIndices = this.model.inputLayersTensorIndices;\n    this.outputLayers = this.model.outputLayers;\n    this.outputLayersNodeIndices = this.model.outputLayersNodeIndices;\n    this.outputLayersTensorIndices = this.model.outputLayersTensorIndices;\n    this.nodesByDepth = this.model.nodesByDepth;\n    this.containerNodes = this.model.containerNodes;\n    this.outputNames = this.model.outputNames;\n    this.inputNames = this.model.inputNames;\n    // TODO(michaelterry): Add feedInputNames, feedInputs, if needed.\n    // TODO(michaelterry): Add callbackModel if needed.\n    this.built = true;\n  }\n\n  countParams(): number {\n    if (!this.built) {\n      this.build();\n    }\n    return super.countParams();\n  }\n\n  /**\n   * Print a text summary of the Sequential model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - The total number of trainable and non-trainable parameters of the\n   * model.\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(\n   *     tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'}));\n   * model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n   *\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      this.build();\n    }\n    super.summary(lineLength, positions, printFn);\n  }\n\n  /**\n   * Sets the weights of the model.\n   *\n   * @param weights Should be a list of Tensors with shapes and types matching\n   *   the output of `model.getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    if (this.model == null) {\n      this.build();\n    }\n    this.model.setWeights(weights);\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   * });\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateConfig`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluate(x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluateDataset(dataset, args);\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([2, 10])).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param conifg A `ModelPredictConfig` object containing optional fields.\n   *\n   * @return `tf.Tensor`(s) of predictions.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predict(x, args);\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * @param x: Input samples, as a Tensor, or list of Tensors (if the model\n   *   has multiple inputs).\n   * @return Tensor(s) of predictions\n   */\n  predictOnBatch(x: Tensor): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predictOnBatch(x);\n  }\n\n  /**\n   * See `LayersModel.compile`.\n   *\n   * @param args\n   */\n  compile(args: ModelCompileArgs): void {\n    this.build();\n    this.model.compile(args);\n    this.optimizer_ = this.model.optimizer;\n    // tslint:disable-next-line:no-any\n    this.isOptimizerOwned = (this.model as any).isOptimizerOwned;\n    this.loss = this.model.loss;\n    this.metrics = this.model.metrics;\n    // TODO(cais): Add this.lossWeights, this.sampleWeightMode,\n    //   this.weightedMetrics, this.targets.\n    this.metricsTensors = this.model.metricsTensors;\n    this.metricsNames = this.model.metricsNames;\n    // TODO(cais): Add sampleWeights.\n  }\n\n  get optimizer(): Optimizer {\n    return this.model == null ? undefined : this.model.optimizer;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    this.model.optimizer = optimizer;\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   *   epochs: 3\n   * });\n   * console.log(history.history.loss[0]);\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you can\n   * also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named, you\n   *  can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args  A `ModelFitConfig`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fit(x, y, args);\n  }\n\n  /**\n   * Trains the model using a dataset object.\n   *\n   * ```js\n   * const xArray = [\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   * ];\n   * const yArray = [1, 1, 1, 1];\n   * // Create a dataset from the JavaScript array.\n   * const xDataset = tf.data.array(xArray);\n   * const yDataset = tf.data.array(yArray);\n   * // Zip combines the `x` and `y` Datasets into a single Dataset, the\n   * // iterator of which will return an object containing of two tensors,\n   * // corresponding to `x` and `y`.  The call to `batch(4)` will bundle\n   * // four such samples into a single object, with the same keys now pointing\n   * // to tensors that hold 4 examples, organized along the batch dimension.\n   * // The call to `shuffle(4)` causes each iteration through the dataset to\n   * // happen in a different order.  The size of the shuffle window is 4.\n   * const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset})\n   *     .batch(4)\n   *     .shuffle(4);\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [9]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fitDataset(xyDataset, {\n   *   epochs: 4,\n   *   callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)}\n   * });\n   * ```\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected to\n   *   generate a dataset iterator object, the `next()` method of which is\n   *   expected to produce data batches for evaluation. The return value of the\n   *   `next()` call ought to contain a boolean `done` field and a `value`\n   *   field.\n   *\n   *   The `value` field is expected to be an object of with fields\n   *   `xs` and `ys`, which point to the feature tensor and the target tensor,\n   *   respectively. This case is for models with exactly one input and one\n   *   output (e.g.. a sequential model). For example:\n   *   ```js\n   *   {value: {xs: xsTensor, ys: ysTensor}, done: false}\n   *   ```\n   *\n   *   If the model has multiple inputs, the `xs` field of `value` should\n   *   be an object mapping input names to their respective feature tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: {\n   *         input_1: xsTensor1,\n   *         input_2: xsTensor2\n   *       },\n   *       ys: ysTensor\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   *   If the model has multiple outputs, the `ys` field of `value` should\n   *   be an object mapping output names to their respective target tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: xsTensor,\n   *       ys: {\n   *         output_1: ysTensor1,\n   *         output_2: ysTensor2\n   *       },\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fitDataset(dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    return this.model.trainOnBatch(x, y);\n  }\n\n  /* See parent class for JsDoc */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    let configArray: serialization.ConfigDictArray;\n    let extraModelConfig: serialization.ConfigDict = {};\n    if (config instanceof Array) {\n      if (!(config[0].className != null) ||\n          config[0]['className'] === 'Merge') {\n        throw new ValueError('Legacy serialization format not supported yet.');\n      }\n      configArray = config;\n    } else {\n      util.assert(\n          config['layers'] != null,\n          () =>\n              `When the config data for a Sequential model is not an Array, ` +\n              `it must be an Object that contains the 'layers' field.`);\n      configArray = config['layers'] as serialization.ConfigDictArray;\n      delete config['layers'];\n      extraModelConfig = config;\n    }\n\n    const model = new cls(extraModelConfig);\n    if (!(model instanceof Sequential)) {\n      throw new NotImplementedError(\n          `Sequential.fromConfig called on non-Sequential input: ${model}`);\n    }\n    for (const conf of configArray) {\n      const customObjects: serialization.ConfigDict = undefined;\n      const layer = deserialize(\n                        conf as serialization.ConfigDict, customObjects,\n                        fastWeightInit) as Layer;\n      if (fastWeightInit) {\n        layer.setFastWeightInitDuringBuild(true);\n      }\n      model.add(layer);\n    }\n    return model;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(tf.layers.dense({units: 1, inputShape: [10]}));\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10 values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    // TODO(cais): When refactoring to remove the composition pattern happens,\n    // remove this method overriding.\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot set the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    this.model.stopTraining = stop;\n  }\n\n  get stopTraining(): boolean {\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot get the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    return this.model.stopTraining;\n  }\n\n  // TODO(cais): Override get trainableWeights() here\n\n  // tslint:disable-next-line:no-any\n  getConfig(): any {\n    // NOTE(cais): We override the return type of getConfig() to `any` here,\n    //   because the `Sequential` class is a special case among `Container`\n    //   subtypes in that its getConfig() method returns an Array (not a\n    //   dict).\n    const layers: serialization.ConfigDict[] = [];\n    for (const layer of this.layers) {\n      const dict: serialization.ConfigDict = {};\n      dict['className'] = layer.getClassName();\n      dict['config'] = layer.getConfig();\n      layers.push(dict);\n    }\n    return {name: this.name, layers};\n  }\n}\nserialization.registerClass(Sequential);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Exported functions.\n */\n\nimport {io} from '@tensorflow/tfjs-core';\n\nimport {BaseCallbackConstructor, CallbackConstructorRegistry} from './base_callbacks';\nimport {ContainerArgs} from './engine/container';\nimport {Input, InputConfig,} from './engine/input_layer';\nimport {SymbolicTensor} from './engine/topology';\nimport {LayersModel} from './engine/training';\nimport {loadLayersModelInternal, Sequential, SequentialArgs} from './models';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// LayersModel and related factory methods.\n\n/**\n * A model is a data structure that consists of `Layers` and defines inputs\n * and outputs.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.model` is more generic, supporting an arbitrary graph (without\n * cycles) of layers. `tf.sequential` is less generic and supports only a linear\n * stack of layers.\n *\n * When creating a `tf.LayersModel`, specify its input(s) and output(s). Layers\n * are used to wire input(s) to output(s).\n *\n * For example, the following code snippet defines a model consisting of\n * two `dense` layers, with 10 and 4 units, respectively.\n *\n * ```js\n * // Define input, which has a size of 5 (not including batch dimension).\n * const input = tf.input({shape: [5]});\n *\n * // First dense layer uses relu activation.\n * const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});\n * // Second dense layer uses softmax activation.\n * const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});\n *\n * // Obtain the output symbolic tensor by applying the layers on the input.\n * const output = denseLayer2.apply(denseLayer1.apply(input));\n *\n * // Create the model based on the inputs.\n * const model = tf.model({inputs: input, outputs: output});\n *\n * // The model can be used for training, evaluation and prediction.\n * // For example, the following line runs prediction with the model on\n * // some fake data.\n * model.predict(tf.ones([2, 5])).print();\n * ```\n * See also:\n *   `tf.sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function model(args: ContainerArgs): LayersModel {\n  return new LayersModel(args);\n}\n\n/**\n * Creates a `tf.Sequential` model.  A sequential model is any model where the\n * outputs of one layer are the inputs to the next layer, i.e. the model\n * topology is a simple 'stack' of layers, with no branching or skipping.\n *\n * This means that the first layer passed to a `tf.Sequential` model should have\n * a defined input shape. What that means is that it should have received an\n * `inputShape` or `batchInputShape` argument, or for some type of layers\n * (recurrent, Dense...) an `inputDim` argument.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.sequential` is less generic, supporting only a linear stack of layers.\n * `tf.model` is more generic and supports an arbitrary graph (without\n * cycles) of layers.\n *\n * Examples:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have an input shape defined.\n * model.add(tf.layers.dense({units: 32, inputShape: [50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output, which equals\n * // `[null, 4]`. The 1st dimension is the undetermined batch dimension; the\n * // 2nd is the output size of the model's last layer.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * It is also possible to specify a batch size (with potentially undetermined\n * batch dimension, denoted by \"null\") for the first layer using the\n * `batchInputShape` key. The following example is equivalent to the above:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have a defined input shape\n * model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * You can also use an `Array` of already-constructed `Layer`s to create\n * a `tf.Sequential` model:\n *\n * ```js\n * const model = tf.sequential({\n *   layers: [tf.layers.dense({units: 32, inputShape: [50]}),\n *            tf.layers.dense({units: 4})]\n * });\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function sequential(config?: SequentialArgs): Sequential {\n  return new Sequential(config);\n}\n\n/**\n * Load a model composed of Layer objects, including its topology and optionally\n * weights. See the Tutorial named \"How to import a Keras Model\" for usage\n * examples.\n *\n * This method is applicable to:\n *\n * 1. Models created with the `tf.layers.*`, `tf.sequential`, and\n * `tf.model` APIs of TensorFlow.js and later saved with the\n * `tf.LayersModel.save` method.\n * 2. Models converted from Keras or TensorFlow tf.keras using the\n * [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).\n *\n * This mode is *not* applicable to TensorFlow `SavedModel`s or their converted\n * forms. For those models, use `tf.loadGraphModel`.\n *\n * Example 1. Load a model from an HTTP server.\n *\n * ```js\n * const model = await tf.loadLayersModel(\n *     'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * Example 2: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 4. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. For file://\n *      (tfjs-node-only), http:// and https:// schemas, the path can be\n *      either absolute or relative.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A function of the signature `(fraction: number) => void',\n *     that can be used as the progress callback for the model loading.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport function loadLayersModel(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  return loadLayersModelInternal(pathOrIOHandler, options);\n}\n\n/**\n * Used to instantiate an input to a model as a `tf.SymbolicTensor`.\n *\n * Users should call the `input` factory function for\n * consistency with other generator functions.\n *\n * Example:\n *\n * ```js\n * // Defines a simple logistic regression model with 32 dimensional input\n * // and 3 dimensional output.\n * const x = tf.input({shape: [32]});\n * const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x);\n * const model = tf.model({inputs: x, outputs: y});\n * model.predict(tf.ones([2, 32])).print();\n * ```\n *\n * Note: `input` is only necessary when using `model`. When using\n * `sequential`, specify `inputShape` for the first layer or use `inputLayer`\n * as the first layer.\n *\n * @doc {heading: 'Models', subheading: 'Inputs'}\n */\nexport function input(config: InputConfig): SymbolicTensor {\n  return Input(config);\n}\n\nexport function registerCallbackConstructor(\n    verbosityLevel: number,\n    callbackConstructor: BaseCallbackConstructor): void {\n  CallbackConstructorRegistry.registerCallbackConstructor(\n      verbosityLevel, callbackConstructor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Layer activation functions\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {ActivationIdentifier} from './keras_format/activation_config';\nimport {deserializeKerasObject} from './utils/generic_utils';\n\n/**\n * Base class for Activations.\n *\n * Special note: due to cross-language compatibility reasons, the\n * static readonly className field in this family of classes must be set to\n * the initialLowerCamelCase name of the activation.\n */\nexport abstract class Activation extends serialization.Serializable {\n  abstract apply(tensor: Tensor, axis?: number): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\n/**\n * Exponential linear unit (ELU).\n * Reference: https://arxiv.org/abs/1511.07289\n */\nexport class Elu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'elu';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x: Input.\n   * @param alpha: Scaling factor the negative section.\n   * @return Output of the ELU activation.\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return K.elu(x, alpha);\n  }\n}\nserialization.registerClass(Elu);\n\n/**\n * Scaled Exponential Linear Unit. (Klambauer et al., 2017).\n * Reference: Self-Normalizing Neural Networks, https://arxiv.org/abs/1706.02515\n * Notes:\n *   - To be used together with the initialization \"lecunNormal\".\n *   - To be used together with the dropout variant \"AlphaDropout\".\n */\nexport class Selu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'selu';\n  apply(x: Tensor): Tensor {\n    return tfc.selu(x);\n  }\n}\nserialization.registerClass(Selu);\n\n/**\n *  Rectified linear unit\n */\nexport class Relu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu';\n  apply(x: Tensor): Tensor {\n    return tfc.relu(x);\n  }\n}\nserialization.registerClass(Relu);\n\n/**\n * Rectified linear unit activation maxing out at 6.0.\n */\nexport class Relu6 extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu6';\n  apply(x: Tensor): Tensor {\n    return tidy(() => tfc.minimum(6.0, tfc.relu(x)));\n  }\n}\nserialization.registerClass(Relu6);\n\n//* Linear activation (no-op) */\nexport class Linear extends Activation {\n  /** @nocollapse */\n  static readonly className = 'linear';\n  apply(x: Tensor): Tensor {\n    return x;\n  }\n}\nserialization.registerClass(Linear);\n\n/**\n * Sigmoid activation function.\n */\nexport class Sigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'sigmoid';\n  apply(x: Tensor): Tensor {\n    return tfc.sigmoid(x);\n  }\n}\nserialization.registerClass(Sigmoid);\n\n/**\n * Segment-wise linear approximation of sigmoid.\n */\nexport class HardSigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'hardSigmoid';\n  apply(x: Tensor): Tensor {\n    return K.hardSigmoid(x);\n  }\n}\nserialization.registerClass(HardSigmoid);\n\n/**\n * Softplus activation function.\n */\nexport class Softplus extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softplus';\n  apply(x: Tensor): Tensor {\n    return tfc.softplus(x);\n  }\n}\nserialization.registerClass(Softplus);\n\n/**\n * Softsign activation function.\n */\nexport class Softsign extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softsign';\n  apply(x: Tensor): Tensor {\n    return K.softsign(x);\n  }\n}\nserialization.registerClass(Softsign);\n\n/**\n * Hyperbolic tangent function.\n */\nexport class Tanh extends Activation {\n  /** @nocollapse */\n  static readonly className = 'tanh';\n  apply(x: Tensor): Tensor {\n    return tfc.tanh(x);\n  }\n}\nserialization.registerClass(Tanh);\n\n/**\n * Softmax activation function\n */\nexport class Softmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softmax';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.softmax(x, axis);\n  }\n}\nserialization.registerClass(Softmax);\n\n/**\n * Log softmax activation function\n */\nexport class LogSoftmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'logSoftmax';\n  /**\n   * Calculate the activation function of log softmax:\n   * log( exp(x_i) / sum(exp(x)) )\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.logSoftmax(x, axis);\n  }\n}\nserialization.registerClass(LogSoftmax);\n\n/**\n * Swish activation function\n */\nexport class Swish extends Activation {\n  /** @nocollapse */\n  static readonly className = 'swish';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param alpha Scaling factor for the sigmoid function.\n   * @returns a Tensor of the same shape as x\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return tidy(() => tfc.sigmoid(x.mul(alpha)).mul(x));\n  }\n}\nserialization.registerClass(Swish);\n\nexport function serializeActivation(activation: Activation): string {\n  return activation.getClassName();\n}\n\nexport function deserializeActivation(\n   config: serialization.ConfigDict,\n   customObjects: serialization.ConfigDict = {}): Activation {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'activation');\n}\n\nexport function getActivation(identifier: ActivationIdentifier|\n                              serialization.ConfigDict|Activation): Activation {\n  if (identifier == null) {\n    const config: serialization.ConfigDict = {};\n    config['className'] = 'linear';\n    config['config'] = {};\n    return deserializeActivation(config);\n  }\n  if (typeof identifier === 'string') {\n    const config: serialization.ConfigDict = {};\n    config['className'] = identifier;\n    config['config'] = {};\n    return deserializeActivation(config);\n  } else if (identifier instanceof Activation) {\n    return identifier;\n  } else {\n    return deserializeActivation(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* original source: keras/regularizers.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {abs, add, Scalar, serialization, sum, Tensor, tidy, zeros} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\nfunction assertObjectArgs(args: L1Args | L2Args | L1L2Args): void {\n  if (args != null && typeof args !== 'object') {\n    throw new Error(\n        `Argument to L1L2 regularizer's constructor is expected to be an ` +\n        `object, but received: ${args}`);\n  }\n}\n\n/**\n * Regularizer base class.\n */\nexport abstract class Regularizer extends serialization.Serializable {\n  abstract apply(x: Tensor): Scalar;\n}\n\nexport interface L1L2Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1?: number;\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2?: number;\n}\n\nexport interface L1Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1: number;\n}\n\nexport interface L2Args {\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2: number;\n}\n\nexport class L1L2 extends Regularizer {\n  /** @nocollapse */\n  static className = 'L1L2';\n\n  private readonly l1: number;\n  private readonly l2: number;\n  private readonly hasL1: boolean;\n  private readonly hasL2: boolean;\n  constructor(args?: L1L2Args) {\n    super();\n\n    assertObjectArgs(args);\n\n    this.l1 = args == null || args.l1 == null ? 0.01 : args.l1;\n    this.l2 = args == null || args.l2 == null ? 0.01 : args.l2;\n    this.hasL1 = this.l1 !== 0;\n    this.hasL2 = this.l2 !== 0;\n  }\n\n  /**\n   * Porting note: Renamed from __call__.\n   * @param x Variable of which to calculate the regularization score.\n   */\n  apply(x: Tensor): Scalar {\n    return tidy(() => {\n      let regularization: Tensor = zeros([1]);\n      if (this.hasL1) {\n        regularization = add(regularization, sum(tfc.mul(this.l1, abs(x))));\n      }\n      if (this.hasL2) {\n        regularization =\n            add(regularization, sum(tfc.mul(this.l2, K.square(x))));\n      }\n      return regularization.asScalar();\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {'l1': this.l1, 'l2': this.l2};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls({l1: config['l1'] as number, l2: config['l2'] as number});\n  }\n}\nserialization.registerClass(L1L2);\n\nexport function l1(args?: L1Args) {\n  assertObjectArgs(args);\n  return new L1L2({l1: args != null ? args.l1 : null, l2: 0});\n}\n\nexport function l2(args: L2Args) {\n  assertObjectArgs(args);\n  return new L1L2({l2: args != null ? args.l2 : null, l1: 0});\n}\n\n/** @docinline */\nexport type RegularizerIdentifier = 'l1l2'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding keras symbols.\nexport const REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in RegularizerIdentifier]: string} = {\n      'l1l2': 'L1L2'\n    };\n\nexport function serializeRegularizer(constraint: Regularizer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeRegularizer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Regularizer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'regularizer');\n}\n\nexport function getRegularizer(identifier: RegularizerIdentifier|\n                               serialization.ConfigDict|\n                               Regularizer): Regularizer {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeRegularizer(config);\n  } else if (identifier instanceof Regularizer) {\n    return identifier;\n  } else {\n    return deserializeRegularizer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n *  Advanced activation layers.\n */\n\nimport {clipByValue, elu, leakyRelu, prelu, relu, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {Softmax as softmaxActivation} from '../activations';\nimport {cast} from '../backend/tfjs_backend';\nimport {Constraint, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface ReLULayerArgs extends LayerArgs {\n  /**\n   * Float, the maximum output value.\n   */\n  maxValue?: number;\n}\n\nexport class ReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ReLU';\n  maxValue: number;\n\n  constructor(args?: ReLULayerArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maxValue = args.maxValue;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    let output = relu(inputs);\n    if (this.maxValue != null) {\n      output = clipByValue(output, 0, this.maxValue);\n    }\n    return output;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {maxValue: this.maxValue};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ReLU);\n\nexport declare interface LeakyReLULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `0.3`.\n   */\n  alpha?: number;\n}\n\nexport class LeakyReLU extends Layer {\n  /** @nocollapse */\n  static className = 'LeakyReLU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 0.3;\n\n  constructor(args?: LeakyReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return leakyRelu(x, this.alpha);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LeakyReLU);\n\nexport declare interface PReLULayerArgs extends LayerArgs {\n  /**\n   * Initializer for the learnable alpha.\n   */\n  alphaInitializer?: Initializer|InitializerIdentifier;\n\n  /**\n   * Regularizer for the learnable alpha.\n   */\n  alphaRegularizer?: Regularizer;\n\n  /**\n   * Constraint for the learnable alpha.\n   */\n  alphaConstraint?: Constraint;\n\n  /**\n   * The axes along which to share learnable parameters for the activation\n   * function. For example, if the incoming feature maps are from a 2D\n   * convolution with output shape `[numExamples, height, width, channels]`,\n   * and you wish to share parameters across space (height and width) so that\n   * each filter channels has only one set of parameters, set\n   * `shared_axes: [1, 2]`.\n   */\n  sharedAxes?: number|number[];\n}\n\nexport class PReLU extends Layer {\n  /** @nocollapse */\n  static className = 'PReLU';\n  private readonly alphaInitializer: Initializer;\n  private readonly alphaRegularizer: Regularizer;\n  private readonly alphaConstraint: Constraint;\n  private readonly sharedAxes: number[];\n  private alpha: LayerVariable;\n\n  readonly DEFAULT_ALPHA_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args?: PReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.supportsMasking = true;\n    this.alphaInitializer =\n        getInitializer(args.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER);\n    this.alphaRegularizer = getRegularizer(args.alphaRegularizer);\n    this.alphaConstraint = getConstraint(args.alphaConstraint);\n    if (args.sharedAxes == null) {\n      this.sharedAxes = null;\n    } else if (Array.isArray(args.sharedAxes)) {\n      this.sharedAxes = args.sharedAxes;\n    } else if (typeof args.sharedAxes === 'number') {\n      this.sharedAxes = [args.sharedAxes];\n    } else {\n      throw new ValueError(\n          `Expected sharedAxes to be a number or an array of numbers, ` +\n          `but got ${args.sharedAxes}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]) {\n    inputShape = getExactlyOneShape(inputShape);\n    const paramShape: Shape = inputShape.slice(1);\n    if (this.sharedAxes != null) {\n      for (const i of this.sharedAxes) {\n        paramShape[i - 1] = 1;\n      }\n    }\n    this.alpha = this.addWeight(\n        'alpha', paramShape, 'float32', this.alphaInitializer,\n        this.alphaRegularizer, true, this.alphaConstraint);\n    // Set input spec.\n    const axes: {[axis: number]: number} = {};\n    if (this.sharedAxes != null) {\n      for (let i = 1; i < inputShape.length; ++i) {\n        axes[i] = inputShape[i];\n      }\n    }\n    this.inputSpec = [new InputSpec({\n      ndim: inputShape.length,\n      axes,\n    })];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    return prelu(inputs, this.alpha.read());\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      alphaInitializer: serializeInitializer(this.alphaInitializer),\n      alphaRegularizer: serializeRegularizer(this.alphaRegularizer),\n      alphaConstraint: serializeConstraint(this.alphaConstraint),\n      sharedAxes: this.sharedAxes\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(PReLU);\n\nexport declare interface ELULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `1.0`.\n   */\n  alpha?: number;\n}\n\nexport class ELU extends Layer {\n  /** @nocollapse */\n  static className = 'ELU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 1.0;\n\n  constructor(args?: ELULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    if (args.alpha != null && args.alpha !== this.DEFAULT_ALPHA) {\n      throw new NotImplementedError(\n          `Non-default alpha value (${args.alpha}) is not supported by the ` +\n          `ELU layer yet.`);\n    }\n\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return elu(x);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ELU);\n\nexport declare interface ThresholdedReLULayerArgs extends LayerArgs {\n  /**\n   * Float >= 0. Threshold location of activation.\n   */\n  theta?: number;\n}\n\nexport class ThresholdedReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ThresholdedReLU';\n  readonly theta: number;\n\n  readonly DEFAULT_THETA = 1.0;\n\n  constructor(args?: ThresholdedReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.theta = args.theta == null ? this.DEFAULT_THETA : args.theta;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return x.mul(cast(x.greater(this.theta), 'float32'));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {theta: this.theta};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ThresholdedReLU);\n\nexport declare interface SoftmaxLayerArgs extends LayerArgs {\n  /**\n   * Integer, axis along which the softmax normalization is applied.\n   * Defaults to `-1` (i.e., the last axis).\n   */\n  axis?: number;\n}\n\nexport class Softmax extends Layer {\n  /** @nocollapse */\n  static className = 'Softmax';\n  readonly axis: number;\n  readonly softmax: (t: Tensor, a?: number) => Tensor;\n  readonly DEFAULT_AXIS = 1.0;\n\n  constructor(args?: SoftmaxLayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.softmax = new softmaxActivation().apply;\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return this.softmax(x, this.axis);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {axis: this.axis};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Softmax);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {ValueError} from '../errors';\nimport {PaddingMode} from '../keras_format/common';\n\nimport {pyListRepeat} from './generic_utils';\nimport {isInteger, max} from './math_utils';\n\n/**\n * Transforms a single number of array of numbers into an array of numbers.\n * @param value\n * @param n: The size of the tuple to be returned.\n * @param name: Name of the parameter, used for generating error messages.\n * @returns An array of numbers.\n */\nexport function normalizeArray(\n    value: number|number[], n: number, name: string): number[] {\n  if (typeof value === 'number') {\n    return pyListRepeat(value, n);\n  } else {\n    if (value.length !== n) {\n      throw new ValueError(\n          `The ${name} argument must be an integer or tuple of ${n} integers.` +\n          ` Received: ${value.length} elements.`);\n    }\n    for (let i = 0; i < n; ++i) {\n      const singleValue = value[i];\n      if (!isInteger(singleValue)) {\n        throw new ValueError(\n            `The ${name} argument must be an integer or tuple of ${n}` +\n            ` integers. Received: ${JSON.stringify(value)} including a` +\n            ` non-integer number ${singleValue}`);\n      }\n    }\n    return value;\n  }\n}\n\n/**\n * Determines output length of a convolution given input length.\n * @param inputLength\n * @param filterSize\n * @param padding\n * @param stride\n * @param dilation: dilation rate.\n */\nexport function convOutputLength(\n    inputLength: number, filterSize: number, padding: PaddingMode,\n    stride: number, dilation = 1): number {\n  if (inputLength == null) {\n    return inputLength;\n  }\n  const dilatedFilterSize = filterSize + (filterSize - 1) * (dilation - 1);\n  let outputLength: number;\n  if (padding === 'same') {\n    outputLength = inputLength;\n  } else {  // VALID\n    outputLength = inputLength - dilatedFilterSize + 1;\n  }\n  return Math.floor((outputLength + stride - 1) / stride);\n}\n\nexport function deconvLength(\n    dimSize: number, strideSize: number, kernelSize: number,\n    padding: PaddingMode): number {\n  if (dimSize == null) {\n    return null;\n  }\n\n  if (padding === 'valid') {\n    dimSize = dimSize * strideSize + max([kernelSize - strideSize, 0]);\n  } else if (padding === 'same') {\n    dimSize = dimSize * strideSize;\n  } else {\n    throw new ValueError(`Unsupport padding mode: ${padding}.`);\n  }\n  return dimSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {fused, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, PaddingMode, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, deconvLength, normalizeArray} from '../utils/conv_utils';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Transpose and cast the input before the conv2d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv2DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  // TODO(cais): Cast type to float32 if not.\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 1]);  // NCHW -> NHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * Transpose and cast the input before the conv3d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv3DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 4, 1]);  // NCDHW -> NDHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * 1D-convolution with bias added.\n *\n * Porting Note: This function does not exist in the Python Keras backend.\n *   It is exactly the same as `conv2d`, except the added `bias`.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.\n * @param bias Bias, rank-3, of shape `[outDepth]`.\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    // Check the ranks of x, kernel and bias.\n    if (x.shape.length !== 3) {\n      throw new ValueError(\n          `The input of a conv1dWithBias operation should be 3, but is ` +\n          `${x.shape.length} instead.`);\n    }\n    if (kernel.shape.length !== 3) {\n      throw new ValueError(\n          `The kernel for a conv1dWithBias operation should be 3, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    if (bias != null && bias.shape.length !== 1) {\n      throw new ValueError(\n          `The bias for a conv1dWithBias operation should be 1, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    // TODO(cais): Support CAUSAL padding mode.\n    if (dataFormat === 'channelsFirst') {\n      x = tfc.transpose(x, [0, 2, 1]);  // NCW -> NWC.\n    }\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    let y: Tensor = tfc.conv1d(\n        x as Tensor2D | Tensor3D, kernel as Tensor3D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias);\n    }\n    return y;\n  });\n}\n\n/**\n * 1D-convolution.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.s\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1d(\n    x: Tensor, kernel: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv1dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 2D pooling.\n */\nexport function conv2d(\n    x: Tensor, kernel: Tensor, strides = [1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv2dWithBiasActivation(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution with an added bias and optional activation.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv2d`, except the added `bias`.\n */\nexport function conv2dWithBiasActivation(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat, dilationRate?: [number, number],\n    activation: fused.Activation = null): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 3 && x.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects input to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    if (kernel.rank !== 3 && kernel.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects kernel to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.fused.conv2d({\n      x: y as Tensor3D | Tensor4D,\n      filter: kernel as Tensor4D,\n      strides: strides as [number, number],\n      pad: padding === 'same' ? 'same' : 'valid',\n      dilations: dilationRate,\n      dataFormat: 'NHWC',\n      bias,\n      activation\n    });\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\n/**\n * 3D Convolution.\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 3D convolution.\n */\nexport function conv3d(\n    x: Tensor, kernel: Tensor, strides = [1, 1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv3dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 3D Convolution with an added bias.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv3d`, except the added `bias`.\n */\nexport function conv3dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 4 && x.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects input to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    if (kernel.rank !== 4 && kernel.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects kernel to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    let y = preprocessConv3DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv3dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.conv3d(\n        y as Tensor4D | tfc.Tensor<tfc.Rank.R5>,\n        kernel as tfc.Tensor<tfc.Rank.R5>, strides as [number, number, number],\n        padding === 'same' ? 'same' : 'valid', 'NDHWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias as Tensor1D);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);\n    }\n    return y;\n  });\n}\n\n/**\n * Base LayerConfig for depthwise and non-depthwise convolutional layers.\n */\nexport declare interface BaseConvLayerArgs extends LayerArgs {\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number]|[number, number, number];\n\n  /**\n   * Activation function of the layer.\n   *\n   * If you don't specify the activation, none is applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector. Defaults to `true`.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the convolutional kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the convolutional kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\n/**\n * LayerConfig for non-depthwise convolutional layers.\n * Applies to non-depthwise convolution of all ranks (e.g, Conv1D, Conv2D,\n * Conv3D).\n */\nexport declare interface ConvLayerArgs extends BaseConvLayerArgs {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n}\n\n/**\n * Abstract convolution layer.\n */\nexport abstract class BaseConv extends Layer {\n  protected readonly rank: number;\n  protected readonly kernelSize: number[];\n  protected readonly strides: number[];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n  protected readonly activation: Activation;\n  protected readonly useBias: boolean;\n  protected readonly dilationRate: number[];\n\n  // Bias-related members are here because all convolution subclasses use the\n  // same configuration parmeters to control bias.  Kernel-related members\n  // are in subclass `Conv` because some subclasses use different parameters to\n  // control kernel properties, for instance, `DepthwiseConv2D` uses\n  // `depthwiseInitializer` instead of `kernelInitializer`.\n  protected readonly biasInitializer?: Initializer;\n  protected readonly biasConstraint?: Constraint;\n  protected readonly biasRegularizer?: Regularizer;\n\n  protected bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(rank: number, args: BaseConvLayerArgs) {\n    super(args as LayerArgs);\n    BaseConv.verifyArgs(args);\n    this.rank = rank;\n    generic_utils.assertPositiveInteger(this.rank, 'rank');\n    if (this.rank !== 1 && this.rank !== 2 && this.rank !== 3) {\n      throw new NotImplementedError(\n          `Convolution layer for rank other than 1, 2, or 3 (${\n              this.rank}) is ` +\n          `not implemented yet.`);\n    }\n    this.kernelSize = normalizeArray(args.kernelSize, rank, 'kernelSize');\n    this.strides = normalizeArray(\n        args.strides == null ? 1 : args.strides, rank, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.activation = getActivation(args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.dilationRate = normalizeArray(\n        args.dilationRate == null ? 1 : args.dilationRate, rank,\n        'dilationRate');\n    if (this.rank === 1 &&\n        (Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)) {\n      throw new ValueError(\n          `dilationRate must be a number or an array of a single number ` +\n          `for 1D convolution, but received ` +\n          `${JSON.stringify(this.dilationRate)}`);\n    } else if (this.rank === 2) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate = [this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 2) {\n        throw new ValueError(\n            `dilationRate must be a number or array of two numbers for 2D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    } else if (this.rank === 3) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate =\n            [this.dilationRate, this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 3) {\n        throw new ValueError(\n            `dilationRate must be a number or array of three numbers for 3D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    }\n  }\n\n  protected static verifyArgs(args: BaseConvLayerArgs) {\n    // Check config.kernelSize type and shape.\n    generic_utils.assert(\n        'kernelSize' in args, `required key 'kernelSize' not in config`);\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 3)) {\n      throw new ValueError(\n          `BaseConv expects config.kernelSize to be number or number[] with ` +\n          `length 1, 2, or 3, but received ${\n              JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      kernelSize: this.kernelSize,\n      strides: this.strides,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\n/**\n * Abstract nD convolution layer.  Ancestor of convolution layers which reduce\n * across channels, i.e., Conv1D and Conv2D, but not DepthwiseConv2D.\n */\nexport abstract class Conv extends BaseConv {\n  protected readonly filters: number;\n\n  protected kernel: LayerVariable = null;\n\n  // Bias-related properties are stored in the superclass `BaseConv` because all\n  // convolution subclasses use the same configuration parameters to control\n  // bias. Kernel-related properties are defined here rather than in the\n  // superclass because some convolution subclasses use different names and\n  // configuration parameters for their internal kernel state.\n  protected readonly kernelInitializer?: Initializer;\n  protected readonly kernelConstraint?: Constraint;\n  protected readonly kernelRegularizer?: Regularizer;\n\n  constructor(rank: number, args: ConvLayerArgs) {\n    super(rank, args as BaseConvLayerArgs);\n    Conv.verifyArgs(args);\n    this.filters = args.filters;\n    generic_utils.assertPositiveInteger(this.filters, 'filters');\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n    const inputDim = inputShape[channelAxis];\n\n    const kernelShape = this.kernelSize.concat([inputDim, this.filters]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.inputSpec = [{ndim: this.rank + 2, axes: {[channelAxis]: inputDim}}];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs: Tensor;\n      const biasValue = this.bias == null ? null : this.bias.read();\n      const fusedActivationName = generic_utils.mapActivationToFusedKernel(\n                                            this.activation.getClassName());\n\n      if (fusedActivationName != null && this.rank === 2) {\n        outputs = conv2dWithBiasActivation(\n            inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n            this.dataFormat, this.dilationRate as [number, number],\n            fusedActivationName);\n      } else {\n        if (this.rank === 1) {\n          outputs = conv1dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides[0],\n              this.padding, this.dataFormat, this.dilationRate[0]);\n        } else if (this.rank === 2) {\n          // TODO(cais): Move up to constructor.\n          outputs = conv2dWithBiasActivation(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number]);\n        } else if (this.rank === 3) {\n          outputs = conv3dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number, number]);\n        } else {\n          throw new NotImplementedError(\n              'convolutions greater than 3D are not implemented yet.');\n        }\n\n        if (this.activation != null) {\n          outputs = this.activation.apply(outputs);\n        }\n      }\n\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const newSpace: number[] = [];\n    const space = (this.dataFormat === 'channelsLast') ?\n        inputShape.slice(1, inputShape.length - 1) :\n        inputShape.slice(2);\n    for (let i = 0; i < space.length; ++i) {\n      const newDim = convOutputLength(\n          space[i], this.kernelSize[i], this.padding, this.strides[i],\n          typeof this.dilationRate === 'number' ? this.dilationRate :\n                                                  this.dilationRate[i]);\n      newSpace.push(newDim);\n    }\n\n    let outputShape = [inputShape[0]];\n    if (this.dataFormat === 'channelsLast') {\n      outputShape = outputShape.concat(newSpace);\n      outputShape.push(this.filters);\n    } else {\n      outputShape.push(this.filters);\n      outputShape = outputShape.concat(newSpace);\n    }\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      filters: this.filters,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // Check config.filters type, shape, and value.\n    if (!('filters' in args) || typeof args.filters !== 'number' ||\n        args.filters < 1) {\n      throw new ValueError(\n          `Convolution layer expected config.filters to be a 'number' > 0 ` +\n          `but got ${JSON.stringify(args.filters)}`);\n    }\n  }\n}\n\nexport class Conv2D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv2D';\n  constructor(args: ConvLayerArgs) {\n    super(2, args);\n    Conv2D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if ((typeof args.kernelSize !== 'number') &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 2)) {\n      throw new ValueError(\n          `Conv2D expects config.kernelSize to be number or number[] with ` +\n          `length 1 or 2, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv2D);\n\nexport class Conv3D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv3D';\n  constructor(args: ConvLayerArgs) {\n    super(3, args);\n    Conv3D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number') {\n      if (!(Array.isArray(args.kernelSize) &&\n            (args.kernelSize.length === 1 || args.kernelSize.length === 3))) {\n        throw new ValueError(\n            `Conv3D expects config.kernelSize to be number or` +\n            ` [number, number, number], but received ${\n                JSON.stringify(args.kernelSize)}.`);\n      }\n    }\n  }\n}\nserialization.registerClass(Conv3D);\n\nexport class Conv2DTranspose extends Conv2D {\n  /** @nocollapse */\n  static className = 'Conv2DTranspose';\n  inputSpec: InputSpec[];\n\n  constructor(args: ConvLayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n\n    if (this.padding !== 'same' && this.padding !== 'valid') {\n      throw new ValueError(\n          `Conv2DTranspose currently supports only padding modes 'same' ` +\n          `and 'valid', but received padding mode ${this.padding}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    if (inputShape.length !== 4) {\n      throw new ValueError(\n          'Input should have rank 4; Received input shape: ' +\n          JSON.stringify(inputShape));\n    }\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          'The channel dimension of the inputs should be defined. ' +\n          'Found `None`.');\n    }\n    const inputDim = inputShape[channelAxis];\n    const kernelShape = this.kernelSize.concat([this.filters, inputDim]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, 'float32', this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    // Set input spec.\n    this.inputSpec =\n        [new InputSpec({ndim: 4, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs);\n      if (input.shape.length !== 4) {\n        throw new ValueError(\n            `Conv2DTranspose.call() expects input tensor to be rank-4, but ` +\n            `received a tensor of rank-${input.shape.length}`);\n      }\n\n      const inputShape = input.shape;\n      const batchSize = inputShape[0];\n\n      let hAxis: number;\n      let wAxis: number;\n      if (this.dataFormat === 'channelsFirst') {\n        hAxis = 2;\n        wAxis = 3;\n      } else {\n        hAxis = 1;\n        wAxis = 2;\n      }\n\n      const height = inputShape[hAxis];\n      const width = inputShape[wAxis];\n      const kernelH = this.kernelSize[0];\n      const kernelW = this.kernelSize[1];\n      const strideH = this.strides[0];\n      const strideW = this.strides[1];\n\n      // Infer the dynamic output shape.\n      const outHeight = deconvLength(height, strideH, kernelH, this.padding);\n      const outWidth = deconvLength(width, strideW, kernelW, this.padding);\n\n      // Porting Note: We don't branch based on `this.dataFormat` here,\n      // because\n      //   the tjfs-core function `conv2dTranspose` called below always\n      //   assumes channelsLast.\n      const outputShape: [number, number, number, number] =\n          [batchSize, outHeight, outWidth, this.filters];\n\n      if (this.dataFormat !== 'channelsLast') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n      }\n      let outputs = tfc.conv2dTranspose(\n          input as Tensor4D, this.kernel.read() as Tensor4D, outputShape,\n          this.strides as [number, number], this.padding as 'same' | 'valid');\n      if (this.dataFormat !== 'channelsLast') {\n        outputs = tfc.transpose(outputs, [0, 3, 1, 2]);\n      }\n\n      if (this.bias != null) {\n        outputs =\n            K.biasAdd(outputs, this.bias.read(), this.dataFormat) as Tensor4D;\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs) as Tensor4D;\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n\n    let channelAxis: number;\n    let heightAxis: number;\n    let widthAxis: number;\n    if (this.dataFormat === 'channelsFirst') {\n      channelAxis = 1;\n      heightAxis = 2;\n      widthAxis = 3;\n    } else {\n      channelAxis = 3;\n      heightAxis = 1;\n      widthAxis = 2;\n    }\n\n    const kernelH = this.kernelSize[0];\n    const kernelW = this.kernelSize[1];\n    const strideH = this.strides[0];\n    const strideW = this.strides[1];\n\n    outputShape[channelAxis] = this.filters;\n    outputShape[heightAxis] =\n        deconvLength(outputShape[heightAxis], strideH, kernelH, this.padding);\n    outputShape[widthAxis] =\n        deconvLength(outputShape[widthAxis], strideW, kernelW, this.padding);\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['dilationRate'];\n    return config;\n  }\n}\nserialization.registerClass(Conv2DTranspose);\n\nexport declare interface SeparableConvLayerArgs extends ConvLayerArgs {\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal\n   * to `filtersIn * depthMultiplier`. Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the pointwise kernel matrix.\n   */\n  pointwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the pointwise kernel matrix.\n   */\n  pointwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the pointwise kernel matrix.\n   */\n  pointwiseConstraint?: ConstraintIdentifier|Constraint;\n}\n\nexport class SeparableConv extends Conv {\n  /** @nocollapse */\n  static className = 'SeparableConv';\n\n  readonly depthMultiplier: number;\n\n  protected readonly depthwiseInitializer?: Initializer;\n  protected readonly depthwiseRegularizer?: Regularizer;\n  protected readonly depthwiseConstraint?: Constraint;\n  protected readonly pointwiseInitializer?: Initializer;\n  protected readonly pointwiseRegularizer?: Regularizer;\n  protected readonly pointwiseConstraint?: Constraint;\n\n  readonly DEFAULT_DEPTHWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n  readonly DEFAULT_POINTWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n\n  protected depthwiseKernel: LayerVariable = null;\n  protected pointwiseKernel: LayerVariable = null;\n\n  constructor(rank: number, config?: SeparableConvLayerArgs) {\n    super(rank, config);\n\n    if (config.filters == null) {\n      throw new ValueError(\n          'The `filters` configuration field is required by SeparableConv, ' +\n          'but is unspecified.');\n    }\n    if (config.kernelInitializer != null || config.kernelRegularizer != null ||\n        config.kernelConstraint != null) {\n      throw new ValueError(\n          'Fields kernelInitializer, kernelRegularizer and kernelConstraint ' +\n          'are invalid for SeparableConv2D. Use depthwiseInitializer, ' +\n          'depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, ' +\n          'pointwiseRegularizer and pointwiseConstraint instead.');\n    }\n    if (config.padding != null && config.padding !== 'same' &&\n        config.padding !== 'valid') {\n      throw new ValueError(\n          `SeparableConv${this.rank}D supports only padding modes: ` +\n          `'same' and 'valid', but received ${JSON.stringify(config.padding)}`);\n    }\n\n    this.depthMultiplier =\n        config.depthMultiplier == null ? 1 : config.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER);\n    this.depthwiseRegularizer = getRegularizer(config.depthwiseRegularizer);\n    this.depthwiseConstraint = getConstraint(config.depthwiseConstraint);\n    this.pointwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER);\n    this.pointwiseRegularizer = getRegularizer(config.pointwiseRegularizer);\n    this.pointwiseConstraint = getConstraint(config.pointwiseConstraint);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < this.rank + 2) {\n      throw new ValueError(\n          `Inputs to SeparableConv${this.rank}D should have rank ` +\n          `${this.rank + 2}, but received input shape: ` +\n          `${JSON.stringify(inputShape)}`);\n    }\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          `The channel dimension of the inputs should be defined, ` +\n          `but found ${JSON.stringify(inputShape[channelAxis])}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape =\n        this.kernelSize.concat([inputDim, this.depthMultiplier]);\n    const pointwiseKernelShape = [];\n    for (let i = 0; i < this.rank; ++i) {\n      pointwiseKernelShape.push(1);\n    }\n    pointwiseKernelShape.push(inputDim * this.depthMultiplier, this.filters);\n\n    const trainable = true;\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, 'float32',\n        this.depthwiseInitializer, this.depthwiseRegularizer, trainable,\n        this.depthwiseConstraint);\n    this.pointwiseKernel = this.addWeight(\n        'pointwise_kernel', pointwiseKernelShape, 'float32',\n        this.pointwiseInitializer, this.pointwiseRegularizer, trainable,\n        this.pointwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, trainable, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n\n    this.inputSpec =\n        [new InputSpec({ndim: this.rank + 2, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      let output: Tensor;\n      if (this.rank === 1) {\n        throw new NotImplementedError(\n            '1D separable convolution is not implemented yet.');\n      } else if (this.rank === 2) {\n        if (this.dataFormat === 'channelsFirst') {\n          inputs = tfc.transpose(inputs, [0, 2, 3, 1]);  // NCHW -> NHWC.\n        }\n\n        output = tfc.separableConv2d(\n            inputs as Tensor4D, this.depthwiseKernel.read() as Tensor4D,\n            this.pointwiseKernel.read() as Tensor4D,\n            this.strides as [number, number], this.padding as 'same' | 'valid',\n            this.dilationRate as [number, number], 'NHWC');\n      }\n\n      if (this.useBias) {\n        output = K.biasAdd(output, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      if (this.dataFormat === 'channelsFirst') {\n        output = tfc.transpose(output, [0, 3, 1, 2]);  // NHWC -> NCHW.\n      }\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['kernelInitializer'];\n    delete config['kernelRegularizer'];\n    delete config['kernelConstraint'];\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['pointwiseInitializer'] =\n        serializeInitializer(this.pointwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['pointwiseRegularizer'] =\n        serializeRegularizer(this.pointwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseConstraint);\n    config['pointwiseConstraint'] =\n        serializeConstraint(this.pointwiseConstraint);\n    return config;\n  }\n}\n\nexport class SeparableConv2D extends SeparableConv {\n  /** @nocollapse */\n  static className = 'SeparableConv2D';\n  constructor(args?: SeparableConvLayerArgs) {\n    super(2, args);\n  }\n}\nserialization.registerClass(SeparableConv2D);\n\nexport class Conv1D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv1D';\n  constructor(args: ConvLayerArgs) {\n    super(1, args);\n    Conv1D.verifyArgs(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['dataFormat'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 1)) {\n      throw new ValueError(\n          `Conv1D expects config.kernelSize to be number or number[] with ` +\n          `length 1, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv1D);\n\nexport declare interface Cropping2DLayerArgs extends LayerArgs {\n  /**\n   * Dimension of the cropping along the width and the height.\n   * - If integer: the same symmetric cropping\n   *  is applied to width and height.\n   * - If list of 2 integers:\n   *   interpreted as two different\n   *   symmetric cropping values for height and width:\n   *   `[symmetric_height_crop, symmetric_width_crop]`.\n   * - If a list of 2 list of 2 integers:\n   *   interpreted as\n   *   `[[top_crop, bottom_crop], [left_crop, right_crop]]`\n   */\n  cropping: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   * `channels_first` corresponds to inputs with shape\n   *   `(batch, channels, ...)`\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class Cropping2D extends Layer {\n  /** @nocollapse */\n  static className = 'Cropping2D';\n  protected readonly cropping: [[number, number], [number, number]];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Cropping2DLayerArgs) {\n    super(args);\n    if (typeof args.cropping === 'number') {\n      this.cropping =\n          [[args.cropping, args.cropping], [args.cropping, args.cropping]];\n    } else if (typeof args.cropping[0] === 'number') {\n      this.cropping = [\n        [args.cropping[0], args.cropping[0]],\n        [args.cropping[1] as number, args.cropping[1] as number]\n      ];\n    } else {\n      this.cropping = args.cropping as [[number, number], [number, number]];\n    }\n    this.dataFormat =\n        args.dataFormat === undefined ? 'channelsLast' : args.dataFormat;\n    this.inputSpec = [{ndim: 4}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      return [\n        inputShape[0], inputShape[1],\n        inputShape[2] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[3] - this.cropping[1][0] - this.cropping[1][1]\n      ];\n    } else {\n      return [\n        inputShape[0],\n        inputShape[1] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[2] - this.cropping[1][0] - this.cropping[1][1], inputShape[3]\n      ];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      if (this.dataFormat === 'channelsLast') {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[1] - this.cropping[0][0] - this.cropping[0][1], 2);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[2] - this.cropping[1][1] - this.cropping[1][0], 3);\n      } else {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[2] - this.cropping[0][0] - this.cropping[0][1], 3);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[3] - this.cropping[1][1] - this.cropping[1][0], 4);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {cropping: this.cropping, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Cropping2D);\n\nexport declare interface UpSampling2DLayerArgs extends LayerArgs {\n  /**\n   * The upsampling factors for rows and columns.\n   *\n   * Defaults to `[2, 2]`.\n   */\n  size?: number[];\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `\"channelsLast\"` corresponds to inputs with shape\n   *   `[batch, ..., channels]`\n   *\n   *  `\"channelsFirst\"` corresponds to inputs with shape `[batch, channels,\n   * ...]`.\n   *\n   * Defaults to `\"channelsLast\"`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class UpSampling2D extends Layer {\n  /** @nocollapse */\n  static className = 'UpSampling2D';\n  protected readonly DEFAULT_SIZE = [2, 2];\n  protected readonly size: number[];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: UpSampling2DLayerArgs) {\n    super(args);\n    this.inputSpec = [{ndim: 4}];\n    this.size = args.size == null ? this.DEFAULT_SIZE : args.size;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      const height =\n          inputShape[2] == null ? null : this.size[0] * inputShape[2];\n      const width = inputShape[3] == null ? null : this.size[1] * inputShape[3];\n      return [inputShape[0], inputShape[1], height, width];\n    } else {\n      const height =\n          inputShape[1] == null ? null : this.size[0] * inputShape[1];\n      const width = inputShape[2] == null ? null : this.size[1] * inputShape[2];\n      return [inputShape[0], height, width, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs) as Tensor4D;\n      const inputShape = input.shape;\n\n      if (this.dataFormat === 'channelsFirst') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n        const height = this.size[0] * inputShape[2];\n        const width = this.size[1] * inputShape[3];\n        const resized = input.resizeNearestNeighbor([height, width]);\n        return tfc.transpose(resized, [0, 3, 1, 2]);\n      } else {\n        const height = this.size[0] * inputShape[1];\n        const width = this.size[1] * inputShape[2];\n        return input.resizeNearestNeighbor([height, width]);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {size: this.size, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(UpSampling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Depthwise Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {BaseConv, BaseConvLayerArgs, ConvLayerArgs, preprocessConv2DInput} from './convolutional';\n\n/**\n * 2D convolution with separable filters.\n * @param x Input tensor.\n * @param depthwiseKernel Convolution kernel for depthwise convolution.\n * @param strides Strides (Array of two integers).\n * @param padding Padding model.\n * @param dataFormat Data format.\n * @param dilationRate Array of two integers, dilation rates for the separable\n *   convolution.\n * @returns Output tensor.\n * @throws ValueError If depthwiseKernel is not a 4D array.\n */\nexport function depthwiseConv2d(\n    x: Tensor, depthwiseKernel: Tensor, strides: [number, number] = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `Input for depthwiseConv2d is required to be 4-D, but is instead ` +\n          `${x.rank}-D`);\n    }\n    if (depthwiseKernel.rank !== 4) {\n      throw new ValueError(\n          `depthwiseKernel is required to be 4-D, but is instead ` +\n          `${depthwiseKernel.rank}-D`);\n    }\n    y = tfc.depthwiseConv2d(\n        y as Tensor4D, depthwiseKernel as Tensor4D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NHWC', dilationRate);\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\nexport declare interface DepthwiseConv2DLayerArgs extends BaseConvLayerArgs {\n  /**\n   * An integer or Array of 2 integers, specifying the width and height of the\n   * 2D convolution window. Can be a single integer to specify the same value\n   * for all spatial dimensions.\n   */\n  kernelSize: number|[number, number];\n\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal to\n   * `filtersIn * depthMultiplier`.\n   * Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   * Default: GlorotNormal.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regulzarizer function for the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class DepthwiseConv2D extends BaseConv {\n  /** @nocollapse */\n  static className = 'DepthwiseConv2D';\n  private readonly depthMultiplier: number;\n  private readonly depthwiseInitializer: Initializer;\n  private readonly depthwiseConstraint: Constraint;\n  private readonly depthwiseRegularizer: Regularizer;\n\n  private depthwiseKernel: LayerVariable = null;\n\n  constructor(args: DepthwiseConv2DLayerArgs) {\n    super(2, args as ConvLayerArgs);\n    this.depthMultiplier =\n        args.depthMultiplier == null ? 1 : args.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        args.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.depthwiseConstraint = getConstraint(args.depthwiseConstraint);\n    this.depthwiseRegularizer = getRegularizer(args.depthwiseRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 4) {\n      throw new ValueError(\n          `Inputs to DepthwiseConv2D should have rank 4. ` +\n          `Received input shape: ${JSON.stringify(inputShape)}.`);\n    }\n    const channelAxis = this.dataFormat === 'channelsFirst' ? 1 : 3;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          'The channel dimension of the inputs to DepthwiseConv2D should ' +\n          `be defined, but is not (${inputShape[channelAxis]}).`);\n    }\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape: Shape = [\n      this.kernelSize[0], this.kernelSize[1], inputDim, this.depthMultiplier\n    ];\n\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, null,\n        this.depthwiseInitializer, this.depthwiseRegularizer, true,\n        this.depthwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [inputDim * this.depthMultiplier], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs = depthwiseConv2d(\n          inputs, this.depthwiseKernel.read(), this.strides as [number, number],\n          this.padding, this.dataFormat, null);\n      // TODO(cais): Add support for dilation.\n      if (this.useBias) {\n        outputs = K.biasAdd(outputs, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs);\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    const cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    const outFilters = this.dataFormat === 'channelsFirst' ?\n        inputShape[1] * this.depthMultiplier :\n        inputShape[3] * this.depthMultiplier;\n    const outRows = convOutputLength(\n        rows, this.kernelSize[0], this.padding, this.strides[0]);\n    const outCols = convOutputLength(\n        cols, this.kernelSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], outFilters, outRows, outCols];\n    } else {\n      // In this case, assume 'channelsLast'.\n      return [inputShape[0], outRows, outCols, outFilters];\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    config['depthMultiplier'] = this.depthMultiplier;\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseRegularizer);\n    return config;\n  }\n}\nserialization.registerClass(DepthwiseConv2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Recurrent Neural Network Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, SymbolicTensor} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, Ones, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs, RnnStepFunction} from '../types';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor, isArrayOfShapes} from '../utils/types_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\nimport {deserialize} from './serialization';\n\n/**\n * Standardize `apply()` args to a single list of tensor inputs.\n *\n * When running a model loaded from file, the input tensors `initialState` and\n * `constants` are passed to `RNN.apply()` as part of `inputs` instead of the\n * dedicated kwargs fields. `inputs` consists of\n * `[inputs, initialState0, initialState1, ..., constant0, constant1]` in this\n * case.\n * This method makes sure that arguments are\n * separated and that `initialState` and `constants` are `Array`s of tensors\n * (or None).\n *\n * @param inputs Tensor or `Array` of  tensors.\n * @param initialState Tensor or `Array` of tensors or `null`/`undefined`.\n * @param constants Tensor or `Array` of tensors or `null`/`undefined`.\n * @returns An object consisting of\n *   inputs: A tensor.\n *   initialState: `Array` of tensors or `null`.\n *   constants: `Array` of tensors or `null`.\n * @throws ValueError, if `inputs` is an `Array` but either `initialState` or\n *   `constants` is provided.\n */\nexport function standardizeArgs(\n    inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    initialState: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    constants: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    numConstants?: number): {\n  inputs: Tensor|SymbolicTensor,\n  initialState: Tensor[]|SymbolicTensor[],\n  constants: Tensor[]|SymbolicTensor[]\n} {\n  if (Array.isArray(inputs)) {\n    if (initialState != null || constants != null) {\n      throw new ValueError(\n          'When inputs is an array, neither initialState or constants ' +\n          'should be provided');\n    }\n    if (numConstants != null) {\n      constants = inputs.slice(inputs.length - numConstants, inputs.length);\n      inputs = inputs.slice(0, inputs.length - numConstants);\n    }\n    if (inputs.length > 1) {\n      initialState = inputs.slice(1, inputs.length);\n    }\n    inputs = inputs[0];\n  }\n\n  function toListOrNull(x: Tensor|Tensor[]|SymbolicTensor|\n                        SymbolicTensor[]): Tensor[]|SymbolicTensor[] {\n    if (x == null || Array.isArray(x)) {\n      return x as Tensor[] | SymbolicTensor[];\n    } else {\n      return [x] as Tensor[] | SymbolicTensor[];\n    }\n  }\n\n  initialState = toListOrNull(initialState);\n  constants = toListOrNull(constants);\n\n  return {inputs, initialState, constants};\n}\n\n/**\n * Iterates over the time dimension of a tensor.\n *\n * @param stepFunction RNN step function.\n *   Parameters:\n *     inputs: tensor with shape `[samples, ...]` (no time dimension),\n *       representing input for the batch of samples at a certain time step.\n *     states: an Array of tensors.\n *   Returns:\n *     outputs: tensor with shape `[samples, outputDim]` (no time dimension).\n *     newStates: list of tensors, same length and shapes as `states`. The first\n *       state in the list must be the output tensor at the previous timestep.\n * @param inputs Tensor of temporal data of shape `[samples, time, ...]` (at\n *   least 3D).\n * @param initialStates Tensor with shape `[samples, outputDim]` (no time\n *   dimension), containing the initial values of the states used in the step\n *   function.\n * @param goBackwards If `true`, do the iteration over the time dimension in\n *   reverse order and return the reversed sequence.\n * @param mask Binary tensor with shape `[sample, time, 1]`, with a zero for\n *   every element that is masked.\n * @param constants An Array of constant values passed at each step.\n * @param unroll Whether to unroll the RNN or to use a symbolic loop. *Not*\n *   applicable to this imperative deeplearn.js backend. Its value is ignored.\n * @param needPerStepOutputs Whether the per-step outputs are to be\n *   concatenated into a single tensor and returned (as the second return\n *   value). Default: `false`. This arg is included so that the relatively\n *   expensive concatenation of the stepwise outputs can be omitted unless\n *   the stepwise outputs need to be kept (e.g., for an LSTM layer of which\n *   `returnSequence` is `true`.)\n * @returns An Array: `[lastOutput, outputs, newStates]`.\n *   lastOutput: the lastest output of the RNN, of shape `[samples, ...]`.\n *   outputs: tensor with shape `[samples, time, ...]` where each entry\n *     `output[s, t]` is the output of the step function at time `t` for sample\n *     `s`. This return value is provided if and only if the\n *     `needPerStepOutputs` is set as `true`. If it is set as `false`, this\n *     return value will be `undefined`.\n *   newStates: Array of tensors, latest states returned by the step function,\n *      of shape `(samples, ...)`.\n * @throws ValueError If input dimension is less than 3.\n *\n * TODO(nielsene): This needs to be tidy-ed.\n */\nexport function rnn(\n    stepFunction: RnnStepFunction, inputs: Tensor, initialStates: Tensor[],\n    goBackwards = false, mask?: Tensor, constants?: Tensor[], unroll = false,\n    needPerStepOutputs = false): [Tensor, Tensor, Tensor[]] {\n  return tfc.tidy(() => {\n    const ndim = inputs.shape.length;\n    if (ndim < 3) {\n      throw new ValueError(`Input should be at least 3D, but is ${ndim}D.`);\n    }\n\n    // Transpose to time-major, i.e., from [batch, time, ...] to [time, batch,\n    // ...].\n    const axes = [1, 0].concat(math_utils.range(2, ndim));\n    inputs = tfc.transpose(inputs, axes);\n\n    if (constants != null) {\n      throw new NotImplementedError(\n          'The rnn() functoin of the deeplearn.js backend does not support ' +\n          'constants yet.');\n    }\n\n    // Porting Note: the unroll option is ignored by the imperative backend.\n    if (unroll) {\n      console.warn(\n          'Backend rnn(): the unroll = true option is not applicable to the ' +\n          'imperative deeplearn.js backend.');\n    }\n\n    if (mask != null) {\n      mask = mask.asType('bool').asType('float32');\n      if (mask.rank === ndim - 1) {\n        mask = tfc.expandDims(mask, -1);\n      }\n      mask = tfc.transpose(mask, axes);\n    }\n\n    if (goBackwards) {\n      inputs = tfc.reverse(inputs, 0);\n      if (mask != null) {\n        mask = tfc.reverse(mask, 0);\n      }\n    }\n\n    // Porting Note: PyKeras with TensorFlow backend uses a symbolic loop\n    //   (tf.while_loop). But for the imperative deeplearn.js backend, we just\n    //   use the usual TypeScript control flow to iterate over the time steps in\n    //   the inputs.\n    // Porting Note: PyKeras patches a \"_use_learning_phase\" attribute to\n    // outputs.\n    //   This is not idiomatic in TypeScript. The info regarding whether we are\n    //   in a learning (i.e., training) phase for RNN is passed in a different\n    //   way.\n\n    const perStepOutputs: Tensor[] = [];\n    let lastOutput: Tensor;\n    let states = initialStates;\n    const timeSteps = inputs.shape[0];\n    const perStepInputs = tfc.unstack(inputs);\n    let perStepMasks: Tensor[];\n    if (mask != null) {\n      perStepMasks = tfc.unstack(mask);\n    }\n\n    for (let t = 0; t < timeSteps; ++t) {\n      const currentInput = perStepInputs[t];\n      const stepOutputs = tfc.tidy(() => stepFunction(currentInput, states));\n\n      if (mask == null) {\n        lastOutput = stepOutputs[0];\n        states = stepOutputs[1];\n      } else {\n        const maskedOutputs = tfc.tidy(() => {\n          const stepMask = perStepMasks[t];\n          const negStepMask = tfc.onesLike(stepMask).sub(stepMask);\n          // TODO(cais): Would tfc.where() be better for performance?\n          const output =\n              stepOutputs[0].mul(stepMask).add(states[0].mul(negStepMask));\n          const newStates = states.map((state, i) => {\n            return stepOutputs[1][i].mul(stepMask).add(state.mul(negStepMask));\n          });\n          return {output, newStates};\n        });\n        lastOutput = maskedOutputs.output;\n        states = maskedOutputs.newStates;\n      }\n\n      if (needPerStepOutputs) {\n        perStepOutputs.push(lastOutput);\n      }\n    }\n    let outputs: Tensor;\n    if (needPerStepOutputs) {\n      const axis = 1;\n      outputs = tfc.stack(perStepOutputs, axis);\n    }\n    return [lastOutput, outputs, states] as [Tensor, Tensor, Tensor[]];\n  });\n}\n\nexport declare interface BaseRNNLayerArgs extends LayerArgs {\n  /**\n   * A RNN cell instance. A RNN cell is a class that has:\n   *   - a `call()` method, which takes `[Tensor, Tensor]` as the\n   *     first input argument. The first item is the input at time t, and\n   *     second item is the cell state at time t.\n   *     The `call()` method returns `[outputAtT, statesAtTPlus1]`.\n   *     The `call()` method of the cell can also take the argument `constants`,\n   *     see section \"Note on passing external constants\" below.\n   *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,\n   *       which are Layer subtypes, to accept two arguments. tfjs-layers does\n   *       not do such overriding. Instead we preseve the `call()` signature,\n   *       which due to its `Tensor|Tensor[]` argument and return value, is\n   *       flexible enough to handle the inputs and states.\n   *   - a `stateSize` attribute. This can be a single integer (single state)\n   *     in which case it is the size of the recurrent state (which should be\n   *     the same as the size of the cell output). This can also be an Array of\n   *     integers (one size per state). In this case, the first entry\n   *     (`stateSize[0]`) should be the same as the size of the cell output.\n   * It is also possible for `cell` to be a list of RNN cell instances, in which\n   * case the cells get stacked on after the other in the RNN, implementing an\n   * efficient stacked RNN.\n   */\n  cell?: RNNCell|RNNCell[];\n\n  /**\n   * Whether to return the last output in the output sequence, or the full\n   * sequence.\n   */\n  returnSequences?: boolean;\n\n  /**\n   * Whether to return the last state in addition to the output.\n   */\n  returnState?: boolean;\n\n  /**\n   * If `true`, process the input sequence backwards and return the reversed\n   * sequence (default: `false`).\n   */\n  goBackwards?: boolean;\n\n  /**\n   * If `true`, the last state for each sample at index i in a batch will be\n   * used as initial state of the sample of index i in the following batch\n   * (default: `false`).\n   *\n   * You can set RNN layers to be \"stateful\", which means that the states\n   * computed for the samples in one batch will be reused as initial states\n   * for the samples in the next batch. This assumes a one-to-one mapping\n   * between samples in different successive batches.\n   *\n   * To enable \"statefulness\":\n   *   - specify `stateful: true` in the layer constructor.\n   *   - specify a fixed batch size for your model, by passing\n   *     - if sequential model:\n   *       `batchInputShape: [...]` to the first layer in your model.\n   *     - else for functional model with 1 or more Input layers:\n   *       `batchShape: [...]` to all the first layers in your model.\n   *     This is the expected shape of your inputs\n   *     *including the batch size*.\n   *     It should be a tuple of integers, e.g., `[32, 10, 100]`.\n   *   - specify `shuffle: false` when calling `LayersModel.fit()`.\n   *\n   * To reset the state of your model, call `resetStates()` on either the\n   * specific layer or on the entire model.\n   */\n  stateful?: boolean;\n  // TODO(cais): Explore whether we can warn users when they fail to set\n  //   `shuffle: false` when training a model consisting of stateful RNNs\n  //   and any stateful Layers in general.\n\n  /**\n   * If `true`, the network will be unrolled, else a symbolic loop will be\n   * used. Unrolling can speed-up a RNN, although it tends to be more memory-\n   * intensive. Unrolling is only suitable for short sequences (default:\n   * `false`).\n   * Porting Note: tfjs-layers has an imperative backend. RNNs are executed with\n   *   normal TypeScript control flow. Hence this property is inapplicable and\n   *   ignored in tfjs-layers.\n   */\n  unroll?: boolean;\n\n  /**\n   * Dimensionality of the input (integer).\n   *   This option (or alternatively, the option `inputShape`) is required when\n   *   this layer is used as the first layer in a model.\n   */\n  inputDim?: number;\n\n  /**\n   * Length of the input sequences, to be specified when it is constant.\n   * This argument is required if you are going to connect `Flatten` then\n   * `Dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed). Note that if the recurrent layer is not the first layer in\n   * your model, you would need to specify the input length at the level of the\n   * first layer (e.g., via the `inputShape` option).\n   */\n  inputLength?: number;\n}\n\nexport class RNN extends Layer {\n  /** @nocollapse */\n  static className = 'RNN';\n  public readonly cell: RNNCell;\n  public readonly returnSequences: boolean;\n  public readonly returnState: boolean;\n  public readonly goBackwards: boolean;\n  public readonly unroll: boolean;\n\n  public stateSpec: InputSpec[];\n  protected states_: Tensor[];\n\n  // NOTE(cais): For stateful RNNs, the old states cannot be disposed right\n  // away when new states are set, because the old states may need to be used\n  // later for backpropagation through time (BPTT) and other purposes. So we\n  // keep them here for final disposal when the state is reset completely\n  // (i.e., through no-arg call to `resetStates()`).\n  protected keptStates: Tensor[][];\n\n  private numConstants: number;\n\n  constructor(args: RNNLayerArgs) {\n    super(args);\n    let cell: RNNCell;\n    if (args.cell == null) {\n      throw new ValueError(\n          'cell property is missing for the constructor of RNN.');\n    } else if (Array.isArray(args.cell)) {\n      cell = new StackedRNNCells({cells: args.cell});\n    } else {\n      cell = args.cell;\n    }\n    if (cell.stateSize == null) {\n      throw new ValueError(\n          'The RNN cell should have an attribute `stateSize` (tuple of ' +\n          'integers, one integer per RNN state).');\n    }\n    this.cell = cell;\n    this.returnSequences =\n        args.returnSequences == null ? false : args.returnSequences;\n    this.returnState = args.returnState == null ? false : args.returnState;\n    this.goBackwards = args.goBackwards == null ? false : args.goBackwards;\n    this._stateful = args.stateful == null ? false : args.stateful;\n    this.unroll = args.unroll == null ? false : args.unroll;\n\n    this.supportsMasking = true;\n    this.inputSpec = [new InputSpec({ndim: 3})];\n    this.stateSpec = null;\n    this.states_ = null;\n    // TODO(cais): Add constantsSpec and numConstants.\n    this.numConstants = null;\n    // TODO(cais): Look into the use of initial_state in the kwargs of the\n    //   constructor.\n\n    this.keptStates = [];\n  }\n\n  // Porting Note: This is the equivalent of `RNN.states` property getter in\n  //   PyKeras.\n  getStates(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      return math_utils.range(0, numStates).map(x => null);\n    } else {\n      return this.states_;\n    }\n  }\n\n  // Porting Note: This is the equivalent of the `RNN.states` property setter in\n  //   PyKeras.\n  setStates(states: Tensor[]): void {\n    this.states_ = states;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    // TODO(cais): Remove the casting once stacked RNN cells become supported.\n    let stateSize = this.cell.stateSize;\n    if (!Array.isArray(stateSize)) {\n      stateSize = [stateSize];\n    }\n    const outputDim = stateSize[0];\n    let outputShape: Shape|Shape[];\n    if (this.returnSequences) {\n      outputShape = [inputShape[0], inputShape[1], outputDim];\n    } else {\n      outputShape = [inputShape[0], outputDim];\n    }\n\n    if (this.returnState) {\n      const stateShape: Shape[] = [];\n      for (const dim of stateSize) {\n        stateShape.push([inputShape[0], dim]);\n      }\n      return [outputShape].concat(stateShape);\n    } else {\n      return outputShape;\n    }\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tfc.tidy(() => {\n      if (Array.isArray(mask)) {\n        mask = mask[0];\n      }\n      const outputMask = this.returnSequences ? mask : null;\n\n      if (this.returnState) {\n        const stateMask = this.states.map(s => null);\n        return [outputMask].concat(stateMask);\n      } else {\n        return outputMask;\n      }\n    });\n  }\n\n  /**\n   * Get the current state tensors of the RNN.\n   *\n   * If the state hasn't been set, return an array of `null`s of the correct\n   * length.\n   */\n  get states(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      const output: Tensor[] = [];\n      for (let i = 0; i < numStates; ++i) {\n        output.push(null);\n      }\n      return output;\n    } else {\n      return this.states_;\n    }\n  }\n\n  set states(s: Tensor[]) {\n    this.states_ = s;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    // Note inputShape will be an Array of Shapes of initial states and\n    // constants if these are passed in apply().\n    const constantShape: Shape[] = null;\n    if (this.numConstants != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    }\n\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    const batchSize: number = this.stateful ? inputShape[0] : null;\n    const inputDim = inputShape.slice(2);\n    this.inputSpec[0] = new InputSpec({shape: [batchSize, null, ...inputDim]});\n\n    // Allow cell (if RNNCell Layer) to build before we set or validate\n    // stateSpec.\n    const stepInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (constantShape != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    } else {\n      this.cell.build(stepInputShape);\n    }\n\n    // Set or validate stateSpec.\n    let stateSize: number[];\n    if (Array.isArray(this.cell.stateSize)) {\n      stateSize = this.cell.stateSize;\n    } else {\n      stateSize = [this.cell.stateSize];\n    }\n\n    if (this.stateSpec != null) {\n      if (!util.arraysEqual(\n              this.stateSpec.map(spec => spec.shape[spec.shape.length - 1]),\n              stateSize)) {\n        throw new ValueError(\n            `An initialState was passed that is not compatible with ` +\n            `cell.stateSize. Received stateSpec=${this.stateSpec}; ` +\n            `However cell.stateSize is ${this.cell.stateSize}`);\n      }\n    } else {\n      this.stateSpec =\n          stateSize.map(dim => new InputSpec({shape: [null, dim]}));\n    }\n    if (this.stateful) {\n      this.resetStates();\n    }\n  }\n\n  /**\n   * Reset the state tensors of the RNN.\n   *\n   * If the `states` argument is `undefined` or `null`, will set the\n   * state tensor(s) of the RNN to all-zero tensors of the appropriate\n   * shape(s).\n   *\n   * If `states` is provided, will set the state tensors of the RNN to its\n   * value.\n   *\n   * @param states Optional externally-provided initial states.\n   * @param training Whether this call is done during training. For stateful\n   *   RNNs, this affects whether the old states are kept or discarded. In\n   *   particular, if `training` is `true`, the old states will be kept so\n   *   that subsequent backpropgataion through time (BPTT) may work properly.\n   *   Else, the old states will be discarded.\n   */\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n      const batchSize = this.inputSpec[0].shape[0];\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n      // Initialize state if null.\n      if (this.states_ == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_ = [tfc.zeros([batchSize, this.cell.stateSize])];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_[0] = tfc.zeros([batchSize, this.cell.stateSize]);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training === true) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n          const dim = Array.isArray(this.cell.stateSize) ?\n              this.cell.stateSize[index] :\n              this.cell.stateSize;\n          const expectedShape = [batchSize, dim];\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n          this.states_[index] = value;\n        }\n      }\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    // TODO(cais): Figure out whether initialState is in kwargs or inputs.\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    // If any of `initial_state` or `constants` are specified and are\n    // `tf.SymbolicTensor`s, then add them to the inputs and temporarily modify\n    // the input_spec to include them.\n\n    let additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    let additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      kwargs['initialState'] = initialState;\n      additionalInputs = additionalInputs.concat(initialState);\n      this.stateSpec = [];\n      for (const state of initialState) {\n        this.stateSpec.push(new InputSpec({shape: state.shape}));\n      }\n      // TODO(cais): Use the following instead.\n      // this.stateSpec = initialState.map(state => new InputSpec({shape:\n      // state.shape}));\n      additionalSpecs = additionalSpecs.concat(this.stateSpec);\n    }\n    if (constants != null) {\n      kwargs['constants'] = constants;\n      additionalInputs = additionalInputs.concat(constants);\n      // TODO(cais): Add this.constantsSpec.\n      this.numConstants = constants.length;\n    }\n\n    const isTensor = additionalInputs[0] instanceof SymbolicTensor;\n    if (isTensor) {\n      // Compute full input spec, including state and constants.\n      const fullInput =\n          [inputs].concat(additionalInputs) as Tensor[] | SymbolicTensor[];\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call with temporarily replaced inputSpec.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output = super.apply(fullInput, kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  // tslint:disable-next-line:no-any\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    // Input shape: `[samples, time (padded with zeros), input_dim]`.\n    // Note that the .build() method of subclasses **must** define\n    // this.inputSpec and this.stateSpec owith complete input shapes.\n    return tidy(() => {\n      const mask = kwargs == null ? null : kwargs['mask'] as Tensor;\n      const training = kwargs == null ? null : kwargs['training'];\n      let initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      inputs = getExactlyOneTensor(inputs);\n      if (initialState == null) {\n        if (this.stateful) {\n          initialState = this.states_;\n        } else {\n          initialState = this.getInitialState(inputs);\n        }\n      }\n\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      if (initialState.length !== numStates) {\n        throw new ValueError(\n            `RNN Layer has ${numStates} state(s) but was passed ` +\n            `${initialState.length} initial state(s).`);\n      }\n      if (this.unroll) {\n        console.warn(\n            'Ignoring unroll = true for RNN layer, due to imperative backend.');\n      }\n\n      const cellCallKwargs: Kwargs = {training};\n\n      // TODO(cais): Add support for constants.\n      const step = (inputs: Tensor, states: Tensor[]) => {\n        // `inputs` and `states` are concatenated to form a single `Array` of\n        // `tf.Tensor`s as the input to `cell.call()`.\n        const outputs =\n            this.cell.call([inputs].concat(states), cellCallKwargs) as Tensor[];\n        // Marshall the return value into output and new states.\n        return [outputs[0], outputs.slice(1)] as [Tensor, Tensor[]];\n      };\n\n      // TODO(cais): Add support for constants.\n\n      const rnnOutputs =\n          rnn(step, inputs, initialState, this.goBackwards, mask, null,\n              this.unroll, this.returnSequences);\n      const lastOutput = rnnOutputs[0];\n      const outputs = rnnOutputs[1];\n      const states = rnnOutputs[2];\n\n      if (this.stateful) {\n        this.resetStates(states, training);\n      }\n\n      const output = this.returnSequences ? outputs : lastOutput;\n\n      // TODO(cais): Porperty set learning phase flag.\n\n      if (this.returnState) {\n        return [output].concat(states);\n      } else {\n        return output;\n      }\n    });\n  }\n\n  getInitialState(inputs: Tensor): Tensor[] {\n    return tidy(() => {\n      // Build an all-zero tensor of shape [samples, outputDim].\n      // [Samples, timeSteps, inputDim].\n      let initialState = tfc.zeros(inputs.shape);\n      // [Samples].\n      initialState = tfc.sum(initialState, [1, 2]);\n      initialState = K.expandDims(initialState);  // [Samples, 1].\n\n      if (Array.isArray(this.cell.stateSize)) {\n        return this.cell.stateSize.map(\n            dim => dim > 1 ? K.tile(initialState, [1, dim]) : initialState);\n      } else {\n        return this.cell.stateSize > 1 ?\n            [K.tile(initialState, [1, this.cell.stateSize])] :\n            [initialState];\n      }\n    });\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    return this.cell.trainableWeights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    if (!this.trainable) {\n      return this.cell.weights;\n    }\n    return this.cell.nonTrainableWeights;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.cell != null) {\n      this.cell.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      returnSequences: this.returnSequences,\n      returnState: this.returnState,\n      goBackwards: this.goBackwards,\n      stateful: this.stateful,\n      unroll: this.unroll,\n    };\n\n    if (this.numConstants != null) {\n      config['numConstants'] = this.numConstants;\n    }\n\n    const cellConfig = this.cell.getConfig();\n\n    if (this.getClassName() === RNN.className) {\n      config['cell'] = {\n        'className': this.cell.getClassName(),\n        'config': cellConfig,\n      } as serialization.ConfigDictValue;\n    }\n\n    // this order is necessary, to prevent cell name from replacing layer name\n    return {...cellConfig, ...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cellConfig = config['cell'] as serialization.ConfigDict;\n    const cell = deserialize(cellConfig, customObjects) as RNNCell;\n    return new cls(Object.assign(config, {cell}));\n  }\n}\nserialization.registerClass(RNN);\n\n// Porting Note: This is a common parent class for RNN cells. There is no\n// equivalent of this in PyKeras. Having a common parent class forgoes the\n//  need for `has_attr(cell, ...)` checks or its TypeScript equivalent.\n/**\n * An RNNCell layer.\n *\n * @doc {heading: 'Layers', subheading: 'Classes'}\n */\nexport abstract class RNNCell extends Layer {\n  /**\n   * Size(s) of the states.\n   * For RNN cells with only a single state, this is a single integer.\n   */\n  public stateSize: number|number[];\n  public dropoutMask: Tensor|Tensor[];\n  public recurrentDropoutMask: Tensor|Tensor[];\n}\n\nexport declare interface SimpleRNNCellLayerArgs extends LayerArgs {\n  /**\n   * units: Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   * Default: hyperbolic tangent ('tanh').\n   * If you pass `null`,  'linear' activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the `kernel` weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the `recurrent_kernel` weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the `kernel` weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the `recurrentKernel` weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraintfunction applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\nexport class SimpleRNNCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'SimpleRNNCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args: SimpleRNNCellLayerArgs) {\n    super(args);\n    this.units = args.units;\n    assertPositiveInteger(this.units, `units`);\n    this.activation = getActivation(\n        args.activation == null ? this.DEFAULT_ACTIVATION : args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    // TODO(cais): Use regularizer.\n    this.kernel = this.addWeight(\n        'kernel', [inputShape[inputShape.length - 1], this.units], null,\n        this.kernelInitializer, this.kernelRegularizer, true,\n        this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  // Porting Note: PyKeras' equivalent of this method takes two tensor inputs:\n  //   `inputs` and `states`. Here, the two tensors are combined into an\n  //   `Tensor[]` Array as the first input argument.\n  //   Similarly, PyKeras' equivalent of this method returns two values:\n  //    `output` and `[output]`. Here the two are combined into one length-2\n  //    `Tensor[]`, consisting of `output` repeated.\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `SimpleRNNCell expects 2 input Tensors, got ${inputs.length}.`);\n      }\n      let prevOutput = inputs[1];\n      inputs = inputs[0];\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training\n                           }) as Tensor;\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(prevOutput),\n                                      rate: this.recurrentDropout,\n                                      training\n                                    }) as Tensor;\n      }\n      let h: Tensor;\n      const dpMask: Tensor = this.dropoutMask as Tensor;\n      const recDpMask: Tensor = this.recurrentDropoutMask as Tensor;\n      if (dpMask != null) {\n        h = K.dot(tfc.mul(inputs, dpMask), this.kernel.read());\n      } else {\n        h = K.dot(inputs, this.kernel.read());\n      }\n      if (this.bias != null) {\n        h = K.biasAdd(h, this.bias.read());\n      }\n      if (recDpMask != null) {\n        prevOutput = tfc.mul(prevOutput, recDpMask);\n      }\n      let output = tfc.add(h, K.dot(prevOutput, this.recurrentKernel.read()));\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      // TODO(cais): Properly set learning phase on output tensor?\n      return [output, output];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(SimpleRNNCell);\n\nexport declare interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {\n  /**\n   * Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   *\n   * Defaults to  hyperbolic tangent (`tanh`)\n   *\n   * If you pass `null`, no activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the recurrentKernel weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the kernel weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the recurrentKernel weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\n/**\n * RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the\n * `cell` property required. This interface is to be used with constructors\n * of concrete RNN layer subtypes.\n */\nexport declare interface RNNLayerArgs extends BaseRNNLayerArgs {\n  cell: RNNCell|RNNCell[];\n}\n\nexport class SimpleRNN extends RNN {\n  /** @nocollapse */\n  static className = 'SimpleRNN';\n  constructor(args: SimpleRNNLayerArgs) {\n    args.cell = new SimpleRNNCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\nserialization.registerClass(SimpleRNN);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n\n  /**\n   * GRU convention (whether to apply reset gate after or before matrix\n   * multiplication). false = \"before\", true = \"after\" (only false is\n   * supported).\n   */\n  resetAfter?: boolean;\n}\n\nexport class GRUCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'GRUCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION: ActivationIdentifier = 'hardSigmoid';\n\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: GRUCellLayerArgs) {\n    super(args);\n    if (args.resetAfter) {\n      throw new ValueError(\n          `GRUCell does not support reset_after parameter set to true.`);\n    }\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 3], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 3], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units * 3], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `GRUCell expects 2 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      let hTMinus1 = inputs[1];  // Previous memory state.\n      inputs = inputs[0];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2, regardless of the actual value of\n      // config.implementation.\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 3\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 3\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor];\n      const recDpMask = this.recurrentDropoutMask as [Tensor, Tensor, Tensor];\n      let z: Tensor;\n      let r: Tensor;\n      let hh: Tensor;\n\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let matrixX = K.dot(inputs, this.kernel.read());\n      if (this.useBias) {\n        matrixX = K.biasAdd(matrixX, this.bias.read());\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n\n      const recurrentKernelValue = this.recurrentKernel.read();\n      const [rk1, rk2] = tfc.split(\n          recurrentKernelValue, [2 * this.units, this.units],\n          recurrentKernelValue.rank - 1);\n      const matrixInner = K.dot(hTMinus1, rk1);\n\n      const [xZ, xR, xH] = tfc.split(matrixX, 3, matrixX.rank - 1);\n      const [recurrentZ, recurrentR] =\n          tfc.split(matrixInner, 2, matrixInner.rank - 1);\n      z = this.recurrentActivation.apply(tfc.add(xZ, recurrentZ));\n      r = this.recurrentActivation.apply(tfc.add(xR, recurrentR));\n\n      const recurrentH = K.dot(tfc.mul(r, hTMinus1), rk2);\n      hh = this.activation.apply(tfc.add(xH, recurrentH));\n\n      const h =\n          tfc.add(tfc.mul(z, hTMinus1), tfc.mul(tfc.add(1, tfc.neg(z)), hh));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n      resetAfter: false\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(GRUCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface GRULayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   * smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class GRU extends RNN {\n  /** @nocollapse */\n  static className = 'GRU';\n  constructor(args: GRULayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new GRUCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(GRU);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class LSTMCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'LSTMCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n  readonly unitForgetBias: boolean;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number[];\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION = 'hardSigmoid';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n\n  readonly DEFAULT_BIAS_INITIALIZER = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: LSTMCellLayerArgs) {\n    super(args);\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.unitForgetBias = args.unitForgetBias;\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = [this.units, this.units];\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 4], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 4], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    let biasInitializer: Initializer;\n    if (this.useBias) {\n      if (this.unitForgetBias) {\n        const capturedBiasInit = this.biasInitializer;\n        const capturedUnits = this.units;\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): Tensor {\n            // TODO(cais): More informative variable names?\n            const bI = capturedBiasInit.apply([capturedUnits]);\n            const bF = (new Ones()).apply([capturedUnits]);\n            const bCAndH = capturedBiasInit.apply([capturedUnits * 2]);\n            return K.concatAlongFirstAxis(\n                K.concatAlongFirstAxis(bI, bF), bCAndH);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n      this.bias = this.addWeight(\n          'bias', [this.units * 4], null, biasInitializer, this.biasRegularizer,\n          true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `LSTMCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n      let hTMinus1 = inputs[1];    // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n      inputs = inputs[0];\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 4\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 4\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor, Tensor];\n      const recDpMask =\n          this.recurrentDropoutMask as [Tensor, Tensor, Tensor, Tensor];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2 regardless of the actual value of\n      // config.implementation.\n      let i: Tensor;\n      let f: Tensor;\n      let c: Tensor;\n      let o: Tensor;\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let z = K.dot(inputs, this.kernel.read());\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n      z = tfc.add(z, K.dot(hTMinus1, this.recurrentKernel.read()));\n      if (this.useBias) {\n        z = K.biasAdd(z, this.bias.read());\n      }\n\n      const [z0, z1, z2, z3] = tfc.split(z, 4, z.rank - 1);\n\n      i = this.recurrentActivation.apply(z0);\n      f = this.recurrentActivation.apply(z1);\n      c = tfc.add(tfc.mul(f, cTMinus1), tfc.mul(i, this.activation.apply(z2)));\n      o = this.recurrentActivation.apply(z3);\n\n      const h = tfc.mul(o, this.activation.apply(c));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      unitForgetBias: this.unitForgetBias,\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(LSTMCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface LSTMLayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *   Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions, whereas mode 2 will\n   *   batch them into fewer, larger operations. These modes will\n   *   have different performance profiles on different hardware and\n   *   for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this config field.\n   */\n  implementation?: number;\n}\n\nexport class LSTM extends RNN {\n  /** @nocollapse */\n  static className = 'LSTM';\n  constructor(args: LSTMLayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new LSTMCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(LSTM);\n\nexport declare interface StackedRNNCellsArgs extends LayerArgs {\n  /**\n   * A `Array` of `RNNCell` instances.\n   */\n  cells: RNNCell[];\n}\n\nexport class StackedRNNCells extends RNNCell {\n  /** @nocollapse */\n  static className = 'StackedRNNCells';\n  protected cells: RNNCell[];\n\n  constructor(args: StackedRNNCellsArgs) {\n    super(args);\n    this.cells = args.cells;\n  }\n\n  get stateSize(): number[] {\n    // States are a flat list in reverse order of the cell stack.\n    // This allows perserving the requirement `stack.statesize[0] ===\n    // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,\n    // assuming one LSTM has states `[h, c]`.\n    const stateSize: number[] = [];\n    for (const cell of this.cells.slice().reverse()) {\n      if (Array.isArray(cell.stateSize)) {\n        stateSize.push(...cell.stateSize);\n      } else {\n        stateSize.push(cell.stateSize);\n      }\n    }\n    return stateSize;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      let states = inputs.slice(1);\n\n      // Recover per-cell states.\n      const nestedStates: Tensor[][] = [];\n      for (const cell of this.cells.slice().reverse()) {\n        if (Array.isArray(cell.stateSize)) {\n          nestedStates.push(states.splice(0, cell.stateSize.length));\n        } else {\n          nestedStates.push(states.splice(0, 1));\n        }\n      }\n      nestedStates.reverse();\n\n      // Call the cells in order and store the returned states.\n      const newNestedStates: Tensor[][] = [];\n      let callInputs: Tensor[];\n      for (let i = 0; i < this.cells.length; ++i) {\n        const cell = this.cells[i];\n        states = nestedStates[i];\n        // TODO(cais): Take care of constants.\n        if (i === 0) {\n          callInputs = [inputs[0]].concat(states);\n        } else {\n          callInputs = [callInputs[0]].concat(states);\n        }\n        callInputs = cell.call(callInputs, kwargs) as Tensor[];\n        newNestedStates.push(callInputs.slice(1));\n      }\n\n      // Format the new states as a flat list in reverse cell order.\n      states = [];\n      for (const cellStates of newNestedStates.slice().reverse()) {\n        states.push(...cellStates);\n      }\n      return [callInputs[0]].concat(states);\n    });\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    if (isArrayOfShapes(inputShape)) {\n      // TODO(cais): Take care of input constants.\n      // const constantShape = inputShape.slice(1);\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n    let outputDim: number;\n    this.cells.forEach((cell, i) => {\n      nameScope(`RNNCell_${i}`, () => {\n        // TODO(cais): Take care of input constants.\n\n        cell.build(inputShape);\n        if (Array.isArray(cell.stateSize)) {\n          outputDim = cell.stateSize[0];\n        } else {\n          outputDim = cell.stateSize;\n        }\n        inputShape = [inputShape[0], outputDim] as Shape;\n      });\n    });\n    this.built = true;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const getCellConfig = (cell: RNNCell) => {\n      return {\n        'className': cell.getClassName(),\n        'config': cell.getConfig(),\n      };\n    };\n\n    const cellConfigs = this.cells.map(getCellConfig);\n\n    const config = {'cells': cellConfigs};\n\n    return {...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cells: RNNCell[] = [];\n    for (const cellConfig of (config['cells'] as serialization.ConfigDict[])) {\n      cells.push(deserialize(cellConfig, customObjects) as RNNCell);\n    }\n    return new cls({cells});\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const cell of this.cells) {\n        trainableWeights.push(...cell.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  /**\n   * Retrieve the weights of a the model.\n   *\n   * @returns A flat `Array` of `tf.Tensor`s.\n   */\n  getWeights(): Tensor[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.weights);\n    }\n    return batchGetValue(weights);\n  }\n\n  /**\n   * Set the weights of the model.\n   *\n   * @param weights An `Array` of `tf.Tensor`s with shapes and types matching\n   *     the output of `getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    const tuples: Array<[LayerVariable, Tensor]> = [];\n    for (const cell of this.cells) {\n      const numParams = cell.weights.length;\n      const inputWeights = weights.splice(numParams);\n      for (let i = 0; i < cell.weights.length; ++i) {\n        tuples.push([cell.weights[i], inputWeights[i]]);\n      }\n    }\n    batchSetValue(tuples);\n  }\n\n  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.\n}\nserialization.registerClass(StackedRNNCells);\n\nexport function generateDropoutMask(args: {\n  ones: () => tfc.Tensor,\n  rate: number,\n  training?: boolean,\n  count?: number,\n}): tfc.Tensor|tfc.Tensor[] {\n  const {ones, rate, training = false, count = 1} = args;\n\n  const droppedInputs = () => K.dropout(ones(), rate);\n\n  const createMask = () => K.inTrainPhase(droppedInputs, ones, training);\n\n  // just in case count is provided with null or undefined\n  if (!count || count <= 1) {\n    return tfc.keep(createMask().clone());\n  }\n\n  const masks = Array(count).fill(undefined).map(createMask);\n\n  return masks.map(m => tfc.keep(m.clone()));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {Activation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint} from '../constraints';\nimport {InputSpec} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {Initializer} from '../initializers';\nimport {DataFormat, DataType, PaddingMode, Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, normalizeArray} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\nimport {BaseRNNLayerArgs, generateDropoutMask, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNNCellLayerArgs} from './recurrent';\n\ndeclare interface ConvRNN2DCellArgs extends\n    Omit<SimpleRNNCellLayerArgs, 'units'> {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number];\n}\n\nabstract class ConvRNN2DCell extends RNNCell {\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n}\n\ndeclare interface ConvRNN2DLayerArgs extends BaseRNNLayerArgs,\n                                             ConvRNN2DCellArgs {}\n\n/**\n * Base class for convolutional-recurrent layers.\n */\nclass ConvRNN2D extends RNN {\n  /** @nocollapse */\n  static className = 'ConvRNN2D';\n\n  readonly cell: ConvRNN2DCell;\n\n  constructor(args: ConvRNN2DLayerArgs) {\n    if (args.unroll) {\n      throw new NotImplementedError(\n          'Unrolling is not possible with convolutional RNNs.');\n    }\n\n    if (Array.isArray(args.cell)) {\n      throw new NotImplementedError(\n          'It is not possible at the moment to stack convolutional cells.');\n    }\n\n    super(args as RNNLayerArgs);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n\n        this.cell.dropoutMask = null;\n      }\n\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n\n        this.cell.recurrentDropoutMask = null;\n      }\n\n      if (kwargs && kwargs['constants']) {\n        throw new ValueError('ConvRNN2D cell does not support constants');\n      }\n\n      const mask = kwargs == null ? null : kwargs['mask'];\n\n      const training = kwargs == null ? null : kwargs['training'];\n\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  computeOutputShape(inputShape: Shape): Shape|Shape[] {\n    let outShape: Shape = this.computeSingleOutputShape(inputShape);\n\n    if (!this.returnSequences) {\n      outShape = [outShape[0], ...outShape.slice(2)];\n    }\n\n    if (this.returnState) {\n      outShape =\n          [outShape, ...Array(2).fill([inputShape[0], ...outShape.slice(-3)])];\n    }\n\n    return outShape;\n  }\n\n  getInitialState(inputs: tfc.Tensor): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      const {stateSize} = this.cell;\n\n      const inputShape = inputs.shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const initialState = tfc.zeros(stateShape);\n\n      if (Array.isArray(stateSize)) {\n        return Array(stateSize.length).fill(initialState);\n      }\n\n      return [initialState];\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tfc.tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n\n      const inputShape = this.inputSpec[0].shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const batchSize = inputShape[0];\n\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n\n      // Initialize state if null.\n      if (this.getStates() == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_ = [tfc.zeros(stateShape)];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_[0] = tfc.zeros(stateShape);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n\n          const expectedShape = stateShape;\n\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n\n          this.states_[index] = value;\n        }\n      }\n\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  protected computeSingleOutputShape(inputShape: Shape): Shape {\n    const {dataFormat, filters, kernelSize, padding, strides, dilationRate} =\n        this.cell;\n\n    const isChannelsFirst = dataFormat === 'channelsFirst';\n\n    const h = inputShape[isChannelsFirst ? 3 : 2];\n    const w = inputShape[isChannelsFirst ? 4 : 3];\n\n    const hOut = convOutputLength(\n        h, kernelSize[0], padding, strides[0], dilationRate[0]);\n    const wOut = convOutputLength(\n        w, kernelSize[1], padding, strides[1], dilationRate[1]);\n\n    const outShape: Shape = [\n      ...inputShape.slice(0, 2),\n      ...(isChannelsFirst ? [filters, hOut, wOut] : [hOut, wOut, filters])\n    ];\n\n    return outShape;\n  }\n}\n\nexport declare interface ConvLSTM2DCellArgs extends\n    Omit<LSTMCellLayerArgs, 'units'>, ConvRNN2DCellArgs {}\n\nexport class ConvLSTM2DCell extends LSTMCell implements ConvRNN2DCell {\n  /** @nocollapse */\n  static className = 'ConvLSTM2DCell';\n\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  constructor(args: ConvLSTM2DCellArgs) {\n    const {\n      filters,\n      kernelSize,\n      strides,\n      padding,\n      dataFormat,\n      dilationRate,\n    } = args;\n\n    super({...args, units: filters});\n\n    this.filters = filters;\n    assertPositiveInteger(this.filters, 'filters');\n\n    this.kernelSize = normalizeArray(kernelSize, 2, 'kernelSize');\n    this.kernelSize.forEach(size => assertPositiveInteger(size, 'kernelSize'));\n\n    this.strides = normalizeArray(strides || 1, 2, 'strides');\n    this.strides.forEach(stride => assertPositiveInteger(stride, 'strides'));\n\n    this.padding = padding || 'valid';\n    checkPaddingMode(this.padding);\n\n    this.dataFormat = dataFormat || 'channelsLast';\n    checkDataFormat(this.dataFormat);\n\n    this.dilationRate = normalizeArray(dilationRate || 1, 2, 'dilationRate');\n    this.dilationRate.forEach(\n        rate => assertPositiveInteger(rate, 'dilationRate'));\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n\n    const numOfKernels = 4;\n\n    const kernelShape =\n        this.kernelSize.concat([inputDim, this.filters * numOfKernels]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n\n    const recurrentKernelShape =\n        this.kernelSize.concat([this.filters, this.filters * numOfKernels]);\n\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', recurrentKernelShape, null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n\n    if (this.useBias) {\n      let biasInitializer: Initializer;\n\n      if (this.unitForgetBias) {\n        const init = this.biasInitializer;\n\n        const filters = this.filters;\n\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): tfc.Tensor {\n            const biasI = init.apply([filters]);\n            const biasF = tfc.ones([filters]);\n            const biasCAndO = init.apply([filters * 2]);\n            return K.concatenate([biasI, biasF, biasCAndO]);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n\n      this.bias = this.addWeight(\n          'bias', [this.filters * numOfKernels], null, biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: tfc.Tensor[], kwargs: Kwargs): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] || false;\n\n      const x = inputs[0];         // Current input\n      const hTMinus1 = inputs[1];  // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n\n      const numOfKernels = 4;\n\n      type DropoutMasks = [tfc.Tensor, tfc.Tensor, tfc.Tensor, tfc.Tensor];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(x),\n                             rate: this.dropout,\n                             training,\n                             count: numOfKernels\n                           }) as tfc.Tensor[];\n      }\n\n      const dropoutMask = this.dropoutMask as DropoutMasks;\n\n      const applyDropout =\n          (x: tfc.Tensor, mask: tfc.Tensor[], index: number) => {\n            if (!mask || !mask[index]) {\n              return x;\n            }\n\n            return tfc.mul(mask[index], x);\n          };\n\n      let xI = applyDropout(x, dropoutMask, 0);\n      let xF = applyDropout(x, dropoutMask, 1);\n      let xC = applyDropout(x, dropoutMask, 2);\n      let xO = applyDropout(x, dropoutMask, 3);\n\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: numOfKernels\n                                    }) as tfc.Tensor[];\n      }\n\n      const recDropoutMask = this.recurrentDropoutMask as DropoutMasks;\n\n      let hI = applyDropout(hTMinus1, recDropoutMask, 0);\n      let hF = applyDropout(hTMinus1, recDropoutMask, 1);\n      let hC = applyDropout(hTMinus1, recDropoutMask, 2);\n      let hO = applyDropout(hTMinus1, recDropoutMask, 3);\n\n      const kernelChannelAxis = 3;\n\n      const [kernelI, kernelF, kernelC, kernelO]: tfc.Tensor[] =\n          tfc.split(this.kernel.read(), numOfKernels, kernelChannelAxis);\n\n      const [biasI, biasF, biasC, biasO]: tfc.Tensor[] = this.useBias ?\n          tfc.split(this.bias.read(), numOfKernels) :\n          [null, null, null, null];\n\n      xI = this.inputConv(xI, kernelI, biasI, this.padding);\n      xF = this.inputConv(xF, kernelF, biasF, this.padding);\n      xC = this.inputConv(xC, kernelC, biasC, this.padding);\n      xO = this.inputConv(xO, kernelO, biasO, this.padding);\n\n      const [recKernelI, recKernelF, recKernelC, recKernelO]: tfc.Tensor[] =\n          tfc.split(\n              this.recurrentKernel.read(), numOfKernels, kernelChannelAxis);\n\n      hI = this.recurrentConv(hI, recKernelI);\n      hF = this.recurrentConv(hF, recKernelF);\n      hC = this.recurrentConv(hC, recKernelC);\n      hO = this.recurrentConv(hO, recKernelO);\n\n      const i = this.recurrentActivation.apply(tfc.add(xI, hI));\n      const f = this.recurrentActivation.apply(tfc.add(xF, hF));\n      const c = tfc.add(\n          tfc.mul(f, cTMinus1),\n          tfc.mul(i, this.activation.apply(tfc.add(xC, hC))));\n      const h = tfc.mul(\n          this.recurrentActivation.apply(tfc.add(xO, hO)),\n          this.activation.apply(c));\n\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): tfc.serialization.ConfigDict {\n    const {'units': _, ...baseConfig} = super.getConfig();\n\n    const config: tfc.serialization.ConfigDict = {\n      filters: this.filters,\n      kernelSize: this.kernelSize,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      strides: this.strides,\n    };\n\n    return {...baseConfig, ...config};\n  }\n\n  inputConv(x: Tensor, w: Tensor, b?: Tensor, padding?: PaddingMode) {\n    const out = tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, this.strides as [number, number],\n        (padding || 'valid') as 'same' | 'valid',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC',\n        this.dilationRate as [number, number]);\n\n    if (b) {\n      return K.biasAdd(out, b, this.dataFormat) as tfc.Tensor3D;\n    }\n\n    return out;\n  }\n\n  recurrentConv(x: Tensor, w: Tensor) {\n    const strides = 1;\n\n    return tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, strides, 'same',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC');\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2DCell);\n\nexport declare interface ConvLSTM2DArgs extends\n    Omit<LSTMLayerArgs, 'units'|'cell'>, ConvRNN2DLayerArgs {}\n\nexport class ConvLSTM2D extends ConvRNN2D {\n  /** @nocollapse */\n  static className = 'ConvLSTM2D';\n\n  constructor(args: ConvLSTM2DArgs) {\n    const cell = new ConvLSTM2DCell(args);\n\n    super({...args, cell} as ConvRNN2DLayerArgs);\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends tfc.serialization.Serializable>(\n      cls: tfc.serialization.SerializableConstructor<T>,\n      config: tfc.serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Basic Layers.\n */\n\nimport {any, notEqual, serialization, Tensor, tidy, transpose, util} from '@tensorflow/tfjs-core';\n\nimport {Activation as ActivationFn, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {DisposeResult, InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {LayerConfig} from '../keras_format/topology_config';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {assertPositiveInteger, mapActivationToFusedKernel} from '../utils/generic_utils';\nimport {arrayProd, range} from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface DropoutLayerArgs extends LayerArgs {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /**\n   * Integer array representing the shape of the binary dropout mask that will\n   * be multiplied with the input.\n   *\n   * For instance, if your inputs have shape `(batchSize, timesteps, features)`\n   * and you want the dropout mask to be the same for all timesteps, you can use\n   * `noise_shape=(batch_size, 1, features)`.\n   */\n  noiseShape?: number[];\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class Dropout extends Layer {\n  /** @nocollapse */\n  static className = 'Dropout';\n  private readonly rate: number;\n  private readonly noiseShape: number[];\n  private readonly seed: number;\n\n  constructor(args: DropoutLayerArgs) {\n    super(args);\n    this.rate = Math.max(Math.min(args.rate, 1), 0);\n    // So that the scalar doesn't get tidied up between executions.\n    this.noiseShape = args.noiseShape;\n    this.seed = args.seed;\n    this.supportsMasking = true;\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    if (this.noiseShape == null) {\n      return this.noiseShape;\n    }\n    const inputShape = input.shape;\n    const noiseShape: Shape = [];\n    for (let i = 0; i < this.noiseShape.length; ++i) {\n      noiseShape.push(\n          this.noiseShape[i] == null ? inputShape[i] : this.noiseShape[i]);\n    }\n    return noiseShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      if (0 < this.rate && this.rate < 1) {\n        const training =\n            kwargs['training'] == null ? false : kwargs['training'];\n        const noiseShape = this.getNoiseShape(input);\n        const output = K.inTrainPhase(\n            () => K.dropout(input, this.rate, noiseShape, this.seed),\n            () => input, training);\n        return output;\n      }\n      return inputs;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      rate: this.rate,\n      noiseShape: this.noiseShape,\n      seed: this.seed,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  dispose(): DisposeResult {\n    return super.dispose();\n  }\n}\nserialization.registerClass(Dropout);\n\nexport declare interface DenseLayerArgs extends LayerArgs {\n  /** Positive integer, dimensionality of the output space. */\n  units: number;\n  /**\n   * Activation function to use.\n   *\n   * If unspecified, no activation is applied.\n   */\n  activation?: ActivationIdentifier;\n  /** Whether to apply a bias. */\n  useBias?: boolean;\n  /**\n   * Initializer for the dense kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * If specified, defines inputShape as `[inputDim]`.\n   */\n  inputDim?: number;\n\n  /**\n   * Constraint for the kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the dense kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport interface SpatialDropout1DLayerConfig extends LayerConfig {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class SpatialDropout1D extends Dropout {\n  /** @nocollapse */\n  static className = 'SpatialDropout1D';\n\n  constructor(args: SpatialDropout1DLayerConfig) {\n    super(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    const inputShape = input.shape;\n    return [inputShape[0], 1, inputShape[2]];\n  }\n}\nserialization.registerClass(SpatialDropout1D);\n\nexport class Dense extends Layer {\n  /** @nocollapse */\n  static className = 'Dense';\n  private units: number;\n  // Default activation: Linear (none).\n  private activation: ActivationFn = null;\n  private useBias = true;\n  private kernelInitializer: Initializer;\n  private biasInitializer: Initializer;\n  private kernel: LayerVariable = null;\n  private bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n  private readonly kernelConstraint?: Constraint;\n  private readonly biasConstraint?: Constraint;\n  private readonly kernelRegularizer?: Regularizer;\n  private readonly biasRegularizer?: Regularizer;\n\n  constructor(args: DenseLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null &&\n        args.inputDim != null) {\n      // This logic is copied from Layer's constructor, since we can't\n      // do exactly what the Python constructor does for Dense().\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      this.batchInputShape = [batchSize, args.inputDim];\n    }\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(args.activation);\n    if (args.useBias != null) {\n      this.useBias = args.useBias;\n    }\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.supportsMasking = true;\n\n    this.inputSpec = [{minNDim: 2}];\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputLastDim = inputShape[inputShape.length - 1];\n    if (this.kernel == null) {\n      this.kernel = this.addWeight(\n          'kernel', [inputLastDim, this.units], null, this.kernelInitializer,\n          this.kernelRegularizer, true, this.kernelConstraint);\n      if (this.useBias) {\n        this.bias = this.addWeight(\n            'bias', [this.units], null, this.biasInitializer,\n            this.biasRegularizer, true, this.biasConstraint);\n      }\n    }\n\n    this.inputSpec = [{minNDim: 2, axes: {[-1]: inputLastDim}}];\n    this.built = true;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    outputShape[outputShape.length - 1] = this.units;\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Dense layer accepts only a single input.\n      const input = getExactlyOneTensor(inputs);\n      const fusedActivationName =\n          mapActivationToFusedKernel(this.activation.getClassName());\n      let output: Tensor;\n\n      if (fusedActivationName != null) {\n        output = K.dot(\n            input, this.kernel.read(), fusedActivationName,\n            this.bias ? this.bias.read() : null);\n      } else {\n        output = K.dot(input, this.kernel.read());\n        if (this.bias != null) {\n          output = K.biasAdd(output, this.bias.read());\n        }\n        if (this.activation != null) {\n          output = this.activation.apply(output);\n        }\n      }\n\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dense);\n\nexport declare interface FlattenLayerArgs extends LayerArgs {\n  /** Image data format: channeLast (default) or channelFirst. */\n  dataFormat?: DataFormat;\n}\n\nexport class Flatten extends Layer {\n  private dataFormat: DataFormat;\n\n  /** @nocollapse */\n  static className = 'Flatten';\n  constructor(args?: FlattenLayerArgs) {\n    args = args || {};\n    super(args);\n    this.inputSpec = [{minNDim: 3}];\n    this.dataFormat = args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    for (const dim of inputShape.slice(1)) {\n      if (dim == null) {\n        throw new ValueError(\n            `The shape of the input to \"Flatten\" is not fully defined ` +\n            `(got ${inputShape.slice(1)}). Make sure to pass a complete ` +\n            `\"input_shape\" or \"batch_input_shape\" argument to the first ` +\n            `layer in your model.`);\n      }\n    }\n    return [inputShape[0], arrayProd(inputShape, 1)];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n\n      let input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsFirst' && input.rank > 1) {\n        const permutation: number[] = [0];\n        for (let i = 2; i < input.rank; ++i) {\n          permutation.push(i);\n        }\n        permutation.push(1);\n        input = input.transpose(permutation);\n      }\n\n      return K.batchFlatten(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {};\n    if (this.dataFormat != null) {\n      config['dataFormat'] = this.dataFormat;\n    }\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Flatten);\n\nexport declare interface ActivationLayerArgs extends LayerArgs {\n  /**\n   * Name of the activation function to use.\n   */\n  activation: ActivationIdentifier;\n}\n\nexport class Activation extends Layer {\n  /** @nocollapse */\n  static className = 'Activation';\n  activation: ActivationFn;\n\n  constructor(args: ActivationLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n    this.activation = getActivation(args.activation);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      return this.activation.apply(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {activation: serializeActivation(this.activation)};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Activation);\n\nexport declare interface ReshapeLayerArgs extends LayerArgs {\n  /** The target shape. Does not include the batch axis. */\n  targetShape: Shape;\n}\n\nexport declare interface RepeatVectorLayerArgs extends LayerArgs {\n  /**\n   * The integer number of times to repeat the input.\n   */\n  n: number;\n}\n\nexport class RepeatVector extends Layer {\n  /** @nocollapse */\n  static className = 'RepeatVector';\n  readonly n: number;\n\n  constructor(args: RepeatVectorLayerArgs) {\n    super(args);\n    this.n = args.n;\n    this.inputSpec = [{ndim: 2}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], this.n, inputShape[1]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      return K.repeat(inputs, this.n);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      n: this.n,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(RepeatVector);\n\nexport class Reshape extends Layer {\n  /** @nocollapse */\n  static className = 'Reshape';\n  private targetShape: Shape;\n\n  constructor(args: ReshapeLayerArgs) {\n    super(args);\n    this.targetShape = args.targetShape;\n\n    // Make sure that all unknown dimensions are represented as `null`.\n    for (let i = 0; i < this.targetShape.length; ++i) {\n      if (this.isUnknown(this.targetShape[i])) {\n        this.targetShape[i] = null;\n      }\n    }\n  }\n\n  private isUnknown(dim: number): boolean {\n    return dim < 0 || dim == null;\n  }\n\n  /**\n   * Finds and replaces a missing dimension in output shape.\n   *\n   * This is a near direct port of the internal Numpy function\n   * `_fix_unknown_dimension` in `numpy/core/src/multiarray/shape.c`.\n   *\n   * @param inputShape: Original shape of array begin reshape.\n   * @param outputShape: Target shape of the array, with at most a single\n   * `null` or negative number, which indicates an underdetermined dimension\n   * that should be derived from `inputShape` and the known dimensions of\n   *   `outputShape`.\n   * @returns: The output shape with `null` replaced with its computed value.\n   * @throws: ValueError: If `inputShape` and `outputShape` do not match.\n   */\n  private fixUnknownDimension(inputShape: Shape, outputShape: Shape): Shape {\n    const errorMsg = 'Total size of new array must be unchanged.';\n    const finalShape = outputShape.slice();\n    let known = 1;\n    let unknown = null;\n    for (let i = 0; i < finalShape.length; ++i) {\n      const dim = finalShape[i];\n      if (this.isUnknown(dim)) {\n        if (unknown === null) {\n          unknown = i;\n        } else {\n          throw new ValueError('Can only specifiy one unknown dimension.');\n        }\n      } else {\n        known *= dim;\n      }\n    }\n\n    const originalSize = arrayProd(inputShape);\n    if (unknown !== null) {\n      if (known === 0 || originalSize % known !== 0) {\n        throw new ValueError(errorMsg);\n      }\n      finalShape[unknown] = originalSize / known;\n    } else if (originalSize !== known) {\n      throw new ValueError(errorMsg);\n    }\n\n    return finalShape;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    let anyUnknownDims = false;\n    for (let i = 0; i < inputShape.length; ++i) {\n      if (this.isUnknown(inputShape[i])) {\n        anyUnknownDims = true;\n        break;\n      }\n    }\n\n    if (anyUnknownDims) {\n      return inputShape.slice(0, 1).concat(this.targetShape);\n    } else {\n      return inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const outputShape = inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n      return input.reshape(outputShape);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      targetShape: this.targetShape,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Reshape);\n\nexport declare interface PermuteLayerArgs extends LayerArgs {\n  /**\n   * Array of integers. Permutation pattern. Does not include the\n   * sample (batch) dimension. Index starts at 1.\n   * For instance, `[2, 1]` permutes the first and second dimensions\n   * of the input.\n   */\n  dims: number[];\n}\n\nexport class Permute extends Layer {\n  /** @nocollapse */\n  static className = 'Permute';\n  readonly dims: number[];\n  private readonly dimsIncludingBatch: number[];\n\n  constructor(args: PermuteLayerArgs) {\n    super(args);\n    if (args.dims == null) {\n      throw new Error(\n          'Required configuration field `dims` is missing during Permute ' +\n          'constructor call.');\n    }\n    if (!Array.isArray(args.dims)) {\n      throw new Error(\n          'Permute constructor requires `dims` to be an Array, but received ' +\n          `${args.dims} instead.`);\n    }\n\n    // Check the validity of the permutation indices.\n    const expectedSortedIndices = range(1, args.dims.length + 1);\n    if (!util.arraysEqual(args.dims.slice().sort(), expectedSortedIndices)) {\n      throw new Error(\n          'Invalid permutation `dims`: ' + JSON.stringify(args.dims) +\n          ' `dims` must contain consecutive integers starting from 1.');\n    }\n\n    this.dims = args.dims;\n    this.dimsIncludingBatch = [0].concat(this.dims);\n    this.inputSpec = [new InputSpec({ndim: this.dims.length + 1})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    this.dims.forEach((dim: number, i: number) => {\n      outputShape[i + 1] = (inputShape as Shape)[dim];\n    });\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return transpose(getExactlyOneTensor(inputs), this.dimsIncludingBatch);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      dims: this.dims,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Permute);\n\nexport declare interface MaskingArgs extends LayerArgs {\n  /**\n   * Masking Value. Defaults to `0.0`.\n   */\n  maskValue?: number;\n}\n\nexport class Masking extends Layer {\n  /** @nocollapse */\n  static className = 'Masking';\n  maskValue: number;\n\n  constructor(args?: MaskingArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maskValue = args.maskValue == null ? 0 : args.maskValue;\n    } else {\n      this.maskValue = 0;\n    }\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig() {\n    const baseConfig = super.getConfig();\n    const config = {maskValue: this.maskValue};\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    const input = getExactlyOneTensor(inputs);\n    const axis = -1;\n    return any(notEqual(input, this.maskValue), axis);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const axis = -1;\n      const keepDims = true;\n      const booleanMask = any(notEqual(input, this.maskValue), axis, keepDims);\n      const output = input.mul(booleanMask.asType(input.dtype));\n      return output;\n    });\n  }\n}\nserialization.registerClass(Masking);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Embedding Layer.\n *\n * Original source: keras/constraints.py\n */\nimport {notEqual, serialization, Tensor, tidy, zerosLike} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface EmbeddingLayerArgs extends LayerArgs {\n  /**\n   * Integer > 0. Size of the vocabulary, i.e. maximum integer index + 1.\n   */\n  inputDim: number;\n  /**\n   * Integer >= 0. Dimension of the dense embedding.\n   */\n  outputDim: number;\n  /**\n   * Initializer for the `embeddings` matrix.\n   */\n  embeddingsInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Regularizer function applied to the `embeddings` matrix.\n   */\n  embeddingsRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Constraint function applied to the `embeddings` matrix.\n   */\n  embeddingsConstraint?: ConstraintIdentifier|Constraint;\n  /**\n   * Whether the input value 0 is a special \"padding\" value that should be\n   * masked out. This is useful when using recurrent layers which may take\n   * variable length input.\n   *\n   * If this is `True` then all subsequent layers in the model need to support\n   * masking or an exception will be raised. If maskZero is set to `True`, as a\n   * consequence, index 0 cannot be used in the vocabulary (inputDim should\n   * equal size of vocabulary + 1).\n   */\n  maskZero?: boolean;\n  /**\n   * Length of input sequences, when it is constant.\n   *\n   * This argument is required if you are going to connect `flatten` then\n   * `dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed).\n   */\n  inputLength?: number|number[];\n}\n\nexport class Embedding extends Layer {\n  /** @nocollapse */\n  static className = 'Embedding';\n  private inputDim: number;\n  private outputDim: number;\n  private embeddingsInitializer: Initializer;\n  private maskZero: boolean;\n  private inputLength: number|number[];\n\n  private embeddings: LayerVariable = null;\n\n  readonly DEFAULT_EMBEDDINGS_INITIALIZER: InitializerIdentifier =\n      'randomUniform';\n  private readonly embeddingsRegularizer?: Regularizer;\n  private readonly embeddingsConstraint?: Constraint;\n\n  constructor(args: EmbeddingLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null) {\n      // Porting Note: This logic is copied from Layer's constructor, since we\n      // can't do exactly what the Python constructor does for Embedding().\n      // Specifically, the super constructor can not be called after the\n      // mutation of the `config` argument.\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      if (args.inputLength == null) {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (None, )\n        this.batchInputShape = [batchSize, null];\n      } else {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (config.inputLength, )\n        this.batchInputShape =\n            [batchSize].concat(generic_utils.toList(args.inputLength));\n      }\n    }\n    this.inputDim = args.inputDim;\n    generic_utils.assertPositiveInteger(this.inputDim, 'inputDim');\n    this.outputDim = args.outputDim;\n    generic_utils.assertPositiveInteger(this.outputDim, 'outputDim');\n    this.embeddingsInitializer = getInitializer(\n        args.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER);\n    this.embeddingsRegularizer = getRegularizer(args.embeddingsRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.embeddingsConstraint = getConstraint(args.embeddingsConstraint);\n    this.maskZero = args.maskZero;\n    this.supportsMasking = args.maskZero;\n    this.inputLength = args.inputLength;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    this.embeddings = this.addWeight(\n        'embeddings', [this.inputDim, this.outputDim], this.dtype,\n        this.embeddingsInitializer, this.embeddingsRegularizer, true,\n        this.embeddingsConstraint);\n    this.built = true;\n  }\n\n  // Override warnOnIncompatibleInputShape because an embedding layer allows\n  // the input to have varying ranks.\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {}\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tidy(() => {\n      if (!this.maskZero) {\n        return null;\n      } else {\n        inputs = getExactlyOneTensor(inputs);\n        return notEqual(inputs, zerosLike(inputs));\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    if (this.inputLength == null) {\n      return [...inputShape, this.outputDim];\n    }\n    // inputLength can be an array if input is 3D or higher.\n    const inLens: number[] = generic_utils.toList(this.inputLength);\n    if (inLens.length !== inputShape.length - 1) {\n      throw new ValueError(\n          `\"inputLength\" is ${this.inputLength}, but received ` +\n          `input shape has shape ${inputShape}`);\n    } else {\n      let i = 0;\n      for (let k = 0; k < inLens.length; ++k) {\n        const s1 = inLens[k];\n        const s2 = inputShape[k + 1];\n        if ((s1 != null) && (s2 != null) && (s1 !== s2)) {\n          throw new ValueError(\n              `\"inputLength\" is ${this.inputLength}, but received ` +\n              `input shape has shape ${inputShape}`);\n        } else if (s1 == null) {\n          inLens[i] = s2;\n        }\n        i++;\n      }\n    }\n    return [inputShape[0], ...inLens, this.outputDim];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Embedding layer accepts only a single input.\n      let input = getExactlyOneTensor(inputs);\n      if (input.dtype !== 'int32') {\n        input = K.cast(input, 'int32');\n      }\n      const output = K.gather(this.embeddings.read(), input.as1D());\n      return output.reshape(\n          getExactlyOneShape(this.computeOutputShape(input.shape)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      inputDim: this.inputDim,\n      outputDim: this.outputDim,\n      embeddingsInitializer: serializeInitializer(this.embeddingsInitializer),\n      embeddingsRegularizer: serializeRegularizer(this.embeddingsRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      embeddingsConstraint: serializeConstraint(this.embeddingsConstraint),\n      maskZero: this.maskZero,\n      inputLength: this.inputLength\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Embedding);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Merge Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {l2Normalize} from '../losses';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as mathUtils from '../utils/math_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\n/**\n * Generic Merge layer for element-wise merge functions.\n *\n * Used to implement `Sum`, `Average`, `Concatenate`, etc.\n */\nexport abstract class Merge extends Layer {\n  protected reshapeRequired: boolean;\n\n  constructor(args?: LayerArgs) {\n    super(args || {});\n    this.supportsMasking = true;\n  }\n\n  /**\n   * Logic for merging multiple tensors, to be overridden by subclasses.\n   * @param inputs\n   */\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    throw new NotImplementedError();\n  }\n\n  /**\n   * Computes the shape of the result of an elementwise operation.\n   *\n   * @param shape1: Shape of the first tensor.\n   * @param shape2: Shape of the second tensor.\n   * @returns Expected output shape when an elementwise operation is carried\n   *   out on 2 tensors with shapes `shape1` and `shape2`.\n   * @throws ValueError: If `shape1` and `shape2` are not compatible for\n   *   element-wise operations.\n   */\n  private computeElementwiseOpOutputShape(shape1: Shape, shape2: Shape): Shape {\n    if (shape1 == null || shape2 == null) {\n      return null;\n    } else if (shape1.length < shape2.length) {\n      return this.computeElementwiseOpOutputShape(shape2, shape1);\n    } else if (shape2.length === 0) {\n      return shape1;\n    }\n    const outputShape: Shape = shape1.slice(0, shape1.length - shape2.length);\n    for (let k = 0; k < shape2.length; ++k) {\n      const i = shape1[shape1.length - shape2.length + k];\n      const j = shape2[k];\n      if (i == null || j == null || i < 0 || j < 0) {\n        outputShape.push(null);\n      } else if (i === 1) {\n        outputShape.push(j);\n      } else if (j === 1) {\n        outputShape.push(i);\n      } else {\n        if (i !== j) {\n          throw new ValueError(\n              'Operands could not be broadcast together with shapes ' +\n              JSON.stringify(shape1) + ' ' + JSON.stringify(shape2));\n        }\n        outputShape.push(i);\n      }\n    }\n    return outputShape;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.\n    if (Array.isArray(inputShape) && !Array.isArray(inputShape[0])) {\n      // Make sure that inputShape is an Array of shape.\n      inputShape = [getExactlyOneShape(inputShape)];\n    }\n    inputShape = inputShape as Shape[];\n    if (inputShape.length < 2) {\n      throw new ValueError(\n          'A merge layer should be called on an Array of at least 2 inputs.' +\n          ` Got ${inputShape.length} input(s).`);\n    }\n\n    // Make sure that there is at most one unique batch size among the input\n    // shapes.\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length > 1) {\n      throw new ValueError(\n          `Can not merge tensors with different batch sizes. ` +\n          `Got tensors with shapes: ${JSON.stringify(inputShape)}.`);\n    }\n\n    let outputShape: Shape =\n        inputShape[0] == null ? null : inputShape[0].slice(1);\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n    // If the inputs have different ranks, we have to reshape them to make them\n    // broadcastable.\n    const allRanks = inputShape.map(shape => shape.length);\n    if (inputShape.indexOf(null) === -1 &&\n        generic_utils.unique(allRanks).length === 1) {\n      this.reshapeRequired = false;\n    } else {\n      this.reshapeRequired = true;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (this.reshapeRequired) {\n        const reshapedInputs: Tensor[] = [];\n        const inputDims = inputs.map(input => input.rank);\n        if (inputDims.indexOf(null) === -1) {\n          // If ranks of all inputs are available, we simply expand each of them\n          // at axis=1 until all of them have the same rank.\n          const maxNDim = mathUtils.max(inputDims);\n          for (let x of inputs) {\n            const xNDim = x.rank;\n            for (let k = 0; k < maxNDim - xNDim; ++k) {\n              x = K.expandDims(x, 1);\n            }\n            reshapedInputs.push(x);\n          }\n          return this.mergeFunction(reshapedInputs);\n        } else {\n          // Transpose all inputs so that batch size is the last dimension.\n          // [batchSize, dim1, dim2, ...] -> [dim1, dim2, ..., batchSize]\n          let transposed = false;\n          for (const x of inputs) {\n            const xNDim = x.rank;\n            if (xNDim == null) {\n              const xShape = x.shape;\n              const batchSize = xShape[0];\n              const newShape = xShape.slice(1).concat([batchSize]);\n              let xTransposed = x.reshape(\n                  [batchSize].concat(mathUtils.arrayProd(xShape.slice(1))));\n              xTransposed = tfc.transpose(xTransposed, [1, 0]);\n              xTransposed = xTransposed.reshape(newShape);\n              reshapedInputs.push(xTransposed);\n              transposed = true;\n            } else if (xNDim > 1) {\n              const dims = mathUtils.range(1, xNDim).concat([0]);\n              reshapedInputs.push(tfc.transpose(x, dims));\n              transposed = true;\n            } else {\n              // We don't transpose inputs if they are 1D vectors or scalars.\n              reshapedInputs.push(x);\n            }\n          }\n          let y = this.mergeFunction(reshapedInputs);\n          const yNDim = y.rank;\n          if (transposed) {\n            // If inputs have been transposed, we have to transpose the output\n            // too.\n            if (yNDim == null) {\n              const yShape = y.shape;\n              const yNDim = yShape.length;\n              const batchSize = yShape[yNDim - 1];\n              const newShape =\n                  [batchSize].concat(yShape.slice(0, yShape.length - 1));\n              y = tfc.transpose(y.reshape([-1, batchSize]), [1, 0])\n                      .reshape(newShape);\n            } else if (yNDim > 1) {\n              const dims = [yNDim - 1].concat(mathUtils.range(0, yNDim - 1));\n              y = tfc.transpose(y, dims);\n            }\n          }\n          return y;\n        }\n      } else {\n        return this.mergeFunction(inputs);\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape[];\n    let outputShape: Shape;\n    if (inputShape[0] == null) {\n      outputShape = null;\n    } else {\n      outputShape = inputShape[0].slice(1);\n    }\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length === 1) {\n      outputShape = batchSizes.concat(outputShape);\n    } else {\n      outputShape = [null].concat(outputShape);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tfc.tidy(() => {\n      if (mask == null) {\n        return null;\n      }\n      if (!Array.isArray(mask)) {\n        throw new ValueError('`mask` should be an Array');\n      }\n      if (!Array.isArray(inputs)) {\n        throw new ValueError('`inputs` should be an Array');\n      }\n      if (mask.length !== inputs.length) {\n        throw new ValueError(\n            `The Array 'inputs' and 'mask' are expected to have the same ` +\n            `length, but have different lengths ` +\n            `(${inputs.length} vs ${mask.length})`);\n      }\n      if (mask.every(m => m == null)) {\n        return null;\n      }\n      mask = mask.map(m => m == null ? m : tfc.expandDims(m, 0));\n      let output = mask[0];\n      for (let i = 1; i < mask.length - 1; ++i) {\n        output = tfc.logicalAnd(output, mask[i]);\n      }\n      return output;\n    });\n  }\n}\n\nexport class Add extends Merge {\n  /** @nocollapse */\n  static className = 'Add';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Add);\n\n/**\n * Calculate the element-wise sum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Add` layer, by using no input argument\n *    or a single configuration argument. The resultant `Add` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const addLayer = tf.layers.add();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = addLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.add([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.add([input1, input2]).print();\n * // Gives [[11, 22], [33, 44]].\n *\n */\nexport function add(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Add({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Add(config);\n  }\n}\n\nexport class Multiply extends Merge {\n  /** @nocollapse */\n  static className = 'Multiply';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.mul(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Multiply);\n\n/**\n * Calculate the element-wise product of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Multiply` layer, by using no input argument\n *    or a single configuration argument. The resultant `Multiply` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const multiplyLayer = tf.layers.multiply();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = multiplyLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.multiply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.multiply([input1, input2]).print();\n * // Gives [[10, 40], [90, 160]].\n *\n */\nexport function multiply(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Multiply({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Multiply(config);\n  }\n}\n\nexport class Average extends Merge {\n  /** @nocollapse */\n  static className = 'Average';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return tfc.mul(1 / inputs.length, output);\n    });\n  }\n}\nserialization.registerClass(Average);\n\n/**\n * Calculate the element-wise arithmetic mean of inputs, which all have the same\n * shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Average` layer, by using no input argument\n *    or a single configuration argument. The resultant `Average` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const averageLayer = tf.layers.average();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = averageLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.average([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.average([input1, input2]).print();\n * // Gives [[5.5, 11], [16.5, 22]].\n *\n */\nexport function average(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Average({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Average(config);\n  }\n}\n\nexport class Maximum extends Merge {\n  /** @nocollapse */\n  static className = 'Maximum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.maximum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Maximum);\n\n/**\n * Calculate the element-wise maximum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Maximum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Maximum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const maximumLayer = tf.layers.maximum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = maximumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.maximum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.maximum([input1, input2]).print();\n * // Gives [[10, 20], [30, 40]].\n *\n */\nexport function maximum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Maximum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Maximum(config);\n  }\n}\n\nexport class Minimum extends Merge {\n  /** @nocollapse */\n  static className = 'Minimum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.minimum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Minimum);\n\n/**\n * Calculate the element-wise minimum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Minimum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Minimum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const minimumLayer = tf.layers.minimum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = minimumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.minimum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.minimum([input1, input2]).print();\n * // Gives [[1, 2], [3, 4]].\n *\n */\nexport function minimum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Minimum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Minimum(config);\n  }\n}\n\nexport declare interface ConcatenateLayerArgs extends LayerArgs {\n  /**\n   * Axis along which to concatenate.\n   */\n  axis?: number;\n}\n\nexport class Concatenate extends Merge {\n  /** @nocollapse */\n  static className = 'Concatenate';\n  readonly DEFAULT_AXIS = -1;\n  private readonly axis: number;\n\n  constructor(args?: ConcatenateLayerArgs) {\n    super(args);\n    if (args == null) {\n      args = {};\n    }\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.]\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0])) ||\n        inputShape.length === 1) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of at least 2 ' +\n          'inputs');\n    }\n    inputShape = inputShape as Shape[];\n\n    let allNoneShape = true;\n    for (const shape of inputShape) {\n      if (shape != null) {\n        allNoneShape = false;\n        break;\n      }\n    }\n    if (allNoneShape) {\n      return;\n    }\n\n    const shapeSet: Shape[] = [];\n    for (let i = 0; i < inputShape.length; ++i) {\n      const shapeWithoutConcatAxis = inputShape[i].slice();\n      shapeWithoutConcatAxis.splice(this.axis, 1);\n      let exists = false;\n      for (const shape of shapeSet) {\n        if (util.arraysEqual(shape, shapeWithoutConcatAxis)) {\n          exists = true;\n          break;\n        }\n      }\n      if (!exists) {\n        shapeSet.push(shapeWithoutConcatAxis);\n      }\n    }\n    if (shapeSet.length > 1) {\n      throw new ValueError(\n          'A `Concatenate` layer requires inputs with matching shapes ' +\n          'except for the concat axis. Got input shapes: ' +\n          JSON.stringify(inputShape));\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      return K.concatenate(inputs, this.axis);\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0]))) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of inputs.');\n    }\n    const inputShapes = inputShape as Shape[];\n    const outputShape = inputShapes[0].slice();\n    const axis = this.axis < 0 ? outputShape.length + this.axis : this.axis;\n    // Porting Note: the line above is because TypeScript doesn't support\n    //   negative indices.\n    for (const shape of inputShapes.slice(1)) {\n      if (outputShape[axis] == null || shape[axis] == null) {\n        outputShape[axis] = null;\n        break;\n      }\n      outputShape[axis] += shape[axis];\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    if (mask == null) {\n      return null;\n    }\n    if (!Array.isArray(mask)) {\n      throw new ValueError('`mask` should be an array for Concatenate');\n    }\n    if (!Array.isArray(inputs)) {\n      throw new ValueError('`inputs` should be an array for Concatenate');\n    }\n    if (mask.length !== inputs.length) {\n      throw new ValueError(\n          `Mismatch in the length of mask (${mask.length}) ` +\n          `and the legnth of inputs (${inputs.length})`);\n    }\n    return tfc.tidy(() => {\n      let allNullMasks = true;\n      mask.forEach(m => {\n        if (m != null) {\n          allNullMasks = false;\n          return;\n        }\n      });\n      if (allNullMasks) {\n        return null;\n      }\n      const outputMasks: Tensor[] = [];\n      for (let i = 0; i < inputs.length; ++i) {\n        if (mask[i] == null) {\n          // Input is unmasked. Append all 1's to masks.\n          outputMasks.push(tfc.onesLike(inputs[i]).asType('bool'));\n        } else if (mask[i].rank < inputs[i].rank) {\n          // Mask is smaller than the input, expand it.\n          outputMasks.push(tfc.expandDims(mask[i], -1));\n        } else {\n          outputMasks.push(mask[i]);\n        }\n      }\n      const concatenatedMasks = tfc.concat(outputMasks, this.axis);\n      return tfc.all(concatenatedMasks, -1, false);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axis': this.axis,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Concatenate);\n\n/**\n * Concatenate an `Array` of inputs.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Concatenate` layer, by using no input argument\n *    or a single configuration argument. The resultant `Concatenate` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const concatLayer = tf.layers.concatenate();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = concatLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 7], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = tf.layers.concatenate([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([[1, 2], [3, 4]], [2, 2]);\n * const input2 = tf.tensor2d([[10, 20], [30, 40]], [2, 2]);\n * tf.layers.concatenate([input1, input2]).print();\n * // Gives [[1, 2, 10, 20], [3, 4, 30, 40]].\n *\n */\nexport function concatenate(config?: SymbolicTensor[]|Tensor[]|\n                            ConcatenateLayerArgs): Layer|SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Concatenate({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Concatenate(config);\n  }\n}\n\nexport declare interface DotLayerArgs extends LayerArgs {\n  /**\n   * Axis or axes along which the dot product will be taken.\n   *\n   * Integer or an Array of integers.\n   */\n  axes: number|[number, number];\n\n  /**\n   * Whether to L2-normalize samples along the dot product axis\n   * before taking the dot product.\n   *\n   * If set to `true`, the output of the dot product isthe cosine\n   * proximity between the two samples.\n   */\n  normalize?: boolean;\n}\n\n/**\n * Interpretable potentially negative axis index.\n *\n * For example, given axis = -1, and dim = 3, this function will return 2.\n *\n * @param axis The axis index, may be a positive, zero or negative integer.\n * @param dim Total number of dimensions, a positive integer.\n * @returns A non-negative axis index equivalent to the input `axis`.\n */\nfunction interpretAxis(axis: number, dim: number): number {\n  while (axis < 0) {\n    axis += dim;\n  }\n  return axis;\n}\n\nfunction batchDot(x: Tensor, y: Tensor, axes: number|[number, number]): Tensor {\n  if (x.shape.length > 3 || y.shape.length > 3) {\n    throw new NotImplementedError(\n        'batchDot is not implemented for tensors of 4D or higher rank yet');\n  }\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of x to be >= 2, ` +\n          `but got ${x.shape.length}`);\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of y to be >= 2, ` +\n          `but got ${y.shape.length}`);\n\n  if (typeof axes === 'number') {\n    axes = [axes, axes];\n  }\n\n  if (x.dtype === 'complex64' || y.dtype === 'complex64') {\n    throw new NotImplementedError(\n        'batchDot is not implemented for complex64-type Tensors yet.');\n  }\n\n  const xNDim = x.shape.length;\n  const yNDim = y.shape.length;\n  if (axes == null) {\n    // Behave like batchMatmul by default.\n    axes = [xNDim - 1, yNDim - 2];\n  }\n  const axesArray = axes as [number, number];\n\n  return tfc.tidy(() => {\n    let diff: number;\n    if (xNDim > yNDim) {\n      diff = xNDim - yNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      y = y.reshape(y.shape.concat(diffShape));\n    } else if (yNDim > xNDim) {\n      diff = yNDim - xNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      x = x.reshape(x.shape.concat(diffShape));\n    } else {\n      diff = 0;\n    }\n\n    let out: Tensor;\n    if (x.shape.length === 2 && y.shape.length === 2) {\n      if (axesArray[0] === axesArray[1]) {\n        out = x.mul(y).sum(axesArray[0]);\n      } else {\n        out = x.transpose([1, 0]).mul(y).sum(axesArray[1]);\n      }\n    } else {\n      const adjX = axesArray[0] !== x.shape.length - 1;\n      const adjY = axesArray[1] === y.shape.length - 1;\n      out = x.matMul(y, adjX, adjY);\n    }\n\n    if (diff > 0) {\n      let idx: number;\n      if (xNDim > yNDim) {\n        idx = xNDim + yNDim - 3;\n      } else {\n        idx = xNDim - 1;\n      }\n      const squeezeAxes: number[] = [];\n      for (let i = idx; i < idx + diff; ++i) {\n        squeezeAxes.push(i);\n      }\n      out = out.squeeze(squeezeAxes);\n    }\n    if (out.shape.length === 1) {\n      out = out.expandDims(1);\n    }\n    return out;\n  });\n}\n\nexport class Dot extends Merge {\n  /** @nocollapse */\n  static className = 'Dot';\n\n  private axes: number|[number, number];\n  private normalize: boolean;\n\n  constructor(args: DotLayerArgs) {\n    super(args);\n    this.axes = args.axes;\n    this.normalize = args.normalize == null ? false : args.normalize;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = inputShape[0] as Shape;\n    const shape2 = inputShape[1] as Shape;\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    if (shape1[axes[0]] !== shape2[axes[1]]) {\n      throw new ValueError(\n          `Dimension incompatibility: ` +\n          `${shape1[axes[0]]} !== ${shape2[axes[1]]}`);\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    if (inputs.length !== 2) {\n      throw new ValueError(\n          'A `Dot` layer must be called on exactly 2 inputs, ' +\n          `but received ${inputs.length} input(s).`);\n    }\n\n    let x1 = inputs[0];\n    let x2 = inputs[1];\n    let axes: [number, number];\n    if (!Array.isArray(this.axes)) {\n      axes = [\n        interpretAxis(this.axes, x1.shape.length),\n        interpretAxis(this.axes, x2.shape.length)\n      ];\n    } else {\n      axes = this.axes.map(\n                 (axis, i) => interpretAxis(\n                     axis, inputs[i].shape.length)) as [number, number];\n    }\n    if (this.normalize) {\n      x1 = l2Normalize(x1, axes[0]);\n      x2 = l2Normalize(x2, axes[1]);\n    }\n    return batchDot(x1, x2, axes);\n  }\n\n  private interpretAxes(shape1: Shape, shape2: Shape): number[] {\n    let axes: number[];\n    if (!Array.isArray(this.axes)) {\n      // `this.axes` is a single integer.\n      axes = [\n        interpretAxis(this.axes, shape1.length),\n        interpretAxis(this.axes, shape2.length)\n      ];\n    } else {\n      // `this.axes` is an Array of integers.\n      axes = this.axes;\n    }\n    return axes;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = (inputShape[0] as Shape).slice();\n    const shape2 = (inputShape[1] as Shape).slice();\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    shape1.splice(axes[0], 1);\n    shape2.splice(axes[1], 1);\n    shape2.splice(0, 1);\n    const outputShape = shape1.concat(shape2);\n    if (outputShape.length === 1) {\n      outputShape.push(1);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return null;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axes': this.axes,\n      'normalize': this.normalize\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dot);\n\n// TODO(cais): Add functional interfaces for the merge layers.\n", "/**\r\n * @license\r\n * Copyright 2018 Google LLC\r\n *\r\n * Use of this source code is governed by an MIT-style\r\n * license that can be found in the LICENSE file or at\r\n * https://opensource.org/licenses/MIT.\r\n * =============================================================================\r\n */\r\n\r\n/**\r\n * TensorFlow.js Layers: Noise Layers.\r\n */\r\n\r\nimport {greaterEqual, randomUniform, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\r\n\r\nimport * as K from '../backend/tfjs_backend';\r\nimport {Layer, LayerArgs} from '../engine/topology';\r\nimport {Shape} from '../keras_format/common';\r\nimport {Kwargs} from '../types';\r\nimport {getExactlyOneTensor} from '../utils/types_utils';\r\n\r\nexport declare interface GaussianNoiseArgs extends LayerArgs {\r\n  /** Standard Deviation.  */\r\n  stddev: number;\r\n}\r\n\r\nexport class GaussianNoise extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianNoise';\r\n  readonly stddev: number;\r\n\r\n  constructor(args: GaussianNoiseArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.stddev = args.stddev;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {stddev: this.stddev};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      const noised = () =>\r\n          K.randomNormal(input.shape, 0, this.stddev).add(input);\r\n      const output =\r\n          K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      return output;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianNoise);\r\n\r\nexport declare interface GaussianDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n}\r\n\r\nexport class GaussianDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianDropout';\r\n  readonly rate: number;\r\n\r\n  constructor(args: GaussianDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      if (this.rate > 0 && this.rate < 1) {\r\n        const noised = () => {\r\n          const stddev = Math.sqrt(this.rate / (1 - this.rate));\r\n          return input.mul(K.randomNormal(input.shape, 1, stddev));\r\n        };\r\n        return K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      }\r\n      return input;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianDropout);\r\n\r\nexport declare interface AlphaDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n  /**\r\n   * A 1-D `Tensor` of type `int32`, representing the\r\n   * shape for randomly generated keep/drop flags.\r\n   */\r\n  noiseShape?: Shape;\r\n}\r\n\r\n/**\r\n * Applies Alpha Dropout to the input.\r\n *\r\n * As it is a regularization layer, it is only active at training time.\r\n *\r\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\r\n * to their original values, in order to ensure the self-normalizing property\r\n * even after this dropout.\r\n * Alpha Dropout fits well to Scaled Exponential Linear Units\r\n * by randomly setting activations to the negative saturation value.\r\n *\r\n * Arguments:\r\n *   - `rate`: float, drop probability (as with `Dropout`).\r\n *     The multiplicative noise will have\r\n *     standard deviation `sqrt(rate / (1 - rate))`.\r\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\r\n *     shape for randomly generated keep/drop flags.\r\n *\r\n * Input shape:\r\n *   Arbitrary. Use the keyword argument `inputShape`\r\n *   (tuple of integers, does not include the samples axis)\r\n *   when using this layer as the first layer in a model.\r\n *\r\n * Output shape:\r\n *   Same shape as input.\r\n *\r\n * References:\r\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\r\n */\r\nexport class AlphaDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'AlphaDropout';\r\n  readonly rate: number;\r\n  readonly noiseShape: Shape;\r\n\r\n  constructor(args: AlphaDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n    this.noiseShape = args.noiseShape;\r\n  }\r\n\r\n  _getNoiseShape(inputs: Tensor|Tensor[]) {\r\n    return this.noiseShape || getExactlyOneTensor(inputs).shape;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      if (this.rate < 1 && this.rate > 0) {\r\n        const noiseShape = this._getNoiseShape(inputs);\r\n\r\n        const droppedInputs = () => {\r\n          const input = getExactlyOneTensor(inputs);\r\n\r\n          const alpha = 1.6732632423543772848170429916717;\r\n          const scale = 1.0507009873554804934193349852946;\r\n\r\n          const alphaP = -alpha * scale;\r\n\r\n          let keptIdx = greaterEqual(randomUniform(noiseShape), this.rate);\r\n\r\n          keptIdx = K.cast(keptIdx, 'float32');  // get default dtype.\r\n\r\n          // Get affine transformation params.\r\n          const a = ((1 - this.rate) * (1 + this.rate * alphaP ** 2)) ** -0.5;\r\n          const b = -a * alphaP * this.rate;\r\n\r\n          // Apply mask.\r\n          const x = input.mul(keptIdx).add(keptIdx.add(-1).mul(alphaP));\r\n\r\n          return x.mul(a).add(b);\r\n        };\r\n        return K.inTrainPhase(\r\n            droppedInputs, () => getExactlyOneTensor(inputs),\r\n            kwargs['training'] || false);\r\n      }\r\n      return inputs;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(AlphaDropout);\r\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Normalization layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {moments, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Applies batch normalization on x given mean, var, beta and gamma.\n *\n * I.e. returns:\n *   `output = (x - mean) / (sqrt(var) + epsilon) * gamma + beta`\n *\n * @param x Input tensor.\n * @param mean Mean of batch.\n * @param variance Variance of batch.\n * @param beta Tensor with which to center the input.\n * @param gamma Tensor by which to scale the input.\n * @param epsilon Fuzz factor.\n * @returns The result of the batch normalization.\n */\nexport function batchNormalization(\n    x: Tensor, mean: Tensor, variance: Tensor, beta?: Tensor, gamma?: Tensor,\n    epsilon = 1e-3): Tensor {\n  let out: Tensor;\n  if (x.rank === 2) {\n    out = tfc.batchNorm2d(\n        x as Tensor2D, mean as Tensor2D | Tensor1D,\n        variance as Tensor2D | Tensor1D, beta as Tensor2D | Tensor1D,\n        gamma as Tensor2D | Tensor1D, epsilon);\n  } else if (x.rank === 3) {\n    // TODO(cais): Check rank; give proper error message.\n    out = tfc.batchNorm3d(\n        x as Tensor3D, mean as Tensor3D | Tensor1D,\n        variance as Tensor3D | Tensor1D, beta as Tensor3D | Tensor1D,\n        gamma as Tensor3D | Tensor1D, epsilon);\n  } else if (x.rank === 4) {\n    out = tfc.batchNorm4d(\n        x as Tensor4D, mean as Tensor4D | Tensor1D,\n        variance as Tensor4D | Tensor1D, beta as Tensor4D | Tensor1D,\n        gamma as Tensor4D | Tensor1D, epsilon);\n  } else {\n    throw new NotImplementedError(\n        `batchNormalization is not implemented for array of rank ${x.rank} ` +\n        `yet`);\n  }\n  return out;\n}\n\n/**\n * Non-broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction regularNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const normed =\n               batchNormalization(x, mean, variance, beta, gamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction broadcastNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const targetShape: number[] = [];\n           for (const axis of math_utils.range(0, x.rank)) {\n             if (reductionAxes.indexOf(axis) !== -1) {\n               targetShape.push(1);\n             } else {\n               targetShape.push(x.shape[axis]);\n             }\n           }\n           const broadcastMean = mean.reshape(targetShape);\n           const broadcastVariance = variance.reshape(targetShape);\n           const broadcastGamma =\n               gamma == null ? null : gamma.reshape(targetShape);\n           const broadcastBeta =\n               beta == null ? null : beta.reshape(targetShape);\n           const normed = batchNormalization(\n               x, broadcastMean, broadcastVariance, broadcastBeta,\n               broadcastGamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Batch normalization for use in training (not inference).\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nexport function normalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  if (util.arraysEqual(\n          reductionAxes.slice().sort(), math_utils.range(0, x.rank - 1))) {\n    return regularNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  } else {\n    return broadcastNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  }\n}\n\nexport declare interface BatchNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The integer axis that should be normalized (typically the features axis).\n   * Defaults to -1.\n   *\n   * For instance, after a `Conv2D` layer with `data_format=\"channels_first\"`,\n   * set `axis=1` in `batchNormalization`.\n   */\n  axis?: number;\n\n  /**\n   * Momentum of the moving average. Defaults to 0.99.\n   */\n  momentum?: number;\n\n  /**\n   * Small float added to the variance to avoid dividing by zero. Defaults to\n   * 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Defaults to `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear (also e.g. `nn.relu`),\n   * this can be disabled since the scaling will be done by the next layer.\n   * Defaults to `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   *  Defaults to 'zeros'.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   *  Defaults to `ones`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving mean.\n   * Defaults to `zeros`\n   */\n  movingMeanInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving variance.\n   *  Defaults to 'Ones'.\n   */\n  movingVarianceInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the beta weight.\n   */\n  betaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for gamma weight.\n   */\n  gammaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer for the beta weight.\n   */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer for the gamma weight.\n   */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class BatchNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'BatchNormalization';\n  private readonly axis: number;\n  private readonly momentum: number;\n  private readonly epsilon: number;\n  private readonly center: boolean;\n  private readonly scale: boolean;\n  private readonly betaInitializer: Initializer;\n  private readonly gammaInitializer: Initializer;\n  private readonly movingMeanInitializer: Initializer;\n  private readonly movingVarianceInitializer: Initializer;\n  private readonly betaConstraint: Constraint;\n  private readonly gammaConstraint: Constraint;\n  private readonly betaRegularizer: Regularizer;\n  private readonly gammaRegularizer: Regularizer;\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n  private movingMean: LayerVariable;\n  private movingVariance: LayerVariable;\n\n  constructor(args?: BatchNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.supportsMasking = true;\n    this.axis = args.axis == null ? -1 : args.axis;\n    this.momentum = args.momentum == null ? 0.99 : args.momentum;\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.movingMeanInitializer =\n        getInitializer(args.movingMeanInitializer || 'zeros');\n    this.movingVarianceInitializer =\n        getInitializer(args.movingVarianceInitializer || 'ones');\n    this.betaConstraint = getConstraint(args.betaConstraint);\n    this.gammaConstraint = getConstraint(args.gammaConstraint);\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const axis = this.axis >= 0 ? this.axis : (this.axis + inputShape.length);\n    const dim = inputShape[axis];\n    if (dim == null) {\n      throw new ValueError(\n          `Axis ${axis} of input tensor should have a defined dimension but ` +\n          `the layer received an input with shape ` +\n          `${JSON.stringify(inputShape)}.`);\n    }\n    this.inputSpec =\n        [new InputSpec({ndim: inputShape.length, axes: {[axis]: dim}})];\n    const shape = [dim];\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', shape, null, this.gammaInitializer, this.gammaRegularizer,\n          true, this.gammaConstraint);\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', shape, null, this.betaInitializer, this.betaRegularizer, true,\n          this.betaConstraint);\n    }\n    this.movingMean = this.addWeight(\n        'moving_mean', shape, null, this.movingMeanInitializer, null, false);\n    this.movingVariance = this.addWeight(\n        'moving_variance', shape, null, this.movingVarianceInitializer, null,\n        false);\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const ndim = inputShape.length;\n      const reductionAxes = math_utils.range(0, ndim);\n      const axis = this.axis >= 0 ? this.axis : (this.axis + ndim);\n      reductionAxes.splice(axis, 1);\n      const broadcastShape = generic_utils.pyListRepeat(1, ndim);\n      broadcastShape[axis] = inputShape[axis];\n\n      const sortedReductionAxes = reductionAxes.slice();\n      sortedReductionAxes.sort();\n      const needsBroadcasting = !util.arraysEqual(\n          sortedReductionAxes, math_utils.range(0, ndim).slice(0, ndim - 1));\n\n      const normalizeInference: () => Tensor = () => {\n        if (needsBroadcasting) {\n          const broadcastMovingMean =\n              this.movingMean.read().reshape(broadcastShape);\n          const broadcastMovingVariance =\n              this.movingVariance.read().reshape(broadcastShape);\n          const broadcastBeta =\n              this.center ? this.beta.read().reshape(broadcastShape) : null;\n          const broadcastGamma =\n              this.scale ? this.gamma.read().reshape(broadcastShape) : null;\n          return batchNormalization(\n              input, broadcastMovingMean, broadcastMovingVariance,\n              broadcastBeta, broadcastGamma, this.epsilon);\n        } else {\n          return batchNormalization(\n              input, this.movingMean.read(), this.movingVariance.read(),\n              this.beta == null ? null : this.beta.read(),\n              this.gamma == null ? null : this.gamma.read(), this.epsilon);\n        }\n      };\n\n      if (!training) {\n        return normalizeInference();\n      }\n\n      const [normedTraining, mean, variance] = normalizeBatchInTraining(\n          input, this.gamma.read(), this.beta.read(), reductionAxes,\n          this.epsilon);\n\n      const doMovingAverage =\n          (variable: LayerVariable, value: Tensor, momentum: number): void => {\n            tfc.tidy(() => {\n              const decay = 1 - momentum;\n              const origValue = variable.read();\n              const updateDelta = origValue.sub(value).mul(decay);\n              variable.write(origValue.sub(updateDelta));\n            });\n          };\n\n      // Perform updates to moving mean and moving variance for training.\n      // Porting Note: In PyKeras, these updates to `movingMean` and\n      //   `movingAverage` are done as a deferred Graph, added to the `Layer`'s\n      //   `update`s using the `add_update()` method. Here we do it imperatively\n      //   and encapsulate the updates in a function that is invoked\n      //   immediately.\n      const updateMovingMeanAndVariance = () => {\n        doMovingAverage(this.movingMean, mean, this.momentum);\n        doMovingAverage(this.movingVariance, variance, this.momentum);\n      };\n      updateMovingMeanAndVariance();\n\n      return normedTraining;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      momentum: this.momentum,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      movingMeanInitializer: serializeInitializer(this.movingMeanInitializer),\n      movingVarianceInitializer:\n          serializeInitializer(this.movingVarianceInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer),\n      betaConstraint: serializeConstraint(this.betaConstraint),\n      gammaConstraint: serializeConstraint(this.gammaConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(BatchNormalization);\n\nexport interface LayerNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The axis or axes that should be normalized (typically, the feature axis.)\n   * Defaults to -1 (the last axis.)\n   */\n  axis?: number|number[];\n\n  /**\n   * A small positive float added to variance to avoid divison by zero.\n   * Defaults to 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Default: `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply output by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear, this can be disabled since scaling will\n   * be done by the next layer.\n   * Default: `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   * Default: `'zeros'`.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   * Default: `'ones'`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /** Regularizer for the beta weight. */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /** Regularizer for the gamma weight. */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class LayerNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'LayerNormalization';\n\n  private axis: number|number[];\n  readonly epsilon: number;\n  readonly center: boolean;\n  readonly scale: boolean;\n  readonly betaInitializer: Initializer;\n  readonly gammaInitializer: Initializer;\n  readonly betaRegularizer: Regularizer;\n  readonly gammaRegularizer: Regularizer;\n\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n\n  constructor(args?: LayerNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.axis = args.axis == null ? -1 : args.axis;\n    if (typeof this.axis === 'number') {\n      if (!Number.isInteger(this.axis)) {\n        throw new Error(\n            `Expected axis to be an integer, but received ${this.axis}`);\n      }\n    } else if (Array.isArray(this.axis)) {\n      for (const axis of this.axis) {\n        if (!Number.isInteger(axis)) {\n          throw new Error(\n              `Expected axis to be an array of integers, ` +\n              `but received ${JSON.stringify(this.axis)}`);\n        }\n      }\n    } else {\n      throw new Error(\n          `Expected axis to be an integer or an array of integers, ` +\n          `but received ${JSON.stringify(this.axis)}`);\n    }\n\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n\n    this.supportsMasking = true;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const nDims = inputShape.length;\n\n    // Convert axis to array and resolve negatives.\n    if (typeof this.axis === 'number') {\n      this.axis = [this.axis];\n    }\n    for (let i = 0; i < this.axis.length; ++i) {\n      if (this.axis[i] < 0) {\n        this.axis[i] += nDims;\n      }\n    }\n\n    // Further validate axes.\n    for (const axis of this.axis) {\n      if (axis < 0 || axis >= nDims) {\n        throw new Error(`Invalid axis: ${axis}`);\n      }\n    }\n    if (this.axis.length !== generic_utils.unique(this.axis).length) {\n      throw new Error(`Found duplicate axes in: ${this.axis}`);\n    }\n\n    const paramShape = this.axis.map(axis => inputShape[axis]) as number[];\n\n    const trainable = true;\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', paramShape, 'float32', this.gammaInitializer,\n          this.gammaRegularizer, trainable);\n    } else {\n      this.gamma = null;\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', paramShape, 'float32', this.betaInitializer,\n          this.betaRegularizer, trainable);\n    } else {\n      this.beta = null;\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const input = getExactlyOneTensor(inputs);\n    const inputShape = input.shape;\n    const nDims = inputShape.length;\n\n    return tidy(() => {\n      const keepDims = true;\n      let {mean, variance} = moments(input, this.axis, keepDims);\n      const broadcastShape = generic_utils.pyListRepeat(1, nDims);\n      for (const dim of this.axis as number[]) {\n        broadcastShape[dim] = inputShape[dim];\n      }\n\n      const broadcast = (v: Tensor) => {\n        if (v != null && v.shape.length !== nDims &&\n            this.axis !== [nDims - 1]) {\n          return v.reshape(broadcastShape);\n        } else {\n          return v;\n        }\n      };\n\n      let scale = broadcast(this.gamma.read());\n      let offset = broadcast(this.beta.read());\n\n      // TODO(https://github.com/tensorflow/tfjs/issues/2120): The tiling below\n      // is a workaround for the limitation of core's batchNormalization?d don't\n      // support broadcasting in their gradients. In addition, the tiling is\n      // necessary to ensure correctness on the browser CPU backend regardless\n      // of forward or backward computation. Remove this workaround once the\n      // limitation is addressed. See .\n      const momentsTiling: number[] = [];\n      const scaleOffsetTiling: number[] = [];\n      for (let i = 0; i < nDims; ++i) {\n        if ((this.axis as number[]).indexOf(i) !== -1) {\n          momentsTiling.push(inputShape[i]);\n          scaleOffsetTiling.push(1);\n        } else {\n          momentsTiling.push(1);\n          scaleOffsetTiling.push(inputShape[i]);\n        }\n      }\n      mean = mean.tile(momentsTiling);\n      variance = variance.tile(momentsTiling);\n      scale = scale.tile(scaleOffsetTiling);\n      offset = offset.tile(scaleOffsetTiling);\n\n      return batchNormalization(\n          input, mean, variance, offset, scale, this.epsilon);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LayerNormalization);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Padding Layers.\n */\n\n// Porting Note: In Python Keras, the padding layers are in convolutional.py,\n//   but we decided to put them in a separate file (padding.ts) for clarity.\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\n/**\n * Pads the middle dimension of a 3D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of 2 integers, how many zeros to add at the start and\n *   end of the middle dimension (i.e., dimension 1).\n * @return A padded 3D `tf.Tensor`.\n */\nexport function temporalPadding(x: Tensor, padding?: [number, number]): Tensor {\n  return tidy(() => {\n    if (x.rank !== 3) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 3-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [1, 1];\n    }\n    if (padding.length !== 2) {\n      throw new ValueError(\n          `temporalPadding expects input padding pattern to be a length-2 ` +\n          `array, but received a length-${padding.length} array.`);\n    }\n\n    const pattern: Array<[number, number]> = [[0, 0], padding, [0, 0]];\n    return tfc.pad(x, pattern);\n  });\n}\n\n/**\n * Pads the 2nd and 3rd dimensions of a 4D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of two `Array`s, each of which is an `Array` of two\n *   integers. The amount of padding at the beginning and end of the 2nd and 3rd\n *   dimensions, respectively.\n * @param dataFormat 'channelsLast' (default) or 'channelsFirst'.\n * @return Padded 4D `tf.Tensor`.\n */\nexport function spatial2dPadding(\n    x: Tensor, padding?: [[number, number], [number, number]],\n    dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 4-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [[1, 1], [1, 1]];\n    }\n    if (padding.length !== 2 || padding[0].length !== 2 ||\n        padding[1].length !== 2) {\n      throw new ValueError(\n          'spatial2dPadding expects `padding` to be an Array of two Arrays, ' +\n          'each of which is an Array of two integers.');\n    }\n\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (dataFormat !== 'channelsLast' && dataFormat !== 'channelsFirst') {\n      throw new ValueError(\n          `Unknown data format: ${dataFormat}. ` +\n          `Supported data formats are 'channelsLast' and 'channelsFirst.`);\n    }\n\n    let pattern: Array<[number, number]>;\n    if (dataFormat === 'channelsFirst') {\n      pattern = [[0, 0], [0, 0], padding[0], padding[1]];\n    } else {\n      pattern = [[0, 0], padding[0], padding[1], [0, 0]];\n    }\n\n    return tfc.pad(x, pattern);\n  });\n}\n\nexport declare interface ZeroPadding2DLayerArgs extends LayerArgs {\n  /**\n   * Integer, or `Array` of 2 integers, or `Array` of 2 `Array`s, each of\n   * which is an `Array` of 2 integers.\n   * - If integer, the same symmetric padding is applied to width and height.\n   * - If Array` of 2 integers, interpreted as two different symmetric values\n   *   for height and width:\n   *   `[symmetricHeightPad, symmetricWidthPad]`.\n   * - If `Array` of 2 `Array`s, interpreted as:\n   *   `[[topPad, bottomPad], [leftPad, rightPad]]`.\n   */\n  padding?: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * One of `'channelsLast'` (default) and `'channelsFirst'`.\n   *\n   * The ordering of the dimensions in the inputs.\n   * `channelsLast` corresponds to inputs with shape\n   * `[batch, height, width, channels]` while `channelsFirst`\n   * corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class ZeroPadding2D extends Layer {\n  /** @nocollapse */\n  static className = 'ZeroPadding2D';\n  readonly dataFormat: DataFormat;\n  readonly padding: [[number, number], [number, number]];\n\n  constructor(args?: ZeroPadding2DLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.dataFormat =\n        args.dataFormat == null ? imageDataFormat() : args.dataFormat;\n    // TODO(cais): Maybe refactor the following logic surrounding `padding`\n    //   into a helper method.\n    if (args.padding == null) {\n      this.padding = [[1, 1], [1, 1]];\n    } else if (typeof args.padding === 'number') {\n      this.padding =\n          [[args.padding, args.padding], [args.padding, args.padding]];\n    } else {\n      args.padding = args.padding;\n      if (args.padding.length !== 2) {\n        throw new ValueError(\n            `ZeroPadding2D expects padding to be a length-2 array, but ` +\n            `received a length-${args.padding.length} array.`);\n      }\n\n      let heightPadding: [number, number];\n      let widthPadding: [number, number];\n      if (typeof args.padding[0] === 'number') {\n        heightPadding = [args.padding[0], args.padding[0]];\n        widthPadding = [args.padding[1] as number, args.padding[1] as number];\n      } else {\n        args.padding = args.padding as [[number, number], [number, number]];\n\n        if (args.padding[0].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects height padding to be a length-2 array, ` +\n              `but received a length-${args.padding[0].length} array.`);\n        }\n        heightPadding = args.padding[0] as [number, number];\n\n        if (args.padding[1].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects width padding to be a length-2 array, ` +\n              `but received a length-${args.padding[1].length} array.`);\n        }\n        widthPadding = args.padding[1] as [number, number];\n      }\n      this.padding = [heightPadding, widthPadding];\n    }\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n\n    let rows: number;\n    let cols: number;\n    if (this.dataFormat === 'channelsFirst') {\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        rows = inputShape[2] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[3] != null && inputShape[3] >= 0) {\n        cols = inputShape[3] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      if (inputShape[1] != null && inputShape[1] >= 0) {\n        rows = inputShape[1] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        cols = inputShape[2] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(\n        () => spatial2dPadding(\n            getExactlyOneTensor(inputs), this.padding, this.dataFormat));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ZeroPadding2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Pooling Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor3D, Tensor4D, Tensor5D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode, checkPoolMode} from '../common';\nimport {InputSpec} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, PaddingMode, PoolMode, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\nimport {preprocessConv2DInput, preprocessConv3DInput} from './convolutional';\n\n/**\n * 2D pooling.\n * @param x\n * @param poolSize\n * @param stridesdes strides. Defaults to [1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 2D pooling.\n */\nexport function pool2d(\n    x: Tensor, poolSize: [number, number], strides?: [number, number],\n    padding?: PaddingMode, dataFormat?: DataFormat,\n    poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // TODO(cais): Remove the preprocessing step once deeplearn.js supports\n    // dataFormat as an input argument.\n    x = preprocessConv2DInput(x, dataFormat);  // x is NHWC after preprocessing.\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      // TODO(cais): Rank check?\n      y = tfc.maxPool(x as Tensor4D, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      // TODO(cais): Check the dtype and rank of x and give clear error message\n      //   if those are incorrect.\n      y = tfc.avgPool(\n          // TODO(cais): Rank check?\n          x as Tensor3D | Tensor4D, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);  // NHWC -> NCHW.\n    }\n    return y;\n  });\n}\n\n/**\n * 3D pooling.\n * @param x\n * @param poolSize. Default to [1, 1, 1].\n * @param strides strides. Defaults to [1, 1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 3D pooling.\n */\nexport function pool3d(\n    x: Tensor5D, poolSize: [number, number, number],\n    strides?: [number, number, number], padding?: PaddingMode,\n    dataFormat?: DataFormat, poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // x is NDHWC after preprocessing.\n    x = preprocessConv3DInput(x as Tensor, dataFormat) as Tensor5D;\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      y = tfc.maxPool3d(x, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      y = tfc.avgPool3d(x, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);  // NDHWC -> NCDHW.\n    }\n    return y;\n  });\n}\n\nexport declare interface Pooling1DLayerArgs extends LayerArgs {\n  /**\n   * Size of the window to pool over, should be an integer.\n   */\n  poolSize?: number|[number];\n  /**\n   * Period at which to sample the pooled values.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number];\n  /** How to fill in data that's not an integer multiple of poolSize. */\n  padding?: PaddingMode;\n}\n\n/**\n * Abstract class for different pooling 1D layers.\n */\nexport abstract class Pooling1D extends Layer {\n  protected readonly poolSize: [number];\n  protected readonly strides: [number];\n  protected readonly padding: PaddingMode;\n\n  /**\n   *\n   * @param args Parameters for the Pooling layer.\n   *\n   * config.poolSize defaults to 2.\n   */\n  constructor(args: Pooling1DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = 2;\n    }\n    super(args);\n    if (typeof args.poolSize === 'number') {\n      this.poolSize = [args.poolSize];\n    } else if (\n        Array.isArray(args.poolSize) &&\n        (args.poolSize as number[]).length === 1 &&\n        typeof (args.poolSize as number[])[0] === 'number') {\n      this.poolSize = args.poolSize;\n    } else {\n      throw new ValueError(\n          `poolSize for 1D convolutional layer must be a number or an ` +\n          `Array of a single number, but received ` +\n          `${JSON.stringify(args.poolSize)}`);\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else {\n      if (typeof args.strides === 'number') {\n        this.strides = [args.strides];\n      } else if (\n          Array.isArray(args.strides) &&\n          (args.strides as number[]).length === 1 &&\n          typeof (args.strides as number[])[0] === 'number') {\n        this.strides = args.strides;\n      } else {\n        throw new ValueError(\n            `strides for 1D convolutional layer must be a number or an ` +\n            `Array of a single number, but received ` +\n            `${JSON.stringify(args.strides)}`);\n      }\n    }\n    assertPositiveInteger(this.strides, 'strides');\n\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const length = convOutputLength(\n        inputShape[1], this.poolSize[0], this.padding, this.strides[0]);\n    return [inputShape[0], length, inputShape[2]];\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Add dummy last dimension.\n      inputs = K.expandDims(getExactlyOneTensor(inputs), 2);\n      const output = this.poolingFunction(\n          getExactlyOneTensor(inputs), [this.poolSize[0], 1],\n          [this.strides[0], 1], this.padding, 'channelsLast');\n      // Remove dummy last dimension.\n      return tfc.squeeze(output, [2]);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'MaxPooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling1D);\n\nexport class AveragePooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'AveragePooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling1D);\n\nexport declare interface Pooling2DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [vertical, horizontal].\n   * Expects an integer or an array of 2 integers.\n   *\n   * For example, `[2, 2]` will halve the input in both spatial dimension.\n   * If only one integer is specified, the same window length\n   * will be used for both dimensions.\n   */\n  poolSize?: number|[number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 2 integers. Integer, tuple of 2 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 2D layers.\n */\nexport abstract class Pooling2D extends Layer {\n  protected readonly poolSize: [number, number];\n  protected readonly strides: [number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling2DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 2) {\n        throw new ValueError(\n            `If the strides property of a 2D pooling layer is an Array, ` +\n            `it is expected to have a length of 2, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    rows =\n        convOutputLength(rows, this.poolSize[0], this.padding, this.strides[0]);\n    cols =\n        convOutputLength(cols, this.poolSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'MaxPooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling2D);\n\nexport class AveragePooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'AveragePooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling2D);\n\nexport declare interface Pooling3DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [depth, height, width].\n   * Expects an integer or an array of 3 integers.\n   *\n   * For example, `[2, 2, 2]` will halve the input in three dimensions.\n   * If only one integer is specified, the same window length\n   * will be used for all dimensions.\n   */\n  poolSize?: number|[number, number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 3 integers. Integer, tuple of 3 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 3D layers.\n */\nexport abstract class Pooling3D extends Layer {\n  protected readonly poolSize: [number, number, number];\n  protected readonly strides: [number, number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling3DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 3) {\n        throw new ValueError(\n            `If the strides property of a 3D pooling layer is an Array, ` +\n            `it is expected to have a length of 3, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let depths =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[4] : inputShape[3];\n    depths = convOutputLength(\n        depths, this.poolSize[0], this.padding, this.strides[0]);\n    rows =\n        convOutputLength(rows, this.poolSize[1], this.padding, this.strides[1]);\n    cols =\n        convOutputLength(cols, this.poolSize[2], this.padding, this.strides[2]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], depths, rows, cols];\n    } else {\n      return [inputShape[0], depths, rows, cols, inputShape[4]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'MaxPooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling3D);\n\nexport class AveragePooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'AveragePooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling3D);\n\n/**\n * Abstract class for different global pooling 1D layers.\n */\nexport abstract class GlobalPooling1D extends Layer {\n  constructor(args: LayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], inputShape[2]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n}\n\nexport class GlobalAveragePooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling1D';\n  constructor(args?: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.mean(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling1D);\n\nexport class GlobalMaxPooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling1D';\n  constructor(args: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.max(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling1D);\n\nexport declare interface GlobalPooling2DLayerArgs extends LayerArgs {\n  /**\n   * One of `CHANNEL_LAST` (default) or `CHANNEL_FIRST`.\n   *\n   * The ordering of the dimensions in the inputs. `CHANNEL_LAST` corresponds\n   * to inputs with shape `[batch, height, width, channels[` while\n   * `CHANNEL_FIRST` corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different global pooling 2D layers.\n */\nexport abstract class GlobalPooling2D extends Layer {\n  protected dataFormat: DataFormat;\n  constructor(args: GlobalPooling2DLayerArgs) {\n    super(args);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape;\n    if (this.dataFormat === 'channelsLast') {\n      return [inputShape[0], inputShape[3]];\n    } else {\n      return [inputShape[0], inputShape[1]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class GlobalAveragePooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.mean(input, [1, 2]);\n      } else {\n        return tfc.mean(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling2D);\n\nexport class GlobalMaxPooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.max(input, [1, 2]);\n      } else {\n        return tfc.max(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Layers that augment the functionality of a base layer.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {InputSpec, Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {BidirectionalMergeMode, Shape, VALID_BIDIRECTIONAL_MERGE_MODES} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {RegularizerFn, RnnStepFunction} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {rnn, RNN, standardizeArgs} from './recurrent';\nimport {deserialize} from './serialization';\n\nexport declare interface WrapperLayerArgs extends LayerArgs {\n  /**\n   * The layer to be wrapped.\n   */\n  layer: Layer;\n}\n\n/**\n * Abstract wrapper base class.\n *\n * Wrappers take another layer and augment it in various ways.\n * Do not use this class as a layer, it is only an abstract base class.\n * Two usable wrappers are the `TimeDistributed` and `Bidirectional` wrappers.\n */\nexport abstract class Wrapper extends Layer {\n  readonly layer: Layer;\n\n  constructor(args: WrapperLayerArgs) {\n    // Porting Note: In PyKeras, `self.layer` is set prior to the calling\n    //   `super()`. But we can't do that here due to TypeScript's restriction.\n    //   See: https://github.com/Microsoft/TypeScript/issues/8277\n    //   As a result, we have to add checks in `get trainable()` and\n    //   `set trainable()` below in order to prevent using `this.layer` when\n    //   its value is `undefined`. The super constructor does use the getter\n    //   and the setter of `this.layer`.\n    super(args);\n    this.layer = args.layer;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    this.built = true;\n  }\n\n  // TODO(cais): Implement activityRegularizer getter.\n\n  get trainable(): boolean {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      return this.layer.trainable;\n    } else {\n      return false;\n    }\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      this.layer.trainable = value;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.layer.trainableWeights;\n  }\n  // TODO(cais): Implement setter for trainableWeights.\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.layer.nonTrainableWeights;\n  }\n  // TODO(cais): Implement setter for nonTrainableWeights.\n\n  get updates(): Tensor[] {\n    // tslint:disable-next-line:no-any\n    return (this.layer as any)._updates;\n  }\n\n  // TODO(cais): Implement getUpdatesFor().\n\n  get losses(): RegularizerFn[] {\n    return this.layer.losses;\n  }\n\n  // TODO(cais): Implement getLossesFor().\n\n  getWeights(): Tensor[] {\n    return this.layer.getWeights();\n  }\n\n  setWeights(weights: Tensor[]): void {\n    this.layer.setWeights(weights);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'layer': {\n        'className': this.layer.getClassName(),\n        'config': this.layer.getConfig(),\n      }\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.layer != null) {\n      this.layer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const layerConfig = config['layer'] as serialization.ConfigDict;\n    const layer = deserialize(layerConfig, customObjects) as Layer;\n    delete config['layer'];\n    const newConfig = {layer};\n    Object.assign(newConfig, config);\n    return new cls(newConfig);\n  }\n}\n\nexport class TimeDistributed extends Wrapper {\n  /** @nocollapse */\n  static className = 'TimeDistributed';\n  constructor(args: WrapperLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 3) {\n      throw new ValueError(\n          `TimeDistributed layer expects an input shape >= 3D, but received ` +\n          `input shape ${JSON.stringify(inputShape)}`);\n    }\n    this.inputSpec = [{shape: inputShape}];\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (!this.layer.built) {\n      this.layer.build(childInputShape);\n      this.layer.built = true;\n    }\n    super.build(inputShape);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    const childOutputShape =\n        this.layer.computeOutputShape(childInputShape) as Shape;\n    const timesteps = inputShape[1];\n    return [childOutputShape[0], timesteps].concat(childOutputShape.slice(1));\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      // TODO(cais): Add 'training' and 'useLearningPhase' to kwargs.\n      inputs = getExactlyOneTensor(inputs);\n      // Porting Note: In tfjs-layers, `inputs` are always concrete tensor\n      // values. Hence the inputs can't have an undetermined first (batch)\n      // dimension, which is why we always use the K.rnn approach here.\n      const step: RnnStepFunction = (inputs: Tensor, states: Tensor[]) => {\n        // TODO(cais): Add useLearningPhase.\n        // NOTE(cais): `layer.call` may return a length-1 array of Tensor in\n        //   some cases (e.g., `layer` is a `Sequential` instance), which is\n        //   why `getExactlyOneTensor` is used below.\n        const output = getExactlyOneTensor(this.layer.call(inputs, kwargs));\n        return [output, []];\n      };\n      const rnnOutputs =\n          rnn(step, inputs, [], false /* goBackwards */, null /* mask */,\n              null /* constants */, false /* unroll */,\n              true /* needPerStepOutputs */);\n      const y = rnnOutputs[1];\n      // TODO(cais): Add activity regularization.\n      // TODO(cais): Add useLearningPhase.\n      return y;\n    });\n  }\n\n  // TODO(cais): Implement detailed computeMask() logic.\n}\nserialization.registerClass(TimeDistributed);\n\nexport function checkBidirectionalMergeMode(value?: string): void {\n  generic_utils.checkStringTypeUnionValue(\n      VALID_BIDIRECTIONAL_MERGE_MODES, 'BidirectionalMergeMode', value);\n}\n\nexport declare interface BidirectionalLayerArgs extends WrapperLayerArgs {\n  /**\n   * The instance of an `RNN` layer to be wrapped.\n   */\n  layer: RNN;\n\n  /**\n   * Mode by which outputs of the forward and backward RNNs are\n   * combined. If `null` or `undefined`, the output will not be\n   * combined, they will be returned as an `Array`.\n   *\n   * If `undefined` (i.e., not provided), defaults to `'concat'`.\n   */\n  mergeMode?: BidirectionalMergeMode;\n}\n\nconst DEFAULT_BIDIRECTIONAL_MERGE_MODE: BidirectionalMergeMode = 'concat';\n\nexport class Bidirectional extends Wrapper {\n  /** @nocollapse */\n  static className = 'Bidirectional';\n  mergeMode: BidirectionalMergeMode;\n  private forwardLayer: RNN;\n  private backwardLayer: RNN;\n  private returnSequences: boolean;\n  private returnState: boolean;\n  private numConstants?: number;\n  private _trainable: boolean;\n\n  constructor(args: BidirectionalLayerArgs) {\n    super(args);\n\n    // Note: When creating `this.forwardLayer`, the original Layer object\n    //   (`config.layer`) ought to be cloned. This is why we call\n    //   `getConfig()` followed by `deserialize()`. Without this cloning,\n    //   the layer names saved during serialization will incorrectly contain\n    //   the 'forward_' prefix. In Python Keras, this is done using\n    //   `copy.copy` (shallow copy), which does not have a simple equivalent\n    //   in JavaScript. JavaScript's `Object.assign()` does not copy\n    //   methods.\n    const layerConfig = args.layer.getConfig();\n    const forwDict: serialization.ConfigDict = {};\n    forwDict['className'] = args.layer.getClassName();\n    forwDict['config'] = layerConfig;\n    this.forwardLayer = deserialize(forwDict) as RNN;\n    layerConfig['goBackwards'] =\n        layerConfig['goBackwards'] === true ? false : true;\n    const backDict: serialization.ConfigDict = {};\n    backDict['className'] = args.layer.getClassName();\n    backDict['config'] = layerConfig;\n    this.backwardLayer = deserialize(backDict) as RNN;\n    this.forwardLayer.name = 'forward_' + this.forwardLayer.name;\n    this.backwardLayer.name = 'backward_' + this.backwardLayer.name;\n\n    this.mergeMode = args.mergeMode === undefined ?\n        DEFAULT_BIDIRECTIONAL_MERGE_MODE :\n        args.mergeMode;\n    checkBidirectionalMergeMode(this.mergeMode);\n    if (args.weights) {\n      throw new NotImplementedError(\n          'weights support is not implemented for Bidirectional layer yet.');\n    }\n    this._stateful = args.layer.stateful;\n    this.returnSequences = args.layer.returnSequences;\n    this.returnState = args.layer.returnState;\n    this.supportsMasking = true;\n    this._trainable = true;\n    this.inputSpec = args.layer.inputSpec;\n    this.numConstants = null;\n  }\n\n  get trainable(): boolean {\n    return this._trainable;\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    this._trainable = value;\n    if (this.forwardLayer != null) {\n      this.forwardLayer.trainable = value;\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.trainable = value;\n    }\n  }\n\n  getWeights(): Tensor[] {\n    return this.forwardLayer.getWeights().concat(\n        this.backwardLayer.getWeights());\n  }\n\n  setWeights(weights: Tensor[]): void {\n    const numWeights = weights.length;\n    const numeightsOver2 = Math.floor(numWeights / 2);\n    this.forwardLayer.setWeights(weights.slice(0, numeightsOver2));\n    this.backwardLayer.setWeights(weights.slice(numeightsOver2));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    let layerShapes: Shape|Shape[] =\n        this.forwardLayer.computeOutputShape(inputShape);\n    if (!(Array.isArray(layerShapes) && Array.isArray(layerShapes[0]))) {\n      layerShapes = [layerShapes as Shape];\n    }\n    layerShapes = layerShapes as Shape[];\n\n    let outputShape: Shape;\n    let outputShapes: Shape[];\n    let stateShape: Shape[];\n    if (this.returnState) {\n      stateShape = layerShapes.slice(1);\n      outputShape = layerShapes[0];\n    } else {\n      outputShape = layerShapes[0];\n    }\n    outputShape = outputShape;\n    if (this.mergeMode === 'concat') {\n      outputShape[outputShape.length - 1] *= 2;\n      outputShapes = [outputShape];\n    } else if (this.mergeMode == null) {\n      outputShapes = [outputShape, outputShape.slice()];\n    } else {\n      outputShapes = [outputShape];\n    }\n\n    if (this.returnState) {\n      if (this.mergeMode == null) {\n        return outputShapes.concat(stateShape).concat(stateShape.slice());\n      }\n      return [outputShape].concat(stateShape).concat(stateShape.slice());\n    }\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    if (Array.isArray(inputs)) {\n      initialState = (inputs as Tensor[] | SymbolicTensor[]).slice(1);\n      inputs = (inputs as Tensor[] | SymbolicTensor[])[0];\n    }\n\n    if ((initialState == null || initialState.length === 0) &&\n        constants == null) {\n      return super.apply(inputs, kwargs);\n    }\n    const additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    const additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      const numStates = initialState.length;\n      if (numStates % 2 > 0) {\n        throw new ValueError(\n            'When passing `initialState` to a Bidrectional RNN, ' +\n            'the state should be an Array containing the states of ' +\n            'the underlying RNNs.');\n      }\n      kwargs['initialState'] = initialState;\n      additionalInputs.push(...initialState);\n      const stateSpecs = (initialState as Array<Tensor|SymbolicTensor>)\n                             .map(state => new InputSpec({shape: state.shape}));\n      this.forwardLayer.stateSpec = stateSpecs.slice(0, numStates / 2);\n      this.backwardLayer.stateSpec = stateSpecs.slice(numStates / 2);\n      additionalSpecs.push(...stateSpecs);\n    }\n    if (constants != null) {\n      throw new NotImplementedError(\n          'Support for constants in Bidirectional layers is not ' +\n          'implemented yet.');\n    }\n\n    const isSymbolicTensor = additionalInputs[0] instanceof SymbolicTensor;\n    for (const tensor of additionalInputs) {\n      if (tensor instanceof SymbolicTensor !== isSymbolicTensor) {\n        throw new ValueError(\n            'The initial state of a Bidirectional layer cannot be ' +\n            'specified as a mix of symbolic and non-symbolic tensors');\n      }\n    }\n\n    if (isSymbolicTensor) {\n      // Compute the full input and specs, including the states.\n      const fullInput = [inputs].concat(additionalInputs);\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call temporarily and replace inputSpec.\n      // Note: with initial states symbolic calls and non-symbolic calls to\n      // this method differ in how the initial states are passed. For\n      // symbolic calls, the initial states are passed in the first arg, as\n      // an Array of SymbolicTensors; for non-symbolic calls, they are\n      // passed in the second arg as a part of the kwargs. Hence the need to\n      // temporarily modify inputSpec here.\n      // TODO(cais): Make refactoring so that this hacky code below is no\n      // longer needed.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output =\n          super.apply(fullInput as Tensor[] | SymbolicTensor[], kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const initialState = kwargs['initialState'];\n\n      let y: Tensor|Tensor[];\n      let yRev: Tensor|Tensor[];\n      if (initialState == null) {\n        y = this.forwardLayer.call(inputs, kwargs);\n        yRev = this.backwardLayer.call(inputs, kwargs);\n      } else {\n        const forwardState = initialState.slice(0, initialState.length / 2);\n        const backwardState = initialState.slice(initialState.length / 2);\n        y = this.forwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: forwardState}));\n        yRev = this.backwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: backwardState}));\n      }\n\n      let states: Tensor[];\n      if (this.returnState) {\n        if (Array.isArray(y)) {\n          states = y.slice(1).concat((yRev as Tensor[]).slice(1));\n        } else {\n        }\n        y = (y as Tensor[])[0];\n        yRev = (yRev as Tensor[])[0];\n      }\n\n      if (this.returnSequences) {\n        yRev = tfc.reverse(yRev as Tensor, 1);\n      }\n\n      let output: Tensor|Tensor[];\n      if (this.mergeMode === 'concat') {\n        output = K.concatenate([y as Tensor, yRev as Tensor]);\n      } else if (this.mergeMode === 'sum') {\n        output = tfc.add(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode === 'ave') {\n        output = tfc.mul(.5, tfc.add(y as Tensor, yRev as Tensor));\n      } else if (this.mergeMode === 'mul') {\n        output = tfc.mul(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode == null) {\n        output = [y as Tensor, yRev as Tensor];\n      }\n\n      // TODO(cais): Properly set learning phase.\n      if (this.returnState) {\n        if (this.mergeMode == null) {\n          return (output as Tensor[]).concat(states);\n        }\n        return [output as Tensor].concat(states);\n      }\n      return output;\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[]): void {\n    this.forwardLayer.resetStates();\n    this.backwardLayer.resetStates();\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    nameScope(this.forwardLayer.name, () => {\n      this.forwardLayer.build(inputShape);\n    });\n    nameScope(this.backwardLayer.name, () => {\n      this.backwardLayer.build(inputShape);\n    });\n    this.built = true;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (Array.isArray(mask)) {\n      mask = mask[0];\n    }\n    let outputMask: Tensor|Tensor[];\n    if (this.returnSequences) {\n      if (this.mergeMode == null) {\n        outputMask = [mask, mask];\n      } else {\n        outputMask = mask;\n      }\n    } else {\n      if (this.mergeMode == null) {\n        outputMask = [null, null];\n      } else {\n        outputMask = null;\n      }\n    }\n    if (this.returnState) {\n      const states = this.forwardLayer.states;\n      const stateMask: Tensor[] = states.map(state => null);\n      if (Array.isArray(outputMask)) {\n        return outputMask.concat(stateMask).concat(stateMask);\n      } else {\n        return [outputMask].concat(stateMask).concat(stateMask);\n      }\n    } else {\n      return outputMask;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.forwardLayer.trainableWeights.concat(\n        this.backwardLayer.trainableWeights);\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.forwardLayer.nonTrainableWeights.concat(\n        this.backwardLayer.nonTrainableWeights);\n  }\n\n  // TODO(cais): Implement constraints().\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.forwardLayer != null) {\n      this.forwardLayer.setFastWeightInitDuringBuild(value);\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'mergeMode': this.mergeMode,\n    };\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    const rnnLayer =\n        deserialize(config['layer'] as serialization.ConfigDict) as RNN;\n    delete config['layer'];\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    if (config['numConstants'] != null) {\n      throw new NotImplementedError(\n          `Deserialization of a Bidirectional layer with numConstants ` +\n          `present is not supported yet.`);\n    }\n    // tslint:disable-next-line:no-any\n    const newConfig: {[key: string]: any} = config;\n    newConfig['layer'] = rnnLayer;\n    return new cls(newConfig);\n  }\n}\nserialization.registerClass(Bidirectional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {InputLayer, InputLayerArgs} from './engine/input_layer';\nimport {Layer, LayerArgs} from './engine/topology';\nimport {input} from './exports';\nimport {ELU, ELULayerArgs, LeakyReLU, LeakyReLULayerArgs, PReLU, PReLULayerArgs, ReLU, ReLULayerArgs, Softmax, SoftmaxLayerArgs, ThresholdedReLU, ThresholdedReLULayerArgs} from './layers/advanced_activations';\nimport {Conv1D, Conv2D, Conv2DTranspose, Conv3D, ConvLayerArgs, Cropping2D, Cropping2DLayerArgs, SeparableConv2D, SeparableConvLayerArgs, UpSampling2D, UpSampling2DLayerArgs} from './layers/convolutional';\nimport {DepthwiseConv2D, DepthwiseConv2DLayerArgs} from './layers/convolutional_depthwise';\nimport {ConvLSTM2D, ConvLSTM2DArgs, ConvLSTM2DCell, ConvLSTM2DCellArgs} from './layers/convolutional_recurrent';\nimport {Activation, ActivationLayerArgs, Dense, DenseLayerArgs, Dropout, DropoutLayerArgs, Flatten, FlattenLayerArgs, Masking, MaskingArgs, Permute, PermuteLayerArgs, RepeatVector, RepeatVectorLayerArgs, Reshape, ReshapeLayerArgs, SpatialDropout1D, SpatialDropout1DLayerConfig} from './layers/core';\nimport {Embedding, EmbeddingLayerArgs} from './layers/embeddings';\nimport {Add, Average, Concatenate, ConcatenateLayerArgs, Dot, DotLayerArgs, Maximum, Minimum, Multiply} from './layers/merge';\nimport {AlphaDropout, AlphaDropoutArgs, GaussianDropout, GaussianDropoutArgs, GaussianNoise, GaussianNoiseArgs} from './layers/noise';\nimport {BatchNormalization, BatchNormalizationLayerArgs, LayerNormalization, LayerNormalizationLayerArgs} from './layers/normalization';\nimport {ZeroPadding2D, ZeroPadding2DLayerArgs} from './layers/padding';\nimport {AveragePooling1D, AveragePooling2D, AveragePooling3D, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalPooling2DLayerArgs, MaxPooling1D, MaxPooling2D, MaxPooling3D, Pooling1DLayerArgs, Pooling2DLayerArgs, Pooling3DLayerArgs} from './layers/pooling';\nimport {GRU, GRUCell, GRUCellLayerArgs, GRULayerArgs, LSTM, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNN, SimpleRNNCell, SimpleRNNCellLayerArgs, SimpleRNNLayerArgs, StackedRNNCells, StackedRNNCellsArgs} from './layers/recurrent';\nimport {Bidirectional, BidirectionalLayerArgs, TimeDistributed, WrapperLayerArgs} from './layers/wrappers';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// Input Layer.\n/**\n * An input layer is an entry point into a `tf.LayersModel`.\n *\n * `InputLayer` is generated automatically for `tf.Sequential`` models by\n * specifying the `inputshape` or `batchInputShape` for the first layer.  It\n * should not be specified explicitly. However, it can be useful sometimes,\n * e.g., when constructing a sequential model from a subset of another\n * sequential model's layers. Like the code snippet below shows.\n *\n * ```js\n * // Define a model which simply adds two inputs.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));\n * model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n * model1.summary();\n * model1.predict(tf.zeros([1, 4])).print();\n *\n * // Construct another model, reusing the second layer of `model1` while\n * // not using the first layer of `model1`. Note that you cannot add the second\n * // layer of `model` directly as the first layer of the new sequential model,\n * // because doing so will lead to an error related to the fact that the layer\n * // is not an input layer. Instead, you need to create an `inputLayer` and add\n * // it to the new sequential model before adding the reused layer.\n * const model2 = tf.sequential();\n * // Use an inputShape that matches the input shape of `model1`'s second\n * // layer.\n * model2.add(tf.layers.inputLayer({inputShape: [3]}));\n * model2.add(model1.layers[1]);\n * model2.summary();\n * model2.predict(tf.zeros([1, 3])).print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}\n */\nexport function inputLayer(args: InputLayerArgs): Layer {\n  return new InputLayer(args);\n}\n\n// Advanced Activation Layers.\n\n/**\n * Exponetial Linear Unit (ELU).\n *\n * It follows:\n * `f(x) =  alpha * (exp(x) - 1.) for x < 0`,\n * `f(x) = x for x >= 0`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Fast and Accurate Deep Network Learning by Exponential Linear Units\n * (ELUs)](https://arxiv.org/abs/1511.07289v1)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function elu(args?: ELULayerArgs): Layer {\n  return new ELU(args);\n}\n\n/**\n * Rectified Linear Unit activation function.\n *\n * Input shape:\n *   Arbitrary. Use the config field `inputShape` (Array of integers, does\n *   not include the sample axis) when using this layer as the first layer\n *   in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function reLU(args?: ReLULayerArgs): Layer {\n  return new ReLU(args);\n}\n\n/**\n * Leaky version of a rectified linear unit.\n *\n * It allows a small gradient when the unit is not active:\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function leakyReLU(args?: LeakyReLULayerArgs): Layer {\n  return new LeakyReLU(args);\n}\n\n/**\n * Parameterized version of a leaky rectified linear unit.\n *\n * It follows\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n * wherein `alpha` is a trainable weight.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function prelu(args?: PReLULayerArgs): Layer {\n  return new PReLU(args);\n}\n\n/**\n * Softmax activation layer.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function softmax(args?: SoftmaxLayerArgs): Layer {\n  return new Softmax(args);\n}\n\n/**\n * Thresholded Rectified Linear Unit.\n *\n * It follows:\n * `f(x) = x for x > theta`,\n * `f(x) = 0 otherwise`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Zero-Bias Autoencoders and the Benefits of Co-Adapting\n * Features](http://arxiv.org/abs/1402.3337)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function thresholdedReLU(args?: ThresholdedReLULayerArgs): Layer {\n  return new ThresholdedReLU(args);\n}\n\n// Convolutional Layers.\n\n/**\n * 1D convolution layer (e.g., temporal convolution).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input over a single spatial (or temporal) dimension\n * to produce a tensor of outputs.\n *\n * If `use_bias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model, provide an\n * `inputShape` argument `Array` or `null`.\n *\n * For example, `inputShape` would be:\n * - `[10, 128]` for sequences of 10 vectors of 128-dimensional vectors\n * - `[null, 128]` for variable-length sequences of 128-dimensional vectors.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional',  namespace: 'layers'}\n */\nexport function conv1d(args: ConvLayerArgs): Layer {\n  return new Conv1D(args);\n}\n\n/**\n * 2D convolution layer (e.g. spatial convolution over images).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 3]` for 128x128 RGB pictures\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2d(args: ConvLayerArgs): Layer {\n  return new Conv2D(args);\n}\n\n/**\n * Transposed convolutional layer (sometimes called Deconvolution).\n *\n * The need for transposed convolutions generally arises\n * from the desire to use a transformation going in the opposite direction of\n * a normal convolution, i.e., from something that has the shape of the output\n * of some convolution to something that has the shape of its input while\n * maintaining a connectivity pattern that is compatible with said\n * convolution.\n *\n * When using this layer as the first layer in a model, provide the\n * configuration `inputShape` (`Array` of integers, does not include the\n * sample axis), e.g., `inputShape: [128, 128, 3]` for 128x128 RGB pictures in\n * `dataFormat: 'channelsLast'`.\n *\n * Input shape:\n *   4D tensor with shape:\n *   `[batch, channels, rows, cols]` if `dataFormat` is `'channelsFirst'`.\n *   or 4D tensor with shape\n *   `[batch, rows, cols, channels]` if `dataFormat` is `'channelsLast`.\n *\n * Output shape:\n *   4D tensor with shape:\n *   `[batch, filters, newRows, newCols]` if `dataFormat` is\n * `'channelsFirst'`. or 4D tensor with shape:\n *   `[batch, newRows, newCols, filters]` if `dataFormat` is `'channelsLast'`.\n *\n * References:\n *   - [A guide to convolution arithmetic for deep\n * learning](https://arxiv.org/abs/1603.07285v1)\n *   - [Deconvolutional\n * Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2dTranspose(args: ConvLayerArgs): Layer {\n  return new Conv2DTranspose(args);\n}\n\n/**\n * 3D convolution layer (e.g. spatial convolution over volumes).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 128, 1]` for 128x128x128 grayscale volumes\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv3d(args: ConvLayerArgs): Layer {\n  return new Conv3D(args);\n}\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Separable convolution consists of first performing\n * a depthwise spatial convolution\n * (which acts on each input channel separately)\n * followed by a pointwise convolution which mixes together the resulting\n * output channels. The `depthMultiplier` argument controls how many\n * output channels are generated per input channel in the depthwise step.\n *\n * Intuitively, separable convolutions can be understood as\n * a way to factorize a convolution kernel into two smaller kernels,\n * or as an extreme version of an Inception block.\n *\n * Input shape:\n *   4D tensor with shape:\n *     `[batch, channels, rows, cols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, rows, cols, channels]` if data_format='channelsLast'.\n *\n * Output shape:\n *   4D tensor with shape:\n *     `[batch, filters, newRows, newCols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, newRows, newCols, filters]` if data_format='channelsLast'.\n *     `rows` and `cols` values might have changed due to padding.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function separableConv2d(args: SeparableConvLayerArgs): Layer {\n  return new SeparableConv2D(args);\n}\n\n/**\n * Cropping layer for 2D input (e.g., image).\n *\n * This layer can crop an input\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, croppedRows, croppedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, croppedRows, croppedCols]`.\n *\n * Examples\n * ```js\n *\n * const model = tf.sequential();\n * model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]],\n *                                inputShape: [128, 128, 3]}));\n * //now output shape is [batch, 124, 124, 3]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function cropping2D(args: Cropping2DLayerArgs): Layer {\n  return new Cropping2D(args);\n}\n\n/**\n * Upsampling layer for 2D inputs.\n *\n * Repeats the rows and columns of the data\n * by size[0] and size[1] respectively.\n *\n *\n * Input shape:\n *    4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *         `[batch, rows, cols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *        `[batch, channels, rows, cols]`\n *\n * Output shape:\n *     4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *        `[batch, upsampledRows, upsampledCols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *         `[batch, channels, upsampledRows, upsampledCols]`\n *\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function upSampling2d(args: UpSampling2DLayerArgs): Layer {\n  return new UpSampling2D(args);\n}\n\n// Convolutional(depthwise) Layers.\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Depthwise Separable convolutions consists in performing just the first step\n * in a depthwise spatial convolution (which acts on each input channel\n * separately). The `depthMultplier` argument controls how many output channels\n * are generated per input channel in the depthwise step.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function depthwiseConv2d(args: DepthwiseConv2DLayerArgs): Layer {\n  return new DepthwiseConv2D(args);\n}\n\n// Basic Layers.\n\n/**\n * Applies an activation function to an output.\n *\n * This layer applies element-wise activation function.  Other layers, notably\n * `dense` can also apply activation functions.  Use this isolated activation\n * function to extract the values before and after the\n * activation. For instance:\n *\n * ```js\n * const input = tf.input({shape: [5]});\n * const denseLayer = tf.layers.dense({units: 1});\n * const activationLayer = tf.layers.activation({activation: 'relu6'});\n *\n * // Obtain the output symbolic tensors by applying the layers in order.\n * const denseOutput = denseLayer.apply(input);\n * const activationOutput = activationLayer.apply(denseOutput);\n *\n * // Create the model based on the inputs.\n * const model = tf.model({\n *     inputs: input,\n *     outputs: [denseOutput, activationOutput]\n * });\n *\n * // Collect both outputs and print separately.\n * const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));\n * denseOut.print();\n * activationOut.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function activation(args: ActivationLayerArgs): Layer {\n  return new Activation(args);\n}\n\n/**\n * Creates a dense (fully connected) layer.\n *\n * This layer implements the operation:\n *   `output = activation(dot(input, kernel) + bias)`\n *\n * `activation` is the element-wise activation function\n *   passed as the `activation` argument.\n *\n * `kernel` is a weights matrix created by the layer.\n *\n * `bias` is a bias vector created by the layer (only applicable if `useBias`\n * is `true`).\n *\n * **Input shape:**\n *\n *   nD `tf.Tensor` with shape: `(batchSize, ..., inputDim)`.\n *\n *   The most common situation would be\n *   a 2D input with shape `(batchSize, inputDim)`.\n *\n * **Output shape:**\n *\n *   nD tensor with shape: `(batchSize, ..., units)`.\n *\n *   For instance, for a 2D input with shape `(batchSize, inputDim)`,\n *   the output would have shape `(batchSize, units)`.\n *\n * Note: if the input to the layer has a rank greater than 2, then it is\n * flattened prior to the initial dot product with the kernel.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dense(args: DenseLayerArgs): Layer {\n  return new Dense(args);\n}\n\n/**\n * Applies\n * [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to\n * the input.\n *\n * Dropout consists in randomly setting a fraction `rate` of input units to 0 at\n * each update during training time, which helps prevent overfitting.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dropout(args: DropoutLayerArgs): Layer {\n  return new Dropout(args);\n}\n\n/**\n * Spatial 1D version of Dropout.\n *\n * This Layer type performs the same function as the Dropout layer, but it drops\n * entire 1D feature maps instead of individual elements. For example, if an\n * input example consists of 3 timesteps and the feature map for each timestep\n * has a size of 4, a `spatialDropout1d` layer may zero out the feature maps\n * of the 1st timesteps and 2nd timesteps completely while sparing all feature\n * elements of the 3rd timestep.\n *\n * If adjacent frames (timesteps) are strongly correlated (as is normally the\n * case in early convolution layers), regular dropout will not regularize the\n * activation and will otherwise just result in merely an effective learning\n * rate decrease. In this case, `spatialDropout1d` will help promote\n * independence among feature maps and should be used instead.\n *\n * **Arguments:**\n *   rate: A floating-point number >=0 and <=1. Fraction of the input elements\n *     to drop.\n *\n * **Input shape:**\n *   3D tensor with shape `(samples, timesteps, channels)`.\n *\n * **Output shape:**\n *   Same as the input shape.\n *\n * References:\n *   - [Efficient Object Localization Using Convolutional\n *      Networks](https://arxiv.org/abs/1411.4280)\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function spatialDropout1d(args: SpatialDropout1DLayerConfig): Layer {\n  return new SpatialDropout1D(args);\n}\n\n/**\n * Flattens the input. Does not affect the batch size.\n *\n * A `Flatten` layer flattens each batch in its inputs to 1D (making the output\n * 2D).\n *\n * For example:\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const flattenLayer = tf.layers.flatten();\n * // Inspect the inferred output shape of the flatten layer, which\n * // equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the\n * // flattening. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(flattenLayer.apply(input).shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function flatten(args?: FlattenLayerArgs): Layer {\n  return new Flatten(args);\n}\n\n/**\n * Repeats the input n times in a new dimension.\n *\n * ```js\n *  const model = tf.sequential();\n *  model.add(tf.layers.repeatVector({n: 4, inputShape: [2]}));\n *  const x = tf.tensor2d([[10, 20]]);\n *  // Use the model to do inference on a data point the model hasn't see\n *  model.predict(x).print();\n *  // output shape is now [batch, 2, 4]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function repeatVector(args: RepeatVectorLayerArgs): Layer {\n  return new RepeatVector(args);\n}\n\n/**\n * Reshapes an input to a certain shape.\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const reshapeLayer = tf.layers.reshape({targetShape: [2, 6]});\n * // Inspect the inferred output shape of the Reshape layer, which\n * // equals `[null, 2, 6]`. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(reshapeLayer.apply(input).shape));\n * ```\n *\n * Input shape:\n *   Arbitrary, although all dimensions in the input shape must be fixed.\n *   Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n *\n * Output shape:\n *   [batchSize, targetShape[0], targetShape[1], ...,\n *    targetShape[targetShape.length - 1]].\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function reshape(args: ReshapeLayerArgs): Layer {\n  return new Reshape(args);\n}\n\n/**\n * Permutes the dimensions of the input according to a given pattern.\n *\n * Useful for, e.g., connecting RNNs and convnets together.\n *\n * Example:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.permute({\n *   dims: [2, 1],\n *   inputShape: [10, 64]\n * }));\n * console.log(model.outputShape);\n * // Now model's output shape is [null, 64, 10], where null is the\n * // unpermuted sample (batch) dimension.\n * ```\n *\n * Input shape:\n *   Arbitrary. Use the configuration field `inputShape` when using this\n *   layer as the first layer in a model.\n *\n * Output shape:\n *   Same rank as the input shape, but with the dimensions re-ordered (i.e.,\n *   permuted) according to the `dims` configuration of this layer.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function permute(args: PermuteLayerArgs): Layer {\n  return new Permute(args);\n}\n\n/**\n * Maps positive integers (indices) into dense vectors of fixed size.\n * eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]\n *\n * **Input shape:** 2D tensor with shape: `[batchSize, sequenceLength]`.\n *\n * **Output shape:** 3D tensor with shape: `[batchSize, sequenceLength,\n * outputDim]`.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function embedding(args: EmbeddingLayerArgs): Layer {\n  return new Embedding(args);\n}\n\n// Merge Layers.\n\n/**\n * Layer that performs element-wise addition on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). The inputs are specified as an\n * `Array` when the `apply` method of the `Add` layer instance is called. For\n * example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const addLayer = tf.layers.add();\n * const sum = addLayer.apply([input1, input2]);\n * console.log(JSON.stringify(sum.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function add(args?: LayerArgs): Layer {\n  return new Add(args);\n}\n\n/**\n * Layer that performs element-wise averaging on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const averageLayer = tf.layers.average();\n * const average = averageLayer.apply([input1, input2]);\n * console.log(JSON.stringify(average.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function average(args?: LayerArgs): Layer {\n  return new Average(args);\n}\n\n/**\n * Layer that concatenates an `Array` of inputs.\n *\n * It takes a list of tensors, all of the same shape except for the\n * concatenation axis, and returns a single tensor, the concatenation\n * of all inputs. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 3]});\n * const concatLayer = tf.layers.concatenate();\n * const output = concatLayer.apply([input1, input2]);\n * console.log(JSON.stringify(output.shape));\n * // You get [null, 2, 5], with the first dimension as the undetermined batch\n * // dimension. The last dimension (5) is the result of concatenating the\n * // last dimensions of the inputs (2 and 3).\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function concatenate(args?: ConcatenateLayerArgs): Layer {\n  return new Concatenate(args);\n}\n\n/**\n * Layer that computes the element-wise maximum an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const maxLayer = tf.layers.maximum();\n * const max = maxLayer.apply([input1, input2]);\n * console.log(JSON.stringify(max.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function maximum(args?: LayerArgs): Layer {\n  return new Maximum(args);\n}\n\n/**\n * Layer that computes the element-wise minimum of an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const minLayer = tf.layers.minimum();\n * const min = minLayer.apply([input1, input2]);\n * console.log(JSON.stringify(min.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function minimum(args?: LayerArgs): Layer {\n  return new Minimum(args);\n}\n\n/**\n * Layer that multiplies (element-wise) an `Array` of inputs.\n *\n * It takes as input an Array of tensors, all of the same\n * shape, and returns a single tensor (also of the same shape).\n * For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const input3 = tf.input({shape: [2, 2]});\n * const multiplyLayer = tf.layers.multiply();\n * const product = multiplyLayer.apply([input1, input2, input3]);\n * console.log(product.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function multiply(args?: LayerArgs): Layer {\n  return new Multiply(args);\n}\n\n/**\n * Layer that computes a dot product between samples in two tensors.\n *\n * E.g., if applied to a list of two tensors `a` and `b` both of shape\n * `[batchSize, n]`, the output will be a tensor of shape `[batchSize, 1]`,\n * where each entry at index `[i, 0]` will be the dot product between\n * `a[i, :]` and `b[i, :]`.\n *\n * Example:\n *\n * ```js\n * const dotLayer = tf.layers.dot({axes: -1});\n * const x1 = tf.tensor2d([[10, 20], [30, 40]]);\n * const x2 = tf.tensor2d([[-1, -2], [-3, -4]]);\n *\n * // Invoke the layer's apply() method in eager (imperative) mode.\n * const y = dotLayer.apply([x1, x2]);\n * y.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function dot(args: DotLayerArgs): Layer {\n  return new Dot(args);\n}\n\n// Normalization Layers.\n\n/**\n * Batch normalization layer (Ioffe and Szegedy, 2014).\n *\n * Normalize the activations of the previous layer at each batch,\n * i.e. applies a transformation that maintains the mean activation\n * close to 0 and the activation standard deviation close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape` (Array of integers, does\n *   not include the sample axis) when calling the constructor of this class,\n *   if this layer is used as a first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Batch Normalization: Accelerating Deep Network Training by Reducing\n * Internal Covariate Shift](https://arxiv.org/abs/1502.03167)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function batchNormalization(args?: BatchNormalizationLayerArgs): Layer {\n  return new BatchNormalization(args);\n}\n\n/**\n * Layer-normalization layer (Ba et al., 2016).\n *\n * Normalizes the activations of the previous layer for each given example in a\n * batch independently, instead of across a batch like in `batchNormalization`.\n * In other words, this layer applies a transformation that maintanis the mean\n * activation within each example close to0 and activation variance close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the argument `inputShape` when using this layer as the first\n *   layer in a model.\n *\n * Output shape:\n *   Same as input.\n *\n * References:\n *   - [Layer Normalization](https://arxiv.org/abs/1607.06450)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function layerNormalization(args?: LayerNormalizationLayerArgs): Layer {\n  return new LayerNormalization(args);\n}\n\n// Padding Layers.\n\n/**\n * Zero-padding layer for 2D input (e.g., image).\n *\n * This layer can add rows and columns of zeros\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, paddedRows, paddedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, paddedRows, paddedCols]`.\n *\n * @doc {heading: 'Layers', subheading: 'Padding', namespace: 'layers'}\n */\nexport function zeroPadding2d(args?: ZeroPadding2DLayerArgs): Layer {\n  return new ZeroPadding2D(args);\n}\n\n// Pooling Layers.\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape: `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * `tf.avgPool1d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling1d(args: Pooling1DLayerArgs): Layer {\n  return new AveragePooling1D(args);\n}\nexport function avgPool1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape:\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, rows, cols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, pooleRows, pooledCols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, pooleRows, pooledCols]`\n *\n * `tf.avgPool2d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling2d(args: Pooling2DLayerArgs): Layer {\n  return new AveragePooling2D(args);\n}\nexport function avgPool2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n\n/**\n * Average pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      4D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling3d(args: Pooling3DLayerArgs): Layer {\n  return new AveragePooling3D(args);\n}\nexport function avgPool3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n\n/**\n * Global average pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling1d(args?: LayerArgs): Layer {\n  return new GlobalAveragePooling1D(args);\n}\n\n/**\n * Global average pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalAveragePooling2D(args);\n}\n\n/**\n * Global max pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling1d(args?: LayerArgs): Layer {\n  return new GlobalMaxPooling1D(args);\n}\n\n/**\n * Global max pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalMaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for temporal data.\n *\n * Input shape:  `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling1d(args: Pooling1DLayerArgs): Layer {\n  return new MaxPooling1D(args);\n}\n\n/**\n * Max pooling operation for spatial data.\n *\n * Input shape\n *   - If `dataFormat === CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, rows, cols, channels]`\n *   - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *       `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, pooleRows, pooledCols, channels]`\n *   - If `dataFormat=CHANNEL_FIRST`:\n *       4D tensor with shape:\n *       `[batchSize, channels, pooleRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling2d(args: Pooling2DLayerArgs): Layer {\n  return new MaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      5D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling3d(args: Pooling3DLayerArgs): Layer {\n  return new MaxPooling3D(args);\n}\n\n// Recurrent Layers.\n\n/**\n * Gated Recurrent Unit - Cho et al. 2014.\n *\n * This is an `RNN` layer consisting of one `GRUCell`. However, unlike\n * the underlying `GRUCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.gru({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `GRUCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gru(args: GRULayerArgs): Layer {\n  return new GRU(args);\n}\n\n/**\n * Cell class for `GRU`.\n *\n * `GRUCell` is distinct from the `RNN` subclass `GRU` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `GRU` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.gruCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `GRUCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.gruCell({units: 4}),\n *   tf.layers.gruCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `gruCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `GRUCell`, use the\n * `tf.layers.gru`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gruCell(args: GRUCellLayerArgs): RNNCell {\n  return new GRUCell(args);\n}\n\n/**\n * Long-Short Term Memory layer - Hochreiter 1997.\n *\n * This is an `RNN` layer consisting of one `LSTMCell`. However, unlike\n * the underlying `LSTMCell`, the `apply` method of `LSTM` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const lstm = tf.layers.lstm({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = lstm.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `LSTMCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstm(args: LSTMLayerArgs): Layer {\n  return new LSTM(args);\n}\n\n/**\n * Cell class for `LSTM`.\n *\n * `LSTMCell` is distinct from the `RNN` subclass `LSTM` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `LSTM` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.lstmCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `LSTMCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.lstmCell({units: 4}),\n *   tf.layers.lstmCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `lstmCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `LSTMCell`, use the\n * `tf.layers.lstm`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstmCell(args: LSTMCellLayerArgs): RNNCell {\n  return new LSTMCell(args);\n}\n\n/**\n * Fully-connected RNN where the output is to be fed back to input.\n *\n * This is an `RNN` layer consisting of one `SimpleRNNCell`. However, unlike\n * the underlying `SimpleRNNCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.simpleRNN({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `SimpleRNNCell`'s number of units.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNN(args: SimpleRNNLayerArgs): Layer {\n  return new SimpleRNN(args);\n}\n\n/**\n * Cell class for `SimpleRNN`.\n *\n * `SimpleRNNCell` is distinct from the `RNN` subclass `SimpleRNN` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `SimpleRNN` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.simpleRNNCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `SimpleRNNCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.simpleRNNCell({units: 4}),\n *   tf.layers.simpleRNNCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `SimpleRNNCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `SimpleRNNCell`, use the\n * `tf.layers.simpleRNN`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNNCell(args: SimpleRNNCellLayerArgs): RNNCell {\n  return new SimpleRNNCell(args);\n}\n\n/**\n * Convolutional LSTM layer - Xingjian Shi 2015.\n *\n * This is an `ConvRNN2D` layer consisting of one `ConvLSTM2DCell`. However,\n * unlike the underlying `ConvLSTM2DCell`, the `apply` method of `ConvLSTM2D`\n * operates on a sequence of inputs. The shape of the input (not including the\n * first, batch dimension) needs to be 4-D, with the first dimension being time\n * steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const batchSize = 4;\n * const sequenceLength = 2;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [batchSize, sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const layer = tf.layers.convLstm2d({filters, kernelSize});\n *\n * const output = layer.apply(input);\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2d(args: ConvLSTM2DArgs): ConvLSTM2D {\n  return new ConvLSTM2D(args);\n}\n\n/**\n * Cell class for `ConvLSTM2D`.\n *\n * `ConvLSTM2DCell` is distinct from the `ConvRNN2D` subclass `ConvLSTM2D` in\n * that its `call` method takes the input data of only a single time step and\n * returns the cell's output at the time step, while `ConvLSTM2D` takes the\n * input data over a number of time steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const sequenceLength = 1;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const cell = tf.layers.convLstm2dCell({filters, kernelSize});\n *\n * cell.build(input.shape);\n *\n * const outputSize = size - kernelSize + 1;\n * const outShape = [sequenceLength, outputSize, outputSize, filters];\n *\n * const initialH = tf.zeros(outShape);\n * const initialC = tf.zeros(outShape);\n *\n * const [o, h, c] = cell.call([input, initialH, initialC], {});\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2dCell(args: ConvLSTM2DCellArgs): ConvLSTM2DCell {\n  return new ConvLSTM2DCell(args);\n}\n\n/**\n * Base class for recurrent layers.\n *\n * Input shape:\n *   3D tensor with shape `[batchSize, timeSteps, inputDim]`.\n *\n * Output shape:\n *   - if `returnState`, an Array of tensors (i.e., `tf.Tensor`s). The first\n *     tensor is the output. The remaining tensors are the states at the\n *     last time step, each with shape `[batchSize, units]`.\n *   - if `returnSequences`, the output will have shape\n *     `[batchSize, timeSteps, units]`.\n *   - else, the output will have shape `[batchSize, units]`.\n *\n * Masking:\n *   This layer supports masking for input data with a variable number\n *   of timesteps. To introduce masks to your data,\n *   use an embedding layer with the `mask_zero` parameter\n *   set to `True`.\n *\n * Notes on using statefulness in RNNs:\n *   You can set RNN layers to be 'stateful', which means that the states\n *   computed for the samples in one batch will be reused as initial states\n *   for the samples in the next batch. This assumes a one-to-one mapping\n *   between samples in different successive batches.\n *\n *   To enable statefulness:\n *     - specify `stateful: true` in the layer constructor.\n *     - specify a fixed batch size for your model, by passing\n *       if sequential model:\n *         `batchInputShape=[...]` to the first layer in your model.\n *       else for functional model with 1 or more Input layers:\n *         `batchShape=[...]` to all the first layers in your model.\n *       This is the expected shape of your inputs *including the batch size*.\n *       It should be a tuple of integers, e.g. `(32, 10, 100)`.\n *     - specify `shuffle=False` when calling fit().\n *\n *   To reset the states of your model, call `.resetStates()` on either\n *   a specific layer, or on your entire model.\n *\n * Note on specifying the initial state of RNNs\n *   You can specify the initial state of RNN layers symbolically by\n *   calling them with the option `initialState`. The value of\n *   `initialState` should be a tensor or list of tensors representing\n *   the initial state of the RNN layer.\n *\n *   You can specify the initial state of RNN layers numerically by\n *   calling `resetStates` with the keyword argument `states`. The value of\n *   `states` should be a numpy array or list of numpy arrays representing\n *   the initial state of the RNN layer.\n *\n * Note on passing external constants to RNNs\n *   You can pass \"external\" constants to the cell using the `constants`\n *   keyword argument of `RNN.call` method. This requires that the `cell.call`\n *   method accepts the same keyword argument `constants`. Such constants\n *   can be used to conditon the cell transformation on additional static inputs\n *   (not changing over time), a.k.a an attention mechanism.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function rnn(args: RNNLayerArgs): Layer {\n  return new RNN(args);\n}\n\n/**\n * Wrapper allowing a stack of RNN cells to behave as a single cell.\n *\n * Used to implement efficient stacked RNNs.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function stackedRNNCells(args: StackedRNNCellsArgs): RNNCell {\n  return new StackedRNNCells(args);\n}\n\n// Wrapper Layers.\n\n/** @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'} */\nexport function bidirectional(args: BidirectionalLayerArgs): Bidirectional {\n  return new Bidirectional(args);\n}\n\n/**\n * This wrapper applies a layer to every temporal slice of an input.\n *\n * The input should be at least 3D,  and the dimension of the index `1` will be\n * considered to be the temporal dimension.\n *\n * Consider a batch of 32 samples, where each sample is a sequence of 10 vectors\n * of 16 dimensions. The batch input shape of the layer is then `[32,  10,\n * 16]`, and the `inputShape`, not including the sample dimension, is\n * `[10, 16]`.\n *\n * You can then use `TimeDistributed` to apply a `Dense` layer to each of the 10\n * timesteps, independently:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.dense({units: 8}),\n *   inputShape: [10, 16],\n * }));\n *\n * // Now model.outputShape = [null, 10, 8].\n * // The output will then have shape `[32, 10, 8]`.\n *\n * // In subsequent layers, there is no need for `inputShape`:\n * model.add(tf.layers.timeDistributed({layer: tf.layers.dense({units: 32})}));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * // Now model.outputShape = [null, 10, 32].\n * ```\n *\n * The output will then have shape `[32, 10, 32]`.\n *\n * `TimeDistributed` can be used with arbitrary layers, not just `Dense`, for\n * instance a `Conv2D` layer.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.conv2d({filters: 64, kernelSize: [3, 3]}),\n *   inputShape: [10, 299, 299, 3],\n * }));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}\n */\nexport function timeDistributed(args: WrapperLayerArgs): Layer {\n  return new TimeDistributed(args);\n}\n\n// Aliases for pooling.\nexport const globalMaxPool1d = globalMaxPooling1d;\nexport const globalMaxPool2d = globalMaxPooling2d;\nexport const maxPool1d = maxPooling1d;\nexport const maxPool2d = maxPooling2d;\n\nexport {Layer, RNN, RNNCell, input /* alias for tf.input */};\n\n/**\n * Apply additive zero-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * This is useful to mitigate overfitting\n * (you could see it as a form of random data augmentation).\n * Gaussian Noise (GS) is a natural choice as corruption process\n * for real valued inputs.\n *\n * # Arguments\n *     stddev: float, standard deviation of the noise distribution.\n *\n * # Input shape\n *         Arbitrary. Use the keyword argument `input_shape`\n *         (tuple of integers, does not include the samples axis)\n *         when using this layer as the first layer in a model.\n *\n * # Output shape\n *         Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianNoise(args: GaussianNoiseArgs): GaussianNoise {\n  return new GaussianNoise(args);\n}\n\n/**\n * Apply multiplicative 1-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting](\n *      http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianDropout(args: GaussianDropoutArgs): GaussianDropout {\n  return new GaussianDropout(args);\n}\n\n/**\n * Applies Alpha Dropout to the input.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\n * to their original values, in order to ensure the self-normalizing property\n * even after this dropout.\n * Alpha Dropout fits well to Scaled Exponential Linear Units\n * by randomly setting activations to the negative saturation value.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\n *     shape for randomly generated keep/drop flags.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function alphaDropout(args: AlphaDropoutArgs): AlphaDropout {\n  return new AlphaDropout(args);\n}\n\n/**\n * Masks a sequence by using a mask value to skip timesteps.\n *\n * If all features for a given sample timestep are equal to `mask_value`,\n * then the sample timestep will be masked (skipped) in all downstream layers\n * (as long as they support masking).\n *\n * If any downstream layer does not support masking yet receives such\n * an input mask, an exception will be raised.\n *\n * Arguments:\n *   - `maskValue`: Either None or mask value to skip.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Mask', namespace: 'layers'}\n */\nexport function masking(args?: MaskingArgs): Layer {\n  return new Masking(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport * as losses from './losses';\nimport * as metrics from './metrics';\n\n/**\n * Binary accuracy metric function.\n *\n * `yTrue` and `yPred` can have 0-1 values. Example:\n * ```js\n * const x = tf.tensor2d([[1, 1, 1, 1], [0, 0, 0, 0]], [2, 4]);\n * const y = tf.tensor2d([[1, 0, 1, 0], [0, 0, 0, 1]], [2, 4]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * `yTrue` and `yPred` can also have floating-number values between 0 and 1, in\n * which case the values will be thresholded at 0.5 to yield 0-1 values (i.e.,\n * a value >= 0.5 and <= 1.0 is interpreted as 1.\n * )\n * Example:\n * ```js\n * const x = tf.tensor1d([1, 1, 1, 1, 0, 0, 0, 0]);\n * const y = tf.tensor1d([0.2, 0.4, 0.6, 0.8, 0.2, 0.3, 0.4, 0.7]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryAccuracy(yTrue, yPred);\n}\n\n/**\n * Binary crossentropy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0], [1], [1], [1]]);\n * const y = tf.tensor2d([[0], [0], [0.5], [1]]);\n * const crossentropy = tf.metrics.binaryCrossentropy(x, y);\n * crossentropy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction, probabilities for the `1` case.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryCrossentropy(yTrue, yPred);\n}\n\n/**\n * Sparse categorical accuracy metric function.\n *\n * Example:\n * ```js\n *\n * const yTrue = tf.tensor1d([1, 1, 2, 2, 0]);\n * const yPred = tf.tensor2d(\n *      [[0, 1, 0], [1, 0, 0], [0, 0.4, 0.6], [0, 0.6, 0.4], [0.7, 0.3, 0]]);\n * const crossentropy = tf.metrics.sparseCategoricalAccuracy(yTrue, yPred);\n * crossentropy.print();\n * ```\n *\n * @param yTrue True labels: indices.\n * @param yPred Predicted probabilities or logits.\n * @returns Accuracy tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.sparseCategoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical accuracy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0, 0, 0, 1], [0, 0, 0, 1]]);\n * const y = tf.tensor2d([[0.1, 0.8, 0.05, 0.05], [0.1, 0.05, 0.05, 0.8]]);\n * const accuracy = tf.metrics.categoricalAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth: one-hot encoding of categories.\n * @param yPred Binary Tensor of prediction: probabilities or logits for the\n *   same categories as in `yTrue`.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical crossentropy between an output tensor and a target tensor.\n *\n * @param target A tensor of the same shape as `output`.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalCrossentropy(yTrue, yPred);\n}\n\n/**\n * Computes the precision of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const precision = tf.metrics.precision(x, y);\n * precision.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Precision Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.precision(yTrue, yPred);\n}\n\n/**\n * Computes the recall of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const recall = tf.metrics.recall(x, y);\n * recall.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Recall Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.recall(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Cosine proximity.\n *\n * Mathematically, cosine proximity is defined as:\n *   `-sum(l2Normalize(yTrue) * l2Normalize(yPred))`,\n * wherein `l2Normalize()` normalizes the L2 norm of the input to 1 and `*`\n * represents element-wise multiplication.\n *\n * ```js\n * const yTrue = tf.tensor2d([[1, 0], [1, 0]]);\n * const yPred = tf.tensor2d([[1 / Math.sqrt(2), 1 / Math.sqrt(2)], [0, 1]]);\n * const proximity = tf.metrics.cosineProximity(yTrue, yPred);\n * proximity.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Cosine proximity Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.cosineProximity(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute error.\n *\n * Mathematically, mean absolute error is defined as:\n *   `mean(abs(yPred - yTrue))`,\n * wherein the `mean` is applied over feature dimensions.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [0, 0], [2, 3]]);\n * const yPred = tf.tensor2d([[0, 1], [0, 1], [-2, -3]]);\n * const mse = tf.metrics.meanAbsoluteError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsoluteError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute percentage error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [10, 20]]);\n * const yPred = tf.tensor2d([[0, 1], [11, 24]]);\n * const mse = tf.metrics.meanAbsolutePercentageError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MAPE`, `tf.metrics.mape`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute percentage error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function MAPE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function mape(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean squared error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [3, 4]]);\n * const yPred = tf.tensor2d([[0, 1], [-3, -4]]);\n * const mse = tf.metrics.meanSquaredError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MSE`, `tf.metrics.mse`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean squared error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function MSE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function mse(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nexport {modelFromJSON} from './models';\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport * as regularizers from './regularizers';\n// tslint:disable-next-line:max-line-length\nimport {L1Args, L1L2, L1L2Args, L2Args, Regularizer} from './regularizers';\n\n/**\n * Regularizer for L1 and L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x)) + sum(l2 * x^2)\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1l2(config?: L1L2Args): Regularizer {\n  return new L1L2(config);\n}\n\n/**\n * Regularizer for L1 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x))\n * @param args l1 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1(config?: L1Args): Regularizer {\n  return regularizers.l1(config);\n}\n\n/**\n * Regularizer for L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l2 * x^2)\n * @param args l2 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l2(config?: L2Args): Regularizer {\n  return regularizers.l2(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {BaseCallback} from './base_callbacks';\nimport {Container} from './engine/container';\nimport {LayersModel} from './engine/training';\nimport {NotImplementedError} from './errors';\nimport {Logs, resolveScalarsInLogs} from './logs';\n\nexport abstract class Callback extends BaseCallback {\n  /** Instance of `keras.models.Model`. Reference of the model being trained. */\n  model: LayersModel = null;\n\n  setModel(model: Container): void {\n    if (!(model instanceof LayersModel)) {\n      throw new Error('model must be a LayersModel, not some other Container');\n    }\n    this.model = model;\n  }\n}\n\nexport interface EarlyStoppingCallbackArgs {\n  /**\n   * Quantity to be monitored.\n   *\n   * Defaults to 'val_loss'.\n   */\n  monitor?: string;\n\n  /**\n   * Minimum change in the monitored quantity to qualify as improvement,\n   * i.e., an absolute change of less than `minDelta` will count as no\n   * improvement.\n   *\n   * Defaults to 0.\n   */\n  minDelta?: number;\n\n  /**\n   * Number of epochs with no improvement after which training will be stopped.\n   *\n   * Defaults to 0.\n   */\n  patience?: number;\n\n  /** Verbosity mode. */\n  verbose?: number;\n\n  /**\n   * Mode: one of 'min', 'max', and 'auto'.\n   * - In 'min' mode, training will be stopped when the quantity monitored has\n   *   stopped decreasing.\n   * - In 'max' mode, training will be stopped when the quantity monitored has\n   *   stopped increasing.\n   * - In 'auto' mode, the direction is inferred automatically from the name of\n   *   the monitored quantity.\n   *\n   * Defaults to 'auto'.\n   */\n  mode?: 'auto'|'min'|'max';\n\n  /**\n   * Baseline value of the monitored quantity.\n   *\n   * If specified, training will be stopped if the model doesn't show\n   * improvement over the baseline.\n   */\n  baseline?: number;\n\n  /**\n   * Whether to restore model weights from the epoch with the best value\n   * of the monitored quantity. If `False`, the model weights obtained at the\n   * at the last step of training are used.\n   *\n   * **`True` is not supported yet.**\n   */\n  restoreBestWeights?: boolean;\n}\n\nfunction less(currVal: number, prevVal: number) {\n  return currVal < prevVal;\n}\n\nfunction greater(currVal: number, prevVal: number) {\n  return currVal > prevVal;\n}\n\n/**\n * A Callback that stops training when a monitored quantity has stopped\n * improving.\n */\nexport class EarlyStopping extends Callback {\n  protected readonly monitor: string;\n  protected readonly minDelta: number;\n  protected readonly patience: number;\n  protected readonly baseline: number;\n  protected readonly verbose: number;\n  protected readonly mode: 'auto'|'min'|'max';\n\n  protected monitorFunc: (currVal: number, prevVal: number) => boolean;\n\n  private wait: number;\n  private stoppedEpoch: number;\n  private best: number;\n\n  constructor(args?: EarlyStoppingCallbackArgs) {\n    super();\n    if (args == null) {\n      args = {};\n    }\n    if (args.restoreBestWeights) {\n      throw new NotImplementedError(\n          'restoreBestWeights = True is not implemented in EarlyStopping yet.');\n    }\n\n    this.monitor = args.monitor || 'val_loss';\n    this.minDelta = Math.abs(args.minDelta || 0);\n    this.patience = args.patience || 0;\n    this.verbose = args.verbose || 0;\n    this.mode = args.mode || 'auto';\n    this.baseline = args.baseline;\n\n    if (['auto', 'min', 'max'].indexOf(this.mode) === -1) {\n      console.warn(\n          `EarlyStopping mode '${this.mode}' is invalid. ` +\n          `Falling back to mode 'auto'.`);\n      this.mode = 'auto';\n    }\n\n    if (this.mode === 'min') {\n      this.monitorFunc = less;\n    } else if (this.mode === 'max') {\n      this.monitorFunc = greater;\n    } else {\n      // For mode === 'auto'.\n      if (this.monitor.indexOf('acc') !== -1) {\n        this.monitorFunc = greater;\n      } else {\n        this.monitorFunc = less;\n      }\n    }\n\n    if (this.monitorFunc === less) {\n      this.minDelta *= -1;\n    }\n  }\n\n  async onTrainBegin(logs?: Logs) {\n    this.wait = 0;\n    this.stoppedEpoch = 0;\n    if (this.baseline != null) {\n      this.best = this.baseline;\n    } else {\n      this.best = this.monitorFunc === less ? Infinity : -Infinity;\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: Logs) {\n    await resolveScalarsInLogs(logs);\n    const current = this.getMonitorValue(logs);\n    if (current == null) {\n      return;\n    }\n\n    if (this.monitorFunc(current - this.minDelta, this.best)) {\n      this.best = current;\n      this.wait = 0;\n      // TODO(cais): Logic for restoreBestWeights.\n    } else {\n      this.wait++;\n      if (this.wait >= this.patience) {\n        this.stoppedEpoch = epoch;\n        this.model.stopTraining = true;\n      }\n      // TODO(cais): Logic for restoreBestWeights.\n    }\n  }\n\n  async onTrainEnd(logs?: Logs) {\n    if (this.stoppedEpoch > 0 && this.verbose) {\n      console.log(`Epoch ${this.stoppedEpoch}: early stopping.`);\n    }\n  }\n\n  private getMonitorValue(logs: Logs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const monitorValue = logs[this.monitor];\n    if (monitorValue == null) {\n      console.warn(\n          `Metric for EarlyStopping ${this.monitor} is not available. ` +\n          `Available metrics are: ${Object.keys(logs)}`);\n    }\n    return monitorValue;\n  }\n}\n\n/**\n * Factory function for a Callback that stops training when a monitored\n * quantity has stopped improving.\n *\n * Early stopping is a type of regularization, and protects model against\n * overfitting.\n *\n * The following example based on fake data illustrates how this callback\n * can be used during `tf.LayersModel.fit()`:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense({\n *   units: 3,\n *   activation: 'softmax',\n *   kernelInitializer: 'ones',\n *   inputShape: [2]\n * }));\n * const xs = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const ys = tf.tensor2d([[1, 0, 0], [0, 1, 0]], [2, 3]);\n * const xsVal = tf.tensor2d([4, 3, 2, 1], [2, 2]);\n * const ysVal = tf.tensor2d([[0, 0, 1], [0, 1, 0]], [2, 3]);\n * model.compile(\n *     {loss: 'categoricalCrossentropy', optimizer: 'sgd', metrics: ['acc']});\n *\n * // Without the EarlyStopping callback, the val_acc value would be:\n * //   0.5, 0.5, 0.5, 0.5, ...\n * // With val_acc being monitored, training should stop after the 2nd epoch.\n * const history = await model.fit(xs, ys, {\n *   epochs: 10,\n *   validationData: [xsVal, ysVal],\n *   callbacks: tf.callbacks.earlyStopping({monitor: 'val_acc'})\n * });\n *\n * // Expect to see a length-2 array.\n * console.log(history.history.val_acc);\n * ```\n *\n * @doc {\n *   heading: 'Callbacks',\n *   namespace: 'callbacks'\n * }\n */\nexport function earlyStopping(args?: EarlyStoppingCallbackArgs) {\n  return new EarlyStopping(args);\n}\n\nexport const callbacks = {earlyStopping};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/* tslint:disable */\n\n/** Properties of an Any. */\nexport declare interface IAny {\n  /** Any typeUrl */\n  typeUrl?: (string|null);\n\n  /** Any value */\n  value?: (Uint8Array|null);\n}\n\n/** DataType enum. */\nexport enum DataType {\n  'DT_INVALID' = 0,\n  'DT_FLOAT' = 1,\n  'DT_DOUBLE' = 2,\n  'DT_INT32' = 3,\n  'DT_UINT8' = 4,\n  'DT_INT16' = 5,\n  'DT_INT8' = 6,\n  'DT_STRING' = 7,\n  'DT_COMPLEX64' = 8,\n  'DT_INT64' = 9,\n  'DT_BOOL' = 10,\n  'DT_QINT8' = 11,\n  'DT_QUINT8' = 12,\n  'DT_QINT32' = 13,\n  'DT_BFLOAT16' = 14,\n  'DT_FLOAT_REF' = 101,\n  'DT_DOUBLE_REF' = 102,\n  'DT_INT32_REF' = 103,\n  'DT_UINT8_REF' = 104,\n  'DT_INT16_REF' = 105,\n  'DT_INT8_REF' = 106,\n  'DT_STRING_REF' = 107,\n  'DT_COMPLEX64_REF' = 108,\n  'DT_INT64_REF' = 109,\n  'DT_BOOL_REF' = 110,\n  'DT_QINT8_REF' = 111,\n  'DT_QUINT8_REF' = 112,\n  'DT_QINT32_REF' = 113,\n  'DT_BFLOAT16_REF' = 114\n}\n\n/** Properties of a TensorShape. */\nexport declare interface ITensorShape {\n  /** TensorShape dim */\n  dim?: (TensorShape.IDim[]|null);\n\n  /** TensorShape unknownRank */\n  unknownRank?: (boolean|null);\n}\n\nexport namespace TensorShape {\n  /** Properties of a Dim. */\n  export declare interface IDim {\n    /** Dim size */\n    size?: (number|string|null);\n\n    /** Dim name */\n    name?: (string|null);\n  }\n}\n\n/** Properties of a Tensor. */\nexport declare interface ITensor {\n  /** Tensor dtype */\n  dtype?: (DataType|null);\n\n  /** Tensor tensorShape */\n  tensorShape?: (ITensorShape|null);\n\n  /** Tensor versionNumber */\n  versionNumber?: (number|null);\n\n  /** Tensor tensorContent */\n  tensorContent?: (Uint8Array|null);\n\n  /** Tensor floatVal */\n  floatVal?: (number[]|null);\n\n  /** Tensor doubleVal */\n  doubleVal?: (number[]|null);\n\n  /** Tensor intVal */\n  intVal?: (number[]|null);\n\n  /** Tensor stringVal */\n  stringVal?: (Uint8Array[]|null);\n\n  /** Tensor scomplexVal */\n  scomplexVal?: (number[]|null);\n\n  /** Tensor int64Val */\n  int64Val?: ((number | string)[]|null);\n\n  /** Tensor boolVal */\n  boolVal?: (boolean[]|null);\n\n  /** Tensor uint32Val */\n  uint32Val?: (number[]|null);\n\n  /** Tensor uint64Val */\n  uint64Val?: ((number | string)[]|null);\n}\n\n/** Properties of an AttrValue. */\nexport declare interface IAttrValue {\n  /** AttrValue list */\n  list?: (AttrValue.IListValue|null);\n\n  /** AttrValue s */\n  s?: (string|null);\n\n  /** AttrValue i */\n  i?: (number|string|null);\n\n  /** AttrValue f */\n  f?: (number|null);\n\n  /** AttrValue b */\n  b?: (boolean|null);\n\n  /** AttrValue type */\n  type?: (DataType|null);\n\n  /** AttrValue shape */\n  shape?: (ITensorShape|null);\n\n  /** AttrValue tensor */\n  tensor?: (ITensor|null);\n\n  /** AttrValue placeholder */\n  placeholder?: (string|null);\n\n  /** AttrValue func */\n  func?: (INameAttrList|null);\n}\n\nexport namespace AttrValue {\n  /** Properties of a ListValue. */\n  export declare interface IListValue {\n    /** ListValue s */\n    s?: (string[]|null);\n\n    /** ListValue i */\n    i?: ((number | string)[]|null);\n\n    /** ListValue f */\n    f?: (number[]|null);\n\n    /** ListValue b */\n    b?: (boolean[]|null);\n\n    /** ListValue type */\n    type?: (DataType[]|null);\n\n    /** ListValue shape */\n    shape?: (ITensorShape[]|null);\n\n    /** ListValue tensor */\n    tensor?: (ITensor[]|null);\n\n    /** ListValue func */\n    func?: (INameAttrList[]|null);\n  }\n}\n\n/** Properties of a NameAttrList. */\nexport declare interface INameAttrList {\n  /** NameAttrList name */\n  name?: (string|null);\n\n  /** NameAttrList attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a NodeDef. */\nexport declare interface INodeDef {\n  /** NodeDef name */\n  name?: (string|null);\n\n  /** NodeDef op */\n  op?: (string|null);\n\n  /** NodeDef input */\n  input?: (string[]|null);\n\n  /** NodeDef device */\n  device?: (string|null);\n\n  /** NodeDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a VersionDef. */\nexport declare interface IVersionDef {\n  /** VersionDef producer */\n  producer?: (number|null);\n\n  /** VersionDef minConsumer */\n  minConsumer?: (number|null);\n\n  /** VersionDef badConsumers */\n  badConsumers?: (number[]|null);\n}\n\n/** Properties of a GraphDef. */\nexport declare interface IGraphDef {\n  /** GraphDef node */\n  node?: (INodeDef[]|null);\n\n  /** GraphDef versions */\n  versions?: (IVersionDef|null);\n\n  /** GraphDef library */\n  library?: (IFunctionDefLibrary|null);\n}\n\n/** Properties of a CollectionDef. */\nexport declare interface ICollectionDef {\n  /** CollectionDef nodeList */\n  nodeList?: (CollectionDef.INodeList|null);\n\n  /** CollectionDef bytesList */\n  bytesList?: (CollectionDef.IBytesList|null);\n\n  /** CollectionDef int64List */\n  int64List?: (CollectionDef.IInt64List|null);\n\n  /** CollectionDef floatList */\n  floatList?: (CollectionDef.IFloatList|null);\n\n  /** CollectionDef anyList */\n  anyList?: (CollectionDef.IAnyList|null);\n}\n\nexport namespace CollectionDef {\n  /** Properties of a NodeList. */\n  export declare interface INodeList {\n    /** NodeList value */\n    value?: (string[]|null);\n  }\n\n  /** Properties of a BytesList. */\n  export declare interface IBytesList {\n    /** BytesList value */\n    value?: (Uint8Array[]|null);\n  }\n\n  /** Properties of an Int64List. */\n  export declare interface IInt64List {\n    /** Int64List value */\n    value?: ((number | string)[]|null);\n  }\n\n  /** Properties of a FloatList. */\n  export declare interface IFloatList {\n    /** FloatList value */\n    value?: (number[]|null);\n  }\n\n  /** Properties of an AnyList. */\n  export declare interface IAnyList {\n    /** AnyList value */\n    value?: (IAny[]|null);\n  }\n}\n\n/** Properties of a SaverDef. */\nexport declare interface ISaverDef {\n  /** SaverDef filenameTensorName */\n  filenameTensorName?: (string|null);\n\n  /** SaverDef saveTensorName */\n  saveTensorName?: (string|null);\n\n  /** SaverDef restoreOpName */\n  restoreOpName?: (string|null);\n\n  /** SaverDef maxToKeep */\n  maxToKeep?: (number|null);\n\n  /** SaverDef sharded */\n  sharded?: (boolean|null);\n\n  /** SaverDef keepCheckpointEveryNHours */\n  keepCheckpointEveryNHours?: (number|null);\n\n  /** SaverDef version */\n  version?: (SaverDef.CheckpointFormatVersion|null);\n}\n\nexport namespace SaverDef {\n  /** CheckpointFormatVersion enum. */\n  export enum CheckpointFormatVersion {'LEGACY' = 0, 'V1' = 1, 'V2' = 2}\n}\n\n/** Properties of a TensorInfo. */\nexport declare interface ITensorInfo {\n  /** TensorInfo name */\n  name?: (string|null);\n\n  /** TensorInfo cooSparse */\n  cooSparse?: (TensorInfo.ICooSparse|null);\n\n  /** TensorInfo dtype */\n  dtype?: (DataType|null);\n\n  /** TensorInfo tensorShape */\n  tensorShape?: (ITensorShape|null);\n}\n\nexport namespace TensorInfo {\n  /** Properties of a CooSparse. */\n  export declare interface ICooSparse {\n    /** CooSparse valuesTensorName */\n    valuesTensorName?: (string|null);\n\n    /** CooSparse indicesTensorName */\n    indicesTensorName?: (string|null);\n\n    /** CooSparse denseShapeTensorName */\n    denseShapeTensorName?: (string|null);\n  }\n}\n\n/** Properties of a SignatureDef. */\nexport declare interface ISignatureDef {\n  /** SignatureDef inputs */\n  inputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef outputs */\n  outputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef methodName */\n  methodName?: (string|null);\n}\n\n/** Properties of an AssetFileDef. */\nexport declare interface IAssetFileDef {\n  /** AssetFileDef tensorInfo */\n  tensorInfo?: (ITensorInfo|null);\n\n  /** AssetFileDef filename */\n  filename?: (string|null);\n}\n\n/** Properties of an OpDef. */\nexport declare interface IOpDef {\n  /** OpDef name */\n  name?: (string|null);\n\n  /** OpDef inputArg */\n  inputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef outputArg */\n  outputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef attr */\n  attr?: (OpDef.IAttrDef[]|null);\n\n  /** OpDef deprecation */\n  deprecation?: (OpDef.IOpDeprecation|null);\n\n  /** OpDef summary */\n  summary?: (string|null);\n\n  /** OpDef description */\n  description?: (string|null);\n\n  /** OpDef isCommutative */\n  isCommutative?: (boolean|null);\n\n  /** OpDef isAggregate */\n  isAggregate?: (boolean|null);\n\n  /** OpDef isStateful */\n  isStateful?: (boolean|null);\n\n  /** OpDef allowsUninitializedInput */\n  allowsUninitializedInput?: (boolean|null);\n}\n\nexport namespace OpDef {\n  /** Properties of an ArgDef. */\n  export declare interface IArgDef {\n    /** ArgDef name */\n    name?: (string|null);\n\n    /** ArgDef description */\n    description?: (string|null);\n\n    /** ArgDef type */\n    type?: (DataType|null);\n\n    /** ArgDef typeAttr */\n    typeAttr?: (string|null);\n\n    /** ArgDef numberAttr */\n    numberAttr?: (string|null);\n\n    /** ArgDef typeListAttr */\n    typeListAttr?: (string|null);\n\n    /** ArgDef isRef */\n    isRef?: (boolean|null);\n  }\n\n  /** Properties of an AttrDef. */\n  export declare interface IAttrDef {\n    /** AttrDef name */\n    name?: (string|null);\n\n    /** AttrDef type */\n    type?: (string|null);\n\n    /** AttrDef defaultValue */\n    defaultValue?: (IAttrValue|null);\n\n    /** AttrDef description */\n    description?: (string|null);\n\n    /** AttrDef hasMinimum */\n    hasMinimum?: (boolean|null);\n\n    /** AttrDef minimum */\n    minimum?: (number|string|null);\n\n    /** AttrDef allowedValues */\n    allowedValues?: (IAttrValue|null);\n  }\n\n  /** Properties of an OpDeprecation. */\n  export declare interface IOpDeprecation {\n    /** OpDeprecation version */\n    version?: (number|null);\n\n    /** OpDeprecation explanation */\n    explanation?: (string|null);\n  }\n}\n\n/** Properties of an OpList. */\nexport declare interface IOpList {\n  /** OpList op */\n  op?: (IOpDef[]|null);\n}\n\n/** Properties of a MetaGraphDef. */\nexport declare interface IMetaGraphDef {\n  /** MetaGraphDef metaInfoDef */\n  metaInfoDef?: (MetaGraphDef.IMetaInfoDef|null);\n\n  /** MetaGraphDef graphDef */\n  graphDef?: (IGraphDef|null);\n\n  /** MetaGraphDef saverDef */\n  saverDef?: (ISaverDef|null);\n\n  /** MetaGraphDef collectionDef */\n  collectionDef?: ({[k: string]: ICollectionDef}|null);\n\n  /** MetaGraphDef signatureDef */\n  signatureDef?: ({[k: string]: ISignatureDef}|null);\n\n  /** MetaGraphDef assetFileDef */\n  assetFileDef?: (IAssetFileDef[]|null);\n}\n\nexport namespace MetaGraphDef {\n  /** Properties of a MetaInfoDef. */\n  export declare interface IMetaInfoDef {\n    /** MetaInfoDef metaGraphVersion */\n    metaGraphVersion?: (string|null);\n\n    /** MetaInfoDef strippedOpList */\n    strippedOpList?: (IOpList|null);\n\n    /** MetaInfoDef anyInfo */\n    anyInfo?: (IAny|null);\n\n    /** MetaInfoDef tags */\n    tags?: (string[]|null);\n\n    /** MetaInfoDef tensorflowVersion */\n    tensorflowVersion?: (string|null);\n\n    /** MetaInfoDef tensorflowGitVersion */\n    tensorflowGitVersion?: (string|null);\n  }\n}\n\n/** Properties of a SavedModel. */\nexport declare interface ISavedModel {\n  /** SavedModel savedModelSchemaVersion */\n  savedModelSchemaVersion?: (number|string|null);\n\n  /** SavedModel metaGraphs */\n  metaGraphs?: (IMetaGraphDef[]|null);\n}\n\n/** Properties of a FunctionDefLibrary. */\nexport declare interface IFunctionDefLibrary {\n  /** FunctionDefLibrary function */\n  'function'?: (IFunctionDef[]|null);\n\n  /** FunctionDefLibrary gradient */\n  gradient?: (IGradientDef[]|null);\n}\n\n/** Properties of a FunctionDef. */\nexport declare interface IFunctionDef {\n  /** FunctionDef signature */\n  signature?: (IOpDef|null);\n\n  /** FunctionDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n\n  /** FunctionDef nodeDef */\n  nodeDef?: (INodeDef[]|null);\n\n  /** FunctionDef ret */\n  ret?: ({[k: string]: string}|null);\n}\n\n/** Properties of a GradientDef. */\nexport declare interface IGradientDef {\n  /** GradientDef functionName */\n  functionName?: (string|null);\n\n  /** GradientDef gradientFunc */\n  gradientFunc?: (string|null);\n}\n", "\n/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpExecutor, OpMapper} from '../types';\n\nconst CUSTOM_OPS: {[key: string]: OpMapper} = {};\n\n/**\n * Register an Op for graph model executor. This allow you to register\n * TensorFlow custom op or override existing op.\n *\n * Here is an example of registering a new MatMul Op.\n * ```js\n * const customMatmul = (node) =>\n *    tf.matMul(\n *        node.inputs[0], node.inputs[1],\n *        node.attrs['transpose_a'], node.attrs['transpose_b']);\n *\n * tf.registerOp('MatMul', customMatmul);\n * ```\n * The inputs and attrs of the node object is based on the TensorFlow op\n * registry.\n *\n * @param name The Tensorflow Op name.\n * @param opFunc An op function which is called with the current graph node\n * during execution and needs to return a tensor or a list of tensors. The node\n * has the following attributes:\n *    - attr: A map from attribute name to its value\n *    - inputs: A list of input tensors\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function registerOp(name: string, opFunc: OpExecutor) {\n  const opMapper: OpMapper = {\n    tfOpName: name,\n    category: 'custom',\n    inputs: [],\n    attrs: [],\n    customExecutor: opFunc\n  };\n\n  CUSTOM_OPS[name] = opMapper;\n}\n\n/**\n * Retrieve the OpMapper object for the registered op.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function getRegisteredOp(name: string): OpMapper {\n  return CUSTOM_OPS[name];\n}\n\n/**\n * Deregister the Op for graph model executor.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function deregisterOp(name: string) {\n  delete CUSTOM_OPS[name];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {clone, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {Node, ValueType} from '../types';\n\nexport function getParamValue(\n    paramName: string, node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  const inputParam = node.inputParams[paramName];\n  if (inputParam && inputParam.inputIndexStart !== undefined) {\n    const start = inputParam.inputIndexStart;\n    const end = inputParam.inputIndexEnd === 0 ?\n        undefined :\n        (inputParam.inputIndexEnd === undefined ? start + 1 :\n                                                  inputParam.inputIndexEnd);\n    if (inputParam.type === 'tensor') {\n      return getTensor(\n          node.inputNames[inputParam.inputIndexStart], tensorMap, context);\n    }\n    if (inputParam.type === 'tensors') {\n      const inputs = node.inputNames.slice(start, end);\n\n      return inputs.map(name => getTensor(name, tensorMap, context));\n    }\n    const tensor =\n        getTensor(node.inputNames.slice(start)[0], tensorMap, context);\n    const data = tensor.dataSync();\n    return inputParam.type === 'number' ?\n        data[0] :\n        util.toNestedArray(tensor.shape, data);\n  }\n  const attrParam = node.attrParams[paramName];\n  return attrParam && attrParam.value;\n}\n\n/**\n * Retrieve the tensor based on input name by extracting the node name and\n * output index information.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensor(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor {\n  const [nodeName, index] = parseNodeName(name);\n  const contextId = context.currentContextIds.find(contextId => {\n    return !!tensorsMap[getNodeNameWithContextId(nodeName, contextId)];\n  });\n\n  return contextId !== undefined ?\n      tensorsMap[getNodeNameWithContextId(nodeName, contextId)][index] :\n      undefined;\n}\n\n/**\n * Retrieve the tensors based on input name for current context.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensorsForCurrentContenxt(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor[] {\n  return tensorsMap[getNodeNameWithContextId(name, context.currentContextId)];\n}\n\n/**\n * Returns the node name and index from the Node input name.\n * @param inputName The input name of the node, in format of\n * node_name:output_index, i.e. MatMul:0, if the output_index is not set, it is\n * default to 0.\n */\nexport function getNodeNameAndIndex(\n    inputName: string, context?: ExecutionContext): [string, number] {\n  const [nodeName, index] = parseNodeName(inputName);\n\n  return [\n    getNodeNameWithContextId(nodeName, context && context.currentContextId),\n    index\n  ];\n}\n\nfunction getNodeNameWithContextId(name: string, contextId?: string): string {\n  return !!contextId ? `${name}-${contextId}` : name;\n}\n\nexport function parseNodeName(name: string): [string, number] {\n  const parts = name.split(':');\n  if (parts.length === 1) {\n    return [name, 0];\n  }\n\n  const nodeName = parts[0];\n  return [nodeName, Number(parts[parts.length - 1])];\n}\n\nexport function split(arr: number[], size: number) {\n  const res = [];\n  for (let i = 0; i < arr.length; i += size) {\n    res.push(arr.slice(i, i + size));\n  }\n  return res;\n}\nexport function getPadding(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  let pad = getParamValue('pad', node, tensorMap, context);\n  if (pad === 'explicit') {\n    // This is 1d array, we need to convert it to 2d array\n    pad = getParamValue('explicitPaddings', node, tensorMap, context);\n    const explicitPadding: [\n      [number, number], [number, number], [number, number], [number, number]\n    ] = [[0, 0], [0, 0], [0, 0], [0, 0]];\n    for (let i = 0; i < 4; i++) {\n      explicitPadding[i][0] = (pad as number[])[i * 2];\n      explicitPadding[i][1] = (pad as number[])[i * 2 + 1];\n    }\n    return explicitPadding;\n  }\n  return pad;\n}\n\n/**\n *  Reuse the tensor if it is marked as keep, otherwise clone the tensor to\n *  avoid disposal. This is important for TensorArray and TensorList ops, since\n *  internally they use a tensor as the id for TensorArray and TensorList, and\n * to simplify lookup, they also use Tensor.id as the key to the internal map.\n * These id tensors have been marked as kept in the backend, we need avoid clone\n * them in order to create new Tensor.id.\n * @param tensor\n */\nexport function cloneTensor(tensor: Tensor): Tensor {\n  return tensor.kept ? tensor : clone(tensor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Add',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddV2',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddN',\n    'category': 'arithmetic',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'BiasAdd',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sub',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'RealDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Div',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'DivNoNan',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mul',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Maximum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Minimum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Pow',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SquaredDifference',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorMod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Abs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan2',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ceil',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ClipByValue',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'clip_value_min', 'name': 'clipValueMin', 'type': 'number'},\n      {'tfName': 'clip_value_max', 'name': 'clipValueMax', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Complex',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'real', 'type': 'tensor'},\n      {'start': 1, 'name': 'imag', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ComplexAbs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Elu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Exp',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Floor',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Imag',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Neg',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Real',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Prelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'alpha', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu6',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'clipValueMin',\n        'name': 'clipValueMin',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'clipValueMax',\n        'name': 'clipValueMax',\n        'type': 'number',\n        'defaultValue': 6\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Selu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sigmoid',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Rsqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Square',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sign',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Round',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Expm1',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log1p',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Reciprocal',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Softplus',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Erf',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axes', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'keep_dims',\n        'name': 'keepDims',\n        'type': 'bool',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LeakyRelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 0.2\n      },\n      {\n        'tfName': 'T',\n        'name': 'dtype',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'LoopCond',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'pred', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Switch',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'data', 'type': 'tensor'},\n      {'start': 1, 'name': 'pred', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Merge',\n    'category': 'control',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Enter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'frame_name', 'name': 'frameName', 'type': 'string'},\n      {'tfName': 'is_constant', 'name': 'isConstant', 'type': 'bool'}\n    ]\n  },\n  {\n    'tfOpName': 'Exit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NextIteration',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'size', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'dynamic_size', 'name': 'dynamicSize', 'type': 'bool'},\n      {'tfName': 'clear_after_read', 'name': 'clearAfterRead', 'type': 'bool'},\n      {\n        'tfName': 'identical_element_shapes',\n        'name': 'identicalElementShapes',\n        'type': 'bool'\n      },\n      {'tfName': 'tensor_array_name', 'name': 'name', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayWriteV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayReadV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{\n      'tfName': 'dtype',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  },\n  {\n    'tfOpName': 'TensorArrayGatherV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayScatterV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArrayConcatV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}, {\n        'tfName': 'element_shape_except0',\n        'name': 'elementShapeExcept0',\n        'type': 'shape',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'TensorArraySplitV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArraySizeV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayCloseV3',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'StatelessIf',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'If',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'StatelessWhile',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'While',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListScatter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListScatterV2',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 3, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGather',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListSetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListReserve',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 1, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListFromTensor',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListStack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs': [\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'},\n      {'tfName': 'num_elements', 'name': 'numElements', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListSplit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListConcat',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListPopBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListPushBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n];\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'AvgPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPoolWithArgmax',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'}, {\n        'tfName': 'include_batch_in_index',\n        'name': 'includeBatchInIndex',\n        'type': 'bool'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AvgPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Conv1D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'stride', 'name': 'stride', 'type': 'number'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NWC'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'dilation',\n        'name': 'dilation',\n        'type': 'number',\n        'defaultValue': 1\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'useCudnnOnGpu', 'name': 'useCudnnOnGpu', 'type': 'bool'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': '_FusedConv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'use_cudnn_on_gpu',\n        'name': 'useCudnnOnGpu',\n        'type': 'bool',\n        'defaultValue': true\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n    ]\n  },\n  {\n    'tfOpName': 'Conv2DBackpropInput',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 2, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 0, 'name': 'outputShape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2d',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'FusedDepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ],\n  },\n  {\n    'tfOpName': 'Dilation2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'rates', 'name': 'dilations', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Fill',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n      {'start': 1, 'name': 'value', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'LinSpace',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'num', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'OneHot',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'depth', 'type': 'number'},\n      {'start': 2, 'name': 'onValue', 'type': 'number', 'defaultValue': 1},\n      {'start': 3, 'name': 'offValue', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [\n      {\n        'tfName': 'axis',\n        'name': 'axis',\n        'type': 'number',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ones',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'OnesLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'RandomUniform',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'minval',\n        'name': 'minval',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'maxval',\n        'name': 'maxval',\n        'type': 'number',\n        'defaultValue': 1\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Range',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'step', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [{'tfName': 'Tidx', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TruncatedNormal',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'means',\n        'name': 'mean',\n        'type': 'number',\n        'defaultValue': 0.0\n      },\n      {\n        'tfName': 'stddev',\n        'name': 'stdDev',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Zeros',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'ZerosLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'Multinomial',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'logits', 'type': 'tensor'},\n      {'start': 1, 'name': 'numSamples', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'},\n      {'tfName': 'seed2', 'name': 'seed2', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'output_dtype', 'name': 'output_dtype', 'type': 'dtype'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'NonMaxSuppressionV2',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV3',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV4',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'T_threshold',\n        'name': 'threshold',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {\n        'tfName': 'pad_to_max_output_size',\n        'name': 'padToMaxOutputSize',\n        'type': 'bool'\n      }\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV5',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'},\n      {'start': 5, 'name': 'softNmsSigma', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Where',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ListDiff',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'TopKV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'k', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'sorted', 'name': 'sorted', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Unique',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n  },\n  {\n    'tfOpName': 'UniqueV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n  },\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'PlaceholderWithDefault',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'default', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'Placeholder',\n    'category': 'graph',\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {'tfOpName': 'Const', 'category': 'graph'}, {\n    'tfOpName': 'Identity',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IdentityN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Snapshot',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Rank',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Size',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Shape',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'ShapeN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Print',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'data', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'message', 'name': 'message', 'type': 'string'}, {\n        'tfName': 'first_n',\n        'name': 'firstN',\n        'type': 'number',\n        'notSupported': true\n      },\n      {\n        'tfName': 'summarize',\n        'name': 'summarize',\n        'type': 'number',\n        'defaultValue': 3\n      }\n    ]\n  },\n  {'tfOpName': 'NoOp', 'category': 'graph', 'inputs': []}, {\n    'tfOpName': 'StopGradient',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'FakeQuantWithMinMaxVars',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'min', 'name': 'min', 'type': 'number'},\n      {'tfName': 'max', 'name': 'max', 'type': 'number'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ResizeBilinear',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ResizeNearestNeighbor',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'CropAndResize',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'image', 'type': 'tensor'},\n      {'start': 1, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 2, 'name': 'boxInd', 'type': 'tensor'},\n      {'start': 3, 'name': 'cropSize', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'method', 'name': 'method', 'type': 'string'}, {\n        'tfName': 'extrapolation_value',\n        'name': 'extrapolationValue',\n        'type': 'number'\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Equal',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NotEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Greater',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'GreaterEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Less',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LessEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalAnd',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalNot',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalOr',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Select',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SelectV2',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': '_FusedMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'}, {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMulV2',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Transpose',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'perm', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FusedBatchNorm',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV2',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV3',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'LRN',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'depth_radius',\n        'name': 'radius',\n        'type': 'number',\n        'defaultValue': 5\n      },\n      {'tfName': 'bias', 'name': 'bias', 'type': 'number', 'defaultValue': 1.0},\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {\n        'tfName': 'beta',\n        'name': 'beta',\n        'type': 'number',\n        'defaultValue': 0.5\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Softmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'LogSoftmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': true,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Max',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Mean',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Min',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Sum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'All',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Any',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'ArgMax',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'ArgMin',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Cumsum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'exclusive', 'name': 'exclusive', 'type': 'bool'},\n      {'tfName': 'reverse', 'name': 'reverse', 'type': 'bool'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ConcatV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': -1, 'name': 'tensors', 'type': 'tensors'},\n      {'start': -1, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs':\n        [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n  },\n  {\n    'tfOpName': 'Concat',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 1, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n      {'start': 0, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs': [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n\n  },\n  {\n    'tfOpName': 'GatherV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Gather',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'validate_indices',\n        'name': 'validateIndices',\n        'type': 'bool',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reverse',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'dims', 'type': 'bool', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ReverseV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Slice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'size', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'StridedSlice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'end', 'type': 'number[]'},\n      {'start': 3, 'name': 'strides', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'begin_mask',\n        'name': 'beginMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'end_mask',\n        'name': 'endMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'new_axis_mask',\n        'name': 'newAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'ellipsis_mask',\n        'name': 'ellipsisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'shrink_axis_mask',\n        'name': 'shrinkAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Pack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Unpack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'num',\n        'name': 'num',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Tile',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'reps', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Split',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'axis', 'type': 'number', 'defaultValue': 0},\n      {'start': 1, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'num_split',\n      'name': 'numOrSizeSplits',\n      'type': 'number',\n      'defaultValue': 1\n    }]\n  },\n  {\n    'tfOpName': 'SplitV',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'numOrSizeSplits', 'type': 'number[]'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'ScatterNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'values', 'type': 'tensor'},\n      {'start': 2, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'GatherNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': false,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IFFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'RFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'IRFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Cast',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'SrcT',\n        'name': 'sdtype',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {'tfName': 'DstT', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'ExpandDims',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Pad',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'constant_value',\n      'name': 'constantValue',\n      'type': 'number',\n      'defaultValue': 0\n    }]\n  },\n  {\n    'tfOpName': 'PadV2',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'}, {\n        'start': 2,\n        'name': 'constantValue',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reshape',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Squeeze',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'axis',\n      'tfDeprecatedName': 'squeeze_dims',\n      'name': 'axis',\n      'type': 'number[]'\n    }]\n  },\n  {\n    'tfOpName': 'SpaceToBatchND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'paddings', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'BatchToSpaceND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'crops', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthToSpace',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'block_size', 'name': 'blockSize', 'type': 'number'},\n      {'tfName': 'data_format', 'name': 'dataFormat', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'BroadcastTo',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': []\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, env} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\n\nimport {getRegisteredOp} from './custom_op/register';\nimport {getNodeNameAndIndex} from './executors/utils';\nimport * as arithmetic from './op_list/arithmetic';\nimport * as basicMath from './op_list/basic_math';\nimport * as control from './op_list/control';\nimport * as convolution from './op_list/convolution';\nimport * as creation from './op_list/creation';\nimport * as dynamic from './op_list/dynamic';\nimport * as evaluation from './op_list/evaluation';\nimport * as graph from './op_list/graph';\nimport * as image from './op_list/image';\nimport * as logical from './op_list/logical';\nimport * as matrices from './op_list/matrices';\nimport * as normalization from './op_list/normalization';\nimport * as reduction from './op_list/reduction';\nimport * as sliceJoin from './op_list/slice_join';\nimport * as spectral from './op_list/spectral';\nimport * as transformation from './op_list/transformation';\nimport {Graph, InputParamValue, Node, OpMapper, ParamValue} from './types';\n\nexport class OperationMapper {\n  private static _instance: OperationMapper;\n\n  private opMappers: {[key: string]: OpMapper};\n\n  // Singleton instance for the mapper\n  public static get Instance() {\n    return this._instance || (this._instance = new this());\n  }\n\n  // Loads the op mapping from the JSON file.\n  private constructor() {\n    const ops = [\n      arithmetic, basicMath, control, convolution, creation, dynamic,\n      evaluation, logical, image, graph, matrices, normalization, reduction,\n      sliceJoin, spectral, transformation\n    ];\n    const mappersJson: OpMapper[] = [].concat(...ops.map(op => op.json));\n\n    this.opMappers = mappersJson.reduce<{[key: string]: OpMapper}>(\n        (map, mapper: OpMapper) => {\n          map[mapper.tfOpName] = mapper;\n          return map;\n        },\n        {});\n  }\n\n  // Converts the model inference graph from Tensorflow GraphDef to local\n  // representation for TensorFlow.js API\n  transformGraph(\n      graph: tensorflow.IGraphDef,\n      signature: tensorflow.ISignatureDef = {}): Graph {\n    const tfNodes = graph.node;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    const initNodes: Node[] = [];\n    const nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n      map[node.name] = this.mapNode(node);\n      if (node.op.startsWith('Placeholder')) {\n        placeholders.push(map[node.name]);\n      } else if (node.op === 'Const') {\n        weights.push(map[node.name]);\n      } else if (node.input == null || node.input.length === 0) {\n        initNodes.push(map[node.name]);\n      }\n      return map;\n    }, {});\n\n    let inputs: Node[] = [];\n    const outputs: Node[] = [];\n    let inputNodeNameToKey: {[key: string]: string} = {};\n    let outputNodeNameToKey: {[key: string]: string} = {};\n    if (signature != null) {\n      inputNodeNameToKey = this.mapSignatureEntries(signature.inputs);\n      outputNodeNameToKey = this.mapSignatureEntries(signature.outputs);\n    }\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    // if signature has not outputs set, add any node that does not have\n    // outputs.\n    if (Object.keys(outputNodeNameToKey).length === 0) {\n      allNodes.forEach(key => {\n        const node = nodes[key];\n        if (node.children.length === 0) {\n          outputs.push(node);\n        }\n      });\n    } else {\n      Object.keys(outputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node != null) {\n          node.signatureKey = outputNodeNameToKey[name];\n          outputs.push(node);\n        }\n      });\n    }\n\n    if (Object.keys(inputNodeNameToKey).length > 0) {\n      Object.keys(inputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node) {\n          node.signatureKey = inputNodeNameToKey[name];\n          inputs.push(node);\n        }\n      });\n    } else {\n      inputs = placeholders;\n    }\n\n    let functions = {};\n    if (graph.library != null && graph.library.function != null) {\n      functions = graph.library.function.reduce((functions, func) => {\n        functions[func.signature.name] = this.mapFunction(func);\n        return functions;\n      }, {} as {[key: string]: Graph});\n    }\n\n    const result: Graph =\n        {nodes, inputs, outputs, weights, placeholders, signature, functions};\n\n    if (initNodes.length > 0) {\n      result.initNodes = initNodes;\n    }\n\n    return result;\n  }\n\n  private mapSignatureEntries(entries: {[k: string]: tensorflow.ITensorInfo}) {\n    return Object.keys(entries || {})\n        .reduce<{[key: string]: string}>((prev, curr) => {\n          prev[entries[curr].name] = curr;\n          return prev;\n        }, {});\n  }\n\n  private mapNode(node: tensorflow.INodeDef): Node {\n    // Unsupported ops will cause an error at run-time (not parse time), since\n    // they may not be used by the actual execution subgraph.\n    const mapper =\n        getRegisteredOp(node.op) || this.opMappers[node.op] || {} as OpMapper;\n    if (node.attr == null) {\n      node.attr = {};\n    }\n\n    const newNode: Node = {\n      name: node.name,\n      op: node.op,\n      category: mapper.category,\n      inputNames:\n          (node.input ||\n           []).map(input => input.startsWith('^') ? input.substr(1) : input),\n      inputs: [],\n      children: [],\n      inputParams: {},\n      attrParams: {},\n      rawAttrs: node.attr\n    };\n\n    if (mapper.inputs != null) {\n      newNode.inputParams =\n          mapper.inputs.reduce<{[key: string]: InputParamValue}>(\n              (map, param) => {\n                map[param.name] = {\n                  type: param.type,\n                  inputIndexStart: param.start,\n                  inputIndexEnd: param.end\n                };\n                return map;\n              },\n              {});\n    }\n    if (mapper.attrs != null) {\n      newNode.attrParams =\n          mapper.attrs.reduce<{[key: string]: ParamValue}>((map, param) => {\n            const type = param.type;\n            let value = undefined;\n            switch (param.type) {\n              case 'string':\n                value = getStringParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'string[]':\n                value = getStringArrayParam(\n                    node.attr, param.tfName, param.defaultValue as string[]);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string[]);\n                }\n                break;\n              case 'number':\n                value = getNumberParam(\n                    node.attr, param.tfName,\n                    (param.defaultValue || 0) as number);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumberParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number);\n                }\n                break;\n              case 'number[]':\n                value = getNumericArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumericArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'bool':\n                value = getBoolParam(\n                    node.attr, param.tfName, param.defaultValue as boolean);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean);\n                }\n                break;\n              case 'bool[]':\n                value = getBoolArrayParam(\n                    node.attr, param.tfName, param.defaultValue as boolean[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean[]);\n                }\n                break;\n              case 'shape':\n                value = getTensorShapeParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'shape[]':\n                value = getTensorShapeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[][]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[][]);\n                }\n                break;\n              case 'dtype':\n                value = getDtypeParam(\n                    node.attr, param.tfName, param.defaultValue as DataType);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType);\n                }\n                break;\n              case 'dtype[]':\n                value = getDtypeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as DataType[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType[]);\n                }\n                break;\n              case 'func':\n                value = getFuncParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getFuncParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'tensor':\n              case 'tensors':\n                break;\n              default:\n                throw new Error(\n                    `Unsupported param type: ${param.type} for op: ${node.op}`);\n            }\n            map[param.name] = {value, type};\n            return map;\n          }, {});\n    }\n    return newNode;\n  }\n\n  // map the TFunctionDef to TFJS graph object\n  private mapFunction(functionDef: tensorflow.IFunctionDef): Graph {\n    const tfNodes = functionDef.nodeDef;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    let nodes: {[key: string]: Node} = {};\n    if (tfNodes != null) {\n      nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n        map[node.name] = this.mapNode(node);\n        if (node.op === 'Const') {\n          weights.push(map[node.name]);\n        }\n        return map;\n      }, {});\n    }\n    const inputs: Node[] = [];\n    const outputs: Node[] = [];\n\n    functionDef.signature.inputArg.forEach(arg => {\n      const [nodeName, ] = getNodeNameAndIndex(arg.name);\n      const node: Node = {\n        name: nodeName,\n        op: 'Placeholder',\n        inputs: [],\n        inputNames: [],\n        category: 'graph',\n        inputParams: {},\n        attrParams: {dtype: {value: parseDtypeParam(arg.type), type: 'dtype'}},\n        children: []\n      };\n      node.signatureKey = arg.name;\n      inputs.push(node);\n      nodes[nodeName] = node;\n    });\n\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    const returnNodeMap = functionDef.ret;\n\n    functionDef.signature.outputArg.forEach(output => {\n      const [nodeName, index] = getNodeNameAndIndex(returnNodeMap[output.name]);\n      const node = nodes[nodeName];\n      if (node != null) {\n        node.defaultOutput = index;\n        outputs.push(node);\n      }\n    });\n\n    const signature = this.mapArgsToSignature(functionDef);\n    return {nodes, inputs, outputs, weights, placeholders, signature};\n  }\n\n  private mapArgsToSignature(functionDef: tensorflow.IFunctionDef):\n      tensorflow.ISignatureDef {\n    return {\n      methodName: functionDef.signature.name,\n      inputs: functionDef.signature.inputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n      outputs: functionDef.signature.outputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg, functionDef.ret);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n    };\n  }\n\n  private mapArgToTensorInfo(\n      arg: tensorflow.OpDef.IArgDef,\n      nameMap?: {[key: string]: string}): tensorflow.ITensorInfo {\n    let name = arg.name;\n    if (nameMap != null) {\n      name = nameMap[name];\n    }\n    return {name, dtype: arg.type};\n  }\n}\n\nexport function decodeBase64(text: string): string {\n  const global = env().global;\n  if (typeof global.atob !== 'undefined') {\n    return global.atob(text);\n  } else if (typeof Buffer !== 'undefined') {\n    return new Buffer(text, 'base64').toString();\n  } else {\n    throw new Error(\n        'Unable to decode base64 in this environment. ' +\n        'Missing built-in atob() or Buffer()');\n  }\n}\n\nexport function parseStringParam(s: []|string, keepCase: boolean): string {\n  const value =\n      Array.isArray(s) ? String.fromCharCode.apply(null, s) : decodeBase64(s);\n  return keepCase ? value : value.toLowerCase();\n}\n\nexport function getStringParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string,\n    keepCase = false): string {\n  const param = attrs[name];\n  if (param != null) {\n    return parseStringParam(param.s, keepCase);\n  }\n  return def;\n}\n\nexport function getBoolParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean): boolean {\n  const param = attrs[name];\n  return param ? param.b : def;\n}\n\nexport function getNumberParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number): number {\n  const param = attrs[name] || {};\n  const value =\n      param['i'] != null ? param['i'] : (param['f'] != null ? param['f'] : def);\n  return (typeof value === 'number') ? value : parseInt(value, 10);\n}\n\nexport function parseDtypeParam(value: string|tensorflow.DataType): DataType {\n  if (typeof (value) === 'string') {\n    // tslint:disable-next-line:no-any\n    value = tensorflow.DataType[value as any];\n  }\n  switch (value) {\n    case tensorflow.DataType.DT_FLOAT:\n      return 'float32';\n    case tensorflow.DataType.DT_INT32:\n    case tensorflow.DataType.DT_INT64:\n    case tensorflow.DataType.DT_INT8:\n    case tensorflow.DataType.DT_UINT8:\n      return 'int32';\n    case tensorflow.DataType.DT_BOOL:\n      return 'bool';\n    case tensorflow.DataType.DT_DOUBLE:\n      return 'float32';\n    case tensorflow.DataType.DT_STRING:\n      return 'string';\n    default:\n      // Unknown dtype error will happen at runtime (instead of parse time),\n      // since these nodes might not be used by the actual subgraph execution.\n      return null;\n  }\n}\n\nexport function getFuncParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: string): string {\n  const param = attrs[name];\n  if (param && param.func) {\n    return param.func.name;\n  }\n  return def;\n}\n\nexport function getDtypeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType): DataType {\n  const param = attrs[name];\n  if (param && param.type) {\n    return parseDtypeParam(param.type);\n  }\n  return def;\n}\n\nexport function getDtypeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType[]): DataType[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.type) {\n    return param.list.type.map(v => parseDtypeParam(v));\n  }\n  return def;\n}\n\nexport function parseTensorShapeParam(shape: tensorflow.ITensorShape): number[]|\n    undefined {\n  if (shape.unknownRank) {\n    return undefined;\n  }\n  if (shape.dim != null) {\n    return shape.dim.map(\n        dim =>\n            (typeof dim.size === 'number') ? dim.size : parseInt(dim.size, 10));\n  }\n  return [];\n}\n\nexport function getTensorShapeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def?: number[]): number[]|undefined {\n  const param = attrs[name];\n  if (param && param.shape) {\n    return parseTensorShapeParam(param.shape);\n  }\n  return def;\n}\n\nexport function getNumericArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[]): number[] {\n  const param = attrs[name];\n  if (param) {\n    return ((param.list.f && param.list.f.length ? param.list.f :\n                                                   param.list.i) ||\n            [])\n        .map(v => (typeof v === 'number') ? v : parseInt(v, 10));\n  }\n  return def;\n}\n\nexport function getStringArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string[],\n    keepCase = false): string[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.s) {\n    return param.list.s.map((v) => {\n      return parseStringParam(v, keepCase);\n    });\n  }\n  return def;\n}\n\nexport function getTensorShapeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[][]): number[][] {\n  const param = attrs[name];\n  if (param && param.list && param.list.shape) {\n    return param.list.shape.map((v) => {\n      return parseTensorShapeParam(v);\n    });\n  }\n  return def;\n}\n\nexport function getBoolArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean[]): boolean[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.b) {\n    return param.list.b;\n  }\n  return def;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {getTensor} from '../executors/utils';\nimport {getBoolArrayParam, getBoolParam, getDtypeArrayParam, getDtypeParam, getNumberParam, getNumericArrayParam, getStringArrayParam, getStringParam, getTensorShapeArrayParam, getTensorShapeParam} from '../operation_mapper';\nimport {GraphNode, Node, ValueType} from '../types';\n\n/**\n * Helper class for lookup inputs and params for nodes in the model graph.\n */\nexport class NodeValueImpl implements GraphNode {\n  public readonly inputs: Tensor[] = [];\n  public readonly attrs: {[key: string]: ValueType} = {};\n  constructor(\n      private node: Node, private tensorMap: NamedTensorsMap,\n      private context: ExecutionContext) {\n    this.inputs = node.inputNames.map(name => this.getInput(name));\n    if (node.rawAttrs != null) {\n      this.attrs = Object.keys(node.rawAttrs)\n                       .reduce((attrs: {[key: string]: ValueType}, key) => {\n                         attrs[key] = this.getAttr(key);\n                         return attrs;\n                       }, {});\n    }\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getInput(name: string): Tensor {\n    return getTensor(name, this.tensorMap, this.context);\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getAttr(name: string, defaultValue?: ValueType): ValueType {\n    const value = this.node.rawAttrs[name];\n    if (value.tensor != null) {\n      return getTensor(name, this.tensorMap, this.context);\n    }\n    if (value.i != null || value.f != null) {\n      return getNumberParam(this.node.rawAttrs, name, defaultValue as number);\n    }\n    if (value.s != null) {\n      return getStringParam(this.node.rawAttrs, name, defaultValue as string);\n    }\n    if (value.b != null) {\n      return getBoolParam(this.node.rawAttrs, name, defaultValue as boolean);\n    }\n    if (value.shape != null) {\n      return getTensorShapeParam(\n          this.node.rawAttrs, name, defaultValue as number[]);\n    }\n    if (value.type != null) {\n      return getDtypeParam(this.node.rawAttrs, name, defaultValue as DataType);\n    }\n    if (value.list != null) {\n      if (value.list.i != null || value.list.f != null) {\n        return getNumericArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[]);\n      }\n      if (value.list.s != null) {\n        return getStringArrayParam(\n            this.node.rawAttrs, name, defaultValue as string[]);\n      }\n      if (value.list.shape != null) {\n        return getTensorShapeArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[][]);\n      }\n      if (value.list.b != null) {\n        return getBoolArrayParam(\n            this.node.rawAttrs, name, defaultValue as boolean[]);\n      }\n      if (value.list.type != null) {\n        return getDtypeArrayParam(\n            this.node.rawAttrs, name, defaultValue as DataType[]);\n      }\n    }\n\n    return defaultValue;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BiasAdd':\n        case 'AddV2':\n        case 'Add': {\n          return [tfOps.add(\n              (getParamValue('a', node, tensorMap, context) as Tensor),\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'AddN': {\n          return [tfOps.addN((\n              getParamValue('tensors', node, tensorMap, context) as Tensor[]))];\n        }\n        case 'FloorMod':\n        case 'Mod':\n          return [tfOps.mod(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'Mul':\n          return [tfOps.mul(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'RealDiv':\n        case 'Div': {\n          return [tfOps.div(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'DivNoNan': {\n          return [tfOps.divNoNan(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'FloorDiv': {\n          return [tfOps.floorDiv(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sub': {\n          return [tfOps.sub(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Minimum': {\n          return [tfOps.minimum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Maximum': {\n          return [tfOps.maximum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Pow': {\n          return [tfOps.pow(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'SquaredDifference': {\n          return [tfOps.squaredDifference(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'arithmetic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Abs':\n        case 'ComplexAbs':\n          return [tfOps.abs(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acos':\n          return [tfOps.acos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acosh':\n          return [tfOps.acosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asin':\n          return [tfOps.asin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asinh':\n          return [tfOps.asinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan':\n          return [tfOps.atan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan2':\n          return [tfOps.atan2(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('y', node, tensorMap, context) as Tensor)];\n        case 'Atanh':\n          return [tfOps.atanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Ceil':\n          return [tfOps.ceil(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Complex':\n          return [tfOps.complex(\n              getParamValue('real', node, tensorMap, context) as Tensor,\n              getParamValue('imag', node, tensorMap, context) as Tensor)];\n        case 'Cos':\n          return [tfOps.cos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Cosh':\n          return [tfOps.cosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Elu':\n          return [tfOps.elu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Erf':\n          return [tfOps.erf(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Exp':\n          return [tfOps.exp(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Expm1': {\n          return [tfOps.expm1(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Floor':\n          return [tfOps.floor(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log':\n          return [tfOps.log(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log1p': {\n          return [tfOps.log1p(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Imag':\n          return [tfOps.imag(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n\n        case 'Neg':\n          return [tfOps.neg(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Reciprocal': {\n          return [tfOps.reciprocal(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Real':\n          return [tfOps.real(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu':\n          return [tfOps.relu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Round': {\n          return [tfOps.round(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Selu':\n          return [tfOps.selu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sigmoid':\n          return [tfOps.sigmoid(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sin':\n          return [tfOps.sin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sign': {\n          return [tfOps.sign(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sinh': {\n          return [tfOps.sinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Softplus': {\n          return [tfOps.softplus(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sqrt': {\n          return [tfOps.sqrt(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Square': {\n          return [tfOps.square(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tanh': {\n          return [tfOps.tanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tan':\n          return [tfOps.tan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu6':\n        case 'ClipByValue':\n          return [tfOps.clipByValue(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('clipValueMin', node, tensorMap, context) as number,\n              getParamValue('clipValueMax', node, tensorMap, context) as\n                  number)];\n        case 'Rsqrt':\n          return [tfOps.rsqrt(\n              getTensor(node.inputNames[0], tensorMap, context))];\n        case 'Prod':\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('axes', node, tensorMap, context) as number[])];\n        case 'LeakyRelu':\n          return [tfOps.leakyRelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as number)];\n        case 'Prelu':\n          return [tfOps.prelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as Tensor)];\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'basic_math';\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n/**\n * This differs from util.assertShapesMatch in that it allows values of\n * negative one, an undefined size of a dimensinon, in a shape to match\n * anything.\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function assertShapesMatchAllowUndefinedSize(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  util.assert(\n      shapesEqualAllowUndefinedSize(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function shapesEqualAllowUndefinedSize(n1: number[], n2: number[]) {\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== -1 && n2[i] !== -1 && n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\nexport interface TensorWithState {\n  tensor?: Tensor;\n  written?: boolean;\n  read?: boolean;\n  cleared?: boolean;\n}\n/**\n * The TensorArray object keeps an array of Tensors.  It\n * allows reading from the array and writing to the array.\n */\nexport class TensorArray {\n  private tensors: TensorWithState[] = [];\n  private closed_ = false;\n  readonly idTensor: Tensor;\n  constructor(\n      readonly name: string, readonly dtype: DataType, private maxSize: number,\n      private elementShape: number[], readonly identicalElementShapes: boolean,\n      readonly dynamicSize: boolean, readonly clearAfterRead: boolean) {\n    this.idTensor = scalar(0);\n    keep(this.idTensor);\n  }\n\n  get id() {\n    return this.idTensor.id;\n  }\n\n  get closed() {\n    return this.closed_;\n  }\n\n  /**\n   * Dispose the tensors and idTensor and mark the TensoryArray as closed.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.tensor.id)) {\n        tensor.tensor.dispose();\n      }\n    });\n    this.tensors = [];\n    this.closed_ = true;\n    this.idTensor.dispose();\n  }\n\n  size(): number {\n    return this.tensors.length;\n  }\n\n  /**\n   * Read the value at location index in the TensorArray.\n   * @param index Number the index to read from.\n   */\n  read(index: number): Tensor {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || index >= this.size()) {\n      throw new Error(`Tried to read from index ${index}, but array size is: ${\n          this.size()}`);\n    }\n\n    const tensorWithState = this.tensors[index];\n    if (tensorWithState.cleared) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not read index ${\n              index} twice because it was cleared after a previous read ` +\n          `(perhaps try setting clear_after_read = false?).`);\n    }\n\n    if (this.clearAfterRead) {\n      tensorWithState.cleared = true;\n    }\n\n    tensorWithState.read = true;\n    return tensorWithState.tensor;\n  }\n\n  /**\n   * Helper method to read multiple tensors from the specified indices.\n   */\n  readMany(indices: number[]): Tensor[] {\n    return indices.map(index => this.read(index));\n  }\n\n  /**\n   * Write value into the index of the TensorArray.\n   * @param index number the index to write to.\n   * @param tensor\n   */\n  write(index: number, tensor: Tensor) {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || !this.dynamicSize && index >= this.maxSize) {\n      throw new Error(`Tried to write to index ${\n          index}, but array is not resizeable and size is: ${this.maxSize}`);\n    }\n\n    const t = this.tensors[index] || {};\n\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray ${\n          this.name}: Could not write to TensorArray index ${index},\n          because the value dtype is ${\n          tensor.dtype}, but TensorArray dtype is ${this.dtype}.`);\n    }\n\n    // Set the shape for the first time write to unknow shape tensor array\n    if (this.size() === 0 &&\n        (this.elementShape == null || this.elementShape.length === 0)) {\n      this.elementShape = tensor.shape;\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape,\n        `TensorArray ${this.name}: Could not write to TensorArray index ${\n            index}.`);\n\n    if (t.read) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been read.`);\n    }\n\n    if (t.written) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been written.`);\n    }\n\n    t.tensor = tensor;\n    keep(tensor);\n    t.written = true;\n\n    this.tensors[index] = t;\n  }\n\n  /**\n   * Helper method to write multiple tensors to the specified indices.\n   */\n  writeMany(indices: number[], tensors: Tensor[]) {\n    if (indices.length !== tensors.length) {\n      throw new Error(\n          `TensorArray ${this.name}: could not write multiple tensors,` +\n          `because the index size: ${\n              indices.length} is not the same as tensors size: ${\n              tensors.length}.`);\n    }\n\n    indices.forEach((i, index) => this.write(i, tensors[index]));\n  }\n\n  /**\n   * Return selected values in the TensorArray as a packed Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param [indices] number[] Optional. Taking values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size(). If not specified returns\n   *    all tensors in the original order.\n   * @param [dtype]\n   */\n  gather(indices?: number[], dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but gather requested dtype ${dtype}`);\n    }\n\n    if (!indices) {\n      indices = [];\n      for (let i = 0; i < this.size(); i++) {\n        indices.push(i);\n      }\n    } else {\n      indices = indices.slice(0, this.size());\n    }\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    // Read all the PersistentTensors into a vector to keep track of\n    // their memory.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape, 'TensorArray shape mismatch: ');\n\n    return stack(tensors, 0);\n  }\n\n  /**\n   * Return the values in the TensorArray as a concatenated Tensor.\n   */\n  concat(dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but concat requested dtype ${dtype}`);\n    }\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    const indices = [];\n    for (let i = 0; i < this.size(); i++) {\n      indices.push(i);\n    }\n    // Collect all the tensors from the tensors array.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape,\n        `TensorArray shape mismatch: tensor array shape (${\n            this.elementShape}) vs first tensor shape (${tensors[0].shape})`);\n\n    return concat(tensors, 0);\n  }\n\n  /**\n   * Scatter the values of a Tensor in specific indices of a TensorArray.\n   * @param indices nummber[] values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size().\n   * @param tensor Tensor input tensor.\n   */\n  scatter(indices: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n\n    if (indices.length !== tensor.shape[0]) {\n      throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n          indices.length} vs. ${tensor.shape[0]}`);\n    }\n\n    const maxIndex = Math.max(...indices);\n\n    if (!this.dynamicSize && maxIndex >= this.maxSize) {\n      throw new Error(\n          `Max index must be < array size (${maxIndex}  vs. ${this.maxSize})`);\n    }\n\n    this.writeMany(indices, unstack(tensor, 0));\n  }\n\n  /**\n   * Split the values of a Tensor into the TensorArray.\n   * @param length number[] with the lengths to use when splitting value along\n   *    its first dimension.\n   * @param tensor Tensor, the tensor to split.\n   */\n  split(length: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n    let totalLength = 0;\n    const cumulativeLengths = length.map(len => {\n      totalLength += len;\n      return totalLength;\n    });\n\n    if (totalLength !== tensor.shape[0]) {\n      throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n    }\n\n    if (!this.dynamicSize && length.length !== this.maxSize) {\n      throw new Error(\n          `TensorArray's size is not equal to the size of lengths (${\n              this.maxSize} vs. ${length.length}), ` +\n          'and the TensorArray is not marked as dynamically resizeable');\n    }\n\n    const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n    const tensors: Tensor[] = [];\n    tidy(() => {\n      tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n      for (let i = 0; i < length.length; ++i) {\n        const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n        const indices = [0, previousLength, 0];\n        const sizes = [1, length[i], elementPerRow];\n        tensors[i] = reshape(slice(tensor, indices, sizes), this.elementShape);\n      }\n      return tensors;\n    });\n    const indices = [];\n    for (let i = 0; i < length.length; i++) {\n      indices[i] = i;\n    }\n    this.writeMany(indices, tensors);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\n/**\n * TensorList stores a container of `tf.Tensor` objects, which are accessible\n * via tensors field.\n *\n * In order to get a copy of the underlying list, use the copy method:\n * ```\n *    TensorList b = a.copy();\n *    b.tensors().pushBack(t);  // This does not modify a.tensors().\n * ```\n *\n * Note that this is not a deep copy: the memory locations of the underlying\n * tensors will still point to the same locations of the corresponding tensors\n * in the original.\n */\n\nexport class TensorList {\n  readonly idTensor: Tensor;\n  maxNumElements: number;\n\n  get id() {\n    return this.idTensor.id;\n  }\n  /**\n   *\n   * @param tensors list of tensors\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param maxNumElements The maximum allowed size of `tensors`. Defaults to -1\n   *   meaning that the size of `tensors` is unbounded.\n   */\n  constructor(\n      readonly tensors: Tensor[], readonly elementShape: number[],\n      readonly elementDtype: DataType, maxNumElements = -1) {\n    if (tensors != null) {\n      tensors.forEach(tensor => {\n        if (elementDtype !== tensor.dtype) {\n          throw new Error(`Invalid data types; op elements ${\n              elementDtype}, but list elements ${tensor.dtype}`);\n        }\n        assertShapesMatchAllowUndefinedSize(\n            elementShape, tensor.shape, 'TensorList shape mismatch: ');\n\n        keep(tensor);\n      });\n    }\n    this.idTensor = scalar(0);\n    this.maxNumElements = maxNumElements;\n    keep(this.idTensor);\n  }\n\n  /**\n   * Get a new TensorList containing a copy of the underlying tensor container.\n   */\n  copy(): TensorList {\n    return new TensorList(\n        [...this.tensors], this.elementShape, this.elementDtype);\n  }\n\n  /**\n   * Dispose the tensors and idTensor and clear the tensor list.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.id)) {\n        tensor.dispose();\n      }\n    });\n    this.tensors.length = 0;\n    this.idTensor.dispose();\n  }\n  /**\n   * The size of the tensors in the tensor list.\n   */\n  size() {\n    return this.tensors.length;\n  }\n\n  /**\n   * Return a tensor that stacks a list of rank-R tf.Tensors into one rank-(R+1)\n   * tf.Tensor.\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param numElements the number of elements to stack\n   */\n  stack(elementShape: number[], elementDtype: DataType, numElements = -1):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (numElements !== -1 && this.tensors.length !== numElements) {\n      throw new Error(`Operation expected a list with ${\n          numElements} elements but got a list with ${\n          this.tensors.length} elements.`);\n    }\n    assertShapesMatchAllowUndefinedSize(\n        elementShape, this.elementShape, 'TensorList shape mismatch: ');\n    return tidy(() => {\n      const reshapedTensors =\n          this.tensors.map(tensor => reshape(tensor, elementShape));\n      return stack(reshapedTensors, 0);\n    });\n  }\n\n  /**\n   * Pop a tensor from the end of the list.\n   * @param elementShape shape of the tensor\n   * @param elementDtype data type of the tensor\n   */\n  popBack(elementShape: number[], elementDtype: DataType): Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (this.size() === 0) {\n      throw new Error('Trying to pop from an empty list.');\n    }\n\n    const tensor = this.tensors.pop();\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, elementShape, 'TensorList shape mismatch: ');\n    return reshape(tensor, elementShape);\n  }\n\n  /**\n   * Push a tensor to the end of the list.\n   * @param tensor Tensor to be pushed.\n   */\n  pushBack(tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, this.elementShape, 'TensorList shape mismatch: ');\n\n    if (this.maxNumElements === this.size()) {\n      throw new Error(`Trying to push element into a full list.`);\n    }\n    keep(tensor);\n    this.tensors.push(tensor);\n  }\n\n  /**\n   * Update the size of the list.\n   * @param size the new size of the list.\n   */\n  resize(size: number) {\n    if (size < 0) {\n      throw new Error(\n          `TensorListResize expects size to be non-negative. Got: ${size}`);\n    }\n\n    if (this.maxNumElements !== -1 && size > this.maxNumElements) {\n      throw new Error(`TensorListResize input size ${\n          size} is greater maxNumElement ${this.maxNumElements}.`);\n    }\n    this.tensors.length = size;\n  }\n\n  /**\n   * Retrieve the element at the provided index\n   * @param elementShape shape of the tensor\n   * @param elementDtype dtype of the tensor\n   * @param elementIndex index of the tensor\n   */\n  getItem(elementIndex: number, elementShape: number[], elementDtype: DataType):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (elementIndex < 0 || elementIndex > this.tensors.length) {\n      throw new Error(`Trying to access element ${\n          elementIndex} in a list with ${this.tensors.length} elements.`);\n    }\n\n    if (this.tensors[elementIndex] == null) {\n      throw new Error(`element at index ${elementIndex} is null.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.tensors[elementIndex].shape, elementShape,\n        'TensorList shape mismatch: ');\n\n    return this.tensors[elementIndex];\n  }\n\n  /**\n   * Set the tensor at the index\n   * @param elementIndex index of the tensor\n   * @param tensor the tensor to be inserted into the list\n   */\n  setItem(elementIndex: number, tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (elementIndex < 0 ||\n        this.maxNumElements !== -1 && elementIndex >= this.maxNumElements) {\n      throw new Error(`Trying to set element ${\n          elementIndex} in a list with max ${this.maxNumElements} elements.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape, 'TensorList shape mismatch: ');\n    keep(tensor);\n    this.tensors[elementIndex] = tensor;\n  }\n\n  /**\n   * Return selected values in the TensorList as a stacked Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param indices indices of tensors to gather\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  gather(indices: number[], elementDtype: DataType, elementShape: number[]):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    // When indices is greater than the size of the list, indices beyond the\n    // size of the list are ignored.\n    indices = indices.slice(0, this.size());\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = indices.map(i => reshape(this.tensors[i], elementShape));\n      return stack(tensors, 0);\n    });\n  }\n\n  /**\n   * Return the values in the TensorList as a concatenated Tensor.\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  concat(elementDtype: DataType, elementShape: number[]): Tensor {\n    if (!!elementDtype && elementDtype !== this.elementDtype) {\n      throw new Error(`TensorList dtype is ${\n          this.elementDtype} but concat requested dtype ${elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = this.tensors.map(t => reshape(t, elementShape));\n      return concat(tensors, 0);\n    });\n  }\n}\n\n/**\n * Creates a TensorList which, when stacked, has the value of tensor.\n * @param tensor from tensor\n * @param elementShape output tensor element shape\n */\nexport function fromTensor(\n    tensor: Tensor, elementShape: number[], elementDtype: DataType) {\n  const dtype = tensor.dtype;\n  if (tensor.shape.length < 1) {\n    throw new Error(\n        `Tensor must be at least a vector, but saw shape: ${tensor.shape}`);\n  }\n  if (tensor.dtype !== elementDtype) {\n    throw new Error(`Invalid data types; op elements ${\n        tensor.dtype}, but list elements ${elementDtype}`);\n  }\n  const outputShape = tensor.shape.slice(1);\n  assertShapesMatchAllowUndefinedSize(\n      outputShape, elementShape, 'TensorList shape mismatch: ');\n\n  const tensorList: Tensor[] = unstack(tensor);\n  return new TensorList(tensorList, elementShape, dtype);\n}\n\n/**\n * Return a TensorList of the given size with empty elements.\n * @param elementShape the shape of the future elements of the list\n * @param elementDtype the desired type of elements in the list\n * @param numElements the number of elements to reserve\n */\nexport function reserve(\n    elementShape: number[], elementDtype: DataType, numElements: number) {\n  return new TensorList([], elementShape, elementDtype, numElements);\n}\n\n/**\n * Put tensors at specific indices of a stacked tensor into a TensorList.\n * @param indices list of indices on how to scatter the tensor.\n * @param tensor input tensor.\n * @param elementShape the shape of the future elements of the list\n * @param numElements the number of elements to scatter\n */\nexport function scatter(\n    tensor: Tensor, indices: number[], elementShape: number[],\n    numElements?: number): TensorList {\n  if (indices.length !== tensor.shape[0]) {\n    throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n        indices.length} vs. ${tensor.shape[0]}`);\n  }\n\n  const maxIndex = Math.max(...indices);\n\n  if (numElements != null && numElements !== -1 && maxIndex >= numElements) {\n    throw new Error(\n        `Max index must be < array size (${maxIndex}  vs. ${numElements})`);\n  }\n\n  const list = new TensorList([], elementShape, tensor.dtype, numElements);\n  const tensors = unstack(tensor, 0);\n  indices.forEach((value, index) => {\n    list.setItem(value, tensors[index]);\n  });\n  return list;\n}\n\n/**\n * Split the values of a Tensor into a TensorList.\n * @param length the lengths to use when splitting value along\n *    its first dimension.\n * @param tensor the tensor to split.\n * @param elementShape the shape of the future elements of the list\n */\nexport function split(\n    tensor: Tensor, length: number[], elementShape: number[]) {\n  let totalLength = 0;\n  const cumulativeLengths = length.map(len => {\n    totalLength += len;\n    return totalLength;\n  });\n\n  if (totalLength !== tensor.shape[0]) {\n    throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n  }\n\n  const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n  const tensors: Tensor[] = tidy(() => {\n    const tensors = [];\n    tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n    for (let i = 0; i < length.length; ++i) {\n      const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n      const indices = [0, previousLength, 0];\n      const sizes = [1, length[i], elementPerRow];\n      tensors[i] = reshape(slice(tensor, indices, sizes), elementShape);\n    }\n    tensor.dispose();\n    return tensors;\n  });\n\n  const list = new TensorList([], elementShape, tensor.dtype, length.length);\n\n  for (let i = 0; i < tensors.length; i++) {\n    list.setItem(i, tensors[i]);\n  }\n  return list;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, scalar, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {TensorArray} from '../../executor/tensor_array';\nimport {fromTensor, reserve, scatter, split} from '../../executor/tensor_list';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'If':\n    case 'StatelessIf': {\n      const thenFunc =\n          getParamValue('thenBranch', node, tensorMap, context) as string;\n      const elseFunc =\n          getParamValue('elseBranch', node, tensorMap, context) as string;\n      const cond = getParamValue('cond', node, tensorMap, context) as Tensor;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n      const condValue = await cond.data();\n      if (condValue[0]) {\n        return context.functionMap[thenFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      } else {\n        return context.functionMap[elseFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      }\n    }\n    case 'While':\n    case 'StatelessWhile': {\n      const bodyFunc =\n          getParamValue('body', node, tensorMap, context) as string;\n      const condFunc =\n          getParamValue('cond', node, tensorMap, context) as string;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n\n      // Calculate the condition of the loop\n      const condResult =\n          (await context.functionMap[condFunc].executeFunctionAsync(\n              args, context.tensorArrayMap, context.tensorListMap));\n      const argIds = args.map(tensor => tensor.id);\n      let condValue = await condResult[0].data();\n      // Dispose the intermediate tensors for condition function\n      condResult.forEach(tensor => {\n        if (!tensor.kept && argIds.indexOf(tensor.id) === -1) {\n          tensor.dispose();\n        }\n      });\n\n      let result: Tensor[] = args;\n\n      while (condValue[0]) {\n        // Record the previous result for intermediate tensor tracking\n        const origResult = result;\n        // Execution the body of the loop\n        result = await context.functionMap[bodyFunc].executeFunctionAsync(\n            result, context.tensorArrayMap, context.tensorListMap);\n        const resultIds = result.map(tensor => tensor.id);\n\n        // Dispose the intermediate tensor for body function that is not global\n        // kept, not input/output of the body function\n        origResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n\n        // Recalcuate the condition of the loop using the latest results.\n        const condResult =\n            (await context.functionMap[condFunc].executeFunctionAsync(\n                result, context.tensorArrayMap, context.tensorListMap));\n        condValue = await condResult[0].data();\n        // Dispose the intermediate tensors for condition function\n        condResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n      }\n      return result;\n    }\n    case 'LoopCond': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      return [cloneTensor(pred)];\n    }\n    case 'Switch': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      let data = getParamValue('data', node, tensorMap, context) as Tensor;\n      if (!data.kept) {\n        data = cloneTensor(data);\n      }\n      // Outputs nodes :0 => false, :1 => true\n      return (await pred.data())[0] ? [undefined, data] : [data, undefined];\n    }\n    case 'Merge': {\n      const inputName = node.inputNames.find(\n          name => getTensor(name, tensorMap, context) !== undefined);\n      if (inputName) {\n        const data = getTensor(inputName, tensorMap, context);\n        return [cloneTensor(data)];\n      }\n      return undefined;\n    }\n    case 'Enter': {\n      const frameId =\n          getParamValue('frameName', node, tensorMap, context) as string;\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.enterFrame(frameId);\n      return [cloneTensor(data)];\n    }\n    case 'Exit': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.exitFrame();\n      return [cloneTensor(data)];\n    }\n    case 'NextIteration': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.nextIteration();\n      return [cloneTensor(data)];\n    }\n    case 'TensorArrayV3': {\n      const size = getParamValue('size', node, tensorMap, context) as number;\n      const dtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const dynamicSize =\n          getParamValue('dynamicSize', node, tensorMap, context) as boolean;\n      const clearAfterRead =\n          getParamValue('clearAfterRead', node, tensorMap, context) as boolean;\n      const identicalElementShapes =\n          getParamValue('identicalElementShapes', node, tensorMap, context) as\n          boolean;\n      const name = getParamValue('name', node, tensorMap, context) as string;\n      const tensorArray = new TensorArray(\n          name, dtype, size, elementShape, identicalElementShapes, dynamicSize,\n          clearAfterRead);\n      context.addTensorArray(tensorArray);\n      return [tensorArray.idTensor, scalar(1.0)];\n    }\n    case 'TensorArrayWriteV3': {\n      const id =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const writeTensorArray = context.getTensorArray(id.id);\n      writeTensorArray.write(index, writeTensor);\n      return [writeTensorArray.idTensor];\n    }\n    case 'TensorArrayReadV3': {\n      const readId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const readTensorArray = context.getTensorArray(readId.id);\n      return [readTensorArray.read(readIndex)];\n    }\n    case 'TensorArrayGatherV3': {\n      const gatherId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const gatherDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const gatherTensorArray = context.getTensorArray(gatherId.id);\n      return [gatherTensorArray.gather(gatherIndices, gatherDtype)];\n    }\n    case 'TensorArrayScatterV3': {\n      const scatterId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const scatterTensorArray = context.getTensorArray(scatterId.id);\n      scatterTensorArray.scatter(scatterIndices, scatterTensor);\n      return [scatterTensorArray.idTensor];\n    }\n    case 'TensorArrayConcatV3': {\n      const concatId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const concatTensorArray = context.getTensorArray(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      return [concatTensorArray.concat(concatDtype)];\n    }\n    case 'TensorArraySplitV3': {\n      const splitId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n      const splitTensorArray = context.getTensorArray(splitId.id);\n      splitTensorArray.split(lengths, splitTensor);\n      return [splitTensorArray.idTensor];\n    }\n    case 'TensorArraySizeV3': {\n      const sizeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const sizeTensorArray = context.getTensorArray(sizeId.id);\n      return [scalar(sizeTensorArray.size(), 'int32')];\n    }\n    case 'TensorArrayCloseV3': {\n      const closeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const closeTensorArray = context.getTensorArray(closeId.id);\n      closeTensorArray.clearAndClose();\n      return [closeTensorArray.idTensor];\n    }\n    case 'TensorListSetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.setItem(index, writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.getItem(readIndex, elementShape, elementDType)];\n    }\n    case 'TensorListScatterV2':\n    case 'TensorListScatter': {\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList =\n          scatter(scatterTensor, scatterIndices, elementShape, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListReserve': {\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = reserve(elementShape, elementDtype, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGather': {\n      const gatherId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(gatherId.id);\n      return [tensorList.gather(gatherIndices, elementDtype, elementShape)];\n    }\n    case 'TensorListStack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.stack(elementShape, elementDtype, numElements)];\n    }\n    case 'TensorListFromTensor': {\n      const tensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = fromTensor(tensor, elementShape, elementDtype);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListConcat': {\n      const concatId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      return [tensorList.concat(concatDtype, elementShape)];\n    }\n    case 'TensorListPushBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.pushBack(writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListPopBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.popBack(elementShape, elementDType)];\n    }\n    case 'TensorListSplit': {\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n\n      const tensorList = split(splitTensor, lengths, elementShape);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'control';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rank, Tensor, Tensor3D, Tensor4D, Tensor5D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getPadding, getParamValue} from './utils';\n\nfunction fusedConvAndDepthWiseParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const [extraOp, activationFunc] =\n      (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n  const isBiasAdd = extraOp === 'biasadd';\n  const isPrelu = activationFunc === 'prelu';\n  const isBatchNorm = extraOp === 'fusedbatchnorm';\n\n  const numArgs =\n      (getParamValue('numArgs', node, tensorMap, context) as number);\n  if (isBiasAdd) {\n    if (isPrelu && numArgs !== 2) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu ' +\n          'must have two extra arguments: bias and alpha.');\n    }\n    if (!isPrelu && numArgs !== 1) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd must have ' +\n          'one extra argument: bias.');\n    }\n  }\n  if (isBatchNorm) {\n    throw new Error(\n        'FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.');\n  }\n  const stride = getParamValue('strides', node, tensorMap, context) as number[];\n  const pad = getPadding(node, tensorMap, context);\n  const dataFormat =\n      (getParamValue('dataFormat', node, tensorMap, context) as string)\n          .toUpperCase();\n  const dilations =\n      getParamValue('dilations', node, tensorMap, context) as number[];\n  const [biasArg, preluArg] =\n      getParamValue('args', node, tensorMap, context) as Tensor[];\n\n  return {\n    stride,\n    pad,\n    dataFormat,\n    dilations,\n    biasArg,\n    preluArg,\n    activationFunc\n  };\n}\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Conv1D': {\n          const stride =\n              getParamValue('stride', node, tensorMap, context) as number;\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilation =\n              getParamValue('dilation', node, tensorMap, context) as number;\n          return [tfOps.conv1d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              stride, pad as 'valid' | 'same', dataFormat as 'NWC' | 'NCW',\n              dilation)];\n        }\n        case 'Conv2D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case '_FusedConv2D': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.conv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n\n        case 'FusedDepthwiseConv2dNative': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.depthwiseConv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n        case 'Conv2DBackpropInput':\n        case 'Conv2dTranspose': {\n          const shape = getParamValue(\n                            'outputShape', node, tensorMap,\n                            context) as [number, number, number] |\n              [number, number, number, number];\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          return [tfOps.conv2dTranspose(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              shape, [stride[1], stride[2]], pad as 'valid' | 'same')];\n        }\n        case 'DepthwiseConv2dNative':\n        case 'DepthwiseConv2d': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n\n          return [tfOps.depthwiseConv2d(\n              getParamValue('input', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case 'Conv3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv3d(\n              getParamValue('x', node, tensorMap, context) as Tensor4D |\n                  Tensor<Rank.R5>,\n              getParamValue('filter', node, tensorMap, context) as\n                  Tensor<Rank.R5>,\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same',\n              dataFormat as 'NDHWC' | 'NCDHW',\n              [dilations[1], dilations[2], dilations[3]])];\n        }\n        case 'AvgPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPoolWithArgmax': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n          const includeBatchInIndex =\n              getParamValue('includeBatchInIndex', node, tensorMap, context) as\n              boolean;\n          const {result, indexes} = tfOps.maxPoolWithArgmax(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same', includeBatchInIndex);\n          return [result, indexes];\n        }\n        case 'AvgPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'MaxPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'Dilation2D': {\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n\n          // strides: [1, stride_height, stride_width, 1].\n          const strideHeight = strides[1];\n          const strideWidth = strides[2];\n\n          // dilations: [1, dilation_height, dilation_width, 1].\n          const dilationHeight = dilations[1];\n          const dilationWidth = dilations[2];\n\n          return [tfOps.dilation2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              [strideHeight, strideWidth], pad as 'valid' | 'same',\n              [dilationHeight, dilationWidth], 'NHWC' /* dataFormat */)];\n        }\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'convolution';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor, Tensor1D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Fill': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const dtype =\n              getParamValue('dtype', node, tensorMap, context) as DataType;\n          const value =\n              getParamValue('value', node, tensorMap, context) as number;\n          return [tfOps.fill(shape, value, dtype)];\n        }\n        case 'LinSpace': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const num = getParamValue('num', node, tensorMap, context) as number;\n          return [tfOps.linspace(start, stop, num)];\n        }\n        case 'Multinomial': {\n          const logits =\n              getParamValue('logits', node, tensorMap, context) as Tensor1D;\n          const numSamples =\n              getParamValue('numSamples', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.multinomial(logits, numSamples, seed)];\n        }\n        case 'OneHot': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          const depth =\n              getParamValue('depth', node, tensorMap, context) as number;\n          const onValue =\n              getParamValue('onValue', node, tensorMap, context) as number;\n          const offValue =\n              getParamValue('offValue', node, tensorMap, context) as number;\n          return [tfOps.oneHot(indices, depth, onValue, offValue)];\n        }\n        case 'Ones': {\n          return [tfOps.ones(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'OnesLike': {\n          return [tfOps.onesLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'RandomUniform': {\n          return [tfOps.randomUniform(\n              // tslint:disable-next-line:no-any\n              getParamValue('shape', node, tensorMap, context) as any,\n              getParamValue('minval', node, tensorMap, context) as number,\n              getParamValue('maxval', node, tensorMap, context) as number,\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'Range': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const step =\n              getParamValue('step', node, tensorMap, context) as number;\n          return [tfOps.range(\n              start, stop, step,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32')];\n        }\n        case 'TruncatedNormal': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const mean =\n              getParamValue('mean', node, tensorMap, context) as number;\n          const stdDev =\n              getParamValue('stdDev', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.truncatedNormal(\n              shape, mean, stdDev,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32',\n              seed)];\n        }\n        case 'Zeros': {\n          return [tfOps.zeros(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'ZerosLike': {\n          return [tfOps.zerosLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'creation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nfunction nmsParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const boxes = getParamValue('boxes', node, tensorMap, context) as Tensor;\n  const scores = getParamValue('scores', node, tensorMap, context) as Tensor;\n  const maxOutputSize =\n      getParamValue('maxOutputSize', node, tensorMap, context) as number;\n  const iouThreshold =\n      getParamValue('iouThreshold', node, tensorMap, context) as number;\n  const scoreThreshold =\n      getParamValue('scoreThreshold', node, tensorMap, context) as number;\n  const softNmsSigma =\n      getParamValue('softNmsSigma', node, tensorMap, context) as number;\n\n  return {\n    boxes,\n    scores,\n    maxOutputSize,\n    iouThreshold,\n    scoreThreshold,\n    softNmsSigma\n  };\n}\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'NonMaxSuppressionV5': {\n      const {\n        boxes,\n        scores,\n        maxOutputSize,\n        iouThreshold,\n        scoreThreshold,\n        softNmsSigma\n      } = nmsParams(node, tensorMap, context);\n\n      const result = await tfOps.image.nonMaxSuppressionWithScoreAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, softNmsSigma);\n\n      return [result.selectedIndices, result.selectedScores];\n    }\n    case 'NonMaxSuppressionV4': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      const padToMaxOutputSize =\n          getParamValue('padToMaxOutputSize', node, tensorMap, context) as\n          boolean;\n\n      const result = await tfOps.image.nonMaxSuppressionPaddedAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, padToMaxOutputSize);\n\n      return [result.selectedIndices, result.validOutputs];\n    }\n    case 'NonMaxSuppressionV3':\n    case 'NonMaxSuppressionV2': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      return [await tfOps.image.nonMaxSuppressionAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold)];\n    }\n    case 'Where': {\n      const condition = tfOps.cast(\n          (getParamValue('condition', node, tensorMap, context) as Tensor),\n          'bool');\n      const result = [await tfOps.whereAsync(condition)];\n      condition.dispose();\n      return result;\n    }\n    case 'ListDiff': {\n      return tfOps.setdiff1dAsync(\n          getParamValue('x', node, tensorMap, context) as Tensor,\n          getParamValue('y', node, tensorMap, context) as Tensor);\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'dynamic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'TopKV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const k = getParamValue('k', node, tensorMap, context) as number;\n              const sorted =\n                  getParamValue('sorted', node, tensorMap, context) as boolean;\n              const result = tfOps.topk(x, k, sorted);\n              return [result.values, result.indices];\n            }\n            case 'Unique': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const result = tfOps.unique(x);\n              return [result.values, result.indices];\n            }\n            case 'UniqueV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const axis =\n                  getParamValue('axis', node, tensorMap, context) as number;\n              const result = tfOps.unique(x, axis);\n              return [result.values, result.indices];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'evaluation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Const': {\n          return tensorMap[node.name];\n        }\n        case 'PlaceholderWithDefault':\n          const def =\n              getParamValue('default', node, tensorMap, context) as Tensor;\n          return [getTensor(node.name, tensorMap, context) || def];\n        case 'Placeholder':\n          return [getTensor(node.name, tensorMap, context)];\n        case 'Identity':\n        case 'StopGradient':\n        case 'FakeQuantWithMinMaxVars': {  // This op is currently ignored.\n          const data = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [cloneTensor(data)];\n        }\n        case 'IdentityN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => cloneTensor(t));\n        case 'Snapshot':\n          const snapshot =\n              (getParamValue('x', node, tensorMap, context) as Tensor);\n          return [cloneTensor(snapshot)];\n        case 'Shape':\n          return [tfOps.tensor1d(\n              (getParamValue('x', node, tensorMap, context) as Tensor).shape,\n              'int32')];\n        case 'ShapeN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => tfOps.tensor1d(t.shape));\n        case 'Size':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).size,\n              'int32')];\n        case 'Rank':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).rank,\n              'int32')];\n        case 'NoOp':\n          return [tfOps.scalar(1)];\n        case 'Print':\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const data =\n              getParamValue('data', node, tensorMap, context) as Tensor[];\n          const message =\n              getParamValue('message', node, tensorMap, context) as string;\n          const summarize =\n              getParamValue('summarize', node, tensorMap, context) as number;\n          console.warn(\n              'The graph has a tf.print() operation,' +\n              'usually used for debugging, which slows down performance.');\n          console.log(message);\n          for (let i = 0; i < data.length; i++) {\n            console.log(Array.prototype.slice.call(data[i].dataSync())\n                            .slice(0, summarize));\n          }\n          return [input];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'graph';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ResizeBilinear': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeBilinear(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'ResizeNearestNeighbor': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeNearestNeighbor(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'CropAndResize': {\n          const image =\n              getParamValue('image', node, tensorMap, context) as Tensor;\n          const boxes =\n              getParamValue('boxes', node, tensorMap, context) as Tensor;\n          const boxInd =\n              getParamValue('boxInd', node, tensorMap, context) as Tensor;\n          const cropSize =\n              getParamValue('cropSize', node, tensorMap, context) as number[];\n          const method =\n              getParamValue('method', node, tensorMap, context) as string;\n          const extrapolationValue =\n              getParamValue('extrapolationValue', node, tensorMap, context) as\n              number;\n          return [tfOps.image.cropAndResize(\n              image as Tensor4D, boxes as Tensor2D, boxInd as Tensor1D,\n              cropSize as [number, number], method as 'bilinear' | 'nearest',\n              extrapolationValue)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'image';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Equal': {\n          return [tfOps.equal(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'NotEqual': {\n          return [tfOps.notEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Greater': {\n          return [tfOps.greater(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'GreaterEqual': {\n          return [tfOps.greaterEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Less': {\n          return [tfOps.less(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LessEqual': {\n          return [tfOps.lessEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalAnd': {\n          return [tfOps.logicalAnd(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalNot': {\n          return [tfOps.logicalNot(\n              getParamValue('a', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalOr': {\n          return [tfOps.logicalOr(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Select':\n        case 'SelectV2': {\n          return [tfOps.where(\n              getParamValue('condition', node, tensorMap, context) as Tensor,\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'logical';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BatchMatMul':\n        case 'BatchMatMulV2':\n        case 'MatMul':\n          return [tfOps.matMul(\n              getParamValue('a', node, tensorMap, context) as Tensor2D,\n              getParamValue('b', node, tensorMap, context) as Tensor2D,\n              getParamValue('transposeA', node, tensorMap, context) as boolean,\n              getParamValue('transposeB', node, tensorMap, context) as\n                  boolean)];\n\n        case 'Transpose':\n          return [tfOps.transpose(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('perm', node, tensorMap, context) as number[])];\n\n        case '_FusedMatMul':\n          const [extraOp, activationFunc] =\n              (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n          const isBiasAdd = extraOp === 'biasadd';\n          const isPrelu = activationFunc === 'prelu';\n\n          const numArgs =\n              (getParamValue('numArgs', node, tensorMap, context) as number);\n          if (isBiasAdd) {\n            if (isPrelu && numArgs !== 2) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd and Prelu must have two ' +\n                  'extra arguments: bias and alpha.');\n            }\n            if (!isPrelu && numArgs !== 1) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd must have one extra argument: bias.');\n            }\n          }\n          const [biasArg, preluArg] =\n              getParamValue('args', node, tensorMap, context) as Tensor[];\n          return [tfOps.fused.matMul({\n            a: getParamValue('a', node, tensorMap, context) as Tensor2D,\n            b: getParamValue('b', node, tensorMap, context) as Tensor2D,\n            transposeA: getParamValue('transposeA', node, tensorMap, context) as\n                boolean,\n            transposeB: getParamValue('transposeB', node, tensorMap, context) as\n                boolean,\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'matrices';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'FusedBatchNorm':\n        case 'FusedBatchNormV2': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'FusedBatchNormV3': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'LRN': {\n          return [tfOps.localResponseNormalization(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('radius', node, tensorMap, context) as number,\n              getParamValue('bias', node, tensorMap, context) as number,\n              getParamValue('alpha', node, tensorMap, context) as number,\n              getParamValue('beta', node, tensorMap, context) as number)];\n        }\n        case 'Softmax': {\n          return [tfOps.softmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogSoftmax': {\n          return [tfOps.logSoftmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'SparseToDense': {\n          return [tfOps.sparseToDense(\n              getParamValue('sparseIndices', node, tensorMap, context) as\n                  Tensor,\n              getParamValue('outputShape', node, tensorMap, context) as Tensor,\n              getParamValue('sparseValues', node, tensorMap, context) as\n                  number[],\n              getParamValue('defaultValue', node, tensorMap, context) as\n                  Scalar)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'normalization';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Max': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.max(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Mean': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.mean(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Min': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.min(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Sum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.sum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'All': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.all(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Any': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.any(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'ArgMax': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMax(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'ArgMin': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMin(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Prod': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Cumsum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const exclusive =\n              getParamValue('exclusive', node, tensorMap, context) as boolean;\n          const reverse =\n              getParamValue('reverse', node, tensorMap, context) as boolean;\n          return [tfOps.cumsum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              exclusive, reverse)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'reduction';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ConcatV2':\n        case 'Concat': {\n          const n = getParamValue('n', node, tensorMap, context) as number;\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          let inputs =\n              getParamValue('tensors', node, tensorMap, context) as Tensor[];\n          inputs = inputs.slice(0, n);\n          return [tfOps.concat(inputs, axis)];\n        }\n        case 'GatherV2':\n        case 'Gather': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          return [tfOps.gather(input, tfOps.cast(indices, 'int32'), axis)];\n        }\n        case 'ReverseV2':\n        case 'Reverse': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [tfOps.reverse(input, axis)];\n        }\n        case 'Slice': {\n          // tslint:disable-next-line:no-any\n          const begin = getParamValue('begin', node, tensorMap, context) as any;\n          // tslint:disable-next-line:no-any\n          const size = getParamValue('size', node, tensorMap, context) as any;\n          return [tfOps.slice(\n              getParamValue('x', node, tensorMap, context) as Tensor, begin,\n              size)];\n        }\n        case 'StridedSlice': {\n          const begin =\n              getParamValue('begin', node, tensorMap, context) as number[];\n          const end =\n              getParamValue('end', node, tensorMap, context) as number[];\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const beginMask =\n              getParamValue('beginMask', node, tensorMap, context) as number;\n          const endMask =\n              getParamValue('endMask', node, tensorMap, context) as number;\n          const ellipsisMask =\n              getParamValue('ellipsisMask', node, tensorMap, context) as number;\n          const newAxisMask =\n              getParamValue('newAxisMask', node, tensorMap, context) as number;\n          const shrinkAxisMask =\n              getParamValue('shrinkAxisMask', node, tensorMap, context) as\n              number;\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return [tfOps.stridedSlice(\n              tensor, begin, end, strides, beginMask, endMask, ellipsisMask,\n              newAxisMask, shrinkAxisMask)];\n        }\n        case 'Pack': {\n          return tidy(() => {\n            const axis =\n                getParamValue('axis', node, tensorMap, context) as number;\n            const tensors =\n                getParamValue('tensors', node, tensorMap, context) as Tensor[];\n            // Reshape the tensors to the first tensor's shape if they don't\n            // match.\n            const shape = tensors[0].shape;\n            const squeezedShape = tfOps.squeeze(tensors[0]).shape;\n            const mapped = tensors.map(tensor => {\n              const sameShape = util.arraysEqual(tensor.shape, shape);\n              if (!sameShape &&\n                  !util.arraysEqual(\n                      tfOps.squeeze(tensor).shape, squeezedShape)) {\n                throw new Error('the input tensors shape does not match');\n              }\n              return sameShape ? tensor : tfOps.reshape(tensor, shape);\n            });\n            return [tfOps.stack(mapped, axis)];\n          });\n        }\n        case 'Unpack': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const tensor =\n              getParamValue('tensor', node, tensorMap, context) as Tensor;\n          return tfOps.unstack(tensor, axis);\n        }\n        case 'Tile': {\n          const reps =\n              getParamValue('reps', node, tensorMap, context) as number[];\n          return [tfOps.tile(\n              getParamValue('x', node, tensorMap, context) as Tensor, reps)];\n        }\n        case 'Split':\n        case 'SplitV': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const numOrSizeSplits =\n              getParamValue('numOrSizeSplits', node, tensorMap, context) as\n                  number |\n              number[];\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return tfOps.split(tensor, numOrSizeSplits, axis);\n        }\n        case 'ScatterNd': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          const values =\n              getParamValue('values', node, tensorMap, context) as Tensor;\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          return [tfOps.scatterND(indices, values, shape)];\n        }\n        case 'GatherNd': {\n          const x = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          return [tfOps.gatherND(x, indices)];\n        }\n        case 'SparseToDense': {\n          const indices =\n              getParamValue('sparseIndices', node, tensorMap, context) as\n              Tensor;\n          const shape =\n              getParamValue('outputShape', node, tensorMap, context) as\n              number[];\n          const sparseValues =\n              getParamValue('sparseValues', node, tensorMap, context) as Tensor;\n          const defaultValue =\n              getParamValue('defaultValue', node, tensorMap, context) as Scalar;\n          return [tfOps.sparseToDense(\n              indices, sparseValues, shape,\n              sparseValues.dtype === defaultValue.dtype ?\n                  defaultValue :\n                  tfOps.cast(defaultValue, sparseValues.dtype))];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'slice_join';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'FFT': {\n              return [tfOps.fft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IFFT': {\n              return [tfOps.ifft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'RFFT': {\n              return [tfOps.rfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IRFFT': {\n              return [tfOps.irfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'spectral';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Cast': {\n          return [tfOps.cast(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('dtype', node, tensorMap, context) as 'int32' |\n                  'float32' | 'bool')];\n        }\n        case 'ExpandDims': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.expandDims(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Squeeze': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          return [tfOps.squeeze(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n\n        case 'Reshape': {\n          return [tfOps.reshape(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        case 'PadV2':\n        case 'Pad': {\n          return [tfOps.pad(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('padding', node, tensorMap, context) as\n                  Array<[number, number]>,\n              getParamValue('constantValue', node, tensorMap, context) as\n                  number)];\n        }\n        case 'SpaceToBatchND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const paddings =\n              getParamValue('paddings', node, tensorMap, context) as number[][];\n          return [tfOps.spaceToBatchND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, paddings)];\n        }\n        case 'BatchToSpaceND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const crops =\n              getParamValue('crops', node, tensorMap, context) as number[][];\n          return [tfOps.batchToSpaceND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, crops)];\n        }\n        case 'DepthToSpace': {\n          const blockSize =\n              getParamValue('blockSize', node, tensorMap, context) as number;\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as\n               string).toUpperCase() as 'NHWC' |\n              'NCHW';\n          return [tfOps.depthToSpace(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              blockSize, dataFormat)];\n        }\n        case 'BroadcastTo': {\n          return [tfOps.broadcastTo(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'transformation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {ExecutionContext} from '../executor/execution_context';\n\nimport {NodeValueImpl} from './custom_op/node_value_impl';\nimport {getRegisteredOp} from './custom_op/register';\nimport * as arithmetic from './executors/arithmetic_executor';\nimport * as basicMath from './executors/basic_math_executor';\nimport * as control from './executors/control_executor';\nimport * as convolution from './executors/convolution_executor';\nimport * as creation from './executors/creation_executor';\nimport * as dynamic from './executors/dynamic_executor';\nimport * as evaluation from './executors/evaluation_executor';\nimport * as graph from './executors/graph_executor';\nimport * as image from './executors/image_executor';\nimport * as logical from './executors/logical_executor';\nimport * as matrices from './executors/matrices_executor';\nimport * as normalization from './executors/normalization_executor';\nimport * as reduction from './executors/reduction_executor';\nimport * as sliceJoin from './executors/slice_join_executor';\nimport * as spectral from './executors/spectral_executor';\nimport * as transformation from './executors/transformation_executor';\nimport {Node} from './types';\n\n/**\n * Executes the op defined by the node object.\n * @param node\n * @param tensorMap contains tensors for executed nodes and weights\n */\nexport function executeOp(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): tfc.Tensor[]|Promise<tfc.Tensor[]> {\n  const value =\n      ((node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) => {\n        switch (node.category) {\n          case 'arithmetic':\n            return tfc.tidy(\n                () => arithmetic.executeOp(node, tensorMap, context));\n          case 'basic_math':\n            return tfc.tidy(\n                () => basicMath.executeOp(node, tensorMap, context));\n          case 'control':\n            return control.executeOp(node, tensorMap, context);\n          case 'convolution':\n            return tfc.tidy(\n                () => convolution.executeOp(node, tensorMap, context));\n          case 'creation':\n            return tfc.tidy(() => creation.executeOp(node, tensorMap, context));\n          case 'dynamic':\n            return dynamic.executeOp(node, tensorMap, context);\n          case 'evaluation':\n            return tfc.tidy(\n                () => evaluation.executeOp(node, tensorMap, context));\n          case 'image':\n            return tfc.tidy(() => image.executeOp(node, tensorMap, context));\n          case 'graph':\n            return tfc.tidy(() => graph.executeOp(node, tensorMap, context));\n          case 'logical':\n            return tfc.tidy(() => logical.executeOp(node, tensorMap, context));\n          case 'matrices':\n            return tfc.tidy(() => matrices.executeOp(node, tensorMap, context));\n          case 'normalization':\n            return tfc.tidy(\n                () => normalization.executeOp(node, tensorMap, context));\n          case 'reduction':\n            return tfc.tidy(\n                () => reduction.executeOp(node, tensorMap, context));\n          case 'slice_join':\n            return tfc.tidy(\n                () => sliceJoin.executeOp(node, tensorMap, context));\n          case 'spectral':\n            return tfc.tidy(() => spectral.executeOp(node, tensorMap, context));\n          case 'transformation':\n            return tfc.tidy(\n                () => transformation.executeOp(node, tensorMap, context));\n          case 'custom':\n            const opMapper = getRegisteredOp(node.op);\n            if (opMapper && opMapper.customExecutor) {\n              return opMapper.customExecutor(\n                  new NodeValueImpl(node, tensorMap, context));\n            } else {\n              throw TypeError(`Custom op ${node.op} is not registered.`);\n            }\n          default:\n            throw TypeError(\n                `Unknown op '${node.op}'. File an issue at ` +\n                `https://github.com/tensorflow/tfjs/issues so we can add it` +\n                `, or register a custom execution with tf.registerOp()`);\n        }\n      })(node, tensorMap, context);\n  if (value instanceof Promise) {\n    return value.then((data) => [].concat(data));\n  }\n  return [].concat(value);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap, TensorArrayMap, TensorListMap} from '../data/types';\n\nimport {TensorArray} from './tensor_array';\nimport {TensorList} from './tensor_list';\nimport {FunctionExecutor} from './types';\n\nexport interface ExecutionContextInfo {\n  id: number;           // the unique id of the context info\n  frameName: string;    // The frame name of the loop, this comes from\n                        // the TensorFlow NodeDef.\n  iterationId: number;  // The iteration id of the loop\n}\n\n/**\n * ExecutionContext captures the runtime environment of the node. It keeps\n * track of the current frame and iteration for the control flow ops.\n *\n * For example, typical Dynamic RNN model may contain loops, for which\n * TensorFlow will generate graphs with Enter/Exit nodes to control the\n * current execution frame, and NextIteration Nodes for iteration id increment.\n * For model with branch logic, TensorFLow will generate Switch/Merge ops.\n */\nexport class ExecutionContext {\n  private rootContext = {id: 0, frameName: '', iterationId: 0};\n  private contexts: ExecutionContextInfo[] = [this.rootContext];\n  private lastId = 0;\n  private _currentContextIds: string[];\n\n  constructor(\n      readonly weightMap: NamedTensorsMap = {},\n      readonly tensorArrayMap: TensorArrayMap = {},\n      readonly tensorListMap: TensorListMap = {},\n      readonly functionMap: {[key: string]: FunctionExecutor} = {}) {\n    this.generateCurrentContextIds();\n  }\n\n  private newFrame(id: number, frameName: string) {\n    return {id, frameName, iterationId: 0};\n  }\n\n  /**\n   * Set the current context\n   * @param contexts: ExecutionContextInfo[] the current path of execution\n   * frames\n   */\n  set currentContext(contexts: ExecutionContextInfo[]) {\n    if (this.contexts !== contexts) {\n      this.contexts = contexts;\n      this.generateCurrentContextIds();\n    }\n  }\n\n  get currentContext(): ExecutionContextInfo[] {\n    return this.contexts;\n  }\n\n  /**\n   * Returns the current context in string format.\n   */\n  get currentContextId(): string {\n    return this._currentContextIds[0];\n  }\n\n  /**\n   * Returns the current context and all parent contexts in string format.\n   * This allow access to the nodes in the current and parent frames.\n   */\n  get currentContextIds(): string[] {\n    return this._currentContextIds;\n  }\n\n  private generateCurrentContextIds() {\n    const names = [];\n    for (let i = 0; i < this.contexts.length - 1; i++) {\n      const contexts = this.contexts.slice(0, this.contexts.length - i);\n      names.push(this.contextIdforContexts(contexts));\n    }\n    names.push('');\n    this._currentContextIds = names;\n  }\n\n  private contextIdforContexts(contexts: ExecutionContextInfo[]) {\n    return contexts ?\n        contexts\n            .map(\n                context => (context.id === 0 && context.iterationId === 0) ?\n                    '' :\n                    `${context.frameName}-${context.iterationId}`)\n            .join('/') :\n        '';\n  }\n\n  /**\n   * Enter a new frame, a new context is pushed on the current context list.\n   * @param frameId new frame id\n   */\n  enterFrame(frameId: string) {\n    if (this.contexts) {\n      this.lastId++;\n      this.contexts = this.contexts.slice();\n      this.contexts.push(this.newFrame(this.lastId, frameId));\n      this._currentContextIds.unshift(this.contextIdforContexts(this.contexts));\n    }\n  }\n\n  /**\n   * Exit the current frame, the last context is removed from the current\n   * context list.\n   */\n  exitFrame() {\n    if (this.contexts && this.contexts.length > 1) {\n      this.contexts = this.contexts.slice();\n      this.contexts.splice(-1);\n      this.currentContextIds.shift();\n    } else {\n      throw new Error('Cannot exit frame, the context is empty');\n    }\n  }\n\n  /**\n   * Enter the next iteration of a loop, the iteration id of last context is\n   * increased.\n   */\n  nextIteration() {\n    if (this.contexts && this.contexts.length > 0) {\n      this.contexts = this.contexts.slice();\n      this.lastId++;\n      const context =\n          Object.assign({}, this.contexts[this.contexts.length - 1]);\n      context.iterationId += 1;\n      context.id = this.lastId;\n      this.contexts.splice(-1, 1, context);\n      this._currentContextIds.splice(\n          0, 1, this.contextIdforContexts(this.contexts));\n    } else {\n      throw new Error('Cannot increase frame iteration, the context is empty');\n    }\n  }\n\n  getWeight(name: string): Tensor[] {\n    return this.weightMap[name];\n  }\n\n  addTensorArray(tensorArray: TensorArray) {\n    this.tensorArrayMap[tensorArray.id] = tensorArray;\n  }\n\n  getTensorArray(id: number): TensorArray {\n    return this.tensorArrayMap[id];\n  }\n\n  addTensorList(tensorList: TensorList) {\n    this.tensorListMap[tensorList.id] = tensorList;\n  }\n\n  getTensorList(id: number): TensorList {\n    return this.tensorListMap[id];\n  }\n\n  dispose(keepIds: Set<number>) {\n    for (const key in this.tensorArrayMap) {\n      this.tensorArrayMap[key].clearAndClose(keepIds);\n    }\n\n    for (const key in this.tensorListMap) {\n      this.tensorListMap[key].clearAndClose(keepIds);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NamedTensorMap} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {parseNodeName} from '../operations/executors/utils';\nimport {Graph, Node} from '../operations/types';\n\nexport interface ExecutionInfo {\n  inputs: NamedTensorMap;\n  outputs: Node[];\n  usedNodes: Set<string>;\n  missingInputs: string[];\n  dynamicNode: Node;\n  syncInputs: string[];\n}\n\n/**\n * Given graph inputs and desired outputs, find the minimal set of nodes\n * to execute in order to compute the outputs. In addition return other useful\n * info such:\n * - Missing inputs needed to compute the output.\n * - Whether the subgraph contains dynamic ops (control flow, dynamic shape).\n * - Alternative inputs in order to avoid async (dynamic op) execution.\n */\nexport function getExecutionSubgraph(\n    inputs: NamedTensorMap, outputs: Node[], weightMap: NamedTensorsMap,\n    initNodes?: Node[]): ExecutionInfo {\n  const usedNodes = new Set<string>();\n  const missingInputs: string[] = [];\n  let dynamicNode: Node = null;\n  let syncInputs: string[] = null;\n\n  // Start with the outputs, going backwards and find all the nodes that are\n  // needed to compute those outputs.\n  const seen = new Set<string>();\n  const inputNodeNames =\n      Object.keys(inputs).map(name => parseNodeName(name)[0]);\n\n  let initNodeNames: string[] = [];\n  if (initNodes != null) {\n    initNodeNames = initNodes.map(node => parseNodeName(node.name)[0]);\n  }\n\n  const frontier = [...outputs];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    if (isControlFlow(node) || isDynamicShape(node)) {\n      if (dynamicNode == null) {\n        dynamicNode = node;\n        syncInputs = dynamicNode.children.map(child => child.name)\n                         .filter(name => usedNodes.has(name));\n      }\n    }\n    usedNodes.add(node.name);\n\n    // Weights are dead end since we already have their values.\n    if (weightMap[node.name] != null) {\n      continue;\n    }\n    // This node is a dead end since it's one of the user-provided inputs.\n    if (inputNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    // This node is a dead end since it doesn't have any inputs.\n    if (initNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    if (node.inputs.length === 0) {\n      missingInputs.push(node.name);\n      continue;\n    }\n    node.inputs.forEach(input => {\n      // Don't add to the frontier if it is already there.\n      if (seen.has(input.name)) {\n        return;\n      }\n      seen.add(input.name);\n      frontier.push(input);\n    });\n  }\n  return {inputs, outputs, usedNodes, missingInputs, dynamicNode, syncInputs};\n}\n\n/**\n * Given the execution info, return a list of nodes in topological order that\n * need to be executed to compute the output.\n */\nexport function getNodesInTopologicalOrder(\n    graph: Graph, weightMap: NamedTensorsMap,\n    executionInfo: ExecutionInfo): Node[] {\n  const {usedNodes, inputs} = executionInfo;\n  const frontier: Node[] = [];\n  const inputNodes = Object.keys(inputs)\n                         .map(name => parseNodeName(name)[0])\n                         .map(name => graph.nodes[name]);\n  const initNodes = graph.initNodes;\n\n  inputNodes.forEach(input => {\n    if (usedNodes.has(input.name)) {\n      frontier.push(input);\n    }\n  });\n  graph.weights.forEach(weight => {\n    if (usedNodes.has(weight.name)) {\n      frontier.push(weight);\n    }\n  });\n  if (initNodes != null) {\n    initNodes.forEach(node => {\n      if (usedNodes.has(node.name)) {\n        frontier.push(node);\n      }\n    });\n  }\n  const seen = new Set<string>();\n  const orderedNodes: Node[] = [];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    seen.add(node.name);\n    if (!weightMap[node.name]) {\n      orderedNodes.push(node);\n    }\n    node.children.forEach(child => {\n      if (!seen.has(child.name) && usedNodes.has(child.name) &&\n          child.inputs.every(input => seen.has(input.name))) {\n        frontier.push(child);\n      }\n    });\n  }\n  return orderedNodes;\n}\n\nconst CONTROL_FLOW_OPS = [\n  'Switch', 'Merge', 'Enter', 'Exit', 'NextIteration', 'StatelessIf',\n  'StatelessWhile', 'if', 'While'\n];\nconst DYNAMIC_SHAPE_OPS = [\n  'NonMaxSuppressionV2', 'NonMaxSuppressionV3', 'NonMaxSuppressionV5', 'Where'\n];\n\nexport function isControlFlow(node: Node) {\n  return CONTROL_FLOW_OPS.indexOf(node.op) >= 0;\n}\n\nexport function isDynamicShape(node: Node) {\n  return DYNAMIC_SHAPE_OPS.indexOf(node.op) >= 0;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NamedTensorMap, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {ISignatureDef} from '../data/compiled_api';\nimport {NamedTensorsMap, TensorArrayMap, TensorInfo, TensorListMap} from '../data/types';\nimport {getNodeNameAndIndex, getParamValue, getTensor, getTensorsForCurrentContenxt, parseNodeName} from '../operations/executors/utils';\nimport {executeOp} from '../operations/operation_executor';\nimport {Graph, Node} from '../operations/types';\n\nimport {ExecutionContext, ExecutionContextInfo} from './execution_context';\nimport {getExecutionSubgraph, getNodesInTopologicalOrder, isControlFlow} from './model_analysis';\nimport {FunctionExecutor} from './types';\n\ninterface NodeWithContexts {\n  contexts: ExecutionContextInfo[];\n  node: Node;\n}\n\nexport class GraphExecutor implements FunctionExecutor {\n  private compiledMap: Map<string, Node[]> = new Map();\n  private _weightMap: NamedTensorsMap = {};\n  private _weightIds: number[];\n  private _signature: ISignatureDef;\n  private _inputs: Node[];\n  private _outputs: Node[];\n  private _initNodes: Node[];\n  private SEPERATOR = ',';\n  private _functions: {[key: string]: Graph} = {};\n  private _functionExecutorMap: {[key: string]: FunctionExecutor} = {};\n\n  get weightIds(): number[] {\n    return this.parent ? this.parent.weightIds : this._weightIds;\n  }\n\n  get functionExecutorMap(): {[key: string]: FunctionExecutor} {\n    return this.parent ? this.parent.functionExecutorMap :\n                         this._functionExecutorMap;\n  }\n\n  get weightMap(): NamedTensorsMap {\n    return this.parent ? this.parent.weightMap : this._weightMap;\n  }\n\n  set weightMap(weightMap: NamedTensorsMap) {\n    const weightIds = Object.keys(weightMap).map(\n        key => weightMap[key].map(tensor => tensor.id));\n    this._weightIds = [].concat(...weightIds);\n    this._weightMap = weightMap;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this._inputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get outputs(): TensorInfo[] {\n    return this._outputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get inputNodes(): string[] {\n    return this._inputs.map(node => node.signatureKey || node.name);\n  }\n\n  get outputNodes(): string[] {\n    return this._outputs.map((node) => {\n      const name = node.signatureKey || node.name;\n      return node.defaultOutput ? (`${name}:${node.defaultOutput}`) : name;\n    });\n  }\n\n  get functions(): {[key: string]: ISignatureDef} {\n    return Object.keys(this._functions).reduce((map, key) => {\n      map[key] = this._functions[key].signature;\n      return map;\n    }, {} as {[key: string]: ISignatureDef});\n  }\n\n  /**\n   *\n   * @param graph Graph the model or function graph to be executed.\n   * @param parent When building function exector you need to set the parent\n   * executor. Since the weights and function executor maps are set at parant\n   * level, that function executor can access the function maps and weight maps\n   * through the parent.\n   */\n  constructor(private graph: Graph, private parent?: GraphExecutor) {\n    this._outputs = graph.outputs;\n    this._inputs = graph.inputs;\n    this._initNodes = graph.initNodes;\n    this._signature = graph.signature;\n    this._functions = graph.functions;\n    // create sub-graph executors\n    if (graph.functions != null) {\n      Object.keys(graph.functions).forEach(name => {\n        this._functionExecutorMap[name] =\n            new GraphExecutor(graph.functions[name], this);\n      });\n    }\n  }\n\n  private getCompilationKey(inputs: Node[], outputs: Node[]): string {\n    const sortedInputs = inputs.map(node => node.name).sort();\n    const sortedOutputs = outputs.map(node => node.name).sort();\n    return sortedInputs.join(this.SEPERATOR) + '--' +\n        sortedOutputs.join(this.SEPERATOR);\n  }\n\n  /**\n   * Compiles the inference graph and returns the minimal set of nodes that are\n   * required for execution, in the correct execution order.\n   */\n  private compile(inputs: NamedTensorMap, outputs: Node[]): Node[] {\n    const executionInfo =\n        getExecutionSubgraph(inputs, outputs, this.weightMap, this._initNodes);\n    const {missingInputs, dynamicNode, syncInputs} = executionInfo;\n    if (dynamicNode != null) {\n      throw new Error(\n          `This execution contains the node '${dynamicNode.name}', which has ` +\n          `the dynamic op '${dynamicNode.op}'. Please use ` +\n          `model.executeAsync() instead. Alternatively, to avoid the ` +\n          `dynamic ops, specify the inputs [${syncInputs}]`);\n    }\n\n    if (missingInputs.length > 0) {\n      const outNames = outputs.map(n => n.name);\n      const inNames = Object.keys(inputs);\n      throw new Error(\n          `Cannot compute the outputs [${outNames}] from the provided inputs ` +\n          `[${inNames}]. Missing the following inputs: [${missingInputs}]`);\n    }\n\n    return getNodesInTopologicalOrder(\n        this.graph, this.weightMap, executionInfo);\n  }\n\n  /**\n   * Executes the inference for given input tensors.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  execute(inputs: NamedTensorMap, outputs: string[]): Tensor[] {\n    inputs = this.mapInputs(inputs);\n    const names = Object.keys(inputs).sort();\n    this.checkInputs(inputs);\n    this.checkInputShapeAndType(inputs);\n    outputs = this.mapOutputs(outputs);\n    this.checkOutputs(outputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputs.map(name => parseNodeName(name)[0]);\n    let outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n\n    // If no outputs are specified, then use the default outputs of the model.\n    if (outputNodes.length === 0) {\n      outputNodes = this._outputs;\n    }\n\n    const compilationKey = this.getCompilationKey(inputNodes, outputNodes);\n\n    // Do nothing if the compiled graph cache contains the input.\n    let orderedNodes = this.compiledMap.get(compilationKey);\n    if (orderedNodes == null) {\n      orderedNodes = this.compile(inputs, outputNodes);\n      this.compiledMap.set(compilationKey, orderedNodes);\n    }\n\n    const tensorArrayMap: TensorArrayMap = {};\n    const tensorListMap: TensorListMap = {};\n    return tidy(() => {\n      const context = new ExecutionContext(\n          this.weightMap, tensorArrayMap, tensorListMap,\n          this.functionExecutorMap);\n      const tensorsMap: NamedTensorsMap = {...this.weightMap};\n\n      Object.keys(inputs).forEach(name => {\n        const [nodeName, index] = parseNodeName(name);\n        const tensors: Tensor[] = [];\n        tensors[index] = inputs[name];\n        tensorsMap[nodeName] = tensors;\n      });\n\n      const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n      const intermediateTensorConsumerCount: {[key: number]: number} = {};\n      for (let i = 0; i < orderedNodes.length; i++) {\n        const node = orderedNodes[i];\n        if (!tensorsMap[node.name]) {\n          const tensors = executeOp(node, tensorsMap, context) as Tensor[];\n          if (tensors instanceof Promise) {\n            throw new Error(\n                `The execution of the op '${node.op}' returned a promise. ` +\n                `Please use model.executeAsync() instead.`);\n          }\n          tensorsMap[node.name] = tensors;\n          this.checkTensorForDisposal(\n              node.name, node, tensorsMap, context, tensorsToKeep,\n              outputNodeNames, intermediateTensorConsumerCount);\n        }\n      }\n      // dispose the context for the root executor\n      if (this.parent == null) {\n        context.dispose(tensorsToKeep);\n      }\n      return outputs.map(name => getTensor(name, tensorsMap, context));\n    });\n  }\n\n  private getFrozenTensorIds(tensorMap: NamedTensorsMap): Set<number> {\n    const ids = [].concat.apply(\n        [],\n        Object.keys(tensorMap)\n            .map(key => tensorMap[key])\n            .map(tensors => tensors.map(tensor => tensor.id)));\n    return new Set(ids);\n  }\n  private checkTensorForDisposal(\n      nodeName: string, node: Node, tensorMap: NamedTensorsMap,\n      context: ExecutionContext, tensorsToKeep: Set<number>,\n      outputNames: string[],\n      intermediateTensorConsumerCount: {[key: string]: number}) {\n    // Skip output nodes and any control flow nodes, since its dependency is\n    // tricky to track correctly.\n    if (node.category === 'control' || outputNames.indexOf(nodeName) !== -1) {\n      return;\n    }\n\n    tensorMap[nodeName].forEach(tensor => {\n      if (tensor != null) {\n        intermediateTensorConsumerCount[tensor.id] =\n            (intermediateTensorConsumerCount[tensor.id] || 0) +\n            node.children.length;\n      }\n    });\n    node.inputs.forEach(input => {\n      // Skip any control flow nodes, since its dependency is tricky to track\n      // correctly.\n      if (input.category !== 'control') {\n        const tensors =\n            getTensorsForCurrentContenxt(input.name, tensorMap, context);\n        if (tensors != null) {\n          tensors.forEach(tensor => {\n            if (tensor && !tensorsToKeep.has(tensor.id)) {\n              const count = intermediateTensorConsumerCount[tensor.id];\n              if (count === 1) {\n                tensor.dispose();\n                delete intermediateTensorConsumerCount[tensor.id];\n              } else if (count != null) {\n                // only intermediate nodes has count set, inputs and weights are\n                // not.\n                intermediateTensorConsumerCount[tensor.id]--;\n              }\n            }\n          });\n        }\n      }\n    });\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  async executeAsync(inputs: NamedTensorMap, outputs: string[]):\n      Promise<Tensor[]> {\n    return this._executeAsync(inputs, outputs);\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   * @param isFunctionExecution Flag for executing a function.\n   * @param tensorArrayMap Optional, global TensorArray map by id. Used for\n   * function execution.\n   * @param tensorArrayMap Optinal global TensorList map by id. Used for\n   * function execution.\n   */\n  private async _executeAsync(\n      inputs: NamedTensorMap, outputs: string[], isFunctionExecution = false,\n      tensorArrayMap: TensorArrayMap = {},\n      tensorListMap: TensorListMap = {}): Promise<Tensor[]> {\n    if (!isFunctionExecution) {\n      inputs = this.mapInputs(inputs);\n      this.checkInputs(inputs);\n      this.checkInputShapeAndType(inputs);\n      outputs = this.mapOutputs(outputs);\n      this.checkOutputs(outputs);\n    }\n\n    const context = new ExecutionContext(\n        this.weightMap, tensorArrayMap, tensorListMap,\n        this.functionExecutorMap);\n\n    // Graph with control flow op requires runtime evaluation of the execution\n    // order, while without control flow the execution order is pre-determined\n    // in the compile method.\n    const tensorMap = await this.executeWithControlFlow(\n        inputs, context, outputs, isFunctionExecution);\n    const results = outputs.map(name => getTensor(name, tensorMap, context));\n\n    // dispose all the intermediate tensors\n    const outputIds = results.map(t => t.id);\n    const inputIds = Object.keys(inputs).map(name => inputs[name].id);\n    const keepIds =\n        new Set<number>([...outputIds, ...inputIds, ...this.weightIds]);\n    Object.keys(tensorMap).forEach(key => {\n      const tensorArray = tensorMap[key];\n      tensorArray.forEach(tensor => {\n        if (tensor && !tensor.isDisposed && !keepIds.has(tensor.id)) {\n          tensor.dispose();\n        }\n      });\n    });\n    // dispose the context for the root executor\n    if (this.parent == null) {\n      context.dispose(keepIds);\n    }\n\n    return results;\n  }\n\n  async executeFunctionAsync(\n      inputs: Tensor[], tensorArrayMap: TensorArrayMap,\n      tensorListMap: TensorListMap): Promise<Tensor[]> {\n    const mappedInputs = inputs.reduce((map, tensor, index) => {\n      map[this.inputs[index].name] = tensor;\n      return map;\n    }, {} as NamedTensorMap);\n\n    return this._executeAsync(\n        mappedInputs, this.outputNodes, true, tensorArrayMap, tensorListMap);\n  }\n  /**\n   * When there are control flow nodes in the graph, the graph execution use\n   * ExecutionContext to keep track of the frames and loop iterators.\n   * @param inputs placeholder tensors for the graph.\n   * @param context the execution context object for current execution.\n   * @param isFunctionExecution Flag for executing a function.\n   */\n  private async executeWithControlFlow(\n      inputs: NamedTensorMap, context: ExecutionContext, outputNames: string[],\n      isFunctionExecution: boolean): Promise<NamedTensorsMap> {\n    const names = Object.keys(inputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputNames.map(name => parseNodeName(name)[0]);\n    const outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n    const {usedNodes, missingInputs, dynamicNode, syncInputs} =\n        getExecutionSubgraph(inputs, outputNodes, this.weightMap);\n\n    const stack: NodeWithContexts[] =\n        [...inputNodes, ...this.graph.weights].map(node => {\n          return {node, contexts: context.currentContext};\n        });\n    const tensorsMap: NamedTensorsMap = {...this.weightMap};\n    Object.keys(inputs).forEach(name => {\n      const [nodeName, index] = parseNodeName(name);\n      const tensors: Tensor[] = [];\n      tensors[index] = inputs[name];\n      tensorsMap[nodeName] = tensors;\n    });\n    const intermediateTensorConsumerCount: {[key: number]: number} = {};\n    const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n    const added: {[key: string]: boolean} = {};\n    while (stack.length > 0) {\n      const promises = this.processStack(\n          inputNodes, stack, context, tensorsMap, added, tensorsToKeep,\n          outputNodeNames, intermediateTensorConsumerCount, usedNodes);\n      await Promise.all(promises);\n    }\n    if (dynamicNode == null && !isFunctionExecution) {\n      console.warn(\n          `This model execution did not contain any nodes with control flow ` +\n          `or dynamic output shapes. You can use model.execute() instead.`);\n    }\n    const missingOutputs =\n        outputNodes\n            .filter(\n                node => !isControlFlow(node) &&\n                    !getTensor(node.name, tensorsMap, context))\n            .map(node => node.name);\n    if (missingOutputs.length > 0) {\n      let alternativeMsg = '';\n      if (dynamicNode != null) {\n        alternativeMsg =\n            `Alternatively, to avoid the dynamic ops, use model.execute() ` +\n            `and specify the inputs [${syncInputs}]`;\n      }\n      throw new Error(\n          `Cannot compute the outputs [${missingOutputs}] from the provided ` +\n          `inputs [${names}]. Consider providing the following inputs: ` +\n          `[${missingInputs}]. ${alternativeMsg}`);\n    }\n    return tensorsMap;\n  }\n\n  private processStack(\n      inputNodes: Node[], stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      tensorsToKeep: Set<number>, outputNames: string[],\n      intermediateTensorConsumerCount: {[key: number]: number},\n      usedNodes: Set<string>) {\n    const promises: Array<Promise<Tensor[]>> = [];\n    while (stack.length > 0) {\n      const item = stack.pop();\n      context.currentContext = item.contexts;\n      let nodeName = '';\n      // The tensor of the Enter op with isConstant set should be set\n      // in the parent scope, so it will be available as constant for the\n      // whole loop.\n      if (item.node.op === 'Enter' &&\n          getParamValue('isConstant', item.node, tensorMap, context)) {\n        [nodeName] = getNodeNameAndIndex(item.node.name, context);\n      }\n\n      // only process nodes that are not provided as input nodes.\n      if (inputNodes.indexOf(item.node) === -1) {\n        const tensors = executeOp(item.node, tensorMap, context);\n        if (!nodeName) {\n          [nodeName] = getNodeNameAndIndex(item.node.name, context);\n        }\n        const currentContext = context.currentContext;\n        if (tensors instanceof Promise) {\n          promises.push(tensors.then(t => {\n            tensorMap[nodeName] = t;\n            context.currentContext = currentContext;\n            this.checkTensorForDisposal(\n                nodeName, item.node, tensorMap, context, tensorsToKeep,\n                outputNames, intermediateTensorConsumerCount);\n            this.processChildNodes(\n                item.node, stack, context, tensorMap, added, usedNodes);\n            return t;\n          }));\n        } else {\n          tensorMap[nodeName] = tensors;\n          this.checkTensorForDisposal(\n              nodeName, item.node, tensorMap, context, tensorsToKeep,\n              outputNames, intermediateTensorConsumerCount);\n          this.processChildNodes(\n              item.node, stack, context, tensorMap, added, usedNodes);\n        }\n      } else {\n        this.processChildNodes(\n            item.node, stack, context, tensorMap, added, usedNodes);\n      }\n    }\n    return promises;\n  }\n\n  private processChildNodes(\n      node: Node, stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      usedNodes: Set<string>) {\n    node.children.forEach((childNode) => {\n      const [nodeName, ] = getNodeNameAndIndex(childNode.name, context);\n      if (added[nodeName] || !usedNodes.has(childNode.name)) {\n        return;\n      }\n      // Merge op can be pushed if any of its inputs has value.\n      if (childNode.op === 'Merge') {\n        if (childNode.inputNames.some(name => {\n              return !!getTensor(name, tensorMap, context);\n            })) {\n          added[nodeName] = true;\n          stack.push({contexts: context.currentContext, node: childNode});\n        }\n      } else  // Otherwise all inputs must to have value.\n          if (childNode.inputNames.every(name => {\n                return !!getTensor(name, tensorMap, context);\n              })) {\n        added[nodeName] = true;\n        stack.push({contexts: context.currentContext, node: childNode});\n      }\n    });\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   */\n  dispose() {\n    Object.keys(this.weightMap)\n        .forEach(\n            key => this.weightMap[key].forEach(tensor => tensor.dispose()));\n  }\n\n  private checkInputShapeAndType(inputs: NamedTensorMap) {\n    Object.keys(inputs).forEach(name => {\n      const input = inputs[name];\n      const [nodeName, ] = parseNodeName(name);\n      const node = this.graph.nodes[nodeName];\n      if (node.attrParams['shape'] && node.attrParams['shape'].value) {\n        const shape = node.attrParams['shape'].value as number[];\n        const match = shape.length === input.shape.length &&\n            input.shape.every(\n                (dim, index) => shape[index] === -1 || shape[index] === dim);\n        util.assert(\n            match,\n            () => `The shape of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be [${shape}], but was ` +\n                `[${input.shape}]`);\n      }\n      if (node.attrParams['dtype'] && node.attrParams['dtype'].value) {\n        util.assert(\n            input.dtype === node.attrParams['dtype'].value as string,\n            () => `The dtype of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be ` +\n                `${node.attrParams['dtype'].value}, but was ${input.dtype}`);\n      }\n    });\n  }\n\n  private mapInputs(inputs: NamedTensorMap) {\n    const result: NamedTensorMap = {};\n    for (const inputName in inputs) {\n      if (this._signature != null && this._signature.inputs != null &&\n          this._signature.inputs[inputName] != null) {\n        const tensor = this._signature.inputs[inputName];\n        result[tensor.name] = inputs[inputName];\n      } else {\n        result[inputName] = inputs[inputName];\n      }\n    }\n    return result;\n  }\n\n  private checkInputs(inputs: NamedTensorMap) {\n    const notInGraph = Object.keys(inputs).filter(name => {\n      const [nodeName] = parseNodeName(name);\n      return this.graph.nodes[nodeName] == null;\n    });\n    if (notInGraph.length > 0) {\n      throw new Error(\n          `The dict provided in model.execute(dict) has ` +\n          `keys: [${notInGraph}] that are not part of graph`);\n    }\n  }\n\n  private mapOutputs(outputs: string[]) {\n    return outputs.map(name => {\n      if (this._signature != null && this._signature.outputs != null &&\n          this._signature.outputs[name] != null) {\n        const tensor = this._signature.outputs[name];\n        return tensor.name;\n      }\n      return name;\n    }, {});\n  }\n\n  private checkOutputs(outputs: string[]): void {\n    outputs.forEach(name => {\n      const [normalizedName] = parseNodeName(name);\n      if (!this.graph.nodes[normalizedName]) {\n        throw new Error(`The output '${name}' is not found in the graph`);\n      }\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {InferenceModel, io, ModelPredictConfig, NamedTensorMap, Tensor} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\nimport {NamedTensorsMap, TensorInfo} from '../data/types';\nimport {OperationMapper} from '../operations/operation_mapper';\n\nimport {GraphExecutor} from './graph_executor';\n\nexport const TFHUB_SEARCH_PARAM = '?tfjs-format=file';\nexport const DEFAULT_MODEL_NAME = 'model.json';\n/**\n * A `tf.GraphModel` is a directed, acyclic graph built from a\n * SavedModel GraphDef and allows inference execution.\n *\n * A `tf.GraphModel` can only be created by loading from a model converted from\n * a [TensorFlow SavedModel](https://www.tensorflow.org/guide/saved_model) using\n * the command line converter tool and loaded via `tf.loadGraphModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class GraphModel implements InferenceModel {\n  private executor: GraphExecutor;\n  private version = 'n/a';\n  private handler: io.IOHandler;\n  private artifacts: io.ModelArtifacts;\n  private initializer: GraphExecutor;\n\n  // Returns the version information for the tensorflow model GraphDef.\n  get modelVersion(): string {\n    return this.version;\n  }\n\n  get inputNodes(): string[] {\n    return this.executor.inputNodes;\n  }\n\n  get outputNodes(): string[] {\n    return this.executor.outputNodes;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this.executor.inputs;\n  }\n\n  get outputs(): TensorInfo[] {\n    return this.executor.outputs;\n  }\n\n  get weights(): NamedTensorsMap {\n    return this.executor.weightMap;\n  }\n\n  /**\n   * @param modelUrl url for the model, or an `io.IOHandler`.\n   * @param weightManifestUrl url for the weight file generated by\n   * scripts/convert.py script.\n   * @param requestOption options for Request, which allows to send credentials\n   * and custom headers.\n   * @param onProgress Optional, progress callback function, fired periodically\n   * before the load is completed.\n   */\n  constructor(\n      private modelUrl: string|io.IOHandler,\n      private loadOptions: io.LoadOptions = {}) {\n    if (loadOptions == null) {\n      this.loadOptions = {};\n    }\n  }\n\n  private findIOHandler() {\n    const path = this.modelUrl;\n    if ((path as io.IOHandler).load != null) {\n      // Path is an IO Handler.\n      this.handler = path as io.IOHandler;\n    } else if (this.loadOptions.requestInit != null) {\n      this.handler = io.browserHTTPRequest(path as string, this.loadOptions);\n    } else {\n      const handlers = io.getLoadHandlers(path as string, this.loadOptions);\n      if (handlers.length === 0) {\n        // For backward compatibility: if no load handler can be found,\n        // assume it is a relative http path.\n        handlers.push(io.browserHTTPRequest(path as string, this.loadOptions));\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) load handlers for ` +\n            `URL '${[path]}'`);\n      }\n      this.handler = handlers[0];\n    }\n  }\n\n  /**\n   * Loads the model and weight files, construct the in memory weight map and\n   * compile the inference graph.\n   */\n  async load(): Promise<boolean> {\n    this.findIOHandler();\n    if (this.handler.load == null) {\n      throw new Error(\n          'Cannot proceed with model loading because the IOHandler provided ' +\n          'does not have the `load` method implemented.');\n    }\n    const artifacts = await this.handler.load();\n\n    return this.loadSync(artifacts);\n  }\n\n  /**\n   * Synchronously construct the in memory weight map and\n   * compile the inference graph. Also initialize hashtable if any.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  loadSync(artifacts: io.ModelArtifacts) {\n    this.artifacts = artifacts;\n    const graph = this.artifacts.modelTopology as tensorflow.IGraphDef;\n    let signature = {};\n    if (this.artifacts.userDefinedMetadata != null) {\n      signature =  // tslint:disable-next-line:no-any\n          (this.artifacts.userDefinedMetadata as any).signature as\n          tensorflow.ISignatureDef;\n    }\n\n    this.version = `${graph.versions.producer}.${graph.versions.minConsumer}`;\n    const weightMap =\n        io.decodeWeights(this.artifacts.weightData, this.artifacts.weightSpecs);\n    this.executor = new GraphExecutor(\n        OperationMapper.Instance.transformGraph(graph, signature));\n    this.executor.weightMap = this.convertTensorMapToTensorsMap(weightMap);\n\n    if (artifacts.modelInitializer != null) {\n      const initializer =\n          OperationMapper.Instance.transformGraph(artifacts.modelInitializer);\n      this.initializer = new GraphExecutor(initializer);\n      this.initializer.weightMap = this.executor.weightMap;\n      this.initializer.execute({}, []);\n    }\n\n    return true;\n  }\n\n  /**\n   * Save the configuration and/or weights of the GraphModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const modelUrl =\n   *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n   * const model = await tf.loadGraphModel(modelUrl);\n   * const zeros = tf.zeros([1, 224, 224, 3]);\n   * model.predict(zeros).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * model.predict(zeros).print();\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new Error(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new Error(\n          'GraphModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    return handlerOrURL.save(this.artifacts);\n  }\n\n  /**\n   * Execute the inference for the input tensors.\n   *\n   * @param input The input tensors, when there is single input for the model,\n   * inputs param should be a `tf.Tensor`. For models with mutliple inputs,\n   * inputs params should be in either `tf.Tensor`[] if the input order is\n   * fixed, or otherwise NamedTensorMap format.\n   *\n   * For model with multiple inputs, we recommend you use NamedTensorMap as the\n   * input type, if you use `tf.Tensor`[], the order of the array needs to\n   * follow the\n   * order of inputNodes array. @see {@link GraphModel.inputNodes}\n   *\n   * You can also feed any intermediate nodes using the NamedTensorMap as the\n   * input type. For example, given the graph\n   *    InputNode => Intermediate => OutputNode,\n   * you can execute the subgraph Intermediate => OutputNode by calling\n   *    model.execute('IntermediateNode' : tf.tensor(...));\n   *\n   * This is useful for models that uses tf.dynamic_rnn, where the intermediate\n   * state needs to be fed manually.\n   *\n   * For batch inference execution, the tensors for each input need to be\n   * concatenated together. For example with mobilenet, the required input shape\n   * is [1, 244, 244, 3], which represents the [batch, height, width, channel].\n   * If we are provide a batched data of 100 images, the input tensor should be\n   * in the shape of [100, 244, 244, 3].\n   *\n   * @param config Prediction configuration for specifying the batch size and\n   * output node names. Currently the batch size option is ignored for graph\n   * model.\n   *\n   * @returns Inference result tensors. The output would be single `tf.Tensor`\n   * if model has single output node, otherwise Tensor[] or NamedTensorMap[]\n   * will be returned for model with multiple outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(inputs: Tensor|Tensor[]|NamedTensorMap, config?: ModelPredictConfig):\n      Tensor|Tensor[]|NamedTensorMap {\n    return this.execute(inputs, this.outputNodes);\n  }\n\n  private normalizeInputs(inputs: Tensor|Tensor[]|\n                          NamedTensorMap): NamedTensorMap {\n    if (!(inputs instanceof Tensor) && !Array.isArray(inputs)) {\n      // The input is already a NamedTensorMap.\n      return inputs;\n    }\n    inputs = Array.isArray(inputs) ? inputs : [inputs];\n    if (inputs.length !== this.inputNodes.length) {\n      throw new Error(\n          'Input tensor count mismatch,' +\n          `the graph model has ${this.inputNodes.length} placeholders, ` +\n          `while there are ${inputs.length} input tensors.`);\n    }\n    return this.inputNodes.reduce((map, inputName, i) => {\n      map[inputName] = (inputs as Tensor[])[i];\n      return map;\n    }, {} as NamedTensorMap);\n  }\n\n  private normalizeOutputs(outputs: string|string[]): string[] {\n    outputs = outputs || this.outputNodes;\n    return !Array.isArray(outputs) ? [outputs] : outputs;\n  }\n\n  /**\n   * Executes inference for the model for given input tensors.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no\n   * outputs are specified, the default outputs of the model would be used.\n   * You can inspect intermediate nodes of the model by adding them to the\n   * outputs array.\n   *\n   * @returns A single tensor if provided with a single output or no outputs\n   * are provided and there is only one default output, otherwise return a\n   * tensor array. The order of the tensor array is the same as the outputs\n   * if provided, otherwise the order of outputNodes attribute of the model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs?: string|string[]):\n      Tensor|Tensor[] {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = this.executor.execute(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n  /**\n   * Executes inference for the model for given input tensors in async\n   * fashion, use this method when your model contains control flow ops.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   *\n   * @returns A Promise of single tensor if provided with a single output or\n   * no outputs are provided and there is only one default output, otherwise\n   * return a tensor map.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async executeAsync(\n      inputs: Tensor|Tensor[]|NamedTensorMap,\n      outputs?: string|string[]): Promise<Tensor|Tensor[]> {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = await this.executor.executeAsync(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n\n  private convertTensorMapToTensorsMap(map: NamedTensorMap): NamedTensorsMap {\n    return Object.keys(map).reduce((newMap: NamedTensorsMap, key) => {\n      newMap[key] = [map[key]];\n      return newMap;\n    }, {});\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  dispose() {\n    this.executor.dispose();\n\n    if (this.initializer) {\n      this.initializer.dispose();\n    }\n  }\n}\n\n/**\n * Load a graph model given a URL to the model definition.\n *\n * Example of loading MobileNetV2 from a URL and making a prediction with a\n * zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n * const model = await tf.loadGraphModel(modelUrl);\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n *\n * Example of loading MobileNetV2 from a TF Hub URL and making a prediction with\n * a zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/2';\n * const model = await tf.loadGraphModel(modelUrl, {fromTFHub: true});\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n * @param modelUrl The url or an `io.IOHandler` that loads the model.\n * @param options Options for the HTTP request, which allows to send credentials\n *    and custom headers.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport async function loadGraphModel(\n    modelUrl: string|io.IOHandler,\n    options: io.LoadOptions = {}): Promise<GraphModel> {\n  if (modelUrl == null) {\n    throw new Error(\n        'modelUrl in loadGraphModel() cannot be null. Please provide a url ' +\n        'or an IOHandler that loads the model');\n  }\n  if (options == null) {\n    options = {};\n  }\n\n  if (options.fromTFHub) {\n    if ((modelUrl as io.IOHandler).load == null) {\n      if (!(modelUrl as string).endsWith('/')) {\n        modelUrl = (modelUrl as string) + '/';\n      }\n      modelUrl = `${modelUrl}${DEFAULT_MODEL_NAME}${TFHUB_SEARCH_PARAM}`;\n    }\n  }\n  const model = new GraphModel(modelUrl, options);\n  await model.load();\n  return model;\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\n\n// tslint:disable:no-any\n\n/**\n * A return value for a mapping function that can be applied via deepMap.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapResult = {\n  value: any,\n  recurse: boolean\n};\n\n/**\n * Apply a mapping function to a nested structure in a recursive manner.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapResult`.  The `DeepMapResult` either provides a\n *   replacement value for that node (i.e., replacing the subtree), or indicates\n *   that the node should be processed recursively.\n */\nexport function deepMap(input: any, mapFn: (x: any) => DeepMapResult): any|\n    any[] {\n  return deepMapInternal(input, mapFn);\n}\n\n/**\n * @param seen: A Map of known object mappings (i.e., memoized results of\n *   `mapFn()`)\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepMapInternal(\n    input: any, mapFn: (x: any) => DeepMapResult,\n    seen: Map<any, any> = new Map(), containedIn: Set<{}> = new Set()): any|\n    any[] {\n  if (input == null) {\n    return null;\n  }\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  if (seen.has(input)) {\n    return seen.get(input);\n  }\n  const result = mapFn(input);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep map function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    seen.set(input, result.value);\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const child = input[k];\n      const childResult = deepMapInternal(child, mapFn, seen, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// TODO(soergel, kangyizhang) Reconsider naming of deepZip() to avoid confusion\n// with zip()\n\n/**\n * Zip nested structures together in a recursive manner.\n *\n * This has the effect of transposing or pivoting data, e.g. converting it from\n * a row-major representation to a column-major representation.\n *\n * For example, `deepZip([{a: 1, b: 2}, {a: 3, b: 4}])` returns\n * `{a: [1, 3], b: [2, 4]}`.\n *\n * The inputs should all have the same nested structure (i.e., of arrays and\n * dicts).  The result is a single object with the same nested structure, where\n * the leaves are arrays collecting the values of the inputs at that location\n * (or, optionally, the result of a custom function applied to those arrays).\n *\n * @param inputs: An array of the objects to zip together.\n * @param zipFn: (optional) A function that expects an array of elements at a\n *   single node of the object tree, and returns a `DeepMapResult`.  The\n *   `DeepMapResult` either provides a result value for that node (i.e.,\n *   representing the subtree), or indicates that the node should be processed\n *   recursively.  The default zipFn recurses as far as possible and places\n *   arrays at the leaves.\n */\nexport function deepZip(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult = zipToList): any|any[] {\n  return deepZipInternal(inputs, zipFn);\n}\n\n/**\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepZipInternal(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult,\n    containedIn: Set<{}> = new Set()): any|any[] {\n  // The recursion follows the structure of input 0; it's assumed that all the\n  // other inputs have the same structure.\n  const input = inputs[0];\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  const result = zipFn(inputs);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep zip function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const children = inputs.map(x => x[k]);\n      const childResult = deepZipInternal(children, zipFn, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// tslint:disable-next-line:no-any\nexport function zipToList(x: any[]): DeepMapResult {\n  if (x === null) {\n    return null;\n  }\n  // TODO(soergel): validate array type?\n\n  if (isIterable(x[0])) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: x, recurse: false};\n  }\n}\n\n/**\n * A return value for an async map function for use with deepMapAndAwaitAll.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapAsyncResult = {\n  value: Promise<any>,\n  recurse: boolean\n};\n\n/**\n * Apply an async mapping function to a nested structure in a recursive manner.\n *\n * This first creates a nested structure of Promises, and then awaits all of\n * those, resulting in a single Promise for a resolved nested structure.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapAsyncResult`.  The `DeepMapAsyncResult` either provides\n *   a `Promise` for a replacement value for that node (i.e., replacing the\n *   subtree), or indicates that the node should be processed recursively.  Note\n *   that the decision whether or not to recurse must be made immediately; only\n *   the mapped value may be promised.\n */\nexport async function deepMapAndAwaitAll(\n    input: any, mapFn: (x: any) => DeepMapAsyncResult): Promise<any|any[]> {\n  const seen: Map<any, Promise<any>> = new Map();\n\n  // First do a normal deepMap, collecting Promises in 'seen' as a side effect.\n  deepMapInternal(input, mapFn, seen);\n\n  // Replace the Promises in 'seen' in place.\n  // Note TypeScript provides no async map iteration, and regular map iteration\n  // is broken too, so sadly we have to do Array.from() to make it work.\n  // (There's no advantage to Promise.all(), and that would be tricky anyway.)\n  for (const key of Array.from(seen.keys())) {\n    const value = seen.get(key);\n    if (value instanceof Promise) {\n      const mappedValue = await value;\n      seen.set(key, mappedValue);\n    }\n  }\n\n  // Normal deepMap again, this time filling in the resolved values.\n  // It's unfortunate that we have to do two passes.\n  // TODO(soergel): test performance and think harder about a fast solution.\n  const result = deepMapInternal(input, mapFn, seen);\n  return result;\n}\n\n/**\n * Determine whether the argument is iterable.\n *\n * @returns true if the argument is an array or any non-Tensor object.\n */\n// tslint:disable-next-line:no-any\nexport function isIterable(obj: any): boolean {\n  return obj != null && (!ArrayBuffer.isView(obj)) &&\n      (Array.isArray(obj) ||\n       (typeof obj === 'object' && !(obj instanceof tf.Tensor)));\n}\n\n/**\n * Determine whether the argument can be converted to Tensor.\n *\n * Tensors, primitives, arrays, and TypedArrays all qualify; anything else does\n * not.\n *\n * @returns true if the argument can be converted to Tensor.\n */\n// tslint:disable-next-line:no-any\nexport function canTensorify(obj: any): boolean {\n  return obj == null || isPrimitive(obj) || Array.isArray(obj) ||\n      (typeof obj === 'object' && (obj instanceof tf.Tensor)) ||\n      tf.util.isTypedArray(obj);\n}\n\n/**\n * Returns true if the given `value` is a primitive type. Otherwise returns\n * false. This is equivalant to node util.isPrimitive\n */\nfunction isPrimitive(value: any): boolean {\n  return (\n      value === null ||\n      (typeof value !== 'object' && typeof value !== 'function'));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {deepMap, DeepMapResult, isIterable} from './deep_map';\n\nexport function deepClone<T>(container: T): T {\n  return deepMap(container, cloneIfTensor);\n}\n\n// tslint:disable-next-line: no-any\nfunction cloneIfTensor(item: any): DeepMapResult {\n  if (item instanceof tf.Tensor) {\n    return ({value: item.clone(), recurse: false});\n  } else if (isIterable(item)) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: item, recurse: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/**\n * A ring buffer, providing O(1) FIFO, LIFO, and related operations.\n */\nexport class RingBuffer<T> {\n  // Note we store the indices in the range 0 <= index < 2*capacity.\n  // This allows us to distinguish the full from the empty case.\n  // See https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/\n  protected begin = 0;  // inclusive\n  protected end = 0;    // exclusive\n  protected doubledCapacity: number;\n\n  protected data: T[];\n\n  /**\n   * Constructs a `RingBuffer`.\n   * @param capacity The number of items that the buffer can accomodate.\n   */\n  constructor(public capacity: number) {\n    if (capacity == null) {\n      throw new RangeError('Can\\'t create a ring buffer of unknown capacity.');\n    }\n    if (capacity < 1) {\n      throw new RangeError('Can\\'t create ring buffer of capacity < 1.');\n    }\n    this.data = new Array<T>(capacity);\n    this.doubledCapacity = 2 * capacity;\n  }\n\n  /**\n   * Map any index into the range 0 <= index < 2*capacity.\n   */\n  protected wrap(index: number) {\n    // don't trust % on negative numbers\n    while (index < 0) {\n      index += this.doubledCapacity;\n    }\n    return index % this.doubledCapacity;\n  }\n\n  protected get(index: number) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t get item at a negative index.');\n    }\n    return this.data[index % this.capacity];\n  }\n\n  protected set(index: number, value: T) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t set item at a negative index.');\n    }\n    this.data[index % this.capacity] = value;\n  }\n\n  /**\n   * Returns the current number of items in the buffer.\n   */\n  length(): number {\n    let length = this.end - this.begin;\n    if (length < 0) {\n      length = this.doubledCapacity + length;\n    }\n    return length;\n  }\n\n  /**\n   * Reports whether the buffer is full.\n   * @returns true if the number of items in the buffer equals its capacity, and\n   *   false otherwise.\n   */\n  isFull() {\n    return this.length() === this.capacity;\n  }\n\n  /**\n   * Reports whether the buffer is empty.\n   * @returns true if the number of items in the buffer equals zero, and\n   *   false otherwise.\n   */\n  isEmpty() {\n    return this.length() === 0;\n  }\n\n  /**\n   * Adds an item to the end of the buffer.\n   */\n  push(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.set(this.end, value);\n    this.end = this.wrap(this.end + 1);\n  }\n\n  /**\n   * Adds many items to the end of the buffer, in order.\n   */\n  pushAll(values: T[]) {\n    for (const value of values) {\n      this.push(value);\n    }\n  }\n\n  /**\n   * Removes and returns the last item in the buffer.\n   */\n  pop(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    this.end = this.wrap(this.end - 1);\n    const result = this.get(this.end);\n    this.set(this.end, undefined);\n    return result;\n  }\n\n  /**\n   * Adds an item to the beginning of the buffer.\n   */\n  unshift(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.begin = this.wrap(this.begin - 1);\n    this.set(this.begin, value);\n  }\n\n  /**\n   * Removes and returns the first item in the buffer.\n   */\n  shift(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const result = this.get(this.begin);\n    this.set(this.begin, undefined);\n    this.begin = this.wrap(this.begin + 1);\n    return result;\n  }\n\n  /**\n   * Removes and returns a specific item in the buffer, and moves the last item\n   * to the vacated slot.  This is useful for implementing a shuffling stream.\n   * Note that this operation necessarily scrambles the original order.\n   *\n   * @param relativeIndex: the index of the item to remove, relative to the\n   *   first item in the buffer (e.g., hiding the ring nature of the underlying\n   *   storage).\n   */\n  shuffleExcise(relativeIndex: number): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const index = this.wrap(this.begin + relativeIndex);\n    const result = this.get(index);\n    this.set(index, this.pop());\n    return result;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {RingBuffer} from './ring_buffer';\n\nexport class GrowingRingBuffer<T> extends RingBuffer<T> {\n  private static INITIAL_CAPACITY = 32;\n\n  /**\n   * Constructs a `GrowingRingBuffer`.\n   */\n  constructor() {\n    super(GrowingRingBuffer.INITIAL_CAPACITY);\n  }\n\n  isFull() {\n    return false;\n  }\n\n  push(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.push(value);\n  }\n\n  unshift(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.unshift(value);\n  }\n\n  /**\n   * Doubles the capacity of the buffer.\n   */\n  private expand() {\n    const newCapacity = this.capacity * 2;\n    const newData = new Array<T>(newCapacity);\n    const len = this.length();\n\n    // Rotate the buffer to start at index 0 again, since we can't just\n    // allocate more space at the end.\n    for (let i = 0; i < len; i++) {\n      newData[i] = this.get(this.wrap(this.begin + i));\n    }\n\n    this.data = newData;\n    this.capacity = newCapacity;\n    this.doubledCapacity = 2 * this.capacity;\n    this.begin = 0;\n    this.end = len;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {Container} from '../types';\nimport {deepClone} from '../util/deep_clone';\nimport {deepMapAndAwaitAll, DeepMapAsyncResult, DeepMapResult, deepZip, zipToList} from '../util/deep_map';\nimport {GrowingRingBuffer} from '../util/growing_ring_buffer';\nimport {RingBuffer} from '../util/ring_buffer';\n\n/**\n * A nested structure of LazyIterators, used as the input to zip().\n */\nexport type IteratorContainer = Container<LazyIterator<tf.TensorContainer>>;\n\n// Here we implement a simple asynchronous iterator.\n// This lets us avoid using either third-party stream libraries or\n// recent TypeScript language support requiring polyfills.\n\n/**\n * Create a `LazyIterator` from an array of items.\n */\nexport function iteratorFromItems<T>(items: T[]): LazyIterator<T> {\n  return new ArrayIterator(items);\n}\n\n/**\n * Create a `LazyIterator` of incrementing integers.\n */\nexport function iteratorFromIncrementing(start: number): LazyIterator<number> {\n  let i = start;\n  return iteratorFromFunction(() => ({value: i++, done: false}));\n}\n\n/**\n * Create a `LazyIterator` from a function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * await iter.forEachAsync(e => console.log(e));\n * ```\n *\n * @param func A function that produces data on each call.\n */\nexport function iteratorFromFunction<T>(\n    func: () =>\n        IteratorResult<T>| Promise<IteratorResult<T>>): LazyIterator<T> {\n  return new FunctionCallIterator(func);\n}\n\n/**\n * Create a `LazyIterator` by concatenating underlying streams, which are\n * themselves provided as a stream.\n *\n * This can also be thought of as a \"stream flatten\" operation.\n *\n * @param baseIterators A stream of streams to be concatenated.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenated<T>(\n    baseIterators: LazyIterator<LazyIterator<T>>,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return new ChainedIterator(baseIterators, baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by concatenating streams produced by calling a\n * stream-generating function a given number of times.\n *\n * Since a `LazyIterator` is read-once, it cannot be repeated, but this\n * function can be used to achieve a similar effect:\n *\n *   LazyIterator.ofConcatenatedFunction(() => new MyIterator(), 6);\n *\n * @param iteratorFunc: A function that produces a new stream on each call.\n * @param count: The number of times to call the function.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenatedFunction<T>(\n    iteratorFunc: () => IteratorResult<LazyIterator<T>>, count: number,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return iteratorFromConcatenated(\n      iteratorFromFunction(iteratorFunc).take(count), baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by zipping together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nexport function iteratorFromZipped<O extends tf.TensorContainer>(\n    iterators: IteratorContainer,\n    mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL): LazyIterator<O> {\n  return new ZipIterator<O>(iterators, mismatchMode);\n}\n\n/**\n * An asynchronous iterator, providing lazy access to a potentially\n * unbounded stream of elements.\n *\n * Iterator can be obtained from a dataset:\n * `const iter = await dataset.iterator();`\n */\nexport abstract class LazyIterator<T> {\n  // This class implements AsyncIterator<T>, but we have not yet set the\n  // TypeScript --downlevelIteration flag to enable that.\n\n  abstract summary(): string;\n\n  /**\n   * Returns a `Promise` for the next element in the stream.\n   *\n   * When an item can be provided successfully, the return value is\n   * `{value:T, done:false}`.\n   *\n   * Calling next() on a closed stream returns `{value:null, done:true}`.\n   */\n  abstract async next(): Promise<IteratorResult<T>>;\n\n  /**\n   * Collect all remaining elements of a bounded stream into an array.\n   * Obviously this will succeed only for small streams that fit in memory.\n   * Useful for testing.\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArray(): Promise<T[]> {\n    const result: T[] = [];\n    let x = await this.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await this.next();\n    }\n    return result;\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArrayForTest(): Promise<T[]> {\n    const stream = this.prefetch(100);\n    const result: T[] = [];\n    let x = await stream.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await stream.next();\n    }\n    return result;\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveFully(): Promise<void> {\n    let x = await this.next();\n    while (!x.done) {\n      x = await this.next();\n    }\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted, or a predicate fails.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveWhile(predicate: (r: T) => boolean): Promise<void> {\n    let x = await this.next();\n    let shouldContinue = predicate(x.value);\n    while ((!x.done) && shouldContinue) {\n      x = await this.next();\n      shouldContinue = predicate(x.value);\n    }\n  }\n\n  /**\n   * Handles errors thrown on this stream using a provided handler function.\n   *\n   * @param handler A function that handles any `Error` thrown during a `next()`\n   *   call and returns true if the stream should continue (dropping the failed\n   *   call) or false if the stream should quietly terminate.  If the handler\n   *   itself throws (or rethrows) an `Error`, that will be propagated.\n   *\n   * @returns A `LazyIterator` of elements passed through from upstream,\n   *   possibly filtering or terminating on upstream `next()` calls that\n   *   throw an `Error`.\n   */\n  handleErrors(handler: (error: Error) => boolean): LazyIterator<T> {\n    return new ErrorHandlingLazyIterator(this, handler);\n  }\n\n  // TODO(soergel): Implement reduce() etc.\n\n  /**\n   * Filters this stream according to `predicate`.\n   *\n   * @param predicate A function mapping a stream element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `LazyIterator` of elements for which the predicate was true.\n   */\n  filter(predicate: (value: T) => boolean): LazyIterator<T> {\n    return new FilterIterator(this, predicate);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  map<O>(transform: (value: T) => O): LazyIterator<O> {\n    return new MapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through an async 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a `Promise` for a\n   *   transformed stream element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  mapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform, forcing serial execution.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  serialMapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform).serial();\n  }\n\n  /**\n   * Maps this stream through a 1-to-many transform.\n   *\n   * @param transform A function mapping a stream element to an array of\n   *   transformed elements.\n   *\n   * @returns A `DataStream` of transformed elements.\n   */\n  flatmap<O>(transform: (value: T) => O[]): LazyIterator<O> {\n    return new FlatmapIterator(this, transform);\n  }\n\n  /**\n   * Apply a function to every element of the stream.\n   *\n   * @param f A function to apply to each stream element.\n   */\n  async forEachAsync(f: (value: T) => void): Promise<void> {\n    return this.map(f).resolveFully();\n  }\n\n  /**\n   * Apply a function to every element of the stream, forcing serial execution.\n   *\n   * @param f A function to apply to each stream element.  Should return 'true'\n   *   to indicate that the stream should continue, or 'false' to cause it to\n   *   terminate.\n   */\n  async serialForEach(f: (value: T) => Promise<boolean>): Promise<void> {\n    return this.serialMapAsync(f).resolveWhile(x => (x === true));\n  }\n\n  /**\n   * Groups elements into batches, represented as arrays of elements.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"Row-major\" means that the resulting batch is simply a collection of\n   * rows: `[row1, row2, row3, ...]`.  This is contrast to the column-major\n   * form, which is needed for vectorized computation.\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `LazyIterator` of batches of elements, represented as arrays\n   *   of the original element type.\n   */\n  rowMajorBatch(batchSize: number, smallLastBatch = true): LazyIterator<T[]> {\n    return new RowMajorBatchIterator(this, batchSize, smallLastBatch);\n  }\n\n  /**\n   * Groups elements into batches, represented in column-major form.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"column-major\" means that the resulting batch is a (potentially\n   * nested) structure representing the columns.  Each column entry, then,\n   * contains a collection of the values found in that column for a range of\n   * input elements.  This representation allows for vectorized computation, in\n   * contrast to the row-major form.\n   *\n   * The inputs should all have the same nested structure (i.e., of arrays and\n   * dicts).  The result is a single object with the same nested structure,\n   * where the leaves are arrays collecting the values of the inputs at that\n   * location (or, optionally, the result of a custom function applied to those\n   * arrays).\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @param zipFn: (optional) A function that expects an array of elements at a\n   *   single node of the object tree, and returns a `DeepMapResult`.  The\n   *   `DeepMapResult` either provides a result value for that node (i.e.,\n   *   representing the subtree), or indicates that the node should be processed\n   *   recursively.  The default zipFn recurses as far as possible and places\n   *   arrays at the leaves.\n   * @returns A `LazyIterator` of batches of elements, represented as an object\n   *   with collections at the leaves.\n   */\n  columnMajorBatch(\n      batchSize: number, smallLastBatch = true,\n      // tslint:disable-next-line:no-any\n      zipFn: (xs: any[]) => DeepMapResult = zipToList):\n      LazyIterator<tf.TensorContainer> {\n    // First collect the desired number of input elements as a row-major batch.\n    const rowBatches = this.rowMajorBatch(batchSize, smallLastBatch);\n    // Now 'rotate' or 'pivot' the data, collecting all values from each column\n    // in the batch (i.e., for each key within the elements) into an array.\n    return rowBatches.map(x => deepZip(x, zipFn));\n  }\n\n  /**\n   * Concatenate this `LazyIterator` with another.\n   *\n   * @param iterator A `LazyIterator` to be concatenated onto this one.\n   * @param baseErrorHandler An optional function that can intercept `Error`s\n   *   raised during a `next()` call on the base stream.  This function can\n   *   decide whether the error should be propagated, whether the error should\n   *   be ignored, or whether the base stream should be terminated.\n   * @returns A `LazyIterator`.\n   */\n  concatenate(\n      iterator: LazyIterator<T>,\n      baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n    return new ChainedIterator(\n        iteratorFromItems([this, iterator]), baseErrorHandler);\n  }\n\n  /**\n   * Limits this stream to return at most `count` items.\n   *\n   * @param count The maximum number of items to provide from the stream. If\n   * a negative or undefined value is given, the entire stream is returned\n   *   unaltered.\n   */\n  take(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new TakeIterator(this, count);\n  }\n\n  /**\n   * Skips the first `count` items in this stream.\n   *\n   * @param count The number of items to skip.  If a negative or undefined\n   * value is given, the entire stream is returned unaltered.\n   */\n  skip(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new SkipIterator(this, count);\n  }\n\n  /**\n   * Prefetch the first `bufferSize` items in this stream.\n   *\n   * Note this prefetches Promises, but makes no guarantees about when those\n   * Promises resolve.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   */\n  prefetch(bufferSize: number): LazyIterator<T> {\n    return new PrefetchIterator(this, bufferSize);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  /**\n   * Randomly shuffles the elements of this stream.\n   *\n   * @param bufferSize: An integer specifying the number of elements from\n   * this stream from which the new stream will sample.\n   * @param seed: (Optional.) An integer specifying the random seed that\n   * will be used to create the distribution.\n   */\n  shuffle(windowSize: number, seed?: string): LazyIterator<T> {\n    return new ShuffleIterator(this, windowSize, seed);\n  }\n\n  /**\n   * Force an iterator to execute serially: each next() call will await the\n   * prior one, so that they cannot execute concurrently.\n   */\n  serial(): LazyIterator<T> {\n    return new SerialIterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on LazyIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// Iterators that just extend LazyIterator directly\n// ============================================================================\n\nclass ArrayIterator<T> extends LazyIterator<T> {\n  private trav = 0;\n  constructor(protected items: T[]) {\n    super();\n  }\n\n  summary() {\n    return `Array of ${this.items.length} items`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.trav >= this.items.length) {\n      return {value: null, done: true};\n    }\n    const item = this.items[this.trav];\n    this.trav++;\n    return {value: deepClone(item), done: false};\n  }\n}\n\nclass FunctionCallIterator<T> extends LazyIterator<T> {\n  constructor(\n      protected nextFn: () => IteratorResult<T>| Promise<IteratorResult<T>>) {\n    super();\n  }\n\n  summary() {\n    return `Function call`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    try {\n      return this.nextFn();\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message =\n          `Error thrown while iterating through a dataset: ${e.message}`;\n      throw e;\n    }\n  }\n}\n\nclass SerialIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(protected upstream: LazyIterator<T>) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Serial`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    return this.upstream.next();\n  }\n}\n\nclass SkipIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  count = 0;\n\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Skip`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider tradeoffs of reading in parallel, eg.\n    // collecting next() promises in an Array and then waiting for\n    // Promise.all() of those. Benefit: pseudo-parallel execution.  Drawback:\n    // maybe delayed GC.\n    while (this.count++ < this.maxCount) {\n      const skipped = await this.upstream.next();\n      // short-circuit if upstream is already empty\n      if (skipped.done) {\n        return skipped;\n      }\n      tf.dispose(skipped.value as {});\n    }\n    return this.upstream.next();\n  }\n}\n\nclass TakeIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Take`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.count++ >= this.maxCount) {\n      return {value: null, done: true};\n    }\n    return this.upstream.next();\n  }\n}\n\n// Note this batch just groups items into row-wise element arrays.\n// Rotating these to a column-wise representation happens only at the dataset\n// level.\nclass RowMajorBatchIterator<T> extends LazyIterator<T[]> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T[]>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected batchSize: number,\n      protected enableSmallLastBatch = true) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> RowMajorBatch`;\n  }\n\n  async next(): Promise<IteratorResult<T[]>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T[]>> {\n    const batch: T[] = [];\n    while (batch.length < this.batchSize) {\n      const item = await this.upstream.next();\n      if (item.done) {\n        if (this.enableSmallLastBatch && batch.length > 0) {\n          return {value: batch, done: false};\n        }\n        return {value: null, done: true};\n      }\n      batch.push(item.value);\n    }\n    return {value: batch, done: false};\n  }\n}\n\nclass FilterIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected predicate: (value: T) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Filter`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      const item = await this.upstream.next();\n      if (item.done || this.predicate(item.value)) {\n        return item;\n      }\n      tf.dispose(item.value as {});\n    }\n  }\n}\n\nclass MapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Map`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\nclass ErrorHandlingLazyIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected handler: (error: Error) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> handleErrors`;\n  }\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      try {\n        return await this.upstream.next();\n      } catch (e) {\n        if (!this.handler(e)) {\n          return {value: null, done: true};\n        }\n        // If the handler returns true, loop and fetch the next upstream item.\n\n        // If the upstream iterator throws an endless stream of errors, and if\n        // the handler says to ignore them, then we loop forever here.  That is\n        // the correct behavior-- it's up to the handler to decide when to stop.\n      }\n    }\n  }\n}\n\nclass AsyncMapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => Promise<O>) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> AsyncMap`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = await this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\n// Iterators that maintain a queue of pending items\n// ============================================================================\n\n/**\n * A base class for transforming streams that operate by maintaining an\n * output queue of elements that are ready to return via next().  This is\n * commonly required when the transformation is 1-to-many:  A call to next()\n * may trigger a call to the underlying stream, which will produce many\n * mapped elements of this stream-- of which we need to return only one, so\n * we have to queue the rest.\n */\nexport abstract class OneToManyIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  protected outputQueue: RingBuffer<T>;\n\n  constructor() {\n    super();\n    this.outputQueue = new GrowingRingBuffer<T>();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  /**\n   * Read one or more chunks from upstream and process them, possibly\n   * reading or writing a carryover, and adding processed items to the\n   * output queue.  Note it's possible that no items are added to the queue\n   * on a given pump() call, even if the upstream stream is not closed\n   * (e.g., because items are filtered).\n   *\n   * @return `true` if any action was taken, i.e. fetching items from the\n   *   upstream source OR adding items to the output queue.  `false` if the\n   *   upstream source is exhausted AND nothing was added to the queue\n   * (i.e., any remaining carryover).\n   */\n  protected abstract async pump(): Promise<boolean>;\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // Fetch so that the queue contains at least one item if possible.\n    // If the upstream source is exhausted, AND there are no items left in\n    // the output queue, then this stream is also exhausted.\n    while (this.outputQueue.length() === 0) {\n      // TODO(soergel): consider parallel reads.\n      if (!await this.pump()) {\n        return {value: null, done: true};\n      }\n    }\n    return {value: this.outputQueue.shift(), done: false};\n  }\n}\nclass FlatmapIterator<I, O> extends OneToManyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O[]) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Flatmap`;\n  }\n\n  async pump(): Promise<boolean> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return false;\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // that's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying any\n    // intermediate Tensors.  Here we are concerned only about the inputs.\n    const mappedArray = this.transform(item.value);\n    const outputTensors =\n        tf.tensor_util.getTensorsInContainer(mappedArray as {});\n    this.outputQueue.pushAll(mappedArray);\n\n    // TODO(soergel) faster intersection, and deduplicate outputTensors\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n\n    return true;\n  }\n}\n\n/**\n * Provides a `LazyIterator` that concatenates a stream of underlying\n * streams.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n */\nexport class ChainedIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>> = null;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private iterator: LazyIterator<T> = null;\n  private moreIterators: LazyIterator<LazyIterator<T>>;\n\n  constructor(\n      iterators: LazyIterator<LazyIterator<T>>,\n      private readonly baseErrorHandler?: (e: Error) => boolean) {\n    super();\n    this.moreIterators = iterators;\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of chained summaries';\n    return `${upstreamSummaries} -> Chained`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    this.lastRead = this.readFromChain(this.lastRead);\n    return this.lastRead;\n  }\n\n  private async readFromChain(lastRead: Promise<IteratorResult<T>>):\n      Promise<IteratorResult<T>> {\n    // Must await on the previous read since the previous read may have advanced\n    // the stream of streams, from which we need to read.\n    // This is unfortunate since we can't parallelize reads. Which means\n    // prefetching of chained streams is a no-op.\n    // One solution is to prefetch immediately upstream of this.\n    await lastRead;\n    if (this.iterator == null) {\n      const iteratorResult = await this.moreIterators.next();\n      if (iteratorResult.done) {\n        // No more streams to stream from.\n        return {value: null, done: true};\n      }\n      this.iterator = iteratorResult.value;\n      if (this.baseErrorHandler != null) {\n        this.iterator = this.iterator.handleErrors(this.baseErrorHandler);\n      }\n    }\n    const itemResult = await this.iterator.next();\n    if (itemResult.done) {\n      this.iterator = null;\n      return this.readFromChain(lastRead);\n    }\n    return itemResult;\n  }\n}\n\nexport enum ZipMismatchMode {\n  FAIL,      // require zipped streams to have the same length\n  SHORTEST,  // terminate zip when the first stream is exhausted\n  LONGEST    // use nulls for exhausted streams; use up the longest stream.\n}\n\n/**\n * Provides a `LazyIterator` that zips together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nclass ZipIterator<O extends tf.TensorContainer> extends LazyIterator<O> {\n  private count = 0;\n  private currentPromise: Promise<IteratorResult<O>> = null;\n\n  constructor(\n      protected readonly iterators: IteratorContainer,\n      protected readonly mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL) {\n    super();\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of zip summaries';\n    return `{${upstreamSummaries}} -> Zip`;\n  }\n\n  private async nextState(afterState: Promise<IteratorResult<O>>):\n      Promise<IteratorResult<O>> {\n    // This chaining ensures that the underlying next() are not even called\n    // before the previous ones have resolved.\n    await afterState;\n\n    // Collect underlying iterator \"done\" signals as a side effect in\n    // getNext()\n    let numIterators = 0;\n    let iteratorsDone = 0;\n\n    function getNext(container: IteratorContainer): DeepMapAsyncResult {\n      if (container instanceof LazyIterator) {\n        const result = container.next();\n        return {\n          value: result.then(x => {\n            numIterators++;\n            if (x.done) {\n              iteratorsDone++;\n            }\n            return x.value;\n          }),\n          recurse: false\n        };\n      } else {\n        return {value: null, recurse: true};\n      }\n    }\n\n    const mapped: O = await deepMapAndAwaitAll(this.iterators, getNext);\n\n    if (numIterators === iteratorsDone) {\n      // The streams have all ended.\n      return {value: null, done: true};\n    }\n    if (iteratorsDone > 0) {\n      switch (this.mismatchMode) {\n        case ZipMismatchMode.FAIL:\n          throw new Error(\n              'Zipped streams should have the same length. ' +\n              `Mismatched at element ${this.count}.`);\n        case ZipMismatchMode.SHORTEST:\n          return {value: null, done: true};\n        case ZipMismatchMode.LONGEST:\n        default:\n          // Continue.  The exhausted streams already produced value: null.\n      }\n    }\n\n    this.count++;\n    return {value: mapped, done: false};\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    this.currentPromise = this.nextState(this.currentPromise);\n    return this.currentPromise;\n  }\n}\n\n// Iterators that maintain a ring buffer of pending promises\n// ============================================================================\n\n/**\n * A stream that prefetches a given number of items from an upstream source,\n * returning them in FIFO order.\n *\n * Note this prefetches Promises, but makes no guarantees about when those\n * Promises resolve.\n */\nexport class PrefetchIterator<T> extends LazyIterator<T> {\n  protected buffer: RingBuffer<Promise<IteratorResult<T>>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected bufferSize: number) {\n    super();\n    this.buffer = new RingBuffer<Promise<IteratorResult<T>>>(bufferSize);\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Prefetch`;\n  }\n\n  /**\n   * Refill the prefetch buffer.  Returns only after the buffer is full, or\n   * the upstream source is exhausted.\n   */\n  protected refill() {\n    while (!this.buffer.isFull()) {\n      const v = this.upstream.next();\n      this.buffer.push(v);\n    }\n  }\n\n  next(): Promise<IteratorResult<T>> {\n    this.refill();\n    // This shift will never throw an error because the buffer is always\n    // full after a refill. If the stream is exhausted, the buffer will be\n    // full of Promises that will resolve to the end-of-stream signal.\n    return this.buffer.shift();\n  }\n}\n\n/**\n * A stream that performs a sliding-window random shuffle on an upstream\n * source. This is like a `PrefetchIterator` except that the items are\n * returned in randomized order.  Mixing naturally improves as the buffer\n * size increases.\n */\nexport class ShuffleIterator<T> extends PrefetchIterator<T> {\n  private readonly random: seedrandom.prng;\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private upstreamExhausted = false;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected windowSize: number,\n      seed?: string) {\n    super(upstream, windowSize);\n    this.random = seedrandom.alea(seed || tf.util.now().toString());\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private randomInt(max: number) {\n    return Math.floor(this.random() * max);\n  }\n\n  protected chooseIndex(): number {\n    return this.randomInt(this.buffer.length());\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider performance\n    if (!this.upstreamExhausted) {\n      this.refill();\n    }\n    while (!this.buffer.isEmpty()) {\n      const chosenIndex = this.chooseIndex();\n      const result = await this.buffer.shuffleExcise(chosenIndex);\n      if (result.done) {\n        this.upstreamExhausted = true;\n      } else {\n        this.refill();\n        return result;\n      }\n    }\n    return {value: null, done: true};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {TensorContainer, TensorLike} from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {iteratorFromConcatenated, iteratorFromFunction, iteratorFromItems, iteratorFromZipped, LazyIterator, ZipMismatchMode} from './iterators/lazy_iterator';\nimport {Container} from './types';\nimport {canTensorify, deepMapAndAwaitAll, DeepMapResult, isIterable} from './util/deep_map';\n\n/**\n * A nested structure of Datasets, used as the input to zip().\n */\nexport type DatasetContainer = Container<Dataset<TensorContainer>>;\n\n// TODO(soergel): consider vectorized operations within the pipeline.\n\n/**\n * Represents a potentially large list of independent data elements (typically\n * 'samples' or 'examples').\n *\n * A 'data example' may be a primitive, an array, a map from string keys to\n * values, or any nested structure of these.\n *\n * A `Dataset` represents an ordered collection of elements, together with a\n * chain of transformations to be performed on those elements. Each\n * transformation is a method of `Dataset` that returns another `Dataset`, so\n * these may be chained, e.g.\n * `const processedDataset = rawDataset.filter(...).map(...).batch(...)`.\n *\n * Data loading and transformation is done in a lazy, streaming fashion.  The\n * dataset may be iterated over multiple times; each iteration starts the data\n * loading anew and recapitulates the transformations.\n *\n * A `Dataset` is typically processed as a stream of unbatched examples --i.e.,\n * its transformations are applied one example at a time. Batching produces a\n * new `Dataset` where each element is a batch. Batching should usually come\n * last in a pipeline, because data transformations are easier to express on a\n * per-example basis than on a per-batch basis.\n *\n * The following code examples are calling `await dataset.forEachAsync(...)` to\n * iterate once over the entire dataset in order to print out the data.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport abstract class Dataset<T extends tf.TensorContainer> {\n  /*\n   * Provide a new stream of elements.  Note this will also start new streams\n   * from any underlying `Dataset`s.\n   *\n   * CAUTION: Any Tensors contained within the elements returned from\n   * this stream *must* be manually disposed to avoid a GPU memory leak.\n   * The tf.tidy() approach cannot be used in an asynchronous context.\n   */\n  abstract async iterator(): Promise<LazyIterator<T>>;\n\n  readonly size: number = null;\n\n  // TODO(soergel): Make Datasets report whether repeated iterator() calls\n  // produce the same result (e.g., reading from a file) or different results\n  // (e.g., from the webcam).  Currently we don't make this distinction but it\n  // could be important for the user to know.\n  // abstract isDeterministic(): boolean;\n\n  /**\n   * Groups elements into batches.\n   *\n   * It is assumed that each of the incoming dataset elements has the same\n   * structure-- i.e. the same set of keys at each location in an object\n   * hierarchy.  For each key, the resulting `Dataset` provides a batched\n   * element collecting all of the incoming values for that key.\n   *\n   *  * Incoming primitives are grouped into a 1-D Tensor.\n   *  * Incoming Tensors are grouped into a new Tensor where the 0'th axis is\n   *    the batch dimension.\n   *  * Incoming arrays are converted to Tensor and then batched.\n   *  * A nested array is interpreted as an n-D Tensor, so the batched result\n   *    has n+1 dimensions.\n   *  * An array that cannot be converted to Tensor produces an error.\n   *\n   * If an array should not be batched as a unit, it should first be converted\n   * to an object with integer keys.\n   *\n   * Here are a few examples:\n   *\n   * Batch a dataset of numbers:\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8]).batch(4);\n   * await a.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of arrays:\n   * ```js\n   * const b = tf.data.array([[1], [2], [3], [4], [5], [6], [7], [8]]).batch(4);\n   * await b.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of objects:\n   * ```js\n   * const c = tf.data.array([{a: 1, b: 11}, {a: 2, b: 12}, {a: 3, b: 13},\n   *   {a: 4, b: 14}, {a: 5, b: 15}, {a: 6, b: 16}, {a: 7, b: 17},\n   *   {a: 8, b: 18}]).batch(4);\n   * await c.forEachAsync(e => {\n   *   console.log('{');\n   *   for(var key in e) {\n   *     console.log(key+':');\n   *     e[key].print();\n   *   }\n   *   console.log('}');\n   * })\n   * ```\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `Dataset`, from which a stream of batches can be obtained.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  batch(batchSize: number, smallLastBatch = true): Dataset<tf.TensorContainer> {\n    const base = this;\n    tf.util.assert(\n        batchSize > 0, () => `batchSize needs to be positive, but it is\n      ${batchSize}`);\n    let size;\n    if (this.size === Infinity || this.size == null) {\n      // If the size of this dataset is infinity or null, the new size keeps the\n      // same.\n      size = this.size;\n    } else if (smallLastBatch) {\n      // If the size of this dataset is known and include small last batch, the\n      // new size is full batch count plus last batch.\n      size = Math.ceil(this.size / batchSize);\n    } else {\n      // If the size of this dataset is known and not include small last batch,\n      // the new size is full batch count.\n      size = Math.floor(this.size / batchSize);\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator())\n          .columnMajorBatch(batchSize, smallLastBatch, deepBatchConcat);\n    }, size);\n  }\n\n  /**\n   * Concatenates this `Dataset` with another.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * const b = tf.data.array([4, 5, 6]);\n   * const c = a.concatenate(b);\n   * await c.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param dataset A `Dataset` to be concatenated onto this one.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  concatenate(dataset: Dataset<T>): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity || dataset.size === Infinity) {\n      // If the size of any of these two dataset is infinity, new size is\n      // infinity.\n      size = Infinity;\n    } else if (this.size != null && dataset.size != null) {\n      // If the size of both datasets are known and not infinity, new size is\n      // sum the size of these two datasets.\n      size = this.size + dataset.size;\n    } else {\n      // If neither of these two datasets has infinite size and any of these two\n      // datasets' size is null, the new size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () =>\n            (await base.iterator()).concatenate(await dataset.iterator()),\n        size);\n  }\n\n  /**\n   * Filters this dataset according to `predicate`.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])\n   *   .filter(x => x%2 === 0);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param predicate A function mapping a dataset element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `Dataset` of elements for which the predicate was true.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  filter(predicate: (value: T) => boolean): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity) {\n      // If the size of this dataset is infinity, new size is infinity\n      size = Infinity;\n    } else {\n      // If this dataset has limited elements, new size is null because it might\n      // exhausted randomly.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).filter(x => tf.tidy(() => predicate(x)));\n    }, size);\n  }\n\n  /**\n   * Apply a function to every element of the dataset.\n   *\n   * After the function is applied to a dataset element, any Tensors contained\n   * within that element are disposed.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param f A function to apply to each dataset element.\n   * @returns A `Promise` that resolves after all elements have been processed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async forEachAsync(f: (input: T) => void): Promise<void> {\n    return (await this.iterator()).forEachAsync(f);\n  }\n\n  /**\n   * Maps this dataset through a 1-to-1 transform.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).map(x => x*x);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a transformed\n   *   dataset element.\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  map<O extends tf.TensorContainer>(transform: (value: T) => O): Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).map(x => tf.tidy(() => transform(x)));\n    }, this.size);\n  }\n\n  /**\n   * Maps this dataset through an async 1-to-1 transform.\n   *\n   * ```js\n   * const a =\n   *  tf.data.array([1, 2, 3]).mapAsync(x => new Promise(function(resolve){\n   *    setTimeout(() => {\n   *      resolve(x * x);\n   *    }, Math.random()*1000 + 500);\n   *  }));\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a `Promise` for a\n   *   transformed dataset element.  This transform is responsible for disposing\n   *   any intermediate `Tensor`s, i.e. by wrapping its computation in\n   *   `tf.tidy()`; that cannot be automated here (as it is in the synchronous\n   *   `map()` case).\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  mapAsync<O extends tf.TensorContainer>(transform: (value: T) => Promise<O>):\n      Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).mapAsync(transform);\n    }, this.size);\n  }\n\n  /**\n   *  Creates a `Dataset` that prefetches elements from this dataset.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  prefetch(bufferSize: number): Dataset<T> {\n    if (bufferSize == null) {\n      throw new RangeError(\n          '`Dataset.prefetch()` requires bufferSize to be specified.');\n    }\n\n    const base = this;\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).prefetch(bufferSize), this.size);\n  }\n\n  /**\n   * Repeats this dataset `count` times.\n   *\n   * NOTE: If this dataset is a function of global state (e.g. a random number\n   * generator), then different repetitions may produce different elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).repeat(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: (Optional) An integer, representing the number of times\n   *   the dataset should be repeated. The default behavior (if `count` is\n   *   `undefined` or negative) is for the dataset be repeated indefinitely.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  repeat(count?: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count > 0) {\n      // If this dataset has size and count is positive, new size is current\n      // size multiply count. This also covers the case that current size is\n      // infinity.\n      size = this.size * count;\n    } else if (count === 0) {\n      // If count is 0, new size is 0.\n      size = 0;\n    } else if (this.size != null && (count === undefined || count < 0)) {\n      // If this dataset has size and count is undefined or negative, the\n      // dataset will be repeated indefinitely and new size is infinity.\n      size = Infinity;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      const iteratorIterator = iteratorFromFunction(\n          async () => ({value: await base.iterator(), done: false}));\n      return iteratorFromConcatenated(iteratorIterator.take(count));\n    }, size);\n  }\n\n  /**\n   * Creates a `Dataset` that skips `count` initial elements from this dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).skip(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be skipped\n   *   to form the new dataset.  If `count` is greater than the size of this\n   *   dataset, the new dataset will contain no elements.  If `count`\n   *   is `undefined` or negative, skips the entire dataset.\n   *\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  skip(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count >= 0 && this.size >= count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is current size minus skipped size.This also covers the case that\n      // current size is infinity.\n      size = this.size - count;\n    } else if (\n        this.size != null &&\n        (this.size < count || count === undefined || count < 0)) {\n      // If the size of this dataset is smaller than count, or count is\n      // undefined or negative, skips the entire dataset and the new size is 0.\n      size = 0;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).skip(count), size);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  static readonly MAX_BUFFER_SIZE = 10000;\n\n  /**\n   * Pseudorandomly shuffles the elements of this dataset. This is done in a\n   * streaming manner, by sampling from a given number of prefetched elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).shuffle(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param bufferSize: An integer specifying the number of elements from this\n   *   dataset from which the new dataset will sample.\n   * @param seed: (Optional) An integer specifying the random seed that will\n   *   be used to create the distribution.\n   * @param reshuffleEachIteration: (Optional) A boolean, which if true\n   *   indicates that the dataset should be pseudorandomly reshuffled each time\n   *   it is iterated over. If false, elements will be returned in the same\n   *   shuffled order on each iteration. (Defaults to `true`.)\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  shuffle(bufferSize: number, seed?: string, reshuffleEachIteration = true):\n      Dataset<T> {\n    if (bufferSize == null || bufferSize < 0) {\n      if (this.size == null) {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.');\n      } else {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.  ' +\n            'If your data fits in main memory (for regular JS objects), ' +\n            'and/or GPU memory (for `tf.Tensor`s), consider setting ' +\n            `bufferSize to the dataset size (${this.size} elements)`);\n      }\n    }\n    const base = this;\n    const random = seedrandom.alea(seed || tf.util.now().toString());\n    return datasetFromIteratorFn(async () => {\n      let seed2 = random.int32();\n      if (reshuffleEachIteration) {\n        seed2 += random.int32();\n      }\n      return (await base.iterator()).shuffle(bufferSize, seed2.toString());\n    }, this.size);\n  }\n\n  /**\n   * Creates a `Dataset` with at most `count` initial elements from this\n   * dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).take(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be taken\n   *   to form the new dataset.  If `count` is `undefined` or negative, or if\n   *   `count` is greater than the size of this dataset, the new dataset will\n   *   contain all elements of this dataset.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  take(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && this.size > count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is count.\n      size = count;\n    } else if (this.size != null && this.size <= count) {\n      // If the size of this dataset is equal or smaller than count, the new\n      // dataset's size is the size of this dataset.\n      size = this.size;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).take(count), size);\n  }\n\n  /**\n   * Collect all elements of this dataset into an array.\n   *\n   * Obviously this will succeed only for small datasets that fit in memory.\n   * Useful for testing and generally should be avoided if possible.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]);\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async toArray() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArray();\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   */\n  async toArrayForTest() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArrayForTest();\n  }\n}\n\n/**\n * Create a `Dataset` defined by a provided iterator() function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * const ds = tf.data.datasetFromIteratorFn(iter);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n */\nexport function datasetFromIteratorFn<T extends tf.TensorContainer>(\n    iteratorFn: () => Promise<LazyIterator<T>>,\n    size: number = null): Dataset<T> {\n  return new class extends Dataset<T> {\n    size = size;\n\n    /*\n     * Provide a new stream of elements.  Note this will also start new streams\n     * from any underlying `Dataset`s.\n     */\n    async iterator(): Promise<LazyIterator<T>> {\n      return iteratorFn();\n    }\n  }\n  ();\n}\n\n/**\n * Create a `Dataset` from an array of elements.\n *\n * Create a Dataset from an array of objects:\n * ```js\n * const a = tf.data.array([{'item': 1}, {'item': 2}, {'item': 3}]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n *\n * Create a Dataset from an array of numbers:\n * ```js\n * const a = tf.data.array([4, 5, 6]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n * @param items An array of elements that will be parsed as items in a dataset.\n *\n * @doc {heading: 'Data', subheading: 'Creation', namespace: 'data'}\n */\nexport function array<T extends tf.TensorContainer>(items: T[]): Dataset<T> {\n  return datasetFromIteratorFn(\n      async () => iteratorFromItems(items), items.length);\n}\n\n/**\n * Create a `Dataset` by zipping together an array, dict, or nested\n * structure of `Dataset`s (and perhaps additional constants).\n * The underlying datasets must provide elements in a consistent order such that\n * they correspond.\n *\n * The number of elements in the resulting dataset is the same as the size of\n * the smallest dataset in datasets.\n *\n * The nested structure of the `datasets` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Note this means that, given an array of two datasets that produce dict\n * elements, the result is a dataset that produces elements that are arrays\n * of two dicts:\n *\n * Zip an array of datasets:\n * ```js\n * console.log('Zip two datasets of objects:');\n * const ds1 = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const ds2 = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const ds3 = tf.data.zip([ds1, ds2]);\n * await ds3.forEachAsync(e => console.log(JSON.stringify(e)));\n *\n * // If the goal is to merge the dicts in order to produce elements like\n * // {a: ..., b: ...}, this requires a second step such as:\n * console.log('Merge the objects:');\n * const ds4 = ds3.map(x => {return {a: x[0].a, b: x[1].b}});\n * await ds4.forEachAsync(e => console.log(e));\n * ```\n *\n * Zip a dict of datasets:\n * ```js\n * const a = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const b = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const c = tf.data.zip({c: a, d: b});\n * await c.forEachAsync(e => console.log(JSON.stringify(e)));\n * ```\n *\n * @doc {heading: 'Data', subheading: 'Operations', namespace: 'data'}\n */\nexport function zip<O extends tf.TensorContainer>(datasets: DatasetContainer):\n    Dataset<O> {\n  // manually type-check the argument for JS users\n  if (!isIterable(datasets)) {\n    throw new Error('The argument to zip() must be an object or array.');\n  }\n  let size;\n  if (Array.isArray(datasets)) {\n    for (let i = 0; i < datasets.length; i++) {\n      size = size == null ? (datasets[i] as Dataset<O>).size :\n                            Math.min(size, (datasets[i] as Dataset<O>).size);\n    }\n  } else if (datasets instanceof Object) {\n    for (const ds in datasets) {\n      size = size == null ? (datasets[ds] as Dataset<O>).size :\n                            Math.min(size, (datasets[ds] as Dataset<O>).size);\n    }\n  }\n  return datasetFromIteratorFn<O>(async () => {\n    const streams = await deepMapAndAwaitAll(datasets, d => {\n      if (d instanceof Dataset) {\n        return {value: d.iterator(), recurse: false};\n      } else if (isIterable(d)) {\n        return {value: null, recurse: true};\n      } else {\n        throw new Error(\n            'Leaves of the structure passed to zip() must be Datasets, ' +\n            'not primitives.');\n      }\n    });\n    return iteratorFromZipped<O>(streams, ZipMismatchMode.SHORTEST);\n  }, size);\n}\n\n/**\n * A zip function for use with deepZip, passed via the columnMajorBatch call.\n *\n * Accepts an array of identically-structured nested elements and either batches\n * them (if they are primitives, numeric arrays, or Tensors) or requests\n * recursion (if not).\n */\n// tslint:disable-next-line:no-any\nfunction deepBatchConcat(rows: any[]): DeepMapResult {\n  if (rows === null) {\n    return null;\n  }\n\n  // use the first item to decide whether to recurse or batch here.\n  const exampleRow = rows[0];\n\n  if (canTensorify(exampleRow)) {\n    // rows is an array of primitives, Tensors, or arrays.  Batch them.\n    const value = batchConcat(rows);\n    return {value, recurse: false};\n  }\n\n  // the example row is an object, so recurse into it.\n  return {value: null, recurse: true};\n}\n\n/**\n * Assembles a list of same-shaped numbers, number arrays, or Tensors\n * into a single new Tensor where axis 0 is the batch dimension.\n */\nfunction batchConcat<T extends(TensorLike | tf.Tensor)>(arrays: T[]):\n    tf.Tensor {\n  if (arrays.length === 0) {\n    // We can't return an empty Tensor because we don't know the element shape.\n    throw new Error('Can\\'t make a batch of zero elements.');\n  }\n\n  if (arrays[0] instanceof tf.Tensor) {\n    // Input is an array of Tensors\n    return tf.stack(arrays as tf.Tensor[]);\n  } else {\n    // Input is a possibly-nested array of numbers.\n    return tf.tensor(arrays as TensorLike);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\n\n/**\n * Represents a potentially large collection of text lines.\n *\n * The results are not batched.\n */\nexport class TextLineDataset extends Dataset<string> {\n  /**\n   * Create a `TextLineDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   */\n  constructor(protected readonly input: DataSource) {\n    super();\n  }\n\n  async iterator(): Promise<LazyIterator<string>> {\n    const inputIterator = await this.input.iterator();\n    const utf8Iterator = inputIterator.decodeUTF8();\n    const lineIterator = utf8Iterator.split('\\n').map(line => {\n      // Windows/DOS format text file has extra line breaker at the end of line.\n      if (line.endsWith('\\r')) {\n        line = line.slice(0, -1);\n      }\n      return line;\n    });\n    return lineIterator;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\nimport {ColumnConfig, CSVConfig} from '../types';\nimport {TextLineDataset} from './text_line_dataset';\n\nconst CODE_QUOTE = '\"';\nconst STATE_OUT = Symbol('out');\nconst STATE_FIELD = Symbol('field');\nconst STATE_QUOTE = Symbol('quote');\nconst STATE_QUOTE_AFTER_QUOTE = Symbol('quoteafterquote');\nconst STATE_WITHIN_QUOTE_IN_QUOTE = Symbol('quoteinquote');\n\n/**\n * Represents a potentially large collection of delimited text records.\n *\n * The produced `TensorContainer`s each contain one key-value pair for\n * every column of the table.  When a field is empty in the incoming data, the\n * resulting value is `undefined`, or throw error if it is required.  Values\n * that can be parsed as numbers are emitted as type `number`, other values\n * are parsed as `string`.\n *\n * The results are not batched.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport class CSVDataset extends Dataset<TensorContainer> {\n  base: TextLineDataset;\n  private hasHeader = true;\n  private fullColumnNames: string[] = null;\n  private columnNamesValidated = false;\n  private columnConfigs: {[key: string]: ColumnConfig} = null;\n  private configuredColumnsOnly = false;\n  private delimiter = ',';\n  private delimWhitespace = false;\n\n  /**\n   * Returns column names of the csv dataset. If `configuredColumnsOnly` is\n   * true, return column names in `columnConfigs`. If `configuredColumnsOnly` is\n   * false and `columnNames` is provided, `columnNames`. If\n   * `configuredColumnsOnly` is false and `columnNames` is not provided, return\n   * all column names parsed from the csv file. For example usage please go to\n   * `tf.data.csv`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async columnNames() {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    return this.configuredColumnsOnly ? Object.keys(this.columnConfigs) :\n                                        this.fullColumnNames;\n  }\n\n  /* 1) If `columnNames` is provided as string[], use this string[] as output\n   * keys in corresponding order. The length must match the number of inferred\n   * columns if `hasHeader` is true .\n   * 2) If `columnNames` is not provided, parse header line as `columnNames` if\n   * hasHeader is true. If `hasHeader` is false, throw an error.\n   * 3) If `columnConfigs` is provided, all the keys in `columnConfigs` must\n   * exist in parsed `columnNames`.\n   */\n  private async setColumnNames() {\n    const columnNamesFromFile = await this.maybeReadHeaderLine();\n    if (!this.fullColumnNames && !columnNamesFromFile) {\n      // Throw an error if columnNames is not provided and no header line.\n      throw new Error(\n          'Column names must be provided if there is no header line.');\n    } else if (this.fullColumnNames && columnNamesFromFile) {\n      // Check provided columnNames match header line.\n      util.assert(\n          columnNamesFromFile.length === this.fullColumnNames.length,\n          () => 'The length of provided columnNames (' +\n              this.fullColumnNames.length.toString() +\n              ') does not match the length of the header line read from ' +\n              'file (' + columnNamesFromFile.length.toString() + ').');\n    }\n    if (!this.fullColumnNames) {\n      this.fullColumnNames = columnNamesFromFile;\n    }\n    // Check if there are duplicate column names.\n    const counts: {[key: string]: number} = this.fullColumnNames.reduce(\n        (countAcc: {[key: string]: number}, name) => {\n          countAcc[name] = (countAcc[name] + 1) || 1;\n          return countAcc;\n        },\n        {});\n    const duplicateNames =\n        Object.keys(counts).filter((name) => (counts[name] > 1));\n    util.assert(\n        duplicateNames.length === 0,\n        () => 'Duplicate column names found: ' + duplicateNames.toString());\n    // Check if keys in columnConfigs match columnNames.\n    if (this.columnConfigs) {\n      for (const key of Object.keys(this.columnConfigs)) {\n        const index = this.fullColumnNames.indexOf(key);\n        if (index === -1) {\n          throw new Error(\n              'The key \"' + key +\n              '\" provided in columnConfigs does not match any of the column ' +\n              'names (' + this.fullColumnNames.toString() + ').');\n        }\n      }\n    }\n    this.columnNamesValidated = true;\n  }\n\n  private async maybeReadHeaderLine() {\n    if (this.hasHeader) {\n      const iter = await this.base.iterator();\n      const firstElement = await iter.next();\n      if (firstElement.done) {\n        throw new Error('No data was found for CSV parsing.');\n      }\n      const firstLine: string = firstElement.value;\n      const headers = this.parseRow(firstLine, false);\n      return headers;\n    } else {\n      return null;\n    }\n  }\n\n  /**\n   * Create a `CSVDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   * @param csvConfig (Optional) A CSVConfig object that contains configurations\n   *     of reading and decoding from CSV file(s).\n   *\n   *     hasHeader: (Optional) A boolean value that indicates whether the first\n   *     row of provided CSV file is a header line with column names, and should\n   *     not be included in the data. Defaults to `true`.\n   *\n   *     columnNames: (Optional) A list of strings that corresponds to\n   *     the CSV column names, in order. If provided, it ignores the column\n   *     names inferred from the header row. If not provided, infers the column\n   *     names from the first row of the records. If hasHeader is false and\n   *     columnNames is not provided, this method throws an error.\n   *\n   *     columnConfigs: (Optional) A dictionary whose key is column names, value\n   *     is an object stating if this column is required, column's data type,\n   *     default value, and if this column is label. If provided, keys must\n   *     correspond to names provided in columnNames or inferred from the file\n   *     header lines. If isLabel is true any column, returns an array of two\n   *     items: the first item is a dict of features key/value pairs, the second\n   *     item is a dict of labels key/value pairs. If no feature is marked as\n   *     label, returns a dict of features only.\n   *\n   *     configuredColumnsOnly (Optional) If true, only columns provided in\n   *     columnConfigs will be parsed and provided during iteration.\n   *\n   *     delimiter (Optional) The string used to parse each line of the input\n   *     file. Defaults to `,`.\n   */\n  constructor(protected readonly input: DataSource, csvConfig?: CSVConfig) {\n    super();\n    this.base = new TextLineDataset(input);\n    if (!csvConfig) {\n      csvConfig = {};\n    }\n    this.hasHeader = csvConfig.hasHeader === false ? false : true;\n    this.fullColumnNames = csvConfig.columnNames;\n    this.columnConfigs = csvConfig.columnConfigs;\n    this.configuredColumnsOnly = csvConfig.configuredColumnsOnly;\n    if (csvConfig.delimWhitespace) {\n      util.assert(\n          csvConfig.delimiter == null,\n          () =>\n              'Delimiter should not be provided when delimWhitespace is true.');\n      this.delimWhitespace = true;\n      this.delimiter = ' ';\n    } else {\n      this.delimiter = csvConfig.delimiter ? csvConfig.delimiter : ',';\n    }\n  }\n\n  async iterator(): Promise<LazyIterator<TensorContainer>> {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    let lines = await this.base.iterator();\n    if (this.hasHeader) {\n      // We previously read the first line to get the columnNames.\n      // Now that we're providing data, skip it.\n      lines = lines.skip(1);\n    }\n    return lines.map(x => this.makeDataElement(x));\n  }\n\n  makeDataElement(line: string): TensorContainer {\n    const values = this.parseRow(line);\n    const features: {[key: string]: TensorContainer} = {};\n    const labels: {[key: string]: TensorContainer} = {};\n\n    for (let i = 0; i < this.fullColumnNames.length; i++) {\n      const key = this.fullColumnNames[i];\n      const config = this.columnConfigs ? this.columnConfigs[key] : null;\n      if (this.configuredColumnsOnly && !config) {\n        // This column is not selected.\n        continue;\n      } else {\n        const value = values[i];\n        let parsedValue = null;\n        if (value === '') {\n          // If default value is provided, use it. If default value is not\n          // provided, set as undefined.\n          if (config && config.default !== undefined) {\n            parsedValue = config.default;\n          } else if (config && (config.required || config.isLabel)) {\n            throw new Error(\n                `Required column ${key} is empty in this line: ${line}`);\n          } else {\n            parsedValue = undefined;\n          }\n        } else {\n          // A value is present, so parse it based on type\n          const valueAsNum = Number(value);\n          if (isNaN(valueAsNum)) {\n            // The value is a string and this column is declared as boolean\n            // in config, parse it as boolean.\n            if (config && config.dtype === 'bool') {\n              parsedValue = this.getBoolean(value);\n            } else {\n              // Set value as string\n              parsedValue = value;\n            }\n          } else if (!config || !config.dtype) {\n            // If this value is a number and no type config is provided, return\n            // it as number.\n            parsedValue = valueAsNum;\n          } else {\n            // If this value is a number and data type is provided, parse it\n            // according to provided data type.\n            switch (config.dtype) {\n              case 'float32':\n                parsedValue = valueAsNum;\n                break;\n              case 'int32':\n                parsedValue = Math.floor(valueAsNum);\n                break;\n              case 'bool':\n                parsedValue = this.getBoolean(value);\n                break;\n              default:\n                parsedValue = valueAsNum;\n            }\n          }\n        }\n        // Check if this column is label.\n        (config && config.isLabel) ? labels[key] = parsedValue :\n                                     features[key] = parsedValue;\n      }\n    }\n    // If label exists, return an object of features and labels as {xs:features,\n    // ys:labels}, otherwise return features only.\n    if (Object.keys(labels).length === 0) {\n      return features;\n\n    } else {\n      return {xs: features, ys: labels};\n    }\n  }\n\n  private getBoolean(value: string): number {\n    if (value === '1' || value.toLowerCase() === 'true') {\n      return 1;\n    } else {\n      return 0;\n    }\n  }\n\n  // adapted from https://beta.observablehq.com/@mbostock/streaming-csv\n  private parseRow(line: string, validateElementCount = true): string[] {\n    const result: string[] = [];\n    let readOffset = 0;\n    const readLength = line.length;\n    let currentState = STATE_OUT;\n    // Goes through the line to parse quote.\n    for (let i = 0; i < readLength; i++) {\n      switch (currentState) {\n        // Before enter a new field\n        case STATE_OUT:\n          switch (line.charAt(i)) {\n            // Enter a quoted field\n            case CODE_QUOTE:\n              readOffset = i + 1;\n              currentState = STATE_QUOTE;\n              break;\n            // Read an empty field\n            case this.delimiter:\n              readOffset = i + 1;\n              // If delimiter is white space and configured to collapse\n              // multiple white spaces, ignore this white space.\n              if (this.delimiter === ' ' && this.delimWhitespace) {\n                break;\n              }\n              result.push('');\n              currentState = STATE_OUT;\n              break;\n            // Enter an unquoted field\n            default:\n              currentState = STATE_FIELD;\n              readOffset = i;\n              break;\n          }\n          break;\n        // In an unquoted field\n        case STATE_FIELD:\n          switch (line.charAt(i)) {\n            // Exit an unquoted field, add it to result\n            case this.delimiter:\n              result.push(line.substring(readOffset, i));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            default:\n          }\n          break;\n        // In a quoted field\n        case STATE_QUOTE:\n          switch (line.charAt(i)) {\n            // Read a quote after a quote\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE_AFTER_QUOTE;\n              break;\n            default:\n          }\n          break;\n        // This state means it's right after a second quote in a field\n        case STATE_QUOTE_AFTER_QUOTE:\n          switch (line.charAt(i)) {\n            // Finished a quoted field\n            case this.delimiter:\n              result.push(line.substring(readOffset, i - 1));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            // Finished a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            // In a quoted part in a quoted field\n            default:\n              currentState = STATE_WITHIN_QUOTE_IN_QUOTE;\n              break;\n          }\n          break;\n        case STATE_WITHIN_QUOTE_IN_QUOTE:\n          switch (line.charAt(i)) {\n            // Exit a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            default:\n          }\n          break;\n        default:\n      }\n    }\n    // Adds last item based on if it is quoted.\n    if (currentState === STATE_QUOTE_AFTER_QUOTE) {\n      result.push(line.substring(readOffset, readLength - 1));\n    } else {\n      result.push(line.substring(readOffset));\n    }\n    // Check if each row has the same number of elements as column names.\n    if (validateElementCount && result.length !== this.fullColumnNames.length) {\n      throw new Error(`Invalid row in csv file. Should have ${\n          this.fullColumnNames.length} elements in a row, but got ${result}`);\n    }\n    return result;\n  }\n}\n\n// TODO(soergel): add more basic datasets for parity with tf.data\n// tf.data.FixedLengthRecordDataset()\n// tf.data.TFRecordDataset()\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env, Tensor, tensor, Tensor2D, Tensor3D, TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {MicrophoneConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of tensors from microphone audio stream. The tensors are\n * representing audio data as frequency-domain spectrogram generated with\n * browser's native FFT. Tensors representing time-domain waveform is available\n * based on configuration. Only works in browser environment.\n */\nexport class MicrophoneIterator extends LazyIterator<TensorContainer> {\n  private isClosed = false;\n  private stream: MediaStream;\n  private readonly fftSize: number;\n  private readonly columnTruncateLength: number;\n  private freqData: Float32Array;\n  private timeData: Float32Array;\n  private readonly numFrames: number;\n  private analyser: AnalyserNode;\n  private audioContext: AudioContext;\n  private sampleRateHz: number;\n  private readonly audioTrackConstraints: MediaTrackConstraints;\n  private readonly smoothingTimeConstant: number;\n  private readonly includeSpectrogram: boolean;\n  private readonly includeWaveform: boolean;\n\n  private constructor(protected readonly microphoneConfig: MicrophoneConfig) {\n    super();\n    this.fftSize = microphoneConfig.fftSize || 1024;\n    const fftSizeLog2 = Math.log2(this.fftSize);\n    if (this.fftSize < 0 || fftSizeLog2 < 4 || fftSizeLog2 > 14 ||\n        !Number.isInteger(fftSizeLog2)) {\n      throw new Error(\n          `Invalid fftSize: it must be a power of 2 between ` +\n          `2 to 4 and 2 to 14, but got ${this.fftSize}`);\n    }\n\n    this.numFrames = microphoneConfig.numFramesPerSpectrogram || 43;\n    this.sampleRateHz = microphoneConfig.sampleRateHz;\n    this.columnTruncateLength =\n        microphoneConfig.columnTruncateLength || this.fftSize;\n    this.audioTrackConstraints = microphoneConfig.audioTrackConstraints;\n    this.smoothingTimeConstant = microphoneConfig.smoothingTimeConstant || 0;\n\n    this.includeSpectrogram =\n        microphoneConfig.includeSpectrogram === false ? false : true;\n    this.includeWaveform =\n        microphoneConfig.includeWaveform === true ? true : false;\n    if (!this.includeSpectrogram && !this.includeWaveform) {\n      throw new Error(\n          'Both includeSpectrogram and includeWaveform are false. ' +\n          'At least one type of data should be returned.');\n    }\n  }\n\n  summary() {\n    return `microphone`;\n  }\n\n  // Construct a MicrophoneIterator and start the audio stream.\n  static async create(microphoneConfig: MicrophoneConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'microphone API is only supported in browser environment.');\n    }\n\n    const microphoneIterator = new MicrophoneIterator(microphoneConfig);\n\n    // Call async function start() to initialize the audio stream.\n    await microphoneIterator.start();\n\n    return microphoneIterator;\n  }\n\n  // Start the audio stream and FFT.\n  async start(): Promise<void> {\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        audio: this.audioTrackConstraints == null ? true :\n                                                    this.audioTrackConstraints,\n        video: false\n      });\n    } catch (e) {\n      throw new Error(\n          `Error thrown while initializing video stream: ${e.message}`);\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain audio from microphone.');\n    }\n\n    const ctxConstructor =\n        // tslint:disable-next-line:no-any\n        (window as any).AudioContext || (window as any).webkitAudioContext;\n    this.audioContext = new ctxConstructor();\n\n    if (!this.sampleRateHz) {\n      // If sample rate is not provided, use the available sample rate on\n      // device.\n      this.sampleRateHz = this.audioContext.sampleRate;\n    } else if (this.audioContext.sampleRate !== this.sampleRateHz) {\n      throw new Error(\n          `Mismatch in sampling rate: ` +\n          `Expected: ${this.sampleRateHz}; ` +\n          `Actual: ${this.audioContext.sampleRate}`);\n    }\n\n    const streamSource = this.audioContext.createMediaStreamSource(this.stream);\n    this.analyser = this.audioContext.createAnalyser();\n    this.analyser.fftSize = this.fftSize * 2;\n    this.analyser.smoothingTimeConstant = this.smoothingTimeConstant;\n    streamSource.connect(this.analyser);\n    this.freqData = new Float32Array(this.fftSize);\n    this.timeData = new Float32Array(this.fftSize);\n    return;\n  }\n\n  async next(): Promise<IteratorResult<TensorContainer>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let spectrogramTensor: Tensor;\n    let waveformTensor: Tensor;\n\n    const audioDataQueue = await this.getAudioData();\n    if (this.includeSpectrogram) {\n      const freqData = this.flattenQueue(audioDataQueue.freqDataQueue);\n      spectrogramTensor = this.getTensorFromAudioDataArray(\n          freqData, [this.numFrames, this.columnTruncateLength, 1]);\n    }\n    if (this.includeWaveform) {\n      const timeData = this.flattenQueue(audioDataQueue.timeDataQueue);\n      waveformTensor = this.getTensorFromAudioDataArray(\n          timeData, [this.numFrames * this.fftSize, 1]);\n    }\n\n    return {\n      value: {'spectrogram': spectrogramTensor, 'waveform': waveformTensor},\n      done: false\n    };\n  }\n\n  // Capture one result from the audio stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<{spectrogram: Tensor3D, waveform: Tensor2D}> {\n    return (await this.next()).value as\n        {spectrogram: Tensor3D, waveform: Tensor2D};\n  }\n\n  private async getAudioData():\n      Promise<{freqDataQueue: Float32Array[], timeDataQueue: Float32Array[]}> {\n    const freqDataQueue: Float32Array[] = [];\n    const timeDataQueue: Float32Array[] = [];\n    let currentFrames = 0;\n    return new Promise(resolve => {\n      const intervalID = setInterval(() => {\n        if (this.includeSpectrogram) {\n          this.analyser.getFloatFrequencyData(this.freqData);\n          // If the audio stream is initializing, return empty queue.\n          if (this.freqData[0] === -Infinity) {\n            resolve({freqDataQueue, timeDataQueue});\n          }\n          freqDataQueue.push(this.freqData.slice(0, this.columnTruncateLength));\n        }\n        if (this.includeWaveform) {\n          this.analyser.getFloatTimeDomainData(this.timeData);\n          timeDataQueue.push(this.timeData.slice());\n        }\n\n        // Clean interval and return when all frames have been collected\n        if (++currentFrames === this.numFrames) {\n          clearInterval(intervalID);\n          resolve({freqDataQueue, timeDataQueue});\n        }\n      }, this.fftSize / this.sampleRateHz * 1e3);\n    });\n  }\n\n  // Stop the audio stream and pause the iterator.\n  stop(): void {\n    if (!this.isClosed) {\n      this.isClosed = true;\n      this.analyser.disconnect();\n      this.audioContext.close();\n      if (this.stream != null && this.stream.getTracks().length > 0) {\n        this.stream.getTracks()[0].stop();\n      }\n    }\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor[]> {\n    throw new Error('Can not convert infinite audio stream to array.');\n  }\n\n  // Return audio sampling rate in Hz\n  getSampleRate(): number {\n    return this.sampleRateHz;\n  }\n\n  private flattenQueue(queue: Float32Array[]): Float32Array {\n    const frameSize = queue[0].length;\n    const freqData = new Float32Array(queue.length * frameSize);\n    queue.forEach((data, i) => freqData.set(data, i * frameSize));\n    return freqData;\n  }\n\n  private getTensorFromAudioDataArray(freqData: Float32Array, shape: number[]):\n      Tensor {\n    const vals = new Float32Array(util.sizeFromShape(shape));\n    // If the data is less than the output shape, the rest is padded with zeros.\n    vals.set(freqData, vals.length - freqData.length);\n    return tensor(vals, shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {browser, env, image, tensor1d, Tensor1D, tensor2d, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\nimport {WebcamConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of image tensors from webcam video stream. Only works in\n * browser environment.\n */\nexport class WebcamIterator extends LazyIterator<Tensor3D> {\n  private isClosed = true;\n  private stream: MediaStream;\n  private resize = false;\n  private cropSize: [number, number];\n  private cropBox: Tensor2D;\n  private cropBoxInd: Tensor1D;\n\n  private constructor(\n      protected readonly webcamVideoElement: HTMLVideoElement,\n      protected readonly webcamConfig: WebcamConfig) {\n    super();\n    if (this.needToResize()) {\n      this.resize = true;\n      this.cropSize =\n          [this.webcamConfig.resizeHeight, this.webcamConfig.resizeWidth];\n      this.cropBoxInd = tensor1d([0], 'int32');\n      if (this.webcamConfig.centerCrop) {\n        // Calculate the box based on resizing shape.\n        const widthCroppingRatio =\n            this.webcamConfig.resizeWidth * 1.0 / this.webcamVideoElement.width;\n        const heightCroppingRatio = this.webcamConfig.resizeHeight * 1.0 /\n            this.webcamVideoElement.height;\n        const widthCropStart = (1 - widthCroppingRatio) / 2;\n        const heightCropStart = (1 - heightCroppingRatio) / 2;\n        const widthCropEnd = widthCropStart + widthCroppingRatio;\n        const heightCropEnd = heightCroppingRatio + heightCropStart;\n        this.cropBox = tensor2d(\n            [heightCropStart, widthCropStart, heightCropEnd, widthCropEnd],\n            [1, 4]);\n      } else {\n        this.cropBox = tensor2d([0, 0, 1, 1], [1, 4]);\n      }\n    }\n  }\n\n  summary() {\n    return `webcam`;\n  }\n\n  // Construct a WebcamIterator and start it's video stream.\n  static async create(\n      webcamVideoElement?: HTMLVideoElement, webcamConfig: WebcamConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'tf.data.webcam is only supported in browser environment.');\n    }\n\n    if (!webcamVideoElement) {\n      // If webcam video element is not provided, create a hidden video element\n      // with provided width and height.\n      webcamVideoElement = document.createElement('video');\n      if (!webcamConfig.resizeWidth || !webcamConfig.resizeHeight) {\n        throw new Error(\n            'Please provide webcam video element, or resizeWidth and ' +\n            'resizeHeight to create a hidden video element.');\n      }\n      webcamVideoElement.width = webcamConfig.resizeWidth;\n      webcamVideoElement.height = webcamConfig.resizeHeight;\n    }\n    const webcamIterator = new WebcamIterator(webcamVideoElement, webcamConfig);\n\n    // Call async function to initialize the video stream.\n    await webcamIterator.start();\n\n    return webcamIterator;\n  }\n\n  // Async function to start video stream.\n  async start(): Promise<void> {\n    if (this.webcamConfig.facingMode) {\n      util.assert(\n          (this.webcamConfig.facingMode === 'user') ||\n              (this.webcamConfig.facingMode === 'environment'),\n          () =>\n              `Invalid webcam facing mode: ${this.webcamConfig.facingMode}. ` +\n              `Please provide 'user' or 'environment'`);\n    }\n\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        video: {\n          deviceId: this.webcamConfig.deviceId,\n          facingMode: this.webcamConfig.facingMode ?\n              this.webcamConfig.facingMode :\n              'user',\n          width: this.webcamVideoElement.width,\n          height: this.webcamVideoElement.height\n        }\n      });\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message = `Error thrown while initializing video stream: ${e.message}`;\n      throw e;\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain video from webcam.');\n    }\n\n    // Older browsers may not have srcObject\n    try {\n      this.webcamVideoElement.srcObject = this.stream;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = window.URL.createObjectURL(this.stream);\n    }\n    // Start the webcam video stream\n    this.webcamVideoElement.play();\n\n    this.isClosed = false;\n\n    return new Promise<void>(resolve => {\n      // Add event listener to make sure the webcam has been fully initialized.\n      this.webcamVideoElement.onloadedmetadata = () => {\n        resolve();\n      };\n    });\n  }\n\n  async next(): Promise<IteratorResult<Tensor3D>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let img;\n    try {\n      img = browser.fromPixels(this.webcamVideoElement);\n    } catch (e) {\n      throw new Error(\n          `Error thrown converting video to pixels: ${JSON.stringify(e)}`);\n    }\n    if (this.resize) {\n      try {\n        return {value: this.cropAndResizeFrame(img), done: false};\n      } catch (e) {\n        throw new Error(`Error thrown cropping the video: ${e.message}`);\n      } finally {\n        img.dispose();\n      }\n    } else {\n      return {value: img, done: false};\n    }\n  }\n\n  private needToResize() {\n    // If resizeWidth and resizeHeight are provided, and different from the\n    // width and height of original HTMLVideoElement, then resizing and cropping\n    // is required.\n    if (this.webcamConfig.resizeWidth && this.webcamConfig.resizeHeight &&\n        (this.webcamVideoElement.width !== this.webcamConfig.resizeWidth ||\n         this.webcamVideoElement.height !== this.webcamConfig.resizeHeight)) {\n      return true;\n    }\n    return false;\n  }\n\n  // Cropping and resizing each frame based on config\n  cropAndResizeFrame(img: Tensor3D): Tensor3D {\n    return tidy(() => {\n      const expandedImage: Tensor4D = img.toFloat().expandDims(0);\n      let resizedImage;\n      resizedImage = image.cropAndResize(\n          expandedImage, this.cropBox, this.cropBoxInd, this.cropSize,\n          'bilinear');\n      // Extract image from batch cropping.\n      const shape = resizedImage.shape;\n      return resizedImage.reshape(shape.slice(1) as [number, number, number]);\n    });\n  }\n\n  // Capture one frame from the video stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<Tensor3D> {\n    return (await this.next()).value;\n  }\n\n  // Stop the video stream and pause webcam iterator.\n  stop(): void {\n    const tracks = this.stream.getTracks();\n\n    tracks.forEach(track => track.stop());\n\n    try {\n      this.webcamVideoElement.srcObject = null;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = null;\n    }\n    this.isClosed = true;\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor3D[]> {\n    throw new Error('Can not convert infinite video stream to array.');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {ByteChunkIterator} from './iterators/byte_chunk_iterator';\n\n/**\n * Represents a data source readable as a stream of binary data chunks.\n *\n * Because `Dataset`s can be read repeatedly (via `Dataset.iterator()`), this\n * provides a means to repeatedly create streams from the underlying data\n * sources.\n */\nexport abstract class DataSource {\n  /**\n   * Obtain a new stream of binary data chunks.\n   *\n   * Starts the new stream from the beginning of the data source, even if other\n   * streams have been obtained previously.\n   */\n  abstract async iterator(): Promise<ByteChunkIterator>;\n\n  // TODO(soergel): consider chainable Dataset construction here\n}\n\n// TODO(soergel): consider convenience factory functions here\n// in combination with chainable source->dataset above, e.g.:\n// tf.data.url(...).asCsvDataset().shuffle().batch()\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\n\nexport abstract class StringIterator extends LazyIterator<string> {\n  /**\n   * Splits a string stream on a given separator.\n   *\n   * It is assumed that the incoming chunk boundaries have no semantic meaning,\n   * so conceptually the incoming stream is treated simply as the concatenation\n   * of its elements.\n   *\n   * The outgoing stream provides chunks corresponding to the results of the\n   * standard string split() operation (even if such a chunk spanned incoming\n   * chunks).  The separators are not included.\n   *\n   * A typical usage is to split a text file (represented as a stream with\n   * arbitrary chunk boundaries) into lines.\n   *\n   * @param upstream A readable stream of strings that can be treated as\n   *   concatenated.\n   * @param separator A character to split on.\n   */\n  split(separator: string): StringIterator {\n    return new SplitIterator(this, separator);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on StringIterator.  Unfortunately they can't be placed in separate files, due\n// to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class SplitIterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass SplitIterator extends StringIterator {\n  private impl: SplitIteratorImpl;\n\n  constructor(protected upstream: LazyIterator<string>, separator: string) {\n    super();\n    this.impl = new SplitIteratorImpl(upstream, separator);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\nclass SplitIteratorImpl extends OneToManyIterator<string> {\n  // A partial string at the end of an upstream chunk\n  carryover = '';\n\n  constructor(\n      protected upstream: LazyIterator<string>, protected separator: string) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Split('${this.separator}')`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    if (chunkResult.done) {\n      if (this.carryover === '') {\n        return false;\n      }\n\n      // Pretend that the pump succeeded in order to emit the small last batch.\n      // The next pump() call will actually fail.\n      this.outputQueue.push(this.carryover);\n      this.carryover = '';\n      return true;\n    }\n    const lines = chunkResult.value.split(this.separator);\n    // Note the behavior: \" ab \".split(' ') === ['', 'ab', '']\n    // Thus the carryover may be '' if the separator falls on a chunk\n    // boundary; this produces the correct result.\n\n    lines[0] = this.carryover + lines[0];\n    for (const line of lines.slice(0, -1)) {\n      this.outputQueue.push(line);\n    }\n    this.carryover = lines[lines.length - 1];\n\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\nimport {StringIterator} from './string_iterator';\n\nexport abstract class ByteChunkIterator extends LazyIterator<Uint8Array> {\n  /**\n   * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n   *\n   * The byte arrays producetd from the ByteChunkIterator on which this is\n   * called will be interpreted as concatenated.  No assumptions are made about\n   * the boundaries of the incoming chunks, so a multi-byte UTF8 encoding of a\n   * character may span the boundary between chunks.  This naturally happens,\n   * for instance, when reading fixed-size byte arrays from a file.\n   */\n  decodeUTF8(): StringIterator {\n    return new Utf8Iterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on ByteChunkIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class Utf8Iterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass Utf8Iterator extends StringIterator {\n  private impl: Utf8IteratorImpl;\n\n  constructor(protected upstream: LazyIterator<Uint8Array>) {\n    super();\n    this.impl = new Utf8IteratorImpl(upstream);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\n/**\n * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n *\n * This is tricky because the incoming byte array boundaries may disrupt a\n * multi-byte UTF8 character. Thus any incomplete character data at the end of\n * a chunk must be carried over and prepended to the next chunk before\n * decoding. Luckily with native decoder, TextDecoder in browser and\n * string_decoder in node, byte array boundaries are handled automatically.\n *\n * In the context of an input pipeline for machine learning, UTF8 decoding is\n * needed to parse text files containing training examples or prediction\n * requests (e.g., formatted as CSV or JSON). We cannot use the built-in\n * decoding provided by FileReader.readAsText() because here we are in a\n * streaming context, which FileReader does not support.\n *\n * @param upstream A `LazyIterator` of `Uint8Arrays` containing UTF8-encoded\n *   text, which should be interpreted as concatenated.  No assumptions are\n *   made about the boundaries of the incoming chunks, so a multi-byte UTF8\n *   encoding of a character may span the boundary between chunks.  This\n *   naturally happens, for instance, when reading fixed-size byte arrays from a\n *   file.\n */\nclass Utf8IteratorImpl extends OneToManyIterator<string> {\n  // `decoder` as `any` here to dynamically assign value based on the\n  // environment.\n  // tslint:disable-next-line:no-any\n  decoder: any;\n\n  constructor(protected readonly upstream: LazyIterator<Uint8Array>) {\n    super();\n    if (env().get('IS_BROWSER')) {\n      this.decoder = new TextDecoder('utf-8');\n    } else {\n      // tslint:disable-next-line:no-require-imports\n      const {StringDecoder} = require('string_decoder');\n      this.decoder = new StringDecoder('utf8');\n    }\n  }\n  summary() {\n    return `${this.upstream.summary()} -> Utf8`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    let chunk;\n    if (chunkResult.done) {\n      return false;\n    } else {\n      chunk = chunkResult.value;\n    }\n\n    let text: string;\n    if (env().get('IS_BROWSER')) {\n      text = this.decoder.decode(chunk, {stream: true});\n    } else {\n      text = this.decoder.write(Buffer.from(chunk.buffer));\n    }\n    this.outputQueue.push(text);\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// inspired by https://github.com/maxogden/filereader-stream\nimport {env, util} from '@tensorflow/tfjs-core';\nimport {FileElement} from '../types';\nimport {ByteChunkIterator} from './byte_chunk_iterator';\n\nexport interface FileChunkIteratorOptions {\n  /** The byte offset at which to begin reading the File or Blob. Default 0. */\n  offset?: number;\n  /** The number of bytes to read at a time. Default 1MB. */\n  chunkSize?: number;\n}\n\n/**\n * Provide a stream of chunks from a File, Blob, or Uint8Array.\n * @param file The source File, Blob or Uint8Array.\n * @param options Optional settings controlling file reading.\n * @returns a lazy Iterator of Uint8Arrays containing sequential chunks of the\n *   input File, Blob or Uint8Array.\n */\nexport class FileChunkIterator extends ByteChunkIterator {\n  offset: number;\n  chunkSize: number;\n\n  constructor(\n      protected file: FileElement,\n      protected options: FileChunkIteratorOptions = {}) {\n    super();\n    util.assert(\n        (file instanceof Uint8Array) ||\n            (env().get('IS_BROWSER') ?\n                 (file instanceof File || file instanceof Blob) :\n                 false),\n        () => 'FileChunkIterator only supports File, Blob and Uint8Array ' +\n            'right now.');\n    this.offset = options.offset || 0;\n    // default 1MB chunk has tolerable perf on large files\n    this.chunkSize = options.chunkSize || 1024 * 1024;\n  }\n\n  summary() {\n    return `FileChunks ${this.file}`;\n  }\n\n  async next(): Promise<IteratorResult<Uint8Array>> {\n    if (this.offset >= ((this.file instanceof Uint8Array) ?\n                            this.file.byteLength :\n                            this.file.size)) {\n      return {value: null, done: true};\n    }\n    const chunk = new Promise<Uint8Array>((resolve, reject) => {\n      const end = this.offset + this.chunkSize;\n      if (this.file instanceof Uint8Array) {\n        // Note if end > this.uint8Array.byteLength, we just get a small last\n        // chunk.\n        resolve(new Uint8Array(this.file.slice(this.offset, end)));\n      } else {\n        // This branch assumes that this.file type is File or Blob, which\n        // means it is in the browser environment.\n\n        // TODO(soergel): is this a performance issue?\n        const fileReader = new FileReader();\n        fileReader.onload = (event) => {\n          let data: string|ArrayBuffer|Uint8Array = fileReader.result;\n          // Not sure we can trust the return type of\n          // FileReader.readAsArrayBuffer See e.g.\n          // https://github.com/node-file-api/FileReader/issues/2\n          if (data instanceof ArrayBuffer) {\n            data = new Uint8Array(data);\n          }\n          if (!(data instanceof Uint8Array)) {\n            return reject(new TypeError('FileReader returned unknown type.'));\n          }\n          resolve(data);\n        };\n        fileReader.onabort = (event) => {\n          return reject(new Error('Aborted'));\n        };\n        fileReader.onerror = (event) => {\n          return reject(new Error(event.type));\n        };\n        // TODO(soergel): better handle onabort, onerror\n        // Note if end > this.file.size, we just get a small last chunk.\n        const slice = this.file.slice(this.offset, end);\n        // We can't use readAsText here (even if we know the file is text)\n        // because the slice boundary may fall within a multi-byte character.\n        fileReader.readAsArrayBuffer(slice);\n      }\n      this.offset = end;\n    });\n    return {value: (await chunk), done: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\nimport {FileChunkIterator, FileChunkIteratorOptions} from './file_chunk_iterator';\n\n/**\n * Provide a stream of chunks from a URL.\n *\n * Note this class first downloads the entire file into memory before providing\n * the first element from the stream.  This is because the Fetch API does not\n * yet reliably provide a reader stream for the response body.\n */\nexport async function urlChunkIterator(\n    url: RequestInfo, options: FileChunkIteratorOptions = {}) {\n  let urlString;\n  let requestInit;\n  if ((typeof url) === 'string') {\n    urlString = url as string;\n  } else {\n    urlString = (url as Request).url;\n    requestInit = getRequestInitFromRequest(url as Request);\n  }\n  const response = await util.fetch(urlString, requestInit);\n  if (response.ok) {\n    const uint8Array = new Uint8Array(await response.arrayBuffer());\n    return new FileChunkIterator(uint8Array, options);\n  } else {\n    throw new Error(response.statusText);\n  }\n}\n\n// Generate RequestInit from Request to match tf.util.fetch signature.\nconst getRequestInitFromRequest = (request: Request) => {\n  const init = {\n    method: request.method,\n    headers: request.headers,\n    body: request.body,\n    mode: request.mode,\n    credentials: request.credentials,\n    cache: request.cache,\n    redirect: request.redirect,\n    referrer: request.referrer,\n    integrity: request.integrity,\n  };\n  return init;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// Skip tslint any type check cause this method is aiming to check type of\n// input.\n// tslint:disable-next-line:no-any\nexport function isLocalPath(source: any): boolean {\n  return (typeof source === 'string') && source.substr(0, 7) === 'file://';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIterator, FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {FileElement} from '../types';\nimport {isLocalPath} from '../util/source_util';\n\n/**\n * Represents a file, blob, or Uint8Array readable as a stream of binary data\n * chunks.\n */\nexport class FileDataSource extends DataSource {\n  /**\n   * Create a `FileDataSource`.\n   *\n   * @param input Local file path, or `File`/`Blob`/`Uint8Array` object to\n   *     read. Local file only works in node environment.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected input: FileElement|string,\n      protected readonly options: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.input) && env().get('IS_NODE')) {\n      // tslint:disable-next-line:no-require-imports\n      const fs = require('fs');\n      this.input = fs.readFileSync((this.input as string).substr(7));\n    }\n    // TODO(kangyizhang): Add LocalFileChunkIterator to split local streaming\n    // with file in browser.\n    return new FileChunkIterator(this.input as FileElement, this.options);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {urlChunkIterator} from '../iterators/url_chunk_iterator';\nimport {isLocalPath} from '../util/source_util';\nimport {FileDataSource} from './file_data_source';\n\n/*\n * Represents a URL readable as a stream of binary data chunks.\n */\nexport class URLDataSource extends DataSource {\n  /**\n   * Create a `URLDataSource`.\n   *\n   * @param url A source URL string, or a `Request` object.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected readonly url: RequestInfo,\n      protected readonly fileOptions: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  // TODO(soergel): provide appropriate caching options.  Currently this\n  // will download the URL anew for each call to iterator().  Since we have\n  // to treat the downloaded file as a blob/buffer anyway, we may as well retain\n  // it-- but that raises GC issues.  Also we may want a persistent disk cache.\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.url)) {\n      return (new FileDataSource(this.url as string, this.fileOptions))\n          .iterator();\n    } else {\n      return urlChunkIterator(this.url, this.fileOptions);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer} from '@tensorflow/tfjs-core';\nimport {Dataset, datasetFromIteratorFn} from './dataset';\nimport {CSVDataset} from './datasets/csv_dataset';\nimport {iteratorFromFunction} from './iterators/lazy_iterator';\nimport {MicrophoneIterator} from './iterators/microphone_iterator';\nimport {WebcamIterator} from './iterators/webcam_iterator';\nimport {URLDataSource} from './sources/url_data_source';\nimport {CSVConfig, MicrophoneConfig, WebcamConfig} from './types';\n\n/**\n * Create a `CSVDataset` by reading and decoding CSV file(s) from provided URL\n * or local path if it's in Node environment.\n *\n * Note: If isLabel in columnConfigs is `true` for at least one column, the\n * element in returned `CSVDataset` will be an object of\n * `{xs:features, ys:labels}`: xs is a dict of features key/value pairs, ys\n * is a dict of labels key/value pairs. If no column is marked as label,\n * returns a dict of features only.\n *\n * ```js\n * const csvUrl =\n * 'https://storage.googleapis.com/tfjs-examples/multivariate-linear-regression/data/boston-housing-train.csv';\n *\n * async function run() {\n *   // We want to predict the column \"medv\", which represents a median value of\n *   // a home (in $1000s), so we mark it as a label.\n *   const csvDataset = tf.data.csv(\n *     csvUrl, {\n *       columnConfigs: {\n *         medv: {\n *           isLabel: true\n *         }\n *       }\n *     });\n *\n *   // Number of features is the number of column names minus one for the label\n *   // column.\n *   const numOfFeatures = (await csvDataset.columnNames()).length - 1;\n *\n *   // Prepare the Dataset for training.\n *   const flattenedDataset =\n *     csvDataset\n *     .map(({xs, ys}) =>\n *       {\n *         // Convert xs(features) and ys(labels) from object form (keyed by\n *         // column name) to array form.\n *         return {xs:Object.values(xs), ys:Object.values(ys)};\n *       })\n *     .batch(10);\n *\n *   // Define the model.\n *   const model = tf.sequential();\n *   model.add(tf.layers.dense({\n *     inputShape: [numOfFeatures],\n *     units: 1\n *   }));\n *   model.compile({\n *     optimizer: tf.train.sgd(0.000001),\n *     loss: 'meanSquaredError'\n *   });\n *\n *   // Fit the model using the prepared Dataset\n *   return model.fitDataset(flattenedDataset, {\n *     epochs: 10,\n *     callbacks: {\n *       onEpochEnd: async (epoch, logs) => {\n *         console.log(epoch + ':' + logs.loss);\n *       }\n *     }\n *   });\n * }\n *\n * await run();\n * ```\n *\n * @param source URL or local path to get CSV file. If it's a local path, it\n * must have prefix `file://` and it only works in node environment.\n * @param csvConfig (Optional) A CSVConfig object that contains configurations\n *     of reading and decoding from CSV file(s).\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function csv(\n    source: RequestInfo, csvConfig: CSVConfig = {}): CSVDataset {\n  return new CSVDataset(new URLDataSource(source), csvConfig);\n}\n\n/**\n * Create a `Dataset` that produces each element by calling a provided function.\n *\n * Note that repeated iterations over this `Dataset` may produce different\n * results, because the function will be called anew for each element of each\n * iteration.\n *\n * Also, beware that the sequence of calls to this function may be out of order\n * in time with respect to the logical order of the Dataset. This is due to the\n * asynchronous lazy nature of stream processing, and depends on downstream\n * transformations (e.g. .shuffle()). If the provided function is pure, this is\n * no problem, but if it is a closure over a mutable state (e.g., a traversal\n * pointer), then the order of the produced elements may be scrambled.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const ds = tf.data.func(func);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param f A function that produces one data element on each call.\n */\nexport function func<T extends TensorContainer>(\n    f: () => IteratorResult<T>| Promise<IteratorResult<T>>): Dataset<T> {\n  const iter = iteratorFromFunction(f);\n  return datasetFromIteratorFn(async () => iter);\n}\n\n/**\n * Create a `Dataset` that produces each element from provided JavaScript\n * generator, which is a function*\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions),\n * or a function that returns an\n * iterator\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions).\n *\n * The returned iterator should have `.next()` function that returns element in\n * format of `{value: TensorContainer, done:boolean}`.\n *\n * Example of creating a dataset from an iterator factory:\n * ```js\n * function makeIterator() {\n *   const numElements = 10;\n *   let index = 0;\n *\n *   const iterator = {\n *     next: () => {\n *       let result;\n *       if (index < numElements) {\n *         result = {value: index, done: false};\n *         index++;\n *         return result;\n *       }\n *       return {value: index, done: true};\n *     }\n *   };\n *   return iterator;\n * }\n * const ds = tf.data.generator(makeIterator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * Example of creating a dataset from a generator:\n * ```js\n * function* dataGenerator() {\n *   const numElements = 10;\n *   let index = 0;\n *   while (index < numElements) {\n *     const x = index;\n *     index++;\n *     yield x;\n *   }\n * }\n *\n * const ds = tf.data.generator(dataGenerator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param generator A Javascript generator function that returns a JavaScript\n *     iterator.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function generator<T extends TensorContainer>(\n    generator: () => Iterator<T>| Promise<Iterator<T>>): Dataset<T> {\n  return datasetFromIteratorFn(async () => {\n    const gen = await generator();\n    return iteratorFromFunction(() => gen.next());\n  });\n}\n\n/**\n * Create an iterator that generate `Tensor`s from webcam video stream. This API\n * only works in Browser environment when the device has webcam.\n *\n * Note: this code snippet only works when the device has a webcam. It will\n * request permission to open the webcam when running.\n * ```js\n * const videoElement = document.createElement('video');\n * videoElement.width = 100;\n * videoElement.height = 100;\n * const cam = await tf.data.webcam(videoElement);\n * const img = await cam.capture();\n * img.print();\n * cam.stop();\n * ```\n *\n * @param webcamVideoElement A `HTMLVideoElement` used to play video from\n *     webcam. If this element is not provided, a hidden `HTMLVideoElement` will\n *     be created. In that case, `resizeWidth` and `resizeHeight` must be\n *     provided to set the generated tensor shape.\n * @param webcamConfig A `WebcamConfig` object that contains configurations of\n *     reading and manipulating data from webcam video stream.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function webcam(\n    webcamVideoElement?: HTMLVideoElement,\n    webcamConfig?: WebcamConfig): Promise<WebcamIterator> {\n  return WebcamIterator.create(webcamVideoElement, webcamConfig);\n}\n\n/**\n * Create an iterator that generate frequency-domain spectrogram `Tensor`s from\n * microphone audio stream with browser's native FFT. This API only works in\n * browser environment when the device has microphone.\n *\n * Note: this code snippet only works when the device has a microphone. It will\n * request permission to open the microphone when running.\n * ```js\n * const mic = await tf.data.microphone({\n *   fftSize: 1024,\n *   columnTruncateLength: 232,\n *   numFramesPerSpectrogram: 43,\n *   sampleRateHz:44100,\n *   includeSpectrogram: true,\n *   includeWaveform: true\n * });\n * const audioData = await mic.capture();\n * const spectrogramTensor = audioData.spectrogram;\n * spectrogramTensor.print();\n * const waveformTensor = audioData.waveform;\n * waveformTensor.print();\n * mic.stop();\n * ```\n *\n * @param microphoneConfig A `MicrophoneConfig` object that contains\n *     configurations of reading audio data from microphone.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function microphone(microphoneConfig?: MicrophoneConfig):\n    Promise<MicrophoneIterator> {\n  return MicrophoneIterator.create(microphoneConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${\n              opName} does not support complex64 tensors in the CPU backend.`);\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {backend_util, BackendTimingInfo, DataStorage, DataType, DataValues, engine, env, kernel_impls, KernelBackend, max, NumericDataType, Rank, Scalar, ShapeMap, slice_util, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, TensorInfo, TypedArray, upcastType, util} from '@tensorflow/tfjs-core';\n\nconst nonMaxSuppressionV3Impl = kernel_impls.nonMaxSuppressionV3Impl;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\nimport * as seedrandom from 'seedrandom';\nimport {assertNotComplex} from './cpu_util';\n\ninterface DataId {}\n\nfunction mapActivation(\n    backend: MathBackendCPU, x: Tensor, activation: backend_util.Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return backend.linear(x);\n  } else if (activation === 'relu') {\n    return backend.relu(x);\n  } else if (activation === 'elu') {\n    return tf.elu(x);\n  } else if (activation === 'relu6') {\n    return backend.relu6(x);\n  } else if (activation === 'prelu') {\n    return backend.prelu(x, preluActivationWeights);\n  }\n  throw new Error(\n      `Activation ${activation} has not been implemented for the CPU backend.`);\n}\n\nexport interface TensorData<D extends DataType> {\n  values?: backend_util.BackendValues;\n  dtype: D;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensorInfos field.\n  complexTensorInfos?: {real: TensorInfo, imag: TensorInfo};\n  // refCount keeps track of how many tensors reference it. Used for memory\n  // management.\n  refCount: number;\n}\n\nexport class MathBackendCPU extends KernelBackend {\n  public blockSize = 48;\n\n  data: DataStorage<TensorData<DataType>>;\n  private firstUse = true;\n\n  constructor() {\n    super();\n    this.data = new DataStorage(this, engine());\n  }\n\n  write(values: backend_util.BackendValues, shape: number[], dtype: DataType):\n      DataId {\n    if (this.firstUse) {\n      this.firstUse = false;\n      if (env().get('IS_NODE')) {\n        backend_util.warn(\n            '\\n============================\\n' +\n            'Hi there \uD83D\uDC4B. Looks like you are running TensorFlow.js in ' +\n            'Node.js. To speed things up dramatically, install our node ' +\n            'backend, which binds to TensorFlow C++, by running ' +\n            'npm i @tensorflow/tfjs-node, ' +\n            'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +\n            'Then call require(\\'@tensorflow/tfjs-node\\'); (-gpu ' +\n            'suffix for CUDA) at the start of your program. ' +\n            'Visit https://github.com/tensorflow/tfjs-node for more details.' +\n            '\\n============================');\n      }\n    }\n    const dataId = {};\n\n    this.data.set(dataId, {values, dtype, refCount: 1});\n\n    return dataId;\n  }\n\n  /**\n   * Create a data bucket in cpu backend.\n   * @param shape Shape of the `TensorInfo`.\n   * @param dtype DType of the `TensorInfo`.\n   * @param values The value of the `TensorInfo` stored as a flattened array.\n   */\n  makeTensorInfo(\n      shape: number[], dtype: DataType,\n      values?: backend_util.BackendValues): TensorInfo {\n    const outId = this.write(values, shape, dtype);\n\n    return {dataId: outId, shape, dtype};\n  }\n\n  /** Increase refCount of a `TensorData`. */\n  incRef(dataId: DataId): void {\n    const tensorData = this.data.get(dataId);\n    tensorData.refCount++;\n  }\n\n  /** Decrease refCount of a `TensorData`. */\n  decRef(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n      tensorData.refCount--;\n    }\n  }\n\n  move(\n      dataId: DataId, values: backend_util.BackendValues, shape: number[],\n      dtype: DataType): void {\n    this.data.set(dataId, {values, dtype, refCount: 1});\n  }\n\n  numDataIds(): number {\n    return this.data.numDataIds();\n  }\n\n  async read(dataId: DataId): Promise<backend_util.BackendValues> {\n    return this.readSync(dataId);\n  }\n  readSync(dataId: DataId): backend_util.BackendValues {\n    const {dtype, complexTensorInfos} = this.data.get(dataId);\n\n    if (dtype === 'complex64') {\n      const realValues =\n          this.readSync(complexTensorInfos.real.dataId) as Float32Array;\n      const imagValues =\n          this.readSync(complexTensorInfos.imag.dataId) as Float32Array;\n      return backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    }\n\n    return this.data.get(dataId).values;\n  }\n\n  private bufferSync<R extends Rank>(t: Tensor<R>): TensorBuffer<R> {\n    const data = this.readSync(t.dataId);\n    let decodedData = data as DataValues;\n    if (t.dtype === 'string') {\n      try {\n        // Decode the bytes into string.\n        decodedData = (data as Uint8Array[]).map(d => util.decodeString(d));\n      } catch {\n        throw new Error('Failed to decode encoded string bytes into utf-8');\n      }\n    }\n    return tf.buffer(t.shape, t.dtype, decodedData) as TensorBuffer<R>;\n  }\n\n  makeOutput<T extends Tensor>(\n      values: backend_util.BackendValues, shape: number[], dtype: DataType): T {\n    const dataId = this.write(values, shape, dtype);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  disposeData(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const {complexTensorInfos} = this.data.get(dataId);\n\n      if (complexTensorInfos != null) {\n        this.disposeData(complexTensorInfos.real.dataId);\n        this.disposeData(complexTensorInfos.imag.dataId);\n      }\n\n      this.data.delete(dataId);\n    }\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n\n      tensorData.refCount--;\n\n      if (tensorData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  async time(f: () => void): Promise<BackendTimingInfo> {\n    const start = util.now();\n    f();\n    const kernelMs = util.now() - start;\n    return {kernelMs};\n  }\n\n  memory() {\n    return {\n      // Unreliable due to automatic gc. The numbers above are cumulative.\n      unreliable: true,\n      reasons:\n          ['The reported memory is an upper bound. Due to automatic garbage ' +\n           'collection, the true allocated memory may be less.']\n    };\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    assertNotComplex(x, 'stridedSlice');\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tf.tensor([], outShape) as T;\n    }\n\n    const buffer = tf.buffer(outShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n    for (let i = 0; i < buffer.size; i++) {\n      const loc = buffer.indexToLoc(i);\n\n      const newLoc: number[] = new Array(loc.length);\n      for (let j = 0; j < newLoc.length; j++) {\n        newLoc[j] = loc[j] * strides[j] + begin[j];\n      }\n      buffer.set(xBuf.get(...newLoc), ...loc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  diag(x: Tensor): Tensor {\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const buffer = tf.buffer([x.size, x.size], x.dtype);\n    const vals = buffer.values;\n    for (let i = 0; i < xVals.length; i++) {\n      vals[i * x.size + i] = xVals[i];\n    }\n    return buffer.toTensor();\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = tf.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    assertNotComplex(x, 'reverse');\n\n    const buffer = tf.buffer(x.shape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < buffer.size; i++) {\n      const outLoc = buffer.indexToLoc(i);\n      const inLoc = outLoc.slice();\n      axis.forEach(ax => inLoc[ax] = x.shape[ax] - 1 - inLoc[ax]);\n      buffer.set(xBuf.get(...inLoc), ...outLoc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'neg');\n\n    // TODO(lina128): Use mul directly once neg is modularized.\n    return tf.mul(tf.scalar(-1), x);\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    assertNotComplex(tensors, 'addN');\n\n    const vals = tensors.map(t => this.readSync(t.dataId) as TypedArray);\n    const result = tf.buffer(tensors[0].shape, tensors[0].dtype as 'float32');\n    const resultVals = result.values;\n    for (let i = 0; i < tensors.length; i++) {\n      const currVals = vals[i];\n      for (let j = 0; j < resultVals.length; j++) {\n        resultVals[j] += currVals[j];\n      }\n    }\n    return result.toTensor() as T;\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n\n    // TODO(lina128): Use sub directly once softmax is modularized.\n    const a = tf.sub(logits, maxLogit.reshape(expandedShape));\n    const b = tf.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax\n    // kernel modularization.\n    return tf.div(b, sumExp);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    assertNotComplex([a, b], 'pow');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.pow(aValue, bValue)) as\n        T;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    assertNotComplex([a, b], 'matMul');\n\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const leftDim = transposeA ? a.shape[2] : a.shape[1];\n    const rightDim = transposeB ? b.shape[1] : b.shape[2];\n    const batchDim = a.shape[0];\n\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const [aBatch, aOuterStep, aInnerStep] = transposeA ?\n        [a.strides[0], 1, a.strides[1]] :\n        [a.strides[0], a.strides[1], 1];\n    const [bInnerStep, bOuterStep, bBatch] = transposeB ?\n        [1, b.strides[1], b.strides[0]] :\n        [b.strides[1], 1, b.strides[0]];\n\n    const size = leftDim * rightDim;\n    const result = tf.buffer([batchDim, leftDim, rightDim], a.dtype);\n    const resVals = result.values as TypedArray;\n    const blockSize = this.blockSize;\n\n    for (let b = 0; b < batchDim; b++) {\n      for (let i0 = 0; i0 < leftDim; i0 += blockSize) {\n        for (let j0 = 0; j0 < rightDim; j0 += blockSize) {\n          for (let k0 = 0; k0 < sharedDim; k0 += blockSize) {\n            // for when blockSize doesn't evenly divide the input\n            const iBlock = Math.min(i0 + blockSize, leftDim);\n            const jBlock = Math.min(j0 + blockSize, rightDim);\n            const kBlock = Math.min(k0 + blockSize, sharedDim);\n\n            for (let i = i0; i < iBlock; i++) {\n              for (let j = j0; j < jBlock; j++) {\n                let sum = 0.0;\n\n                for (let k = k0; k < kBlock; k++) {\n                  sum += aValues[b * aBatch + i * aOuterStep + k * aInnerStep] *\n                      bValues[k * bInnerStep + j * bOuterStep + b * bBatch];\n                }\n                resVals[b * size + (i * rightDim + j)] += sum;\n              }\n            }\n          }\n        }\n      }\n    }\n    return result.toTensor() as Tensor3D;\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    let result = this.batchMatMul(a, b, transposeA, transposeB);\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedBatchMatMul is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor3D;\n    }\n\n    return result;\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'floorDiv');\n\n    const op = (a: number, b: number) => Math.floor(a / b);\n    const outputDtype = 'int32';\n    return this.broadcastedBinaryOp(a, b, outputDtype, op);\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let sum = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        sum += aVals[offset + j];\n      }\n      vals[i] = sum;\n    }\n    return result;\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let prod = 1;\n      for (let j = 0; j < reduceSize; ++j) {\n        prod *= aVals[offset + j];\n      }\n      vals[i] = prod;\n    }\n    return result;\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    assertNotComplex(x, 'unsortedSegmentSum');\n\n    const res = [];\n\n    // Reshape the segment id's so that they can be broadcast with\n    // x. The new shape should be [segmentIds.shape, 1, ..., 1]\n    const numIters = x.rank - segmentIds.rank;\n    for (let i = 0; i < numIters; ++i) {\n      segmentIds = segmentIds.expandDims(i + 1);\n    }\n\n    for (let i = 0; i < numSegments; ++i) {\n      const segmentId = tf.scalar(i, 'int32');\n      const mask = tf.equal(segmentId, segmentIds).asType('float32');\n      const sum = mask.mul(x).sum(0);\n      res.push(sum);\n    }\n\n    return tf.stack(res);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMin');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMin', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      let minIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n          minIndex = j;\n        }\n      }\n      vals[i] = minIndex;\n    }\n    return result;\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMax');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMax', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let max = aVals[offset];\n      let maxIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value > max) {\n          max = value;\n          maxIndex = j;\n        }\n      }\n      vals[i] = maxIndex;\n    }\n    return result;\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    assertNotComplex(x, 'cumsum');\n\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `backend.cumsum in CPU expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(x.shape, resultDtype);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    const finalDim = x.shape[x.rank - 1];\n    const indexAdjuster = reverse ?\n        (i: number, j: number) => i + finalDim - j - 1 :\n        (i: number, j: number) => i + j;\n    for (let i = 0; i < aVals.length; i += finalDim) {\n      for (let j = 0; j < finalDim; j++) {\n        const idx = indexAdjuster(i, j);\n        if (j === 0) {\n          vals[idx] = exclusive ? 0 : aVals[idx];\n        } else {\n          const prevIdx = indexAdjuster(i, j - 1);\n          vals[idx] = exclusive ? aVals[prevIdx] + vals[prevIdx] :\n                                  aVals[idx] + vals[prevIdx];\n        }\n      }\n    }\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'equal');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal === bVal) ? 1 : 0;\n    });\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'notEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal !== bVal) ? 1 : 0;\n    });\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'less');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal < bVal) ? 1 : 0;\n    });\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'lessEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal <= bVal) ? 1 : 0;\n    });\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greater');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal > bVal) ? 1 : 0;\n    });\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greaterEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal >= bVal) ? 1 : 0;\n    });\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalAnd');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal && bVal;\n    });\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalOr');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal || bVal;\n    });\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([condition, a, b], 'select');\n\n    const values = this.readSync(condition.dataId) as TypedArray;\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const result = tf.zeros(a.shape, upcastType(a.dtype, b.dtype));\n    const newValues = this.readSync(result.dataId) as TypedArray;\n    let index = 0;\n    const offset = condition.rank === 0 || condition.rank > 1 || a.rank === 1 ?\n        1 :\n        util.sizeFromShape(a.shape.slice(1));\n\n    for (let i = 0; i < values.length; i++) {\n      for (let j = 0; j < offset; j++) {\n        if (values[i] === 1) {\n          newValues[index++] = aValues[i];\n        } else {\n          newValues[index++] = bValues[i];\n        }\n      }\n    }\n\n    return result;\n  }\n\n  where(condition: Tensor): Tensor2D {\n    assertNotComplex([condition], 'where');\n\n    const condVals = this.readSync(condition.dataId) as TypedArray;\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    assertNotComplex(x, 'topk');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'min');\n\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n        }\n      }\n      vals[i] = min;\n    }\n    return result;\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'minimum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.min(aVal, bVal));\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'mod');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const rem = aVal % bVal;\n      if ((aVal < 0 && bVal < 0) || (aVal >= 0 && bVal >= 0)) {\n        return rem;\n      } else {\n        return (rem + bVal) % bVal;\n      }\n    });\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'maximum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.max(aVal, bVal));\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'all');\n\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let all = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        all = all && value;\n      }\n      vals[i] = all;\n    }\n    return result;\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'any');\n\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let anyVal = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        anyVal = anyVal || value;\n      }\n      vals[i] = anyVal;\n    }\n    return result;\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'squaredDifference');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const diff = aVal - bVal;\n      return diff * diff;\n    });\n  }\n\n  linear<T extends Tensor>(x: T): T {\n    return x;\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.max(0, inVals[i]);\n    }\n    return res as T;\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.min(Math.max(0, inVals[i]), 6);\n    }\n    return res as T;\n  }\n\n  prelu<T extends Tensor>(x: T, a: T): T {\n    assertNotComplex([x, a], 'prelu');\n\n    return this.broadcastedBinaryOp(\n               x, a, x.dtype,\n               (xValue, aValue) => xValue < 0 ? aValue * xValue : xValue) as T;\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    assertNotComplex([dy, y], 'eluDer');\n\n    const resultValues = new Float32Array(y.size);\n    const values = this.readSync(y.dataId) as TypedArray;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    for (let i = 0; i < values.length; ++i) {\n      const v = values[i];\n      if (v >= 1) {\n        resultValues[i] = dyValues[i];\n      } else {\n        resultValues[i] = dyValues[i] * (v + 1);\n      }\n    }\n    return this.makeOutput(resultValues, y.shape, 'float32');\n  }\n\n  atan2<T extends Tensor>(a: T, b: T): T {\n    assertNotComplex([a, b], 'atan2');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.atan2(aValue, bValue)) as\n        T;\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.conv2d(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedConv2d is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, filter], 'conv2d');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n\n    const xBatchStride = x.strides[0];\n    const xRowStride = isChannelsLast ? x.strides[1] : x.strides[2];\n    const xColStride = isChannelsLast ? x.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : x.strides[1];\n    const yBatchStride = y.strides[0];\n    const yRowStride = isChannelsLast ? y.strides[1] : y.strides[2];\n    const yColStride = isChannelsLast ? y.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : y.strides[1];\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * xBatchStride;\n      const yOffset1 = b * yBatchStride;\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * yRowStride;\n        const xRCorner = yR * convInfo.strideHeight - padTop;\n        for (let wR = 0; wR < filterHeight; wR++) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * xRowStride;\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * yColStride;\n            const xCCorner = yC * convInfo.strideWidth - padLeft;\n            for (let wC = 0; wC < filterWidth; wC++) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * xColStride;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1 * xChannelStride];\n                for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                  yVals[yOffset3 + d2 * yChannelStride] +=\n                      xVal * wVals[wOffset3 + d2];\n                }\n                wOffset3 += convInfo.outChannels;\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor() as Tensor4D;\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padFront = convInfo.padInfo.front;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const y = tf.buffer<Rank.R5>(convInfo.outShape, x.dtype as 'float32');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yF = 0; yF < convInfo.outDepth; ++yF) {\n        const yOffset2 = yOffset1 + yF * y.strides[1];\n        const xFCorner = yF * convInfo.strideDepth - padFront;\n        for (let wF = 0; wF < filterDepth; wF++) {\n          const xF = xFCorner + wF * dilationDepth;\n          if (xF < 0 || xF >= convInfo.inDepth) {\n            continue;\n          }\n          const wOffset1 = wF * filter.strides[0];\n          const xOffset2 = xOffset1 + xF * x.strides[1];\n\n          for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n            const yOffset3 = yOffset2 + yR * y.strides[2];\n            const xRCorner = yR * convInfo.strideHeight - padTop;\n            for (let wR = 0; wR < filterHeight; wR++) {\n              const xR = xRCorner + wR * dilationHeight;\n              if (xR < 0 || xR >= convInfo.inHeight) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wR * filter.strides[1];\n              const xOffset3 = xOffset2 + xR * x.strides[2];\n              for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n                const yOffset4 = yOffset3 + yC * convInfo.outChannels;\n                const xCCorner = yC * convInfo.strideWidth - padLeft;\n                for (let wC = 0; wC < filterWidth; wC++) {\n                  const xC = xCCorner + wC * dilationWidth;\n                  if (xC < 0 || xC >= convInfo.inWidth) {\n                    continue;\n                  }\n                  const wOffset3 = wOffset2 + wC * filter.strides[2];\n                  const xOffset4 = xOffset3 + xC * convInfo.inChannels;\n                  let wOffset4 = wOffset3;\n                  for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                    const xVal = xVals[xOffset4 + d1];\n                    for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                      yVals[yOffset4 + d2] += xVal * wVals[wOffset4 + d2];\n                    }\n                    wOffset4 += convInfo.outChannels;\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor();\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'conv2dDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth,\n      dataFormat\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const xBatchStride = dx.strides[0];\n    const xRowStride = isChannelsLast ? dx.strides[1] : dx.strides[2];\n    const xColStride = isChannelsLast ? dx.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : dx.strides[1];\n    const yBatchStride = dy.strides[0];\n    const yRowStride = isChannelsLast ? dy.strides[1] : dy.strides[2];\n    const yColStride = isChannelsLast ? dy.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : dy.strides[1];\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset =\n                    yBatchStride * b + yRowStride * yR + yColStride * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let d2 = 0; d2 < outChannels; ++d2) {\n                  const pixel = dyValues[dyOffset + yChannelStride * d2];\n                  const weight = fltValues[fltOffset + d2];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            const dxOffset = xBatchStride * b + xRowStride * xR +\n                xColStride * xC + xChannelStride * d1;\n            dxValues[dxOffset] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const dx = tf.buffer<Rank.R5>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2, dxS3] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2, fltS3] = filter.strides;\n    const {\n      batchSize,\n      filterDepth,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inDepth,\n      inHeight,\n      inWidth,\n      outChannels,\n      outDepth,\n      outHeight,\n      outWidth,\n      strideDepth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const frontPad = filterDepth - 1 - convInfo.padInfo.front;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        // Frames of depth\n        for (let xF = 0; xF < inDepth; ++xF) {\n          const xFCorner = xF - frontPad;\n          const xFMin = Math.max(0, Math.ceil(xFCorner / strideDepth));\n          const yFMax =\n              Math.min(outDepth, (filterDepth + xFCorner) / strideDepth);\n\n          // Rows as per standard 2d matrix notation\n          for (let xR = 0; xR < inHeight; ++xR) {\n            const xRCorner = xR - topPad;\n            const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n            const yRMax =\n                Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n            // Columns as per standard 2d matrix notation\n            for (let xC = 0; xC < inWidth; ++xC) {\n              const xCCorner = xC - leftPad;\n              const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n              const yCMax =\n                  Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n              let dotProd = 0;\n              for (let yF = xFMin; yF < yFMax; ++yF) {\n                const wF = yF * strideDepth - xFCorner;\n\n                for (let yR = xRMin; yR < yRMax; ++yR) {\n                  const wR = yR * strideHeight - xRCorner;\n\n                  for (let yC = xCMin; yC < yCMax; ++yC) {\n                    const wC = yC * strideWidth - xCCorner;\n                    const dyOffset =\n                        dyS0 * b + dyS1 * yF + dyS2 * yR + dyS3 * yC;\n                    const fltOffset = fltS0 * (filterDepth - 1 - wF) +\n                        fltS1 * (filterHeight - 1 - wR) +\n                        fltS2 * (filterWidth - 1 - wC) + fltS3 * d1;\n\n                    for (let d2 = 0; d2 < outChannels; ++d2) {\n                      const pixel = dyValues[dyOffset + d2];\n                      const weight = fltValues[fltOffset + d2];\n                      dotProd += pixel * weight;\n                    }\n                  }\n                }\n              }\n              dxValues[dxS0 * b + dxS1 * xF + dxS2 * xR + dxS3 * xC + d1] =\n                  dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, dy], 'conv2dDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n          for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n            // Need to convolve.\n            let dotProd = 0;\n            for (let b = 0; b < convInfo.batchSize; ++b) {\n              for (let yR = yRMin; yR < yRMax; ++yR) {\n                const xR = wR + yR * strideHeight - topPad;\n                for (let yC = yCMin; yC < yCMax; ++yC) {\n                  const xC = wC + yC * strideWidth - leftPad;\n                  if (isChannelsLast) {\n                    dotProd +=\n                        xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n                  } else {\n                    dotProd +=\n                        xBuf.get(b, d1, xR, xC) * dyBuf.get(b, d2, yR, yC);\n                  }\n                }\n              }\n            }\n            dW.set(dotProd, wR, wC, d1, d2);\n          }\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const dw = tf.buffer<Rank.R5>(convInfo.filterShape, 'float32');\n    const dwValues = dw.values;\n    const [dwS0, dwS1, dwS2, dwS3] = dw.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const [xS0, xS1, xS2, xS3] = x.strides;\n\n    const frontPad = convInfo.padInfo.front;\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n\n    for (let wF = 0; wF < filterDepth; ++wF) {\n      const yFMin = Math.max(0, Math.ceil((frontPad - wF) / strideDepth));\n      const yFMax = Math.min(\n          convInfo.outDepth, (convInfo.inDepth + frontPad - wF) / strideDepth);\n      const wOffset1 = wF * dwS0;\n\n      for (let wR = 0; wR < filterHeight; ++wR) {\n        const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n        const yRMax = Math.min(\n            convInfo.outHeight,\n            (convInfo.inHeight + topPad - wR) / strideHeight);\n        const wOffset2 = wR * dwS1 + wOffset1;\n\n        for (let wC = 0; wC < filterWidth; ++wC) {\n          const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n          const yCMax = Math.min(\n              convInfo.outWidth,\n              (convInfo.inWidth + leftPad - wC) / strideWidth);\n          const wOffset3 = wC * dwS2 + wOffset2;\n\n          for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n            const wOffset4 = d1 * dwS3 + wOffset3;\n\n            for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n              let dotProd = 0;\n              for (let b = 0; b < convInfo.batchSize; ++b) {\n                const xOffset1 = b * xS0;\n                const yOffset1 = b * dyS0;\n\n                for (let yF = yFMin; yF < yFMax; ++yF) {\n                  const xF = wF + yF * strideDepth - frontPad;\n                  const xOffset2 = xF * xS1 + xOffset1;\n                  const yOffset2 = yF * dyS1 + yOffset1;\n\n                  for (let yR = yRMin; yR < yRMax; ++yR) {\n                    const xR = wR + yR * strideHeight - topPad;\n                    const xOffset3 = xR * xS2 + xOffset2;\n                    const yOffset3 = yR * dyS2 + yOffset2;\n\n                    for (let yC = yCMin; yC < yCMax; ++yC) {\n                      const xC = wC + yC * strideWidth - leftPad;\n                      const xOffset4 = xC * xS3 + xOffset3;\n                      const yOffset4 = yC * dyS3 + yOffset3;\n\n                      dotProd +=\n                          xValues[xOffset4 + d1] * dyValues[yOffset4 + d2];\n                    }\n                  }\n                }\n              }\n              dwValues[wOffset4 + d2] = dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dw.toTensor();\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.depthwiseConv2D(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedDepthwiseConv2D is\n      // modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, filter], 'depthwiseConv2D');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * y.strides[1];\n        const xRCorner = yR * convInfo.strideHeight - padLeft;\n        for (let wR = 0; wR < filterHeight; ++wR) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * x.strides[1];\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * y.strides[2];\n            const xCCorner = yC * convInfo.strideWidth - padTop;\n            for (let wC = 0; wC < filterWidth; ++wC) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * convInfo.inChannels;\n              let yOffset4 = yOffset3;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1];\n                for (let q = 0; q < chMul; ++q) {\n                  yVals[yOffset4 + q] += xVal * wVals[wOffset3 + q];\n                }\n                yOffset4 += chMul;\n                wOffset3 += chMul;\n              }\n            }\n          }\n        }\n      }\n    }\n\n    return y.toTensor() as Tensor4D;\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'depthwiseConv2DDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n    const chMul = outChannels / inChannels;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset = dyS0 * b + dyS1 * yR + dyS2 * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let dm = 0; dm < chMul; ++dm) {\n                  const d2 = d1 * chMul + dm;\n                  const pixel = dyValues[dyOffset + d2];\n                  const weight = fltValues[fltOffset + dm];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            dxValues[dxS0 * b + dxS1 * xR + dxS2 * xC + d1] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, dy], 'depthwiseConv2DDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n          const d1 = Math.trunc(d2 / chMul);\n          const dm = d2 % chMul;\n\n          let dotProd = 0;\n          for (let b = 0; b < convInfo.batchSize; ++b) {\n            for (let yR = yRMin; yR < yRMax; ++yR) {\n              const xR = wR + yR * strideHeight - topPad;\n              for (let yC = yCMin; yC < yCMax; ++yC) {\n                const xC = wC + yC * strideWidth - leftPad;\n                dotProd += xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n              }\n            }\n          }\n          dW.set(dotProd, wR, wC, d1, dm);\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    assertNotComplex(x, 'tile');\n    return tile(this.bufferSync(x), reps) as T;\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    assertNotComplex([x, indices], 'gather');\n\n    const newShape: number[] = x.shape.slice();\n    const indicesValues = this.readSync(indices.dataId) as TypedArray;\n    newShape[axis] = indicesValues.length;\n    const result = tf.buffer(newShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < result.size; ++i) {\n      const newLoc = result.indexToLoc(i);\n\n      const originalLoc: number[] = newLoc.slice();\n      originalLoc[axis] = indicesValues[newLoc[axis]];\n\n      const originalIndex = xBuf.locToIndex(originalLoc);\n      result.values[i] = xBuf.values[originalIndex];\n    }\n    return result.toTensor() as T;\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    assertNotComplex([x], 'batchToSpaceND');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return tf.transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  private pool3d(\n      x: Tensor5D, convInfo: backend_util.Conv3DInfo,\n      poolType: 'max'|'avg'): Tensor5D {\n    assertNotComplex(x, 'pool3d');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const initialValue =\n        (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                              Number.POSITIVE_INFINITY);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = tf.buffer(convInfo.outShape, x.dtype);\n    const outputVals = output.values;\n\n    const outputBatchStrides = convInfo.outShape[1] * convInfo.outShape[2] *\n        convInfo.outShape[3] * convInfo.outShape[4];\n    const outputDepthStrides =\n        convInfo.outShape[2] * convInfo.outShape[3] * convInfo.outShape[4];\n    const outputRowStrides = convInfo.outShape[3] * convInfo.outShape[4];\n    const outputColStrides = convInfo.outShape[4];\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      const outputBatchOffset = batch * outputBatchStrides;\n      const inputBatchOffset = batch * x.strides[0];\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          const outputDepthOffset =\n              outputBatchOffset + yDepth * outputDepthStrides;\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            const outputRowOffset = outputDepthOffset + yRow * outputRowStrides;\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n              // Shader code begins\n              const outputColOffset = outputRowOffset + yCol * outputColStrides;\n              let minMaxValue = initialValue;\n              let avgValue = 0;\n              let count = 0;\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const xDepthOffset = inputBatchOffset + xDepth * x.strides[1];\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const xRowOffset = xDepthOffset + xRow * x.strides[2];\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const xColOffset = xRowOffset + xCol * x.strides[3];\n                    const pixel = xValues[xColOffset + channel];\n                    if ((poolType === 'max' && pixel > minMaxValue)) {\n                      minMaxValue = pixel;\n                    } else if (poolType === 'avg') {\n                      avgValue += pixel;\n                      count++;\n                    }\n                    if (isNaN(minMaxValue)) {\n                      break;\n                    }\n                  }\n                  if (isNaN(minMaxValue)) {\n                    break;\n                  }\n                }\n                if (isNaN(minMaxValue)) {\n                  break;\n                }\n              }\n              const outputOffset = outputColOffset + channel;\n              outputVals[outputOffset] =\n                  poolType === 'avg' ? avgValue / count : minMaxValue;\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor5D;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'avgPool3d');\n\n    return this.pool3d(x, convInfo, 'avg').toFloat();\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([dy, x], 'avgPool3dBackprop');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins.\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel;\n                  }\n                }\n              }\n              dx.set(\n                  dotProd * avgMultiplier, batch, dxDepth, dxRow, dxCol,\n                  channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'maxPool3d');\n\n    return this.pool3d(x, convInfo, 'max').toFloat();\n  }\n\n  private maxPool3dPositions(x: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const maxPositions = tf.buffer(convInfo.outShape, 'int32');\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const xBuf = this.bufferSync(x);\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n\n              // Shader code begins\n              let maxValue = Number.NEGATIVE_INFINITY;\n              let maxPosition = -1;\n\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const wDepth = xDepth - xDepthCorner;\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const wRow = xRow - xRowCorner;\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const wCol = xCol - xColCorner;\n                    const pixel = xBuf.get(batch, xDepth, xRow, xCol, channel);\n                    if (pixel >= maxValue) {\n                      maxValue = pixel;\n                      maxPosition = wDepth * effectiveFilterHeight *\n                              effectiveFilterWidth +\n                          wRow * effectiveFilterHeight + wCol;\n                    }\n                  }\n                }\n              }\n\n              maxPositions.set(maxPosition, batch, yDepth, yRow, yCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return maxPositions.toTensor() as Tensor5D;\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([x, y], 'maxPool3dBackprop');\n\n    const maxPositions = this.maxPool3dPositions(x, convInfo);\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const maxPosBuf = this.bufferSync(maxPositions);\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const maxPos = effectiveFilterDepth *\n                            effectiveFilterHeight * effectiveFilterWidth -\n                        1 -\n                        maxPosBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    const curPos =\n                        wDepth * effectiveFilterHeight * effectiveFilterWidth +\n                        wRow * effectiveFilterWidth + wCol;\n\n                    const mask = maxPos === curPos ? 1 : 0;\n                    if (mask === 0) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel * mask;\n                  }\n                }\n              }\n              dx.set(dotProd, batch, dxDepth, dxRow, dxCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeBilinear');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result = new Float32Array(\n        util.sizeFromShape([batch, newHeight, newWidth, numChannels]));\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n    let outputIdx = 0;\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n    for (let b = 0; b < batch; b++) {\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceRowFloor = Math.floor(sourceFracRow);\n        const rowFrac = sourceFracRow - sourceRowFloor;\n        const sourceRowCeil = Math.min(oldHeight - 1, Math.ceil(sourceFracRow));\n        const topRowOffset = b * x.strides[0] + sourceRowFloor * x.strides[1];\n        const botRowOffset = b * x.strides[0] + sourceRowCeil * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceColFloor = Math.floor(sourceFracCol);\n          const colFrac = sourceFracCol - sourceColFloor;\n          const sourceColCeil =\n              Math.min(oldWidth - 1, Math.ceil(sourceFracCol));\n          const topLeftOffest = topRowOffset + sourceColFloor * x.strides[2];\n          const botLeftOffset = botRowOffset + sourceColFloor * x.strides[2];\n          const topRightOffset = topRowOffset + sourceColCeil * x.strides[2];\n          const botRightOffest = botRowOffset + sourceColCeil * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n\n            // Compute the fractional index of the source.\n            const topLeft = xValues[topLeftOffest + d];\n            const bottomLeft = xValues[botLeftOffset + d];\n            const topRight = xValues[topRightOffset + d];\n            const bottomRight = xValues[botRightOffest + d];\n\n            const top = topLeft + (topRight - topLeft) * colFrac;\n            const bottom = bottomLeft + (bottomRight - bottomLeft) * colFrac;\n            const newValue = top + (bottom - top) * rowFrac;\n\n            result[outputIdx++] = newValue;\n          }\n        }\n      }\n    }\n    return tf.tensor(result, [batch, newHeight, newWidth, numChannels]);\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeBilinearBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass and add the\n    // corresponding coefficient from dy to the gradient (with some\n    // interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/3039375c86a5bbc9610c7725dcaa95d635f87ba2/tensorflow/core/kernels/resize_bilinear_op.cc#L275\n\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    let offset = 0;\n    for (let b = 0; b < batch; b++) {\n      const bOffset = b * x.strides[0];\n      for (let r = 0; r < yHeight; r++) {\n        const dxR = r * heightScale;\n        const topDxRIndex = Math.floor(dxR);\n        const bottomDxRIndex = Math.min(Math.ceil(dxR), xHeight - 1);\n\n        const topDxROffset = bOffset + topDxRIndex * x.strides[1];\n        const bottomDxROffset = bOffset + bottomDxRIndex * x.strides[1];\n\n        const dxRLerp = dxR - topDxRIndex;\n        const inverseDxRLerp = 1.0 - dxRLerp;\n        for (let c = 0; c < yWidth; c++) {\n          const dxC = c * widthScale;\n          const leftDxCIndex = Math.floor(dxC);\n          const rightDxCIndex = Math.min(Math.ceil(dxC), xWidth - 1);\n          const dxCLerp = dxC - leftDxCIndex;\n          const inverseDxCLerp = 1.0 - dxCLerp;\n\n          const topLeftRCOffset = topDxROffset + leftDxCIndex * x.strides[2];\n          const topRightRCOffset = topDxROffset + rightDxCIndex * x.strides[2];\n          const bottomLeftRCOffset =\n              bottomDxROffset + leftDxCIndex * x.strides[2];\n          const bottomRightRCOffset =\n              bottomDxROffset + rightDxCIndex * x.strides[2];\n\n          const inverseDxRLerpTimesInverseDxCLerp =\n              inverseDxRLerp * inverseDxCLerp;\n          const inverseDxRLerpTimesDxCLerp = inverseDxRLerp * dxCLerp;\n          const dxRLerpTimesInverseDxCLerp = dxRLerp * inverseDxCLerp;\n          const dxRLerpTimesDxCLerp = dxRLerp * dxCLerp;\n          for (let d = 0; d < depth; d++) {\n            const dyVal = dyValues[offset++];\n            output[topLeftRCOffset + d] +=\n                dyVal * inverseDxRLerpTimesInverseDxCLerp;\n            output[topRightRCOffset + d] += dyVal * inverseDxRLerpTimesDxCLerp;\n            output[bottomLeftRCOffset + d] +=\n                dyVal * dxRLerpTimesInverseDxCLerp;\n            output[bottomRightRCOffset + d] += dyVal * dxRLerpTimesDxCLerp;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, [batch, xWidth, xHeight, depth], x.dtype);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeNearestNeighbor');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = new Float32Array(batch * newHeight * newWidth * numChannels);\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n\n    let outputOffset = 0;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceNearestRow = Math.min(\n            oldHeight - 1,\n            alignCorners ? Math.round(sourceFracRow) :\n                           Math.floor(sourceFracRow));\n        const rowOffset = batchOffset + sourceNearestRow * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceNearestCol = Math.min(\n              oldWidth - 1,\n              alignCorners ? Math.round(sourceFracCol) :\n                             Math.floor(sourceFracCol));\n          const colOffset = rowOffset + sourceNearestCol * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n            // Compute the fractional index of the source.\n            const newVal = xValues[colOffset + d];\n            output[outputOffset++] = newVal;\n          }\n        }\n      }\n    }\n    return tf.tensor(\n        output, [batch, newHeight, newWidth, numChannels], x.dtype);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeNearestNeighborBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    // Loop over the output space.\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < xHeight; r++) {\n        const rowOffset = batchOffset + r * x.strides[1];\n\n        // Compute bounds for where in dy we will look\n        const startRLerp = Math.floor(r * invHeightScale);\n        const startDyR = Math.floor(startRLerp - (winHeight / 2));\n        for (let c = 0; c < xWidth; c++) {\n          const colOffset = rowOffset + c * x.strides[2];\n\n          // Compute bounds for where in dy we will look\n          const startCLerp = Math.floor(c * invWidthScale);\n          const startDyC = Math.floor(startCLerp - (winWidth / 2));\n\n          for (let d = 0; d < depth; d++) {\n            let accum = 0;\n            // loop over dy\n\n            for (let dyRIndex = 0; dyRIndex < winHeight; dyRIndex++) {\n              const dyR = dyRIndex + startDyR;\n              // Guard against the window exceeding the bounds of dy\n              if (dyR < 0 || dyR >= yHeight) {\n                continue;\n              }\n\n              const dyROffset = batchOffset + dyR * dy.strides[1];\n              const sourceFracRow = dyR * heightScale;\n              const sourceNearestRow = Math.min(\n                  xHeight - 1,\n                  alignCorners ? Math.round(sourceFracRow) :\n                                 Math.floor(sourceFracRow));\n              if (r !== sourceNearestRow) {\n                continue;\n              }\n              for (let dyCIndex = 0; dyCIndex < winWidth; dyCIndex++) {\n                const dyC = dyCIndex + startDyC;\n                // Guard against the window exceeding the bounds of dy\n                if (dyC < 0 || dyC >= yWidth) {\n                  continue;\n                }\n\n                const dyCOffset = dyROffset + dyC * dy.strides[2];\n                const sourceFracCol = dyC * widthScale;\n                const sourceNearestCol = Math.min(\n                    xWidth - 1,\n                    alignCorners ? Math.round(sourceFracCol) :\n                                   Math.floor(sourceFracCol));\n\n                if (c === sourceNearestCol) {\n                  accum += dyValues[dyCOffset + d];\n                }\n              }\n            }\n            output[colOffset + d] = accum;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, x.shape, x.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(x, 'localResponseNormalization4D');\n\n    const channels = x.shape[3];\n    const maxD = channels - 1;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const size = x.size;\n    const result = new Float32Array(size);\n\n    function sumAcrossChannels(offset: number) {\n      const currentChannel = offset % channels;\n      let beginSumOffset =\n          offset - currentChannel + Math.max(0, currentChannel - depthRadius);\n      const endSumOffset = offset - currentChannel +\n          Math.min(currentChannel + depthRadius, maxD);\n\n      let sum = 0.0;\n      for (; beginSumOffset <= endSumOffset; beginSumOffset++) {\n        const z = xValues[beginSumOffset];\n        sum += z * z;\n      }\n      return sum;\n    }\n\n    for (let offset = 0; offset < size; offset++) {\n      const sum = sumAcrossChannels(offset);\n      const val = xValues[offset] * Math.pow(bias + alpha * sum, -beta);\n      result[offset] = val;\n    }\n\n    return tf.tensor4d(result, x.shape);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(dy, 'LRNGrad');\n    const channels = dy.shape[3];\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const inputImageValues = this.readSync(inputImage.dataId) as TypedArray;\n    const outputImageValues = this.readSync(outputImage.dataId) as TypedArray;\n    const result = new Float32Array(dy.size);\n    const size = dy.size;\n\n    for (let offset = 0; offset < size; offset++) {\n      const currentChannel = offset % channels;\n      const depthBegin =\n          (offset - currentChannel) + Math.max(0, currentChannel - depthRadius);\n      const depthEnd = (offset - currentChannel) +\n          Math.min(channels, currentChannel + depthRadius + 1);\n\n      let norm = 0;\n      for (let k = depthBegin; k < depthEnd; k++) {\n        norm += Math.pow(inputImageValues[k], 2);\n      }\n      norm = alpha * norm + bias;\n\n      for (let k = depthBegin; k < depthEnd; k++) {\n        let dyi = -2 * alpha * beta * inputImageValues[k] *\n            outputImageValues[offset] / norm;\n        if (offset === k) {\n          dyi += Math.pow(norm, -beta);\n        }\n        dyi *= dyValues[offset];\n        result[k] += dyi;\n      }\n    }\n    return tf.tensor4d(result, dy.shape);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    assertNotComplex(logits, 'multinomial');\n\n    const probabilities = normalized ? logits : tf.softmax(logits);\n    const batchSize = probabilities.shape[0];\n    const numEvents = probabilities.shape[1];\n    const res = tf.zeros<Rank.R2>([batchSize, numSamples], 'int32');\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const probVals = this.readSync(probabilities.dataId) as TypedArray;\n\n    for (let b = 0; b < batchSize; ++b) {\n      const offset = b * numEvents;\n      // The cdf won't include the last event. It will be implicit if no other\n      // event happened.\n      const cdf = new Float32Array(numEvents - 1);\n      cdf[0] = probVals[offset];\n      for (let event = 1; event < cdf.length; ++event) {\n        cdf[event] = cdf[event - 1] + probVals[offset + event];\n      }\n\n      const random = seedrandom.alea(seed.toString());\n      const outOffset = b * numSamples;\n      for (let sampleId = 0; sampleId < numSamples; ++sampleId) {\n        const r = random();\n\n        // Assume last event happened by default.\n        resVals[outOffset + sampleId] = cdf.length;\n\n        for (let event = 0; event < cdf.length; event++) {\n          if (r < cdf[event]) {\n            resVals[outOffset + sampleId] = event;\n            break;\n          }\n        }\n      }\n    }\n    return res;\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    assertNotComplex(indices, 'oneHot');\n\n    const res = new Float32Array(indices.size * depth);\n    res.fill(offValue);\n    const indicesVal = this.readSync(indices.dataId) as TypedArray;\n\n    for (let event = 0; event < indices.size; ++event) {\n      if (indicesVal[event] >= 0 && indicesVal[event] < depth) {\n        res[event * depth + indicesVal[event]] = onValue;\n      }\n    }\n    return tf.tensor2d(res, [indices.size, depth], 'int32');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold: number): Tensor1D {\n    assertNotComplex(boxes, 'nonMaxSuppression');\n\n    const boxesVals = this.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = this.readSync(scores.dataId) as TypedArray;\n    return nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        dataFormat === 'NHWC',\n        () => `Only NHWC dataFormat supported on CPU for depthToSpace. Got ${\n            dataFormat}`);\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = x.shape[1];\n    const inputWidth = x.shape[2];\n    const inputDepth = x.shape[3];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result =\n        new Float32Array(batchSize * outputHeight * outputWidth * outputDepth);\n\n    let outputIdx = 0;\n    for (let b = 0; b < batchSize; ++b) {\n      for (let h = 0; h < outputHeight; ++h) {\n        const inH = Math.floor(h / blockSize);\n        const offsetH = (h % blockSize);\n        for (let w = 0; w < outputWidth; ++w) {\n          const inW = Math.floor(w / blockSize);\n          const offsetW = (w % blockSize);\n          const offsetD = (offsetH * blockSize + offsetW) * outputDepth;\n          for (let d = 0; d < outputDepth; ++d) {\n            const inD = d + offsetD;\n            const inputIdx =\n                inD + inputDepth * (inW + inputWidth * (inH + inputHeight * b));\n            result[outputIdx++] = xValues[inputIdx];\n          }\n        }\n      }\n    }\n    return tf.tensor4d(\n        result, [batchSize, outputHeight, outputWidth, outputDepth]);\n  }\n\n  private broadcastedBinaryOp(\n      a: Tensor, b: Tensor, dtype: DataType,\n      op: (a: number, b: number) => number): Tensor {\n    const newShape = backend_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const result = tf.buffer(newShape, dtype);\n    const aVals = this.readSync(a.dataId) as TypedArray;\n    const bVals = this.readSync(b.dataId) as TypedArray;\n    const aBroadcastDims = backend_util.getBroadcastDims(a.shape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(b.shape, newShape);\n\n    const resVals = result.values;\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resVals.length; ++i) {\n        resVals[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      const aBuf = this.bufferSync(a);\n      const bBuf = this.bufferSync(b);\n      for (let i = 0; i < resVals.length; ++i) {\n        const loc = result.indexToLoc(i);\n\n        const aLoc = loc.slice(-a.rank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = aBuf.locToIndex(aLoc);\n\n        const bLoc = loc.slice(-b.rank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = bBuf.locToIndex(bLoc);\n\n        resVals[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n    return result.toTensor();\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  dispose() {}\n\n  floatPrecision(): 16|32 {\n    return 32;\n  }\n\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return super.epsilon();\n  }\n\n  cropAndResize(\n      images: Tensor4D,\n      boxes: Tensor2D,\n      boxIndex: Tensor1D,\n      cropSize: [number, number],\n      method: string,\n      extrapolationValue: number,\n  ) {\n    const [batch, imageHeight, imageWidth, numChannels] = images.shape;\n    const numBoxes = boxes.shape[0];\n\n    const [cropHeight, cropWidth] = cropSize;\n    const output =\n        tf.buffer([numBoxes, cropHeight, cropWidth, numChannels], 'float32');\n\n    const boxVals = this.readSync(boxes.dataId) as TypedArray;\n    const boxIndVals = this.readSync(boxIndex.dataId) as TypedArray;\n    const imageVals = this.readSync(images.dataId) as TypedArray;\n\n    const inStride = images.strides;   // to calculate flat indexes into image\n    const outStride = output.strides;  // to calculate flat indexes into output\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op.cc\n    for (let b = 0; b < numBoxes; b++) {\n      const startInd = b * 4;\n      const y1 = boxVals[startInd];\n      const x1 = boxVals[startInd + 1];\n      const y2 = boxVals[startInd + 2];\n      const x2 = boxVals[startInd + 3];\n\n      const bInd: number = boxIndVals[b];\n      if (bInd >= batch) {\n        continue;\n      }\n\n      const heightScale = (cropHeight > 1) ?\n          (y2 - y1) * (imageHeight - 1) / (cropHeight - 1) :\n          0;\n      const widthScale =\n          (cropWidth > 1) ? (x2 - x1) * (imageWidth - 1) / (cropWidth - 1) : 0;\n\n      for (let y = 0; y < cropHeight; y++) {\n        const yInd: number = (cropHeight > 1) ?\n            y1 * (imageHeight - 1) + y * (heightScale) :\n            0.5 * (y1 + y2) * (imageHeight - 1);\n\n        if (yInd < 0 || yInd > imageHeight - 1) {\n          for (let x = 0; x < cropWidth; x++) {\n            for (let c = 0; c < numChannels; c++) {\n              const ind =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = extrapolationValue;\n            }\n          }\n          continue;\n        }\n\n        if (method === 'bilinear') {\n          const topInd = Math.floor(yInd);\n          const bottomInd = Math.ceil(yInd);\n          const yLerp = yInd - topInd;\n\n          for (let x = 0; x < cropWidth; x++) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const leftInd = Math.floor(xInd);\n            const rightInd = Math.ceil(xInd);\n            const xLerp = xInd - leftInd;\n\n            for (let c = 0; c < numChannels; c++) {\n              let ind = c + leftInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topRight = imageVals[ind];\n\n              ind = c + leftInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomRight = imageVals[ind];\n\n              const top = topLeft + (topRight - topLeft) * xLerp;\n              const bottom = bottomLeft + (bottomRight - bottomLeft) * xLerp;\n\n              ind = c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = top + ((bottom - top) * yLerp);\n            }\n          }\n        } else {  // method == \"nearest\"\n          for (let x = 0; x < cropWidth; ++x) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const closestX = Math.round(xInd);\n            const closestY = Math.round(yInd);\n            for (let c = 0; c < numChannels; c++) {\n              const inInd = c + closestX * inStride[2] +\n                  closestY * inStride[1] + bInd * inStride[0];\n              const outInd =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[outInd] = imageVals[inInd];\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor4D;\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n    const sumDupeIndices = false;\n    return this.scatter(\n        sparseIndices, sparseValues, outputShape, outputSize, sliceSize,\n        numUpdates, sliceRank, strides, defaultValue, sumDupeIndices);\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n    if (numSlices === 0) {\n      return tf.tensor([], resultShape, x.dtype);\n    }\n\n    const buffer = new TensorBuffer([numSlices, sliceSize], x.dtype);\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const xData = this.readSync(x.dataId) as TypedArray;\n\n    for (let i = 0; i < numSlices; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        flattenIndex += dim * strides[j];\n        index.push(dim);\n      }\n      if (flattenIndex < 0 || flattenIndex >= x.size / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${x.shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        buffer.values[i * sliceSize + k] = xData[flattenIndex * sliceSize + k];\n      }\n    }\n    return buffer.toTensor().reshape(resultShape);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n    const defaultValue = tf.scalar(0);\n    const sumDupeIndices = true;\n    return this.scatter(\n        indices, updates, shape, outputSize, sliceSize, numUpdates, sliceRank,\n        strides, defaultValue, sumDupeIndices);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n    const values =\n        util.getArrayFromDType(dtype, util.sizeFromShape(shape)) as TypedArray;\n    values.fill(value as number);\n    return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported for string tensors');\n    } else {\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    const values = util.getArrayFromDType(\n                       x.dtype, util.sizeFromShape(x.shape)) as TypedArray;\n    return this.makeOutput(values, x.shape, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  private scatter<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R], outputSize: number,\n      sliceSize: number, numUpdates: number, sliceRank: number,\n      strides: number[], defaultValue: Scalar,\n      sumDupeIndices: boolean): Tensor<R> {\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const updatesData = this.readSync(updates.dataId) as TypedArray;\n\n    if (outputSize === 0) {\n      return tf.tensor([], shape, updates.dtype);\n    }\n\n    const buffer = new TensorBuffer(flattenShape, updates.dtype as 'float32');\n    buffer.values.fill((this.readSync(defaultValue.dataId) as TypedArray)[0]);\n\n    for (let i = 0; i < numUpdates; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        index.push(dim);\n        flattenIndex += dim * strides[j];\n      }\n\n      if (flattenIndex < 0 || flattenIndex >= outputSize / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        if (sumDupeIndices) {\n          buffer.values[flattenIndex * sliceSize + k] +=\n              updatesData[i * sliceSize + k];\n        } else {\n          buffer.values[flattenIndex * sliceSize + k] = updates.rank === 0 ?\n              updatesData[0] :\n              updatesData[i * sliceSize + k];\n        }\n      }\n    }\n    return buffer.toTensor().reshape(shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs, AbsInputs, KernelConfig, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function simpleAbsImpl(vals: TypedArray): Float32Array {\n  const resultValues = new Float32Array(vals.length);\n  for (let i = 0; i < vals.length; ++i) {\n    resultValues[i] = Math.abs(vals[i]);\n  }\n  return resultValues;\n}\n\nexport const absKernelFunc =\n    (args: {inputs: AbsInputs, backend: MathBackendCPU}) => {\n      const {x} = args.inputs;\n      const cpuBackend = args.backend;\n      let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n      if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n        resultValues = simpleAbsImpl(values);\n      } else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals =\n            cpuBackend.data.get(real.dataId).values as Float32Array;\n        const imagVals =\n            cpuBackend.data.get(imag.dataId).values as Float32Array;\n        for (let i = 0; i < realVals.length; i++) {\n          const real = realVals[i];\n          const imag = imagVals[i];\n          resultValues[i] = Math.hypot(real, imag);\n        }\n      }\n      return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n    };\n\nexport const absConfig: KernelConfig = {\n  kernelName: Abs,\n  backendName: 'cpu',\n  kernelFunc: absKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleBinaryKernelImpl, SimpleBinaryOperation} from './binary_types';\n\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op: SimpleBinaryOperation):\n    SimpleBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aVals: TypedArray,\n          bVals: TypedArray, dtype: DataType): [TypedArray, number[]] => {\n    const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    const resultRank = newShape.length;\n    const resultStrides = util.computeStrides(newShape);\n    const resultSize = util.sizeFromShape(newShape);\n\n    const result =\n        util.getTypedArrayFromDType(dtype as NumericDataType, resultSize);\n\n    const aRank = aShape.length;\n    const bRank = bShape.length;\n\n    const aStrides = util.computeStrides(aShape);\n    const bStrides = util.computeStrides(bShape);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < result.length; ++i) {\n        result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      for (let i = 0; i < result.length; ++i) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        result[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n\n    return [result, newShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Complex, ComplexInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function complex(args: {inputs: ComplexInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {real, imag} = inputs;\n\n  const realVals = backend.data.get(real.dataId).values as TypedArray;\n  const imagVals = backend.data.get(imag.dataId).values as TypedArray;\n\n  const complexInfo = backend.makeTensorInfo(real.shape, 'complex64');\n\n  const complex = backend.data.get(complexInfo.dataId);\n\n  // The complex tensor owns the underlying real and imag tensorInfos, only the\n  // complex tensor tracks refCount, when complexData is disposed the\n  // underlying tensorData will be disposed.\n  complex.complexTensorInfos = {\n    real: backend.makeTensorInfo(real.shape, 'float32', realVals),\n    imag: backend.makeTensorInfo(imag.shape, 'float32', imagVals)\n  };\n\n  return complexInfo;\n}\n\nexport const complexConfig: KernelConfig = {\n  kernelName: Complex,\n  backendName: 'cpu',\n  kernelFunc: complex as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendCPU}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'cpu',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Real, RealInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function real(args: {inputs: RealInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const real = backend.data.get(input.dataId).complexTensorInfos.real;\n  const realVal = backend.data.get(real.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the real value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(real.shape, real.dtype, realVal);\n}\n\nexport const realConfig: KernelConfig = {\n  kernelName: Real,\n  backendName: 'cpu',\n  kernelFunc: real as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport * as tf from '@tensorflow/tfjs-core';\nimport {Cast, CastAttrs, CastInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\n\nimport {complex} from './Complex';\nimport {identity} from './Identity';\nimport {real} from './Real';\n\nexport function cast(\n    args: {inputs: CastInputs, backend: MathBackendCPU, attrs: CastAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {dtype} = attrs;\n\n  // Casting to complex64.\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return identity({inputs: {x}, backend});\n    }\n\n    // TODO(lina128): Import kernel function once zeros is modularized.\n    const zerosTensor = tf.zeros(x.shape);\n    const floatX = cast({inputs: {x}, backend, attrs: {dtype: 'float32'}});\n\n    const result =\n        complex({inputs: {real: floatX, imag: zerosTensor}, backend});\n\n    zerosTensor.dispose();\n    backend.disposeIntermediateTensorInfo(floatX);\n\n    return result;\n  }\n\n  // Casting from complex64\n  if (x.dtype === 'complex64') {\n    const realPart = real({inputs: {input: x}, backend});\n    const result = cast({inputs: {x: realPart}, backend, attrs: {dtype}});\n\n    backend.disposeIntermediateTensorInfo(realPart);\n\n    return result;\n  }\n\n  if (!util.hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    const result = identity({inputs: {x}, backend});\n    return {dataId: result.dataId, shape: result.shape, dtype};\n  }\n\n  if (dtype === 'int32') {\n    const values = backend.data.get(x.dataId).values as TypedArray;\n    const resultValues = Int32Array.from(values);\n    return backend.makeTensorInfo(x.shape, 'int32', resultValues);\n  }\n\n  if (dtype === 'bool') {\n    // This is essentially the result of notEqual(x, 0). We avoid using\n    // kernel notEqual to avoid circular dependency, i.e. binary_utils ->\n    // cast -> notEqual -> binary_utils.\n    const xVals = backend.data.get(x.dataId).values as TypedArray;\n    const zero = util.toTypedArray([0], x.dtype);\n\n    const [resultData, resultShape] = createSimpleBinaryKernelImpl(\n        (a, b) => (a !== b) ? 1 : 0)(x.shape, [], xVals, zero, 'bool');\n\n    return backend.makeTensorInfo(resultShape, 'bool', resultData);\n  }\n\n  throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n}\n\nexport const castConfig: KernelConfig = {\n  kernelName: Cast,\n  backendName: 'cpu',\n  kernelFunc: cast as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, BinaryInputs, DataType, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {cast} from '../kernels/Cast';\nimport {complex} from '../kernels/Complex';\n\nimport {ComplexBinaryKernelImpl, ComplexBinaryOperation, SimpleBinaryKernelImpl} from './binary_types';\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param name Kernel name.\n * @param binaryKernelImpl A `SimpleBinaryKernelImpl` for the kernel.\n * @param binaryKernelComplexImpl Optional. If exists, represents a\n *     `ComplexBinaryKernelImpl` for the kernel, will be used when input dtype\n *     is `complex64`.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    name: string, simpleImpl: SimpleBinaryKernelImpl,\n    complexImpl?: ComplexBinaryKernelImpl, dtype?: DataType): KernelFunc {\n  if (complexImpl == null) {\n    return ({inputs, backend}) => {\n      const {a, b} = inputs as BinaryInputs;\n      const cpuBackend = backend as MathBackendCPU;\n\n      assertNotComplex([a, b], name);\n\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    };\n  }\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    if (a.dtype === 'complex64' || b.dtype === 'complex64') {\n      const $aComplex = cast(\n          {inputs: {x: a}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $aComplexVals = cpuBackend.data.get($aComplex.dataId);\n\n      const aReal = $aComplexVals.complexTensorInfos.real;\n      const aImag = $aComplexVals.complexTensorInfos.imag;\n\n      const aRealVals =\n          cpuBackend.data.get(aReal.dataId).values as Float32Array;\n      const aImagVals =\n          cpuBackend.data.get(aImag.dataId).values as Float32Array;\n\n      const $bComplex = cast(\n          {inputs: {x: b}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $bComplexVals = cpuBackend.data.get($bComplex.dataId);\n\n      const bReal = $bComplexVals.complexTensorInfos.real;\n      const bImag = $bComplexVals.complexTensorInfos.imag;\n\n      const bRealVals =\n          cpuBackend.data.get(bReal.dataId).values as Float32Array;\n      const bImagVals =\n          cpuBackend.data.get(bImag.dataId).values as Float32Array;\n\n      const [resultRealData, resultImagData, resultShape] = complexImpl(\n          a.shape, b.shape, aRealVals, aImagVals, bRealVals, bImagVals);\n\n      const resultReal =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultRealData);\n\n      const resultImag =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultImagData);\n\n      const result = complex(\n          {inputs: {real: resultReal, imag: resultImag}, backend: cpuBackend});\n\n      cpuBackend.disposeIntermediateTensorInfo($aComplex);\n      cpuBackend.disposeIntermediateTensorInfo($bComplex);\n      cpuBackend.disposeIntermediateTensorInfo(resultReal);\n      cpuBackend.disposeIntermediateTensorInfo(resultImag);\n\n      return result;\n    } else {\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    }\n  };\n}\n\n/**\n * Template that creates the complex type implementation for binary ops.\n * Supports broadcast.\n */\nexport function createComplexBinaryKernelImpl(op: ComplexBinaryOperation):\n    ComplexBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aRealVals: Float32Array,\n          aImagVals: Float32Array, bRealVals: Float32Array,\n          bImagVals: Float32Array): [TypedArray, TypedArray, number[]] => {\n    const resultShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const resultSize = util.sizeFromShape(resultShape);\n    const resultRank = resultShape.length;\n    const resultStrides = util.computeStrides(resultShape);\n\n    const resultRealVals = util.getTypedArrayFromDType('float32', resultSize);\n    const resultImagVals = util.getTypedArrayFromDType('float32', resultSize);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, resultShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, resultShape);\n\n    const aVals = backend_util.mergeRealAndImagArrays(aRealVals, aImagVals);\n    const bVals = backend_util.mergeRealAndImagArrays(bRealVals, bImagVals);\n\n    const aRank = aShape.length;\n    const aStrides = util.computeStrides(aShape);\n\n    const bRank = bShape.length;\n    const bStrides = util.computeStrides(bShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const aIdx = i % aVals.length;\n        const bIdx = i % bVals.length;\n\n        const result =\n            op(aVals[aIdx * 2], aVals[aIdx * 2 + 1], bVals[bIdx * 2],\n               bVals[bIdx * 2 + 1]);\n\n        resultRealVals[i] = result.real;\n        resultImagVals[i] = result.imag;\n      }\n    } else {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        const opResult =\n            op(aVals[aIndex * 2], aVals[aIndex * 2 + 1], bVals[bIndex * 2],\n               bVals[bIndex * 2 + 1]);\n\n        resultRealVals[i] = opResult.real;\n        resultImagVals[i] = opResult.imag;\n      }\n    }\n    return [resultRealVals, resultImagVals, resultShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Add, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal + bReal, imag: aImag + bImag};\n    }));\n\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\n\nexport const addConfig: KernelConfig = {\n  kernelName: Add,\n  backendName: 'cpu',\n  kernelFunc: add\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NumericDataType, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op: SimpleUnaryOperation):\n    SimpleUnaryImpl {\n  return (values, dtype, attrs) => {\n    const newValues =\n        util.getTypedArrayFromDType(dtype as NumericDataType, values.length);\n    for (let i = 0; i < values.length; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return newValues;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelFunc, TypedArray, UnaryInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param name Kernel name.\n * @param op A `SimpleUnaryOperation` for the kernel.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFunc(\n    name: string, op: SimpleUnaryOperation, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xSize = util.sizeFromShape(x.shape);\n    const $dtype = dtype || x.dtype;\n    const newValues = util.getArrayFromDType($dtype, xSize);\n    for (let i = 0; i < xSize; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for unary ops from the given\n * `SimpleUnaryImpl`..\n * @param name Kernel name.\n * @param unaryImpl A `SimpleUnaryImpl` that implements the op.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFuncFromImpl(\n    name: string, unaryImpl: SimpleUnaryImpl, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const $dtype = dtype || x.dtype;\n    const newValues = unaryImpl(values, $dtype, attrs);\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\n\nexport const ceilConfig: KernelConfig = {\n  kernelName: Ceil,\n  backendName: 'cpu',\n  kernelFunc: ceilKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\n\nexport const expConfig: KernelConfig = {\n  kernelName: Exp,\n  backendName: 'cpu',\n  kernelFunc: expKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\n\nexport const expm1Config: KernelConfig = {\n  kernelName: Expm1,\n  backendName: 'cpu',\n  kernelFunc: expm1KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\n\nexport const floorConfig: KernelConfig = {\n  kernelName: Floor,\n  backendName: 'cpu',\n  kernelFunc: floorKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\n\nexport const logConfig: KernelConfig = {\n  kernelName: Log,\n  backendName: 'cpu',\n  kernelFunc: logKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function maxImpl(\n    aVals: TypedArray, reduceSize: number, outShape: number[],\n    dtype: DataType): TypedArray {\n  const vals = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(outShape));\n\n  for (let i = 0; i < vals.length; ++i) {\n    const offset = i * reduceSize;\n    let max = aVals[offset];\n    for (let j = 0; j < reduceSize; ++j) {\n      const value = aVals[offset + j];\n      if (value > max) {\n        max = value;\n      }\n    }\n    vals[i] = max;\n  }\n  return vals;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Multiply} from '@tensorflow/tfjs-core';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const multiplyImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n      };\n    }));\n\nexport const multiply =\n    binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\n\nexport const multiplyConfig: KernelConfig = {\n  kernelName: Multiply,\n  backendName: 'cpu',\n  kernelFunc: multiply\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Rsqrt} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\n\nexport const rsqrtConfig: KernelConfig = {\n  kernelName: Rsqrt,\n  backendName: 'cpu',\n  kernelFunc: rsqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelConfig, KernelFunc, NumericDataType, Slice, slice_util, SliceAttrs, SliceInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function sliceImpl(\n    vals: TypedArray, begin: number[], size: number[], shape: number[],\n    dtype: DataType): TypedArray {\n  const isContinous = slice_util.isSliceContinous(shape, begin, size);\n  const length = util.sizeFromShape(size);\n  const xStrides = util.computeStrides(shape);\n\n  if (isContinous) {\n    const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n    return vals.subarray(flatOffset, flatOffset + length);\n  }\n\n  const outVals = util.getTypedArrayFromDType(dtype as NumericDataType, length);\n  for (let i = 0; i < length; ++i) {\n    const rank = size.length;\n    const strides = util.computeStrides(size);\n    const loc = util.indexToLoc(i, rank, strides);\n    const xLoc = loc.map((idx: number, j) => idx + begin[j]);\n    const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n    outVals[i] = vals[xIndex];\n  }\n  return outVals;\n}\n\nexport function slice(\n    args: {inputs: SliceInputs, backend: MathBackendCPU, attrs: SliceAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {begin, size} = attrs;\n\n  assertNotComplex(x, 'slice');\n\n  const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n  slice_util.assertParamsValid(x, $begin, $size);\n\n  const vals = backend.data.get(x.dataId).values as TypedArray;\n  const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n  return backend.makeTensorInfo($size, x.dtype, outVals);\n}\n\nexport const sliceConfig: KernelConfig = {\n  kernelName: Slice,\n  backendName: 'cpu',\n  kernelFunc: slice as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sub} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const subImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal - bReal, imag: aImag - bImag};\n    }));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\n\nexport const subConfig: KernelConfig = {\n  kernelName: Sub,\n  backendName: 'cpu',\n  kernelFunc: sub\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function transposeImpl(\n    xVals: TypedArray, xShape: number[], dtype: DataType, perm: number[],\n    newShape: number[]): TypedArray {\n  const xRank = xShape.length;\n  const xSize = util.sizeFromShape(xShape);\n  const xStrides = util.computeStrides(xShape);\n  const newStrides = util.computeStrides(newShape);\n\n  const result = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(newShape));\n\n  for (let i = 0; i < xSize; ++i) {\n    const loc = util.indexToLoc(i, xRank, xStrides);\n\n    // Permute location.\n    const newLoc: number[] = new Array(loc.length);\n    for (let i = 0; i < newLoc.length; i++) {\n      newLoc[i] = loc[perm[i]];\n    }\n\n    const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n    result[newIndex] = xVals[i];\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendValues, DataType, TensorBuffer, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function uniqueImpl(\n    values: BackendValues, axis: number, shape: number[], dtype: DataType): {\n  outputValues: BackendValues,\n  outputShape: number[],\n  indices: BackendValues\n} {\n  // Normalize and validate axis.\n  const $axis = util.parseAxisParam(axis, shape)[0];\n\n  // Calculate the new shape that is suitable for extracting data along the\n  // given axis.\n  //\n  // The rank is 3.\n  // The size of the 1st dimension is the size of all the axes < the given axis.\n  // The size of the 2nd dimension is the same as the size of the given axis.\n  // The size of the 3rd dimension is the size of all the axes > the given axis.\n  //\n  // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n  // newShape would be: [2*3, 5, 4].\n  //\n  // Note that this is not the final output shape. This will be the shape for an\n  // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n  // values along the given axis. To demonstrate how it works, consider the\n  // following example:\n  //\n  // Input: a 3D tensor, with shape [1, 2, 3]\n  // [\n  //   [\n  //      [1,2,3],\n  //      [4,5,6]\n  //   ]\n  // ]\n  // Axis: 2 (the last axis).\n  // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n  //\n  // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n  // 1*2. The re-shaped data would look like:\n  //\n  // [\n  //   [\n  //     [1], [2], [3]\n  //   ],\n  //   [\n  //     [4], [5], [6]\n  //   ]\n  // ]\n  //\n  // Then, we can construct a 3-level nested loop by the following dimension\n  // order to extract the values along the axis (dimension1):\n  // i: dimension1       // 0,1,2 (newShape[1])\n  //   m: dimension0     // 0,1   (newShape[0])\n  //     n: dimension2   // 0     (newShape[2])\n  //\n  //                       m, i, n\n  //                      ---------\n  // Iteration 0: data at [0, 0, 0] => \"1\"\n  // Iteration 1: data at [1, 0, 0] => \"4\"\n  // We got [1,4].\n  // Iteration 2: data at [0, 1, 0] => \"2\"\n  // Iteration 3: data at [1, 1, 0] => \"5\"\n  // We got [2,5].\n  // Iteration 4: data at [0, 2, 0] => \"3\"\n  // Iteration 5: data at [1, 2, 0] => \"6\"\n  // We got [3,6].\n  const newShape = [1, shape[0], 1];\n  for (let i = 0; i < $axis; i++) {\n    newShape[0] *= shape[i];\n  }\n  newShape[1] = shape[$axis];\n  for (let i = $axis + 1; i < shape.length; i++) {\n    newShape[2] *= shape[i];\n  }\n\n  // A map from unique elements (their string representations) to their values\n  // in \"indices\" (below).\n  const uniqueElements: {[key: string]: number} = {};\n  // The indices of each unique element in the original tensor along the given\n  // axis. It is 1D and has the same size as the given axis.\n  const indices = new Int32Array(shape[$axis]);\n  // Create a buffer so we can easily extract value at a given location.\n  const inputBuffer = new TensorBuffer(newShape, dtype, values as TypedArray);\n  // The indices along the given axis that have unique elements. This is a\n  // de-duped version of \"indices\" above.\n  const uniqueIndices: number[] = [];\n  const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n  for (let i = 0; i < shape[$axis]; i++) {\n    // Extract values along the axis.\n    let element: string;\n    if (is1DTensor) {\n      // Fast path for 1D tensor input.\n      element = values[i].toString();\n    } else {\n      const axisValues = [];\n      for (let m = 0; m < newShape[0]; m++) {\n        for (let n = 0; n < newShape[2]; n++) {\n          axisValues.push(inputBuffer.get(m, i, n));\n        }\n      }\n      element = axisValues.join(',');\n    }\n\n    // Dedup and update various indices.\n    if (uniqueElements[element] !== undefined) {\n      indices[i] = uniqueElements[element];\n    } else {\n      const uniqueIndex = Object.keys(uniqueElements).length;\n      uniqueElements[element] = uniqueIndex;\n      indices[i] = uniqueIndex;\n      uniqueIndices.push(i);\n    }\n  }\n\n  // Now we know where each of the unique elements are located along the axis\n  // (uniqueIndices). Extract them from input buffer and store them in the\n  // output buffer.\n  const outputTmpShape = newShape.slice();\n  outputTmpShape[1] = Object.keys(uniqueElements).length;\n  const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n  uniqueIndices.forEach((uniqueElementIndex, i) => {\n    for (let m = 0; m < newShape[0]; m++) {\n      for (let n = 0; n < newShape[2]; n++) {\n        outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n      }\n    }\n  });\n\n  // The output shape can be calculated from the input shape with the size of\n  // the given axis replaced by the number of unique elements along that axis.\n  const outputShape = shape.slice();\n  outputShape[$axis] = outputTmpShape[1];\n\n  return {\n    outputValues: outputBuffer.values as BackendValues,\n    outputShape,\n    indices,\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Shared kernel impls for use in other backends.\nexport {simpleAbsImpl} from './kernels/Abs';\nexport {addImpl} from './kernels/Add';\nexport {ceilImpl} from './kernels/Ceil';\nexport {expImpl} from './kernels/Exp';\nexport {expm1Impl} from './kernels/Expm1';\nexport {floorImpl} from './kernels/Floor';\nexport {logImpl} from './kernels/Log';\nexport {maxImpl} from './kernels/Max_impl';\nexport {multiplyImpl} from './kernels/Multiply';\nexport {rsqrtImpl} from './kernels/Rsqrt';\nexport {sliceImpl} from './kernels/Slice';\nexport {subImpl} from './kernels/Sub';\nexport {transposeImpl} from './kernels/Transpose_impl';\nexport {uniqueImpl} from './kernels/Unique_impl';\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/*\n * base.ts contains all the exports from tfjs-backend-cpu\n * without auto-kernel registration\n */\nimport {registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from './backend_cpu';\nimport * as shared from './shared';\n\nexport {MathBackendCPU} from './backend_cpu';\nexport {version as version_cpu} from './version';\nexport {shared};\n\n// Side effects for default initialization of MathBackendCPU\nregisterBackend('cpu', () => new MathBackendCPU(), 1 /* priority */);\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acosKernelFunc = unaryKernelFunc(Acos, (xi) => Math.acos(xi));\n\nexport const acosConfig: KernelConfig = {\n  kernelName: Acos,\n  backendName: 'cpu',\n  kernelFunc: acosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acoshKernelFunc = unaryKernelFunc(Acosh, (xi) => Math.acosh(xi));\n\nexport const acoshConfig: KernelConfig = {\n  kernelName: Acosh,\n  backendName: 'cpu',\n  kernelFunc: acoshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinKernelFunc = unaryKernelFunc(Asin, (xi) => Math.asin(xi));\n\nexport const asinConfig: KernelConfig = {\n  kernelName: Asin,\n  backendName: 'cpu',\n  kernelFunc: asinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinhKernelFunc = unaryKernelFunc(Asinh, (xi) => Math.asinh(xi));\n\nexport const asinhConfig: KernelConfig = {\n  kernelName: Asinh,\n  backendName: 'cpu',\n  kernelFunc: asinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanKernelFunc = unaryKernelFunc(Atan, (xi) => Math.atan(xi));\n\nexport const atanConfig: KernelConfig = {\n  kernelName: Atan,\n  backendName: 'cpu',\n  kernelFunc: atanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanhKernelFunc = unaryKernelFunc(Atanh, (xi) => Math.atanh(xi));\n\nexport const atanhConfig: KernelConfig = {\n  kernelName: Atanh,\n  backendName: 'cpu',\n  kernelFunc: atanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, buffer, DataType, Rank, TensorBuffer, TypedArray} from '@tensorflow/tfjs-core';\n\nexport function pool(\n    xValues: TypedArray, xShape: number[], dtype: DataType, strides: number[],\n    convInfo: backend_util.Conv2DInfo,\n    poolType: 'max'|'avg'): TensorBuffer<Rank, DataType> {\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const initialValue =\n      (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                            Number.POSITIVE_INFINITY);\n\n  const output = buffer(convInfo.outShape, dtype);\n  const outputVals = output.values;\n\n  const outputBatchStrides =\n      convInfo.outShape[1] * convInfo.outShape[2] * convInfo.outShape[3];\n  const outputRowStrides = convInfo.outShape[2] * convInfo.outShape[3];\n  const outputColStrides = convInfo.outShape[3];\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    const outputBatchOffset = b * outputBatchStrides;\n    const inputBatchOffset = b * strides[0];\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        const outputRowOffset = outputBatchOffset + yR * outputRowStrides;\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          const xCMin = Math.max(0, xCCorner);\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let minMaxValue = initialValue;\n          let avgValue = 0;\n          let count = 0;\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const xROffset = inputBatchOffset + xR * strides[1];\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const xCOffset = xROffset + xC * strides[2];\n              const pixel = xValues[xCOffset + d];\n              if ((poolType === 'max' && pixel > minMaxValue)) {\n                minMaxValue = pixel;\n              } else if (poolType === 'avg') {\n                avgValue += pixel;\n                count++;\n              }\n            }\n            if (isNaN(minMaxValue)) {\n              break;\n            }\n          }\n          const outputOffset = outputRowOffset + yC * outputColStrides + d;\n          outputVals[outputOffset] =\n              poolType === 'avg' ? avgValue / count : minMaxValue;\n        }\n      }\n    }\n  }\n  return output;\n}\n\nexport function maxPoolPositions(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    convInfo: backend_util.Conv2DInfo, flattenPositions = false,\n    includeBatchInIndex = false): TensorBuffer<Rank, 'int32'> {\n  const maxPositions = buffer(convInfo.outShape, 'int32');\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const xBuf = buffer(xShape, dtype, xValues);\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        let xRMin = xRCorner;\n        while (xRMin < 0) {\n          xRMin += dilationHeight;\n        }\n        // const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          let xCMin = xCCorner;\n          while (xCMin < 0) {\n            xCMin += dilationWidth;\n          }\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let maxValue = Number.NEGATIVE_INFINITY;\n          let maxPosition = -1;\n\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const wR = xR - xRCorner;\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const wC = xC - xCCorner;\n              const pixel = xBuf.get(b, xR, xC, d);\n              if (pixel > maxValue) {\n                maxValue = pixel as number;\n                if (flattenPositions) {\n                  maxPosition = includeBatchInIndex ?\n                      ((b * convInfo.inHeight + xR) * convInfo.inWidth + xC) *\n                              convInfo.inChannels +\n                          d :\n                      (xR * convInfo.inWidth + xC) * convInfo.inChannels + d;\n                } else {\n                  maxPosition = wR * effectiveFilterWidth + wC;\n                }\n              }\n            }\n          }\n          maxPositions.set(maxPosition, b, yR, yC, d);\n        }\n      }\n    }\n  }\n  return maxPositions;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function avgPool(\n    args:\n        {inputs: AvgPoolInputs, backend: MathBackendCPU, attrs: AvgPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'avg');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'cpu',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, buffer, KernelConfig, KernelFunc, Rank, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const filterHeight = convInfo.filterHeight;\n  const filterWidth = convInfo.filterWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel;\n            }\n          }\n          dx.set(dotProd * avgMultiplier, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function batchNormKernelFunc(args: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendCPU,\n  attrs: FusedBatchNormAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x, scale, offset, mean, variance} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  assertNotComplex([x, mean, variance, scale, offset], 'batchNorm');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const mVals = backend.data.get(mean.dataId).values as TypedArray;\n  const varVals = backend.data.get(variance.dataId).values as TypedArray;\n  const sVals = scale ? backend.data.get(scale.dataId).values as TypedArray :\n                        new Float32Array([1]);\n  const offVals = offset ?\n      backend.data.get(offset.dataId).values as TypedArray :\n      new Float32Array([0]);\n  const outVals = new Float32Array(xVals.length);\n\n  const offValsLength = offVals.length;\n  const sValsLength = sVals.length;\n  const varValsLength = varVals.length;\n  const mValsLength = mVals.length;\n\n  let offi = 0;\n  let mi = 0;\n  let si = 0;\n  let vi = 0;\n  for (let i = 0; i < xVals.length; ++i) {\n    outVals[i] = offVals[offi++] +\n        (xVals[i] - mVals[mi++]) * sVals[si++] /\n            Math.sqrt(varVals[vi++] + varianceEpsilon);\n    if (offi >= offValsLength) {\n      offi = 0;\n    }\n    if (mi >= mValsLength) {\n      mi = 0;\n    }\n    if (si >= sValsLength) {\n      si = 0;\n    }\n    if (vi >= varValsLength) {\n      vi = 0;\n    }\n  }\n  return backend.makeTensorInfo(x.shape, x.dtype, outVals);\n}\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'cpu',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const clipKernelFunc = unaryKernelFunc(ClipByValue, (xi, attrs) => {\n  const clipAttrs = attrs as {} as ClipByValueAttrs;\n  if (xi > clipAttrs.clipValueMax) {\n    return clipAttrs.clipValueMax;\n  }\n  return xi < clipAttrs.clipValueMin ? clipAttrs.clipValueMin : xi;\n});\n\nexport const clipConfig: KernelConfig = {\n  kernelName: ClipByValue,\n  backendName: 'cpu',\n  kernelFunc: clipKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Imag, ImagInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function imag(args: {inputs: ImagInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const imag = backend.data.get(input.dataId).complexTensorInfos.imag;\n  const imagVal = backend.data.get(imag.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the imag value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(imag.shape, imag.dtype, imagVal);\n}\n\nexport const imagConfig: KernelConfig = {\n  kernelName: Imag,\n  backendName: 'cpu',\n  kernelFunc: imag as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function reshape(\n    args:\n        {inputs: ReshapeInputs, backend: MathBackendCPU, attrs: ReshapeAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  backend.incRef(x.dataId);\n\n  const xData = backend.data.get(x.dataId);\n\n  if (xData.complexTensorInfos != null) {\n    const real = xData.complexTensorInfos.real;\n    const imag = xData.complexTensorInfos.imag;\n\n    real.shape = $shape;\n    imag.shape = $shape;\n  }\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'cpu',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Concat, ConcatAttrs, ConcatInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nimport {complex} from './Complex';\nimport {imag} from './Imag';\nimport {real} from './Real';\nimport {reshape} from './Reshape';\n\nexport function concat(\n    args: {inputs: ConcatInputs, backend: MathBackendCPU, attrs: ConcatAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {axis} = attrs;\n\n  const $axis = util.parseAxisParam(axis, inputs[0].shape)[0];\n  let outShape = backend_util.computeOutShape(inputs.map(t => t.shape), $axis);\n\n  if (util.sizeFromShape(outShape) === 0) {\n    return backend.makeTensorInfo(outShape, inputs[0].dtype, []);\n  }\n\n  // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n  const $inputs = inputs.filter(t => util.sizeFromShape(t.shape) > 0);\n  if ($inputs.length === 1) {\n    return $inputs[0];\n  }\n\n  const shapes = $inputs.map(t => t.shape);\n  backend_util.assertParamsConsistent(shapes, $axis);\n\n  if ($inputs[0].dtype === 'complex64') {\n    const reals = $inputs.map((t) => real({inputs: {input: t}, backend}));\n    const imags = $inputs.map((t) => imag({inputs: {input: t}, backend}));\n\n    const realConcated = concat({inputs: reals, backend, attrs: {axis}});\n    const imagConcated = concat({inputs: imags, backend, attrs: {axis}});\n\n    const result =\n        complex({inputs: {real: realConcated, imag: imagConcated}, backend});\n\n    reals.forEach(r => backend.disposeIntermediateTensorInfo(r));\n    imags.forEach(i => backend.disposeIntermediateTensorInfo(i));\n    backend.disposeIntermediateTensorInfo(realConcated);\n    backend.disposeIntermediateTensorInfo(imagConcated);\n\n    return result;\n  }\n\n  // Any concat of n-dimensional tensors across any axis can be reduced to\n  // a concatenation of two-dimensional tensors across the axis 1 by first\n  // partitioning the axes of the original tensors into those less than the\n  // axis to be concatenated and the rest. Then reshape the tensors\n  // into a two-dimensional tensor by collapsing these two sets of axes and\n  // concatenate the resulting matrices across the axis 1, finally reshaping\n  // the result to have the proper shape.\n  const inputs2D = $inputs.map(t => {\n    const innerSize = util.sizeFromShape(t.shape.slice($axis));\n    const shape = [-1, innerSize];\n    return reshape({inputs: {x: t}, backend, attrs: {shape}});\n  });\n\n  // Concats 2d tensors along axis=1.\n  outShape =\n      backend_util.computeOutShape(inputs2D.map(t => t.shape), 1 /* axis */);\n\n  const outVals = util.getTypedArrayFromDType(\n      $inputs[0].dtype as 'float32', util.sizeFromShape(outShape));\n\n  if (inputs2D[0].shape[0] === 1) {\n    // Use built-in TypedArray.set() method for speed.\n    let offset = 0;\n    inputs2D.forEach(t => {\n      const val = backend.data.get(t.dataId).values as TypedArray;\n      const size = util.sizeFromShape(t.shape);\n\n      outVals.set(val, offset);\n      offset += size;\n    });\n  } else {\n    let colOffset = 0;\n\n    inputs2D.forEach(t => {\n      const tVals = backend.data.get(t.dataId).values as TypedArray;\n\n      let tIdx = 0;\n\n      for (let row = 0; row < t.shape[0]; ++row) {\n        const resIdx = row * outShape[1] + colOffset;\n        for (let col = 0; col < t.shape[1]; ++col) {\n          outVals[resIdx + col] = tVals[tIdx++];\n        }\n      }\n\n      colOffset += t.shape[1];\n    });\n  }\n\n  const finalOutShape =\n      backend_util.computeOutShape($inputs.map(t => t.shape), $axis);\n\n  const outInfo =\n      backend.makeTensorInfo(finalOutShape, inputs[0].dtype, outVals);\n\n  inputs2D.forEach(t => backend.disposeIntermediateTensorInfo(t));\n\n  return outInfo;\n}\n\nexport const concatConfig: KernelConfig = {\n  kernelName: Concat,\n  backendName: 'cpu',\n  kernelFunc: concat as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const cosKernelFunc = unaryKernelFunc(Cos, (xi) => Math.cos(xi));\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'cpu',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const coshKernelFunc = unaryKernelFunc(Cosh, (xi) => Math.cosh(xi));\n\nexport const coshConfig: KernelConfig = {\n  kernelName: Cosh,\n  backendName: 'cpu',\n  kernelFunc: coshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2D, Dilation2DAttrs, Dilation2DInputs, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dConfig: KernelConfig = {\n  kernelName: Dilation2D,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter} = inputs as Dilation2DInputs;\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xRank = x.shape.length;\n\n    const filterVals = cpuBackend.data.get(filter.dataId).values as TypedArray;\n    const filterRank = filter.shape.length;\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    const outSize = util.sizeFromShape(outShape);\n    const outRank = outShape.length;\n    const outputVals = util.getArrayFromDType(x.dtype, outSize);\n\n    // Upsampling the input by fill in `dilation size - 1` values between each\n    // input value.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const xIndex = util.locToIndex(\n                        [b, hIn, wIn, d], xRank, util.computeStrides(x.shape));\n                    const filterIndex = util.locToIndex(\n                        [h, w, d], filterRank,\n                        util.computeStrides(filter.shape));\n                    const val = xVals[xIndex] + filterVals[filterIndex];\n                    if (val > curVal) {\n                      curVal = val;\n                    }\n                  }\n                }\n              }\n            }\n            const outputIndex = util.locToIndex(\n                [b, hOut, wOut, d], outRank, util.computeStrides(outShape));\n            outputVals[outputIndex] = curVal;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(outputVals, x.dtype), outShape, x.dtype);\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropFilter, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropFilterConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropFilter,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropFilter}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed filter gradients has the same dimensions as the filter:\n    // [filterHeight, filterWidth, depth]\n    const gradients = util.makeZerosNestedTypedArray(\n                          filter.shape, filter.dtype) as number[][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hMax = 0;\n            let wMax = 0;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hMax = h;\n                      wMax = w;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[hMax][wMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), filter.shape, filter.dtype);\n\n    return {dataId, shape: filter.shape, dtype: filter.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropInput, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropInputConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropInput,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropInput}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed gradients has the same dimensions as the input:\n    // [batch, inputHeight, inputCols, inChannel]\n    const gradients =\n        util.makeZerosNestedTypedArray(x.shape, x.dtype) as number[][][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hInMax = (hBeg < 0) ? 0 : hBeg;\n            let wInMax = (wBeg < 0) ? 0 : wBeg;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hInMax = hIn;\n                      wInMax = wIn;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[b][hInMax][wInMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), x.shape, x.dtype);\n\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const divImpl =\n    createSimpleBinaryKernelImpl((a: number, b: number) => a / b);\nexport const div = binaryKernelFunc(Div, divImpl);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'cpu',\n  kernelFunc: div\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Elu, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const eluKernelFunc =\n    unaryKernelFunc(Elu, (xi) => xi >= 0 ? xi : (Math.exp(xi) - 1));\n\nexport const eluConfig: KernelConfig = {\n  kernelName: Elu,\n  backendName: 'cpu',\n  kernelFunc: eluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Erf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst p = backend_util.ERF_P;\nconst a1 = backend_util.ERF_A1;\nconst a2 = backend_util.ERF_A2;\nconst a3 = backend_util.ERF_A3;\nconst a4 = backend_util.ERF_A4;\nconst a5 = backend_util.ERF_A5;\n\nexport const erfKernelFunc = unaryKernelFunc(\n    Erf,\n    (xi) => {\n      const sign = Math.sign(xi);\n      const v = Math.abs(xi);\n      const t = 1.0 / (1.0 + p * v);\n      return sign *\n          (1.0 -\n           (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t *\n               Math.exp(-v * v));\n    },\n);\n\nexport const erfConfig: KernelConfig = {\n  kernelName: Erf,\n  backendName: 'cpu',\n  kernelFunc: erfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Tensor, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {add} from '../kernels/Add';\nimport {complex} from '../kernels/Complex';\nimport {concat} from '../kernels/Concat';\nimport {divConfig} from '../kernels/Div';\nimport {identity} from '../kernels/Identity';\nimport {imag} from '../kernels/Imag';\nimport {multiply} from '../kernels/Multiply';\nimport {real} from '../kernels/Real';\nimport {slice} from '../kernels/Slice';\nimport {sub} from '../kernels/Sub';\n\n/**\n * Calculate FFT of inner most elements of batch tensor.\n */\nexport function fftBatch(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): TensorInfo {\n  const inputShape = input.shape;\n  const batch = inputShape[0];\n  const innerDim = inputShape[1];\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const real2D = inputVals.complexTensorInfos.real;\n  const imag2D = inputVals.complexTensorInfos.imag;\n\n  // Collects real and imaginary values separately.\n  const resultShape = [batch, innerDim];\n  const resultSize = util.sizeFromShape(resultShape);\n  const resultReal = util.getTypedArrayFromDType('float32', resultSize);\n  const resultImag = util.getTypedArrayFromDType('float32', resultSize);\n\n  for (let b = 0; b < batch; b++) {\n    // TODO: Support slice ops for complex type.\n    const r = slice({\n      inputs: {x: real2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n    const i = slice({\n      inputs: {x: imag2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n\n    const input = complex({inputs: {real: r, imag: i}, backend: cpuBackend});\n\n    // Run FFT by batch element.\n    const {real, imag} = fftImpl(input, inverse, cpuBackend);\n    const res = backend_util.mergeRealAndImagArrays(real, imag);\n\n    for (let d = 0; d < innerDim; d++) {\n      const c = backend_util.getComplexWithIndex(res, d);\n      resultReal[b * innerDim + d] = c.real;\n      resultImag[b * innerDim + d] = c.imag;\n    }\n\n    cpuBackend.disposeIntermediateTensorInfo(r);\n    cpuBackend.disposeIntermediateTensorInfo(i);\n    cpuBackend.disposeIntermediateTensorInfo(input);\n  }\n\n  const $realInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultReal);\n  const $imagInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultImag);\n\n  const result = complex(\n      {inputs: {real: $realInfo, imag: $imagInfo}, backend: cpuBackend});\n\n  cpuBackend.disposeIntermediateTensorInfo($realInfo);\n  cpuBackend.disposeIntermediateTensorInfo($imagInfo);\n\n  return result;\n}\n\nexport function fftImpl(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  const inputSize = util.sizeFromShape(input.shape);\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const realVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.real.dataId).values as\n      Float32Array;\n\n  const imagVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.imag.dataId).values as\n      Float32Array;\n\n  if (isExponentOf2(inputSize)) {\n    const result =\n        fftRadix2(realVals, imagVals, inputSize, inverse, cpuBackend);\n\n    const resultShape = [input.shape[0], input.shape[1]];\n\n    if (inverse) {\n      const realInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.real);\n      const imagInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.imag);\n\n      const sizeInfo: TensorInfo = cpuBackend.makeTensorInfo(\n          [], 'float32',\n          util.createScalarValue(inputSize as {} as 'float32', 'float32'));\n      const sizeInfoCopy =\n          identity({inputs: {x: sizeInfo}, backend: cpuBackend});\n\n      const divRealInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: realInfo, b: sizeInfo}, backend: cpuBackend}) as\n          TensorInfo;\n      const divImagInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: imagInfo, b: sizeInfoCopy}, backend: cpuBackend}) as\n          TensorInfo;\n\n      const divRealVals =\n          cpuBackend.data.get(divRealInfo.dataId).values as Float32Array;\n      const divImagVals =\n          cpuBackend.data.get(divImagInfo.dataId).values as Float32Array;\n\n      cpuBackend.disposeIntermediateTensorInfo(realInfo);\n      cpuBackend.disposeIntermediateTensorInfo(imagInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfoCopy);\n      cpuBackend.disposeIntermediateTensorInfo(divRealInfo);\n      cpuBackend.disposeIntermediateTensorInfo(divImagInfo);\n\n      return {real: divRealVals, imag: divImagVals};\n    }\n\n    return result;\n  } else {\n    const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n    const rawOutput =\n        fourierTransformByMatmul(data, inputSize, inverse) as Float32Array;\n\n    return backend_util.splitRealAndImagArrays(rawOutput);\n  }\n}\n\nfunction isExponentOf2(size: number): boolean {\n  return (size & size - 1) === 0;\n}\n\n// FFT using Cooley-Tukey algorithm on radix 2 dimensional input.\nfunction fftRadix2(\n    realVals: Float32Array, imagVals: Float32Array, size: number,\n    inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  if (size === 1) {\n    return {real: realVals, imag: imagVals};\n  }\n\n  const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n  const half = size / 2;\n\n  const evenComplex = backend_util.complexWithEvenIndex(data);\n\n  const evenRealVals = evenComplex.real;\n  const evenImagVals = evenComplex.imag;\n\n  const evenShape = [evenRealVals.length];\n\n  const evenRealInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenRealVals);\n  const evenImagInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenImagVals);\n\n  const evenTensorInfo = complex(\n      {inputs: {real: evenRealInfo, imag: evenImagInfo}, backend: cpuBackend});\n\n  const oddComplex = backend_util.complexWithOddIndex(data);\n\n  const oddRealVals = oddComplex.real;\n  const oddImagVals = oddComplex.imag;\n\n  const oddShape = [oddRealVals.length];\n\n  const oddRealInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddRealVals);\n  const oddImagInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddImagVals);\n\n  const oddTensorInfo = complex(\n      {inputs: {real: oddRealInfo, imag: oddImagInfo}, backend: cpuBackend});\n\n  // Recursive call for half part of original input.\n  const $evenComplex =\n      fftRadix2(evenRealVals, evenImagVals, half, inverse, cpuBackend);\n\n  const $evenRealVals = $evenComplex.real;\n  const $evenImagVals = $evenComplex.imag;\n\n  const $evenShape = [$evenRealVals.length];\n\n  const $evenRealInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenRealVals);\n  const $evenImagInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenImagVals);\n\n  const $evenTensorInfo = complex({\n    inputs: {real: $evenRealInfo, imag: $evenImagInfo},\n    backend: cpuBackend\n  });\n\n  const $oddComplex =\n      fftRadix2(oddRealVals, oddImagVals, half, inverse, cpuBackend);\n\n  const $oddRealVals = $oddComplex.real;\n  const $oddImagVals = $oddComplex.imag;\n\n  const $oddShape = [$oddRealVals.length];\n\n  const $oddRealInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddRealVals);\n  const $oddImagInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddImagVals);\n\n  const $oddTensorInfo = complex(\n      {inputs: {real: $oddRealInfo, imag: $oddImagInfo}, backend: cpuBackend});\n\n  const e = backend_util.exponents(size, inverse);\n  const eShape = [e.real.length];\n\n  const eRealInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.real);\n  const eImagInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.imag);\n\n  const complexInfo = complex(\n      {inputs: {real: eRealInfo, imag: eImagInfo}, backend: cpuBackend});\n\n  const exponentInfo =\n      multiply(\n          {inputs: {a: complexInfo, b: $oddTensorInfo}, backend: cpuBackend}) as\n      TensorInfo;\n\n  const addPart = add({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n  const subPart = sub({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n\n  const addPartReal = real({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartReal = real({inputs: {input: subPart}, backend: cpuBackend});\n\n  const addPartImag = imag({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartImag = imag({inputs: {input: subPart}, backend: cpuBackend});\n\n  const $real = concat({\n    inputs: [addPartReal as Tensor, subPartReal as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n  const $imag = concat({\n    inputs: [addPartImag as Tensor, subPartImag as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n\n  const $realVals = cpuBackend.data.get($real.dataId).values as Float32Array;\n  const $imagVals = cpuBackend.data.get($imag.dataId).values as Float32Array;\n\n  cpuBackend.disposeIntermediateTensorInfo(evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(complexInfo);\n  cpuBackend.disposeIntermediateTensorInfo(exponentInfo);\n  cpuBackend.disposeIntermediateTensorInfo(addPart);\n  cpuBackend.disposeIntermediateTensorInfo(subPart);\n  cpuBackend.disposeIntermediateTensorInfo(addPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(addPartImag);\n  cpuBackend.disposeIntermediateTensorInfo(subPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(subPartImag);\n  cpuBackend.disposeIntermediateTensorInfo($real);\n  cpuBackend.disposeIntermediateTensorInfo($imag);\n\n  return {real: $realVals, imag: $imagVals};\n}\n\n// Calculate fourier transform by multplying sinusoid matrix.\nfunction fourierTransformByMatmul(\n    data: TypedArray, size: number, inverse: boolean): TypedArray {\n  const ret = new Float32Array(size * 2);\n  // TODO: Use matmul instead once it supports complex64 type.\n  for (let r = 0; r < size; r++) {\n    let real = 0.0;\n    let imag = 0.0;\n    for (let c = 0; c < size; c++) {\n      const e = backend_util.exponent(r * c, size, inverse);\n      const term = backend_util.getComplexWithIndex(data as Float32Array, c);\n      real += term.real * e.real - term.imag * e.imag;\n      imag += term.real * e.imag + term.imag * e.real;\n    }\n    if (inverse) {\n      real /= size;\n      imag /= size;\n    }\n    backend_util.assignToTypedArray(ret, real, imag, r);\n  }\n  return ret;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FFT, FFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function fft(args: {inputs: FFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, false, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const fftConfig: KernelConfig = {\n  kernelName: FFT,\n  backendName: 'cpu',\n  kernelFunc: fft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n\n            const coordX = Math.round(imageWidth - x);\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n\n            let outputValue = imageVals[outIdx];\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth) {\n              // set the output to the image value at the coordinate position.\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n            output[outIdx] = outputValue;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IFFT, IFFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function ifft(args: {inputs: IFFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, true, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const ifftConfig: KernelConfig = {\n  kernelName: IFFT,\n  backendName: 'cpu',\n  kernelFunc: ifft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isFiniteKernelFunc =\n    unaryKernelFunc(IsFinite, (xi) => Number.isFinite(xi) ? 1 : 0, 'bool');\n\nexport const isFiniteConfig: KernelConfig = {\n  kernelName: IsFinite,\n  backendName: 'cpu',\n  kernelFunc: isFiniteKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isInfKernelFunc =\n    unaryKernelFunc(IsInf, (xi) => Math.abs(xi) === Infinity ? 1 : 0, 'bool');\n\nexport const isInfConfig: KernelConfig = {\n  kernelName: IsInf,\n  backendName: 'cpu',\n  kernelFunc: isInfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isNaNKernelFunc =\n    unaryKernelFunc(IsNan, (xi) => Number.isNaN(xi) ? 1 : 0, 'bool');\n\nexport const isNaNConfig: KernelConfig = {\n  kernelName: IsNan,\n  backendName: 'cpu',\n  kernelFunc: isNaNKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log1p} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const log1pKernelFunc = unaryKernelFunc(Log1p, (xi) => Math.log1p(xi));\n\nexport const log1pConfig: KernelConfig = {\n  kernelName: Log1p,\n  backendName: 'cpu',\n  kernelFunc: log1pKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, LogicalNot} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const logicalNotKernelFunc =\n    unaryKernelFunc(LogicalNot, (xi) => xi ? 0 : 1, 'bool');\n\nexport const logicalNotConfig: KernelConfig = {\n  kernelName: LogicalNot,\n  backendName: 'cpu',\n  kernelFunc: logicalNotKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig} from '@tensorflow/tfjs-core';\nimport {TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    let xShape = x.shape;\n    const xRank = xShape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, xShape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    let xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    if (permutedAxes != null) {\n      const newShape: number[] = new Array(xRank);\n      for (let i = 0; i < newShape.length; i++) {\n        newShape[i] = xShape[permutedAxes[i]];\n      }\n\n      xVals = transposeImpl(xVals, xShape, x.dtype, permutedAxes, newShape);\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n\n      xShape = newShape;\n    }\n\n    assertNotComplex(x, 'max');\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(xShape, axes);\n\n    const reduceSize = util.sizeFromShape(reduceShape);\n\n    const result = maxImpl(xVals, reduceSize, maxOutShape, x.dtype);\n    const dataId = cpuBackend.write(result, maxOutShape, x.dtype);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // reshape\n      const newShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n      outShape = newShape;\n    }\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function maxPool(\n    args:\n        {inputs: MaxPoolInputs, backend: MathBackendCPU, attrs: MaxPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'max');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'cpu',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, buffer, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, Rank, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {maxPoolPositions} from '../utils/pool_utils';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const xValues = backend.data.get(x.dataId).values as TypedArray;\n  const maxPosBuf = buffer(\n      convInfo.outShape, x.dtype,\n      maxPoolPositions(xValues, x.shape, x.dtype, convInfo).values);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n              const maxPos = effectiveFilterHeight * effectiveFilterWidth - 1 -\n                  (maxPosBuf.get(b, dyR, dyC, d) as number);\n              const curPos = wR * effectiveFilterWidth + wC;\n\n              const mask = maxPos === curPos ? 1 : 0;\n              if (mask === 0) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel * mask;\n            }\n          }\n          dx.set(dotProd, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, DataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {maxPoolPositions, pool} from '../utils/pool_utils';\nexport function maxPoolWithArgmaxImpl(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    includeBatchInIndex: boolean, convInfo: backend_util.Conv2DInfo) {\n  const strides = util.computeStrides(xShape);\n  const maxPools = pool(xValues, xShape, dtype, strides, convInfo, 'max');\n  const maxPositions = maxPoolPositions(\n      xValues, xShape, dtype, convInfo, true, includeBatchInIndex);\n\n  return [maxPools.values, maxPositions.values];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'MaxPoolWithArgmax');\n\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        [1, 1], pad);\n    const [pooled, indexes] = maxPoolWithArgmaxImpl(\n        values, x.shape, x.dtype, includeBatchInIndex, convInfo);\n\n    const pooledDataId =\n        cpuBackend.write(pooled as Float32Array, convInfo.outShape, x.dtype);\n    const indexesDataId =\n        cpuBackend.write(indexes as Int32Array, convInfo.outShape, x.dtype);\n    return [\n      {dataId: pooledDataId, shape: convInfo.outShape, dtype: x.dtype},\n      {dataId: indexesDataId, shape: convInfo.outShape, dtype: 'int32'}\n    ];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionPadded');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionWithScore');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NotEqual} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const notEqualImpl =\n    createSimpleBinaryKernelImpl(((a, b) => (a !== b) ? 1 : 0));\nexport const notEqual =\n    binaryKernelFunc(NotEqual, notEqualImpl, null /* complexOp */, 'bool');\n\nexport const notEqualConfig: KernelConfig = {\n  kernelName: NotEqual,\n  backendName: 'cpu',\n  kernelFunc: notEqual\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, NumericDataType, PadV2, PadV2Attrs, PadV2Inputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function padV2(\n    args: {inputs: PadV2Inputs, backend: MathBackendCPU, attrs: PadV2Attrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {paddings, constantValue} = attrs;\n\n  assertNotComplex(x, 'pad');\n\n  const outShape = paddings.map(\n      (p, i) => p[0] /* beforePad */ + x.shape[i] + p[1] /* afterPad */);\n\n  const start = paddings.map(p => p[0]);\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const xSize = util.sizeFromShape(x.shape);\n  const xRank = x.shape.length;\n  const xStrides = util.computeStrides(x.shape);\n\n  const resultSize = util.sizeFromShape(outShape);\n  const resultRank = outShape.length;\n  const resultStrides = util.computeStrides(outShape);\n  const resVals =\n      util.getTypedArrayFromDType(x.dtype as NumericDataType, resultSize);\n\n  if (constantValue !== 0) {\n    resVals.fill(constantValue);\n  }\n\n  for (let i = 0; i < xSize; i++) {\n    const coords = util.indexToLoc(i, xRank, xStrides);\n    const outCoords = coords.map((c, i) => c + start[i]);\n    const outIndex = util.locToIndex(outCoords, resultRank, resultStrides);\n\n    resVals[outIndex] = xVals[i];\n  }\n\n  const outId = backend.write(resVals, outShape, x.dtype);\n\n  return {dataId: outId, shape: outShape, dtype: x.dtype};\n}\n\nexport const padV2Config: KernelConfig = {\n  kernelName: PadV2,\n  backendName: 'cpu',\n  kernelFunc: padV2 as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Reciprocal} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const reciprocalKernelFunc = unaryKernelFunc(Reciprocal, (xi) => 1 / xi);\n\nexport const reciprocalConfig: KernelConfig = {\n  kernelName: Reciprocal,\n  backendName: 'cpu',\n  kernelFunc: reciprocalKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {backend_util, RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const fullOpacityValue = 255;\n\n    const sinFactor = Math.sin(radians);\n    const cosFactor = Math.cos(radians);\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n            const y = coords[1];\n\n            // coordX/coordY are the result of rotating and translating x/y.\n            let coordX = (x - centerX) * cosFactor - (y - centerY) * sinFactor;\n            let coordY = (x - centerX) * sinFactor + (y - centerY) * cosFactor;\n            coordX = Math.round(coordX + centerX);\n            coordY = Math.round(coordY + centerY);\n\n            let outputValue = fillValue;\n            if (typeof fillValue !== 'number') {\n              if (channel === 3) {\n                outputValue = fullOpacityValue;\n              } else {\n                outputValue = fillValue[channel];\n              }\n            }\n\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth && coordY >= 0 &&\n                coordY < imageHeight) {\n              // set the output to the image value at the coordinate position.\n              const rotatedRowOffset = coordY * (imageWidth * numChannels);\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rotatedRowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n            output[outIdx] = outputValue as number;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Round} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const roundKernelFunc = unaryKernelFunc(Round, (xi) => {\n  // The algorithm is based on banker's rounding.\n  const base = Math.floor(xi);\n  if (xi - base < 0.5) {\n    return Math.floor(xi);\n  } else if (xi - base > 0.5) {\n    return Math.ceil(xi);\n  } else {\n    if (base % 2.0 === 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n});\n\nexport const roundConfig: KernelConfig = {\n  kernelName: Round,\n  backendName: 'cpu',\n  kernelFunc: roundKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, Selu} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst scaleAlpha = backend_util.SELU_SCALEALPHA;\nconst scale = backend_util.SELU_SCALE;\n\nexport const seluKernelFunc = unaryKernelFunc(Selu, (xi) => {\n  if (xi >= 0) {\n    return scale * xi;\n  } else {\n    return scaleAlpha * (Math.exp(xi) - 1);\n  }\n});\n\nexport const seluConfig: KernelConfig = {\n  kernelName: Selu,\n  backendName: 'cpu',\n  kernelFunc: seluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sigmoid} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sigmoidKernelFunc =\n    unaryKernelFunc(Sigmoid, (xi) => 1 / (1 + Math.exp(-xi)));\n\nexport const sigmoidConfig: KernelConfig = {\n  kernelName: Sigmoid,\n  backendName: 'cpu',\n  kernelFunc: sigmoidKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sign} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const signKernelFunc = unaryKernelFunc(Sign, (xi) => {\n  if (xi < 0) {\n    return -1;\n  } else if (xi > 0) {\n    return 1;\n  } else {\n    return 0;\n  }\n});\n\nexport const signConfig: KernelConfig = {\n  kernelName: Sign,\n  backendName: 'cpu',\n  kernelFunc: signKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinKernelFunc = unaryKernelFunc(Sin, (xi) => Math.sin(xi));\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'cpu',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sinh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinhKernelFunc = unaryKernelFunc(Sinh, (xi) => Math.sinh(xi));\n\nexport const sinhConfig: KernelConfig = {\n  kernelName: Sinh,\n  backendName: 'cpu',\n  kernelFunc: sinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Softplus} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\n// mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n\n// epsilon is the difference between 1.0 and the next representable float.\n// For a single precision 32 bit float this should be 2^-23, see:\n// https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\nconst epsilon = 1.1920928955078125e-7;\nconst threshold = Math.log(epsilon) + 2.0;\n\nexport const softplusKernelFunc = unaryKernelFunc(Softplus, (xi) => {\n  // Value above which exp(x) may overflow, but softplus(x) == x\n  // is within machine epsilon.\n  const tooLarge = xi > -threshold;\n\n  // Value below which exp(x) may underflow, but softplus(x) == exp(x)\n  // is within machine epsilon.\n  const tooSmall = xi < threshold;\n\n  const expX = Math.exp(xi);\n  let result;\n\n  if (tooSmall) {\n    result = expX;\n  } else if (tooLarge) {\n    result = xi;\n  } else {\n    result = Math.log(1.0 + expX);\n  }\n  return result;\n});\n\nexport const softplusConfig: KernelConfig = {\n  kernelName: Softplus,\n  backendName: 'cpu',\n  kernelFunc: softplusKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {transposeImpl} from './Transpose_impl';\n\nexport function transpose(args: {\n  inputs: TransposeInputs,\n  attrs: TransposeAttrs,\n  backend: MathBackendCPU\n}): TensorInfo {\n  const {inputs, attrs, backend} = args;\n  const {x} = inputs;\n  const {perm} = attrs;\n\n  assertNotComplex(x, 'transpose');\n\n  const xRank = x.shape.length;\n\n  const newShape: number[] = new Array(xRank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = x.shape[perm[i]];\n  }\n\n  const values = backend.data.get(x.dataId).values as TypedArray;\n  const result = transposeImpl(values, x.shape, x.dtype, perm, newShape);\n\n  const dataId = backend.write(result, newShape, x.dtype);\n  return {dataId, shape: newShape, dtype: x.dtype};\n}\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'cpu',\n  kernelFunc: transpose as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, KernelFunc, ReshapeAttrs, ReshapeInputs, SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs, TensorInfo, TransposeAttrs, TransposeInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {padV2Config} from './PadV2';\nimport {reshape} from './Reshape';\nimport {transpose} from './Transpose';\n\nexport function spaceToBatchND(args: {\n  inputs: SpaceToBatchNDInputs,\n  backend: MathBackendCPU,\n  attrs: SpaceToBatchNDAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {blockShape, paddings} = attrs;\n\n  assertNotComplex([x], 'spaceToBatchND');\n\n  const prod = util.sizeFromShape(blockShape);\n\n  const completePaddings: Array<[number, number]> = [[0, 0]];\n  completePaddings.push(...(paddings as Array<[number, number]>));\n\n  for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n    completePaddings.push([0, 0]);\n  }\n\n  const paddedX = padV2Config.kernelFunc({\n    inputs: {x},\n    backend,\n    attrs: {paddings: completePaddings, constantValue: 0}\n  }) as TensorInfo;\n\n  const reshapedPaddedShape =\n      backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n  const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n      reshapedPaddedShape.length, blockShape.length, false);\n\n  const flattenShape =\n      backend_util.getReshapedPermuted(paddedX.shape, blockShape, prod, false);\n\n  const reshapeInputs: ReshapeInputs = {x: paddedX};\n  const reshapeAttrs: ReshapeAttrs = {shape: reshapedPaddedShape};\n  const paddedXReshaped =\n      reshape({inputs: reshapeInputs, backend, attrs: reshapeAttrs});\n\n  const transposeInputs: TransposeInputs = {x: paddedXReshaped};\n  const transposeAttrs:\n      TransposeAttrs = {perm: permutedReshapedPaddedPermutation};\n  const paddedXT =\n      transpose({inputs: transposeInputs, backend, attrs: transposeAttrs});\n\n  const resultReshapeInputs: ReshapeInputs = {x: paddedXT};\n  const resultReshapeAttrs: ReshapeAttrs = {shape: flattenShape};\n  const result = reshape(\n      {inputs: resultReshapeInputs, backend, attrs: resultReshapeAttrs});\n\n  backend.disposeIntermediateTensorInfo(paddedX);\n  backend.disposeIntermediateTensorInfo(paddedXReshaped);\n  backend.disposeIntermediateTensorInfo(paddedXT);\n\n  return result;\n}\n\nexport const spaceToBatchNDConfig: KernelConfig = {\n  kernelName: SpaceToBatchND,\n  backendName: 'cpu',\n  kernelFunc: spaceToBatchND as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sqrt} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sqrtKernelFunc = unaryKernelFunc(Sqrt, (xi) => Math.sqrt(xi));\n\nexport const sqrtConfig: KernelConfig = {\n  kernelName: Sqrt,\n  backendName: 'cpu',\n  kernelFunc: sqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square, SquareInputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend}) => {\n    const {x} = inputs as SquareInputs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'square');\n\n    const values = cpuBackend.data.get(x.dataId).values as Float32Array;\n    const newValues = new Float32Array(values.length);\n    for (let i = 0; i < values.length; ++i) {\n      const value = values[i];\n      newValues[i] = value * value;\n    }\n    const dataId = cpuBackend.write(newValues, x.shape, x.dtype);\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const squaredDifferenceImpl = createSimpleBinaryKernelImpl(((a, b) => {\n  const diff = a - b;\n  return diff * diff;\n}));\nexport const squaredDifference =\n    binaryKernelFunc(SquaredDifference, squaredDifferenceImpl);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'cpu',\n  kernelFunc: squaredDifference\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Step, StepAttrs} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const stepKernelFunc = unaryKernelFunc(Step, (xi, attrs) => {\n  const stepAttrs = attrs as {} as StepAttrs;\n  if (isNaN(xi)) {\n    return NaN;\n  } else {\n    return xi > 0 ? 1 : stepAttrs.alpha;\n  }\n});\n\nexport const stepConfig: KernelConfig = {\n  kernelName: Step,\n  backendName: 'cpu',\n  kernelFunc: stepKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanKernelFunc = unaryKernelFunc(Tan, (xi) => Math.tan(xi));\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'cpu',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tanh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanhKernelFunc = unaryKernelFunc(Tanh, (xi) => Math.tanh(xi));\n\nexport const tanhConfig: KernelConfig = {\n  kernelName: Tanh,\n  backendName: 'cpu',\n  kernelFunc: tanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {uniqueImpl} from './Unique_impl';\n\nexport function unique(\n    args: {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendCPU}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  const values = backend.data.get(x.dataId).values;\n  const {outputValues, outputShape, indices} =\n      uniqueImpl(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'cpu',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// We explicitly import the modular kernels so they get registered in the\n// global registry when we compile the library. A modular build would replace\n// the contents of this file and import only the kernels that are needed.\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {absConfig} from './kernels/Abs';\nimport {acosConfig} from './kernels/Acos';\nimport {acoshConfig} from './kernels/Acosh';\nimport {addConfig} from './kernels/Add';\nimport {asinConfig} from './kernels/Asin';\nimport {asinhConfig} from './kernels/Asinh';\nimport {atanConfig} from './kernels/Atan';\nimport {atanhConfig} from './kernels/Atanh';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {castConfig} from './kernels/Cast';\nimport {ceilConfig} from './kernels/Ceil';\nimport {clipConfig} from './kernels/Clip';\nimport {complexConfig} from './kernels/Complex';\nimport {concatConfig} from './kernels/Concat';\nimport {cosConfig} from './kernels/Cos';\nimport {coshConfig} from './kernels/Cosh';\nimport {dilation2dConfig} from './kernels/Dilation2D';\nimport {dilation2dBackpropFilterConfig} from './kernels/Dilation2DBackpropFilter';\nimport {dilation2dBackpropInputConfig} from './kernels/Dilation2DBackpropInput';\nimport {divConfig} from './kernels/Div';\nimport {eluConfig} from './kernels/Elu';\nimport {erfConfig} from './kernels/Erf';\nimport {expConfig} from './kernels/Exp';\nimport {expm1Config} from './kernels/Expm1';\nimport {fftConfig} from './kernels/FFT';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {floorConfig} from './kernels/Floor';\nimport {identityConfig} from './kernels/Identity';\nimport {ifftConfig} from './kernels/IFFT';\nimport {imagConfig} from './kernels/Imag';\nimport {isFiniteConfig} from './kernels/IsFinite';\nimport {isInfConfig} from './kernels/IsInf';\nimport {isNaNConfig} from './kernels/IsNaN';\nimport {logConfig} from './kernels/Log';\nimport {log1pConfig} from './kernels/Log1p';\nimport {logicalNotConfig} from './kernels/LogicalNot';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {multiplyConfig} from './kernels/Multiply';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {notEqualConfig} from './kernels/NotEqual';\nimport {padV2Config} from './kernels/PadV2';\nimport {realConfig} from './kernels/Real';\nimport {reciprocalConfig} from './kernels/Reciprocal';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {roundConfig} from './kernels/Round';\nimport {rsqrtConfig} from './kernels/Rsqrt';\nimport {seluConfig} from './kernels/Selu';\nimport {sigmoidConfig} from './kernels/Sigmoid';\nimport {signConfig} from './kernels/Sign';\nimport {sinConfig} from './kernels/Sin';\nimport {sinhConfig} from './kernels/Sinh';\nimport {sliceConfig} from './kernels/Slice';\nimport {softplusConfig} from './kernels/Softplus';\nimport {spaceToBatchNDConfig} from './kernels/SpaceToBatchND';\nimport {sqrtConfig} from './kernels/Sqrt';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {stepConfig} from './kernels/Step';\nimport {subConfig} from './kernels/Sub';\nimport {tanConfig} from './kernels/Tan';\nimport {tanhConfig} from './kernels/Tanh';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  absConfig,\n  acosConfig,\n  acoshConfig,\n  addConfig,\n  asinConfig,\n  asinhConfig,\n  atanConfig,\n  atanhConfig,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  castConfig,\n  ceilConfig,\n  clipConfig,\n  complexConfig,\n  concatConfig,\n  cosConfig,\n  coshConfig,\n  dilation2dConfig,\n  dilation2dBackpropInputConfig,\n  dilation2dBackpropFilterConfig,\n  divConfig,\n  eluConfig,\n  erfConfig,\n  expConfig,\n  expm1Config,\n  fftConfig,\n  flipLeftRightConfig,\n  floorConfig,\n  identityConfig,\n  ifftConfig,\n  imagConfig,\n  isFiniteConfig,\n  isInfConfig,\n  isNaNConfig,\n  logConfig,\n  log1pConfig,\n  logicalNotConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  maxConfig,\n  multiplyConfig,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  notEqualConfig,\n  padV2Config,\n  realConfig,\n  reciprocalConfig,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  roundConfig,\n  rsqrtConfig,\n  seluConfig,\n  sigmoidConfig,\n  signConfig,\n  sinConfig,\n  sinhConfig,\n  sliceConfig,\n  softplusConfig,\n  spaceToBatchNDConfig,\n  sqrtConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  stepConfig,\n  subConfig,\n  tanConfig,\n  tanhConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst contexts: {[key: string]: WebGLRenderingContext} = {};\n\nconst WEBGL_ATTRIBUTES: WebGLContextAttributes = {\n  alpha: false,\n  antialias: false,\n  premultipliedAlpha: false,\n  preserveDrawingBuffer: false,\n  depth: false,\n  stencil: false,\n  failIfMajorPerformanceCaveat: true\n};\n\nexport function setWebGLContext(\n    webGLVersion: number, gl: WebGLRenderingContext) {\n  contexts[webGLVersion] = gl;\n}\n\nexport function getWebGLContext(webGLVersion: number): WebGLRenderingContext {\n  if (!(webGLVersion in contexts)) {\n    const newCtx = getWebGLRenderingContext(webGLVersion);\n    if (newCtx !== null) {\n      contexts[webGLVersion] = newCtx;\n    } else {\n      console.log('Could not get context for WebGL version', webGLVersion);\n      return null;\n    }\n  }\n  const gl = contexts[webGLVersion];\n  if (gl.isContextLost()) {\n    delete contexts[webGLVersion];\n    return getWebGLContext(webGLVersion);\n  }\n\n  gl.disable(gl.DEPTH_TEST);\n  gl.disable(gl.STENCIL_TEST);\n  gl.disable(gl.BLEND);\n  gl.disable(gl.DITHER);\n  gl.disable(gl.POLYGON_OFFSET_FILL);\n  gl.disable(gl.SAMPLE_COVERAGE);\n  gl.enable(gl.SCISSOR_TEST);\n  gl.enable(gl.CULL_FACE);\n  gl.cullFace(gl.BACK);\n\n  return contexts[webGLVersion];\n}\n\nfunction createCanvas(webGLVersion: number) {\n  if (typeof OffscreenCanvas !== 'undefined' && webGLVersion === 2) {\n    return new OffscreenCanvas(300, 150);\n  } else if (typeof document !== 'undefined') {\n    return document.createElement('canvas');\n  } else {\n    throw new Error('Cannot create a canvas in this context');\n  }\n}\n\nfunction getWebGLRenderingContext(webGLVersion: number): WebGLRenderingContext {\n  if (webGLVersion !== 1 && webGLVersion !== 2) {\n    throw new Error('Cannot get WebGL rendering context, WebGL is disabled.');\n  }\n  const canvas = createCanvas(webGLVersion);\n\n  canvas.addEventListener('webglcontextlost', (ev: Event) => {\n    ev.preventDefault();\n    delete contexts[webGLVersion];\n  }, false);\n  if (webGLVersion === 1) {\n    return (canvas.getContext('webgl', WEBGL_ATTRIBUTES) ||\n            canvas.getContext('experimental-webgl', WEBGL_ATTRIBUTES)) as\n        WebGLRenderingContext;\n  }\n  return canvas.getContext('webgl2', WEBGL_ATTRIBUTES) as WebGLRenderingContext;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {backend_util, DataId, DataType, Tensor, util} from '@tensorflow/tfjs-core';\n\nexport enum PackingScheme {\n  /**\n   * All values in a single texel are densely packed without any constraints.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 4]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   010|011   020|021\n   * -------   -------   -------\n   * 002|003   012|013   022|023\n   *\n   * 100|101   110|111   120|121\n   * -------   -------   -------\n   * 102|103   112|113   122|123\n   *\n   */\n  DENSE,\n\n  /**\n   * Single texels contain only values from the same batch, and from adjacent\n   * rows and columns.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 5]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   002|003   004|xxx   020|021   022|023   024|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   * 100|101   102|103   104|xxx   120|121   122|123   124|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   */\n  SHARED_BATCH\n}\n\nexport enum TextureUsage {\n  RENDER,\n  UPLOAD,\n  PIXELS,\n  DOWNLOAD\n}\n\nexport enum PhysicalTextureType {\n  UNPACKED_FLOAT16,\n  UNPACKED_FLOAT32,\n  PACKED_4X1_UNSIGNED_BYTE,\n  PACKED_2X2_FLOAT32,\n  PACKED_2X2_FLOAT16\n}\n\nexport interface TextureData {\n  // Required.\n  shape: number[];\n  dtype: DataType;\n\n  // Optional.\n  values?: backend_util.BackendValues;\n  texture?: WebGLTexture;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensors field. When this is defined, texture will be null.\n  complexTensors?: {real: Tensor, imag: Tensor};\n  /** [rows, columns] shape of the texture. */\n  texShape?: [number, number];\n  usage?: TextureUsage;\n  isPacked?: boolean;\n\n  refCount: number;\n\n  // Available when the tensor has been sliced.\n  slice?: {\n    // Offset in the 'flat index' space.\n    flatOffset: number;\n    // Used for counting how many sliced tensors point to the same texture.\n    origDataId: DataId;\n  };\n}\n\nexport function getUnpackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns, rows];\n}\n\nexport function getUnpackedArraySizeFromMatrixSize(\n    matrixSize: number, channelsPerTexture: number): number {\n  return matrixSize * channelsPerTexture;\n}\n\nexport function getColorMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns * 4, rows];\n}\n\n/**\n * Get shape for densely packed RGBA texture.\n */\nexport function getDenseTexShape(shape: number[]): [number, number] {\n  const size = util.sizeFromShape(shape);\n  const texelsNeeded = Math.ceil(size / 4);\n  return util.sizeToSquarishShape(texelsNeeded);\n}\n\nexport function getMatrixSizeFromUnpackedArraySize(\n    unpackedSize: number, channelsPerTexture: number): number {\n  if (unpackedSize % channelsPerTexture !== 0) {\n    throw new Error(\n        `unpackedSize (${unpackedSize}) must be a multiple of ` +\n        `${channelsPerTexture}`);\n  }\n  return unpackedSize / channelsPerTexture;\n}\n\nexport function decodeMatrixFromUnpackedColorRGBAArray(\n    unpackedArray: Float32Array, matrix: Float32Array, channels: number) {\n  const requiredSize = unpackedArray.length * channels / 4;\n  if (matrix.length < requiredSize) {\n    throw new Error(\n        `matrix length (${matrix.length}) must be >= ${requiredSize}`);\n  }\n  let dst = 0;\n  for (let src = 0; src < unpackedArray.length; src += 4) {\n    for (let c = 0; c < channels; c++) {\n      matrix[dst++] = unpackedArray[src + c];\n    }\n  }\n}\n\nexport function getPackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [\n    Math.max(1, Math.ceil(columns / 2)), Math.max(1, Math.ceil(rows / 2))\n  ];\n}\n\nexport function getPackedRGBAArraySizeFromMatrixShape(\n    rows: number, columns: number): number {\n  const [w, h] = getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return w * h * 4;\n}\n\nexport interface TextureConfig {\n  internalFormatFloat: number;\n  textureFormatFloat: number;\n  internalFormatPackedHalfFloat: number;\n  internalFormatHalfFloat: number;\n  internalFormatPackedFloat: number;\n\n  // The format to use during a gl.readPixels call.\n  downloadTextureFormat: number;\n  // How many channels need to be unpacked after a gl.readPixels call.\n  downloadUnpackNumChannels: number;\n\n  defaultNumChannels: number;\n  textureTypeHalfFloat: number;\n  textureTypeFloat: number;\n}\n\nexport function getTextureConfig(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension?: any): TextureConfig {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n\n  let internalFormatFloat: number;\n  let internalFormatHalfFloat: number;\n  let internalFormatPackedHalfFloat: number;\n  let internalFormatPackedFloat: number;\n  let textureFormatFloat: number;\n\n  let downloadTextureFormat: number;\n  let downloadUnpackNumChannels: number;\n\n  let defaultNumChannels: number;\n  let textureTypeHalfFloat: number;\n  let textureTypeFloat: number;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    internalFormatFloat = glany.R32F;\n    internalFormatHalfFloat = glany.R16F;\n    internalFormatPackedHalfFloat = glany.RGBA16F;\n    internalFormatPackedFloat = glany.RGBA32F;\n    textureFormatFloat = glany.RED;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 1;\n    textureTypeHalfFloat = glany.HALF_FLOAT;\n    textureTypeFloat = glany.FLOAT;\n  } else {\n    internalFormatFloat = gl.RGBA;\n    internalFormatHalfFloat = gl.RGBA;\n    internalFormatPackedHalfFloat = gl.RGBA;\n    internalFormatPackedFloat = glany.RGBA;\n    textureFormatFloat = gl.RGBA;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 4;\n    textureTypeHalfFloat = textureHalfFloatExtension != null ?\n        textureHalfFloatExtension.HALF_FLOAT_OES :\n        null;\n    textureTypeFloat = gl.FLOAT;\n  }\n  downloadTextureFormat = gl.RGBA;\n\n  return {\n    internalFormatFloat,\n    internalFormatHalfFloat,\n    internalFormatPackedHalfFloat,\n    internalFormatPackedFloat,\n    textureFormatFloat,\n    downloadTextureFormat,\n    downloadUnpackNumChannels,\n    defaultNumChannels,\n    textureTypeHalfFloat,\n    textureTypeFloat\n  };\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext} from './canvas_util';\nimport {getTextureConfig} from './tex_util';\n\nexport function callAndCheck<T>(gl: WebGLRenderingContext, func: () => T): T {\n  const returnValue = func();\n  if (env().getBool('DEBUG')) {\n    checkWebGLError(gl);\n  }\n  return returnValue;\n}\n\nfunction checkWebGLError(gl: WebGLRenderingContext) {\n  const error = gl.getError();\n  if (error !== gl.NO_ERROR) {\n    throw new Error('WebGL Error: ' + getWebGLErrorMessage(gl, error));\n  }\n}\n\n// https://en.wikipedia.org/wiki/Half-precision_floating-point_format\nconst MIN_FLOAT16 = 5.96e-8;\nconst MAX_FLOAT16 = 65504;\n\nexport function canBeRepresented(num: number): boolean {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED') || num === 0 ||\n      (MIN_FLOAT16 < Math.abs(num) && Math.abs(num) < MAX_FLOAT16)) {\n    return true;\n  }\n  return false;\n}\n\nexport function getWebGLErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.NO_ERROR:\n      return 'NO_ERROR';\n    case gl.INVALID_ENUM:\n      return 'INVALID_ENUM';\n    case gl.INVALID_VALUE:\n      return 'INVALID_VALUE';\n    case gl.INVALID_OPERATION:\n      return 'INVALID_OPERATION';\n    case gl.INVALID_FRAMEBUFFER_OPERATION:\n      return 'INVALID_FRAMEBUFFER_OPERATION';\n    case gl.OUT_OF_MEMORY:\n      return 'OUT_OF_MEMORY';\n    case gl.CONTEXT_LOST_WEBGL:\n      return 'CONTEXT_LOST_WEBGL';\n    default:\n      return `Unknown error code ${status}`;\n  }\n}\n\nexport function getExtensionOrThrow(\n    gl: WebGLRenderingContext, extensionName: string): {} {\n  return throwIfNull<{}>(\n      gl, () => gl.getExtension(extensionName),\n      'Extension \"' + extensionName + '\" not supported on this browser.');\n}\n\nexport function createVertexShader(\n    gl: WebGLRenderingContext, vertexShaderSource: string): WebGLShader {\n  const vertexShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.VERTEX_SHADER),\n      'Unable to create vertex WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(vertexShader, vertexShaderSource));\n  callAndCheck(gl, () => gl.compileShader(vertexShader));\n  if (gl.getShaderParameter(vertexShader, gl.COMPILE_STATUS) === false) {\n    console.log(gl.getShaderInfoLog(vertexShader));\n    throw new Error('Failed to compile vertex shader.');\n  }\n  return vertexShader;\n}\n\nexport function createFragmentShader(\n    gl: WebGLRenderingContext, fragmentShaderSource: string): WebGLShader {\n  const fragmentShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.FRAGMENT_SHADER),\n      'Unable to create fragment WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(fragmentShader, fragmentShaderSource));\n  callAndCheck(gl, () => gl.compileShader(fragmentShader));\n  if (gl.getShaderParameter(fragmentShader, gl.COMPILE_STATUS) === false) {\n    logShaderSourceAndInfoLog(\n        fragmentShaderSource, gl.getShaderInfoLog(fragmentShader));\n    throw new Error('Failed to compile fragment shader.');\n  }\n  return fragmentShader;\n}\n\nconst lineNumberRegex = /ERROR: [0-9]+:([0-9]+):/g;\nfunction logShaderSourceAndInfoLog(\n    shaderSource: string, shaderInfoLog: string) {\n  const lineNumberRegexResult = lineNumberRegex.exec(shaderInfoLog);\n  if (lineNumberRegexResult == null) {\n    console.log(`Couldn't parse line number in error: ${shaderInfoLog}`);\n    console.log(shaderSource);\n    return;\n  }\n\n  const lineNumber = +lineNumberRegexResult[1];\n\n  const shaderLines = shaderSource.split('\\n');\n  const pad = shaderLines.length.toString().length + 2;\n  const linesWithLineNumbers = shaderLines.map(\n      (line, lineNumber) =>\n          util.rightPad((lineNumber + 1).toString(), pad) + line);\n  let maxLineLength = 0;\n  for (let i = 0; i < linesWithLineNumbers.length; i++) {\n    maxLineLength = Math.max(linesWithLineNumbers[i].length, maxLineLength);\n  }\n\n  const beforeErrorLines = linesWithLineNumbers.slice(0, lineNumber - 1);\n  const errorLine = linesWithLineNumbers.slice(lineNumber - 1, lineNumber);\n  const afterErrorLines = linesWithLineNumbers.slice(lineNumber);\n\n  console.log(beforeErrorLines.join('\\n'));\n  console.log(shaderInfoLog.split('\\n')[0]);\n  console.log(\n      `%c ${util.rightPad(errorLine[0], maxLineLength)}`,\n      'border:1px solid red; background-color:#e3d2d2; color:#a61717');\n  console.log(afterErrorLines.join('\\n'));\n}\n\nexport function createProgram(gl: WebGLRenderingContext): WebGLProgram {\n  return throwIfNull<WebGLProgram>(\n      gl, () => gl.createProgram(), 'Unable to create WebGLProgram.');\n}\n\nexport function linkProgram(gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.linkProgram(program));\n  if (gl.getProgramParameter(program, gl.LINK_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Failed to link vertex and fragment shaders.');\n  }\n}\n\nexport function validateProgram(\n    gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.validateProgram(program));\n  if (gl.getProgramParameter(program, gl.VALIDATE_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Shader program validation failed.');\n  }\n}\n\nexport function createStaticVertexBuffer(\n    gl: WebGLRenderingContext, data: Float32Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(gl, () => gl.bufferData(gl.ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function createStaticIndexBuffer(\n    gl: WebGLRenderingContext, data: Uint16Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl, () => gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function getNumChannels(): number {\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    return 1;\n  }\n  return 4;\n}\n\nexport function createTexture(gl: WebGLRenderingContext): WebGLTexture {\n  return throwIfNull<WebGLTexture>(\n      gl, () => gl.createTexture(), 'Unable to create WebGLTexture.');\n}\n\nexport function validateTextureSize(width: number, height: number) {\n  const maxTextureSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if ((width <= 0) || (height <= 0)) {\n    const requested = `[${width}x${height}]`;\n    throw new Error('Requested texture size ' + requested + ' is invalid.');\n  }\n  if ((width > maxTextureSize) || (height > maxTextureSize)) {\n    const requested = `[${width}x${height}]`;\n    const max = `[${maxTextureSize}x${maxTextureSize}]`;\n    throw new Error(\n        'Requested texture size ' + requested +\n        ' greater than WebGL maximum on this browser / GPU ' + max + '.');\n  }\n}\n\nexport function createFramebuffer(gl: WebGLRenderingContext): WebGLFramebuffer {\n  return throwIfNull<WebGLFramebuffer>(\n      gl, () => gl.createFramebuffer(), 'Unable to create WebGLFramebuffer.');\n}\n\nexport function bindVertexBufferToProgramAttribute(\n    gl: WebGLRenderingContext, program: WebGLProgram, attribute: string,\n    buffer: WebGLBuffer, arrayEntriesPerItem: number, itemStrideInBytes: number,\n    itemOffsetInBytes: number): boolean {\n  const loc = gl.getAttribLocation(program, attribute);\n  if (loc === -1) {\n    // The GPU compiler decided to strip out this attribute because it's unused,\n    // thus no need to bind.\n    return false;\n  }\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl,\n      () => gl.vertexAttribPointer(\n          loc, arrayEntriesPerItem, gl.FLOAT, false, itemStrideInBytes,\n          itemOffsetInBytes));\n  callAndCheck(gl, () => gl.enableVertexAttribArray(loc));\n  return true;\n}\n\nexport function bindTextureUnit(\n    gl: WebGLRenderingContext, texture: WebGLTexture, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n}\n\nexport function unbindTextureUnit(\n    gl: WebGLRenderingContext, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function getProgramUniformLocationOrThrow(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return throwIfNull<WebGLUniformLocation>(\n      gl, () => gl.getUniformLocation(program, uniformName),\n      'uniform \"' + uniformName + '\" not present in program.');\n}\n\nexport function getProgramUniformLocation(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return gl.getUniformLocation(program, uniformName);\n}\n\nexport function bindTextureToProgramUniformSampler(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    uniformSamplerLocation: WebGLUniformLocation, textureUnit: number) {\n  callAndCheck(gl, () => bindTextureUnit(gl, texture, textureUnit));\n  callAndCheck(gl, () => gl.uniform1i(uniformSamplerLocation, textureUnit));\n}\n\nexport function bindCanvasToFramebuffer(gl: WebGLRenderingContext) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n  callAndCheck(gl, () => gl.viewport(0, 0, gl.canvas.width, gl.canvas.height));\n  callAndCheck(gl, () => gl.scissor(0, 0, gl.canvas.width, gl.canvas.height));\n}\n\nexport function bindColorTextureToFramebuffer(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0));\n}\n\nexport function unbindColorTextureFromFramebuffer(\n    gl: WebGLRenderingContext, framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, null, 0));\n}\n\nexport function validateFramebuffer(gl: WebGLRenderingContext) {\n  const status = gl.checkFramebufferStatus(gl.FRAMEBUFFER);\n  if (status !== gl.FRAMEBUFFER_COMPLETE) {\n    throw new Error(\n        'Error binding framebuffer: ' + getFramebufferErrorMessage(gl, status));\n  }\n}\n\nexport function getFramebufferErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:\n      return 'FRAMEBUFFER_INCOMPLETE_DIMENSIONS';\n    case gl.FRAMEBUFFER_UNSUPPORTED:\n      return 'FRAMEBUFFER_UNSUPPORTED';\n    default:\n      return `unknown error ${status}`;\n  }\n}\n\nfunction throwIfNull<T>(\n    gl: WebGLRenderingContext, returnTOrNull: () => T | null,\n    failureMessage: string): T {\n  const tOrNull: T|null = callAndCheck(gl, () => returnTOrNull());\n  if (tOrNull == null) {\n    throw new Error(failureMessage);\n  }\n  return tOrNull;\n}\n\nfunction validateTextureUnit(gl: WebGLRenderingContext, textureUnit: number) {\n  const maxTextureUnit = gl.MAX_COMBINED_TEXTURE_IMAGE_UNITS - 1;\n  const glTextureUnit = textureUnit + gl.TEXTURE0;\n  if (glTextureUnit < gl.TEXTURE0 || glTextureUnit > maxTextureUnit) {\n    const textureUnitRange = `[gl.TEXTURE0, gl.TEXTURE${maxTextureUnit}]`;\n    throw new Error(`textureUnit must be in ${textureUnitRange}.`);\n  }\n}\n\nexport function getBatchDim(shape: number[], dimsToSkip = 2): number {\n  return util.sizeFromShape(shape.slice(0, shape.length - dimsToSkip));\n}\n\nexport function getRowsCols(shape: number[]): [number, number] {\n  if (shape.length === 0) {\n    throw Error('Cannot get rows and columns of an empty shape array.');\n  }\n\n  return [\n    shape.length > 1 ? shape[shape.length - 2] : 1, shape[shape.length - 1]\n  ];\n}\n\nexport function getShapeAs3D(shape: number[]): [number, number, number] {\n  let shapeAs3D: [number, number, number] = [1, 1, 1];\n  const isScalar = shape.length === 0 || (shape.length === 1 && shape[0] === 1);\n  if (!isScalar) {\n    shapeAs3D =\n        [getBatchDim(shape), ...getRowsCols(shape)] as [number, number, number];\n  }\n  return shapeAs3D;\n}\n\nexport function getTextureShapeFromLogicalShape(\n    logShape: number[], isPacked = false): [number, number] {\n  let maxTexSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if (isPacked) {\n    maxTexSize = maxTexSize * 2;\n\n    // This logic ensures we accurately count the number of packed texels needed\n    // to accommodate the tensor. We can only pack values in the same texel if\n    // they are from adjacent pairs of rows/cols within the same batch. So if a\n    // tensor has 3 rows, we pretend it has 4 rows in order to account for the\n    // fact that the texels containing the third row are half empty.\n    logShape = logShape.map(\n        (d, i) => i >= logShape.length - 2 ?\n            util.nearestLargerEven(logShape[i]) :\n            logShape[i]);\n\n    // Packed texture height is at least 2 (the channel height of a single\n    // texel).\n    if (logShape.length === 1) {\n      logShape = [2, logShape[0]];\n    }\n  }\n\n  // If logical shape is 2, we don't squeeze, since we want to match physical.\n  if (logShape.length !== 2) {\n    const squeezeResult = util.squeezeShape(logShape);\n    logShape = squeezeResult.newShape;\n  }\n\n  let size = util.sizeFromShape(logShape);\n  if (logShape.length <= 1 && size <= maxTexSize) {\n    return [1, size];\n  } else if (\n      logShape.length === 2 && logShape[0] <= maxTexSize &&\n      logShape[1] <= maxTexSize) {\n    return logShape as [number, number];\n  } else if (\n      logShape.length === 3 && logShape[0] * logShape[1] <= maxTexSize &&\n      logShape[2] <= maxTexSize) {\n    return [logShape[0] * logShape[1], logShape[2]];\n  } else if (\n      logShape.length === 3 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2]];\n  } else if (\n      logShape.length === 4 &&\n      logShape[0] * logShape[1] * logShape[2] <= maxTexSize &&\n      logShape[3] <= maxTexSize) {\n    return [logShape[0] * logShape[1] * logShape[2], logShape[3]];\n  } else if (\n      logShape.length === 4 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] * logShape[3] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2] * logShape[3]];\n  } else {\n    if (isPacked) {\n      // For packed textures size equals the number of channels required to\n      // accommodate the texture data. However in order to squarify such that\n      // inner dimensions stay even, we rewrite size to equal the number of\n      // texels. Then in the return statement we rehydrate the squarified\n      // dimensions to channel units.\n\n      const batchDim = getBatchDim(logShape);\n      let rows = 2, cols = 2;\n      if (logShape.length) {\n        [rows, cols] = getRowsCols(logShape);\n      }\n      size = batchDim * (rows / 2) * (cols / 2);\n      return util.sizeToSquarishShape(size).map(d => d * 2) as [number, number];\n    }\n    return util.sizeToSquarishShape(size);\n  }\n}\n\nfunction isEven(n: number): boolean {\n  return n % 2 === 0;\n}\n\n/**\n * This determines whether reshaping a packed texture requires rearranging\n * the data within the texture, assuming 2x2 packing.\n */\nexport function isReshapeFree(shape1: number[], shape2: number[]): boolean {\n  shape1 = shape1.slice(-2);\n  shape2 = shape2.slice(-2);\n\n  if (util.arraysEqual(shape1, shape2)) {\n    return true;\n  }\n\n  if (!shape1.length || !shape2.length) {  // One of the shapes is a scalar.\n    return true;\n  }\n\n  if (shape1[0] === 0 || shape1[1] === 0 || shape2[0] === 0 ||\n      shape2[1] === 0) {\n    return true;\n  }\n\n  if (shape1.length !== shape2.length) {  // One of the shapes is a vector.\n    const shape1Cols = shape1.slice(-1)[0];\n    const shape2Cols = shape2.slice(-1)[0];\n    if (shape1Cols === shape2Cols) {\n      return true;\n    }\n\n    if (isEven(shape1Cols) && isEven(shape2Cols) &&\n        (shape1[0] === 1 || shape2[0] === 1)) {\n      return true;\n    }\n  }\n  return shape1[1] === shape2[1] && isEven(shape1[0]) && isEven(shape2[0]);\n}\n\n// We cache webgl params because the environment gets reset between\n// unit tests and we don't want to constantly query the WebGLContext for\n// MAX_TEXTURE_SIZE.\nlet MAX_TEXTURE_SIZE: number;\nlet MAX_TEXTURES_IN_SHADER: number;\n\nexport function getWebGLMaxTextureSize(webGLVersion: number): number {\n  if (MAX_TEXTURE_SIZE == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURE_SIZE = gl.getParameter(gl.MAX_TEXTURE_SIZE);\n  }\n  return MAX_TEXTURE_SIZE;\n}\n\nexport function resetMaxTextureSize() {\n  MAX_TEXTURE_SIZE = null;\n}\nexport function resetMaxTexturesInShader() {\n  MAX_TEXTURES_IN_SHADER = null;\n}\n\nexport function getMaxTexturesInShader(webGLVersion: number): number {\n  if (MAX_TEXTURES_IN_SHADER == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURES_IN_SHADER = gl.getParameter(gl.MAX_TEXTURE_IMAGE_UNITS);\n  }\n  // We cap at 16 to avoid spurious runtime \"memory exhausted\" error.\n  return Math.min(16, MAX_TEXTURES_IN_SHADER);\n}\n\nexport function getWebGLDisjointQueryTimerVersion(webGLVersion: number):\n    number {\n  if (webGLVersion === 0) {\n    return 0;\n  }\n\n  let queryTimerVersion: number;\n  const gl = getWebGLContext(webGLVersion);\n\n  if (hasExtension(gl, 'EXT_disjoint_timer_query_webgl2') &&\n      webGLVersion === 2) {\n    queryTimerVersion = 2;\n  } else if (hasExtension(gl, 'EXT_disjoint_timer_query')) {\n    queryTimerVersion = 1;\n  } else {\n    queryTimerVersion = 0;\n  }\n  return queryTimerVersion;\n}\n\nexport function hasExtension(gl: WebGLRenderingContext, extensionName: string) {\n  const ext = gl.getExtension(extensionName);\n  return ext != null;\n}\n\nexport function isWebGLVersionEnabled(webGLVersion: 1|2) {\n  try {\n    const gl = getWebGLContext(webGLVersion);\n    if (gl != null) {\n      return true;\n    }\n  } catch (e) {\n    console.log('Error when getting WebGL context: ', e);\n    return false;\n  }\n  return false;\n}\n\nexport function isCapableOfRenderingToFloatTexture(webGLVersion: number):\n    boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n  } else {\n    if (!hasExtension(gl, 'EXT_color_buffer_float')) {\n      return false;\n    }\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\n/**\n * Check if we can download values from a float/half-float texture.\n *\n * Note that for performance reasons we use binding a texture to a framebuffer\n * as a proxy for ability to download float values later using readPixels. The\n * texture params of this texture will not match those in readPixels exactly\n * but if we are unable to bind some kind of float texture to the frameBuffer\n * then we definitely will not be able to read float values from it.\n */\nexport function isDownloadFloatTextureEnabled(webGLVersion: number): boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n    if (!hasExtension(gl, 'WEBGL_color_buffer_float')) {\n      return false;\n    }\n  } else {\n    if (hasExtension(gl, 'EXT_color_buffer_float')) {\n      return createFloatTextureAndBindToFramebuffer(gl);\n    }\n\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (hasExtension(gl, COLOR_BUFFER_HALF_FLOAT)) {\n      const textureHalfFloatExtension =\n          gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      return createHalfFloatTextureAndBindToFramebuffer(\n          gl, textureHalfFloatExtension);\n    }\n\n    return false;\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\nfunction createFloatTextureAndBindToFramebuffer(gl: WebGLRenderingContext):\n    boolean {\n  const texConfig = getTextureConfig(gl);\n\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nfunction createHalfFloatTextureAndBindToFramebuffer(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension: any): boolean {\n  const texConfig = getTextureConfig(gl, textureHalfFloatExtension);\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatHalfFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeHalfFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nexport function isWebGLFenceEnabled(webGLVersion: number) {\n  if (webGLVersion !== 2) {\n    return false;\n  }\n  const gl = getWebGLContext(webGLVersion);\n\n  // tslint:disable-next-line:no-any\n  const isEnabled = (gl as any).fenceSync != null;\n  return isEnabled;\n}\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${opName} does not support complex64 tensors ` +\n              'in the WebGL backend.');\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {device_util, env} from '@tensorflow/tfjs-core';\n\nimport {getMaxTexturesInShader, getWebGLDisjointQueryTimerVersion, getWebGLMaxTextureSize, isCapableOfRenderingToFloatTexture, isDownloadFloatTextureEnabled, isWebGLFenceEnabled, isWebGLVersionEnabled} from './webgl_util';\n\nconst ENV = env();\n\n/**\n * This file contains WebGL-specific flag registrations.\n */\n\n/**\n * True if WebGL is supported.\n */\nENV.registerFlag('HAS_WEBGL', () => ENV.getNumber('WEBGL_VERSION') > 0);\n\n/** 0: No WebGL, 1: WebGL 1.0, 2: WebGL 2.0. */\nENV.registerFlag('WEBGL_VERSION', () => {\n  if (isWebGLVersionEnabled(2)) {\n    return 2;\n  } else if (isWebGLVersionEnabled(1)) {\n    return 1;\n  }\n  return 0;\n});\n\n/** Whether to check for numerical representation problems. */\nENV.registerFlag('WEBGL_CHECK_NUMERICAL_PROBLEMS', () => false);\n\nENV.registerFlag(\n    'WEBGL_BUFFER_SUPPORTED', () => ENV.get('WEBGL_VERSION') === 2);\n\n/** Whether the WebGL backend will sometimes forward ops to the CPU. */\nENV.registerFlag('WEBGL_CPU_FORWARD', () => true);\n\n/** Whether the WebGL backend will always use f16 textures for rendering. */\nENV.registerFlag('WEBGL_FORCE_F16_TEXTURES', () => false);\n\n/** Whether to turn all packing related flags on. */\nENV.registerFlag('WEBGL_PACK', () => ENV.getBool('HAS_WEBGL'));\n\n/** Whether we will pack the batchnormalization op. */\nENV.registerFlag('WEBGL_PACK_NORMALIZATION', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the clip op. */\nENV.registerFlag('WEBGL_PACK_CLIP', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the depthwise conv op. */\n// TODO: https://github.com/tensorflow/tfjs/issues/1679\nENV.registerFlag('WEBGL_PACK_DEPTHWISECONV', () => false);\n\n/** Whether we will pack binary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_BINARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack unary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_UNARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack array ops. */\nENV.registerFlag(\n    'WEBGL_PACK_ARRAY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack image ops. */\nENV.registerFlag(\n    'WEBGL_PACK_IMAGE_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack reduce ops. */\nENV.registerFlag('WEBGL_PACK_REDUCE', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether packed WebGL kernels lazily unpack their outputs. */\nENV.registerFlag('WEBGL_LAZILY_UNPACK', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will use the im2col algorithm to speed up convolutions. */\nENV.registerFlag('WEBGL_CONV_IM2COL', () => ENV.getBool('WEBGL_PACK'));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURE_SIZE',\n    () => getWebGLMaxTextureSize(ENV.getNumber('WEBGL_VERSION')));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURES_IN_SHADER',\n    () => getMaxTexturesInShader(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * The disjoint_query_timer extension version.\n * 0: disabled, 1: EXT_disjoint_timer_query, 2:\n * EXT_disjoint_timer_query_webgl2.\n * In Firefox with WebGL 2.0,\n * EXT_disjoint_timer_query_webgl2 is not available, so we must use the\n * WebGL 1.0 extension.\n */\nENV.registerFlag('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION', () => {\n  const webGLVersion = ENV.getNumber('WEBGL_VERSION');\n\n  if (webGLVersion === 0) {\n    return 0;\n  }\n  return getWebGLDisjointQueryTimerVersion(webGLVersion);\n});\n\n/**\n * Whether the timer object from the disjoint_query_timer extension gives\n * timing information that is reliable.\n */\nENV.registerFlag(\n    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE',\n    () => ENV.getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0 &&\n        !device_util.isMobile());\n\n/**\n * Whether the device is physically capable of rendering to float32 textures.\n */\nENV.registerFlag(\n    'WEBGL_RENDER_FLOAT32_CAPABLE',\n    () => isCapableOfRenderingToFloatTexture(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Whether rendering to float32 textures is enabled. If disabled, renders to\n * float16 textures.\n */\nENV.registerFlag('WEBGL_RENDER_FLOAT32_ENABLED', () => {\n  return ENV.getBool('WEBGL_FORCE_F16_TEXTURES') ?\n      false :\n      ENV.getBool('WEBGL_RENDER_FLOAT32_CAPABLE');\n});\n\n/**\n * Whether downloading float textures is enabled (16 or 32 bit). If disabled,\n * uses IEEE 754 encoding of the float32 values to 4 uint8 when downloading.\n */\nENV.registerFlag(\n    'WEBGL_DOWNLOAD_FLOAT_ENABLED',\n    () => isDownloadFloatTextureEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/** Whether the fence API is available. */\nENV.registerFlag(\n    'WEBGL_FENCE_API_ENABLED',\n    () => isWebGLFenceEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Tensors with size <= than this will be uploaded as uniforms, not textures.\n */\nENV.registerFlag('WEBGL_SIZE_UPLOAD_UNIFORM', () => {\n  // Use uniform uploads only when 32bit floats are supported. In\n  // 16bit\n  // environments there are problems with comparing a 16bit texture value\n  // with a 32bit uniform value.\n  const useUniforms = ENV.getBool('WEBGL_RENDER_FLOAT32_ENABLED');\n  return useUniforms ? 4 : 0;\n});\n\n/**\n * If the total number of bytes allocated on the GPU is greater than this\n * number, we will aggressively delete textures upon disposal with\n * gl.deleteMatrixTexture, rather than making them available for reuse.\n *\n * Default value -1 indicates that we will never aggressively delete textures.\n */\nENV.registerFlag(\n    'WEBGL_DELETE_TEXTURE_THRESHOLD',\n    () => {\n      return -1;\n    },\n    threshold => {\n      if (threshold < 0 && threshold !== -1) {\n        throw new Error(\n            `WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never ` +\n            `delete) or at least 0, but got ${threshold}.`);\n      }\n    });\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Abs, util } from '@tensorflow/tfjs-core';\nexport function simpleAbsImpl(vals) {\n    const resultValues = new Float32Array(vals.length);\n    for (let i = 0; i < vals.length; ++i) {\n        resultValues[i] = Math.abs(vals[i]);\n    }\n    return resultValues;\n}\nexport const absKernelFunc = (args) => {\n    const { x } = args.inputs;\n    const cpuBackend = args.backend;\n    let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n    if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values;\n        resultValues = simpleAbsImpl(values);\n    }\n    else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals = cpuBackend.data.get(real.dataId).values;\n        const imagVals = cpuBackend.data.get(imag.dataId).values;\n        for (let i = 0; i < realVals.length; i++) {\n            const real = realVals[i];\n            const imag = imagVals[i];\n            resultValues[i] = Math.hypot(real, imag);\n        }\n    }\n    return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n};\nexport const absConfig = {\n    kernelName: Abs,\n    backendName: 'cpu',\n    kernelFunc: absKernelFunc,\n};\n//# sourceMappingURL=Abs.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { backend_util, util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op) {\n    return (aShape, bShape, aVals, bVals, dtype) => {\n        const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n        const resultRank = newShape.length;\n        const resultStrides = util.computeStrides(newShape);\n        const resultSize = util.sizeFromShape(newShape);\n        const result = util.getTypedArrayFromDType(dtype, resultSize);\n        const aRank = aShape.length;\n        const bRank = bShape.length;\n        const aStrides = util.computeStrides(aShape);\n        const bStrides = util.computeStrides(bShape);\n        const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n        const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n        if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n            for (let i = 0; i < result.length; ++i) {\n                result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n            }\n        }\n        else {\n            for (let i = 0; i < result.length; ++i) {\n                const loc = util.indexToLoc(i, resultRank, resultStrides);\n                const aLoc = loc.slice(-aRank);\n                aBroadcastDims.forEach(d => aLoc[d] = 0);\n                const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n                const bLoc = loc.slice(-bRank);\n                bBroadcastDims.forEach(d => bLoc[d] = 0);\n                const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n                result[i] = op(aVals[aIndex], bVals[bIndex]);\n            }\n        }\n        return [result, newShape];\n    };\n}\n//# sourceMappingURL=binary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Add } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal + bReal, imag: aImag + bImag };\n}));\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\nexport const addConfig = {\n    kernelName: Add,\n    backendName: 'cpu',\n    kernelFunc: add\n};\n//# sourceMappingURL=Add.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op) {\n    return (values, dtype, attrs) => {\n        const newValues = util.getTypedArrayFromDType(dtype, values.length);\n        for (let i = 0; i < values.length; ++i) {\n            newValues[i] = op(values[i], attrs);\n        }\n        return newValues;\n    };\n}\n//# sourceMappingURL=unary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Ceil } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\nexport const ceilConfig = {\n    kernelName: Ceil,\n    backendName: 'cpu',\n    kernelFunc: ceilKernelFunc,\n};\n//# sourceMappingURL=Ceil.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Exp } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\nexport const expConfig = {\n    kernelName: Exp,\n    backendName: 'cpu',\n    kernelFunc: expKernelFunc,\n};\n//# sourceMappingURL=Exp.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Expm1 } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\nexport const expm1Config = {\n    kernelName: Expm1,\n    backendName: 'cpu',\n    kernelFunc: expm1KernelFunc,\n};\n//# sourceMappingURL=Expm1.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Floor } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\nexport const floorConfig = {\n    kernelName: Floor,\n    backendName: 'cpu',\n    kernelFunc: floorKernelFunc,\n};\n//# sourceMappingURL=Floor.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Log } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\nexport const logConfig = {\n    kernelName: Log,\n    backendName: 'cpu',\n    kernelFunc: logKernelFunc,\n};\n//# sourceMappingURL=Log.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function maxImpl(aVals, reduceSize, outShape, dtype) {\n    const vals = util.getTypedArrayFromDType(dtype, util.sizeFromShape(outShape));\n    for (let i = 0; i < vals.length; ++i) {\n        const offset = i * reduceSize;\n        let max = aVals[offset];\n        for (let j = 0; j < reduceSize; ++j) {\n            const value = aVals[offset + j];\n            if (value > max) {\n                max = value;\n            }\n        }\n        vals[i] = max;\n    }\n    return vals;\n}\n//# sourceMappingURL=Max_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Multiply } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const multiplyImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n    };\n}));\nexport const multiply = binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\nexport const multiplyConfig = {\n    kernelName: Multiply,\n    backendName: 'cpu',\n    kernelFunc: multiply\n};\n//# sourceMappingURL=Multiply.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Rsqrt } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\nexport const rsqrtConfig = {\n    kernelName: Rsqrt,\n    backendName: 'cpu',\n    kernelFunc: rsqrtKernelFunc,\n};\n//# sourceMappingURL=Rsqrt.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Slice, slice_util, util } from '@tensorflow/tfjs-core';\nimport { assertNotComplex } from '../cpu_util';\nexport function sliceImpl(vals, begin, size, shape, dtype) {\n    const isContinous = slice_util.isSliceContinous(shape, begin, size);\n    const length = util.sizeFromShape(size);\n    const xStrides = util.computeStrides(shape);\n    if (isContinous) {\n        const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n        return vals.subarray(flatOffset, flatOffset + length);\n    }\n    const outVals = util.getTypedArrayFromDType(dtype, length);\n    for (let i = 0; i < length; ++i) {\n        const rank = size.length;\n        const strides = util.computeStrides(size);\n        const loc = util.indexToLoc(i, rank, strides);\n        const xLoc = loc.map((idx, j) => idx + begin[j]);\n        const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n        outVals[i] = vals[xIndex];\n    }\n    return outVals;\n}\nexport function slice(args) {\n    const { inputs, backend, attrs } = args;\n    const { x } = inputs;\n    const { begin, size } = attrs;\n    assertNotComplex(x, 'slice');\n    const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n    slice_util.assertParamsValid(x, $begin, $size);\n    const vals = backend.data.get(x.dataId).values;\n    const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n    return backend.makeTensorInfo($size, x.dtype, outVals);\n}\nexport const sliceConfig = {\n    kernelName: Slice,\n    backendName: 'cpu',\n    kernelFunc: slice\n};\n//# sourceMappingURL=Slice.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Sub } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const subImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal - bReal, imag: aImag - bImag };\n}));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\nexport const subConfig = {\n    kernelName: Sub,\n    backendName: 'cpu',\n    kernelFunc: sub\n};\n//# sourceMappingURL=Sub.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function transposeImpl(xVals, xShape, dtype, perm, newShape) {\n    const xRank = xShape.length;\n    const xSize = util.sizeFromShape(xShape);\n    const xStrides = util.computeStrides(xShape);\n    const newStrides = util.computeStrides(newShape);\n    const result = util.getTypedArrayFromDType(dtype, util.sizeFromShape(newShape));\n    for (let i = 0; i < xSize; ++i) {\n        const loc = util.indexToLoc(i, xRank, xStrides);\n        // Permute location.\n        const newLoc = new Array(loc.length);\n        for (let i = 0; i < newLoc.length; i++) {\n            newLoc[i] = loc[perm[i]];\n        }\n        const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n        result[newIndex] = xVals[i];\n    }\n    return result;\n}\n//# sourceMappingURL=Transpose_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { TensorBuffer, util } from '@tensorflow/tfjs-core';\nexport function uniqueImpl(values, axis, shape, dtype) {\n    // Normalize and validate axis.\n    const $axis = util.parseAxisParam(axis, shape)[0];\n    // Calculate the new shape that is suitable for extracting data along the\n    // given axis.\n    //\n    // The rank is 3.\n    // The size of the 1st dimension is the size of all the axes < the given axis.\n    // The size of the 2nd dimension is the same as the size of the given axis.\n    // The size of the 3rd dimension is the size of all the axes > the given axis.\n    //\n    // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n    // newShape would be: [2*3, 5, 4].\n    //\n    // Note that this is not the final output shape. This will be the shape for an\n    // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n    // values along the given axis. To demonstrate how it works, consider the\n    // following example:\n    //\n    // Input: a 3D tensor, with shape [1, 2, 3]\n    // [\n    //   [\n    //      [1,2,3],\n    //      [4,5,6]\n    //   ]\n    // ]\n    // Axis: 2 (the last axis).\n    // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n    //\n    // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n    // 1*2. The re-shaped data would look like:\n    //\n    // [\n    //   [\n    //     [1], [2], [3]\n    //   ],\n    //   [\n    //     [4], [5], [6]\n    //   ]\n    // ]\n    //\n    // Then, we can construct a 3-level nested loop by the following dimension\n    // order to extract the values along the axis (dimension1):\n    // i: dimension1       // 0,1,2 (newShape[1])\n    //   m: dimension0     // 0,1   (newShape[0])\n    //     n: dimension2   // 0     (newShape[2])\n    //\n    //                       m, i, n\n    //                      ---------\n    // Iteration 0: data at [0, 0, 0] => \"1\"\n    // Iteration 1: data at [1, 0, 0] => \"4\"\n    // We got [1,4].\n    // Iteration 2: data at [0, 1, 0] => \"2\"\n    // Iteration 3: data at [1, 1, 0] => \"5\"\n    // We got [2,5].\n    // Iteration 4: data at [0, 2, 0] => \"3\"\n    // Iteration 5: data at [1, 2, 0] => \"6\"\n    // We got [3,6].\n    const newShape = [1, shape[0], 1];\n    for (let i = 0; i < $axis; i++) {\n        newShape[0] *= shape[i];\n    }\n    newShape[1] = shape[$axis];\n    for (let i = $axis + 1; i < shape.length; i++) {\n        newShape[2] *= shape[i];\n    }\n    // A map from unique elements (their string representations) to their values\n    // in \"indices\" (below).\n    const uniqueElements = {};\n    // The indices of each unique element in the original tensor along the given\n    // axis. It is 1D and has the same size as the given axis.\n    const indices = new Int32Array(shape[$axis]);\n    // Create a buffer so we can easily extract value at a given location.\n    const inputBuffer = new TensorBuffer(newShape, dtype, values);\n    // The indices along the given axis that have unique elements. This is a\n    // de-duped version of \"indices\" above.\n    const uniqueIndices = [];\n    const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n    for (let i = 0; i < shape[$axis]; i++) {\n        // Extract values along the axis.\n        let element;\n        if (is1DTensor) {\n            // Fast path for 1D tensor input.\n            element = values[i].toString();\n        }\n        else {\n            const axisValues = [];\n            for (let m = 0; m < newShape[0]; m++) {\n                for (let n = 0; n < newShape[2]; n++) {\n                    axisValues.push(inputBuffer.get(m, i, n));\n                }\n            }\n            element = axisValues.join(',');\n        }\n        // Dedup and update various indices.\n        if (uniqueElements[element] !== undefined) {\n            indices[i] = uniqueElements[element];\n        }\n        else {\n            const uniqueIndex = Object.keys(uniqueElements).length;\n            uniqueElements[element] = uniqueIndex;\n            indices[i] = uniqueIndex;\n            uniqueIndices.push(i);\n        }\n    }\n    // Now we know where each of the unique elements are located along the axis\n    // (uniqueIndices). Extract them from input buffer and store them in the\n    // output buffer.\n    const outputTmpShape = newShape.slice();\n    outputTmpShape[1] = Object.keys(uniqueElements).length;\n    const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n    uniqueIndices.forEach((uniqueElementIndex, i) => {\n        for (let m = 0; m < newShape[0]; m++) {\n            for (let n = 0; n < newShape[2]; n++) {\n                outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n            }\n        }\n    });\n    // The output shape can be calculated from the input shape with the size of\n    // the given axis replaced by the number of unique elements along that axis.\n    const outputShape = shape.slice();\n    outputShape[$axis] = outputTmpShape[1];\n    return {\n        outputValues: outputBuffer.values,\n        outputShape,\n        indices,\n    };\n}\n//# sourceMappingURL=Unique_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import shared functionality from tfjs-backend-cpu without triggering\n// side effects.\n// tslint:disable-next-line: no-imports-from-dist\nimport * as shared from '@tensorflow/tfjs-backend-cpu/dist/shared';\n\nconst {\n  simpleAbsImpl: simpleAbsImplCPU,\n  addImpl: addImplCPU,\n  ceilImpl: ceilImplCPU,\n  expImpl: expImplCPU,\n  expm1Impl: expm1ImplCPU,\n  floorImpl: floorImplCPU,\n  logImpl: logImplCPU,\n  maxImpl: maxImplCPU,\n  multiplyImpl: multiplyImplCPU,\n  rsqrtImpl: rsqrtImplCPU,\n  sliceImpl: sliceImplCPU,\n  subImpl: subImplCPU,\n  transposeImpl: transposeImplCPU,\n  uniqueImpl: uniqueImplCPU,\n} = shared;\n\nexport {\n  simpleAbsImplCPU,\n  addImplCPU,\n  ceilImplCPU,\n  expImplCPU,\n  expm1ImplCPU,\n  logImplCPU,\n  multiplyImplCPU,\n  sliceImplCPU,\n  subImplCPU,\n  floorImplCPU,\n  maxImplCPU,\n  rsqrtImplCPU,\n  transposeImplCPU,\n  uniqueImplCPU,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`float v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        float result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`vec4 v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        vec4 result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ArgMinMaxProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo, op: 'max'|'min',\n      firstPass: boolean) {\n    const {windowSize, batchSize, outSize} = reduceInfo;\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    this.outputShape = [batchSize, outSize];\n    const compOp = (op === 'max') ? '>' : '<';\n    const indexSnippet = firstPass ?\n        'inOffset + i;' :\n        'round(getBestIndicesA(batch, inOffset + i));';\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        int bestIndex = inOffset;\n        float bestValue = getA(batch, bestIndex);\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          int inIdx = ${indexSnippet};\n          float candidate = getA(batch, inIdx);\n          if (candidate ${compOp} bestValue) {\n            bestValue = candidate;\n            bestIndex = inIdx;\n          }\n        }\n        setOutput(float(bestIndex));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport function getVecChannels(name: string, rank: number): string[] {\n  return ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank).map(d => `${name}.${d}`);\n}\n\nexport function getChannels(name: string, rank: number): string[] {\n  if (rank === 1) {\n    return [name];\n  }\n  return getVecChannels(name, rank);\n}\n\nexport function getSourceCoords(rank: number, dims: string[]): string {\n  if (rank === 1) {\n    return 'rc';\n  }\n\n  let coords = '';\n  for (let i = 0; i < rank; i++) {\n    coords += dims[i];\n    if (i < rank - 1) {\n      coords += ',';\n    }\n  }\n  return coords;\n}", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '@tensorflow/tfjs-core';\n\nexport type GLSL = {\n  version: string,\n  attribute: string,\n  varyingVs: string,\n  varyingFs: string,\n  texture2D: string,\n  output: string,\n  defineOutput: string,\n  defineSpecialNaN: string,\n  defineSpecialInf: string,\n  defineRound: string\n};\n\nexport function getGlslDifferences(): GLSL {\n  let version: string;\n  let attribute: string;\n  let varyingVs: string;\n  let varyingFs: string;\n  let texture2D: string;\n  let output: string;\n  let defineOutput: string;\n  let defineSpecialNaN: string;\n  let defineSpecialInf: string;\n  let defineRound: string;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    version = '#version 300 es';\n    attribute = 'in';\n    varyingVs = 'out';\n    varyingFs = 'in';\n    texture2D = 'texture';\n    output = 'outputColor';\n    defineOutput = 'out vec4 outputColor;';\n\n    // Use custom isnan definition to work across differences between\n    // implementations on various platforms. While this should happen in ANGLE\n    // we still see differences between android and windows (on chrome) when\n    // using isnan directly.\n    defineSpecialNaN = `\n      bool isnan_custom(float val) {\n        return (val > 0.0 || val < 0.0) ? false : val != 0.0;\n      }\n\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan_custom(val.x),\n          isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));\n      }\n\n      #define isnan(value) isnan_custom(value)\n    `;\n    // In webgl 2 we do not need to specify a custom isinf so there is no\n    // need for a special INFINITY constant.\n    defineSpecialInf = ``;\n    defineRound = `\n      #define round(value) newRound(value)\n      int newRound(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 newRound(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  } else {\n    version = '';\n    attribute = 'attribute';\n    varyingVs = 'varying';\n    varyingFs = 'varying';\n    texture2D = 'texture2D';\n    output = 'gl_FragColor';\n    defineOutput = '';\n    // WebGL1 has no built in isnan so we define one here.\n    defineSpecialNaN = `\n      #define isnan(value) isnan_custom(value)\n      bool isnan_custom(float val) {\n        return (val > 0. || val < 1. || val == 0.) ? false : true;\n      }\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));\n      }\n    `;\n    defineSpecialInf = `\n      uniform float INFINITY;\n\n      bool isinf(float val) {\n        return abs(val) == INFINITY;\n      }\n      bvec4 isinf(vec4 val) {\n        return equal(abs(val), vec4(INFINITY));\n      }\n    `;\n    defineRound = `\n      int round(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 round(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  }\n\n  return {\n    version,\n    attribute,\n    varyingVs,\n    varyingFs,\n    texture2D,\n    output,\n    defineOutput,\n    defineSpecialNaN,\n    defineSpecialInf,\n    defineRound\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\n/**\n * Produces GLSL code that derives logical coordinates from a flat\n * index. The code performs integer division with each stride and decrements\n * the index until the index equals the final dimension coordinate.\n */\nexport function getLogicalCoordinatesFromFlatIndex(\n    coords: string[], shape: number[], index = 'index'): string {\n  const strides = util.computeStrides(shape);\n  return strides\n      .map((stride, i) => {\n        const line1 = `int ${coords[i]} = ${index} / ${stride}`;\n        const line2 = i === strides.length - 1 ?\n            `int ${coords[i + 1]} = ${index} - ${coords[i]} * ${stride}` :\n            `index -= ${coords[i]} * ${stride}`;\n        return `${line1}; ${line2};`;\n      })\n      .join('');\n}\n\nfunction buildVec(x: string[]): string {\n  if (x.length === 1) {\n    return `${x[0]}`;\n  }\n  return `vec${x.length}(${x.join(',')})`;\n}\n\n/**\n * Produces GLSL code that computes the dot product of the input x and y\n * vectors. Handles splitting inputs into increments of vec4s when necessary.\n */\nexport function dotify(x: string[], y: string[]): string {\n  if (x.length !== y.length) {\n    throw new Error(\n        `Vectors to be dotted must be of the same length -` +\n        `got ${x.length} and ${y.length}`);\n  }\n\n  const slices: string[] = [];\n  const nearestVec4 = Math.floor(x.length / 4);\n  const nearestVec4Remainder = x.length % 4;\n\n  for (let i = 0; i < nearestVec4; i++) {\n    const xSlice = x.slice(i * 4, i * 4 + 4);\n    const ySlice = y.slice(i * 4, i * 4 + 4);\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  if (nearestVec4Remainder !== 0) {\n    let xSlice = x.slice(nearestVec4 * 4);\n    let ySlice = y.slice(nearestVec4 * 4);\n    if (xSlice.length === 1) {\n      xSlice = xSlice.map(d => `float(${d})`);\n      ySlice = ySlice.map(d => `float(${d})`);\n    }\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  return slices.map((d, i) => `dot(${d})`).join('+');\n}\n\n/**\n * Produces GLSL that computes the flat index from 3D coordinates.\n */\nexport function getFlatIndexFrom3D(shape: [number, number, number]): string {\n  const strides = util.computeStrides(shape).map(d => d.toString());\n\n  return `\n  int getFlatIndex(ivec3 coords) {\n    return coords.x * ${strides[0]} + coords.y * ${strides[1]} + coords.z;\n  }\n`;\n}\n\nexport const ENCODE_FLOAT_SNIPPET = `\n  const float FLOAT_MAX = 1.70141184e38;\n  const float FLOAT_MIN = 1.17549435e-38;\n\n  lowp vec4 encode_float(highp float v) {\n    if (isnan(v)) {\n      return vec4(255, 255, 255, 255);\n    }\n\n    highp float av = abs(v);\n\n    if(av < FLOAT_MIN) {\n      return vec4(0.0, 0.0, 0.0, 0.0);\n    } else if(v > FLOAT_MAX) {\n      return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;\n    } else if(v < -FLOAT_MAX) {\n      return vec4(0.0, 0.0,  128.0, 255.0) / 255.0;\n    }\n\n    highp vec4 c = vec4(0,0,0,0);\n\n    highp float e = floor(log2(av));\n    highp float m = exp2(fract(log2(av))) - 1.0;\n\n    c[2] = floor(128.0 * m);\n    m -= c[2] / 128.0;\n    c[1] = floor(32768.0 * m);\n    m -= c[1] / 32768.0;\n    c[0] = floor(8388608.0 * m);\n\n    highp float ebias = e + 127.0;\n    c[3] = floor(ebias / 2.0);\n    ebias -= c[3] * 2.0;\n    c[2] += floor(ebias) * 128.0;\n\n    c[3] += 128.0 * step(0.0, -v);\n\n    return c / 255.0;\n  }\n`;\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\nconst {getBroadcastDims} = backend_util;\nimport {getGlslDifferences, GLSL} from './glsl_version';\nimport * as shader_util from './shader_compiler_util';\n\nexport type ShapeInfo = {\n  logicalShape: number[],\n  texShape: [number, number],\n  isUniform: boolean,\n  isPacked: boolean,\n  flatOffset: number\n};\n\nexport type InputInfo = {\n  name: string,\n  shapeInfo: ShapeInfo\n};\n\nexport function makeShader(\n    inputsInfo: InputInfo[], outputShape: ShapeInfo, userCode: string,\n    usesPackedTextures: boolean): string {\n  const prefixSnippets: string[] = [];\n  inputsInfo.forEach(x => {\n    const size = util.sizeFromShape(x.shapeInfo.logicalShape);\n\n    // Snippet when we decided to upload the values as uniform.\n    if (x.shapeInfo.isUniform) {\n      prefixSnippets.push(\n          `uniform float ${x.name}${size > 1 ? `[${size}]` : ''};`);\n    } else {\n      prefixSnippets.push(`uniform sampler2D ${x.name};`);\n      prefixSnippets.push(`uniform int offset${x.name};`);\n    }\n  });\n  const inputPrefixSnippet = prefixSnippets.join('\\n');\n\n  const inputSamplingSnippet =\n      inputsInfo\n          .map(x => getInputSamplingSnippet(x, outputShape, usesPackedTextures))\n          .join('\\n');\n  const outTexShape = outputShape.texShape;\n  const glsl = getGlslDifferences();\n  const floatTextureSampleSnippet = getFloatTextureSampleSnippet(glsl);\n  let outputSamplingSnippet: string;\n  let floatTextureSetOutputSnippet: string;\n  let shaderPrefix = getShaderPrefix(glsl);\n\n  if (outputShape.isPacked) {\n    outputSamplingSnippet =\n        getPackedOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRGBASnippet(glsl);\n  } else {\n    outputSamplingSnippet =\n        getOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRSnippet(glsl);\n  }\n\n  if (usesPackedTextures) {\n    shaderPrefix += SHADER_PACKED_PREFIX;\n  }\n\n  const source = [\n    shaderPrefix, floatTextureSampleSnippet, floatTextureSetOutputSnippet,\n    inputPrefixSnippet, outputSamplingSnippet, inputSamplingSnippet, userCode\n  ].join('\\n');\n  return source;\n}\n\nfunction getSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getSamplerScalar(inInfo);\n    case 1:\n      return getSampler1D(inInfo);\n    case 2:\n      return getSampler2D(inInfo);\n    case 3:\n      return getSampler3D(inInfo);\n    case 4:\n      return getSampler4D(inInfo);\n    case 5:\n      return getSampler5D(inInfo);\n    case 6:\n      return getSampler6D(inInfo);\n    default:\n      throw new Error(\n          `${shape.length}-D input sampling` +\n          ` is not yet supported`);\n  }\n}\n\nfunction getPackedSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getPackedSamplerScalar(inInfo);\n    case 1:\n      return getPackedSampler1D(inInfo);\n    case 2:\n      return getPackedSampler2D(inInfo);\n    case 3:\n      return getPackedSampler3D(inInfo);\n    default:\n      return getPackedSamplerND(inInfo);\n  }\n}\n\nfunction getInputSamplingSnippet(\n    inInfo: InputInfo, outShapeInfo: ShapeInfo,\n    usesPackedTextures = false): string {\n  let res = '';\n  if (usesPackedTextures) {\n    res += getPackedSamplerFromInInfo(inInfo);\n  } else {\n    res += getSamplerFromInInfo(inInfo);\n  }\n\n  const inShape = inInfo.shapeInfo.logicalShape;\n  const outShape = outShapeInfo.logicalShape;\n  if (inShape.length <= outShape.length) {\n    if (usesPackedTextures) {\n      res += getPackedSamplerAtOutputCoords(inInfo, outShapeInfo);\n    } else {\n      res += getSamplerAtOutputCoords(inInfo, outShapeInfo);\n    }\n  }\n  return res;\n}\n\nfunction getPackedOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutputPacked1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutputPacked2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutputPacked3DCoords(\n          outShape as [number, number, number], outTexShape);\n    default:\n      return getOutputPackedNDCoords(outShape, outTexShape);\n  }\n}\n\nfunction getOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutput1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutput2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutput3DCoords(\n          outShape as [number, number, number], outTexShape);\n    case 4:\n      return getOutput4DCoords(\n          outShape as [number, number, number, number], outTexShape);\n    case 5:\n      return getOutput5DCoords(\n          outShape as [number, number, number, number, number], outTexShape);\n    case 6:\n      return getOutput6DCoords(\n          outShape as [number, number, number, number, number, number],\n          outTexShape);\n    default:\n      throw new Error(\n          `${outShape.length}-D output sampling is not yet supported`);\n  }\n}\n\nfunction getFloatTextureSampleSnippet(glsl: GLSL): string {\n  return `\n    float sampleTexture(sampler2D textureSampler, vec2 uv) {\n      return ${glsl.texture2D}(textureSampler, uv).r;\n    }\n  `;\n}\n\nfunction getFloatTextureSetRSnippet(glsl: GLSL): string {\n  return `\n    void setOutput(float val) {\n      ${glsl.output} = vec4(val, 0, 0, 0);\n    }\n  `;\n}\n\nfunction getFloatTextureSetRGBASnippet(glsl: GLSL): string {\n  return `\n    void setOutput(vec4 val) {\n      ${glsl.output} = val;\n    }\n  `;\n}\n\nfunction getShaderPrefix(glsl: GLSL): string {\n  const SHADER_PREFIX = `${glsl.version}\n    precision highp float;\n    precision highp int;\n    precision highp sampler2D;\n    ${glsl.varyingFs} vec2 resultUV;\n    ${glsl.defineOutput}\n    const vec2 halfCR = vec2(0.5, 0.5);\n\n    struct ivec5\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n    };\n\n    struct ivec6\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n      int v;\n    };\n\n    uniform float NAN;\n    ${glsl.defineSpecialNaN}\n    ${glsl.defineSpecialInf}\n    ${glsl.defineRound}\n\n    int imod(int x, int y) {\n      return x - y * (x / y);\n    }\n\n    int idiv(int a, int b, float sign) {\n      int res = a / b;\n      int mod = imod(a, b);\n      if (sign < 0. && mod != 0) {\n        res -= 1;\n      }\n      return res;\n    }\n\n    //Based on the work of Dave Hoskins\n    //https://www.shadertoy.com/view/4djSRW\n    #define HASHSCALE1 443.8975\n    float random(float seed){\n      vec2 p = resultUV * seed;\n      vec3 p3  = fract(vec3(p.xyx) * HASHSCALE1);\n      p3 += dot(p3, p3.yzx + 19.19);\n      return fract((p3.x + p3.y) * p3.z);\n    }\n\n    ${SAMPLE_1D_SNIPPET}\n    ${SAMPLE_2D_SNIPPET}\n    ${SAMPLE_3D_SNIPPET}\n  `;\n\n  return SHADER_PREFIX;\n}\n\nconst SAMPLE_1D_SNIPPET = `\nvec2 uvFromFlat(int texNumR, int texNumC, int index) {\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\nvec2 packedUVfrom1D(int texNumR, int texNumC, int index) {\n  int texelIndex = index / 2;\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_2D_SNIPPET = `\nvec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,\n  int texNumC, int row, int col) {\n  int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_3D_SNIPPET = `\nvec2 packedUVfrom3D(int texNumR, int texNumC,\n    int texelsInBatch, int texelsInLogicalRow, int b,\n    int row, int col) {\n  int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SHADER_PACKED_PREFIX = `\n  float getChannel(vec4 frag, vec2 innerDims) {\n    vec2 modCoord = mod(innerDims, 2.);\n    return modCoord.x == 0. ?\n      (modCoord.y == 0. ? frag.r : frag.g) :\n      (modCoord.y == 0. ? frag.b : frag.a);\n  }\n  float getChannel(vec4 frag, int dim) {\n    float modCoord = mod(float(dim), 2.);\n    return modCoord == 0. ? frag.r : frag.g;\n  }\n`;\n\nfunction getOutputScalarCoords() {\n  return `\n    int getOutputCoords() {\n      return 0;\n    }\n  `;\n}\n\nfunction getOutputPacked1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (packedTexShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.x * ${packedTexShape[1]}.0);\n      }\n    `;\n  }\n\n  if (packedTexShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.y * ${packedTexShape[0]}.0);\n      }\n    `;\n  }\n\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      return 2 * (resTexRC.x * ${packedTexShape[1]} + resTexRC.y);\n    }\n  `;\n}\n\nfunction getOutput1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  if (texShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.x * ${texShape[1]}.0);\n      }\n    `;\n  }\n  if (texShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.y * ${texShape[0]}.0);\n      }\n    `;\n  }\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      return resTexRC.x * ${texShape[1]} + resTexRC.y;\n    }\n  `;\n}\n\nfunction getOutputPacked3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texelsInLogicalRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = texelsInLogicalRow * Math.ceil(shape[1] / 2);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec3(b, r, c);\n    }\n  `;\n}\n\nfunction getOutput3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n\nfunction getOutputPackedNDCoords(\n    shape: number[], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  const texelsInLogicalRow = Math.ceil(shape[shape.length - 1] / 2);\n  const texelsInBatch =\n      texelsInLogicalRow * Math.ceil(shape[shape.length - 2] / 2);\n  let texelsInBatchN = texelsInBatch;\n  let batches = ``;\n  let coords = 'b, r, c';\n\n  for (let b = 2; b < shape.length - 1; b++) {\n    texelsInBatchN *= shape[shape.length - b - 1];\n    batches = `\n      int b${b} = index / ${texelsInBatchN};\n      index -= b${b} * ${texelsInBatchN};\n    ` + batches;\n    coords = `b${b}, ` + coords;\n  }\n\n  return `\n    ivec${shape.length} getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      ${batches}\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec${shape.length}(${coords});\n    }\n  `;\n}\n\nfunction getOutput4DCoords(\n    shape: [number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2'], shape);\n\n  return `\n    ivec4 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec4(r, c, d, d2);\n    }\n  `;\n}\n\nfunction getOutput5DCoords(\n    shape: [number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3'], shape);\n\n  return `\n    ivec5 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx * vec2(${texShape[0]},\n                             ${texShape[1]}));\n\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec5 outShape = ivec5(r, c, d, d2, d3);\n      return outShape;\n    }\n  `;\n}\n\nfunction getOutput6DCoords(\n    shape: [number, number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3', 'd4'], shape);\n\n  return `\n    ivec6 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec6 result = ivec6(r, c, d, d2, d3, d4);\n      return result;\n    }\n  `;\n}\n\nfunction getOutputPacked2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return 2 * ivec2(resultUV.yx * vec2(${packedTexShape[0]}, ${\n        packedTexShape[1]}));\n      }\n    `;\n  }\n\n  // texels needed to accommodate a logical row\n  const texelsInLogicalRow = Math.ceil(shape[1] / 2);\n\n  /**\n   * getOutputCoords\n   *\n   * resTexRC: The rows and columns of the texels. If you move over one\n   * texel to the right in the packed texture, you are moving over one column\n   * (not two).\n   *\n   * index: The texel index\n   */\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getOutput2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return ivec2(resultUV.yx * vec2(${texShape[0]}, ${texShape[1]}));\n      }\n    `;\n  }\n  if (shape[1] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(index, 0);\n      }\n    `;\n  }\n  if (shape[0] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(0, index);\n      }\n    `;\n  }\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      int r = index / ${shape[1]};\n      int c = index - r * ${shape[1]};\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getFlatOffsetUniformName(texName: string): string {\n  return `offset${texName}`;\n}\n\nfunction getPackedSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}() {\n      return ${glsl.texture2D}(${texName}, halfCR);\n    }\n  `;\n}\n\nfunction getSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  if (inputInfo.shapeInfo.isUniform) {\n    return `float ${funcName}() {return ${texName};}`;\n  }\n  const [texNumR, texNumC] = inputInfo.shapeInfo.texShape;\n  if (texNumR === 1 && texNumC === 1) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n\n  const [tNumR, tNumC] = inputInfo.shapeInfo.texShape;\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}() {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int index) {\n      vec2 uv = packedUVfrom1D(\n        ${packedTexShape[0]}, ${packedTexShape[1]}, index);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int index) {\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const tNumR = texShape[0];\n  const tNumC = texShape[1];\n\n  if (tNumC === 1 && tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  if (tNumC === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2(0.5, (float(index + ${offset}) + 0.5) / ${tNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2((float(index + ${offset}) + 0.5) / ${tNumC}.0, 0.5);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  return `\n    float ${funcName}(int index) {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const glsl = getGlslDifferences();\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    return `\n      vec4 ${funcName}(int row, int col) {\n        vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n\n        return ${glsl.texture2D}(${texName}, uv);\n      }\n    `;\n  }\n\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const valuesPerRow = Math.ceil(shape[1] / 2);\n\n  return `\n    vec4 ${funcName}(int row, int col) {\n      vec2 uv = packedUVfrom2D(${valuesPerRow}, ${packedTexShape[0]}, ${\n      packedTexShape[1]}, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    const texNumR = texShape[0];\n    const texNumC = texShape[1];\n    return `\n    float ${funcName}(int row, int col) {\n      vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col) {\n        int index = round(dot(vec2(row, col), vec2(${shape[1]}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const offset = getFlatOffsetUniformName(texName);\n  if (texNumC === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2(0.5, (index + 0.5) / ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n  if (texNumR === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2((index + 0.5) / ${texNumC}.0, 0.5);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  return `\n  float ${funcName}(int row, int col) {\n    // Explicitly use integer operations as dot() only works on floats.\n    int index = row * ${shape[1]} + col + ${offset};\n    vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n    return sampleTexture(${texName}, uv);\n  }\n`;\n}\n\nfunction getPackedSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  if (shape[0] === 1) {\n    const squeezedShape = shape.slice(1);\n    const keptDims = [1, 2];\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['b', 'row', 'col'];\n    return `\n        ${getPackedSamplerFromInInfo(newInputInfo)}\n        vec4 ${funcName}(int b, int row, int col) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = valuesPerRow * Math.ceil(shape[1] / 2);\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int b, int row, int col) {\n      vec2 uv = packedUVfrom3D(\n        ${texNumR}, ${texNumC}, ${texelsInBatch}, ${valuesPerRow}, b, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride0 = shape[1] * shape[2];\n  const stride1 = shape[2];\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col', 'depth'];\n    return `\n        ${getSamplerFromInInfo(newInputInfo)}\n        float ${funcName}(int row, int col, int depth) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth) {\n        int index = round(dot(vec3(row, col, depth),\n                          vec3(${stride0}, ${stride1}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n        float ${funcName}(int row, int col, int depth) {\n          float texR = float(row);\n          float texC = dot(vec2(col, depth), vec2(${stride1}, 1));\n          vec2 uv = (vec2(texC, texR) + halfCR) /\n                     vec2(${texNumC}.0, ${texNumR}.0);\n          return sampleTexture(${texName}, uv);\n        }\n      `;\n  }\n\n  if (texNumC === stride1 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col, int depth) {\n      float texR = dot(vec2(row, col), vec2(${shape[1]}, 1));\n      float texC = float(depth);\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n      float ${funcName}(int row, int col, int depth) {\n        // Explicitly use integer operations as dot() only works on floats.\n        int index = row * ${stride0} + col * ${stride1} + depth + ${offset};\n        vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n        return sampleTexture(${texName}, uv);\n      }\n  `;\n}\n\nfunction getPackedSamplerND(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const rank = shape.length;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[rank - 1] / 2);\n  let texelsInBatch = valuesPerRow * Math.ceil(shape[rank - 2] / 2);\n  let params = `int b, int row, int col`;\n  let index = `b * ${texelsInBatch} + (row / 2) * ${valuesPerRow} + (col / 2)`;\n  for (let b = 2; b < rank - 1; b++) {\n    params = `int b${b}, ` + params;\n    texelsInBatch *= shape[rank - b - 1];\n    index = `b${b} * ${texelsInBatch} + ` + index;\n  }\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}(${params}) {\n      int index = ${index};\n      int texR = index / ${texNumC};\n      int texC = index - texR * ${texNumC};\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}, ${texNumR});\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler4D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride2 = shape[3];\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        int index = round(dot(vec4(row, col, depth, depth2),\n                          vec4(${stride0}, ${stride1}, ${stride2}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = float(row);\n        float texC =\n            dot(vec3(col, depth, depth2),\n                vec3(${stride1}, ${stride2}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride2 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = dot(vec3(row, col, depth),\n                         vec3(${shape[1] * shape[2]}, ${shape[2]}, 1));\n        float texC = float(depth2);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} +\n          depth * ${stride2} + depth2;\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler5D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride3 = shape[4];\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float index = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          depth3;\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n                         vec4(${stride1}, ${stride2}, ${stride3}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  if (texNumC === stride3 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float texR = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3]},\n               ${shape[2] * shape[3]}, ${shape[3]}, 1));\n        int texC = depth3;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler6D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3', 'depth4'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  const stride4 = shape[5];\n  const stride3 = shape[4] * stride4;\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n        int index = round(dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          dot(\n            vec2(depth3, depth4),\n            vec2(${stride4}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n          vec4(${stride1}, ${stride2}, ${stride3}, ${stride4})) +\n               float(depth4);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride4 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        float texR = dot(vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3] * shape[4]},\n               ${shape[2] * shape[3] * shape[4]},\n               ${shape[3] * shape[4]},\n               ${shape[4]})) + float(depth3);\n        int texC = depth4;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 * ${stride4} + depth4 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getUniformSampler(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n\n  if (inSize < 2) {\n    return `return ${texName};`;\n  }\n  return `\n    for (int i = 0; i < ${inSize}; i++) {\n      if (i == index) {\n        return ${texName}[i];\n      }\n    }\n  `;\n}\n\nfunction getPackedSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n\n  const type = getCoordsDataType(outRank);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  let output = `return outputValue;`;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n  const isInputScalar = inSize === 1;\n  const outSize = util.sizeFromShape(outShapeInfo.logicalShape);\n  const isOutputScalar = outSize === 1;\n\n  if (inRank === 1 && !isInputScalar && !isOutputScalar) {\n    output = `\n      return vec4(outputValue.xy, outputValue.xy);\n    `;\n  } else if (isInputScalar && !isOutputScalar) {\n    if (outRank === 1) {\n      output = `\n        return vec4(outputValue.x, outputValue.x, 0., 0.);\n      `;\n    } else {\n      output = `\n        return vec4(outputValue.x);\n      `;\n    }\n  } else if (broadcastDims.length) {\n    const rows = inRank - 2;\n    const cols = inRank - 1;\n\n    if (broadcastDims.indexOf(rows) > -1 && broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.x);`;\n    } else if (broadcastDims.indexOf(rows) > -1) {\n      output = `return vec4(outputValue.x, outputValue.y, ` +\n          `outputValue.x, outputValue.y);`;\n    } else if (broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.xx, outputValue.zz);`;\n    }\n  }\n\n  return `\n    vec4 ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      vec4 outputValue = get${texFuncSnippet}(${unpackedCoordsSnippet});\n      ${output}\n    }\n  `;\n}\n\nfunction getSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const outTexShape = outShapeInfo.texShape;\n  const inTexShape = inputInfo.shapeInfo.texShape;\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  if (!inputInfo.shapeInfo.isUniform && inRank === outRank &&\n      inputInfo.shapeInfo.flatOffset == null &&\n      util.arraysEqual(inTexShape, outTexShape)) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, resultUV);\n      }\n    `;\n  }\n\n  const type = getCoordsDataType(outRank);\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  return `\n    float ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      return get${texFuncSnippet}(${unpackedCoordsSnippet});\n    }\n  `;\n}\n\nexport function getCoordsDataType(rank: number): string {\n  if (rank <= 1) {\n    return 'int';\n  } else if (rank === 2) {\n    return 'ivec2';\n  } else if (rank === 3) {\n    return 'ivec3';\n  } else if (rank === 4) {\n    return 'ivec4';\n  } else if (rank === 5) {\n    return 'ivec5';\n  } else if (rank === 6) {\n    return 'ivec6';\n  } else {\n    throw Error(`GPU for rank ${rank} is not yet supported`);\n  }\n}\n\n/** Returns a new input info (a copy) that has a squeezed logical shape. */\nfunction squeezeInputInfo(\n    inInfo: InputInfo, squeezedShape: number[]): InputInfo {\n  // Deep copy.\n  const newInputInfo: InputInfo = JSON.parse(JSON.stringify(inInfo));\n  newInputInfo.shapeInfo.logicalShape = squeezedShape;\n  return newInputInfo;\n}\n\nfunction getSqueezedParams(params: string[], keptDims: number[]): string {\n  return keptDims.map(d => params[d]).join(', ');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ArgMinMaxPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      shape: number[], windowSize: number, op: 'max'|'min',\n      firstPass: boolean) {\n    util.assert(\n        shape.length > 2,\n        () => `Packed arg${\n            op.charAt(0).toUpperCase() +\n            op.slice(1)} supports only inputs with rank above 2.`);\n    const inSize = shape[shape.length - 1];\n    const outSize = Math.ceil(inSize / windowSize);\n    this.outputShape = shape.slice(0, -1);\n    if (outSize > 1) {\n      this.outputShape.push(outSize);\n    }\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    const outShape = this.outputShape;\n    const rank = outShape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n\n    let sourceLocSetup;\n    let sourceRank;\n    if (outSize === 1) {\n      sourceRank = rank + 1;\n      const sourceLocDType = getCoordsDataType(sourceRank);\n      sourceLocSetup = `\n        ${sourceLocDType} sourceLocR = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 1]};\n        ${sourceLocDType} sourceLocG = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 2]};\n        ${sourceLocDType} sourceLocA = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 1]};\n        ${sourceLocDType} sourceLocB = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 2]};`;\n    } else {\n      sourceRank = rank;\n      sourceLocSetup = `\n        ${dtype} sourceLocR = coords;\n        ++${coords[rank - 1]};\n        ${dtype} sourceLocG = coords;\n        ++${coords[rank - 2]};\n        ${dtype} sourceLocA = coords;\n        --${coords[rank - 1]};\n        ${dtype} sourceLocB = coords;\n        --${coords[rank - 2]};`;\n    }\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, sourceRank);\n    const inChannel = '.' + channels[sourceRank - 1];  // e.g. \".b\" for rank 3.\n    const intChannels = channels.map(x => 'int ' + x);\n    const srcRCoords =\n        getChannels('sourceLocR', sourceRank - 1).concat('inIdx.r');\n    const srcGCoords =\n        getChannels('sourceLocG', sourceRank - 1).concat('inIdx.g');\n    const srcBCoords =\n        getChannels('sourceLocB', sourceRank - 1).concat('inIdx.b');\n    const srcACoords =\n        getChannels('sourceLocA', sourceRank - 1).concat('inIdx.a');\n\n    const compOp = (op === 'max') ? 'greaterThan' : 'lessThan';\n    const fetchCandidateIdx = firstPass ? '' : `\n          inIdx = round(vec4(getBestIndicesAChannel(${srcRCoords.join()}),\n                             getBestIndicesAChannel(${srcGCoords.join()}),\n                             getBestIndicesAChannel(${srcBCoords.join()}),\n                             getBestIndicesAChannel(${srcACoords.join()})));`;\n\n    const fetchValue = `vec4(\n            getAChannel(${srcRCoords.join()}),\n            hasNextCol ? getAChannel(${srcGCoords.join()}) : 0.,\n            hasNextRow ? getAChannel(${srcBCoords.join()}) : 0.,\n            hasNextRow && hasNextCol ? getAChannel(${srcACoords.join()}) : 0.)`;\n\n    const getBestIndicesAChannelSnippet = firstPass ? '' : `\n      float getBestIndicesAChannel(${intChannels.join()}) {\n        return getChannel(getBestIndicesA(${channels.join()}),\n                                          vec2(${channels.slice(-2).join()}));\n      }`;\n\n    this.userCode = `\n      float getAChannel(${intChannels.join()}) {\n        return getChannel(getA(${channels.join()}),\n                               vec2(${channels.slice(-2).join()}));\n      }\n      ${getBestIndicesAChannelSnippet}\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        bool hasNextCol = ${coords[rank - 1]} < ${outShape[rank - 1] - 1};\n        bool hasNextRow = ${coords[rank - 2]} < ${outShape[rank - 2] - 1};\n        ${sourceLocSetup}\n        ivec4 srcIdx = ivec4(sourceLocR${inChannel}, sourceLocG${inChannel},\n          sourceLocB${inChannel}, sourceLocA${inChannel}) * ${windowSize};\n        ivec4 inIdx = srcIdx;\n        vec4 bestIndex = vec4(inIdx);\n        vec4 bestValue = ${fetchValue};\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          inIdx = srcIdx;\n          ${fetchCandidateIdx}\n          vec4 candidate = ${fetchValue};\n          bvec4 nan = isnan(candidate);\n          bvec4 replace = bvec4(\n            vec4(${compOp}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));\n\n          bestValue = vec4(replace.x  ? candidate.x : bestValue.x,\n                           replace.y  ? candidate.y : bestValue.y,\n                           replace.z  ? candidate.z : bestValue.z,\n                           replace.w  ? candidate.w : bestValue.w);\n          bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));\n          srcIdx++;\n        }\n        setOutput(bestIndex);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AvgPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth};\n            wC+= ${dilationWidth}) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n\n            dotProd += dyValue * avgMultiplier;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class AvgPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n\n              dotProd += dyValue * avgMultiplier;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\n// (Ar + Ai)(Br + Bi) =\n// ArBr + ArBi + AiBr + AiBi = ArBr - AB + ArBi + AiBr\n// Yr = ArBr - AB\n// Yi = ArBi + AiBr\nexport const COMPLEX_MULTIPLY = {\n  REAL: 'return areal * breal - aimag * bimag;',\n  IMAG: 'return areal * bimag + aimag * breal;'\n};\n\nexport class BinaryOpComplexProgram implements GPGPUProgram {\n  variableNames = ['AReal', 'AImag', 'BReal', 'BImag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    this.userCode = `\n      float binaryOpComplex(\n          float areal, float aimag, float breal, float bimag) {\n        ${op}\n      }\n\n      void main() {\n        float areal = getARealAtOutCoords();\n        float aimag = getAImagAtOutCoords();\n        float breal = getBRealAtOutCoords();\n        float bimag = getBImagAtOutCoords();\n        setOutput(binaryOpComplex(areal, aimag, breal, bimag));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nconst CHECK_NAN_SNIPPET = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const ADD = 'return a + b;';\nexport const SUB = 'return a - b;';\nexport const MUL = 'return a * b;';\n\n// We use native integer division to deal with floating point imprecision. Since\n// we implement floor division and glsl implements truncated division, we\n// correct for this by subtracting 1 from result when the result is negative and\n// there is a remainder.\nexport const INT_DIV = `\n  float s = sign(a) * sign(b);\n  int ia = round(a);\n  int ib = round(b);\n  if (ib != 0) {\n    // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n    return float(idiv(ia, ib, s));\n  } else {\n    return NAN;\n  }\n`;\n\nexport const POW = `\nif(a < 0.0 && floor(b) < b){\n  return NAN;\n}\nif (b == 0.0) {\n  return 1.0;\n}\nreturn (round(mod(b, 2.0)) != 1) ?\n    pow(abs(a), b) : sign(a) * pow(abs(a), b);\n`;\nexport const SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const EQUAL = `return float(a == b);`;\n\nexport const NOT_EQUAL = `return float(a != b);`;\n\nexport const LESS = `return float(a < b);`;\n\nexport const LESS_EQUAL = `return float(a <= b);`;\n\nexport const GREATER = `return float(a > b);`;\n\nexport const GREATER_EQUAL = `return float(a >= b);`;\n\nexport const LOGICAL_AND = `return float(a >= 1.0 && b >= 1.0);`;\n\nexport const LOGICAL_OR = `return float(a >= 1.0 || b >= 1.0);`;\n\nexport const MAX = CHECK_NAN_SNIPPET + `\n  return max(a, b);\n`;\nexport const MIN = CHECK_NAN_SNIPPET + `\n  return min(a, b);\n`;\nexport const MOD = `if (b == 0.0) return NAN;\n  return mod(a, b);`;\n\nexport const ELU_DER = `return (b >= 1.0) ? a : a * (b + 1.0);`;\n\nexport const PRELU = `return (a < 0.) ? b * a : a;`;\n\nexport class BinaryOpProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    this.userCode = `\n      float binaryOperation(float a, float b) {\n        ${op}\n      }\n\n      void main() {\n        float a = getAAtOutCoords();\n        float b = getBAtOutCoords();\n        setOutput(binaryOperation(a, b));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nconst CHECK_NAN_SNIPPET = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\nexport const INT_DIV = `\n  ivec4 ia = round(a);\n  ivec4 ib = round(b);\n  bvec4 cond = notEqual(ib, ivec4(0));\n  ivec4 result = ivec4(0);\n  vec4 s = sign(a) * sign(b);\n\n  // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n  if (cond[0]) {\n    result[0] = idiv(ia[0], ib[0], s[0]);\n  }\n  if (cond[1]) {\n    result[1] = idiv(ia[1], ib[1], s[1]);\n  }\n  if (cond[2]) {\n    result[2] = idiv(ia[2], ib[2], s[2]);\n  }\n  if (cond[3]) {\n    result[3] = idiv(ia[3], ib[3], s[3]);\n  }\n  return vec4(result);\n`;\n\nexport const POW = `\n  // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.\n  vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));\n  vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);\n  vec4 result = multiplier * pow(abs(a), b);\n\n  // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS\n  bvec4 isExpZero = equal(b, vec4(0.0));\n  result.r = isExpZero.r ? 1.0 : result.r;\n  result.g = isExpZero.g ? 1.0 : result.g;\n  result.b = isExpZero.b ? 1.0 : result.b;\n  result.a = isExpZero.a ? 1.0 : result.a;\n\n  vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const PRELU = `\n  vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));\n  return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);\n`;\n\nexport const ELU_DER = `\n  vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));\n  return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));\n`;\n\nexport const EQUAL = `\n  return vec4(equal(a, b));\n`;\n\nexport const NOT_EQUAL = `\n  return vec4(notEqual(a, b));\n`;\n\nexport const LESS = `\n  return vec4(lessThan(a, b));\n`;\n\nexport const LESS_EQUAL = `\n  return vec4(lessThanEqual(a, b));\n`;\n\nexport const GREATER = `\n  return vec4(greaterThan(a, b));\n`;\n\nexport const GREATER_EQUAL = `\n  return vec4(greaterThanEqual(a, b));\n`;\n\nexport const LOGICAL_AND = `\n  return vec4(\n    vec4(greaterThanEqual(a, vec4(1.0))) *\n    vec4(greaterThanEqual(b, vec4(1.0))));\n`;\n\nexport const LOGICAL_OR = `\n  return min(\n    vec4(greaterThanEqual(a, vec4(1.0))) +\n    vec4(greaterThanEqual(b, vec4(1.0))),\n    vec4(1.0));\n`;\n\nexport const MAX = `\n  vec4 result = vec4(max(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MIN = `\n  vec4 result = vec4(min(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MOD = `\n  vec4 result = mod(a, b);\n  vec4 isNaN = vec4(equal(b, vec4(0.0)));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport class BinaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n  supportsBroadcasting = true;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      op: string, aShape: number[], bShape: number[],\n      checkOutOfBounds = false) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const rank = this.outputShape.length;\n    let checkOutOfBoundsString = '';\n    if (checkOutOfBounds) {\n      if (rank === 0 || util.sizeFromShape(this.outputShape) === 1) {\n        checkOutOfBoundsString = `\n          result.y = 0.;\n          result.z = 0.;\n          result.w = 0.;\n        `;\n      } else {\n        const dtype = getCoordsDataType(rank);\n        checkOutOfBoundsString = `\n          ${dtype} coords = getOutputCoords();\n        `;\n        if (rank === 1) {\n          checkOutOfBoundsString += `\n            result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;\n            result.z = 0.;\n            result.w = 0.;\n          `;\n        } else {\n          const channels = getChannels('coords', rank);\n          checkOutOfBoundsString += `\n            bool nextRowOutOfBounds =\n              (${channels[rank - 2]} + 1) >= ${this.outputShape[rank - 2]};\n            bool nextColOutOfBounds =\n              (${channels[rank - 1]} + 1) >= ${this.outputShape[rank - 1]};\n            result.y = nextColOutOfBounds ? 0. : result.y;\n            result.z = nextRowOutOfBounds ? 0. : result.z;\n            result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;\n          `;\n        }\n      }\n    }\n\n    this.userCode = `\n      vec4 binaryOperation(vec4 a, vec4 b) {\n        ${op}\n      }\n\n      void main() {\n        vec4 a = getAAtOutCoords();\n        vec4 b = getBAtOutCoords();\n\n        vec4 result = binaryOperation(a, b);\n        ${checkOutOfBoundsString}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        float value = getAAtOutCoords();\n        if (isnan(value)) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, minVal, maxVal));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        vec4 value = getAAtOutCoords();\n\n        if (any(isnan(value))) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, vec4(minVal), vec4(maxVal)));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ComplexAbsProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(shape: number[]) {\n    this.outputShape = shape;\n    this.userCode = `\n      void main() {\n        float re = abs(getRealAtOutCoords());\n        float im = abs(getImagAtOutCoords());\n        float mx = max(re, im);\n\n        // sadly the length function in glsl is not underflow-safe\n        // (at least not on Intel GPUs). So the safe solution is\n        // to ensure underflow-safety in all cases.\n        setOutput(\n          mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))\n        );\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ConcatProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  // Concats 2d tensors along axis=1. See comments in MathBackendWebGL.concat().\n  constructor(shapes: Array<[number, number]>) {\n    this.outputShape = backend_util.computeOutShape(shapes, 1 /* axis */);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][1];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][1];\n    }\n\n    const snippets = [`if (yC < ${offsets[0]}) setOutput(getT0(yR, yC));`];\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      snippets.push(\n          `else if (yC < ${offsets[i]}) ` +\n          `setOutput(getT${i}(yR, yC-${shift}));`);\n    }\n    const lastIndex = offsets.length;\n    const lastShift = offsets[offsets.length - 1];\n    snippets.push(`else setOutput(getT${lastIndex}(yR, yC-${lastShift}));`);\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int yR = coords.x;\n        int yC = coords.y;\n\n        ${snippets.join('\\n        ')}\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ConcatPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(shapes: number[][], axis: number) {\n    this.outputShape = backend_util.computeOutShape(shapes, axis);\n    const shape = this.outputShape;\n    const rank = shape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][axis];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][axis];\n    }\n\n    const channel = channels[axis];\n    const lastChannels = channels.slice(-2);\n    const allChannels = channels.join();\n\n    let getValueSnippet = `if (${channel} < ${offsets[0]}) {\n        return getChannel(\n            getT0(${allChannels}), vec2(${lastChannels.join()}));\n        }`;\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      // Note: the >= comparison below may seem unnecessary given the check\n      // above but is needed to workaround branch execution issues on some\n      // devices. It makes all the conditions exclusive without relying on\n      // execution order.\n      getValueSnippet += `\n        if (${channel} < ${offsets[i]}  && ${channel} >= ${offsets[i - 1]}) {\n          return getChannel(\n            getT${i}(${shiftedChannels(channels, channel, shift)}),\n            vec2(${shiftedChannels(lastChannels, channel, shift)}));\n        }`;\n    }\n    const lastIndex = offsets.length;\n    const shift = offsets[offsets.length - 1];\n    getValueSnippet += `\n        return getChannel(\n          getT${lastIndex}(${shiftedChannels(channels, channel, shift)}),\n          vec2(${shiftedChannels(lastChannels, channel, shift)}));`;\n\n    this.userCode = `\n      float getValue(${channels.map(x => 'int ' + x)}) {\n        ${getValueSnippet}\n      }\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        vec4 result = vec4(getValue(${coords}), 0., 0., 0.);\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} + 1;\n        if (${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.g = getValue(${coords});\n        }\n\n        ${coords[rank - 2]} = ${coords[rank - 2]} + 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]}) {\n          result.a = getValue(${coords});\n        }\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} - 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]} &&\n            ${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.b = getValue(${coords});\n        }\n        setOutput(result);\n      }\n    `;\n  }\n}\n\n/**\n * Return an expression for coordinates into a vector where a given channel\n * will be offset by [shift].\n *\n * @param channels the channels to consider\n * @param channel the channel we want shifted\n * @param shift  the amount to subtract from the channel.\n *\n * @returns a string of the form 'x, y-[shift], z' where any one channel can\n * have the shift applied.\n */\nfunction shiftedChannels(channels: string[], channel: string, shift: number) {\n  const channelIdx = channels.indexOf(channel);\n  const res = channels.map((c, idx) => {\n    if (idx === channelIdx) {\n      return `${c} - ${shift}`;\n    } else {\n      return c;\n    }\n  });\n  return res.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int d2 = coords.w;\n\n        // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              if (${isChannelsLast}) {\n                float dyValue = getDy(b, yR, yC, d2);\n                float xValue = getX(b, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              } else {\n                float dyValue = getDy(b, d2, yR, yC);\n                float xValue = getX(b, d1, xR, xC);\n                dotProd += (xValue * dyValue);\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[${channelDim}];\n\n        ivec2 dyCorner = ivec2(coords[${rowDim}], coords[${colDim}]) - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n\n              if (${isChannelsLast}) {\n                float xValue = getDy(batch, idyR, idyC, d2);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              } else {\n                float xValue = getDy(batch, d2, idyR, idyC);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    this.userCode = `\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int wF = coords.x;\n        int wR = coords.y;\n        int wC = coords.z;\n        int d1 = coords.w;\n        int d2 = coords.u;\n\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yF = 0; yF < ${convInfo.outDepth}; yF++) {\n            int xF = wF + yF * ${strideDepth} - ${padFront};\n\n            if (xF < 0 || xF >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n              int xR = wR + yR * ${strideHeight} - ${padTop};\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n                int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float dyValue = getDy(b, yF, yR, yC, d2);\n                float xValue = getX(b, xF, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padFront = filterDepth - 1 - convInfo.padInfo.front;\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d1 = coords.u;\n\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyFCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          float dyF = float(dyFCorner + wF) / ${strideDepth}.0;\n\n          if (dyF < 0.0 || dyF >= ${convInfo.outDepth}.0 || fract(dyF) > 0.0) {\n            continue;\n          }\n          int idyF = int(dyF);\n\n          int wFPerm = ${filterDepth} - 1 - wF;\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n              fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            int wRPerm = ${filterHeight} - 1 - wR;\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              int wCPerm = ${filterWidth} - 1 - wC;\n\n              for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n                float xValue = getDy(batch, idyF, idyR, idyC, d2);\n                float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int dm = coords.w;\n        int d2 = d1 * ${channelMul} + dm;\n\n        float dotProd = 0.0;\n\n        // TO DO: Vec4 over the batch size\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float dyValue = getDy(b, yR, yC, d2);\n              float xValue = getX(b, xR, xC, d1);\n              dotProd += (xValue * dyValue);\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class DepthwiseConv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[3];\n        ivec2 dyCorner = coords.yz - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        float dotProd = 0.0;\n\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            // TO DO: Vec4 over the channelMul\n            for (int dm = 0; dm < ${channelMul}; dm++) {\n              int d2 = d1 * ${channelMul} + dm;\n              float xValue = getDy(batch, idyR, idyC, d2);\n              float wValue = getW(wRPerm, wCPerm, d1, dm);\n              dotProd += xValue * wValue;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivationWeights = false) {\n    this.outputShape = convInfo.outShape;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivationWeights) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivationWeights) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d2 = coords[${channelDim}];\n\n        ivec2 xRCCorner =\n            ivec2(coords[${rowDim}], coords[${colDim}]) * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${convInfo.inWidth}) {\n              continue;\n            }\n\n            for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n              vec4 wValues = vec4(\n                getW(wR, wC, d1, d2),\n                getW(wR, wC, d1 + 1, d2),\n                getW(wR, wC, d1 + 2, d2),\n                getW(wR, wC, d1 + 3, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec4 xValues = vec4(\n                  getX(batch, xR, xC, d1),\n                  getX(batch, xR, xC, d1 + 1),\n                  getX(batch, xR, xC, d1 + 2),\n                  getX(batch, xR, xC, d1 + 3)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec4 xValues = vec4(\n                  getX(batch, d1, xR, xC),\n                  getX(batch, d1 + 1, xR, xC),\n                  getX(batch, d1 + 2, xR, xC),\n                  getX(batch, d1 + 3, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n\n            if (${inputDepthVec4Remainder === 1}) {\n\n              if (${isChannelsLast}) {\n                dotProd +=\n                    getX(batch, xR, xC, ${inputDepthNearestVec4}) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              } else {\n                dotProd +=\n                    getX(batch, ${inputDepthNearestVec4}, xR, xC) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              }\n\n            } else if (${inputDepthVec4Remainder === 2}) {\n              vec2 wValues = vec2(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec2 xValues = vec2(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec2 xValues = vec2(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            } else if (${inputDepthVec4Remainder === 3}) {\n              vec3 wValues = vec3(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 2, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec3 xValues = vec3(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec3 xValues = vec3(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 2, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            }\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nexport class Conv3DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.outShape;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n\n    this.userCode = `\n      const ivec3 strides = ivec3(${strideDepth}, ${strideHeight}, ${\n        strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d2 = coords.u;\n\n        ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xFCorner = xFRCCorner.x;\n        int xRCorner = xFRCCorner.y;\n        int xCCorner = xFRCCorner.z;\n\n        // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get\n        // y(yF, yR, yC, d2). ? = to be determined. : = across all\n        // values in that axis.\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          int xF = xFCorner + wF * ${dilationDepth};\n\n          if (xF < 0 || xF >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            int xR = xRCorner + wR * ${dilationHeight};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n                vec4 xValues = vec4(\n                  getX(batch, xF, xR, xC, d1),\n                  getX(batch, xF, xR, xC, d1 + 1),\n                  getX(batch, xF, xR, xC, d1 + 2),\n                  getX(batch, xF, xR, xC, d1 + 3)\n                );\n                vec4 wValues = vec4(\n                  getW(wF, wR, wC, d1, d2),\n                  getW(wF, wR, wC, d1 + 1, d2),\n                  getW(wF, wR, wC, d1 + 2, d2),\n                  getW(wF, wR, wC, d1 + 3, d2)\n                );\n\n                dotProd += dot(xValues, wValues);\n              }\n\n              if (${inputDepthVec4Remainder === 1}) {\n                dotProd +=\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}) *\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2);\n              } else if (${inputDepthVec4Remainder === 2}) {\n                vec2 xValues = vec2(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                vec2 wValues = vec2(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else if (${inputDepthVec4Remainder === 3}) {\n                vec3 xValues = vec3(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                vec3 wValues = vec3(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 2, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2 / ${channelMul};\n        int q = d2 - d1 * ${channelMul};\n\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${xNumRows}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${xNumCols}) {\n              continue;\n            }\n\n            float xVal = getX(batch, xR, xC, d1);\n            float wVal = getW(wR, wC, d1, q);\n            dotProd += xVal * wVal;\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConvPacked2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const texelsAcross = filterWidth;\n\n    let mainLoop = `int xR; int xC; int xCOffset;`;\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `\n          vec4 xTexelR${r}C${c * 2} = vec4(0.);\n          vec4 wR${r}C${c} = vec4(0.);\n          vec4 xR${r}C${c} = vec4(0.);`;\n      }\n    }\n\n    /**\n     * This vectorized implementation works by gathering the values needed for\n     * each output channel's dot product into vec4's and then multiplying them\n     * all together (this happens in the final double for-loop below). Most of\n     * the main loop consists of constructing these vec4's with the minimum\n     * number of texture2D calls, which means making use of all four returned\n     * values from a texture2D call at once.\n     */\n    for (let r = 0; r < filterHeight; r++) {\n      for (let texelC = 0; texelC < texelsAcross; texelC++) {\n        const c = texelC * 2;\n\n        mainLoop += `\n          xR = xRCorner + ${r * dilationHeight};\n          xC = xCCorner + ${c * dilationWidth};\n        `;\n\n        if (strideWidth === 1) {\n          if (c < filterWidth) {\n            // If padding is odd, the outer texels have to be composed.\n            if (padLeft % 2 === 1) {\n              // TODO: Ensure vec4 previous does not result in redundant sample,\n              // and avoid setting xTexelRC's that exceed the boundary in the\n              // first place rather than resetting them to vec4(0)).\n\n              // To compute xCOffset:\n              // - If padding is odd, we must add 1 to ensure we ask for an\n              // even-numbered row.\n              // - We subtract 2 to access the previous texel.\n\n              mainLoop += `\n                xCOffset = xC + 1;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    xTexelR${r}C${c}.zw = vec2(0.);\n                  }\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + 1 - 2;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  vec4 previous = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    previous.zw = vec2(0.);\n                  }\n\n                  xR${r}C${c} = vec4(previous.zw, xTexelR${r}C${c}.xy);\n                } else {\n                  xR${r}C${c} = vec4(0, 0, xTexelR${r}C${c}.xy);\n                }\n              `;\n            } else {\n              // Padding is even, so xRC corresponds to a single texel.\n              mainLoop += `\n                if(xR >= 0 && xR < ${xNumRows} && xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xR${r}C${c} = xTexelR${r}C${c};\n              `;\n            }\n\n            if (c + 1 < filterWidth) {\n              // If dilation is even, the second entry should match the first\n              // (either both are composed or both are single samples). But if\n              // dilation is odd, then the second entry should be the opposite\n              // of the first (if the first is composed, the second is a single\n              // sample, and vice versa.)\n\n              const nextTexelOffset = padLeft % 2 === 0 ?\n                  util.nearestLargerEven(dilationWidth) :\n                  dilationWidth;\n\n              if ((dilationWidth % 2 === 0 && padLeft % 2 === 1) ||\n                  (dilationWidth % 2 !== 0 && padLeft % 2 !== 1)) {\n                mainLoop += `\n                  xCOffset = xC + ${padLeft % 2} + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n                `;\n\n                // If dilation > 1 then the xRC's will not be able to share any\n                // values, so each xRC will require two unique calls to getX.\n                if (dilationWidth > 1) {\n                  mainLoop += `\n                    xCOffset -= 2;\n                    if(xR >= 0 && xR < ${xNumRows} &&\n                      xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                      xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                    } else {\n                      xTexelR${r}C${c} = vec4(0.);\n                    }\n                  `;\n                }\n\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.xy);\n                `;\n              } else {\n                mainLoop += `\n                  xCOffset = xC + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n\n                  xR${r}C${c + 1} = xTexelR${r}C${c + 2};\n                `;\n              }\n            }\n          }\n        } else {  // stride > 1\n          if (c < filterWidth) {\n            mainLoop += `\n              if(xR >= 0 && xR < ${xNumRows}) {\n            `;\n\n            // Depending on whether padLeft is even or odd, we want either the\n            // xy or zw channels from X texels for xR${r}C${c}. If padLeft is\n            // even, xR${r}C${c + 1} is simply the zw channels of texels we've\n            // already sampled. But if padLeft is odd, xR${r}C{$c + 1}.zw will\n            // need to come from the xy channels of a new texel, hence the `vec4\n            // final` initialized below.\n            if (padLeft % 2 === 1) {\n              mainLoop += `\n                xCOffset = xC + 1 - ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                if(xC + 1 >= 0 && xC + 1 < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xC + 1, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  vec4 final = vec4(0.);\n                  xCOffset = xC + 1 + ${strideWidth};\n                  if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    final = getX(batch, xR, xCOffset, d1);\n                  }\n                  xR${r}C${c + 1} = vec4(xTexelR${r}C${c + 2}.xy, final.xy);\n                `;\n              }\n            } else {\n              mainLoop += `\n                if(xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.xy, xTexelR${r}C${c + 2}.xy);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n                `;\n              }\n            }\n\n            mainLoop += `}`;\n          }\n        }\n\n        if (c < filterWidth) {\n          mainLoop += `\n            vec4 wTexelR${r}C${c} = getW(${r}, ${c}, d1, q);\n            wR${r}C${c} = vec4(wTexelR${r}C${c}.xz, wTexelR${r}C${c}.xz);\n          `;\n\n          if (c + 1 < filterWidth) {\n            mainLoop += `\n              vec4 wTexelR${r}C${c + 1} = getW(${r}, ${c + 1}, d1, q);\n              wR${r}C${c + 1} =\n                vec4(wTexelR${r}C${c + 1}.xz, wTexelR${r}C${c + 1}.xz);`;\n          }\n        }\n      }\n    }\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `dotProd += xR${r}C${c} * wR${r}C${c};`;\n      }\n    }\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2;\n        int q = 0;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        vec4 dotProd = vec4(0.);\n\n        ${mainLoop}\n\n        vec4 result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class CropAndResizeProgram implements GPGPUProgram {\n  variableNames = ['Image', 'Boxes', 'BoxInd'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], boxShape: [number, number],\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number) {\n    const [batch, imageHeight, imageWidth, depth] = imageShape;\n    const [numBoxes, ] = boxShape;\n    const [cropHeight, cropWidth] = cropSize;\n    this.outputShape = [numBoxes, cropHeight, cropWidth, depth];\n    const methodId = method === 'bilinear' ? 1 : 0;\n\n    const [inputHeightFloat, inputWidthFloat] =\n        [`${imageHeight - 1}.0`, `${imageWidth - 1}.0`];\n\n    const [heightRatio, heightScale, inY] = cropHeight > 1 ?\n        [\n          `${(imageHeight - 1) / (cropHeight - 1)}`,\n          '(y2-y1) * height_ratio',\n          `y1*${inputHeightFloat} + float(y)*(height_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (y1+y2) * ${inputHeightFloat}`,\n        ];\n    const [widthRatio, widthScale, inX] = cropWidth > 1 ?\n        [\n          `${(imageWidth - 1) / (cropWidth - 1)}`,\n          '(x2-x1) * width_ratio',\n          `x1*${inputWidthFloat} + float(x)*(width_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (x1+x2) * ${inputWidthFloat}`,\n        ];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc\n    this.userCode = `\n      const float height_ratio = float(${heightRatio});\n      const float width_ratio = float(${widthRatio});\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int y = coords[1];\n        int x = coords[2];\n        int d = coords[3];\n\n        // get box vals\n        float y1 = getBoxes(b,0);\n        float x1 = getBoxes(b,1);\n        float y2 = getBoxes(b,2);\n        float x2 = getBoxes(b,3);\n\n        // get image in batch index\n        int bInd = round(getBoxInd(b));\n        if(bInd < 0 || bInd >= ${batch}) {\n          return;\n        }\n\n        float height_scale = ${heightScale};\n        float width_scale = ${widthScale};\n\n        float in_y = ${inY};\n        if( in_y < 0.0 || in_y > ${inputHeightFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n        float in_x = ${inX};\n        if( in_x < 0.0 || in_x > ${inputWidthFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n\n        vec2 sourceFracIndexCR = vec2(in_x,in_y);\n        if(${methodId} == 1) {\n          // Compute the four integer indices.\n          ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);\n          ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));\n\n          float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);\n          float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);\n          float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);\n          float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);\n\n          vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);\n\n          float top = topLeft + (topRight - topLeft) * fracCR.x;\n          float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;\n          float newValue = top + (bottom - top) * fracCR.y;\n          setOutput(newValue);\n        } else {\n          // Compute the coordinators of nearest neighbor point.\n          ivec2 sourceNearestCR = ivec2(floor(\n            sourceFracIndexCR + vec2(0.5,0.5)));\n          float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);\n          setOutput(newValue);\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class CumSumProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  index: WebGLUniformLocation;\n\n  constructor(shape: number[], exclusive: boolean, reverse: boolean) {\n    this.outputShape = shape;\n    const rank = shape.length;\n    const val = exclusive ? '0.0' : `getX(${getCoords(rank, 'coords')})`;\n    const length = shape[shape.length - 1];\n    let condition = '';\n    let idxString = '';\n    // When exclusive is set, the cumsum op becomes roll op that copies the\n    // value from the previous index based on the direction specified by the\n    // reverse flag.\n    if (exclusive) {\n      condition = reverse ? `end != ${length - 1}` : 'end != 0';\n      idxString = reverse ? 'end + 1' : 'end - 1';\n    } else {\n      condition = reverse ? `end + pow2 < ${length}` : 'end >= pow2';\n      idxString = (reverse ? 'end + pow2' : 'end - pow2');\n    }\n\n    this.userCode = `\n      uniform float index;\n      void main() {\n        ${getCoordsDataType(rank)} coords = getOutputCoords();\n        int end = ${getFinalCoord(rank, 'coords')};\n        float val = ${val};\n        int pow2 = int(pow(2.0, index));\n        if (${condition}) {\n          int idx = ${idxString};\n          ${getFinalCoord(rank, 'coords')} = idx;\n          val += getX(${getCoords(rank, 'coords')});\n        }\n        setOutput(val);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(index: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.index == null) {\n        this.index = gpgpu.getUniformLocation(webGLProgram, 'index');\n      }\n      gpgpu.gl.uniform1f(this.index, index);\n    };\n  }\n}\n\nfunction getCoords(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.x, ${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.x, ${name}.y, ${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.x, ${name}.y, ${name}.z, ${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n\nfunction getFinalCoord(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: [number, number, number];\n  packedInputs = false;\n  packedOutput = true;\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getA(rc.x, rc.y, rc.z);\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: [number, number, number];\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthToSpaceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  blockSize: number;\n  dataFormat: string;\n\n  constructor(\n      outputShape: number[], blockSize: number, dataFormat: 'NHWC'|'NCHW') {\n    this.outputShape = outputShape;\n    this.blockSize = blockSize;\n    this.dataFormat = dataFormat;\n    this.userCode = `\n    void main() {\n      ivec4 coords = getOutputCoords();\n      int b = coords[0];\n      int h = ${this.getHeightCoordString()};\n      int w = ${this.getWidthCoordString()};\n      int d = ${this.getDepthCoordString()};\n\n      int in_h = h / ${blockSize};\n      int offset_h = imod(h, ${blockSize});\n      int in_w = w / ${blockSize};\n      int offset_w = imod(w, ${blockSize});\n      int offset_d = (offset_h * ${blockSize} + offset_w) *\n        ${this.getOutputDepthSize()};\n      int in_d = d + offset_d;\n\n      float result = ${this.getInputSamplingString()};\n      setOutput(result);\n    }\n  `;\n  }\n\n  private getHeightCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[1]`;\n    } else {\n      return `coords[2]`;\n    }\n  }\n\n  private getWidthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[2]`;\n    } else {\n      return `coords[3]`;\n    }\n  }\n\n  private getDepthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[3]`;\n    } else {\n      return `coords[1]`;\n    }\n  }\n\n  private getOutputDepthSize(): number {\n    if (this.dataFormat === 'NHWC') {\n      return this.outputShape[3];\n    } else {\n      return this.outputShape[1];\n    }\n  }\n\n  private getInputSamplingString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `getX(b, in_h, in_w, in_d)`;\n    } else {\n      return `getX(b, in_d, in_h, in_w)`;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DiagProgram implements GPGPUProgram {\n  variableNames = ['X'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(size: number) {\n    this.outputShape = [size, size];\n    this.userCode = `\n      void main() {\n          ivec2 coords = getOutputCoords();\n          float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;\n          setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        float x = getAAtOutCoords();\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = false;\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: [number, number, number]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n        float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class EncodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let output = `result`;\n    if (inputIsUnsignedByte) {\n      output = `floor(result * 255. + 0.5)`;\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        int flatIndex = getFlatIndex(coords);\n        int offset = imod(flatIndex, 4);\n\n        flatIndex = idiv(flatIndex, 4, 1.);\n\n        int r = flatIndex / ${width};\n        int c = imod(flatIndex, ${width});\n        vec2 uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n        vec4 values = ${glsl.texture2D}(A, uv);\n\n        float result;\n\n        if(offset == 0) {\n          result = values[0];\n        } else if(offset == 1) {\n          result = values[1];\n        } else if(offset == 2) {\n          result = values[2];\n        } else {\n          result = values[3];\n        }\n\n        ${glsl.output} = vec4(${output}, 0., 0., 0.);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\n/*\nThis is how the shader encodes a tensor with shape = [2, 3, 5]\n(indices are [batch, row, col]).\n\n000|001   002|003   004|xxx   020|021   022|023   024|xxx\n-------   -------   -------   -------   -------   -------\n010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\n100|101   102|103   104|xxx   120|121   122|123   124|xxx\n-------   -------   -------   -------   -------   -------\n110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\nSingle texels contain only values from the same batch, and from adjacent rows\nand columns.\n */\n\nexport class EncodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let mainLoop = '';\n    let output = 'result';\n    if (inputIsUnsignedByte) {\n      output = 'floor(result * 255. + 0.5)';\n    }\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        const channel = row * 2 + col;\n\n        mainLoop += `\n          localCoords = coords;\n          if(localCoords[2] + ${col} < ${outputShape[2]}) {\n            localCoords[2] += ${col};\n            if(localCoords[1] + ${row} < ${outputShape[1]}) {\n              localCoords[1] += ${row};\n\n              flatIndex = getFlatIndex(localCoords);\n              offset = imod(flatIndex, 4);\n\n              flatIndex = idiv(flatIndex, 4, 1.);\n\n              r = flatIndex / ${width};\n              c = imod(flatIndex, ${width});\n              uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n              values = ${glsl.texture2D}(A, uv);\n\n              if(offset == 0) {\n                result[${channel}] = values[0];\n              } else if(offset == 1) {\n                result[${channel}] = values[1];\n              } else if(offset == 2) {\n                result[${channel}] = values[2];\n              } else {\n                result[${channel}] = values[3];\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        vec4 result = vec4(0.);\n        int flatIndex, r, c, offset;\n        ivec3 localCoords;\n        vec2 uv;\n        vec4 values;\n\n        ${mainLoop}\n\n        ${glsl.output} = ${output};\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const COMPLEX_FFT = {\n  REAL: 'return real * expR - imag * expI;',\n  IMAG: 'return real * expI + imag * expR;'\n};\n\nexport class FFTProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, inputShape: [number, number], inverse: boolean) {\n    const innerDim = inputShape[1];\n    this.outputShape = inputShape;\n\n    const exponentMultiplierSnippet =\n        inverse ? `2.0 * ${Math.PI}` : `-2.0 * ${Math.PI}`;\n    const resultDenominator = inverse ? `${innerDim}.0` : '1.0';\n\n    this.userCode = `\n      const float exponentMultiplier = ${exponentMultiplierSnippet};\n\n      float unaryOpComplex(float real, float expR, float imag, float expI) {\n        ${op}\n      }\n\n      float mulMatDFT(int batch, int index) {\n        float indexRatio = float(index) / float(${innerDim});\n        float exponentMultiplierTimesIndexRatio =\n            exponentMultiplier * indexRatio;\n\n        float result = 0.0;\n\n        for (int i = 0; i < ${innerDim}; i++) {\n          // x = (-2|2 * PI / N) * index * i;\n          float x = exponentMultiplierTimesIndexRatio * float(i);\n          float expR = cos(x);\n          float expI = sin(x);\n          float real = getReal(batch, i);\n          float imag = getImag(batch, i);\n\n          result +=\n              unaryOpComplex(real, expR, imag, expI) / ${resultDenominator};\n        }\n\n        return result;\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        setOutput(mulMatDFT(coords[0], coords[1]));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FillProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  valueLoc: WebGLUniformLocation;\n\n  constructor(shape: number[], value: number) {\n    this.variableNames = ['x'];\n    this.outputShape = shape;\n\n    this.userCode = `\n      uniform float value;\n      void main() {\n        // Input can be obtained from uniform value.\n        setOutput(value);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(value: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.valueLoc == null) {\n        this.valueLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'value');\n      }\n      gpgpu.gl.uniform1f(this.valueLoc, value);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherProgram implements GPGPUProgram {\n  variableNames = ['A', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], indicesLength: number, axis: number) {\n    const outputShape: number[] = aShape.slice();\n    outputShape[axis] = indicesLength;\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape, axis);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[], axis: number): string {\n  const rank = aShape.length;\n  if (rank > 4) {\n    throw Error(`Gather for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `int(getIndices(resRC))`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    if (i === axis) {\n      sourceCoords.push(`int(getIndices(${currentCoords[i]}))`);\n    } else {\n      sourceCoords.push(`${currentCoords[i]}`);\n    }\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherNDProgram implements GPGPUProgram {\n  variableNames = ['x', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  constructor(\n      private sliceDim: number, private strides: number[], shape: number[]) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    const strideString = this.sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${this.strides});\n         void main() {\n          ${dtype} coords = getOutputCoords();\n          int flattenIndex = 0;\n          for (int j = 0; j < ${this.sliceDim}; j++) {\n            int index = round(getIndices(coords[0], j));\n            flattenIndex += index * ${strideString};\n          }\n          setOutput(getX(flattenIndex, coords[1]));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PixelData, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {getGlslDifferences} from './glsl_version';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport * as webgl_util from './webgl_util';\n\nexport function createVertexShader(gl: WebGLRenderingContext): WebGLShader {\n  const glsl = getGlslDifferences();\n  const vertexShaderSource = `${glsl.version}\n    precision highp float;\n    ${glsl.attribute} vec3 clipSpacePos;\n    ${glsl.attribute} vec2 uv;\n    ${glsl.varyingVs} vec2 resultUV;\n\n    void main() {\n      gl_Position = vec4(clipSpacePos, 1);\n      resultUV = uv;\n    }`;\n  return webgl_util.createVertexShader(gl, vertexShaderSource);\n}\n\nexport function createVertexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // [x y z u v] * [upper-left, lower-left, upper-right, lower-right]\n  const vertexArray = new Float32Array(\n      [-1, 1, 0, 0, 1, -1, -1, 0, 0, 0, 1, 1, 0, 1, 1, 1, -1, 0, 1, 0]);\n  return webgl_util.createStaticVertexBuffer(gl, vertexArray);\n}\n\nexport function createIndexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // OpenGL (and WebGL) have \"CCW == front\" winding\n  const triangleVertexIndices = new Uint16Array([0, 1, 2, 2, 1, 3]);\n  return webgl_util.createStaticIndexBuffer(gl, triangleVertexIndices);\n}\n\nfunction createAndConfigureTexture(\n    gl: WebGLRenderingContext, width: number, height: number,\n    internalFormat: number, textureFormat: number,\n    textureType: number): WebGLTexture {\n  webgl_util.validateTextureSize(width, height);\n  const texture = webgl_util.createTexture(gl);\n\n  const tex2d = gl.TEXTURE_2D;\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(tex2d, texture));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MIN_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MAG_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          tex2d, 0, internalFormat, width, height, 0, textureFormat,\n          textureType, null));\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n  return texture;\n}\n\nexport function getInternalFormatForFloat32MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatFloat;\n}\n\nexport function createFloat32MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat32MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatHalfFloat;\n}\n\nexport function createFloat16MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, textureConfig.textureTypeHalfFloat);\n}\n\nexport function getInternalFormatForUnsignedBytesMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.downloadTextureFormat;\n}\n\nexport function createUnsignedBytesMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForUnsignedBytesMatrixTexture(textureConfig), gl.RGBA,\n      gl.UNSIGNED_BYTE);\n}\n\nexport function getInternalFormatForPackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedFloat;\n}\n\nexport function createPackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height, getInternalFormatForPackedMatrixTexture(textureConfig),\n      gl.RGBA, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16PackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedHalfFloat;\n}\n\nexport function createFloat16PackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16PackedMatrixTexture(textureConfig), gl.RGBA,\n      textureConfig.textureTypeHalfFloat);\n}\n\nexport function bindVertexProgramAttributeStreams(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    vertexBuffer: WebGLBuffer): boolean {\n  const posOffset = 0;               // x is the first buffer element\n  const uvOffset = 3 * 4;            // uv comes after [x y z]\n  const stride = (3 * 4) + (2 * 4);  // xyz + uv, each entry is 4-byte float.\n  webgl_util.callAndCheck(\n      gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, vertexBuffer));\n  const success = webgl_util.bindVertexBufferToProgramAttribute(\n      gl, program, 'clipSpacePos', vertexBuffer, 3, stride, posOffset);\n  return success &&\n      webgl_util.bindVertexBufferToProgramAttribute(\n          gl, program, 'uv', vertexBuffer, 2, stride, uvOffset);\n}\n\nexport function uploadDenseMatrixToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture, width: number,\n    height: number, data: TypedArray, textureConfig: TextureConfig) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n\n  let dataForUpload: TypedArray, texelDataType: number, internalFormat: number;\n  if (data instanceof Uint8Array) {\n    dataForUpload = new Uint8Array(width * height * 4);\n    texelDataType = gl.UNSIGNED_BYTE;\n    internalFormat = gl.RGBA;\n  } else {\n    dataForUpload = new Float32Array(width * height * 4);\n    texelDataType = gl.FLOAT;\n    internalFormat = textureConfig.internalFormatPackedFloat;\n  }\n\n  dataForUpload.set(data);\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          gl.TEXTURE_2D, 0, internalFormat, width, height, 0, gl.RGBA,\n          texelDataType, dataForUpload));\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function uploadPixelDataToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, pixels.width, pixels.height, 0, gl.RGBA,\n            gl.UNSIGNED_BYTE, (pixels as PixelData).data));\n  } else {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, gl.RGBA, gl.UNSIGNED_BYTE,\n            pixels as ImageData | HTMLImageElement | HTMLCanvasElement |\n                HTMLVideoElement));\n  }\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function createBufferFromOutputTexture(\n    gl2: WebGL2RenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLBuffer {\n  // Create and bind the buffer.\n  const buffer = gl2.createBuffer();\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer));\n\n  // Initialize the buffer to the size of the texture in bytes.\n  const bytesPerFloat = 4;\n  const valuesPerTexel = 4;\n  const bufferSizeBytes = bytesPerFloat * valuesPerTexel * rows * columns;\n\n  webgl_util.callAndCheck(\n      gl2,\n      () => gl2.bufferData(\n          gl2.PIXEL_PACK_BUFFER, bufferSizeBytes, gl2.STREAM_READ));\n\n  // Enqueue a command on the GPU command queue to copy of texture into the\n  // buffer.\n  webgl_util.callAndCheck(\n      gl2, () => gl2.readPixels(0, 0, columns, rows, gl2.RGBA, gl2.FLOAT, 0));\n\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null));\n\n  return buffer;\n}\n\nexport function downloadFloat32MatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer,\n    size: number): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget = new Float32Array(size);\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadByteEncodedFloatMatrixFromOutputTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig) {\n  const [w, h] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n\n  const numChannels = 4;\n  const downloadTarget = new Uint8Array(\n      tex_util.getUnpackedArraySizeFromMatrixSize(rows * columns, numChannels));\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, w, h, textureConfig.downloadTextureFormat, gl.UNSIGNED_BYTE,\n          downloadTarget));\n\n  // By wrapping the buffer in a Float32Array, we use native browser IEEE 754\n  // decoding of the 4 bytes that back each 32 bit float.\n  return new Float32Array(downloadTarget.buffer);\n}\n\nexport function downloadPackedMatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer, batch: number, rows: number,\n    cols: number, physicalRows: number, physicalCols: number,\n    textureConfig: TextureConfig): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget =\n      new Float32Array(tex_util.getPackedRGBAArraySizeFromMatrixShape(\n          physicalRows, physicalCols));\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadMatrixFromPackedOutputTexture(\n    gl: WebGLRenderingContext, physicalRows: number,\n    physicalCols: number): Float32Array {\n  const packedRGBA = new Float32Array(physicalRows * physicalCols * 4);\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, physicalCols, physicalRows, gl.RGBA, gl.FLOAT, packedRGBA));\n\n  return packedRGBA;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, PixelData, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext, setWebGLContext} from './canvas_util';\nimport * as gpgpu_util from './gpgpu_util';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport {WebGL1DisjointQueryTimerExtension, WebGL2DisjointQueryTimerExtension} from './webgl_types';\nimport * as webgl_util from './webgl_util';\n\nexport interface FenceContext {\n  query: WebGLQuery|WebGLSync;\n  isFencePassed(): boolean;\n}\n\nexport class GPGPUContext {\n  gl: WebGLRenderingContext;\n  textureFloatExtension: {};\n  textureHalfFloatExtension: {};\n  colorBufferFloatExtension: {};\n  colorBufferHalfFloatExtension: {};\n  disjointQueryTimerExtension: WebGL2DisjointQueryTimerExtension|\n      WebGL1DisjointQueryTimerExtension;\n  vertexBuffer: WebGLBuffer;\n  indexBuffer: WebGLBuffer;\n  framebuffer: WebGLFramebuffer;\n  outputTexture: WebGLTexture|null = null;\n  program: WebGLProgram|null = null;\n  private disposed = false;\n  private disjoint: boolean;\n  textureConfig: TextureConfig;\n\n  constructor(gl?: WebGLRenderingContext) {\n    const glVersion = env().getNumber('WEBGL_VERSION');\n    if (gl != null) {\n      this.gl = gl;\n      setWebGLContext(glVersion, gl);\n    } else {\n      this.gl = getWebGLContext(glVersion);\n    }\n    // WebGL 2.0 enables texture floats without an extension.\n    let COLOR_BUFFER_FLOAT = 'WEBGL_color_buffer_float';\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (env().getNumber('WEBGL_VERSION') === 1) {\n      const TEXTURE_FLOAT = 'OES_texture_float';\n      const TEXTURE_HALF_FLOAT = 'OES_texture_half_float';\n\n      this.textureFloatExtension =\n          webgl_util.getExtensionOrThrow(this.gl, TEXTURE_FLOAT);\n      if (webgl_util.hasExtension(this.gl, TEXTURE_HALF_FLOAT)) {\n        this.textureHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, TEXTURE_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support half float textures, yet the ' +\n            'environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n\n      this.colorBufferFloatExtension = this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, COLOR_BUFFER_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support color renderable half floats, yet ' +\n            'the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n    } else {\n      COLOR_BUFFER_FLOAT = 'EXT_color_buffer_float';\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_FLOAT)) {\n        this.colorBufferFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      } else if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      } else {\n        throw new Error('GL context does not support color renderable floats');\n      }\n    }\n\n    this.vertexBuffer = gpgpu_util.createVertexBuffer(this.gl);\n    this.indexBuffer = gpgpu_util.createIndexBuffer(this.gl);\n    this.framebuffer = webgl_util.createFramebuffer(this.gl);\n\n    this.textureConfig =\n        tex_util.getTextureConfig(this.gl, this.textureHalfFloatExtension);\n  }\n\n  private get debug(): boolean {\n    return env().getBool('DEBUG');\n  }\n\n  public dispose() {\n    if (this.disposed) {\n      return;\n    }\n    if (this.program != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound WebGLProgram.' +\n          ' This is probably a resource leak, delete the program with ' +\n          'GPGPUContext.deleteProgram before disposing.');\n    }\n    if (this.outputTexture != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound output matrix ' +\n          'texture.  This is probably a resource leak, delete the output ' +\n          'matrix texture with GPGPUContext.deleteMatrixTexture before ' +\n          'disposing.');\n    }\n    const gl = this.gl;\n    webgl_util.callAndCheck(gl, () => gl.finish());\n    webgl_util.callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteFramebuffer(this.framebuffer));\n    webgl_util.callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(\n        gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteBuffer(this.indexBuffer));\n    this.disposed = true;\n  }\n\n  public createFloat32MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat32MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createFloat16MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createUnsignedBytesMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createUnsignedBytesMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public uploadPixelDataToTexture(\n      texture: WebGLTexture,\n      pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadPixelDataToTexture(this.gl, texture, pixels);\n  }\n\n  public uploadDenseMatrixToTexture(\n      texture: WebGLTexture, width: number, height: number, data: TypedArray) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadDenseMatrixToTexture(\n        this.gl, texture, width, height, data, this.textureConfig);\n  }\n\n  public createFloat16PackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16PackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createPackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createPackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public deleteMatrixTexture(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    if (this.outputTexture === texture) {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n      this.outputTexture = null;\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.deleteTexture(texture));\n  }\n\n  public downloadByteEncodedFloatMatrixFromOutputTexture(\n      texture: WebGLTexture, rows: number, columns: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadByteEncodedFloatMatrixFromOutputTexture(\n            this.gl, rows, columns, this.textureConfig));\n  }\n\n  public downloadPackedMatrixFromBuffer(\n      buffer: WebGLBuffer, batch: number, rows: number, columns: number,\n      physicalRows: number, physicalCols: number): Float32Array {\n    return gpgpu_util.downloadPackedMatrixFromBuffer(\n        this.gl, buffer, batch, rows, columns, physicalRows, physicalCols,\n        this.textureConfig);\n  }\n\n  public downloadFloat32MatrixFromBuffer(buffer: WebGLBuffer, size: number):\n      Float32Array {\n    return gpgpu_util.downloadFloat32MatrixFromBuffer(this.gl, buffer, size);\n  }\n\n  public createBufferFromTexture(\n      texture: WebGLTexture, rows: number, columns: number): WebGLBuffer {\n    this.bindTextureToFrameBuffer(texture);\n    const result = gpgpu_util.createBufferFromOutputTexture(\n        this.gl as WebGL2RenderingContext, rows, columns, this.textureConfig);\n    this.unbindTextureToFrameBuffer();\n    return result;\n  }\n\n  public createAndWaitForFence(): Promise<void> {\n    const fenceContext = this.createFence(this.gl);\n    return this.pollFence(fenceContext);\n  }\n\n  private createFence(gl: WebGLRenderingContext): FenceContext {\n    let query: WebGLQuery|WebGLSync;\n    let isFencePassed: () => boolean;\n\n    if (env().getBool('WEBGL_FENCE_API_ENABLED')) {\n      const gl2 = gl as WebGL2RenderingContext;\n\n      const sync = gl2.fenceSync(gl2.SYNC_GPU_COMMANDS_COMPLETE, 0);\n      gl.flush();\n\n      isFencePassed = () => {\n        const status = gl2.clientWaitSync(sync, 0, 0);\n        return status === gl2.ALREADY_SIGNALED ||\n            status === gl2.CONDITION_SATISFIED;\n      };\n\n      query = sync;\n    } else if (\n        env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0) {\n      query = this.beginQuery();\n      this.endQuery();\n      isFencePassed = () => this.isQueryAvailable(\n          query,\n          env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n    } else {\n      // If we have no way to fence, return true immediately. This will fire in\n      // WebGL 1.0 when there is no disjoint query timer. In this case, because\n      // the fence passes immediately, we'll immediately ask for a download of\n      // the texture, which will cause the UI thread to hang.\n      isFencePassed = () => true;\n    }\n\n    return {query, isFencePassed};\n  }\n\n  public downloadMatrixFromPackedTexture(\n      texture: WebGLTexture, physicalRows: number,\n      physicalCols: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadMatrixFromPackedOutputTexture(\n            this.gl, physicalRows, physicalCols));\n  }\n\n  private vertexAttrsAreBound = false;\n\n  public createProgram(fragmentShaderSource: string): WebGLProgram {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    const fragmentShader: WebGLShader =\n        webgl_util.createFragmentShader(gl, fragmentShaderSource);\n    const vertexShader: WebGLShader = gpgpu_util.createVertexShader(gl);\n    const program: WebGLProgram = webgl_util.createProgram(gl);\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, vertexShader));\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, fragmentShader));\n    webgl_util.linkProgram(gl, program);\n    if (this.debug) {\n      webgl_util.validateProgram(gl, program);\n    }\n    if (!this.vertexAttrsAreBound) {\n      this.setProgram(program);\n      this.vertexAttrsAreBound = gpgpu_util.bindVertexProgramAttributeStreams(\n          gl, this.program, this.vertexBuffer);\n    }\n    return program;\n  }\n\n  public deleteProgram(program: WebGLProgram) {\n    this.throwIfDisposed();\n    if (program === this.program) {\n      this.program = null;\n    }\n    if (program != null) {\n      webgl_util.callAndCheck(this.gl, () => this.gl.deleteProgram(program));\n    }\n  }\n\n  public setProgram(program: WebGLProgram|null) {\n    this.throwIfDisposed();\n    this.program = program;\n    if ((this.program != null) && this.debug) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.useProgram(program));\n  }\n\n  public getUniformLocation(\n      program: WebGLProgram, uniformName: string,\n      shouldThrow = true): WebGLUniformLocation {\n    this.throwIfDisposed();\n    if (shouldThrow) {\n      return webgl_util.getProgramUniformLocationOrThrow(\n          this.gl, program, uniformName);\n    } else {\n      return webgl_util.getProgramUniformLocation(\n          this.gl, program, uniformName);\n    }\n  }\n\n  public getAttributeLocation(program: WebGLProgram, attribute: string):\n      number {\n    this.throwIfDisposed();\n    return webgl_util.callAndCheck(\n        this.gl, () => this.gl.getAttribLocation(program, attribute));\n  }\n\n  public getUniformLocationNoThrow(program: WebGLProgram, uniformName: string):\n      WebGLUniformLocation {\n    this.throwIfDisposed();\n    return this.gl.getUniformLocation(program, uniformName);\n  }\n\n  public setInputMatrixTexture(\n      inputMatrixTexture: WebGLTexture, uniformLocation: WebGLUniformLocation,\n      textureUnit: number) {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    webgl_util.bindTextureToProgramUniformSampler(\n        this.gl, inputMatrixTexture, uniformLocation, textureUnit);\n  }\n\n  public setOutputMatrixTexture(\n      outputMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.setOutputMatrixTextureDriver(outputMatrixTexture, columns, rows);\n  }\n\n  public setOutputPackedMatrixTexture(\n      outputPackedMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.throwIfDisposed();\n    const [width, height] =\n        tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n    this.setOutputMatrixTextureDriver(outputPackedMatrixTexture, width, height);\n  }\n\n  public setOutputMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    this.setOutputMatrixWriteRegionDriver(\n        startColumn, startRow, numColumns, numRows);\n  }\n\n  public setOutputPackedMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    throw new Error('setOutputPackedMatrixWriteRegion not implemented.');\n  }\n\n  public debugValidate() {\n    if (this.program != null) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.validateFramebuffer(this.gl);\n  }\n\n  public executeProgram() {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    const gl = this.gl;\n    if (this.debug) {\n      this.debugValidate();\n    }\n    webgl_util.callAndCheck(\n        gl, () => gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_SHORT, 0));\n  }\n\n  public blockUntilAllProgramsCompleted() {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(this.gl, () => this.gl.finish());\n  }\n\n  private getQueryTimerExtension(): WebGL1DisjointQueryTimerExtension\n      |WebGL2DisjointQueryTimerExtension {\n    if (this.disjointQueryTimerExtension == null) {\n      this.disjointQueryTimerExtension =\n          webgl_util.getExtensionOrThrow(\n              this.gl,\n              env().getNumber(\n                  'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2 ?\n                  'EXT_disjoint_timer_query_webgl2' :\n                  'EXT_disjoint_timer_query') as\n              WebGL1DisjointQueryTimerExtension |\n          WebGL2DisjointQueryTimerExtension;\n    }\n    return this.disjointQueryTimerExtension;\n  }\n\n  private getQueryTimerExtensionWebGL2(): WebGL2DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension();\n  }\n\n  private getQueryTimerExtensionWebGL1(): WebGL1DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension() as WebGL1DisjointQueryTimerExtension;\n  }\n\n  beginQuery(): WebGLQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const query = gl2.createQuery();\n      gl2.beginQuery(ext.TIME_ELAPSED_EXT, query);\n      return query;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    const query = ext.createQueryEXT() as WebGLQuery;\n    ext.beginQueryEXT(ext.TIME_ELAPSED_EXT, query);\n    return query;\n  }\n\n  endQuery() {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n      gl2.endQuery(ext.TIME_ELAPSED_EXT);\n      return;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    ext.endQueryEXT(ext.TIME_ELAPSED_EXT);\n  }\n\n  public async waitForQueryAndGetTime(query: WebGLQuery): Promise<number> {\n    await util.repeatedTry(\n        () => this.disposed ||  // while testing contexts are created / disposed\n                                // in rapid succession, so without this check we\n                                // may poll for the query timer indefinitely\n            this.isQueryAvailable(\n                query,\n                env().getNumber(\n                    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION')));\n    return this.getQueryTime(\n        query, env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n  }\n\n  private getQueryTime(query: WebGLQuery, queryTimerVersion: number): number {\n    if (queryTimerVersion === 0) {\n      return null;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n\n      const timeElapsedNanos = gl2.getQueryParameter(query, gl2.QUERY_RESULT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const timeElapsedNanos =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_EXT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    }\n  }\n\n  private isQueryAvailable(query: WebGLQuery, queryTimerVersion: number):\n      boolean {\n    if (queryTimerVersion === 0) {\n      return true;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const available =\n          gl2.getQueryParameter(query, gl2.QUERY_RESULT_AVAILABLE);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const available =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_AVAILABLE_EXT);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    }\n  }\n\n  pollFence(fenceContext: FenceContext) {\n    return new Promise<void>(resolve => {\n      this.addItemToPoll(() => fenceContext.isFencePassed(), () => resolve());\n    });\n  }\n\n  private itemsToPoll: PollItem[] = [];\n\n  pollItems(): void {\n    // Find the last query that has finished.\n    const index = linearSearchLastTrue(this.itemsToPoll.map(x => x.isDoneFn));\n    for (let i = 0; i <= index; ++i) {\n      const {resolveFn} = this.itemsToPoll[i];\n      resolveFn();\n    }\n    this.itemsToPoll = this.itemsToPoll.slice(index + 1);\n  }\n\n  private addItemToPoll(isDoneFn: () => boolean, resolveFn: () => void) {\n    this.itemsToPoll.push({isDoneFn, resolveFn});\n    if (this.itemsToPoll.length > 1) {\n      // We already have a running loop that polls.\n      return;\n    }\n    // Start a new loop that polls.\n    util.repeatedTry(() => {\n      this.pollItems();\n      // End the loop if no more items to poll.\n      return this.itemsToPoll.length === 0;\n    });\n  }\n\n  private bindTextureToFrameBuffer(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    webgl_util.bindColorTextureToFramebuffer(\n        this.gl, texture, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(this.gl);\n    }\n  }\n\n  private unbindTextureToFrameBuffer() {\n    if (this.outputTexture != null) {\n      webgl_util.bindColorTextureToFramebuffer(\n          this.gl, this.outputTexture, this.framebuffer);\n      if (this.debug) {\n        webgl_util.validateFramebuffer(this.gl);\n      }\n    } else {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n    }\n  }\n\n  private downloadMatrixDriver(\n      texture: WebGLTexture,\n      downloadAndDecode: () => Float32Array): Float32Array {\n    this.bindTextureToFrameBuffer(texture);\n    const result = downloadAndDecode();\n    this.unbindTextureToFrameBuffer();\n\n    return result;\n  }\n\n  private setOutputMatrixTextureDriver(\n      outputMatrixTextureMaybePacked: WebGLTexture, width: number,\n      height: number) {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    webgl_util.bindColorTextureToFramebuffer(\n        gl, outputMatrixTextureMaybePacked, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(gl);\n    }\n    this.outputTexture = outputMatrixTextureMaybePacked;\n    webgl_util.callAndCheck(gl, () => gl.viewport(0, 0, width, height));\n    webgl_util.callAndCheck(gl, () => gl.scissor(0, 0, width, height));\n  }\n\n  private setOutputMatrixWriteRegionDriver(\n      x: number, y: number, width: number, height: number) {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(\n        this.gl, () => this.gl.scissor(x, y, width, height));\n  }\n\n  private throwIfDisposed() {\n    if (this.disposed) {\n      throw new Error('Attempted to use disposed GPGPUContext.');\n    }\n  }\n\n  private throwIfNoProgram() {\n    if (this.program == null) {\n      throw new Error('No GPU program is currently set.');\n    }\n  }\n}\n\ntype PollItem = {\n  isDoneFn: () => boolean,\n  resolveFn: () => void\n};\n\n/**\n * Finds the index of the last true element using linear search.\n * Note: We can't do binary search because Chrome expects us to explicitly\n * test all fences before download:\n * https://github.com/tensorflow/tfjs/issues/1145\n */\nexport function linearSearchLastTrue(arr: Array<() => boolean>): number {\n  let i = 0;\n  for (; i < arr.length; ++i) {\n    const isDone = arr[i]();\n    if (!isDone) {\n      break;\n    }\n  }\n  return i - 1;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, Tensor, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport * as shader_compiler from './shader_compiler';\nimport {InputInfo, ShapeInfo} from './shader_compiler';\nimport {PackingScheme, TextureData, TextureUsage} from './tex_util';\n\nexport interface GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  /** If true, this program expects packed input textures. Defaults to false. */\n  packedInputs?: boolean;\n  /** If true, this program produces a packed texture. Defaults to false. */\n  packedOutput?: boolean;\n  /**\n   * Affects what type of texture we allocate for the output. Defaults to\n   * `TextureUsage.RENDER`.\n   */\n  outTexUsage?: TextureUsage;\n  /**\n   * The type of scheme to use when packing texels for the output values.\n   * See `PackingScheme` for details. Defaults to `PackingScheme.SHARED_BATCH`.\n   */\n  outPackingScheme?: PackingScheme;\n}\n\nexport interface GPGPUBinary {\n  webGLProgram: WebGLProgram;\n  program: GPGPUProgram;\n  uniformLocations: {[name: string]: WebGLUniformLocation};\n  source: string;\n  inShapeInfos: ShapeInfo[];\n  outShapeInfo: ShapeInfo;\n  infLoc: WebGLUniformLocation;\n  nanLoc: WebGLUniformLocation;\n}\n\nexport interface TensorData {\n  shape: number[];\n  texData: TextureData;\n  isUniform: boolean;\n  // Available when we decide to upload as uniform instead of texture.\n  uniformValues?: TypedArray;\n}\n\nexport function compileProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, program: GPGPUProgram, inputs: TensorData[],\n    output: TensorData): GPGPUBinary {\n  const userCode = program.userCode;\n  const inputInfos: InputInfo[] = inputs.map((input, i) => {\n    const shapeInfo: ShapeInfo = {\n      logicalShape: input.shape,\n      texShape: input.isUniform ? null : input.texData.texShape,\n      isUniform: input.isUniform,\n      isPacked: input.isUniform ? false : input.texData.isPacked,\n      flatOffset: null\n    };\n    if (input.texData != null && input.texData.slice != null &&\n        input.texData.slice.flatOffset > 0) {\n      shapeInfo.flatOffset = input.texData.slice.flatOffset;\n    }\n    return {name: program.variableNames[i], shapeInfo};\n  });\n  const inShapeInfos = inputInfos.map(x => x.shapeInfo);\n  const outShapeInfo: ShapeInfo = {\n    logicalShape: output.shape,\n    texShape: output.texData.texShape,\n    isUniform: false,\n    isPacked: output.texData.isPacked,\n    flatOffset: null\n  };\n  const source = shader_compiler.makeShader(\n      inputInfos, outShapeInfo, userCode, program.packedInputs);\n\n  const webGLProgram = gpgpu.createProgram(source);\n\n  // Add special uniforms (NAN, INFINITY)\n  let infLoc: WebGLUniformLocation = null;\n  const nanLoc = gpgpu.getUniformLocation(webGLProgram, 'NAN', false);\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    infLoc = gpgpu.getUniformLocation(webGLProgram, 'INFINITY', false);\n  }\n\n  // Add user-defined uniforms\n  const uniformLocations: {[name: string]: WebGLUniformLocation} = {};\n  for (let i = 0; i < program.variableNames.length; i++) {\n    const varName = program.variableNames[i];\n    const shouldThrow = false;\n    uniformLocations[varName] =\n        gpgpu.getUniformLocation(webGLProgram, varName, shouldThrow);\n    uniformLocations[`offset${varName}`] =\n        gpgpu.getUniformLocation(webGLProgram, `offset${varName}`, shouldThrow);\n  }\n\n  return {\n    program,\n    source,\n    webGLProgram,\n    uniformLocations,\n    inShapeInfos,\n    outShapeInfo,\n    infLoc,\n    nanLoc,\n  };\n}\n\nfunction validateBinaryAndProgram(\n    shapeInfos: ShapeInfo[], inputs: TensorData[]) {\n  if (shapeInfos.length !== inputs.length) {\n    throw Error(\n        `Binary was compiled with ${shapeInfos.length} inputs, but ` +\n        `was executed with ${inputs.length} inputs`);\n  }\n\n  shapeInfos.forEach((s, i) => {\n    const shapeA = s.logicalShape;\n    const input = inputs[i];\n    const shapeB = input.shape;\n\n    if (!util.arraysEqual(shapeA, shapeB)) {\n      throw Error(\n          `Binary was compiled with different shapes than ` +\n          `the current args. Shapes ${shapeA} and ${shapeB} must match`);\n    }\n    // The input is uploaded as uniform.\n    if (s.isUniform && input.isUniform) {\n      return;\n    }\n\n    const texShapeA = s.texShape;\n    const texShapeB = input.isUniform ? null : input.texData.texShape;\n    if (!util.arraysEqual(texShapeA, texShapeB)) {\n      throw Error(\n          `Binary was compiled with different texture shapes than the` +\n          ` current args. Shape ${texShapeA} and ${texShapeB} must match`);\n    }\n  });\n}\n\nexport function runProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, binary: GPGPUBinary, inputs: TensorData[],\n    output: TensorData,\n    customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) =>\n        void): void {\n  validateBinaryAndProgram(binary.inShapeInfos, inputs);\n  validateBinaryAndProgram([binary.outShapeInfo], [output]);\n\n  const outTex = output.texData.texture;\n  const outTexShape = output.texData.texShape;\n  if (output.texData.isPacked) {\n    gpgpu.setOutputPackedMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  } else {\n    gpgpu.setOutputMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  }\n  gpgpu.setProgram(binary.webGLProgram);\n\n  // Set special uniforms (NAN, INFINITY)\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    if (binary.infLoc !== null) {\n      gpgpu.gl.uniform1f(binary.infLoc, Infinity);\n    }\n  }\n  if (binary.nanLoc !== null) {\n    gpgpu.gl.uniform1f(binary.nanLoc, NaN);\n  }\n\n  // Set user-defined inputs\n  inputs.forEach((input, i) => {\n    const varName = binary.program.variableNames[i];\n    const varLoc = binary.uniformLocations[varName];\n    const varOffsetLoc = binary.uniformLocations[`offset${varName}`];\n\n    if (varLoc == null) {\n      // The compiler inferred that this variable is not used in this shader.\n      return;\n    }\n\n    if (input.isUniform) {\n      // Upload the values of the tensor as uniform.\n      if (util.sizeFromShape(input.shape) < 2) {\n        gpgpu.gl.uniform1f(varLoc, input.uniformValues[0]);\n      } else {\n        let vals = input.uniformValues;\n        if (!(vals instanceof Float32Array)) {\n          vals = new Float32Array(vals);\n        }\n        gpgpu.gl.uniform1fv(varLoc, vals);\n      }\n      return;\n    }\n\n    // If the input was sliced, upload the flat offset index.\n    if (input.texData.slice != null && varOffsetLoc != null) {\n      gpgpu.gl.uniform1i(varOffsetLoc, input.texData.slice.flatOffset);\n    }\n\n    gpgpu.setInputMatrixTexture(input.texData.texture, varLoc, i);\n  });\n\n  if (customSetup != null) {\n    customSetup(gpgpu, binary.webGLProgram);\n  }\n  gpgpu.executeProgram();\n}\n\nexport function makeShaderKey(\n    program: GPGPUProgram, inputs: TensorData[], output: TensorData): string {\n  let keyInputs = '';\n  inputs.concat(output).forEach(x => {\n    const hasOffset = x.texData != null && x.texData.slice != null &&\n        x.texData.slice.flatOffset > 0;\n    const texShape = x.isUniform ? 'uniform' : x.texData.texShape;\n    keyInputs += `${x.shape}_${texShape}_${hasOffset}`;\n  });\n  const keyUserCode = program.userCode;\n  let key = program.constructor.name;\n  // Fast string concat. See https://jsperf.com/string-concatenation/14.\n  key += '_' + keyInputs + '_' + keyUserCode;\n  return key;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Im2ColPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      outputShape: number[], inputShape: number[],\n      convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = outputShape;\n\n    const {\n      filterWidth,\n      inChannels,\n      strideWidth,\n      strideHeight,\n      padInfo,\n      outWidth,\n      dilationWidth,\n      dilationHeight,\n      dataFormat\n    } = convInfo;\n    const {left, top} = padInfo;\n    const itemsPerBlockRow = inChannels * filterWidth;\n    const glsl = getGlslDifferences();\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const rowDim = isChannelsLast ? 0 : 1;\n    const colDim = isChannelsLast ? 1 : 2;\n\n    let unrolled = ``;\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        unrolled += `\n          blockIndex = rc.y + ${col};\n          pos = rc.x + ${row};\n\n          if(blockIndex < ${outputShape[1]} && pos < ${outputShape[0]}) {\n            offsetY = int(blockIndex / (${outWidth})) * ${strideHeight} - ${\n            top};\n            d0 = offsetY + ${dilationHeight} * (pos / ${itemsPerBlockRow});\n\n            if(d0 < ${inputShape[rowDim]} && d0 >= 0) {\n\n              offsetX = int(mod(float(blockIndex), ${outWidth}.) * ${\n            strideWidth}. - ${left}.);\n              d1 = offsetX + ${dilationWidth} * (int(mod(float(pos), ${\n            itemsPerBlockRow}.) / ${inChannels}.));\n\n              if(d1 < ${inputShape[colDim]} && d1 >= 0) {\n\n                ch = int(mod(float(pos), ${inChannels}.));\n\n                if (${isChannelsLast}) {\n                  innerDims = vec2(d1, ch);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(d0, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                } else {\n                  innerDims = vec2(d0, d1);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(ch, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                }\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      void main() {\n        ivec2 rc = getOutputCoords();\n\n        vec4 result = vec4(0);\n\n        int blockIndex, pos, offsetY, d0, offsetX, d1, ch;\n        vec2 innerDims;\n\n        ${unrolled}\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n        int d = coords[3];\n        float x = getX(b, r, c, d);\n        float sum = 0.0;\n        for (int j = -${rad}; j <= ${rad}; j++) {\n          int idx = d + j;\n          if (idx >= 0 && idx <=  ${maxD}) {\n            float z = getX(b, r, c, idx);\n            sum += z * z;\n          }\n        }\n        float val = x * ${powOperator};\n        setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNGradProgram implements GPGPUProgram {\n  variableNames = ['inputImage', 'outputImage', 'dy'];\n  outputShape: number[] = [];\n  userCode: string;\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n  depth: number;\n\n  constructor(\n      inputShape: number[], depthRadius: number, bias: number, alpha: number,\n      beta: number) {\n    this.outputShape = inputShape;\n    this.depth = inputShape[3];\n    this.depthRadius = depthRadius;\n    this.bias = bias;\n    this.alpha = alpha;\n    this.beta = beta;\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n\n        float result = 0.0;\n        for (int d = 0; d < ${this.depth}; ++d) {\n          int depthBegin = int(max(0.0, float(d - ${depthRadius})));\n          int depthEnd = int(min(float(${this.depth}),\n              float(d + ${depthRadius} + 1)));\n\n          const int MIN_DEPTH_BEGIN = 0;\n          const int MAX_DEPTH_END = ${this.depth};\n\n          float norm = 0.0;\n          for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd) {\n              norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);\n            }\n            else {\n              break;\n            }\n          }\n\n          norm = float(${alpha}) * norm + float(${bias});\n\n          for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd){\n              float dyi = -2.0 * float(${alpha})\n                * float(${beta})\n                * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)\n                / norm;\n              if (k == d) {\n                dyi += pow(norm, -1.0 * ${beta});\n              }\n              if (k == coords[3]) {\n                dyi *= getDy(b, r, c, d);\n                result += dyi;\n              }\n            }\n            else {\n              break;\n            }\n          }\n      }\n      setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords.x;\n        int r = coords.y;\n        int c = coords.z;\n        int d = coords.w;\n\n        bool hasNextCol = d < ${this.outputShape[3]};\n        bool hasNextRow = c < ${this.outputShape[2]};\n\n        vec4 sum = vec4(0.);\n        vec4 xFragAtOutputCoords = getX(b, r, c, d);\n\n        vec4 xAtOutputCoords = vec4(\n          getChannel(xFragAtOutputCoords, vec2(c, d)),\n          hasNextCol ?\n            getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,\n          hasNextRow ?\n            getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0\n        );\n\n        int firstChannel = d - ${rad};\n        vec2 cache = vec2(0.);\n        if(firstChannel >= 0){\n          vec4 firstChannelFrag = getX(b, r, c, firstChannel);\n          cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));\n            if(hasNextRow){\n              cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));\n            }\n        }\n\n        ivec2 depth = ivec2(d, d + 1);\n        for (int j = - ${rad}; j <= ${rad}; j++) {\n          ivec2 idx = depth + j;\n          bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));\n          bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${maxD}));\n\n          bool depthInRange = aboveLowerBound.x && belowUpperBound.x;\n          bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;\n\n          if(depthInRange || depthPlusOneInRange){\n            vec4 z = vec4(0.);\n            vec4 xFragAtCurrentDepth;\n            z.xz = cache.xy;\n            if(depthPlusOneInRange && hasNextCol){\n              xFragAtCurrentDepth = idx.y != d ?\n                getX(b, r, c, idx.y) : xFragAtOutputCoords;\n              z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));\n              if(hasNextRow){\n                z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));\n              }\n            }\n            cache.xy = z.yw;\n            sum += z * z;\n          }\n        }\n        vec4 result = xAtOutputCoords * ${powOperator};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MaxPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex = effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n          wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n            int maxPosValue = ${lastIndex} - int(getMaxPos(b, idyR, idyC, d));\n\n            // Get the current value, check it against the value from the\n            // position matrix.\n            int curPosValue = wR * ${effectiveFilterWidth} + wC;\n            float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n            dotProd += dyValue * mask;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class MaxPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex =\n        effectiveFilterDepth * effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n           wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n              int maxPosValue = ${lastIndex} -\n                  int(getMaxPos(batch, idyD, idyR, idyC, ch));\n\n              // Get the current value, check it against the value from the\n              // position matrix.\n              int curPosValue =\n                  wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                  wR * ${effectiveFilterWidth} + wC;\n              float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n              dotProd += dyValue * mask;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MatMulPackedProgram implements GPGPUProgram {\n  variableNames = ['matrixA', 'matrixB'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      aShape: [number, number, number], outputShape: [number, number, number],\n      transposeA = false, transposeB = false, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = outputShape;\n\n    const sharedDim = transposeA ? aShape[1] : aShape[2];\n    const sharedDimensionPacked = Math.ceil(sharedDim / 2);\n\n    const aSample = transposeA ? 'i * 2, rc.y' : 'rc.y, i * 2';\n    const bSample = transposeB ? 'rc.z, i * 2' : 'i * 2, rc.z';\n    const aSwizzle = transposeA ? ['a.xxyy', 'a.zzww'] : ['a.xxzz', 'a.yyww'];\n    const bSwizzle = transposeB ? ['b.xzxz', 'b.ywyw'] : ['b.xyxy', 'b.zwzw'];\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const float sharedDimension = ${sharedDimensionPacked}.0;\n\n      vec4 dot2x2ARowBCol(ivec3 rc) {\n        vec4 result = vec4(0);\n        for (int i = 0; i < ${sharedDimensionPacked}; i++) {\n          vec4 a = getMatrixA(rc.x, ${aSample});\n          vec4 b = getMatrixB(rc.x, ${bSample});\n\n          // These swizzled products need to be separately added.\n          // See: https://github.com/tensorflow/tfjs/issues/1735\n          result += (${aSwizzle[0]} * ${bSwizzle[0]});\n          result += (${aSwizzle[1]} * ${bSwizzle[1]});\n        }\n        return result;\n      }\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n        vec4 result = dot2x2ARowBCol(rc);\n\n        ${addBiasSnippet}\n\n        ${applyActivationSnippet}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MultinomialProgram implements GPGPUProgram {\n  variableNames = ['probs'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(batchSize: number, numOutcomes: number, numSamples: number) {\n    this.outputShape = [batchSize, numSamples];\n\n    this.userCode = `\n      uniform float seed;\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n\n        float r = random(seed);\n        float cdf = 0.0;\n\n        for (int i = 0; i < ${numOutcomes - 1}; i++) {\n          cdf += getProbs(batch, i);\n\n          if (r < cdf) {\n            setOutput(float(i));\n            return;\n          }\n        }\n\n        // If no other event happened, last event happened.\n        setOutput(float(${numOutcomes - 1}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(seed: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.seedLoc == null) {\n        this.seedLoc = gpgpu.getUniformLocation(webGLProgram, 'seed');\n      }\n      gpgpu.gl.uniform1f(this.seedLoc, seed);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class OneHotProgram implements GPGPUProgram {\n  variableNames = ['indices'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(\n      numIndices: number, depth: number, onValue: number, offValue: number) {\n    this.outputShape = [numIndices, depth];\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int index = round(getIndices(coords.x));\n        setOutput(mix(float(${offValue}), float(${onValue}),\n                      float(index == coords.y)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape:\n          number[]) {  // TODO(https://github.com/tensorflow/tfjs/issues/893):\n                       // Only input / output 3D tensors.\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    if (rank === 0) {\n      this.userCode = `\n        void main() {\n          setOutput(vec4(getA(), 0., 0., 0.));\n        }\n      `;\n    } else {\n      const channels = getChannels('rc', rank);\n      const dtype = getCoordsDataType(rank);\n      const outOfBoundsCondition =\n          getOutOfBoundsCondition(rank, outputShape, channels);\n      const setup = getSetup(\n          rank, outputShape[outputShape.length - 1],\n          outputShape[outputShape.length - 2], channels);\n      const output = getOutput(outputShape, channels);\n\n      this.userCode = `\n        void main() {\n          ${dtype} rc = getOutputCoords();\n\n          if(${outOfBoundsCondition}) {\n            setOutput(vec4(0));\n          } else {\n            ${setup}\n\n            setOutput(vec4(${output}));\n          }\n        }\n      `;\n    }\n  }\n}\n\nfunction getSourceCoordsArr(rank: number, dims: string[]): string[] {\n  const coords = [];\n\n  for (let row = 0; row <= 1; row++) {\n    for (let col = 0; col <= 1; col++) {\n      let coord = `${row === 0 ? 'r' : 'rp1'}, ${col === 0 ? 'c' : 'cp1'}`;\n\n      for (let d = 2; d < rank; d++) {\n        coord = `${dims[dims.length - 1 - d]},` + coord;\n      }\n\n      coords.push(coord);\n    }\n  }\n  return coords;\n}\n\nfunction getOutOfBoundsCondition(\n    rank: number, shape: number[], dims: string[]): string {\n  if (rank === 1) {\n    return `rc > ${shape[0]}`;\n  }\n\n  let cond = '';\n  for (let i = rank - 2; i < rank; i++) {\n    cond += `${dims[i]} >= ${shape[i]}`;\n    if (i < rank - 1) {\n      cond += '||';\n    }\n  }\n\n  return cond;\n}\n\nfunction getSetup(\n    rank: number, cols: number, rows: number, dims: string[]): string {\n  if (rank === 1) {\n    return '';\n  }\n\n  const innerDims = dims.slice(-2);\n\n  return `\n    int r = ${innerDims[0]};\n    int c = ${innerDims[1]};\n    int rp1 = r + 1;\n    int cp1 = c + 1;\n\n    bool cEdge = cp1 >= ${cols};\n    bool rEdge = rp1 >= ${rows};\n  `;\n}\n\nfunction getOutput(shape: number[], dims: string[]): string {\n  const rank = shape.length;\n  const sourceCoords = getSourceCoordsArr(rank, dims);\n  if (rank === 1) {\n    return `getA(rc),\n            rc + 1 >= ${shape[0]} ? 0. : getA(rc + 1),\n            0, 0`;\n  }\n\n  return `getA(${sourceCoords[0]}),\n          cEdge ? 0. : getA(${sourceCoords[1]}),\n          rEdge ? 0. : getA(${sourceCoords[2]}),\n          rEdge || cEdge ? 0. : getA(${sourceCoords[3]})`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const type = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const unpackedCoords =\n        ['coords[0]', 'coords[1]', 'coords[2]', 'coords[3]'].slice(0, rank);\n\n    if (rank === 1) {\n      this.userCode = `\n        int start = ${start};\n        int end = ${end};\n\n        void main() {\n          int outC = getOutputCoords();\n          if (outC < start || outC >= end) {\n            setOutput(float(${constantValue}));\n          } else {\n            setOutput(getX(outC - start));\n          }\n        }\n      `;\n      return;\n    }\n    this.userCode = `\n      ${type} start = ${type}(${start});\n      ${type} end = ${type}(${end});\n\n      void main() {\n        ${type} outC = getOutputCoords();\n        if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {\n          setOutput(float(${constantValue}));\n        } else {\n          ${type} coords = outC - start;\n          setOutput(getX(${unpackedCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const dtype = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const coords = getChannels('rc', rank);\n    const source = getChannels('source', rank);\n    const cLimit = `${coords[rank - 1]} < ${this.outputShape[rank - 1]}`;\n    const innerDims =\n        rank === 1 ? 'source' : `vec2(${source.slice(-2).join()})`;\n\n    const componentSetup = [\n      `${dtype} rc = outputLoc;`, `${coords[rank - 1]} += 1;\n       if(${cLimit}) {\n      `,\n      rank === 1 ? '' : `}\n       rc = outputLoc;\n       ${coords[rank - 2]} += 1;\n       if(${coords[rank - 2]} < ${this.outputShape[rank - 2]}) {`,\n      rank === 1 ? '' : `  ${coords[rank - 1]} += 1;\n         if(${cLimit}) {`\n    ];\n\n    const paddingArea = rank === 1 ?\n        'rc < start || rc >= end' :\n        'any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))';\n    let mainLoop = '';\n    for (let i = 0, j = rank === 1 ? 2 : 4; i < j; i++) {\n      mainLoop += `\n        ${componentSetup[i]}\n        if (${paddingArea}) {\n          result[${i}] = float(${constantValue});\n        } else {\n          ${dtype} source = rc - start;\n          result[${i}] = getChannel(getX(${source.join()}), ${innerDims});\n        }\n      `;\n    }\n    mainLoop += (rank === 1 ? `} ` : `}}`);\n\n    this.userCode = `\n      const ${dtype} start = ${dtype}(${start});\n      const ${dtype} end = ${dtype}(${end});\n\n      void main() {\n        ${dtype} outputLoc = getOutputCoords();\n        vec4 result = vec4(0.);\n        ${mainLoop}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Pool2DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n    const batchFlattenPositionStr = `((batch  * ${convInfo.inHeight} + xR) * ${\n        convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n    const flattenPositionStr =\n        `(xR * ${convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n        const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int batch = coords[0];\n          int d = coords[3];\n\n          ivec2 xRCCorner = coords.yz * strides - pads;\n          int xRCorner = xRCCorner.x;\n          int xCCorner = xRCCorner.y;\n\n          // max/min x(?, ?, d) to get y(yR, yC, d).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n          float avgValue = 0.0;\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              int xC = xCCorner + wC;\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float value = getX(batch, xR, xC, d);\n\n              // If a min / max value has already been found, use it. If not,\n              // use the current value.\n              float currMinMaxValue = mix(\n                  value, minMaxValue, minMaxValueFound);\n              if (value ${compareOp} currMinMaxValue) {\n                minMaxValue = value;\n                minMaxValueFound = 1.0;\n                minMaxPosition = ${\n          flattenPositions ? (includeBatchInIndex ? batchFlattenPositionStr :\n                                                    flattenPositionStr) :\n                             `wR * ${effectiveFilterWidth} + wC`};\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xR, int xC, int d) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xR, xC, d);\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d = coords[3];\n\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // max/min x(?, ?, d) to get y(yR, yC, d).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          int xR = xRCorner + wR;\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              getValue(batch, xR, xC + 3 * ${dilationWidth}, d)\n            );\n\n            ${updateSnippet}\n          }\n\n          int xC = xCCorner + ${filterWidthNearestVec4};\n          if (${filterWidthVec4Remainder === 1}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              initializationValue,\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 2}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 3}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              initializationValue\n            );\n\n            ${updateSnippet}\n          }\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n\nexport class Pool3DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv3DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec3 strides =\n            ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n        const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n        void main() {\n          ivec5 coords = getOutputCoords();\n          int batch = coords.x;\n          int ch = coords.u;\n\n          ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n          int xDCorner = xCorner.x;\n          int xRCorner = xCorner.y;\n          int xCCorner = xCorner.z;\n\n          // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n\n          for (int wD = 0; wD < ${effectiveFilterDepth};\n              wD += ${dilationDepth}) {\n            int xD = xDCorner + wD;\n\n            if (xD < 0 || xD >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int wR = 0; wR < ${effectiveFilterHeight};\n                wR += ${dilationHeight}) {\n              int xR = xRCorner + wR;\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int wC = 0; wC < ${effectiveFilterWidth};\n                  wC += ${dilationWidth}) {\n                int xC = xCCorner + wC;\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float value = getX(batch, xD, xR, xC, ch);\n\n                // If a min / max value has already been found, use it. If not,\n                // use the current value.\n                float currMinMaxValue = mix(\n                    value, minMaxValue, minMaxValueFound);\n                if (value ${compareOp} currMinMaxValue) {\n                  minMaxValue = value;\n                  minMaxValueFound = 1.0;\n                  minMaxPosition = ${\n          flattenPositions ?\n              (includeBatchInIndex ?\n                   `(((batch * ${convInfo.inDepth} + xD) * ${\n                       convInfo.inHeight} + xR) * ${convInfo.inWidth} + xC) * ${\n                       convInfo.inChannels} + ch` :\n                   `((xD * ${convInfo.inHeight} + xR) * ${\n                       convInfo.inWidth} + xC) * ${convInfo.inChannels} + ch`) :\n              `wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                      wR * ${effectiveFilterWidth} + wC`};\n                }\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec3 strides =\n        ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xD, int xR, int xC, int ch) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xD, xR, xC, ch);\n      }\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xDCorner = xCorner.x;\n        int xRCorner = xCorner.y;\n        int xCCorner = xCorner.z;\n\n        // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          int xD = xDCorner + wD;\n\n          if (xD < 0 || xD >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 3 * ${dilationWidth}, ch)\n              );\n\n              ${updateSnippet}\n            }\n\n            int xC = xCCorner + ${filterWidthNearestVec4};\n            if (${filterWidthVec4Remainder === 1}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                initializationValue,\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 2}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 3}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                initializationValue\n              );\n\n              ${updateSnippet}\n            }\n          }\n          setOutput(${returnValue});\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ReduceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo,\n      reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod') {\n    const {windowSize, batchSize, inSize, outSize} = reduceInfo;\n    this.outputShape = [batchSize, outSize];\n\n    let initializationValue = '0.0';\n    let compareOp = ``;\n\n    if (reduceType === 'prod') {\n      initializationValue = '1.0';\n    } else if (reduceType === 'min') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '1.0 / 1e-20';\n      compareOp = `min`;\n    } else if (reduceType === 'max') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n      compareOp = `max`;\n    }\n\n    let returnValue = `${reduceType}(${reduceType}(${reduceType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n\n    if (reduceType === 'sum') {\n      returnValue = `sumValue`;\n    } else if (reduceType === 'prod') {\n      returnValue = `prodValue`;\n    } else if (reduceType === 'all') {\n      returnValue = `allValue`;\n    } else if (reduceType === 'any') {\n      returnValue = `anyValue`;\n    }\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    let updateSnippet = `\n      if (${reduceType === 'sum'}) {\n        sumValue += dot(values, ones);\n      } else if (${reduceType === 'prod'}) {\n        vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);\n        prodValue *= tmp[0] * tmp[1];\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    let vecType = `vec4`;\n\n    if (reduceType === 'all') {\n      initializationValue = '1.0';\n      updateSnippet = `\n        bool reducedAllValue = all(values);\n        float floatedReducedAllValue = float(reducedAllValue);\n        allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    } else if (reduceType === 'any') {\n      initializationValue = '0.0';\n      updateSnippet = `\n        bool reducedAnyValue = any(values);\n        float floatedReducedAnyValue = float(reducedAnyValue);\n        anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    }\n\n    let checkOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float getValue(int batch, int inIdx) {\n        ${checkOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        vec4 minMaxValue = vec4(${initializationValue});\n        float prodValue = 1.0;\n        float sumValue = 0.0;\n        float allValue = 1.0;\n        float anyValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class ReshapePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: [number, number, number], inputShape: [\n    number, number, number\n  ]) {\n    this.outputShape = outputShape;\n\n    let mainLoop = ``;\n    for (let i = 0; i < 4; i++) {\n      let thisRC = `thisRC = rc;`;\n      if (i % 2 === 1) {\n        thisRC += `thisRC.z += 1;`;\n      }\n      if (i > 1) {\n        thisRC += `thisRC.y += 1;`;\n      }\n\n      mainLoop += `\n        ${thisRC}\n        ${i > 0 ? `if(thisRC.y < rows && thisRC.z < cols){` : ''}\n          int flatIndex = getFlatIndex(thisRC);\n\n          ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);\n          vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));\n\n          result[${i}] =\n            getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);\n        ${i > 0 ? '}' : ''}\n      `;\n    }\n\n    this.userCode = `\n      ${getReshapedInputCoords(inputShape)}\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n\n        vec4 result = vec4(0.);\n\n        ivec3 thisRC;\n        int rows = ${outputShape[1]};\n        int cols = ${outputShape[2]};\n\n        ${mainLoop}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nfunction getReshapedInputCoords(shape: [number, number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 inputCoordsFromReshapedOutCoords(int index) {\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(startRLerp - float(winHeight / 2));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(startCLerp - float(winWidth / 2));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float dxR = float(dyR) * heightScale;\n            int topDxRIndex = int(floor(dxR));\n            int bottomDxRIndex = int(min(ceil(dxR), ${xHeight - 1}.0));\n            float dxRLerp = dxR - float(topDxRIndex);\n            float inverseDxRLerp = 1.0 - dxRLerp;\n\n            float dxC = float(dyC) * widthScale;\n            int leftDxCIndex = int(floor(dxC));\n            int rightDxCIndex = int(min(ceil(dxC), ${xWidth - 1}.0));\n            float dxCLerp = dxC - float(leftDxCIndex);\n            float inverseDxCLerp = 1.0 - dxCLerp;\n\n            if (r == topDxRIndex && c == leftDxCIndex) {\n              // topLeft\n              accumulator +=\n                getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;\n            }\n\n            if (r == topDxRIndex && c == rightDxCIndex) {\n              // topRight\n              accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == leftDxCIndex) {\n              // bottomLeft\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == rightDxCIndex) {\n              // bottomRight\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec2 sourceFloorRC = ivec2(sourceFracIndexRC);\n        ivec2 sourceCeilRC = ivec2(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);\n        float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);\n        float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);\n        float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);\n\n        vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);\n\n        float top = topLeft + (topRight - topLeft) * fracRC.y;\n        float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;\n        float newValue = top + (bottom - top) * fracRC.x;\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec3 effectiveInputOverOutputRatioRC = vec3(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec3 inputShapeRC = vec3(${oldHeight}.0, ${oldWidth}.0,\n                                     ${oldWidth}.0);\n\n      float getAValue(int b, int r, int c, int d) {\n        return getChannel(getA(b, r, c, d), vec2(c, d));\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        // Calculate values for next column in yRC.z.\n        ivec3 yRC = coords.yzz + ivec3(0, 0, 1);\n\n        // Fractional source index.\n        vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec3 sourceFloorRC = ivec3(sourceFracIndexRC);\n        ivec3 sourceCeilRC = ivec3(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        // Should we calculate next column and row elements in 2x2 packed cell.\n        bool hasNextCol = d < ${depth - 1};\n        bool hasNextRow = coords.z < ${newWidth - 1};\n\n        // In parallel, construct four corners for all four components in\n        // packed 2x2 cell.\n        vec4 topLeft = vec4(\n          getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 bottomLeft = vec4(\n          getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 topRight = vec4(\n          getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec4 bottomRight = vec4(\n          getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);\n\n        vec4 top = mix(topLeft, topRight, fracRC.yyzz);\n        vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);\n        vec4 newValue = mix(top, bottom, fracRC.x);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeigborBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(floor(startRLerp - float(winHeight / 2)));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(floor(startCLerp - float(winWidth / 2)));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float sourceFracRow =\n              float(${effectiveXSize[0]}) *\n                (float(dyR) / float(${effectiveYSize[0]}));\n\n            float sourceFracCol =\n                float(${effectiveXSize[1]}) *\n                  (float(dyC) / float(${effectiveYSize[1]}));\n\n            int sourceNearestRow = int(min(\n                float(int(${xHeight}) - 1),\n                ${alignCorners} ? float(round(sourceFracRow)) :\n                                  float(floor(sourceFracRow))));\n\n            int sourceNearestCol = int(min(\n                float(int(${xWidth}) - 1),\n                ${alignCorners} ? float(round(sourceFracCol)) :\n                                  float(floor(sourceFracCol))));\n\n            if (r == sourceNearestRow && c == sourceNearestCol) {\n              accumulator += getDy(b, dyR, dyC, d);\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeighborProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    // When align corners is false, we rounds the value with floor.\n    const roundBase = alignCorners ? '0.5' : '0.0';\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the coordinators of nearest neighbor point.\n        ivec2 sourceNearestRC = ivec2(\n          min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${roundBase})));\n\n        float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReverseProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n\n    if (rank === 1) {\n      this.userCode = `\n        void main() {\n          int coord = getOutputCoords();\n          setOutput(getX(${xShape[0]} - coord - 1));\n        }\n      `;\n      return;\n    }\n    const getInCoord = (i: number) => {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - coords[${i}] - 1`;\n      }\n      return `coords[${i}]`;\n    };\n    const inCoords = xShape.map((_, i) => getInCoord(i)).join(',');\n    const type = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${type} coords = getOutputCoords();\n        setOutput(getX(${inCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReversePackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n    const channels = getChannels('rc', rank);\n    const nextColumn =\n        `${channels[rank - 1]} + 1 < ${this.outputShape[rank - 1]}`;\n    const nextRow = `${channels[rank - 2]} + 1 < ${this.outputShape[rank - 2]}`;\n    const type = getCoordsDataType(rank);\n    if (rank === 1) {\n      this.userCode = `\n        void main(){\n          int rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = getChannel(getX(${xShape[0]} - rc - 1),\n            ${xShape[0]} - rc - 1);\n          if(${nextColumn}){\n              result.g = getChannel(getX(${xShape[0]} - (rc  + 1) - 1),\n                ${xShape[0]} - (rc  + 1) - 1);\n          }\n          setOutput(result);\n        }\n      `;\n    } else {\n      this.userCode = `\n        void main() {\n          ${type} rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = ${getR(channels.slice())};\n          if(${nextColumn}){\n            result.g = ${getG(channels.slice())};\n          }\n          if(${nextRow}) {\n            result.b = ${getB(channels.slice())};\n            if(${nextColumn}) {\n              result.a = ${getA(channels.slice())};\n            }\n          }\n          setOutput(result);\n        }\n    `;\n    }\n\n    function getR(channels: string[]): string {\n      return getChannel(channels);\n    }\n\n    function getG(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getB(channels: string[]): string {\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getA(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getChannel(channels: string[]): string {\n      const inCoordsArray = xShape.map((_, i) => getInCoord(i, channels));\n      const inCoords = inCoordsArray.join(',');\n      const innerDims = inCoordsArray.slice(-2).join(',');\n      return `getChannel(getX(${inCoords}), vec2(${innerDims}))`;\n    }\n\n    function getInCoord(i: number, channels1: string[]): string {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - ${channels1[i]} - 1`;\n      } else {\n        return `${channels1[i]}`;\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ScatterProgram implements GPGPUProgram {\n  variableNames = ['updates', 'indices', 'defaultValue'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      updateSize: number, sliceDim: number, indicesRank: number,\n      updatesRank: number, strides: number[], shape: number[],\n      summingDupeIndex = true) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    let indicesString = '';\n    if (indicesRank === 1) {\n      indicesString = 'i';\n    } else if (indicesRank === 2) {\n      indicesString = 'i, j';\n    }\n    const indicesSnippet = `getIndices(${indicesString})`;\n\n    let updatesString = '';\n    if (updatesRank === 1) {\n      updatesString = 'i';\n    } else if (updatesRank === 2) {\n      updatesString = 'i, coords[1]';\n    }\n    const updatesSnippet = `getUpdates(${updatesString})`;\n\n    const strideString = sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${strides});\n\n        void main() {\n          ${dtype} coords = getOutputCoords();\n          float sum = 0.0;\n          bool found = false;\n          for (int i = 0; i < ${updateSize}; i++) {\n            int flattenedIndex = 0;\n            for (int j = 0; j < ${sliceDim}; j++) {\n              int index = round(${indicesSnippet});\n              flattenedIndex += index * ${strideString};\n            }\n            if (flattenedIndex == coords[0]) {\n              sum += ${updatesSnippet};\n              found = true;\n            }\n          }\n          setOutput(mix(getDefaultValue(), sum, float(found)));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class SegmentOpProgram implements GPGPUProgram {\n  variableNames = ['x', 'segmentIds'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      segOpInfo: backend_util.segment_util.SegOpInfo,\n      segOpType: 'unsortedSegmentSum') {\n    const windowSize = segOpInfo.windowSize;\n    const batchSize = segOpInfo.batchSize;\n    const inSize = segOpInfo.inSize;\n    const numSegments = segOpInfo.numSegments;\n    const outSize = numSegments * Math.ceil(inSize / windowSize);\n    this.outputShape = [batchSize, outSize];\n\n    const initializationValue = '0.0';\n    const returnValue = `sumValue`;\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    const updateSnippet = `\n        sumValue += dot(values, segFilter);\n    `;\n\n    let checkValueOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkValueOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n\n    let checkSegmentIdOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkSegmentIdOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return -1.0;\n        }\n      `;\n    }\n\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n\n      float getValue(int batch, int inIdx) {\n        ${checkValueOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      float getSegmentIdAtIndex(int inIdx) {\n        ${checkSegmentIdOutOfBounds}\n        return getSegmentIds(inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = int(floor(float(outIdx) / float(\n          ${numSegments})) * float(${windowSize}));\n        int currentSeg = int(mod(float(outIdx), float(${numSegments})));\n\n        float sumValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          int inIdxSeg = int(getSegmentIdAtIndex(inIdx));\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            0,\n            0,\n            0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n              0,\n              0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            0\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SelectProgram implements GPGPUProgram {\n  variableNames = ['c', 'a', 'b'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(cRank: number, shape: number[], rank: number) {\n    this.outputShape = shape;\n\n    let cCoords;\n    let abCoords;\n    if (rank > 4) {\n      throw Error(`Where for rank ${rank} is not yet supported`);\n    }\n\n    if (rank === 1) {\n      abCoords = `resRC`;\n      cCoords = `resRC`;\n    } else {\n      const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n      const cCoordVars = [];\n      const abCoordVars = [];\n      for (let i = 0; i < shape.length; i++) {\n        abCoordVars.push(`${currentCoords[i]}`);\n        if (i < cRank) {\n          cCoordVars.push(`${currentCoords[i]}`);\n        }\n      }\n      cCoords = cCoordVars.join();\n      abCoords = abCoordVars.join();\n    }\n\n    const dtype = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        float cVal = getC(${cCoords});\n        if (cVal >= 1.0) {\n          setOutput(getA(${abCoords}));\n        } else {\n          setOutput(getB(${abCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SliceProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const uniformPart = `uniform int start[${this.rank}];`;\n    const sourceCoords = getCoords(this.rank);\n\n    let body: string;\n    const coordSum = destSize.map((_, i) => {\n      return `sourceLoc.${coords[i]} = start[${i}] + coords.${coords[i]};`;\n    });\n    body = `\n        ${dtype} sourceLoc;\n        ${dtype} coords = getOutputCoords();\n        ${coordSum.join('\\n')}\n      `;\n    this.userCode = `\n      ${uniformPart}\n      void main() {\n        ${body}\n        setOutput(getSource(${sourceCoords}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n\nconst coords = ['x', 'y', 'z', 'w', 'u', 'v'];\n\nfunction getCoords(rank: number): string {\n  if (rank === 1) {\n    return 'sourceLoc';\n  } else if (rank <= 6) {\n    return coords.slice(0, rank).map(x => 'sourceLoc.' + x).join(',');\n  } else {\n    throw Error(`Slicing for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SlicePackedProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const coords = getChannels('coords', this.rank);\n    const sourceLoc = getChannels('sourceLoc', this.rank);\n\n    const innerDims =\n        this.rank === 1 ? 'sourceLoc' : `vec2(${sourceLoc.slice(-2).join()})`;\n    const getChannel =\n        `getChannel(getSource(${sourceLoc.join()}), ${innerDims})`;\n    const upperRow = `\n      result.x = ${getChannel};\n      if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n        ++${sourceLoc[this.rank - 1]};\n        result.y = ${getChannel};\n        --${sourceLoc[this.rank - 1]};\n      }\n    `;\n    const lowerRow = this.rank === 1 ? '' : `\n      --${coords[this.rank - 1]};\n      if (++${coords[this.rank - 2]} < ${destSize[this.rank - 2]}) {\n        ++${sourceLoc[this.rank - 2]};\n        result.z = ${getChannel};\n        if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n          ++${sourceLoc[this.rank - 1]};\n          result.w = ${getChannel};\n        }\n      }\n    `;\n\n    const sourceLocSetup = this.rank <= 4 ?\n        `sourceLoc = coords +\n            ${dtype}(${destSize.map((_, i) => `start[${i}]`).join()});` :\n        destSize.map((_, i) => `${sourceLoc[i]} = ${coords[i]} + start[${i}];`)\n            .join('\\n');\n    this.userCode = `\n      uniform int start[${this.rank}];\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        ${dtype} sourceLoc;\n        ${sourceLocSetup}\n        vec4 result = vec4(0.);\n        ${upperRow}\n        ${lowerRow}\n        setOutput(result);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class StridedSliceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(begin: number[], strides: number[], size: number[]) {\n    this.outputShape = size;\n    const rank = size.length;\n    const inputDtype = getCoordsDataType(size.length);\n    const dtype = getCoordsDataType(size.length);\n\n    let newCoords = '';\n    if (rank === 1) {\n      newCoords = 'coords * strides + begin';\n    } else {\n      let outputAxis = 0;\n      newCoords =\n          size.map((_, i) => {\n                outputAxis++;\n                return size.length === 1 ?\n                    `coords * strides[${i}] + begin[${i}]` :\n                    `coords[${outputAxis - 1}] * strides[${i}] + begin[${i}]`;\n              })\n              .join(',');\n    }\n\n    this.userCode = `\n      ${inputDtype} begin = ${inputDtype}(${begin});\n      ${inputDtype} strides = ${inputDtype}(${strides});\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        setOutput(getX(${newCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {getInternalFormatForFloat16MatrixTexture, getInternalFormatForFloat16PackedMatrixTexture, getInternalFormatForFloat32MatrixTexture, getInternalFormatForPackedMatrixTexture, getInternalFormatForUnsignedBytesMatrixTexture} from './gpgpu_util';\nimport {getPackedMatrixTextureShapeWidthHeight, getUnpackedMatrixTextureShapeWidthHeight, PhysicalTextureType, TextureConfig, TextureUsage} from './tex_util';\n\nexport class TextureManager {\n  private numUsedTextures = 0;\n  private numFreeTextures = 0;\n  private _numBytesAllocated = 0;\n  private _numBytesFree = 0;  // How many bytes that have been allocated\n                              // are available for reuse.\n  private freeTextures: {[shape: string]: WebGLTexture[]} = {};\n  private logEnabled = false;\n  private usedTextures: {[shape: string]: WebGLTexture[]} = {};\n\n  constructor(private gpgpu: GPGPUContext) {}\n\n  acquireTexture(\n      shapeRC: [number, number], usage: TextureUsage,\n      isPacked: boolean): WebGLTexture {\n    const physicalTexType = getPhysicalFromLogicalTextureType(usage, isPacked);\n\n    const shapeKey = getKeyFromTextureShape(shapeRC, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n    if (!(shapeKey in this.usedTextures)) {\n      this.usedTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shapeRC, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n\n    if (this.freeTextures[shapeKey].length > 0) {\n      this.numFreeTextures--;\n      this.numUsedTextures++;\n      this._numBytesFree -= texBytes;\n      this.log();\n      const newTexture = this.freeTextures[shapeKey].shift();\n      this.usedTextures[shapeKey].push(newTexture);\n      return newTexture;\n    }\n\n    let newTexture: WebGLTexture;\n    if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT32) {\n      newTexture = this.gpgpu.createPackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16PackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT32) {\n      newTexture =\n          this.gpgpu.createFloat32MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (\n        physicalTexType === PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE) {\n      newTexture =\n          this.gpgpu.createUnsignedBytesMatrixTexture(shapeRC[0], shapeRC[1]);\n    }\n    this.usedTextures[shapeKey].push(newTexture);\n\n    this.numUsedTextures++;\n    this._numBytesAllocated += texBytes;\n    this.log();\n\n    return newTexture;\n  }\n\n  releaseTexture(\n      texture: WebGLTexture, shape: [number, number],\n      logicalTexType: TextureUsage, isPacked: boolean): void {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    const physicalTexType =\n        getPhysicalFromLogicalTextureType(logicalTexType, isPacked);\n    const shapeKey = getKeyFromTextureShape(shape, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shape, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n    const deleteTexThreshold = env().get('WEBGL_DELETE_TEXTURE_THRESHOLD');\n    if (deleteTexThreshold !== -1 &&\n        this._numBytesAllocated > deleteTexThreshold) {\n      this.gpgpu.deleteMatrixTexture(texture);\n      this._numBytesAllocated -= texBytes;\n    } else {\n      this.freeTextures[shapeKey].push(texture);\n      this.numFreeTextures++;\n      this._numBytesFree += texBytes;\n    }\n\n    this.numUsedTextures--;\n\n    const texList = this.usedTextures[shapeKey];\n    const texIndex = texList.indexOf(texture);\n    if (texIndex < 0) {\n      throw new Error(\n          'Cannot release a texture that was never provided by this ' +\n          'texture manager');\n    }\n    texList.splice(texIndex, 1);\n    this.log();\n  }\n\n  private log() {\n    if (!this.logEnabled) {\n      return;\n    }\n    const total = this.numFreeTextures + this.numUsedTextures;\n    console.log(\n        'Free/Used', `${this.numFreeTextures} / ${this.numUsedTextures}`,\n        `(${total})`);\n    const freeRatio = this._numBytesFree / this._numBytesAllocated;\n    console.log(`Bytes allocated: ${this._numBytesAllocated}`);\n    console.log(`Bytes unused: ${this._numBytesFree} (${\n        Math.round(100 * freeRatio)}%)`);\n  }\n\n  get numBytesAllocated(): number {\n    return this._numBytesAllocated;\n  }\n\n  get numBytesFree(): number {\n    return this._numBytesFree;\n  }\n\n  getNumUsedTextures(): number {\n    return this.numUsedTextures;\n  }\n\n  getNumFreeTextures(): number {\n    return this.numFreeTextures;\n  }\n\n  dispose() {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    for (const texShape in this.freeTextures) {\n      this.freeTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    for (const texShape in this.usedTextures) {\n      this.usedTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    this.freeTextures = null;\n    this.usedTextures = null;\n    this.numUsedTextures = 0;\n    this.numFreeTextures = 0;\n    this._numBytesAllocated = 0;\n    this._numBytesFree = 0;\n  }\n}\n\nfunction numBytesForInternalFormat(\n    gl: WebGLRenderingContext, internalFormat: number): number {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n  if (internalFormat === glany.R32F) {\n    return 4;\n  } else if (internalFormat === glany.R16F) {\n    return 2;\n  } else if (internalFormat === glany.RGBA32F) {\n    return 16;\n  } else if (internalFormat === gl.RGBA) {\n    return 16;\n  } else if (internalFormat === glany.RGBA16F) {\n    return 8;\n  }\n  throw new Error(`Unknown internal format ${internalFormat}`);\n}\n\nexport function computeBytes(\n    shape: [number, number], physicalTexType: PhysicalTextureType,\n    gl: WebGLRenderingContext, textureConfig: TextureConfig,\n    isPacked: boolean): number {\n  // It is not possible to infer packed status from the texture type because\n  // depending on the textureConfig, different  texture types may resolve to the\n  // same internal format (e.g. in WebGL1, the internal format for\n  // UNPACKED_FLOAT16 textures is gl.RGBA). Therefore we pass in `isPacked`\n  // explicitly.\n  const internalFormat =\n      internalFormatForPhysicalTexType(physicalTexType, textureConfig);\n\n  let numElements: number;\n  if (isPacked) {\n    const [packedWidth, packedHeight] =\n        getPackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = packedWidth * packedHeight;\n\n  } else {\n    const [width, height] =\n        getUnpackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = width * height;\n  }\n\n  const bytesPerElement = numBytesForInternalFormat(gl, internalFormat);\n  return numElements * bytesPerElement;\n}\n\nfunction internalFormatForPhysicalTexType(\n    physicalTexType: PhysicalTextureType,\n    textureConfig: TextureConfig): number {\n  switch (physicalTexType) {\n    case PhysicalTextureType.PACKED_2X2_FLOAT32:\n      return getInternalFormatForPackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_2X2_FLOAT16:\n      return getInternalFormatForFloat16PackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT32:\n      return getInternalFormatForFloat32MatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT16:\n      return getInternalFormatForFloat16MatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE:\n      return getInternalFormatForUnsignedBytesMatrixTexture(textureConfig);\n    default:\n      throw new Error(`Unknown physical texture type ${physicalTexType}`);\n  }\n}\n\nfunction getPhysicalTextureForRendering(isPacked: boolean):\n    PhysicalTextureType {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED')) {\n    if (isPacked) {\n      return PhysicalTextureType.PACKED_2X2_FLOAT32;\n    }\n    return PhysicalTextureType.UNPACKED_FLOAT32;\n  }\n\n  if (isPacked) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT16;\n  }\n  return PhysicalTextureType.UNPACKED_FLOAT16;\n}\n\nfunction getPhysicalFromLogicalTextureType(\n    logicalTexType: TextureUsage, isPacked: boolean): PhysicalTextureType {\n  if (logicalTexType === TextureUsage.UPLOAD) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT32;\n  } else if (logicalTexType === TextureUsage.RENDER || logicalTexType == null) {\n    return getPhysicalTextureForRendering(isPacked);\n  } else if (\n      logicalTexType === TextureUsage.DOWNLOAD ||\n      logicalTexType === TextureUsage.PIXELS) {\n    return PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE;\n  }\n  throw new Error(`Unknown logical texture type ${logicalTexType}`);\n}\n\nfunction getKeyFromTextureShape(\n    shapeRowsCol: [number, number], physicalTexType: PhysicalTextureType,\n    isPacked: boolean): string {\n  return `${shapeRowsCol[0]}_${shapeRowsCol[1]}_${physicalTexType}_${isPacked}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TileProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], reps: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[i] * reps[i];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[]): string {\n  const rank = aShape.length;\n  if (rank > 5) {\n    throw Error(`Tile for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `imod(resRC, ${aShape[0]})`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    sourceCoords.push(`imod(${currentCoords[i]}, ${aShape[i]})`);\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class UnaryOpProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      float unaryOperation(float x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        float x = getAAtOutCoords();\n        float y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n\nconst CHECK_NAN_SNIPPET = `if (isnan(x)) return x;`;\n\nexport const LINEAR = `return x;`;\n\nexport const ABS = `return abs(x);`;\n\nexport const RELU = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : x;\n`;\n\nexport const RELU6 = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : min(6.0, x);\n`;\n\nexport const ELU = `return (x >= 0.0) ? x : (exp(x) - 1.0);`;\n\nexport const SELU = `\n  // Stable and Attracting Fixed Point (0, 1) for Normalized Weights.\n  // see: https://arxiv.org/abs/1706.02515\n  float scaleAlpha = ${backend_util.SELU_SCALEALPHA};\n  float scale = ${backend_util.SELU_SCALE};\n  return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);\n`;\n\nexport function STEP(alpha = 0.0) {\n  return CHECK_NAN_SNIPPET + `\n    return x > 0.0 ? 1.0 : float(${alpha});\n  `;\n}\n\nexport const NEG = `return -x;`;\n\nexport const CEIL = `return ceil(x);`;\n\nexport const FLOOR = `return floor(x);`;\n\nexport const SIGN = `\n  if (isnan(x)) { return 0.0; }\n  return sign(x);\n`;\n\nexport const IS_NAN = `return float(isnan(x));`;\n\nexport const IS_INF = `return float(isinf(x));`;\n\nexport const IS_FINITE = `return float(!isnan(x) && !isinf(x));`;\n\nexport const ROUND = `\n  // OpenGL ES does not support round function.\n  // The algorithm is based on banker's rounding.\n  float base = floor(x);\n  if ((x - base) < 0.5) {\n    return floor(x);\n  } else if ((x - base) > 0.5) {\n    return ceil(x);\n  } else {\n    if (mod(base, 2.0) == 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n`;\n\nexport const EXP = `return exp(x);`;\n\nexport const EXPM1 = `return exp(x) - 1.0;`;\n\nexport const LOG = `if (x < 0.0) return NAN;\n  return log(x);`;\n\nexport const LOG1P = `return log(1.0 + x);`;\n\nexport const SQRT = `return sqrt(x);`;\n\nexport const RSQRT = `return inversesqrt(x);`;\n\nexport const SIGMOID = `return 1.0 / (1.0 + exp(-1.0 * x));`;\n\n/**\n * mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n *\n * epsilon is the difference between 1.0 and the next representable\n * float. For a single precision 32 bit float this should be 2^-23, see:\n * https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\n *\n * too_large = (x > -threshold) is value above which exp(x) may overflow\n * but softplus(x) == x is within machine epsilon\n *\n * too_small = (x < threshold) is value below which exp(x) may underflow,\n * but softplus(x) == exp(x) is within machine epsilon.\n */\nexport const SOFTPLUS = `\n  float epsilon = 1.1920928955078125e-7;\n  float threshold = log(epsilon) + 2.0;\n\n  bool too_large = x > -threshold;\n  bool too_small = x < threshold;\n\n  float result;\n  float exp_x = exp(x);\n\n  if (too_large){\n    result = x;\n  }\n  else if (too_small){\n    result = exp_x;\n  }\n  else{\n    result = log(exp_x + 1.0);\n  }\n  return result;\n`;\n\nexport const ASIN = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return asin(x);\n`;\n\nexport const ACOS = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return acos(x);\n`;\n\nexport const ATAN = CHECK_NAN_SNIPPET + `\n  return atan(x);\n`;\n\nexport const SINH = `\n  float e2x = exp(x);\n  return (e2x - 1.0 / e2x) / 2.0;\n`;\n\nexport const COSH = `\n  float e2x = exp(-x);\n  return (e2x + 1.0 / e2x) / 2.0;\n`;\n\nexport const TANH = `\n  float e2x = exp(-2.0 * abs(x));\n  return sign(x) * (1.0 - e2x) / (1.0 + e2x);\n`;\n\nexport const ASINH = CHECK_NAN_SNIPPET + `return log(x + sqrt(x * x + 1.0));`;\n\nexport const ACOSH = CHECK_NAN_SNIPPET + `\n  if (x < 1.0) return NAN;\n  return log(x + sqrt(x * x - 1.0));`;\n\nexport const ATANH = CHECK_NAN_SNIPPET + `\n  if ((x < -1.0) || (x > 1.0)) return NAN;\n  return (log(1.0 + x) - log(1.0 - x)) / 2.0;`;\n\nexport const ERF = `\n  // Error function is calculated approximately with elementary function.\n  // See \"Handbook of Mathematical Functions with Formulas,\n  // Graphs, and Mathematical Tables\", Abramowitz and Stegun.\n  float p = ${backend_util.ERF_P};\n  float a1 = ${backend_util.ERF_A1};\n  float a2 = ${backend_util.ERF_A2};\n  float a3 = ${backend_util.ERF_A3};\n  float a4 = ${backend_util.ERF_A4};\n  float a5 = ${backend_util.ERF_A5};\n\n  float sign = sign(x);\n  x = abs(x);\n  float t = 1.0 / (1.0 + p * x);\n  return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));\n`;\n\nexport const RECIPROCAL = `return 1.0 / x;`;\n\nexport const LOGICAL_NOT = `return float(!(x >= 1.0));`;\n\nexport const TO_INT = `return float(int(x));`;\n\nexport const CLONE = 'return x;';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const LINEAR = `return x;`;\n\nexport const LOG = `\n  vec4 result = log(x);\n  vec4 isNaN = vec4(lessThan(x, vec4(0.0)));\n  result.r = isNaN.r == 1.0 ? NAN : result.r;\n  result.g = isNaN.g == 1.0 ? NAN : result.g;\n  result.b = isNaN.b == 1.0 ? NAN : result.b;\n  result.a = isNaN.a == 1.0 ? NAN : result.a;\n\n  return result;\n`;\n\nexport const RELU = `\n  vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const RELU6 = `\n  vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const ELU = `\n  vec4 result;\n\n  result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);\n  result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);\n  result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);\n  result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);\n\n  return result;\n`;\n\nexport class UnaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      vec4 unaryOperation(vec4 x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        vec4 x = getAAtOutCoords();\n        vec4 y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels, getSourceCoords} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class UnpackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = false;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: number[]) {\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    const channels = getChannels('rc', rank);\n    const dtype = getCoordsDataType(rank);\n    const sourceCoords = getSourceCoords(rank, channels);\n    const innerDims = channels.slice(-2);\n    const coords = rank <= 1 ? 'rc' : `vec2(${innerDims.join(',')})`;\n\n    this.userCode = `\n      void main() {\n        ${dtype} rc = getOutputCoords();\n        vec4 packedInput = getA(${sourceCoords});\n\n        setOutput(getChannel(packedInput, ${coords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import webgl flags.\nimport './flags_webgl';\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {complex, DataId, div, engine, env, imag, max, MemoryInfo, range, real, RecursiveArray, reshape, scalar, softmax, tensor, tidy, TimingInfo, transpose} from '@tensorflow/tfjs-core';\nimport {backend_util, buffer, kernel_impls, slice_util, util} from '@tensorflow/tfjs-core';\nimport {DataStorage, DataType, KernelBackend, NumericDataType, Rank, Scalar, ShapeMap, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorInfo, TypedArray, upcastType} from '@tensorflow/tfjs-core';\n\nimport {addImplCPU, ceilImplCPU, expImplCPU, expm1ImplCPU, floorImplCPU, logImplCPU, multiplyImplCPU, rsqrtImplCPU, simpleAbsImplCPU, sliceImplCPU, subImplCPU} from './kernel_utils/shared';\n\nconst {segment_util} = backend_util;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\n\nimport {AddNProgram} from './addn_gpu';\nimport {AddNPackedProgram} from './addn_packed_gpu';\nimport {ArgMinMaxProgram} from './argminmax_gpu';\nimport {ArgMinMaxPackedProgram} from './argminmax_packed_gpu';\nimport {AvgPool3DBackpropProgram} from './avg_pool_backprop_gpu';\nimport * as binaryop_complex_gpu from './binaryop_complex_gpu';\nimport {BinaryOpComplexProgram} from './binaryop_complex_gpu';\nimport * as binaryop_gpu from './binaryop_gpu';\nimport {BinaryOpProgram} from './binaryop_gpu';\nimport * as binaryop_packed_gpu from './binaryop_packed_gpu';\nimport {BinaryOpPackedProgram} from './binaryop_packed_gpu';\nimport {getWebGLContext} from './canvas_util';\nimport {ClipProgram} from './clip_gpu';\nimport {ClipPackedProgram} from './clip_packed_gpu';\nimport {ComplexAbsProgram} from './complex_abs_gpu';\nimport {ConcatProgram} from './concat_gpu';\nimport {ConcatPackedProgram} from './concat_packed_gpu';\nimport {Conv2DDerFilterProgram, Conv2DDerInputProgram, Conv3DDerFilterProgram, Conv3DDerInputProgram} from './conv_backprop_gpu';\nimport {DepthwiseConv2DDerFilterProgram, DepthwiseConv2DDerInputProgram} from './conv_backprop_gpu_depthwise';\nimport {Conv2DProgram, Conv3DProgram} from './conv_gpu';\nimport {DepthwiseConv2DProgram} from './conv_gpu_depthwise';\nimport {DepthwiseConvPacked2DProgram} from './conv_packed_gpu_depthwise';\nimport {CropAndResizeProgram} from './crop_and_resize_gpu';\nimport {CumSumProgram} from './cumsum_gpu';\nimport {DecodeMatrixProgram} from './decode_matrix_gpu';\nimport {DecodeMatrixPackedProgram} from './decode_matrix_packed_gpu';\nimport {DepthToSpaceProgram} from './depth_to_space_gpu';\nimport {DiagProgram} from './diag_gpu';\nimport {EncodeFloatProgram} from './encode_float_gpu';\nimport {EncodeFloatPackedProgram} from './encode_float_packed_gpu';\nimport {EncodeMatrixProgram} from './encode_matrix_gpu';\nimport {EncodeMatrixPackedProgram} from './encode_matrix_packed_gpu';\nimport * as fft_gpu from './fft_gpu';\nimport {FFTProgram} from './fft_gpu';\nimport {FillProgram} from './fill_gpu';\nimport {GatherProgram} from './gather_gpu';\nimport {GatherNDProgram} from './gather_nd_gpu';\nimport {GPGPUContext} from './gpgpu_context';\nimport * as gpgpu_math from './gpgpu_math';\nimport {GPGPUBinary, GPGPUProgram, TensorData} from './gpgpu_math';\nimport {Im2ColPackedProgram} from './im2col_packed_gpu';\nimport {LRNProgram} from './lrn_gpu';\nimport {LRNGradProgram} from './lrn_grad_gpu';\nimport {LRNPackedProgram} from './lrn_packed_gpu';\nimport {MaxPool3DBackpropProgram} from './max_pool_backprop_gpu';\nimport {MatMulPackedProgram} from './mulmat_packed_gpu';\nimport {MultinomialProgram} from './multinomial_gpu';\nimport {OneHotProgram} from './onehot_gpu';\nimport {PackProgram} from './pack_gpu';\nimport {PadProgram} from './pad_gpu';\nimport {PadPackedProgram} from './pad_packed_gpu';\nimport {Pool3DProgram} from './pool_gpu';\nimport {ReduceProgram} from './reduce_gpu';\nimport {ReshapePackedProgram} from './reshape_packed_gpu';\nimport {ResizeBilinearBackpropProgram} from './resize_bilinear_backprop_gpu';\nimport {ResizeBilinearProgram} from './resize_bilinear_gpu';\nimport {ResizeBilinearPackedProgram} from './resize_bilinear_packed_gpu';\nimport {ResizeNearestNeigborBackpropProgram} from './resize_nearest_neighbor_backprop_gpu';\nimport {ResizeNearestNeighborProgram} from './resize_nearest_neighbor_gpu';\nimport {ReverseProgram} from './reverse_gpu';\nimport {ReversePackedProgram} from './reverse_packed_gpu';\nimport {ScatterProgram} from './scatter_gpu';\nimport {SegmentOpProgram} from './segment_gpu';\nimport {SelectProgram} from './select_gpu';\nimport {SliceProgram} from './slice_gpu';\nimport {SlicePackedProgram} from './slice_packed_gpu';\nimport {StridedSliceProgram} from './strided_slice_gpu';\nimport * as tex_util from './tex_util';\nimport {TextureData, TextureUsage} from './tex_util';\nimport {TextureManager} from './texture_manager';\nimport {TileProgram} from './tile_gpu';\nimport * as unary_op from './unaryop_gpu';\nimport {UnaryOpProgram} from './unaryop_gpu';\nimport * as unary_packed_op from './unaryop_packed_gpu';\nimport {UnaryOpPackedProgram} from './unaryop_packed_gpu';\nimport {UnpackProgram} from './unpack_gpu';\nimport * as webgl_util from './webgl_util';\nimport {BackendValues} from '@tensorflow/tfjs-core';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\ntype KernelInfo = {\n  name: string; query: Promise<number>;\n};\n\nexport type TimerNode = RecursiveArray<KernelInfo>|KernelInfo;\nexport interface CPUTimerQuery {\n  startMs: number;\n  endMs?: number;\n}\n\nexport interface WebGLMemoryInfo extends MemoryInfo {\n  numBytesInGPU: number;\n  // Tracks the total number of bytes allocated on the GPU, accounting for the\n  // physical texture type.\n  numBytesInGPUAllocated: number;\n  // Tracks byte size of textures that were created and then made available for\n  // reuse (disposed).\n  numBytesInGPUFree: number;\n  unreliable: boolean;\n}\n\nexport interface WebGLTimingInfo extends TimingInfo {\n  uploadWaitMs: number;\n  downloadWaitMs: number;\n}\n\nconst binaryCaches: {[webGLVersion: string]: {[key: string]: GPGPUBinary}} = {};\n\nexport function getBinaryCache(webGLVersion: number) {\n  if (webGLVersion in binaryCaches) {\n    return binaryCaches[webGLVersion];\n  }\n  binaryCaches[webGLVersion] = {};\n  return binaryCaches[webGLVersion];\n}\n\nfunction mapActivationToShaderProgram(\n    activation: backend_util.Activation, packed = false): string {\n  if (activation === 'linear') {\n    if (packed) {\n      return unary_packed_op.LINEAR;\n    }\n    return unary_op.LINEAR;\n  } else if (activation === 'relu') {\n    if (packed) {\n      return unary_packed_op.RELU;\n    }\n    return unary_op.RELU;\n  } else if (activation === 'elu') {\n    if (packed) {\n      return unary_packed_op.ELU;\n    }\n    return unary_op.ELU;\n  } else if (activation === 'relu6') {\n    if (packed) {\n      return unary_packed_op.RELU6;\n    }\n    return unary_op.RELU6;\n  } else if (activation === 'prelu') {\n    if (packed) {\n      return binaryop_packed_gpu.PRELU;\n    }\n    return binaryop_gpu.PRELU;\n  }\n  throw new Error(`Activation ${\n      activation} has not been implemented for the WebGL backend.`);\n}\n\n// Empirically determined constant used to determine size threshold for handing\n// off execution to the CPU.\nconst CPU_HANDOFF_SIZE_THRESHOLD = 128;\n\n// Empirically determined constant used to decide the number of MB on GPU\n// before we warn about high memory use. The MB are this constant * screen area\n// * dpi / 1024 / 1024.\nconst BEFORE_PAGING_CONSTANT = 600;\nfunction numMBBeforeWarning(): number {\n  if (env().global.screen == null) {\n    return 1024;  // 1 GB.\n  }\n  return (env().global.screen.height * env().global.screen.width *\n          window.devicePixelRatio) *\n      BEFORE_PAGING_CONSTANT / 1024 / 1024;\n}\n\n// Empirically determined minimal shared dimension in matmul before we forward\n// to a.mul(b).sum() in order to take advantage of GPU parallelism. See\n// https://github.com/tensorflow/tfjs-core/pull/1379 for benchmarks.\nexport const MATMUL_SHARED_DIM_THRESHOLD = 1000;\n\nexport class MathBackendWebGL extends KernelBackend {\n  texData: DataStorage<TextureData>;\n  gpgpu: GPGPUContext;\n\n  // Maps data ids that have a pending read operation, to list of subscribers.\n  private pendingRead = new WeakMap<DataId, Array<(arr: TypedArray) => void>>();\n  // List of data ids that are scheduled for disposal, but are waiting on a\n  // pending read operation.\n  private pendingDisposal = new WeakSet<DataId>();\n  // Used to count the number of 'shallow' sliced tensors that point to the\n  // same data id.\n  private dataRefCount = new WeakMap<DataId, number>();\n  private numBytesInGPU = 0;\n\n  private canvas: HTMLCanvasElement|OffscreenCanvas;\n\n  private programTimersStack: TimerNode[];\n  private activeTimers: TimerNode[];\n  // Accumulated time spent (including blocking) in uploading data to webgl.\n  private uploadWaitMs = 0;\n  // Accumulated time spent (including blocking in downloading data from webgl.\n  private downloadWaitMs = 0;\n  private cpuBackend: KernelBackend;\n\n  // Number of bits of precision of this backend.\n  private floatPrecisionValue: 32|16;\n\n  private textureManager: TextureManager;\n  private binaryCache: {[key: string]: GPGPUBinary};\n  private gpgpuCreatedLocally: boolean;\n  private numMBBeforeWarning: number;\n  private warnedAboutMemory = false;\n  private warnedAboutCPUBackend = false;\n\n  constructor(gpgpu?: GPGPUContext) {\n    super();\n    if (!env().getBool('HAS_WEBGL')) {\n      throw new Error('WebGL is not supported on this device');\n    }\n\n    if (gpgpu == null) {\n      const gl = getWebGLContext(env().getNumber('WEBGL_VERSION'));\n      this.binaryCache = getBinaryCache(env().getNumber('WEBGL_VERSION'));\n      this.gpgpu = new GPGPUContext(gl);\n      this.canvas = gl.canvas;\n      this.gpgpuCreatedLocally = true;\n    } else {\n      this.gpgpu = gpgpu;\n      this.binaryCache = {};\n      this.gpgpuCreatedLocally = false;\n      this.canvas = gpgpu.gl.canvas;\n    }\n    this.textureManager = new TextureManager(this.gpgpu);\n    this.numMBBeforeWarning = numMBBeforeWarning();\n\n    this.texData = new DataStorage(this, engine());\n  }\n\n  numDataIds() {\n    return this.texData.numDataIds() +\n        (this.cpuBackend ? this.cpuBackend.numDataIds() : 0) -\n        this.pendingDeletes;\n  }\n\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    if (env().getBool('WEBGL_CHECK_NUMERICAL_PROBLEMS') ||\n        env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64' && values != null) {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    const dataId = {};\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n    return dataId;\n  }\n\n  /** Increase refCount of a `TextureData`. */\n  incRef(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    texData.refCount++;\n  }\n\n  /** Decrease refCount of a `TextureData`. */\n  decRef(dataId: DataId): void {\n    if (this.texData.has(dataId)) {\n      const texData = this.texData.get(dataId);\n      texData.refCount--;\n    }\n  }\n\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    if (env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.texData.has(dataId)) {\n      const textureData = this.texData.get(dataId);\n\n      textureData.refCount--;\n\n      if (textureData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    const texData = this.texData.get(dataId);\n    const {values, dtype, complexTensors, slice, shape, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.readSync(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n    if (dtype === 'string') {\n      return values;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let result: Float32Array;\n    if (dtype === 'complex64') {\n      const realValues = complexTensors.real.dataSync() as Float32Array;\n      const imagValues = complexTensors.imag.dataSync() as Float32Array;\n      result = backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    } else {\n      result = this.getValuesFromTexture(dataId);\n    }\n\n    if (shouldTimeProgram) {\n      this.downloadWaitMs += util.now() - start;\n    }\n    return this.convertAndCacheOnCPU(dataId, result);\n  }\n\n  async read(dataId: DataId): Promise<BackendValues> {\n    if (this.pendingRead.has(dataId)) {\n      const subscribers = this.pendingRead.get(dataId);\n      return new Promise<TypedArray>(resolve => subscribers.push(resolve));\n    }\n    const texData = this.texData.get(dataId);\n    const {values, shape, slice, dtype, complexTensors, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.read(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n\n    if (!env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED') &&\n        env().getNumber('WEBGL_VERSION') === 2) {\n      throw new Error(\n          `tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and ` +\n          `WEBGL_VERSION=2 not yet supported.`);\n    }\n\n    let buffer = null;\n    let tmpDownloadTarget: TensorInfo;\n\n    if (dtype !== 'complex64' && env().get('WEBGL_BUFFER_SUPPORTED')) {\n      // Possibly copy the texture into a buffer before inserting a fence.\n      tmpDownloadTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpDownloadTarget.dataId);\n\n      buffer = this.gpgpu.createBufferFromTexture(\n          tmpData.texture, ...tex_util.getDenseTexShape(shape));\n    }\n\n    this.pendingRead.set(dataId, []);\n\n    if (dtype !== 'complex64') {\n      // Create a fence and wait for it to resolve.\n      await this.gpgpu.createAndWaitForFence();\n    }\n\n    // Download the values from the GPU.\n    let vals: Float32Array;\n    if (dtype === 'complex64') {\n      const ps = await Promise.all(\n          [complexTensors.real.data(), complexTensors.imag.data()]);\n      const realValues = ps[0];\n      const imagValues = ps[1];\n      vals = backend_util.mergeRealAndImagArrays(\n          realValues as Float32Array, imagValues as Float32Array);\n    } else if (buffer == null) {\n      vals = this.getValuesFromTexture(dataId);\n    } else {\n      const size = util.sizeFromShape(shape);\n      vals = this.gpgpu.downloadFloat32MatrixFromBuffer(buffer, size);\n    }\n    if (tmpDownloadTarget != null) {\n      this.disposeIntermediateTensorInfo(tmpDownloadTarget);\n    }\n    const dTypeVals = this.convertAndCacheOnCPU(dataId, vals);\n\n    const subscribers = this.pendingRead.get(dataId);\n    this.pendingRead.delete(dataId);\n\n    // Notify all pending reads.\n    subscribers.forEach(resolve => resolve(dTypeVals));\n    if (this.pendingDisposal.has(dataId)) {\n      this.pendingDisposal.delete(dataId);\n      this.disposeData(dataId);\n      this.pendingDeletes--;\n    }\n    return dTypeVals;\n  }\n\n  private checkNumericalProblems(values: BackendValues): void {\n    if (values == null) {\n      return;\n    }\n    for (let i = 0; i < values.length; i++) {\n      const num = values[i] as number;\n      if (!webgl_util.canBeRepresented(num)) {\n        if (env().getBool('WEBGL_RENDER_FLOAT32_CAPABLE')) {\n          throw Error(\n              `The value ${num} cannot be represented with your ` +\n              `current settings. Consider enabling float32 rendering: ` +\n              `'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`);\n        }\n        throw Error(`The value ${num} cannot be represented on this device.`);\n      }\n    }\n  }\n\n  private getValuesFromTexture(dataId: DataId): Float32Array {\n    const {shape, dtype, isPacked} = this.texData.get(dataId);\n    const size = util.sizeFromShape(shape);\n    if (env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED')) {\n      const tmpTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpTarget.dataId);\n      const vals = this.gpgpu\n                       .downloadMatrixFromPackedTexture(\n                           tmpData.texture, ...tex_util.getDenseTexShape(shape))\n                       .subarray(0, size);\n\n      this.disposeIntermediateTensorInfo(tmpTarget);\n\n      return vals;\n    }\n\n    const shouldUsePackedProgram =\n        env().getBool('WEBGL_PACK') && isPacked === true;\n    const outputShape =\n        shouldUsePackedProgram ? webgl_util.getShapeAs3D(shape) : shape;\n    const program = shouldUsePackedProgram ?\n        new EncodeFloatPackedProgram(outputShape as [number, number, number]) :\n        new EncodeFloatProgram(outputShape);\n    const output = this.runWebGLProgram(\n        program, [{shape: outputShape, dtype, dataId}], 'float32');\n    const tmpData = this.texData.get(output.dataId);\n    const vals =\n        this.gpgpu\n            .downloadByteEncodedFloatMatrixFromOutputTexture(\n                tmpData.texture, tmpData.texShape[0], tmpData.texShape[1])\n            .subarray(0, size);\n    this.disposeIntermediateTensorInfo(output);\n\n    return vals;\n  }\n\n  async time(f: () => void): Promise<WebGLTimingInfo> {\n    const oldActiveTimers = this.activeTimers;\n    const newActiveTimers: TimerNode[] = [];\n\n    let outerMostTime = false;\n    if (this.programTimersStack == null) {\n      this.programTimersStack = newActiveTimers;\n      outerMostTime = true;\n    } else {\n      this.activeTimers.push(newActiveTimers);\n    }\n    this.activeTimers = newActiveTimers;\n\n    f();\n\n    // needing to split these up because util.flatten only accepts certain types\n    const flattenedActiveTimerQueries =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.query))\n            .filter(d => d != null);\n    const flattenedActiveTimerNames =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.name))\n            .filter(d => d != null);\n\n    this.activeTimers = oldActiveTimers;\n\n    if (outerMostTime) {\n      this.programTimersStack = null;\n    }\n\n    const res: WebGLTimingInfo = {\n      uploadWaitMs: this.uploadWaitMs,\n      downloadWaitMs: this.downloadWaitMs,\n      kernelMs: null,\n      wallMs: null  // will be filled by the engine\n    };\n\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      const kernelMs = await Promise.all(flattenedActiveTimerQueries);\n\n      res['kernelMs'] = util.sum(kernelMs);\n      res['getExtraProfileInfo'] = () =>\n          kernelMs.map((d, i) => ({name: flattenedActiveTimerNames[i], ms: d}))\n              .map(d => `${d.name}: ${d.ms}`)\n              .join(', ');\n    } else {\n      res['kernelMs'] = {\n        error: 'WebGL query timers are not supported in this environment.'\n      };\n    }\n\n    this.uploadWaitMs = 0;\n    this.downloadWaitMs = 0;\n    return res;\n  }\n  memory(): WebGLMemoryInfo {\n    return {\n      unreliable: false,\n      numBytesInGPU: this.numBytesInGPU,\n      numBytesInGPUAllocated: this.textureManager.numBytesAllocated,\n      numBytesInGPUFree: this.textureManager.numBytesFree\n    } as WebGLMemoryInfo;\n  }\n\n  private startTimer(): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.beginQuery();\n    }\n    return {startMs: util.now(), endMs: null};\n  }\n\n  private endTimer(query: WebGLQuery|CPUTimerQuery): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      this.gpgpu.endQuery();\n      return query;\n    }\n    (query as CPUTimerQuery).endMs = util.now();\n    return query;\n  }\n\n  private async getQueryTime(query: WebGLQuery|CPUTimerQuery): Promise<number> {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.waitForQueryAndGetTime(query as WebGLQuery);\n    }\n    const timerQuery = query as CPUTimerQuery;\n    return timerQuery.endMs - timerQuery.startMs;\n  }\n\n  private pendingDeletes = 0;\n\n  disposeData(dataId: DataId): void {\n    if (this.pendingDisposal.has(dataId)) {\n      return;\n    }\n    if (this.pendingRead.has(dataId)) {\n      this.pendingDisposal.add(dataId);\n      this.pendingDeletes++;\n      return;\n    }\n    // No-op if already disposed.\n    if (!this.texData.has(dataId)) {\n      return;\n    }\n\n    this.releaseGPUData(dataId);\n    const {complexTensors} = this.texData.get(dataId);\n    if (complexTensors != null) {\n      complexTensors.real.dispose();\n      complexTensors.imag.dispose();\n    }\n    this.texData.delete(dataId);\n  }\n\n  private releaseGPUData(dataId: DataId): void {\n    const {texture, dtype, texShape, usage, isPacked, slice} =\n        this.texData.get(dataId);\n    const key = slice && slice.origDataId || dataId;\n    const refCount = this.dataRefCount.get(key);\n    if (refCount > 1) {\n      this.dataRefCount.set(key, refCount - 1);\n    } else {\n      this.dataRefCount.delete(key);\n      if (texture != null) {\n        this.numBytesInGPU -= this.computeBytes(texShape, dtype);\n        this.textureManager.releaseTexture(texture, texShape, usage, isPacked);\n      }\n    }\n    const texData = this.texData.get(dataId);\n    texData.texture = null;\n    texData.texShape = null;\n    texData.isPacked = false;\n    texData.slice = null;\n  }\n\n  getTexture(dataId: DataId): WebGLTexture {\n    this.uploadToGPU(dataId);\n    return this.texData.get(dataId).texture;\n  }\n\n  /**\n   * Returns internal information for the specific data bucket. Used in unit\n   * tests.\n   */\n  getDataInfo(dataId: DataId): TextureData {\n    return this.texData.get(dataId);\n  }\n\n  private getCPUBackend(): KernelBackend|null {\n    if (!env().getBool('WEBGL_CPU_FORWARD')) {\n      return null;\n    }\n\n    if (this.cpuBackend == null) {\n      this.cpuBackend = engine().findBackend('cpu');\n    }\n\n    return this.cpuBackend;\n  }\n\n  /*\n  Tests whether all the inputs to an op are small and on the CPU. This heuristic\n  determines when it would be faster to execute a kernel on the CPU. WebGL\n  kernels opt into running this check and forwarding when appropriate.\n  TODO(https://github.com/tensorflow/tfjs/issues/872): Develop a more\n  sustainable strategy for optimizing backend execution of ops.\n   */\n  shouldExecuteOnCPU(\n      inputs: TensorInfo[],\n      sizeThreshold = CPU_HANDOFF_SIZE_THRESHOLD): boolean {\n    const cpuBackend = this.getCPUBackend();\n    if (!this.warnedAboutCPUBackend && cpuBackend == null) {\n      console.warn(\n          'Your application contains ops that are small enough to be ' +\n          'executed on the CPU backend, however the CPU backend cannot ' +\n          'be found. Consider importing the CPU backend ' +\n          '(@tensorflow/tfjs-backend-cpu) for better performance.');\n\n      this.warnedAboutCPUBackend = true;\n    }\n\n    return cpuBackend != null &&\n        inputs.every(\n            input => this.texData.get(input.dataId).texture == null &&\n                util.sizeFromShape(input.shape) < sizeThreshold);\n  }\n\n  getGPGPUContext(): GPGPUContext {\n    return this.gpgpu;\n  }\n\n  complex<T extends Tensor>(real: T, imag: T): T {\n    const result = this.makeOutput(real.shape, 'complex64');\n    const resultData = this.texData.get(result.dataId);\n    // The backend owns the reference to the underlying real and imaginary\n    // clones. These will explicitly get disposed when the complex tensor is\n    // disposed.\n    resultData.complexTensors = {\n      real: engine().keep(real.clone()),\n      imag: engine().keep(imag.clone())\n    };\n\n    return result as T;\n  }\n  real<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.real.clone() as T;\n  }\n  imag<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.imag.clone() as T;\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = sliceImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, begin, size, x.shape,\n          x.dtype);\n      return this.makeOutput(size, x.dtype, outValues);\n    }\n    // Short-circuit computation if the slice is zero-sized.\n    if (util.sizeFromShape(size) === 0) {\n      return tensor([], size, x.dtype) as T;\n    }\n    const {isPacked} = this.texData.get(x.dataId);\n    const isContinous = slice_util.isSliceContinous(x.shape, begin, size);\n    if (isPacked || !isContinous) {\n      const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n          new SlicePackedProgram(size) :\n          new SliceProgram(size);\n      const customSetup = program.getCustomSetupFunc(begin);\n      return this.compileAndRun(program, [x], null, customSetup);\n    }\n    this.uploadToGPU(x.dataId);\n    return this.shallowSlice(x, begin, size) as T;\n  }\n\n  private shallowSlice(x: Tensor, begin: number[], size: number[]): Tensor {\n    const xTexData = this.texData.get(x.dataId);\n    const t = this.makeOutput(size, x.dtype);\n    const newTexData = this.texData.get(t.dataId);\n    // Copy texture data from the original tensor.\n    Object.assign(newTexData, xTexData);\n    newTexData.shape = size;\n    newTexData.dtype = x.dtype;\n    let flatOffset = slice_util.computeFlatOffset(begin, x.strides);\n    if (xTexData.slice) {\n      // We are slicing an already sliced tensor, so we have to accumulate\n      // the offset.\n      flatOffset += xTexData.slice.flatOffset;\n    }\n    newTexData.slice = {\n      flatOffset,\n      // Point to the original dataId, which is used to do ref counting.\n      origDataId: xTexData.slice && xTexData.slice.origDataId || x.dataId\n    };\n\n    // Increase the ref count for that data bucket.\n    const refCount = this.dataRefCount.get(newTexData.slice.origDataId) || 1;\n    this.dataRefCount.set(newTexData.slice.origDataId, refCount + 1);\n\n    return t;\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x], () => this.cpuBackend.stridedSlice(x, begin, end, strides));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tensor([], outShape) as T;\n    }\n\n    const program = new StridedSliceProgram(begin, strides, outShape);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new ReversePackedProgram(x.shape, axis) :\n        new ReverseProgram(x.shape, axis);\n    return this.compileAndRun(program, [x]);\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    if (tensors[0].dtype === 'complex64') {\n      const reals = tensors.map((t) => real(t));\n      const imags = tensors.map((t) => imag(t));\n      return complex(this.concat(reals, axis), this.concat(imags, axis));\n    }\n\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n    if (tensors.length > env().getNumber('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.concat(tensors.slice(0, midIndex), axis);\n      const rightSide = this.concat(tensors.slice(midIndex), axis);\n      return this.concat([leftSide, rightSide], axis);\n    }\n    if (env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') && tensors[0].rank > 1) {\n      const program = new ConcatPackedProgram(tensors.map(t => t.shape), axis);\n      return this.compileAndRun(program, tensors);\n    }\n    // Any concat of n-dimensional tensors across any axis can be reduced to\n    // a concatenation of two-dimensional tensors across the axis 1 by first\n    // partitioning the axes of the original tensors into those less than the\n    // axis to be concatenated and the rest. Then reshape the tensors\n    // into a two-dimensional tensor by collapsing these two sets of axes and\n    // concatenate the resulting matrices across the axis 1, finally reshaping\n    // the result to have the proper shape.\n    const outShape =\n        backend_util.computeOutShape(tensors.map(t => t.shape), axis);\n    const tensors2D =\n        tensors.map(t => t.as2D(-1, util.sizeFromShape(t.shape.slice(axis))));\n    const program = new ConcatProgram(tensors2D.map(t => t.shape));\n    const res: Tensor = this.compileAndRun(program, tensors2D);\n    return res.reshape(outShape);\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    const cpuRes = this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.neg(x));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.NEG, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.NEG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const [batch, , ] = a.shape;\n\n    // Since the matrices are vectors, it is faster to call mul().sum()\n    // because sum() is O(sqrt(N)) due to divide-and-conquer.\n    if ((outerShapeA === 1 || outerShapeB === 1) &&\n        sharedDim > MATMUL_SHARED_DIM_THRESHOLD) {\n      if (transposeA) {\n        a = transpose(a, [0, 2, 1]);\n      }\n      if (transposeB) {\n        b = transpose(b, [0, 2, 1]);\n      }\n\n      const a3D = outerShapeB === 1 ? a : a.as3D(batch, sharedDim, 1);\n      const axis = outerShapeB === 1 ? 2 : 1;\n      const b3D = outerShapeB === 1 ? b.as3D(batch, 1, sharedDim) : b;\n      return this.multiply(a3D, b3D).sum(axis, true /* keepDims */);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB);\n    return this.compileAndRun<Tensor3D>(program, [a, b], dtype);\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const [batch, , ] = a.shape;\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB,\n        hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [a, b];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun<Tensor3D>(program, inputs, dtype);\n  }\n\n  multiply(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64') {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n\n      const realProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.REAL, a.shape, b.shape);\n      const imagProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.IMAG, a.shape, b.shape);\n\n      const inputs = [\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.imag),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.imag)\n      ];\n      const real = this.compileAndRun<Tensor>(realProgram, inputs);\n      const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n\n      const complex = this.complex(real, imag);\n      real.dispose();\n      imag.dispose();\n      return complex;\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = multiplyImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.MUL, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.MUL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], a.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n        new LRNPackedProgram(x.shape, radius, bias, alpha, beta) :\n        new LRNProgram(x.shape, radius, bias, alpha, beta);\n    return this.compileAndRun(program, [x]);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program =\n        new LRNGradProgram(inputImage.shape, depthRadius, bias, alpha, beta);\n    return this.compileAndRun(program, [inputImage, outputImage, dy]);\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    if (x.dtype === 'string') {\n      const data = this.readSync(x.dataId) as Uint8Array[];\n      const decodedData = data.map(d => util.decodeString(d));\n      const buf = buffer(x.shape, x.dtype, decodedData);\n      return tile(buf, reps) as T;\n    }\n    const program = new TileProgram(x.shape, reps);\n    return this.compileAndRun(program, [x]);\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new PadPackedProgram(x.shape, paddings, constantValue) :\n        new PadProgram(x.shape, paddings, constantValue);\n    return this.compileAndRun(program, [x]);\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x, indices], () => this.cpuBackend.gather(x, indices, axis));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = new GatherProgram(x.shape, indices.size, axis);\n    return this.compileAndRun(program, [x, indices]);\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'batchToSpaceND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: Array<[number, number]>): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'spaceToBatchND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const completePaddings: Array<[number, number]> = [[0, 0]];\n    completePaddings.push(...paddings);\n    for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n      completePaddings.push([0, 0]);\n    }\n\n    const paddedX = x.pad(completePaddings);\n\n    const reshapedPaddedShape =\n        backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n    const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n        reshapedPaddedShape.length, blockShape.length, false);\n\n    const flattenShape = backend_util.getReshapedPermuted(\n        paddedX.shape, blockShape, prod, false);\n\n    const paddedXT = transpose(\n        paddedX.reshape(reshapedPaddedShape),\n        permutedReshapedPaddedPermutation);\n    return reshape(paddedXT, flattenShape) as T;\n  }\n\n  private reduce(\n      x: Tensor2D, reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod',\n      dtype: DataType): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const outSize = Math.ceil(inSize / windowSize);\n    const reduceInfo = {windowSize, inSize, batchSize, outSize};\n    const program = new ReduceProgram(reduceInfo, reduceType);\n    const output = this.compileAndRun<Tensor2D>(program, [x], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.reduce(output, reduceType, dtype);\n  }\n\n  private argReduce(\n      x: Tensor2D, reduceType: 'max'|'min',\n      bestIndicesA: Tensor2D = null): Tensor2D {\n    let batchSize = x.shape[0];\n    let inSize = x.shape[1];\n    if (bestIndicesA != null) {\n      batchSize = bestIndicesA.shape[0];\n      inSize = bestIndicesA.shape[1];\n    }\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const reduceInfo = {\n      windowSize,\n      inSize,\n      batchSize,\n      outSize: Math.ceil(inSize / windowSize)\n    };\n    const program =\n        new ArgMinMaxProgram(reduceInfo, reduceType, bestIndicesA == null);\n    const inputs = [x];\n    if (bestIndicesA != null) {\n      inputs.push(bestIndicesA);\n    }\n    const output = this.compileAndRun<Tensor2D>(program, inputs, 'int32');\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.argReduce(x, reduceType, output);\n  }\n\n  private argReducePacked(\n      x: Tensor, reduceType: 'max'|'min', bestIndicesA: Tensor = null): Tensor {\n    const inShape = bestIndicesA != null ? bestIndicesA.shape : x.shape;\n    const inSize = inShape[inShape.length - 1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const program = new ArgMinMaxPackedProgram(\n        inShape, windowSize, reduceType, bestIndicesA == null);\n    const inputs = bestIndicesA == null ? [x] : [x, bestIndicesA];\n    const output = this.compileAndRun<Tensor>(program, inputs, 'int32');\n    if (output.rank === x.rank) {\n      return this.argReducePacked(x, reduceType, output);\n    }\n    return output;\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'sum', outputDType).reshape(outShape);\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.prod(x, axes));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'prod', outputDType).reshape(outShape);\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    let axis = 0;\n    const permutation = backend_util.getAxesPermutation([axis], x.rank);\n    let permutedX = x;\n    if (permutation != null) {\n      permutedX = transpose(x, permutation);\n      axis = backend_util.getInnerMostAxes(1, x.rank)[0];\n    }\n\n    const outShape =\n        segment_util.computeOutShape(permutedX.shape, axis, numSegments);\n    const inSize = util.sizeFromShape([permutedX.shape[axis]]);\n    const a2D = permutedX.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    let result =\n        this.segOpCompute(\n                a2D, 'unsortedSegmentSum', segmentIds, outputDType, numSegments)\n            .reshape(outShape);\n    if (permutation != null) {\n      result =\n          transpose(result, backend_util.getUndoAxesPermutation(permutation));\n    }\n    return result;\n  }\n\n  private segOpCompute(\n      x: Tensor2D, segOpType: 'unsortedSegmentSum', segmentIds: Tensor1D,\n      dtype: DataType, numSegments: number): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize =\n        segment_util.segOpComputeOptimalWindowSize(inSize, numSegments);\n    const segOpInfo = {windowSize, inSize, batchSize, numSegments};\n    const program = new SegmentOpProgram(segOpInfo, segOpType);\n    const output =\n        this.compileAndRun<Tensor2D>(program, [x, segmentIds], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === numSegments) {\n      return output;\n    }\n    segmentIds = range(0, numSegments).tile([inSize / windowSize]);\n    return this.segOpCompute(output, segOpType, segmentIds, dtype, numSegments);\n  }\n\n  private argMinMaxReduce(x: Tensor, axis: number, reduceType: 'min'|'max'):\n      Tensor {\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims(\n        'arg' + reduceType.charAt(0).toUpperCase() + reduceType.slice(1), axes,\n        x.rank);\n    if (!env().getBool('WEBGL_PACK_REDUCE') || x.rank <= 2) {\n      const [outShape, reduceShape] =\n          backend_util.computeOutAndReduceShapes(x.shape, axes);\n      const inSize = util.sizeFromShape(reduceShape);\n      const a2D = x.as2D(-1, inSize);\n      return this.argReduce(a2D, reduceType).reshape(outShape);\n    }\n    return this.argReducePacked(x, reduceType);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'min');\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'max');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `WebGL cumsum shader expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const size = x.shape[axis];\n    let result = x;\n    // Use cumsum parallel algorithm, ref:\n    // https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-39-parallel-prefix-sum-scan-cuda\n    for (let i = 0; i <= Math.ceil(Math.log2(size)) - 1; i++) {\n      const program = new CumSumProgram(x.shape, false, reverse);\n      const customSetup = program.getCustomSetupFunc(i);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result], result.dtype, customSetup);\n      prevResult.dispose();\n    }\n    // For exclusive cumsum, shift the end result in the direction of sum and\n    // add 0 to the front index.\n    if (exclusive) {\n      const program = new CumSumProgram(x.shape, exclusive, reverse);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result]);\n      prevResult.dispose();\n    }\n\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.EQUAL, 'bool');\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.NOT_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.NOT_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.less(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.LESS, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LESS_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.greater(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.GREATER, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.GREATER, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.GREATER_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.GREATER_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalNot<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOGICAL_NOT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_AND, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_AND, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_OR, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_OR, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    const program = new SelectProgram(condition.rank, a.shape, a.rank);\n    return this.compileAndRun(\n        program, [condition, a, b], upcastType(a.dtype, b.dtype));\n  }\n\n  where(condition: Tensor): Tensor2D {\n    backend_util.warn(\n        'tf.where() in webgl locks the UI thread. ' +\n        'Call tf.whereAsync() instead');\n    const condVals = condition.dataSync();\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    const xVals = x.dataSync();\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'min', a2D.dtype).reshape(outShape);\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.minimum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MIN, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MIN, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MOD, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MOD, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.maximum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MAX, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MAX, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'all', a2D.dtype).reshape(outShape);\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'any', a2D.dtype).reshape(outShape);\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    const op = binaryop_gpu.INT_DIV;\n    const outputDtype = 'int32';\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.INT_DIV, outputDtype);\n    }\n    const program = new BinaryOpProgram(op, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], outputDtype);\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.ADD);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = addImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.ADD, dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.ADD, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  private packedUnaryOp(x: TensorInfo, op: string, dtype: DataType) {\n    const program = new UnaryOpPackedProgram(x.shape, op);\n    return this.compileAndRun<Tensor>(program, [x], dtype);\n  }\n\n  private packedBinaryOp(\n      a: TensorInfo, b: TensorInfo, op: string, dtype: DataType,\n      checkOutOfBounds = false) {\n    const program =\n        new BinaryOpPackedProgram(op, a.shape, b.shape, checkOutOfBounds);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  /**\n   * Computes a complex binary operation that can be decomposed into a simple\n   * binary operation on both the real and imagary parts.\n   */\n  private complexSeparableBinaryOp(a: Tensor, b: Tensor, op: string): Tensor {\n    const aData = this.texData.get(a.dataId);\n    const bData = this.texData.get(b.dataId);\n\n    const [real, imag] = [\n      [aData.complexTensors.real, bData.complexTensors.real],\n      [aData.complexTensors.imag, bData.complexTensors.imag]\n    ].map(complexParts => {\n      const [aPart, bPart] = complexParts;\n\n      const aHandle = this.makeComplexComponentTensorInfo(a, aPart);\n      const bHandle = this.makeComplexComponentTensorInfo(b, bPart);\n\n      const program = new BinaryOpProgram(op, a.shape, b.shape);\n      return this.compileAndRun<Tensor>(\n          program, [aHandle, bHandle], upcastType(aPart.dtype, bPart.dtype));\n    });\n\n    const complex = this.complex(real, imag);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  // Returns a TensorInfo with the complex shape and the dataId of the\n  // underlying part. We need to do this because a reshaped complex tensor is\n  // not reflected in its parts.\n  private makeComplexComponentTensorInfo(\n      complexTensor: Tensor, complexPart: Tensor): TensorInfo {\n    return {\n      dataId: complexPart.dataId,\n      dtype: complexPart.dtype,\n      shape: complexTensor.shape\n    };\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n\n    // Limit the number of uploaded textures for optimization.\n    if (tensors.length > env().get('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.addN(tensors.slice(0, midIndex));\n      const rightSide = this.addN(tensors.slice(midIndex));\n      return this.addN([leftSide, rightSide]);\n    }\n\n    const dtype =\n        tensors.map(t => t.dtype).reduce((d1, d2) => upcastType(d1, d2));\n    const shapes = tensors.map(t => t.shape);\n    // We can make sure shapes are identical in op level.\n    const usePackedOp = env().getBool('WEBGL_PACK');\n    const program = usePackedOp ?\n        new AddNPackedProgram(tensors[0].shape, shapes) :\n        new AddNProgram(tensors[0].shape, shapes);\n    return this.compileAndRun<T>(program, tensors, dtype);\n  }\n\n  subtract(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.SUB);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = subImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.SUB, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.SUB, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    const usePackedOp = env().getBool('WEBGL_PACK_BINARY_OPERATIONS');\n    const program = usePackedOp ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.POW, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.POW, a.shape, b.shape);\n    const dtype = upcastType(a.dtype, b.dtype);\n    return this.compileAndRun<T>(program, [a, b], dtype);\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          ceilImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.CEIL, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.CEIL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  floor<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = floorImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.FLOOR, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.FLOOR);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_NAN);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_INF);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_FINITE);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n\n  round<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ROUND);\n    return this.compileAndRun(program, [x]);\n  }\n\n  exp<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          expImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXP, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXP);\n    return this.compileAndRun(program, [x]);\n  }\n\n  expm1<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = expm1ImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXPM1, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXPM1);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n    const a = this.subtract(logits, maxLogit.reshape(expandedShape));\n    const b = this.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax kernel\n    // modularization.\n    return div(b, sumExp);\n  }\n\n  log<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          logImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.LOG, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  log1p<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG1P);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sqrt<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  rsqrt<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = rsqrtImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.RSQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reciprocal<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.RECIPROCAL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU6);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU6);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  prelu<T extends Tensor>(x: T, alpha: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.PRELU, x.shape, alpha.shape) :\n        new BinaryOpProgram(binaryop_gpu.PRELU, x.shape, alpha.shape);\n    return this.compileAndRun(program, [x, alpha]);\n  }\n\n  elu<T extends Tensor>(x: T): T {\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.ELU, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.ELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.ELU_DER, dy.shape, y.shape) :\n        new BinaryOpProgram(binaryop_gpu.ELU_DER, dy.shape, y.shape);\n    return this.compileAndRun(program, [dy, y]);\n  }\n\n  selu<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  int<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TO_INT);\n    return this.compileAndRun(program, [x], 'int32');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    let program;\n    if (env().getBool('WEBGL_PACK_CLIP')) {\n      program = new ClipPackedProgram(x.shape);\n    } else {\n      program = new ClipProgram(x.shape);\n    }\n    const customSetup = program.getCustomSetupFunc(min, max);\n    return this.compileAndRun(program, [x], null, customSetup);\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    // TODO: handle cases when x is complex.\n    if (this.shouldExecuteOnCPU([x]) && x.dtype !== 'complex64') {\n      const outValues =\n          simpleAbsImplCPU(this.texData.get(x.dataId).values as TypedArray);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.ABS, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.ABS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  complexAbs<T extends Tensor>(x: T): T {\n    const xData = this.texData.get(x.dataId);\n\n    const program = new ComplexAbsProgram(x.shape);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    return this.compileAndRun<Tensor>(program, inputs) as T;\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGMOID);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SOFTPLUS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASIN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acos<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atan<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATAN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.COSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  tanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ERF);\n    return this.compileAndRun(program, [x]);\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.STEP(alpha));\n    return this.compileAndRun(program, [x]);\n  }\n\n  private conv2dByMatMul(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Reshapes conv2D input to 2D tensors, uses matMul and then reshape the\n    // result from 2D to 4D.\n    const xShape = x.shape;\n    const xTexData = this.texData.get(x.dataId);\n    const sharedMatMulDim = convInfo.inChannels;\n    const outerShapeX = xShape[0] * xShape[1] * xShape[2];\n    const outerShapeFilter = convInfo.outChannels;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const transposeA = false;\n    const transposeB = false;\n\n    // TODO: Once reduction ops are packed, batchMatMul will always be packed\n    // and we can remove this condition.\n    const batchMatMulWillBeUnpacked =\n        (outerShapeX === 1 || outerShapeFilter === 1) &&\n        sharedMatMulDim > MATMUL_SHARED_DIM_THRESHOLD;\n    const reshapeWillBeExpensive = xShape[2] % 2 !== 0 && !!xTexData.isPacked;\n\n    if (batchMatMulWillBeUnpacked || !env().getBool('WEBGL_LAZILY_UNPACK') ||\n        !env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ||\n        !reshapeWillBeExpensive) {\n      const targetShape = isChannelsLast ? xShape[0] * xShape[1] * xShape[2] :\n                                           xShape[0] * xShape[2] * xShape[3];\n      const xReshaped = reshape(x, [1, targetShape, convInfo.inChannels]);\n      const filterReshaped =\n          reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n      const result = this.fusedBatchMatMul({\n        a: xReshaped as Tensor3D,\n        b: filterReshaped as Tensor3D,\n        transposeA,\n        transposeB,\n        bias,\n        activation,\n        preluActivationWeights\n      });\n      return reshape(result, convInfo.outShape);\n    }\n\n    // Following optimization is specific to packed |x| with odd row count\n    // (For example, in channelLast mode, 'row count' refers to x.shape[2]):\n    // we avoid expensive packed 2x2 reshape by padding row count to next,\n    // even number. When x.shape[2] is odd, the result of packed batchMatMul is\n    // the same (has the same texture layout and and values in the texture) as\n    // it is for even x.shape[2] + 1. We make the odd-rows tensor to look like\n    // even-rows tensor before the operation and, after the batchMatMul,\n    // fix the even-rows result to have odd number of rows.\n    const targetShape = isChannelsLast ?\n        xShape[0] * xShape[1] * (xShape[2] + 1) :\n        xShape[0] * xShape[2] * (xShape[3] + 1);\n    const xReshaped: TensorInfo = {\n      dataId: x.dataId,\n      shape: [1, targetShape, convInfo.inChannels],\n      dtype: x.dtype\n    };\n    // xTexData.shape gets referenced from GPGPUBinary.inShapeInfos.\n    // Decrementing row count, after batchMatMul->...->compileProgram leads to\n    // invalid row count within the reference in GPGPUBinary.inShapeInfos.\n    // Alternative fix would be to provide a copy to GPGPUBinary.inShapeInfos\n    // in compileProgram method, but that would affect compilation of all\n    // programs - instead, provide a copy here, with even row count, before\n    // calling batchMatMul->...->compileProgram and after that, the original\n    // xTexData.shape is restored.\n    const originalXTexDataShape = xTexData.shape;\n    xTexData.shape = xTexData.shape.slice();\n    xTexData.shape[xTexData.shape.length - 2]++;\n    util.assert(\n        webgl_util.isReshapeFree(xTexData.shape, xReshaped.shape),\n        () => `packed reshape ${xTexData.shape} to ${\n            xReshaped.shape} isn't free`);\n    const filterReshaped =\n        reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n    const pointwiseConv = this.fusedBatchMatMul({\n      a: xReshaped as Tensor3D,\n      b: filterReshaped as Tensor3D,\n      transposeA,\n      transposeB,\n      bias,\n      activation,\n      preluActivationWeights\n    });\n    const pointwiseConvTexData = this.texData.get(pointwiseConv.dataId);\n    util.assert(\n        pointwiseConvTexData.isPacked,\n        () => 'batchMatMul result is expected to be packed');\n    // Restore the input shape to original.\n    xTexData.shape = originalXTexDataShape;\n    // Set the output shape - there is no need for expensive reshape as data\n    // layout is already correct.\n    pointwiseConvTexData.shape = convInfo.outShape;\n    return engine().makeTensorFromDataId(\n               pointwiseConv.dataId, convInfo.outShape, pointwiseConv.dtype) as\n        Tensor4D;\n  }\n\n  private conv2dWithIm2Row(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Rearranges conv2d input so each block to be convolved over forms the\n    // column of a new matrix with shape [filterWidth * filterHeight *\n    // inChannels, outHeight * outWidth]. The filter is also rearranged so each\n    // output channel forms a row of a new matrix with shape [outChannels,\n    // filterWidth * filterHeight * inChannels]. The convolution is then\n    // computed by multiplying these matrices and reshaping the result.\n    const {\n      filterWidth,\n      filterHeight,\n      inChannels,\n      outWidth,\n      outHeight,\n      dataFormat\n    } = convInfo;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n\n    const sharedDim = filterWidth * filterHeight * inChannels;\n    const numCols = outHeight * outWidth;\n    const x2ColShape = [sharedDim, numCols];\n    const transposeA = true;\n    const transposeB = false;\n\n    const xSqueezed = x.squeeze([0]);\n    const w2Row = filter.reshape([1, sharedDim, -1]);\n\n    const im2ColProgram =\n        new Im2ColPackedProgram(x2ColShape, xSqueezed.shape, convInfo);\n    const im2Col: Tensor3D =\n        this.compileAndRun<Tensor2D>(im2ColProgram, [xSqueezed]).reshape([\n          1, x2ColShape[0], x2ColShape[1]\n        ]);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const matmulProgram = new MatMulPackedProgram(\n        im2Col.shape, [1, numCols, convInfo.outChannels], transposeA,\n        transposeB, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [im2Col, w2Row];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    const product = this.compileAndRun<Tensor4D>(matmulProgram, inputs);\n\n    if (isChannelsLast) {\n      return product.reshape([1, outHeight, outWidth, convInfo.outChannels]);\n    } else {\n      return product.reshape([1, convInfo.outChannels, outHeight, outWidth]);\n    }\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && input.shape[0] === 1) {\n      return this.conv2dWithIm2Row(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, false) : null;\n    const program = new Conv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [input, filter];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun(program, inputs);\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(x, filter, convInfo);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && x.shape[0] === 1) {\n      return this.conv2dWithIm2Row(x, filter, convInfo);\n    }\n    const program = new Conv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new Conv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    const program = new Conv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    const shouldPackDepthwiseConv = env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1;\n    const fusedActivation = activation ?\n        mapActivationToShaderProgram(activation, shouldPackDepthwiseConv) :\n        null;\n    const inputs: Tensor[] = [input, filter];\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    if (hasBias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (shouldPackDepthwiseConv) {\n      program = new DepthwiseConvPacked2DProgram(\n          convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n      return this.compileAndRun(program, inputs);\n    }\n\n    program = new DepthwiseConv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    return this.compileAndRun(program, inputs);\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1) {\n      program = new DepthwiseConvPacked2DProgram(convInfo);\n      return this.compileAndRun(program, [x, filter]);\n    }\n\n    program = new DepthwiseConv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Conv3DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return backend_util.castTensor(x, dtype, this);\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = this.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'avg', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const avgPool3dBackpropProgram = new AvgPool3DBackpropProgram(convInfo);\n    return this.compileAndRun(avgPool3dBackpropProgram, [dy], x.dtype);\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'max', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const getPositions = true;\n    const maxPool3dPositionsProgram =\n        new Pool3DProgram(convInfo, 'max', getPositions);\n    const maxPool3dPositions: Tensor5D =\n        this.compileAndRun(maxPool3dPositionsProgram, [x]);\n    const maxPool3dBackPropProgram = new MaxPool3DBackpropProgram(convInfo);\n    const result = this.compileAndRun(\n        maxPool3dBackPropProgram, [dy, maxPool3dPositions], x.dtype);\n    maxPool3dPositions.dispose();\n    return result as Tensor5D;\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_IMAGE_OPERATIONS') ?\n        new ResizeBilinearPackedProgram(\n            x.shape, newHeight, newWidth, alignCorners) :\n        new ResizeBilinearProgram(x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    const program = new ResizeBilinearBackpropProgram(dy, x, alignCorners);\n\n    return this.compileAndRun(program, [dy]);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = new ResizeNearestNeighborProgram(\n        x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x]);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    const program =\n        new ResizeNearestNeigborBackpropProgram(dy, x, alignCorners);\n    return this.compileAndRun(program, [dy]);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    const probs = normalized ? logits : softmax(logits);\n    const batchSize = probs.shape[0];\n    const numOutcomes = probs.shape[1];\n    const program = new MultinomialProgram(batchSize, numOutcomes, numSamples);\n    const customSetup = program.getCustomSetupFunc(seed);\n    return this.compileAndRun(program, [probs], 'int32', customSetup);\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    const program = new OneHotProgram(indices.size, depth, onValue, offValue);\n    return this.compileAndRun(program, [indices]);\n  }\n\n  diag(x: Tensor): Tensor {\n    const program = new DiagProgram(x.size);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    const program = new CropAndResizeProgram(\n        image.shape, boxes.shape, cropSize, method, extrapolationValue);\n    return this.compileAndRun(program, [image, boxes, boxIndex], 'float32');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = (dataFormat === 'NHWC') ? x.shape[1] : x.shape[2];\n    const inputWidth = (dataFormat === 'NHWC') ? x.shape[2] : x.shape[3];\n    const inputDepth = (dataFormat === 'NHWC') ? x.shape[3] : x.shape[1];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const outputShape = (dataFormat === 'NHWC') ?\n        [batchSize, outputHeight, outputWidth, outputDepth] :\n        [batchSize, outputDepth, outputHeight, outputWidth];\n\n    const program = new DepthToSpaceProgram(outputShape, blockSize, dataFormat);\n    return this.compileAndRun(program, [x]);\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n    const flattenIndices = indices.reshape([numUpdates, sliceRank]);\n    const flattenX = updates.reshape([numUpdates, sliceSize]);\n\n    if (outputSize === 0) {\n      return backend_util.reshapeTensor(tensor([]), shape);\n    }\n    const defaultValue = scalar(0);\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, flattenIndices.rank, flattenX.rank, strides,\n        flattenShape);\n    const res: Tensor =\n        this.compileAndRun(program, [flattenX, flattenIndices, defaultValue]);\n    return res.reshape(shape);\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n\n    const sumDupeIndices = false;\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, sparseIndices.rank, sparseValues.rank, strides,\n        [outputSize, 1], sumDupeIndices);\n    const res: Tensor = this.compileAndRun(\n        program, [sparseValues, sparseIndices, defaultValue]);\n    return res.reshape(outputShape);\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    const inverse = false;\n    return this.fftImpl(x, inverse);\n  }\n\n  ifft(x: Tensor2D): Tensor2D {\n    const inverse = true;\n    return this.fftImpl(x, inverse);\n  }\n\n  private fftImpl(x: Tensor2D, inverse: boolean): Tensor2D {\n    const xData = this.texData.get(x.dataId);\n\n    const realProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.REAL, x.shape, inverse);\n    const imagProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.IMAG, x.shape, inverse);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    const real = this.compileAndRun<Tensor>(realProgram, inputs);\n    const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n    const complex = this.complex(real, imag).as2D(x.shape[0], x.shape[1]);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n\n    const flattenIndices = indices.reshape([numSlices, sliceRank]);\n    const flattenX = x.reshape([x.size / sliceSize, sliceSize]);\n    const program =\n        new GatherNDProgram(sliceRank, strides, [numSlices, sliceSize]);\n    const res: Tensor = this.compileAndRun(program, [flattenX, flattenIndices]);\n    return res.reshape(resultShape);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n\n    if (dtype === 'string') {\n      // String type should be handled in CPU memory.\n      const values = util.getArrayFromDType(dtype, util.sizeFromShape(shape));\n      values.fill(value as string);\n      return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n    } else {\n      const program = new FillProgram(shape, value as number);\n      const customSetup = program.getCustomSetupFunc(value as number);\n      return this.compileAndRun(program, [], dtype, customSetup);\n    }\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported under string dtype');\n    } else {\n      // TODO(cais, smilkov): Add WebGL shader for onesLike:\n      //   https://github.com/tensorflow/tfjs/issues/1293\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return this.fill(x.shape, x.dtype === 'string' ? '' : 0, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    // TODO: Use CPU implementation due to the precision problem in Safari.\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  makeTensorInfo(shape: number[], dtype: DataType, values?: BackendValues):\n      TensorInfo {\n    const dataId = this.write(values, shape, dtype);\n    this.texData.get(dataId).usage = null;\n    return {dataId, shape, dtype};\n  }\n\n  private makeOutput<T extends Tensor>(\n      shape: number[], dtype: DataType, values?: BackendValues): T {\n    const {dataId} = this.makeTensorInfo(shape, dtype, values);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  private unpackTensor(input: TensorInfo): TensorInfo {\n    const program = new UnpackProgram(input.shape);\n    return this.runWebGLProgram(program, [input], input.dtype);\n  }\n\n  private packTensor(input: TensorInfo): TensorInfo {\n    const program = new PackProgram(input.shape);\n    const preventEagerUnpackingOutput = true;\n    return this.runWebGLProgram(\n        program, [input], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOutput);\n  }\n\n  private packedReshape(input: TensorInfo, afterShape: number[]): TensorInfo {\n    const input3DShape = [\n      webgl_util.getBatchDim(input.shape),\n      ...webgl_util.getRowsCols(input.shape)\n    ] as [number, number, number];\n    const input3D: TensorInfo = {\n      dtype: input.dtype,\n      shape: input3DShape,\n      dataId: input.dataId\n    };\n    const afterShapeAs3D = [\n      webgl_util.getBatchDim(afterShape), ...webgl_util.getRowsCols(afterShape)\n    ] as [number, number, number];\n\n    const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n    const preventEagerUnpackingOfOutput = true;\n    const output = this.runWebGLProgram(\n        program, [input3D], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOfOutput);\n    return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n  }\n\n  private decode(dataId: DataId): TensorInfo {\n    const texData = this.texData.get(dataId);\n    const {isPacked, shape, dtype} = texData;\n    const shapeAs3D =\n        webgl_util.getShapeAs3D(shape) as [number, number, number];\n    let program;\n    if (isPacked) {\n      program = new DecodeMatrixPackedProgram(shapeAs3D);\n    } else {\n      program = new DecodeMatrixProgram(shapeAs3D);\n    }\n    const preventEagerUnpackingOfOutput = true;\n    const out = this.runWebGLProgram(\n        program, [{shape: shapeAs3D, dtype, dataId}], dtype,\n        null /* customSetup */, preventEagerUnpackingOfOutput);\n    return {dtype, shape, dataId: out.dataId};\n  }\n\n  runWebGLProgram(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): TensorInfo {\n    const output = this.makeTensorInfo(program.outputShape, outputDtype);\n    const outData = this.texData.get(output.dataId);\n    if (program.packedOutput) {\n      outData.isPacked = true;\n    }\n    if (program.outPackingScheme === tex_util.PackingScheme.DENSE) {\n      const texelShape = tex_util.getDenseTexShape(program.outputShape);\n      // For a densely packed output, we explicitly set texShape\n      // so it doesn't get assigned later according to our typical packing\n      // scheme wherein a single texel can only contain values from adjacent\n      // rows/cols.\n      outData.texShape = texelShape.map(d => d * 2) as [number, number];\n    }\n    if (program.outTexUsage != null) {\n      outData.usage = program.outTexUsage;\n    }\n    if (util.sizeFromShape(output.shape) === 0) {\n      // Short-circuit the computation since the result is empty (has 0 in its\n      // shape).\n      outData.values =\n          util.getTypedArrayFromDType(output.dtype as 'float32', 0);\n      return output;\n    }\n\n    const dataToDispose: TensorInfo[] = [];\n    const inputsData: TensorData[] = inputs.map(input => {\n      if (input.dtype === 'complex64') {\n        throw new Error(\n            `GPGPUProgram does not support complex64 input. For complex64 ` +\n            `dtypes, please separate the program into real and imaginary ` +\n            `parts.`);\n      }\n\n      let texData = this.texData.get(input.dataId);\n\n      if (texData.texture == null) {\n        if (!program.packedInputs &&\n            util.sizeFromShape(input.shape) <=\n                env().getNumber('WEBGL_SIZE_UPLOAD_UNIFORM')) {\n          // Upload small tensors that live on the CPU as uniforms, not as\n          // textures. Do this only when the environment supports 32bit floats\n          // due to problems when comparing 16bit floats with 32bit floats.\n          // TODO(https://github.com/tensorflow/tfjs/issues/821): Make it\n          // possible for packed shaders to sample from uniforms.\n          return {\n            shape: input.shape,\n            texData: null,\n            isUniform: true,\n            uniformValues: texData.values as TypedArray\n          };\n        }\n\n        // This ensures that if a packed program's inputs have not yet been\n        // uploaded to the GPU, they get uploaded as packed right off the bat.\n        if (program.packedInputs) {\n          texData.isPacked = true;\n          texData.shape = input.shape;\n        }\n      } else if (!!texData.isPacked !== !!program.packedInputs) {\n        input = texData.isPacked ? this.unpackTensor(input) :\n                                   this.packTensor(input);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n      } else if (\n          texData.isPacked &&\n          !webgl_util.isReshapeFree(texData.shape, input.shape)) {\n        // This is a special case where a texture exists for a tensor\n        // but the shapes are incompatible (due to packing constraints) because\n        // the tensor did not have a chance to go through the packed reshape\n        // shader. This only happens when we reshape the *same* tensor to form\n        // *distinct* inputs to an op, e.g. dotting a vector with itself. This\n        // case will disappear once packed uploading is the default.\n\n        const savedInput = input;\n        const targetShape = input.shape;\n\n        input.shape = texData.shape;\n        input = this.packedReshape(input as Tensor, targetShape);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n\n        savedInput.shape = targetShape;\n      }\n\n      this.uploadToGPU(input.dataId);\n      return {shape: input.shape, texData, isUniform: false};\n    });\n\n    this.uploadToGPU(output.dataId);\n    const outputData:\n        TensorData = {shape: output.shape, texData: outData, isUniform: false};\n    const key = gpgpu_math.makeShaderKey(program, inputsData, outputData);\n    const binary = this.getAndSaveBinary(key, () => {\n      return gpgpu_math.compileProgram(\n          this.gpgpu, program, inputsData, outputData);\n    });\n    const shouldTimeProgram = this.activeTimers != null;\n    let query: WebGLQuery|CPUTimerQuery;\n    if (shouldTimeProgram) {\n      query = this.startTimer();\n    }\n\n    gpgpu_math.runProgram(\n        this.gpgpu, binary, inputsData, outputData, customSetup);\n\n    dataToDispose.forEach(info => this.disposeIntermediateTensorInfo(info));\n\n    if (shouldTimeProgram) {\n      query = this.endTimer(query);\n      this.activeTimers.push(\n          {name: program.constructor.name, query: this.getQueryTime(query)});\n    }\n\n    if (!env().getBool('WEBGL_LAZILY_UNPACK') && outData.isPacked &&\n        preventEagerUnpackingOfOutput === false) {\n      const unpacked = this.unpackTensor(output);\n      this.disposeIntermediateTensorInfo(output);\n      return unpacked;\n    }\n    return output;\n  }\n\n  compileAndRun<K extends TensorInfo>(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype?: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): K {\n    outputDtype = outputDtype || inputs[0].dtype;\n    const outInfo = this.runWebGLProgram(\n        program, inputs, outputDtype, customSetup,\n        preventEagerUnpackingOfOutput);\n    return engine().makeTensorFromDataId(\n               outInfo.dataId, outInfo.shape, outInfo.dtype) as {} as K;\n  }\n\n  private getAndSaveBinary(key: string, getBinary: () => GPGPUBinary):\n      GPGPUBinary {\n    if (!(key in this.binaryCache)) {\n      this.binaryCache[key] = getBinary();\n    }\n    return this.binaryCache[key];\n  }\n\n  getTextureManager(): TextureManager {\n    return this.textureManager;\n  }\n\n  private disposed = false;\n\n  dispose() {\n    if (this.disposed) {\n      return;\n    }\n    // Avoid disposing the compiled webgl programs during unit testing because\n    // it slows down test execution.\n    if (!env().getBool('IS_TEST')) {\n      const allKeys = Object.keys(this.binaryCache);\n      allKeys.forEach(key => {\n        this.gpgpu.deleteProgram(this.binaryCache[key].webGLProgram);\n        delete this.binaryCache[key];\n      });\n    }\n    this.textureManager.dispose();\n    if (this.canvas != null &&\n        (typeof (HTMLCanvasElement) !== 'undefined' &&\n         this.canvas instanceof HTMLCanvasElement)) {\n      this.canvas.remove();\n    } else {\n      this.canvas = null;\n    }\n    if (this.gpgpuCreatedLocally) {\n      this.gpgpu.program = null;\n      this.gpgpu.dispose();\n    }\n    this.disposed = true;\n  }\n\n  floatPrecision(): 16|32 {\n    if (this.floatPrecisionValue == null) {\n      this.floatPrecisionValue = tidy(() => {\n        if (!env().get('WEBGL_RENDER_FLOAT32_ENABLED')) {\n          // Momentarily switching DEBUG flag to false so we don't throw an\n          // error trying to upload a small value.\n          const debugFlag = env().getBool('DEBUG');\n          env().set('DEBUG', false);\n          const underflowCheckValue = this.abs(scalar(1e-8)).dataSync()[0];\n          env().set('DEBUG', debugFlag);\n\n          if (underflowCheckValue > 0) {\n            return 32;\n          }\n        }\n        return 16;\n      });\n    }\n    return this.floatPrecisionValue;\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  private uploadToGPU(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    const {shape, dtype, values, texture, usage, isPacked} = texData;\n\n    if (texture != null) {\n      // Array is already on GPU. No-op.\n      return;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let texShape = texData.texShape;\n    if (texShape == null) {\n      texShape = webgl_util.getTextureShapeFromLogicalShape(shape, isPacked);\n      texData.texShape = texShape;\n    }\n\n    if (values != null) {\n      const shapeAs3D = webgl_util.getShapeAs3D(shape);\n\n      let program;\n      let width = texShape[1], height = texShape[0];\n      const isByteArray = values instanceof Uint8Array;\n\n      if (isPacked) {\n        [width, height] = tex_util.getPackedMatrixTextureShapeWidthHeight(\n            texShape[0], texShape[1]);\n        program = new EncodeMatrixPackedProgram(\n            shapeAs3D, [height, width], isByteArray);\n      } else {\n        program =\n            new EncodeMatrixProgram(shapeAs3D, [height, width], isByteArray);\n      }\n\n      const tempDenseInputHandle = this.makeTensorInfo([height, width], dtype);\n      if (isByteArray) {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.PIXELS;\n      } else {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.UPLOAD;\n      }\n      this.gpgpu.uploadDenseMatrixToTexture(\n          this.getTexture(tempDenseInputHandle.dataId), width, height,\n          values as TypedArray);\n\n      // We want the output to remain packed regardless of the value of\n      // WEBGL_PACK.\n      const preventEagerUnpacking = true;\n      const encodedOutputTarget = this.runWebGLProgram(\n          program, [tempDenseInputHandle], dtype, null, preventEagerUnpacking);\n\n      // Have the original texture assume the identity of the encoded output.\n      const outputTexData = this.texData.get(encodedOutputTarget.dataId);\n      texData.texture = outputTexData.texture;\n      texData.texShape = outputTexData.texShape;\n      texData.isPacked = outputTexData.isPacked;\n      texData.usage = outputTexData.usage;\n\n      this.disposeIntermediateTensorInfo(tempDenseInputHandle);\n      this.texData.delete(encodedOutputTarget.dataId);\n\n      // Once uploaded, don't store the values on cpu.\n      texData.values = null;\n      if (shouldTimeProgram) {\n        this.uploadWaitMs += util.now() - start;\n      }\n    } else {\n      const newTexture = this.acquireTexture(texShape, usage, dtype, isPacked);\n      texData.texture = newTexture;\n    }\n  }\n\n  private convertAndCacheOnCPU(dataId: DataId, float32Values?: Float32Array):\n      TypedArray {\n    const texData = this.texData.get(dataId);\n    const {dtype} = texData;\n\n    this.releaseGPUData(dataId);\n\n    if (float32Values != null) {\n      texData.values = float32ToTypedArray(float32Values, dtype as 'float32');\n    }\n    return texData.values as TypedArray;\n  }\n\n  private acquireTexture(\n      texShape: [number, number], texType: TextureUsage, dtype: DataType,\n      isPacked: boolean): WebGLTexture {\n    this.numBytesInGPU += this.computeBytes(texShape, dtype);\n    if (!this.warnedAboutMemory &&\n        this.numBytesInGPU > this.numMBBeforeWarning * 1024 * 1024) {\n      const mb = (this.numBytesInGPU / 1024 / 1024).toFixed(2);\n      this.warnedAboutMemory = true;\n      console.warn(\n          `High memory usage in GPU: ${mb} MB, ` +\n          `most likely due to a memory leak`);\n    }\n    return this.textureManager.acquireTexture(texShape, texType, isPacked);\n  }\n\n  private computeBytes(shape: [number, number], dtype: DataType) {\n    return shape[0] * shape[1] * util.bytesPerElement(dtype);\n  }\n\n  private tryRunOnCpuOrThrow<T extends Tensor>(\n      inputs: TensorInfo[], fn: () => T): T|null {\n    if (this.shouldExecuteOnCPU(inputs)) {\n      try {\n        return fn();\n      } catch (e) {\n        if (env().getBool('IS_TEST')) {\n          throw new Error('CPU forwarding failed');\n        }\n      }\n    }\n    return null;\n  }\n}\n\nfunction float32ToTypedArray<D extends NumericDataType>(\n    a: Float32Array, dtype: D): tf.DataTypeMap[D] {\n  if (dtype === 'float32' || dtype === 'complex64') {\n    return a as tf.DataTypeMap[D];\n  } else if (dtype === 'int32' || dtype === 'bool') {\n    const result = (dtype === 'int32') ? new Int32Array(a.length) :\n                                         new Uint8Array(a.length);\n    for (let i = 0; i < result.length; ++i) {\n      result[i] = Math.round(a[i]);\n    }\n    return result as tf.DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport * as gpgpu_util from './gpgpu_util';\nimport * as webgl_util from './webgl_util';\n\nexport {MathBackendWebGL, WebGLMemoryInfo, WebGLTimingInfo} from './backend_webgl';\nexport {setWebGLContext} from './canvas_util';\nexport {GPGPUContext} from './gpgpu_context';\nexport {GPGPUProgram} from './gpgpu_math';\n// WebGL specific utils.\nexport {gpgpu_util, webgl_util};\n\n/**\n * Enforce use of half precision textures if available on the platform.\n *\n * @doc {heading: 'Environment', namespace: 'webgl'}\n */\nexport function forceHalfFloat(): void {\n  env().set('WEBGL_FORCE_F16_TEXTURES', true);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// base.ts is the webgl backend without auto kernel registration.\n\nimport {device_util, registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from './backend_webgl';\nexport {version as version_webgl} from './version';\n\nif (device_util.isBrowser()) {\n  registerBackend('webgl', () => new MathBackendWebGL(), 2 /* priority */);\n}\n\n// Export webgl utilities\nexport * from './webgl';\n\n// Export forceHalfFlost under webgl namespace for the union bundle.\nimport {forceHalfFloat} from './webgl';\nexport const webgl = {forceHalfFloat};\n", "import {BinaryInputs, DataType, env, KernelFunc, UnaryInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BinaryOpProgram} from '../binaryop_gpu';\nimport {BinaryOpPackedProgram} from '../binaryop_packed_gpu';\nimport {UnaryOpProgram} from '../unaryop_gpu';\n\nexport const CHECK_NAN_SNIPPET_UNARY = `if (isnan(x)) return x;`;\n\nexport const CHECK_NAN_SNIPPET_BINARY = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const CHECK_NAN_SNIPPET_BINARY_PACKED = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param opSnippets Op snippet to create `UnaryOpProgram`.\n */\nexport function unaryKernelFunc(opSnippet: string): KernelFunc {\n  return ({inputs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = new UnaryOpProgram(x.shape, opSnippet);\n    return webglBackend.runWebGLProgram(program, [x], x.dtype);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param opSnippet Op snippet to create `BinaryOpProgram`.\n * @param packedOpSnippet Op snippet to create `BinaryOpPackedProgram`.\n * @param checkOutOfBoundsForPackedProgram Whether to set checkOutOfBounds=true\n *     when creating BinaryOpPackedProgram.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    opSnippet: string, packedOpSnippet: string,\n    checkOutOfBoundsForPackedProgram?: boolean, dtype?: DataType): KernelFunc {\n  // TODO(jingjin): handle complex64.\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            packedOpSnippet, a.shape, b.shape,\n            !!checkOutOfBoundsForPackedProgram) :\n        new BinaryOpProgram(opSnippet, a.shape, b.shape);\n    const $dtype = dtype || a.dtype;\n    const output = webglBackend.runWebGLProgram(program, [a, b], $dtype);\n    return output;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc, CHECK_NAN_SNIPPET_BINARY, CHECK_NAN_SNIPPET_BINARY_PACKED} from '../kernel_utils/kernel_funcs_utils';\n\nconst ATAN2 = CHECK_NAN_SNIPPET_BINARY + `\n  return atan(a, b);\n`;\n\nconst ATAN2_PACKED = `\n  vec4 result = atan(a, b);\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET_BINARY_PACKED + `\n  return result;\n`;\n\nexport const atan2KernelFunc = binaryKernelFunc(ATAN2, ATAN2_PACKED);\n\nexport const atan2Config: KernelConfig = {\n  kernelName: Atan2,\n  backendName: 'webgl',\n  kernelFunc: atan2KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendWebGL}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'webgl',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function avgPool(args: {\n  inputs: AvgPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const avgPoolProgram = new Pool2DProgram(convInfo, 'avg', false);\n  return backend.runWebGLProgram(avgPoolProgram, [x], 'float32');\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'webgl',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {AvgPool2DBackpropProgram} from '../avg_pool_backprop_gpu';\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const avgPoolBackpropProgram = new AvgPool2DBackpropProgram(convInfo);\n  return backend.runWebGLProgram(avgPoolBackpropProgram, [dy], x.dtype);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = '0.0';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = '1.0';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        float x = getXAtOutCoords();\n        float mean = getMeanAtOutCoords();\n        float variance = getVarianceAtOutCoords();\n        float offset = ${offsetSnippet};\n        float scale = ${scaleSnippet};\n        float inv = scale * inversesqrt(variance + float(${varianceEpsilon}));\n        setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = 'vec4(0.0)';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = 'vec4(1.0)';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        vec4 offset = ${offsetSnippet};\n        vec4 scale = ${scaleSnippet};\n\n        vec4 x = getXAtOutCoords();\n        vec4 mean = getMeanAtOutCoords();\n        vec4 variance = getVarianceAtOutCoords();\n\n        vec4 inv = scale * inversesqrt(variance + vec4(${varianceEpsilon}));\n\n        setOutput((x - mean) * inv + offset);\n      }\n    `;\n  }\n}\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BatchNormProgram} from '../batchnorm_gpu';\nimport {BatchNormPackedProgram} from '../batchnorm_packed_gpu';\n\nexport const batchNormKernelFunc: (params: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendWebGL,\n  attrs: FusedBatchNormAttrs\n}) => TensorInfo = ({inputs, backend, attrs}) => {\n  const {x, mean, variance, offset, scale} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const finalInputs = [x, mean, variance];\n\n  let offsetShape = null;\n  if (offset != null) {\n    offsetShape = offset.shape;\n    finalInputs.push(offset);\n  }\n\n  let scaleShape = null;\n  if (scale != null) {\n    scaleShape = scale.shape;\n    finalInputs.push(scale);\n  }\n\n  const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n      new BatchNormPackedProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon) :\n      new BatchNormProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon);\n  const output =\n      backend.runWebGLProgram(program, finalInputs, finalInputs[0].dtype);\n\n  return output;\n};\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'webgl',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst COS = CHECK_NAN_SNIPPET_UNARY + `\n  return cos(x);\n`;\n\nexport const cosKernelFunc = unaryKernelFunc(COS);\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'webgl',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\n// Without the equality check div produces 0.9999 for a = b, which when\n// floored can cause errors.\nconst DIV = `\nif (a == b) {\n  return 1.0;\n};\nreturn a / b;`;\n\n// We do the same as in ./binaryop_gpu, with vec4 and ivec4.\n// On Linux, the vectorized implementation produces NaNs when a and b are 0.\nconst DIV_PACKED = `\n  // vec4 one = vec4(equal(a, b));\n  // return one + (vec4(1.0) - one) * a / b;\n  vec4 result = a / b;\n  if(a.x == b.x) {\n    result.x = 1.;\n  }\n  if(a.y == b.y) {\n    result.y = 1.;\n  }\n  if(a.z == b.z) {\n    result.z = 1.;\n  }\n  if(a.w == b.w) {\n    result.w = 1.;\n  }\n\n  return result;\n`;\n\nexport const divKernelFunc = binaryKernelFunc(\n    DIV, DIV_PACKED, true /* checkOutOfBoundsForPackedProgram */);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'webgl',\n  kernelFunc: divKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FlipLeftRightProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(imageShape: [number, number, number, number]) {\n    const imageWidth = imageShape[2];\n    this.outputShape = imageShape;\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n\n          int coordX = ${imageWidth} - x;\n          float outputValue;\n          if(coordX >= 0 && coordX < ${imageWidth}) {\n            outputValue = getImage(coords[0], coords[1], coordX, coords[3]);\n          } else {\n            outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {FlipLeftRightProgram} from '../flip_left_right_gpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new FlipLeftRightProgram((image as Tensor4D).shape);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n        vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${width}.0, ${height}.0);\n\n        vec4 values = ${glsl.texture2D}(A, uv);\n        float value;\n        if (depth == 0) {\n          value = values.r;\n        } else if (depth == 1) {\n          value = values.g;\n        } else if (depth == 2) {\n          value = values.b;\n        } else if (depth == 3) {\n          value = values.a;\n        }\n\n        setOutput(floor(value * 255.0 + 0.5));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n\n        vec4 result = vec4(0.);\n\n        for(int row=0; row<=1; row++) {\n          for(int col=0; col<=1; col++) {\n            texC = coords[1] + row;\n            depth = coords[2] + col;\n\n            vec2 uv = (vec2(texC, texR) + halfCR) /\n                       vec2(${width}.0, ${height}.0);\n            vec4 values = ${glsl.texture2D}(A, uv);\n            float value;\n            if (depth == 0) {\n              value = values.r;\n            } else if (depth == 1) {\n              value = values.g;\n            } else if (depth == 2) {\n              value = values.b;\n            } else if (depth == 3) {\n              value = values.a;\n            }\n\n            result[row * 2 + col] = floor(value * 255.0 + 0.5);\n          }\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {TextureUsage} from '../tex_util';\n\nimport {FromPixelsProgram} from './FromPixels_utils/from_pixels_gpu';\nimport {FromPixelsPackedProgram} from './FromPixels_utils/from_pixels_packed_gpu';\n\nexport const fromPixelsConfig: KernelConfig = {\n  kernelName: FromPixels,\n  backendName: 'webgl',\n  kernelFunc: fromPixels as {} as KernelFunc,\n};\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\nfunction fromPixels(args: {\n  inputs: FromPixelsInputs,\n  backend: MathBackendWebGL,\n  attrs: FromPixelsAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  let {pixels} = inputs;\n  const {numChannels} = attrs;\n\n  const isVideo = typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement;\n  const isImage = typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement;\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n\n  const texShape: [number, number] = [height, width];\n  const outShape = [height, width, numChannels];\n\n  if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement | HTMLImageElement, 0, 0, width, height);\n    pixels = fromPixels2DContext.canvas;\n  }\n\n  const tempPixelHandle = backend.makeTensorInfo(texShape, 'int32');\n  // This is a byte texture with pixels.\n  backend.texData.get(tempPixelHandle.dataId).usage = TextureUsage.PIXELS;\n  backend.gpgpu.uploadPixelDataToTexture(\n      backend.getTexture(tempPixelHandle.dataId), pixels as ImageData);\n  const program = env().getBool('WEBGL_PACK') ?\n      new FromPixelsPackedProgram(outShape) :\n      new FromPixelsProgram(outShape);\n  const res = backend.runWebGLProgram(program, [tempPixelHandle], 'int32');\n  backend.disposeData(tempPixelHandle.dataId);\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReduceProgram} from '../reduce_gpu';\n\ntype ReduceTypes = 'all'|'any'|'max'|'min'|'sum'|'prod';\n\n// Returns an array of configuration objects that describe each stage of the\n// reduction.\nfunction getReductionStages(inShape: number[]):\n    Array<{inSize: number, windowSize: number, outSize: number}> {\n  const stages = [];\n\n  while (stages.length === 0 || stages[stages.length - 1].outSize !== 1) {\n    const outSize: number =\n        stages.length ? stages[stages.length - 1].outSize : inShape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(outSize);\n    stages.push({\n      inSize: outSize,\n      windowSize,\n      outSize: Math.ceil(outSize / windowSize)\n    });\n  }\n\n  return stages;\n}\n\nexport function reduce(\n    x: TensorInfo, dtype: DataType, reductionType: ReduceTypes,\n    backend: MathBackendWebGL): TensorInfo {\n  const reductionStages = getReductionStages(x.shape);\n\n  let result = x;\n  for (let i = 0; i < reductionStages.length; i++) {\n    const {inSize, windowSize, outSize} = reductionStages[i];\n\n    const program = new ReduceProgram(\n        {windowSize, inSize, batchSize: x.shape[0], outSize}, reductionType);\n    const previousResult = result;\n    result = backend.runWebGLProgram(program, [result], dtype);\n\n    if (previousResult.dataId !== x.dataId) {\n      backend.disposeData(previousResult.dataId);\n    }\n  }\n\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReshapePackedProgram} from '../reshape_packed_gpu';\nimport {getBatchDim, getRowsCols} from '../webgl_util';\n\nexport function packedReshape(\n    input: TensorInfo, afterShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const input3DShape =\n      [getBatchDim(input.shape),\n       ...getRowsCols(input.shape)] as [number, number, number];\n  const input3D: TensorInfo = {\n    dtype: input.dtype,\n    shape: input3DShape,\n    dataId: input.dataId\n  };\n  const afterShapeAs3D =\n      [getBatchDim(afterShape),\n       ...getRowsCols(afterShape)] as [number, number, number];\n\n  const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n  const preventEagerUnpackingOfOutput = true;\n  const output = backend.runWebGLProgram(\n      program, [input3D], input.dtype, null /* customSetup */,\n      preventEagerUnpackingOfOutput);\n  return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {packedReshape} from '../kernel_utils/reshape';\nimport {isReshapeFree} from '../webgl_util';\n\nexport function reshape(args: {\n  inputs: ReshapeInputs,\n  backend: MathBackendWebGL,\n  attrs: ReshapeAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n  const webglBackend = backend;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  const xTexData = webglBackend.texData.get(x.dataId);\n  if (xTexData.isPacked && !isReshapeFree(x.shape, $shape) &&\n      !(xTexData.texture !== null && isReshapeFree(xTexData.shape, $shape))) {\n    return packedReshape(x, $shape, webglBackend);\n  }\n\n  webglBackend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'webgl',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {reduce} from '../kernel_utils/reduce';\nimport {reshape} from '../kernels/Reshape';\n\nexport function maxImpl(\n    x: TensorInfo, reduceShape: number[], outShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const inSize = util.sizeFromShape(reduceShape);\n  const xSize = util.sizeFromShape(x.shape);\n  const batchSize = xSize / inSize;\n  const reshapedInput =\n      reshape({inputs: {x}, attrs: {shape: [batchSize, inSize]}, backend});\n\n  const reduced = reduce(reshapedInput, x.dtype, 'max', backend);\n  const reshapedOutput =\n      reshape({inputs: {x: reduced}, attrs: {shape: outShape}, backend});\n\n  backend.disposeIntermediateTensorInfo(reshapedInput);\n  backend.disposeIntermediateTensorInfo(reduced);\n\n  return reshapedOutput;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposeProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const switched = getSwitchedCoords(newDim);\n\n    this.userCode = `\n    void main() {\n      ${dtype} resRC = getOutputCoords();\n      setOutput(getA(${switched}));\n    }\n    `;\n  }\n}\n\nfunction getSwitchedCoords(newDim: number[]): string {\n  const rank = newDim.length;\n  if (rank > 6) {\n    throw Error(`Transpose for rank ${rank} is not yet supported`);\n  }\n  const originalOrder =\n      ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u', 'resRC.v'];\n  const switchedCoords = new Array(rank);\n  for (let i = 0; i < newDim.length; i++) {\n    switchedCoords[newDim[i]] = originalOrder[i];\n  }\n  return switchedCoords.join();\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getVecChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    if (this.rank > 6) {\n      throw Error(\n          `Packed transpose for rank ${this.rank} is not yet supported.`);\n    }\n    const dtype = getCoordsDataType(this.rank);\n\n    const outputOrder = getVecChannels('rc', this.rank);\n    const switchedOrder = new Array(this.rank);\n    for (let i = 0; i < newDim.length; i++) {\n      switchedOrder[newDim[i]] = outputOrder[i];\n    }\n    const innerDims = `vec2(${switchedOrder.slice(-2).join()})`;\n    const nextColumn =\n        `++${outputOrder[this.rank - 1]} < ${outputShape[this.rank - 1]}`;\n    const getc = `getChannel(getA(${switchedOrder.join()}), ${innerDims})`;\n\n    this.userCode = `\n    void main() {\n      ${dtype} rc = getOutputCoords();\n      vec4 result = vec4(0.);\n      result[0] = ${getc};\n      if(${nextColumn}) {\n        result[1] = ${getc};\n      }\n      --${outputOrder[this.rank - 1]};\n      if(++${outputOrder[this.rank - 2]} < ${outputShape[this.rank - 2]}) {\n        result[2] = ${getc};\n        if(${nextColumn}) {\n          result[3] = ${getc};\n        }\n      }\n      setOutput(result);\n    }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {transposeImplCPU} from '../kernel_utils/shared';\nimport {TransposeProgram} from '../transpose_gpu';\nimport {TransposePackedProgram} from '../transpose_packed_gpu';\n\nexport function transposeImpl(\n    x: TensorInfo, perm: number[], backend: MathBackendWebGL): TensorInfo {\n  const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n      new TransposePackedProgram(x.shape, perm) :\n      new TransposeProgram(x.shape, perm);\n  return backend.runWebGLProgram(program, [x], x.dtype);\n}\n\nexport {transposeImplCPU};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {maxImplCPU} from '../kernel_utils/shared';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl, transposeImplCPU} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    const maxInputIsTransposed = permutedAxes != null;\n    const shouldExecuteOnCPU = webglBackend.shouldExecuteOnCPU([x]);\n\n    let maxInput = x;\n    if (maxInputIsTransposed) {\n      if (shouldExecuteOnCPU) {\n        const xTexData = webglBackend.texData.get(maxInput.dataId);\n        const values = xTexData.values as TypedArray;\n\n        const newShape: number[] = new Array(xRank);\n        for (let i = 0; i < newShape.length; i++) {\n          newShape[i] = x.shape[permutedAxes[i]];\n        }\n        const maxInputValues =\n            transposeImplCPU(values, x.shape, x.dtype, permutedAxes, newShape);\n\n        maxInput = webglBackend.makeTensorInfo(newShape, x.dtype);\n        const maxInputData = webglBackend.texData.get(maxInput.dataId);\n        maxInputData.values = maxInputValues;\n      } else {\n        maxInput = transposeImpl(x, permutedAxes, webglBackend);\n      }\n\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n    }\n\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(maxInput.shape, axes);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // rather than reshape at the end, set the target shape here.\n      outShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n    }\n\n    let out;\n    if (shouldExecuteOnCPU) {\n      const xTexData = webglBackend.texData.get(maxInput.dataId);\n      const values = xTexData.values as TypedArray;\n\n      const outValues = maxImplCPU(\n          values, util.sizeFromShape(reduceShape), outShape, x.dtype);\n\n      out = webglBackend.makeTensorInfo(outShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = maxImpl(maxInput, reduceShape, outShape, webglBackend);\n    }\n\n    if (maxInputIsTransposed) {\n      webglBackend.disposeIntermediateTensorInfo(maxInput);\n    }\n\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function maxPool(args: {\n  inputs: MaxPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const maxPoolProgram = new Pool2DProgram(convInfo, 'max', false);\n  return backend.runWebGLProgram(maxPoolProgram, [x], x.dtype);\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'webgl',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {MaxPool2DBackpropProgram} from '../max_pool_backprop_gpu';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const getPositions = true;\n  const maxPoolPositionsProgram =\n      new Pool2DProgram(convInfo, 'max', getPositions);\n  const maxPoolPositions: TensorInfo =\n      backend.runWebGLProgram(maxPoolPositionsProgram, [x], x.dtype);\n\n  const maxPoolBackPropProgram = new MaxPool2DBackpropProgram(convInfo);\n  const result = backend.runWebGLProgram(\n      maxPoolBackPropProgram, [dy, maxPoolPositions], x.dtype);\n  backend.disposeIntermediateTensorInfo(maxPoolPositions);\n  return result;\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\n\nexport function maxPoolWithArgmaxImpl(\n    x: TensorInfo, includeBatchInIndex: boolean,\n    convInfo: backend_util.Conv2DInfo,\n    backend: MathBackendWebGL): TensorInfo[] {\n  let program = new Pool2DProgram(convInfo, 'max', false);\n  const poolOutput = backend.runWebGLProgram(program, [x], 'float32');\n\n  program = new Pool2DProgram(convInfo, 'max', true, true, includeBatchInIndex);\n  const indexOutput = backend.runWebGLProgram(program, [x], 'float32');\n  return [poolOutput, indexOutput];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    util.assert(\n        x.shape.length === 4,\n        () => `Error in maxPool: input must be rank 4 but got rank ${\n            x.shape.length}.`);\n    const dilations: [number, number] = [1, 1];\n    util.assert(\n        backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n            `Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        dilations, pad);\n\n    const [result, indexes] =\n        maxPoolWithArgmaxImpl(x, includeBatchInIndex, convInfo, webglBackend);\n    return [result, indexes];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV3, NonMaxSuppressionV3Attrs, NonMaxSuppressionV3Inputs, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV3Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV3,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV3Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold} =\n        attrs as unknown as NonMaxSuppressionV3Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n\n    return kernel_impls.nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal);\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs, TypedArray} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class RotateProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], radians: number,\n      fillValue: number|[number, number, number],\n      center: number|[number, number]) {\n    const imageHeight = imageShape[1];\n    const imageWidth = imageShape[2];\n    const sinFactor = Math.sin(radians).toFixed(3);\n    const cosFactor = Math.cos(radians).toFixed(3);\n    this.outputShape = imageShape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const centerXString = centerX.toFixed(3);\n    const centerYString = centerY.toFixed(3);\n\n    let fillSnippet = '';\n    if (typeof fillValue === 'number') {\n      fillSnippet = `float outputValue = ${fillValue.toFixed(2)};`;\n    } else {\n      fillSnippet = `\n        vec3 fill = vec3(${fillValue.join(',')});\n        float outputValue = fill[coords[3]];`;\n    }\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n          int y = coords[1];\n          float coordXFloat = (float(x) - ${centerXString}) * ${\n        cosFactor} - (float(y) - ${centerYString}) * ${sinFactor};\n          float coordYFloat = (float(x) - ${centerXString}) * ${\n        sinFactor} + (float(y) - ${centerYString}) * ${cosFactor};\n          int coordX = int(round(coordXFloat + ${centerXString}));\n          int coordY = int(round(coordYFloat + ${centerYString}));\n          ${fillSnippet}\n          if(coordX >= 0 && coordX < ${imageWidth} && coordY >= 0 && coordY < ${\n        imageHeight}) {\n            outputValue = getImage(coords[0], coordY, coordX, coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {RotateProgram} from '../rotate_gpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new RotateProgram(\n        (image as Tensor4D).shape, radians, fillValue, center);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SIN = CHECK_NAN_SNIPPET_UNARY + `\n  return sin(x);\n`;\n\nexport const sinKernelFunc = unaryKernelFunc(SIN);\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'webgl',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Square} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARE = `return x * x;`;\n\nexport const squareKernelFunc = unaryKernelFunc(SQUARE);\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'webgl',\n  kernelFunc: squareKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const squaredDifferenceKernelFunc =\n    binaryKernelFunc(SQUARED_DIFFERENCE, SQUARED_DIFFERENCE);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'webgl',\n  kernelFunc: squaredDifferenceKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst TAN = `return tan(x);`;\n\nexport const tanKernelFunc = unaryKernelFunc(TAN);\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'webgl',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {transposeImpl} from './Transpose_impl';\nimport {transposeImplCPU as cpuTranspose} from './Transpose_impl';\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as TransposeInputs;\n    const {perm} = attrs as {} as TransposeAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const newShape: number[] = new Array(xRank);\n    for (let i = 0; i < newShape.length; i++) {\n      newShape[i] = x.shape[perm[i]];\n    }\n\n    let out: TensorInfo;\n    if (webglBackend.shouldExecuteOnCPU([x])) {\n      const xTexData = webglBackend.texData.get(x.dataId);\n      const values = xTexData.values as TypedArray;\n      const outValues = cpuTranspose(values, x.shape, x.dtype, perm, newShape);\n\n      out = webglBackend.makeTensorInfo(newShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = transposeImpl(x, perm, webglBackend);\n    }\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {uniqueImplCPU} from '../kernel_utils/shared';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function unique(\n    args:\n        {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendWebGL}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  // For now, always forward calculation to the CPU backend.\n  console.warn(\n      'WARNING: ',\n      'UI might be locked temporarily as data is being downloaded');\n  const values = backend.readSync(x.dataId);\n  const {outputValues, outputShape, indices} =\n      uniqueImplCPU(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'webgl',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {atan2Config} from './kernels/Atan2';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {cosConfig} from './kernels/Cos';\nimport {divConfig} from './kernels/Div';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {fromPixelsConfig} from './kernels/FromPixels';\nimport {identityConfig} from './kernels/Identity';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {nonMaxSuppressionV3Config} from './kernels/NonMaxSuppressionV3';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {sinConfig} from './kernels/Sin';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {tanConfig} from './kernels/Tan';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  atan2Config,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  cosConfig,\n  divConfig,\n  flipLeftRightConfig,\n  fromPixelsConfig,\n  identityConfig,\n  maxConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  nonMaxSuppressionV3Config,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  sinConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  tanConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport * from '@tensorflow/tfjs-core';\nexport * from '@tensorflow/tfjs-layers';\nexport * from '@tensorflow/tfjs-converter';\n\n// Export data api as tf.data\nimport * as data from '@tensorflow/tfjs-data';\nexport {data};\n\n// Import and register backends.\nimport '@tensorflow/tfjs-backend-cpu';\nimport '@tensorflow/tfjs-backend-webgl';\n\n// Import versions of all sub-packages.\nimport {version_core} from '@tensorflow/tfjs-core';\nimport {version_cpu} from '@tensorflow/tfjs-backend-cpu';\nimport {version_webgl} from '@tensorflow/tfjs-backend-webgl';\nimport {version_data} from '@tensorflow/tfjs-data';\nimport {version_layers} from '@tensorflow/tfjs-layers';\nimport {version_converter} from '@tensorflow/tfjs-converter';\nimport {version as version_union} from './version';\n\nexport const version = {\n  'tfjs-core': version_core,\n  'tfjs-backend-cpu': version_cpu,\n  'tfjs-backend-webgl': version_webgl,\n  'tfjs-data': version_data,\n  'tfjs-layers': version_layers,\n  'tfjs-converter': version_converter,\n  'tfjs': version_union\n};\n", "const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  if (frame > config.face.age.skipFrames) {\n    frame = 0;\n  } else {\n    frame += 1;\n  }\n  if (frame === 0) return last;\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;\n    if (confidence > config.face.gender.minConfidence) {\n      obj.gender = data[0] <= 0.5 ? 'female' : 'male';\n      obj.confidence = confidence;\n    }\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "/* eslint-disable indent */\n/* eslint-disable no-multi-spaces */\n\nexport default {\n  backend: 'webgl',          // select tfjs backend to use\n  console: true,             // enable debugging output to console\n  face: {\n    enabled: true,           // controls if specified modul is enabled\n                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion\n                             // (note: module is not loaded until it is required)\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.\n                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.\n      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'\n      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192,        // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3,    // empiric tuning\n      inputSize: 64,         // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'\n                                                         // which determines training set for model\n      inputSize: 64,         // fixed value\n      skipFrames: 10,        // how many frames to go without re-running the detector\n    },\n    gender: {\n      enabled: true,\n      minConfidence: 0.8,    // threshold for discarding a prediction\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64,         // fixed value\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      skipFrames: 10,        // how many frames to go without re-running the detector\n      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257,    // fixed value\n    outputStride: 16,        // fixed value\n    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256,          // fixed value\n    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n    minConfidence: 0.5,      // threshold for discarding a prediction\n    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace\n    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    // let savedWebglPackDepthwiseConvFlag;\n    // if (tf.getBackend() === 'webgl') {\n    //   savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n    //  tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    // }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          agConfidence: ssrData.confidence,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    // tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
+  "mappings": "ucAAA,gHAAmB,sBACnB,GAAiB,oBACjB,GAAgB,mBAChB,GAAkB,qBAClB,GAAiB,oBAKX,GAAW,WAAO,SAElB,GAAS,OAAO,UAChB,GAAO,OAAO,QAEpB,SACC,cACC,KAAK,IAAQ,GAEb,KAAM,GAAY,UAAU,GACtB,EAAU,UAAU,GAEpB,EAAU,GAChB,GAAI,GAAO,EAEX,GAAI,GACH,KAAM,GAAI,EACJ,EAAS,OAAO,EAAE,QACxB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAC3B,KAAM,GAAU,EAAE,GAClB,GAAI,GACJ,AAAI,YAAmB,QACtB,EAAS,EACH,AAAI,YAAY,OAAO,GAC7B,EAAS,OAAO,KAAK,EAAQ,OAAQ,EAAQ,WAAY,EAAQ,YAC3D,AAAI,YAAmB,aAC7B,EAAS,OAAO,KAAK,GACf,AAAI,YAAmB,IAC7B,EAAS,EAAQ,IAEjB,EAAS,OAAO,KAAK,MAAO,IAAY,SAAW,EAAU,OAAO,IAErE,GAAQ,EAAO,OACf,EAAQ,KAAK,IAIf,KAAK,IAAU,OAAO,OAAO,GAE7B,GAAI,GAAO,GAAW,EAAQ,OAAS,QAAa,OAAO,EAAQ,MAAM,cACzE,AAAI,GAAQ,CAAC,mBAAmB,KAAK,IACpC,MAAK,IAAQ,MAGX,QACH,MAAO,MAAK,IAAQ,UAEjB,QACH,MAAO,MAAK,IAEb,OACC,MAAO,SAAQ,QAAQ,KAAK,IAAQ,YAErC,cACC,KAAM,GAAM,KAAK,IACX,EAAK,EAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YACjE,MAAO,SAAQ,QAAQ,GAExB,SACC,KAAM,GAAW,GAAI,IACrB,SAAS,MAAQ,aACjB,EAAS,KAAK,KAAK,KACnB,EAAS,KAAK,MACP,EAER,WACC,MAAO,gBAER,QACC,KAAM,GAAO,KAAK,KAEZ,EAAQ,UAAU,GAClB,EAAM,UAAU,GACtB,GAAI,GAAe,EACnB,AAAI,IAAU,OACb,EAAgB,EACV,AAAI,EAAQ,EAClB,EAAgB,KAAK,IAAI,EAAO,EAAO,GAEvC,EAAgB,KAAK,IAAI,EAAO,GAEjC,AAAI,IAAQ,OACX,EAAc,EACR,AAAI,EAAM,EAChB,EAAc,KAAK,IAAI,EAAO,EAAK,GAEnC,EAAc,KAAK,IAAI,EAAK,GAE7B,KAAM,GAAO,KAAK,IAAI,EAAc,EAAe,GAE7C,EAAS,KAAK,IACd,EAAe,EAAO,MAAM,EAAe,EAAgB,GAC3D,EAAO,GAAI,IAAK,GAAI,CAAE,KAAM,UAAU,KAC5C,SAAK,IAAU,EACR,GAIT,OAAO,iBAAiB,GAAK,UAAW,CACvC,KAAM,CAAE,WAAY,IACpB,KAAM,CAAE,WAAY,IACpB,MAAO,CAAE,WAAY,MAGtB,OAAO,eAAe,GAAK,UAAW,OAAO,YAAa,CACzD,MAAO,OACP,SAAU,GACV,WAAY,GACZ,aAAc,KAiBf,YAAoB,EAAS,EAAM,GACjC,MAAM,KAAK,KAAM,GAEjB,KAAK,QAAU,EACf,KAAK,KAAO,EAGZ,AAAI,GACF,MAAK,KAAO,KAAK,MAAQ,EAAY,MAIvC,MAAM,kBAAkB,KAAM,KAAK,aAGrC,GAAW,UAAY,OAAO,OAAO,MAAM,WAC3C,GAAW,UAAU,YAAc,GACnC,GAAW,UAAU,KAAO,aAE5B,GAAI,IACJ,IACC,GAAU,oBAAoB,cACtB,IAET,KAAM,IAAY,OAAO,kBAGnB,GAAc,WAAO,YAW3B,YAAc,GACb,GAAI,GAAQ,KAER,EAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,GAC3E,EAAY,EAAK,KAErB,GAAI,GAAO,IAAc,OAAY,EAAI,EACzC,GAAI,GAAe,EAAK,QACxB,GAAI,GAAU,IAAiB,OAAY,EAAI,EAE/C,AAAI,GAAQ,KAEX,EAAO,KACD,AAAI,GAAkB,GAE5B,EAAO,OAAO,KAAK,EAAK,YAClB,AAAI,GAAO,IAAc,CAAI,OAAO,SAAS,IAAc,CAAI,OAAO,UAAU,SAAS,KAAK,KAAU,uBAE9G,EAAO,OAAO,KAAK,GACb,AAAI,YAAY,OAAO,GAE7B,EAAO,OAAO,KAAK,EAAK,OAAQ,EAAK,WAAY,EAAK,YAChD,AAAI,YAAgB,aAG1B,GAAO,OAAO,KAAK,OAAO,OAE3B,KAAK,IAAa,CACjB,OACA,UAAW,GACX,MAAO,MAER,KAAK,KAAO,EACZ,KAAK,QAAU,EAEf,AAAI,YAAgB,aACnB,EAAK,GAAG,QAAS,SAAU,GAC1B,KAAM,GAAQ,EAAI,OAAS,aAAe,EAAM,GAAI,IAAW,+CAA+C,EAAM,QAAQ,EAAI,UAAW,SAAU,GACrJ,EAAM,IAAW,MAAQ,IAK5B,GAAK,UAAY,IACZ,QACH,MAAO,MAAK,IAAW,SAGpB,YACH,MAAO,MAAK,IAAW,WAQxB,cACC,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,eAS/D,OACC,GAAI,GAAK,KAAK,SAAW,KAAK,QAAQ,IAAI,iBAAmB,GAC7D,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,QAAO,OAEd,GAAI,IAAK,GAAI,CACZ,KAAM,EAAG,gBACN,EACF,IAAS,OAUb,OACC,GAAI,GAAS,KAEb,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,IACC,MAAO,MAAK,MAAM,EAAO,kBACjB,GACR,MAAO,IAAK,QAAQ,OAAO,GAAI,IAAW,iCAAiC,EAAO,eAAe,EAAI,UAAW,qBAUnH,OACC,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,GAAO,cAShB,SACC,MAAO,IAAY,KAAK,OASzB,gBACC,GAAI,GAAS,KAEb,MAAO,IAAY,KAAK,MAAM,KAAK,SAAU,GAC5C,MAAO,IAAY,EAAQ,EAAO,aAMrC,OAAO,iBAAiB,GAAK,UAAW,CACvC,KAAM,CAAE,WAAY,IACpB,SAAU,CAAE,WAAY,IACxB,YAAa,CAAE,WAAY,IAC3B,KAAM,CAAE,WAAY,IACpB,KAAM,CAAE,WAAY,IACpB,KAAM,CAAE,WAAY,MAGrB,GAAK,MAAQ,SAAU,GACtB,SAAW,KAAQ,QAAO,oBAAoB,GAAK,WAElD,GAAI,CAAE,KAAQ,KACb,KAAM,GAAO,OAAO,yBAAyB,GAAK,UAAW,GAC7D,OAAO,eAAe,EAAO,EAAM,KAYtC,cACC,GAAI,GAAS,KAEb,GAAI,KAAK,IAAW,UACnB,MAAO,IAAK,QAAQ,OAAO,GAAI,WAAU,0BAA0B,KAAK,QAKzE,GAFA,KAAK,IAAW,UAAY,GAExB,KAAK,IAAW,MACnB,MAAO,IAAK,QAAQ,OAAO,KAAK,IAAW,OAG5C,GAAI,GAAO,KAAK,KAGhB,GAAI,IAAS,KACZ,MAAO,IAAK,QAAQ,QAAQ,OAAO,MAAM,IAS1C,GALA,AAAI,GAAO,IACV,GAAO,EAAK,UAIT,OAAO,SAAS,GACnB,MAAO,IAAK,QAAQ,QAAQ,GAI7B,GAAI,CAAE,aAAgB,aACrB,MAAO,IAAK,QAAQ,QAAQ,OAAO,MAAM,IAK1C,GAAI,GAAQ,GACR,EAAa,EACb,EAAQ,GAEZ,MAAO,IAAI,IAAK,QAAQ,SAAU,EAAS,GAC1C,GAAI,GAGJ,AAAI,EAAO,SACV,GAAa,WAAW,WACvB,EAAQ,GACR,EAAO,GAAI,IAAW,0CAA0C,EAAO,aAAa,EAAO,aAAc,kBACvG,EAAO,UAIX,EAAK,GAAG,QAAS,SAAU,GAC1B,AAAI,EAAI,OAAS,aAEhB,GAAQ,GACR,EAAO,IAGP,EAAO,GAAI,IAAW,+CAA+C,EAAO,QAAQ,EAAI,UAAW,SAAU,MAI/G,EAAK,GAAG,OAAQ,SAAU,GACzB,GAAI,GAAS,IAAU,KACtB,OAGD,GAAI,EAAO,MAAQ,EAAa,EAAM,OAAS,EAAO,MACrD,EAAQ,GACR,EAAO,GAAI,IAAW,mBAAmB,EAAO,mBAAmB,EAAO,OAAQ,aAClF,OAGD,GAAc,EAAM,OACpB,EAAM,KAAK,KAGZ,EAAK,GAAG,MAAO,WACd,GAAI,EACH,OAGD,aAAa,GAEb,IACC,EAAQ,OAAO,OAAO,EAAO,UACrB,GAER,EAAO,GAAI,IAAW,kDAAkD,EAAO,QAAQ,EAAI,UAAW,SAAU,SAcpH,YAAqB,EAAQ,GAC5B,GAAI,MAAO,KAAY,WACtB,KAAM,IAAI,OAAM,gFAGjB,KAAM,GAAK,EAAQ,IAAI,gBACvB,GAAI,GAAU,QACV,EAAK,EAGT,MAAI,IACH,GAAM,mBAAmB,KAAK,IAI/B,EAAM,EAAO,MAAM,EAAG,MAAM,WAG5B,AAAI,CAAC,GAAO,GACX,GAAM,iCAAiC,KAAK,IAI7C,AAAI,CAAC,GAAO,GACX,GAAM,yEAAyE,KAAK,GACpF,AAAK,GACJ,GAAM,yEAAyE,KAAK,GACpF,AAAI,GACH,EAAI,OAIN,AAAI,GACH,GAAM,gBAAgB,KAAK,EAAI,SAKjC,AAAI,CAAC,GAAO,GACX,GAAM,mCAAmC,KAAK,IAI/C,AAAI,GACH,GAAU,EAAI,MAId,AAAI,KAAY,UAAY,IAAY,QACvC,GAAU,YAKL,GAAQ,EAAQ,QAAS,GAAS,WAU1C,YAA2B,GAE1B,MAAI,OAAO,IAAQ,UAAY,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,KAAQ,YAAc,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,KAAQ,YAAc,MAAO,GAAI,KAAQ,WACxN,GAID,EAAI,YAAY,OAAS,mBAAqB,OAAO,UAAU,SAAS,KAAK,KAAS,4BAA8B,MAAO,GAAI,MAAS,WAQhJ,YAAgB,GACf,MAAO,OAAO,IAAQ,UAAY,MAAO,GAAI,aAAgB,YAAc,MAAO,GAAI,MAAS,UAAY,MAAO,GAAI,QAAW,YAAc,MAAO,GAAI,aAAgB,YAAc,MAAO,GAAI,YAAY,MAAS,UAAY,gBAAgB,KAAK,EAAI,YAAY,OAAS,gBAAgB,KAAK,EAAI,OAAO,cASnT,YAAe,GACd,GAAI,GAAI,EACJ,EAAO,EAAS,KAGpB,GAAI,EAAS,SACZ,KAAM,IAAI,OAAM,sCAKjB,MAAI,aAAgB,aAAU,MAAO,GAAK,aAAgB,YAEzD,GAAK,GAAI,IACT,EAAK,GAAI,IACT,EAAK,KAAK,GACV,EAAK,KAAK,GAEV,EAAS,IAAW,KAAO,EAC3B,EAAO,GAGD,EAYR,YAA4B,GAC3B,MAAI,KAAS,KAEL,KACG,MAAO,IAAS,SAEnB,2BACG,GAAkB,GAErB,kDACG,GAAO,GAEV,EAAK,MAAQ,KACV,OAAO,SAAS,IAGhB,QAAO,UAAU,SAAS,KAAK,KAAU,wBAGzC,YAAY,OAAO,IAJtB,KAOG,MAAO,GAAK,aAAgB,WAE/B,gCAAgC,EAAK,gBAClC,YAAgB,YAGnB,KAGA,2BAaT,YAAuB,GACtB,KAAM,GAAO,EAAS,KAGtB,MAAI,KAAS,KAEL,EACG,GAAO,GACV,EAAK,KACF,OAAO,SAAS,GAEnB,EAAK,OACF,GAAQ,MAAO,GAAK,eAAkB,YAE5C,GAAK,mBAAqB,EAAK,kBAAkB,QAAU,GAC/D,EAAK,gBAAkB,EAAK,kBAEpB,EAAK,gBAKN,KAUT,YAAuB,EAAM,GAC5B,KAAM,GAAO,EAAS,KAGtB,AAAI,IAAS,KAEZ,EAAK,MACC,AAAI,GAAO,GACjB,EAAK,SAAS,KAAK,GACb,AAAI,OAAO,SAAS,GAE1B,GAAK,MAAM,GACX,EAAK,OAGL,EAAK,KAAK,GAKZ,GAAK,QAAU,OAAO,QAQtB,KAAM,IAAoB,gCACpB,GAAyB,0BAE/B,YAAsB,GAErB,GADA,EAAO,GAAG,IACN,GAAkB,KAAK,IAAS,IAAS,GAC5C,KAAM,IAAI,WAAU,GAAG,qCAIzB,YAAuB,GAEtB,GADA,EAAQ,GAAG,IACP,GAAuB,KAAK,GAC/B,KAAM,IAAI,WAAU,GAAG,sCAYzB,YAAc,EAAK,GAClB,EAAO,EAAK,cACZ,SAAW,KAAO,GACjB,GAAI,EAAI,gBAAkB,EACzB,MAAO,GAGT,OAGD,KAAM,IAAM,OAAO,OACnB,SAOC,cACC,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,OAI/E,GAFA,KAAK,IAAO,OAAO,OAAO,MAEtB,YAAgB,KACnB,KAAM,GAAa,EAAK,MAClB,EAAc,OAAO,KAAK,GAEhC,SAAW,KAAc,GACxB,SAAW,KAAS,GAAW,GAC9B,KAAK,OAAO,EAAY,GAI1B,OAKD,GAAI,KAAQ,MAAa,GAAI,MAAO,IAAS,UAC5C,KAAM,GAAS,EAAK,OAAO,UAC3B,GAAI,GAAU,MACb,GAAI,MAAO,IAAW,WACrB,KAAM,IAAI,WAAU,iCAKrB,KAAM,GAAQ,GACd,SAAW,KAAQ,IAClB,GAAI,MAAO,IAAS,UAAY,MAAO,GAAK,OAAO,WAAc,WAChE,KAAM,IAAI,WAAU,qCAErB,EAAM,KAAK,MAAM,KAAK,IAGvB,SAAW,KAAQ,IAClB,GAAI,EAAK,SAAW,EACnB,KAAM,IAAI,WAAU,+CAErB,KAAK,OAAO,EAAK,GAAI,EAAK,SAI3B,UAAW,KAAO,QAAO,KAAK,IAC7B,KAAM,GAAQ,EAAK,GACnB,KAAK,OAAO,EAAK,QAInB,MAAM,IAAI,WAAU,0CAUtB,IAAI,GACH,EAAO,GAAG,IACV,GAAa,GACb,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,MAAI,KAAQ,OACJ,KAGD,KAAK,IAAK,GAAK,KAAK,MAU5B,QAAQ,GACP,GAAI,GAAU,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,OAE9E,EAAQ,GAAW,MACnB,EAAI,EACR,KAAO,EAAI,EAAM,SAChB,GAAI,GAAW,EAAM,GACrB,KAAM,GAAO,EAAS,GAChB,EAAQ,EAAS,GAEvB,EAAS,KAAK,EAAS,EAAO,EAAM,MACpC,EAAQ,GAAW,MACnB,KAWF,IAAI,EAAM,GACT,EAAO,GAAG,IACV,EAAQ,GAAG,IACX,GAAa,GACb,GAAc,GACd,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,KAAK,IAAK,IAAQ,OAAY,EAAM,GAAQ,CAAC,GAU9C,OAAO,EAAM,GACZ,EAAO,GAAG,IACV,EAAQ,GAAG,IACX,GAAa,GACb,GAAc,GACd,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,AAAI,IAAQ,OACX,KAAK,IAAK,GAAK,KAAK,GAEpB,KAAK,IAAK,GAAQ,CAAC,GAUrB,IAAI,GACH,SAAO,GAAG,IACV,GAAa,GACN,GAAK,KAAK,IAAM,KAAU,OASlC,OAAO,GACN,EAAO,GAAG,IACV,GAAa,GACb,KAAM,GAAM,GAAK,KAAK,IAAM,GAC5B,AAAI,IAAQ,QACX,MAAO,MAAK,IAAK,GASnB,MACC,MAAO,MAAK,IAQb,OACC,MAAO,IAAsB,KAAM,OAQpC,SACC,MAAO,IAAsB,KAAM,UAUnC,OAAO,YACP,MAAO,IAAsB,KAAM,cAGrC,GAAQ,UAAU,QAAU,GAAQ,UAAU,OAAO,UAErD,OAAO,eAAe,GAAQ,UAAW,OAAO,YAAa,CAC5D,MAAO,UACP,SAAU,GACV,WAAY,GACZ,aAAc,KAGf,OAAO,iBAAiB,GAAQ,UAAW,CAC1C,IAAK,CAAE,WAAY,IACnB,QAAS,CAAE,WAAY,IACvB,IAAK,CAAE,WAAY,IACnB,OAAQ,CAAE,WAAY,IACtB,IAAK,CAAE,WAAY,IACnB,OAAQ,CAAE,WAAY,IACtB,KAAM,CAAE,WAAY,IACpB,OAAQ,CAAE,WAAY,IACtB,QAAS,CAAE,WAAY,MAGxB,YAAoB,GACnB,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,YAE/E,KAAM,GAAO,OAAO,KAAK,EAAQ,KAAM,OACvC,MAAO,GAAK,IAAI,IAAS,MAAQ,SAAU,GAC1C,MAAO,GAAE,eACN,IAAS,QAAU,SAAU,GAChC,MAAO,GAAQ,IAAK,GAAG,KAAK,OACzB,SAAU,GACb,MAAO,CAAC,EAAE,cAAe,EAAQ,IAAK,GAAG,KAAK,SAIhD,KAAM,IAAW,OAAO,YAExB,YAA+B,EAAQ,GACtC,KAAM,GAAW,OAAO,OAAO,IAC/B,SAAS,IAAY,CACpB,SACA,OACA,MAAO,GAED,EAGR,KAAM,IAA2B,OAAO,eAAe,CACtD,OAEC,GAAI,CAAC,MAAQ,OAAO,eAAe,QAAU,GAC5C,KAAM,IAAI,WAAU,4CAGrB,GAAI,GAAY,KAAK,IACrB,KAAM,GAAS,EAAU,OACnB,EAAO,EAAU,KACjB,EAAQ,EAAU,MAElB,EAAS,GAAW,EAAQ,GAC5B,EAAM,EAAO,OACnB,MAAI,IAAS,EACL,CACN,MAAO,OACP,KAAM,IAIR,MAAK,IAAU,MAAQ,EAAQ,EAExB,CACN,MAAO,EAAO,GACd,KAAM,OAGN,OAAO,eAAe,OAAO,eAAe,GAAG,OAAO,eAEzD,OAAO,eAAe,GAA0B,OAAO,YAAa,CACnE,MAAO,kBACP,SAAU,GACV,WAAY,GACZ,aAAc,KASf,YAAqC,GACpC,KAAM,GAAM,OAAO,OAAO,CAAE,UAAW,MAAQ,EAAQ,KAIjD,EAAgB,GAAK,EAAQ,IAAM,QACzC,MAAI,KAAkB,QACrB,GAAI,GAAiB,EAAI,GAAe,IAGlC,EAUR,YAA8B,GAC7B,KAAM,GAAU,GAAI,IACpB,SAAW,KAAQ,QAAO,KAAK,IAC9B,GAAI,GAAkB,KAAK,GAC1B,SAED,GAAI,MAAM,QAAQ,EAAI,IACrB,SAAW,KAAO,GAAI,IACrB,GAAI,GAAuB,KAAK,GAC/B,SAED,AAAI,EAAQ,IAAK,KAAU,OAC1B,EAAQ,IAAK,GAAQ,CAAC,GAEtB,EAAQ,IAAK,GAAM,KAAK,OAGpB,AAAK,IAAuB,KAAK,EAAI,KAC3C,GAAQ,IAAK,GAAQ,CAAC,EAAI,KAG5B,MAAO,GAGR,KAAM,IAAc,OAAO,sBAGrB,GAAe,WAAK,aAS1B,SACC,cACC,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,KAC3E,EAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,GAE/E,GAAK,KAAK,KAAM,EAAM,GAEtB,KAAM,GAAS,EAAK,QAAU,IACxB,EAAU,GAAI,IAAQ,EAAK,SAEjC,GAAI,GAAQ,MAAQ,CAAC,EAAQ,IAAI,iBAChC,KAAM,GAAc,GAAmB,GACvC,AAAI,GACH,EAAQ,OAAO,eAAgB,GAIjC,KAAK,IAAe,CACnB,IAAK,EAAK,IACV,SACA,WAAY,EAAK,YAAc,GAAa,GAC5C,UACA,QAAS,EAAK,YAIZ,OACH,MAAO,MAAK,IAAa,KAAO,MAG7B,UACH,MAAO,MAAK,IAAa,UAMtB,MACH,MAAO,MAAK,IAAa,QAAU,KAAO,KAAK,IAAa,OAAS,OAGlE,cACH,MAAO,MAAK,IAAa,QAAU,KAGhC,cACH,MAAO,MAAK,IAAa,cAGtB,WACH,MAAO,MAAK,IAAa,QAQ1B,QACC,MAAO,IAAI,IAAS,GAAM,MAAO,CAChC,IAAK,KAAK,IACV,OAAQ,KAAK,OACb,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,GAAI,KAAK,GACT,WAAY,KAAK,cAKpB,GAAK,MAAM,GAAS,WAEpB,OAAO,iBAAiB,GAAS,UAAW,CAC3C,IAAK,CAAE,WAAY,IACnB,OAAQ,CAAE,WAAY,IACtB,GAAI,CAAE,WAAY,IAClB,WAAY,CAAE,WAAY,IAC1B,WAAY,CAAE,WAAY,IAC1B,QAAS,CAAE,WAAY,IACvB,MAAO,CAAE,WAAY,MAGtB,OAAO,eAAe,GAAS,UAAW,OAAO,YAAa,CAC7D,MAAO,WACP,SAAU,GACV,WAAY,GACZ,aAAc,KAGf,KAAM,IAAc,OAAO,qBAGrB,GAAY,WAAI,MAChB,GAAa,WAAI,OAEjB,GAA6B,WAAa,YAAO,SAAS,UAQhE,YAAmB,GAClB,MAAO,OAAO,IAAU,UAAY,MAAO,GAAM,KAAiB,SAGnE,YAAuB,GACtB,KAAM,GAAQ,GAAU,MAAO,IAAW,UAAY,OAAO,eAAe,GAC5E,MAAO,CAAC,CAAE,IAAS,EAAM,YAAY,OAAS,eAU/C,SACC,YAAY,GACX,GAAI,GAAO,UAAU,OAAS,GAAK,UAAU,KAAO,OAAY,UAAU,GAAK,GAE3E,EAGJ,AAAK,GAAU,GAYd,EAAY,GAAU,EAAM,KAX5B,CAAI,GAAS,EAAM,KAIlB,EAAY,GAAU,EAAM,MAG5B,EAAY,GAAU,GAAG,KAE1B,EAAQ,IAKT,GAAI,GAAS,EAAK,QAAU,EAAM,QAAU,MAG5C,GAFA,EAAS,EAAO,cAEX,GAAK,MAAQ,MAAQ,GAAU,IAAU,EAAM,OAAS,OAAU,KAAW,OAAS,IAAW,QACrG,KAAM,IAAI,WAAU,iDAGrB,GAAI,GAAY,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAU,IAAU,EAAM,OAAS,KAAO,GAAM,GAAS,KAEzG,GAAK,KAAK,KAAM,EAAW,CAC1B,QAAS,EAAK,SAAW,EAAM,SAAW,EAC1C,KAAM,EAAK,MAAQ,EAAM,MAAQ,IAGlC,KAAM,GAAU,GAAI,IAAQ,EAAK,SAAW,EAAM,SAAW,IAE7D,GAAI,GAAa,MAAQ,CAAC,EAAQ,IAAI,iBACrC,KAAM,GAAc,GAAmB,GACvC,AAAI,GACH,EAAQ,OAAO,eAAgB,GAIjC,GAAI,GAAS,GAAU,GAAS,EAAM,OAAS,KAG/C,GAFA,AAAI,UAAY,IAAM,GAAS,EAAK,QAEhC,GAAU,MAAQ,CAAC,GAAc,GACpC,KAAM,IAAI,WAAU,mDAGrB,KAAK,IAAe,CACnB,SACA,SAAU,EAAK,UAAY,EAAM,UAAY,SAC7C,UACA,YACA,UAID,KAAK,OAAS,EAAK,SAAW,OAAY,EAAK,OAAS,EAAM,SAAW,OAAY,EAAM,OAAS,GACpG,KAAK,SAAW,EAAK,WAAa,OAAY,EAAK,SAAW,EAAM,WAAa,OAAY,EAAM,SAAW,GAC9G,KAAK,QAAU,EAAK,SAAW,EAAM,SAAW,EAChD,KAAK,MAAQ,EAAK,OAAS,EAAM,SAG9B,UACH,MAAO,MAAK,IAAa,UAGtB,OACH,MAAO,IAAW,KAAK,IAAa,cAGjC,WACH,MAAO,MAAK,IAAa,WAGtB,YACH,MAAO,MAAK,IAAa,YAGtB,UACH,MAAO,MAAK,IAAa,OAQ1B,QACC,MAAO,IAAI,IAAQ,OAIrB,GAAK,MAAM,GAAQ,WAEnB,OAAO,eAAe,GAAQ,UAAW,OAAO,YAAa,CAC5D,MAAO,UACP,SAAU,GACV,WAAY,GACZ,aAAc,KAGf,OAAO,iBAAiB,GAAQ,UAAW,CAC1C,OAAQ,CAAE,WAAY,IACtB,IAAK,CAAE,WAAY,IACnB,QAAS,CAAE,WAAY,IACvB,SAAU,CAAE,WAAY,IACxB,MAAO,CAAE,WAAY,IACrB,OAAQ,CAAE,WAAY,MASvB,YAA+B,GAC9B,KAAM,GAAY,EAAQ,IAAa,UACjC,EAAU,GAAI,IAAQ,EAAQ,IAAa,SAQjD,GALA,AAAK,EAAQ,IAAI,WAChB,EAAQ,IAAI,SAAU,OAInB,CAAC,EAAU,UAAY,CAAC,EAAU,SACrC,KAAM,IAAI,WAAU,oCAGrB,GAAI,CAAC,YAAY,KAAK,EAAU,UAC/B,KAAM,IAAI,WAAU,wCAGrB,GAAI,EAAQ,QAAU,EAAQ,eAAgB,YAAO,UAAY,CAAC,GACjE,KAAM,IAAI,OAAM,mFAIjB,GAAI,GAAqB,KAIzB,GAHA,AAAI,EAAQ,MAAQ,MAAQ,gBAAgB,KAAK,EAAQ,SACxD,GAAqB,KAElB,EAAQ,MAAQ,MACnB,KAAM,GAAa,GAAc,GACjC,AAAI,MAAO,IAAe,UACzB,GAAqB,OAAO,IAG9B,AAAI,GACH,EAAQ,IAAI,iBAAkB,GAI/B,AAAK,EAAQ,IAAI,eAChB,EAAQ,IAAI,aAAc,0DAI3B,AAAI,EAAQ,UAAY,CAAC,EAAQ,IAAI,oBACpC,EAAQ,IAAI,kBAAmB,gBAGhC,GAAI,GAAQ,EAAQ,MACpB,MAAI,OAAO,IAAU,YACpB,GAAQ,EAAM,IAGf,AAAI,CAAC,EAAQ,IAAI,eAAiB,CAAC,GAClC,EAAQ,IAAI,aAAc,SAMpB,OAAO,OAAO,GAAI,EAAW,CACnC,OAAQ,EAAQ,OAChB,QAAS,GAA4B,GACrC,UAgBF,YAAoB,GAClB,MAAM,KAAK,KAAM,GAEjB,KAAK,KAAO,UACZ,KAAK,QAAU,EAGf,MAAM,kBAAkB,KAAM,KAAK,aAGrC,GAAW,UAAY,OAAO,OAAO,MAAM,WAC3C,GAAW,UAAU,YAAc,GACnC,GAAW,UAAU,KAAO,aAG5B,KAAM,IAAgB,WAAO,YACvB,GAAc,WAAI,QASxB,YAAe,EAAK,GAGnB,GAAI,CAAC,GAAM,QACV,KAAM,IAAI,OAAM,0EAGjB,UAAK,QAAU,GAAM,QAGd,GAAI,IAAM,QAAQ,SAAU,EAAS,GAE3C,KAAM,GAAU,GAAI,IAAQ,EAAK,GAC3B,EAAU,GAAsB,GAEhC,EAAQ,GAAQ,WAAa,SAAW,WAAQ,YAAM,QACtD,EAAS,EAAQ,OAEvB,GAAI,GAAW,KAEf,KAAM,GAAQ,WACb,GAAI,GAAQ,GAAI,IAAW,+BAK3B,GAJA,EAAO,GACP,AAAI,EAAQ,MAAQ,EAAQ,eAAgB,YAAO,UAClD,EAAQ,KAAK,QAAQ,GAElB,CAAC,GAAY,CAAC,EAAS,KAAM,OACjC,EAAS,KAAK,KAAK,QAAS,IAG7B,GAAI,GAAU,EAAO,SACpB,IACA,OAGD,KAAM,GAAmB,WACxB,IACA,KAIK,EAAM,EAAK,GACjB,GAAI,GAEJ,AAAI,GACH,EAAO,iBAAiB,QAAS,GAGlC,aACC,EAAI,QACJ,AAAI,GAAQ,EAAO,oBAAoB,QAAS,GAChD,aAAa,GAGd,AAAI,EAAQ,SACX,EAAI,KAAK,SAAU,SAAU,GAC5B,EAAa,WAAW,WACvB,EAAO,GAAI,IAAW,uBAAuB,EAAQ,MAAO,oBAC5D,KACE,EAAQ,WAIb,EAAI,GAAG,QAAS,SAAU,GACzB,EAAO,GAAI,IAAW,cAAc,EAAQ,uBAAuB,EAAI,UAAW,SAAU,IAC5F,MAGD,EAAI,GAAG,WAAY,SAAU,GAC5B,aAAa,GAEb,KAAM,GAAU,GAAqB,EAAI,SAGzC,GAAI,GAAM,WAAW,EAAI,aAExB,KAAM,GAAW,EAAQ,IAAI,YAGvB,EAAc,IAAa,KAAO,KAAO,GAAY,EAAQ,IAAK,GAGxE,OAAQ,EAAQ,cACV,QACJ,EAAO,GAAI,IAAW,0EAA0E,EAAQ,MAAO,gBAC/G,IACA,WACI,SAEJ,GAAI,IAAgB,KAEnB,IACC,EAAQ,IAAI,WAAY,SAChB,GAER,EAAO,GAGT,UACI,SAEJ,GAAI,IAAgB,KACnB,MAID,GAAI,EAAQ,SAAW,EAAQ,QAC9B,EAAO,GAAI,IAAW,gCAAgC,EAAQ,MAAO,iBACrE,IACA,OAKD,KAAM,GAAc,CACnB,QAAS,GAAI,IAAQ,EAAQ,SAC7B,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QAAU,EAC3B,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,OAAQ,EAAQ,OAChB,KAAM,EAAQ,KACd,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,MAIf,GAAI,EAAI,aAAe,KAAO,EAAQ,MAAQ,GAAc,KAAa,MACxE,EAAO,GAAI,IAAW,2DAA4D,yBAClF,IACA,OAID,AAAI,GAAI,aAAe,KAAQ,GAAI,aAAe,KAAO,EAAI,aAAe,MAAQ,EAAQ,SAAW,SACtG,GAAY,OAAS,MACrB,EAAY,KAAO,OACnB,EAAY,QAAQ,OAAO,mBAI5B,EAAQ,GAAM,GAAI,IAAQ,EAAa,KACvC,IACA,QAKH,EAAI,KAAK,MAAO,WACf,AAAI,GAAQ,EAAO,oBAAoB,QAAS,KAEjD,GAAI,GAAO,EAAI,KAAK,GAAI,KAExB,KAAM,GAAmB,CACxB,IAAK,EAAQ,IACb,OAAQ,EAAI,WACZ,WAAY,EAAI,cAChB,QAAS,EACT,KAAM,EAAQ,KACd,QAAS,EAAQ,QACjB,QAAS,EAAQ,SAIZ,EAAU,EAAQ,IAAI,oBAU5B,GAAI,CAAC,EAAQ,UAAY,EAAQ,SAAW,QAAU,IAAY,MAAQ,EAAI,aAAe,KAAO,EAAI,aAAe,KACtH,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,GACR,OAQD,KAAM,GAAc,CACnB,MAAO,WAAK,aACZ,YAAa,WAAK,cAInB,GAAI,GAAW,QAAU,GAAW,UACnC,EAAO,EAAK,KAAK,WAAK,aAAa,IACnC,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,GACR,OAID,GAAI,GAAW,WAAa,GAAW,aAGtC,KAAM,GAAM,EAAI,KAAK,GAAI,KACzB,EAAI,KAAK,OAAQ,SAAU,GAE1B,AAAK,GAAM,GAAK,MAAU,EACzB,EAAO,EAAK,KAAK,WAAK,iBAEtB,EAAO,EAAK,KAAK,WAAK,oBAEvB,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,KAET,OAID,GAAI,GAAW,MAAQ,MAAO,YAAK,wBAA2B,YAC7D,EAAO,EAAK,KAAK,WAAK,0BACtB,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,GACR,OAID,EAAW,GAAI,IAAS,EAAM,GAC9B,EAAQ,KAGT,GAAc,EAAK,KASrB,GAAM,WAAa,SAAU,GAC5B,MAAO,KAAS,KAAO,IAAS,KAAO,IAAS,KAAO,IAAS,KAAO,IAAS,KAIjF,GAAM,QAAU,OAAO,QAEvB,GAAO,IAAQ,6rDCpmDf,AAsBO,GAAM,IAAkB,KAClB,GAAkB,mBA0B7B,WAAoB,EAAgC,GAAhC,KAAA,QAAA,EAAgC,KAAA,UAAA,EAH5C,KAAA,KAAO,GAAI,SACX,KAAA,aAAe,EAIvB,SAAA,UAAA,IAAA,SAAI,GACF,MAAK,MAAK,KAAK,IAAI,IACjB,KAAK,UAAU,SAAS,KAAK,QAAS,GAEjC,KAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,IAAA,SAAI,EAAgB,GAClB,KAAK,eACL,KAAK,KAAK,IAAI,EAAQ,IAGxB,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,MAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,OAAA,SAAO,GACL,YAAK,eACE,KAAK,KAAK,OAAO,IAG1B,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,cAEhB,mBAqBA,cACE,SAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,WACE,MAAO,GAAkB,eAE3B,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,GAAkB,gBAE3B,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAC5C,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAE3D,MAAO,GAAkB,SAE3B,EAAA,UAAA,OAAA,WACE,MAAO,GAAkB,WAG3B,EAAA,UAAA,eAAA,WACE,MAAO,GAAkB,mBAG3B,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAG1D,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAEnD,MAAO,GAAkB,qBAG3B,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,MAAO,GAAkB,UAE3B,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,MAAO,GAAkB,iBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,GAAkB,WAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAG3B,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,uBAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAE3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAG3B,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,MAAO,GAAkB,iBAG3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,MAAO,GAAkB,WAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,MAAO,GAAkB,sBAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,GAAkB,WAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,MAAO,GAAkB,WAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAG3B,EAAA,UAAA,QAAA,SAA0B,GACxB,MAAO,GAAkB,YAG3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAG3B,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,yBAG3B,EAAA,UAAA,gBAAA,SAAgB,EAAiB,EAAkB,GAEjD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,wBAAA,SAAwB,EAAc,EAAkB,GAEtD,MAAO,GAAkB,4BAE3B,EAAA,UAAA,yBAAA,SAAyB,EAAa,EAAc,GAElD,MAAO,GAAkB,6BAE3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,EAAa,GAEtD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SAAkB,EAAc,EAAa,GAC3C,MAAO,GAAkB,sBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAAa,GAC1C,MAAO,GAAkB,sBAG3B,EAAA,UAAA,QAAA,SAA0C,EAAM,GAE9C,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,MAAO,GAAkB,QAG3B,EAAA,UAAA,UAAA,SAA4B,EAAM,GAChC,MAAO,GAAkB,cAG3B,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,MAAO,GAAkB,WAG3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GACpC,MAAO,GAAkB,cAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,mBAG3B,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,MAAO,GAAkB,2BAG3B,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,0BAG3B,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,MAAO,GAAkB,kCAG3B,EAAA,UAAA,UAAA,SACI,EAAa,EAAyB,EACtC,EAA4B,EAC5B,GACF,MAAO,GAAkB,cAG3B,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,MAAO,GAAkB,iCAG3B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EAAuB,EAC3D,EAAc,EAAe,GAC/B,MAAO,GAAkB,YAG3B,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,MAAO,GAAkB,WAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,MAAO,GAAkB,WAG3B,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,MAAO,GAAkB,sBAG3B,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAC3C,MAAO,GAAkB,iBAI3B,EAAA,UAAA,MAAA,SAAwB,EAAU,EAAsB,GACtD,MAAO,GAAkB,UAG3B,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,MAAO,GAAkB,SAG3B,EAAA,UAAA,SAAA,SAAyB,GACvB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,GAAkB,cAG3B,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAkB,aAG3B,EAAA,UAAA,QAAA,WACE,MAAO,GAAkB,YAE7B,KAEA,WAA2B,GACzB,KAAM,IAAI,OACN,IAAI,EAAU,4HCzpBpB,AAoBA,GAAM,IAA4B,0BA6BhC,WAAmB,GAAA,KAAA,OAAA,EATX,KAAA,MAAe,GACf,KAAA,aAAwD,GAExD,KAAA,SAAkB,GAOxB,KAAK,mBAGP,SAAA,UAAA,YAAA,SAAY,EAAsB,GAChC,AAAI,KAAK,UAAY,MACnB,QAAQ,KACJ,YAAY,KAAK,aAAY,0BAC7B,kCAAiC,EAAQ,MAE/C,KAAK,aAAe,EACpB,KAAK,SAAW,GAGlB,EAAA,UAAA,aAAA,SACI,EAAkB,EAClB,GAKF,GAJA,KAAK,aAAa,GAAY,CAAC,aAAY,EAAE,QAAO,GAIhD,KAAK,SAAS,IAAa,MAC7B,GAAM,GAAY,KAAK,SAAS,GAChC,QAAQ,KACJ,qCAAqC,EAAQ,KAAK,EAAS,KAC/D,KAAK,IAAI,EAAU,KAIjB,EAAA,UAAA,SAAN,SAAe,qGACb,MAAI,KAAY,MAAK,MACnB,CAAA,EAAO,KAAK,MAAM,IAGpB,GAAA,KAAK,MAAM,EAAA,EAAY,CAAA,EAAM,KAAK,aAAa,YAA/C,SAAA,GAAuB,EAAA,OACvB,CAAA,EAAO,KAAK,MAAM,UAGpB,EAAA,UAAA,IAAA,SAAI,GACF,GAAI,IAAY,MAAK,MACnB,MAAO,MAAK,MAAM,GAGpB,GAAM,GAAY,KAAK,aAAa,GACpC,GAAI,YAAqB,SACvB,KAAM,IAAI,OACN,QAAQ,EAAQ,sEAItB,YAAK,MAAM,GAAY,EAEhB,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,QAAA,SAAQ,GACN,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,OAGd,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,uCAGd,EAAA,UAAA,IAAA,SAAI,EAAkB,GACpB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,mBAAmB,EAAQ,mCAEjC,KAAK,MAAM,GAAY,EACvB,AAAI,KAAK,aAAa,GAAU,SAAW,MACzC,KAAK,aAAa,GAAU,QAAQ,IAIhC,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,yBAAyB,EAAQ,oCAEvC,MAAO,MAAK,aAAa,GAAU,gBAGrC,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,MAAQ,OAAO,OAAO,GAAI,IAGjC,EAAA,UAAA,MAAA,WACE,KAAK,MAAQ,GACb,KAAK,SAAW,GAChB,KAAK,oBAGC,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,GAAI,MAAO,MAAK,QAAW,aACvB,MAAO,MAAK,OAAO,UAAa,aAChC,MAAO,MAAK,OAAO,SAAS,QAAW,YACzC,OAGF,GAAM,GAAY,GAAe,KAAK,OAAO,SAAS,QACtD,GAAI,KAA6B,IAC/B,GAAM,GAAY,EAAU,IAA2B,MAAM,KAC7D,EAAU,QAAQ,SAAA,GACV,GAAA,GAAA,EAAA,MAAA,KAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,SAAS,GAAO,GAAW,EAAK,OAI7C,iBAE+B,GAC7B,GAAM,GAAS,GACf,SAAY,QAAQ,8BAA+B,SAAC,UAAG,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GACrD,UAAY,EAAQ,EAAE,GAAI,EAAE,IACrB,EAAE,KAAK,OAET,EAGT,YACI,EAAiC,EAAc,GACjD,EAAO,mBAAmB,IAAS,mBAAmB,GAAS,IAGjE,YAAoB,EAAkB,GAEpC,GADA,EAAQ,EAAM,cACV,IAAU,QAAU,IAAU,QAChC,MAAO,KAAU,OACZ,GAAI,GAAG,EAAE,IAAY,EAC1B,MAAO,CAAC,EAEV,KAAM,IAAI,OACN,oCAAoC,EAAK,aAAa,EAAQ,mBAYlE,MAAO,GAAA,IAGE,EAAA,IAAmB,iBACO,GACnC,EAAA,IAAM,EC9MR,AAqBA,GAAI,kBAGF,GAAI,IAAmB,MAErB,GAAI,GAAE,OACN,GAAI,MAAQ,SAAY,YACtB,EAAK,eACI,MAAQ,SAAY,YAC7B,EAAK,eACI,MAAQ,UAAa,YAC9B,EAAK,gBACI,MAAQ,OAAU,YAC3B,EAAK,SAEL,MAAM,IAAI,OAAM,kCAElB,GAAkB,EAEpB,MAAO,IAIT,cACE,GAAM,GAAK,KACX,MAAI,GAAG,YAAc,MACnB,GAAG,WAAa,GAAI,MAEf,EAAG,uBAUiB,EAAa,GACxC,GAAM,GAAY,KAClB,GAAI,EAAU,IAAI,GAChB,MAAO,GAAU,IAAI,GAErB,GAAM,GAAY,IAClB,SAAU,IAAI,EAAK,GACZ,EAAU,IAAI,MCzCZ,IAAM,MAGN,GAAO,OAGP,GAAQ,QAGR,GAAM,MAGN,GAAO,OAGP,GAAM,MAON,GAAM,MAON,GAAS,SAMT,GAAS,SAMT,GAAO,OAGP,GAAQ,QAGR,GAAO,OAGP,GAAQ,QAGR,GAAQ,QAGR,GAAU,UASV,GAAkB,kBAQlB,GAAY,YAWZ,GAAoB,oBAUpB,GAAc,cAOd,GAAiB,iBASjB,GAAc,cAOd,GAAO,OAMP,GAAO,OAGP,GAAc,cAOd,GAAU,UAGV,GAAS,SAMT,GAAS,SAUT,GAAuB,uBASvB,GAAsB,sBAUtB,GAAS,SAST,GAAyB,yBAQzB,GAAwB,wBAMxB,GAAM,MAGN,GAAO,OAGP,GAAS,SAQT,GAAgB,gBAShB,GAAe,eAOf,GAAwB,wBAWxB,GACT,sCAIS,GACT,qCAIS,GAAO,OAGP,GAAa,aAQb,GAA0B,0BAI1B,GAA2B,2BAI3B,GAAM,MAGN,GAAM,MAGN,GAAU,UAGV,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAO,OAOP,GAAgB,gBAGhB,GAAQ,QAGR,GAAW,WAGX,GAAiB,iBAOjB,GAAW,WAMX,GAAW,WAGX,GAAU,UAGV,GAAe,eAGf,GAAW,WAGX,GAAO,OAGP,GAAO,OAGP,GAAW,WAGX,GAAQ,QAGR,GAAQ,QAGR,GAAO,OAGP,GAAY,YAGZ,GAAW,WAMX,GAAM,MAGN,GAAQ,QAGR,GAAa,aAGb,GAAa,aAGb,GAAY,YAGZ,GAAa,aAMb,GAAM,MASN,GAAc,cASd,GAAM,MAON,GAAU,UAGV,GAAU,UASV,GAAkB,kBAUlB,GAAY,YAWZ,GAAoB,oBAWpB,GAAoB,oBASpB,GAAO,OAOP,GAAM,MAON,GAAU,UAGV,GAAM,MAGN,GAAW,WAGX,GAAS,SAGT,GAAW,WAGX,GAAsB,sBAStB,GAAsB,sBAUtB,GAAsB,sBAUtB,GAAW,WAGX,GAAS,SAQT,GAAQ,QAOR,GAAO,OAGP,GAAM,MAGN,GAAQ,QAGR,GAAO,OAOP,GAAQ,QAQR,GAAO,OAGP,GAAa,aAGb,GAAO,OAGP,GAAU,UAMV,GAAwB,wBAOxB,GAA4B,4BAI5B,GAAiB,iBAOjB,GAAqB,qBAGrB,GAAQ,QAGR,GAAU,UAMV,GAAQ,QAGR,GAAQ,QAGR,GAAY,YAMZ,GAAW,WAGX,GAAO,OAGP,GAAQ,QAMR,GAAM,MAGN,GAAO,OAGP,GAAO,OAGP,GAAU,UAGV,GAAW,WAGX,GAAO,OAGP,GAAM,MAON,GAAiB,iBAOjB,GAAS,SAOT,GAAU,UAMV,GAAoB,oBAGpB,GAAS,SAGT,GAAM,MAGN,GAAgB,gBAOhB,GAAe,eAaf,GAAM,MAGN,GAAO,OAGP,GAAO,OAMP,GAAO,OAOP,GAAY,YAMZ,GAAS,SAQT,GAAS,SAMT,GAAqB,qBAOrB,GAAY,YAMZ,GAAO,OAMP,GAAa,aASb,GAAmB,mBAQnB,GAAe,eAef,GAAc,cAgBd,GAAuB,uBC7zBpC,AAuBA,GAAM,IACF,GAAU,iBAAkB,WAAM,MAAA,IAAI,OACpC,GACF,GAAU,eAAgB,WAAM,MAAA,IAAI,mBAqEpC,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,MAAO,IAAe,IAAI,eAOA,GAC1B,MAAO,IAAa,IAAI,eAGW,GAInC,OAHM,GAAK,GAAe,UACpB,EAAyB,MAGvB,GAAA,GAAA,EAAA,OAAC,EAAA,EAAA,KAAM,EAAA,EAAA,MACb,GAAI,EACF,MAEK,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GACL,EAAA,EAAA,MAAA,KAAA,GACP,AAAI,IAAY,GACd,EAAO,KAAK,GAGhB,MAAO,eAcsB,GACtB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,YACb,EAAM,GAAQ,EAAY,GAChC,AAAI,GAAe,IAAI,IACrB,QAAQ,KACJ,eAAe,EAAU,iBACzB,KAAI,EAAW,4BAErB,GAAe,IAAI,EAAK,eAWO,GACxB,GAAA,GAAA,EAAA,WAEP,AAAI,GAAa,IAAI,IAGnB,CAAI,KAAM,QAAQ,UAChB,QAAQ,KAAK,gCAAgC,EAAU,MAG3D,GAAa,IAAI,EAAY,eAW3B,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,GAAI,CAAC,GAAe,IAAI,GACtB,KAAM,IAAI,OACN,eAAe,EAAU,iBACzB,KAAI,EAAW,wBAErB,GAAe,OAAO,eAIW,GACjC,GAAI,CAAC,GAAa,IAAI,GACpB,KAAM,IAAI,OACN,iBAAiB,EAAU,mCAEjC,GAAa,OAAO,eAUlB,EAA+B,GACjC,GAAM,GAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,GAAM,GACF,OAAO,OAAO,GAAI,EAAc,CAAC,YAAa,IAClD,GAAe,KAInB,YAAiB,EAAoB,GACnC,MAAU,GAAW,IAAI,ECrN3B,YAkCwB,GAMtB,OAJI,GAAU,EAAM,OAChB,EAAO,EACP,EAAQ,EAEL,EAAU,GAEf,EAAS,KAAK,SAAW,EAAW,EAEpC,IAEA,EAAO,EAAM,GACb,EAAM,GAAW,EAAM,GACvB,EAAM,GAAS,cAKG,EAAa,EAAW,GAC5C,MAAO,MAAK,IAAI,EAAK,KAAK,IAAI,EAAG,gBAGD,GAChC,MAAO,GAAM,IAAM,EAAI,EAAM,EAAM,cAGjB,GAElB,OADI,GAAM,EACD,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,GAAO,EAAI,GAEb,MAAO,eAUmB,EAAW,GACrC,GAAM,GAAI,KAAK,SACf,MAAQ,GAAI,EAAM,GAAI,GAAK,cAID,EAAe,GAEzC,OADI,GAAS,EACJ,EAAI,EAAG,EAAI,EAAE,OAAQ,KAC5B,GAAM,GAAO,OAAO,EAAE,IAAM,OAAO,EAAE,IACrC,GAAU,EAAO,EAEnB,MAAO,cAkBc,EAAe,GACpC,GAAI,CAAC,EACH,KAAM,IAAI,OAAM,MAAO,IAAQ,SAAW,EAAM,iBAKhD,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EACI,GAAY,EAAQ,GACpB,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlC,GAC5B,EACI,GAAK,KACL,WAAM,MAAA,8EAwBR,EAA0B,EAAkB,GAI9C,GAJ4B,AAAA,IAAA,QAAA,GAAA,IAAkB,AAAA,IAAA,QAAA,GAAA,IAC9C,AAAI,GAAU,MACZ,GAAS,IAEP,MAAM,QAAQ,IAAQ,GAAa,IAAQ,CAAC,EAC9C,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAAQ,EAAI,GAAI,EAAQ,OAG1B,GAAO,KAAK,GAEd,MAAO,eAcqB,GAC5B,GAAI,EAAM,SAAW,EAEnB,MAAO,GAGT,OADI,GAAO,EAAM,GACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,GAAQ,EAAM,GAEhB,MAAO,eAGqB,GAC5B,MAAO,GAAM,SAAW,cAGE,EAAgB,GAC1C,GAAI,IAAO,EACT,MAAO,GAET,GAAI,GAAM,MAAQ,GAAM,KACtB,MAAO,GAGT,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,EAAG,GACf,MAAO,GAGX,MAAO,eAGa,GACpB,MAAO,GAAI,IAAM,cAGE,GAEnB,GAAK,KAAa,MAAQ,KAExB,MAAQ,MAAa,KAAK,GAE5B,GAAI,IAAM,SACR,MAAO,GACF,GAAI,IAAM,UACf,MAAO,GAEP,GAAM,GAAM,KAAK,IAAI,EAAI,GACzB,MAAQ,GAAM,GAAM,GAAM,eAIM,GAClC,GAAM,GAAQ,KAAK,KAAK,KAAK,KAAK,IAClC,MAAO,CAAC,EAAO,KAAK,KAAK,EAAO,gBAeI,GAEpC,OADM,GAAkB,GAAI,aAAY,GAC/B,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAgB,GAAK,EAEvB,UAAQ,GACD,cAGgB,EAAW,GAClC,MAAI,IAAQ,EAAE,OACL,EAEF,EAAI,IAAI,OAAO,EAAO,EAAE,oBAI7B,EAAwB,EACxB,GADwB,MAAA,KAAA,QAAA,GAAA,SAAW,GAAoB,MAAA,KAElD,GAAI,SAAc,SAAC,EAAS,GACjC,GAAI,GAAW,EAET,EAAQ,WACZ,GAAI,KACF,IACA,OAGF,IAEA,GAAM,GAAc,EAAQ,GAE5B,GAAI,GAAc,MAAQ,GAAY,GACpC,IACA,OAEF,WAAW,EAAO,IAGpB,kBAcA,EAAiB,GAInB,OAHI,GAAY,EACZ,EAAc,GAET,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,GAAI,EAAM,IAAM,EACd,GAAa,EAAM,WACV,EAAM,KAAO,IACtB,GAAI,IAAgB,GAClB,KAAM,OACF,yCACA,oBAAmB,EAAW,YAAY,IAEhD,EAAc,UACL,EAAM,GAAK,EACpB,KAAM,OAAM,gCAAgC,EAAM,GAAE,WAAW,GAInE,GAAI,IAAgB,IAClB,GAAI,EAAO,GAAK,IAAS,EACvB,KAAM,OAAM,QAAQ,EAAI,qCAAqC,GAE/D,MAAO,GAGT,GAAI,IAAc,EAChB,KAAM,OACF,qCAAqC,EAAK,+BAGhD,GAAI,EAAO,IAAc,EACvB,KAAM,OACF,oDACA,QAAO,EAAI,MAAM,IAGvB,GAAM,GAAW,EAAM,QACvB,SAAS,GAAe,EAAO,EACxB,cAIL,EAAuB,GACzB,GAAM,GAAO,EAAM,OAGnB,SAAO,GAAQ,KAAO,EAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAK,GAAG,OAAO,GAGzD,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,CAAC,GAAQ,EAAK,IACrC,WACI,MAAA,+CAA+C,EAAI,KAAK,EAAI,SAC5D,aAAY,KAGpB,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,KACvB,WAAM,MAAA,iDACF,aAAY,KAGb,EAAK,IAAI,SAAA,GAAK,MAAA,GAAI,EAAI,EAAO,EAAI,gBAIb,EAAiB,GAS5C,OAPM,GAAqB,GACrB,EAAqB,GACrB,EAAe,GAAQ,MAAQ,MAAM,QAAQ,IAAS,EAAK,SAAW,EACtE,EAAQ,GAAQ,MAAQ,EAC1B,KACA,GAAe,EAAM,GAAO,OAC5B,EAAI,EACC,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,GAAQ,MACV,GAAI,EAAK,KAAO,GAAK,EAAM,KAAO,EAChC,KAAM,IAAI,OACN,sBAAsB,EAAC,mBAAmB,EAAM,GAAE,cAExD,AAAK,GAAK,IAAM,MAAQ,EAAK,GAAK,IAAM,EAAM,KAAO,GACnD,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAEhB,AAAI,EAAK,IAAM,GACb,IAGJ,AAAI,EAAM,KAAO,GACf,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAGlB,MAAO,CAAC,SAAQ,EAAE,SAAQ,eAIxB,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,SACnB,EAAS,GAAI,OAAgB,OAE7B,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAA+B,GACjC,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAC1B,KAAM,OAAM,oBAAoB,EAAK,4BAA4B,EAAG,kBAM7C,GAC3B,MAAO,KAAU,QAAU,IAAU,aAAe,IAAU,WAC1D,IAAU,SAAW,IAAU,qBAOL,EAAmB,GACjD,MAAI,KAAY,aAGZ,KAAY,WAAa,IAAY,aAGrC,IAAY,SAAW,IAAY,WAAa,IAAY,aALvD,GAQL,MAAY,QAAU,IAAY,oBAMX,GAC3B,MAAO,aAAa,eAAgB,YAAa,aAC7C,YAAa,wBAGa,GAC9B,GAAI,IAAU,WAAa,IAAU,QACnC,MAAO,GACF,GAAI,IAAU,YACnB,MAAO,GACF,GAAI,IAAU,OACnB,MAAO,GAEP,KAAM,IAAI,OAAM,iBAAiB,eAUA,GACnC,GAAI,GAAO,KACT,MAAO,GAET,GAAI,GAAQ,EACZ,SAAI,QAAQ,SAAA,GAAK,MAAA,IAAS,EAAE,SACrB,cAIgB,GACvB,MAAO,OAAO,IAAU,UAAY,YAAiB,oBAG7B,GACxB,MAAO,OAAO,IAAU,sBAGD,GACvB,MAAO,OAAO,IAAU,qBAGC,GACzB,MAAI,OAAM,QAAQ,GACT,GAAW,EAAO,IAEvB,YAAkB,cACb,UACE,YAAkB,aAAc,YAAkB,YACpD,QACE,GAAS,GACX,UACE,GAAS,GACX,SACE,GAAU,GACZ,OAEF,sBAGkB,GACzB,MAAO,CAAC,CAAE,IAAK,EAAE,aAAe,EAAE,MAAQ,EAAE,mBAGf,EAAc,GAC3C,OAAS,GAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,GAAI,EAAO,IAAM,EACf,MAAO,GAGX,MAAO,eAGsB,GAC7B,GAAM,GAAO,EAAM,OACnB,GAAI,EAAO,EACT,MAAO,GAKT,GAAM,GAAU,GAAI,OAAM,EAAO,GACjC,EAAQ,EAAO,GAAK,EAAM,EAAO,GACjC,OAAS,GAAI,EAAO,EAAG,GAAK,EAAG,EAAE,EAC/B,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAM,EAAI,GAE1C,MAAO,eAOL,EAAiB,GACnB,MAAI,KAAU,SACL,GAAa,GAGf,GAAa,CAAC,GAAQ,eAGF,EAAe,GAC1C,GAAI,IAAU,SACZ,KAAM,IAAI,OAAM,6CASlB,GAPA,AAAI,MAAM,QAAQ,IAChB,GAAI,GAAQ,IAGd,AAAI,KAAM,QAAQ,UAChB,GAAyB,EAAe,GAEtC,GAAmB,EAAG,GACxB,MAAO,GAET,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,QAEnB,OADM,GAAO,GAAI,YAAY,EAAe,QACnC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,AAAI,KAAK,MAAO,EAAe,MAAQ,GACrC,GAAK,GAAK,GAGd,MAAO,OAEP,MAAM,IAAI,OAAM,qBAAqB,GAIzC,YAA2B,EAAgB,EAAiB,GAC1D,GAAM,GAAM,GAAI,OAChB,GAAI,EAAM,SAAW,EAEnB,OADM,GAAI,EAAM,GACP,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,EAAE,EAAS,OAMtB,QAHM,GAAI,EAAM,GACV,EAAO,EAAM,MAAM,GACnB,EAAM,EAAK,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IACjC,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,GAAkB,EAAS,EAAI,EAAK,EAAM,GAGvD,MAAO,eAIqB,EAAiB,GAC7C,GAAI,EAAM,SAAW,EAEnB,MAAO,GAAE,GAEX,GAAM,GAAO,EAAM,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IAC5C,GAAI,IAAS,EAEX,MAAO,GAET,GAAI,IAAS,EAAE,OACb,KAAM,IAAI,OAAM,IAAI,EAAK,mCAAmC,EAAE,OAAM,KAGtE,MAAO,IAAkB,EAAG,EAAO,GAGrC,YAA4B,EAAe,GACzC,MAAQ,aAAa,eAAgB,IAAU,WAC1C,YAAa,aAAc,IAAU,SACrC,YAAa,aAAc,IAAU,mBAIxC,EAAc,GAEhB,OADM,GAAQ,GAAoB,EAAM,GAC/B,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAM,GAAK,EAEb,MAAO,eAIL,EAAc,GAChB,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,OACnB,MAAO,IAAI,YAAW,GAEtB,KAAM,IAAI,OAAM,qBAAqB,eAUrC,EAAiB,GACnB,GAAM,GAAO,EAAM,OAAO,SAAC,EAAM,GAAS,MAAA,GAAO,GAAM,GACvD,GAAI,GAAS,MAAQ,IAAU,UAC7B,MAAO,IAAc,EAAO,GAAI,cAAa,IACxC,GAAI,IAAU,QACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IACtC,GAAI,IAAU,OACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IAE3C,KAAM,IAAI,OAAM,qBAAqB,iBAgBvC,MAAO,MAAM,SAAS,kBAG2B,GACjD,EAAM,QAAQ,SAAA,GACZ,EACI,OAAO,UAAU,IAAY,GAAW,EACxC,WACI,MAAA,mEACA,WAAU,EAAK,sBAqBvB,EAAc,GAChB,MAAO,MAAM,SAAS,MAAM,EAAM,eAWP,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,SACtC,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAG,eAWL,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,SAC9C,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAO,eAYlC,EAAgB,EAAc,GAChC,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,EAAQ,GAAK,EAAK,GAE7B,MAAO,eAYL,EAAe,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,EAAQ,IACrC,GAAS,EAAK,GAAK,EAAQ,GAE7B,SAAK,EAAK,OAAS,GAAK,EACjB,22BC3xBT,AA+BA,GAAA,IAAA,WACE,WAAoB,EAAoC,GAApC,KAAA,aAAA,EAAoC,KAAA,OAAA,EACtD,AAAI,GAAU,MACZ,MAAK,OAAS,GAAI,KAItB,SAAA,UAAA,cAAA,SAAc,EAAoB,EAAwB,GAQxD,OANI,GACE,EAAsB,WAC1B,EAAU,KAEN,EAAQ,KAAK,aAAa,KAAK,cAE5B,GACP,GAAM,GAAS,EAAQ,GAGvB,EAAO,OAAO,KAAK,SAAA,GACjB,GAA0B,EAAY,EAAO,MAAO,MAL/C,EAAI,EAAG,EAAI,EAAQ,OAAQ,MAA3B,GAST,GAAM,GAAgB,CACpB,WAAU,EACV,QAAO,EACP,OAAM,EACN,OAAQ,EAAM,KAAK,SAAA,GAAU,MAAA,GAAO,WACpC,UAAW,EAAM,KACb,SAAA,GAAU,MAAA,GAAO,qBAAuB,KACpC,EAAO,sBACP,MAEV,MAAO,IAGT,EAAA,UAAA,iBAAA,SAAiB,GAAjB,GAAA,GAAA,KACS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,OAAQ,EAAA,EAAA,OAAQ,EAAA,EAAA,UAE5C,EAAQ,QAAQ,SAAA,GACd,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,IAAY,KAAK,SAAA,GACnD,EAAK,OAAO,iBACR,EAAY,EAAQ,EAAe,GAAI,EAAe,GAAI,EAC1D,EAAe,SAI3B,iBAGI,EAAsB,EAAU,GAClC,GAAI,IAAU,UAEZ,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAE1B,eAAQ,KAAK,SAAS,EAAG,sBAAsB,EAAU,KAClD,GAGX,MAAO,GAGT,GAAA,IAAA,WAAA,cACE,SAAA,UAAA,iBAAA,SACI,EAAc,EAAgB,EAC9B,EAAgC,EAChC,GACF,GAAM,GAAO,MAAO,IAAW,SAAW,GAAiB,EAAM,KAAM,GAC7B,EAAO,MAC3C,EAAa,GAAc,EAAM,IACjC,EAAO,EAAO,KACd,EAAO,EAAO,KACd,EAAQ,GAAc,EAAO,MAAM,WAAY,IACjD,EAAyB,GAE7B,OAAW,KAAQ,IACjB,GAAM,GAAQ,EAAO,GACrB,GAAI,GAAS,MAGX,GAAM,GAAa,EAAM,OAAS,EAAO,MACnC,EAAY,EAAW,OAC7B,GACO,EAAI,KAAK,EAAS,KAAK,GAAY,EAAI,EAAa,IAAE,KAIjE,QAAQ,IACJ,KAAK,EAAU,MAAO,EAAI,MAAO,EAAI,KAAK,EAAK,MAAO,EAAI,MACtD,EAAsB,MAAO,EACjC,mBAAoB,YAAa,aAAc,gBAC/C,eAAgB,qBAExB,KCjIA,YA4CI,EAAkB,EAAc,GAKlC,OAFM,GAA8C,GAC9C,EAA0C,GACvC,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,EAAa,EAAG,GAAG,IAAM,GAG3B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GACZ,EAAa,EAAK,OACxB,OAAW,KAAa,IAItB,OAHM,GAAQ,EAAW,GAErB,EAAgB,GACX,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAa,EAAM,KACrB,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAa,EAAO,IAAM,KACzD,EAAgB,GAChB,EAAW,EAAK,IAAM,GACtB,MAIJ,GAAI,EACF,OAMN,GAAM,GAAgD,GACtD,EAAe,EAAE,IAAM,GAGvB,OAFM,GAAwC,GAErC,EAAI,EAAK,OAAS,EAAG,GAAK,EAAG,IAKpC,OAJM,GAAO,EAAK,GACZ,EAAa,EAAK,OAGf,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,IACvC,GAAI,EAAe,EAAK,QAAQ,GAAG,KACjC,OAAW,KAAa,GACtB,EAAe,EAAW,GAAW,IAAM,GAC3C,EAAS,EAAK,IAAM,GAEtB,MAON,OADM,GAA2B,GACxB,EAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GAElB,GAAI,EAAW,EAAK,KAAO,EAAS,EAAK,KAEvC,GAAM,GAA8C,GACpD,OAAW,KAAa,GAAK,QAC3B,GAAM,GAAY,EAAK,OAAO,GAC9B,AAAI,EAAa,EAAU,KACzB,GAAa,GAAa,GAK9B,GAAM,GAAa,OAAO,OAAO,GAAI,GACrC,EAAW,OAAS,EACpB,EAAW,QAAU,EAAK,QAE1B,EAAa,KAAK,IAItB,MAAO,eAWL,EACA,EAA0B,EAC1B,GAEF,mBAAS,GACP,GAAM,GAAO,EAAa,GAEpB,EAAgB,GAYtB,GAXA,EAAK,QAAQ,QAAQ,SAAA,GACnB,GAAM,GAAa,EAA6B,EAAE,IAClD,AAAI,GAAc,KAChB,EAAI,KAAK,GAIT,EAAI,KAAK,QAIT,EAAK,UAAY,KACnB,KAAM,IAAI,OACN,wDACA,QAAO,EAAK,WAAU,MAI5B,GAAM,GAAiB,EAAK,SAAS,cAE1B,GACT,GAAI,CAAE,KAAa,IACjB,KAAM,IAAI,OACN,iCAAiC,EAAS,KAC1C,+BAA8B,OAAO,KAAK,GAAe,MAI/D,GAAM,GAAK,EAAK,WAAM,MAAA,GAAe,OACrC,GAAI,EAAG,QAAU,UACf,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BAChB,GAAS,wCAAwC,EAAG,MAAK,MAElE,GAAM,GAAI,EAAK,OAAO,GACtB,GAAI,CAAC,GAAiB,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BACnB,KAAI,EAAS,gBAAgB,EAAG,MAAK,4BACrC,4BAA2B,EAAE,MAAK,MAGxC,GAAI,EAA6B,EAAE,KAAO,KACxC,EAA6B,EAAE,IAAM,OAErC,GAAM,GAAc,EAA6B,EAAE,IACnD,EAA6B,EAAE,IAAM,EAAI,EAAa,GACtD,EAAY,YA7BhB,OAAW,KAAa,GAAK,SAAlB,IAxBJ,EAAI,EAAa,OAAS,EAAG,GAAK,EAAG,MAArC,GCvIX,AAqBA,GAAM,IAAwB,GAExB,GAA6B,EAE7B,GAAwB,cAG1B,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAU,GAAe,GACzB,EAAY,GAAwB,EAAM,EAAO,EAAO,GACxD,EAAO,EAAM,OACb,EAAY,GAAkB,EAAM,EAAO,EAAO,EAAS,GAC3D,EAAQ,CAAC,UACf,MAAI,IACF,GAAM,KAAK,YAAY,GACvB,EAAM,KAAK,WAAW,GACtB,EAAM,KAAK,aAAa,EAAK,KAC7B,EAAM,KAAK,cAEb,EAAM,KAAK,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,IAAG,KAAK;IACxC,EAAM,KAAK;GAGpB,YACI,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAI,GAAc,GAClB,EAAU,EAAQ,EAAQ,OAAS,GACnC,EAAY,GAAI,OAAM,GAAS,KAAK,GACpC,EAAO,EAAM,OACb,EACF,IAAU,YAAc,GAAoB,GAAQ,EAExD,GAAI,EAAO,EACT,OAAS,GAAM,EAAG,EAAM,EAAI,EAAS,IAEnC,OADM,GAAS,EAAM,EACZ,EAAI,EAAG,EAAI,EAAS,IAC3B,EAAU,GAAK,KAAK,IAChB,EAAU,GACV,GAAY,EAAe,EAAS,GAAI,EAAG,GAAO,QAI5D,MAAO,GAGT,YACI,EAAqC,EAAa,GACpD,GAAI,GACJ,MAAI,OAAM,QAAQ,GAChB,EAAY,WAAW,EAAI,GAAG,QAAQ,KAAuB,MACtD,YAAW,EAAI,GAAG,QAAQ,KAAuB,KACnD,AAAI,GAAS,GAClB,EAAS,IAAI,EAAG,IACX,AAAI,IAAU,OACnB,EAAS,GAAgB,GAEzB,EAAS,WAAW,EAAI,QAAQ,KAAwB,WAGnD,GAAS,EAAQ,GAG1B,YAAyB,GACvB,MAAO,KAAM,EAAI,QAAU,OAG7B,YACI,EAA2B,EAAiB,EAC5C,EAAmB,EAAqB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAM,GAAoB,IAAU,YAAc,EAAI,EAEhD,EAAO,EAAM,GACb,EAAO,EAAM,OACnB,GAAI,IAAS,GACX,GAAI,IAAU,aACZ,GAAM,GAAe,GAAoB,GACzC,MAAO,CAAC,GAAY,EAAa,GAAI,EAAG,IAE1C,MAAI,KAAU,OACL,CAAC,GAAgB,EAAK,KAExB,CAAC,EAAK,GAAG,YAGlB,GAAI,IAAS,GACX,GAAI,EAAO,IACT,GAAM,GAAgB,GAA6B,EAE/C,EAAY,MAAM,KAClB,EAAK,MAAM,EAAG,IACd,EAAW,MAAM,KAAqC,EAAK,MAC1D,GAAO,IAA8B,EACtC,EAAO,IACX,MAAI,KAAU,aACZ,GAAY,GAAoB,GAChC,EAAW,GAAoB,IAE1B,CACL,IACA,EAAU,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAChD,KAAK,MACV,UACA,EACK,IACG,SAAC,EAAG,GAAM,MAAA,IACN,EAAG,EAAU,EAAO,GAA6B,GAAI,KAC5D,KAAK,MACV,KAGJ,GAAM,GACF,IAAU,YAAc,GAAoB,GACpB,MAAM,KAAoB,GAEtD,MAAO,CACL,IACA,EAAY,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAClD,KAAK,MACV,KAKJ,GAAM,GAAW,EAAM,MAAM,GACvB,EAAa,EAAQ,MAAM,GAC3B,EAAS,EAAQ,GAAK,EACtB,EAAkB,GACxB,GAAI,EAAO,IACT,OAAS,GAAI,EAAG,EAAI,GAA4B,KAC9C,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,KAEN,EAAM,KAAK,OACX,OAAS,GAAI,EAAO,GAA4B,EAAI,EAAM,KACxD,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,SAGnB,QAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,IAGrB,GAAM,GAAM,IAAS,EAAI,IAAM,GAC/B,EAAM,GAAK,IAAM,EAAM,GAAK,EAC5B,OAAS,GAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,EAAM,GAAK,IAAM,EAAM,GAAK,EAG9B,OADI,GAAa;EACR,EAAI,EAAG,EAAI,EAAM,IACxB,GAAc;EAEhB,SAAM,EAAM,OAAS,GACjB,IAAM,EAAM,EAAM,OAAS,GAAK,IAAO,GAAS,GAAK,GAClD,EAGT,YAA6B,GAG3B,OADM,GAAyC,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,GAAK,EACpC,EAAc,KAAK,CAAC,EAAK,GAAI,EAAK,EAAI,KAExC,MAAO,GCnMT,kBA6CE,WAAY,EAA2B,EAAU,GAAjD,GAAA,GAAA,KAIE,GAJqC,KAAA,MAAA,EACrC,KAAK,MAAQ,EAAM,QACnB,KAAK,KAAO,GAAmB,GAE3B,GAAU,MACZ,GAAM,GAAI,EAAO,OACjB,EACI,IAAM,KAAK,KACX,WAAM,MAAA,qBAAqB,EAAC,6BACxB,2BAA0B,EAAK,KAAI,QAE7C,GAAI,IAAU,YACZ,KAAM,IAAI,OACN,8JAIN,KAAK,OAAS,GAAU,GAAuB,EAAO,KAAK,MAC3D,KAAK,QAAU,GAAe,GAWhC,SAAA,UAAA,IAAA,SAAI,UAAJ,GAAA,KAA8B,EAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GAC5B,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAEV,EACI,EAAK,SAAW,KAAK,KACrB,WAAM,MAAA,uCAAuC,EAAK,OAAM,UACpD,oBAAmB,EAAK,KAAI,OAEpC,GAAM,GAAQ,KAAK,WAAW,GAC9B,KAAK,OAAO,GAAS,GAUvB,EAAA,UAAA,IAAA,kBAAI,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACF,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAGV,OADI,GAAI,EACU,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,EAAM,GAAK,GAAO,KAAK,MAAM,IAC/B,GAAM,GAAM,qCAAqC,EAAI,KACjD,mBAAkB,KAAK,OAC3B,KAAM,IAAI,OAAM,GAElB,IAGF,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,MAAK,OAAO,IAGrB,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,IAGT,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,KAAK,MAAM,QACnC,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,KAAK,QAAQ,IAC1C,GAAS,EAAK,GAAK,KAAK,QAAQ,GAElC,SAAK,EAAK,OAAS,GAAK,EACjB,GAGT,OAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAY,WAAW,KAAK,OAAQ,KAAK,MAAO,KAAK,QAGhE,KA8BI,GAAiC,KAEjC,GAAuB,iBAaM,GAC/B,GAAY,cAQe,GAC3B,GAAY,mBA4DZ,WAAY,EAAoB,EAAiB,EAAgB,GAXjE,KAAA,KAAO,GAoIG,KAAA,mBAAqB,GAxH7B,KAAK,MAAQ,EAAM,QACnB,KAAK,MAAQ,GAAS,UACtB,KAAK,KAAO,GAAmB,GAC/B,KAAK,QAAU,GAAe,GAC9B,KAAK,OAAS,EACd,KAAK,GAAK,EACV,KAAK,SAAY,KAAK,KAAO,EAAI,KAAK,KAAK,WAAa,SAG1D,cAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQd,EAAA,UAAA,OAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAU,OAAO,KAAK,MAAO,KAAK,MAAY,UAOvD,EAAA,UAAA,WAAA,WACE,MAAO,IAAU,OAAO,KAAK,MAAO,KAAK,MAAY,KAAK,aAStD,EAAA,UAAA,MAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAc,KAAK,MAAO,UASnC,EAAA,UAAA,UAAA,WACE,MAAO,IAAc,KAAK,MAAO,KAAK,aASlC,EAAA,UAAA,KAAN,mHACE,MAAK,kBACC,EAAO,KAAY,KAAK,KAAK,QAC/B,KAAK,QAAU,SACH,CAAA,EAAM,GADlB,CAAA,EAAA,UACI,EAAQ,EAAA,OACd,IACE,MAAA,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,IAAkB,YACxC,GACA,KAAM,IAAI,OACN,kHAIR,MAAA,CAAA,EAAO,SAST,EAAA,UAAA,SAAA,WACE,KAAK,kBACL,GAAM,GAAO,KAAY,SAAS,KAAK,QACvC,GAAI,KAAK,QAAU,SACjB,IACE,MAAQ,GAAsB,IAAI,SAAA,GAAK,MAAA,IAAkB,WAEzD,GACA,KAAM,IAAI,OACN,iGAIR,MAAO,IAIH,EAAA,UAAA,MAAN,2GACE,YAAK,kBACQ,CAAA,EAAM,KAAY,KAAK,KAAK,gBACzC,MADM,GAAO,EAAA,OACT,KAAK,QAAU,SACjB,CAAA,EAAO,GAEP,CAAA,EAAO,GAAI,YAAY,EAAoB,eAS/C,EAAA,UAAA,QAAA,WACE,GAAI,KAAK,WACP,OAEF,KAAY,cAAc,MAC1B,KAAK,mBAAqB,IAI5B,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,oDAGd,EAAA,UAAA,gBAAA,WACE,GAAI,KAAK,WACP,KAAM,IAAI,OAAM,wBAYpB,EAAA,UAAA,MAAA,SAAM,GAAA,MAAA,KAAA,QAAA,GAAA,IACG,GAAU,MAAM,KAAM,IAO/B,EAAA,UAAA,MAAA,WACE,YAAK,kBACE,GAAU,MAAM,OAQzB,EAAA,UAAA,SAAA,SAAS,GAAA,AAAA,IAAA,QAAA,GAAA,IACP,GAAM,GAAO,KAAK,WAClB,MAAO,IAAe,EAAM,KAAK,MAAO,KAAK,MAAO,IAGtD,EAAA,UAAA,KAAA,SAAqB,GACnB,YAAK,kBACE,GAAU,KAAK,KAAW,IAEnC,EAAA,UAAA,SAAA,SAAS,EAAkB,EAAe,GAAjC,MAAA,KAAA,QAAA,GAAA,IACP,KAAK,kBACE,KAAY,aAAa,KAAM,EAAW,EAAM,IAG3D,KACA,OAAO,eAAe,EAAQ,OAAO,YAAa,CAChD,MAAO,SAAC,GAMN,MAAO,CAAC,CAAC,GAAY,EAAS,MAAQ,MAAQ,EAAS,UAAY,MAC/D,EAAS,iBAAmB,2BAoCiB,GAAA,EAAA,GAGnD,WACI,EAAgC,EAAoB,EACpD,GAFJ,GAAA,GAGE,EAAA,KAAA,KACI,EAAa,MAAO,EAAa,MAAO,EAAa,OAAQ,IAAS,KAHxC,SAAA,UAAA,EAIlC,EAAK,KAAO,IAWd,SAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAS,QAAU,KAAK,MAC1B,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,GAAI,CAAC,GAAiB,EAAS,MAAO,KAAK,OACzC,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,KAAY,cAAc,MAC1B,KAAK,OAAS,EAAS,OACvB,KAAY,OAAO,KAAM,OAG3B,EAAA,UAAA,QAAA,WACE,KAAY,gBAAgB,MAC5B,KAAK,mBAAqB,IAE9B,GAvCqD,GAyCrD,OAAO,eAAe,GAAU,OAAO,YAAa,CAClD,MAAO,SAAC,GACN,MAAO,aAAoB,IAAU,EAAS,QAAU,MACpD,EAAS,iBAAkB,aCnhBnC,AAgEA,AAAA,UAAY,GACV,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,OAPU,EAAA,MAAA,GAAA,KAAI,KAqBhB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,QACA,EAAA,UAAA,cAJG,IAAA,IAAiB,KAOtB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,OACA,EAAA,UAAA,cAJG,IAAA,IAAgB,KAOrB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,UACA,EAAA,KAAA,UACA,EAAA,UAAA,cAJG,IAAA,IAAmB,KAOxB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,YACA,EAAA,MAAA,YACA,EAAA,KAAA,YACA,EAAA,UAAA,cAJG,IAAA,IAAqB,KAO1B,GAAM,IAAgB,CACpB,QAAW,GACX,MAAS,GACT,KAAQ,GACR,UAAa,gBAGY,EAAiB,GAC1C,GAAI,IAAU,UAAY,IAAU,UAClC,GAAI,IAAU,UAAY,IAAU,SAClC,MAAO,SAET,KAAM,IAAI,OAAM,kBAAkB,EAAK,SAAS,GAElD,MAAO,IAAc,GAAO,eAIH,GACzB,MAAO,IAAW,EAAM,SCpI1B,YAsBiD,EAAM,GACrD,GAAI,EAAE,QAAU,EAAE,MAChB,MAAO,CAAC,EAAG,GAEb,GAAM,GAAQ,GAAW,EAAE,MAAO,EAAE,OACpC,MAAO,CAAC,EAAE,KAAK,GAAQ,EAAE,KAAK,gBAGC,EAAW,GAC1C,EACI,EAAE,QAAU,EAAE,MACd,WAAM,MAAA,2BAA2B,EAAE,MAAK,QACpC,YAAW,EAAE,MAAK,oCAGG,EAAgB,GAC7C,MAAO,GAAW,KAAK,SAAA,GAAK,MAAA,GAAE,KAAO,EAAO,iBAeR,GACpC,GAAM,GAAiB,GACjB,EAAO,GAAI,KACjB,UAAoB,EAAQ,EAAM,GAC3B,EAGT,YACI,EAA4B,EAAgB,GAC9C,GAAI,GAAa,KACf,OAEF,GAAI,YAAqB,IACvB,EAAK,KAAK,GACV,OAEF,GAAI,CAAC,GAAW,GACd,OAGF,GAAM,GAAW,EACjB,OAAW,KAAK,IACd,GAAM,GAAM,EAAS,GACrB,AAAK,EAAK,IAAI,IACZ,GAAK,IAAI,GACT,GAAoB,EAAK,EAAM,KAMrC,YAAoB,GAClB,MAAO,OAAM,QAAQ,IAAQ,MAAO,IAAQ,kHCrF9C,AAmFA,GAAA,IAAA,WAAA,aAEE,KAAA,oBAAwC,GAExC,KAAA,eAAiB,EACjB,KAAA,SAAW,EACX,KAAA,WAAa,EACb,KAAA,iBAAmB,EACnB,KAAA,eAAiB,EAMjB,KAAA,cAAgB,EAGhB,KAAA,YAAc,EAId,KAAA,WAA2B,GAK3B,KAAA,kBAA8B,GAC9B,KAAA,YAAc,EAEd,KAAA,WAAa,GAAI,SAQjB,KAAA,UAAY,GACZ,KAAA,cACI,CAAC,SAAU,EAAG,WAAY,EAAG,UAAW,EAAG,QAAS,GAAI,OAAQ,MAEpE,SAAA,UAAA,QAAA,WACE,OAAW,KAAgB,MAAK,oBAC9B,KAAK,oBAAoB,GAAc,WAG7C,KAEA,GAAA,WAgBE,WAAmB,GAAA,KAAA,IAAA,EAbnB,KAAA,SAA0C,GAC1C,KAAA,gBAKI,GAKI,KAAA,qBAAuB,EAG7B,KAAK,MAAQ,GAAI,IAGb,SAAA,UAAA,MAAN,iHACE,GAAI,KAAK,oBAAsB,KAC7B,MAAA,CAAA,EAAO,KAAK,mBAAmB,KAAK,eAEtC,GAAI,KAAK,iBAAmB,KAC1B,MAAA,CAAA,GAEI,EAAiB,KAAK,oBAEnB,EAAI,yBAAG,GAAI,EAAe,OAC3B,GAAc,EAAe,GACnB,CAAA,EAAM,KAAK,kBAAkB,GAAa,UAFnB,CAAA,EAAA,gBAEjC,GAAU,EAAA,OACZ,EACF,CAAA,EAAM,KAAK,WAAW,IADpB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,UALuC,wBAS3C,KAAM,IAAI,OACN,gFAIN,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,GAAI,KAAK,oBAAsB,KAC7B,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,uHAIlC,GAAI,KAAK,iBAAmB,MACpB,GAAA,GAAA,KAAA,kCAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UACb,GAAI,EACF,KAAM,IAAI,OACN,iCAAiC,EAAI,uHAI3C,KAAK,WAAW,GAElB,MAAO,MAAK,iDAGd,EAAA,UAAA,aAAA,WACE,MAAO,QAAO,KAAK,KAAK,kBAG1B,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,CAAE,KAAe,MAAK,UAGxB,GAAI,IAAe,MAAK,iBACf,GAAA,GAAA,KAAA,kBAAA,GAAA,UACP,GAAI,EAEF,MAAO,UAGT,OAAO,MAGX,MAAO,MAAK,SAAS,IAGvB,EAAA,UAAA,mBAAA,SAAmB,GAEjB,MAAM,KAAe,MAAK,gBAGnB,KAAK,gBAAgB,GAAa,QAFhC,MAKX,EAAA,UAAA,gBAAA,SACI,EACA,EACA,GACF,MADE,AAAA,KAAA,QAAA,GAAA,GACE,IAAe,MAAK,gBACtB,SAAQ,KACD,EAAW,sEAEX,IAET,MAAK,gBAAgB,GAAe,CAAC,QAAO,EAAE,SAAQ,GAC/C,KAGH,EAAA,UAAA,WAAN,SAAiB,2GACf,GAAI,KAAK,gBAAgB,IAAgB,KACvC,KAAM,IAAI,OAAM,iBAAiB,EAAW,iCAE9C,MAAK,YAAc,EACf,KAAK,SAAS,IAAgB,KAChC,MAAK,gBAAkB,KACjB,EAAuB,KAAK,kBAAkB,GAA7C,EAAO,EAAA,QAAE,EAAS,EAAA,UACV,EAAY,CAAA,EAAM,GAAlB,CAAA,EAAA,IAHb,CAAA,EAAA,UAGyB,SAAA,EAAA,oBAAgB,EAAA,mBAC3C,GADM,EAAM,EACR,CAAC,EACH,MAAA,CAAA,EAAO,qBAGX,YAAK,gBAAkB,KAAK,SAAS,GACrC,KAAK,yBAEL,KAAK,SAAW,GAAI,IAAS,KAAK,iBAElC,CAAA,EAAO,UAGD,EAAA,UAAA,uBAAR,WAAA,GAAA,GAAA,KACQ,EAAU,GAAqB,KAAK,aAC1C,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,WAAa,MACtB,EAAO,UAAU,EAAK,oBAKpB,EAAA,UAAA,yBAAR,SAAiC,GAAjC,GAAA,GAAA,KACQ,EAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,aAAe,MACxB,EAAO,YAAY,EAAK,SAAS,OAW/B,EAAA,UAAA,kBAAR,SAA0B,GAA1B,GAAA,GAAA,KAEQ,EAAuB,KAAK,gBAAgB,GAClD,GAAI,GAAwB,KAC1B,KAAM,IAAI,OACN,6BAA6B,EAAW,4BAG9C,IACE,GAAM,GAAU,EAAqB,UAMrC,GAAI,GAAW,CAAE,aAAmB,MAC7B,MAAO,GAAQ,MAAS,YAC7B,GAAM,GAAY,EAAE,KAAK,qBACnB,EACF,EACK,KAAK,SAAA,GAEJ,MAAI,GAAY,EAAK,qBACZ,GAET,GAAK,SAAS,GAAe,EAC7B,EAAK,mBAAqB,KACnB,MAER,MAAM,SAAA,GAEL,MAAI,GAAY,EAAK,sBAGrB,GAAK,mBAAqB,KAC1B,QAAQ,KACJ,6BAA6B,EAAW,WAC5C,QAAQ,KAAK,EAAI,OAAS,EAAI,UACvB,KAEjB,YAAK,mBAAqB,EACnB,CAAC,QAAO,EAAE,UAAW,QAE5B,aAAK,SAAS,GAAe,EACtB,CAAC,QAAS,GAAM,UAAW,UAE7B,GACP,eAAQ,KAAK,6BAA6B,EAAW,WACrD,QAAQ,KAAK,EAAI,OAAS,EAAI,SACvB,CAAC,QAAS,GAAO,UAAW,MAIvC,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAE,KAAe,MAAK,iBACxB,KAAM,IAAI,OAAS,EAAW,kCAEhC,AAAI,KAAK,cAAgB,GAAe,KAAK,oBAAsB,MAGjE,KAAK,uBAGP,AAAI,IAAe,MAAK,UACtB,MAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,IAGvB,MAAO,MAAK,gBAAgB,GAG5B,AAAI,KAAK,cAAgB,GACvB,MAAK,mBAAqB,KAC1B,KAAK,YAAc,KACnB,KAAK,gBAAkB,OAInB,EAAA,UAAA,kBAAR,WAAA,GAAA,GAAA,KACE,GAAI,OAAO,KAAK,KAAK,iBAAiB,SAAW,EAC/C,KAAM,IAAI,OAAM,iCAElB,MAAO,QAAO,KAAK,KAAK,iBAAiB,KAAK,SAAC,EAAW,GAExD,MAAO,GAAK,gBAAgB,GAAG,SAC3B,EAAK,gBAAgB,GAAG,YAIxB,EAAA,UAAA,gCAAR,WAIE,OAFM,GAAiB,KAAK,oBAEnB,EAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAc,EAAe,GAC7B,EAAA,KAAA,kBAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAChB,GAAI,GAAa,EACf,MAAO,CAAC,KAAM,EAAa,UAAS,GAGxC,KAAM,IAAI,OACN,2EAIN,EAAA,UAAA,SAAA,SAAS,EAAwB,GAC/B,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAa,EAAK,QAClB,EAAS,KAAK,SAAS,GAG7B,EAAW,YAAY,GACvB,EAAK,QAAU,EACf,EAAQ,KAAK,EAAQ,EAAQ,EAAK,MAAO,EAAK,OAC9C,AAAI,KAAK,0BAGP,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,MAIvE,EAAA,UAAA,KAAA,SAAgC,EAA6B,GAA7D,GAAA,GAAA,KAEM,EAAe,KACnB,GAAI,GAAM,MAER,GAAI,MAAO,IAAa,WACtB,KAAM,IAAI,OAAM,uCAElB,EAAK,OAGL,GAAI,MAAO,IAAa,UAAY,CAAE,aAAoB,SACxD,KAAM,IAAI,OACN,kFAGN,GAAI,MAAO,IAAO,WAChB,KAAM,IAAI,OACN,kFAGN,EAAO,EAIT,GAAI,GACJ,MAAO,MAAK,UACR,WAAM,MAAA,GAAK,WAAW,IAAO,WAAM,MAAA,GAAK,SAAS,IAAS,WACxD,SAAS,IACT,AAAI,YAAkB,UACpB,QAAQ,MAAM,2CAET,KAIP,EAAA,UAAA,UAAR,SAAqB,EAAmB,EAAiB,GACvD,IACA,IACE,GAAM,GAAM,IACZ,WACO,QACA,GACP,UACM,IAKF,EAAA,UAAA,aAAR,WACE,MAAO,GAAO,gBAIR,EAAA,UAAA,eAAR,WACE,MAAO,GAAO,kBAYR,EAAA,UAAA,MAAR,SAAc,GACZ,GAAM,GAAI,KAAK,qBAAqB,EAAE,OAAQ,EAAE,MAAO,EAAE,OACnD,EAAS,CAAC,EAAC,GACX,EAAO,SAAC,GAAe,MAAC,CAC5B,EAAG,WACD,GAAM,GAAQ,UACR,EAAa,CAAC,EAAG,GACjB,EAAQ,CAAC,MAAK,GAEpB,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAC5B,EAAoC,KAAiB,GACrD,MAGF,EAAkB,GACxB,YAAK,YAAY,KAAK,MAAM,YAAY,KAAM,EAAQ,CAAC,GAAI,EAAM,EAAO,IACjE,GAgBT,EAAA,UAAA,UAAA,SACI,EAAoB,EAAwB,EAC5C,EAAyB,GAC3B,GAAM,GAAoB,KACpB,EAAsB,KAI5B,MAAO,MAAK,cACR,EAAa,EAAQ,EAAe,EAAY,EAAO,EACvD,IAGE,EAAA,UAAA,uBAAR,WACE,MAAO,MAAK,IAAI,QAAQ,YAGlB,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAkB,KAAK,QAAQ,aAGjC,EAAmB,EACvB,EAAS,QAAQ,SAAA,GAGf,GAAqB,EAAK,QAAU,YAAc,EAAI,IAQxD,GAAM,GACF,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,GACjE,EACF,EAAkB,EAAmB,EAAmB,EAC5D,GAAI,EAAgB,EAClB,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,iCAC5B,KAAI,EAAa,6BAA6B,EAAU,OAQhE,EAAA,UAAA,cAAA,SACI,EAA6B,EAC7B,EACA,EAAqB,EAAsB,EAC3C,GAJJ,GAAA,GAAA,KAKM,EACA,EAAkB,GAChB,EAAW,KAAK,WACtB,AAAI,GAAc,MAChB,GACI,KAAK,MAAM,aAAe,KAAO,KAAK,MAAM,YAAY,KAAO,IAGrE,GAAM,GAAoB,KAAK,MAAM,SAC/B,EAAqB,KAAK,MAAM,WAEtC,AAAI,KAAK,0BACP,KAAK,MAAM,kBAAkB,KAAK,GAGpC,GAAI,GACE,EAAS,GAAU,EAAY,KAAK,aACtC,EACJ,GAAI,GAAU,KACZ,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAO,WAAW,CAAC,OAAM,EAAE,MAAK,EAAE,QAAS,EAAK,UACtD,GAAM,GAAW,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC7C,AAAI,EAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAE3D,GAAM,GAAa,EAAS,IACxB,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,MACb,MAAA,GAAK,qBAAqB,EAAQ,EAAO,KAMjD,GAAI,GACF,GAAI,GACA,EAAK,sBAAsB,EAAY,EAAQ,GACnD,GAAI,GAAiB,MAKnB,AAAI,GAAiB,MACnB,GAAgB,IAElB,GAAM,GAAa,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAC7D,EAAiB,IAAgB,IAAI,QAAQ,OAAO,GAEtD,EAAQ,EAAK,2BAA2B,GAE1C,MAAO,SAGT,GAAM,GAAyB,SAAC,GAI9B,GAAI,CAAC,EACH,OAEF,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,OAGrD,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAK,KAAK,WAAM,MAAA,GAAY,EAAK,QAAS,KAChD,GAAM,GAAQ,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC1C,MAAI,GAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAEpD,GAKX,GAAI,GACJ,YAAK,UACD,WAAM,MAAA,GAAK,MAAM,eAAe,WAAM,MAAA,GAAK,MAAM,eAAe,WAC9D,AAAI,CAAC,EAAK,IAAI,QAAQ,UAAY,CAAC,EAAK,MAAM,UAC5C,EAAU,IAEV,GAAgB,EAAK,SAAS,cAC1B,EAAY,EAAQ,WAAM,MAAA,OAC9B,AAAI,EAAK,IAAI,QAAQ,UACnB,EAAK,SAAS,iBAAiB,GAEjC,EAAU,EAAc,WAIhC,AAAI,GACF,KAAK,YACD,EAAY,EAAQ,EAAS,EAAe,EAAO,GAGzD,AAAI,KAAK,MAAM,WACb,KAAK,MAAM,cAAc,QAAQ,KAAK,CACpC,KAAM,EACN,WAAY,KAAK,MAAM,SAAW,EAClC,mBAAoB,KAAK,MAAM,SAC/B,aAAc,KAAK,MAAM,WAAa,EACtC,qBAAsB,KAAK,MAAM,WACjC,YAAa,OAAO,KAAK,GAAQ,IAC7B,SAAA,GAAO,MAAA,GAAO,IAAQ,KAAO,EAAO,GAAK,MAAQ,OACrD,aAAc,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,QACvC,aAAc,EAAc,OAC5B,UAAW,EAAc,YAGrB,MAAM,QAAQ,GAAO,EAAU,EAAQ,IAQzC,EAAA,UAAA,2BAAR,SAAmC,GAAnC,GAAA,GAAA,KACQ,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,MACzD,MAAO,IAaD,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAa,GAAY,GAC/B,GAAI,GAAc,MAChB,GAAM,GAAyB,EAAW,cAAgB,GACpD,EAA2B,EAAW,eAAiB,GAIzD,EAAkB,OACtB,AAAI,EAAW,cACb,GACI,MAAM,QAAQ,GACd,WAAM,MAAA,2DAEV,EAAqB,OAAO,KAAK,GAAQ,IAAI,SAAC,GAAQ,MAAA,GAAO,MAE7D,EAAqB,EAAa,IAAI,SAAC,GAAc,MAAA,GAAO,KAG9D,GAAM,GACF,EAAQ,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAE3C,MAAO,GAAmB,OAAO,GAInC,MAAO,OAQT,EAAA,UAAA,WAAA,SACI,EAAoB,EAAiB,EACrC,GACF,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,iDAElB,EAAQ,GAAS,UACjB,EAAU,GAAW,KAAK,QAC1B,GAAI,GAAc,EAClB,AAAI,IAAU,UAAY,GAAc,EAAO,KAC7C,GAAe,EAAoB,IAAI,SAAA,GAAK,MAAA,IAAkB,MAEhE,GAAM,GAAS,EAAQ,MAAM,EAAa,EAAO,GAC3C,EAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAIhD,GAHA,KAAK,OAAO,EAAG,GAGX,IAAU,UACZ,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAW,GAAqB,GACtC,KAAK,MAAM,UAAY,EAAW,EAAK,MACvC,EAAK,MAAQ,EAEf,MAAO,IAQT,EAAA,UAAA,qBAAA,SACI,EAAgB,EAAiB,EACjC,GACF,EAAQ,GAAS,UACjB,GAAM,GAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAChD,YAAK,OAAO,EAAG,GACR,GAGT,EAAA,UAAA,aAAA,SACI,EAAsB,EAAkB,EACxC,GADsB,AAAA,IAAA,QAAA,GAAA,IAExB,EAAO,GAAQ,KAAK,iBAAiB,WACrC,AAAI,GAAS,MAAQ,IAAU,EAAa,OAC1C,GAAe,EAAa,KAAK,IAEnC,GAAM,GAAI,GAAI,IAAS,EAAc,EAAW,EAAM,KAAK,gBAC3D,GAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,KAC5C,KAAM,IAAI,OAAM,sBAAsB,EAAE,KAAI,2BAE9C,YAAK,MAAM,oBAAoB,EAAE,MAAQ,EACzC,KAAK,OAAO,EAAG,KAAK,SACb,GAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAM,GAAW,KAAK,MAAM,WAAW,IAAI,EAAE,QACzC,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,SACpC,EAKJ,GAJA,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAET,IAAa,GACf,KAAK,MAAM,iBAIX,GAAI,GAAQ,EACZ,AAAI,EAAE,QAAU,aAAe,EAAE,QAAU,UACzC,GAAQ,EAAE,KAAO,GAAqB,EAAE,QAE1C,KAAK,MAAM,WAAW,IAAI,EAAE,OAAQ,CAClC,QAAS,GAAW,KAAK,QACzB,MAAO,EAAE,MACT,MAAO,EAAE,MACT,MAAK,EACL,SAAU,IAEZ,KAAK,MAAM,UAAY,EAGzB,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,WAEpC,AAAM,YAAa,KACjB,KAAK,MAAM,IAIf,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAC,KAAK,MAAM,WAAW,IAAI,EAAE,QAC/B,OAGF,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAEb,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,EAAE,QACnC,EAAW,EAAK,SAEtB,AAAI,GAAY,EAGd,CAAI,EAAE,QAAU,aACd,MAAK,MAAM,UAAY,EAAK,OAE9B,KAAK,MAAM,iBAEX,EAAK,QAAQ,YAAY,EAAE,QAC3B,KAAK,MAAM,WAAW,OAAO,EAAE,SAE/B,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,YAOxC,EAAA,UAAA,iBAAA,WACE,OAAW,KAAW,MAAK,MAAM,qBAC/B,GAAM,GAAI,KAAK,MAAM,oBAAoB,GACzC,KAAK,gBAAgB,KAIzB,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,cAAc,GACnB,AAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,MAC5C,MAAO,MAAK,MAAM,oBAAoB,EAAE,OAI5C,EAAA,UAAA,OAAA,WACE,GAAM,GAAO,KAAK,QAAQ,SAC1B,SAAK,WAAa,KAAK,MAAM,WAC7B,EAAK,eAAiB,KAAK,MAAM,eACjC,EAAK,SAAW,KAAK,MAAM,SAC3B,AAAI,KAAK,MAAM,iBAAmB,GAChC,GAAK,WAAa,GAClB,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,IAEjB,EAAK,QAAQ,KACT,0EAGC,GAGH,EAAA,UAAA,QAAN,SAAc,iHAEZ,YAAK,MAAM,UAAY,GAEjB,EAAa,KAAK,MAAM,SACxB,EAAkB,KAAK,MAAM,WAEnC,KAAK,MAAM,cAAc,QAAU,GACnC,EAAA,KAAK,MAAM,cAAuB,CAAA,EAAM,YAAxC,EAAyB,OAAS,EAAA,OAElC,KAAK,MAAM,UAAY,GAEvB,KAAK,MAAM,cAAc,UAAY,KAAK,IAAG,MAAR,KAC9B,KAAK,MAAM,cAAc,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,sBACnD,KAAK,MAAM,cAAc,SAAW,KAAK,MAAM,SAAW,EAC1D,KAAK,MAAM,cAAc,WACrB,KAAK,MAAM,WAAa,MACP,EAAA,KAAK,MAAM,cAAc,+BAAzB,GAAA,EAAA,OAAV,GAAM,EAAA,GACf,EAAA,EAAsB,CAAA,EAAM,EAAO,eADgB,CAAA,EAAA,UACnD,SAAO,aAAe,EAAA,OACtB,EAAA,EAAmB,CAAA,EAAM,EAAO,kBAAhC,EAAO,UAAY,EAAA,wBAFA,wBAIrB,MAAA,CAAA,EAAO,KAAK,MAAM,qBAGpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,MAAM,cAAgB,GAAK,KAAK,MAAM,cAAgB,GAG5D,EAAA,UAAA,YAAR,SACI,EAAoB,EAAwB,EAC5C,EAAyB,EAAiB,GAF9C,GAAA,GAAA,KAGQ,EACF,CAAC,GAAI,KAAK,MAAM,iBAAkB,WAAU,EAAE,OAAM,EAAE,QAAO,EAAE,MAAK,GAElE,EAAa,GAAY,GAC/B,AAAI,GAAc,MAChB,GAAgB,EAAW,UAE7B,AAAI,GAAiB,MACnB,GAAS,SAAW,SAAC,GAGnB,SAAM,EAAI,IAAI,SAAC,EAAI,GACjB,GAAI,GAAM,MACR,GAAM,GAAS,EAAQ,GACjB,EAAO,GAAyB,EAAO,KAAM,EAAO,OAC1D,MAAO,GAAK,WAAW,EAAM,EAAO,MAAO,EAAO,OAEpD,MAAO,KAIF,EAAc,EAAI,OAAS,EAAI,EAAM,EAAI,GAAI,EAAO,KAG/D,KAAK,MAAM,WAAW,KAAK,IAG7B,EAAA,UAAA,KAAA,SAAuB,GACrB,SAAO,KAAO,GACP,GAGD,EAAA,UAAA,UAAR,WACE,AAAI,KAAK,MAAM,gBAAkB,GAC/B,MAAK,MAAM,WAAa,IAE1B,KAAK,MAAM,iBAGL,EAAA,UAAA,QAAR,WACE,KAAK,MAAM,iBAOb,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAwB,CAC5B,MAAO,GACP,KAAM,gBACN,GAAI,KAAK,MAAM,eAEjB,AAAI,GACF,GAAU,KAAO,GAEnB,KAAK,MAAM,WAAW,KAAK,GAC3B,KAAK,MAAM,YAAc,GAO3B,EAAA,UAAA,SAAA,SAAS,GAMP,OANF,GAAA,KACQ,EAAyB,GAAsB,GAC/C,EACF,GAAI,KAAI,EAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,MAGrC,EAAI,EAAG,EAAI,KAAK,MAAM,YAAY,MAAM,OAAQ,KACvD,GAAM,GAAS,KAAK,MAAM,YAAY,MAAM,GAC5C,AAAI,CAAC,EAAO,MAAQ,CAAC,EAA0B,IAAI,EAAO,KACxD,EAAO,UAIX,GAAM,GAAW,KAAK,MAAM,WAAW,MACvC,KAAK,MAAM,YAAc,KAAK,MAAM,WAAW,SAAW,EACtD,KACA,KAAK,MAAM,WAAW,KAAK,MAAM,WAAW,OAAS,GAGzD,EAAuB,QAAQ,SAAA,GAG7B,AAAI,CAAC,EAAO,MAAQ,EAAO,UAAY,EAAS,IAC9C,EAAK,MAAM,MAWjB,EAAA,UAAA,UAAA,SACI,EAAY,EAAc,EAC1B,GAFJ,GAAA,GAAA,KAKE,GAHE,AAAA,IAAA,QAAA,GAAA,IACF,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,8CACrB,GAAM,MAAQ,EAAG,QAAU,UAC7B,KAAM,IAAI,OAAM,0CAA0C,EAAG,MAAK,KAGpE,GAAM,GAAI,KAAK,UACX,WAAM,MAAA,GAAK,aAAa,WAAM,MAAA,GAAK,WACnC,WAAM,MAAA,GAAK,KAAK,UAAW,KAE/B,EACI,YAAa,GACb,WAAM,MAAA,mDAEV,GAAM,GAAe,GAAqB,KAAK,MAAM,WAAY,EAAI,GACrE,GAAI,CAAC,GAAoB,EAAa,SAAW,GAAK,EAAG,OAAS,EAChE,KAAM,IAAI,OACN,uIAKN,MAAO,MAAK,KAAK,WAAY,WAC3B,GAAM,GAAuD,GAC7D,EAAuB,EAAE,IAAO,GAAc,GAAK,EAAE,OAGrD,GACI,EAAwB,EAExB,SAAA,GAAK,MAAA,GAAK,KAAK,IAEf,IACJ,GAAM,GAAQ,EAAG,IAAI,SAAA,GAAK,MAAA,GAAuB,EAAE,MAEnD,MAAI,GAAK,MAAM,gBAAkB,GAG/B,GAAK,MAAM,WAAW,QAAQ,SAAA,GAC5B,OAAqB,GAAA,EAAA,EAAA,EAAK,MAAL,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,EAAO,aAGX,EAAK,MAAM,WAAa,MAEnB,CAAC,MAAO,EAAG,MAAK,MAI3B,EAAA,UAAA,WAAA,SAA6B,GAA7B,GAAA,GAAA,KAEE,SACI,GAAgB,GAChB,WAAM,MAAA,sDACH,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACN,EACI,EAAO,MAAM,SAAA,GAAK,MAAA,aAAa,KAC/B,WAAM,MAAA,qEAGV,GAAI,GAIE,EAA2B,GACjC,SAAO,QAAQ,SAAC,EAAO,GACrB,EAAS,GAAK,IAET,EAAK,cACR,SAAC,EAAG,GACF,SAAM,EAAC,MAAA,OAAQ,EAAM,OAAA,CAAE,KACvB,EACI,EAAI,gBAAiB,GACrB,WAAM,MAAA,+FAEV,EACI,GAAgB,EAAI,UACpB,WAAM,MAAA,qGAEH,EAAI,OAEb,EACA,SAAC,EAAO,GACN,GAAM,GAAU,EAAI,SAAS,EAAI,GAC3B,EACF,MAAM,QAAQ,GAAW,EAAU,CAAC,GACxC,EACI,EAAM,SAAW,EAAO,OACxB,WAAM,MAAA,wKAGV,EACI,EAAM,MAAM,SAAA,GAAK,MAAA,aAAa,KAC9B,WAAM,MAAA,yIAGV,GAAM,GAAyC,GAC/C,SAAM,QAAQ,SAAC,EAAM,GACnB,EAAQ,GAAK,WAAM,MAAA,MAEd,MAKjB,EAAA,UAAA,SAAA,SAAS,GAEP,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,SAAS,IAE/B,EAAA,UAAA,KAAA,SAAK,GAEH,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,KAAK,IAGrB,EAAA,UAAA,KAAN,SAAW,qGACH,SAAQ,KACK,CAAA,EAAM,KAAK,QAAQ,KAAK,WAArC,SAAa,EAAA,OACnB,EAAW,OAAS,KAAQ,EAC5B,CAAA,EAAO,SASD,EAAA,UAAA,MAAR,SAAgC,GAC9B,MAAI,MAAK,MAAM,aAAe,MAC5B,GAAO,QAAU,KAAK,MAAM,YAAY,GACxC,KAAK,MAAM,YAAY,MAAM,KAAK,IAG7B,GAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAOpB,EAAA,UAAA,MAAA,WAEE,KAAK,uBAEL,KAAK,MAAM,UACX,KAAK,IAAI,QACT,KAAK,MAAQ,GAAI,IAEjB,OAAW,KAAe,MAAK,SAC7B,KAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,GAEvB,KAAK,YAAc,KACnB,KAAK,gBAAkB,KACvB,KAAK,mBAAqB,MA/sBb,EAAA,aAAe,EAKf,EAAA,eAAiB,EA4sBlC,KAEA,YAAc,GACZ,GAAM,GAAS,GAAmB,GAAc,GAAQ,WACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,yBAIxC,GAAM,GAAK,KACX,GAAI,EAAG,WAAa,MAClB,GAAM,GAAc,GAAI,IAAY,GACpC,EAAG,UAAY,GAAI,IAAO,GAE5B,UAAqB,EAAG,UAAU,KAIlC,GAAiB,WAAM,MAAA,GAAG,YACnB,EAAG,UAGL,GAAM,GAAS,iBAQF,EAAW,GAE7B,GAAM,GAAS,CAAC,EAAC,EAAE,EAAC,GACpB,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,EAAG,GAC3B,SAAK,CAAC,EAAG,IACF,GACN,EAAgC,KAAqB,ICzrC1D,AAkBA,cACE,MAAO,OAAO,YAAc,aAAe,WAAa,mBAIxD,GAAI,MAEF,GAAM,GAAI,UAAU,WAAa,UAAU,QAAW,OAAe,MAErE,MAAO,2TACK,KAAK,IAEb,0kDACK,KAAK,EAAE,OAAO,EAAG,IAE5B,MAAO,iBAIP,MAAQ,OAAO,SAAW,aAAe,OAAO,UAAY,MAEvD,MAAO,oBAAsB,6DCvCpC,AAqBA,GAAM,IAAM,KAOZ,GAAI,aAAa,QAAS,WAAM,MAAA,IAAO,SAAA,GACrC,AAAI,GACF,QAAQ,KACJ,iJAOR,GAAI,aAAa,aAAc,WAAM,MAAA,QAGrC,GAAI,aACA,UACA,WAAM,MAAC,OAAO,UAAY,aACrB,MAAO,SAAQ,UAAa,aAC5B,MAAO,SAAQ,SAAS,MAAS,cAG1C,GAAI,aACA,YACA,WAAM,MAAA,OAAO,YAAc,aAAe,WAAa,MACnD,UAAU,WAAa,MAAQ,SAAS,KAAK,UAAU,YACvD,aAAa,KAAK,UAAU,UAMpC,GAAI,aAAa,OAAQ,WAAM,MAAA,KAM/B,GAAI,aACA,qCAAsC,WAAM,MAAA,IAAI,QAAQ,WAG5D,GAAI,aAAa,+BAAgC,WAAM,MAAA,KAGvD,GAAI,aAAa,UAAW,WAAM,MAAA,KCvElC,YAuB2B,EAAiB,GAC1C,GAAI,GAAwB,EAE5B,GAAI,GAAa,GACf,MAAO,KAAU,SAAW,GAAK,CAAC,EAAI,QAExC,GAAI,CAAC,MAAM,QAAQ,GACjB,MAAO,GAIT,OAFM,GAAkB,GAEjB,MAAM,QAAQ,IACd,GAAa,IAAc,IAAU,UAC1C,EAAM,KAAK,EAAU,QACrB,EAAY,EAAU,GAExB,MAAI,OAAM,QAAQ,IACd,KAAM,QAAQ,uCAChB,GAA2B,EAAK,EAAO,IAGlC,EAGT,YACI,EAAiB,EAAiB,GAEpC,GADA,EAAU,GAAW,GACjB,CAAE,MAAM,QAAQ,IAAS,CAAC,GAAa,IACzC,EACI,EAAM,SAAW,EACjB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,qBACnC,yCAAwC,EAAM,GAAE,eACxD,OAEF,EACI,EAAM,OAAS,EACf,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,4BACnC,uBAAsB,EAAI,OAAM,eACxC,EACI,EAAI,SAAW,EAAM,GACrB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,iBAAiB,EAAM,GAAE,IAC5D,sBAAqB,EAAI,OAAM,eAEvC,OADM,GAAW,EAAM,MAAM,GACpB,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAA2B,EAAI,GAAI,EAAU,EAAQ,OAAO,IAIhE,YACI,EAAmC,EAAuB,EAC1D,GACF,GAAI,GAAiB,KACnB,OAEF,GAAI,IAAkB,WAAa,IAAkB,GACjD,IAAkB,WAAa,IAAgB,SACjD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,UAChD,OAAM,EAAa,oBAAoB,EAAW,uBAKtD,EAAiB,EAAiB,EAClC,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,YAAa,GACf,UAAY,EAAc,EAAE,MAAO,EAAS,GACrC,EAET,GAAI,GAAgB,GAAW,GAS/B,GANA,AAAI,IAAkB,UAClB,CAAC,OAAQ,QAAS,WAAW,QAAQ,IAAiB,GACxD,GAAgB,GAElB,GAAY,EAAc,EAAe,EAAS,GAE7C,GAAK,MACL,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IAAM,MAAO,IAAM,UACtD,MAAO,IAAM,WAAa,MAAO,IAAM,UAC1C,GAAM,GAAO,GAAK,KAAO,OAAU,EAAS,YAAY,KACxD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,eAChD,mCAAkC,EAAI,MAE5C,GAAM,GAAgB,GAAW,EAAG,GACpC,AAAI,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IACrC,GAAI,CAAC,IAEP,GAAM,GAAiB,GACjB,EAAS,IAAkB,SAC7B,GAAa,EAAG,GAChB,GAAQ,EAAe,GAAI,GAC/B,MAAO,GAAO,WAAW,EAAQ,EAAe,eAI9C,EAA0B,EAAiB,EAC3C,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,OACN,YAAY,EAAO,cAAc,EAAY,2CAGnD,GAAM,GAAU,EAChB,MAAO,GAAQ,IACX,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAM,EAAO,IAAI,EAAC,IAAK,IACjD,GCnIN,GAkBa,IAAkB,kBAOQ,GACrC,GAAM,GAAO,OAAO,KAAK,GACzB,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,OACN,yGAEG,GAAK,OAAM,WAGpB,GAAI,GAAS,EAAK,GACZ,EAAK,EAAE,GAGb,AAAI,EAAO,SAAS,MAClB,GAAS,EAAO,UAAU,EAAG,EAAO,OAAS,IAI/C,EAAS,EAAS,GAGlB,GAAM,GAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,EAAO,WAAW,GAClB,IACE,GAAM,GAAS,EAAE,MAAA,OAAI,GACrB,MAAI,aAAkB,UACpB,QAAQ,MAAM,2CAEhB,EAAO,SAAS,GACT,QACA,GACP,QAAO,SAAS,MACV,IAGV,cAAO,eAAe,EAAI,OAAQ,CAAC,MAAO,EAAQ,aAAc,KAGzD,EC/DT,AA8CA,YAAoC,EAAoB,GACtD,GAAM,GAAQ,EAAgB,EAAM,OAAQ,WACtC,EAAQ,EAAgB,EAAM,OAAQ,WAC5C,GACI,EAAM,MAAO,EAAM,MACnB,yBAAyB,EAAM,MAAK,QAAQ,EAAM,MAAK,yCAG3D,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,QAAQ,EAAO,IAE1B,EAAwB,CAAC,KAAM,EAAO,KAAM,GAClD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC/DnC,YAyBI,EAAoB,EAAiB,EACrC,GAIF,GAHA,AAAI,GAAS,MACX,GAAQ,GAAW,IAEjB,IAAU,YACZ,KAAM,IAAI,OACN,oFAGN,GAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IACxC,MAAO,IAAW,UAAY,MAAO,IAAW,WAChD,MAAO,IAAW,SACpB,KAAM,IAAI,OACN,4HAGN,GAAI,GAAS,MACX,GAAmC,GAEnC,GAAM,GAAe,GAAc,GAC7B,EAAe,GAAc,GACnC,EACI,IAAiB,EACjB,WACI,MAAA,iCAAiC,EAAK,6BACnC,GAAY,mBAAmB,KAE1C,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,EAAE,GAC1C,GAAM,GAAW,EAAc,GACzB,EAAoB,IAAM,EAAc,OAAS,EACnD,IAAa,GAAc,EAAM,MAAM,IACvC,GACJ,EACI,EAAc,KAAO,EAAM,IAAM,CAAC,EAClC,WAAM,MAAA,+CACF,KAAI,EAAa,kCACjB,WAAU,EAAK,UAI3B,MAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IAC1C,GAAS,CAAC,IAGZ,EAAQ,GAAS,EACjB,EAAS,IAAU,SACf,GAAa,EAAQ,GACrB,GAAQ,EAAoB,GAAI,IAC7B,EAAO,WAAW,EAAsB,EAAO,GC1ExD,YAqDI,EAAoB,EAAqB,GAC3C,GAAM,GAAgB,GAAW,EAAQ,GACzC,MAAO,IAAW,EAAQ,EAAO,EAAe,GCvDlD,AAsBO,GAAM,IAAkD,CAC7D,QAAW,EACX,QAAW,EACX,MAAS,EACT,OAAU,EACV,MAAS,EACT,KAAQ,EACR,UAAa,GC7Bf,AA2BA,GAAM,IAA0B,cAoB5B,EAAuC,oHAUzC,IAPM,EAAgC,GAChC,EAA2C,GAE3C,EAAkB,MAAM,QAAQ,GAClC,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,OAC7B,OAAO,KAAK,cAEP,GACP,GAAM,GAAO,EAAM,GACb,EAAI,MAAM,QAAQ,GAAW,EAAQ,GAAG,OAAS,EAAQ,GAC/D,GAAI,EAAE,QAAU,WAAa,EAAE,QAAU,SAAW,EAAE,QAAU,QAC5D,EAAE,QAAU,UAAY,EAAE,QAAU,YACtC,KAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,EAAE,OAE9D,GAAM,GAA6B,CAAC,KAAI,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,OACnE,GAAI,EAAE,QAAU,UACd,GAAM,GAAY,GAAI,SAAoB,SAAM,GAAO,MAAA,IAAA,EAAA,OAAA,OAAA,+EACxC,MAAA,CAAA,EAAM,EAAE,gBAKrB,IALM,EAAO,EAAA,OACP,EAAgB,EAAK,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAE,QAAQ,GACtD,GAA0B,EAAK,OAC7B,EAAQ,GAAI,YAAW,GACzB,EAAS,EACJ,EAAI,EAAG,EAAI,EAAK,OAAQ,IACzB,EAAM,EAAK,GACX,EACF,GAAI,YAAW,GAAI,aAAY,CAAC,EAAI,SAAS,QACjD,EAAM,IAAI,EAAe,GACzB,GAAU,GACV,EAAM,IAAI,EAAK,GACf,GAAU,EAAI,OAEhB,SAAQ,cAEV,EAAa,KAAK,OAElB,GAAa,KAAK,EAAE,QAEtB,AAAI,GAAS,MACX,GAAK,MAAQ,GAEf,EAAM,KAAK,IAjCJ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,IAA3B,GAoCY,MAAA,CAAA,EAAM,QAAQ,IAAI,WAAjC,SAAe,EAAA,OACrB,CAAA,EAAO,CAAC,KAAM,GAAuB,GAAe,MAAK,qBAmBvD,EAAqB,GAKvB,OAHM,GAAsB,GACxB,EACA,EAAS,EACM,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAO,EAAK,KACZ,EAAQ,EAAK,MACb,EAAQ,EAAK,MACb,EAAO,GAAc,GACvB,EAAM,OAEV,GAAI,gBAAkB,IACpB,GAAM,GAAe,EAAK,aAC1B,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,GAAI,CAAE,QAAS,IAAgB,SAAW,IACxC,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,+DAGxD,EAAa,QAAU,WAChC,GAAI,IAAU,UACZ,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,IAC3D,oDAAmD,EAAK,UAG9D,MAAM,IAAI,OACN,UAAU,EAAK,KAAI,gBACnB,uBAAsB,EAAa,MAAK,MACxC,wEAGN,GAAM,GAAyB,GAAqB,EAAa,OAC3D,EACF,EAAO,MAAM,EAAQ,EAAS,EAAO,GACnC,EAAkB,EAAa,QAAU,QAC3C,GAAI,YAAW,GACf,GAAI,aAAY,GACpB,GAAI,IAAU,UACZ,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,EAAS,GAAI,cAAa,EAAe,QACzC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,EAAI,EAAa,MAAQ,EAAa,aAE3C,EAAa,QAAU,UAChC,AAAI,IAAkB,QACpB,GAAgB,MAElB,EAAS,EAAc,OAEvB,MAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,qCAGhD,IAAU,SACnB,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,SAC3D,KAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,2BAGzD,EAAS,GAAI,YAAW,EAAe,QACvC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,KAAK,MAAM,EAAI,EAAa,MAAQ,EAAa,UAG/D,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,UACR,IAAU,UACnB,GAAM,GAAO,GAAc,EAAK,OAChC,EAAS,GACT,OAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAa,GAAI,aACnB,EAAO,MAAM,EAAQ,EAAS,KAA0B,GAC5D,GAAU,GACV,GAAM,GAAQ,GAAI,YAAW,EAAO,MAAM,EAAQ,EAAS,IAC1D,EAAwB,KAAK,GAC9B,GAAU,QAGZ,GAAM,GAAc,GAAqB,GACnC,EAAa,EAAO,MAAM,EAAQ,EAAS,EAAO,GAExD,GAAI,IAAU,UACZ,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,aACnB,EAAS,GAAI,cAAa,GAG1B,OAFM,GAAO,GAAI,cAAa,EAAO,OAAS,GACxC,EAAQ,GAAI,cAAa,EAAO,OAAS,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAK,GAAK,EAAO,EAAI,GACrB,EAAM,GAAK,EAAO,EAAI,EAAI,GAE5B,GAAM,GAAa,GAAO,EAAM,EAAO,WACjC,EAAc,GAAO,EAAO,EAAO,WACzC,EAAI,GAAQ,GAAQ,EAAY,OAEhC,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,EAEnB,AAAI,IAAU,aACZ,GAAI,GAAQ,GAAO,EAAQ,EAAO,IAGtC,MAAO,eAM8B,GAErC,GAAI,IAAO,KACT,KAAM,IAAI,OAAM,wBAAwB,KAAK,UAAU,IAGzD,GAAI,GAAkB,EAShB,EAA6B,GACnC,EAAG,QAAQ,SAAC,GAMV,GALA,GAAmB,EAAE,WAErB,EAAa,KACT,EAAE,aAAe,EAAE,OAAO,WAAa,EACA,GAAK,GAAE,YAAoB,IAClE,CAAE,aAAoB,eAAgB,YAAoB,aACxD,YAAoB,aACxB,KAAM,IAAI,OAAM,mCAAmC,EAAE,YAAY,QAKrE,GAAM,GAAI,GAAI,YAAW,GACrB,EAAS,EACb,SAAa,QAAQ,SAAC,GACpB,EAAE,IAAI,GAAI,YAAW,EAAE,QAAS,GAChC,GAAU,EAAE,aAGP,EAAE,OAIX,GAAM,IAAgB,MAAO,SAAW,aACnC,OAAO,OAAS,aAAe,MAAO,OAAS,aAC/C,MAAO,OAAS,yBAWY,GAC/B,MAAI,IACK,OAAO,WAAW,GAEpB,GAAI,MAAK,CAAC,IAAM,iBASiB,GACxC,GAAI,GACF,MAAO,QAAO,KAAK,GAAQ,SAAS,UAItC,OAFM,GAAM,GAAI,YAAW,GACvB,EAAI,GACC,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAI,EAAG,IACrC,GAAK,OAAO,aAAa,EAAI,IAE/B,MAAO,MAAK,eAS4B,GACxC,GAAI,IACF,GAAM,GAAM,OAAO,KAAK,EAAK,UAC7B,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YAI/D,OAFM,GAAI,KAAK,GACT,EAAS,GAAI,YAAW,EAAE,QACvB,EAAI,EAAG,EAAI,EAAE,OAAQ,EAAE,EAC9B,EAAO,IAAI,CAAC,EAAE,WAAW,IAAK,GAEhC,MAAO,GAAO,mBASwB,GACtC,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAI,GAAkB,EACtB,EAAQ,QAAQ,SAAC,GACf,GAAmB,EAAO,aAG5B,GAAM,GAAO,GAAI,YAAW,GACxB,EAAS,EACb,SAAQ,QAAQ,SAAC,GACf,EAAK,IAAI,GAAI,YAAW,GAAS,GACjC,GAAU,EAAO,aAEZ,EAAK,mBAUW,GACvB,GAAM,GAAY,IAElB,IADA,EAAO,EAAK,OACL,EAAK,SAAS,IACnB,EAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAErC,GAAM,GAAQ,EAAK,MAAM,GACzB,MAAO,GAAM,EAAM,OAAS,eAQe,GAE3C,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OAAM,uDAGlB,MAAO,CACL,UAAW,GAAI,MACf,kBAAmB,OACnB,mBAAoB,EAAe,eAAiB,KAChD,EACA,GAAiB,KAAK,UAAU,EAAe,gBACnD,iBAAkB,EAAe,aAAe,KAC5C,EACA,GAAiB,KAAK,UAAU,EAAe,cACnD,gBAAiB,EAAe,YAAc,KAC1C,EACA,EAAe,WAAW,YAUlC,cACE,GAAM,GAAkB,SAAC,GAIvB,OAHI,GAAI,GAAK,GACT,EAAI,EAEA,GAAI,WAAgB,GAC1B,GAAK,QACL,IAAM,EAER,UAAK,CAAC,QACN,GAAK,UAEE,EAAI,GAGP,EAAe,GAAI,aAAY,MAErC,EAAa,GAAK,EAClB,OAAS,GAAI,EAAG,EAAI,KAAM,IACxB,EAAa,GAAK,EAAgB,GAEpC,OAAS,GAAI,KAAM,EAAI,KAAM,IAC3B,EAAa,GAAK,UAAe,GAAI,MAAS,IAGhD,MAAO,GAST,cACE,GAAM,GAAgB,GAAI,aAAY,IAEtC,EAAc,GAAK,EACnB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,OAAS,GAAI,EAAG,EAAI,GAAI,IACtB,EAAc,GAAK,GAAK,GAE1B,OAAS,GAAI,GAAI,EAAI,GAAI,IACvB,EAAc,GAAK,WAAe,GAAI,IAAO,IAG/C,MAAO,GAST,cAGE,OAFM,GAAc,GAAI,aAAY,IAE3B,EAAI,EAAG,EAAI,GAAI,IACtB,EAAY,GAAK,KAEnB,SAAY,GAAK,EAAY,IAAM,EAE5B,gBAeP,GAAM,GAAe,KACf,EAAgB,KAChB,EAAc,KAEpB,MAAO,UAAC,GAGN,OAFM,GAAS,GAAI,aAAY,EAAI,EAAe,QAC5C,EAAmB,GAAI,aAAY,GAChC,EAAQ,EAAG,EAAQ,EAAe,OAAQ,KACjD,GAAM,GAAc,EAAe,GAC7B,EACF,EAAa,EAAY,GAAe,IAAO,GAAc,OAC7D,EAAc,GAAe,IACjC,EAAiB,GAAS,EAE5B,MAAO,IAAI,cAAa,IChf5B,AAsBA,GAAA,IAAA,WAOE,aACE,KAAK,YAAc,GACnB,KAAK,YAAc,GAGN,SAAA,YAAf,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UASnB,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAS3C,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAW3C,EAAA,gBAAP,SAAuB,GACrB,MAAO,GAAiB,YAAY,EAAK,SAWpC,EAAA,gBAAP,SAAuB,EAAsB,GAE3C,MAAO,GAAiB,YAAY,EAAK,OAAQ,IAGpC,EAAA,YAAf,SACI,EAAsB,EACtB,GACF,GAAM,GAA6B,GAC7B,EAAU,IAAgB,OAC5B,EAAiB,cAAc,YAC/B,EAAiB,cAAc,YACnC,SAAQ,QAAQ,SAAA,GACd,GAAM,GAAU,EAAO,EAAK,GAC5B,AAAI,IAAY,MACd,EAAc,KAAK,KAGhB,GAEX,KAEa,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAkB,SAAC,GAC5B,MAAA,IAAiB,gBAAgB,IACxB,GACT,SAAC,EAAsB,GACnB,MAAA,IAAiB,gBAAgB,EAAK,IC/G9C,AAyBA,GAAM,IAAgB,eAChB,GAAmB,EAKnB,GAAmB,eAInB,GAAkB,mBAexB,cACE,GAAI,CAAC,KAAM,QAAQ,cAIjB,KAAM,IAAI,OACN,2FAIN,GAAM,GAAiB,MAAO,SAAW,YAAc,KAAO,OACxD,EAAU,EAAU,WAAa,EAAU,cAC7C,EAAU,iBAAmB,EAAU,aACvC,EAAU,cACd,GAAI,GAAW,KACb,KAAM,IAAI,OACN,6DAEN,MAAO,GAGT,YAAuB,GACrB,GAAM,GAAK,EAAY,OACvB,EAAG,kBAAkB,GAAkB,CAAC,QAAS,cACjD,EAAG,kBAAkB,GAAiB,CAAC,QAAS,cAQlD,GAAA,IAAA,WAME,WAAY,GAGV,GAFA,KAAK,UAAY,KAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,kEAEN,KAAK,UAAY,EAGb,SAAA,UAAA,KAAN,SAAW,sEAET,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAIN,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,UAAW,SAIvC,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,iBAiB1B,EAAA,UAAA,eAAR,SAAuB,EAAmB,GAA1C,GAAA,GAAA,KAEE,MAAO,IAAI,SAAmC,SAAC,EAAS,GACtD,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OAEvB,GAAI,GAAkB,MAEpB,GAAM,GAAU,EAAG,YAAY,GAAkB,YAC3C,EAAa,EAAQ,YAAY,IACjC,EAAa,EAAW,IAAI,EAAK,WACvC,EAAW,UAAY,WACrB,GAAI,EAAW,QAAU,KACvB,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAK,UAAS,oBAGlD,EAAQ,EAAW,OAAO,iBAG9B,EAAW,QAAU,SAAA,GACnB,SAAG,QACI,EAAO,EAAW,QAE3B,EAAQ,WAAa,WAAM,MAAA,GAAG,cAG9B,GAAM,GACF,GAA6B,GAE3B,EAAS,EAAG,YAAY,GAAiB,aAC3C,EAAY,EAAO,YAAY,IAC7B,EACF,EAAU,IAAI,CAAC,UAAW,EAAK,UAAW,mBAAkB,IAC5D,EACJ,EAAe,UAAY,WAEzB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAkB,EAAW,IAAI,CACrC,UAAW,EAAK,UAChB,eAAc,EACd,mBAAkB,IAEpB,EAAgB,UAAY,WAAM,MAAA,GAAQ,CAAC,mBAAkB,KAC7D,EAAgB,QAAU,SAAA,GAGxB,EAAY,EAAO,YAAY,IAC/B,GAAM,GAAoB,EAAU,OAAO,EAAK,WAChD,EAAkB,UAAY,WAC5B,SAAG,QACI,EAAO,EAAgB,QAEhC,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAgB,UAIpC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAE/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,YAKtC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,WAvHtC,EAAA,WAAa,eA0H/B,KAEa,GAA4B,SAAC,GACxC,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAmBH,GAC/B,MAAO,IAAI,IAAiB,GAG9B,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAiB,YACnC,EAAI,MAAM,GAAiB,WAAW,QACtC,EAGN,GAAA,IAAA,WAGE,aACE,KAAK,UAAY,KAGb,SAAA,UAAA,WAAN,yFACE,MAAA,CAAA,EAAO,GAAI,SACP,SAAC,EAAS,GACR,GAAM,GACF,EAAK,UAAU,KAAK,GAAe,IACvC,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAK,EAAG,YAAY,GAAiB,YACrC,EAAQ,EAAG,YAAY,IASvB,EAAqB,EAAc,SACzC,EAAkB,UAAY,WAE5B,OADM,GAA4C,GAC/B,EAAA,EAAA,EAAA,EAAkB,OAAlB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAI,EAAK,WAAa,EAAK,mBAE7B,EAAQ,IAEV,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAkB,QAElC,EAAG,WAAa,WAAM,MAAA,GAAG,UAE3B,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAIpD,EAAA,UAAA,YAAN,SAAkB,iFAChB,SAAO,GAAiB,GACxB,CAAA,EAAO,GAAI,SAA4B,SAAC,EAAS,GAC/C,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAS,EAAG,YAAY,GAAiB,aACzC,EAAY,EAAO,YAAY,IAE/B,EAAiB,EAAU,IAAI,GACjC,EACJ,EAAe,UAAY,WACzB,GAAI,EAAe,QAAU,KAC3B,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAI,oBAIxC,GAAM,GAAoB,EAAU,OAAO,GACrC,EAAkB,WAEtB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAqB,EAAW,OAAO,GAC7C,EAAmB,UAAY,WAC3B,MAAA,GAAQ,EAAe,OAAO,qBAClC,EAAmB,QAAU,SAAA,GACzB,MAAA,GAAO,EAAe,SAI5B,EAAkB,UAAY,EAC9B,EAAkB,QAAU,SAAA,GAC1B,WACA,EAAG,QACI,EAAO,EAAe,SAInC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAG/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,WAIpC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAGxD,KCjWA,AAyBA,GAAM,IAAiB,IACjB,GAAc,sBACd,GAAc,OACd,GAAwB,iBACxB,GAAsB,eACtB,GAAqB,cACrB,GAAwB,iBA8B9B,YAAsB,GAOpB,MAAO,CACL,KAAM,CAAC,GAAa,EAAM,IAAa,KAAK,IAC5C,SAAU,CAAC,GAAa,EAAM,IAAuB,KAAK,IAC1D,YAAa,CAAC,GAAa,EAAM,IAAqB,KAAK,IAC3D,WAAY,CAAC,GAAa,EAAM,IAAoB,KAAK,IACzD,cACI,CAAC,GAAa,EAAM,IAAuB,KAAK,KAWxD,YAA6B,GAC3B,GAAM,GAAQ,EAAI,MAAM,IACxB,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OAAM,uBAAuB,GAEzC,MAAO,GAAM,MAAM,EAAG,EAAM,OAAS,GAAG,KAAK,IAG/C,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAoB,YACtC,EAAI,MAAM,GAAoB,WAAW,QACzC,EAgBN,GAAA,IAAA,WAOE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,eAAiB,MAAO,SAAW,aAClD,MAAO,QAAO,cAAiB,YAKjC,KAAM,IAAI,OACN,2DAIN,GAFA,KAAK,GAAK,OAAO,aAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,sEAEN,KAAK,UAAY,EACjB,KAAK,KAAO,GAAa,KAAK,WAY1B,SAAA,UAAA,KAAN,SAAW,gFACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAGE,EAAW,KAAK,UAAU,EAAe,eACzC,EAAc,KAAK,UAAU,EAAe,aAE5C,EACF,GAA6B,GAEjC,IACE,YAAK,GAAG,QAAQ,KAAK,KAAK,KAAM,KAAK,UAAU,IAC/C,KAAK,GAAG,QAAQ,KAAK,KAAK,SAAU,GACpC,KAAK,GAAG,QAAQ,KAAK,KAAK,YAAa,GACvC,KAAK,GAAG,QACJ,KAAK,KAAK,WACV,GAA0B,EAAe,aAC7C,KAAK,GAAG,QAAQ,KAAK,KAAK,cAAe,KAAK,UAAU,CACtD,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,uBAGtC,CAAA,EAAO,CAAC,mBAAkB,UACnB,GAEP,WAAK,GAAG,WAAW,KAAK,KAAK,MAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,UAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,aAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,YAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,eAEvB,GAAI,OACN,yBAAyB,KAAK,UAAS,sFAEvC,uBAAsB,EAAmB,mBAAkB,MAC3D,qBAAoB,EAAmB,iBAAgB,MACvD,oBAAmB,EAAmB,gBAAe,qBAazD,EAAA,UAAA,KAAN,gGAGE,GAFM,EACF,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,OACrC,GAAQ,KACV,KAAM,IAAI,OACN,kDAAkD,KAAK,UAAS,KAGtE,GAAI,EAAK,oBAAsB,OAC7B,KAAM,IAAI,OACN,6EAQN,GAJM,EAAsB,GAGtB,EAAW,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,WAClD,GAAY,KACd,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAS,iBAOhE,GAJA,EAAI,cAAgB,EAGd,EAAc,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,cACrD,GAAe,KACjB,KAAM,IAAI,OACN,gDAAgD,KAAK,UAAS,kBAiBpE,GAdA,EAAI,YAAc,EAGZ,EAAiB,KAAK,GAAG,QAAQ,KAAK,KAAK,eACjD,AAAI,GAAkB,MACd,GAAW,KAAK,MAAM,GAC5B,EAAI,OAAS,EAAS,OACtB,EAAI,YAAc,EAAS,YAC3B,EAAI,YAAc,EAAS,YAC3B,EAAI,oBAAsB,EAAS,qBAI/B,EAAmB,KAAK,GAAG,QAAQ,KAAK,KAAK,YAC/C,GAAoB,KACtB,KAAM,IAAI,OACN,uDACA,KAAI,KAAK,UAAS,mBAExB,SAAI,WAAa,GAA0B,GAE3C,CAAA,EAAO,QAzIO,EAAA,WAAa,kBA2I/B,KAEa,GAA+B,SAAC,GAC3C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAoB,aACrD,GACH,EAAI,MAAM,GAAoB,WAAW,SAJxC,MAUX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBA0BA,GAClC,MAAO,IAAI,IAAoB,GAGjC,GAAA,IAAA,WAGE,aACE,EACI,KAAM,QAAQ,cACd,WAAM,MAAA,6CACV,EACI,MAAO,SAAW,aACd,MAAO,QAAO,cAAiB,YACnC,WAAM,MAAA,4DACV,KAAK,GAAK,OAAO,aAGb,SAAA,UAAA,WAAN,8FAIE,IAHM,EAA4C,GAC5C,EAAS,GAAc,GACvB,EAAS,GAAiB,GACvB,EAAI,EAAG,EAAI,KAAK,GAAG,OAAQ,EAAE,EAC9B,EAAM,KAAK,GAAG,IAAI,GACxB,AAAI,EAAI,WAAW,IAAW,EAAI,SAAS,IACnC,GAAY,GAAoB,GACtC,EAAI,GAAa,KAAK,MAAM,KAAK,GAAG,QAAQ,KAGhD,MAAA,CAAA,EAAO,QAGH,EAAA,UAAA,YAAN,SAAkB,8EAGhB,GAFA,EAAO,GAAiB,GAClB,EAAO,GAAa,GACtB,KAAK,GAAG,QAAQ,EAAK,OAAS,KAChC,KAAM,IAAI,OAAM,8BAA8B,EAAI,KAE9C,SAAO,KAAK,MAAM,KAAK,GAAG,QAAQ,EAAK,OAE7C,KAAK,GAAG,WAAW,EAAK,MACxB,KAAK,GAAG,WAAW,EAAK,UACxB,KAAK,GAAG,WAAW,EAAK,aACxB,KAAK,GAAG,WAAW,EAAK,YACxB,CAAA,EAAO,QAEX,KCvVA,AAiCA,GAAM,IAAoB,MAE1B,GAAA,WAME,aACE,KAAK,SAAW,GAGH,SAAA,YAAf,WACE,MAAI,GAA0B,UAAY,MACxC,GAA0B,SAAW,GAAI,IAEpC,EAA0B,UAS5B,EAAA,gBAAP,SAAuB,EAAgB,GACrC,EAAO,GAAU,KAAM,WAAM,MAAA,0CAC7B,AAAI,EAAO,SAAS,KAClB,GAAS,EAAO,MAAM,EAAG,EAAO,QAAQ,MAE1C,EAAO,EAAO,OAAS,EAAG,WAAM,MAAA,wCAChC,GAAM,GAAW,EAA0B,cAC3C,EACI,EAAS,SAAS,IAAW,KAC7B,WAAM,MAAA,2DACF,EAAM,OACd,EAAS,SAAS,GAAU,GAGvB,EAAA,WAAP,SAAkB,GAChB,GAAM,GAAU,KAAK,cAAc,SAAS,GAC5C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,yCAAyC,EAAM,KAEjE,MAAO,IAGF,EAAA,WAAP,WACE,MAAO,QAAO,KAAK,KAAK,cAAc,WAE1C,KAUA,YAAkB,GAChB,GAAI,EAAI,QAAQ,MAAuB,GACrC,KAAM,IAAI,OACN,6EAEA,IAAG,GAA0B,aAAa,KAAK,OAErD,MAAO,CACL,OAAQ,EAAI,MAAM,IAAmB,GACrC,KAAM,EAAI,MAAM,IAAmB,IAIvC,YACI,EAAmB,EACnB,GAAA,MAAA,KAAA,QAAA,GAAA,6GACF,SACI,IAAc,EACd,WAAM,MAAA,wCAAwC,EAAS,MAErD,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,kEACF,EAAS,MACjB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,iCAAgC,EAAS,OAC3C,EAAc,EAAa,GAE3B,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,mEACF,QAAO,EAAO,OACtB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,sCAAqC,EAAO,OAC9C,EAAc,EAAa,GAE3B,EAAe,GAAS,GAAW,OACnC,EAAa,GAAS,GAAW,KACjC,EAAa,IAAiB,GAAS,GAAW,OAEjC,CAAA,EAAM,EAAY,qBAAnC,GAAiB,EAAA,OAKnB,GAAgB,EAClB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIiB,MAAA,CAAA,EAAM,EAAY,KAAK,iBAApC,GAAa,EAAA,OAKf,GAAgB,CAAC,EACnB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIF,MAAA,CAAA,EAAO,EAAW,yBAsCpB,4HACQ,EAAU,GAA0B,aACpC,EAA2C,OAC5B,EAAA,yBAAA,GAAA,EAAA,OAAV,GAAM,EAAA,GAEX,CAAA,EAAM,GAA0B,WAAW,GAAQ,eAF7B,CAAA,EAAA,UACpB,EACF,EAAA,OACJ,IAAW,IAAQ,GACX,EAAM,EAAS,GAAoB,EACzC,EAAI,GAAO,EAAU,oBALJ,wBAQrB,MAAA,CAAA,EAAO,QAoCT,YAA2B,8EACnB,SAAgB,GAAS,GACzB,EAAU,GAA0B,WAAW,EAAc,QACnE,CAAA,EAAO,EAAQ,YAAY,EAAc,WAkD3C,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QAiDhD,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QChWhD,AA0BA,GAAA,IAAA,WAAA,cAKE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,OAAM,EAAM,IAGrB,EAAA,UAAA,IAAA,WACE,MAAO,aAAY,OAGrB,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,kDAAkD,GAExD,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAAI,cAElB,KAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,IAAI,aAAY,GAAU,OAAO,IAE5C,KAEA,GAAI,KAAM,IAAI,eACZ,KAAM,YAAY,UAAW,GAAI,KAGjC,IACE,GAA0B,gBACtB,GAAoB,WAAY,GAAI,WACjC,IAIT,IACE,GAA0B,gBACtB,GAAiB,WAAY,GAAI,WAC9B,KCpEX,AAqBO,GAAM,IAAe,CAE1B,YAAa,WAAM,MAAA,QAIjB,GAaJ,GAAA,WAKE,aAEE,KAAK,KAAO,gBAGZ,KAAK,YAAc,GAAI,MAAK,KAAK,YAGnC,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAI,MAAM,OAAO,OAAS,KACjB,KAAM,OAAO,MAAM,EAAM,GAGlC,CAAI,IAAe,MACjB,IAAc,GAAa,eAEtB,GAAY,EAAM,KAG3B,EAAA,UAAA,IAAA,WACE,GAAM,GAAO,QAAQ,SACrB,MAAO,GAAK,GAAK,IAAO,EAAK,GAAK,KAGpC,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,sDAAsD,GAE5D,MAAO,MAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAI,GAAM,SAAW,EACZ,GAEF,GAAI,MAAK,KAAK,YAAY,GAAU,OAAO,IAEtD,KAEA,AAAI,KAAM,IAAI,YACZ,KAAM,YAAY,OAAQ,GAAI,KCrFhC,YAgDI,EAAoB,EACpB,GADoB,MAAA,KAAA,QAAA,GAAW,WAEjC,EAAQ,GAAS,UACjB,GAAwC,GACjC,GAAI,IAAmB,EAAO,EAAO,GCpD9C,AAuCA,YAAiC,EAAiB,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAGnC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,mCAAmC,GAErD,GAAI,IAAU,UAAY,EAAG,QAAU,UACnC,IAAU,UAAY,EAAG,QAAU,SACrC,KAAM,IAAI,OAAM,yCAGlB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAGhB,IAAO,EAAG,CAAC,MAAK,KC3D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,QAAS,MACtC,EAAU,WACZ,MAAA,GAAO,qBAAqB,EAAG,OAAQ,EAAG,MAAO,EAAG,QAElD,EAAyB,CAAC,EAAG,GAInC,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAQ,EAAG,CAAC,OAAM,KCrD/B,YAgCwC,EAAM,GAAA,AAAA,IAAA,QAAA,GAAA,IAC5C,QAAQ,IAAI,EAAE,SAAS,ICjCzB,AAqBA,KAcA,GAAM,IAAuB,CAC3B,OAAM,GACN,KAAI,GACJ,MAAK,GACL,MAAK,IAEP,GAAa,ICzCb,AA6BA,GAAM,IAA2B,QAC3B,GAA8B,QAC9B,GAAqC,eAE3C,YAAkB,GAChB,MAAO,IAAI,SAAQ,SAAA,GAAW,MAAA,YAAW,KAAU,KAAK,GAG1D,GAAA,IAAA,WAQE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,cAGjB,KAAM,IAAI,OACN,uFAIN,AAAI,EAAe,WAAW,EAAiB,aAC7C,GAAiB,EAAe,MAAM,EAAiB,WAAW,SAEpE,AAAI,IAAkB,MAAQ,EAAe,SAAW,IACtD,GAAiB,IAGnB,KAAK,sBAAwB,EAAiB,GAC9C,KAAK,mBACD,EAAiB,GAGjB,SAAA,UAAA,KAAN,SAAW,6GACT,GAAI,MAAQ,WAAc,YACxB,KAAM,IAAI,OACN,8FAGA,EAAa,OAAO,IAAI,gBAAgB,GAAI,MAC9C,CAAC,EAAe,YAAa,CAAC,KAAM,+BAEpC,GAAe,wBAAyB,cAAxC,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,OACN,gGAGE,SAAyC,CAAC,CAC9C,MAAO,CAAC,KAAO,KAAK,oBACpB,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,gBAAe,GAEX,EACF,OAAO,IAAI,gBAAgB,GAAI,MAC3B,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,sBAIT,EAAa,KAAK,YAAc,KAAO,SAAS,cAAc,KACvB,KAAK,WAClD,EAAW,SAAW,KAAK,sBAC3B,EAAW,KAAO,EAIlB,CAAA,EAAM,GAAM,WAAM,MAAA,GAAW,cAAc,GAAI,YAAW,0BAA1D,GAAA,OAEI,EAAe,YAAc,KACzB,GAAmB,KAAK,kBAAoB,KAC9C,SAAS,cAAc,KACvB,KAAK,iBACT,EAAiB,SAAW,KAAK,mBACjC,EAAiB,KAAO,EACxB,CAAA,EAAM,GACF,WAAM,MAAA,GAAiB,cAAc,GAAI,YAAW,cAPtD,CAAA,EAAA,UAMF,EAAA,wBAIF,MAAA,CAAA,EAAO,CAAC,mBAAoB,GAA6B,WA1E7C,EAAA,WAAa,eA6E/B,KAEA,GAAA,WAGE,WAAY,GACV,GAAI,GAAS,MAAQ,EAAM,OAAS,EAClC,KAAM,IAAI,OACN,2DACA,iBAAgB,IAEtB,KAAK,MAAQ,EAGT,SAAA,UAAA,KAAN,6FACQ,SAAW,KAAK,MAAM,GACtB,EAAc,KAAK,MAAM,MAAM,GAErC,CAAA,EAAO,GAAI,SAAwB,SAAC,EAAS,GAC3C,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GAEnB,GAAM,GAAY,KAAK,MAAO,EAAM,OAAe,QAC7C,EAAgB,EAAU,cAChC,GAAI,GAAiB,MACnB,EAAO,GAAI,OACP,4CAA4C,EAAS,OACzD,OAGF,AAAI,EAAY,SAAW,GACzB,EAAQ,CAAC,cAAa,IAGxB,GAAM,GAAkB,EAAU,gBAClC,GAAI,GAAmB,MACrB,EAAO,GAAI,OACP,6CAA6C,EAAS,OAC1D,OAGF,GAAI,GACJ,IACE,EACI,EAAK,4BAA4B,EAAiB,SAC/C,GACP,EAAO,GACP,OAGF,GAAM,GAAsC,GACtC,EAAkB,GAClB,EAAgC,GACtC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,EAAM,KAAK,GACX,EAAe,KAAK,QAEtB,EAAY,KAAI,MAAhB,EAAoB,EAAa,WAGnC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,GAAM,GAAmB,GAAI,YAC7B,EAAiB,OAAS,SAAC,GAEzB,GAAM,GAAc,EAAM,OAAe,OACnC,EAAQ,EAAM,QAAQ,GAC5B,EAAe,GAAS,EACxB,AAAI,EAAe,QAAQ,QAAU,IACnC,EAAQ,CACN,cAAa,EACb,YAAW,EACX,WAAY,GAAwB,GACpC,OAAQ,EAAU,OAClB,YAAa,EAAU,YACvB,YAAa,EAAU,YACvB,oBAAqB,EAAU,uBAIrC,EAAiB,QAAU,SAAA,GACvB,MAAA,GAAO,6CAA6C,EAAI,OAC5D,EAAiB,kBAAkB,EAAW,SAIpD,EAAW,QAAU,SAAA,GAAS,MAAA,GAC1B,2DACA,eAAc,EAAS,KAAI,qCAC3B,yCACJ,EAAW,WAAW,WAOlB,EAAA,UAAA,4BAAR,SACI,EAAiC,GAInC,OAHM,GAAsB,GACtB,EAAY,EAAM,IAAI,SAAA,GAAQ,MAAA,IAAS,EAAK,QAC5C,EAAqC,GACvB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAM,MAAM,QAAQ,SAAA,GAClB,GAAM,GAAe,GAAS,GAC9B,GAAI,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,sDACA,KAAI,EAAY,MAGtB,GADA,EAAU,KAAK,GACX,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,8BAA8B,EAAY,sBAE9C,EAAW,GAAQ,EAAM,EAAU,QAAQ,MAKjD,GAAI,EAAU,SAAW,EAAM,OAC7B,KAAM,IAAI,OACN,uDACA,KAAI,EAAU,OAAM,8CACpB,KAAI,EAAM,OAAM,OAEtB,MAAO,IAEX,KAEa,GAAmC,SAAC,GAC/C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,gBAyCH,GAAA,MAAA,KAAA,QAAA,GAAA,SACxB,GAAI,IAAiB,eA2CD,GAC3B,MAAO,IAAI,IAAa,GC5V1B,YA8BI,EAAmC,EACnC,EAAwB,GAC1B,EAAc,GACd,EAAgB,GAAwB,EACxC,EAAc,GAAsB,EACpC,EAAc,EAAe,GAC7B,GAAI,GAAkB,EAEhB,EAAkB,SAAC,GACvB,SAAQ,KAAK,SAAA,GACX,GAAM,GAAW,EACb,GAAE,EAAkB,EAAS,OAAU,GAAc,GAEzD,SAAW,GACJ,IAEF,GAGT,WAAuB,GACrB,EACI,GAAY,MAAQ,MAAM,QAAQ,IAAa,EAAS,OAAS,EACjE,WAAM,MAAA,wCAGZ,WAAuB,EAAuB,GAC5C,EACI,GAAiB,GAAK,GAAiB,EACvC,WAAM,MAAA,kDACF,sBAAqB,KAC7B,EACI,GAAe,GAAK,GAAe,EACnC,WAAM,MAAA,kDACF,oBAAmB,KAC3B,EACI,GAAe,EACf,WAAM,MAAA,uDACF,sBAAqB,EAAa,qBAClC,IAAG,KAGb,MAAO,SAAQ,IAAI,EAAS,IAAI,ICvElC,YAqCI,EAAqB,6HACvB,AAAI,IAAe,MACjB,GAAc,IAGV,EAAY,EAAY,WAAa,KAAO,KAAM,SAAS,MACf,EAAY,UAGxD,EAAW,EAAU,IACvB,SAAA,GACI,MAAA,GAAU,EAAU,EAAY,YAAa,CAAC,SAAU,OAE1D,EAAqB,EACrB,EAAmB,GAEP,EAAY,YAAc,KACxC,CAAA,EAAM,QAAQ,IAAI,IADJ,CAAA,EAAA,UACd,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAU,EAAY,WAAY,EAClC,WAFJ,EAAA,EAAA,8BAFE,GAAS,EAMT,EAAiB,EAAU,IAAI,SAAA,GAAY,MAAA,GAAS,gBAEpD,EAAsB,GACtB,EAAoB,EAEV,EAAY,YAAc,KACtC,CAAA,EAAM,QAAQ,IAAI,IADN,CAAA,EAAA,UACZ,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAgB,EAAY,WAAY,EACxC,WAFJ,EAAA,EAAA,wBAFE,SAAO,EAKb,CAAA,EAAO,oBAaL,EAAiC,EACjC,EACA,GAFiC,MAAA,KAAA,QAAA,GAAA,wEAS7B,SAAe,SAAC,GAClB,MAAA,IAAyB,EAAW,CAAC,YAAW,KAC9C,EAAc,GAAqB,GAEzC,CAAA,EAAO,EAAY,EAAU,EAAgB,oBA4B3C,GADJ,GAAA,GAAA,KAIE,MAAO,UACI,EAAiC,EACjC,GADiC,MAAA,KAAA,QAAA,GAAA,4GAqD1C,GAjDM,EAAyB,EAAS,IAAI,WAAM,MAAA,KAC5C,EAKF,GACE,EACF,GAAe,KAAO,EAAY,IAAI,WAAM,MAAA,KAAS,GACnD,EAAmC,GACzC,EAAS,QAAQ,SAAC,EAAqB,GACrC,GAAI,GAAc,EAClB,EAAoB,QAAQ,QAAQ,SAAA,GAClC,GAAM,GAAY,gBAAkB,GAChC,EAAa,aAAa,MAC1B,EAAa,MAEX,EAAe,GAAqB,GACtC,GAAmB,EAAa,OAE9B,EAA8B,WAClC,EAAuB,GAAc,GACrC,AAAI,EAAoB,IAAe,MACrC,GAAoB,GAAc,IAGpC,EAAoB,GAAY,KAAK,CACnC,cAAe,EACf,YAAW,EACX,UAAW,KAIf,AAAI,GAAe,KACjB,EAAY,QAAQ,SAAC,EAAY,GAC/B,AAAI,IAAe,EAAa,MAC9B,KACA,EAAa,GAAe,MAIhC,IAGF,EAAuB,KAAK,EAAa,MACzC,GAAe,MAIf,CAAC,EAAa,MAAM,SAAA,GAAS,MAAA,KACzB,QAAkB,EAAY,OAAO,SAAC,EAAG,GAAM,MAAA,CAAC,EAAa,KAC7D,GAAI,OACN,kDACG,GAAgB,KAAK,MAAK;GAC7B,yCACG,GAAuB,KAAK,MAAK,MAKpC,SACF,EAAuB,OAAO,SAAC,EAAa,EAAa,GACvD,MAAI,IACF,EAAY,KAAK,GAEZ,GACN,IAED,EAAsB,GAC5B,EAAoB,QAAQ,SAAA,GAC1B,EAAS,GAAG,MAAM,QAAQ,SAAA,GACxB,GAAM,GAAW,EACZ,CAAC,EAAe,SAAS,KAAa,GAAN,KAAY,EACjD,EAAU,KAAK,OAGH,CAAA,EAAM,EAAqB,WAArC,SAAU,EAAA,OAEV,EAAmC,GACrC,EAAoB,EACxB,EAAoB,QAAQ,SAAA,GAI1B,OAHM,GAAa,EAAS,GAAG,MAAM,OAEjC,EAAa,EACR,EAAI,EAAG,EAAI,EAAY,IAC9B,GAAc,EAAQ,EAAoB,GAAG,WAO/C,OAHM,GAAc,GAAI,aAAY,GAC9B,EAAkB,GAAI,YAAW,GACnC,EAAoB,EACf,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAS,GAAI,YAAW,EAAQ,EAAoB,IAC1D,EAAgB,IAAI,EAAQ,GAC5B,GAAqB,EAAO,WAG9B,GAAM,GAAiB,EAAoB,GAC3C,EAAe,QAAQ,SAAA,GACrB,GAAM,GAAa,EAAY,MAC3B,EAAa,YACb,EAAa,YAAc,EAAa,WACtC,EACF,GAAc,EAAY,CAAC,EAAa,gBAC5C,OAAW,KAAQ,GACjB,EAAiB,GAAQ,EAAgB,KAI7C,GAAqB,IAGvB,CAAA,EAAO,SCpPX,AA+BA,GAAM,IAAyB,2BACzB,GAAY,mBAClB,GAAA,WAcE,WAAY,EAAc,GAgCxB,GAvCO,KAAA,eAAiB,OAQxB,AAAI,GAAe,MACjB,GAAc,IAEhB,KAAK,iBAAmB,EAAY,iBACpC,KAAK,WAAa,EAAY,WAC9B,KAAK,mBAAqB,EAAY,mBAEtC,AAAI,EAAY,WAAa,KAC3B,GACI,MAAO,GAAY,WAAc,WACjC,WAAM,MAAA,gIAGV,KAAK,MAAQ,EAAY,WAEzB,KAAK,MAAQ,KAAM,SAAS,MAG9B,EACI,GAAQ,MAAQ,EAAK,OAAS,EAC9B,WAAM,MAAA,4DAGV,AAAI,MAAM,QAAQ,IAChB,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,+CACF,sBAAqB,EAAK,OAAM,QAE1C,KAAK,KAAO,EAER,EAAY,aAAe,MAC3B,EAAY,YAAY,MAAQ,KAClC,KAAM,IAAI,OACN,sEAEN,KAAK,YAAc,EAAY,aAAe,GAG1C,SAAA,UAAA,KAAN,SAAW,yGACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,2FAIA,SAAO,OAAO,OAAO,CAAC,OAAQ,KAAK,gBAAiB,KAAK,aAC/D,EAAK,KAAO,GAAI,UAEV,EAAyC,CAAC,CAC9C,MAAO,CAAC,uBACR,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,oBACpC,gBAAe,GAGjB,EAAK,KAAK,OACN,aACA,GAAI,MACA,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,KACX,cAEJ,AAAI,EAAe,YAAc,MAC/B,EAAK,KAAK,OACN,oBACA,GAAI,MAAK,CAAC,EAAe,YAAa,CAAC,KAAM,KAC7C,qBAGW,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,WAE7C,GAFM,EAAW,EAAA,OAEb,EAAS,GACX,MAAA,CAAA,EAAO,CACL,mBAAoB,GAA6B,GACjD,UAAW,CAAC,KAGd,KAAM,IAAI,OACN,gEACG,GAAS,OAAM,YAYpB,EAAA,UAAA,KAAN,uIAC6B,MAAA,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,KAAK,qBAE5D,GAFM,EAAqB,EAAA,OAEvB,CAAC,EAAmB,GACtB,KAAM,IAAI,OACN,cAAc,KAAK,KAAI,4BACpB,GAAmB,OAAM,uCAC5B,sFAIU,CAAA,EAAM,EAAmB,eAAvC,SAAc,EAAA,qCAEV,EAAU,+CAA+C,KAAK,KAAI,IAGtE,AAAI,KAAK,KAAK,SAAS,OACrB,GAAW,+UAOX,GAAW,uEAGP,GAAI,OAAM,UAUlB,GARM,EAAgB,EAAY,cAC5B,EAAkB,EAAY,gBAC9B,EAAc,EAAY,YAC1B,EAAc,EAAY,YAC1B,EAAS,EAAY,OACrB,EAAsB,EAAY,oBAGpC,GAAiB,MAAQ,GAAmB,KAC9C,KAAM,IAAI,OACN,2BAA2B,KAAK,KAAI,mEAMtC,IAAmB,KACL,CAAA,EAAM,KAAK,YAAY,IADrC,CAAA,EAAA,UACI,EAAU,EAAA,OACf,EAAA,EAAA,GAAa,EAAA,EAAA,oBAGV,SAA4B,CAChC,cAAa,EACb,YAAW,EACX,WAAU,EACV,oBAAmB,EACnB,YAAW,EACX,YAAW,EACX,OAAM,GAGF,EAAc,EAAY,iBAChC,AAAI,GACF,GAAU,iBAAmB,GAG/B,CAAA,EAAO,SAGK,EAAA,UAAA,YAAd,SAA0B,2IAOxB,IALM,EAAa,MAAM,QAAQ,KAAK,MAAQ,KAAK,KAAK,GAAK,KAAK,KAC5D,EAAmB,GAAS,GAA3B,EAAM,EAAA,GAAE,EAAM,EAAA,GACf,EAAa,KAAK,kBAAoB,EAEtC,EAAc,GACpB,EAAA,EAAoB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAT,EAAK,EAAA,GACd,EAAY,KAAI,MAAhB,EAAoB,EAAM,SAK5B,IAFM,EAAsB,GACtB,EAAsC,GAC5C,EAAA,EAA2B,EAAA,EAAA,EAAA,EAAA,OAAA,IACzB,IADS,EAAY,EAAA,GACrB,EAAA,EAAmB,EAAA,EAAa,MAAb,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,AAAI,KAAK,oBAAsB,KAC7B,EAAY,KAAK,KAAK,mBAAmB,IAEzC,EAAU,KAAK,EAAa,EAAO,SAKrC,MAAK,sBACP,GAAA,EAAU,MAAI,SAAd,GAAkB,CAAA,EAAM,QAAQ,IAAI,KADlC,CAAA,EAAA,UACF,EAAA,MAAA,EAAA,EAAA,OAAA,CAAkB,EAAA,2BAGJ,MAAA,CAAA,EAAM,GAAyB,EAAW,CACxD,YAAa,KAAK,YAClB,UAAW,KAAK,MAChB,WAAY,KAAK,qBAHb,SAAU,EAAA,OAKhB,CAAA,EAAO,CAAC,EAAa,GAAwB,WA5M/B,EAAA,iBAAmB,eA8MrC,iBAayB,GACvB,GAAM,GAAY,EAAI,YAAY,KAC5B,EAAkB,EAAI,YAAY,KAClC,EAAS,EAAI,UAAU,EAAG,GAC1B,EACF,EAAkB,EAAY,EAAI,UAAU,GAAmB,GACnE,MAAO,CAAC,EAAS,IAAK,eAGK,GAC3B,MAAO,GAAI,MAAM,GAAY,mBAAqB,KAG7C,GAAM,IACT,SAAC,EAAa,GACZ,GAAI,MAAO,QAAU,aAChB,IAAe,MAAQ,EAAY,WAAa,MAInD,MAAO,MAEP,GAAI,GAAS,GAMb,MALA,AAAI,OAAM,QAAQ,GAChB,EAAS,EAAI,MAAM,SAAA,GAAW,MAAA,IAAa,KAE3C,EAAS,GAAa,GAEpB,EACK,GAAK,EAAK,GAGd,MAEb,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAwEf,EAAc,GACjC,MAAO,IAAI,IAAY,EAAM,eAS3B,EAAc,GAChB,MAAO,IAAK,EAAM,GC3XpB,AAuBA,GAAA,IAAA,WACE,WAA6B,GAAA,KAAA,eAAA,EAEvB,SAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,qBAEhB,KAEA,GAAA,WACE,WACqB,GAAA,KAAA,YAAA,EAGf,SAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,YAAY,SAE5B,iBAwBI,EAAmC,EACnC,EAA0B,GAC5B,GAAI,UAAU,SAAW,GACvB,GAAM,GACD,EAAkC,eAAiB,MACnD,EAAkC,aAAe,KACtD,MAAI,GACK,GAAI,IAAkB,GAI7B,SAAQ,KACJ,yNAIG,GAAI,IAAkB,CAAC,cAAe,SAK/C,gBAAQ,KACJ,yNAIG,GAAI,IAAkB,CAC3B,cAAe,EACf,YAAW,EACX,WAAU,EACV,eAAc,gBAqBhB,GAEF,MAAO,IAAI,IAAiB,GCpH9B,oYCAA,AAuDA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MAExC,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,MAAK,GAC5B,EACW,SAAC,EAAwB,GACxC,SAAQ,GAA4B,EAAO,EAAG,MAC9C,EACI,EAAG,OAAS,GAAmB,GAC/B,WAAM,MAAA,mEACV,EAAK,CAAC,IACC,EAAQ,QAAQ,EAAI,IAE7B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAEO,GAAU,EAAG,CAAC,SAAQ,KC1EnC,AA6CA,YACI,EAAiB,EAAiB,EAClC,SADkC,AAAA,IAAA,QAAA,GAAA,IAClC,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAK,EAAgB,EAAG,IAAK,UACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WAAM,MAAA,kEACF,cAAa,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE3C,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,sCAAsC,EAAU,UAC/C,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,kCAAkC,EAAW,UAC5C,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACxD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAExD,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,IAEJ,EAAQ,YACX,EAAiB,EAAiB,EAAY,IAG9C,EAA4B,CAAC,EAAG,EAAK,EAAG,GAExC,EAA0B,CAAC,WAAU,EAAE,WAAU,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAO,GAAQ,EAAK,MAGT,IAAS,EAAG,CAAC,QAAO,KC9GjC,AA+CA,YACI,EAA4B,EAAe,EAC3C,GACF,GAF6C,AAAA,IAAA,QAAA,GAAA,GAC3C,AAAA,IAAA,QAAA,GAAA,GACE,EAAQ,EACV,KAAM,IAAI,OAAM,iDAAiD,GAEnE,GAAM,GAAW,EAAgB,EAAS,UAAW,SAAU,SACzD,EAAe,EAAS,MAAK,OAAA,CAAE,IAE/B,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,IACC,EACH,EAAQ,OACJ,EAAQ,EAAU,CAAC,EAAS,OAAQ,EAAO,EAAS,GACxD,IAGA,EAAuB,CAAC,QAAS,GACjC,EAAqB,CAAC,MAAK,EAAE,QAAO,EAAE,SAAQ,GAEpD,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAS,EAAG,CAAC,QAAO,KCxEjC,AA+CA,YAAsC,EAAiB,GACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAgBnC,GAdA,AAAI,GAAQ,MACV,GAAO,EAAG,MAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAG,WAEnC,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OACzC,EAAK,QAAQ,SAAA,GACX,EACI,GAAQ,GAAK,EAAO,EAAG,KACvB,WAAM,MAAA,+CAA+C,GAAG,KAAO,GAC3D,aAAY,OAGlB,EAAG,MAAQ,EACb,MAAO,GAAG,QAGZ,GAAM,GAA0B,CAAC,EAAG,GAC9B,EAAwB,CAAC,KAAI,GAEnC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IAAO,EACxC,KAAqB,GAAW,MAGzB,IAAY,EAAG,CAAC,WAAU,KC5EvC,YA0DI,EAA6B,EAC7B,GACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,mBAC5C,EACF,EAAgB,EAAa,cAAe,mBAEhD,EACI,GAAc,MAAQ,EAAa,GAAK,OAAO,UAAU,GACzD,WAAM,MAAA,uDACF,YAAW,KACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,gDAAgD,EAAQ,OAClE,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,6CACF,YAAW,EAAa,QAChC,EACI,EAAQ,MAAM,KAAO,EAAa,MAAM,GACxC,WAAM,MAAA,uCACC,GAAQ,MAAM,GAAE,QAAQ,EAAa,MAAM,GAAE,MAChD,oEACR,EACI,EAAa,GAAK,OAAO,UAAU,GACnC,WAAM,MAAA,4DACF,IAAG,KAIX,GAAM,GAAe,GAAO,GAAK,EAAS,SAAU,GAC9C,EACF,GAAO,GAAK,EAAc,SAAU,GAClC,EAA0B,GAAU,GAC1C,MAAO,IAAK,GAAO,EAAe,GAAoB,SAGjD,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KC9FnD,2CCAA,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,mDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,oEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,AA6BA,GAAI,IAyBJ,YACI,EAEA,GAEF,GAFE,AAAA,IAAA,QAAA,GAAA,GAEE,EAAc,EAChB,KAAM,IAAI,OACN,kEAEN,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,4DAElB,GAAI,GAAc,GACd,EAAc,GACd,EAAU,GACV,EAAU,GACV,EAAe,GACnB,GAAK,EAAqB,eAAgB,YACxC,EAAc,WAEZ,MAAQ,YAAe,aAAe,YAAkB,WAC1D,EAAc,WAEZ,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAER,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAEA,EAAe,YAAc,KACvC,EAAe,OAEf,MAAM,IAAI,OACN,6OAIA,YAAY,EAAc,YAAY,OAE5C,GAAI,GACF,GAAM,GAAgC,EACtC,GAAI,GACC,EAA4B,WACzB,EACN,KAAM,IAAI,OACN,yGAMR,GAAM,GAAS,GAAU,GAAY,EAAO,aAC5C,GAAI,GAAU,MACZ,GAAM,GAA2B,CAAC,OAAM,GAClC,EAAyB,CAAC,YAAW,GAC3C,MAAO,GAAO,UACH,GAAY,EACZ,GAGP,GAAA,GAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAMV,EAEJ,AAAI,EACF,EAEK,EAAe,WAAW,MAAM,aAAa,EAAG,EAAG,EAAO,GAAQ,KAClE,AAAI,GAAe,EACxB,EAAQ,EAAiC,KACpC,AAAI,IAAW,IACpB,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAEpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA4B,EAAG,EAAG,EAAO,GAC7C,EAAO,GAAoB,aAAa,EAAG,EAAG,EAAO,GAAQ,MAE/D,GAAI,GACJ,GAAI,IAAgB,EAClB,EAAS,GAAI,YAAW,QAExB,GAAM,GAAY,EAAQ,EAC1B,EAAS,GAAI,YAAW,EAAY,GACpC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAU,EAAG,EAAU,EAAa,EAAE,EAC7C,EAAO,EAAI,EAAc,GAAW,EAAK,EAAI,EAAI,GAIvD,GAAM,GAAqC,CAAC,EAAQ,EAAO,GAC3D,MAAO,IAAS,EAAQ,EAAU,qBAuBhC,EACA,iIAQF,GAPI,EAAO,EAAgB,EAAK,MAAO,YACvC,AAAM,YAAe,IAEb,GAAoB,EAC1B,EAAO,GAAK,EAAmB,SAC/B,EAAkB,WAEhB,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,OACN,wDAAwD,EAAK,KAAI,KAKvE,GAHM,EAAkB,EAAK,MAAM,MAAM,EAAG,GAArC,EAAM,EAAA,GAAE,EAAK,EAAA,GACd,EAAQ,EAAK,OAAS,EAAI,EAAI,EAAK,MAAM,GAE3C,EAAQ,GAAK,IAAU,EACzB,KAAM,IAAI,OACN,wCACA,sBAAqB,IAG3B,GAAI,EAAK,QAAU,WAAa,EAAK,QAAU,QAC7C,KAAM,IAAI,OACN,kCAAkC,EAAK,MAAK,0CAIrC,MAAA,CAAA,EAAM,EAAK,eAIxB,IAJM,EAAO,EAAA,OACP,EAAa,EAAK,QAAU,UAAY,IAAM,EAC9C,EAAQ,GAAI,mBAAkB,EAAQ,EAAS,GAE5C,EAAI,EAAG,EAAI,EAAS,EAAO,EAAE,GAGpC,IAFM,EAAO,CAAC,EAAG,EAAG,EAAG,KAEd,EAAI,EAAG,EAAI,EAAO,KAGzB,GAFM,EAAQ,EAAK,EAAI,EAAQ,GAE3B,EAAK,QAAU,WACjB,GAAI,EAAQ,GAAK,EAAQ,EACvB,KAAM,IAAI,OACN,qDACA,kCAAiC,EAAK,cAEnC,EAAK,QAAU,SACpB,GAAQ,GAAK,EAAQ,KACvB,KAAM,IAAI,OACN,mDACA,oCAAmC,EAAK,MAIhD,AAAI,IAAU,EACZ,GAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,GAElB,EAAK,GAAK,EAAQ,EAIhB,EAAI,EAAI,EACd,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAGjC,MAAI,IAAU,MACZ,GAAO,MAAQ,EACf,EAAO,OAAS,EACV,EAAM,EAAO,WAAW,MACxB,EAAY,GAAI,WAAU,EAAO,EAAO,GAC9C,EAAI,aAAa,EAAW,EAAG,IAEjC,AAAI,IAAS,GACX,EAAK,UAEP,CAAA,EAAO,QAGF,GAAM,IAAa,EAAG,CAAC,YAAW,+DCrOrC,EAAgB,GAClB,GAAI,EAAO,KAAO,EAChB,KAAM,IAAI,OACN,0DACA,sBAAqB,EAAO,KAAI,MAEtC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,4DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OACN,sDACA,uBAAsB,EAAQ,MAAK,MAEzC,GAAI,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAAO,KAC3C,KAAM,IAAI,OACN,iEACG,GAAQ,MAAM,EAAQ,KAAO,GAAE,QAAQ,EAAO,OAGvD,GAAI,EAAO,OAAS,EAClB,KAAM,IAAI,OACN,qDACA,kBAAiB,EAAO,MAAK,MASnC,OANM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAIjD,EAAU,EACL,EAAI,EAAG,EAAI,EAAa,OAAS,EAAG,EAAE,EAC7C,GAAW,EAAa,GAG1B,GAAM,GAAa,EAAO,MAEpB,EAAc,EAAa,QACjC,EAAY,MAGZ,OADI,GAAY,EACP,EAAI,EAAW,EAAI,EAAO,KAAM,EAAE,EACzC,GAAa,EAAW,GACxB,EAAY,KAAK,EAAW,IAG9B,GAAM,GACE,GAAe,EAAO,OAAO,IAAI,SAAA,GAAU,MAAA,GAAS,IAAU,OAAA,CACjE,IAAG,MAAM,EAAG,GAEjB,MAAO,CAAC,EAAa,EAAS,EAAW,6DCtDvC,EAAiB,EAAiB,GACpC,GAAM,GAAY,EAAQ,KAAO,EAAK,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAClE,EAAY,EAAQ,KAAO,EAAK,EAAQ,KAAO,EAAI,EAEnD,EAAa,wDACf,yCAAwC,EAAQ,OAChD,qBAAoB,EAAQ,MAAK,YAAY,GAC7C,gBAAe,EAAQ,mBAAmB,EAAQ,KAEtD,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OAAM,EAAa,mBAAkB,EAAQ,OAEzD,GAAI,EAAM,OAAS,EAAY,GAAQ,KAAO,GAC5C,KAAM,IAAI,OACN,EACA,2BAA0B,GAAY,GAAQ,KAAO,MAE3D,GAAI,EAAQ,OAAS,EAAW,EAAM,OAAS,EAC7C,KAAM,IAAI,OACN,EAAa,oBAAmB,GAAW,EAAM,OAAS,KAEhE,OAAS,GAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,GAAI,EAAQ,MAAM,KAAO,EAAQ,MAAM,GACrC,KAAM,IAAI,OACN,EACA,mBAAkB,EAAC,MAAM,EAAQ,MAAM,GAAE,sBAAsB,EAAC,MAC5D,EAAQ,MAAM,GAAE,OAG5B,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAO,EAAU,EAAE,EAC7C,GAAI,EAAQ,MAAM,EAAI,KAAc,EAAM,EAAI,GAC5C,KAAM,IAAI,OACN,EACA,mBAAkB,GAAI,GAAQ,MAC1B,EAAQ,MAAM,EAAI,GAAS,cAAc,GAAI,GAAQ,MACrD,EAAM,EAAI,GAAS,kBAoB7B,EAAiB,EAAiB,GACpC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OAAM,0DACZ,EAAQ,OAEd,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OACN,6DAA6D,GAGnE,GAAI,EAAM,SAAW,GACnB,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAEd,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAIhB,GAAoB,EAAO,EAAS,eAalC,EAAqB,EACrB,GAWF,OATM,GAAc,EAAQ,MAAM,OAC5B,EAAa,EAAc,EAAK,EAAQ,MAAM,EAAc,GAAK,EAKjE,EAAU,EAAM,OAElB,EAAY,EACP,EAAI,EAAW,EAAI,EAAS,EAAE,EACrC,GAAa,EAAM,GAGrB,GAAM,GAAgB,EAAY,EAAK,EAAI,EACrC,EAAa,GAAc,EAAQ,OAAS,EAE5C,EAAc,GAAe,EAAM,MAAM,EAAG,IAAW,OAAA,CAAE,IACzD,EAAa,GAAc,GACjC,MAAO,CAAC,UAAS,EAAE,WAAU,EAAE,UAAS,EAAE,QAAO,EAAE,WAAU,sFCnJ/D,YAqBI,EAAmB,EAAiB,GACtC,GAAM,GAAY,EAAM,MAAM,OAC9B,EACI,IAAc,EAAM,OACpB,WAAM,MAAA,iBAAiB,EAAS,sBAAsB,EAAK,SACvD,iCAAgC,EAAS,QACjD,EACI,IAAc,EAAK,OACnB,WAAM,MAAA,iBAAiB,EAAS,qBAAqB,EAAI,SACrD,iCAAgC,EAAS,QAEjD,mBAAS,GACP,EACI,EAAM,GAAK,EAAK,IAAM,EAAM,MAAM,GAClC,WAAM,MAAA,iBAAiB,EAAS,YAAY,EAAC,YAAY,EAAC,KACtD,KAAI,GAAM,GAAK,EAAK,IAAE,gCAAgC,EAAC,MACjD,EAAM,MAAM,GAAE,QALrB,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,eAUgB,GAGzB,OAFM,GAAO,GACT,EAAO,EACJ,EAAO,GACZ,AAAI,EAAO,GACT,EAAK,KAAK,GAEZ,GAAQ,EACR,IAEF,MAAO,eAKL,EAAiB,EAAe,GAElC,OADM,GAAO,GACJ,EAAO,EAAG,EAAO,EAAM,OAAQ,IACtC,EAAK,GAAQ,KAAK,KAAM,GAAI,GAAQ,EAAM,IAAS,EAAQ,IAE7D,MAAO,eAML,EAAmB,EAAgC,EACnD,GAEF,OADM,GAAiB,EAAO,QACrB,EAAI,EAAW,OAAQ,EAAI,EAAW,OAAQ,IACrD,EAAW,KAAK,GAElB,OAAS,GAAI,EAAG,EAAI,EAAe,IACjC,AAAI,IAAM,EACR,EAAW,GAA0B,EAErC,GAAW,OACP,EAAwB,EACxB,GACJ,EAAW,OAGf,MAAO,GAGT,YACI,EAAgC,EAChC,GACF,MAAI,IAAkB,EACb,EAGF,EAAkB,GAAgB,GAG3C,YAAuB,EAAuB,GAE5C,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAe,IACjC,EAAW,KAAK,EAAyB,GAE3C,MAAO,eAKL,EAAsB,EAAwB,EAC9C,EAAiB,EAAe,EAAmB,EACnD,EACA,GACF,GAAM,GAAY,EAAW,OACzB,EAAkB,GAAI,OAAM,GAC5B,EAAgB,GAAI,OAAM,GAC1B,EAAoB,GAAI,OAAM,GAClC,GAAI,EAAa,QAAU,EAAsB,GAC/C,GAAM,GAAY,EAAa,GAIzB,EAAgB,EAAsB,EAC5C,EAAkB,GACd,EAAW,EAAW,EAAe,EAAO,GAChD,EAAgB,GACZ,EAAS,EAAW,EAAe,EAAK,GAC5C,EACI,GAAsB,EAAS,EAAW,EAAe,OAE7D,QAAS,GAAO,EAAG,EAAO,EAAW,IACnC,EAAgB,GAAQ,GACpB,EAAW,EAAO,EAAS,EAAY,EAAM,GACjD,EAAc,GACV,GAAY,EAAS,EAAK,EAAS,EAAY,EAAM,GACzD,EAAkB,GAAQ,GAAe,EAAS,EAAM,GAI5D,MAAO,CACL,MAAO,EACP,IAAK,EACL,QAAS,eAOT,EAAmB,EAAgC,EACnD,EAAyB,GAI3B,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAEnB,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAc,GAClC,AAAI,EAAY,GAAK,GACnB,GAAgB,GAGlB,EAAW,GAAQ,EAGvB,MAAO,eAML,EAAiB,EAAgC,EACjD,EAAuB,GAIzB,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAAO,sBAE1B,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAY,GAChC,AAAI,EAAU,GAAK,GACjB,GAAgB,OAAO,kBAEzB,EAAW,GAAQ,EAIvB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,KAErC,GAAM,GAAW,EAAW,GAC5B,AAAI,EAAW,GAAK,GAClB,GAAW,IAAM,GAEnB,EAAW,GAAK,GAAW,EAAG,EAAW,GAAI,EAAW,IAE1D,MAAO,eAIL,EAAmB,EAAc,GACnC,GAAI,GAAS,EAAQ,GACrB,MAAI,GAAgB,GAAK,GAAS,GAAU,OAC1C,GAAS,GAGJ,cAIL,EAAmB,EAAwB,EAC3C,EAAsB,EAAc,GAEtC,GAAI,GAAQ,EAAa,GACnB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAY,GAAK,GAAQ,EAAe,GAAK,GAAQ,GAAS,OAChE,CAAI,EAAS,EAIX,EAAQ,OAAO,iBAGf,EAAQ,OAAO,kBAKnB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAQ,GACV,IAAS,GAIX,EAAQ,GAAW,EAAG,EAAO,EAAW,GAEjC,cAIL,EAAiB,EAAuB,EACxC,EAAsB,EAAc,GAEtC,GAAI,GAAO,EAAY,GACjB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAW,GAAK,GAAS,EAAgB,GAAK,GAAS,GAAQ,OACjE,CAAI,EAAS,EAGX,EAAO,OAAO,iBAGd,EAAO,OAAO,kBAKlB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAO,GACT,IAAQ,GAMV,AAAI,EAAS,EAEX,EAAO,GAAW,EAAG,EAAM,GAG3B,EAAO,GAAW,GAAI,EAAM,EAAW,GAGlC,cAQL,EAAiB,EAAiB,GAGpC,OADI,GAAkB,EAAK,OAClB,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,GAAI,EAAK,GAAK,GACZ,EAAkB,EAClB,MAIJ,OAAS,GAAI,EAAkB,EAAG,EAAI,EAAK,OAAQ,IACjD,GAAI,EAAM,GAAK,GAAK,EAAK,KAAO,EAAM,GACpC,MAAO,GAGX,MAAO,eAGyB,EAAiB,GAEjD,OADI,GAAa,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EACrD,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAc,EAAM,GAAK,EAAQ,GAEnC,MAAO,eAIL,EAAe,EAAwB,GAEzC,GAAI,GACE,EAAQ,EAAE,MAAM,OACtB,AAAI,MAAO,IAAU,SACnB,EAAM,CAAI,GAAK,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,IACzC,AAAI,EAAM,OAAS,EACxB,EAAS,EAAM,OAAO,GAAI,OAAM,EAAQ,EAAM,QAAQ,KAAK,IAE3D,EAAS,EAAM,QAEjB,EAAO,QAAQ,SAAA,GACb,EACI,IAAM,GAAI,WAAM,MAAA,wDAEtB,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAAI,OAAM,GAAO,KAAK,IACzB,AAAI,MAAO,IAAS,SACzB,EAAK,CAAI,GAAI,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,KACvC,AAAI,EAAK,OAAS,EACvB,EAAQ,EAAK,OAAO,GAAI,OAAM,EAAQ,EAAK,QAAQ,KAAK,KAExD,EAAQ,EAEV,EAAQ,EAAM,IAAI,SAAC,EAAG,GACpB,MAAI,IAAK,EACA,EAEP,GACI,IAAM,GACN,WAAM,MAAA,qDACC,GAAC,kCAAkC,EAAC,OACxC,EAAE,MAAM,GAAK,EAAO,MAGxB,CAAC,EAAQ,wSCjWlB,AA8DA,GAAA,IAAA,WAAA,cAYE,SAAA,UAAA,aAAA,WACE,MAAQ,MAAK,YACR,WAiBA,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,IAEnB,KASA,GAAA,WAOE,aACE,KAAK,aAAe,GAMf,SAAA,OAAP,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UAMnB,EAAA,SAAP,SAAwC,GACtC,EAAiB,SAAS,aAAa,EAAI,WACvC,CAAC,EAAK,EAAI,aAElB,iBA2BI,GACF,EACI,EAAI,WAAa,KACjB,WAAM,MAAA,gFAEV,EACI,MAAO,GAAI,WAAc,SACzB,WAAM,MAAA,sDACF,MAAO,GAAI,YACnB,EACI,EAAI,UAAU,OAAS,EACvB,WAAM,MAAA,sFAGV,GAAiB,SAAS,gFC/K5B,AAsBA,GAAM,IAAuB,KAChB,GAAuB,eAGhC,EACA,EAAoD,GACtD,MAAI,IAAW,MACb,GAAU,MAEL,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,mBAInE,MAAO,GAAO,QAAQ,mBAAqB,GAAK,GACA,GAGlD,YACI,EAAoB,EACpB,GACF,GAAI,GAAiB,GAOrB,GANA,AAAI,IAAa,IAAW,GAAa,KACvC,GAAiB,IAEnB,AAAI,GAAa,IAAW,GAAa,IACvC,GAAiB,IAEf,GACF,GAAM,GAAQ,EAAO,YAAY,KAC3B,EAAQ,EAAS,YAAY,KAEnC,GAAI,IAAU,EACZ,KAAM,IAAI,OACN,yCAAyC,EAAK,KAC9C,cAAa,IAIrB,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,IACzC,GAAM,GAAc,GAAW,GACzB,EAAgB,GAAW,GACjC,GAAI,CAAC,GAAY,EAAa,GAC5B,KAAM,IAAI,OACN,iCACA,aAAY,EAAW,iBAAiB,EAAa,MAI7D,GAAM,GACF,GAAa,GAAU,EAAS,GAAQ,GACtC,EAAe,GAAa,GAC9B,EACA,GAAQ,GAEZ,GAAI,EAAW,SAAW,EAAa,OACrC,KAAM,IAAI,OACN,yCAAyC,EAAW,OAAM,OAC1D,cAAa,EAAa,OAAM;GAChC,cAAa,EAAU;GACvB,cAAa,EAAY,MAE/B,OAAS,GAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAI,EAAW,GACf,EAAI,EAAa,GAEvB,GAAI,CAAC,EAAU,EAAG,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAC,OAAO,EAAC,cAAc,EAAC,OAAO,EAAC;EACzD,cAAa,EAAU;GACvB,cAAa,EAAY,mBAUC,EAAuB,GACzD,IAAK,KAAK,WAAM,MAAA,GAAK,QAAQ,WAAM,MAAA,mBAGH,EAAoB,GACpD,GAAM,GAAM,MAAO,IAAa,UAAY,MAAO,IAAa,UACxD,MAAO,IAAa,UACxB,CAAC,GACD,EACJ,MAAI,IAAS,IAAW,GAAU,EAAoB,KAClD,GAAS,IAAa,GAAU,EAAsB,IAEjD,GAAsB,EAAQ,EAAK,SAAC,EAAG,GAAM,MAAA,IAAK,IAEpD,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,iBAGlC,EAAW,EAAW,GAIvD,GAHA,AAAI,GAAW,MACb,GAAU,MAER,CAAC,GAAS,EAAG,EAAG,GAClB,KAAM,IAAI,OAAM,8BAA8B,EAAC,kBAAkB,GAIrE,YAAkB,EAAW,EAAW,GACtC,MAAI,CAAC,SAAS,IAAM,CAAC,SAAS,GACrB,GAEL,QAAM,IAAM,MAAM,IAAM,KAAK,IAAI,EAAI,GAAK,eAO5C,EAA6B,EAAa,GAC5C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,GAAI,EAAO,GAAK,GAAO,EAAO,GAAK,EACjC,KAAM,IAAI,OACN,sBAAsB,EAAO,GAAE,SAAS,EAAG,WAAW,eAM5D,EAAqB,GAGvB,OAAO,GAAI,cAAa,IAAS,QAAQ,GAAI,cAAa,4MCzJ5D,GAGM,IAAU,QCHhB,cAiCE,KAAM,IAAI,OAAQ,kBAkBlB,KAAM,IAAI,QAAS,kBAKnB,KAAM,IAAI,+BAAgC,IAC1C,QAAQ,KAAK,sEAIiB,GAC9B,AAAI,KAAM,QAAQ,iCAChB,QAAQ,KACJ,EAAM,6FAYZ,EAAO,iCASP,MAAO,iBAyBP,MAAO,GAAO,qBAgCQ,GAEtB,MAAO,GAAO,QAAQ,eA4CpB,EAA6B,GAC/B,MAAO,GAAO,KAAK,EAAU,eAcP,GACtB,GAAM,GAAU,GAAsB,GACtC,EAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,wBAmCI,GACrC,MAAO,GAAO,KAAK,eA4BA,GACnB,MAAO,GAAO,KAAK,eAkBM,GACzB,MAAO,GAAO,WAAW,iBAWzB,MAAO,GAAO,sBAUd,MAAO,GAAO,wBAQc,GAC5B,EAAO,cAAc,eAOK,GAC1B,MAAO,GAAO,YAAY,eAQO,GAEjC,MAAO,GAAO,mBAAmB,eAmB/B,EAAc,EACd,GAAA,MAAA,KAAA,QAAA,GAAA,GACK,EAAO,gBAAgB,EAAM,EAAS,iBAY7C,MAAO,GAAO,oBASY,EAAsB,GAChD,KAAM,YAAY,EAAc,GC5XlC,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCnE3B,AAqDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCvErC,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OAGjC,GAFA,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAED,EAAG,QAAU,SAAW,EAAG,QAAU,QACvC,MAAO,IAAS,EAAI,GAGtB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,WAAW,EAAI,GACnC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAC/B,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KC3E3B,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAEA,GAAM,EAAG,CAAC,KAAI,KCpE3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GAEpC,MADA,GAAK,CAAC,IACF,EAAG,QAAU,YACR,EAAQ,WAAW,GAGrB,EAAQ,IAAI,IAClB,EAAgC,KAAiB,OAGzC,IAAM,EAAG,CAAC,KAAI,KCrD3B,AAoCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAuCA,YAAiC,GAC/B,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,+DACV,EACI,EAAQ,QAAU,EAClB,WAAM,MAAA,uDACF,IAAG,EAAQ,UAEnB,GAAM,GACF,EAAQ,IAAI,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAG,UAAU,EAAK,UAEtD,EAAc,EAAS,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAI,EAAE,QAAU,EAAY,MAC1B,KAAM,IAAI,OACN,8DAIR,EAAS,QAAQ,SAAA,GACf,GAAI,CAAC,GAAiB,EAAE,MAAO,EAAY,OACzC,KAAM,IAAI,OACN,8DAIR,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,GACE,GAGH,EAAqB,EAE3B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC/E7B,YAuBqC,EAAgB,GACnD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,GAAI,EAAK,EAAK,OAAS,EAAI,KAAO,EAAO,EAAI,EAC3C,MAAO,GAGX,MAAO,eAIL,EAAqB,EAAqB,GAK1C,OAJI,GAAO,EAAU,OAAS,EAAU,OACpC,EAAM,GACR,EAAS,EACT,EAAY,EACL,EAAM,EAAG,EAAM,EAAM,IAC9B,AAAI,EAAK,QAAQ,KAAS,GACxB,EAAI,KAAK,EAAU,MAEnB,EAAI,KAAK,EAAU,MAGvB,MAAO,eAIL,EAAkB,GAGpB,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,EAAK,QAAQ,KAAS,IACxB,EAAS,KAAK,EAAO,IAGzB,GAAM,GAAc,EAAK,IAAI,SAAA,GAAO,MAAA,GAAO,KAC3C,MAAO,CAAC,EAAU,eAIhB,EAAiB,GACnB,GAAM,GAAiB,EAAK,IAAI,SAAA,GAAK,MAAA,KACrC,MAAO,IAAiB,EAAO,EAAgB,eAI7C,EAAa,EAAgB,GAC/B,EACI,GAAqB,EAAM,GAC3B,WAAM,MAAG,GAAG,2CACR,aAAY,EAAI,aAAa,EAAI,yBAQR,EAAgB,GAEjD,GAAI,GAAqB,EAAM,GAC7B,MAAO,MAGT,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,AAAI,EAAK,QAAQ,KAAO,IACtB,EAAO,KAAK,GAGhB,SAAK,QAAQ,SAAA,GAAQ,MAAA,GAAO,KAAK,KAC1B,cAI8B,GACrC,MAAO,GAAK,IAAI,SAAC,EAAM,GAAM,MAAA,CAAC,EAAG,KAC5B,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAE,KACxB,IAAI,SAAA,GAAK,MAAA,GAAE,iBAGe,EAAiB,GAEhD,OADM,GAAgB,GACb,EAAI,EAAO,EAAS,EAAI,EAAM,EAAE,EACvC,EAAI,KAAK,GAEX,MAAO,GC3GT,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KCxF3B,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAIA,IAAM,EAAG,CAAC,KAAI,KCzF3B,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC7EjC,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,AAAI,GAAQ,MACV,GAAO,GAET,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KChFjC,AAqCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC/C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA2CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAEH,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAQ,EAAG,CAAC,OAAM,KC7D/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,YAoGI,EACA,EAAuC,EACvC,EAA4B,EAC5B,GAD4B,AAAA,IAAA,QAAA,GAAA,QAM9B,GAAM,GAAgB,EAAW,GAC3B,EACE,EAAW,OAAA,CAAE,IACf,EAAc,GAAwB,GAE5C,MAAO,IACH,EAAY,EAAc,EAAS,EAAW,EAC9C,KAAyB,KAAsB,eAIjD,EACA,EAAqC,EACrC,EAAoC,EACpC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,gBACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GAEjB,EACJ,GAAI,IAAe,eACjB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACrD,IAAe,gBACxB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAE9D,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,EAAc,GAC7D,eAOF,EACA,EACA,EACA,EAA4C,EAC5C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,SACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAE9B,EACA,EACJ,GAAI,IAAe,QACjB,EAAc,eACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACxD,IAAe,QACxB,EAAc,gBACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAEjE,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,GAAO,EACtD,eAQF,EACA,EACA,EAAkC,EAClC,EACA,EAAuC,EACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACvC,AAAA,IAAA,QAAA,GAAA,gBACE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GACnC,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WACtB,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAU,EAAA,EAAA,OAElC,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAC9B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GACf,EAAA,GAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAEjB,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,SAIrB,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAW,GAC1C,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAW,EAAU,IAGvC,CACL,UAAS,EACT,WAAU,EACV,SAAQ,EACR,QAAO,EACP,WAAU,EACV,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,aAAY,EACZ,YAAW,EACX,aAAY,EACZ,YAAW,EACX,sBAAqB,EACrB,qBAAoB,EACpB,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,eA2CX,EACA,EACA,EACA,EAA4C,EAC5C,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,gBAEE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GAE5C,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WAC/B,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,OAE3C,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAE3C,EAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAC5B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAe,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGhC,EACF,GAAuB,EAAa,GAClC,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,SAAU,EAAA,EAAA,UAAW,EAAA,EAAA,SAK/B,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAU,EAAW,GACpD,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAU,EAAW,EAAU,IAGjD,CACL,UAAS,EACT,WAAU,EACV,QAAO,EACP,SAAQ,EACR,QAAO,EACP,WAAU,EACV,SAAQ,EACR,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,YAAW,EACX,aAAY,EACZ,YAAW,EACX,YAAW,EACX,aAAY,EACZ,YAAW,EACX,qBAAoB,EACpB,sBAAqB,EACrB,qBAAoB,EACpB,cAAa,EACb,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,GAIf,YACI,EAA2B,EAAmB,EAC9C,EAAkB,GACpB,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAY,GAGtB,YACI,EAA2C,EAC3C,EAAqB,EAAgB,EACrC,GACF,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAa,EAAQ,GACrB,EAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAe,GAChB,GAAa,EAAY,EAAI,GAAW,EAAS,EAAG,GACzD,EACI,GAAW,GACX,WAAM,MAAA,2BAA2B,EAAY,uEAGjD,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAc,EAAY,EAAY,eAI5C,EACA,EAAmB,EAAgB,GAAA,AAAA,IAAA,QAAA,GAAA,GACrC,GAAM,GAAqB,GAAuB,EAAW,GAC7D,MAAO,MAAK,MACP,GAAW,GAAM,GAAS,GAAK,EAAS,GAAsB,GAGrE,YAAyB,GACvB,MAAI,OAAO,IAAU,SACZ,CAAC,EAAO,EAAO,GAEpB,EAAM,SAAW,EACZ,CAAC,EAAM,GAAI,EAAM,GAAI,GAEvB,EAGT,YAA0B,GAExB,MAAO,OAAO,IAAU,SAAW,CAAC,EAAO,EAAO,GAAS,EAc7D,YAAgC,EAAoB,GAClD,MAAI,IAAY,EACP,EAGF,EAAc,GAAa,GAAM,GAAW,GAGrD,YACI,EAA4C,EAC5C,EAAiB,EAAsB,EACvC,EAAsB,EACtB,EACA,GAEF,GAAI,GACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CAAC,IAAK,EAAK,OAAQ,EAAK,KAAM,EAAK,MAAO,EAAK,KAAM,GAC/D,GAAM,GAAW,GACb,CAAC,EAAU,GAAU,EAAc,EAAc,EAAK,GAC1D,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GACF,KAAK,IAAI,EAAI,GAAY,GAAK,EAAe,EAAe,GAC1D,EACF,KAAK,IAAI,EAAI,GAAW,GAAK,EAAc,EAAc,GACvD,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAC9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,gBAClC,IAAQ,QACjB,EAAU,CAAC,IAAK,EAAG,OAAQ,EAAG,KAAM,EAAG,MAAO,EAAG,KAAM,SACvD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,WAC1C,MAAO,IAAQ,UACxB,GAAM,GAAM,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GACzD,EAAS,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC5D,EAAO,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC1D,EAAQ,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC3D,EAAW,IAAQ,GAAK,IAAW,GAAK,IAAS,GAAK,IAAU,EAClE,QACA,WACJ,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,GAC3C,EAAY,GACP,GAAW,EAAe,EAAM,GAAU,EAAe,EAC1D,GACJ,EAAW,GACN,GAAU,EAAc,EAAO,GAAS,EAAc,EAAG,OAE9D,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,UAAS,EAAE,SAAQ,GAGtC,YACI,EAA4B,EAAiB,EAC7C,EAAiB,EAAqB,EACtC,EAAqB,EAAqB,EAC1C,EAAqB,GAMvB,GAAI,GACA,EACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,GAER,GAAM,GAAW,GACb,CAAC,EAAS,EAAU,EAAS,GAAI,EAAa,EAAG,EAAa,EAC9D,GACJ,EAAW,EAAS,GACpB,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAW,KAAK,KAAK,EAAU,GAC/B,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EACD,GAAY,GAAK,EAAe,EAAe,EAC9C,EAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EAAQ,KAAK,MAAM,EAAgB,GACnC,EAAO,EAAgB,EACvB,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAE9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,MAAK,EAAE,KAAI,EAAE,KAAM,gBAC/C,IAAQ,QACjB,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,SAER,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,GACnD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,OAEnD,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,SAAQ,EAAE,UAAS,EAAE,SAAQ,GAQhD,YACI,EAAe,GACjB,GAAI,CAAC,EACH,MAAO,GAET,OAAQ,OACD,QAEH,MAAO,MAAK,MAAM,OACf,OAEH,MAAO,MAAK,KAAK,OACd,QACH,MAAO,MAAK,MAAM,WAElB,KAAM,IAAI,OAAM,wBAAwB,gBAIZ,GAC1B,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,MAAO,KAAS,GAAK,IAAS,GAAK,IAAS,cAI1C,EAA0B,GAC5B,MAAO,IAAkB,IAAY,GAAkB,eAUjB,GAEtC,GAAI,IAAe,OACjB,MAAO,eACF,GAAI,IAAe,OACxB,MAAO,gBAEP,KAAM,IAAI,OAAM,sBAAsB,GC5mB1C,AAoDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,WACxC,EAAY,EAElB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,mDAAmD,EAAI,KAAI,MAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAIJ,MAFA,GAAK,CAAC,IAEF,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UACvC,EAAI,QAGN,EAAQ,QAAQ,EAAK,IAGxB,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAElE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAG,OAEf,EACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KClHnC,AA6EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,WAE5C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAGrB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GAEJ,SAAK,CAAC,IAEC,EAAQ,UAAU,EAAK,IAG1B,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAEjE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAI,OAEhB,EACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KCzJvC,YAmBuC,EAAoB,GACzD,GAAM,GAAO,EAAO,GAAG,OACvB,EAAO,QAAQ,SAAC,EAAO,GACrB,EACI,EAAM,SAAW,EACjB,WACI,MAAA,kBAAkB,EAAI,sBAAsB,EAAC,sBAC7C,6BAA4B,EAAI,SAG1C,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,kBAAkB,EAAI,iCAAiC,GAAO,GAAC,MAEzE,GAAM,GAAa,EAAO,GAC1B,EAAO,QAAQ,SAAC,EAAO,GACrB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EACK,IAAM,GAAU,EAAM,KAAO,EAAW,GACzC,WAAM,MAAA,kBAAkB,EAAI,uBAAuB,EAAC,MAAM,EAAK,KAC3D,0CAAyC,EAAU,MACnD,oCAAmC,EAAC,qBAKlB,EAAoB,GAElD,OADM,GAAc,EAAO,GAAG,QACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAY,IAAS,EAAO,GAAG,GAEjC,MAAO,GClDT,AAqEA,YAAmC,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/D,EAAO,EAAQ,QAAU,EAAG,WAAM,MAAA,uCAElC,GAAI,GAAW,GAAqB,EAAS,UAAW,UAExD,AAAI,EAAS,GAAG,QAAU,aACxB,EAAS,QAAQ,SAAA,GACf,GAAI,EAAO,QAAU,YACnB,KAAM,IAAI,OAAM;uBACD,EAAO,MAAK,QAKjC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAS,GAAG,OAAO,GAChD,EAAW,GAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAC7D,GAAI,GAAc,KAAc,EAC9B,MAAO,IAAO,GAAI,GAIpB,GADA,EAAW,EAAS,OAAO,SAAA,GAAK,MAAA,GAAE,KAAO,IACrC,EAAS,SAAW,EACtB,MAAO,GAAS,GAGlB,GAAM,GAAS,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QACnC,GAAuB,EAAQ,GAE/B,GAAM,GAAM,EAAQ,OAAO,EAAU,GACrC,SAAK,GACE,GAGH,EAAuB,EACvB,EAAoB,CAAC,KAAI,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC/GjC,AAsCA,YAAoC,GAClC,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAAwB,CAAC,EAAG,GAElC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,QAAQ,GAC5B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAU,EAAG,CAAC,SAAQ,KCjDnC,AA8DA,YACI,EAAiB,EAAwB,GAC3C,GAAM,GAAK,EAAgB,EAAG,IAAK,SAEnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,kCAGlB,GAAM,GAA+B,SAAC,EAAS,GACvC,GAAA,GAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,UAA6B,EAAI,EAAQ,GACzC,EAAK,CAAC,IACC,EAAQ,MAAM,EAAI,EAAQ,IAG7B,EAAsB,CAAC,EAAG,GAC1B,EAAoB,CAAC,MAAK,EAAE,KAAI,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCrF/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAI,EAAQ,KAAK,GACvB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA8CA,YACI,EAA+B,EAC/B,EAA+B,EAC/B,EAAwB,GAC1B,GAAM,GACF,EAAgB,EAAY,aAAc,iBACxC,EACF,EAAgB,EAAY,aAAc,iBACxC,EAAY,EAAgB,EAAU,WAAY,iBAClD,EAAQ,EAAgB,EAAM,OAAQ,iBACtC,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAE7B,EAAW,GAAO,CAAC,EAAO,GAAK,GAC/B,EAAW,GAAO,EAAU,GAC5B,EAAgB,GAAI,EAAU,GAG9B,EAAY,EAAI,MAAM,GACtB,EAAY,EAAI,MAAM,GAAK,EAC3B,EAA8B,CAAC,EAAW,GAC1C,EAAI,GAAM,EAAK,CAAC,EAAG,GAAI,GACvB,EAAI,GAAM,EAAK,CAAC,EAAG,GAAY,GAC/B,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GACnC,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GAEnC,EACF,GAAI,EAAI,GAAQ,GAAI,GAAK,IACrB,EAAI,EAAI,GAAQ,GAAI,EAAa,MACnC,EAAiB,EAAI,GAAK,GAAO,GAAQ,IAC/C,MAAO,CAAC,EAAM,MAGH,IAAgB,EAAG,CAAC,eAAc,KC/E/C,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAE7C,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,iBAAiB,EAAG,KAAI,2CAC1B,EAAW,SAEnB,EACI,EAAM,SAAW,EAAW,OAC5B,WAAM,MAAA,mBACF,EAAM,OAAM,8CACZ,EAAW,SAEnB,EACI,EAAG,MAAM,GAAK,IAAS,EACvB,WAAM,MAAA,yBACI,EAAG,MAAM,GAAE,2CACjB,+BAA8B,EAAW,KAAK,OAAM,QAAQ,KAEpE,GAAM,GAA0B,SAAA,GAC9B,MAAO,GAAQ,eAAe,EAAI,EAAY,IAG1C,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,MAAK,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,iBC1FX,GACpC,GAAI,GACJ,MAAI,GAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAG,EAAE,OACxB,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACvC,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAErD,EAAM,EAGD,EChCT,AAwDA,YACI,EAAyB,EACzB,EACA,EACA,EACA,GACF,AAAI,GAAmB,MACrB,GAAkB,MAEpB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,AAAI,GAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAG9C,EACI,EAAM,OAAS,EAAU,KACzB,WAAM,MAAA,iFAEV,EACI,GAAW,MAAQ,EAAM,OAAS,EAAQ,KAC1C,WAAM,MAAA,+EAEV,EACI,GAAU,MAAQ,EAAM,OAAS,EAAO,KACxC,WAAM,MAAA,8EAGV,GAAM,GAAgB,GAAM,GAEtB,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,EAAO,EAAW,IAEtB,EAAQ,UACX,EAAK,GAAS,GAAQ,GAAS,GAAY,GAAS,GACpD,GAAS,GAAS,IAGlB,EAA+B,CACnC,EAAG,EACH,MAAO,EACP,OAAQ,EACR,KAAM,EACN,SAAU,GAGN,EAA6B,CAAC,gBAAe,GAE7C,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAO,GAAQ,EAAK,EAAG,OAGzB,YAAkB,GAChB,MAAI,IAAK,KACA,KAEL,EAAE,OAAS,EAEN,EAAQ,EAAG,CAAC,EAAE,OACZ,EAAE,OAAS,EACb,EACE,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACpC,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAEjD,KAGI,IAAY,EAAG,CAAC,WAAU,KCrGvC,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAE/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC9E3C,AA4CA,YACI,EAAsB,GACxB,GAAI,GAAQ,EAAgB,EAAG,cAAe,KACxC,EAAS,EAAM,MAErB,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,CAAE,GAAI,IAAM,EAAI,IAAM,IACxC,KAAM,IAAI,OAAM,2CAA2C,EAAK,MAGlE,GAAI,EAAM,OAAS,EAAM,KACvB,KAAM,IAAI,OAAM,+BAA+B,EAAM,OAAM,iBACvD,EAAM,KAAI,KAGhB,GAAI,EAAM,OAAS,EAAM,MAEvB,OADM,GAAW,EAAM,MAAM,QACtB,EAAS,OAAS,EAAM,QAC7B,EAAS,QAAQ,GAEnB,EAAQ,EAAQ,EAAO,GAKzB,OAFM,GAAa,EAAM,MACnB,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAM,OAAS,EAAG,GAAK,EAAG,IACrC,GAAI,EAAW,KAAO,EAAM,GAC1B,EAAK,GAAK,UACD,EAAM,MAAM,KAAO,EAC5B,KAAM,IAAI,OACN,mBAAmB,EAAM,6BAA6B,EAAK,MAGnE,GAAM,GAAO,EAAK,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,KAAI,OAAO,SAAA,GAAK,MAAA,IAAK,IAEjE,GAAI,EAAK,SAAW,EAClB,MAAO,IAAM,GAGf,GAAM,GAAU,SAAC,GAA2B,MAAA,GAAQ,KAAK,EAAO,IAE1D,EAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,MAAK,EAAE,WAAU,GAElD,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAa,MAGb,IAAc,EAAG,CAAC,aAAY,KC5F3C,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAyCA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,eACnC,EACK,GAAgB,EACjB,WAAM,MAAA,uBAAuB,EAAY,aACrC,+BAA8B,EAAY,QAElD,GAAM,GAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,aAAY,EAAE,aAAY,GAE3D,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAM,EAAQ,KAAK,EAAI,EAAc,GAC3C,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAc,EAAG,CAAC,aAAY,KC7B3C,YAAmB,GACjB,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCYrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCDrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KC5BrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCnCrC,AA8DA,YACI,EAAiB,EACjB,EACA,EACA,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GACjE,EACI,IAAY,EAAQ,MAAM,GAC1B,WAAM,MAAA,oCAAoC,EAAO,gBAC7C,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,GAAO,GACL,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GACxC,EACY,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAS,EAAG,CAAC,QAAO,KC3EjC,YACI,EAAiB,EAA6B,EAC9C,EACA,EAAiC,EACjC,GADA,AAAA,IAAA,QAAA,GAAA,OAAiC,AAAA,IAAA,QAAA,GAAA,GAEnC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,MAG9C,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAQ,GACjD,WAAM,MAAA,yDACF,eAAc,EAAM,kBAAkB,EAAQ,OACtD,EACI,IAAe,MACf,WAAM,MAAA,sCACF,EAAU,0CAElB,GAAM,GAAW,EACb,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,KAC7D,EAAU,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAA4B,CAAC,EAAG,GAChC,EAA8B,CAAC,EAAG,GAElC,EAAmB,OAEnB,EAAM,GACP,EAAuB,EAAuB,EAAS,EACxD,EAAkB,EAAW,GAEjC,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAGxC,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,QAGhD,IAAS,EAAG,CAAC,QAAO,KCpHjC,AAoDA,YACI,EAAmE,EACnE,EAAkB,EAClB,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KAC1D,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,KAG9C,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,GAAM,GAAU,IAAe,OAAS,EAAS,GAAK,EAAS,GACzD,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAY,EAEZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEE,EAAM,EAAQ,eAAe,EAAM,EAAQ,GAEjD,SAAK,CAAC,EAAM,IAEL,GAGH,EAAoC,CAAC,GAAI,EAAM,OAAM,GAErD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,EAAE,WAAY,GAEtD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAqB,GAEzB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxF3D,YACI,EAAiB,EACjB,EACA,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IACH,EAAa,EAAI,EAAS,EAAS,EAAK,OAAQ,MAGzC,IAAkB,EAAG,CAAC,iBAAgB,KCxDnD,AA8DA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAExE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAA+B,EAAS,GACxC,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,QACf,WAAM,MAAA,sCACF,EAAU,4CAElB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,GAC5C,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GAExC,EAAqB,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,GAEzD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,KAGI,IAAS,EAAG,CAAC,QAAO,KC3HjC,AA6CA,YACI,EAGA,EAAO,EAAkB,EACzB,GACF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACvE,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,EAAO,KAGzD,GAAM,GAAU,EAAS,GACnB,EAAW,EAAK,MAAM,GAC5B,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,GAEhD,MAAO,GAAQ,eAAe,EAAM,EAAQ,IAGxC,EAAoC,CAAC,GAAI,GAEzC,EAAkC,CAAC,IAAG,GAEtC,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxE3D,YACI,EAAiB,EACjB,EAGA,EAA0C,GAC5C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IAAoB,EAAa,EAAI,EAAS,EAAS,MAGnD,IAAkB,EAAG,CAAC,iBAAgB,KCtDnD,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AAqDA,YACI,EAAsB,EAAU,EAAmB,GAA7B,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,IAAmB,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAG,MAC9C,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,IAE5B,GAAM,GAAe,GAAiB,EAAG,EAAG,MAAM,GAC9C,EAAQ,EAAQ,OAAO,EAAW,EAAc,EAAW,GAG/D,GAFA,EAAK,CAAC,IAEF,GAAe,MACjB,GAAM,GAAqB,GAAuB,GAClD,EAAQ,GAAU,EAAO,GAE3B,MAAO,IAGP,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,EAAE,UAAS,EAAE,QAAO,GAEpD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KCnFjC,AAiEA,YACI,EAA0B,EAC1B,GAAA,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,gBAE7B,EAAe,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC/D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC9D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAEpE,EACI,EAAc,GAAa,EAC3B,WAAM,MAAA;MACN,EAAW,QAAQ,EAAS;MAC5B,EAAG,QAEP,EACI,EAAa,GAAa,EAC1B,WAAM,MAAA;MACN,EAAU,QAAQ,EAAS;UACvB,EAAG,QAEX,EACK,EAAc,GAAY,KAAe,EAC1C,WAAM,MAAA,8CACF,EAAY,EAAS,WACrB,EAAU,sCAAsC,EAAG,QAE3D,GAAM,GAAiC,SAAA,GACnC,MAAA,GAAQ,aAAa,EAAI,EAAW,IAElC,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAAC,UAAS,EAAE,WAAU,GAEvD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAc,MAGP,IAAe,EAAG,CAAC,cAAa,KCvG7C,AA2EA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,sDACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAElC,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAGlB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,iEACF,mBAAkB,EAAO,mBAAmB,EAAS,OAE7D,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IACE,EAAM,EAAQ,gBAAgB,EAAK,EAAS,GAClD,SAAK,CAAC,EAAK,IACJ,GAGH,EAAsC,CAAC,EAAG,EAAK,OAAQ,GACvD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KC/InD,AA+CA,YAAe,GACb,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAA,GACnC,GAAM,GAAO,EAAQ,EAAI,CAAC,EAAG,OACvB,EAAS,EAAQ,KAAK,GACtB,EAAe,EAAE,MAAK,OAAK,EAAE,OAEnC,MAAO,GAAQ,EAAQ,IAGnB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAO,EAAG,CAAC,MAAK,KChE7B,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,GADA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IACzC,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAU,EAAgB,EAAQ,SAAU,cAElD,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA,gEACC,GAAG,KAAI,OAClB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,4DACC,GAAQ,KAAI,OACvB,EACI,IAAe,OACf,WAAM,MAAA,0DACF,0BAAyB,KAEjC,GAAI,GAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACzD,EAAe,IAGjB,GAAM,GAA2B,CAAC,EAAG,EAAK,OAAQ,GAC5C,EAAyB,CAAC,QAAO,EAAE,IAAG,EAAE,UAAS,GAEjD,EAAM,EAAO,UACH,GAAY,EACZ,GAEhB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAa,EAAG,CAAC,YAAW,KCvGzC,YA2BI,EAAmB,GAGrB,OAFM,GAAS,EAAQ,OACjB,EAAiB,GACd,EAAI,EAAG,EAAI,EAAQ,KAC1B,GAAM,GAAM,EAAS,EAAI,EACnB,EAAI,EAAQ,IAAQ,EACpB,EAAI,EAAS,EAAS,OAAS,EAAI,IAAM,EAC/C,AAAI,EAAI,GAAK,IAAM,GACjB,EAAK,QAAQ,GAGjB,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAQ,EAAQ,EAAQ,OAAS,EAAI,GACrC,EAAU,EAAS,OAAS,EAAI,EAChC,EAAS,EAAS,GACxB,AAAI,IAAS,MAAS,IAAU,GAAK,EAAS,IAC5C,EAAO,QAAQ,GAGnB,MAAO,eAIL,EAAkB,GAIpB,OAHM,GAAmB,GACnB,EAAI,KAAK,IAAI,EAAO,OAAQ,EAAO,QAEhC,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GACnC,AAAI,GAAK,MACP,GAAI,GAEN,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GAInC,GAHA,AAAI,GAAK,MACP,GAAI,GAEF,IAAM,EACR,EAAO,QAAQ,WACN,IAAM,EACf,EAAO,QAAQ,WACN,IAAM,GACf,GAAM,GAAS,wDACR,GAAM,QAAQ,EAAM,KAC3B,KAAM,OAAM,OAEZ,GAAO,QAAQ,GAGnB,MAAO,GCrFT,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,MAAM,EAAI,IAE5D,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,OAG/C,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AAoDA,YACI,EAA8B,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAa,EAAgB,EAAW,YAAa,QAAS,QAI9D,EAAiB,GAA2B,EAAG,MAAO,EAAG,OACzD,EAAgB,GAAY,EAAI,GAChC,EAAgB,GAAY,EAAI,GACtC,AAAI,EAAW,OAAS,GAGtB,EACI,EAAW,MAAM,KAAO,EAAG,MAAM,GACjC,WAAM,MAAA,mEAGZ,AAAI,EAAW,OAAS,GAEtB,GACI,EAAW,MAAO,EAAc,MAAO,oBAG7C,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,OAAO,EAAY,EAAe,GACtD,SAAK,CAAC,IACC,GAEH,EAAyB,CAC7B,UAAW,EACX,EAAG,EACH,EAAG,GAEL,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,OAGA,IAAQ,EAAG,CAAC,OAAM,KC5F/B,AAuCA,YAAsC,GACpC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAA0B,CAAC,EAAG,GACpC,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,IAAK,EAClC,KAAiB,OAEjB,IAAY,EAAG,CAAC,WAAU,KC9CvC,AA0DA,YACI,EAAsB,SAEpB,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAAY,GAAI,EAAI,GACpB,EAAQ,GAAU,GAClB,EAAc,GAAM,EAAI,GAC9B,MAAO,IAAM,EAAa,EAAO,MAGtB,IAAW,EAAG,CAAC,UAAS,KCvErC,AA2CA,YAAc,EAAuB,GACnC,GAAM,GAAM,EAAgB,EAAI,KAAM,OAChC,EAAM,EAAgB,EAAI,KAAM,OAEtC,EACK,GAAI,OAAS,GAAK,EAAI,OAAS,IAAO,GAAI,OAAS,GAAK,EAAI,OAAS,GACtE,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GACjD,EAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GAOvD,GALA,EACI,IAAY,EACZ,WAAM,MAAA,gEACC,GAAO,QAAQ,EAAO,OAE7B,EAAI,OAAS,GAAK,EAAI,OAAS,GACjC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,YACZ,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,CAAC,EAAK,eAClB,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAK,GACzB,MAAO,GAAQ,EAAM,CAAC,EAAK,YAE3B,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAK,GACzB,MAAO,OAIE,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,GACtB,SAAK,CAAC,IACC,GAGH,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCtD3B,AAyCA,YAAgC,GAC9B,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,EACI,EAAG,QAAU,SAAW,EAAG,QAAU,UACrC,WAAM,MAAA,8CAEV,AAAI,EAAG,QAAU,SACf,GAAK,GAAK,EAAI,YAGhB,GAAM,GAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KC1D3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAyCA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GAC3D,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,aAAc,GAEjD,EAAY,GAAQ,EAAG,KAAM,WAAM,MAAA,uCACnC,GAAM,GAAW,EAAG,MAAM,QAC1B,MAAI,GAAO,GAET,GACI,CAAE,GAAG,KAAO,IAAM,EAClB,WAAM,MAAA,iCAAiC,CAAG,GAAG,KAAO,GAAE,KAAK,EAAG,KAAI,MACtE,EAAO,EAAG,KAAO,EAAO,GAE1B,EAAS,OAAO,EAAM,EAAG,GAClB,EAAQ,EAAI,MAGR,IAAa,EAAG,CAAC,YAAW,KC1DzC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAqDA,YAAiC,EAAiB,GAChD,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,OAAQ,GAC3C,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OAEzC,GAAM,GAA0B,SAAC,EAAS,GACxC,GAAM,GAAM,EAAQ,KAAK,EAAI,GAC7B,SAAK,CAAC,IACC,GAGH,EAAe,CAAC,GAChB,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,GAE9B,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,EAAkC,MAG3B,IAAO,EAAG,CAAC,MAAK,KC5E7B,AAwCA,YACI,EAAiB,EACjB,EAKA,GAAA,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAc,MAChB,GAAa,GAIf,OAFM,GAAO,GAAO,CAAC,EAAS,GAAa,GACrC,EAAI,GAAW,EAAa,EAAU,EACnC,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAK,IAAI,EAAG,EAAG,GAEjB,GAAM,GAAgB,EAAQ,EAAK,WAAY,CAAC,EAAS,IACzD,GAAI,GAAc,KAChB,MAAO,GAEP,GAAI,EAAW,SAAW,EACxB,MAAO,IAAK,GAAW,EAAK,GAAI,CAAC,EAAW,GAAI,EAAG,IAC9C,GAAI,EAAW,SAAW,EAE/B,MAAO,IACI,GAAW,GAAW,EAAK,GAAI,GAC/B,CAAC,EAAW,GAAI,EAAW,GAAI,EAAG,IACxC,GAAI,EAAW,SAAW,EAE/B,MAAO,IAAK,GAAW,GAAW,GAAW,EAAK,GAAI,GAAI,GAAI,CACrD,EAAW,GAAI,EAAW,GAAI,EAAW,GAAI,EAAG,IAGzD,KAAM,IAAI,OACN,2CAEA,8BAA8B,EAAmB,OAAM,UAKpD,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAqCA,YACI,EAAoB,EAAsB,GAC5C,GAAM,GAAmB,CAAC,MAAK,EAAE,MAAK,EAAE,MAAK,GAE7C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAO,EAAO,IAAQ,GAAI,KAAM,GACxD,GC3CN,AAqCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC7C/B,AAuBO,GAAM,IAAwB,eASI,GACvC,MAAI,IAAU,GACL,EAEF,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KCpCrD,YA8BI,EAAgB,GAClB,GAAI,GAAO,GACP,EASJ,IAPA,AAAI,GAAU,GACZ,GAAM,EACN,EAAO,IAEP,EAAM,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KAG7C,CAAC,GACN,AAAI,EAAM,GAAe,IAAQ,EAC/B,EAAO,GAEP,EAAM,GAAe,EAAQ,EAAM,GAGvC,MAAO,eAIL,EAAkB,EAAc,GAGlC,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,IAAQ,EACV,EAAS,KAAK,EAAO,IAErB,EAAS,KAAK,GAGlB,MAAO,eAUL,EAAW,EAAiB,GAM9B,OALM,GAAU,EAAE,MAAM,GAElB,EAAwB,GAC1B,EAAY,EACZ,EAAY,EACP,EAAI,EAAG,EAAI,EAAM,IACxB,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAM,IAChC,EAAY,KAAK,EAAQ,MAAM,IAGjC,OAAS,GAAI,EAAO,EAAG,EAAI,EAAE,KAAM,IACjC,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,MAAO,CAAC,UAAS,EAAE,UAAS,EAAE,QAAO,EAAE,YAAW,2GC5FpD,AAoDA,YACI,EAAiB,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAW,EAAgB,EAAS,UAAW,SAAU,SAEzD,EAAyB,CAAC,EAAG,EAAI,QAAS,GAC1C,EAAuB,CAAC,KAAI,GAE5B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAa,GAAe,EAAM,EAAG,OAAO,GAC5C,EAAY,GAAyB,EAAI,EAAU,GAEnD,EACF,EAAQ,OAAO,EAAI,EAAQ,EAAU,CAAC,EAAS,OAAQ,GAC3D,SAAK,CAAC,EAAI,IAEH,EAAQ,EAAK,EAAU,cAGhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC5EjC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,QAAQ,EAAI,IAE9D,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC3DnC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,aAAa,EAAI,GACrC,SAAK,CAAC,EAAI,IACH,GAGH,EAA6B,CAAC,EAAG,EAAI,EAAG,GAE9C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAe,EAAG,CAAC,cAAa,KC/D7C,AAsCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCnD7B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GAEnC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,SAAS,IAAK,EACnC,KAAiB,OAEV,IAAW,EAAG,CAAC,UAAS,KC/CrC,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC/C/B,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC9C/B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAEhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAGH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,YAsCI,EAAyC,GAC3C,GAAM,IAAa,IAAU,IAAU,UAAa,MAAM,QAAQ,KAC9D,IAAU,YACZ,KAAM,IAAI,OACN,kFAGN,GAAI,IAAU,UAAY,GAAa,IACnC,CAAE,aAAiB,aACrB,KAAM,IAAI,OACN,6EAGN,GAAM,GAAkB,GAClB,EAA0B,GAChC,MAAO,IAAW,EAAO,EAAO,EAAe,GCrDjD,AA2CA,YAAsC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aACnC,MAAO,IAAQ,EAAI,GAAO,GAAQ,GAAK,MAG5B,IAAY,EAAG,CAAC,WAAU,KChDvC,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,QAC7B,EAAK,EAAgB,EAAG,IAAK,QACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAE3D,EAAqB,CAAC,EAAG,EAAI,EAAG,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC1D7B,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,UAAU,EAAI,GAClC,SAAK,CAAC,EAAI,IACH,GAGH,EAA0B,CAAC,EAAG,EAAI,EAAG,GAE3C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAY,EAAG,CAAC,WAAU,KC/DvC,YAkCyB,EAAe,EAAc,GACpD,GAAI,GAAO,EACT,KAAM,IAAI,OAAM,4CAGlB,GAAM,GAAuB,CAAC,MAAK,EAAE,KAAI,EAAE,IAAG,GAC9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,SAAS,EAAO,EAAM,IAAM,GAC/C,KAAiB,GAAU,GC1CjC,AA4CA,YACI,EAAiB,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IACzD,GAAM,GAAK,EAAgB,EAAG,IAAK,8BACnC,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA;sBACU,EAAG,KAAI,MAC3B,EACI,GAAW,GACX,WAAM,MAAA,+DACF,gCAA+B,EAAW,OAElD,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,6BACd,EAAK,EAAa,EAAM,EAAO,GAEnC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAoB,CAAC,EAAG,GAExB,EAAkB,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEpD,KAIE,IAA6B,EAAG,CAAC,4BAA2B,KCvFzE,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AA0DA,YAAc,GAEZ,SACI,GAAgB,GAAI,WAAM,MAAA,+CACvB,SAAC,EAAsB,GAE5B,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MACxC,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,WAAa,KAC1D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,kFAGN,GAAW,GACJ,EAAM,MAkCnB,YAAe,GAEb,SACI,GAAgB,GAAI,WAAM,MAAA,gDACvB,SAAC,EAAgC,GACtC,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,qFAGV,GAAM,GAAQ,GAAqB,EAAM,OAAQ,WAAY,MACvD,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,YAAc,KAC3D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,MAAA,OAAA,IAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,iGAGN,GAAW,GACJ,KA8Bb,YAA0D,GAKxD,SACI,GAAgB,GAChB,WAAM,MAAA,uDACH,SAAC,EAAM,GACZ,EACI,YAAa,GACb,WAAM,MAAA,wDACV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,6DACJ,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,UAAW,GACJ,CAAC,KAAM,EAAM,GAAS,MAAK,IAmCtC,YAAyC,GAKvC,SACI,GAAgB,GAChB,WAAM,MAAA,wDACH,SAAC,EAAgB,GACtB,EACI,MAAM,QAAQ,IAAS,EAAK,MAAM,SAAA,GAAO,MAAA,aAAe,KACxD,WAAM,MAAA,uEAEV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,iEACV,GAAM,GAAM,EAAO,UAAU,WAAM,MAAA,GAAC,MAAA,OAAI,IAAO,EAAM,GACrD,MAAI,IAAM,MACR,GACI,EAAI,MAAM,MAAO,EAAG,MACpB,yGAGN,GAAW,EAAI,OACR,GAkCX,YAAuB,EAAiB,GAEtC,EACI,GAAgB,GAChB,WAAM,MAAA,wDACV,EACI,GAAW,MACP,MAAM,QAAQ,IAAY,EAAQ,MAAM,SAAA,GAAK,MAAA,aAAa,MAC9D,WACI,MAAA,kFAGR,GAAM,GAAmB,GAAW,KACpC,GAAI,CAAC,GAEH,EAAU,GACV,OAAW,KAAW,GAAO,oBAC3B,EAAQ,KAAK,EAAO,oBAAoB,IAI5C,GAAM,GACF,EAAmB,EAAQ,OAAO,SAAA,GAAY,MAAA,CAAC,EAAS,YAAa,KAGnE,EAAmB,EAAQ,OACjC,EAAU,EAAQ,OAAO,SAAA,GAAY,MAAA,GAAS,YAC9C,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,kEACF,kCAAiC,EAAgB,kBACjD,eAER,GAAM,GAAmB,GACnB,EAAA,EAAA,UAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,EACI,EAAM,KAAK,SAAA,GAAK,MAAA,IAAK,OACrB,WAAM,MAAA,iMAGV,EACI,EAAM,OAAS,EACf,WAAM,MAAA,iEACF,oBAAmB,EAAM,KAAI,aAErC,GAAM,GAA6B,GACnC,SAAQ,QAAQ,SAAC,EAAG,GAClB,AAAI,EAAM,IAAM,MACd,GAAW,EAAE,MAAQ,EAAM,MAG/B,AAAI,GAAyB,MAG3B,EAAsB,QAAQ,SAAA,GAAK,MAAA,GAAW,EAAE,MAAQ,OAEnD,CAAC,MAAK,EAAE,MAAO,GA2CxB,YAAsC,GAEpC,MAAO,GAAO,WAAW,GAG3B,YAAoB,GAClB,GAAM,GAAmB,EAAM,OAAO,SAAA,GAAK,MAAA,IAAK,OAAM,OACtD,GAAI,EAAmB,EACrB,KAAM,IAAI,OACN;sEC7XR,AAuCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAuB,CAAC,EAAG,GACjC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,IAAI,IAAK,EAC5B,KAAiB,OAEV,IAAM,EAAG,CAAC,KAAI,KC/C3B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,SAAS,GAC7B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAW,EAAG,CAAC,UAAS,KChDrC,AAyCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAK7B,EAAW,GAAW,SAAC,GAI3B,GAAM,GAAQ,GAAI,GAAS,GAAI,KAEzB,EAAW,SAAC,GAChB,GAAM,GAAO,EAAI,EAAI,GAAQ,GAAI,KACjC,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAEL,IAAa,EAAG,CAAC,YAAW,KC9DzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAoB,EAAM,EAAG,OAC1C,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAS,OAG1D,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GAAgB,GAClB,EAAI,MAAO,GAAoB,EAAM,EAAG,QAC5C,EAAM,EAAQ,EAAK,GACnB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,iBAAkB,EAAM,SAAQ,GAEzD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCnE3B,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAGhB,GAAM,GAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IACN,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,IAAI,EAAW,GACnC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAEzB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KC/F3B,AAsDA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAAU,EAAgB,EAAQ,SAAU,cAKlD,GAHA,AAAI,IAAS,IACX,GAAO,EAAQ,KAAO,GAEpB,IAAS,EAAQ,KAAO,EAC1B,KAAM,OACF,gEACA,oBAAmB,EAAQ,KAAI,iBAAiB,IAGtD,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACX,EAAO,GAAI,EAAQ,EAAM,IACzB,EAAU,GAAI,EAAQ,GACtB,EACF,GAAI,GAAK,EAAS,WAAY,GAAI,GAAI,GAAI,GAAU,EAAM,KAC9D,SAAK,CAAC,IACC,GAGH,EAA2B,CAAC,OAAQ,GACpC,EAAyB,CAAC,KAAI,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAY,MAGZ,IAAa,EAAG,CAAC,YAAW,KCpFzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAO,GAAI,EAAI,EAAM,IACrB,EAAI,GAAI,EAAI,GACZ,EAAI,GAAI,GACR,EAAI,GAAI,EAAG,GACX,EAAI,GAAI,GACR,EAAM,GAAI,EAAQ,EAAM,EAAE,OAAQ,GAExC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,MAGI,IAAY,EAAG,CAAC,WAAU,KChFvC,AAyCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA2B,CAAC,EAAG,EAAI,EAAG,GAE5C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,WAAW,EAAI,IAClC,EAAgC,KAAiB,OAGjD,IAAa,EAAG,CAAC,YAAW,KCtDzC,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,WAAW,IAAK,EACnC,KAAiB,OAGV,IAAa,EAAG,CAAC,YAAW,KC9CzC,AAwCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,QAC1C,EAAK,EAAgB,EAAG,IAAK,YAAa,QAChD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA0B,CAAC,EAAG,EAAI,EAAG,GAC3C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IACjC,EAAgC,KAAiB,OAEjD,IAAY,EAAG,CAAC,WAAU,KCnDvC,AA0CA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,UAA2B,EAAG,MAAO,EAAG,OAGjC,GAAW,GAAU,EAAG,GAAI,GAAW,GAAW,EAAG,QAGjD,IAAa,EAAG,CAAC,YAAW,KCpDzC,AAwDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAC7B,EAAY,EAEd,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAEA,EAEJ,MAAI,GAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UAC9C,EAAI,EAAI,QAER,EAAI,EAAQ,QAAQ,EAAK,GAG3B,EAAK,CAAC,EAAK,IAEJ,GAGH,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAEhE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KCrHnC,AA2EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GAJiB,AAAA,IAAA,QAAA,GAAA,CAA+C,EAAG,EAAG,IAGtE,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE/B,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAErB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GACE,EAAI,EAAQ,UAAU,EAAK,GACjC,SAAK,CAAC,EAAK,IACJ,GAGH,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAE/D,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KClJvC,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAK,EAAgB,EAAG,IAAK,qBAE7B,EAAkC,CAAC,EAAG,GACtC,EACuB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,oBAAmB,GAErE,EAAS,EAAO,UACH,GAAmB,EACnB,GAEnB,MAAO,CAAC,OAAQ,EAAO,GAAI,QAAS,EAAO,OAGhC,IAAoB,EAAG,CAAC,mBAAkB,KC7EvD,YAsCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAM,EAAO,WACpB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAoB,GAAc,GAAQ,GACzD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC7C1C,YAuCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAK,EAAO,WACnB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAmB,GAAc,GAAQ,GACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC9C1C,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAS,GAA0B,EAAG,MAAO,GAC7C,EAAc,EAAO,GACrB,EAAa,GAAc,GAI3B,EAAW,GAAW,SAAC,GAC3B,GAAM,GAAmB,GAAO,GAE1B,EAAU,EAAiB,QAAU,EAAE,MACzC,EACA,GAAK,EAAG,EAAiB,OACvB,EAAM,GAAI,EAAS,GACnB,EAAQ,GAAI,EAAK,EAAM,GAEvB,EAAW,SAAC,GAChB,GAAM,GAAkB,EAAE,MAAM,QAChC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,GAAI,EAAI,EAAY,GAAK,EAAE,MAAO,YAAa,GAC5D,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAGL,IAAO,EAAG,CAAC,MAAK,KCpC7B,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAGpD,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GACF,GAA+B,EAAI,MAAO,GAC9C,EAAM,EAAQ,EAAG,GACjB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAGhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAEH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,AAuDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAEH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AAmCA,YAAmC,GACjC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAQ,GACR,EAAe,CAAC,GAChB,EAA2B,GACjC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,SAAK,CAAC,IACC,EAAQ,OAAO,IACrB,CAAC,EAAG,GAAK,KAAiB,SAAU,EAAO,EAAc,MAGjD,IAAS,EAAG,CAAC,QAAO,KC9CjC,AA4CA,YACI,EAAsB,EACtB,GADsB,AAAA,IAAA,QAAA,GAAA,MACtB,AAAA,IAAA,QAAA,GAAA,IACF,EAAI,EAAgB,EAAG,IAAK,WAC5B,GAAM,GAAO,GAAe,EAAM,EAAE,OAC9B,EAAQ,GAAK,EAAG,EAAM,GACxB,EAAgB,EAAM,MAC1B,AAAK,GACH,GAAgB,GAAqB,EAAM,MAAO,IAEpD,GAAM,GACF,GAAO,GAAI,GAAK,EAAG,WAAY,EAAQ,EAAO,KAC5C,EAAW,GAAK,EAAY,EAAM,GACxC,MAAO,CAAC,KAAM,EAAO,SAAQ,MAGlB,IAAU,EAAG,CAAC,SAAQ,KChBnC,YACI,EAA2B,EAC3B,EACA,GAOF,OANM,GAAQ,EAAgB,EAAM,OAAQ,gBACtC,EAAK,GAAqB,EAAG,IAAK,gBAClC,EAAK,GAAqB,EAAG,IAAK,gBAEpC,EAAQ,EACN,EAAY,GACT,EAAI,EAAG,EAAI,EAAU,OAAQ,KACpC,GAAM,GAAS,EAAU,GAAG,EAAO,EAAG,GAAI,EAAG,IAC7C,EAAU,KAAK,EAAO,IACtB,EAAU,KAAK,EAAO,IACtB,EAAQ,EAAO,GAIjB,OAFM,GAAmB,GACnB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,GAAK,EACzC,EAAK,KAAK,EAAU,IACpB,EAAK,KAAK,EAAU,EAAI,IAE1B,MAAO,CAAC,EAAM,MAEH,IAAe,EAAG,CAAC,cAAa,KCpE7C,AA4CA,YACI,EAAsC,EAAoB,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,eAC5C,EAAc,EAAQ,KACtB,EAAW,EAAQ,KACzB,GAAI,EAAc,EAChB,KAAM,IAAI,OACN,+DACG,GAAW,MAEpB,GAAI,EAAW,EACb,KAAM,IAAI,OAAM,gDAAgD,GAElE,EAAO,GAAQ,KAAK,SACpB,GAAM,GACF,IAAa,EAAI,EAAQ,EAAS,CAAC,EAAG,KAAO,EAC3C,EAAM,EAAO,cACf,SAAA,GAAW,MAAA,GAAQ,YAAY,EAAU,EAAY,EAAY,IACjE,CAAC,SAAQ,IAEb,MAAO,KAAa,EAAI,EAAQ,EAAK,CAAC,EAAI,OAAqB,KAGpD,IAAc,EAAG,CAAC,aAAY,KCpE3C,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,GAAY,MAAA,GAAQ,SAAS,EAAI,IAEjE,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KC1DrC,AAwCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCrD7B,AA0CA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAI,EAAG,QAAU,aACf,GAAM,GAAI,GAAS,GAAK,IAClB,EAAI,GAAU,GAAK,IACzB,MAAO,IAAQ,EAAG,GAGpB,MAAO,GAAQ,SAAS,IAGpB,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCtBrC,YACI,EAAyB,GAC3B,GAAM,GAAM,EAAgB,EAAI,KAAM,gBAChC,EAAM,EAAgB,EAAI,KAAM,gBAEtC,EACI,EAAI,OAAS,GAAK,EAAI,OAAS,EAC/B,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,EAAQ,EAAK,CAAC,EAAG,KAC9B,MAAO,IAAO,EAAM,MAGT,IAAe,EAAG,CAAC,cAAa,KCtD7C,AAmDA,YACI,EAAiB,EAAmC,GAAA,AAAA,IAAA,QAAA,GAAA,GACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAA0B,SAAC,EAAS,GACxC,SAAK,CAAC,IACC,EAAQ,IAAI,EAAI,EAAU,IAG7B,EAAoB,CAAC,SAAQ,EAAE,cAAa,GAC5C,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAM,EAAG,CAAC,KAAI,KC5C3B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,qDACH,GAAI,EAAG,CAAC,GAAW,MAGf,IAAQ,EAAG,CAAC,OAAM,KCT/B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCV/B,YACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,EACvD,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCX/B,YACI,EACA,EAIA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,GACnD,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCxC/B,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAEnC,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,cAAc,EAAG,KAAI,kCACvB,EAAW,SAEnB,EACI,EAAS,SAAW,EAAW,OAC/B,WAAM,MAAA,qBACF,EAAS,OAAM,kCAAkC,EAAW,SAEpE,EACI,EAAG,MAAM,OACL,SAAC,EAAG,EAAG,GACL,MAAI,GAAI,GAAK,GAAK,EAAW,OACpB,GACD,GAAI,EAAS,EAAI,GAAG,GAAK,EAAS,EAAI,GAAG,IACtC,EAAW,EAAI,KACnB,EAEA,GAET,IACJ,WAAM,MAAA,4BAA4B,EAAG,MAAM,MAAM,GAAE,kBAC/C,EAAS,WAAU,qCACnB,EAAW,aAEnB,GAAM,GAA0B,SAAA,GAC5B,MAAA,GAAQ,eAAe,EAAI,EAAY,IAErC,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,SAAQ,GAExD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,KCrHjD,AAwDA,YACI,EAAqB,EACrB,EAA0B,EAC1B,EAAqC,GACvC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,AAAI,GAAW,MACb,GAAU,GAEZ,AAAI,IAAQ,GACV,GAAM,SAGR,GAAM,GAAK,EAAgB,EAAO,IAAK,WACnC,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,yDACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAC1C,EACF,CAAC,EAAS,eAAgB,EAAS,eAOnC,EACJ,AAAI,IAAQ,OACV,EAAc,GACV,CAAC,EAAS,aAAc,EAAS,aAAc,GAEnD,EAAc,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,IAG7B,GAAM,GAAgB,EAAS,KAAO,GAAK,EAAS,KAAO,EACrD,EAAA,GAAA,CAAA,EAAA,SAAA,EAAA,SAAA,EAAA,GAAC,EAAA,EAAA,GAAiB,EAAA,EAAA,GAElB,EAAe,EAAgB,EAAM,QACrC,EACF,EAAgB,EAAM,GAAe,EAAK,EAAU,GAElD,EAAY,IAAgB,MAC9B,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAChD,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAC9C,EAAI,IAEJ,EAAM,EAAgB,EAAI,GAAe,EAAG,EAAU,GAE5D,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,EAMT,YACI,EAA8B,EAC9B,GACF,GAAM,GAAW,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,EAAa,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KACpC,EAAiB,EAAW,OAAO,EAAU,GAC7C,EAAc,EAAW,IAAI,SAAC,EAAG,GAAM,MAAC,GAAI,EAAe,GAAK,GAAK,IACrE,EAAS,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAY,KAClD,EAAW,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAS,GAAI,EAAO,MACzD,EAAQ,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAG,EAAY,MACvD,MAAO,CAAC,EAAU,GAMpB,YACI,EAA+B,GAGjC,GAAM,GAAqB,EAAY,IAAI,SAAC,EAAG,GAC7C,MAAO,GAAK,GAAI,GAAM,GAAS,GAAK,KAEhC,EAAgB,EAAmB,IAAI,SAAA,GAAK,MAAA,GAAI,IAIhD,EAAgB,EAAc,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAI,KACtD,EAAc,EAAc,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAc,KAClE,MAAO,GAAc,IAAI,SAAC,EAAG,GAC3B,MAAO,CAAC,EAAc,GAAI,EAAY,SAI7B,IAAO,EAAG,CAAC,MAAK,KChK7B,AAsDA,YACI,EAAyB,SACvB,EAAQ,EAAgB,EAAM,OAAQ,OACtC,EAAO,EAAgB,EAAK,MAAO,OACvC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,GAAM,GAAoB,CAAC,EAAG,EAAO,EAAG,GAClC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,EAAO,GAC7B,SAAK,CAAC,EAAO,EAAM,IACZ,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AA0CA,YAAkC,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAS,EAAgB,EAAO,QAAS,SAEzC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAGH,EAAsB,CAAC,EAAG,EAAI,MAAO,GAC3C,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AA8DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,QAE3B,EAA+B,SAAC,GACpC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAEhB,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,KAAK,EAAW,GACpC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAGzB,MAAO,IAGH,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,EAAE,SAAQ,GAExC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAO,EAAG,CAAC,MAAK,KChG7B,AAiCA,YACI,EAAoB,EACpB,GACF,GAAM,GAAO,GAAc,GACvB,EAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,IAEd,MAAO,GAAO,WAAW,EAAQ,EAAO,MAG7B,IAAO,EAAG,CAAC,MAAK,6OC1B7B,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,cAEJ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,wBACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,AAyCA,GAAA,IAAA,WAUE,WACI,EAAc,EAAsB,EACpC,EAAqB,GACvB,KAAK,KAAO,EACZ,KAAK,OAAS,EACd,KAAK,MAAQ,EACb,KAAK,QAAU,IACf,KAAK,UAAY,EACjB,AAAI,KAAK,WACP,MAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,EACvC,KAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,GAEzC,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,OAAS,GAAgB,EAAU,YAInC,SAAA,UAAA,UAAP,WACE,GAAI,CAAC,MAAM,KAAK,UACd,GAAM,GAAQ,KAAK,QACnB,YAAK,QAAU,IACR,EAKT,OAFI,GAAiB,EACjB,EAAU,GACP,CAAC,IACN,GAAI,GAAE,OAAU,EAAE,OAAU,EAAC,OAC7B,EACE,GAAK,EAAI,KAAK,SAAW,EACzB,EAAK,EAAI,KAAK,SAAW,EACzB,EAAI,EAAK,EAAK,EAAK,QACZ,GAAK,GAAK,IAAM,GAEzB,GAAM,GAAM,KAAK,KAAK,GAAO,KAAK,IAAI,GAAK,GAC3C,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EACzC,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EAEzC,AAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,GAAU,IAId,MAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,MAAK,QAAU,KAAK,aAAa,IAE5B,KAAK,aAAa,IAInB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,OAAS,MAAQ,KAAK,QAAU,UAChC,EAEF,KAAK,MAAM,IAIZ,EAAA,UAAA,iBAAR,SAAyB,GACvB,MAAO,IAAS,KAAK,OAAS,GAAS,KAAK,OAEhD,KAIA,GAAA,WASE,WACI,EAAe,EAAc,EAC7B,GACF,KAAK,MAAQ,EACb,KAAK,KAAO,EAAI,EAChB,KAAK,MAAQ,EAEb,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,MAAQ,GAAgB,EAAU,YACvC,KAAK,MAAQ,GAAI,IAAY,EAAG,EAAG,EAAO,GAAO,KAAK,SAEtD,AAAI,EAAQ,EACV,KAAK,EAAI,EAAS,EAAI,EAEtB,KAAK,EAAI,EAAS,EAAI,EAExB,KAAK,EAAI,EAAI,KAAK,KAAK,EAAI,KAAK,GAI3B,SAAA,UAAA,UAAP,WAEE,OADI,GAAY,EAAY,EAAY,EAAW,EAAW,KAE5D,EACE,GAAI,KAAK,MAAM,YACf,EAAI,EAAK,KAAK,EAAI,QACX,GAAK,GAMd,GALA,GAAK,EAAI,EACT,EAAK,EAAI,EACT,EAAK,EAAK,KAAQ,EAAK,EACvB,EAAM,GAAM,EAAO,KAAK,EAAK,GAAI,EAAI,KAAK,IAAI,IAC9C,EAAI,KAAK,QACL,EAAI,GAAM,KAAK,IAAI,GAAK,EAC1B,MAGJ,SAAK,EAAI,KAAK,KAAQ,KAAK,EAAI,EAC/B,AAAI,KAAK,MAAQ,GACf,IAAK,KAAK,IAAI,KAAK,QAAS,EAAI,KAAK,QAEhC,KAAK,aAAa,IAGnB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,QAAU,UACV,EAEF,KAAK,MAAM,IAEtB,KAEA,GAAA,WAME,WACI,EAAS,EAAS,EAClB,GAFJ,GAAA,GAAA,KAaE,GAZE,AAAA,IAAA,QAAA,GAAA,GAAS,AAAA,IAAA,QAAA,GAAA,GAoBL,KAAA,eAAiB,WACrB,MAAC,GAAK,OAAS,MAAQ,EAAK,QAAU,WAnBxC,KAAK,IAAM,EACX,KAAK,MAAQ,EAAM,EACnB,KAAK,MAAQ,EACb,AAAI,GAAQ,MACV,GAAO,KAAK,UAEd,AAAI,MAAO,IAAS,UAClB,GAAO,EAAK,YAGV,CAAC,KAAK,kBAAoB,KAAK,OAAS,EAC1C,KAAM,IAAI,OACN,0BAA0B,EAAG,MAAM,EAAG,gCAE5C,KAAK,OAAS,GAAgB,GAOxB,SAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,iBACA,EAEF,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,WACE,MAAO,MAAK,aAAa,KAAK,IAAM,KAAK,MAAQ,KAAK,WAE1D,KCxNA,AAwCA,YACI,EAAoB,EAAe,EACnC,EAAsC,GAOxC,GARqC,AAAA,IAAA,QAAA,GAAA,GACnC,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAQ,MACV,GAAO,GAET,AAAI,GAAS,MACX,GAAQ,WAEN,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,OAAM,yBAAyB,GAI3C,OAFM,GAAS,GAAI,IAAU,EAAO,EAAM,EAAO,GAC3C,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAc,EAAG,CAAC,aAAY,KC5D3C,AAuCA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,yBAAyB,GAK3C,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAuB,GAC1D,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAe,EAAG,CAAC,cAAa,KCtD7C,AA4CA,YACI,EAAoB,EAAY,EAAY,EAC5C,GADoB,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,WAI9C,OAFM,GAAM,GAAO,EAAO,GACpB,EAAS,GAAI,IAAc,EAAQ,EAAQ,KAAM,GAC9C,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAgB,EAAG,CAAC,eAAc,KCvD/C,YAwCyB,EAAsB,GAC7C,GAAc,GACd,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,EAC3B,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAkB,KACxB,MAAO,IAAW,EAAQ,EAAO,EAAe,GC/ClD,YA8CI,EAAe,EAAc,EAC7B,GACF,GAF+B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAA,WACE,IAAS,EACX,KAAM,IAAI,OAAM,8BAGlB,GAAM,GAA+B,WACnC,GAAM,GAAgB,IAAU,EAC1B,EAA8B,EAAQ,GAAQ,EAAO,EACrD,EAA8B,EAAO,GAAS,EAAO,EAE3D,GAAI,GAAiB,GACjB,EACF,MAAO,IAAM,CAAC,GAAI,GAGpB,GAAM,GAAc,KAAK,IAAI,KAAK,KAAM,GAAO,GAAS,IAClD,EAAS,GAAoB,EAAa,GAEhD,AAAI,EAAO,GAAS,IAAS,GAG3B,GAAO,IAGT,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,IAGpB,EAAoB,CAAC,MAAK,EAAE,KAAI,EAAE,KAAI,EAAE,MAAK,GAEnD,MAAO,GAAO,cACH,EAAS,GAAiB,KAAiB,GAC3C,GCnFb,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAE7B,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,WAAW,GAC/B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAa,EAAG,CAAC,YAAW,KChDzC,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,KAAK,IAGhB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,MAAM,IAGjB,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC5D/B,AA6DA,YACI,EAAiB,GACnB,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAA+B,SAAC,GACpC,GAAM,GAAO,GAAe,EAAM,EAAG,OACrC,GAAI,EAAG,OAAS,EACd,MAAO,IAAM,GAEf,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,MAAO,GAAQ,EAAK,EAAG,QAGnB,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,KAAM,GAEnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAS,MAGT,IAAU,EAAG,CAAC,SAAQ,KClFnC,AA6BA,YAAoB,GAClB,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCrCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAGV,IAAQ,EAAG,CAAC,OAAM,KChD/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GAGH,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KCe7B,YACI,EAAiB,EACjB,EAAsC,EACtC,EAAqB,EACrB,GADqB,AAAA,IAAA,QAAA,GAAA,CAAqC,EAAG,IAC7D,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EACF,EAAgB,EAAiB,kBAAmB,mBAClD,EACF,EAAgB,EAAiB,kBAAmB,mBAEpD,EAAM,EACN,EAAe,GAMnB,GALA,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGvD,IAAe,OACjB,KAAM,IAAI,OACN,sFAIN,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,MAAM,KAAO,EAC9B,WACI,MAAA,qEACA,wBAAuB,EAAiB,MAAM,GAAE,OACxD,EACI,EAAiB,MAAM,KAAO,EAC9B,WAAM,MAAA,+DACF,8BAA6B,EAAiB,MAAM,GAAE,OAE9D,GAAM,GAAa,EAAiB,MAAM,GACpC,EAAoB,EAAiB,MAAM,GACjD,EACI,EAAiB,MAAM,KAAO,EAAa,EAC3C,WACI,MAAA,qEACA,YAAW,EAAa,EAAiB,MACzC,YAAW,EAAiB,MAAM,GAAE,OAE5C,GAAM,GAAY,GACd,EAAK,EAAkB,EAAS,EAAK,EAAY,GAC/C,EAAkB,EAClB,EACF,GAAO,EAAW,EAAkB,EAAiB,QAAS,GAElE,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KCxFnD,YACI,EAAsB,uHAClB,SAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aAEnC,EACI,EAAG,QAAU,EAAG,MAChB,WAAM,MAAA,kDACF,EAAG,MAAK,YAAY,EAAG,MAAK,OAEpC,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAEtE,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAExD,CAAA,EAAM,EAAG,eAAjB,SAAQ,EAAA,OACA,CAAA,EAAM,EAAG,eAIvB,IAJM,EAAQ,EAAA,OACR,EAAO,GAAI,KAAI,GAEjB,EAAa,EACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,AAAK,EAAK,IAAI,EAAM,KAClB,IAMJ,IAFM,EAAS,GAAI,IAAa,CAAC,GAAa,EAAG,OAC3C,EAAU,GAAI,IAAa,CAAC,GAAa,SACtC,EAAI,EAAG,EAAI,EAAG,EAAI,EAAM,OAAQ,IACvC,AAAK,EAAK,IAAI,EAAM,KAClB,GAAO,OAAO,GAAK,EAAM,GACzB,EAAQ,OAAO,GAAK,EACpB,KAGJ,MAAA,CAAA,EAAO,CAAC,EAAO,WAAY,EAAQ,qBAExB,IAAiB,GCxF9B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC7C7B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AA6BA,YACI,EAAwB,EAAe,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,CAAC,GAAQ,CAAC,OAEhB,IAAU,EAAG,CAAC,SAAQ,KCtCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AAgDA,YAAoC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxD,GAAM,GAAU,EAAgB,EAAQ,SAAU,UAAW,WAK7D,GAHA,AAAI,IAAQ,IACV,GAAM,EAAQ,KAAO,GAEnB,IAAQ,EAAQ,KAAO,EACzB,KAAM,OACF,4DACA,oBAAmB,EAAQ,KAAI,gBAAgB,IAGrD,GAAM,GAAwB,CAAC,OAAQ,GACjC,EAAsB,CAAC,IAAG,GAEhC,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAI,EAAQ,QAAQ,EAAS,GACnC,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAU,EAAG,CAAC,SAAQ,KCzEnC,AAyCA,YAAc,GACZ,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,qDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAoB,CAAC,MAAK,GAEhC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAU,EAAM,KAAK,EAAO,GAC5B,EAAS,EAAQ,IAAI,GAC3B,MAAO,GAAO,QAAQ,EAAM,QAC3B,EAAgC,KAAqB,OAG7C,IAAM,EAAG,CAAC,KAAI,KC5D3B,AA0CA,YAAe,GACb,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,sDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAqB,CAAC,MAAK,GAEjC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAoB,EAAQ,EAAO,CAAC,EAAO,IAC3C,EAAS,EAAQ,KAAK,GAC5B,MAAO,GAAQ,EAAQ,EAAM,QAC5B,EAAgC,KAAqB,OAG7C,IAAO,EAAG,CAAC,MAAK,KC7D7B,AAgDA,YAAgB,GACd,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EACvB,EACJ,GAAI,GAAsB,GACxB,GAAM,GAAe,EAAQ,EAAO,CAAC,EAAO,IAC5C,EAAM,GAAK,QAIX,GAAM,GAAc,CAAC,EAAO,EAAK,GAAqB,IAChD,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IACzC,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IAEzC,EACF,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACjE,EAA0B,EAC5B,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACnE,GAAO,KAEL,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EACF,EAAQ,GAAQ,EAAG,GAAI,CAAC,EAAY,GAAI,EAAY,KACxD,EAAM,GAAK,GAIb,GAFA,EAAM,GAAK,GAEP,EAAM,OAAS,GAAK,EAAM,MAAM,KAAO,GACzC,GAAM,GAAO,EACP,EAAQ,EAAM,MAAM,GAC1B,EAAM,EAAQ,EAAK,CAAC,EAAO,EAAI,MAAM,GAAK,EAAO,EAAI,MAAM,KAC3D,EAAK,UAEP,MAAO,MAGI,IAAQ,EAAG,CAAC,OAAM,iBC3D3B,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAa,GACjB,GAAI,MAAQ,IAAqB,SAC/B,EACI,EAAE,MAAM,GAAQ,IAAoB,EACpC,WAAM,MAAA,kDACV,EACI,GAAI,OAAM,GAAiB,KAAK,EAAE,MAAM,GAAQ,QAEpD,GAAM,GAAY,EAAgB,OAAO,SAAC,EAAO,GAC/C,MAAI,KAAU,IACZ,IAAS,GAEJ,GACN,GACH,EACI,GAAa,EACb,WAAM,MAAA,4DACV,GAAM,GAAW,EAAgB,QAAQ,IAGzC,GAAI,IAAa,IACf,GAAM,GAAQ,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,EAAI,IAC/D,EAAgB,GAAY,EAAE,MAAM,GAAQ,EAE9C,EACI,EAAE,MAAM,KAAU,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IACvD,WAAM,MAAA,gEACV,EAAa,EAGf,MAAO,GC1DT,AA+DA,YACI,EAAsB,EAAkC,GAAA,AAAA,IAAA,QAAA,GAAA,GAC1D,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAG,OAAO,GACvC,EAAa,GAAiB,EAAI,EAAiB,GACzD,MAAO,GAAQ,MAAM,EAAI,EAAY,IAGjC,EAAuB,CAAC,EAAG,GAC3B,EAAoB,CAAC,gBAAe,EAAE,KAAI,GAEhD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCjF/B,AA+CA,YAAe,EAAe,GAC5B,EACI,EAAM,QAAU,UAChB,WAAM,MAAA,mDAAmD,EAAM,QAEnE,GAAI,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACpD,EAAQ,EAAM,KAAO,EAEvB,EACJ,GAAI,GAAa,MAAQ,EAAY,GAEnC,GAAM,GAAQ,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAC7B,EAAO,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAClC,EAAK,EAAM,MAAM,OAAS,GAAK,EAC/B,EAAgB,GAAM,EAAO,EAAO,GACpC,EAAqB,UACZ,GAAa,MAAQ,EAAY,GAE1C,GAAM,GAAa,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KACxC,EAAW,EAAM,MAAM,OAAS,GAAK,EAAY,EACjD,EAAgB,GAAO,CAAC,EAAO,GAAM,IAAc,EAAM,MAAM,OAAS,GACxE,EAAqB,MAErB,GAAgB,EAIlB,GAAM,GAAa,GAAU,GACvB,EACF,EAAQ,GAAQ,EAAe,GAAa,CAAC,EAAO,IAElD,EAAM,GAAI,GAGV,EAAO,KAAK,MAAM,EAAqB,GAAK,EAC5C,EAAa,GAAK,GAClB,EAAa,GAAK,GAClB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GACxB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GAExB,EAAc,EAAc,MAAM,QACxC,SAAY,EAAc,MAAM,OAAS,GAAK,EAEvC,EACH,GAAQ,EAAqB,GAAI,EAAqB,IAAK,MAGpD,IAAO,EAAG,CAAC,MAAK,KClG7B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AAoDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,qBAC7B,EAAK,EAAgB,EAAG,IAAK,qBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,kBAAkB,EAAI,GAC1C,SAAK,CAAC,EAAI,IACH,GAGH,EAAkC,CAAC,EAAG,EAAI,EAAG,GAC7C,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAmB,MAGnB,IAAoB,EAAG,CAAC,mBAAkB,KC1EvD,AAwCA,YAAoC,EAAsB,GACxD,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,MAAO,GAAQ,EAAI,GAAa,EAAG,MAAO,GAAM,aAGrC,IAAU,EAAG,CAAC,SAAQ,KC7CnC,AAyCA,YACI,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAChC,GAAM,GAAW,GAAqB,EAAS,UAAW,SAK1D,GAHA,EACI,EAAS,QAAU,EAAG,WAAM,MAAA,yCAE5B,EAAS,SAAW,EACtB,MAAO,IAAW,EAAS,GAAI,GAGjC,GAAM,GAAO,EAAS,GAAG,KACnB,EAAQ,EAAS,GAAG,MACpB,EAAQ,EAAS,GAAG,MAE1B,EAAY,GAAQ,EAAM,WAAM,MAAA,uCAEhC,EAAS,QAAQ,SAAA,GACf,GACI,EAAO,EAAE,MACT,yDACJ,EACI,IAAU,EAAE,MACZ,WAAM,MAAA,4DAGZ,GAAM,GAAkB,EAAS,IAAI,SAAA,GAAK,MAAA,IAAW,EAAG,KAOxD,MAAO,IAAO,EAAiB,MAGpB,IAAQ,EAAG,CAAC,OAAM,KC7E/B,AAwCA,YAAiC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAEhB,IAAO,EAAG,CAAC,MAAK,KClD7B,AA+DA,YACI,EAAsB,EAAiB,EAAe,EACtD,EAAe,EAAa,EAAkB,EAC9C,GADA,AAAA,IAAA,QAAA,GAAA,GAAe,AAAA,IAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAAkB,AAAA,IAAA,QAAA,GAAA,GAC9C,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAE3B,EAA+B,SAAC,GACpC,AAAI,GAAW,MACb,GAAU,GAAI,OAAM,EAAM,SAG5B,GAAM,GAAe,GAAW,GAChC,GAAI,EAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAI,IAAiB,GAAK,IAAgB,EACxC,KAAM,IAAI,OACN,iEAGN,GAAI,IAAiB,GAAK,IAAmB,EAC3C,KAAM,IAAI,OACN,oEAGN,GAAM,GAAsB,EAAG,KAAO,EAAM,OAGtC,EAAa,GAAW,GACxB,EAAW,EAAG,MAAM,QAC1B,EAAW,QAAQ,SAAA,GACjB,EAAM,GAAQ,EACd,EAAI,GAAQ,EACZ,EAAS,OAAO,EAAM,EAAG,KAE3B,EAAK,EAAQ,EAAI,GAEX,GAAA,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GACJ,EAAA,EAAA,MACA,EAAA,EAAA,IACA,EAAA,EAAA,QAKF,EAAQ,EACR,EAAM,EACN,EAAU,EAEV,GAAM,GAAa,GAAW,GAE9B,EAAW,QAAQ,SAAA,GACjB,EAAI,GAAQ,EAAM,GAAQ,EAC1B,EAAQ,GAAQ,IAIlB,GAAM,GAAO,GAAgB,EAAO,EAAK,GAEnC,EAAW,EAAK,OAAO,SAAC,EAAG,GAAS,MAAA,GAAW,QAAQ,KAAU,KAEjE,EAAa,EAAQ,MAAM,SAAA,GAAK,MAAA,KAAM,IAC5C,GAAI,EACF,MAAO,GAAQ,GAAM,EAAI,EAAO,GAAO,GAGzC,GAAM,GAAM,EAAQ,aAAa,EAAI,EAAO,EAAK,GACjD,MAAO,GAAQ,EAAK,IAGhB,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAC/B,MAAK,EACL,IAAG,EACH,QAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,YAAW,EACX,eAAc,GAGhB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAe,EAAG,CAAC,cAAa,KCvJ7C,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,kEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,gFAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,sEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,wEAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GCjElD,YAgDI,EACA,EACA,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,0EAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,SAAQ,GACJ,EACG,GAAW,EAAQ,EAAO,EAAe,GCpElD,AAmDA,YACI,EAAiB,EAAO,GAAP,AAAA,IAAA,QAAA,GAAA,GAAO,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAK,EAAgB,EAAG,IAAK,QACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAU,EAAG,MAAM,EAAG,MAAM,OAAS,GAC3C,GAAI,EAAI,EACN,KAAM,IAAI,OACN,uDAAuD,EAAO,KAC9D,YAAW,IAGjB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,EAAC,EAAE,OAAM,GAE7B,EAAA,EAAA,cAAA,SAAA,GAAA,MAAA,GAAA,KAAA,EAAA,EAAA,IAAA,EAAA,KAAA,GAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAKf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAO,EAAG,CAAC,MAAK,KC3E7B,AA4CA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,qCAKlB,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAsB,GACzD,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAkB,EAAG,CAAC,iBAAgB,KC3DnD,AA6EA,YACI,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAEnB,GAAM,GAAK,EAAgB,EAAG,IAAK,SAAU,MAC7C,EAAO,EAAG,KAAO,EAAG,WAAM,MAAA,yCAE1B,GAAM,GAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAC1B,EAAA,EAAA,UAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAS,EAAG,CAAC,QAAO,KC3FjC,AA8CA,YACI,EAAiB,EAAiC,GACpD,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EACF,EAAgB,EAAY,aAAc,qBAAsB,SACpE,EAAO,GAAM,GAAc,WAAM,MAAA,qCAEjC,GAAM,GAAmC,CAAC,EAAG,EAAI,WAAY,GACvD,EAAiC,CAAC,YAAW,GAE7C,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,mBAAmB,EAAI,EAAa,GACxD,SAAK,CAAC,IACC,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAoB,MAGpB,IAAqB,EAAG,CAAC,oBAAmB,KCnEzD,AA2CA,YAAkB,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACtC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,EACI,GAAQ,CAAC,EAAG,MAAM,QAAU,EAAO,EAAG,MAAM,OAC5C,WACI,MAAA,UAAU,EAAI,gBAAgB,EAAG,MAAM,OAAM,KAAK,EAAG,MAAM,OAAM,MACzE,AAAI,EAAO,GACT,IAAQ,EAAG,MAAM,QAEnB,GAAM,GAAuB,CAAC,MAAO,GAC/B,EAAqB,CAAC,KAAI,GAC1B,EAAiC,SAAC,GACpC,MAAA,GAAQ,QAAQ,EAAI,IACxB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAU,EAAG,CAAC,SAAQ,KC7DnC,YAsCI,EAAyB,EAAkB,EAC3C,GADyB,MAAA,KAAA,QAAA,GAAA,IAEpB,EAAO,aAAa,EAAc,EAAW,EAAM,GCxC5D,YAuB0B,EAAqB,GAE7C,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,AAAI,EAAS,IACX,EAAQ,KAAK,GAOjB,OAHM,GAAW,GAAO,EAAW,SAE7B,EAAM,GAAO,CAAC,EAAQ,OAAQ,EAAU,QAAS,SAC9C,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAM,EAAS,WAAW,EAAQ,IAClC,EAAS,EAAI,EAAU,OAC7B,EAAI,OAAO,IAAI,EAAK,GAEtB,MAAO,GAAI,WCEb,YAA2B,uGACnB,SACF,EAAgB,EAAW,YAAa,aAAc,QAC7C,CAAA,EAAM,EAAW,eAAxB,SAAO,EAAA,OACP,EAAM,GAAU,EAAW,MAAO,GACxC,AAAI,IAAc,GAChB,EAAW,UAEb,CAAA,EAAO,WAGI,IAAa,GCpD1B,AA6CA,YACI,EAA2B,EAC3B,2HAcF,IAbM,EAAU,EAAgB,EAAQ,SAAU,YAC5C,EAAQ,EAAgB,EAAM,OAAQ,WAAY,QAElD,EAAW,GAAe,EAC1B,EAAU,EAAM,KAChB,EAAc,EAAQ,MAE5B,EAAY,EAAU,EAAG,WAAM,MAAA,0BAC/B,GACI,EAAY,MAAM,EAAU,EAAW,GAAU,EAAM,MACvD,qEAEA,EAAc,EACT,EAAI,EAAU,EAAI,EAAW,EAAS,IAC7C,GAAe,EAAY,GAEvB,SACF,EAAY,MAAM,EAAG,GAChB,OAAO,CAAC,GAAc,EAAY,MAAM,EAAW,IACtD,EAAiB,EAAQ,EAAS,GAClC,EAAe,EAAQ,EAAO,CAAC,KACX,CAAA,EAAM,GAAW,WAArC,SAAoB,EAAA,OACpB,EAAU,GAAQ,EAAmB,CAAC,IAEtC,EAAM,GAAO,EAAgB,EAAS,GAG5C,AAAI,IAAW,GACb,EAAQ,UAEV,AAAI,IAAS,GACX,EAAM,UAER,EAAQ,UACR,EAAe,UACf,EAAa,UACb,EAAkB,UAElB,CAAA,EAAO,WAGI,IAAmB,GCzFhC,AAuCA,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAK,EAAgB,EAAG,IAAK,kBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,6BAC/B,GAAS,EAAI,GAYtB,YAAuC,EAAiB,GACtD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAK,EAAgB,EAAG,IAAK,cACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,yBAC/B,GAAK,EAAI,GAGlB,YAAwC,EAAiB,GACvD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,eAC7B,EAAK,EAAgB,EAAG,IAAK,eACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,0BAC/B,GAAM,EAAI,GAGnB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAK,EAAgB,EAAG,IAAK,mBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,8BAC/B,GAAU,EAAI,GAGvB,YAA0C,EAAiB,GACzD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,4BAC/B,GAAQ,EAAI,GAGrB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EAAK,EAAgB,EAAG,IAAK,sBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,iCAC/B,GAAa,EAAI,MAGb,IAAc,EAAG,CAAC,aAAY,KAC9B,GAAqB,EAAG,CAAC,oBAAmB,KAC5C,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAkB,EAAG,CAAC,iBAAgB,KACtC,GAAa,EAAG,CAAC,YAAW,KAC5B,GAAiB,EAAG,CAAC,gBAAe,KCpHjD,AA2CA,YAAsC,EAAiB,GACrD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAS,GAC7C,UACI,6EAGJ,GAAuB,EAAK,MAAO,EAAI,MAAO,wBACvC,GAAI,EAAM,GAanB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,6BACpC,EAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,2BACpC,GAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAWjB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAWrB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAarB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,2BAC7B,EAAK,EAAgB,EAAG,IAAK,2BACnC,UACI,EAAG,MAAO,EAAG,MAAO,sCACjB,GAAkB,EAAI,MAGlB,IAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAA0B,EAAG,CAAC,yBAAwB,KACtD,GAAY,EAAG,CAAC,WAAU,KC5NvC,AAwEA,YACI,EAAsB,EACtB,EAA8B,GADR,AAAA,IAAA,QAAA,GAAA,aACtB,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IAChC,EAAI,EAAgB,EAAG,IAAK,QAE5B,GAAM,GAAO,GAAS,EAAG,EAAK,GAC1B,EAAgB,EAAK,MACzB,GAAI,GACF,GAAM,GAAO,GAAe,EAAM,EAAE,OACpC,EAAgB,GAA+B,EAAK,MAAO,GAE7D,MAAO,GAAQ,EAAM,GAGvB,YACI,EAAW,EAAkB,GAC/B,GAD+B,AAAA,IAAA,QAAA,GAAA,MAC3B,EAAE,OAAS,EACb,MAAO,IAAI,GAIb,GAAI,EAAE,OAAS,GAAK,IAAS,KAC3B,MAAO,IAAS,EAAQ,EAAG,CAAC,KAAM,EAAG,GAIvC,GAAI,EAAE,OAAS,GAAK,MAAO,IAAS,UAChC,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,aAAe,IAAM,EAE7B,MAAO,IAAK,GAAI,GAAI,GAAI,GAAI,GAAO,EAAG,UAAW,IAGnD,KAAM,IAAI,OAAM,qCAAqC,GAIvD,GAAI,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,GAAK,GAE7C,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,OAAS,IAAM,YAEvB,MAAO,IAAK,GAAI,GAAO,GAAI,IAG7B,KAAM,IAAI,OAAM,qCAAqC,GAGvD,KAAM,IAAI,OAAM,gCAAgC,MAGrC,IAAO,EAAG,CAAC,MAAK,KC3I7B,AA2DA,YACI,EAAiB,EAAiB,EAClC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAS,EAAgB,EAAO,QAAS,iBAE/C,GAAiB,EAAI,GACrB,EACI,GAAiB,EAAG,MAAO,EAAG,OAAQ,WAAM,MAAA,8BAEhD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAI,EAAK,GAE3B,EAAS,EAAI,GAAI,EAAI,GAAK,GAC9B,GAAI,GACF,EACI,GAAQ,KAAM,WAAM,MAAA,mDACxB,GAAM,GAAQ,EAAgB,EAAM,OAAQ,iBAC5C,EAAS,GAAI,EAAQ,GAAI,EAAK,GAAI,EAAQ,KAE5C,MAAO,IAAI,EAAI,MAGJ,IAAgB,EAAG,CAAC,eAAc,KCnF/C,AA+CA,YACI,EAA4B,EAC5B,GACF,GAAM,GAAW,EAAgB,EAAS,UAAW,YAAa,SAC5D,EAAW,EAAgB,EAAS,UAAW,aACrD,GAA8B,EAAU,EAAU,GAElD,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,UAAU,EAAU,EAAU,IAGzC,EAA0B,CAAC,QAAS,EAAU,QAAS,GACvD,EAAwB,CAAC,MAAK,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAW,MAGX,IAAY,EAAG,CAAC,WAAU,iBClCnC,EAAuB,EAAsB,EAC7C,GACF,GAAI,EAAc,QAAU,QAC1B,KAAM,IAAI,OACN,2DACA,uBAAsB,EAAc,MAAK,MAE/C,GAAI,EAAc,KAAO,EACvB,KAAM,IAAI,OACN,uDACA,mBAAkB,EAAc,MAAK,MAG3C,GAAM,GAAW,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAC7D,EAAU,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAElE,GAAI,EAAY,SAAW,EACzB,KAAM,IAAI,OACN,iDACA,KAAI,EAAY,OAAM,gBAAgB,EAAO,MAGnD,GAAM,GAAY,EAAa,KAC/B,GAAI,CAAE,GAAa,OAAS,GACtB,EAAa,OAAS,GAAK,IAAc,GAC7C,KAAM,IAAI,OACN,oCACG,GAAa,MAAK,sBAAsB,EAAQ,MAGzD,GAAI,EAAa,QAAU,EAAc,MACvC,KAAM,IAAI,OAAM,qDC/DpB,AAkEA,YACI,EAAkC,EAClC,EAA0B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC5B,GAAM,GACF,EAAgB,EAAe,gBAAiB,gBAAiB,SAC/D,EACF,EAAgB,EAAc,eAAgB,iBAC5C,EAAgB,EAClB,EAAc,eAAgB,gBAAiB,EAAc,OAEjE,GACI,EAAgB,EAAe,EAAa,GAEhD,GAAM,GAA8B,CAClC,cAAe,EACf,aAAc,EACd,aAAc,GAGV,EAA4B,CAAC,YAAW,GAE9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,cACf,EAAgB,EAAe,EAAa,IAChD,EAAgC,KAAiB,GACjD,MAGO,IAAgB,EAAG,CAAC,eAAc,KC9F/C,AA8DA,YAAmB,EAAsB,GACvC,GAAM,GAAW,EAAgB,EAAS,UAAW,WAAY,SAC3D,EAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,GACpC,MAAO,GAAQ,SAAS,EAAI,IAGxB,EAAyB,CAAC,OAAQ,EAAI,QAAS,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KAAqB,OAGvD,IAAW,EAAG,CAAC,UAAS,KC5ErC,YA4B8B,EAAW,GACvC,GAAI,GAAc,KAChB,MAAO,GAAE,MAAM,QAEjB,GAAI,GAAiB,EAAE,MAAO,GAC5B,MAAO,GAET,GAAI,EAAE,MAAM,SAAW,EAAW,QAEhC,OADM,GAAyB,GACtB,EAAI,EAAG,EAAI,EAAE,MAAM,OAAQ,IAClC,AAAI,EAAW,IAAM,MAAQ,EAAE,MAAM,IAAM,KACzC,EAAa,KAAK,EAAE,MAAM,IAE1B,EAAa,KAAK,EAAW,IAGjC,MAAO,GAGT,MAAO,GC/CT,AAoDA,YACI,EAAsB,EAAc,EACpC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAUnC,GARA,EACI,EAAG,QAAU,UACb,WAAM,MAAA,8DACF,sBAAqB,EAAG,MAAK,sBACrC,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,qDAAqD,EAAI,MAE/D,IAAS,EACX,MAAO,aAAa,GAAS,EAAG,QAAU,EAG5C,GAAM,GAAc,GAAc,EAAI,GAChC,EAAW,EAAI,EACf,EAAa,GACf,GAAM,GAAI,GAAc,EAAa,EAAG,EAAG,UAAW,GAAO,IAC7D,GAEJ,MAAO,GAAI,EAAI,MAGJ,IAAU,EAAG,CAAC,SAAQ,KC9EnC,YAoBoC,GAElC,MAAO,MAAK,MAAM,KAAK,IAAI,EAAG,KAAK,KAAK,KAAK,IAAI,GAAS,KAAK,IAAI,kBAIjE,EAAsB,EAAW,GAGnC,OAFM,GAAO,EAAI,EAAe,EAC1B,EAAY,GAAI,cAAa,GAC1B,EAAI,EAAG,EAAI,EAAc,EAAE,GAClC,GAAM,GAAU,EAAM,KAAK,GAAK,EAAM,GAAe,EAAO,GAC5D,EAAU,GAAK,EAAI,EAAI,KAAK,IAAI,GAElC,MAAO,IAAS,EAAW,WCjC7B,AAwCA,YACI,EAA2B,EAAuB,GAAA,MAAA,KAAA,QAAA,GAAA,wHAC9C,SAAe,EAAgB,EAAa,cAAe,UAC3D,EAAW,EAAgB,EAAS,UAAW,UAErD,EACI,EAAa,KAAO,EACpB,WAAM,MAAA,+DACF,YAAW,EAAa,QAChC,EACI,EAAa,KAAO,IAAM,EAAS,KACnC,WAAM,MAAA,mFAEC,GAAa,KAAI,qBAAqB,EAAS,QAC1D,GACI,EAAa,MAAM,MAAM,EAAG,EAAa,MAAM,OAAS,GACxD,EAAS,MACT,2FAEE,EAAU,EAAa,MAAM,EAAa,MAAM,OAAS,GAC/D,EACI,EAAI,GAAK,GAAK,EACd,WAAM,MAAA,iEACF,eAAc,EAAO,cAAc,KAEnB,CAAA,EAAM,EAAa,eAArC,SAAkB,EAAA,OACJ,CAAA,EAAM,EAAS,eAOnC,IAPM,EAAc,EAAA,OAId,EAAgB,CAAC,EAAgB,OAAS,EAAS,GAAlD,EAAK,EAAA,GAAE,EAAI,EAAA,GACZ,EAAY,GAAuB,OAAQ,GAExC,EAAI,EAAG,EAAI,EAAO,KAIzB,IAHM,EAAS,EAAI,EACb,EAAO,EAAgB,SAAS,EAAQ,EAAS,GACjD,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAKzC,IAHA,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAErC,EAAU,GAAK,EACN,EAAI,EAAG,EAAI,EAAG,IACrB,GAAI,EAAU,GAAG,QAAU,EAAY,IACrC,EAAU,GAAK,EACf,OAKN,MAAI,KAAgB,GAClB,EAAa,UAEf,AAAI,IAAY,GACd,EAAS,UAIX,CAAA,EAAO,GAAO,EAAW,EAAS,MAAO,iBAG9B,IAAc,GCtG3B,AAiDA,YACI,EAAM,EAAO,EACb,EACA,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE5D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GAC3D,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAY,GACxB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAY,GAAE,OACtD,EACI,IAAa,EAAY,GACzB,WAAM,MAAA,0CAA0C,EAAQ,UACpD,mCAAkC,EAAY,GAAE,QACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EACZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,GAAI,GAClD,EAC0B,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,GAE1E,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAsB,GAG5B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC/G7D,YAgCI,EAAY,EAAW,GACzB,GAAI,GAAc,MAAQ,IAAe,SACvC,MAAO,GAET,GAAI,IAAe,OACjB,MAAO,GAAI,EAAI,GAAK,IAEtB,KAAM,IAAI,OACN,gDAAgD,EAAU,iBAK5D,EAAc,GAChB,GAAI,GAAM,EACJ,EACF,GAAgC,EAAK,MAAO,EAAa,OAC7D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,mBAIvB,EAAW,EACX,GACF,GAAI,IAAe,SACjB,MAAO,GACF,GAAI,IAAe,OACxB,MAAO,IAAK,GACP,GAAI,IAAe,MACxB,MAAO,IAAI,GACN,GAAI,IAAe,QACxB,MAAO,IAAM,GACR,GAAI,IAAe,QACxB,MAAO,IAAM,EAAG,GAElB,KAAM,IAAI,OAAM,4BAA4B,EAAU,KAIjD,GAAM,IAAa,SAAC,EAAuB,GAChD,GAAM,GAAe,EAAgB,EACrC,MAAO,CAAC,GAAgB,IAAe,UC3EzC,AA8FA,YAAmD,MACjD,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAeA,GAFA,EAAa,GAAc,SAEvB,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,6DACC,GAAI,KAAI,OACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,8DACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,6DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,OACf,WAAM,MAAA,sCACF,EAAU,2CAElB,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,GAEnD,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACpB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAGlB,EAAe,GAAqB,EAAI,EAAG,GAEjD,EACI,GAA4B,GAC5B,WAAM,MAAA,oEAEF,uDAAsD,EAAS,OAEvE,GAAM,GACF,GAAoB,EAAI,MAAO,EAAc,EAAS,EAAS,GAC7D,EACF,GAAqB,EAAK,EAAc,EAAQ,MAAO,EAAS,GAC9D,GAAgB,CAAC,EAAM,GAE7B,GAAI,GAAS,MACX,GAAM,IAAU,GAAqB,EAAO,GAC5C,GAAI,KAAK,IAEX,MAAO,KAGH,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,YAAY,CAC9B,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA4B,CAChC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAGpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAS,EAAG,CAAC,aAAY,KC5QtC,AAyBA,YACI,EAAM,EAAO,EACb,GACF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,yBAAyB,EAAK,EAAM,IAE1C,EAAoD,CAAC,EAAG,EAAK,GAAI,GAEvE,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,IAGN,GAAM,IACT,EAAG,CAAC,qCAAoC,KChD5C,AAyBA,YACI,EAAmE,EACnE,EAAkB,GACpB,GAAI,GAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,wBAAwB,EAAM,EAAQ,IAE5C,EAAmD,CAAC,GAAI,GAExD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,IAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IACT,EAAG,CAAC,oCAAmC,KCnD3C,AAyFA,YAA4D,MAC1D,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAaA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,0DACF,iBAAgB,EAAQ,KAAI,OACpC,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,4DACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAClC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,EACI,GAAyC,EAAS,GAClD,WACI,MAAA,oEACA,sBAAqB,EAAO,mBAAmB,EAAS,OAEhE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,0BAAyB,EAAe,gBAAgB,EAAG,OAGrE,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IAEA,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,0BAG/C,GAAM,GAAO,SAAC,EAAc,GAC1B,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAS,OACd,GAAA,GAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAElB,EAAe,GAAqB,EAAI,EAAG,GAE3C,EAAO,GACR,EAAiB,MAAO,EAAc,EAAqB,GAC1D,EAAY,GACd,EAAiB,EAAe,EAAqB,MAAO,GAEhE,GAAI,GAAQ,MACV,GAAM,GAAU,GAAqB,EAAO,GAC5C,MAAO,CAAC,EAAM,EAAW,GAE3B,MAAO,CAAC,EAAM,IAGV,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,qBAAqB,CACvC,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAAqC,CACzC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAEpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAkB,EAAG,CAAC,sBAAqB,KClQxD,AAwDA,YAAwC,SACtC,EAAA,EAAA,EACA,EAAA,EAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAUA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GAAc,EAAG,EAAG,EAAY,GAC7C,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WACI,MAAA,qEACA,iBAAgB,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE9C,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAU,UACrD,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,wCAAwC,EAAW,UAClD,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACnC,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAErC,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAU,EAAM,QAG5D,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACnB,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAAK,GAAA,EAAA,GAAG,GAAA,EAAA,GAId,GACF,GAAqB,EAAQ,EAAI,GAAE,OAAQ,GAAG,GAC9C,GACA,GAgBJ,GAdA,AAAI,CAAC,GAAc,CAAC,EAClB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAM,KACzC,AAAI,CAAC,GAAc,EACxB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,GAAc,EAAK,GAAM,KACzC,AAAI,GAAc,CAAC,EACxB,IAAO,GAAc,EAAK,GAAc,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAO,KAE/C,IAAO,GAAc,EAAK,GAAc,GAAM,IAC9C,GAAO,GAAc,GAAc,EAAK,GAAM,KAG5C,GAAQ,MACV,GAAM,IAAU,GAAqB,GAAO,IAC5C,MAAO,CAAC,GAAM,GAAM,QAEpB,OAAO,CAAC,GAAM,KAIZ,EAA+B,SAAC,GACpC,GAAM,GAAI,EAAQ,iBAAiB,CACjC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA6B,CACjC,EAAG,EACH,EAAG,EACH,KAAM,EACN,uBAAwB,GAEpB,EAA2B,CAAC,WAAU,EAAE,WAAU,EAAE,WAAU,GAIpE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAe,GACxC,GAAM,GAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,IAET,CAAC,MAAO,EAAQ,EAAK,GAAW,SAAU,KAEvD,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAe,EAAe,GAC5C,GAAM,IAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,GAAK,IAEd,CAAC,MAAO,EAAQ,GAAK,GAAW,SAAU,KAGvD,MAAO,GAAiB,EAAK,EAAK,IAI/B,GAAM,IAAS,EAAG,CAAC,aAAY,KClOtC,+DCAA,AAiCA,YAAwB,GACtB,MAAO,IAAa,EAAc,IAAM,KAEnC,GAAM,IAAgB,EAAG,CAAC,eAAc,KCpC/C,AAiCA,YAAqB,GACnB,MAAO,IAAa,EAAc,GAAK,IAGlC,GAAM,IAAa,EAAG,CAAC,YAAW,KCrCzC,AAyCA,YACI,EAAkB,EAAqB,EAAmB,EAC1D,GAD0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,GAGF,OAFI,GAAQ,EACN,EAAmB,GAClB,EAAQ,GAAe,EAAO,MACnC,EAAO,KAAK,GAAM,EAAQ,EAAO,IACjC,GAAS,EAGX,GAAI,EACF,KAAO,EAAQ,EAAO,OACpB,GAAM,GAAU,EAAQ,EAAe,EAAO,KACxC,EAAM,GAAO,CACjB,GAAM,EAAQ,EAAO,EAAc,GAAS,GAAK,CAAC,GAAS,KAE7D,EAAO,KAAK,GACZ,GAAS,EAIb,MAAI,GAAO,SAAW,EACb,GAAS,GAAI,CAAC,EAAG,IAGnB,EAAQ,GAAO,GAAS,CAAC,EAAO,OAAQ,IAE1C,GAAM,IAAQ,EAAG,CAAC,OAAM,KCpE/B,AA4CA,YACI,EAAkB,EAAqB,EACvC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,AAAI,GAAa,MACf,GAAY,GAAoB,IAKlC,OAHM,GAAe,GAAM,EAAQ,EAAa,GAC1C,EAAiB,EAAI,EAAc,EAAS,IAC5C,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAa,MAAM,GAAI,IACzC,EAAO,KACH,GAAK,GAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,EAAG,IAAe,IAE5D,MAAO,IAAO,GAET,GAAM,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAmDA,YACI,EACA,EACA,EACA,EACA,EACA,GAEF,GAAM,GAAS,EAAgB,EAAO,QAAS,iBACzC,EAAS,EAAgB,EAAO,QAAS,gBAAiB,WAC1D,EAAU,EAAgB,EAAQ,SAAU,gBAAiB,SACnE,EAAS,GAAU,WACnB,EAAqB,GAAsB,EAE3C,GAAM,GAAW,EAAO,MAAM,GAE9B,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OACnC,EACI,EAAO,OAAS,GAAK,EAAO,MAAM,KAAO,EACzC,WAAM,MAAA,oDAAoD,EAAQ,OAC9D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAQ,OAAS,GAAK,EAAQ,MAAM,KAAO,EAC3C,WAAM,MAAA,qDAAqD,EAAQ,KAC/D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,iEACF,WAAU,EAAS,OAAM,OACjC,EACI,EAAS,IAAM,GAAK,EAAS,IAAM,EACnC,WAAM,MAAA,2CAA2C,IACrD,EACI,IAAW,YAAc,IAAW,UACpC,WAAM,MAAA,+CAA+C,IAEzD,GAAM,GAAiC,SAAC,GAAY,MAAA,GAAQ,cACxD,EAAQ,EAAQ,EAAS,EAAU,EAAQ,IAEzC,EACoB,CAAC,MAAO,EAAQ,MAAO,EAAQ,OAAQ,GAC3D,EAA4B,CAAC,OAAM,EAAE,mBAAkB,EAAE,SAAQ,GACjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GACJ,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KCtG/C,AAiCA,YAAwB,GACtB,GAAM,GAAS,EAAgB,EAAO,QAAS,gBAAiB,WAEhE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAA8B,CAAC,MAAO,GACtC,EACF,EAAO,UAAU,GAAe,EAAgC,IACpE,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KC/C/C,AA4CA,YACI,EAA4B,EAC5B,EACA,GADA,AAAA,IAAA,QAAA,GAAA,GACA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,mBAAoB,WAEnE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,mDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAAiC,CAAC,MAAO,GACzC,EAA+B,CAAC,QAAO,EAAE,UAAS,EAAE,OAAM,GAC1D,EAAM,EAAO,UACf,GAAkB,EAClB,GACJ,MAAO,GAGF,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC/DrD,AAoBA,YACI,EAAiB,EAAkB,EACnC,EAAsB,EAAwB,GAMhD,AAAI,GAAgB,MAClB,GAAe,IAEjB,AAAI,GAAkB,MACpB,GAAiB,OAAO,mBAE1B,AAAI,GAAgB,MAClB,GAAe,GAGjB,GAAM,GAAW,EAAM,MAAM,GAC7B,SAAgB,KAAK,IAAI,EAAe,GAExC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAClE,EACI,EAAM,OAAS,EACf,WAAM,MAAA,+CAA+C,EAAM,KAAI,MACnE,EACI,EAAM,MAAM,KAAO,EACnB,WACI,MAAA,oDAAoD,EAAM,MAAM,KACxE,EAAY,EAAO,OAAS,EAAG,WAAM,MAAA,+BACrC,EACI,EAAO,MAAM,KAAO,EACpB,WAAM,MAAA,sDAAsD,EAAQ,KAChE,YAAW,EAAO,MAAM,MAChC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAC3D,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GC3DnE,AA0BA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBAC1B,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAExB,GAAM,GAAQ,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,GAC1D,MAAO,GAAO,cACV,SAAA,GAAK,MAAA,GAAE,kBACH,EAAQ,EAAS,EAAe,EAAc,IAClD,CAAC,MAAO,EAAQ,OAAQ,GAAU,KAAiB,GACnD,GAGC,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC/CvD,YA+BI,EAAU,EAAY,GACxB,GAAM,GAAQ,GAAa,EAAK,EAAS,GACnC,EAAiB,EAAQ,EAAI,CAAE,GAAQ,GAAK,EAClD,EAAI,OAAO,EAAgB,EAAG,eAmB5B,EAAU,EAAW,GACvB,MAAO,IAAc,EAAK,EAAQ,GAAc,IAUlD,YAA8B,EAAM,GAClC,MAAO,GAAI,EAAI,EAAI,EAAI,EAAI,GAAK,EAGlC,YACI,EAAU,EAAW,GAKvB,OAJI,GAAO,EACP,EAAQ,EAAI,OACZ,EAAS,EACT,EAAQ,GACL,EAAO,IACZ,EAAS,EAAS,GAAQ,IAAU,GACpC,GAAM,GAAgB,EAAW,EAAQ,EAAI,IAC7C,AAAI,EAAgB,EAClB,EAAO,EAAS,EAEhB,GAAQ,EAGR,EAAQ,CAAC,GAIb,MAAO,GAAQ,EAAO,CAAC,EAAO,ECvFhC,YAoCI,EAAmB,EAAoB,EACvC,EAAsB,GACxB,MAAO,IACI,EAAO,EAAQ,EAAe,EAAc,EAC5C,GACC,4BAIV,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAC5C,EAAsB,GACtB,EAA6C,gBAK/C,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAAgB,EAC5D,IAGN,YACI,EAAmB,EAAoB,EACvC,EAAsB,EAAwB,EAC9C,EAA4B,EAC5B,GADA,AAAA,IAAA,QAAA,GAAA,IAA4B,AAAA,IAAA,QAAA,GAAA,IAC5B,AAAA,IAAA,QAAA,GAAA,IAKF,OAFM,GAAa,GAEV,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,EAAO,GAAK,GACd,EAAW,KAAK,CAAC,MAAO,EAAO,GAAI,SAAU,EAAG,mBAAoB,IAIxE,EAAW,KAAK,IAShB,OALM,GAAQ,EAAe,EAAK,IAAO,EAAgB,EAEnD,EAA4B,GAC5B,EAA2B,GAE1B,EAAgB,OAAS,GAAiB,EAAW,OAAS,IACnE,GAAM,GAAY,EAAW,MACtB,EAAA,EAAA,MAAsB,EAAA,EAAA,SAAU,EAAA,EAAA,mBAEvC,GAAI,EAAgB,EAClB,MAUF,OADI,GAAkB,GACb,EAAI,EAAgB,OAAS,EAAG,GAAK,EAAoB,EAAE,GAClE,GAAM,GAAM,GAAsB,EAAO,EAAU,EAAgB,IAEnE,GAAI,GAAO,GACT,EAAkB,GAClB,MAMF,GAHA,EAAU,MACN,EAAU,MAAQ,GAAe,EAAc,EAAO,GAEtD,EAAU,OAAS,EACrB,MAWJ,EAAU,mBAAqB,EAAgB,OAE/C,AAAK,GAGH,CAAI,EAAU,QAAU,EACtB,GAAgB,KAAK,GACrB,EAAe,KAAK,EAAU,QACzB,AAAI,EAAU,MAAQ,GAG3B,GAAa,EAAY,EAAW,KAM1C,GAAM,GAAe,EAAgB,OAC/B,EAAa,EAAgB,EAEnC,AAAI,GAAsB,EAAa,GACrC,GAAgB,KAAI,MAApB,EAAwB,GAAI,OAAM,GAAY,KAAK,IACnD,EAAe,KAAI,MAAnB,EAAuB,GAAI,OAAM,GAAY,KAAK,KAGpD,GAAM,GACe,CAAC,gBAAiB,GAAS,EAAiB,UAEjE,MAAI,IACF,GAAO,eAAoB,GAAS,EAAgB,YAGtD,AAAI,GACF,GAAO,aAAkB,GAAO,EAAc,UAGzC,EAGT,YAA+B,EAAmB,EAAW,GAC3D,GAAM,GAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAS,GAAQ,GAAU,GAAQ,GACnC,EAAS,GAAQ,GAAU,GAAQ,GACzC,GAAI,GAAS,GAAK,GAAS,EACzB,MAAO,GAET,GAAM,GAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAmB,EAAkB,GACnE,KAAK,IAAI,EAAmB,EAAkB,GAClD,MAAO,GAAoB,GAAQ,EAAQ,GAO7C,YAAwB,EAAsB,EAAe,GAC3D,GAAM,GAAS,KAAK,IAAI,EAAQ,EAAM,GACtC,MAAO,IAAO,EAAe,EAAS,EAGxC,YAA6B,EAAe,GAK1C,MAAQ,GAAG,MAAQ,EAAG,OAChB,EAAG,QAAU,EAAG,OAAW,EAAG,SAAW,EAAG,SC7MpD,AA2CA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,wHACpB,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAED,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,GACxD,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAyB,GC1EtC,AAyDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,GACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEtB,GAAM,GAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EACF,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GAExD,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,eAAgB,EAAO,IAGtD,GAAM,IAA6B,EAAG,CAAC,4BAA2B,KChCzE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,wGACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEC,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,EACpD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAkC,GCxF/C,AAmDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAEzB,EAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EAAkC,CACtC,cAAe,EACf,aAAc,EACd,eAAgB,EAChB,mBAAkB,GAGd,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,aAAc,EAAO,IAGpD,GAAM,IAA0B,EAAG,CAAC,yBAAwB,KCnCnE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,+GACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAG3B,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBADxC,SACF,EAAA,OADG,EAAS,EAAA,GAAE,EAAU,EAAA,GAMtB,EAAM,GACR,EAAW,EAAY,EAAgB,EAAe,EACtD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAA+B,GChF5C,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,2DACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,6DACC,GAAI,OAEf,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAG9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GACZ,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,eACX,EAAa,EAAW,EAAU,IAGlC,EAA+B,CAAC,OAAQ,GACxC,EAA6B,CAAC,aAAY,EAAE,KAAI,GAEhD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCpFjD,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,yBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,kEACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WACI,MAAA,oEACG,GAAI,OACf,EACI,EAAQ,QAAU,WAAa,EAAQ,QAAU,QACjD,WAAM,MAAA,qDAEV,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAE9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GAEZ,EAAsC,CAAC,OAAQ,GAC/C,EAAoC,CAAC,aAAY,EAAE,KAAI,GAEvD,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,sBACX,EAAa,EAAW,EAAU,IAGlC,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAwB,EAAG,CAAC,uBAAsB,KCxF/D,AAyEA,YACI,EAAiB,EAAkB,GACrC,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAClE,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAElE,GAAM,GAAK,EAAgB,EAAG,IAAK,YAEnC,EACI,EAAG,MAAQ,EACX,WAAM,MAAA,4CAA4C,EAAG,KAAI,MAE7D,GAAM,GAAQ,EAAG,MACX,EAAA,EAAA,MAAA,MAAA,IAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,kDAAiD,EAAC,OAExD,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,qDAAoD,EAAC,OAG3D,AAAI,EAAW,GACb,GAAW,GAEb,AAAI,EAAW,GACb,GAAW,GAGb,GAAM,GAAI,EAAQ,GAAM,EAAG,EAAG,EAAG,SAAU,CAAC,GAAI,IAC1C,EAAI,GAAM,EAAG,EAAG,EAAG,SACnB,EAAK,GAAI,EAAG,GAEZ,EAAS,GACX,GAAU,EAAI,GAAO,CAAC,EAAU,UAChC,GAAa,EAAI,GAAO,CAAC,EAAU,WAEjC,EAAO,GAAM,CAAC,EAAG,GAAI,EAAG,OAE9B,MAAO,GACI,GAAM,GAAQ,EAAQ,EAAI,CAAC,GAAI,EAAG,KACvB,IAAI,SAAA,GAAO,MAAA,IAAM,EAAQ,EAAK,MACzC,GAGN,GAAM,IAAW,EAAG,CAAC,UAAS,KC7HrC,AA2DA,YAAsB,GACpB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,EAAkB,GAClB,EACI,GAAM,MAAQ,EAAG,OAAS,EAC1B,WAAM,MAAA,sEAGV,OADM,GAAM,EAAG,GAAG,MAAM,cACf,GACP,EACI,EAAG,GAAG,MAAM,KAAO,EACnB,WACI,MAAA,gEACA,KAAK,EAAkB,GAAG,MAAM,GAAE,QAAQ,EAAG,QAL9C,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,OAQT,GAAkB,GAClB,EAAK,GAAM,EAAI,EAAG,MAAM,GAAI,GAAG,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,CAAC,MAGtD,EACI,EAAG,QAAU,EAAG,GAAG,MAAM,GACzB,WAAM,MAAA,oCACK,EAAkB,OAAM,aAC/B,0BAA0B,EAAkB,GAAG,MAAM,GAAE,QAI/D,OAFM,GAAiB,GACjB,EAAO,aACJ,GACP,EAAG,KAAK,EAAO,KAAK,WAClB,GAAI,GAAI,EAAK,GACb,GAAI,EAAI,EACN,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAO,EAAI,GAAI,EAAI,EAAG,GAAI,IAAK,EAAG,IACxC,EAAI,GAAI,EAAG,GAGf,MAAO,IAAI,EAAG,GAAK,EAAG,kBATjB,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,GAaT,MAAI,GACK,GAAM,EAAI,GAEV,EAIJ,GAAM,IAAc,EAAG,CAAC,aAAY,KC5G3C,AAmFA,YAAa,EAAW,GAMtB,GANsB,AAAA,IAAA,QAAA,GAAA,IACtB,EACI,EAAE,MAAQ,EACV,WAAM,MAAA,gEACF,EAAE,OAEN,EAAE,OAAS,EACb,MAAO,IAAK,EAAe,GAM3B,GAAM,GAAgB,EAAE,MAAM,MAAM,EAAG,EAAE,MAAM,OAAS,GAC7B,OAAO,SAAC,EAAO,GAAS,MAAA,GAAQ,IACrD,EAAO,GACT,EACI,EACA,CACE,EAAe,EAAE,MAAM,EAAE,MAAM,OAAS,GACxC,EAAE,MAAM,EAAE,MAAM,OAAS,KAE/B,GACE,EAAmB,GACnB,EAAmB,GACzB,EAAK,QAAQ,SAAA,GACL,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,KAAK,GACV,EAAK,KAAK,KAEZ,GAAM,GAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OAC9B,EAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OACpC,MAAO,CAAC,EAAG,GAIf,YAAc,EAAa,GAAA,MAAA,KAAA,QAAA,GAAA,IAClB,EAAO,KAAK,WACjB,EACI,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,0CACF,EAAE,MAAM,OAAM,cAYtB,OAVM,GAAI,EAAE,MAAM,GACZ,EAAI,EAAE,MAAM,GAEd,EAAI,GAAI,GACR,EAAI,GAAM,GAER,EAAQ,GAAS,CAAC,CAAC,IAAK,CAAC,EAAG,IAC9B,EAAc,GAAM,GAElB,EAAQ,GAAK,EAAI,EAAI,aAClB,SAGD,EAAQ,EACR,EAAQ,EACR,EAAQ,EACd,EAAA,EAAA,KAAA,wfAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GA8CP,GAAQ,CAAC,EAAO,EAAO,KApDhB,EAAI,EAAG,EAAI,EAAO,EAAE,IAApB,GAuDT,MAAI,CAAC,GAAgB,EAAI,GACvB,GAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,IACzB,EAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,KAGpB,CAAC,EAAG,KAIR,GAAM,IAAK,EAAG,CAAC,IAAG,KCxMzB,AAiBA,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,IAAA,GAAA,MACA,EAAA,EAAA,uBAAA,GAAA,2BAJU,EAAA,WAAA,GAAA,UAAS,KCyBrB,YACI,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GAAM,GAAgB,GAAY,KAAQ,EAAU,EAAI,EAAS,GAEjE,GAAI,IAAc,EAAA,UAAU,KAC1B,MAAO,GAET,GAAI,IAAc,EAAA,UAAU,IAC1B,MAAO,IAAI,GAEb,GAAI,IAAc,EAAA,UAAU,MAC1B,GAAI,GAAY,KACd,MAAO,IAAK,GAEZ,GAAM,GAAkB,EAAQ,KAAO,EAAS,KAC1C,EAAS,GAAI,GAAI,GAAe,GAAI,IAC1C,MAAO,GAAkB,EAAI,GAAI,EAAQ,GAAO,IACnB,EAGjC,GAAI,IAAc,EAAA,UAAU,wBAC1B,GAAI,GAAY,KACd,MAAO,IAAI,GAAI,GAAe,GAAO,EAAQ,OAE7C,GAAM,GAAqB,EAAI,EAAU,GAAK,EAAQ,QAEhD,EACF,GAAK,GAAI,GAAS,EAAoB,GAAO,KAAM,WACvD,MAAO,IAAI,GAAI,GAAe,GAIlC,KAAM,OAAM,sBAAsB,GAE7B,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCnF3D,AA2CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,sBAC5C,EACF,EAAgB,EAAa,cAAe,sBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,uBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,iCAEvC,GAAM,GAAS,GAAI,GAAI,EAAS,IAChC,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAqB,EAAG,CAAC,oBAAmB,KChBzD,YACI,EAAsB,EAA2B,EACjD,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAC5C,EACF,EAAgB,EAAa,cAAe,kBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,mBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,6BAEvC,GAAM,GAAM,GAAO,GACb,EAAS,GAAI,EAAK,GAAI,EAAI,EAAS,GAAe,EAAM,KAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCnBjD,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAU,EAAgB,EAAQ,SAAU,aAC1C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAM,GAAO,GAEnB,EAAU,GAAI,EAAI,GAAO,GAAI,GAAU,GACvC,GAAM,GAAS,GAAK,GAAI,EAAK,EAAI,EAAS,KAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KC9DvC,AAiDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,aAC5C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAc,GAAO,GACrB,EAAQ,GAAI,GAAI,EAAc,IAC9B,EAAY,GAAQ,EAAO,GAC3B,EAAS,GAAI,EAAO,GAEpB,EACF,GAAI,EAAI,GAAO,IAAM,GAAO,IAAa,EAAI,EAAa,IAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KCtEvC,AAgDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,MAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,WAC5C,EAAe,EAAgB,EAAa,cAAe,WAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,YAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,sBAErD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAO,GAEvB,EAAK,GAAI,EAAI,EAAS,GAAI,GAAI,EAAc,MAC5C,EACF,EAAI,GAAI,EAAK,GAAU,GAAI,GAAI,GAAI,EAAK,GAAe,KACrD,EAAS,GAAI,EAAI,GACvB,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAU,EAAG,CAAC,SAAQ,KCrEnC,AA0CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,oBAC5C,EACF,EAAgB,EAAa,cAAe,oBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,qBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,+BAEvC,GAAM,GAAS,GAAkB,EAAS,GAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC3DrD,AAmCA,YACI,EAAsB,GACxB,GAAM,GACF,EAAgB,EAAQ,SAAU,iCAChC,EACF,EAAgB,EAAQ,SAAU,iCACtC,GACI,EAAQ,MAAO,EAAQ,MAAO,4CAsBlC,GAAM,GAAY,GAAK,GACjB,EAAgB,EAAI,EAAS,GAC7B,EAAgB,GAAM,GAAI,GAAI,GAAI,MAExC,MAAO,IAAI,GAAI,EAAW,GAAgB,GAwB5C,YACI,EAAgC,EAChC,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAoB,EACpB,EAAkB,mBAAoB,uBACpC,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KAOvB,GANA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAEjD,GACI,EAAkB,MAAO,EAAQ,MAAO,kCAExC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAO,GAAO,IAEpB,EACI,GAAI,EAAI,EAAmB,GAAI,EAAK,IAChC,EAAI,EAAM,IAEpB,GAAM,GAAS,GAA+B,EAAmB,GAEjE,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxH3D,AA+DA,YACI,EAAW,EAAW,GAKxB,GALwB,AAAA,IAAA,QAAA,GAAA,IACxB,AAAI,IAAQ,IACV,GAAM,EAAO,KAAO,GAGlB,IAAQ,EAAO,KAAO,EACxB,KAAM,OACF,+DACA,wCAAuC,EAAO,KAAI,KAClD,gBAAe,IAGrB,GAAM,GACF,GAAW,SAAC,EAAgB,EAAgB,GAI1C,GAAM,GAAW,GACX,EAAM,GAAU,EAAQ,CAAC,GAAM,GAC/B,EAAY,GAAI,GAAK,EAAQ,WAAY,GAC/C,EAAK,CAAC,EAAQ,IAEd,GAAM,GAAa,GAAI,EAAI,EAAW,IAChC,EAAW,GAAI,EAAY,CAAC,IAE5B,EAAW,SAAC,EAAO,GAChB,GAAA,GAAA,EAAA,GAAQ,EAAA,EAAA,GACT,EAAU,GAAqB,EAAG,MAAO,CAAC,IAChD,MAAO,CACL,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAK,EAAQ,WAAY,GAAI,KACrC,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAI,GAAY,GAAK,EAAQ,eAGzC,MAAO,CAAC,MAAK,EAAE,SAAQ,KAG7B,MAAO,GAAS,EAAQ,GAsB1B,YACI,EAA4B,EAC5B,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GACA,EAAgB,EAAc,eAAgB,uBAC5C,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KASvB,GAPA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GACI,EAAc,MAAO,EAAQ,MAAO,kCAEpC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAa,GAAO,EAAc,MAAM,IAE9C,EACI,GAAI,EAAI,EAAe,GAAI,EAAK,IAC5B,GAAI,EAAsB,IAGpC,GAAM,GAAS,GAA+B,EAAe,GAE7D,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KC3J3D,GAuNM,IAAW,CACf,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,MAAK,IASD,GAAS,CACb,cAAa,GACb,WAAU,GACV,MAAK,GACL,KAAI,IAeA,GAAQ,CACZ,cAAa,GACb,sBAAqB,GACrB,eAAc,GACd,iBAAgB,GAChB,cAAa,GACb,kBAAiB,GACjB,uBAAsB,GACtB,2BAA0B,GAC1B,gCAA+B,GAC/B,wBAAuB,GACvB,6BAA4B,IAOxB,GAAS,CACb,SAAQ,GACR,YAAW,GACX,GAAE,IAaE,GAAS,CACb,mBAAkB,GAClB,oBAAmB,GACnB,eAAc,GACd,UAAS,GACT,UAAS,GACT,QAAO,GACP,iBAAgB,GAChB,oBAAmB,GACnB,oBAAmB,IClSrB,mBAyCwC,GAAA,EAAA,GAAxC,4DAiBE,SAAA,UAAA,SAAA,SAAS,EAAiB,EAAoB,GAApB,AAAA,IAAA,QAAA,GAAA,IAElB,GAAA,GAAA,KAAA,iBAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,GAAI,GAAW,MACb,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,KAAM,OAAQ,EAAM,EAAE,SACrD,KAAK,eAAe,OAEpB,MAAK,eAAe,GAMtB,MAFA,IAAQ,GAEJ,EACK,EAEP,GAAM,UACC,OAOX,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAEd,KAAK,6CAGJ,EAAA,UAAA,oBAAV,WACE,KAAK,YAAc,KAAK,WAAa,GAgBvC,EAAA,UAAA,iBAAA,SAAiB,EAAiB,GAEhC,MAAO,IAAc,EAAG,IAgB1B,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,aAAe,MACtB,GAAQ,KAAK,cAIX,EAAA,UAAA,eAAN,8EACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAErB,CAAA,EAAO,CACL,KAAM,OAEN,OAAQ,GAAO,KAAK,YAAa,gBAI/B,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,gEAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OACN,4DACA,IAAG,KAAK,sBAUE,EAAA,UAAA,kBAAhB,SAAkC,mGAEhC,SAAA,KAAoB,CAAA,EAAM,EAAa,GAAG,OAAO,eAAjD,SAAK,YAAe,EAAA,OAAqC,GACzD,CAAA,EAAO,EAAa,MAAM,UAE9B,GA5HwC,IA8HxC,OAAO,eAAe,GAAW,OAAO,YAAa,CACnD,MAAO,SAAC,GACN,MAAO,GAAS,UAAY,MAAQ,EAAS,kBAAoB,MAC7D,EAAS,gBAAkB,QC1KnC,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAAgC,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,MAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,IAAA,EAChC,EAAA,QAAA,EALN,EAAA,iBAAwC,GACxC,EAAA,mBAA0C,GAOhD,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,cACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,aACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAC3C,EAAoB,EAAK,mBAAmB,GAAG,SAErD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAiB,EAAK,KAC1B,EAAI,GAAO,GAAW,EAAI,EAAK,MAEjC,EACF,EAAI,GAAI,GAAK,GAAI,EAAmB,EAAK,UACjC,GAAK,GAAI,EAAiB,EAAK,WACnC,GAEF,EACF,GAAI,EAAI,EAAmB,EAAK,KAC5B,EAAI,GAAO,GAAU,EAAI,EAAK,MAEtC,EAAgB,OAAO,GACvB,EAAkB,OAAO,GAEzB,GAAM,GAAW,GAAI,EAAI,EAAS,CAAC,EAAK,cAAe,GACvD,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,oBAAsB,MAC7B,IAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,YACzC,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,cAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,iBAAgB,OAAK,KAAK,oBAC/B,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,iBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,IAAO,KAAK,IACZ,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,IAAQ,EAAO,UAjHxD,EAAA,UAAY,WAmHrB,GArHuC,IAsHvC,GAAc,ICrJd,mBA+BsC,GAAA,EAAA,GAMpC,WACc,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,IAD5C,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,aAAA,EAA8B,EAAA,wBAAA,EAHpC,EAAA,iBAAwC,KAOhD,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAM,GAAY,GAClB,EAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,eACrB,SAAU,GACN,WAAM,MAAA,IAAK,EAAM,MAAO,EAAK,yBAClB,SAAS,MAI5B,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAEjD,GAAK,WACH,GAAM,GAAqB,GAAI,EAAiB,GAAO,IACvD,EAAgB,OAAO,GAEvB,GAAM,GAAW,GACb,EAAI,GAAI,EACA,GAAK,GAAI,EAAoB,EAAO,QAAQ,aAChD,CAAC,EAAK,cACV,GACJ,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,kBAAoB,MAC3B,GAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIvC,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,iBAAiB,IAC9D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,iBAAmB,EAAa,IACjC,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,wBAA2B,KAAK,0BAK7B,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,0BA/EzC,EAAA,UAAY,UAiFrB,GAnFsC,IAoFtC,GAAc,ICnHd,mBAkCmC,GAAA,EAAA,GASjC,WACc,EAAgC,EAChC,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MAFvC,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EAL/B,EAAA,uBAA8C,GAC9C,EAAA,wBAA+C,GAMrD,GAAK,WAEH,EAAK,SAAW,GAAO,GAAO,WAC9B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EAAmB,GAAI,EAAG,EAAK,UAErC,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,wBAAwB,IAAM,MACrC,GAAK,wBAAwB,GAAK,CAChC,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAe,EAAK,wBAAwB,GAAG,SAE/C,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QACvD,EACF,GAAI,EAAI,EAAc,EAAK,OACvB,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EAA2B,GAAI,EAAgB,GAC/C,EACF,GAAI,EAAiB,GAEzB,EAAY,OAAO,GACnB,EAAa,OAAO,GAEpB,GAAM,GACF,GAAI,EAAI,GAAI,EACA,GAAI,GAAK,GAA4B,EAAK,UAC9C,CAAC,EAAK,cACV,GACR,EAAM,OAAO,KAGf,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,QAC7C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,SAAS,UAEd,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,yBAA2B,MAClC,GAAQ,KAAK,wBAAwB,IAAI,SAAA,GAAK,MAAA,GAAE,aAI9C,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,yBACrC,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,4GACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACf,GAAK,WACH,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,IACxD,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,MAGpD,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,wBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,UA5IN,EAAA,UAAY,OA8IrB,GAhJmC,IAiJnC,GAAc,ICnLb,mBAiCoC,GAAA,EAAA,GASnC,WACc,EAAgC,EAChC,EAAyB,EACzB,GADyB,AAAA,IAAA,QAAA,GAAA,MACzB,AAAA,IAAA,QAAA,GAAA,GAHd,GAAA,GAIE,EAAA,KAAA,OAAO,KAHK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EACzB,EAAA,MAAA,EANN,EAAA,uBAA8C,GAC9C,EAAA,2BAAkD,GAQxD,GAAK,WACH,EAAK,UAAY,GAAO,GAAG,WAC3B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EACF,GAAI,CAAC,EAAK,aAAc,GAAI,EAAI,EAAK,UAAW,EAAK,OAAQ,IAEjE,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAGxC,AAAI,EAAK,2BAA2B,IAAM,MACxC,GAAK,2BAA2B,GAAK,CACnC,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAIxC,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAkB,EAAK,2BAA2B,GAAG,SAErD,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QAEvD,EAAM,EAAI,EAAiB,EAAK,OAChC,EAAM,GAAI,GAEV,EAAqB,GAAQ,EAAK,GAExC,EAAY,OAAO,GACnB,EAAgB,OAAO,GAEvB,GAAM,GACF,GAAI,EAAI,GAAI,EAAI,GACR,GAAI,EAAgB,GAAI,EAAoB,EAAK,WACrD,GAER,EAAM,OAAO,KAGf,EAAK,UAAU,OAAO,GAAI,EAAK,UAAW,IAC1C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,UAAU,UAEf,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,4BAA8B,MACrC,GAAQ,KAAK,2BAA2B,IAAI,SAAA,GAAK,MAAA,GAAE,aAIjD,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,wDAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OAAM,wDAGlB,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,QAChB,MAAS,KAAK,QAKX,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,QAAY,EAAO,QAvHzB,EAAA,UAAY,SAyHrB,GA3HqC,IA4HrC,GAAc,IC7Jd,mBA6BkC,GAAA,EAAA,GAKhC,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,aAAA,EAEpB,EAAK,gBAAgB,KAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAEF,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAK,WACH,GAAM,GAAW,GAAI,EAAI,EAAK,EAAG,GAAW,GAC5C,EAAM,OAAO,OAGjB,KAAK,uBAMP,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,aAAe,EACpB,AAAI,KAAK,GAAK,MACZ,KAAK,EAAE,UAET,KAAK,EAAI,GAAK,GAAO,CAAC,KAGxB,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,WAGH,EAAA,UAAA,WAAN,qGACU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAA,CAAQ,EAAA,eAGJ,EAAA,UAAA,WAAN,SAAiB,6FACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAC5C,GADA,EAAe,EAAA,OACX,EAAa,SAAW,EAC1B,KAAM,IAAI,OAAM,iEAIpB,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,aAAgB,KAAK,eAIxB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,eA7DjB,EAAA,UAAY,MA+DrB,GAjEkC,IAkElC,GAAc,IC/Fd,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAA8B,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFZ,GAAA,GAGE,EAAA,KAAA,KAAM,IAAa,KAFP,SAAA,aAAA,EAA8B,EAAA,SAAA,EAChC,EAAA,YAAA,EAJJ,EAAA,cAAqC,GAM3C,EAAK,EAAI,GAAO,EAAK,YAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,cAAc,IAAM,MAC3B,GAAM,GAAY,GAClB,EAAK,cAAc,GAAK,CACtB,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,MAInD,GAAM,GAAe,EAAK,cAAc,GAAG,SACrC,EAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAK,WACH,GAAI,GACE,EAAkB,GAAI,EAAI,EAAK,EAAG,GAAe,GACvD,AAAI,EAAK,YACP,EAAW,GACP,EAAI,EAAK,EAAG,GAAI,EAAU,EAAI,EAAiB,EAAK,KAAM,GAE9D,EAAW,GAAI,EAAI,EAAK,EAAG,GAAkB,GAE/C,EAAa,OAAO,GACpB,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,UACP,AAAI,KAAK,eAAiB,MACxB,GAAQ,KAAK,cAAc,IAAI,SAAA,GAAK,MAAA,GAAE,aAS1C,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,SAAW,GAGZ,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,cAAc,IAC3D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,cAAgB,EAAa,IAC9B,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,SAAY,KAAK,SACjB,YAAe,KAAK,cAKjB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,SAAa,EAAO,cA3FlD,EAAA,UAAY,WA6FrB,GA/FuC,IAgGvC,GAAc,IC/Hd,mBAgCsC,GAAA,EAAA,GASpC,WACc,EAAgC,EAChC,EAA0B,EACpC,GAF0C,AAAA,IAAA,QAAA,GAAA,IAChC,AAAA,IAAA,QAAA,GAAA,GAA0B,AAAA,IAAA,QAAA,GAAA,MACpC,AAAA,IAAA,QAAA,GAAA,IAHJ,GAAA,GAIE,EAAA,KAAA,OAAO,KAOP,GAVY,EAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,SAAA,EAA0B,EAAA,QAAA,EANhC,EAAA,uBAA8C,GAC9C,EAAA,mBAA0C,GAC1C,EAAA,qBAA4C,GAQlD,EAAK,SAAW,EAEhB,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,WAE5B,GAAgB,KAClB,KAAM,IAAI,OAAM,+DAIpB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,OACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,qBAAqB,IAAM,MAAQ,EAAK,UAC/C,GAAK,qBAAqB,GAAK,CAC7B,aAAiB,EAAI,MACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAwB,EAAK,uBAAuB,GAAG,SACvD,EAAqB,EAAK,mBAAmB,GAAG,SACtD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEvC,GAAI,EAAK,UACP,GAAM,GAAsB,EAAK,qBAAqB,GAAG,SAEnD,EACF,GAAI,EAAI,EAAqB,EAAK,OAC9B,EAAI,EAAU,EAAI,EAAK,QAEzB,EACF,GAAI,EAAI,EAAU,EAAK,cACnB,GACI,GAAI,EACA,GAAI,GAAO,GAAyB,EAAK,YACnD,EACF,GAAI,EAAI,EAAoB,EAAK,UAAW,GAEhD,EAAsB,OAAO,GAC7B,EAAoB,OAAO,GAC3B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAGb,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EACF,GAAI,EAAI,EAAoB,EAAK,UAC7B,GAAI,EAAI,EAAU,EAAK,cACnB,GAAK,GAAI,EAA0B,EAAK,YAEpD,EAAsB,OAAO,GAC7B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAInB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,sBAAwB,MAAQ,KAAK,UAC5C,GAAQ,KAAK,qBAAqB,IAAI,SAAA,GAAK,MAAA,GAAE,YAE/C,AAAI,KAAK,oBAAsB,MAC7B,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,oBAC7C,AAAI,KAAK,UACP,EAAU,KAAI,MAAd,EAAkB,KAAK,sBAEjB,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EACF,KAAK,SAAW,EAAa,OAAS,EAAI,EAAa,OAAS,EAC9D,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,MAE3C,AAAI,KAAK,UACP,MAAK,qBACD,EAAa,MAAM,EAAgB,EAAG,EAAgB,GACjD,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,iBAK/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,SAAY,KAAK,SACjB,QAAW,KAAK,QAChB,SAAY,KAAK,WAKd,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,SAChD,EAAO,QAAY,EAAO,WA5KzB,EAAA,UAAY,UA8KrB,GAhLsC,IAiLtC,GAAc,ICjNd,AAyBA,GAAA,IAAA,WAAA,cAsCS,SAAA,IAAP,SAAW,GACT,MAAO,IAAI,IAAa,IAkBnB,EAAA,SAAP,SAAgB,EAAsB,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE/C,GAAI,IAAkB,EAAc,EAAU,IAuBhD,EAAA,QAAP,SACI,EAAsB,EAAY,EAAgB,EAClD,GADsB,MAAA,KAAA,QAAA,GAAA,IAAY,AAAA,IAAA,QAAA,GAAA,GAAgB,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,IACK,GAAI,IACP,EAAc,EAAO,EAAU,EAAS,IAevC,EAAA,KAAP,SACI,EAAsB,EAAa,EACnC,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MACnC,AAAA,IAAA,QAAA,GAAA,MACK,GAAI,IAAc,EAAc,EAAO,EAAO,IAehD,EAAA,SAAP,SAAgB,EAAqB,EAAW,GAAhC,MAAA,KAAA,QAAA,GAAA,MAAqB,AAAA,IAAA,QAAA,GAAA,KAAW,AAAA,IAAA,QAAA,GAAA,MAEvC,GAAI,IAAkB,EAAc,EAAK,IAgB3C,EAAA,OAAP,SACI,EAAsB,EAAa,EAAe,EAClD,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MAAe,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,GACK,GAAI,IAAgB,EAAc,EAAO,EAAO,EAAS,IAmB3D,EAAA,QAAP,SAAe,EAAsB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE5B,GAAI,IAAiB,EAAc,IAE9C,KC5LA,GA+Ba,IAAQ,CACnB,IAAK,GAAsB,IAC3B,SAAU,GAAsB,SAChC,SAAU,GAAsB,SAChC,QAAS,GAAsB,QAC/B,QAAS,GAAsB,QAC/B,OAAQ,GAAsB,OAC9B,KAAM,GAAsB,MCtC9B,AAiBA,GAAM,IAA2B,WAC/B,MAAI,OAAO,wBAA0B,YAC5B,sBACE,MAAO,eAAiB,YAC1B,aAEF,SAAC,GAAgB,MAAA,SAa1B,cACE,MAAO,IAAI,SAAc,SAAA,GAAW,MAAA,IAAc,WAAM,MAAA,SCrC1D,YAmBI,EAAiC,EACjC,GACF,GAAM,GACF,EAAc,OAAO,IAAW,SAAW,EAAS,EAAO,IACzD,EACF,EAAe,OAAO,IAAW,SAAW,EAAS,EAAO,IAChE,MAAO,CAAC,EAAS,GCzBnB,YA0BI,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAqB,GACzB,GAAI,EACF,EAAW,EAAS,OAAO,EAAW,MAAM,IAC5C,EAAS,KAAK,EAAW,GAAK,GAC9B,EAAW,EAAS,OAAO,EAAW,MAAM,SAE5C,EAAW,EAAS,OAAO,EAAW,IAEtC,OADM,GAAgB,EAAW,OACxB,EAAI,EAAG,EAAI,EAAe,EAAE,EACnC,EACI,EAAS,OAAO,CAAC,EAAW,EAAI,GAAK,EAAW,GAAI,EAAW,KAErE,EAAW,EAAS,OAAO,EAAW,MAAM,EAAgB,IAE9D,MAAO,eAaL,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAW,GACjB,GAAI,GACF,EAAS,KAAK,GACd,OAAS,GAAI,EAAiB,EAAG,EAAI,EAAc,EAAE,EACnD,AAAI,GAAK,EAAI,EACX,GAAS,KAAK,GACd,EAAS,KAAK,EAAK,GAAiB,KAEpC,EAAS,KAAK,QAMlB,OAFM,GAAsB,GACtB,EAAqB,GAClB,EAAI,EAAG,EAAI,EAAc,EAAE,EAClC,AAAI,GAAK,EAAiB,EAAI,GAAK,EAAI,IAAM,EAC3C,EAAmB,KAAK,GAExB,EAAoB,KAAK,GAG7B,EAAS,KAAI,MAAb,EAAiB,GACjB,EAAS,KAAK,GACd,EAAS,KAAI,MAAb,EAAiB,GAEnB,MAAO,eAaL,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAmB,GAEzB,AAAI,EACF,EAAiB,KAAK,EAAW,GAAK,GAEtC,EAAiB,KAAK,EAAW,GAAK,GAGxC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,GAAK,EAAW,OAClB,AAAI,EACF,EAAiB,KAAK,EAAW,EAAI,GAAK,EAAW,IAErD,EAAiB,KAAK,EAAW,GAAK,EAAW,EAAI,IAGvD,EAAiB,KAAK,EAAW,IAIrC,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,CAAC,GACjB,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAiB,KAAK,EAAM,GAAG,IAEjC,MAAO,eAeL,EAA0B,EAAmB,GAE/C,OADM,GAAY,EAAe,MAAM,EAAG,GACjC,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAU,KAAK,EAAe,EAAI,GAAK,EAAM,GAAG,GAAK,EAAM,GAAG,IAGhE,MAAO,GCvJT,AAiBO,GAAM,IAAkB,mBAClB,GAAa,mBClB1B,AAiBO,GAAM,IAAQ,SACR,GAAS,WACT,GAAS,YACT,GAAS,YACT,GAAS,aACT,GAAS,YCtBtB,qBAmBqB,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACnB,AAAK,KAAM,QAAQ,YACjB,QAAQ,KAAI,MAAZ,QAAgB,wBAIA,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GAClB,AAAK,KAAM,QAAQ,YACjB,QAAQ,IAAG,MAAX,QAAe,GC3BnB,YAiCI,EAAoB,GACtB,GAAI,EAAK,SAAW,EAAK,OACvB,KAAM,IAAI,OACN,gEACG,GAAK,OAAM,WAAW,EAAK,OAAM,MAG1C,OADM,GAAS,GAAI,cAAa,EAAK,OAAS,GACrC,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EACtC,EAAO,GAAK,EAAK,EAAI,GACrB,EAAO,EAAI,GAAK,EAAK,EAAI,GAE3B,MAAO,eAiB8B,GAIrC,OAFM,GAAO,GAAI,cAAa,EAAQ,OAAS,GACzC,EAAO,GAAI,cAAa,EAAQ,OAAS,GACtC,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,EAAI,GAAK,EAAQ,GACtB,EAAK,EAAI,GAAK,EAAQ,EAAI,GAE5B,MAAO,CAAC,KAAI,EAAE,KAAI,eAOiB,GAKnC,OAHM,GAAM,KAAK,KAAK,EAAQ,OAAS,GACjC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAOgB,GAKlC,OAHM,GAAM,KAAK,MAAM,EAAQ,OAAS,GAClC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAShB,EAAuB,GACzB,GAAM,GAAO,EAAQ,EAAQ,GACvB,EAAO,EAAQ,EAAQ,EAAI,GACjC,MAAO,CAAC,KAAI,EAAE,KAAI,eAUhB,EAAkB,EAAc,EAAc,GAChD,EAAK,EAAQ,GAAK,EAClB,EAAK,EAAQ,EAAI,GAAK,cAOpB,EAAW,GAGb,OAFM,GAAO,GAAI,cAAa,EAAI,GAC5B,EAAO,GAAI,cAAa,EAAI,GACzB,EAAI,EAAG,EAAI,KAAK,KAAK,EAAI,GAAI,KACpC,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GAC9C,EAAK,GAAK,KAAK,IAAI,GACnB,EAAK,GAAK,KAAK,IAAI,GAErB,MAAO,CAAC,KAAI,EAAE,KAAI,eAOhB,EAAW,EAAW,GACxB,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GACxC,EAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GACtB,MAAO,CAAC,KAAI,EAAE,KAAI,GCvJpB,YAyDI,EAAM,EAAiB,GACzB,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,GAAE,QAEX,GAAM,GAAc,GAAM,EAAE,OACtB,EAAS,GAAK,EAAG,WACjB,EAAS,EAAQ,QAAQ,EAAQ,GACvC,SAAY,UACZ,EAAO,UACA,EAGT,GAAI,CAAC,GAAgB,EAAE,MAAO,GAG5B,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAE,MAAO,GAExD,GAAI,EAAE,QAAU,aACd,GAAM,GAAO,EAAQ,KAAK,GACpB,EAAS,GAAK,EAAM,GAC1B,SAAK,UACE,EAET,GAAI,IAAU,QACZ,MAAO,GAAQ,IAAI,GACd,GAAI,IAAU,QACnB,GAAM,GAAO,GAAO,EAAG,EAAE,OACnB,EAAS,EAAQ,SAAS,EAAG,GACnC,SAAK,UACE,MAEP,MAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,eAK/D,EAAM,GACR,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAO,EAAE,mBAG3B,EAAe,EAAc,GACxD,GAAM,GAAQ,GAAO,GAAU,GAAM,GAE/B,EAAS,GAAoB,EAAK,WACxC,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,qvCC3G1B,YAyBI,EAAM,EAAsB,GAC9B,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,MAAO,GAAW,IAAI,SAAA,GACpB,GAAM,GAAgB,EAAI,QAC1B,EAAU,GAAQ,EAClB,GAAM,GAAS,GAAM,EAAG,EAAO,GAC/B,SAAM,IAAS,EACR,ICjCX,YA2BI,EAAiC,GAEnC,OADM,GAAqB,GAAI,OAAM,EAAK,MACjC,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAK,MAAM,GAAK,EAAK,GAGrC,OADM,GAAS,GAAO,EAAU,EAAK,OAC5B,EAAI,EAAG,EAAI,EAAO,OAAO,OAAQ,EAAE,GAI1C,OAHM,GAAS,EAAO,WAAW,GAE3B,EAAwB,GAAI,OAAM,EAAK,MACpC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,MAAM,GAG1C,GAAM,GAAgB,EAAK,WAAW,GAEtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,WC7ChB,YAyBI,EAAe,EAAkB,EAAyB,EAC1D,GAOF,OALM,GAAU,EAAO,EAAO,OAAS,GACjC,EAAA,CAAA,EAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAc,GAAuB,EAAQ,EAAQ,GACrD,EAAiB,GAAuB,QAAS,EAAQ,GAEtD,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHM,GAAS,EAAI,EACb,EAAO,EAAE,SAAS,EAAQ,EAAS,GACnC,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAEzC,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAKrC,OAHM,GAAY,EAAI,EAChB,EAAW,EAAY,SAAS,EAAW,EAAY,GACvD,EAAc,EAAe,SAAS,EAAW,EAAY,GAC1D,EAAI,EAAG,EAAI,EAAG,IACrB,EAAS,GAAK,EAAU,GAAG,MAC3B,EAAY,GAAK,EAAU,GAAG,MAKlC,GAAM,GAAc,EAAO,QAC3B,SAAY,EAAY,OAAS,GAAK,EAC/B,CACL,GAAO,EAAa,EAAa,GACjC,GAAO,EAAgB,EAAa,UCxDxC,mJCAA,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,WAAY,SC7BtD,AA4BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAO,GAAK,EAAG,YACnB,EAAI,GAAK,GAAI,GAAO,GAAI,IAC9B,MAAO,IAAI,GAAI,EAAI,QCtC3B,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAK,EAAG,YAAa,IAC/C,MAAO,IAAI,EAAI,OCnCvB,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAGxB,MAAO,CAAC,EAAS,EAAG,KChDxB,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAsC,GAC5C,SAAM,QAAQ,SAAC,EAAG,GAChB,EAAK,GAAK,WAAM,MAAA,GAAG,WAEd,IC7BX,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,mBChChE,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,cAC7C,MAAO,IAAI,EAAI,OCpCvB,AA6BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,EAAI,EAAI,GAAI,EAAG,IACnB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,GAAI,EAAI,EAAI,GAAI,EAAG,KACvB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAExB,MAAO,CAAC,EAAS,EAAG,KCtDxB,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAK,EAAG,YAAa,QC/B7D,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,kBC/B3D,AA2DA,YACI,EAAkB,EAClB,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBAE3C,EAAO,EACP,EAAU,EACV,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,MAIvE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OAEvB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,IAG5C,EAAkC,CAAC,GAAI,EAAM,MAAO,GAEpD,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KChIvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAY,EAAS,EAAY,EACpD,OCpCV,AA8CA,YACI,EAAkB,EAAqB,EACvC,EAAkC,GACpC,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBAE/C,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,gCAC/B,EAAI,KAAI,MAEhB,GAAI,GAAU,EACV,EAAO,EACP,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EACI,EAAQ,EAAQ,CAAC,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KACvE,EAAO,EAAQ,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGhE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,4DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,+DACC,GAAQ,KAAI,OAEvB,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAmB,GAE3D,MAAO,GAAQ,gBAAgB,EAAM,EAAS,IAG1C,EAAgC,CAAC,GAAI,EAAM,MAAO,GAElD,EAA8B,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,GAEvD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAC5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAY,EAAS,OC7BvE,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,WAEnB,MAAI,CAAC,GAAc,CAAC,EACX,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,MAEtB,CAAC,GAAc,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,MAEtB,GAAc,CAAC,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,MAGzB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,KAC7B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,QChDrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,MACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAQtC,OAPM,GACF,EAEE,EAAa,EAAiB,WAC9B,EAAc,EAAiB,MAE/B,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAW,OAAS,EAAG,GAAK,EAAG,IAC1C,GAAI,EAAW,KAAO,EAAY,GAChC,EAAK,GAAK,UACD,EAAW,KAAO,EAC3B,KAAM,IAAI,OAAM,mBACZ,EAAU,6BAA6B,EAAW,MAI1D,OADM,GAAiB,GACd,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,EAAK,GAAK,GACZ,EAAK,KAAK,GAId,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAM,QC/CnC,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,GAAG,YCxBxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAET,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AA0BO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,aAAc,EAAA,EAAA,aACrB,MAAO,CACL,EAAG,WAAM,MAAA,IACL,GAAW,GAAa,EAAG,GAAe,GAAU,EAAG,IACvD,EAAI,GAAU,QCnCxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAS,EAAM,IAAI,SAAA,GAAK,MAAA,GAAE,QACzB,EAAA,EAAA,KACD,EAAQ,GAAe,EAAM,EAAM,GAAG,OAAO,GAC7C,EAAa,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAa,GAAM,EAAI,EAAY,GACzC,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,YAAM,MAAA,QC/BrC,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACN,EAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAEhC,SACI,GAA4B,GAC5B,WAAM,MAAA,8DACF,uDAAsD,EAAS,OAEhE,CACL,EAAG,WACC,MAAA,IAAoB,EAAI,MAAO,EAAI,EAAS,EAAS,EAAK,IAC9D,OAAQ,WACJ,MAAA,IAAqB,EAAK,EAAI,EAAQ,MAAO,EAAS,EAAK,OCxCrE,AAsBO,GAAM,IAA4C,CACvD,WAAY,GACZ,aAAc,CAAC,KAAM,UACrB,SAAU,SAAC,EAAe,EAAiB,GACnC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAAY,EAAA,EAAA,gBAGjC,MAAO,CACL,GAAI,WAAM,MAAA,IACN,EAAK,EAAQ,EAAS,EAAK,EAAY,EACvC,IACJ,OAAQ,WAAM,MAAA,IACV,EAAK,EAAI,EAAO,MAAO,EAAS,EAAK,EAAY,OCpC3D,AA2CA,YACI,EAAM,EAAO,EACb,EAA0C,GAC5C,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEnE,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAEzE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,EACI,EAAI,MAAM,KAAO,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAI,MAAM,GAAE,UAC1D,iCAAgC,EAAY,GAAE,OACtD,EACI,EAAK,MAAM,KAAO,EAAY,GAC9B,WAAM,MAAA,0CAA0C,EAAK,MAAM,GAAE,UACzD,mCAAkC,EAAY,GAAE,QAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAEhD,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,EAAG,GAEjD,EAAmC,CAAC,QAAO,EAAE,IAAG,GAEtD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAwB,GAG9B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC7F7D,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAC3B,EACI,GAAkB,GAClB,WACI,MAAA,kEACA,mDAAkD,EAAS,OAE5D,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAEZ,MAAO,CACL,EAAG,WAAM,MAAA,IACJ,EAAiB,MAAO,EAAI,EAAqB,EAAS,IAC/D,OAAQ,WAAM,MAAA,IACV,EAAiB,EAAK,EAAqB,MAAO,EAAS,OCzCrE,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAI,GAAK,EAAG,aAAc,OC/BvD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UAAW,EAAA,EAAA,QAExB,MAAO,CACL,EAAG,WACD,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAE,MAE7C,EAAM,GAAO,EAAI,EAAM,EAAW,CAAC,GAEvC,MAAI,IAAe,MACjB,GAAM,GAAU,EAAK,IAGhB,MCzCf,AAwBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAG1B,EAAa,GAAoB,CAAC,EAAG,GAE3C,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAU,OAEhB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,EACI,EAAE,OAAS,EACX,WAAM,MAAA,6DACF,yBAAwB,EAAE,KAAI,OACtC,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,8DACF,yBAAwB,EAAO,KAAI,OAC3C,EACI,EAAE,MAAM,KAAO,EAAO,MAAM,GAC5B,WAAM,MAAA,yDACF,cAAa,EAAE,MAAM,GAAE,0CACvB,cAAa,EAAO,MAAM,GAAE,OAEpC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,sCAAqC,EAAO,mBAC5C,KAAI,EAAU,QAEtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WACI,MAAA,gEACA,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAW,GACb,EAAE,MAAO,EAAO,MAAO,EAAS,EAChC,EAAK,EAAiB,IAE1B,MAAO,CACL,EAAG,WACC,MAAA,IAAmC,EAAE,MAAO,EAAI,EAAQ,IAC5D,OAAQ,WACJ,MAAA,IAAoC,EAAG,EAAI,EAAO,MAAO,OC7EnE,AAuBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAA6C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAC3D,EAA+C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAEnE,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,UACH,GAAyB,EACzB,IACb,OAAQ,WAAM,MAAA,GAAO,UACH,GACA,EAAsC,OCtC9D,AA6BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OACnD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAgC,EAAE,MAAO,GAC5D,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GAC/C,MAAO,GAAQ,OAAO,EAAI,IAGtB,EAAwB,CAAC,GAAE,EAAE,EAAC,GAEpC,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,cACZ,EAAoB,EAAgC,KACpD,QCrCV,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAI,EAAI,GAAI,GAAI,GAAO,KAAM,EAAI,KAAK,KAAK,KAAK,KACtD,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC/B7B,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC3B7B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAI,QC5BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AA6BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAiB,EAAE,MAAO,GAC7C,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AA8BO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,IAAK,OAAQ,WAAY,SACxC,SAAU,SACN,EAAY,EAAiB,GACxB,GAAA,GAAA,EAAA,gBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GAAM,EAAA,EAAA,GAAU,EAAA,EAAA,GAEpB,EAAa,GAAgB,GAAO,GACpC,EAAgB,GAAiB,EAAK,MAAO,EAAE,OAC/C,EAAsB,GAC5B,GAAI,EAAK,OAAS,GAChB,OAAS,GAAI,EAAG,EAAI,EAAE,MAAM,OAAS,EAAG,EAAE,EACxC,EAAU,KAAK,EAAE,MAAM,IAEzB,EAAU,KAAK,GAGjB,GAAM,GAAa,GAAI,EAAG,GACpB,EAAoB,EAAI,EAAI,GAC5B,EAAsB,GAAM,GAAI,EAAU,GAAO,KACjD,EAAiB,EACnB,EAAI,EAAI,EAAqB,GAAsB,GACnD,GAAO,MAEL,EAAO,WACX,MAAI,GAAK,OAAS,EACT,EACH,EAAI,EAAI,EACA,GACI,EAAQ,EAAqB,CAAC,EAAG,EAAG,EAAG,EAAK,MAAM,KAClD,IACR,GACJ,EAAE,OAEC,EAAQ,EAAI,EAAI,EAAI,GAAsB,GAAa,EAAE,QAG9D,EAAU,WACd,GAAI,GACA,EAAI,EAAI,EAAqB,GAAO,KAAM,GAC9C,MAAI,GAAK,OAAS,GAChB,GAAU,GAAI,EAAS,IAElB,EAAQ,EAAS,EAAK,QAEzB,EAAc,WAClB,GAAI,GAAc,EAAI,EAAI,EAAgB,GAAa,GAEvD,MAAI,GAAK,OAAS,GAChB,GAAc,GAAI,EAAa,IAE1B,EAAQ,EAAa,EAAK,QAE7B,EAAW,WACf,GAAM,GAAwB,EAAI,EAAY,GAE1C,EAAW,EAAI,EAAI,GACvB,MAAI,GAAK,OAAS,GAChB,GAAW,GAAI,EAAU,IAEpB,EAAQ,EAAU,EAAK,QAE1B,EAAY,WAChB,GAAI,GAAY,EAChB,MAAI,GAAK,OAAS,GAChB,GAAY,GAAI,EAAW,IAEtB,EAAQ,EAAW,EAAK,QAGjC,MAAO,CACL,EAAG,EACH,KAAM,EACN,SAAU,EACV,MAAO,EACP,OAAQ,KC1Gd,AA0BO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,WACpB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACH,EAAA,EAAA,KAED,EAAa,GAAe,EAAM,EAAE,OAAO,GAE3C,EAAO,WACX,GAAM,GAAc,EAAE,MAChB,EAAc,EAAQ,KAEtB,EAAa,EAAY,MAAM,EAAG,GAClC,EAAY,EAAW,OACvB,EAAa,EAAY,MAAM,EAAM,EAAY,QAAQ,MAAM,GAC/D,EAAY,EAAW,OAEvB,EAAmB,GAAW,EAAG,GACjC,EACF,GAAW,EAAY,EAAG,EAAY,EAAI,GAExC,EAAc,GAAY,CAAC,EAAY,CAAC,GAAc,IAEtD,EAAS,EAAQ,EAAI,GACrB,EAAkB,EAAQ,EAAS,CAAC,IAEpC,EACF,GAAY,CAAC,CAAC,GAAY,EAAkB,IAC1C,EAAkB,GAAU,EAAQ,GACtC,EAAa,GACb,EAAiB,EAA6B,EAAE,MAAM,IAEpD,EAAsB,GAAuB,GACnD,SAAa,GAAU,EAAY,GAE5B,GAET,MAAO,CAAC,EAAG,EAAM,QAAS,WAAM,MAAA,OAIpC,YAAoB,EAAe,GAEjC,OADM,GAAS,GACN,EAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,EAAO,KAAK,GAEd,MAAO,GAGT,YAAqB,GAEnB,OADM,GAAS,GACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,EACtC,EAAO,KAAK,EAAO,GAAG,IAG1B,MAAO,GCnFT,AAqBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,IAAI,EAAG,WAAM,MAAA,IAAU,OC1BtD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAK,EAAI,eCzB9B,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC5B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,EAAG,QC5BpC,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,EAAG,gBC5BrC,AAyBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,GACd,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KACP,MAAO,CACL,OAAQ,WACN,GAAM,GAAW,GACX,EAAU,GAAI,GACpB,MAAO,IAAI,EAAI,EAAI,GAAI,EAAI,EAAM,GAAW,QCpCpD,AAyBA,YACI,EAAM,EAAM,EAAO,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,QAAQ,EAAI,EAAG,EAAG,EAAa,EAAM,EAAO,IAElD,EAA4B,CAAC,EAAC,EAAE,EAAC,EAAE,GAAE,GAErC,EAA0B,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAE/D,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAa,GAGnB,GAAM,IACT,EAAG,CAAC,oCAAmC,KCxC3C,AAqBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,YAAa,EAAA,EAAA,KAAM,EAAA,EAAA,MAAO,EAAA,EAAA,KAEjC,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAG,EAAG,EAAI,EAAa,EAAM,EAAO,OC/B9C,YA6BI,EAAO,EAAM,EAAe,EAAoB,GAClD,MAAI,GAAE,KAAO,EAAM,MACjB,GAAI,EAAQ,EAAG,GAA+B,EAAE,MAAO,KAEzD,AAAI,EAAG,KAAO,EAAM,MAClB,GAAK,EAAQ,EAAI,GAA+B,EAAG,MAAO,KAErD,CACL,EAAG,WACD,GAAM,GAAK,EAAI,EAAI,GAAK,GAAM,EAAO,GAAI,EAAG,QAC5C,MAAO,IAAgB,KAAO,EAAK,GAAU,EAAI,KCvCvD,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,iBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAkB,EAAE,OACnD,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAa,EAAG,GAAI,aAC9C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,GAAI,aAC5C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA6DA,YACI,EAAkB,EAAqB,EACvC,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE9C,EAAO,EACP,EAAU,EACV,EAAW,EACX,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KAErE,EAAW,EAAQ,EAAS,CAC1B,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAI3E,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAS,OAAS,EAClB,WAAM,MAAA,kEACC,GAAS,KAAI,OAExB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OACrE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,EAAU,IAGtD,EACwB,CAAC,GAAI,EAAM,MAAO,EAAS,OAAQ,GAE3D,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC1IvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAG,EAAY,EAAS,EAAY,EACvD,OCrCV,AAkDA,YACI,EAAyB,EACzB,EAA6B,EAC7B,EAAkC,EAClC,GACF,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBACzC,EAAU,EAAgB,EAAQ,SAAU,mBAElD,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,+BAC/B,KAAI,EAAI,KAAI,OAEpB,EACI,EAAI,OAAS,EACb,WAAM,MAAA,4DACC,GAAI,KAAI,OACnB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,+DACC,GAAO,KAAI,OACtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAO,MAAO,EAAY,EAAS,EAAmB,EACtD,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAQ,EAAS,IAGjD,EACsB,CAAC,GAAI,EAAK,MAAO,EAAQ,OAAQ,GAEvD,EACqB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAErE,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,GAC/C,GAGN,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAE5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAG,EAAY,EAAS,OC/B1E,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,KACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAM,EAAE,OACvC,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAU,EAAG,GAAI,aAC3C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAQ,EAAG,GAAI,aAC/C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA4BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAI,GAAa,EAAE,OAEjC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAI,GAAM,GAAI,EAAG,MAC/B,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KClDxB,AA0BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,OCzBzB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,WACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAU,EAAM,GACtB,MAAO,CAAC,QAAS,WAAM,MAAA,IAAM,EAAQ,MAAO,eC3BhD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAGtC,GAAM,GAAI,EAAM,GACT,EAAA,EAAA,SACD,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAM,EAAI,EAAO,EAAE,WC/BxC,AAgCO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GACP,EAAO,EACP,EAAM,EACN,EACF,GAA0C,EAAK,MAAO,EAAI,OAExD,EAAU,WACd,GAAM,GAAW,GAAK,EAAK,WACvB,EAAM,EAAI,EAAI,EAAI,EAAU,GAAI,EAAM,GAAI,EAAU,GAAO,OACzD,EAAa,GAAgC,EAAK,MAAO,GAC/D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,QAErB,EAAS,WACb,GAAM,GAAY,GAAQ,EAAM,GAC1B,EAAU,GAAM,EAAW,GAAI,GAAO,GAAU,IAClD,EAAM,EAAI,EAAI,EAAI,EAAG,IACnB,EAAa,GAAgC,EAAI,MAAO,GAC9D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAI,QAE1B,MAAO,CAAC,EAAG,EAAS,EAAG,KC9D3B,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,SACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,GAAQ,EAAG,GAExB,MAAO,CACL,EAAG,WAAM,MAAA,IAAM,EAAM,EAAI,EAAI,EAAI,KACjC,MAAO,WACL,GAAI,GAAM,GAAM,EAAM,GAAU,GAAK,EAAI,EAAI,IACvC,EAAa,GAAiB,EAAM,MAAO,EAAG,OACpD,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAM,WC1ClC,AAwBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,SC7BxC,AAwBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAO,EAAI,GAAU,EAAG,GAAI,GAAK,IAEvC,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAM,gBC/BxC,AAuBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,GAAI,gBC5B3C,AAqBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAQ,EAAI,EAAE,WC1BnC,AAuBO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,uBACX,EAAI,EAAoB,IAGxB,EAAmC,CAAC,OAAM,GAC1C,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAAoB,IAExB,MAAO,CAAC,OAAQ,KCxCpB,AAuBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,8BACX,EAAI,EAAoB,IAGxB,EAA0C,CAAC,OAAM,GACjD,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAA2B,IAE/B,MAAO,CAAC,OAAQ,KCxCpB,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,KACD,EAAO,GAAe,EAAM,EAAG,OACrC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAQ,EAAI,OC3BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAyBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,GAAI,EAAI,EAAI,GAAI,EAAG,KAAM,SC9BlD,AAyBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,aACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAGL,UAAW,WAAM,MAAA,IAAK,GAAU,GAAY,YAC5C,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAW,EAAG,SACpC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAW,GAAY,EAAG,YCnCtD,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CACL,EAAG,WACD,GAAM,GAAO,GAAQ,EAAG,GAAO,IAEzB,EAAa,GAAO,IACpB,EAAQ,GAAO,IAEf,EAAqB,EAAI,EAAI,GAC7B,EACF,EAAI,EAAI,EAAI,GAAa,GAAI,GAAK,EAAG,aAEzC,MAAO,IAAM,EAAM,EAAoB,OC3C/C,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,EAAG,GAAI,GAAO,GAAI,SC9BnD,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAK,EAAG,YAAa,OC7BlD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GActC,OAbO,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,KAER,EAAa,EAAE,MACf,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAQT,EAAoC,GACjC,EAAI,EAAG,EAAI,EAAG,KAAM,IAC3B,EAAS,KAAK,CAAC,EAAO,GAAI,EAAW,GAAK,EAAO,GAAK,EAAM,KAE9D,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,OC3C7B,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,IACD,EAAW,GAEX,EAAW,EAAI,EAAI,GACzB,MAAO,CACL,OAAQ,WAAM,MAAA,IAAI,EAAU,EAAI,GAAI,EAAU,CAAC,GAAM,GAAW,QClCtE,AAuBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAQ,QC7BrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,SACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,KAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAO,EAAI,OC1BhC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAI,GAAK,GAAK,EAAG,YAAa,QC/B3D,AAuBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,GAAK,EAAG,WAAY,QC5BrD,AAwBO,GAAM,IAA0C,CACrD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAM,GAAO,GACb,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MACrC,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MAC3C,MAAO,CAAC,EAAS,EAAG,KChCxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,GAAI,GAAM,EAAE,QAG7B,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAkB,EAAE,MAAM,QACzB,EAAA,EAAA,KAED,EAAO,GAAe,EAAM,EAAE,OACpC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,EAAI,EAAY,GAAK,EAAE,MAAO,YAE3C,MAAO,CAAC,EAAG,WAAM,MAAA,OCxCrB,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAO,GAAI,SC9BxC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAO,GAAI,GAAO,IAAK,OC/BpD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KAED,EAAO,WACX,GAAI,GAAQ,GAAU,GAGtB,GAAI,EAAE,OAAS,EACb,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,IAAK,CAAC,EAAE,MAAM,cAEjD,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IAAK,CACjD,EAAE,MAAM,GAAI,EAAE,MAAM,cAI7B,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IACjD,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,cAI1C,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EACA,CACE,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAC5C,EAAI,EAAE,MAAM,IAEd,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,UAMjE,MAAM,IAAI,OACN,2DACG,GAAE,KAAI,kBAEf,MAAO,IAET,MAAO,CAAC,EAAG,KCpFf,AAuBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAiC,EAChC,EAAA,EAAA,KACD,EAAW,GAAiC,GAClD,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,EAAI,OC7BnC,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACxC,GAAM,GAA2B,EAC1B,EAAA,EAAA,KACP,MAAO,CAAC,MAAO,WAAM,MAAA,IAAM,EAAI,OC1BnC,AA8BO,GAAM,IAA2C,CACtD,WAAY,GACZ,aAAc,CAAC,cACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAAO,WACX,MAAO,IAAoB,EAAI,IAEjC,MAAO,CAAC,EAAG,KAIf,YAA+C,EAAM,GAQnD,OAJM,GAAqB,GAAQ,EAAS,GAAU,IAChD,EAAW,GAAO,EAAG,GACvB,EAAa,GAAa,EAAS,GAAO,EAAG,UAC3C,EAAW,EAAS,KAAO,EAAW,KACnC,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,GAAW,EAAY,EAAI,GAE1C,EAAa,GAAW,EAAY,GAAK,EAAS,MAAO,SACzD,GAAM,GAAY,GAAU,GAC5B,MAAO,IAAM,EAAY,EAAU,GCxDrC,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAkHA,GAAM,IAA4B,CAChC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA6B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAxB,GAAM,IAAc,GAAA,IACvB,GAAiB,ICxNnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA+BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACL,EAAO,KAAK,OAAS,EAAG,WAAM,MAAA,wCACvB,EAAQ,KAAM,KClCvB,AAmCA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAQ,KAAM,ICtCvB,AA8BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AAkCA,EAAO,UAAU,KAAO,SACpB,EAAc,GAChB,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,KCrC9B,AAoCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,GACjC,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,KCvCvC,AAqCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,GAChD,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,KCxC9C,AAwCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,EAC9C,GACF,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,EAAQ,KC5CtD,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA+BA,EAAO,UAAU,UAAY,SACzB,EACA,EACA,EACA,EACA,GACF,YAAK,kBACE,GAAU,KAAM,EAAM,EAAU,EAAQ,EAAO,ICtCxD,AA2BA,EAAO,UAAU,YAAc,SAAyB,GAEtD,YAAK,kBACE,GAAY,KAAM,IC9B3B,AA4BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,YAAc,SAC3B,EAAa,GACf,YAAK,kBACE,GAAY,KAAM,EAAK,IC/BhC,AA0BA,EAAO,UAAU,OAAS,SACtB,EAA0B,GAC5B,YAAK,kBACL,AAAI,YAAa,IACf,GAAI,CAAC,IAEA,GAAM,CAAE,MAAI,OAAK,GAAI,IChC9B,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4C,EAC5C,EAAmB,GACrB,YAAK,kBACE,GACI,KAAM,EAAQ,EAAQ,EAAK,EAAY,EACvC,ICrCb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EACA,EAAkC,EAClC,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAa,EAAS,EAAK,ICrCtD,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAqB,GACtC,YAAK,kBACE,GAAO,KAAM,EAAM,EAAW,IC9BvC,AA2BA,EAAO,UAAU,aAAe,SAC5B,EAAmB,GACrB,YAAK,kBACE,GAAa,KAAM,EAAW,IC9BvC,AAkCA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,UAAgB,8DAChB,KAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,IC3Cb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA6BA,EAAO,UAAU,WAAa,SAC1B,EAA+B,EAC/B,EAAqB,EACrB,GACF,YAAK,kBACE,GAAW,KAAM,EAAQ,EAAS,EAAK,EAAW,IClC3D,AA2BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC9BxB,AA4BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC/BzB,AA2BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC7BnB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC5Bb,AA+BA,EAAO,UAAU,YAAc,SAClB,GACX,YAAK,kBACE,GAAY,KAAM,IClC3B,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA0BA,EAAO,UAAU,WAAa,SAA2B,GACvD,YAAK,kBACE,GAAW,KAAM,IC5B1B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA4BA,EAAO,UAAU,OAAS,SACb,EAA4B,GACvC,YAAK,kBACE,GAAO,KAAM,EAAS,IC/B/B,AA+BA,EAAO,UAAU,mBAAqB,SACzB,GACX,YAAK,kBACE,GAAmB,KAAM,IClClC,AA0BA,EAAO,UAAU,aAAe,SAA2B,GAEzD,YAAK,kBACE,GAAa,KAAM,IC7B5B,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA+BA,EAAO,UAAU,gBAAkB,SACtB,GACX,YAAK,kBACE,GAAgB,KAAM,IClC/B,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA0BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC5BpB,AA2BA,EAAO,UAAU,2BAA6B,SAC1C,EAAsB,EAAe,EAAgB,GACvD,YAAK,kBACE,GAA2B,KAAM,EAAa,EAAM,EAAO,IC9BpE,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA2BA,EAAO,UAAU,UAAY,SAChB,EAAwB,GACnC,YAAK,kBACE,GAAU,KAAM,EAAM,IC9B/B,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA0BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC5BpB,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA2BA,EAAO,UAAU,OAAS,SACb,EAAiB,EAAsB,GAClD,YAAK,kBACE,GAAO,KAAM,EAAG,EAAY,IC9BrC,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,KAAO,SACpB,EAAwB,GAC1B,YAAK,kBACE,GAAK,KAAM,EAAM,IC7B1B,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,EAAI,KAAM,IC5BnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,KAAO,SACpB,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAK,EAAM,IClC/B,AA+BA,EAAO,UAAU,eAAiB,SACrB,GACX,YAAK,kBACE,GAAe,KAAM,IClC9B,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAa,GAAb,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAC9B,KAAK,kBACE,GAAO,KAAM,EAAO,EAAS,IC9BtC,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,IAAM,SACV,EAAmC,GAC9C,YAAK,kBACE,GAAI,KAAM,EAAU,IC/B7B,AA6BA,EAAO,UAAU,KAAO,SACX,EAAsC,EAC/C,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAa,EAAa,EAAS,EAAc,IClCrE,AA+BA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,GAAU,KAAM,ICjCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,MAAQ,SACZ,GACX,YAAK,kBACE,GAAM,KAAM,IC7BrB,AA2BA,EAAO,UAAU,KAAO,SACX,EAAwB,GACnC,YAAK,kBACE,GAAK,KAAM,EAAM,IC9B1B,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA0BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC5Bf,AAkCA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,EAAQ,KAAM,EAAE,QCpCzB,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,EAAQ,KAAM,IC5BvB,AA2BA,EAAO,UAAU,eAAiB,SACrB,EAA8B,GACzC,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,sBAAwB,SAC5B,EAA8B,GACzC,YAAK,kBACE,GAAsB,KAAM,EAAY,IC9BjD,AA0BA,EAAO,UAAU,QAAU,SACd,GACX,YAAK,kBACE,GAAQ,KAAM,IC7BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EAAsC,EACtC,EAAqB,EACrB,GACF,YAAK,kBACE,GACI,KAAM,EAAiB,EAAiB,EAAS,EAAK,EACtD,ICtCb,AA4BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,OC9BjB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA6BA,EAAO,UAAU,MAAQ,SACZ,EAAwB,GACnC,YAAK,kBACE,GAAM,KAAM,EAAO,IChC5B,AA4BA,EAAO,UAAU,QAAU,SAAoC,GAC7D,YAAK,kBACE,GAAQ,KAAM,IC9BvB,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,MAAQ,SACrB,EAAkC,GACpC,YAAK,kBACE,GAAM,KAAM,EAAiB,IC9BtC,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAO,OC9BhB,AA2BA,EAAO,UAAU,kBAAoB,SAA2B,GAE9D,YAAK,kBACE,GAAkB,KAAM,IC9BjC,AA+BA,EAAO,UAAU,wBAA0B,SAC9B,GACX,YAAK,kBACE,GAAwB,KAAM,IClCvC,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,MAAQ,SACrB,EAAoB,GACtB,KAAK,kBACL,GAAM,GAAqB,YAAa,GAAS,CAAC,KAAM,GAAE,CAAI,MAAI,OAAK,GACvE,MAAO,IAAM,EAAoB,IC9BnC,AA4BA,EAAO,UAAU,KAAO,SAAoC,GAC1D,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA+BA,EAAO,UAAU,aAAe,SACd,EAAiB,EAAe,EAC9C,EAAoB,EAAkB,EACtC,EAAsB,GACxB,YAAK,kBACE,GACI,KAAM,EAAO,EAAK,EAAS,EAAW,EAAS,EAC/C,EAAa,ICtC1B,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC7BzB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC7BpB,AAgCA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAQ,KAAM,SClCvB,AAgCA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,KAAM,YClCvB,AAgCA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAQ,KAAM,UClCvB,AA4BA,EAAO,UAAU,KAAO,SACX,EAAY,GACvB,YAAK,kBACE,GAAK,KAAM,EAAG,IC/BvB,AA2BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC9BzB,AA2BA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAO,KAAM,IC9BtB,AA4BA,EAAO,UAAU,mBAAqB,SACzB,EAAmC,GAC9C,YAAK,kBACE,GAAmB,KAAM,EAAY,IC/B9C,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,MAAQ,SACrB,EAA8B,GAChC,YAAK,kBACE,GAAM,EAAW,KAAM,IC/BhC,AA4BA,EAAO,UAAU,UAAY,WAC3B,YAAK,kBACE,GAAU,4vQC9BnB,AAaA,GAAI,kBAMF,MAAI,KAAY,MACd,IAAW,EAAA,UAAU,WAEhB,iBAeP,MAAO,eCrCT,AAuBA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAChC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAa,aAE7C,MAAA,IANkC,OAWlC,EAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OAWhC,GAAA,SAAA,GAAyC,EAAA,EAAA,GACvC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAoB,aAEpD,MAAA,IANyC,OAWzC,GAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OC9EhC,YAuB6B,EAAY,GACvC,GAAI,MAAM,QAAQ,IAGhB,OADI,GAAkB,GACb,EAAI,EAAG,EAAI,EAAW,IAC7B,EAAW,EAAS,OAAO,GAE7B,MAAO,QAEP,GAAM,GAAW,GAAI,OAAM,GAC3B,SAAS,KAAK,GACP,eAIY,EAAc,GACnC,GAAI,CAAC,EACH,KAAM,IAAI,IAAe,eAOJ,EAAY,GAEnC,OADI,GAAU,EACK,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,IAAS,GACX,IAGJ,MAAO,eAQ2B,GAClC,MAAI,GAAG,SAAW,EACT,EAAG,GAEL,cAYc,GACrB,MAAI,OAAM,QAAQ,GACT,EAEF,CAAC,eA0BkB,GAC1B,GAAM,GAAe,EAAK,QAAQ,uBAAwB,SACpD,EACF,EAAa,QAAQ,kBAAmB,SAAS,cAKrD,MAAI,GAAS,KAAO,IACX,EAEF,UAAY,cAGO,GAE1B,MAAI,GAAW,QAAU,GAIrB,EAAW,QAAQ,OAAS,GAHvB,EAMF,EAAW,QAAQ,cAAe,SAAC,EAAG,GAAO,MAAA,GAAG,gBAIzD,GAAI,IAAyB,eAEQ,GAEnC,GAAI,GAAa,KACf,MAAO,MAET,GAAM,GAAsC,GAC5C,SAAK,UAAe,EAAS,eAC7B,EAAK,OAAY,EAAS,YACnB,EAcT,YAAuC,GAErC,GAAI,GAAU,MAAQ,MAAO,IAAW,SACtC,OACK,GAAI,MAAM,QAAQ,GACvB,EAAO,QAAQ,SAAA,GAAc,MAAA,IAA8B,SAG3D,QADM,GAAS,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAO,GACrB,AAAI,GAAS,MAAQ,MAAO,IAAU,UACpC,CAAI,CAAC,MAAM,QAAQ,IAAU,EAAM,OAAY,WAC3C,MAAO,GAAM,OAAa,SAC5B,EAAO,GAAS,EAAM,MAEtB,GAA8B,iBAqBpC,EACA,EACA,EACA,EAAgC,aAElC,GAJE,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,UAAgC,AAAA,IAAA,QAAA,GAAA,IAE9B,MAAO,IAAe,UACxB,GAAM,GAAe,EACjB,EAAE,OACN,GAAI,IAAgB,GAClB,EAAK,EAAc,WACV,IAAgB,IACzB,EAAK,GAAuB,WAE5B,EAAK,EAAc,GACf,GAAM,KACR,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAU;EAE7C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAKR,MAAO,QAGP,GAAM,GAAS,EACf,GAAI,EAAO,WAAgB,MAAQ,EAAO,QAAa,KACrD,KAAM,IAAI,GACH,EAAmB,6BACnB,MAAK,UAAU,GAAO;GACzB,sCAEN,GAAM,GAAY,EAAO,UACrB,EAAG,OAAE,EAAU,OAQnB,GAPA,AAAI,IAAa,GACf,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,GAAA,UAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IAEJ,GAAO,KACT,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAS;EAE5C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAIN,GAAI,GAAc,MAQhB,OADM,GAAwB,GACZ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,GAAuB,GAEtD,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,EAAc,GAG7C,GAAM,GAAe,EAAO,OAC5B,EAAa,cAAmB,EAGhC,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAE9C,GAA8B,EAAO,QACrC,GAAM,GACF,EAAW,EAAK,EAAO,OAAW,EAAe,GACrD,UAAsB,GAAA,GAAO,GAEtB,OAMP,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAK9C,GAAM,GAAY,GAAI,GAAI,EAAO,QACjC,UAAsB,GAAA,GAAO,GACtB,gBAUiB,EAAW,GACvC,MAAQ,GAAI,EAAK,GAAO,EAAI,EAAK,EAAI,cAQF,EAAW,GAC9C,MAAO,GAAK,GAAc,EAAG,eA2CL,GACxB,GAAI,GAAM,KACR,MAAO,GAIT,OAFM,GAAW,GAED,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAI,QAAQ,KAAO,IACrB,EAAI,KAAK,GAGb,MAAO,eASqB,GAC5B,GAAI,GAAO,KACT,KAAM,IAAI,GAAW,yBAAyB,KAAK,UAAU,IAE/D,OAAW,KAAO,GAChB,GAAI,EAAI,eAAe,GACrB,MAAO,GAGX,MAAO,eAWL,EAAkB,EAAe,GACnC,GAAI,GAAS,KACX,OAEF,GAAI,EAAO,QAAQ,GAAS,EAC1B,KAAM,IAAI,GAAc,EAAK,mBAAmB,EAAK,uBACjD,EAAM,mCAoBV,EAAQ,EAAsB,EAC9B,GAD8B,MAAA,KAAA,QAAA,GAAA,GAC9B,AAAA,IAAA,QAAA,GAAA,UACF,GAAO,GAAa,GACpB,GAAO,GAAa,GAEhB,MAAM,QAAQ,IAAM,EAAE,QAAU,GAAa,EAAE,QAAU,GACzD,EAAE,MAAM,SAAA,GAAK,MAAA,OAAO,KAAM,gBAWM,EAAwB,GAC5D,AAAI,MAAM,QAAQ,GAChB,GAAA,KAAK,OACD,EAAM,OAAS,EAAG,WAAM,MAAG,GAAI,qCACnC,EAAM,QACF,SAAC,EAAG,GAAM,MAAA,IAAsB,EAAG,WAAW,GAAI,GAAC,OAAO,MAE9D,EAAA,KAAK,OACD,OAAO,UAAU,IAAU,EAAQ,EACnC,WAAM,MAAA,YAAY,EAAI,sCACf,IAAuB,GAAM,mBAeL,GACrC,MAAI,KAAU,KACL,OACE,MAAM,QAAQ,GAChB,IAAM,EAAM,IAAI,SAAA,GAAK,MAAA,IAAuB,KAAI,KAAK,KAAO,IAC1D,MAAO,IAAU,SACnB,IAAI,EAAK,IAET,GAAG,cAcV,EAA8B,GAChC,GAAI,GAAW,EAAA,KAAK,MAChB,EACE,EAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,GAAM,GAAM,EAAA,KAAK,MACjB,MAAI,GAAM,EAAW,GAGrB,GAAW,EACX,EAAa,EAAC,MAAA,OAAI,IACX,GAET,MAAO,eASkC,GAEzC,MAAI,KAAmB,OACd,OAEL,IAAmB,SACd,SAEL,IAAmB,MACd,MAEF,KCvgBT,AAoBA,YAAqB,EAAW,GAC9B,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAI,EAAM,OAY1D,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAGE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GANyC,EAAA,cAAc,cA8BvD,GAAA,SAAA,GAA6B,EAAA,EAAA,GAQ3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAJQ,SAAA,gBAAkB,EAClB,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,YAAgB,EAAO,EAAG,EAAK,UAC/C,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,SAAU,KAAK,SAAU,KAAM,KAAK,OAtB9B,EAAA,UAAY,UAwB9B,GA1B6B,IA2B7B,EAAA,cAAc,cAAc,IAoB5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAK5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAFQ,SAAA,YAAc,EAG7B,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,KAAW,GAAY,EAAG,EAAK,WAG9D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OAdL,EAAA,UAAY,WAgB9B,GAlB8B,IAmB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAIE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAHF,EAAA,UAAY,SAK9B,GAP4B,IAQ5B,EAAA,cAAc,cAAc,IAoC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAY9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KANQ,SAAA,gBAAkB,EAClB,EAAA,gBAAkB,EAClB,EAAA,YAAc,EACd,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,YACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,IACZ,EAAA,IACI,EAAK,KAAM,EAAA,YAAgB,EAAO,EAAK,SAAU,EAAK,WAC1D,EAAA,IAAQ,EAAM,EAAK,KAAM,IAC7B,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,SAAU,KAAK,SACf,SAAU,KAAK,SACf,KAAM,KAAK,KACX,KAAM,KAAK,OApCC,EAAA,UAAY,aAuC9B,GAzCgC,IA0ChC,EAAA,cAAc,cAAc,IAQrB,GAAM,IACwC,CAC/C,QAAW,UACX,WAAc,aACd,OAAU,SACV,SAAY,wBAGkB,GAElC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA0C,GAC1C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCpQjC,YAyBwB,GACtB,MAAO,IAAI,IAAQ,eAQI,GACvB,MAAO,IAAI,IAAS,iBASpB,MAAO,IAAI,gBAIc,GACzB,MAAO,IAAI,IAAW,0ECjDxB,AAqBO,GAAM,IAA2B,CAAC,gBAAiB,gBAO7C,GAA4B,CAAC,QAAS,OAAQ,UAI9C,GAAyB,CAAC,MAAO,OAIjC,GAAkC,CAAC,MAAO,MAAO,SAAU,OCpCxE,AAmBA,GAAM,IAA+B,GAAI,iBAET,GAC9B,GAA0B,GAA0B,aAAc,eAGnC,GAC/B,GAA0B,GAA2B,cAAe,eAGxC,GAC5B,GAA0B,GAAwB,WAAY,GAGhE,GAAM,IAA4B,GAC5B,GAAoB,gBAKG,EAAc,GACzC,GAAgB,KAAK,GACrB,IACE,GAAM,GAAS,IACf,UAAgB,MACT,QACA,GACP,SAAgB,MACV,GAOV,cACE,MAAI,IAAgB,SAAW,EACtB,GAEA,GAAgB,KAAK,IAAqB,eASjB,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,MAAO,MAA2B,cAaA,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,AAAK,GAAQ,IAAI,IACf,GAAQ,IAAI,EAAY,GAE1B,GAAM,GAAQ,GAAQ,IAAI,GAG1B,GAFA,GAAQ,IAAI,EAAY,GAAQ,IAAI,GAAc,GAE9C,EAAQ,GACV,GAAM,GAAY,EAAU,IAAI,EAGhC,UAAQ,IAAI,EAAQ,GACb,MAEP,OAAO,GAIX,GAAM,IAAkB,GAAI,QAAO,+CAOD,GAChC,MAAO,CAAC,CAAC,EAAK,MAAM,ICjHtB,YA8B0B,GACxB,MAAO,KAAM,SAAS,EAAE,WAAY,gBAWlC,EAA4B,EAAgB,GAC9C,AAAI,GAAS,MACX,GAAQ,GAEV,AAAI,GAAO,MACT,GAAM,EAAM,QAId,OADI,GAAO,EACF,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,GAAQ,EAAM,GAEhB,MAAO,GAQT,YAAmB,GACjB,SAAQ,MAAM,QAAQ,GAAS,GAAI,cAAa,GAAS,EAClD,EAAA,SAAS,eAQE,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAQ1B,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAsDxB,EAAe,GACnC,GAAI,EAAM,EACR,KAAM,IAAI,GAAW,QAAQ,EAAG,cAAc,EAAK,mBAGrD,OADM,GAAgB,GACb,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,EAAI,KAAK,GAEX,MAAO,GChJT,YAwEqB,EAAW,GAC9B,MAAO,GAAE,OAAO,eASS,EAAW,GAAA,AAAA,IAAA,QAAA,GAAA,IACpC,GAAM,GAAW,EAAE,MAAM,QACzB,MAAI,GAAO,GACT,GAAO,EAAS,OAAS,EAAO,GAElC,EAAS,OAAO,EAAM,EAAG,GAClB,EAAE,QAAQ,eAcI,EAAW,GAChC,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,oDACA,SAAQ,EAAE,MAAM,OAAM,aAE5B,GAAM,GAAI,GAAW,EAAG,GACxB,MAAO,IAAK,EAAG,CAAC,EAAG,EAAG,kBASF,GACtB,GAAM,GAAW,CAAC,GAAqB,EAAE,QACzC,MAAO,GAAE,QAAQ,eAWU,GAC3B,GAAI,EAAE,MAAQ,EACZ,KAAM,IAAI,GACN,wDAAwD,EAAE,KAAI,KAEpE,GAAM,GAAW,CAAC,EAAE,MAAM,GAAI,GAAqB,EAAE,MAAO,IAC5D,MAAO,GAAE,QAAQ,eAYf,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,GAAI,CAAC,EAAM,EAAM,MAAM,SACnD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,GAC9B,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,EAAG,GACjC,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,MAAU,EAAmB,CAAC,EAAO,EAAG,EAAG,EAAG,GAAI,CACvD,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SAEjE,GACH,MAAO,GAAA,MAAU,EAAO,CAAC,EAAO,EAAG,EAAG,EAAG,EAAG,GAAI,CAC9C,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAClE,EAAM,MAAM,aAGd,KAAM,IAAI,GACN,8DACA,IAAG,EAAM,sBAcjB,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,GAAQ,CAAC,EAAM,MAAM,GAAI,QACjD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,GAC1B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,QAClC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAG,GAC7B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,YAErD,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAejB,EAAe,EAAe,EAAc,GAC9C,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,GAC9B,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,EAAG,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAO,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpD,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,YAGX,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAWO,EAAmB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC7C,GAAI,GACJ,MAAI,GAAO,GACT,GAAO,EAAQ,GAAG,KAClB,AAAI,IAAS,EACX,EAAO,EAEP,EAAO,GAGX,AAAI,IAAS,EAAQ,GAAG,MAGtB,GAAO,IAGF,EAAA,OAAW,EAAS,eAUQ,EAAW,GAC9C,OAAQ,EAAE,UACH,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,QACjC,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,WAEpD,KAAM,IAAI,GACN,kDACA,iBAAgB,EAAE,oBAWP,EAAW,GAI9B,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAI,CAAC,IAEH,EAAE,OAAS,EAAE,OACf,KAAM,IAAI,GACN,0BAA0B,EAAE,OAAM,oBAClC,yCAAwC,EAAE,KAAI,MAEpD,MAAO,GAAA,KAAS,EAAG,eAgBjB,EAAc,EAAY,EAAc,EACxC,GADc,MAAA,KAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAErB,EAAA,aAAiB,EAAO,EAAM,EAAQ,EAAO,eAsBlD,EAAW,EAAW,EACtB,GACF,GAAK,EAAE,KAAO,GAAO,EAAE,KAAO,EAC5B,KAAM,IAAI,IACN,2CACA,uBAAsB,EAAE,MAAK,kBAAkB,EAAE,QAEvD,GAAI,EAAE,MAAQ,GACZ,GAAM,GAAW,EAAE,MAAM,MAAM,IAAI,GAC7B,EAAiB,EAAE,MAAM,MAAM,IAAI,GACzC,GAAI,IAAa,EACf,KAAM,IAAI,IACN,2CACA,yDACI,EAAE,MAAK,SACX,eAAc,EAAE,QAIxB,GAAK,EAAE,OAAS,GAAO,EAAE,OAAS,GAChC,GAAM,GAAa,GACb,EAAa,GAInB,MAAO,GAAA,MAAU,OAAO,CACtB,EAAC,EACD,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,SAIZ,GAAM,GAAa,EAAE,MAAM,QACrB,EAAW,EAAW,MAC5B,EAAI,EAAE,QAAQ,CAAC,GAAI,IAInB,GAAM,GAAS,EAAE,MAAM,QACjB,EAAW,EAAO,MAClB,EAAiB,EAAO,MACxB,EAAiB,EAAM,OAAA,CAAE,IAGzB,EAAO,MAAM,KAAK,CAAC,OAAQ,EAAE,MAAO,SAAC,EAAG,GAC5C,MAAI,KAAM,EACD,EAAE,KAAO,EACP,GAAK,EAAE,KAAO,EAChB,EAAI,EAEN,IAET,EAAI,EAAE,UAAU,GAAM,QAAQ,CAAC,EAAgB,KAG/C,GAAM,GAAkB,EAAU,OAAK,GACjC,EAAa,GACb,EAAa,GACnB,MAAO,GAAA,MACF,OAAO,CACN,EAAC,EACD,EAAC,EACD,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,IAEX,QAAQ,gBAyDb,EAAmB,EAA4B,GACjD,MAAO,GAAA,KAAK,WACV,MAAI,OAAM,QAAQ,GAChB,EAAU,EAAA,SAAS,EAAS,SAE5B,EAAU,EAAQ,QAEb,EAAA,OAAW,EAAW,EAAS,iBASnB,GACrB,MAAO,GAAA,IAAQ,EAAG,GA+BpB,YAAqB,EAAe,EAAc,GAChD,GAAM,GAAY,EAAK,MAEvB,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,GACN,+BAA+B,EAAK,KACpC,6BAA4B,IAGlC,GAAI,IAAU,GACZ,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,EAAG,IAErC,EAAK,QACR,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,GAAI,EAAU,KAEzD,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAG,EAAU,KAEpC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,IAElC,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,KAE3D,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAU,KAEjC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,IAE/B,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,KAE7C,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAU,KAE9B,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,EAAQ,EACjB,MAAO,GAET,KAAM,IAAI,GAAW,sCAAsC,EAAK,kBAc9D,EAAW,EAAc,GAC3B,MAAO,GAAA,KAAK,WACV,MAAI,IAAc,MAChB,GAAa,MAEf,GAAgB,GAET,EAAE,IAAI,GAAY,EAAE,KAAM,EAAM,kBAUvB,EAAW,GAE7B,GAF6B,AAAA,IAAA,QAAA,GAAA,GAEzB,IAAU,EACZ,KAAM,IAAI,IACN,0CAA0C,EAAK,6BAGrD,MAAO,GAAA,IAAQ,eAWQ,GACvB,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,GAAG,IAAI,kBAc1C,EAAW,EAAe,EAAuB,GACnD,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAG,EAAO,EAAY,iBAY1B,GAC1B,MAAO,GAAA,KAAK,WACV,GAAM,GAAI,EAAA,IAAQ,GAAI,EAAA,IAAQ,GAAI,IAClC,MAAO,GAAA,YAAgB,EAAG,EAAG,iBAiBD,EAAY,EAAc,GAAA,MAAA,KAAA,QAAA,GAAA,IACjD,EAAW,IAAM,IC7sB1B,AAgBO,GAAM,IAAwB,CAAC,QAAS,SAAU,UAO5C,GACT,CAAC,SAAU,UAAW,mBCxB1B,YAoB6B,GAC3B,GAA0B,GAAuB,UAAW,eAG5B,GAChC,GAA0B,GAA2B,eAAgB,GASvE,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DACS,SAAA,UAAA,4BAAP,WACE,MAAO,IAUT,EAAA,UAAA,UAAA,WACE,MAAO,IAEX,GAf0C,EAAA,cAAc,cAiBxD,GAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,MAAM,EAAO,IAHf,EAAA,UAAY,QAKrB,GAP2B,IAQ3B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,KAAK,EAAO,IAHd,EAAA,UAAY,OAKrB,GAP0B,IAQ1B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,GACN,oDAAoD,GAE1D,GAAI,EAAK,QAAU,OACjB,KAAM,IAAI,GAAW,sCAAsC,GAE7D,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAI,EAAA,OAAO,EAAK,OAAQ,EAAA,KAAK,EAAO,OAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,QApBT,EAAA,UAAY,WAuBrB,GAzB8B,IA0B9B,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GASjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,eAAiB,KACjB,EAAA,eAAiB,IAOxB,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,cAAc,EAAO,KAAK,OAAQ,KAAK,OAAQ,IAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,OAAQ,KAAK,OAAQ,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAnBxD,EAAA,UAAY,gBAqBrB,GAvBmC,IAwBnC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAShC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,uCAAuC,EAAK,KAGlD,MAAO,IAAe,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGnE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,eA2BrB,GA7BkC,IA8BlC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,0CAA0C,EAAK,KAErD,MAAO,GAAA,gBAAgB,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGpE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,kBA2BrB,GA7BqC,IA8BrC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,IAG9C,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,SAAW,GAAK,EAAM,KAAO,EAAM,GAC3C,KAAM,IAAI,GACN,wEAGJ,MAAO,GAAA,IAAI,EAAK,KAAM,EAAA,IAAI,EAAM,QAKtC,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OApBd,EAAA,UAAY,WAsBrB,GAxB8B,IAyB9B,EAAA,cAAc,cAAc,IAU5B,YACI,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,gBAChB,GAAI,GACA,EAEJ,GADA,GAAgB,GACZ,EAAM,SAAW,EACnB,EAAQ,EAAM,GACd,EAAS,EAAM,WACN,CAAC,EAAG,EAAG,GAAG,QAAQ,EAAM,UAAY,IAC7C,GAAI,IAAe,iBACjB,GAAM,GAAqB,GAAU,EAAO,GAC5C,EAAQ,EAAM,GAAK,EACnB,EAAS,EAAM,GAAK,UACX,IAAe,gBACxB,GAAM,GAAqB,GAAU,EAAO,EAAG,EAAM,OAAS,GAC9D,EAAQ,EAAM,EAAM,OAAS,GAAK,EAClC,EAAS,EAAM,EAAM,OAAS,GAAK,QAGrC,GAAM,GAAY,GAAU,GAC5B,EAAQ,KAAK,KAAK,GAClB,EAAS,KAAK,KAAK,GAGrB,MAAO,CAAC,EAAO,GAiBjB,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAYnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,EAAK,MAAQ,EACf,KAAM,IAAI,GACN,wCAAwC,EAAK,OAEnD,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAM,EAAK,MAC7C,EAAK,KAAO,EAAK,MAAQ,KAAO,QAAU,EAAK,KAC/C,GAAa,EAAK,MAClB,EAAK,aACD,EAAK,cAAgB,KAAO,SAAW,EAAK,aAChD,GAAkB,EAAK,cACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAO,GAAY,GACnB,EAAQ,EAAK,GACb,EAAS,EAAK,GAChB,EAAQ,KAAK,MASjB,GARA,AAAI,KAAK,OAAS,QAChB,GAAS,KAAK,IAAI,EAAG,GAChB,AAAI,KAAK,OAAS,SACvB,GAAS,KAAK,IAAI,EAAG,GAErB,GAAS,KAAK,IAAI,EAAI,GAAQ,GAAU,GAGtC,KAAK,eAAiB,UACxB,GAAM,GAAS,KAAK,KAAK,GAEzB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACH,KAAK,eAAc,2BAA2B,EAAK,KAE5D,MAAO,GAAA,gBAAgB,EAAO,EAAG,EAAQ,EAAO,KAAK,WAErD,GAAM,GAAQ,KAAK,KAAK,EAAI,GAC5B,MAAO,GAAA,cAAc,EAAO,CAAC,EAAO,EAAO,KAI/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,MACZ,KAAM,KAAK,KACX,aAAc,KAAK,aACnB,KAAM,KAAK,OAzDR,EAAA,UAAY,kBA4DrB,GA9DqC,IA+DrC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,gBAwBrB,GA1BmC,IA2BnC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAWhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,eAwBrB,GA1BkC,IA2BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,WAiBrB,GAnB8B,IAoB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAI7B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,YAiBrB,GAnB+B,IAoB/B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAI/B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBiC,IAoBjC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAIhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBkC,IAoBlC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAO9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GATO,EAAA,aAAe,EAMtB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,KAAO,EAAK,KAEb,EAAK,MAAQ,KACf,KAAM,IAAI,IACN,2EAIR,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,IAAoB,8BAEhC,AAAI,EAAM,GAAK,EAAM,GAAK,KACxB,QAAQ,KACJ,gEACA,eAAc,EAAM,GAAK,EAAM,GAAE,gBACjC,wBAIN,GAAM,GACF,EAAM,GAAK,EAAM,GAAK,CAAC,EAAM,GAAI,EAAM,IAAM,EAC3C,EAAI,GAAe,EAAiB,EAAG,EAAG,WAC5C,EAAI,EAAA,OAAO,YAAY,GAC3B,MAAI,GAAM,GAAK,EAAM,IACnB,GAAI,EAAE,aAED,EAAA,IAAI,EAAK,KAAM,MAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CACL,KAAM,KAAK,KACX,KAAM,KAAK,OA3CR,EAAA,UAAY,aA8CrB,GAhDgC,IAiDhC,EAAA,cAAc,cAAc,IAUrB,GAAM,IACyC,CAChD,SAAY,WACZ,aAAgB,eAChB,cAAiB,gBACjB,SAAY,WACZ,UAAa,YACb,SAAY,WACZ,YAAe,cACf,aAAgB,eAChB,KAAQ,OACR,WAAc,aACd,aAAgB,eAChB,cAAiB,gBACjB,gBAAmB,kBACnB,gBAAmB,kBACnB,MAAS,SAGf,YACI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGgB,GAEnC,MAAO,IAAqB,eAGC,GAE7B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EAIJ,GAAI,IAAc,eAChB,MAAO,IAAI,IACN,GAAI,IAAc,gBACvB,MAAO,IAAI,IACN,GAAI,IAAc,WACvB,MAAO,IAAI,IACN,GAAI,IAAc,YACvB,MAAO,IAAI,IACN,GAAI,IAAc,cACvB,MAAO,IAAI,IACN,GAAI,IAAc,eACvB,MAAO,IAAI,IAEX,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAuB,OAE3B,OAAI,aAAsB,IACxB,EAEA,GAAuB,GCxoBlC,cAkBE,MAAO,IAAI,kBASX,MAAO,IAAI,gBAQY,GACvB,MAAO,IAAI,IAAS,eAYQ,GAC5B,MAAO,IAAI,IAAc,eASE,GAC3B,MAAO,IAAI,IAAa,eAaM,GAC9B,MAAO,IAAI,IAAgB,eASJ,GACvB,MAAO,IAAI,IAAS,eAgBU,GAC9B,MAAO,IAAI,IAAgB,eAgBC,GAC5B,MAAO,IAAI,IAAc,eAgBE,GAC3B,MAAO,IAAI,IAAa,eAeD,GACvB,MAAO,IAAI,IAAS,eAeI,GACxB,MAAO,IAAI,IAAU,eAgBK,GAC1B,MAAO,IAAI,IAAY,eAYI,GAC3B,MAAO,IAAI,IAAa,eAWC,GACzB,MAAO,IAAI,IAAW,kPChNxB,AAmBA,GAAI,IAAsB,gBAGxB,MAAO,MAGT,GAAM,IAA2C,eAO1B,GAAA,MAAA,KAAA,QAAA,GAAA,IACrB,AAAM,IAAU,KACd,IAAa,GAAU,GAEzB,GAAa,IAAW,EACjB,EAAS,GAAa,GAAQ,WCrCvC,YAoBgC,GAC9B,MAAO,OAAM,QAAQ,IAAM,MAAM,QAAQ,EAAE,gBASV,GACjC,MAAI,GAAE,SAAW,EACR,GAEJ,MAAM,QAAQ,EAAE,IAGd,EAFE,CAAC,eAWwB,GAClC,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAG,SAAW,EAChB,KAAM,IAAI,GAAW,uCAAuC,EAAG,QAEjE,EAAI,EAAG,OAEP,GAAI,EAEN,MAAO,eAa0B,GACjC,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,EAAO,KAChD,GAAI,EAAO,SAAW,EACpB,SAAS,EACF,EAAO,GAEd,KAAM,IAAI,GAAW,iCAAiC,EAAO,YAG/D,OAAO,GC9EX,YAmBqC,GAEnC,OADI,GAAQ,EACS,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,AAAI,EAAO,MAAM,SAAW,EAC1B,GAAS,EAET,GAAS,EAAO,MAAM,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAG/C,MAAO,GC5BT,AAoBA,GAAM,IAA+B,yBAqCnC,WACI,EAAa,EACb,EAAqC,EACrC,GAFa,AAAA,IAAA,QAAA,GAAA,WACb,AAAA,IAAA,QAAA,GAAA,IAAqC,AAAA,IAAA,QAAA,GAAA,IACrC,AAAA,IAAA,QAAA,GAAA,MACF,KAAK,MAAQ,GAAgB,UAC7B,KAAK,MAAQ,EAAI,MACjB,KAAK,GAAK,KAEV,EAAO,GAAe,GACtB,KAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,cAErC,KAAK,WAAa,EAClB,KAAK,WAAa,EAElB,KAAK,IAAM,EAAA,SAAa,EAAK,KAAK,WAAY,KAAK,KAAM,KAAK,OAUhE,SAAA,UAAA,KAAA,WACE,YAAK,oBACE,KAAK,KAUd,EAAA,UAAA,MAAA,SAAM,GAEJ,YAAK,oBACL,GAAiB,KAAK,IAAK,GAE3B,AAAI,KAAK,IAAI,KAAO,EAAO,IACzB,MAAK,IAAI,OAAO,GAChB,AAAI,KAAK,YAAc,MACrB,KAAK,IAAI,OAAO,KAAK,WAAW,MAAM,KAAK,OAGxC,MAMT,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,KAAK,IAAI,WAGD,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,IAAI,WACX,KAAM,IAAI,OAAM,kBAAkB,KAAK,KAAI,0BAI/C,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,WAAa,EAClB,KAAK,IAAI,UAAY,mCAEzB,KAEA,YAA0B,EAAa,GACrC,GAAI,EAAE,MAAM,aAAe,EAAE,MAAM,WACjC,KAAM,IAAI,OACN,mBAAqB,KAAK,UAAU,EAAE,OAAS,QAC/C,KAAK,UAAU,EAAE,oBA0LK,GAC5B,MAAO,GAAG,IAAI,SAAA,GAAK,MAAA,GAAE,qBAYnB,GACF,EAAmB,QAAQ,SAAA,GACzB,GAAM,GAA0B,EAAiB,GACjD,EAAS,MAAM,EAAiB,MClVpC,kBAuEE,WAAY,GACV,KAAK,MAAQ,EAAK,MAClB,KAAK,MAAQ,EAAK,MAKlB,AAAI,EAAK,OAAS,KAChB,KAAK,KAAO,EAAK,MAAM,OAEvB,KAAK,KAAO,EAAK,KAEnB,KAAK,QAAU,EAAK,QACpB,KAAK,QAAU,EAAK,QACpB,KAAK,KAAO,EAAK,MAAQ,GAE7B,MAAA,oBA6CE,WACa,EAA0B,EAC5B,EAA6B,EAC3B,EAAkB,EAClB,GAHA,KAAA,MAAA,EAA0B,KAAA,MAAA,EAC5B,KAAA,YAAA,EAA6B,KAAA,OAAA,EAC3B,KAAA,SAAA,EACA,KAAA,kBAAA,EACX,KAAK,GAAK,KACV,AAAI,GAAQ,MACV,MAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,eAEvC,KAAK,KAAO,EAAM,OAEtB,MAAA,MA2DI,GAAc,EAsBlB,GAAA,WAwCE,WACI,EAEO,GAAA,KAAA,SAAA,EACT,KAAK,GAAK,KAQV,KAAK,cAAgB,EAAK,cAU1B,KAAK,cAAgB,EAAK,cAE1B,KAAK,YAAc,EAAK,YAExB,KAAK,cAAgB,EAAK,cAQ1B,KAAK,aAAe,EAAK,aAEzB,KAAK,cAAgB,EAAK,cAM1B,KAAK,WAAa,EAAK,WAEvB,KAAK,YAAc,EAAK,YAKxB,KAAK,YAAc,EAAK,YAExB,KAAK,aAAe,EAAK,aAGzB,OAAoB,GAAA,EAAA,EAAA,EAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MACX,EAAM,cAAc,KAAK,MAG7B,EAAK,cAAc,aAAa,KAAK,MAGvC,SAAA,UAAA,UAAA,WAEE,OADM,GAAyB,GACX,EAAA,EAAA,EAAA,KAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,KACX,EAAa,KAAK,EAAM,MAExB,EAAa,KAAK,MAGtB,MAAO,CACL,cAAe,KAAK,cAAgB,KAAK,cAAc,KAAO,KAC9D,cAAe,EACf,YAAa,KAAK,YAClB,cAAe,KAAK,gBAG1B,KAkDI,GAAe,EAWnB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAmDlC,WAAY,GAAA,AAAA,IAAA,QAAA,GAAA,IAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAtBD,EAAA,UAAsB,KAEtB,EAAA,kBAA8B,GAQ5B,EAAA,UAAY,GAapB,EAAK,GAAK,KAEV,EAAK,oBAAsB,KAE3B,EAAK,UAAY,KACjB,EAAK,gBAAkB,GAGvB,EAAK,kBAAoB,GACzB,EAAK,qBAAuB,GAC5B,EAAK,QAAU,GACf,EAAK,SAAW,GAChB,EAAK,OAAS,GAMd,EAAK,aAAe,GACpB,EAAK,cAAgB,GAErB,GAAI,GAAO,EAAK,KAChB,GAAI,CAAC,GACH,GAAM,GAAS,EAAK,eACpB,EAAO,GAA0B,GAAU,IAAM,GAAO,GAM1D,GAJA,EAAK,KAAO,EAEZ,EAAK,WAAa,EAAK,WAAa,KAAO,GAAO,EAAK,UAEnD,EAAK,YAAc,MAAQ,EAAK,iBAAmB,MAKrD,GAAI,GAAe,OACnB,GAAI,EAAK,iBAAmB,KAC1B,EAAkB,EAAK,wBACd,EAAK,YAAc,MAC5B,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAkB,CAAC,GAAW,OAAO,EAAK,YAE5C,EAAK,gBAAkB,EAGvB,GAAI,GAAQ,EAAK,MACjB,AAAI,GAAS,MACX,GAAQ,EAAK,YAEf,AAAI,GAAS,MACX,GAAQ,WAEV,EAAK,MAAQ,EAGf,MAAI,GAAK,SAAW,KAClB,EAAK,eAAiB,EAAK,QAE3B,EAAK,eAAiB,KAKxB,EAAK,UAAY,KAEjB,EAAK,0BAA4B,KAYlB,SAAA,QAAjB,SAAyB,EAAc,GACrC,MAAO,GAAM,KAAO,OAAS,EAAU,YAUjC,EAAA,UAAA,eAAR,SAAuB,EAAmB,GACxC,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,mCACA,4BAA2B,EAAQ,MAEzC,GAAI,KAAK,aAAa,QAAU,EAC9B,KAAM,IAAI,GACN,gBAAgB,EAAQ,YAAY,EAAS,KAC7C,2BAA0B,KAAK,aAAa,OAAM,oBAExD,MAAO,MAAK,aAAa,IAY3B,EAAA,UAAA,WAAA,SAAW,GACT,MAAO,IACH,KAAK,eAAe,EAAW,SAAS,eAY9C,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,IACH,KAAK,eAAe,EAAW,UAAU,gBAgB/C,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,uHAIC,GAAI,KAAK,aAAa,SAAW,EACtC,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0CAEN,MAAO,IACH,KAAK,eAAe,EAAG,SAAS,+CActC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0BAEN,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,yHAKN,MAAO,IACH,KAAK,eAAe,EAAG,UAAU,gDAGvC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAQd,EAAA,UAAA,gBAAA,WAKE,MAAO,MAAK,OAAO,IAAI,SAAA,GAAU,MAAA,QAGnC,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,0CAGd,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,MAAO,MAAK,YAGd,SAAU,GACR,KAAK,OAAS,mCAGhB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,kBAAkB,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,IAClD,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAI,MAAK,WACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,GAAE,YAErC,QAIX,SAAqB,GACnB,KAAK,kBAAoB,mCAG3B,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAI,MAAK,UACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,CAAC,EAAE,YACxC,OAAO,KAAK,sBAEV,KAAK,kBAAkB,OAAO,KAAK,2BAI9C,SAAwB,GACtB,KAAK,qBAAuB,mCAO9B,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAG3C,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,2CAUd,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,SACR,KAAM,IAAI,OACN,yEAiBE,EAAA,UAAA,yBAAV,SAAmC,GAGjC,GADA,EAAS,GAAqB,GAC1B,KAAK,WAAa,MAAQ,KAAK,UAAU,SAAW,EACtD,OAEF,GAAM,GAAY,GAAqB,KAAK,WAC5C,GAAI,EAAO,SAAW,EAAU,OAC9B,KAAM,IAAI,GACN,SAAS,KAAK,KAAI,YAAY,EAAU,OAAM,YAC9C,oBAAmB,EAAO,OAAM,oBAChC,oBAAmB,IAEzB,OAAS,GAAa,EAAG,EAAa,EAAO,OAAQ,KACnD,GAAM,GAAI,EAAO,GACX,EAAkB,EAAU,GAClC,GAAI,GAAQ,KACV,SAIF,GAAM,GAAO,EAAE,KACf,GAAI,EAAK,MAAQ,MACX,IAAS,EAAK,KAChB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,KAC3D,kBAAiB,EAAK,KAAI,gBAAgB,IAGlD,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,IAG3D,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,EAAI,MAK/D,GAAI,EAAK,OAAS,MACZ,EAAE,QAAU,EAAK,MACnB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,IAC3D,qBAAoB,EAAK,MAAK,iBAAiB,EAAE,MAAK,MAK9D,GAAI,EAAK,MACP,GAAM,GAAS,EAAE,MACjB,OAAW,KAAO,GAAK,MACrB,GAAM,GAAO,OAAO,GACd,EAAQ,EAAK,KAAK,GAIlB,EACF,GAAQ,EAAI,EAAO,GAAQ,EAAO,EAAO,OAAS,GACtD,GAAI,GAAS,MAAQ,CAAC,EAAO,MAAM,QAAQ,KAAkB,GAC3D,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,mBAAmB,EAAI,uBACnC,eAAc,EAAK,kBAAkB,EAAM,OAMrD,GAAI,EAAK,OAAS,KAChB,OAAS,GAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAU,EAAK,MAAM,GACrB,EAAM,EAAE,MAAM,GACpB,GAAI,GAAW,MAAQ,GAAO,MACxB,IAAY,EACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,oBAAoB,EAAK,MAAK,MAC1C,gBAAe,EAAE,MAAK,SAgBtC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,IAGC,EAAA,UAAA,eAAV,SAAyB,EAAyB,GAChD,AAAI,KAAK,WAAa,MACpB,KAAK,UAAU,EAAQ,IAS3B,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,UAAY,GAOnB,EAAA,UAAA,cAAA,WACE,KAAK,UAAY,MAwEnB,EAAA,UAAA,MAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAGE,EAAS,GAAU,GAEnB,KAAK,oBAML,OAHM,GAAa,GAAqB,GAEpC,EAAiB,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,CAAE,aAAiB,MACrB,EAAiB,GACjB,OAIJ,OADI,GAAkB,GACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,YAAiB,KACnB,EAAkB,GAClB,OAIJ,GAAI,IAAmB,EACrB,KAAM,IAAI,GACN,mEAKN,MAAO,IAAU,KAAK,KAAM,WAE1B,GAAI,CAAC,EAAK,OAKR,EAAK,yBAAyB,GAI9B,OADM,GAAuB,GACT,EAAA,EAAA,EAAA,GAAqB,GAArB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,KAAK,EAAM,OAEzB,EAAK,MAAM,GAA+B,IAC1C,EAAK,MAAQ,GAGb,AAAI,EAAK,gBACP,EAAK,WAAW,EAAK,gBAGvB,AAAI,EAAK,YAAc,MAAQ,GAI7B,GAAK,UAAY,GAcrB,GANA,EAAK,yBAAyB,GAM1B,GAUF,OATI,GAAS,EAAK,KAAK,EAA6B,GAK9C,EAAuB,GAAqB,GAC5C,EAA2B,GAGnB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAT,GAAI,GAAC,EAAA,GACR,AAAI,EAAW,QAAQ,KAAO,IAC5B,GAAI,EAAE,SAER,EAAe,KAAK,GAItB,GAFA,EAAS,GAA+B,GAEpC,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAKN,MAAO,QAEP,GAAM,GAAa,GAAkB,GAC/B,EAAc,EAAK,mBAAmB,GACxC,EAAM,OACJ,EAAc,GAAiB,GAgCrC,GA/BA,EAAK,6BACD,MAAM,QAAQ,GAAU,EAAW,GACX,GAE5B,AAAI,GAAe,MAAQ,EAAY,OAAS,GAC5C,MAAM,QAAQ,EAAY,IAE5B,EAAU,EACI,IACG,SAAC,EAAO,GAAU,MAAA,IAAI,IAClB,EAAa,EAAO,EACpB,GAAqB,GAAS,EAAQ,EAAK,KAC3C,KAErB,EAAS,GAAI,IACT,EAAa,EAAsB,EACnC,GAAqB,GAAS,EAAQ,EAAK,MAUjD,EAAK,eACD,EAA6C,EAAQ,KAAM,KAC3D,EAAY,EAAa,GAC7B,EAAK,YAED,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAIN,MAAO,OAYH,EAAA,UAAA,6BAAV,SAAuC,GACrC,GAAI,KAAK,iBAAmB,KAC1B,OACK,GAAI,EAAW,SAAW,KAAK,gBAAgB,OACpD,QAAQ,KACJ,iDACG,MAAK,UAAU,GAAW,iCAC7B,qBAAoB,KAAK,UAAU,KAAK,iBAAgB,MACxD,iBAAgB,KAAK,YAEzB,GAAI,GAAc,GAClB,KAAK,gBAAgB,QAAQ,SAAC,EAAW,GACvC,AAAI,GAAa,MAAQ,EAAW,IAAM,MACtC,EAAW,KAAO,GACpB,GAAc,MAGlB,AAAI,GACF,QAAQ,KACJ,iCACA,KAAI,KAAK,UAAU,GAAW,eAC9B,mCAAkC,KAAK,KAAI,MAC3C,IAAG,KAAK,UAAU,KAAK,qBAiBjC,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,GAAI,KAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,EAC5D,KAAM,IAAI,IACN,aAAa,KAAK,KAAI,gEAI5B,OADM,GAA4B,GACf,EAAA,EAAA,EAAA,KAAK,aAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAc,KAAK,UAAU,EAAK,cACxC,AAAI,EAAgB,QAAQ,KAAiB,IAC3C,EAAgB,KAAK,GAGzB,GAAI,EAAgB,SAAW,GAC7B,GAAM,GAAe,KAAK,aAAa,GAAG,aAC1C,MAAI,OAAM,QAAQ,IAAiB,MAAM,QAAQ,EAAa,KAC1D,EAAa,SAAW,EAClB,EAAyB,GAE1B,MAIT,MAAM,IAAI,IACN,aAAa,KAAK,KAAI,+JAiB9B,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,sCAAsC,KAAK,KAAI,uFAIrD,MAAO,IAAoC,KAAK,UAclD,EAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAWf,EAAA,UAAA,WAAA,SAAW,GAAA,MAAA,KAAA,QAAA,GAAA,IACF,GAAc,EAAgB,KAAK,iBAAmB,KAAK,UAepE,EAAA,UAAA,WAAA,SAAW,GAAX,GAAA,GAAA,KACE,EAAA,KAAK,WACH,GAAM,GAAS,EAAK,QACpB,GAAI,EAAO,SAAW,EAAQ,OAK5B,KAAM,IAAI,GACN,4CAA4C,EAAK,KAAI,KACrD,iCAAgC,EAAQ,OAAM,MAC9C,gCAA+B,EAAO,OAAM,cAC5C,sBAAqB,EAAO,QAElC,GAAI,EAAO,SAAW,EACpB,OAIF,OAFM,GAAoD,GACpD,EAAc,GAAc,GACzB,EAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,GACxC,GAAM,GAAK,EAAY,GACjB,EAAI,EAAO,GACX,EAAI,EAAQ,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,GACN,sBAAsB,EAAG,MAAK,IAC9B,8CAA6C,EAAE,QAErD,EAAkB,KAAK,CAAC,EAAG,IAE7B,GAAc,MAmBR,EAAA,UAAA,UAAV,SACI,EAAc,EAAc,EAAkB,EAC9C,EAA2B,EAC3B,GAEF,GAAI,KAAK,kBAAkB,QAAQ,KAAU,GAC3C,KAAM,IAAI,GACN,yBAAyB,EAAI,cAAc,KAAK,MAEtD,KAAK,kBAAkB,KAAK,GAE5B,AAAI,GAAS,MACX,GAAQ,WAGV,AAAI,KAAK,2BACP,GAAc,GAAe,UAE/B,GAAM,GAAY,EAAY,MAAM,EAAO,GACrC,EACF,GAAI,IAAc,EAAW,EAAO,EAAM,EAAW,GACzD,SAAU,UAEV,AAAI,GAAe,MACjB,KAAK,QAAQ,WAAM,MAAA,GAAY,MAAM,EAAO,UAE9C,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,EACF,KAAK,kBAAkB,KAAK,GAE5B,KAAK,qBAAqB,KAAK,GAE1B,GAaT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,KAAK,0BAA4B,GAWnC,EAAA,UAAA,QAAA,SAAQ,SACN,GAAI,GAAU,MAAQ,MAAM,QAAQ,IAAW,EAAO,SAAW,EAC/D,OAGF,EAAS,GAAqB,GAC9B,AAAI,KAAK,UAAY,QAAa,KAAK,UAAY,MACjD,GAAA,KAAK,QAAO,KAAI,MAAA,EAAI,IAexB,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAYT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,GAAI,CAAC,KAAK,iBACR,GAAI,GAAQ,KACV,GAAI,MAAM,QAAQ,GAChB,EAAK,QAAQ,SAAA,GACX,GAAI,GAAe,KACjB,KAAM,IAAI,WACN,SAAS,EAAK,KAAI,iEAK1B,MAAM,IAAI,WACN,SAAS,KAAK,KAAI,2DAK1B,MAAO,MAIT,MAAO,IAeD,EAAA,UAAA,eAAR,SACI,EACA,EACA,EAA6B,EAC7B,EAA4B,EAC5B,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAM,GACF,GAAqB,GACzB,EAAgB,GAAqB,GACrC,EAAa,GAAqB,GAClC,EAAc,GAAqB,GACnC,EAAc,GAA+B,GAC7C,EAAe,GAA+B,GAM9C,OAHM,GAAyB,GACzB,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GAKV,EAAc,KAAK,EAAE,aACrB,EAAY,KAAK,EAAE,WACnB,EAAc,KAAK,EAAE,aAMvB,GAAI,IACA,CACE,cAAe,KACf,cAAa,EACb,YAAW,EACX,cAAa,EACb,aAAc,EACd,cAAa,EACb,WAAU,EACV,YAAW,EACX,YAAW,EACX,aAAY,GAEd,GAGJ,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,IAExC,EAAc,GAAG,YAAc,KAC/B,EAAc,GAAG,UAAY,KAAK,aAAa,OAAS,EACxD,EAAc,GAAG,YAAc,GAyBnC,EAAA,UAAA,UAAA,WACE,GAAM,GACyB,CAAC,KAAM,KAAK,KAAM,UAAW,KAAK,WACjE,MAAI,MAAK,iBAAmB,MAC1B,GAAO,gBAAqB,KAAK,iBAEnC,AAAI,KAAK,OAAS,MAChB,GAAO,MAAW,KAAK,OAElB,GAQC,EAAA,UAAA,eAAV,WACE,YAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,YAC/B,KAAK,QAAQ,QAGZ,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,UAAU,KAAK,KAAI,2BAkCvC,EAAA,UAAA,QAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,uCAIvC,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,sCAIvC,KAAK,oBAEL,GAAI,GAAuB,EAC3B,MAAI,EAAE,KAAK,YAAc,GACvB,GAAuB,KAAK,kBAGvB,CAAC,qBAAsB,KAAK,UAAW,qBAAoB,IAEtE,GAxoCoC,EAAA,cAAc,cAopClD,YAA2B,GAEzB,EACI,GAAqB,GAEzB,OADM,GAAkB,GACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAO,KAAK,EAAE,OAEhB,MAAO,IAA+B,GAYxC,YAA0B,GAExB,MAAO,sBAeL,EAAwB,EACxB,GAKF,GAJA,AAAI,IAAS,MAAS,GAAa,MAAQ,EAAY,IACrD,GAAQ,EAAO,YACf,EAAY,EAAO,WAEjB,EAAM,aAAa,SAAW,EAChC,MAAO,CAAC,GAER,GAAM,GAAO,EAAM,aAAa,GAChC,GAAI,EAAK,cAAc,SAAW,EAChC,MAAO,GAAK,aAGZ,OADM,GAAkC,GAC/B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,IAM7C,OALM,GAAI,EAAK,aAAa,GACtB,EAAQ,EAAK,cAAc,GAC3B,EAAY,EAAK,YAAY,GAC7B,EAAkB,GAAgB,EAAG,EAAO,GAElC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAc,QAAQ,KAAO,IAC/B,EAAc,KAAK,GAIzB,MAAO,GCvmDb,AA4CA,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CACJ,MAAO,EAAK,MACZ,KAAM,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAO,SAAS,cACtD,KAaF,GAXA,AAAI,EAAK,WAAa,MACpB,GAAK,UAAY,MAEnB,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,IAGhB,EAAK,UAAY,GACjB,EAAK,MAAQ,GACb,EAAK,OAAS,EAAK,OAEf,EAAK,YAAc,MAAQ,EAAK,iBAAmB,KACrD,KAAM,IAAI,GACN,qGAGN,GAAI,GAAkB,EAAK,gBAC3B,GAAI,GAAmB,MACrB,GAAI,EAAK,YAAc,KACrB,KAAM,IAAI,GACN,iFAGJ,EAAkB,CAAC,EAAK,WAAW,OAAO,EAAK,oBAI7C,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,yFAKR,GAAM,GAAQ,EAAK,OAAS,UAE5B,EAAK,gBAAkB,EACvB,EAAK,MAAQ,EAEb,EAAK,UAAY,CAAC,CAAC,MAAO,IAE1B,GAAM,GAAc,GAAI,IACpB,EAAK,MAAO,EAAK,gBAAiB,EAAM,GAAI,GAAI,EAAK,MACzD,SAAY,UAAY,EACxB,EAAY,YAAc,EAK1B,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,CAAC,GACf,cAAe,CAAC,GAChB,WAAY,CAAC,MACb,YAAa,CAAC,MACd,YAAa,CAAC,GACd,aAAc,CAAC,OAInB,SAAA,UAAA,MAAA,SACI,EACA,GACF,KAAM,IAAI,GACN,+BACA,kDAAiD,KAAK,QAG5D,EAAA,UAAA,QAAA,WAEE,MAAO,CAAC,qBAAsB,KAAK,UAAW,qBAAsB,IAGtE,EAAA,UAAA,UAAA,WACE,MAAO,CACL,gBAAiB,KAAK,gBACtB,MAAO,KAAK,MACZ,OAAQ,KAAK,OACb,KAAM,KAAK,OAzFC,EAAA,UAAY,aA4F9B,GA9FgC,IA+FhC,EAAA,cAAc,cAAc,gBAmCN,GACpB,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAC/C,KAAM,IAAI,OACN,gIAKN,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAE/C,KAAM,IAAI,GACN,oFAGN,GAAI,GAAa,EAAO,WACxB,AAAI,EAAO,OAAS,MAAQ,GAAc,MACxC,GAAa,CAAC,MAAM,OAAO,EAAO,QAGpC,GAAI,GAAQ,EAAO,MACnB,AAAI,GAAS,MACX,GAAQ,WAGV,GAAM,GAAa,GAAI,IAAW,CAChC,gBAAiB,EACjB,KAAM,EAAO,KACb,MAAK,EACL,OAAQ,EAAO,SAGX,EAAU,EAAW,aAAa,GAAG,cAC3C,MAAO,GAAQ,GC9MjB,YA0B2C,iHACzC,GAAI,GAAQ,KACV,MAAA,CAAA,GAEI,EAA+D,GAC/D,EAAiB,GACjB,EAA6B,GACnC,IAAW,IAAO,GACV,EAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACb,GAAc,EACpB,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAiB,KAAK,UAGtB,GAAS,OAAS,EACL,CAAA,EAAM,QAAQ,IAAI,IAD/B,CAAA,EAAA,UAEF,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAK,EAAK,IAAM,EAAO,GAAG,GAG5B,EAAA,QAAQ,+CAUyB,GACnC,GAAI,GAAQ,KACV,OAEF,OAAW,KAAO,IAChB,GAAM,GAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACnB,EAAM,WCjEZ,AAoBA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,QAAA,GAAA,YAFU,IAAA,IAAqB,KAM1B,GAAM,IAAyB,IA0BtC,GAAA,WAAA,aAEE,KAAA,eAAkC,KAMlC,SAAA,UAAA,UAAA,SAAU,GACR,KAAK,OAAS,GAGV,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,qFAEb,EAAA,UAAA,WAAN,SAAiB,qFASjB,EAAA,UAAA,SAAA,SAAS,KAGX,mBAqBE,WAAY,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,IAGtC,AAAI,GAAa,MACf,GAAY,IAEd,KAAK,UAAY,EACjB,KAAK,YAAc,EAGrB,SAAA,UAAA,OAAA,SAAO,GACL,KAAK,UAAU,KAAK,IAGtB,EAAA,UAAA,UAAA,SAAU,GACR,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,UAAU,KAIvB,EAAA,UAAA,SAAA,SAAS,GACP,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,SAAS,KAShB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,aAAN,SAAmB,uGACjB,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,KADO,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,WAAN,SAAiB,uGACf,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,KADS,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAI3B,KAOA,GAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,mBACE,GAAA,KAAA,OAAO,KAGH,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,OAAS,YAGV,EAAA,UAAA,WAAN,SAAiB,EAAe,yFAC9B,AAAI,GAAQ,MACV,GAAO,IAEH,EAAY,EAAK,MAAW,KAAO,EAAI,EAAK,KAClD,KAAK,MAAQ,aACF,GACT,GAAM,GAAQ,EAAK,GACnB,GAAI,MAAO,IAAU,SACnB,AAAK,EAAK,OAAO,eAAe,IAC9B,GAAK,OAAO,GAAO,GAErB,EAAK,OAAO,GAAO,EAAK,OAAO,GAAiB,EAAQ,OAExD,GAAI,GAAkB,OACtB,AAAI,IAAO,GAAK,OACd,EAAqB,EAAK,OAAO,GAEjC,EAAK,OAAO,GAAO,EAErB,GAAM,GACF,EAAA,KAAK,WAAM,MAAA,GAAA,IAAK,EAAK,OAAO,GAAO,EAAA,IAAI,EAAO,MAClD,EAAK,OAAO,GAAO,EACnB,AAAI,GAAsB,MACxB,EAAmB,mBAlBzB,IAAW,IAAO,KAAP,kBAwBP,EAAA,UAAA,WAAN,SAAiB,EAAe,2FAC9B,GAAI,GAAQ,KACV,eAAW,GACT,GAAI,EAAK,OAAO,IAAQ,sBAGxB,AAAI,MAAO,GAAK,OAAO,IAAS,SAC9B,EAAK,GAAO,EAAK,OAAO,GAAiB,EAAK,KAE9C,EAAA,KAAK,WACH,GAAM,GAAc,EAAA,IAAI,EAAA,IAAI,EAAG,EAAK,MAAO,EAAK,OAAO,IACvD,EAAK,GAAO,EACX,EAAK,OAAO,GAAgB,UAC7B,EAAA,KAAK,EAAK,cAXhB,EAAA,EAAkB,EAAA,KAAK,OAAO,QAAZ,EAAA,EAAA,OAAA,IAAP,EAAG,EAAA,KAAH,kBAiBjB,GA9DgC,mBAqEH,EAAA,EAAA,GAA7B,4DAIQ,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,MAAQ,GACb,KAAK,QAAU,YAGX,EAAA,UAAA,WAAN,SAAiB,EAAe,4EAC9B,AAAI,GAAQ,MACV,GAAO,IAET,KAAK,MAAM,KAAK,GAChB,IAAW,IAAO,GAChB,AAAI,KAAK,QAAQ,IAAQ,MACvB,MAAK,QAAQ,GAAO,IAEtB,KAAK,QAAQ,GAAK,KAAK,EAAK,mBAO1B,EAAA,UAAA,SAAN,6HACQ,EAA+D,GAC/D,EAAiB,GACjB,EAAoB,GAC1B,IAAW,IAAO,MAAK,QAErB,IADM,EAAa,KAAK,QAAQ,GACvB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,MAAO,GAAW,IAAO,UACrB,GAAc,EAAW,GAC/B,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAQ,KAAK,IAIJ,MAAA,CAAA,EAAM,QAAQ,IAAI,WACjC,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EAC7B,EAAkB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IACtD,EAAgB,UAChB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IAAM,EAAO,GAAG,mBAGpD,GA/C6B,mBA8DO,EAAA,EAAA,GAiBlC,WAAY,EAA0B,GAAtC,GAAA,GACE,EAAA,KAAA,OAAO,KAKP,GARM,EAAA,aAAe,EAIrB,EAAK,WAAa,GAAc,OAChC,AAAI,EAAK,aAAe,QACtB,GAAK,WAAa,IAEhB,EAAK,aAAe,SAAW,EAAK,SAAW,KACjD,KAAM,IAAI,OACN,mHAGN,MAAI,GAAA,KAAK,SAAS,EAAK,aAGrB,GAAK,UAAY,GACb,EAAK,UAAU,KAAK,GAAO,EAAK,aAEtC,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,MAAQ,EAAK,UAGd,SAAA,UAAA,UAAN,SAAgB,EAAe,EAAe,yGACtC,GAAgC,GAClC,KAAK,OAAS,KAChB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,MAAM,EAAO,EAAO,qBAEnC,SAAG,KAAK,EAAA,aACR,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAChC,MAAK,aAAe,EAChB,KAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,SACtB,EAAG,KAAK,EAAA,aAEV,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAC5B,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,QACtB,EAAG,KAAK,EAAA,aACH,AAAI,EAAA,KAAK,SAAS,KAAK,aAC5B,EAAG,KAAK,KAAK,UAAU,KAAK,aAAc,EAAO,IAEnD,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,mGACb,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,WAAtB,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,mGACX,MAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,SAAS,WAApB,EAAA,wCAGN,GA3GoC,gBAiHhC,EAEA,GAIF,GAHA,AAAI,GAAa,MACf,GAAY,IAEV,YAAqB,IACvB,MAAO,CAAC,GAEV,GAAI,MAAM,QAAQ,IAAc,EAAU,YAAc,IACtD,MAAO,GAGT,GAAM,GACF,GAAqB,GACzB,MAAO,GAAgB,IACnB,SAAA,GAAkB,MAAA,IAAI,IAAe,EAAgB,KAW3D,GAAA,IAAA,WAOE,cAcO,SAAA,4BAAP,SACI,EAAwB,GAC1B,EAAA,KAAK,OACD,GAAkB,GAAK,OAAO,UAAU,GACxC,WAAM,MAAA,sDACF,YAAW,KACnB,EAA4B,kBAAkB,GAC9C,AAAI,EAA4B,aAAa,IAAmB,MAC9D,GAA4B,aAAa,GAAkB,IAE7D,EAA4B,aAAa,GAAgB,KACrD,IAGS,EAAA,kBAAf,SAAiC,GAE/B,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAe,EAA4B,aAAa,CAAC,GAC/D,EAAa,QAAQ,SAAA,GACnB,GAAI,IAAS,EACX,KAAM,IAAI,GAAW,uCASZ,EAAA,MAAjB,WACE,EAA4B,aAAe,IAWtC,EAAA,gBAAP,SAAuB,GACrB,GAAM,GAA0C,GAChD,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAQ,CAAC,EACf,AAAI,GAAkB,GACpB,EAAa,KAAI,MAAjB,EAAqB,EAA4B,aAAa,IAGlE,MAAO,GAAa,IAAI,SAAA,GAAQ,MAAA,IAAI,MArEvB,EAAA,aAC6C,GAsE9D,iBAGI,EAA2B,EAAgC,EAC3D,EAAsB,EAAyB,EAC/C,EAAmB,EACnB,GACF,GAAM,GAAU,GAAI,IACd,EAAe,CACnB,GAAI,YAAiB,GAA4B,gBAAgB,IAEnE,AAAI,GAAa,MACf,EAAgB,KAAI,MAApB,EAAwB,GAE1B,EAAgB,KAAK,GACrB,GAAM,GAAe,GAAI,IAAa,GAMtC,SAAa,UAAU,CACrB,OAAM,EACN,aAAY,EACZ,QAAS,EACT,MAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,QAAS,IAEJ,CAAC,aAAY,EAAE,QAAO,GCnlB/B,YA2BI,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,QAAS,GChC9B,YAwB4B,EAAW,GACrC,MAAO,GAAA,KAAK,WACV,AAAI,EAAE,QAAU,WACd,GAAI,EAAE,OAAO,YAEf,GAAM,GAAY,EAAA,IAAQ,GAAS,GAAI,EAAM,IACvC,EAAgB,EAAA,KAAS,EAAU,MAAO,MAC1C,EAAO,EAAA,KAAS,EAAA,QAAY,EAAW,IAC7C,MAAO,GAAA,IAAQ,EAAG,iBAIW,EAAe,GAC9C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAO,IAAS,kBAG5B,EAAe,GAC/C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAO,IAAS,kBAIzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,EAAA,IAAQ,EAAO,GACtB,EACF,EAAA,YAAgB,EAAA,IAAQ,GAAQ,KAAW,OAAO,WAChD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAM,IACxC,MAAO,GAAA,IAAQ,IAAK,EAAA,KAAS,EAAW,mBAKxC,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAE9B,EAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAErC,MAAO,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAU,IAAa,kBAI/B,EAAe,GAC1C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,GAAS,GAAY,kBAInB,EAAe,GACnC,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,EAAW,kBAIE,EAAe,GAC9C,MAAO,GAAA,KAAK,WACV,GAAM,GAAM,EAAA,IAAQ,EAAA,IAAQ,EAAO,GAAQ,IACrC,EAAM,EAAA,IAAQ,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAQ,GAAQ,IACvD,MAAO,GAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAK,mBAY1B,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,KAAK,IAAI,GAChB,EAAiB,EAAA,IAAQ,EAAO,GAChC,EAAgB,EAAA,IAClB,EAAA,IAAQ,EAAgB,EAAA,SAAa,EAAA,IAAQ,GAAI,KACjD,GACJ,MAAO,GAAA,KAAS,EAAe,kBAK/B,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAI,EACF,EAAS,EAAA,QAAY,QAGrB,GAAM,GAAY,EAAA,IAAQ,EAAQ,EAAO,MAAM,OAAS,EAAG,IAC3D,EAAS,EAAA,IAAQ,EAAQ,GAE3B,SAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MACzC,EAAA,IAAQ,EAAA,IACX,EAAA,IAAQ,EAAO,UAAW,EAAA,IAAQ,IAAU,EAAO,MAAM,OAAS,kBActE,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAM,GAAa,EAAA,MAAU,GAAU,IAAS,QAChD,EAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MAChD,GAAM,GAAc,EAAO,MACrB,EACF,EAAA,OAAW,EAAY,EAAY,EAAY,OAAS,IACnD,QAAQ,GACjB,MAAO,IAAwB,EAAc,EAAQ,iBA0BrD,EAAgB,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAO,MAAO,EAAO,OACzC,KAAM,IAAI,GACN,8DACG,MAAK,UAAU,EAAO,OAAM,QAAQ,KAAK,UAAU,EAAO,SAEnE,MAAO,GAAA,KAAK,WAOV,GAAM,GAAa,EAAO,OACpB,EAAe,EAAO,MAAM,MAClC,MAAO,GAAW,IAAI,EAAO,IAAI,IAAS,IAAI,EAAa,MAAM,uBAIlC,EAAe,GAChD,MAAO,GAAA,KAAK,WACV,GAAI,GACJ,SAAI,EAAA,YAAgB,EAAO,KAAW,EAAI,MAC1C,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAG,KAC3B,EAAA,KAAS,GAA8B,EAAO,GAAI,kBAKzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,GAChD,EAAc,EAAA,YAAgB,EAAO,KAAW,GACtD,MAAO,GAAA,IACH,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAA,IAAQ,EAAa,KAAgB,kBAI5C,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,EAAA,IAAQ,EAAA,IAAQ,KAAW,IAC3C,MAAO,GAAA,KAAS,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAO,IAAW,kBAI7B,EAAe,GAC7C,MAAO,GAAA,KAAK,WACV,GAAM,GAAiB,GAAY,EAAO,IACpC,EAAiB,GAAY,EAAO,IACpC,EAAY,EAAA,IAAQ,EAAgB,GAC1C,MAAO,GAAA,IAAQ,EAAA,IAAQ,EAAW,OAkB/B,GAAM,IAAsD,CACjE,iBAAgB,GAChB,kBAAiB,GACjB,4BAA2B,GAC3B,4BAA2B,GAC3B,aAAY,GACZ,MAAK,GACL,iBAAgB,GAChB,QAAO,GACP,wBAAuB,GACvB,8BAA6B,GAC7B,mBAAkB,GAClB,0BAAyB,GACzB,QAAO,GACP,gBAAe,gBAKG,GAClB,GAAI,MAAO,IAAmB,UAC5B,GAAI,IAAkB,IACpB,MAAO,IAAU,GAEnB,GAAI,GAAS,gBAAgB,EAC7B,KAAI,GAAe,cAAc,SAAS,wBACxC,GAAS,gBAAgB,EAAc,wFAInC,GAAI,GAAW,OAErB,OAAO,GC7QX,YAyB+B,EAAe,GAC5C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,IAAQ,GAAI,EAAA,SAAa,IACrC,EAAmB,GAAO,EAAA,QAAY,EAAO,GAAY,EAAM,OACrE,MAAO,GAAA,KAAS,EAAA,MAAU,EAAO,GAAmB,kBAIpB,EAAe,GACjD,MAAO,GAAA,KACH,WAAM,MAAA,IACF,EAAA,MAAU,EAAA,OAAW,EAAO,IAAK,EAAA,OAAW,EAAO,KAAM,aAGnE,YAAuB,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,yBAI3C,EAAe,GACvC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIS,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIqB,EAAe,GAChD,MAAO,IAAuB,EAAO,eAInC,EAAe,GACjB,MAAI,GAAM,OAAS,EAAM,MACvB,GAAQ,EAAM,QAAQ,CAAC,EAAM,KAAO,KAEtC,EAAQ,EAAM,OAAO,IACrB,AAAI,EAAM,QAAU,EAAM,OACxB,GAAQ,EAAM,OAAO,EAAM,QAEtB,EAAA,MAAU,EAAO,GAAO,OAAO,WAajC,GAAM,IAAM,GACN,GAAM,GACN,GAAM,GACN,GAAM,GACN,GAAO,GACP,GAAO,GACP,GAA0B,GAC1B,GAAS,GACT,GAAgC,GAIhC,GAAuD,CAClE,eAAc,GACd,oBAAmB,GACnB,UAAS,GACT,wBAAuB,GACvB,8BAA6B,GAC7B,IAAG,GACH,IAAG,GACH,IAAG,GACH,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,OAAM,gBAGY,GAClB,GAAI,MAAO,IAAe,UAAY,IAAc,IAClD,MAAO,IAAW,GACb,GAAI,MAAO,IAAe,UAAY,GAAc,KACzD,MAAO,GAEP,KAAM,IAAI,GAAW,kBAAkB,eAqBP,GAElC,GADA,GAAY,IAAO,KAAM,0BAA0B,GAC/C,MAAO,IAAO,SAChB,MAAO,GAGP,OADI,GAAM,OACQ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAU,KAAS,GACrB,EAAS,EACT,OAGJ,GAAI,IAAW,OACb,MAAO,GAET,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAW,KAAS,GACtB,EAAS,EACT,OAGJ,MAAI,KAAW,OACN,EAED,EAAgB,KCzL5B,YAwB6B,GAC3B,GAAM,GAA2D,CAC/D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,MAC/B,SAAY,WAAM,MAAA,GAAA,MAAM,SAAS,EAAG,IAAM,OAC1C,KAAQ,WAAM,MAAA,GAAA,MAAM,KAAK,KAAO,GAAK,KAAO,OAC5C,OAAU,WAAM,MAAA,GAAA,MAAM,OAAO,KAAO,GAAK,KAAO,KAAW,IAC3D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,KAAO,GAAK,EAAG,OAC9C,IAAO,WAAM,MAAA,GAAA,MAAM,IAAI,OASzB,GAPA,EAAa,QAAa,EAAa,QACvC,EAAa,SAAc,EAAa,SACxC,EAAa,KAAU,EAAa,KACpC,EAAa,OAAY,EAAa,OACtC,EAAa,QAAa,EAAa,QACvC,EAAa,IAAS,EAAa,IAE/B,IAAc,GAChB,MAAO,GAAa,KAEtB,KAAM,IAAI,GAAW,qBAAqB,GC3C5C,AAeO,GAAM,IAA8C,EAAI,KAAO,iBAelE,EAAyB,EAAmB,GAC9C,GAD8C,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAuB,MACvB,MAAO,IAAwB,UAC/B,OAAO,eAAe,KAAyB,OAAO,WACtD,CAAC,GAAiB,GACpB,KAAM,IAAI,OACN,sEAGN,GAAI,GACF,GAAM,GAAM,KAAK,UAAU,GAC3B,AAAI,EAAI,OAAS,IACf,QAAQ,KACJ,mCAAmC,EAAS,qBAC5C,iBAAgB,EAAI,OAAM,iCAC1B,kHAEG,IAA2C,mBAgBvB,GAC/B,GAAI,IAAM,KAER,MAAO,GACF,GAAI,MAAO,IAAM,SACtB,GAAI,OAAO,eAAe,KAAO,OAAO,WAGtC,OADM,GAAO,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,MAAO,IAAQ,SAEjB,MAAO,GAET,GAAI,CAAC,GAAiB,EAAE,IACtB,MAAO,GAGX,MAAO,WAGH,MAAM,QAAQ,IAEhB,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,GAAiB,GACpB,MAAO,GAGX,MAAO,OAKP,OAAO,QAKX,GAAM,GAAQ,MAAO,GACrB,MAAO,KAAU,UAAY,IAAU,UAAY,IAAU,WCrGjE,YA8BI,EAAkB,EAAqB,EAEvC,GAAA,AAAA,IAAA,QAAA,GACI,QAAQ,KACd,GAAM,GAAiB,GAAsB,GAGvC,EAAsB,CAAC,eAAgB,eAAgB,WAC7D,AAAI,EACF,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAEtC,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAAM,IAI9C,AAAI,EAAU,EAAU,OAAS,IAAM,GAErC,GAAY,EAAU,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAa,MAGzD,GAAI,GACJ,GAAI,CAAC,GACH,EAAU,KAAK,mBACf,EAAgB,GAChB,OAAW,KAAS,GAAM,aACxB,EAAc,KAAI,MAAlB,EAAsB,EAAM,aAAa,IAI7C,EAAQ,IAAI,OAAO,IACnB,GAAS,EAAW,EAAW,GAC/B,EAAQ,IAAI,OAAO,IAGnB,OADM,GAAS,EAAM,OACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EACF,GAAkB,EAAO,GAAI,EAAW,GAExC,GACI,EAAO,GAAI,EAAW,EAAe,GAE3C,EAAS,KAAM,EAAO,OAAS,EAAI,IAAM,KAAK,OAAO,IAItD,EAAc,mCAEf,GAAM,GAAiB,GAAqB,GACtC,EAAoB,GAAqB,EAAM,qBAErD,EAAQ,iBAAiB,GAAiB,IAC1C,EAAQ,qBAAqB,GAC7B,EAAQ,yBAAyB,GACjC,EAAQ,IAAI,OAAO,IAGrB,YAA8B,GAC5B,GAAI,GAEJ,MAAK,GAAc,2BAA6B,KAC9C,EACI,GAAsB,EAAc,2BAExC,EAAiB,GAAqB,EAAM,kBAGvC,EAGT,YAA+B,GAC7B,GAAI,GAAiB,GACf,EAAyB,GACzB,EAAgB,GACtB,OAAW,KAAS,GAAM,aACxB,EAAa,KAAK,EAAM,aAAa,IAEvC,OAAyB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,EAAW,OAAS,GACpB,EAAW,SAAW,GAAK,EAAW,GAAG,cAAc,OAAS,GAClE,EAAiB,GACjB,MAEF,EAAM,KAAI,MAAV,EAAc,GAEhB,GAAI,EAEF,OAAoB,GAAA,EAAA,EAAA,EAAM,OAAN,EAAA,EAAA,OAAA,KAElB,OAFS,GAAK,EAAA,GACV,EAAO,GACQ,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAM,QAAQ,KAAU,GAC1B,GAAI,GACF,EAAiB,GACjB,UAEA,GAAO,GAIb,GAAI,CAAC,EACH,MAIN,MAAO,GAGT,YACI,EAAkB,EAElB,GAAA,AAAA,IAAA,QAAA,GAA6D,QAAQ,KAEvE,OADI,GAAO,GACF,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAI,GACN,GAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAAK,KAE1C,GAAQ,EAAO,GACf,EAAO,EAAK,MAAM,EAAG,EAAU,IAC/B,GAAQ,IAAI,OAAO,EAAU,GAAK,EAAK,QAEzC,EAAQ,GAQV,YACI,EAAc,EAEd,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAGhB,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EACF,CAAI,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,YAChE,GAAS,EAAQ,EAAW,GAM9B,YACI,EAAc,EAAqB,EAEnC,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAIhB,OADM,GAAwB,GACX,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,GAAiB,MAAQ,EAAc,OAAS,GAChD,EAAc,QAAQ,KAAU,GAClC,SAEF,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAAG,KACrC,EAAoB,EAAK,YAAY,GACrC,EAAqB,EAAK,cAAc,GAC9C,EAAY,KACL,EAAY,IAAI,EAAiB,KAAK,EAAkB,MAGnE,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EAAkB,EAAY,SAAW,EAAI,GAAK,EAAY,GAC9D,EAAmB,CACpB,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,WAC3D,GAGF,GAAS,EAAQ,EAAW,GAC5B,OAAS,GAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,EACxC,GAAS,CAAC,GAAI,GAAI,GAAI,EAAY,IAAK,EAAW,GCvNtD,AA8BA,YACI,EAAa,EAAe,GAC9B,MAAQ,KAAQ,gBAAkB,IAAQ,gBAClC,IAAQ,gBACZ,IAAU,GAAK,MAAO,IAAU,qBAUlC,EAA6B,GAC/B,GAAI,IAAmB,KACrB,MAAO,MACF,GAAI,MAAO,IAAmB,SACnC,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAmB,UAC1B,MAAO,IAAmB,UAC7B,MAAO,GACF,GAAI,YAA0B,QAGnC,OAFM,GAAU,GACV,EAAc,EAAe,OAC1B,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAe,GAC5B,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACf,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAArB,GAAM,GAAW,EAAA,GACd,EAAgB,EAAe,GACrC,GAAI,IAAgB,QAAU,MAAO,IAAkB,SAIrD,EAAO,GAAe,OAEtB,GAAM,GAAQ,GAA0B,GACxC,EAAO,GAAS,GAAoB,EAAe,IAGvD,MAAO,gBAWP,EAAyC,GAC3C,GAAI,GAAa,KACf,MAAO,MACF,GAAI,MAAO,IAAa,SAC7B,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAa,UAAc,MAAO,IAAa,UACzD,MAAO,GACF,GAAI,YAAoB,QAG7B,OAFM,GAAU,GACV,EAAc,EAAS,OACpB,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAS,GACtB,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACrB,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAU,EAAS,GACnB,EAAQ,GAA0B,GACxC,AAAK,KAAU,QAAU,IAAU,cAC/B,MAAO,IAAY,SAIrB,EAAO,GAAS,EAEhB,EAAO,GAAS,GAAoB,EAAS,GAGjD,MAAO,IC7HX,GAGM,IAAU,QCHhB,AA0BA,YAAiC,EAAqB,GAEpD,GAAI,EAAI,OAAS,MAAQ,EAAI,QAAU,EAAI,MAEzC,MAAO,GAET,IAEE,MAAO,GAAA,KAAK,EAAK,EAAI,aACd,GAEP,KAAM,IAAI,GACN,0BAA0B,EAAI,MAAK,kCACnC,gBAAe,EAAI,KAAI,MAAM,EAAI,MAAK,QAgB9C,GAAA,IAAA,WAUE,WAAY,GACV,GAVM,KAAA,SAAmC,GACnC,KAAA,QAAkC,GAClC,KAAA,QAAoC,GAQtC,YAAiB,GACnB,OAAW,KAAM,GAAM,SACrB,KAAK,SAAS,GAAM,EAAM,SAAS,GACnC,AAAI,IAAM,GAAM,SACd,MAAK,QAAQ,GAAM,EAAM,QAAQ,SAIrC,GAAI,GAAS,KACX,OAEF,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,KAAK,IAAI,EAAK,IAAK,EAAK,SAe9B,SAAA,UAAA,IAAA,SAAI,EAAqB,EAAe,GACtC,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAK,SAAS,EAAI,IAAM,GAAwB,EAAK,GACrD,KAAK,QAAQ,EAAI,MAAQ,EAAI,GAC7B,AAAI,GAAQ,MACV,MAAK,QAAQ,EAAI,IAAM,OAGzB,MAAM,IAAI,GAAW,uBAAuB,EAAI,KAAI,QAAQ,EAAI,IAElE,MAAO,OAQT,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,IAAI,EAAK,IAAK,EAAK,QAO1B,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,MAAK,SAAS,EAAI,KAAO,MAMlC,EAAA,UAAA,MAAA,WACE,MAAO,QAAO,KAAK,KAAK,UAU1B,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,SAAS,EAAI,SAG3B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,SAAS,KAWzB,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,QAAQ,EAAI,SAG1B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,QAAQ,KAKxB,EAAA,UAAA,aAAA,WACE,AAAI,KAAK,SAAW,MAClB,EAAA,QAAQ,KAAK,UAGnB,KAIM,GAA+D,GAG/D,GAC4D,eA4C9D,EAA0C,EAC1C,EAAiB,GAWnB,OATM,GAAoB,GAAU,KAAO,GAAQ,EAAO,SAEpD,EAAe,MAAM,QAAQ,GAC7B,EACF,EAAe,EAA8B,CAAC,GAE5C,EAAc,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,OACpC,EAAyB,GACzB,EAAY,EAAS,QACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,AAAI,EAAU,QAAQ,KAAgB,GACpC,EAAa,KAAK,EAAS,SAAS,IAEpC,EAAa,KAAK,MAItB,AAAI,GAAS,MAEX,GAAM,cAAgB,UACtB,EAAM,cAAgB,UAIxB,GAAM,GACF,EAAY,KAAK,KAAO,IAAM,EAAS,QAAQ,KAAK,KACpD,EACA,EACJ,GAAI,GAAa,IAAoB,MAGnC,GAAM,GAAM,GAAqC,EAAY,GAC7D,EAAS,EAAI,OACb,EAAkB,EAAI,gBAGtB,GAAa,GAAmB,EAChC,GAAsB,GAAmB,EAE3C,EAAS,GAAa,GACtB,EAAkB,GAClB,AAAK,GACH,OAAO,OAAO,EAAiB,GAAsB,IAMvD,OAHM,GAAmB,GAAI,IAAS,GAG7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAI,GAAS,MAEX,GAAM,GAAa,EAAA,SAAS,WAC5B,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAExB,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAI1B,GAAM,GAAW,EAAO,GAClB,EAAW,EAAS,YAC1B,GAAI,YAAoB,IACtB,SAOF,OALM,GAAwB,GACxB,EAAuB,GACvB,EAA6B,GAE/B,EAAa,GACG,EAAA,EAAA,EAAA,EAAS,OAAT,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAiB,SAAS,GAClC,EAAO,EAAiB,QAAQ,GACtC,EAAY,KAAK,GACjB,EAAW,KAAK,GAChB,AAAI,GAAQ,MACV,GAAa,IAEf,AAAK,GACH,GAAgB,EAAM,QACtB,AAAI,EAAgB,EAAM,QAAU,GAAK,CAAC,EAAS,OAAO,IACtD,EAAY,QAAQ,EAAM,QAAU,IAAM,CAAC,EAAM,YACjD,EAAM,YAAY,WAAa,IACjC,EAAiB,KAAK,IAK5B,AAAI,GACF,GAAS,GAAU,GACnB,EAAO,KAAU,EAAW,IAE9B,GAAM,GACF,GAAO,EAAS,MAAM,EAAa,IACnC,EAA8B,KAClC,AAAI,EAAS,iBACX,GAAa,EAAS,YAAY,EAAa,IAKjD,OAHM,GAAe,GAAe,GAC9B,EACF,MAAM,QAAQ,GAAgB,EAAe,CAAC,GACzC,EAAI,EAAG,EAAI,EAAsB,OAAQ,EAAE,GAClD,AAAK,EAAiB,OAAO,EAAsB,KACjD,EAAiB,IACb,EAAsB,GAAI,EAAc,GACxC,MAAM,QAAQ,GAAc,EAAW,GAAK,GAElD,GAAM,GAAQ,EAAY,QAAQ,EAAsB,GAAG,MAC3D,AAAI,IAAU,IACZ,GAAa,GAAS,EAAc,IAIxC,AAAK,GAEH,EAAA,QAAQ,GAQZ,SAAiB,eAEV,EAAe,EAAe,EAAa,GAsBpD,YACI,EAA2B,GAE7B,EAAA,KAAK,OACD,GAAW,MAAQ,EAAQ,OAAS,EACpC,WAAM,MAAA,0CAEV,GAAI,GAAgC,GAChC,EAAkC,GACtC,GAAI,EAAQ,SAAW,GAErB,GAAM,GACF,GAAgD,EAAQ,GAAI,GAChE,EAAc,EAAI,OAClB,EAAoB,EAAI,iBAGxB,QADM,GAAU,GAAI,KACA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAKlB,OALS,GAAK,EAAA,GACR,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,aAIc,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAxB,GAAM,GAAc,EAAA,GACvB,AAAK,EAAQ,IAAI,EAAe,OAC9B,GAAY,KAAK,GACjB,EAAQ,IAAI,EAAe,sBAKpB,GACT,AAAI,EAAkB,IAAS,MAC7B,GAAkB,GAAQ,GAAI,MAEhC,EAAa,GAAM,QACf,SAAA,GAAa,MAAA,GAAkB,GAAM,IAAI,MAL/C,OAAW,KAAQ,KAAR,GASf,MAAO,CACL,OAAQ,EACR,gBAAiB,GAAoB,IAIzC,YAA6B,GAC3B,GAAM,GAAmC,GACzC,OAAW,KAAQ,GACjB,EAAgB,GAAQ,EAAa,GAAM,KAE7C,MAAO,eAeL,EAAuB,GASzB,OAPM,GAAU,GAAI,KACd,EAA2B,GAC3B,EAA6B,GAKjB,EAAA,EAAA,EAAA,EAAS,QAAT,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAQ,IAAI,GAGd,GAAM,GAA0B,GAC1B,EAAkB,GAKxB,IAFA,EAAM,KAAK,GAEJ,EAAM,OAAS,IACpB,GAAM,GAAM,EAAM,EAAM,OAAS,GACjC,GAAI,EAAQ,IAAI,EAAI,OAClB,EAAM,MACN,SAEF,GAAM,GAAc,EAAM,EAAM,OAAS,KAAO,EAAM,OAAS,EAC/D,GAAI,EAAI,OAAO,SAAW,GAAK,EAE7B,EAAM,MACN,EAAO,KAAK,GACZ,EAAQ,IAAI,EAAI,MAChB,AAAI,GACF,EAAM,WAKR,EAAM,KAAK,EAAM,OAAS,GAC1B,OAAoB,GAAA,EAAA,EAAA,EAAI,OAAJ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GAQd,GALA,AAAI,EAAa,EAAM,OAAS,MAC9B,GAAa,EAAM,MAAQ,GAAI,MAEjC,EAAa,EAAM,MAAM,IAAI,EAAI,MAE7B,EAAQ,IAAI,EAAM,MACpB,SAEF,EAAM,KAAK,KAIjB,MAAO,CAAC,OAAM,EAAE,aAAY,GAS9B,YAAwB,GAEtB,GAAI,GACJ,GAAI,EAAM,YAAY,aAAa,SAAW,EAC5C,EAAe,EAAM,YAAY,YAGjC,OADI,GAAoB,KACf,EAAI,EAAG,EAAI,EAAM,YAAY,aAAa,OAAQ,EAAE,EAC3D,OAA2B,GAAA,EAAA,EAAA,EAAM,YAAY,aAAa,GAChD,cADiB,EAAA,EAAA,OAAA,KAAtB,GAAM,GAAY,EAAA,GAErB,GAAI,EAAa,KAAO,EAAM,IAC5B,EAAY,EACZ,OAIN,EAAe,EAAM,YAAY,YAAY,GAE/C,MAAO,GCjhBT,AA6CA,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAoCtC,WAAY,GAAZ,GAAA,GAEE,EAAA,KAAA,KAAM,KAAG,KAET,GAtBF,EAAA,eAAiB,GAAI,KAqBnB,EAAK,KAAO,EAAK,KACb,EAAK,MAAQ,MACf,GAAM,GAAS,EAAK,eAAe,cACnC,EAAK,KAAO,GAAO,GAqBrB,GAlBA,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAKlB,AAAI,MAAM,QAAQ,EAAK,QACrB,EAAK,OAAS,EAAK,OAAO,QAE1B,EAAK,OAAS,CAAC,EAAK,QAEtB,AAAI,MAAM,QAAQ,EAAK,SACrB,EAAK,QAAU,EAAK,QAAQ,QAE5B,EAAK,QAAU,CAAC,EAAK,SAInB,GAAqB,EAAK,QAAQ,SAAW,EAAK,OAAO,OAC3D,KAAM,IAAI,GACN,mGAEA,IAAG,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,SAIhC,AAAI,GAAqB,EAAK,SAAS,SAAW,EAAK,QAAQ,QAC7D,QAAQ,KACJ,qGAEA,IAAG,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,SAOjC,EAAK,YAAc,GACnB,EAAK,uBAAyB,GAC9B,EAAK,yBAA2B,GAKhC,EAAK,aAAe,GACpB,EAAK,wBAA0B,GAC/B,EAAK,0BAA4B,GAKjC,EAAK,OAAS,GAMd,EAAK,sBAAwB,GAe7B,OAAgB,GAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YACtB,EAAK,aAAa,KAAK,GACvB,EAAK,wBAAwB,KAAK,GAClC,EAAK,0BAA0B,KAAK,GAMtC,OAAgB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YAKtB,GAAqB,IAAc,EAAG,4BACtC,GAAqB,IAAgB,EAAG,8BACxC,EAAK,YAAY,KAAK,GACtB,EAAK,uBAAuB,KAAK,GACjC,EAAK,yBAAyB,KAAK,GAIrC,EAAK,WAAa,GAClB,EAAK,YAAc,GACnB,EAAK,gBAAkB,GACvB,EAAK,eAAiB,GACtB,EAAK,gBAAkB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,EAAK,YAAY,GAE/B,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,WACN,6DACA,qBAAoB,EAAK,OAAM,MAC/B,UAAS,EAAC,0BACV,oBAAmB,EAAM,eAAc,MAE7C,EAAK,WAAW,KAAK,EAAM,MAC3B,EAAK,gBAAgB,KAAK,EAAM,iBAEhC,EAAK,eAAe,KAAK,EAAM,MAEjC,OAAoB,GAAA,EAAA,EAAA,EAAK,aAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,YAAY,KAAK,EAAM,MAG9B,EAAK,oBAAsB,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QAClD,EAAK,qBAAuB,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,QAwFpD,OAjFM,GAA0C,GAE1C,EAAyC,GACzC,EAA4C,GAE5C,EAA6C,GAC7C,EAA4C,GAC5C,EAAiC,GAoBjC,EACF,SAAC,GAAwB,GAAuB,GAC/C,GAAe,GAAoB,IAClC,AAAI,KAAS,MAAQ,IAAa,MAAQ,IAAe,OACvD,IAAQ,GAAO,YACf,GAAY,GAAO,UACnB,GAAc,GAAO,aAEvB,GAAM,IAAO,GAAM,aAAa,IAGhC,GAAI,GAAgB,QAAQ,MAAU,GACpC,KAAM,IAAI,IACN,cAAc,GAAO,KAAI,cAAc,GAAM,KAAI,yBAKvD,GAAI,GAAc,QAAQ,MAAU,GAClC,OAIF,EAAK,eAAe,IAAI,EAAU,QAAQ,GAAO,KAGjD,AAAM,GAAM,KAAM,IAChB,GAAa,GAAM,IAAM,OAAO,KAAK,GAAc,QAGrD,AAAI,GAAgB,QAAQ,MAAU,IACpC,GAAgB,KAAK,IAKvB,OADM,IAAmB,GAAK,cAAc,OACnC,GAAI,EAAG,GAAI,GAAkB,MACpC,GAAM,IAAI,GAAK,aAAa,IACtB,GAAQ,GAAK,cAAc,IAC3B,GAAY,GAAK,YAAY,IAC7B,GAAc,GAAK,cAAc,IACvC,EACI,GAAG,GAAe,GAAiB,GAAO,GAC1C,IAGN,IADA,GAAc,KAAK,IACZ,GAAgB,QAAQ,KAAS,GACtC,GAAgB,OAAO,GAAgB,QAAQ,IAAO,GAExD,EAAuB,KAAK,KAG5B,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAgB,EAAG,EAAe,GAKpC,OAFM,GACF,EAAuB,QAAQ,UAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAa,EAAK,IAAM,EAExB,AAAM,EAAK,KAAM,IACf,GAAY,EAAK,IAAM,GAEzB,GAAI,GAAQ,EAAY,EAAK,IAGvB,EACD,EAAa,EAAK,cAAc,KAAO,KACnC,EACA,EAAa,EAAK,cAAc,IAOzC,EAAQ,KAAK,IAAI,EAAO,GACxB,EAAa,EAAK,cAAc,IAAM,EACtC,EAAe,EAAK,cAAc,IAAM,EAAK,cAC7C,EAAY,EAAK,IAAM,EAGvB,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAa,aAAa,GACxC,EACD,EAAY,EAAY,KAAO,KAAO,EACA,EAAY,EAAY,IACnE,EAAY,EAAY,IAAM,KAAK,IAAI,EAAQ,EAAG,GAClD,EAAa,EAAY,IAAM,GAKnC,GAAM,GAA0C,GAChD,OAAW,KAAU,IACnB,GAAM,GAAQ,EAAY,GAC1B,AAAM,IAAS,IACb,GAAa,GAAS,IAExB,EAAa,GAAO,KAAK,EAAa,IAIxC,GAAM,IAA4C,GAClD,OAAW,MAAW,IACpB,GAAM,GAAQ,EAAa,IAC3B,AAAM,IAAS,KACb,IAAc,GAAS,IAEzB,GAAc,GAAO,KAAK,EAAe,KAI3C,GAAI,IAAY,OAAO,KAAK,IACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAG1B,EAAK,OAAS,GACd,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACR,GAAiB,GAAc,GAGrC,GAAe,KAAK,SAAC,GAAG,IACtB,GAAM,IAAS,EAAa,GAAE,IACxB,GAAS,EAAa,GAAE,IAC9B,MAAI,IAAS,GACJ,GAEL,GAAS,GACJ,EAEF,IAET,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACd,AAAI,YAAiB,IACnB,EAAK,sBAAsB,KAAK,GAElC,EAAK,OAAO,KAAK,IAGrB,EAAK,cAAgB,GAGrB,GAAY,OAAO,KAAK,GACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAStB,OAJM,IAAoB,EAAK,OAAO,QAGhC,GAAoC,GACtB,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,KAClB,OADS,GAAK,GAAA,IACK,GAAA,EAAA,GAAA,EAAa,GAAb,GAAA,GAAA,OAAA,MAAd,GAAM,GAAI,GAAA,IACP,EAAQ,EAAK,cACnB,GAAI,GAAS,MACX,OAAgB,IAAA,EAAA,GAAA,EAAK,aAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAI,GAAkB,QAAQ,KAAO,GACnC,KAAM,IAAI,IACN,sDAAsD,EACtD,eAAc,EAAM,KAAI,OACxB,uDACA,WAAU,KAGlB,OAAgB,IAAA,EAAA,GAAA,EAAK,cAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAkB,KAAK,GAEzB,GAAwB,KAAK,EAAM,OAMzC,EAAK,aAAe,EAKpB,OADM,IAAW,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,mBAC7B,IACT,GAAM,IAAiB,GAAS,OAAO,SAAA,IAAK,MAAA,MAAM,KAAM,OACxD,GAAI,KAAmB,EACrB,KAAM,IAAI,IACN,aAAa,GAAI,aAAa,GAAc,uEAE5C,KAAK,UAAU,MANJ,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAd,GAAM,IAAI,GAAA,OAAJ,IAcX,SAAK,cAAgB,GAErB,EAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,EAAK,OACnB,cAAe,EAAK,QACpB,WAAY,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,QACjC,YAAa,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,QACnC,YAAa,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QACpC,aAAc,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,UAExC,EAAK,MAAQ,GACb,EAAK,UAAY,IAGT,SAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,cAAc,KAAK,KAAI,2BA8B3C,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,GAAM,GACc,CAAC,qBAAsB,KAAM,qBAAsB,GACvE,GAAI,EAAE,KAAK,YAAc,GACvB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAO,sBAAwB,EAAM,UAAU,qBAKjD,OAAwB,GAAA,EAAA,EAAA,KAAK,sBAAL,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAO,sBAAwB,EAAU,UAAU,sBAGvD,SAAO,qBAAuB,KAAK,UAC5B,GAGT,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,OAAO,QAAQ,SAAA,GAEhB,EAAc,kBACX,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,MAElC,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WAIE,GAAI,KAAK,kBAAkB,OAAS,EAClC,KAAM,IAAI,GACN,wNAMN,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADI,GAA2B,GACX,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAU,EAAQ,OAAO,EAAM,kBAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACb,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAQ,KAAI,MAAZ,EAAgB,EAAM,qBAExB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACtB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAiB,KAAI,MAArB,EAAyB,EAAM,kBAEjC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAkB3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,IAGnC,OAFM,GAAgD,GAClD,EAAoB,EACJ,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACO,EAAA,EAAA,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,EAAa,EAAO,eAAiB,KACvC,KAAM,IAAI,GAAW,0BAA0B,EAAO,cAExD,EAAa,EAAO,cAAgB,EACpC,IAIJ,GAAM,GAAoD,GAC1D,OAAW,KAAQ,IAIjB,GAAI,GAAgB,EACpB,GAAI,EAAa,IAAS,MACxB,GAAM,GAAS,EAAK,MAAM,KACpB,EACF,EAAO,MAAM,EAAG,IAAI,OAAO,CAAC,EAAO,EAAO,OAAS,KACvD,EAAgB,EAAiB,KAAK,KAExC,GAAI,EAAa,IAAkB,KACjC,EAAkB,KAAK,CAAC,EAAa,GAAgB,EAAQ,aACpD,EACT,KAAM,IAAI,GACN,gDAAgD,GAEtD,MAAO,GAAa,GAGtB,GAAI,GAEF,GAAM,GAAuB,GAC7B,OAAW,KAAQ,GACjB,EAAW,KAAK,GAElB,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,GACH,EAAW,OAAM,OAChB,EAAiB,yBACrB,IAAG,IAIX,GAAc,IAON,EAAA,UAAA,cAAV,WACE,GAAM,GAAY,KAAK,YACjB,EAAwC,GAC9C,SAAY,UAAe,KAAK,eAChC,EAAY,OAAY,EACxB,EAAY,aAAkB,eAAe,GAG7C,EAAY,QAAa,gBAClB,GAeT,EAAA,UAAA,OAAA,SAAO,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IACnB,GAAM,GAAc,GAAoB,KAAK,iBAC7C,MAAO,GAAe,KAAK,UAAU,GAAe,GAgBtD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAE9B,OADM,GAAW,GAAI,IACZ,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,EAAK,OAAO,GAAI,EAAO,IAEtC,MAAO,IAAQ,EAAK,QAAS,EAAU,MAa3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAC9B,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAA2B,KAAM,EAAO,QAEhD,EAAQ,GAAqB,GAGxB,EAAK,iBAAiB,EAAQ,GAAO,MAahD,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAM,GAAc,GAA+B,GACnD,GAAI,EAAY,SAAW,KAAK,YAAY,OAC1C,KAAM,IAAI,GACN,+BAA+B,EAAU,KACzC,cAAa,KAAK,YAAY,OAAM,oBAK1C,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,EAAY,OAAQ,KACtC,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAa,EAAY,GAGzB,EAAW,EAAM,KAAO,OAC9B,EAAqB,GAAY,EAGnC,GAAM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,GAAK,MAAA,UAAS,EAAG,MACrB,KAAK,IAE5B,GAAI,EAAU,OAAS,EACrB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GAEP,EAAQ,EAAK,cACnB,GAAI,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,QAAQ,EAAM,MAAQ,GAExD,SAIF,OADM,GAAuB,GACpB,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAc,EAAa,KAAI,IAAI,EAAS,IAAI,EAChD,EAAa,EAAqB,GACxC,EAAY,KAAK,GAQnB,OALM,GAAc,EAAM,mBACtB,GAA+B,IAE7B,EAAe,GAA+B,GAC9C,EAAY,EAAM,aAAa,QAAQ,GACpC,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAqB,GAAY,EAAa,IAStD,OAFM,GAAwB,GACxB,EAA4B,GACzB,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GACzC,EAAc,KAAK,0BAA0B,GAC7C,EAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAgB,KAAK,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAgB,OAAQ,KAC1C,GAAM,GAAM,EAAgB,GAC5B,GAAqB,IAAO,IAC5B,EAAa,KAAK,EAAqB,IAIzC,MAAO,IAA+B,IAa9B,EAAA,UAAA,iBAAV,SAA2B,EAAkB,GAE3C,AAAI,GAAS,MACX,GAAQ,GAA2B,KAAM,EAAO,SASlD,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,GACxC,GAAM,GAAI,KAAK,OAAO,GAChB,EAAI,EAAO,GACX,EAAO,EAAM,GACnB,EAAU,EAAE,IAAM,CAAC,EAAG,GAMxB,OAHM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAUjB,OAVS,GAAI,EAAA,GAEP,EAAQ,EAAK,cACb,EAAwB,EAAK,aAC7B,EAAyB,EAAK,cAK9B,EAAe,GAAI,OACT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAE,KAAM,IACV,EAAa,KAAK,EAAU,EAAE,KAGlC,GAAI,EAAa,SAAW,EAAsB,QAEhD,GAAI,GAAiB,GACjB,EAAe,OACf,EAAa,OACb,EAAa,OACb,EAAW,OAKf,GAHA,AAAI,EAAK,UAAY,MACnB,GAAS,EAAK,UAEZ,EAAa,SAAW,GACpB,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GACvB,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAgB,IACpD,EAAc,GACV,EAAM,YAAY,EAAgB,IACtC,EAAkB,CAAC,GACnB,EAAgB,CAAC,OAEjB,GAAkB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KAC1C,EAAgB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KACxC,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAiB,IACrD,EAAc,GACV,EAAM,YAAY,EAAiB,IAGzC,GAAI,EAAM,oBACR,KAAM,IAAI,IACN,yHAMN,OAAS,GAAI,EAAG,EAAI,EAAuB,OAAQ,EAAE,GACnD,GAAM,GAAI,EAAuB,GAC3B,EAAI,EAAc,GAClB,EAAO,EAAY,GACzB,EAAU,EAAE,IAAM,CAAC,EAAG,KAS9B,OAHM,GAA0B,GAC1B,EAAwB,GACxB,EAAwB,GACd,EAAA,EAAA,EAAA,KAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,GACI,EAAE,KAAM,GAAW,4BAA4B,EAAE,KAAI,MAAM,EAAE,IAC3D,GAAA,GAAA,EAAA,EAAA,IAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAa,KAAK,EAAO,OACzB,EAAc,KAAK,GACnB,EAAY,KAAK,GAInB,MAAO,CAAC,EAAe,EAAa,IAW9B,EAAA,UAAA,uBAAR,SAA+B,GAG7B,OAFM,GAAiD,GACnD,EACgB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,YAAiB,GAAY,EAAI,EAC7C,OAAS,GAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,KAAK,eAAe,IAAI,IAE1B,GAAkB,GAAW,EAC7B,GAAa,IAInB,MAAO,IAsBT,EAAA,UAAA,SAAA,SAAS,EAAe,GACtB,GAAI,GAAS,MACX,GAAI,KAAK,OAAO,QAAU,EACxB,KAAM,IAAI,GACN,wCAAwC,EAAK,oBAC7C,QAAO,KAAK,OAAO,OAAM,eAE7B,MAAO,MAAK,OAAO,WAGjB,GAAQ,KACV,KAAM,IAAI,GAAW,8CAIzB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,OAAS,EACjB,MAAO,GAGX,KAAM,IAAI,GAAW,kBAAkB,IAQzC,EAAA,UAAA,gBAAA,WAAA,GAAA,GAAA,KAKE,MAAO,GAAA,KAAK,WAEV,OADM,GAAmB,GACL,EAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACL,EAAY,EAAG,EAAY,EAAM,aAAa,OAClD,EAAE,GACL,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,EAAK,eAAe,IAAI,IAC1B,EAAO,KAAI,MAAX,EAAe,EAAM,mBAK3B,MAAO,MAIX,EAAA,UAAA,UAAA,WAWE,OAVM,GAAmC,CAAC,KAAM,KAAK,MAK/C,EACF,KAAK,uBAAuB,KAAK,QAG/B,EAAe,GACD,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EAAiB,EAAM,eACvB,EAAc,EAAM,YACpB,EAAuB,GACpB,EAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAO,EAAM,aAAa,GAC1B,EAAU,EAAU,QAAQ,EAAO,GACrC,EAAS,GACb,GAAI,KAAK,eAAe,IAAI,IAG1B,GAAI,EAAK,SACP,IACE,KAAK,UAAU,EAAK,UACpB,EAAS,EAAK,eACP,GACP,QAAQ,KACJ,SAAS,EAAM,KAAI,mDAEhB,GAAK,SAAQ,gCAChB,+EAEJ,EAAS,GAGb,GAAI,EAAK,cAAc,OAAS,GAE9B,OADM,GAAW,GACR,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAU,EAAU,QAAQ,EAAc,GAC5C,EAAe,EAAkB,GACrC,AAAI,GAAgB,MAClB,GAAe,GAEjB,EAAS,KACL,CAAC,EAAa,KAAM,EAAc,EAAa,IAErD,EAAqB,KAAK,KAIhC,GAAM,GAAiC,GACvC,EAAK,KAAU,EAAM,KACrB,EAAK,UAAe,EACpB,EAAK,OAAY,EACjB,EAAK,aAAkB,EACvB,EAAa,KAAK,GAEpB,EAAO,OAAY,EAGnB,OADM,GAAc,GACX,EAAI,EAAG,EAAI,KAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAY,KAAK,uBAAuB,GAExC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,yBAAyB,GAClD,EAAY,KAAK,CAAC,EAAM,KAAM,EAAc,IAE9C,EAAO,YAAiB,EAGxB,OADM,GAAe,GACZ,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GAEzC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,0BAA0B,GACnD,EAAa,KAAK,CAAC,EAAM,KAAM,EAAc,IAE/C,SAAO,aAAkB,EAClB,GAgBF,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAA8C,GAO9C,EAAkE,GACxE,WACI,EAAc,GAChB,AAAM,EAAM,OAAQ,GAGlB,EAAiB,EAAM,MAAM,KAAK,GAFlC,EAAiB,EAAM,MAAQ,CAAC,GAMpC,WAAqB,EAAc,GAGjC,OAFM,GAAiC,GACnC,GACoB,GAAA,EAAA,GAAA,EAAA,GAAA,GAAA,OAAA,MAAnB,GAAM,IAAS,GAAA,IACZ,GAAmB,GAAU,GAC7B,GAAmB,GAAU,GAC7B,GAAqB,GAAU,GAKrC,GAHA,GAAS,GAAU,IAAM,KACrB,GACA,GAAU,GACV,CAAE,MAAoB,KACxB,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAe,EAAc,IACnC,GAAI,GAAa,aAAa,QAAU,IACtC,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAc,GAAa,aAAa,IAC9C,EAAa,KAAK,GAAY,cAAc,KAK9C,AAAI,EAAa,OAAS,GACxB,EAAM,MACF,GAA+B,GAC/B,IAUR,WAAsB,GACpB,GAAM,GAAY,EAAU,KAEtB,EACF,GACI,EACA,EAAO,eAAoB,KACvB,EAAO,cACP,IACZ,EAAM,6BAA6B,GACnC,EAAc,GAAa,EAE3B,GAAM,IACF,EAAU,aACd,GAAiB,QAAQ,SAAA,IACvB,GAAI,CAAE,cAAoB,QACxB,KAAM,IAAI,GACN,yDACI,IAMV,EAAmB,EAAO,MAO9B,OAFM,GAAO,EAAO,KACd,EAAmB,EAAO,OACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAa,GAOf,KAAO,CAAC,GAA4B,IAClC,OAAwB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAQ,EAAc,EAAU,MACtC,GAAI,EAAM,OAAQ,IAChB,GAAM,GAAkC,EAAiB,EAAM,MAC/D,MAAO,GAAiB,EAAM,MAC9B,OAAuB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAY,EAAO,KAU3B,OAJM,GAAiC,GACjC,EAAkC,GAClC,EACF,EAAO,YACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAa,KAAK,EAAmB,IAIvC,OAFM,GACF,EAAO,aACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAc,KAAK,EAAmB,IAExC,MAAO,IAAI,GAAI,CAAC,OAAQ,EAAc,QAAS,EAAe,KAAI,KASpE,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WAGE,GAAI,KAAK,UACP,KAAM,IAAI,GACN,wLAIN,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,SACR,MAAO,GAGX,MAAO,oCAST,EAAA,UAAA,YAAA,WAAA,GAAA,GAAA,KACE,EAAA,KAAK,WACH,EAAK,OAAO,QAAQ,SAAA,GAElB,AAAI,EAAM,UACR,EAAM,mBAMhB,GA3vCwC,IC7CxC,AAiCA,YACI,EAAmD,EACnD,GACF,GAAM,GAAa,EAAY,OAC/B,GAAI,GAAW,MAAS,MAAM,QAAQ,IAAY,EAAQ,SAAW,EACnE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,QAEjC,GAAI,IAAe,EACjB,MAAI,OAAM,QAAQ,IAAY,EAAQ,SAAW,EACxC,EACE,MAAO,IAAY,UAAY,EAAY,IAAM,GACnD,CAAE,EAA2B,EAAY,KAEzC,CAAC,GAGZ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAQ,SAAW,EACrB,KAAM,IAAI,OACN,YAAY,EAAU,mBAAmB,EAAQ,OAAM,IACvD,kCAAiC,EAAU,cAC3C,iEAEN,MAAO,WAEL,MAAO,IAAY,UAAY,OAAO,KAAK,GAAS,OAAS,GAC7D,MAAQ,GAA2B,OAAO,KAAK,GAAS,KACpD,UACN,GAAM,GAAwB,GAC9B,SAAY,QAAQ,SAAA,GAClB,AAAI,IAAc,GAChB,EAAO,KAAM,EAA2B,IAExC,EAAO,KAAK,QAGT,MAEP,MAAM,IAAI,OACN,2BAA2B,EAAU,cACrC,OAAM,EAAU,kCACb,GAAU,+BAA+B,EAAW,WACvD,aAAY,EAAU,oBAAoB,KAAK,UAAU,iBAkB7D,EACA,GACF,MAAO,IACH,EAAa,EAAa,2BA6B5B,EAAW,EAAuB,EAClC,2GACF,GAAI,GAAgB,MAAQ,GAAoB,KAG9C,KAAM,IAAI,OAAM,qDAGd,IAAe,KAEX,GAAqB,EAAA,KAAK,WAC9B,GAAI,EAAE,MAAM,SAAW,EAErB,MAAO,GAAE,QACJ,GAAI,EAAE,MAAM,SAAW,EAC5B,GAAI,EAAE,MAAM,GAAK,GAEf,GAAM,GAAO,EACb,MAAO,GAAE,OAAO,QACX,GAAI,EAAE,MAAM,KAAO,EAExB,MAAO,GAAE,QAAQ,CAAC,EAAE,MAAM,KAE1B,KAAM,IAAI,OACN,+CAA+C,EAAE,MAAM,GAAE,4EAK/D,MAAM,IAAI,OACN,yCAAyC,EAAE,KAAI,4EAKjC,EAAA,GAAA,OAAM,KAAK,CAAA,EAAM,EAAS,SA3B9C,CAAA,EAAA,UA2BI,SAAgB,EAAA,MAAA,EAAA,CAAW,EAAA,SACjC,EAAA,QAAQ,GACF,EAA8B,GACpC,EAAc,QAAQ,SAAA,GACpB,GAAI,EAAY,IAAe,KAC7B,KAAM,IAAI,OACN,8DACA,cAAa,EAAU,mCACvB,eAEJ,EAAkB,KAAK,EAAY,MAIvC,CAAA,EAAO,EAAA,SAAS,EAAmB,mBAEnC,MAAA,CAAA,EAAO,uBAWyB,EAAgB,GAClD,MAAO,GAAA,IAAI,EAAQ,GC5LrB,AAyLA,GAAM,IAAgC,GAgBtC,YAII,EAAY,GACd,GAAI,GACA,EAEE,EAAiB,EACvB,EAAK,EAAe,GACpB,EAAK,EAAe,GACpB,EAAA,KAAS,OACL,GAAM,MAAQ,GAAM,KACpB,WAAM,MAAA,+OAIF,IAAG,KAEX,GAAM,GACF,GAA0B,QAAS,EAAM,WAAY,GACnD,EACF,GAA0B,SAAU,EAAM,YAAa,GAErD,EAAoB,EAAY,GAAG,MAAM,GAE/C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,OAAO,OACpC,WAAM,MAAA,mBAAmB,EAAM,OAAO,OAAM,4BACxC,aAAY,EAAY,OAAM,oCAC3B,MAAK,UAAU,EAAM,YAAW,OAE3C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,QAAQ,OACrC,WACI,MAAA,mBAAmB,EAAM,QAAQ,OAAM,6BACvC,aAAY,EAAY,OAAM,sCAC3B,MAAK,UAAU,EAAM,aAAY,OAE5C,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,8BACC,GAAM,WAAW,GAAO,QACrB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,+BACC,GAAM,YAAY,GAAO,QACtB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,MAAO,CAAC,GAAI,EAAa,GAAI,GAG/B,YACI,EAAuB,EAAiB,GAC1C,GAAI,YAAkB,GAAA,OACpB,MAAO,CAAC,GACH,GAAI,MAAM,QAAQ,GACvB,SAAA,KAAS,OACL,EAAO,SAAW,EAAM,OACxB,WAAM,MAAA,wBAAwB,EAAO,OAAM,0BACvC,EAAM,OAAM,iBAAiB,EAAa,SAAS,EAAK,MACzD,EAIP,OAFM,GAAuB,GAEV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,IAAS,KAClB,KAAM,IAAI,GACN,gEACG,GAAa,SAAS,EAAI,OAEnC,EAAO,KAAK,EAAO,IAErB,MAAO,GAIX,YACI,GAMF,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,IACN,0DAEN,MAAO,CAAC,GAAI,EAAK,GAAI,GAAI,EAAK,gBAO5B,EAAY,EACZ,yKA0BF,GAzBM,EAAqB,EAAK,iBAAmB,KACnD,EAAA,KAAS,OACL,EAAM,WAAa,KACnB,WAAM,MAAA,mGAGV,EAAA,KAAS,OACL,GAAQ,KACR,WAAM,MAAA,kGAEV,EAAA,KAAS,OACL,EAAK,QAAU,MAAQ,EAAK,OAAS,GAAK,OAAO,UAAU,EAAK,QAChE,WAAM,MAAA,gEACF,qBAAoB,EAAK,UACjC,EAAA,KAAS,OACL,CAAC,GACI,EAAK,gBAAkB,GAAK,OAAO,UAAU,EAAK,iBACvD,WAAM,MAAA,gEACF,2CAA0C,EAAK,mBACvD,EAAA,KAAS,OAEJ,EAAa,iBAAsB,KACpC,WAAM,MAAA,sFAGN,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,mDAGX,EAAe,EAAK,gBAAkB,KACxC,EAAK,OACL,EAAK,OACT,AAAI,GACF,CAAI,GAAgB,EAAK,gBACvB,EAAA,KAAS,OACL,EAAK,mBAAqB,MACrB,EAAK,kBAAoB,GACzB,OAAO,UAAU,EAAK,mBAC3B,WAAM,MAAA,yIAGF,YAAW,EAAK,qBAElB,GAAiB,GACnB,EAAK,gBAMT,EAAQ,EAAe,GACvB,EAAQ,EAAe,KAIrB,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAe,OACnB,AAAI,EACF,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,IAAK,MAAA,OAAS,MAEzD,EAAkB,EAAU,QAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YACtD,EAAU,EAAK,SAAW,KAAO,EAAI,EAAK,QAC1C,EAA0B,GAC5B,EAAW,EAAS,EAAK,OAAQ,KAAM,KACvC,GAAiB,EAAS,GAC1B,KACA,EAAc,GAJX,EAAY,EAAA,aAAE,EAAA,EAAA,QAKrB,EAAa,SAAS,GACtB,EAAM,QAAU,EAEhB,CAAA,EAAM,EAAa,uBAAnB,SAAA,OACA,EAAM,cAAgB,GAClB,EAAQ,EAAK,cAAgB,KAAO,EAAI,EAAK,aAE9B,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BACZ,GAAQ,EAAK,OACZ,GAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,KAFR,CAAA,EAAA,iBAExB,GAAA,OACI,EAAY,EACZ,EAAa,EACZ,EAAD,CAAA,EAAA,GACa,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BAEV,GAAqB,EAAY,EAAK,gBAAkB,IACzC,CAAA,EAAM,EAAa,QAD0B,CAAA,EAAA,WAKjE,MAJM,GAAc,EAAA,OAIhB,GAAsB,EAAY,KACpC,SAAQ,KACJ,qCACG,GAAK,gBAAe,MACvB,mDACG,GAAS,cACZ,8HAGG,GAAK,gBAAkB,EAAK,OAAM,eACrC,yEAEJ,CAAA,EAAA,KAGE,EAAY,OAAS,KACjB,GACF,GAA8B,EAAO,EAAY,OAD9C,EAAE,EAAA,GAAE,EAAE,EAAA,GAEP,EAA4B,GAClC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAG,GAAG,MAAM,GAEhC,CAAA,EAAM,EAAa,aAAa,EAAY,KAP1C,CAAA,EAAA,cAOF,EAAA,OAEM,EAA8B,IAChC,GAAK,aAAe,MAApB,MAAA,CAAA,EAAA,IACI,EACF,GAAwB,EAAK,YAAa,EAAM,aAC3C,EAAI,2BAAG,GAAI,EAAqB,OACvC,GAAA,GAAA,GAAc,KAAK,CAAA,EAAM,GACrB,EAAG,GAAI,KAAM,EAAqB,MAFO,CAAA,EAAA,YAC7C,EAAA,MAAA,EAAA,CAAmB,EAAA,4BAD4B,QAAE,iBAUrD,IAHM,EAAM,EAAG,OAAO,GAAI,OAAO,GAC3B,EAAO,EAAc,GAC3B,EAAA,QAAY,GACH,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EAChC,EAAQ,EAAU,GAClB,EAAM,EAAK,GACjB,EAAU,GAAS,EACnB,EAAA,KAAS,GAGX,MAAA,CAAA,EAAM,EAAa,WAAW,EAAY,YAA1C,EAAA,OACA,GAAqB,GAErB,IACA,6BAGE,GAAqB,GAAa,EAAK,gBAClB,EAAY,MAE/B,EACE,GAAO,OACP,GAAgB,EAAK,gBACb,GAAA,GAAO,CAAA,EAAM,EAAM,gBACzB,EAAK,eAAgB,CAAC,QAAS,EAAK,sBAFtC,CAAA,EAAA,KAFF,CAAA,EAAA,IAHF,CAAA,EAAA,YAME,SAAU,EAAA,MAAA,OAAA,CAAO,EAAA,wBAGjB,EAAU,GAAO,EAAM,SAAS,EAAO,EAAO,CAC5C,UAAW,EAAK,qBAAuB,KACnC,GACA,EAAK,oBACT,QAAS,wBAGb,IAAS,EAAI,EAAG,EAAI,EAAM,aAAa,OAAQ,EAAE,EAC/C,EAAU,OAAO,EAAM,aAAa,IAAQ,EAAQ,sBAQxD,MAAA,CAAA,EAAA,YAGF,MAAI,GAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,YAErC,MAFA,GAAA,OACA,IACI,EAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,sBAAnB,SAAA,OACA,CAAA,EAAM,EAAM,QAAQ,oBAApB,SAAA,OACA,CAAA,EAAO,EAAM,iBAEb,SAAM,WAAa,8BAKvB,YACI,EAAqB,GAEvB,GAAI,GAAwB,KAC5B,MAAI,GAAK,iBAAmB,KAC1B,EAAgB,EAAK,gBAChB,AAAI,OAAO,SAAS,EAAQ,OACjC,GAAgB,EAAQ,MAEnB,EAKT,YACI,GAKF,MAAQ,OAAQ,GAAuB,UAAa,WAKtD,YAAiC,GAE/B,MAAQ,OAAQ,GAA6B,MAAS,uBAOpD,EAAY,EACZ,uHAKF,GAJA,EAAO,GAAQ,GACT,EAAa,EAAK,SAAW,KAC7B,EAAI,EAAM,aACZ,EAAqB,GACrB,EAAK,QAAU,EACjB,KAAM,IAAI,IAAoB,8CAGhC,GAAA,KAAS,OACL,CAAC,GAAe,EAAK,QAAU,GAAK,OAAO,UAAU,EAAK,SAC1D,WAAM,MAAA,6DACF,aAAY,KAAK,UAAU,EAAK,YACnB,GAAqB,GACtC,GAAA,SADiB,CAAA,EAAA,UAEjB,MAAA,CAAA,EAAO,EAAuB,mBAA9B,EAAA,EAAA,wBAFE,EAAY,EAId,EAAc,EACd,EAAQ,uEAGU,MAAA,CAAA,EAAM,EAAa,eAmCvC,MAnCM,GAAc,EAAA,OACpB,EAAO,EAAA,KAAS,WACd,GAAI,EAAY,OAGR,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAU,EAAG,OAAO,GACpB,EAAY,EAAA,KAAS,WAAM,MAAA,GAAE,KAGnC,GAFA,EAAA,QAAY,GAER,IAAU,EACZ,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAKrB,OADM,GAAY,EAAQ,GAAG,MAAM,cAC1B,GACP,GAAM,GAAW,EAAU,GACrB,EAAY,EAAK,GACvB,EAAK,GACD,EAAA,KAAS,WAAM,MAAA,GAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,MACvD,AAAI,EAAQ,GACV,EAAA,QAAY,IANP,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,IAA/B,GAST,EAAA,QAAY,GACZ,GAAe,EAEf,EAAE,EAEJ,MAAO,KAGL,EAAY,KACd,CAAI,GACF,QAAQ,KACJ,uJAGA,2BAA0B,EAAK,QAAO,eACtC,qHA3CH,GAAa,EAAQ,EAAK,QAAU,YAAI,CAAA,EAAA,2DAkD/C,IAAS,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC3B,EAAY,EAAK,GACvB,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAC3B,EAAA,QAAY,GAGd,MAAA,CAAA,EAAO,GAAiB,SCrmB1B,YAiK+B,GAC7B,EAAA,KAAS,OACL,EAAY,GAAK,OAAO,UAAU,GAClC,WAAM,MAAA,2DACF,gBAiBN,EAAyB,EAAe,GAC1C,MAAI,IAAU,KACL,CAAC,MACC,MAAM,QAAQ,GAChB,EAAO,IAAI,SAAA,GAAS,MAAA,IAAoB,EAAO,EAAO,EAAO,KAE7D,GAAoB,EAAQ,EAAO,EAAO,eAkBjD,EAAyB,GAC3B,MAAO,GAAA,KAAS,WACd,MAAI,IAAU,KACL,KACE,MAAM,QAAQ,GAChB,EAAO,IACV,SAAA,GAAS,MAAC,IAAqB,EAAO,KAInC,GACH,EAAQ,EAAQ,QAAU,QAAU,EAAU,EAAQ,uBAc5D,EAAc,GAIhB,OAHM,GAAkC,GACpC,EAAa,EACb,EAAmB,KAChB,EAAa,GAClB,EAAW,EAAa,EACxB,AAAI,GAAY,GACd,GAAW,GAEb,EAAO,KAAK,CAAC,EAAY,IACzB,EAAa,EAEf,MAAO,GA8BT,YAGI,EAAY,EAAiC,EAC7C,EAAsB,EAAoB,EAAiB,EAC3D,EAA4B,EAC5B,EAAmB,EAA0B,EAC7C,EAAuB,EACvB,mHAoBF,GAnBA,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,GAAU,MACZ,GAAS,GAEX,AAAI,GAAW,MACb,GAAU,IAEZ,AAAI,GAAgB,MAClB,GAAe,GAIb,EAAe,GACnB,AAAI,GAAQ,MAAQ,GAAU,MAC5B,GAAe,IAGb,GAAmB,MACrB,GAAe,GACX,GAAiB,MACnB,KAAM,IAAI,GACN,oGAKF,SACF,EAAM,gBAAgB,EAAK,EAAW,EAAe,mBAEzD,AAAI,GAAmB,MACrB,GAAa,GAAM,EAAG,IAGxB,AAAI,GAAW,MACb,GAAU,GAGN,EAA0B,GAC5B,EAAW,EAAS,EAAQ,EAAc,EAAiB,EAC3D,EAAW,EAAc,GAFtB,EAAY,EAAA,aAAE,EAAO,EAAA,QAG5B,EAAa,SAAS,GACtB,EAAM,QAAU,EAChB,CAAA,EAAM,EAAa,uBAAnB,EAAA,OACA,EAAM,cAAgB,cAIb,qEACP,MAAA,CAAA,EAAM,EAAa,aAAa,cAAhC,EAAA,OACM,EAA4B,IAC9B,IAAiB,MAAjB,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,IACN,qDAEJ,GAAI,IAAY,QACd,KAAM,IAAI,IAAoB,0CACzB,AAAI,GACT,EAAA,KAAK,QAAQ,GAIT,EAAoB,EAAA,SAAS,GAE7B,EAAU,GAAY,EAAiB,cACpC,2DACD,SAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,EAAY,WAA5C,SAAA,OAEA,EAAA,KAAS,WACP,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,GAAW,GACI,EAAmB,EACnB,EAAW,GAChC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAW,EAM/B,OAFM,IAAW,GAAqB,EAAK,IACrC,GAAO,EAAE,IACN,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAK,IACjB,EAAU,IAAS,GACnB,EAAA,KAAS,IAIX,GAAI,IAAe,EAAQ,OAAS,GAC9B,EAGF,OAFM,IAAU,EAAM,SAAS,EAAM,EAAQ,GAEpC,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAQ,IACpB,EAAA,KAAS,IAET,EAAU,OAAS,IAAS,MAMpC,CAAA,EAAM,EAAa,WAAW,EAAY,WAG1C,MAHA,GAAA,OACA,GAAqB,GAEjB,EAAM,kCA3CH,EAAa,yBAAG,GAAa,EAAQ,YAArC,IAA2C,CAAA,EAAA,iEAAE,QAAE,eAiDxD,EAAkB,2BAGpB,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,WACrC,MADA,GAAA,OACI,EAAM,kCAtEH,EAAQ,yBAAc,GAAQ,OAA9B,IAAoC,CAAA,EAAA,iEAAE,QAAE,eA0EjD,MAAA,CAAA,EAAM,EAAa,qBAAnB,SAAA,OAEA,CAAA,EAAM,EAAM,QAAQ,mBAApB,SAAA,OACA,CAAA,EAAO,EAAM,0BAMX,EAAY,EACZ,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,yIACF,GAAI,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,iDASX,EAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIT,EAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAG,EAAG,EAAK,aAAc,EAAK,YAAa,EAC3C,cAHF,EACF,EAAA,OAGJ,EAAS,EAAiB,GAC1B,EAAU,EAAiB,GAC3B,EAAgB,EAAiB,GAG7B,EAAe,GACf,EAAM,QACN,GAAK,gBAAkB,MAAQ,EAAK,eAAe,OAAS,GAA5D,MAAA,CAAA,EAAA,GAEF,GADA,EAAe,GACX,EAAK,eAAe,SAAW,EAEjC,EAAY,EAAK,eAAe,GAChC,EAAY,EAAK,eAAe,OAC3B,MAAI,GAAK,eAAe,SAAW,EAClC,GAAI,IACN,iEAEE,GAAI,GACN,0GAEG,GAAK,eAAc,iBAGtB,SAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAW,EAAW,KACtB,KACA,EAAgB,WAJlB,SACF,EAAA,OAIJ,EAAO,EAAgB,GACvB,EAAO,EAAgB,GACvB,EAAS,EAAK,OAAO,gBAEhB,AACH,EAAK,iBAAmB,MAAQ,EAAK,gBAAkB,GACvD,EAAK,gBAAkB,EACzB,GAAe,GAET,EACF,KAAK,MAAM,EAAO,GAAG,MAAM,GAAM,GAAI,EAAK,kBACxC,EAAoB,EAAO,GAAG,MAAM,GAC1C,EAAO,GAAY,EAAQ,EAAS,GACpC,EAAS,GAAY,EAAQ,EAAG,GAChC,EAAO,GAAY,EAAS,EAAS,GACrC,EAAU,GAAY,EAAS,EAAG,GAGlC,EAAS,EAAK,OAAO,IAGhB,AAAI,EAAK,iBAAmB,MACjC,GAAe,qBAIX,SAAM,EAAO,OAAO,GAAS,OAAO,GAE1C,EAAM,mCAcA,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAW,OACX,EAAe,OACnB,AAAI,EACF,GAAM,mBACN,EAAc,EAAM,aACpB,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,MAEzD,GAAc,KACd,EAAS,GACT,EAAkB,EAAU,SAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YAChD,CAAA,EAAM,GACd,EAAO,EAAe,EAAK,EAAW,EAAW,EAAK,OACtD,EAAK,QAAS,EAAW,EAAa,EAAQ,EAAK,QACnD,EAAiB,EAAK,aAAc,KAAM,cAHxC,SAAM,EAAA,OAIZ,CAAA,EAAO,UAEP,SAAM,WAAa,GAEnB,GAAkB,EAAQ,GAC1B,GAAkB,EAAS,GAC3B,GAAkB,EAAkB,GACpC,GAAkB,EAAkB,GACpC,AAAI,GAAiB,MACnB,EAAA,QAAY,yCAYyB,GACzC,GAAM,GAAiB,GACvB,AAAI,YAAmB,GAAA,QACrB,GAAU,CAAC,IAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAS,EAAQ,GACvB,GAAI,EAAO,OAAS,EAClB,EAAK,KAAK,GAAW,EAAQ,SACxB,GAAI,EAAO,OAAS,EACzB,KAAM,IAAI,OACN,yEAGJ,EAAK,KAAK,IAGd,MAAO,eAgBL,EACA,GACF,GAAI,GAAW,KACb,OAEF,GAAM,GAAyB,GAC/B,GAAI,YAAsB,GAAA,OACxB,EAAa,KAAK,EAAW,YACpB,MAAM,QAAQ,GACvB,EAAW,QAAQ,SAAA,GAAK,MAAA,GAAa,KAAK,EAAE,cACnC,GAAc,KAEvB,OAAW,KAAQ,IACjB,GAAM,GAAY,EAAW,GAC7B,EAAa,KAAK,EAAU,IAIhC,GAAM,GAA6B,GACnC,GAAI,YAAmB,GAAA,OACrB,AAAI,EAAa,QAAQ,EAAQ,MAAQ,IACvC,EAAiB,KAAK,WAEf,MAAM,QAAQ,GACvB,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAa,QAAQ,EAAE,MAAQ,IACjC,EAAiB,KAAK,aAGjB,GAAW,KAEpB,OAAW,KAAQ,IACjB,GAAM,GAAS,EAAQ,GACvB,AAAI,EAAa,QAAQ,EAAO,MAAQ,IACtC,EAAiB,KAAK,GAK5B,EAAiB,QAAQ,SAAA,GACvB,AAAK,EAAE,YACL,EAAE,YCrnBR,YA+C6B,GAE3B,MAAO,aAAa,GAAA,mBAMM,GAE1B,MAAO,OAAM,QAAQ,eAMI,GAEzB,MAAO,CAAC,GAAa,IAAM,CAAC,GAAY,eAetC,EAAqD,EACrD,EAAkB,EAAuB,GAC3C,GADoB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACvC,GAAS,MAAQ,EAAM,SAAW,GAGpC,GAAI,GAAQ,MACV,GAAI,GAAoB,GACxB,GAAI,GAAY,IAAU,EAAkB,OAAS,EACnD,EAAoB,WACX,GAAW,IACpB,OAAW,KAAO,GAChB,GAAI,EAAK,eAAe,IACtB,EAAoB,GACpB,WAKJ,GAAoB,GAEtB,GAAI,EACF,KAAM,IAAI,GACN,6BAA6B,EAAe,sBAC5C,YAAW,IAGnB,MAAO,GAET,GAAI,GAAQ,KACV,MAAO,GAAM,IAAI,SAAA,GAAQ,MAAA,QAG3B,GAAI,GACJ,GAAI,GAAW,IACb,EAAO,EACP,EAAS,GACT,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAK,IAAS,KAChB,KAAM,IAAI,GACN,yBAAyB,EAAI,iCAC7B,IAAG,IAET,EAAO,KAAK,EAAK,aAEV,GAAY,IAErB,GADA,EAAO,EACH,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,oCAAmC,EAAM,OAAM,oBAC/C,iDAAgD,IAEtD,EAAS,OAGT,GADA,EAAO,EACH,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,aAAa,EAAe,YAAY,EAAM,OAAM,eACpD,2DACI,EAAK,QAEf,EAAS,CAAC,GAMZ,GAHA,EAAS,GAA2B,GAGhC,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,EAAM,QAErB,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EAEd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MAAQ,GAAU,GAAK,IAAQ,EAC3C,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,mBAAkB,EAAO,GAAE,gCAC3B,KAAI,EAAM,MAAK,QAK3B,MAAO,eAWL,EAAkB,EAAmB,GACvC,GAAM,GAAO,GAAO,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,MAAM,MACpD,EAAK,OACL,GAAM,GAAO,GAAO,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,MAAM,MAGvD,GAFA,EAAK,OAED,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,mFAEA,IAAG,KAAK,UAAU,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,WAElD,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,oFAEA,IAAG,KAAK,UAAU,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,WAErD,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,CAAC,EAAA,KAAK,YAAY,EAAM,GAChE,KAAM,IAAI,GACN,kEACA,mBAAkB,EAAK,GAAE,wBAAwB,EAAK,GAAE,YACxD,cAaR,YACI,EAAmB,EAA2B,GAMhD,OAJM,GAAY,CAChB,GAAyB,GACzB,IAEO,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAI,EAAQ,GACZ,EAAO,EAAQ,GACf,EAAQ,EAAa,GAC3B,GAAI,GAAQ,KACV,SAEF,GAAI,IAAS,IACP,EAAE,MAAM,EAAE,MAAM,OAAS,KAAO,EAClC,KAAM,IAAI,GACN,2CAA2C,EAAE,MAAK,4JAO1D,GAAI,EAAU,QAAQ,KAAU,GAG9B,OAFM,GAAe,EAAE,MAAM,MAAM,GAC7B,EAAc,EAAM,MAAM,GACvB,EAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAY,EAAa,GACzB,EAAS,EAAY,GAC3B,GAAI,GAAU,MAAQ,IAAc,EAClC,KAAM,IAAI,GACN,8BAA8B,EAAE,MAAK,sBACrC,oBAAmB,EAAK,uCACxB,2DAiCd,YACI,EAAuB,EAAiB,EACxC,EAAuB,GAAvB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACzB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,wCAAuC,EAAM,OAAM,eACnD,qBAAoB,EAAK,OAAM,iBAErC,EAAS,OAET,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,qBAAqB,EAAM,OAAM,IAAI,EAAe,mEAEjD,MAAK,UAAU,EAAK,OAAM,MAEnC,EAAS,CAAC,GAGZ,GAAI,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,KAAK,UAAU,EAAM,SAEpC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EACd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MACR,IAAW,EACb,KAAM,IAAI,GACN,uBAAuB,EAAe,cACnC,GAAM,GAAE,kBAAkB,KAAK,UAAU,EAAO,IAAG,SACtD,yBAAwB,KAAK,UAAU,EAAM,OAAM,oBAsB/D,EAEA,GACF,GAAI,GAAW,MAAQ,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAClE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,KAGjC,GAAI,GAEJ,GAAI,MAAO,IAAY,UAAY,MAAO,IAAY,WACpD,EAAiB,CAAC,WACT,MAAM,QAAQ,IAAY,MAAO,IAAY,SACtD,EAAiB,MAGjB,MAAM,IAAI,WACN,+DACA,uCAAsC,IAG5C,GAAI,MAAM,QAAQ,GAEhB,MAAO,GAAY,IACf,SAAA,GAAQ,MAAA,KAIZ,OADM,GAAqD,GACxC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACT,EACA,EAAe,eAAe,GAAQ,EAAe,GAAQ,GACjE,AAAK,MAAM,QAAQ,IACjB,GAAgB,CAAC,IAEnB,EAAc,KAAK,GAErB,MAAO,GA6DX,GAAM,IAA2B,8BAcA,EAAA,EAAA,GA4C/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WAAa,KAsCpB,SAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAGF,GAHE,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC1D,CAAC,KAAK,MACR,KAAM,IAAI,GACN,gLAIN,GAAa,KAAM,EAAY,EAAW,IAa5C,EAAA,UAAA,QAAA,SAAQ,GAAR,GAAA,GAAA,KAME,GALA,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAEd,KAAK,KAAO,EAAK,KAEb,MAAO,GAAK,WAAc,SAC5B,KAAK,WAAa,GAAwB,EAAK,WAC/C,KAAK,iBAAmB,QAExB,GAAI,CAAE,GAAK,oBAAqB,GAAA,WAC9B,KAAM,IAAI,GACN,+DAEN,KAAK,WAAa,EAAK,UACvB,KAAK,iBAAmB,GAO1B,GAAI,GAAkC,GACtC,GAAI,CAAC,MAAM,QAAQ,EAAK,OAAS,MAAO,GAAK,MAAS,UAClD,MAAO,GAAK,MAAS,YACvB,EAAK,KAAO,EAAK,KACjB,OAAW,KAAQ,GAAK,KACtB,GAAI,KAAK,YAAY,QAAQ,KAAU,GACrC,KAAM,IAAI,GACN,sCAAsC,EAAI,MAC1C,sCAAqC,KAAK,cAGlD,OAAmB,GAAA,EAAA,EAAA,KAAK,YAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAK,KAAK,IAAS,MACrB,QAAQ,KACJ,WAAW,EAAI,4GAEf,oBAAmB,EAAI,qBAE7B,EAAc,KAAK,GAAW,EAAK,KAAK,cAEjC,MAAM,QAAQ,EAAK,OAC5B,GAAI,EAAK,KAAK,SAAW,KAAK,QAAQ,OACpC,KAAM,IAAI,GACN,+DACA,gCAA+B,KAAK,QAAQ,OAAM,gBAClD,wBAAuB,EAAK,KAAI,MAEtC,GAAM,GAAY,EAAK,KACvB,EAAgB,EAAU,IAAI,SAAA,GAAK,MAAA,IAAW,UAE9C,GAAM,GAAe,GAAW,EAAK,MACrC,KAAK,QAAQ,QAAQ,SAAA,GACnB,EAAc,KAAK,KAIvB,KAAK,cAAgB,EAErB,KAAK,gBAAkB,GACvB,KAAK,iBAAmB,GACxB,KAAK,YAAc,GACnB,OAAS,GAAI,EAAG,EAAI,KAAK,QAAQ,OAAQ,EAAE,GAEzC,GAAM,GAAQ,KAAK,qBAAqB,GAClC,EAAO,KAAK,YAAY,GAC9B,KAAK,gBAAgB,KAAK,GAC1B,KAAK,iBAAiB,KAAK,GAC3B,KAAK,YAAY,KAAK,KAAK,cAAc,IAK3C,GAAM,GAA8B,GAGpC,KAAK,QAAU,EAAK,QAEpB,KAAK,aAAe,CAAC,QACrB,KAAK,eAAiB,GAMtB,GAAU,OAAQ,WAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,GACzC,GAAI,EAAkB,QAAQ,KAAO,GACnC,SAIF,GAAM,GAAe,EAAK,cAAc,GACxC,AAAI,EAAK,QAAQ,OAAS,GACxB,GAAK,eAAe,KAAK,CAAC,EAAc,IACxC,EAAK,aAAa,KAAK,EAAK,YAAY,GAAK,aAQnD,GAAM,GAAgB,GAAe,EAAK,QAAS,KAAK,aAMlD,EACF,SAAC,EAAqB,EACrB,GACC,AAAI,EAAK,YAAY,OAAS,GAC5B,GAAa,EAAK,YAAY,GAAe,IAAM,GAErD,EAAK,aAAa,KAAK,GACvB,EAAK,eAAe,KAAK,CAAC,EAAc,KAG9C,GAAU,SAAU,WAClB,mBAAS,GACP,GAAI,EAAkB,QAAQ,KAAO,oBAGrC,GAAM,GAAgB,EAAc,GAI9B,EAAgB,SAAC,GAOrB,OANM,GAAmB,GACrB,EACA,EACA,aAGO,GACT,GAAI,MAAO,IAAW,UAClB,CAAC,WAAY,MAAO,eAAgB,MAAM,QAAQ,KAC9C,IACN,GAAM,GAAc,EAAK,qBAAqB,GAE9C,AAAI,EAAY,EAAY,OAAS,KAAO,GACxC,EAAK,cAAc,KAAO,GAE5B,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAEL,AACH,EAAK,cAAc,KACnB,GAGF,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAIV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAGZ,GAAI,GAAM,OACV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAS,MACJ,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAS,MAGX,EAAmB,EACnB,EAAa,EAAmB,OAEhC,GAAM,GAAW,GAAY,GAE7B,EAAmB,EACnB,EACI,EAAmB,GAA4B,GAIrD,GAAI,GACJ,GAAU,EAAY,WACpB,EAAe,IAEjB,EAAa,EAAG,EAAY,IAtDT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,KAAN,KA0Db,EAAc,IAzEP,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,IAAlC,KAgFX,KAAK,0BAA4B,KAAK,kBAY9B,EAAA,UAAA,iCAAV,WACE,GAAI,KAAK,2BAA6B,KACpC,OAEF,AAAI,KAAK,iBAAiB,SACtB,KAAK,0BAA0B,QACjC,QAAQ,KACJ,sJAqCR,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIf,GAAM,GAAiB,GACjB,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GACrD,IAGE,GAAM,GAAM,EAAiB,GAAG,OAAO,EAAiB,IACxD,KAAK,mBACL,GAAM,GAAI,KAAK,aACT,EACF,KAAK,SAAS,EAAG,EAAK,EAAW,EAAK,QAAS,EAAK,OACxD,MAAO,IAAiB,WAExB,GAAkB,EAAiB,GAAI,GACvC,GAAkB,EAAiB,GAAI,KA0BrC,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,YAAK,mBACL,CAAA,EAAO,GAAgB,KAAM,EAAS,SAahC,EAAA,UAAA,gBAAR,SACI,EAAsB,EAAoB,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,SACF,GAAI,GACJ,GAAI,GAAS,MAEX,GADA,EAAa,KACT,GAAa,KACf,KAAM,IAAI,GACN,MAAM,EAAS,gDACf,oBAAmB,YAEhB,GAAO,KAChB,AAAI,MAAM,QAAQ,GAChB,EAAa,EAAI,GAAG,MAAM,GAE1B,EAAa,EAAI,MAAM,OAGzB,MAAM,IAAI,GACN,yDACG,GAAS,yBAElB,MAAO,IAUT,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,GAAI,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAC/C,KAAM,IAAI,GACN,sDAGN,GAAM,GAAiB,MAAM,QAAQ,GAC/B,EACD,EAAiB,EAAsB,CAAC,GACvC,EAAwB,KAAK,wBAAwB,GAGrD,EAAW,GAAI,IAIrB,GAHA,AAAI,YAAkB,GAAA,QACpB,GAAS,CAAC,IAER,MAAM,QAAQ,IAChB,GAAI,EAAO,SAAW,KAAK,OAAO,OAChC,KAAM,IAAI,GACN,kCAAkC,EAAO,OAAM,uDAE/C,KAAI,KAAK,OAAO,OAAM,OAE5B,OAAS,GAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,KAAK,OAAO,GAAI,EAAO,QAGtC,QAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAc,EAAO,EAAM,MACjC,GAAI,GAAe,KACjB,KAAM,IAAI,GACN,8CAA8C,EAAM,MAE1D,EAAS,IAAI,EAAO,GAKxB,GAAM,GAAiB,GAAQ,EAAuB,GACtD,MAAO,GAAiB,EAAiB,EAAe,IAMlD,EAAA,UAAA,wBAAR,SAAgC,GAK9B,OAHM,GACF,GAAa,KAAM,EAAoB,QACvC,EAAmB,EAAoB,OACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EACF,MAAM,QAAQ,EAAM,QAAU,EAAM,OAAS,CAAC,EAAM,QAClD,EAAmB,EAAa,IAAI,SAAA,GAAU,MAAA,GAAO,OAClD,EAAI,EAAG,EAAI,EAAoB,OAAQ,EAAE,GAChD,GAAM,GAAQ,EAAiB,QAAQ,EAAoB,IAK3D,GAJA,AAAI,IAAU,IACZ,GAAsB,GAAK,EAAa,GACxC,KAEE,IAAqB,EACvB,MAGJ,GAAI,IAAqB,EACvB,MAIJ,GAAI,EAAmB,GACrB,GAAM,GAA2B,GACjC,QAAsB,QAAQ,SAAC,EAAQ,GACrC,AAAI,GAAU,MACZ,EAAe,KAAK,EAAoB,MAGtC,GAAI,GACN,mDACA,IAAG,KAAK,UAAU,KAExB,MAAO,IAgBD,EAAA,UAAA,YAAR,SAAoB,EAAsB,EAAgB,GAA1D,GAAA,GAAA,KAA0C,MAAA,KAAA,QAAA,GAAA,IAAgB,AAAA,IAAA,QAAA,GAAA,IAEjD,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,GACxC,GAAI,EACF,KAAM,IAAI,IACN,iDAYN,OAJM,GAAU,GAAY,EAAY,GAClC,EAA0B,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,gBAGlD,GACP,GAAM,GAAY,EAAA,KAAS,WACzB,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAG/B,EAAW,GAAY,EAAK,EAAY,GAGxC,EAAQ,GACd,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,EAAE,EACrC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAS,SAGnD,GAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,IAE1C,GAAM,GAAW,GAAI,IAAS,GAC9B,MAAO,IAAQ,EAAK,QAAS,KAE/B,EAAU,QAAQ,SAAC,EAAU,GAAM,MAAA,GAAY,GAAG,KAAK,MApBhD,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,IAA/C,GAsBT,MAAO,IACH,EAAY,IAAI,SAAA,GAAW,MAAA,GAAA,OAAW,EAAS,SA+BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAkB,GAA2B,GACnD,GACI,EAAiB,KAAK,WAAY,KAAK,gBAAiB,IAC5D,IAKE,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,UAAe,GACR,KAAK,YAAY,EAAiB,WAEzC,GAAkB,EAAiB,KAmBvC,EAAA,UAAA,eAAA,SAAe,GACb,GAAe,EAAG,KAAK,WAAY,KAAK,gBAAiB,IAGzD,GAAM,GAAa,OAAM,QAAQ,GAAK,EAAE,GAAK,GAAG,MAAM,GACtD,MAAO,MAAK,YAAY,EAAG,IAGnB,EAAA,UAAA,sBAAV,SACI,EACA,EAAkD,EAClD,GAEF,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,IACN,gGAIN,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,KAAK,iBAAiB,OAAQ,EAAE,GAClD,GAAM,GAAc,KAAK,iBAAiB,GACpC,EAAS,KAAK,YAAY,GAChC,AAAI,IAAW,GACb,EAAa,KACT,EAAY,MAAM,EAAG,EAAY,OAAS,GAAG,OAAO,CAAC,KAGzD,EAAa,KAAK,GAWtB,GARA,EAAI,GACA,EAAG,KAAK,eAAgB,KAAK,gBAAiB,GAAO,SACzD,EAAI,GACA,EAAG,KAAK,gBAAiB,EAAc,GAAO,UAElD,GAAkB,EAAG,GAErB,GAAgC,EAAG,KAAK,YAAa,KAAK,kBACtD,KAAK,UAAY,GAAa,MAAQ,EAAY,GAChD,EAAE,GAAG,MAAM,GAAK,IAAc,EAChC,KAAM,IAAI,GACN,mHAEG,GAAS,YAAY,EAAE,GAAG,MAAM,GAAE,gBAG7C,MAAO,CAAC,EAAG,IAGG,EAAA,UAAA,oBAAhB,SACI,EACA,EACA,EACA,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAA,2GAKF,GAHM,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GAD9C,EAAU,EAAA,GAAE,EAAU,EAAA,GAGzB,GAAgB,KAClB,KAAM,IAAI,OAAM,0CAGd,EAAkC,MAClC,IAAe,MAAf,MAAA,CAAA,EAAA,GACI,EACF,GAAwB,EAAa,KAAK,aAC9C,EAAwB,GACf,EAAI,yBAAG,GAAI,EAAa,OAC/B,GAAA,GAAA,GAAsB,KAClB,CAAA,EAAM,GAAmB,EAAW,GAAI,KAAM,EAAa,MAF1B,CAAA,EAAA,UACrC,EAAA,MAAA,EAAA,CACI,EAAA,0BAFmC,QAAE,eAO7C,MAAA,CAAA,EAAO,CAAC,EAAY,EAAY,UAc1B,EAAA,UAAA,SAAR,SACI,EAAiC,EAAe,EAChD,EAAa,GAFjB,GAAA,GAAA,KAEI,MAAA,KAAA,QAAA,GAAA,GACK,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,EAAK,EAAW,EAAO,SACzD,EAAiB,GACvB,GAAI,EAAU,EACZ,KAAM,IAAI,IAAoB,wCAGhC,GAAI,GAAS,KACX,KAAM,IAAI,IACN,mDAIJ,OAFM,GAAU,GAAY,EAAY,GAClC,EAAa,EAAA,SAAS,GAAM,EAAG,IAC5B,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,GACtD,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,EACF,GACI,EAAY,EAAY,EAAW,GAGrC,EAAW,GAAqB,EAAK,GACrC,EAAY,EAAE,GACpB,GAAI,IAAe,EACjB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAGrB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAW,EAAU,GAC3B,EAAK,GACD,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,EAAY,KAGxD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAG/B,MAAO,MAID,EAAA,UAAA,uBAAV,WAKE,OAJM,GAAY,KAAK,aAGjB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAQ,EAAU,GACpB,EAAW,EACf,GAAI,GAAM,EAAW,GAAS,GAC5B,GAAM,GAAW,GAAM,EAAU,MAAM,EAAG,GAAI,GAC9C,GAAY,IAAI,EAElB,EAAiB,KAAK,GAExB,MAAO,IAaC,EAAA,UAAA,kBAAV,WAAA,GAAA,GAAA,KACE,MAAO,UAAC,GACN,GAAM,GAAuB,GAEvB,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAgB,EAAK,MACvB,EAAK,OAAO,OAAS,EAAK,QAAQ,OAClC,EAAK,OAAO,OAAS,EAAK,QAAQ,OAAS,GAEzC,EAA0B,GAK1B,EAAoB,WAExB,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KASjD,OAPM,GAAW,GAAI,IAAS,GACxB,EACF,GAAQ,EAAK,QAAS,EAAU,CAAC,SAAY,KAI7C,EACK,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GACpC,EAAO,EAAa,EAAQ,GAAI,EAAQ,IAC5C,AAAI,EAAc,IAAM,MACtB,GAAO,GAAoB,EAAM,EAAc,KAIjD,GAAM,GAAmB,EAAA,KAAS,GAElC,EAAW,KAAK,GAChB,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAOnC,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAI,GAAc,OAElB,GAAI,EAAK,QAAQ,OAAS,GAAK,EAAI,EAAK,QAAQ,OAC9C,EAAiB,EAAW,QAE5B,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAC3C,EACI,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAGpD,EAAA,KAAS,GAET,EAAc,KAAK,GAGrB,SAAY,EAAA,KAAS,GAGrB,EAAK,kBAAkB,QAAQ,SAAA,GAC7B,EAAY,EAAA,IAAQ,EAAW,KAG1B,GAGH,EAAY,EAAK,0BAA0B,IAC7C,SAAA,GAAS,MAAA,GAAM,SACb,EAAa,GACb,EACF,EAAK,WAAW,SAAS,EAAmB,EAAY,GAE5D,MAAO,CAAC,GAAgB,OAAO,KAS3B,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,KAAK,aAAe,SAAC,GACnB,MAAO,GAAA,KAAS,WAOd,OANM,GAAuB,GACzB,EACE,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KAKjD,OAHM,GAAW,GAAI,IAAS,GACxB,EAAU,GAAQ,EAAK,QAAS,GAE7B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAGlC,EAAe,EAAA,KAAS,EAAa,EAAQ,GAAI,EAAQ,KAC/D,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAEjC,EAAW,KAAK,GAGlB,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAErC,EACF,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAClD,EAAW,KAAK,GAElB,MAAO,OAuCP,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,MAAA,CAAA,EAAO,GAAW,KAAM,EAAG,EAAG,SA0B1B,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,MAAA,CAAA,EAAO,GAAW,KAAM,EAAS,SA0B7B,EAAA,UAAA,aAAN,SACI,EACA,qHAIqB,MAAA,CAAA,EAAM,KAAK,oBAAoB,EAAG,WAAnD,EAAiB,EAAA,OACjB,EAAS,EAAe,GACxB,EAAU,EAAe,GACzB,EAAgB,KAAK,oBACrB,EAAS,EAAc,EAAO,OAAO,IACrC,EAAuB,OACV,EAAA,yBAAA,GAAA,EAAA,OAAR,GAAI,EAAA,GACH,CAAA,EAAM,EAAK,SADE,CAAA,EAAA,UACjB,EAAI,EAAA,OACV,EAAW,KAAK,EAAE,qBAFD,wBAInB,SAAA,QAAY,GACZ,CAAA,EAAO,GAAiB,UAYhB,EAAA,UAAA,gBAAV,SAA0B,GAMxB,OALM,GAA8B,GAE9B,EAAgB,GAAU,MAAQ,EAAO,cACzC,EAAU,EAAgB,KAAK,iBAAmB,KAAK,QACvD,EAAe,KAAK,WAAW,GAC5B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAI,GAAiB,CAAC,EAAQ,GAAG,UAE/B,SAEF,EAAa,KACT,CAAC,KAAM,EAAQ,GAAG,aAAc,OAAQ,EAAa,KAE3D,MAAO,IAiCT,OAAA,eAAI,EAAA,UAAA,eAAY,KAIhB,WACE,MAAO,MAAK,mBALd,SAAiB,GACf,KAAK,cAAgB,mCAOvB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,AAAI,KAAK,aAAe,GACtB,MAAK,WAAa,EAClB,KAAK,iBAAmB,qCAI5B,EAAA,UAAA,QAAA,WACE,GAAM,GAAS,EAAA,UAAM,QAAO,KAAA,MAC5B,GAAI,EAAO,uBAAyB,GAAK,KAAK,WAAa,MACvD,KAAK,kBACP,GAAM,GAAmC,EAAA,SAAa,WACtD,KAAK,WAAW,UAChB,EAAO,sBACH,EAAmC,EAAA,SAAa,WAEtD,MAAO,IAGD,EAAA,UAAA,mBAAR,WAEE,GAAI,GAEJ,GAAI,MAAO,MAAK,MAAS,SACvB,EAAY,GAAY,KAAK,cACpB,MAAM,QAAQ,KAAK,OAC5B,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,OAAM,sDAGpB,EAAa,KAAK,KAAkB,IAAI,SAAA,GAAQ,MAAA,IAAY,UAG5D,GAAM,GAAc,OAAO,KAAK,KAAK,MACrC,EAAY,GAGZ,OAFM,GACF,KAAK,KACgB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,MAAO,GAAO,IAAgB,SAChC,EAAU,GACN,GAAY,EAAO,QAEvB,MAAM,IAAI,OAAM,uDAItB,MAAO,IAGD,EAAA,UAAA,qBAAR,WAEE,GAAI,MAAO,MAAK,SAAY,UACxB,MAAO,MAAK,SAAY,WAC1B,MAAO,CAAC,GAAY,GAA4B,KAAK,WAChD,GAAI,MAAM,QAAQ,KAAK,SAC5B,MAAO,MAAK,QAAQ,IAChB,SAAA,GAAU,MAAA,IAAY,GAA4B,MAEtD,GAAM,GAAyD,GAC/D,OAAW,KAAO,MAAK,QACrB,EAAmB,GACf,GAAY,GAA4B,KAAK,QAAQ,KAE3D,MAAO,IAID,EAAA,UAAA,kBAAV,WACE,MAAO,CACL,KAAM,KAAK,qBACX,QAAS,KAAK,uBACd,iBAAkB,CAChB,WAAY,KAAK,UAAU,eAC3B,OAAQ,KAAK,UAAU,eAQ7B,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,EAAe,kBAAoB,KACrC,KAAM,IAAI,OAAM,gDAElB,GAAI,EAAe,cAAgB,KACjC,KAAM,IAAI,OAAM,8CAElB,GAAI,EAAe,oBAAsB,KACvC,KAAM,IAAI,OAAM,oDAGlB,GAAM,GAAW,GAAoB,EAAe,kBAE9C,EAAY,GAAY,GAE1B,EACJ,GAAI,MAAO,GAAe,MAAS,SACjC,EAAO,GAAY,EAAe,cACzB,MAAM,QAAQ,EAAe,MACtC,EAAO,EAAe,KAAK,IAAI,SAAA,GAAa,MAAA,IAAY,aAC/C,EAAe,MAAQ,MAChC,EAAO,GACP,OAAW,KAAO,GAAe,KAC/B,EAAK,GAAO,GAAY,EAAe,KAAK,IAIhD,GAAI,GACJ,GAAI,MAAM,QAAQ,EAAe,SAC/B,EAAU,EAAe,QAAQ,IAAI,SAAA,GAAU,MAAA,IAAY,aAClD,EAAe,SAAW,MACnC,EAAU,GACV,OAAW,KAAO,GAAe,QAC/B,EAAQ,GAAO,GAAY,EAAe,QAAQ,IAItD,KAAK,QAAQ,CAAC,KAAI,EAAE,QAAO,EAAE,UAAS,KAoFlC,EAAA,UAAA,KAAN,SAAW,EAAmC,+HAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,GACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,GACN,gHAKF,MAAA,CAAA,EAAM,EAAA,GAAG,cAAc,KAAK,gBAAgB,kBAD1C,GACF,EAAA,OAEE,EAAe,GACf,EAAgB,KAChB,EAAc,KAAK,OAAO,EAAW,GACrC,EAAoC,CACxC,cAAe,EACf,OAAQ,GACR,YAAa,8BAA8B,GAC3C,YAAa,MAGT,EAAmB,GAAU,KAAO,GAAQ,EAAO,iBACrD,GAAoB,KAAK,WAAa,KACxC,GAAe,eAAiB,KAAK,oBAC/B,EAAa,YAET,EAAA,GAAA,EAAA,IAAG,cAAc,CAAA,EAAM,KAAK,UAAU,eAJ9C,CAAA,EAAA,UAIE,MAAA,CAAA,EAAM,EAAA,MAAA,EAAA,CAAiB,EAAA,OAAmC,YADxD,EACF,EAAA,OADS,EAAmB,EAAA,KAAS,EAAoB,EAAA,MAE7D,GAAA,EAAmB,OAAM,KAAI,MAAA,EAAI,GACjC,EAAmB,KAAO,EAAA,GAAG,wBACzB,CAAC,EAAmB,KAAM,qBAGhC,MAAI,MAAK,qBAAuB,MAExB,GAAY,GAClB,GAAyB,KAAK,oBAAqB,KAAK,KAAM,GAC9D,EAAe,oBAAsB,KAAK,qBAG5C,EAAe,WAAa,EAAmB,KAC/C,EAAe,YAAc,EAAmB,MAChD,CAAA,EAAO,EAAa,KAAK,UAW3B,EAAA,UAAA,uBAAA,SAAuB,GACrB,GAAyB,EAAqB,KAAK,MACnD,KAAK,oBAAsB,GAc7B,EAAA,UAAA,uBAAA,WACE,MAAO,MAAK,qBAv4CP,EAAA,UAAY,QAy4CrB,GA74CiC,IA84CjC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DACS,SAAA,UAAY,aACrB,GAFgC,IAGhC,EAAA,cAAc,cAAc,IC72D5B,YA8DI,EACA,uHACF,AAAM,iBAAmB,IACvB,GAAwB,CAAC,cAAe,IAE1C,EAAwB,EAEpB,EAAgB,EAAsB,cAC1C,AAAI,EAAc,cAAmB,MAMnC,GAAgB,EAAc,cAE1B,EACF,GAAoB,GAClB,EAAQ,GAAY,EAAU,GAEhC,EAAsB,iBAAmB,KAItB,CAAA,EAAM,EAAA,GAAG,YAC1B,EAAsB,gBAAiB,EAAsB,WAC7D,EAAM,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,iBANrC,CAAA,EAAA,UAUF,IANM,EAAe,EAAA,OAKf,EAAqC,GAC3C,EAAA,EAAqB,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,IAAV,EAAM,EAAA,GACf,EAAmB,EAAO,cACtB,EAAa,EAAO,cAG1B,EAAM,YAAY,GAElB,EAAA,QAAQ,oBAEV,MAAA,CAAA,EAAO,oBAsIL,EACA,4EAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,MAAO,IAAoB,UAE7B,GADM,EAAW,EAAA,GAAG,gBAAgB,EAAiB,GACjD,EAAS,SAAW,EAKtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAiB,YAC5C,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAe,MAE7B,EAAkB,EAAS,GAE7B,MAAA,CAAA,EAAO,GAA6B,EAAiB,OAAW,oBAc9D,EAAuB,EACvB,mHAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,EAAQ,MAAQ,KAClB,KAAM,IAAI,GACN,iHAGY,MAAA,CAAA,EAAM,EAAQ,kBAA1B,EAAY,EAAA,OACd,EAAgB,EAAU,cAC9B,AAAI,EAAc,cAAmB,MACnC,GAAgB,EAAc,cAG1B,EAAS,EAAQ,QAAU,KAAO,GAAO,EAAQ,OAMjD,EACF,EAAU,YAAc,MAAQ,EAAU,aAAe,MAAQ,EAC/D,EACF,GACI,GAAoB,GACpB,EAAe,GAEjB,EAAiB,EAAU,eACjC,AAAI,GAAkB,MACpB,EAAM,mBAAmB,GAE3B,AAAI,EAAU,qBAAuB,MACnC,EAAM,uBAAuB,EAAU,sBAIrC,GAAU,YAAc,MAAxB,MAAA,CAAA,EAAA,GAEF,GAAI,EAAU,aAAe,KAC3B,KAAM,IAAI,GACN,wHAIA,GAAmC,GACrC,EAAU,WAAY,EAAU,aAD7B,EAAY,EAAA,aAAE,EAAgB,EAAA,iBAErC,EAAM,YAAY,EAAc,GAE5B,EAAM,WAAa,MAAQ,EAAiB,OAAS,EACvD,CAAA,EAAM,EAAM,UAAU,WAAW,IAD/B,CAAA,EAAA,UACF,EAAA,wBAIF,EAAA,QAAQ,GACR,EAAA,QAAQ,EAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,2BAEtC,MAAA,CAAA,EAAO,QAGT,YACI,EAAqB,GAEvB,GAAM,GAAc,EAAA,GAAG,cAAc,EAAQ,GACvC,EAA+B,GAC/B,EAAkC,GACxC,SAAM,QAAQ,SAAA,GACZ,AAAI,EAAK,QAAU,YACjB,EAAiB,KAAK,CAAC,KAAM,EAAK,KAAM,OAAQ,EAAY,EAAK,QAEjE,EAAa,EAAK,MAAQ,EAAY,EAAK,QAGxC,CAAC,aAAY,EAAE,iBAAgB,sBAwCR,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CAAC,OAAQ,GAAI,QAAS,MAAI,KAUhC,GATA,EAAO,GAAQ,GAEf,EAAK,UAAY,GACjB,EAAK,MAAQ,GAGb,EAAK,KAAQ,EAAK,MAAQ,KAAQ,EAAK,KAAO,GAAO,eAGjD,EAAK,QAAU,KACjB,OAAoB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,IAAI,YAOP,SAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAQ,EAAM,aAAa,GAAG,cAAc,GAAG,MACrD,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,GAAI,IACtB,KAAM,IAAI,GACN,kDACG,GAAM,KAAI,uBACV,GAAM,aAAa,GAAG,aAAa,GAAG,MAAK,OAyBtD,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GACF,YAAiB,IAAc,YAAiB,IAChD,EACJ,GAAI,GAEF,GADA,EAAa,EACT,EAAW,QAAQ,SAAW,EAChC,KAAM,IAAI,GACN,yHAKN,GAAI,EAAW,OAAO,SAAW,EAC/B,KAAM,IAAI,GACN,uHAOR,GAAI,KAAK,QAAQ,SAAW,GAE1B,GAAI,EAAM,aAAa,SAAW,GAEhC,GAAI,EAAM,iBAAmB,KAC3B,KAAM,IAAI,GACN,iGAIN,GAAM,GAAI,GAAM,CACd,WAAY,EAAM,gBAClB,MAAO,EAAM,MACb,KAAM,EAAM,KAAO,WAIrB,EAAM,MAAM,GAGd,GAAI,EACF,KAAK,QAAU,EAAW,QAC1B,KAAK,OAAS,EAAW,YAEzB,GAAI,EAAM,aAAa,SAAW,EAChC,KAAM,IAAI,GACN,2DACA,yDACI,EAAM,KAAI,KACd,cAAa,EAAM,aAAa,OAAM,0BACtC,gBAGN,GAAI,EAAM,aAAa,GAAG,cAAc,SAAW,EACjD,KAAM,IAAI,GACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,EAAM,aAAa,GAAG,cAAc,IACpD,KAAK,OAAS,GAAgB,KAAK,QAAQ,IAG7C,KAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,KACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,KAAK,OACnB,cAAe,KAAK,QAEpB,WAAY,GAA2B,KAAM,KAAK,OAAO,QACzD,YAAa,CAAC,MACd,YAAa,KAAK,OAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QACpC,aAAc,KAAK,QAAQ,GAAG,aAGhC,GAAM,GAAe,EAAM,MAAM,KAAK,QAAQ,IAC9C,GAAI,MAAM,QAAQ,GAChB,KAAM,IAAI,WACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,GAEhB,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,OAGvD,KAAK,OAAO,KAAK,GACjB,KAAK,MAAQ,IAQf,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,OAAO,SAAW,EACzB,KAAM,IAAI,WAAU,qCAItB,GADA,KAAK,OAAO,MACR,KAAK,OAAO,SAAW,EACzB,KAAK,QAAU,GACf,KAAK,aAAe,GACpB,KAAK,cAAgB,QAErB,GAAM,GAAiB,KAAK,OAAO,OAAS,EAC5C,KAAK,OAAO,GAAgB,cAAgB,GAC5C,KAAK,QAAU,CAAC,KAAK,OAAO,GAAgB,QAE5C,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,SAIzD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,KAAK,EAAQ,IAGjC,EAAA,UAAA,MAAA,SAAM,GAKJ,GAFA,GAAmB,GAEf,KAAK,OAAO,SAAW,GAAK,KAAK,QAAQ,SAAW,EACtD,KAAM,IAAI,WACN,4EAIN,KAAK,MAAQ,GAAI,IAAY,CAC3B,OAAQ,KAAK,OACb,QAAS,KAAK,QAAQ,GACtB,KAAM,KAAK,KAAO,WAEpB,KAAK,MAAM,UAAY,KAAK,UAG5B,KAAK,gBAAkB,KAAK,MAAM,gBAElC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,uBAAyB,KAAK,MAAM,uBACzC,KAAK,yBAA2B,KAAK,MAAM,yBAC3C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,wBAA0B,KAAK,MAAM,wBAC1C,KAAK,0BAA4B,KAAK,MAAM,0BAC5C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,WAAa,KAAK,MAAM,WAG7B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,WACE,MAAK,MAAK,OACR,KAAK,QAEA,EAAA,UAAM,YAAW,KAAA,OAiC1B,EAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAAA,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC9D,AAAK,KAAK,OACR,KAAK,QAEP,EAAA,UAAM,QAAO,KAAA,KAAC,EAAY,EAAW,IASvC,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,OAAS,MAChB,KAAK,QAEP,KAAK,MAAM,WAAW,IAmCxB,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GACF,GADE,AAAA,IAAA,QAAA,GAAA,IACE,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAO,MAAK,MAAM,SAAS,EAAG,EAAG,IAyB7B,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAA,CAAA,EAAO,KAAK,MAAM,gBAAgB,EAAS,SA8B7C,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC1B,AAAI,KAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,QAAQ,EAAG,IAU/B,EAAA,UAAA,eAAA,SAAe,GACb,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,eAAe,IAQnC,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,QACL,KAAK,MAAM,QAAQ,GACnB,KAAK,WAAa,KAAK,MAAM,UAE7B,KAAK,iBAAoB,KAAK,MAAc,iBAC5C,KAAK,KAAO,KAAK,MAAM,KACvB,KAAK,QAAU,KAAK,MAAM,QAG1B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,aAAe,KAAK,MAAM,cAIjC,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAO,OAAY,KAAK,MAAM,eAGrD,SAAc,GACZ,KAAK,MAAM,UAAY,mCAkCnB,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,IAAI,EAAG,EAAG,SAwFxB,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,WAAW,EAAS,SA0BlC,EAAA,UAAA,aAAN,SACI,EACA,sEAEF,MAAA,CAAA,EAAO,KAAK,MAAM,aAAa,EAAG,SAK7B,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GACA,EAA6C,GACjD,GAAI,YAAkB,QACpB,GAAI,CAAE,GAAO,GAAG,WAAa,OACzB,EAAO,GAAG,YAAiB,QAC7B,KAAM,IAAI,GAAW,kDAEvB,EAAc,MAEd,GAAA,KAAK,OACD,EAAO,QAAa,KACpB,WACI,MAAA,wHAER,EAAc,EAAO,OACrB,MAAO,GAAO,OACd,EAAmB,EAGrB,GAAM,GAAQ,GAAI,GAAI,GACtB,GAAI,CAAE,aAAiB,IACrB,KAAM,IAAI,IACN,yDAAyD,GAE/D,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAA0C,OAC1C,EAAQ,GACI,EAAkC,EAClC,GAClB,AAAI,GACF,EAAM,6BAA6B,IAErC,EAAM,IAAI,GAEZ,MAAO,IA+BT,OAAA,eAAI,EAAA,UAAA,eAAY,KAWhB,WACE,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,MAAO,MAAK,MAAM,kBAjBpB,SAAiB,GAGf,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,KAAK,MAAM,aAAe,mCAe5B,EAAA,UAAA,UAAA,WAME,OADM,GAAqC,GACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAiC,GACvC,EAAK,UAAe,EAAM,eAC1B,EAAK,OAAY,EAAM,YACvB,EAAO,KAAK,GAEd,MAAO,CAAC,KAAM,KAAK,KAAM,OAAM,IAvsB1B,EAAA,UAAY,aAysBrB,GA3sBgC,IA4sBhC,EAAA,cAAc,cAAc,IC3kC5B,YAqEsB,GACpB,MAAO,IAAI,IAAY,eA+DE,GACzB,MAAO,IAAI,IAAW,eA+FpB,EACA,GACF,MAAI,IAAW,MACb,GAAU,IAEL,GAAwB,EAAiB,eA0B5B,GACpB,MAAO,IAAM,eAIX,EACA,GACF,GAA4B,4BACxB,EAAgB,GC5QtB,AAwBA,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAEE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GALyC,EAAA,cAAc,cAWvD,GAAA,SAAA,GAAyB,EAAA,EAAA,GAAzB,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,GAAM,EAAG,IATF,EAAA,UAAY,MAW9B,GAbyB,IAczB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAK,EAAA,KAAS,OAF9B,EAAA,UAAY,QAI9B,GAN2B,IAO3B,EAAA,cAAc,cAAc,IAG5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAFO,EAAA,UAAY,SAI9B,GAN4B,IAO5B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,QAAY,IAFL,EAAA,UAAY,UAI9B,GAN6B,IAO7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAAjC,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAc,IAFP,EAAA,UAAY,cAI9B,GANiC,IAOjC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,SAAa,IAFN,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAW,IAFJ,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAeE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,QAAY,EAAG,IAdR,EAAA,UAAY,UAgB9B,GAlB6B,IAmB7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DAgBE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,WAAe,EAAG,IAfX,EAAA,UAAY,aAiB9B,GAnBgC,IAoBhC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,EAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAE,IAAI,IAAQ,IAAI,MATlC,EAAA,UAAY,QAW9B,GAb2B,IAc3B,EAAA,cAAc,cAAc,gBAEQ,GAClC,MAAO,GAAW,2BAIjB,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACM,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,MAChB,GAAM,GAAmC,GACzC,SAAO,UAAe,SACtB,EAAO,OAAY,GACZ,GAAsB,GAE/B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCjQjC,AAiBA,YAA0B,GACxB,GAAI,GAAQ,MAAQ,MAAO,IAAS,SAClC,KAAM,IAAI,OACN,mEACA,0BAAyB,IAOjC,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DAEA,MAAA,IAF0C,EAAA,cAAc,cAqBxD,GAAA,SAAA,GAA0B,EAAA,EAAA,GAQxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAEP,UAAiB,GAEjB,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,MAAQ,EAAK,KAAO,EACzB,EAAK,MAAQ,EAAK,KAAO,IAO3B,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,GAAyB,EAAA,MAAM,CAAC,IACpC,MAAI,GAAK,OACP,GAAiB,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAI,OAEhE,AAAI,EAAK,OACP,GACI,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,GAAS,OAEjD,EAAe,cAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,GAAM,KAAK,GAAI,GAAM,KAAK,KAI7B,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,CAAC,GAAI,EAAO,GAAiB,GAAI,EAAO,MA3ClD,EAAA,UAAY,OA6CrB,GA/C0B,IAgD1B,EAAA,cAAc,cAAc,gBAET,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,gBAGvC,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,IAOnD,GAAM,IACyC,CAChD,KAAQ,oBAGuB,GAEnC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGU,GAG7B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAuB,OACzB,OAAI,aAAsB,IACxB,EAEA,GAAuB,GClJlC,AAmCA,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAKxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,MACV,GAAK,SAAW,EAAK,YAIzB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,EAAS,GAAoB,GAC7B,GAAI,GAAS,EAAA,KAAK,GAClB,MAAI,MAAK,UAAY,MACnB,GAAS,EAAA,YAAY,EAAQ,EAAG,KAAK,WAEhC,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,SAAU,KAAK,UACnD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA5BF,EAAA,UAAY,OA8BrB,GAhC0B,IAiC1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAO7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,GAIvB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,UAAU,EAAG,KAAK,QAG3B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,YA4BrB,GA9B+B,IA+B/B,EAAA,cAAc,cAAc,IA6B5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAWzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAU/B,GAbO,EAAA,0BAAmD,QAI1D,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,gBAAkB,GACvB,EAAK,iBACD,GAAe,EAAK,kBAAoB,EAAK,2BACjD,EAAK,iBAAmB,GAAe,EAAK,kBAC5C,EAAK,gBAAkB,GAAc,EAAK,iBACtC,EAAK,YAAc,KACrB,EAAK,WAAa,aACT,MAAM,QAAQ,EAAK,YAC5B,EAAK,WAAa,EAAK,mBACd,MAAO,GAAK,YAAe,SACpC,EAAK,WAAa,CAAC,EAAK,gBAExB,MAAM,IAAI,GACN,8DACA,YAAW,EAAK,sBAIxB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAChC,GAAM,GAAoB,EAAW,MAAM,GAC3C,GAAI,KAAK,YAAc,KACrB,OAAgB,GAAA,EAAA,EAAA,KAAK,WAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAW,EAAI,GAAK,EAGxB,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAAM,KAAK,iBAEtC,GAAM,GAAiC,GACvC,GAAI,KAAK,YAAc,KACrB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,EAAK,GAAK,EAAW,GAGzB,KAAK,UAAY,CAAC,GAAI,IAAU,CAC9B,KAAM,EAAW,OACjB,KAAI,KAEN,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,SAAS,GAAoB,GACtB,EAAA,MAAM,EAAQ,KAAK,MAAM,SAGlC,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,iBAAkB,GAAqB,KAAK,kBAC5C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAoB,KAAK,iBAC1C,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAxEF,EAAA,UAAY,QA0ErB,GA5E2B,IA6E3B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAK/B,GARO,EAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGL,EAAK,OAAS,MAAQ,EAAK,QAAU,EAAK,cAC5C,KAAM,IAAI,IACN,4BAA4B,EAAK,MAAK,4CAI5C,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,IAAI,IAGb,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjCF,EAAA,UAAY,MAmCrB,GArCyB,IAsCzB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAOnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAE,IAAI,GAAK,EAAE,QAAQ,KAAK,OAAQ,aAG3C,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,kBA6BrB,GA/BqC,IAgCrC,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,aAAe,EAItB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,QAAU,GAAI,MAAoB,MACvC,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,OAG3D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,MAAK,QAAQ,EAAG,KAAK,OAG9B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,KAAM,KAAK,MAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,UA6BrB,GA/B6B,IAgC7B,EAAA,cAAc,cAAc,ICxV5B,YAwBI,EAAwB,EAAW,GACrC,GAAI,MAAO,IAAU,SACnB,MAAO,IAAa,EAAO,GAE3B,GAAI,EAAM,SAAW,EACnB,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EAAC,aACxD,eAAc,EAAM,OAAM,eAEhC,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAc,EAAM,GAC1B,GAAI,CAAC,GAAU,GACb,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EACvD,yBAAwB,KAAK,UAAU,GAAM,gBAC7C,wBAAuB,IAG/B,MAAO,eAaP,EAAqB,EAAoB,EACzC,EAAgB,GAClB,GADkB,AAAA,IAAA,QAAA,GAAA,GACd,GAAe,KACjB,MAAO,GAET,GAAM,GAAoB,EAAc,GAAa,GAAM,GAAW,GAClE,EACJ,MAAI,KAAY,OACd,EAAe,EAEf,EAAe,EAAc,EAAoB,EAE5C,KAAK,MAAO,GAAe,EAAS,GAAK,eAI9C,EAAiB,EAAoB,EACrC,GACF,GAAI,GAAW,KACb,MAAO,MAGT,GAAI,IAAY,QACd,EAAU,EAAU,EAAa,GAAI,CAAC,EAAa,EAAY,YACtD,IAAY,OACrB,EAAU,EAAU,MAEpB,MAAM,IAAI,GAAW,2BAA2B,EAAO,KAEzD,MAAO,GCpFT,YAwCI,EAAW,GAEb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,IAE3B,gBAWT,EAAW,GACb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,IAE9B,gBAsBT,EAAW,EAAgB,EAAc,EAAa,EACtD,EAAyB,GADgB,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,SAC7B,AAAA,IAAA,QAAA,GAAA,GACpB,EAAA,KAAK,WAMV,GALA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAEZ,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,+DACG,GAAE,MAAM,OAAM,cAEvB,GAAI,EAAO,MAAM,SAAW,EAC1B,KAAM,IAAI,GACN,iEACG,GAAO,MAAM,OAAM,aAE5B,GAAI,GAAQ,MAAQ,EAAK,MAAM,SAAW,EACxC,KAAM,IAAI,GACN,+DACG,GAAO,MAAM,OAAM,aAM5B,GAHA,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,KAE1B,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,GAAI,GAAY,EAAA,OACZ,EAA0B,EAAoB,EAC9C,IAAY,OAAS,OAAS,QAAS,MAAO,GAClD,MAAI,IAAQ,MACV,GAAI,GAAU,EAAG,IAEZ,gBAoDP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EAAyB,EAC5C,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,IACvD,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,gEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,iEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,MAAU,OAAO,CACnB,EAAG,EACH,OAAQ,EACR,QAAS,EACT,IAAK,IAAY,OAAS,OAAS,QACnC,UAAW,EACX,WAAY,OACZ,KAAI,EACJ,WAAU,IAEZ,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBA8BP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EACnB,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,EAAG,IAC1D,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,MAEf,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,oEACG,GAAE,KAAI,MAEf,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,OACA,EACA,EAAmC,EACnC,IAAY,OAAS,OAAS,QAAS,QAAS,GACpD,AAAI,GAAQ,MACV,GAAI,GAAU,EAAG,IAEnB,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAmHX,GAAA,IAAA,SAAA,GAAuC,EAAA,EAAA,GAwBrC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,IAAkB,KAIxB,GAVQ,EAAA,KAAsB,KAEvB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAIzD,EAAS,WAAW,GACpB,EAAK,KAAO,EACZ,GAAoC,EAAK,KAAM,QAC3C,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,EAAK,OAAS,EACtD,KAAM,IAAI,IACN,qDACI,EAAK,KAAI,6BAqBnB,GAlBA,EAAK,WAAa,GAAe,EAAK,WAAY,EAAM,cACxD,EAAK,QAAU,GACX,EAAK,SAAW,KAAO,EAAI,EAAK,QAAS,EAAM,WACnD,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,WAAa,GAAc,EAAK,YACrC,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAClD,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,aAAe,GAChB,EAAK,cAAgB,KAAO,EAAI,EAAK,aAAc,EACnD,gBACA,EAAK,OAAS,GACb,MAAM,QAAQ,EAAK,eAAiB,EAAK,aAAa,SAAW,EACpE,KAAM,IAAI,GACN,iGAEA,IAAG,KAAK,UAAU,EAAK,gBACtB,GAAI,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aAAe,CAAC,EAAK,aAAc,EAAK,sBACpC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,gEACA,8BAA6B,KAAK,UAAU,EAAK,wBAE9C,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aACD,CAAC,EAAK,aAAc,EAAK,aAAc,EAAK,sBACvC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,kEACA,8BAA6B,KAAK,UAAU,EAAK,yBAK1C,SAAA,WAAjB,SAA4B,GAI1B,GAFA,GACI,cAAgB,GAAM,2CACtB,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,oEACA,oCACI,KAAK,UAAU,EAAK,YAAW,OAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,gBAAiB,GAAqB,KAAK,iBAC3C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA/GuC,IAqHvC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAcjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAA0B,KAZ9B,SAAA,OAAwB,KAahC,EAAK,WAAW,GAChB,EAAK,QAAU,EAAK,QACpB,GAAoC,EAAK,QAAS,WAClD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,kBAAoB,GAAe,EAAK,qBAG/C,SAAA,UAAA,MAAA,SAAM,SACJ,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAE1B,GAAM,GAAW,EAAW,GAEtB,EAAc,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,UAE3D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,KAAM,KAAK,gBACnC,KAAK,gBAAiB,GAAM,KAAK,iBAGvC,KAAK,UAAY,CAAC,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,KACtE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GACE,EAAY,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACjD,EAAsB,GACU,EAAK,WAAW,gBAEtD,GAAI,GAAuB,MAAQ,EAAK,OAAS,EAC/C,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,aACtB,QAEJ,GAAI,EAAK,OAAS,EAChB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAQ,GACpD,EAAK,QAAS,EAAK,WAAY,EAAK,aAAa,YAC5C,EAAK,OAAS,EAEvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,sBACjB,EAAK,OAAS,EACvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,kBAE1B,MAAM,IAAI,IACN,yDAGN,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAIpC,MAAO,MAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAKhC,OAJM,GAAqB,GACrB,EAAS,KAAK,aAAe,eAC/B,EAAW,MAAM,EAAG,EAAW,OAAS,GACxC,EAAW,MAAM,GACZ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAM,GAAS,GACX,EAAM,GAAI,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,GACzD,MAAO,MAAK,cAAiB,SAAW,KAAK,aACL,KAAK,aAAa,IAC9D,EAAS,KAAK,GAGhB,GAAI,GAAc,CAAC,EAAW,IAC9B,MAAI,MAAK,aAAe,eACtB,GAAc,EAAY,OAAO,GACjC,EAAY,KAAK,KAAK,UAEtB,GAAY,KAAK,KAAK,SACtB,EAAc,EAAY,OAAO,IAE5B,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,kBAAmB,GAAqB,KAAK,mBAC7C,iBAAkB,GAAoB,KAAK,mBAEvC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,CAAE,YAAa,KAAS,MAAO,GAAK,SAAY,UAChD,EAAK,QAAU,EACjB,KAAM,IAAI,GACN,kEACA,YAAW,KAAK,UAAU,EAAK,YAGzC,GA1ImC,IA4InC,GAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAK,MAAO,GAAK,YAAe,UAC5B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,gCAA+B,KAAK,UAAU,EAAK,YAAW,OAnB/D,EAAA,UAAY,SAsBrB,GAxB4B,IAyB5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UACzB,CAAE,OAAM,QAAQ,EAAK,aAClB,GAAK,WAAW,SAAW,GAAK,EAAK,WAAW,SAAW,IAChE,KAAM,IAAI,GACN,mDACA,4CACI,KAAK,UAAU,EAAK,YAAW,OApBtC,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAGX,GAFA,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KAEnC,EAAK,UAAY,QAAU,EAAK,UAAY,QAC9C,KAAM,IAAI,GACN,gEACA,2CAA0C,EAAK,mBAIvD,SAAA,UAAA,MAAA,SAAM,SAGJ,GAFA,EAAa,GAAmB,GAE5B,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,mDACA,KAAK,UAAU,IAGrB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,wEAGN,GAAM,GAAW,EAAW,GACtB,EAAc,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,IAE1D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,UAAW,KAAK,kBACvC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,GAAM,KAAK,iBAIvC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MAC3D,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAM,MAAM,SAAW,EACzB,KAAM,IAAI,GACN,iEACA,8BAA6B,EAAM,MAAM,SAG/C,GAAM,GAAa,EAAM,MACnB,EAAY,EAAW,GAEzB,EACA,EACJ,AAAI,EAAK,aAAe,gBACtB,GAAQ,EACR,EAAQ,GAER,GAAQ,EACR,EAAQ,GAGV,GAAM,GAAS,EAAW,GACpB,EAAQ,EAAW,GACnB,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,QAAQ,GACvB,EAAU,EAAK,QAAQ,GAGvB,EAAY,GAAa,EAAQ,EAAS,EAAS,EAAK,SACxD,EAAW,GAAa,EAAO,EAAS,EAAS,EAAK,SAMtD,EACF,CAAC,EAAW,EAAW,EAAU,EAAK,SAE1C,AAAI,EAAK,aAAe,gBACtB,GAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,KAEzC,GAAI,GAAU,EAAA,gBACV,EAAmB,EAAK,OAAO,OAAoB,EACnD,EAAK,QAA6B,EAAK,SAC3C,MAAI,GAAK,aAAe,gBACtB,GAAU,EAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,KAG7C,AAAI,EAAK,MAAQ,MACf,GACI,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEhD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAE3B,EACA,EACA,EACJ,AAAI,KAAK,aAAe,gBACtB,GAAc,EACd,EAAa,EACb,EAAY,GAEZ,GAAc,EACd,EAAa,EACb,EAAY,GAGd,GAAM,GAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,QAAQ,GACvB,EAAU,KAAK,QAAQ,GAE7B,SAAY,GAAe,KAAK,QAChC,EAAY,GACR,GAAa,EAAY,GAAa,EAAS,EAAS,KAAK,SACjE,EAAY,GACR,GAAa,EAAY,GAAY,EAAS,EAAS,KAAK,SACzD,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,aACP,GA9IF,EAAA,UAAY,kBAgJrB,GAlJqC,IAmJrC,EAAA,cAAc,cAAc,IA0C5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAqBjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAAO,KAEnB,GAXO,EAAA,8BACL,gBACK,EAAA,8BACL,gBAEM,EAAA,gBAAiC,KACjC,EAAA,gBAAiC,KAKrC,EAAO,SAAW,KACpB,KAAM,IAAI,GACN,uFAGN,GAAI,EAAO,mBAAqB,MAAQ,EAAO,mBAAqB,MAChE,EAAO,kBAAoB,KAC7B,KAAM,IAAI,GACN,sPAKN,GAAI,EAAO,SAAW,MAAQ,EAAO,UAAY,QAC7C,EAAO,UAAY,QACrB,KAAM,IAAI,GACN,gBAAgB,EAAK,KAAI,kCACzB,qCAAoC,KAAK,UAAU,EAAO,WAGhE,SAAK,gBACD,EAAO,iBAAmB,KAAO,EAAI,EAAO,gBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,qBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,uBAGlD,SAAA,UAAA,MAAA,SAAM,SAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,KAAK,KAAO,EAClC,KAAM,IAAI,GACN,0BAA0B,KAAK,KAAI,sBAChC,MAAK,KAAO,EAAC,gCAChB,IAAG,KAAK,UAAU,KAExB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,0DACA,cAAa,KAAK,UAAU,EAAW,MAO7C,OAJM,GAAW,EAAW,GACtB,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,kBACrC,EAAuB,GACpB,EAAI,EAAG,EAAI,KAAK,KAAM,EAAE,EAC/B,EAAqB,KAAK,GAE5B,EAAqB,KAAK,EAAW,KAAK,gBAAiB,KAAK,SAEhE,GAAM,GAAY,GAClB,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,EAAW,KAAK,gBAE1C,KAAK,KAAO,KAGd,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MACvE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAE7B,GAAI,GACJ,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,IACN,oDACC,MAAI,GAAK,OAAS,GACvB,CAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAG3C,EAAS,EAAA,gBACL,EAAoB,EAAK,gBAAgB,OACzC,EAAK,gBAAgB,OACrB,EAAK,QAA6B,EAAK,QACvC,EAAK,aAAkC,SAG7C,AAAI,EAAK,SACP,GAAS,GAAU,EAAQ,EAAK,KAAK,OAAQ,EAAK,aAEpD,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAGjC,AAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAEpC,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,kBACd,MAAO,GAAO,kBACd,MAAO,GAAO,iBACd,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,qBAC7B,EAAO,oBACH,GAAoB,KAAK,qBACtB,GAzJF,EAAA,UAAY,gBA2JrB,GA7JmC,IA+JnC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,SACV,GAAA,KAAA,KAAM,EAAG,IAAK,KAFT,SAAA,UAAY,kBAIrB,GANqC,IAOrC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,GAClB,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,WACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,2BAA0B,KAAK,UAAU,EAAK,YAAW,OArB1D,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAgC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAM9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,MAAI,OAAO,GAAK,UAAa,SAC3B,EAAK,SACD,CAAC,CAAC,EAAK,SAAU,EAAK,UAAW,CAAC,EAAK,SAAU,EAAK,WACrD,AAAI,MAAO,GAAK,SAAS,IAAO,SACrC,EAAK,SAAW,CACd,CAAC,EAAK,SAAS,GAAI,EAAK,SAAS,IACjC,CAAC,EAAK,SAAS,GAAc,EAAK,SAAS,KAG7C,EAAK,SAAW,EAAK,SAEvB,EAAK,WACD,EAAK,aAAe,OAAY,eAAiB,EAAK,WAC1D,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAI,MAAK,aAAe,gBACf,CACL,EAAW,GAAI,EAAW,GAC1B,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,IAGlD,CACL,EAAW,GACX,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GAAI,EAAW,KAK5E,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAFA,EAAS,GAAoB,GAEzB,EAAK,aAAe,gBACtB,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,QAEjE,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,OAKvE,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,SAAU,KAAK,SAAU,WAAY,KAAK,YACpD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAhEF,EAAA,UAAY,aAkErB,GApEgC,IAqEhC,EAAA,cAAc,cAAc,IAwB5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAOhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KALM,SAAA,aAAe,CAAC,EAAG,GAMpC,EAAK,UAAY,CAAC,CAAC,KAAM,IACzB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,aAGtD,SAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,KAAK,aAAe,iBACtB,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,QAE9C,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAO,EAAW,MAIrD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAC1B,EAAa,EAAM,MAEzB,GAAI,EAAK,aAAe,iBACtB,EAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,IACvC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GAClC,EAAU,EAAM,sBAAsB,CAAC,EAAQ,IACrD,MAAO,GAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,SAExC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GACxC,MAAO,GAAM,sBAAsB,CAAC,EAAQ,QAKlD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,KAAM,KAAK,KAAM,WAAY,KAAK,YAC5C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlDF,EAAA,UAAY,eAoDrB,GAtDkC,IAuDlC,EAAA,cAAc,cAAc,ICrvC5B,YA6CI,EAAW,EAAyB,EACpC,EAAmB,EACnB,GAFoC,MAAA,KAAA,QAAA,GAAA,CAA6B,EAAG,IACpE,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WACV,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAChB,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,OAEf,GAAI,EAAgB,OAAS,EAC3B,KAAM,IAAI,GACN,yDACG,GAAgB,KAAI,OAE7B,SAAI,EAAA,gBACA,EAAe,EAA6B,EAC5C,IAAY,OAAS,OAAS,QAAS,OAAQ,GACnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,IAsCX,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAsB,KAHzB,SAAA,gBAAiC,KAIvC,EAAK,gBACD,EAAK,iBAAmB,KAAO,EAAI,EAAK,gBAC5C,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,4BACtC,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,qBAAuB,GAAe,EAAK,wBAGlD,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,iDACA,0BAAyB,KAAK,UAAU,GAAW,MAEzD,GAAM,GAAc,KAAK,aAAe,gBAAkB,EAAI,EAC9D,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,iEACA,4BAA2B,EAAW,GAAY,OAExD,GAAM,GAAW,EAAW,GACtB,EAA8B,CAClC,KAAK,WAAW,GAAI,KAAK,WAAW,GAAI,EAAU,KAAK,iBAGzD,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,EAAW,KAAK,iBAAkB,KAAM,KAAK,gBACtD,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GAAU,GACV,EAAQ,EAAK,gBAAgB,OAAQ,EAAK,QAC1C,EAAK,QAAS,EAAK,WAAY,MAEnC,MAAI,GAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEtD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EAAa,KAAK,aAAe,gBACnC,EAAW,GAAK,KAAK,gBACrB,EAAW,GAAK,KAAK,gBACnB,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACnD,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,MAAI,MAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAY,EAAS,GAGrC,CAAC,EAAW,GAAI,EAAS,EAAS,IAI7C,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,SAAO,gBAAqB,KAAK,gBACjC,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,sBACtB,GAjGF,EAAA,UAAY,kBAmGrB,GArGqC,IAsGrC,EAAA,cAAc,cAAc,IClN5B,YA0DI,EACA,EACA,EACA,GAKF,GAAI,MAAM,QAAQ,IAChB,GAAI,GAAgB,MAAQ,GAAa,KACvC,KAAM,IAAI,GACN,iFAGN,AAAI,GAAgB,MAClB,GAAY,EAAO,MAAM,EAAO,OAAS,EAAc,EAAO,QAC9D,EAAS,EAAO,MAAM,EAAG,EAAO,OAAS,IAE3C,AAAI,EAAO,OAAS,GAClB,GAAe,EAAO,MAAM,EAAG,EAAO,SAExC,EAAS,EAAO,GAGlB,WAAsB,GAEpB,MAAI,IAAK,MAAQ,MAAM,QAAQ,GACtB,EAEA,CAAC,GAIZ,SAAe,EAAa,GAC5B,EAAY,EAAa,GAElB,CAAC,OAAM,EAAE,aAAY,EAAE,UAAS,eA+CrC,EAA+B,EAAgB,EAC/C,EAAqB,EAAe,EAAsB,EAC1D,GADA,MAAA,KAAA,QAAA,GAAA,IAA0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,IACK,EAAA,KAAS,WACd,GAAM,GAAO,EAAO,MAAM,OAC1B,GAAI,EAAO,EACT,KAAM,IAAI,GAAW,uCAAuC,EAAI,MAKlE,GAAM,GAAO,CAAC,EAAG,GAAG,OAAO,GAAiB,EAAG,IAG/C,GAFA,EAAS,EAAA,UAAc,EAAQ,GAE3B,GAAa,KACf,KAAM,IAAI,IACN,kFAKN,AAAI,GACF,QAAQ,KACJ,qGAIN,AAAI,GAAQ,MACV,GAAO,EAAK,OAAO,QAAQ,OAAO,WAClC,AAAI,EAAK,OAAS,EAAO,GACvB,GAAO,EAAA,WAAe,EAAM,KAE9B,EAAO,EAAA,UAAc,EAAM,IAG7B,AAAI,GACF,GAAS,EAAA,QAAY,EAAQ,GAC7B,AAAI,GAAQ,MACV,GAAO,EAAA,QAAY,EAAM,KAc7B,GAAM,GAA2B,GAC7B,EACA,EAAS,EACP,EAAY,EAAO,MAAM,GACzB,EAAgB,EAAA,QAAY,GAC9B,EACJ,AAAI,GAAQ,MACV,GAAe,EAAA,QAAY,IAG7B,mBAAS,GACP,GAAM,GAAe,EAAc,GAC7B,EAAc,EAAA,KAAS,WAAM,MAAA,GAAa,EAAc,KAE9D,GAAI,GAAQ,KACV,EAAa,EAAY,GACzB,EAAS,EAAY,QAErB,GAAM,GAAgB,EAAA,KAAS,WAC7B,GAAM,GAAW,EAAa,GACxB,EAAc,EAAA,SAAa,GAAU,IAAI,GAEzC,EACF,EAAY,GAAG,IAAI,GAAU,IAAI,EAAO,GAAG,IAAI,IAC7C,EAAY,EAAO,IAAI,SAAC,EAAO,GACnC,MAAO,GAAY,GAAG,GAAG,IAAI,GAAU,IAAI,EAAM,IAAI,MAEvD,MAAO,CAAC,OAAM,EAAE,UAAS,KAE3B,EAAa,EAAc,OAC3B,EAAS,EAAc,UAGzB,AAAI,GACF,EAAe,KAAK,IAxBf,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,GA2BT,GAAI,GACJ,GAAI,GACF,GAAM,GAAO,EACb,EAAU,EAAA,MAAU,EAAgB,GAEtC,MAAO,CAAC,EAAY,EAAS,wBAyGR,EAAA,EAAA,GAqBvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACP,EACJ,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,GACN,wDAMN,GALO,AAAI,MAAM,QAAQ,EAAK,MAC5B,EAAO,GAAI,IAAgB,CAAC,MAAO,EAAK,OAExC,EAAO,EAAK,KAEV,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,qGAGN,SAAK,KAAO,EACZ,EAAK,gBACD,EAAK,iBAAmB,KAAO,GAAQ,EAAK,gBAChD,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,UAAY,EAAK,UAAY,KAAO,GAAQ,EAAK,SACtD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAQ,EAAK,OAEjD,EAAK,gBAAkB,GACvB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KACvC,EAAK,UAAY,KACjB,EAAK,QAAU,KAEf,EAAK,aAAe,KAIpB,EAAK,WAAa,KAKpB,SAAA,UAAA,UAAA,WACE,GAAI,KAAK,SAAW,MAClB,GAAM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EACtE,MAAO,IAAiB,EAAG,GAAW,IAAI,SAAA,GAAK,MAAA,YAE/C,OAAO,MAAK,SAMhB,EAAA,UAAA,UAAA,SAAU,GACR,KAAK,QAAU,GAGjB,EAAA,UAAA,mBAAA,SAAmB,GACjB,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAGb,GAAI,GAAY,KAAK,KAAK,UAC1B,AAAK,MAAM,QAAQ,IACjB,GAAY,CAAC,IAEf,GAAM,GAAY,EAAU,GACxB,EAOJ,GANA,AAAI,KAAK,gBACP,EAAc,CAAC,EAAW,GAAI,EAAW,GAAI,GAE7C,EAAc,CAAC,EAAW,GAAI,GAG5B,KAAK,aAEP,OADM,GAAsB,GACV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAW,KAAK,CAAC,EAAW,GAAI,IAElC,MAAO,CAAC,GAAa,OAAO,OAE5B,OAAO,IAIX,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAS,WACd,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAM,GAAa,EAAK,gBAAkB,EAAO,KAEjD,GAAI,EAAK,aACP,GAAM,GAAY,EAAK,OAAO,IAAI,SAAA,GAAK,MAAA,QACvC,MAAO,CAAC,GAAY,OAAO,OAE3B,OAAO,MAWb,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,SAAW,MAIlB,OAHM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EAChE,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,EAAO,KAAK,MAEd,MAAO,OAEP,OAAO,MAAK,aAIhB,SAAW,GACT,KAAK,QAAU,mCAGV,EAAA,UAAA,MAAP,SAAa,GAIX,GAAI,KAAK,cAAgB,KACvB,KAAM,IAAI,IACN,oDAGN,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAEb,GAAM,GAAoB,KAAK,SAAW,EAAW,GAAK,KACpD,EAAW,EAAW,MAAM,GAClC,KAAK,UAAU,GAAK,GAAI,IAAU,CAAC,MAAK,CAAG,EAAW,MAAI,OAAK,KAI/D,GAAM,GAAiB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAK7D,KAAK,KAAK,MAAM,GAIlB,GAAI,GAOJ,GANA,AAAI,MAAM,QAAQ,KAAK,KAAK,WAC1B,EAAY,KAAK,KAAK,UAEtB,EAAY,CAAC,KAAK,KAAK,WAGrB,KAAK,WAAa,MACpB,GAAI,CAAC,EAAA,KAAK,YACF,KAAK,UAAU,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,EAAK,MAAM,OAAS,KAC1D,GACN,KAAM,IAAI,GACN,0DACA,uCAAsC,KAAK,UAAS,MACpD,8BAA6B,KAAK,KAAK,gBAG7C,MAAK,UACD,EAAU,IAAI,SAAA,GAAO,MAAA,IAAI,IAAU,CAAC,MAAO,CAAC,KAAM,OAExD,AAAI,KAAK,UACP,KAAK,eAqBT,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAK,WACH,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAEN,GAAM,GAAY,EAAK,UAAU,GAAG,MAAM,GAC1C,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUAQN,GAAI,EAAK,SAAW,KAClB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAU,CAAC,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,qBAEzC,GAAU,KAEnB,EAAA,QAAY,EAAK,SAEjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAQ,GAAK,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,iBAMpD,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAER,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,IAAa,GAKf,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GACf,EAAM,MAAM,QAAQ,EAAK,KAAK,WAChC,EAAK,KAAK,UAAU,GACpB,EAAK,KAAK,UACR,EAAgB,CAAC,EAAW,GAClC,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAEhB,EAAK,QAAQ,GAAS,GAG1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAI5D,EAAA,UAAA,MAAA,SACI,EACA,GAEF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAGX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAC1D,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAMzB,GAAI,GAAiD,GACjD,EAA+B,GACnC,GAAI,GAAgB,MAClB,EAAO,aAAkB,EACzB,EAAmB,EAAiB,OAAO,GAC3C,KAAK,UAAY,GACjB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,UAAU,KAAK,GAAI,IAAU,CAAC,MAAO,EAAM,SAKlD,EAAkB,EAAgB,OAAO,KAAK,WAEhD,AAAI,GAAa,MACf,GAAO,UAAe,EACtB,EAAmB,EAAiB,OAAO,GAE3C,KAAK,aAAe,EAAU,QAGhC,GAAM,GAAW,EAAiB,YAAc,IAChD,GAAI,GAEF,GAAM,GACF,CAAC,GAAQ,OAAO,GACd,EAAgB,KAAK,UAAU,OAAO,GAEtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GAAS,EAAA,UAAM,MAAK,KAAA,KAAC,EAAW,GACtC,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAK/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KAIE,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC5C,EACA,GAAU,KAAO,KAAO,EAAO,aAEnC,EAAS,GAAoB,GAC7B,AAAI,GAAgB,MAClB,CAAI,EAAK,SACP,EAAe,EAAK,QAEpB,EAAe,EAAK,gBAAgB,IAIxC,GAAM,GACF,MAAM,QAAQ,EAAK,KAAK,WAAa,EAAK,KAAK,UAAU,OAAS,EACtE,GAAI,EAAa,SAAW,EAC1B,KAAM,IAAI,GACN,iBAAiB,EAAS,4BACvB,GAAa,OAAM,uBAE5B,AAAI,EAAK,QACP,QAAQ,KACJ,oEAGN,GAAM,GAAyB,CAAC,SAAQ,GAGlC,EAAO,SAAC,EAAgB,GAG5B,GAAM,GACF,EAAK,KAAK,KAAK,CAAC,GAAQ,OAAO,GAAS,GAE5C,MAAO,CAAC,EAAQ,GAAI,EAAQ,MAAM,KAK9B,EACF,GAAI,EAAM,EAAQ,EAAc,EAAK,YAAa,EAAM,KACpD,EAAK,OAAQ,EAAK,iBACpB,EAAa,EAAW,GACxB,EAAU,EAAW,GACrB,EAAS,EAAW,GAE1B,AAAI,EAAK,UACP,EAAK,YAAY,EAAQ,GAG3B,GAAM,GAAS,EAAK,gBAAkB,EAAU,EAIhD,MAAI,GAAK,YACA,CAAC,GAAQ,OAAO,GAEhB,KAKb,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAAI,GAAe,EAAA,MAAU,EAAO,OAKpC,MAHA,GAAe,EAAA,IAAQ,EAAc,CAAC,EAAG,IACzC,EAAe,GAAa,GAExB,MAAM,QAAQ,EAAK,KAAK,WACnB,EAAK,KAAK,UAAU,IACvB,SAAA,GAAO,MAAA,GAAM,EAAI,GAAO,EAAc,CAAC,EAAG,IAAQ,IAE/C,EAAK,KAAK,UAAY,EACzB,CAAC,GAAO,EAAc,CAAC,EAAG,EAAK,KAAK,aACpC,CAAC,MAKX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAK,MAAK,UAIH,KAAK,KAAK,iBAHR,oCAMX,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,MAAK,MAAK,UAGH,KAAK,KAAK,oBAFR,KAAK,KAAK,yCAKrB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,MAAQ,MACf,KAAK,KAAK,6BAA6B,IAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,gBAAiB,KAAK,gBACtB,YAAa,KAAK,YAClB,YAAa,KAAK,YAClB,SAAU,KAAK,SACf,OAAQ,KAAK,QAGf,AAAI,KAAK,cAAgB,MACvB,GAAO,aAAkB,KAAK,cAGhC,GAAM,GAAa,KAAK,KAAK,YAE7B,MAAI,MAAK,iBAAmB,EAAI,WAC9B,GAAO,KAAU,CACf,UAAa,KAAK,KAAK,eACvB,OAAU,IAKd,GAAA,GAAW,EAAe,EAAe,IAIpC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAa,EAAO,KACpB,EAAO,GAAY,EAAY,GACrC,MAAO,IAAI,GAAI,OAAO,OAAO,EAAQ,CAAC,KAAI,MArfrC,EAAA,UAAY,MAufrB,GAzfyB,IA0fzB,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAAtC,4DAQA,MAAA,IARsC,IAwFtC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAiCjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KANJ,SAAA,mBAAqB,OACrB,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAIzD,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,YAAc,KAAO,EAAK,mBAAqB,EAAK,YAC7D,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAG9B,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAEhC,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAW,EAAW,OAAS,GAAI,KAAK,OAAQ,KAC3D,KAAK,kBAAmB,KAAK,kBAAmB,GAChD,KAAK,kBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,OAAQ,KAC9C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IASf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8CAA8C,EAAO,OAAM,KAEjE,GAAI,GAAa,EAAO,GACxB,EAAS,EAAO,GAChB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,KAG/B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,KAGxC,GAAI,GACE,EAAiB,EAAK,YACtB,EAAoB,EAAK,qBAC/B,AAAI,GAAU,KACZ,EAAI,GAAM,EAAA,IAAQ,EAAQ,GAAS,EAAK,OAAO,QAE/C,EAAI,GAAM,EAAQ,EAAK,OAAO,QAEhC,AAAI,EAAK,MAAQ,MACf,GAAI,GAAU,EAAG,EAAK,KAAK,SAE7B,AAAI,GAAa,MACf,GAAa,EAAA,IAAQ,EAAY,IAEnC,GAAI,GAAS,EAAA,IAAQ,EAAG,GAAM,EAAY,EAAK,gBAAgB,SAC/D,MAAI,GAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAI1B,CAAC,EAAQ,MAIpB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,kBAGzB,MAAA,IAAA,GAAW,EAAe,IArKrB,EAAA,UAAY,gBAuKrB,GAzKmC,IA0KnC,EAAA,cAAc,cAAc,IA2F5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAG7B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAK,KAAO,GAAI,IAAc,GAC9B,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IA7BV,EAAA,UAAY,YA+BrB,GAjC+B,IAkC/B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAqC3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAbO,EAAA,mBAAqB,OACrB,EAAA,6BAAqD,cAErD,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAQrD,EAAK,WACP,KAAM,IAAI,GACN,+DAEN,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,KAAK,gBACrC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,uDACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACzD,EAAW,EAAO,GACtB,EAAS,EAAO,GAKhB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EAAY,EAAK,qBACnB,EACA,EACA,EAEJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAU,GAAM,EAAQ,EAAK,OAAO,QACxC,AAAI,EAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,SAEzC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAGzC,GAAM,GAAuB,EAAK,gBAAgB,OAC5C,EAAA,EAAA,MAAA,EAAA,CAAA,EAAA,EAAA,MAAA,EAAA,OAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAGN,EAAc,GAAM,EAAU,GAE9B,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GACT,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAE/C,GAAM,GAAa,GAAM,EAAA,IAAQ,EAAG,GAAW,GAC/C,EAAK,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,IAEvC,GAAM,GACF,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,IAAK,IAElE,MAAO,CAAC,EAAG,MAIf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,eACrB,WAAY,IAGd,MAAA,IAAA,GAAW,EAAe,IAvMrB,EAAA,UAAY,UAyMrB,GA3M6B,IA4M7B,EAAA,cAAc,cAAc,IA8B5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAQ,GACxB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,MAuCrB,GAzCyB,IA0CzB,EAAA,cAAc,cAAc,IAuC5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAsC5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAZJ,SAAA,mBAAqB,OACrB,EAAA,6BAA+B,cAC/B,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAEhC,EAAA,yBAA2B,QASlC,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,EAAK,eAE3B,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,CAAC,EAAK,MAAO,EAAK,OACnC,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,GAAI,GACJ,GAAI,KAAK,SACP,GAAI,KAAK,gBACP,GAAM,GAAmB,KAAK,gBACxB,EAAgB,KAAK,MAC3B,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GAAM,GAAK,EAAiB,MAAM,CAAC,IAC7B,EAAM,GAAI,MAAQ,MAAM,CAAC,IACzB,EAAS,EAAiB,MAAM,CAAC,EAAgB,IACvD,MAAO,IACH,GAAuB,EAAI,GAAK,IAExC,GAZgD,IAEvC,EAAA,UAAY,oBAYrB,GAAkB,KAAK,gBAEzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,EAAiB,KAAK,gBACtD,GAAM,KAAK,oBAEf,MAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,wDACG,GAAO,OAAM,MAEtB,GAAI,GAAW,EAAO,GAChB,EAAW,EAAO,GACxB,EAAS,EAAO,GAChB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EACF,EAAK,qBAKL,EACA,EACA,EACA,EACJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAI,GAAM,EAAQ,EAAK,OAAO,QAClC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAEzC,EAAI,EAAA,IAAQ,EAAG,GAAM,EAAU,EAAK,gBAAgB,SACpD,AAAI,EAAK,SACP,GAAI,GAAU,EAAG,EAAK,KAAK,SAGvB,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,KACnE,EAAI,EAAK,oBAAoB,MAAM,GAEnC,GAAM,GAAI,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,IAE3C,MAAO,CAAC,EAAG,EAAG,MAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,eAAgB,KAAK,eACrB,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,gBAGvB,MAAA,IAAA,GAAW,EAAe,IAnNrB,EAAA,UAAY,WAqNrB,GAvN8B,IAwN9B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAGxB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAS,GACzB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,OAuCrB,GAzC0B,IA0C1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAME,OADM,GAAsB,GACT,EAAA,EAAA,EAAA,KAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAU,KAAI,MAAd,EAAkB,EAAK,WAEvB,EAAU,KAAK,EAAK,WAGxB,MAAO,oCAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,EAKT,OAJI,GAAS,EAAO,MAAM,GAGpB,EAA2B,GACd,EAAA,EAAA,EAAA,EAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAa,KAAK,EAAO,OAAO,EAAG,EAAK,UAAU,SAElD,EAAa,KAAK,EAAO,OAAO,EAAG,IAGvC,EAAa,UAKb,OAFM,GAA8B,GAChC,EACK,EAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAO,EAAK,MAAM,GACxB,EAAS,EAAa,GAEtB,AAAI,IAAM,EACR,EAAa,CAAC,EAAO,IAAI,OAAO,GAEhC,EAAa,CAAC,EAAW,IAAI,OAAO,GAEtC,EAAa,EAAK,KAAK,EAAY,GACnC,EAAgB,KAAK,EAAW,MAAM,IAIxC,EAAS,GACT,OAAyB,GAAA,EAAA,EAAA,EAAgB,QAAQ,UAAxB,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAO,KAAI,MAAX,EAAe,GAEjB,MAAO,CAAC,EAAW,IAAI,OAAO,MAI3B,EAAA,UAAA,MAAP,SAAa,GACX,AAAI,GAAgB,IAGlB,GAAc,EAAuB,IAEvC,EAAa,EACb,GAAI,GACJ,KAAK,MAAM,QAAQ,SAAC,EAAM,GACxB,GAAU,WAAW,EAAK,WAGxB,EAAK,MAAM,GACX,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAY,EAAK,UAAU,GAE3B,EAAY,EAAK,UAEnB,EAAa,CAAC,EAAW,GAAI,OAGjC,KAAK,MAAQ,IAGf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAgB,SAAC,GACrB,MAAO,CACL,UAAa,EAAK,eAClB,OAAU,EAAK,cAIb,EAAc,KAAK,MAAM,IAAI,GAE7B,EAAS,CAAC,MAAS,GAEzB,MAAA,IAAA,GAAW,EAAe,IAIrB,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAElB,OADM,GAAmB,GACA,EAAA,EAAA,EAAC,EAAO,MAAR,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAM,KAAK,GAAY,EAAY,IAErC,MAAO,IAAI,GAAI,CAAC,MAAK,KAGvB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,kBAEvB,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,qBAEvB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACvB,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAiB,KAAI,MAArB,EAAyB,EAAK,kBAEhC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAQT,EAAA,UAAA,WAAA,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,SAEvB,MAAO,IAAc,IASvB,EAAA,UAAA,WAAA,SAAW,GAET,OADM,GAAyC,GAC5B,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,IAGjB,OAHS,GAAI,EAAA,GACP,EAAY,EAAK,QAAQ,OACzB,EAAe,EAAQ,OAAO,GAC3B,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,EACzC,EAAO,KAAK,CAAC,EAAK,QAAQ,GAAI,EAAa,KAG/C,GAAc,IA5KT,EAAA,UAAY,kBAgLrB,GAlLqC,IAmLrC,EAAA,cAAc,cAAc,gBAEQ,GAM3B,GAAA,GAAA,EAAA,KAAM,EAAA,EAAA,KAAM,EAAA,EAAA,SAAA,EAAA,IAAA,OAAA,GAAA,EAAkB,EAAA,EAAA,MAAA,EAAA,IAAA,OAAA,EAAA,EAE/B,EAAgB,WAAM,MAAA,IAAU,IAAQ,IAExC,EAAa,WAAM,MAAA,IAAe,EAAe,EAAM,IAG7D,GAAI,CAAC,GAAS,GAAS,EACrB,MAAO,GAAA,KAAS,IAAa,SAG/B,GAAM,GAAQ,MAAM,GAAO,KAAK,QAAW,IAAI,GAE/C,MAAO,GAAM,IAAI,SAAA,GAAK,MAAA,GAAA,KAAS,EAAE,WC/iEnC,AAiFA,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAArC,4DAyBA,MAAA,IAzBqC,IAiCrC,GAAA,SAAA,GAAwB,EAAA,EAAA,GAMtB,WAAY,GAAZ,GAAA,GAAA,KACE,GAAI,EAAK,OACP,KAAM,IAAI,IACN,sDAGN,GAAI,MAAM,QAAQ,EAAK,MACrB,KAAM,IAAI,IACN,kEAGN,SAAA,EAAA,KAAA,KAAM,IAAqB,KAE3B,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WAad,GAZA,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aAEtB,EAAK,KAAK,YAAc,MAG1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBAEtB,EAAK,KAAK,qBAAuB,MAG/B,GAAU,EAAO,UACnB,KAAM,IAAI,GAAW,6CAGvB,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KAEtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAE1C,EACF,GAAU,KAAO,KAAO,EAAO,aAEnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAI3D,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GAAkB,KAAK,yBAAyB,GAEpD,MAAK,MAAK,iBACR,GAAQ,CAAI,EAAS,IAAE,OAAK,EAAS,MAAM,KAG7C,AAAI,KAAK,aACP,GAAQ,CACH,GAAQ,OAAK,MAAM,GAAG,KAAI,CAAE,EAAW,IAAE,OAAK,EAAS,MAAM,QAG7D,GAGT,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACP,GAAA,GAAA,EAAA,KAAA,UAED,EAAa,EAAO,MAEpB,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAe,EAAA,MAAU,GAE/B,MAAI,OAAM,QAAQ,GACT,MAAM,EAAU,QAAQ,KAAK,GAG/B,CAAC,MAIZ,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAS,WACP,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAGN,GAAM,GAAa,EAAK,UAAU,GAAG,MAE/B,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAY,EAAW,GAE7B,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUASN,GAAI,EAAK,aAAe,KACtB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAU,CAAC,EAAA,MAAU,YAEnB,GAAU,KAEnB,EAAA,QAAY,EAAK,SAGjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAQ,GAAK,EAAA,MAAU,QAO9B,GAJA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAGR,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,EAKF,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GAEf,EAAgB,EAEtB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAGhB,EAAK,QAAQ,GAAS,GAI1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAIlD,EAAA,UAAA,yBAAV,SAAmC,GAC3B,GAAA,GAAA,KAAA,KAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,QAAS,EAAA,EAAA,aAGpD,EAAkB,IAAe,gBAEjC,EAAI,EAAW,EAAkB,EAAI,GACrC,EAAI,EAAW,EAAkB,EAAI,GAErC,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAClD,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAElD,EACD,EAAW,MAAM,EAAG,GAAE,OACrB,EAAkB,CAAC,EAAS,EAAM,GAAQ,CAAC,EAAM,EAAM,IAG7D,MAAO,IAhMF,EAAA,UAAY,YAkMrB,GApMwB,IAyMxB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAWlC,WAAY,GAAZ,GAAA,GAAA,KAEI,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,aAGF,SAAA,EAAA,KAAA,KAAA,GAAA,GAAU,EAAI,CAAE,MAAO,MAAS,KAEhC,EAAK,QAAU,EACf,GAAsB,EAAK,QAAS,WAEpC,EAAK,WAAa,GAAe,EAAY,EAAG,cAChD,EAAK,WAAW,QAAQ,SAAA,GAAQ,MAAA,IAAsB,EAAM,gBAE5D,EAAK,QAAU,GAAe,GAAW,EAAG,EAAG,WAC/C,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,IAAsB,EAAQ,aAE7D,EAAK,QAAU,GAAW,QAC1B,GAAiB,EAAK,SAEtB,EAAK,WAAa,GAAc,eAChC,GAAgB,EAAK,YAErB,EAAK,aAAe,GAAe,GAAgB,EAAG,EAAG,gBACzD,EAAK,aAAa,QACd,SAAA,GAAQ,MAAA,IAAsB,EAAM,oBAGnC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAEhC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAElE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAG1B,GAAM,GAAW,EAAW,GAEtB,EAAe,EAEf,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,QAAU,IAErD,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBAEvC,GAAM,GACF,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,KAAK,QAAU,IAOzD,GALA,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBAEL,KAAK,SACP,GAAI,GAAe,OAEnB,GAAI,KAAK,gBACP,GAAM,GAAO,KAAK,gBAEZ,EAAU,KAAK,QAErB,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAQ,EAAK,MAAM,CAAC,IACpB,EAAQ,EAAA,KAAS,CAAC,IAClB,EAAY,EAAK,MAAM,CAAC,EAAU,IACxC,MAAO,IAAc,CAAC,EAAO,EAAO,KAExC,GAVgD,IAEvC,EAAA,UAAY,oBAUrB,GAAkB,KAAK,gBAGzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,QAAU,GAAe,KAAM,EAC7C,KAAK,gBAAiB,GAAM,KAAK,gBAGvC,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAsB,GAA3B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8DACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,GAEjC,EAAI,EAAO,GACX,EAAW,EAAO,GAClB,EAAW,EAAO,GAElB,EAAe,EAIrB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAI9B,GAAM,GAAc,EAAK,YAEnB,EACF,SAAC,GAAe,GAAoB,IAClC,MAAI,CAAC,IAAQ,CAAC,GAAK,IACV,GAGF,EAAA,IAAQ,GAAK,IAAQ,KAG9B,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAEtC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAIvC,GAAM,GAAiB,EAAK,qBAExB,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAE1C,EAAoB,EAEpB,EAAA,EAAA,MAAA,EAAA,OAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAG5B,EAAA,EAAA,uDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAI5B,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAEvC,GAAA,GAAA,EAAA,MAAA,EAAA,gBAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAI3C,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAE5B,GAAM,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAA,IACN,EAAA,IAAQ,GAAG,GACX,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,MAC3C,GAAI,EAAA,IACN,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC3C,EAAK,WAAW,MAAM,KAE1B,MAAO,CAAC,GAAG,GAAG,OAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAA,EAAA,UAAA,UAAA,KAAA,MAAC,EAAA,EAAA,MAAY,EAAA,GAAA,EAAA,CAAA,UAEb,EAAuC,CAC3C,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,QAAS,KAAK,SAGhB,MAAA,IAAA,GAAW,EAAe,IAG5B,EAAA,UAAA,UAAA,SAAU,EAAW,EAAW,EAAY,GAC1C,GAAM,GAAM,EAAA,OACR,EAAmB,EAAmB,KAAK,QAC1C,GAAW,QACZ,KAAK,aAAe,gBAAkB,OAAS,OAC/C,KAAK,cAET,MAAI,GACK,GAAU,EAAK,EAAG,KAAK,YAGzB,GAGT,EAAA,UAAA,cAAA,SAAc,EAAW,GACvB,GAAM,GAAU,EAEhB,MAAO,GAAA,OACH,EAAmB,EAAmB,EAAS,OAC/C,KAAK,aAAe,gBAAkB,OAAS,SAzO9C,EAAA,UAAY,iBA2OrB,GA7OoC,IA+OpC,EAAA,cAAkB,cAAc,IAKhC,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,GAAI,IAAe,GAEhC,SAAA,EAAA,KAAA,KAAM,GAAA,GAAI,EAAI,CAAE,KAAI,MAAwB,OAIvC,SAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IAZV,EAAA,UAAY,aAcrB,GAhBgC,IAkBhC,EAAA,cAAkB,cAAc,ICjkBhC,AAkDA,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,KAAK,IAAI,KAAK,IAAI,EAAK,KAAM,GAAI,GAE7C,EAAK,WAAa,EAAK,WACvB,EAAK,KAAO,EAAK,KACjB,EAAK,gBAAkB,KAGf,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,KAAK,YAAc,KACrB,MAAO,MAAK,WAId,OAFM,GAAa,EAAM,MACnB,EAAoB,GACjB,EAAI,EAAG,EAAI,KAAK,WAAW,OAAQ,EAAE,EAC5C,EAAW,KACP,KAAK,WAAW,IAAM,KAAO,EAAW,GAAK,KAAK,WAAW,IAEnE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAI,EAAK,MAAQ,EAAK,KAAO,GAC/B,GAAM,GACF,EAAO,UAAe,KAAO,GAAQ,EAAO,SAC1C,EAAa,EAAK,cAAc,GAChC,EAAS,GACX,WAAM,MAAA,IAAU,EAAO,EAAK,KAAM,EAAY,EAAK,OACnD,WAAM,MAAA,IAAO,GACjB,MAAO,GAET,MAAO,MAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,KACX,WAAY,KAAK,WACjB,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAxDf,EAAA,UAAY,UA0DrB,GA5D6B,IA6D7B,EAAA,cAAc,cAAc,IA4D5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAIpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,CAAC,KAAM,MAGjB,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAM,GAAa,EAAM,MACzB,MAAO,CAAC,EAAW,GAAI,EAAG,EAAW,KAThC,EAAA,UAAY,mBAWrB,GAbsC,IActC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAmBzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAhBM,EAAA,WAA2B,KAC3B,EAAA,QAAU,GAGV,EAAA,OAAwB,KACxB,EAAA,KAAsB,KAErB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAQrD,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MACnD,EAAK,UAAY,MAGnB,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAK,gBAAkB,CAAC,EAAW,EAAK,UAG1C,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GAAc,EAAK,YACrC,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAK,SAEtB,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,gBAAkB,GAEvB,EAAK,UAAY,CAAC,CAAC,QAAS,MAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAe,EAAW,EAAW,OAAS,GACpD,AAAI,KAAK,QAAU,MACjB,MAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAc,KAAK,OAAQ,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,kBAIzC,KAAK,UAAY,CAAC,CAAC,QAAS,EAAG,KAAI,GAAA,GAAG,EAAC,IAAK,EAAY,KACxD,KAAK,MAAQ,IAGf,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,SAAY,EAAY,OAAS,GAAK,KAAK,MACpC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAM,GAAQ,GAAoB,GAC5B,EACF,GAA2B,EAAK,WAAW,gBAC3C,EAEJ,MAAI,IAAuB,KACzB,EAAS,GACL,EAAO,EAAK,OAAO,OAAQ,EAC3B,EAAK,KAAO,EAAK,KAAK,OAAS,MAEnC,GAAS,GAAM,EAAO,EAAK,OAAO,QAClC,AAAI,EAAK,MAAQ,MACf,GAAS,GAAU,EAAQ,EAAK,KAAK,SAEvC,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,KAI5B,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArHF,EAAA,UAAY,QAuHrB,GAzH2B,IA0H3B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAO,GAAQ,GACf,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,UAAY,CAAC,CAAC,QAAS,IAC5B,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,OAAkB,GAAA,EAAA,EAAA,EAAW,MAAM,GAAjB,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAO,KACT,KAAM,IAAI,GACN,4DACA,SAAQ,EAAW,MAAM,GAAE,oCAC3B,mFAIR,MAAO,CAAC,EAAW,GAAI,GAAU,EAAY,KAG/C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAK,aAAe,iBAAmB,EAAM,KAAO,GAEtD,OADM,GAAwB,CAAC,GACtB,EAAI,EAAG,EAAI,EAAM,KAAM,EAAE,EAChC,EAAY,KAAK,GAEnB,EAAY,KAAK,GACjB,EAAQ,EAAM,UAAU,GAG1B,MAAO,IAAe,MAI1B,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,GACzC,AAAI,KAAK,YAAc,MACrB,GAAO,WAAgB,KAAK,YAE9B,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA/CF,EAAA,UAAY,UAiDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAK9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,WAAa,GAAc,EAAK,cAGvC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAK,WAAW,MAAM,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,GAAoB,KAAK,aAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArBF,EAAA,UAAY,aAuBrB,GAzBgC,IA0BhC,EAAA,cAAc,cAAc,IAc5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAKhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,EAAI,EAAK,EACd,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,KAAK,EAAG,EAAW,KAG5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAS,GAAoB,GACtB,GAAS,EAAQ,EAAK,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,EAAG,KAAK,GAEJ,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,eA4BrB,GA9BkC,IA+BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,YAAc,EAAK,YAGxB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,EAAE,EAC7C,AAAI,EAAK,UAAU,EAAK,YAAY,KAClC,GAAK,YAAY,GAAK,eAKpB,SAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,GAAM,GAAK,GAAO,MAiBnB,EAAA,UAAA,oBAAR,SAA4B,EAAmB,GAK7C,OAJM,GAAW,6CACX,EAAa,EAAY,QAC3B,EAAQ,EACR,EAAU,KACL,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAM,EAAW,GACvB,GAAI,KAAK,UAAU,GACjB,GAAI,IAAY,KACd,EAAU,MAEV,MAAM,IAAI,GAAW,gDAGvB,IAAS,EAIb,GAAM,GAAe,GAAU,GAC/B,GAAI,IAAY,MACd,GAAI,IAAU,GAAK,EAAe,IAAU,EAC1C,KAAM,IAAI,GAAW,GAEvB,EAAW,GAAW,EAAe,UAC5B,IAAiB,EAC1B,KAAM,IAAI,GAAW,GAGvB,MAAO,IAGT,EAAA,UAAA,mBAAA,SAAmB,GAEjB,OADI,GAAiB,GACZ,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,GAAI,KAAK,UAAU,EAAW,KAC5B,EAAiB,GACjB,MAIJ,MAAI,GACK,EAAW,MAAM,EAAG,GAAG,OAAO,KAAK,aAEnC,EAAW,MAAM,EAAG,GAAG,OAC1B,KAAK,oBAAoB,EAAW,MAAM,GAAI,KAAK,eAI3D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAc,EAAW,MAAM,EAAG,GAAG,OACvC,EAAK,oBAAoB,EAAW,MAAM,GAAI,EAAK,cACvD,MAAO,GAAM,QAAQ,MAIzB,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlGF,EAAA,UAAY,UAoGrB,GAtG6B,IAuG7B,EAAA,cAAc,cAAc,IAY5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAM3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,OACN,mFAGN,GAAI,CAAC,MAAM,QAAQ,EAAK,MACtB,KAAM,IAAI,OACN,oEACG,GAAK,KAAI,cAIlB,GAAM,GAAwB,GAAM,EAAG,EAAK,KAAK,OAAS,GAC1D,GAAI,CAAC,EAAA,KAAK,YAAY,EAAK,KAAK,QAAQ,OAAQ,GAC9C,KAAM,IAAI,OACN,+BAAiC,KAAK,UAAU,EAAK,MACrD,8DAGN,SAAK,KAAO,EAAK,KACjB,EAAK,mBAAqB,CAAC,GAAG,OAAO,EAAK,MAC1C,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,EAAK,KAAK,OAAS,OAG5D,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,YAAK,KAAK,QAAQ,SAAC,EAAa,GAC9B,EAAY,EAAI,GAAM,EAAqB,KAEtC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,UAAU,GAAoB,GAAS,KAAK,qBAGrD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjDF,EAAA,UAAY,UAmDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,KACV,EAAK,UAAY,EAAK,WAAa,KAAO,EAAI,EAAK,UAEnD,EAAK,UAAY,IAIrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,UAAW,KAAK,WAChC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACb,MAAO,GAAA,IAAI,EAAA,SAAS,EAAO,KAAK,WAAY,IAG9C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACP,EAAW,GACX,EAAc,EAAA,IAAI,EAAA,SAAS,EAAO,EAAK,WAAY,EAAM,GACzD,EAAS,EAAM,IAAI,EAAY,OAAO,EAAM,QAClD,MAAO,MAtCJ,EAAA,UAAY,UAyCrB,GA3C6B,IA4C7B,EAAA,cAAc,cAAc,IClqB5B,AA2EA,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAgB7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GATM,EAAA,WAA4B,KAE3B,EAAA,+BACL,gBAME,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MAKrD,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,AAAI,EAAK,aAAe,KAGtB,EAAK,gBAAkB,CAAC,EAAW,MAInC,EAAK,gBACD,CAAC,GAAW,OAAO,GAAqB,EAAK,cAGrD,SAAK,SAAW,EAAK,SACrB,GAAoC,EAAK,SAAU,YACnD,EAAK,UAAY,EAAK,UACtB,GAAoC,EAAK,UAAW,aACpD,EAAK,sBAAwB,GACzB,EAAK,uBAAyB,EAAK,gCACvC,EAAK,sBAAwB,GAAe,EAAK,uBACjD,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,qBAAuB,GAAc,EAAK,sBAC/C,EAAK,SAAW,EAAK,SACrB,EAAK,gBAAkB,EAAK,SAC5B,EAAK,YAAc,EAAK,cAGnB,SAAA,UAAA,MAAP,SAAa,GACX,KAAK,WAAa,KAAK,UACnB,aAAc,CAAC,KAAK,SAAU,KAAK,WAAY,KAAK,MACpD,KAAK,sBAAuB,KAAK,sBAAuB,GACxD,KAAK,sBACT,KAAK,MAAQ,IAKL,EAAA,UAAA,6BAAV,SAAuC,KAEvC,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAK,GAAK,SAGR,GAAS,GAAoB,GACtB,EAAA,SAAS,EAAQ,EAAA,UAAU,KAH3B,QAQb,EAAA,UAAA,mBAAA,SAAmB,GAEjB,GADA,EAAa,GAAmB,GAC5B,KAAK,aAAe,KACtB,MAAW,GAAU,OAAA,CAAE,KAAK,YAG9B,GAAM,GAAmB,GAAqB,KAAK,aACnD,GAAI,EAAO,SAAW,EAAW,OAAS,EACxC,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IAG7B,OADI,GAAI,EACC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAK,EAAO,GACZ,EAAK,EAAW,EAAI,GAC1B,GAAK,GAAM,MAAU,GAAM,MAAU,IAAO,EAC1C,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IACxB,AAAI,GAAM,MACf,GAAO,GAAK,GAEd,IAGJ,MAAA,CAAQ,EAAW,IAAE,OAAK,EAAM,CAAE,KAAK,aAGzC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,AAAI,EAAM,QAAU,SAClB,GAAQ,GAAO,EAAO,UAExB,GAAM,GAAS,GAAS,EAAK,WAAW,OAAQ,EAAM,QACtD,MAAO,GAAO,QACV,GAAmB,EAAK,mBAAmB,EAAM,YAIzD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,UAAW,KAAK,UAChB,sBAAuB,GAAqB,KAAK,uBACjD,sBAAuB,GAAqB,KAAK,uBACjD,oBAAqB,GAAqB,KAAK,qBAC/C,qBAAsB,GAAoB,KAAK,sBAC/C,SAAU,KAAK,SACf,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjIF,EAAA,UAAY,YAmIrB,GArI+B,IAsI/B,EAAA,cAAc,cAAc,ICjN5B,AA+BA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAGlC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAQ,KAAG,KACjB,SAAK,gBAAkB,KAOf,SAAA,UAAA,cAAV,SAAwB,GACtB,KAAM,IAAI,KAaJ,EAAA,UAAA,gCAAR,SAAwC,EAAe,GACrD,GAAI,GAAU,MAAQ,GAAU,KAC9B,MAAO,MACF,GAAI,EAAO,OAAS,EAAO,OAChC,MAAO,MAAK,gCAAgC,EAAQ,GAC/C,GAAI,EAAO,SAAW,EAC3B,MAAO,GAGT,OADM,GAAqB,EAAO,MAAM,EAAG,EAAO,OAAS,EAAO,QACzD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,EAAO,OAAS,EAAO,OAAS,GAC3C,EAAI,EAAO,GACjB,GAAI,GAAK,MAAQ,GAAK,MAAQ,EAAI,GAAK,EAAI,EACzC,EAAY,KAAK,cACR,IAAM,EACf,EAAY,KAAK,WACR,IAAM,EACf,EAAY,KAAK,QAEjB,GAAI,IAAM,EACR,KAAM,IAAI,GACN,wDACA,KAAK,UAAU,GAAU,IAAM,KAAK,UAAU,IAEpD,EAAY,KAAK,IAGrB,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GAOJ,GALA,AAAI,MAAM,QAAQ,IAAe,CAAC,MAAM,QAAQ,EAAW,KAEzD,GAAa,CAAC,GAAmB,KAEnC,EAAa,EACT,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,mEACA,SAAQ,EAAW,OAAM,eAM/B,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAI1B,GADA,EAAa,GAAqB,GAC9B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,qDACA,6BAA4B,KAAK,UAAU,GAAW,MAK5D,OAFI,GACA,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GAC9C,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,GAAM,GAAW,EAAW,IAAI,SAAA,GAAS,MAAA,GAAM,SAC/C,AAAI,EAAW,QAAQ,QAAU,IAC7B,GAAqB,GAAU,SAAW,EAC5C,KAAK,gBAAkB,GAEvB,KAAK,gBAAkB,IAI3B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAK,iBACP,GAAM,GAA2B,GAC3B,EAAY,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,OAC5C,GAAI,EAAU,QAAQ,QAAU,IAI9B,OADM,GAAU,GAAc,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAEZ,OAFO,GAAC,EAAA,GACF,EAAQ,EAAE,KACP,EAAI,EAAG,EAAI,EAAU,EAAO,EAAE,EACrC,EAAI,GAAa,EAAG,GAEtB,EAAe,KAAK,GAEtB,MAAO,GAAK,cAAc,QAK1B,OADI,GAAa,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,KAChB,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAY,EAAO,GACnB,EAAW,EAAO,MAAM,GAAG,OAAO,CAAC,IACrC,EAAc,EAAE,QAChB,CAAC,GAAW,OAAO,GAAoB,EAAO,MAAM,MACxD,EAAc,EAAA,UAAc,EAAa,CAAC,EAAG,IAC7C,EAAc,EAAY,QAAQ,GAClC,EAAe,KAAK,GACpB,EAAa,WACJ,EAAQ,GACjB,GAAM,GAAO,GAAgB,EAAG,GAAO,OAAO,CAAC,IAC/C,EAAe,KAAK,EAAA,UAAc,EAAG,IACrC,EAAa,OAGb,GAAe,KAAK,GAGxB,GAAI,GAAI,EAAK,cAAc,GACrB,EAAQ,EAAE,KAChB,GAAI,GAGF,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAQ,EAAO,OACf,EAAY,EAAO,EAAQ,GAC3B,EACF,CAAC,GAAW,OAAO,EAAO,MAAM,EAAG,EAAO,OAAS,IACvD,EAAI,EAAA,UAAc,EAAE,QAAQ,CAAC,GAAI,IAAa,CAAC,EAAG,IACzC,QAAQ,WACR,EAAQ,GACjB,GAAM,GAAO,CAAC,EAAQ,GAAG,OAAO,GAAgB,EAAG,EAAQ,IAC3D,EAAI,EAAA,UAAc,EAAG,IAGzB,MAAO,QAGT,OAAO,GAAK,cAAc,MAKhC,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,EACb,GAAI,GACJ,AAAI,EAAW,IAAM,KACnB,EAAc,KAEd,EAAc,EAAW,GAAG,MAAM,GAEpC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAG1B,SAAa,GAAqB,GAClC,AAAI,EAAW,SAAW,EACxB,EAAc,EAAW,OAAO,GAEhC,EAAc,CAAC,MAAM,OAAO,GAEvB,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6BAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+BAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,kGAEA,KAAI,EAAO,OAAM,OAAO,EAAK,OAAM,MAEzC,GAAI,EAAK,MAAM,SAAA,GAAK,MAAA,IAAK,OACvB,MAAO,MAET,EAAO,EAAK,IAAI,SAAA,GAAK,MAAA,IAAK,KAAO,EAAI,EAAA,WAAe,EAAG,KAEvD,OADI,GAAS,EAAK,GACT,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAS,EAAA,WAAe,EAAQ,EAAK,IAEvC,MAAO,MAGb,GAlOoC,IAoOpC,GAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,MAcrB,GAhByB,IAiBzB,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAG5B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,WAcrB,GAhB8B,IAiB9B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,GAAA,IAAQ,EAAI,EAAO,OAAQ,MAX/B,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA2D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IAiE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAM/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAJJ,SAAA,aAAe,GAKtB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,MACxD,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,yEAGN,EAAa,EAGb,OADI,GAAe,GACC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,GAAS,MACX,EAAe,GACf,OAGJ,GAAI,EACF,OAIF,OADM,GAAoB,GACjB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAyB,EAAW,GAAG,QAC7C,EAAuB,OAAO,KAAK,KAAM,GAEzC,OADI,GAAS,GACO,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAA,KAAK,YAAY,EAAO,IAC1B,EAAS,GACT,OAGJ,AAAK,GACH,EAAS,KAAK,GAGlB,GAAI,EAAS,OAAS,EACpB,KAAM,IAAI,GACN,4GAEA,KAAK,UAAU,KAIb,EAAA,UAAA,cAAV,SAAwB,GAAxB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAO,IAAc,EAAQ,EAAK,SAItC,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,KAC1D,KAAM,IAAI,GACN,+DAON,OALM,GAAc,EACd,EAAc,EAAY,GAAG,QAC7B,EAAO,KAAK,KAAO,EAAI,EAAY,OAAS,KAAK,KAAO,KAAK,KAG/C,EAAA,EAAA,EAAA,EAAY,MAAM,GAAlB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAY,IAAS,MAAQ,EAAM,IAAS,MAC9C,EAAY,GAAQ,KACpB,MAEF,EAAY,IAAS,EAAM,GAE7B,MAAO,IAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6CAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+CAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,mCAAmC,EAAK,OAAM,KAC9C,8BAA6B,EAAO,OAAM,MAEhD,MAAO,GAAA,KAAS,WACd,GAAI,GAAe,GAOnB,GANA,EAAK,QAAQ,SAAA,GACX,GAAI,GAAK,MACP,EAAe,GACf,UAGA,EACF,MAAO,MAGT,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAK,IAAM,KAEb,EAAY,KAAK,EAAA,SAAa,EAAO,IAAI,OAAO,SAC3C,AAAI,EAAK,GAAG,KAAO,EAAO,GAAG,KAElC,EAAY,KAAK,EAAA,WAAe,EAAK,GAAI,KAEzC,EAAY,KAAK,EAAK,IAG1B,GAAM,GAAoB,EAAA,OAAW,EAAa,EAAK,MACvD,MAAO,GAAA,IAAQ,EAAmB,GAAI,OAI1C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,MAET,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArIF,EAAA,UAAY,cAuIrB,GAzIiC,IA0IjC,EAAA,cAAc,cAAc,IAuF5B,YAAuB,EAAc,GACnC,KAAO,EAAO,GACZ,GAAQ,EAEV,MAAO,GAGT,YAAkB,EAAW,EAAW,GACtC,GAAI,EAAE,MAAM,OAAS,GAAK,EAAE,MAAM,OAAS,EACzC,KAAM,IAAI,IACN,oEAeN,GAbA,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAC3B,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAE3B,AAAI,MAAO,IAAS,UAClB,GAAO,CAAC,EAAM,IAGZ,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,KAAM,IAAI,IACN,+DAGN,GAAM,GAAQ,EAAE,MAAM,OAChB,EAAQ,EAAE,MAAM,OACtB,AAAI,GAAQ,MAEV,GAAO,CAAC,EAAQ,EAAG,EAAQ,IAE7B,GAAM,GAAY,EAElB,MAAO,GAAA,KAAS,WACd,GAAI,GACJ,GAAI,EAAQ,GACV,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,YACpB,EAAQ,GACjB,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,QAE7B,GAAO,EAGT,GAAI,GACJ,GAAI,EAAE,MAAM,SAAW,GAAK,EAAE,MAAM,SAAW,EAC7C,AAAI,EAAU,KAAO,EAAU,GAC7B,EAAM,EAAE,IAAI,GAAG,IAAI,EAAU,IAE7B,EAAM,EAAE,UAAU,CAAC,EAAG,IAAI,IAAI,GAAG,IAAI,EAAU,SAGjD,GAAM,GAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EACzC,EAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EAC/C,EAAM,EAAE,OAAO,EAAG,EAAM,GAG1B,GAAI,EAAO,GACT,GAAI,GAAG,OACP,AAAI,EAAQ,EACV,EAAM,EAAQ,EAAQ,EAEtB,EAAM,EAAQ,EAGhB,OADM,GAAwB,GACrB,EAAI,EAAK,EAAI,EAAM,EAAM,EAAE,EAClC,EAAY,KAAK,GAEnB,EAAM,EAAI,QAAQ,GAEpB,MAAI,GAAI,MAAM,SAAW,GACvB,GAAM,EAAI,WAAW,IAEhB,IAIX,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,EAAK,KACjB,EAAK,UAAY,EAAK,WAAa,KAAO,GAAQ,EAAK,UACvD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAS,EAAW,GACpB,EAAS,EAAW,GAC1B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,GAAI,EAAO,EAAK,MAAQ,EAAO,EAAK,IAClC,KAAM,IAAI,GACN,8BACG,GAAO,EAAK,IAAG,QAAQ,EAAO,EAAK,OAIpC,EAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,qDACA,iBAAgB,EAAO,OAAM,eAGnC,GAAI,GAAK,EAAO,GACZ,EAAK,EAAO,GACZ,EACJ,MAAK,OAAM,QAAQ,KAAK,MAMtB,EAAO,KAAK,KAAK,IACN,SAAC,EAAM,GAAM,MAAA,IACT,EAAM,EAAO,GAAG,MAAM,UAPrC,EAAO,CACL,GAAc,KAAK,KAAM,EAAG,MAAM,QAClC,GAAc,KAAK,KAAM,EAAG,MAAM,SAOtC,AAAI,KAAK,WACP,GAAK,GAAY,EAAI,EAAK,IAC1B,EAAK,GAAY,EAAI,EAAK,KAErB,GAAS,EAAI,EAAI,IAGlB,EAAA,UAAA,cAAR,SAAsB,EAAe,GACnC,GAAI,GACJ,MAAK,OAAM,QAAQ,KAAK,MAQtB,EAAO,KAAK,KANZ,EAAO,CACL,GAAc,KAAK,KAAM,EAAO,QAChC,GAAc,KAAK,KAAM,EAAO,SAM7B,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAU,EAAW,GAAa,QAClC,EAAU,EAAW,GAAa,QACxC,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAG,GACjB,GAAM,GAAc,EAAO,OAAO,GAClC,MAAI,GAAY,SAAW,GACzB,EAAY,KAAK,GAEZ,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,OAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,KACb,UAAa,KAAK,WAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA7GF,EAAA,UAAY,MA+GrB,GAjHyB,IAkHzB,EAAA,cAAc,cAAc,IChjC5B,AA2BA,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAKjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,OAAS,EAAK,SAGrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,OAAQ,KAAK,QAC7B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAS,WACX,MAAA,IAAe,EAAM,MAAO,EAAG,EAAK,QAAQ,IAAI,IAC9C,EACF,GAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAC9D,MAAO,MA5BJ,EAAA,UAAY,gBA+BrB,GAjCmC,IAkCnC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAS,WACb,GAAM,GAAS,KAAK,KAAK,EAAK,KAAQ,GAAI,EAAK,OAC/C,MAAO,GAAM,IAAI,GAAe,EAAM,MAAO,EAAG,KAElD,MAAO,IAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAEnE,MAAO,MA/BJ,EAAA,UAAY,kBAkCrB,GApCqC,IAqCrC,EAAA,cAAc,cAAc,IAyC5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAMhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,KACjB,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,YAAc,GAAoB,GAAQ,OAGxD,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAa,EAAK,eAAe,GAEjC,EAAgB,WACpB,GAAM,GAAQ,GAAoB,GAE5B,EAAQ,mBACR,EAAQ,mBAER,EAAS,CAAC,EAAQ,EAEpB,EAAU,EAAA,aAAa,EAAA,cAAc,GAAa,EAAK,MAE3D,EAAU,GAAO,EAAS,WAG1B,GAAM,GAAI,KAAA,IAAE,GAAI,EAAK,MAAS,GAAI,EAAK,KAAO,KAAA,IAAA,EAAU,IAAO,KACzD,EAAI,CAAC,EAAI,EAAS,EAAK,KAGvB,EAAI,EAAM,IAAI,GAAS,IAAI,EAAQ,IAAI,IAAI,IAAI,IAErD,MAAO,GAAE,IAAI,GAAG,IAAI,IAEtB,MAAO,IACH,EAAe,WAAM,MAAA,IAAoB,IACzC,EAAO,UAAe,IAE5B,MAAO,MAxDJ,EAAA,UAAY,eA2DrB,GA7DkC,IA8DlC,EAAA,cAAc,cAAc,IChN5B,YA4CI,EAAW,EAAc,EAAkB,EAAe,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GACJ,GAAI,EAAE,OAAS,EACb,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EAEpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EACpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,OAElC,MAAM,IAAI,IACN,2DAA2D,EAAE,KAAI,QAGvE,MAAO,GAoBT,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WACH,GAAM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EACF,GAAmB,EAAG,EAAM,EAAU,EAAM,EAAO,GACvD,MAAO,CAAC,EAAQ,EAAM,KAqBjC,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKH,OAJM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EAAwB,GACX,EAAA,EAAA,EAAA,GAAiB,EAAG,EAAE,MAAtB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAc,QAAQ,KAAU,GAClC,EAAY,KAAK,GAEjB,EAAY,KAAK,EAAE,MAAM,IAG7B,GAAM,GAAgB,EAAK,QAAQ,GAC7B,EAAoB,EAAS,QAAQ,GACrC,EACF,GAAS,KAAO,KAAO,EAAM,QAAQ,GACnC,EACF,GAAQ,KAAO,KAAO,EAAK,QAAQ,GACjC,EAAS,GACX,EAAG,EAAe,EAAmB,EACrC,EAAgB,GACpB,MAAO,CAAC,EAAQ,EAAM,iBAgB7B,EAAW,EAAe,EAAc,EACxC,GACF,MADE,AAAA,KAAA,QAAA,GAAA,MACE,EAAA,KAAK,YACD,EAAc,QAAQ,OAAQ,GAAiB,EAAG,EAAE,KAAO,IAC1D,GACH,EAAG,EAAO,EAAM,EAAe,GAE5B,GACH,EAAG,EAAO,EAAM,EAAe,GAsFvC,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAqBtC,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,IAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KAC1C,EAAK,SAAW,EAAK,UAAY,KAAO,IAAO,EAAK,SACpD,EAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,sBACD,GAAe,EAAK,uBAAyB,SACjD,EAAK,0BACD,GAAe,EAAK,2BAA6B,QACrD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAc,EAAK,iBAC1C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,oBAGvC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAO,KAAK,MAAQ,EAAI,KAAK,KAAQ,KAAK,KAAO,EAAW,OAC5D,EAAM,EAAW,GACvB,GAAI,GAAO,KACT,KAAM,IAAI,GACN,QAAQ,EAAI,+FAET,MAAK,UAAU,GAAW,MAEnC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAW,OAAQ,KAAI,GAAA,GAAG,EAAC,GAAO,EAAG,MAC/D,GAAM,GAAQ,CAAC,GACf,AAAI,KAAK,OACP,MAAK,MAAQ,KAAK,UACd,QAAS,EAAO,KAAM,KAAK,iBAAkB,KAAK,iBAClD,GAAM,KAAK,kBAEjB,AAAI,KAAK,QACP,MAAK,KAAO,KAAK,UACb,OAAQ,EAAO,KAAM,KAAK,gBAAiB,KAAK,gBAAiB,GACjE,KAAK,iBAEX,KAAK,WAAa,KAAK,UACnB,cAAe,EAAO,KAAM,KAAK,sBAAuB,KAAM,IAClE,KAAK,eAAiB,KAAK,UACvB,kBAAmB,EAAO,KAAM,KAAK,0BAA2B,KAChE,IACJ,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACvD,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAO,EAAW,OAClB,EAAgB,GAAiB,EAAG,GACpC,EAAO,EAAK,MAAQ,EAAI,EAAK,KAAQ,EAAK,KAAO,EACvD,EAAc,OAAO,EAAM,GAC3B,GAAM,GAAiB,GAA2B,EAAG,GACrD,EAAe,GAAQ,EAAW,GAElC,GAAM,GAAsB,EAAc,QAC1C,EAAoB,OACpB,GAAM,GAAoB,CAAC,EAAA,KAAK,YAC5B,EAAqB,GAAiB,EAAG,GAAM,MAAM,EAAG,EAAO,IAE7D,EAAmC,WACvC,GAAI,GACF,GAAM,GACF,EAAK,WAAW,OAAO,QAAQ,GAC7B,EACF,EAAK,eAAe,OAAO,QAAQ,GACjC,EACF,EAAK,OAAS,EAAK,KAAK,OAAO,QAAQ,GAAkB,KACvD,EACF,EAAK,MAAQ,EAAK,MAAM,OAAO,QAAQ,GAAkB,KAC7D,MAAO,IACH,EAAO,EAAqB,EAC5B,EAAe,EAAgB,EAAK,aAExC,OAAO,IACH,EAAO,EAAK,WAAW,OAAQ,EAAK,eAAe,OACnD,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACrC,EAAK,OAAS,KAAO,KAAO,EAAK,MAAM,OAAQ,EAAK,UAI5D,GAAI,CAAC,EACH,MAAO,KAGH,GAAA,GAAA,GAAA,EAAA,EAAA,MAAA,OAAA,EAAA,KAAA,OAAA,EAAA,EAAA,SAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAM,EAAA,EAAA,GAIvB,EACF,SAAC,EAAyB,EAAe,GACvC,EAAA,KAAS,WACP,GAAM,GAAQ,EAAI,EACZ,EAAY,EAAS,OACrB,EAAc,EAAU,IAAI,GAAO,IAAI,GAC7C,EAAS,MAAM,EAAU,IAAI,OAU/B,EAA8B,WAClC,EAAgB,EAAK,WAAY,EAAM,EAAK,UAC5C,EAAgB,EAAK,eAAgB,EAAU,EAAK,WAEtD,WAEO,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,SAAU,KAAK,SACf,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,sBAAuB,GAAqB,KAAK,uBACjD,0BACI,GAAqB,KAAK,2BAC9B,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,eAAgB,GAAoB,KAAK,gBACzC,gBAAiB,GAAoB,KAAK,kBAEtC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArKF,EAAA,UAAY,qBAuKrB,GAzKwC,IA0KxC,EAAA,cAAc,cAAc,IAkD5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAgBtC,WAAY,GAAZ,GAAA,GAAA,KAOE,GANA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KACtC,MAAO,GAAK,MAAS,UACvB,GAAI,CAAC,OAAO,UAAU,EAAK,MACzB,KAAM,IAAI,OACN,gDAAgD,EAAK,cAElD,MAAM,QAAQ,EAAK,MAC5B,OAAmB,GAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,6CACA,iBAAgB,KAAK,UAAU,EAAK,YAI5C,MAAM,IAAI,OACN,2DACA,iBAAgB,KAAK,UAAU,EAAK,QAG1C,SAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,kBAE5C,EAAK,gBAAkB,KAGlB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAQ,EAAW,OAGzB,AAAI,MAAO,MAAK,MAAS,UACvB,MAAK,KAAO,CAAC,KAAK,OAEpB,OAAS,GAAI,EAAG,EAAI,KAAK,KAAK,OAAQ,EAAE,EACtC,AAAI,KAAK,KAAK,GAAK,GACjB,MAAK,KAAK,IAAM,GAKpB,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,GAAK,GAAQ,EACtB,KAAM,IAAI,OAAM,iBAAiB,GAGrC,GAAI,KAAK,KAAK,SAAW,GAAqB,KAAK,MAAM,OACvD,KAAM,IAAI,OAAM,4BAA4B,KAAK,MAGnD,GAAM,GAAa,KAAK,KAAK,IAAI,SAAA,GAAQ,MAAA,GAAW,KAE9C,EAAY,GAClB,AAAI,KAAK,MACP,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAE3B,KAAK,MAAQ,KAEf,AAAI,KAAK,OACP,KAAK,KAAO,KAAK,UACb,OAAQ,EAAY,UAAW,KAAK,gBACpC,KAAK,gBAAiB,GAE1B,KAAK,KAAO,KAGd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACQ,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAQ,EAAW,OAEzB,MAAO,GAAA,KAAK,WAIV,OAHM,GAAW,GACb,EAAA,EAAA,QAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,SACL,EAAiB,GAA2B,EAAG,GACnC,EAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAe,GAAO,EAAW,GAuBnC,OApBM,GAAY,SAAC,GACjB,MAAI,IAAK,MAAQ,EAAE,MAAM,SAAW,GAChC,EAAK,OAAS,CAAC,EAAQ,GAClB,EAAE,QAAQ,GAEV,GAIP,EAAQ,EAAU,EAAK,MAAM,QAC7B,EAAS,EAAU,EAAK,KAAK,QAQ3B,EAA0B,GAC1B,EAA8B,GAC3B,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,AAAK,EAAK,KAAkB,QAAQ,KAAO,GACzC,GAAc,KAAK,EAAW,IAC9B,EAAkB,KAAK,IAEvB,GAAc,KAAK,GACnB,EAAkB,KAAK,EAAW,KAGtC,SAAO,EAAK,KAAK,GACjB,EAAW,EAAS,KAAK,GACzB,EAAQ,EAAM,KAAK,GACnB,EAAS,EAAO,KAAK,GAEd,GACH,EAAO,EAAM,EAAU,EAAQ,EAAO,EAAK,YAInD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,mBAExC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjKF,EAAA,UAAY,qBAmKrB,GArKwC,IAsKxC,EAAA,cAAc,cAAc,IC9nB5B,YAoEI,EAAW,EACX,GACF,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,kEACG,GAAE,KAAI,eAMf,GAHA,AAAI,GAAW,MACb,GAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,KAErB,EAAQ,SAAW,GAAK,EAAQ,GAAG,SAAW,GAC9C,EAAQ,GAAG,SAAW,EACxB,KAAM,IAAI,GACN,+GAON,GAHA,AAAI,GAAc,MAChB,GAAa,MAEX,IAAe,gBAAkB,IAAe,gBAClD,KAAM,IAAI,GACN,wBAAwB,EAAU,mEAIxC,GAAI,GACJ,MAAI,KAAe,gBACjB,EAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,IAE/C,EAAU,CAAC,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,GAAI,CAAC,EAAG,IAG1C,EAAA,IAAQ,EAAG,KA6BtB,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAMjC,WAAY,GAAZ,GAAA,GAAA,KAUE,GATA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,WACD,EAAK,YAAc,KAAO,KAAoB,EAAK,WAGnD,EAAK,SAAW,KAClB,EAAK,QAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,YACnB,MAAO,GAAK,SAAY,SACjC,EAAK,QACD,CAAC,CAAC,EAAK,QAAS,EAAK,SAAU,CAAC,EAAK,QAAS,EAAK,eAGvD,GADA,EAAK,QAAU,EAAK,QAChB,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,6DACA,sBAAqB,EAAK,QAAQ,OAAM,YAG9C,GAAI,GAAa,OACb,EAAY,OAChB,GAAI,MAAO,GAAK,QAAQ,IAAO,SAC7B,EAAgB,CAAC,EAAK,QAAQ,GAAI,EAAK,QAAQ,IAC/C,EAAe,CAAC,EAAK,QAAQ,GAAc,EAAK,QAAQ,SAIxD,GAFA,EAAK,QAAU,EAAK,QAEhB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,gEACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAIrD,GAFA,EAAgB,EAAK,QAAQ,GAEzB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,+DACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAErD,EAAe,EAAK,QAAQ,GAE9B,EAAK,QAAU,CAAC,EAAe,GAEjC,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAEhC,GAAI,GACA,EACJ,MAAI,MAAK,aAAe,gBACtB,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,IAE5C,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,MAIlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,IACF,GAAoB,GAAS,EAAK,QAAS,EAAK,eAG1D,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAnGF,EAAA,UAAY,gBAqGrB,GAvGmC,IAwGnC,EAAA,cAAc,cAAc,IC5O5B,YA0CI,EAAW,EAA4B,EACvC,EAAuB,EACvB,GACF,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,IAEhB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAKb,EAAI,GAAsB,EAAG,GAC7B,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MAEf,EAAI,EAAA,QAAY,EAAe,EAAU,EAAS,GAIlD,EAAI,EAAA,QAEA,EAA0B,EAAU,EAAS,GAEnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBAeP,EAAa,EACb,EAAoC,EACpC,EAAyB,GAC3B,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,EAAG,IAEnB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAIb,EAAI,GAAsB,EAAa,GACvC,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MACf,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAExC,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAE1C,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAsBX,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAWtC,WAAY,GAAZ,GAAA,GAAA,KAKE,GAJA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,GAElB,EAAA,EAAA,KAAA,KAAM,IAAK,KACP,MAAO,GAAK,UAAa,SAC3B,EAAK,SAAW,CAAC,EAAK,kBAEpB,MAAM,QAAQ,EAAK,WAClB,EAAK,SAAsB,SAAW,GACvC,MAAQ,GAAK,SAAsB,IAAO,SAC5C,EAAK,SAAW,EAAK,aAErB,MAAM,IAAI,GACN,qGAEA,IAAG,KAAK,UAAU,EAAK,YAG7B,GADA,GAAsB,EAAK,SAAU,YACjC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBAEhB,MAAO,GAAK,SAAY,SAC1B,EAAK,QAAU,CAAC,EAAK,iBAEnB,MAAM,QAAQ,EAAK,UAClB,EAAK,QAAqB,SAAW,GACtC,MAAQ,GAAK,QAAqB,IAAO,SAC3C,EAAK,QAAU,EAAK,YAEpB,MAAM,IAAI,GACN,oGAEA,IAAG,KAAK,UAAU,EAAK,WAG/B,UAAsB,EAAK,QAAS,WAEpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAS,GACX,EAAW,GAAI,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IAChE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAW,KAO5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,EAAS,GAAa,GAAoB,GAAS,GACnD,GAAM,GAAS,EAAK,gBAChB,GAAoB,GAAS,CAAC,EAAK,SAAS,GAAI,GAChD,CAAC,EAAK,QAAQ,GAAI,GAAI,EAAK,QAAS,gBAExC,MAAO,GAAA,QAAY,EAAQ,CAAC,OAIhC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,SAEV,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAxFwC,IA0FxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,IAEtB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,UACrB,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,SAErC,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAKrE,MAJA,GACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,GAErC,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,KAQlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAhFwC,IAkFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,EAAG,IAEzB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,SAAU,EAAK,UACpC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,QAAS,EAAK,SAEnD,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAOrE,MANA,GAAS,GACL,EAAQ,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,EAAM,GAE7C,CAAC,EAAW,GAAI,EAAQ,EAAM,EAAM,EAAW,KAS1D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GArFwC,IAuFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,eAcrB,GAhBkC,IAiBlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,mBAcrB,GAhBsC,IAiBtC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAC5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,EAAW,KAGpC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAEd,GAb8C,IAe9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAG1C,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,KAAS,EAAO,MARpB,EAAA,UAAY,yBAWrB,GAb4C,IAc5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAGtC,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,IAAQ,EAAO,MARnB,EAAA,UAAY,qBAWrB,GAbwC,IAcxC,EAAA,cAAc,cAAc,IAiB5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAE5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GAEjB,MADA,GAAa,EACT,KAAK,aAAe,eACf,CAAC,EAAW,GAAI,EAAW,IAE3B,CAAC,EAAW,GAAI,EAAW,KAItC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAGZ,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,KAAK,YAC3B,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA7B8C,IA+B9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAA5C,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,KAAS,EAAO,CAAC,EAAG,IAEpB,EAAA,KAAS,EAAO,CAAC,EAAG,OAR1B,EAAA,UAAY,yBAYrB,GAd4C,IAe5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAAxC,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,IAAQ,EAAO,CAAC,EAAG,IAEnB,EAAA,IAAQ,EAAO,CAAC,EAAG,OARzB,EAAA,UAAY,qBAYrB,GAdwC,IAexC,EAAA,cAAc,cAAc,IC5rB5B,AA4CA,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,GAAZ,GAAA,GAQE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAKf,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAIE,MAAI,MAAK,OAAS,KACT,KAAK,MAAM,UAEX,QAIX,SAAc,GAIZ,AAAI,KAAK,OAAS,MAChB,MAAK,MAAM,UAAY,oCAI3B,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,MAAM,kDAIpB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAIpB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WAEE,MAAQ,MAAK,MAAc,0CAK7B,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,MAAM,wCAKpB,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,MAAM,cAGpB,EAAA,UAAA,WAAA,SAAW,GACT,KAAK,MAAM,WAAW,IAGxB,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAS,CACP,UAAa,KAAK,MAAM,eACxB,OAAU,KAAK,MAAM,cAGnB,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,OAAS,MAChB,KAAK,MAAM,6BAA6B,IAKrC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAc,EAAO,MACrB,EAAQ,GAAY,EAAa,GACvC,MAAO,GAAO,MACd,GAAM,GAAY,CAAC,MAAK,GACxB,cAAO,OAAO,EAAW,GAClB,GAAI,GAAI,IAEnB,GAvGsC,IAyGtC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,oEACA,gBAAe,KAAK,UAAU,KAEpC,KAAK,UAAY,CAAC,CAAC,MAAO,IAC1B,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAChE,AAAK,KAAK,MAAM,OACd,MAAK,MAAM,MAAM,GACjB,KAAK,MAAM,MAAQ,IAErB,EAAA,UAAM,MAAK,KAAA,KAAC,IAGd,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAC1D,EACF,KAAK,MAAM,mBAAmB,GAC5B,EAAY,EAAW,GAC7B,MAAO,CAAC,EAAiB,GAAI,GAAW,OAAO,EAAiB,MAAM,KAGxE,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,EAAS,GAAoB,GAI7B,GAAM,GAAwB,SAAC,EAAgB,GAK7C,GAAM,GAAS,GAAoB,EAAK,MAAM,KAAK,EAAQ,IAC3D,MAAO,CAAC,EAAQ,KAEZ,EACF,GAAI,EAAM,EAAQ,GAAI,GAAyB,KAC3C,KAAsB,GACtB,IACF,EAAI,EAAW,GAGrB,MAAO,MArDJ,EAAA,UAAY,kBA0DrB,GA5DqC,IA6DrC,EAAA,cAAc,cAAc,gBAEgB,GAC1C,GACI,GAAiC,yBAA0B,GAmBjE,GAAM,IAA2D,SAEjE,GAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAUL,EAAc,EAAK,MAAM,YACzB,EAAqC,GAC3C,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,aAAe,GAAY,GAChC,EAAY,YACR,IAAY,cAAmB,IACnC,GAAM,GAAqC,GAW3C,GAVA,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,cAAgB,GAAY,GACjC,EAAK,aAAa,KAAO,WAAa,EAAK,aAAa,KACxD,EAAK,cAAc,KAAO,YAAc,EAAK,cAAc,KAE3D,EAAK,UAAY,EAAK,YAAc,OAChC,GACA,EAAK,UACT,GAA4B,EAAK,WAC7B,EAAK,QACP,KAAM,IAAI,IACN,mEAEN,SAAK,UAAY,EAAK,MAAM,SAC5B,EAAK,gBAAkB,EAAK,MAAM,gBAClC,EAAK,YAAc,EAAK,MAAM,YAC9B,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAClB,EAAK,UAAY,EAAK,MAAM,UAC5B,EAAK,aAAe,OAGtB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GAIZ,KAAK,WAAa,EAClB,AAAI,KAAK,cAAgB,MACvB,MAAK,aAAa,UAAY,GAEhC,AAAI,KAAK,eAAiB,MACxB,MAAK,cAAc,UAAY,oCAInC,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,aAAa,aAAa,OAClC,KAAK,cAAc,eAGzB,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAa,EAAQ,OACrB,EAAiB,KAAK,MAAM,EAAa,GAC/C,KAAK,aAAa,WAAW,EAAQ,MAAM,EAAG,IAC9C,KAAK,cAAc,WAAW,EAAQ,MAAM,KAG9C,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GACA,KAAK,aAAa,mBAAmB,GACzC,AAAM,MAAM,QAAQ,IAAgB,MAAM,QAAQ,EAAY,KAC5D,GAAc,CAAC,IAEjB,EAAc,EAEd,GAAI,GACA,EACA,EAiBJ,MAhBA,AAAI,MAAK,aACP,GAAa,EAAY,MAAM,IAC/B,EAAc,EAAY,GAI5B,EAAc,EACd,AAAI,KAAK,YAAc,SACrB,GAAY,EAAY,OAAS,IAAM,EACvC,EAAe,CAAC,IACX,AAAI,KAAK,WAAa,KAC3B,EAAe,CAAC,EAAa,EAAY,SAEzC,EAAe,CAAC,GAGd,KAAK,YACH,KAAK,WAAa,KACb,EAAa,OAAO,GAAY,OAAO,EAAW,SAEpD,CAAC,GAAa,OAAO,GAAY,OAAO,EAAW,SAErD,GAA+B,IAGxC,EAAA,UAAA,MAAA,SACI,EACA,GACF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAEX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAU1D,GATA,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAEzB,AAAI,MAAM,QAAQ,IAChB,GAAgB,EAAuC,MAAM,GAC7D,EAAU,EAAuC,IAG9C,IAAgB,MAAQ,EAAa,SAAW,IACjD,GAAa,KACf,MAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,GAE7B,GAAM,GAAiD,GACjD,EAA+B,GACrC,GAAI,GAAgB,MAClB,GAAM,GAAY,EAAa,OAC/B,GAAI,EAAY,EAAI,EAClB,KAAM,IAAI,GACN,iIAIN,EAAO,aAAkB,EACzB,EAAiB,KAAI,MAArB,EAAyB,GACzB,GAAM,GAAc,EACI,IAAI,SAAA,GAAS,MAAA,IAAI,IAAU,CAAC,MAAO,EAAM,UACjE,KAAK,aAAa,UAAY,EAAW,MAAM,EAAG,EAAY,GAC9D,KAAK,cAAc,UAAY,EAAW,MAAM,EAAY,GAC5D,EAAgB,KAAI,MAApB,EAAwB,GAE1B,GAAI,GAAa,KACf,KAAM,IAAI,IACN,yEAKN,OADM,GAAmB,EAAiB,YAAc,IACnC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,YAAkB,MAAmB,EACvC,KAAM,IAAI,GACN,gHAKR,GAAI,GAEF,GAAM,GAAY,CAAC,GAAQ,OAAO,GAC5B,EAAgB,KAAK,UAAU,OAAO,GAUtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GACF,EAAA,UAAM,MAAK,KAAA,KAAC,EAA0C,GAC1D,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAI/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAe,EAAO,aAExB,EACA,EACJ,GAAI,GAAgB,KAClB,EAAI,EAAK,aAAa,KAAK,EAAQ,GACnC,EAAO,EAAK,cAAc,KAAK,EAAQ,QAEvC,GAAM,GAAe,EAAa,MAAM,EAAG,EAAa,OAAS,GAC3D,EAAgB,EAAa,MAAM,EAAa,OAAS,GAC/D,EAAI,EAAK,aAAa,KAClB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KACjD,EAAO,EAAK,cAAc,KACtB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KAGnD,GAAI,GACJ,AAAI,EAAK,aACP,CAAI,MAAM,QAAQ,IAChB,GAAS,EAAE,MAAM,GAAG,OAAQ,EAAkB,MAAM,KAGtD,EAAK,EAAe,GACpB,EAAQ,EAAkB,IAG5B,AAAI,EAAK,iBACP,GAAO,EAAA,QAAY,EAAgB,IAGrC,GAAI,GAcJ,MAbA,AAAI,GAAK,YAAc,SACrB,EAAS,GAAc,CAAC,EAAa,IAChC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,GAAI,EAAA,IAAQ,EAAa,IACrC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,WAAa,MAC3B,GAAS,CAAC,EAAa,IAIrB,EAAK,YACH,EAAK,WAAa,KACZ,EAAoB,OAAO,GAE9B,CAAC,GAAkB,OAAO,GAE5B,KAIX,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,aAAa,cAClB,KAAK,cAAc,eAGrB,EAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,GAAU,KAAK,aAAa,KAAM,WAChC,EAAK,aAAa,MAAM,KAE1B,GAAU,KAAK,cAAc,KAAM,WACjC,EAAK,cAAc,MAAM,KAE3B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,SAAY,EAAyB,GAEnC,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAI,GAcJ,GAbA,AAAI,KAAK,gBACP,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,EAAM,GAEpB,EAAa,EAGf,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,KAAM,MAEpB,EAAa,KAGb,KAAK,aACP,GAAM,GAAS,KAAK,aAAa,OAC3B,EAAsB,EAAO,IAAI,SAAA,GAAS,MAAA,QAChD,MAAI,OAAM,QAAQ,GACT,EAAW,OAAO,GAAW,OAAO,GAEpC,CAAC,GAAY,OAAO,GAAW,OAAO,OAG/C,OAAO,IAIX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,aAAa,iBAAiB,OACtC,KAAK,cAAc,mDAGzB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,aAAa,oBAAoB,OACzC,KAAK,cAAc,sDAKzB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,cAAgB,MACvB,KAAK,aAAa,6BAA6B,GAEjD,AAAI,KAAK,eAAiB,MACxB,KAAK,cAAc,6BAA6B,IAIpD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,UAAa,KAAK,WAGd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAIF,EAAA,WAAP,SACI,EACA,GACF,GAAM,GACF,GAAY,EAAO,OAGvB,GAFA,MAAO,GAAO,MAEV,EAAO,cAAmB,KAC5B,KAAM,IAAI,IACN,4FAIN,GAAM,GAAkC,EACxC,SAAU,MAAW,EACd,GAAI,GAAI,IA7VV,EAAA,UAAY,gBA+VrB,GAjWmC,IAkWnC,EAAA,cAAc,cAAc,IC7kB5B,YAkE2B,GACzB,MAAO,IAAI,IAAW,eA6BJ,GAClB,MAAO,IAAI,IAAI,eAoBI,GACnB,MAAO,IAAI,IAAK,eAuBQ,GACxB,MAAO,IAAI,IAAU,eAwBD,GACpB,MAAO,IAAI,IAAM,eAmBK,GACtB,MAAO,IAAI,IAAQ,eA2BW,GAC9B,MAAO,IAAI,IAAgB,eAyBN,GACrB,MAAO,IAAI,IAAO,eAqBG,GACrB,MAAO,IAAI,IAAO,eAsCY,GAC9B,MAAO,IAAI,IAAgB,eAqBN,GACrB,MAAO,IAAI,IAAO,eAgCY,GAC9B,MAAO,IAAI,IAAgB,eAkCF,GACzB,MAAO,IAAI,IAAW,eA2BK,GAC3B,MAAO,IAAI,IAAa,eAeM,GAC9B,MAAO,IAAI,IAAgB,eAoCF,GACzB,MAAO,IAAI,IAAW,eAoCF,GACpB,MAAO,IAAI,IAAM,eAaK,GACtB,MAAO,IAAI,IAAQ,eAmCY,GAC/B,MAAO,IAAI,IAAiB,eAsBN,GACtB,MAAO,IAAI,IAAQ,eAiBQ,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,eA+BG,GACtB,MAAO,IAAI,IAAQ,eAcK,GACxB,MAAO,IAAI,IAAU,eAyBH,GAClB,MAAO,IAAI,IAAI,eAqBO,GACtB,MAAO,IAAI,IAAQ,eAuBO,GAC1B,MAAO,IAAI,IAAY,eAqBD,GACtB,MAAO,IAAI,IAAQ,eAqBG,GACtB,MAAO,IAAI,IAAQ,eAsBI,GACvB,MAAO,IAAI,IAAS,eAyBF,GAClB,MAAO,IAAI,IAAI,eA0BkB,GACjC,MAAO,IAAI,IAAmB,eAuBG,GACjC,MAAO,IAAI,IAAmB,eA2BF,GAC5B,MAAO,IAAI,IAAc,eAgBM,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eA0BO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAwBO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAYa,GACrC,MAAO,IAAI,IAAuB,eAiBG,GACrC,MAAO,IAAI,IAAuB,eAYD,GACjC,MAAO,IAAI,IAAmB,eAiBG,GACjC,MAAO,IAAI,IAAmB,eAYH,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eA4BN,GAClB,MAAO,IAAI,IAAI,eAgDO,GACtB,MAAO,IAAI,IAAQ,eA0BA,GACnB,MAAO,IAAI,IAAK,eAgDO,GACvB,MAAO,IAAI,IAAS,eA2BI,GACxB,MAAO,IAAI,IAAU,eAgDO,GAC5B,MAAO,IAAI,IAAc,eA8BA,GACzB,MAAO,IAAI,IAAW,eAoCO,GAC7B,MAAO,IAAI,IAAe,eA+DR,GAClB,MAAO,IAAI,IAAI,eAUe,GAC9B,MAAO,IAAI,IAAgB,eAMC,GAC5B,MAAO,IAAI,IAAc,eAiDK,GAC9B,MAAO,IAAI,IAAgB,GAItB,GAAM,IAAkB,GAClB,GAAkB,GAClB,GAAY,GACZ,GAAY,eA2BK,GAC5B,MAAO,IAAI,IAAc,eA2BK,GAC9B,MAAO,IAAI,IAAgB,eAkCA,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,glCClnDU,EAAe,GAC5C,MAAO,IAAuB,EAAO,eAoBJ,EAAe,GAChD,MAAO,IAA2B,EAAO,eAuBvC,EAAe,GACjB,MAAO,IAAkC,EAAO,eAqBd,EAAe,GACjD,MAAO,IAA4B,EAAO,eAcJ,EAAe,GACrD,MAAO,IAAgC,EAAO,eAsCtB,EAAe,GACvC,MAAO,IAAkB,EAAO,eAsCX,EAAe,GACpC,MAAO,IAAe,EAAO,eAwBC,EAAe,GAC7C,MAAO,IAAuB,EAAO,eAuBL,EAAe,GAC/C,MAAO,IAAyB,EAAO,eAsBrC,EAAe,GACjB,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAqBlB,EAAe,GAC9C,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,kSC1TxC,yCCAA,YAqBqB,GACnB,MAAO,IAAI,IAAK,eAYC,GACjB,MAAO,IAAgB,eAYN,GACjB,MAAO,IAAgB,+CChDzB,mBAkBuC,EAAA,EAAA,GAAvC,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KAEE,SAAA,MAAqB,OAErB,SAAA,UAAA,SAAA,SAAS,GACP,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,OAAM,yDAElB,KAAK,MAAQ,GAEjB,GAVuC,IAsEvC,YAAc,EAAiB,GAC7B,MAAO,GAAU,EAGnB,YAAiB,EAAiB,GAChC,MAAO,GAAU,qBAOgB,EAAA,EAAA,GAcjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GAHA,AAAI,GAAQ,MACV,GAAO,IAEL,EAAK,mBACP,KAAM,IAAI,IACN,sEAGN,SAAK,QAAU,EAAK,SAAW,WAC/B,EAAK,SAAW,KAAK,IAAI,EAAK,UAAY,GAC1C,EAAK,SAAW,EAAK,UAAY,EACjC,EAAK,QAAU,EAAK,SAAW,EAC/B,EAAK,KAAO,EAAK,MAAQ,OACzB,EAAK,SAAW,EAAK,SAErB,AAAI,CAAC,OAAQ,MAAO,OAAO,QAAQ,EAAK,QAAU,IAChD,SAAQ,KACJ,uBAAuB,EAAK,KAAI,8CAEpC,EAAK,KAAO,QAGd,AAAI,EAAK,OAAS,MAChB,EAAK,YAAc,GACd,AAAI,EAAK,OAAS,OAInB,EAAK,QAAQ,QAAQ,SAAW,GAHpC,EAAK,YAAc,GAMjB,EAAK,YAAc,GAIvB,AAAI,EAAK,cAAgB,IACvB,GAAK,UAAY,MAIf,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,aAAe,EACpB,AAAI,KAAK,UAAY,KACnB,KAAK,KAAO,KAAK,SAEjB,KAAK,KAAO,KAAK,cAAgB,GAAO,SAAW,mBAIjD,EAAA,UAAA,WAAN,SAAiB,EAAe,mGAC9B,MAAA,CAAA,EAAM,GAAqB,WAE3B,MAFA,GAAA,OACM,EAAU,KAAK,gBAAgB,GACjC,GAAW,KACb,CAAA,GAGF,CAAI,KAAK,YAAY,EAAU,KAAK,SAAU,KAAK,MACjD,MAAK,KAAO,EACZ,KAAK,KAAO,GAGZ,MAAK,OACL,AAAI,KAAK,MAAQ,KAAK,UACpB,MAAK,aAAe,EACpB,KAAK,MAAM,aAAe,gBAM1B,EAAA,UAAA,WAAN,SAAiB,sEACf,MAAI,MAAK,aAAe,GAAK,KAAK,SAChC,QAAQ,IAAI,SAAS,KAAK,aAAY,8BAIlC,EAAA,UAAA,gBAAR,SAAwB,GACtB,AAAI,GAAQ,MACV,GAAO,IAET,GAAM,GAAe,EAAK,KAAK,SAC/B,MAAI,IAAgB,MAClB,QAAQ,KACJ,4BAA4B,KAAK,QAAO,sBACxC,2BAA0B,OAAO,KAAK,KAErC,GAEX,GAzGmC,gBAsJL,GAC5B,MAAO,IAAI,IAAc,MAGd,IAAY,CAAC,cAAa,i8DC9PvC,AA8BA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,WAAA,GAAA,aACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,UACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,aAAA,GAAA,eACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,IAAA,UACA,EAAA,EAAA,SAAA,IAAA,WACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,YAAA,IAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,iBAAA,KAAA,mBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,gBAAA,KAAA,oBA7BU,IAAA,IAAQ,QAyRH,IAAjB,AAAA,UAAiB,GAEf,GAAY,GAAZ,AAAA,UAAY,GAAyB,EAAA,EAAA,OAAA,GAAA,SAAc,EAAA,EAAA,GAAA,GAAA,KAAU,EAAA,EAAA,GAAA,GAAA,OAAjD,EAAA,EAAA,yBAAA,GAAA,wBAAuB,OAFpB,IAAA,IAAQ,KCtTzB,AAmBA,GAAM,IAAwC,eA2BnB,EAAc,GACvC,GAAM,GAAqB,CACzB,SAAU,EACV,SAAU,SACV,OAAQ,GACR,MAAO,GACP,eAAgB,GAGlB,GAAW,GAAQ,cAUW,GAC9B,MAAO,IAAW,eAUS,GAC3B,MAAO,IAAW,GC9EpB,WAwBI,EAAmB,EAAY,EAC/B,GACF,GAAM,GAAa,EAAK,YAAY,GACpC,GAAI,GAAc,EAAW,kBAAoB,QAC/C,GAAM,GAAQ,EAAW,gBACnB,EAAM,EAAW,gBAAkB,EACrC,OACC,EAAW,gBAAkB,OAAY,EAAQ,EACR,EAAW,cACzD,GAAI,EAAW,OAAS,SACtB,MAAO,IACH,EAAK,WAAW,EAAW,iBAAkB,EAAW,GAE9D,GAAI,EAAW,OAAS,WACtB,GAAM,GAAS,EAAK,WAAW,MAAM,EAAO,GAE5C,MAAO,GAAO,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAEvD,GAAM,GACF,GAAU,EAAK,WAAW,MAAM,GAAO,GAAI,EAAW,GACpD,EAAO,EAAO,WACpB,MAAO,GAAW,OAAS,SACvB,EAAK,GACL,EAAA,KAAK,cAAc,EAAO,MAAO,GAEvC,GAAM,GAAY,EAAK,WAAW,GAClC,MAAO,IAAa,EAAU,kBAU5B,EAAc,EACd,GACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAY,EAAQ,kBAAkB,KAAK,SAAA,GAC/C,MAAO,CAAC,CAAC,EAAW,GAAyB,EAAU,MAGzD,MAAO,KAAc,OACjB,EAAW,GAAyB,EAAU,IAAY,GAC1D,mBASF,EAAc,EACd,GACF,MAAO,GAAW,GAAyB,EAAM,EAAQ,+BAUvD,EAAmB,GACf,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEjB,MAAO,CACL,GAAyB,EAAU,GAAW,EAAQ,kBACtD,GAIJ,YAAkC,EAAc,GAC9C,MAAO,AAAE,GAAe,EAAI,IAAI,EAAc,cAGlB,GAC5B,GAAM,GAAQ,EAAK,MAAM,KACzB,GAAI,EAAM,SAAW,EACnB,MAAO,CAAC,EAAM,GAGhB,GAAM,GAAW,EAAM,GACvB,MAAO,CAAC,EAAU,OAAO,EAAM,EAAM,OAAS,iBAW5C,EAAY,EACZ,GACF,GAAI,GAAM,EAAc,MAAO,EAAM,EAAW,GAChD,GAAI,IAAQ,YAEV,EAAM,EAAc,mBAAoB,EAAM,EAAW,GAIzD,OAHM,GAEF,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,IACxB,EAAI,EAAG,EAAI,EAAG,IACrB,EAAgB,GAAG,GAAM,EAAiB,EAAI,GAC9C,EAAgB,GAAG,GAAM,EAAiB,EAAI,EAAI,GAEpD,MAAO,GAET,MAAO,eAYmB,GAC1B,MAAO,GAAO,KAAO,EAAS,EAAA,MAAM,GCrJtC,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,mCCnLxE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,UAC7D,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,YAGjE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC7dxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAElD,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,QAC1D,CAAC,OAAU,mBAAoB,KAAQ,iBAAkB,KAAQ,QACjE,CACE,OAAU,2BACV,KAAQ,yBACR,KAAQ,QAEV,CAAC,OAAU,oBAAqB,KAAQ,OAAQ,KAAQ,YAG5D,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CACR,OAAU,QACV,KAAQ,QACR,KAAQ,QACR,aAAgB,MAGpB,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAAU,CACrD,OAAU,wBACV,KAAQ,sBACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,YAG3C,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,YAE3D,CACE,SAAY,cACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,KACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,iBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,WAG9D,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,wCCnWrE,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAAa,CAC7D,OAAU,yBACV,KAAQ,sBACR,KAAQ,QAEV,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAC/C,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,OAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,WACV,KAAQ,WACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,gBAAiB,KAAQ,gBAAiB,KAAQ,QAAS,CACpE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,eACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,mBACV,KAAQ,gBACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,QAItB,CACE,SAAY,sBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,aAE9C,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,kBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,wBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,6BACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,aACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,QAAS,KAAQ,YAAa,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,yCCpVnD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,WAEtC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,SAAU,aAAgB,GAClE,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,SAAU,aAAgB,IAErE,MAAS,CACP,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAEzD,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,IAEjE,MAAS,CAAC,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAExD,CACE,SAAY,kBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,OACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAAW,CACpD,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,WAE7C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAC3C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,UAC7C,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,SACzC,CAAC,OAAU,eAAgB,KAAQ,eAAgB,KAAQ,wCC3KjE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,YAGnD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,WAEjD,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,cACV,KAAQ,YACR,KAAQ,QACR,aAAgB,IAElB,CACE,OAAU,yBACV,KAAQ,qBACR,KAAQ,UAId,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,UAC/C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,WAE5C,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCChGtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAE3D,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,yCCvC3C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,yBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CACE,SAAY,cACZ,SAAY,QACZ,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CAAC,SAAY,QAAS,SAAY,SAAU,CAC1C,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,YACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,SACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAEvC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,UAAW,CAC1D,OAAU,UACV,KAAQ,SACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,SACR,aAAgB,KAItB,CAAC,SAAY,OAAQ,SAAY,QAAS,OAAU,IAAK,CACvD,SAAY,eACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,0BACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,UACzC,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,yCC3G/C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,wBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,aAE3C,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAAW,CACxD,OAAU,sBACV,KAAQ,qBACR,KAAQ,yCCvDhB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,eACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC9ItB,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,eACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAAW,CAC3D,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCCtItB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,eACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GACrE,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,aACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,gBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCCjJtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,QACrD,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,uCC1GvD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,GAAI,KAAQ,UAAW,KAAQ,WACnD,CAAC,MAAS,GAAI,KAAQ,OAAQ,KAAQ,WAExC,MACI,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAEtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,WAClD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAG3E,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,OAAQ,aAAgB,MAGjE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,eACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,YACpC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CACP,CACE,OAAU,aACV,KAAQ,YACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,WACV,KAAQ,UACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,cACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,mBACV,KAAQ,iBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,YAEpD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGzE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,MACV,KAAQ,MACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,MAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAC/D,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,YACV,KAAQ,kBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,kBAAmB,KAAQ,YAChD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAG5C,CACE,SAAY,gBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCC1NtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,mCChDxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,OACV,KAAQ,SACR,KAAQ,QACR,aAAgB,IAElB,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAGhD,CACE,SAAY,aACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,MACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CAAC,CACR,OAAU,iBACV,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,QACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAAa,CACnD,MAAS,EACT,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,OACV,iBAAoB,eACpB,KAAQ,OACR,KAAQ,cAGZ,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,cAG7C,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,eACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,YAG5D,CACE,SAAY,cACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,iCChIb,AAyCA,GAAA,IAAA,WAWE,aACE,GAAM,GAAM,CACV,GAAY,GAAW,GAAS,GAAa,GAAU,GACvD,GAAY,GAAS,GAAO,GAAO,GAAU,GAAe,GAC5D,GAAW,GAAU,IAEjB,EAA0B,GAAG,OAAM,MAAT,GAAa,EAAI,IAAI,SAAA,GAAM,MAAA,GAAG,QAE9D,KAAK,UAAY,EAAY,OACzB,SAAC,EAAK,GACJ,SAAI,EAAO,UAAY,EAChB,GAET,IAlBN,cAAA,eAAkB,EAAA,WAAQ,KAA1B,WACE,MAAO,MAAK,WAAc,MAAK,UAAY,GAAI,wCAsBjD,EAAA,UAAA,eAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAM,KAChB,EAAuB,GACvB,EAAkB,GAClB,EAAoB,GACpB,EAAQ,EAAQ,OAA8B,SAAC,EAAK,GACxD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,GAAG,WAAW,eACrB,EAAa,KAAK,EAAI,EAAK,OACtB,AAAI,EAAK,KAAO,QACrB,EAAQ,KAAK,EAAI,EAAK,OACjB,AAAI,GAAK,OAAS,MAAQ,EAAK,MAAM,SAAW,IACrD,EAAU,KAAK,EAAI,EAAK,OAEnB,GACN,IAEC,EAAiB,GACf,EAAkB,GACpB,EAA8C,GAC9C,EAA+C,GACnD,AAAI,GAAa,MACf,GAAqB,KAAK,oBAAoB,EAAU,QACxD,EAAsB,KAAK,oBAAoB,EAAU,UAE3D,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAMlC,AAAI,OAAO,KAAK,GAAqB,SAAW,EAC9C,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,AAAI,EAAK,SAAS,SAAW,GAC3B,EAAQ,KAAK,KAIjB,OAAO,KAAK,GAAqB,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,aAAe,EAAoB,GACxC,EAAQ,KAAK,MAKnB,AAAI,OAAO,KAAK,GAAoB,OAAS,EAC3C,OAAO,KAAK,GAAoB,QAAQ,SAAA,GAC/B,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GACF,GAAK,aAAe,EAAmB,GACvC,EAAO,KAAK,MAIhB,EAAS,EAGX,GAAI,GAAY,GAChB,AAAI,EAAM,SAAW,MAAQ,EAAM,QAAQ,UAAY,MACrD,GAAY,EAAM,QAAQ,SAAS,OAAO,SAAC,EAAW,GACpD,SAAU,EAAK,UAAU,MAAQ,EAAK,YAAY,GAC3C,GACN,KAGL,GAAM,GACF,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,EAAE,UAAS,GAExE,MAAI,GAAU,OAAS,GACrB,GAAO,UAAY,GAGd,GAGD,EAAA,UAAA,oBAAR,SAA4B,GAC1B,MAAO,QAAO,KAAK,GAAW,IACzB,OAAgC,SAAC,EAAM,GACtC,SAAK,EAAQ,GAAM,MAAQ,EACpB,GACN,KAGD,EAAA,UAAA,QAAR,SAAgB,GAGd,GAAM,GACF,GAAgB,EAAK,KAAO,KAAK,UAAU,EAAK,KAAO,GAC3D,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAGd,GAAM,GAAgB,CACpB,KAAM,EAAK,KACX,GAAI,EAAK,GACT,SAAU,EAAO,SACjB,WACK,GAAK,OACL,IAAI,IAAI,SAAA,GAAS,MAAA,GAAM,WAAW,KAAO,EAAM,OAAO,GAAK,IAChE,OAAQ,GACR,SAAU,GACV,YAAa,GACb,WAAY,GACZ,SAAU,EAAK,MAGjB,MAAI,GAAO,QAAU,MACnB,GAAQ,YACJ,EAAO,OAAO,OACV,SAAC,EAAK,GACJ,SAAI,EAAM,MAAQ,CAChB,KAAM,EAAM,KACZ,gBAAiB,EAAM,MACvB,cAAe,EAAM,KAEhB,GAET,KAEV,AAAI,EAAO,OAAS,MAClB,GAAQ,WACJ,EAAO,MAAM,OAAoC,SAAC,EAAK,GACrD,GAAM,GAAO,EAAM,KACf,EAAQ,OACZ,OAAQ,EAAM,UACP,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAChB,EAAM,cAAgB,GAC3B,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,aACA,UACH,cAEA,KAAM,IAAI,OACN,2BAA2B,EAAM,KAAI,YAAY,EAAK,IAE9D,SAAI,EAAM,MAAQ,CAAC,MAAK,EAAE,KAAI,GACvB,GACN,KAEF,GAID,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAU,EAAY,QACtB,EAAuB,GACvB,EAAkB,GACpB,EAA+B,GACnC,AAAI,GAAW,MACb,GAAQ,EAAQ,OAA8B,SAAC,EAAK,GAClD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,KAAO,SACd,EAAQ,KAAK,EAAI,EAAK,OAEjB,GACN,KAEL,GAAM,GAAiB,GACjB,EAAkB,GAExB,EAAY,UAAU,SAAS,QAAQ,SAAA,GAC9B,GAAA,GAAA,GAAA,EAAA,MAAA,GACD,EAAa,CACjB,KAAM,EACN,GAAI,cACJ,OAAQ,GACR,WAAY,GACZ,SAAU,QACV,YAAa,GACb,WAAY,CAAC,MAAO,CAAC,MAAO,GAAgB,EAAI,MAAO,KAAM,UAC7D,SAAU,IAEZ,EAAK,aAAe,EAAI,KACxB,EAAO,KAAK,GACZ,EAAM,GAAY,IAGpB,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAIlC,GAAM,GAAgB,EAAY,IAElC,EAAY,UAAU,UAAU,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,cAAgB,EACrB,EAAQ,KAAK,MAIjB,GAAM,GAAY,KAAK,mBAAmB,GAC1C,MAAO,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,IAG1D,EAAA,UAAA,mBAAR,SAA2B,GAA3B,GAAA,GAAA,KAEE,MAAO,CACL,WAAY,EAAY,UAAU,KAClC,OAAQ,EAAY,UAAU,SAAS,OACnC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,GACjC,GAET,IACJ,QAAS,EAAY,UAAU,UAAU,OACrC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,EAAK,EAAY,KAClD,GAET,MAIA,EAAA,UAAA,mBAAR,SACI,EACA,GACF,GAAI,GAAO,EAAI,KACf,MAAI,IAAW,MACb,GAAO,EAAQ,IAEV,CAAC,KAAI,EAAE,MAAO,EAAI,OAE7B,iBAE6B,GAC3B,GAAM,GAAS,EAAA,MAAM,OACrB,GAAI,MAAO,GAAO,MAAS,YACzB,MAAO,GAAO,KAAK,GACd,GAAI,MAAO,SAAW,YAC3B,MAAO,IAAI,QAAO,EAAM,UAAU,WAElC,KAAM,IAAI,OACN,gGAKyB,EAAc,GAC7C,GAAM,GACF,MAAM,QAAQ,GAAK,OAAO,aAAa,MAAM,KAAM,GAAK,GAAa,GACzE,MAAO,GAAW,EAAQ,EAAM,0BAI9B,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,KACJ,GAAiB,EAAM,EAAG,GAE5B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAO,GAAQ,EAAM,EAAI,cAIvB,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,IAAS,GACvB,EACF,EAAM,GAAQ,KAAO,EAAM,EAAQ,EAAM,GAAQ,KAAO,EAAM,EAAO,EACzE,MAAQ,OAAO,IAAU,SAAY,EAAQ,SAAS,EAAO,gBAG/B,GAC9B,AAAI,MAAQ,IAAW,UAErB,GAAQ,GAAoB,IAE9B,OAAQ,OACD,IAAoB,SACvB,MAAO,cACJ,IAAoB,aACpB,IAAoB,aACpB,IAAoB,YACpB,IAAoB,SACvB,MAAO,YACJ,IAAoB,QACvB,MAAO,WACJ,IAAoB,UACvB,MAAO,cACJ,IAAoB,UACvB,MAAO,iBAIP,MAAO,mBAKT,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,EAAM,KAAK,KAEb,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,GAAgB,EAAM,MAExB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,KAC7B,EAAM,KAAK,KAAK,IAAI,SAAA,GAAK,MAAA,IAAgB,KAE3C,cAG6B,GAEpC,MAAI,GAAM,YACR,OAEE,EAAM,KAAO,KACR,EAAM,IAAI,IACb,SAAA,GACI,MAAC,OAAO,GAAI,MAAS,SAAY,EAAI,KAAO,SAAS,EAAI,KAAM,MAElE,eAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MACV,GAAsB,EAAM,OAE9B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,GACO,IAAM,KAAK,GAAK,EAAM,KAAK,EAAE,OAAS,EAAM,KAAK,EACX,EAAM,KAAK,IAClD,IACH,IAAI,SAAA,GAAK,MAAC,OAAO,IAAM,SAAY,EAAI,SAAS,EAAG,MAEnD,cAIL,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAAE,IAAI,SAAC,GACvB,MAAO,IAAiB,EAAG,KAGxB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,MAC7B,EAAM,KAAK,MAAM,IAAI,SAAC,GAC3B,MAAO,IAAsB,KAG1B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAEb,ECtkBT,AA4BA,GAAA,IAAA,WAGE,WACY,EAAoB,EACpB,GAFZ,GAAA,GAAA,KACY,KAAA,KAAA,EAAoB,KAAA,UAAA,EACpB,KAAA,QAAA,EAJI,KAAA,OAAmB,GACnB,KAAA,MAAoC,GAIlD,KAAK,OAAS,EAAK,WAAW,IAAI,SAAA,GAAQ,MAAA,GAAK,SAAS,KACxD,AAAI,EAAK,UAAY,MACnB,MAAK,MAAQ,OAAO,KAAK,EAAK,UACZ,OAAO,SAAC,EAAmC,GAC1C,SAAM,GAAO,EAAK,QAAQ,GACnB,GACN,KAQhB,SAAA,UAAA,SAAR,SAAiB,GACf,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,UAOtC,EAAA,UAAA,QAAR,SAAgB,EAAc,GAC5B,GAAM,GAAQ,KAAK,KAAK,SAAS,GACjC,GAAI,EAAM,QAAU,KAClB,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,SAE9C,GAAI,EAAM,GAAK,MAAQ,EAAM,GAAK,KAChC,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAa,KAAK,KAAK,SAAU,EAAM,GAEhD,GAAI,EAAM,OAAS,KACjB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,MAAQ,KAChB,MAAO,IAAc,KAAK,KAAK,SAAU,EAAM,GAEjD,GAAI,EAAM,MAAQ,MAChB,GAAI,EAAM,KAAK,GAAK,MAAQ,EAAM,KAAK,GAAK,KAC1C,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,OAAS,KACtB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,MAAQ,KACrB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAIlC,MAAO,IAEX,KCtGA,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,cACA,YACA,MACH,MAAO,CAAC,EAAA,IACH,EAAc,IAAK,EAAM,EAAW,GACrC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,UAAW,EAAM,EAAW,SAE3C,eACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,cACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,oBACH,MAAO,CAAC,EAAA,kBACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC7F9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACA,aACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,OAAQ,EAAM,EAAW,SACxC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,YACA,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,eAAgB,EAAM,EAAW,GAC/C,EAAc,eAAgB,EAAM,EAAW,SAEhD,QACH,MAAO,CAAC,EAAA,MACJ,GAAU,EAAK,WAAW,GAAI,EAAW,SAC1C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SACxC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SACzC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAE5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC5K9C,YAyBI,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EAAA,KAAK,OACD,GAA8B,EAAQ,GACtC,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlB,EAAc,GAC1D,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,IAAM,EAAG,KAAO,IAAM,EAAG,KAAO,EAAG,GAC/C,MAAO,GAGX,MAAO,GCzCT,AA+BA,GAAA,IAAA,WAIE,WACa,EAAuB,EAAyB,EACjD,EAAiC,EAChC,EAA+B,GAF/B,KAAA,KAAA,EAAuB,KAAA,MAAA,EAAyB,KAAA,QAAA,EACjD,KAAA,aAAA,EAAiC,KAAA,uBAAA,EAChC,KAAA,YAAA,EAA+B,KAAA,eAAA,EANpC,KAAA,QAA6B,GAC7B,KAAA,QAAU,GAMhB,KAAK,SAAW,EAAA,OAAO,GACvB,EAAA,KAAK,KAAK,UAGZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAMd,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,OAAO,MAChD,EAAO,OAAO,YAGlB,KAAK,QAAU,GACf,KAAK,QAAU,GACf,KAAK,SAAS,WAGhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAOtB,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,GAAS,KAAK,OAC7B,KAAM,IAAI,OAAM,4BAA4B,EAAK,wBAC7C,KAAK,QAGX,GAAM,GAAkB,KAAK,QAAQ,GACrC,GAAI,EAAgB,QAClB,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0BACpB,EAAK,wGAIf,MAAI,MAAK,gBACP,GAAgB,QAAU,IAG5B,EAAgB,KAAO,GAChB,EAAgB,QAMzB,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GAAS,MAAA,GAAK,KAAK,MAQxC,EAAA,UAAA,MAAA,SAAM,EAAe,GACnB,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,CAAC,KAAK,aAAe,GAAS,KAAK,QAClD,KAAM,IAAI,OAAM,2BACZ,EAAK,8CAA8C,KAAK,SAG9D,GAAM,GAAI,KAAK,QAAQ,IAAU,GAEjC,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,eACZ,KAAK,KAAI,0CAA0C,EAAK;uCAExD,EAAO,MAAK,8BAA8B,KAAK,MAAK,KAc1D,GAVA,AAAI,KAAK,SAAW,GACf,MAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,IAC7D,MAAK,aAAe,EAAO,OAG7B,GACI,KAAK,aAAc,EAAO,MAC1B,eAAe,KAAK,KAAI,0CACpB,EAAK,KAET,EAAE,KACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,uCAGf,GAAI,EAAE,QACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,0CAGf,EAAE,OAAS,EACX,EAAA,KAAK,GACL,EAAE,QAAU,GAEZ,KAAK,QAAQ,GAAS,GAMxB,EAAA,UAAA,UAAA,SAAU,EAAmB,GAA7B,GAAA,GAAA,KACE,GAAI,EAAQ,SAAW,EAAQ,OAC7B,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,sCACxB,4BACI,EAAQ,OAAM,qCACd,EAAQ,OAAM,MAGxB,EAAQ,QAAQ,SAAC,EAAG,GAAU,MAAA,GAAK,MAAM,EAAG,EAAQ,OAWtD,EAAA,UAAA,OAAA,SAAO,EAAoB,GACzB,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAK,EAMH,EAAU,EAAQ,MAAM,EAAG,KAAK,aALhC,EAAU,GACV,OAAS,GAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAMjB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAKpC,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAAO,gCAElC,EAAA,MAAM,EAAS,IAMxB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAI,KAAK,SAAW,EAClB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAIpC,OADM,GAAU,GACP,EAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAGf,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAC9B,mDACI,KAAK,aAAY,4BAA4B,EAAQ,GAAG,MAAK,KAE9D,EAAA,OAAO,EAAS,IASzB,EAAA,UAAA,QAAA,SAAQ,EAAmB,GACzB,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAGhD,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,CAAC,KAAK,aAAe,GAAY,KAAK,QACxC,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,KAAK,QAAO,KAGtE,KAAK,UAAU,EAAS,EAAA,QAAQ,EAAQ,KAS1C,EAAA,UAAA,MAAA,SAAM,EAAkB,GAAxB,GAAA,GAAA,KACE,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAEhD,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEZ,EAAW,4BAA4B,EAAO,OAGpD,GAAI,CAAC,KAAK,aAAe,EAAO,SAAW,KAAK,QAC9C,KAAM,IAAI,OACN,2DACI,KAAK,QAAO,QAAQ,EAAO,OAAM,kEAI3C,GAAM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,GAC1B,EAAA,KAAK,WACH,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,EAAK,cAE3D,MAAO,KAGT,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAQ,GAAK,EAEf,KAAK,UAAU,EAAS,IAE5B,KC3TA,AAoCA,GAAA,IAAA,WAeE,WACa,EAA4B,EAC5B,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IADxB,KAAA,QAAA,EAA4B,KAAA,aAAA,EAC5B,KAAA,aAAA,EACX,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,IAAiB,EAAO,MAC1B,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,EAAO,OAEhD,GACI,EAAc,EAAO,MAAO,+BAEhC,EAAA,KAAK,KAGT,KAAK,SAAW,EAAA,OAAO,GACvB,KAAK,eAAiB,EACtB,EAAA,KAAK,KAAK,UA5BZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAiCvB,EAAA,UAAA,KAAA,WACE,MAAO,IAAI,GACH,KAAK,QAAO,QAAG,KAAK,aAAc,KAAK,eAMjD,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,MACzC,EAAO,YAGX,KAAK,QAAQ,OAAS,EACtB,KAAK,SAAS,WAKhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAUtB,EAAA,UAAA,MAAA,SAAM,EAAwB,EAAwB,GAAtD,GAAA,GAAA,KAEE,GAFoD,AAAA,IAAA,QAAA,GAAA,IAEhD,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,IAAgB,IAAM,KAAK,QAAQ,SAAW,EAChD,KAAM,IAAI,OAAM,kCACZ,EAAW,iCACX,KAAK,QAAQ,OAAM,cAEzB,UACI,EAAc,KAAK,aAAc,+BAC9B,EAAA,KAAK,WACV,GAAM,GACF,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAA,QAAQ,EAAQ,KAC/C,MAAO,GAAA,MAAM,EAAiB,MASlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAC9B,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAG9C,GAAI,KAAK,SAAW,EAClB,KAAM,IAAI,OAAM,qCAGlB,GAAM,GAAS,KAAK,QAAQ,MAC5B,UACI,EAAO,MAAO,EAAc,+BACzB,EAAA,QAAQ,EAAQ,IAOzB,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAM9C,GAHA,GACI,EAAO,MAAO,KAAK,aAAc,+BAEjC,KAAK,iBAAmB,KAAK,OAC/B,KAAM,IAAI,OAAM,4CAElB,EAAA,KAAK,GACL,KAAK,QAAQ,KAAK,IAOpB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAO,EACT,KAAM,IAAI,OACN,0DAA0D,GAGhE,GAAI,KAAK,iBAAmB,IAAM,EAAO,KAAK,eAC5C,KAAM,IAAI,OAAM,+BACZ,EAAI,6BAA6B,KAAK,eAAc,KAE1D,KAAK,QAAQ,OAAS,GASxB,EAAA,UAAA,QAAA,SAAQ,EAAsB,EAAwB,GAEpD,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,EAAe,GAAK,EAAe,KAAK,QAAQ,OAClD,KAAM,IAAI,OAAM,4BACZ,EAAY,mBAAmB,KAAK,QAAQ,OAAM,cAGxD,GAAI,KAAK,QAAQ,IAAiB,KAChC,KAAM,IAAI,OAAM,oBAAoB,EAAY,aAGlD,UACI,KAAK,QAAQ,GAAc,MAAO,EAClC,+BAEG,KAAK,QAAQ,IAQtB,EAAA,UAAA,QAAA,SAAQ,EAAsB,GAC5B,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAG9C,GAAI,EAAe,GACf,KAAK,iBAAmB,IAAM,GAAgB,KAAK,eACrD,KAAM,IAAI,OAAM,yBACZ,EAAY,uBAAuB,KAAK,eAAc,cAG5D,GACI,KAAK,aAAc,EAAO,MAAO,+BACrC,EAAA,KAAK,GACL,KAAK,QAAQ,GAAgB,GAU/B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAwB,GAAlD,GAAA,GAAA,KAEE,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAU9C,MAPA,IACI,KAAK,aAAc,EAAc,+BAIrC,EAAU,EAAQ,MAAM,EAAG,KAAK,QAE5B,EAAQ,SAAW,EACd,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAK,QAAQ,GAAI,KAC1D,MAAO,GAAA,MAAM,EAAS,MAS1B,EAAA,UAAA,OAAA,SAAO,EAAwB,GAA/B,GAAA,GAAA,KACE,GAAI,CAAC,CAAC,GAAgB,IAAiB,KAAK,aAC1C,KAAM,IAAI,OAAM,uBACZ,KAAK,aAAY,+BAA+B,GAMtD,MAHA,IACI,KAAK,aAAc,EAAc,+BAEjC,KAAK,SAAW,EACX,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAK,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAG,KACjD,MAAO,GAAA,OAAO,EAAS,MAG7B,iBAQI,EAAgB,EAAwB,GAC1C,GAAM,GAAQ,EAAO,MACrB,GAAI,EAAO,MAAM,OAAS,EACxB,KAAM,IAAI,OACN,oDAAoD,EAAO,OAEjE,GAAI,EAAO,QAAU,EACnB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,GAEzC,GAAM,GAAc,EAAO,MAAM,MAAM,GACvC,GACI,EAAa,EAAc,+BAE/B,GAAM,GAAuB,EAAA,QAAQ,GACrC,MAAO,IAAI,IAAW,EAAY,EAAc,eAU9C,EAAwB,EAAwB,GAClD,MAAO,IAAI,IAAW,GAAI,EAAc,EAAc,eAWpD,EAAgB,EAAmB,EACnC,GACF,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,GAAe,MAAQ,IAAgB,IAAM,GAAY,EAC3D,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,EAAW,KAGrE,GAAM,GAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,GACtD,EAAU,EAAA,QAAQ,EAAQ,GAChC,SAAQ,QAAQ,SAAC,EAAO,GACtB,EAAK,QAAQ,EAAO,EAAQ,MAEvB,cAWL,EAAgB,EAAkB,GACpC,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEV,EAAW,4BAA4B,EAAO,OAmBtD,OAhBM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,EAAA,KAAK,WAC7B,GAAM,GAAU,GAChB,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,GAEtD,SAAO,UACA,IAGH,EAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,EAAO,QAE1D,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAK,QAAQ,EAAG,EAAQ,IAE1B,MAAO,GC3YT,AAiBA,GAAA,IAAA,OAUa,GAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,kUACnB,EAAA,EAAK,iBACN,KAAA,MAAA,CAAA,EAAA,OACA,cAAA,MAAA,CAAA,EAAA,OAgBA,QAAA,MAAA,CAAA,EAAA,OACA,iBAAA,MAAA,CAAA,EAAA,OAsDA,WAAA,MAAA,CAAA,EAAA,OAIA,SAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QAOA,OAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAoBA,qBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAQA,sBAAA,MAAA,CAAA,EAAA,QAUA,uBAAA,MAAA,CAAA,EAAA,QAWA,sBAAA,MAAA,CAAA,EAAA,QAQA,qBAAA,MAAA,CAAA,EAAA,QAWA,oBAAA,MAAA,CAAA,EAAA,QAMA,qBAAA,MAAA,CAAA,EAAA,QAOA,oBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAaA,sBAAA,MAAA,CAAA,EAAA,QACA,oBAAA,MAAA,CAAA,EAAA,QAcA,oBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAYA,kBAAA,MAAA,CAAA,EAAA,QAYA,uBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAUA,qBAAA,MAAA,CAAA,EAAA,QASA,oBAAA,MAAA,CAAA,EAAA,QAUA,kBAAA,MAAA,CAAA,EAAA,wBAzTG,SACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClC,CAAA,EAAM,EAAK,eAC7B,MADM,GAAY,GAAA,OACd,EAAU,GACZ,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,gBAE1C,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAKtC,SACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAO,EAAc,OAAQ,EAAM,EAAW,GAI/C,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAFxC,SACD,GAAA,OAEC,EAAS,EAAK,IAAI,SAAA,IAAU,MAAA,IAAO,KACzB,CAAA,EAAM,EAAW,GAAG,eAAhC,EAAY,GAAA,OAEhB,EAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAChD,GAAO,YAIP,EAAmB,gFAIf,UAAa,EAEV,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACzC,EAAQ,EAAQ,eAAgB,EAAQ,uBAD5C,SAAS,GAAA,OAEH,GAAY,EAAO,IAAI,SAAA,IAAU,MAAA,IAAO,KAI9C,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,YAMN,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAQ,EAAQ,eAAgB,EAAQ,uBAF1C,UACD,GAAA,OAEO,CAAA,EAAM,GAAW,GAAG,eAAhC,SAAY,GAAA,OAEZ,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,4CA1BN,GAAU,WAAE,CAAA,EAAA,wCA8BnB,MAAA,CAAA,EAAO,UAGD,SAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAChD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClD,AAAK,EAAK,MACR,GAAO,GAAY,IAGb,CAAA,EAAM,EAAK,gBAAnB,MAAA,CAAA,EAAQ,GAAA,OAAmB,GAAK,CAAC,OAAW,GAAQ,CAAC,EAAM,iBAK3D,MAFM,GAAY,EAAK,WAAW,KAC9B,SAAA,IAAQ,MAAA,IAAU,GAAM,EAAW,KAAa,SAChD,EACI,GAAO,GAAU,EAAW,EAAW,GAC7C,CAAA,EAAO,CAAC,GAAY,MAEtB,CAAA,EAAO,gBAGD,SACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,WAAW,GACnB,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,YACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,gBACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,yBAA0B,EAAM,EAAW,GAEvD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAc,GAAI,IACpB,EAAM,EAAO,EAAM,EAAc,EAAwB,EACzD,GACJ,EAAQ,eAAe,GACvB,CAAA,EAAO,CAAC,EAAY,SAAU,EAAA,OAAO,aAG/B,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAmB,EAAQ,eAAe,EAAG,IACnD,EAAiB,MAAM,EAAO,GAC9B,CAAA,EAAO,CAAC,EAAiB,mBAGnB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAkB,EAAQ,eAAe,EAAO,IACtD,CAAA,EAAO,CAAC,EAAgB,KAAK,aAGvB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAoB,EAAQ,eAAe,EAAS,IAC1D,CAAA,EAAO,CAAC,EAAkB,OAAO,EAAe,aAG1C,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAqB,EAAQ,eAAe,GAAU,IAC5D,GAAmB,QAAQ,GAAgB,IAC3C,CAAA,EAAO,CAAC,GAAmB,mBAGrB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAoB,EAAQ,eAAe,GAAS,IACpD,GACF,EAAc,QAAS,EAAM,EAAW,GAC5C,CAAA,EAAO,CAAC,GAAkB,OAAO,cAG3B,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,MAAM,GAAS,IAChC,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAkB,EAAQ,eAAe,GAAO,IACtD,CAAA,EAAO,CAAC,EAAA,OAAO,GAAgB,OAAQ,mBAGjC,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,gBACjB,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,QAAQ,EAAO,GAC1B,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAE7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAW,EAAc,cAI9C,UACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GACF,GAAQ,GAAe,GAAgB,EAAc,IACzD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,GAAQ,EAAc,GAAc,IACvD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,EAAS,IAClD,CAAA,EAAO,CAAC,GAAW,OAAO,EAAe,GAAc,aAGjD,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,MAAM,EAAc,GAAc,cAG/C,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,GAAW,GAAQ,EAAc,IACpD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAC5C,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,CAAA,EAAO,CAAC,GAAW,OAAO,GAAa,aAGjC,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,SAAS,GACpB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAc,cAGnC,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,UAAW,EAAM,EAAW,GAExC,GAAa,GAAM,GAAa,GAAS,GAC/C,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGnB,KAAM,WAAU,aAAa,EAAK,GAAE,6BCvW1C,AA2BA,YACI,EAAY,EAA4B,GACpC,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAC7B,EAAc,IAAY,iBAE1B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,yGAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,oFAIR,GAAI,EACF,KAAM,IAAI,OACN,yEAEN,GAAM,GAAS,EAAc,UAAW,EAAM,EAAW,GACnD,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGhB,MAAO,CACL,OAAM,EACN,IAAG,EACH,WAAU,EACV,UAAS,EACT,QAAO,EACP,SAAQ,EACR,eAAc,GAIX,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAQ,EAAyB,EACjC,QAED,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,gBACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAIvB,8BACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,gBAAgB,CAClC,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAGvB,0BACA,mBACH,GAAM,GAAQ,EACI,cAAe,EAAM,EACrB,GAEZ,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GACxC,MAAO,CAAC,EAAA,gBACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAO,CAAC,EAAO,GAAI,EAAO,IAAK,QAEhC,4BACA,mBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cAET,MAAO,CAAC,EAAA,gBACJ,EAAc,QAAS,EAAM,EAAW,GAExC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GAEzC,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,EACnC,EACA,CAAC,EAAU,GAAI,EAAU,GAAI,EAAU,UAExC,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,qBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,sBAAuB,EAAM,EAAW,GAEpD,EAAA,EAAA,kBAAA,EAAA,IAAA,EAAA,EAAA,GAAA,CAAA,EAAA,GAAA,EAAA,IAAA,CAAA,EAAA,GAAA,EAAA,IAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,QAIf,MAAO,CAAC,EAAQ,OAEb,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,cACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,YAAa,EAAM,EAAW,GAG1C,EAAe,EAAQ,GACvB,EAAc,EAAQ,GAGtB,EAAiB,EAAU,GAC3B,EAAgB,EAAU,GAEhC,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAc,GAAc,EAC7B,CAAC,EAAgB,GAAgB,iBAIrC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC/S9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,KAAW,EAAO,EAAO,QAE9B,YACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAM,EAAc,MAAO,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,SAAe,EAAO,EAAM,QAEjC,eACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,YAAkB,EAAQ,EAAY,QAE3C,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OAAa,EAAS,EAAO,EAAS,QAE3C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cAEJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,SAEzC,SACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,MACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,SAGzC,mBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,gBACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,GAExC,QAED,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1H9C,AAiBA,GAAA,IAAA,OAUA,YACI,EAAY,EAA4B,GAC1C,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EAAS,EAAc,SAAU,EAAM,EAAW,GAClD,EACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CACL,MAAK,EACL,OAAM,EACN,cAAa,EACb,aAAY,EACZ,eAAc,EACd,aAAY,GAIT,GAAM,IAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,mHACnB,EAAA,EAAK,iBACN,sBAAA,MAAA,CAAA,EAAA,OAgBA,sBAAA,MAAA,CAAA,EAAA,OAcA,sBAAA,MAAA,CAAA,EAAA,OACA,sBAAA,MAAA,CAAA,EAAA,OAQA,QAAA,MAAA,CAAA,EAAA,OAQA,WAAA,MAAA,CAAA,EAAA,uBA9CG,SAOF,GAAU,EAAM,EAAW,GAN7B,EAAK,EAAA,MACL,EAAM,EAAA,OACN,EAAa,EAAA,cACb,EAAY,EAAA,aACZ,EAAc,EAAA,eACd,EAAY,EAAA,aAGC,CAAA,EAAM,EAAA,MAAY,gCAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,wBAGjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAG3D,EACF,EAAc,qBAAsB,EAAM,EAAW,GAG1C,CAAA,EAAM,EAAA,MAAY,6BAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,sBAIjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAGzD,CAAA,EAAM,EAAA,MAAY,uBACtB,EAAmB,EAAoB,EAAe,EACtD,WAFJ,MAAA,CAAA,EAAA,CAAQ,EAAA,gBAKF,SAAY,EAAA,KACb,EAAc,YAAa,EAAM,EAAW,GAC7C,QACY,CAAA,EAAM,EAAA,WAAiB,WAAjC,SAAM,CAAI,EAAA,QAChB,EAAU,UACV,CAAA,EAAO,UAGP,MAAA,CAAA,EAAO,EAAA,eACH,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,6BC3G1C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAS,EAAA,KAAW,EAAG,EAAG,GAChC,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAS,EAAA,OAAa,GAC5B,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAS,EAAA,OAAa,EAAG,GAC/B,MAAO,CAAC,EAAO,OAAQ,EAAO,iBAG9B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCpDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,GAAU,EAAK,UAEnB,yBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,IAAY,OACjD,cACH,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,QACrC,eACA,mBACA,2BACH,GAAM,GAAO,EAAc,IAAK,EAAM,EAAW,GACjD,MAAO,CAAC,GAAY,QAEjB,YACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,IAAY,SACjC,WACH,GAAM,GACD,EAAc,IAAK,EAAM,EAAW,GACzC,MAAO,CAAC,GAAY,QACjB,QACH,MAAO,CAAC,EAAA,SACH,EAAc,IAAK,EAAM,EAAW,GAAoB,MACzD,cACD,SACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,GAAA,SAAe,EAAE,aACtC,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OAAa,QAClB,QACH,GAAM,GAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,QAAQ,KACJ,kGAEJ,QAAQ,IAAI,GACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,QAAQ,IAAI,MAAM,UAAU,MAAM,KAAK,EAAK,GAAG,YAC9B,MAAM,EAAG,IAE5B,MAAO,CAAC,WAGR,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,eAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,yBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,sBAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,iBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,WAAY,EAAM,EAAW,GACzC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,qBAAsB,EAAM,EAAW,GAEzD,MAAO,CAAC,EAAA,MAAY,cAChB,EAAmB,EAAmB,EACtC,EAA8B,EAC9B,YAGJ,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzE9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,eACH,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACA,WACH,MAAO,CAAC,EAAA,MACJ,EAAc,YAAa,EAAM,EAAW,GAC5C,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACA,oBACA,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,aAAc,EAAM,EAAW,GAC7C,EAAc,aAAc,EAAM,EAAW,SAG9C,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SAExC,eACG,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAE7B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,sFAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,iEAGF,GAAA,GAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEhB,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,KAAM,EACN,WAAY,EACZ,uBAAwB,aAI1B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCjF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,qBACA,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,MACH,MAAO,CAAC,EAAA,2BACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,OAAQ,EAAM,EAAW,SAExC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cACJ,EAAc,gBAAiB,EAAM,EAAW,GAEhD,EAAc,cAAe,EAAM,EAAW,GAC9C,EAAc,eAAgB,EAAM,EAAW,GAE/C,EAAc,eAAgB,EAAM,EAAW,aAInD,KAAM,WAAU,aAAa,EAAK,GAAE,yBC9E9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,EAAW,YAGf,KAAM,WAAU,aAAa,EAAK,GAAE,yBCtH9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,eACA,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACvC,EACA,EAAc,UAAW,EAAM,EAAW,GAC9C,SAAS,EAAO,MAAM,EAAG,GAClB,CAAC,EAAA,OAAa,EAAQ,QAE1B,eACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OAAa,EAAO,EAAA,KAAW,EAAS,SAAU,QAEvD,gBACA,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,QAAc,EAAO,QAE1B,SAEH,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAEhD,EAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,gBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,MAAO,EAAM,EAAW,GACpC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAE/C,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,aACJ,EAAQ,EAAO,EAAK,EAAS,EAAW,EAAS,EACjD,EAAa,QAEd,OACH,MAAO,GAAA,KAAK,WACV,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GAGxC,EAAQ,EAAQ,GAAG,MACnB,EAAgB,EAAA,QAAc,EAAQ,IAAI,MAC1C,EAAS,EAAQ,IAAI,SAAA,GACzB,GAAM,GAAY,EAAA,KAAK,YAAY,EAAO,MAAO,GACjD,GAAI,CAAC,GACD,CAAC,EAAA,KAAK,YACF,EAAA,QAAc,GAAQ,MAAO,GACnC,KAAM,IAAI,OAAM,0CAElB,MAAO,GAAY,EAAS,EAAA,QAAc,EAAQ,KAEpD,MAAO,CAAC,EAAA,MAAY,EAAQ,UAG3B,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GAC7C,MAAO,GAAA,QAAc,EAAQ,OAE1B,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,YACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,kBAAmB,EAAM,EAAW,GAGhD,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,GAAA,MAAY,EAAQ,EAAiB,OAEzC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,UAAgB,EAAS,EAAQ,QAEtC,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,SAAe,EAAG,QAEvB,iBACH,GAAM,GACF,EAAc,gBAAiB,EAAM,EAAW,GAE9C,EACF,EAAc,cAAe,EAAM,EAAW,GAE5C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,MAAO,CAAC,EAAA,cACJ,EAAS,EAAc,EACvB,EAAa,QAAU,EAAa,MAChC,EACA,EAAA,KAAW,EAAc,EAAa,iBAG9C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1K9C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBChDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAGzC,cACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAGzD,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,UAAW,EAAM,EAAW,GAE1C,EAAc,gBAAiB,EAAM,EAAW,SAGjD,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,gBACH,GAAM,GACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACrC,cAEb,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAW,QAEZ,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAG5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnG9C,YAgDI,EAAY,EACZ,GACF,GAAM,GACD,SAAC,EAAY,EAA4B,GACxC,OAAQ,EAAK,cACN,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,cACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAsB,EAAM,EAAW,SAC9C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,UACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAkB,EAAM,EAAW,SACtD,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,gBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAwB,EAAM,EAAW,SAChD,YACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,iBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAyB,EAAM,EAAW,SACjD,SACH,GAAM,GAAW,GAAgB,EAAK,IACtC,GAAI,GAAY,EAAS,eACvB,MAAO,GAAS,eACZ,GAAI,IAAc,EAAM,EAAW,IAEvC,KAAM,WAAU,aAAa,EAAK,GAAE,+BAGtC,KAAM,WACF,eAAe,EAAK,GAAE,yIAI7B,EAAM,EAAW,GACxB,MAAI,aAAiB,SACZ,EAAM,KAAK,SAAC,GAAS,MAAA,GAAG,OAAO,KAEjC,GAAG,OAAO,GCvEnB,GAAA,IAAA,WAME,WACa,EACA,EACA,EACA,GAHA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IAHA,KAAA,UAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,YAAA,EATL,KAAA,YAAc,CAAC,GAAI,EAAG,UAAW,GAAI,YAAa,GAClD,KAAA,SAAmC,CAAC,KAAK,aACzC,KAAA,OAAS,EAQf,KAAK,4BAGC,SAAA,UAAA,SAAR,SAAiB,EAAY,GAC3B,MAAO,CAAC,GAAE,EAAE,UAAS,EAAE,YAAa,IAQtC,OAAA,eAAI,EAAA,UAAA,iBAAc,KAOlB,WACE,MAAO,MAAK,cARd,SAAmB,GACjB,AAAI,KAAK,WAAa,GACpB,MAAK,SAAW,EAChB,KAAK,8DAWT,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,mBAAmB,oCAOjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGN,EAAA,UAAA,0BAAR,WAEE,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,KAAK,SAAS,OAAS,EAAG,KAC5C,GAAM,GAAW,KAAK,SAAS,MAAM,EAAG,KAAK,SAAS,OAAS,GAC/D,EAAM,KAAK,KAAK,qBAAqB,IAEvC,EAAM,KAAK,IACX,KAAK,mBAAqB,GAGpB,EAAA,UAAA,qBAAR,SAA6B,GAC3B,MAAO,GACH,EACK,IACG,SAAA,GAAW,MAAC,GAAQ,KAAO,GAAK,EAAQ,cAAgB,EACpD,GACG,EAAQ,UAAS,IAAI,EAAQ,cACvC,KAAK,KACV,IAON,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,UACP,MAAK,SACL,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,KAAK,KAAK,SAAS,KAAK,OAAQ,IAC9C,KAAK,mBAAmB,QAAQ,KAAK,qBAAqB,KAAK,aAQnE,EAAA,UAAA,UAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,EAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,OAAO,IACrB,KAAK,kBAAkB,YAEvB,MAAM,IAAI,OAAM,4CAQpB,EAAA,UAAA,cAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,GAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SACL,GAAM,GACF,OAAO,OAAO,GAAI,KAAK,SAAS,KAAK,SAAS,OAAS,IAC3D,EAAQ,aAAe,EACvB,EAAQ,GAAK,KAAK,OAClB,KAAK,SAAS,OAAO,GAAI,EAAG,GAC5B,KAAK,mBAAmB,OACpB,EAAG,EAAG,KAAK,qBAAqB,KAAK,eAEzC,MAAM,IAAI,OAAM,0DAIpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,UAAU,IAGxB,EAAA,UAAA,eAAA,SAAe,GACb,KAAK,eAAe,EAAY,IAAM,GAGxC,EAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,eAAe,IAG7B,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,cAAc,EAAW,IAAM,GAGtC,EAAA,UAAA,cAAA,SAAc,GACZ,MAAO,MAAK,cAAc,IAG5B,EAAA,UAAA,QAAA,SAAQ,GACN,OAAW,KAAO,MAAK,eACrB,KAAK,eAAe,GAAK,cAAc,GAGzC,OAAW,KAAO,MAAK,cACrB,KAAK,cAAc,GAAK,cAAc,IAG5C,KC1LA,YAyCI,EAAwB,EAAiB,EACzC,GACF,GAAM,GAAY,GAAI,KAChB,EAA0B,GAC5B,EAAoB,KACpB,EAAuB,KAIrB,EAAO,GAAI,KACX,EACF,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAEpD,EAA0B,GAC9B,AAAI,GAAa,MACf,GAAgB,EAAU,IAAI,SAAA,GAAQ,MAAA,IAAc,EAAK,MAAM,MAIjE,OADM,GAAe,EAAO,QACrB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MAWtB,GAVA,AAAI,IAAc,IAAS,GAAe,KACxC,CAAI,GAAe,MACjB,GAAc,EACd,EAAa,EAAY,SAAS,IAAI,SAAA,GAAS,MAAA,GAAM,OACnC,OAAO,SAAA,GAAQ,MAAA,GAAU,IAAI,OAGnD,EAAU,IAAI,EAAK,MAGf,EAAU,EAAK,OAAS,KAC1B,SAGF,GAAI,EAAe,QAAQ,EAAK,QAAU,GACxC,SAGF,GAAI,EAAc,QAAQ,EAAK,QAAU,GACvC,SAEF,GAAI,EAAK,OAAO,SAAW,GACzB,EAAc,KAAK,EAAK,MACxB,SAEF,EAAK,OAAO,QAAQ,SAAA,GAElB,GAAI,EAAK,IAAI,EAAM,MACjB,OAEF,EAAK,IAAI,EAAM,MACf,EAAS,KAAK,KAGlB,MAAO,CAAC,OAAM,EAAE,QAAO,EAAE,UAAS,EAAE,cAAa,EAAE,YAAW,EAAE,WAAU,eAQxE,EAAc,EACd,GACK,GAAA,GAAA,EAAA,UAAW,EAAA,EAAA,OACZ,EAAmB,GACnB,EAAa,OAAO,KAAK,GACP,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAChC,IAAI,SAAA,GAAQ,MAAA,GAAM,MAAM,KAC1C,EAAY,EAAM,UAExB,EAAW,QAAQ,SAAA,GACjB,AAAI,EAAU,IAAI,EAAM,OACtB,EAAS,KAAK,KAGlB,EAAM,QAAQ,QAAQ,SAAA,GACpB,AAAI,EAAU,IAAI,EAAO,OACvB,EAAS,KAAK,KAGlB,AAAI,GAAa,MACf,EAAU,QAAQ,SAAA,GAChB,AAAI,EAAU,IAAI,EAAK,OACrB,EAAS,KAAK,KAMpB,OAFM,GAAO,GAAI,KACX,EAAuB,GACtB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MACtB,EAAK,IAAI,EAAK,MACd,AAAK,EAAU,EAAK,OAClB,EAAa,KAAK,GAEpB,EAAK,SAAS,QAAQ,SAAA,GACpB,AAAI,CAAC,EAAK,IAAI,EAAM,OAAS,EAAU,IAAI,EAAM,OAC7C,EAAM,OAAO,MAAM,SAAA,GAAS,MAAA,GAAK,IAAI,EAAM,SAC7C,EAAS,KAAK,KAIpB,MAAO,GAGT,GAAM,IAAmB,CACvB,SAAU,QAAS,QAAS,OAAQ,gBAAiB,cACrD,iBAAkB,KAAM,SAEpB,GAAoB,CACxB,sBAAuB,sBAAuB,sBAAuB,qBAGzC,GAC5B,MAAO,IAAiB,QAAQ,EAAK,KAAO,cAGf,GAC7B,MAAO,IAAkB,QAAQ,EAAK,KAAO,ECjK/C,AAkCA,GAAA,IAAA,WAsFE,WAAoB,EAAsB,GAA1C,GAAA,GAAA,KAAoB,KAAA,MAAA,EAAsB,KAAA,OAAA,EArFlC,KAAA,YAAmC,GAAI,KACvC,KAAA,WAA8B,GAM9B,KAAA,UAAY,IACZ,KAAA,WAAqC,GACrC,KAAA,qBAA0D,GA6EhE,KAAK,SAAW,EAAM,QACtB,KAAK,QAAU,EAAM,OACrB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UAExB,AAAI,EAAM,WAAa,MACrB,OAAO,KAAK,EAAM,WAAW,QAAQ,SAAA,GACnC,EAAK,qBAAqB,GACtB,GAAI,GAAc,EAAM,UAAU,GAAO,KApFnD,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,4CAGpD,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,oBACZ,KAAK,sDAG5B,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,gBAGpD,SAAc,GACZ,GAAM,GAAY,OAAO,KAAK,GAAW,IACrC,SAAA,GAAO,MAAA,GAAU,GAAK,IAAI,SAAA,GAAU,MAAA,GAAO,OAC/C,KAAK,WAAa,GAAG,OAAM,MAAT,GAAa,GAC/B,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GACtB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,IAAI,SAAA,GACvB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,cAAgB,EAAK,wCAG5D,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,IAAI,SAAC,GACxB,GAAM,GAAO,EAAK,cAAgB,EAAK,KACvC,MAAO,GAAK,cAAoB,EAAI,IAAI,EAAK,cAAmB,qCAIpE,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAAA,GAAA,GAAA,KACE,MAAO,QAAO,KAAK,KAAK,YAAY,OAAO,SAAC,EAAK,GAC/C,SAAI,GAAO,EAAK,WAAW,GAAK,UACzB,GACN,qCA0BG,EAAA,UAAA,kBAAR,SAA0B,EAAgB,GACxC,GAAM,GAAe,EAAO,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OAC7C,EAAgB,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OACrD,MAAO,GAAa,KAAK,KAAK,WAAa,KACvC,EAAc,KAAK,KAAK,YAOtB,EAAA,UAAA,QAAR,SAAgB,EAAwB,GACtC,GAAM,GACF,GAAqB,EAAQ,EAAS,KAAK,UAAW,KAAK,YACxD,EAAA,EAAA,cAAe,EAAA,EAAA,YAAa,EAAA,EAAA,WACnC,GAAI,GAAe,KACjB,KAAM,IAAI,OACN,qCAAqC,EAAY,KAAI,gBACrD,oBAAmB,EAAY,GAAE,kBACjC,6DACA,qCAAoC,EAAU,MAGpD,GAAI,EAAc,OAAS,GACzB,GAAM,GAAW,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,OAC9B,EAAU,OAAO,KAAK,GAC5B,KAAM,IAAI,OACN,+BAA+B,EAAQ,8BACvC,KAAI,EAAO,qCAAqC,EAAa,MAGnE,MAAO,IACH,KAAK,MAAO,KAAK,UAAW,IAYlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAAhC,GAAA,GAAA,KACE,EAAS,KAAK,UAAU,GACxB,GAAM,GAAQ,OAAO,KAAK,GAAQ,OAClC,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,GAClB,GAAM,GACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC5D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAG/D,AAAI,EAAY,SAAW,GACzB,GAAc,KAAK,UAGrB,GAAM,GAAiB,KAAK,kBAAkB,EAAY,GAGtD,EAAe,KAAK,YAAY,IAAI,GACxC,AAAI,GAAgB,MAClB,GAAe,KAAK,QAAQ,EAAQ,GACpC,KAAK,YAAY,IAAI,EAAgB,IAGvC,GAAM,GAAiC,GACjC,EAA+B,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,GAAI,IAChB,EAAK,UAAW,EAAgB,EAChC,EAAK,qBACH,EAAU,GAAA,GAAwB,EAAK,WAE7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAKzB,OAFM,GAAgB,EAAK,mBAAmB,GACxC,EAA2D,GACxD,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAO,EAAa,GAC1B,GAAI,CAAC,EAAW,EAAK,OACnB,GAAM,GAAU,GAAU,EAAM,EAAY,GAC5C,GAAI,YAAmB,SACrB,KAAM,IAAI,OACN,4BAA4B,EAAK,GAAE,kEAGzC,EAAW,EAAK,MAAQ,EACxB,EAAK,uBACD,EAAK,KAAM,EAAM,EAAY,EAAS,EACtC,EAAiB,IAIzB,MAAI,GAAK,QAAU,MACjB,EAAQ,QAAQ,GAEX,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAY,QAInD,EAAA,UAAA,mBAAR,SAA2B,GACzB,GAAM,GAAM,GAAG,OAAO,MAClB,GACA,OAAO,KAAK,GACP,IAAI,SAAA,GAAO,MAAA,GAAU,KACrB,IAAI,SAAA,GAAW,MAAA,GAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,QACrD,MAAO,IAAI,KAAI,IAET,EAAA,UAAA,uBAAR,SACI,EAAkB,EAAY,EAC9B,EAA2B,EAC3B,EACA,GAGF,GAAI,EAAK,WAAa,WAAa,EAAY,QAAQ,KAAc,GACnE,OAGF,EAAU,GAAU,QAAQ,SAAA,GAC1B,AAAI,GAAU,MACZ,GAAgC,EAAO,IAClC,GAAgC,EAAO,KAAO,GAC/C,EAAK,SAAS,UAGtB,EAAK,OAAO,QAAQ,SAAA,GAGlB,GAAI,EAAM,WAAa,WACrB,GAAM,GACF,GAA6B,EAAM,KAAM,EAAW,GACxD,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,GAAU,CAAC,EAAc,IAAI,EAAO,KACtC,GAAM,GAAQ,EAAgC,EAAO,IACrD,AAAI,IAAU,EACZ,GAAO,UACP,MAAO,GAAgC,EAAO,KACzC,AAAI,GAAS,MAGlB,EAAgC,EAAO,aAkB/C,EAAA,UAAA,aAAN,SAAmB,EAAwB,sEAEzC,MAAA,CAAA,EAAO,KAAK,cAAc,EAAQ,SAiBtB,EAAA,UAAA,cAAd,SACI,EAAwB,EAAmB,EAC3C,EACA,GAF2C,MAAA,KAAA,QAAA,GAAA,IAC3C,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,uGACF,MAAK,IACH,GAAS,KAAK,UAAU,GACxB,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,IAGd,EAAU,GAAI,IAChB,KAAK,UAAW,EAAgB,EAChC,KAAK,qBAKS,CAAA,EAAM,KAAK,uBACzB,EAAQ,EAAS,EAAS,WADxB,SAAY,EAAA,OAEZ,EAAU,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAGzD,EAAY,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAW,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,GAAO,GAAM,KACxD,EACF,GAAI,KAAgB,EAAS,OAAK,EAAa,KAAK,YACxD,OAAO,KAAK,GAAW,QAAQ,SAAA,GAC7B,GAAM,GAAc,EAAU,GAC9B,EAAY,QAAQ,SAAA,GAClB,AAAI,GAAU,CAAC,EAAO,YAAc,CAAC,EAAQ,IAAI,EAAO,KACtD,EAAO,cAKb,AAAI,KAAK,QAAU,MACjB,EAAQ,QAAQ,GAGlB,CAAA,EAAO,SAGH,EAAA,UAAA,qBAAN,SACI,EAAkB,EAClB,mFACI,SAAe,EAAO,OAAO,SAAC,EAAK,EAAQ,GAC/C,SAAI,EAAK,OAAO,GAAO,MAAQ,EACxB,GACN,IAEH,CAAA,EAAO,KAAK,cACR,EAAc,KAAK,YAAa,GAAM,EAAgB,SAS9C,EAAA,UAAA,uBAAd,SACI,EAAwB,EAA2B,EACnD,0IACI,EAAQ,OAAO,KAAK,GACpB,EACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAY,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC9D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAC3D,EACF,GAAqB,EAAQ,EAAa,KAAK,WAD5C,EAAS,EAAA,UAAE,EAAa,EAAA,cAAE,EAAW,EAAA,YAAE,EAAU,EAAA,WAGlD,EACE,EAAU,OAAK,KAAK,MAAM,SAAS,IAAI,SAAA,GACzC,MAAO,CAAC,KAAI,EAAE,SAAU,EAAQ,kBAEhC,EAAU,GAAA,GAAwB,KAAK,WAC7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAEnB,EAA2D,GAC3D,EAAgB,KAAK,mBAAmB,GACxC,EAAkC,0BACjC,GAAM,OAAS,EACd,GAAW,KAAK,aAClB,EAAY,EAAO,EAAS,EAAY,EAAO,EAC/C,EAAiB,EAAiC,GACtD,CAAA,EAAM,QAAQ,IAAI,KAJG,CAAA,EAAA,UAIrB,SAAA,oBAaF,GAXA,AAAI,GAAe,MAAQ,CAAC,GAC1B,QAAQ,KACJ,mIAGA,EACF,EACK,OACG,SAAA,GAAQ,MAAA,CAAC,GAAc,IACnB,CAAC,GAAU,EAAK,KAAM,EAAY,KACzC,IAAI,SAAA,GAAQ,MAAA,GAAK,OACtB,EAAe,OAAS,EACtB,QAAiB,GACrB,AAAI,GAAe,MACjB,GACI,gEACA,4BAA2B,EAAU,MAErC,GAAI,OACN,+BAA+B,EAAc,uBAC7C,YAAW,EAAK,gDAChB,KAAI,EAAa,MAAM,IAE7B,MAAA,CAAA,EAAO,SAGD,EAAA,UAAA,aAAR,SACI,EAAoB,EAA2B,EAC/C,EAA4B,EAC5B,EAA4B,EAC5B,EACA,GAEF,OAPF,GAAA,KAMQ,EAAqC,gBAEzC,GAAM,GAAO,EAAM,MACnB,EAAQ,eAAiB,EAAK,SAC9B,GAAI,GAAW,GAUf,GANA,AAAI,EAAK,KAAK,KAAO,SACjB,EAAc,aAAc,EAAK,KAAM,EAAW,IACnD,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAIC,EAAW,QAAQ,EAAK,QAAU,IACpC,GAAM,GAAU,GAAU,EAAK,KAAM,EAAW,GAChD,AAAK,GACF,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAEH,GAAM,GAAiB,EAAQ,eAC/B,AAAI,YAAmB,SACrB,EAAS,KAAK,EAAQ,KAAK,SAAA,GACzB,SAAU,GAAY,EACtB,EAAQ,eAAiB,EACzB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,GAC1C,KAGT,GAAU,GAAY,EACtB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,QAGnD,GAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,WAxC9C,EAAM,OAAS,OA2CtB,MAAO,IAGD,EAAA,UAAA,kBAAR,SACI,EAAY,EAA2B,EACvC,EAA4B,EAC5B,GACF,EAAK,SAAS,QAAQ,SAAC,GACd,GAAA,GAAA,GAAA,EAAA,KAAA,GAAA,GACP,GAAI,EAAM,IAAa,CAAC,EAAU,IAAI,EAAU,MAC9C,OAGF,AAAI,EAAU,KAAO,QACnB,AAAI,EAAU,WAAW,KAAK,SAAA,GACxB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAExC,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,KAGpD,AAAI,EAAU,WAAW,MAAM,SAAA,GACzB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAE5C,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,QAQ1D,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,OAAO,KAAK,KAAK,WACZ,QACG,SAAA,GAAO,MAAA,GAAK,UAAU,GAAK,QAAQ,SAAA,GAAU,MAAA,GAAO,eAGtD,EAAA,UAAA,uBAAR,SAA+B,GAA/B,GAAA,GAAA,KACE,OAAO,KAAK,GAAQ,QAAQ,SAAA,GAC1B,GAAM,GAAQ,EAAO,GACd,EAAA,GAAA,GAAA,GACD,EAAO,EAAK,MAAM,MAAM,GAC9B,GAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,GAAM,GAAQ,EAAK,WAAW,MAAS,MACjC,EAAQ,EAAM,SAAW,EAAM,MAAM,QACvC,EAAM,MAAM,MACR,SAAC,EAAK,GAAU,MAAA,GAAM,KAAW,IAAM,EAAM,KAAW,IAChE,EAAA,KAAK,OACD,EACA,WAAM,MAAA,sBAAsB,EAAK,KAAI,kBACjC,iCAAgC,EAAK,eACrC,KAAI,EAAM,MAAK,OAEzB,AAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,EAAA,KAAK,OACD,EAAM,QAAU,EAAK,WAAW,MAAS,MACzC,WAAM,MAAA,sBAAsB,EAAK,KAAI,8CAE9B,GAAK,WAAW,MAAS,MAAK,aAAa,EAAM,YAK1D,EAAA,UAAA,UAAR,SAAkB,GAChB,GAAM,GAAyB,GAC/B,OAAW,KAAa,GACtB,GAAI,KAAK,YAAc,MAAQ,KAAK,WAAW,QAAU,MACrD,KAAK,WAAW,OAAO,IAAc,MACvC,GAAM,GAAS,KAAK,WAAW,OAAO,GACtC,EAAO,EAAO,MAAQ,EAAO,OAE7B,GAAO,GAAa,EAAO,GAG/B,MAAO,IAGD,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAa,OAAO,KAAK,GAAQ,OAAO,SAAA,GACrC,GAAA,GAAA,GAAA,GAAA,GACP,MAAO,GAAK,MAAM,MAAM,IAAa,OAEvC,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,OACN,gDACA,WAAU,EAAU,kCAIpB,EAAA,UAAA,WAAR,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GACjB,GAAI,EAAK,YAAc,MAAQ,EAAK,WAAW,SAAW,MACtD,EAAK,WAAW,QAAQ,IAAS,MACnC,GAAM,GAAS,EAAK,WAAW,QAAQ,GACvC,MAAO,GAAO,KAEhB,MAAO,IACN,KAGG,EAAA,UAAA,aAAR,SAAqB,GAArB,GAAA,GAAA,KACE,EAAQ,QAAQ,SAAA,GACP,GAAA,GAAA,GAAA,GAAA,GACP,GAAI,CAAC,EAAK,MAAM,MAAM,GACpB,KAAM,IAAI,OAAM,eAAe,EAAI,kCAI3C,KC3lBA,AAyBO,GAAM,IAAqB,oBACrB,GAAqB,2BAoDhC,WACY,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IADA,KAAA,SAAA,EACA,KAAA,YAAA,EAzCJ,KAAA,QAAU,MA0ChB,AAAI,GAAe,MACjB,MAAK,YAAc,IArCvB,cAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,yCAGd,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,SAAS,4CAGvB,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,6CAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,SAAS,wCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,yCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,2CAoBf,EAAA,UAAA,cAAR,WACE,GAAM,GAAO,KAAK,SAClB,GAAK,EAAsB,MAAQ,KAEjC,KAAK,QAAU,UACN,KAAK,YAAY,aAAe,KACzC,KAAK,QAAU,EAAA,GAAG,mBAAmB,EAAgB,KAAK,kBAE1D,GAAM,GAAW,EAAA,GAAG,gBAAgB,EAAgB,KAAK,aACzD,GAAI,EAAS,SAAW,EAGtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAgB,KAAK,sBAChD,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,CAAC,GAAK,MAEpB,KAAK,QAAU,EAAS,KAQtB,EAAA,UAAA,KAAN,2GAEE,GADA,KAAK,gBACD,KAAK,QAAQ,MAAQ,KACvB,KAAM,IAAI,OACN,iHAGY,MAAA,CAAA,EAAM,KAAK,QAAQ,eAA/B,SAAY,EAAA,OAElB,CAAA,EAAO,KAAK,SAAS,UASvB,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,UAAY,EACjB,GAAM,GAAQ,KAAK,UAAU,cACzB,EAAY,GAChB,AAAI,KAAK,UAAU,qBAAuB,MACxC,GACK,KAAK,UAAU,oBAA4B,WAIlD,KAAK,QAAa,EAAM,SAAS,SAAQ,IAAI,EAAM,SAAS,YAC5D,GAAM,GACF,EAAA,GAAG,cAAc,KAAK,UAAU,WAAY,KAAK,UAAU,aAK/D,GAJA,KAAK,SAAW,GAAI,IAChB,GAAgB,SAAS,eAAe,EAAO,IACnD,KAAK,SAAS,UAAY,KAAK,6BAA6B,GAExD,EAAU,kBAAoB,MAChC,GAAM,GACF,GAAgB,SAAS,eAAe,EAAU,kBACtD,KAAK,YAAc,GAAI,IAAc,GACrC,KAAK,YAAY,UAAY,KAAK,SAAS,UAC3C,KAAK,YAAY,QAAQ,GAAI,IAG/B,MAAO,IA+CH,EAAA,UAAA,KAAN,SAAW,EAAmC,4EAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,OACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,OACN,+GAIN,MAAA,CAAA,EAAO,EAAa,KAAK,KAAK,iBAyChC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,MAAO,MAAK,QAAQ,EAAQ,KAAK,cAG3B,EAAA,UAAA,gBAAR,SAAwB,GAEtB,GAAI,CAAE,aAAkB,GAAA,SAAW,CAAC,MAAM,QAAQ,GAEhD,MAAO,GAGT,GADA,EAAS,MAAM,QAAQ,GAAU,EAAS,CAAC,GACvC,EAAO,SAAW,KAAK,WAAW,OACpC,KAAM,IAAI,OACN,+BACA,wBAAuB,KAAK,WAAW,OAAM,mBAC7C,oBAAmB,EAAO,OAAM,oBAEtC,MAAO,MAAK,WAAW,OAAO,SAAC,EAAK,EAAW,GAC7C,SAAI,GAAc,EAAoB,GAC/B,GACN,KAGG,EAAA,UAAA,iBAAR,SAAyB,GACvB,SAAU,GAAW,KAAK,YACnB,AAAC,MAAM,QAAQ,GAAuB,EAAZ,CAAC,IAmBpC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,EAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GAChC,GAAM,GAAS,KAAK,SAAS,QAAQ,EAAQ,GAC7C,MAAO,GAAO,OAAS,EAAI,EAAS,EAAO,IAkBvC,EAAA,UAAA,aAAN,SACI,EACA,mGACF,SAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GACjB,CAAA,EAAM,KAAK,SAAS,aAAa,EAAQ,WAAlD,SAAS,EAAA,OACf,CAAA,EAAO,EAAO,OAAS,EAAI,EAAS,EAAO,UAGrC,EAAA,UAAA,6BAAR,SAAqC,GACnC,MAAO,QAAO,KAAK,GAAK,OAAO,SAAC,EAAyB,GACvD,SAAO,GAAO,CAAC,EAAI,IACZ,GACN,KAQL,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UAEd,AAAI,KAAK,aACP,KAAK,YAAY,WAGvB,iBAiCI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,6FACF,GAAI,GAAY,KACd,KAAM,IAAI,OACN,0GAGN,MAAI,IAAW,MACb,GAAU,IAGZ,AAAI,EAAQ,WACV,CAAK,EAA0B,MAAQ,MACrC,CAAM,EAAoB,SAAS,MACjC,GAAY,EAAsB,KAEpC,EAAW,GAAG,EAAW,GAAqB,KAG5C,EAAQ,GAAI,IAAW,EAAU,GACvC,CAAA,EAAM,EAAM,eAAZ,SAAA,OACA,CAAA,EAAO,QC7ZT,GAGM,IAAU,8hECwBhB,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAEZ,GAAQ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,wBACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,YAkDwB,EAAY,GAElC,MAAO,IAAgB,EAAO,GAShC,YACI,EAAY,EACZ,EAAiC,GAEnC,GAFE,AAAA,IAAA,QAAA,GAAA,GAA0B,MAAO,AAAA,IAAA,QAAA,GAAA,GAA2B,MAE1D,GAAS,KACX,MAAO,MAET,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAI,EAAK,IAAI,GACX,MAAO,GAAK,IAAI,GAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAGL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,GAChB,OAAW,KAAK,IACd,GAAM,GAAQ,EAAM,GACd,EAAc,GAAgB,EAAO,EAAO,EAAM,GACxD,EAAe,GAAK,EAEtB,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAdzD,UAAK,IAAI,EAAO,EAAO,OAChB,EAAO,kBA2Cd,EAAe,GAAA,MAAA,KAAA,QAAA,GAAA,IACV,GAAgB,EAAQ,GAOjC,YACI,EAAe,EACf,GAAA,AAAA,IAAA,QAAA,GAAA,GAA2B,MAG7B,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAEL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,kBACL,GACT,GAAM,GAAW,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC7B,EAAc,GAAgB,EAAU,EAAO,GACrD,EAAe,GAAK,GAHtB,OAAW,KAAK,KAAL,GAKX,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAbzD,OAAO,GAAO,kBAkBQ,GACxB,MAAI,KAAM,KACD,KAIL,GAAW,EAAE,IACR,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAG,QAAS,gBAsC3B,EAAY,+GACR,EAA+B,GAAI,KAGzC,GAAgB,EAAO,EAAO,OAMZ,EAAA,MAAM,KAAK,EAAK,+BAAhB,GAAA,EAAA,OAAP,GAAG,EAAA,GACN,EAAQ,EAAK,IAAI,GACnB,YAAiB,SACC,CAAA,EAAM,GADxB,CAAA,EAAA,IAFmC,CAAA,EAAA,UAG/B,EAAc,EAAA,OACpB,EAAK,IAAI,EAAK,oBAJA,wBAWZ,SAAS,GAAgB,EAAO,EAAO,GAC7C,CAAA,EAAO,oBASkB,GACzB,MAAO,IAAO,MAAS,CAAC,YAAY,OAAO,IACtC,OAAM,QAAQ,IACb,MAAO,IAAQ,UAAY,CAAE,aAAe,IAAA,qBAYvB,GAC3B,MAAO,IAAO,MAAQ,GAAY,IAAQ,MAAM,QAAQ,IACnD,MAAO,IAAQ,UAAa,YAAe,IAAA,QAC5C,GAAA,KAAQ,aAAa,GAO3B,YAAqB,GACnB,MACI,KAAU,MACT,MAAO,IAAU,UAAY,MAAO,IAAU,WCnRrD,YAqB6B,GAC3B,MAAO,IAAQ,EAAW,IAI5B,YAAuB,GACrB,MAAI,aAAgB,IAAA,OACV,CAAC,MAAO,EAAK,QAAS,QAAS,IAC9B,GAAW,GACb,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAM,QAAS,IChClC,AAqBA,GAAA,IAAA,WAcE,WAAmB,GACjB,GADiB,KAAA,SAAA,EAVT,KAAA,MAAQ,EACR,KAAA,IAAM,EAUV,GAAY,KACd,KAAM,IAAI,YAAW,mDAEvB,GAAI,EAAW,EACb,KAAM,IAAI,YAAW,6CAEvB,KAAK,KAAO,GAAI,OAAS,GACzB,KAAK,gBAAkB,EAAI,EAMnB,SAAA,UAAA,KAAV,SAAe,GAEb,KAAO,EAAQ,GACb,GAAS,KAAK,gBAEhB,MAAO,GAAQ,KAAK,iBAGZ,EAAA,UAAA,IAAV,SAAc,GACZ,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,MAAO,MAAK,KAAK,EAAQ,KAAK,WAGtB,EAAA,UAAA,IAAV,SAAc,EAAe,GAC3B,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,KAAK,KAAK,EAAQ,KAAK,UAAY,GAMrC,EAAA,UAAA,OAAA,WACE,GAAI,GAAS,KAAK,IAAM,KAAK,MAC7B,MAAI,GAAS,GACX,GAAS,KAAK,gBAAkB,GAE3B,GAQT,EAAA,UAAA,OAAA,WACE,MAAO,MAAK,WAAa,KAAK,UAQhC,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,WAAa,GAM3B,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,IAAI,KAAK,IAAK,GACnB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,IAMlC,EAAA,UAAA,QAAA,SAAQ,GACN,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,KAAK,KAOd,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,GAChC,GAAM,GAAS,KAAK,IAAI,KAAK,KAC7B,YAAK,IAAI,KAAK,IAAK,QACZ,GAMT,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GACpC,KAAK,IAAI,KAAK,MAAO,IAMvB,EAAA,UAAA,MAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAS,KAAK,IAAI,KAAK,OAC7B,YAAK,IAAI,KAAK,MAAO,QACrB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GAC7B,GAYT,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAQ,KAAK,KAAK,KAAK,MAAQ,GAC/B,EAAS,KAAK,IAAI,GACxB,YAAK,IAAI,EAAO,KAAK,OACd,GAEX,KC/KA,AAoBA,GAAA,IAAA,SAAA,GAA0C,GAAA,EAAA,GAMxC,mBACE,GAAA,KAAA,KAAM,EAAkB,mBAAiB,KAG3C,SAAA,UAAA,OAAA,WACE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,GACH,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,KAAI,KAAA,KAAC,IAGb,EAAA,UAAA,QAAA,SAAQ,GACN,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,QAAO,KAAA,KAAC,IAMR,EAAA,UAAA,OAAR,WAOE,OANM,GAAc,KAAK,SAAW,EAC9B,EAAU,GAAI,OAAS,GACvB,EAAM,KAAK,SAIR,EAAI,EAAG,EAAI,EAAK,IACvB,EAAQ,GAAK,KAAK,IAAI,KAAK,KAAK,KAAK,MAAQ,IAG/C,KAAK,KAAO,EACZ,KAAK,SAAW,EAChB,KAAK,gBAAkB,EAAI,KAAK,SAChC,KAAK,MAAQ,EACb,KAAK,IAAM,GA7CE,EAAA,iBAAmB,GA+CpC,GAhD0C,ICpB1C,YAuCqC,GACnC,MAAO,IAAI,IAAc,eAyBvB,GAEF,MAAO,IAAI,IAAqB,eAgB9B,EACA,GACF,MAAO,IAAI,IAAgB,EAAe,eAmDxC,EACA,GAAA,MAAA,KAAA,QAAA,GAAgC,GAAgB,MAC3C,GAAI,IAAe,EAAW,GAUvC,GAAA,IAAA,WAAA,cAwBQ,SAAA,UAAA,QAAN,6GACQ,SAAc,GACZ,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,KAAK,gBAAf,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAcH,EAAA,UAAA,eAAN,+GACQ,SAAS,KAAK,SAAS,KACvB,EAAc,GACZ,CAAA,EAAM,EAAO,eAAjB,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,EAAO,gBAAjB,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAUH,EAAA,UAAA,aAAN,2GACU,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACR,CAAA,EAAM,KAAK,eAAf,SAAI,EAAA,oCAWF,EAAA,UAAA,aAAN,SAAmB,qGACT,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,8BACzB,CAAC,EAAE,MAAS,EACd,CAAA,EAAM,KAAK,QADiB,CAAA,EAAA,UAChC,SAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,oCAgBjC,EAAA,UAAA,aAAA,SAAa,GACX,MAAO,IAAI,IAA0B,KAAM,IAa7C,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,IAAI,IAAe,KAAM,IAWlC,EAAA,UAAA,IAAA,SAAO,GACL,MAAO,IAAI,IAAY,KAAM,IAW/B,EAAA,UAAA,SAAA,SAAY,GACV,MAAO,IAAI,IAAiB,KAAM,IAWpC,EAAA,UAAA,eAAA,SAAkB,GAChB,MAAO,IAAI,IAAiB,KAAM,GAAW,UAW/C,EAAA,UAAA,QAAA,SAAW,GACT,MAAO,IAAI,IAAgB,KAAM,IAQ7B,EAAA,UAAA,aAAN,SAAmB,sEACjB,MAAA,CAAA,EAAO,KAAK,IAAI,GAAG,qBAUf,EAAA,UAAA,cAAN,SAAoB,sEAClB,MAAA,CAAA,EAAO,KAAK,eAAe,GAAG,aAAa,SAAA,GAAK,MAAC,KAAM,WAqBzD,EAAA,UAAA,cAAA,SAAc,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,IACxB,GAAI,IAAsB,KAAM,EAAW,IAmCpD,EAAA,UAAA,iBAAA,SACI,EAAmB,EAEnB,GAFmB,AAAA,IAAA,QAAA,GAAA,IAEnB,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAAa,KAAK,cAAc,EAAW,GAGjD,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,MAaxC,EAAA,UAAA,YAAA,SACI,EACA,GACF,MAAO,IAAI,IACP,GAAkB,CAAC,KAAM,IAAY,IAU3C,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAShC,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAYhC,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,IAAI,IAAiB,KAAM,IAapC,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAC1B,MAAO,IAAI,IAAgB,KAAM,EAAY,IAO/C,EAAA,UAAA,OAAA,WACE,MAAO,IAAI,IAAe,OAE9B,KAWA,GAAA,SAAA,GAA+B,GAAA,EAAA,GAE7B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,MAAA,EADd,EAAA,KAAO,IAKf,SAAA,UAAA,QAAA,WACE,MAAO,YAAY,KAAK,MAAM,OAAM,UAGhC,EAAA,UAAA,KAAN,oFACE,MAAI,MAAK,MAAQ,KAAK,MAAM,OAC1B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAO,KAAK,MAAM,KAAK,MAC7B,KAAK,OACL,CAAA,EAAO,CAAC,MAAO,GAAU,GAAO,KAAM,WAE1C,GAlB+B,IAoB/B,GAAA,SAAA,GAAsC,GAAA,EAAA,GACpC,WACc,GADd,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,OAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAO,iBAGH,EAAA,UAAA,KAAN,8EACE,IACE,MAAA,CAAA,EAAO,KAAK,gBACL,GAEP,QAAE,QACE,mDAAmD,EAAE,QACnD,iBAGZ,GApBsC,IAsBtC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,8EACE,MAAA,CAAA,EAAO,KAAK,SAAS,aAEzB,GA1BgC,IA4BhC,GAAA,SAAA,GAA8B,GAAA,EAAA,GAQ5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAF3D,EAAA,MAAQ,EAIN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,iHAKS,MAAK,QAAU,KAAK,SACT,CAAA,EAAM,KAAK,SAAS,QADH,CAAA,EAAA,UAGjC,MAFM,GAAU,EAAA,OAEZ,EAAQ,KACV,CAAA,EAAO,GAET,IAAA,QAAW,EAAQ,qBAErB,MAAA,CAAA,EAAO,KAAK,SAAS,cAEzB,GAzC8B,IA2C9B,GAAA,SAAA,GAA8B,GAAA,EAAA,GAE5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAD3D,EAAA,MAAQ,IAKR,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,8EACE,MAAI,MAAK,SAAW,KAAK,SACvB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,CAAA,EAAO,KAAK,SAAS,aAEzB,GAhB8B,IAqB9B,GAAA,SAAA,GAAuC,GAAA,EAAA,GAKrC,WACc,EAAqC,EACrC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EAAqC,EAAA,UAAA,EACrC,EAAA,qBAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,qBAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,6GACQ,EAAa,0BACZ,GAAM,OAAS,KAAK,UACZ,CAAA,EAAM,KAAK,SAAS,QADC,CAAA,EAAA,UAElC,MADM,GAAO,EAAA,OACT,EAAK,KACH,KAAK,sBAAwB,EAAM,OAAS,EAC9C,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,KAE9B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAM,KAAK,EAAK,qBAElB,MAAA,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,WAEhC,GAvCuC,IAyCvC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,2GAEiB,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,MAAQ,KAAK,UAAU,EAAK,OACnC,CAAA,EAAO,GAET,IAAA,QAAW,EAAK,qCAGtB,GAlCgC,IAoChC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAC9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,WAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAc7B,IAZM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAOzD,EAAS,KAAK,UAAU,EAAK,OAC7B,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCgC,IAqChC,GAAA,SAAA,GAA2C,GAAA,EAAA,GAEzC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,QAAA,EAHd,EAAA,MAAQ,EAKN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,oBAO7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGR,EAAA,UAAA,WAAN,yJAGa,CAAA,EAAM,KAAK,SAAS,eAA3B,MAAA,CAAA,EAAO,EAAA,eAEP,kBAAI,AAAC,KAAK,QAAQ,SAChB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,+CAUrC,GA1C2C,IA4C3C,GAAA,SAAA,GAAqC,GAAA,EAAA,GACnC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,KACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAe,GAAA,YAAe,sBAAsB,EAAK,OAOhD,CAAA,EAAM,KAAK,UAAU,EAAK,gBAKzC,IALM,EAAS,EAAA,OACT,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCqC,IAgDrC,GAAA,SAAA,GAAmD,GAAA,EAAA,GAQjD,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,YAAc,GAAI,IACvB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAiBR,EAAA,UAAA,WAAN,2GAIS,MAAK,YAAY,WAAa,EAE9B,CAAA,EAAM,KAAK,QAFoB,CAAA,EAAA,UAEpC,MAAK,GAAA,aACH,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,YAG/B,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,YAAY,QAAS,KAAM,WAEnD,GAjDmD,IAkDnD,GAAA,SAAA,GAAoC,GAAA,EAAA,GAClC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,eAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,IAeT,IAbM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAMzD,EAAc,KAAK,UAAU,EAAK,OAClC,EACF,GAAA,YAAe,sBAAsB,GACzC,KAAK,YAAY,QAAQ,GAIzB,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAIN,MAAA,CAAA,EAAO,UAEX,GArCoC,IAgDpC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAStC,WACI,EACiB,GAFrB,GAAA,GAGE,EAAA,KAAA,OAAO,KADY,SAAA,iBAAA,EARb,EAAA,SAAuC,KAGvC,EAAA,SAA4B,KAOlC,EAAK,cAAgB,IAGvB,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,8CAC1B,MAAU,GAAiB,eAGvB,EAAA,UAAA,KAAN,8EACE,YAAK,SAAW,KAAK,cAAc,KAAK,UACxC,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,cAAd,SAA4B,qGAO1B,MAAA,CAAA,EAAM,gBAAN,GAAA,OACI,KAAK,UAAY,KACI,CAAA,EAAM,KAAK,cAAc,QAD9C,CAAA,EAAA,UAEF,GADM,EAAiB,EAAA,OACnB,EAAe,KAEjB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,KAAK,SAAW,EAAe,MAC/B,AAAI,KAAK,kBAAoB,MAC3B,MAAK,SAAW,KAAK,SAAS,aAAa,KAAK,oCAGjC,MAAA,CAAA,EAAM,KAAK,SAAS,eACvC,MADM,GAAa,EAAA,OACf,EAAW,KACb,MAAK,SAAW,KAChB,CAAA,EAAO,KAAK,cAAc,KAE5B,CAAA,EAAO,SAEX,GApDwC,IAsD5B,GAAZ,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,YAHU,IAAA,IAAe,KAmC3B,GAAA,IAAA,SAAA,GAAwD,GAAA,EAAA,GAItD,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAgC,GAAgB,MAFvE,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,UAAA,EACA,EAAA,aAAA,EALf,EAAA,MAAQ,EACR,EAAA,eAA6C,OAQrD,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,0CAC1B,MAAO,IAAI,EAAiB,YAGhB,EAAA,UAAA,UAAd,SAAwB,2CAWtB,WAAiB,GACf,GAAI,YAAqB,KACvB,GAAM,GAAS,EAAU,OACzB,MAAO,CACL,MAAO,EAAO,KAAK,SAAA,GACjB,WACA,AAAI,EAAE,MACJ,IAEK,EAAE,QAEX,QAAS,QAGX,OAAO,CAAC,MAAO,KAAM,QAAS,gEArBlC,MAAA,CAAA,EAAM,UAAN,SAAA,OAII,EAAe,EACf,EAAgB,EAoBF,CAAA,EAAM,GAAmB,KAAK,UAAW,WAE3D,GAFM,EAAY,EAAA,OAEd,IAAiB,EAEnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAI,EAAgB,EAClB,OAAQ,KAAK,kBACN,IAAgB,KACnB,KAAM,IAAI,OACN,+CACA,0BAAyB,KAAK,MAAK,UACpC,IAAgB,SACnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,SACxB,IAAgB,SAMzB,YAAK,QACL,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAGzB,EAAA,UAAA,KAAN,8EACE,YAAK,eAAiB,KAAK,UAAU,KAAK,gBAC1C,CAAA,EAAO,KAAK,qBAEhB,GAxEwD,IAoFxD,GAAA,SAAA,GAAyC,GAAA,EAAA,GAGvC,WACc,EAAqC,GADnD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAAqC,EAAA,WAAA,EAEjD,EAAK,OAAS,GAAI,IAAuC,KAG3D,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAOzB,EAAA,UAAA,OAAV,WACE,KAAO,CAAC,KAAK,OAAO,WAClB,GAAM,GAAI,KAAK,SAAS,OACxB,KAAK,OAAO,KAAK,KAIrB,EAAA,UAAA,KAAA,WACE,YAAK,SAIE,KAAK,OAAO,SAEvB,GA/ByC,IAuCzC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAUtC,WACc,EAAqC,EAC/C,GAFJ,GAAA,GAGE,EAAA,KAAA,KAAM,EAAU,IAAW,KAFf,SAAA,SAAA,EAAqC,EAAA,WAAA,EAH3C,EAAA,kBAAoB,GAM1B,EAAK,OAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACpD,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGN,EAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,MAAK,MAAM,KAAK,SAAW,IAG1B,EAAA,UAAA,YAAV,WACE,MAAO,MAAK,UAAU,KAAK,OAAO,WAG9B,EAAA,UAAA,WAAN,6GAEE,AAAK,KAAK,mBACR,KAAK,gCAEC,MAAK,OAAO,UAAS,CAAA,EAAA,GACrB,GAAc,KAAK,cACV,CAAA,EAAM,KAAK,OAAO,cAAc,YAC/C,GADM,EAAS,EAAA,OACX,EAAO,KACT,KAAK,kBAAoB,OAEzB,aAAK,SACL,CAAA,EAAO,sBAGX,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,WAE/B,GApDwC,ICjnCxC,kBA6DA,aAWW,KAAA,KAAe,KA+DxB,SAAA,UAAA,MAAA,SAAM,EAAmB,GAAzB,GAAA,GAAA,KAAyB,AAAA,IAAA,QAAA,GAAA,IACvB,GAAM,GAAO,KACb,GAAA,KAAQ,OACJ,EAAY,EAAG,WAAM,MAAA;QACrB,IACJ,GAAI,GACJ,MAAI,MAAK,OAAS,UAAY,KAAK,MAAQ,KAGzC,EAAO,KAAK,KACP,AAAI,EAGT,EAAO,KAAK,KAAK,KAAK,KAAO,GAI7B,EAAO,KAAK,MAAM,KAAK,KAAO,GAEzB,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OACH,iBAAiB,EAAW,EAAgB,WAChD,IAkBL,EAAA,UAAA,YAAA,SAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,UAAY,EAAQ,OAAS,SAG7C,EAAO,SACF,AAAI,KAAK,MAAQ,MAAQ,EAAQ,MAAQ,KAG9C,EAAO,KAAK,KAAO,EAAQ,KAI3B,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,EAAA,MAAA,IAAA,KAAA,SAAA,0BACK,MAAA,CAAA,EAAM,EAAK,mBAAZ,SAAA,GAAC,EAAA,QAAuB,YAAY,CAAA,EAAM,EAAQ,mBAAlD,MAAA,CAAA,EAAA,EAAA,MAAA,EAAA,CAAoC,EAAA,gBACxC,IAmBN,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,SAEhB,EAAO,SAIP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,OAAO,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAClE,IAmBC,EAAA,UAAA,aAAN,SAAmB,6FACT,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,aAAa,UAkB9C,EAAA,UAAA,IAAA,SAAkC,GAAlC,GAAA,GAAA,KACQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,IAAI,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAC/D,KAAK,OA0BV,EAAA,UAAA,SAAA,SAAuC,GAAvC,GAAA,GAAA,KAEQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,SAAS,UACvC,KAAK,OAYV,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,GAAI,GAAc,KAChB,KAAM,IAAI,YACN,6DAGN,GAAM,GAAO,KACb,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,SAAS,UAAa,KAAK,OAqBrE,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,EAAQ,EAI/B,EAAO,KAAK,KAAO,EACd,AAAI,IAAU,EAEnB,EAAO,EACF,AAAI,KAAK,MAAQ,MAAS,KAAU,QAAa,EAAQ,GAG9D,EAAO,SAGP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mDACrB,SAAmB,GACrB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,MAAA,IAAA,KAAA,SAAA,sCAAqB,CAAA,EAAM,EAAK,mBAApB,MAAA,CAAA,EAAE,GAAA,MAAO,EAAA,OAAuB,EAAA,KAAM,GAAK,WAC3D,CAAA,EAAO,GAAyB,EAAiB,KAAK,UACrD,IAoBL,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,GAAS,GAAK,KAAK,MAAQ,EAIlD,EAAO,KAAK,KAAO,EACd,AACH,KAAK,MAAQ,MACZ,MAAK,KAAO,GAAS,IAAU,QAAa,EAAQ,GAGvD,EAAO,EAGP,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IA4BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,EAAe,GAA3C,GAAA,GAAA,KAEE,GAFyC,AAAA,IAAA,QAAA,GAAA,IAErC,GAAc,MAAQ,EAAa,EACrC,KAAI,MAAK,MAAQ,KACT,GAAI,YACN,4DAEE,GAAI,YACN,+KAGA,oCAAmC,KAAK,KAAI,eAGpD,GAAM,GAAO,KACP,EAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACrD,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACvB,SAAQ,EAAO,QACnB,AAAI,GACF,IAAS,EAAO,SAEV,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,QAAQ,EAAY,EAAM,mBACxD,KAAK,OAoBV,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,KAAK,KAAO,EAGnC,EAAO,EACF,AAAI,KAAK,MAAQ,MAAQ,KAAK,MAAQ,EAG3C,EAAO,KAAK,KAGZ,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IAmBjD,EAAA,UAAA,QAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,iBAc3B,EAAA,UAAA,eAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,wBA1HjB,EAAA,gBAAkB,IA4HpC,iBAeI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,GAAA,UAAA,GAAkB,GAAA,EAAA,GAAd,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KACT,SAAA,KAAO,IAMD,SAAA,UAAA,SAAN,8EACE,MAAA,CAAA,EAAO,UAEX,GAVyB,iBAgCyB,GAApD,GAAA,GAAA,KACE,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,GAAkB,SAAQ,EAAM,oBA4CA,GAAlD,GAAA,GAAA,KAGE,GAAI,CAAC,GAAW,GACd,KAAM,IAAI,OAAM,qDAElB,GAAI,GACJ,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAO,GAAQ,KAAQ,EAAS,GAAkB,KAC5B,KAAK,IAAI,EAAO,EAAS,GAAkB,cAE1D,YAAoB,QAC7B,OAAW,KAAM,GACf,EAAO,GAAQ,KAAQ,EAAS,GAAmB,KAC7B,KAAK,IAAI,EAAO,EAAS,GAAmB,MAGtE,MAAO,IAAyB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACd,MAAA,CAAA,EAAM,GAAmB,EAAU,SAAA,GACjD,GAAI,YAAa,IACf,MAAO,CAAC,MAAO,EAAE,WAAY,QAAS,IACjC,GAAI,GAAW,GACpB,MAAO,CAAC,MAAO,KAAM,QAAS,IAE9B,KAAM,IAAI,OACN,uFAPF,SAAU,EAAA,OAWhB,CAAA,EAAO,GAAsB,EAAS,GAAgB,iBACrD,GAWL,YAAyB,GACvB,GAAI,IAAS,KACX,MAAO,MAIT,GAAM,GAAa,EAAK,GAExB,GAAI,GAAa,IAEf,GAAM,GAAQ,GAAY,GAC1B,MAAO,CAAC,MAAK,EAAE,QAAS,IAI1B,MAAO,CAAC,MAAO,KAAM,QAAS,IAOhC,YAAwD,GAEtD,GAAI,EAAO,SAAW,EAEpB,KAAM,IAAI,OAAM,wCAGlB,MAAI,GAAO,YAAc,IAAA,OAEhB,GAAA,MAAS,GAGT,GAAA,OAAU,GC9rBrB,mBA2BqC,GAAA,EAAA,GAMnC,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,IAIzB,SAAA,UAAA,SAAN,+GACwB,MAAA,CAAA,EAAM,KAAK,MAAM,mBAAjC,SAAgB,EAAA,OAChB,EAAe,EAAc,aAC7B,EAAe,EAAa,MAAM;GAAM,IAAI,SAAA,GAEhD,MAAI,GAAK,SAAS,OAChB,GAAO,EAAK,MAAM,EAAG,KAEhB,IAET,CAAA,EAAO,SAEX,GAtBqC,IC3BrC,AAyBA,GAAM,IAAa,IACb,GAAY,OAAO,OACnB,GAAc,OAAO,SACrB,GAAc,OAAO,SACrB,GAA0B,OAAO,mBACjC,GAA8B,OAAO,+BAeX,GAAA,EAAA,GAgI9B,WAA+B,EAAmB,GAAlD,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,EA9HvB,EAAA,UAAY,GACZ,EAAA,gBAA4B,KAC5B,EAAA,qBAAuB,GACvB,EAAA,cAA+C,KAC/C,EAAA,sBAAwB,GACxB,EAAA,UAAY,IACZ,EAAA,gBAAkB,GA0HxB,EAAK,KAAO,GAAI,IAAgB,GAChC,AAAK,GACH,GAAY,IAEd,EAAK,UAAY,IAAU,YAAc,IACzC,EAAK,gBAAkB,EAAU,YACjC,EAAK,cAAgB,EAAU,cAC/B,EAAK,sBAAwB,EAAU,sBACvC,AAAI,EAAU,gBACZ,IAAA,KAAK,OACD,EAAU,WAAa,KACvB,WACI,MAAA,mEACR,EAAK,gBAAkB,GACvB,EAAK,UAAY,KAEjB,EAAK,UAAY,EAAU,UAAY,EAAU,UAAY,MA9H3D,SAAA,UAAA,YAAN,2GACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEF,MAAA,CAAA,EAAO,KAAK,sBAAwB,OAAO,KAAK,KAAK,eACjB,KAAK,uBAW7B,EAAA,UAAA,eAAd,8HAC8B,MAAA,CAAA,EAAM,KAAK,8BACvC,GADM,EAAsB,EAAA,OACxB,CAAC,KAAK,iBAAmB,CAAC,EAE5B,KAAM,IAAI,OACN,6DA0BN,GAzBO,AAAI,KAAK,iBAAmB,GAEjC,GAAA,KAAK,OACD,EAAoB,SAAW,KAAK,gBAAgB,OACpD,WAAM,MAAA,uCACF,EAAK,gBAAgB,OAAO,WAC5B,kEACW,EAAoB,OAAO,WAAa,OAE7D,AAAK,KAAK,iBACR,MAAK,gBAAkB,GAGnB,EAAkC,KAAK,gBAAgB,OACzD,SAAC,EAAmC,GAClC,SAAS,GAAS,EAAS,GAAQ,GAAM,EAClC,GAET,IACE,EACF,OAAO,KAAK,GAAQ,OAAO,SAAC,GAAS,MAAC,GAAO,GAAQ,IACzD,GAAA,KAAK,OACD,EAAe,SAAW,EAC1B,WAAM,MAAA,iCAAmC,EAAe,aAExD,KAAK,eACP,IAAA,EAAA,EAAkB,EAAA,OAAO,KAAK,KAAK,eAAjB,EAAA,EAAA,OAAA,IAEhB,GAFS,EAAG,EAAA,GACN,EAAQ,KAAK,gBAAgB,QAAQ,GACvC,IAAU,GACZ,KAAM,IAAI,OACN,YAAc,EACd,uEACY,KAAK,gBAAgB,WAAa,MAIxD,YAAK,qBAAuB,aAGhB,EAAA,UAAA,oBAAd,uHACM,MAAK,UACM,CAAA,EAAM,KAAK,KAAK,YAD3B,CAAA,EAAA,UACI,SAAO,EAAA,OACQ,CAAA,EAAM,EAAK,eAChC,GADM,EAAe,EAAA,OACjB,EAAa,KACf,KAAM,IAAI,OAAM,sCAEZ,SAAoB,EAAa,MACjC,EAAU,KAAK,SAAS,EAAW,IACzC,CAAA,EAAO,UAEP,MAAA,CAAA,EAAO,YA0DL,EAAA,UAAA,SAAN,wHACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEU,MAAA,CAAA,EAAM,KAAK,KAAK,mBAAxB,SAAQ,EAAA,OACZ,AAAI,KAAK,WAGP,GAAQ,EAAM,KAAK,IAErB,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,GAAK,gBAAgB,YAG7C,EAAA,UAAA,gBAAA,SAAgB,GAKd,OAJM,GAAS,KAAK,SAAS,GACvB,EAA6C,GAC7C,EAA2C,GAExC,EAAI,EAAG,EAAI,KAAK,gBAAgB,OAAQ,KAC/C,GAAM,GAAM,KAAK,gBAAgB,GAC3B,EAAS,KAAK,cAAgB,KAAK,cAAc,GAAO,KAC9D,GAAI,KAAK,uBAAyB,CAAC,EAEjC,SAEA,GAAM,GAAQ,EAAO,GACjB,EAAc,KAClB,GAAI,IAAU,GAGZ,GAAI,GAAU,EAAO,UAAY,OAC/B,EAAc,EAAO,aAChB,GAAI,GAAW,GAAO,UAAY,EAAO,SAC9C,KAAM,IAAI,OACN,mBAAmB,EAAG,2BAA2B,GAErD,EAAc,YAIhB,GAAM,GAAa,OAAO,GAC1B,GAAI,MAAM,GAGR,AAAI,GAAU,EAAO,QAAU,OAC7B,EAAc,KAAK,WAAW,GAG9B,EAAc,UAEP,CAAC,GAAU,CAAC,EAAO,MAG5B,EAAc,MAId,QAAQ,EAAO,WACR,UACH,EAAc,EACd,UACG,QACH,EAAc,KAAK,MAAM,GACzB,UACG,OACH,EAAc,KAAK,WAAW,GAC9B,cAEA,EAAc,GAKtB,AAAC,GAAU,EAAO,QAAW,EAAO,GAAO,EACd,EAAS,GAAO,EAKjD,MAAI,QAAO,KAAK,GAAQ,SAAW,EAC1B,EAGA,CAAC,GAAI,EAAU,GAAI,IAItB,EAAA,UAAA,WAAR,SAAmB,GACjB,MAAI,KAAU,KAAO,EAAM,gBAAkB,OACpC,EAEA,GAKH,EAAA,UAAA,SAAR,SAAiB,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IAM7B,OALM,GAAmB,GACrB,EAAa,EACX,EAAa,EAAK,OACpB,EAAe,GAEV,EAAI,EAAG,EAAI,EAAY,IAC9B,OAAQ,OAED,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAa,EAAI,EACjB,EAAe,GACf,UAEG,MAAK,UAIR,GAHA,EAAa,EAAI,EAGb,KAAK,YAAc,KAAO,KAAK,gBACjC,MAEF,EAAO,KAAK,IACZ,EAAe,GACf,cAGA,EAAe,GACf,EAAa,EACb,MAEJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,IACvC,EAAe,GACf,EAAa,EAAI,EACjB,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,EAAI,IAC3C,EAAe,GACf,EAAa,EAAI,EACjB,UAEG,IACH,EAAe,GACf,cAGA,EAAe,GACf,MAEJ,UACG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,MAWN,GANA,AAAI,IAAiB,GACnB,EAAO,KAAK,EAAK,UAAU,EAAY,EAAa,IAEpD,EAAO,KAAK,EAAK,UAAU,IAGzB,GAAwB,EAAO,SAAW,KAAK,gBAAgB,OACjE,KAAM,IAAI,OAAM,wCACZ,KAAK,gBAAgB,OAAM,+BAA+B,GAEhE,MAAO,IAEX,GA1VgC,IC7ChC,AA4BA,GAAA,IAAA,SAAA,GAAwC,GAAA,EAAA,GAgBtC,WAAuC,GAAvC,GAAA,GACE,EAAA,KAAA,OAAO,KAD8B,EAAA,iBAAA,EAf/B,EAAA,SAAW,GAiBjB,EAAK,QAAU,EAAiB,SAAW,KAC3C,GAAM,GAAc,KAAK,KAAK,EAAK,SACnC,GAAI,EAAK,QAAU,GAAK,EAAc,GAAK,EAAc,IACrD,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,oDACA,gCAA+B,EAAK,UAc1C,GAXA,EAAK,UAAY,EAAiB,yBAA2B,GAC7D,EAAK,aAAe,EAAiB,aACrC,EAAK,qBACD,EAAiB,sBAAwB,EAAK,QAClD,EAAK,sBAAwB,EAAiB,sBAC9C,EAAK,sBAAwB,EAAiB,uBAAyB,EAEvE,EAAK,mBACD,IAAiB,qBAAuB,IAC5C,EAAK,gBACD,EAAiB,kBAAoB,GACrC,CAAC,EAAK,oBAAsB,CAAC,EAAK,gBACpC,KAAM,IAAI,OACN,iHAKR,SAAA,UAAA,QAAA,WACE,MAAO,cAII,EAAA,OAAb,SAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,6FAClB,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGA,SAAqB,GAAI,GAAmB,GAGlD,CAAA,EAAM,EAAmB,gBAAzB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,8IAEI,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,KAAK,uBAAyB,KAAO,GACA,KAAK,sBACjD,MAAO,aAHT,SAAK,OAAS,EAAA,qCAMR,GAAI,OACN,iDAAiD,EAAE,gBAGzD,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,2CAQlB,GALM,EAED,OAAe,cAAiB,OAAe,mBACpD,KAAK,aAAe,GAAI,GAEpB,CAAC,KAAK,aAGR,KAAK,aAAe,KAAK,aAAa,mBAC7B,KAAK,aAAa,aAAe,KAAK,aAC/C,KAAM,IAAI,OACN,8BACA,cAAa,KAAK,aAAY,MAC9B,YAAW,KAAK,aAAa,aAG7B,SAAe,KAAK,aAAa,wBAAwB,KAAK,QACpE,KAAK,SAAW,KAAK,aAAa,iBAClC,KAAK,SAAS,QAAU,KAAK,QAAU,EACvC,KAAK,SAAS,sBAAwB,KAAK,sBAC3C,EAAa,QAAQ,KAAK,UAC1B,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,CAAA,SAGI,EAAA,UAAA,KAAN,mHACE,MAAI,MAAK,SACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAMN,CAAA,EAAM,KAAK,uBAA5B,SAAiB,EAAA,OACvB,AAAI,KAAK,oBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAoB,KAAK,4BACrB,EAAU,CAAC,KAAK,UAAW,KAAK,qBAAsB,KAE5D,AAAI,KAAK,iBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAiB,KAAK,4BAClB,EAAU,CAAC,KAAK,UAAY,KAAK,QAAS,KAGhD,CAAA,EAAO,CACL,MAAO,CAAC,YAAe,EAAmB,SAAY,GACtD,KAAM,WAMJ,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAIf,EAAA,UAAA,aAAd,+FAEQ,SAAgC,GAChC,EAAgC,GAClC,EAAgB,EACpB,CAAA,EAAO,GAAI,SAAQ,SAAA,GACjB,GAAM,GAAa,YAAY,WAC7B,AAAI,EAAK,oBACP,GAAK,SAAS,sBAAsB,EAAK,UAEzC,AAAI,EAAK,SAAS,KAAO,WACvB,EAAQ,CAAC,cAAa,EAAE,cAAa,IAEvC,EAAc,KAAK,EAAK,SAAS,MAAM,EAAG,EAAK,wBAEjD,AAAI,EAAK,iBACP,GAAK,SAAS,uBAAuB,EAAK,UAC1C,EAAc,KAAK,EAAK,SAAS,UAInC,AAAI,EAAE,IAAkB,EAAK,WAC3B,eAAc,GACd,EAAQ,CAAC,cAAa,EAAE,cAAa,MAEtC,EAAK,QAAU,EAAK,aAAe,aAK1C,EAAA,UAAA,KAAA,WACE,AAAK,KAAK,UACR,MAAK,SAAW,GAChB,KAAK,SAAS,aACd,KAAK,aAAa,QAClB,AAAI,KAAK,QAAU,MAAQ,KAAK,OAAO,YAAY,OAAS,GAC1D,KAAK,OAAO,YAAY,GAAG,SAMjC,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAIlB,EAAA,UAAA,cAAA,WACE,MAAO,MAAK,cAGN,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAY,EAAM,GAAG,OACrB,EAAW,GAAI,cAAa,EAAM,OAAS,GACjD,SAAM,QAAQ,SAAC,EAAM,GAAM,MAAA,GAAS,IAAI,EAAM,EAAI,KAC3C,GAGD,EAAA,UAAA,4BAAR,SAAoC,EAAwB,GAE1D,GAAM,GAAO,GAAI,cAAa,GAAA,KAAK,cAAc,IAEjD,SAAK,IAAI,EAAU,EAAK,OAAS,EAAS,QACnC,GAAA,OAAO,EAAM,IAExB,GA7MwC,IC5BxC,AA0BA,GAAA,IAAA,SAAA,GAAoC,GAAA,EAAA,GAQlC,WACuB,EACA,GAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KACP,GAHqB,EAAA,mBAAA,EACA,EAAA,aAAA,EATf,EAAA,SAAW,GAEX,EAAA,OAAS,GASX,EAAK,eAKP,GAJA,EAAK,OAAS,GACd,EAAK,SACD,CAAC,EAAK,aAAa,aAAc,EAAK,aAAa,aACvD,EAAK,WAAa,GAAA,SAAS,CAAC,GAAI,SAC5B,EAAK,aAAa,YAEpB,GAAM,GACF,EAAK,aAAa,YAAc,EAAM,EAAK,mBAAmB,MAC5D,EAAsB,EAAK,aAAa,aAAe,EACzD,EAAK,mBAAmB,OACtB,EAAkB,GAAI,GAAsB,EAC5C,EAAmB,GAAI,GAAuB,EAC9C,EAAe,EAAiB,EAChC,EAAgB,EAAsB,EAC5C,EAAK,QAAU,GAAA,SACX,CAAC,EAAiB,EAAgB,EAAe,GACjD,CAAC,EAAG,QAER,GAAK,QAAU,GAAA,SAAS,CAAC,EAAG,EAAG,EAAG,GAAI,CAAC,EAAG,aAKhD,SAAA,UAAA,QAAA,WACE,MAAO,UAII,EAAA,OAAb,SACI,EAAuC,GAAA,MAAA,KAAA,QAAA,GAAA,6FACzC,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGN,GAAI,CAAC,GAIH,GADA,EAAqB,SAAS,cAAc,SACxC,CAAC,EAAa,aAAe,CAAC,EAAa,aAC7C,KAAM,IAAI,OACN,0GAGN,EAAmB,MAAQ,EAAa,YACxC,EAAmB,OAAS,EAAa,aAErC,SAAiB,GAAI,GAAe,EAAoB,GAG9D,CAAA,EAAM,EAAe,gBAArB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,oHACE,AAAI,KAAK,aAAa,YACpB,GAAA,KAAK,OACA,KAAK,aAAa,aAAe,QAC7B,KAAK,aAAa,aAAe,cACtC,WACI,MAAA,+BAA+B,EAAK,aAAa,WAAU,2FAKnE,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,CACL,SAAU,KAAK,aAAa,SAC5B,WAAY,KAAK,aAAa,WAC1B,KAAK,aAAa,WAClB,OACJ,MAAO,KAAK,mBAAmB,MAC/B,OAAQ,KAAK,mBAAmB,kBAPpC,SAAK,OAAS,EAAA,qCAYd,EAAE,QAAU,iDAAiD,EAAE,QACzD,SAGR,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,uCAIlB,IACE,KAAK,mBAAmB,UAAY,KAAK,aAClC,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,OAAO,IAAI,gBAAgB,KAAK,QAGhE,YAAK,mBAAmB,OAExB,KAAK,SAAW,GAEhB,CAAA,EAAO,GAAI,SAAc,SAAA,GAEvB,EAAK,mBAAmB,iBAAmB,WACzC,cAKA,EAAA,UAAA,KAAN,oFACE,GAAI,KAAK,SACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAI7B,IACE,EAAM,GAAA,QAAQ,WAAW,KAAK,0BACvB,GACP,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAU,IAEjE,GAAI,KAAK,OACP,IACE,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,mBAAmB,GAAM,KAAM,WAC5C,GACP,KAAM,IAAI,OAAM,oCAAoC,EAAE,iBAEtD,EAAI,cAGN,OAAA,CAAA,EAAO,CAAC,MAAO,EAAK,KAAM,oBAItB,EAAA,UAAA,aAAR,WAIE,MAAI,QAAK,aAAa,aAAe,KAAK,aAAa,cAClD,MAAK,mBAAmB,QAAU,KAAK,aAAa,aACpD,KAAK,mBAAmB,SAAW,KAAK,aAAa,gBAO5D,EAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,IAAA,KAAK,WACV,GAAM,GAA0B,EAAI,UAAU,WAAW,GACrD,EACJ,EAAe,GAAA,MAAM,cACjB,EAAe,EAAK,QAAS,EAAK,WAAY,EAAK,SACnD,YAEJ,GAAM,GAAQ,EAAa,MAC3B,MAAO,GAAa,QAAQ,EAAM,MAAM,OAMtC,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAI7B,EAAA,UAAA,KAAA,WACE,GAAM,GAAS,KAAK,OAAO,YAE3B,EAAO,QAAQ,SAAA,GAAS,MAAA,GAAM,SAE9B,IACE,KAAK,mBAAmB,UAAY,WAC7B,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,KAEhC,KAAK,SAAW,IAIlB,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAEpB,GApMoC,IC1BpC,AA2BA,GAAA,IAAA,WAAA,cAUA,MAAA,MCrCA,AAoBA,GAAA,IAAA,SAAA,GAA6C,GAAA,EAAA,GAA7C,4DAmBE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAI,IAAc,KAAM,IAEnC,GAtB6C,IAmC7C,GAAA,SAAA,GAA4B,GAAA,EAAA,GAG1B,WAAsB,EAAgC,GAAtD,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAkB,EAAU,KAG9C,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf4B,IAiB5B,GAAA,SAAA,GAAgC,GAAA,EAAA,GAI9B,WACc,EAA0C,GADxD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAA0C,EAAA,UAAA,EAHxD,EAAA,UAAY,KAOZ,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAAc,KAAK,UAAS,MAGzD,EAAA,UAAA,KAAN,mHACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eACxC,GADM,EAAc,EAAA,OAChB,EAAY,KACd,MAAI,MAAK,YAAc,GACrB,CAAA,EAAO,IAKT,MAAK,YAAY,KAAK,KAAK,WAC3B,KAAK,UAAY,GACjB,CAAA,EAAO,KAQT,IANM,EAAQ,EAAY,MAAM,MAAM,KAAK,WAK3C,EAAM,GAAK,KAAK,UAAY,EAAM,GAClC,EAAA,EAAmB,EAAA,EAAM,MAAM,EAAG,IAAf,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,KAAK,YAAY,KAAK,GAExB,YAAK,UAAY,EAAM,EAAM,OAAS,GAEtC,CAAA,EAAO,UAEX,GAvCgC,ICxEhC,AAsBA,GAAA,IAAA,SAAA,GAAgD,GAAA,EAAA,GAAhD,4DAUE,SAAA,UAAA,WAAA,WACE,MAAO,IAAI,IAAa,OAE5B,GAbgD,IA0BhD,GAAA,SAAA,GAA2B,GAAA,EAAA,GAGzB,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAiB,KAGnC,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf2B,IAuC3B,GAAA,SAAA,GAA+B,GAAA,EAAA,GAM7B,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAF6B,EAAA,SAAA,EAEzB,GAAA,MAAM,IAAI,cACZ,EAAK,QAAU,GAAI,aAAY,cAGxB,GAAA,GAAA,0BAAA,cACP,EAAK,QAAU,GAAI,GAAc,iBAGrC,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,+GACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eAExC,MAFM,GAAc,EAAA,OAEhB,EAAY,KACd,CAAA,EAAO,IAEP,GAAQ,EAAY,MAItB,AAAI,GAAA,MAAM,IAAI,cACZ,EAAO,KAAK,QAAQ,OAAO,EAAO,CAAC,OAAQ,KAE3C,EAAO,KAAK,QAAQ,MAAM,OAAO,KAAK,EAAM,SAE9C,KAAK,YAAY,KAAK,GACtB,CAAA,EAAO,WAEX,GAtC+B,ICvF/B,AAqCA,GAAA,IAAA,SAAA,GAAuC,GAAA,EAAA,GAIrC,WACc,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,KAAA,EACA,EAAA,QAAA,EAEZ,GAAA,KAAK,OACA,YAAgB,aACZ,IAAA,MAAM,IAAI,cACL,YAAgB,OAAQ,YAAgB,MACzC,IACT,WAAM,MAAA,yEAEV,EAAK,OAAS,EAAQ,QAAU,EAEhC,EAAK,UAAY,EAAQ,WAAa,KAAO,OAG/C,SAAA,UAAA,QAAA,WACE,MAAO,cAAc,KAAK,MAGtB,EAAA,UAAA,KAAN,oHACE,MAAI,MAAK,QAAY,MAAK,eAAgB,YAClB,KAAK,KAAK,WACV,KAAK,KAAK,MAChC,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAQ,GAAI,SAAoB,SAAC,EAAS,GAC9C,GAAM,GAAM,EAAK,OAAS,EAAK,UAC/B,GAAI,EAAK,eAAgB,YAGvB,EAAQ,GAAI,YAAW,EAAK,KAAK,MAAM,EAAK,OAAQ,UAMpD,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GACnB,GAAI,GAAsC,EAAW,OAOrD,GAHA,AAAI,YAAgB,cAClB,GAAO,GAAI,YAAW,IAEpB,CAAE,aAAgB,aACpB,MAAO,GAAO,GAAI,WAAU,sCAE9B,EAAQ,IAEV,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,aAE1B,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,EAAM,QAIhC,GAAM,GAAQ,EAAK,KAAK,MAAM,EAAK,OAAQ,GAG3C,EAAW,kBAAkB,GAE/B,EAAK,OAAS,SAEA,CAAA,EAAM,WAAtB,MAAA,CAAA,EAAQ,GAAA,MAAQ,EAAA,OAAc,EAAA,KAAM,GAAK,UAE7C,GAxEuC,ICrCvC,YA6BI,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,qGAGpB,MAAK,OAAO,IAAS,SACnB,EAAY,EAEZ,GAAa,EAAgB,IAC7B,EAAc,GAA0B,IAEzB,CAAA,EAAM,GAAA,KAAK,MAAM,EAAW,iBAAvC,GAAW,EAAA,OACb,EAAS,MACY,WAAU,KAAC,CAAA,EAAM,EAAS,gBAD/C,CAAA,EAAA,UACI,SAAa,GAAA,GAAA,MAAI,WAAU,CAAA,OAAC,EAAA,UAClC,CAAA,EAAO,GAAI,IAAkB,EAAY,WAEzC,KAAM,IAAI,OAAM,EAAS,iBAK7B,GAAM,IAA4B,SAAC,GACjC,GAAM,GAAO,CACX,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,KACd,KAAM,EAAQ,KACd,YAAa,EAAQ,YACrB,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,SAAU,EAAQ,SAClB,UAAW,EAAQ,WAErB,MAAO,IC5DT,YAqB4B,GAC1B,MAAQ,OAAO,IAAW,UAAa,EAAO,OAAO,EAAG,KAAO,UCtBjE,mBA6BoC,GAAA,EAAA,GASlC,WACc,EACS,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,MAAA,EACS,EAAA,QAAA,IAIjB,SAAA,UAAA,SAAN,oFACE,MAAI,IAAY,KAAK,QAAU,GAAA,MAAM,IAAI,YAEjC,GAAK,cACX,KAAK,MAAQ,EAAG,aAAc,KAAK,MAAiB,OAAO,KAI7D,CAAA,EAAO,GAAI,IAAkB,KAAK,MAAsB,KAAK,eAEjE,GAzBoC,IC7BpC,mBA4BmC,GAAA,EAAA,GAQjC,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,IAAA,EACA,EAAA,YAAA,IAQjB,SAAA,UAAA,SAAN,8EACE,MAAI,IAAY,KAAK,KACnB,CAAA,EAAQ,GAAI,IAAe,KAAK,IAAe,KAAK,aAC/C,YAEL,CAAA,EAAO,GAAiB,KAAK,IAAK,KAAK,mBAG7C,GA1BmC,IC5BnC,YA0GI,EAAqB,GAAA,MAAA,KAAA,QAAA,GAAA,IAChB,GAAI,IAAW,GAAI,IAAc,GAAS,eA4B/C,GADJ,GAAA,GAAA,KAEQ,EAAO,GAAqB,GAClC,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,qBAgEvC,GADJ,GAAA,GAAA,KAEE,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACf,MAAA,CAAA,EAAM,YAAZ,SAAM,EAAA,OACZ,CAAA,EAAO,GAAqB,WAAM,MAAA,GAAI,6BAmCtC,EACA,sEACF,MAAA,CAAA,EAAO,GAAe,OAAO,EAAoB,oBAqClB,sEAE/B,MAAA,CAAA,EAAO,GAAmB,OAAO,QCxRnC,GAGM,IAAU,oNCHhB,oBA2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9GnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC7EnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UClFnC,oBAKA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAER,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAG,EAC9B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9FnC,oBAyBA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UChJnC,oBAIA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UCnGnC,oBAwBA,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMjB,MAAO,KAAW,UAAY,GAAO,SACxC,GAAO,QAAU,EAEjB,IACE,EAAa,wBACN,SACJ,AAAK,OAAO,SAAW,YAAc,OAAO,KACjD,OAAO,WAAa,MAAO,OAK3B,GACA,QCrPF,oBAYA,GAAI,IAAO,KAKP,GAAS,KAKT,GAAS,KAQT,GAAY,KASZ,GAAU,KAOV,GAAS,KAIT,GAAK,KAET,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAO,QAAU,6vDC3DjB,YAoBI,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MACF,GAAM,8DC7BpB,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBACvC,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAM/B,YACI,EAAyB,EAAW,EACpC,GACF,GAAI,IAAe,SACjB,MAAO,GAAQ,OAAO,GACjB,GAAI,IAAe,OACxB,MAAO,GAAQ,KAAK,GACf,GAAI,IAAe,MACxB,MAAO,GAAA,IAAO,GACT,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,GAChB,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,EAAG,GAE1B,KAAM,IAAI,OACN,cAAc,EAAU,qEAeM,GAAA,EAAA,GAMlC,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KANF,SAAA,UAAY,GAGX,EAAA,SAAW,GAIjB,EAAK,KAAO,GAAI,GAAA,YAAY,EAAM,EAAA,YAGpC,SAAA,UAAA,MAAA,SAAM,EAAoC,EAAiB,GAEzD,AAAI,KAAK,UACP,MAAK,SAAW,GAChB,AAAI,EAAA,MAAM,IAAI,YACZ,EAAA,aAAa,KACT;;;gCAYR,GAAM,GAAS,GAEf,YAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,IAEzC,GAST,EAAA,UAAA,eAAA,SACI,EAAiB,EACjB,GACF,GAAM,GAAQ,KAAK,MAAM,EAAQ,EAAO,GAExC,MAAO,CAAC,OAAQ,EAAO,MAAK,EAAE,MAAK,IAIrC,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,YAIb,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,aAIf,EAAA,UAAA,KAAA,SACI,EAAgB,EAAoC,EACpD,GACF,KAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,KAGlD,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,KAAK,cAGb,EAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,SAAS,SAEvB,EAAA,UAAA,SAAA,SAAS,GACD,GAAA,GAAA,KAAA,KAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,mBAEd,GAAI,IAAU,aACZ,GAAM,GACF,KAAK,SAAS,EAAmB,KAAK,QACpC,EACF,KAAK,SAAS,EAAmB,KAAK,QAC1C,MAAO,GAAA,aAAa,uBAAuB,EAAY,GAGzD,MAAO,MAAK,KAAK,IAAI,GAAQ,QAGvB,EAAA,UAAA,WAAR,SAAmC,GACjC,GAAM,GAAO,KAAK,SAAS,EAAE,QACzB,EAAc,EAClB,GAAI,EAAE,QAAU,SACd,IAEE,EAAe,EAAsB,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,WAChE,GACA,KAAM,IAAI,OAAM,oDAGpB,MAAO,GAAA,OAAU,EAAE,MAAO,EAAE,MAAO,IAGrC,EAAA,UAAA,WAAA,SACI,EAAoC,EAAiB,GACvD,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAG7D,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,KAAK,IAAI,IACT,GAAA,GAAA,KAAA,KAAA,IAAA,GAAA,mBAEP,AAAI,GAAsB,MACxB,MAAK,YAAY,EAAmB,KAAK,QACzC,KAAK,YAAY,EAAmB,KAAK,SAG3C,KAAK,KAAK,OAAO,KAIrB,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GAEjC,EAAW,WAEX,AAAI,EAAW,SAAW,GACxB,KAAK,YAAY,KAKjB,EAAA,UAAA,KAAN,SAAW,8EACH,SAAQ,EAAA,KAAK,MACnB,IACM,EAAW,EAAA,KAAK,MAAQ,EAC9B,CAAA,EAAO,CAAC,SAAQ,SAGlB,EAAA,UAAA,OAAA,WACE,MAAO,CAEL,WAAY,GACZ,QACI,CAAC,wHAKT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,GAAiB,EAAG,gBAEpB,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAU,GAAI,GAKvB,OAFM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GACpB,EAAI,EAAG,EAAI,EAAO,KAAM,KAI/B,OAHM,GAAM,EAAO,WAAW,GAExB,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,GAAK,EAAQ,GAAK,EAAM,GAE1C,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAM,OAAM,IAGrC,MAAO,GAAO,YAGhB,EAAA,UAAA,KAAA,SAAK,GAIH,OAHM,GAAQ,KAAK,SAAS,EAAE,QACxB,EAAS,EAAA,OAAU,CAAC,EAAE,KAAM,EAAE,MAAO,EAAE,OACvC,EAAO,EAAO,OACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAK,EAAI,EAAE,KAAO,GAAK,EAAM,GAE/B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,EAAA,MAAS,EAAG,EAAO,GAAM,QAAQ,GAE5C,MAAO,IAGT,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAiB,EAAG,WAKpB,OAHM,GAAS,EAAA,OAAU,EAAE,MAAO,EAAE,OAC9B,EAAO,KAAK,WAAW,cAEpB,GACP,GAAM,GAAS,EAAO,WAAW,GAC3B,EAAQ,EAAO,QACrB,EAAK,QAAQ,SAAA,GAAM,MAAA,GAAM,GAAM,EAAE,MAAM,GAAM,EAAI,EAAM,KACvD,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAK,OAAM,KAJ3B,EAAI,EAAG,EAAI,EAAO,KAAM,MAAxB,GAOT,MAAO,GAAO,YAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,UAAiB,EAAG,OAGb,EAAA,IAAO,EAAA,OAAU,IAAK,IAG/B,EAAA,UAAA,KAAA,SAAuB,GAAvB,GAAA,GAAA,KACE,GAAiB,EAAS,QAK1B,OAHM,GAAO,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAK,SAAS,EAAE,UACxC,EAAS,EAAA,OAAU,EAAQ,GAAG,MAAO,EAAQ,GAAG,OAChD,EAAa,EAAO,OACjB,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAElC,OADM,GAAW,EAAK,GACb,EAAI,EAAG,EAAI,EAAW,OAAQ,IACrC,EAAW,IAAM,EAAS,GAG9B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAGhD,EAAI,EAAA,IAAO,EAAQ,EAAS,QAAQ,IACpC,EAAI,EAAA,IAAO,GACX,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAO,EAAG,IAGnB,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,IAAI,EAAQ,MAIjE,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAiB,CAAC,EAAG,GAAI,UAqBzB,OAnBM,GAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAU,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5C,EAAW,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7C,EAAW,EAAE,MAAM,GAEnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,8DAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAAY,EAAA,EAAA,GAGrB,EAAA,8DAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAIzB,EAAO,EAAU,EACjB,EAAS,EAAA,OAAU,CAAC,EAAU,EAAS,GAAW,EAAE,OACpD,EAAU,EAAO,OACjB,EAAY,KAAK,UAEd,EAAI,EAAG,EAAI,EAAU,IAC5B,OAAS,GAAK,EAAG,EAAK,EAAS,GAAM,EACnC,OAAS,GAAK,EAAG,EAAK,EAAU,GAAM,EACpC,OAAS,GAAK,EAAG,EAAK,EAAW,GAAM,EAMrC,OAJM,GAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAE/B,EAAI,EAAI,EAAI,EAAQ,IAC3B,OAAS,GAAI,EAAI,EAAI,EAAQ,KAG3B,OAFI,GAAM,EAED,EAAI,EAAI,EAAI,EAAQ,IAC3B,GAAO,EAAQ,EAAI,EAAS,EAAI,EAAa,EAAI,GAC7C,EAAQ,EAAI,EAAa,EAAI,EAAa,EAAI,GAEpD,EAAQ,EAAI,EAAQ,GAAI,EAAW,KAAO,EAOtD,MAAO,GAAO,YAGhB,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE/C,EAAS,KAAK,YAAY,EAAG,EAAG,EAAY,GAChD,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAIvC,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAiB,CAAC,EAAG,GAAI,YAEzB,GAAM,GAAK,SAAC,EAAW,GAAc,MAAA,MAAK,MAAM,EAAI,IAC9C,EAAc,QACpB,MAAO,MAAK,oBAAoB,EAAG,EAAG,EAAa,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MASvD,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EACD,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAO,EAAM,EAAS,GAExB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,GAAiB,EAAG,OAUpB,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAO,EACF,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAQ,EAAM,EAAS,GAEzB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,EAAG,sBAOpB,OALM,GAAM,GAIN,EAAW,EAAE,KAAO,EAAW,KAC5B,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,EAAW,WAAW,EAAI,GAGzC,OAAS,GAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAY,EAAA,OAAU,EAAG,SACzB,EAAO,EAAA,MAAS,EAAW,GAAY,OAAO,WAC9C,EAAM,EAAK,IAAI,GAAG,IAAI,GAC5B,EAAI,KAAK,GAGX,MAAO,GAAA,MAAS,IAGlB,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAIlD,GAFA,GAAiB,EAAG,UAEhB,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,oDAAoD,GAAE,KAAO,GAAC,IAC9D,iBAAgB,IAWtB,OATM,GAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAE,MAAO,GAC3B,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAW,EAAE,MAAM,EAAE,KAAO,GAC5B,EAAgB,EAClB,SAAC,EAAW,GAAc,MAAA,GAAI,EAAW,EAAI,GAC7C,SAAC,EAAW,GAAc,MAAA,GAAI,GACzB,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACrC,OAAS,GAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAM,EAAc,EAAG,GAC7B,GAAI,IAAM,EACR,EAAK,GAAO,EAAY,EAAI,EAAM,QAElC,GAAM,GAAU,EAAc,EAAG,EAAI,GACrC,EAAK,GAAO,EAAY,EAAM,GAAW,EAAK,GACtB,EAAM,GAAO,EAAK,IAIhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,UAAiB,CAAC,EAAG,GAAI,YAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,UAAiB,CAAC,EAAG,GAAI,QAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,UAAiB,CAAC,EAAG,GAAI,gBAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,UAAiB,CAAC,EAAG,GAAI,cAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAiB,CAAC,EAAW,EAAG,GAAI,UAYpC,OAVM,GAAS,KAAK,SAAS,EAAU,QACjC,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,MAAS,EAAE,MAAO,EAAA,WAAW,EAAE,MAAO,EAAE,QACjD,EAAY,KAAK,SAAS,EAAO,QACnC,EAAQ,EACN,EAAS,EAAU,OAAS,GAAK,EAAU,KAAO,GAAK,EAAE,OAAS,EACpE,EACA,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAE5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,OAAS,GAAI,EAAG,EAAI,EAAQ,IAC1B,AAAI,EAAO,KAAO,EAChB,EAAU,KAAW,EAAQ,GAE7B,EAAU,KAAW,EAAQ,GAKnC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GACJ,GAAiB,CAAC,GAAY,SAE9B,GAAM,GAAW,KAAK,SAAS,EAAU,QACzC,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAiB,EAAG,QAEpB,GAAM,GAAQ,KAAK,SAAS,EAAE,QAC9B,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAM,EAAO,EACnB,MAAK,GAAO,GAAK,EAAO,GAAO,GAAQ,GAAK,GAAQ,EAC3C,EAEC,GAAM,GAAQ,KAK5B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAM,GAAO,EAEf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAS,EAAM,GACV,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAS,GAAU,EAErB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,UAAiB,CAAC,EAAG,GAAI,qBAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAO,EAAO,EACpB,MAAO,GAAO,KAIlB,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,IAGT,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,EAAG,EAAO,IAElC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,KAAK,IAAI,EAAG,EAAO,IAAK,GAEhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MACR,SAAC,EAAQ,GAAW,MAAA,GAAS,EAAI,EAAS,EAAS,KAGhE,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAiB,CAAC,EAAI,GAAI,UAK1B,OAHM,GAAe,GAAI,cAAa,EAAE,MAClC,EAAS,KAAK,SAAS,EAAE,QACzB,EAAW,KAAK,SAAS,EAAG,QACzB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,GACjB,AAAI,GAAK,EACP,EAAa,GAAK,EAAS,GAE3B,EAAa,GAAK,EAAS,GAAM,GAAI,GAGzC,MAAO,MAAK,WAAW,EAAc,EAAE,MAAO,YAGhD,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,MAAM,EAAQ,MAInE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,OAAO,EAAO,EAAQ,GAExC,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAiB,CAAC,EAAG,GAAS,UAyB9B,OAvBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAiB,EAAS,aAAe,eAEzC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OAEnC,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAChD,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAEhD,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EACf,EAAW,EAAI,EACZ,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EACxB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,IAAK,EAAW,EAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAC7B,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,GAAK,GAC1B,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,GAAK,IAClB,GAAO,EAAM,GAAW,IAE9B,IAAY,EAAS,cAOjC,MAAO,GAAE,YAGX,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAiBpC,OAfM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAI,EAAA,OAAmB,EAAS,SAAU,EAAE,OAE5C,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAKF,OAHM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAElC,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAS,YACpC,EAAW,EAAK,EAAS,YAAc,EACpC,GAAK,EAAG,GAAK,EAAa,MACjC,GAAM,IAAK,EAAW,GAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,GAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAAS,WACtC,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,IACrB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,KAAO,GAAO,EAAM,GAAW,IAElD,IAAY,EAAS,eASrC,MAAO,GAAE,YAGX,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,kBAkC/B,OAhCM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAiB,IAAe,eAChC,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GACjD,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GAE9C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,IAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAOjC,OANM,IAAK,GAAK,EAAc,GACxB,GACF,EAAe,EAAI,EAAa,GAAK,EAAa,GAChD,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,EAAiB,IAC7C,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,GAAM,IAAW,EAAe,EAAI,EAAa,GAC7C,EAAa,GAAK,EAAiB,EACvC,EAAS,IAAY,GAK7B,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GA6BF,OA5BM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAE1B,EAAA,EAAA,UACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,GAAA,EAAA,YAEI,GAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,GAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,GAAU,EAAc,EAAI,EAAS,QAAQ,KAE1C,GAAI,EAAG,GAAI,EAAW,EAAE,GAC/B,OAAS,IAAK,EAAG,GAAK,EAAY,EAAE,GAElC,OAAS,IAAK,EAAG,GAAK,EAAS,EAAE,GAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAGzC,GAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,KACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,IAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAc,GAErB,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAQjC,OAPM,IAAK,GAAK,GAAc,GACxB,GACF,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GACxC,GAAY,EAAS,GAAc,EAAI,IACzC,EAAS,GAAe,EAAI,IAC5B,EAAS,GAAc,EAAI,IAAM,EAAQ,GAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAK3B,EAAS,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GAAK,IACpD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAiB,CAAC,EAAG,GAAK,mBAa1B,OAXM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eACzC,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,EAC3C,OAAS,GAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAG5C,OADI,GAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,AAAI,EACF,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAEnD,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAK3D,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAKpC,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAqBzC,OAnBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAC9C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAEhB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAEvB,EAAK,EAAG,EAAK,EAAa,EAAE,EAMnC,OALM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAW,GAAM,IAChD,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAW,GAAM,GACtD,EAAW,EAAK,EAEb,EAAK,EAAG,EAAK,EAAc,EAAE,EAOpC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,GAAQ,KAAK,IACf,EAAS,UACR,GAAS,SAAW,EAAS,GAAM,GAClC,GAAW,EAAK,EAAO,EAEpB,GAAK,EAAG,GAAK,EAAa,EAAE,GAOnC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,IAAM,IAC/C,GAAQ,KAAK,IACf,EAAS,SACR,GAAS,QAAU,EAAU,IAAM,GAClC,GAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,GAG3C,OAFM,IAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,IAE5C,OADI,IAAU,EACL,GAAI,EAAG,GAAI,EAAS,UAAW,EAAE,GAIxC,OAHM,IAAW,GAAI,EACf,GAAW,GAAI,EAEZ,GAAK,EAAO,GAAK,EAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAe,EAC9B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,IACjC,GAAM,IAAK,GAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAE7B,IACI,EAAQ,GAAW,IAAM,EAAS,GAAW,IAKzD,EAAS,GAAW,IAAM,GAMpC,MAAO,GAAG,YAGZ,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,gBAAgB,EAAO,EAAQ,GAEjD,MAAI,IAGF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAiB,CAAC,EAAG,GAAS,mBAc9B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WACxC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OACnC,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,EAAE,GACpC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,EAAE,GACnC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAMF,OAJM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAS,WACtC,EAAW,EACX,EAAW,EACN,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,GAE3C,OADM,GAAO,EAAM,EAAW,GACrB,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAM,EAAW,IAAM,EAAO,EAAM,EAAW,GAEjD,GAAY,EACZ,GAAY,IAQxB,MAAO,GAAE,YAGX,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,2BA0B/B,OAxBM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAQ,EAAc,EAEnB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,GAAK,EAAG,EAAK,EAAU,EAAE,EAMhC,OALM,IAAW,EAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAMjC,OALM,IAAK,GAAK,EAAc,GACxB,GAAW,EAAO,EAAI,EAAO,GAAK,EAAO,GACzC,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAO,EAAE,IAC7B,GAAM,IAAK,EAAK,EAAQ,GAClB,GAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,EAAS,EAAO,EAAI,EAAO,EAAK,EAAO,GAAK,GAAM,GAK1D,MAAO,GAAG,YAGZ,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAiB,CAAC,EAAG,GAAK,4BAc1B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WAExC,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAK5C,OAJM,GAAK,KAAK,MAAM,EAAK,GACrB,EAAK,EAAK,EAEZ,EAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,GAAW,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAIhE,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAIlC,MAAO,GAAG,YAGZ,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,UAAiB,EAAG,QACb,GAAK,KAAK,WAAW,GAAI,IAGlC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,GAAiB,CAAC,EAAG,GAAU,UAE/B,GAAM,GAAqB,EAAE,MAAM,QAC7B,EAAgB,KAAK,SAAS,EAAQ,QAC5C,EAAS,GAAQ,EAAc,OAI/B,OAHM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GAEpB,EAAI,EAAG,EAAI,EAAO,KAAM,EAAE,GACjC,GAAM,GAAS,EAAO,WAAW,GAE3B,EAAwB,EAAO,QACrC,EAAY,GAAQ,EAAc,EAAO,IAEzC,GAAM,GAAgB,EAAK,WAAW,GACtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,YAGhB,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAa,EAAE,QAAQ,GAAW,GAC7B,QAAQ,GACR,MAAM,EAAkB,IAG9B,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAiB,EAAG,UA8BpB,OA5BM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,OAAU,EAAS,SAAU,EAAE,OACxC,EAAa,EAAO,OAEpB,EAAqB,EAAS,SAAS,GAAK,EAAS,SAAS,GAChE,EAAS,SAAS,GAAK,EAAS,SAAS,GACvC,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAGhD,OAFM,GAAoB,EAAQ,EAC5B,EAAmB,EAAQ,EAAE,QAAQ,GAClC,EAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAMf,OAJM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAChD,EACF,EAAoB,EAAS,EACxB,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAKb,OAHM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,GAAkB,EAAoB,EAAO,EAC1C,GAAO,EAAG,GAAO,EAAS,SAAU,EAAE,IAG7C,OAFM,IAAa,GAAO,EAAc,EACpC,GAAU,GACP,GAAU,GACf,IAAW,EASb,OAPM,IACF,KAAK,IAAI,EAAS,QAAS,EAAuB,IAEhD,GAAkB,GAAkB,GAAO,EAC7C,GAAc,EACd,GAAW,EACX,GAAQ,EACH,GAAS,EAAW,GAAS,EACjC,IAAU,GAEb,OADM,IAAe,EAAmB,GAAS,EAAE,QAAQ,GAClD,GAAO,EAAS,GAAO,EAC3B,IAAQ,GAEX,OADM,IAAa,GAAe,GAAO,EAAE,QAAQ,GAC1C,GAAO,GAAS,GAAO,GAC3B,IAAQ,GACX,GAAM,IAAa,GAAa,GAAO,EAAE,QAAQ,GAC3C,GAAQ,EAAQ,GAAa,GAOnC,GANA,AAAK,IAAa,OAAS,GAAQ,GACjC,GAAc,GACT,AAAI,IAAa,OACtB,KAAY,GACZ,MAEE,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAM,IAAe,GAAkB,EACvC,EAAW,IACP,IAAa,MAAQ,GAAW,GAAQ,KAMtD,MAAO,GAAO,YAGhB,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGzC,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,qBAuB1B,OArBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAgB,EAAK,GAAc,EAAe,GAElD,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,KAIjB,EAAG,IACC,EAAU,EAAe,EAAO,EAAS,EAAO,EAChD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGjC,EAAA,UAAA,mBAAR,SAA2B,EAAa,GAiBtC,OAfM,GAAe,EAAA,OAAU,EAAS,SAAU,SAC5C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,KAAK,WAAW,GACpB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAIf,OAFM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAC7C,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAIb,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAO,EAAG,EAAO,EAAS,SAAU,EAAE,GAG7C,OAFM,GAAa,EAAO,EAAc,EACpC,EAAU,EACP,EAAU,GACf,GAAW,EASb,OAPM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAGlD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAS,EAAW,EAAS,EACjC,GAAU,EAEb,OADM,GAAS,EAAS,EACf,EAAO,EAAS,EAAO,EAC3B,GAAQ,EAEX,OADM,GAAO,EAAO,EACX,GAAO,EAAS,GAAO,EAC3B,IAAQ,GACX,GAAM,IAAO,GAAO,EACd,GAAQ,EAAK,IAAI,EAAO,EAAQ,EAAM,GAAM,GAClD,AAAI,IAAS,GACX,GAAW,GACX,EAAc,EAAS,EACf,EACJ,EAAO,EAAwB,IAM3C,EAAa,IAAI,EAAa,EAAO,EAAQ,EAAM,EAAM,KAMnE,MAAO,GAAa,YAGtB,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAiB,CAAC,EAAG,GAAI,qBAoBzB,OAlBM,GAAe,KAAK,mBAAmB,EAAG,GAC1C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAY,KAAK,WAAW,GAC5B,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,GAAS,EACP,EAAwB,EAC5B,EACA,EAAU,IAAI,EAAO,EAAS,EAAO,EAAO,GAC1C,GACF,EAAS,EAAwB,EACjC,EAAO,EAAuB,EAE5B,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,GAAQ,KAIzB,EAAG,IAAI,EAAS,EAAO,EAAS,EAAO,EAAO,GAMxD,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,kBAqBpB,OAnBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cACf,EAAA,KAAK,cAAc,CAAC,EAAO,EAAW,EAAU,KAE9C,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAE9C,EAAY,EACV,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GACvC,EAAI,EAAG,EAAI,EAAO,IACzB,OAAS,GAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EAAgB,KAAK,IAAI,EAAY,EAAG,KAAK,KAAK,IAClD,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAiB,EAAE,QAAQ,GAC7D,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAgB,EAAE,QAAQ,GACzD,EAAI,EAAG,EAAI,EAAU,IAU5B,OATM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EACF,KAAK,IAAI,EAAW,EAAG,KAAK,KAAK,IAC/B,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GACvD,EAAI,EAAG,EAAI,EAAa,KAI/B,GAAM,GAAU,EAAQ,EAAgB,GAClC,EAAa,EAAQ,EAAgB,GACrC,GAAW,EAAQ,EAAiB,GACpC,GAAc,EAAQ,EAAiB,GAEvC,GAAM,EAAW,IAAW,GAAW,EACvC,GAAS,EAAc,IAAc,GAAc,EACnD,GAAW,GAAO,IAAS,IAAO,EAExC,EAAO,KAAe,GAK9B,MAAO,GAAA,OAAU,EAAQ,CAAC,EAAO,EAAW,EAAU,KAGxD,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAChD,GAAiB,CAAC,EAAI,GAAI,0BA+B1B,OA7BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GAOrD,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAMhD,EAAW,KAAK,SAAS,EAAG,QAC9B,EAAS,EACJ,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAU,EAAI,EAAE,QAAQ,GACrB,EAAI,EAAG,EAAI,EAAS,IAU3B,OATM,GAAM,EAAI,EACV,EAAc,KAAK,MAAM,GACzB,EAAiB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAU,GAEpD,EAAe,EAAU,EAAc,EAAE,QAAQ,GACjD,EAAkB,EAAU,EAAiB,EAAE,QAAQ,GAEvD,EAAU,EAAM,EAChB,EAAiB,EAAM,EACpB,EAAI,EAAG,EAAI,EAAQ,IAmB1B,OAlBM,GAAM,EAAI,EACV,EAAe,KAAK,MAAM,GAC1B,EAAgB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAS,GAClD,EAAU,EAAM,EAChB,EAAiB,EAAM,EAEvB,EAAkB,EAAe,EAAe,EAAE,QAAQ,GAC1D,EAAmB,EAAe,EAAgB,EAAE,QAAQ,GAC5D,GACF,EAAkB,EAAe,EAAE,QAAQ,GACzC,GACF,EAAkB,EAAgB,EAAE,QAAQ,GAE1C,GACF,EAAiB,EACf,GAA6B,EAAiB,EAC9C,GAA6B,EAAU,EACvC,GAAsB,EAAU,EAC7B,GAAI,EAAG,GAAI,EAAO,MACzB,GAAM,IAAQ,EAAS,KACvB,EAAO,EAAkB,KACrB,GAAQ,GACZ,EAAO,EAAmB,KAAM,GAAQ,GACxC,EAAO,GAAqB,KACxB,GAAQ,GACZ,EAAO,GAAsB,KAAM,GAAQ,GAKnD,MAAO,GAAA,SAAY,EAAQ,CAAC,EAAO,EAAQ,EAAS,GAAQ,EAAE,QAGhE,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,yBAsBpB,OApBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cAAa,EAAQ,EAAY,EAAW,GAEzD,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GAE5C,EAAe,EACV,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAY,EACZ,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAc,EAAmB,EAAE,QAAQ,GACpD,EAAI,EAAG,EAAI,EAAU,IAO5B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAW,EACX,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAY,EAAmB,EAAE,QAAQ,GAClD,EAAI,EAAG,EAAI,EAAa,KAG/B,GAAM,GAAS,EAAQ,EAAY,GACnC,EAAO,KAAkB,EAKjC,MAAO,GAAA,OACH,EAAQ,CAAC,EAAO,EAAW,EAAU,GAAc,EAAE,QAG3D,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,iCAiC1B,OA/BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GACrD,EAAW,KAAK,SAAS,EAAG,QAK5B,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAGzC,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAS,IAM3B,OALM,GAAY,EAAc,EAAI,EAAE,QAAQ,GAGxC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAY,GAC7C,EAAI,EAAG,EAAI,EAAQ,IAO1B,OANM,GAAY,EAAY,EAAI,EAAE,QAAQ,GAGtC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAW,GAE5C,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHI,GAAQ,EAGH,EAAW,EAAG,EAAW,EAAW,KAC3C,GAAM,GAAM,EAAW,EAEvB,GAAI,EAAM,GAAK,GAAO,EACpB,SAGF,GAAM,GAAY,EAAc,EAAM,EAAG,QAAQ,GAC3C,GAAgB,EAAM,EACtB,GAAmB,KAAK,IAC1B,EAAU,EACV,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAC9B,GAAI,IAAM,GACR,SAEF,OAAS,IAAW,EAAG,GAAW,EAAU,MAC1C,GAAM,IAAM,GAAW,EAEvB,GAAI,GAAM,GAAK,IAAO,EACpB,SAGF,GAAM,IAAY,EAAY,GAAM,EAAG,QAAQ,GACzC,GAAgB,GAAM,EACtB,GAAmB,KAAK,IAC1B,EAAS,EACT,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAE9B,AAAI,IAAM,IACR,IAAS,EAAS,GAAY,KAIpC,EAAO,EAAY,GAAK,EAKhC,MAAO,GAAA,SAAY,EAAQ,EAAE,MAAO,EAAE,QAGxC,EAAA,UAAA,6BAAA,SACI,EAAa,EAAqB,EAAc,EAChD,GACF,GAAiB,EAAG,gCAEpB,GAAM,GAAW,EAAE,MAAM,GACnB,EAAO,EAAW,EAClB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAO,EAAE,KACT,EAAS,GAAI,cAAa,GAEhC,WAA2B,GAQzB,OAPM,GAAiB,EAAS,EAC5B,EACA,EAAS,EAAiB,KAAK,IAAI,EAAG,EAAiB,GACrD,EAAe,EAAS,EAC1B,KAAK,IAAI,EAAiB,EAAa,GAEvC,EAAM,EACH,GAAkB,EAAc,KACrC,GAAM,GAAI,EAAQ,GAClB,GAAO,EAAI,EAEb,MAAO,GAGT,OAAS,GAAS,EAAG,EAAS,EAAM,KAClC,GAAM,GAAM,EAAkB,GACxB,EAAM,EAAQ,GAAU,KAAK,IAAI,EAAO,EAAQ,EAAK,CAAC,GAC5D,EAAO,GAAU,EAGnB,MAAO,GAAA,SAAY,EAAQ,EAAE,QAG/B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAiB,EAAI,WAQrB,OAPM,GAAW,EAAG,MAAM,GACpB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAmB,KAAK,SAAS,EAAW,QAC5C,EAAoB,KAAK,SAAS,EAAY,QAC9C,EAAS,GAAI,cAAa,EAAG,MAC7B,EAAO,EAAG,KAEP,EAAS,EAAG,EAAS,EAAM,KAQlC,OAPM,GAAiB,EAAS,EAC1B,EACD,EAAS,EAAkB,KAAK,IAAI,EAAG,EAAiB,GACvD,EAAY,EAAS,EACvB,KAAK,IAAI,EAAU,EAAiB,EAAc,GAElD,EAAO,EACF,EAAI,EAAY,EAAI,EAAU,IACrC,GAAQ,KAAK,IAAI,EAAiB,GAAI,GAExC,EAAO,EAAQ,EAAO,EAEtB,OAAS,GAAI,EAAY,EAAI,EAAU,KACrC,GAAI,GAAM,GAAK,EAAQ,EAAO,EAAiB,GAC3C,EAAkB,GAAU,EAChC,AAAI,IAAW,GACb,IAAO,KAAK,IAAI,EAAM,CAAC,IAEzB,GAAO,EAAS,GAChB,EAAO,IAAM,GAGjB,MAAO,GAAA,SAAY,EAAQ,EAAG,QAGhC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAiB,EAAQ,eASzB,OAPM,GAAgB,EAAa,EAAS,EAAA,QAAW,GACjD,EAAY,EAAc,MAAM,GAChC,EAAY,EAAc,MAAM,GAChC,EAAM,EAAA,MAAkB,CAAC,EAAW,GAAa,SACjD,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAW,KAAK,SAAS,EAAc,QAEpC,EAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,GAAS,EAAI,EAGb,EAAM,GAAI,cAAa,EAAY,GACzC,EAAI,GAAK,EAAS,GAClB,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,EAAE,EACxC,EAAI,GAAS,EAAI,EAAQ,GAAK,EAAS,EAAS,GAKlD,OAFM,GAAS,GAAA,KAAgB,EAAK,YAC9B,EAAY,EAAI,EACb,EAAW,EAAG,EAAW,EAAY,EAAE,GAC9C,GAAM,GAAI,IAGV,EAAQ,EAAY,GAAY,EAAI,OAEpC,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,IACtC,GAAI,EAAI,EAAI,IACV,EAAQ,EAAY,GAAY,EAChC,QAKR,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAiB,EAAS,UAE1B,GAAM,GAAM,GAAI,cAAa,EAAQ,KAAO,GAC5C,EAAI,KAAK,GAGT,OAFM,GAAa,KAAK,SAAS,EAAQ,QAEhC,EAAQ,EAAG,EAAQ,EAAQ,KAAM,EAAE,EAC1C,AAAI,EAAW,IAAU,GAAK,EAAW,GAAS,GAChD,GAAI,EAAQ,EAAQ,EAAW,IAAU,GAG7C,MAAO,GAAA,SAAY,EAAK,CAAC,EAAQ,KAAM,GAAQ,UAGjD,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,GAAiB,EAAO,qBAExB,GAAM,GAAY,KAAK,SAAS,EAAM,QAChC,EAAa,KAAK,SAAS,EAAO,QACxC,MAAO,IACH,EAAW,EAAY,EAAe,EAAc,IAG1D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,IAAe,OACf,WAAM,MAAA,+DACF,IACR,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAgB9D,OAdM,GAAY,EAAE,MAAM,GACpB,EAAc,EAAE,MAAM,GACtB,EAAa,EAAE,MAAM,GACrB,EAAa,EAAE,MAAM,GAErB,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAU,KAAK,SAAS,EAAE,QAC1B,EACF,GAAI,cAAa,EAAY,EAAe,EAAc,GAE1D,EAAY,EACP,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAI,EAAG,EAAI,EAAc,EAAE,EAGlC,OAFM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACZ,EAAI,EAAG,EAAI,EAAa,EAAE,EAIjC,OAHM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACf,EAAW,GAAU,EAAY,GAAW,EACzC,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAM,EAAI,EACV,EACF,EAAM,EAAc,GAAM,EAAc,GAAM,EAAc,IAChE,EAAO,KAAe,EAAQ,GAKtC,MAAO,GAAA,SACH,EAAQ,CAAC,EAAW,EAAc,EAAa,KAG7C,EAAA,UAAA,oBAAR,SACI,EAAW,EAAW,EACtB,GACF,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAE,MAAO,EAAE,OAC9D,EAAS,EAAA,OAAU,EAAU,GAC7B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GACxD,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GAExD,EAAU,EAAO,OACvB,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,EACpC,EAAQ,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAK3D,QAFM,GAAO,KAAK,WAAW,GACvB,EAAO,KAAK,WAAW,cACpB,GACP,GAAM,GAAM,EAAO,WAAW,GAExB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAEzB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAE/B,EAAQ,GAAK,EAAG,EAAM,GAAS,EAAM,KAX9B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,IAA7B,GAcX,MAAO,GAAO,YAGhB,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,QAAA,aAEA,EAAA,UAAA,eAAA,WACE,MAAO,KAIT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAGtB,EAAA,UAAA,cAAA,SACI,EACA,EACA,EACA,EACA,EACA,GAmBF,OAjBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GACjC,EAAW,EAAM,MAAM,GAEtB,EAAA,EAAA,GAAY,EAAA,EAAA,GACb,EACF,EAAA,OAAU,CAAC,EAAU,EAAY,EAAW,GAAc,WAExD,EAAU,KAAK,SAAS,EAAM,QAC9B,EAAa,KAAK,SAAS,EAAS,QACpC,EAAY,KAAK,SAAS,EAAO,QAEjC,EAAW,EAAO,QAClB,EAAY,EAAO,QAKhB,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAW,EAAI,EACf,EAAK,EAAQ,GACb,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GAExB,EAAe,EAAW,GAChC,GAAI,GAAQ,EACV,SASF,OANM,GAAe,EAAa,EAC7B,GAAK,GAAO,GAAc,GAAM,GAAa,GAC9C,EACE,EACD,EAAY,EAAM,GAAK,GAAO,GAAa,GAAM,GAAY,GAAK,EAE9D,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAgB,EAAa,EAC/B,EAAM,GAAc,GAAK,EAAK,EAC9B,GAAO,GAAK,GAAO,GAAc,GAErC,GAAI,EAAO,GAAK,EAAO,EAAc,GACnC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAGzB,SAGF,GAAI,IAAW,WAKb,OAJM,GAAS,KAAK,MAAM,GACpB,EAAY,KAAK,KAAK,GACtB,GAAQ,EAAO,EAEZ,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAOF,OAJM,IAAU,KAAK,MAAM,IACrB,GAAW,KAAK,KAAK,IACrB,GAAQ,GAAO,GAEZ,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAI,GAAM,EAAI,GAAU,EAAS,GAAK,EAAS,EAAS,GACpD,EAAO,EAAS,GACd,GAAU,EAAU,GAE1B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAS,EAAS,GACjD,EAAO,EAAS,GACpB,GAAM,IAAW,EAAU,GAE3B,EAAM,EAAI,GAAU,EAAS,GAAK,EAAY,EAAS,GACnD,EAAO,EAAS,GACpB,GAAM,IAAa,EAAU,GAE7B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAY,EAAS,GACpD,EAAO,EAAS,GACpB,GAAM,IAAc,EAAU,GAExB,GAAM,GAAW,IAAW,IAAW,GACvC,GAAS,GAAc,IAAc,IAAc,GAEzD,EAAM,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC9D,EAAO,OAAO,GAAO,GAAQ,IAAS,IAAO,QAIjD,QAAS,GAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAKF,OAFM,IAAW,KAAK,MAAM,IACtB,GAAW,KAAK,MAAM,GACnB,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,IAAQ,EAAI,GAAW,EAAS,GAClC,GAAW,EAAS,GAAK,EAAO,EAAS,GACvC,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,IAAU,EAAU,OAM5C,MAAO,GAAO,YAGhB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAiB,GACvB,MAAO,MAAK,QACR,EAAe,EAAc,EAAa,EAAY,EACtD,EAAY,EAAW,EAAS,EAAc,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAE1C,GAAI,IAAc,EAChB,MAAO,GAAA,OAAU,GAAI,EAAa,EAAE,OAOtC,OAJM,GAAS,GAAI,GAAA,aAAa,CAAC,EAAW,GAAY,EAAE,OACpD,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAQ,KAAK,SAAS,EAAE,QAErB,EAAI,EAAG,EAAI,EAAW,KAG7B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,GAAgB,EAAM,EAAQ,GAC9B,EAAM,KAAK,GAEb,GAAI,EAAe,GAAK,GAAgB,EAAE,KAAO,EAC/C,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,EAAE,OAGzD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,EAAO,OAAO,EAAI,EAAY,GAAK,EAAM,EAAe,EAAY,GAGxE,MAAO,GAAO,WAAW,QAAQ,IAGnC,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAe,EAAA,OAAU,GACzB,EAAiB,GACvB,MAAO,MAAK,QACR,EAAS,EAAS,EAAO,EAAY,EAAW,EAAY,EAC5D,EAAS,EAAc,IAG7B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,EAAQ,GAAS,EAAA,KAAK,WAAW,GACjC,GAAM,GACF,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IACrD,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,OAGnD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAEhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,GAAM,GAAS,EAAA,KAAK,kBACD,EAAE,MAAO,EAAA,KAAK,cAAc,EAAE,QACjD,MAAO,MAAK,WAAW,EAAQ,EAAE,MAAO,EAAE,QAG5C,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGxC,EAAA,UAAA,QAAR,SACI,EAAiB,EAAiB,EAAoB,EACtD,EAAmB,EAAoB,EACvC,EAAmB,EACnB,GACF,GAAM,GAAe,CAAC,EAAa,EAAW,GAExC,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAc,KAAK,SAAS,EAAQ,QAE1C,GAAI,IAAe,EACjB,MAAO,GAAA,OAAU,GAAI,EAAO,EAAQ,OAGtC,GAAM,GAAS,GAAI,GAAA,aAAa,EAAc,EAAQ,OACtD,EAAO,OAAO,KAAM,KAAK,SAAS,EAAa,QAAuB,IAEtE,OAAS,GAAI,EAAG,EAAI,EAAY,KAG9B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,EAAM,KAAK,GACX,GAAgB,EAAM,EAAQ,GAGhC,GAAI,EAAe,GAAK,GAAgB,EAAa,EACnD,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,GAGvD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,AAAI,EACF,EAAO,OAAO,EAAe,EAAY,IACrC,EAAY,EAAI,EAAY,GAEhC,EAAO,OAAO,EAAe,EAAY,GAAK,EAAQ,OAAS,EAC3D,EAAY,GACZ,EAAY,EAAI,EAAY,GAItC,MAAO,GAAO,WAAW,QAAQ,IAErC,GAzjFoC,EAAA,eC5DpC,YAqB8B,GAE5B,OADM,GAAe,GAAI,cAAa,EAAK,QAClC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAa,GAAK,KAAK,IAAI,EAAK,IAElC,MAAO,GAGF,GAAM,IACT,SAAC,GACQ,GAAA,GAAA,EAAA,OAAA,EACD,EAAa,EAAK,QACpB,EAAe,GAAI,cAAa,EAAA,KAAK,cAAc,EAAE,QACzD,GAAI,EAAE,QAAU,aACd,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC7C,EAAe,GAAc,OAS7B,QAPM,GAAc,EAAW,KAAK,IAAI,EAAE,QACpC,EAAO,EAAY,mBAAmB,KACtC,EAAO,EAAY,mBAAmB,KACtC,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC/B,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAO,EAAS,GAChB,EAAO,EAAS,GACtB,EAAa,GAAK,KAAK,MAAM,EAAM,GAGvC,MAAO,GAAW,WAAW,EAAc,EAAE,MAAO,YAG7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICzDd,YAwB6C,GAE3C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAmB,GACzB,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAE3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAEhC,EACF,EAAA,KAAK,uBAAuB,EAA0B,GAEpD,EAAQ,EAAO,OACf,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAE7D,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAG1D,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAE5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,KAX7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,IAA5B,GAeX,MAAO,CAAC,EAAQ,IClEpB,YAqBwB,GAEf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,KAAM,EAAA,EAAA,KAEP,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACzC,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAEzC,EAAc,EAAQ,eAAe,EAAK,MAAO,aAEjD,EAAU,EAAQ,KAAK,IAAI,EAAY,QAK7C,SAAQ,mBAAqB,CAC3B,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,GACpD,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,IAG/C,EAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC/Cd,YAsBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IClCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAGP,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAIvC,GAAM,GAAc,EAAA,MAAS,EAAE,OACzB,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,EAAE,MAAO,CAAC,MAAO,aAEpD,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAQ,KAAM,GAAc,QAAO,IAE/D,SAAY,UACZ,EAAQ,8BAA8B,GAE/B,EAIT,GAAI,EAAE,QAAU,aACd,GAAM,GAAW,GAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,IAC5C,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAO,EAAE,MAAO,CAAC,MAAK,KAElE,SAAQ,8BAA8B,GAE/B,EAGT,GAAI,CAAC,EAAA,KAAK,gBAAgB,EAAE,MAAO,IAGjC,GAAM,GAAS,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAC7C,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAO,MAAO,MAAK,GAG3D,GAAI,IAAU,SACZ,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAe,WAAW,KAAK,GACrC,MAAO,GAAQ,eAAe,EAAE,MAAO,QAAS,GAGlD,GAAI,IAAU,QAIZ,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAO,EAAA,KAAK,aAAa,CAAC,GAAI,EAAE,OAEhC,EAAA,GAAA,SAAA,EAAA,GAAA,MAAA,KAAA,EAAA,EAAA,IAAA,EAAA,MAAA,GAAA,EAAA,EAAA,QAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAQ,eAAe,EAAa,OAAQ,GAGrD,KAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,GAG1D,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC9Fd,YAsCI,EAAc,EACd,EAAuC,GACzC,MAAI,IAAe,KACV,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAiB,CAAC,EAAG,GAAI,GAEzB,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,IAInD,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,aACzC,GAAM,GAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGjC,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAY,KAAM,GAAa,QAAS,IAE5D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,OAEP,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,iBASd,GAE5C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAyB,EACzB,GACN,GAAM,GAAc,EAAA,aAAa,2BAA2B,EAAQ,GAC9D,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAY,OACzB,EAAgB,EAAA,KAAK,eAAe,GAEpC,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GACxD,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GAExD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAEvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GACvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GAEvD,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAO,EAAI,EAAM,OACjB,EAAO,EAAI,EAAM,OAEjB,EACF,EAAG,EAAM,EAAO,GAAI,EAAM,EAAO,EAAI,GAAI,EAAM,EAAO,GACnD,EAAM,EAAO,EAAI,IAExB,EAAe,GAAK,EAAO,KAC3B,EAAe,GAAK,EAAO,SAG7B,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EACF,EAAG,EAAM,EAAS,GAAI,EAAM,EAAS,EAAI,GAAI,EAAM,EAAS,GACzD,EAAM,EAAS,EAAI,IAE1B,EAAe,GAAK,EAAS,KAC7B,EAAe,GAAK,EAAS,MAhBtB,EAAI,EAAG,EAAI,EAAe,OAAQ,MAAlC,GAmBX,MAAO,CAAC,EAAgB,EAAgB,ICvL5C,AAsBO,GAAM,IAAU,GAA8B,SAAC,EAAG,GAAM,MAAA,GAAI,IACtD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAGpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAwBsC,GAEpC,MAAO,UAAC,EAAQ,EAAO,GAGrB,OAFM,GACF,EAAA,KAAK,uBAAuB,EAA0B,EAAO,QACxD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,IChCX,YAiCI,EAAc,EAA0B,GAC1C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAQlB,OALM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,GAAS,EAAE,MACpB,EAAY,EAAA,KAAK,kBAAkB,EAAQ,GACxC,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,gBAclD,EAAc,EAA4B,GAC5C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAGlB,GAAM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAS,GAAS,EAAE,MACpB,EAAY,EAAU,EAAQ,EAAQ,GAC5C,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,IC3EtD,AAsBO,GAAM,IAAW,GAAsB,SAAC,GAAO,MAAA,MAAK,KAAK,KACnD,GAAiB,GAAwB,EAAA,KAAM,IAE/C,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,YAoBI,EAAmB,EAAoB,EACvC,GAIF,OAHM,GAAO,EAAA,KAAK,uBACd,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,GCpCT,AAqBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CACL,KAAM,EAAQ,EAAQ,EAAQ,EAC9B,KAAM,EAAQ,EAAQ,EAAQ,KAIzB,GACT,GAAiB,EAAA,SAAU,GAAc,IAEhC,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICrCd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,GAAI,KAAK,KAAK,KACxD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,YAuBI,EAAkB,EAAiB,EAAgB,EACnD,GACF,GAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,GACF,GAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAIhD,OADM,GAAU,EAAA,KAAK,uBAAuB,EAA0B,GAC7D,EAAI,EAAG,EAAI,EAAQ,EAAE,GAC5B,GAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,SAAC,EAAa,GAAM,MAAA,GAAM,EAAM,KAC/C,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEpB,MAAO,eAIL,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAAO,EAAA,EAAA,KAEd,GAAiB,EAAG,SAEd,GAAA,GAAA,EAAA,WAAA,iBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAA,WAAW,kBAAkB,EAAG,EAAQ,GAExC,GAAM,GAAO,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAClC,EAAU,GAAU,EAAM,EAAQ,EAAO,EAAE,MAAO,EAAE,OAC1D,MAAO,GAAQ,eAAe,EAAO,EAAE,MAAO,GAGzC,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IClEd,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAEpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAqBI,EAAmB,EAAkB,EAAiB,EACtD,GASF,OARM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GAEjC,EAAS,EAAA,KAAK,uBAChB,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAO,EAAE,GAK3B,OAJM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAGhC,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,EAAK,IAGvB,GAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE3B,MAAO,GC3CT,YAoBI,EAAuB,EAAc,EAAiB,GAgExD,OA1DM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAyDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GACtB,EAAI,EAAG,EAAI,EAAO,IACzB,EAAS,IAAM,EAAM,GAEvB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACxC,EAAS,IAAM,EAAM,GAevB,OAVM,GAA0C,GAG1C,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAA0B,GAC1B,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EAC/C,EAAI,EAAG,EAAI,EAAM,GAAQ,KAEhC,GAAI,GAAO,OACX,GAAI,EAEF,EAAU,EAAO,GAAG,gBAGpB,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG1C,EAAU,EAAW,KAAK,KAI5B,GAAI,EAAe,KAAa,OAC9B,EAAQ,GAAK,EAAe,QAE5B,GAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAOvB,GAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,GAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,SAAC,EAAoB,GACzC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAOxE,GAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAE7B,CACL,aAAc,EAAa,OAC3B,YAAW,EACX,QAAO,GCzJX,+MCAA,GAGM,IAAU,QCHhB,AA8BA,EAAA,gBAAgB,MAAO,WAAM,MAAA,IAAI,KAAkB,GC9BnD,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,YAoBI,EAAqB,EAAkB,EAAiB,EACxD,EACA,GAsBF,OArBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAS,EAAA,OAAO,EAAS,SAAU,GACnC,EAAa,EAAO,OAEpB,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAoB,EAAI,EACxB,EAAmB,EAAI,EAAQ,GAC5B,EAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAM1C,OALM,GAAW,EAAK,EAAe,EAC/B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,EAAkB,EAAoB,EAAK,EACxC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAQzC,OAPM,GAAW,EAAK,EAAc,EAC9B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAc,EACd,EAAW,EACX,EAAQ,EACH,EAAK,EAAO,EAAK,EAAO,GAAM,GAErC,OADM,GAAW,EAAmB,EAAK,EAAQ,GACxC,GAAK,EAAO,GAAK,EAAO,IAAM,GACrC,GAAM,IAAW,EAAW,GAAK,EAAQ,GACnC,GAAQ,EAAQ,GAAW,GACjC,AAAK,IAAa,OAAS,GAAQ,EACjC,EAAc,GACT,AAAI,IAAa,OACtB,IAAY,GACZ,KAGJ,GAAI,MAAM,GACR,MAGJ,GAAM,IAAe,EAAkB,EAAK,EAAmB,EAC/D,EAAW,IACP,IAAa,MAAQ,EAAW,EAAQ,EAKpD,MAAO,eAIL,EAAqB,EAAkB,EACvC,EAAmC,EACnC,GADmC,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,IAYF,OAXM,GAAe,EAAA,OAAO,EAAS,SAAU,SACzC,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,EAAA,OAAO,EAAQ,EAAO,GAC1B,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,GAG1C,OAFM,GAAW,EAAK,EAAe,EACjC,EAAQ,EACL,EAAQ,GACb,GAAS,EAKX,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAGzC,OAFM,GAAW,EAAK,EAAc,EAChC,EAAQ,EACL,EAAQ,GACb,GAAS,EAOX,OALM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAK,EAAO,EAAK,EAAO,GAAM,EAErC,OADM,GAAK,EAAK,EACP,EAAK,EAAO,EAAK,EAAO,GAAM,GACrC,GAAM,GAAK,EAAK,EACV,EAAQ,EAAK,IAAI,EAAG,EAAI,EAAI,GAClC,AAAI,EAAQ,GACV,GAAW,EACX,AAAI,EACF,EAAc,EACR,IAAI,EAAS,SAAW,GAAM,EAAS,QAAU,GAC3C,EAAS,WACb,EACH,GAAK,EAAS,QAAU,GAAM,EAAS,WAAa,EAEzD,EAAc,EAAK,EAAuB,GAKlD,EAAa,IAAI,EAAa,EAAG,EAAI,EAAI,IAKjD,MAAO,GCrJT,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAqBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBAyB9B,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAgB,EAAK,GAAe,GAEpC,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAGF,GAAM,GAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAGf,EAAG,IAAI,EAAU,EAAe,EAAG,EAAK,EAAK,GAKrD,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,IC1Fd,YAsBoC,GAK3B,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EAAG,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,KAAM,EAAA,EAAA,SAE/B,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGV,GAAiB,CAAC,EAAG,EAAM,EAAU,EAAO,GAAS,aAEhD,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAsBpB,OAnBM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACtC,EAAU,EAAQ,KAAK,IAAI,EAAS,QAAQ,OAC5C,EAAQ,EAAQ,EAAQ,KAAK,IAAI,EAAM,QAAQ,OAC/B,GAAI,cAAa,CAAC,IAClC,EAAU,EACZ,EAAQ,KAAK,IAAI,EAAO,QAAQ,OAChC,GAAI,cAAa,CAAC,IAChB,EAAU,GAAI,cAAa,EAAM,QAEjC,EAAgB,EAAQ,OACxB,EAAc,EAAM,OACpB,EAAgB,EAAQ,OACxB,EAAc,EAAM,OAEtB,EAAO,EACP,EAAK,EACL,EAAK,EACL,EAAK,EACA,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,EAAQ,GAAK,EAAQ,KAChB,GAAM,GAAK,EAAM,MAAS,EAAM,KAC7B,KAAK,KAAK,EAAQ,KAAQ,GAClC,AAAI,GAAQ,GACV,GAAO,GAET,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAGT,MAAO,GAAQ,eAAe,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAM,IAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,IC5Fd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,YAAa,SAAC,EAAI,GAC9D,GAAM,GAAY,EAClB,MAAI,GAAK,EAAU,aACV,EAAU,aAEZ,EAAK,EAAU,aAAe,EAAU,aAAe,IAGnD,GAA2B,CACtC,WAAY,EAAA,YACZ,YAAa,MACb,WAAY,IChCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YAsBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAED,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,EAAQ,OAAO,EAAE,QAEjB,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAEjC,GAAI,EAAM,oBAAsB,MAC9B,GAAM,GAAO,EAAM,mBAAmB,KAChC,EAAO,EAAM,mBAAmB,KAEtC,EAAK,MAAQ,EACb,EAAK,MAAQ,EAGf,MAAO,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,OAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,ICzDd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,KAED,EAAQ,EAAA,KAAK,eAAe,EAAM,EAAO,GAAG,OAAO,GACrD,EAAW,EAAA,aAAa,gBAAgB,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtE,GAAI,EAAA,KAAK,cAAc,KAAc,EACnC,MAAO,GAAQ,eAAe,EAAU,EAAO,GAAG,MAAO,IAI3D,GAAM,GAAU,EAAO,OAAO,SAAA,GAAK,MAAA,GAAA,KAAK,cAAc,EAAE,OAAS,IACjE,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAM,GAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAGlC,GAFA,EAAA,aAAa,uBAAuB,EAAQ,GAExC,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAC5D,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAE5D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAC3D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAE3D,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAO,IAEtE,SAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAUT,GAAM,GAAW,EAAQ,IAAI,SAAA,GAC3B,GAAM,GAAY,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAC7C,EAAQ,CAAC,GAAI,GACnB,MAAO,IAAQ,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAO,EAAE,MAAO,CAAC,MAAK,OAIxD,EACI,EAAA,aAAa,gBAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAE7D,GAAM,GAAU,EAAA,KAAK,uBACjB,EAAQ,GAAG,MAAoB,EAAA,KAAK,cAAc,IAEtD,GAAI,EAAS,GAAG,MAAM,KAAO,GAE3B,GAAI,GAAS,EACb,EAAS,QAAQ,SAAA,GACf,GAAM,GAAM,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACjC,EAAO,EAAA,KAAK,cAAc,EAAE,OAElC,EAAQ,IAAI,EAAK,GACjB,GAAU,SAGZ,GAAI,GAAY,EAEhB,EAAS,QAAQ,SAAA,GAKf,OAJM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAErC,EAAO,EAEF,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EAEpC,OADM,GAAS,EAAM,EAAS,GAAK,EAC1B,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EACpC,EAAQ,EAAS,GAAO,EAAM,KAIlC,GAAa,EAAE,MAAM,KAIzB,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtD,EACF,EAAQ,eAAe,EAAe,EAAO,GAAG,MAAO,GAE3D,SAAS,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KAErD,EAGF,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,ICjId,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,SAAC,GAwCX,OAxCY,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OACJ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAE,MAAM,OAEhB,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAChD,EAAa,EAAO,MAAM,OAE1B,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOI,EAAU,EAAA,KAAK,cAAc,GAC7B,EAAU,EAAS,OACnB,EAAa,EAAA,KAAK,kBAAkB,EAAE,MAAO,GAM1C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,IAAO,EAAO,EAAc,EAAQ,KACjC,GAAI,EAAG,GAAI,EAAY,EAAE,IAEhC,OADI,IAAS,OAAO,iBACX,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,GAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAS,EAAA,KAAK,WAChB,CAAC,EAAG,GAAK,GAAK,IAAI,EAAO,EAAA,KAAK,eAAe,EAAE,QAC7C,GAAc,EAAA,KAAK,WACrB,CAAC,GAAG,GAAG,IAAI,EACX,EAAA,KAAK,eAAe,EAAO,QACzB,GAAM,EAAM,IAAU,EAAW,IACvC,AAAI,GAAM,IACR,IAAS,MAMnB,GAAM,IAAc,EAAA,KAAK,WACrB,CAAC,EAAG,EAAM,EAAM,IAAI,EAAS,EAAA,KAAK,eAAe,IACrD,EAAW,IAAe,GAMlC,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAY,EAAE,OAAQ,EAAU,EAAE,OAExD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAU,MAAO,EAAE,SCrG9C,AAsBO,GAAM,IAA+C,CAC1D,WAAY,EAAA,yBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,yBAAwB,QACtC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EAAY,EAAA,KAAK,0BACD,EAAO,MAAO,EAAO,OAOlC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAO,EACP,GAAO,EACF,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAO,GACP,GAAO,MAMjB,EAAU,IAAM,IAAM,IAAM,EAAI,GAAG,GAAM,GAAM,GAMvD,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAO,MAAO,EAAO,OAEhE,MAAO,CAAC,OAAM,GAAE,MAAO,EAAO,MAAO,MAAO,EAAO,SCrHvD,AAsBO,GAAM,IAA8C,CACzD,WAAY,EAAA,wBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,wBAAuB,QACrC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EACF,EAAA,KAAK,0BAA0B,EAAE,MAAO,EAAE,OAOrC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAU,EAAO,EAAK,EAAI,EAC1B,GAAU,EAAO,EAAK,EAAI,EACrB,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAS,GACT,GAAS,MAMnB,EAAU,GAAG,IAAQ,IAAQ,IAAM,EAAI,GAAG,GAAM,GAAM,GAM9D,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAE,MAAO,EAAE,OAEtD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrH7C,AAsBO,GAAM,IACT,GAA6B,SAAC,EAAW,GAAc,MAAA,GAAI,IAClD,GAAM,GAAiB,EAAA,IAAK,IAE5B,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC7Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,IAAM,EAAI,EAAM,KAAK,IAAI,GAAM,IAEnD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC3Bd,AAqBA,GAAM,IAAI,EAAA,aAAa,MACjB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAEX,GAAgB,GACzB,EAAA,IACA,SAAC,GACC,GAAM,GAAO,KAAK,KAAK,GACjB,EAAI,KAAK,IAAI,GACb,EAAI,EAAO,GAAM,GAAI,GAC3B,MAAO,GACF,GACK,OAAK,EAAI,IAAM,EAAK,IAAM,EAAI,IAAM,EAAI,IAAM,EAC/C,KAAK,IAAI,CAAC,EAAI,MAIhB,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Cd,YAmCI,EAAmB,EACnB,GAgBF,OAfM,GAAa,EAAM,MACnB,EAAQ,EAAW,GACnB,EAAW,EAAW,GAEtB,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EAAS,EAAU,mBAAmB,KACtC,EAAS,EAAU,mBAAmB,KAGtC,EAAc,CAAC,EAAO,GACtB,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAA,KAAK,uBAAuB,UAAW,GACpD,EAAa,EAAA,KAAK,uBAAuB,UAAW,GAEjD,EAAI,EAAG,EAAI,EAAO,KAmBzB,OAjBM,GAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAE7B,EAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAG7B,EAAQ,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAG,KAAM,GAAI,QAAS,IAGtD,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,KACP,EAAM,EAAA,aAAa,uBAAuB,EAAM,GAE7C,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAI,EAAA,aAAa,oBAAoB,EAAK,GAChD,EAAW,EAAI,EAAW,GAAK,EAAE,KACjC,EAAW,EAAI,EAAW,GAAK,EAAE,KAGnC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAG3C,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,GAChD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,GAAY,QAAS,IAE1D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,cAIL,EAAmB,EACnB,GACF,GAAM,GAAY,EAAA,KAAK,cAAc,EAAM,OAErC,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAG5D,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAGlE,GAAI,GAAc,IAChB,GAAM,GACF,GAAU,EAAU,EAAU,EAAW,EAAS,GAEhD,EAAc,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAEjD,GAAI,GACF,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,EAAO,MACvD,EACF,EAAW,eAAe,EAAa,UAAW,EAAO,MAEvD,EAAuB,EAAW,eACpC,GAAI,UACJ,EAAA,KAAK,kBAAkB,EAA8B,YACnD,EACF,GAAS,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAS,IAExC,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAW,QAAS,IAEhD,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAe,QAAS,IAGpD,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OACtC,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OAE5C,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,CAAC,KAAM,EAAa,KAAM,GAGnC,MAAO,QAEP,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EACF,GAAyB,EAAM,EAAW,GAE9C,MAAO,GAAA,aAAa,uBAAuB,IAI/C,YAAuB,GACrB,MAAQ,GAAO,EAAO,KAAO,EAI/B,YACI,EAAwB,EAAwB,EAChD,EACA,GACF,GAAI,IAAS,EACX,MAAO,CAAC,KAAM,EAAU,KAAM,GAGhC,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EAAO,EAAO,EAEd,EAAc,EAAA,aAAa,qBAAqB,GAEhD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAa,EAAA,aAAa,oBAAoB,GAE9C,EAAc,EAAW,KACzB,EAAc,EAAW,KAEzB,EAAW,CAAC,EAAY,QAExB,EACF,EAAW,eAAe,EAAU,UAAW,GAC7C,EACF,EAAW,eAAe,EAAU,UAAW,GAE7C,EAAgB,GAClB,CAAC,OAAQ,CAAC,KAAM,EAAa,KAAM,GAAc,QAAS,IAGxD,EACF,GAAU,EAAc,EAAc,EAAM,EAAS,GAEnD,EAAgB,EAAa,KAC7B,EAAgB,EAAa,KAE7B,EAAa,CAAC,EAAc,QAE5B,EACF,EAAW,eAAe,EAAY,UAAW,GAC/C,EACF,EAAW,eAAe,EAAY,UAAW,GAE/C,EAAkB,GAAQ,CAC9B,OAAQ,CAAC,KAAM,EAAe,KAAM,GACpC,QAAS,IAGL,EACF,GAAU,EAAa,EAAa,EAAM,EAAS,GAEjD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAI,EAAA,aAAa,UAAU,EAAM,GACjC,EAAS,CAAC,EAAE,KAAK,QAEjB,EAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAC3D,GAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAE3D,GAAc,GAChB,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,IAAY,QAAS,IAEpD,GACF,GACI,CAAC,OAAQ,CAAC,EAAG,GAAa,EAAG,GAAiB,QAAS,IAGzD,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAErB,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAGrB,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAEV,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAGV,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAC9C,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAEpD,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IAElC,CAAC,KAAM,GAAW,KAAM,IAIjC,YACI,EAAkB,EAAc,GAGlC,OAFM,GAAM,GAAI,cAAa,EAAO,GAE3B,EAAI,EAAG,EAAI,EAAM,KAGxB,OAFI,GAAO,EACP,EAAO,EACF,EAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAI,EAAA,aAAa,SAAS,EAAI,EAAG,EAAM,GACvC,EAAO,EAAA,aAAa,oBAAoB,EAAsB,GACpE,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAC3C,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAE7C,AAAI,GACF,IAAQ,EACR,GAAQ,GAEV,EAAA,aAAa,mBAAmB,EAAK,EAAM,EAAM,GAEnD,MAAO,GCjVT,YAuBoB,GAEX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAO,GAElC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICtDd,AAsBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,MACb,WAAY,SAAC,GAUX,OAVY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GAEX,EAAS,KAAK,MAAM,EAAa,GACjC,EAAS,EAAc,EAAY,EAAY,EAEjD,EAAc,EAAU,GAE5B,GAAI,GAAU,GAAK,EAAS,GAE1B,GAAM,GAAmB,EAAS,EAC5B,EACF,EAAc,EAAY,EAAmB,EACjD,EAAc,EAAU,GAE1B,EAAO,GAAU,EAMzB,GAAM,GAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,EAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SCpErD,YAuBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAM,GAEjC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtDd,AAqBO,GAAM,IACT,GAAgB,EAAA,SAAU,SAAC,GAAO,MAAA,QAAO,SAAS,GAAM,EAAI,GAAG,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,IAAI,KAAQ,SAAW,EAAI,GAAG,QAEzD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,QAAO,MAAM,GAAM,EAAI,GAAG,QAEhD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAK,EAAI,GAAG,QAEvC,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC3Bd,AA2BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAa,EACf,EAAS,EAAE,MACT,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,EAAkB,GACnD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACvD,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC1C,GAAI,GAAgB,MAElB,OADM,GAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAO,EAAa,IAGpC,EAAQ,GAAc,EAAO,EAAQ,EAAE,MAAO,EAAc,GAC5D,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAElD,EAAS,EAGX,GAAiB,EAAG,OACpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGd,EAAa,EAAA,KAAK,cAAc,GAEhC,EAAS,GAAQ,EAAO,EAAY,EAAa,EAAE,OACnD,EAAS,EAAW,MAAM,EAAQ,EAAa,EAAE,OAEnD,EAAW,EACf,GAAI,GAEF,GAAM,GAAW,EAAA,aAAa,qBAAqB,EAAa,GAChE,EAAW,EAGb,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,SCtE9C,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAyBlC,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAY,EAAA,OACd,EAAS,SAAU,EAAE,MACrB,GAAiB,EAAS,EAAE,MAAO,EAAE,MAAO,GAAU,QACpD,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,GAAM,GAAS,EAAwB,EAAuB,EACzD,EAAU,IAAI,EAAG,EAAK,EAAK,GAC1B,GAAS,EAAK,EAAuB,EAErC,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAAQ,IAGvB,EAAG,IAAI,EAAS,EAAG,EAAK,EAAK,GAKrC,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,ICnGd,YAoBI,EAAqB,EAAkB,EACvC,EAA8B,GAChC,GAAM,GAAU,EAAA,KAAK,eAAe,GAC9B,EAAW,GAAK,EAAS,EAAQ,EAAO,EAAS,EAAU,OAC3D,EAAe,GACjB,EAAS,EAAQ,EAAO,EAAU,GAAM,GAE5C,MAAO,CAAC,EAAS,OAAQ,EAAa,QC3BxC,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAa,EACnB,GAAiB,EAAG,qBAEpB,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,CAAC,EAAG,GAAI,GACN,EAAA,GAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGT,EACF,EAAW,MAAM,EAAwB,EAAS,SAAU,EAAE,OAC5D,EACF,EAAW,MAAM,EAAuB,EAAS,SAAU,EAAE,OACjE,MAAO,CACL,CAAC,OAAQ,EAAc,MAAO,EAAS,SAAU,MAAO,EAAE,OAC1D,CAAC,OAAQ,EAAe,MAAO,EAAS,SAAU,MAAO,YC/C/D,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEnB,GAAiB,EAAO,2BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC3C7B,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEnB,GAAiB,EAAO,8BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAG,GAAM,MAAC,KAAM,EAAK,EAAI,IAC/C,GACT,GAAiB,EAAA,SAAU,GAAc,KAAsB,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC9Bd,YAuBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,SAAU,EAAA,EAAA,cAEjB,GAAiB,EAAG,OAEpB,GAAM,GAAW,EAAS,IACtB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAE,MAAM,GAAK,EAAE,KAE9C,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAE5B,EAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAQ,EAAE,MAAM,OAChB,EAAW,EAAA,KAAK,eAAe,EAAE,OAEjC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EACF,EAAA,KAAK,uBAAuB,EAAE,MAA0B,GAE5D,AAAI,IAAkB,GACpB,EAAQ,KAAK,GAGf,OAAS,GAAI,EAAG,EAAI,EAAO,KACzB,GAAM,GAAS,EAAA,KAAK,WAAW,EAAG,EAAO,GACnC,EAAY,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAM,KAC3C,EAAW,EAAA,KAAK,WAAW,EAAW,EAAY,GAExD,EAAQ,GAAY,EAAM,GAG5B,GAAM,GAAQ,EAAQ,MAAM,EAAS,EAAU,EAAE,OAEjD,MAAO,CAAC,OAAQ,EAAO,MAAO,EAAU,MAAO,EAAE,OAG5C,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICnEd,AAqBO,GAAM,IAAuB,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAI,IAE/D,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,MACb,WAAY,SAAC,GAiBX,OAjBY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAmB,IAEnB,EAAY,KAAK,IAAI,GACrB,EAAY,KAAK,IAAI,GACrB,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GACX,EAAI,EAAO,GAGb,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACrD,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACzD,EAAS,KAAK,MAAM,EAAS,GAC7B,EAAS,KAAK,MAAM,EAAS,GAE7B,GAAI,GAAc,EAUlB,GATA,AAAI,MAAO,IAAc,UACvB,CAAI,IAAY,EACd,EAAc,EAEd,EAAc,EAAU,IAKxB,GAAU,GAAK,EAAS,GAAc,GAAU,GAChD,EAAS,GAEX,GAAM,GAAmB,EAAU,GAAa,GAC1C,EAAmB,EAAS,EAC5B,GACF,EAAc,EAAmB,EAAmB,EACxD,EAAc,EAAU,IAG1B,GAAM,IAAS,EAAc,EAAY,EAAY,EACrD,EAAO,IAAU,EAMzB,GAAM,IAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,GAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SC3FrD,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAErD,GAAM,GAAO,KAAK,MAAM,GACxB,MAAI,GAAK,EAAO,GACP,KAAK,MAAM,GACT,EAAK,EAAO,GACd,KAAK,KAAK,GAEb,EAAO,IAAQ,EACV,EAEA,EAAO,IAKP,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICxCd,AAqBA,GAAM,IAAa,EAAA,aAAa,gBAC1B,GAAQ,EAAA,aAAa,WAEd,GAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,IAAM,EACD,GAAQ,EAER,GAAc,MAAK,IAAI,GAAM,KAI3B,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICnCd,AAqBO,GAAM,IACT,GAAgB,EAAA,QAAS,SAAC,GAAO,MAAA,GAAK,GAAI,KAAK,IAAI,CAAC,MAE3C,GAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,GAAK,EACA,GACE,EAAK,EACP,EAEA,IAIE,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IClCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AA0BA,GAAM,IAAU,sBACV,GAAY,KAAK,IAAI,IAAW,EAEzB,GAAqB,GAAgB,EAAA,SAAU,SAAC,GAG3D,GAAM,GAAW,EAAK,CAAC,GAIjB,EAAW,EAAK,GAEhB,EAAO,KAAK,IAAI,GAClB,EAEJ,MAAI,GACF,EAAS,EACJ,AAAI,EACT,EAAS,EAET,EAAS,KAAK,IAAI,EAAM,GAEnB,IAGI,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICtDd,YAwB0B,GAKjB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EACA,EAAA,EAAA,KAEP,GAAiB,EAAG,aAKpB,OAHM,GAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAS,GAAc,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAEvD,EAAS,EAAQ,MAAM,EAAQ,EAAU,EAAE,OACjD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,OAGrC,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,MACb,WAAY,ICpDd,YA0B+B,GAKtB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,WAAY,EAAA,EAAA,SAEnB,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAA,KAAK,cAAc,GAE1B,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAA0B,GAE1B,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,GAAY,WAAW,CACrC,OAAQ,CAAC,EAAC,GACV,QAAO,EACP,MAAO,CAAC,SAAU,EAAkB,cAAe,KAG/C,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EACF,EAAA,aAAa,oBAAoB,EAAQ,MAAO,EAAY,EAAM,IAEhE,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,MAAO,GACrC,EACF,GAAQ,CAAC,OAAQ,EAAe,QAAO,EAAE,MAAO,IAE9C,EAAmC,CAAC,EAAG,GACvC,EACe,CAAC,KAAM,GACtB,EACF,GAAU,CAAC,OAAQ,EAAiB,QAAO,EAAE,MAAO,IAElD,EAAqC,CAAC,EAAG,GACzC,EAAmC,CAAC,MAAO,GAC3C,EAAS,GACX,CAAC,OAAQ,EAAqB,QAAO,EAAE,MAAO,IAElD,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAAqC,CAChD,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,ICvFd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,EACD,EAAa,EACnB,GAAiB,EAAG,UAIpB,OAFM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAY,GAAI,cAAa,EAAO,QACjC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAQ,EAAO,GACrB,EAAU,GAAK,EAAQ,EAEzB,GAAM,GAAS,EAAW,MAAM,EAAW,EAAE,MAAO,EAAE,OACtD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrC7C,AAsBO,GAAM,IAAwB,GAA8B,SAAC,EAAG,GACrE,GAAM,GAAO,EAAI,EACjB,MAAO,GAAO,IAEH,GACT,GAAiB,EAAA,kBAAmB,IAE3B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,IChCd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,EAAI,GACvD,GAAM,GAAY,EAClB,MAAI,OAAM,GACD,IAEA,EAAK,EAAI,EAAI,EAAU,QAIrB,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICjCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,YAyBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAEpB,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,IC5Cd,AA6FA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,0yDCvKjB,AAiBA,GAAM,IAAmD,GAEnD,GAA2C,CAC/C,MAAO,GACP,UAAW,GACX,mBAAoB,GACpB,sBAAuB,GACvB,MAAO,GACP,QAAS,GACT,6BAA8B,gBAI5B,EAAsB,GACxB,GAAS,GAAgB,cAGK,GAC9B,GAAI,CAAE,KAAgB,MACpB,GAAM,GAAS,GAAyB,GACxC,GAAI,IAAW,KACb,GAAS,GAAgB,MAEzB,gBAAQ,IAAI,0CAA2C,GAChD,KAGX,GAAM,GAAK,GAAS,GACpB,MAAI,GAAG,gBACL,OAAO,IAAS,GACT,GAAgB,IAGzB,GAAG,QAAQ,EAAG,YACd,EAAG,QAAQ,EAAG,cACd,EAAG,QAAQ,EAAG,OACd,EAAG,QAAQ,EAAG,QACd,EAAG,QAAQ,EAAG,qBACd,EAAG,QAAQ,EAAG,iBACd,EAAG,OAAO,EAAG,cACb,EAAG,OAAO,EAAG,WACb,EAAG,SAAS,EAAG,MAER,GAAS,IAGlB,YAAsB,GACpB,GAAI,MAAO,kBAAoB,aAAe,IAAiB,EAC7D,MAAO,IAAI,iBAAgB,IAAK,KAC3B,GAAI,MAAO,WAAa,YAC7B,MAAO,UAAS,cAAc,UAE9B,KAAM,IAAI,OAAM,0CAIpB,YAAkC,GAChC,GAAI,IAAiB,GAAK,IAAiB,EACzC,KAAM,IAAI,OAAM,0DAElB,GAAM,GAAS,GAAa,GAM5B,MAJA,GAAO,iBAAiB,mBAAoB,SAAC,GAC3C,EAAG,iBACH,MAAO,IAAS,IACf,IACC,IAAiB,EACX,EAAO,WAAW,QAAS,KAC3B,EAAO,WAAW,qBAAsB,IAG3C,EAAO,WAAW,SAAU,ICxFrC,AAoBA,GAAY,IAAZ,AAAA,UAAY,GAgBV,EAAA,EAAA,MAAA,GAAA,QAkBA,EAAA,EAAA,aAAA,GAAA,iBAlCU,IAAA,IAAa,KAqCzB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,SAAA,GAAA,aAJU,IAAA,IAAY,KAOxB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,yBAAA,GAAA,2BACA,EAAA,EAAA,mBAAA,GAAA,qBACA,EAAA,EAAA,mBAAA,GAAA,uBALU,IAAA,IAAmB,iBAqC3B,EAAc,GAChB,MAAO,CAAC,EAAS,eAIf,EAAoB,GACtB,MAAO,GAAa,cAWW,GAC/B,GAAM,GAAO,EAAA,KAAK,cAAc,GAC1B,EAAe,KAAK,KAAK,EAAO,GACtC,MAAO,GAAA,KAAK,oBAAoB,eA6B9B,EAAc,GAChB,MAAO,CACL,KAAK,IAAI,EAAG,KAAK,KAAK,EAAU,IAAK,KAAK,IAAI,EAAG,KAAK,KAAK,EAAO,iBAKlE,EAAc,GACV,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,GAAI,EAAI,cAsBb,EAA2B,GAE7B,GAAM,GAAQ,EAEV,EACA,EACA,EACA,EACA,EAEA,EACA,EAEA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAsB,EAAM,KAC5B,EAA0B,EAAM,KAChC,EAAgC,EAAM,QACtC,EAA4B,EAAM,QAClC,EAAqB,EAAM,IAC3B,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,EAAM,WAC7B,EAAmB,EAAM,OAEzB,GAAsB,EAAG,KACzB,EAA0B,EAAG,KAC7B,EAAgC,EAAG,KACnC,EAA4B,EAAM,KAClC,EAAqB,EAAG,KACxB,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,GAA6B,KAChD,EAA0B,eAC1B,KACJ,EAAmB,EAAG,OAExB,EAAwB,EAAG,KAEpB,CACL,oBAAmB,EACnB,wBAAuB,EACvB,8BAA6B,EAC7B,0BAAyB,EACzB,mBAAkB,EAClB,sBAAqB,EACrB,0BAAyB,EACzB,mBAAkB,EAClB,qBAAoB,EACpB,iBAAgB,GCzOpB,YAsBgC,EAA2B,GACzD,GAAM,GAAc,IACpB,MAAI,GAAA,MAAM,QAAQ,UAChB,GAAgB,GAEX,EAGT,YAAyB,GACvB,GAAM,GAAQ,EAAG,WACjB,GAAI,IAAU,EAAG,SACf,KAAM,IAAI,OAAM,gBAAkB,GAAqB,EAAI,IAK/D,GAAM,IAAc,QACd,GAAc,kBAEa,GAC/B,MAAI,KAAA,MAAM,QAAQ,iCAAmC,IAAQ,GACxD,GAAc,KAAK,IAAI,IAAQ,KAAK,IAAI,GAAO,gBAOlD,EAA2B,GAC7B,OAAQ,OACD,GAAG,SACN,MAAO,eACJ,GAAG,aACN,MAAO,mBACJ,GAAG,cACN,MAAO,oBACJ,GAAG,kBACN,MAAO,wBACJ,GAAG,8BACN,MAAO,oCACJ,GAAG,cACN,MAAO,oBACJ,GAAG,mBACN,MAAO,6BAEP,MAAO,sBAAsB,eAK/B,EAA2B,GAC7B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,aAAa,IAC1B,cAAgB,EAAgB,gDAIlC,EAA2B,GAC7B,GAAM,GAA4B,GAC9B,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,gBAC7B,wCAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAc,KACrD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAc,EAAG,kBAAoB,GAC7D,cAAQ,IAAI,EAAG,iBAAiB,IAC1B,GAAI,OAAM,oCAElB,MAAO,eAIL,EAA2B,GAC7B,GAAM,GAA8B,GAChC,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,kBAC7B,0CAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAgB,KACvD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAgB,EAAG,kBAAoB,GAC/D,SACI,EAAsB,EAAG,iBAAiB,IACxC,GAAI,OAAM,sCAElB,MAAO,GAGT,GAAM,IAAkB,2BACxB,YACI,EAAsB,GACxB,GAAM,GAAwB,GAAgB,KAAK,GACnD,GAAI,GAAyB,MAC3B,QAAQ,IAAI,wCAAwC,GACpD,QAAQ,IAAI,GACZ,OAWF,OARM,GAAa,CAAC,EAAsB,GAEpC,EAAc,EAAa,MAAM;GACjC,EAAM,EAAY,OAAO,WAAW,OAAS,EAC7C,EAAuB,EAAY,IACrC,SAAC,EAAM,GACH,MAAA,GAAA,KAAK,SAAU,GAAa,GAAG,WAAY,GAAO,IACtD,EAAgB,EACX,EAAI,EAAG,EAAI,EAAqB,OAAQ,IAC/C,EAAgB,KAAK,IAAI,EAAqB,GAAG,OAAQ,GAG3D,GAAM,GAAmB,EAAqB,MAAM,EAAG,EAAa,GAC9D,EAAY,EAAqB,MAAM,EAAa,EAAG,GACvD,EAAkB,EAAqB,MAAM,GAEnD,QAAQ,IAAI,EAAiB,KAAK;IAClC,QAAQ,IAAI,EAAc,MAAM;GAAM,IACtC,QAAQ,IACJ,MAAM,EAAA,KAAK,SAAS,EAAU,GAAI,GAClC,iEACJ,QAAQ,IAAI,EAAgB,KAAK;gBAGL,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGR,EAA2B,GAErD,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,KAClC,EAAG,oBAAoB,EAAS,EAAG,eAAiB,GACtD,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,2DAKhB,EAA2B,GAE7B,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,KACtC,EAAG,oBAAoB,EAAS,EAAG,mBAAqB,GAC1D,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,iDAKhB,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,EAAM,EAAG,eACxD,cAIL,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,KAC9D,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,EAAM,EAAG,eACvD,gBAIP,MAAI,GAAA,MAAM,UAAU,mBAAqB,EAChC,EAEF,cAGqB,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGA,EAAe,GACjD,GAAM,GAAiB,EAAA,MAAM,UAAU,0BACvC,GAAK,GAAS,GAAO,GAAU,GAC7B,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IACrC,KAAM,IAAI,OAAM,0BAA4B,EAAY,gBAE1D,GAAK,EAAQ,GAAoB,EAAS,GACxC,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IAC/B,EAAM,IAAI,EAAc,IAAI,EAAc,IAChD,KAAM,IAAI,OACN,0BAA4B,EAC5B,qDAAuD,EAAM,kBAInC,GAChC,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,qBAAqB,kDAIpC,EAA2B,EAAuB,EAClD,EAAqB,EAA6B,EAClD,GACF,GAAM,GAAM,EAAG,kBAAkB,EAAS,GAC1C,MAAI,KAAQ,GAGH,GAET,IAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GACI,EACA,WAAM,MAAA,GAAG,oBACL,EAAK,EAAqB,EAAG,MAAO,GAAO,EAC3C,KACR,GAAa,EAAI,WAAM,MAAA,GAAG,wBAAwB,KAC3C,gBAIL,EAA2B,EAAuB,GACpD,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,iBAInD,EAA2B,GAC7B,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAInD,EAA2B,EAC3B,GACF,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,mBAAmB,EAAS,IACzC,YAAc,EAAc,yCAI9B,EAA2B,EAC3B,GACF,MAAO,GAAG,mBAAmB,EAAS,eAIpC,EAA2B,EAC3B,EAA8C,GAChD,GAAa,EAAI,WAAM,MAAA,IAAgB,EAAI,EAAS,KACpD,GAAa,EAAI,WAAM,MAAA,GAAG,UAAU,EAAwB,iBAGtB,GACtC,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QAC1D,GAAa,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,UACpE,GAAa,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,sBAIjE,EAA2B,EAC3B,GACF,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,iBAIpE,EAA2B,GAC7B,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,KAAM,iBAGjC,GAClC,GAAM,GAAS,EAAG,uBAAuB,EAAG,aAC5C,GAAI,IAAW,EAAG,qBAChB,KAAM,IAAI,OACN,8BAAgC,GAA2B,EAAI,gBAKnE,EAA2B,GAC7B,OAAQ,OACD,GAAG,kCACN,MAAO,wCACJ,GAAG,0CACN,MAAO,gDACJ,GAAG,kCACN,MAAO,wCACJ,GAAG,wBACN,MAAO,kCAEP,MAAO,iBAAiB,GAI9B,YACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAa,EAAI,WAAM,MAAA,OAC/C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,GAElB,MAAO,GAGT,YAA6B,EAA2B,GACtD,GAAM,GAAiB,EAAG,iCAAmC,EACvD,EAAgB,EAAc,EAAG,SACvC,GAAI,EAAgB,EAAG,UAAY,EAAgB,GACjD,GAAM,GAAmB,2BAA2B,EAAc,IAClE,KAAM,IAAI,OAAM,0BAA0B,EAAgB,kBAIlC,EAAiB,GAAA,MAAA,KAAA,QAAA,GAAA,GACpC,EAAA,KAAK,cAAc,EAAM,MAAM,EAAG,EAAM,OAAS,gBAG9B,GAC1B,GAAI,EAAM,SAAW,EACnB,KAAM,OAAM,wDAGd,MAAO,CACL,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EAAG,EAAM,EAAM,OAAS,gBAI5C,GAC3B,GAAI,GAAsC,CAAC,EAAG,EAAG,GAC3C,EAAW,EAAM,SAAW,GAAM,EAAM,SAAW,GAAK,EAAM,KAAO,EAC3E,MAAK,IACH,GACI,CAAC,GAAY,IAAM,OAAK,GAAY,KAEnC,cAIL,EAAoB,SAAA,AAAA,IAAA,QAAA,GAAA,IACtB,GAAI,GAAa,EAAA,MAAM,UAAU,0BAsBjC,GArBA,AAAI,GACF,GAAa,EAAa,EAO1B,EAAW,EAAS,IAChB,SAAC,EAAG,GAAM,MAAA,IAAK,EAAS,OAAS,EAC7B,EAAA,KAAK,kBAAkB,EAAS,IAChC,EAAS,KAIjB,AAAI,EAAS,SAAW,GACtB,GAAW,CAAC,EAAG,EAAS,MAKxB,EAAS,SAAW,GACtB,GAAM,GAAgB,EAAA,KAAK,aAAa,GACxC,EAAW,EAAc,SAG3B,GAAI,GAAO,EAAA,KAAK,cAAc,GAC9B,GAAI,EAAS,QAAU,GAAK,GAAQ,EAClC,MAAO,CAAC,EAAG,GACN,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,IAAM,EACjB,MAAO,GACF,GACH,EAAS,SAAW,GAAK,EAAS,GAAK,EAAS,IAAM,GACtD,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAI,EAAS,IACvC,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,IAAM,EAC/B,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,IACvC,GACH,EAAS,SAAW,GACpB,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,GAC3C,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAK,EAAS,GAAI,EAAS,IACrD,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,EAC7C,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,GAAK,EAAS,IAE1D,GAAI,GAOF,GAAM,GAAW,GAAY,GACzB,EAAO,EAAG,EAAO,EACrB,MAAI,GAAS,QACX,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,IAET,EAAO,EAAY,GAAO,GAAM,GAAO,GAChC,EAAA,KAAK,oBAAoB,GAAM,IAAI,SAAA,GAAK,MAAA,GAAI,IAErD,MAAO,GAAA,KAAK,oBAAoB,GAIpC,YAAgB,GACd,MAAO,GAAI,IAAM,cAOW,EAAkB,GAI9C,GAHA,EAAS,EAAO,MAAM,IACtB,EAAS,EAAO,MAAM,IAElB,EAAA,KAAK,YAAY,EAAQ,GAC3B,MAAO,GAGT,GAAI,CAAC,EAAO,QAAU,CAAC,EAAO,OAC5B,MAAO,GAGT,GAAI,EAAO,KAAO,GAAK,EAAO,KAAO,GAAK,EAAO,KAAO,GACpD,EAAO,KAAO,EAChB,MAAO,GAGT,GAAI,EAAO,SAAW,EAAO,QAC3B,GAAM,GAAa,EAAO,MAAM,IAAI,GAC9B,EAAa,EAAO,MAAM,IAAI,GACpC,GAAI,IAAe,EACjB,MAAO,GAGT,GAAI,GAAO,IAAe,GAAO,IAC5B,GAAO,KAAO,GAAK,EAAO,KAAO,GACpC,MAAO,GAGX,MAAO,GAAO,KAAO,EAAO,IAAM,GAAO,EAAO,KAAO,GAAO,EAAO,IAMvE,GAAI,IACA,eAEmC,GACrC,GAAI,IAAoB,MACtB,GAAM,GAAK,GAAgB,GAC3B,GAAmB,EAAG,aAAa,EAAG,kBAExC,MAAO,kBAIP,GAAmB,mBAGnB,GAAyB,iBAGY,GACrC,GAAI,IAA0B,MAC5B,GAAM,GAAK,GAAgB,GAC3B,GAAyB,EAAG,aAAa,EAAG,yBAG9C,MAAO,MAAK,IAAI,GAAI,gBAG4B,GAEhD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAI,GACE,EAAK,GAAgB,GAE3B,MAAI,IAAa,EAAI,oCACjB,IAAiB,EACnB,EAAoB,EACf,AAAI,GAAa,EAAI,4BAC1B,EAAoB,EAEpB,EAAoB,EAEf,cAGoB,EAA2B,GACtD,GAAM,GAAM,EAAG,aAAa,GAC5B,MAAO,IAAO,iBAGsB,GACpC,IACE,GAAM,GAAK,GAAgB,GAC3B,GAAI,GAAM,KACR,MAAO,SAEF,GACP,eAAQ,IAAI,qCAAsC,GAC3C,GAET,MAAO,eAG0C,GAEjD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,WAGL,CAAC,GAAa,EAAI,0BACpB,MAAO,GAIX,GAAM,GAAwB,GAAuC,GACrE,MAAO,eAYqC,GAC5C,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,GAET,GAAI,CAAC,GAAa,EAAI,4BACpB,MAAO,QAGT,GAAI,GAAa,EAAI,0BACnB,MAAO,IAAuC,GAGhD,GAAM,GAA0B,8BAChC,GAAI,GAAa,EAAI,IACnB,GAAM,GACF,EAAG,aAAa,GACpB,MAAO,IACH,EAAI,GAGV,MAAO,GAGT,GAAM,GAAwB,GAAuC,GACrE,MAAO,GAGT,YAAgD,GAE9C,GAAM,GAAY,GAAiB,GAE7B,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,oBAAqB,EAAO,EAAQ,EAChE,EAAU,mBAAoB,EAAU,iBAAkB,MAE9D,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,EAGT,YAEI,EAA2B,GAC7B,GAAM,GAAY,GAAiB,EAAI,GACjC,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,wBAAyB,EAAO,EAAQ,EACpE,EAAU,mBAAoB,EAAU,qBAAsB,MAElE,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,cAG2B,GAClC,GAAI,IAAiB,EACnB,MAAO,GAET,GAAM,GAAK,GAAgB,GAGrB,EAAa,EAAW,WAAa,KAC3C,MAAO,eAIL,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MAAG,GAAM,smCC7qBzB,AAqBA,GAAM,IAAM,EAAA,MASZ,GAAI,aAAa,YAAa,WAAM,MAAA,IAAI,UAAU,iBAAmB,IAGrE,GAAI,aAAa,gBAAiB,WAChC,MAAI,IAAsB,GACjB,EACE,GAAsB,GACxB,EAEF,IAIT,GAAI,aAAa,iCAAkC,WAAM,MAAA,KAEzD,GAAI,aACA,yBAA0B,WAAM,MAAA,IAAI,IAAI,mBAAqB,IAGjE,GAAI,aAAa,oBAAqB,WAAM,MAAA,KAG5C,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aAAa,aAAc,WAAM,MAAA,IAAI,QAAQ,eAGjD,GAAI,aAAa,2BAA4B,WAAM,MAAA,IAAI,QAAQ,gBAG/D,GAAI,aAAa,kBAAmB,WAAM,MAAA,IAAI,QAAQ,gBAItD,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aACA,+BAAgC,WAAM,MAAA,IAAI,QAAQ,gBAGtD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aAAa,sBAAuB,WAAM,MAAA,IAAI,QAAQ,gBAG1D,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aACA,yBACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAG/C,GAAI,aACA,+BACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAU/C,GAAI,aAAa,+CAAgD,WAC/D,GAAM,GAAe,GAAI,UAAU,iBAEnC,MAAI,KAAiB,EACZ,EAEF,GAAkC,KAO3C,GAAI,aACA,gDACA,WAAM,MAAA,IAAI,UAAU,gDAAkD,GAClE,CAAC,EAAA,YAAY,aAKrB,GAAI,aACA,+BACA,WAAM,MAAA,IAAmC,GAAI,UAAU,oBAM3D,GAAI,aAAa,+BAAgC,WAC/C,MAAO,IAAI,QAAQ,4BACf,GACA,GAAI,QAAQ,kCAOlB,GAAI,aACA,+BACA,WAAM,MAAA,IAA8B,GAAI,UAAU,oBAGtD,GAAI,aACA,0BACA,WAAM,MAAA,IAAoB,GAAI,UAAU,oBAK5C,GAAI,aAAa,4BAA6B,WAK5C,GAAM,GAAc,GAAI,QAAQ,gCAChC,MAAO,GAAc,EAAI,IAU3B,GAAI,aACA,iCACA,WACE,MAAO,IAET,SAAA,GACE,GAAI,EAAY,GAAK,IAAc,GACjC,KAAM,IAAI,OACN,+DACA,mCAAkC,EAAS,QC1LvD,AAiBO,YAAuB,GAC1B,KAAM,GAAe,GAAI,cAAa,EAAK,QAC3C,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC/B,EAAa,GAAK,KAAK,IAAI,EAAK,IAEpC,MAAO,GCtBX,AAoBO,YAAsC,GACzC,MAAO,CAAC,EAAQ,EAAQ,EAAO,EAAO,KAClC,KAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAC3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAQ,EAAO,OACf,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAC7D,GAAI,EAAe,OAAS,EAAe,SAAW,EAClD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAI5D,QAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACjC,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GACrC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GACtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAC5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,IAG5C,MAAO,CAAC,EAAQ,IClDxB,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAG,IAAM,EAAI,GCnBnE,AAoBO,YAA+B,GAClC,MAAO,CAAC,EAAQ,EAAO,KACnB,KAAM,GAAY,EAAA,KAAK,uBAAuB,EAAO,EAAO,QAC5D,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAU,GAAK,EAAG,EAAO,GAAI,GAEjC,MAAO,IC1Bf,AAmBO,KAAM,IAAW,GAAsB,AAAC,GAAO,KAAK,KAAK,ICnBhE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAiBO,YAAiB,EAAO,EAAY,EAAU,GACjD,KAAM,GAAO,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACnE,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAC/B,KAAM,GAAS,EAAI,EACnB,GAAI,GAAM,EAAM,GAChB,OAAS,GAAI,EAAG,EAAI,EAAY,EAAE,GAC9B,KAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACR,GAAM,GAGd,EAAK,GAAK,EAEd,MAAO,GC9BX,AAmBO,KAAM,IAAe,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBvF,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,EAAI,KAAK,KAAK,ICnBrE,AAkBO,YAAmB,EAAM,EAAO,EAAM,EAAO,GAChD,KAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GACrC,GAAI,GACA,KAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAElD,KAAM,GAAU,EAAA,KAAK,uBAAuB,EAAO,GACnD,OAAS,GAAI,EAAG,EAAI,EAAQ,EAAE,GAC1B,KAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,CAAC,EAAK,IAAM,EAAM,EAAM,IACvC,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEtB,MAAO,GCnCX,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBlF,AAiBO,YAAuB,EAAO,EAAQ,EAAO,EAAM,GACtD,KAAM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GACjC,EAAS,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACrE,OAAS,GAAI,EAAG,EAAI,EAAO,EAAE,GACzB,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAEhC,EAAS,GAAI,OAAM,EAAI,QAC7B,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAO,GAAK,EAAI,EAAK,IAEzB,KAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE7B,MAAO,GCjCX,AAiBO,YAAoB,EAAQ,EAAM,EAAO,GAE5C,KAAM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAwDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GAC/B,OAAS,GAAI,EAAG,EAAI,EAAO,IACvB,EAAS,IAAM,EAAM,GAEzB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACtC,EAAS,IAAM,EAAM,GAIzB,KAAM,GAAiB,GAGjB,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAAgB,GAChB,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EACxD,OAAS,GAAI,EAAG,EAAI,EAAM,GAAQ,KAE9B,GAAI,GACJ,GAAI,EAEA,EAAU,EAAO,GAAG,gBAGpB,KAAM,GAAa,GACnB,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG9C,EAAU,EAAW,KAAK,KAG9B,GAAI,EAAe,KAAa,OAC5B,EAAQ,GAAK,EAAe,QAG5B,KAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAM3B,KAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,KAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,CAAC,EAAoB,KACvC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAM9E,KAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAC7B,CACH,aAAc,EAAa,OAC3B,cACA,WC9IR,AAuBE,GAAA,IAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GCpCF,AAmBA,GAAA,IAAA,WAKE,WAAY,EAAuB,GAHnC,KAAA,YAAwB,GAItB,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,UAAU,EAAQ,SAAS,EAAQ,oBAInD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;yBAEZ,EAAS;;;MAKlC,MAAA,MClDA,AAmBA,GAAA,IAAA,WAOE,WAAY,EAAuB,GALnC,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,SAAS,EAAQ,SAAS,EAAQ,oBAIlD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;wBAEb,EAAS;;;MAKjC,MAAA,MCpDA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAqC,EACrC,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAC9B,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,KAAK,YAAc,CAAC,EAAW,GAC/B,GAAM,GAAU,IAAO,MAAS,IAAM,IAChC,EAAe,EACjB,gBACA,+CAEJ,KAAK,SAAW;;;;;kCAKc,EAAU;;;;;8BAKd,EAAU;wBAChB,EAAY;;0BAEV,EAAM;;;;;;;MAShC,MAAA,MC5DA,YAiB+B,EAAc,GAC3C,MAAO,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAG,GAAI,IAAI,gBAG/C,EAAc,GACxC,MAAI,KAAS,EACJ,CAAC,GAEH,GAAe,EAAM,eAGE,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,KAIT,OADI,GAAS,GACJ,EAAI,EAAG,EAAI,EAAM,IACxB,GAAU,EAAK,GACf,AAAI,EAAI,EAAO,GACb,IAAU,KAGd,MAAO,GCxCT,cAgCE,GAAI,GACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAU,kBACV,EAAY,KACZ,EAAY,MACZ,EAAY,KACZ,EAAY,UACZ,EAAS,cACT,EAAe,wBAMf,EAAmB;;;;;;;;;;;MAcnB,EAAmB,GACnB,EAAc;;;;;;;;;OAWd,GAAU,GACV,EAAY,YACZ,EAAY,UACZ,EAAY,UACZ,EAAY,YACZ,EAAS,eACT,EAAe,GAEf,EAAmB;;;;;;;;MASnB,EAAmB;;;;;;;;;MAUnB,EAAc;;;;;;;;OAWT,CACL,QAAO,EACP,UAAS,EACT,UAAS,EACT,UAAS,EACT,UAAS,EACT,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,iBAAgB,EAChB,YAAW,GClIf,YAyBI,EAAkB,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,SACrC,GAAM,GAAU,EAAA,KAAK,eAAe,GACpC,MAAO,GACF,IAAI,SAAC,EAAQ,GACZ,GAAM,GAAQ,OAAO,EAAO,GAAE,MAAM,EAAK,MAAM,EACzC,EAAQ,IAAM,EAAQ,OAAS,EACjC,OAAO,EAAO,EAAI,GAAE,MAAM,EAAK,MAAM,EAAO,GAAE,MAAM,EACpD,YAAY,EAAO,GAAE,MAAM,EAC/B,MAAU,GAAK,KAAK,EAAK,MAE1B,KAAK,gBA+CuB,GACjC,GAAM,GAAU,EAAA,KAAK,eAAe,GAAO,IAAI,SAAA,GAAK,MAAA,GAAE,aAEtD,MAAO;;wBAEe,EAAQ,GAAE,iBAAiB,EAAQ,GAAE;;EAKtD,GAAM,IAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EC5FpC,AAkBO,GAAA,IAAA,EAAA,aAAA,6BAkBH,EAAyB,EAAwB,EACjD,GACF,GAAM,GAA2B,GACjC,EAAW,QAAQ,SAAA,GACjB,GAAM,GAAO,EAAA,KAAK,cAAc,EAAE,UAAU,cAG5C,AAAI,EAAE,UAAU,UACd,EAAe,KACX,iBAAiB,EAAE,KAAO,GAAO,EAAI,IAAI,EAAI,IAAM,IAAE,KAEzD,GAAe,KAAK,qBAAqB,EAAE,KAAI,KAC/C,EAAe,KAAK,qBAAqB,EAAE,KAAI,QAGnD,GAAM,GAAqB,EAAe,KAAK;GAEzC,EACF,EACK,IAAI,SAAA,GAAK,MAAA,IAAwB,EAAG,EAAa,KACjD,KAAK;GACR,EAAc,EAAY,SAC1B,EAAO,KACP,EAA4B,GAA6B,GAC3D,EACA,EACA,EAAe,GAAgB,GAEnC,AAAI,EAAY,SACd,GACI,GAA+B,EAAY,aAAc,GAC7D,EAA+B,GAA8B,IAE7D,GACI,GAAyB,EAAY,aAAc,GACvD,EAA+B,GAA2B,IAG5D,AAAI,GACF,IAAgB,IAGlB,GAAM,GAAS,CACb,EAAc,EAA2B,EACzC,EAAoB,EAAuB,EAAsB,GACjE,KAAK;GACP,MAAO,GAGT,YAA8B,GAC5B,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAiB,OACrB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,WAEpB,KAAM,IAAI,OACH,EAAM,OAAM,2CAKzB,YAAoC,GAClC,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAuB,OAC3B,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,WAE1B,MAAO,IAAmB,IAIhC,YACI,EAAmB,EACnB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAM,GACV,AAAI,EACF,GAAO,GAA2B,GAElC,GAAO,GAAqB,GAG9B,GAAM,GAAU,EAAO,UAAU,aAC3B,EAAW,EAAa,aAC9B,MAAI,GAAQ,QAAU,EAAS,QAC7B,CAAI,EACF,GAAO,GAA+B,EAAQ,GAE9C,GAAO,GAAyB,EAAQ,IAGrC,EAGT,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAwB,EAAsB,OAClD,GACH,MAAO,IAAwB,EAA8B,OAC1D,GACH,MAAO,IACH,EAAsC,WAE1C,MAAO,IAAwB,EAAU,IAI/C,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAkB,EAAsB,OAC5C,GACH,MAAO,IAAkB,EAA8B,OACpD,GACH,MAAO,IACH,EAAsC,OACvC,GACH,MAAO,IACH,EAA8C,OAC/C,GACH,MAAO,IACH,EAAsD,OACvD,GACH,MAAO,IACH,EACA,WAEJ,KAAM,IAAI,OACH,EAAS,OAAM,4CAI5B,YAAsC,GACpC,MAAO;;eAEM,EAAK,UAAS;;IAK7B,YAAoC,GAClC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAuC,GACrC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAyB,GACvB,GAAM,GAAmB,EAAK,QAAO;;;;MAIjC,EAAK,UAAS;MACd,EAAK,aAAY;;;;;;;;;;;;;;;;;;;;;;;MAuBjB,EAAK,iBAAgB;MACrB,EAAK,iBAAgB;MACrB,EAAK,YAAW;;;;;;;;;;;;;;;;;;;;;;;;;MAyBhB,GAAiB;MACjB,GAAiB;MACjB,GAAiB;IAGrB,MAAO,GAGT,GAAM,IAAoB;;;;;;;;;;;;EAcpB,GAAoB;;;;;;;;EAUpB,GAAoB;;;;;;;;;EAWpB,GAAuB;;;;;;;;;;;EAa7B,cACE,MAAO;;;;IAOT,YACI,EAAiB,GACnB,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,MAAI,GAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAKjD,EAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAK9C;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;IAKlD,YACI,EAAiB,GACnB,MAAI,GAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIvC,EAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIpC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;4BACnC,EAAS,GAAE;;IAKvC,YACI,EAAiC,GACnC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAqB,KAAK,KAAK,EAAM,GAAK,GAC1C,EAAgB,EAAqB,KAAK,KAAK,EAAM,GAAK,GAEhE,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;wBAE1B,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAiC,GACnC,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EAAiB,GAWnB,OAVM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEnD,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EACF,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EAAiB,EACjB,EAAU,GACV,EAAS,UAEJ,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAkB,EAAM,EAAM,OAAS,EAAI,GAC3C,EAAU;aACD,EAAC,cAAc,EAAc;kBACxB,EAAC,MAAM,EAAc;MAC/B,EACJ,EAAS,IAAI,EAAC,KAAO,EAGvB,MAAO;UACC,EAAM,OAAM;;oCAEc,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;QAE1C,EAAO;;wBAES,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;mBAE3B,EAAM,OAAM,IAAI,EAAM;;IAKzC,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,MAAO,GAE3B,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,MAAO,GAEjC,MAAO;;kDAEyC,EAAS,GAAE;+BAC9B,EAAS,GAAE;;iCAET,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,KAAM,MAAO,GAEvC,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EAAyB,GAC3B,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,GAAI,EAAA,KAAK,YAAY,EAAO,GAC1B,MAAO;;8CAEmC,EAAe,GAAE,KACvD,EAAe,GAAE;;MAMvB,GAAM,GAAqB,KAAK,KAAK,EAAM,GAAK,GAWhD,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;;iCAE1C,EAAe,GAAE;6BACrB,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAyB,GAC3B,MAAI,GAAA,KAAK,YAAY,EAAO,GACnB;;0CAE+B,EAAS,GAAE,KAAK,EAAS,GAAE;;MAI/D,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKxC,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKrC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;wBACpB,EAAM,GAAE;4BACJ,EAAM,GAAE;;;IAMpC,YAAkC,GAChC,MAAO,SAAS,EAGlB,YAAgC,GAC9B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAO,KACb,MAAO;WACE,EAAQ;eACJ,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAA0B,GACxB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACzE,GAAI,EAAU,UAAU,UACtB,MAAO,SAAS,EAAQ,cAAc,EAAO,KAEzC,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAChB,GAAI,IAAY,GAAK,IAAY,EAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAK9B,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,KAAK,EAAM;6BAC1B,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAe,GAAE,KAAK,EAAe,GAAE;eAClC,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEzE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;UACZ,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAQ,EAAS,GACjB,EAAQ,EAAS,GAEvB,GAAI,IAAU,GAAK,IAAU,EAC3B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAI,KAAU,EACL;cACG,EAAQ;6CACuB,EAAM,cAAc,EAAK;+BACvC,EAAO;;MAIhC,IAAU,EACL;cACG,EAAQ;wCACkB,EAAM,cAAc,EAAK;+BAClC,EAAO;;MAI7B;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,aAAa,EAAM;6BAClC,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAE/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAO,KACb,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,GAC9C,MAAO;aACE,EAAQ;qDACgC,EAAO,OAAO,EAAO;;iBAEzD,EAAK,UAAS,IAAI,EAAO;;MAKxC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAe,KAAK,KAAK,EAAM,GAAK,GAE1C,MAAO;WACE,EAAQ;iCACc,EAAY,KAAK,EAAe,GAAE,KAC7D,EAAe,GAAE;eACR,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAErC,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,IAC9C,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACzB,MAAO;YACC,EAAQ;mDAC+B,EAAO,OAAO,EAAO;6BAC3C,EAAO;;IAK5B,GAAA,GAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,OACvB,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;qDAC+B,EAAM,GAAE;UACnD,GAAkB,GAAU;;MAKpC,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAS,GAAyB,GACxC,MAAI,KAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;4CACtB,EAAO;6BACtB,EAAO;;IAI9B,IAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;uCAC3B,EAAO;6BACjB,EAAO;;IAK3B;UACC,EAAQ;;wBAEM,EAAM,GAAE,YAAY,EAAM;2BACvB,EAAO,KAAK,EAAO;2BACnB,EAAO;;EAKlC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEzD,GAAI,EAAM,KAAO,GACf,GAAM,GAAgB,EAAM,MAAM,GAC5B,EAAW,CAAC,EAAG,GACf,EAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,IAAK,MAAO,OAC5B,MAAO;UACD,GAA2B,GAAa;eACnC,EAAQ;mBACJ,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAM,GAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpC,EAAgB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAO,KAAK,EAAO,KAAK,EAAa,KAAK,EAAY;eACjD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAAK,EAAM,GAC3B,EAAU,EAAM,GAEhB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,SAC9B,MAAO;UACD,GAAqB,GAAa;gBAC5B,EAAQ;mBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO;UAC1C,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAa,EAAU,UAAU,WACvC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;gBACK,EAAQ;;oDAE4B,EAAO;;4BAE/B,EAAO,OAAO,EAAO;iCAChB,EAAO;;QAKtC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;YACC,EAAQ;8CAC0B,EAAM,GAAE;;qDAED,EAAO,OAAO,EAAO;6BAC7C,EAAO;;IAKlC,GAAM,GAAS,GAAyB,GACxC,MAAO;cACK,EAAQ;;4BAEM,EAAO,YAAY,EAAO,cAAc,EAAM;+BAC3C,EAAO,KAAK,EAAO;+BACnB,EAAO;;IAKtC,YAA4B,GAe1B,OAdM,GAAQ,EAAU,UAAU,aAC5B,EAAO,EAAM,OACb,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC7C,EAAgB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC3D,EAAS,0BACT,EAAQ,OAAO,EAAa,kBAAkB,EAAY,eACrD,EAAI,EAAG,EAAI,EAAO,EAAG,IAC5B,EAAS,QAAQ,EAAC,KAAO,EACzB,GAAiB,EAAM,EAAO,EAAI,GAClC,EAAQ,IAAI,EAAC,MAAM,EAAa,MAAQ,EAE1C,GAAM,GAAO,KACb,MAAO;WACE,EAAQ,IAAI,EAAM;oBACT,EAAK;2BACE,EAAO;kCACA,EAAO;qDACY,EAAO,KAAK,EAAO;eACzD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,UACvC,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO,KAAK,EAAO;UACtD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;uBAIC,EAAO,KAAK,EAAO;;0BAEhB,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;gCAEU,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGvC,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO;oBAChC,EAAO;6BACE,EAAO,KAAK,EAAO,aAAa,EAAM;6BACtC,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,UACjD,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;iBAGL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;UAElD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;gCAGU,EAAO,KAAK,EAAO,KAAK,EAAO;;0BAErC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAKpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGxB,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAM;6BACpB,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEnE,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,SAAU,UAC3D,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;;iBAEL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAM,GAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAE3B,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;mBAGzC,EAAO;UAChB,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;0BAGlC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBACzC,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE;iBACnB,EAAM,GAAE;;;yBAGA,EAAO,OAAO,EAAO;+BACf,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;;0BAGM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAO,eAAe,EAAM;6BAC1C,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAA2B,GACzB,GAAM,GAAU,EAAU,KACpB,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAEtD,MAAI,GAAS,EACJ,UAAU,EAAO,IAEnB;0BACiB,EAAM;;iBAEf,EAAO;;;IAMxB,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAEpC,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAE7C,EAAO,GAAkB,GACzB,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,AAAI,EAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAGpC,GAAI,GAAS,sBACP,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAChD,EAAgB,IAAW,EAC3B,EAAU,EAAA,KAAK,cAAc,EAAa,cAC1C,EAAiB,IAAY,EAEnC,GAAI,IAAW,GAAK,CAAC,GAAiB,CAAC,EACrC,EAAS;;cAGA,GAAiB,CAAC,EAC3B,AAAI,IAAY,EACd,EAAS;;QAIT,EAAS;;gBAIF,EAAc,QACvB,GAAM,GAAO,EAAS,EAChB,EAAO,EAAS,EAEtB,AAAI,EAAc,QAAQ,GAAQ,IAAM,EAAc,QAAQ,GAAQ,GACpE,EAAS,8BACJ,AAAI,EAAc,QAAQ,GAAQ,GACvC,EAAS,2EAEJ,AAAI,EAAc,QAAQ,GAAQ,IACvC,GAAS,gDAIb,MAAO;WACE,EAAQ;QACX,EAAI;QACJ,EAAa;8BACS,EAAc,IAAI,EAAqB;QAC7D,EAAM;;IAKd,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAc,EAAa,SAC3B,EAAa,EAAU,UAAU,SACjC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAE1C,GAAI,CAAC,EAAU,UAAU,WAAa,IAAW,GAC7C,EAAU,UAAU,YAAc,MAClC,EAAA,KAAK,YAAY,EAAY,GAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAKpC,GAAM,GAAO,GAAkB,GACzB,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAC7C,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,MAAI,GAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAG7B;YACG,EAAQ;QACZ,EAAI;QACJ,EAAa;kBACH,EAAc,IAAI,EAAqB;;gBAKvB,GAChC,GAAI,GAAQ,EACV,MAAO,MACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QAEP,KAAM,OAAM,gBAAgB,EAAI,yBAKpC,YACI,EAAmB,GAErB,GAAM,GAA0B,KAAK,MAAM,KAAK,UAAU,IAC1D,SAAa,UAAU,aAAe,EAC/B,EAGT,YAA2B,EAAkB,GAC3C,MAAO,GAAS,IAAI,SAAA,GAAK,MAAA,GAAO,KAAI,KAAK,MCt2C3C,AAuBA,GAAA,IAAA,WAOE,WACI,EAAiB,EAAoB,EACrC,GARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,EAAA,KAAK,OACD,EAAM,OAAS,EACf,WAAM,MAAA,aACF,GAAG,OAAO,GAAG,cACb,EAAG,MAAM,IAAE,6CACnB,GAAM,GAAS,EAAM,EAAM,OAAS,GAC9B,EAAU,KAAK,KAAK,EAAS,GACnC,KAAK,YAAc,EAAM,MAAM,EAAG,IAClC,AAAI,EAAU,GACZ,KAAK,YAAY,KAAK,GAExB,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,GAAM,GAAW,KAAK,YAChB,EAAO,EAAS,OAChB,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAEjC,EACA,EACJ,GAAI,IAAY,GACd,EAAa,EAAO,EACpB,GAAM,GAAiB,GAAkB,GACzC,EAAiB;UACb,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE,QAEtB,GAAa,EACb,EAAiB;UACb,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE,IAExB,GAAM,GAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACnD,EAAY,IAAM,EAAS,EAAa,GACxC,EAAc,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IACzC,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAE/C,EAAU,IAAO,MAAS,cAAgB,WAC1C,EAAoB,EAAY,GAAK;sDACO,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM,OAE7D,EAAa;0BACG,EAAW,OAAM;uCACJ,EAAW,OAAM;uCACjB,EAAW,OAAM;qDACH,EAAW,OAAM,UAE5D,EAAgC,EAAY,GAAK;qCACtB,EAAY,OAAM;4CACX,EAAS,OAAM;iDACV,EAAS,MAAM,IAAI,OAAM;SAGtE,KAAK,SAAW;0BACM,EAAY,OAAM;iCACX,EAAS,OAAM;sCACV,EAAS,MAAM,IAAI,OAAM;;QAEvD,EAA6B;;UAE3B,EAAK;4BACa,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;4BAC5C,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;UAC9D,EAAc;yCACiB,EAAS,eAAe,EAAS;sBACpD,EAAS,eAAe,EAAS,OAAO,EAAU;;;2BAG7C,EAAU;;8BAEP,EAAU;;YAE5B,EAAiB;6BACA,EAAU;;;mBAGpB,EAAM;;;;;;;;;;;MAazB,MAAA,MCjJA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAe,GAE1C,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;0CACT,EAAa;;;;;;;;;;;;;;gCAcvB,EAAqB;oBACjC,EAAc;gDACc,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;mBACnC,EAAa;kDACkB,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;;;;;;;MAevD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAc,EAAe,GAExD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAa;;;;;;;;;;;;;;;;;gCAiBvB,EAAoB;oBAChC,EAAa;gDACe,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;;;;;;;;MAgBzD,MAAA,MC3KA,AAwBO,GAAM,IAAmB,CAC9B,KAAM,wCACN,KAAM,yCAGR,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,QAAS,QAAS,QAAS,SAK1C,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GAEnE,KAAK,SAAW;;;UAGV,EAAE;;;;;;;;;;MAYZ,MAAA,MCpDA,AAqBA,GAAM,IAAoB;;;EAKb,GAAM,gBACN,GAAM,gBACN,GAAM,gBAMN,GAAU;;;;;;;;;;EAYV,GAAM;;;;;;;;;EAYN,GAAQ,wBAER,GAAY,wBAEZ,GAAO,uBAEP,GAAa,wBAEb,GAAU,uBAEV,GAAgB,wBAEhB,GAAc,sCAEd,GAAa,sCAEb,GAAM,GAAoB;;EAG1B,GAAM,GAAoB;;EAG1B,GAAM;qBAGN,GAAU,yCAEV,GAAQ,+BAErB,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;MAUZ,MAAA,MC1GA,AAuBA,GAAM,IAAoB;;;;;EAOb,GAAU;;;;;;;;;;;;;;;;;;;;;EAuBV,GAAM;;;;;;;;;;;;;;IAef,GAAoB;;EAIX,GAAQ;;;EAKR,GAAU;;;EAKV,GAAQ;;EAIR,GAAY;;EAIZ,GAAO;;EAIP,GAAa;;EAIb,GAAU;;EAIV,GAAgB;;EAIhB,GAAc;;;;EAMd,GAAa;;;;;EAOb,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIxB,GAAA,WAQE,WACI,EAAY,EAAkB,EAC9B,GAAA,AAAA,IAAA,QAAA,GAAA,IATJ,KAAA,cAAgB,CAAC,IAAK,KAGtB,KAAA,qBAAuB,GACvB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,GAAM,GAAO,KAAK,YAAY,OAC1B,EAAyB,GAC7B,GAAI,EACF,GAAI,IAAS,GAAK,EAAA,KAAK,cAAc,KAAK,eAAiB,EACzD,EAAyB;;;;eAMzB,GAAM,GAAQ,GAAkB,GAIhC,GAHA,EAAyB;YACrB,EAAK;UAEL,IAAS,EACX,GAA0B;yCACK,KAAK,YAAY,GAAE;;;iBAKlD,GAAM,GAAW,GAAY,SAAU,GACvC,GAA0B;;iBAEnB,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;iBAExD,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;;;aASrE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;UAQF,EAAsB;;;;MAMhC,MAAA,MC9MA,AAoBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,KASf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAgBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KCzDA,AAoBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;MAiBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KC5DA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,OAAQ,QAKvB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAepB,MAAA,MCzCA,AAoBA,GAAA,IAAA,WAME,WAAY,GAJZ,KAAA,YAAwB,GAKtB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAI1C,OADM,GAAW,CAAC,YAAY,EAAQ,GAAE,+BAC/B,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAC1B,EAAS,KACL,iBAAiB,EAAQ,GAAE,KAC3B,kBAAiB,EAAC,WAAW,EAAK,QAExC,GAAM,GAAY,EAAQ,OACpB,EAAY,EAAQ,EAAQ,OAAS,GAC3C,EAAS,KAAK,sBAAsB,EAAS,WAAW,EAAS,OAEjE,KAAK,SAAW;;;;;;UAMV,EAAS,KAAK;WAAa;;MAIrC,MAAA,MCzDA,AAuBA,GAAA,IAAA,WAOE,WAAY,EAAoB,GALhC,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,GAAM,GAAQ,KAAK,YACb,EAAO,EAAM,OACb,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAC/B,EAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACzD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAW1C,OARM,GAAU,EAAS,GACnB,EAAe,EAAS,MAAM,IAC9B,EAAc,EAAS,OAEzB,EAAkB,OAAO,EAAO,MAAM,EAAQ,GAAE;;oBAEpC,EAAW,WAAW,EAAa,OAAM;WAEhD,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAK1B,GAAmB;cACX,EAAO,MAAM,EAAQ,GAAE,QAAQ,EAAO,OAAO,EAAQ,EAAI,GAAE;;kBAEvD,EAAC,IAAI,GAAgB,EAAU,EAAS,GAAM;mBAC7C,GAAgB,EAAc,EAAS,GAAM;WAG5D,GAAM,GAAY,EAAQ,OACpB,EAAQ,EAAQ,EAAQ,OAAS,GACvC,GAAmB;;gBAEP,EAAS,IAAI,GAAgB,EAAU,EAAS,GAAM;iBACrD,GAAgB,EAAc,EAAS,GAAM,MAE1D,KAAK,SAAW;uBACG,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IAAE;UAC1C,EAAe;;;;UAIf,EAAK;sCACuB,EAAM;;UAElC,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;cACrC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;;MAMtC,MAAA,MAaA,YAAyB,EAAoB,EAAiB,GAC5D,GAAM,GAAa,EAAS,QAAQ,GAC9B,EAAM,EAAS,IAAI,SAAC,EAAG,GAC3B,MAAI,KAAQ,EACA,EAAC,MAAM,EAEV,IAGX,MAAO,GAAI,OC3Hb,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAiB,EAAS,aAAe,eAE/C,KAAK,SAAW;;;;;;;;;;;;8BAYU,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;oBAIhC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAExC,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;wCAEI,EAAM,aAAa,EAAM;;;;;;;gCAOjC,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;oCAEF,EAAS,YAAW;;oBAEpC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAEjC,KAAK,SAAW;;;;;;;;;;;8BAWU,EAAS,UAAS;kCACd,EAAS,SAAQ;iCAClB,EAAW,MAAM,EAAQ;;kCAExB,EAAS,QAAO;;;;oCAId,EAAS,UAAS;mCACnB,EAAY,MAAM,EAAM;;oCAEvB,EAAS,SAAQ;;;;sCAIf,EAAS,SAAQ;qCAClB,EAAW,MAAM,EAAO;;sCAEvB,EAAS,QAAO;;;;;;;;;;;;;MAetD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAEnD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAchC,EAAW;gDACK,EAAW;;oCAEvB,EAAS,SAAQ;;;;;yBAK5B,EAAW;;kCAEF,EAAY;kDACI,EAAY;;sCAExB,EAAS,UAAS;;;;;;2BAM7B,EAAY;;oCAEH,EAAW;oDACK,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;6BAM5B,EAAW;;sCAEF,EAAS,YAAW;;;;;;;;;;MAY1D,MAAA,MCzSA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;;;;;;;wBAOI,EAAU;;;;;8BAKJ,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;;;;;;;;;MAcpD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;gCAYnB,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;;oCAGF,EAAU;8BAChB,EAAU;;;;;;;;;MAWxC,MAAA,MCzIA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAChD,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAEpC,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;;2BAGT,EAAM,aAAa,EAAM;;;;;;;gCAOpB,EAAY;qCACP,EAAc;;gCAEnB,EAAS,SAAQ;;;;kCAIf,EAAW;uCACN,EAAa;;kCAElB,EAAS,QAAO;;;;oCAId,EAAqB;;;;;;;;oBAQrC,EAAc;;;;;;;;;;;;;;;;;;;kBAmBhB,KAA4B,GAAC;;oBAE3B,EAAc;;0CAEQ,EAAqB;mCAC5B,EAAqB;;;kCAGtB,EAAqB;mCACpB,EAAqB;;;yBAG/B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;;;;;yBAK5B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;gCACrB,EAAqB;;;;;;;;;;UAU3C,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAEtD,KAAK,SAAW;oCACgB,EAAW,KAAK,EAAY,KACxD,EAAW;iCACc,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;gCAgBhC,EAAW;qCACN,EAAa;;gCAElB,EAAS,QAAO;;;;kCAId,EAAY;uCACP,EAAc;;kCAEnB,EAAS,SAAQ;;;;oCAIf,EAAW;yCACN,EAAa;;oCAElB,EAAS,QAAO;;;;sCAId,EAAqB;;;;;;;;;;;;;;;;;oBAiBvC,KAA4B,GAAC;;4CAEL,EAAqB;qCAC5B,EAAqB;2BAC/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;;;2BAG/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;qCACrB,EAAqB;;;;;;;;;MAW1D,MAAA,MC3TA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAa,EAAS,YAAc,EAAS,WAE/C,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;wBAO3B,EAAU;4BACN,EAAU;;;;;;;;;gCASN,EAAY;qCACP,EAAc;;gCAEnB,EAAQ;;;;kCAIN,EAAW;uCACN,EAAa;;kCAElB,EAAQ;;;;;;;;;;;UAWhC,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MCrHA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAR/B,KAAA,cAAgB,CAAC,IAAK,KACtB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,SAgB5B,OAdM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAEjB,EAAW,gCAEN,EAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY;wBACI,EAAC,IAAI,EAAI,EAAC;mBACf,EAAC,IAAI,EAAC;mBACN,EAAC,IAAI,EAAC,eAYrB,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAS,EAAG,EAAS,EAAc,KAC1C,GAAM,GAAI,EAAS,EAOnB,GALA,GAAY;4BACQ,EAAI,EAAc;4BAClB,EAAI,EAAa;UAGjC,IAAgB,GAClB,GAAI,EAAI,GAEN,CAAI,EAAU,IAAM,EAUlB,GAAY;;qCAEW,EAAQ,mCAC3B,EAAQ;2BACC,EAAC,IAAI,EAAC;;;;uCAIM,EAAQ;6BAClB,EAAC,IAAI,EAAC;;;2BAGR,EAAC,IAAI,EAAC;;;;qCAII,EAAQ,mCAC3B,EAAQ;;;;;uCAKa,EAAQ;;;;sBAIzB,EAAC,IAAI,EAAC,+BAA+B,EAAC,IAAI,EAAC;;sBAE3C,EAAC,IAAI,EAAC,wBAAwB,EAAC,IAAI,EAAC;;gBAK5C,GAAY;qCACW,EAAQ,uBAAuB,EAAQ;2BACjD,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;oBAGb,EAAC,IAAI,EAAC,aAAa,EAAC,IAAI,EAAC;gBAI7B,EAAI,EAAI,IAOV,GAAM,GAAkB,EAAU,IAAM,EACpC,EAAA,KAAK,kBAAkB,GACvB,EAEJ,AAAK,EAAgB,IAAM,GAAK,EAAU,IAAM,GAC3C,EAAgB,IAAM,GAAK,EAAU,IAAM,EAC9C,IAAY;oCACQ,EAAU,EAAC,MAAM,EAAe;;uCAE7B,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;kBAMvB,AAAI,EAAgB,GAClB,IAAY;;yCAEW,EAAQ;oDACG,EAAQ;+BAC7B,EAAC,IAAI,EAAC;;+BAEN,EAAC,IAAI,EAAC;;qBAKrB,GAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;mBAG5C,GAAY;oCACQ,EAAe;;uCAEZ,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;;sBAGjB,EAAC,IAAI,GAAI,GAAC,aAAa,EAAC,IAAI,GAAI,GAAC;uBAM7C,AAAI,GAAI,GACN,IAAY;mCACW,EAAQ;cAS/B,AAAI,EAAU,IAAM,EAClB,IAAY;sCACY,EAAW;iDACA,EAAQ;2BAC9B,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;6CAGY,EAAQ;2BAC1B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;;wCAEY,EAAW;mDACA,EAAQ;;;sBAGrC,EAAC,IAAI,GAAI,GAAC,kBAAkB,EAAC,IAAI,GAAI,GAAC;oBAI9C,IAAY;qCACW,EAAQ;2BAClB,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;kCAGC,EAAW;iDACI,EAAQ;2BAC9B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;oBAKhD,GAAY,KAIhB,AAAI,EAAI,GACN,IAAY;0BACI,EAAC,IAAI,EAAC,WAAW,EAAC,KAAK,EAAC;gBAClC,EAAC,IAAI,EAAC,kBAAkB,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,EAAC;YAGzD,AAAI,EAAI,EAAI,GACV,IAAY;4BACI,EAAC,IAAI,GAAI,GAAC,WAAW,EAAC,KAAK,GAAI,GAAC;kBAC1C,EAAC,IAAI,GAAI,GAAC;8BACE,EAAC,IAAI,GAAI,GAAC,eAAe,EAAC,IAAI,GAAI,GAAC,UAM7D,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY,gBAAgB,EAAC,IAAI,EAAC,QAAQ,EAAC,IAAI,EAAC,IAIpD,GAAI,GAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;UAezC,EAAQ;;;UAGR,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MC3UA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAA4B,EAC5B,GAPJ,KAAA,cAAgB,CAAC,QAAS,QAAS,UACnC,KAAA,YAAwB,GAOf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAChC,EAAA,EAAA,GACA,EAAA,EAAA,GAAY,EAAA,EAAA,GACnB,KAAK,YAAc,CAAC,EAAU,EAAY,EAAW,GACrD,GAAM,GAAW,IAAW,WAAa,EAAI,EAEvC,EAAA,CAAA,EAAA,EAAA,KAAA,EAAA,EAAA,MAAC,EAAA,EAAA,GAAkB,EAAA,EAAA,GAGnB,EAAA,EAAA,oHAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAa,EAAA,EAAA,GAW3B,EAAA,EAAA,kHAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAe/B,KAAK,SAAW;yCACqB,EAAW;wCACZ,EAAU;;;;;;;;;;;;;;;;iCAgBjB,EAAK;;;;+BAIP,EAAW;8BACZ,EAAU;;uBAEjB,EAAG;mCACS,EAAgB;4BACvB,EAAkB;;;uBAGvB,EAAG;mCACS,EAAe;4BACtB,EAAkB;;;;;aAKjC,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;MA0BrB,MAAA,MC1GA,GAAA,WAQE,WAAY,EAAiB,EAAoB,GAPjD,KAAA,cAAgB,CAAC,KAQf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAM,OACb,EAAM,EAAY,MAAQ,QAAQ,GAAU,EAAM,UAAS,IAC3D,EAAS,EAAM,EAAM,OAAS,GAChC,EAAY,GACZ,EAAY,GAIhB,AAAI,EACF,GAAY,EAAU,UAAU,GAAS,GAAM,WAC/C,EAAY,EAAU,UAAY,WAElC,GAAY,EAAU,gBAAgB,EAAW,cACjD,EAAa,EAAU,aAAe,cAGxC,KAAK,SAAW;;;UAGV,GAAkB,GAAK;oBACb,GAAc,EAAM,UAAS;sBAC3B,EAAG;;cAEX,EAAS;sBACD,EAAS;YACnB,GAAc,EAAM,UAAS;wBACjB,GAAU,EAAM,UAAS;;;;MAO/C,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,OAAS,MAChB,GAAK,MAAQ,EAAM,mBAAmB,EAAc,UAEtD,EAAM,GAAG,UAAU,EAAK,MAAO,KAGrC,KAEA,YAAmB,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,KACpB,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,KAC/B,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,OAAO,EAAI,KAE/C,KAAM,OAAM,2BAA2B,EAAI,yBAI/C,YAAuB,EAAc,GACnC,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KAEd,KAAM,OAAM,2BAA2B,EAAI,yBCjG/C,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAEjB,KAAA,aAAe,GACf,KAAA,aAAe,GAEf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAuB,EAAmB,GAP9C,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAOtB,KAAK,YAAc,EACnB,KAAK,UAAY,EACjB,KAAK,WAAa,EAClB,KAAK,SAAW;;;;gBAIJ,KAAK,uBAAsB;gBAC3B,KAAK,sBAAqB;gBAC1B,KAAK,sBAAqB;;uBAEnB,EAAS;+BACD,EAAS;uBACjB,EAAS;+BACD,EAAS;mCACL,EAAS;UAClC,KAAK,qBAAoB;;;uBAGZ,KAAK,yBAAwB;;;IAM1C,SAAA,UAAA,qBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,mBAAR,WACE,MAAI,MAAK,aAAe,OACf,KAAK,YAAY,GAEjB,KAAK,YAAY,IAIpB,EAAA,UAAA,uBAAR,WACE,MAAI,MAAK,aAAe,OACf,4BAEA,6BAGb,KC5FA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,CAAC,EAAM,GAC1B,KAAK,SAAW;;;;;;MAQpB,MAAA,MClCA,AAsBA,GAAA,IAAA,WAME,WAAY,GALZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;UAIlB,EAAK,OAAM;;MAIrB,MAAA,MCxCA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;;UAKlB,EAAK,OAAM;;MAIrB,MAAA,MC3CA,AAqBA,GAAA,IAAA,WAKE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IANJ,KAAA,cAAgB,CAAC,KAOf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAS,SACb,AAAI,GACF,GAAS,8BAGX,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;8BAUrB,EAAK;kCACD,EAAK;iDACU,EAAK,OAAO,EAAM;wBAC3C,EAAK,UAAS;;;;;;;;;;;;;;UAc5B,EAAK,OAAM,WAAW,EAAM;;MAItC,MAAA,MCtEA,AAqCA,GAAA,IAAA,WAOE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAW,GACX,EAAS,SACb,AAAI,GACF,GAAS,8BAGX,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAC1B,GAAM,GAAU,EAAM,EAAI,EAE1B,GAAY;;gCAEY,EAAG,MAAM,EAAY,GAAE;gCACvB,EAAG;kCACD,EAAG,MAAM,EAAY,GAAE;kCACvB,EAAG;;;;;;;gCAOL,EAAK;oCACD,EAAK;kDACS,EAAK,OAAO,EAAM;yBAC3C,EAAK,UAAS;;;yBAGd,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;;;UAQ5B,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;;UAWzC,EAAQ;;UAER,EAAK,OAAM,MAAM,EAAM;;MAIjC,MAAA,MC/GA,AAmBO,GAAM,IAAc,CACzB,KAAM,oCACN,KAAM,qCAGR,GAAA,WAKE,WAAY,EAAY,EAA8B,GAJtD,KAAA,cAAgB,CAAC,OAAQ,QAKvB,GAAM,GAAW,EAAW,GAC5B,KAAK,YAAc,EAEnB,GAAM,GACF,EAAU,SAAS,KAAK,GAAO,UAAU,KAAK,GAC5C,EAAoB,EAAa,EAAQ,KAAO,MAEtD,KAAK,SAAW;yCACqB,EAAyB;;;UAGxD,EAAE;;;;kDAIsC,EAAQ;;;;;;8BAM5B,EAAQ;;;;;;;;;yDASmB,EAAiB;;;;;;;;;;MAY1E,MAAA,MCxEA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAiB,GAL7B,KAAA,YAAwB,GAMtB,KAAK,cAAgB,CAAC,KACtB,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;;MASlB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,UAEhE,EAAM,GAAG,UAAU,EAAK,SAAU,KAGxC,KChDA,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,EAAuB,GALrD,KAAA,cAAgB,CAAC,IAAK,WAMpB,GAAM,GAAwB,EAAO,QACrC,EAAY,GAAQ,EACpB,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,EAAQ,GAE7C,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,EAAkB,GACzC,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,mBAAmB,EAAI,yBAErC,GAAI,IAAS,EACX,MAAO,yBAMT,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAElD,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,IAAM,EACR,EAAa,KAAK,kBAAkB,EAAc,GAAE,MAEpD,EAAa,KAAK,GAAG,EAAc,IAGvC,MAAO,GAAa,OC3CtB,GAAA,IAAA,WAIE,WACY,EAA0B,EAAmB,GAA7C,KAAA,SAAA,EAA0B,KAAA,QAAA,EAJtC,KAAA,cAAgB,CAAC,IAAK,WAKpB,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAChC,EAAe,KAAK,SAAW,EAAI,aAAe,UACxD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,KAAK,QAAO;;YAElD,EAAK;;gCAEe,KAAK,SAAQ;;sCAEP,EAAY;;;;QAMlD,MAAA,MC1CA,YAwBmC,GACjC,GAAM,GAAO,KACP,EAAwB,EAAK,QAAO;;MAEtC,EAAK,UAAS;MACd,EAAK,UAAS;MACd,EAAK,UAAS;;;;;OAMlB,MAAO,IAA8B,EAAI,eAGR,GAEjC,GAAM,GAAc,GAAI,cACpB,CAAC,GAAI,EAAG,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,GAAI,EAAG,EAAG,IAClE,MAAO,IAAoC,EAAI,eAGf,GAEhC,GAAM,GAAwB,GAAI,aAAY,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,IAC9D,MAAO,IAAmC,EAAI,GAGhD,YACI,EAA2B,EAAe,EAC1C,EAAwB,EACxB,GACF,GAA+B,EAAO,GACtC,GAAM,GAAU,GAAyB,GAEnC,EAAQ,EAAG,WACjB,UAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAO,KACxD,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAO,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAC5C,EAAa,QACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,QACzD,cAIL,GACF,MAAO,GAAc,gCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAG,mBAIvC,GACF,MAAO,GAAc,oCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAc,kCAIlD,GACF,MAAO,GAAc,kCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAG,2BAIL,GACF,MAAO,GAAc,sCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EAAQ,GAAwC,GAC3D,EAAG,KAAM,EAAG,mBAId,GACF,MAAO,GAAc,0CAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAc,kCAIhB,EAA2B,EAC3B,GACF,GAAM,GAAY,EACZ,EAAW,EAAI,EACf,EAAU,EAAI,EAAM,EAAI,EAC9B,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KAC7C,GAAM,GAAU,GACZ,EAAI,EAAS,eAAgB,EAAc,EAAG,EAAQ,GAC1D,MAAO,IACH,GACI,EAAI,EAAS,KAAM,EAAc,EAAG,EAAQ,eAIlD,EAA2B,EAAuB,EAClD,EAAgB,EAAkB,GACpC,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAEhE,GAAI,GAA2B,EAAuB,EACtD,AAAI,YAAgB,YAClB,GAAgB,GAAI,YAAW,EAAQ,EAAS,GAChD,EAAgB,EAAG,cACnB,EAAiB,EAAG,MAEpB,GAAgB,GAAI,cAAa,EAAQ,EAAS,GAClD,EAAgB,EAAG,MACnB,EAAiB,EAAc,2BAGjC,EAAc,IAAI,GAElB,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAAG,KACvD,EAAe,KAEvB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA2B,EAC3B,GAEF,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAChE,AAAK,EAAqB,eAAgB,YACxC,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAO,MAAO,EAAO,OAAQ,EAAG,EAAG,KAC9D,EAAG,cAAgB,EAAqB,QAEhD,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAG,KAAM,EAAG,cACvC,KAIV,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA6B,EAAc,EAC3C,GAEF,GAAM,GAAS,EAAI,eACnB,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,KAGrD,GAAM,GAAgB,EAChB,EAAiB,EACjB,EAAkB,EAAgB,EAAiB,EAAO,EAEhE,UACI,EACA,WAAM,MAAA,GAAI,WACN,EAAI,kBAAmB,EAAiB,EAAI,eAIpD,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAG,EAAG,EAAS,EAAM,EAAI,KAAM,EAAI,MAAO,KAExE,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,QAE9C,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAM,EAEN,EAAiB,GAAI,cAAa,GAExC,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAGJ,EAAc,EACd,EAAiB,GAAI,YACvB,GAA4C,EAAO,EAAS,IAEhE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAG,EAAG,EAAc,sBAAuB,EAAG,cACpD,KAID,GAAI,cAAa,EAAe,oBAIrC,EAA2B,EAAqB,EAAe,EAC/D,EAAc,EAAsB,EACpC,GACF,GAAM,GAAM,EAEN,EACF,GAAI,cAAa,GACb,EAAc,IAEtB,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAa,GAAI,cAAa,EAAe,EAAe,GAClE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAc,EAAc,EAAG,KAAM,EAAG,MAAO,KAEtD,owBCxTT,kBAgDE,WAAY,GANZ,KAAA,cAAmC,KACnC,KAAA,QAA6B,KACrB,KAAA,SAAW,GAqOX,KAAA,oBAAsB,GAsPtB,KAAA,YAA0B,GAtdhC,GAAM,GAAY,EAAA,MAAM,UAAU,iBAClC,AAAI,GAAM,KACR,MAAK,GAAK,EACV,GAAgB,EAAW,IAE3B,KAAK,GAAK,GAAgB,GAG5B,GAAI,GAAqB,2BACnB,EAA0B,8BAChC,GAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAM,GAAgB,oBAChB,EAAqB,yBAI3B,GAFA,KAAK,sBACD,GAA+B,KAAK,GAAI,GACxC,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,sHAKN,GADA,KAAK,0BAA4B,KAAK,GAAG,aAAa,GAClD,GAAwB,KAAK,GAAI,GACnC,KAAK,8BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,uIAIN,EAAqB,yBACjB,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,KAAK,GAAG,aAAa,WAChB,GAAwB,KAAK,GAAI,GAC1C,KAAK,8BACD,KAAK,GAAG,aAAa,OAEzB,MAAM,IAAI,OAAM,uDAIpB,KAAK,aAAe,GAA8B,KAAK,IACvD,KAAK,YAAc,GAA6B,KAAK,IACrD,KAAK,YAAc,GAA6B,KAAK,IAErD,KAAK,cACD,GAA0B,KAAK,GAAI,KAAK,2BAG9C,cAAA,eAAY,EAAA,UAAA,QAAK,KAAjB,WACE,MAAO,GAAA,MAAM,QAAQ,0CAGhB,EAAA,UAAA,QAAP,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAEF,AAAI,KAAK,SAAW,MAClB,QAAQ,KACJ,wKAIN,AAAI,KAAK,eAAiB,MACxB,QAAQ,KACJ,sMAKN,GAAM,GAAK,KAAK,GAChB,GAAwB,EAAI,WAAM,MAAA,GAAG,WACrC,GAAwB,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QACrE,GAAwB,EAAI,WAAM,MAAA,GAAG,kBAAkB,EAAK,eAC5D,GAAwB,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,QACjE,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,QACrD,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAK,eACvD,KAAK,SAAW,IAGX,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,yBAAP,SACI,EACA,GACF,KAAK,kBACL,GAAoC,KAAK,GAAI,EAAS,IAGjD,EAAA,UAAA,2BAAP,SACI,EAAuB,EAAe,EAAgB,GACxD,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,EAAO,EAAQ,EAAM,KAAK,gBAG3C,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,0BAAP,SAAiC,EAAc,GAE7C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,oBAAP,SAA2B,GAA3B,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,KAAK,gBAAkB,GACzB,IAA6C,KAAK,GAAI,KAAK,aAC3D,KAAK,cAAgB,MAEvB,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAGxD,EAAA,UAAA,gDAAP,SACI,EAAuB,EAAc,GADzC,GAAA,GAAA,KAEE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAM,EAAS,EAAK,kBAGhC,EAAA,UAAA,+BAAP,SACI,EAAqB,EAAe,EAAc,EAClD,EAAsB,GACxB,MAAO,IACH,KAAK,GAAI,EAAQ,EAAO,EAAM,EAAS,EAAc,EACrD,KAAK,gBAGJ,EAAA,UAAA,gCAAP,SAAuC,EAAqB,GAE1D,MAAO,IAA2C,KAAK,GAAI,EAAQ,IAG9D,EAAA,UAAA,wBAAP,SACI,EAAuB,EAAc,GACvC,KAAK,yBAAyB,GAC9B,GAAM,GAAS,GACX,KAAK,GAA8B,EAAM,EAAS,KAAK,eAC3D,YAAK,6BACE,GAGF,EAAA,UAAA,sBAAP,WACE,GAAM,GAAe,KAAK,YAAY,KAAK,IAC3C,MAAO,MAAK,UAAU,IAGhB,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACM,EACA,EAEJ,GAAI,EAAA,MAAM,QAAQ,4BAChB,GAAM,GAAM,EAEN,EAAO,EAAI,UAAU,EAAI,2BAA4B,GAC3D,EAAG,QAEH,EAAgB,WACd,GAAM,GAAS,EAAI,eAAe,EAAM,EAAG,GAC3C,MAAO,KAAW,EAAI,kBAClB,IAAW,EAAI,qBAGrB,EAAQ,MACH,AACH,GAAA,MAAM,UAAU,gDAAkD,EACpE,GAAQ,KAAK,aACb,KAAK,WACL,EAAgB,WAAM,MAAA,GAAK,iBACvB,EACA,EAAA,MAAM,UAAU,mDAMpB,EAAgB,WAAM,MAAA,IAGxB,MAAO,CAAC,MAAK,EAAE,cAAa,IAGvB,EAAA,UAAA,gCAAP,SACI,EAAuB,EACvB,GAFJ,GAAA,GAAA,KAGE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAc,MAK1B,EAAA,UAAA,cAAP,SAAqB,GACnB,KAAK,kBACL,GAAM,GAAK,KAAK,GACV,EACF,GAAgC,EAAI,GAClC,EAA4B,GAA8B,GAC1D,EAAwB,GAAyB,GACvD,UAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAuB,EAAI,GAC3B,AAAI,KAAK,OACP,GAA2B,EAAI,GAEjC,AAAK,KAAK,qBACR,MAAK,WAAW,GAChB,KAAK,oBAAsB,GACvB,EAAI,KAAK,QAAS,KAAK,eAEtB,GAGF,EAAA,UAAA,cAAP,SAAqB,GAArB,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,IAAY,KAAK,SACnB,MAAK,QAAU,MAEjB,AAAI,GAAW,MACb,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAI1D,EAAA,UAAA,WAAP,SAAkB,GAAlB,GAAA,GAAA,KACE,KAAK,kBACL,KAAK,QAAU,EACf,AAAK,KAAK,SAAW,MAAS,KAAK,OACjC,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,WAAW,MAGrD,EAAA,UAAA,mBAAP,SACI,EAAuB,EACvB,GAEF,MAFE,AAAA,KAAA,QAAA,GAAA,IACF,KAAK,kBACD,EACK,GACH,KAAK,GAAI,EAAS,GAEf,GACH,KAAK,GAAI,EAAS,IAInB,EAAA,UAAA,qBAAP,SAA4B,EAAuB,GAAnD,GAAA,GAAA,KAEE,YAAK,kBACE,GACH,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,kBAAkB,EAAS,MAGjD,EAAA,UAAA,0BAAP,SAAiC,EAAuB,GAEtD,YAAK,kBACE,KAAK,GAAG,mBAAmB,EAAS,IAGtC,EAAA,UAAA,sBAAP,SACI,EAAkC,EAClC,GACF,KAAK,kBACL,KAAK,mBACL,GACI,KAAK,GAAI,EAAoB,EAAiB,IAG7C,EAAA,UAAA,uBAAP,SACI,EAAmC,EAAc,GACnD,KAAK,6BAA6B,EAAqB,EAAS,IAG3D,EAAA,UAAA,6BAAP,SACI,EAAyC,EAAc,GACzD,KAAK,kBACC,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,KAAK,6BAA6B,EAA2B,EAAO,IAG/D,EAAA,UAAA,2BAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAK,iCACD,EAAa,EAAU,EAAY,IAGlC,EAAA,UAAA,iCAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAM,IAAI,OAAM,sDAGX,EAAA,UAAA,cAAP,WACE,AAAI,KAAK,SAAW,MAClB,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAA+B,KAAK,KAG/B,EAAA,UAAA,eAAP,WACE,KAAK,kBACL,KAAK,mBACL,GAAM,GAAK,KAAK,GAChB,AAAI,KAAK,OACP,KAAK,gBAEP,GACI,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,UAAW,EAAG,EAAG,eAAgB,MAG7D,EAAA,UAAA,+BAAP,WAAA,GAAA,GAAA,KACE,KAAK,kBACL,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,YAGzC,EAAA,UAAA,uBAAR,WAEE,MAAI,MAAK,6BAA+B,MACtC,MAAK,4BACD,GACI,KAAK,GACL,EAAA,MAAM,UACF,kDAAoD,EACpD,kCACA,6BAIP,KAAK,6BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGd,EAAA,UAAA,WAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EAAQ,EAAI,cAClB,SAAI,WAAW,EAAI,iBAAkB,GAC9B,EAET,GAAM,GAAM,KAAK,+BACX,EAAQ,EAAI,iBAClB,SAAI,cAAc,EAAI,iBAAkB,GACjC,GAGT,EAAA,UAAA,SAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BACjB,EAAI,SAAS,EAAI,kBACjB,OAEF,GAAM,GAAM,KAAK,+BACjB,EAAI,YAAY,EAAI,mBAGT,EAAA,UAAA,uBAAb,SAAoC,wGAClC,MAAA,CAAA,EAAM,EAAA,KAAK,YACP,WAAM,MAAA,GAAK,UAGP,EAAK,iBACD,EACA,EAAA,MAAM,UACF,2DAPhB,SAAA,OAQA,CAAA,EAAO,KAAK,aACR,EAAO,EAAA,MAAM,UAAU,wDAGrB,EAAA,UAAA,aAAR,SAAqB,EAAmB,GACtC,GAAI,IAAsB,EACxB,MAAO,MAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GAEX,EAAmB,EAAI,kBAAkB,EAAO,EAAI,cAE1D,MAAO,GAAmB,SAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,kBAErC,MAAO,GAAmB,MAItB,EAAA,UAAA,iBAAR,SAAyB,EAAmB,GAE1C,GAAI,IAAsB,EACxB,MAAO,GAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,wBACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,cAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,4BACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,WAI9B,EAAA,UAAA,UAAA,SAAU,GAAV,GAAA,GAAA,KACE,MAAO,IAAI,SAAc,SAAA,GACvB,EAAK,cAAc,WAAM,MAAA,GAAa,iBAAiB,WAAM,MAAA,UAMjE,EAAA,UAAA,UAAA,WAGE,OADM,GAAQ,GAAqB,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,YACtD,EAAI,EAAG,GAAK,EAAO,EAAE,GACrB,GAAA,GAAA,KAAA,YAAA,GAAA,UACP,IAEF,KAAK,YAAc,KAAK,YAAY,MAAM,EAAQ,IAG5C,EAAA,UAAA,cAAR,SAAsB,EAAyB,GAA/C,GAAA,GAAA,KAEE,GADA,KAAK,YAAY,KAAK,CAAC,SAAQ,EAAE,UAAS,IACtC,KAAK,YAAY,OAAS,EAE5B,OAGF,EAAA,KAAK,YAAY,WACf,SAAK,YAEE,EAAK,YAAY,SAAW,KAI/B,EAAA,UAAA,yBAAR,SAAiC,GAC/B,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,KAAK,aAC3B,AAAI,KAAK,OACP,GAA+B,KAAK,KAIhC,EAAA,UAAA,2BAAR,WACE,AAAI,KAAK,eAAiB,KACxB,IACI,KAAK,GAAI,KAAK,cAAe,KAAK,aACtC,AAAI,KAAK,OACP,GAA+B,KAAK,KAGtC,GAA6C,KAAK,GAAI,KAAK,cAIvD,EAAA,UAAA,qBAAR,SACI,EACA,GACF,KAAK,yBAAyB,GAC9B,GAAM,GAAS,IACf,YAAK,6BAEE,GAGD,EAAA,UAAA,6BAAR,SACI,EAA8C,EAC9C,GACF,KAAK,kBACL,GAAM,GAAK,KAAK,GAChB,GACI,EAAI,EAAgC,KAAK,aAC7C,AAAI,KAAK,OACP,GAA+B,GAEjC,KAAK,cAAgB,EACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAO,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAO,MAGpD,EAAA,UAAA,iCAAR,SACI,EAAW,EAAW,EAAe,GADzC,GAAA,GAAA,KAEE,KAAK,kBACL,GACI,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,QAAQ,EAAG,EAAG,EAAO,MAG1C,EAAA,UAAA,gBAAR,WACE,GAAI,KAAK,SACP,KAAM,IAAI,OAAM,4CAIZ,EAAA,UAAA,iBAAR,WACE,GAAI,KAAK,SAAW,KAClB,KAAM,IAAI,OAAM,qCAGtB,iBAaqC,GAEnC,OADI,GAAI,EACD,EAAI,EAAI,OAAQ,EAAE,GACvB,GAAM,GAAS,EAAI,KACnB,GAAI,CAAC,EACH,MAGJ,MAAO,GAAI,ECtnBb,YAgEI,EAAqB,EAAuB,EAC5C,GACF,GAAM,GAAW,EAAQ,SACnB,EAA0B,EAAO,IAAI,SAAC,EAAO,GACjD,GAAM,GAAuB,CAC3B,aAAc,EAAM,MACpB,SAAU,EAAM,UAAY,KAAO,EAAM,QAAQ,SACjD,UAAW,EAAM,UACjB,SAAU,EAAM,UAAY,GAAQ,EAAM,QAAQ,SAClD,WAAY,MAEd,MAAI,GAAM,SAAW,MAAQ,EAAM,QAAQ,OAAS,MAChD,EAAM,QAAQ,MAAM,WAAa,GACnC,GAAU,WAAa,EAAM,QAAQ,MAAM,YAEtC,CAAC,KAAM,EAAQ,cAAc,GAAI,UAAS,KAE7C,EAAe,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,YACrC,EAA0B,CAC9B,aAAc,EAAO,MACrB,SAAU,EAAO,QAAQ,SACzB,UAAW,GACX,SAAU,EAAO,QAAQ,SACzB,WAAY,MAER,EAAS,GACX,EAAY,EAAc,EAAU,EAAQ,cAE1C,EAAe,EAAM,cAAc,GAGrC,EAA+B,KAC7B,EAAS,EAAM,mBAAmB,EAAc,MAAO,IAC7D,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAS,EAAM,mBAAmB,EAAc,WAAY,KAK9D,OADM,GAA2D,GACxD,EAAI,EAAG,EAAI,EAAQ,cAAc,OAAQ,KAChD,GAAM,GAAU,EAAQ,cAAc,GAChC,EAAc,GACpB,EAAiB,GACb,EAAM,mBAAmB,EAAc,EAAS,GACpD,EAAiB,SAAS,GACtB,EAAM,mBAAmB,EAAc,SAAS,EAAW,GAGjE,MAAO,CACL,QAAO,EACP,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,aAAY,EACZ,aAAY,EACZ,OAAM,EACN,OAAM,GAIV,YACI,EAAyB,GAC3B,GAAI,EAAW,SAAW,EAAO,OAC/B,KAAM,OACF,4BAA4B,EAAW,OAAM,gBAC7C,sBAAqB,EAAO,OAAM,YAGxC,EAAW,QAAQ,SAAC,EAAG,GACrB,GAAM,GAAS,EAAE,aACX,EAAQ,EAAO,GACf,EAAS,EAAM,MAErB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAQ,GAC5B,KAAM,OACF,kDACA,6BAA4B,EAAM,QAAQ,EAAM,gBAGtD,GAAI,EAAE,WAAa,EAAM,UACvB,OAGF,GAAM,GAAY,EAAE,SACd,EAAY,EAAM,UAAY,KAAO,EAAM,QAAQ,SACzD,GAAI,CAAC,EAAA,KAAK,YAAY,EAAW,GAC/B,KAAM,OACF,6DACA,yBAAwB,EAAS,QAAQ,EAAS,8BAMxD,EAAqB,EAAqB,EAC1C,EACA,GAEF,GAAyB,EAAO,aAAc,GAC9C,GAAyB,CAAC,EAAO,cAAe,CAAC,IAEjD,GAAM,GAAS,EAAO,QAAQ,QACxB,EAAc,EAAO,QAAQ,SACnC,AAAI,EAAO,QAAQ,SACjB,EAAM,6BAA6B,EAAQ,EAAY,GAAI,EAAY,IAEvE,EAAM,uBAAuB,EAAQ,EAAY,GAAI,EAAY,IAEnE,EAAM,WAAW,EAAO,cAGxB,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,CAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,WAGtC,AAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,KAIpC,EAAO,QAAQ,SAAC,EAAO,GACrB,GAAM,GAAU,EAAO,QAAQ,cAAc,GACvC,EAAS,EAAO,iBAAiB,GACjC,EAAe,EAAO,iBAAiB,SAAS,GAEtD,GAAI,GAAU,KAEZ,OAGF,GAAI,EAAM,WAER,GAAI,EAAA,KAAK,cAAc,EAAM,OAAS,EACpC,EAAM,GAAG,UAAU,EAAQ,EAAM,cAAc,SAE/C,GAAI,GAAO,EAAM,cACjB,AAAM,YAAgB,eACpB,GAAO,GAAI,cAAa,IAE1B,EAAM,GAAG,WAAW,EAAQ,GAE9B,OAIF,AAAI,EAAM,QAAQ,OAAS,MAAQ,GAAgB,MACjD,EAAM,GAAG,UAAU,EAAc,EAAM,QAAQ,MAAM,YAGvD,EAAM,sBAAsB,EAAM,QAAQ,QAAS,EAAQ,KAG7D,AAAI,GAAe,MACjB,EAAY,EAAO,EAAO,cAE5B,EAAM,6BAIJ,EAAuB,EAAsB,GAC/C,GAAI,GAAY,GAChB,EAAO,OAAO,GAAQ,QAAQ,SAAA,GAC5B,GAAM,GAAY,EAAE,SAAW,MAAQ,EAAE,QAAQ,OAAS,MACtD,EAAE,QAAQ,MAAM,WAAa,EAC3B,EAAW,EAAE,UAAY,UAAY,EAAE,QAAQ,SACrD,GAAgB,EAAE,MAAK,IAAI,EAAQ,IAAI,IAEzC,GAAM,GAAc,EAAQ,SACxB,EAAM,EAAQ,YAAY,KAE9B,UAAO,IAAM,EAAY,IAAM,EACxB,EC5OT,AAqBA,GAAA,IAAA,WAOE,WACI,EAAuB,EACvB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAsBnB,OAnBE,GAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,cACA,EAAA,EAAA,eACA,EAAA,EAAA,WAEK,EAAA,EAAA,KAAM,EAAA,EAAA,IACP,EAAmB,EAAa,EAChC,EAAO,KACP,EAAiB,IAAe,eAChC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAEhC,EAAW,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,GAAY;gCACY,EAAG;yBACV,EAAG;;4BAEA,EAAY,GAAE,aAAa,EAAY,GAAE;0CAC3B,EAAQ,QAAQ,EAAY,MAC1D,EAAG;6BACc,EAAc,aAAa,EAAgB;;sBAElD,EAAW,GAAO;;qDAEa,EAAQ,QACjD,EAAW,OAAO,EAAI;+BACH,EAAa,2BAChC,EAAgB,QAAQ,EAAU;;wBAEtB,EAAW,GAAO;;2CAEC,EAAU;;sBAE/B,EAAc;;2BAET,GAAM,EAAI,GAAG;;;;;2BAKb,GAAM,EAAI,GAAG;;;;;;;UAWpC,KAAK,SAAW;;;;;;;;;UASV,EAAQ;;UAER,EAAK,OAAM;;MAIrB,MAAA,MC5GA,AAmBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAgB,EAAc,EAChD,GANJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMtB,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;;wBASI,EAAG,UAAU,EAAG;;oCAEJ,EAAI;;;;;0BAKd,EAAW;;;MAKrC,MAAA,MClEA,AAmBA,GAAA,IAAA,WAUE,WACI,EAAsB,EAAqB,EAAc,EACzD,GAXJ,KAAA,cAAgB,CAAC,aAAc,cAAe,MAC9C,KAAA,YAAwB,GAWtB,KAAK,YAAc,EACnB,KAAK,MAAQ,EAAW,GACxB,KAAK,YAAc,EACnB,KAAK,KAAO,EACZ,KAAK,MAAQ,EACb,KAAK,KAAO,EACZ,KAAK,SAAW;;;;;;;;8BAQU,KAAK,MAAK;oDACY,EAAW;yCACtB,KAAK,MAAK;0BACzB,EAAW;;;sCAGC,KAAK,MAAK;;;;;;;;;;;;;;;yBAevB,EAAK,oBAAoB,EAAI;;;;;;;yCAOb,EAAK;0BACpB,EAAI;;;;0CAIY,EAAI;;;;;;;;;;;;;;MAgB9C,MAAA,MC/FA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAgB,EAAc,EAChD,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;gCAQY,KAAK,YAAY,GAAE;gCACnB,KAAK,YAAY,GAAE;;;;;;;;;;;;;;;iCAelB,EAAG;;;;;;;;;;;yBAWX,EAAG,UAAU,EAAG;;;6DAGoB,EAAI;;;;;;;;;;;;;;;;;;;;;0CAqBvB,EAAW;;;MAKrD,MAAA,MC/GA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAY,EAAwB,EAAuB,EACjE,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAcnB,EAAqB;kBACnC,EAAc;gDACgB,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;kDACJ,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;gCAOvB,EAAS;;;;qCAIJ,EAAoB;;;;;;;;MAUzD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EACF,EAAuB,EAAwB,EAAuB,EAC1E,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;gCAiBhC,EAAoB;mBACjC,EAAa;gDACgB,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;kCAOvB,EAAS;;;;;;yBAMlB,EAAqB,MAAM,EAAoB;yBAC/C,EAAoB;;;;;;;;;MAW7C,MAAA,MChLA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkC,EAClC,EAAoB,EAAoB,EACxC,EAA2B,GAD3B,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IACxC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAT/B,KAAA,cAAgB,CAAC,UAAW,WAC5B,KAAA,aAAe,GACf,KAAA,aAAe,GAQb,KAAK,YAAc,EAEnB,GAAM,GAAY,EAAa,EAAO,GAAK,EAAO,GAC5C,EAAwB,KAAK,KAAK,EAAY,GAE9C,EAAU,EAAa,cAAgB,cACvC,EAAU,EAAa,cAAgB,cACvC,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAC1D,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAE5D,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;sCAEa,EAAqB;;;;8BAI7B,EAAqB;sCACb,EAAO;sCACP,EAAO;;;;uBAItB,EAAS,GAAE,MAAM,EAAS,GAAE;uBAC5B,EAAS,GAAE,MAAM,EAAS,GAAE;;;;;;;;;UASzC,EAAc;;UAEd,EAAsB;;;;MAMhC,MAAA,MChGA,AAoBA,GAAA,IAAA,WAQE,WAAY,EAAmB,EAAqB,GAPpD,KAAA,cAAgB,CAAC,SAQf,KAAK,YAAc,CAAC,EAAW,GAE/B,KAAK,SAAW;;;;;;;;;;8BAUU,GAAc,GAAC;;;;;;;;;;0BAUnB,GAAc,GAAC;;MAKvC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAM,mBAAmB,EAAc,SAExD,EAAM,GAAG,UAAU,EAAK,QAAS,KAGvC,KChEA,AAmBA,GAAA,IAAA,WAQE,WACI,EAAoB,EAAe,EAAiB,GARxD,KAAA,cAAgB,CAAC,WASf,KAAK,YAAc,CAAC,EAAY,GAEhC,KAAK,SAAW;;;;8BAIU,EAAQ,YAAY,EAAO;;;MAKzD,MAAA,MCxCA,AAqBA,GAAA,IAAA,WAOE,WACI,GAPJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEzB,GAAI,IAAS,EACX,KAAK,SAAW;;;;aAMhB,GAAM,GAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EACF,GAAwB,EAAM,EAAa,GACzC,EAAQ,GACV,EAAM,EAAY,EAAY,OAAS,GACvC,EAAY,EAAY,OAAS,GAAI,GACnC,EAAS,GAAU,EAAa,GAEtC,KAAK,SAAW;;YAEV,EAAK;;eAEF,EAAoB;;;cAGrB,EAAK;;6BAEU,EAAM;;;SAMnC,MAAA,MAEA,YAA4B,EAAc,GAGxC,OAFM,GAAS,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAG1B,OAFI,GAAW,KAAQ,EAAI,IAAM,OAAK,KAAK,KAAQ,EAAI,IAAM,OAEpD,EAAI,EAAG,EAAI,EAAM,IACxB,EAAW,EAAK,EAAK,OAAS,EAAI,GAAE,IAAM,EAG5C,EAAO,KAAK,GAGhB,MAAO,GAGT,YACI,EAAc,EAAiB,GACjC,GAAI,IAAS,EACX,MAAO,QAAQ,EAAM,GAIvB,OADI,GAAO,GACF,EAAI,EAAO,EAAG,EAAI,EAAM,IAC/B,GAAW,EAAK,GAAE,OAAO,EAAM,GAC/B,AAAI,EAAI,EAAO,GACb,IAAQ,MAIZ,MAAO,GAGT,YACI,EAAc,EAAc,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,GAGT,GAAM,GAAY,EAAK,MAAM,IAE7B,MAAO;cACK,EAAU,GAAE;cACZ,EAAU,GAAE;;;;0BAIA,EAAI;0BACJ,EAAI;IAI9B,YAAmB,EAAiB,GAClC,GAAM,GAAO,EAAM,OACb,EAAe,GAAmB,EAAM,GAC9C,MAAI,KAAS,EACJ;wBACa,EAAM,GAAE;kBAIvB,QAAQ,EAAa,GAAE;8BACF,EAAa,GAAE;8BACf,EAAa,GAAE;uCACN,EAAa,GAAE,ICrItD,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAClB,GANJ,KAAA,cAAgB,CAAC,KAOf,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KACnD,GAAM,GAAO,EAAO,OACd,EAAO,GAAkB,GAEzB,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EACF,CAAC,YAAa,YAAa,YAAa,aAAa,MAAM,EAAG,GAElE,GAAI,IAAS,GACX,KAAK,SAAW;sBACA,EAAK;oBACP,EAAG;;;;;8BAKO,EAAa;;;;;QAMrC,OAEF,KAAK,SAAW;QACZ,EAAI,YAAY,EAAI,IAAI,EAAK;QAC7B,EAAI,UAAU,EAAI,IAAI,EAAG;;;UAGvB,EAAI;;4BAEc,EAAa;;YAE7B,EAAI;2BACW,EAAc;;;MAKzC,MAAA,MCrEA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAClB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KA4BnD,OA3BM,GAAO,EAAO,OACd,EAAQ,GAAkB,GAE1B,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EAAS,GAAY,KAAM,GAC3B,EAAS,GAAY,SAAU,GAC/B,EAAY,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAC1D,EACF,IAAS,EAAI,SAAW,QAAQ,EAAO,MAAM,IAAI,OAAM,IAErD,EAAiB,CAClB,EAAK,mBAAuB,EAAO,EAAO,GAAE;YACzC,EAAM;QAEZ,IAAS,EAAI,GAAK;;SAEf,EAAO,EAAO,GAAE;YACb,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAAE,MACtD,IAAS,EAAI,GAAK,KAAK,EAAO,EAAO,GAAE;cAC/B,EAAM,OAGV,EAAc,IAAS,EACzB,0BACA,6DACA,EAAW,GACN,EAAI,EAAG,EAAI,IAAS,EAAI,EAAI,EAAG,EAAI,EAAG,IAC7C,GAAY;UACR,EAAe,GAAE;cACb,EAAW;mBACN,EAAC,aAAa,EAAa;;YAElC,EAAK;mBACE,EAAC,uBAAuB,EAAO,OAAM,MAAM,EAAS;;QAInE,GAAa,IAAS,EAAI,KAAO,KAEjC,KAAK,SAAW;cACN,EAAK,YAAY,EAAK,IAAI,EAAK;cAC/B,EAAK,UAAU,EAAK,IAAI,EAAG;;;UAG/B,EAAK;;UAEL,EAAQ;;;MAKlB,MAAA,MCrFA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MACzB,EAA0B,cAAc,EAAS,SAAQ,YAC3D,EAAS,QAAO,YAAY,EAAS,WAAU,OAC7C,EACF,SAAS,EAAS,QAAO,YAAY,EAAS,WAAU,OAExD,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;sCACgB,EAAY,KAAK,EAAW;mCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBnB,EAAqB;sBACjC,EAAc;;;kCAGF,EAAS,SAAQ;;;;oCAIf,EAAoB;wBAChC,EAAa;;;oCAGD,EAAS,QAAO;;;;;;;;;;0BAU1B,EAAS;;;mCAIzB,GAAoB,EAAsB,EACA,EACvB,QAAQ,EAAoB,SAAO;;;;;;QAO1D,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;oCACgB,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;0CACT,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;kCAkBZ,EAAmB;;;;gCAIrB,EAAqB;oBACjC,EAAc;;;gCAGF,EAAS,SAAQ;;;;kCAIf,EAAsB;uCACjB,EAAa;;;;yCAIX,EAAa;6CACT,EAAa;6CACb,EAAa;;;cAG5C,EAAa;;;gCAGK,EAAsB;gBACtC,KAA6B,GAAC;;;;;;;;cAQhC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;;;;;cAKxC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;6CACT,EAAa;;;;cAI5C,EAAa;;;oBAGP,EAAW;;MAI/B,MAAA,MAEA,GAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MAE3B,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;;oBAEF,EAAW,KAAK,EAAY,KAAK,EAAW;mCAC7B,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBhC,EAAoB;sBAChC,EAAa;;;kCAGD,EAAS,QAAO;;;;oCAId,EAAqB;wBACjC,EAAc;;;oCAGF,EAAS,SAAQ;;;;sCAIf,EAAoB;0BAChC,EAAa;;;sCAGD,EAAS,QAAO;;;;;;;;;;4BAU1B,EAAS;;;qCAI3B,GACK,EACI,cAAc,EAAS,QAAO,YAC1B,EAAS,SAAQ,YAAY,EAAS,QAAO,YAC7C,EAAS,WAAU,QACvB,UAAU,EAAS,SAAQ,YACvB,EAAS,QAAO,YAAY,EAAS,WAAU,QACxD,QAAQ,EAAqB,MAAM,EAAoB;6BACxC,EAAoB,SAAO;;;;;;;QAQlD,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;;gBAEJ,EAAW,KAAK,EAAY,KAAK,EAAW;iCAC3B,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;;kCAmBZ,EAAmB;;;;gCAIrB,EAAoB;oBAChC,EAAa;;;gCAGD,EAAS,QAAO;;;;kCAId,EAAqB;oBACnC,EAAc;;;kCAGA,EAAS,SAAQ;;;;oCAIf,EAAsB;yCACjB,EAAa;;;;+CAIP,EAAa;mDACT,EAAa;mDACb,EAAa;;;gBAGhD,EAAa;;;kCAGK,EAAsB;kBACtC,KAA6B,GAAC;;;;;;;;gBAQhC,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;;;;;gBAK5C,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;mDACT,EAAa;;;;gBAIhD,EAAa;;;sBAGP,EAAW;;;MAKjC,MAAA,MCvcA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,OAAQ,EAAA,EAAA,QACtC,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAI,GAAsB,MACtB,EAAY,GAEhB,AAAI,IAAe,OACjB,EAAsB,MACjB,AAAI,IAAe,MAExB,GAAsB,cACtB,EAAY,OACP,AAAI,IAAe,OAExB,GAAsB,eACtB,EAAY,OAGd,GAAI,GAAiB,EAAU,IAAI,EAAU,IAAI,EAAU,qEAG3D,AAAI,IAAe,MACjB,EAAc,WACT,AAAI,IAAe,OACxB,EAAc,YACT,AAAI,IAAe,MACxB,EAAc,WACT,AAAI,IAAe,OACxB,GAAc,YAGhB,GAAM,GAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEzC,EAAgB;YACZ,KAAe,OAAK;;mBAEb,KAAe,QAAM;;;;wBAIhB,EAAS;;MAIzB,EAAU,OAEd,AAAI,IAAe,MACjB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SACL,AAAI,IAAe,OACxB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SAGZ,GAAI,GAAmB,GACvB,AAAI,EAAS,EAAa,GACxB,GAAmB;oCACW,EAAM;;;SAKtC,KAAK,SAAW;0CACsB,EAAmB;;;;UAInD,EAAgB;;;;;;;;kCAQQ,EAAU;;kCAEV,EAAmB;;;;;;8BAMvB,EAAqB;;YAEvC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;YAC/B,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCvKA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAuC,GANnD,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAGnB,OADI,GAAW,GACN,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAS,eACb,AAAI,EAAI,IAAM,GACZ,IAAU,kBAEZ,AAAI,EAAI,GACN,IAAU,kBAGZ,GAAY;UACR,EAAM;UACN,GAAI,EAAI,0CAA4C,IAAE;;;;;;mBAM7C,EAAC;;UAEV,GAAI,EAAI,IAAM,IAAE;QAItB,KAAK,SAAW;QACZ,GAAuB,GAAW;QAClC,GAA+B,GAAY;;;;;;;;qBAQ9B,EAAY,GAAE;qBACd,EAAY,GAAE;;UAEzB,EAAQ;;;;MAMlB,MAAA,MAEA,YAAgC,GAC9B,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;QAED,EAAsB;;;ICnF9B,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;;sDAMY,GAAU,GAAC;;;;;;qDAMZ,GAAS,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAgC/D,MAAA,MC1IA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MA+B/D,MAAA,MC3EA,AAmBA,GAAA,IAAA,WAOE,WACI,EAA8C,EAC9C,EAAkB,GARtB,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;uCACxB,EAAQ;;;;;;;;;;;;;;;;;;;;;;gCAsBf,GAAQ,GAAC;uCACF,GAAW,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAkDnD,MAAA,MCzHA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;sBAKpB,EAAe,GAAE;sCACD,EAAe,GAAE;;;wBAG/B,EAAe,GAAE;wCACD,EAAe,GAAE;;;4BAG7B,EAAO;kBACjB,EAAY;;;;4BAIF,EAAM;kBAChB,EAAY;;;;;;;;;;;;MAc9B,MAAA,MC/HA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAI5C,EAAY,EAAe,MAAQ,MAEzC,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;8DAaD,EAAS;;;;;;MAQvE,MAAA,MCpEA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAI5C,GAFA,KAAK,YAAc,EAEf,IAAS,GACX,KAAK,SAAW;;;2BAGK,EAAO,GAAE;;QAG9B,OAEF,GAAM,GAAa,SAAC,GAClB,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,aAAa,EAAC,QAE5B,UAAU,EAAC,KAEd,EAAW,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,KAAI,KAAK,KACpD,EAAO,GAAkB,GAE/B,KAAK,SAAW;;UAEV,EAAI;yBACW,EAAQ;;MAIjC,MAAA,MC1DA,AAqBA,GAAA,IAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAE5C,KAAK,YAAc,EACnB,GAAM,GAAW,GAAY,KAAM,GAC7B,EACC,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACrD,EAAa,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACjE,EAAO,GAAkB,GAC/B,AAAI,IAAS,EACX,KAAK,SAAW;;;;uCAIiB,EAAO,GAAE;cAClC,EAAO,GAAE;eACR,EAAU;2CACkB,EAAO,GAAE;kBAClC,EAAO,GAAE;;;;QAMrB,KAAK,SAAW;;YAEV,EAAI;;uBAEO,EAAK,EAAS,SAAQ;eAC9B,EAAU;yBACA,EAAK,EAAS,SAAQ;;eAEhC,EAAO;yBACG,EAAK,EAAS,SAAQ;iBAC9B,EAAU;2BACA,EAAK,EAAS,SAAQ;;;;;MAQ7C,WAAc,GACZ,MAAO,GAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QAChD,EAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAoB,GAClB,GAAM,GAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,EAAG,KACnD,EAAW,EAAc,KAAK,KAC9B,EAAY,EAAc,MAAM,IAAI,KAAK,KAC/C,MAAO,mBAAmB,EAAQ,WAAW,EAAS,KAGxD,WAAoB,EAAW,GAC7B,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,MAAM,EAAU,GAAE,OAE9B,GAAG,EAAU,IAI5B,MAAA,MC7GA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAoB,EAAkB,EACtC,EAAqB,EAAmB,EACxC,GAPJ,KAAA,cAAgB,CAAC,UAAW,UAAW,gBAQrC,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAClC,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,QAElB,GAAM,GAAiB,cAAc,EAAa,IAE9C,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,gBAElB,GAAM,GAAiB,cAAc,EAAa,IAE5C,EAAe,EAAW,EAAI,aAAe,UACnD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,EAAO;;;YAG7C,EAAK;;;gCAGe,EAAU;;kCAER,EAAQ;kCACR,EAAc;0CACN,EAAY;;;uBAG/B,EAAc;;;;;;QAQrC,MAAA,MCvEA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,IAAK,cAOpB,GAAM,GAAa,EAAU,WACvB,EAAY,EAAU,UACtB,EAAS,EAAU,OACnB,EAAc,EAAU,YACxB,EAAU,EAAc,KAAK,KAAK,EAAS,GACjD,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAM,GAAsB,MACtB,EAAc,WAEd,EAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEvC,EAAgB;;MAIlB,EAAwB,GAC5B,AAAI,EAAS,EAAa,GACxB,GAAwB;oCACM,EAAM;;;SAMtC,GAAI,GAA4B,GAChC,AAAI,EAAS,EAAa,GACxB,GAA4B;oCACE,EAAM;;;SAMtC,KAAK,SAAW;0CACsB,EAAmB;;;UAGnD,EAAqB;;;;;UAKrB,EAAyB;;;;;;;;;YASvB,EAAW,cAAc,EAAU;wDACS,EAAW;;;;8BAIrC,EAAqB;;;;;;;;;;;;;;;;YAgBvC,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;;;;;;;;;;;;;;;;;YAiB/B,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCjKA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAe,EAAiB,GAJ5C,KAAA,cAAgB,CAAC,IAAK,IAAK,KAKzB,KAAK,YAAc,EAEnB,GAAI,GACA,EACJ,GAAI,EAAO,EACT,KAAM,OAAM,kBAAkB,EAAI,yBAGpC,GAAI,IAAS,EACX,EAAW,QACX,EAAU,aAKV,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAClD,EAAa,GACb,EAAc,GACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAY,KAAK,GAAG,EAAc,IAClC,AAAI,EAAI,GACN,EAAW,KAAK,GAAG,EAAc,IAGrC,EAAU,EAAW,OACrB,EAAW,EAAY,OAGzB,GAAM,GAAQ,GAAkB,GAEhC,KAAK,SAAW;;UAEV,EAAK;4BACa,EAAO;;2BAER,EAAQ;;2BAER,EAAQ;;;MAKnC,MAAA,MCjEA,AAqBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,UASf,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAc,qBAAqB,KAAK,KAAI,KAC5C,EAAe,GAAU,KAAK,MAEhC,EACE,EAAW,EAAS,IAAI,SAAC,EAAG,GAChC,MAAO,aAAa,GAAO,GAAE,YAAY,EAAC,cAAc,GAAO,GAAE,MAEnE,EAAO;UACD,EAAK;UACL,EAAK;UACL,EAAS,KAAK;GAAK;QAEzB,KAAK,SAAW;QACZ,EAAW;;UAET,EAAI;8BACgB,EAAY;;MAKxC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KAEM,GAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,YAAmB,GACjB,GAAI,IAAS,EACX,MAAO,YACF,GAAI,GAAQ,EACjB,MAAO,IAAO,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAA,aAAe,IAAG,KAAK,KAE7D,KAAM,OAAM,oBAAoB,EAAI,yBCpFxC,AAsBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,UACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAS,GAAY,SAAU,KAAK,MACpC,EAAY,GAAY,YAAa,KAAK,MAE1C,EACF,KAAK,OAAS,EAAI,YAAc,QAAQ,EAAU,MAAM,IAAI,OAAM,IAChE,EACF,wBAAwB,EAAU,OAAM,MAAM,EAAS,IACrD,EAAW;mBACF,EAAU;cACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;YACnB,EAAU,KAAK,KAAO,GAAE;;MAG1B,EAAW,KAAK,OAAS,EAAI,GAAK;UAClC,EAAO,KAAK,KAAO,GAAE;cACjB,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;gBACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;cACpD,EAAU,KAAK,KAAO,GAAE;uBACf,EAAU;;;MAKvB,EAAiB,KAAK,MAAQ,EAChC;cACM,EAAK,IAAI,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,SAAS,EAAC,MAAK,OAAM,KAC3D,EAAS,IAAI,SAAC,EAAG,GAAM,MAAG,GAAU,GAAE,MAAM,EAAO,GAAE,YAAY,EAAC,OAC7D,KAAK;GACd,KAAK,SAAW;0BACM,KAAK,KAAI;;UAEzB,EAAK;UACL,EAAK;UACL,EAAc;;UAEd,EAAQ;UACR,EAAQ;;;MAMhB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KCtGA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAiB,EAAmB,GAJhD,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAK,OACZ,EAAa,GAAkB,EAAK,QACpC,EAAQ,GAAkB,EAAK,QAEjC,EAAY,GAChB,GAAI,IAAS,EACX,EAAY,gCAEZ,GAAI,GAAa,EACjB,EACI,EAAK,IAAI,SAAC,EAAG,GACP,WACO,EAAK,SAAW,EACnB,oBAAoB,EAAC,aAAa,EAAC,IACnC,UAAU,GAAa,GAAC,eAAe,EAAC,aAAa,EAAC,MAE3D,KAAK,KAGhB,KAAK,SAAW;QACZ,EAAU,YAAY,EAAU,IAAI,EAAK;QACzC,EAAU,cAAc,EAAU,IAAI,EAAO;;;UAG3C,EAAK;yBACU,EAAS;;MAIlC,MAAA,MCxDA,AAuBA,GAAA,IAAA,WAUE,WAAoB,GAAA,KAAA,MAAA,EATZ,KAAA,gBAAkB,EAClB,KAAA,gBAAkB,EAClB,KAAA,mBAAqB,EACrB,KAAA,cAAgB,EAEhB,KAAA,aAAkD,GAClD,KAAA,WAAa,GACb,KAAA,aAAkD,GAI1D,SAAA,UAAA,eAAA,SACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAkC,EAAO,GAE3D,EAAW,GAAuB,EAAS,EAAiB,GAClE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAEhC,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAS,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cACpD,GAEJ,GAAI,KAAK,aAAa,GAAU,OAAS,GACvC,KAAK,kBACL,KAAK,kBACL,KAAK,eAAiB,EACtB,KAAK,MACL,GAAM,GAAa,KAAK,aAAa,GAAU,QAC/C,YAAK,aAAa,GAAU,KAAK,GAC1B,EAGT,GAAI,GACJ,MAAI,KAAoB,GAAoB,mBAC1C,EAAa,KAAK,MAAM,0BAA0B,EAAQ,GAAI,EAAQ,IACjE,AAAI,IAAoB,GAAoB,mBACjD,EACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,IAC/D,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AACH,IAAoB,GAAoB,0BAC1C,GACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,KAEtE,KAAK,aAAa,GAAU,KAAK,GAEjC,KAAK,kBACL,KAAK,oBAAsB,EAC3B,KAAK,MAEE,GAGT,EAAA,UAAA,eAAA,SACI,EAAuB,EACvB,EAA8B,GAChC,GAAI,KAAK,cAAgB,KAEvB,OAEF,GAAM,GACF,GAAkC,EAAgB,GAChD,EAAW,GAAuB,EAAO,EAAiB,GAChE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAO,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cAClD,GACE,EAAqB,EAAA,MAAM,IAAI,kCACrC,AAAI,IAAuB,IACvB,KAAK,mBAAqB,EAC5B,MAAK,MAAM,oBAAoB,GAC/B,KAAK,oBAAsB,GAE3B,MAAK,aAAa,GAAU,KAAK,GACjC,KAAK,kBACL,KAAK,eAAiB,GAGxB,KAAK,kBAEL,GAAM,GAAU,KAAK,aAAa,GAC5B,EAAW,EAAQ,QAAQ,GACjC,GAAI,EAAW,EACb,KAAM,IAAI,OACN,4EAGN,EAAQ,OAAO,EAAU,GACzB,KAAK,OAGC,EAAA,UAAA,IAAR,WACE,GAAI,CAAC,KAAK,WACR,OAEF,GAAM,GAAQ,KAAK,gBAAkB,KAAK,gBAC1C,QAAQ,IACJ,YAAgB,KAAK,gBAAe,MAAM,KAAK,gBAC/C,IAAI,EAAK,KACb,GAAM,GAAY,KAAK,cAAgB,KAAK,mBAC5C,QAAQ,IAAI,oBAAoB,KAAK,oBACrC,QAAQ,IAAI,iBAAiB,KAAK,cAAa,KAC3C,KAAK,MAAM,IAAM,GAAU,OAGjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGd,OAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,+CAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,cAAgB,KAEvB,OAEF,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,KAAK,aAAe,KACpB,KAAK,aAAe,KACpB,KAAK,gBAAkB,EACvB,KAAK,gBAAkB,EACvB,KAAK,mBAAqB,EAC1B,KAAK,cAAgB,GAEzB,KAEA,YACI,EAA2B,GAE7B,GAAM,GAAQ,EACd,GAAI,IAAmB,EAAM,KAC3B,MAAO,GACF,GAAI,IAAmB,EAAM,KAClC,MAAO,GACF,GAAI,IAAmB,EAAM,QAClC,MAAO,IACF,GAAI,IAAmB,EAAG,KAC/B,MAAO,IACF,GAAI,IAAmB,EAAM,QAClC,MAAO,GAET,KAAM,IAAI,OAAM,2BAA2B,eAIzC,EAAyB,EACzB,EAA2B,EAC3B,GAMF,GAAM,GACF,GAAiC,EAAiB,GAElD,EACJ,GAAI,GACI,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAEpB,EAAc,EAAc,OAGtB,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,EAAc,EAAQ,EAGxB,GAAM,GAAkB,GAA0B,EAAI,GACtD,MAAO,GAAc,EAGvB,YACI,EACA,GACF,OAAQ,OACD,IAAoB,mBACvB,MAAO,IAAwC,OAC5C,IAAoB,mBACvB,MAAO,IAA+C,OACnD,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,yBACvB,MAAO,IAA+C,WAEtD,KAAM,IAAI,OAAM,iCAAiC,IAIvD,YAAwC,GAEtC,MAAI,GAAA,MAAM,QAAQ,gCACZ,EACK,GAAoB,mBAEtB,GAAoB,iBAGzB,EACK,GAAoB,mBAEtB,GAAoB,iBAG7B,YACI,EAA8B,GAChC,GAAI,IAAmB,GAAa,OAClC,MAAO,IAAoB,mBACtB,GAAI,IAAmB,GAAa,QAAU,GAAkB,KACrE,MAAO,IAA+B,GACjC,GACH,IAAmB,GAAa,UAChC,IAAmB,GAAa,OAClC,MAAO,IAAoB,yBAE7B,KAAM,IAAI,OAAM,gCAAgC,GAGlD,YACI,EAAgC,EAChC,GACF,MAAU,GAAa,GAAE,IAAI,EAAa,GAAE,IAAI,EAAe,IAAI,ECxRrE,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,GAEpC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,GAErC,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,GACvB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,iBAAiB,EAAI,yBAEnC,GAAI,IAAS,EACX,MAAO,eAAe,EAAO,GAAE,IAMjC,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,UAAW,WAE7D,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAa,KAAK,QAAQ,EAAc,GAAE,KAAK,EAAO,GAAE,KAE1D,MAAO,GAAa,OC5DtB,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MAEM,GAAoB,0BAEb,GAAS,YAET,GAAM,iBAEN,GAAO,GAAoB;;EAI3B,GAAQ,GAAoB;;EAI5B,GAAM,0CAEN,GAAO;;;uBAGG,EAAA,aAAa,gBAAe;kBACjC,EAAA,aAAa,WAAU;;cAIpB,GAAA,MAAA,KAAA,QAAA,GAAA,GACZ,GAAoB;mCACM,EAAK;KAIjC,GAAM,IAAM,aAEN,GAAO,kBAEP,GAAQ,mBAER,GAAO;;;EAKP,GAAS,0BAET,GAAS,0BAET,GAAY,wCAEZ,GAAQ;;;;;;;;;;;;;;;EAiBR,GAAM,iBAEN,GAAQ,uBAER,GAAM;kBAGN,GAAQ,uBAER,GAAO,kBAEP,GAAQ,yBAER,GAAU,sCAeV,GAAW;;;;;;;;;;;;;;;;;;;;EAsBX,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;EAI3B,GAAO;;;EAKP,GAAO;;;EAKP,GAAO;;;EAKP,GAAQ,GAAoB,qCAE5B,GAAQ,GAAoB;;sCAI5B,GAAQ,GAAoB;;+CAI5B,GAAM;;;;cAIL,EAAA,aAAa,MAAK;eACjB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;;;;;;EAQrB,GAAa,kBAEb,GAAc,6BAEd,GAAS,wBAET,GAAQ,YC9NrB,AAmBO,GAAM,IAAS,YAET,GAAM;;;;;;;;;EAWN,GAAO;;;;;;;;;;EAYP,GAAQ;;;;;;;;;;EAYR,GAAM;;;;;;;;;EAWnB,GAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MCzFA,AAqBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEnB,EAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EAAe,GAAgB,EAAM,GACrC,EAAY,EAAS,MAAM,IAC3B,EAAS,GAAQ,EAAI,KAAO,QAAQ,EAAU,KAAK,KAAI,IAE7D,KAAK,SAAW;;UAEV,EAAK;kCACmB,EAAY;;4CAEF,EAAM;;MAIlD,MAAA,MC/CA,AA2BO,GAAA,IAAA,EAAA,aAAA,aACD,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAiFlB,GAAkB,KAClB,GAAkB,KA4BzB,GAAuE,eAE9C,GAC7B,MAAI,KAAgB,KAGpB,IAAa,GAAgB,IACtB,GAAa,GAGtB,YACI,EAAqC,GACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACnC,IAAe,SACjB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,OACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,MACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GAET,KAAM,IAAI,OAAM,cACZ,EAAU,oDAKhB,GAAM,IAA6B,IAK7B,GAAyB,IAC/B,cACE,MAAI,GAAA,MAAM,OAAO,QAAU,KAClB,KAED,EAAA,MAAM,OAAO,OAAO,OAAS,EAAA,MAAM,OAAO,OAAO,MACjD,OAAO,iBACX,GAAyB,KAAO,KAM/B,GAAM,IAA8B,mBAEL,GAAA,EAAA,GAkCpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GA/BM,EAAA,YAAc,GAAI,SAGlB,EAAA,gBAAkB,GAAI,SAGtB,EAAA,aAAe,GAAI,SACnB,EAAA,cAAgB,EAOhB,EAAA,aAAe,EAEf,EAAA,eAAiB,EAUjB,EAAA,kBAAoB,GACpB,EAAA,sBAAwB,GAgXxB,EAAA,eAAiB,EAq+DjB,EAAA,SAAW,GAj1Eb,CAAC,EAAA,MAAM,QAAQ,aACjB,KAAM,IAAI,OAAM,yCAGlB,GAAI,GAAS,MACX,GAAM,GAAK,GAAgB,EAAA,MAAM,UAAU,kBAC3C,EAAK,YAAc,GAAe,EAAA,MAAM,UAAU,kBAClD,EAAK,MAAQ,GAAI,IAAa,GAC9B,EAAK,OAAS,EAAG,OACjB,EAAK,oBAAsB,OAE3B,GAAK,MAAQ,EACb,EAAK,YAAc,GACnB,EAAK,oBAAsB,GAC3B,EAAK,OAAS,EAAM,GAAG,OAEzB,SAAK,eAAiB,GAAI,IAAe,EAAK,OAC9C,EAAK,mBAAqB,KAE1B,EAAK,QAAU,GAAI,GAAA,YAAY,EAAM,EAAA,YAGvC,SAAA,UAAA,WAAA,WACE,MAAO,MAAK,QAAQ,aACf,MAAK,WAAa,KAAK,WAAW,aAAe,GAClD,KAAK,gBAGX,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAK5C,GAJA,AAAI,GAAA,MAAM,QAAQ,mCACd,EAAA,MAAM,QAAQ,WAChB,KAAK,uBAAuB,GAE1B,IAAU,aAAe,GAAU,KACrC,KAAM,IAAI,OACN,yEAGN,GAAM,GAAS,GACf,YAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,IAC1D,GAIT,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,YAIV,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,aAIZ,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAK3D,GAHA,AAAI,EAAA,MAAM,QAAQ,UAChB,KAAK,uBAAuB,GAE1B,IAAU,YACZ,KAAM,IAAI,OACN,yEAGN,KAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,KAGnE,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAc,KAAK,QAAQ,IAAI,GAErC,EAAY,WAEZ,AAAI,EAAY,SAAW,GACzB,KAAK,YAAY,KAKvB,EAAA,UAAA,SAAA,SAAS,GACP,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,eAAgB,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SAKpD,GAAI,GAAS,MACX,GAAI,GAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEtC,GAAM,GACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,SAAS,EAAI,QAC/B,YAAK,8BAA8B,GAC5B,EAET,GAAI,GAAU,KACZ,MAAO,MAAK,qBAAqB,GAEnC,GAAI,IAAU,SACZ,MAAO,GAET,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GACJ,GAAI,IAAU,aACZ,GAAM,GAAa,EAAe,KAAK,WACjC,EAAa,EAAe,KAAK,WACvC,EAAS,EAAA,aAAa,uBAAuB,EAAY,OAEzD,GAAS,KAAK,qBAAqB,GAGrC,MAAI,IACF,MAAK,gBAAkB,EAAA,KAAK,MAAQ,GAE/B,KAAK,qBAAqB,EAAQ,IAGrC,EAAA,UAAA,KAAN,SAAW,6IACT,GAAI,KAAK,YAAY,IAAI,GACjB,SAAc,KAAK,YAAY,IAAI,GACzC,CAAA,EAAO,GAAI,SAAoB,SAAA,GAAW,MAAA,GAAY,KAAK,MAQ7D,GANM,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAyD,EAAO,OAAxD,EAAiD,EAAO,MAAjD,EAA0C,EAAO,MAA1C,EAAmC,EAAO,MAAnC,EAA4B,EAAO,eAAnB,EAAY,EAAO,SAKnE,GAAS,KACP,SAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEhC,EACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,KAAK,EAAI,QAC3B,KAAK,8BAA8B,GACnC,CAAA,EAAO,GAGT,GAAI,GAAU,KACZ,MAAA,CAAA,EAAO,KAAK,qBAAqB,IAGnC,GAAI,CAAC,EAAA,MAAM,QAAQ,iCACf,EAAA,MAAM,UAAU,mBAAqB,EACvC,KAAM,IAAI,OACN,sGAIF,GAAS,KAGb,AAAI,IAAU,aAAe,EAAA,MAAM,IAAI,2BAErC,GAAoB,KAAK,OAAO,GAC1B,EAAU,KAAK,QAAQ,IAAI,EAAkB,QAEnD,EAAS,GAAA,KAAK,OAAM,wBAAuB,MAAA,EAAA,CACvC,EAAQ,SAAO,OAAK,GAA0B,MAGpD,KAAK,YAAY,IAAI,EAAQ,IAEzB,IAAU,YAEZ,CAAA,EAAM,KAAK,MAAM,yBAFf,CAAA,EAAA,UAEF,EAAA,8BAKE,KAAU,YACD,CAAA,EAAM,QAAQ,IACrB,CAAC,EAAe,KAAK,OAAQ,EAAe,KAAK,UAFnD,CAAA,EAAA,UACI,SAAK,EAAA,OAEL,EAAa,EAAG,GAChB,EAAa,EAAG,GACtB,EAAO,EAAA,aAAa,uBAChB,EAA4B,gBAC3B,AAAI,GAAU,KACnB,EAAO,KAAK,qBAAqB,GAE3B,GAAO,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,MAAM,gCAAgC,EAAQ,qBAE5D,MAAI,IAAqB,MACvB,KAAK,8BAA8B,GAE/B,EAAY,KAAK,qBAAqB,EAAQ,GAE9C,EAAc,KAAK,YAAY,IAAI,GACzC,KAAK,YAAY,OAAO,GAGxB,EAAY,QAAQ,SAAA,GAAW,MAAA,GAAQ,KACvC,AAAI,KAAK,gBAAgB,IAAI,IAC3B,MAAK,gBAAgB,OAAO,GAC5B,KAAK,YAAY,GACjB,KAAK,kBAEP,CAAA,EAAO,SAGD,EAAA,UAAA,uBAAR,SAA+B,GAC7B,GAAI,GAAU,KACZ,OAEF,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,KACjC,GAAM,GAAM,EAAO,GACnB,GAAI,CAAC,GAA4B,GAC/B,KAAI,GAAA,MAAM,QAAQ,gCACV,MACF,aAAa,EAAG,iJAIhB,MAAM,aAAa,EAAG,4CAK1B,EAAA,UAAA,qBAAR,SAA6B,SACrB,EAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SACf,EAAO,EAAA,KAAK,cAAc,GAChC,GAAI,EAAA,MAAM,QAAQ,iCAChB,GAAM,GAAY,KAAK,OAAO,GACxB,EAAU,KAAK,QAAQ,IAAI,EAAU,QACrC,EAAO,GAAA,KAAK,OACA,gCAA+B,MAAA,EAAA,CAC5B,EAAQ,SAAO,OAAK,GAA0B,KACjD,SAAS,EAAG,GAE9B,YAAK,8BAA8B,GAE5B,EAGT,GAAM,GACF,EAAA,MAAM,QAAQ,eAAiB,IAAa,GAC1C,EACF,EAAyB,GAAwB,GAAS,EACxD,EAAU,EACZ,GAAI,IAAyB,GAC7B,GAAI,IAAmB,GACrB,EAAS,KAAK,gBAChB,EAAS,CAAC,CAAC,MAAO,EAAa,MAAK,EAAE,OAAM,IAAI,WAC9C,EAAU,KAAK,QAAQ,IAAI,EAAO,QAClC,EACF,KAAK,MACA,gDACG,EAAQ,QAAS,EAAQ,SAAS,GAAI,EAAQ,SAAS,IAC1D,SAAS,EAAG,GACrB,YAAK,8BAA8B,GAE5B,GAGH,EAAA,UAAA,KAAN,SAAW,qHACH,GAAkB,KAAK,aACvB,EAA+B,GAEjC,EAAgB,GACpB,AAAI,KAAK,oBAAsB,KAC7B,MAAK,mBAAqB,EAC1B,EAAgB,IAEhB,KAAK,aAAa,KAAK,GAEzB,KAAK,aAAe,EAEpB,IAGM,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,SACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OACpB,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,QACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OAE1B,KAAK,aAAe,EAEpB,AAAI,GACF,MAAK,mBAAqB,MAGtB,EAAuB,CAC3B,aAAc,KAAK,aACnB,eAAgB,KAAK,eACrB,SAAU,KACV,OAAQ,MAGN,EAAA,MAAM,UAAU,iDAAmD,EACpD,CAAA,EAAM,QAAQ,IAAI,IADjC,CAAA,EAAA,UACI,SAAW,EAAA,OAEjB,EAAI,SAAc,EAAA,KAAK,IAAI,GAC3B,EAAI,oBAAyB,WACzB,MAAA,GAAS,IAAI,SAAC,EAAG,GAAM,MAAC,CAAC,KAAM,EAA0B,GAAI,GAAI,KAC5D,IAAI,SAAA,GAAK,MAAG,GAAE,KAAI,KAAK,EAAE,KACzB,KAAK,oBAEd,EAAI,SAAc,CAChB,MAAO,8EAIX,YAAK,aAAe,EACpB,KAAK,eAAiB,EACtB,CAAA,EAAO,SAET,EAAA,UAAA,OAAA,WACE,MAAO,CACL,WAAY,GACZ,cAAe,KAAK,cACpB,uBAAwB,KAAK,eAAe,kBAC5C,kBAAmB,KAAK,eAAe,eAInC,EAAA,UAAA,WAAR,WACE,MAAI,GAAA,MAAM,UAAU,iDAAmD,EAC9D,KAAK,MAAM,aAEb,CAAC,QAAS,EAAA,KAAK,MAAO,MAAO,OAG9B,EAAA,UAAA,SAAR,SAAiB,GACf,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,MAAK,MAAM,WACJ,GAER,GAAwB,MAAQ,EAAA,KAAK,MAC/B,IAGK,EAAA,UAAA,aAAd,SAA2B,4EACzB,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,CAAA,EAAO,KAAK,MAAM,uBAAuB,IAErC,GAAa,EACnB,CAAA,EAAO,EAAW,MAAQ,EAAW,eAKvC,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,gBAAgB,IAAI,GAC3B,OAEF,GAAI,KAAK,YAAY,IAAI,IACvB,KAAK,gBAAgB,IAAI,GACzB,KAAK,iBACL,OAGF,GAAI,CAAC,KAAK,QAAQ,IAAI,GACpB,OAGF,KAAK,eAAe,GACb,GAAA,GAAA,KAAA,QAAA,IAAA,GAAA,eACP,AAAI,GAAkB,MACpB,GAAe,KAAK,UACpB,EAAe,KAAK,WAEtB,KAAK,QAAQ,OAAO,IAGd,EAAA,UAAA,eAAR,SAAuB,GACf,GAAA,GAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAE5C,EAAM,GAAS,EAAM,YAAc,EACnC,EAAW,KAAK,aAAa,IAAI,GACvC,AAAI,EAAW,EACb,KAAK,aAAa,IAAI,EAAK,EAAW,GAEtC,MAAK,aAAa,OAAO,GACzB,AAAI,GAAW,MACb,MAAK,eAAiB,KAAK,aAAa,EAAU,GAClD,KAAK,eAAe,eAAe,EAAS,EAAU,EAAO,KAGjE,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,QAAU,KAClB,EAAQ,SAAW,KACnB,EAAQ,SAAW,GACnB,EAAQ,MAAQ,MAGlB,EAAA,UAAA,WAAA,SAAW,GACT,YAAK,YAAY,GACV,KAAK,QAAQ,IAAI,GAAQ,SAOlC,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,MAAK,QAAQ,IAAI,IAGlB,EAAA,UAAA,cAAR,WACE,MAAK,GAAA,MAAM,QAAQ,qBAInB,CAAI,KAAK,YAAc,MACrB,MAAK,WAAa,EAAA,SAAS,YAAY,QAGlC,KAAK,YAPH,MAiBX,EAAA,UAAA,mBAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAa,KAAK,gBACxB,MAAI,CAAC,KAAK,uBAAyB,GAAc,MAC/C,SAAQ,KACJ,6NAKJ,KAAK,sBAAwB,IAGxB,GAAc,MACjB,EAAO,MACH,SAAA,GAAS,MAAA,GAAK,QAAQ,IAAI,EAAM,QAAQ,SAAW,MAC/C,EAAA,KAAK,cAAc,EAAM,OAAS,KAGhD,EAAA,UAAA,gBAAA,WACE,MAAO,MAAK,OAGd,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,GAAM,GAAS,KAAK,WAAW,EAAK,MAAO,aACrC,EAAa,KAAK,QAAQ,IAAI,EAAO,QAI3C,SAAW,eAAiB,CAC1B,KAAM,EAAA,SAAS,KAAK,EAAK,SACzB,KAAM,EAAA,SAAS,KAAK,EAAK,UAGpB,GAET,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAExC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAGxC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAO,EAAM,EAAE,MAChE,EAAE,OACN,MAAO,MAAK,WAAW,EAAM,EAAE,MAAO,GAGxC,GAAI,EAAA,KAAK,cAAc,KAAU,EAC/B,MAAO,GAAA,OAAO,GAAI,EAAM,EAAE,OAErB,GAAA,GAAA,KAAA,QAAA,IAAA,EAAA,QAAA,SACD,EAAc,EAAA,WAAW,iBAAiB,EAAE,MAAO,EAAO,GAChE,GAAI,GAAY,CAAC,GACf,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAmB,GACvB,GAAI,IAAa,GACf,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,GAEhD,YAAK,YAAY,EAAE,QACZ,KAAK,aAAa,EAAG,EAAO,IAG7B,EAAA,UAAA,aAAR,SAAqB,EAAW,EAAiB,GAC/C,GAAM,GAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAI,KAAK,WAAW,EAAM,EAAE,OAC5B,EAAa,KAAK,QAAQ,IAAI,EAAE,QAEtC,OAAO,OAAO,EAAY,GAC1B,EAAW,MAAQ,EACnB,EAAW,MAAQ,EAAE,MACrB,GAAI,GAAa,EAAA,WAAW,kBAAkB,EAAO,EAAE,SACvD,AAAI,EAAS,OAGX,IAAc,EAAS,MAAM,YAE/B,EAAW,MAAQ,CACjB,WAAU,EAEV,WAAY,EAAS,OAAS,EAAS,MAAM,YAAc,EAAE,QAI/D,GAAM,GAAW,KAAK,aAAa,IAAI,EAAW,MAAM,aAAe,EACvE,YAAK,aAAa,IAAI,EAAW,MAAM,WAAY,EAAW,GAEvD,GAGT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GAD1C,GAAA,GAAA,KAEQ,EAAS,KAAK,mBAChB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,aAAa,EAAG,EAAO,EAAK,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAO,GAAI,GAGpB,GAAM,GAAU,GAAI,IAAoB,EAAO,EAAS,GACxD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAqB,EAAE,MAAO,GAClC,GAAI,IAAe,EAAE,MAAO,GAChC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,GAAI,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KAChC,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KACtC,MAAO,GAAA,QAAQ,KAAK,OAAO,EAAO,GAAO,KAAK,OAAO,EAAO,IAG9D,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAEjB,GAAI,EAAQ,OAAS,EAAA,MAAM,UAAU,iCACnC,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,OAAO,EAAQ,MAAM,EAAG,GAAW,GACnD,EAAY,KAAK,OAAO,EAAQ,MAAM,GAAW,GACvD,MAAO,MAAK,OAAO,CAAC,EAAU,GAAY,GAE5C,GAAI,EAAA,MAAM,QAAQ,gCAAkC,EAAQ,GAAG,KAAO,GACpE,GAAM,GAAU,GAAI,IAAoB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACnE,MAAO,MAAK,cAAc,EAAS,GASrC,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACtD,EACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAAK,GAAI,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,OAC3D,EAAU,GAAI,IAAc,EAAU,IAAI,SAAA,GAAK,MAAA,GAAE,SACjD,EAAc,KAAK,cAAc,EAAS,GAChD,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAsB,GAAtB,GAAA,GAAA,KACQ,EAAS,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,IAAI,KACtE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAE/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAM,GAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAA,EAAA,MAAC,EAAA,EAAA,GAIP,GAAK,KAAgB,GAAK,IAAgB,IACtC,EAAY,IACd,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAE1B,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAG1B,GAAM,GAAM,IAAgB,EAAI,EAAI,EAAE,KAAK,EAAO,EAAW,GACvD,EAAO,IAAgB,EAAI,EAAI,EAC/B,EAAM,IAAgB,EAAI,EAAE,KAAK,EAAO,EAAG,GAAa,EAC9D,MAAO,MAAK,SAAS,EAAK,GAAK,IAAI,EAAM,IAG3C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,GAC5D,MAAO,MAAK,cAAwB,EAAS,CAAC,EAAG,GAAI,IAGvD,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE7C,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAA,EAAA,MAAC,EAAA,EAAA,GAED,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,EACxD,EAAS,EAAiB,GACxB,EAAuB,CAAC,EAAG,GACjC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAwB,EAAS,EAAQ,IAGvD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aACd,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OACrD,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OAErD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAExD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAE/C,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,EAGT,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,EAAE,QAG/C,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAQ,EAAM,EAAO,GACnD,GAAI,IAAW,EAAE,MAAO,EAAQ,EAAM,EAAO,GACjD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAM,GACF,GAAI,IAAe,EAAW,MAAO,EAAa,EAAM,EAAO,GACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAY,EAAa,KAG/D,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAI,EAAE,QAAU,UACd,GAAM,GAAO,KAAK,SAAS,EAAE,QACvB,EAAc,EAAK,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,KAC9C,EAAM,EAAA,OAAO,EAAE,MAAO,EAAE,MAAO,GACrC,MAAO,IAAK,EAAK,GAEnB,GAAM,GAAU,GAAI,IAAY,EAAE,MAAO,GACzC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAU,GACxC,GAAI,IAAW,EAAE,MAAO,EAAU,GACtC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAAlD,GAAA,GAAA,KACQ,EAAS,KAAK,mBAChB,CAAC,EAAG,GAAU,WAAM,MAAA,GAAK,WAAW,OAAO,EAAG,EAAS,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAQ,KAAM,GACzD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAEV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAU,EAAE,QAAQ,GAAW,GAC1B,QAAQ,GACR,MAAM,EAAkB,IAGtC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAGV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAAyB,GACzB,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,EAAE,IAAI,GAEhB,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EAAe,EAAA,aAAa,oBAC9B,EAAQ,MAAO,EAAY,EAAM,IAE/B,EAAW,EAAA,UACb,EAAQ,QAAQ,GAChB,GACJ,MAAO,GAAA,QAAQ,EAAU,IAGnB,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,KAAK,KAAK,EAAS,GAC7B,EAAa,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,QAAO,GACpD,EAAU,GAAI,IAAc,EAAY,GACxC,EAAS,KAAK,cAAwB,EAAS,CAAC,GAAI,GAE1D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,OAAO,EAAQ,EAAY,IAGjC,EAAA,UAAA,UAAR,SACI,EAAa,EACb,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACrB,AAAI,GAAgB,MAClB,GAAY,EAAa,MAAM,GAC/B,EAAS,EAAa,MAAM,IAE9B,GAAM,GAAa,EAAA,aAAa,yBAAyB,GACnD,EAAa,CACjB,WAAU,EACV,OAAM,EACN,UAAS,EACT,QAAS,KAAK,KAAK,EAAS,IAExB,EACF,GAAI,IAAiB,EAAY,EAAY,GAAgB,MAC3D,EAAS,CAAC,GAChB,AAAI,GAAgB,MAClB,EAAO,KAAK,GAEd,GAAM,GAAS,KAAK,cAAwB,EAAS,EAAQ,SAE7D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,UAAU,EAAG,EAAY,IAG/B,EAAA,UAAA,gBAAR,SACI,EAAW,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MACtC,GAAM,GAAU,GAAgB,KAAO,EAAa,MAAQ,EAAE,MACxD,EAAS,EAAQ,EAAQ,OAAS,GAClC,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,GAAI,IAChB,EAAS,EAAY,EAAY,GAAgB,MAC/C,EAAS,GAAgB,KAAO,CAAC,GAAK,CAAC,EAAG,GAC1C,EAAS,KAAK,cAAsB,EAAS,EAAQ,SAC3D,MAAI,GAAO,OAAS,EAAE,KACb,KAAK,gBAAgB,EAAG,EAAY,GAEtC,GAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,MAAO,GAAa,QAAQ,IAGtD,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAC/D,GAAI,EACF,MAAO,GAGH,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,OAAQ,GAAa,QAAQ,IAGvD,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAI,GAAO,EACL,EAAc,EAAA,aAAa,mBAAmB,CAAC,GAAO,EAAE,MAC1D,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,EAAA,UAAU,EAAG,GACzB,EAAO,EAAA,aAAa,iBAAiB,EAAG,EAAE,MAAM,IAGlD,GAAM,GACF,GAAa,gBAAgB,EAAU,MAAO,EAAM,GAClD,EAAS,EAAA,KAAK,cAAc,CAAC,EAAU,MAAM,KAC7C,EAAM,EAAU,KAAK,GAAI,GACzB,EAAc,EAAA,WAAc,EAAE,OAChC,EACA,KAAK,aACG,EAAK,qBAAsB,EAAY,EAAa,GACvD,QAAQ,GACjB,MAAI,IAAe,MACjB,GACI,EAAA,UAAU,EAAQ,EAAA,aAAa,uBAAuB,KAErD,GAGD,EAAA,UAAA,aAAR,SACI,EAAa,EAAiC,EAC9C,EAAiB,GACnB,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EACF,GAAa,8BAA8B,EAAQ,GACjD,EAAY,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,YAAW,GACvD,EAAU,GAAI,IAAiB,EAAW,GAC1C,EACF,KAAK,cAAwB,EAAS,CAAC,EAAG,GAAa,GAE3D,MAAI,GAAO,MAAM,KAAO,EACf,EAET,GAAa,EAAA,MAAM,EAAG,GAAa,KAAK,CAAC,EAAS,IAC3C,KAAK,aAAa,EAAQ,EAAW,EAAY,EAAO,KAGzD,EAAA,UAAA,gBAAR,SAAwB,EAAW,EAAc,GAE/C,GAAM,GAAO,CAAC,GAId,GAHA,EAAA,aAAa,2BACT,MAAQ,EAAW,OAAO,GAAG,cAAgB,EAAW,MAAM,GAAI,EAClE,EAAE,MACF,CAAC,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAQ,GAC7C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,UAAU,EAAK,GAAY,QAAQ,GAEjD,MAAO,MAAK,gBAAgB,EAAG,IAGjC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,GAAI,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,kDAAkD,GAAE,KAAO,GAAC,IAC5D,iBAAgB,IAMtB,OAJM,GAAO,EAAE,MAAM,GACjB,EAAS,EAGJ,EAAI,EAAG,GAAK,KAAK,KAAK,KAAK,KAAK,IAAS,EAAG,KACnD,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,GAAO,GAC5C,EAAc,EAAQ,mBAAmB,GACzC,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,GAAS,EAAO,MAAO,GAC7D,EAAW,UAIb,GAAI,GACF,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAW,GAChD,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,IACtC,EAAW,UAGb,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA2B,QAE9D,GAAM,GAAU,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAE,OACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA+B,QAElE,GAAM,GACF,GAAI,IAAgB,GAAwB,EAAE,MAAO,EAAE,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAClE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA0B,QAG7D,GAAM,GAAU,GAAI,IAAgB,GAAmB,EAAE,MAAO,EAAE,OAClE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA6B,QAGhE,GAAM,GAAU,GAAI,IAAgB,GAAsB,EAAE,MAAO,EAAE,OACrE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAAmC,QAE/C,GAAM,GACF,GAAI,IAAgB,GAA4B,EAAE,MAAO,EAAE,OAC/D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAiC,QAEpE,GAAM,GACF,GAAI,IAAgB,GAA0B,EAAE,MAAO,EAAE,OAC7D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAM,GAAU,GAAI,IAAc,EAAU,KAAM,EAAE,MAAO,EAAE,MAC7D,MAAO,MAAK,cACR,EAAS,CAAC,EAAW,EAAG,GAAI,EAAA,WAAW,EAAE,MAAO,EAAE,SAGxD,EAAA,UAAA,MAAA,SAAM,GACJ,EAAA,aAAa,KACT,yEAEJ,GAAM,GAAW,EAAU,WAC3B,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAM,GAAQ,EAAE,WAChB,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAK,GACL,EAAc,QACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAA6B,GAEzC,GAAM,GAAU,GAAI,IAAgB,EAAI,EAAE,MAAO,EAAE,OACnD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAG1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,GAErD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAG7C,EAAA,UAAA,cAAR,SAAsB,EAAe,EAAY,GAC/C,GAAM,GAAU,GAAI,IAAqB,EAAE,MAAO,GAClD,MAAO,MAAK,cAAsB,EAAS,CAAC,GAAI,IAG1C,EAAA,UAAA,eAAR,SACI,EAAe,EAAe,EAAY,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GACF,GAAI,IAAsB,EAAI,EAAE,MAAO,EAAE,MAAO,GACpD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAO7C,EAAA,UAAA,yBAAR,SAAiC,EAAW,EAAW,GAAvD,GAAA,GAAA,KACQ,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAA,4SAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAcP,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,GAMD,EAAA,UAAA,+BAAR,SACI,EAAuB,GACzB,MAAO,CACL,OAAQ,EAAY,OACpB,MAAO,EAAY,MACnB,MAAO,EAAc,QAIzB,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAIjB,GAAI,EAAQ,OAAS,EAAA,MAAM,IAAI,iCAC7B,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,KAAK,EAAQ,MAAM,EAAG,IACtC,EAAY,KAAK,KAAK,EAAQ,MAAM,IAC1C,MAAO,MAAK,KAAK,CAAC,EAAU,IAG9B,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAO,OAAO,SAAC,EAAI,GAAO,MAAA,GAAA,WAAW,EAAI,KAC1D,EAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAE5B,EAAc,EAAA,MAAM,QAAQ,cAC5B,EAAU,EACZ,GAAI,IAAkB,EAAQ,GAAG,MAAO,GACxC,GAAI,IAAY,EAAQ,GAAG,MAAO,GACtC,MAAO,MAAK,cAAiB,EAAS,EAAS,IAGjD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,GAAM,GAAc,EAAA,MAAM,QAAQ,gCAC5B,EAAU,EACZ,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OAC/C,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,MAAO,MAAK,cAAiB,EAAS,CAAC,EAAG,GAAI,IAGhD,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAY,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACnE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAe,EAAE,OAGhD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAG1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAChD,EAAI,KAAK,SAAS,EAAQ,EAAS,QAAQ,IAC3C,EAAI,KAAK,IAAI,GACb,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAI,EAAG,IAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAGtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAE3C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA2B,EAAE,MAAO,EAAM,OAC9C,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAM,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAEtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA6B,EAAG,MAAO,EAAE,OAC7C,GAAI,IAAgB,GAAsB,EAAG,MAAO,EAAE,OAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,UAG1C,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,GAAI,GACJ,AAAI,EAAA,MAAM,QAAQ,mBAChB,EAAU,GAAI,IAAkB,EAAE,OAElC,EAAU,GAAI,IAAY,EAAE,OAE9B,GAAM,GAAc,EAAQ,mBAAmB,EAAK,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,IAGhD,EAAA,UAAA,IAAA,SAAsB,GAEpB,GAAI,KAAK,mBAAmB,CAAC,KAAO,EAAE,QAAU,aAC9C,GAAM,GACF,GAAiB,KAAK,QAAQ,IAAI,EAAE,QAAQ,QAChD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAU,GAAI,IAAkB,EAAE,OAClC,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAG9D,MAAO,MAAK,cAAsB,EAAS,IAG7C,EAAA,UAAA,QAAA,SAA0B,GACxB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,GAAc,IAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,KAG9B,EAAA,UAAA,eAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAGF,GAAM,GAAS,EAAE,MACX,EAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAkB,EAAS,WAC3B,EAAc,EAAO,GAAK,EAAO,GAAK,EAAO,GAC7C,EAAmB,EAAS,YAC5B,EAAiB,EAAS,aAAe,eACzC,EAAa,GACb,EAAa,GAIb,EACD,KAAgB,GAAK,IAAqB,IAC3C,EAAkB,GAChB,EAAyB,EAAO,GAAK,IAAM,GAAK,CAAC,CAAC,EAAS,SAEjE,GAAI,GAA6B,CAAC,EAAA,MAAM,QAAQ,wBAC5C,CAAC,EAAA,MAAM,QAAQ,iCACf,CAAC,GACH,GAAM,GAAc,EAAiB,EAAO,GAAK,EAAO,GAAK,EAAO,GAC/B,EAAO,GAAK,EAAO,GAAK,EAAO,GAC9D,EAAY,EAAA,QAAQ,EAAG,CAAC,EAAG,EAAa,EAAS,aACjD,EACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAS,KAAK,iBAAiB,CACnC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAExB,MAAO,GAAA,QAAQ,EAAQ,EAAS,UAWlC,GAAM,GAAc,EAChB,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACrC,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACnC,EAAwB,CAC5B,OAAQ,EAAE,OACV,MAAO,CAAC,EAAG,EAAa,EAAS,YACjC,MAAO,EAAE,OAUL,EAAwB,EAAS,MACvC,EAAS,MAAQ,EAAS,MAAM,QAChC,EAAS,MAAM,EAAS,MAAM,OAAS,KACvC,EAAA,KAAK,OACD,GAAyB,EAAS,MAAO,EAAU,OACnD,WAAM,MAAA,kBAAkB,EAAS,MAAK,OAClC,EAAU,MAAK,gBACvB,GAAM,GACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAgB,KAAK,iBAAiB,CAC1C,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAElB,EAAuB,KAAK,QAAQ,IAAI,EAAc,QAC5D,SAAA,KAAK,OACD,EAAqB,SACrB,WAAM,MAAA,gDAEV,EAAS,MAAQ,EAGjB,EAAqB,MAAQ,EAAS,SAC/B,EAAA,SAAS,qBACL,EAAc,OAAQ,EAAS,SAAU,EAAc,QAI5D,EAAA,UAAA,iBAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAQA,GAAA,GAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,WAGI,EAAiB,IAAe,eAEhC,EAAY,EAAc,EAAe,EACzC,EAAU,EAAY,EACtB,EAAa,CAAC,EAAW,GACzB,EAAa,GACb,EAAa,GAEb,EAAY,EAAE,QAAQ,CAAC,IACvB,EAAQ,EAAO,QAAQ,CAAC,EAAG,EAAW,KAEtC,EACF,GAAI,IAAoB,EAAY,EAAU,MAAO,GACnD,EACF,KAAK,cAAwB,EAAe,CAAC,IAAY,QAAQ,CAC/D,EAAG,EAAW,GAAI,EAAW,KAG7B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAgB,GAAI,IACtB,EAAO,MAAO,CAAC,EAAG,EAAS,EAAS,aAAc,EAClD,EAAY,EAAS,EAAiB,GACpC,EAAuB,CAAC,EAAQ,GACtC,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEd,GAAM,GAAU,KAAK,cAAwB,EAAe,GAE5D,MAAI,GACK,EAAQ,QAAQ,CAAC,EAAG,EAAW,EAAU,EAAS,cAElD,EAAQ,QAAQ,CAAC,EAAG,EAAS,YAAa,EAAW,KAIhE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAEjD,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAM,MAAM,KAAO,EAC3D,MAAO,MAAK,iBACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAGjD,GAAM,GAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAS,KAC7D,EAAU,GAAI,IAChB,EAAU,EAAS,EAAiB,GAClC,EAAuB,CAAC,EAAO,GACrC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAc,EAAS,IAGrC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eAAe,EAAG,EAAQ,GAExC,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAM,KAAO,EACvD,MAAO,MAAK,iBAAiB,EAAG,EAAQ,GAE1C,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAExC,EAA0B,EAAA,MAAM,QAAQ,6BAC1C,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EAC7C,EAAkB,EACpB,GAA6B,EAAY,GACzC,KACE,EAAmB,CAAC,EAAO,GAE3B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KAC5D,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAGd,GAAI,GACJ,MAAI,GACF,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,IAGrC,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,KAGrC,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,6BACd,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EACjD,GAAU,GAAI,IAA6B,GACpC,KAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,GAAU,GAAI,IAAuB,GAC9B,KAAK,cAAc,EAAS,CAAC,EAAG,MAGzC,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAA+B,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAM,GAAU,GAAI,IAAgC,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAA,aAAa,WAAW,EAAG,EAAO,OAG3C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,KAAK,MAAM,EAAG,EAAO,GAAM,QAAQ,GAE9C,MAAO,IAGT,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GAA2B,GAAI,IAAyB,GAC9D,MAAO,MAAK,cAAc,EAA0B,CAAC,GAAK,EAAE,QAG9D,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAM,GAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,KAAK,cAAc,EAA2B,CAAC,IAC7C,EAA2B,GAAI,IAAyB,GACxD,EAAS,KAAK,cAChB,EAA0B,CAAC,EAAI,GAAqB,EAAE,OAC1D,SAAmB,UACZ,GAGT,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IACA,EAAE,MAAO,EAAW,EAAU,GAClC,GAAI,IAAsB,EAAE,MAAO,EAAW,EAAU,GAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,GAAM,GAAU,GAAI,IAA8B,EAAI,EAAG,GAEzD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,GAAI,IAChB,EAAE,MAAO,EAAW,EAAU,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GACF,GAAI,IAAoC,EAAI,EAAG,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAM,GAAQ,EAAa,EAAS,EAAA,QAAQ,GACtC,EAAY,EAAM,MAAM,GACxB,EAAc,EAAM,MAAM,GAC1B,EAAU,GAAI,IAAmB,EAAW,EAAa,GACzD,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAQ,QAAS,IAGvD,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAM,GAAU,GAAI,IAAc,EAAQ,KAAM,EAAO,EAAS,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAAI,IAAY,EAAE,MAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,GAAM,GAAU,GAAI,IAChB,EAAM,MAAO,EAAM,MAAO,EAAU,EAAQ,GAChD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAO,EAAO,GAAW,YAG/D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAE9D,GAAM,GAAY,EAAE,MAAM,GACpB,EAAe,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAE5D,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAe,IAAe,OAChC,CAAC,EAAW,EAAc,EAAa,GACvC,CAAC,EAAW,EAAa,EAAc,GAErC,EAAU,GAAI,IAAoB,EAAa,EAAW,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAG5C,EAAe,CAAC,EAAa,EAAW,GACxC,EAAiB,EAAQ,QAAQ,CAAC,EAAY,IAC9C,EAAW,EAAQ,QAAQ,CAAC,EAAY,IAE9C,GAAI,IAAe,EACjB,MAAO,GAAA,aAAa,cAAc,EAAA,OAAO,IAAK,GAEhD,GAAM,GAAe,EAAA,OAAO,GACtB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAe,KAAM,EAAS,KAAM,EAC3D,GACE,EACF,KAAK,cAAc,EAAS,CAAC,EAAU,EAAgB,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAGjC,EAAiB,GACjB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAc,KAAM,EAAa,KAAM,EAC9D,CAAC,EAAY,GAAI,GACf,EAAc,KAAK,cACrB,EAAS,CAAC,EAAc,EAAe,IAC3C,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGzB,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGjB,EAAA,UAAA,QAAR,SAAgB,EAAa,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAGxD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAU,KAAK,QAAQ,EAAM,GAAM,KAAK,EAAE,MAAM,GAAI,EAAE,MAAM,IAClE,SAAK,UACL,EAAK,UACE,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAGpC,EAAiB,EAAQ,QAAQ,CAAC,EAAW,IAC7C,EAAW,EAAE,QAAQ,CAAC,EAAE,KAAO,EAAW,IAC1C,EACF,GAAI,IAAgB,EAAW,EAAS,CAAC,EAAW,IAClD,EAAc,KAAK,cAAc,EAAS,CAAC,EAAU,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAG5C,GAFA,EAAQ,GAAS,EAAA,KAAK,WAAW,GAE7B,IAAU,UAEZ,GAAM,GAAS,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IAChE,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,WAEjD,GAAM,GAAU,GAAI,IAAY,EAAO,GACjC,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,GAAI,EAAO,KAIlD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAIhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAE,QAAU,SAAW,GAAK,EAAG,EAAE,QAG7D,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GAEpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGhD,EAAA,UAAA,eAAA,SAAe,EAAiB,EAAiB,GAE/C,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,YAAK,QAAQ,IAAI,GAAQ,MAAQ,KAC1B,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAGtB,EAAA,UAAA,WAAR,SACI,EAAiB,EAAiB,GAC7B,GAAA,GAAA,KAAA,eAAA,EAAA,EAAA,GAAA,OACP,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAGrD,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAU,GAAI,IAAc,EAAM,OACxC,MAAO,MAAK,gBAAgB,EAAS,CAAC,GAAQ,EAAM,QAG9C,EAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAU,GAAI,IAAY,EAAM,OAChC,EAA8B,GACpC,MAAO,MAAK,gBACR,EAAS,CAAC,GAAQ,EAAM,MAAO,KAC/B,IAGE,EAAA,UAAA,cAAR,SAAsB,EAAmB,GACvC,GAAM,GAAe,CACnB,GAAuB,EAAM,eAC1B,GAAuB,EAAM,QAE5B,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EAAiB,CACrB,GAAuB,WAAgB,GAAuB,IAG1D,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,KAAK,gBAChB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,QAG1D,EAAA,UAAA,OAAR,SAAe,GACb,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,MAClB,EACF,GAAwB,GACxB,EACJ,AAAI,EACF,EAAU,GAAI,IAA0B,GAExC,EAAU,GAAI,IAAoB,GAEpC,GAAM,GAAgC,GAChC,EAAM,KAAK,gBACb,EAAS,CAAC,CAAC,MAAO,EAAW,MAAK,EAAE,OAAM,IAAI,EAC9C,KAAwB,GAC5B,MAAO,CAAC,MAAK,EAAE,MAAK,EAAE,OAAQ,EAAI,SAGpC,EAAA,UAAA,gBAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAHJ,GAAA,GAAA,KAGI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,KAAK,eAAe,EAAQ,YAAa,GAClD,EAAU,KAAK,QAAQ,IAAI,EAAO,QAIxC,GAHA,AAAI,EAAQ,cACV,GAAQ,SAAW,IAEjB,EAAQ,mBAAqB,GAAuB,OACtD,GAAM,GAAa,GAA0B,EAAQ,aAKrD,EAAQ,SAAW,EAAW,IAAI,SAAA,GAAK,MAAA,GAAI,IAK7C,GAHA,AAAI,EAAQ,aAAe,MACzB,GAAQ,MAAQ,EAAQ,aAEtB,EAAA,KAAK,cAAc,EAAO,SAAW,EAGvC,SAAQ,OACJ,EAAA,KAAK,uBAAuB,EAAO,MAAoB,GACpD,EAGT,GAAM,GAA8B,GAC9B,EAA2B,EAAO,IAAI,SAAA,GAC1C,GAAI,EAAM,QAAU,YAClB,KAAM,IAAI,OACN,mIAKN,GAAI,GAAU,EAAK,QAAQ,IAAI,EAAM,QAErC,GAAI,EAAQ,SAAW,MACrB,GAAI,CAAC,EAAQ,cACT,EAAA,KAAK,cAAc,EAAM,QACrB,EAAA,MAAM,UAAU,6BAMtB,MAAO,CACL,MAAO,EAAM,MACb,QAAS,KACT,UAAW,GACX,cAAe,EAAQ,QAM3B,AAAI,EAAQ,cACV,GAAQ,SAAW,GACnB,EAAQ,MAAQ,EAAM,eAEf,CAAC,CAAC,EAAQ,WAAa,CAAC,CAAC,EAAQ,aAC1C,EAAQ,EAAQ,SAAW,EAAK,aAAa,GAClB,EAAK,WAAW,GAC3C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,gBAE/B,EAAQ,UACR,CAAC,GAAyB,EAAQ,MAAO,EAAM,QAQjD,GAAM,GAAa,EACb,EAAc,EAAM,MAE1B,EAAM,MAAQ,EAAQ,MACtB,EAAQ,EAAK,cAAc,EAAiB,GAC5C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,QAEjC,EAAW,MAAQ,EAGrB,SAAK,YAAY,EAAM,QAChB,CAAC,MAAO,EAAM,MAAO,QAAO,EAAE,UAAW,MAGlD,KAAK,YAAY,EAAO,QACxB,GAAM,GACW,CAAC,MAAO,EAAO,MAAO,QAAS,EAAS,UAAW,IAC9D,EAAM,GAAyB,EAAS,EAAY,GACpD,EAAS,KAAK,iBAAiB,EAAK,WACxC,MAAO,IACH,EAAK,MAAO,EAAS,EAAY,KAEjC,EAAoB,KAAK,cAAgB,KAC3C,EAgBJ,GAfA,AAAI,GACF,GAAQ,KAAK,cAGf,GACI,KAAK,MAAO,EAAQ,EAAY,EAAY,GAEhD,EAAc,QAAQ,SAAA,GAAQ,MAAA,GAAK,8BAA8B,KAEjE,AAAI,GACF,GAAQ,KAAK,SAAS,GACtB,KAAK,aAAa,KACd,CAAC,KAAM,EAAQ,YAAY,KAAM,MAAO,KAAK,aAAa,MAG5D,CAAC,EAAA,MAAM,QAAQ,wBAA0B,EAAQ,UACjD,IAAkC,IACpC,GAAM,GAAW,KAAK,aAAa,GACnC,YAAK,8BAA8B,GAC5B,EAET,MAAO,IAGT,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,EAAc,GAAe,EAAO,GAAG,MACvC,GAAM,GAAU,KAAK,gBACjB,EAAS,EAAQ,EAAa,EAC9B,GACJ,MAAO,GAAA,SAAS,qBACL,EAAQ,OAAQ,EAAQ,MAAO,EAAQ,QAG5C,EAAA,UAAA,iBAAR,SAAyB,EAAa,GAEpC,MAAM,KAAO,MAAK,aAChB,MAAK,YAAY,GAAO,KAEnB,KAAK,YAAY,IAG1B,EAAA,UAAA,kBAAA,WACE,MAAO,MAAK,gBAKd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAIF,GAAI,CAAC,EAAA,MAAM,QAAQ,YACjB,GAAM,GAAU,OAAO,KAAK,KAAK,aACjC,EAAQ,QAAQ,SAAA,GACd,EAAK,MAAM,cAAc,EAAK,YAAY,GAAK,cAC/C,MAAO,GAAK,YAAY,KAG5B,KAAK,eAAe,UACpB,AAAI,KAAK,QAAU,MACd,MAAQ,oBAAuB,aAC/B,KAAK,iBAAkB,mBAC1B,KAAK,OAAO,SAEZ,KAAK,OAAS,KAEhB,AAAI,KAAK,qBACP,MAAK,MAAM,QAAU,KACrB,KAAK,MAAM,WAEb,KAAK,SAAW,IAGlB,EAAA,UAAA,eAAA,WAAA,GAAA,GAAA,KACE,MAAI,MAAK,qBAAuB,MAC9B,MAAK,oBAAsB,EAAA,KAAK,WAC9B,GAAI,CAAC,EAAA,MAAM,IAAI,iCAGb,GAAM,GAAY,EAAA,MAAM,QAAQ,SAChC,EAAA,MAAM,IAAI,QAAS,IACnB,GAAM,GAAsB,EAAK,IAAI,EAAA,OAAO,OAAO,WAAW,GAG9D,GAFA,EAAA,MAAM,IAAI,QAAS,GAEf,EAAsB,EACxB,MAAO,IAGX,MAAO,OAGJ,KAAK,qBAGd,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAGlD,EAAA,UAAA,YAAR,SAAoB,SACZ,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAE7C,GAAI,GAAW,KAEb,OAEF,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GAAW,EAAQ,SAMvB,GALA,AAAI,GAAY,MACd,GAAW,GAA2C,EAAO,GAC7D,EAAQ,SAAW,GAGjB,GAAU,MACZ,GAAM,GAAY,GAAwB,GAEtC,EAAO,OACP,EAAQ,EAAS,GAAI,EAAS,EAAS,GACrC,EAAc,YAAkB,YAEtC,AAAI,EACF,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,EAAU,GAAI,IACV,EAAW,CAAC,EAAQ,GAAQ,IAEhC,EACI,GAAI,IAAoB,EAAW,CAAC,EAAQ,GAAQ,GAG1D,GAAM,GAAuB,KAAK,eAAe,CAAC,EAAQ,GAAQ,GAClE,AAAI,EACF,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEjB,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEnB,KAAK,MAAM,2BACP,KAAK,WAAW,EAAqB,QAAS,EAAO,EACrD,GAIJ,GAAM,GAAwB,GACxB,EAAsB,KAAK,gBAC7B,EAAS,CAAC,GAAuB,EAAO,KAAM,GAG5C,EAAgB,KAAK,QAAQ,IAAI,EAAoB,QAC3D,EAAQ,QAAU,EAAc,QAChC,EAAQ,SAAW,EAAc,SACjC,EAAQ,SAAW,EAAc,SACjC,EAAQ,MAAQ,EAAc,MAE9B,KAAK,8BAA8B,GACnC,KAAK,QAAQ,OAAO,EAAoB,QAGxC,EAAQ,OAAS,KACjB,AAAI,GACF,MAAK,cAAgB,EAAA,KAAK,MAAQ,QAGpC,GAAM,GAAa,KAAK,eAAe,EAAU,EAAO,EAAO,GAC/D,EAAQ,QAAU,IAId,EAAA,UAAA,qBAAR,SAA6B,EAAgB,GAE3C,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAEP,YAAK,eAAe,GAEpB,AAAI,GAAiB,MACnB,GAAQ,OAAS,GAAoB,EAAe,IAE/C,EAAQ,QAGT,EAAA,UAAA,eAAR,SACI,EAA4B,EAAuB,EACnD,GAEF,GADA,KAAK,eAAiB,KAAK,aAAa,EAAU,GAC9C,CAAC,KAAK,mBACN,KAAK,cAAgB,KAAK,mBAAqB,KAAO,MACxD,GAAM,GAAM,MAAK,cAAgB,KAAO,MAAM,QAAQ,GACtD,KAAK,kBAAoB,GACzB,QAAQ,KACJ,6BAA6B,EAAE,yCAGrC,MAAO,MAAK,eAAe,eAAe,EAAU,EAAS,IAGvD,EAAA,UAAA,aAAR,SAAqB,EAAyB,GAC5C,MAAO,GAAM,GAAK,EAAM,GAAK,EAAA,KAAK,gBAAgB,IAG5C,EAAA,UAAA,mBAAR,SACI,EAAsB,GACxB,GAAI,KAAK,mBAAmB,GAC1B,IACE,MAAO,WACA,GACP,GAAI,EAAA,MAAM,QAAQ,WAChB,KAAM,IAAI,OAAM,yBAItB,MAAO,OAEX,GAriFsC,EAAA,eAuiFtC,YACI,EAAiB,GACnB,GAAI,IAAU,WAAa,IAAU,YACnC,MAAO,GACF,GAAI,IAAU,SAAW,IAAU,QAGxC,OAFM,GAAU,IAAU,QAAW,GAAI,YAAW,EAAE,QACjB,GAAI,YAAW,EAAE,QAC7C,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,KAAK,MAAM,EAAE,IAE3B,MAAO,OAEP,MAAM,IAAI,OAAM,iBAAiB,GChwFrC,GAGM,IAAU,QCHhB,cAmCE,EAAA,MAAM,IAAI,2BAA4B,ICnCxC,AAuBA,AAAI,EAAA,YAAY,aACd,EAAA,gBAAgB,QAAS,WAAM,MAAA,IAAI,KAAoB,MAQ5C,IAAQ,CAAC,eAAc,ICzBvB,GAA0B,0BAE1B,GAA2B;;;EAK3B,GAAkC;;;;;cAWf,GAC9B,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EACD,EAAe,EACf,EAAU,GAAI,IAAe,EAAE,MAAO,GAC5C,MAAO,GAAa,gBAAgB,EAAS,CAAC,GAAI,EAAE,oBAepD,EAAmB,EACnB,EAA4C,GAG9C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAe,EACf,EAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,EAAiB,EAAE,MAAO,EAAE,MAC5B,CAAC,CAAC,GACN,GAAI,IAAgB,EAAW,EAAE,MAAO,EAAE,OACxC,EAAS,GAAS,EAAE,MACpB,EAAS,EAAa,gBAAgB,EAAS,CAAC,EAAG,GAAI,GAC7D,MAAO,IC3DX,AAsBA,GAAM,IAAQ,GAA2B;;EAInC,GAAe;;;IAIjB,GAAkC;;EAIzB,GAAkB,GAAiB,GAAO,IAE1C,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,QACb,WAAY,ICvCd,YAqBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,QACb,WAAY,ICjCd,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,WAG/C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBACvB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAyB,GAAI,IAAyB,GAC5D,MAAO,GAAQ,gBAAgB,EAAwB,CAAC,GAAK,EAAE,OAG1D,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,IC3Cd,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,YAAwB,GAOtB,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,MACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,MACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;yBAKK,EAAa;wBACd,EAAY;2DACuB,EAAe;;;MAK1E,MAAA,MC5DA,AAoBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,YACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,YACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;wBAEI,EAAa;uBACd,EAAY;;;;;;yDAMsB,EAAe;;;;MAMxE,MAAA,MChEA,AAuBO,GAAM,IAIM,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,EAAA,EAAG,EAAA,EAAA,KAAM,EAAA,EAAA,SAAU,EAAA,EAAA,OAAQ,EAAA,EAAA,MAElC,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGL,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAGpB,GAAM,GAAc,CAAC,EAAG,EAAM,GAE1B,EAAc,KAClB,AAAI,GAAU,MACZ,GAAc,EAAO,MACrB,EAAY,KAAK,IAGnB,GAAI,GAAa,KACjB,AAAI,GAAS,MACX,GAAa,EAAM,MACnB,EAAY,KAAK,IAGnB,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACJ,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACF,EACF,EAAQ,gBAAgB,EAAS,EAAa,EAAY,GAAG,OAEjE,MAAO,IAGI,GAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,QACb,WAAY,IC/Ed,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAuBA,GAAM,IAAM;;;;eAQN,GAAa;;;;;;;;;;;;;;;;;;EAoBN,GAAgB,GACzB,GAAK,GAAY,IAER,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,ICzDd,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAItB,GAAM,GAAa,EAAW,GAC9B,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;yBAKK,EAAU;;uCAEI,EAAU;;;;;;;MASjD,MAAA,MC5CA,AAuBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,MACD,EAAe,EAEf,EAAU,GAAI,IAAsB,EAAmB,OACvD,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KChCX,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;uDAMmC,EAAK,OAAO,EAAM;;wBAEjD,EAAK,UAAS;;;;;;;;;;;;;;MAgBtC,MAAA,MCrDA,AAoBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;;8BAeU,EAAK,OAAO,EAAM;4BACpB,EAAK,UAAS;;;;;;;;;;;;;;;;UAgBhC,EAAK,OAAM;;MAIrB,MAAA,MCnEA,AA0BO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,QACb,WAAY,IAGV,GAEJ,YAAoB,GAKX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACnB,EAAA,EAAA,OACE,EAAA,EAAA,YAED,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAOR,EAA6B,CAAC,EAAQ,GACtC,EAAW,CAAC,EAAQ,EAAO,GAEjC,AAAI,IAAW,IACb,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAGpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA+C,EAAG,EAAG,EAAO,GAChE,EAAS,GAAoB,QAG/B,GAAM,GAAkB,EAAQ,eAAe,EAAU,SAEzD,EAAQ,QAAQ,IAAI,EAAgB,QAAQ,MAAQ,GAAa,OACjE,EAAQ,MAAM,yBACV,EAAQ,WAAW,EAAgB,QAAS,GAChD,GAAM,GAAU,EAAA,MAAM,QAAQ,cAC1B,GAAI,IAAwB,GAC5B,GAAI,IAAkB,GACpB,EAAM,EAAQ,gBAAgB,EAAS,CAAC,GAAkB,SAChE,SAAQ,YAAY,EAAgB,QAC7B,EC/ET,AA0BA,YAA4B,GAI1B,OAFM,GAAS,GAER,EAAO,SAAW,GAAK,EAAO,EAAO,OAAS,GAAG,UAAY,IAClE,GAAM,GACF,EAAO,OAAS,EAAO,EAAO,OAAS,GAAG,QAAU,EAAQ,GAC1D,EAAa,EAAA,aAAa,yBAAyB,GACzD,EAAO,KAAK,CACV,OAAQ,EACR,WAAU,EACV,QAAS,KAAK,KAAK,EAAU,KAIjC,MAAO,eAIL,EAAe,EAAiB,EAChC,GAIF,OAHM,GAAkB,GAAmB,EAAE,OAEzC,EAAS,EACJ,EAAI,EAAG,EAAI,EAAgB,OAAQ,KACpC,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,WAAY,EAAA,EAAA,QAErB,EAAU,GAAI,IAChB,CAAC,WAAU,EAAE,OAAM,EAAE,UAAW,EAAE,MAAM,GAAI,QAAO,GAAG,GACpD,EAAiB,EACvB,EAAS,EAAQ,gBAAgB,EAAS,CAAC,GAAS,GAEpD,AAAI,EAAe,SAAW,EAAE,QAC9B,EAAQ,YAAY,EAAe,QAIvC,MAAO,GC/DT,YAwBI,EAAmB,EACnB,GACF,GAAM,GACF,CAAC,GAAY,EAAM,QAAM,OACrB,GAAY,EAAM,QACpB,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EACF,CAAC,GAAY,IAAW,OACpB,GAAY,IAEd,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,EAAQ,gBACnB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,OC3ClE,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MACD,EAAe,EAEf,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QAC5C,MAAI,GAAS,UAAY,CAAC,GAAc,EAAE,MAAO,IAC7C,CAAE,GAAS,UAAY,MAAQ,GAAc,EAAS,MAAO,IACxD,GAAc,EAAG,EAAQ,GAGlC,GAAa,OAAO,EAAE,QAEf,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,QAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICzDd,YAwBI,EAAe,EAAuB,EACtC,GACF,GAAM,GAAS,EAAA,KAAK,cAAc,GAC5B,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAY,EAAQ,EACpB,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAC,GAAG,MAAO,CAAC,MAAO,CAAC,EAAW,IAAU,QAAO,IAEhE,EAAU,GAAO,EAAe,EAAE,MAAO,MAAO,GAChD,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAU,MAAO,CAAC,MAAO,GAAW,QAAO,IAEpE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,ECvCT,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAEjC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAW,GAAkB,GAEnC,KAAK,SAAW;;QAEZ,EAAK;uBACU,EAAQ;;MAI/B,MAAA,MAEA,YAA2B,GACzB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,sBAAsB,EAAI,yBAKxC,OAHM,GACF,CAAC,UAAW,UAAW,UAAW,UAAW,UAAW,WACtD,EAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAe,EAAO,IAAM,EAAc,GAE5C,MAAO,GAAe,OCxDxB,AAqBA,GAAA,IAAA,WAQE,WAAY,EAAkB,GAP9B,KAAA,cAAgB,CAAC,KAIjB,KAAA,aAAe,GACf,KAAA,aAAe,GAIb,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAIjC,GAFA,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACpB,KAAK,KAAO,EACd,KAAM,OACF,6BAA6B,KAAK,KAAI,0BAM5C,OAJM,GAAQ,GAAkB,KAAK,MAE/B,EAAc,GAAe,KAAM,KAAK,MACxC,EAAgB,GAAI,OAAM,KAAK,MAC5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAc,EAAO,IAAM,EAAY,GAEzC,GAAM,GAAY,QAAQ,EAAc,MAAM,IAAI,OAAM,IAClD,EACF,KAAK,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAC3D,EAAO,mBAAmB,EAAc,OAAM,MAAM,EAAS,IAEnE,KAAK,SAAW;;QAEZ,EAAK;;oBAEO,EAAI;WACb,EAAU;sBACC,EAAI;;UAEhB,EAAY,KAAK,KAAO,GAAE;aACvB,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAAE;sBACjD,EAAI;aACb,EAAU;wBACC,EAAI;;;;;MAO5B,MAAA,MCvEA,YAyBI,EAAe,EAAgB,GACjC,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAuB,EAAE,MAAO,GACpC,GAAI,IAAiB,EAAE,MAAO,GAClC,MAAO,GAAQ,gBAAgB,EAAS,CAAC,GAAI,EAAE,OC7BjD,AA0BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAW,EAAA,KAAK,eAAe,EAAkB,EAAE,OACrD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACrD,EAAuB,GAAgB,KACvC,EAAqB,EAAa,mBAAmB,CAAC,IAExD,EAAW,EACf,GAAI,GACF,GAAI,GAKF,OAJM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAa,IAErC,GAAM,GACF,GAAiB,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAc,GAE7D,EAAW,EAAa,eAAe,EAAU,EAAE,OACnD,GAAM,GAAe,EAAa,QAAQ,IAAI,EAAS,QACvD,EAAa,OAAS,MAEtB,GAAW,GAAc,EAAG,EAAc,GAG5C,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAGpD,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGhB,EAAW,EACf,AAAI,GAEF,GAAW,EAAA,aAAa,qBAAqB,EAAa,IAG5D,GAAI,GACJ,GAAI,GACF,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAY,GACd,EAAQ,EAAA,KAAK,cAAc,GAAc,EAAU,EAAE,OAEzD,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAQ,EAAU,EAAa,EAAU,GAGjD,MAAI,IACF,EAAa,8BAA8B,GAGtC,IC9FX,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,EAAE,OAGjD,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAuBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAC3B,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,EAAQ,gBAAgB,EAAyB,CAAC,GAAI,EAAE,OAEtD,EAAyB,GAAI,IAAyB,GACtD,EAAS,EAAQ,gBACnB,EAAwB,CAAC,EAAI,GAAmB,EAAE,OACtD,SAAQ,8BAA8B,GAC/B,EAGF,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,ICrDd,YAuBI,EAAe,EACf,EACA,GACF,GAAI,GAAU,GAAI,IAAc,EAAU,MAAO,IAC3C,EAAa,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAEzD,EAAU,GAAI,IAAc,EAAU,MAAO,GAAM,GAAM,GACzD,GAAM,GAAc,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAC1D,MAAO,CAAC,EAAY,GC/BtB,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAe,EAErB,EAAA,KAAK,OACD,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,uDACF,EAAE,MAAM,OAAM,MACtB,GAAM,GAA8B,CAAC,EAAG,GACxC,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,GAET,EAAA,GAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAEf,MAAO,CAAC,EAAQ,KCjDpB,AAoBO,GAAM,IAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAG9B,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EAE1B,MAAO,GAAA,aAAa,wBAChB,EAAW,EAAY,EAAkB,EACzC,KC3CR,AAkBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC1C7B,AAmBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAqBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EACA,GAPJ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAOtB,GAAM,GAAc,EAAW,GACzB,EAAa,EAAW,GACxB,EAAY,KAAK,IAAI,GAAS,QAAQ,GACtC,EAAY,KAAK,IAAI,GAAS,QAAQ,GAC5C,KAAK,YAAc,EAEb,GAAA,GAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAgB,EAAQ,QAAQ,GAChC,EAAgB,EAAQ,QAAQ,GAElC,EAAc,GAClB,AAAI,MAAO,IAAc,SACvB,EAAc,uBAAuB,EAAU,QAAQ,GAAE,IAEzD,EAAc;2BACO,EAAU,KAAK,KAAI;8CAI1C,KAAK,SAAW;;;;;4CAKwB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;4CACpB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;iDACf,EAAa;iDACb,EAAa;YAClD,EAAW;uCACgB,EAAU,+BACzC,EAAW;;;;;MAOnB,MAAA,MCtEA,AAuBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAe,EAEf,EAAU,GAAI,IACf,EAAmB,MAAO,EAAS,EAAW,GAC7C,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KClCX,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAqBA,GAAM,IAAS,gBAEF,GAAmB,GAAgB,IAEnC,GAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IC5Bd,AAqBA,GAAM,IAAqB,4BAEd,GACT,GAAiB,GAAoB,IAE5B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,IC7Bd,AAqBA,GAAM,IAAM,iBAEC,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC5Bd,AAwBO,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,QACb,WAAY,SAAC,GAQX,OARY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACA,EAAA,EAAA,KACD,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAI,GACJ,GAAI,EAAa,mBAAmB,CAAC,KACnC,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QACtC,EAAS,EAAS,OAClB,EAAY,GAAa,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAE/D,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAc,EAAG,EAAM,GAE/B,MAAO,KCnDX,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAGpB,QAAQ,KACJ,YACA,8DACJ,GAAM,GAAS,EAAQ,SAAS,EAAE,QAC5B,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IChDd,AA4CA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,+RCxEjB,AAGA,GAAM,IAAU,QCHhB,GAsCa,IAAU,CACrB,YAAa,GAAA,aACb,mBAAoB,GAAA,YACpB,qBAAsB,GAAA,cACtB,YAAa,GAAA,aACb,cAAe,GAAA,eACf,iBAAkB,GAAA,kBAClB,KAAQ,yZC7CV,oBAAM,IAAK,KAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,GAAG,IAAI,EAAI,WAAY,GAChC,EAAO,GAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,GAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,GAAG,IAAI,EAAW,GAC5B,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,GAAG,IAAI,EAAU,GACtC,EAAoB,GAAG,IAAI,EAAS,GACpC,EAAc,GAAG,IAAI,EAAoB,GACzC,EAAS,GAAG,IAAI,EAAmB,GACnC,EAAO,GAAG,IAAI,EAAmB,GACjC,EAAkB,GAAG,IAAI,EAAQ,GACjC,EAAgB,GAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,IAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,IAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,GAAG,SAAS,KAAK,aAChC,KAAK,UAAY,GAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,GAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,GAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,GAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,GAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,GACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,GAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,IAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,kBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,oBAAM,IAAK,KAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,GAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,YAAc,KClDtB,kBAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,GAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,GAAQ,wBAA0B,KCvFlC,eACA,KAAM,IAAK,KACL,GAAW,KACX,GAAY,KACZ,GAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,GAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,GAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,GAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,GAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,GAAU,yBAAyB,GACtD,EAAkB,GAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,GAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,GAAS,YAAY,GAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,GAAS,WAAW,GACpC,GAAI,GAAO,GAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,GAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,GAAS,oBAAoB,EAAe,GACxD,EAAc,GAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,GAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,GAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,GAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,GAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,GAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,EAAgB,KAAM,GAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,EAAa,QAAS,EAAiB,KAAM,GAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,EAAkB,KAAK,UAAU,QAAQ,GAAG,OAAO,CAAC,EAAa,KACjE,EAAqB,EAAe,WAC1C,EAAe,UACf,KAAM,GAAc,EAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,EAAkB,KAAM,GAAsB,KAAK,aAAa,EAAa,EAAY,EAAgB,IACtH,EAAe,EAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,EAAmB,KAAM,GAAuB,KAAK,aAAa,EAAc,EAAa,GAC1G,EAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,GAAiC,GAC5C,IAAsB,EAAW,EAAkB,QACnD,GAAsB,EAAW,EAAmB,UAE/C,AAAI,EAAgC,EACzC,GAAsB,EAAW,EAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,EAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,GAAyB,KAAK,sBAAsB,EAAW,EAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,EAAoB,SAC1F,EAAY,EAAU,OAAO,GAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,GAAG,QAAQ,GACX,KAAM,GAAe,GAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,GAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,GAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,GAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,kBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,0CAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,oBAAM,IAAK,KACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QAAS,GAAQ,GAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,GAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,GAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,oBAAM,IAAK,KAEL,GAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,GAAG,QAAQ,WAAW,GAC/B,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,IAAO,KAAK,IAAO,IAAM,KAAM,IAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,GAAO,IAGhB,kBAA0B,GACxB,MAAK,IAAO,QAAQ,IAAO,OAAS,KAAM,IAAG,eAAe,EAAO,KAAK,OAAO,YACxE,GAAO,OAGhB,kBAAuB,EAAO,GAM5B,GALA,AAAI,GAAQ,EAAO,KAAK,IAAI,WAC1B,GAAQ,EAER,IAAS,EAEP,KAAU,EAAG,MAAO,IACxB,GAAI,GACJ,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,GAAG,IAAI,EAAQ,CAAC,MAC1B,GAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,IAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,GAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,IAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OACrB,EAAa,KAAK,MAAM,KAAK,IAAI,IAAM,IAAO,GAAK,GAAK,MAAS,IACvE,AAAI,EAAa,EAAO,KAAK,OAAO,eAClC,GAAI,OAAS,EAAK,IAAM,GAAM,SAAW,OACzC,EAAI,WAAa,GAEnB,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KC9DrB,oBAAM,IAAK,KAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,GAAG,KAAK,KACrB,KAAM,GAAS,GAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,IAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,GAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,GAAG,IAAI,EAAG,CAAC,QAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAY,GAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,oBAAM,IAAK,KAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,oBAAM,IAAK,KACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,eACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,oBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,kBAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,GAAQ,cAAgB,GAAQ,UAAU,OAC1C,GAAQ,QAAU,GAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,GAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,GAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,GAAQ,QAAQ,GAAa,GAAQ,QAAQ,KACnI,GAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,oBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,GAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,GAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,GAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,GAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,GAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,GAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,GAAQ,YAAc,KCnDtB,oBAAM,IAAY,KACZ,GAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,GAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,GAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,GAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,GAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,oBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,oBAAM,IAAK,KACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,GAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,GAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,GAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,GAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,GAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,GAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,IAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,GAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,IAAG,OAAS,EAAQ,GAAG,QAAQ,WAAW,GAEpE,GAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,IAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,GAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,GAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,GAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,GAAQ,kBAAoB,KCrH5B,oBAAM,IAAK,KACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,oBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,GAAQ,KAAO,GAAa,KAC5B,GAAQ,QAAU,GAAa,QAE/B,GAAQ,UAAY,GAAe,UACnC,GAAQ,oBAAsB,GAAe,oBAC7C,GAAQ,aAAe,GAAU,aACjC,GAAQ,QAAU,GAAU,QAC5B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,eAAiB,GAAK,eAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,kBAAoB,GAAK,kBACjC,GAAQ,UAAY,GAAK,YCnBzB,oBAAM,IAAK,KAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,GAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,SAAW,KCtEnB,oBAAM,IAAK,KACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,GAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,GAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,GAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,IAAG,KAAK,KACb,KAAM,GAAa,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,GAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,IAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,IAAG,KAAK,KACb,KAAM,GAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,IAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,GAAG,KAAK,IAAM,GAAG,IAAI,GAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,GAAG,KAAK,IAAM,GAAG,QAAQ,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,GAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,GAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,GAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,kBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,2BAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,KCzEtB,oBAAM,IAAK,KACL,GAAW,KACX,GAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,IAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,IAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,GAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,GAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,GAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,GAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,GAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,oBAAM,IAAK,KACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,GAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,IAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,uCAGA,GAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GAGT,SAAU,CACR,UAAW,sCAEX,UAAW,IACX,SAAU,GACV,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCAEX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,cAAe,GACf,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,GACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,GACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,6vEC/EjB,KAAM,IAAK,KACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,IAGJ,KAAM,IAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,GAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,GAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,GAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,GAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,IAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,GAAG,eAAiB,GAAO,SAC7B,IAAI,iCAAkC,GAAO,SAC7C,KAAM,IAAG,WAAW,GAAO,SAC3B,KAAM,IAAG,SAUX,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,GAAO,OACvF,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,IAAI,SAAW,CAAC,GAAO,KAAK,IAAO,IAAM,KAAM,IAAO,QAAQ,KACrG,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,OAAO,SAAW,CAAC,GAAO,QAAQ,IAAO,OAAS,KAAM,IAAO,WAAW,KACjH,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,QAAQ,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,KAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,QAAQ,cAAc,EAAO,GAAO,MAAQ,GAC/F,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAAQ,GAChG,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,GAAO,KAAK,SACd,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAChE,EAAK,KAAO,KAAK,MAAM,KAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,KACZ,KAAM,GAAW,GAAO,KAAK,IAAI,SAAW,GAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,IAAU,GACrH,EAAK,UAAY,KAAK,MAAM,KAAQ,GAEpC,EAAY,KACZ,KAAM,GAAc,GAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,IAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,KAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,aAAc,EAAQ,WACtB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,GAAG,SAAS,WAOd,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,QAAQ,OAAS,GACjB,QAAQ,SAAW,GACnB,QAAQ,OAAS,GACjB,QAAQ,OAAS,GACjB,QAAQ,SAAW,GACnB,QAAQ,OAAS,GACjB,QAAQ,QAAU,GAClB,QAAQ,SAAW,GACnB,QAAQ,GAAK,GACb,QAAQ,QAAU,GAAI",
   "names": []
 }
diff --git a/dist/human.esm-nobundle.js b/dist/human.esm-nobundle.js
index 8bcf4ad1..a72adea0 100644
--- a/dist/human.esm-nobundle.js
+++ b/dist/human.esm-nobundle.js
@@ -1,2 +1,2 @@
-var We=Object.defineProperty;var y=(e,t)=>()=>(t||(t={exports:{}},e(t.exports,t)),t.exports),B0=e=>We(e,"__esModule",{value:!0}),Ke=(e,t)=>{B0(e);for(var n in t)We(e,n,{get:t[n],enumerable:!0})};var Qe=y(se=>{const f=require("@tensorflow/tfjs"),Ye=6;function T0(e){const t={strides:[e/16,e/8],anchors:[2,6]},n=[];for(let o=0;o<t.strides.length;o++){const i=t.strides[o],s=Math.floor((e+i-1)/i),r=Math.floor((e+i-1)/i),a=t.anchors[o];for(let c=0;c<s;c++){const l=i*(c+.5);for(let h=0;h<r;h++){const u=i*(h+.5);for(let d=0;d<a;d++)n.push([u,l])}}}return n}const Ge=e=>{e.startEndTensor.dispose(),e.startPoint.dispose(),e.endPoint.dispose()},Xe=e=>({startEndTensor:e,startPoint:f.slice(e,[0,0],[-1,2]),endPoint:f.slice(e,[0,2],[-1,2])}),k0=(e,t)=>{const n=f.mul(e.startPoint,t),o=f.mul(e.endPoint,t),i=f.concat2d([n,o],1);return Xe(i)};function _0(e,t,n){const o=f.slice(e,[0,1],[-1,2]),i=f.add(o,t),s=f.slice(e,[0,3],[-1,2]),r=f.div(s,n),a=f.div(i,n),c=f.div(r,2),l=f.sub(a,c),h=f.add(a,c),u=f.mul(l,n),d=f.mul(h,n),m=1;return f.concat2d([u,d],m)}function z0(e,t){return f.tidy(()=>{const n=e.box?e.box:e;return k0(n,t).startEndTensor.squeeze()})}class Ve{constructor(e,t){this.blazeFaceModel=e,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=T0(t.detector.inputSize),this.anchors=f.tensor2d(this.anchorsData),this.inputSize=f.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(e){if(!e||e.isDisposedInternal||e.shape.length!==4||e.shape[1]<1||e.shape[2]<1)return null;const[t,n,o]=f.tidy(()=>{const c=e.resizeBilinear([this.width,this.height]),l=f.mul(f.sub(c.div(255),.5),2),h=this.blazeFaceModel.predict(l);let u;if(Array.isArray(h)){const x=h.sort((N,v)=>N.size-v.size),E=f.concat([x[0],x[2]],2),P=f.concat([x[1],x[3]],2),w=f.concat([P,E],1);u=w.squeeze(0)}else u=h.squeeze();const d=_0(u,this.anchors,this.inputSize),m=f.slice(u,[0,0],[-1,1]),p=f.sigmoid(m).squeeze();return[u,d,p]}),i=await f.image.nonMaxSuppressionAsync(n,o,this.maxFaces,this.iouThreshold,this.scoreThreshold),s=await i.array();i.dispose();let r=s.map(c=>f.slice(n,[c,0],[1,-1]));r=await Promise.all(r.map(async c=>{const l=await c.array();return c.dispose(),l}));const a=[];for(let c=0;c<r.length;c++){const l=r[c],h=f.tidy(()=>{const u=Xe(l),d=s[c],m=this.anchorsData[d],p=f.slice(t,[d,Ye-1],[1,-1]).squeeze().reshape([Ye,-1]),x=f.slice(o,[d],[1]);return{box:u,landmarks:p,probability:x,anchor:m}});a.push(h)}return n.dispose(),o.dispose(),t.dispose(),{boxes:a,scaleFactor:[e.shape[2]/this.width,e.shape[1]/this.height]}}async estimateFaces(e){const t=f.tidy(()=>(e instanceof f.Tensor||(e=f.browser.fromPixels(e)),e.toFloat().expandDims(0))),{boxes:n,scaleFactor:o}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(n.map(async i=>{const s=z0(i,o),[r,a,c]=await Promise.all([i.landmarks,s,i.probability].map(async p=>p.array())),l=i.anchor,[h,u]=o,d=r.map(p=>[(p[0]+l[0])*h,(p[1]+l[1])*u]),m={topLeft:a.slice(0,2),bottomRight:a.slice(2),landmarks:d,probability:c};return Ge(i.box),i.landmarks.dispose(),i.probability.dispose(),s.dispose(),m}))}}async function A0(e){const t=await f.loadGraphModel(e.detector.modelPath,{fromTFHub:e.detector.modelPath.includes("tfhub.dev")}),n=new Ve(t,e);return n}se.load=A0;se.BlazeFaceModel=Ve;se.disposeBox=Ge});var pe=y(me=>{me.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};me.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var Ze=y(K=>{const N0=require("@tensorflow/tfjs");function R0(e,t){const n=[e.startPoint[0]*t[0],e.startPoint[1]*t[1]],o=[e.endPoint[0]*t[0],e.endPoint[1]*t[1]];return{startPoint:n,endPoint:o}}K.scaleBoxCoordinates=R0;function fe(e){return[Math.abs(e.endPoint[0]-e.startPoint[0]),Math.abs(e.endPoint[1]-e.startPoint[1])]}K.getBoxSize=fe;function ge(e){return[e.startPoint[0]+(e.endPoint[0]-e.startPoint[0])/2,e.startPoint[1]+(e.endPoint[1]-e.startPoint[1])/2]}K.getBoxCenter=ge;function O0(e,t,n){const o=t.shape[1],i=t.shape[2],s=[[e.startPoint[1]/o,e.startPoint[0]/i,e.endPoint[1]/o,e.endPoint[0]/i]];return N0.image.cropAndResize(t,s,[0],n)}K.cutBoxFromImageAndResize=O0;function C0(e,t=1.5){const n=ge(e),o=fe(e),i=[t*o[0]/2,t*o[1]/2],s=[n[0]-i[0],n[1]-i[1]],r=[n[0]+i[0],n[1]+i[1]];return{startPoint:s,endPoint:r,landmarks:e.landmarks}}K.enlargeBox=C0;function v0(e){const t=ge(e),n=fe(e),o=Math.max(...n),i=o/2,s=[t[0]-i,t[1]-i],r=[t[0]+i,t[1]+i];return{startPoint:s,endPoint:r,landmarks:e.landmarks}}K.squarifyBox=v0});var nt=y(_=>{_.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function $e(e){return e-2*Math.PI*Math.floor((e+Math.PI)/(2*Math.PI))}_.normalizeRadians=$e;function D0(e,t){const n=Math.PI/2-Math.atan2(-(t[1]-e[1]),t[0]-e[0]);return $e(n)}_.computeRotation=D0;function F0(e){return e*180/Math.PI}_.radToDegrees=F0;function Je(e,t){return[[1,0,e],[0,1,t],[0,0,1]]}function V(e,t){let n=0;for(let o=0;o<e.length;o++)n+=e[o]*t[o];return n}_.dot=V;function et(e,t){const n=[];for(let o=0;o<e.length;o++)n.push(e[o][t]);return n}_.getColumnFrom2DArr=et;function tt(e,t){const n=[],o=e.length;for(let i=0;i<o;i++){n.push([]);for(let s=0;s<o;s++)n[i].push(V(e[i],et(t,s)))}return n}function L0(e,t){const n=Math.cos(e),o=Math.sin(e),i=[[n,-o,0],[o,n,0],[0,0,1]],s=Je(t[0],t[1]),r=tt(s,i),a=Je(-t[0],-t[1]);return tt(r,a)}_.buildRotationMatrix=L0;function H0(e){const t=[[e[0][0],e[1][0]],[e[0][1],e[1][1]]],n=[e[0][2],e[1][2]],o=[-V(t[0],n),-V(t[1],n)];return[t[0].concat(o[0]),t[1].concat(o[1]),[0,0,1]]}_.invertTransformMatrix=H0;function q0(e,t){return[V(e,t[0]),V(e,t[1])]}_.rotatePoint=q0;function j0(e,t){return Math.sqrt((e[0]-t[0])**2+(e[1]-t[1])**2)}_.xyDistanceBetweenPoints=j0});var st=y(ot=>{const D=require("@tensorflow/tfjs"),R=Ze(),F=pe(),L=nt(),U0=468,W0=.25,K0=13,Y0=[K0,F.MESH_ANNOTATIONS.midwayBetweenEyes[0]],G0=3,X0=2,V0=[G0,X0],be=F.MESH_ANNOTATIONS.leftEyeLower0,ye=[be[0],be[be.length-1]],xe=F.MESH_ANNOTATIONS.rightEyeLower0,we=[xe[0],xe[xe.length-1]],Q0=3,Z0=4,$0=71,Pe=76;function ie(e,t,n,o){for(let i=0;i<F.MESH_TO_IRIS_INDICES_MAP.length;i++){const{key:s,indices:r}=F.MESH_TO_IRIS_INDICES_MAP[i],a=F.MESH_ANNOTATIONS[`${n}${s}`],c=o==null;if(c||o.includes(s))for(let l=0;l<r.length;l++){const h=r[l];e[a[l]]=[t[h][0],t[h][1],(t[h][2]+e[a[l]][2])/2]}}}class J0{constructor(e,t,n,o){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=e,this.meshDetector=t,this.irisModel=n,this.meshWidth=o.mesh.inputSize,this.meshHeight=o.mesh.inputSize,this.irisSize=o.iris.inputSize,this.irisEnlarge=o.iris.enlargeFactor}transformRawCoords(e,t,n,o){const i=R.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),s=[i[0]/this.meshWidth,i[1]/this.meshHeight],r=e.map(d=>[s[0]*(d[0]-this.meshWidth/2),s[1]*(d[1]-this.meshHeight/2),d[2]]),a=L.buildRotationMatrix(n,[0,0]),c=r.map(d=>[...L.rotatePoint(d,a),d[2]]),l=L.invertTransformMatrix(o),h=[...R.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],u=[L.dot(h,l[0]),L.dot(h,l[1])];return c.map(d=>[d[0]+u[0],d[1]+u[1],d[2]])}getLeftToRightEyeDepthDifference(e){const t=e[ye[0]][2],n=e[we[0]][2];return t-n}getEyeBox(e,t,n,o,i=!1){const s=R.squarifyBox(R.enlargeBox(this.calculateLandmarksBoundingBox([e[n],e[o]]),this.irisEnlarge)),r=R.getBoxSize(s);let a=D.image.cropAndResize(t,[[s.startPoint[1]/this.meshHeight,s.startPoint[0]/this.meshWidth,s.endPoint[1]/this.meshHeight,s.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return i&&(a=D.image.flipLeftRight(a)),{box:s,boxSize:r,crop:a}}getEyeCoords(e,t,n,o=!1){const i=[];for(let s=0;s<Pe;s++){const r=e[s*3],a=e[s*3+1],c=e[s*3+2];i.push([(o?1-r/this.irisSize:r/this.irisSize)*n[0]+t.startPoint[0],a/this.irisSize*n[1]+t.startPoint[1],c])}return{rawCoords:i,iris:i.slice($0)}}getAdjustedIrisCoords(e,t,n){const o=e[F.MESH_ANNOTATIONS[`${n}EyeUpper0`][Q0]][2],i=e[F.MESH_ANNOTATIONS[`${n}EyeLower0`][Z0]][2],s=(o+i)/2;return t.map((r,a)=>{let c=s;return a===2?c=o:a===4&&(c=i),[r[0],r[1],c]})}async predict(e,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:o,scaleFactor:i}=await this.boundingBoxDetector.getBoundingBoxes(e);if(o.length===0)return this.regionsOfInterest=[],null;const s=o.map(r=>{const a={startPoint:r.box.startPoint.squeeze().arraySync(),endPoint:r.box.endPoint.squeeze().arraySync()};r.box.startPoint.dispose(),r.box.endPoint.dispose();const c=R.scaleBoxCoordinates(a,i),l=R.enlargeBox(c),h=r.landmarks.arraySync();return r.landmarks.dispose(),r.probability.dispose(),{...l,landmarks:h}});this.updateRegionsOfInterest(s),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const n=D.tidy(()=>this.regionsOfInterest.map((o,i)=>{let s=0;const r=o.landmarks.length>=U0;let[a,c]=Y0;r===!1&&([a,c]=V0),s=L.computeRotation(o.landmarks[a],o.landmarks[c]);const l=R.getBoxCenter({startPoint:o.startPoint,endPoint:o.endPoint}),h=[l[0]/e.shape[2],l[1]/e.shape[1]];let u=e,d=L.IDENTITY_MATRIX;s!==0&&(u=D.image.rotateWithOffset(e,s,0,h),d=L.buildRotationMatrix(-s,l));const m={startPoint:o.startPoint,endPoint:o.endPoint},p=R.cutBoxFromImageAndResize(m,u,[this.meshHeight,this.meshWidth]).div(255),[,x,E]=this.meshDetector.predict(p),P=D.reshape(E,[-1,3]);let w=P.arraySync();if(t.iris.enabled){const{box:X,boxSize:ue,crop:g0}=this.getEyeBox(w,p,ye[0],ye[1],!0),{box:b0,boxSize:y0,crop:x0}=this.getEyeBox(w,p,we[0],we[1]),Le=this.irisModel.predict(D.concat([g0,x0])),He=Le.dataSync();Le.dispose();const w0=He.slice(0,Pe*3),{rawCoords:qe,iris:P0}=this.getEyeCoords(w0,X,ue,!0),E0=He.slice(Pe*3),{rawCoords:je,iris:M0}=this.getEyeCoords(E0,b0,y0),Ue=this.getLeftToRightEyeDepthDifference(w);Math.abs(Ue)<30?(ie(w,qe,"left"),ie(w,je,"right")):Ue<1?ie(w,qe,"left",["EyeUpper0","EyeLower0"]):ie(w,je,"right",["EyeUpper0","EyeLower0"]);const I0=this.getAdjustedIrisCoords(w,P0,"left"),S0=this.getAdjustedIrisCoords(w,M0,"right");w=w.concat(I0).concat(S0)}const N=this.transformRawCoords(w,o,s,d);D.dispose(w);const v=R.enlargeBox(this.calculateLandmarksBoundingBox(N));if(t.mesh.enabled){const X=D.tensor2d(N);this.regionsOfInterest[i]={...v,landmarks:X.arraySync()};const ue={coords:X,box:v,confidence:x.squeeze(),image:p};return ue}const J={coords:null,box:v,confidence:x.squeeze(),image:p};return J}));return n}updateRegionsOfInterest(e){for(let t=0;t<e.length;t++){const n=e[t],o=this.regionsOfInterest[t];let i=0;if(o&&o.startPoint){const[s,r]=n.startPoint,[a,c]=n.endPoint,[l,h]=o.startPoint,[u,d]=o.endPoint,m=Math.max(s,l),p=Math.max(r,h),x=Math.min(a,u),E=Math.min(c,d),P=(x-m)*(E-p),w=(a-s)*(c-r),N=(u-l)*(d-r);i=P/(w+N-P)}i<W0&&(this.regionsOfInterest[t]=n)}this.regionsOfInterest=this.regionsOfInterest.slice(0,e.length)}clearRegionOfInterest(e){this.regionsOfInterest[e]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,e),...this.regionsOfInterest.slice(e+1)])}shouldUpdateRegionsOfInterest(){const e=this.regionsOfInterest.length,t=e===0;return this.maxFaces===1||t?t:e!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(e){const t=e.map(s=>s[0]),n=e.map(s=>s[1]),o=[Math.min(...t),Math.min(...n)],i=[Math.max(...t),Math.max(...n)];return{startPoint:o,endPoint:i}}}ot.Pipeline=J0});var rt=y(it=>{it.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var at=y(en=>{Ke(en,{default:()=>tn});var tn=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var lt=y(ee=>{const Y=require("@tensorflow/tfjs"),nn=Qe(),ct=pe(),on=st(),sn=rt(),rn=at().default;class dt{constructor(e,t,n,o){this.pipeline=new on.Pipeline(e,t,n,o),o&&(this.config=o)}async estimateFaces(e,t){t&&(this.config=t);const n=Y.tidy(()=>(e instanceof Y.Tensor||(e=Y.browser.fromPixels(e)),e.toFloat().expandDims(0))),o=await this.pipeline.predict(n,t);Y.dispose(n);const i=[];for(const s of o||[]){const r=s.confidence.arraySync();if(r>=this.config.detector.minConfidence){const a=s.coords?s.coords.arraySync():null,c={};if(a&&a.length>0)for(const l in ct.MESH_ANNOTATIONS)(this.config.iris.enabled||l.includes("Iris")===!1)&&(c[l]=ct.MESH_ANNOTATIONS[l].map(h=>a[h]));i.push({confidence:r||0,box:s.box?[s.box.startPoint[0],s.box.startPoint[1],s.box.endPoint[0]-s.box.startPoint[0],s.box.endPoint[1]-s.box.startPoint[1]]:0,mesh:a,annotations:c,image:s.image?Y.clone(s.image):null})}s.confidence.dispose(),s.image.dispose()}return i}}async function an(e){const t=await Promise.all([nn.load(e),Y.loadGraphModel(e.mesh.modelPath,{fromTFHub:e.mesh.modelPath.includes("tfhub.dev")}),Y.loadGraphModel(e.iris.modelPath,{fromTFHub:e.iris.modelPath.includes("tfhub.dev")})]),n=new dt(t[0],t[1],t[2],e);return n}ee.load=an;ee.MediaPipeFaceMesh=dt;ee.uv_coords=sn;ee.triangulation=rn});var ut=y(re=>{const T=require("@tensorflow/tfjs"),H={};let ht={age:0,gender:""},Ee=0;async function cn(e,t){const n=T.browser.fromPixels(e),o=T.image.resizeBilinear(n,[t,t]),i=T.cast(T.expandDims(o,0),"float32");return i}async function dn(e){return H.age||(H.age=await T.loadGraphModel(e.face.age.modelPath)),H.age}async function ln(e){return H.gender||(H.gender=await T.loadGraphModel(e.face.gender.modelPath)),H.gender}async function hn(e,t){if(Ee+=1,Ee>=t.face.age.skipFrames)return Ee=0,ht;let n;if(e instanceof T.Tensor){const i=T.image.resizeBilinear(e,[t.face.age.inputSize,t.face.age.inputSize],!1);n=T.mul(i,[255]),T.dispose(i)}else n=await cn(e,t.face.age.inputSize);const o={};if(t.face.age.enabled){const i=await H.age.predict(n),s=await i.data();o.age=Math.trunc(10*s[0])/10,T.dispose(i)}if(t.face.gender.enabled){const i=await H.gender.predict(n),s=await i.data();o.gender=Math.trunc(100*s[0])<50?"female":"male",T.dispose(i)}return T.dispose(n),ht=o,o}re.predict=hn;re.loadAge=dn;re.loadGender=ln});var ft=y(Me=>{const I=require("@tensorflow/tfjs"),un=["angry","discust","fear","happy","sad","surpise","neutral"],ae={};let mt=[],Ie=0;const pt=1.5;function mn(e,t){const n=I.tidy(()=>{const o=I.browser.fromPixels(e,1),i=I.image.resizeBilinear(o,[t,t]),s=I.cast(I.expandDims(i,0),"float32");return s});return n}async function pn(e){return ae.emotion||(ae.emotion=await I.loadGraphModel(e.face.emotion.modelPath)),ae.emotion}async function fn(e,t){if(Ie+=1,Ie>=t.face.emotion.skipFrames)return Ie=0,mt;const n=I.tidy(()=>{if(e instanceof I.Tensor){const i=I.image.resizeBilinear(e,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[s,r,a]=I.split(i,3,3);if(t.face.emotion.useGrayscale){const c=I.mul(s,[.2989]),l=I.mul(r,[.587]),h=I.mul(a,[.114]),u=I.addN([c,l,h]);return u}return r}return mn(e,t.face.emotion.inputSize)}),o=[];if(t.face.emotion.enabled){const i=await ae.emotion.predict(n),s=await i.data();for(let r=0;r<s.length;r++)pt*s[r]>t.face.emotion.minConfidence&&o.push({score:Math.min(.99,Math.trunc(100*pt*s[r])/100),emotion:un[r]});o.sort((r,a)=>a.score-r.score),I.dispose(i)}return I.dispose(n),mt=o,o}Me.predict=fn;Me.load=pn});var yt=y(gt=>{const bt=require("@tensorflow/tfjs");class gn{constructor(e,t){this.model=e,this.outputStride=t;const n=this.model.inputs[0].shape;bt.util.assert(n[1]===-1&&n[2]===-1,()=>`Input shape [${n[1]}, ${n[2]}] must both be equal to or -1`)}predict(e){return bt.tidy(()=>{const t=this.preprocessInput(e.toFloat()),n=t.expandDims(0),o=this.model.predict(n),i=o.map(r=>r.squeeze([0])),s=this.nameOutputResults(i);return{heatmapScores:s.heatmap.sigmoid(),offsets:s.offsets,displacementFwd:s.displacementFwd,displacementBwd:s.displacementBwd}})}dispose(){this.model.dispose()}}gt.BaseModel=gn});var Se=y(xt=>{const wt=require("@tensorflow/tfjs"),bn=yt();class yn extends bn.BaseModel{preprocessInput(e){return wt.tidy(()=>wt.div(e,127.5).sub(1))}nameOutputResults(e){const[t,n,o,i]=e;return{offsets:t,heatmap:n,displacementFwd:o,displacementBwd:i}}}xt.MobileNet=yn});var Et=y(Pt=>{function Be(e){return Math.floor(e/2)}class xn{constructor(e,t){this.priorityQueue=new Array(e),this.numberOfElements=-1,this.getElementValue=t}enqueue(e){this.priorityQueue[++this.numberOfElements]=e,this.swim(this.numberOfElements)}dequeue(){const e=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,e}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(e){for(;e>0&&this.less(Be(e),e);)this.exchange(e,Be(e)),e=Be(e)}sink(e){for(;2*e<=this.numberOfElements;){let t=2*e;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(e,t))break;this.exchange(e,t),e=t}}getValueAt(e){return this.getElementValue(this.priorityQueue[e])}less(e,t){return this.getValueAt(e)<this.getValueAt(t)}exchange(e,t){const n=this.priorityQueue[e];this.priorityQueue[e]=this.priorityQueue[t],this.priorityQueue[t]=n}}Pt.MaxHeap=xn});var It=y(Mt=>{const wn=Et();function Pn(e,t,n,o,i,s){const[r,a]=s.shape;let c=!0;const l=Math.max(n-i,0),h=Math.min(n+i+1,r);for(let u=l;u<h;++u){const d=Math.max(o-i,0),m=Math.min(o+i+1,a);for(let p=d;p<m;++p)if(s.get(u,p,e)>t){c=!1;break}if(!c)break}return c}function En(e,t,n){const[o,i,s]=n.shape,r=new wn.MaxHeap(o*i*s,({score:a})=>a);for(let a=0;a<o;++a)for(let c=0;c<i;++c)for(let l=0;l<s;++l){const h=n.get(a,c,l);if(h<e)continue;Pn(l,h,a,c,t,n)&&r.enqueue({score:h,part:{heatmapY:a,heatmapX:c,id:l}})}return r}Mt.buildPartWithScoreQueue=En});var te=y(z=>{z.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];z.NUM_KEYPOINTS=z.partNames.length;z.partIds=z.partNames.reduce((e,t,n)=>(e[t]=n,e),{});const Mn=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];z.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];z.connectedPartIndices=Mn.map(([e,t])=>[z.partIds[e],z.partIds[t]]);z.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var ke=y(q=>{const In=te();function St(e,t,n,o){return{y:o.get(e,t,n),x:o.get(e,t,n+In.NUM_KEYPOINTS)}}q.getOffsetPoint=St;function Sn(e,t,n){const{heatmapY:o,heatmapX:i,id:s}=e,{y:r,x:a}=St(o,i,s,n);return{x:e.heatmapX*t+a,y:e.heatmapY*t+r}}q.getImageCoords=Sn;function Bn(e,t){const n=new Array(t);for(let o=0;o<t;o++)n[o]=e;return n}q.fillArray=Bn;function Te(e,t,n){return e<t?t:e>n?n:e}q.clamp=Te;function Tn(e,t,n,o){const i=n-e,s=o-t;return i*i+s*s}q.squaredDistance=Tn;function kn(e,t){return{x:e.x+t.x,y:e.y+t.y}}q.addVectors=kn;function _n(e,t,n){return{y:Te(e.y,t,n),x:Te(e.x,t,n)}}q.clampVector=_n});var zt=y(Bt=>{const ne=te(),Q=ke(),Tt=ne.poseChain.map(([e,t])=>[ne.partIds[e],ne.partIds[t]]),_e=Tt.map(([,e])=>e),kt=Tt.map(([e])=>e);function zn(e,t,n){const o=n.shape[2]/2;return{y:n.get(t.y,t.x,e),x:n.get(t.y,t.x,o+e)}}function ze(e,t,n,o){return{y:Q.clamp(Math.round(e.y/t),0,n-1),x:Q.clamp(Math.round(e.x/t),0,o-1)}}function _t(e,t,n,o,i,s,r,a=2){const[c,l]=o.shape,h=ze(t.position,s,c,l),u=zn(e,h,r),d=Q.addVectors(t.position,u);let m=d;for(let E=0;E<a;E++){const P=ze(m,s,c,l),w=Q.getOffsetPoint(P.y,P.x,n,i);m=Q.addVectors({x:P.x*s,y:P.y*s},{x:w.x,y:w.y})}const p=ze(m,s,c,l),x=o.get(p.y,p.x,n);return{position:m,part:ne.partNames[n],score:x}}function An(e,t,n,o,i,s){const r=t.shape[2],a=_e.length,c=new Array(r),{part:l,score:h}=e,u=Q.getImageCoords(l,o,n);c[l.id]={score:h,part:ne.partNames[l.id],position:u};for(let d=a-1;d>=0;--d){const m=_e[d],p=kt[d];c[m]&&!c[p]&&(c[p]=_t(d,c[m],p,t,n,o,s))}for(let d=0;d<a;++d){const m=kt[d],p=_e[d];c[m]&&!c[p]&&(c[p]=_t(d,c[m],p,t,n,o,i))}return c}Bt.decodePose=An});var Ae=y(At=>{const Nn=It(),Rn=zt(),Nt=ke();function Rt(e,t,{x:n,y:o},i){return e.some(({keypoints:s})=>{const r=s[i].position;return Nt.squaredDistance(o,n,r.y,r.x)<=t})}function On(e,t,n){const o=n.reduce((i,{position:s,score:r},a)=>(Rt(e,t,s,a)||(i+=r),i),0);return o/n.length}const Cn=1;function vn(e,t,n,o,i,s,r=.5,a=20){const c=[],l=Nn.buildPartWithScoreQueue(r,Cn,e),h=a*a;for(;c.length<s&&!l.empty();){const u=l.dequeue(),d=Nt.getImageCoords(u.part,i,t);if(Rt(c,h,d,u.part.id))continue;const m=Rn.decodePose(u,e,t,i,n,o),p=On(c,h,m);c.push({keypoints:m,score:p})}return c}At.decodeMultiplePoses=vn});var Re=y(k=>{const Z=require("@tensorflow/tfjs"),Dn=te();function Fn(e,t,n){return e<n||t<n}function Ln(e,t){return Dn.connectedPartIndices.reduce((n,[o,i])=>(Fn(e[o].score,e[i].score,t)||n.push([e[o],e[i]]),n),[])}k.getAdjacentKeyPoints=Ln;const{NEGATIVE_INFINITY:Ot,POSITIVE_INFINITY:Ct}=Number;function vt(e){return e.reduce(({maxX:t,maxY:n,minX:o,minY:i},{position:{x:s,y:r}})=>({maxX:Math.max(t,s),maxY:Math.max(n,r),minX:Math.min(o,s),minY:Math.min(i,r)}),{maxX:Ot,maxY:Ot,minX:Ct,minY:Ct})}k.getBoundingBox=vt;function Hn(e){const{minX:t,minY:n,maxX:o,maxY:i}=vt(e);return[{x:t,y:n},{x:o,y:n},{x:o,y:i},{x:t,y:i}]}k.getBoundingBoxPoints=Hn;async function qn(e){return Promise.all(e.map(t=>t.buffer()))}k.toTensorBuffers3D=qn;function Dt(e,t,n,o=0,i=0){return{score:e.score,keypoints:e.keypoints.map(({score:s,part:r,position:a})=>({score:s,part:r,position:{x:a.x*n+i,y:a.y*t+o}}))}}k.scalePose=Dt;function Ft(e,t,n,o=0,i=0){return n===1&&t===1&&o===0&&i===0?e:e.map(s=>Dt(s,t,n,o,i))}k.scalePoses=Ft;function Lt(e){return e instanceof Z.Tensor?[e.shape[0],e.shape[1]]:[e.height,e.width]}k.getInputTensorDimensions=Lt;function Ne(e){return e instanceof Z.Tensor?e:Z.browser.fromPixels(e)}k.toInputTensor=Ne;function jn(e,t,n){return Z.tidy(()=>{const o=Ne(e);return o.resizeBilinear([t,n])})}k.toResizedInputTensor=jn;function Un(e,[t,n]){const[o,i]=Lt(e),s=n/t,r=i/o;let[a,c,l,h]=[0,0,0,0];r<s?(a=0,c=0,l=Math.round(.5*(s*o-i)),h=Math.round(.5*(s*o-i))):(a=Math.round(.5*(1/s*i-o)),c=Math.round(.5*(1/s*i-o)),l=0,h=0);const u=Z.tidy(()=>{let d=Ne(e);return d=Z.pad3d(d,[[a,c],[l,h],[0,0]]),d.resizeBilinear([t,n])});return{resized:u,padding:{top:a,left:l,right:h,bottom:c}}}k.padAndResizeTo=Un;function Wn(e,[t,n],[o,i],s){const r=(t+s.top+s.bottom)/o,a=(n+s.left+s.right)/i,c=Ft(e,r,a,-s.top,-s.left);return c}k.scaleAndFlipPoses=Wn});var qt=y(Oe=>{const Kn=require("@tensorflow/tfjs"),Yn=Se(),Gn=Ae(),ce=Re();class Ht{constructor(e){this.baseModel=e}async estimatePoses(e,t){const n=t.outputStride,[o,i]=ce.getInputTensorDimensions(e),{resized:s,padding:r}=ce.padAndResizeTo(e,[t.inputResolution,t.inputResolution]),{heatmapScores:a,offsets:c,displacementFwd:l,displacementBwd:h}=this.baseModel.predict(s),u=await ce.toTensorBuffers3D([a,c,l,h]),d=u[0],m=u[1],p=u[2],x=u[3],E=await Gn.decodeMultiplePoses(d,m,p,x,n,t.maxDetections,t.scoreThreshold,t.nmsRadius),P=ce.scaleAndFlipPoses(E,[o,i],[t.inputResolution,t.inputResolution],r);return a.dispose(),c.dispose(),l.dispose(),h.dispose(),s.dispose(),P}dispose(){this.baseModel.dispose()}}Oe.PoseNet=Ht;async function Xn(e){const t=await Kn.loadGraphModel(e.modelPath),n=new Yn.MobileNet(t,e.outputStride);return new Ht(n)}async function Vn(e){return Xn(e)}Oe.load=Vn});var Ut=y(B=>{const Qn=Se(),jt=qt(),Zn=Ae(),de=te(),oe=Re();B.load=jt.load;B.PoseNet=jt.PoseNet;B.MobileNet=Qn.MobileNet;B.decodeMultiplePoses=Zn.decodeMultiplePoses;B.partChannels=de.partChannels;B.partIds=de.partIds;B.partNames=de.partNames;B.poseChain=de.poseChain;B.getAdjacentKeyPoints=oe.getAdjacentKeyPoints;B.getBoundingBox=oe.getBoundingBox;B.getBoundingBoxPoints=oe.getBoundingBoxPoints;B.scaleAndFlipPoses=oe.scaleAndFlipPoses;B.scalePose=oe.scalePose});var De=y(j=>{const $n=require("@tensorflow/tfjs");function Ce(e){return[Math.abs(e.endPoint[0]-e.startPoint[0]),Math.abs(e.endPoint[1]-e.startPoint[1])]}j.getBoxSize=Ce;function ve(e){return[e.startPoint[0]+(e.endPoint[0]-e.startPoint[0])/2,e.startPoint[1]+(e.endPoint[1]-e.startPoint[1])/2]}j.getBoxCenter=ve;function Jn(e,t,n){const o=t.shape[1],i=t.shape[2],s=[[e.startPoint[1]/o,e.startPoint[0]/i,e.endPoint[1]/o,e.endPoint[0]/i]];return $n.image.cropAndResize(t,s,[0],n)}j.cutBoxFromImageAndResize=Jn;function eo(e,t){const n=[e.startPoint[0]*t[0],e.startPoint[1]*t[1]],o=[e.endPoint[0]*t[0],e.endPoint[1]*t[1]],i=e.palmLandmarks.map(s=>{const r=[s[0]*t[0],s[1]*t[1]];return r});return{startPoint:n,endPoint:o,palmLandmarks:i}}j.scaleBoxCoordinates=eo;function to(e,t=1.5){const n=ve(e),o=Ce(e),i=[t*o[0]/2,t*o[1]/2],s=[n[0]-i[0],n[1]-i[1]],r=[n[0]+i[0],n[1]+i[1]];return{startPoint:s,endPoint:r,palmLandmarks:e.palmLandmarks}}j.enlargeBox=to;function no(e){const t=ve(e),n=Ce(e),o=Math.max(...n),i=o/2,s=[t[0]-i,t[1]-i],r=[t[0]+i,t[1]+i];return{startPoint:s,endPoint:r,palmLandmarks:e.palmLandmarks}}j.squarifyBox=no;function oo(e,t){const n=[e.endPoint[0]-e.startPoint[0],e.endPoint[1]-e.startPoint[1]],o=[n[0]*t[0],n[1]*t[1]],i=[e.startPoint[0]+o[0],e.startPoint[1]+o[1]],s=[e.endPoint[0]+o[0],e.endPoint[1]+o[1]];return{startPoint:i,endPoint:s,palmLandmarks:e.palmLandmarks}}j.shiftBox=oo});var Kt=y(Wt=>{const b=require("@tensorflow/tfjs"),so=De();class io{constructor(e,t,n){this.model=e,this.width=n.inputSize,this.height=n.inputSize,this.anchors=t.map(o=>[o.x_center,o.y_center]),this.anchorsTensor=b.tensor2d(this.anchors),this.inputSizeTensor=b.tensor1d([n.inputSize,n.inputSize]),this.doubleInputSizeTensor=b.tensor1d([n.inputSize*2,n.inputSize*2])}normalizeBoxes(e){return b.tidy(()=>{const t=b.slice(e,[0,0],[-1,2]),n=b.slice(e,[0,2],[-1,2]),o=b.add(b.div(t,this.inputSizeTensor),this.anchorsTensor),i=b.div(n,this.doubleInputSizeTensor),s=b.mul(b.sub(o,i),this.inputSizeTensor),r=b.mul(b.add(o,i),this.inputSizeTensor);return b.concat2d([s,r],1)})}normalizeLandmarks(e,t){return b.tidy(()=>{const n=b.add(b.div(e.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return b.mul(n,this.inputSizeTensor)})}async getBoundingBoxes(e){const t=b.tidy(()=>b.mul(b.sub(e,.5),2)),n=this.model.predict(t),o=n.squeeze(),i=b.tidy(()=>b.sigmoid(b.slice(o,[0,0],[-1,1])).squeeze()),s=b.slice(o,[0,1],[-1,4]),r=this.normalizeBoxes(s),a=await b.image.nonMaxSuppressionAsync(r,i,this.maxHands,this.iouThreshold,this.scoreThreshold),c=await a.array(),l=[t,n,a,o,r,s,i];if(c.length===0)return l.forEach(u=>u.dispose()),null;const h=b.tidy(()=>{const u=[];for(const d in c){const m=c[d],p=b.slice(r,[m,0],[1,-1]),x=b.slice(o,[m,5],[1,14]),E=b.tidy(()=>this.normalizeLandmarks(x,m).reshape([-1,2]));u.push({boxes:p,palmLandmarks:E})}return u});return h}async estimateHandBounds(e,t){const n=e.shape[1],o=e.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const i=b.tidy(()=>e.resizeBilinear([this.width,this.height]).div(255)),s=await this.getBoundingBoxes(i);if(i.dispose(),!s||s.length===0)return null;const r=[];for(const a in s){const c=s[a],l=await c.boxes.array(),h=l[0].slice(0,2),u=l[0].slice(2,4),d=await c.palmLandmarks.array();c.boxes.dispose(),c.palmLandmarks.dispose(),r.push(so.scaleBoxCoordinates({startPoint:h,endPoint:u,palmLandmarks:d},[o/this.width,n/this.height]))}return r}}Wt.HandDetector=io});var Gt=y(Yt=>{Yt.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var $t=y(U=>{function Xt(e){return e-2*Math.PI*Math.floor((e+Math.PI)/(2*Math.PI))}U.normalizeRadians=Xt;function ro(e,t){const n=Math.PI/2-Math.atan2(-(t[1]-e[1]),t[0]-e[0]);return Xt(n)}U.computeRotation=ro;const Vt=(e,t)=>[[1,0,e],[0,1,t],[0,0,1]];function $(e,t){let n=0;for(let o=0;o<e.length;o++)n+=e[o]*t[o];return n}U.dot=$;function Qt(e,t){const n=[];for(let o=0;o<e.length;o++)n.push(e[o][t]);return n}U.getColumnFrom2DArr=Qt;function Zt(e,t){const n=[],o=e.length;for(let i=0;i<o;i++){n.push([]);for(let s=0;s<o;s++)n[i].push($(e[i],Qt(t,s)))}return n}function ao(e,t){const n=Math.cos(e),o=Math.sin(e),i=[[n,-o,0],[o,n,0],[0,0,1]],s=Vt(t[0],t[1]),r=Zt(s,i),a=Vt(-t[0],-t[1]);return Zt(r,a)}U.buildRotationMatrix=ao;function co(e){const t=[[e[0][0],e[1][0]],[e[0][1],e[1][1]]],n=[e[0][2],e[1][2]],o=[-$(t[0],n),-$(t[1],n)];return[t[0].concat(o[0]),t[1].concat(o[1]),[0,0,1]]}U.invertTransformMatrix=co;function lo(e,t){return[$(e,t[0]),$(e,t[1])]}U.rotatePoint=lo});var n0=y(Jt=>{const e0=require("@tensorflow/tfjs"),O=De(),W=$t(),ho=.8,uo=[0,-.4],mo=[0,-.1],po=1.65,t0=[0,5,9,13,17,1,2],fo=0,go=2;class bo{constructor(e,t,n){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=e,this.meshDetector=t,this.meshWidth=n.inputSize,this.meshHeight=n.inputSize,this.enlargeFactor=n.enlargeFactor}getBoxForPalmLandmarks(e,t){const n=e.map(i=>{const s=[...i,1];return W.rotatePoint(s,t)}),o=this.calculateLandmarksBoundingBox(n);return O.enlargeBox(O.squarifyBox(O.shiftBox(o,uo)),this.enlargeFactor)}getBoxForHandLandmarks(e){const t=this.calculateLandmarksBoundingBox(e),n=O.enlargeBox(O.squarifyBox(O.shiftBox(t,mo)),po),o=[];for(let i=0;i<t0.length;i++)o.push(e[t0[i]].slice(0,2));return n.palmLandmarks=o,n}transformRawCoords(e,t,n,o){const i=O.getBoxSize(t),s=[i[0]/this.meshWidth,i[1]/this.meshHeight],r=e.map(d=>[s[0]*(d[0]-this.meshWidth/2),s[1]*(d[1]-this.meshHeight/2),d[2]]),a=W.buildRotationMatrix(n,[0,0]),c=r.map(d=>{const m=W.rotatePoint(d,a);return[...m,d[2]]}),l=W.invertTransformMatrix(o),h=[...O.getBoxCenter(t),1],u=[W.dot(h,l[0]),W.dot(h,l[1])];return c.map(d=>[d[0]+u[0],d[1]+u[1],d[2]])}async estimateHands(e,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const n=this.shouldUpdateRegionsOfInterest();if(n===!0){const i=await this.boundingBoxDetector.estimateHandBounds(e,t);this.regionsOfInterest=[];for(const s in i)this.updateRegionsOfInterest(i[s],!0,s);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const o=[];if(!this.regionsOfInterest)return o;for(const i in this.regionsOfInterest){const s=this.regionsOfInterest[i][0];if(!s)return o;const r=W.computeRotation(s.palmLandmarks[fo],s.palmLandmarks[go]),a=O.getBoxCenter(s),c=[a[0]/e.shape[2],a[1]/e.shape[1]],l=e0.image.rotateWithOffset(e,r,0,c),h=W.buildRotationMatrix(-r,a),u=n?this.getBoxForPalmLandmarks(s.palmLandmarks,h):s,d=O.cutBoxFromImageAndResize(u,l,[this.meshWidth,this.meshHeight]),m=d.div(255);d.dispose(),l.dispose();const p=this.meshDetector.predict(m),[x,E]=p;m.dispose();const P=x.dataSync()[0];if(x.dispose(),P<t.minConfidence)return E.dispose(),this.regionsOfInterest[i]=[],o;const w=e0.reshape(E,[-1,3]),N=await w.array();E.dispose(),w.dispose();const v=this.transformRawCoords(N,u,r,h),J=this.getBoxForHandLandmarks(v);this.updateRegionsOfInterest(J,!1,i);const X={landmarks:v,confidence:P,box:{topLeft:J.startPoint,bottomRight:J.endPoint}};o.push(X)}return o}calculateLandmarksBoundingBox(e){const t=e.map(s=>s[0]),n=e.map(s=>s[1]),o=[Math.min(...t),Math.min(...n)],i=[Math.max(...t),Math.max(...n)];return{startPoint:o,endPoint:i}}updateRegionsOfInterest(e,t,n){if(t)this.regionsOfInterest[n]=[e];else{const o=this.regionsOfInterest[n][0];let i=0;if(o!=null&&o.startPoint!=null){const[s,r]=e.startPoint,[a,c]=e.endPoint,[l,h]=o.startPoint,[u,d]=o.endPoint,m=Math.max(s,l),p=Math.max(r,h),x=Math.min(a,u),E=Math.min(c,d),P=(x-m)*(E-p),w=(a-s)*(c-r),N=(u-l)*(d-r);i=P/(w+N-P)}this.regionsOfInterest[n][0]=i>ho?o:e}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}Jt.HandPipeline=bo});var i0=y(Fe=>{const G=require("@tensorflow/tfjs"),yo=Kt(),o0=Gt(),xo=n0();class s0{constructor(e){this.pipeline=e}async estimateHands(e,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const n=G.tidy(()=>(e instanceof G.Tensor||(e=G.browser.fromPixels(e)),e.toFloat().expandDims(0))),o=await this.pipeline.estimateHands(n,t);n.dispose();const i=[];if(!o)return i;for(const s of o){if(!s)return[];const r={};for(const a of Object.keys(o0.MESH_ANNOTATIONS))r[a]=o0.MESH_ANNOTATIONS[a].map(c=>s.landmarks[c]);i.push({confidence:s.confidence||0,box:s.box?[s.box.topLeft[0],s.box.topLeft[1],s.box.bottomRight[0]-s.box.topLeft[0],s.box.bottomRight[1]-s.box.topLeft[1]]:0,landmarks:s.landmarks,annotations:r})}return i}}Fe.HandPose=s0;async function wo(e){if(G.env().features.IS_NODE){const t=require("fs"),n=await t.readFileSync(e.replace("file://",""));return JSON.parse(n)}return G.util.fetch(e).then(t=>t.json())}async function Po(e){const[t,n,o]=await Promise.all([wo(e.detector.anchors),G.loadGraphModel(e.detector.modelPath,{fromTFHub:e.detector.modelPath.includes("tfhub.dev")}),G.loadGraphModel(e.skeleton.modelPath,{fromTFHub:e.skeleton.modelPath.includes("tfhub.dev")})]),i=new yo.HandDetector(n,t,e),s=new xo.HandPipeline(i,o,e),r=new s0(s);return r}Fe.load=Po});var r0=y(Eo=>{Ke(Eo,{default:()=>Mo});var Mo={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:5,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:2,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var c0=y((ns,a0)=>{a0.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});var f0=y(A=>{const S=require("@tensorflow/tfjs"),d0=lt(),le=ut(),l0=ft(),h0=Ut(),u0=i0(),m0=r0().default,Io=c0();let g;const M={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},C=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),he=(...e)=>{g.console&&console.log(...e)};function p0(...e){const t=n=>n&&typeof n=="object";return e.reduce((n,o)=>(Object.keys(o||{}).forEach(i=>{const s=n[i],r=o[i];Array.isArray(s)&&Array.isArray(r)?n[i]=s.concat(...r):t(s)&&t(r)?n[i]=p0(s,r):n[i]=r}),n),{})}function So(e){if(!e)return"input is not defined";const t=e.naturalWidth||e.videoWidth||e.width||e.shape&&e.shape[1]>0;if(!t||t===0)return"input is empty";if(e.readyState&&e.readyState<=2)return"input is not ready";try{S.getBackend()}catch{return"backend not loaded"}return null}async function Bo(e,t){g=p0(m0,t);const n=So(e);return n?(he(n,e),{error:n}):new Promise(async o=>{const i=Object.values(M).filter(u=>u).length;i===0&&he("Human library starting"),S.getBackend()!==g.backend&&(he("Human library setting backend:",g.backend),await S.setBackend(g.backend),await S.ready());let s;S.getBackend()==="webgl"&&(s=S.env().get("WEBGL_PACK_DEPTHWISECONV"),S.env().set("WEBGL_PACK_DEPTHWISECONV",!0)),g.face.enabled&&!M.facemesh&&(M.facemesh=await d0.load(g.face)),g.body.enabled&&!M.posenet&&(M.posenet=await h0.load(g.body)),g.hand.enabled&&!M.handpose&&(M.handpose=await u0.load(g.hand)),g.face.enabled&&g.face.age.enabled&&!M.age&&(M.age=await le.loadAge(g)),g.face.enabled&&g.face.gender.enabled&&!M.gender&&(M.gender=await le.loadGender(g)),g.face.enabled&&g.face.emotion.enabled&&!M.emotion&&(M.emotion=await l0.load(g));const r={};let a;a=C(),S.engine().startScope();const c=g.body.enabled?await M.posenet.estimatePoses(e,g.body):[];S.engine().endScope(),r.body=Math.trunc(C()-a),a=C(),S.engine().startScope();const l=g.hand.enabled?await M.handpose.estimateHands(e,g.hand):[];S.engine().endScope(),r.hand=Math.trunc(C()-a);const h=[];if(g.face.enabled){a=C(),S.engine().startScope();const u=await M.facemesh.estimateFaces(e,g.face);r.face=Math.trunc(C()-a);for(const d of u){if(!d.image||d.image.isDisposedInternal){he("face object is disposed:",d.image);continue}a=C();const m=g.face.age.enabled||g.face.gender.enabled?await le.predict(d.image,g):{};r.agegender=Math.trunc(C()-a),a=C();const p=g.face.emotion.enabled?await l0.predict(d.image,g):{};r.emotion=Math.trunc(C()-a),d.image.dispose();const x=d.annotations.leftEyeIris&&d.annotations.rightEyeIris?Math.max(d.annotations.leftEyeIris[3][0]-d.annotations.leftEyeIris[1][0],d.annotations.rightEyeIris[3][0]-d.annotations.rightEyeIris[1][0]):0;h.push({confidence:d.confidence,box:d.box,mesh:d.mesh,annotations:d.annotations,age:m.age,gender:m.gender,emotion:p,iris:x!==0?Math.trunc(100*11.7/x)/100:0})}S.engine().endScope()}S.env().set("WEBGL_PACK_DEPTHWISECONV",s),r.total=Object.values(r).reduce((u,d)=>u+d),o({face:h,body:c,hand:l,performance:r})})}A.detect=Bo;A.defaults=m0;A.config=g;A.models=M;A.facemesh=d0;A.ssrnet=le;A.posenet=h0;A.handpose=u0;A.tf=S;A.version=Io.version});export default f0();
+var Ue=Object.defineProperty;var y=(e,t)=>()=>(t||(t={exports:{}},e(t.exports,t)),t.exports),B0=e=>Ue(e,"__esModule",{value:!0}),Ke=(e,t)=>{B0(e);for(var n in t)Ue(e,n,{get:t[n],enumerable:!0})};var Qe=y(se=>{const f=require("@tensorflow/tfjs"),Ye=6;function T0(e){const t={strides:[e/16,e/8],anchors:[2,6]},n=[];for(let o=0;o<t.strides.length;o++){const i=t.strides[o],s=Math.floor((e+i-1)/i),r=Math.floor((e+i-1)/i),c=t.anchors[o];for(let a=0;a<s;a++){const d=i*(a+.5);for(let u=0;u<r;u++){const l=i*(u+.5);for(let h=0;h<c;h++)n.push([l,d])}}}return n}const Ge=e=>{e.startEndTensor.dispose(),e.startPoint.dispose(),e.endPoint.dispose()},Xe=e=>({startEndTensor:e,startPoint:f.slice(e,[0,0],[-1,2]),endPoint:f.slice(e,[0,2],[-1,2])}),k0=(e,t)=>{const n=f.mul(e.startPoint,t),o=f.mul(e.endPoint,t),i=f.concat2d([n,o],1);return Xe(i)};function _0(e,t,n){const o=f.slice(e,[0,1],[-1,2]),i=f.add(o,t),s=f.slice(e,[0,3],[-1,2]),r=f.div(s,n),c=f.div(i,n),a=f.div(r,2),d=f.sub(c,a),u=f.add(c,a),l=f.mul(d,n),h=f.mul(u,n),p=1;return f.concat2d([l,h],p)}function z0(e,t){return f.tidy(()=>{const n=e.box?e.box:e;return k0(n,t).startEndTensor.squeeze()})}class Ve{constructor(e,t){this.blazeFaceModel=e,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=T0(t.detector.inputSize),this.anchors=f.tensor2d(this.anchorsData),this.inputSize=f.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(e){if(!e||e.isDisposedInternal||e.shape.length!==4||e.shape[1]<1||e.shape[2]<1)return null;const[t,n,o]=f.tidy(()=>{const a=e.resizeBilinear([this.width,this.height]),d=f.mul(f.sub(a.div(255),.5),2),u=this.blazeFaceModel.predict(d);let l;if(Array.isArray(u)){const w=u.sort((N,C)=>N.size-C.size),E=f.concat([w[0],w[2]],2),P=f.concat([w[1],w[3]],2),x=f.concat([P,E],1);l=x.squeeze(0)}else l=u.squeeze();const h=_0(l,this.anchors,this.inputSize),p=f.slice(l,[0,0],[-1,1]),m=f.sigmoid(p).squeeze();return[l,h,m]}),i=await f.image.nonMaxSuppressionAsync(n,o,this.maxFaces,this.iouThreshold,this.scoreThreshold),s=await i.array();i.dispose();let r=s.map(a=>f.slice(n,[a,0],[1,-1]));r=await Promise.all(r.map(async a=>{const d=await a.array();return a.dispose(),d}));const c=[];for(let a=0;a<r.length;a++){const d=r[a],u=f.tidy(()=>{const l=Xe(d),h=s[a],p=this.anchorsData[h],m=f.slice(t,[h,Ye-1],[1,-1]).squeeze().reshape([Ye,-1]),w=f.slice(o,[h],[1]);return{box:l,landmarks:m,probability:w,anchor:p}});c.push(u)}return n.dispose(),o.dispose(),t.dispose(),{boxes:c,scaleFactor:[e.shape[2]/this.width,e.shape[1]/this.height]}}async estimateFaces(e){const t=f.tidy(()=>(e instanceof f.Tensor||(e=f.browser.fromPixels(e)),e.toFloat().expandDims(0))),{boxes:n,scaleFactor:o}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(n.map(async i=>{const s=z0(i,o),[r,c,a]=await Promise.all([i.landmarks,s,i.probability].map(async m=>m.array())),d=i.anchor,[u,l]=o,h=r.map(m=>[(m[0]+d[0])*u,(m[1]+d[1])*l]),p={topLeft:c.slice(0,2),bottomRight:c.slice(2),landmarks:h,probability:a};return Ge(i.box),i.landmarks.dispose(),i.probability.dispose(),s.dispose(),p}))}}async function A0(e){const t=await f.loadGraphModel(e.detector.modelPath,{fromTFHub:e.detector.modelPath.includes("tfhub.dev")}),n=new Ve(t,e);return n}se.load=A0;se.BlazeFaceModel=Ve;se.disposeBox=Ge});var fe=y(pe=>{pe.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};pe.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var Ze=y(K=>{const N0=require("@tensorflow/tfjs");function R0(e,t){const n=[e.startPoint[0]*t[0],e.startPoint[1]*t[1]],o=[e.endPoint[0]*t[0],e.endPoint[1]*t[1]];return{startPoint:n,endPoint:o}}K.scaleBoxCoordinates=R0;function ge(e){return[Math.abs(e.endPoint[0]-e.startPoint[0]),Math.abs(e.endPoint[1]-e.startPoint[1])]}K.getBoxSize=ge;function be(e){return[e.startPoint[0]+(e.endPoint[0]-e.startPoint[0])/2,e.startPoint[1]+(e.endPoint[1]-e.startPoint[1])/2]}K.getBoxCenter=be;function O0(e,t,n){const o=t.shape[1],i=t.shape[2],s=[[e.startPoint[1]/o,e.startPoint[0]/i,e.endPoint[1]/o,e.endPoint[0]/i]];return N0.image.cropAndResize(t,s,[0],n)}K.cutBoxFromImageAndResize=O0;function v0(e,t=1.5){const n=be(e),o=ge(e),i=[t*o[0]/2,t*o[1]/2],s=[n[0]-i[0],n[1]-i[1]],r=[n[0]+i[0],n[1]+i[1]];return{startPoint:s,endPoint:r,landmarks:e.landmarks}}K.enlargeBox=v0;function C0(e){const t=be(e),n=ge(e),o=Math.max(...n),i=o/2,s=[t[0]-i,t[1]-i],r=[t[0]+i,t[1]+i];return{startPoint:s,endPoint:r,landmarks:e.landmarks}}K.squarifyBox=C0});var nt=y(k=>{k.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function $e(e){return e-2*Math.PI*Math.floor((e+Math.PI)/(2*Math.PI))}k.normalizeRadians=$e;function D0(e,t){const n=Math.PI/2-Math.atan2(-(t[1]-e[1]),t[0]-e[0]);return $e(n)}k.computeRotation=D0;function F0(e){return e*180/Math.PI}k.radToDegrees=F0;function Je(e,t){return[[1,0,e],[0,1,t],[0,0,1]]}function V(e,t){let n=0;for(let o=0;o<e.length;o++)n+=e[o]*t[o];return n}k.dot=V;function et(e,t){const n=[];for(let o=0;o<e.length;o++)n.push(e[o][t]);return n}k.getColumnFrom2DArr=et;function tt(e,t){const n=[],o=e.length;for(let i=0;i<o;i++){n.push([]);for(let s=0;s<o;s++)n[i].push(V(e[i],et(t,s)))}return n}function L0(e,t){const n=Math.cos(e),o=Math.sin(e),i=[[n,-o,0],[o,n,0],[0,0,1]],s=Je(t[0],t[1]),r=tt(s,i),c=Je(-t[0],-t[1]);return tt(r,c)}k.buildRotationMatrix=L0;function H0(e){const t=[[e[0][0],e[1][0]],[e[0][1],e[1][1]]],n=[e[0][2],e[1][2]],o=[-V(t[0],n),-V(t[1],n)];return[t[0].concat(o[0]),t[1].concat(o[1]),[0,0,1]]}k.invertTransformMatrix=H0;function q0(e,t){return[V(e,t[0]),V(e,t[1])]}k.rotatePoint=q0;function j0(e,t){return Math.sqrt((e[0]-t[0])**2+(e[1]-t[1])**2)}k.xyDistanceBetweenPoints=j0});var st=y(ot=>{const D=require("@tensorflow/tfjs"),R=Ze(),F=fe(),L=nt(),W0=468,U0=.25,K0=13,Y0=[K0,F.MESH_ANNOTATIONS.midwayBetweenEyes[0]],G0=3,X0=2,V0=[G0,X0],ye=F.MESH_ANNOTATIONS.leftEyeLower0,xe=[ye[0],ye[ye.length-1]],we=F.MESH_ANNOTATIONS.rightEyeLower0,Pe=[we[0],we[we.length-1]],Q0=3,Z0=4,$0=71,Ee=76;function ie(e,t,n,o){for(let i=0;i<F.MESH_TO_IRIS_INDICES_MAP.length;i++){const{key:s,indices:r}=F.MESH_TO_IRIS_INDICES_MAP[i],c=F.MESH_ANNOTATIONS[`${n}${s}`],a=o==null;if(a||o.includes(s))for(let d=0;d<r.length;d++){const u=r[d];e[c[d]]=[t[u][0],t[u][1],(t[u][2]+e[c[d]][2])/2]}}}class J0{constructor(e,t,n,o){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=e,this.meshDetector=t,this.irisModel=n,this.meshWidth=o.mesh.inputSize,this.meshHeight=o.mesh.inputSize,this.irisSize=o.iris.inputSize,this.irisEnlarge=o.iris.enlargeFactor}transformRawCoords(e,t,n,o){const i=R.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),s=[i[0]/this.meshWidth,i[1]/this.meshHeight],r=e.map(h=>[s[0]*(h[0]-this.meshWidth/2),s[1]*(h[1]-this.meshHeight/2),h[2]]),c=L.buildRotationMatrix(n,[0,0]),a=r.map(h=>[...L.rotatePoint(h,c),h[2]]),d=L.invertTransformMatrix(o),u=[...R.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],l=[L.dot(u,d[0]),L.dot(u,d[1])];return a.map(h=>[h[0]+l[0],h[1]+l[1],h[2]])}getLeftToRightEyeDepthDifference(e){const t=e[xe[0]][2],n=e[Pe[0]][2];return t-n}getEyeBox(e,t,n,o,i=!1){const s=R.squarifyBox(R.enlargeBox(this.calculateLandmarksBoundingBox([e[n],e[o]]),this.irisEnlarge)),r=R.getBoxSize(s);let c=D.image.cropAndResize(t,[[s.startPoint[1]/this.meshHeight,s.startPoint[0]/this.meshWidth,s.endPoint[1]/this.meshHeight,s.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return i&&(c=D.image.flipLeftRight(c)),{box:s,boxSize:r,crop:c}}getEyeCoords(e,t,n,o=!1){const i=[];for(let s=0;s<Ee;s++){const r=e[s*3],c=e[s*3+1],a=e[s*3+2];i.push([(o?1-r/this.irisSize:r/this.irisSize)*n[0]+t.startPoint[0],c/this.irisSize*n[1]+t.startPoint[1],a])}return{rawCoords:i,iris:i.slice($0)}}getAdjustedIrisCoords(e,t,n){const o=e[F.MESH_ANNOTATIONS[`${n}EyeUpper0`][Q0]][2],i=e[F.MESH_ANNOTATIONS[`${n}EyeLower0`][Z0]][2],s=(o+i)/2;return t.map((r,c)=>{let a=s;return c===2?a=o:c===4&&(a=i),[r[0],r[1],a]})}async predict(e,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:o,scaleFactor:i}=await this.boundingBoxDetector.getBoundingBoxes(e);if(o.length===0)return this.regionsOfInterest=[],null;const s=o.map(r=>{const c={startPoint:r.box.startPoint.squeeze().arraySync(),endPoint:r.box.endPoint.squeeze().arraySync()};r.box.startPoint.dispose(),r.box.endPoint.dispose();const a=R.scaleBoxCoordinates(c,i),d=R.enlargeBox(a),u=r.landmarks.arraySync();return r.landmarks.dispose(),r.probability.dispose(),{...d,landmarks:u}});this.updateRegionsOfInterest(s),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const n=D.tidy(()=>this.regionsOfInterest.map((o,i)=>{let s=0;const r=o.landmarks.length>=W0;let[c,a]=Y0;r===!1&&([c,a]=V0),s=L.computeRotation(o.landmarks[c],o.landmarks[a]);const d=R.getBoxCenter({startPoint:o.startPoint,endPoint:o.endPoint}),u=[d[0]/e.shape[2],d[1]/e.shape[1]];let l=e,h=L.IDENTITY_MATRIX;s!==0&&(l=D.image.rotateWithOffset(e,s,0,u),h=L.buildRotationMatrix(-s,d));const p={startPoint:o.startPoint,endPoint:o.endPoint},m=R.cutBoxFromImageAndResize(p,l,[this.meshHeight,this.meshWidth]).div(255),[,w,E]=this.meshDetector.predict(m),P=D.reshape(E,[-1,3]);let x=P.arraySync();if(t.iris.enabled){const{box:X,boxSize:me,crop:g0}=this.getEyeBox(x,m,xe[0],xe[1],!0),{box:b0,boxSize:y0,crop:x0}=this.getEyeBox(x,m,Pe[0],Pe[1]),Le=this.irisModel.predict(D.concat([g0,x0])),He=Le.dataSync();Le.dispose();const w0=He.slice(0,Ee*3),{rawCoords:qe,iris:P0}=this.getEyeCoords(w0,X,me,!0),E0=He.slice(Ee*3),{rawCoords:je,iris:M0}=this.getEyeCoords(E0,b0,y0),We=this.getLeftToRightEyeDepthDifference(x);Math.abs(We)<30?(ie(x,qe,"left"),ie(x,je,"right")):We<1?ie(x,qe,"left",["EyeUpper0","EyeLower0"]):ie(x,je,"right",["EyeUpper0","EyeLower0"]);const I0=this.getAdjustedIrisCoords(x,P0,"left"),S0=this.getAdjustedIrisCoords(x,M0,"right");x=x.concat(I0).concat(S0)}const N=this.transformRawCoords(x,o,s,h);D.dispose(x);const C=R.enlargeBox(this.calculateLandmarksBoundingBox(N));if(t.mesh.enabled){const X=D.tensor2d(N);this.regionsOfInterest[i]={...C,landmarks:X.arraySync()};const me={coords:X,box:C,confidence:w.squeeze(),image:m};return me}const J={coords:null,box:C,confidence:w.squeeze(),image:m};return J}));return n}updateRegionsOfInterest(e){for(let t=0;t<e.length;t++){const n=e[t],o=this.regionsOfInterest[t];let i=0;if(o&&o.startPoint){const[s,r]=n.startPoint,[c,a]=n.endPoint,[d,u]=o.startPoint,[l,h]=o.endPoint,p=Math.max(s,d),m=Math.max(r,u),w=Math.min(c,l),E=Math.min(a,h),P=(w-p)*(E-m),x=(c-s)*(a-r),N=(l-d)*(h-r);i=P/(x+N-P)}i<U0&&(this.regionsOfInterest[t]=n)}this.regionsOfInterest=this.regionsOfInterest.slice(0,e.length)}clearRegionOfInterest(e){this.regionsOfInterest[e]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,e),...this.regionsOfInterest.slice(e+1)])}shouldUpdateRegionsOfInterest(){const e=this.regionsOfInterest.length,t=e===0;return this.maxFaces===1||t?t:e!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(e){const t=e.map(s=>s[0]),n=e.map(s=>s[1]),o=[Math.min(...t),Math.min(...n)],i=[Math.max(...t),Math.max(...n)];return{startPoint:o,endPoint:i}}}ot.Pipeline=J0});var rt=y(it=>{it.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var at=y(en=>{Ke(en,{default:()=>tn});var tn=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var lt=y(ee=>{const Y=require("@tensorflow/tfjs"),nn=Qe(),ct=fe(),on=st(),sn=rt(),rn=at().default;class dt{constructor(e,t,n,o){this.pipeline=new on.Pipeline(e,t,n,o),o&&(this.config=o)}async estimateFaces(e,t){t&&(this.config=t);const n=Y.tidy(()=>(e instanceof Y.Tensor||(e=Y.browser.fromPixels(e)),e.toFloat().expandDims(0))),o=await this.pipeline.predict(n,t);Y.dispose(n);const i=[];for(const s of o||[]){const r=s.confidence.arraySync();if(r>=this.config.detector.minConfidence){const c=s.coords?s.coords.arraySync():null,a={};if(c&&c.length>0)for(const d in ct.MESH_ANNOTATIONS)(this.config.iris.enabled||d.includes("Iris")===!1)&&(a[d]=ct.MESH_ANNOTATIONS[d].map(u=>c[u]));i.push({confidence:r||0,box:s.box?[s.box.startPoint[0],s.box.startPoint[1],s.box.endPoint[0]-s.box.startPoint[0],s.box.endPoint[1]-s.box.startPoint[1]]:0,mesh:c,annotations:a,image:s.image?Y.clone(s.image):null})}s.confidence.dispose(),s.image.dispose()}return i}}async function an(e){const t=await Promise.all([nn.load(e),Y.loadGraphModel(e.mesh.modelPath,{fromTFHub:e.mesh.modelPath.includes("tfhub.dev")}),Y.loadGraphModel(e.iris.modelPath,{fromTFHub:e.iris.modelPath.includes("tfhub.dev")})]),n=new dt(t[0],t[1],t[2],e);return n}ee.load=an;ee.MediaPipeFaceMesh=dt;ee.uv_coords=sn;ee.triangulation=rn});var ut=y(re=>{const B=require("@tensorflow/tfjs"),H={};let ht={age:0,gender:""},ae=0;async function cn(e,t){const n=B.browser.fromPixels(e),o=B.image.resizeBilinear(n,[t,t]),i=B.cast(B.expandDims(o,0),"float32");return i}async function dn(e){return H.age||(H.age=await B.loadGraphModel(e.face.age.modelPath)),H.age}async function ln(e){return H.gender||(H.gender=await B.loadGraphModel(e.face.gender.modelPath)),H.gender}async function hn(e,t){if(ae>t.face.age.skipFrames?ae=0:ae+=1,ae===0)return ht;let n;if(e instanceof B.Tensor){const i=B.image.resizeBilinear(e,[t.face.age.inputSize,t.face.age.inputSize],!1);n=B.mul(i,[255]),B.dispose(i)}else n=await cn(e,t.face.age.inputSize);const o={};if(t.face.age.enabled){const i=await H.age.predict(n),s=await i.data();o.age=Math.trunc(10*s[0])/10,B.dispose(i)}if(t.face.gender.enabled){const i=await H.gender.predict(n),s=await i.data(),r=Math.trunc(Math.abs(1.9*100*(s[0]-.5)))/100;r>t.face.gender.minConfidence&&(o.gender=s[0]<=.5?"female":"male",o.confidence=r),B.dispose(i)}return B.dispose(n),ht=o,o}re.predict=hn;re.loadAge=dn;re.loadGender=ln});var ft=y(Me=>{const I=require("@tensorflow/tfjs"),un=["angry","discust","fear","happy","sad","surpise","neutral"],ce={};let mt=[],Ie=0;const pt=1.5;function mn(e,t){const n=I.tidy(()=>{const o=I.browser.fromPixels(e,1),i=I.image.resizeBilinear(o,[t,t]),s=I.cast(I.expandDims(i,0),"float32");return s});return n}async function pn(e){return ce.emotion||(ce.emotion=await I.loadGraphModel(e.face.emotion.modelPath)),ce.emotion}async function fn(e,t){if(Ie+=1,Ie>=t.face.emotion.skipFrames)return Ie=0,mt;const n=I.tidy(()=>{if(e instanceof I.Tensor){const i=I.image.resizeBilinear(e,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[s,r,c]=I.split(i,3,3);if(t.face.emotion.useGrayscale){const a=I.mul(s,[.2989]),d=I.mul(r,[.587]),u=I.mul(c,[.114]),l=I.addN([a,d,u]);return l}return r}return mn(e,t.face.emotion.inputSize)}),o=[];if(t.face.emotion.enabled){const i=await ce.emotion.predict(n),s=await i.data();for(let r=0;r<s.length;r++)pt*s[r]>t.face.emotion.minConfidence&&o.push({score:Math.min(.99,Math.trunc(100*pt*s[r])/100),emotion:un[r]});o.sort((r,c)=>c.score-r.score),I.dispose(i)}return I.dispose(n),mt=o,o}Me.predict=fn;Me.load=pn});var yt=y(gt=>{const bt=require("@tensorflow/tfjs");class gn{constructor(e,t){this.model=e,this.outputStride=t;const n=this.model.inputs[0].shape;bt.util.assert(n[1]===-1&&n[2]===-1,()=>`Input shape [${n[1]}, ${n[2]}] must both be equal to or -1`)}predict(e){return bt.tidy(()=>{const t=this.preprocessInput(e.toFloat()),n=t.expandDims(0),o=this.model.predict(n),i=o.map(r=>r.squeeze([0])),s=this.nameOutputResults(i);return{heatmapScores:s.heatmap.sigmoid(),offsets:s.offsets,displacementFwd:s.displacementFwd,displacementBwd:s.displacementBwd}})}dispose(){this.model.dispose()}}gt.BaseModel=gn});var Se=y(xt=>{const wt=require("@tensorflow/tfjs"),bn=yt();class yn extends bn.BaseModel{preprocessInput(e){return wt.tidy(()=>wt.div(e,127.5).sub(1))}nameOutputResults(e){const[t,n,o,i]=e;return{offsets:t,heatmap:n,displacementFwd:o,displacementBwd:i}}}xt.MobileNet=yn});var Et=y(Pt=>{function Be(e){return Math.floor(e/2)}class xn{constructor(e,t){this.priorityQueue=new Array(e),this.numberOfElements=-1,this.getElementValue=t}enqueue(e){this.priorityQueue[++this.numberOfElements]=e,this.swim(this.numberOfElements)}dequeue(){const e=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,e}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(e){for(;e>0&&this.less(Be(e),e);)this.exchange(e,Be(e)),e=Be(e)}sink(e){for(;2*e<=this.numberOfElements;){let t=2*e;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(e,t))break;this.exchange(e,t),e=t}}getValueAt(e){return this.getElementValue(this.priorityQueue[e])}less(e,t){return this.getValueAt(e)<this.getValueAt(t)}exchange(e,t){const n=this.priorityQueue[e];this.priorityQueue[e]=this.priorityQueue[t],this.priorityQueue[t]=n}}Pt.MaxHeap=xn});var It=y(Mt=>{const wn=Et();function Pn(e,t,n,o,i,s){const[r,c]=s.shape;let a=!0;const d=Math.max(n-i,0),u=Math.min(n+i+1,r);for(let l=d;l<u;++l){const h=Math.max(o-i,0),p=Math.min(o+i+1,c);for(let m=h;m<p;++m)if(s.get(l,m,e)>t){a=!1;break}if(!a)break}return a}function En(e,t,n){const[o,i,s]=n.shape,r=new wn.MaxHeap(o*i*s,({score:c})=>c);for(let c=0;c<o;++c)for(let a=0;a<i;++a)for(let d=0;d<s;++d){const u=n.get(c,a,d);if(u<e)continue;Pn(d,u,c,a,t,n)&&r.enqueue({score:u,part:{heatmapY:c,heatmapX:a,id:d}})}return r}Mt.buildPartWithScoreQueue=En});var te=y(_=>{_.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];_.NUM_KEYPOINTS=_.partNames.length;_.partIds=_.partNames.reduce((e,t,n)=>(e[t]=n,e),{});const Mn=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];_.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];_.connectedPartIndices=Mn.map(([e,t])=>[_.partIds[e],_.partIds[t]]);_.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var ke=y(q=>{const In=te();function St(e,t,n,o){return{y:o.get(e,t,n),x:o.get(e,t,n+In.NUM_KEYPOINTS)}}q.getOffsetPoint=St;function Sn(e,t,n){const{heatmapY:o,heatmapX:i,id:s}=e,{y:r,x:c}=St(o,i,s,n);return{x:e.heatmapX*t+c,y:e.heatmapY*t+r}}q.getImageCoords=Sn;function Bn(e,t){const n=new Array(t);for(let o=0;o<t;o++)n[o]=e;return n}q.fillArray=Bn;function Te(e,t,n){return e<t?t:e>n?n:e}q.clamp=Te;function Tn(e,t,n,o){const i=n-e,s=o-t;return i*i+s*s}q.squaredDistance=Tn;function kn(e,t){return{x:e.x+t.x,y:e.y+t.y}}q.addVectors=kn;function _n(e,t,n){return{y:Te(e.y,t,n),x:Te(e.x,t,n)}}q.clampVector=_n});var zt=y(Bt=>{const ne=te(),Q=ke(),Tt=ne.poseChain.map(([e,t])=>[ne.partIds[e],ne.partIds[t]]),_e=Tt.map(([,e])=>e),kt=Tt.map(([e])=>e);function zn(e,t,n){const o=n.shape[2]/2;return{y:n.get(t.y,t.x,e),x:n.get(t.y,t.x,o+e)}}function ze(e,t,n,o){return{y:Q.clamp(Math.round(e.y/t),0,n-1),x:Q.clamp(Math.round(e.x/t),0,o-1)}}function _t(e,t,n,o,i,s,r,c=2){const[a,d]=o.shape,u=ze(t.position,s,a,d),l=zn(e,u,r),h=Q.addVectors(t.position,l);let p=h;for(let E=0;E<c;E++){const P=ze(p,s,a,d),x=Q.getOffsetPoint(P.y,P.x,n,i);p=Q.addVectors({x:P.x*s,y:P.y*s},{x:x.x,y:x.y})}const m=ze(p,s,a,d),w=o.get(m.y,m.x,n);return{position:p,part:ne.partNames[n],score:w}}function An(e,t,n,o,i,s){const r=t.shape[2],c=_e.length,a=new Array(r),{part:d,score:u}=e,l=Q.getImageCoords(d,o,n);a[d.id]={score:u,part:ne.partNames[d.id],position:l};for(let h=c-1;h>=0;--h){const p=_e[h],m=kt[h];a[p]&&!a[m]&&(a[m]=_t(h,a[p],m,t,n,o,s))}for(let h=0;h<c;++h){const p=kt[h],m=_e[h];a[p]&&!a[m]&&(a[m]=_t(h,a[p],m,t,n,o,i))}return a}Bt.decodePose=An});var Ae=y(At=>{const Nn=It(),Rn=zt(),Nt=ke();function Rt(e,t,{x:n,y:o},i){return e.some(({keypoints:s})=>{const r=s[i].position;return Nt.squaredDistance(o,n,r.y,r.x)<=t})}function On(e,t,n){const o=n.reduce((i,{position:s,score:r},c)=>(Rt(e,t,s,c)||(i+=r),i),0);return o/n.length}const vn=1;function Cn(e,t,n,o,i,s,r=.5,c=20){const a=[],d=Nn.buildPartWithScoreQueue(r,vn,e),u=c*c;for(;a.length<s&&!d.empty();){const l=d.dequeue(),h=Nt.getImageCoords(l.part,i,t);if(Rt(a,u,h,l.part.id))continue;const p=Rn.decodePose(l,e,t,i,n,o),m=On(a,u,p);a.push({keypoints:p,score:m})}return a}At.decodeMultiplePoses=Cn});var Re=y(T=>{const Z=require("@tensorflow/tfjs"),Dn=te();function Fn(e,t,n){return e<n||t<n}function Ln(e,t){return Dn.connectedPartIndices.reduce((n,[o,i])=>(Fn(e[o].score,e[i].score,t)||n.push([e[o],e[i]]),n),[])}T.getAdjacentKeyPoints=Ln;const{NEGATIVE_INFINITY:Ot,POSITIVE_INFINITY:vt}=Number;function Ct(e){return e.reduce(({maxX:t,maxY:n,minX:o,minY:i},{position:{x:s,y:r}})=>({maxX:Math.max(t,s),maxY:Math.max(n,r),minX:Math.min(o,s),minY:Math.min(i,r)}),{maxX:Ot,maxY:Ot,minX:vt,minY:vt})}T.getBoundingBox=Ct;function Hn(e){const{minX:t,minY:n,maxX:o,maxY:i}=Ct(e);return[{x:t,y:n},{x:o,y:n},{x:o,y:i},{x:t,y:i}]}T.getBoundingBoxPoints=Hn;async function qn(e){return Promise.all(e.map(t=>t.buffer()))}T.toTensorBuffers3D=qn;function Dt(e,t,n,o=0,i=0){return{score:e.score,keypoints:e.keypoints.map(({score:s,part:r,position:c})=>({score:s,part:r,position:{x:c.x*n+i,y:c.y*t+o}}))}}T.scalePose=Dt;function Ft(e,t,n,o=0,i=0){return n===1&&t===1&&o===0&&i===0?e:e.map(s=>Dt(s,t,n,o,i))}T.scalePoses=Ft;function Lt(e){return e instanceof Z.Tensor?[e.shape[0],e.shape[1]]:[e.height,e.width]}T.getInputTensorDimensions=Lt;function Ne(e){return e instanceof Z.Tensor?e:Z.browser.fromPixels(e)}T.toInputTensor=Ne;function jn(e,t,n){return Z.tidy(()=>{const o=Ne(e);return o.resizeBilinear([t,n])})}T.toResizedInputTensor=jn;function Wn(e,[t,n]){const[o,i]=Lt(e),s=n/t,r=i/o;let[c,a,d,u]=[0,0,0,0];r<s?(c=0,a=0,d=Math.round(.5*(s*o-i)),u=Math.round(.5*(s*o-i))):(c=Math.round(.5*(1/s*i-o)),a=Math.round(.5*(1/s*i-o)),d=0,u=0);const l=Z.tidy(()=>{let h=Ne(e);return h=Z.pad3d(h,[[c,a],[d,u],[0,0]]),h.resizeBilinear([t,n])});return{resized:l,padding:{top:c,left:d,right:u,bottom:a}}}T.padAndResizeTo=Wn;function Un(e,[t,n],[o,i],s){const r=(t+s.top+s.bottom)/o,c=(n+s.left+s.right)/i,a=Ft(e,r,c,-s.top,-s.left);return a}T.scaleAndFlipPoses=Un});var qt=y(Oe=>{const Kn=require("@tensorflow/tfjs"),Yn=Se(),Gn=Ae(),de=Re();class Ht{constructor(e){this.baseModel=e}async estimatePoses(e,t){const n=t.outputStride,[o,i]=de.getInputTensorDimensions(e),{resized:s,padding:r}=de.padAndResizeTo(e,[t.inputResolution,t.inputResolution]),{heatmapScores:c,offsets:a,displacementFwd:d,displacementBwd:u}=this.baseModel.predict(s),l=await de.toTensorBuffers3D([c,a,d,u]),h=l[0],p=l[1],m=l[2],w=l[3],E=await Gn.decodeMultiplePoses(h,p,m,w,n,t.maxDetections,t.scoreThreshold,t.nmsRadius),P=de.scaleAndFlipPoses(E,[o,i],[t.inputResolution,t.inputResolution],r);return c.dispose(),a.dispose(),d.dispose(),u.dispose(),s.dispose(),P}dispose(){this.baseModel.dispose()}}Oe.PoseNet=Ht;async function Xn(e){const t=await Kn.loadGraphModel(e.modelPath),n=new Yn.MobileNet(t,e.outputStride);return new Ht(n)}async function Vn(e){return Xn(e)}Oe.load=Vn});var Wt=y(S=>{const Qn=Se(),jt=qt(),Zn=Ae(),le=te(),oe=Re();S.load=jt.load;S.PoseNet=jt.PoseNet;S.MobileNet=Qn.MobileNet;S.decodeMultiplePoses=Zn.decodeMultiplePoses;S.partChannels=le.partChannels;S.partIds=le.partIds;S.partNames=le.partNames;S.poseChain=le.poseChain;S.getAdjacentKeyPoints=oe.getAdjacentKeyPoints;S.getBoundingBox=oe.getBoundingBox;S.getBoundingBoxPoints=oe.getBoundingBoxPoints;S.scaleAndFlipPoses=oe.scaleAndFlipPoses;S.scalePose=oe.scalePose});var De=y(j=>{const $n=require("@tensorflow/tfjs");function ve(e){return[Math.abs(e.endPoint[0]-e.startPoint[0]),Math.abs(e.endPoint[1]-e.startPoint[1])]}j.getBoxSize=ve;function Ce(e){return[e.startPoint[0]+(e.endPoint[0]-e.startPoint[0])/2,e.startPoint[1]+(e.endPoint[1]-e.startPoint[1])/2]}j.getBoxCenter=Ce;function Jn(e,t,n){const o=t.shape[1],i=t.shape[2],s=[[e.startPoint[1]/o,e.startPoint[0]/i,e.endPoint[1]/o,e.endPoint[0]/i]];return $n.image.cropAndResize(t,s,[0],n)}j.cutBoxFromImageAndResize=Jn;function eo(e,t){const n=[e.startPoint[0]*t[0],e.startPoint[1]*t[1]],o=[e.endPoint[0]*t[0],e.endPoint[1]*t[1]],i=e.palmLandmarks.map(s=>{const r=[s[0]*t[0],s[1]*t[1]];return r});return{startPoint:n,endPoint:o,palmLandmarks:i}}j.scaleBoxCoordinates=eo;function to(e,t=1.5){const n=Ce(e),o=ve(e),i=[t*o[0]/2,t*o[1]/2],s=[n[0]-i[0],n[1]-i[1]],r=[n[0]+i[0],n[1]+i[1]];return{startPoint:s,endPoint:r,palmLandmarks:e.palmLandmarks}}j.enlargeBox=to;function no(e){const t=Ce(e),n=ve(e),o=Math.max(...n),i=o/2,s=[t[0]-i,t[1]-i],r=[t[0]+i,t[1]+i];return{startPoint:s,endPoint:r,palmLandmarks:e.palmLandmarks}}j.squarifyBox=no;function oo(e,t){const n=[e.endPoint[0]-e.startPoint[0],e.endPoint[1]-e.startPoint[1]],o=[n[0]*t[0],n[1]*t[1]],i=[e.startPoint[0]+o[0],e.startPoint[1]+o[1]],s=[e.endPoint[0]+o[0],e.endPoint[1]+o[1]];return{startPoint:i,endPoint:s,palmLandmarks:e.palmLandmarks}}j.shiftBox=oo});var Kt=y(Ut=>{const b=require("@tensorflow/tfjs"),so=De();class io{constructor(e,t,n){this.model=e,this.width=n.inputSize,this.height=n.inputSize,this.anchors=t.map(o=>[o.x_center,o.y_center]),this.anchorsTensor=b.tensor2d(this.anchors),this.inputSizeTensor=b.tensor1d([n.inputSize,n.inputSize]),this.doubleInputSizeTensor=b.tensor1d([n.inputSize*2,n.inputSize*2])}normalizeBoxes(e){return b.tidy(()=>{const t=b.slice(e,[0,0],[-1,2]),n=b.slice(e,[0,2],[-1,2]),o=b.add(b.div(t,this.inputSizeTensor),this.anchorsTensor),i=b.div(n,this.doubleInputSizeTensor),s=b.mul(b.sub(o,i),this.inputSizeTensor),r=b.mul(b.add(o,i),this.inputSizeTensor);return b.concat2d([s,r],1)})}normalizeLandmarks(e,t){return b.tidy(()=>{const n=b.add(b.div(e.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return b.mul(n,this.inputSizeTensor)})}async getBoundingBoxes(e){const t=b.tidy(()=>b.mul(b.sub(e,.5),2)),n=this.model.predict(t),o=n.squeeze(),i=b.tidy(()=>b.sigmoid(b.slice(o,[0,0],[-1,1])).squeeze()),s=b.slice(o,[0,1],[-1,4]),r=this.normalizeBoxes(s),c=await b.image.nonMaxSuppressionAsync(r,i,this.maxHands,this.iouThreshold,this.scoreThreshold),a=await c.array(),d=[t,n,c,o,r,s,i];if(a.length===0)return d.forEach(l=>l.dispose()),null;const u=b.tidy(()=>{const l=[];for(const h in a){const p=a[h],m=b.slice(r,[p,0],[1,-1]),w=b.slice(o,[p,5],[1,14]),E=b.tidy(()=>this.normalizeLandmarks(w,p).reshape([-1,2]));l.push({boxes:m,palmLandmarks:E})}return l});return u}async estimateHandBounds(e,t){const n=e.shape[1],o=e.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const i=b.tidy(()=>e.resizeBilinear([this.width,this.height]).div(255)),s=await this.getBoundingBoxes(i);if(i.dispose(),!s||s.length===0)return null;const r=[];for(const c in s){const a=s[c],d=await a.boxes.array(),u=d[0].slice(0,2),l=d[0].slice(2,4),h=await a.palmLandmarks.array();a.boxes.dispose(),a.palmLandmarks.dispose(),r.push(so.scaleBoxCoordinates({startPoint:u,endPoint:l,palmLandmarks:h},[o/this.width,n/this.height]))}return r}}Ut.HandDetector=io});var Gt=y(Yt=>{Yt.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var $t=y(W=>{function Xt(e){return e-2*Math.PI*Math.floor((e+Math.PI)/(2*Math.PI))}W.normalizeRadians=Xt;function ro(e,t){const n=Math.PI/2-Math.atan2(-(t[1]-e[1]),t[0]-e[0]);return Xt(n)}W.computeRotation=ro;const Vt=(e,t)=>[[1,0,e],[0,1,t],[0,0,1]];function $(e,t){let n=0;for(let o=0;o<e.length;o++)n+=e[o]*t[o];return n}W.dot=$;function Qt(e,t){const n=[];for(let o=0;o<e.length;o++)n.push(e[o][t]);return n}W.getColumnFrom2DArr=Qt;function Zt(e,t){const n=[],o=e.length;for(let i=0;i<o;i++){n.push([]);for(let s=0;s<o;s++)n[i].push($(e[i],Qt(t,s)))}return n}function ao(e,t){const n=Math.cos(e),o=Math.sin(e),i=[[n,-o,0],[o,n,0],[0,0,1]],s=Vt(t[0],t[1]),r=Zt(s,i),c=Vt(-t[0],-t[1]);return Zt(r,c)}W.buildRotationMatrix=ao;function co(e){const t=[[e[0][0],e[1][0]],[e[0][1],e[1][1]]],n=[e[0][2],e[1][2]],o=[-$(t[0],n),-$(t[1],n)];return[t[0].concat(o[0]),t[1].concat(o[1]),[0,0,1]]}W.invertTransformMatrix=co;function lo(e,t){return[$(e,t[0]),$(e,t[1])]}W.rotatePoint=lo});var n0=y(Jt=>{const e0=require("@tensorflow/tfjs"),O=De(),U=$t(),ho=.8,uo=[0,-.4],mo=[0,-.1],po=1.65,t0=[0,5,9,13,17,1,2],fo=0,go=2;class bo{constructor(e,t,n){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=e,this.meshDetector=t,this.meshWidth=n.inputSize,this.meshHeight=n.inputSize,this.enlargeFactor=n.enlargeFactor}getBoxForPalmLandmarks(e,t){const n=e.map(i=>{const s=[...i,1];return U.rotatePoint(s,t)}),o=this.calculateLandmarksBoundingBox(n);return O.enlargeBox(O.squarifyBox(O.shiftBox(o,uo)),this.enlargeFactor)}getBoxForHandLandmarks(e){const t=this.calculateLandmarksBoundingBox(e),n=O.enlargeBox(O.squarifyBox(O.shiftBox(t,mo)),po),o=[];for(let i=0;i<t0.length;i++)o.push(e[t0[i]].slice(0,2));return n.palmLandmarks=o,n}transformRawCoords(e,t,n,o){const i=O.getBoxSize(t),s=[i[0]/this.meshWidth,i[1]/this.meshHeight],r=e.map(h=>[s[0]*(h[0]-this.meshWidth/2),s[1]*(h[1]-this.meshHeight/2),h[2]]),c=U.buildRotationMatrix(n,[0,0]),a=r.map(h=>{const p=U.rotatePoint(h,c);return[...p,h[2]]}),d=U.invertTransformMatrix(o),u=[...O.getBoxCenter(t),1],l=[U.dot(u,d[0]),U.dot(u,d[1])];return a.map(h=>[h[0]+l[0],h[1]+l[1],h[2]])}async estimateHands(e,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const n=this.shouldUpdateRegionsOfInterest();if(n===!0){const i=await this.boundingBoxDetector.estimateHandBounds(e,t);this.regionsOfInterest=[];for(const s in i)this.updateRegionsOfInterest(i[s],!0,s);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const o=[];if(!this.regionsOfInterest)return o;for(const i in this.regionsOfInterest){const s=this.regionsOfInterest[i][0];if(!s)return o;const r=U.computeRotation(s.palmLandmarks[fo],s.palmLandmarks[go]),c=O.getBoxCenter(s),a=[c[0]/e.shape[2],c[1]/e.shape[1]],d=e0.image.rotateWithOffset(e,r,0,a),u=U.buildRotationMatrix(-r,c),l=n?this.getBoxForPalmLandmarks(s.palmLandmarks,u):s,h=O.cutBoxFromImageAndResize(l,d,[this.meshWidth,this.meshHeight]),p=h.div(255);h.dispose(),d.dispose();const m=this.meshDetector.predict(p),[w,E]=m;p.dispose();const P=w.dataSync()[0];if(w.dispose(),P<t.minConfidence)return E.dispose(),this.regionsOfInterest[i]=[],o;const x=e0.reshape(E,[-1,3]),N=await x.array();E.dispose(),x.dispose();const C=this.transformRawCoords(N,l,r,u),J=this.getBoxForHandLandmarks(C);this.updateRegionsOfInterest(J,!1,i);const X={landmarks:C,confidence:P,box:{topLeft:J.startPoint,bottomRight:J.endPoint}};o.push(X)}return o}calculateLandmarksBoundingBox(e){const t=e.map(s=>s[0]),n=e.map(s=>s[1]),o=[Math.min(...t),Math.min(...n)],i=[Math.max(...t),Math.max(...n)];return{startPoint:o,endPoint:i}}updateRegionsOfInterest(e,t,n){if(t)this.regionsOfInterest[n]=[e];else{const o=this.regionsOfInterest[n][0];let i=0;if(o!=null&&o.startPoint!=null){const[s,r]=e.startPoint,[c,a]=e.endPoint,[d,u]=o.startPoint,[l,h]=o.endPoint,p=Math.max(s,d),m=Math.max(r,u),w=Math.min(c,l),E=Math.min(a,h),P=(w-p)*(E-m),x=(c-s)*(a-r),N=(l-d)*(h-r);i=P/(x+N-P)}this.regionsOfInterest[n][0]=i>ho?o:e}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}Jt.HandPipeline=bo});var i0=y(Fe=>{const G=require("@tensorflow/tfjs"),yo=Kt(),o0=Gt(),xo=n0();class s0{constructor(e){this.pipeline=e}async estimateHands(e,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const n=G.tidy(()=>(e instanceof G.Tensor||(e=G.browser.fromPixels(e)),e.toFloat().expandDims(0))),o=await this.pipeline.estimateHands(n,t);n.dispose();const i=[];if(!o)return i;for(const s of o){if(!s)return[];const r={};for(const c of Object.keys(o0.MESH_ANNOTATIONS))r[c]=o0.MESH_ANNOTATIONS[c].map(a=>s.landmarks[a]);i.push({confidence:s.confidence||0,box:s.box?[s.box.topLeft[0],s.box.topLeft[1],s.box.bottomRight[0]-s.box.topLeft[0],s.box.bottomRight[1]-s.box.topLeft[1]]:0,landmarks:s.landmarks,annotations:r})}return i}}Fe.HandPose=s0;async function wo(e){if(G.env().features.IS_NODE){const t=require("fs"),n=await t.readFileSync(e.replace("file://",""));return JSON.parse(n)}return G.util.fetch(e).then(t=>t.json())}async function Po(e){const[t,n,o]=await Promise.all([wo(e.detector.anchors),G.loadGraphModel(e.detector.modelPath,{fromTFHub:e.detector.modelPath.includes("tfhub.dev")}),G.loadGraphModel(e.skeleton.modelPath,{fromTFHub:e.skeleton.modelPath.includes("tfhub.dev")})]),i=new yo.HandDetector(n,t,e),s=new xo.HandPipeline(i,o,e),r=new s0(s);return r}Fe.load=Po});var r0=y(Eo=>{Ke(Eo,{default:()=>Mo});var Mo={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,minConfidence:.8,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:10,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:10,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var c0=y((ns,a0)=>{a0.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});var f0=y(z=>{const A=require("@tensorflow/tfjs"),d0=lt(),he=ut(),l0=ft(),h0=Wt(),u0=i0(),m0=r0().default,Io=c0();let g;const M={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},v=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),ue=(...e)=>{g.console&&console.log(...e)};function p0(...e){const t=n=>n&&typeof n=="object";return e.reduce((n,o)=>(Object.keys(o||{}).forEach(i=>{const s=n[i],r=o[i];Array.isArray(s)&&Array.isArray(r)?n[i]=s.concat(...r):t(s)&&t(r)?n[i]=p0(s,r):n[i]=r}),n),{})}function So(e){if(!e)return"input is not defined";const t=e.naturalWidth||e.videoWidth||e.width||e.shape&&e.shape[1]>0;if(!t||t===0)return"input is empty";if(e.readyState&&e.readyState<=2)return"input is not ready";try{A.getBackend()}catch{return"backend not loaded"}return null}async function Bo(e,t){g=p0(m0,t);const n=So(e);return n?(ue(n,e),{error:n}):new Promise(async o=>{const i=Object.values(M).filter(u=>u).length;i===0&&ue("Human library starting"),A.getBackend()!==g.backend&&(ue("Human library setting backend:",g.backend),await A.setBackend(g.backend),await A.ready()),g.face.enabled&&!M.facemesh&&(M.facemesh=await d0.load(g.face)),g.body.enabled&&!M.posenet&&(M.posenet=await h0.load(g.body)),g.hand.enabled&&!M.handpose&&(M.handpose=await u0.load(g.hand)),g.face.enabled&&g.face.age.enabled&&!M.age&&(M.age=await he.loadAge(g)),g.face.enabled&&g.face.gender.enabled&&!M.gender&&(M.gender=await he.loadGender(g)),g.face.enabled&&g.face.emotion.enabled&&!M.emotion&&(M.emotion=await l0.load(g));const s={};let r;r=v(),A.engine().startScope();const c=g.body.enabled?await M.posenet.estimatePoses(e,g.body):[];A.engine().endScope(),s.body=Math.trunc(v()-r),r=v(),A.engine().startScope();const a=g.hand.enabled?await M.handpose.estimateHands(e,g.hand):[];A.engine().endScope(),s.hand=Math.trunc(v()-r);const d=[];if(g.face.enabled){r=v(),A.engine().startScope();const u=await M.facemesh.estimateFaces(e,g.face);s.face=Math.trunc(v()-r);for(const l of u){if(!l.image||l.image.isDisposedInternal){ue("face object is disposed:",l.image);continue}r=v();const h=g.face.age.enabled||g.face.gender.enabled?await he.predict(l.image,g):{};s.agegender=Math.trunc(v()-r),r=v();const p=g.face.emotion.enabled?await l0.predict(l.image,g):{};s.emotion=Math.trunc(v()-r),l.image.dispose();const m=l.annotations.leftEyeIris&&l.annotations.rightEyeIris?Math.max(l.annotations.leftEyeIris[3][0]-l.annotations.leftEyeIris[1][0],l.annotations.rightEyeIris[3][0]-l.annotations.rightEyeIris[1][0]):0;d.push({confidence:l.confidence,box:l.box,mesh:l.mesh,annotations:l.annotations,age:h.age,gender:h.gender,agConfidence:h.confidence,emotion:p,iris:m!==0?Math.trunc(100*11.7/m)/100:0})}A.engine().endScope()}s.total=Object.values(s).reduce((u,l)=>u+l),o({face:d,body:c,hand:a,performance:s})})}z.detect=Bo;z.defaults=m0;z.config=g;z.models=M;z.facemesh=d0;z.ssrnet=he;z.posenet=h0;z.handpose=u0;z.tf=A;z.version=Io.version});export default f0();
 //# sourceMappingURL=human.esm-nobundle.js.map
diff --git a/dist/human.esm-nobundle.js.map b/dist/human.esm-nobundle.js.map
index 1d1f90ef..7b0d8bfb 100644
--- a/dist/human.esm-nobundle.js.map
+++ b/dist/human.esm-nobundle.js.map
@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../src/facemesh/blazeface.js", "../src/facemesh/keypoints.js", "../src/facemesh/box.js", "../src/facemesh/util.js", "../src/facemesh/pipeline.js", "../src/facemesh/uvcoords.js", "../src/facemesh/triangulation.js", "../src/facemesh/facemesh.js", "../src/ssrnet/ssrnet.js", "../src/emotion/emotion.js", "../src/posenet/modelBase.js", "../src/posenet/modelMobileNet.js", "../src/posenet/heapSort.js", "../src/posenet/buildParts.js", "../src/posenet/keypoints.js", "../src/posenet/vectors.js", "../src/posenet/decodePose.js", "../src/posenet/decodeMultiple.js", "../src/posenet/util.js", "../src/posenet/modelPoseNet.js", "../src/posenet/posenet.js", "../src/handpose/box.js", "../src/handpose/handdetector.js", "../src/handpose/keypoints.js", "../src/handpose/util.js", "../src/handpose/pipeline.js", "../src/handpose/handpose.js", "../src/config.js", "../src/index.js"],
-  "sourcesContent": ["const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.age.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    obj.gender = Math.trunc(100 * data[0]) < 50 ? 'female' : 'male';\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "export default {\n  backend: 'webgl',\n  console: true,\n  face: {\n    enabled: true, // refers to detector, but since all other face modules rely on detector, it should be a global\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be blazeface-front or blazeface-back\n      inputSize: 256, // fixed value: 128 for front and tfhub and 256 for back\n      maxFaces: 10, // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10, // how many frames to go without running the bounding box detector\n      minConfidence: 0.5, // threshold for discarding a prediction\n      iouThreshold: 0.3, // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7, // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192, // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3, // empiric tuning\n      inputSize: 64, // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json',\n      inputSize: 64, // fixed value\n      skipFrames: 10,\n    },\n    gender: {\n      enabled: true,\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64, // fixed value\n      minConfidence: 0.5,\n      skipFrames: 10,\n      useGrayscale: true,\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257, // fixed value\n    outputStride: 16, // fixed value\n    maxDetections: 5,\n    scoreThreshold: 0.7,\n    nmsRadius: 20,\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256, // fixed value\n    skipFrames: 10,\n    minConfidence: 0.5,\n    iouThreshold: 0.3,\n    scoreThreshold: 0.7,\n    enlargeFactor: 1.65, // empiric tuning\n    maxHands: 2,\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    let savedWebglPackDepthwiseConvFlag;\n    if (tf.getBackend() === 'webgl') {\n      savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n      tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
-  "mappings": "mMAAA,mBAAM,GAAK,4BAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,EAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,EAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,EAAG,IAAI,EAAI,WAAY,GAChC,EAAO,EAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,EAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,EAAG,IAAI,EAAW,GAC5B,EAAW,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,EAAG,IAAI,EAAU,GACtC,EAAoB,EAAG,IAAI,EAAS,GACpC,EAAc,EAAG,IAAI,EAAoB,GACzC,EAAS,EAAG,IAAI,EAAmB,GACnC,EAAO,EAAG,IAAI,EAAmB,GACjC,EAAkB,EAAG,IAAI,EAAQ,GACjC,EAAgB,EAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,GAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,GAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,EAAG,SAAS,KAAK,aAChC,KAAK,UAAY,EAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,EAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,EAAG,IAAI,EAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,EAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,EAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,EAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,EAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,GAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,EAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,EAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,EACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,EAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,EAAG,KAAK,IACpB,CAAM,YAAiB,GAAG,QACxB,GAAQ,EAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,iBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,kBAAM,IAAK,4BAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,EAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,EAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,EAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,EAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,EAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,EAAQ,YAAc,KClDtB,eAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,EAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,EAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,EAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,WAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,EAAQ,IAAM,EACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,EAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,EAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,EAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,EAAI,EAAkB,GAAI,GAC3B,CAAC,EAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,EAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,EAAI,EAAuB,EAAe,IAC1C,EAAI,EAAuB,EAAe,KAG9C,EAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,EAAQ,wBAA0B,KCvFlC,cACA,KAAM,GAAK,4BACL,EAAW,KACX,EAAY,KACZ,EAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,EAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,EAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,EAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,EAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,EAAU,yBAAyB,GACtD,EAAkB,EAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,EAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,EAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,EAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,EAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,EAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,EAAK,IAAI,EAAW,EAAsB,IAC1C,EAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,EAAS,YAAY,EAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,EAAS,WAAW,GACpC,GAAI,GAAO,EAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,EAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,EAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,EAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,EAAS,oBAAoB,EAAe,GACxD,EAAc,EAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,EAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,EAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,EAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,EAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,EAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,EAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,EAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,EAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,GAAgB,KAAM,IAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,GAAa,QAAS,GAAiB,KAAM,IAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,GAAkB,KAAK,UAAU,QAAQ,EAAG,OAAO,CAAC,GAAa,MACjE,GAAqB,GAAe,WAC1C,GAAe,UACf,KAAM,IAAc,GAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,GAAkB,KAAM,IAAsB,KAAK,aAAa,GAAa,EAAY,GAAgB,IACtH,GAAe,GAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,GAAmB,KAAM,IAAuB,KAAK,aAAa,GAAc,GAAa,IAC1G,GAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,IAAiC,GAC5C,IAAsB,EAAW,GAAkB,QACnD,GAAsB,EAAW,GAAmB,UAE/C,AAAI,GAAgC,EACzC,GAAsB,EAAW,GAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,GAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,IAAyB,KAAK,sBAAsB,EAAW,GAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,GAAoB,SAC1F,EAAY,EAAU,OAAO,IAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,EAAG,QAAQ,GACX,KAAM,GAAe,EAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,EAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,IAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,IAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,iBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,yCAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,mBAAM,GAAK,4BACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,EAAG,KAAK,IACpB,CAAM,YAAiB,GAAG,QAAS,GAAQ,EAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,EAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,EAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,EAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,EAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,mBAAM,GAAK,4BAEL,EAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,EAAG,QAAQ,WAAW,GAC/B,EAAS,EAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,EAAG,KAAK,EAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,GAAO,KAAK,GAAO,IAAM,KAAM,GAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,EAAO,IAGhB,kBAA0B,GACxB,MAAK,GAAO,QAAQ,GAAO,OAAS,KAAM,GAAG,eAAe,EAAO,KAAK,OAAO,YACxE,EAAO,OAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,IAAI,WAC3B,UAAQ,EACD,GAET,GAAI,GACJ,GAAI,YAAiB,GAAG,QACtB,KAAM,GAAS,EAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,EAAG,IAAI,EAAQ,CAAC,MAC1B,EAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,GAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,EAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,GAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OAC3B,EAAI,OAAS,KAAK,MAAM,IAAM,EAAK,IAAM,GAAK,SAAW,OACzD,EAAG,QAAQ,GAEb,SAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KCzDrB,mBAAM,GAAK,4BAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,EAAG,KAAK,KACrB,KAAM,GAAS,EAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,EAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,EAAG,KAAK,EAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,GAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,EAAG,KAAK,KACtB,GAAI,YAAiB,GAAG,QACtB,KAAM,GAAS,EAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,EAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,EAAG,IAAI,EAAG,CAAC,QAChB,EAAK,EAAG,IAAI,EAAG,CAAC,OAChB,EAAK,EAAG,IAAI,EAAG,CAAC,OAChB,EAAY,EAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,EAAG,QAAQ,GAEb,SAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,mBAAM,IAAK,4BAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,mBAAM,IAAK,4BACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,cACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,mBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,eAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,EAAQ,cAAgB,EAAQ,UAAU,OAC1C,EAAQ,QAAU,EAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,EAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,EAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,EAAQ,QAAQ,GAAa,EAAQ,QAAQ,KACnI,EAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,kBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,EAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,EAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,EAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,EAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,EAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,EAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,EAAQ,YAAc,KCnDtB,mBAAM,IAAY,KACZ,EAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,EAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,EAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,EAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,EAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,EAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,EAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,mBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,kBAAM,GAAK,4BACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,EAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,EAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,EAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,EAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,EAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,EAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,GAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,EAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,GAAG,OAAS,EAAQ,EAAG,QAAQ,WAAW,GAEpE,EAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,GAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,EAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,EAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,EAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,EAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,EAAQ,kBAAoB,KCrH5B,mBAAM,IAAK,4BACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,kBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,EAAQ,KAAO,GAAa,KAC5B,EAAQ,QAAU,GAAa,QAE/B,EAAQ,UAAY,GAAe,UACnC,EAAQ,oBAAsB,GAAe,oBAC7C,EAAQ,aAAe,GAAU,aACjC,EAAQ,QAAU,GAAU,QAC5B,EAAQ,UAAY,GAAU,UAC9B,EAAQ,UAAY,GAAU,UAC9B,EAAQ,qBAAuB,GAAK,qBACpC,EAAQ,eAAiB,GAAK,eAC9B,EAAQ,qBAAuB,GAAK,qBACpC,EAAQ,kBAAoB,GAAK,kBACjC,EAAQ,UAAY,GAAK,YCnBzB,kBAAM,IAAK,4BAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,EAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,EAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,EAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,EAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,EAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,EAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,EAAQ,SAAW,KCtEnB,mBAAM,GAAK,4BACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,EAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,EAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,EAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,GAAG,KAAK,KACb,KAAM,GAAa,EAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,EAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,EAAG,IAAI,EAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,EAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,EAAG,IAAI,EAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,EAAG,IAAI,EAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,GAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,GAAG,KAAK,KACb,KAAM,GAAY,EAAG,IAAI,EAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,GAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,EAAG,KAAK,IAAM,EAAG,IAAI,EAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,EAAG,KAAK,IAAM,EAAG,QAAQ,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,GAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,EAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,EAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,EAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,EAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,EAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,iBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,yBAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,EAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,EAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,WAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,EAAQ,IAAM,EAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,EAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,EAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,EAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,EAAI,EAAkB,GAAI,GAC3B,CAAC,EAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,EAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,EAAI,EAAuB,EAAe,IAC1C,EAAI,EAAuB,EAAe,KAG9C,EAAQ,YAAc,KCzEtB,mBAAM,IAAK,4BACL,EAAW,KACX,EAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,GAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,GAAS,WAAW,EAAS,YAAY,EAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,EAAS,WAAW,EAAS,YAAY,EAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,EAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,EAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,EAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,EAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,EAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,EAAK,IAAI,EAAW,EAAsB,IAC1C,EAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,EAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,EAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,EAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,EAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,mBAAM,GAAK,4BACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,EAAG,KAAK,IACpB,CAAM,YAAiB,GAAG,QACxB,GAAQ,EAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,EAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,GAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,EAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,EAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,yCAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GACT,SAAU,CACR,UAAW,sCACX,UAAW,IACX,SAAU,GACV,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCACX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,EACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,EACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,4vECnEjB,kBAAM,GAAK,4BACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,GAGJ,KAAM,GAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,EAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,EAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,EAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,EAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,GAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,EAAG,eAAiB,EAAO,SAC7B,IAAI,iCAAkC,EAAO,SAC7C,KAAM,GAAG,WAAW,EAAO,SAC3B,KAAM,GAAG,SAGX,GAAI,GACJ,AAAI,EAAG,eAAiB,SACtB,GAAkC,EAAG,MAAM,IAAI,4BAC/C,EAAG,MAAM,IAAI,2BAA4B,KAI3C,AAAI,EAAO,KAAK,SAAW,CAAC,EAAO,UAAU,GAAO,SAAW,KAAM,IAAS,KAAK,EAAO,OAC1F,AAAI,EAAO,KAAK,SAAW,CAAC,EAAO,SAAS,GAAO,QAAU,KAAM,IAAQ,KAAK,EAAO,OACvF,AAAI,EAAO,KAAK,SAAW,CAAC,EAAO,UAAU,GAAO,SAAW,KAAM,IAAS,KAAK,EAAO,OAC1F,AAAI,EAAO,KAAK,SAAW,EAAO,KAAK,IAAI,SAAW,CAAC,EAAO,KAAK,GAAO,IAAM,KAAM,IAAO,QAAQ,IACrG,AAAI,EAAO,KAAK,SAAW,EAAO,KAAK,OAAO,SAAW,CAAC,EAAO,QAAQ,GAAO,OAAS,KAAM,IAAO,WAAW,IACjH,AAAI,EAAO,KAAK,SAAW,EAAO,KAAK,QAAQ,SAAW,CAAC,EAAO,SAAS,GAAO,QAAU,KAAM,IAAQ,KAAK,IAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,IACZ,EAAG,SAAS,aACZ,KAAM,GAAU,EAAO,KAAK,QAAU,KAAM,GAAO,QAAQ,cAAc,EAAO,EAAO,MAAQ,GAC/F,EAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,IAAQ,GAG/B,EAAY,IACZ,EAAG,SAAS,aACZ,KAAM,GAAU,EAAO,KAAK,QAAU,KAAM,GAAO,SAAS,cAAc,EAAO,EAAO,MAAQ,GAChG,EAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,IAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,EAAO,KAAK,SACd,EAAY,IACZ,EAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,GAAO,SAAS,cAAc,EAAO,EAAO,MAChE,EAAK,KAAO,KAAK,MAAM,IAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,IACZ,KAAM,GAAW,EAAO,KAAK,IAAI,SAAW,EAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,GAAU,GACrH,EAAK,UAAY,KAAK,MAAM,IAAQ,GAEpC,EAAY,IACZ,KAAM,GAAc,EAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,GAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,IAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,EAAG,SAAS,WAId,EAAG,MAAM,IAAI,2BAA4B,GAGzC,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,EAAQ,OAAS,GACjB,EAAQ,SAAW,GACnB,EAAQ,OAAS,EACjB,EAAQ,OAAS,EACjB,EAAQ,SAAW,GACnB,EAAQ,OAAS,GACjB,EAAQ,QAAU,GAClB,EAAQ,SAAW,GACnB,EAAQ,GAAK,EACb,EAAQ,QAAU,GAAI",
+  "sourcesContent": ["const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  if (frame > config.face.age.skipFrames) {\n    frame = 0;\n  } else {\n    frame += 1;\n  }\n  if (frame === 0) return last;\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;\n    if (confidence > config.face.gender.minConfidence) {\n      obj.gender = data[0] <= 0.5 ? 'female' : 'male';\n      obj.confidence = confidence;\n    }\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "/* eslint-disable indent */\n/* eslint-disable no-multi-spaces */\n\nexport default {\n  backend: 'webgl',          // select tfjs backend to use\n  console: true,             // enable debugging output to console\n  face: {\n    enabled: true,           // controls if specified modul is enabled\n                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion\n                             // (note: module is not loaded until it is required)\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.\n                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.\n      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'\n      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192,        // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3,    // empiric tuning\n      inputSize: 64,         // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'\n                                                         // which determines training set for model\n      inputSize: 64,         // fixed value\n      skipFrames: 10,        // how many frames to go without re-running the detector\n    },\n    gender: {\n      enabled: true,\n      minConfidence: 0.8,    // threshold for discarding a prediction\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64,         // fixed value\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      skipFrames: 10,        // how many frames to go without re-running the detector\n      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257,    // fixed value\n    outputStride: 16,        // fixed value\n    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256,          // fixed value\n    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n    minConfidence: 0.5,      // threshold for discarding a prediction\n    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace\n    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    // let savedWebglPackDepthwiseConvFlag;\n    // if (tf.getBackend() === 'webgl') {\n    //   savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n    //  tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    // }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          agConfidence: ssrData.confidence,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    // tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
+  "mappings": "mMAAA,mBAAM,GAAK,4BAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,EAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,EAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,EAAG,IAAI,EAAI,WAAY,GAChC,EAAO,EAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,EAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,EAAG,IAAI,EAAW,GAC5B,EAAW,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,EAAG,IAAI,EAAU,GACtC,EAAoB,EAAG,IAAI,EAAS,GACpC,EAAc,EAAG,IAAI,EAAoB,GACzC,EAAS,EAAG,IAAI,EAAmB,GACnC,EAAO,EAAG,IAAI,EAAmB,GACjC,EAAkB,EAAG,IAAI,EAAQ,GACjC,EAAgB,EAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,GAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,GAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,EAAG,SAAS,KAAK,aAChC,KAAK,UAAY,EAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,EAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,EAAG,IAAI,EAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,EAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,EAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,EAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,EAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,GAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,EAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,EAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,EACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,EAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,EAAG,KAAK,IACpB,CAAM,YAAiB,GAAG,QACxB,GAAQ,EAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,iBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,kBAAM,IAAK,4BAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,EAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,EAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,EAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,EAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,EAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,EAAQ,YAAc,KClDtB,eAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,EAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,EAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,EAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,WAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,EAAQ,IAAM,EACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,EAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,EAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,EAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,EAAI,EAAkB,GAAI,GAC3B,CAAC,EAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,EAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,EAAI,EAAuB,EAAe,IAC1C,EAAI,EAAuB,EAAe,KAG9C,EAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,EAAQ,wBAA0B,KCvFlC,cACA,KAAM,GAAK,4BACL,EAAW,KACX,EAAY,KACZ,EAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,EAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,EAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,EAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,EAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,EAAU,yBAAyB,GACtD,EAAkB,EAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,EAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,EAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,EAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,EAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,EAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,EAAK,IAAI,EAAW,EAAsB,IAC1C,EAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,EAAS,YAAY,EAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,EAAS,WAAW,GACpC,GAAI,GAAO,EAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,EAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,EAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,EAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,EAAS,oBAAoB,EAAe,GACxD,EAAc,EAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,EAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,EAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,EAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,EAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,EAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,EAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,EAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,EAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,GAAgB,KAAM,IAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,GAAa,QAAS,GAAiB,KAAM,IAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,GAAkB,KAAK,UAAU,QAAQ,EAAG,OAAO,CAAC,GAAa,MACjE,GAAqB,GAAe,WAC1C,GAAe,UACf,KAAM,IAAc,GAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,GAAkB,KAAM,IAAsB,KAAK,aAAa,GAAa,EAAY,GAAgB,IACtH,GAAe,GAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,GAAmB,KAAM,IAAuB,KAAK,aAAa,GAAc,GAAa,IAC1G,GAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,IAAiC,GAC5C,IAAsB,EAAW,GAAkB,QACnD,GAAsB,EAAW,GAAmB,UAE/C,AAAI,GAAgC,EACzC,GAAsB,EAAW,GAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,GAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,IAAyB,KAAK,sBAAsB,EAAW,GAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,GAAoB,SAC1F,EAAY,EAAU,OAAO,IAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,EAAG,QAAQ,GACX,KAAM,GAAe,EAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,EAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,IAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,IAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,iBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,yCAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,mBAAM,GAAK,4BACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,EAAG,KAAK,IACpB,CAAM,YAAiB,GAAG,QAAS,GAAQ,EAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,EAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,EAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,EAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,EAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,mBAAM,GAAK,4BAEL,EAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,EAAG,QAAQ,WAAW,GAC/B,EAAS,EAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,EAAG,KAAK,EAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,GAAO,KAAK,GAAO,IAAM,KAAM,GAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,EAAO,IAGhB,kBAA0B,GACxB,MAAK,GAAO,QAAQ,GAAO,OAAS,KAAM,GAAG,eAAe,EAAO,KAAK,OAAO,YACxE,EAAO,OAGhB,kBAAuB,EAAO,GAM5B,GALA,AAAI,GAAQ,EAAO,KAAK,IAAI,WAC1B,GAAQ,EAER,IAAS,EAEP,KAAU,EAAG,MAAO,IACxB,GAAI,GACJ,GAAI,YAAiB,GAAG,QACtB,KAAM,GAAS,EAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,EAAG,IAAI,EAAQ,CAAC,MAC1B,EAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,GAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,EAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,GAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OACrB,EAAa,KAAK,MAAM,KAAK,IAAI,IAAM,IAAO,GAAK,GAAK,MAAS,IACvE,AAAI,EAAa,EAAO,KAAK,OAAO,eAClC,GAAI,OAAS,EAAK,IAAM,GAAM,SAAW,OACzC,EAAI,WAAa,GAEnB,EAAG,QAAQ,GAEb,SAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KC9DrB,mBAAM,GAAK,4BAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,EAAG,KAAK,KACrB,KAAM,GAAS,EAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,EAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,EAAG,KAAK,EAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,GAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,EAAG,KAAK,KACtB,GAAI,YAAiB,GAAG,QACtB,KAAM,GAAS,EAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,EAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,EAAG,IAAI,EAAG,CAAC,QAChB,EAAK,EAAG,IAAI,EAAG,CAAC,OAChB,EAAK,EAAG,IAAI,EAAG,CAAC,OAChB,EAAY,EAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,EAAG,QAAQ,GAEb,SAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,mBAAM,IAAK,4BAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,mBAAM,IAAK,4BACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,cACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,mBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,eAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,EAAQ,cAAgB,EAAQ,UAAU,OAC1C,EAAQ,QAAU,EAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,EAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,EAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,EAAQ,QAAQ,GAAa,EAAQ,QAAQ,KACnI,EAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,kBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,EAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,EAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,EAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,EAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,EAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,EAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,EAAQ,YAAc,KCnDtB,mBAAM,IAAY,KACZ,EAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,EAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,EAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,EAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,EAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,EAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,EAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,mBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,kBAAM,GAAK,4BACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,EAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,EAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,EAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,EAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,EAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,EAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,GAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,EAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,GAAG,OAAS,EAAQ,EAAG,QAAQ,WAAW,GAEpE,EAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,GAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,EAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,EAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,EAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,EAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,EAAQ,kBAAoB,KCrH5B,mBAAM,IAAK,4BACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,kBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,EAAQ,KAAO,GAAa,KAC5B,EAAQ,QAAU,GAAa,QAE/B,EAAQ,UAAY,GAAe,UACnC,EAAQ,oBAAsB,GAAe,oBAC7C,EAAQ,aAAe,GAAU,aACjC,EAAQ,QAAU,GAAU,QAC5B,EAAQ,UAAY,GAAU,UAC9B,EAAQ,UAAY,GAAU,UAC9B,EAAQ,qBAAuB,GAAK,qBACpC,EAAQ,eAAiB,GAAK,eAC9B,EAAQ,qBAAuB,GAAK,qBACpC,EAAQ,kBAAoB,GAAK,kBACjC,EAAQ,UAAY,GAAK,YCnBzB,kBAAM,IAAK,4BAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,EAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,EAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,EAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,EAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,EAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,EAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,EAAQ,SAAW,KCtEnB,mBAAM,GAAK,4BACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,EAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,EAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,EAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,GAAG,KAAK,KACb,KAAM,GAAa,EAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,EAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,EAAG,IAAI,EAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,EAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,EAAG,IAAI,EAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,EAAG,IAAI,EAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,GAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,GAAG,KAAK,KACb,KAAM,GAAY,EAAG,IAAI,EAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,GAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,EAAG,KAAK,IAAM,EAAG,IAAI,EAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,EAAG,KAAK,IAAM,EAAG,QAAQ,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,EAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,GAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,EAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,EAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,EAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,EAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,EAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,iBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,yBAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,EAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,EAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,WAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,EAAQ,IAAM,EAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,EAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,EAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,EAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,EAAI,EAAkB,GAAI,GAC3B,CAAC,EAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,EAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,EAAI,EAAuB,EAAe,IAC1C,EAAI,EAAuB,EAAe,KAG9C,EAAQ,YAAc,KCzEtB,mBAAM,IAAK,4BACL,EAAW,KACX,EAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,GAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,GAAS,WAAW,EAAS,YAAY,EAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,EAAS,WAAW,EAAS,YAAY,EAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,EAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,EAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,EAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,EAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,EAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,EAAK,IAAI,EAAW,EAAsB,IAC1C,EAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,EAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,EAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,EAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,EAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,mBAAM,GAAK,4BACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,EAAG,KAAK,IACpB,CAAM,YAAiB,GAAG,QACxB,GAAQ,EAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,EAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,GAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,EAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,EAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,sCAGA,GAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GAGT,SAAU,CACR,UAAW,sCAEX,UAAW,IACX,SAAU,GACV,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCAEX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,cAAe,GACf,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,GACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,GACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,4vEC/EjB,kBAAM,GAAK,4BACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,GAGJ,KAAM,GAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,EAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,EAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,EAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,EAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,GAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,EAAG,eAAiB,EAAO,SAC7B,IAAI,iCAAkC,EAAO,SAC7C,KAAM,GAAG,WAAW,EAAO,SAC3B,KAAM,GAAG,SAUX,AAAI,EAAO,KAAK,SAAW,CAAC,EAAO,UAAU,GAAO,SAAW,KAAM,IAAS,KAAK,EAAO,OAC1F,AAAI,EAAO,KAAK,SAAW,CAAC,EAAO,SAAS,GAAO,QAAU,KAAM,IAAQ,KAAK,EAAO,OACvF,AAAI,EAAO,KAAK,SAAW,CAAC,EAAO,UAAU,GAAO,SAAW,KAAM,IAAS,KAAK,EAAO,OAC1F,AAAI,EAAO,KAAK,SAAW,EAAO,KAAK,IAAI,SAAW,CAAC,EAAO,KAAK,GAAO,IAAM,KAAM,IAAO,QAAQ,IACrG,AAAI,EAAO,KAAK,SAAW,EAAO,KAAK,OAAO,SAAW,CAAC,EAAO,QAAQ,GAAO,OAAS,KAAM,IAAO,WAAW,IACjH,AAAI,EAAO,KAAK,SAAW,EAAO,KAAK,QAAQ,SAAW,CAAC,EAAO,SAAS,GAAO,QAAU,KAAM,IAAQ,KAAK,IAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,IACZ,EAAG,SAAS,aACZ,KAAM,GAAU,EAAO,KAAK,QAAU,KAAM,GAAO,QAAQ,cAAc,EAAO,EAAO,MAAQ,GAC/F,EAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,IAAQ,GAG/B,EAAY,IACZ,EAAG,SAAS,aACZ,KAAM,GAAU,EAAO,KAAK,QAAU,KAAM,GAAO,SAAS,cAAc,EAAO,EAAO,MAAQ,GAChG,EAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,IAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,EAAO,KAAK,SACd,EAAY,IACZ,EAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,GAAO,SAAS,cAAc,EAAO,EAAO,MAChE,EAAK,KAAO,KAAK,MAAM,IAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,IACZ,KAAM,GAAW,EAAO,KAAK,IAAI,SAAW,EAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,GAAU,GACrH,EAAK,UAAY,KAAK,MAAM,IAAQ,GAEpC,EAAY,IACZ,KAAM,GAAc,EAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,GAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,IAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,aAAc,EAAQ,WACtB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,EAAG,SAAS,WAOd,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,EAAQ,OAAS,GACjB,EAAQ,SAAW,GACnB,EAAQ,OAAS,EACjB,EAAQ,OAAS,EACjB,EAAQ,SAAW,GACnB,EAAQ,OAAS,GACjB,EAAQ,QAAU,GAClB,EAAQ,SAAW,GACnB,EAAQ,GAAK,EACb,EAAQ,QAAU,GAAI",
   "names": []
 }
diff --git a/dist/human.esm.js b/dist/human.esm.js
index 3e818b73..64cb7fad 100644
--- a/dist/human.esm.js
+++ b/dist/human.esm.js
@@ -1,39 +1,39 @@
-var Np=Object.defineProperty;var be=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),CI=n=>Np(n,"__esModule",{value:!0}),xp=(n,t)=>{CI(n);for(var e in t)Np(n,e,{get:t[e],enumerable:!0})};var Cp=be(()=>{});var Rp=be(()=>{});var ds=be(()=>{});var Xi=be(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var el=function(n,t){return el=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},el(n,t)};function qn(n,t){el(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var RI=1e-7,OI=1e-4,EI=function(){function n(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}return n.prototype.get=function(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)},n.prototype.set=function(t,e){this.dataIdsCount++,this.data.set(t,e)},n.prototype.has=function(t){return this.data.has(t)},n.prototype.delete=function(t){return this.dataIdsCount--,this.data.delete(t)},n.prototype.numDataIds=function(){return this.dataIdsCount},n}(),Op=function(){function n(){}return n.prototype.time=function(t){return $("time")},n.prototype.read=function(t){return $("read")},n.prototype.readSync=function(t){return $("readSync")},n.prototype.numDataIds=function(){return $("numDataIds")},n.prototype.disposeData=function(t){return $("disposeData")},n.prototype.write=function(t,e,i){return $("write")},n.prototype.move=function(t,e,i,r){return $("move")},n.prototype.memory=function(){return $("memory")},n.prototype.floatPrecision=function(){return $("floatPrecision")},n.prototype.epsilon=function(){return this.floatPrecision()===32?RI:OI},n.prototype.batchMatMul=function(t,e,i,r){return $("batchMatMul")},n.prototype.fusedBatchMatMul=function(t){var e=t.a,i=t.b,r=t.transposeA,a=t.transposeB,s=t.bias,o=t.activation,l=t.preluActivationWeights;return $("fusedBatchMatMul")},n.prototype.slice=function(t,e,i){return $("slice")},n.prototype.stridedSlice=function(t,e,i,r){return $("stridedSlice")},n.prototype.unstack=function(t,e){return $("unstack")},n.prototype.reverse=function(t,e){return $("reverse")},n.prototype.concat=function(t,e){return $("concat")},n.prototype.neg=function(t){return $("neg")},n.prototype.add=function(t,e){return $("add")},n.prototype.addN=function(t){return $("addN")},n.prototype.subtract=function(t,e){return $("subtract")},n.prototype.multiply=function(t,e){return $("multiply")},n.prototype.realDivide=function(t,e){return $("realDivide")},n.prototype.floorDiv=function(t,e){return $("floorDiv")},n.prototype.sum=function(t,e){return $("sum")},n.prototype.prod=function(t,e){return $("prod")},n.prototype.unsortedSegmentSum=function(t,e,i){return $("unsortedSegmentSum")},n.prototype.argMin=function(t,e){return $("argMin")},n.prototype.argMax=function(t,e){return $("argMax")},n.prototype.equal=function(t,e){return $("equal")},n.prototype.notEqual=function(t,e){return $("notEqual")},n.prototype.less=function(t,e){return $("less")},n.prototype.lessEqual=function(t,e){return $("lessEqual")},n.prototype.greater=function(t,e){return $("greater")},n.prototype.greaterEqual=function(t,e){return $("greaterEqual")},n.prototype.logicalNot=function(t){return $("logicalNot")},n.prototype.logicalAnd=function(t,e){return $("logicalAnd")},n.prototype.logicalOr=function(t,e){return $("logicalOr")},n.prototype.where=function(t){return $("where")},n.prototype.select=function(t,e,i){return $("select")},n.prototype.topk=function(t,e,i){return $("topk")},n.prototype.min=function(t,e){return $("min")},n.prototype.minimum=function(t,e){return $("minimum")},n.prototype.mod=function(t,e){return $("mod")},n.prototype.max=function(t,e){return $("max")},n.prototype.maximum=function(t,e){return $("maximum")},n.prototype.all=function(t,e){return $("all")},n.prototype.any=function(t,e){return $("any")},n.prototype.squaredDifference=function(t,e){return $("squaredDifference")},n.prototype.ceil=function(t){return $("ceil")},n.prototype.floor=function(t){return $("floor")},n.prototype.round=function(t){return $("round")},n.prototype.sign=function(t){return $("sign")},n.prototype.isNaN=function(t){return $("isNaN")},n.prototype.isInf=function(t){return $("isInf")},n.prototype.isFinite=function(t){return $("isFinite")},n.prototype.pow=function(t,e){return $("pow")},n.prototype.exp=function(t){return $("exp")},n.prototype.expm1=function(t){return $("expm1")},n.prototype.softmax=function(t,e){return $("softmax")},n.prototype.log=function(t){return $("log")},n.prototype.log1p=function(t){return $("log1p")},n.prototype.sqrt=function(t){return $("sqrt")},n.prototype.rsqrt=function(t){return $("rsqrt")},n.prototype.square=function(t){return $("square")},n.prototype.reciprocal=function(t){return $("reciprocal")},n.prototype.relu=function(t){return $("relu")},n.prototype.relu6=function(t){return $("relu6")},n.prototype.prelu=function(t,e){return $("prelu")},n.prototype.elu=function(t){return $("elu")},n.prototype.eluDer=function(t,e){return $("eluDer")},n.prototype.selu=function(t){return $("selu")},n.prototype.int=function(t){return $("int")},n.prototype.clip=function(t,e,i){return $("clip")},n.prototype.abs=function(t){return $("abs")},n.prototype.complexAbs=function(t){return $("complexAbs")},n.prototype.sigmoid=function(t){return $("sigmoid")},n.prototype.softplus=function(t){return $("softplus")},n.prototype.sin=function(t){return $("sin")},n.prototype.cos=function(t){return $("cos")},n.prototype.tan=function(t){return $("tan")},n.prototype.asin=function(t){return $("asin")},n.prototype.acos=function(t){return $("acos")},n.prototype.atan=function(t){return $("atan")},n.prototype.atan2=function(t,e){return $("atan2")},n.prototype.sinh=function(t){return $("sinh")},n.prototype.cosh=function(t){return $("cosh")},n.prototype.tanh=function(t){return $("tanh")},n.prototype.asinh=function(t){return $("asinh")},n.prototype.acosh=function(t){return $("acosh")},n.prototype.atanh=function(t){return $("atanh")},n.prototype.erf=function(t){return $("erf")},n.prototype.step=function(t,e){return $("step")},n.prototype.fusedConv2d=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedConv2d")},n.prototype.conv2d=function(t,e,i){return $("conv2d")},n.prototype.conv2dDerInput=function(t,e,i){return $("conv2dDerInput")},n.prototype.conv2dDerFilter=function(t,e,i){return $("conv2dDerFilter")},n.prototype.fusedDepthwiseConv2D=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedDepthwiseConv2D")},n.prototype.depthwiseConv2D=function(t,e,i){return $("depthwiseConv2D")},n.prototype.depthwiseConv2DDerInput=function(t,e,i){return $("depthwiseConv2DDerInput")},n.prototype.depthwiseConv2DDerFilter=function(t,e,i){return $("depthwiseConv2DDerFilter")},n.prototype.conv3d=function(t,e,i){return $("conv3d")},n.prototype.conv3dDerInput=function(t,e,i){return $("conv3dDerInput")},n.prototype.conv3dDerFilter=function(t,e,i){return $("conv3dDerFilter")},n.prototype.maxPool=function(t,e){return $("maxPool")},n.prototype.maxPoolBackprop=function(t,e,i,r){return $("maxPoolBackprop")},n.prototype.avgPool=function(t,e){return $("avgPool")},n.prototype.avgPoolBackprop=function(t,e,i){return $("avgPoolBackprop")},n.prototype.avgPool3d=function(t,e){return $("avgPool3d")},n.prototype.avgPool3dBackprop=function(t,e,i){return $("avgPool3dBackprop")},n.prototype.maxPool3d=function(t,e){return $("maxPool3d")},n.prototype.maxPool3dBackprop=function(t,e,i,r){return $("maxPool3dBackprop")},n.prototype.reshape=function(t,e){return $("reshape")},n.prototype.cast=function(t,e){return $("cast")},n.prototype.tile=function(t,e){return $("tile")},n.prototype.pad=function(t,e,i){return $("pad")},n.prototype.transpose=function(t,e){return $("transpose")},n.prototype.gather=function(t,e,i){return $("gather")},n.prototype.gatherND=function(t,e){return $("gatherND")},n.prototype.scatterND=function(t,e,i){return $("scatterND")},n.prototype.batchToSpaceND=function(t,e,i){return $("batchToSpaceND")},n.prototype.spaceToBatchND=function(t,e,i){return $("spaceToBatchND")},n.prototype.resizeBilinear=function(t,e,i,r){return $("resizeBilinear")},n.prototype.resizeBilinearBackprop=function(t,e,i){return $("resizeBilinearBackprop")},n.prototype.resizeNearestNeighbor=function(t,e,i,r){return $("resizeNearestNeighbor")},n.prototype.resizeNearestNeighborBackprop=function(t,e,i){return $("resizeNearestNeighborBackprop")},n.prototype.batchNorm=function(t,e,i,r,a,s){return $("batchNorm")},n.prototype.localResponseNormalization4D=function(t,e,i,r,a){return $("localResponseNormalization4D")},n.prototype.LRNGrad=function(t,e,i,r,a,s,o){return $("LRNGrad")},n.prototype.multinomial=function(t,e,i,r){return $("multinomial")},n.prototype.oneHot=function(t,e,i,r){return $("oneHot")},n.prototype.cumsum=function(t,e,i,r){return $("cumsum")},n.prototype.nonMaxSuppression=function(t,e,i,r,a){return $("nonMaxSuppression")},n.prototype.fft=function(t){return $("fft")},n.prototype.ifft=function(t){return $("ifft")},n.prototype.complex=function(t,e){return $("complex")},n.prototype.real=function(t){return $("real")},n.prototype.imag=function(t){return $("imag")},n.prototype.cropAndResize=function(t,e,i,r,a,s){return $("cropAndResize")},n.prototype.depthToSpace=function(t,e,i){return $("depthToSpace")},n.prototype.split=function(t,e,i){return $("split")},n.prototype.sparseToDense=function(t,e,i,r){return $("sparseToDense")},n.prototype.diag=function(t){return $("diag")},n.prototype.fill=function(t,e,i){return $("fill")},n.prototype.onesLike=function(t){return $("onesLike")},n.prototype.zerosLike=function(t){return $("zerosLike")},n.prototype.linspace=function(t,e,i){return $("linspace")},n.prototype.dispose=function(){return $("dispose")},n}();function $(n){throw new Error("'"+n+"' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen")}var Ep="tfjsflags",Dp=function(){function n(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}return n.prototype.setPlatform=function(t,e){this.platform!=null&&console.warn("Platform "+this.platformName+" has already been set. "+("Overwriting the platform with "+e+".")),this.platformName=t,this.platform=e},n.prototype.registerFlag=function(t,e,i){if(this.flagRegistry[t]={evaluationFn:e,setHook:i},this.urlFlags[t]!=null){var r=this.urlFlags[t];console.warn("Setting feature override from URL "+t+": "+r+"."),this.set(t,r)}},n.prototype.getAsync=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return t in this.flags?[2,this.flags[t]]:(e=this.flags,i=t,[4,this.evaluateFlag(t)]);case 1:return e[i]=r.sent(),[2,this.flags[t]]}})})},n.prototype.get=function(t){if(t in this.flags)return this.flags[t];var e=this.evaluateFlag(t);if(e instanceof Promise)throw new Error("Flag "+t+" cannot be synchronously evaluated. Please use getAsync() instead.");return this.flags[t]=e,this.flags[t]},n.prototype.getNumber=function(t){return this.get(t)},n.prototype.getBool=function(t){return this.get(t)},n.prototype.getFlags=function(){return this.flags},Object.defineProperty(n.prototype,"features",{get:function(){return this.flags},enumerable:!0,configurable:!0}),n.prototype.set=function(t,e){if(this.flagRegistry[t]==null)throw new Error("Cannot set flag "+t+" as it has not been registered.");this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)},n.prototype.evaluateFlag=function(t){if(this.flagRegistry[t]==null)throw new Error("Cannot evaluate flag '"+t+"': no evaluation function found.");return this.flagRegistry[t].evaluationFn()},n.prototype.setFlags=function(t){this.flags=Object.assign({},t)},n.prototype.reset=function(){this.flags={},this.urlFlags={},this.populateURLFlags()},n.prototype.populateURLFlags=function(){var t=this;if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;var e=DI(this.global.location.search);if(Ep in e){var i=e[Ep].split(",");i.forEach(function(r){var a=r.split(":"),s=a[0],o=a[1];t.urlFlags[s]=kI(s,o)})}},n}();function DI(n){var t={};return n.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,function(e){for(var i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];return FI(t,i[0],i[1]),i.join("=")}),t}function FI(n,t,e){n[decodeURIComponent(t)]=decodeURIComponent(e||"")}function kI(n,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(""+ +t===t)return+t;throw new Error("Could not parse value flag value "+t+" for flag "+n+".")}function He(){return I.ENV}I.ENV=null;function WI(n){I.ENV=n}var tl;function kp(){if(tl==null){var n=void 0;if(typeof window!="undefined")n=window;else if(typeof global!="undefined")n=global;else if(typeof process!="undefined")n=process;else if(typeof self!="undefined")n=self;else throw new Error("Could not find a global object");tl=n}return tl}function UI(){var n=kp();return n._tfGlobals==null&&(n._tfGlobals=new Map),n._tfGlobals}function Fp(n,t){var e=UI();if(e.has(n))return e.get(n);var i=t();return e.set(n,i),e.get(n)}var nl="Abs",il="Acos",rl="Acosh",ps="Add",al="AddN",Wp="All",Up="Any",sl="ArgMax",ol="ArgMin",ll="Asin",ul="Asinh",cl="Atan",hl="Atanh",dl="Atan2",pl="AvgPool",Bp="AvgPoolBackprop",fl="AvgPool3D",zp="AvgPool3DBackprop",ml="BatchMatMul",gl="BatchToSpaceND",vl="BroadcastTo",fs="Cast",yl="Ceil",bl="ClipByValue",Pp="Complex",wl="Concat",Sl="Conv2D",_p="Conv2DBackpropFilter",Ll="Conv2DBackpropInput",Il="Conv3D",Mp="Conv3DBackpropFilterV2",Hp="Conv3DBackpropInputV2",Al="Cos",Tl="Cosh",Nl="Cumsum",Vp="CropAndResize",qp="DepthToSpace",xl="DepthwiseConv2dNative",Gp="DepthwiseConv2dNativeBackpropFilter",Yp="DepthwiseConv2dNativeBackpropInput",Kp="Diag",Cl="Dilation2D",jp="Dilation2DBackpropInput",$p="Dilation2DBackpropFilter",Rl="Div",Ol="Elu",Xp="EluGrad",El="Erf",Jp="Equal",Dl="Exp",kl="Expm1",Zp="FFT",Qp="Fill",ef="FlipLeftRight",Fl="Floor",Wl="FloorDiv",Ul="FusedBatchNorm",Bl="GatherV2",tf="GatherNd",nf="Greater",zl="GreaterEqual",Pl="Identity",rf="IFFT",af="Imag",_l="IsFinite",Ml="IsInf",Hl="IsNan",sf="Less",of="LessEqual",lf="LinSpace",Vl="Log",ql="Log1p",uf="LogicalAnd",cf="LogicalNot",hf="LogicalOr",Gl="LogSoftmax",Yl="LRN",df="LRNBackprop",Kl="Max",jl="Maximum",$l="MaxPool",pf="MaxPoolBackprop",Xl="MaxPool3D",ff="MaxPool3DBackprop",mf="MaxPoolWithArgmax",BI="Mean",Jl="Min",Zl="Minimum",Ql="Mod",eu="Multiply",tu="Negate",gf="NotEqual",vf="NonMaxSuppressionV3",yf="NonMaxSuppressionV4",bf="NonMaxSuppressionV5",nu="OnesLike",iu="OneHot",ru="PadV2",zI="Pool",au="Pow",su="Prelu",wf="Prod",Sf="Range",Lf="Real",ou="Reciprocal",lu="Relu",uu="Reshape",cu="ResizeNearestNeighbor",If="ResizeNearestNeighborGrad",hu="ResizeBilinear",Af="ResizeBilinearGrad",du="Relu6",pu="Reverse",fu="Round",mu="Rsqrt",Tf="ScatterNd",gu="SelectV2",vu="Selu",yu="Slice",bu="Sin",wu="Sinh",Su="Sign",Lu="Sigmoid",Iu="Softplus",Au="Sqrt",Tu="Sum",Nu="SpaceToBatchND",xu="SplitV",Cu="Softmax",Ru="SquaredDifference",Nf="Square",Ou="Sub",xf="SparseToDense",Cf="StridedSlice",Eu="Tan",Du="Tanh",ku="Tile",Rf="TopK",Fu="Transpose",Of="Unique",Wu="Unpack",Uu="UnsortedSegmentSum",Bu="ZerosLike",zu="Step",Pu="FromPixels",Ef="RotateWithOffset",_u="_FusedMatMul",Mu="FusedConv2D",Hu="FusedDepthwiseConv2D";var br=Fp("kernelRegistry",function(){return new Map}),da=Fp("gradRegistry",function(){return new Map});function qu(n,t){var e=Vu(n,t);return br.get(e)}function Gu(n){return da.get(n)}function ms(n){for(var t=br.entries(),e=[];;){var i=t.next(),r=i.done,a=i.value;if(r)break;var s=a[0],o=a[1],l=s.split("_")[0];l===n&&e.push(o)}return e}function Df(n){var t=n.kernelName,e=n.backendName,i=Vu(t,e);br.has(i)&&console.warn("The kernel '"+t+"' for backend "+("'"+e+"' is already registered")),br.set(i,n)}function kf(n){var t=n.kernelName;da.has(t)&&(He().getBool("DEBUG")&&console.warn("Overriding the gradient for '"+t+"'")),da.set(t,n)}function PI(n,t){var e=Vu(n,t);if(!br.has(e))throw new Error("The kernel '"+n+"' for backend "+("'"+t+"' is not registered"));br.delete(e)}function _I(n){if(!da.has(n))throw new Error("The gradient '"+n+"' for backend is not registered");da.delete(n)}function MI(n,t){var e=ms(n);e.forEach(function(i){var r=Object.assign({},i,{backendName:t});Df(r)})}function Vu(n,t){return t+"_"+n}function Ff(n){for(var t=n.length,e=0,i=0;t>0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function pa(n,t,e){return Math.max(n,Math.min(t,e))}function HI(n){return n%2===0?n:n+1}function VI(n){for(var t=0,e=0;e<n.length;e++)t+=n[e];return t}function qI(n,t){var e=Math.random();return t*e+(1-e)*n}function GI(n,t){for(var e=0,i=0;i<n.length;i++){var r=Number(n[i])-Number(t[i]);e+=r*r}return e}function E(n,t){if(!n)throw new Error(typeof t=="string"?t:t())}function Be(n,t,e){e===void 0&&(e=""),E(ln(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function Fi(n){E(n!=null,function(){return"The input to the tensor constructor must be a non-null value."})}function Wi(n,t,e){if(t===void 0&&(t=[]),e===void 0&&(e=!1),t==null&&(t=[]),Array.isArray(n)||Ot(n)&&!e)for(var i=0;i<n.length;++i)Wi(n[i],t,e);else t.push(n);return t}function ot(n){if(n.length===0)return 1;for(var t=n[0],e=1;e<n.length;e++)t*=n[e];return t}function YI(n){return n.length===0}function ln(n,t){if(n===t)return!0;if(n==null||t==null)return!1;if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==t[e])return!1;return!0}function rt(n){return n%1===0}function KI(n){if(Math.tanh!=null)return Math.tanh(n);if(n===Infinity)return 1;if(n===-Infinity)return-1;var t=Math.exp(2*n);return(t-1)/(t+1)}function jI(n){var t=Math.ceil(Math.sqrt(n));return[t,Math.ceil(n/t)]}function $I(n){for(var t=new Uint32Array(n),e=0;e<n;++e)t[e]=e;return Ff(t),t}function fa(n,t){return t<=n.length?n:n+" ".repeat(t-n.length)}function XI(n,t,e){return t===void 0&&(t=function(i){return 0}),new Promise(function(i,r){var a=0,s=function(){if(n()){i();return}a++;var o=t(a);if(e!=null&&a>=e){r();return}setTimeout(s,o)};s()})}function Wf(n,t){for(var e=1,i=-1,r=0;r<n.length;++r)if(n[r]>=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Qe(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&i<e}),function(){return"All values in axis param must be in range [-"+e+", "+e+") but "+("got axis "+n)}),E(n.every(function(i){return rt(i)}),function(){return"All values in axis param must be integers but "+("got axis "+n)}),n.map(function(i){return i<0?e+i:i})}function Uf(n,t){for(var e=[],i=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Qe(t,n).sort(),s=0,o=0;o<n.length;++o){if(a!=null){if(a[s]===o&&n[o]!==1)throw new Error("Can't squeeze axis "+o+" since its dim '"+n[o]+"' is not 1");(a[s]==null||a[s]>o)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function gs(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function Bf(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function zf(n,t){for(var e=0;e<n.length;e++){var i=n[e];if(isNaN(i)||!isFinite(i))throw Error("A tensor of type "+t+" being uploaded contains "+i+".")}}function Pf(n){return n==="bool"||n==="complex64"||n==="float32"||n==="int32"||n==="string"}function _f(n,t){return t==="complex64"||(t==="float32"&&n!=="complex64"||t==="int32"&&n!=="float32"&&n!=="complex64")?!1:!(t==="bool"&&n==="bool")}function Ot(n){return n instanceof Float32Array||n instanceof Int32Array||n instanceof Uint8Array}function Mf(n){if(n==="float32"||n==="int32")return 4;if(n==="complex64")return 8;if(n==="bool")return 1;throw new Error("Unknown dtype "+n)}function Hf(n){if(n==null)return 0;var t=0;return n.forEach(function(e){return t+=e.length}),t}function ni(n){return typeof n=="string"||n instanceof String}function Vf(n){return typeof n=="boolean"}function qf(n){return typeof n=="number"}function vs(n){return Array.isArray(n)?vs(n[0]):n instanceof Float32Array?"float32":n instanceof Int32Array||n instanceof Uint8Array?"int32":qf(n)?"float32":ni(n)?"string":Vf(n)?"bool":"float32"}function ii(n){return!!(n&&n.constructor&&n.call&&n.apply)}function ys(n,t){for(var e=t;e<n;++e)if(n%e===0)return e;return n}function wr(n){var t=n.length;if(t<2)return[];var e=new Array(t-1);e[t-2]=n[t-1];for(var i=t-3;i>=0;--i)e[i]=e[i+1]*n[i+1];return e}function JI(n,t){return t==="string"?Yu(n):bs([n],t)}function bs(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=Wi(n)),He().getBool("DEBUG")&&zf(n,t),ZI(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i<e.length;++i)Math.round(n[i])!==0&&(e[i]=1);return e}else throw new Error("Unknown data type "+t)}function Gf(n,t,e){var i=new Array;if(t.length===1)for(var r=t[0],a=0;a<r;a++)i[a]=e[n+a];else for(var r=t[0],s=t.slice(1),o=s.reduce(function(u,c){return u*c}),a=0;a<r;a++)i[a]=Gf(n+a*o,s,e);return i}function Sr(n,t){if(n.length===0)return t[0];var e=n.reduce(function(i,r){return i*r});if(e===0)return[];if(e!==t.length)throw new Error("["+n+"] does not match the input size "+t.length+".");return Gf(0,n,t)}function ZI(n,t){return n instanceof Float32Array&&t==="float32"||n instanceof Int32Array&&t==="int32"||n instanceof Uint8Array&&t==="bool"}function Ku(n,t){for(var e=Lr(n,t),i=0;i<e.length;i++)e[i]=1;return e}function Lr(n,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool")return new Uint8Array(n);throw new Error("Unknown data type "+t)}function QI(n,t){var e=n.reduce(function(i,r){return i*r},1);if(t==null||t==="float32")return Sr(n,new Float32Array(e));if(t==="int32")return Sr(n,new Int32Array(e));if(t==="bool")return Sr(n,new Uint8Array(e));throw new Error("Unknown data type "+t)}function ju(){return He().platform.now()}function $u(n){n.forEach(function(t){E(Number.isInteger(t)&&t>=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function e2(n,t){return He().platform.fetch(n,t)}function Yu(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.encode(n,t)}function Xu(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.decode(n,t)}function t2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r<n.length-1;++r)i+=e[r]*n[r];return i}function n2(n,t,e){if(t===0)return[];if(t===1)return[n];for(var i=new Array(t),r=0;r<i.length-1;++r)i[r]=Math.floor(n/e[r]),n-=i[r]*e[r];return i[i.length-1]=n,i}var i2={__proto__:null,shuffle:Ff,clamp:pa,nearestLargerEven:HI,sum:VI,randUniform:qI,distSquared:GI,assert:E,assertShapesMatch:Be,assertNonNull:Fi,flatten:Wi,sizeFromShape:ot,isScalarShape:YI,arraysEqual:ln,isInt:rt,tanh:KI,sizeToSquarishShape:jI,createShuffledIndices:$I,rightPad:fa,repeatedTry:XI,inferFromImplicitShape:Wf,parseAxisParam:Qe,squeezeShape:Uf,getTypedArrayFromDType:gs,getArrayFromDType:Bf,checkConversionForErrors:zf,isValidDtype:Pf,hasEncodingLoss:_f,isTypedArray:Ot,bytesPerElement:Mf,bytesFromStringArray:Hf,isString:ni,isBoolean:Vf,isNumber:qf,inferDtype:vs,isFunction:ii,nearestDivisor:ys,computeStrides:wr,createScalarValue:JI,toTypedArray:bs,toNestedArray:Sr,makeOnesTypedArray:Ku,makeZerosTypedArray:Lr,makeZerosNestedTypedArray:QI,now:ju,assertNonNegativeIntegerDimensions:$u,fetch:e2,encodeString:Yu,decodeString:Xu,locToIndex:t2,indexToLoc:n2};var s2=function(){function n(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new a2)}return n.prototype.profileKernel=function(t,e,i){for(var r,a=function(){r=i()},s=this.backendTimer.time(a),o=function(c){var h=r[c];h.data().then(function(d){r2(d,h.dtype,t)})},l=0;l<r.length;l++)o(l);var u={kernelName:t,outputs:r,inputs:e,timeMs:s.then(function(c){return c.kernelMs}),extraInfo:s.then(function(c){return c.getExtraProfileInfo!=null?c.getExtraProfileInfo():""})};return u},n.prototype.logKernelProfile=function(t){var e=this,i=t.kernelName,r=t.outputs,a=t.timeMs,s=t.inputs,o=t.extraInfo;r.forEach(function(l){Promise.all([l.data(),a,o]).then(function(u){e.logger.logKernelProfile(i,l,u[0],u[1],s,u[2])})})},n}();function r2(n,t,e){if(t!=="float32")return!1;for(var i=0;i<n.length;i++){var r=n[i];if(isNaN(r)||!isFinite(r))return console.warn("Found "+r+" in the result of '"+e+"'"),!0}return!1}var a2=function(){function n(){}return n.prototype.logKernelProfile=function(t,e,i,r,a,s){var o=typeof r=="number"?fa(r+"ms",9):r.error,l=fa(t,25),u=e.rank,c=e.size,h=fa(e.shape.toString(),14),d="";for(var p in a){var f=a[p];if(f!=null){var m=f.shape||e.shape,g=m.length;d+=p+": "+g+"D "+(g>0?m:"")+" "}}console.log("%c"+l+"	%c"+o+"	%c"+u+"D "+h+"	%c"+c+"	%c"+d+"	%c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function o2(n,t,e){for(var i={},r={},a=0;a<t.length;a++)i[t[a].id]=!0;for(var a=0;a<n.length;a++){var s=n[a],o=s.inputs;for(var l in o){for(var u=o[l],c=!1,h=0;h<t.length;h++)if(i[u.id]){s.outputs.forEach(function(w){return i[w.id]=!0}),c=!0,r[s.id]=!0;break}if(c)break}}var d={};d[e.id]=!0;for(var p={},a=n.length-1;a>=0;a--)for(var s=n[a],o=s.inputs,h=0;h<s.outputs.length;h++)if(d[s.outputs[h].id]){for(var l in o)d[o[l].id]=!0,p[s.id]=!0;break}for(var f=[],a=0;a<n.length;a++){var s=n[a];if(r[s.id]&&p[s.id]){var m={};for(var l in s.inputs){var g=s.inputs[l];i[g.id]&&(m[l]=g)}var v=Object.assign({},s);v.inputs=m,v.outputs=s.outputs,f.push(v)}}return f}function l2(n,t,e,i){for(var r=function(s){var o=t[s],l=[];if(o.outputs.forEach(function(d){var p=n[d.id];p!=null?l.push(p):l.push(null)}),o.gradient==null)throw new Error("Cannot compute gradient: gradient function not found "+("for "+o.kernelName+"."));var u=o.gradient(l),c=function(d){if(!(d in u))throw new Error("Cannot backprop through input "+d+". "+("Available gradients found: "+Object.keys(u)+"."));var p=e(function(){return u[d]()});if(p.dtype!=="float32")throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+(d+" must have 'float32' dtype, but has '"+p.dtype+"'"));var f=o.inputs[d];if(!ln(p.shape,f.shape))throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+("'"+d+"' has shape '"+p.shape+"', which does not match ")+("the shape of the input '"+f.shape+"'"));if(n[f.id]==null)n[f.id]=p;else{var m=n[f.id];n[f.id]=i(m,p),m.dispose()}};for(var h in o.inputs)c(h)},a=t.length-1;a>=0;a--)r(a)}var Yf=20,ma=3,Ju=7;function c2(n,t,e,i){var r=wr(t),a=u2(n,t,e,r),s=t.length,o=ws(n,t,e,r,a),l=["Tensor"];return i&&(l.push("  dtype: "+e),l.push("  rank: "+s),l.push("  shape: ["+t+"]"),l.push("  values:")),l.push(o.map(function(u){return"    "+u}).join(`
+var Np=Object.defineProperty;var be=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),CI=n=>Np(n,"__esModule",{value:!0}),xp=(n,t)=>{CI(n);for(var e in t)Np(n,e,{get:t[e],enumerable:!0})};var Cp=be(()=>{});var Rp=be(()=>{});var ds=be(()=>{});var Xi=be(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var tl=function(n,t){return tl=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},tl(n,t)};function qn(n,t){tl(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var RI=1e-7,OI=1e-4,EI=function(){function n(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}return n.prototype.get=function(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)},n.prototype.set=function(t,e){this.dataIdsCount++,this.data.set(t,e)},n.prototype.has=function(t){return this.data.has(t)},n.prototype.delete=function(t){return this.dataIdsCount--,this.data.delete(t)},n.prototype.numDataIds=function(){return this.dataIdsCount},n}(),Op=function(){function n(){}return n.prototype.time=function(t){return $("time")},n.prototype.read=function(t){return $("read")},n.prototype.readSync=function(t){return $("readSync")},n.prototype.numDataIds=function(){return $("numDataIds")},n.prototype.disposeData=function(t){return $("disposeData")},n.prototype.write=function(t,e,i){return $("write")},n.prototype.move=function(t,e,i,r){return $("move")},n.prototype.memory=function(){return $("memory")},n.prototype.floatPrecision=function(){return $("floatPrecision")},n.prototype.epsilon=function(){return this.floatPrecision()===32?RI:OI},n.prototype.batchMatMul=function(t,e,i,r){return $("batchMatMul")},n.prototype.fusedBatchMatMul=function(t){var e=t.a,i=t.b,r=t.transposeA,a=t.transposeB,s=t.bias,o=t.activation,l=t.preluActivationWeights;return $("fusedBatchMatMul")},n.prototype.slice=function(t,e,i){return $("slice")},n.prototype.stridedSlice=function(t,e,i,r){return $("stridedSlice")},n.prototype.unstack=function(t,e){return $("unstack")},n.prototype.reverse=function(t,e){return $("reverse")},n.prototype.concat=function(t,e){return $("concat")},n.prototype.neg=function(t){return $("neg")},n.prototype.add=function(t,e){return $("add")},n.prototype.addN=function(t){return $("addN")},n.prototype.subtract=function(t,e){return $("subtract")},n.prototype.multiply=function(t,e){return $("multiply")},n.prototype.realDivide=function(t,e){return $("realDivide")},n.prototype.floorDiv=function(t,e){return $("floorDiv")},n.prototype.sum=function(t,e){return $("sum")},n.prototype.prod=function(t,e){return $("prod")},n.prototype.unsortedSegmentSum=function(t,e,i){return $("unsortedSegmentSum")},n.prototype.argMin=function(t,e){return $("argMin")},n.prototype.argMax=function(t,e){return $("argMax")},n.prototype.equal=function(t,e){return $("equal")},n.prototype.notEqual=function(t,e){return $("notEqual")},n.prototype.less=function(t,e){return $("less")},n.prototype.lessEqual=function(t,e){return $("lessEqual")},n.prototype.greater=function(t,e){return $("greater")},n.prototype.greaterEqual=function(t,e){return $("greaterEqual")},n.prototype.logicalNot=function(t){return $("logicalNot")},n.prototype.logicalAnd=function(t,e){return $("logicalAnd")},n.prototype.logicalOr=function(t,e){return $("logicalOr")},n.prototype.where=function(t){return $("where")},n.prototype.select=function(t,e,i){return $("select")},n.prototype.topk=function(t,e,i){return $("topk")},n.prototype.min=function(t,e){return $("min")},n.prototype.minimum=function(t,e){return $("minimum")},n.prototype.mod=function(t,e){return $("mod")},n.prototype.max=function(t,e){return $("max")},n.prototype.maximum=function(t,e){return $("maximum")},n.prototype.all=function(t,e){return $("all")},n.prototype.any=function(t,e){return $("any")},n.prototype.squaredDifference=function(t,e){return $("squaredDifference")},n.prototype.ceil=function(t){return $("ceil")},n.prototype.floor=function(t){return $("floor")},n.prototype.round=function(t){return $("round")},n.prototype.sign=function(t){return $("sign")},n.prototype.isNaN=function(t){return $("isNaN")},n.prototype.isInf=function(t){return $("isInf")},n.prototype.isFinite=function(t){return $("isFinite")},n.prototype.pow=function(t,e){return $("pow")},n.prototype.exp=function(t){return $("exp")},n.prototype.expm1=function(t){return $("expm1")},n.prototype.softmax=function(t,e){return $("softmax")},n.prototype.log=function(t){return $("log")},n.prototype.log1p=function(t){return $("log1p")},n.prototype.sqrt=function(t){return $("sqrt")},n.prototype.rsqrt=function(t){return $("rsqrt")},n.prototype.square=function(t){return $("square")},n.prototype.reciprocal=function(t){return $("reciprocal")},n.prototype.relu=function(t){return $("relu")},n.prototype.relu6=function(t){return $("relu6")},n.prototype.prelu=function(t,e){return $("prelu")},n.prototype.elu=function(t){return $("elu")},n.prototype.eluDer=function(t,e){return $("eluDer")},n.prototype.selu=function(t){return $("selu")},n.prototype.int=function(t){return $("int")},n.prototype.clip=function(t,e,i){return $("clip")},n.prototype.abs=function(t){return $("abs")},n.prototype.complexAbs=function(t){return $("complexAbs")},n.prototype.sigmoid=function(t){return $("sigmoid")},n.prototype.softplus=function(t){return $("softplus")},n.prototype.sin=function(t){return $("sin")},n.prototype.cos=function(t){return $("cos")},n.prototype.tan=function(t){return $("tan")},n.prototype.asin=function(t){return $("asin")},n.prototype.acos=function(t){return $("acos")},n.prototype.atan=function(t){return $("atan")},n.prototype.atan2=function(t,e){return $("atan2")},n.prototype.sinh=function(t){return $("sinh")},n.prototype.cosh=function(t){return $("cosh")},n.prototype.tanh=function(t){return $("tanh")},n.prototype.asinh=function(t){return $("asinh")},n.prototype.acosh=function(t){return $("acosh")},n.prototype.atanh=function(t){return $("atanh")},n.prototype.erf=function(t){return $("erf")},n.prototype.step=function(t,e){return $("step")},n.prototype.fusedConv2d=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedConv2d")},n.prototype.conv2d=function(t,e,i){return $("conv2d")},n.prototype.conv2dDerInput=function(t,e,i){return $("conv2dDerInput")},n.prototype.conv2dDerFilter=function(t,e,i){return $("conv2dDerFilter")},n.prototype.fusedDepthwiseConv2D=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedDepthwiseConv2D")},n.prototype.depthwiseConv2D=function(t,e,i){return $("depthwiseConv2D")},n.prototype.depthwiseConv2DDerInput=function(t,e,i){return $("depthwiseConv2DDerInput")},n.prototype.depthwiseConv2DDerFilter=function(t,e,i){return $("depthwiseConv2DDerFilter")},n.prototype.conv3d=function(t,e,i){return $("conv3d")},n.prototype.conv3dDerInput=function(t,e,i){return $("conv3dDerInput")},n.prototype.conv3dDerFilter=function(t,e,i){return $("conv3dDerFilter")},n.prototype.maxPool=function(t,e){return $("maxPool")},n.prototype.maxPoolBackprop=function(t,e,i,r){return $("maxPoolBackprop")},n.prototype.avgPool=function(t,e){return $("avgPool")},n.prototype.avgPoolBackprop=function(t,e,i){return $("avgPoolBackprop")},n.prototype.avgPool3d=function(t,e){return $("avgPool3d")},n.prototype.avgPool3dBackprop=function(t,e,i){return $("avgPool3dBackprop")},n.prototype.maxPool3d=function(t,e){return $("maxPool3d")},n.prototype.maxPool3dBackprop=function(t,e,i,r){return $("maxPool3dBackprop")},n.prototype.reshape=function(t,e){return $("reshape")},n.prototype.cast=function(t,e){return $("cast")},n.prototype.tile=function(t,e){return $("tile")},n.prototype.pad=function(t,e,i){return $("pad")},n.prototype.transpose=function(t,e){return $("transpose")},n.prototype.gather=function(t,e,i){return $("gather")},n.prototype.gatherND=function(t,e){return $("gatherND")},n.prototype.scatterND=function(t,e,i){return $("scatterND")},n.prototype.batchToSpaceND=function(t,e,i){return $("batchToSpaceND")},n.prototype.spaceToBatchND=function(t,e,i){return $("spaceToBatchND")},n.prototype.resizeBilinear=function(t,e,i,r){return $("resizeBilinear")},n.prototype.resizeBilinearBackprop=function(t,e,i){return $("resizeBilinearBackprop")},n.prototype.resizeNearestNeighbor=function(t,e,i,r){return $("resizeNearestNeighbor")},n.prototype.resizeNearestNeighborBackprop=function(t,e,i){return $("resizeNearestNeighborBackprop")},n.prototype.batchNorm=function(t,e,i,r,a,s){return $("batchNorm")},n.prototype.localResponseNormalization4D=function(t,e,i,r,a){return $("localResponseNormalization4D")},n.prototype.LRNGrad=function(t,e,i,r,a,s,o){return $("LRNGrad")},n.prototype.multinomial=function(t,e,i,r){return $("multinomial")},n.prototype.oneHot=function(t,e,i,r){return $("oneHot")},n.prototype.cumsum=function(t,e,i,r){return $("cumsum")},n.prototype.nonMaxSuppression=function(t,e,i,r,a){return $("nonMaxSuppression")},n.prototype.fft=function(t){return $("fft")},n.prototype.ifft=function(t){return $("ifft")},n.prototype.complex=function(t,e){return $("complex")},n.prototype.real=function(t){return $("real")},n.prototype.imag=function(t){return $("imag")},n.prototype.cropAndResize=function(t,e,i,r,a,s){return $("cropAndResize")},n.prototype.depthToSpace=function(t,e,i){return $("depthToSpace")},n.prototype.split=function(t,e,i){return $("split")},n.prototype.sparseToDense=function(t,e,i,r){return $("sparseToDense")},n.prototype.diag=function(t){return $("diag")},n.prototype.fill=function(t,e,i){return $("fill")},n.prototype.onesLike=function(t){return $("onesLike")},n.prototype.zerosLike=function(t){return $("zerosLike")},n.prototype.linspace=function(t,e,i){return $("linspace")},n.prototype.dispose=function(){return $("dispose")},n}();function $(n){throw new Error("'"+n+"' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen")}var Ep="tfjsflags",Dp=function(){function n(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}return n.prototype.setPlatform=function(t,e){this.platform!=null&&console.warn("Platform "+this.platformName+" has already been set. "+("Overwriting the platform with "+e+".")),this.platformName=t,this.platform=e},n.prototype.registerFlag=function(t,e,i){if(this.flagRegistry[t]={evaluationFn:e,setHook:i},this.urlFlags[t]!=null){var r=this.urlFlags[t];console.warn("Setting feature override from URL "+t+": "+r+"."),this.set(t,r)}},n.prototype.getAsync=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return t in this.flags?[2,this.flags[t]]:(e=this.flags,i=t,[4,this.evaluateFlag(t)]);case 1:return e[i]=r.sent(),[2,this.flags[t]]}})})},n.prototype.get=function(t){if(t in this.flags)return this.flags[t];var e=this.evaluateFlag(t);if(e instanceof Promise)throw new Error("Flag "+t+" cannot be synchronously evaluated. Please use getAsync() instead.");return this.flags[t]=e,this.flags[t]},n.prototype.getNumber=function(t){return this.get(t)},n.prototype.getBool=function(t){return this.get(t)},n.prototype.getFlags=function(){return this.flags},Object.defineProperty(n.prototype,"features",{get:function(){return this.flags},enumerable:!0,configurable:!0}),n.prototype.set=function(t,e){if(this.flagRegistry[t]==null)throw new Error("Cannot set flag "+t+" as it has not been registered.");this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)},n.prototype.evaluateFlag=function(t){if(this.flagRegistry[t]==null)throw new Error("Cannot evaluate flag '"+t+"': no evaluation function found.");return this.flagRegistry[t].evaluationFn()},n.prototype.setFlags=function(t){this.flags=Object.assign({},t)},n.prototype.reset=function(){this.flags={},this.urlFlags={},this.populateURLFlags()},n.prototype.populateURLFlags=function(){var t=this;if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;var e=DI(this.global.location.search);if(Ep in e){var i=e[Ep].split(",");i.forEach(function(r){var a=r.split(":"),s=a[0],o=a[1];t.urlFlags[s]=kI(s,o)})}},n}();function DI(n){var t={};return n.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,function(e){for(var i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];return FI(t,i[0],i[1]),i.join("=")}),t}function FI(n,t,e){n[decodeURIComponent(t)]=decodeURIComponent(e||"")}function kI(n,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(""+ +t===t)return+t;throw new Error("Could not parse value flag value "+t+" for flag "+n+".")}function He(){return I.ENV}I.ENV=null;function WI(n){I.ENV=n}var nl;function kp(){if(nl==null){var n=void 0;if(typeof window!="undefined")n=window;else if(typeof global!="undefined")n=global;else if(typeof process!="undefined")n=process;else if(typeof self!="undefined")n=self;else throw new Error("Could not find a global object");nl=n}return nl}function UI(){var n=kp();return n._tfGlobals==null&&(n._tfGlobals=new Map),n._tfGlobals}function Fp(n,t){var e=UI();if(e.has(n))return e.get(n);var i=t();return e.set(n,i),e.get(n)}var il="Abs",rl="Acos",al="Acosh",ps="Add",sl="AddN",Wp="All",Up="Any",ol="ArgMax",ll="ArgMin",ul="Asin",cl="Asinh",hl="Atan",dl="Atanh",pl="Atan2",fl="AvgPool",Bp="AvgPoolBackprop",ml="AvgPool3D",zp="AvgPool3DBackprop",gl="BatchMatMul",vl="BatchToSpaceND",yl="BroadcastTo",fs="Cast",bl="Ceil",wl="ClipByValue",Pp="Complex",Sl="Concat",Ll="Conv2D",_p="Conv2DBackpropFilter",Il="Conv2DBackpropInput",Al="Conv3D",Mp="Conv3DBackpropFilterV2",Hp="Conv3DBackpropInputV2",Tl="Cos",Nl="Cosh",xl="Cumsum",Vp="CropAndResize",qp="DepthToSpace",Cl="DepthwiseConv2dNative",Gp="DepthwiseConv2dNativeBackpropFilter",Yp="DepthwiseConv2dNativeBackpropInput",Kp="Diag",Rl="Dilation2D",jp="Dilation2DBackpropInput",$p="Dilation2DBackpropFilter",Ol="Div",El="Elu",Xp="EluGrad",Dl="Erf",Jp="Equal",kl="Exp",Fl="Expm1",Zp="FFT",Qp="Fill",ef="FlipLeftRight",Wl="Floor",Ul="FloorDiv",Bl="FusedBatchNorm",zl="GatherV2",tf="GatherNd",nf="Greater",Pl="GreaterEqual",_l="Identity",rf="IFFT",af="Imag",Ml="IsFinite",Hl="IsInf",Vl="IsNan",sf="Less",of="LessEqual",lf="LinSpace",ql="Log",Gl="Log1p",uf="LogicalAnd",cf="LogicalNot",hf="LogicalOr",Yl="LogSoftmax",Kl="LRN",df="LRNBackprop",jl="Max",$l="Maximum",Xl="MaxPool",pf="MaxPoolBackprop",Jl="MaxPool3D",ff="MaxPool3DBackprop",mf="MaxPoolWithArgmax",BI="Mean",Zl="Min",Ql="Minimum",eu="Mod",tu="Multiply",nu="Negate",gf="NotEqual",vf="NonMaxSuppressionV3",yf="NonMaxSuppressionV4",bf="NonMaxSuppressionV5",iu="OnesLike",ru="OneHot",au="PadV2",zI="Pool",su="Pow",ou="Prelu",wf="Prod",Sf="Range",Lf="Real",lu="Reciprocal",uu="Relu",cu="Reshape",hu="ResizeNearestNeighbor",If="ResizeNearestNeighborGrad",du="ResizeBilinear",Af="ResizeBilinearGrad",pu="Relu6",fu="Reverse",mu="Round",gu="Rsqrt",Tf="ScatterNd",vu="SelectV2",yu="Selu",bu="Slice",wu="Sin",Su="Sinh",Lu="Sign",Iu="Sigmoid",Au="Softplus",Tu="Sqrt",Nu="Sum",xu="SpaceToBatchND",Cu="SplitV",Ru="Softmax",Ou="SquaredDifference",Nf="Square",Eu="Sub",xf="SparseToDense",Cf="StridedSlice",Du="Tan",ku="Tanh",Fu="Tile",Rf="TopK",Wu="Transpose",Of="Unique",Uu="Unpack",Bu="UnsortedSegmentSum",zu="ZerosLike",Pu="Step",_u="FromPixels",Ef="RotateWithOffset",Mu="_FusedMatMul",Hu="FusedConv2D",Vu="FusedDepthwiseConv2D";var br=Fp("kernelRegistry",function(){return new Map}),da=Fp("gradRegistry",function(){return new Map});function Gu(n,t){var e=qu(n,t);return br.get(e)}function Yu(n){return da.get(n)}function ms(n){for(var t=br.entries(),e=[];;){var i=t.next(),r=i.done,a=i.value;if(r)break;var s=a[0],o=a[1],l=s.split("_")[0];l===n&&e.push(o)}return e}function Df(n){var t=n.kernelName,e=n.backendName,i=qu(t,e);br.has(i)&&console.warn("The kernel '"+t+"' for backend "+("'"+e+"' is already registered")),br.set(i,n)}function kf(n){var t=n.kernelName;da.has(t)&&(He().getBool("DEBUG")&&console.warn("Overriding the gradient for '"+t+"'")),da.set(t,n)}function PI(n,t){var e=qu(n,t);if(!br.has(e))throw new Error("The kernel '"+n+"' for backend "+("'"+t+"' is not registered"));br.delete(e)}function _I(n){if(!da.has(n))throw new Error("The gradient '"+n+"' for backend is not registered");da.delete(n)}function MI(n,t){var e=ms(n);e.forEach(function(i){var r=Object.assign({},i,{backendName:t});Df(r)})}function qu(n,t){return t+"_"+n}function Ff(n){for(var t=n.length,e=0,i=0;t>0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function pa(n,t,e){return Math.max(n,Math.min(t,e))}function HI(n){return n%2===0?n:n+1}function VI(n){for(var t=0,e=0;e<n.length;e++)t+=n[e];return t}function qI(n,t){var e=Math.random();return t*e+(1-e)*n}function GI(n,t){for(var e=0,i=0;i<n.length;i++){var r=Number(n[i])-Number(t[i]);e+=r*r}return e}function E(n,t){if(!n)throw new Error(typeof t=="string"?t:t())}function Be(n,t,e){e===void 0&&(e=""),E(on(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function Fi(n){E(n!=null,function(){return"The input to the tensor constructor must be a non-null value."})}function Wi(n,t,e){if(t===void 0&&(t=[]),e===void 0&&(e=!1),t==null&&(t=[]),Array.isArray(n)||Ot(n)&&!e)for(var i=0;i<n.length;++i)Wi(n[i],t,e);else t.push(n);return t}function ot(n){if(n.length===0)return 1;for(var t=n[0],e=1;e<n.length;e++)t*=n[e];return t}function YI(n){return n.length===0}function on(n,t){if(n===t)return!0;if(n==null||t==null)return!1;if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==t[e])return!1;return!0}function rt(n){return n%1===0}function KI(n){if(Math.tanh!=null)return Math.tanh(n);if(n===Infinity)return 1;if(n===-Infinity)return-1;var t=Math.exp(2*n);return(t-1)/(t+1)}function jI(n){var t=Math.ceil(Math.sqrt(n));return[t,Math.ceil(n/t)]}function $I(n){for(var t=new Uint32Array(n),e=0;e<n;++e)t[e]=e;return Ff(t),t}function fa(n,t){return t<=n.length?n:n+" ".repeat(t-n.length)}function XI(n,t,e){return t===void 0&&(t=function(i){return 0}),new Promise(function(i,r){var a=0,s=function(){if(n()){i();return}a++;var o=t(a);if(e!=null&&a>=e){r();return}setTimeout(s,o)};s()})}function Wf(n,t){for(var e=1,i=-1,r=0;r<n.length;++r)if(n[r]>=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Qe(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&i<e}),function(){return"All values in axis param must be in range [-"+e+", "+e+") but "+("got axis "+n)}),E(n.every(function(i){return rt(i)}),function(){return"All values in axis param must be integers but "+("got axis "+n)}),n.map(function(i){return i<0?e+i:i})}function Uf(n,t){for(var e=[],i=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Qe(t,n).sort(),s=0,o=0;o<n.length;++o){if(a!=null){if(a[s]===o&&n[o]!==1)throw new Error("Can't squeeze axis "+o+" since its dim '"+n[o]+"' is not 1");(a[s]==null||a[s]>o)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function gs(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function Bf(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function zf(n,t){for(var e=0;e<n.length;e++){var i=n[e];if(isNaN(i)||!isFinite(i))throw Error("A tensor of type "+t+" being uploaded contains "+i+".")}}function Pf(n){return n==="bool"||n==="complex64"||n==="float32"||n==="int32"||n==="string"}function _f(n,t){return t==="complex64"||(t==="float32"&&n!=="complex64"||t==="int32"&&n!=="float32"&&n!=="complex64")?!1:!(t==="bool"&&n==="bool")}function Ot(n){return n instanceof Float32Array||n instanceof Int32Array||n instanceof Uint8Array}function Mf(n){if(n==="float32"||n==="int32")return 4;if(n==="complex64")return 8;if(n==="bool")return 1;throw new Error("Unknown dtype "+n)}function Hf(n){if(n==null)return 0;var t=0;return n.forEach(function(e){return t+=e.length}),t}function ni(n){return typeof n=="string"||n instanceof String}function Vf(n){return typeof n=="boolean"}function qf(n){return typeof n=="number"}function vs(n){return Array.isArray(n)?vs(n[0]):n instanceof Float32Array?"float32":n instanceof Int32Array||n instanceof Uint8Array?"int32":qf(n)?"float32":ni(n)?"string":Vf(n)?"bool":"float32"}function ii(n){return!!(n&&n.constructor&&n.call&&n.apply)}function ys(n,t){for(var e=t;e<n;++e)if(n%e===0)return e;return n}function wr(n){var t=n.length;if(t<2)return[];var e=new Array(t-1);e[t-2]=n[t-1];for(var i=t-3;i>=0;--i)e[i]=e[i+1]*n[i+1];return e}function JI(n,t){return t==="string"?Ku(n):bs([n],t)}function bs(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=Wi(n)),He().getBool("DEBUG")&&zf(n,t),ZI(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i<e.length;++i)Math.round(n[i])!==0&&(e[i]=1);return e}else throw new Error("Unknown data type "+t)}function Gf(n,t,e){var i=new Array;if(t.length===1)for(var r=t[0],a=0;a<r;a++)i[a]=e[n+a];else for(var r=t[0],s=t.slice(1),o=s.reduce(function(u,c){return u*c}),a=0;a<r;a++)i[a]=Gf(n+a*o,s,e);return i}function Sr(n,t){if(n.length===0)return t[0];var e=n.reduce(function(i,r){return i*r});if(e===0)return[];if(e!==t.length)throw new Error("["+n+"] does not match the input size "+t.length+".");return Gf(0,n,t)}function ZI(n,t){return n instanceof Float32Array&&t==="float32"||n instanceof Int32Array&&t==="int32"||n instanceof Uint8Array&&t==="bool"}function ju(n,t){for(var e=Lr(n,t),i=0;i<e.length;i++)e[i]=1;return e}function Lr(n,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool")return new Uint8Array(n);throw new Error("Unknown data type "+t)}function QI(n,t){var e=n.reduce(function(i,r){return i*r},1);if(t==null||t==="float32")return Sr(n,new Float32Array(e));if(t==="int32")return Sr(n,new Int32Array(e));if(t==="bool")return Sr(n,new Uint8Array(e));throw new Error("Unknown data type "+t)}function $u(){return He().platform.now()}function Xu(n){n.forEach(function(t){E(Number.isInteger(t)&&t>=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function e2(n,t){return He().platform.fetch(n,t)}function Ku(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.encode(n,t)}function Ju(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.decode(n,t)}function t2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r<n.length-1;++r)i+=e[r]*n[r];return i}function n2(n,t,e){if(t===0)return[];if(t===1)return[n];for(var i=new Array(t),r=0;r<i.length-1;++r)i[r]=Math.floor(n/e[r]),n-=i[r]*e[r];return i[i.length-1]=n,i}var i2={__proto__:null,shuffle:Ff,clamp:pa,nearestLargerEven:HI,sum:VI,randUniform:qI,distSquared:GI,assert:E,assertShapesMatch:Be,assertNonNull:Fi,flatten:Wi,sizeFromShape:ot,isScalarShape:YI,arraysEqual:on,isInt:rt,tanh:KI,sizeToSquarishShape:jI,createShuffledIndices:$I,rightPad:fa,repeatedTry:XI,inferFromImplicitShape:Wf,parseAxisParam:Qe,squeezeShape:Uf,getTypedArrayFromDType:gs,getArrayFromDType:Bf,checkConversionForErrors:zf,isValidDtype:Pf,hasEncodingLoss:_f,isTypedArray:Ot,bytesPerElement:Mf,bytesFromStringArray:Hf,isString:ni,isBoolean:Vf,isNumber:qf,inferDtype:vs,isFunction:ii,nearestDivisor:ys,computeStrides:wr,createScalarValue:JI,toTypedArray:bs,toNestedArray:Sr,makeOnesTypedArray:ju,makeZerosTypedArray:Lr,makeZerosNestedTypedArray:QI,now:$u,assertNonNegativeIntegerDimensions:Xu,fetch:e2,encodeString:Ku,decodeString:Ju,locToIndex:t2,indexToLoc:n2};var s2=function(){function n(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new a2)}return n.prototype.profileKernel=function(t,e,i){for(var r,a=function(){r=i()},s=this.backendTimer.time(a),o=function(c){var h=r[c];h.data().then(function(d){r2(d,h.dtype,t)})},l=0;l<r.length;l++)o(l);var u={kernelName:t,outputs:r,inputs:e,timeMs:s.then(function(c){return c.kernelMs}),extraInfo:s.then(function(c){return c.getExtraProfileInfo!=null?c.getExtraProfileInfo():""})};return u},n.prototype.logKernelProfile=function(t){var e=this,i=t.kernelName,r=t.outputs,a=t.timeMs,s=t.inputs,o=t.extraInfo;r.forEach(function(l){Promise.all([l.data(),a,o]).then(function(u){e.logger.logKernelProfile(i,l,u[0],u[1],s,u[2])})})},n}();function r2(n,t,e){if(t!=="float32")return!1;for(var i=0;i<n.length;i++){var r=n[i];if(isNaN(r)||!isFinite(r))return console.warn("Found "+r+" in the result of '"+e+"'"),!0}return!1}var a2=function(){function n(){}return n.prototype.logKernelProfile=function(t,e,i,r,a,s){var o=typeof r=="number"?fa(r+"ms",9):r.error,l=fa(t,25),u=e.rank,c=e.size,h=fa(e.shape.toString(),14),d="";for(var p in a){var f=a[p];if(f!=null){var m=f.shape||e.shape,g=m.length;d+=p+": "+g+"D "+(g>0?m:"")+" "}}console.log("%c"+l+"	%c"+o+"	%c"+u+"D "+h+"	%c"+c+"	%c"+d+"	%c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function o2(n,t,e){for(var i={},r={},a=0;a<t.length;a++)i[t[a].id]=!0;for(var a=0;a<n.length;a++){var s=n[a],o=s.inputs;for(var l in o){for(var u=o[l],c=!1,h=0;h<t.length;h++)if(i[u.id]){s.outputs.forEach(function(w){return i[w.id]=!0}),c=!0,r[s.id]=!0;break}if(c)break}}var d={};d[e.id]=!0;for(var p={},a=n.length-1;a>=0;a--)for(var s=n[a],o=s.inputs,h=0;h<s.outputs.length;h++)if(d[s.outputs[h].id]){for(var l in o)d[o[l].id]=!0,p[s.id]=!0;break}for(var f=[],a=0;a<n.length;a++){var s=n[a];if(r[s.id]&&p[s.id]){var m={};for(var l in s.inputs){var g=s.inputs[l];i[g.id]&&(m[l]=g)}var v=Object.assign({},s);v.inputs=m,v.outputs=s.outputs,f.push(v)}}return f}function l2(n,t,e,i){for(var r=function(s){var o=t[s],l=[];if(o.outputs.forEach(function(d){var p=n[d.id];p!=null?l.push(p):l.push(null)}),o.gradient==null)throw new Error("Cannot compute gradient: gradient function not found "+("for "+o.kernelName+"."));var u=o.gradient(l),c=function(d){if(!(d in u))throw new Error("Cannot backprop through input "+d+". "+("Available gradients found: "+Object.keys(u)+"."));var p=e(function(){return u[d]()});if(p.dtype!=="float32")throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+(d+" must have 'float32' dtype, but has '"+p.dtype+"'"));var f=o.inputs[d];if(!on(p.shape,f.shape))throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+("'"+d+"' has shape '"+p.shape+"', which does not match ")+("the shape of the input '"+f.shape+"'"));if(n[f.id]==null)n[f.id]=p;else{var m=n[f.id];n[f.id]=i(m,p),m.dispose()}};for(var h in o.inputs)c(h)},a=t.length-1;a>=0;a--)r(a)}var Yf=20,ma=3,Zu=7;function c2(n,t,e,i){var r=wr(t),a=u2(n,t,e,r),s=t.length,o=ws(n,t,e,r,a),l=["Tensor"];return i&&(l.push("  dtype: "+e),l.push("  rank: "+s),l.push("  shape: ["+t+"]"),l.push("  values:")),l.push(o.map(function(u){return"    "+u}).join(`
 `)),l.join(`
-`)}function u2(n,t,e,i){var r=ot(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?va(n):n;if(o>1)for(var u=0;u<r/a;u++)for(var c=u*a,h=0;h<a;h++)s[h]=Math.max(s[h],ga(l[c+h],0,e).length);return s}function ga(n,t,e){var i;return Array.isArray(n)?i=parseFloat(n[0].toFixed(Ju))+" + "+(parseFloat(n[1].toFixed(Ju))+"j"):ni(n)?i="'"+n+"'":e==="bool"?i=Kf(n):i=parseFloat(n.toFixed(Ju)).toString(),fa(i,t)}function Kf(n){return n===0?"false":"true"}function ws(n,t,e,i,r,a){a===void 0&&(a=!0);var s=e==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(e==="complex64"){var u=va(n);return[ga(u[0],0,e)]}return e==="bool"?[Kf(n[0])]:[n[0].toString()]}if(l===1){if(o>Yf){var c=ma*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-ma)*s,o*s));return e==="complex64"&&(h=va(h),d=va(d)),["["+h.map(function(C,R){return ga(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return ga(C,r[o-ma+R],e)}).join(", ")+"]"]}var p=e==="complex64"?va(n):Array.from(n);return["["+p.map(function(C,R){return ga(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>Yf){for(var b=0;b<ma;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,!1))}v.push("...");for(var b=o-ma;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}}else for(var b=0;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}var L=l===2?",":"";v[0]="["+v[0]+L;for(var b=1;b<v.length-1;b++)v[b]=" "+v[b]+L;for(var x=`,
+`)}function u2(n,t,e,i){var r=ot(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?va(n):n;if(o>1)for(var u=0;u<r/a;u++)for(var c=u*a,h=0;h<a;h++)s[h]=Math.max(s[h],ga(l[c+h],0,e).length);return s}function ga(n,t,e){var i;return Array.isArray(n)?i=parseFloat(n[0].toFixed(Zu))+" + "+(parseFloat(n[1].toFixed(Zu))+"j"):ni(n)?i="'"+n+"'":e==="bool"?i=Kf(n):i=parseFloat(n.toFixed(Zu)).toString(),fa(i,t)}function Kf(n){return n===0?"false":"true"}function ws(n,t,e,i,r,a){a===void 0&&(a=!0);var s=e==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(e==="complex64"){var u=va(n);return[ga(u[0],0,e)]}return e==="bool"?[Kf(n[0])]:[n[0].toString()]}if(l===1){if(o>Yf){var c=ma*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-ma)*s,o*s));return e==="complex64"&&(h=va(h),d=va(d)),["["+h.map(function(C,R){return ga(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return ga(C,r[o-ma+R],e)}).join(", ")+"]"]}var p=e==="complex64"?va(n):Array.from(n);return["["+p.map(function(C,R){return ga(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>Yf){for(var b=0;b<ma;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,!1))}v.push("...");for(var b=o-ma;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}}else for(var b=0;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}var L=l===2?",":"";v[0]="["+v[0]+L;for(var b=1;b<v.length-1;b++)v[b]=" "+v[b]+L;for(var x=`,
 `,b=2;b<l;b++)x+=`
-`;return v[v.length-1]=" "+v[v.length-1]+"]"+(a?"":x),v}function va(n){for(var t=[],e=0;e<n.length;e+=2)t.push([n[e],n[e+1]]);return t}var Ss=function(){function n(t,e,i){var r=this;if(this.dtype=e,this.shape=t.slice(),this.size=ot(t),i!=null){var a=i.length;E(a===this.size,function(){return"Length of values '"+a+"' does not match the size "+("inferred by the shape '"+r.size+"'.")})}if(e==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=i||Bf(e,this.size),this.strides=wr(t)}return n.prototype.set=function(t){for(var e=this,i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];i.length===0&&(i=[0]),E(i.length===this.rank,function(){return"The number of provided coordinates ("+i.length+") must "+("match the rank ("+e.rank+")")});var a=this.locToIndex(i);this.values[a]=t},n.prototype.get=function(){for(var t=[],e=0;e<arguments.length;e++)t[e]=arguments[e];t.length===0&&(t=[0]);for(var i=0,r=0,a=t;r<a.length;r++){var s=a[r];if(s<0||s>=this.shape[i]){var o="Requested out of range element at "+t+". "+("  Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u<t.length-1;++u)l+=this.strides[u]*t[u];return this.values[l]},n.prototype.locToIndex=function(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];for(var e=t[t.length-1],i=0;i<t.length-1;++i)e+=this.strides[i]*t[i];return e},n.prototype.indexToLoc=function(t){if(this.rank===0)return[];if(this.rank===1)return[t];for(var e=new Array(this.shape.length),i=0;i<e.length-1;++i)e[i]=Math.floor(t/this.strides[i]),t-=e[i]*this.strides[i];return e[e.length-1]=t,e},Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.toTensor=function(){return Cn().makeTensor(this.values,this.shape,this.dtype)},n}(),Cn=null,Ir=null;function h2(n){Cn=n}function d2(n){Ir=n}var Y=function(){function n(t,e,i,r){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=ot(t),this.strides=wr(t),this.dataId=i,this.id=r,this.rankType=this.rank<5?this.rank.toString():"higher"}return Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.buffer=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Ir.buffer(this.shape,this.dtype,t)]}})})},n.prototype.bufferSync=function(){return Ir.buffer(this.shape,this.dtype,this.dataSync())},n.prototype.array=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Sr(this.shape,t)]}})})},n.prototype.arraySync=function(){return Sr(this.shape,this.dataSync())},n.prototype.data=function(){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){switch(i.label){case 0:return this.throwIfDisposed(),t=Cn().read(this.dataId),this.dtype==="string"?[4,t]:[3,2];case 1:e=i.sent();try{return[2,e.map(function(r){return Xu(r)})]}catch(r){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}i.label=2;case 2:return[2,t]}})})},n.prototype.dataSync=function(){this.throwIfDisposed();var t=Cn().readSync(this.dataId);if(this.dtype==="string")try{return t.map(function(e){return Xu(e)})}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t},n.prototype.bytes=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return this.throwIfDisposed(),[4,Cn().read(this.dataId)];case 1:return t=e.sent(),this.dtype==="string"?[2,t]:[2,new Uint8Array(t.buffer)]}})})},n.prototype.dispose=function(){if(this.isDisposed)return;Cn().disposeTensor(this),this.isDisposedInternal=!0},Object.defineProperty(n.prototype,"isDisposed",{get:function(){return this.isDisposedInternal},enumerable:!0,configurable:!0}),n.prototype.throwIfDisposed=function(){if(this.isDisposed)throw new Error("Tensor is disposed.")},n.prototype.print=function(t){return t===void 0&&(t=!1),Ir.print(this,t)},n.prototype.clone=function(){return this.throwIfDisposed(),Ir.clone(this)},n.prototype.toString=function(t){t===void 0&&(t=!1);var e=this.dataSync();return c2(e,this.shape,this.dtype,t)},n.prototype.cast=function(t){return this.throwIfDisposed(),Ir.cast(this,t)},n.prototype.variable=function(t,e,i){return t===void 0&&(t=!0),this.throwIfDisposed(),Cn().makeVariable(this,t,e,i)},n}();Object.defineProperty(Y,Symbol.hasInstance,{value:function(n){return!!n&&n.data!=null&&n.dataSync!=null&&n.throwIfDisposed!=null}});var ya=function(n){qn(t,n);function t(e,i,r,a){var s=n.call(this,e.shape,e.dtype,e.dataId,a)||this;return s.trainable=i,s.name=r,s}return t.prototype.assign=function(e){if(e.dtype!==this.dtype)throw new Error("dtype of the new value ("+e.dtype+") and "+("previous value ("+this.dtype+") must match"));if(!ln(e.shape,this.shape))throw new Error("shape of the new value ("+e.shape+") and "+("previous value ("+this.shape+") must match"));Cn().disposeTensor(this),this.dataId=e.dataId,Cn().incRef(this,null)},t.prototype.dispose=function(){Cn().disposeVariable(this),this.isDisposedInternal=!0},t}(Y);Object.defineProperty(ya,Symbol.hasInstance,{value:function(n){return n instanceof Y&&n.assign!=null&&n.assign instanceof Function}});(function(n){n.R0="R0",n.R1="R1",n.R2="R2",n.R3="R3",n.R4="R4",n.R5="R5",n.R6="R6"})(I.Rank||(I.Rank={}));var Zu;(function(n){n.float32="float32",n.int32="int32",n.bool="int32",n.complex64="complex64"})(Zu||(Zu={}));var Qu;(function(n){n.float32="float32",n.int32="int32",n.bool="bool",n.complex64="complex64"})(Qu||(Qu={}));var ec;(function(n){n.float32="float32",n.int32="float32",n.bool="float32",n.complex64="complex64"})(ec||(ec={}));var tc;(function(n){n.float32="complex64",n.int32="complex64",n.bool="complex64",n.complex64="complex64"})(tc||(tc={}));var p2={float32:ec,int32:Zu,bool:Qu,complex64:tc};function Ls(n,t){if(n==="string"||t==="string"){if(n==="string"&&t==="string")return"string";throw new Error("Can not upcast "+n+" with "+t)}return p2[n][t]}function f2(n){return Ls(n,"int32")}function at(n,t){if(n.dtype===t.dtype)return[n,t];var e=Ls(n.dtype,t.dtype);return[n.cast(e),t.cast(e)]}function jf(n,t){E(n.dtype===t.dtype,function(){return"The dtypes of the first("+n.dtype+") and"+(" second("+t.dtype+") input must match")})}function m2(n,t){return t.some(function(e){return e.id===n.id})}function nc(n){var t=[],e=new Set;return $f(n,t,e),t}function $f(n,t,e){if(n==null)return;if(n instanceof Y){t.push(n);return}if(!g2(n))return;var i=n;for(var r in i){var a=i[r];e.has(a)||(e.add(a),$f(a,t,e))}}function g2(n){return Array.isArray(n)||typeof n=="object"}var v2={__proto__:null,makeTypesMatch:at,assertTypesMatch:jf,isTensorInList:m2,getTensorsInContainer:nc};var Xf=function(){function n(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null}}return n.prototype.dispose=function(){for(var t in this.registeredVariables)this.registeredVariables[t].dispose()},n}(),w2=function(){function n(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Xf}return n.prototype.ready=function(){return de(this,void 0,void 0,function(){var t,e,i,r;return pe(this,function(a){switch(a.label){case 0:if(this.pendingBackendInit!=null)return[2,this.pendingBackendInit.then(function(){})];if(this.backendInstance!=null)return[2];t=this.getSortedBackends(),e=0,a.label=1;case 1:return e<t.length?(i=t[e],[4,this.initializeBackend(i).success]):[3,5];case 2:return r=a.sent(),r?[4,this.setBackend(i)]:[3,4];case 3:return a.sent(),[2];case 4:return e++,[3,1];case 5:throw new Error("Could not initialize any backends, all backend initializations failed.")}})})},Object.defineProperty(n.prototype,"backend",{get:function(){if(this.pendingBackendInit!=null)throw new Error("Backend '"+this.backendName+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");if(this.backendInstance==null){var t=this.initializeBackendsAndReturnBest(),e=t.name,i=t.asyncInit;if(i)throw new Error("The highest priority backend '"+e+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");this.setBackend(e)}return this.backendInstance},enumerable:!0,configurable:!0}),n.prototype.backendNames=function(){return Object.keys(this.registryFactory)},n.prototype.findBackend=function(t){if(!(t in this.registry))if(t in this.registryFactory){var e=this.initializeBackend(t).asyncInit;if(e)return null}else return null;return this.registry[t]},n.prototype.findBackendFactory=function(t){return t in this.registryFactory?this.registryFactory[t].factory:null},n.prototype.registerBackend=function(t,e,i){return i===void 0&&(i=1),t in this.registryFactory?(console.warn(t+" backend was already registered. Reusing existing backend factory."),!1):(this.registryFactory[t]={factory:e,priority:i},!0)},n.prototype.setBackend=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s;return pe(this,function(o){switch(o.label){case 0:if(this.registryFactory[t]==null)throw new Error("Backend name '"+t+"' not found in registry");return this.backendName=t,this.registry[t]==null?(this.backendInstance=null,e=this.initializeBackend(t),i=e.success,r=e.asyncInit,r?[4,i]:[3,2]):[3,4];case 1:return s=o.sent(),[3,3];case 2:s=i,o.label=3;case 3:if(a=s,!a)return[2,!1];o.label=4;case 4:return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new s2(this.backendInstance),[2,!0]}})})},n.prototype.setupRegisteredKernels=function(){var t=this,e=ms(this.backendName);e.forEach(function(i){i.setupFunc!=null&&i.setupFunc(t.backendInstance)})},n.prototype.disposeRegisteredKernels=function(t){var e=this,i=ms(t);i.forEach(function(r){r.disposeFunc!=null&&r.disposeFunc(e.registry[t])})},n.prototype.initializeBackend=function(t){var e=this,i=this.registryFactory[t];if(i==null)throw new Error("Cannot initialize backend "+t+", no registration found.");try{var r=i.factory();if(r&&!(r instanceof Op)&&typeof r.then=="function"){var a=++this.pendingBackendInitId,s=r.then(function(o){return a<e.pendingBackendInitId?!1:(e.registry[t]=o,e.pendingBackendInit=null,!0)}).catch(function(o){return a<e.pendingBackendInitId||(e.pendingBackendInit=null,console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message)),!1});return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=r,{success:!0,asyncInit:!1}}catch(o){return console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message),{success:!1,asyncInit:!1}}},n.prototype.removeBackend=function(t){if(!(t in this.registryFactory))throw new Error(t+" backend not found in registry");this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)},n.prototype.getSortedBackends=function(){var t=this;if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(function(e,i){return t.registryFactory[i].priority-t.registryFactory[e].priority})},n.prototype.initializeBackendsAndReturnBest=function(){for(var t=this.getSortedBackends(),e=0;e<t.length;e++){var i=t[e],r=this.initializeBackend(i),a=r.success,s=r.asyncInit;if(s||a)return{name:i,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")},n.prototype.moveData=function(t,e){var i=this.state.tensorInfo.get(e),r=i.backend,a=this.readSync(e);r.disposeData(e),i.backend=t,t.move(e,a,i.shape,i.dtype),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++},n.prototype.tidy=function(t,e){var i=this,r=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");r=t}var a;return this.scopedRun(function(){return i.startScope(r)},function(){return i.endScope(a)},function(){return a=e(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a})},n.prototype.scopedRun=function(t,e,i){t();try{var r=i();return e(),r}catch(a){throw e(),a}},n.prototype.nextTensorId=function(){return n.nextTensorId++},n.prototype.nextVariableId=function(){return n.nextVariableId++},n.prototype.clone=function(t){var e=this.makeTensorFromDataId(t.dataId,t.shape,t.dtype),i={x:t},r=function(s){return{x:function(){var o="float32",l={x:s},u={dtype:o};return z.runKernelFunc(function(c){return c.cast(s,o)},l,null,fs,u)}}},a=[];return this.addTapeNode(this.state.activeScope.name,i,[e],r,a,{}),e},n.prototype.runKernel=function(t,e,i,r,a){var s=null,o=null;return this.runKernelFunc(s,e,o,t,i,r,a)},n.prototype.shouldCheckForMemLeaks=function(){return this.ENV.getBool("IS_TEST")},n.prototype.checkKernelForMemLeak=function(t,e,i){var r=this.backend.numDataIds(),a=0;i.forEach(function(l){a+=l.dtype==="complex64"?3:1});var s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=r-e-a-s;if(o>0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=qu(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var x=l.getTensorsForGradient(r,e,L);if(x==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});x=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(x)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=Gu(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&ni(t[0])&&(a=t.map(function(c){return Yu(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Hf(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new ya(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Mf(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof ya||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a<s.length?(o=s[a],l=o,[4,o.kernelTimeMs]):[3,6];case 3:return l.kernelTimeMs=c.sent(),u=o,[4,o.extraInfo];case 4:u.extraInfo=c.sent(),c.label=5;case 5:return a++,[3,2];case 6:return[2,this.state.activeProfile]}})})},n.prototype.isTapeOn=function(){return this.state.gradientDepth>0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=Gu(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=Lr(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=nc(t),r=new Set(i.map(function(l){return l.id})),a=0;a<this.state.activeScope.track.length;a++){var s=this.state.activeScope.track[a];!s.kept&&!r.has(s.id)&&s.dispose()}var o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],i.forEach(function(l){!l.kept&&l.scopeId===o.id&&e.track(l)})},n.prototype.gradients=function(t,e,i,r){var a=this;if(r===void 0&&(r=!1),E(e.length>0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=o2(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??y2(s.shape),l2(l,o,function(c){return a.tidy(c)},b2);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h<d.length;h++){var p=d[h];p.dispose()}}),a.state.activeTape=null),{value:s,grads:u}})},n.prototype.customGrad=function(t){var e=this;return E(ii(t),function(){return"The f passed in customGrad(f) must be a function."}),function(){for(var i=[],r=0;r<arguments.length;r++)i[r]=arguments[r];E(i.every(function(o){return o instanceof Y}),function(){return"The args passed in customGrad(f)(x1, x2,...) must all be tensors"});var a,s={};return i.forEach(function(o,l){s[l]=o}),e.runKernelFunc(function(o,l){return a=t.apply(void 0,i.concat([l])),E(a.value instanceof Y,function(){return"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"}),E(ii(a.gradFunc),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."}),a.value},s,function(o,l){var u=a.gradFunc(o,l),c=Array.isArray(u)?u:[u];E(c.length===i.length,function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."}),E(c.every(function(d){return d instanceof Y}),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."});var h={};return c.forEach(function(d,p){h[p]=function(){return d}}),h})}},n.prototype.readSync=function(t){var e=this.state.tensorInfo.get(t);return e.backend.readSync(t)},n.prototype.read=function(t){var e=this.state.tensorInfo.get(t);return e.backend.read(t)},n.prototype.time=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return e=ju(),[4,this.backend.time(t)];case 1:return i=r.sent(),i.wallMs=ju()-e,[2,i]}})})},n.prototype.track=function(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t},Object.defineProperty(n.prototype,"registeredVariables",{get:function(){return this.state.registeredVariables},enumerable:!0,configurable:!0}),n.prototype.reset=function(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Xf;for(var t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null},n.nextTensorId=0,n.nextVariableId=0,n}();function y2(n){var t=Ku(ot(n),"float32");return z.makeTensor(t,n,"float32")}function Jf(){var n=kp();if(n._tfengine==null){var t=new Dp(n);n._tfengine=new w2(t)}return WI(n._tfengine.ENV),h2(function(){return n._tfengine}),n._tfengine}var z=Jf();function b2(n,t){var e={a:n,b:t};return z.runKernelFunc(function(i,r){var a=i.add(n,t);return r([n,t]),a},e,null,ps)}function S2(){return typeof navigator!="undefined"&&navigator!=null}function L2(){if(S2()){var n=navigator.userAgent||navigator.vendor||window.opera;return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(n)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(n.substr(0,4))}return!1}function Zf(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var I2={__proto__:null,isMobile:L2,isBrowser:Zf};var Gn=He();Gn.registerFlag("DEBUG",function(){return!1},function(n){n&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});Gn.registerFlag("IS_BROWSER",function(){return Zf()});Gn.registerFlag("IS_NODE",function(){return typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined"});Gn.registerFlag("IS_CHROME",function(){return typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor)});Gn.registerFlag("PROD",function(){return!1});Gn.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",function(){return Gn.getBool("DEBUG")});Gn.registerFlag("DEPRECATION_WARNINGS_ENABLED",function(){return!0});Gn.registerFlag("IS_TEST",function(){return!1});function Rn(n,t){var e=n;if(Ot(n))return t==="string"?[]:[n.length];if(!Array.isArray(n))return[];for(var i=[];Array.isArray(e)||Ot(e)&&t!=="string";)i.push(e.length),e=e[0];return Array.isArray(n)&&He().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Qf(n,i,[]),i}function Qf(n,t,e){if(e=e||[],!Array.isArray(n)&&!Ot(n)){E(t.length===0,function(){return"Element arr["+e.join("][")+"] is a primitive, "+("but should be an array/TypedArray of "+t[0]+" elements")});return}E(t.length>0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r<n.length;++r)Qf(n[r],i,e.concat(r))}function em(n,t,e,i){if(n==null)return;if(n!=="numeric"&&n!==t||n==="numeric"&&t==="string")throw new Error("Argument '"+e+"' passed to '"+i+"' must "+("be "+n+" tensor, but got "+t+" tensor"))}function O(n,t,e,i){if(i===void 0&&(i="numeric"),n instanceof Y)return em(i,n.dtype,t,e),n;var r=vs(n);if(r!=="string"&&["bool","int32","float32"].indexOf(i)>=0&&(r=i),em(i,r,t,e),n==null||!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=Rn(n,r);!Ot(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?bs(n,r):Wi(n,[],o);return z.makeTensor(l,s,r)}function ba(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var tm="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+tm;var r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];z.startScope(e);try{var o=i.apply(void 0,a);return o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),z.endScope(o),o}catch(l){throw z.endScope(null),l}};return Object.defineProperty(r,"name",{value:e,configurable:!0}),r}function A2(n,t){var e=O(n,"real","complex"),i=O(t,"imag","complex");Be(e.shape,i.shape,"real and imag shapes, "+e.shape+" and "+i.shape+", must match in call to tf.complex().");var r=function(s){return s.complex(e,i)},a={real:e,imag:i};return z.runKernelFunc(r,a,null,Pp)}var ri=U({complex_:A2});function ai(n,t,e,i){if(i==null&&(i=vs(n)),i==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){$u(t);var r=ot(t),a=ot(e);E(r===a,function(){return"Based on the provided shape, ["+t+"], the tensor should have "+(r+" values but has "+a)});for(var s=0;s<e.length;++s){var o=e[s],l=s===e.length-1?o!==ot(t.slice(s)):!0;E(e[s]===t[s]||!l,function(){return"Error creating a new Tensor. Inferred shape "+("("+e+") does not match the provided ")+("shape ("+t+"). ")})}}return!Ot(n)&&!Array.isArray(n)&&(n=[n]),t=t||e,n=i!=="string"?bs(n,i):Wi(n,[],!0),z.makeTensor(n,t,i)}function si(n,t,e){var i=Rn(n,e);return ai(n,t,i,e)}var ic={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var Is=4;function N2(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l=this;return pe(this,function(u){switch(u.label){case 0:for(e=[],i=[],r=Array.isArray(n)?n.map(function(c){return c.name}):Object.keys(n),a=function(c){var h=r[c],d=Array.isArray(n)?n[c].tensor:n[h];if(d.dtype!=="float32"&&d.dtype!=="int32"&&d.dtype!=="bool"&&d.dtype!=="string"&&d.dtype!=="complex64")throw new Error("Unsupported dtype in weight '"+h+"': "+d.dtype);var p={name:h,shape:d.shape,dtype:d.dtype};if(d.dtype==="string"){var f=new Promise(function(m){return de(l,void 0,void 0,function(){var g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:return[4,d.bytes()];case 1:for(g=C.sent(),v=g.reduce(function(R,D){return R+D.length},0)+Is*g.length,b=new Uint8Array(v),w=0,S=0;S<g.length;S++)L=g[S],x=new Uint8Array(new Uint32Array([L.length]).buffer),b.set(x,w),w+=Is,b.set(L,w),w+=L.length;return m(b),[2]}})})});i.push(f)}else i.push(d.data());t!=null&&(p.group=t),e.push(p)},s=0;s<r.length;++s)a(s);return[4,Promise.all(i)];case 1:return o=u.sent(),[2,{data:T2(o),specs:e}]}})})}function nm(n,t){for(var e={},i,r=0,a=0,s=t;a<s.length;a++){var o=s[a],l=o.name,u=o.dtype,c=o.shape,h=ot(c),d=void 0;if("quantization"in o){var p=o.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error("Weight "+o.name+" with quantization "+p.dtype+" doesn't have corresponding metadata min and scale.")}else if(p.dtype==="float16"){if(u!=="float32")throw new Error("Weight "+o.name+" is quantized with "+p.dtype+" "+("which only supports weights of type float32 not "+u+"."))}else throw new Error("Weight "+o.name+" has unknown "+("quantization dtype "+p.dtype+". ")+"Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.");var f=ic[p.dtype],m=n.slice(r,r+h*f),g=p.dtype==="uint8"?new Uint8Array(m):new Uint16Array(m);if(u==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){d=new Float32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=b*p.scale+p.min}}else if(p.dtype==="float16")i===void 0&&(i=x2()),d=i(g);else throw new Error("Unsupported quantization type "+p.dtype+" for weight type float32.");else if(u==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error("Unsupported quantization type "+p.dtype+" for weight type int32.");d=new Int32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=Math.round(b*p.scale+p.min)}}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*f}else if(u==="string"){var w=ot(o.shape);d=[];for(var v=0;v<w;v++){var S=new Uint32Array(n.slice(r,r+Is))[0];r+=Is;var L=new Uint8Array(n.slice(r,r+S));d.push(L),r+=S}}else{var x=ic[u],m=n.slice(r,r+h*x);if(u==="float32")d=new Float32Array(m);else if(u==="int32")d=new Int32Array(m);else if(u==="bool")d=new Uint8Array(m);else if(u==="complex64"){d=new Float32Array(m);for(var C=new Float32Array(d.length/2),R=new Float32Array(d.length/2),v=0;v<C.length;v++)C[v]=d[v*2],R[v]=d[v*2+1];var D=si(C,c,"float32"),k=si(R,c,"float32");e[l]=ri(D,k)}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*x}u!=="complex64"&&(e[l]=si(d,c,u))}return e}function T2(n){if(n===null)throw new Error("Invalid input value: "+JSON.stringify(n));var t=0,e=[];n.forEach(function(a){if(t+=a.byteLength,e.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error("Unsupported TypedArray subtype: "+a.constructor.name)});var i=new Uint8Array(t),r=0;return e.forEach(function(a){i.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),i.buffer}var rc=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function im(n){return rc?Buffer.byteLength(n):new Blob([n]).size}function C2(n){if(rc)return Buffer.from(n).toString("base64");for(var t=new Uint8Array(n),e="",i=0,r=t.length;i<r;i++)e+=String.fromCharCode(t[i]);return btoa(e)}function R2(n){if(rc){var t=Buffer.from(n,"base64");return t.buffer.slice(t.byteOffset,t.byteOffset+t.byteLength)}for(var e=atob(n),i=new Uint8Array(e.length),r=0;r<e.length;++r)i.set([e.charCodeAt(r)],r);return i.buffer}function ac(n){if(n.length===1)return n[0];var t=0;n.forEach(function(r){t+=r.byteLength});var e=new Uint8Array(t),i=0;return n.forEach(function(r){e.set(new Uint8Array(r),i),i+=r.byteLength}),e.buffer}function rm(n){var t="/";for(n=n.trim();n.endsWith(t);)n=n.slice(0,n.length-1);var e=n.split(t);return e[e.length-1]}function wa(n){if(n.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:n.modelTopology==null?0:im(JSON.stringify(n.modelTopology)),weightSpecsBytes:n.weightSpecs==null?0:im(JSON.stringify(n.weightSpecs)),weightDataBytes:n.weightData==null?0:n.weightData.byteLength}}function O2(){var n=function(i){for(var r=i<<13,a=0;(r&8388608)===0;)a-=8388608,r<<=1;return r&=~8388608,a+=947912704,r|a},t=new Uint32Array(2048);t[0]=0;for(var e=1;e<1024;e++)t[e]=n(e);for(var e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function E2(){var n=new Uint32Array(64);n[0]=0,n[31]=1199570944,n[32]=2147483648,n[63]=3347054592;for(var t=1;t<31;t++)n[t]=t<<23;for(var t=33;t<63;t++)n[t]=2147483648+(t-32<<23);return n}function D2(){for(var n=new Uint32Array(64),t=0;t<64;t++)n[t]=1024;return n[0]=n[32]=0,n}function x2(){var n=O2(),t=E2(),e=D2();return function(i){for(var r=new ArrayBuffer(4*i.length),a=new Uint32Array(r),s=0;s<i.length;s++){var o=i[s],l=n[e[o>>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var Qt=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),k2=function(n){return Qt.registerSaveRouter(n)},F2=function(n){return Qt.registerLoadRouter(n)},W2=function(n){return Qt.getSaveHandlers(n)},U2=function(n,t){return Qt.getLoadHandlers(n,t)};var sc="tensorflowjs",oc=1,Ui="models_store",oi="model_info_store";function am(){if(!He().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function lc(n){var t=n.result;t.createObjectStore(Ui,{keyPath:"modelPath"}),t.createObjectStore(oi,{keyPath:"modelPath"})}var Ar=function(){function n(t){if(this.indexedDB=am(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(sc,oc);s.onupgradeneeded=function(){return lc(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(Ui,"readonly"),u=l.objectStore(Ui),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=wa(e),d=o.transaction(oi,"readwrite"),p=d.objectStore(oi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(Ui,"readwrite");var g=m.objectStore(Ui),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(oi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),sm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Ar.URL_SCHEME))?B2(n.slice(Ar.URL_SCHEME.length)):null};Qt.registerSaveRouter(sm);Qt.registerLoadRouter(sm);function B2(n){return new Ar(n)}function z2(n){return n.startsWith(Ar.URL_SCHEME)?n.slice(Ar.URL_SCHEME.length):n}var P2=function(){function n(){this.indexedDB=am()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(sc,oc);a.onupgradeneeded=function(){return lc(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(oi,"readonly"),l=o.objectStore(oi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h<d.length;h++){var p=d[h];c[p.modelPath]=p.modelArtifactsInfo}i(c)},u.onerror=function(c){return s.close(),r(u.error)},o.oncomplete=function(){return s.close()}},a.onerror=function(s){return r(a.error)}})]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e=this;return pe(this,function(i){return t=z2(t),[2,new Promise(function(r,a){var s=e.indexedDB.open(sc,oc);s.onupgradeneeded=function(){return lc(s)},s.onsuccess=function(){var o=s.result,l=o.transaction(oi,"readwrite"),u=l.objectStore(oi),c=u.get(t),h;c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+t+"' in IndexedDB."));var d=u.delete(t),p=function(){h=o.transaction(Ui,"readwrite");var f=h.objectStore(Ui),m=f.delete(t);m.onsuccess=function(){return r(c.result.modelArtifactsInfo)},m.onerror=function(g){return a(c.error)}};d.onsuccess=p,d.onerror=function(f){return p(),o.close(),a(c.error)}},c.onerror=function(d){return o.close(),a(c.error)},l.oncomplete=function(){h==null?o.close():h.oncomplete=function(){return o.close()}}},s.onerror=function(o){return a(s.error)}})]})})},n}();var Yn="/",Tr="tensorflowjs_models",om="info",_2="model_topology",M2="weight_specs",H2="weight_data",V2="model_metadata";function lm(n){return{info:[Tr,n,om].join(Yn),topology:[Tr,n,_2].join(Yn),weightSpecs:[Tr,n,M2].join(Yn),weightData:[Tr,n,H2].join(Yn),modelMetadata:[Tr,n,V2].join(Yn)}}function q2(n){var t=n.split(Yn);if(t.length<3)throw new Error("Invalid key format: "+n);return t.slice(1,t.length-1).join(Yn)}function G2(n){return n.startsWith(Nr.URL_SCHEME)?n.slice(Nr.URL_SCHEME.length):n}var Nr=function(){function n(t){if(!He().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=lm(this.modelPath)}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r;return pe(this,function(a){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");e=JSON.stringify(t.modelTopology),i=JSON.stringify(t.weightSpecs),r=wa(t);try{return this.LS.setItem(this.keys.info,JSON.stringify(r)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,i),this.LS.setItem(this.keys.weightData,C2(t.weightData)),this.LS.setItem(this.keys.modelMetadata,JSON.stringify({format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata})),[2,{modelArtifactsInfo:r}]}catch(s){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error("Failed to save model '"+this.modelPath+"' to local storage: size quota being exceeded is a possible cause of this failure: "+("modelTopologyBytes="+r.modelTopologyBytes+", ")+("weightSpecsBytes="+r.weightSpecsBytes+", ")+("weightDataBytes="+r.weightDataBytes+"."))}return[2]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o;return pe(this,function(l){if(t=JSON.parse(this.LS.getItem(this.keys.info)),t==null)throw new Error("In local storage, there is no model with name '"+this.modelPath+"'");if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");if(e={},i=JSON.parse(this.LS.getItem(this.keys.topology)),i==null)throw new Error("In local storage, the topology of model '"+this.modelPath+"' is missing.");if(e.modelTopology=i,r=JSON.parse(this.LS.getItem(this.keys.weightSpecs)),r==null)throw new Error("In local storage, the weight specs of model '"+this.modelPath+"' are missing.");if(e.weightSpecs=r,a=this.LS.getItem(this.keys.modelMetadata),a!=null&&(s=JSON.parse(a),e.format=s.format,e.generatedBy=s.generatedBy,e.convertedBy=s.convertedBy,e.userDefinedMetadata=s.userDefinedMetadata),o=this.LS.getItem(this.keys.weightData),o==null)throw new Error("In local storage, the binary weight values of model "+("'"+this.modelPath+"' are missing."));return e.weightData=R2(o),[2,e]})})},n.URL_SCHEME="localstorage://",n}(),um=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Nr.URL_SCHEME))?Y2(n.slice(Nr.URL_SCHEME.length)):null};Qt.registerSaveRouter(um);Qt.registerLoadRouter(um);function Y2(n){return new Nr(n)}var K2=function(){function n(){E(He().getBool("IS_BROWSER"),function(){return"Current environment is not a web browser"}),E(typeof window=="undefined"||typeof window.localStorage!="undefined",function(){return"Current browser does not appear to support localStorage"}),this.LS=window.localStorage}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s;return pe(this,function(o){for(t={},e=Tr+Yn,i=Yn+om,r=0;r<this.LS.length;++r)a=this.LS.key(r),a.startsWith(e)&&a.endsWith(i)&&(s=q2(a),t[s]=JSON.parse(this.LS.getItem(a)));return[2,t]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){if(t=G2(t),e=lm(t),this.LS.getItem(e.info)==null)throw new Error("Cannot find model at path '"+t+"'");return i=JSON.parse(this.LS.getItem(e.info)),this.LS.removeItem(e.info),this.LS.removeItem(e.topology),this.LS.removeItem(e.weightSpecs),this.LS.removeItem(e.weightData),[2,i]})})},n}();var xr="://",li=function(){function n(){this.managers={}}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerManager=function(t,e){E(t!=null,function(){return"scheme must not be undefined or null."}),t.endsWith(xr)&&(t=t.slice(0,t.indexOf(xr))),E(t.length>0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function As(n){if(n.indexOf(xr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+li.getSchemes().join(",")));return{scheme:n.split(xr)[0],path:n.split(xr)[1]}}function cm(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=Qt.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=Qt.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=As(n).scheme,l=As(n).path,u=o===As(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,li.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,li.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function j2(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=li.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e<i.length?(r=i[e],[4,li.getManager(r).listModels()]):[3,4];case 2:a=l.sent();for(s in a)o=r+xr+s,t[o]=a[s];l.label=3;case 3:return e++,[3,1];case 4:return[2,t]}})})}function $2(n){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){return t=As(n),e=li.getManager(t.scheme),[2,e.removeModel(t.path)]})})}function X2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!1,[2,cm(n,t,e)]})})}function J2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!0,[2,cm(n,t,e)]})})}var Z2=function(){function n(){}return n.prototype.fetch=function(t,e){return fetch(t,e)},n.prototype.now=function(){return performance.now()},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Browser's encoder only supports utf-8, but got "+e);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return new TextDecoder(e).decode(t)},n}();if(He().get("IS_BROWSER")){He().setPlatform("browser",new Z2);try{li.registerManager(Nr.URL_SCHEME,new K2)}catch(n){}try{li.registerManager(Ar.URL_SCHEME,new P2)}catch(n){}}var Q2={importFetch:function(){return Cp()}},uc,eA=function(){function n(){this.util=Rp(),this.textEncoder=new this.util.TextEncoder}return n.prototype.fetch=function(t,e){return He().global.fetch!=null?He().global.fetch(t,e):(uc==null&&(uc=Q2.importFetch()),uc(t,e))},n.prototype.now=function(){var t=process.hrtime();return t[0]*1e3+t[1]/1e6},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Node built-in encoder only supports utf-8, but got "+e);return this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)},n}();He().get("IS_NODE")&&He().setPlatform("node",new eA);function On(n,t,e){return t===void 0&&(t="float32"),t=t||"float32",$u(n),new Ss(n,t,e)}function tA(n,t){var e=O(n,"x","cast");if(!Pf(t))throw new Error("Failed to cast to unknown dtype "+t);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");var i={x:e},r={dtype:t};return z.runKernelFunc(function(a){return a.cast(e,t)},i,null,fs,r)}var ue=U({cast_:tA});function nA(n){var t=O(n,"x","clone",null),e=function(){return z.makeTensorFromDataId(t.dataId,t.shape,t.dtype)},i={x:t};return z.runKernelFunc(e,i,null,Pl)}var Bi=U({clone_:nA});function hm(n,t){t===void 0&&(t=!1),console.log(n.toString(t))}Jf();var iA={buffer:On,cast:ue,clone:Bi,print:hm};d2(iA);var rA="model",aA=".json",sA=".weights.bin";function dm(n){return new Promise(function(t){return setTimeout(t)}).then(n)}var cc=function(){function n(t){if(!He().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(n.URL_SCHEME)&&(t=t.slice(n.URL_SCHEME.length)),(t==null||t.length===0)&&(t=rA),this.modelTopologyFileName=t+aA,this.weightDataFileName=t+sA}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");if(e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"})),!(t.modelTopology instanceof ArrayBuffer))return[3,1];throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");case 1:return i=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:i},a=window.URL.createObjectURL(new Blob([JSON.stringify(r)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor,s.download=this.modelTopologyFileName,s.href=a,[4,dm(function(){return s.dispatchEvent(new MouseEvent("click"))})];case 2:return l.sent(),t.weightData!=null?(o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor,o.download=this.weightDataFileName,o.href=e,[4,dm(function(){return o.dispatchEvent(new MouseEvent("click"))})]):[3,4];case 3:l.sent(),l.label=4;case 4:return[2,{modelArtifactsInfo:wa(t)}]}})})},n.URL_SCHEME="downloads://",n}(),oA=function(){function n(t){if(t==null||t.length<1)throw new Error("When calling browserFiles, at least 1 file is required, "+("but received "+t));this.files=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i=this;return pe(this,function(r){return t=this.files[0],e=this.files.slice(1),[2,new Promise(function(a,s){var o=new FileReader;o.onload=function(l){var u=JSON.parse(l.target.result),c=u.modelTopology;if(c==null){s(new Error("modelTopology field is missing from file "+t.name));return}e.length===0&&a({modelTopology:c});var h=u.weightsManifest;if(h==null){s(new Error("weightManifest field is missing from file "+t.name));return}var d;try{d=i.checkManifestAndWeightFiles(h,e)}catch(g){s(g);return}var p=[],f=[],m=[];h.forEach(function(g){g.paths.forEach(function(v){f.push(v),m.push(null)}),p.push.apply(p,g.weights)}),h.forEach(function(g){g.paths.forEach(function(v){var b=new FileReader;b.onload=function(w){var S=w.target.result,L=f.indexOf(v);m[L]=S,m.indexOf(null)===-1&&a({modelTopology:c,weightSpecs:p,weightData:ac(m),format:u.format,generatedBy:u.generatedBy,convertedBy:u.convertedBy,userDefinedMetadata:u.userDefinedMetadata})},b.onerror=function(w){return s("Failed to weights data from file of path '"+v+"'.")},b.readAsArrayBuffer(d[v])})})},o.onerror=function(l){return s("Failed to read model topology and weights manifest JSON "+("from file '"+t.name+"'. BrowserFiles supports loading ")+"Keras-style tf.Model artifacts only.")},o.readAsText(t)})]})})},n.prototype.checkManifestAndWeightFiles=function(t,e){for(var i=[],r=e.map(function(u){return rm(u.name)}),a={},s=0,o=t;s<o.length;s++){var l=o[s];l.paths.forEach(function(u){var c=rm(u);if(i.indexOf(c)!==-1)throw new Error("Duplicate file basename found in weights manifest: "+("'"+c+"'"));if(i.push(c),r.indexOf(c)===-1)throw new Error("Weight file with basename '"+c+"' is not provided.");a[u]=e[r.indexOf(c)]})}if(i.length!==e.length)throw new Error("Mismatch in the number of files in weights manifest "+("("+i.length+") and the number of weight files provided ")+("("+e.length+")."));return a},n}(),uA=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(cc.URL_SCHEME))?lA(n.slice(cc.URL_SCHEME.length)):null};Qt.registerSaveRouter(uA);function lA(n){return n===void 0&&(n="model"),new cc(n)}function cA(n){return new oA(n)}function pm(n,t,e,i){s(n),e=e??0,i=i??1,o(e,i);var r=0,a=function(l){return l.then(function(u){var c=e+ ++r/n.length*(i-e);return t(c),u}),l};function s(l){E(l!=null&&Array.isArray(l)&&l.length>0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function fm(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?He().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,pm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,pm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function hA(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return fm(o,{requestInit:i})},a=mm(r),[2,a(n,t,e)]})})}function mm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=ic[S]*ot(w.shape),x=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(x(),o[R]=!0)}):x(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. 
-`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w<v;w++)b+=d[f+w].byteLength;for(var S=new ArrayBuffer(b),L=new Uint8Array(S),x=0,C=0;C<v;C++){var R=new Uint8Array(d[f+C]);L.set(R,x),x+=R.byteLength}var D=s[g];D.forEach(function(k){var W=S.slice(k.groupOffset,k.groupOffset+k.sizeBytes),F=nm(W,[k.manifestEntry]);for(var P in F)p[P]=F[P]}),f+=v}),[2,p]}})})}}var dA="application/octet-stream",pA="application/json",gm=function(){function n(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(E(typeof e.fetchFunc=="function",function(){return"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"}),this.fetch=e.fetchFunc):this.fetch=He().platform.fetch,E(t!=null&&t.length>0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:pA}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:dA}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:wa(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=fA(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;l<u.length;l++)c=u[l],o.push.apply(o,c.weights);for(h=[],d=[],p=0,f=t;p<f.length;p++)for(m=f[p],g=0,v=m.paths;g<v.length;g++)b=v[g],this.weightUrlConverter!=null?d.push(this.weightUrlConverter(b)):h.push(s+b+a);return this.weightUrlConverter?(S=(w=h.push).apply,L=[h],[4,Promise.all(d)]):[3,2];case 1:S.apply(w,L.concat([C.sent()])),C.label=2;case 2:return[4,fm(h,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress})];case 3:return x=C.sent(),[2,[o,ac(x)]]}})})},n.URL_SCHEME_REGEX=/^https?:\/\//,n}();function fA(n){var t=n.lastIndexOf("/"),e=n.lastIndexOf("?"),i=n.substring(0,t),r=e>t?n.substring(e):"";return[i+"/",r]}function hc(n){return n.match(gm.URL_SCHEME_REGEX)!=null}var vm=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return hc(i)}):e=hc(n),e?dc(n,t):null};Qt.registerSaveRouter(vm);Qt.registerLoadRouter(vm);function dc(n,t){return new gm(n,t)}function mA(n,t){return dc(n,t)}var pc=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),gA=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function vA(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new pc(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new pc({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new pc({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function yA(n){return new gA(n)}var bA={__proto__:null,browserFiles:cA,browserHTTPRequest:mA,concatenateArrayBuffers:ac,decodeWeights:nm,encodeWeights:N2,fromMemory:vA,getLoadHandlers:U2,getModelArtifactsInfoForJSON:wa,getSaveHandlers:W2,http:dc,isHTTPScheme:hc,loadWeights:hA,registerLoadRouter:F2,registerSaveRouter:k2,weightsLoaderFactory:mm,withSaveHandler:yA,copyModel:X2,listModels:j2,moveModel:J2,removeModel:$2};function wA(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=Wf(t,e.size),E(e.size===ot(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,uu,r)}var V=U({reshape_:wA});function SA(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=at(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=ot(h),f=ot(d);E(ln(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(x,C){return C([g,v]),x.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,ml,S);return V(L,m)}var We=U({matMul_:SA});function LA(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,iu,l)}var Ts=U({oneHot_:LA});function IA(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a<e.rank,function(){return"All entries in 'perm' must be between 0 and "+(e.rank-1)+(" but got "+t)})}),e.rank<=1)return e.clone();var i={x:e},r={perm:t};return z.runKernelFunc(function(a){return a.transpose(e,t)},i,null,Fu,r)}var ut=U({transpose_:IA});function AA(n,t,e){var i=O(n,"labels","confusionMatrix"),r=O(t,"predictions","confusionMatrix");E(e==null||e>0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=Ts(ue(i,"int32"),e),s=Ts(ue(r,"int32"),e),o=ut(a);return ue(We(o,s),"int32")}var TA=U({confusionMatrix_:AA});var NA={__proto__:null,confusionMatrix:TA};function ym(n,t,e){if(Fi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=Rn(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}var Cr;function xA(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState<o)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}var l=qu(Pu,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(Pu,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Cr==null&&(Cr=document.createElement("canvas").getContext("2d")),Cr.canvas.width=d,Cr.canvas.height=p,Cr.drawImage(n,0,0,d,p),f=Cr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v<g;v++)for(var b=0;b<t;++b)m[v*t+b]=f[v*4+b]}var w=[p,d,t];return ym(m,w,"int32")}function CA(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v;return pe(this,function(b){switch(b.label){case 0:if(e=O(n,"img","toPixels"),n instanceof Y||(i=e,e=ue(i,"int32"),i.dispose()),e.rank!==2&&e.rank!==3)throw new Error("toPixels only supports rank 2 or 3 tensors, got rank "+e.rank+".");if(r=e.shape.slice(0,2),a=r[0],s=r[1],o=e.rank===2?1:e.shape[2],o>4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h<a*s;++h){for(d=[0,0,0,255],p=0;p<o;p++){if(f=l[h*o+p],e.dtype==="float32"){if(f<0||f>1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var RA=U({fromPixels_:xA}),OA={__proto__:null,toPixels:CA,fromPixels:RA};function bm(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a<e.length-1;++a)r*=e[a];var s=n.shape,o=e.slice();o.pop();for(var l=1,a=i;a<n.rank;++a)l*=s[a],o.push(s[a]);var u=wr(n.shape).map(function(c){return c/l}).concat([1]).slice(0,i);return[o,r,l,u]}var EA={__proto__:null,prepareAndValidate:bm};function fc(n,t,e){var i=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank<r)throw new Error(a+(" update.rank < "+r+". "));if(n.length<i+(e.rank-r))throw new Error(a+(" Output shape length < "+(i+(e.rank-r))));if(e.rank!==r+n.length-i)throw new Error(a+(" update.rank != "+(r+n.length-i)));for(var s=0;s<r;++s)if(e.shape[s]!==t.shape[s])throw new Error(a+(" updates.shape["+s+"] ("+e.shape[s]+") != indices.shape["+s+"] ("+t.shape[s]+")."));for(var s=0;s<e.rank-r;++s)if(e.shape[s+r]!==n[s+i])throw new Error(a+(" updates.shape["+(s+r)+"] ("+e.shape[s+r]+") != shape["+(s+r)+"] ("+n[s+r]+")"))}function mc(n,t,e){if(t.rank<1)throw new Error("tf.scatterND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(n.rank<1)throw new Error("tf.scatterND() expects the updates to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.dtype!=="int32")throw new Error("The dtype of 'indices' should be int32, but got dtype: "+t.dtype);if(e.length<1)throw new Error("Output rank must be greater or equal to 1, but got shape: "+e);if(e.length===0){if(t.size===0)throw new Error("Indices specified for empty output. indices shape: "+t.shape);if(n.size===0)throw new Error("Updates specified for empty output. updates shape: "+n.shape)}fc(e,t,n)}function wm(n,t,e){for(var i=t.shape.length,r=i>1?t.shape[i-1]:1,a=e.length,s=1,o=r;o<a;++o)s*=e[o];var l=r<1?1:r,u=ot(t.shape)/l,c=wr(e.slice(0,r)).concat([1]),h=ot(e);return{sliceRank:r,numUpdates:u,sliceSize:s,strides:c,outputSize:h}}var DA={__proto__:null,validateUpdateShape:fc,validateInput:mc,calculateShapes:wm};function Sm(n,t,e){var i=n.shape.length;E(i===t.length,function(){return"Error in slice"+i+"D: Length of begin "+t+" must "+("match the rank of the array ("+i+").")}),E(i===e.length,function(){return"Error in slice"+i+"D: Length of size "+e+" must "+("match the rank of the array ("+i+").")});for(var r=function(s){E(t[s]+e[s]<=n.shape[s],function(){return"Error in slice"+i+"D: begin["+s+"] + size["+s+"] "+("("+(t[s]+e[s])+") would overflow input.shape["+s+"] ("+n.shape[s]+")")})},a=0;a<i;++a)r(a)}function Ns(n){for(var t=[],e=0;n>0;)n&1&&t.push(e),n/=2,e++;return t}function Lm(n,t,e){for(var i=[],r=0;r<n.length;r++)i[r]=Math.ceil((t[r]-n[r])/e[r]);return i}function Im(n,t,e,i){for(var r=n.slice(),a=r.length;a<i.length;a++)r.push(1);for(var a=0;a<e;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function Am(n,t,e){return e<=n?e:e-(t-1)}function Tm(n,t){for(var e=[],i=0;i<n;i++)e.push(t+i);return e}function Em(n,t,e,i,r,a,s,o,l){var u=n.length,c=new Array(u),h=new Array(u),d=new Array(u);if(t.length&&e>0){var p=t[0],f=e+1;c=Nm(s,p,f,i,n),h=xm(o,p,f,r,n),d=Im(a,p,f,n)}else for(var m=0;m<u;m++)c[m]=Rm(s,i,a,n,m,l),h[m]=Om(o,r,a,n,m,l),d[m]=Cm(a,m,l);return{begin:c,end:h,strides:d}}function Nm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=0;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=0),a[o]=u}return a}function xm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=Number.MAX_SAFE_INTEGER),a[o]=u}for(var c=0;c<a.length;c++){var h=r[c];a[c]<0&&(a[c]+=h),a[c]=pa(0,a[c],r[c])}return a}function Cm(n,t,e){var i=n[t];return(e&1<<t||i==null)&&(i=1),i}function Rm(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=pa(0,s,l-1),s}function Om(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=pa(0,s,l):s=pa(-1,s,l-1),s}function kA(n,t,e){for(var i=e.length,r=0;r<e.length;r++)if(e[r]>1){i=r;break}for(var r=i+1;r<e.length;r++)if(t[r]>0||e[r]!==n[r])return!1;return!0}function FA(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i<n.length-1;i++)e+=n[i]*t[i];return e}function gc(n,t,e){var i,r=n.shape.length;typeof t=="number"?i=[t].concat(new Array(r-1).fill(0)):t.length<r?i=t.concat(new Array(r-t.length).fill(0)):i=t.slice(),i.forEach(function(s){E(s!==-1,function(){return"slice() does not support negative begin indexing."})});var a;return e==null?a=new Array(r).fill(-1):typeof e=="number"?a=[e].concat(new Array(r-1).fill(-1)):e.length<r?a=e.concat(new Array(r-e.length).fill(-1)):a=e,a=a.map(function(s,o){return s>=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var Dm={__proto__:null,assertParamsValid:Sm,maskToAxes:Ns,computeOutShape:Lm,stridesWithElidedDims:Im,getNormalizedAxes:Em,startIndicesWithElidedDims:Nm,stopIndicesWithElidedDims:xm,stridesForAxis:Cm,startForAxis:Rm,stopForAxis:Om,isSliceContinous:kA,computeFlatOffset:FA,parseSliceParams:gc};var km=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),Fm=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function ui(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),Fm.register(n)}var WA={__proto__:null,Serializable:km,SerializationMap:Fm,registerClass:ui};var UA=.001,Wm=.1;function BA(n,t,e){return e==null&&(e=vc()),yc(n,t,function(i,r){return bc(i,r,e)})}function vc(){return z.backend.floatPrecision()===32?UA:Wm}function yc(n,t,e){var i=!0;if((Ot(n)||Ot(t))&&(i=!1),Ot(n)&&Ot(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=Rn(n),o=Rn(t);if(!ln(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Ot(n)?n:Wi(n),u=Ot(t)?t:Wi(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`.
+`;return v[v.length-1]=" "+v[v.length-1]+"]"+(a?"":x),v}function va(n){for(var t=[],e=0;e<n.length;e+=2)t.push([n[e],n[e+1]]);return t}var Ss=function(){function n(t,e,i){var r=this;if(this.dtype=e,this.shape=t.slice(),this.size=ot(t),i!=null){var a=i.length;E(a===this.size,function(){return"Length of values '"+a+"' does not match the size "+("inferred by the shape '"+r.size+"'.")})}if(e==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=i||Bf(e,this.size),this.strides=wr(t)}return n.prototype.set=function(t){for(var e=this,i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];i.length===0&&(i=[0]),E(i.length===this.rank,function(){return"The number of provided coordinates ("+i.length+") must "+("match the rank ("+e.rank+")")});var a=this.locToIndex(i);this.values[a]=t},n.prototype.get=function(){for(var t=[],e=0;e<arguments.length;e++)t[e]=arguments[e];t.length===0&&(t=[0]);for(var i=0,r=0,a=t;r<a.length;r++){var s=a[r];if(s<0||s>=this.shape[i]){var o="Requested out of range element at "+t+". "+("  Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u<t.length-1;++u)l+=this.strides[u]*t[u];return this.values[l]},n.prototype.locToIndex=function(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];for(var e=t[t.length-1],i=0;i<t.length-1;++i)e+=this.strides[i]*t[i];return e},n.prototype.indexToLoc=function(t){if(this.rank===0)return[];if(this.rank===1)return[t];for(var e=new Array(this.shape.length),i=0;i<e.length-1;++i)e[i]=Math.floor(t/this.strides[i]),t-=e[i]*this.strides[i];return e[e.length-1]=t,e},Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.toTensor=function(){return Cn().makeTensor(this.values,this.shape,this.dtype)},n}(),Cn=null,Ir=null;function h2(n){Cn=n}function d2(n){Ir=n}var Y=function(){function n(t,e,i,r){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=ot(t),this.strides=wr(t),this.dataId=i,this.id=r,this.rankType=this.rank<5?this.rank.toString():"higher"}return Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.buffer=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Ir.buffer(this.shape,this.dtype,t)]}})})},n.prototype.bufferSync=function(){return Ir.buffer(this.shape,this.dtype,this.dataSync())},n.prototype.array=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Sr(this.shape,t)]}})})},n.prototype.arraySync=function(){return Sr(this.shape,this.dataSync())},n.prototype.data=function(){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){switch(i.label){case 0:return this.throwIfDisposed(),t=Cn().read(this.dataId),this.dtype==="string"?[4,t]:[3,2];case 1:e=i.sent();try{return[2,e.map(function(r){return Ju(r)})]}catch(r){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}i.label=2;case 2:return[2,t]}})})},n.prototype.dataSync=function(){this.throwIfDisposed();var t=Cn().readSync(this.dataId);if(this.dtype==="string")try{return t.map(function(e){return Ju(e)})}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t},n.prototype.bytes=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return this.throwIfDisposed(),[4,Cn().read(this.dataId)];case 1:return t=e.sent(),this.dtype==="string"?[2,t]:[2,new Uint8Array(t.buffer)]}})})},n.prototype.dispose=function(){if(this.isDisposed)return;Cn().disposeTensor(this),this.isDisposedInternal=!0},Object.defineProperty(n.prototype,"isDisposed",{get:function(){return this.isDisposedInternal},enumerable:!0,configurable:!0}),n.prototype.throwIfDisposed=function(){if(this.isDisposed)throw new Error("Tensor is disposed.")},n.prototype.print=function(t){return t===void 0&&(t=!1),Ir.print(this,t)},n.prototype.clone=function(){return this.throwIfDisposed(),Ir.clone(this)},n.prototype.toString=function(t){t===void 0&&(t=!1);var e=this.dataSync();return c2(e,this.shape,this.dtype,t)},n.prototype.cast=function(t){return this.throwIfDisposed(),Ir.cast(this,t)},n.prototype.variable=function(t,e,i){return t===void 0&&(t=!0),this.throwIfDisposed(),Cn().makeVariable(this,t,e,i)},n}();Object.defineProperty(Y,Symbol.hasInstance,{value:function(n){return!!n&&n.data!=null&&n.dataSync!=null&&n.throwIfDisposed!=null}});var ya=function(n){qn(t,n);function t(e,i,r,a){var s=n.call(this,e.shape,e.dtype,e.dataId,a)||this;return s.trainable=i,s.name=r,s}return t.prototype.assign=function(e){if(e.dtype!==this.dtype)throw new Error("dtype of the new value ("+e.dtype+") and "+("previous value ("+this.dtype+") must match"));if(!on(e.shape,this.shape))throw new Error("shape of the new value ("+e.shape+") and "+("previous value ("+this.shape+") must match"));Cn().disposeTensor(this),this.dataId=e.dataId,Cn().incRef(this,null)},t.prototype.dispose=function(){Cn().disposeVariable(this),this.isDisposedInternal=!0},t}(Y);Object.defineProperty(ya,Symbol.hasInstance,{value:function(n){return n instanceof Y&&n.assign!=null&&n.assign instanceof Function}});(function(n){n.R0="R0",n.R1="R1",n.R2="R2",n.R3="R3",n.R4="R4",n.R5="R5",n.R6="R6"})(I.Rank||(I.Rank={}));var Qu;(function(n){n.float32="float32",n.int32="int32",n.bool="int32",n.complex64="complex64"})(Qu||(Qu={}));var ec;(function(n){n.float32="float32",n.int32="int32",n.bool="bool",n.complex64="complex64"})(ec||(ec={}));var tc;(function(n){n.float32="float32",n.int32="float32",n.bool="float32",n.complex64="complex64"})(tc||(tc={}));var nc;(function(n){n.float32="complex64",n.int32="complex64",n.bool="complex64",n.complex64="complex64"})(nc||(nc={}));var p2={float32:tc,int32:Qu,bool:ec,complex64:nc};function Ls(n,t){if(n==="string"||t==="string"){if(n==="string"&&t==="string")return"string";throw new Error("Can not upcast "+n+" with "+t)}return p2[n][t]}function f2(n){return Ls(n,"int32")}function at(n,t){if(n.dtype===t.dtype)return[n,t];var e=Ls(n.dtype,t.dtype);return[n.cast(e),t.cast(e)]}function jf(n,t){E(n.dtype===t.dtype,function(){return"The dtypes of the first("+n.dtype+") and"+(" second("+t.dtype+") input must match")})}function m2(n,t){return t.some(function(e){return e.id===n.id})}function ic(n){var t=[],e=new Set;return $f(n,t,e),t}function $f(n,t,e){if(n==null)return;if(n instanceof Y){t.push(n);return}if(!g2(n))return;var i=n;for(var r in i){var a=i[r];e.has(a)||(e.add(a),$f(a,t,e))}}function g2(n){return Array.isArray(n)||typeof n=="object"}var v2={__proto__:null,makeTypesMatch:at,assertTypesMatch:jf,isTensorInList:m2,getTensorsInContainer:ic};var Xf=function(){function n(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null}}return n.prototype.dispose=function(){for(var t in this.registeredVariables)this.registeredVariables[t].dispose()},n}(),w2=function(){function n(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Xf}return n.prototype.ready=function(){return de(this,void 0,void 0,function(){var t,e,i,r;return pe(this,function(a){switch(a.label){case 0:if(this.pendingBackendInit!=null)return[2,this.pendingBackendInit.then(function(){})];if(this.backendInstance!=null)return[2];t=this.getSortedBackends(),e=0,a.label=1;case 1:return e<t.length?(i=t[e],[4,this.initializeBackend(i).success]):[3,5];case 2:return r=a.sent(),r?[4,this.setBackend(i)]:[3,4];case 3:return a.sent(),[2];case 4:return e++,[3,1];case 5:throw new Error("Could not initialize any backends, all backend initializations failed.")}})})},Object.defineProperty(n.prototype,"backend",{get:function(){if(this.pendingBackendInit!=null)throw new Error("Backend '"+this.backendName+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");if(this.backendInstance==null){var t=this.initializeBackendsAndReturnBest(),e=t.name,i=t.asyncInit;if(i)throw new Error("The highest priority backend '"+e+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");this.setBackend(e)}return this.backendInstance},enumerable:!0,configurable:!0}),n.prototype.backendNames=function(){return Object.keys(this.registryFactory)},n.prototype.findBackend=function(t){if(!(t in this.registry))if(t in this.registryFactory){var e=this.initializeBackend(t).asyncInit;if(e)return null}else return null;return this.registry[t]},n.prototype.findBackendFactory=function(t){return t in this.registryFactory?this.registryFactory[t].factory:null},n.prototype.registerBackend=function(t,e,i){return i===void 0&&(i=1),t in this.registryFactory?(console.warn(t+" backend was already registered. Reusing existing backend factory."),!1):(this.registryFactory[t]={factory:e,priority:i},!0)},n.prototype.setBackend=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s;return pe(this,function(o){switch(o.label){case 0:if(this.registryFactory[t]==null)throw new Error("Backend name '"+t+"' not found in registry");return this.backendName=t,this.registry[t]==null?(this.backendInstance=null,e=this.initializeBackend(t),i=e.success,r=e.asyncInit,r?[4,i]:[3,2]):[3,4];case 1:return s=o.sent(),[3,3];case 2:s=i,o.label=3;case 3:if(a=s,!a)return[2,!1];o.label=4;case 4:return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new s2(this.backendInstance),[2,!0]}})})},n.prototype.setupRegisteredKernels=function(){var t=this,e=ms(this.backendName);e.forEach(function(i){i.setupFunc!=null&&i.setupFunc(t.backendInstance)})},n.prototype.disposeRegisteredKernels=function(t){var e=this,i=ms(t);i.forEach(function(r){r.disposeFunc!=null&&r.disposeFunc(e.registry[t])})},n.prototype.initializeBackend=function(t){var e=this,i=this.registryFactory[t];if(i==null)throw new Error("Cannot initialize backend "+t+", no registration found.");try{var r=i.factory();if(r&&!(r instanceof Op)&&typeof r.then=="function"){var a=++this.pendingBackendInitId,s=r.then(function(o){return a<e.pendingBackendInitId?!1:(e.registry[t]=o,e.pendingBackendInit=null,!0)}).catch(function(o){return a<e.pendingBackendInitId||(e.pendingBackendInit=null,console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message)),!1});return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=r,{success:!0,asyncInit:!1}}catch(o){return console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message),{success:!1,asyncInit:!1}}},n.prototype.removeBackend=function(t){if(!(t in this.registryFactory))throw new Error(t+" backend not found in registry");this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)},n.prototype.getSortedBackends=function(){var t=this;if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(function(e,i){return t.registryFactory[i].priority-t.registryFactory[e].priority})},n.prototype.initializeBackendsAndReturnBest=function(){for(var t=this.getSortedBackends(),e=0;e<t.length;e++){var i=t[e],r=this.initializeBackend(i),a=r.success,s=r.asyncInit;if(s||a)return{name:i,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")},n.prototype.moveData=function(t,e){var i=this.state.tensorInfo.get(e),r=i.backend,a=this.readSync(e);r.disposeData(e),i.backend=t,t.move(e,a,i.shape,i.dtype),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++},n.prototype.tidy=function(t,e){var i=this,r=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");r=t}var a;return this.scopedRun(function(){return i.startScope(r)},function(){return i.endScope(a)},function(){return a=e(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a})},n.prototype.scopedRun=function(t,e,i){t();try{var r=i();return e(),r}catch(a){throw e(),a}},n.prototype.nextTensorId=function(){return n.nextTensorId++},n.prototype.nextVariableId=function(){return n.nextVariableId++},n.prototype.clone=function(t){var e=this.makeTensorFromDataId(t.dataId,t.shape,t.dtype),i={x:t},r=function(s){return{x:function(){var o="float32",l={x:s},u={dtype:o};return z.runKernelFunc(function(c){return c.cast(s,o)},l,null,fs,u)}}},a=[];return this.addTapeNode(this.state.activeScope.name,i,[e],r,a,{}),e},n.prototype.runKernel=function(t,e,i,r,a){var s=null,o=null;return this.runKernelFunc(s,e,o,t,i,r,a)},n.prototype.shouldCheckForMemLeaks=function(){return this.ENV.getBool("IS_TEST")},n.prototype.checkKernelForMemLeak=function(t,e,i){var r=this.backend.numDataIds(),a=0;i.forEach(function(l){a+=l.dtype==="complex64"?3:1});var s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=r-e-a-s;if(o>0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=Gu(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var x=l.getTensorsForGradient(r,e,L);if(x==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});x=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(x)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=Yu(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&ni(t[0])&&(a=t.map(function(c){return Ku(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Hf(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new ya(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Mf(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof ya||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a<s.length?(o=s[a],l=o,[4,o.kernelTimeMs]):[3,6];case 3:return l.kernelTimeMs=c.sent(),u=o,[4,o.extraInfo];case 4:u.extraInfo=c.sent(),c.label=5;case 5:return a++,[3,2];case 6:return[2,this.state.activeProfile]}})})},n.prototype.isTapeOn=function(){return this.state.gradientDepth>0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=Yu(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=Lr(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=ic(t),r=new Set(i.map(function(l){return l.id})),a=0;a<this.state.activeScope.track.length;a++){var s=this.state.activeScope.track[a];!s.kept&&!r.has(s.id)&&s.dispose()}var o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],i.forEach(function(l){!l.kept&&l.scopeId===o.id&&e.track(l)})},n.prototype.gradients=function(t,e,i,r){var a=this;if(r===void 0&&(r=!1),E(e.length>0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=o2(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??y2(s.shape),l2(l,o,function(c){return a.tidy(c)},b2);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h<d.length;h++){var p=d[h];p.dispose()}}),a.state.activeTape=null),{value:s,grads:u}})},n.prototype.customGrad=function(t){var e=this;return E(ii(t),function(){return"The f passed in customGrad(f) must be a function."}),function(){for(var i=[],r=0;r<arguments.length;r++)i[r]=arguments[r];E(i.every(function(o){return o instanceof Y}),function(){return"The args passed in customGrad(f)(x1, x2,...) must all be tensors"});var a,s={};return i.forEach(function(o,l){s[l]=o}),e.runKernelFunc(function(o,l){return a=t.apply(void 0,i.concat([l])),E(a.value instanceof Y,function(){return"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"}),E(ii(a.gradFunc),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."}),a.value},s,function(o,l){var u=a.gradFunc(o,l),c=Array.isArray(u)?u:[u];E(c.length===i.length,function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."}),E(c.every(function(d){return d instanceof Y}),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."});var h={};return c.forEach(function(d,p){h[p]=function(){return d}}),h})}},n.prototype.readSync=function(t){var e=this.state.tensorInfo.get(t);return e.backend.readSync(t)},n.prototype.read=function(t){var e=this.state.tensorInfo.get(t);return e.backend.read(t)},n.prototype.time=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return e=$u(),[4,this.backend.time(t)];case 1:return i=r.sent(),i.wallMs=$u()-e,[2,i]}})})},n.prototype.track=function(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t},Object.defineProperty(n.prototype,"registeredVariables",{get:function(){return this.state.registeredVariables},enumerable:!0,configurable:!0}),n.prototype.reset=function(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Xf;for(var t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null},n.nextTensorId=0,n.nextVariableId=0,n}();function y2(n){var t=ju(ot(n),"float32");return z.makeTensor(t,n,"float32")}function Jf(){var n=kp();if(n._tfengine==null){var t=new Dp(n);n._tfengine=new w2(t)}return WI(n._tfengine.ENV),h2(function(){return n._tfengine}),n._tfengine}var z=Jf();function b2(n,t){var e={a:n,b:t};return z.runKernelFunc(function(i,r){var a=i.add(n,t);return r([n,t]),a},e,null,ps)}function S2(){return typeof navigator!="undefined"&&navigator!=null}function L2(){if(S2()){var n=navigator.userAgent||navigator.vendor||window.opera;return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(n)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(n.substr(0,4))}return!1}function Zf(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var I2={__proto__:null,isMobile:L2,isBrowser:Zf};var Gn=He();Gn.registerFlag("DEBUG",function(){return!1},function(n){n&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});Gn.registerFlag("IS_BROWSER",function(){return Zf()});Gn.registerFlag("IS_NODE",function(){return typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined"});Gn.registerFlag("IS_CHROME",function(){return typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor)});Gn.registerFlag("PROD",function(){return!1});Gn.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",function(){return Gn.getBool("DEBUG")});Gn.registerFlag("DEPRECATION_WARNINGS_ENABLED",function(){return!0});Gn.registerFlag("IS_TEST",function(){return!1});function Rn(n,t){var e=n;if(Ot(n))return t==="string"?[]:[n.length];if(!Array.isArray(n))return[];for(var i=[];Array.isArray(e)||Ot(e)&&t!=="string";)i.push(e.length),e=e[0];return Array.isArray(n)&&He().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Qf(n,i,[]),i}function Qf(n,t,e){if(e=e||[],!Array.isArray(n)&&!Ot(n)){E(t.length===0,function(){return"Element arr["+e.join("][")+"] is a primitive, "+("but should be an array/TypedArray of "+t[0]+" elements")});return}E(t.length>0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r<n.length;++r)Qf(n[r],i,e.concat(r))}function em(n,t,e,i){if(n==null)return;if(n!=="numeric"&&n!==t||n==="numeric"&&t==="string")throw new Error("Argument '"+e+"' passed to '"+i+"' must "+("be "+n+" tensor, but got "+t+" tensor"))}function O(n,t,e,i){if(i===void 0&&(i="numeric"),n instanceof Y)return em(i,n.dtype,t,e),n;var r=vs(n);if(r!=="string"&&["bool","int32","float32"].indexOf(i)>=0&&(r=i),em(i,r,t,e),n==null||!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=Rn(n,r);!Ot(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?bs(n,r):Wi(n,[],o);return z.makeTensor(l,s,r)}function ba(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var tm="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+tm;var r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];z.startScope(e);try{var o=i.apply(void 0,a);return o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),z.endScope(o),o}catch(l){throw z.endScope(null),l}};return Object.defineProperty(r,"name",{value:e,configurable:!0}),r}function A2(n,t){var e=O(n,"real","complex"),i=O(t,"imag","complex");Be(e.shape,i.shape,"real and imag shapes, "+e.shape+" and "+i.shape+", must match in call to tf.complex().");var r=function(s){return s.complex(e,i)},a={real:e,imag:i};return z.runKernelFunc(r,a,null,Pp)}var ri=U({complex_:A2});function ai(n,t,e,i){if(i==null&&(i=vs(n)),i==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){Xu(t);var r=ot(t),a=ot(e);E(r===a,function(){return"Based on the provided shape, ["+t+"], the tensor should have "+(r+" values but has "+a)});for(var s=0;s<e.length;++s){var o=e[s],l=s===e.length-1?o!==ot(t.slice(s)):!0;E(e[s]===t[s]||!l,function(){return"Error creating a new Tensor. Inferred shape "+("("+e+") does not match the provided ")+("shape ("+t+"). ")})}}return!Ot(n)&&!Array.isArray(n)&&(n=[n]),t=t||e,n=i!=="string"?bs(n,i):Wi(n,[],!0),z.makeTensor(n,t,i)}function si(n,t,e){var i=Rn(n,e);return ai(n,t,i,e)}var rc={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var Is=4;function N2(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l=this;return pe(this,function(u){switch(u.label){case 0:for(e=[],i=[],r=Array.isArray(n)?n.map(function(c){return c.name}):Object.keys(n),a=function(c){var h=r[c],d=Array.isArray(n)?n[c].tensor:n[h];if(d.dtype!=="float32"&&d.dtype!=="int32"&&d.dtype!=="bool"&&d.dtype!=="string"&&d.dtype!=="complex64")throw new Error("Unsupported dtype in weight '"+h+"': "+d.dtype);var p={name:h,shape:d.shape,dtype:d.dtype};if(d.dtype==="string"){var f=new Promise(function(m){return de(l,void 0,void 0,function(){var g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:return[4,d.bytes()];case 1:for(g=C.sent(),v=g.reduce(function(R,D){return R+D.length},0)+Is*g.length,b=new Uint8Array(v),w=0,S=0;S<g.length;S++)L=g[S],x=new Uint8Array(new Uint32Array([L.length]).buffer),b.set(x,w),w+=Is,b.set(L,w),w+=L.length;return m(b),[2]}})})});i.push(f)}else i.push(d.data());t!=null&&(p.group=t),e.push(p)},s=0;s<r.length;++s)a(s);return[4,Promise.all(i)];case 1:return o=u.sent(),[2,{data:T2(o),specs:e}]}})})}function nm(n,t){for(var e={},i,r=0,a=0,s=t;a<s.length;a++){var o=s[a],l=o.name,u=o.dtype,c=o.shape,h=ot(c),d=void 0;if("quantization"in o){var p=o.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error("Weight "+o.name+" with quantization "+p.dtype+" doesn't have corresponding metadata min and scale.")}else if(p.dtype==="float16"){if(u!=="float32")throw new Error("Weight "+o.name+" is quantized with "+p.dtype+" "+("which only supports weights of type float32 not "+u+"."))}else throw new Error("Weight "+o.name+" has unknown "+("quantization dtype "+p.dtype+". ")+"Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.");var f=rc[p.dtype],m=n.slice(r,r+h*f),g=p.dtype==="uint8"?new Uint8Array(m):new Uint16Array(m);if(u==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){d=new Float32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=b*p.scale+p.min}}else if(p.dtype==="float16")i===void 0&&(i=x2()),d=i(g);else throw new Error("Unsupported quantization type "+p.dtype+" for weight type float32.");else if(u==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error("Unsupported quantization type "+p.dtype+" for weight type int32.");d=new Int32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=Math.round(b*p.scale+p.min)}}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*f}else if(u==="string"){var w=ot(o.shape);d=[];for(var v=0;v<w;v++){var S=new Uint32Array(n.slice(r,r+Is))[0];r+=Is;var L=new Uint8Array(n.slice(r,r+S));d.push(L),r+=S}}else{var x=rc[u],m=n.slice(r,r+h*x);if(u==="float32")d=new Float32Array(m);else if(u==="int32")d=new Int32Array(m);else if(u==="bool")d=new Uint8Array(m);else if(u==="complex64"){d=new Float32Array(m);for(var C=new Float32Array(d.length/2),R=new Float32Array(d.length/2),v=0;v<C.length;v++)C[v]=d[v*2],R[v]=d[v*2+1];var D=si(C,c,"float32"),k=si(R,c,"float32");e[l]=ri(D,k)}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*x}u!=="complex64"&&(e[l]=si(d,c,u))}return e}function T2(n){if(n===null)throw new Error("Invalid input value: "+JSON.stringify(n));var t=0,e=[];n.forEach(function(a){if(t+=a.byteLength,e.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error("Unsupported TypedArray subtype: "+a.constructor.name)});var i=new Uint8Array(t),r=0;return e.forEach(function(a){i.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),i.buffer}var ac=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function im(n){return ac?Buffer.byteLength(n):new Blob([n]).size}function C2(n){if(ac)return Buffer.from(n).toString("base64");for(var t=new Uint8Array(n),e="",i=0,r=t.length;i<r;i++)e+=String.fromCharCode(t[i]);return btoa(e)}function R2(n){if(ac){var t=Buffer.from(n,"base64");return t.buffer.slice(t.byteOffset,t.byteOffset+t.byteLength)}for(var e=atob(n),i=new Uint8Array(e.length),r=0;r<e.length;++r)i.set([e.charCodeAt(r)],r);return i.buffer}function sc(n){if(n.length===1)return n[0];var t=0;n.forEach(function(r){t+=r.byteLength});var e=new Uint8Array(t),i=0;return n.forEach(function(r){e.set(new Uint8Array(r),i),i+=r.byteLength}),e.buffer}function rm(n){var t="/";for(n=n.trim();n.endsWith(t);)n=n.slice(0,n.length-1);var e=n.split(t);return e[e.length-1]}function wa(n){if(n.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:n.modelTopology==null?0:im(JSON.stringify(n.modelTopology)),weightSpecsBytes:n.weightSpecs==null?0:im(JSON.stringify(n.weightSpecs)),weightDataBytes:n.weightData==null?0:n.weightData.byteLength}}function O2(){var n=function(i){for(var r=i<<13,a=0;(r&8388608)===0;)a-=8388608,r<<=1;return r&=~8388608,a+=947912704,r|a},t=new Uint32Array(2048);t[0]=0;for(var e=1;e<1024;e++)t[e]=n(e);for(var e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function E2(){var n=new Uint32Array(64);n[0]=0,n[31]=1199570944,n[32]=2147483648,n[63]=3347054592;for(var t=1;t<31;t++)n[t]=t<<23;for(var t=33;t<63;t++)n[t]=2147483648+(t-32<<23);return n}function D2(){for(var n=new Uint32Array(64),t=0;t<64;t++)n[t]=1024;return n[0]=n[32]=0,n}function x2(){var n=O2(),t=E2(),e=D2();return function(i){for(var r=new ArrayBuffer(4*i.length),a=new Uint32Array(r),s=0;s<i.length;s++){var o=i[s],l=n[e[o>>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var Zt=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),k2=function(n){return Zt.registerSaveRouter(n)},F2=function(n){return Zt.registerLoadRouter(n)},W2=function(n){return Zt.getSaveHandlers(n)},U2=function(n,t){return Zt.getLoadHandlers(n,t)};var oc="tensorflowjs",lc=1,Ui="models_store",oi="model_info_store";function am(){if(!He().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function uc(n){var t=n.result;t.createObjectStore(Ui,{keyPath:"modelPath"}),t.createObjectStore(oi,{keyPath:"modelPath"})}var Ar=function(){function n(t){if(this.indexedDB=am(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(oc,lc);s.onupgradeneeded=function(){return uc(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(Ui,"readonly"),u=l.objectStore(Ui),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=wa(e),d=o.transaction(oi,"readwrite"),p=d.objectStore(oi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(Ui,"readwrite");var g=m.objectStore(Ui),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(oi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),sm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Ar.URL_SCHEME))?B2(n.slice(Ar.URL_SCHEME.length)):null};Zt.registerSaveRouter(sm);Zt.registerLoadRouter(sm);function B2(n){return new Ar(n)}function z2(n){return n.startsWith(Ar.URL_SCHEME)?n.slice(Ar.URL_SCHEME.length):n}var P2=function(){function n(){this.indexedDB=am()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(oc,lc);a.onupgradeneeded=function(){return uc(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(oi,"readonly"),l=o.objectStore(oi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h<d.length;h++){var p=d[h];c[p.modelPath]=p.modelArtifactsInfo}i(c)},u.onerror=function(c){return s.close(),r(u.error)},o.oncomplete=function(){return s.close()}},a.onerror=function(s){return r(a.error)}})]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e=this;return pe(this,function(i){return t=z2(t),[2,new Promise(function(r,a){var s=e.indexedDB.open(oc,lc);s.onupgradeneeded=function(){return uc(s)},s.onsuccess=function(){var o=s.result,l=o.transaction(oi,"readwrite"),u=l.objectStore(oi),c=u.get(t),h;c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+t+"' in IndexedDB."));var d=u.delete(t),p=function(){h=o.transaction(Ui,"readwrite");var f=h.objectStore(Ui),m=f.delete(t);m.onsuccess=function(){return r(c.result.modelArtifactsInfo)},m.onerror=function(g){return a(c.error)}};d.onsuccess=p,d.onerror=function(f){return p(),o.close(),a(c.error)}},c.onerror=function(d){return o.close(),a(c.error)},l.oncomplete=function(){h==null?o.close():h.oncomplete=function(){return o.close()}}},s.onerror=function(o){return a(s.error)}})]})})},n}();var Yn="/",Tr="tensorflowjs_models",om="info",_2="model_topology",M2="weight_specs",H2="weight_data",V2="model_metadata";function lm(n){return{info:[Tr,n,om].join(Yn),topology:[Tr,n,_2].join(Yn),weightSpecs:[Tr,n,M2].join(Yn),weightData:[Tr,n,H2].join(Yn),modelMetadata:[Tr,n,V2].join(Yn)}}function q2(n){var t=n.split(Yn);if(t.length<3)throw new Error("Invalid key format: "+n);return t.slice(1,t.length-1).join(Yn)}function G2(n){return n.startsWith(Nr.URL_SCHEME)?n.slice(Nr.URL_SCHEME.length):n}var Nr=function(){function n(t){if(!He().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=lm(this.modelPath)}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r;return pe(this,function(a){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");e=JSON.stringify(t.modelTopology),i=JSON.stringify(t.weightSpecs),r=wa(t);try{return this.LS.setItem(this.keys.info,JSON.stringify(r)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,i),this.LS.setItem(this.keys.weightData,C2(t.weightData)),this.LS.setItem(this.keys.modelMetadata,JSON.stringify({format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata})),[2,{modelArtifactsInfo:r}]}catch(s){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error("Failed to save model '"+this.modelPath+"' to local storage: size quota being exceeded is a possible cause of this failure: "+("modelTopologyBytes="+r.modelTopologyBytes+", ")+("weightSpecsBytes="+r.weightSpecsBytes+", ")+("weightDataBytes="+r.weightDataBytes+"."))}return[2]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o;return pe(this,function(l){if(t=JSON.parse(this.LS.getItem(this.keys.info)),t==null)throw new Error("In local storage, there is no model with name '"+this.modelPath+"'");if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");if(e={},i=JSON.parse(this.LS.getItem(this.keys.topology)),i==null)throw new Error("In local storage, the topology of model '"+this.modelPath+"' is missing.");if(e.modelTopology=i,r=JSON.parse(this.LS.getItem(this.keys.weightSpecs)),r==null)throw new Error("In local storage, the weight specs of model '"+this.modelPath+"' are missing.");if(e.weightSpecs=r,a=this.LS.getItem(this.keys.modelMetadata),a!=null&&(s=JSON.parse(a),e.format=s.format,e.generatedBy=s.generatedBy,e.convertedBy=s.convertedBy,e.userDefinedMetadata=s.userDefinedMetadata),o=this.LS.getItem(this.keys.weightData),o==null)throw new Error("In local storage, the binary weight values of model "+("'"+this.modelPath+"' are missing."));return e.weightData=R2(o),[2,e]})})},n.URL_SCHEME="localstorage://",n}(),um=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Nr.URL_SCHEME))?Y2(n.slice(Nr.URL_SCHEME.length)):null};Zt.registerSaveRouter(um);Zt.registerLoadRouter(um);function Y2(n){return new Nr(n)}var K2=function(){function n(){E(He().getBool("IS_BROWSER"),function(){return"Current environment is not a web browser"}),E(typeof window=="undefined"||typeof window.localStorage!="undefined",function(){return"Current browser does not appear to support localStorage"}),this.LS=window.localStorage}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s;return pe(this,function(o){for(t={},e=Tr+Yn,i=Yn+om,r=0;r<this.LS.length;++r)a=this.LS.key(r),a.startsWith(e)&&a.endsWith(i)&&(s=q2(a),t[s]=JSON.parse(this.LS.getItem(a)));return[2,t]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){if(t=G2(t),e=lm(t),this.LS.getItem(e.info)==null)throw new Error("Cannot find model at path '"+t+"'");return i=JSON.parse(this.LS.getItem(e.info)),this.LS.removeItem(e.info),this.LS.removeItem(e.topology),this.LS.removeItem(e.weightSpecs),this.LS.removeItem(e.weightData),[2,i]})})},n}();var xr="://",li=function(){function n(){this.managers={}}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerManager=function(t,e){E(t!=null,function(){return"scheme must not be undefined or null."}),t.endsWith(xr)&&(t=t.slice(0,t.indexOf(xr))),E(t.length>0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function As(n){if(n.indexOf(xr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+li.getSchemes().join(",")));return{scheme:n.split(xr)[0],path:n.split(xr)[1]}}function cm(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=Zt.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=Zt.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=As(n).scheme,l=As(n).path,u=o===As(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,li.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,li.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function j2(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=li.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e<i.length?(r=i[e],[4,li.getManager(r).listModels()]):[3,4];case 2:a=l.sent();for(s in a)o=r+xr+s,t[o]=a[s];l.label=3;case 3:return e++,[3,1];case 4:return[2,t]}})})}function $2(n){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){return t=As(n),e=li.getManager(t.scheme),[2,e.removeModel(t.path)]})})}function X2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!1,[2,cm(n,t,e)]})})}function J2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!0,[2,cm(n,t,e)]})})}var Z2=function(){function n(){}return n.prototype.fetch=function(t,e){return fetch(t,e)},n.prototype.now=function(){return performance.now()},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Browser's encoder only supports utf-8, but got "+e);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return new TextDecoder(e).decode(t)},n}();if(He().get("IS_BROWSER")){He().setPlatform("browser",new Z2);try{li.registerManager(Nr.URL_SCHEME,new K2)}catch(n){}try{li.registerManager(Ar.URL_SCHEME,new P2)}catch(n){}}var Q2={importFetch:function(){return Cp()}},cc,eA=function(){function n(){this.util=Rp(),this.textEncoder=new this.util.TextEncoder}return n.prototype.fetch=function(t,e){return He().global.fetch!=null?He().global.fetch(t,e):(cc==null&&(cc=Q2.importFetch()),cc(t,e))},n.prototype.now=function(){var t=process.hrtime();return t[0]*1e3+t[1]/1e6},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Node built-in encoder only supports utf-8, but got "+e);return this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)},n}();He().get("IS_NODE")&&He().setPlatform("node",new eA);function On(n,t,e){return t===void 0&&(t="float32"),t=t||"float32",Xu(n),new Ss(n,t,e)}function tA(n,t){var e=O(n,"x","cast");if(!Pf(t))throw new Error("Failed to cast to unknown dtype "+t);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");var i={x:e},r={dtype:t};return z.runKernelFunc(function(a){return a.cast(e,t)},i,null,fs,r)}var ue=U({cast_:tA});function nA(n){var t=O(n,"x","clone",null),e=function(){return z.makeTensorFromDataId(t.dataId,t.shape,t.dtype)},i={x:t};return z.runKernelFunc(e,i,null,_l)}var Bi=U({clone_:nA});function hm(n,t){t===void 0&&(t=!1),console.log(n.toString(t))}Jf();var iA={buffer:On,cast:ue,clone:Bi,print:hm};d2(iA);var rA="model",aA=".json",sA=".weights.bin";function dm(n){return new Promise(function(t){return setTimeout(t)}).then(n)}var hc=function(){function n(t){if(!He().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(n.URL_SCHEME)&&(t=t.slice(n.URL_SCHEME.length)),(t==null||t.length===0)&&(t=rA),this.modelTopologyFileName=t+aA,this.weightDataFileName=t+sA}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");if(e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"})),!(t.modelTopology instanceof ArrayBuffer))return[3,1];throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");case 1:return i=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:i},a=window.URL.createObjectURL(new Blob([JSON.stringify(r)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor,s.download=this.modelTopologyFileName,s.href=a,[4,dm(function(){return s.dispatchEvent(new MouseEvent("click"))})];case 2:return l.sent(),t.weightData!=null?(o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor,o.download=this.weightDataFileName,o.href=e,[4,dm(function(){return o.dispatchEvent(new MouseEvent("click"))})]):[3,4];case 3:l.sent(),l.label=4;case 4:return[2,{modelArtifactsInfo:wa(t)}]}})})},n.URL_SCHEME="downloads://",n}(),oA=function(){function n(t){if(t==null||t.length<1)throw new Error("When calling browserFiles, at least 1 file is required, "+("but received "+t));this.files=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i=this;return pe(this,function(r){return t=this.files[0],e=this.files.slice(1),[2,new Promise(function(a,s){var o=new FileReader;o.onload=function(l){var u=JSON.parse(l.target.result),c=u.modelTopology;if(c==null){s(new Error("modelTopology field is missing from file "+t.name));return}e.length===0&&a({modelTopology:c});var h=u.weightsManifest;if(h==null){s(new Error("weightManifest field is missing from file "+t.name));return}var d;try{d=i.checkManifestAndWeightFiles(h,e)}catch(g){s(g);return}var p=[],f=[],m=[];h.forEach(function(g){g.paths.forEach(function(v){f.push(v),m.push(null)}),p.push.apply(p,g.weights)}),h.forEach(function(g){g.paths.forEach(function(v){var b=new FileReader;b.onload=function(w){var S=w.target.result,L=f.indexOf(v);m[L]=S,m.indexOf(null)===-1&&a({modelTopology:c,weightSpecs:p,weightData:sc(m),format:u.format,generatedBy:u.generatedBy,convertedBy:u.convertedBy,userDefinedMetadata:u.userDefinedMetadata})},b.onerror=function(w){return s("Failed to weights data from file of path '"+v+"'.")},b.readAsArrayBuffer(d[v])})})},o.onerror=function(l){return s("Failed to read model topology and weights manifest JSON "+("from file '"+t.name+"'. BrowserFiles supports loading ")+"Keras-style tf.Model artifacts only.")},o.readAsText(t)})]})})},n.prototype.checkManifestAndWeightFiles=function(t,e){for(var i=[],r=e.map(function(u){return rm(u.name)}),a={},s=0,o=t;s<o.length;s++){var l=o[s];l.paths.forEach(function(u){var c=rm(u);if(i.indexOf(c)!==-1)throw new Error("Duplicate file basename found in weights manifest: "+("'"+c+"'"));if(i.push(c),r.indexOf(c)===-1)throw new Error("Weight file with basename '"+c+"' is not provided.");a[u]=e[r.indexOf(c)]})}if(i.length!==e.length)throw new Error("Mismatch in the number of files in weights manifest "+("("+i.length+") and the number of weight files provided ")+("("+e.length+")."));return a},n}(),uA=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(hc.URL_SCHEME))?lA(n.slice(hc.URL_SCHEME.length)):null};Zt.registerSaveRouter(uA);function lA(n){return n===void 0&&(n="model"),new hc(n)}function cA(n){return new oA(n)}function pm(n,t,e,i){s(n),e=e??0,i=i??1,o(e,i);var r=0,a=function(l){return l.then(function(u){var c=e+ ++r/n.length*(i-e);return t(c),u}),l};function s(l){E(l!=null&&Array.isArray(l)&&l.length>0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function fm(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?He().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,pm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,pm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function hA(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return fm(o,{requestInit:i})},a=mm(r),[2,a(n,t,e)]})})}function mm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=rc[S]*ot(w.shape),x=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(x(),o[R]=!0)}):x(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. 
+`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w<v;w++)b+=d[f+w].byteLength;for(var S=new ArrayBuffer(b),L=new Uint8Array(S),x=0,C=0;C<v;C++){var R=new Uint8Array(d[f+C]);L.set(R,x),x+=R.byteLength}var D=s[g];D.forEach(function(k){var W=S.slice(k.groupOffset,k.groupOffset+k.sizeBytes),F=nm(W,[k.manifestEntry]);for(var P in F)p[P]=F[P]}),f+=v}),[2,p]}})})}}var dA="application/octet-stream",pA="application/json",gm=function(){function n(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(E(typeof e.fetchFunc=="function",function(){return"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"}),this.fetch=e.fetchFunc):this.fetch=He().platform.fetch,E(t!=null&&t.length>0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:pA}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:dA}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:wa(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=fA(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;l<u.length;l++)c=u[l],o.push.apply(o,c.weights);for(h=[],d=[],p=0,f=t;p<f.length;p++)for(m=f[p],g=0,v=m.paths;g<v.length;g++)b=v[g],this.weightUrlConverter!=null?d.push(this.weightUrlConverter(b)):h.push(s+b+a);return this.weightUrlConverter?(S=(w=h.push).apply,L=[h],[4,Promise.all(d)]):[3,2];case 1:S.apply(w,L.concat([C.sent()])),C.label=2;case 2:return[4,fm(h,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress})];case 3:return x=C.sent(),[2,[o,sc(x)]]}})})},n.URL_SCHEME_REGEX=/^https?:\/\//,n}();function fA(n){var t=n.lastIndexOf("/"),e=n.lastIndexOf("?"),i=n.substring(0,t),r=e>t?n.substring(e):"";return[i+"/",r]}function dc(n){return n.match(gm.URL_SCHEME_REGEX)!=null}var vm=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return dc(i)}):e=dc(n),e?pc(n,t):null};Zt.registerSaveRouter(vm);Zt.registerLoadRouter(vm);function pc(n,t){return new gm(n,t)}function mA(n,t){return pc(n,t)}var fc=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),gA=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function vA(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new fc(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new fc({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new fc({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function yA(n){return new gA(n)}var bA={__proto__:null,browserFiles:cA,browserHTTPRequest:mA,concatenateArrayBuffers:sc,decodeWeights:nm,encodeWeights:N2,fromMemory:vA,getLoadHandlers:U2,getModelArtifactsInfoForJSON:wa,getSaveHandlers:W2,http:pc,isHTTPScheme:dc,loadWeights:hA,registerLoadRouter:F2,registerSaveRouter:k2,weightsLoaderFactory:mm,withSaveHandler:yA,copyModel:X2,listModels:j2,moveModel:J2,removeModel:$2};function wA(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=Wf(t,e.size),E(e.size===ot(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,cu,r)}var V=U({reshape_:wA});function SA(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=at(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=ot(h),f=ot(d);E(on(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(x,C){return C([g,v]),x.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,gl,S);return V(L,m)}var We=U({matMul_:SA});function LA(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,ru,l)}var Ts=U({oneHot_:LA});function IA(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a<e.rank,function(){return"All entries in 'perm' must be between 0 and "+(e.rank-1)+(" but got "+t)})}),e.rank<=1)return e.clone();var i={x:e},r={perm:t};return z.runKernelFunc(function(a){return a.transpose(e,t)},i,null,Wu,r)}var ut=U({transpose_:IA});function AA(n,t,e){var i=O(n,"labels","confusionMatrix"),r=O(t,"predictions","confusionMatrix");E(e==null||e>0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=Ts(ue(i,"int32"),e),s=Ts(ue(r,"int32"),e),o=ut(a);return ue(We(o,s),"int32")}var TA=U({confusionMatrix_:AA});var NA={__proto__:null,confusionMatrix:TA};function ym(n,t,e){if(Fi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=Rn(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}var Cr;function xA(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState<o)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}var l=Gu(_u,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(_u,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Cr==null&&(Cr=document.createElement("canvas").getContext("2d")),Cr.canvas.width=d,Cr.canvas.height=p,Cr.drawImage(n,0,0,d,p),f=Cr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v<g;v++)for(var b=0;b<t;++b)m[v*t+b]=f[v*4+b]}var w=[p,d,t];return ym(m,w,"int32")}function CA(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v;return pe(this,function(b){switch(b.label){case 0:if(e=O(n,"img","toPixels"),n instanceof Y||(i=e,e=ue(i,"int32"),i.dispose()),e.rank!==2&&e.rank!==3)throw new Error("toPixels only supports rank 2 or 3 tensors, got rank "+e.rank+".");if(r=e.shape.slice(0,2),a=r[0],s=r[1],o=e.rank===2?1:e.shape[2],o>4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h<a*s;++h){for(d=[0,0,0,255],p=0;p<o;p++){if(f=l[h*o+p],e.dtype==="float32"){if(f<0||f>1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var RA=U({fromPixels_:xA}),OA={__proto__:null,toPixels:CA,fromPixels:RA};function bm(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a<e.length-1;++a)r*=e[a];var s=n.shape,o=e.slice();o.pop();for(var l=1,a=i;a<n.rank;++a)l*=s[a],o.push(s[a]);var u=wr(n.shape).map(function(c){return c/l}).concat([1]).slice(0,i);return[o,r,l,u]}var EA={__proto__:null,prepareAndValidate:bm};function mc(n,t,e){var i=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank<r)throw new Error(a+(" update.rank < "+r+". "));if(n.length<i+(e.rank-r))throw new Error(a+(" Output shape length < "+(i+(e.rank-r))));if(e.rank!==r+n.length-i)throw new Error(a+(" update.rank != "+(r+n.length-i)));for(var s=0;s<r;++s)if(e.shape[s]!==t.shape[s])throw new Error(a+(" updates.shape["+s+"] ("+e.shape[s]+") != indices.shape["+s+"] ("+t.shape[s]+")."));for(var s=0;s<e.rank-r;++s)if(e.shape[s+r]!==n[s+i])throw new Error(a+(" updates.shape["+(s+r)+"] ("+e.shape[s+r]+") != shape["+(s+r)+"] ("+n[s+r]+")"))}function gc(n,t,e){if(t.rank<1)throw new Error("tf.scatterND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(n.rank<1)throw new Error("tf.scatterND() expects the updates to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.dtype!=="int32")throw new Error("The dtype of 'indices' should be int32, but got dtype: "+t.dtype);if(e.length<1)throw new Error("Output rank must be greater or equal to 1, but got shape: "+e);if(e.length===0){if(t.size===0)throw new Error("Indices specified for empty output. indices shape: "+t.shape);if(n.size===0)throw new Error("Updates specified for empty output. updates shape: "+n.shape)}mc(e,t,n)}function wm(n,t,e){for(var i=t.shape.length,r=i>1?t.shape[i-1]:1,a=e.length,s=1,o=r;o<a;++o)s*=e[o];var l=r<1?1:r,u=ot(t.shape)/l,c=wr(e.slice(0,r)).concat([1]),h=ot(e);return{sliceRank:r,numUpdates:u,sliceSize:s,strides:c,outputSize:h}}var DA={__proto__:null,validateUpdateShape:mc,validateInput:gc,calculateShapes:wm};function Sm(n,t,e){var i=n.shape.length;E(i===t.length,function(){return"Error in slice"+i+"D: Length of begin "+t+" must "+("match the rank of the array ("+i+").")}),E(i===e.length,function(){return"Error in slice"+i+"D: Length of size "+e+" must "+("match the rank of the array ("+i+").")});for(var r=function(s){E(t[s]+e[s]<=n.shape[s],function(){return"Error in slice"+i+"D: begin["+s+"] + size["+s+"] "+("("+(t[s]+e[s])+") would overflow input.shape["+s+"] ("+n.shape[s]+")")})},a=0;a<i;++a)r(a)}function Ns(n){for(var t=[],e=0;n>0;)n&1&&t.push(e),n/=2,e++;return t}function Lm(n,t,e){for(var i=[],r=0;r<n.length;r++)i[r]=Math.ceil((t[r]-n[r])/e[r]);return i}function Im(n,t,e,i){for(var r=n.slice(),a=r.length;a<i.length;a++)r.push(1);for(var a=0;a<e;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function Am(n,t,e){return e<=n?e:e-(t-1)}function Tm(n,t){for(var e=[],i=0;i<n;i++)e.push(t+i);return e}function Em(n,t,e,i,r,a,s,o,l){var u=n.length,c=new Array(u),h=new Array(u),d=new Array(u);if(t.length&&e>0){var p=t[0],f=e+1;c=Nm(s,p,f,i,n),h=xm(o,p,f,r,n),d=Im(a,p,f,n)}else for(var m=0;m<u;m++)c[m]=Rm(s,i,a,n,m,l),h[m]=Om(o,r,a,n,m,l),d[m]=Cm(a,m,l);return{begin:c,end:h,strides:d}}function Nm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=0;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=0),a[o]=u}return a}function xm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=Number.MAX_SAFE_INTEGER),a[o]=u}for(var c=0;c<a.length;c++){var h=r[c];a[c]<0&&(a[c]+=h),a[c]=pa(0,a[c],r[c])}return a}function Cm(n,t,e){var i=n[t];return(e&1<<t||i==null)&&(i=1),i}function Rm(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=pa(0,s,l-1),s}function Om(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=pa(0,s,l):s=pa(-1,s,l-1),s}function kA(n,t,e){for(var i=e.length,r=0;r<e.length;r++)if(e[r]>1){i=r;break}for(var r=i+1;r<e.length;r++)if(t[r]>0||e[r]!==n[r])return!1;return!0}function FA(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i<n.length-1;i++)e+=n[i]*t[i];return e}function vc(n,t,e){var i,r=n.shape.length;typeof t=="number"?i=[t].concat(new Array(r-1).fill(0)):t.length<r?i=t.concat(new Array(r-t.length).fill(0)):i=t.slice(),i.forEach(function(s){E(s!==-1,function(){return"slice() does not support negative begin indexing."})});var a;return e==null?a=new Array(r).fill(-1):typeof e=="number"?a=[e].concat(new Array(r-1).fill(-1)):e.length<r?a=e.concat(new Array(r-e.length).fill(-1)):a=e,a=a.map(function(s,o){return s>=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var Dm={__proto__:null,assertParamsValid:Sm,maskToAxes:Ns,computeOutShape:Lm,stridesWithElidedDims:Im,getNormalizedAxes:Em,startIndicesWithElidedDims:Nm,stopIndicesWithElidedDims:xm,stridesForAxis:Cm,startForAxis:Rm,stopForAxis:Om,isSliceContinous:kA,computeFlatOffset:FA,parseSliceParams:vc};var km=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),Fm=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function ui(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),Fm.register(n)}var WA={__proto__:null,Serializable:km,SerializationMap:Fm,registerClass:ui};var UA=.001,Wm=.1;function BA(n,t,e){return e==null&&(e=yc()),bc(n,t,function(i,r){return wc(i,r,e)})}function yc(){return z.backend.floatPrecision()===32?UA:Wm}function bc(n,t,e){var i=!0;if((Ot(n)||Ot(t))&&(i=!1),Ot(n)&&Ot(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=Rn(n),o=Rn(t);if(!on(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Ot(n)?n:Wi(n),u=Ot(t)?t:Wi(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`.
 `)+("Actual:   "+l+`.
 `)+("Expected: "+u+"."));for(var c=0;c<u.length;++c){var h=l[c],d=u[c];if(!e(h,d))throw new Error("Arrays differ: actual["+c+"] = "+h+", expected["+c+"] = "+d+`.
 `+("Actual:   "+l+`.
-`)+("Expected: "+u+"."))}}function zA(n,t){n().then(function(){return t.fail()},function(){return t()})}function PA(n,t){var e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return ni(n)||ni(n[0])||ni(t)||ni(t[0])?yc(n,e,function(i,r){return i==r}):yc(n,t,function(i,r){return bc(i,r,0)})}function _A(n,t,e){if(e==null&&(e=vc()),!bc(n,t,e))throw new Error("Numbers differ: actual === "+n+", expected === "+t)}function bc(n,t,e){return!isFinite(n)&&!isFinite(t)?!0:!(isNaN(n)||isNaN(t)||Math.abs(n-t)>e)}function MA(n,t,e){for(var i=0;i<n.length;i++)if(n[i]<t||n[i]>e)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function HA(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var VA={__proto__:null,TEST_EPSILON_FLOAT16:Wm,expectArraysClose:BA,testEpsilon:vc,expectPromiseToFail:zA,expectArraysEqual:PA,expectNumbersClose:_A,expectValuesInRange:MA,expectArrayBuffersEqual:HA};var qA="2.6.0";function GA(){He().set("PROD",!0)}function YA(){He().set("DEBUG",!0)}function KA(){He().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function It(n){He().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function jA(){z.disposeVariables()}function $A(){return z}function XA(){return z.memory()}function JA(n){return z.profile(n)}function ft(n,t){return z.tidy(n,t)}function zt(n){var t=nc(n);t.forEach(function(e){return e.dispose()})}function Um(n){return z.keep(n)}function ZA(n){return z.time(n)}function QA(n){return z.setBackend(n)}function eT(){return z.ready()}function tT(){return z.backendName}function nT(n){z.removeBackend(n)}function iT(n){return z.findBackend(n)}function rT(n){return z.findBackendFactory(n)}function aT(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function sT(){return z.backend}function oT(n,t){He().setPlatform(n,t)}function lT(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,ps)}var fe=U({add_:lT});function uT(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Wl)}var wc=U({floorDiv_:uT});function cT(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=at(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return wc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Rl,o)}var Ie=U({div_:cT});function hT(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,eu)}var J=U({mul_:hT});function dT(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,nl)}var Gt=U({abs_:dT});function pT(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,il)}var Bm=U({acos_:pT});function fT(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,rl)}var zm=U({acosh_:fT});function mT(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!ln(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,al)}var gT=U({addN_:mT});function Sc(n,t){for(var e=0;e<n.length;++e)if(n[n.length-e-1]!==t-1-e)return!1;return!0}function Pm(n,t,e){for(var i=n.length+t.length,r=[],a=0,s=0,o=0;o<i;o++)e.indexOf(o)===-1?r.push(n[a++]):r.push(t[s++]);return r}function _m(n,t){for(var e=[],i=n.length,r=0;r<i;r++)t.indexOf(r)===-1&&e.push(n[r]);var a=t.map(function(s){return n[s]});return[e,a]}function en(n,t){var e=t.map(function(i){return 1});return Pm(n,e,t)}function vT(n,t,e){E(Sc(t,e),function(){return n+" supports only inner-most axes for now. "+("Got axes "+t+" and rank-"+e+" input.")})}function tn(n,t){if(Sc(n,t))return null;for(var e=[],i=0;i<t;++i)n.indexOf(i)===-1&&e.push(i);return n.forEach(function(r){return e.push(r)}),e}function xs(n){return n.map(function(t,e){return[e,t]}).sort(function(t,e){return t[1]-e[1]}).map(function(t){return t[0]})}function En(n,t){for(var e=[],i=t-n;i<t;++i)e.push(i);return e}function yT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","all","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=tn(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.all(i,u);if(e){var d=en(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Wp,s)}var Mm=U({all_:yT});function bT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","any","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=tn(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.any(i,u);if(e){var d=en(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Up,s)}var Hm=U({any_:bT});function wT(n,t){t===void 0&&(t=0);var e=O(n,"x","argMax"),i=function(s,o){o([e]);var l=Qe(t,e.shape),u=tn(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMax(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,sl,a)}var Vm=U({argMax_:wT});function ST(n,t){t===void 0&&(t=0);var e=O(n,"x","argMin"),i=function(s,o){o([e]),t==null&&(t=0);var l=Qe(t,e.shape),u=tn(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMin(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,ol,a)}var qm=U({argMin_:ST});function LT(n){var t=O(n,"x","asin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asin(t);return r([t]),a},e,null,ll)}var Gm=U({asin_:LT});function IT(n){var t=O(n,"x","asinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asinh(t);return r([t]),a},e,null,ul)}var Ym=U({asinh_:IT});function AT(n){var t=O(n,"x","atan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atan(t);return r([t]),a},e,null,cl)}var Km=U({atan_:AT});function TT(n,t){var e,i=O(n,"a","atan2"),r=O(t,"b","atan2");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.atan2(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,dl)}var jm=U({atan2_:TT});function NT(n){var t=O(n,"x","atanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atanh(t);return r([t]),a},e,null,hl)}var $m=U({atanh_:NT});function xT(n,t,e,i,r,a){r===void 0&&(r="NHWC");var s=n[3],o=t.concat([s]),l=Sa(r);return Dn(n,o,e,a,i,null,null,l)}function Rr(n,t,e,i,r,a,s){s===void 0&&(s="channelsLast");var o=Cs(t),l=o[0],u=o[1],c;if(s==="channelsLast")c=[l,u,n[3],n[3]];else if(s==="channelsFirst")c=[l,u,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Dn(n,c,e,i,r,a,!1,s)}function Ia(n,t,e,i,r,a,s){s===void 0&&(s="NDHWC");var o=Lc(t),l=o[0],u=o[1],c=o[2],h,d;if(s==="NDHWC")d="channelsLast",h=[l,u,c,n[4],n[4]];else if(s==="NCDHW")d="channelsFirst",h=[l,u,c,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return La(n,h,e,i,r,!1,d,a)}function Dn(n,t,e,i,r,a,s,o){s===void 0&&(s=!1),o===void 0&&(o="channelsLast");var l=[-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3];if(o==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3];else if(o==="channelsFirst")u=n[0],d=n[1],c=n[2],h=n[3];else throw new Error("Unknown dataFormat "+o);var p=t[0],f=t[1],m=t[3],g=Cs(e),v=g[0],b=g[1],w=Cs(i),S=w[0],L=w[1],x=Or(p,S),C=Or(f,L),R=CT(r,c,h,v,b,x,C,a,o),D=R.padInfo,k=R.outHeight,W=R.outWidth,F=s?m*d:m,P;return o==="channelsFirst"?P=[u,F,k,W]:o==="channelsLast"&&(P=[u,k,W,F]),{batchSize:u,dataFormat:o,inHeight:c,inWidth:h,inChannels:d,outHeight:k,outWidth:W,outChannels:F,padInfo:D,strideHeight:v,strideWidth:b,filterHeight:p,filterWidth:f,effectiveFilterHeight:x,effectiveFilterWidth:C,dilationHeight:S,dilationWidth:L,inShape:n,outShape:P,filterShape:t}}function La(n,t,e,i,r,a,s,o){a===void 0&&(a=!1),s===void 0&&(s="channelsLast");var l=[-1,-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3],p=l[4];if(s==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3],p=n[4];else if(s==="channelsFirst")u=n[0],p=n[1],c=n[2],h=n[3],d=n[4];else throw new Error("Unknown dataFormat "+s);var f=t[0],m=t[1],g=t[2],v=t[4],b=Lc(e),w=b[0],S=b[1],L=b[2],x=Lc(i),C=x[0],R=x[1],D=x[2],k=Or(f,C),W=Or(m,R),F=Or(g,D),P=RT(r,c,h,d,w,S,L,k,W,F,o),H=P.padInfo,_=P.outDepth,K=P.outHeight,j=P.outWidth,q=a?v*p:v,G;return s==="channelsFirst"?G=[u,q,_,K,j]:s==="channelsLast"&&(G=[u,_,K,j,q]),{batchSize:u,dataFormat:s,inDepth:c,inHeight:h,inWidth:d,inChannels:p,outDepth:_,outHeight:K,outWidth:j,outChannels:q,padInfo:H,strideDepth:w,strideHeight:S,strideWidth:L,filterDepth:f,filterHeight:m,filterWidth:g,effectiveFilterDepth:k,effectiveFilterHeight:W,effectiveFilterWidth:F,dilationDepth:C,dilationHeight:R,dilationWidth:D,inShape:n,outShape:G,filterShape:t}}function OT(n,t,e,i,r){i==null&&(i=Ic(n,t,e));var a=n[0],s=n[1],o=zi((a-t+2*i)/e+1,r);E(rt(o),function(){return"The output # of rows ("+o+") must be an integer. Change the stride and/or zero pad parameters"});var l=zi((s-t+2*i)/e+1,r);return E(rt(l),function(){return"The output # of columns ("+l+") must be an integer. Change the stride and/or zero pad parameters"}),[o,l]}function ET(n,t,e,i,r,a){r==null&&(r=Ic(n,t,i));var s=n[0],o=n[1],l=n[2],u=zi((s-t+2*r)/i+1,a);E(rt(u),function(){return"The output # of depths ("+u+") must be an integer. Change the stride and/or zero pad parameters"});var c=zi((o-t+2*r)/i+1,a);E(rt(c),function(){return"The output # of rows ("+c+") must be an integer. Change the stride and/or zero pad parameters"});var h=zi((l-t+2*r)/i+1,a);return E(rt(h),function(){return"The output # of columns ("+h+") must be an integer. Change the stride and/or zero pad parameters"}),[u,c,h,e]}function Ic(n,t,e,i){i===void 0&&(i=1);var r=Or(t,i);return Math.floor((n[0]*(e-1)-e+r)/2)}function Cs(n){return typeof n=="number"?[n,n,n]:n.length===2?[n[0],n[1],1]:n}function Lc(n){return typeof n=="number"?[n,n,n]:n}function Or(n,t){return t<=1?n:n+(n-1)*(t-1)}function CT(n,t,e,i,r,a,s,o,l){var u,c,h;if(typeof n=="number"){var d=n===0?"VALID":"NUMBER";u={top:n,bottom:n,left:n,right:n,type:d};var p=OT([t,e],a,i,n,o);c=p[0],h=p[1]}else if(n==="same"){c=Math.ceil(t/i),h=Math.ceil(e/r);var f=Math.max(0,(c-1)*i+a-t),m=Math.max(0,(h-1)*r+s-e),g=Math.floor(f/2),v=f-g,b=Math.floor(m/2),w=m-b;u={top:g,bottom:v,left:b,right:w,type:"SAME"}}else if(n==="valid")u={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-a+1)/i),h=Math.ceil((e-s+1)/r);else if(typeof n=="object"){var S=l==="channelsLast"?n[1][0]:n[2][0],v=l==="channelsLast"?n[1][1]:n[2][1],b=l==="channelsLast"?n[2][0]:n[3][0],w=l==="channelsLast"?n[2][1]:n[3][1],d=S===0&&v===0&&b===0&&w===0?"VALID":"EXPLICIT";u={top:S,bottom:v,left:b,right:w,type:d},c=zi((t-a+S+v)/i+1,o),h=zi((e-s+b+w)/r+1,o)}else throw Error("Unknown padding parameter: "+n);return{padInfo:u,outHeight:c,outWidth:h}}function RT(n,t,e,i,r,a,s,o,l,u,c){var h,d,p,f;if(typeof n=="number"){var m=n===0?"VALID":"NUMBER";h={top:n,bottom:n,left:n,right:n,front:n,back:n,type:m};var g=ET([t,e,i,1],o,1,r,n,c);d=g[0],p=g[1],f=g[2]}else if(n==="same"){d=Math.ceil(t/r),p=Math.ceil(e/a),f=Math.ceil(i/s);var v=(d-1)*r+o-t,b=(p-1)*a+l-e,w=(f-1)*s+u-i,S=Math.floor(v/2),L=v-S,x=Math.floor(b/2),C=b-x,R=Math.floor(w/2),D=w-R;h={top:x,bottom:C,left:R,right:D,front:S,back:L,type:"SAME"}}else if(n==="valid")h={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),p=Math.ceil((e-l+1)/a),f=Math.ceil((i-u+1)/s);else throw Error("Unknown padding parameter: "+n);return{padInfo:h,outDepth:d,outHeight:p,outWidth:f}}function zi(n,t){if(!t)return n;switch(t){case"round":return Math.round(n);case"ceil":return Math.ceil(n);case"floor":return Math.floor(n);default:throw new Error("Unknown roundingMode "+t)}}function ci(n){var t=Cs(n),e=t[0],i=t[1],r=t[2];return e===1&&i===1&&r===1}function Pt(n,t){return ci(n)||ci(t)}function Sa(n){if(n==="NHWC")return"channelsLast";if(n==="NCHW")return"channelsFirst";throw new Error("Unknown dataFormat "+n)}function DT(n,t,e,i,r){var a=O(n,"x","avgPool","float32"),s=1;E(Pt(e,s),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")});var o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in avgPool: x must be rank 4 but got rank "+o.rank+"."}),r!=null&&E(rt(i),function(){return"Error in avgPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r);return f([o]),m.filterWidth===1&&m.filterHeight===1&&ln(m.inShape,m.outShape)?o.clone():p.avgPool(o,m)},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,pl,h);return d=ue(d,a.dtype),l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Ac=U({avgPool_:DT});function kT(n,t,e,i,r,a,s){a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","avgPool3d","float32"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in avgPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in avgPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in avgPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in avgPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a);return m([l]),f.avgPool3d(l,g)},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,fl,d);return p=ue(p,l.dtype),u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var FT=U({avgPool3d_:kT});function Xm(n,t){var e=n[0].length;n.forEach(function(r,a){E(r.length===e,function(){return"Error in concat"+e+"D: rank of tensors["+a+"] must be the same "+("as the rank of the rest ("+e+")")})}),E(t>=0&&t<e,function(){return"Error in concat"+e+"D: axis must be between 0 and "+(e-1)+"."});var i=n[0];n.forEach(function(r,a){for(var s=0;s<e;s++)E(s===t||r[s]===i[s],function(){return"Error in concat"+e+"D: Shape of tensors["+a+"] ("+r+") "+("does not match the shape of the rest ("+i+") ")+("along the non-concatenated axis "+a+".")})})}function Jm(n,t){for(var e=n[0].slice(),i=1;i<n.length;i++)e[t]+=n[i][t];return e}function WT(n,t){t===void 0&&(t=0),E(n.length>=1,function(){return"Pass at least one tensor to concat"});var e=ba(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
-          with dtype `+s.dtype+". ")});var i=function(s,o){var l=Qe(t,e[0].shape)[0],u=Jm(e.map(function(d){return d.shape}),l);if(ot(u)===0)return si([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Xm(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,wl,a)}var xt=U({concat_:WT});function UT(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,Lu)}var Pi=U({sigmoid_:UT});function BT(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=gc(i,t,e),c=u[0],h=u[1];return Sm(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,yu,s)}var ze=U({slice_:BT});function zT(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,Du)}var Rs=U({tanh_:zT});function PT(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=xt([u,h],1),p=We(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),x=fe(J(Pi(b),Rs(w)),J(c,Pi(fe(s,S)))),C=J(Rs(x),Pi(L));return[x,C]}var _T=U({basicLSTMCell_:PT});function MT(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length  "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,gl,o)}var Os=U({batchToSpaceND_:MT});function HT(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function VT(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=HT(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Es(o),Es(l),Es(c),Es(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,Ul,f);return V(m,s.shape)}function Es(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Aa=U({batchNorm_:VT});function qT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var GT=U({batchNorm2d_:qT});function YT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var KT=U({batchNorm3d_:YT});function jT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var $T=U({batchNorm4d_:jT});function XT(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.length<e.rank)throw new Error("broadcastTo(): shape.length="+t.length+" < input.rank="+e.rank+".");if(t.length>e.rank){for(var r=e.shape.slice();r.length<t.length;)r.unshift(1);e=V(e,r)}for(var a=e.shape,s=Array.from(t),o=t.length-1;o>=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return Bi(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,vl,h)}var Ds=U({broadcastTo_:XT});function JT(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,yl)}var Zm=U({ceil_:JT});function ZT(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,bl,a)}var Qm=U({clipByValue_:ZT});function QT(n){return xt(n,0)}var e1=U({concat1d_:QT});function t1(n,t){return xt(n,t)}var n1=U({concat2d_:t1});function i1(n,t){return xt(n,t)}var r1=U({concat3d_:i1});function a1(n,t){return xt(n,t)}var s1=U({concat4d_:a1});function o1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(Pt(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=Sa(r),w=Dn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,Sl,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Er=U({conv2d_:o1});function l1(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(Pt(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Er(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var eg=U({conv1d_:l1});function u1(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=Sa(a),S=Dn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,Ll,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Tc=U({conv2DBackpropInput_:u1});function c1(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return Tc(e,s,o,i,r,"NHWC",a)}var tg=U({conv2dTranspose_:c1});function h1(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(Pt(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=La(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,Il,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var d1=U({conv3d_:h1});function p1(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=La(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,Hp,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var ng=U({conv3DBackpropInput_:p1});function f1(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return ng(e,a,s,i,r)}var m1=U({conv3dTranspose_:f1});function g1(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,Al)}var ks=U({cos_:g1});function v1(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,Tl)}var Nc=U({cosh_:v1});function y1(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=tn([t],r.rank),h=r;c!=null&&(h=ut(r,c));var d=En(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=xs(c);p=ut(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,Nl,o)}var xc=U({cumsum_:y1});function b1(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
+`)+("Expected: "+u+"."))}}function zA(n,t){n().then(function(){return t.fail()},function(){return t()})}function PA(n,t){var e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return ni(n)||ni(n[0])||ni(t)||ni(t[0])?bc(n,e,function(i,r){return i==r}):bc(n,t,function(i,r){return wc(i,r,0)})}function _A(n,t,e){if(e==null&&(e=yc()),!wc(n,t,e))throw new Error("Numbers differ: actual === "+n+", expected === "+t)}function wc(n,t,e){return!isFinite(n)&&!isFinite(t)?!0:!(isNaN(n)||isNaN(t)||Math.abs(n-t)>e)}function MA(n,t,e){for(var i=0;i<n.length;i++)if(n[i]<t||n[i]>e)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function HA(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var VA={__proto__:null,TEST_EPSILON_FLOAT16:Wm,expectArraysClose:BA,testEpsilon:yc,expectPromiseToFail:zA,expectArraysEqual:PA,expectNumbersClose:_A,expectValuesInRange:MA,expectArrayBuffersEqual:HA};var qA="2.6.0";function GA(){He().set("PROD",!0)}function YA(){He().set("DEBUG",!0)}function KA(){He().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function It(n){He().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function jA(){z.disposeVariables()}function $A(){return z}function XA(){return z.memory()}function JA(n){return z.profile(n)}function ft(n,t){return z.tidy(n,t)}function zt(n){var t=ic(n);t.forEach(function(e){return e.dispose()})}function Um(n){return z.keep(n)}function ZA(n){return z.time(n)}function QA(n){return z.setBackend(n)}function eT(){return z.ready()}function tT(){return z.backendName}function nT(n){z.removeBackend(n)}function iT(n){return z.findBackend(n)}function rT(n){return z.findBackendFactory(n)}function aT(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function sT(){return z.backend}function oT(n,t){He().setPlatform(n,t)}function lT(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,ps)}var fe=U({add_:lT});function uT(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ul)}var Sc=U({floorDiv_:uT});function cT(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=at(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return Sc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Ol,o)}var Ie=U({div_:cT});function hT(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,tu)}var J=U({mul_:hT});function dT(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,il)}var qt=U({abs_:dT});function pT(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,rl)}var Bm=U({acos_:pT});function fT(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,al)}var zm=U({acosh_:fT});function mT(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!on(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,sl)}var gT=U({addN_:mT});function Lc(n,t){for(var e=0;e<n.length;++e)if(n[n.length-e-1]!==t-1-e)return!1;return!0}function Pm(n,t,e){for(var i=n.length+t.length,r=[],a=0,s=0,o=0;o<i;o++)e.indexOf(o)===-1?r.push(n[a++]):r.push(t[s++]);return r}function _m(n,t){for(var e=[],i=n.length,r=0;r<i;r++)t.indexOf(r)===-1&&e.push(n[r]);var a=t.map(function(s){return n[s]});return[e,a]}function Qt(n,t){var e=t.map(function(i){return 1});return Pm(n,e,t)}function vT(n,t,e){E(Lc(t,e),function(){return n+" supports only inner-most axes for now. "+("Got axes "+t+" and rank-"+e+" input.")})}function en(n,t){if(Lc(n,t))return null;for(var e=[],i=0;i<t;++i)n.indexOf(i)===-1&&e.push(i);return n.forEach(function(r){return e.push(r)}),e}function xs(n){return n.map(function(t,e){return[e,t]}).sort(function(t,e){return t[1]-e[1]}).map(function(t){return t[0]})}function En(n,t){for(var e=[],i=t-n;i<t;++i)e.push(i);return e}function yT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","all","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=en(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.all(i,u);if(e){var d=Qt(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Wp,s)}var Mm=U({all_:yT});function bT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","any","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=en(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.any(i,u);if(e){var d=Qt(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Up,s)}var Hm=U({any_:bT});function wT(n,t){t===void 0&&(t=0);var e=O(n,"x","argMax"),i=function(s,o){o([e]);var l=Qe(t,e.shape),u=en(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMax(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,ol,a)}var Vm=U({argMax_:wT});function ST(n,t){t===void 0&&(t=0);var e=O(n,"x","argMin"),i=function(s,o){o([e]),t==null&&(t=0);var l=Qe(t,e.shape),u=en(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMin(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,ll,a)}var qm=U({argMin_:ST});function LT(n){var t=O(n,"x","asin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asin(t);return r([t]),a},e,null,ul)}var Gm=U({asin_:LT});function IT(n){var t=O(n,"x","asinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asinh(t);return r([t]),a},e,null,cl)}var Ym=U({asinh_:IT});function AT(n){var t=O(n,"x","atan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atan(t);return r([t]),a},e,null,hl)}var Km=U({atan_:AT});function TT(n,t){var e,i=O(n,"a","atan2"),r=O(t,"b","atan2");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.atan2(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,pl)}var jm=U({atan2_:TT});function NT(n){var t=O(n,"x","atanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atanh(t);return r([t]),a},e,null,dl)}var $m=U({atanh_:NT});function xT(n,t,e,i,r,a){r===void 0&&(r="NHWC");var s=n[3],o=t.concat([s]),l=Sa(r);return Dn(n,o,e,a,i,null,null,l)}function Rr(n,t,e,i,r,a,s){s===void 0&&(s="channelsLast");var o=Cs(t),l=o[0],u=o[1],c;if(s==="channelsLast")c=[l,u,n[3],n[3]];else if(s==="channelsFirst")c=[l,u,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Dn(n,c,e,i,r,a,!1,s)}function Ia(n,t,e,i,r,a,s){s===void 0&&(s="NDHWC");var o=Ic(t),l=o[0],u=o[1],c=o[2],h,d;if(s==="NDHWC")d="channelsLast",h=[l,u,c,n[4],n[4]];else if(s==="NCDHW")d="channelsFirst",h=[l,u,c,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return La(n,h,e,i,r,!1,d,a)}function Dn(n,t,e,i,r,a,s,o){s===void 0&&(s=!1),o===void 0&&(o="channelsLast");var l=[-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3];if(o==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3];else if(o==="channelsFirst")u=n[0],d=n[1],c=n[2],h=n[3];else throw new Error("Unknown dataFormat "+o);var p=t[0],f=t[1],m=t[3],g=Cs(e),v=g[0],b=g[1],w=Cs(i),S=w[0],L=w[1],x=Or(p,S),C=Or(f,L),R=CT(r,c,h,v,b,x,C,a,o),D=R.padInfo,k=R.outHeight,W=R.outWidth,F=s?m*d:m,P;return o==="channelsFirst"?P=[u,F,k,W]:o==="channelsLast"&&(P=[u,k,W,F]),{batchSize:u,dataFormat:o,inHeight:c,inWidth:h,inChannels:d,outHeight:k,outWidth:W,outChannels:F,padInfo:D,strideHeight:v,strideWidth:b,filterHeight:p,filterWidth:f,effectiveFilterHeight:x,effectiveFilterWidth:C,dilationHeight:S,dilationWidth:L,inShape:n,outShape:P,filterShape:t}}function La(n,t,e,i,r,a,s,o){a===void 0&&(a=!1),s===void 0&&(s="channelsLast");var l=[-1,-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3],p=l[4];if(s==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3],p=n[4];else if(s==="channelsFirst")u=n[0],p=n[1],c=n[2],h=n[3],d=n[4];else throw new Error("Unknown dataFormat "+s);var f=t[0],m=t[1],g=t[2],v=t[4],b=Ic(e),w=b[0],S=b[1],L=b[2],x=Ic(i),C=x[0],R=x[1],D=x[2],k=Or(f,C),W=Or(m,R),F=Or(g,D),P=RT(r,c,h,d,w,S,L,k,W,F,o),H=P.padInfo,_=P.outDepth,K=P.outHeight,j=P.outWidth,q=a?v*p:v,G;return s==="channelsFirst"?G=[u,q,_,K,j]:s==="channelsLast"&&(G=[u,_,K,j,q]),{batchSize:u,dataFormat:s,inDepth:c,inHeight:h,inWidth:d,inChannels:p,outDepth:_,outHeight:K,outWidth:j,outChannels:q,padInfo:H,strideDepth:w,strideHeight:S,strideWidth:L,filterDepth:f,filterHeight:m,filterWidth:g,effectiveFilterDepth:k,effectiveFilterHeight:W,effectiveFilterWidth:F,dilationDepth:C,dilationHeight:R,dilationWidth:D,inShape:n,outShape:G,filterShape:t}}function OT(n,t,e,i,r){i==null&&(i=Ac(n,t,e));var a=n[0],s=n[1],o=zi((a-t+2*i)/e+1,r);E(rt(o),function(){return"The output # of rows ("+o+") must be an integer. Change the stride and/or zero pad parameters"});var l=zi((s-t+2*i)/e+1,r);return E(rt(l),function(){return"The output # of columns ("+l+") must be an integer. Change the stride and/or zero pad parameters"}),[o,l]}function ET(n,t,e,i,r,a){r==null&&(r=Ac(n,t,i));var s=n[0],o=n[1],l=n[2],u=zi((s-t+2*r)/i+1,a);E(rt(u),function(){return"The output # of depths ("+u+") must be an integer. Change the stride and/or zero pad parameters"});var c=zi((o-t+2*r)/i+1,a);E(rt(c),function(){return"The output # of rows ("+c+") must be an integer. Change the stride and/or zero pad parameters"});var h=zi((l-t+2*r)/i+1,a);return E(rt(h),function(){return"The output # of columns ("+h+") must be an integer. Change the stride and/or zero pad parameters"}),[u,c,h,e]}function Ac(n,t,e,i){i===void 0&&(i=1);var r=Or(t,i);return Math.floor((n[0]*(e-1)-e+r)/2)}function Cs(n){return typeof n=="number"?[n,n,n]:n.length===2?[n[0],n[1],1]:n}function Ic(n){return typeof n=="number"?[n,n,n]:n}function Or(n,t){return t<=1?n:n+(n-1)*(t-1)}function CT(n,t,e,i,r,a,s,o,l){var u,c,h;if(typeof n=="number"){var d=n===0?"VALID":"NUMBER";u={top:n,bottom:n,left:n,right:n,type:d};var p=OT([t,e],a,i,n,o);c=p[0],h=p[1]}else if(n==="same"){c=Math.ceil(t/i),h=Math.ceil(e/r);var f=Math.max(0,(c-1)*i+a-t),m=Math.max(0,(h-1)*r+s-e),g=Math.floor(f/2),v=f-g,b=Math.floor(m/2),w=m-b;u={top:g,bottom:v,left:b,right:w,type:"SAME"}}else if(n==="valid")u={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-a+1)/i),h=Math.ceil((e-s+1)/r);else if(typeof n=="object"){var S=l==="channelsLast"?n[1][0]:n[2][0],v=l==="channelsLast"?n[1][1]:n[2][1],b=l==="channelsLast"?n[2][0]:n[3][0],w=l==="channelsLast"?n[2][1]:n[3][1],d=S===0&&v===0&&b===0&&w===0?"VALID":"EXPLICIT";u={top:S,bottom:v,left:b,right:w,type:d},c=zi((t-a+S+v)/i+1,o),h=zi((e-s+b+w)/r+1,o)}else throw Error("Unknown padding parameter: "+n);return{padInfo:u,outHeight:c,outWidth:h}}function RT(n,t,e,i,r,a,s,o,l,u,c){var h,d,p,f;if(typeof n=="number"){var m=n===0?"VALID":"NUMBER";h={top:n,bottom:n,left:n,right:n,front:n,back:n,type:m};var g=ET([t,e,i,1],o,1,r,n,c);d=g[0],p=g[1],f=g[2]}else if(n==="same"){d=Math.ceil(t/r),p=Math.ceil(e/a),f=Math.ceil(i/s);var v=(d-1)*r+o-t,b=(p-1)*a+l-e,w=(f-1)*s+u-i,S=Math.floor(v/2),L=v-S,x=Math.floor(b/2),C=b-x,R=Math.floor(w/2),D=w-R;h={top:x,bottom:C,left:R,right:D,front:S,back:L,type:"SAME"}}else if(n==="valid")h={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),p=Math.ceil((e-l+1)/a),f=Math.ceil((i-u+1)/s);else throw Error("Unknown padding parameter: "+n);return{padInfo:h,outDepth:d,outHeight:p,outWidth:f}}function zi(n,t){if(!t)return n;switch(t){case"round":return Math.round(n);case"ceil":return Math.ceil(n);case"floor":return Math.floor(n);default:throw new Error("Unknown roundingMode "+t)}}function ci(n){var t=Cs(n),e=t[0],i=t[1],r=t[2];return e===1&&i===1&&r===1}function Pt(n,t){return ci(n)||ci(t)}function Sa(n){if(n==="NHWC")return"channelsLast";if(n==="NCHW")return"channelsFirst";throw new Error("Unknown dataFormat "+n)}function DT(n,t,e,i,r){var a=O(n,"x","avgPool","float32"),s=1;E(Pt(e,s),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")});var o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in avgPool: x must be rank 4 but got rank "+o.rank+"."}),r!=null&&E(rt(i),function(){return"Error in avgPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r);return f([o]),m.filterWidth===1&&m.filterHeight===1&&on(m.inShape,m.outShape)?o.clone():p.avgPool(o,m)},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,fl,h);return d=ue(d,a.dtype),l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Tc=U({avgPool_:DT});function kT(n,t,e,i,r,a,s){a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","avgPool3d","float32"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in avgPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in avgPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in avgPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in avgPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a);return m([l]),f.avgPool3d(l,g)},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,ml,d);return p=ue(p,l.dtype),u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var FT=U({avgPool3d_:kT});function Xm(n,t){var e=n[0].length;n.forEach(function(r,a){E(r.length===e,function(){return"Error in concat"+e+"D: rank of tensors["+a+"] must be the same "+("as the rank of the rest ("+e+")")})}),E(t>=0&&t<e,function(){return"Error in concat"+e+"D: axis must be between 0 and "+(e-1)+"."});var i=n[0];n.forEach(function(r,a){for(var s=0;s<e;s++)E(s===t||r[s]===i[s],function(){return"Error in concat"+e+"D: Shape of tensors["+a+"] ("+r+") "+("does not match the shape of the rest ("+i+") ")+("along the non-concatenated axis "+a+".")})})}function Jm(n,t){for(var e=n[0].slice(),i=1;i<n.length;i++)e[t]+=n[i][t];return e}function WT(n,t){t===void 0&&(t=0),E(n.length>=1,function(){return"Pass at least one tensor to concat"});var e=ba(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
+          with dtype `+s.dtype+". ")});var i=function(s,o){var l=Qe(t,e[0].shape)[0],u=Jm(e.map(function(d){return d.shape}),l);if(ot(u)===0)return si([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Xm(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,Sl,a)}var xt=U({concat_:WT});function UT(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,Iu)}var Pi=U({sigmoid_:UT});function BT(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=vc(i,t,e),c=u[0],h=u[1];return Sm(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,bu,s)}var ze=U({slice_:BT});function zT(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,ku)}var Rs=U({tanh_:zT});function PT(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=xt([u,h],1),p=We(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),x=fe(J(Pi(b),Rs(w)),J(c,Pi(fe(s,S)))),C=J(Rs(x),Pi(L));return[x,C]}var _T=U({basicLSTMCell_:PT});function MT(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length  "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,vl,o)}var Os=U({batchToSpaceND_:MT});function HT(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function VT(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=HT(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Es(o),Es(l),Es(c),Es(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,Bl,f);return V(m,s.shape)}function Es(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Aa=U({batchNorm_:VT});function qT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var GT=U({batchNorm2d_:qT});function YT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var KT=U({batchNorm3d_:YT});function jT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var $T=U({batchNorm4d_:jT});function XT(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.length<e.rank)throw new Error("broadcastTo(): shape.length="+t.length+" < input.rank="+e.rank+".");if(t.length>e.rank){for(var r=e.shape.slice();r.length<t.length;)r.unshift(1);e=V(e,r)}for(var a=e.shape,s=Array.from(t),o=t.length-1;o>=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return Bi(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,yl,h)}var Ds=U({broadcastTo_:XT});function JT(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,bl)}var Zm=U({ceil_:JT});function ZT(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,wl,a)}var Qm=U({clipByValue_:ZT});function QT(n){return xt(n,0)}var e1=U({concat1d_:QT});function t1(n,t){return xt(n,t)}var n1=U({concat2d_:t1});function i1(n,t){return xt(n,t)}var r1=U({concat3d_:i1});function a1(n,t){return xt(n,t)}var s1=U({concat4d_:a1});function o1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(Pt(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=Sa(r),w=Dn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,Ll,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Er=U({conv2d_:o1});function l1(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(Pt(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Er(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var eg=U({conv1d_:l1});function u1(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=Sa(a),S=Dn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,Il,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Nc=U({conv2DBackpropInput_:u1});function c1(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return Nc(e,s,o,i,r,"NHWC",a)}var tg=U({conv2dTranspose_:c1});function h1(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(Pt(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=La(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,Al,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var d1=U({conv3d_:h1});function p1(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=La(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,Hp,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var ng=U({conv3DBackpropInput_:p1});function f1(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return ng(e,a,s,i,r)}var m1=U({conv3dTranspose_:f1});function g1(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,Tl)}var ks=U({cos_:g1});function v1(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,Nl)}var xc=U({cosh_:v1});function y1(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=en([t],r.rank),h=r;c!=null&&(h=ut(r,c));var d=En(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=xs(c);p=ut(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,xl,o)}var Cc=U({cumsum_:y1});function b1(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
     `+r+" and "+t+`  for depthToSpace with input shape
     `+i.shape}),E(a*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
     `+a+" and "+t+` for depthToSpace with input shape
-        `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,qp,u)}var ig=U({depthToSpace_:b1});function w1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(rt(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(Pt(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Dn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,xl,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ta=U({depthwiseConv2d_:w1});function S1(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Kp)}var L1=U({diag_:S1});function I1(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(Cl,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var rg=U({dilation2d_:I1});function A1(n,t){for(var e=n.length,i=[],r=0;r<e;r++){var a=e-1-r,s=n[a]||1,o=t[t.length-1-r]||1;o>1&&s===1&&i.unshift(a)}return i}function mt(n,t){for(var e=[],i=0;i<t.length;i++){var r=n[n.length-i-1],a=t.length-i-1,s=t[a];(r==null||r===1&&s>1)&&e.unshift(a)}return e}function et(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;r<i;r++){var a=n[n.length-r-1];a==null&&(a=1);var s=t[t.length-r-1];if(s==null&&(s=1),a===1)e.unshift(s);else if(s===1)e.unshift(a);else if(a!==s){var o="Operands could not be broadcast together with shapes "+(n+" and "+t+".");throw Error(o)}else e.unshift(a)}return e}function T1(n,t){var e,i=O(n,"a","equal"),r=O(t,"b","equal");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.equal(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Jp)}var Na=U({equal_:T1});function N1(n,t,e){var i=O(t,"a","where"),r=O(e,"b","where"),a=O(n,"condition","where","bool"),s=et(i.shape,r.shape),o=Ds(i,s),l=Ds(r,s);a.rank===1&&E(a.shape[0]===i.shape[0],function(){return"The first dimension of `a` must match the size of `condition`."}),a.rank!==1&&Be(a.shape,l.shape,"Error in where: ");var u=function(h,d){var p=h.select(a,o,l);return d([a]),p},c={condition:a,t:o,e:l};return z.runKernelFunc(u,c,null,gu)}var pn=U({where_:N1});function x1(n){var t=O(n,"x","zerosLike"),e={x:t};return z.runKernelFunc(function(i){return i.zerosLike(t)},e,null,Bu)}var Ee=U({zerosLike_:x1});function C1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");e=at(i,r),i=e[0],r=e[1];var a=Ie(i,r),s=Ee(a),o=Na(r,s);return pn(o,s,a)}var ag=U({divNoNan_:C1});function R1(n,t){var e=O(n,"t1","dot"),i=O(t,"t2","dot");E((e.rank===1||e.rank===2)&&(i.rank===1||i.rank===2),function(){return"Error in dot: inputs must all be rank 1 or 2, but got ranks "+(e.rank+" and "+i.rank+".")});var r=e.rank===1?e.size:e.shape[1],a=i.rank===1?i.size:i.shape[0];if(E(r===a,function(){return"Error in dot: inner dimensions of inputs must match, but got "+(r+" and "+a+".")}),e.rank===1&&i.rank===1){var s=V(e,[1,-1]),o=V(i,[-1,1]),l=We(s,o);return V(l,[])}else if(e.rank===1&&i.rank===2){var s=V(e,[1,-1]),o=V(i,[i.shape[0],i.shape[1]]),l=We(s,o);return V(l,[l.size])}else if(e.rank===2&&i.rank===1){var o=V(i,[-1,1]),l=We(e,o);return V(l,[l.size])}else{var o=V(i,[i.shape[0],i.shape[1]]),l=We(e,o);return l}}var sg=U({dot_:R1});function O1(n){var t=O(n,"x","elu"),e=function(r,a){var s=r.elu(t);return a([s]),s},i={x:t};return z.runKernelFunc(e,i,null,Ol)}var Cc=U({elu_:O1});function E1(n){var t=O(n,"x","erf");E(t.dtype==="int32"||t.dtype==="float32",function(){return"Input dtype must be `int32` or `float32`."}),t.dtype==="int32"&&(t=ue(t,"float32"));var e={x:t};return z.runKernelFunc(function(i,r){var a=i.erf(t);return r([t]),a},e,null,El)}var og=U({erf_:E1});function D1(n){var t=O(n,"x","exp"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.exp(t);return r([a]),a},e,null,Dl)}var fn=U({exp_:D1});function k1(n,t){t===void 0&&(t=0);var e=null,i=O(n,"x","expandDims",e);E(t<=i.rank,function(){return"Axis must be <= rank of the tensor"});var r=i.shape.slice();return t<0&&(E(-(i.rank+1)<=t,function(){return"Axis must be in the interval ["+-(i.rank+1)+", "+i.rank+"]"}),t=i.rank+t+1),r.splice(t,0,1),V(i,r)}var mn=U({expandDims_:k1});function F1(n){var t=O(n,"x","expm1"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.expm1(t);return r([t]),a},e,null,kl)}var lg=U({expm1_:F1});function W1(n,t){var e=null,i=O(n,"x","tile",e);E(i.rank===t.length,function(){return"Error in transpose: rank of input "+i.rank+" "+("must match length of reps "+t+".")});var r=function(l,u){var c=l.tile(i,t);return u([i]),c},a=[i],s={x:i},o={reps:t};return z.runKernelFunc(r,s,null,ku,o,a)}var Dr=U({tile_:W1});function U1(n,t,e,i){i===void 0&&(i="float32"),t==null&&(t=n);for(var r=On([n,t],i),a=n<=t?n:t,s=0;s<a;++s)r.set(1,s,s);var o=V(r.toTensor(),[n,t]);if(e==null)return o;if(e.length===1)return Dr(mn(o,0),[e[0],1,1]);if(e.length===2)return Dr(mn(mn(o,0),0),[e[0],e[1],1,1]);if(e.length===3)return Dr(mn(mn(mn(o,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error("eye() currently supports only 1D and 2D "+("batchShapes, but received "+e.length+"D."))}var ug=U({eye_:U1});function Rc(n,t,e){var i={shape:n,value:t,dtype:e};return z.runKernelFunc(function(r){return r.fill(n,t,e)},{},null,Qp,i)}function B1(n){var t=O(n,"x","floor"),e={x:t};return z.runKernelFunc(function(i){return i.floor(t)},e,null,Fl)}var Fs=U({floor_:B1});var Oc=30;function z1(n){return n<=Oc?n:ys(n,Math.floor(Math.sqrt(n)))}function P1(n,t){var e=!1,i;for(n<=Oc?(i=n,e=!0):i=ys(n,Math.floor(Math.sqrt(n)));!e;)i>t||i===n?e=!0:i=ys(n,i+1);return i}function _1(n,t,e){for(var i=[],r=n.length,a=0;a<r;a++)a!==t?i.push(n[a]):i.push(e);return i}function cg(n,t,e){for(var i=n.shape[e],r=[],a=1,s=1,o=0;o<e;o++)r.push(n.shape[o]),a*=n.shape[o];for(var o=0;o<t.rank;o++)r.push(t.shape[o]);for(var o=e+1;o<n.rank;o++)r.push(n.shape[o]),s*=n.shape[o];return{batchSize:a,sliceSize:s,dimSize:i,outputShape:r}}var M1={__proto__:null,segOpComputeOptimalWindowSize:P1,computeOutShape:_1,collectGatherOpShapeInfo:cg};function H1(n,t,e){e===void 0&&(e=0);var i=O(n,"x","gather"),r=O(t,"indices","gather","int32"),a={x:i,indices:r},s={axis:e},o=function(l,u){var c=Qe(e,i.shape)[0],h=cg(i,r,c),d=l.gather(i,V(r,[r.size]),c);return u([i,r]),V(d,h.outputShape)};return z.runKernelFunc(o,a,null,Bl,s)}var Ws=U({gather_:H1});function V1(n,t){var e,i=O(n,"a","greater"),r=O(t,"b","greater");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.greater(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,nf)}var hi=U({greater_:V1});function q1(n,t){var e,i=O(n,"a","greaterEqual"),r=O(t,"b","greaterEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.greaterEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,zl)}var _i=U({greaterEqual_:q1});function G1(n){var t=O(n,"input","imag"),e=function(r){return r.imag(t)},i={input:t};return z.runKernelFunc(e,i,null,af)}var Us=U({imag_:G1});function Y1(n){var t=O(n,"x","isFinite"),e={x:t};return z.runKernelFunc(function(i){return i.isFinite(t)},e,null,_l)}var hg=U({isFinite_:Y1});function K1(n){var t=O(n,"x","isInf"),e={x:t};return z.runKernelFunc(function(i){return i.isInf(t)},e,null,Ml)}var dg=U({isInf_:K1});function j1(n){var t=O(n,"x","isNaN"),e={x:t};return z.runKernelFunc(function(i){return i.isNaN(t)},e,null,Hl)}var pg=U({isNaN_:j1});function $1(n,t){var e,i=O(n,"a","maximum"),r=O(t,"b","maximum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.maximum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,jl)}var kr=U({maximum_:$1});function we(n,t){if((Ot(n)&&t!=="string"||Array.isArray(n))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&Ot(n)&&!(n instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");var e=[],i=[];return ai(n,e,i,t)}function X1(n,t){t===void 0&&(t=.2);var e=O(n,"x","leakyRelu");return kr(J(we(t),e),e)}var fg=U({leakyRelu_:X1});function J1(n,t){var e,i=O(n,"a","less"),r=O(t,"b","less");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.less(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,sf)}var Bs=U({less_:J1});function Z1(n,t){var e,i=O(n,"a","lessEqual"),r=O(t,"b","lessEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.lessEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,of)}var Mi=U({lessEqual_:Z1});function Q1(n,t,e){if(e<=0)throw new Error("The number of values should be positive.");var i={start:n,stop:t,num:e};return z.runKernelFunc(function(r){return r.linspace(n,t,e)},{},null,lf,i)}function eN(n,t,e,i,r){t===void 0&&(t=5),e===void 0&&(e=1),i===void 0&&(i=1),r===void 0&&(r=.5);var a=O(n,"x","localResponseNormalization");E(a.rank===4||a.rank===3,function(){return`Error in localResponseNormalization: x must be rank 3 or 4 but got
-               rank `+a.rank+"."}),E(rt(t),function(){return"Error in localResponseNormalization: depthRadius must be an "+("integer but got depthRadius "+t+".")});var s=a,o=!1;a.rank===3&&(o=!0,s=V(a,[1,a.shape[0],a.shape[1],a.shape[2]]));var l=function(d,p){var f=d.localResponseNormalization4D(s,t,e,i,r);return p([s,f]),f},u={x:s},c={depthRadius:t,bias:e,alpha:i,beta:r},h=z.runKernelFunc(l,u,null,Yl,c);return o?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var mg=U({localResponseNormalization_:eN});function tN(n){var t=O(n,"x","log"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log(t);return r([t]),a},e,null,Vl)}var Hi=U({log_:tN});function nN(n){var t=O(n,"x","log1p"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log1p(t);return r([t]),a},e,null,ql)}var Ec=U({log1p_:nN});function iN(n){return E(ii(n),function(){return"The f passed in grad(f) must be a function"}),function(t,e){var i=O(t,"x","tf.grad",null),r=e!=null?O(e,"dy","tf.grad"):null;return z.tidy(function(){var a=z.gradients(function(){return n(i)},[i],r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),zs(o),o[0]})}}function rN(n){return E(ii(n),function(){return"The f passed in grads(f) must be a function"}),function(t,e){E(Array.isArray(t),function(){return"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s"});var i=ba(t,"args","tf.grads",null),r=e!=null?O(e,"dy","tf.grads"):null;return z.tidy(function(){var a=z.gradients(function(){return n.apply(void 0,i)},i,r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(o),o})}}function aN(n){return E(ii(n),function(){return"The f passed in valueAndGrad(f) must be a function"}),function(t,e){E(t instanceof Y,function(){return"The x passed in valueAndGrad(f)(x) must be a tensor"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrad(f)(x, dy) must be a tensor"});var i=z.gradients(function(){return n(t)},[t],e),r=i.grads,a=i.value;return zs(r),{grad:r[0],value:a}}}function sN(n){return E(ii(n),function(){return"The f passed in valueAndGrads(f) must be a function"}),function(t,e){E(Array.isArray(t)&&t.every(function(r){return r instanceof Y}),function(){return"The args passed in valueAndGrads(f)(args) must be array of tensors"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrads(f)(args, dy) must be a tensor"});var i=z.gradients(function(){return n.apply(void 0,t)},t,e);return e!=null&&Be(i.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(i.grads),i}}function gg(n,t){E(ii(n),function(){return"The f passed in variableGrads(f) must be a function"}),E(t==null||Array.isArray(t)&&t.every(function(h){return h instanceof ya}),function(){return"The varList passed in variableGrads(f, varList) must be an array of variables"});var e=t!=null;if(!e){t=[];for(var i in z.registeredVariables)t.push(z.registeredVariables[i])}var r=e?t.filter(function(h){return!h.trainable}):null,a=t.length;t=t.filter(function(h){return h.trainable}),E(t.length>0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function kn(n){return z.customGrad(n)}function zs(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
-    the f you passed encloses all operations that lead from x to y.`)}function oN(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,tu)}var gt=U({neg_:oN});function lN(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,Iu)}var Dc=U({softplus_:lN});function uN(n){var t=O(n,"x","logSigmoid"),e=kn(function(i){var r=gt(Dc(gt(i))),a=function(s){var o=J(s,Pi(gt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var vg=U({logSigmoid_:uN});function cN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=tn(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=en(f.shape,Qe(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,Kl,s)}var Vi=U({max_:cN});function hN(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ou)}var ge=U({sub_:hN});function dN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Qe(t,i.shape),c=tn(u,i.rank),h=u,d=i;c!=null&&(d=ut(i,c),h=En(h.length,i.rank));var p=o.sum(d,h);if(e){var f=en(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Tu,s)}var Ae=U({sum_:dN});function pN(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=Vi(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),Hi(Ae(fn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,Gl,a)}var yg=U({logSoftmax_:pN});function fN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Qe(t,i.shape),a=Vi(i,r,!0),s=ge(i,a),o=fn(s),l=Ae(o,r),u=Hi(l),c=fe(V(a,u.shape),u);if(e){var h=en(c.shape,r);return V(c,h)}return c}var kc=U({logSumExp_:fN});function mN(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,uf)}var qi=U({logicalAnd_:mN});function gN(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,cf)}var Ps=U({logicalNot_:gN});function vN(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,hf)}var Fc=U({logicalOr_:vN});function yN(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return et(e.shape,i.shape),qi(Fc(n,t),Ps(qi(n,t)))}var bg=U({logicalXor_:yN});function bN(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(Pt(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&ln(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,$l,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Wc=U({maxPool_:bN});function wN(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Xl,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var SN=U({maxPool3d_:wN});function LN(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(mf,s,o);return{result:l[0],indexes:l[1]}}var IN=U({maxPoolWithArgmax_:LN});function Kn(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Kn(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=Lr(ot(n),t);return z.makeTensor(r,n,t)}function Fr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Fr(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=Ku(ot(n),t);return z.makeTensor(r,n,t)}function AN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Qe(t,i.shape),a=_m(i.shape,r),s=a[1],o=ot(s),l=kn(function(u){var c=we(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Ie(h,c),p=Ae(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Ie(J(v,Fr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var xa=U({mean_:AN});function TN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=tn(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=en(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Jl,s)}var _s=U({min_:TN});function NN(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Zl)}var Ms=U({minimum_:NN});function xN(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ql)}var Uc=U({mod_:xN});function CN(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ye=U({square_:CN});function RN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Qe(t,n.shape),r=xa(n,i,e),a=r.shape;e||(a=en(r.shape,i));var s=Ye(ge(ue(n,"float32"),V(r,a))),o=xa(s,i,e);return{mean:r,variance:o}}var ON=U({moments_:RN});function EN(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=ba(e,"c","multiRNNCell"),s=ba(i,"h","multiRNNCell"),o=r,l=[],u=0;u<n.length;u++){var c=n[u](o,a[u],s[u]);l.push(c[0]),l.push(c[1]),o=c[1]}for(var h=[],d=[],u=0;u<l.length;u+=2)h.push(l[u]),d.push(l[u+1]);return[h,d]}var DN=U({multiRNNCell_:EN});function kN(n,t,e,i){i===void 0&&(i=!1);var r=O(n,"logits","multinomial"),a=r.size,s=r.rank;if(a<2)throw new Error("Error in multinomial: you need at least 2 outcomes, but got "+(a+"."));if(s>2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var FN=U({multinomial_:kN});function WN(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,gf)}var Hs=U({notEqual_:WN});function UN(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,Lf)}var Ca=U({real_:UN});function BN(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=Bc(Ca(t)),o=Ee(Us(t));return ri(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,nu)}var Bc=U({onesLike_:BN});function zN(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return We(r,a)}var PN=U({outerProduct_:zN});function _N(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,ru,a)}var Gi=U({pad_:_N});function MN(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),Gi(n,[t],e)}var HN=U({pad1d_:MN});function VN(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var qN=U({pad2d_:VN});function GN(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var YN=U({pad3d_:GN});function KN(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var jN=U({pad4d_:KN});function $N(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,Nu,s)}var Vs=U({spaceToBatchND_:$N});function ZN(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(Pt(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=Rr(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=JN([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=XN([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:Vs(o,c,f),b=e==="avg"?function(){return Ac(v,t,a,g)}:function(){return Wc(v,t,a,g)},w=b(),S=d?w:Os(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function XN(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function JN(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var wg=U({pool_:ZN});function QN(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=at(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,au)}var jn=U({pow_:QN});function ex(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,su)}var zc=U({prelu_:ex});function tx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Qe(t,i.shape),u=tn(l,i.rank),c=l,h=i;u!=null&&(h=ut(i,u),c=En(c.length,i.rank));var d=o.prod(h,c);if(e){var p=en(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,wf,s)}var Sg=U({prod_:tx});function nx(n,t,e){var i=ot(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a<i;a++)r[a]=t();return z.makeTensor(r,n,e)}var ix=U({rand_:nx}),Wr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function Yi(n,t){return t={exports:{}},n(t,t.exports),t.exports}var rx=Yi(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(Wr,n,!1)}),ax=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(Wr,n,!1)}),sx=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(Wr,n,!1)}),ox=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(Wr,n,!1)}),lx=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(Wr,n,!1)}),ux=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(Wr,n,!1)}),Ki=Yi(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});Ki.alea=rx;Ki.xor128=ax;Ki.xorwow=sx;Ki.xorshift7=ox;Ki.xor4096=lx;Ki.tychei=ux;var cx=Ki,Pc=cx.alea;var _c=function(){function n(t,e,i,r,a){this.mean=t,this.stdDev=e,this.dtype=i,this.nextVal=NaN,this.truncated=r,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);var s=a||Math.random();this.random=Pc(s.toString())}return n.prototype.nextValue=function(){if(!isNaN(this.nextVal)){var t=this.nextVal;return this.nextVal=NaN,t}for(var e,i,r=!1;!r;){var a=void 0,s=void 0,o=void 0;do a=2*this.random()-1,s=2*this.random()-1,o=a*a+s*s;while(o>=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),hx=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=Pc(a.toString()),this.randn=new _c(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a<e||Math.log(a)<i)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)},n.prototype.convertValue=function(t){return this.dtype==="float32"?t:Math.round(t)},n}(),dx=function(){function n(t,e,i,r){var a=this;if(t===void 0&&(t=0),e===void 0&&(e=1),this.canReturnFloat=function(){return a.dtype==null||a.dtype==="float32"},this.min=t,this.range=e-t,this.dtype=i,r==null&&(r=Math.random()),typeof r=="number"&&(r=r.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error("The difference between "+t+" - "+e+" <= 1 and dtype is not float");this.random=Pc(r)}return n.prototype.convertValue=function(t){return this.canReturnFloat()?t:Math.round(t)},n.prototype.nextValue=function(){return this.convertValue(this.min+this.range*this.random())},n}();function px(n,t,e,i,r){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e==null&&(e=1),i==null&&(i="float32"),i!=="float32"&&i!=="int32")throw new Error("Unsupported data type "+i);for(var a=new hx(t,e,i,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var fx=U({randomGamma_:px});function mx(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type "+i);for(var a=new _c(t,e,i,!1,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var gx=U({randomNormal_:mx});function vx(n,t,e,i,r){t===void 0&&(t=0),e===void 0&&(e=1),i===void 0&&(i="float32");for(var a=On(n,i),s=new dx(t,e,null,r),o=0;o<a.values.length;o++)a.values[o]=s.nextValue();return a.toTensor()}var Lg=U({randomUniform_:vx});function Ur(n,t){Fi(n);var e=Rn(n,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");var i=null;return ai(n,i,e,t)}function Mc(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e===0)throw new Error("Cannot have a step of zero");var r=function(){var s=n===t,o=n<t&&e<0,l=t<n&&e>1;if(s||o||l)return Kn([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=Lr(u,i);t<n&&e===1&&(e=-1),c[0]=n;for(var h=1;h<c.length;h++)c[h]=c[h-1]+e;return Ur(c,i)},a={start:n,stop:t,step:e,dtype:i};return z.runKernelFunc(r,{},null,Sf,a)}function yx(n){var t=O(n,"x","reciprocal"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.reciprocal(t);return r([t]),a},e,null,ou)}var Ig=U({reciprocal_:yx});function bx(n){var t=O(n,"x","relu"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu(t)},i={x:t};return z.runKernelFunc(e,i,null,lu)}var Ra=U({relu_:bx});function wx(n){var t=O(n,"x","relu6"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu6(t)},i={x:t};return z.runKernelFunc(e,i,null,du)}var Hc=U({relu6_:wx});function Sx(n,t){var e=O(n,"x","reverse"),i=function(s){var o=Qe(t,e.shape);if(e.rank===0)return Bi(e);var l=s.reverse(e,o);return V(l,e.shape)},r={x:e},a={dims:t};return z.runKernelFunc(i,r,null,pu,a)}var $n=U({reverse_:Sx});function Lx(n){var t=O(n,"x","reverse");return E(t.rank===1,function(){return"Error in reverse1D: x must be rank 1 but got rank "+t.rank+"."}),$n(t,0)}var Ix=U({reverse1d_:Lx});function Ax(n,t){var e=O(n,"x","reverse");return E(e.rank===2,function(){return"Error in reverse2D: x must be rank 2 but got rank "+e.rank+"."}),$n(e,t)}var Tx=U({reverse2d_:Ax});function Nx(n,t){var e=O(n,"x","reverse");return E(e.rank===3,function(){return"Error in reverse3D: x must be rank 3 but got rank "+e.rank+"."}),$n(e,t)}var xx=U({reverse3d_:Nx});function Cx(n,t){var e=O(n,"x","reverse");return E(e.rank===4,function(){return"Error in reverse4D: x must be rank 4 but got rank "+e.rank+"."}),$n(e,t)}var Rx=U({reverse4d_:Cx});function Ox(n){var t=O(n,"x","round"),e={x:t};return z.runKernelFunc(function(i){return i.round(t)},e,null,fu)}var Ag=U({round_:Ox});function Ex(n){var t=O(n,"x","rsqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.rsqrt(t);return r([t]),a},e,null,mu)}var Vc=U({rsqrt_:Ex});function Dx(n){var t=O(n,"x","selu"),e=function(r,a){var s=r.selu(t);return a([t]),s},i={x:t};return z.runKernelFunc(e,i,null,vu)}var Tg=U({selu_:Dx});function kx(n,t,e,i,r,a,s){a===void 0&&(a=[1,1]),s===void 0&&(s="NHWC");var o=O(n,"x","separableConv2d"),l=O(t,"depthwiseFilter","separableConv2d"),u=O(e,"pointwiseFilter","separableConv2d"),c=o,h=!1;if(o.rank===3&&(h=!0,c=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),s==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");E(c.rank===4,function(){return"Error in separableConv2d: input must be rank 4, but got "+("rank "+c.rank+".")}),E(l.rank===4,function(){return"Error in separableConv2d: depthwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.rank===4,function(){return"Error in separableConv2d: pointwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.shape[0]===1,function(){return"Error in separableConv2d: the first dimension of pointwise filter "+(" must be 1, but got "+u.shape[0]+".")}),E(u.shape[1]===1,function(){return"Error in separableConv2d: the second dimension of pointwise "+("filter must be 1, but got "+u.shape[1]+".")});var d=l.shape[2],p=l.shape[3];E(u.shape[2]===d*p,function(){return"Error in separableConv2d: the third dimension of pointwise filter "+("must be "+d*p+", ")+("but got "+u.shape[2]+".")});var f=Ta(c,l,i,r,s,a),m=1,g=Er(f,u,m,"valid",s);return h?V(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var Ng=U({separableConv2d_:kx});function Fx(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,l,h;return pe(this,function(d){switch(d.label){case 0:return e=O(n,"x","setdiff1d"),i=O(t,"y","setdiff1d"),E(e.dtype===i.dtype,function(){return"x and y should have the same dtype, but got x ("+e.dtype+") and y ("+i.dtype+")."}),E(e.rank===1,function(){return"x should be 1D tensor, but got x ("+e.shape+")."}),E(i.rank===1,function(){return"y should be 1D tensor, but got y ("+i.shape+")."}),[4,e.data()];case 1:return r=d.sent(),[4,i.data()];case 2:for(a=d.sent(),s=new Set(a),o=0,l=0;l<r.length;l++)s.has(r[l])||o++;for(u=new Ss([o],e.dtype),c=new Ss([o],"int32"),l=0,h=0;l<r.length;l++)s.has(r[l])||(u.values[h]=r[l],c.values[h]=l,h++);return[2,[u.toTensor(),c.toTensor()]]}})})}var Wx=Fx;function Ux(n){var t=O(n,"x","sign"),e={x:t};return z.runKernelFunc(function(i){return i.sign(t)},e,null,Su)}var xg=U({sign_:Ux});function Bx(n){var t=O(n,"x","sin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sin(t);return r([t]),a},e,null,bu)}var qc=U({sin_:Bx});function zx(n){var t=O(n,"x","sinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sinh(t);return r([t]),a},e,null,wu)}var Gc=U({sinh_:zx});function Px(n,t,e){var i=O(n,"x","slice1d");return E(i.rank===1,function(){return"slice1d expects a rank-1 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,[t],[e])}var _x=U({slice1d_:Px});function Mx(n,t,e){var i=O(n,"x","slice2d");return E(i.rank===2,function(){return"slice2d expects a rank-2 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Hx=U({slice2d_:Mx});function Vx(n,t,e){var i=O(n,"x","slice3d");return E(i.rank===3,function(){return"slice3d expects a rank-3 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var qx=U({slice3d_:Vx});function Gx(n,t,e){var i=O(n,"x","slice4d");return E(i.rank===4,function(){return"slice4d expects a rank-4 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Yx=U({slice4d_:Gx});function Kx(n,t){t===void 0&&(t=-1);var e=O(n,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and dim was "+t));var i={logits:e},r={dim:t};return z.runKernelFunc(function(a,s){var o=a.softmax(e,t);return s([o]),o},i,null,Cu,r)}var Cg=U({softmax_:Kx});function jx(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.fft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=n.as2D(r,i),s=e.fft(a);return s.reshape(n.shape)},t,null,Zp)}var qs=U({fft_:jx});function $x(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.ifft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=V(n,[r,i]),s=e.ifft(a);return V(s,n.shape)},t,null,rf)}var Oa=U({ifft_:$x});function Xx(n){var t=n.shape[n.shape.length-1],e=n.size/t,i;if(t<=2){var r=V(n,[e,t]);i=Oa(r)}else{var a=[e,2*(t-1)],s=V(Ca(n),[e,t]),o=V(Us(n),[e,t]),l=$n(ze(s,[0,1],[e,t-2]),1),u=J($n(ze(o,[0,1],[e,t-2]),1),we(-1)),c=xt([s,l],1),h=xt([o,u],1),r=V(ri(c,h),[a[0],a[1]]);i=Oa(r)}if(i=Ca(i),n.rank===3&&n.shape[0]!==0){var d=i,p=n.shape[0];i=V(i,[p,i.shape[0]/p,i.shape[1]]),d.dispose()}return i}var Yc=U({irfft_:Xx});function Rg(n,t,e){e===void 0&&(e=0);var i=[];if(typeof t=="number")E(n.shape[e]%t===0,function(){return"Number of splits must evenly divide the axis."}),i=new Array(t).fill(n.shape[e]/t);else{var r=t.reduce(function(o,l){return l===-1&&(o+=1),o},0);E(r<=1,function(){return"There should be only one negative value in split array."});var a=t.indexOf(-1);if(a!==-1){var s=t.reduce(function(o,l){return l>0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function Jx(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Qe(e,i.shape)[0],c=Rg(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,xu,s)}var Br=U({split_:Jx});function Zx(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&t<e){var a=n.shape.map(function(v){return 0}),s=n.shape.map(function(v){return v});s[n.shape.length-1]=t,r=ze(n,a,s),e=t}else if(t!=null&&t>e){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=xt([n,Kn(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(ri(r,l),[i,e]),c=qs(u),h=Math.floor(e/2)+1,d=Ca(c),p=Us(c),f=Br(d,[h,e-h],d.shape.length-1),m=Br(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(ri(f[0],m[0]),g)}var Gs=U({rfft_:Zx});function Qx(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,Au)}var _t=U({sqrt_:Qx});function eC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Ru,o)}var Ys=U({squaredDifference_:eC});function tC(n,t){var e=O(n,"x","squeeze");return V(e,Uf(e.shape,t).newShape)}var Ks=U({squeeze_:tC});function nC(n,t){t===void 0&&(t=0);var e=ba(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return mn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Be(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return mn(o,t)});return xt(s,t)}var ji=U({stack_:nC});function iC(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,zu,r)}var zr=U({step_:iC});function rC(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=Ns(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=Ns(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=Em(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var x=Ns(l);x.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=Lm(t,e,i),R=C.filter(function(W,F){return x.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,Cf,d)}var Og=U({stridedSlice_:rC});function aC(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,Eu)}var Eg=U({tan_:aC});function Ea(n,t,e){if(Fi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=Rn(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return ai(n,t,i,e)}function sC(n,t,e){if(Fi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=Rn(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function oC(n,t,e){if(Fi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=Rn(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function lC(n,t,e){if(Fi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=Rn(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,ai(n,t,i,e)}function uC(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,Rf,s),l=o[0],u=o[1];return{values:l,indices:u}}var Dg=U({topk_:uC});function cC(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new _c(t,e,i,!0,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var hC=U({truncatedNormal_:cC});function dC(n,t){t===void 0&&(t=0);var e=O(n,"x","unique",null);E(e.rank>0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(Of,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var kg=U({unique_:dC});function pC(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(rt(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,Uu,s)}var Kc=U({unsortedSegmentSum_:pC});function fC(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t<e.shape.length,function(){return"Axis = "+t+" is not in [-"+e.shape.length+", "+e.shape.length+")"}),t<0&&(t+=e.shape.length);var i={value:e},r={axis:t},a=function(s){return s.unstack(e,t)};return z.runKernelFunc(a,i,null,Wu,r)}var js=U({unstack_:fC});function mC(n,t,e,i){return t===void 0&&(t=!0),z.makeVariable(n,t,e,i)}function Fg(n,t){for(var e=[],i=0;i<t.length;i++)t[i]&&e.push(i);for(var r=On(n,"int32"),a=On([e.length,n.length],"int32"),i=0;i<e.length;i++){var s=r.indexToLoc(e[i]),o=i*n.length;a.values.set(s,o)}return a.toTensor()}function gC(n){return de(this,void 0,void 0,function(){var t,e,i;return pe(this,function(r){switch(r.label){case 0:return t=O(n,"condition","whereAsync","bool"),[4,t.data()];case 1:return e=r.sent(),i=Fg(t.shape,e),n!==t&&t.dispose(),[2,i]}})})}var Wg=gC;function vC(n,t,e){return de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:for(i=O(n,"tensor","boolMask"),r=O(t,"mask","boolMask","bool"),a=e??0,s=r.rank,o=i.shape,E(s>0,function(){return"mask cannot be scalar"}),Be(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u<a+s;u++)l*=o[u];return c=o.slice(0,a).concat([l],o.slice(a+s)),h=V(i,c),d=V(r,[-1]),[4,Wg(d)];case 1:return p=g.sent(),f=Ks(p,[1]),m=Ws(h,f,a),n!==i&&i.dispose(),t!==r&&r.dispose(),f.dispose(),h.dispose(),d.dispose(),p.dispose(),[2,m]}})})}var yC=vC;function bC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","notEqualStrict"),i=O(t,"b","notEqualStrict");return Be(e.shape,i.shape,"Error in notEqualStrict: "),Hs(e,i)}function wC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessStrict"),i=O(t,"b","lessStrict");return Be(e.shape,i.shape,"Error in lessStrict: "),Bs(e,i)}function SC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","equalStrict"),i=O(t,"b","equalStrict");return Be(e.shape,i.shape,"Error in equalStrict: "),Na(e,i)}function LC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessEqualStrict"),i=O(t,"b","lessEqualStrict");return Be(e.shape,i.shape,"Error in lessEqualStrict: "),Mi(e,i)}function IC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterStrict"),i=O(t,"b","greaterStrict");return Be(e.shape,i.shape,"Error in greaterStrict: "),hi(e,i)}function AC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterEqualStrict"),i=O(t,"b","greaterEqualStrict");return Be(e.shape,i.shape,"Error in greaterEqualStrict: "),_i(e,i)}var Ug=U({equalStrict_:SC}),Bg=U({greaterEqualStrict_:AC}),zg=U({greaterStrict_:IC}),Pg=U({lessEqualStrict_:LC}),_g=U({lessStrict_:wC}),Mg=U({notEqualStrict_:bC});function TC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","addStrict"),i=O(t,"b","addStrict");return Be(e.shape,i.shape,"Error in addStrict: "),fe(e,i)}function NC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","subStrict"),i=O(t,"b","subStrict");return Be(e.shape,i.shape,"Error in subStrict: "),ge(e,i)}function xC(n,t){return It("strict variants of ops have been deprecated and will be removed in future"),Be(n.shape,t.shape,"Error in powStrict: "),jn(n,t)}function CC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","mul"),i=O(t,"b","mul");return Be(e.shape,i.shape,"Error in multiplyStrict: "),J(e,i)}function RC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","div"),i=O(t,"b","div");return Be(e.shape,i.shape,"Error in divideStrict: "),Ie(e,i)}function OC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","modStrict"),i=O(t,"b","modStrict");return Be(e.shape,i.shape,"Error in modStrict: "),Uc(e,i)}function EC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","minimumStrict"),i=O(t,"b","minimumStrict");return Be(e.shape,i.shape,"Error in minimumStrict: "),Ms(e,i)}function DC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","maximumStrict"),i=O(t,"b","maximumStrict");return Be(e.shape,i.shape,"Error in maximumStrict: "),kr(e,i)}function kC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","squaredDifferenceStrict"),i=O(t,"b","squaredDifferenceStrict");return Be(e.shape,i.shape,"Error in squaredDifferenceStrict: "),Ys(e,i)}var Hg=U({addStrict_:TC}),Vg=U({divStrict_:RC}),qg=U({maximumStrict_:DC}),Gg=U({minimumStrict_:EC}),Yg=U({modStrict_:OC}),Kg=U({mulStrict_:CC}),jg=U({powStrict_:xC}),$g=U({squaredDifferenceStrict_:kC}),Xg=U({subStrict_:NC});function FC(n,t,e,i){t===void 0&&(t="euclidean"),e===void 0&&(e=null),i===void 0&&(i=!1),n=O(n,"x","norm");var r=Jg(n,t,e),a=r.shape;if(i){var s=Qe(e,n.shape);a=en(r.shape,s)}return V(r,a)}function Jg(n,t,e){if(e===void 0&&(e=null),n.rank===0)return Gt(n);if(n.rank!==1&&e===null)return Jg(V(n,[-1]),t,e);if(n.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return Ae(Gt(n),e);if(t===Infinity)return Vi(Gt(n),e);if(t===-Infinity)return _s(Gt(n),e);if(t==="euclidean"||t===2)return _t(Ae(jn(Gt(n),we(2,"int32")),e));throw new Error("Error in norm: invalid ord value: "+t)}if(Array.isArray(e)&&e.length===2){if(t===1)return Vi(Ae(Gt(n),e[0]),e[1]-1);if(t===Infinity)return Vi(Ae(Gt(n),e[1]),e[0]);if(t===-Infinity)return _s(Ae(Gt(n),e[1]),e[0]);if(t==="fro"||t==="euclidean")return _t(Ae(Ye(n),e));throw new Error("Error in norm: invalid ord value: "+t)}throw new Error("Error in norm: invalid axis: "+e)}var $s=U({norm_:FC});function WC(n,t,e,i,r){r===void 0&&(r=!0);var a=O(n,"v","movingAverage"),s=O(t,"x","movingAverage"),o=O(e,"decay","movingAverage");jf(a,s),E(ln(a.shape,s.shape),function(){return"Shape mismatch in v and x"});var l=we(1),u=ge(l,o),c=J(ge(s,a),u);if(r){E(i!=null,function(){return"When using zeroDebias: true, step is required."});var h=O(i,"step","movingAverage");c=Ie(c,ge(l,jn(o,h)))}return fe(a,c)}var UC=U({movingAverage_:WC});function BC(n,t,e){var i=O(n,"indices","scatterND","int32"),r=O(t,"updates","scatterND");mc(r,i,e);var a=function(l){return l.scatterND(i,r,e)},s={indices:i,updates:r},o={shape:e};return z.runKernelFunc(a,s,null,Tf,o)}var zC=U({scatterND_:BC});function PC(n,t,e,i){if(n.dtype!=="int32")throw new Error("tf.sparseToDense() expects the indices to be int32 type,"+(" but the dtype was "+n.dtype+"."));if(n.rank>2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function _C(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);PC(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,xf,l)}var MC=U({sparseToDense_:_C});function HC(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,tf)}var VC=U({gatherND_:HC});function qC(n,t){if(t==null)return n.shape.slice();if(ln(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i<n.shape.length;i++)t[i]==null&&n.shape[i]!=null?e.push(n.shape[i]):e.push(t[i]);return e}return t}function GC(n,t,e,i){var r=O(n,"x","dropout");if(E(r.dtype==="float32",function(){return"x has to be a floating point tensor since it's going to be "+("scaled, but got a "+r.dtype+" tensor instead.")}),E(t>=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=qC(r,e),s=1-t,o=Ie(Fs(fe(Lg(a,0,1,"float32",i),s)),s);return J(r,o)}var YC=U({dropout_:GC});function Zg(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function jc(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a<n;++a){var s=2*Math.PI*a/(n+i-1);r[a]=t-e*Math.cos(s)}return Ur(r,"float32")}function KC(n,t,e){return e===void 0&&(e=1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,g;return pe(this,function(v){switch(v.label){case 0:return i=O(n,"predictions","inTopK"),r=O(t,"targets","inTopK"),E(i.rank>1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Be(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=gs("bool",u),d=0;d<u;d++){for(p=d*c,f=s.subarray(p,p+c),m=[],g=0;g<f.length;g++)m.push({value:f[g],index:g});for(m.sort(function(b,w){return w.value-b.value}),h[d]=0,g=0;g<e;g++)if(m[g].index===o[d]){h[d]=1;break}}return n!==i&&i.dispose(),t!==r&&r.dispose(),[2,si(h,r.shape,"bool")]}})})}var jC=KC;function $C(n,t,e,i,r,a,s){a===void 0&&(a="NHWC");var o=n;n.rank===3&&(o=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var l=t;l.rank===3&&(l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]])),E(o.rank===4,function(){return"Error in conv2dDerFilter: input must be rank 4, but got shape "+(o.shape+".")}),E(l.rank===4,function(){return"Error in conv2dDerFilter: dy must be rank 4, but got shape "+(l.shape+".")}),E(e.length===4,function(){return"Error in conv2dDerFilter: filterShape must be length 4, but got "+(e+".")});var u=a==="NHWC"?o.shape[3]:o.shape[1],c=a==="NHWC"?l.shape[3]:l.shape[1];E(u===e[2],function(){return"Error in conv2dDerFilter: depth of input "+u+") must "+("match input depth in filter ("+e[2]+".")}),E(c===e[3],function(){return"Error in conv2dDerFilter: depth of dy ("+c+") must "+("match output depth for filter ("+e[3]+").")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerFilter: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var h=function(f){var m=1,g=Sa(a),v=Dn(o.shape,e,i,m,r,s,!1,g);return f.conv2dDerFilter(o,l,v)},d={x:o,dy:l},p={strides:i,pad:r,dataFormat:a,dimRoundingMode:s};return z.runKernelFunc(h,d,null,_p,p)}var $c=U({conv2DBackpropFilter_:$C});function Xs(n,t,e){if(e==null||e==="linear")return n;if(e==="relu")return J(n,zr(t));throw new Error("Cannot compute gradient for fused activation "+e+".")}function Js(n,t){var e=t,i=mt(n.shape,t.shape);return i.length>0&&(e=Ae(e,i)),V(e,n.shape)}function Zs(n,t,e){if(t==="linear")return n;if(t==="relu")return Ra(n);if(t==="elu")return Cc(n);if(t==="relu6")return Hc(n);if(t==="prelu")return zc(n,e);throw new Error("Unknown fused activation "+t+".")}var Qs=function(n,t){var e=n>0;return!e||t==="linear"};function XC(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",Qs(z.state.gradientDepth,d)===!1){var f=Er(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(rt(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(Pt(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Dn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var x=function(F,P){var H=P,_=H[0],K=H[1],j=H[2],q=H[3],G=Xs(F,j,d);E(ci(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=Tc(K.shape,G,_,i,r),X=$c(K,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=Js(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Mu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Mu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var JC=U({fusedConv2d_:XC});function ZC(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,Gp)}var Qg=U({depthwiseConv2dNativeBackpropFilter_:ZC});function QC(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Yp);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var ev=U({depthwiseConv2dNativeBackpropInput_:QC});function eR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(Qs(z.state.gradientDepth,d)===!1){var f=Ta(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(Pt(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(rt(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Dn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var x=function(F,P){E(ci(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],K=P[2],j=P[3],q=Xs(F,K,d),G=ev(_.shape,q,H,w),Z=Qg(_,q,H.shape,w);if(j!=null){var X=Js(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Hu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Hu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var tR=U({fusedDepthwiseConv2d_:eR});function nR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(Qs(z.state.gradientDepth,c)===!1){var d=We(e,i,a,o);return l!=null&&(d=fe(d,l)),Zs(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=at(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=ot(w),x=ot(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(ln(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[x,b,g]):V(f,[x,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=at(k,p)[0],et(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=Xs(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=We(ie,X,!1,!0),re=We(Z,ie,!0,!1)):!a&&o?(te=We(ie,X,!1,!1),re=We(ie,Z,!0,!1)):a&&!o?(te=We(X,ie,!1,!0),re=We(Z,ie,!1,!1)):(te=We(X,ie,!0,!0),re=We(ie,Z,!0,!0)),l!=null){var le=Js(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var K=kn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,_u,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return K(R,D)}else{var j=kn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,_u,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return j(R,D,k)}}var iR=U({fusedMatMul_:nR});var rR={__proto__:null,conv2d:JC,depthwiseConv2d:tR,matMul:iR};function aR(n){return jc(n,.54,.46)}var sR=U({hammingWindow_:aR});function oR(n){return jc(n,.5,.5)}var tv=U({hannWindow_:oR});function lR(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a<n.size;){var o=a+t-n.size,l=xt([ze(n,a,t-o),Rc([o],r)]);s.push(l),a+=e}return s.length===0?Ea([],[0,t]):V(xt(s),[s.length,t])}var nv=U({frame_:lR});function uR(n,t,e,i,r){r===void 0&&(r=tv),i==null&&(i=Zg(t));for(var a=nv(n,t,e),s=J(a,r(t)),o=[],l=0;l<a.shape[0];l++)o.push(Gs(ze(s,[l,0],[1,t]),i));return xt(o)}var cR=U({stft_:uR});function hR(n,t,e,i,r,a){var s=O(n,"image","cropAndResize"),o=O(t,"boxes","cropAndResize","float32"),l=O(e,"boxInd","cropAndResize","int32");r=r||"bilinear",a=a||0;var u=o.shape[0];E(s.rank===4,function(){return"Error in cropAndResize: image must be rank 4,"+("but got rank "+s.rank+".")}),E(o.rank===2&&o.shape[1]===4,function(){return"Error in cropAndResize: boxes must be have size ["+u+",4] "+("but had shape "+o.shape+".")}),E(l.rank===1&&l.shape[0]===u,function(){return"Error in cropAndResize: boxInd must be have size ["+u+"] "+("but had shape "+o.shape+".")}),E(i.length===2,function(){return"Error in cropAndResize: cropSize must be of length 2, but got "+("length "+i.length+".")}),E(i[0]>=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Vp,d);return p}var dR=U({cropAndResize_:hR});function pR(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(ef,e,{});return i}var fR=U({flipLeftRight_:pR});function mR(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(Ef,a,s);return o}var gR=U({rotateWithOffset_:mR});function Pr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function vR(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Pr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,vf,l)}var yR=U({nonMaxSuppression_:vR});function wR(n,t,e){var i=bR(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function bR(n,t,e){return LR(n,t,e||SR)}function SR(n,t){return n>t?1:n<t?-1:0}function LR(n,t,e){for(var i=0,r=n.length,a=0,s=!1;i<r;){a=i+(r-i>>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function iv(n,t,e,i,r){return Xc(n,t,e,i,r,0).selectedIndices}function rv(n,t,e,i,r,a){return Xc(n,t,e,i,r,0,!1,a,!0)}function av(n,t,e,i,r,a){return Xc(n,t,e,i,r,a,!0)}function Xc(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;c<t.length;c++)t[c]>r&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(sv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length<e&&u.length>0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m<r)break;for(var b=!1,w=d.length-1;w>=v;--w){var S=IR(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*AR(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&wR(u,f,sv))}var L=d.length,x=e-L;o&&x>0&&(d.push.apply(d,new Array(x).fill(0)),p.push.apply(p,new Array(x).fill(0)));var C={selectedIndices:Ur(d,"int32")};return s&&(C.selectedScores=Ur(p,"float32")),l&&(C.validOutputs=we(L,"int32")),C}function IR(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function AR(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function sv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function TR(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Pr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=iv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var NR=TR;function xR(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(bf,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var CR=U({nonMaxSuppressionWithScore_:xR});function RR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=av(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var OR=RR;function ER(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(yf,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var DR=U({nonMaxSuppressionPadded_:ER});function kR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=rv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var FR=kR;function WR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,hu,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var ov=U({resizeBilinear_:WR});function UR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,cu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var lv=U({resizeNearestNeighbor_:UR});function BR(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(Mc(0,s,1,"int32"),[-1,1]),u=Mc(0,o,1,"int32"),c=ge(l,u),h=qi(Mi(c,we(+t,"int32")),_i(c,we(-e,"int32"))),d=Kn([s,o],i.dtype);return V(ji(js(V(i,[-1,s,o])).map(function(p){return pn(h,p,d)})),r)}var zR=U({bandPart_:BR});function PR(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r<n.length;++r)i(r)}else t=!0,n=Br(n,n.shape[0],0).map(function(l){return Ks(l,[0])});E(n.length<=n[0].shape[0],function(){return"Gram-Schmidt: Number of vectors ("+n.length+") exceeds "+("number of dimensions ("+n[0].shape[0]+").")});for(var a=[],s=n,o=function(l){a.push(z.tidy(function(){var u=s[l];if(l>0)for(var c=0;c<l;++c){var h=J(Ae(J(a[c],u)),a[c]);u=ge(u,h)}return Ie(u,$s(u,"euclidean"))}))},r=0;r<n.length;++r)o(r);return t?ji(a,0):a}var _R=U({gramSchmidt_:PR});function MR(n,t){if(t===void 0&&(t=!1),E(n.rank>=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return uv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=js(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=uv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(ji(r,0),n.shape),o=V(ji(a,0),n.shape);return[s,o]}function uv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=ug(e),a=Bi(n),s=Ea([[1]],[1,1]),o=Bi(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=$s(g),b=ze(a,[h,h],[1,1]),w=pn(hi(b,0),Ea([[-1]]),Ea([[1]])),S=ge(b,J(w,v)),L=Ie(g,S);L.shape[0]===1?o=Bi(s):o=xt([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var x=gt(Ie(We(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(x,o),D=ut(o);if(h===0)a=ge(C,We(R,We(D,C)));else{var k=ge(C,We(R,We(D,C)));a=xt([ze(a,[0,0],[h,i]),k],0)}var W=ut(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,We(We(F,o),W));else{var P=ge(F,We(We(F,o),W));r=xt([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],zt([p,f,m])},c=0;c<l;++c)u(c);return!t&&e>i&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var HR=U({qr_:MR});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function VR(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ae(a);if(e===I.Reduction.MEAN){if(r==null)return xa(a);var s=i.size/r.size,o=Ie(Ae(a),Ae(r));return s>1?Ie(o,we(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Ie(Ae(a),we(i.size));var l=J(r,Fr(i.shape)),u=ue(Ae(Hs(l,we(0))),"float32");return Ie(Ae(a),u)}throw Error("Unknown reduction: "+e)}var Xn=U({computeWeightedLoss_:VR});function qR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Be(r.shape,a.shape,"Error in absoluteDifference: ");var o=Gt(ge(r,a));return Xn(o,s,i)}var GR=U({absoluteDifference_:qR});function YR(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Be(a.shape,s.shape,"Error in cosineDistance: ");var l=we(1),u=ge(l,Ae(J(a,s),e,!0));return Xn(u,o,r)}var KR=U({cosineDistance_:YR});function jR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Be(r.shape,a.shape,"Error in hingeLoss: ");var o=we(1);r=ge(J(we(2),r),o);var l=Ra(ge(o,J(r,a)));return Xn(l,s,i)}var $R=U({hingeLoss_:jR});function XR(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Be(a.shape,s.shape,"Error in huberLoss: ");var l=we(i),u=Gt(ge(s,a)),c=Ms(u,l),h=ge(u,c),d=fe(J(we(.5),Ye(c)),J(l,h));return Xn(d,o,r)}var JR=U({huberLoss_:XR});function ZR(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Be(a.shape,s.shape,"Error in logLoss: ");var l=we(1),u=we(i),c=gt(J(a,Hi(fe(s,u)))),h=J(ge(l,a),Hi(fe(ge(l,s),u))),d=ge(c,h);return Xn(d,o,r)}var QR=U({logLoss_:ZR});function eO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Be(r.shape,a.shape,"Error in meanSquaredError: ");var o=Ys(r,a);return Xn(o,s,i)}var tO=U({meanSquaredError_:eO});function nO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Be(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ra(i),a=J(i,e),s=Ec(fn(gt(Gt(i))));return fe(ge(r,a),s)}function iO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Be(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=nO(a,s);return Xn(h,o,r)}var rO=U({sigmoidCrossEntropy_:iO});function aO(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=kn(function(r,a,s){var o=!0,l=kc(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=gt(J(u,r)),h=Ae(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=en(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),fn(g))),J(V(p,v),ge(fn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function sO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Be(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(a.shape[1]);a=fe(J(a,ge(u,l)),Ie(l,c))}var h=aO(a,s);return Xn(h,o,r)}var oO=U({softmaxCrossEntropy_:sO});var lO={fft:qs,ifft:Oa,rfft:Gs,irfft:Yc},uO={hammingWindow:sR,hannWindow:tv,frame:nv,stft:cR},cO={flipLeftRight:fR,resizeNearestNeighbor:lv,resizeBilinear:ov,rotateWithOffset:gR,cropAndResize:dR,nonMaxSuppression:yR,nonMaxSuppressionAsync:NR,nonMaxSuppressionWithScore:CR,nonMaxSuppressionWithScoreAsync:OR,nonMaxSuppressionPadded:DR,nonMaxSuppressionPaddedAsync:FR},hO={bandPart:zR,gramSchmidt:_R,qr:HR},dO={absoluteDifference:GR,computeWeightedLoss:Xn,cosineDistance:KR,hingeLoss:$R,huberLoss:JR,logLoss:QR,meanSquaredError:tO,sigmoidCrossEntropy:rO,softmaxCrossEntropy:oO};var di=function(n){qn(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return zt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return gg(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&zt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:we(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(km);Object.defineProperty(di,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Jc=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;ft(function(){var d=fe(J(c,i.rho),J(Ye(u),1-i.rho)),p=J(Ie(_t(fe(h,i.epsilon)),_t(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ye(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(zt(this.accumulatedGrads.map(function(e){return e.variable})),zt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(di);ui(Jc);var Zc=function(n){qn(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:ft(function(){return Rc(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;ft(function(){var h=fe(c,Ye(u));c.assign(h);var d=fe(J(Ie(u,_t(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&zt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(di);ui(Zc);var Qc=function(n){qn(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],ft(function(){s.accBeta1=we(i).variable(),s.accBeta2=we(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:ft(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:ft(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ye(h),1-i.beta2)),g=Ie(f,a),v=Ie(m,s);d.assign(f),p.assign(m);var b=fe(J(Ie(g,fe(_t(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&zt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),ft(function(){a.accBeta1.assign(jn(a.beta1,a.iterations_+1)),a.accBeta2.assign(jn(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(di);ui(Qc);var eh=function(n){qn(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],ft(function(){o.iteration=we(0).variable(),o.accBeta1=we(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=Ie(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=Gt(h),v=kr(m,g);d.assign(f),p.assign(v);var b=fe(J(Ie(s,a),Ie(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&zt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(di);ui(eh);var eo=function(n){qn(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];ft(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=Um(we(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(di);ui(eo);var th=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=we(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;ft(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&zt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(eo);ui(th);var nh=function(n){qn(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;ft(function(){var d=fe(J(c,i.decay),J(Ye(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Ie(J(u,i.learningRate),_t(ge(d,fe(Ye(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ye(u),1-i.decay)),g=fe(J(h,i.momentum),Ie(J(u,i.learningRate),_t(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&zt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&zt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&zt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(di);ui(nh);var $i=function(){function n(){}return n.sgd=function(t){return new eo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new th(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new nh(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new Qc(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Jc(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new eh(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Zc(t,e)},n}();var pO={sgd:$i.sgd,momentum:$i.momentum,adadelta:$i.adadelta,adagrad:$i.adagrad,rmsprop:$i.rmsprop,adamax:$i.adamax,adam:$i.adam};var fO=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function mO(){return new Promise(function(n){return fO(function(){return n()})})}function gO(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function vO(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s<a;++s)r=r.concat([n[s+1]/t[s],t[s]]);r=r.concat(n.slice(a+1))}return r}function yO(n,t,e){e===void 0&&(e=!0);var i=[];if(e){i.push(t);for(var r=t+1;r<n;++r)r<=2*t?(i.push(r),i.push(r-(t+1))):i.push(r)}else{for(var a=[],s=[],r=1;r<n;++r)r>=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function bO(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a<n.length;++a)a<=t.length?i?r.push(t[a-1]*n[a]):r.push(n[a]/t[a-1]):r.push(n[a]);return r}function wO(n,t){for(var e=[0],i=0;i<t;++i)e.push(n[i][0]);return e}function SO(n,t,e){for(var i=n.slice(0,1),r=0;r<e;++r)i.push(n[r+1]-t[r][0]-t[r][1]);return i}var cv=1.7580993408473768,hv=1.0507009873554805;var LO=.3275911,IO=.254829592,AO=-.284496736,TO=1.421413741,NO=-1.453152027,xO=1.061405429;function CO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.warn.apply(console,n)}function RO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.log.apply(console,n)}function OO(n,t){if(n.length!==t.length)throw new Error("Cannot merge real and imag arrays of different lengths. real:"+(n.length+", imag: "+t.length+"."));for(var e=new Float32Array(n.length*2),i=0;i<e.length;i+=2)e[i]=n[i/2],e[i+1]=t[i/2];return e}function EO(n){for(var t=new Float32Array(n.length/2),e=new Float32Array(n.length/2),i=0;i<n.length;i+=2)t[i/2]=n[i],e[i/2]=n[i+1];return{real:t,imag:e}}function DO(n){for(var t=Math.ceil(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=0;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function kO(n){for(var t=Math.floor(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=2;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function FO(n,t){var e=n[t*2],i=n[t*2+1];return{real:e,imag:i}}function WO(n,t,e,i){n[i*2]=t,n[i*2+1]=e}function UO(n,t){for(var e=new Float32Array(n/2),i=new Float32Array(n/2),r=0;r<Math.ceil(n/2);r++){var a=(t?2:-2)*Math.PI*(r/n);e[r]=Math.cos(a),i[r]=Math.sin(a)}return{real:e,imag:i}}function BO(n,t,e){var i=(e?2:-2)*Math.PI*(n/t),r=Math.cos(i),a=Math.sin(i);return{real:r,imag:a}}function zO(n,t,e){if(t==="complex64"){if(n.dtype==="complex64")return n.clone();var i=Kn(n.shape),r=ue(n,"float32"),a=e.complex(r,i);return i.dispose(),r.dispose(),a}if(!_f(n.dtype,t))return z.makeTensorFromDataId(n.dataId,n.shape,t);if(n.dtype==="complex64"){var s=e.real(n),a=ue(s,t);return s.dispose(),a}if(t==="int32")return e.int(n);if(t==="bool"){var o=we(0,n.dtype),a=e.notEqual(n,o);return o.dispose(),a}else throw new Error("Error in Cast: failed to cast "+n.dtype+" to "+t)}function PO(n,t){return z.makeTensorFromDataId(n.dataId,t,n.dtype)}function _O(n,t,e){var i=(t-n)/(e-1),r=Lr(e,"float32");r[0]=n;for(var a=1;a<r.length;a++)r[a]=r[a-1]+i;return Ur(r,"float32")}var MO={__proto__:null,slice_util:Dm,segment_util:M1,castTensor:zO,reshapeTensor:PO,linspaceImpl:_O,upcastType:Ls,axesAreInnerMostDims:Sc,combineLocations:Pm,computeOutAndReduceShapes:_m,expandShapeToKeepDim:en,assertAxesAreInnerMostDims:vT,getAxesPermutation:tn,getUndoAxesPermutation:xs,getInnerMostAxes:En,getBroadcastDims:A1,getReductionAxes:mt,assertAndGetBroadcastShape:et,assertParamsConsistent:Xm,computeOutShape:Jm,computeDilation2DInfo:xT,computePool2DInfo:Rr,computePool3DInfo:Ia,computeConv2DInfo:Dn,computeConv3DInfo:La,computeDefaultPad:Ic,tupleValuesAreOne:ci,eitherStridesOrDilationsAreOne:Pt,convertConv2DDataFormat:Sa,getFusedDyActivation:Xs,getFusedBiasGradient:Js,applyActivation:Zs,shouldFuse:Qs,PARALLELIZE_THRESHOLD:Oc,computeOptimalWindowSize:z1,getImageCenter:gO,getReshaped:vO,getPermuted:yO,getReshapedPermuted:bO,getSliceBeginCoords:wO,getSliceSize:SO,prepareAndValidate:bm,validateUpdateShape:fc,validateInput:mc,calculateShapes:wm,SELU_SCALEALPHA:cv,SELU_SCALE:hv,ERF_P:LO,ERF_A1:IO,ERF_A2:AO,ERF_A3:TO,ERF_A4:NO,ERF_A5:xO,warn:CO,log:RO,mergeRealAndImagArrays:OO,splitRealAndImagArrays:EO,complexWithEvenIndex:DO,complexWithOddIndex:kO,getComplexWithIndex:FO,assignToTypedArray:WO,exponents:UO,exponent:BO,prepareSplitSize:Rg};function HO(n,t,e){var i=new Array(n.rank).fill(0),r=n.shape.slice();return t.map(function(a){var s=r.slice();s[e]=a;var o=ze(n,i,s);return i[e]+=a,o})}function VO(n,t){for(var e=new Array(n.rank),i=0;i<e.length;i++)e[i]=n.shape[i]*t[i];for(var r=On(e,n.dtype),i=0;i<r.values.length;++i){for(var a=r.indexToLoc(i),s=new Array(n.rank),o=0;o<s.length;o++)s[o]=a[o]%n.shape[o];var l=n.locToIndex(s);r.values[i]=n.values[l]}return r.toTensor()}function qO(n,t,e,i,r){for(var a=t[t.length-1],s=[n.length/a,a],o=s[0],l=s[1],u=gs(e,o*i),c=gs("int32",o*i),h=0;h<o;h++){for(var d=h*l,p=n.subarray(d,d+l),f=[],m=0;m<p.length;m++)f.push({value:p[m],index:m});f.sort(function(S,L){return L.value-S.value});for(var g=h*i,v=u.subarray(g,g+i),b=c.subarray(g,g+i),m=0;m<i;m++)v[m]=f[m].value,b[m]=f[m].index}var w=t.slice();return w[w.length-1]=i,[si(u,w,e),si(c,w,"int32")]}var GO={__proto__:null,nonMaxSuppressionV3Impl:iv,nonMaxSuppressionV4Impl:rv,nonMaxSuppressionV5Impl:av,split:HO,tile:VO,topkImpl:qO,whereImpl:Fg};var YO={kernelName:nl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,zr(ue(e,"float32"),-1))}}}};var KO={kernelName:il,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Ye(ue(e,"float32")),r=_t(ge(we(1),i));return gt(Ie(n,r))}}}};var jO={kernelName:rl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(ge(Ye(ue(e,"float32")),1));return Ie(n,i)}}}};var $O={kernelName:ps,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,i.shape)};return{a,b:s}}};var XO={kernelName:al,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var JO={kernelName:sl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var ZO={kernelName:ol,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var QO={kernelName:ll,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,_t(ge(we(1),Ye(ue(e,"float32")))))}}}};var eE={kernelName:ul,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(fe(we(1),Ye(ue(e,"float32"))));return Ie(n,i)}}}};var tE={kernelName:dl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=fe(Ye(e),Ye(i)),l=J(n,Ie(i,o)),u=mt(e.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,e.shape)},s=function(){var o=fe(Ye(e),Ye(i)),l=gt(J(n,Ie(e,o))),u=mt(i.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,i.shape)};return{a,b:s}}};var nE={kernelName:cl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(Ye(ue(e,"float32")),1))}}}};var iE={kernelName:hl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ge(we(1),Ye(ue(e,"float32"))))}}}};function rE(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(Pt(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(rt(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=Ia(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,zp,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var aE=U({avgPool3dBackprop_:rE});var sE={kernelName:fl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return aE(n,i,a,s,c,l,u)}}}};function oE(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=Rr(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,Bp,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var lE=U({avgPoolBackprop_:oE});var uE={kernelName:pl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return lE(n,i,a,s,o)}}}};var cE={kernelName:ml,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return We(n,a,!1,!0)},b:function(){return We(r,n,!0,!1)}}:!o&&l?{a:function(){return We(n,a,!1,!1)},b:function(){return We(n,r,!0,!1)}}:o&&!l?{a:function(){return We(a,n,!1,!0)},b:function(){return We(r,n,!1,!1)}}:{a:function(){return We(a,n,!0,!0)},b:function(){return We(n,r,!0,!0)}}}};var hE={kernelName:gl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return Vs(n,r,a)}}}};var dE={kernelName:vl,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o<s.length;o++)s[o]>1&&l.push(o);return{x:function(){return Ae(n,l,!0)}}}};var pE={kernelName:fs,gradFunc:function(n){return{x:function(){return n.clone()}}}};var fE={kernelName:yl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var mE={kernelName:bl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return pn(qi(_i(i,a),Mi(i,s)),n,Ee(n))}}}};var gE={kernelName:wl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Qe(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=Br(n,s,a);return o.map(function(l){return function(){return l}})}};var vE={kernelName:Sl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(ci(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return Tc(r.shape,n,a,l,u,c)},filter:function(){return $c(r,n,a.shape,l,u,c)}}}};var yE={kernelName:Ll,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Er(n,a,o,l,u,1,c)},filter:function(){return $c(n,r,a.shape,o,l,u,c)}}}};function bE(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=La(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,Mp,u)}var wE=U({conv3DBackpropFilter_:bE});var SE={kernelName:Il,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(ci(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return ng(o.shape,n,l,a,s)},filter:function(){return wE(o,n,l.shape,a,s)}}}};var LE={kernelName:Al,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(gt(qc(ue(e,"float32"))),n)}}}};var IE={kernelName:Tl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Gc(ue(e,"float32")),n)}}}};var AE={kernelName:Nl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=tn([a],i.rank),u=xc(n,a,s,!o);return l!=null&&(u=ut(u,l)),u}}}};var TE={kernelName:xl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(ci(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(Pt(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be  1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(rt(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Dn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return ev(c.shape,n,h,d)},filter:function(){return Qg(c,n,h.shape,d)}}}};var NE={kernelName:Cl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(jp,s,e)},filter:function(){return z.runKernel($p,o,e)}}}};var xE={kernelName:Rl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var CE={kernelName:Ol,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Xp)}}}};var RE={kernelName:El,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(fn(gt(Ye(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var OE={kernelName:Dl,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var EE={kernelName:kl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,fn(e))}}}};var DE={kernelName:Fl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var kE={kernelName:Wl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var FE={kernelName:Ul,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??we(1),u=mt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h<r.shape.length-1;++h)c.push(r.shape[h]);c.push(1)}var d=ge(r,a),p=J(n,l),f=Vc(fe(s,we(i))),m=J(J(J(f,f),f),we(-.5)),g=function(){return a.rank===1?V(J(J(n,Dr(V(f,[1,1,1,a.shape[0]]),c)),l),r.shape):V(J(J(n,f),l),r.shape)},v=function(){var L=J(J(f,we(-1)),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},b=function(){var L=J(J(m,d),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},w=function(){var L=J(d,f),x=J(n,L);return a.rank===1&&(x=Ae(x,u)),V(x,a.shape)},S=function(){var L=n;return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)};return{x:g,mean:v,variance:b,scale:w,offset:S}}};var WE={kernelName:Bl,inputsToSave:["x","indices"],gradFunc:function(n,t,e){var i=t[0],r=t[1],a=e.axis,s=Qe(a,i.shape)[0],o=function(){var l=i.shape,u=r.size,c=l.slice(0,s),h=c.length,d=l.slice(a,l.length).slice(1),p=d.length,f=dv(0,h),m=dv(h+1,h+1+p),g=pv([c,[u],d]),v=V(n,g),b=V(r,[u]),w=pv([[h],f,m]),S=ut(v,w),L=Kc(S,b,i.shape[s]),x=xs(w);return L=ut(L,x),L};return{x:o,indices:function(){return r}}}};function dv(n,t){for(var e=[],i=n;i<t;++i)e.push(i);return e}function pv(n){for(var t=[],e=0;e<n.length;++e)for(var i=0;i<n[e].length;++i)t.push(n[e][i]);return t}var UE={kernelName:zl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1];return{a:function(){return Ee(e)},b:function(){return Ee(i)}}}};var BE={kernelName:Pl,gradFunc:function(n){return{x:function(){return ue(n,"float32")}}}};var zE={kernelName:_l,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var PE={kernelName:Ml,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var _E={kernelName:Hl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ME={kernelName:ql,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(e,1))}}}};var HE={kernelName:Vl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ue(e,"float32"))}}}};var VE={kernelName:Gl,inputsToSave:[],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.axis;return{logits:function(){var a=!0,s=fn(i);return ge(n,J(Ae(n,r,a),s))}}}};function qE(n,t,e,i,r,a,s){i===void 0&&(i=5),r===void 0&&(r=1),a===void 0&&(a=1),s===void 0&&(s=.5);var o=function(c){return c.LRNGrad(e,n,t,i,r,a,s)},l={x:n,y:t,dy:e},u={depthRadius:i,bias:r,alpha:a,beta:s};return z.runKernelFunc(o,l,null,df,u)}var GE=U({localResponseNormalizationBackprop_:qE});var YE={kernelName:Yl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.depthRadius,l=s.bias,u=s.alpha,c=s.beta;return{x:function(){return GE(r,a,n,o,l,u,c)}}}};function fv(n,t,e,i,r){return t.rank<e.rank&&(t=V(t,en(t.shape,i))),n.rank<e.rank&&(n=V(n,en(n.shape,i))),{x:function(){var a=J(n,ue(Na(e,t),n.dtype));return r==null?a:ut(a,r)}}}var mv={kernelName:Kl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.reductionIndices,a=t[0],s=t[1],o=Qe(r,a.shape),l=tn(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var KE={kernelName:jl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(_i(e,i),"float32"))},a=function(){return J(n,ue(Bs(e,i),"float32"))};return{a:r,b:a}}};function jE(n,t,e,i,r,a,s,o){a===void 0&&(a=[1,1,1]);var l=O(n,"dy","maxPool3dBackprop"),u=O(t,"input","maxPool3dBackprop"),c=O(e,"output","maxPool3dBackprop"),h=l,d=u,p=c,f=!1;u.rank===4&&(f=!0,h=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=V(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),p=V(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),E(h.rank===5,function(){return"Error in maxPool3dBackprop: dy must be rank 5 but got rank "+(h.rank+".")}),E(d.rank===5,function(){return"Error in maxPool3dBackprop: input must be rank 5 but got rank "+(d.rank+".")}),E(p.rank===5,function(){return"Error in maxPool3dBackprop: output must be rank 5 but got rank "+(p.rank+".")}),E(Pt(r,a),function(){return"Error in maxPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+r+" and dilations '"+a+"'")}),o!=null&&E(rt(s),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+o+" but got pad "+s+".")});var m=function(w){var S=Ia(d.shape,i,r,a,s,o);return w.maxPool3dBackprop(h,d,p,S)},g={dy:h,input:d,output:p},v={filterSize:i,strides:r,dilations:a,pad:s,dimRoundingMode:o},b=z.runKernelFunc(m,g,null,ff,v);return f?V(b,[b.shape[1],b.shape[2],b.shape[3],b.shape[4]]):b}var $E=U({maxPool3dBackprop_:jE});var XE={kernelName:Xl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.dilations,c=s.pad,h=s.dimRoundingMode,d=u??[1,1,1];return{x:function(){return $E(n,r,a,o,l,d,c,h)}}}};function JE(n,t,e,i,r,a,s){var o=O(n,"dy","maxPoolBackprop"),l=O(t,"input","maxPoolBackprop"),u=O(e,"output","maxPoolBackprop");E(l.rank===o.rank,function(){return"Rank of input ("+l.rank+") does not match rank of dy "+("("+o.rank+")")}),E(o.rank===4,function(){return"Error in maxPoolBackprop: dy must be rank 4 but got rank "+(o.rank+".")}),E(l.rank===4,function(){return"Error in maxPoolBackprop: input must be rank 4 but got rank "+(l.rank+".")}),s!=null&&E(rt(a),function(){return"Error in maxPoolBackprop: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+a+".")});var c=function(p){var f=Rr(l.shape,i,r,1,a,s);return p.maxPoolBackprop(o,l,u,f)},h={dy:o,input:l,output:u},d={filterSize:i,strides:r,pad:a,dimRoundingMode:s};return z.runKernelFunc(c,h,null,pf,d)}var ZE=U({maxPoolBackprop_:JE});var QE={kernelName:$l,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.pad;return{x:function(){return ZE(n,r,a,o,l,u)}}}};var eD={kernelName:Jl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.axis,a=t[0],s=t[1],o=Qe(r,a.shape),l=tn(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var tD={kernelName:Zl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Mi(e,i),"float32"))},a=function(){return J(n,ue(hi(e,i),"float32"))};return{a:r,b:a}}};var nD={kernelName:Ql,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=mt(e.shape,r);return o.length>0?V(Ae(n,o),e.shape):n},s=function(){var o=J(n,gt(Fs(Ie(e,i)))),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var iD={kernelName:eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var rD={kernelName:tu,gradFunc:function(n){return{x:function(){return gt(n)}}}};var aD={kernelName:iu,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return Kn(e.shape,"float32")}}}};var sD={kernelName:nu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var gv={kernelName:ru,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var oD={kernelName:au,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=et(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,jn(a,ge(c,we(1))))),d=mt(a.shape,o);return d.length>0&&(h=Ae(h,d)),V(h,a.shape)},u=function(){var c=hi(a,0),h=pn(c,Hi(a),Ee(a)),d=J(n,J(r,h)),p=mt(s.shape,o);return p.length>0&&(d=Ae(d,p)),V(d,s.shape)};return{a:l,b:u}}};var lD={kernelName:su,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=hi(e,0);return{x:function(){return pn(r,n,J(n,i))},alpha:function(){var a=pn(r,Ee(n),J(n,e)),s=mt(i.shape,n.shape);return s.length>0&&(a=Ae(a,s)),V(a,i.shape)}}}};var uD={kernelName:ou,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,gt(Ye(e)))}}}};var cD={kernelName:du,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(Mi(e,6),zr(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var hD={kernelName:lu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue(zr(e),"float32"))}}}};var dD={kernelName:uu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var pD={kernelName:hu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,Af,e)};return{images:s}}};var fD={kernelName:cu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,If,e)};return{images:s}}};var mD={kernelName:pu,gradFunc:function(n,t,e){var i=e.dims,r=Qe(i,n.shape);return{x:function(){return $n(n,r)}}}};var gD={kernelName:fu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var vD={kernelName:mu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return gt(Ie(n,J(jn(e,1.5),2)))}}}};var yD={kernelName:gu,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(Ps(e),n.dtype))}}}};var bD={kernelName:vu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=hi(e,we(0)),r=we(cv),a=we(hv),s=J(n,a),o=J(J(n,r),fn(ue(e,"float32")));return pn(i,s,o)}}}};var wD={kernelName:Lu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(we(1),e)))}}}};var SD={kernelName:Su,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var LD={kernelName:bu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ks(ue(e,"float32")),n)}}}};var ID={kernelName:wu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Nc(ue(e,"float32")),n)}}}};var AD={kernelName:yu,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=gc(i,a,s),u=l[0],c=l[1],h=[],d=0;d<n.rank;d++)h.push([u[d],o[d]-u[d]-c[d]]);return{x:function(){return Gi(n,h)}}}};var TD={kernelName:Cu,outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.dim,a=!0,s=J(n,i);return{logits:function(){return ge(s,J(Ae(s,[r],a),i))}}}};var ND={kernelName:Iu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Pi(e))}}}};var vv={kernelName:Nu,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.paddings;return{x:function(){return Os(n,r,a)}}}};var yv={kernelName:xu,gradFunc:function(n,t,e){var i=e.axis;return{x:function(){return xt(n,i)}}}};var xD={kernelName:Au,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,J(_t(ue(e,"float32")),2))}}}};var CD={kernelName:Nf,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(ue(e,"float32"),2))}}}};var RD={kernelName:Ru,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=we(2),a=function(){return J(n,J(r,ge(e,i)))},s=function(){return J(n,J(r,ge(i,e)))};return{a,b:s}}};var OD={kernelName:zu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ED={kernelName:Ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(gt(o),i.shape)};return{a,b:s}}};var DD={kernelName:Tu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Qe(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Fr(i.shape,"float32"));return{x:function(){return l}}}};var kD={kernelName:Eu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Ye(ks(e)))}}}};var FD={kernelName:Du,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(we(1),Ye(e)),n)}}}};var WD={kernelName:ku,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o<r[0];++o)s=fe(s,ze(n,[o*i.shape[0]],[i.shape[0]]));else if(i.rank===2)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1]],[i.shape[0],i.shape[1]]));else if(i.rank===3)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2]],[i.shape[0],i.shape[1],i.shape[2]]));else if(i.rank===4)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)for(var c=0;c<r[3];++c)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2],c*i.shape[3]],[i.shape[0],i.shape[1],i.shape[2],i.shape[3]]));else throw new Error("Gradient for tile operation is not implemented for rank-"+(i.rank+" tensors yet."));return s};return{x:a}}};var UD={kernelName:Fu,gradFunc:function(n,t,e){var i=e,r=i.perm,a=xs(r);return{x:function(){return ut(n,a)}}}};var BD={kernelName:Wu,gradFunc:function(n,t,e){var i=e,r=i.axis;return{value:function(){return ji(n,r)}}}};var PD={kernelName:Uu,inputsToSave:["segmentIds"],gradFunc:function(n,t){var e=t[0],i=function(){return zD(n,e)};return{x:i}}};function zD(n,t){for(var e=kr(t,Ee(t)),i=Ws(n,e),r=_i(t,we(0,"int32")),a=i.rank-r.rank,s=0;s<a;++s)r=mn(r,s+1);r=qi(r,Fr(i.shape,"bool"));var o=Ee(i);return pn(r,i,o)}var _D={kernelName:Bu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var MD=[YO,KO,jO,$O,XO,JO,ZO,QO,eE,tE,nE,iE,sE,uE,cE,hE,dE,pE,fE,mE,gE,yE,vE,SE,LE,IE,AE,TE,NE,xE,CE,RE,OE,EE,kE,DE,FE,WE,UE,BE,zE,PE,_E,ME,HE,VE,YE,mv,mv,KE,XE,QE,eD,tD,nD,iD,rD,aD,sD,gv,gv,oD,lD,uD,cD,hD,dD,pD,fD,mD,gD,vD,yD,bD,wD,SD,LD,ID,AD,TD,ND,vv,vv,yv,yv,xD,RD,CD,OD,ED,DD,kD,FD,WD,UD,BD,PD,_D];for(var ih=0,bv=MD;ih<bv.length;ih++){var HD=bv[ih];kf(HD)}Y.prototype.abs=function(){return this.throwIfDisposed(),Gt(this)};Y.prototype.acos=function(){return this.throwIfDisposed(),Bm(this)};Y.prototype.acosh=function(){return this.throwIfDisposed(),zm(this)};Y.prototype.addStrict=function(n){return this.throwIfDisposed(),Hg(this,n)};Y.prototype.add=function(n){return this.throwIfDisposed(),fe(this,n)};Y.prototype.all=function(n,t){return this.throwIfDisposed(),Mm(this,n,t)};Y.prototype.any=function(n,t){return this.throwIfDisposed(),Hm(this,n,t)};Y.prototype.argMax=function(n){return this.throwIfDisposed(),Vm(this,n)};Y.prototype.argMin=function(n){return this.throwIfDisposed(),qm(this,n)};Y.prototype.asScalar=function(){return this.throwIfDisposed(),E(this.size===1,function(){return"The array must have only 1 element."}),V(this,[])};Y.prototype.asType=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.as1D=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.as2D=function(n,t){return this.throwIfDisposed(),V(this,[n,t])};Y.prototype.as3D=function(n,t,e){return this.throwIfDisposed(),V(this,[n,t,e])};Y.prototype.as4D=function(n,t,e,i){return this.throwIfDisposed(),V(this,[n,t,e,i])};Y.prototype.as5D=function(n,t,e,i,r){return this.throwIfDisposed(),V(this,[n,t,e,i,r])};Y.prototype.asin=function(){return this.throwIfDisposed(),Gm(this)};Y.prototype.asinh=function(){return this.throwIfDisposed(),Ym(this)};Y.prototype.atan=function(){return this.throwIfDisposed(),Km(this)};Y.prototype.atan2=function(n){return this.throwIfDisposed(),jm(this,n)};Y.prototype.atanh=function(){return this.throwIfDisposed(),$m(this)};Y.prototype.avgPool=function(n,t,e,i){return this.throwIfDisposed(),Ac(this,n,t,e,i)};Y.prototype.batchToSpaceND=function(n,t){return this.throwIfDisposed(),Os(this,n,t)};Y.prototype.batchNorm=function(n,t,e,i,r){return this.throwIfDisposed(),Aa(this,n,t,e,i,r)};Y.prototype.broadcastTo=function(n){return this.throwIfDisposed(),Ds(this,n)};Y.prototype.cast=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.ceil=function(){return this.throwIfDisposed(),Zm(this)};Y.prototype.clipByValue=function(n,t){return this.throwIfDisposed(),Qm(this,n,t)};Y.prototype.concat=function(n,t){return this.throwIfDisposed(),n instanceof Y&&(n=[n]),xt([this].concat(n),t)};Y.prototype.conv1d=function(n,t,e,i,r,a){return this.throwIfDisposed(),eg(this,n,t,e,i,r,a)};Y.prototype.conv2dTranspose=function(n,t,e,i,r){return this.throwIfDisposed(),tg(this,n,t,e,i,r)};Y.prototype.conv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Er(this,n,t,e,i,r,a)};Y.prototype.cos=function(){return this.throwIfDisposed(),ks(this)};Y.prototype.cosh=function(){return this.throwIfDisposed(),Nc(this)};Y.prototype.cumsum=function(n,t,e){return this.throwIfDisposed(),xc(this,n,t,e)};Y.prototype.depthToSpace=function(n,t){return this.throwIfDisposed(),ig(this,n,t)};Y.prototype.depthwiseConv2D=function(n,t,e,i,r,a){return It("depthwiseConv2D is deprecated, use depthwiseConv2d instead"),this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.depthwiseConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.dilation2d=function(n,t,e,i,r){return this.throwIfDisposed(),rg(this,n,t,e,i,r)};Y.prototype.divNoNan=function(n){return this.throwIfDisposed(),ag(this,n)};Y.prototype.divStrict=function(n){return this.throwIfDisposed(),Vg(this,n)};Y.prototype.div=function(n){return this.throwIfDisposed(),Ie(this,n)};Y.prototype.dot=function(n){return this.throwIfDisposed(),sg(this,n)};Y.prototype.elu=function(){return this.throwIfDisposed(),Cc(this)};Y.prototype.equalStrict=function(n){return this.throwIfDisposed(),Ug(this,n)};Y.prototype.equal=function(n){return this.throwIfDisposed(),Na(this,n)};Y.prototype.erf=function(){return this.throwIfDisposed(),og(this)};Y.prototype.exp=function(){return this.throwIfDisposed(),fn(this)};Y.prototype.expandDims=function(n){return this.throwIfDisposed(),mn(this,n)};Y.prototype.expm1=function(){return this.throwIfDisposed(),lg(this)};Y.prototype.fft=function(){return this.throwIfDisposed(),qs(this)};Y.prototype.flatten=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.floor=function(){return this.throwIfDisposed(),Fs(this)};Y.prototype.floorDiv=function(n){return this.throwIfDisposed(),wc(this,n)};Y.prototype.gather=function(n,t){return this.throwIfDisposed(),Ws(this,n,t)};Y.prototype.greaterEqualStrict=function(n){return this.throwIfDisposed(),Bg(this,n)};Y.prototype.greaterEqual=function(n){return this.throwIfDisposed(),_i(this,n)};Y.prototype.greaterStrict=function(n){return this.throwIfDisposed(),zg(this,n)};Y.prototype.greater=function(n){return this.throwIfDisposed(),hi(this,n)};Y.prototype.ifft=function(){return this.throwIfDisposed(),Oa(this)};Y.prototype.irfft=function(){return this.throwIfDisposed(),Yc(this)};Y.prototype.isFinite=function(){return this.throwIfDisposed(),hg(this)};Y.prototype.isInf=function(){return this.throwIfDisposed(),dg(this)};Y.prototype.isNaN=function(){return this.throwIfDisposed(),pg(this)};Y.prototype.leakyRelu=function(n){return this.throwIfDisposed(),fg(this,n)};Y.prototype.lessEqualStrict=function(n){return this.throwIfDisposed(),Pg(this,n)};Y.prototype.lessEqual=function(n){return this.throwIfDisposed(),Mi(this,n)};Y.prototype.lessStrict=function(n){return this.throwIfDisposed(),_g(this,n)};Y.prototype.less=function(n){return this.throwIfDisposed(),Bs(this,n)};Y.prototype.localResponseNormalization=function(n,t,e,i){return this.throwIfDisposed(),mg(this,n,t,e,i)};Y.prototype.logSigmoid=function(){return this.throwIfDisposed(),vg(this)};Y.prototype.logSoftmax=function(n){return this.throwIfDisposed(),yg(this,n)};Y.prototype.logSumExp=function(n,t){return this.throwIfDisposed(),kc(this,n,t)};Y.prototype.log=function(){return this.throwIfDisposed(),Hi(this)};Y.prototype.log1p=function(){return this.throwIfDisposed(),Ec(this)};Y.prototype.logicalAnd=function(n){return this.throwIfDisposed(),qi(this,n)};Y.prototype.logicalNot=function(){return this.throwIfDisposed(),Ps(this)};Y.prototype.logicalOr=function(n){return this.throwIfDisposed(),Fc(this,n)};Y.prototype.logicalXor=function(n){return this.throwIfDisposed(),bg(this,n)};Y.prototype.matMul=function(n,t,e){return this.throwIfDisposed(),We(this,n,t,e)};Y.prototype.maxPool=function(n,t,e,i){return this.throwIfDisposed(),Wc(this,n,t,e,i)};Y.prototype.max=function(n,t){return this.throwIfDisposed(),Vi(this,n,t)};Y.prototype.maximumStrict=function(n){return this.throwIfDisposed(),qg(this,n)};Y.prototype.maximum=function(n){return this.throwIfDisposed(),kr(this,n)};Y.prototype.mean=function(n,t){return this.throwIfDisposed(),xa(this,n,t)};Y.prototype.min=function(n,t){return this.throwIfDisposed(),_s(this,n,t)};Y.prototype.minimumStrict=function(n){return this.throwIfDisposed(),Gg(this,n)};Y.prototype.minimum=function(n){return this.throwIfDisposed(),Ms(this,n)};Y.prototype.modStrict=function(n){return this.throwIfDisposed(),Yg(this,n)};Y.prototype.mod=function(n){return this.throwIfDisposed(),Uc(this,n)};Y.prototype.mulStrict=function(n){return this.throwIfDisposed(),Kg(this,n)};Y.prototype.mul=function(n){return this.throwIfDisposed(),J(this,n)};Y.prototype.neg=function(){return this.throwIfDisposed(),gt(this)};Y.prototype.norm=function(n,t,e){return this.throwIfDisposed(),$s(this,n,t,e)};Y.prototype.notEqualStrict=function(n){return this.throwIfDisposed(),Mg(this,n)};Y.prototype.notEqual=function(n){return this.throwIfDisposed(),Hs(this,n)};Y.prototype.oneHot=function(n,t,e){return t===void 0&&(t=1),e===void 0&&(e=0),this.throwIfDisposed(),Ts(this,n,t,e)};Y.prototype.onesLike=function(){return this.throwIfDisposed(),Bc(this)};Y.prototype.pad=function(n,t){return this.throwIfDisposed(),Gi(this,n,t)};Y.prototype.pool=function(n,t,e,i,r){return this.throwIfDisposed(),wg(this,n,t,e,i,r)};Y.prototype.powStrict=function(n){return this.throwIfDisposed(),jg(this,n)};Y.prototype.pow=function(n){return this.throwIfDisposed(),jn(this,n)};Y.prototype.prelu=function(n){return this.throwIfDisposed(),zc(this,n)};Y.prototype.prod=function(n,t){return this.throwIfDisposed(),Sg(this,n,t)};Y.prototype.reciprocal=function(){return this.throwIfDisposed(),Ig(this)};Y.prototype.relu=function(){return this.throwIfDisposed(),Ra(this)};Y.prototype.relu6=function(){return this.throwIfDisposed(),Hc(this)};Y.prototype.reshapeAs=function(n){return this.throwIfDisposed(),V(this,n.shape)};Y.prototype.reshape=function(n){return this.throwIfDisposed(),V(this,n)};Y.prototype.resizeBilinear=function(n,t){return this.throwIfDisposed(),ov(this,n,t)};Y.prototype.resizeNearestNeighbor=function(n,t){return this.throwIfDisposed(),lv(this,n,t)};Y.prototype.reverse=function(n){return this.throwIfDisposed(),$n(this,n)};Y.prototype.rfft=function(){return this.throwIfDisposed(),Gs(this)};Y.prototype.round=function(){return this.throwIfDisposed(),Ag(this)};Y.prototype.rsqrt=function(){return this.throwIfDisposed(),Vc(this)};Y.prototype.selu=function(){return this.throwIfDisposed(),Tg(this)};Y.prototype.separableConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ng(this,n,t,e,i,r,a)};Y.prototype.sigmoid=function(){return this.throwIfDisposed(),Pi(this)};Y.prototype.sign=function(){return this.throwIfDisposed(),xg(this)};Y.prototype.sin=function(){return this.throwIfDisposed(),qc(this)};Y.prototype.sinh=function(){return this.throwIfDisposed(),Gc(this)};Y.prototype.slice=function(n,t){return this.throwIfDisposed(),ze(this,n,t)};Y.prototype.softmax=function(n){return this.throwIfDisposed(),Cg(this,n)};Y.prototype.softplus=function(){return this.throwIfDisposed(),Dc(this)};Y.prototype.spaceToBatchND=function(n,t){return this.throwIfDisposed(),Vs(this,n,t)};Y.prototype.split=function(n,t){return this.throwIfDisposed(),Br(this,n,t)};Y.prototype.sqrt=function(){return this.throwIfDisposed(),_t(this)};Y.prototype.square=function(){return this.throwIfDisposed(),Ye(this)};Y.prototype.squaredDifference=function(n){return this.throwIfDisposed(),Ys(this,n)};Y.prototype.squaredDifferenceStrict=function(n){return this.throwIfDisposed(),$g(this,n)};Y.prototype.squeeze=function(n){return this.throwIfDisposed(),Ks(this,n)};Y.prototype.stack=function(n,t){this.throwIfDisposed();var e=n instanceof Y?[this,n]:[this].concat(n);return ji(e,t)};Y.prototype.step=function(n){return this.throwIfDisposed(),zr(this,n)};Y.prototype.stridedSlice=function(n,t,e,i,r,a,s,o){return this.throwIfDisposed(),Og(this,n,t,e,i,r,a,s,o)};Y.prototype.subStrict=function(n){return this.throwIfDisposed(),Xg(this,n)};Y.prototype.sub=function(n){return this.throwIfDisposed(),ge(this,n)};Y.prototype.sum=function(n,t){return this.throwIfDisposed(),Ae(this,n,t)};Y.prototype.tan=function(){return this.throwIfDisposed(),Eg(this)};Y.prototype.tanh=function(){return this.throwIfDisposed(),Rs(this)};Y.prototype.tile=function(n){return this.throwIfDisposed(),Dr(this,n)};Y.prototype.toBool=function(){return this.throwIfDisposed(),ue(this,"bool")};Y.prototype.toFloat=function(){return this.throwIfDisposed(),ue(this,"float32")};Y.prototype.toInt=function(){return this.throwIfDisposed(),ue(this,"int32")};Y.prototype.topk=function(n,t){return this.throwIfDisposed(),Dg(this,n,t)};Y.prototype.transpose=function(n){return this.throwIfDisposed(),ut(this,n)};Y.prototype.unique=function(n){return this.throwIfDisposed(),kg(this,n)};Y.prototype.unsortedSegmentSum=function(n,t){return this.throwIfDisposed(),Kc(this,n,t)};Y.prototype.unstack=function(n){return this.throwIfDisposed(),js(this,n)};Y.prototype.where=function(n,t){return this.throwIfDisposed(),pn(n,this,t)};Y.prototype.zerosLike=function(){return this.throwIfDisposed(),Ee(this)};I.Abs=nl;I.Acos=il;I.Acosh=rl;I.AdadeltaOptimizer=Jc;I.AdagradOptimizer=Zc;I.AdamOptimizer=Qc;I.AdamaxOptimizer=eh;I.Add=ps;I.AddN=al;I.All=Wp;I.Any=Up;I.ArgMax=sl;I.ArgMin=ol;I.Asin=ll;I.Asinh=ul;I.Atan=cl;I.Atan2=dl;I.Atanh=hl;I.AvgPool=pl;I.AvgPool3D=fl;I.AvgPool3DBackprop=zp;I.AvgPoolBackprop=Bp;I.BatchMatMul=ml;I.BatchToSpaceND=gl;I.BroadcastTo=vl;I.Cast=fs;I.Ceil=yl;I.ClipByValue=bl;I.Complex=Pp;I.Concat=wl;I.Conv2D=Sl;I.Conv2DBackpropFilter=_p;I.Conv2DBackpropInput=Ll;I.Conv3D=Il;I.Conv3DBackpropFilterV2=Mp;I.Conv3DBackpropInputV2=Hp;I.Cos=Al;I.Cosh=Tl;I.CropAndResize=Vp;I.Cumsum=Nl;I.DataStorage=EI;I.DepthToSpace=qp;I.DepthwiseConv2dNative=xl;I.DepthwiseConv2dNativeBackpropFilter=Gp;I.DepthwiseConv2dNativeBackpropInput=Yp;I.Diag=Kp;I.Dilation2D=Cl;I.Dilation2DBackpropFilter=$p;I.Dilation2DBackpropInput=jp;I.Div=Rl;I.Elu=Ol;I.EluGrad=Xp;I.Environment=Dp;I.Equal=Jp;I.Erf=El;I.Exp=Dl;I.Expm1=kl;I.FFT=Zp;I.Fill=Qp;I.FlipLeftRight=ef;I.Floor=Fl;I.FloorDiv=Wl;I.FromPixels=Pu;I.FusedBatchNorm=Ul;I.FusedConv2D=Mu;I.FusedDepthwiseConv2D=Hu;I.GatherNd=tf;I.GatherV2=Bl;I.Greater=nf;I.GreaterEqual=zl;I.IFFT=rf;I.Identity=Pl;I.Imag=af;I.IsFinite=_l;I.IsInf=Ml;I.IsNan=Hl;I.KernelBackend=Op;I.LRN=Yl;I.LRNBackprop=df;I.Less=sf;I.LessEqual=of;I.LinSpace=lf;I.Log=Vl;I.Log1p=ql;I.LogSoftmax=Gl;I.LogicalAnd=uf;I.LogicalNot=cf;I.LogicalOr=hf;I.Max=Kl;I.MaxPool=$l;I.MaxPool3D=Xl;I.MaxPool3DBackprop=ff;I.MaxPoolBackprop=pf;I.MaxPoolWithArgmax=mf;I.Maximum=jl;I.Mean=BI;I.Min=Jl;I.Minimum=Zl;I.Mod=Ql;I.MomentumOptimizer=th;I.Multiply=eu;I.Negate=tu;I.NonMaxSuppressionV3=vf;I.NonMaxSuppressionV4=yf;I.NonMaxSuppressionV5=bf;I.NotEqual=gf;I.OP_SCOPE_SUFFIX=tm;I.OneHot=iu;I.OnesLike=nu;I.Optimizer=di;I.PadV2=ru;I.Pool=zI;I.Pow=au;I.Prelu=su;I.Prod=wf;I.RMSPropOptimizer=nh;I.Range=Sf;I.Real=Lf;I.Reciprocal=ou;I.Relu=lu;I.Relu6=du;I.Reshape=uu;I.ResizeBilinear=hu;I.ResizeBilinearGrad=Af;I.ResizeNearestNeighbor=cu;I.ResizeNearestNeighborGrad=If;I.Reverse=pu;I.RotateWithOffset=Ef;I.Round=fu;I.Rsqrt=mu;I.SGDOptimizer=eo;I.ScatterNd=Tf;I.SelectV2=gu;I.Selu=vu;I.Sigmoid=Lu;I.Sign=Su;I.Sin=bu;I.Sinh=wu;I.Slice=yu;I.Softmax=Cu;I.Softplus=Iu;I.SpaceToBatchND=Nu;I.SparseToDense=xf;I.SplitV=xu;I.Sqrt=Au;I.Square=Nf;I.SquaredDifference=Ru;I.Step=zu;I.StridedSlice=Cf;I.Sub=Ou;I.Sum=Tu;I.Tan=Eu;I.Tanh=Du;I.Tensor=Y;I.TensorBuffer=Ss;I.Tile=ku;I.TopK=Rf;I.Transpose=Fu;I.Unique=Of;I.Unpack=Wu;I.UnsortedSegmentSum=Uu;I.Variable=ya;I.ZerosLike=Bu;I._FusedMatMul=_u;I.abs=Gt;I.acos=Bm;I.acosh=zm;I.add=fe;I.addN=gT;I.addStrict=Hg;I.all=Mm;I.any=Hm;I.argMax=Vm;I.argMin=qm;I.asin=Gm;I.asinh=Ym;I.atan=Km;I.atan2=jm;I.atanh=$m;I.avgPool=Ac;I.avgPool3d=FT;I.backend=sT;I.backend_util=MO;I.basicLSTMCell=_T;I.batchNorm=Aa;I.batchNorm2d=GT;I.batchNorm3d=KT;I.batchNorm4d=$T;I.batchToSpaceND=Os;I.booleanMaskAsync=yC;I.broadcastTo=Ds;I.browser=OA;I.buffer=On;I.cast=ue;I.ceil=Zm;I.clipByValue=Qm;I.clone=Bi;I.complex=ri;I.concat=xt;I.concat1d=e1;I.concat2d=n1;I.concat3d=r1;I.concat4d=s1;I.conv1d=eg;I.conv2d=Er;I.conv2dTranspose=tg;I.conv3d=d1;I.conv3dTranspose=m1;I.copyRegisteredKernels=MI;I.cos=ks;I.cosh=Nc;I.cosineWindow=jc;I.cumsum=xc;I.customGrad=kn;I.deprecationWarn=It;I.depthToSpace=ig;I.depthwiseConv2d=Ta;I.device_util=I2;I.diag=L1;I.dilation2d=rg;I.disableDeprecationWarnings=KA;I.dispose=zt;I.disposeVariables=jA;I.div=Ie;I.divNoNan=ag;I.divStrict=Vg;I.dot=sg;I.dropout=YC;I.elu=Cc;I.enableDebugMode=YA;I.enableProdMode=GA;I.enclosingPowerOfTwo=Zg;I.engine=$A;I.env=He;I.equal=Na;I.equalStrict=Ug;I.erf=og;I.exp=fn;I.expandDims=mn;I.expm1=lg;I.eye=ug;I.fft=qs;I.fill=Rc;I.findBackend=iT;I.findBackendFactory=rT;I.floor=Fs;I.floorDiv=wc;I.fused=rR;I.gather=Ws;I.gatherND=VC;I.gather_util=EA;I.getBackend=tT;I.getGradient=Gu;I.getKernel=qu;I.getKernelsForBackend=ms;I.grad=iN;I.grads=rN;I.greater=hi;I.greaterEqual=_i;I.greaterEqualStrict=Bg;I.greaterStrict=zg;I.ifft=Oa;I.imag=Us;I.image=cO;I.inTopKAsync=jC;I.io=bA;I.irfft=Yc;I.isFinite=hg;I.isInf=dg;I.isNaN=pg;I.keep=Um;I.kernel_impls=GO;I.leakyRelu=fg;I.less=Bs;I.lessEqual=Mi;I.lessEqualStrict=Pg;I.lessStrict=_g;I.linalg=hO;I.linspace=Q1;I.localResponseNormalization=mg;I.log=Hi;I.log1p=Ec;I.logSigmoid=vg;I.logSoftmax=yg;I.logSumExp=kc;I.logicalAnd=qi;I.logicalNot=Ps;I.logicalOr=Fc;I.logicalXor=bg;I.losses=dO;I.matMul=We;I.math=NA;I.max=Vi;I.maxPool=Wc;I.maxPool3d=SN;I.maxPoolWithArgmax=IN;I.maximum=kr;I.maximumStrict=qg;I.mean=xa;I.memory=XA;I.min=_s;I.minimum=Ms;I.minimumStrict=Gg;I.mod=Uc;I.modStrict=Yg;I.moments=ON;I.movingAverage=UC;I.mul=J;I.mulStrict=Kg;I.multiRNNCell=DN;I.multinomial=FN;I.neg=gt;I.nextFrame=mO;I.norm=$s;I.notEqual=Hs;I.notEqualStrict=Mg;I.oneHot=Ts;I.ones=Fr;I.onesLike=Bc;I.op=U;I.outerProduct=PN;I.pad=Gi;I.pad1d=HN;I.pad2d=qN;I.pad3d=YN;I.pad4d=jN;I.pool=wg;I.pow=jn;I.powStrict=jg;I.prelu=zc;I.print=hm;I.prod=Sg;I.profile=JA;I.rand=ix;I.randomGamma=fx;I.randomNormal=gx;I.randomUniform=Lg;I.range=Mc;I.ready=eT;I.real=Ca;I.reciprocal=Ig;I.registerBackend=aT;I.registerGradient=kf;I.registerKernel=Df;I.relu=Ra;I.relu6=Hc;I.removeBackend=nT;I.reshape=V;I.reverse=$n;I.reverse1d=Ix;I.reverse2d=Tx;I.reverse3d=xx;I.reverse4d=Rx;I.rfft=Gs;I.round=Ag;I.rsqrt=Vc;I.scalar=we;I.scatterND=zC;I.scatter_util=DA;I.selu=Tg;I.separableConv2d=Ng;I.serialization=WA;I.setBackend=QA;I.setPlatform=oT;I.setdiff1dAsync=Wx;I.sigmoid=Pi;I.sign=xg;I.signal=uO;I.sin=qc;I.sinh=Gc;I.slice=ze;I.slice1d=_x;I.slice2d=Hx;I.slice3d=qx;I.slice4d=Yx;I.slice_util=Dm;I.softmax=Cg;I.softplus=Dc;I.spaceToBatchND=Vs;I.sparseToDense=MC;I.spectral=lO;I.split=Br;I.sqrt=_t;I.square=Ye;I.squaredDifference=Ys;I.squaredDifferenceStrict=$g;I.squeeze=Ks;I.stack=ji;I.step=zr;I.stridedSlice=Og;I.sub=ge;I.subStrict=Xg;I.sum=Ae;I.sumOutType=f2;I.tan=Eg;I.tanh=Rs;I.tensor=si;I.tensor1d=Ur;I.tensor2d=Ea;I.tensor3d=ym;I.tensor4d=sC;I.tensor5d=oC;I.tensor6d=lC;I.tensor_util=v2;I.test_util=VA;I.tidy=ft;I.tile=Dr;I.time=ZA;I.topk=Dg;I.train=pO;I.transpose=ut;I.truncatedNormal=hC;I.unique=kg;I.unregisterGradient=_I;I.unregisterKernel=PI;I.unsortedSegmentSum=Kc;I.unstack=js;I.upcastType=Ls;I.util=i2;I.valueAndGrad=aN;I.valueAndGrads=sN;I.variable=mC;I.variableGrads=gg;I.version_core=qA;I.where=pn;I.whereAsync=Wg;I.zeros=Kn;I.zerosLike=Ee});var Yb=be(Xe=>{"use strict";Object.defineProperty(Xe,"__esModule",{value:!0});var y=Xi();var rh=function(n,t){return rh=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},rh(n,t)};function Q(n,t){rh(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var Yt=function(){return Yt=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t},Yt.apply(this,arguments)};function VD(n,t){var e={};for(var i in n)Object.prototype.hasOwnProperty.call(n,i)&&t.indexOf(i)<0&&(e[i]=n[i]);if(n!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,i=Object.getOwnPropertySymbols(n);r<i.length;r++)t.indexOf(i[r])<0&&(e[i[r]]=n[i[r]]);return e}function Le(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function ve(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var ah;function vt(){return ah==null&&(ah=y.backend().epsilon()),ah}function gn(){return"channelsLast"}var pi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Fn=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),M=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Te=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),qD=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),SH=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error);function Ji(n,t){if(Array.isArray(n)){for(var e=[],i=0;i<t;i++)e=e.concat(n);return e}else{var e=new Array(t);return e.fill(n),e}}function Wn(n,t){if(!n)throw new qD(t)}function wv(n,t){for(var e=0,i=0,r=n;i<r.length;i++){var a=r[i];a===t&&e++}return e}function Mt(n){return n.length===1?n[0]:n}function Je(n){return Array.isArray(n)?n:[n]}function Jn(n){var t=n.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2"),e=t.replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function Zi(n){return n.length<=1||n.indexOf("_")===-1?n:n.replace(/[_]+(\w|$)/g,function(t,e){return e.toUpperCase()})}var un={};function sh(n){if(n==null)return null;var t={};return t.className=n.getClassName(),t.config=n.getConfig(),t}function oh(n){if(n==null||typeof n!="object")return;if(Array.isArray(n))n.forEach(function(s){return oh(s)});else for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e],a=n[r];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?n[r]=a.value:oh(a))}}function Da(n,t,e,i,r){var a,s,o;if(t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i="object"),r===void 0&&(r=!1),typeof n=="string"){var l=n,u=void 0;if(l in e)u=e[l];else if(l in un)u=un[l];else if(u=t[l],u==null)throw new M("Unknown "+i+": "+n+`. This may be due to one of the following reasons:
+        `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,qp,u)}var ig=U({depthToSpace_:b1});function w1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(rt(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(Pt(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Dn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,Cl,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ta=U({depthwiseConv2d_:w1});function S1(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Kp)}var L1=U({diag_:S1});function I1(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(Rl,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var rg=U({dilation2d_:I1});function A1(n,t){for(var e=n.length,i=[],r=0;r<e;r++){var a=e-1-r,s=n[a]||1,o=t[t.length-1-r]||1;o>1&&s===1&&i.unshift(a)}return i}function mt(n,t){for(var e=[],i=0;i<t.length;i++){var r=n[n.length-i-1],a=t.length-i-1,s=t[a];(r==null||r===1&&s>1)&&e.unshift(a)}return e}function et(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;r<i;r++){var a=n[n.length-r-1];a==null&&(a=1);var s=t[t.length-r-1];if(s==null&&(s=1),a===1)e.unshift(s);else if(s===1)e.unshift(a);else if(a!==s){var o="Operands could not be broadcast together with shapes "+(n+" and "+t+".");throw Error(o)}else e.unshift(a)}return e}function T1(n,t){var e,i=O(n,"a","equal"),r=O(t,"b","equal");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.equal(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Jp)}var Na=U({equal_:T1});function N1(n,t,e){var i=O(t,"a","where"),r=O(e,"b","where"),a=O(n,"condition","where","bool"),s=et(i.shape,r.shape),o=Ds(i,s),l=Ds(r,s);a.rank===1&&E(a.shape[0]===i.shape[0],function(){return"The first dimension of `a` must match the size of `condition`."}),a.rank!==1&&Be(a.shape,l.shape,"Error in where: ");var u=function(h,d){var p=h.select(a,o,l);return d([a]),p},c={condition:a,t:o,e:l};return z.runKernelFunc(u,c,null,vu)}var dn=U({where_:N1});function x1(n){var t=O(n,"x","zerosLike"),e={x:t};return z.runKernelFunc(function(i){return i.zerosLike(t)},e,null,zu)}var Ee=U({zerosLike_:x1});function C1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");e=at(i,r),i=e[0],r=e[1];var a=Ie(i,r),s=Ee(a),o=Na(r,s);return dn(o,s,a)}var ag=U({divNoNan_:C1});function R1(n,t){var e=O(n,"t1","dot"),i=O(t,"t2","dot");E((e.rank===1||e.rank===2)&&(i.rank===1||i.rank===2),function(){return"Error in dot: inputs must all be rank 1 or 2, but got ranks "+(e.rank+" and "+i.rank+".")});var r=e.rank===1?e.size:e.shape[1],a=i.rank===1?i.size:i.shape[0];if(E(r===a,function(){return"Error in dot: inner dimensions of inputs must match, but got "+(r+" and "+a+".")}),e.rank===1&&i.rank===1){var s=V(e,[1,-1]),o=V(i,[-1,1]),l=We(s,o);return V(l,[])}else if(e.rank===1&&i.rank===2){var s=V(e,[1,-1]),o=V(i,[i.shape[0],i.shape[1]]),l=We(s,o);return V(l,[l.size])}else if(e.rank===2&&i.rank===1){var o=V(i,[-1,1]),l=We(e,o);return V(l,[l.size])}else{var o=V(i,[i.shape[0],i.shape[1]]),l=We(e,o);return l}}var sg=U({dot_:R1});function O1(n){var t=O(n,"x","elu"),e=function(r,a){var s=r.elu(t);return a([s]),s},i={x:t};return z.runKernelFunc(e,i,null,El)}var Rc=U({elu_:O1});function E1(n){var t=O(n,"x","erf");E(t.dtype==="int32"||t.dtype==="float32",function(){return"Input dtype must be `int32` or `float32`."}),t.dtype==="int32"&&(t=ue(t,"float32"));var e={x:t};return z.runKernelFunc(function(i,r){var a=i.erf(t);return r([t]),a},e,null,Dl)}var og=U({erf_:E1});function D1(n){var t=O(n,"x","exp"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.exp(t);return r([a]),a},e,null,kl)}var pn=U({exp_:D1});function k1(n,t){t===void 0&&(t=0);var e=null,i=O(n,"x","expandDims",e);E(t<=i.rank,function(){return"Axis must be <= rank of the tensor"});var r=i.shape.slice();return t<0&&(E(-(i.rank+1)<=t,function(){return"Axis must be in the interval ["+-(i.rank+1)+", "+i.rank+"]"}),t=i.rank+t+1),r.splice(t,0,1),V(i,r)}var fn=U({expandDims_:k1});function F1(n){var t=O(n,"x","expm1"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.expm1(t);return r([t]),a},e,null,Fl)}var lg=U({expm1_:F1});function W1(n,t){var e=null,i=O(n,"x","tile",e);E(i.rank===t.length,function(){return"Error in transpose: rank of input "+i.rank+" "+("must match length of reps "+t+".")});var r=function(l,u){var c=l.tile(i,t);return u([i]),c},a=[i],s={x:i},o={reps:t};return z.runKernelFunc(r,s,null,Fu,o,a)}var Dr=U({tile_:W1});function U1(n,t,e,i){i===void 0&&(i="float32"),t==null&&(t=n);for(var r=On([n,t],i),a=n<=t?n:t,s=0;s<a;++s)r.set(1,s,s);var o=V(r.toTensor(),[n,t]);if(e==null)return o;if(e.length===1)return Dr(fn(o,0),[e[0],1,1]);if(e.length===2)return Dr(fn(fn(o,0),0),[e[0],e[1],1,1]);if(e.length===3)return Dr(fn(fn(fn(o,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error("eye() currently supports only 1D and 2D "+("batchShapes, but received "+e.length+"D."))}var ug=U({eye_:U1});function Oc(n,t,e){var i={shape:n,value:t,dtype:e};return z.runKernelFunc(function(r){return r.fill(n,t,e)},{},null,Qp,i)}function B1(n){var t=O(n,"x","floor"),e={x:t};return z.runKernelFunc(function(i){return i.floor(t)},e,null,Wl)}var Fs=U({floor_:B1});var Ec=30;function z1(n){return n<=Ec?n:ys(n,Math.floor(Math.sqrt(n)))}function P1(n,t){var e=!1,i;for(n<=Ec?(i=n,e=!0):i=ys(n,Math.floor(Math.sqrt(n)));!e;)i>t||i===n?e=!0:i=ys(n,i+1);return i}function _1(n,t,e){for(var i=[],r=n.length,a=0;a<r;a++)a!==t?i.push(n[a]):i.push(e);return i}function cg(n,t,e){for(var i=n.shape[e],r=[],a=1,s=1,o=0;o<e;o++)r.push(n.shape[o]),a*=n.shape[o];for(var o=0;o<t.rank;o++)r.push(t.shape[o]);for(var o=e+1;o<n.rank;o++)r.push(n.shape[o]),s*=n.shape[o];return{batchSize:a,sliceSize:s,dimSize:i,outputShape:r}}var M1={__proto__:null,segOpComputeOptimalWindowSize:P1,computeOutShape:_1,collectGatherOpShapeInfo:cg};function H1(n,t,e){e===void 0&&(e=0);var i=O(n,"x","gather"),r=O(t,"indices","gather","int32"),a={x:i,indices:r},s={axis:e},o=function(l,u){var c=Qe(e,i.shape)[0],h=cg(i,r,c),d=l.gather(i,V(r,[r.size]),c);return u([i,r]),V(d,h.outputShape)};return z.runKernelFunc(o,a,null,zl,s)}var Ws=U({gather_:H1});function V1(n,t){var e,i=O(n,"a","greater"),r=O(t,"b","greater");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.greater(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,nf)}var hi=U({greater_:V1});function q1(n,t){var e,i=O(n,"a","greaterEqual"),r=O(t,"b","greaterEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.greaterEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Pl)}var _i=U({greaterEqual_:q1});function G1(n){var t=O(n,"input","imag"),e=function(r){return r.imag(t)},i={input:t};return z.runKernelFunc(e,i,null,af)}var Us=U({imag_:G1});function Y1(n){var t=O(n,"x","isFinite"),e={x:t};return z.runKernelFunc(function(i){return i.isFinite(t)},e,null,Ml)}var hg=U({isFinite_:Y1});function K1(n){var t=O(n,"x","isInf"),e={x:t};return z.runKernelFunc(function(i){return i.isInf(t)},e,null,Hl)}var dg=U({isInf_:K1});function j1(n){var t=O(n,"x","isNaN"),e={x:t};return z.runKernelFunc(function(i){return i.isNaN(t)},e,null,Vl)}var pg=U({isNaN_:j1});function $1(n,t){var e,i=O(n,"a","maximum"),r=O(t,"b","maximum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.maximum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,$l)}var kr=U({maximum_:$1});function we(n,t){if((Ot(n)&&t!=="string"||Array.isArray(n))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&Ot(n)&&!(n instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");var e=[],i=[];return ai(n,e,i,t)}function X1(n,t){t===void 0&&(t=.2);var e=O(n,"x","leakyRelu");return kr(J(we(t),e),e)}var fg=U({leakyRelu_:X1});function J1(n,t){var e,i=O(n,"a","less"),r=O(t,"b","less");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.less(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,sf)}var Bs=U({less_:J1});function Z1(n,t){var e,i=O(n,"a","lessEqual"),r=O(t,"b","lessEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.lessEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,of)}var Mi=U({lessEqual_:Z1});function Q1(n,t,e){if(e<=0)throw new Error("The number of values should be positive.");var i={start:n,stop:t,num:e};return z.runKernelFunc(function(r){return r.linspace(n,t,e)},{},null,lf,i)}function eN(n,t,e,i,r){t===void 0&&(t=5),e===void 0&&(e=1),i===void 0&&(i=1),r===void 0&&(r=.5);var a=O(n,"x","localResponseNormalization");E(a.rank===4||a.rank===3,function(){return`Error in localResponseNormalization: x must be rank 3 or 4 but got
+               rank `+a.rank+"."}),E(rt(t),function(){return"Error in localResponseNormalization: depthRadius must be an "+("integer but got depthRadius "+t+".")});var s=a,o=!1;a.rank===3&&(o=!0,s=V(a,[1,a.shape[0],a.shape[1],a.shape[2]]));var l=function(d,p){var f=d.localResponseNormalization4D(s,t,e,i,r);return p([s,f]),f},u={x:s},c={depthRadius:t,bias:e,alpha:i,beta:r},h=z.runKernelFunc(l,u,null,Kl,c);return o?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var mg=U({localResponseNormalization_:eN});function tN(n){var t=O(n,"x","log"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log(t);return r([t]),a},e,null,ql)}var Hi=U({log_:tN});function nN(n){var t=O(n,"x","log1p"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log1p(t);return r([t]),a},e,null,Gl)}var Dc=U({log1p_:nN});function iN(n){return E(ii(n),function(){return"The f passed in grad(f) must be a function"}),function(t,e){var i=O(t,"x","tf.grad",null),r=e!=null?O(e,"dy","tf.grad"):null;return z.tidy(function(){var a=z.gradients(function(){return n(i)},[i],r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),zs(o),o[0]})}}function rN(n){return E(ii(n),function(){return"The f passed in grads(f) must be a function"}),function(t,e){E(Array.isArray(t),function(){return"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s"});var i=ba(t,"args","tf.grads",null),r=e!=null?O(e,"dy","tf.grads"):null;return z.tidy(function(){var a=z.gradients(function(){return n.apply(void 0,i)},i,r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(o),o})}}function aN(n){return E(ii(n),function(){return"The f passed in valueAndGrad(f) must be a function"}),function(t,e){E(t instanceof Y,function(){return"The x passed in valueAndGrad(f)(x) must be a tensor"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrad(f)(x, dy) must be a tensor"});var i=z.gradients(function(){return n(t)},[t],e),r=i.grads,a=i.value;return zs(r),{grad:r[0],value:a}}}function sN(n){return E(ii(n),function(){return"The f passed in valueAndGrads(f) must be a function"}),function(t,e){E(Array.isArray(t)&&t.every(function(r){return r instanceof Y}),function(){return"The args passed in valueAndGrads(f)(args) must be array of tensors"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrads(f)(args, dy) must be a tensor"});var i=z.gradients(function(){return n.apply(void 0,t)},t,e);return e!=null&&Be(i.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(i.grads),i}}function gg(n,t){E(ii(n),function(){return"The f passed in variableGrads(f) must be a function"}),E(t==null||Array.isArray(t)&&t.every(function(h){return h instanceof ya}),function(){return"The varList passed in variableGrads(f, varList) must be an array of variables"});var e=t!=null;if(!e){t=[];for(var i in z.registeredVariables)t.push(z.registeredVariables[i])}var r=e?t.filter(function(h){return!h.trainable}):null,a=t.length;t=t.filter(function(h){return h.trainable}),E(t.length>0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function kn(n){return z.customGrad(n)}function zs(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
+    the f you passed encloses all operations that lead from x to y.`)}function oN(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,nu)}var gt=U({neg_:oN});function lN(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,Au)}var kc=U({softplus_:lN});function uN(n){var t=O(n,"x","logSigmoid"),e=kn(function(i){var r=gt(kc(gt(i))),a=function(s){var o=J(s,Pi(gt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var vg=U({logSigmoid_:uN});function cN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=en(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=Qt(f.shape,Qe(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,jl,s)}var Vi=U({max_:cN});function hN(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Eu)}var ge=U({sub_:hN});function dN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Qe(t,i.shape),c=en(u,i.rank),h=u,d=i;c!=null&&(d=ut(i,c),h=En(h.length,i.rank));var p=o.sum(d,h);if(e){var f=Qt(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Nu,s)}var Ae=U({sum_:dN});function pN(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=Vi(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),Hi(Ae(pn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,Yl,a)}var yg=U({logSoftmax_:pN});function fN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Qe(t,i.shape),a=Vi(i,r,!0),s=ge(i,a),o=pn(s),l=Ae(o,r),u=Hi(l),c=fe(V(a,u.shape),u);if(e){var h=Qt(c.shape,r);return V(c,h)}return c}var Fc=U({logSumExp_:fN});function mN(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,uf)}var qi=U({logicalAnd_:mN});function gN(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,cf)}var Ps=U({logicalNot_:gN});function vN(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,hf)}var Wc=U({logicalOr_:vN});function yN(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return et(e.shape,i.shape),qi(Wc(n,t),Ps(qi(n,t)))}var bg=U({logicalXor_:yN});function bN(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(Pt(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&on(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,Xl,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Uc=U({maxPool_:bN});function wN(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Jl,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var SN=U({maxPool3d_:wN});function LN(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(mf,s,o);return{result:l[0],indexes:l[1]}}var IN=U({maxPoolWithArgmax_:LN});function Kn(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Kn(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=Lr(ot(n),t);return z.makeTensor(r,n,t)}function Fr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Fr(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=ju(ot(n),t);return z.makeTensor(r,n,t)}function AN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Qe(t,i.shape),a=_m(i.shape,r),s=a[1],o=ot(s),l=kn(function(u){var c=we(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Ie(h,c),p=Ae(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Ie(J(v,Fr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var xa=U({mean_:AN});function TN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=en(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=Qt(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Zl,s)}var _s=U({min_:TN});function NN(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ql)}var Ms=U({minimum_:NN});function xN(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,eu)}var Bc=U({mod_:xN});function CN(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ye=U({square_:CN});function RN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Qe(t,n.shape),r=xa(n,i,e),a=r.shape;e||(a=Qt(r.shape,i));var s=Ye(ge(ue(n,"float32"),V(r,a))),o=xa(s,i,e);return{mean:r,variance:o}}var ON=U({moments_:RN});function EN(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=ba(e,"c","multiRNNCell"),s=ba(i,"h","multiRNNCell"),o=r,l=[],u=0;u<n.length;u++){var c=n[u](o,a[u],s[u]);l.push(c[0]),l.push(c[1]),o=c[1]}for(var h=[],d=[],u=0;u<l.length;u+=2)h.push(l[u]),d.push(l[u+1]);return[h,d]}var DN=U({multiRNNCell_:EN});function kN(n,t,e,i){i===void 0&&(i=!1);var r=O(n,"logits","multinomial"),a=r.size,s=r.rank;if(a<2)throw new Error("Error in multinomial: you need at least 2 outcomes, but got "+(a+"."));if(s>2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var FN=U({multinomial_:kN});function WN(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,gf)}var Hs=U({notEqual_:WN});function UN(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,Lf)}var Ca=U({real_:UN});function BN(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=zc(Ca(t)),o=Ee(Us(t));return ri(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,iu)}var zc=U({onesLike_:BN});function zN(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return We(r,a)}var PN=U({outerProduct_:zN});function _N(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,au,a)}var Gi=U({pad_:_N});function MN(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),Gi(n,[t],e)}var HN=U({pad1d_:MN});function VN(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var qN=U({pad2d_:VN});function GN(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var YN=U({pad3d_:GN});function KN(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var jN=U({pad4d_:KN});function $N(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,xu,s)}var Vs=U({spaceToBatchND_:$N});function ZN(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(Pt(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=Rr(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=JN([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=XN([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:Vs(o,c,f),b=e==="avg"?function(){return Tc(v,t,a,g)}:function(){return Uc(v,t,a,g)},w=b(),S=d?w:Os(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function XN(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function JN(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var wg=U({pool_:ZN});function QN(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=at(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,su)}var jn=U({pow_:QN});function ex(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,ou)}var Pc=U({prelu_:ex});function tx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Qe(t,i.shape),u=en(l,i.rank),c=l,h=i;u!=null&&(h=ut(i,u),c=En(c.length,i.rank));var d=o.prod(h,c);if(e){var p=Qt(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,wf,s)}var Sg=U({prod_:tx});function nx(n,t,e){var i=ot(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a<i;a++)r[a]=t();return z.makeTensor(r,n,e)}var ix=U({rand_:nx}),Wr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function Yi(n,t){return t={exports:{}},n(t,t.exports),t.exports}var rx=Yi(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(Wr,n,!1)}),ax=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(Wr,n,!1)}),sx=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(Wr,n,!1)}),ox=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(Wr,n,!1)}),lx=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(Wr,n,!1)}),ux=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(Wr,n,!1)}),Ki=Yi(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});Ki.alea=rx;Ki.xor128=ax;Ki.xorwow=sx;Ki.xorshift7=ox;Ki.xor4096=lx;Ki.tychei=ux;var cx=Ki,_c=cx.alea;var Mc=function(){function n(t,e,i,r,a){this.mean=t,this.stdDev=e,this.dtype=i,this.nextVal=NaN,this.truncated=r,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);var s=a||Math.random();this.random=_c(s.toString())}return n.prototype.nextValue=function(){if(!isNaN(this.nextVal)){var t=this.nextVal;return this.nextVal=NaN,t}for(var e,i,r=!1;!r;){var a=void 0,s=void 0,o=void 0;do a=2*this.random()-1,s=2*this.random()-1,o=a*a+s*s;while(o>=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),hx=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=_c(a.toString()),this.randn=new Mc(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a<e||Math.log(a)<i)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)},n.prototype.convertValue=function(t){return this.dtype==="float32"?t:Math.round(t)},n}(),dx=function(){function n(t,e,i,r){var a=this;if(t===void 0&&(t=0),e===void 0&&(e=1),this.canReturnFloat=function(){return a.dtype==null||a.dtype==="float32"},this.min=t,this.range=e-t,this.dtype=i,r==null&&(r=Math.random()),typeof r=="number"&&(r=r.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error("The difference between "+t+" - "+e+" <= 1 and dtype is not float");this.random=_c(r)}return n.prototype.convertValue=function(t){return this.canReturnFloat()?t:Math.round(t)},n.prototype.nextValue=function(){return this.convertValue(this.min+this.range*this.random())},n}();function px(n,t,e,i,r){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e==null&&(e=1),i==null&&(i="float32"),i!=="float32"&&i!=="int32")throw new Error("Unsupported data type "+i);for(var a=new hx(t,e,i,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var fx=U({randomGamma_:px});function mx(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type "+i);for(var a=new Mc(t,e,i,!1,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var gx=U({randomNormal_:mx});function vx(n,t,e,i,r){t===void 0&&(t=0),e===void 0&&(e=1),i===void 0&&(i="float32");for(var a=On(n,i),s=new dx(t,e,null,r),o=0;o<a.values.length;o++)a.values[o]=s.nextValue();return a.toTensor()}var Lg=U({randomUniform_:vx});function Ur(n,t){Fi(n);var e=Rn(n,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");var i=null;return ai(n,i,e,t)}function Hc(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e===0)throw new Error("Cannot have a step of zero");var r=function(){var s=n===t,o=n<t&&e<0,l=t<n&&e>1;if(s||o||l)return Kn([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=Lr(u,i);t<n&&e===1&&(e=-1),c[0]=n;for(var h=1;h<c.length;h++)c[h]=c[h-1]+e;return Ur(c,i)},a={start:n,stop:t,step:e,dtype:i};return z.runKernelFunc(r,{},null,Sf,a)}function yx(n){var t=O(n,"x","reciprocal"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.reciprocal(t);return r([t]),a},e,null,lu)}var Ig=U({reciprocal_:yx});function bx(n){var t=O(n,"x","relu"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu(t)},i={x:t};return z.runKernelFunc(e,i,null,uu)}var Ra=U({relu_:bx});function wx(n){var t=O(n,"x","relu6"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu6(t)},i={x:t};return z.runKernelFunc(e,i,null,pu)}var Vc=U({relu6_:wx});function Sx(n,t){var e=O(n,"x","reverse"),i=function(s){var o=Qe(t,e.shape);if(e.rank===0)return Bi(e);var l=s.reverse(e,o);return V(l,e.shape)},r={x:e},a={dims:t};return z.runKernelFunc(i,r,null,fu,a)}var $n=U({reverse_:Sx});function Lx(n){var t=O(n,"x","reverse");return E(t.rank===1,function(){return"Error in reverse1D: x must be rank 1 but got rank "+t.rank+"."}),$n(t,0)}var Ix=U({reverse1d_:Lx});function Ax(n,t){var e=O(n,"x","reverse");return E(e.rank===2,function(){return"Error in reverse2D: x must be rank 2 but got rank "+e.rank+"."}),$n(e,t)}var Tx=U({reverse2d_:Ax});function Nx(n,t){var e=O(n,"x","reverse");return E(e.rank===3,function(){return"Error in reverse3D: x must be rank 3 but got rank "+e.rank+"."}),$n(e,t)}var xx=U({reverse3d_:Nx});function Cx(n,t){var e=O(n,"x","reverse");return E(e.rank===4,function(){return"Error in reverse4D: x must be rank 4 but got rank "+e.rank+"."}),$n(e,t)}var Rx=U({reverse4d_:Cx});function Ox(n){var t=O(n,"x","round"),e={x:t};return z.runKernelFunc(function(i){return i.round(t)},e,null,mu)}var Ag=U({round_:Ox});function Ex(n){var t=O(n,"x","rsqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.rsqrt(t);return r([t]),a},e,null,gu)}var qc=U({rsqrt_:Ex});function Dx(n){var t=O(n,"x","selu"),e=function(r,a){var s=r.selu(t);return a([t]),s},i={x:t};return z.runKernelFunc(e,i,null,yu)}var Tg=U({selu_:Dx});function kx(n,t,e,i,r,a,s){a===void 0&&(a=[1,1]),s===void 0&&(s="NHWC");var o=O(n,"x","separableConv2d"),l=O(t,"depthwiseFilter","separableConv2d"),u=O(e,"pointwiseFilter","separableConv2d"),c=o,h=!1;if(o.rank===3&&(h=!0,c=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),s==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");E(c.rank===4,function(){return"Error in separableConv2d: input must be rank 4, but got "+("rank "+c.rank+".")}),E(l.rank===4,function(){return"Error in separableConv2d: depthwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.rank===4,function(){return"Error in separableConv2d: pointwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.shape[0]===1,function(){return"Error in separableConv2d: the first dimension of pointwise filter "+(" must be 1, but got "+u.shape[0]+".")}),E(u.shape[1]===1,function(){return"Error in separableConv2d: the second dimension of pointwise "+("filter must be 1, but got "+u.shape[1]+".")});var d=l.shape[2],p=l.shape[3];E(u.shape[2]===d*p,function(){return"Error in separableConv2d: the third dimension of pointwise filter "+("must be "+d*p+", ")+("but got "+u.shape[2]+".")});var f=Ta(c,l,i,r,s,a),m=1,g=Er(f,u,m,"valid",s);return h?V(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var Ng=U({separableConv2d_:kx});function Fx(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,l,h;return pe(this,function(d){switch(d.label){case 0:return e=O(n,"x","setdiff1d"),i=O(t,"y","setdiff1d"),E(e.dtype===i.dtype,function(){return"x and y should have the same dtype, but got x ("+e.dtype+") and y ("+i.dtype+")."}),E(e.rank===1,function(){return"x should be 1D tensor, but got x ("+e.shape+")."}),E(i.rank===1,function(){return"y should be 1D tensor, but got y ("+i.shape+")."}),[4,e.data()];case 1:return r=d.sent(),[4,i.data()];case 2:for(a=d.sent(),s=new Set(a),o=0,l=0;l<r.length;l++)s.has(r[l])||o++;for(u=new Ss([o],e.dtype),c=new Ss([o],"int32"),l=0,h=0;l<r.length;l++)s.has(r[l])||(u.values[h]=r[l],c.values[h]=l,h++);return[2,[u.toTensor(),c.toTensor()]]}})})}var Wx=Fx;function Ux(n){var t=O(n,"x","sign"),e={x:t};return z.runKernelFunc(function(i){return i.sign(t)},e,null,Lu)}var xg=U({sign_:Ux});function Bx(n){var t=O(n,"x","sin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sin(t);return r([t]),a},e,null,wu)}var Gc=U({sin_:Bx});function zx(n){var t=O(n,"x","sinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sinh(t);return r([t]),a},e,null,Su)}var Yc=U({sinh_:zx});function Px(n,t,e){var i=O(n,"x","slice1d");return E(i.rank===1,function(){return"slice1d expects a rank-1 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,[t],[e])}var _x=U({slice1d_:Px});function Mx(n,t,e){var i=O(n,"x","slice2d");return E(i.rank===2,function(){return"slice2d expects a rank-2 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Hx=U({slice2d_:Mx});function Vx(n,t,e){var i=O(n,"x","slice3d");return E(i.rank===3,function(){return"slice3d expects a rank-3 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var qx=U({slice3d_:Vx});function Gx(n,t,e){var i=O(n,"x","slice4d");return E(i.rank===4,function(){return"slice4d expects a rank-4 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Yx=U({slice4d_:Gx});function Kx(n,t){t===void 0&&(t=-1);var e=O(n,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and dim was "+t));var i={logits:e},r={dim:t};return z.runKernelFunc(function(a,s){var o=a.softmax(e,t);return s([o]),o},i,null,Ru,r)}var Cg=U({softmax_:Kx});function jx(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.fft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=n.as2D(r,i),s=e.fft(a);return s.reshape(n.shape)},t,null,Zp)}var qs=U({fft_:jx});function $x(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.ifft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=V(n,[r,i]),s=e.ifft(a);return V(s,n.shape)},t,null,rf)}var Oa=U({ifft_:$x});function Xx(n){var t=n.shape[n.shape.length-1],e=n.size/t,i;if(t<=2){var r=V(n,[e,t]);i=Oa(r)}else{var a=[e,2*(t-1)],s=V(Ca(n),[e,t]),o=V(Us(n),[e,t]),l=$n(ze(s,[0,1],[e,t-2]),1),u=J($n(ze(o,[0,1],[e,t-2]),1),we(-1)),c=xt([s,l],1),h=xt([o,u],1),r=V(ri(c,h),[a[0],a[1]]);i=Oa(r)}if(i=Ca(i),n.rank===3&&n.shape[0]!==0){var d=i,p=n.shape[0];i=V(i,[p,i.shape[0]/p,i.shape[1]]),d.dispose()}return i}var Kc=U({irfft_:Xx});function Rg(n,t,e){e===void 0&&(e=0);var i=[];if(typeof t=="number")E(n.shape[e]%t===0,function(){return"Number of splits must evenly divide the axis."}),i=new Array(t).fill(n.shape[e]/t);else{var r=t.reduce(function(o,l){return l===-1&&(o+=1),o},0);E(r<=1,function(){return"There should be only one negative value in split array."});var a=t.indexOf(-1);if(a!==-1){var s=t.reduce(function(o,l){return l>0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function Jx(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Qe(e,i.shape)[0],c=Rg(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,Cu,s)}var Br=U({split_:Jx});function Zx(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&t<e){var a=n.shape.map(function(v){return 0}),s=n.shape.map(function(v){return v});s[n.shape.length-1]=t,r=ze(n,a,s),e=t}else if(t!=null&&t>e){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=xt([n,Kn(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(ri(r,l),[i,e]),c=qs(u),h=Math.floor(e/2)+1,d=Ca(c),p=Us(c),f=Br(d,[h,e-h],d.shape.length-1),m=Br(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(ri(f[0],m[0]),g)}var Gs=U({rfft_:Zx});function Qx(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,Tu)}var _t=U({sqrt_:Qx});function eC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Ou,o)}var Ys=U({squaredDifference_:eC});function tC(n,t){var e=O(n,"x","squeeze");return V(e,Uf(e.shape,t).newShape)}var Ks=U({squeeze_:tC});function nC(n,t){t===void 0&&(t=0);var e=ba(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return fn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Be(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return fn(o,t)});return xt(s,t)}var ji=U({stack_:nC});function iC(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,Pu,r)}var zr=U({step_:iC});function rC(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=Ns(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=Ns(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=Em(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var x=Ns(l);x.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=Lm(t,e,i),R=C.filter(function(W,F){return x.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,Cf,d)}var Og=U({stridedSlice_:rC});function aC(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,Du)}var Eg=U({tan_:aC});function Ea(n,t,e){if(Fi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=Rn(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return ai(n,t,i,e)}function sC(n,t,e){if(Fi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=Rn(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function oC(n,t,e){if(Fi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=Rn(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function lC(n,t,e){if(Fi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=Rn(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,ai(n,t,i,e)}function uC(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,Rf,s),l=o[0],u=o[1];return{values:l,indices:u}}var Dg=U({topk_:uC});function cC(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new Mc(t,e,i,!0,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var hC=U({truncatedNormal_:cC});function dC(n,t){t===void 0&&(t=0);var e=O(n,"x","unique",null);E(e.rank>0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(Of,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var kg=U({unique_:dC});function pC(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(rt(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,Bu,s)}var jc=U({unsortedSegmentSum_:pC});function fC(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t<e.shape.length,function(){return"Axis = "+t+" is not in [-"+e.shape.length+", "+e.shape.length+")"}),t<0&&(t+=e.shape.length);var i={value:e},r={axis:t},a=function(s){return s.unstack(e,t)};return z.runKernelFunc(a,i,null,Uu,r)}var js=U({unstack_:fC});function mC(n,t,e,i){return t===void 0&&(t=!0),z.makeVariable(n,t,e,i)}function Fg(n,t){for(var e=[],i=0;i<t.length;i++)t[i]&&e.push(i);for(var r=On(n,"int32"),a=On([e.length,n.length],"int32"),i=0;i<e.length;i++){var s=r.indexToLoc(e[i]),o=i*n.length;a.values.set(s,o)}return a.toTensor()}function gC(n){return de(this,void 0,void 0,function(){var t,e,i;return pe(this,function(r){switch(r.label){case 0:return t=O(n,"condition","whereAsync","bool"),[4,t.data()];case 1:return e=r.sent(),i=Fg(t.shape,e),n!==t&&t.dispose(),[2,i]}})})}var Wg=gC;function vC(n,t,e){return de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:for(i=O(n,"tensor","boolMask"),r=O(t,"mask","boolMask","bool"),a=e??0,s=r.rank,o=i.shape,E(s>0,function(){return"mask cannot be scalar"}),Be(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u<a+s;u++)l*=o[u];return c=o.slice(0,a).concat([l],o.slice(a+s)),h=V(i,c),d=V(r,[-1]),[4,Wg(d)];case 1:return p=g.sent(),f=Ks(p,[1]),m=Ws(h,f,a),n!==i&&i.dispose(),t!==r&&r.dispose(),f.dispose(),h.dispose(),d.dispose(),p.dispose(),[2,m]}})})}var yC=vC;function bC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","notEqualStrict"),i=O(t,"b","notEqualStrict");return Be(e.shape,i.shape,"Error in notEqualStrict: "),Hs(e,i)}function wC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessStrict"),i=O(t,"b","lessStrict");return Be(e.shape,i.shape,"Error in lessStrict: "),Bs(e,i)}function SC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","equalStrict"),i=O(t,"b","equalStrict");return Be(e.shape,i.shape,"Error in equalStrict: "),Na(e,i)}function LC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessEqualStrict"),i=O(t,"b","lessEqualStrict");return Be(e.shape,i.shape,"Error in lessEqualStrict: "),Mi(e,i)}function IC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterStrict"),i=O(t,"b","greaterStrict");return Be(e.shape,i.shape,"Error in greaterStrict: "),hi(e,i)}function AC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterEqualStrict"),i=O(t,"b","greaterEqualStrict");return Be(e.shape,i.shape,"Error in greaterEqualStrict: "),_i(e,i)}var Ug=U({equalStrict_:SC}),Bg=U({greaterEqualStrict_:AC}),zg=U({greaterStrict_:IC}),Pg=U({lessEqualStrict_:LC}),_g=U({lessStrict_:wC}),Mg=U({notEqualStrict_:bC});function TC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","addStrict"),i=O(t,"b","addStrict");return Be(e.shape,i.shape,"Error in addStrict: "),fe(e,i)}function NC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","subStrict"),i=O(t,"b","subStrict");return Be(e.shape,i.shape,"Error in subStrict: "),ge(e,i)}function xC(n,t){return It("strict variants of ops have been deprecated and will be removed in future"),Be(n.shape,t.shape,"Error in powStrict: "),jn(n,t)}function CC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","mul"),i=O(t,"b","mul");return Be(e.shape,i.shape,"Error in multiplyStrict: "),J(e,i)}function RC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","div"),i=O(t,"b","div");return Be(e.shape,i.shape,"Error in divideStrict: "),Ie(e,i)}function OC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","modStrict"),i=O(t,"b","modStrict");return Be(e.shape,i.shape,"Error in modStrict: "),Bc(e,i)}function EC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","minimumStrict"),i=O(t,"b","minimumStrict");return Be(e.shape,i.shape,"Error in minimumStrict: "),Ms(e,i)}function DC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","maximumStrict"),i=O(t,"b","maximumStrict");return Be(e.shape,i.shape,"Error in maximumStrict: "),kr(e,i)}function kC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","squaredDifferenceStrict"),i=O(t,"b","squaredDifferenceStrict");return Be(e.shape,i.shape,"Error in squaredDifferenceStrict: "),Ys(e,i)}var Hg=U({addStrict_:TC}),Vg=U({divStrict_:RC}),qg=U({maximumStrict_:DC}),Gg=U({minimumStrict_:EC}),Yg=U({modStrict_:OC}),Kg=U({mulStrict_:CC}),jg=U({powStrict_:xC}),$g=U({squaredDifferenceStrict_:kC}),Xg=U({subStrict_:NC});function FC(n,t,e,i){t===void 0&&(t="euclidean"),e===void 0&&(e=null),i===void 0&&(i=!1),n=O(n,"x","norm");var r=Jg(n,t,e),a=r.shape;if(i){var s=Qe(e,n.shape);a=Qt(r.shape,s)}return V(r,a)}function Jg(n,t,e){if(e===void 0&&(e=null),n.rank===0)return qt(n);if(n.rank!==1&&e===null)return Jg(V(n,[-1]),t,e);if(n.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return Ae(qt(n),e);if(t===Infinity)return Vi(qt(n),e);if(t===-Infinity)return _s(qt(n),e);if(t==="euclidean"||t===2)return _t(Ae(jn(qt(n),we(2,"int32")),e));throw new Error("Error in norm: invalid ord value: "+t)}if(Array.isArray(e)&&e.length===2){if(t===1)return Vi(Ae(qt(n),e[0]),e[1]-1);if(t===Infinity)return Vi(Ae(qt(n),e[1]),e[0]);if(t===-Infinity)return _s(Ae(qt(n),e[1]),e[0]);if(t==="fro"||t==="euclidean")return _t(Ae(Ye(n),e));throw new Error("Error in norm: invalid ord value: "+t)}throw new Error("Error in norm: invalid axis: "+e)}var $s=U({norm_:FC});function WC(n,t,e,i,r){r===void 0&&(r=!0);var a=O(n,"v","movingAverage"),s=O(t,"x","movingAverage"),o=O(e,"decay","movingAverage");jf(a,s),E(on(a.shape,s.shape),function(){return"Shape mismatch in v and x"});var l=we(1),u=ge(l,o),c=J(ge(s,a),u);if(r){E(i!=null,function(){return"When using zeroDebias: true, step is required."});var h=O(i,"step","movingAverage");c=Ie(c,ge(l,jn(o,h)))}return fe(a,c)}var UC=U({movingAverage_:WC});function BC(n,t,e){var i=O(n,"indices","scatterND","int32"),r=O(t,"updates","scatterND");gc(r,i,e);var a=function(l){return l.scatterND(i,r,e)},s={indices:i,updates:r},o={shape:e};return z.runKernelFunc(a,s,null,Tf,o)}var zC=U({scatterND_:BC});function PC(n,t,e,i){if(n.dtype!=="int32")throw new Error("tf.sparseToDense() expects the indices to be int32 type,"+(" but the dtype was "+n.dtype+"."));if(n.rank>2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function _C(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);PC(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,xf,l)}var MC=U({sparseToDense_:_C});function HC(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,tf)}var VC=U({gatherND_:HC});function qC(n,t){if(t==null)return n.shape.slice();if(on(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i<n.shape.length;i++)t[i]==null&&n.shape[i]!=null?e.push(n.shape[i]):e.push(t[i]);return e}return t}function GC(n,t,e,i){var r=O(n,"x","dropout");if(E(r.dtype==="float32",function(){return"x has to be a floating point tensor since it's going to be "+("scaled, but got a "+r.dtype+" tensor instead.")}),E(t>=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=qC(r,e),s=1-t,o=Ie(Fs(fe(Lg(a,0,1,"float32",i),s)),s);return J(r,o)}var YC=U({dropout_:GC});function Zg(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function $c(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a<n;++a){var s=2*Math.PI*a/(n+i-1);r[a]=t-e*Math.cos(s)}return Ur(r,"float32")}function KC(n,t,e){return e===void 0&&(e=1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,g;return pe(this,function(v){switch(v.label){case 0:return i=O(n,"predictions","inTopK"),r=O(t,"targets","inTopK"),E(i.rank>1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Be(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=gs("bool",u),d=0;d<u;d++){for(p=d*c,f=s.subarray(p,p+c),m=[],g=0;g<f.length;g++)m.push({value:f[g],index:g});for(m.sort(function(b,w){return w.value-b.value}),h[d]=0,g=0;g<e;g++)if(m[g].index===o[d]){h[d]=1;break}}return n!==i&&i.dispose(),t!==r&&r.dispose(),[2,si(h,r.shape,"bool")]}})})}var jC=KC;function $C(n,t,e,i,r,a,s){a===void 0&&(a="NHWC");var o=n;n.rank===3&&(o=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var l=t;l.rank===3&&(l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]])),E(o.rank===4,function(){return"Error in conv2dDerFilter: input must be rank 4, but got shape "+(o.shape+".")}),E(l.rank===4,function(){return"Error in conv2dDerFilter: dy must be rank 4, but got shape "+(l.shape+".")}),E(e.length===4,function(){return"Error in conv2dDerFilter: filterShape must be length 4, but got "+(e+".")});var u=a==="NHWC"?o.shape[3]:o.shape[1],c=a==="NHWC"?l.shape[3]:l.shape[1];E(u===e[2],function(){return"Error in conv2dDerFilter: depth of input "+u+") must "+("match input depth in filter ("+e[2]+".")}),E(c===e[3],function(){return"Error in conv2dDerFilter: depth of dy ("+c+") must "+("match output depth for filter ("+e[3]+").")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerFilter: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var h=function(f){var m=1,g=Sa(a),v=Dn(o.shape,e,i,m,r,s,!1,g);return f.conv2dDerFilter(o,l,v)},d={x:o,dy:l},p={strides:i,pad:r,dataFormat:a,dimRoundingMode:s};return z.runKernelFunc(h,d,null,_p,p)}var Xc=U({conv2DBackpropFilter_:$C});function Xs(n,t,e){if(e==null||e==="linear")return n;if(e==="relu")return J(n,zr(t));throw new Error("Cannot compute gradient for fused activation "+e+".")}function Js(n,t){var e=t,i=mt(n.shape,t.shape);return i.length>0&&(e=Ae(e,i)),V(e,n.shape)}function Zs(n,t,e){if(t==="linear")return n;if(t==="relu")return Ra(n);if(t==="elu")return Rc(n);if(t==="relu6")return Vc(n);if(t==="prelu")return Pc(n,e);throw new Error("Unknown fused activation "+t+".")}var Qs=function(n,t){var e=n>0;return!e||t==="linear"};function XC(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",Qs(z.state.gradientDepth,d)===!1){var f=Er(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(rt(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(Pt(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Dn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var x=function(F,P){var H=P,_=H[0],K=H[1],j=H[2],q=H[3],G=Xs(F,j,d);E(ci(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=Nc(K.shape,G,_,i,r),X=Xc(K,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=Js(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Hu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Hu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var JC=U({fusedConv2d_:XC});function ZC(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,Gp)}var Qg=U({depthwiseConv2dNativeBackpropFilter_:ZC});function QC(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Yp);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var ev=U({depthwiseConv2dNativeBackpropInput_:QC});function eR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(Qs(z.state.gradientDepth,d)===!1){var f=Ta(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(Pt(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(rt(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Dn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var x=function(F,P){E(ci(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],K=P[2],j=P[3],q=Xs(F,K,d),G=ev(_.shape,q,H,w),Z=Qg(_,q,H.shape,w);if(j!=null){var X=Js(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Vu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Vu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var tR=U({fusedDepthwiseConv2d_:eR});function nR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(Qs(z.state.gradientDepth,c)===!1){var d=We(e,i,a,o);return l!=null&&(d=fe(d,l)),Zs(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=at(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=ot(w),x=ot(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(on(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[x,b,g]):V(f,[x,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=at(k,p)[0],et(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=Xs(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=We(ie,X,!1,!0),re=We(Z,ie,!0,!1)):!a&&o?(te=We(ie,X,!1,!1),re=We(ie,Z,!0,!1)):a&&!o?(te=We(X,ie,!1,!0),re=We(Z,ie,!1,!1)):(te=We(X,ie,!0,!0),re=We(ie,Z,!0,!0)),l!=null){var le=Js(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var K=kn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,Mu,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return K(R,D)}else{var j=kn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,Mu,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return j(R,D,k)}}var iR=U({fusedMatMul_:nR});var rR={__proto__:null,conv2d:JC,depthwiseConv2d:tR,matMul:iR};function aR(n){return $c(n,.54,.46)}var sR=U({hammingWindow_:aR});function oR(n){return $c(n,.5,.5)}var tv=U({hannWindow_:oR});function lR(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a<n.size;){var o=a+t-n.size,l=xt([ze(n,a,t-o),Oc([o],r)]);s.push(l),a+=e}return s.length===0?Ea([],[0,t]):V(xt(s),[s.length,t])}var nv=U({frame_:lR});function uR(n,t,e,i,r){r===void 0&&(r=tv),i==null&&(i=Zg(t));for(var a=nv(n,t,e),s=J(a,r(t)),o=[],l=0;l<a.shape[0];l++)o.push(Gs(ze(s,[l,0],[1,t]),i));return xt(o)}var cR=U({stft_:uR});function hR(n,t,e,i,r,a){var s=O(n,"image","cropAndResize"),o=O(t,"boxes","cropAndResize","float32"),l=O(e,"boxInd","cropAndResize","int32");r=r||"bilinear",a=a||0;var u=o.shape[0];E(s.rank===4,function(){return"Error in cropAndResize: image must be rank 4,"+("but got rank "+s.rank+".")}),E(o.rank===2&&o.shape[1]===4,function(){return"Error in cropAndResize: boxes must be have size ["+u+",4] "+("but had shape "+o.shape+".")}),E(l.rank===1&&l.shape[0]===u,function(){return"Error in cropAndResize: boxInd must be have size ["+u+"] "+("but had shape "+o.shape+".")}),E(i.length===2,function(){return"Error in cropAndResize: cropSize must be of length 2, but got "+("length "+i.length+".")}),E(i[0]>=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Vp,d);return p}var dR=U({cropAndResize_:hR});function pR(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(ef,e,{});return i}var fR=U({flipLeftRight_:pR});function mR(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(Ef,a,s);return o}var gR=U({rotateWithOffset_:mR});function Pr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function vR(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Pr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,vf,l)}var yR=U({nonMaxSuppression_:vR});function wR(n,t,e){var i=bR(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function bR(n,t,e){return LR(n,t,e||SR)}function SR(n,t){return n>t?1:n<t?-1:0}function LR(n,t,e){for(var i=0,r=n.length,a=0,s=!1;i<r;){a=i+(r-i>>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function iv(n,t,e,i,r){return Jc(n,t,e,i,r,0).selectedIndices}function rv(n,t,e,i,r,a){return Jc(n,t,e,i,r,0,!1,a,!0)}function av(n,t,e,i,r,a){return Jc(n,t,e,i,r,a,!0)}function Jc(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;c<t.length;c++)t[c]>r&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(sv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length<e&&u.length>0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m<r)break;for(var b=!1,w=d.length-1;w>=v;--w){var S=IR(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*AR(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&wR(u,f,sv))}var L=d.length,x=e-L;o&&x>0&&(d.push.apply(d,new Array(x).fill(0)),p.push.apply(p,new Array(x).fill(0)));var C={selectedIndices:Ur(d,"int32")};return s&&(C.selectedScores=Ur(p,"float32")),l&&(C.validOutputs=we(L,"int32")),C}function IR(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function AR(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function sv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function TR(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Pr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=iv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var NR=TR;function xR(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(bf,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var CR=U({nonMaxSuppressionWithScore_:xR});function RR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=av(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var OR=RR;function ER(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(yf,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var DR=U({nonMaxSuppressionPadded_:ER});function kR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=rv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var FR=kR;function WR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,du,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var ov=U({resizeBilinear_:WR});function UR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,hu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var lv=U({resizeNearestNeighbor_:UR});function BR(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(Hc(0,s,1,"int32"),[-1,1]),u=Hc(0,o,1,"int32"),c=ge(l,u),h=qi(Mi(c,we(+t,"int32")),_i(c,we(-e,"int32"))),d=Kn([s,o],i.dtype);return V(ji(js(V(i,[-1,s,o])).map(function(p){return dn(h,p,d)})),r)}var zR=U({bandPart_:BR});function PR(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r<n.length;++r)i(r)}else t=!0,n=Br(n,n.shape[0],0).map(function(l){return Ks(l,[0])});E(n.length<=n[0].shape[0],function(){return"Gram-Schmidt: Number of vectors ("+n.length+") exceeds "+("number of dimensions ("+n[0].shape[0]+").")});for(var a=[],s=n,o=function(l){a.push(z.tidy(function(){var u=s[l];if(l>0)for(var c=0;c<l;++c){var h=J(Ae(J(a[c],u)),a[c]);u=ge(u,h)}return Ie(u,$s(u,"euclidean"))}))},r=0;r<n.length;++r)o(r);return t?ji(a,0):a}var _R=U({gramSchmidt_:PR});function MR(n,t){if(t===void 0&&(t=!1),E(n.rank>=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return uv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=js(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=uv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(ji(r,0),n.shape),o=V(ji(a,0),n.shape);return[s,o]}function uv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=ug(e),a=Bi(n),s=Ea([[1]],[1,1]),o=Bi(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=$s(g),b=ze(a,[h,h],[1,1]),w=dn(hi(b,0),Ea([[-1]]),Ea([[1]])),S=ge(b,J(w,v)),L=Ie(g,S);L.shape[0]===1?o=Bi(s):o=xt([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var x=gt(Ie(We(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(x,o),D=ut(o);if(h===0)a=ge(C,We(R,We(D,C)));else{var k=ge(C,We(R,We(D,C)));a=xt([ze(a,[0,0],[h,i]),k],0)}var W=ut(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,We(We(F,o),W));else{var P=ge(F,We(We(F,o),W));r=xt([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],zt([p,f,m])},c=0;c<l;++c)u(c);return!t&&e>i&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var HR=U({qr_:MR});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function VR(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ae(a);if(e===I.Reduction.MEAN){if(r==null)return xa(a);var s=i.size/r.size,o=Ie(Ae(a),Ae(r));return s>1?Ie(o,we(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Ie(Ae(a),we(i.size));var l=J(r,Fr(i.shape)),u=ue(Ae(Hs(l,we(0))),"float32");return Ie(Ae(a),u)}throw Error("Unknown reduction: "+e)}var Xn=U({computeWeightedLoss_:VR});function qR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Be(r.shape,a.shape,"Error in absoluteDifference: ");var o=qt(ge(r,a));return Xn(o,s,i)}var GR=U({absoluteDifference_:qR});function YR(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Be(a.shape,s.shape,"Error in cosineDistance: ");var l=we(1),u=ge(l,Ae(J(a,s),e,!0));return Xn(u,o,r)}var KR=U({cosineDistance_:YR});function jR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Be(r.shape,a.shape,"Error in hingeLoss: ");var o=we(1);r=ge(J(we(2),r),o);var l=Ra(ge(o,J(r,a)));return Xn(l,s,i)}var $R=U({hingeLoss_:jR});function XR(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Be(a.shape,s.shape,"Error in huberLoss: ");var l=we(i),u=qt(ge(s,a)),c=Ms(u,l),h=ge(u,c),d=fe(J(we(.5),Ye(c)),J(l,h));return Xn(d,o,r)}var JR=U({huberLoss_:XR});function ZR(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Be(a.shape,s.shape,"Error in logLoss: ");var l=we(1),u=we(i),c=gt(J(a,Hi(fe(s,u)))),h=J(ge(l,a),Hi(fe(ge(l,s),u))),d=ge(c,h);return Xn(d,o,r)}var QR=U({logLoss_:ZR});function eO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Be(r.shape,a.shape,"Error in meanSquaredError: ");var o=Ys(r,a);return Xn(o,s,i)}var tO=U({meanSquaredError_:eO});function nO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Be(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ra(i),a=J(i,e),s=Dc(pn(gt(qt(i))));return fe(ge(r,a),s)}function iO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Be(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=nO(a,s);return Xn(h,o,r)}var rO=U({sigmoidCrossEntropy_:iO});function aO(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=kn(function(r,a,s){var o=!0,l=Fc(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=gt(J(u,r)),h=Ae(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=Qt(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),pn(g))),J(V(p,v),ge(pn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function sO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Be(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(a.shape[1]);a=fe(J(a,ge(u,l)),Ie(l,c))}var h=aO(a,s);return Xn(h,o,r)}var oO=U({softmaxCrossEntropy_:sO});var lO={fft:qs,ifft:Oa,rfft:Gs,irfft:Kc},uO={hammingWindow:sR,hannWindow:tv,frame:nv,stft:cR},cO={flipLeftRight:fR,resizeNearestNeighbor:lv,resizeBilinear:ov,rotateWithOffset:gR,cropAndResize:dR,nonMaxSuppression:yR,nonMaxSuppressionAsync:NR,nonMaxSuppressionWithScore:CR,nonMaxSuppressionWithScoreAsync:OR,nonMaxSuppressionPadded:DR,nonMaxSuppressionPaddedAsync:FR},hO={bandPart:zR,gramSchmidt:_R,qr:HR},dO={absoluteDifference:GR,computeWeightedLoss:Xn,cosineDistance:KR,hingeLoss:$R,huberLoss:JR,logLoss:QR,meanSquaredError:tO,sigmoidCrossEntropy:rO,softmaxCrossEntropy:oO};var di=function(n){qn(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return zt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return gg(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&zt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:we(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(km);Object.defineProperty(di,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Zc=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;ft(function(){var d=fe(J(c,i.rho),J(Ye(u),1-i.rho)),p=J(Ie(_t(fe(h,i.epsilon)),_t(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ye(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(zt(this.accumulatedGrads.map(function(e){return e.variable})),zt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(di);ui(Zc);var Qc=function(n){qn(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:ft(function(){return Oc(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;ft(function(){var h=fe(c,Ye(u));c.assign(h);var d=fe(J(Ie(u,_t(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&zt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(di);ui(Qc);var eh=function(n){qn(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],ft(function(){s.accBeta1=we(i).variable(),s.accBeta2=we(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:ft(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:ft(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ye(h),1-i.beta2)),g=Ie(f,a),v=Ie(m,s);d.assign(f),p.assign(m);var b=fe(J(Ie(g,fe(_t(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&zt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),ft(function(){a.accBeta1.assign(jn(a.beta1,a.iterations_+1)),a.accBeta2.assign(jn(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(di);ui(eh);var th=function(n){qn(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],ft(function(){o.iteration=we(0).variable(),o.accBeta1=we(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=Ie(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=qt(h),v=kr(m,g);d.assign(f),p.assign(v);var b=fe(J(Ie(s,a),Ie(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&zt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(di);ui(th);var eo=function(n){qn(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];ft(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=Um(we(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(di);ui(eo);var nh=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=we(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;ft(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&zt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(eo);ui(nh);var ih=function(n){qn(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;ft(function(){var d=fe(J(c,i.decay),J(Ye(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Ie(J(u,i.learningRate),_t(ge(d,fe(Ye(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ye(u),1-i.decay)),g=fe(J(h,i.momentum),Ie(J(u,i.learningRate),_t(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&zt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&zt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&zt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(di);ui(ih);var $i=function(){function n(){}return n.sgd=function(t){return new eo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new nh(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new ih(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new eh(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Zc(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new th(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Qc(t,e)},n}();var pO={sgd:$i.sgd,momentum:$i.momentum,adadelta:$i.adadelta,adagrad:$i.adagrad,rmsprop:$i.rmsprop,adamax:$i.adamax,adam:$i.adam};var fO=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function mO(){return new Promise(function(n){return fO(function(){return n()})})}function gO(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function vO(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s<a;++s)r=r.concat([n[s+1]/t[s],t[s]]);r=r.concat(n.slice(a+1))}return r}function yO(n,t,e){e===void 0&&(e=!0);var i=[];if(e){i.push(t);for(var r=t+1;r<n;++r)r<=2*t?(i.push(r),i.push(r-(t+1))):i.push(r)}else{for(var a=[],s=[],r=1;r<n;++r)r>=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function bO(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a<n.length;++a)a<=t.length?i?r.push(t[a-1]*n[a]):r.push(n[a]/t[a-1]):r.push(n[a]);return r}function wO(n,t){for(var e=[0],i=0;i<t;++i)e.push(n[i][0]);return e}function SO(n,t,e){for(var i=n.slice(0,1),r=0;r<e;++r)i.push(n[r+1]-t[r][0]-t[r][1]);return i}var cv=1.7580993408473768,hv=1.0507009873554805;var LO=.3275911,IO=.254829592,AO=-.284496736,TO=1.421413741,NO=-1.453152027,xO=1.061405429;function CO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.warn.apply(console,n)}function RO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.log.apply(console,n)}function OO(n,t){if(n.length!==t.length)throw new Error("Cannot merge real and imag arrays of different lengths. real:"+(n.length+", imag: "+t.length+"."));for(var e=new Float32Array(n.length*2),i=0;i<e.length;i+=2)e[i]=n[i/2],e[i+1]=t[i/2];return e}function EO(n){for(var t=new Float32Array(n.length/2),e=new Float32Array(n.length/2),i=0;i<n.length;i+=2)t[i/2]=n[i],e[i/2]=n[i+1];return{real:t,imag:e}}function DO(n){for(var t=Math.ceil(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=0;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function kO(n){for(var t=Math.floor(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=2;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function FO(n,t){var e=n[t*2],i=n[t*2+1];return{real:e,imag:i}}function WO(n,t,e,i){n[i*2]=t,n[i*2+1]=e}function UO(n,t){for(var e=new Float32Array(n/2),i=new Float32Array(n/2),r=0;r<Math.ceil(n/2);r++){var a=(t?2:-2)*Math.PI*(r/n);e[r]=Math.cos(a),i[r]=Math.sin(a)}return{real:e,imag:i}}function BO(n,t,e){var i=(e?2:-2)*Math.PI*(n/t),r=Math.cos(i),a=Math.sin(i);return{real:r,imag:a}}function zO(n,t,e){if(t==="complex64"){if(n.dtype==="complex64")return n.clone();var i=Kn(n.shape),r=ue(n,"float32"),a=e.complex(r,i);return i.dispose(),r.dispose(),a}if(!_f(n.dtype,t))return z.makeTensorFromDataId(n.dataId,n.shape,t);if(n.dtype==="complex64"){var s=e.real(n),a=ue(s,t);return s.dispose(),a}if(t==="int32")return e.int(n);if(t==="bool"){var o=we(0,n.dtype),a=e.notEqual(n,o);return o.dispose(),a}else throw new Error("Error in Cast: failed to cast "+n.dtype+" to "+t)}function PO(n,t){return z.makeTensorFromDataId(n.dataId,t,n.dtype)}function _O(n,t,e){var i=(t-n)/(e-1),r=Lr(e,"float32");r[0]=n;for(var a=1;a<r.length;a++)r[a]=r[a-1]+i;return Ur(r,"float32")}var MO={__proto__:null,slice_util:Dm,segment_util:M1,castTensor:zO,reshapeTensor:PO,linspaceImpl:_O,upcastType:Ls,axesAreInnerMostDims:Lc,combineLocations:Pm,computeOutAndReduceShapes:_m,expandShapeToKeepDim:Qt,assertAxesAreInnerMostDims:vT,getAxesPermutation:en,getUndoAxesPermutation:xs,getInnerMostAxes:En,getBroadcastDims:A1,getReductionAxes:mt,assertAndGetBroadcastShape:et,assertParamsConsistent:Xm,computeOutShape:Jm,computeDilation2DInfo:xT,computePool2DInfo:Rr,computePool3DInfo:Ia,computeConv2DInfo:Dn,computeConv3DInfo:La,computeDefaultPad:Ac,tupleValuesAreOne:ci,eitherStridesOrDilationsAreOne:Pt,convertConv2DDataFormat:Sa,getFusedDyActivation:Xs,getFusedBiasGradient:Js,applyActivation:Zs,shouldFuse:Qs,PARALLELIZE_THRESHOLD:Ec,computeOptimalWindowSize:z1,getImageCenter:gO,getReshaped:vO,getPermuted:yO,getReshapedPermuted:bO,getSliceBeginCoords:wO,getSliceSize:SO,prepareAndValidate:bm,validateUpdateShape:mc,validateInput:gc,calculateShapes:wm,SELU_SCALEALPHA:cv,SELU_SCALE:hv,ERF_P:LO,ERF_A1:IO,ERF_A2:AO,ERF_A3:TO,ERF_A4:NO,ERF_A5:xO,warn:CO,log:RO,mergeRealAndImagArrays:OO,splitRealAndImagArrays:EO,complexWithEvenIndex:DO,complexWithOddIndex:kO,getComplexWithIndex:FO,assignToTypedArray:WO,exponents:UO,exponent:BO,prepareSplitSize:Rg};function HO(n,t,e){var i=new Array(n.rank).fill(0),r=n.shape.slice();return t.map(function(a){var s=r.slice();s[e]=a;var o=ze(n,i,s);return i[e]+=a,o})}function VO(n,t){for(var e=new Array(n.rank),i=0;i<e.length;i++)e[i]=n.shape[i]*t[i];for(var r=On(e,n.dtype),i=0;i<r.values.length;++i){for(var a=r.indexToLoc(i),s=new Array(n.rank),o=0;o<s.length;o++)s[o]=a[o]%n.shape[o];var l=n.locToIndex(s);r.values[i]=n.values[l]}return r.toTensor()}function qO(n,t,e,i,r){for(var a=t[t.length-1],s=[n.length/a,a],o=s[0],l=s[1],u=gs(e,o*i),c=gs("int32",o*i),h=0;h<o;h++){for(var d=h*l,p=n.subarray(d,d+l),f=[],m=0;m<p.length;m++)f.push({value:p[m],index:m});f.sort(function(S,L){return L.value-S.value});for(var g=h*i,v=u.subarray(g,g+i),b=c.subarray(g,g+i),m=0;m<i;m++)v[m]=f[m].value,b[m]=f[m].index}var w=t.slice();return w[w.length-1]=i,[si(u,w,e),si(c,w,"int32")]}var GO={__proto__:null,nonMaxSuppressionV3Impl:iv,nonMaxSuppressionV4Impl:rv,nonMaxSuppressionV5Impl:av,split:HO,tile:VO,topkImpl:qO,whereImpl:Fg};var YO={kernelName:il,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,zr(ue(e,"float32"),-1))}}}};var KO={kernelName:rl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Ye(ue(e,"float32")),r=_t(ge(we(1),i));return gt(Ie(n,r))}}}};var jO={kernelName:al,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(ge(Ye(ue(e,"float32")),1));return Ie(n,i)}}}};var $O={kernelName:ps,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,i.shape)};return{a,b:s}}};var XO={kernelName:sl,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var JO={kernelName:ol,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var ZO={kernelName:ll,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var QO={kernelName:ul,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,_t(ge(we(1),Ye(ue(e,"float32")))))}}}};var eE={kernelName:cl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(fe(we(1),Ye(ue(e,"float32"))));return Ie(n,i)}}}};var tE={kernelName:pl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=fe(Ye(e),Ye(i)),l=J(n,Ie(i,o)),u=mt(e.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,e.shape)},s=function(){var o=fe(Ye(e),Ye(i)),l=gt(J(n,Ie(e,o))),u=mt(i.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,i.shape)};return{a,b:s}}};var nE={kernelName:hl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(Ye(ue(e,"float32")),1))}}}};var iE={kernelName:dl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ge(we(1),Ye(ue(e,"float32"))))}}}};function rE(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(Pt(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(rt(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=Ia(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,zp,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var aE=U({avgPool3dBackprop_:rE});var sE={kernelName:ml,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return aE(n,i,a,s,c,l,u)}}}};function oE(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=Rr(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,Bp,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var lE=U({avgPoolBackprop_:oE});var uE={kernelName:fl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return lE(n,i,a,s,o)}}}};var cE={kernelName:gl,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return We(n,a,!1,!0)},b:function(){return We(r,n,!0,!1)}}:!o&&l?{a:function(){return We(n,a,!1,!1)},b:function(){return We(n,r,!0,!1)}}:o&&!l?{a:function(){return We(a,n,!1,!0)},b:function(){return We(r,n,!1,!1)}}:{a:function(){return We(a,n,!0,!0)},b:function(){return We(n,r,!0,!0)}}}};var hE={kernelName:vl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return Vs(n,r,a)}}}};var dE={kernelName:yl,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o<s.length;o++)s[o]>1&&l.push(o);return{x:function(){return Ae(n,l,!0)}}}};var pE={kernelName:fs,gradFunc:function(n){return{x:function(){return n.clone()}}}};var fE={kernelName:bl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var mE={kernelName:wl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return dn(qi(_i(i,a),Mi(i,s)),n,Ee(n))}}}};var gE={kernelName:Sl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Qe(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=Br(n,s,a);return o.map(function(l){return function(){return l}})}};var vE={kernelName:Ll,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(ci(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return Nc(r.shape,n,a,l,u,c)},filter:function(){return Xc(r,n,a.shape,l,u,c)}}}};var yE={kernelName:Il,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Er(n,a,o,l,u,1,c)},filter:function(){return Xc(n,r,a.shape,o,l,u,c)}}}};function bE(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=La(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,Mp,u)}var wE=U({conv3DBackpropFilter_:bE});var SE={kernelName:Al,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(ci(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return ng(o.shape,n,l,a,s)},filter:function(){return wE(o,n,l.shape,a,s)}}}};var LE={kernelName:Tl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(gt(Gc(ue(e,"float32"))),n)}}}};var IE={kernelName:Nl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Yc(ue(e,"float32")),n)}}}};var AE={kernelName:xl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=en([a],i.rank),u=Cc(n,a,s,!o);return l!=null&&(u=ut(u,l)),u}}}};var TE={kernelName:Cl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(ci(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(Pt(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be  1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(rt(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Dn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return ev(c.shape,n,h,d)},filter:function(){return Qg(c,n,h.shape,d)}}}};var NE={kernelName:Rl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(jp,s,e)},filter:function(){return z.runKernel($p,o,e)}}}};var xE={kernelName:Ol,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var CE={kernelName:El,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Xp)}}}};var RE={kernelName:Dl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(pn(gt(Ye(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var OE={kernelName:kl,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var EE={kernelName:Fl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,pn(e))}}}};var DE={kernelName:Wl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var kE={kernelName:Ul,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var FE={kernelName:Bl,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??we(1),u=mt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h<r.shape.length-1;++h)c.push(r.shape[h]);c.push(1)}var d=ge(r,a),p=J(n,l),f=qc(fe(s,we(i))),m=J(J(J(f,f),f),we(-.5)),g=function(){return a.rank===1?V(J(J(n,Dr(V(f,[1,1,1,a.shape[0]]),c)),l),r.shape):V(J(J(n,f),l),r.shape)},v=function(){var L=J(J(f,we(-1)),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},b=function(){var L=J(J(m,d),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},w=function(){var L=J(d,f),x=J(n,L);return a.rank===1&&(x=Ae(x,u)),V(x,a.shape)},S=function(){var L=n;return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)};return{x:g,mean:v,variance:b,scale:w,offset:S}}};var WE={kernelName:zl,inputsToSave:["x","indices"],gradFunc:function(n,t,e){var i=t[0],r=t[1],a=e.axis,s=Qe(a,i.shape)[0],o=function(){var l=i.shape,u=r.size,c=l.slice(0,s),h=c.length,d=l.slice(a,l.length).slice(1),p=d.length,f=dv(0,h),m=dv(h+1,h+1+p),g=pv([c,[u],d]),v=V(n,g),b=V(r,[u]),w=pv([[h],f,m]),S=ut(v,w),L=jc(S,b,i.shape[s]),x=xs(w);return L=ut(L,x),L};return{x:o,indices:function(){return r}}}};function dv(n,t){for(var e=[],i=n;i<t;++i)e.push(i);return e}function pv(n){for(var t=[],e=0;e<n.length;++e)for(var i=0;i<n[e].length;++i)t.push(n[e][i]);return t}var UE={kernelName:Pl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1];return{a:function(){return Ee(e)},b:function(){return Ee(i)}}}};var BE={kernelName:_l,gradFunc:function(n){return{x:function(){return ue(n,"float32")}}}};var zE={kernelName:Ml,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var PE={kernelName:Hl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var _E={kernelName:Vl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ME={kernelName:Gl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(e,1))}}}};var HE={kernelName:ql,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ue(e,"float32"))}}}};var VE={kernelName:Yl,inputsToSave:[],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.axis;return{logits:function(){var a=!0,s=pn(i);return ge(n,J(Ae(n,r,a),s))}}}};function qE(n,t,e,i,r,a,s){i===void 0&&(i=5),r===void 0&&(r=1),a===void 0&&(a=1),s===void 0&&(s=.5);var o=function(c){return c.LRNGrad(e,n,t,i,r,a,s)},l={x:n,y:t,dy:e},u={depthRadius:i,bias:r,alpha:a,beta:s};return z.runKernelFunc(o,l,null,df,u)}var GE=U({localResponseNormalizationBackprop_:qE});var YE={kernelName:Kl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.depthRadius,l=s.bias,u=s.alpha,c=s.beta;return{x:function(){return GE(r,a,n,o,l,u,c)}}}};function fv(n,t,e,i,r){return t.rank<e.rank&&(t=V(t,Qt(t.shape,i))),n.rank<e.rank&&(n=V(n,Qt(n.shape,i))),{x:function(){var a=J(n,ue(Na(e,t),n.dtype));return r==null?a:ut(a,r)}}}var mv={kernelName:jl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.reductionIndices,a=t[0],s=t[1],o=Qe(r,a.shape),l=en(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var KE={kernelName:$l,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(_i(e,i),"float32"))},a=function(){return J(n,ue(Bs(e,i),"float32"))};return{a:r,b:a}}};function jE(n,t,e,i,r,a,s,o){a===void 0&&(a=[1,1,1]);var l=O(n,"dy","maxPool3dBackprop"),u=O(t,"input","maxPool3dBackprop"),c=O(e,"output","maxPool3dBackprop"),h=l,d=u,p=c,f=!1;u.rank===4&&(f=!0,h=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=V(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),p=V(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),E(h.rank===5,function(){return"Error in maxPool3dBackprop: dy must be rank 5 but got rank "+(h.rank+".")}),E(d.rank===5,function(){return"Error in maxPool3dBackprop: input must be rank 5 but got rank "+(d.rank+".")}),E(p.rank===5,function(){return"Error in maxPool3dBackprop: output must be rank 5 but got rank "+(p.rank+".")}),E(Pt(r,a),function(){return"Error in maxPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+r+" and dilations '"+a+"'")}),o!=null&&E(rt(s),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+o+" but got pad "+s+".")});var m=function(w){var S=Ia(d.shape,i,r,a,s,o);return w.maxPool3dBackprop(h,d,p,S)},g={dy:h,input:d,output:p},v={filterSize:i,strides:r,dilations:a,pad:s,dimRoundingMode:o},b=z.runKernelFunc(m,g,null,ff,v);return f?V(b,[b.shape[1],b.shape[2],b.shape[3],b.shape[4]]):b}var $E=U({maxPool3dBackprop_:jE});var XE={kernelName:Jl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.dilations,c=s.pad,h=s.dimRoundingMode,d=u??[1,1,1];return{x:function(){return $E(n,r,a,o,l,d,c,h)}}}};function JE(n,t,e,i,r,a,s){var o=O(n,"dy","maxPoolBackprop"),l=O(t,"input","maxPoolBackprop"),u=O(e,"output","maxPoolBackprop");E(l.rank===o.rank,function(){return"Rank of input ("+l.rank+") does not match rank of dy "+("("+o.rank+")")}),E(o.rank===4,function(){return"Error in maxPoolBackprop: dy must be rank 4 but got rank "+(o.rank+".")}),E(l.rank===4,function(){return"Error in maxPoolBackprop: input must be rank 4 but got rank "+(l.rank+".")}),s!=null&&E(rt(a),function(){return"Error in maxPoolBackprop: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+a+".")});var c=function(p){var f=Rr(l.shape,i,r,1,a,s);return p.maxPoolBackprop(o,l,u,f)},h={dy:o,input:l,output:u},d={filterSize:i,strides:r,pad:a,dimRoundingMode:s};return z.runKernelFunc(c,h,null,pf,d)}var ZE=U({maxPoolBackprop_:JE});var QE={kernelName:Xl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.pad;return{x:function(){return ZE(n,r,a,o,l,u)}}}};var eD={kernelName:Zl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.axis,a=t[0],s=t[1],o=Qe(r,a.shape),l=en(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var tD={kernelName:Ql,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Mi(e,i),"float32"))},a=function(){return J(n,ue(hi(e,i),"float32"))};return{a:r,b:a}}};var nD={kernelName:eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=mt(e.shape,r);return o.length>0?V(Ae(n,o),e.shape):n},s=function(){var o=J(n,gt(Fs(Ie(e,i)))),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var iD={kernelName:tu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var rD={kernelName:nu,gradFunc:function(n){return{x:function(){return gt(n)}}}};var aD={kernelName:ru,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return Kn(e.shape,"float32")}}}};var sD={kernelName:iu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var gv={kernelName:au,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var oD={kernelName:su,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=et(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,jn(a,ge(c,we(1))))),d=mt(a.shape,o);return d.length>0&&(h=Ae(h,d)),V(h,a.shape)},u=function(){var c=hi(a,0),h=dn(c,Hi(a),Ee(a)),d=J(n,J(r,h)),p=mt(s.shape,o);return p.length>0&&(d=Ae(d,p)),V(d,s.shape)};return{a:l,b:u}}};var lD={kernelName:ou,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=hi(e,0);return{x:function(){return dn(r,n,J(n,i))},alpha:function(){var a=dn(r,Ee(n),J(n,e)),s=mt(i.shape,n.shape);return s.length>0&&(a=Ae(a,s)),V(a,i.shape)}}}};var uD={kernelName:lu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,gt(Ye(e)))}}}};var cD={kernelName:pu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(Mi(e,6),zr(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var hD={kernelName:uu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue(zr(e),"float32"))}}}};var dD={kernelName:cu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var pD={kernelName:du,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,Af,e)};return{images:s}}};var fD={kernelName:hu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,If,e)};return{images:s}}};var mD={kernelName:fu,gradFunc:function(n,t,e){var i=e.dims,r=Qe(i,n.shape);return{x:function(){return $n(n,r)}}}};var gD={kernelName:mu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var vD={kernelName:gu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return gt(Ie(n,J(jn(e,1.5),2)))}}}};var yD={kernelName:vu,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(Ps(e),n.dtype))}}}};var bD={kernelName:yu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=hi(e,we(0)),r=we(cv),a=we(hv),s=J(n,a),o=J(J(n,r),pn(ue(e,"float32")));return dn(i,s,o)}}}};var wD={kernelName:Iu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(we(1),e)))}}}};var SD={kernelName:Lu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var LD={kernelName:wu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ks(ue(e,"float32")),n)}}}};var ID={kernelName:Su,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(xc(ue(e,"float32")),n)}}}};var AD={kernelName:bu,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=vc(i,a,s),u=l[0],c=l[1],h=[],d=0;d<n.rank;d++)h.push([u[d],o[d]-u[d]-c[d]]);return{x:function(){return Gi(n,h)}}}};var TD={kernelName:Ru,outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.dim,a=!0,s=J(n,i);return{logits:function(){return ge(s,J(Ae(s,[r],a),i))}}}};var ND={kernelName:Au,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Pi(e))}}}};var vv={kernelName:xu,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.paddings;return{x:function(){return Os(n,r,a)}}}};var yv={kernelName:Cu,gradFunc:function(n,t,e){var i=e.axis;return{x:function(){return xt(n,i)}}}};var xD={kernelName:Tu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,J(_t(ue(e,"float32")),2))}}}};var CD={kernelName:Nf,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(ue(e,"float32"),2))}}}};var RD={kernelName:Ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=we(2),a=function(){return J(n,J(r,ge(e,i)))},s=function(){return J(n,J(r,ge(i,e)))};return{a,b:s}}};var OD={kernelName:Pu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ED={kernelName:Eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(gt(o),i.shape)};return{a,b:s}}};var DD={kernelName:Nu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Qe(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Fr(i.shape,"float32"));return{x:function(){return l}}}};var kD={kernelName:Du,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Ye(ks(e)))}}}};var FD={kernelName:ku,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(we(1),Ye(e)),n)}}}};var WD={kernelName:Fu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o<r[0];++o)s=fe(s,ze(n,[o*i.shape[0]],[i.shape[0]]));else if(i.rank===2)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1]],[i.shape[0],i.shape[1]]));else if(i.rank===3)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2]],[i.shape[0],i.shape[1],i.shape[2]]));else if(i.rank===4)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)for(var c=0;c<r[3];++c)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2],c*i.shape[3]],[i.shape[0],i.shape[1],i.shape[2],i.shape[3]]));else throw new Error("Gradient for tile operation is not implemented for rank-"+(i.rank+" tensors yet."));return s};return{x:a}}};var UD={kernelName:Wu,gradFunc:function(n,t,e){var i=e,r=i.perm,a=xs(r);return{x:function(){return ut(n,a)}}}};var BD={kernelName:Uu,gradFunc:function(n,t,e){var i=e,r=i.axis;return{value:function(){return ji(n,r)}}}};var PD={kernelName:Bu,inputsToSave:["segmentIds"],gradFunc:function(n,t){var e=t[0],i=function(){return zD(n,e)};return{x:i}}};function zD(n,t){for(var e=kr(t,Ee(t)),i=Ws(n,e),r=_i(t,we(0,"int32")),a=i.rank-r.rank,s=0;s<a;++s)r=fn(r,s+1);r=qi(r,Fr(i.shape,"bool"));var o=Ee(i);return dn(r,i,o)}var _D={kernelName:zu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var MD=[YO,KO,jO,$O,XO,JO,ZO,QO,eE,tE,nE,iE,sE,uE,cE,hE,dE,pE,fE,mE,gE,yE,vE,SE,LE,IE,AE,TE,NE,xE,CE,RE,OE,EE,kE,DE,FE,WE,UE,BE,zE,PE,_E,ME,HE,VE,YE,mv,mv,KE,XE,QE,eD,tD,nD,iD,rD,aD,sD,gv,gv,oD,lD,uD,cD,hD,dD,pD,fD,mD,gD,vD,yD,bD,wD,SD,LD,ID,AD,TD,ND,vv,vv,yv,yv,xD,RD,CD,OD,ED,DD,kD,FD,WD,UD,BD,PD,_D];for(var rh=0,bv=MD;rh<bv.length;rh++){var HD=bv[rh];kf(HD)}Y.prototype.abs=function(){return this.throwIfDisposed(),qt(this)};Y.prototype.acos=function(){return this.throwIfDisposed(),Bm(this)};Y.prototype.acosh=function(){return this.throwIfDisposed(),zm(this)};Y.prototype.addStrict=function(n){return this.throwIfDisposed(),Hg(this,n)};Y.prototype.add=function(n){return this.throwIfDisposed(),fe(this,n)};Y.prototype.all=function(n,t){return this.throwIfDisposed(),Mm(this,n,t)};Y.prototype.any=function(n,t){return this.throwIfDisposed(),Hm(this,n,t)};Y.prototype.argMax=function(n){return this.throwIfDisposed(),Vm(this,n)};Y.prototype.argMin=function(n){return this.throwIfDisposed(),qm(this,n)};Y.prototype.asScalar=function(){return this.throwIfDisposed(),E(this.size===1,function(){return"The array must have only 1 element."}),V(this,[])};Y.prototype.asType=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.as1D=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.as2D=function(n,t){return this.throwIfDisposed(),V(this,[n,t])};Y.prototype.as3D=function(n,t,e){return this.throwIfDisposed(),V(this,[n,t,e])};Y.prototype.as4D=function(n,t,e,i){return this.throwIfDisposed(),V(this,[n,t,e,i])};Y.prototype.as5D=function(n,t,e,i,r){return this.throwIfDisposed(),V(this,[n,t,e,i,r])};Y.prototype.asin=function(){return this.throwIfDisposed(),Gm(this)};Y.prototype.asinh=function(){return this.throwIfDisposed(),Ym(this)};Y.prototype.atan=function(){return this.throwIfDisposed(),Km(this)};Y.prototype.atan2=function(n){return this.throwIfDisposed(),jm(this,n)};Y.prototype.atanh=function(){return this.throwIfDisposed(),$m(this)};Y.prototype.avgPool=function(n,t,e,i){return this.throwIfDisposed(),Tc(this,n,t,e,i)};Y.prototype.batchToSpaceND=function(n,t){return this.throwIfDisposed(),Os(this,n,t)};Y.prototype.batchNorm=function(n,t,e,i,r){return this.throwIfDisposed(),Aa(this,n,t,e,i,r)};Y.prototype.broadcastTo=function(n){return this.throwIfDisposed(),Ds(this,n)};Y.prototype.cast=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.ceil=function(){return this.throwIfDisposed(),Zm(this)};Y.prototype.clipByValue=function(n,t){return this.throwIfDisposed(),Qm(this,n,t)};Y.prototype.concat=function(n,t){return this.throwIfDisposed(),n instanceof Y&&(n=[n]),xt([this].concat(n),t)};Y.prototype.conv1d=function(n,t,e,i,r,a){return this.throwIfDisposed(),eg(this,n,t,e,i,r,a)};Y.prototype.conv2dTranspose=function(n,t,e,i,r){return this.throwIfDisposed(),tg(this,n,t,e,i,r)};Y.prototype.conv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Er(this,n,t,e,i,r,a)};Y.prototype.cos=function(){return this.throwIfDisposed(),ks(this)};Y.prototype.cosh=function(){return this.throwIfDisposed(),xc(this)};Y.prototype.cumsum=function(n,t,e){return this.throwIfDisposed(),Cc(this,n,t,e)};Y.prototype.depthToSpace=function(n,t){return this.throwIfDisposed(),ig(this,n,t)};Y.prototype.depthwiseConv2D=function(n,t,e,i,r,a){return It("depthwiseConv2D is deprecated, use depthwiseConv2d instead"),this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.depthwiseConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.dilation2d=function(n,t,e,i,r){return this.throwIfDisposed(),rg(this,n,t,e,i,r)};Y.prototype.divNoNan=function(n){return this.throwIfDisposed(),ag(this,n)};Y.prototype.divStrict=function(n){return this.throwIfDisposed(),Vg(this,n)};Y.prototype.div=function(n){return this.throwIfDisposed(),Ie(this,n)};Y.prototype.dot=function(n){return this.throwIfDisposed(),sg(this,n)};Y.prototype.elu=function(){return this.throwIfDisposed(),Rc(this)};Y.prototype.equalStrict=function(n){return this.throwIfDisposed(),Ug(this,n)};Y.prototype.equal=function(n){return this.throwIfDisposed(),Na(this,n)};Y.prototype.erf=function(){return this.throwIfDisposed(),og(this)};Y.prototype.exp=function(){return this.throwIfDisposed(),pn(this)};Y.prototype.expandDims=function(n){return this.throwIfDisposed(),fn(this,n)};Y.prototype.expm1=function(){return this.throwIfDisposed(),lg(this)};Y.prototype.fft=function(){return this.throwIfDisposed(),qs(this)};Y.prototype.flatten=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.floor=function(){return this.throwIfDisposed(),Fs(this)};Y.prototype.floorDiv=function(n){return this.throwIfDisposed(),Sc(this,n)};Y.prototype.gather=function(n,t){return this.throwIfDisposed(),Ws(this,n,t)};Y.prototype.greaterEqualStrict=function(n){return this.throwIfDisposed(),Bg(this,n)};Y.prototype.greaterEqual=function(n){return this.throwIfDisposed(),_i(this,n)};Y.prototype.greaterStrict=function(n){return this.throwIfDisposed(),zg(this,n)};Y.prototype.greater=function(n){return this.throwIfDisposed(),hi(this,n)};Y.prototype.ifft=function(){return this.throwIfDisposed(),Oa(this)};Y.prototype.irfft=function(){return this.throwIfDisposed(),Kc(this)};Y.prototype.isFinite=function(){return this.throwIfDisposed(),hg(this)};Y.prototype.isInf=function(){return this.throwIfDisposed(),dg(this)};Y.prototype.isNaN=function(){return this.throwIfDisposed(),pg(this)};Y.prototype.leakyRelu=function(n){return this.throwIfDisposed(),fg(this,n)};Y.prototype.lessEqualStrict=function(n){return this.throwIfDisposed(),Pg(this,n)};Y.prototype.lessEqual=function(n){return this.throwIfDisposed(),Mi(this,n)};Y.prototype.lessStrict=function(n){return this.throwIfDisposed(),_g(this,n)};Y.prototype.less=function(n){return this.throwIfDisposed(),Bs(this,n)};Y.prototype.localResponseNormalization=function(n,t,e,i){return this.throwIfDisposed(),mg(this,n,t,e,i)};Y.prototype.logSigmoid=function(){return this.throwIfDisposed(),vg(this)};Y.prototype.logSoftmax=function(n){return this.throwIfDisposed(),yg(this,n)};Y.prototype.logSumExp=function(n,t){return this.throwIfDisposed(),Fc(this,n,t)};Y.prototype.log=function(){return this.throwIfDisposed(),Hi(this)};Y.prototype.log1p=function(){return this.throwIfDisposed(),Dc(this)};Y.prototype.logicalAnd=function(n){return this.throwIfDisposed(),qi(this,n)};Y.prototype.logicalNot=function(){return this.throwIfDisposed(),Ps(this)};Y.prototype.logicalOr=function(n){return this.throwIfDisposed(),Wc(this,n)};Y.prototype.logicalXor=function(n){return this.throwIfDisposed(),bg(this,n)};Y.prototype.matMul=function(n,t,e){return this.throwIfDisposed(),We(this,n,t,e)};Y.prototype.maxPool=function(n,t,e,i){return this.throwIfDisposed(),Uc(this,n,t,e,i)};Y.prototype.max=function(n,t){return this.throwIfDisposed(),Vi(this,n,t)};Y.prototype.maximumStrict=function(n){return this.throwIfDisposed(),qg(this,n)};Y.prototype.maximum=function(n){return this.throwIfDisposed(),kr(this,n)};Y.prototype.mean=function(n,t){return this.throwIfDisposed(),xa(this,n,t)};Y.prototype.min=function(n,t){return this.throwIfDisposed(),_s(this,n,t)};Y.prototype.minimumStrict=function(n){return this.throwIfDisposed(),Gg(this,n)};Y.prototype.minimum=function(n){return this.throwIfDisposed(),Ms(this,n)};Y.prototype.modStrict=function(n){return this.throwIfDisposed(),Yg(this,n)};Y.prototype.mod=function(n){return this.throwIfDisposed(),Bc(this,n)};Y.prototype.mulStrict=function(n){return this.throwIfDisposed(),Kg(this,n)};Y.prototype.mul=function(n){return this.throwIfDisposed(),J(this,n)};Y.prototype.neg=function(){return this.throwIfDisposed(),gt(this)};Y.prototype.norm=function(n,t,e){return this.throwIfDisposed(),$s(this,n,t,e)};Y.prototype.notEqualStrict=function(n){return this.throwIfDisposed(),Mg(this,n)};Y.prototype.notEqual=function(n){return this.throwIfDisposed(),Hs(this,n)};Y.prototype.oneHot=function(n,t,e){return t===void 0&&(t=1),e===void 0&&(e=0),this.throwIfDisposed(),Ts(this,n,t,e)};Y.prototype.onesLike=function(){return this.throwIfDisposed(),zc(this)};Y.prototype.pad=function(n,t){return this.throwIfDisposed(),Gi(this,n,t)};Y.prototype.pool=function(n,t,e,i,r){return this.throwIfDisposed(),wg(this,n,t,e,i,r)};Y.prototype.powStrict=function(n){return this.throwIfDisposed(),jg(this,n)};Y.prototype.pow=function(n){return this.throwIfDisposed(),jn(this,n)};Y.prototype.prelu=function(n){return this.throwIfDisposed(),Pc(this,n)};Y.prototype.prod=function(n,t){return this.throwIfDisposed(),Sg(this,n,t)};Y.prototype.reciprocal=function(){return this.throwIfDisposed(),Ig(this)};Y.prototype.relu=function(){return this.throwIfDisposed(),Ra(this)};Y.prototype.relu6=function(){return this.throwIfDisposed(),Vc(this)};Y.prototype.reshapeAs=function(n){return this.throwIfDisposed(),V(this,n.shape)};Y.prototype.reshape=function(n){return this.throwIfDisposed(),V(this,n)};Y.prototype.resizeBilinear=function(n,t){return this.throwIfDisposed(),ov(this,n,t)};Y.prototype.resizeNearestNeighbor=function(n,t){return this.throwIfDisposed(),lv(this,n,t)};Y.prototype.reverse=function(n){return this.throwIfDisposed(),$n(this,n)};Y.prototype.rfft=function(){return this.throwIfDisposed(),Gs(this)};Y.prototype.round=function(){return this.throwIfDisposed(),Ag(this)};Y.prototype.rsqrt=function(){return this.throwIfDisposed(),qc(this)};Y.prototype.selu=function(){return this.throwIfDisposed(),Tg(this)};Y.prototype.separableConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ng(this,n,t,e,i,r,a)};Y.prototype.sigmoid=function(){return this.throwIfDisposed(),Pi(this)};Y.prototype.sign=function(){return this.throwIfDisposed(),xg(this)};Y.prototype.sin=function(){return this.throwIfDisposed(),Gc(this)};Y.prototype.sinh=function(){return this.throwIfDisposed(),Yc(this)};Y.prototype.slice=function(n,t){return this.throwIfDisposed(),ze(this,n,t)};Y.prototype.softmax=function(n){return this.throwIfDisposed(),Cg(this,n)};Y.prototype.softplus=function(){return this.throwIfDisposed(),kc(this)};Y.prototype.spaceToBatchND=function(n,t){return this.throwIfDisposed(),Vs(this,n,t)};Y.prototype.split=function(n,t){return this.throwIfDisposed(),Br(this,n,t)};Y.prototype.sqrt=function(){return this.throwIfDisposed(),_t(this)};Y.prototype.square=function(){return this.throwIfDisposed(),Ye(this)};Y.prototype.squaredDifference=function(n){return this.throwIfDisposed(),Ys(this,n)};Y.prototype.squaredDifferenceStrict=function(n){return this.throwIfDisposed(),$g(this,n)};Y.prototype.squeeze=function(n){return this.throwIfDisposed(),Ks(this,n)};Y.prototype.stack=function(n,t){this.throwIfDisposed();var e=n instanceof Y?[this,n]:[this].concat(n);return ji(e,t)};Y.prototype.step=function(n){return this.throwIfDisposed(),zr(this,n)};Y.prototype.stridedSlice=function(n,t,e,i,r,a,s,o){return this.throwIfDisposed(),Og(this,n,t,e,i,r,a,s,o)};Y.prototype.subStrict=function(n){return this.throwIfDisposed(),Xg(this,n)};Y.prototype.sub=function(n){return this.throwIfDisposed(),ge(this,n)};Y.prototype.sum=function(n,t){return this.throwIfDisposed(),Ae(this,n,t)};Y.prototype.tan=function(){return this.throwIfDisposed(),Eg(this)};Y.prototype.tanh=function(){return this.throwIfDisposed(),Rs(this)};Y.prototype.tile=function(n){return this.throwIfDisposed(),Dr(this,n)};Y.prototype.toBool=function(){return this.throwIfDisposed(),ue(this,"bool")};Y.prototype.toFloat=function(){return this.throwIfDisposed(),ue(this,"float32")};Y.prototype.toInt=function(){return this.throwIfDisposed(),ue(this,"int32")};Y.prototype.topk=function(n,t){return this.throwIfDisposed(),Dg(this,n,t)};Y.prototype.transpose=function(n){return this.throwIfDisposed(),ut(this,n)};Y.prototype.unique=function(n){return this.throwIfDisposed(),kg(this,n)};Y.prototype.unsortedSegmentSum=function(n,t){return this.throwIfDisposed(),jc(this,n,t)};Y.prototype.unstack=function(n){return this.throwIfDisposed(),js(this,n)};Y.prototype.where=function(n,t){return this.throwIfDisposed(),dn(n,this,t)};Y.prototype.zerosLike=function(){return this.throwIfDisposed(),Ee(this)};I.Abs=il;I.Acos=rl;I.Acosh=al;I.AdadeltaOptimizer=Zc;I.AdagradOptimizer=Qc;I.AdamOptimizer=eh;I.AdamaxOptimizer=th;I.Add=ps;I.AddN=sl;I.All=Wp;I.Any=Up;I.ArgMax=ol;I.ArgMin=ll;I.Asin=ul;I.Asinh=cl;I.Atan=hl;I.Atan2=pl;I.Atanh=dl;I.AvgPool=fl;I.AvgPool3D=ml;I.AvgPool3DBackprop=zp;I.AvgPoolBackprop=Bp;I.BatchMatMul=gl;I.BatchToSpaceND=vl;I.BroadcastTo=yl;I.Cast=fs;I.Ceil=bl;I.ClipByValue=wl;I.Complex=Pp;I.Concat=Sl;I.Conv2D=Ll;I.Conv2DBackpropFilter=_p;I.Conv2DBackpropInput=Il;I.Conv3D=Al;I.Conv3DBackpropFilterV2=Mp;I.Conv3DBackpropInputV2=Hp;I.Cos=Tl;I.Cosh=Nl;I.CropAndResize=Vp;I.Cumsum=xl;I.DataStorage=EI;I.DepthToSpace=qp;I.DepthwiseConv2dNative=Cl;I.DepthwiseConv2dNativeBackpropFilter=Gp;I.DepthwiseConv2dNativeBackpropInput=Yp;I.Diag=Kp;I.Dilation2D=Rl;I.Dilation2DBackpropFilter=$p;I.Dilation2DBackpropInput=jp;I.Div=Ol;I.Elu=El;I.EluGrad=Xp;I.Environment=Dp;I.Equal=Jp;I.Erf=Dl;I.Exp=kl;I.Expm1=Fl;I.FFT=Zp;I.Fill=Qp;I.FlipLeftRight=ef;I.Floor=Wl;I.FloorDiv=Ul;I.FromPixels=_u;I.FusedBatchNorm=Bl;I.FusedConv2D=Hu;I.FusedDepthwiseConv2D=Vu;I.GatherNd=tf;I.GatherV2=zl;I.Greater=nf;I.GreaterEqual=Pl;I.IFFT=rf;I.Identity=_l;I.Imag=af;I.IsFinite=Ml;I.IsInf=Hl;I.IsNan=Vl;I.KernelBackend=Op;I.LRN=Kl;I.LRNBackprop=df;I.Less=sf;I.LessEqual=of;I.LinSpace=lf;I.Log=ql;I.Log1p=Gl;I.LogSoftmax=Yl;I.LogicalAnd=uf;I.LogicalNot=cf;I.LogicalOr=hf;I.Max=jl;I.MaxPool=Xl;I.MaxPool3D=Jl;I.MaxPool3DBackprop=ff;I.MaxPoolBackprop=pf;I.MaxPoolWithArgmax=mf;I.Maximum=$l;I.Mean=BI;I.Min=Zl;I.Minimum=Ql;I.Mod=eu;I.MomentumOptimizer=nh;I.Multiply=tu;I.Negate=nu;I.NonMaxSuppressionV3=vf;I.NonMaxSuppressionV4=yf;I.NonMaxSuppressionV5=bf;I.NotEqual=gf;I.OP_SCOPE_SUFFIX=tm;I.OneHot=ru;I.OnesLike=iu;I.Optimizer=di;I.PadV2=au;I.Pool=zI;I.Pow=su;I.Prelu=ou;I.Prod=wf;I.RMSPropOptimizer=ih;I.Range=Sf;I.Real=Lf;I.Reciprocal=lu;I.Relu=uu;I.Relu6=pu;I.Reshape=cu;I.ResizeBilinear=du;I.ResizeBilinearGrad=Af;I.ResizeNearestNeighbor=hu;I.ResizeNearestNeighborGrad=If;I.Reverse=fu;I.RotateWithOffset=Ef;I.Round=mu;I.Rsqrt=gu;I.SGDOptimizer=eo;I.ScatterNd=Tf;I.SelectV2=vu;I.Selu=yu;I.Sigmoid=Iu;I.Sign=Lu;I.Sin=wu;I.Sinh=Su;I.Slice=bu;I.Softmax=Ru;I.Softplus=Au;I.SpaceToBatchND=xu;I.SparseToDense=xf;I.SplitV=Cu;I.Sqrt=Tu;I.Square=Nf;I.SquaredDifference=Ou;I.Step=Pu;I.StridedSlice=Cf;I.Sub=Eu;I.Sum=Nu;I.Tan=Du;I.Tanh=ku;I.Tensor=Y;I.TensorBuffer=Ss;I.Tile=Fu;I.TopK=Rf;I.Transpose=Wu;I.Unique=Of;I.Unpack=Uu;I.UnsortedSegmentSum=Bu;I.Variable=ya;I.ZerosLike=zu;I._FusedMatMul=Mu;I.abs=qt;I.acos=Bm;I.acosh=zm;I.add=fe;I.addN=gT;I.addStrict=Hg;I.all=Mm;I.any=Hm;I.argMax=Vm;I.argMin=qm;I.asin=Gm;I.asinh=Ym;I.atan=Km;I.atan2=jm;I.atanh=$m;I.avgPool=Tc;I.avgPool3d=FT;I.backend=sT;I.backend_util=MO;I.basicLSTMCell=_T;I.batchNorm=Aa;I.batchNorm2d=GT;I.batchNorm3d=KT;I.batchNorm4d=$T;I.batchToSpaceND=Os;I.booleanMaskAsync=yC;I.broadcastTo=Ds;I.browser=OA;I.buffer=On;I.cast=ue;I.ceil=Zm;I.clipByValue=Qm;I.clone=Bi;I.complex=ri;I.concat=xt;I.concat1d=e1;I.concat2d=n1;I.concat3d=r1;I.concat4d=s1;I.conv1d=eg;I.conv2d=Er;I.conv2dTranspose=tg;I.conv3d=d1;I.conv3dTranspose=m1;I.copyRegisteredKernels=MI;I.cos=ks;I.cosh=xc;I.cosineWindow=$c;I.cumsum=Cc;I.customGrad=kn;I.deprecationWarn=It;I.depthToSpace=ig;I.depthwiseConv2d=Ta;I.device_util=I2;I.diag=L1;I.dilation2d=rg;I.disableDeprecationWarnings=KA;I.dispose=zt;I.disposeVariables=jA;I.div=Ie;I.divNoNan=ag;I.divStrict=Vg;I.dot=sg;I.dropout=YC;I.elu=Rc;I.enableDebugMode=YA;I.enableProdMode=GA;I.enclosingPowerOfTwo=Zg;I.engine=$A;I.env=He;I.equal=Na;I.equalStrict=Ug;I.erf=og;I.exp=pn;I.expandDims=fn;I.expm1=lg;I.eye=ug;I.fft=qs;I.fill=Oc;I.findBackend=iT;I.findBackendFactory=rT;I.floor=Fs;I.floorDiv=Sc;I.fused=rR;I.gather=Ws;I.gatherND=VC;I.gather_util=EA;I.getBackend=tT;I.getGradient=Yu;I.getKernel=Gu;I.getKernelsForBackend=ms;I.grad=iN;I.grads=rN;I.greater=hi;I.greaterEqual=_i;I.greaterEqualStrict=Bg;I.greaterStrict=zg;I.ifft=Oa;I.imag=Us;I.image=cO;I.inTopKAsync=jC;I.io=bA;I.irfft=Kc;I.isFinite=hg;I.isInf=dg;I.isNaN=pg;I.keep=Um;I.kernel_impls=GO;I.leakyRelu=fg;I.less=Bs;I.lessEqual=Mi;I.lessEqualStrict=Pg;I.lessStrict=_g;I.linalg=hO;I.linspace=Q1;I.localResponseNormalization=mg;I.log=Hi;I.log1p=Dc;I.logSigmoid=vg;I.logSoftmax=yg;I.logSumExp=Fc;I.logicalAnd=qi;I.logicalNot=Ps;I.logicalOr=Wc;I.logicalXor=bg;I.losses=dO;I.matMul=We;I.math=NA;I.max=Vi;I.maxPool=Uc;I.maxPool3d=SN;I.maxPoolWithArgmax=IN;I.maximum=kr;I.maximumStrict=qg;I.mean=xa;I.memory=XA;I.min=_s;I.minimum=Ms;I.minimumStrict=Gg;I.mod=Bc;I.modStrict=Yg;I.moments=ON;I.movingAverage=UC;I.mul=J;I.mulStrict=Kg;I.multiRNNCell=DN;I.multinomial=FN;I.neg=gt;I.nextFrame=mO;I.norm=$s;I.notEqual=Hs;I.notEqualStrict=Mg;I.oneHot=Ts;I.ones=Fr;I.onesLike=zc;I.op=U;I.outerProduct=PN;I.pad=Gi;I.pad1d=HN;I.pad2d=qN;I.pad3d=YN;I.pad4d=jN;I.pool=wg;I.pow=jn;I.powStrict=jg;I.prelu=Pc;I.print=hm;I.prod=Sg;I.profile=JA;I.rand=ix;I.randomGamma=fx;I.randomNormal=gx;I.randomUniform=Lg;I.range=Hc;I.ready=eT;I.real=Ca;I.reciprocal=Ig;I.registerBackend=aT;I.registerGradient=kf;I.registerKernel=Df;I.relu=Ra;I.relu6=Vc;I.removeBackend=nT;I.reshape=V;I.reverse=$n;I.reverse1d=Ix;I.reverse2d=Tx;I.reverse3d=xx;I.reverse4d=Rx;I.rfft=Gs;I.round=Ag;I.rsqrt=qc;I.scalar=we;I.scatterND=zC;I.scatter_util=DA;I.selu=Tg;I.separableConv2d=Ng;I.serialization=WA;I.setBackend=QA;I.setPlatform=oT;I.setdiff1dAsync=Wx;I.sigmoid=Pi;I.sign=xg;I.signal=uO;I.sin=Gc;I.sinh=Yc;I.slice=ze;I.slice1d=_x;I.slice2d=Hx;I.slice3d=qx;I.slice4d=Yx;I.slice_util=Dm;I.softmax=Cg;I.softplus=kc;I.spaceToBatchND=Vs;I.sparseToDense=MC;I.spectral=lO;I.split=Br;I.sqrt=_t;I.square=Ye;I.squaredDifference=Ys;I.squaredDifferenceStrict=$g;I.squeeze=Ks;I.stack=ji;I.step=zr;I.stridedSlice=Og;I.sub=ge;I.subStrict=Xg;I.sum=Ae;I.sumOutType=f2;I.tan=Eg;I.tanh=Rs;I.tensor=si;I.tensor1d=Ur;I.tensor2d=Ea;I.tensor3d=ym;I.tensor4d=sC;I.tensor5d=oC;I.tensor6d=lC;I.tensor_util=v2;I.test_util=VA;I.tidy=ft;I.tile=Dr;I.time=ZA;I.topk=Dg;I.train=pO;I.transpose=ut;I.truncatedNormal=hC;I.unique=kg;I.unregisterGradient=_I;I.unregisterKernel=PI;I.unsortedSegmentSum=jc;I.unstack=js;I.upcastType=Ls;I.util=i2;I.valueAndGrad=aN;I.valueAndGrads=sN;I.variable=mC;I.variableGrads=gg;I.version_core=qA;I.where=dn;I.whereAsync=Wg;I.zeros=Kn;I.zerosLike=Ee});var Yb=be(Xe=>{"use strict";Object.defineProperty(Xe,"__esModule",{value:!0});var y=Xi();var ah=function(n,t){return ah=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},ah(n,t)};function Q(n,t){ah(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var Gt=function(){return Gt=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t},Gt.apply(this,arguments)};function VD(n,t){var e={};for(var i in n)Object.prototype.hasOwnProperty.call(n,i)&&t.indexOf(i)<0&&(e[i]=n[i]);if(n!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,i=Object.getOwnPropertySymbols(n);r<i.length;r++)t.indexOf(i[r])<0&&(e[i[r]]=n[i[r]]);return e}function Le(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function ve(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var sh;function vt(){return sh==null&&(sh=y.backend().epsilon()),sh}function mn(){return"channelsLast"}var pi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Fn=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),M=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Te=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),qD=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),SH=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error);function Ji(n,t){if(Array.isArray(n)){for(var e=[],i=0;i<t;i++)e=e.concat(n);return e}else{var e=new Array(t);return e.fill(n),e}}function Wn(n,t){if(!n)throw new qD(t)}function wv(n,t){for(var e=0,i=0,r=n;i<r.length;i++){var a=r[i];a===t&&e++}return e}function Mt(n){return n.length===1?n[0]:n}function Je(n){return Array.isArray(n)?n:[n]}function Jn(n){var t=n.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2"),e=t.replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function Zi(n){return n.length<=1||n.indexOf("_")===-1?n:n.replace(/[_]+(\w|$)/g,function(t,e){return e.toUpperCase()})}var ln={};function oh(n){if(n==null)return null;var t={};return t.className=n.getClassName(),t.config=n.getConfig(),t}function lh(n){if(n==null||typeof n!="object")return;if(Array.isArray(n))n.forEach(function(s){return lh(s)});else for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e],a=n[r];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?n[r]=a.value:lh(a))}}function Da(n,t,e,i,r){var a,s,o;if(t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i="object"),r===void 0&&(r=!1),typeof n=="string"){var l=n,u=void 0;if(l in e)u=e[l];else if(l in ln)u=ln[l];else if(u=t[l],u==null)throw new M("Unknown "+i+": "+n+`. This may be due to one of the following reasons:
 `+("1. The "+i+" is defined in Python, in which ")+`case it needs to be ported to TensorFlow.js or your JavaScript code.
 `+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");return u}else{var c=n;if(c.className==null||c.config==null)throw new M(i+": Improper config format: "+(JSON.stringify(c)+`.
-`)+"'className' and 'config' must set.");var h=c.className,d=void 0,p=void 0;if(h in e?(a=e[h],d=a[0],p=a[1]):h in un?(s=un.className,d=s[0],p=s[1]):h in t&&(o=t[h],d=o[0],p=o[1]),d==null)throw new M("Unknown "+i+": "+h+`. This may be due to one of the following reasons:
+`)+"'className' and 'config' must set.");var h=c.className,d=void 0,p=void 0;if(h in e?(a=e[h],d=a[0],p=a[1]):h in ln?(s=ln.className,d=s[0],p=s[1]):h in t&&(o=t[h],d=o[0],p=o[1]),d==null)throw new M("Unknown "+i+": "+h+`. This may be due to one of the following reasons:
 `+("1. The "+i+" is defined in Python, in which ")+`case it needs to be ported to TensorFlow.js or your JavaScript code.
-`+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");if(p!=null){for(var f={},m=0,g=Object.keys(un);m<g.length;m++){var v=g[m];f[v]=un[v]}for(var b=0,w=Object.keys(e);b<w.length;b++){var v=w[b];f[v]=e[v]}var S=c.config;S.customObjects=f;for(var L=Yt({},un),x=0,C=Object.keys(e);x<C.length;x++){var v=C[x];un[v]=e[v]}oh(c.config);var R=p(d,c.config,e,r);return un=Yt({},L),R}else{for(var L=Yt({},un),D=0,k=Object.keys(e);D<k.length;D++){var v=k[D];un[v]=e[v]}var R=new d(c.config);return un=Yt({},L),R}}}function GD(n,t){return n<t?-1:n>t?1:0}function to(n,t){return-1*GD(n,t)}function fi(n){if(n==null)return n;for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.indexOf(r)===-1&&t.push(r)}return t}function YD(n){if(n==null)throw new M("Invalid value in obj: "+JSON.stringify(n));for(var t in n)if(n.hasOwnProperty(t))return!1;return!0}function _r(n,t,e){if(e==null)return;if(n.indexOf(e)<0)throw new M(e+" is not a valid "+t+".  Valid values are "+n+" or null/undefined.")}function lh(n,t,e,i){return e===void 0&&(e=0),i===void 0&&(i=Infinity),Wn(e>=0),Wn(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function At(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return At(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(Sv(n)+".")})}function Sv(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return Sv(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function KD(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];var o=y.util.now();return o-e<t||(e=o,i=n.apply(void 0,a)),i};return r}function Lv(n){return n==="relu"?"relu":n==="linear"?"linear":n==="elu"?"elu":null}function uh(n,t){return y.tidy(function(){return y.sqrt(y.sum(y.mul(n,n),t,!0))})}var ka=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),Iv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMaxValue=2,i.defaultAxis=0,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=uh(e,i.axis),a=y.clipByValue(r,0,i.maxValue);return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{maxValue:this.maxValue,axis:this.axis}},t.className="MaxNorm",t}(ka);y.serialization.registerClass(Iv);var Av=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultAxis=0,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){return y.div(e,y.add(vt(),uh(e,i.axis)))})},t.prototype.getConfig=function(){return{axis:this.axis}},t.className="UnitNorm",t}(ka);y.serialization.registerClass(Av);var Tv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="NonNeg",t}(ka);y.serialization.registerClass(Tv);var Nv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMinValue=0,i.defaultMaxValue=1,i.defaultRate=1,i.defaultAxis=0,i.minValue=e.minValue!=null?e.minValue:i.defaultMinValue,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.rate=e.rate!=null?e.rate:i.defaultRate,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=uh(e,i.axis),a=y.add(y.mul(i.rate,y.clipByValue(r,i.minValue,i.maxValue)),y.mul(1-i.rate,r));return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}},t.className="MinMaxNorm",t}(ka);y.serialization.registerClass(Nv);var xv={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function yt(n){return sh(n)}function Cv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"constraint")}function bt(n){if(n==null)return null;if(typeof n=="string"){var t=n in xv?xv[n]:n,e={className:t,config:{}};return Cv(e)}else return n instanceof ka?n:Cv(n)}function jD(n){return new Iv(n)}function $D(n){return new Av(n)}function XD(){return new Tv}function JD(n){return new Nv(n)}var ZD={__proto__:null,maxNorm:jD,unitNorm:$D,nonNeg:XD,minMaxNorm:JD};var QD=["channelsFirst","channelsLast"],ek=["valid","same","causal"],tk=["max","avg"],nk=["sum","mul","concat","ave"];var Mr=new Map;function ct(n){_r(QD,"DataFormat",n)}function nn(n){_r(ek,"PaddingMode",n)}function Rv(n){_r(tk,"PoolMode",n)}var Fa=[],Ov="/";function Qi(n,t){Fa.push(n);try{var e=t();return Fa.pop(),e}catch(i){throw Fa.pop(),i}}function ik(){return Fa.length===0?"":Fa.join(Ov)+Ov}function Dv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");return ik()+n}function kv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");Mr.has(n)||Mr.set(n,0);var t=Mr.get(n);if(Mr.set(n,Mr.get(n)+1),t>0){var e=n+"_"+t;return Mr.set(e,1),e}else return n}var rk=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function Ev(n){return!!n.match(rk)}function ak(n){return n===parseInt(n.toString(),10)}function mi(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r<e;++r)i*=n[r];return i}function Fv(n){return n=Array.isArray(n)?new Float32Array(n):n,y.tensor1d(n)}function Hr(n){return y.min(Fv(n)).dataSync()[0]}function gi(n){return y.max(Fv(n)).dataSync()[0]}function vn(n,t){if(t<n)throw new M("end ("+t+") < begin ("+n+") is forbidden.");for(var e=[],i=n;i<t;++i)e.push(i);return e}function Wa(n,t){return n.asType(t)}function Ua(n,t){t===void 0&&(t=-1);var e=n.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),n.reshape(e)}function sk(n,t){return y.tidy(function(){if(n.shape.length!==2)throw new M("repeat() expects a rank-2 tensor, but received a "+("rank-"+n.shape.length+" tensor."));var e=Ua(n,1);return ch(e,[1,t,1])})}function ok(n){var t=[mi(n.shape)];return n.reshape(t)}function lk(n){if(n.rank<=1)throw new M("batchFlatten requires a minimum rank of 2. Got rank: "+n.rank+".");var t=[n.shape[0],mi(n.shape,1)];return n.reshape(t)}function er(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[t,0],[e,n.shape[1]]);case 3:return y.slice3d(n,[t,0,0],[e,n.shape[1],n.shape[2]]);case 4:return y.slice4d(n,[t,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3]]);case 5:return y.slice(n,[t,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4]]);case 6:return y.slice(n,[t,0,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4],n.shape[5]]);default:throw new M("sliceAlongFirstAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function hh(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[0,t],[n.shape[0],e]);case 3:return y.slice3d(n,[0,0,t],[n.shape[0],n.shape[1],e]);case 4:return y.slice4d(n,[0,0,0,t],[n.shape[0],n.shape[1],n.shape[2],e]);default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function no(n,t,e,i){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:switch(i){case 1:return er(n,t,e);case 2:return hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 3:switch(i){case 1:return er(n,t,e);case 2:return y.slice3d(n,[0,t,0],[n.shape[0],e,n.shape[2]]);case 3:return hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 4:switch(i){case 1:return er(n,t,e);case 2:return y.slice4d(n,[0,t,0,0],[n.shape[0],e,n.shape[2],n.shape[3]]);case 3:return y.slice4d(n,[0,0,t,0],[n.shape[0],n.shape[1],e,n.shape[3]]);case 4:return hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function dh(n,t){t===void 0&&(t=-1);var e;return t<0&&(e=n[0].rank,e!==0?t=e:t=0),t===n[0].rank&&(t=-1),y.concat(n,t)}function Wv(n,t){switch(n.rank){case 1:return y.concat1d([n,t]);case 2:return y.concat2d([n,t],0);case 3:return y.concat3d([n,t],0);case 4:return y.concat4d([n,t],0);default:throw new M("concatAlongFirstAxis() received an unsupported "+("tensor rank: "+n.rank))}}function ch(n,t){if(Array.isArray(t)||(t=[t]),n.rank!==t.length)throw new M("The length of input n ("+t.length+") does not match "+("the number of dimensions in input x ("+n.rank+")"));return y.tile(n,t)}function io(n,t,e,i,r){return t===void 0&&(t=0),e===void 0&&(e=1),y.randomNormal(n,t,e,i,r)}function Un(n,t,e,i){if(n.rank<2||t.rank<2)throw new Te("dot requires both inputs to be rank >= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?ph(n.rank,i,gn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?ph(n.rank,i,gn()):null,activation:e}).reshape(f)}}function Uv(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function Ba(n){return y.mul(n,n)}function ph(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Bn(n,t,e){return y.tidy(function(){return e==null&&(e=gn()),ct(e),n.add(ph(n.rank,t,e))})}function uk(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function ck(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function Bv(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function hk(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function za(n,t,e){return e===void 0&&(e=!1),e?n():t()}var dk=["fanIn","fanOut","fanAvg"],pk=["normal","uniform","truncatedNormal"];function fk(n){_r(dk,"FanMode",n)}function mk(n){_r(pk,"Distribution",n)}var cn=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),zv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(cn);y.serialization.registerClass(zv);var fh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(cn);y.serialization.registerClass(fh);var Pv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(cn);y.serialization.registerClass(Pv);var _v=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(cn);y.serialization.registerClass(_v);var Mv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return io(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(cn);y.serialization.registerClass(Mv);var Hv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(cn);y.serialization.registerClass(Hv);var Vv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(cn);y.serialization.registerClass(Vv);function gk(n,t){t===void 0&&(t="channelsLast");var e,i;if(ct(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=mi(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=mi(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=mi(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var Kt=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,fk(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,mk(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=gk(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(cn);y.serialization.registerClass(Kt);var mh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="GlorotUniform",t}(Kt);y.serialization.registerClass(mh);var gh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="GlorotNormal",t}(Kt);y.serialization.registerClass(gh);var vh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="HeNormal",t}(Kt);y.serialization.registerClass(vh);var yh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="HeUniform",t}(Kt);y.serialization.registerClass(yh);var bh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="LeCunNormal",t}(Kt);y.serialization.registerClass(bh);var wh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="LeCunNormal",t}(Kt);y.serialization.registerClass(wh);var qv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=io(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(cn);y.serialization.registerClass(qv);var Gv={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function Yv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function st(n){return sh(n)}function tt(n){if(typeof n=="string"){var t=n in Gv?Gv[n]:n;if(t==="GlorotNormal")return new gh;if(t==="GlorotUniform")return new mh;if(t==="HeNormal")return new vh;if(t==="HeUniform")return new yh;if(t==="LeCunNormal")return new bh;if(t==="LeCunUniform")return new wh;var e={};return e.className=t,e.config={},Yv(e)}else return n instanceof cn?n:Yv(n)}function vk(){return new zv}function yk(){return new fh}function bk(n){return new Pv(n)}function wk(n){return new _v(n)}function Sk(n){return new Mv(n)}function Lk(n){return new Hv(n)}function Ik(n){return new Vv(n)}function Ak(n){return new Kt(n)}function Tk(n){return new mh(n)}function Nk(n){return new gh(n)}function xk(n){return new vh(n)}function Ck(n){return new yh(n)}function Rk(n){return new bh(n)}function Ok(n){return new wh(n)}function Ek(n){return new qv(n)}var Dk={__proto__:null,zeros:vk,ones:yk,constant:bk,randomUniform:wk,randomNormal:Sk,truncatedNormal:Lk,identity:Ik,varianceScaling:Ak,glorotUniform:Tk,glorotNormal:Nk,heNormal:xk,heUniform:Ck,leCunNormal:Rk,leCunUniform:Ok,orthogonal:Ek};var kk=0;function Kv(){return kk++}var ro={};function ao(n){return n===void 0&&(n=""),n in ro||(ro[n]=0),ro[n]+=1,n+ro[n].toString()}function Sh(n){return Array.isArray(n)&&Array.isArray(n[0])}function so(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function Ke(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function oo(n){for(var t=0,e=0,i=n;e<i.length;e++){var r=i[e];r.shape.length===0?t+=1:t+=r.shape.reduce(function(a,s){return a*s})}return t}var jv="Variable",$v=function(){function n(t,e,i,r,a){e===void 0&&(e="float32"),i===void 0&&(i=jv),r===void 0&&(r=!0),a===void 0&&(a=null),this.dtype=e??"float32",this.shape=t.shape,this.id=Kv(),i=i??jv,this.originalName=Dv(i),this.name=kv(this.originalName),this.trainable_=r,this.constraint=a,this.val=y.variable(t,this.trainable_,this.name,this.dtype)}return n.prototype.read=function(){return this.assertNotDisposed(),this.val},n.prototype.write=function(t){return this.assertNotDisposed(),Fk(this.val,t),this.val.id!==t.id&&(this.val.assign(t),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this},n.prototype.dispose=function(){this.assertNotDisposed(),this.val.dispose()},n.prototype.assertNotDisposed=function(){if(this.val.isDisposed)throw new Error("LayersVariable "+this.name+" is already disposed.")},Object.defineProperty(n.prototype,"trainable",{get:function(){return this.trainable_},set:function(t){this.trainable_=t,this.val.trainable=t},enumerable:!0,configurable:!0}),n}();function Fk(n,t){if(n.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(n.shape)+" vs. "+JSON.stringify(t.shape))}function Lh(n){return n.map(function(t){return t.read()})}function Ih(n){n.forEach(function(t){var e=t[0];e.write(t[1])})}var Tt=function(){function n(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}return n}(),yn=function(){function n(t,e,i,r,a,s,o){this.dtype=t,this.shape=e,this.sourceLayer=i,this.inputs=r,this.callArgs=a,this.outputTensorIndex=o,this.id=Kv(),s!=null&&(this.originalName=Dv(s),this.name=kv(this.originalName)),this.rank=e.length}return n}(),Wk=0,lo=function(){function n(t,e){this.callArgs=e,this.id=Wk++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(var i=0,r=t.inboundLayers;i<r.length;i++){var a=r[i];a!=null&&a.outboundNodes.push(this)}t.outboundLayer.inboundNodes.push(this)}return n.prototype.getConfig=function(){for(var t=[],e=0,i=this.inboundLayers;e<i.length;e++){var r=i[e];r!=null?t.push(r.name):t.push(null)}return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}},n}(),Uk=0,De=function(n){Q(t,n);function t(e){e===void 0&&(e={});var i=n.call(this)||this;i._callHook=null,i._addedWeightNames=[],i._stateful=!1,i.id=Uk++,i.activityRegularizer=null,i.inputSpec=null,i.supportsMasking=!1,i._trainableWeights=[],i._nonTrainableWeights=[],i._losses=[],i._updates=[],i._built=!1,i.inboundNodes=[],i.outboundNodes=[];var r=e.name;if(!r){var a=i.getClassName();r=Jn(a)+"_"+ao(a)}if(i.name=r,i.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){var s=void 0;if(e.batchInputShape!=null)s=e.batchInputShape;else if(e.inputShape!=null){var o=null;e.batchSize!=null&&(o=e.batchSize),s=[o].concat(e.inputShape)}i.batchInputShape=s;var l=e.dtype;l==null&&(l=e.inputDType),l==null&&(l="float32"),i.dtype=l}return e.weights!=null?i.initialWeights=e.weights:i.initialWeights=null,i._refCount=null,i.fastWeightInitDuringBuild=!1,i}return t.nodeKey=function(e,i){return e.name+"_ib-"+i.toString()},t.prototype.getNodeAtIndex=function(e,i){if(this.inboundNodes.length===0)throw new Fn("The layer has never been called "+("and thus has no defined "+i+"."));if(this.inboundNodes.length<=e)throw new M("Asked to get "+i+" at node "+e+", "+("but the layer has only "+this.inboundNodes.length+" inbound nodes."));return this.inboundNodes[e]},t.prototype.getInputAt=function(e){return Mt(this.getNodeAtIndex(e,"input").inputTensors)},t.prototype.getOutputAt=function(e){return Mt(this.getNodeAtIndex(e,"output").outputTensors)},Object.defineProperty(t.prototype,"input",{get:function(){if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" is not connected, no input to return.");return Mt(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return Mt(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Je(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Je(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r<e.length;r++){var a=e[r],s=i[r];if(s==null)continue;var o=a.rank;if(s.ndim!=null&&o!==s.ndim)throw new M("Input "+r+" is incompatible with layer "+this.name+": "+("expected ndim="+s.ndim+", found ndim="+o));if(s.maxNDim!=null&&o>s.maxNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected max_ndim="+s.maxNDim+", found ndim="+o));if(s.minNDim!=null&&o<s.minNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected min_ndim="+s.minNDim+", found ndim="+o+"."));if(s.dtype!=null&&a.dtype!==s.dtype)throw new M("Input "+r+" is incompatible with layer "+this.name+" "+(": expected dtype="+s.dtype+", found dtype="+a.dtype+"."));if(s.axes){var l=a.shape;for(var u in s.axes){var c=Number(u),h=s.axes[u],d=c>=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p<s.shape.length;++p){var f=s.shape[p],m=a.shape[p];if(f!=null&&m!=null&&f!==m)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected shape="+s.shape+", ")+("found shape="+a.shape+"."))}}},t.prototype.call=function(e,i){return e},t.prototype.invokeCallHook=function(e,i){this._callHook!=null&&this._callHook(e,i)},t.prototype.setCallHook=function(e){this._callHook=e},t.prototype.clearCallHook=function(){this._callHook=null},t.prototype.apply=function(e,i){var r=this;i=i||{},this.assertNotDisposed();for(var a=Je(e),s=!0,o=0,l=a;o<l.length;o++){var u=l[o];if(!(u instanceof yn)){s=!1;break}}for(var c=!0,h=0,d=a;h<d.length;h++){var u=d[h];if(u instanceof yn){c=!1;break}}if(s===c)throw new M("Arguments to apply() must be all SymbolicTensors or all Tensors");return Qi(this.name,function(){if(!r.built){r.assertInputCompatibility(e);for(var p=[],f=0,m=Je(e);f<m.length;f++){var g=m[f];p.push(g.shape)}r.build(Mt(p)),r.built=!0,r.initialWeights&&r.setWeights(r.initialWeights),r._refCount===null&&c&&(r._refCount=1)}if(r.assertInputCompatibility(e),c){for(var v=r.call(e,i),b=Je(v),w=[],S=0,L=b;S<L.length;S++){var x=L[S];a.indexOf(x)!==-1&&(x=x.clone()),w.push(x)}if(v=Mt(w),r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}else{var C=Bk(e),R=r.computeOutputShape(C),v=void 0,D=zk(e);if(r.warnOnIncompatibleInputShape(Array.isArray(e)?C[0]:C),R!=null&&R.length>0&&Array.isArray(R[0])?v=R.map(function(W,F){return new yn(D,W,r,Je(e),i,r.name,F)}):v=new yn(D,R,r,Je(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new pi("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i<r.length;i++){var a=r[i],s=JSON.stringify(a.outputShapes);e.indexOf(s)===-1&&e.push(s)}if(e.length===1){var o=this.inboundNodes[0].outputShapes;return Array.isArray(o)&&Array.isArray(o[0])&&o.length===1?o[0]:o}else throw new pi("The layer "+this.name+' has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.')},enumerable:!0,configurable:!0}),t.prototype.countParams=function(){if(!this.built)throw new Fn("You tried to call countParams() on "+this.name+", but the layer is not built yet. Build it first by calling build(batchInputShape).");return oo(this.weights)},t.prototype.build=function(e){this.built=!0},t.prototype.getWeights=function(e){return e===void 0&&(e=!1),Lh(e?this.trainableWeights:this.weights)},t.prototype.setWeights=function(e){var i=this;y.tidy(function(){var r=i.weights;if(r.length!==e.length)throw new M('You called setWeights(weights) on layer "'+i.name+'" '+("with a weight list of length "+e.length+", ")+("but the layer was expecting "+r.length+" weights. ")+("Provided weights: "+e+"..."));if(r.length===0)return;for(var a=[],s=Lh(r),o=0;o<s.length;++o){var l=s[o],u=r[o],c=e[o];if(!y.util.arraysEqual(l.shape,c.shape))throw new M("Layer weight shape "+l.shape+" "+("not compatible with provided weight shape "+c.shape));a.push([u,c])}Ih(a)})},t.prototype.addWeight=function(e,i,r,a,s,o,l){if(this._addedWeightNames.indexOf(e)!==-1)throw new M("Duplicate weight name "+e+" for layer "+this.name);this._addedWeightNames.push(e),r==null&&(r="float32"),this.fastWeightInitDuringBuild&&(a=tt("zeros"));var u=a.apply(i,r),c=new $v(u,r,e,o,l);return u.dispose(),s!=null&&this.addLoss(function(){return s.apply(c.read())}),o==null&&(o=!0),o?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c},t.prototype.setFastWeightInitDuringBuild=function(e){this.fastWeightInitDuringBuild=e},t.prototype.addLoss=function(e){var i;if(e==null||Array.isArray(e)&&e.length===0)return;e=Je(e),this._losses!==void 0&&this._losses!==null&&(i=this.losses).push.apply(i,e)},t.prototype.computeOutputShape=function(e){return e},t.prototype.computeMask=function(e,i){var r=this;if(!this.supportsMasking){if(i!=null)if(Array.isArray(i))i.forEach(function(a){if(a!=null)throw new TypeError("Layer "+r.name+" does not support masking, but was passed an inputMask.")});else throw new TypeError("Layer "+this.name+" does not support masking, but was passed an inputMask.");return null}return i},t.prototype.addInboundNode=function(e,i,r,a,s,o,l){l===void 0&&(l=null);var u=Je(e);i=Je(i),r=Je(r),a=Je(a),s=so(s),o=so(o);for(var c=[],h=[],d=[],p=0,f=u;p<f.length;p++){var m=f[p];c.push(m.sourceLayer),h.push(m.nodeIndex),d.push(m.tensorIndex)}new lo({outboundLayer:this,inboundLayers:c,nodeIndices:h,tensorIndices:d,inputTensors:u,outputTensors:i,inputMasks:r,outputMasks:a,inputShapes:s,outputShapes:o},l);for(var g=0;g<i.length;g++)i[g].sourceLayer=this,i[g].nodeIndex=this.inboundNodes.length-1,i[g].tensorIndex=g},t.prototype.getConfig=function(){var e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e},t.prototype.disposeWeights=function(){return this.weights.forEach(function(e){return e.dispose()}),this.weights.length},t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Layer '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){if(!this.built)throw new Error("Cannot dispose Layer "+this.name+" because it has not been built yet.");if(this._refCount===null)throw new Error("Cannot dispose Layer "+this.name+" because it has not been used yet.");this.assertNotDisposed();var e=0;return--this._refCount===0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}},t}(y.serialization.Serializable);function Bk(n){n=Je(n);for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.push(r.shape)}return Mt(t)}function zk(n){return"float32"}function Xv(n,t,e){if((t==null||e!=null&&e>0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a<i.inboundLayers.length;a++)for(var s=i.inputTensors[a],o=i.inboundLayers[a],l=i.nodeIndices[a],u=Xv(s,o,l),c=0,h=u;c<h.length;c++){var d=h[c];r.indexOf(d)===-1&&r.push(d)}return r}var Pa=function(n){Q(t,n);function t(e){var i=n.call(this,{dtype:e.dtype,name:e.name!=null?e.name:ao("input").toString()})||this;if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),i.trainable=!1,i.built=!0,i.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new M("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");var r=e.batchInputShape;if(r==null){if(e.inputShape==null)throw new M("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");r=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new M("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");var a=e.dtype||"float32";i.batchInputShape=r,i.dtype=a,i.inputSpec=[{shape:r}];var s=new yn(i.dtype,i.batchInputShape,i,[],{},i.name);return s.nodeIndex=0,s.tensorIndex=0,new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[r],outputShapes:[r]}),i}return t.prototype.apply=function(e,i){throw new M("Cannot pass any input to an "+("InputLayer's apply() method. InputLayer name: "+this.name))},t.prototype.dispose=function(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}},t.prototype.getConfig=function(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}},t.className="InputLayer",t}(De);y.serialization.registerClass(Pa);function Jv(n){if(n.batchShape==null&&n.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(n.batchShape!=null&&n.shape!=null)throw new M("Please provide either a `shape` or `batchShape` argument to Input, but not both.");var t=n.batchShape;n.shape!=null&&t==null&&(t=[null].concat(n.shape));var e=n.dtype;e==null&&(e="float32");var i=new Pa({batchInputShape:t,name:n.name,dtype:e,sparse:n.sparse}),r=i.inboundNodes[0].outputTensors;return r[0]}function vi(n){return Le(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(n==null)return[2];t=[],e=[],i=[];for(r in n)a=n[r],typeof a!="number"&&(s=a,t.push(s.data()),e.push(r),i.push(s));return t.length>0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l<o.length;++l)n[e[l]]=o[l][0];y.dispose(i),u.label=2;case 2:return[2]}})})}function Zv(n){if(n==null)return;for(var t in n){var e=n[t];typeof e!="number"&&e.dispose()}}var Qv;(function(n){n[n.SILENT=0]="SILENT",n[n.VERBOSE=1]="VERBOSE"})(Qv||(Qv={}));var Pk=125,Vr=function(){function n(){this.validationData=null}return n.prototype.setParams=function(t){this.params=t},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.setModel=function(t){},n}(),ey=function(){function n(t,e){e===void 0&&(e=10),t==null&&(t=[]),this.callbacks=t,this.queueLength=e}return n.prototype.append=function(t){this.callbacks.push(t)},n.prototype.setParams=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setParams(t)}},n.prototype.setModel=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setModel(t)}},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainBegin(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainEnd(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n}(),_k=function(n){Q(t,n);function t(){return n.call(this)||this}return t.prototype.onEpochBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.seen=0,this.totals={},[2]})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l=this;return ve(this,function(u){i==null&&(i={}),r=i.size==null?0:i.size,this.seen+=r,a=function(c){var h=i[c];if(typeof h=="number")s.totals.hasOwnProperty(c)||(s.totals[c]=0),s.totals[c]=s.totals[c]+h*r;else{var d=void 0;c in s.totals?d=s.totals[c]:s.totals[c]=0;var p=y.tidy(function(){return y.add(l.totals[c],y.mul(h,r))});s.totals[c]=p,d!=null&&d.dispose()}},s=this;for(o in i)a(o);return[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u=this;return ve(this,function(c){if(i!=null)for(r=function(h){if(a.totals[h]==null)return"continue";typeof a.totals[h]=="number"?i[h]=a.totals[h]/a.seen:y.tidy(function(){var d=y.mul(y.div(1,u.seen),u.totals[h]);i[h]=d,u.totals[h].dispose(),y.keep(i[h])})},a=this,s=0,o=this.params.metrics;s<o.length;s++)l=o[s],r(l);return[2]})})},t}(Vr),ty=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.epoch=[],this.history={},[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){i==null&&(i={}),this.epoch.push(e);for(r in i)this.history[r]==null&&(this.history[r]=[]),this.history[r].push(i[r]);return[2]})})},t.prototype.syncData=function(){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:e=[],i=[],r=[];for(a in this.history)for(s=this.history[a],o=0;o<s.length;++o)typeof s[o]!="number"&&(l=s[o],e.push(l.data()),i.push(a),r.push(o));return[4,Promise.all(e)];case 1:for(u=d.sent(),c=0;c<u.length;++c)h=this.history[i[c]][r[c]],h.dispose(),this.history[i[c]][r[c]]=u[c][0];return[2]}})})},t}(Vr),ny=function(n){Q(t,n);function t(e,i){var r=n.call(this)||this;if(r.currentEpoch=0,r.yieldEvery=i||"auto",r.yieldEvery==="auto"&&(r.yieldEvery=Pk),r.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");return y.util.isNumber(r.yieldEvery)&&(r.maybeWait=KD(r.maybeWait.bind(r),r.yieldEvery)),r.trainBegin=e.onTrainBegin,r.trainEnd=e.onTrainEnd,r.epochBegin=e.onEpochBegin,r.epochEnd=e.onEpochEnd,r.batchBegin=e.onBatchBegin,r.batchEnd=e.onBatchEnd,r.yield=e.onYield,r}return t.prototype.maybeWait=function(e,i,r){return Le(this,void 0,void 0,function(){var a;return ve(this,function(s){switch(s.label){case 0:return a=[],this.yield!=null?[4,vi(r)]:[3,2];case 1:s.sent(),a.push(this.yield(e,i,r)),s.label=2;case 2:return a.push(y.nextFrame()),[4,Promise.all(a)];case 3:return s.sent(),[2]}})})},t.prototype.onEpochBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.currentEpoch=e,this.epochBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.epochBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.epochEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.epochEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="epoch"&&r.push(y.nextFrame()),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onBatchBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.batchBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.batchBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.batchEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.batchEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="batch"?r.push(y.nextFrame()):y.util.isNumber(this.yieldEvery)&&r.push(this.maybeWait(this.currentEpoch,e,i)),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainBegin!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainBegin(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainEnd!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainEnd(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t}(Vr);function iy(n,t){if(n==null&&(n={}),n instanceof Vr)return[n];if(Array.isArray(n)&&n[0]instanceof Vr)return n;var e=Je(n);return e.map(function(i){return new ny(i,t)})}var ry=function(){function n(){}return n.registerCallbackConstructor=function(t,e){y.util.assert(t>=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function ay(n,t,e,i,r,a,s,o,l){var u=new ty,c=[new _k].concat(ry.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new ey(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function bn(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function uo(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(Ba(n),t,!0),i=y.fill(e.shape,vt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function tr(n,t){return y.tidy(function(){return y.mean(Ba(y.sub(t,n)),-1)})}function co(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function qr(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),vt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function Mk(n,t){return y.tidy(function(){var e=y.clipByValue(t,vt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,vt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(Ba(y.sub(i,a)),-1)})}function Hk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(Ba(e),-1)})}function Vk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function qk(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function Gk(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function _a(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,vt(),1-vt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function ho(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(ok(n)).toInt();t=y.clipByValue(t,vt(),1-vt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return _a(a,t,e)})}function Yk(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function po(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,vt(),1-vt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(Yk(n,e),-1)})}function Kk(n,t){return y.tidy(function(){var e=y.clipByValue(n,vt(),1),i=y.clipByValue(t,vt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function jk(n,t){return y.tidy(function(){var e=y.log(y.add(vt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function Ah(n,t){return y.tidy(function(){var e=uo(n,-1),i=uo(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var fo={meanSquaredError:tr,meanAbsoluteError:co,meanAbsolutePercentageError:qr,meanSquaredLogarithmicError:Mk,squaredHinge:Hk,hinge:Vk,categoricalHinge:qk,logcosh:Gk,categoricalCrossentropy:_a,sparseCategoricalCrossentropy:ho,binaryCrossentropy:po,kullbackLeiblerDivergence:Kk,poisson:jk,cosineProximity:Ah};function Th(n){if(typeof n=="string"){if(n in fo)return fo[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function Nh(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=Wa(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function xh(n,t){return y.tidy(function(){return Wa(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function sy(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function $k(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function Xk(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function oy(n,t){return y.tidy(function(){var e=sy(n,t),i=Xk(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Jk(n,t){return y.tidy(function(){var e=sy(n,t),i=$k(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function ly(n,t){return po(n,t)}function uy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var Zk=tr,Qk=tr,e3=co,t3=co,n3=qr,i3=qr,Ch=_a,r3=Ah,cy=ho,mo={binaryAccuracy:Nh,categoricalAccuracy:xh,precision:oy,categoricalCrossentropy:Ch,sparseCategoricalCrossentropy:cy,mse:Zk,MSE:Qk,mae:e3,MAE:t3,mape:n3,MAPE:i3,cosine:r3};function a3(n){if(typeof n=="string"&&n in mo)return mo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function go(n){if(Wn(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(fo);e<i.length;e++){var r=i[e];if(fo[r]===n){t=r;break}}if(t!==void 0)return t;for(var a=0,s=Object.keys(mo);a<s.length;a++){var r=s[a];if(mo[r]===n){t=r;break}}return t!==void 0?t:n.name}function s3(n){var t={Adagrad:function(){return y.train.adagrad(.01)},Adadelta:function(){return y.train.adadelta(1,.95,vt())},Adam:function(){return y.train.adam(.001,.9,.999,vt())},Adamax:function(){return y.train.adamax(.002,.9,.999,vt(),0)},RMSProp:function(){return y.train.rmsprop(.001,.9,0,vt())},SGD:function(){return y.train.sgd(.01)}};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,n in t)return t[n]();throw new M("Unknown Optimizer "+n)}var hy=1*1024*1024;function dy(n,t,e){if(e===void 0&&(e=!1),n==null||typeof n!="object"||Object.getPrototypeOf(n)!==Object.prototype||!Rh(n))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){var i=JSON.stringify(n);i.length>hy&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+(hy+"."))}}function Rh(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e];if(typeof r!="string")return!1;if(!Rh(n[r]))return!1}return!0}else if(Array.isArray(n)){for(var a=0,s=n;a<s.length;a++){var o=s[a];if(!Rh(o))return!1}return!0}else return!1;else{var l=typeof n;return l==="string"||l==="number"||l==="boolean"}}function h3(n,t,e,i){i===void 0&&(i=console.log);var r=l3(n),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,e=e||[.45,.85,1]):(t=t||98,e=e||[.33,.55,.67,1]),e[e.length-1]<=1&&(e=e.map(function(d){return Math.floor(t*d)}));var s;if(!r){a.push("Receives inputs"),s=[];for(var o in n.nodesByDepth)s.push.apply(s,n.nodesByDepth[o])}i("_".repeat(t)),vo(a,e,i),i("=".repeat(t));for(var l=n.layers,u=0;u<l.length;++u)r?u3(l[u],e,i):c3(l[u],e,s,i),i((u===l.length-1?"=":"_").repeat(t));n.checkTrainableWeightsConsistency();var c=o3(n),h=oo(n.nonTrainableWeights);i("Total params: "+(c+h)),i("Trainable params: "+c),i("Non-trainable params: "+h),i("_".repeat(t))}function o3(n){var t;return n.collectedTrainableWeights!=null?t=oo(n.collectedTrainableWeights):t=oo(n.trainableWeights),t}function l3(n){var t=!0,e=[],i=[];for(var r in n.nodesByDepth)e.push(n.nodesByDepth[r]);for(var a=0,s=e;a<s.length;a++){var o=s[a];if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l<u.length;l++){for(var c=u[l],h=!1,d=0,p=c.inboundNodes;d<p.length;d++){var f=p[d];if(i.indexOf(f)!==-1)if(h){t=!1;break}else h=!0}if(!t)break}return t}function vo(n,t,e){e===void 0&&(e=console.log);for(var i="",r=0;r<n.length;++r)r>0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function u3(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];vo(s,t,e)}function c3(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s<o.length;s++){var l=o[s];if(e!=null&&e.length>0&&e.indexOf(l)===-1)continue;for(var u=0;u<l.inboundLayers.length;++u){var c=l.inboundLayers[u].name,h=l.nodeIndices[u],d=l.tensorIndices[u];a.push(c+"["+h+"]["+d+"]")}}var p=n.name,f=n.getClassName(),m=a.length===0?"":a[0],g=[p+" ("+f+")",r,n.countParams().toString(),m];vo(g,t,i);for(var u=1;u<a.length;++u)vo(["","","",a[u]],t,i)}function py(n,t,e){return(n==="inboundNodes"||n==="outputLayers"||n==="inputLayers")&&t===0&&typeof e=="string"}function Ma(n,t){if(n===null)return null;if(typeof n=="string")return Zi(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Ma(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u];if(u==="name"&&typeof c=="string")s[u]=c;else{var h=Zi(u);s[h]=Ma(c,h)}}return s}}function Oh(n,t){if(n==null)return null;if(typeof n=="string")return Jn(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Oh(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u],h=Jn(u);(u==="name"||u==="className")&&typeof c=="string"?s[h]=c:s[h]=Oh(c,u)}return s}}var Eh="2.6.0";function d3(n,t){if(n.dtype==null||n.dtype===t.dtype)return t;try{return y.cast(t,n.dtype)}catch(e){throw new M("The dtype of the feed ("+t.dtype+") can not be cast to the dtype "+("of the key '"+n.name+"' ("+n.dtype+")."))}}var Gr=function(){function n(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof n)for(var e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(var i=0,r=t;i<r.length;i++){var a=r[i];this.add(a.key,a.value)}}}return n.prototype.add=function(t,e,i){if(this.id2Value[t.id]==null)this.id2Value[t.id]=d3(t,e),this.name2Id[t.name]=t.id,i!=null&&(this.id2Mask[t.id]=i);else throw new M("Duplicate key: name="+t.name+", id="+t.id);return this},n.prototype.addFeed=function(t){this.add(t.key,t.value)},n.prototype.hasKey=function(t){return this.id2Value[t.id]!=null},n.prototype.names=function(){return Object.keys(this.name2Id)},n.prototype.getValue=function(t){if(t instanceof yn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Value[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Value[e]}},n.prototype.getMask=function(t){if(t instanceof yn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Mask[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Mask[e]}},n.prototype.disposeMasks=function(){this.id2Mask!=null&&y.dispose(this.id2Mask)},n}(),Dh={},fy={};function Ha(n,t,e,i){for(var r=e==null?!1:e.training,a=Array.isArray(n),s=a?n:[n],o=s.map(function(X){return X.name}),l=[],u=t.names(),c=0,h=o;c<h.length;c++){var d=h[c];u.indexOf(d)!==-1?l.push(t.getValue(d)):l.push(null)}i!=null&&(i.maxNumTensors=-Infinity,i.minNumTensors=Infinity);var p=o.join(",")+"|"+t.names().join(","),f,m;if(Dh[p]==null){var g=p3(s,t);f=g.sorted,m=g.recipientCounts,Dh[p]=f,fy[p]=m}f=Dh[p],m={},r||Object.assign(m,fy[p]);for(var v=new Gr(t),b=0;b<f.length;++b){if(i!=null){var w=y.memory().numTensors;w>i.maxNumTensors&&(i.maxNumTensors=w),w<i.minNumTensors&&(i.minNumTensors=w)}var S=f[b],L=S.sourceLayer;if(L instanceof Pa)continue;for(var x=[],C=[],R=[],D=!1,k=0,W=S.inputs;k<W.length;k++){var F=W[k],P=v.getValue(F),H=v.getMask(F);x.push(P),C.push(H),H!=null&&(D=!0),r||(m[F.name]--,m[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!P.isDisposed&&F.sourceLayer.stateful!==!0&&R.push(P))}D&&(e=e||{},e.mask=C[0]);var _=Je(L.apply(x,e)),K=null;L.supportsMasking&&(K=L.computeMask(x,C));for(var j=f3(S),q=Array.isArray(j)?j:[j],G=0;G<q.length;++G){v.hasKey(q[G])||v.add(q[G],_[G],Array.isArray(K)?K[0]:K);var Z=o.indexOf(q[G].name);Z!==-1&&(l[Z]=_[G])}r||y.dispose(R)}return v.disposeMasks(),a?l:l[0]}function p3(n,t){y.util.assert(n!=null&&n.length>0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=my(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s<o.length;s++){for(var l=o[s],u=my(l,t),c=u.sorted,h=u.recipientMap,d=0,p=c;d<p.length;d++){var f=p[d];a.has(f.name)||(e.push(f),a.add(f.name))}var m=function(v){i[v]==null&&(i[v]=new Set),h[v].forEach(function(b){return i[v].add(b)})};for(var g in h)m(g)}return{sorted:e,recipientCounts:m3(i)}}function m3(n){var t={};for(var e in n)t[e]=n[e].size;return t}function my(n,t){for(var e=new Set,i=[],r={},a=0,s=t.names();a<s.length;a++){var o=s[a];e.add(o)}var l=[],u=[];for(l.push(n);l.length>0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d<p.length;d++){var f=p[d];if(r[f.name]==null&&(r[f.name]=new Set),r[f.name].add(c.name),e.has(f.name))continue;l.push(f)}}}return{sorted:i,recipientMap:r}}function f3(n){var t;if(n.sourceLayer.inboundNodes.length===1)t=n.sourceLayer.output;else{for(var e=null,i=0;i<n.sourceLayer.inboundNodes.length;++i)for(var r=0,a=n.sourceLayer.inboundNodes[i].outputTensors;r<a.length;r++){var s=a[r];if(s.id===n.id){e=i;break}}t=n.sourceLayer.getOutputAt(e)}return t}var g3=function(n){Q(t,n);function t(e){var i=n.call(this,{})||this;if(i.containerNodes=new Set,i.name=e.name,i.name==null){var r=i.getClassName().toLowerCase();i.name=ao(r)}if(i.supportsMasking=!1,i.trainable_=!0,Array.isArray(e.inputs)?i.inputs=e.inputs.slice():i.inputs=[e.inputs],Array.isArray(e.outputs)?i.outputs=e.outputs.slice():i.outputs=[e.outputs],fi(i.inputs).length!==i.inputs.length)throw new M("The list of inputs passed to the model is redundant. All inputs should only appear once. Found: "+(""+i.inputs.map(function(Ce){return Ce.name})));fi(i.outputs).length!==i.outputs.length&&console.warn("The list of outputs passed to the model is redundant. All outputs should only appear once. Found: "+(""+i.outputs.map(function(Ce){return Ce.name}))),i.inputLayers=[],i.inputLayersNodeIndices=[],i.inputLayersTensorIndices=[],i.outputLayers=[],i.outputLayersNodeIndices=[],i.outputLayersTensorIndices=[],i.layers=[],i.internalContainerRefs=[];for(var a=0,s=i.outputs;a<s.length;a++){var o=s[a],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;i.outputLayers.push(l),i.outputLayersNodeIndices.push(u),i.outputLayersTensorIndices.push(c)}for(var h=0,d=i.inputs;h<d.length;h++){var o=d[h],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;Wn(u===0,"input layer has >1 nodes"),Wn(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p<i.inputLayers.length;p++){var l=i.inputLayers[p];if(!(l instanceof Pa))throw new TypeError("Input layers to a LayersModel must be InputLayer objects. "+("Received inputs: "+e.inputs+". ")+("Input "+p+" (0-based) originates ")+("from layer type "+l.getClassName()+"."));i.inputNames.push(l.name),i.feedInputShapes.push(l.batchInputShape),i.feedInputNames.push(l.name)}for(var f=0,m=i.outputLayers;f<m.length;f++){var l=m[f];i.outputNames.push(l.name)}i.internalInputShapes=i.inputs.map(function(Ce){return Ce.shape}),i.internalOutputShapes=i.outputs.map(function(Ce){return Ce.shape});for(var g={},v={},b={},w={},S={},L=[],x=function(Ce,kt,pt,it,on,hs){(it==null||on==null||hs==null)&&(it=Ce.sourceLayer,on=Ce.nodeIndex,hs=Ce.tensorIndex);var Bt=it.inboundNodes[on];if(pt.indexOf(Bt)!==-1)throw new Fn("The tensor "+Ce.name+' at layer "'+it.name+'" is part of a cycle.');if(kt.indexOf(Bt)!==-1)return;i.containerNodes.add(t.nodeKey(it,on)),it.id in S||(S[it.id]=Object.keys(S).length),pt.indexOf(Bt)===-1&&pt.push(Bt);for(var II=Bt.inboundLayers.length,yr=0;yr<II;yr++){var AI=Bt.inputTensors[yr],TI=Bt.inboundLayers[yr],NI=Bt.nodeIndices[yr],xI=Bt.tensorIndices[yr];x(AI,kt,pt,TI,NI,xI)}for(kt.push(Bt);pt.indexOf(Bt)>=0;)pt.splice(pt.indexOf(Bt),1);L.push(Bt)},C=[],R=[],D=0,k=i.outputs;D<k.length;D++){var o=k[D];x(o,C,R)}for(var W=L.slice().reverse(),F=0,P=W;F<P.length;F++){var H=P[F];v[H.id]=H,H.id in g||(g[H.id]=0);var _=g[H.id],K=b[H.outboundLayer.id]==null?0:b[H.outboundLayer.id];_=Math.max(_,K),b[H.outboundLayer.id]=_,w[H.outboundLayer.id]=H.outboundLayer,g[H.id]=_;for(var p=0;p<H.inboundLayers.length;p++){var j=H.inboundLayers[p],u=H.nodeIndices[p],q=j.inboundNodes[u],G=g[q.id]==null?0:g[q.id];g[q.id]=Math.max(_+1,G),v[q.id]=q}}var Z={};for(var X in g){var _=g[X];_ in Z||(Z[_]=[]),Z[_].push(v[X])}var ee={};for(var ne in b){var _=b[ne];_ in ee||(ee[_]=[]),ee[_].push(w[ne])}var ie=Object.keys(ee).map(function(Ce){return parseInt(Ce,10)}).sort(to);i.layers=[];for(var te=0,re=ie;te<re.length;te++){var _=re[te],le=ee[_];le.sort(function(kt,pt){var it=S[kt.id],on=S[pt.id];return it<on?-1:it>on?1:0});for(var he=0,ye=le;he<ye.length;he++){var l=ye[he];l instanceof t&&i.internalContainerRefs.push(l),i.layers.push(l)}}i.layersByDepth=ee,ie=Object.keys(Z).map(function(Ce){return parseInt(Ce,10)}).sort(to);for(var Oe=i.inputs.slice(),Fe=[],_e=0,Me=ie;_e<Me.length;_e++)for(var _=Me[_e],ht=0,St=Z[_];ht<St.length;ht++){var H=St[ht],l=H.outboundLayer;if(l!=null){for(var Ge=0,Ue=H.inputTensors;Ge<Ue.length;Ge++){var o=Ue[Ge];if(Oe.indexOf(o)===-1)throw new Fn("Graph disconnected: cannot obtain value for tensor "+o+(' at layer "'+l.name+'". ')+"The following previous layers were accessed without "+("issue: "+Fe))}for(var me=0,Dt=H.outputTensors;me<Dt.length;me++){var o=Dt[me];Oe.push(o)}Fe.push(l.name)}}i.nodesByDepth=Z;for(var lt=i.layers.map(function(Ce){return Ce.name}),dt=function(Ce){var kt=lt.filter(function(pt){return pt===Ce}).length;if(kt!==1)throw new Fn('The name "'+Ce+'" is used '+kt+" times in the model. All layer names should be unique. Layer names: "+JSON.stringify(lt))},xn=0,Rt=lt;xn<Rt.length;xn++){var Lt=Rt[xn];dt(Lt)}return i.outboundNodes=[],i.inboundNodes=[],new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:i.inputs,outputTensors:i.outputs,inputMasks:i.inputs.map(function(Ce){return null}),outputMasks:i.outputs.map(function(Ce){return null}),inputShapes:i.inputs.map(function(Ce){return Ce.shape}),outputShapes:i.outputs.map(function(Ce){return Ce.shape})}),i.built=!0,i._refCount=1,i}return t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Container '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){this.assertNotDisposed();var e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(var i=0,r=this.layers;i<r.length;i++){var a=r[i];e.numDisposedVariables+=a.dispose().numDisposedVariables}for(var s=0,o=this.internalContainerRefs;s<o.length;s++){var l=o[s];e.numDisposedVariables+=l.dispose().numDisposedVariables}}return e.refCountAfterDispose=this._refCount,e},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this.layers.forEach(function(i){i._trainableWeights.forEach(function(r){return r.trainable=e})}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(this._trainableWeights.length>0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e=e.concat(a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.layers;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.loadWeights=function(e,i){i===void 0&&(i=!0);for(var r={},a=0,s=0,o=this.layers;s<o.length;s++)for(var l=o[s],u=0,c=l.weights;u<c.length;u++){var h=c[u];if(r[h.originalName]!=null)throw new M("Duplicate weight name: "+h.originalName);r[h.originalName]=h,a++}var d=[];for(var p in e){var f=p;if(r[p]==null){var m=p.split("/"),g=m.slice(0,-2).concat([m[m.length-1]]);f=g.join("/")}if(r[f]!=null)d.push([r[f],e[p]]);else if(i)throw new M("Provided weight data has no target variable: "+p);delete r[f]}if(i){var v=[];for(var b in r)v.push(b);if(v.length>0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}Ih(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+Eh,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=Oh(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Je(e);for(var a=new Gr,s=0;s<r.inputs.length;++s)a.add(r.inputs[s],e[s]);return Ha(r.outputs,a,i)})},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){e=Je(e);var a;return i==null?a=Ji(null,e.length):a=Je(i),r.runInternalGraph(e,a)[1]})},t.prototype.computeOutputShape=function(e){var i=so(e);if(i.length!==this.inputLayers.length)throw new M("Invalid inputShape argument "+e+": "+("model has "+this.inputLayers.length+" tensor inputs."));for(var r={},a=0;a<i.length;a++){var s=this.inputLayers[a],o=i[a],l=s.name+"_0_0";r[l]=o}var u=Object.keys(this.nodesByDepth).map(function(P){return parseInt(P,10)}).sort(to);if(u.length>1)for(var c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){var g=m[f],s=g.outboundLayer;if(this.inputLayers.map(function(H){return H.id}).indexOf(s.id)!==-1)continue;for(var v=[],b=0;b<g.inboundLayers.length;b++){var w=g.inboundLayers[b],S=g.nodeIndices[b],L=g.tensorIndices[b],l=w.name+"_"+S+"_"+L,x=r[l];v.push(x)}for(var C=s.computeOutputShape(Mt(v)),R=so(C),D=s.inboundNodes.indexOf(g),b=0;b<R.length;b++){var l=s.name+"_"+D+"_"+b;r[l]=R[b]}}for(var k=[],W=[],a=0;a<this.outputLayers.length;a++){var s=this.outputLayers[a],D=this.outputLayersNodeIndices[a],L=this.outputLayersTensorIndices[a],l=s.name+"_"+D+"_"+L;W.push(l)}for(var a=0;a<W.length;a++){var F=W[a];Wn(F in r),k.push(r[F])}return Mt(k)},t.prototype.runInternalGraph=function(e,i){i==null&&(i=Ji(null,e.length));for(var r={},a=0;a<this.inputs.length;++a){var s=this.inputs[a],o=e[a],l=i[a];r[s.id]=[o,l]}for(var u=Object.keys(this.nodesByDepth).map(function(ee){return parseInt(ee,10)}).sort(to),c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){for(var g=m[f],v=g.outboundLayer,b=g.inputTensors,w=g.outputTensors,S=new Array,L=0,x=b;L<x.length;L++){var s=x[L];s.id in r&&S.push(r[s.id])}if(S.length===b.length){var C={},R=void 0,D=void 0,k=void 0,W=void 0;if(g.callArgs!=null&&(C=g.callArgs),S.length===1){var F=S[0],P=F[0],H=F[1];C.mask==null&&(C.mask=H),k=Je(v.call(P,C)),W=Je(v.computeMask(P,H)),R=[P],D=[H]}else R=S.map(function(ee){return ee[0]}),D=S.map(function(ee){return ee[1]}),C.mask==null&&(C.mask=D),k=Je(v.call(R,C)),W=Je(v.computeMask(R,D));if(v.activityRegularizer)throw new Te("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(var a=0;a<w.length;++a){var s=w[a],o=k[a],l=W[a];r[s.id]=[o,l]}}}for(var _=[],K=[],j=[],q=0,G=this.outputs;q<G.length;q++){var s=G[q];Wn(s.id in r,"Could not compute output "+s.name+" : "+s.id);var Z=r[s.id],X=Z[0],l=Z[1];j.push(X.shape),_.push(X),K.push(l)}return[_,K,j]},t.prototype.buildNodeConversionMap=function(e){for(var i={},r,a=0,s=this.layers;a<s.length;a++){var o=s[a];r=o instanceof t?1:0;for(var l=0;l<o.inboundNodes.length;l++){var u=t.nodeKey(o,l);this.containerNodes.has(u)&&(i[u]=r,r+=1)}}return i},t.prototype.getLayer=function(e,i){if(i!=null){if(this.layers.length<=i)throw new M("Was asked to retrieve layer at index "+i+", but model only "+("has "+this.layers.length+" layer(s)."));return this.layers[i]}else if(e==null)throw new M("Provide either a layer name or layer index");for(var r=0,a=this.layers;r<a.length;r++){var s=a[r];if(s.name===e)return s}throw new M("No such layer: "+e)},t.prototype.calculateLosses=function(){var e=this;return y.tidy(function(){for(var i=[],r=0,a=e.layers;r<a.length;r++)for(var s=a[r],o=0;o<s.inboundNodes.length;++o){var l=t.nodeKey(s,o);e.containerNodes.has(l)&&i.push.apply(i,s.calculateLosses())}return i})},t.prototype.getConfig=function(){for(var e={name:this.name},i=this.buildNodeConversionMap(this.layers),r=[],a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=o.getClassName(),u=o.getConfig(),c=[],h=0;h<o.inboundNodes.length;h++){var d=o.inboundNodes[h],p=t.nodeKey(o,h),f={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),f=d.callArgs}catch(D){console.warn("Layer "+o.name+" was passed non-serializable keyword arguments: "+(d.callArgs+". They will not be included ")+"in the serialized model (and thus will be missing at deserialization time)."),f={}}if(d.inboundLayers.length>0){for(var m=[],g=0;g<d.inboundLayers.length;g++){var v=d.inboundLayers[g],b=d.nodeIndices[g],w=d.tensorIndices[g],S=t.nodeKey(v,b),L=i[S];L==null&&(L=0),m.push([v.name,L,w,f])}c.push(m)}}}var x={};x.name=o.name,x.className=l,x.config=u,x.inboundNodes=c,r.push(x)}e.layers=r;for(var C=[],g=0;g<this.inputLayers.length;g++){var o=this.inputLayers[g],b=this.inputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.inputLayersTensorIndices[g];C.push([o.name,L,w])}e.inputLayers=C;for(var R=[],g=0;g<this.outputLayers.length;g++){var o=this.outputLayers[g],b=this.outputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.outputLayersTensorIndices[g];R.push([o.name,L,w])}return e.outputLayers=R,e},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s={},o={};function l(G,Z){G.name in o?o[G.name].push(Z):o[G.name]=[Z]}function u(G,Z){for(var X=[],ee,ne=0,ie=Z;ne<ie.length;ne++){var te=ie[ne],re=te[0],le=te[1],he=te[2];if(ee=te[3]==null?{}:te[3],!(re in s)){l(G,Z);return}var ye=s[re];if(ye.inboundNodes.length<=le){l(G,Z);return}var Oe=ye.inboundNodes[le];X.push(Oe.outputTensors[he])}X.length>0&&G.apply(Mt(X),ee)}function c(G){var Z=G.name,X=bn(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p<f.length;p++){var m=f[p];c(m)}for(;!YD(o);)for(var g=0,v=d;g<v.length;g++){var m=v[g],b=s[m.name];if(b.name in o){var w=o[b.name];delete o[b.name];for(var S=0,L=w;S<L.length;S++){var x=L[S];u(b,x)}}}for(var C=[],R=[],D=i.inputLayers,k=0,W=D;k<W.length;k++){var m=W[k],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;C.push(_[H])}for(var K=i.outputLayers,j=0,q=K;j<q.length;j++){var m=q[j],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;R.push(_[H])}return new e({inputs:C,outputs:R,name:h})},Object.defineProperty(t.prototype,"stateful",{get:function(){if(this._stateful)throw new M("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(var e=0,i=this.layers;e<i.length;e++){var r=i[e];if(r.stateful)return!0}return!1},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){var e=this;y.tidy(function(){e.layers.forEach(function(i){i.stateful&&i.resetStates()})})},t}(De);function v3(n,t,e){var i=t.length;if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(a){return null});if(i===1)return Array.isArray(n)&&n.length===1?n:typeof n=="object"&&t[0]in n?[n[t[0]]]:[n];if(Array.isArray(n)){if(n.length!==i)throw new Error("Provided "+e+" is an array of "+n.length+" "+("element(s), but the model has "+i+" outputs. ")+"Make sure a set of weights is provided for each model output.");return n}else if(typeof n=="object"&&Object.keys(n).length>0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function gy(n,t){return v3(n,t,"classWeight")}function vy(n,t,e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function y3(n,t){return y.mul(n,t)}var b3=32;function by(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor.  The provided Dataset instead generates "+(""+t)});var a=yy("input",n.inputNames,e),s=yy("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs.  (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs.  (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},u=0;u<a.length;u++)l(u);for(var c=function(d){y.util.assert(s[d].shape[0]===o,function(){return"Batch size mismatch: output "+(n.outputNames[d]+" has "+s[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},h=0;h<s.length;h++)c(h);return{xs:a,ys:s}}function yy(n,t,e){if(e instanceof y.Tensor)return[e];if(Array.isArray(e))return y.util.assert(e.length===t.length,function(){return"Received an array of "+e.length+" Tensors, but expected "+t.length+" to match the "+n+" keys "+t+"."}),e;for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r];if(e[s]==null)throw new M("The feature data generated by the dataset lacks the required "+(n+" key '"+s+"'."));i.push(e[s])}return i}function w3(n){if(n.length===3)throw new Te("Validation with sample weights is not implemented yet.");return{xs:n[0],ys:n[1]}}function L3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k,W,F,P,H,_,K,F,j,q,G,Z,F;return ve(this,function(X){switch(X.label){case 0:if(i=e.batchesPerEpoch!=null,y.util.assert(n.optimizer!=null,function(){return"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."}),y.util.assert(e!=null,function(){return"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."}),y.util.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(wy(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=w3(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=iy(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=ay(h,d,e.epochs,null,null,S3(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g<e.epochs?(b={},[4,f.onEpochBegin(g)]):[3,23];case 5:return X.sent(),w=0,S=0,i?[3,7]:[4,t.iterator()];case 6:v=X.sent(),X.label=7;case 7:return(i?w<e.batchesPerEpoch:!0)?[4,v.next()]:[3,21];case 8:return L=X.sent(),i&&L.done?(console.warn("You provided `batchesPerEpoch` as "+(e.batchesPerEpoch+", ")+"but your dataset iterator ran out of data after "+(w+" batches; ")+"interrupting training. Make sure that your dataset can generate at least `batchesPerEpoch * epochs` batches (in this case, "+(e.batchesPerEpoch*e.epochs+" batches). ")+"You may need to use the repeat() function when building your dataset."),[3,21]):L.value!=null?(x=by(n,L.value),C=x.xs,R=x.ys,D={},D.batch=S,D.size=C[0].shape[0],[4,f.onBatchBegin(S,D)]):[3,15];case 9:if(X.sent(),k=[],!(e.classWeight!=null))return[3,13];W=gy(e.classWeight,n.outputNames),F=0,X.label=10;case 10:return F<W.length?(H=(P=k).push,[4,vy(R[F],null,W[F])]):[3,13];case 11:H.apply(P,[X.sent()]),X.label=12;case 12:return++F,[3,10];case 13:for(_=C.concat(R).concat(k),K=l(_),y.dispose(_),F=0;F<u.length;++F)j=u[F],q=K[F],D[j]=q,y.keep(q);return[4,f.onBatchEnd(S,D)];case 14:X.sent(),Zv(D),S++,w++,X.label=15;case 15:return(i?w>=e.batchesPerEpoch:L.done)?r?(G=void 0,wy(e.validationData)?(Z=Je,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Je(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?b3:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F<n.metricsNames.length;++F)b["val_"+n.metricsNames[F]]=G[F];X.label=19;case 19:return[3,21];case 20:return n.stopTraining_?[3,21]:[3,7];case 21:return[4,f.onEpochEnd(g,b)];case 22:return X.sent(),g++,n.stopTraining_?[3,23]:[3,4];case 23:return[4,f.onTrainEnd()];case 24:return X.sent(),[4,n.history.syncData()];case 25:return X.sent(),[2,n.history];case 26:return n.isTraining=!1,[7];case 27:return[2]}})})}function S3(n,t){var e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(n.size)&&(e=n.size),e}function wy(n){return typeof n.iterator=="function"}function I3(n){return typeof n.next=="function"}function A3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:if(e=e||{},i=e.batches!=null,r=n.testFunction,a=[],e.verbose>0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),I3(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=by(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var x=0;x<L.length;++x)a.push(y.scalar(0));for(var C=S[0].shape[0],R=function(D){var k=L[D],W=a[D];a[D]=y.tidy(function(){return y.add(a[D],y.mul(C,k))}),u>0&&y.dispose(W)},x=0;x<L.length;++x)R(x);y.dispose(L),l+=C,++u}return a}),m.done?(i&&console.warn("Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least `batches` "+("batches (in this case, "+e.batches+" batches). ")+"You may need to use the repeat() function when building your dataset."),[2,"break"]):[2]}})},f.label=4;case 4:return(i?u<e.batches:!0)?[5,c()]:[3,6];case 5:return h=f.sent(),h==="break"?[3,6]:[3,4];case 6:for(d=0;d<a.length;++d)p=a[d],a[d]=y.div(a[d],l),y.dispose(p);return[2,Mt(a)]}})})}function kh(n){y.util.assert(n>0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function Va(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return er(i,t,e-t)}):er(n,t,e-t)}function Fh(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return Fh(e,t)}):Uv(n,t.dtype==="int32"?t:t.toInt())})}function Wh(n,t){for(var e=[],i=0,r=null;i<n;)r=i+t,r>=n&&(r=n),e.push([i,r]),i=r;return e}function T3(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Le(this,void 0,void 0,function(){var m,g,v,b,w,S,L,x,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=vn(0,g)),s==null&&(s=1),b=ay(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(K){switch(K.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(K.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=Wh(g,r),P=function(j){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(j,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[j][0],X=F[j][1],ee=er(W,Z,X-Z);q.batch=j,q.size=X-Z;for(var ne=Fh(e,ee),ie=t(ne),te=0;te<i.length;++te){var re=i[te],le=ie[te];q[re]=le,y.keep(le)}if(j===F.length-1&&m)for(var he=n.testLoop(l,u,r),te=0;te<i.length;++te){var re=i[te],le=he[te];y.keep(le),k["val_"+re]=le}}),[4,w.onBatchEnd(j,q)];case 2:return G.sent(),Zv(q),n.stopTraining_?[2,"break"]:[2]}})},H=0,K.label=3;case 3:return H<F.length?[5,P(H)]:[3,6];case 4:if(_=K.sent(),_==="break")return[3,6];K.label=5;case 5:return++H,[3,3];case 6:W.dispose(),K.label=7;case 7:return[4,w.onEpochEnd(D,k)];case 8:return K.sent(),n.stopTraining_?[2,"break"]:[2]}})},x=d,R.label=2;case 2:return x<a?[5,L(x)]:[3,5];case 3:if(C=R.sent(),C==="break")return[3,5];R.label=4;case 4:return++x,[3,2];case 5:return[4,w.onTrainEnd()];case 6:return R.sent(),[4,n.history.syncData()];case 7:return R.sent(),[2,n.history]}})})}function N3(n,t,e,i){return i===void 0&&(i={}),Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k;return ve(this,function(W){switch(W.label){case 0:if(n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,W.label=1;case 1:return W.trys.push([1,,7,8]),h=i.batchSize==null?32:i.batchSize,kh(h),d=!1,[4,n.standardizeUserData(t,e,i.sampleWeight,i.classWeight,d,h)];case 2:if(p=W.sent(),r=p[0],a=p[1],c=p[2],f=!1,m=void 0,!(i.validationData!=null&&i.validationData.length>0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=Va(r,b,w),r=Va(r,0,b),u=Va(a,b,w),a=Va(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),x=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=x.slice().concat(x.map(function(F){return"val_"+F}))):(C=null,m=[],R=x.slice()),D=iy(i.callbacks,i.yieldEvery),[4,T3(n,L,S,x,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,nr(r,t),nr(a,e),nr(l,s),nr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function Sy(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e<n.length;++e){var i=n[e];if(i.rank===1)t.push(Ua(i,1));else{if(i.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(i)}}return t}function nr(n,t){if(n==null)return;var e=[];if(t instanceof y.Tensor)e.push(t.id);else if(Array.isArray(t))t.forEach(function(l){return e.push(l.id)});else if(t!=null)for(var i in t){var r=t[i];e.push(r.id)}var a=[];if(n instanceof y.Tensor)e.indexOf(n.id)===-1&&a.push(n);else if(Array.isArray(n))n.forEach(function(l){e.indexOf(l.id)===-1&&a.push(l)});else if(n!=null)for(var s in n){var o=n[s];e.indexOf(o.id)===-1&&a.push(o)}a.forEach(function(l){l.isDisposed||l.dispose()})}function x3(n){return n instanceof y.Tensor}function Uh(n){return Array.isArray(n)}function Ly(n){return!x3(n)&&!Uh(n)}function Iy(n,t,e,i,r){if(i===void 0&&(i=!0),r===void 0&&(r=""),t==null||t.length===0){if(n!=null){var a=!1;if(Uh(n)&&n.length>0)a=!0;else if(Ly(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(Ly(n)){n=n,o=[];for(var l=0,u=t;l<u.length;l++){var c=u[l];if(n[c]==null)throw new M('No data provided for "'+c+'". Need data for each key in: '+(""+t));o.push(n[c])}}else if(Uh(n)){if(n=n,n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("model expected. Expected to see "+t.length+" Tensor(s), but ")+("instead got the following list of Tensor(s): "+n));o=n}else{if(n=n,t.length>1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=Sy(o),e!=null)for(var h=0;h<t.length;++h){if(e[h]==null)continue;var d=o[h];if(d.shape.length!==e[h].length)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have "+e[h].length+" dimension(s). but got array with ")+("shape "+d.shape));for(var p=0;p<e[h].length;++p){if(p===0&&!i)continue;var f=d.shape[p],m=e[h][p];if(m!=null&&m>=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function C3(n,t,e){var i=fi(n.map(function(a){return a.shape[0]}));i.sort();var r=fi(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function R3(n,t,e){for(var i=[tr,po,_a],r=0;r<n.length;++r){var a=n[r],s=t[r],o=e[r];if(s==null)continue;if(s===_a&&a.shape[a.shape.length-1]===1)throw new M("You are passing a target array of shape "+a.shape+" while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].");if(i.indexOf(s)!==-1)for(var l=a.shape.slice(1),u=o.slice(1),c=0;c<l.length;++c){var h=l[c],d=u[c];if(d!=null&&h!==d)throw new M("A target Tensor with shape "+a.shape+" was passed for an "+("output of shape "+o+", while using a loss function that ")+"expects targets to have the same shape as the output.")}}}function Ay(n,t,e,i,r){i===void 0&&(i=!0),r===void 0&&(r="");var a;if(Array.isArray(n)){if(n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("the model expected. Expected to see "+t.length+" Tensor(s),")+(" but instead got "+n.length+" Tensors(s)."));a=n}else{if(t.length>1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s<t.length;++s){if(e[s]==null)continue;var o=a[s];if(o.shape.length!==e[s].length)throw new M("Error when checking "+r+": expected "+t[s]+" "+("to have "+e[s].length+" dimension(s), but got array with ")+("shape "+JSON.stringify(o.shape)));for(var l=0;l<e[s].length;++l){if(l===0&&!i)continue;var u=o.shape[l],c=e[s][l];if(c!=null&&c!==u)throw new M("Error when checking "+r+": expected "+(t[s]+" to have shape "+JSON.stringify(e[s])+" but ")+("got array with shape "+JSON.stringify(o.shape)+"."))}}}function O3(n,t){if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(l){return[]});var e;if(typeof n=="string"||typeof n=="function")e=[n];else if(Array.isArray(n)||typeof n=="object")e=n;else throw new TypeError("Type of metrics argument not understood. Expected an string,"+("function, Array, or Object, found: "+n));if(Array.isArray(e))return t.map(function(l){return e});for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r],o=e.hasOwnProperty(s)?e[s]:[];Array.isArray(o)||(o=[o]),i.push(o)}return i}var E3="layers-model",yi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.isTraining=!1,i}return t.prototype.summary=function(e,i,r){if(r===void 0&&(r=console.log),!this.built)throw new M("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");h3(this,e,i,r)},t.prototype.compile=function(e){var i=this;if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=s3(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof y.Optimizer))throw new M("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}var r=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(var a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new M('Unknown entry in loss dictionary: "'+a+'". '+("Only expected the following keys: "+this.outputNames));for(var s=0,o=this.outputNames;s<o.length;s++){var l=o[s];e.loss[l]==null&&console.warn('Output "'+l+'" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data '+("to be passed to "+l+" during training")),r.push(Th(e.loss[l]))}}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new M("When passing an Array as loss, it should have one entry per "+("model output. The model has "+this.outputs.length+" output(s), ")+("but you passed loss="+e.loss+"."));var u=e.loss;r=u.map(function(v){return Th(v)})}else{var c=Th(e.loss);this.outputs.forEach(function(v){r.push(c)})}this.lossFunctions=r,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(var h=0;h<this.outputs.length;++h){var d=this.internalOutputShapes[h],p=this.outputNames[h];this.feedOutputNames.push(p),this.feedOutputShapes.push(d),this.feedLossFns.push(this.lossFunctions[h])}var f=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],Qi("loss",function(){for(var v=0;v<i.outputs.length;++v){if(f.indexOf(v)!==-1)continue;var b=i.lossFunctions[v];i.outputs.length>1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=O3(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};Qi("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(x){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var K=i.internalOutputShapes[w];K[K.length-1]===1||i.lossFunctions[w]===po?["accuracy","acc"].indexOf(_)!==-1?D=Nh:["crossentropy","ce"].indexOf(_)!==-1&&(D=ly):i.lossFunctions[w]===ho?["accuracy","acc"].indexOf(_)!==-1?D=uy:["crossentropy","ce"].indexOf(_)!==-1&&(D=cy):["accuracy","acc"].indexOf(_)!==-1?D=xh:["crossentropy","ce"].indexOf(_)!==-1&&(D=Ch);var j=void 0;["accuracy","acc"].indexOf(_)!==-1?j="acc":["crossentropy","ce"].indexOf(_)!==-1&&(j="ce"),k=D,R=C+j}else{var q=a3(_);k=q,R=C+go(_)}var G;Qi(R,function(){G=k}),g(w,R,G)},F=0,P=x;F<P.length;F++){var H=P[F];W(H)}};L(S)},b=0;b<i.outputs.length;++b)v(b)}),this.collectedTrainableWeights=this.trainableWeights},t.prototype.checkTrainableWeightsConsistency=function(){if(this.collectedTrainableWeights==null)return;this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")},t.prototype.evaluate=function(e,i,r){r===void 0&&(r={});var a=r.batchSize==null?32:r.batchSize;kh(a);var s=!0,o=this.standardizeUserDataXY(e,i,s,a);try{var l=o[0].concat(o[1]);this.makeTestFunction();var u=this.testFunction,c=this.testLoop(u,l,a,r.verbose,r.steps);return Mt(c)}finally{nr(o[0],e),nr(o[1],i)}},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return this.makeTestFunction(),[2,A3(this,e,i)]})})},t.prototype.checkNumSamples=function(e,i,r,a){a===void 0&&(a="steps");var s;if(r!=null){if(s=null,i!=null)throw new M("If "+a+" is set, batchSize must be null or undefined."+("Got batchSize = "+i))}else if(e!=null)Array.isArray(e)?s=e[0].shape[0]:s=e.shape[0];else throw new M("Either the input data should have a defined shape, or "+(a+" shoud be specified."));return s},t.prototype.execute=function(e,i){if(Array.isArray(i)&&i.length===0)throw new M("`outputs` is an empty Array, which is not allowed.");var r=Array.isArray(i),a=r?i:[i],s=this.retrieveSymbolicTensors(a),o=new Gr;if(e instanceof y.Tensor&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new M("The number of inputs provided ("+e.length+") does not match the number of inputs of this model "+("("+this.inputs.length+")."));for(var l=0;l<this.inputs.length;++l)o.add(this.inputs[l],e[l])}else for(var u=0,c=this.inputs;u<c.length;u++){var h=c[u],d=e[h.name];if(d==null)throw new M("No value is provided for the model's input "+h.name);o.add(h,d)}var p=Ha(s,o);return r?p:p[0]},t.prototype.retrieveSymbolicTensors=function(e){for(var i=Ji(null,e.length),r=e.length,a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=Array.isArray(o.output)?o.output:[o.output],u=l.map(function(p){return p.name}),c=0;c<e.length;++c){var h=u.indexOf(e[c]);if(h!==-1&&(i[c]=l[h],r--),r===0)break}if(r===0)break}if(r>0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=Wh(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=Va(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v<m.length;++v)g.push({key:a.inputs[v],value:m[v]});else g.push({key:a.inputs[0],value:m});var b=new Gr(g);return Ha(a.outputs,b)});d.forEach(function(p,f){return l[f].push(p)})},c=0;c<o.length;++c)u(c);return Mt(l.map(function(h){return y.concat(h,0)}))})},t.prototype.predict=function(e,i){i===void 0&&(i={});var r=Sy(e);Ay(r,this.inputNames,this.feedInputShapes,!1);try{var a=i.batchSize==null?32:i.batchSize;return kh(a),this.predictLoop(r,a)}finally{nr(r,e)}},t.prototype.predictOnBatch=function(e){Ay(e,this.inputNames,this.feedInputShapes,!0);var i=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,i)},t.prototype.standardizeUserDataXY=function(e,i,r,a){if(this.optimizer_==null)throw new Fn("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");for(var s=[],o=0;o<this.feedOutputShapes.length;++o){var l=this.feedOutputShapes[o],u=this.feedLossFns[o];u===ho?s.push(l.slice(0,l.length-1).concat([1])):s.push(l)}if(e=Iy(e,this.feedInputNames,this.feedInputShapes,!1,"input"),i=Iy(i,this.feedOutputNames,s,!1,"target"),C3(e,i),R3(i,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Le(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=gy(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p<d.length?(m=(f=h).push,[4,vy(c[p],null,d[p])]):[3,4];case 2:m.apply(f,[g.sent()]),g.label=3;case 3:return++p,[3,1];case 4:return[2,[u,c,h]]}})})},t.prototype.testLoop=function(e,i,r,a,s){var o=this;return a===void 0&&(a=0),y.tidy(function(){var l=o.checkNumSamples(i,r,s,"steps"),u=[];if(a>0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=Wh(l,r),h=y.tensor1d(vn(0,l)),d=0;d<c.length;++d){var p=c[d][0],f=c[d][1],m=er(h,p,f-p),g=Fh(i,m),v=e(g);if(d===0)for(var b=0;b<v.length;++b)u.push(y.scalar(0));for(var b=0;b<v.length;++b){var w=v[b];u[b]=y.add(u[b],y.mul(f-p,w))}}for(var b=0;b<u.length;++b)u[b]=y.div(u[b],l);return u})},t.prototype.getDedupedMetricsNames=function(){for(var e=this.metricsNames,i=[],r=0;r<e.length;++r){var a=e[r],s=a;if(wv(e,a)>1){var o=wv(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f<e.inputs.length;++f)p.push({key:e.inputs[f],value:a[f]});for(var m=new Gr(p),g=Ha(e.outputs,m,{training:!0}),v,f=0;f<e.lossFunctions.length;++f){var b=e.lossFunctions[f],w=b(s[f],g[f]);o[f]!=null&&(w=y3(w,o[f]));var S=y.mean(w);r.push(S),f===0?v=w:v=y.add(v,w)}for(var f=0;f<e.metricsTensors.length;++f){var L=void 0;if(e.outputs.length>1&&f<e.outputs.length)L=r[f];else{var x=e.metricsTensors[f][0],C=e.metricsTensors[f][1];L=y.mean(x(s[C],g[C]))}y.keep(L),l.push(L)}return v=y.mean(v),e.calculateLosses().forEach(function(R){v=y.add(v,R)}),v},c=e.collectedTrainableWeights.map(function(p){return p.read()}),h=!0,d=e.optimizer_.minimize(u,h,c);return[d].concat(l)}},t.prototype.makeTestFunction=function(){var e=this;this.testFunction=function(i){return y.tidy(function(){for(var r=[],a,s=i.slice(0,e.inputs.length),o=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),l=[],u=0;u<e.inputs.length;++u)l.push({key:e.inputs[u],value:s[u]});for(var c=new Gr(l),h=Ha(e.outputs,c),u=0;u<e.lossFunctions.length;++u){var d=e.lossFunctions[u],p=y.mean(d(o[u],h[u]));u===0?a=p:a=y.add(a,p),r.push(a)}for(var u=0;u<e.metricsTensors.length;++u){var f=e.metricsTensors[u][0],m=e.metricsTensors[u][1],g=y.mean(f(o[m],h[m]));r.push(g)}return r})}},t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){return[2,N3(this,e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,L3(this,e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:return[4,this.standardizeUserData(e,i)];case 1:r=f.sent(),a=r[0],s=r[1],o=this.makeTrainFunction(),l=o(a.concat(s)),u=[],c=0,h=l,f.label=2;case 2:return c<h.length?(d=h[c],[4,d.data()]):[3,5];case 3:p=f.sent(),u.push(p[0]),f.label=4;case 4:return c++,[3,2];case 5:return y.dispose(l),[2,Mt(u)]}})})},t.prototype.getNamedWeights=function(e){for(var i=[],r=e!=null&&e.trainableOnly,a=r?this.trainableWeights:this.weights,s=this.getWeights(r),o=0;o<a.length;++o){if(r&&!a[o].trainable)continue;i.push({name:a[o].originalName,tensor:s[o]})}return i},Object.defineProperty(t.prototype,"stopTraining",{get:function(){return this.stopTraining_},set:function(e){this.stopTraining_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.optimizer_},set:function(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)},enumerable:!0,configurable:!0}),t.prototype.dispose=function(){var e=n.prototype.dispose.call(this);if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){var i=y.memory().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=i-y.memory().numTensors}return e},t.prototype.getLossIdentifiers=function(){var e;if(typeof this.loss=="string")e=Jn(this.loss);else if(Array.isArray(this.loss)){for(var i=0,r=this.loss;i<r.length;i++){var a=r[i];if(typeof a!="string")throw new Error("Serialization of non-string loss is not supported.")}e=this.loss.map(function(h){return Jn(h)})}else{var s=Object.keys(this.loss);e={};for(var o=this.loss,l=0,u=s;l<u.length;l++){var c=u[l];if(typeof o[c]=="string")e[c]=Jn(o[c]);else throw new Error("Serialization of non-string loss is not supported.")}}return e},t.prototype.getMetricIdentifiers=function(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[Jn(go(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(function(r){return Jn(go(r))});var e={};for(var i in this.metrics)e[i]=Jn(go(this.metrics[i]));return e},t.prototype.getTrainingConfig=function(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}},t.prototype.loadTrainingConfig=function(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");var i=Ma(e.optimizer_config),r=bn(i),a;if(typeof e.loss=="string")a=Zi(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(function(l){return Zi(l)});else if(e.loss!=null){a={};for(var s in e.loss)a[s]=Zi(e.loss[s])}var o;if(Array.isArray(e.metrics))o=e.metrics.map(function(l){return Zi(l)});else if(e.metrics!=null){o={};for(var s in e.metrics)o[s]=Zi(e.metrics[s])}this.compile({loss:a,metrics:o,optimizer:r})},t.prototype.save=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b;return ve(this,function(w){switch(w.label){case 0:if(typeof e=="string"){if(r=y.io.getSaveHandlers(e),r.length===0)throw new M("Cannot find any save handlers for URL '"+e+"'");if(r.length>1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:E3,generatedBy:"TensorFlow.js tfjs-layers v"+Eh,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,dy(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){dy(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(g3);y.serialization.registerClass(yi);var D3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(yi);y.serialization.registerClass(D3);function k3(n,t){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=Ma(e),r=bn(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o<l.length;o++)u=l[o],s[u.originalName]=a[u.originalName];r.loadWeights(s),y.dispose(a),c.label=2;case 2:return[2,r]}})})}function W3(n,t){return Le(this,void 0,void 0,function(){var e;return ve(this,function(i){if(t==null&&(t={}),typeof n=="string"){if(e=y.io.getLoadHandlers(n,t),e.length===0)e.push(y.io.browserHTTPRequest(n,t));else if(e.length>1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,F3(n,void 0,t)]})})}function F3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=bn(Ma(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=U3(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function U3(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var Bh=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:ao("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r<a.length;r++){var s=a[r];i.add(s)}return i}return t.prototype.checkShape=function(e){var i=e.inboundNodes[0].outputTensors[0].shape;if(i.some(function(r){return r<0}))throw new M("Negative dimension size caused by adding layer "+(e.name+" with input shape [")+(e.inboundNodes[0].inputTensors[0].shape+"]"))},t.prototype.add=function(e){var i=e instanceof t||e instanceof yi,r;if(i){if(r=e,r.outputs.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(r.inputs.length!==1)throw new M("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new M("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");var a=Jv({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(i)this.outputs=r.outputs,this.inputs=r.inputs;else{if(e.inboundNodes.length!==1)throw new M("A layer added to a Sequential model must not already be "+("connected somewhere else. LayersModel received layer "+e.name+" ")+("which has "+e.inboundNodes.length+" pre-existing inbound ")+"connections.");if(e.inboundNodes[0].outputTensors.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Xv(this.outputs[0])}this.inboundNodes=[],new lo({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:Ji(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(function(o){return o.shape}),outputShapes:this.outputs[0].shape})}else{var s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1},t.prototype.pop=function(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{var e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}},t.prototype.call=function(e,i){return this.model==null&&this.build(),this.model.call(e,i)},t.prototype.build=function(e){if(Ke(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new yi({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0},t.prototype.countParams=function(){return this.built||this.build(),n.prototype.countParams.call(this)},t.prototype.summary=function(e,i,r){r===void 0&&(r=console.log),this.built||this.build(),n.prototype.summary.call(this,e,i,r)},t.prototype.setWeights=function(e){this.model==null&&this.build(),this.model.setWeights(e)},t.prototype.evaluate=function(e,i,r){if(r===void 0&&(r={}),!this.built)throw new Fn("The model needs to be compiled before being used.");return this.model.evaluate(e,i,r)},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.evaluateDataset(e,i)]})})},t.prototype.predict=function(e,i){return i===void 0&&(i={}),this.model==null&&this.build(),this.model.predict(e,i)},t.prototype.predictOnBatch=function(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)},t.prototype.compile=function(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames},Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.model==null?void 0:this.model.optimizer},set:function(e){this.model.optimizer=e},enumerable:!0,configurable:!0}),t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fit(e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fitDataset(e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,this.model.trainOnBatch(e,i)]})})},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s,o={};if(i instanceof Array){if(!(i[0].className!=null)||i[0].className==="Merge")throw new M("Legacy serialization format not supported yet.");s=i}else y.util.assert(i.layers!=null,function(){return"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."}),s=i.layers,delete i.layers,o=i;var l=new e(o);if(!(l instanceof t))throw new Te("Sequential.fromConfig called on non-Sequential input: "+l);for(var u=0,c=s;u<c.length;u++){var h=c[u],d=void 0,p=bn(h,d,a);a&&p.setFastWeightInitDuringBuild(!0),l.add(p)}return l},Object.defineProperty(t.prototype,"stopTraining",{get:function(){if(this.model==null)throw new M("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining},set:function(e){if(this.model==null)throw new M("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e},enumerable:!0,configurable:!0}),t.prototype.getConfig=function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i],s={};s.className=a.getClassName(),s.config=a.getConfig(),e.push(s)}return{name:this.name,layers:e}},t.className="Sequential",t}(yi);y.serialization.registerClass(Bh);function B3(n){return new yi(n)}function z3(n){return new Bh(n)}function P3(n,t){return t==null&&(t={}),W3(n,t)}function Ty(n){return Jv(n)}function _3(n,t){ry.registerCallbackConstructor(n,t)}var jt=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),M3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),uk(e,i)},t.className="elu",t}(jt);y.serialization.registerClass(M3);var H3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.selu(e)},t.className="selu",t}(jt);y.serialization.registerClass(H3);var V3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="relu",t}(jt);y.serialization.registerClass(V3);var q3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tidy(function(){return y.minimum(6,y.relu(e))})},t.className="relu6",t}(jt);y.serialization.registerClass(q3);var G3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return e},t.className="linear",t}(jt);y.serialization.registerClass(G3);var Y3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.sigmoid(e)},t.className="sigmoid",t}(jt);y.serialization.registerClass(Y3);var K3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return hk(e)},t.className="hardSigmoid",t}(jt);y.serialization.registerClass(K3);var j3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.softplus(e)},t.className="softplus",t}(jt);y.serialization.registerClass(j3);var $3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return ck(e)},t.className="softsign",t}(jt);y.serialization.registerClass($3);var X3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tanh(e)},t.className="tanh",t}(jt);y.serialization.registerClass(X3);var Ny=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.softmax(e,i)},t.className="softmax",t}(jt);y.serialization.registerClass(Ny);var J3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.logSoftmax(e,i)},t.className="logSoftmax",t}(jt);y.serialization.registerClass(J3);var Z3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),y.tidy(function(){return y.sigmoid(e.mul(i)).mul(e)})},t.className="swish",t}(jt);y.serialization.registerClass(Z3);function bi(n){return n.getClassName()}function zh(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"activation")}function wi(n){if(n==null){var t={};return t.className="linear",t.config={},zh(t)}if(typeof n=="string"){var t={};return t.className=n,t.config={},zh(t)}else return n instanceof jt?n:zh(n)}function Ph(n){if(n!=null&&typeof n!="object")throw new Error("Argument to L1L2 regularizer's constructor is expected to be an "+("object, but received: "+n))}var xy=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(y.serialization.Serializable),yo=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return Ph(e),i.l1=e==null||e.l1==null?.01:e.l1,i.l2=e==null||e.l2==null?.01:e.l2,i.hasL1=i.l1!==0,i.hasL2=i.l2!==0,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=y.zeros([1]);return i.hasL1&&(r=y.add(r,y.sum(y.mul(i.l1,y.abs(e))))),i.hasL2&&(r=y.add(r,y.sum(y.mul(i.l2,Ba(e))))),r.asScalar()})},t.prototype.getConfig=function(){return{l1:this.l1,l2:this.l2}},t.fromConfig=function(e,i){return new e({l1:i.l1,l2:i.l2})},t.className="L1L2",t}(xy);y.serialization.registerClass(yo);function Q3(n){return Ph(n),new yo({l1:n!=null?n.l1:null,l2:0})}function e4(n){return Ph(n),new yo({l2:n!=null?n.l2:null,l1:0})}var Cy={l1l2:"L1L2"};function je(n){return sh(n)}function Ry(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"regularizer")}function nt(n){if(n==null)return null;if(typeof n=="string"){var t=n in Cy?Cy[n]:n,e={className:t,config:{}};return Ry(e)}else return n instanceof xy?n:Ry(n)}var Oy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null&&(i.maxValue=e.maxValue),i}return t.prototype.call=function(e,i){e=xe(e);var r=y.relu(e);return this.maxValue!=null&&(r=y.clipByValue(r,0,this.maxValue)),r},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={maxValue:this.maxValue},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ReLU",t}(De);y.serialization.registerClass(Oy);var Ey=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_ALPHA=.3,e==null&&(e={}),i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.leakyRelu(r,this.alpha)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LeakyReLU",t}(De);y.serialization.registerClass(Ey);var Dy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),i.supportsMasking=!0,i.alphaInitializer=tt(e.alphaInitializer||i.DEFAULT_ALPHA_INITIALIZER),i.alphaRegularizer=nt(e.alphaRegularizer),i.alphaConstraint=bt(e.alphaConstraint),e.sharedAxes==null)i.sharedAxes=null;else if(Array.isArray(e.sharedAxes))i.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")i.sharedAxes=[e.sharedAxes];else throw new M("Expected sharedAxes to be a number or an array of numbers, "+("but got "+e.sharedAxes));return i}return t.prototype.build=function(e){e=Ke(e);var i=e.slice(1);if(this.sharedAxes!=null)for(var r=0,a=this.sharedAxes;r<a.length;r++){var s=a[r];i[s-1]=1}this.alpha=this.addWeight("alpha",i,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);var o={};if(this.sharedAxes!=null)for(var s=1;s<e.length;++s)o[s]=e[s];this.inputSpec=[new Tt({ndim:e.length,axes:o})],this.built=!0},t.prototype.call=function(e,i){return e=xe(e),y.prelu(e,this.alpha.read())},t.prototype.getConfig=function(){var e={alphaInitializer:st(this.alphaInitializer),alphaRegularizer:je(this.alphaRegularizer),alphaConstraint:yt(this.alphaConstraint),sharedAxes:this.sharedAxes},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="PReLU",t}(De);y.serialization.registerClass(Dy);var ky=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==i.DEFAULT_ALPHA)throw new Te("Non-default alpha value ("+e.alpha+") is not supported by the ELU layer yet.");return i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.elu(r)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ELU",t}(De);y.serialization.registerClass(ky);var Fy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_THETA=1,e==null&&(e={}),i.theta=e.theta==null?i.DEFAULT_THETA:e.theta,i}return t.prototype.call=function(e,i){var r=xe(e);return r.mul(Wa(r.greater(this.theta),"float32"))},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={theta:this.theta},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ThresholdedReLU",t}(De);y.serialization.registerClass(Fy);var Wy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_AXIS=1,e==null&&(e={}),i.softmax=new Ny().apply,i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i}return t.prototype.call=function(e,i){var r=xe(e);return this.softmax(r,this.axis)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Softmax",t}(De);y.serialization.registerClass(Wy);function Yr(n,t,e){if(typeof n=="number")return Ji(n,t);if(n.length!==t)throw new M("The "+e+" argument must be an integer or tuple of "+t+" integers."+(" Received: "+n.length+" elements."));for(var i=0;i<t;++i){var r=n[i];if(!ak(r))throw new M("The "+e+" argument must be an integer or tuple of "+t+(" integers. Received: "+JSON.stringify(n)+" including a")+(" non-integer number "+r))}return n}function wn(n,t,e,i,r){if(r===void 0&&(r=1),n==null)return n;var a=t+(t-1)*(r-1),s;return e==="same"?s=n:s=n-a+1,Math.floor((s+i-1)/i)}function bo(n,t,e,i){if(n==null)return null;if(i==="valid")n=n*t+gi([e-t,0]);else if(i==="same")n=n*t;else throw new M("Unsupport padding mode: "+i+".");return n}function _h(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,1]):n})}function Uy(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,4,1]):n})}function t4(n,t,e,i,r,a,s){return i===void 0&&(i=1),r===void 0&&(r="valid"),s===void 0&&(s=1),y.tidy(function(){if(a==null&&(a=gn()),ct(a),n.shape.length!==3)throw new M("The input of a conv1dWithBias operation should be 3, but is "+(n.shape.length+" instead."));if(t.shape.length!==3)throw new M("The kernel for a conv1dWithBias operation should be 3, but is "+(t.shape.length+" instead"));if(e!=null&&e.shape.length!==1)throw new M("The bias for a conv1dWithBias operation should be 1, but is "+(t.shape.length+" instead"));if(a==="channelsFirst"&&(n=y.transpose(n,[0,2,1])),r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");var o=y.conv1d(n,t,i,r==="same"?"same":"valid","NWC",s);return e!=null&&(o=Bn(o,e)),o})}function By(n,t,e,i,r,a,s,o){return i===void 0&&(i=[1,1]),r===void 0&&(r="valid"),o===void 0&&(o=null),y.tidy(function(){if(a==null&&(a=gn()),ct(a),n.rank!==3&&n.rank!==4)throw new M("conv2dWithBiasActivation expects input to be of rank 3 or 4, "+("but received "+n.rank+"."));if(t.rank!==3&&t.rank!==4)throw new M("conv2dWithBiasActivation expects kernel to be of rank 3 or 4, "+("but received "+n.rank+"."));var l=_h(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=y.fused.conv2d({x:l,filter:t,strides:i,pad:r==="same"?"same":"valid",dilations:s,dataFormat:"NHWC",bias:e,activation:o}),a==="channelsFirst"&&(l=y.transpose(l,[0,3,1,2])),l})}function n4(n,t,e,i,r,a,s){return i===void 0&&(i=[1,1,1]),r===void 0&&(r="valid"),y.tidy(function(){if(a==null&&(a=gn()),ct(a),n.rank!==4&&n.rank!==5)throw new M("conv3dWithBias expects input to be of rank 4 or 5, but received "+(n.rank+"."));if(t.rank!==4&&t.rank!==5)throw new M("conv3dWithBias expects kernel to be of rank 4 or 5, but received "+(n.rank+"."));var o=Uy(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=y.conv3d(o,t,i,r==="same"?"same":"valid","NDHWC",s),e!=null&&(o=Bn(o,e)),a==="channelsFirst"&&(o=y.transpose(o,[0,4,1,2,3])),o})}var zy=function(n){Q(t,n);function t(e,i){var r=n.call(this,i)||this;if(r.bias=null,r.DEFAULT_KERNEL_INITIALIZER="glorotNormal",r.DEFAULT_BIAS_INITIALIZER="zeros",t.verifyArgs(i),r.rank=e,At(r.rank,"rank"),r.rank!==1&&r.rank!==2&&r.rank!==3)throw new Te("Convolution layer for rank other than 1, 2, or 3 ("+r.rank+") is not implemented yet.");if(r.kernelSize=Yr(i.kernelSize,e,"kernelSize"),r.strides=Yr(i.strides==null?1:i.strides,e,"strides"),r.padding=i.padding==null?"valid":i.padding,nn(r.padding),r.dataFormat=i.dataFormat==null?"channelsLast":i.dataFormat,ct(r.dataFormat),r.activation=wi(i.activation),r.useBias=i.useBias==null?!0:i.useBias,r.biasInitializer=tt(i.biasInitializer||r.DEFAULT_BIAS_INITIALIZER),r.biasConstraint=bt(i.biasConstraint),r.biasRegularizer=nt(i.biasRegularizer),r.activityRegularizer=nt(i.activityRegularizer),r.dilationRate=Yr(i.dilationRate==null?1:i.dilationRate,e,"dilationRate"),r.rank===1&&Array.isArray(r.dilationRate)&&r.dilationRate.length!==1)throw new M("dilationRate must be a number or an array of a single number for 1D convolution, but received "+(""+JSON.stringify(r.dilationRate)));if(r.rank===2){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==2)throw new M("dilationRate must be a number or array of two numbers for 2D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}else if(r.rank===3){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==3)throw new M("dilationRate must be a number or array of three numbers for 3D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}return r}return t.verifyArgs=function(e){if(Wn("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!lh(e.kernelSize,"number",1,3))throw new M("BaseConv expects config.kernelSize to be number or number[] with "+("length 1, 2, or 3, but received "+JSON.stringify(e.kernelSize)+"."))},t.prototype.getConfig=function(){var e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:bi(this.activation),useBias:this.useBias,biasInitializer:st(this.biasInitializer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),wo=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;return r.kernel=null,t.verifyArgs(i),r.filters=i.filters,At(r.filters,"filters"),r.kernelInitializer=tt(i.kernelInitializer||r.DEFAULT_KERNEL_INITIALIZER),r.kernelConstraint=bt(i.kernelConstraint),r.kernelRegularizer=nt(i.kernelRegularizer),r}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=this.kernelSize.concat([a,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:(i={},i[r]=a,i)}],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a,s=r.bias==null?null:r.bias.read(),o=Lv(r.activation.getClassName());if(o!=null&&r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate,o);else{if(r.rank===1)a=t4(e,r.kernel.read(),s,r.strides[0],r.padding,r.dataFormat,r.dilationRate[0]);else if(r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else if(r.rank===3)a=n4(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else throw new Te("convolutions greater than 3D are not implemented yet.");r.activation!=null&&(a=r.activation.apply(a))}return a})},t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=[],r=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2),a=0;a<r.length;++a){var s=wn(r[a],this.kernelSize[a],this.padding,this.strides[a],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[a]);i.push(s)}var o=[e[0]];return this.dataFormat==="channelsLast"?(o=o.concat(i),o.push(this.filters)):(o.push(this.filters),o=o.concat(i)),o},t.prototype.getConfig=function(){var e={filters:this.filters,kernelInitializer:st(this.kernelInitializer),kernelRegularizer:je(this.kernelRegularizer),kernelConstraint:yt(this.kernelConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.verifyArgs=function(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new M("Convolution layer expected config.filters to be a 'number' > 0 "+("but got "+JSON.stringify(e.filters)))},t}(zy),Mh=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!lh(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(wo);y.serialization.registerClass(Mh);var Py=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(wo);y.serialization.registerClass(Py);var _y=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Tt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Tt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=bo(c,f,d,r.padding),v=bo(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Bn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=bo(i[a],u,o,this.padding),i[s]=bo(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(Mh);y.serialization.registerClass(_y);var i4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=nt(i.depthwiseRegularizer),r.depthwiseConstraint=bt(i.depthwiseConstraint),r.pointwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=nt(i.pointwiseRegularizer),r.pointwiseConstraint=bt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length<this.rank+2)throw new M("Inputs to SeparableConv"+this.rank+"D should have rank "+(this.rank+2+", but received input shape: ")+(""+JSON.stringify(e)));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null||e[r]<0)throw new M("The channel dimension of the inputs should be defined, "+("but found "+JSON.stringify(e[r])));for(var a=e[r],s=this.kernelSize.concat([a,this.depthMultiplier]),o=[],l=0;l<this.rank;++l)o.push(1);o.push(a*this.depthMultiplier,this.filters);var u=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,u,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",o,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,u,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,u,this.biasConstraint):this.bias=null,this.inputSpec=[new Tt({ndim:this.rank+2,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a;if(r.rank===1)throw new Te("1D separable convolution is not implemented yet.");return r.rank===2&&(r.dataFormat==="channelsFirst"&&(e=y.transpose(e,[0,2,3,1])),a=y.separableConv2d(e,r.depthwiseKernel.read(),r.pointwiseKernel.read(),r.strides,r.padding,r.dilationRate,"NHWC")),r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),r.dataFormat==="channelsFirst"&&(a=y.transpose(a,[0,3,1,2])),a})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=st(this.depthwiseInitializer),e.pointwiseInitializer=st(this.pointwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.pointwiseRegularizer=je(this.pointwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseConstraint),e.pointwiseConstraint=yt(this.pointwiseConstraint),e},t.className="SeparableConv",t}(wo),My=function(n){Q(t,n);function t(e){return n.call(this,2,e)||this}return t.className="SeparableConv2D",t}(i4);y.serialization.registerClass(My);var Hy=function(n){Q(t,n);function t(e){var i=n.call(this,1,e)||this;return t.verifyArgs(e),i.inputSpec=[{ndim:3}],i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.dataFormat,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!lh(e.kernelSize,"number",1,1))throw new M("Conv1D expects config.kernelSize to be number or number[] with "+("length 1, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv1D",t}(wo);y.serialization.registerClass(Hy);var Vy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return typeof e.cropping=="number"?i.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?i.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:i.cropping=e.cropping,i.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,i.inputSpec=[{ndim:4}],i}return t.prototype.computeOutputShape=function(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=xe(e),r.dataFormat==="channelsLast"){var a=no(e,r.cropping[0][0],e.shape[1]-r.cropping[0][0]-r.cropping[0][1],2);return no(a,r.cropping[1][0],e.shape[2]-r.cropping[1][1]-r.cropping[1][0],3)}else{var a=no(e,r.cropping[0][0],e.shape[2]-r.cropping[0][0]-r.cropping[0][1],3);return no(a,r.cropping[1][0],e.shape[3]-r.cropping[1][1]-r.cropping[1][0],4)}})},t.prototype.getConfig=function(){var e={cropping:this.cropping,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Cropping2D",t}(De);y.serialization.registerClass(Vy);var qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_SIZE=[2,2],i.inputSpec=[{ndim:4}],i.size=e.size==null?i.DEFAULT_SIZE:e.size,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,i}return t.prototype.computeOutputShape=function(e){if(this.dataFormat==="channelsFirst"){var i=e[2]==null?null:this.size[0]*e[2],r=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],i,r]}else{var i=e[1]==null?null:this.size[0]*e[1],r=e[2]==null?null:this.size[1]*e[2];return[e[0],i,r,e[3]]}},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e),s=a.shape;if(r.dataFormat==="channelsFirst"){a=y.transpose(a,[0,2,3,1]);var o=r.size[0]*s[2],l=r.size[1]*s[3],u=a.resizeNearestNeighbor([o,l]);return y.transpose(u,[0,3,1,2])}else{var o=r.size[0]*s[1],l=r.size[1]*s[2];return a.resizeNearestNeighbor([o,l])}})},t.prototype.getConfig=function(){var e={size:this.size,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="UpSampling2D",t}(De);y.serialization.registerClass(qy);function r4(n,t,e,i,r,a){return e===void 0&&(e=[1,1]),i===void 0&&(i="valid"),y.tidy(function(){r==null&&(r=gn()),ct(r);var s=_h(n,r);if(n.rank!==4)throw new M("Input for depthwiseConv2d is required to be 4-D, but is instead "+(n.rank+"-D"));if(t.rank!==4)throw new M("depthwiseKernel is required to be 4-D, but is instead "+(t.rank+"-D"));return s=y.depthwiseConv2d(s,t,e,i==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}var Gy=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return i.depthwiseKernel=null,i.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,i.depthwiseInitializer=tt(e.depthwiseInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.depthwiseConstraint=bt(e.depthwiseConstraint),i.depthwiseRegularizer=nt(e.depthwiseRegularizer),i}return t.prototype.build=function(e){if(e=Ke(e),e.length<4)throw new M("Inputs to DepthwiseConv2D should have rank 4. "+("Received input shape: "+JSON.stringify(e)+"."));var i=this.dataFormat==="channelsFirst"?1:3;if(e[i]==null||e[i]<0)throw new M("The channel dimension of the inputs to DepthwiseConv2D should "+("be defined, but is not ("+e[i]+")."));var r=e[i],a=[this.kernelSize[0],this.kernelSize[1],r,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[r*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=r4(e,r.depthwiseKernel.read(),r.strides,r.padding,r.dataFormat,null);return r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),a})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,s=wn(i,this.kernelSize[0],this.padding,this.strides[0]),o=wn(r,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,s,o]:[e[0],s,o,a]},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=st(this.depthwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseRegularizer),e},t.className="DepthwiseConv2D",t}(zy);y.serialization.registerClass(Gy);function Yy(n,t,e,i){if(Array.isArray(n)){if(t!=null||e!=null)throw new M("When inputs is an array, neither initialState or constants should be provided");i!=null&&(e=n.slice(n.length-i,n.length),n=n.slice(0,n.length-i)),n.length>1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Ky(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(vn(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],x=y.tidy(function(){return n(L,d)});if(r==null)h=x[0],d=x[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=x[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return x[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v<p;++v)g(v);var b;if(o){var w=1;b=y.stack(c,w)}return[h,b,d]})}var Si=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r;if(e.cell==null)throw new M("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?r=new Hh({cells:e.cell}):r=e.cell,r.stateSize==null)throw new M("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");return i.cell=r,i.returnSequences=e.returnSequences==null?!1:e.returnSequences,i.returnState=e.returnState==null?!1:e.returnState,i.goBackwards=e.goBackwards==null?!1:e.goBackwards,i._stateful=e.stateful==null?!1:e.stateful,i.unroll=e.unroll==null?!1:e.unroll,i.supportsMasking=!0,i.inputSpec=[new Tt({ndim:3})],i.stateSpec=null,i.states_=null,i.numConstants=null,i.keptStates=[],i}return t.prototype.getStates=function(){if(this.states_==null){var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return vn(0,e).map(function(i){return null})}else return this.states_},t.prototype.setStates=function(e){this.states_=e},t.prototype.computeOutputShape=function(e){Sh(e)&&(e=e[0]),e=e;var i=this.cell.stateSize;Array.isArray(i)||(i=[i]);var r=i[0],a;if(this.returnSequences?a=[e[0],e[1],r]:a=[e[0],r],this.returnState){for(var s=[],o=0,l=i;o<l.length;o++){var u=l[o];s.push([e[0],u])}return[a].concat(s)}else return a},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){Array.isArray(i)&&(i=i[0]);var a=r.returnSequences?i:null;if(r.returnState){var s=r.states.map(function(o){return null});return[a].concat(s)}else return a})},Object.defineProperty(t.prototype,"states",{get:function(){if(this.states_==null){for(var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,i=[],r=0;r<e;++r)i.push(null);return i}else return this.states_},set:function(e){this.states_=e},enumerable:!0,configurable:!0}),t.prototype.build=function(e){if(this.numConstants!=null)throw new Te("Constants support is not implemented in RNN yet.");Sh(e)&&(e=e[0]),e=e;var i=this.stateful?e[0]:null,r=e.slice(2);this.inputSpec[0]=new Tt({shape:[i,null].concat(r)});var a=[e[0]].concat(e.slice(2));this.cell.build(a);var s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!y.util.arraysEqual(this.stateSpec.map(function(o){return o.shape[o.shape.length-1]}),s))throw new M("An initialState was passed that is not compatible with "+("cell.stateSize. Received stateSpec="+this.stateSpec+"; ")+("However cell.stateSize is "+this.cell.stateSize))}else this.stateSpec=s.map(function(o){return new Tt({shape:[null,o]})});this.stateful&&this.resetStates()},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape[0];if(a==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.states_==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_=[y.zeros([a,r.cell.stateSize])];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_[0]=y.zeros([a,r.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i===!0?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var s=0;s<r.states_.length;++s){var o=e[s],l=Array.isArray(r.cell.stateSize)?r.cell.stateSize[s]:r.cell.stateSize,u=[a,l];if(!y.util.arraysEqual(o.shape,u))throw new M("State "+s+" is incompatible with layer "+r.name+": "+("expected shape="+u+", received shape="+o.shape));r.states_[s]=o}}r.states_=r.states_.map(function(c){return y.keep(c.clone())})})},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);e=s.inputs,r=s.initialState,a=s.constants;var o=[],l=[];if(r!=null){i.initialState=r,o=o.concat(r),this.stateSpec=[];for(var u=0,c=r;u<c.length;u++){var h=c[u];this.stateSpec.push(new Tt({shape:h.shape}))}l=l.concat(this.stateSpec)}a!=null&&(i.constants=a,o=o.concat(a),this.numConstants=a.length);var d=o[0]instanceof yn;if(d){var p=[e].concat(o),f=this.inputSpec.concat(l),m=this.inputSpec;this.inputSpec=f;var g=n.prototype.apply.call(this,p,i);return this.inputSpec=m,g}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;e=xe(e),o==null&&(r.stateful?o=r.states_:o=r.getInitialState(e));var l=Array.isArray(r.cell.stateSize)?r.cell.stateSize.length:1;if(o.length!==l)throw new M("RNN Layer has "+l+" state(s) but was passed "+(o.length+" initial state(s)."));r.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");var u={training:s},c=function(g,v){var b=r.cell.call([g].concat(v),u);return[b[0],b.slice(1)]},h=Ky(c,e,o,r.goBackwards,a,null,r.unroll,r.returnSequences),d=h[0],p=h[1],f=h[2];r.stateful&&r.resetStates(f,s);var m=r.returnSequences?p:d;return r.returnState?[m].concat(f):m})},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=y.zeros(e.shape);return r=y.sum(r,[1,2]),r=Ua(r),Array.isArray(i.cell.stateSize)?i.cell.stateSize.map(function(a){return a>1?ch(r,[1,a]):r}):i.cell.stateSize>1?[ch(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),Yt({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=bn(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Si);var Kr=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),Vh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:s})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(a)},rate:r.recurrentDropout,training:s}));var o,l=r.dropoutMask,u=r.recurrentDropoutMask;l!=null?o=Un(y.mul(e,l),r.kernel.read()):o=Un(e,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),u!=null&&(a=y.mul(a,u));var c=y.add(o,Un(a,r.recurrentKernel.read()));return r.activation!=null&&(c=r.activation.apply(c)),[c,c]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Yt({},e,i)},t.className="SimpleRNNCell",t}(Kr);y.serialization.registerClass(Vh);var jy=function(n){Q(t,n);function t(e){var i=this;return e.cell=new Vh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return new e(i)},t.className="SimpleRNN",t}(Si);y.serialization.registerClass(jy);var qh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new M("GRUCell does not support reset_after parameter set to true.");return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e);var i=e[e.length-1];this.kernel=this.addWeight("kernel",[i,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("GRUCell expects 2 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training==null?!1:i.training,s=e[1];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:3})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:3}));var o=r.dropoutMask,l=r.recurrentDropoutMask,u,c,h;0<r.dropout&&r.dropout<1&&(e=y.mul(e,o[0]));var d=Un(e,r.kernel.read());r.useBias&&(d=Bn(d,r.bias.read())),0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,l[0]));var p=r.recurrentKernel.read(),f=y.split(p,[2*r.units,r.units],p.rank-1),m=f[0],g=f[1],v=Un(s,m),b=y.split(d,3,d.rank-1),w=b[0],S=b[1],L=b[2],x=y.split(v,2,v.rank-1),C=x[0],R=x[1];u=r.recurrentActivation.apply(y.add(w,C)),c=r.recurrentActivation.apply(y.add(S,R));var D=Un(y.mul(c,s),g);h=r.activation.apply(y.add(L,D));var k=y.add(y.mul(u,s),y.mul(y.add(1,y.neg(u)),h));return[k,k]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Yt({},e,i)},t.className="GRUCell",t}(Kr);y.serialization.registerClass(qh);var $y=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new qh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="GRU",t}(Si);y.serialization.registerClass($y);var So=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.unitForgetBias=e.unitForgetBias,i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=[i.units,i.units],i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel=this.addWeight("kernel",[r,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);var a;if(this.useBias){if(this.unitForgetBias){var s=this.biasInitializer,o=this.units;a=new(i=function(l){Q(u,l);function u(){return l!==null&&l.apply(this,arguments)||this}return u.prototype.apply=function(c,h){var d=s.apply([o]),p=new fh().apply([o]),f=s.apply([o*2]);return Wv(Wv(d,p),f)},u}(cn),i.className="CustomInit",i)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training;if(e=e,e.length!==3)throw new M("LSTMCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var s=e[1],o=e[2];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:4})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:4}));var l=r.dropoutMask,u=r.recurrentDropoutMask,c,h,d,p;0<r.dropout&&r.dropout<1&&(e=y.mul(e,l[0]));var f=Un(e,r.kernel.read());0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,u[0])),f=y.add(f,Un(s,r.recurrentKernel.read())),r.useBias&&(f=Bn(f,r.bias.read()));var m=y.split(f,4,f.rank-1),g=m[0],v=m[1],b=m[2],w=m[3];c=r.recurrentActivation.apply(g),h=r.recurrentActivation.apply(v),d=y.add(y.mul(h,o),y.mul(c,r.activation.apply(b))),p=r.recurrentActivation.apply(w);var S=y.mul(p,r.activation.apply(d));return[S,S,d]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Yt({},e,i)},t.className="LSTMCell",t}(Kr);y.serialization.registerClass(So);var Xy=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new So(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="LSTM",t}(Si);y.serialization.registerClass(Xy);var Hh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.cells=e.cells,i}return Object.defineProperty(t.prototype,"stateSize",{get:function(){for(var e=[],i=0,r=this.cells.slice().reverse();i<r.length;i++){var a=r[i];Array.isArray(a.stateSize)?e.push.apply(e,a.stateSize):e.push(a.stateSize)}return e},enumerable:!0,configurable:!0}),t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=e;for(var a=e.slice(1),s=[],o=0,l=r.cells.slice().reverse();o<l.length;o++){var u=l[o];Array.isArray(u.stateSize)?s.push(a.splice(0,u.stateSize.length)):s.push(a.splice(0,1))}s.reverse();for(var c=[],h,d=0;d<r.cells.length;++d){var u=r.cells[d];a=s[d],d===0?h=[e[0]].concat(a):h=[h[0]].concat(a),h=u.call(h,i),c.push(h.slice(1))}a=[];for(var p=0,f=c.slice().reverse();p<f.length;p++){var m=f[p];a.push.apply(a,m)}return[h[0]].concat(a)})},t.prototype.build=function(e){Sh(e)&&(e=e[0]),e=e;var i;this.cells.forEach(function(r,a){Qi("RNNCell_"+a,function(){r.build(e),Array.isArray(r.stateSize)?i=r.stateSize[0]:i=r.stateSize,e=[e[0],i]})}),this.built=!0},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=function(s){return{className:s.getClassName(),config:s.getConfig()}},r=this.cells.map(i),a={cells:r};return Yt({},e,a)},t.fromConfig=function(e,i,r){r===void 0&&(r={});for(var a=[],s=0,o=i.cells;s<o.length;s++){var l=o[s];a.push(bn(l,r))}return new e({cells:a})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(!this.trainable)return[];for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.cells;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.weights)}return Lh(e)},t.prototype.setWeights=function(e){for(var i=[],r=0,a=this.cells;r<a.length;r++)for(var s=a[r],o=s.weights.length,l=e.splice(o),u=0;u<s.weights.length;++u)i.push([s.weights[u],l[u]]);Ih(i)},t.className="StackedRNNCells",t}(Kr);y.serialization.registerClass(Hh);function Li(n){var t=n.ones,e=n.rate,i=n.training,r=i===void 0?!1:i,a=n.count,s=a===void 0?1:a,o=function(){return Bv(t(),e)},l=function(){return za(o,t,r)};if(!s||s<=1)return y.keep(l().clone());var u=Array(s).fill(void 0).map(l);return u.map(function(c){return y.keep(c.clone())})}var LH=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(Kr),a4=function(n){Q(t,n);function t(e){var i=this;if(e.unroll)throw new Te("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Te("It is not possible at the moment to stack convolutional cells.");return i=n.call(this,e)||this,i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null),i&&i.constants)throw new M("ConvRNN2D cell does not support constants");var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.prototype.computeOutputShape=function(e){var i=this.computeSingleOutputShape(e);return this.returnSequences||(i=[i[0]].concat(i.slice(2))),this.returnState&&(i=[i].concat(Array(2).fill([e[0]].concat(i.slice(-3))))),i},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=i.cell.stateSize,a=e.shape,s=i.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=y.zeros(o);return Array.isArray(r)?Array(r.length).fill(l):[l]})},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape,s=r.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=a[0];if(l==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.getStates()==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_=[y.zeros(o)];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_[0]=y.zeros(o);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var u=0;u<r.states_.length;++u){var c=e[u],h=o;if(!y.util.arraysEqual(c.shape,h))throw new M("State "+u+" is incompatible with layer "+r.name+": "+("expected shape="+h+", received shape="+c.shape));r.states_[u]=c}}r.states_=r.states_.map(function(d){return y.keep(d.clone())})})},t.prototype.computeSingleOutputShape=function(e){var i=this.cell,r=i.dataFormat,a=i.filters,s=i.kernelSize,o=i.padding,l=i.strides,u=i.dilationRate,c=r==="channelsFirst",h=e[c?3:2],d=e[c?4:3],p=wn(h,s[0],o,l[0],u[0]),f=wn(d,s[1],o,l[1],u[1]),m=e.slice(0,2).concat(c?[a,p,f]:[p,f,a]);return m},t.className="ConvRNN2D",t}(Si),Gh=function(n){Q(t,n);function t(e){var i=this,r=e.filters,a=e.kernelSize,s=e.strides,o=e.padding,l=e.dataFormat,u=e.dilationRate;return i=n.call(this,Yt({},e,{units:r}))||this,i.filters=r,At(i.filters,"filters"),i.kernelSize=Yr(a,2,"kernelSize"),i.kernelSize.forEach(function(c){return At(c,"kernelSize")}),i.strides=Yr(s||1,2,"strides"),i.strides.forEach(function(c){return At(c,"strides")}),i.padding=o||"valid",nn(i.padding),i.dataFormat=l||"channelsLast",ct(i.dataFormat),i.dilationRate=Yr(u||1,2,"dilationRate"),i.dilationRate.forEach(function(c){return At(c,"dilationRate")}),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=4,o=this.kernelSize.concat([a,this.filters*s]);this.kernel=this.addWeight("kernel",o,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);var l=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",l,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){var u=void 0;if(this.unitForgetBias){var c=this.biasInitializer,h=this.filters;u=new(i=function(d){Q(p,d);function p(){return d!==null&&d.apply(this,arguments)||this}return p.prototype.apply=function(f,m){var g=c.apply([h]),v=y.ones([h]),b=c.apply([h*2]);return dh([g,v,b])},p}(cn),i.className="CustomInit",i)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e.length!==3)throw new M("ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training||!1,s=e[0],o=e[1],l=e[2],u=4;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.dropout,training:a,count:u}));var c=r.dropoutMask,h=function(te,re,le){return!re||!re[le]?te:y.mul(re[le],te)},d=h(s,c,0),p=h(s,c,1),f=h(s,c,2),m=h(s,c,3);0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(o)},rate:r.recurrentDropout,training:a,count:u}));var g=r.recurrentDropoutMask,v=h(o,g,0),b=h(o,g,1),w=h(o,g,2),S=h(o,g,3),L=3,x=y.split(r.kernel.read(),u,L),C=x[0],R=x[1],D=x[2],k=x[3],W=r.useBias?y.split(r.bias.read(),u):[null,null,null,null],F=W[0],P=W[1],H=W[2],_=W[3];d=r.inputConv(d,C,F,r.padding),p=r.inputConv(p,R,P,r.padding),f=r.inputConv(f,D,H,r.padding),m=r.inputConv(m,k,_,r.padding);var K=y.split(r.recurrentKernel.read(),u,L),j=K[0],q=K[1],G=K[2],Z=K[3];v=r.recurrentConv(v,j),b=r.recurrentConv(b,q),w=r.recurrentConv(w,G),S=r.recurrentConv(S,Z);var X=r.recurrentActivation.apply(y.add(d,v)),ee=r.recurrentActivation.apply(y.add(p,b)),ne=y.add(y.mul(ee,l),y.mul(X,r.activation.apply(y.add(f,w)))),ie=y.mul(r.recurrentActivation.apply(y.add(m,S)),r.activation.apply(ne));return[ie,ie,ne]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=e.units,r=VD(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Yt({},r,a)},t.prototype.inputConv=function(e,i,r,a){var s=y.conv2d(e,i,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return r?Bn(s,r,this.dataFormat):s},t.prototype.recurrentConv=function(e,i){var r=1;return y.conv2d(e,i,r,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")},t.className="ConvLSTM2DCell",t}(So);y.serialization.registerClass(Gh);var Jy=function(n){Q(t,n);function t(e){var i=this,r=new Gh(e);return i=n.call(this,Yt({},e,{cell:r}))||this,i}return t.fromConfig=function(e,i){return new e(i)},t.className="ConvLSTM2D",t}(a4);y.serialization.registerClass(Jy);var Yh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.rate=Math.max(Math.min(e.rate,1),0),i.noiseShape=e.noiseShape,i.seed=e.seed,i.supportsMasking=!0,i}return t.prototype.getNoiseShape=function(e){if(this.noiseShape==null)return this.noiseShape;for(var i=e.shape,r=[],a=0;a<this.noiseShape.length;++a)r.push(this.noiseShape[a]==null?i[a]:this.noiseShape[a]);return r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(0<r.rate&&r.rate<1){var s=i.training==null?!1:i.training,o=r.getNoiseShape(a),l=za(function(){return Bv(a,r.rate,o,r.seed)},function(){return a},s);return l}return e})},t.prototype.getConfig=function(){var e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.dispose=function(){return n.prototype.dispose.call(this)},t.className="Dropout",t}(De);y.serialization.registerClass(Yh);var Zy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[{ndim:3}],i}return t.prototype.getNoiseShape=function(e){var i=e.shape;return[i[0],1,i[2]]},t.className="SpatialDropout1D",t}(Yh);y.serialization.registerClass(Zy);var Qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.activation=null,i.useBias=!0,i.kernel=null,i.bias=null,i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){var r=null;e.batchSize!=null&&(r=e.batchSize),i.batchInputShape=[r,e.inputDim]}return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation),e.useBias!=null&&(i.useBias=e.useBias),i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelConstraint=bt(e.kernelConstraint),i.biasConstraint=bt(e.biasConstraint),i.kernelRegularizer=nt(e.kernelRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.supportsMasking=!0,i.inputSpec=[{minNDim:2}],i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[r,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:(i={},i[-1]=r,i)}],this.built=!0},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return i[i.length-1]=this.units,i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=Lv(r.activation.getClassName()),o;return s!=null?o=Un(a,r.kernel.read(),s,r.bias?r.bias.read():null):(o=Un(a,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),r.activation!=null&&(o=r.activation.apply(o))),o})},t.prototype.getConfig=function(){var e={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dense",t}(De);y.serialization.registerClass(Qy);var eb=function(n){Q(t,n);function t(e){var i=this;return e=e||{},i=n.call(this,e)||this,i.inputSpec=[{minNDim:3}],i.dataFormat=e.dataFormat,i}return t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=0,r=e.slice(1);i<r.length;i++){var a=r[i];if(a==null)throw new M('The shape of the input to "Flatten" is not fully defined '+("(got "+e.slice(1)+"). Make sure to pass a complete ")+'"input_shape" or "batch_input_shape" argument to the first layer in your model.')}return[e[0],mi(e,1)]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.dataFormat==="channelsFirst"&&a.rank>1){for(var s=[0],o=2;o<a.rank;++o)s.push(o);s.push(1),a=a.transpose(s)}return lk(a)})},t.prototype.getConfig=function(){var e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);var i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Flatten",t}(De);y.serialization.registerClass(eb);var tb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.activation=wi(e.activation),i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);return r.activation.apply(a)})},t.prototype.getConfig=function(){var e={activation:bi(this.activation)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Activation",t}(De);y.serialization.registerClass(tb);var nb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.n=e.n,i.inputSpec=[{ndim:2}],i}return t.prototype.computeOutputShape=function(e){return[e[0],this.n,e[1]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return e=xe(e),sk(e,r.n)})},t.prototype.getConfig=function(){var e={n:this.n},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="RepeatVector",t}(De);y.serialization.registerClass(nb);var ib=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;i.targetShape=e.targetShape;for(var r=0;r<i.targetShape.length;++r)i.isUnknown(i.targetShape[r])&&(i.targetShape[r]=null);return i}return t.prototype.isUnknown=function(e){return e<0||e==null},t.prototype.fixUnknownDimension=function(e,i){for(var r="Total size of new array must be unchanged.",a=i.slice(),s=1,o=null,l=0;l<a.length;++l){var u=a[l];if(this.isUnknown(u))if(o===null)o=l;else throw new M("Can only specifiy one unknown dimension.");else s*=u}var c=mi(e);if(o!==null){if(s===0||c%s!==0)throw new M(r);a[o]=c/s}else if(c!==s)throw new M(r);return a},t.prototype.computeOutputShape=function(e){for(var i=!1,r=0;r<e.length;++r)if(this.isUnknown(e[r])){i=!0;break}return i?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=a.shape,o=s.slice(0,1).concat(r.fixUnknownDimension(s.slice(1),r.targetShape));return a.reshape(o)})},t.prototype.getConfig=function(){var e={targetShape:this.targetShape},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Reshape",t}(De);y.serialization.registerClass(ib);var rb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error("Permute constructor requires `dims` to be an Array, but received "+(e.dims+" instead."));var r=vn(1,e.dims.length+1);if(!y.util.arraysEqual(e.dims.slice().sort(),r))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");return i.dims=e.dims,i.dimsIncludingBatch=[0].concat(i.dims),i.inputSpec=[new Tt({ndim:i.dims.length+1})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return this.dims.forEach(function(r,a){i[a+1]=e[r]}),i},t.prototype.call=function(e,i){return y.transpose(xe(e),this.dimsIncludingBatch)},t.prototype.getConfig=function(){var e={dims:this.dims},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Permute",t}(De);y.serialization.registerClass(rb);var ab=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null?i.maskValue=e.maskValue==null?0:e.maskValue:i.maskValue=0,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={maskValue:this.maskValue};return Object.assign(i,e),i},t.prototype.computeMask=function(e,i){var r=xe(e),a=-1;return y.any(y.notEqual(r,this.maskValue),a)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=-1,o=!0,l=y.any(y.notEqual(a,r.maskValue),s,o),u=a.mul(l.asType(a.dtype));return u})},t.className="Masking",t}(De);y.serialization.registerClass(ab);var sb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.embeddings=null,i.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){var r=null;e.batchSize!=null&&(r=e.batchSize),e.inputLength==null?i.batchInputShape=[r,null]:i.batchInputShape=[r].concat(Je(e.inputLength))}return i.inputDim=e.inputDim,At(i.inputDim,"inputDim"),i.outputDim=e.outputDim,At(i.outputDim,"outputDim"),i.embeddingsInitializer=tt(e.embeddingsInitializer||i.DEFAULT_EMBEDDINGS_INITIALIZER),i.embeddingsRegularizer=nt(e.embeddingsRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.embeddingsConstraint=bt(e.embeddingsConstraint),i.maskZero=e.maskZero,i.supportsMasking=e.maskZero,i.inputLength=e.inputLength,i}return t.prototype.build=function(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0},t.prototype.warnOnIncompatibleInputShape=function(e){},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){return r.maskZero?(e=xe(e),y.notEqual(e,y.zerosLike(e))):null})},t.prototype.computeOutputShape=function(e){if(e=Ke(e),this.inputLength==null)return e.concat([this.outputDim]);var i=Je(this.inputLength);if(i.length!==e.length-1)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));for(var r=0,a=0;a<i.length;++a){var s=i[a],o=e[a+1];if(s!=null&&o!=null&&s!==o)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));s==null&&(i[r]=o),r++}return[e[0]].concat(i,[this.outputDim])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);a.dtype!=="int32"&&(a=Wa(a,"int32"));var s=Uv(r.embeddings.read(),a.as1D());return s.reshape(Ke(r.computeOutputShape(a.shape)))})},t.prototype.getConfig=function(){var e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:st(this.embeddingsInitializer),embeddingsRegularizer:je(this.embeddingsRegularizer),activityRegularizer:je(this.activityRegularizer),embeddingsConstraint:yt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Embedding",t}(De);y.serialization.registerClass(sb);var ir=function(n){Q(t,n);function t(e){var i=n.call(this,e||{})||this;return i.supportsMasking=!0,i}return t.prototype.mergeFunction=function(e){throw new Te},t.prototype.computeElementwiseOpOutputShape=function(e,i){if(e==null||i==null)return null;if(e.length<i.length)return this.computeElementwiseOpOutputShape(i,e);if(i.length===0)return e;for(var r=e.slice(0,e.length-i.length),a=0;a<i.length;++a){var s=e[e.length-i.length+a],o=i[a];if(s==null||o==null||s<0||o<0)r.push(null);else if(s===1)r.push(o);else if(o===1)r.push(s);else{if(s!==o)throw new M("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(i));r.push(s)}}return r},t.prototype.build=function(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[Ke(e)]),e=e,e.length<2)throw new M("A merge layer should be called on an Array of at least 2 inputs."+(" Got "+e.length+" input(s)."));for(var i=[],r=0,a=e;r<a.length;r++){var s=a[r];s!=null&&s[0]!==null&&i.push(s[0])}if(i=fi(i),i.length>1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l<e.length;++l){var s=e[l]==null?null:e[l].slice(1);o=this.computeElementwiseOpOutputShape(o,s)}var u=e.map(function(c){return c.length});e.indexOf(null)===-1&&fi(u).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,r.reshapeRequired){var a=[],s=e.map(function(D){return D.rank});if(s.indexOf(null)===-1){for(var o=gi(s),l=0,u=e;l<u.length;l++){for(var c=u[l],h=c.rank,d=0;d<o-h;++d)c=Ua(c,1);a.push(c)}return r.mergeFunction(a)}else{for(var p=!1,f=0,m=e;f<m.length;f++){var c=m[f],h=c.rank;if(h==null){var g=c.shape,v=g[0],b=g.slice(1).concat([v]),w=c.reshape([v].concat(mi(g.slice(1))));w=y.transpose(w,[1,0]),w=w.reshape(b),a.push(w),p=!0}else if(h>1){var S=vn(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),x=L.rank;if(p){if(x==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(x>1){var S=[x-1].concat(vn(0,x-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r<e.length;++r){var a=e[r]==null?null:e[r].slice(1);i=this.computeElementwiseOpOutputShape(i,a)}for(var s=[],o=0,l=e;o<l.length;o++){var a=l[o];a!=null&&a[0]!==null&&s.push(a[0])}return s=fi(s),s.length===1?i=s.concat(i):i=[null].concat(i),i},t.prototype.computeMask=function(e,i){return y.tidy(function(){if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an Array");if(!Array.isArray(e))throw new M("`inputs` should be an Array");if(i.length!==e.length)throw new M("The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths "+("("+e.length+" vs "+i.length+")"));if(i.every(function(s){return s==null}))return null;i=i.map(function(s){return s==null?s:y.expandDims(s,0)});for(var r=i[0],a=1;a<i.length-1;++a)r=y.logicalAnd(r,i[a]);return r})},t}(De),ob=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return i})},t.className="Add",t}(ir);y.serialization.registerClass(ob);var lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.mul(i,e[r]);return i})},t.className="Multiply",t}(ir);y.serialization.registerClass(lb);var ub=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return y.mul(1/e.length,i)})},t.className="Average",t}(ir);y.serialization.registerClass(ub);var cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.maximum(i,e[r]);return i})},t.className="Maximum",t}(ir);y.serialization.registerClass(cb);var hb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.minimum(i,e[r]);return i})},t.className="Minimum",t}(ir);y.serialization.registerClass(hb);var db=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_AXIS=-1,e==null&&(e={}),i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new M("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;for(var i=!0,r=0,a=e;r<a.length;r++){var s=a[r];if(s!=null){i=!1;break}}if(i)return;for(var o=[],l=0;l<e.length;++l){var u=e[l].slice();u.splice(this.axis,1);for(var c=!1,h=0,d=o;h<d.length;h++){var s=d[h];if(y.util.arraysEqual(s,u)){c=!0;break}}c||o.push(u)}if(o.length>1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return dh(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s<o.length;s++){var l=o[s];if(r[a]==null||l[a]==null){r[a]=null;break}r[a]+=l[a]}return r},t.prototype.computeMask=function(e,i){var r=this;if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new M("`inputs` should be an array for Concatenate");if(i.length!==e.length)throw new M("Mismatch in the length of mask ("+i.length+") "+("and the legnth of inputs ("+e.length+")"));return y.tidy(function(){var a=!0;if(i.forEach(function(u){if(u!=null){a=!1;return}}),a)return null;for(var s=[],o=0;o<e.length;++o)i[o]==null?s.push(y.onesLike(e[o]).asType("bool")):i[o].rank<e[o].rank?s.push(y.expandDims(i[o],-1)):s.push(i[o]);var l=y.concat(s,r.axis);return y.all(l,-1,!1)})},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Concatenate",t}(ir);y.serialization.registerClass(db);function qa(n,t){for(;n<0;)n+=t;return n}function s4(n,t,e){if(n.shape.length>3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;l<s;++l)o.push(1);t=t.reshape(t.shape.concat(o))}else if(r>i){s=r-i;for(var o=[],l=0;l<s;++l)o.push(1);n=n.reshape(n.shape.concat(o))}else s=0;var u;if(n.shape.length===2&&t.shape.length===2)a[0]===a[1]?u=n.mul(t).sum(a[0]):u=n.transpose([1,0]).mul(t).sum(a[1]);else{var c=a[0]!==n.shape.length-1,h=a[1]===t.shape.length-1;u=n.matMul(t,c,h)}if(s>0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l<d+s;++l)p.push(l);u=u.squeeze(p)}return u.shape.length===1&&(u=u.expandDims(1)),u})}var pb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.axes=e.axes,i.normalize=e.normalize==null?!1:e.normalize,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0],r=e[1];if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return qa(s,e[o].shape.length)}):a=[qa(this.axes,i.shape.length),qa(this.axes,r.shape.length)],this.normalize&&(i=uo(i,a[0]),r=uo(r,a[1])),s4(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[qa(this.axes,e.length),qa(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(ir);y.serialization.registerClass(pb);var fb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return io(a.shape,0,r.stddev).add(a)},o=za(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass(fb);var mb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(io(a.shape,1,o))};return za(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(mb);var gb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=Wa(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return za(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(gb);function Ga(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function o4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=Ga(n,s,o,e,t,r);return[l,s,o]})}function l4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=vn(0,n.rank);u<c.length;u++){var h=c[u];i.indexOf(h)!==-1?l.push(1):l.push(n.shape[h])}var d=s.reshape(l),p=o.reshape(l),f=t==null?null:t.reshape(l),m=e==null?null:e.reshape(l),g=Ga(n,d,p,m,f,r);return[g,s,o]})}function u4(n,t,e,i,r){return r===void 0&&(r=.001),y.util.arraysEqual(i.slice().sort(),vn(0,n.rank-1))?o4(n,t,e,i,r):l4(n,t,e,i,r)}var vb=function(n){Q(t,n);function t(e){var i=this;return e==null&&(e={}),i=n.call(this,e)||this,i.supportsMasking=!0,i.axis=e.axis==null?-1:e.axis,i.momentum=e.momentum==null?.99:e.momentum,i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.movingMeanInitializer=tt(e.movingMeanInitializer||"zeros"),i.movingVarianceInitializer=tt(e.movingVarianceInitializer||"ones"),i.betaConstraint=bt(e.betaConstraint),i.gammaConstraint=bt(e.gammaConstraint),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.axis>=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Tt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=vn(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=Ji(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,vn(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),x=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return Ga(s,L,x,C,R,r.epsilon)}else return Ga(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=u4(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,x,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(x).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),movingMeanInitializer:st(this.movingMeanInitializer),movingVarianceInitializer:st(this.movingVarianceInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer),betaConstraint:yt(this.betaConstraint),gammaConstraint:yt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(vb);var yb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r<a.length;r++){var s=a[r];if(!Number.isInteger(s))throw new Error("Expected axis to be an array of integers, "+("but received "+JSON.stringify(i.axis)))}else throw new Error("Expected axis to be an integer or an array of integers, "+("but received "+JSON.stringify(i.axis)));return i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i.supportsMasking=!0,i}return t.prototype.build=function(e){e=Ke(e);var i=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(var r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=i);for(var a=0,s=this.axis;a<s.length;a++){var o=s[a];if(o<0||o>=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==fi(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=Ji(1,o),p=0,f=r.axis;p<f.length;p++){var m=f[p];d[m]=s[m]}for(var g=function(x){return x!=null&&x.shape.length!==o&&r.axis!==[o-1]?x.reshape(d):x},v=g(r.gamma.read()),b=g(r.beta.read()),w=[],S=[],L=0;L<o;++L)r.axis.indexOf(L)!==-1?(w.push(s[L]),S.push(1)):(w.push(1),S.push(s[L]));return c=c.tile(w),h=h.tile(w),v=v.tile(S),b=b.tile(S),Ga(a,c,h,b,v,r.epsilon)})},t.prototype.getConfig=function(){var e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LayerNormalization",t}(De);y.serialization.registerClass(yb);function c4(n,t,e){return y.tidy(function(){if(n.rank!==4)throw new M("temporalPadding expects input tensor to be 4-D, but received a "+(n.rank+"-D tensor."));if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new M("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(e==null&&(e=gn()),e!=="channelsLast"&&e!=="channelsFirst")throw new M("Unknown data format: "+e+". Supported data formats are 'channelsLast' and 'channelsFirst.");var i;return e==="channelsFirst"?i=[[0,0],[0,0],t[0],t[1]]:i=[[0,0],t[0],t[1],[0,0]],y.pad(n,i)})}var bb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.dataFormat=e.dataFormat==null?gn():e.dataFormat,e.padding==null)i.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")i.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new M("ZeroPadding2D expects padding to be a length-2 array, but "+("received a length-"+e.padding.length+" array."));var r=void 0,a=void 0;if(typeof e.padding[0]=="number")r=[e.padding[0],e.padding[0]],a=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new M("ZeroPadding2D expects height padding to be a length-2 array, "+("but received a length-"+e.padding[0].length+" array."));if(r=e.padding[0],e.padding[1].length!==2)throw new M("ZeroPadding2D expects width padding to be a length-2 array, "+("but received a length-"+e.padding[1].length+" array."));a=e.padding[1]}i.padding=[r,a]}return i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i,r;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return c4(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(bb);function Lo(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),nn(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=gn()),a==null&&(a="max"),n=_h(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function wb(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),nn(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=gn()),a==null&&(a="max"),n=Uy(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var Sb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(At(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,nn(i.padding),i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=wn(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=Ua(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(Sb);y.serialization.registerClass(Lb);var Ib=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(Sb);y.serialization.registerClass(Ib);var Ab=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),nn(i.padding),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=wn(i,this.poolSize[0],this.padding,this.strides[0]),r=wn(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Tb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(Ab);y.serialization.registerClass(Tb);var Nb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(Ab);y.serialization.registerClass(Nb);var xb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),nn(i.padding),i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=wn(i,this.poolSize[0],this.padding,this.strides[0]),r=wn(r,this.poolSize[1],this.padding,this.strides[1]),a=wn(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),wb(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(xb);y.serialization.registerClass(Cb);var Rb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),wb(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(xb);y.serialization.registerClass(Rb);var Ob=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),Eb=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(Ob);y.serialization.registerClass(Eb);var Db=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(Ob);y.serialization.registerClass(Db);var kb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Fb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(kb);y.serialization.registerClass(Fb);var Wb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(kb);y.serialization.registerClass(Wb);var Ub=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=bn(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),Bb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=Ke(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Ky(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(Ub);y.serialization.registerClass(Bb);function h4(n){_r(nk,"BidirectionalMergeMode",n)}var d4="concat",zb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=bn(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=bn(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?d4:e.mergeMode,h4(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):Mt(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Tt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof yn,d=0,p=o;d<p.length;d++){var f=p[d];if(f instanceof yn!==h)throw new M("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors")}if(h){var m=[e].concat(o),g=this.inputSpec.concat(l),v=this.inputSpec;this.inputSpec=g;var b=n.prototype.apply.call(this,m,i);return this.inputSpec=v,b}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.initialState,s,o;if(a==null)s=r.forwardLayer.call(e,i),o=r.backwardLayer.call(e,i);else{var l=a.slice(0,a.length/2),u=a.slice(a.length/2);s=r.forwardLayer.call(e,Object.assign(i,{initialState:l})),o=r.backwardLayer.call(e,Object.assign(i,{initialState:u}))}var c;r.returnState&&(Array.isArray(s)&&(c=s.slice(1).concat(o.slice(1))),s=s[0],o=o[0]),r.returnSequences&&(o=y.reverse(o,1));var h;return r.mergeMode==="concat"?h=dh([s,o]):r.mergeMode==="sum"?h=y.add(s,o):r.mergeMode==="ave"?h=y.mul(.5,y.add(s,o)):r.mergeMode==="mul"?h=y.mul(s,o):r.mergeMode==null&&(h=[s,o]),r.returnState?r.mergeMode==null?h.concat(c):[h].concat(c):h})},t.prototype.resetStates=function(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()},t.prototype.build=function(e){var i=this;Qi(this.forwardLayer.name,function(){i.forwardLayer.build(e)}),Qi(this.backwardLayer.name,function(){i.backwardLayer.build(e)}),this.built=!0},t.prototype.computeMask=function(e,i){Array.isArray(i)&&(i=i[0]);var r;if(this.returnSequences?this.mergeMode==null?r=[i,i]:r=i:this.mergeMode==null?r=[null,null]:r=null,this.returnState){var a=this.forwardLayer.states,s=a.map(function(o){return null});return Array.isArray(r)?r.concat(s).concat(s):[r].concat(s).concat(s)}else return r},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e={mergeMode:this.mergeMode},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.fromConfig=function(e,i){var r=bn(i.layer);if(delete i.layer,i.numConstants!=null)throw new Te("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");var a=i;return a.layer=r,new e(a)},t.className="Bidirectional",t}(Ub);y.serialization.registerClass(zb);function p4(n){return new Pa(n)}function f4(n){return new ky(n)}function m4(n){return new Oy(n)}function g4(n){return new Ey(n)}function v4(n){return new Dy(n)}function y4(n){return new Wy(n)}function b4(n){return new Fy(n)}function w4(n){return new Hy(n)}function S4(n){return new Mh(n)}function L4(n){return new _y(n)}function I4(n){return new Py(n)}function A4(n){return new My(n)}function T4(n){return new Vy(n)}function N4(n){return new qy(n)}function x4(n){return new Gy(n)}function C4(n){return new tb(n)}function R4(n){return new Qy(n)}function O4(n){return new Yh(n)}function E4(n){return new Zy(n)}function D4(n){return new eb(n)}function k4(n){return new nb(n)}function F4(n){return new ib(n)}function W4(n){return new rb(n)}function U4(n){return new sb(n)}function B4(n){return new ob(n)}function z4(n){return new ub(n)}function P4(n){return new db(n)}function _4(n){return new cb(n)}function M4(n){return new hb(n)}function H4(n){return new lb(n)}function V4(n){return new pb(n)}function q4(n){return new vb(n)}function G4(n){return new yb(n)}function Y4(n){return new bb(n)}function Kh(n){return new Ib(n)}function K4(n){return Kh(n)}function j4(n){return Kh(n)}function jh(n){return new Nb(n)}function $4(n){return jh(n)}function X4(n){return jh(n)}function $h(n){return new Rb(n)}function J4(n){return $h(n)}function Z4(n){return $h(n)}function Q4(n){return new Eb(n)}function eF(n){return new Fb(n)}function Pb(n){return new Db(n)}function _b(n){return new Wb(n)}function Mb(n){return new Lb(n)}function Hb(n){return new Tb(n)}function tF(n){return new Cb(n)}function nF(n){return new $y(n)}function iF(n){return new qh(n)}function rF(n){return new Xy(n)}function aF(n){return new So(n)}function sF(n){return new jy(n)}function oF(n){return new Vh(n)}function lF(n){return new Jy(n)}function uF(n){return new Gh(n)}function cF(n){return new Si(n)}function hF(n){return new Hh(n)}function dF(n){return new zb(n)}function pF(n){return new Bb(n)}var fF=Pb,mF=_b,gF=Mb,vF=Hb;function yF(n){return new fb(n)}function bF(n){return new mb(n)}function wF(n){return new gb(n)}function SF(n){return new ab(n)}var LF={__proto__:null,inputLayer:p4,elu:f4,reLU:m4,leakyReLU:g4,prelu:v4,softmax:y4,thresholdedReLU:b4,conv1d:w4,conv2d:S4,conv2dTranspose:L4,conv3d:I4,separableConv2d:A4,cropping2D:T4,upSampling2d:N4,depthwiseConv2d:x4,activation:C4,dense:R4,dropout:O4,spatialDropout1d:E4,flatten:D4,repeatVector:k4,reshape:F4,permute:W4,embedding:U4,add:B4,average:z4,concatenate:P4,maximum:_4,minimum:M4,multiply:H4,dot:V4,batchNormalization:q4,layerNormalization:G4,zeroPadding2d:Y4,averagePooling1d:Kh,avgPool1d:K4,avgPooling1d:j4,averagePooling2d:jh,avgPool2d:$4,avgPooling2d:X4,averagePooling3d:$h,avgPool3d:J4,avgPooling3d:Z4,globalAveragePooling1d:Q4,globalAveragePooling2d:eF,globalMaxPooling1d:Pb,globalMaxPooling2d:_b,maxPooling1d:Mb,maxPooling2d:Hb,maxPooling3d:tF,gru:nF,gruCell:iF,lstm:rF,lstmCell:aF,simpleRNN:sF,simpleRNNCell:oF,convLstm2d:lF,convLstm2dCell:uF,rnn:cF,stackedRNNCells:hF,bidirectional:dF,timeDistributed:pF,globalMaxPool1d:fF,globalMaxPool2d:mF,maxPool1d:gF,maxPool2d:vF,Layer:De,RNN:Si,RNNCell:Kr,input:Ty,gaussianNoise:yF,gaussianDropout:bF,alphaDropout:wF,masking:SF};function IF(n,t){return Nh(n,t)}function AF(n,t){return ly(n,t)}function TF(n,t){return uy(n,t)}function NF(n,t){return xh(n,t)}function xF(n,t){return Ch(n,t)}function CF(n,t){return oy(n,t)}function RF(n,t){return Jk(n,t)}function OF(n,t){return Ah(n,t)}function EF(n,t){return co(n,t)}function DF(n,t){return qr(n,t)}function kF(n,t){return qr(n,t)}function FF(n,t){return qr(n,t)}function WF(n,t){return tr(n,t)}function UF(n,t){return tr(n,t)}function BF(n,t){return tr(n,t)}var zF={__proto__:null,binaryAccuracy:IF,binaryCrossentropy:AF,sparseCategoricalAccuracy:TF,categoricalAccuracy:NF,categoricalCrossentropy:xF,precision:CF,recall:RF,cosineProximity:OF,meanAbsoluteError:EF,meanAbsolutePercentageError:DF,MAPE:kF,mape:FF,meanSquaredError:WF,MSE:UF,mse:BF};var PF={__proto__:null,modelFromJSON:k3};function _F(n){return new yo(n)}function MF(n){return Q3(n)}function HF(n){return e4(n)}var VF={__proto__:null,l1l2:_F,l1:MF,l2:HF};var Vb=function(n){Q(t,n);function t(){var e=n!==null&&n.apply(this,arguments)||this;return e.model=null,e}return t.prototype.setModel=function(e){if(!(e instanceof yi))throw new Error("model must be a LayersModel, not some other Container");this.model=e},t}(Vr);function Io(n,t){return n<t}function qb(n,t){return n>t}var Gb=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=Io:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=qb:i.monitorFunc=Io,i.monitorFunc===Io&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===Io?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,vi(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(Vb);function qF(n){return new Gb(n)}var GF={earlyStopping:qF};Xe.Callback=Vb;Xe.CallbackList=ey;Xe.CustomCallback=ny;Xe.EarlyStopping=Gb;Xe.History=ty;Xe.InputSpec=Tt;Xe.LayerVariable=$v;Xe.LayersModel=yi;Xe.RNN=Si;Xe.Sequential=Bh;Xe.SymbolicTensor=yn;Xe.callbacks=GF;Xe.constraints=ZD;Xe.initializers=Dk;Xe.input=Ty;Xe.layers=LF;Xe.loadLayersModel=P3;Xe.metrics=zF;Xe.model=B3;Xe.models=PF;Xe.registerCallbackConstructor=_3;Xe.regularizers=VF;Xe.sequential=z3;Xe.version_layers=Eh});var ow=be(rr=>{"use strict";Object.defineProperty(rr,"__esModule",{value:!0});var B=Xi();var Kb=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t};function zn(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function Sn(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=r[u[0]&2?"return":u[0]?"throw":"next"])&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[0,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Ln;(function(n){n[n.DT_INVALID=0]="DT_INVALID",n[n.DT_FLOAT=1]="DT_FLOAT",n[n.DT_DOUBLE=2]="DT_DOUBLE",n[n.DT_INT32=3]="DT_INT32",n[n.DT_UINT8=4]="DT_UINT8",n[n.DT_INT16=5]="DT_INT16",n[n.DT_INT8=6]="DT_INT8",n[n.DT_STRING=7]="DT_STRING",n[n.DT_COMPLEX64=8]="DT_COMPLEX64",n[n.DT_INT64=9]="DT_INT64",n[n.DT_BOOL=10]="DT_BOOL",n[n.DT_QINT8=11]="DT_QINT8",n[n.DT_QUINT8=12]="DT_QUINT8",n[n.DT_QINT32=13]="DT_QINT32",n[n.DT_BFLOAT16=14]="DT_BFLOAT16",n[n.DT_FLOAT_REF=101]="DT_FLOAT_REF",n[n.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",n[n.DT_INT32_REF=103]="DT_INT32_REF",n[n.DT_UINT8_REF=104]="DT_UINT8_REF",n[n.DT_INT16_REF=105]="DT_INT16_REF",n[n.DT_INT8_REF=106]="DT_INT8_REF",n[n.DT_STRING_REF=107]="DT_STRING_REF",n[n.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",n[n.DT_INT64_REF=109]="DT_INT64_REF",n[n.DT_BOOL_REF=110]="DT_BOOL_REF",n[n.DT_QINT8_REF=111]="DT_QINT8_REF",n[n.DT_QUINT8_REF=112]="DT_QUINT8_REF",n[n.DT_QINT32_REF=113]="DT_QINT32_REF",n[n.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(Ln||(Ln={}));var jb;(function(n){var t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=n.CheckpointFormatVersion||(n.CheckpointFormatVersion={}))})(jb||(jb={}));var Xh={};function YF(n,t){var e={tfOpName:n,category:"custom",inputs:[],attrs:[],customExecutor:t};Xh[n]=e}function $b(n){return Xh[n]}function KF(n){delete Xh[n]}function A(n,t,e,i){var r=t.inputParams[n];if(r&&r.inputIndexStart!==void 0){var a=r.inputIndexStart,s=r.inputIndexEnd===0?void 0:r.inputIndexEnd===void 0?a+1:r.inputIndexEnd;if(r.type==="tensor")return Ht(t.inputNames[r.inputIndexStart],e,i);if(r.type==="tensors"){var o=t.inputNames.slice(a,s);return o.map(function(h){return Ht(h,e,i)})}var l=Ht(t.inputNames.slice(a)[0],e,i),u=l.dataSync();return r.type==="number"?u[0]:B.util.toNestedArray(l.shape,u)}var c=t.attrParams[n];return c&&c.value}function Ht(n,t,e){var i=$t(n),r=i[0],a=i[1],s=e.currentContextIds.find(function(o){return!!t[Ao(r,o)]});return s!==void 0?t[Ao(r,s)][a]:void 0}function jF(n,t,e){return t[Ao(n,e.currentContextId)]}function Zn(n,t){var e=$t(n),i=e[0],r=e[1];return[Ao(i,t&&t.currentContextId),r]}function Ao(n,t){return t?n+"-"+t:n}function $t(n){var t=n.split(":");if(t.length===1)return[n,0];var e=t[0];return[e,Number(t[t.length-1])]}function To(n,t,e){var i=A("pad",n,t,e);if(i==="explicit"){i=A("explicitPaddings",n,t,e);for(var r=[[0,0],[0,0],[0,0],[0,0]],a=0;a<4;a++)r[a][0]=i[a*2],r[a][1]=i[a*2+1];return r}return i}function Qn(n){return n.kept?n:B.clone(n)}var $F=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],XF={__proto__:null,json:$F};var JF=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"clip_value_min",name:"clipValueMin",type:"number"},{tfName:"clip_value_max",name:"clipValueMax",type:"number"}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"clipValueMin",name:"clipValueMin",type:"number",defaultValue:0},{tfName:"clipValueMax",name:"clipValueMax",type:"number",defaultValue:6}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],ZF={__proto__:null,json:JF};var QF=[{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],eW={__proto__:null,json:QF};var tW=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],nW={__proto__:null,json:tW};var iW=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],rW={__proto__:null,json:iW};var aW=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],sW={__proto__:null,json:aW};var oW=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],lW={__proto__:null,json:oW};var uW=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],cW={__proto__:null,json:uW};var hW=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],dW={__proto__:null,json:hW};var pW=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],fW={__proto__:null,json:pW};var mW=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],gW={__proto__:null,json:mW};var vW=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],yW={__proto__:null,json:vW};var bW=[{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],wW={__proto__:null,json:bW};var SW=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool",notSupported:!0}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],LW={__proto__:null,json:SW};var IW=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],AW={__proto__:null,json:IW};var TW=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]}],NW={__proto__:null,json:TW};var Jb=function(){function n(){var t=[XF,ZF,eW,nW,rW,sW,lW,fW,dW,cW,gW,yW,wW,LW,AW,NW],e=[].concat.apply([],t.map(function(i){return i.json}));this.opMappers=e.reduce(function(i,r){return i[r.tfOpName]=r,i},{})}return Object.defineProperty(n,"Instance",{get:function(){return this._instance||(this._instance=new this)},enumerable:!0,configurable:!0}),n.prototype.transformGraph=function(t,e){var i=this;e===void 0&&(e={});var r=t.node,a=[],s=[],o=[],l=r.reduce(function(g,v){return g[v.name]=i.mapNode(v),v.op.startsWith("Placeholder")?a.push(g[v.name]):v.op==="Const"?s.push(g[v.name]):(v.input==null||v.input.length===0)&&o.push(g[v.name]),g},{}),u=[],c=[],h={},d={};e!=null&&(h=this.mapSignatureEntries(e.inputs),d=this.mapSignatureEntries(e.outputs));var p=Object.keys(l);p.forEach(function(g){var v=l[g];v.inputNames.forEach(function(b){var w=Zn(b)[0];v.inputs.push(l[w]),l[w].children.push(v)})}),Object.keys(d).length===0?p.forEach(function(g){var v=l[g];v.children.length===0&&c.push(v)}):Object.keys(d).forEach(function(g){var v=Zn(g)[0],b=l[v];b!=null&&(b.signatureKey=d[g],c.push(b))}),Object.keys(h).length>0?Object.keys(h).forEach(function(g){var v=Zn(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=$b(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Jh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Jh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=ad(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=ad(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=Qh(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=Qh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Zh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Zh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=od(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=id(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=id(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=sd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=sd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=td(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=td(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=nd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=nd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Xb(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Xb(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=Zn(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:ed(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=Zn(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=Zn(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function xW(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function Zb(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):xW(n);return t?e:e.toLowerCase()}function Jh(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?Zb(r.s,i):e}function Zh(n,t,e){var i=n[t];return i?i.b:e}function Qh(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function ed(n){typeof n=="string"&&(n=Ln[n]);switch(n){case Ln.DT_FLOAT:return"float32";case Ln.DT_INT32:case Ln.DT_INT64:case Ln.DT_INT8:case Ln.DT_UINT8:return"int32";case Ln.DT_BOOL:return"bool";case Ln.DT_DOUBLE:return"float32";case Ln.DT_STRING:return"string";default:return null}}function Xb(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function td(n,t,e){var i=n[t];return i&&i.type?ed(i.type):e}function nd(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return ed(r)}):e}function Qb(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function id(n,t,e){var i=n[t];return i&&i.shape?Qb(i.shape):e}function rd(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function ad(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return Zb(a,i)}):e}function sd(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Qb(r)}):e}function od(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var CW=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Ht(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Ht(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return Qh(this.node.rawAttrs,t,e);if(i.s!=null)return Jh(this.node.rawAttrs,t,e);if(i.b!=null)return Zh(this.node.rawAttrs,t,e);if(i.shape!=null)return id(this.node.rawAttrs,t,e);if(i.type!=null)return td(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return rd(this.node.rawAttrs,t,e);if(i.list.s!=null)return ad(this.node.rawAttrs,t,e);if(i.list.shape!=null)return sd(this.node.rawAttrs,t,e);if(i.list.b!=null)return od(this.node.rawAttrs,t,e);if(i.list.type!=null)return nd(this.node.rawAttrs,t,e)}return e},n}();var RW=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var OW=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Ht(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function hn(n,t,e){e===void 0&&(e=""),B.util.assert(EW(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function EW(n,t){if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==-1&&t[e]!==-1&&n[e]!==t[e])return!1;return!0}var DW=function(){function n(t,e,i,r,a,s,o){this.name=t,this.dtype=e,this.maxSize=i,this.elementShape=r,this.identicalElementShapes=a,this.dynamicSize=s,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=B.scalar(0),B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"closed",{get:function(){return this.closed_},enumerable:!0,configurable:!0}),n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.read=function(t){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||t>=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`,
-          because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),hn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i<this.size();i++)t.push(i)}if(t.length===0)return B.tensor([],[0].concat(this.elementShape));var r=this.readMany(t);return hn(this.elementShape,r[0].shape,"TensorArray shape mismatch: "),B.stack(r,0)},n.prototype.concat=function(t){if(!!t&&t!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but concat requested dtype "+t);if(this.size()===0)return B.tensor([],[0].concat(this.elementShape));for(var e=[],i=0;i<this.size();i++)e.push(i);var r=this.readMany(e);return hn(this.elementShape,r[0].shape,"TensorArray shape mismatch: tensor array shape ("+this.elementShape+") vs first tensor shape ("+r[0].shape+")"),B.concat(r,0)},n.prototype.scatter=function(t,e){if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);if(t.length!==e.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+e.shape[0]);var i=Math.max.apply(Math,t);if(!this.dynamicSize&&i>=this.maxSize)throw new Error("Max index must be < array size ("+i+"  vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
+`+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");if(p!=null){for(var f={},m=0,g=Object.keys(ln);m<g.length;m++){var v=g[m];f[v]=ln[v]}for(var b=0,w=Object.keys(e);b<w.length;b++){var v=w[b];f[v]=e[v]}var S=c.config;S.customObjects=f;for(var L=Gt({},ln),x=0,C=Object.keys(e);x<C.length;x++){var v=C[x];ln[v]=e[v]}lh(c.config);var R=p(d,c.config,e,r);return ln=Gt({},L),R}else{for(var L=Gt({},ln),D=0,k=Object.keys(e);D<k.length;D++){var v=k[D];ln[v]=e[v]}var R=new d(c.config);return ln=Gt({},L),R}}}function GD(n,t){return n<t?-1:n>t?1:0}function to(n,t){return-1*GD(n,t)}function fi(n){if(n==null)return n;for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.indexOf(r)===-1&&t.push(r)}return t}function YD(n){if(n==null)throw new M("Invalid value in obj: "+JSON.stringify(n));for(var t in n)if(n.hasOwnProperty(t))return!1;return!0}function _r(n,t,e){if(e==null)return;if(n.indexOf(e)<0)throw new M(e+" is not a valid "+t+".  Valid values are "+n+" or null/undefined.")}function uh(n,t,e,i){return e===void 0&&(e=0),i===void 0&&(i=Infinity),Wn(e>=0),Wn(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function At(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return At(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(Sv(n)+".")})}function Sv(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return Sv(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function KD(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];var o=y.util.now();return o-e<t||(e=o,i=n.apply(void 0,a)),i};return r}function Lv(n){return n==="relu"?"relu":n==="linear"?"linear":n==="elu"?"elu":null}function ch(n,t){return y.tidy(function(){return y.sqrt(y.sum(y.mul(n,n),t,!0))})}var ka=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),Iv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMaxValue=2,i.defaultAxis=0,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=ch(e,i.axis),a=y.clipByValue(r,0,i.maxValue);return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{maxValue:this.maxValue,axis:this.axis}},t.className="MaxNorm",t}(ka);y.serialization.registerClass(Iv);var Av=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultAxis=0,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){return y.div(e,y.add(vt(),ch(e,i.axis)))})},t.prototype.getConfig=function(){return{axis:this.axis}},t.className="UnitNorm",t}(ka);y.serialization.registerClass(Av);var Tv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="NonNeg",t}(ka);y.serialization.registerClass(Tv);var Nv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMinValue=0,i.defaultMaxValue=1,i.defaultRate=1,i.defaultAxis=0,i.minValue=e.minValue!=null?e.minValue:i.defaultMinValue,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.rate=e.rate!=null?e.rate:i.defaultRate,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=ch(e,i.axis),a=y.add(y.mul(i.rate,y.clipByValue(r,i.minValue,i.maxValue)),y.mul(1-i.rate,r));return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}},t.className="MinMaxNorm",t}(ka);y.serialization.registerClass(Nv);var xv={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function yt(n){return oh(n)}function Cv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"constraint")}function bt(n){if(n==null)return null;if(typeof n=="string"){var t=n in xv?xv[n]:n,e={className:t,config:{}};return Cv(e)}else return n instanceof ka?n:Cv(n)}function jD(n){return new Iv(n)}function $D(n){return new Av(n)}function XD(){return new Tv}function JD(n){return new Nv(n)}var ZD={__proto__:null,maxNorm:jD,unitNorm:$D,nonNeg:XD,minMaxNorm:JD};var QD=["channelsFirst","channelsLast"],ek=["valid","same","causal"],tk=["max","avg"],nk=["sum","mul","concat","ave"];var Mr=new Map;function ct(n){_r(QD,"DataFormat",n)}function tn(n){_r(ek,"PaddingMode",n)}function Rv(n){_r(tk,"PoolMode",n)}var Fa=[],Ov="/";function Qi(n,t){Fa.push(n);try{var e=t();return Fa.pop(),e}catch(i){throw Fa.pop(),i}}function ik(){return Fa.length===0?"":Fa.join(Ov)+Ov}function Dv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");return ik()+n}function kv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");Mr.has(n)||Mr.set(n,0);var t=Mr.get(n);if(Mr.set(n,Mr.get(n)+1),t>0){var e=n+"_"+t;return Mr.set(e,1),e}else return n}var rk=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function Ev(n){return!!n.match(rk)}function ak(n){return n===parseInt(n.toString(),10)}function mi(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r<e;++r)i*=n[r];return i}function Fv(n){return n=Array.isArray(n)?new Float32Array(n):n,y.tensor1d(n)}function Hr(n){return y.min(Fv(n)).dataSync()[0]}function gi(n){return y.max(Fv(n)).dataSync()[0]}function gn(n,t){if(t<n)throw new M("end ("+t+") < begin ("+n+") is forbidden.");for(var e=[],i=n;i<t;++i)e.push(i);return e}function Wa(n,t){return n.asType(t)}function Ua(n,t){t===void 0&&(t=-1);var e=n.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),n.reshape(e)}function sk(n,t){return y.tidy(function(){if(n.shape.length!==2)throw new M("repeat() expects a rank-2 tensor, but received a "+("rank-"+n.shape.length+" tensor."));var e=Ua(n,1);return hh(e,[1,t,1])})}function ok(n){var t=[mi(n.shape)];return n.reshape(t)}function lk(n){if(n.rank<=1)throw new M("batchFlatten requires a minimum rank of 2. Got rank: "+n.rank+".");var t=[n.shape[0],mi(n.shape,1)];return n.reshape(t)}function er(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[t,0],[e,n.shape[1]]);case 3:return y.slice3d(n,[t,0,0],[e,n.shape[1],n.shape[2]]);case 4:return y.slice4d(n,[t,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3]]);case 5:return y.slice(n,[t,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4]]);case 6:return y.slice(n,[t,0,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4],n.shape[5]]);default:throw new M("sliceAlongFirstAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function dh(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[0,t],[n.shape[0],e]);case 3:return y.slice3d(n,[0,0,t],[n.shape[0],n.shape[1],e]);case 4:return y.slice4d(n,[0,0,0,t],[n.shape[0],n.shape[1],n.shape[2],e]);default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function no(n,t,e,i){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:switch(i){case 1:return er(n,t,e);case 2:return dh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 3:switch(i){case 1:return er(n,t,e);case 2:return y.slice3d(n,[0,t,0],[n.shape[0],e,n.shape[2]]);case 3:return dh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 4:switch(i){case 1:return er(n,t,e);case 2:return y.slice4d(n,[0,t,0,0],[n.shape[0],e,n.shape[2],n.shape[3]]);case 3:return y.slice4d(n,[0,0,t,0],[n.shape[0],n.shape[1],e,n.shape[3]]);case 4:return dh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function ph(n,t){t===void 0&&(t=-1);var e;return t<0&&(e=n[0].rank,e!==0?t=e:t=0),t===n[0].rank&&(t=-1),y.concat(n,t)}function Wv(n,t){switch(n.rank){case 1:return y.concat1d([n,t]);case 2:return y.concat2d([n,t],0);case 3:return y.concat3d([n,t],0);case 4:return y.concat4d([n,t],0);default:throw new M("concatAlongFirstAxis() received an unsupported "+("tensor rank: "+n.rank))}}function hh(n,t){if(Array.isArray(t)||(t=[t]),n.rank!==t.length)throw new M("The length of input n ("+t.length+") does not match "+("the number of dimensions in input x ("+n.rank+")"));return y.tile(n,t)}function io(n,t,e,i,r){return t===void 0&&(t=0),e===void 0&&(e=1),y.randomNormal(n,t,e,i,r)}function Un(n,t,e,i){if(n.rank<2||t.rank<2)throw new Te("dot requires both inputs to be rank >= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?fh(n.rank,i,mn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?fh(n.rank,i,mn()):null,activation:e}).reshape(f)}}function Uv(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function Ba(n){return y.mul(n,n)}function fh(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Bn(n,t,e){return y.tidy(function(){return e==null&&(e=mn()),ct(e),n.add(fh(n.rank,t,e))})}function uk(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function ck(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function Bv(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function hk(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function za(n,t,e){return e===void 0&&(e=!1),e?n():t()}var dk=["fanIn","fanOut","fanAvg"],pk=["normal","uniform","truncatedNormal"];function fk(n){_r(dk,"FanMode",n)}function mk(n){_r(pk,"Distribution",n)}var un=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),zv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(un);y.serialization.registerClass(zv);var mh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(un);y.serialization.registerClass(mh);var Pv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(un);y.serialization.registerClass(Pv);var _v=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(un);y.serialization.registerClass(_v);var Mv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return io(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(un);y.serialization.registerClass(Mv);var Hv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(un);y.serialization.registerClass(Hv);var Vv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(un);y.serialization.registerClass(Vv);function gk(n,t){t===void 0&&(t="channelsLast");var e,i;if(ct(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=mi(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=mi(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=mi(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var Yt=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,fk(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,mk(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=gk(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(un);y.serialization.registerClass(Yt);var gh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="GlorotUniform",t}(Yt);y.serialization.registerClass(gh);var vh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="GlorotNormal",t}(Yt);y.serialization.registerClass(vh);var yh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="HeNormal",t}(Yt);y.serialization.registerClass(yh);var bh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="HeUniform",t}(Yt);y.serialization.registerClass(bh);var wh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="LeCunNormal",t}(Yt);y.serialization.registerClass(wh);var Sh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="LeCunNormal",t}(Yt);y.serialization.registerClass(Sh);var qv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=io(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(un);y.serialization.registerClass(qv);var Gv={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function Yv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function st(n){return oh(n)}function tt(n){if(typeof n=="string"){var t=n in Gv?Gv[n]:n;if(t==="GlorotNormal")return new vh;if(t==="GlorotUniform")return new gh;if(t==="HeNormal")return new yh;if(t==="HeUniform")return new bh;if(t==="LeCunNormal")return new wh;if(t==="LeCunUniform")return new Sh;var e={};return e.className=t,e.config={},Yv(e)}else return n instanceof un?n:Yv(n)}function vk(){return new zv}function yk(){return new mh}function bk(n){return new Pv(n)}function wk(n){return new _v(n)}function Sk(n){return new Mv(n)}function Lk(n){return new Hv(n)}function Ik(n){return new Vv(n)}function Ak(n){return new Yt(n)}function Tk(n){return new gh(n)}function Nk(n){return new vh(n)}function xk(n){return new yh(n)}function Ck(n){return new bh(n)}function Rk(n){return new wh(n)}function Ok(n){return new Sh(n)}function Ek(n){return new qv(n)}var Dk={__proto__:null,zeros:vk,ones:yk,constant:bk,randomUniform:wk,randomNormal:Sk,truncatedNormal:Lk,identity:Ik,varianceScaling:Ak,glorotUniform:Tk,glorotNormal:Nk,heNormal:xk,heUniform:Ck,leCunNormal:Rk,leCunUniform:Ok,orthogonal:Ek};var kk=0;function Kv(){return kk++}var ro={};function ao(n){return n===void 0&&(n=""),n in ro||(ro[n]=0),ro[n]+=1,n+ro[n].toString()}function Lh(n){return Array.isArray(n)&&Array.isArray(n[0])}function so(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function Ke(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function oo(n){for(var t=0,e=0,i=n;e<i.length;e++){var r=i[e];r.shape.length===0?t+=1:t+=r.shape.reduce(function(a,s){return a*s})}return t}var jv="Variable",$v=function(){function n(t,e,i,r,a){e===void 0&&(e="float32"),i===void 0&&(i=jv),r===void 0&&(r=!0),a===void 0&&(a=null),this.dtype=e??"float32",this.shape=t.shape,this.id=Kv(),i=i??jv,this.originalName=Dv(i),this.name=kv(this.originalName),this.trainable_=r,this.constraint=a,this.val=y.variable(t,this.trainable_,this.name,this.dtype)}return n.prototype.read=function(){return this.assertNotDisposed(),this.val},n.prototype.write=function(t){return this.assertNotDisposed(),Fk(this.val,t),this.val.id!==t.id&&(this.val.assign(t),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this},n.prototype.dispose=function(){this.assertNotDisposed(),this.val.dispose()},n.prototype.assertNotDisposed=function(){if(this.val.isDisposed)throw new Error("LayersVariable "+this.name+" is already disposed.")},Object.defineProperty(n.prototype,"trainable",{get:function(){return this.trainable_},set:function(t){this.trainable_=t,this.val.trainable=t},enumerable:!0,configurable:!0}),n}();function Fk(n,t){if(n.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(n.shape)+" vs. "+JSON.stringify(t.shape))}function Ih(n){return n.map(function(t){return t.read()})}function Ah(n){n.forEach(function(t){var e=t[0];e.write(t[1])})}var Tt=function(){function n(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}return n}(),vn=function(){function n(t,e,i,r,a,s,o){this.dtype=t,this.shape=e,this.sourceLayer=i,this.inputs=r,this.callArgs=a,this.outputTensorIndex=o,this.id=Kv(),s!=null&&(this.originalName=Dv(s),this.name=kv(this.originalName)),this.rank=e.length}return n}(),Wk=0,lo=function(){function n(t,e){this.callArgs=e,this.id=Wk++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(var i=0,r=t.inboundLayers;i<r.length;i++){var a=r[i];a!=null&&a.outboundNodes.push(this)}t.outboundLayer.inboundNodes.push(this)}return n.prototype.getConfig=function(){for(var t=[],e=0,i=this.inboundLayers;e<i.length;e++){var r=i[e];r!=null?t.push(r.name):t.push(null)}return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}},n}(),Uk=0,De=function(n){Q(t,n);function t(e){e===void 0&&(e={});var i=n.call(this)||this;i._callHook=null,i._addedWeightNames=[],i._stateful=!1,i.id=Uk++,i.activityRegularizer=null,i.inputSpec=null,i.supportsMasking=!1,i._trainableWeights=[],i._nonTrainableWeights=[],i._losses=[],i._updates=[],i._built=!1,i.inboundNodes=[],i.outboundNodes=[];var r=e.name;if(!r){var a=i.getClassName();r=Jn(a)+"_"+ao(a)}if(i.name=r,i.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){var s=void 0;if(e.batchInputShape!=null)s=e.batchInputShape;else if(e.inputShape!=null){var o=null;e.batchSize!=null&&(o=e.batchSize),s=[o].concat(e.inputShape)}i.batchInputShape=s;var l=e.dtype;l==null&&(l=e.inputDType),l==null&&(l="float32"),i.dtype=l}return e.weights!=null?i.initialWeights=e.weights:i.initialWeights=null,i._refCount=null,i.fastWeightInitDuringBuild=!1,i}return t.nodeKey=function(e,i){return e.name+"_ib-"+i.toString()},t.prototype.getNodeAtIndex=function(e,i){if(this.inboundNodes.length===0)throw new Fn("The layer has never been called "+("and thus has no defined "+i+"."));if(this.inboundNodes.length<=e)throw new M("Asked to get "+i+" at node "+e+", "+("but the layer has only "+this.inboundNodes.length+" inbound nodes."));return this.inboundNodes[e]},t.prototype.getInputAt=function(e){return Mt(this.getNodeAtIndex(e,"input").inputTensors)},t.prototype.getOutputAt=function(e){return Mt(this.getNodeAtIndex(e,"output").outputTensors)},Object.defineProperty(t.prototype,"input",{get:function(){if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" is not connected, no input to return.");return Mt(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return Mt(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Je(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Je(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r<e.length;r++){var a=e[r],s=i[r];if(s==null)continue;var o=a.rank;if(s.ndim!=null&&o!==s.ndim)throw new M("Input "+r+" is incompatible with layer "+this.name+": "+("expected ndim="+s.ndim+", found ndim="+o));if(s.maxNDim!=null&&o>s.maxNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected max_ndim="+s.maxNDim+", found ndim="+o));if(s.minNDim!=null&&o<s.minNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected min_ndim="+s.minNDim+", found ndim="+o+"."));if(s.dtype!=null&&a.dtype!==s.dtype)throw new M("Input "+r+" is incompatible with layer "+this.name+" "+(": expected dtype="+s.dtype+", found dtype="+a.dtype+"."));if(s.axes){var l=a.shape;for(var u in s.axes){var c=Number(u),h=s.axes[u],d=c>=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p<s.shape.length;++p){var f=s.shape[p],m=a.shape[p];if(f!=null&&m!=null&&f!==m)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected shape="+s.shape+", ")+("found shape="+a.shape+"."))}}},t.prototype.call=function(e,i){return e},t.prototype.invokeCallHook=function(e,i){this._callHook!=null&&this._callHook(e,i)},t.prototype.setCallHook=function(e){this._callHook=e},t.prototype.clearCallHook=function(){this._callHook=null},t.prototype.apply=function(e,i){var r=this;i=i||{},this.assertNotDisposed();for(var a=Je(e),s=!0,o=0,l=a;o<l.length;o++){var u=l[o];if(!(u instanceof vn)){s=!1;break}}for(var c=!0,h=0,d=a;h<d.length;h++){var u=d[h];if(u instanceof vn){c=!1;break}}if(s===c)throw new M("Arguments to apply() must be all SymbolicTensors or all Tensors");return Qi(this.name,function(){if(!r.built){r.assertInputCompatibility(e);for(var p=[],f=0,m=Je(e);f<m.length;f++){var g=m[f];p.push(g.shape)}r.build(Mt(p)),r.built=!0,r.initialWeights&&r.setWeights(r.initialWeights),r._refCount===null&&c&&(r._refCount=1)}if(r.assertInputCompatibility(e),c){for(var v=r.call(e,i),b=Je(v),w=[],S=0,L=b;S<L.length;S++){var x=L[S];a.indexOf(x)!==-1&&(x=x.clone()),w.push(x)}if(v=Mt(w),r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}else{var C=Bk(e),R=r.computeOutputShape(C),v=void 0,D=zk(e);if(r.warnOnIncompatibleInputShape(Array.isArray(e)?C[0]:C),R!=null&&R.length>0&&Array.isArray(R[0])?v=R.map(function(W,F){return new vn(D,W,r,Je(e),i,r.name,F)}):v=new vn(D,R,r,Je(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new pi("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i<r.length;i++){var a=r[i],s=JSON.stringify(a.outputShapes);e.indexOf(s)===-1&&e.push(s)}if(e.length===1){var o=this.inboundNodes[0].outputShapes;return Array.isArray(o)&&Array.isArray(o[0])&&o.length===1?o[0]:o}else throw new pi("The layer "+this.name+' has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.')},enumerable:!0,configurable:!0}),t.prototype.countParams=function(){if(!this.built)throw new Fn("You tried to call countParams() on "+this.name+", but the layer is not built yet. Build it first by calling build(batchInputShape).");return oo(this.weights)},t.prototype.build=function(e){this.built=!0},t.prototype.getWeights=function(e){return e===void 0&&(e=!1),Ih(e?this.trainableWeights:this.weights)},t.prototype.setWeights=function(e){var i=this;y.tidy(function(){var r=i.weights;if(r.length!==e.length)throw new M('You called setWeights(weights) on layer "'+i.name+'" '+("with a weight list of length "+e.length+", ")+("but the layer was expecting "+r.length+" weights. ")+("Provided weights: "+e+"..."));if(r.length===0)return;for(var a=[],s=Ih(r),o=0;o<s.length;++o){var l=s[o],u=r[o],c=e[o];if(!y.util.arraysEqual(l.shape,c.shape))throw new M("Layer weight shape "+l.shape+" "+("not compatible with provided weight shape "+c.shape));a.push([u,c])}Ah(a)})},t.prototype.addWeight=function(e,i,r,a,s,o,l){if(this._addedWeightNames.indexOf(e)!==-1)throw new M("Duplicate weight name "+e+" for layer "+this.name);this._addedWeightNames.push(e),r==null&&(r="float32"),this.fastWeightInitDuringBuild&&(a=tt("zeros"));var u=a.apply(i,r),c=new $v(u,r,e,o,l);return u.dispose(),s!=null&&this.addLoss(function(){return s.apply(c.read())}),o==null&&(o=!0),o?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c},t.prototype.setFastWeightInitDuringBuild=function(e){this.fastWeightInitDuringBuild=e},t.prototype.addLoss=function(e){var i;if(e==null||Array.isArray(e)&&e.length===0)return;e=Je(e),this._losses!==void 0&&this._losses!==null&&(i=this.losses).push.apply(i,e)},t.prototype.computeOutputShape=function(e){return e},t.prototype.computeMask=function(e,i){var r=this;if(!this.supportsMasking){if(i!=null)if(Array.isArray(i))i.forEach(function(a){if(a!=null)throw new TypeError("Layer "+r.name+" does not support masking, but was passed an inputMask.")});else throw new TypeError("Layer "+this.name+" does not support masking, but was passed an inputMask.");return null}return i},t.prototype.addInboundNode=function(e,i,r,a,s,o,l){l===void 0&&(l=null);var u=Je(e);i=Je(i),r=Je(r),a=Je(a),s=so(s),o=so(o);for(var c=[],h=[],d=[],p=0,f=u;p<f.length;p++){var m=f[p];c.push(m.sourceLayer),h.push(m.nodeIndex),d.push(m.tensorIndex)}new lo({outboundLayer:this,inboundLayers:c,nodeIndices:h,tensorIndices:d,inputTensors:u,outputTensors:i,inputMasks:r,outputMasks:a,inputShapes:s,outputShapes:o},l);for(var g=0;g<i.length;g++)i[g].sourceLayer=this,i[g].nodeIndex=this.inboundNodes.length-1,i[g].tensorIndex=g},t.prototype.getConfig=function(){var e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e},t.prototype.disposeWeights=function(){return this.weights.forEach(function(e){return e.dispose()}),this.weights.length},t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Layer '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){if(!this.built)throw new Error("Cannot dispose Layer "+this.name+" because it has not been built yet.");if(this._refCount===null)throw new Error("Cannot dispose Layer "+this.name+" because it has not been used yet.");this.assertNotDisposed();var e=0;return--this._refCount===0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}},t}(y.serialization.Serializable);function Bk(n){n=Je(n);for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.push(r.shape)}return Mt(t)}function zk(n){return"float32"}function Xv(n,t,e){if((t==null||e!=null&&e>0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a<i.inboundLayers.length;a++)for(var s=i.inputTensors[a],o=i.inboundLayers[a],l=i.nodeIndices[a],u=Xv(s,o,l),c=0,h=u;c<h.length;c++){var d=h[c];r.indexOf(d)===-1&&r.push(d)}return r}var Pa=function(n){Q(t,n);function t(e){var i=n.call(this,{dtype:e.dtype,name:e.name!=null?e.name:ao("input").toString()})||this;if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),i.trainable=!1,i.built=!0,i.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new M("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");var r=e.batchInputShape;if(r==null){if(e.inputShape==null)throw new M("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");r=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new M("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");var a=e.dtype||"float32";i.batchInputShape=r,i.dtype=a,i.inputSpec=[{shape:r}];var s=new vn(i.dtype,i.batchInputShape,i,[],{},i.name);return s.nodeIndex=0,s.tensorIndex=0,new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[r],outputShapes:[r]}),i}return t.prototype.apply=function(e,i){throw new M("Cannot pass any input to an "+("InputLayer's apply() method. InputLayer name: "+this.name))},t.prototype.dispose=function(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}},t.prototype.getConfig=function(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}},t.className="InputLayer",t}(De);y.serialization.registerClass(Pa);function Jv(n){if(n.batchShape==null&&n.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(n.batchShape!=null&&n.shape!=null)throw new M("Please provide either a `shape` or `batchShape` argument to Input, but not both.");var t=n.batchShape;n.shape!=null&&t==null&&(t=[null].concat(n.shape));var e=n.dtype;e==null&&(e="float32");var i=new Pa({batchInputShape:t,name:n.name,dtype:e,sparse:n.sparse}),r=i.inboundNodes[0].outputTensors;return r[0]}function vi(n){return Le(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(n==null)return[2];t=[],e=[],i=[];for(r in n)a=n[r],typeof a!="number"&&(s=a,t.push(s.data()),e.push(r),i.push(s));return t.length>0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l<o.length;++l)n[e[l]]=o[l][0];y.dispose(i),u.label=2;case 2:return[2]}})})}function Zv(n){if(n==null)return;for(var t in n){var e=n[t];typeof e!="number"&&e.dispose()}}var Qv;(function(n){n[n.SILENT=0]="SILENT",n[n.VERBOSE=1]="VERBOSE"})(Qv||(Qv={}));var Pk=125,Vr=function(){function n(){this.validationData=null}return n.prototype.setParams=function(t){this.params=t},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.setModel=function(t){},n}(),ey=function(){function n(t,e){e===void 0&&(e=10),t==null&&(t=[]),this.callbacks=t,this.queueLength=e}return n.prototype.append=function(t){this.callbacks.push(t)},n.prototype.setParams=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setParams(t)}},n.prototype.setModel=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setModel(t)}},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainBegin(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainEnd(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n}(),_k=function(n){Q(t,n);function t(){return n.call(this)||this}return t.prototype.onEpochBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.seen=0,this.totals={},[2]})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l=this;return ve(this,function(u){i==null&&(i={}),r=i.size==null?0:i.size,this.seen+=r,a=function(c){var h=i[c];if(typeof h=="number")s.totals.hasOwnProperty(c)||(s.totals[c]=0),s.totals[c]=s.totals[c]+h*r;else{var d=void 0;c in s.totals?d=s.totals[c]:s.totals[c]=0;var p=y.tidy(function(){return y.add(l.totals[c],y.mul(h,r))});s.totals[c]=p,d!=null&&d.dispose()}},s=this;for(o in i)a(o);return[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u=this;return ve(this,function(c){if(i!=null)for(r=function(h){if(a.totals[h]==null)return"continue";typeof a.totals[h]=="number"?i[h]=a.totals[h]/a.seen:y.tidy(function(){var d=y.mul(y.div(1,u.seen),u.totals[h]);i[h]=d,u.totals[h].dispose(),y.keep(i[h])})},a=this,s=0,o=this.params.metrics;s<o.length;s++)l=o[s],r(l);return[2]})})},t}(Vr),ty=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.epoch=[],this.history={},[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){i==null&&(i={}),this.epoch.push(e);for(r in i)this.history[r]==null&&(this.history[r]=[]),this.history[r].push(i[r]);return[2]})})},t.prototype.syncData=function(){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:e=[],i=[],r=[];for(a in this.history)for(s=this.history[a],o=0;o<s.length;++o)typeof s[o]!="number"&&(l=s[o],e.push(l.data()),i.push(a),r.push(o));return[4,Promise.all(e)];case 1:for(u=d.sent(),c=0;c<u.length;++c)h=this.history[i[c]][r[c]],h.dispose(),this.history[i[c]][r[c]]=u[c][0];return[2]}})})},t}(Vr),ny=function(n){Q(t,n);function t(e,i){var r=n.call(this)||this;if(r.currentEpoch=0,r.yieldEvery=i||"auto",r.yieldEvery==="auto"&&(r.yieldEvery=Pk),r.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");return y.util.isNumber(r.yieldEvery)&&(r.maybeWait=KD(r.maybeWait.bind(r),r.yieldEvery)),r.trainBegin=e.onTrainBegin,r.trainEnd=e.onTrainEnd,r.epochBegin=e.onEpochBegin,r.epochEnd=e.onEpochEnd,r.batchBegin=e.onBatchBegin,r.batchEnd=e.onBatchEnd,r.yield=e.onYield,r}return t.prototype.maybeWait=function(e,i,r){return Le(this,void 0,void 0,function(){var a;return ve(this,function(s){switch(s.label){case 0:return a=[],this.yield!=null?[4,vi(r)]:[3,2];case 1:s.sent(),a.push(this.yield(e,i,r)),s.label=2;case 2:return a.push(y.nextFrame()),[4,Promise.all(a)];case 3:return s.sent(),[2]}})})},t.prototype.onEpochBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.currentEpoch=e,this.epochBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.epochBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.epochEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.epochEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="epoch"&&r.push(y.nextFrame()),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onBatchBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.batchBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.batchBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.batchEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.batchEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="batch"?r.push(y.nextFrame()):y.util.isNumber(this.yieldEvery)&&r.push(this.maybeWait(this.currentEpoch,e,i)),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainBegin!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainBegin(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainEnd!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainEnd(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t}(Vr);function iy(n,t){if(n==null&&(n={}),n instanceof Vr)return[n];if(Array.isArray(n)&&n[0]instanceof Vr)return n;var e=Je(n);return e.map(function(i){return new ny(i,t)})}var ry=function(){function n(){}return n.registerCallbackConstructor=function(t,e){y.util.assert(t>=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function ay(n,t,e,i,r,a,s,o,l){var u=new ty,c=[new _k].concat(ry.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new ey(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function yn(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function uo(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(Ba(n),t,!0),i=y.fill(e.shape,vt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function tr(n,t){return y.tidy(function(){return y.mean(Ba(y.sub(t,n)),-1)})}function co(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function qr(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),vt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function Mk(n,t){return y.tidy(function(){var e=y.clipByValue(t,vt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,vt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(Ba(y.sub(i,a)),-1)})}function Hk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(Ba(e),-1)})}function Vk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function qk(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function Gk(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function _a(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,vt(),1-vt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function ho(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(ok(n)).toInt();t=y.clipByValue(t,vt(),1-vt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return _a(a,t,e)})}function Yk(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function po(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,vt(),1-vt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(Yk(n,e),-1)})}function Kk(n,t){return y.tidy(function(){var e=y.clipByValue(n,vt(),1),i=y.clipByValue(t,vt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function jk(n,t){return y.tidy(function(){var e=y.log(y.add(vt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function Th(n,t){return y.tidy(function(){var e=uo(n,-1),i=uo(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var fo={meanSquaredError:tr,meanAbsoluteError:co,meanAbsolutePercentageError:qr,meanSquaredLogarithmicError:Mk,squaredHinge:Hk,hinge:Vk,categoricalHinge:qk,logcosh:Gk,categoricalCrossentropy:_a,sparseCategoricalCrossentropy:ho,binaryCrossentropy:po,kullbackLeiblerDivergence:Kk,poisson:jk,cosineProximity:Th};function Nh(n){if(typeof n=="string"){if(n in fo)return fo[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function xh(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=Wa(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function Ch(n,t){return y.tidy(function(){return Wa(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function sy(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function $k(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function Xk(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function oy(n,t){return y.tidy(function(){var e=sy(n,t),i=Xk(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Jk(n,t){return y.tidy(function(){var e=sy(n,t),i=$k(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function ly(n,t){return po(n,t)}function uy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var Zk=tr,Qk=tr,e3=co,t3=co,n3=qr,i3=qr,Rh=_a,r3=Th,cy=ho,mo={binaryAccuracy:xh,categoricalAccuracy:Ch,precision:oy,categoricalCrossentropy:Rh,sparseCategoricalCrossentropy:cy,mse:Zk,MSE:Qk,mae:e3,MAE:t3,mape:n3,MAPE:i3,cosine:r3};function a3(n){if(typeof n=="string"&&n in mo)return mo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function go(n){if(Wn(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(fo);e<i.length;e++){var r=i[e];if(fo[r]===n){t=r;break}}if(t!==void 0)return t;for(var a=0,s=Object.keys(mo);a<s.length;a++){var r=s[a];if(mo[r]===n){t=r;break}}return t!==void 0?t:n.name}function s3(n){var t={Adagrad:function(){return y.train.adagrad(.01)},Adadelta:function(){return y.train.adadelta(1,.95,vt())},Adam:function(){return y.train.adam(.001,.9,.999,vt())},Adamax:function(){return y.train.adamax(.002,.9,.999,vt(),0)},RMSProp:function(){return y.train.rmsprop(.001,.9,0,vt())},SGD:function(){return y.train.sgd(.01)}};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,n in t)return t[n]();throw new M("Unknown Optimizer "+n)}var hy=1*1024*1024;function dy(n,t,e){if(e===void 0&&(e=!1),n==null||typeof n!="object"||Object.getPrototypeOf(n)!==Object.prototype||!Oh(n))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){var i=JSON.stringify(n);i.length>hy&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+(hy+"."))}}function Oh(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e];if(typeof r!="string")return!1;if(!Oh(n[r]))return!1}return!0}else if(Array.isArray(n)){for(var a=0,s=n;a<s.length;a++){var o=s[a];if(!Oh(o))return!1}return!0}else return!1;else{var l=typeof n;return l==="string"||l==="number"||l==="boolean"}}function h3(n,t,e,i){i===void 0&&(i=console.log);var r=l3(n),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,e=e||[.45,.85,1]):(t=t||98,e=e||[.33,.55,.67,1]),e[e.length-1]<=1&&(e=e.map(function(d){return Math.floor(t*d)}));var s;if(!r){a.push("Receives inputs"),s=[];for(var o in n.nodesByDepth)s.push.apply(s,n.nodesByDepth[o])}i("_".repeat(t)),vo(a,e,i),i("=".repeat(t));for(var l=n.layers,u=0;u<l.length;++u)r?u3(l[u],e,i):c3(l[u],e,s,i),i((u===l.length-1?"=":"_").repeat(t));n.checkTrainableWeightsConsistency();var c=o3(n),h=oo(n.nonTrainableWeights);i("Total params: "+(c+h)),i("Trainable params: "+c),i("Non-trainable params: "+h),i("_".repeat(t))}function o3(n){var t;return n.collectedTrainableWeights!=null?t=oo(n.collectedTrainableWeights):t=oo(n.trainableWeights),t}function l3(n){var t=!0,e=[],i=[];for(var r in n.nodesByDepth)e.push(n.nodesByDepth[r]);for(var a=0,s=e;a<s.length;a++){var o=s[a];if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l<u.length;l++){for(var c=u[l],h=!1,d=0,p=c.inboundNodes;d<p.length;d++){var f=p[d];if(i.indexOf(f)!==-1)if(h){t=!1;break}else h=!0}if(!t)break}return t}function vo(n,t,e){e===void 0&&(e=console.log);for(var i="",r=0;r<n.length;++r)r>0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function u3(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];vo(s,t,e)}function c3(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s<o.length;s++){var l=o[s];if(e!=null&&e.length>0&&e.indexOf(l)===-1)continue;for(var u=0;u<l.inboundLayers.length;++u){var c=l.inboundLayers[u].name,h=l.nodeIndices[u],d=l.tensorIndices[u];a.push(c+"["+h+"]["+d+"]")}}var p=n.name,f=n.getClassName(),m=a.length===0?"":a[0],g=[p+" ("+f+")",r,n.countParams().toString(),m];vo(g,t,i);for(var u=1;u<a.length;++u)vo(["","","",a[u]],t,i)}function py(n,t,e){return(n==="inboundNodes"||n==="outputLayers"||n==="inputLayers")&&t===0&&typeof e=="string"}function Ma(n,t){if(n===null)return null;if(typeof n=="string")return Zi(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Ma(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u];if(u==="name"&&typeof c=="string")s[u]=c;else{var h=Zi(u);s[h]=Ma(c,h)}}return s}}function Eh(n,t){if(n==null)return null;if(typeof n=="string")return Jn(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Eh(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u],h=Jn(u);(u==="name"||u==="className")&&typeof c=="string"?s[h]=c:s[h]=Eh(c,u)}return s}}var Dh="2.6.0";function d3(n,t){if(n.dtype==null||n.dtype===t.dtype)return t;try{return y.cast(t,n.dtype)}catch(e){throw new M("The dtype of the feed ("+t.dtype+") can not be cast to the dtype "+("of the key '"+n.name+"' ("+n.dtype+")."))}}var Gr=function(){function n(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof n)for(var e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(var i=0,r=t;i<r.length;i++){var a=r[i];this.add(a.key,a.value)}}}return n.prototype.add=function(t,e,i){if(this.id2Value[t.id]==null)this.id2Value[t.id]=d3(t,e),this.name2Id[t.name]=t.id,i!=null&&(this.id2Mask[t.id]=i);else throw new M("Duplicate key: name="+t.name+", id="+t.id);return this},n.prototype.addFeed=function(t){this.add(t.key,t.value)},n.prototype.hasKey=function(t){return this.id2Value[t.id]!=null},n.prototype.names=function(){return Object.keys(this.name2Id)},n.prototype.getValue=function(t){if(t instanceof vn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Value[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Value[e]}},n.prototype.getMask=function(t){if(t instanceof vn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Mask[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Mask[e]}},n.prototype.disposeMasks=function(){this.id2Mask!=null&&y.dispose(this.id2Mask)},n}(),kh={},fy={};function Ha(n,t,e,i){for(var r=e==null?!1:e.training,a=Array.isArray(n),s=a?n:[n],o=s.map(function(X){return X.name}),l=[],u=t.names(),c=0,h=o;c<h.length;c++){var d=h[c];u.indexOf(d)!==-1?l.push(t.getValue(d)):l.push(null)}i!=null&&(i.maxNumTensors=-Infinity,i.minNumTensors=Infinity);var p=o.join(",")+"|"+t.names().join(","),f,m;if(kh[p]==null){var g=p3(s,t);f=g.sorted,m=g.recipientCounts,kh[p]=f,fy[p]=m}f=kh[p],m={},r||Object.assign(m,fy[p]);for(var v=new Gr(t),b=0;b<f.length;++b){if(i!=null){var w=y.memory().numTensors;w>i.maxNumTensors&&(i.maxNumTensors=w),w<i.minNumTensors&&(i.minNumTensors=w)}var S=f[b],L=S.sourceLayer;if(L instanceof Pa)continue;for(var x=[],C=[],R=[],D=!1,k=0,W=S.inputs;k<W.length;k++){var F=W[k],P=v.getValue(F),H=v.getMask(F);x.push(P),C.push(H),H!=null&&(D=!0),r||(m[F.name]--,m[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!P.isDisposed&&F.sourceLayer.stateful!==!0&&R.push(P))}D&&(e=e||{},e.mask=C[0]);var _=Je(L.apply(x,e)),K=null;L.supportsMasking&&(K=L.computeMask(x,C));for(var j=f3(S),q=Array.isArray(j)?j:[j],G=0;G<q.length;++G){v.hasKey(q[G])||v.add(q[G],_[G],Array.isArray(K)?K[0]:K);var Z=o.indexOf(q[G].name);Z!==-1&&(l[Z]=_[G])}r||y.dispose(R)}return v.disposeMasks(),a?l:l[0]}function p3(n,t){y.util.assert(n!=null&&n.length>0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=my(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s<o.length;s++){for(var l=o[s],u=my(l,t),c=u.sorted,h=u.recipientMap,d=0,p=c;d<p.length;d++){var f=p[d];a.has(f.name)||(e.push(f),a.add(f.name))}var m=function(v){i[v]==null&&(i[v]=new Set),h[v].forEach(function(b){return i[v].add(b)})};for(var g in h)m(g)}return{sorted:e,recipientCounts:m3(i)}}function m3(n){var t={};for(var e in n)t[e]=n[e].size;return t}function my(n,t){for(var e=new Set,i=[],r={},a=0,s=t.names();a<s.length;a++){var o=s[a];e.add(o)}var l=[],u=[];for(l.push(n);l.length>0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d<p.length;d++){var f=p[d];if(r[f.name]==null&&(r[f.name]=new Set),r[f.name].add(c.name),e.has(f.name))continue;l.push(f)}}}return{sorted:i,recipientMap:r}}function f3(n){var t;if(n.sourceLayer.inboundNodes.length===1)t=n.sourceLayer.output;else{for(var e=null,i=0;i<n.sourceLayer.inboundNodes.length;++i)for(var r=0,a=n.sourceLayer.inboundNodes[i].outputTensors;r<a.length;r++){var s=a[r];if(s.id===n.id){e=i;break}}t=n.sourceLayer.getOutputAt(e)}return t}var g3=function(n){Q(t,n);function t(e){var i=n.call(this,{})||this;if(i.containerNodes=new Set,i.name=e.name,i.name==null){var r=i.getClassName().toLowerCase();i.name=ao(r)}if(i.supportsMasking=!1,i.trainable_=!0,Array.isArray(e.inputs)?i.inputs=e.inputs.slice():i.inputs=[e.inputs],Array.isArray(e.outputs)?i.outputs=e.outputs.slice():i.outputs=[e.outputs],fi(i.inputs).length!==i.inputs.length)throw new M("The list of inputs passed to the model is redundant. All inputs should only appear once. Found: "+(""+i.inputs.map(function(Ce){return Ce.name})));fi(i.outputs).length!==i.outputs.length&&console.warn("The list of outputs passed to the model is redundant. All outputs should only appear once. Found: "+(""+i.outputs.map(function(Ce){return Ce.name}))),i.inputLayers=[],i.inputLayersNodeIndices=[],i.inputLayersTensorIndices=[],i.outputLayers=[],i.outputLayersNodeIndices=[],i.outputLayersTensorIndices=[],i.layers=[],i.internalContainerRefs=[];for(var a=0,s=i.outputs;a<s.length;a++){var o=s[a],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;i.outputLayers.push(l),i.outputLayersNodeIndices.push(u),i.outputLayersTensorIndices.push(c)}for(var h=0,d=i.inputs;h<d.length;h++){var o=d[h],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;Wn(u===0,"input layer has >1 nodes"),Wn(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p<i.inputLayers.length;p++){var l=i.inputLayers[p];if(!(l instanceof Pa))throw new TypeError("Input layers to a LayersModel must be InputLayer objects. "+("Received inputs: "+e.inputs+". ")+("Input "+p+" (0-based) originates ")+("from layer type "+l.getClassName()+"."));i.inputNames.push(l.name),i.feedInputShapes.push(l.batchInputShape),i.feedInputNames.push(l.name)}for(var f=0,m=i.outputLayers;f<m.length;f++){var l=m[f];i.outputNames.push(l.name)}i.internalInputShapes=i.inputs.map(function(Ce){return Ce.shape}),i.internalOutputShapes=i.outputs.map(function(Ce){return Ce.shape});for(var g={},v={},b={},w={},S={},L=[],x=function(Ce,kt,pt,it,sn,hs){(it==null||sn==null||hs==null)&&(it=Ce.sourceLayer,sn=Ce.nodeIndex,hs=Ce.tensorIndex);var Bt=it.inboundNodes[sn];if(pt.indexOf(Bt)!==-1)throw new Fn("The tensor "+Ce.name+' at layer "'+it.name+'" is part of a cycle.');if(kt.indexOf(Bt)!==-1)return;i.containerNodes.add(t.nodeKey(it,sn)),it.id in S||(S[it.id]=Object.keys(S).length),pt.indexOf(Bt)===-1&&pt.push(Bt);for(var II=Bt.inboundLayers.length,yr=0;yr<II;yr++){var AI=Bt.inputTensors[yr],TI=Bt.inboundLayers[yr],NI=Bt.nodeIndices[yr],xI=Bt.tensorIndices[yr];x(AI,kt,pt,TI,NI,xI)}for(kt.push(Bt);pt.indexOf(Bt)>=0;)pt.splice(pt.indexOf(Bt),1);L.push(Bt)},C=[],R=[],D=0,k=i.outputs;D<k.length;D++){var o=k[D];x(o,C,R)}for(var W=L.slice().reverse(),F=0,P=W;F<P.length;F++){var H=P[F];v[H.id]=H,H.id in g||(g[H.id]=0);var _=g[H.id],K=b[H.outboundLayer.id]==null?0:b[H.outboundLayer.id];_=Math.max(_,K),b[H.outboundLayer.id]=_,w[H.outboundLayer.id]=H.outboundLayer,g[H.id]=_;for(var p=0;p<H.inboundLayers.length;p++){var j=H.inboundLayers[p],u=H.nodeIndices[p],q=j.inboundNodes[u],G=g[q.id]==null?0:g[q.id];g[q.id]=Math.max(_+1,G),v[q.id]=q}}var Z={};for(var X in g){var _=g[X];_ in Z||(Z[_]=[]),Z[_].push(v[X])}var ee={};for(var ne in b){var _=b[ne];_ in ee||(ee[_]=[]),ee[_].push(w[ne])}var ie=Object.keys(ee).map(function(Ce){return parseInt(Ce,10)}).sort(to);i.layers=[];for(var te=0,re=ie;te<re.length;te++){var _=re[te],le=ee[_];le.sort(function(kt,pt){var it=S[kt.id],sn=S[pt.id];return it<sn?-1:it>sn?1:0});for(var he=0,ye=le;he<ye.length;he++){var l=ye[he];l instanceof t&&i.internalContainerRefs.push(l),i.layers.push(l)}}i.layersByDepth=ee,ie=Object.keys(Z).map(function(Ce){return parseInt(Ce,10)}).sort(to);for(var Oe=i.inputs.slice(),Fe=[],_e=0,Me=ie;_e<Me.length;_e++)for(var _=Me[_e],ht=0,St=Z[_];ht<St.length;ht++){var H=St[ht],l=H.outboundLayer;if(l!=null){for(var Ge=0,Ue=H.inputTensors;Ge<Ue.length;Ge++){var o=Ue[Ge];if(Oe.indexOf(o)===-1)throw new Fn("Graph disconnected: cannot obtain value for tensor "+o+(' at layer "'+l.name+'". ')+"The following previous layers were accessed without "+("issue: "+Fe))}for(var me=0,Dt=H.outputTensors;me<Dt.length;me++){var o=Dt[me];Oe.push(o)}Fe.push(l.name)}}i.nodesByDepth=Z;for(var lt=i.layers.map(function(Ce){return Ce.name}),dt=function(Ce){var kt=lt.filter(function(pt){return pt===Ce}).length;if(kt!==1)throw new Fn('The name "'+Ce+'" is used '+kt+" times in the model. All layer names should be unique. Layer names: "+JSON.stringify(lt))},xn=0,Rt=lt;xn<Rt.length;xn++){var Lt=Rt[xn];dt(Lt)}return i.outboundNodes=[],i.inboundNodes=[],new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:i.inputs,outputTensors:i.outputs,inputMasks:i.inputs.map(function(Ce){return null}),outputMasks:i.outputs.map(function(Ce){return null}),inputShapes:i.inputs.map(function(Ce){return Ce.shape}),outputShapes:i.outputs.map(function(Ce){return Ce.shape})}),i.built=!0,i._refCount=1,i}return t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Container '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){this.assertNotDisposed();var e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(var i=0,r=this.layers;i<r.length;i++){var a=r[i];e.numDisposedVariables+=a.dispose().numDisposedVariables}for(var s=0,o=this.internalContainerRefs;s<o.length;s++){var l=o[s];e.numDisposedVariables+=l.dispose().numDisposedVariables}}return e.refCountAfterDispose=this._refCount,e},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this.layers.forEach(function(i){i._trainableWeights.forEach(function(r){return r.trainable=e})}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(this._trainableWeights.length>0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e=e.concat(a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.layers;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.loadWeights=function(e,i){i===void 0&&(i=!0);for(var r={},a=0,s=0,o=this.layers;s<o.length;s++)for(var l=o[s],u=0,c=l.weights;u<c.length;u++){var h=c[u];if(r[h.originalName]!=null)throw new M("Duplicate weight name: "+h.originalName);r[h.originalName]=h,a++}var d=[];for(var p in e){var f=p;if(r[p]==null){var m=p.split("/"),g=m.slice(0,-2).concat([m[m.length-1]]);f=g.join("/")}if(r[f]!=null)d.push([r[f],e[p]]);else if(i)throw new M("Provided weight data has no target variable: "+p);delete r[f]}if(i){var v=[];for(var b in r)v.push(b);if(v.length>0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}Ah(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+Dh,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=Eh(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Je(e);for(var a=new Gr,s=0;s<r.inputs.length;++s)a.add(r.inputs[s],e[s]);return Ha(r.outputs,a,i)})},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){e=Je(e);var a;return i==null?a=Ji(null,e.length):a=Je(i),r.runInternalGraph(e,a)[1]})},t.prototype.computeOutputShape=function(e){var i=so(e);if(i.length!==this.inputLayers.length)throw new M("Invalid inputShape argument "+e+": "+("model has "+this.inputLayers.length+" tensor inputs."));for(var r={},a=0;a<i.length;a++){var s=this.inputLayers[a],o=i[a],l=s.name+"_0_0";r[l]=o}var u=Object.keys(this.nodesByDepth).map(function(P){return parseInt(P,10)}).sort(to);if(u.length>1)for(var c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){var g=m[f],s=g.outboundLayer;if(this.inputLayers.map(function(H){return H.id}).indexOf(s.id)!==-1)continue;for(var v=[],b=0;b<g.inboundLayers.length;b++){var w=g.inboundLayers[b],S=g.nodeIndices[b],L=g.tensorIndices[b],l=w.name+"_"+S+"_"+L,x=r[l];v.push(x)}for(var C=s.computeOutputShape(Mt(v)),R=so(C),D=s.inboundNodes.indexOf(g),b=0;b<R.length;b++){var l=s.name+"_"+D+"_"+b;r[l]=R[b]}}for(var k=[],W=[],a=0;a<this.outputLayers.length;a++){var s=this.outputLayers[a],D=this.outputLayersNodeIndices[a],L=this.outputLayersTensorIndices[a],l=s.name+"_"+D+"_"+L;W.push(l)}for(var a=0;a<W.length;a++){var F=W[a];Wn(F in r),k.push(r[F])}return Mt(k)},t.prototype.runInternalGraph=function(e,i){i==null&&(i=Ji(null,e.length));for(var r={},a=0;a<this.inputs.length;++a){var s=this.inputs[a],o=e[a],l=i[a];r[s.id]=[o,l]}for(var u=Object.keys(this.nodesByDepth).map(function(ee){return parseInt(ee,10)}).sort(to),c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){for(var g=m[f],v=g.outboundLayer,b=g.inputTensors,w=g.outputTensors,S=new Array,L=0,x=b;L<x.length;L++){var s=x[L];s.id in r&&S.push(r[s.id])}if(S.length===b.length){var C={},R=void 0,D=void 0,k=void 0,W=void 0;if(g.callArgs!=null&&(C=g.callArgs),S.length===1){var F=S[0],P=F[0],H=F[1];C.mask==null&&(C.mask=H),k=Je(v.call(P,C)),W=Je(v.computeMask(P,H)),R=[P],D=[H]}else R=S.map(function(ee){return ee[0]}),D=S.map(function(ee){return ee[1]}),C.mask==null&&(C.mask=D),k=Je(v.call(R,C)),W=Je(v.computeMask(R,D));if(v.activityRegularizer)throw new Te("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(var a=0;a<w.length;++a){var s=w[a],o=k[a],l=W[a];r[s.id]=[o,l]}}}for(var _=[],K=[],j=[],q=0,G=this.outputs;q<G.length;q++){var s=G[q];Wn(s.id in r,"Could not compute output "+s.name+" : "+s.id);var Z=r[s.id],X=Z[0],l=Z[1];j.push(X.shape),_.push(X),K.push(l)}return[_,K,j]},t.prototype.buildNodeConversionMap=function(e){for(var i={},r,a=0,s=this.layers;a<s.length;a++){var o=s[a];r=o instanceof t?1:0;for(var l=0;l<o.inboundNodes.length;l++){var u=t.nodeKey(o,l);this.containerNodes.has(u)&&(i[u]=r,r+=1)}}return i},t.prototype.getLayer=function(e,i){if(i!=null){if(this.layers.length<=i)throw new M("Was asked to retrieve layer at index "+i+", but model only "+("has "+this.layers.length+" layer(s)."));return this.layers[i]}else if(e==null)throw new M("Provide either a layer name or layer index");for(var r=0,a=this.layers;r<a.length;r++){var s=a[r];if(s.name===e)return s}throw new M("No such layer: "+e)},t.prototype.calculateLosses=function(){var e=this;return y.tidy(function(){for(var i=[],r=0,a=e.layers;r<a.length;r++)for(var s=a[r],o=0;o<s.inboundNodes.length;++o){var l=t.nodeKey(s,o);e.containerNodes.has(l)&&i.push.apply(i,s.calculateLosses())}return i})},t.prototype.getConfig=function(){for(var e={name:this.name},i=this.buildNodeConversionMap(this.layers),r=[],a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=o.getClassName(),u=o.getConfig(),c=[],h=0;h<o.inboundNodes.length;h++){var d=o.inboundNodes[h],p=t.nodeKey(o,h),f={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),f=d.callArgs}catch(D){console.warn("Layer "+o.name+" was passed non-serializable keyword arguments: "+(d.callArgs+". They will not be included ")+"in the serialized model (and thus will be missing at deserialization time)."),f={}}if(d.inboundLayers.length>0){for(var m=[],g=0;g<d.inboundLayers.length;g++){var v=d.inboundLayers[g],b=d.nodeIndices[g],w=d.tensorIndices[g],S=t.nodeKey(v,b),L=i[S];L==null&&(L=0),m.push([v.name,L,w,f])}c.push(m)}}}var x={};x.name=o.name,x.className=l,x.config=u,x.inboundNodes=c,r.push(x)}e.layers=r;for(var C=[],g=0;g<this.inputLayers.length;g++){var o=this.inputLayers[g],b=this.inputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.inputLayersTensorIndices[g];C.push([o.name,L,w])}e.inputLayers=C;for(var R=[],g=0;g<this.outputLayers.length;g++){var o=this.outputLayers[g],b=this.outputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.outputLayersTensorIndices[g];R.push([o.name,L,w])}return e.outputLayers=R,e},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s={},o={};function l(G,Z){G.name in o?o[G.name].push(Z):o[G.name]=[Z]}function u(G,Z){for(var X=[],ee,ne=0,ie=Z;ne<ie.length;ne++){var te=ie[ne],re=te[0],le=te[1],he=te[2];if(ee=te[3]==null?{}:te[3],!(re in s)){l(G,Z);return}var ye=s[re];if(ye.inboundNodes.length<=le){l(G,Z);return}var Oe=ye.inboundNodes[le];X.push(Oe.outputTensors[he])}X.length>0&&G.apply(Mt(X),ee)}function c(G){var Z=G.name,X=yn(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p<f.length;p++){var m=f[p];c(m)}for(;!YD(o);)for(var g=0,v=d;g<v.length;g++){var m=v[g],b=s[m.name];if(b.name in o){var w=o[b.name];delete o[b.name];for(var S=0,L=w;S<L.length;S++){var x=L[S];u(b,x)}}}for(var C=[],R=[],D=i.inputLayers,k=0,W=D;k<W.length;k++){var m=W[k],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;C.push(_[H])}for(var K=i.outputLayers,j=0,q=K;j<q.length;j++){var m=q[j],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;R.push(_[H])}return new e({inputs:C,outputs:R,name:h})},Object.defineProperty(t.prototype,"stateful",{get:function(){if(this._stateful)throw new M("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(var e=0,i=this.layers;e<i.length;e++){var r=i[e];if(r.stateful)return!0}return!1},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){var e=this;y.tidy(function(){e.layers.forEach(function(i){i.stateful&&i.resetStates()})})},t}(De);function v3(n,t,e){var i=t.length;if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(a){return null});if(i===1)return Array.isArray(n)&&n.length===1?n:typeof n=="object"&&t[0]in n?[n[t[0]]]:[n];if(Array.isArray(n)){if(n.length!==i)throw new Error("Provided "+e+" is an array of "+n.length+" "+("element(s), but the model has "+i+" outputs. ")+"Make sure a set of weights is provided for each model output.");return n}else if(typeof n=="object"&&Object.keys(n).length>0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function gy(n,t){return v3(n,t,"classWeight")}function vy(n,t,e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function y3(n,t){return y.mul(n,t)}var b3=32;function by(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor.  The provided Dataset instead generates "+(""+t)});var a=yy("input",n.inputNames,e),s=yy("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs.  (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs.  (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},u=0;u<a.length;u++)l(u);for(var c=function(d){y.util.assert(s[d].shape[0]===o,function(){return"Batch size mismatch: output "+(n.outputNames[d]+" has "+s[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},h=0;h<s.length;h++)c(h);return{xs:a,ys:s}}function yy(n,t,e){if(e instanceof y.Tensor)return[e];if(Array.isArray(e))return y.util.assert(e.length===t.length,function(){return"Received an array of "+e.length+" Tensors, but expected "+t.length+" to match the "+n+" keys "+t+"."}),e;for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r];if(e[s]==null)throw new M("The feature data generated by the dataset lacks the required "+(n+" key '"+s+"'."));i.push(e[s])}return i}function w3(n){if(n.length===3)throw new Te("Validation with sample weights is not implemented yet.");return{xs:n[0],ys:n[1]}}function L3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k,W,F,P,H,_,K,F,j,q,G,Z,F;return ve(this,function(X){switch(X.label){case 0:if(i=e.batchesPerEpoch!=null,y.util.assert(n.optimizer!=null,function(){return"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."}),y.util.assert(e!=null,function(){return"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."}),y.util.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(wy(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=w3(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=iy(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=ay(h,d,e.epochs,null,null,S3(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g<e.epochs?(b={},[4,f.onEpochBegin(g)]):[3,23];case 5:return X.sent(),w=0,S=0,i?[3,7]:[4,t.iterator()];case 6:v=X.sent(),X.label=7;case 7:return(i?w<e.batchesPerEpoch:!0)?[4,v.next()]:[3,21];case 8:return L=X.sent(),i&&L.done?(console.warn("You provided `batchesPerEpoch` as "+(e.batchesPerEpoch+", ")+"but your dataset iterator ran out of data after "+(w+" batches; ")+"interrupting training. Make sure that your dataset can generate at least `batchesPerEpoch * epochs` batches (in this case, "+(e.batchesPerEpoch*e.epochs+" batches). ")+"You may need to use the repeat() function when building your dataset."),[3,21]):L.value!=null?(x=by(n,L.value),C=x.xs,R=x.ys,D={},D.batch=S,D.size=C[0].shape[0],[4,f.onBatchBegin(S,D)]):[3,15];case 9:if(X.sent(),k=[],!(e.classWeight!=null))return[3,13];W=gy(e.classWeight,n.outputNames),F=0,X.label=10;case 10:return F<W.length?(H=(P=k).push,[4,vy(R[F],null,W[F])]):[3,13];case 11:H.apply(P,[X.sent()]),X.label=12;case 12:return++F,[3,10];case 13:for(_=C.concat(R).concat(k),K=l(_),y.dispose(_),F=0;F<u.length;++F)j=u[F],q=K[F],D[j]=q,y.keep(q);return[4,f.onBatchEnd(S,D)];case 14:X.sent(),Zv(D),S++,w++,X.label=15;case 15:return(i?w>=e.batchesPerEpoch:L.done)?r?(G=void 0,wy(e.validationData)?(Z=Je,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Je(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?b3:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F<n.metricsNames.length;++F)b["val_"+n.metricsNames[F]]=G[F];X.label=19;case 19:return[3,21];case 20:return n.stopTraining_?[3,21]:[3,7];case 21:return[4,f.onEpochEnd(g,b)];case 22:return X.sent(),g++,n.stopTraining_?[3,23]:[3,4];case 23:return[4,f.onTrainEnd()];case 24:return X.sent(),[4,n.history.syncData()];case 25:return X.sent(),[2,n.history];case 26:return n.isTraining=!1,[7];case 27:return[2]}})})}function S3(n,t){var e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(n.size)&&(e=n.size),e}function wy(n){return typeof n.iterator=="function"}function I3(n){return typeof n.next=="function"}function A3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:if(e=e||{},i=e.batches!=null,r=n.testFunction,a=[],e.verbose>0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),I3(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=by(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var x=0;x<L.length;++x)a.push(y.scalar(0));for(var C=S[0].shape[0],R=function(D){var k=L[D],W=a[D];a[D]=y.tidy(function(){return y.add(a[D],y.mul(C,k))}),u>0&&y.dispose(W)},x=0;x<L.length;++x)R(x);y.dispose(L),l+=C,++u}return a}),m.done?(i&&console.warn("Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least `batches` "+("batches (in this case, "+e.batches+" batches). ")+"You may need to use the repeat() function when building your dataset."),[2,"break"]):[2]}})},f.label=4;case 4:return(i?u<e.batches:!0)?[5,c()]:[3,6];case 5:return h=f.sent(),h==="break"?[3,6]:[3,4];case 6:for(d=0;d<a.length;++d)p=a[d],a[d]=y.div(a[d],l),y.dispose(p);return[2,Mt(a)]}})})}function Fh(n){y.util.assert(n>0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function Va(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return er(i,t,e-t)}):er(n,t,e-t)}function Wh(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return Wh(e,t)}):Uv(n,t.dtype==="int32"?t:t.toInt())})}function Uh(n,t){for(var e=[],i=0,r=null;i<n;)r=i+t,r>=n&&(r=n),e.push([i,r]),i=r;return e}function T3(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Le(this,void 0,void 0,function(){var m,g,v,b,w,S,L,x,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=gn(0,g)),s==null&&(s=1),b=ay(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(K){switch(K.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(K.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=Uh(g,r),P=function(j){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(j,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[j][0],X=F[j][1],ee=er(W,Z,X-Z);q.batch=j,q.size=X-Z;for(var ne=Wh(e,ee),ie=t(ne),te=0;te<i.length;++te){var re=i[te],le=ie[te];q[re]=le,y.keep(le)}if(j===F.length-1&&m)for(var he=n.testLoop(l,u,r),te=0;te<i.length;++te){var re=i[te],le=he[te];y.keep(le),k["val_"+re]=le}}),[4,w.onBatchEnd(j,q)];case 2:return G.sent(),Zv(q),n.stopTraining_?[2,"break"]:[2]}})},H=0,K.label=3;case 3:return H<F.length?[5,P(H)]:[3,6];case 4:if(_=K.sent(),_==="break")return[3,6];K.label=5;case 5:return++H,[3,3];case 6:W.dispose(),K.label=7;case 7:return[4,w.onEpochEnd(D,k)];case 8:return K.sent(),n.stopTraining_?[2,"break"]:[2]}})},x=d,R.label=2;case 2:return x<a?[5,L(x)]:[3,5];case 3:if(C=R.sent(),C==="break")return[3,5];R.label=4;case 4:return++x,[3,2];case 5:return[4,w.onTrainEnd()];case 6:return R.sent(),[4,n.history.syncData()];case 7:return R.sent(),[2,n.history]}})})}function N3(n,t,e,i){return i===void 0&&(i={}),Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k;return ve(this,function(W){switch(W.label){case 0:if(n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,W.label=1;case 1:return W.trys.push([1,,7,8]),h=i.batchSize==null?32:i.batchSize,Fh(h),d=!1,[4,n.standardizeUserData(t,e,i.sampleWeight,i.classWeight,d,h)];case 2:if(p=W.sent(),r=p[0],a=p[1],c=p[2],f=!1,m=void 0,!(i.validationData!=null&&i.validationData.length>0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=Va(r,b,w),r=Va(r,0,b),u=Va(a,b,w),a=Va(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),x=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=x.slice().concat(x.map(function(F){return"val_"+F}))):(C=null,m=[],R=x.slice()),D=iy(i.callbacks,i.yieldEvery),[4,T3(n,L,S,x,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,nr(r,t),nr(a,e),nr(l,s),nr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function Sy(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e<n.length;++e){var i=n[e];if(i.rank===1)t.push(Ua(i,1));else{if(i.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(i)}}return t}function nr(n,t){if(n==null)return;var e=[];if(t instanceof y.Tensor)e.push(t.id);else if(Array.isArray(t))t.forEach(function(l){return e.push(l.id)});else if(t!=null)for(var i in t){var r=t[i];e.push(r.id)}var a=[];if(n instanceof y.Tensor)e.indexOf(n.id)===-1&&a.push(n);else if(Array.isArray(n))n.forEach(function(l){e.indexOf(l.id)===-1&&a.push(l)});else if(n!=null)for(var s in n){var o=n[s];e.indexOf(o.id)===-1&&a.push(o)}a.forEach(function(l){l.isDisposed||l.dispose()})}function x3(n){return n instanceof y.Tensor}function Bh(n){return Array.isArray(n)}function Ly(n){return!x3(n)&&!Bh(n)}function Iy(n,t,e,i,r){if(i===void 0&&(i=!0),r===void 0&&(r=""),t==null||t.length===0){if(n!=null){var a=!1;if(Bh(n)&&n.length>0)a=!0;else if(Ly(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(Ly(n)){n=n,o=[];for(var l=0,u=t;l<u.length;l++){var c=u[l];if(n[c]==null)throw new M('No data provided for "'+c+'". Need data for each key in: '+(""+t));o.push(n[c])}}else if(Bh(n)){if(n=n,n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("model expected. Expected to see "+t.length+" Tensor(s), but ")+("instead got the following list of Tensor(s): "+n));o=n}else{if(n=n,t.length>1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=Sy(o),e!=null)for(var h=0;h<t.length;++h){if(e[h]==null)continue;var d=o[h];if(d.shape.length!==e[h].length)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have "+e[h].length+" dimension(s). but got array with ")+("shape "+d.shape));for(var p=0;p<e[h].length;++p){if(p===0&&!i)continue;var f=d.shape[p],m=e[h][p];if(m!=null&&m>=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function C3(n,t,e){var i=fi(n.map(function(a){return a.shape[0]}));i.sort();var r=fi(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function R3(n,t,e){for(var i=[tr,po,_a],r=0;r<n.length;++r){var a=n[r],s=t[r],o=e[r];if(s==null)continue;if(s===_a&&a.shape[a.shape.length-1]===1)throw new M("You are passing a target array of shape "+a.shape+" while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].");if(i.indexOf(s)!==-1)for(var l=a.shape.slice(1),u=o.slice(1),c=0;c<l.length;++c){var h=l[c],d=u[c];if(d!=null&&h!==d)throw new M("A target Tensor with shape "+a.shape+" was passed for an "+("output of shape "+o+", while using a loss function that ")+"expects targets to have the same shape as the output.")}}}function Ay(n,t,e,i,r){i===void 0&&(i=!0),r===void 0&&(r="");var a;if(Array.isArray(n)){if(n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("the model expected. Expected to see "+t.length+" Tensor(s),")+(" but instead got "+n.length+" Tensors(s)."));a=n}else{if(t.length>1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s<t.length;++s){if(e[s]==null)continue;var o=a[s];if(o.shape.length!==e[s].length)throw new M("Error when checking "+r+": expected "+t[s]+" "+("to have "+e[s].length+" dimension(s), but got array with ")+("shape "+JSON.stringify(o.shape)));for(var l=0;l<e[s].length;++l){if(l===0&&!i)continue;var u=o.shape[l],c=e[s][l];if(c!=null&&c!==u)throw new M("Error when checking "+r+": expected "+(t[s]+" to have shape "+JSON.stringify(e[s])+" but ")+("got array with shape "+JSON.stringify(o.shape)+"."))}}}function O3(n,t){if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(l){return[]});var e;if(typeof n=="string"||typeof n=="function")e=[n];else if(Array.isArray(n)||typeof n=="object")e=n;else throw new TypeError("Type of metrics argument not understood. Expected an string,"+("function, Array, or Object, found: "+n));if(Array.isArray(e))return t.map(function(l){return e});for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r],o=e.hasOwnProperty(s)?e[s]:[];Array.isArray(o)||(o=[o]),i.push(o)}return i}var E3="layers-model",yi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.isTraining=!1,i}return t.prototype.summary=function(e,i,r){if(r===void 0&&(r=console.log),!this.built)throw new M("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");h3(this,e,i,r)},t.prototype.compile=function(e){var i=this;if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=s3(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof y.Optimizer))throw new M("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}var r=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(var a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new M('Unknown entry in loss dictionary: "'+a+'". '+("Only expected the following keys: "+this.outputNames));for(var s=0,o=this.outputNames;s<o.length;s++){var l=o[s];e.loss[l]==null&&console.warn('Output "'+l+'" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data '+("to be passed to "+l+" during training")),r.push(Nh(e.loss[l]))}}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new M("When passing an Array as loss, it should have one entry per "+("model output. The model has "+this.outputs.length+" output(s), ")+("but you passed loss="+e.loss+"."));var u=e.loss;r=u.map(function(v){return Nh(v)})}else{var c=Nh(e.loss);this.outputs.forEach(function(v){r.push(c)})}this.lossFunctions=r,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(var h=0;h<this.outputs.length;++h){var d=this.internalOutputShapes[h],p=this.outputNames[h];this.feedOutputNames.push(p),this.feedOutputShapes.push(d),this.feedLossFns.push(this.lossFunctions[h])}var f=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],Qi("loss",function(){for(var v=0;v<i.outputs.length;++v){if(f.indexOf(v)!==-1)continue;var b=i.lossFunctions[v];i.outputs.length>1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=O3(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};Qi("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(x){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var K=i.internalOutputShapes[w];K[K.length-1]===1||i.lossFunctions[w]===po?["accuracy","acc"].indexOf(_)!==-1?D=xh:["crossentropy","ce"].indexOf(_)!==-1&&(D=ly):i.lossFunctions[w]===ho?["accuracy","acc"].indexOf(_)!==-1?D=uy:["crossentropy","ce"].indexOf(_)!==-1&&(D=cy):["accuracy","acc"].indexOf(_)!==-1?D=Ch:["crossentropy","ce"].indexOf(_)!==-1&&(D=Rh);var j=void 0;["accuracy","acc"].indexOf(_)!==-1?j="acc":["crossentropy","ce"].indexOf(_)!==-1&&(j="ce"),k=D,R=C+j}else{var q=a3(_);k=q,R=C+go(_)}var G;Qi(R,function(){G=k}),g(w,R,G)},F=0,P=x;F<P.length;F++){var H=P[F];W(H)}};L(S)},b=0;b<i.outputs.length;++b)v(b)}),this.collectedTrainableWeights=this.trainableWeights},t.prototype.checkTrainableWeightsConsistency=function(){if(this.collectedTrainableWeights==null)return;this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")},t.prototype.evaluate=function(e,i,r){r===void 0&&(r={});var a=r.batchSize==null?32:r.batchSize;Fh(a);var s=!0,o=this.standardizeUserDataXY(e,i,s,a);try{var l=o[0].concat(o[1]);this.makeTestFunction();var u=this.testFunction,c=this.testLoop(u,l,a,r.verbose,r.steps);return Mt(c)}finally{nr(o[0],e),nr(o[1],i)}},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return this.makeTestFunction(),[2,A3(this,e,i)]})})},t.prototype.checkNumSamples=function(e,i,r,a){a===void 0&&(a="steps");var s;if(r!=null){if(s=null,i!=null)throw new M("If "+a+" is set, batchSize must be null or undefined."+("Got batchSize = "+i))}else if(e!=null)Array.isArray(e)?s=e[0].shape[0]:s=e.shape[0];else throw new M("Either the input data should have a defined shape, or "+(a+" shoud be specified."));return s},t.prototype.execute=function(e,i){if(Array.isArray(i)&&i.length===0)throw new M("`outputs` is an empty Array, which is not allowed.");var r=Array.isArray(i),a=r?i:[i],s=this.retrieveSymbolicTensors(a),o=new Gr;if(e instanceof y.Tensor&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new M("The number of inputs provided ("+e.length+") does not match the number of inputs of this model "+("("+this.inputs.length+")."));for(var l=0;l<this.inputs.length;++l)o.add(this.inputs[l],e[l])}else for(var u=0,c=this.inputs;u<c.length;u++){var h=c[u],d=e[h.name];if(d==null)throw new M("No value is provided for the model's input "+h.name);o.add(h,d)}var p=Ha(s,o);return r?p:p[0]},t.prototype.retrieveSymbolicTensors=function(e){for(var i=Ji(null,e.length),r=e.length,a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=Array.isArray(o.output)?o.output:[o.output],u=l.map(function(p){return p.name}),c=0;c<e.length;++c){var h=u.indexOf(e[c]);if(h!==-1&&(i[c]=l[h],r--),r===0)break}if(r===0)break}if(r>0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=Uh(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=Va(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v<m.length;++v)g.push({key:a.inputs[v],value:m[v]});else g.push({key:a.inputs[0],value:m});var b=new Gr(g);return Ha(a.outputs,b)});d.forEach(function(p,f){return l[f].push(p)})},c=0;c<o.length;++c)u(c);return Mt(l.map(function(h){return y.concat(h,0)}))})},t.prototype.predict=function(e,i){i===void 0&&(i={});var r=Sy(e);Ay(r,this.inputNames,this.feedInputShapes,!1);try{var a=i.batchSize==null?32:i.batchSize;return Fh(a),this.predictLoop(r,a)}finally{nr(r,e)}},t.prototype.predictOnBatch=function(e){Ay(e,this.inputNames,this.feedInputShapes,!0);var i=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,i)},t.prototype.standardizeUserDataXY=function(e,i,r,a){if(this.optimizer_==null)throw new Fn("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");for(var s=[],o=0;o<this.feedOutputShapes.length;++o){var l=this.feedOutputShapes[o],u=this.feedLossFns[o];u===ho?s.push(l.slice(0,l.length-1).concat([1])):s.push(l)}if(e=Iy(e,this.feedInputNames,this.feedInputShapes,!1,"input"),i=Iy(i,this.feedOutputNames,s,!1,"target"),C3(e,i),R3(i,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Le(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=gy(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p<d.length?(m=(f=h).push,[4,vy(c[p],null,d[p])]):[3,4];case 2:m.apply(f,[g.sent()]),g.label=3;case 3:return++p,[3,1];case 4:return[2,[u,c,h]]}})})},t.prototype.testLoop=function(e,i,r,a,s){var o=this;return a===void 0&&(a=0),y.tidy(function(){var l=o.checkNumSamples(i,r,s,"steps"),u=[];if(a>0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=Uh(l,r),h=y.tensor1d(gn(0,l)),d=0;d<c.length;++d){var p=c[d][0],f=c[d][1],m=er(h,p,f-p),g=Wh(i,m),v=e(g);if(d===0)for(var b=0;b<v.length;++b)u.push(y.scalar(0));for(var b=0;b<v.length;++b){var w=v[b];u[b]=y.add(u[b],y.mul(f-p,w))}}for(var b=0;b<u.length;++b)u[b]=y.div(u[b],l);return u})},t.prototype.getDedupedMetricsNames=function(){for(var e=this.metricsNames,i=[],r=0;r<e.length;++r){var a=e[r],s=a;if(wv(e,a)>1){var o=wv(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f<e.inputs.length;++f)p.push({key:e.inputs[f],value:a[f]});for(var m=new Gr(p),g=Ha(e.outputs,m,{training:!0}),v,f=0;f<e.lossFunctions.length;++f){var b=e.lossFunctions[f],w=b(s[f],g[f]);o[f]!=null&&(w=y3(w,o[f]));var S=y.mean(w);r.push(S),f===0?v=w:v=y.add(v,w)}for(var f=0;f<e.metricsTensors.length;++f){var L=void 0;if(e.outputs.length>1&&f<e.outputs.length)L=r[f];else{var x=e.metricsTensors[f][0],C=e.metricsTensors[f][1];L=y.mean(x(s[C],g[C]))}y.keep(L),l.push(L)}return v=y.mean(v),e.calculateLosses().forEach(function(R){v=y.add(v,R)}),v},c=e.collectedTrainableWeights.map(function(p){return p.read()}),h=!0,d=e.optimizer_.minimize(u,h,c);return[d].concat(l)}},t.prototype.makeTestFunction=function(){var e=this;this.testFunction=function(i){return y.tidy(function(){for(var r=[],a,s=i.slice(0,e.inputs.length),o=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),l=[],u=0;u<e.inputs.length;++u)l.push({key:e.inputs[u],value:s[u]});for(var c=new Gr(l),h=Ha(e.outputs,c),u=0;u<e.lossFunctions.length;++u){var d=e.lossFunctions[u],p=y.mean(d(o[u],h[u]));u===0?a=p:a=y.add(a,p),r.push(a)}for(var u=0;u<e.metricsTensors.length;++u){var f=e.metricsTensors[u][0],m=e.metricsTensors[u][1],g=y.mean(f(o[m],h[m]));r.push(g)}return r})}},t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){return[2,N3(this,e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,L3(this,e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:return[4,this.standardizeUserData(e,i)];case 1:r=f.sent(),a=r[0],s=r[1],o=this.makeTrainFunction(),l=o(a.concat(s)),u=[],c=0,h=l,f.label=2;case 2:return c<h.length?(d=h[c],[4,d.data()]):[3,5];case 3:p=f.sent(),u.push(p[0]),f.label=4;case 4:return c++,[3,2];case 5:return y.dispose(l),[2,Mt(u)]}})})},t.prototype.getNamedWeights=function(e){for(var i=[],r=e!=null&&e.trainableOnly,a=r?this.trainableWeights:this.weights,s=this.getWeights(r),o=0;o<a.length;++o){if(r&&!a[o].trainable)continue;i.push({name:a[o].originalName,tensor:s[o]})}return i},Object.defineProperty(t.prototype,"stopTraining",{get:function(){return this.stopTraining_},set:function(e){this.stopTraining_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.optimizer_},set:function(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)},enumerable:!0,configurable:!0}),t.prototype.dispose=function(){var e=n.prototype.dispose.call(this);if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){var i=y.memory().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=i-y.memory().numTensors}return e},t.prototype.getLossIdentifiers=function(){var e;if(typeof this.loss=="string")e=Jn(this.loss);else if(Array.isArray(this.loss)){for(var i=0,r=this.loss;i<r.length;i++){var a=r[i];if(typeof a!="string")throw new Error("Serialization of non-string loss is not supported.")}e=this.loss.map(function(h){return Jn(h)})}else{var s=Object.keys(this.loss);e={};for(var o=this.loss,l=0,u=s;l<u.length;l++){var c=u[l];if(typeof o[c]=="string")e[c]=Jn(o[c]);else throw new Error("Serialization of non-string loss is not supported.")}}return e},t.prototype.getMetricIdentifiers=function(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[Jn(go(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(function(r){return Jn(go(r))});var e={};for(var i in this.metrics)e[i]=Jn(go(this.metrics[i]));return e},t.prototype.getTrainingConfig=function(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}},t.prototype.loadTrainingConfig=function(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");var i=Ma(e.optimizer_config),r=yn(i),a;if(typeof e.loss=="string")a=Zi(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(function(l){return Zi(l)});else if(e.loss!=null){a={};for(var s in e.loss)a[s]=Zi(e.loss[s])}var o;if(Array.isArray(e.metrics))o=e.metrics.map(function(l){return Zi(l)});else if(e.metrics!=null){o={};for(var s in e.metrics)o[s]=Zi(e.metrics[s])}this.compile({loss:a,metrics:o,optimizer:r})},t.prototype.save=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b;return ve(this,function(w){switch(w.label){case 0:if(typeof e=="string"){if(r=y.io.getSaveHandlers(e),r.length===0)throw new M("Cannot find any save handlers for URL '"+e+"'");if(r.length>1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:E3,generatedBy:"TensorFlow.js tfjs-layers v"+Dh,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,dy(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){dy(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(g3);y.serialization.registerClass(yi);var D3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(yi);y.serialization.registerClass(D3);function k3(n,t){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=Ma(e),r=yn(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o<l.length;o++)u=l[o],s[u.originalName]=a[u.originalName];r.loadWeights(s),y.dispose(a),c.label=2;case 2:return[2,r]}})})}function W3(n,t){return Le(this,void 0,void 0,function(){var e;return ve(this,function(i){if(t==null&&(t={}),typeof n=="string"){if(e=y.io.getLoadHandlers(n,t),e.length===0)e.push(y.io.browserHTTPRequest(n,t));else if(e.length>1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,F3(n,void 0,t)]})})}function F3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=yn(Ma(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=U3(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function U3(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var zh=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:ao("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r<a.length;r++){var s=a[r];i.add(s)}return i}return t.prototype.checkShape=function(e){var i=e.inboundNodes[0].outputTensors[0].shape;if(i.some(function(r){return r<0}))throw new M("Negative dimension size caused by adding layer "+(e.name+" with input shape [")+(e.inboundNodes[0].inputTensors[0].shape+"]"))},t.prototype.add=function(e){var i=e instanceof t||e instanceof yi,r;if(i){if(r=e,r.outputs.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(r.inputs.length!==1)throw new M("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new M("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");var a=Jv({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(i)this.outputs=r.outputs,this.inputs=r.inputs;else{if(e.inboundNodes.length!==1)throw new M("A layer added to a Sequential model must not already be "+("connected somewhere else. LayersModel received layer "+e.name+" ")+("which has "+e.inboundNodes.length+" pre-existing inbound ")+"connections.");if(e.inboundNodes[0].outputTensors.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Xv(this.outputs[0])}this.inboundNodes=[],new lo({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:Ji(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(function(o){return o.shape}),outputShapes:this.outputs[0].shape})}else{var s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1},t.prototype.pop=function(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{var e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}},t.prototype.call=function(e,i){return this.model==null&&this.build(),this.model.call(e,i)},t.prototype.build=function(e){if(Ke(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new yi({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0},t.prototype.countParams=function(){return this.built||this.build(),n.prototype.countParams.call(this)},t.prototype.summary=function(e,i,r){r===void 0&&(r=console.log),this.built||this.build(),n.prototype.summary.call(this,e,i,r)},t.prototype.setWeights=function(e){this.model==null&&this.build(),this.model.setWeights(e)},t.prototype.evaluate=function(e,i,r){if(r===void 0&&(r={}),!this.built)throw new Fn("The model needs to be compiled before being used.");return this.model.evaluate(e,i,r)},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.evaluateDataset(e,i)]})})},t.prototype.predict=function(e,i){return i===void 0&&(i={}),this.model==null&&this.build(),this.model.predict(e,i)},t.prototype.predictOnBatch=function(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)},t.prototype.compile=function(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames},Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.model==null?void 0:this.model.optimizer},set:function(e){this.model.optimizer=e},enumerable:!0,configurable:!0}),t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fit(e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fitDataset(e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,this.model.trainOnBatch(e,i)]})})},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s,o={};if(i instanceof Array){if(!(i[0].className!=null)||i[0].className==="Merge")throw new M("Legacy serialization format not supported yet.");s=i}else y.util.assert(i.layers!=null,function(){return"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."}),s=i.layers,delete i.layers,o=i;var l=new e(o);if(!(l instanceof t))throw new Te("Sequential.fromConfig called on non-Sequential input: "+l);for(var u=0,c=s;u<c.length;u++){var h=c[u],d=void 0,p=yn(h,d,a);a&&p.setFastWeightInitDuringBuild(!0),l.add(p)}return l},Object.defineProperty(t.prototype,"stopTraining",{get:function(){if(this.model==null)throw new M("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining},set:function(e){if(this.model==null)throw new M("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e},enumerable:!0,configurable:!0}),t.prototype.getConfig=function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i],s={};s.className=a.getClassName(),s.config=a.getConfig(),e.push(s)}return{name:this.name,layers:e}},t.className="Sequential",t}(yi);y.serialization.registerClass(zh);function B3(n){return new yi(n)}function z3(n){return new zh(n)}function P3(n,t){return t==null&&(t={}),W3(n,t)}function Ty(n){return Jv(n)}function _3(n,t){ry.registerCallbackConstructor(n,t)}var Kt=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),M3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),uk(e,i)},t.className="elu",t}(Kt);y.serialization.registerClass(M3);var H3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.selu(e)},t.className="selu",t}(Kt);y.serialization.registerClass(H3);var V3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="relu",t}(Kt);y.serialization.registerClass(V3);var q3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tidy(function(){return y.minimum(6,y.relu(e))})},t.className="relu6",t}(Kt);y.serialization.registerClass(q3);var G3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return e},t.className="linear",t}(Kt);y.serialization.registerClass(G3);var Y3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.sigmoid(e)},t.className="sigmoid",t}(Kt);y.serialization.registerClass(Y3);var K3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return hk(e)},t.className="hardSigmoid",t}(Kt);y.serialization.registerClass(K3);var j3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.softplus(e)},t.className="softplus",t}(Kt);y.serialization.registerClass(j3);var $3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return ck(e)},t.className="softsign",t}(Kt);y.serialization.registerClass($3);var X3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tanh(e)},t.className="tanh",t}(Kt);y.serialization.registerClass(X3);var Ny=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.softmax(e,i)},t.className="softmax",t}(Kt);y.serialization.registerClass(Ny);var J3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.logSoftmax(e,i)},t.className="logSoftmax",t}(Kt);y.serialization.registerClass(J3);var Z3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),y.tidy(function(){return y.sigmoid(e.mul(i)).mul(e)})},t.className="swish",t}(Kt);y.serialization.registerClass(Z3);function bi(n){return n.getClassName()}function Ph(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"activation")}function wi(n){if(n==null){var t={};return t.className="linear",t.config={},Ph(t)}if(typeof n=="string"){var t={};return t.className=n,t.config={},Ph(t)}else return n instanceof Kt?n:Ph(n)}function _h(n){if(n!=null&&typeof n!="object")throw new Error("Argument to L1L2 regularizer's constructor is expected to be an "+("object, but received: "+n))}var xy=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(y.serialization.Serializable),yo=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return _h(e),i.l1=e==null||e.l1==null?.01:e.l1,i.l2=e==null||e.l2==null?.01:e.l2,i.hasL1=i.l1!==0,i.hasL2=i.l2!==0,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=y.zeros([1]);return i.hasL1&&(r=y.add(r,y.sum(y.mul(i.l1,y.abs(e))))),i.hasL2&&(r=y.add(r,y.sum(y.mul(i.l2,Ba(e))))),r.asScalar()})},t.prototype.getConfig=function(){return{l1:this.l1,l2:this.l2}},t.fromConfig=function(e,i){return new e({l1:i.l1,l2:i.l2})},t.className="L1L2",t}(xy);y.serialization.registerClass(yo);function Q3(n){return _h(n),new yo({l1:n!=null?n.l1:null,l2:0})}function e4(n){return _h(n),new yo({l2:n!=null?n.l2:null,l1:0})}var Cy={l1l2:"L1L2"};function je(n){return oh(n)}function Ry(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"regularizer")}function nt(n){if(n==null)return null;if(typeof n=="string"){var t=n in Cy?Cy[n]:n,e={className:t,config:{}};return Ry(e)}else return n instanceof xy?n:Ry(n)}var Oy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null&&(i.maxValue=e.maxValue),i}return t.prototype.call=function(e,i){e=xe(e);var r=y.relu(e);return this.maxValue!=null&&(r=y.clipByValue(r,0,this.maxValue)),r},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={maxValue:this.maxValue},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ReLU",t}(De);y.serialization.registerClass(Oy);var Ey=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_ALPHA=.3,e==null&&(e={}),i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.leakyRelu(r,this.alpha)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LeakyReLU",t}(De);y.serialization.registerClass(Ey);var Dy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),i.supportsMasking=!0,i.alphaInitializer=tt(e.alphaInitializer||i.DEFAULT_ALPHA_INITIALIZER),i.alphaRegularizer=nt(e.alphaRegularizer),i.alphaConstraint=bt(e.alphaConstraint),e.sharedAxes==null)i.sharedAxes=null;else if(Array.isArray(e.sharedAxes))i.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")i.sharedAxes=[e.sharedAxes];else throw new M("Expected sharedAxes to be a number or an array of numbers, "+("but got "+e.sharedAxes));return i}return t.prototype.build=function(e){e=Ke(e);var i=e.slice(1);if(this.sharedAxes!=null)for(var r=0,a=this.sharedAxes;r<a.length;r++){var s=a[r];i[s-1]=1}this.alpha=this.addWeight("alpha",i,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);var o={};if(this.sharedAxes!=null)for(var s=1;s<e.length;++s)o[s]=e[s];this.inputSpec=[new Tt({ndim:e.length,axes:o})],this.built=!0},t.prototype.call=function(e,i){return e=xe(e),y.prelu(e,this.alpha.read())},t.prototype.getConfig=function(){var e={alphaInitializer:st(this.alphaInitializer),alphaRegularizer:je(this.alphaRegularizer),alphaConstraint:yt(this.alphaConstraint),sharedAxes:this.sharedAxes},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="PReLU",t}(De);y.serialization.registerClass(Dy);var ky=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==i.DEFAULT_ALPHA)throw new Te("Non-default alpha value ("+e.alpha+") is not supported by the ELU layer yet.");return i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.elu(r)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ELU",t}(De);y.serialization.registerClass(ky);var Fy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_THETA=1,e==null&&(e={}),i.theta=e.theta==null?i.DEFAULT_THETA:e.theta,i}return t.prototype.call=function(e,i){var r=xe(e);return r.mul(Wa(r.greater(this.theta),"float32"))},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={theta:this.theta},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ThresholdedReLU",t}(De);y.serialization.registerClass(Fy);var Wy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_AXIS=1,e==null&&(e={}),i.softmax=new Ny().apply,i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i}return t.prototype.call=function(e,i){var r=xe(e);return this.softmax(r,this.axis)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Softmax",t}(De);y.serialization.registerClass(Wy);function Yr(n,t,e){if(typeof n=="number")return Ji(n,t);if(n.length!==t)throw new M("The "+e+" argument must be an integer or tuple of "+t+" integers."+(" Received: "+n.length+" elements."));for(var i=0;i<t;++i){var r=n[i];if(!ak(r))throw new M("The "+e+" argument must be an integer or tuple of "+t+(" integers. Received: "+JSON.stringify(n)+" including a")+(" non-integer number "+r))}return n}function bn(n,t,e,i,r){if(r===void 0&&(r=1),n==null)return n;var a=t+(t-1)*(r-1),s;return e==="same"?s=n:s=n-a+1,Math.floor((s+i-1)/i)}function bo(n,t,e,i){if(n==null)return null;if(i==="valid")n=n*t+gi([e-t,0]);else if(i==="same")n=n*t;else throw new M("Unsupport padding mode: "+i+".");return n}function Mh(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,1]):n})}function Uy(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,4,1]):n})}function t4(n,t,e,i,r,a,s){return i===void 0&&(i=1),r===void 0&&(r="valid"),s===void 0&&(s=1),y.tidy(function(){if(a==null&&(a=mn()),ct(a),n.shape.length!==3)throw new M("The input of a conv1dWithBias operation should be 3, but is "+(n.shape.length+" instead."));if(t.shape.length!==3)throw new M("The kernel for a conv1dWithBias operation should be 3, but is "+(t.shape.length+" instead"));if(e!=null&&e.shape.length!==1)throw new M("The bias for a conv1dWithBias operation should be 1, but is "+(t.shape.length+" instead"));if(a==="channelsFirst"&&(n=y.transpose(n,[0,2,1])),r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");var o=y.conv1d(n,t,i,r==="same"?"same":"valid","NWC",s);return e!=null&&(o=Bn(o,e)),o})}function By(n,t,e,i,r,a,s,o){return i===void 0&&(i=[1,1]),r===void 0&&(r="valid"),o===void 0&&(o=null),y.tidy(function(){if(a==null&&(a=mn()),ct(a),n.rank!==3&&n.rank!==4)throw new M("conv2dWithBiasActivation expects input to be of rank 3 or 4, "+("but received "+n.rank+"."));if(t.rank!==3&&t.rank!==4)throw new M("conv2dWithBiasActivation expects kernel to be of rank 3 or 4, "+("but received "+n.rank+"."));var l=Mh(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=y.fused.conv2d({x:l,filter:t,strides:i,pad:r==="same"?"same":"valid",dilations:s,dataFormat:"NHWC",bias:e,activation:o}),a==="channelsFirst"&&(l=y.transpose(l,[0,3,1,2])),l})}function n4(n,t,e,i,r,a,s){return i===void 0&&(i=[1,1,1]),r===void 0&&(r="valid"),y.tidy(function(){if(a==null&&(a=mn()),ct(a),n.rank!==4&&n.rank!==5)throw new M("conv3dWithBias expects input to be of rank 4 or 5, but received "+(n.rank+"."));if(t.rank!==4&&t.rank!==5)throw new M("conv3dWithBias expects kernel to be of rank 4 or 5, but received "+(n.rank+"."));var o=Uy(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=y.conv3d(o,t,i,r==="same"?"same":"valid","NDHWC",s),e!=null&&(o=Bn(o,e)),a==="channelsFirst"&&(o=y.transpose(o,[0,4,1,2,3])),o})}var zy=function(n){Q(t,n);function t(e,i){var r=n.call(this,i)||this;if(r.bias=null,r.DEFAULT_KERNEL_INITIALIZER="glorotNormal",r.DEFAULT_BIAS_INITIALIZER="zeros",t.verifyArgs(i),r.rank=e,At(r.rank,"rank"),r.rank!==1&&r.rank!==2&&r.rank!==3)throw new Te("Convolution layer for rank other than 1, 2, or 3 ("+r.rank+") is not implemented yet.");if(r.kernelSize=Yr(i.kernelSize,e,"kernelSize"),r.strides=Yr(i.strides==null?1:i.strides,e,"strides"),r.padding=i.padding==null?"valid":i.padding,tn(r.padding),r.dataFormat=i.dataFormat==null?"channelsLast":i.dataFormat,ct(r.dataFormat),r.activation=wi(i.activation),r.useBias=i.useBias==null?!0:i.useBias,r.biasInitializer=tt(i.biasInitializer||r.DEFAULT_BIAS_INITIALIZER),r.biasConstraint=bt(i.biasConstraint),r.biasRegularizer=nt(i.biasRegularizer),r.activityRegularizer=nt(i.activityRegularizer),r.dilationRate=Yr(i.dilationRate==null?1:i.dilationRate,e,"dilationRate"),r.rank===1&&Array.isArray(r.dilationRate)&&r.dilationRate.length!==1)throw new M("dilationRate must be a number or an array of a single number for 1D convolution, but received "+(""+JSON.stringify(r.dilationRate)));if(r.rank===2){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==2)throw new M("dilationRate must be a number or array of two numbers for 2D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}else if(r.rank===3){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==3)throw new M("dilationRate must be a number or array of three numbers for 3D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}return r}return t.verifyArgs=function(e){if(Wn("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!uh(e.kernelSize,"number",1,3))throw new M("BaseConv expects config.kernelSize to be number or number[] with "+("length 1, 2, or 3, but received "+JSON.stringify(e.kernelSize)+"."))},t.prototype.getConfig=function(){var e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:bi(this.activation),useBias:this.useBias,biasInitializer:st(this.biasInitializer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),wo=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;return r.kernel=null,t.verifyArgs(i),r.filters=i.filters,At(r.filters,"filters"),r.kernelInitializer=tt(i.kernelInitializer||r.DEFAULT_KERNEL_INITIALIZER),r.kernelConstraint=bt(i.kernelConstraint),r.kernelRegularizer=nt(i.kernelRegularizer),r}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=this.kernelSize.concat([a,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:(i={},i[r]=a,i)}],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a,s=r.bias==null?null:r.bias.read(),o=Lv(r.activation.getClassName());if(o!=null&&r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate,o);else{if(r.rank===1)a=t4(e,r.kernel.read(),s,r.strides[0],r.padding,r.dataFormat,r.dilationRate[0]);else if(r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else if(r.rank===3)a=n4(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else throw new Te("convolutions greater than 3D are not implemented yet.");r.activation!=null&&(a=r.activation.apply(a))}return a})},t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=[],r=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2),a=0;a<r.length;++a){var s=bn(r[a],this.kernelSize[a],this.padding,this.strides[a],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[a]);i.push(s)}var o=[e[0]];return this.dataFormat==="channelsLast"?(o=o.concat(i),o.push(this.filters)):(o.push(this.filters),o=o.concat(i)),o},t.prototype.getConfig=function(){var e={filters:this.filters,kernelInitializer:st(this.kernelInitializer),kernelRegularizer:je(this.kernelRegularizer),kernelConstraint:yt(this.kernelConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.verifyArgs=function(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new M("Convolution layer expected config.filters to be a 'number' > 0 "+("but got "+JSON.stringify(e.filters)))},t}(zy),Hh=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!uh(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(wo);y.serialization.registerClass(Hh);var Py=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(wo);y.serialization.registerClass(Py);var _y=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Tt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Tt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=bo(c,f,d,r.padding),v=bo(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Bn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=bo(i[a],u,o,this.padding),i[s]=bo(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(Hh);y.serialization.registerClass(_y);var i4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=nt(i.depthwiseRegularizer),r.depthwiseConstraint=bt(i.depthwiseConstraint),r.pointwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=nt(i.pointwiseRegularizer),r.pointwiseConstraint=bt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length<this.rank+2)throw new M("Inputs to SeparableConv"+this.rank+"D should have rank "+(this.rank+2+", but received input shape: ")+(""+JSON.stringify(e)));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null||e[r]<0)throw new M("The channel dimension of the inputs should be defined, "+("but found "+JSON.stringify(e[r])));for(var a=e[r],s=this.kernelSize.concat([a,this.depthMultiplier]),o=[],l=0;l<this.rank;++l)o.push(1);o.push(a*this.depthMultiplier,this.filters);var u=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,u,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",o,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,u,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,u,this.biasConstraint):this.bias=null,this.inputSpec=[new Tt({ndim:this.rank+2,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a;if(r.rank===1)throw new Te("1D separable convolution is not implemented yet.");return r.rank===2&&(r.dataFormat==="channelsFirst"&&(e=y.transpose(e,[0,2,3,1])),a=y.separableConv2d(e,r.depthwiseKernel.read(),r.pointwiseKernel.read(),r.strides,r.padding,r.dilationRate,"NHWC")),r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),r.dataFormat==="channelsFirst"&&(a=y.transpose(a,[0,3,1,2])),a})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=st(this.depthwiseInitializer),e.pointwiseInitializer=st(this.pointwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.pointwiseRegularizer=je(this.pointwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseConstraint),e.pointwiseConstraint=yt(this.pointwiseConstraint),e},t.className="SeparableConv",t}(wo),My=function(n){Q(t,n);function t(e){return n.call(this,2,e)||this}return t.className="SeparableConv2D",t}(i4);y.serialization.registerClass(My);var Hy=function(n){Q(t,n);function t(e){var i=n.call(this,1,e)||this;return t.verifyArgs(e),i.inputSpec=[{ndim:3}],i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.dataFormat,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!uh(e.kernelSize,"number",1,1))throw new M("Conv1D expects config.kernelSize to be number or number[] with "+("length 1, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv1D",t}(wo);y.serialization.registerClass(Hy);var Vy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return typeof e.cropping=="number"?i.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?i.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:i.cropping=e.cropping,i.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,i.inputSpec=[{ndim:4}],i}return t.prototype.computeOutputShape=function(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=xe(e),r.dataFormat==="channelsLast"){var a=no(e,r.cropping[0][0],e.shape[1]-r.cropping[0][0]-r.cropping[0][1],2);return no(a,r.cropping[1][0],e.shape[2]-r.cropping[1][1]-r.cropping[1][0],3)}else{var a=no(e,r.cropping[0][0],e.shape[2]-r.cropping[0][0]-r.cropping[0][1],3);return no(a,r.cropping[1][0],e.shape[3]-r.cropping[1][1]-r.cropping[1][0],4)}})},t.prototype.getConfig=function(){var e={cropping:this.cropping,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Cropping2D",t}(De);y.serialization.registerClass(Vy);var qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_SIZE=[2,2],i.inputSpec=[{ndim:4}],i.size=e.size==null?i.DEFAULT_SIZE:e.size,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,i}return t.prototype.computeOutputShape=function(e){if(this.dataFormat==="channelsFirst"){var i=e[2]==null?null:this.size[0]*e[2],r=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],i,r]}else{var i=e[1]==null?null:this.size[0]*e[1],r=e[2]==null?null:this.size[1]*e[2];return[e[0],i,r,e[3]]}},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e),s=a.shape;if(r.dataFormat==="channelsFirst"){a=y.transpose(a,[0,2,3,1]);var o=r.size[0]*s[2],l=r.size[1]*s[3],u=a.resizeNearestNeighbor([o,l]);return y.transpose(u,[0,3,1,2])}else{var o=r.size[0]*s[1],l=r.size[1]*s[2];return a.resizeNearestNeighbor([o,l])}})},t.prototype.getConfig=function(){var e={size:this.size,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="UpSampling2D",t}(De);y.serialization.registerClass(qy);function r4(n,t,e,i,r,a){return e===void 0&&(e=[1,1]),i===void 0&&(i="valid"),y.tidy(function(){r==null&&(r=mn()),ct(r);var s=Mh(n,r);if(n.rank!==4)throw new M("Input for depthwiseConv2d is required to be 4-D, but is instead "+(n.rank+"-D"));if(t.rank!==4)throw new M("depthwiseKernel is required to be 4-D, but is instead "+(t.rank+"-D"));return s=y.depthwiseConv2d(s,t,e,i==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}var Gy=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return i.depthwiseKernel=null,i.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,i.depthwiseInitializer=tt(e.depthwiseInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.depthwiseConstraint=bt(e.depthwiseConstraint),i.depthwiseRegularizer=nt(e.depthwiseRegularizer),i}return t.prototype.build=function(e){if(e=Ke(e),e.length<4)throw new M("Inputs to DepthwiseConv2D should have rank 4. "+("Received input shape: "+JSON.stringify(e)+"."));var i=this.dataFormat==="channelsFirst"?1:3;if(e[i]==null||e[i]<0)throw new M("The channel dimension of the inputs to DepthwiseConv2D should "+("be defined, but is not ("+e[i]+")."));var r=e[i],a=[this.kernelSize[0],this.kernelSize[1],r,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[r*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=r4(e,r.depthwiseKernel.read(),r.strides,r.padding,r.dataFormat,null);return r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),a})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,s=bn(i,this.kernelSize[0],this.padding,this.strides[0]),o=bn(r,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,s,o]:[e[0],s,o,a]},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=st(this.depthwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseRegularizer),e},t.className="DepthwiseConv2D",t}(zy);y.serialization.registerClass(Gy);function Yy(n,t,e,i){if(Array.isArray(n)){if(t!=null||e!=null)throw new M("When inputs is an array, neither initialState or constants should be provided");i!=null&&(e=n.slice(n.length-i,n.length),n=n.slice(0,n.length-i)),n.length>1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Ky(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(gn(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],x=y.tidy(function(){return n(L,d)});if(r==null)h=x[0],d=x[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=x[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return x[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v<p;++v)g(v);var b;if(o){var w=1;b=y.stack(c,w)}return[h,b,d]})}var Si=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r;if(e.cell==null)throw new M("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?r=new Vh({cells:e.cell}):r=e.cell,r.stateSize==null)throw new M("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");return i.cell=r,i.returnSequences=e.returnSequences==null?!1:e.returnSequences,i.returnState=e.returnState==null?!1:e.returnState,i.goBackwards=e.goBackwards==null?!1:e.goBackwards,i._stateful=e.stateful==null?!1:e.stateful,i.unroll=e.unroll==null?!1:e.unroll,i.supportsMasking=!0,i.inputSpec=[new Tt({ndim:3})],i.stateSpec=null,i.states_=null,i.numConstants=null,i.keptStates=[],i}return t.prototype.getStates=function(){if(this.states_==null){var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return gn(0,e).map(function(i){return null})}else return this.states_},t.prototype.setStates=function(e){this.states_=e},t.prototype.computeOutputShape=function(e){Lh(e)&&(e=e[0]),e=e;var i=this.cell.stateSize;Array.isArray(i)||(i=[i]);var r=i[0],a;if(this.returnSequences?a=[e[0],e[1],r]:a=[e[0],r],this.returnState){for(var s=[],o=0,l=i;o<l.length;o++){var u=l[o];s.push([e[0],u])}return[a].concat(s)}else return a},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){Array.isArray(i)&&(i=i[0]);var a=r.returnSequences?i:null;if(r.returnState){var s=r.states.map(function(o){return null});return[a].concat(s)}else return a})},Object.defineProperty(t.prototype,"states",{get:function(){if(this.states_==null){for(var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,i=[],r=0;r<e;++r)i.push(null);return i}else return this.states_},set:function(e){this.states_=e},enumerable:!0,configurable:!0}),t.prototype.build=function(e){if(this.numConstants!=null)throw new Te("Constants support is not implemented in RNN yet.");Lh(e)&&(e=e[0]),e=e;var i=this.stateful?e[0]:null,r=e.slice(2);this.inputSpec[0]=new Tt({shape:[i,null].concat(r)});var a=[e[0]].concat(e.slice(2));this.cell.build(a);var s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!y.util.arraysEqual(this.stateSpec.map(function(o){return o.shape[o.shape.length-1]}),s))throw new M("An initialState was passed that is not compatible with "+("cell.stateSize. Received stateSpec="+this.stateSpec+"; ")+("However cell.stateSize is "+this.cell.stateSize))}else this.stateSpec=s.map(function(o){return new Tt({shape:[null,o]})});this.stateful&&this.resetStates()},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape[0];if(a==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.states_==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_=[y.zeros([a,r.cell.stateSize])];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_[0]=y.zeros([a,r.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i===!0?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var s=0;s<r.states_.length;++s){var o=e[s],l=Array.isArray(r.cell.stateSize)?r.cell.stateSize[s]:r.cell.stateSize,u=[a,l];if(!y.util.arraysEqual(o.shape,u))throw new M("State "+s+" is incompatible with layer "+r.name+": "+("expected shape="+u+", received shape="+o.shape));r.states_[s]=o}}r.states_=r.states_.map(function(c){return y.keep(c.clone())})})},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);e=s.inputs,r=s.initialState,a=s.constants;var o=[],l=[];if(r!=null){i.initialState=r,o=o.concat(r),this.stateSpec=[];for(var u=0,c=r;u<c.length;u++){var h=c[u];this.stateSpec.push(new Tt({shape:h.shape}))}l=l.concat(this.stateSpec)}a!=null&&(i.constants=a,o=o.concat(a),this.numConstants=a.length);var d=o[0]instanceof vn;if(d){var p=[e].concat(o),f=this.inputSpec.concat(l),m=this.inputSpec;this.inputSpec=f;var g=n.prototype.apply.call(this,p,i);return this.inputSpec=m,g}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;e=xe(e),o==null&&(r.stateful?o=r.states_:o=r.getInitialState(e));var l=Array.isArray(r.cell.stateSize)?r.cell.stateSize.length:1;if(o.length!==l)throw new M("RNN Layer has "+l+" state(s) but was passed "+(o.length+" initial state(s)."));r.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");var u={training:s},c=function(g,v){var b=r.cell.call([g].concat(v),u);return[b[0],b.slice(1)]},h=Ky(c,e,o,r.goBackwards,a,null,r.unroll,r.returnSequences),d=h[0],p=h[1],f=h[2];r.stateful&&r.resetStates(f,s);var m=r.returnSequences?p:d;return r.returnState?[m].concat(f):m})},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=y.zeros(e.shape);return r=y.sum(r,[1,2]),r=Ua(r),Array.isArray(i.cell.stateSize)?i.cell.stateSize.map(function(a){return a>1?hh(r,[1,a]):r}):i.cell.stateSize>1?[hh(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),Gt({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=yn(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Si);var Kr=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),qh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:s})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(a)},rate:r.recurrentDropout,training:s}));var o,l=r.dropoutMask,u=r.recurrentDropoutMask;l!=null?o=Un(y.mul(e,l),r.kernel.read()):o=Un(e,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),u!=null&&(a=y.mul(a,u));var c=y.add(o,Un(a,r.recurrentKernel.read()));return r.activation!=null&&(c=r.activation.apply(c)),[c,c]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Gt({},e,i)},t.className="SimpleRNNCell",t}(Kr);y.serialization.registerClass(qh);var jy=function(n){Q(t,n);function t(e){var i=this;return e.cell=new qh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return new e(i)},t.className="SimpleRNN",t}(Si);y.serialization.registerClass(jy);var Gh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new M("GRUCell does not support reset_after parameter set to true.");return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e);var i=e[e.length-1];this.kernel=this.addWeight("kernel",[i,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("GRUCell expects 2 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training==null?!1:i.training,s=e[1];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:3})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:3}));var o=r.dropoutMask,l=r.recurrentDropoutMask,u,c,h;0<r.dropout&&r.dropout<1&&(e=y.mul(e,o[0]));var d=Un(e,r.kernel.read());r.useBias&&(d=Bn(d,r.bias.read())),0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,l[0]));var p=r.recurrentKernel.read(),f=y.split(p,[2*r.units,r.units],p.rank-1),m=f[0],g=f[1],v=Un(s,m),b=y.split(d,3,d.rank-1),w=b[0],S=b[1],L=b[2],x=y.split(v,2,v.rank-1),C=x[0],R=x[1];u=r.recurrentActivation.apply(y.add(w,C)),c=r.recurrentActivation.apply(y.add(S,R));var D=Un(y.mul(c,s),g);h=r.activation.apply(y.add(L,D));var k=y.add(y.mul(u,s),y.mul(y.add(1,y.neg(u)),h));return[k,k]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Gt({},e,i)},t.className="GRUCell",t}(Kr);y.serialization.registerClass(Gh);var $y=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Gh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="GRU",t}(Si);y.serialization.registerClass($y);var So=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.unitForgetBias=e.unitForgetBias,i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=[i.units,i.units],i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel=this.addWeight("kernel",[r,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);var a;if(this.useBias){if(this.unitForgetBias){var s=this.biasInitializer,o=this.units;a=new(i=function(l){Q(u,l);function u(){return l!==null&&l.apply(this,arguments)||this}return u.prototype.apply=function(c,h){var d=s.apply([o]),p=new mh().apply([o]),f=s.apply([o*2]);return Wv(Wv(d,p),f)},u}(un),i.className="CustomInit",i)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training;if(e=e,e.length!==3)throw new M("LSTMCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var s=e[1],o=e[2];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:4})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:4}));var l=r.dropoutMask,u=r.recurrentDropoutMask,c,h,d,p;0<r.dropout&&r.dropout<1&&(e=y.mul(e,l[0]));var f=Un(e,r.kernel.read());0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,u[0])),f=y.add(f,Un(s,r.recurrentKernel.read())),r.useBias&&(f=Bn(f,r.bias.read()));var m=y.split(f,4,f.rank-1),g=m[0],v=m[1],b=m[2],w=m[3];c=r.recurrentActivation.apply(g),h=r.recurrentActivation.apply(v),d=y.add(y.mul(h,o),y.mul(c,r.activation.apply(b))),p=r.recurrentActivation.apply(w);var S=y.mul(p,r.activation.apply(d));return[S,S,d]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Gt({},e,i)},t.className="LSTMCell",t}(Kr);y.serialization.registerClass(So);var Xy=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new So(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="LSTM",t}(Si);y.serialization.registerClass(Xy);var Vh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.cells=e.cells,i}return Object.defineProperty(t.prototype,"stateSize",{get:function(){for(var e=[],i=0,r=this.cells.slice().reverse();i<r.length;i++){var a=r[i];Array.isArray(a.stateSize)?e.push.apply(e,a.stateSize):e.push(a.stateSize)}return e},enumerable:!0,configurable:!0}),t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=e;for(var a=e.slice(1),s=[],o=0,l=r.cells.slice().reverse();o<l.length;o++){var u=l[o];Array.isArray(u.stateSize)?s.push(a.splice(0,u.stateSize.length)):s.push(a.splice(0,1))}s.reverse();for(var c=[],h,d=0;d<r.cells.length;++d){var u=r.cells[d];a=s[d],d===0?h=[e[0]].concat(a):h=[h[0]].concat(a),h=u.call(h,i),c.push(h.slice(1))}a=[];for(var p=0,f=c.slice().reverse();p<f.length;p++){var m=f[p];a.push.apply(a,m)}return[h[0]].concat(a)})},t.prototype.build=function(e){Lh(e)&&(e=e[0]),e=e;var i;this.cells.forEach(function(r,a){Qi("RNNCell_"+a,function(){r.build(e),Array.isArray(r.stateSize)?i=r.stateSize[0]:i=r.stateSize,e=[e[0],i]})}),this.built=!0},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=function(s){return{className:s.getClassName(),config:s.getConfig()}},r=this.cells.map(i),a={cells:r};return Gt({},e,a)},t.fromConfig=function(e,i,r){r===void 0&&(r={});for(var a=[],s=0,o=i.cells;s<o.length;s++){var l=o[s];a.push(yn(l,r))}return new e({cells:a})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(!this.trainable)return[];for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.cells;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.weights)}return Ih(e)},t.prototype.setWeights=function(e){for(var i=[],r=0,a=this.cells;r<a.length;r++)for(var s=a[r],o=s.weights.length,l=e.splice(o),u=0;u<s.weights.length;++u)i.push([s.weights[u],l[u]]);Ah(i)},t.className="StackedRNNCells",t}(Kr);y.serialization.registerClass(Vh);function Li(n){var t=n.ones,e=n.rate,i=n.training,r=i===void 0?!1:i,a=n.count,s=a===void 0?1:a,o=function(){return Bv(t(),e)},l=function(){return za(o,t,r)};if(!s||s<=1)return y.keep(l().clone());var u=Array(s).fill(void 0).map(l);return u.map(function(c){return y.keep(c.clone())})}var LH=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(Kr),a4=function(n){Q(t,n);function t(e){var i=this;if(e.unroll)throw new Te("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Te("It is not possible at the moment to stack convolutional cells.");return i=n.call(this,e)||this,i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null),i&&i.constants)throw new M("ConvRNN2D cell does not support constants");var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.prototype.computeOutputShape=function(e){var i=this.computeSingleOutputShape(e);return this.returnSequences||(i=[i[0]].concat(i.slice(2))),this.returnState&&(i=[i].concat(Array(2).fill([e[0]].concat(i.slice(-3))))),i},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=i.cell.stateSize,a=e.shape,s=i.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=y.zeros(o);return Array.isArray(r)?Array(r.length).fill(l):[l]})},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape,s=r.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=a[0];if(l==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.getStates()==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_=[y.zeros(o)];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_[0]=y.zeros(o);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var u=0;u<r.states_.length;++u){var c=e[u],h=o;if(!y.util.arraysEqual(c.shape,h))throw new M("State "+u+" is incompatible with layer "+r.name+": "+("expected shape="+h+", received shape="+c.shape));r.states_[u]=c}}r.states_=r.states_.map(function(d){return y.keep(d.clone())})})},t.prototype.computeSingleOutputShape=function(e){var i=this.cell,r=i.dataFormat,a=i.filters,s=i.kernelSize,o=i.padding,l=i.strides,u=i.dilationRate,c=r==="channelsFirst",h=e[c?3:2],d=e[c?4:3],p=bn(h,s[0],o,l[0],u[0]),f=bn(d,s[1],o,l[1],u[1]),m=e.slice(0,2).concat(c?[a,p,f]:[p,f,a]);return m},t.className="ConvRNN2D",t}(Si),Yh=function(n){Q(t,n);function t(e){var i=this,r=e.filters,a=e.kernelSize,s=e.strides,o=e.padding,l=e.dataFormat,u=e.dilationRate;return i=n.call(this,Gt({},e,{units:r}))||this,i.filters=r,At(i.filters,"filters"),i.kernelSize=Yr(a,2,"kernelSize"),i.kernelSize.forEach(function(c){return At(c,"kernelSize")}),i.strides=Yr(s||1,2,"strides"),i.strides.forEach(function(c){return At(c,"strides")}),i.padding=o||"valid",tn(i.padding),i.dataFormat=l||"channelsLast",ct(i.dataFormat),i.dilationRate=Yr(u||1,2,"dilationRate"),i.dilationRate.forEach(function(c){return At(c,"dilationRate")}),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=4,o=this.kernelSize.concat([a,this.filters*s]);this.kernel=this.addWeight("kernel",o,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);var l=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",l,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){var u=void 0;if(this.unitForgetBias){var c=this.biasInitializer,h=this.filters;u=new(i=function(d){Q(p,d);function p(){return d!==null&&d.apply(this,arguments)||this}return p.prototype.apply=function(f,m){var g=c.apply([h]),v=y.ones([h]),b=c.apply([h*2]);return ph([g,v,b])},p}(un),i.className="CustomInit",i)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e.length!==3)throw new M("ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training||!1,s=e[0],o=e[1],l=e[2],u=4;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.dropout,training:a,count:u}));var c=r.dropoutMask,h=function(te,re,le){return!re||!re[le]?te:y.mul(re[le],te)},d=h(s,c,0),p=h(s,c,1),f=h(s,c,2),m=h(s,c,3);0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(o)},rate:r.recurrentDropout,training:a,count:u}));var g=r.recurrentDropoutMask,v=h(o,g,0),b=h(o,g,1),w=h(o,g,2),S=h(o,g,3),L=3,x=y.split(r.kernel.read(),u,L),C=x[0],R=x[1],D=x[2],k=x[3],W=r.useBias?y.split(r.bias.read(),u):[null,null,null,null],F=W[0],P=W[1],H=W[2],_=W[3];d=r.inputConv(d,C,F,r.padding),p=r.inputConv(p,R,P,r.padding),f=r.inputConv(f,D,H,r.padding),m=r.inputConv(m,k,_,r.padding);var K=y.split(r.recurrentKernel.read(),u,L),j=K[0],q=K[1],G=K[2],Z=K[3];v=r.recurrentConv(v,j),b=r.recurrentConv(b,q),w=r.recurrentConv(w,G),S=r.recurrentConv(S,Z);var X=r.recurrentActivation.apply(y.add(d,v)),ee=r.recurrentActivation.apply(y.add(p,b)),ne=y.add(y.mul(ee,l),y.mul(X,r.activation.apply(y.add(f,w)))),ie=y.mul(r.recurrentActivation.apply(y.add(m,S)),r.activation.apply(ne));return[ie,ie,ne]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=e.units,r=VD(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Gt({},r,a)},t.prototype.inputConv=function(e,i,r,a){var s=y.conv2d(e,i,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return r?Bn(s,r,this.dataFormat):s},t.prototype.recurrentConv=function(e,i){var r=1;return y.conv2d(e,i,r,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")},t.className="ConvLSTM2DCell",t}(So);y.serialization.registerClass(Yh);var Jy=function(n){Q(t,n);function t(e){var i=this,r=new Yh(e);return i=n.call(this,Gt({},e,{cell:r}))||this,i}return t.fromConfig=function(e,i){return new e(i)},t.className="ConvLSTM2D",t}(a4);y.serialization.registerClass(Jy);var Kh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.rate=Math.max(Math.min(e.rate,1),0),i.noiseShape=e.noiseShape,i.seed=e.seed,i.supportsMasking=!0,i}return t.prototype.getNoiseShape=function(e){if(this.noiseShape==null)return this.noiseShape;for(var i=e.shape,r=[],a=0;a<this.noiseShape.length;++a)r.push(this.noiseShape[a]==null?i[a]:this.noiseShape[a]);return r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(0<r.rate&&r.rate<1){var s=i.training==null?!1:i.training,o=r.getNoiseShape(a),l=za(function(){return Bv(a,r.rate,o,r.seed)},function(){return a},s);return l}return e})},t.prototype.getConfig=function(){var e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.dispose=function(){return n.prototype.dispose.call(this)},t.className="Dropout",t}(De);y.serialization.registerClass(Kh);var Zy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[{ndim:3}],i}return t.prototype.getNoiseShape=function(e){var i=e.shape;return[i[0],1,i[2]]},t.className="SpatialDropout1D",t}(Kh);y.serialization.registerClass(Zy);var Qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.activation=null,i.useBias=!0,i.kernel=null,i.bias=null,i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){var r=null;e.batchSize!=null&&(r=e.batchSize),i.batchInputShape=[r,e.inputDim]}return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation),e.useBias!=null&&(i.useBias=e.useBias),i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelConstraint=bt(e.kernelConstraint),i.biasConstraint=bt(e.biasConstraint),i.kernelRegularizer=nt(e.kernelRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.supportsMasking=!0,i.inputSpec=[{minNDim:2}],i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[r,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:(i={},i[-1]=r,i)}],this.built=!0},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return i[i.length-1]=this.units,i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=Lv(r.activation.getClassName()),o;return s!=null?o=Un(a,r.kernel.read(),s,r.bias?r.bias.read():null):(o=Un(a,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),r.activation!=null&&(o=r.activation.apply(o))),o})},t.prototype.getConfig=function(){var e={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dense",t}(De);y.serialization.registerClass(Qy);var eb=function(n){Q(t,n);function t(e){var i=this;return e=e||{},i=n.call(this,e)||this,i.inputSpec=[{minNDim:3}],i.dataFormat=e.dataFormat,i}return t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=0,r=e.slice(1);i<r.length;i++){var a=r[i];if(a==null)throw new M('The shape of the input to "Flatten" is not fully defined '+("(got "+e.slice(1)+"). Make sure to pass a complete ")+'"input_shape" or "batch_input_shape" argument to the first layer in your model.')}return[e[0],mi(e,1)]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.dataFormat==="channelsFirst"&&a.rank>1){for(var s=[0],o=2;o<a.rank;++o)s.push(o);s.push(1),a=a.transpose(s)}return lk(a)})},t.prototype.getConfig=function(){var e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);var i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Flatten",t}(De);y.serialization.registerClass(eb);var tb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.activation=wi(e.activation),i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);return r.activation.apply(a)})},t.prototype.getConfig=function(){var e={activation:bi(this.activation)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Activation",t}(De);y.serialization.registerClass(tb);var nb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.n=e.n,i.inputSpec=[{ndim:2}],i}return t.prototype.computeOutputShape=function(e){return[e[0],this.n,e[1]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return e=xe(e),sk(e,r.n)})},t.prototype.getConfig=function(){var e={n:this.n},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="RepeatVector",t}(De);y.serialization.registerClass(nb);var ib=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;i.targetShape=e.targetShape;for(var r=0;r<i.targetShape.length;++r)i.isUnknown(i.targetShape[r])&&(i.targetShape[r]=null);return i}return t.prototype.isUnknown=function(e){return e<0||e==null},t.prototype.fixUnknownDimension=function(e,i){for(var r="Total size of new array must be unchanged.",a=i.slice(),s=1,o=null,l=0;l<a.length;++l){var u=a[l];if(this.isUnknown(u))if(o===null)o=l;else throw new M("Can only specifiy one unknown dimension.");else s*=u}var c=mi(e);if(o!==null){if(s===0||c%s!==0)throw new M(r);a[o]=c/s}else if(c!==s)throw new M(r);return a},t.prototype.computeOutputShape=function(e){for(var i=!1,r=0;r<e.length;++r)if(this.isUnknown(e[r])){i=!0;break}return i?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=a.shape,o=s.slice(0,1).concat(r.fixUnknownDimension(s.slice(1),r.targetShape));return a.reshape(o)})},t.prototype.getConfig=function(){var e={targetShape:this.targetShape},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Reshape",t}(De);y.serialization.registerClass(ib);var rb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error("Permute constructor requires `dims` to be an Array, but received "+(e.dims+" instead."));var r=gn(1,e.dims.length+1);if(!y.util.arraysEqual(e.dims.slice().sort(),r))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");return i.dims=e.dims,i.dimsIncludingBatch=[0].concat(i.dims),i.inputSpec=[new Tt({ndim:i.dims.length+1})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return this.dims.forEach(function(r,a){i[a+1]=e[r]}),i},t.prototype.call=function(e,i){return y.transpose(xe(e),this.dimsIncludingBatch)},t.prototype.getConfig=function(){var e={dims:this.dims},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Permute",t}(De);y.serialization.registerClass(rb);var ab=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null?i.maskValue=e.maskValue==null?0:e.maskValue:i.maskValue=0,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={maskValue:this.maskValue};return Object.assign(i,e),i},t.prototype.computeMask=function(e,i){var r=xe(e),a=-1;return y.any(y.notEqual(r,this.maskValue),a)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=-1,o=!0,l=y.any(y.notEqual(a,r.maskValue),s,o),u=a.mul(l.asType(a.dtype));return u})},t.className="Masking",t}(De);y.serialization.registerClass(ab);var sb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.embeddings=null,i.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){var r=null;e.batchSize!=null&&(r=e.batchSize),e.inputLength==null?i.batchInputShape=[r,null]:i.batchInputShape=[r].concat(Je(e.inputLength))}return i.inputDim=e.inputDim,At(i.inputDim,"inputDim"),i.outputDim=e.outputDim,At(i.outputDim,"outputDim"),i.embeddingsInitializer=tt(e.embeddingsInitializer||i.DEFAULT_EMBEDDINGS_INITIALIZER),i.embeddingsRegularizer=nt(e.embeddingsRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.embeddingsConstraint=bt(e.embeddingsConstraint),i.maskZero=e.maskZero,i.supportsMasking=e.maskZero,i.inputLength=e.inputLength,i}return t.prototype.build=function(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0},t.prototype.warnOnIncompatibleInputShape=function(e){},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){return r.maskZero?(e=xe(e),y.notEqual(e,y.zerosLike(e))):null})},t.prototype.computeOutputShape=function(e){if(e=Ke(e),this.inputLength==null)return e.concat([this.outputDim]);var i=Je(this.inputLength);if(i.length!==e.length-1)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));for(var r=0,a=0;a<i.length;++a){var s=i[a],o=e[a+1];if(s!=null&&o!=null&&s!==o)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));s==null&&(i[r]=o),r++}return[e[0]].concat(i,[this.outputDim])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);a.dtype!=="int32"&&(a=Wa(a,"int32"));var s=Uv(r.embeddings.read(),a.as1D());return s.reshape(Ke(r.computeOutputShape(a.shape)))})},t.prototype.getConfig=function(){var e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:st(this.embeddingsInitializer),embeddingsRegularizer:je(this.embeddingsRegularizer),activityRegularizer:je(this.activityRegularizer),embeddingsConstraint:yt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Embedding",t}(De);y.serialization.registerClass(sb);var ir=function(n){Q(t,n);function t(e){var i=n.call(this,e||{})||this;return i.supportsMasking=!0,i}return t.prototype.mergeFunction=function(e){throw new Te},t.prototype.computeElementwiseOpOutputShape=function(e,i){if(e==null||i==null)return null;if(e.length<i.length)return this.computeElementwiseOpOutputShape(i,e);if(i.length===0)return e;for(var r=e.slice(0,e.length-i.length),a=0;a<i.length;++a){var s=e[e.length-i.length+a],o=i[a];if(s==null||o==null||s<0||o<0)r.push(null);else if(s===1)r.push(o);else if(o===1)r.push(s);else{if(s!==o)throw new M("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(i));r.push(s)}}return r},t.prototype.build=function(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[Ke(e)]),e=e,e.length<2)throw new M("A merge layer should be called on an Array of at least 2 inputs."+(" Got "+e.length+" input(s)."));for(var i=[],r=0,a=e;r<a.length;r++){var s=a[r];s!=null&&s[0]!==null&&i.push(s[0])}if(i=fi(i),i.length>1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l<e.length;++l){var s=e[l]==null?null:e[l].slice(1);o=this.computeElementwiseOpOutputShape(o,s)}var u=e.map(function(c){return c.length});e.indexOf(null)===-1&&fi(u).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,r.reshapeRequired){var a=[],s=e.map(function(D){return D.rank});if(s.indexOf(null)===-1){for(var o=gi(s),l=0,u=e;l<u.length;l++){for(var c=u[l],h=c.rank,d=0;d<o-h;++d)c=Ua(c,1);a.push(c)}return r.mergeFunction(a)}else{for(var p=!1,f=0,m=e;f<m.length;f++){var c=m[f],h=c.rank;if(h==null){var g=c.shape,v=g[0],b=g.slice(1).concat([v]),w=c.reshape([v].concat(mi(g.slice(1))));w=y.transpose(w,[1,0]),w=w.reshape(b),a.push(w),p=!0}else if(h>1){var S=gn(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),x=L.rank;if(p){if(x==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(x>1){var S=[x-1].concat(gn(0,x-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r<e.length;++r){var a=e[r]==null?null:e[r].slice(1);i=this.computeElementwiseOpOutputShape(i,a)}for(var s=[],o=0,l=e;o<l.length;o++){var a=l[o];a!=null&&a[0]!==null&&s.push(a[0])}return s=fi(s),s.length===1?i=s.concat(i):i=[null].concat(i),i},t.prototype.computeMask=function(e,i){return y.tidy(function(){if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an Array");if(!Array.isArray(e))throw new M("`inputs` should be an Array");if(i.length!==e.length)throw new M("The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths "+("("+e.length+" vs "+i.length+")"));if(i.every(function(s){return s==null}))return null;i=i.map(function(s){return s==null?s:y.expandDims(s,0)});for(var r=i[0],a=1;a<i.length-1;++a)r=y.logicalAnd(r,i[a]);return r})},t}(De),ob=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return i})},t.className="Add",t}(ir);y.serialization.registerClass(ob);var lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.mul(i,e[r]);return i})},t.className="Multiply",t}(ir);y.serialization.registerClass(lb);var ub=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return y.mul(1/e.length,i)})},t.className="Average",t}(ir);y.serialization.registerClass(ub);var cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.maximum(i,e[r]);return i})},t.className="Maximum",t}(ir);y.serialization.registerClass(cb);var hb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.minimum(i,e[r]);return i})},t.className="Minimum",t}(ir);y.serialization.registerClass(hb);var db=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_AXIS=-1,e==null&&(e={}),i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new M("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;for(var i=!0,r=0,a=e;r<a.length;r++){var s=a[r];if(s!=null){i=!1;break}}if(i)return;for(var o=[],l=0;l<e.length;++l){var u=e[l].slice();u.splice(this.axis,1);for(var c=!1,h=0,d=o;h<d.length;h++){var s=d[h];if(y.util.arraysEqual(s,u)){c=!0;break}}c||o.push(u)}if(o.length>1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return ph(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s<o.length;s++){var l=o[s];if(r[a]==null||l[a]==null){r[a]=null;break}r[a]+=l[a]}return r},t.prototype.computeMask=function(e,i){var r=this;if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new M("`inputs` should be an array for Concatenate");if(i.length!==e.length)throw new M("Mismatch in the length of mask ("+i.length+") "+("and the legnth of inputs ("+e.length+")"));return y.tidy(function(){var a=!0;if(i.forEach(function(u){if(u!=null){a=!1;return}}),a)return null;for(var s=[],o=0;o<e.length;++o)i[o]==null?s.push(y.onesLike(e[o]).asType("bool")):i[o].rank<e[o].rank?s.push(y.expandDims(i[o],-1)):s.push(i[o]);var l=y.concat(s,r.axis);return y.all(l,-1,!1)})},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Concatenate",t}(ir);y.serialization.registerClass(db);function qa(n,t){for(;n<0;)n+=t;return n}function s4(n,t,e){if(n.shape.length>3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;l<s;++l)o.push(1);t=t.reshape(t.shape.concat(o))}else if(r>i){s=r-i;for(var o=[],l=0;l<s;++l)o.push(1);n=n.reshape(n.shape.concat(o))}else s=0;var u;if(n.shape.length===2&&t.shape.length===2)a[0]===a[1]?u=n.mul(t).sum(a[0]):u=n.transpose([1,0]).mul(t).sum(a[1]);else{var c=a[0]!==n.shape.length-1,h=a[1]===t.shape.length-1;u=n.matMul(t,c,h)}if(s>0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l<d+s;++l)p.push(l);u=u.squeeze(p)}return u.shape.length===1&&(u=u.expandDims(1)),u})}var pb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.axes=e.axes,i.normalize=e.normalize==null?!1:e.normalize,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0],r=e[1];if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return qa(s,e[o].shape.length)}):a=[qa(this.axes,i.shape.length),qa(this.axes,r.shape.length)],this.normalize&&(i=uo(i,a[0]),r=uo(r,a[1])),s4(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[qa(this.axes,e.length),qa(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(ir);y.serialization.registerClass(pb);var fb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return io(a.shape,0,r.stddev).add(a)},o=za(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass(fb);var mb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(io(a.shape,1,o))};return za(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(mb);var gb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=Wa(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return za(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(gb);function Ga(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function o4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=Ga(n,s,o,e,t,r);return[l,s,o]})}function l4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=gn(0,n.rank);u<c.length;u++){var h=c[u];i.indexOf(h)!==-1?l.push(1):l.push(n.shape[h])}var d=s.reshape(l),p=o.reshape(l),f=t==null?null:t.reshape(l),m=e==null?null:e.reshape(l),g=Ga(n,d,p,m,f,r);return[g,s,o]})}function u4(n,t,e,i,r){return r===void 0&&(r=.001),y.util.arraysEqual(i.slice().sort(),gn(0,n.rank-1))?o4(n,t,e,i,r):l4(n,t,e,i,r)}var vb=function(n){Q(t,n);function t(e){var i=this;return e==null&&(e={}),i=n.call(this,e)||this,i.supportsMasking=!0,i.axis=e.axis==null?-1:e.axis,i.momentum=e.momentum==null?.99:e.momentum,i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.movingMeanInitializer=tt(e.movingMeanInitializer||"zeros"),i.movingVarianceInitializer=tt(e.movingVarianceInitializer||"ones"),i.betaConstraint=bt(e.betaConstraint),i.gammaConstraint=bt(e.gammaConstraint),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.axis>=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Tt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=gn(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=Ji(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,gn(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),x=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return Ga(s,L,x,C,R,r.epsilon)}else return Ga(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=u4(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,x,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(x).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),movingMeanInitializer:st(this.movingMeanInitializer),movingVarianceInitializer:st(this.movingVarianceInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer),betaConstraint:yt(this.betaConstraint),gammaConstraint:yt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(vb);var yb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r<a.length;r++){var s=a[r];if(!Number.isInteger(s))throw new Error("Expected axis to be an array of integers, "+("but received "+JSON.stringify(i.axis)))}else throw new Error("Expected axis to be an integer or an array of integers, "+("but received "+JSON.stringify(i.axis)));return i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i.supportsMasking=!0,i}return t.prototype.build=function(e){e=Ke(e);var i=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(var r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=i);for(var a=0,s=this.axis;a<s.length;a++){var o=s[a];if(o<0||o>=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==fi(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=Ji(1,o),p=0,f=r.axis;p<f.length;p++){var m=f[p];d[m]=s[m]}for(var g=function(x){return x!=null&&x.shape.length!==o&&r.axis!==[o-1]?x.reshape(d):x},v=g(r.gamma.read()),b=g(r.beta.read()),w=[],S=[],L=0;L<o;++L)r.axis.indexOf(L)!==-1?(w.push(s[L]),S.push(1)):(w.push(1),S.push(s[L]));return c=c.tile(w),h=h.tile(w),v=v.tile(S),b=b.tile(S),Ga(a,c,h,b,v,r.epsilon)})},t.prototype.getConfig=function(){var e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LayerNormalization",t}(De);y.serialization.registerClass(yb);function c4(n,t,e){return y.tidy(function(){if(n.rank!==4)throw new M("temporalPadding expects input tensor to be 4-D, but received a "+(n.rank+"-D tensor."));if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new M("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(e==null&&(e=mn()),e!=="channelsLast"&&e!=="channelsFirst")throw new M("Unknown data format: "+e+". Supported data formats are 'channelsLast' and 'channelsFirst.");var i;return e==="channelsFirst"?i=[[0,0],[0,0],t[0],t[1]]:i=[[0,0],t[0],t[1],[0,0]],y.pad(n,i)})}var bb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.dataFormat=e.dataFormat==null?mn():e.dataFormat,e.padding==null)i.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")i.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new M("ZeroPadding2D expects padding to be a length-2 array, but "+("received a length-"+e.padding.length+" array."));var r=void 0,a=void 0;if(typeof e.padding[0]=="number")r=[e.padding[0],e.padding[0]],a=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new M("ZeroPadding2D expects height padding to be a length-2 array, "+("but received a length-"+e.padding[0].length+" array."));if(r=e.padding[0],e.padding[1].length!==2)throw new M("ZeroPadding2D expects width padding to be a length-2 array, "+("but received a length-"+e.padding[1].length+" array."));a=e.padding[1]}i.padding=[r,a]}return i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i,r;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return c4(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(bb);function Lo(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),tn(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=mn()),a==null&&(a="max"),n=Mh(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function wb(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),tn(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=mn()),a==null&&(a="max"),n=Uy(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var Sb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(At(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,tn(i.padding),i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=bn(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=Ua(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(Sb);y.serialization.registerClass(Lb);var Ib=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(Sb);y.serialization.registerClass(Ib);var Ab=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),tn(i.padding),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=bn(i,this.poolSize[0],this.padding,this.strides[0]),r=bn(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Tb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(Ab);y.serialization.registerClass(Tb);var Nb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(Ab);y.serialization.registerClass(Nb);var xb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),tn(i.padding),i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=bn(i,this.poolSize[0],this.padding,this.strides[0]),r=bn(r,this.poolSize[1],this.padding,this.strides[1]),a=bn(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),wb(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(xb);y.serialization.registerClass(Cb);var Rb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),wb(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(xb);y.serialization.registerClass(Rb);var Ob=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),Eb=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(Ob);y.serialization.registerClass(Eb);var Db=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(Ob);y.serialization.registerClass(Db);var kb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Fb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(kb);y.serialization.registerClass(Fb);var Wb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(kb);y.serialization.registerClass(Wb);var Ub=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=yn(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),Bb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=Ke(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Ky(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(Ub);y.serialization.registerClass(Bb);function h4(n){_r(nk,"BidirectionalMergeMode",n)}var d4="concat",zb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=yn(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=yn(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?d4:e.mergeMode,h4(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):Mt(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Tt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof vn,d=0,p=o;d<p.length;d++){var f=p[d];if(f instanceof vn!==h)throw new M("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors")}if(h){var m=[e].concat(o),g=this.inputSpec.concat(l),v=this.inputSpec;this.inputSpec=g;var b=n.prototype.apply.call(this,m,i);return this.inputSpec=v,b}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.initialState,s,o;if(a==null)s=r.forwardLayer.call(e,i),o=r.backwardLayer.call(e,i);else{var l=a.slice(0,a.length/2),u=a.slice(a.length/2);s=r.forwardLayer.call(e,Object.assign(i,{initialState:l})),o=r.backwardLayer.call(e,Object.assign(i,{initialState:u}))}var c;r.returnState&&(Array.isArray(s)&&(c=s.slice(1).concat(o.slice(1))),s=s[0],o=o[0]),r.returnSequences&&(o=y.reverse(o,1));var h;return r.mergeMode==="concat"?h=ph([s,o]):r.mergeMode==="sum"?h=y.add(s,o):r.mergeMode==="ave"?h=y.mul(.5,y.add(s,o)):r.mergeMode==="mul"?h=y.mul(s,o):r.mergeMode==null&&(h=[s,o]),r.returnState?r.mergeMode==null?h.concat(c):[h].concat(c):h})},t.prototype.resetStates=function(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()},t.prototype.build=function(e){var i=this;Qi(this.forwardLayer.name,function(){i.forwardLayer.build(e)}),Qi(this.backwardLayer.name,function(){i.backwardLayer.build(e)}),this.built=!0},t.prototype.computeMask=function(e,i){Array.isArray(i)&&(i=i[0]);var r;if(this.returnSequences?this.mergeMode==null?r=[i,i]:r=i:this.mergeMode==null?r=[null,null]:r=null,this.returnState){var a=this.forwardLayer.states,s=a.map(function(o){return null});return Array.isArray(r)?r.concat(s).concat(s):[r].concat(s).concat(s)}else return r},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e={mergeMode:this.mergeMode},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.fromConfig=function(e,i){var r=yn(i.layer);if(delete i.layer,i.numConstants!=null)throw new Te("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");var a=i;return a.layer=r,new e(a)},t.className="Bidirectional",t}(Ub);y.serialization.registerClass(zb);function p4(n){return new Pa(n)}function f4(n){return new ky(n)}function m4(n){return new Oy(n)}function g4(n){return new Ey(n)}function v4(n){return new Dy(n)}function y4(n){return new Wy(n)}function b4(n){return new Fy(n)}function w4(n){return new Hy(n)}function S4(n){return new Hh(n)}function L4(n){return new _y(n)}function I4(n){return new Py(n)}function A4(n){return new My(n)}function T4(n){return new Vy(n)}function N4(n){return new qy(n)}function x4(n){return new Gy(n)}function C4(n){return new tb(n)}function R4(n){return new Qy(n)}function O4(n){return new Kh(n)}function E4(n){return new Zy(n)}function D4(n){return new eb(n)}function k4(n){return new nb(n)}function F4(n){return new ib(n)}function W4(n){return new rb(n)}function U4(n){return new sb(n)}function B4(n){return new ob(n)}function z4(n){return new ub(n)}function P4(n){return new db(n)}function _4(n){return new cb(n)}function M4(n){return new hb(n)}function H4(n){return new lb(n)}function V4(n){return new pb(n)}function q4(n){return new vb(n)}function G4(n){return new yb(n)}function Y4(n){return new bb(n)}function jh(n){return new Ib(n)}function K4(n){return jh(n)}function j4(n){return jh(n)}function $h(n){return new Nb(n)}function $4(n){return $h(n)}function X4(n){return $h(n)}function Xh(n){return new Rb(n)}function J4(n){return Xh(n)}function Z4(n){return Xh(n)}function Q4(n){return new Eb(n)}function eF(n){return new Fb(n)}function Pb(n){return new Db(n)}function _b(n){return new Wb(n)}function Mb(n){return new Lb(n)}function Hb(n){return new Tb(n)}function tF(n){return new Cb(n)}function nF(n){return new $y(n)}function iF(n){return new Gh(n)}function rF(n){return new Xy(n)}function aF(n){return new So(n)}function sF(n){return new jy(n)}function oF(n){return new qh(n)}function lF(n){return new Jy(n)}function uF(n){return new Yh(n)}function cF(n){return new Si(n)}function hF(n){return new Vh(n)}function dF(n){return new zb(n)}function pF(n){return new Bb(n)}var fF=Pb,mF=_b,gF=Mb,vF=Hb;function yF(n){return new fb(n)}function bF(n){return new mb(n)}function wF(n){return new gb(n)}function SF(n){return new ab(n)}var LF={__proto__:null,inputLayer:p4,elu:f4,reLU:m4,leakyReLU:g4,prelu:v4,softmax:y4,thresholdedReLU:b4,conv1d:w4,conv2d:S4,conv2dTranspose:L4,conv3d:I4,separableConv2d:A4,cropping2D:T4,upSampling2d:N4,depthwiseConv2d:x4,activation:C4,dense:R4,dropout:O4,spatialDropout1d:E4,flatten:D4,repeatVector:k4,reshape:F4,permute:W4,embedding:U4,add:B4,average:z4,concatenate:P4,maximum:_4,minimum:M4,multiply:H4,dot:V4,batchNormalization:q4,layerNormalization:G4,zeroPadding2d:Y4,averagePooling1d:jh,avgPool1d:K4,avgPooling1d:j4,averagePooling2d:$h,avgPool2d:$4,avgPooling2d:X4,averagePooling3d:Xh,avgPool3d:J4,avgPooling3d:Z4,globalAveragePooling1d:Q4,globalAveragePooling2d:eF,globalMaxPooling1d:Pb,globalMaxPooling2d:_b,maxPooling1d:Mb,maxPooling2d:Hb,maxPooling3d:tF,gru:nF,gruCell:iF,lstm:rF,lstmCell:aF,simpleRNN:sF,simpleRNNCell:oF,convLstm2d:lF,convLstm2dCell:uF,rnn:cF,stackedRNNCells:hF,bidirectional:dF,timeDistributed:pF,globalMaxPool1d:fF,globalMaxPool2d:mF,maxPool1d:gF,maxPool2d:vF,Layer:De,RNN:Si,RNNCell:Kr,input:Ty,gaussianNoise:yF,gaussianDropout:bF,alphaDropout:wF,masking:SF};function IF(n,t){return xh(n,t)}function AF(n,t){return ly(n,t)}function TF(n,t){return uy(n,t)}function NF(n,t){return Ch(n,t)}function xF(n,t){return Rh(n,t)}function CF(n,t){return oy(n,t)}function RF(n,t){return Jk(n,t)}function OF(n,t){return Th(n,t)}function EF(n,t){return co(n,t)}function DF(n,t){return qr(n,t)}function kF(n,t){return qr(n,t)}function FF(n,t){return qr(n,t)}function WF(n,t){return tr(n,t)}function UF(n,t){return tr(n,t)}function BF(n,t){return tr(n,t)}var zF={__proto__:null,binaryAccuracy:IF,binaryCrossentropy:AF,sparseCategoricalAccuracy:TF,categoricalAccuracy:NF,categoricalCrossentropy:xF,precision:CF,recall:RF,cosineProximity:OF,meanAbsoluteError:EF,meanAbsolutePercentageError:DF,MAPE:kF,mape:FF,meanSquaredError:WF,MSE:UF,mse:BF};var PF={__proto__:null,modelFromJSON:k3};function _F(n){return new yo(n)}function MF(n){return Q3(n)}function HF(n){return e4(n)}var VF={__proto__:null,l1l2:_F,l1:MF,l2:HF};var Vb=function(n){Q(t,n);function t(){var e=n!==null&&n.apply(this,arguments)||this;return e.model=null,e}return t.prototype.setModel=function(e){if(!(e instanceof yi))throw new Error("model must be a LayersModel, not some other Container");this.model=e},t}(Vr);function Io(n,t){return n<t}function qb(n,t){return n>t}var Gb=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=Io:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=qb:i.monitorFunc=Io,i.monitorFunc===Io&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===Io?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,vi(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(Vb);function qF(n){return new Gb(n)}var GF={earlyStopping:qF};Xe.Callback=Vb;Xe.CallbackList=ey;Xe.CustomCallback=ny;Xe.EarlyStopping=Gb;Xe.History=ty;Xe.InputSpec=Tt;Xe.LayerVariable=$v;Xe.LayersModel=yi;Xe.RNN=Si;Xe.Sequential=zh;Xe.SymbolicTensor=vn;Xe.callbacks=GF;Xe.constraints=ZD;Xe.initializers=Dk;Xe.input=Ty;Xe.layers=LF;Xe.loadLayersModel=P3;Xe.metrics=zF;Xe.model=B3;Xe.models=PF;Xe.registerCallbackConstructor=_3;Xe.regularizers=VF;Xe.sequential=z3;Xe.version_layers=Dh});var ow=be(rr=>{"use strict";Object.defineProperty(rr,"__esModule",{value:!0});var B=Xi();var Kb=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t};function zn(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function wn(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=r[u[0]&2?"return":u[0]?"throw":"next"])&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[0,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Sn;(function(n){n[n.DT_INVALID=0]="DT_INVALID",n[n.DT_FLOAT=1]="DT_FLOAT",n[n.DT_DOUBLE=2]="DT_DOUBLE",n[n.DT_INT32=3]="DT_INT32",n[n.DT_UINT8=4]="DT_UINT8",n[n.DT_INT16=5]="DT_INT16",n[n.DT_INT8=6]="DT_INT8",n[n.DT_STRING=7]="DT_STRING",n[n.DT_COMPLEX64=8]="DT_COMPLEX64",n[n.DT_INT64=9]="DT_INT64",n[n.DT_BOOL=10]="DT_BOOL",n[n.DT_QINT8=11]="DT_QINT8",n[n.DT_QUINT8=12]="DT_QUINT8",n[n.DT_QINT32=13]="DT_QINT32",n[n.DT_BFLOAT16=14]="DT_BFLOAT16",n[n.DT_FLOAT_REF=101]="DT_FLOAT_REF",n[n.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",n[n.DT_INT32_REF=103]="DT_INT32_REF",n[n.DT_UINT8_REF=104]="DT_UINT8_REF",n[n.DT_INT16_REF=105]="DT_INT16_REF",n[n.DT_INT8_REF=106]="DT_INT8_REF",n[n.DT_STRING_REF=107]="DT_STRING_REF",n[n.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",n[n.DT_INT64_REF=109]="DT_INT64_REF",n[n.DT_BOOL_REF=110]="DT_BOOL_REF",n[n.DT_QINT8_REF=111]="DT_QINT8_REF",n[n.DT_QUINT8_REF=112]="DT_QUINT8_REF",n[n.DT_QINT32_REF=113]="DT_QINT32_REF",n[n.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(Sn||(Sn={}));var jb;(function(n){var t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=n.CheckpointFormatVersion||(n.CheckpointFormatVersion={}))})(jb||(jb={}));var Jh={};function YF(n,t){var e={tfOpName:n,category:"custom",inputs:[],attrs:[],customExecutor:t};Jh[n]=e}function $b(n){return Jh[n]}function KF(n){delete Jh[n]}function A(n,t,e,i){var r=t.inputParams[n];if(r&&r.inputIndexStart!==void 0){var a=r.inputIndexStart,s=r.inputIndexEnd===0?void 0:r.inputIndexEnd===void 0?a+1:r.inputIndexEnd;if(r.type==="tensor")return Ht(t.inputNames[r.inputIndexStart],e,i);if(r.type==="tensors"){var o=t.inputNames.slice(a,s);return o.map(function(h){return Ht(h,e,i)})}var l=Ht(t.inputNames.slice(a)[0],e,i),u=l.dataSync();return r.type==="number"?u[0]:B.util.toNestedArray(l.shape,u)}var c=t.attrParams[n];return c&&c.value}function Ht(n,t,e){var i=jt(n),r=i[0],a=i[1],s=e.currentContextIds.find(function(o){return!!t[Ao(r,o)]});return s!==void 0?t[Ao(r,s)][a]:void 0}function jF(n,t,e){return t[Ao(n,e.currentContextId)]}function Zn(n,t){var e=jt(n),i=e[0],r=e[1];return[Ao(i,t&&t.currentContextId),r]}function Ao(n,t){return t?n+"-"+t:n}function jt(n){var t=n.split(":");if(t.length===1)return[n,0];var e=t[0];return[e,Number(t[t.length-1])]}function To(n,t,e){var i=A("pad",n,t,e);if(i==="explicit"){i=A("explicitPaddings",n,t,e);for(var r=[[0,0],[0,0],[0,0],[0,0]],a=0;a<4;a++)r[a][0]=i[a*2],r[a][1]=i[a*2+1];return r}return i}function Qn(n){return n.kept?n:B.clone(n)}var $F=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],XF={__proto__:null,json:$F};var JF=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"clip_value_min",name:"clipValueMin",type:"number"},{tfName:"clip_value_max",name:"clipValueMax",type:"number"}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"clipValueMin",name:"clipValueMin",type:"number",defaultValue:0},{tfName:"clipValueMax",name:"clipValueMax",type:"number",defaultValue:6}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],ZF={__proto__:null,json:JF};var QF=[{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],eW={__proto__:null,json:QF};var tW=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],nW={__proto__:null,json:tW};var iW=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],rW={__proto__:null,json:iW};var aW=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],sW={__proto__:null,json:aW};var oW=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],lW={__proto__:null,json:oW};var uW=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],cW={__proto__:null,json:uW};var hW=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],dW={__proto__:null,json:hW};var pW=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],fW={__proto__:null,json:pW};var mW=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],gW={__proto__:null,json:mW};var vW=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],yW={__proto__:null,json:vW};var bW=[{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],wW={__proto__:null,json:bW};var SW=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool",notSupported:!0}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],LW={__proto__:null,json:SW};var IW=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],AW={__proto__:null,json:IW};var TW=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]}],NW={__proto__:null,json:TW};var Jb=function(){function n(){var t=[XF,ZF,eW,nW,rW,sW,lW,fW,dW,cW,gW,yW,wW,LW,AW,NW],e=[].concat.apply([],t.map(function(i){return i.json}));this.opMappers=e.reduce(function(i,r){return i[r.tfOpName]=r,i},{})}return Object.defineProperty(n,"Instance",{get:function(){return this._instance||(this._instance=new this)},enumerable:!0,configurable:!0}),n.prototype.transformGraph=function(t,e){var i=this;e===void 0&&(e={});var r=t.node,a=[],s=[],o=[],l=r.reduce(function(g,v){return g[v.name]=i.mapNode(v),v.op.startsWith("Placeholder")?a.push(g[v.name]):v.op==="Const"?s.push(g[v.name]):(v.input==null||v.input.length===0)&&o.push(g[v.name]),g},{}),u=[],c=[],h={},d={};e!=null&&(h=this.mapSignatureEntries(e.inputs),d=this.mapSignatureEntries(e.outputs));var p=Object.keys(l);p.forEach(function(g){var v=l[g];v.inputNames.forEach(function(b){var w=Zn(b)[0];v.inputs.push(l[w]),l[w].children.push(v)})}),Object.keys(d).length===0?p.forEach(function(g){var v=l[g];v.children.length===0&&c.push(v)}):Object.keys(d).forEach(function(g){var v=Zn(g)[0],b=l[v];b!=null&&(b.signatureKey=d[g],c.push(b))}),Object.keys(h).length>0?Object.keys(h).forEach(function(g){var v=Zn(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=$b(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Zh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Zh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=sd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=sd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=ed(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=ed(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=ad(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=ad(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Qh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Qh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=ld(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=ld(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=od(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=nd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=nd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=id(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=id(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Xb(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Xb(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=Zn(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:td(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=Zn(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=Zn(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function xW(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function Zb(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):xW(n);return t?e:e.toLowerCase()}function Zh(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?Zb(r.s,i):e}function Qh(n,t,e){var i=n[t];return i?i.b:e}function ed(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function td(n){typeof n=="string"&&(n=Sn[n]);switch(n){case Sn.DT_FLOAT:return"float32";case Sn.DT_INT32:case Sn.DT_INT64:case Sn.DT_INT8:case Sn.DT_UINT8:return"int32";case Sn.DT_BOOL:return"bool";case Sn.DT_DOUBLE:return"float32";case Sn.DT_STRING:return"string";default:return null}}function Xb(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function nd(n,t,e){var i=n[t];return i&&i.type?td(i.type):e}function id(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return td(r)}):e}function Qb(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function rd(n,t,e){var i=n[t];return i&&i.shape?Qb(i.shape):e}function ad(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function sd(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return Zb(a,i)}):e}function od(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Qb(r)}):e}function ld(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var CW=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Ht(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Ht(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return ed(this.node.rawAttrs,t,e);if(i.s!=null)return Zh(this.node.rawAttrs,t,e);if(i.b!=null)return Qh(this.node.rawAttrs,t,e);if(i.shape!=null)return rd(this.node.rawAttrs,t,e);if(i.type!=null)return nd(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return ad(this.node.rawAttrs,t,e);if(i.list.s!=null)return sd(this.node.rawAttrs,t,e);if(i.list.shape!=null)return od(this.node.rawAttrs,t,e);if(i.list.b!=null)return ld(this.node.rawAttrs,t,e);if(i.list.type!=null)return id(this.node.rawAttrs,t,e)}return e},n}();var RW=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var OW=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Ht(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function cn(n,t,e){e===void 0&&(e=""),B.util.assert(EW(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function EW(n,t){if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==-1&&t[e]!==-1&&n[e]!==t[e])return!1;return!0}var DW=function(){function n(t,e,i,r,a,s,o){this.name=t,this.dtype=e,this.maxSize=i,this.elementShape=r,this.identicalElementShapes=a,this.dynamicSize=s,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=B.scalar(0),B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"closed",{get:function(){return this.closed_},enumerable:!0,configurable:!0}),n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.read=function(t){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||t>=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`,
+          because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),cn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i<this.size();i++)t.push(i)}if(t.length===0)return B.tensor([],[0].concat(this.elementShape));var r=this.readMany(t);return cn(this.elementShape,r[0].shape,"TensorArray shape mismatch: "),B.stack(r,0)},n.prototype.concat=function(t){if(!!t&&t!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but concat requested dtype "+t);if(this.size()===0)return B.tensor([],[0].concat(this.elementShape));for(var e=[],i=0;i<this.size();i++)e.push(i);var r=this.readMany(e);return cn(this.elementShape,r[0].shape,"TensorArray shape mismatch: tensor array shape ("+this.elementShape+") vs first tensor shape ("+r[0].shape+")"),B.concat(r,0)},n.prototype.scatter=function(t,e){if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);if(t.length!==e.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+e.shape[0]);var i=Math.max.apply(Math,t);if(!this.dynamicSize&&i>=this.maxSize)throw new Error("Max index must be < array size ("+i+"  vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
           tensor.shape[0], but sum of lengths is
-        `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;c<t.length;++c){var h=c===0?0:a[c-1],d=[0,h,0],p=[1,t[c],s];o[c]=B.reshape(B.slice(e,d,p),i.elementShape)}return o});for(var l=[],u=0;u<t.length;u++)l[u]=u;this.writeMany(l,o)},n}();var No=function(){function n(t,e,i,r){r===void 0&&(r=-1),this.tensors=t,this.elementShape=e,this.elementDtype=i,t!=null&&t.forEach(function(a){if(i!==a.dtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+a.dtype);hn(e,a.shape,"TensorList shape mismatch: "),B.keep(a)}),this.idTensor=B.scalar(0),this.maxNumElements=r,B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),n.prototype.copy=function(){return new n(this.tensors.slice(),this.elementShape,this.elementDtype)},n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.stack=function(t,e,i){var r=this;if(i===void 0&&(i=-1),e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(i!==-1&&this.tensors.length!==i)throw new Error("Operation expected a list with "+i+" elements but got a list with "+this.tensors.length+" elements.");return hn(t,this.elementShape,"TensorList shape mismatch: "),B.tidy(function(){var a=r.tensors.map(function(s){return B.reshape(s,t)});return B.stack(a,0)})},n.prototype.popBack=function(t,e){if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(this.size()===0)throw new Error("Trying to pop from an empty list.");var i=this.tensors.pop();return hn(i.shape,t,"TensorList shape mismatch: "),B.reshape(i,t)},n.prototype.pushBack=function(t){if(t.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+t.dtype+", but list elements "+this.elementDtype);if(hn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");B.keep(t),this.tensors.push(t)},n.prototype.resize=function(t){if(t<0)throw new Error("TensorListResize expects size to be non-negative. Got: "+t);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return hn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");hn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return hn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return hn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function kW(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);hn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new No(a,t,i)}function FW(n,t,e){return new No([],n,t,e)}function WW(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+"  vs. "+i+")");var a=new No([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function UW(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to
+        `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;c<t.length;++c){var h=c===0?0:a[c-1],d=[0,h,0],p=[1,t[c],s];o[c]=B.reshape(B.slice(e,d,p),i.elementShape)}return o});for(var l=[],u=0;u<t.length;u++)l[u]=u;this.writeMany(l,o)},n}();var No=function(){function n(t,e,i,r){r===void 0&&(r=-1),this.tensors=t,this.elementShape=e,this.elementDtype=i,t!=null&&t.forEach(function(a){if(i!==a.dtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+a.dtype);cn(e,a.shape,"TensorList shape mismatch: "),B.keep(a)}),this.idTensor=B.scalar(0),this.maxNumElements=r,B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),n.prototype.copy=function(){return new n(this.tensors.slice(),this.elementShape,this.elementDtype)},n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.stack=function(t,e,i){var r=this;if(i===void 0&&(i=-1),e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(i!==-1&&this.tensors.length!==i)throw new Error("Operation expected a list with "+i+" elements but got a list with "+this.tensors.length+" elements.");return cn(t,this.elementShape,"TensorList shape mismatch: "),B.tidy(function(){var a=r.tensors.map(function(s){return B.reshape(s,t)});return B.stack(a,0)})},n.prototype.popBack=function(t,e){if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(this.size()===0)throw new Error("Trying to pop from an empty list.");var i=this.tensors.pop();return cn(i.shape,t,"TensorList shape mismatch: "),B.reshape(i,t)},n.prototype.pushBack=function(t){if(t.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+t.dtype+", but list elements "+this.elementDtype);if(cn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");B.keep(t),this.tensors.push(t)},n.prototype.resize=function(t){if(t<0)throw new Error("TensorListResize expects size to be non-negative. Got: "+t);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return cn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");cn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return cn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return cn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function kW(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);cn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new No(a,t,i)}function FW(n,t,e){return new No([],n,t,e)}function WW(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+"  vs. "+i+")");var a=new No([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function UW(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to
           tensor.shape[0], but sum of lengths is
-        `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c<t.length;++c){var h=c===0?0:r[c-1],d=[0,h,0],p=[1,t[c],a];u[c]=B.reshape(B.slice(n,d,p),e)}return n.dispose(),u}),o=new No([],e,n.dtype,t.length),l=0;l<s.length;l++)o.setItem(l,s[l]);return o}var BW=void 0,zW=function(n,t,e){return zn(BW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,o,h,d,l,p,f,m,m,g,v,g,b,g,g,g,w,S,L,x,C,R,D,k,W,F,P,H,_,K,j,q,G,Z,X,ee,ne,ie,te,re,le,he,ye,Oe,Fe,_e,Me,ht,St,Ge,Ue,F,P,me,Ue,K,L,Dt,me,ne,ie,L,lt,me,L,dt,lt,me,q,G,L,dt,me,Ue,L,dt,lt,me,xn,L,dt,me,re,me,he,L,Ue,P,me,Ue,L,Dt,me,Oe,L,Fe,me;return Sn(this,function(Rt){switch(Rt.label){case 0:i=n.op;switch(i){case"If":return[3,1];case"StatelessIf":return[3,1];case"While":return[3,3];case"StatelessWhile":return[3,3];case"LoopCond":return[3,9];case"Switch":return[3,10];case"Merge":return[3,12];case"Enter":return[3,13];case"Exit":return[3,14];case"NextIteration":return[3,15];case"TensorArrayV3":return[3,16];case"TensorArrayWriteV3":return[3,17];case"TensorArrayReadV3":return[3,18];case"TensorArrayGatherV3":return[3,19];case"TensorArrayScatterV3":return[3,20];case"TensorArrayConcatV3":return[3,21];case"TensorArraySplitV3":return[3,22];case"TensorArraySizeV3":return[3,23];case"TensorArrayCloseV3":return[3,24];case"TensorListSetItem":return[3,25];case"TensorListGetItem":return[3,26];case"TensorListScatterV2":return[3,27];case"TensorListScatter":return[3,27];case"TensorListReserve":return[3,28];case"TensorListGather":return[3,29];case"TensorListStack":return[3,30];case"TensorListFromTensor":return[3,31];case"TensorListConcat":return[3,32];case"TensorListPushBack":return[3,33];case"TensorListPopBack":return[3,34];case"TensorListSplit":return[3,35]}return[3,36];case 1:return r=A("thenBranch",n,t,e),a=A("elseBranch",n,t,e),s=A("cond",n,t,e),o=A("args",n,t,e),[4,s.data()];case 2:return l=Rt.sent(),l[0]?[2,e.functionMap[r].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)]:[2,e.functionMap[a].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 3:return u=A("body",n,t,e),c=A("cond",n,t,e),o=A("args",n,t,e),[4,e.functionMap[c].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 4:return h=Rt.sent(),d=o.map(function(Lt){return Lt.id}),[4,h[0].data()];case 5:l=Rt.sent(),h.forEach(function(Lt){!Lt.kept&&d.indexOf(Lt.id)===-1&&Lt.dispose()}),p=o,f=function(){var Lt,Ce,kt;return Sn(this,function(pt){switch(pt.label){case 0:return Lt=p,[4,e.functionMap[u].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 1:return p=pt.sent(),Ce=p.map(function(it){return it.id}),Lt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[4,e.functionMap[c].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 2:return kt=pt.sent(),[4,kt[0].data()];case 3:return l=pt.sent(),kt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[2]}})},Rt.label=6;case 6:return l[0]?[5,f()]:[3,8];case 7:return Rt.sent(),[3,6];case 8:return[2,p];case 9:return m=A("pred",n,t,e),[2,[Qn(m)]];case 10:return m=A("pred",n,t,e),g=A("data",n,t,e),g.kept||(g=Qn(g)),[4,m.data()];case 11:return[2,Rt.sent()[0]?[void 0,g]:[g,void 0]];case 12:return v=n.inputNames.find(function(Lt){return Ht(Lt,t,e)!==void 0}),v?(g=Ht(v,t,e),[2,[Qn(g)]]):[2,void 0];case 13:return b=A("frameName",n,t,e),g=A("tensor",n,t,e),e.enterFrame(b),[2,[Qn(g)]];case 14:return g=A("tensor",n,t,e),e.exitFrame(),[2,[Qn(g)]];case 15:return g=A("tensor",n,t,e),e.nextIteration(),[2,[Qn(g)]];case 16:return w=A("size",n,t,e),S=A("dtype",n,t,e),L=A("elementShape",n,t,e),x=A("dynamicSize",n,t,e),C=A("clearAfterRead",n,t,e),R=A("identicalElementShapes",n,t,e),D=A("name",n,t,e),k=new DW(D,S,w,L,R,x,C),e.addTensorArray(k),[2,[k.idTensor,B.scalar(1)]];case 17:return W=A("tensorArrayId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),H=e.getTensorArray(W.id),H.write(F,P),[2,[H.idTensor]];case 18:return _=A("tensorArrayId",n,t,e),K=A("index",n,t,e),j=e.getTensorArray(_.id),[2,[j.read(K)]];case 19:return q=A("tensorArrayId",n,t,e),G=A("indices",n,t,e),Z=A("dtype",n,t,e),X=e.getTensorArray(q.id),[2,[X.gather(G,Z)]];case 20:return ee=A("tensorArrayId",n,t,e),ne=A("indices",n,t,e),ie=A("tensor",n,t,e),te=e.getTensorArray(ee.id),te.scatter(ne,ie),[2,[te.idTensor]];case 21:return re=A("tensorArrayId",n,t,e),le=e.getTensorArray(re.id),he=A("dtype",n,t,e),[2,[le.concat(he)]];case 22:return ye=A("tensorArrayId",n,t,e),Oe=A("tensor",n,t,e),Fe=A("lengths",n,t,e),_e=e.getTensorArray(ye.id),_e.split(Fe,Oe),[2,[_e.idTensor]];case 23:return Me=A("tensorArrayId",n,t,e),ht=e.getTensorArray(Me.id),[2,[B.scalar(ht.size(),"int32")]];case 24:return St=A("tensorArrayId",n,t,e),Ge=e.getTensorArray(St.id),Ge.clearAndClose(),[2,[Ge.idTensor]];case 25:return Ue=A("tensorListId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.setItem(F,P),[2,[me.idTensor]];case 26:return Ue=A("tensorListId",n,t,e),K=A("index",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.getItem(K,L,Dt)]];case 27:return ne=A("indices",n,t,e),ie=A("tensor",n,t,e),L=A("elementShape",n,t,e),lt=A("numElements",n,t,e),me=WW(ie,ne,L,lt),e.addTensorList(me),[2,[me.idTensor]];case 28:return L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=FW(L,dt,lt),e.addTensorList(me),[2,[me.idTensor]];case 29:return q=A("tensorListId",n,t,e),G=A("indices",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=e.getTensorList(q.id),[2,[me.gather(G,dt,L)]];case 30:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=e.getTensorList(Ue.id),[2,[me.stack(L,dt,lt)]];case 31:return xn=A("tensor",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=kW(xn,L,dt),e.addTensorList(me),[2,[me.idTensor]];case 32:return re=A("tensorListId",n,t,e),me=e.getTensorList(re.id),he=A("dtype",n,t,e),L=A("elementShape",n,t,e),[2,[me.concat(he,L)]];case 33:return Ue=A("tensorListId",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.pushBack(P),[2,[me.idTensor]];case 34:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.popBack(L,Dt)]];case 35:return Oe=A("tensor",n,t,e),L=A("elementShape",n,t,e),Fe=A("lengths",n,t,e),me=UW(Oe,Fe,L),e.addTensorList(me),[2,[me.idTensor]];case 36:throw TypeError("Node type "+n.op+" is not implemented")}})})};function ew(n,t,e){var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=r==="fusedbatchnorm",u=A("numArgs",n,t,e);if(s){if(o&&u!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&u!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");var c=A("strides",n,t,e),h=To(n,t,e),d=A("dataFormat",n,t,e).toUpperCase(),p=A("dilations",n,t,e),f=A("args",n,t,e),m=f[0],g=f[1];return{stride:c,pad:h,dataFormat:d,dilations:p,biasArg:m,preluArg:g,activationFunc:a}}var PW=function(n,t,e){switch(n.op){case"Conv1D":{var i=A("stride",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),s=A("dilation",n,t,e);return[B.conv1d(A("x",n,t,e),A("filter",n,t,e),i,r,a,s)]}case"Conv2D":{var i=A("strides",n,t,e),r=To(n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv2d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{var l=ew(n,t,e),i=l.stride,r=l.pad,a=l.dataFormat,o=l.dilations,u=l.biasArg,c=l.preluArg,h=l.activationFunc;return[B.fused.conv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"FusedDepthwiseConv2dNative":{var d=ew(n,t,e),i=d.stride,r=d.pad,a=d.dataFormat,o=d.dilations,u=d.biasArg,c=d.preluArg,h=d.activationFunc;return[B.fused.depthwiseConv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{var p=A("outputShape",n,t,e),i=A("strides",n,t,e),r=To(n,t,e);return[B.conv2dTranspose(A("x",n,t,e),A("filter",n,t,e),p,[i[1],i[2]],r)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{var i=A("strides",n,t,e),r=To(n,t,e),o=A("dilations",n,t,e),a=A("dataFormat",n,t,e).toUpperCase();return[B.depthwiseConv2d(A("input",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"Conv3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv3d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2],i[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPoolWithArgmax":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e),m=A("includeBatchInIndex",n,t,e),g=B.maxPoolWithArgmax(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r,m),v=g.result,b=g.indexes;return[v,b]}case"AvgPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"MaxPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"Dilation2D":{var w=A("strides",n,t,e),r=A("pad",n,t,e),o=A("dilations",n,t,e),S=w[1],L=w[2],x=o[1],C=o[2];return[B.dilation2d(A("x",n,t,e),A("filter",n,t,e),[S,L],r,[x,C],"NHWC")]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var _W=function(n,t,e){switch(n.op){case"Fill":{var i=A("shape",n,t,e),r=A("dtype",n,t,e),a=A("value",n,t,e);return[B.fill(i,a,r)]}case"LinSpace":{var s=A("start",n,t,e),o=A("stop",n,t,e),l=A("num",n,t,e);return[B.linspace(s,o,l)]}case"Multinomial":{var u=A("logits",n,t,e),c=A("numSamples",n,t,e),h=A("seed",n,t,e);return[B.multinomial(u,c,h)]}case"OneHot":{var d=A("indices",n,t,e),p=A("depth",n,t,e),f=A("onValue",n,t,e),m=A("offValue",n,t,e);return[B.oneHot(d,p,f,m)]}case"Ones":return[B.ones(A("shape",n,t,e),A("dtype",n,t,e))];case"OnesLike":return[B.onesLike(A("x",n,t,e))];case"RandomUniform":return[B.randomUniform(A("shape",n,t,e),A("minval",n,t,e),A("maxval",n,t,e),A("dtype",n,t,e))];case"Range":{var s=A("start",n,t,e),g=A("stop",n,t,e),v=A("step",n,t,e);return[B.range(s,g,v,A("dtype",n,t,e))]}case"TruncatedNormal":{var i=A("shape",n,t,e),b=A("mean",n,t,e),w=A("stdDev",n,t,e),h=A("seed",n,t,e);return[B.truncatedNormal(i,b,w,A("dtype",n,t,e),h)]}case"Zeros":return[B.zeros(A("shape",n,t,e),A("dtype",n,t,e))];case"ZerosLike":return[B.zerosLike(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var MW=void 0;function ld(n,t,e){var i=A("boxes",n,t,e),r=A("scores",n,t,e),a=A("maxOutputSize",n,t,e),s=A("iouThreshold",n,t,e),o=A("scoreThreshold",n,t,e),l=A("softNmsSigma",n,t,e);return{boxes:i,scores:r,maxOutputSize:a,iouThreshold:s,scoreThreshold:o,softNmsSigma:l}}var HW=function(n,t,e){return zn(MW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,a,s,o,l,u,p,h,f,a,s,o,l,u,m,h;return Sn(this,function(g){switch(g.label){case 0:i=n.op;switch(i){case"NonMaxSuppressionV5":return[3,1];case"NonMaxSuppressionV4":return[3,3];case"NonMaxSuppressionV3":return[3,5];case"NonMaxSuppressionV2":return[3,5];case"Where":return[3,7];case"ListDiff":return[3,9]}return[3,10];case 1:return r=ld(n,t,e),a=r.boxes,s=r.scores,o=r.maxOutputSize,l=r.iouThreshold,u=r.scoreThreshold,c=r.softNmsSigma,[4,B.image.nonMaxSuppressionWithScoreAsync(a,s,o,l,u,c)];case 2:return h=g.sent(),[2,[h.selectedIndices,h.selectedScores]];case 3:return d=ld(n,t,e),a=d.boxes,s=d.scores,o=d.maxOutputSize,l=d.iouThreshold,u=d.scoreThreshold,p=A("padToMaxOutputSize",n,t,e),[4,B.image.nonMaxSuppressionPaddedAsync(a,s,o,l,u,p)];case 4:return h=g.sent(),[2,[h.selectedIndices,h.validOutputs]];case 5:return f=ld(n,t,e),a=f.boxes,s=f.scores,o=f.maxOutputSize,l=f.iouThreshold,u=f.scoreThreshold,[4,B.image.nonMaxSuppressionAsync(a,s,o,l,u)];case 6:return[2,[g.sent()]];case 7:return m=B.cast(A("condition",n,t,e),"bool"),[4,B.whereAsync(m)];case 8:return h=[g.sent()],m.dispose(),[2,h];case 9:return[2,B.setdiff1dAsync(A("x",n,t,e),A("y",n,t,e))];case 10:throw TypeError("Node type "+n.op+" is not implemented")}})})};var VW=function(n,t,e){switch(n.op){case"TopKV2":{var i=A("x",n,t,e),r=A("k",n,t,e),a=A("sorted",n,t,e),s=B.topk(i,r,a);return[s.values,s.indices]}case"Unique":{var i=A("x",n,t,e),s=B.unique(i);return[s.values,s.indices]}case"UniqueV2":{var i=A("x",n,t,e),o=A("axis",n,t,e),s=B.unique(i,o);return[s.values,s.indices]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var qW=function(n,t,e){switch(n.op){case"Const":return t[n.name];case"PlaceholderWithDefault":var i=A("default",n,t,e);return[Ht(n.name,t,e)||i];case"Placeholder":return[Ht(n.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{var r=A("x",n,t,e);return[Qn(r)]}case"IdentityN":return A("x",n,t,e).map(function(h){return Qn(h)});case"Snapshot":var a=A("x",n,t,e);return[Qn(a)];case"Shape":return[B.tensor1d(A("x",n,t,e).shape,"int32")];case"ShapeN":return A("x",n,t,e).map(function(h){return B.tensor1d(h.shape)});case"Size":return[B.scalar(A("x",n,t,e).size,"int32")];case"Rank":return[B.scalar(A("x",n,t,e).rank,"int32")];case"NoOp":return[B.scalar(1)];case"Print":var s=A("x",n,t,e),o=A("data",n,t,e),l=A("message",n,t,e),u=A("summarize",n,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(l);for(var c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,u));return[s];default:throw TypeError("Node type "+n.op+" is not implemented")}};var GW=function(n,t,e){switch(n.op){case"ResizeBilinear":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeBilinear(i,[r[0],r[1]],a)]}case"ResizeNearestNeighbor":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeNearestNeighbor(i,[r[0],r[1]],a)]}case"CropAndResize":{var s=A("image",n,t,e),o=A("boxes",n,t,e),l=A("boxInd",n,t,e),u=A("cropSize",n,t,e),c=A("method",n,t,e),h=A("extrapolationValue",n,t,e);return[B.image.cropAndResize(s,o,l,u,c,h)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var YW=function(n,t,e){switch(n.op){case"Equal":return[B.equal(A("a",n,t,e),A("b",n,t,e))];case"NotEqual":return[B.notEqual(A("a",n,t,e),A("b",n,t,e))];case"Greater":return[B.greater(A("a",n,t,e),A("b",n,t,e))];case"GreaterEqual":return[B.greaterEqual(A("a",n,t,e),A("b",n,t,e))];case"Less":return[B.less(A("a",n,t,e),A("b",n,t,e))];case"LessEqual":return[B.lessEqual(A("a",n,t,e),A("b",n,t,e))];case"LogicalAnd":return[B.logicalAnd(A("a",n,t,e),A("b",n,t,e))];case"LogicalNot":return[B.logicalNot(A("a",n,t,e))];case"LogicalOr":return[B.logicalOr(A("a",n,t,e),A("b",n,t,e))];case"Select":case"SelectV2":return[B.where(A("condition",n,t,e),A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var KW=function(n,t,e){switch(n.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[B.matMul(A("a",n,t,e),A("b",n,t,e),A("transposeA",n,t,e),A("transposeB",n,t,e))];case"Transpose":return[B.transpose(A("x",n,t,e),A("perm",n,t,e))];case"_FusedMatMul":var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=A("numArgs",n,t,e);if(s){if(o&&l!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&l!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}var u=A("args",n,t,e),c=u[0],h=u[1];return[B.fused.matMul({a:A("a",n,t,e),b:A("b",n,t,e),transposeA:A("transposeA",n,t,e),transposeB:A("transposeB",n,t,e),bias:c,activation:a,preluActivationWeights:h})];default:throw TypeError("Node type "+n.op+" is not implemented")}};var jW=function(n,t,e){switch(n.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"FusedBatchNormV3":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"LRN":return[B.localResponseNormalization(A("x",n,t,e),A("radius",n,t,e),A("bias",n,t,e),A("alpha",n,t,e),A("beta",n,t,e))];case"Softmax":return[B.softmax(A("x",n,t,e))];case"LogSoftmax":return[B.logSoftmax(A("x",n,t,e))];case"SparseToDense":return[B.sparseToDense(A("sparseIndices",n,t,e),A("outputShape",n,t,e),A("sparseValues",n,t,e),A("defaultValue",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var $W=function(n,t,e){switch(n.op){case"Max":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.max(A("x",n,t,e),i,r)]}case"Mean":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.mean(A("x",n,t,e),i,r)]}case"Min":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.min(A("x",n,t,e),i,r)]}case"Sum":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.sum(A("x",n,t,e),i,r)]}case"All":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.all(A("x",n,t,e),i,r)]}case"Any":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.any(A("x",n,t,e),i,r)]}case"ArgMax":{var i=A("axis",n,t,e);return[B.argMax(A("x",n,t,e),i)]}case"ArgMin":{var i=A("axis",n,t,e);return[B.argMin(A("x",n,t,e),i)]}case"Prod":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.prod(A("x",n,t,e),i,r)]}case"Cumsum":{var i=A("axis",n,t,e),a=A("exclusive",n,t,e),s=A("reverse",n,t,e);return[B.cumsum(A("x",n,t,e),i,a,s)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var XW=function(n,t,e){switch(n.op){case"ConcatV2":case"Concat":{var i=A("n",n,t,e),r=A("axis",n,t,e),a=A("tensors",n,t,e);return a=a.slice(0,i),[B.concat(a,r)]}case"GatherV2":case"Gather":{var r=A("axis",n,t,e),s=A("x",n,t,e),o=A("indices",n,t,e);return[B.gather(s,B.cast(o,"int32"),r)]}case"ReverseV2":case"Reverse":{var r=A("axis",n,t,e),s=A("x",n,t,e);return[B.reverse(s,r)]}case"Slice":{var l=A("begin",n,t,e),u=A("size",n,t,e);return[B.slice(A("x",n,t,e),l,u)]}case"StridedSlice":{var l=A("begin",n,t,e),c=A("end",n,t,e),h=A("strides",n,t,e),d=A("beginMask",n,t,e),p=A("endMask",n,t,e),f=A("ellipsisMask",n,t,e),m=A("newAxisMask",n,t,e),g=A("shrinkAxisMask",n,t,e),v=A("x",n,t,e);return[B.stridedSlice(v,l,c,h,d,p,f,m,g)]}case"Pack":return B.tidy(function(){var D=A("axis",n,t,e),k=A("tensors",n,t,e),W=k[0].shape,F=B.squeeze(k[0]).shape,P=k.map(function(H){var _=B.util.arraysEqual(H.shape,W);if(!_&&!B.util.arraysEqual(B.squeeze(H).shape,F))throw new Error("the input tensors shape does not match");return _?H:B.reshape(H,W)});return[B.stack(P,D)]});case"Unpack":{var r=A("axis",n,t,e),v=A("tensor",n,t,e);return B.unstack(v,r)}case"Tile":{var b=A("reps",n,t,e);return[B.tile(A("x",n,t,e),b)]}case"Split":case"SplitV":{var r=A("axis",n,t,e),w=A("numOrSizeSplits",n,t,e),v=A("x",n,t,e);return B.split(v,w,r)}case"ScatterNd":{var o=A("indices",n,t,e),S=A("values",n,t,e),L=A("shape",n,t,e);return[B.scatterND(o,S,L)]}case"GatherNd":{var x=A("x",n,t,e),o=A("indices",n,t,e);return[B.gatherND(x,o)]}case"SparseToDense":{var o=A("sparseIndices",n,t,e),L=A("outputShape",n,t,e),C=A("sparseValues",n,t,e),R=A("defaultValue",n,t,e);return[B.sparseToDense(o,C,L,C.dtype===R.dtype?R:B.cast(R,C.dtype))]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var JW=function(n,t,e){switch(n.op){case"FFT":return[B.fft(A("x",n,t,e))];case"IFFT":return[B.ifft(A("x",n,t,e))];case"RFFT":return[B.rfft(A("x",n,t,e))];case"IRFFT":return[B.irfft(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var ZW=function(n,t,e){switch(n.op){case"Cast":return[B.cast(A("x",n,t,e),A("dtype",n,t,e))];case"ExpandDims":{var i=A("axis",n,t,e);return[B.expandDims(A("x",n,t,e),i)]}case"Squeeze":{var i=A("axis",n,t,e);return[B.squeeze(A("x",n,t,e),i)]}case"Reshape":return[B.reshape(A("x",n,t,e),A("shape",n,t,e))];case"PadV2":case"Pad":return[B.pad(A("x",n,t,e),A("padding",n,t,e),A("constantValue",n,t,e))];case"SpaceToBatchND":{var r=A("blockShape",n,t,e),a=A("paddings",n,t,e);return[B.spaceToBatchND(A("x",n,t,e),r,a)]}case"BatchToSpaceND":{var r=A("blockShape",n,t,e),s=A("crops",n,t,e);return[B.batchToSpaceND(A("x",n,t,e),r,s)]}case"DepthToSpace":{var o=A("blockSize",n,t,e),l=A("dataFormat",n,t,e).toUpperCase();return[B.depthToSpace(A("x",n,t,e),o,l)]}case"BroadcastTo":return[B.broadcastTo(A("x",n,t,e),A("shape",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function tw(n,t,e){var i=function(r,a,s){switch(r.category){case"arithmetic":return B.tidy(function(){return RW(r,a,s)});case"basic_math":return B.tidy(function(){return OW(r,a,s)});case"control":return zW(r,a,s);case"convolution":return B.tidy(function(){return PW(r,a,s)});case"creation":return B.tidy(function(){return _W(r,a,s)});case"dynamic":return HW(r,a,s);case"evaluation":return B.tidy(function(){return VW(r,a,s)});case"image":return B.tidy(function(){return GW(r,a,s)});case"graph":return B.tidy(function(){return qW(r,a,s)});case"logical":return B.tidy(function(){return YW(r,a,s)});case"matrices":return B.tidy(function(){return KW(r,a,s)});case"normalization":return B.tidy(function(){return jW(r,a,s)});case"reduction":return B.tidy(function(){return $W(r,a,s)});case"slice_join":return B.tidy(function(){return XW(r,a,s)});case"spectral":return B.tidy(function(){return JW(r,a,s)});case"transformation":return B.tidy(function(){return ZW(r,a,s)});case"custom":var o=$b(r.op);if(o&&o.customExecutor)return o.customExecutor(new CW(r,a,s));throw TypeError("Custom op "+r.op+" is not registered.");default:throw TypeError("Unknown op '"+r.op+"'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()")}}(n,t,e);return i instanceof Promise?i.then(function(r){return[].concat(r)}):[].concat(i)}var nw=function(){function n(t,e,i,r){t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i={}),r===void 0&&(r={}),this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=i,this.functionMap=r,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}return n.prototype.newFrame=function(t,e){return{id:t,frameName:e,iterationId:0}},Object.defineProperty(n.prototype,"currentContext",{get:function(){return this.contexts},set:function(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextId",{get:function(){return this._currentContextIds[0]},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextIds",{get:function(){return this._currentContextIds},enumerable:!0,configurable:!0}),n.prototype.generateCurrentContextIds=function(){for(var t=[],e=0;e<this.contexts.length-1;e++){var i=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(i))}t.push(""),this._currentContextIds=t},n.prototype.contextIdforContexts=function(t){return t?t.map(function(e){return e.id===0&&e.iterationId===0?"":e.frameName+"-"+e.iterationId}).join("/"):""},n.prototype.enterFrame=function(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))},n.prototype.exitFrame=function(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function rw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return $t(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return $t(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((iw(d)||QW(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function eU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return $t(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var tU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],nU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function iw(n){return tU.indexOf(n.op)>=0}function QW(n){return nU.indexOf(n.op)>=0}var aw=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=rw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return eU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[$t(d)[0]]}),s=e.map(function(d){return $t(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new nw(i.weightMap,c,h,i.functionExecutorMap),p=Kb({},i.weightMap);Object.keys(t).forEach(function(w){var S=$t(w),L=S[0],x=S[1],C=[];C[x]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g<u.length;g++){var v=u[g];if(!p[v.name]){var b=tw(v,p,d);if(b instanceof Promise)throw new Error("The execution of the op '"+v.op+"' returned a promise. Please use model.executeAsync() instead.");p[v.name]=b,i.checkTensorForDisposal(v.name,v,p,d,f,s,m)}}return i.parent==null&&d.dispose(f),e.map(function(w){return Ht(w,p,d)})})},n.prototype.getFrozenTensorIds=function(t){var e=[].concat.apply([],Object.keys(t).map(function(i){return t[i]}).map(function(i){return i.map(function(r){return r.id})}));return new Set(e)},n.prototype.checkTensorForDisposal=function(t,e,i,r,a,s,o){if(e.category==="control"||s.indexOf(t)!==-1)return;i[t].forEach(function(l){l!=null&&(o[l.id]=(o[l.id]||0)+e.children.length)}),e.inputs.forEach(function(l){if(l.category!=="control"){var u=jF(l.name,i,r);u!=null&&u.forEach(function(c){if(c&&!a.has(c.id)){var h=o[c.id];h===1?(c.dispose(),delete o[c.id]):h!=null&&o[c.id]--}})}})},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){return Sn(this,function(i){return[2,this._executeAsync(t,e)]})})},n.prototype._executeAsync=function(t,e,i,r,a){return i===void 0&&(i=!1),r===void 0&&(r={}),a===void 0&&(a={}),zn(this,void 0,void 0,function(){var s,o,l,u,c,h;return Sn(this,function(d){switch(d.label){case 0:return i||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e)),s=new nw(this.weightMap,r,a,this.functionExecutorMap),[4,this.executeWithControlFlow(t,s,e,i)];case 1:return o=d.sent(),l=e.map(function(p){return Ht(p,o,s)}),u=l.map(function(p){return p.id}),c=Object.keys(t).map(function(p){return t[p].id}),h=new Set(u.concat(c,this.weightIds)),Object.keys(o).forEach(function(p){var f=o[p];f.forEach(function(m){m&&!m.isDisposed&&!h.has(m.id)&&m.dispose()})}),this.parent==null&&s.dispose(h),[2,l]}})})},n.prototype.executeFunctionAsync=function(t,e,i){return zn(this,void 0,void 0,function(){var r,a=this;return Sn(this,function(s){return r=t.reduce(function(o,l,u){return o[a.inputs[u].name]=l,o},{}),[2,this._executeAsync(r,this.outputNodes,!0,e,i)]})})},n.prototype.executeWithControlFlow=function(t,e,i,r){return zn(this,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x=this;return Sn(this,function(C){switch(C.label){case 0:a=Object.keys(t),s=a.map(function(R){return x.graph.nodes[$t(R)[0]]}),o=i.map(function(R){return $t(R)[0]}),l=o.map(function(R){return x.graph.nodes[R]}),u=rw(t,l,this.weightMap),c=u.usedNodes,h=u.missingInputs,d=u.dynamicNode,p=u.syncInputs,f=s.concat(this.graph.weights).map(function(R){return{node:R,contexts:e.currentContext}}),m=Kb({},this.weightMap),Object.keys(t).forEach(function(R){var D=$t(R),k=D[0],W=D[1],F=[];F[W]=t[R],m[k]=F}),g={},v=this.getFrozenTensorIds(m),b={},C.label=1;case 1:return f.length>0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!iw(R)&&!Ht(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=Zn(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=tw(f.node,r,i);m||(m=Zn(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=Zn(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=$t(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=$t(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=$t(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var iU="?tfjs-format=file",rU="model.json",sw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return zn(this,void 0,void 0,function(){var t;return Sn(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new aw(Jb.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Jb.Instance.transformGraph(t.modelInitializer);this.initializer=new aw(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return zn(this,void 0,void 0,function(){var i;return Sn(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){var i;return Sn(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function aU(n,t){return t===void 0&&(t={}),zn(this,void 0,void 0,function(){var e;return Sn(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+rU+iU)),e=new sw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var sU="2.6.0";rr.GraphModel=sw;rr.deregisterOp=KF;rr.loadGraphModel=aU;rr.registerOp=YF;rr.version_converter=sU});var lw=be(()=>{});var Cw=be(Vt=>{"use strict";Object.defineProperty(Vt,"__esModule",{value:!0});var Ne=Xi();var ud=function(n,t){return ud=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},ud(n,t)};function Ve(n,t){ud(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var jr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function ar(n,t){return t={exports:{}},n(t,t.exports),t.exports}var oU=ar(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(jr,n,!1)}),lU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(jr,n,!1)}),uU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(jr,n,!1)}),cU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(jr,n,!1)}),hU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(jr,n,!1)}),dU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(jr,n,!1)}),sr=ar(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});sr.alea=oU;sr.xor128=lU;sr.xorwow=uU;sr.xorshift7=cU;sr.xor4096=hU;sr.tychei=dU;var pU=sr,uw=pU.alea;function fU(n,t){return xo(n,t)}function xo(n,t,e,i){if(e===void 0&&(e=new Map),i===void 0&&(i=new Set),n==null)return null;if(i.has(n))throw new Error("Circular references are not supported.");if(e.has(n))return e.get(n);var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if($r(n)){var a=Array.isArray(n)?[]:{};i.add(n);for(var s in n){var o=n[s],l=xo(o,t,e,i);a[s]=l}return i.delete(n),a}else throw new Error("Can't recurse into non-iterable type: "+n);else return e.set(n,r.value),r.value}function mU(n,t){return t===void 0&&(t=hw),cw(n,t)}function cw(n,t,e){e===void 0&&(e=new Set);var i=n[0];if(e.has(i))throw new Error("Circular references are not supported.");var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if($r(i)){var a=Array.isArray(i)?[]:{};e.add(i);var s=function(l){var u=n.map(function(h){return h[l]}),c=cw(u,t,e);a[l]=c};for(var o in i)s(o);return e.delete(i),a}else throw new Error("Can't recurse into non-iterable type: "+i);else return r.value}function hw(n){return n===null?null:$r(n[0])?{value:null,recurse:!0}:{value:n,recurse:!1}}function dw(n,t){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:e=new Map,xo(n,t,e),i=0,r=Array.from(e.keys()),u.label=1;case 1:return i<r.length?(a=r[i],s=e.get(a),s instanceof Promise?[4,s]:[3,3]):[3,4];case 2:o=u.sent(),e.set(a,o),u.label=3;case 3:return i++,[3,1];case 4:return l=xo(n,t,e),[2,l]}})})}function $r(n){return n!=null&&!ArrayBuffer.isView(n)&&(Array.isArray(n)||typeof n=="object"&&!(n instanceof Ne.Tensor))}function vU(n){return n==null||gU(n)||Array.isArray(n)||typeof n=="object"&&n instanceof Ne.Tensor||Ne.util.isTypedArray(n)}function gU(n){return n===null||typeof n!="object"&&typeof n!="function"}function bU(n){return fU(n,yU)}function yU(n){return n instanceof Ne.Tensor?{value:n.clone(),recurse:!1}:$r(n)?{value:null,recurse:!0}:{value:n,recurse:!1}}var pw=function(){function n(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}return n.prototype.wrap=function(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity},n.prototype.get=function(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]},n.prototype.set=function(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e},n.prototype.length=function(){var t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t},n.prototype.isFull=function(){return this.length()===this.capacity},n.prototype.isEmpty=function(){return this.length()===0},n.prototype.push=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)},n.prototype.pushAll=function(t){for(var e=0,i=t;e<i.length;e++){var r=i[e];this.push(r)}},n.prototype.pop=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);var t=this.get(this.end);return this.set(this.end,void 0),t},n.prototype.unshift=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)},n.prototype.shift=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t},n.prototype.shuffleExcise=function(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var e=this.wrap(this.begin+t),i=this.get(e);return this.set(e,this.pop()),i},n}();var wU=function(n){Ve(t,n);function t(){return n.call(this,t.INITIAL_CAPACITY)||this}return t.prototype.isFull=function(){return!1},t.prototype.push=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.push.call(this,e)},t.prototype.unshift=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.unshift.call(this,e)},t.prototype.expand=function(){for(var e=this.capacity*2,i=new Array(e),r=this.length(),a=0;a<r;a++)i[a]=this.get(this.wrap(this.begin+a));this.data=i,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=r},t.INITIAL_CAPACITY=32,t}(pw);function fw(n){return new SU(n)}function cd(n){return new LU(n)}function IU(n,t){return new mw(n,t)}function TU(n,t){return t===void 0&&(t=Ii.FAIL),new AU(n,t)}var Nt=function(){function n(){}return n.prototype.toArray=function(){return se(this,void 0,void 0,function(){var t,e;return oe(this,function(i){switch(i.label){case 0:return t=[],[4,this.next()];case 1:e=i.sent(),i.label=2;case 2:return e.done?[3,4]:(t.push(e.value),[4,this.next()]);case 3:return e=i.sent(),[3,2];case 4:return[2,t]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){var t,e,i;return oe(this,function(r){switch(r.label){case 0:return t=this.prefetch(100),e=[],[4,t.next()];case 1:i=r.sent(),r.label=2;case 2:return i.done?[3,4]:(e.push(i.value),[4,t.next()]);case 3:return i=r.sent(),[3,2];case 4:return[2,e]}})})},n.prototype.resolveFully=function(){return se(this,void 0,void 0,function(){var t;return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:t=e.sent(),e.label=2;case 2:return t.done?[3,4]:[4,this.next()];case 3:return t=e.sent(),[3,2];case 4:return[2]}})})},n.prototype.resolveWhile=function(t){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:return[4,this.next()];case 1:e=r.sent(),i=t(e.value),r.label=2;case 2:return!e.done&&i?[4,this.next()]:[3,4];case 3:return e=r.sent(),i=t(e.value),[3,2];case 4:return[2]}})})},n.prototype.handleErrors=function(t){return new DU(this,t)},n.prototype.filter=function(t){return new OU(this,t)},n.prototype.map=function(t){return new EU(this,t)},n.prototype.mapAsync=function(t){return new gw(this,t)},n.prototype.serialMapAsync=function(t){return new gw(this,t).serial()},n.prototype.flatmap=function(t){return new kU(this,t)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.map(t).resolveFully()]})})},n.prototype.serialForEach=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.serialMapAsync(t).resolveWhile(function(i){return i===!0})]})})},n.prototype.rowMajorBatch=function(t,e){return e===void 0&&(e=!0),new RU(this,t,e)},n.prototype.columnMajorBatch=function(t,e,i){e===void 0&&(e=!0),i===void 0&&(i=hw);var r=this.rowMajorBatch(t,e);return r.map(function(a){return mU(a,i)})},n.prototype.concatenate=function(t,e){return new mw(fw([this,t]),e)},n.prototype.take=function(t){return t<0||t==null?this:new CU(this,t)},n.prototype.skip=function(t){return t<0||t==null?this:new xU(this,t)},n.prototype.prefetch=function(t){return new vw(this,t)},n.prototype.shuffle=function(t,e){return new FU(this,t,e)},n.prototype.serial=function(){return new NU(this)},n}(),SU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.items=e,i.trav=0,i}return t.prototype.summary=function(){return"Array of "+this.items.length+" items"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return this.trav>=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:bU(e),done:!1}])})})},t}(Nt),LU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(Nt),NU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(Nt),xU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++<this.maxCount?[4,this.upstream.next()]:[3,2];case 1:return e=i.sent(),e.done?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2,this.upstream.next()]}})})},t}(Nt),CU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(Nt),RU=function(n){Ve(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length<this.batchSize?[4,this.upstream.next()]:[3,3];case 2:return i=r.sent(),i.done?this.enableSmallLastBatch&&e.length>0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(Nt),OU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(Nt),EU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),DU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.handler=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(Nt),gw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),hd=function(n){Ve(t,n);function t(){var e=n.call(this)||this;return e.outputQueue=new wU,e.lastRead=Promise.resolve({value:null,done:!1}),e}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.outputQueue.length()===0?[4,this.pump()]:[3,2];case 1:return e.sent()?[3,0]:[2,{value:null,done:!0}];case 2:return[2,{value:this.outputQueue.shift(),done:!1}]}})})},t}(Nt),kU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,!0]}})})},t}(hd),mw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.baseErrorHandler=i,r.lastRead=null,r.iterator=null,r.moreIterators=e,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of chained summaries";return e+" -> Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(Nt),Ii;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Ii||(Ii={}));var AU=function(n){Ve(t,n);function t(e,i){i===void 0&&(i=Ii.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof Nt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,dw(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Ii.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Ii.SHORTEST:return[2,{value:null,done:!0}];case Ii.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(Nt),vw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new pw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(Nt),FU=function(n){Ve(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=uw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(vw);var Ya=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is
-      `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),Xt(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,WU)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,Xt(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,Xt(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=cd(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,IU(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?r=0:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().skip(t)]}})})},r)},n.prototype.shuffle=function(t,e,i){var r=this;if(i===void 0&&(i=!0),t==null||t<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError("`Dataset.shuffle()` requires bufferSize to be specified.  If your data fits in main memory (for regular JS objects), and/or GPU memory (for `tf.Tensor`s), consider setting "+("bufferSize to the dataset size ("+this.size+" elements)"));var a=this,s=uw(e||Ne.util.now().toString());return Xt(function(){return se(r,void 0,void 0,function(){var o;return oe(this,function(l){switch(l.label){case 0:return o=s.int32(),i&&(o+=s.int32()),[4,a.iterator()];case 1:return[2,l.sent().shuffle(t,o.toString())]}})})},this.size)},n.prototype.take=function(t){var e=this,i=this,r;return this.size!=null&&this.size>t?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function Xt(n,t){return t===void 0&&(t=null),new(function(e){Ve(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(Ya))}function UU(n){var t=this;return Xt(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,fw(n)]})})},n.length)}function BU(n){var t=this;if(!$r(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i<n.length;i++)e=e==null?n[i].size:Math.min(e,n[i].size);else if(n instanceof Object)for(var r in n)e=e==null?n[r].size:Math.min(e,n[r].size);return Xt(function(){return se(t,void 0,void 0,function(){var a;return oe(this,function(s){switch(s.label){case 0:return[4,dw(n,function(o){if(o instanceof Ya)return{value:o.iterator(),recurse:!1};if($r(o))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")})];case 1:return a=s.sent(),[2,TU(a,Ii.SHORTEST)]}})})},e)}function WU(n){if(n===null)return null;var t=n[0];if(vU(t)){var e=zU(n);return{value:e,recurse:!1}}return{value:null,recurse:!0}}function zU(n){if(n.length===0)throw new Error("Can't make a batch of zero elements.");return n[0]instanceof Ne.Tensor?Ne.stack(n):Ne.tensor(n)}var yw=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.input=e,i}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.input.iterator()];case 1:return e=a.sent(),i=e.decodeUTF8(),r=i.split(`
-`).map(function(s){return s.endsWith("\r")&&(s=s.slice(0,-1)),s}),[2,r]}})})},t}(Ya);var Co='"',Ka=Symbol("out"),bw=Symbol("field"),Ro=Symbol("quote"),dd=Symbol("quoteafterquote"),ww=Symbol("quoteinquote"),Sw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.input=e,r.hasHeader=!0,r.fullColumnNames=null,r.columnNamesValidated=!1,r.columnConfigs=null,r.configuredColumnsOnly=!1,r.delimiter=",",r.delimWhitespace=!1,r.base=new yw(e),i||(i={}),r.hasHeader=!(i.hasHeader===!1),r.fullColumnNames=i.columnNames,r.columnConfigs=i.columnConfigs,r.configuredColumnsOnly=i.configuredColumnsOnly,i.delimWhitespace?(Ne.util.assert(i.delimiter==null,function(){return"Delimiter should not be provided when delimWhitespace is true."}),r.delimWhitespace=!0,r.delimiter=" "):r.delimiter=i.delimiter?i.delimiter:",",r}return t.prototype.columnNames=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:e.sent(),e.label=2;case 2:return[2,this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames]}})})},t.prototype.setColumnNames=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u=this;return oe(this,function(c){switch(c.label){case 0:return[4,this.maybeReadHeaderLine()];case 1:if(e=c.sent(),!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");if(this.fullColumnNames&&e&&Ne.util.assert(e.length===this.fullColumnNames.length,function(){return"The length of provided columnNames ("+u.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."}),this.fullColumnNames||(this.fullColumnNames=e),i=this.fullColumnNames.reduce(function(h,d){return h[d]=h[d]+1||1,h},{}),r=Object.keys(i).filter(function(h){return i[h]>1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a<s.length;a++)if(o=s[a],l=this.fullColumnNames.indexOf(o),l===-1)throw new Error('The key "'+o+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}return this.columnNamesValidated=!0,[2]}})})},t.prototype.maybeReadHeaderLine=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return this.hasHeader?[4,this.base.iterator()]:[3,3];case 1:return e=s.sent(),[4,e.next()];case 2:if(i=s.sent(),i.done)throw new Error("No data was found for CSV parsing.");return r=i.value,a=this.parseRow(r,!1),[2,a];case 3:return[2,null]}})})},t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i=this;return oe(this,function(r){switch(r.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:r.sent(),r.label=2;case 2:return[4,this.base.iterator()];case 3:return e=r.sent(),this.hasHeader&&(e=e.skip(1)),[2,e.map(function(a){return i.makeDataElement(a)})]}})})},t.prototype.makeDataElement=function(e){for(var i=this.parseRow(e),r={},a={},s=0;s<this.fullColumnNames.length;s++){var o=this.fullColumnNames[s],l=this.columnConfigs?this.columnConfigs[o]:null;if(this.configuredColumnsOnly&&!l)continue;var u=i[s],c=null;if(u==="")if(l&&l.default!==void 0)c=l.default;else{if(l&&(l.required||l.isLabel))throw new Error("Required column "+o+" is empty in this line: "+e);c=void 0}else{var h=Number(u);if(isNaN(h))l&&l.dtype==="bool"?c=this.getBoolean(u):c=u;else if(!l||!l.dtype)c=h;else switch(l.dtype){case"float32":c=h;break;case"int32":c=Math.floor(h);break;case"bool":c=this.getBoolean(u);break;default:c=h}}l&&l.isLabel?a[o]=c:r[o]=c}return Object.keys(a).length===0?r:{xs:r,ys:a}},t.prototype.getBoolean=function(e){return e==="1"||e.toLowerCase()==="true"?1:0},t.prototype.parseRow=function(e,i){i===void 0&&(i=!0);for(var r=[],a=0,s=e.length,o=Ka,l=0;l<s;l++)switch(o){case Ka:switch(e.charAt(l)){case Co:a=l+1,o=Ro;break;case this.delimiter:if(a=l+1,this.delimiter===" "&&this.delimWhitespace)break;r.push(""),o=Ka;break;default:o=bw,a=l;break}break;case bw:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l)),o=Ka,a=l+1;break}break;case Ro:switch(e.charAt(l)){case Co:o=dd;break}break;case dd:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l-1)),o=Ka,a=l+1;break;case Co:o=Ro;break;default:o=ww;break}break;case ww:switch(e.charAt(l)){case Co:o=Ro;break}break}if(o===dd?r.push(e.substring(a,s-1)):r.push(e.substring(a)),i&&r.length!==this.fullColumnNames.length)throw new Error("Invalid row in csv file. Should have "+this.fullColumnNames.length+" elements in a row, but got "+r);return r},t}(Ya);var PU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;i.microphoneConfig=e,i.isClosed=!1,i.fftSize=e.fftSize||1024;var r=Math.log2(i.fftSize);if(i.fftSize<0||r<4||r>14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(Nt);var _U=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(Nt);var Lw=function(){function n(){}return n}();var Iw=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new MU(this,e)},t}(Nt),MU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new HU(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),HU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r<a.length;r++)s=a[r],this.outputQueue.push(s);return this.carryover=i[i.length-1],[2,!0]}})})},t}(hd);var qU=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.decodeUTF8=function(){return new VU(this)},t}(Nt),VU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.impl=new GU(e),i}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),GU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;if(i.upstream=e,Ne.env().get("IS_BROWSER"))i.decoder=new TextDecoder("utf-8");else{var r=lw().StringDecoder;i.decoder=new r("utf8")}return i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(hd);var Aw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(qU);function KU(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=YU(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new Aw(a,t)];case 3:throw new Error(r.statusText)}})})}var YU=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function Tw(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var Nw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return Tw(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new Aw(this.input,this.options)]})})},t}(Lw);var xw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return Tw(this.url)?[2,new Nw(this.url,this.fileOptions).iterator()]:[2,KU(this.url,this.fileOptions)]})})},t}(Lw);function jU(n,t){return t===void 0&&(t={}),new Sw(new xw(n),t)}function $U(n){var t=this,e=cd(n);return Xt(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function XU(n){var t=this;return Xt(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,cd(function(){return e.next()})]}})})})}function JU(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,_U.create(n,t)]})})}function ZU(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,PU.create(n)]})})}var QU="2.6.0";Vt.CSVDataset=Sw;Vt.Dataset=Ya;Vt.FileDataSource=Nw;Vt.TextLineDataset=yw;Vt.URLDataSource=xw;Vt.array=UU;Vt.csv=jU;Vt.func=$U;Vt.generator=XU;Vt.microphone=ZU;Vt.version_data=QU;Vt.webcam=JU;Vt.zip=BU});var Ow=be((Rw,pd)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c<u.length;c++){o+=u.charCodeAt(c);var h=.02519603282416938*o;o=h>>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(Rw,typeof pd=="object"&&pd,typeof define=="function"&&define)});var Dw=be((Ew,fd)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(Ew,typeof fd=="object"&&fd,typeof define=="function"&&define)});var Fw=be((kw,md)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,u==l.length&&(o.d=o.x<<10^o.x>>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(kw,typeof md=="object"&&md,typeof define=="function"&&define)});var Uw=be((Ww,gd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h<c.length;++h)p[h&7]=p[h&7]<<15^c.charCodeAt(h)+p[h+1&7]<<13;for(;p.length<8;)p.push(0);for(h=0;h<8&&p[h]===0;++h);for(h==8?d=p[7]=-1:d=p[h],u.x=p,u.i=0,h=256;h>0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(Ww,typeof gd=="object"&&gd,typeof define=="function"&&define)});var zw=be((Bw,vd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f<v;++f)c&&(d^=c.charCodeAt((f+32)%c.length)),f===0&&(m=d),d^=d<<10,d^=d>>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(Bw,typeof vd=="object"&&vd,typeof define=="function"&&define)});var _w=be((Pw,yd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u<l.length+20;u++)o.b^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.a=s.a,o.b=s.b,o.c=s.c,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(Pw,typeof yd=="object"&&yd,typeof define=="function"&&define)});var Mw=be((CH,Oo)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var x=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),x),R=new p(x),D=function(){for(var k=R.g(r),W=o,F=0;k<l;)k=(k+F)*i,W*=i,F=R.g(1);for(;k>=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,x=this,C=0,R=x.i=x.j=0,D=x.S=[];for(L||(w=[L++]);C<i;)D[C]=C++;for(C=0;C<i;C++)D[C]=D[R=c&R+w[C%L]+(S=D[C])],D[R]=S;(x.g=function(k){for(var W,F=0,P=x.i,H=x.j,_=x.S;k--;)W=_[P=c&P+1],F=F*i+_[c&(_[P]=_[H=c&H+W])+(_[H]=W)];return x.i=P,x.j=H,F})(i)}function f(w,S){return S.i=w.i,S.j=w.j,S.S=w.S.slice(),S}function m(w,S){var L=[],x=typeof w,C;if(S&&x=="object")for(C in w)try{L.push(m(w[C],S-1))}catch(R){}return L.length?L:x=="string"?w:w+"\0"}function g(w,S){for(var L=w+"",x,C=0;C<L.length;)S[c&C]=c&(x^=S[c&C]*19)+L.charCodeAt(C++);return b(S)}function v(){try{var w;return h&&(w=h.randomBytes)?w=w(i):(w=new Uint8Array(i),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(x){var S=e.navigator,L=S&&S.plugins;return[+new Date,e,L,e.screen,b(n)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),n),typeof Oo=="object"&&Oo.exports){Oo.exports=d;try{h=ds()}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return d})})([],Math)});var Vw=be((RH,Hw)=>{var eB=Ow(),tB=Dw(),nB=Fw(),iB=Uw(),rB=zw(),aB=_w(),or=Mw();or.alea=eB;or.xor128=tB;or.xorwow=nB;or.xorshift7=iB;or.xor4096=rB;or.tychei=aB;Hw.exports=or});var m0=be(ja=>{"use strict";Object.defineProperty(ja,"__esModule",{value:!0});var T=Xi(),sB=Vw();var bd=function(n,t){return bd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},bd(n,t)};function oB(n,t){bd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function qw(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function Gw(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}function ae(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&T.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the CPU backend."})})}var lB=T.kernel_impls.nonMaxSuppressionV3Impl,uB=T.kernel_impls.split,cB=T.kernel_impls.tile,hB=T.kernel_impls.topkImpl,dB=T.kernel_impls.whereImpl;function wd(n,t,e,i){if(e==="linear")return n.linear(t);if(e==="relu")return n.relu(t);if(e==="elu")return T.elu(t);if(e==="relu6")return n.relu6(t);if(e==="prelu")return n.prelu(t,i);throw new Error("Activation "+e+" has not been implemented for the CPU backend.")}var Yw=function(n){oB(t,n);function t(){var e=n.call(this)||this;return e.blockSize=48,e.firstUse=!0,e.data=new T.DataStorage(e,T.engine()),e}return t.prototype.write=function(e,i,r){this.firstUse&&(this.firstUse=!1,T.env().get("IS_NODE")&&T.backend_util.warn(`
+        `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c<t.length;++c){var h=c===0?0:r[c-1],d=[0,h,0],p=[1,t[c],a];u[c]=B.reshape(B.slice(n,d,p),e)}return n.dispose(),u}),o=new No([],e,n.dtype,t.length),l=0;l<s.length;l++)o.setItem(l,s[l]);return o}var BW=void 0,zW=function(n,t,e){return zn(BW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,o,h,d,l,p,f,m,m,g,v,g,b,g,g,g,w,S,L,x,C,R,D,k,W,F,P,H,_,K,j,q,G,Z,X,ee,ne,ie,te,re,le,he,ye,Oe,Fe,_e,Me,ht,St,Ge,Ue,F,P,me,Ue,K,L,Dt,me,ne,ie,L,lt,me,L,dt,lt,me,q,G,L,dt,me,Ue,L,dt,lt,me,xn,L,dt,me,re,me,he,L,Ue,P,me,Ue,L,Dt,me,Oe,L,Fe,me;return wn(this,function(Rt){switch(Rt.label){case 0:i=n.op;switch(i){case"If":return[3,1];case"StatelessIf":return[3,1];case"While":return[3,3];case"StatelessWhile":return[3,3];case"LoopCond":return[3,9];case"Switch":return[3,10];case"Merge":return[3,12];case"Enter":return[3,13];case"Exit":return[3,14];case"NextIteration":return[3,15];case"TensorArrayV3":return[3,16];case"TensorArrayWriteV3":return[3,17];case"TensorArrayReadV3":return[3,18];case"TensorArrayGatherV3":return[3,19];case"TensorArrayScatterV3":return[3,20];case"TensorArrayConcatV3":return[3,21];case"TensorArraySplitV3":return[3,22];case"TensorArraySizeV3":return[3,23];case"TensorArrayCloseV3":return[3,24];case"TensorListSetItem":return[3,25];case"TensorListGetItem":return[3,26];case"TensorListScatterV2":return[3,27];case"TensorListScatter":return[3,27];case"TensorListReserve":return[3,28];case"TensorListGather":return[3,29];case"TensorListStack":return[3,30];case"TensorListFromTensor":return[3,31];case"TensorListConcat":return[3,32];case"TensorListPushBack":return[3,33];case"TensorListPopBack":return[3,34];case"TensorListSplit":return[3,35]}return[3,36];case 1:return r=A("thenBranch",n,t,e),a=A("elseBranch",n,t,e),s=A("cond",n,t,e),o=A("args",n,t,e),[4,s.data()];case 2:return l=Rt.sent(),l[0]?[2,e.functionMap[r].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)]:[2,e.functionMap[a].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 3:return u=A("body",n,t,e),c=A("cond",n,t,e),o=A("args",n,t,e),[4,e.functionMap[c].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 4:return h=Rt.sent(),d=o.map(function(Lt){return Lt.id}),[4,h[0].data()];case 5:l=Rt.sent(),h.forEach(function(Lt){!Lt.kept&&d.indexOf(Lt.id)===-1&&Lt.dispose()}),p=o,f=function(){var Lt,Ce,kt;return wn(this,function(pt){switch(pt.label){case 0:return Lt=p,[4,e.functionMap[u].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 1:return p=pt.sent(),Ce=p.map(function(it){return it.id}),Lt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[4,e.functionMap[c].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 2:return kt=pt.sent(),[4,kt[0].data()];case 3:return l=pt.sent(),kt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[2]}})},Rt.label=6;case 6:return l[0]?[5,f()]:[3,8];case 7:return Rt.sent(),[3,6];case 8:return[2,p];case 9:return m=A("pred",n,t,e),[2,[Qn(m)]];case 10:return m=A("pred",n,t,e),g=A("data",n,t,e),g.kept||(g=Qn(g)),[4,m.data()];case 11:return[2,Rt.sent()[0]?[void 0,g]:[g,void 0]];case 12:return v=n.inputNames.find(function(Lt){return Ht(Lt,t,e)!==void 0}),v?(g=Ht(v,t,e),[2,[Qn(g)]]):[2,void 0];case 13:return b=A("frameName",n,t,e),g=A("tensor",n,t,e),e.enterFrame(b),[2,[Qn(g)]];case 14:return g=A("tensor",n,t,e),e.exitFrame(),[2,[Qn(g)]];case 15:return g=A("tensor",n,t,e),e.nextIteration(),[2,[Qn(g)]];case 16:return w=A("size",n,t,e),S=A("dtype",n,t,e),L=A("elementShape",n,t,e),x=A("dynamicSize",n,t,e),C=A("clearAfterRead",n,t,e),R=A("identicalElementShapes",n,t,e),D=A("name",n,t,e),k=new DW(D,S,w,L,R,x,C),e.addTensorArray(k),[2,[k.idTensor,B.scalar(1)]];case 17:return W=A("tensorArrayId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),H=e.getTensorArray(W.id),H.write(F,P),[2,[H.idTensor]];case 18:return _=A("tensorArrayId",n,t,e),K=A("index",n,t,e),j=e.getTensorArray(_.id),[2,[j.read(K)]];case 19:return q=A("tensorArrayId",n,t,e),G=A("indices",n,t,e),Z=A("dtype",n,t,e),X=e.getTensorArray(q.id),[2,[X.gather(G,Z)]];case 20:return ee=A("tensorArrayId",n,t,e),ne=A("indices",n,t,e),ie=A("tensor",n,t,e),te=e.getTensorArray(ee.id),te.scatter(ne,ie),[2,[te.idTensor]];case 21:return re=A("tensorArrayId",n,t,e),le=e.getTensorArray(re.id),he=A("dtype",n,t,e),[2,[le.concat(he)]];case 22:return ye=A("tensorArrayId",n,t,e),Oe=A("tensor",n,t,e),Fe=A("lengths",n,t,e),_e=e.getTensorArray(ye.id),_e.split(Fe,Oe),[2,[_e.idTensor]];case 23:return Me=A("tensorArrayId",n,t,e),ht=e.getTensorArray(Me.id),[2,[B.scalar(ht.size(),"int32")]];case 24:return St=A("tensorArrayId",n,t,e),Ge=e.getTensorArray(St.id),Ge.clearAndClose(),[2,[Ge.idTensor]];case 25:return Ue=A("tensorListId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.setItem(F,P),[2,[me.idTensor]];case 26:return Ue=A("tensorListId",n,t,e),K=A("index",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.getItem(K,L,Dt)]];case 27:return ne=A("indices",n,t,e),ie=A("tensor",n,t,e),L=A("elementShape",n,t,e),lt=A("numElements",n,t,e),me=WW(ie,ne,L,lt),e.addTensorList(me),[2,[me.idTensor]];case 28:return L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=FW(L,dt,lt),e.addTensorList(me),[2,[me.idTensor]];case 29:return q=A("tensorListId",n,t,e),G=A("indices",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=e.getTensorList(q.id),[2,[me.gather(G,dt,L)]];case 30:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=e.getTensorList(Ue.id),[2,[me.stack(L,dt,lt)]];case 31:return xn=A("tensor",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=kW(xn,L,dt),e.addTensorList(me),[2,[me.idTensor]];case 32:return re=A("tensorListId",n,t,e),me=e.getTensorList(re.id),he=A("dtype",n,t,e),L=A("elementShape",n,t,e),[2,[me.concat(he,L)]];case 33:return Ue=A("tensorListId",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.pushBack(P),[2,[me.idTensor]];case 34:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.popBack(L,Dt)]];case 35:return Oe=A("tensor",n,t,e),L=A("elementShape",n,t,e),Fe=A("lengths",n,t,e),me=UW(Oe,Fe,L),e.addTensorList(me),[2,[me.idTensor]];case 36:throw TypeError("Node type "+n.op+" is not implemented")}})})};function ew(n,t,e){var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=r==="fusedbatchnorm",u=A("numArgs",n,t,e);if(s){if(o&&u!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&u!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");var c=A("strides",n,t,e),h=To(n,t,e),d=A("dataFormat",n,t,e).toUpperCase(),p=A("dilations",n,t,e),f=A("args",n,t,e),m=f[0],g=f[1];return{stride:c,pad:h,dataFormat:d,dilations:p,biasArg:m,preluArg:g,activationFunc:a}}var PW=function(n,t,e){switch(n.op){case"Conv1D":{var i=A("stride",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),s=A("dilation",n,t,e);return[B.conv1d(A("x",n,t,e),A("filter",n,t,e),i,r,a,s)]}case"Conv2D":{var i=A("strides",n,t,e),r=To(n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv2d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{var l=ew(n,t,e),i=l.stride,r=l.pad,a=l.dataFormat,o=l.dilations,u=l.biasArg,c=l.preluArg,h=l.activationFunc;return[B.fused.conv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"FusedDepthwiseConv2dNative":{var d=ew(n,t,e),i=d.stride,r=d.pad,a=d.dataFormat,o=d.dilations,u=d.biasArg,c=d.preluArg,h=d.activationFunc;return[B.fused.depthwiseConv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{var p=A("outputShape",n,t,e),i=A("strides",n,t,e),r=To(n,t,e);return[B.conv2dTranspose(A("x",n,t,e),A("filter",n,t,e),p,[i[1],i[2]],r)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{var i=A("strides",n,t,e),r=To(n,t,e),o=A("dilations",n,t,e),a=A("dataFormat",n,t,e).toUpperCase();return[B.depthwiseConv2d(A("input",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"Conv3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv3d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2],i[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPoolWithArgmax":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e),m=A("includeBatchInIndex",n,t,e),g=B.maxPoolWithArgmax(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r,m),v=g.result,b=g.indexes;return[v,b]}case"AvgPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"MaxPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"Dilation2D":{var w=A("strides",n,t,e),r=A("pad",n,t,e),o=A("dilations",n,t,e),S=w[1],L=w[2],x=o[1],C=o[2];return[B.dilation2d(A("x",n,t,e),A("filter",n,t,e),[S,L],r,[x,C],"NHWC")]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var _W=function(n,t,e){switch(n.op){case"Fill":{var i=A("shape",n,t,e),r=A("dtype",n,t,e),a=A("value",n,t,e);return[B.fill(i,a,r)]}case"LinSpace":{var s=A("start",n,t,e),o=A("stop",n,t,e),l=A("num",n,t,e);return[B.linspace(s,o,l)]}case"Multinomial":{var u=A("logits",n,t,e),c=A("numSamples",n,t,e),h=A("seed",n,t,e);return[B.multinomial(u,c,h)]}case"OneHot":{var d=A("indices",n,t,e),p=A("depth",n,t,e),f=A("onValue",n,t,e),m=A("offValue",n,t,e);return[B.oneHot(d,p,f,m)]}case"Ones":return[B.ones(A("shape",n,t,e),A("dtype",n,t,e))];case"OnesLike":return[B.onesLike(A("x",n,t,e))];case"RandomUniform":return[B.randomUniform(A("shape",n,t,e),A("minval",n,t,e),A("maxval",n,t,e),A("dtype",n,t,e))];case"Range":{var s=A("start",n,t,e),g=A("stop",n,t,e),v=A("step",n,t,e);return[B.range(s,g,v,A("dtype",n,t,e))]}case"TruncatedNormal":{var i=A("shape",n,t,e),b=A("mean",n,t,e),w=A("stdDev",n,t,e),h=A("seed",n,t,e);return[B.truncatedNormal(i,b,w,A("dtype",n,t,e),h)]}case"Zeros":return[B.zeros(A("shape",n,t,e),A("dtype",n,t,e))];case"ZerosLike":return[B.zerosLike(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var MW=void 0;function ud(n,t,e){var i=A("boxes",n,t,e),r=A("scores",n,t,e),a=A("maxOutputSize",n,t,e),s=A("iouThreshold",n,t,e),o=A("scoreThreshold",n,t,e),l=A("softNmsSigma",n,t,e);return{boxes:i,scores:r,maxOutputSize:a,iouThreshold:s,scoreThreshold:o,softNmsSigma:l}}var HW=function(n,t,e){return zn(MW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,a,s,o,l,u,p,h,f,a,s,o,l,u,m,h;return wn(this,function(g){switch(g.label){case 0:i=n.op;switch(i){case"NonMaxSuppressionV5":return[3,1];case"NonMaxSuppressionV4":return[3,3];case"NonMaxSuppressionV3":return[3,5];case"NonMaxSuppressionV2":return[3,5];case"Where":return[3,7];case"ListDiff":return[3,9]}return[3,10];case 1:return r=ud(n,t,e),a=r.boxes,s=r.scores,o=r.maxOutputSize,l=r.iouThreshold,u=r.scoreThreshold,c=r.softNmsSigma,[4,B.image.nonMaxSuppressionWithScoreAsync(a,s,o,l,u,c)];case 2:return h=g.sent(),[2,[h.selectedIndices,h.selectedScores]];case 3:return d=ud(n,t,e),a=d.boxes,s=d.scores,o=d.maxOutputSize,l=d.iouThreshold,u=d.scoreThreshold,p=A("padToMaxOutputSize",n,t,e),[4,B.image.nonMaxSuppressionPaddedAsync(a,s,o,l,u,p)];case 4:return h=g.sent(),[2,[h.selectedIndices,h.validOutputs]];case 5:return f=ud(n,t,e),a=f.boxes,s=f.scores,o=f.maxOutputSize,l=f.iouThreshold,u=f.scoreThreshold,[4,B.image.nonMaxSuppressionAsync(a,s,o,l,u)];case 6:return[2,[g.sent()]];case 7:return m=B.cast(A("condition",n,t,e),"bool"),[4,B.whereAsync(m)];case 8:return h=[g.sent()],m.dispose(),[2,h];case 9:return[2,B.setdiff1dAsync(A("x",n,t,e),A("y",n,t,e))];case 10:throw TypeError("Node type "+n.op+" is not implemented")}})})};var VW=function(n,t,e){switch(n.op){case"TopKV2":{var i=A("x",n,t,e),r=A("k",n,t,e),a=A("sorted",n,t,e),s=B.topk(i,r,a);return[s.values,s.indices]}case"Unique":{var i=A("x",n,t,e),s=B.unique(i);return[s.values,s.indices]}case"UniqueV2":{var i=A("x",n,t,e),o=A("axis",n,t,e),s=B.unique(i,o);return[s.values,s.indices]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var qW=function(n,t,e){switch(n.op){case"Const":return t[n.name];case"PlaceholderWithDefault":var i=A("default",n,t,e);return[Ht(n.name,t,e)||i];case"Placeholder":return[Ht(n.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{var r=A("x",n,t,e);return[Qn(r)]}case"IdentityN":return A("x",n,t,e).map(function(h){return Qn(h)});case"Snapshot":var a=A("x",n,t,e);return[Qn(a)];case"Shape":return[B.tensor1d(A("x",n,t,e).shape,"int32")];case"ShapeN":return A("x",n,t,e).map(function(h){return B.tensor1d(h.shape)});case"Size":return[B.scalar(A("x",n,t,e).size,"int32")];case"Rank":return[B.scalar(A("x",n,t,e).rank,"int32")];case"NoOp":return[B.scalar(1)];case"Print":var s=A("x",n,t,e),o=A("data",n,t,e),l=A("message",n,t,e),u=A("summarize",n,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(l);for(var c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,u));return[s];default:throw TypeError("Node type "+n.op+" is not implemented")}};var GW=function(n,t,e){switch(n.op){case"ResizeBilinear":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeBilinear(i,[r[0],r[1]],a)]}case"ResizeNearestNeighbor":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeNearestNeighbor(i,[r[0],r[1]],a)]}case"CropAndResize":{var s=A("image",n,t,e),o=A("boxes",n,t,e),l=A("boxInd",n,t,e),u=A("cropSize",n,t,e),c=A("method",n,t,e),h=A("extrapolationValue",n,t,e);return[B.image.cropAndResize(s,o,l,u,c,h)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var YW=function(n,t,e){switch(n.op){case"Equal":return[B.equal(A("a",n,t,e),A("b",n,t,e))];case"NotEqual":return[B.notEqual(A("a",n,t,e),A("b",n,t,e))];case"Greater":return[B.greater(A("a",n,t,e),A("b",n,t,e))];case"GreaterEqual":return[B.greaterEqual(A("a",n,t,e),A("b",n,t,e))];case"Less":return[B.less(A("a",n,t,e),A("b",n,t,e))];case"LessEqual":return[B.lessEqual(A("a",n,t,e),A("b",n,t,e))];case"LogicalAnd":return[B.logicalAnd(A("a",n,t,e),A("b",n,t,e))];case"LogicalNot":return[B.logicalNot(A("a",n,t,e))];case"LogicalOr":return[B.logicalOr(A("a",n,t,e),A("b",n,t,e))];case"Select":case"SelectV2":return[B.where(A("condition",n,t,e),A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var KW=function(n,t,e){switch(n.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[B.matMul(A("a",n,t,e),A("b",n,t,e),A("transposeA",n,t,e),A("transposeB",n,t,e))];case"Transpose":return[B.transpose(A("x",n,t,e),A("perm",n,t,e))];case"_FusedMatMul":var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=A("numArgs",n,t,e);if(s){if(o&&l!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&l!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}var u=A("args",n,t,e),c=u[0],h=u[1];return[B.fused.matMul({a:A("a",n,t,e),b:A("b",n,t,e),transposeA:A("transposeA",n,t,e),transposeB:A("transposeB",n,t,e),bias:c,activation:a,preluActivationWeights:h})];default:throw TypeError("Node type "+n.op+" is not implemented")}};var jW=function(n,t,e){switch(n.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"FusedBatchNormV3":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"LRN":return[B.localResponseNormalization(A("x",n,t,e),A("radius",n,t,e),A("bias",n,t,e),A("alpha",n,t,e),A("beta",n,t,e))];case"Softmax":return[B.softmax(A("x",n,t,e))];case"LogSoftmax":return[B.logSoftmax(A("x",n,t,e))];case"SparseToDense":return[B.sparseToDense(A("sparseIndices",n,t,e),A("outputShape",n,t,e),A("sparseValues",n,t,e),A("defaultValue",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var $W=function(n,t,e){switch(n.op){case"Max":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.max(A("x",n,t,e),i,r)]}case"Mean":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.mean(A("x",n,t,e),i,r)]}case"Min":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.min(A("x",n,t,e),i,r)]}case"Sum":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.sum(A("x",n,t,e),i,r)]}case"All":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.all(A("x",n,t,e),i,r)]}case"Any":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.any(A("x",n,t,e),i,r)]}case"ArgMax":{var i=A("axis",n,t,e);return[B.argMax(A("x",n,t,e),i)]}case"ArgMin":{var i=A("axis",n,t,e);return[B.argMin(A("x",n,t,e),i)]}case"Prod":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.prod(A("x",n,t,e),i,r)]}case"Cumsum":{var i=A("axis",n,t,e),a=A("exclusive",n,t,e),s=A("reverse",n,t,e);return[B.cumsum(A("x",n,t,e),i,a,s)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var XW=function(n,t,e){switch(n.op){case"ConcatV2":case"Concat":{var i=A("n",n,t,e),r=A("axis",n,t,e),a=A("tensors",n,t,e);return a=a.slice(0,i),[B.concat(a,r)]}case"GatherV2":case"Gather":{var r=A("axis",n,t,e),s=A("x",n,t,e),o=A("indices",n,t,e);return[B.gather(s,B.cast(o,"int32"),r)]}case"ReverseV2":case"Reverse":{var r=A("axis",n,t,e),s=A("x",n,t,e);return[B.reverse(s,r)]}case"Slice":{var l=A("begin",n,t,e),u=A("size",n,t,e);return[B.slice(A("x",n,t,e),l,u)]}case"StridedSlice":{var l=A("begin",n,t,e),c=A("end",n,t,e),h=A("strides",n,t,e),d=A("beginMask",n,t,e),p=A("endMask",n,t,e),f=A("ellipsisMask",n,t,e),m=A("newAxisMask",n,t,e),g=A("shrinkAxisMask",n,t,e),v=A("x",n,t,e);return[B.stridedSlice(v,l,c,h,d,p,f,m,g)]}case"Pack":return B.tidy(function(){var D=A("axis",n,t,e),k=A("tensors",n,t,e),W=k[0].shape,F=B.squeeze(k[0]).shape,P=k.map(function(H){var _=B.util.arraysEqual(H.shape,W);if(!_&&!B.util.arraysEqual(B.squeeze(H).shape,F))throw new Error("the input tensors shape does not match");return _?H:B.reshape(H,W)});return[B.stack(P,D)]});case"Unpack":{var r=A("axis",n,t,e),v=A("tensor",n,t,e);return B.unstack(v,r)}case"Tile":{var b=A("reps",n,t,e);return[B.tile(A("x",n,t,e),b)]}case"Split":case"SplitV":{var r=A("axis",n,t,e),w=A("numOrSizeSplits",n,t,e),v=A("x",n,t,e);return B.split(v,w,r)}case"ScatterNd":{var o=A("indices",n,t,e),S=A("values",n,t,e),L=A("shape",n,t,e);return[B.scatterND(o,S,L)]}case"GatherNd":{var x=A("x",n,t,e),o=A("indices",n,t,e);return[B.gatherND(x,o)]}case"SparseToDense":{var o=A("sparseIndices",n,t,e),L=A("outputShape",n,t,e),C=A("sparseValues",n,t,e),R=A("defaultValue",n,t,e);return[B.sparseToDense(o,C,L,C.dtype===R.dtype?R:B.cast(R,C.dtype))]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var JW=function(n,t,e){switch(n.op){case"FFT":return[B.fft(A("x",n,t,e))];case"IFFT":return[B.ifft(A("x",n,t,e))];case"RFFT":return[B.rfft(A("x",n,t,e))];case"IRFFT":return[B.irfft(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var ZW=function(n,t,e){switch(n.op){case"Cast":return[B.cast(A("x",n,t,e),A("dtype",n,t,e))];case"ExpandDims":{var i=A("axis",n,t,e);return[B.expandDims(A("x",n,t,e),i)]}case"Squeeze":{var i=A("axis",n,t,e);return[B.squeeze(A("x",n,t,e),i)]}case"Reshape":return[B.reshape(A("x",n,t,e),A("shape",n,t,e))];case"PadV2":case"Pad":return[B.pad(A("x",n,t,e),A("padding",n,t,e),A("constantValue",n,t,e))];case"SpaceToBatchND":{var r=A("blockShape",n,t,e),a=A("paddings",n,t,e);return[B.spaceToBatchND(A("x",n,t,e),r,a)]}case"BatchToSpaceND":{var r=A("blockShape",n,t,e),s=A("crops",n,t,e);return[B.batchToSpaceND(A("x",n,t,e),r,s)]}case"DepthToSpace":{var o=A("blockSize",n,t,e),l=A("dataFormat",n,t,e).toUpperCase();return[B.depthToSpace(A("x",n,t,e),o,l)]}case"BroadcastTo":return[B.broadcastTo(A("x",n,t,e),A("shape",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function tw(n,t,e){var i=function(r,a,s){switch(r.category){case"arithmetic":return B.tidy(function(){return RW(r,a,s)});case"basic_math":return B.tidy(function(){return OW(r,a,s)});case"control":return zW(r,a,s);case"convolution":return B.tidy(function(){return PW(r,a,s)});case"creation":return B.tidy(function(){return _W(r,a,s)});case"dynamic":return HW(r,a,s);case"evaluation":return B.tidy(function(){return VW(r,a,s)});case"image":return B.tidy(function(){return GW(r,a,s)});case"graph":return B.tidy(function(){return qW(r,a,s)});case"logical":return B.tidy(function(){return YW(r,a,s)});case"matrices":return B.tidy(function(){return KW(r,a,s)});case"normalization":return B.tidy(function(){return jW(r,a,s)});case"reduction":return B.tidy(function(){return $W(r,a,s)});case"slice_join":return B.tidy(function(){return XW(r,a,s)});case"spectral":return B.tidy(function(){return JW(r,a,s)});case"transformation":return B.tidy(function(){return ZW(r,a,s)});case"custom":var o=$b(r.op);if(o&&o.customExecutor)return o.customExecutor(new CW(r,a,s));throw TypeError("Custom op "+r.op+" is not registered.");default:throw TypeError("Unknown op '"+r.op+"'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()")}}(n,t,e);return i instanceof Promise?i.then(function(r){return[].concat(r)}):[].concat(i)}var nw=function(){function n(t,e,i,r){t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i={}),r===void 0&&(r={}),this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=i,this.functionMap=r,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}return n.prototype.newFrame=function(t,e){return{id:t,frameName:e,iterationId:0}},Object.defineProperty(n.prototype,"currentContext",{get:function(){return this.contexts},set:function(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextId",{get:function(){return this._currentContextIds[0]},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextIds",{get:function(){return this._currentContextIds},enumerable:!0,configurable:!0}),n.prototype.generateCurrentContextIds=function(){for(var t=[],e=0;e<this.contexts.length-1;e++){var i=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(i))}t.push(""),this._currentContextIds=t},n.prototype.contextIdforContexts=function(t){return t?t.map(function(e){return e.id===0&&e.iterationId===0?"":e.frameName+"-"+e.iterationId}).join("/"):""},n.prototype.enterFrame=function(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))},n.prototype.exitFrame=function(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function rw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return jt(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return jt(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((iw(d)||QW(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function eU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return jt(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var tU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],nU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function iw(n){return tU.indexOf(n.op)>=0}function QW(n){return nU.indexOf(n.op)>=0}var aw=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=rw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return eU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[jt(d)[0]]}),s=e.map(function(d){return jt(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new nw(i.weightMap,c,h,i.functionExecutorMap),p=Kb({},i.weightMap);Object.keys(t).forEach(function(w){var S=jt(w),L=S[0],x=S[1],C=[];C[x]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g<u.length;g++){var v=u[g];if(!p[v.name]){var b=tw(v,p,d);if(b instanceof Promise)throw new Error("The execution of the op '"+v.op+"' returned a promise. Please use model.executeAsync() instead.");p[v.name]=b,i.checkTensorForDisposal(v.name,v,p,d,f,s,m)}}return i.parent==null&&d.dispose(f),e.map(function(w){return Ht(w,p,d)})})},n.prototype.getFrozenTensorIds=function(t){var e=[].concat.apply([],Object.keys(t).map(function(i){return t[i]}).map(function(i){return i.map(function(r){return r.id})}));return new Set(e)},n.prototype.checkTensorForDisposal=function(t,e,i,r,a,s,o){if(e.category==="control"||s.indexOf(t)!==-1)return;i[t].forEach(function(l){l!=null&&(o[l.id]=(o[l.id]||0)+e.children.length)}),e.inputs.forEach(function(l){if(l.category!=="control"){var u=jF(l.name,i,r);u!=null&&u.forEach(function(c){if(c&&!a.has(c.id)){var h=o[c.id];h===1?(c.dispose(),delete o[c.id]):h!=null&&o[c.id]--}})}})},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){return wn(this,function(i){return[2,this._executeAsync(t,e)]})})},n.prototype._executeAsync=function(t,e,i,r,a){return i===void 0&&(i=!1),r===void 0&&(r={}),a===void 0&&(a={}),zn(this,void 0,void 0,function(){var s,o,l,u,c,h;return wn(this,function(d){switch(d.label){case 0:return i||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e)),s=new nw(this.weightMap,r,a,this.functionExecutorMap),[4,this.executeWithControlFlow(t,s,e,i)];case 1:return o=d.sent(),l=e.map(function(p){return Ht(p,o,s)}),u=l.map(function(p){return p.id}),c=Object.keys(t).map(function(p){return t[p].id}),h=new Set(u.concat(c,this.weightIds)),Object.keys(o).forEach(function(p){var f=o[p];f.forEach(function(m){m&&!m.isDisposed&&!h.has(m.id)&&m.dispose()})}),this.parent==null&&s.dispose(h),[2,l]}})})},n.prototype.executeFunctionAsync=function(t,e,i){return zn(this,void 0,void 0,function(){var r,a=this;return wn(this,function(s){return r=t.reduce(function(o,l,u){return o[a.inputs[u].name]=l,o},{}),[2,this._executeAsync(r,this.outputNodes,!0,e,i)]})})},n.prototype.executeWithControlFlow=function(t,e,i,r){return zn(this,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x=this;return wn(this,function(C){switch(C.label){case 0:a=Object.keys(t),s=a.map(function(R){return x.graph.nodes[jt(R)[0]]}),o=i.map(function(R){return jt(R)[0]}),l=o.map(function(R){return x.graph.nodes[R]}),u=rw(t,l,this.weightMap),c=u.usedNodes,h=u.missingInputs,d=u.dynamicNode,p=u.syncInputs,f=s.concat(this.graph.weights).map(function(R){return{node:R,contexts:e.currentContext}}),m=Kb({},this.weightMap),Object.keys(t).forEach(function(R){var D=jt(R),k=D[0],W=D[1],F=[];F[W]=t[R],m[k]=F}),g={},v=this.getFrozenTensorIds(m),b={},C.label=1;case 1:return f.length>0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!iw(R)&&!Ht(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=Zn(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=tw(f.node,r,i);m||(m=Zn(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=Zn(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=jt(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=jt(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=jt(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var iU="?tfjs-format=file",rU="model.json",sw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return zn(this,void 0,void 0,function(){var t;return wn(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new aw(Jb.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Jb.Instance.transformGraph(t.modelInitializer);this.initializer=new aw(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return zn(this,void 0,void 0,function(){var i;return wn(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){var i;return wn(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function aU(n,t){return t===void 0&&(t={}),zn(this,void 0,void 0,function(){var e;return wn(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+rU+iU)),e=new sw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var sU="2.6.0";rr.GraphModel=sw;rr.deregisterOp=KF;rr.loadGraphModel=aU;rr.registerOp=YF;rr.version_converter=sU});var lw=be(()=>{});var Cw=be(Vt=>{"use strict";Object.defineProperty(Vt,"__esModule",{value:!0});var Ne=Xi();var cd=function(n,t){return cd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},cd(n,t)};function Ve(n,t){cd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var jr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function ar(n,t){return t={exports:{}},n(t,t.exports),t.exports}var oU=ar(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(jr,n,!1)}),lU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(jr,n,!1)}),uU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(jr,n,!1)}),cU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(jr,n,!1)}),hU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(jr,n,!1)}),dU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(jr,n,!1)}),sr=ar(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});sr.alea=oU;sr.xor128=lU;sr.xorwow=uU;sr.xorshift7=cU;sr.xor4096=hU;sr.tychei=dU;var pU=sr,uw=pU.alea;function fU(n,t){return xo(n,t)}function xo(n,t,e,i){if(e===void 0&&(e=new Map),i===void 0&&(i=new Set),n==null)return null;if(i.has(n))throw new Error("Circular references are not supported.");if(e.has(n))return e.get(n);var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if($r(n)){var a=Array.isArray(n)?[]:{};i.add(n);for(var s in n){var o=n[s],l=xo(o,t,e,i);a[s]=l}return i.delete(n),a}else throw new Error("Can't recurse into non-iterable type: "+n);else return e.set(n,r.value),r.value}function mU(n,t){return t===void 0&&(t=hw),cw(n,t)}function cw(n,t,e){e===void 0&&(e=new Set);var i=n[0];if(e.has(i))throw new Error("Circular references are not supported.");var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if($r(i)){var a=Array.isArray(i)?[]:{};e.add(i);var s=function(l){var u=n.map(function(h){return h[l]}),c=cw(u,t,e);a[l]=c};for(var o in i)s(o);return e.delete(i),a}else throw new Error("Can't recurse into non-iterable type: "+i);else return r.value}function hw(n){return n===null?null:$r(n[0])?{value:null,recurse:!0}:{value:n,recurse:!1}}function dw(n,t){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:e=new Map,xo(n,t,e),i=0,r=Array.from(e.keys()),u.label=1;case 1:return i<r.length?(a=r[i],s=e.get(a),s instanceof Promise?[4,s]:[3,3]):[3,4];case 2:o=u.sent(),e.set(a,o),u.label=3;case 3:return i++,[3,1];case 4:return l=xo(n,t,e),[2,l]}})})}function $r(n){return n!=null&&!ArrayBuffer.isView(n)&&(Array.isArray(n)||typeof n=="object"&&!(n instanceof Ne.Tensor))}function vU(n){return n==null||gU(n)||Array.isArray(n)||typeof n=="object"&&n instanceof Ne.Tensor||Ne.util.isTypedArray(n)}function gU(n){return n===null||typeof n!="object"&&typeof n!="function"}function bU(n){return fU(n,yU)}function yU(n){return n instanceof Ne.Tensor?{value:n.clone(),recurse:!1}:$r(n)?{value:null,recurse:!0}:{value:n,recurse:!1}}var pw=function(){function n(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}return n.prototype.wrap=function(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity},n.prototype.get=function(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]},n.prototype.set=function(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e},n.prototype.length=function(){var t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t},n.prototype.isFull=function(){return this.length()===this.capacity},n.prototype.isEmpty=function(){return this.length()===0},n.prototype.push=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)},n.prototype.pushAll=function(t){for(var e=0,i=t;e<i.length;e++){var r=i[e];this.push(r)}},n.prototype.pop=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);var t=this.get(this.end);return this.set(this.end,void 0),t},n.prototype.unshift=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)},n.prototype.shift=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t},n.prototype.shuffleExcise=function(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var e=this.wrap(this.begin+t),i=this.get(e);return this.set(e,this.pop()),i},n}();var wU=function(n){Ve(t,n);function t(){return n.call(this,t.INITIAL_CAPACITY)||this}return t.prototype.isFull=function(){return!1},t.prototype.push=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.push.call(this,e)},t.prototype.unshift=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.unshift.call(this,e)},t.prototype.expand=function(){for(var e=this.capacity*2,i=new Array(e),r=this.length(),a=0;a<r;a++)i[a]=this.get(this.wrap(this.begin+a));this.data=i,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=r},t.INITIAL_CAPACITY=32,t}(pw);function fw(n){return new SU(n)}function hd(n){return new LU(n)}function IU(n,t){return new mw(n,t)}function TU(n,t){return t===void 0&&(t=Ii.FAIL),new AU(n,t)}var Nt=function(){function n(){}return n.prototype.toArray=function(){return se(this,void 0,void 0,function(){var t,e;return oe(this,function(i){switch(i.label){case 0:return t=[],[4,this.next()];case 1:e=i.sent(),i.label=2;case 2:return e.done?[3,4]:(t.push(e.value),[4,this.next()]);case 3:return e=i.sent(),[3,2];case 4:return[2,t]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){var t,e,i;return oe(this,function(r){switch(r.label){case 0:return t=this.prefetch(100),e=[],[4,t.next()];case 1:i=r.sent(),r.label=2;case 2:return i.done?[3,4]:(e.push(i.value),[4,t.next()]);case 3:return i=r.sent(),[3,2];case 4:return[2,e]}})})},n.prototype.resolveFully=function(){return se(this,void 0,void 0,function(){var t;return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:t=e.sent(),e.label=2;case 2:return t.done?[3,4]:[4,this.next()];case 3:return t=e.sent(),[3,2];case 4:return[2]}})})},n.prototype.resolveWhile=function(t){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:return[4,this.next()];case 1:e=r.sent(),i=t(e.value),r.label=2;case 2:return!e.done&&i?[4,this.next()]:[3,4];case 3:return e=r.sent(),i=t(e.value),[3,2];case 4:return[2]}})})},n.prototype.handleErrors=function(t){return new DU(this,t)},n.prototype.filter=function(t){return new OU(this,t)},n.prototype.map=function(t){return new EU(this,t)},n.prototype.mapAsync=function(t){return new gw(this,t)},n.prototype.serialMapAsync=function(t){return new gw(this,t).serial()},n.prototype.flatmap=function(t){return new kU(this,t)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.map(t).resolveFully()]})})},n.prototype.serialForEach=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.serialMapAsync(t).resolveWhile(function(i){return i===!0})]})})},n.prototype.rowMajorBatch=function(t,e){return e===void 0&&(e=!0),new RU(this,t,e)},n.prototype.columnMajorBatch=function(t,e,i){e===void 0&&(e=!0),i===void 0&&(i=hw);var r=this.rowMajorBatch(t,e);return r.map(function(a){return mU(a,i)})},n.prototype.concatenate=function(t,e){return new mw(fw([this,t]),e)},n.prototype.take=function(t){return t<0||t==null?this:new CU(this,t)},n.prototype.skip=function(t){return t<0||t==null?this:new xU(this,t)},n.prototype.prefetch=function(t){return new vw(this,t)},n.prototype.shuffle=function(t,e){return new FU(this,t,e)},n.prototype.serial=function(){return new NU(this)},n}(),SU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.items=e,i.trav=0,i}return t.prototype.summary=function(){return"Array of "+this.items.length+" items"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return this.trav>=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:bU(e),done:!1}])})})},t}(Nt),LU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(Nt),NU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(Nt),xU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++<this.maxCount?[4,this.upstream.next()]:[3,2];case 1:return e=i.sent(),e.done?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2,this.upstream.next()]}})})},t}(Nt),CU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(Nt),RU=function(n){Ve(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length<this.batchSize?[4,this.upstream.next()]:[3,3];case 2:return i=r.sent(),i.done?this.enableSmallLastBatch&&e.length>0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(Nt),OU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(Nt),EU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),DU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.handler=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(Nt),gw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),dd=function(n){Ve(t,n);function t(){var e=n.call(this)||this;return e.outputQueue=new wU,e.lastRead=Promise.resolve({value:null,done:!1}),e}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.outputQueue.length()===0?[4,this.pump()]:[3,2];case 1:return e.sent()?[3,0]:[2,{value:null,done:!0}];case 2:return[2,{value:this.outputQueue.shift(),done:!1}]}})})},t}(Nt),kU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,!0]}})})},t}(dd),mw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.baseErrorHandler=i,r.lastRead=null,r.iterator=null,r.moreIterators=e,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of chained summaries";return e+" -> Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(Nt),Ii;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Ii||(Ii={}));var AU=function(n){Ve(t,n);function t(e,i){i===void 0&&(i=Ii.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof Nt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,dw(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Ii.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Ii.SHORTEST:return[2,{value:null,done:!0}];case Ii.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(Nt),vw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new pw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(Nt),FU=function(n){Ve(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=uw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(vw);var Ya=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is
+      `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),$t(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,WU)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,$t(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,$t(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return $t(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return $t(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return $t(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,$t(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=hd(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,IU(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?r=0:r=null,$t(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().skip(t)]}})})},r)},n.prototype.shuffle=function(t,e,i){var r=this;if(i===void 0&&(i=!0),t==null||t<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError("`Dataset.shuffle()` requires bufferSize to be specified.  If your data fits in main memory (for regular JS objects), and/or GPU memory (for `tf.Tensor`s), consider setting "+("bufferSize to the dataset size ("+this.size+" elements)"));var a=this,s=uw(e||Ne.util.now().toString());return $t(function(){return se(r,void 0,void 0,function(){var o;return oe(this,function(l){switch(l.label){case 0:return o=s.int32(),i&&(o+=s.int32()),[4,a.iterator()];case 1:return[2,l.sent().shuffle(t,o.toString())]}})})},this.size)},n.prototype.take=function(t){var e=this,i=this,r;return this.size!=null&&this.size>t?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,$t(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function $t(n,t){return t===void 0&&(t=null),new(function(e){Ve(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(Ya))}function UU(n){var t=this;return $t(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,fw(n)]})})},n.length)}function BU(n){var t=this;if(!$r(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i<n.length;i++)e=e==null?n[i].size:Math.min(e,n[i].size);else if(n instanceof Object)for(var r in n)e=e==null?n[r].size:Math.min(e,n[r].size);return $t(function(){return se(t,void 0,void 0,function(){var a;return oe(this,function(s){switch(s.label){case 0:return[4,dw(n,function(o){if(o instanceof Ya)return{value:o.iterator(),recurse:!1};if($r(o))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")})];case 1:return a=s.sent(),[2,TU(a,Ii.SHORTEST)]}})})},e)}function WU(n){if(n===null)return null;var t=n[0];if(vU(t)){var e=zU(n);return{value:e,recurse:!1}}return{value:null,recurse:!0}}function zU(n){if(n.length===0)throw new Error("Can't make a batch of zero elements.");return n[0]instanceof Ne.Tensor?Ne.stack(n):Ne.tensor(n)}var yw=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.input=e,i}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.input.iterator()];case 1:return e=a.sent(),i=e.decodeUTF8(),r=i.split(`
+`).map(function(s){return s.endsWith("\r")&&(s=s.slice(0,-1)),s}),[2,r]}})})},t}(Ya);var Co='"',Ka=Symbol("out"),bw=Symbol("field"),Ro=Symbol("quote"),pd=Symbol("quoteafterquote"),ww=Symbol("quoteinquote"),Sw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.input=e,r.hasHeader=!0,r.fullColumnNames=null,r.columnNamesValidated=!1,r.columnConfigs=null,r.configuredColumnsOnly=!1,r.delimiter=",",r.delimWhitespace=!1,r.base=new yw(e),i||(i={}),r.hasHeader=!(i.hasHeader===!1),r.fullColumnNames=i.columnNames,r.columnConfigs=i.columnConfigs,r.configuredColumnsOnly=i.configuredColumnsOnly,i.delimWhitespace?(Ne.util.assert(i.delimiter==null,function(){return"Delimiter should not be provided when delimWhitespace is true."}),r.delimWhitespace=!0,r.delimiter=" "):r.delimiter=i.delimiter?i.delimiter:",",r}return t.prototype.columnNames=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:e.sent(),e.label=2;case 2:return[2,this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames]}})})},t.prototype.setColumnNames=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u=this;return oe(this,function(c){switch(c.label){case 0:return[4,this.maybeReadHeaderLine()];case 1:if(e=c.sent(),!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");if(this.fullColumnNames&&e&&Ne.util.assert(e.length===this.fullColumnNames.length,function(){return"The length of provided columnNames ("+u.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."}),this.fullColumnNames||(this.fullColumnNames=e),i=this.fullColumnNames.reduce(function(h,d){return h[d]=h[d]+1||1,h},{}),r=Object.keys(i).filter(function(h){return i[h]>1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a<s.length;a++)if(o=s[a],l=this.fullColumnNames.indexOf(o),l===-1)throw new Error('The key "'+o+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}return this.columnNamesValidated=!0,[2]}})})},t.prototype.maybeReadHeaderLine=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return this.hasHeader?[4,this.base.iterator()]:[3,3];case 1:return e=s.sent(),[4,e.next()];case 2:if(i=s.sent(),i.done)throw new Error("No data was found for CSV parsing.");return r=i.value,a=this.parseRow(r,!1),[2,a];case 3:return[2,null]}})})},t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i=this;return oe(this,function(r){switch(r.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:r.sent(),r.label=2;case 2:return[4,this.base.iterator()];case 3:return e=r.sent(),this.hasHeader&&(e=e.skip(1)),[2,e.map(function(a){return i.makeDataElement(a)})]}})})},t.prototype.makeDataElement=function(e){for(var i=this.parseRow(e),r={},a={},s=0;s<this.fullColumnNames.length;s++){var o=this.fullColumnNames[s],l=this.columnConfigs?this.columnConfigs[o]:null;if(this.configuredColumnsOnly&&!l)continue;var u=i[s],c=null;if(u==="")if(l&&l.default!==void 0)c=l.default;else{if(l&&(l.required||l.isLabel))throw new Error("Required column "+o+" is empty in this line: "+e);c=void 0}else{var h=Number(u);if(isNaN(h))l&&l.dtype==="bool"?c=this.getBoolean(u):c=u;else if(!l||!l.dtype)c=h;else switch(l.dtype){case"float32":c=h;break;case"int32":c=Math.floor(h);break;case"bool":c=this.getBoolean(u);break;default:c=h}}l&&l.isLabel?a[o]=c:r[o]=c}return Object.keys(a).length===0?r:{xs:r,ys:a}},t.prototype.getBoolean=function(e){return e==="1"||e.toLowerCase()==="true"?1:0},t.prototype.parseRow=function(e,i){i===void 0&&(i=!0);for(var r=[],a=0,s=e.length,o=Ka,l=0;l<s;l++)switch(o){case Ka:switch(e.charAt(l)){case Co:a=l+1,o=Ro;break;case this.delimiter:if(a=l+1,this.delimiter===" "&&this.delimWhitespace)break;r.push(""),o=Ka;break;default:o=bw,a=l;break}break;case bw:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l)),o=Ka,a=l+1;break}break;case Ro:switch(e.charAt(l)){case Co:o=pd;break}break;case pd:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l-1)),o=Ka,a=l+1;break;case Co:o=Ro;break;default:o=ww;break}break;case ww:switch(e.charAt(l)){case Co:o=Ro;break}break}if(o===pd?r.push(e.substring(a,s-1)):r.push(e.substring(a)),i&&r.length!==this.fullColumnNames.length)throw new Error("Invalid row in csv file. Should have "+this.fullColumnNames.length+" elements in a row, but got "+r);return r},t}(Ya);var PU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;i.microphoneConfig=e,i.isClosed=!1,i.fftSize=e.fftSize||1024;var r=Math.log2(i.fftSize);if(i.fftSize<0||r<4||r>14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(Nt);var _U=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(Nt);var Lw=function(){function n(){}return n}();var Iw=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new MU(this,e)},t}(Nt),MU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new HU(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),HU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r<a.length;r++)s=a[r],this.outputQueue.push(s);return this.carryover=i[i.length-1],[2,!0]}})})},t}(dd);var qU=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.decodeUTF8=function(){return new VU(this)},t}(Nt),VU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.impl=new GU(e),i}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),GU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;if(i.upstream=e,Ne.env().get("IS_BROWSER"))i.decoder=new TextDecoder("utf-8");else{var r=lw().StringDecoder;i.decoder=new r("utf8")}return i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(dd);var Aw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(qU);function KU(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=YU(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new Aw(a,t)];case 3:throw new Error(r.statusText)}})})}var YU=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function Tw(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var Nw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return Tw(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new Aw(this.input,this.options)]})})},t}(Lw);var xw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return Tw(this.url)?[2,new Nw(this.url,this.fileOptions).iterator()]:[2,KU(this.url,this.fileOptions)]})})},t}(Lw);function jU(n,t){return t===void 0&&(t={}),new Sw(new xw(n),t)}function $U(n){var t=this,e=hd(n);return $t(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function XU(n){var t=this;return $t(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,hd(function(){return e.next()})]}})})})}function JU(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,_U.create(n,t)]})})}function ZU(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,PU.create(n)]})})}var QU="2.6.0";Vt.CSVDataset=Sw;Vt.Dataset=Ya;Vt.FileDataSource=Nw;Vt.TextLineDataset=yw;Vt.URLDataSource=xw;Vt.array=UU;Vt.csv=jU;Vt.func=$U;Vt.generator=XU;Vt.microphone=ZU;Vt.version_data=QU;Vt.webcam=JU;Vt.zip=BU});var Ow=be((Rw,fd)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c<u.length;c++){o+=u.charCodeAt(c);var h=.02519603282416938*o;o=h>>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(Rw,typeof fd=="object"&&fd,typeof define=="function"&&define)});var Dw=be((Ew,md)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(Ew,typeof md=="object"&&md,typeof define=="function"&&define)});var Fw=be((kw,gd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,u==l.length&&(o.d=o.x<<10^o.x>>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(kw,typeof gd=="object"&&gd,typeof define=="function"&&define)});var Uw=be((Ww,vd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h<c.length;++h)p[h&7]=p[h&7]<<15^c.charCodeAt(h)+p[h+1&7]<<13;for(;p.length<8;)p.push(0);for(h=0;h<8&&p[h]===0;++h);for(h==8?d=p[7]=-1:d=p[h],u.x=p,u.i=0,h=256;h>0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(Ww,typeof vd=="object"&&vd,typeof define=="function"&&define)});var zw=be((Bw,yd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f<v;++f)c&&(d^=c.charCodeAt((f+32)%c.length)),f===0&&(m=d),d^=d<<10,d^=d>>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(Bw,typeof yd=="object"&&yd,typeof define=="function"&&define)});var _w=be((Pw,bd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u<l.length+20;u++)o.b^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.a=s.a,o.b=s.b,o.c=s.c,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(Pw,typeof bd=="object"&&bd,typeof define=="function"&&define)});var Mw=be((CH,Oo)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var x=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),x),R=new p(x),D=function(){for(var k=R.g(r),W=o,F=0;k<l;)k=(k+F)*i,W*=i,F=R.g(1);for(;k>=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,x=this,C=0,R=x.i=x.j=0,D=x.S=[];for(L||(w=[L++]);C<i;)D[C]=C++;for(C=0;C<i;C++)D[C]=D[R=c&R+w[C%L]+(S=D[C])],D[R]=S;(x.g=function(k){for(var W,F=0,P=x.i,H=x.j,_=x.S;k--;)W=_[P=c&P+1],F=F*i+_[c&(_[P]=_[H=c&H+W])+(_[H]=W)];return x.i=P,x.j=H,F})(i)}function f(w,S){return S.i=w.i,S.j=w.j,S.S=w.S.slice(),S}function m(w,S){var L=[],x=typeof w,C;if(S&&x=="object")for(C in w)try{L.push(m(w[C],S-1))}catch(R){}return L.length?L:x=="string"?w:w+"\0"}function g(w,S){for(var L=w+"",x,C=0;C<L.length;)S[c&C]=c&(x^=S[c&C]*19)+L.charCodeAt(C++);return b(S)}function v(){try{var w;return h&&(w=h.randomBytes)?w=w(i):(w=new Uint8Array(i),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(x){var S=e.navigator,L=S&&S.plugins;return[+new Date,e,L,e.screen,b(n)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),n),typeof Oo=="object"&&Oo.exports){Oo.exports=d;try{h=ds()}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return d})})([],Math)});var Vw=be((RH,Hw)=>{var eB=Ow(),tB=Dw(),nB=Fw(),iB=Uw(),rB=zw(),aB=_w(),or=Mw();or.alea=eB;or.xor128=tB;or.xorwow=nB;or.xorshift7=iB;or.xor4096=rB;or.tychei=aB;Hw.exports=or});var m0=be(ja=>{"use strict";Object.defineProperty(ja,"__esModule",{value:!0});var T=Xi(),sB=Vw();var wd=function(n,t){return wd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},wd(n,t)};function oB(n,t){wd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function qw(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function Gw(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}function ae(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&T.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the CPU backend."})})}var lB=T.kernel_impls.nonMaxSuppressionV3Impl,uB=T.kernel_impls.split,cB=T.kernel_impls.tile,hB=T.kernel_impls.topkImpl,dB=T.kernel_impls.whereImpl;function Sd(n,t,e,i){if(e==="linear")return n.linear(t);if(e==="relu")return n.relu(t);if(e==="elu")return T.elu(t);if(e==="relu6")return n.relu6(t);if(e==="prelu")return n.prelu(t,i);throw new Error("Activation "+e+" has not been implemented for the CPU backend.")}var Yw=function(n){oB(t,n);function t(){var e=n.call(this)||this;return e.blockSize=48,e.firstUse=!0,e.data=new T.DataStorage(e,T.engine()),e}return t.prototype.write=function(e,i,r){this.firstUse&&(this.firstUse=!1,T.env().get("IS_NODE")&&T.backend_util.warn(`
 ============================
 Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed things up dramatically, install our node backend, which binds to TensorFlow C++, by running npm i @tensorflow/tfjs-node, or npm i @tensorflow/tfjs-node-gpu if you have CUDA. Then call require('@tensorflow/tfjs-node'); (-gpu suffix for CUDA) at the start of your program. Visit https://github.com/tensorflow/tfjs-node for more details.
-============================`));var a={};return this.data.set(a,{values:e,dtype:r,refCount:1}),a},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return{dataId:a,shape:e,dtype:i}},t.prototype.incRef=function(e){var i=this.data.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.data.has(e)){var i=this.data.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){this.data.set(e,{values:i,dtype:a,refCount:1})},t.prototype.numDataIds=function(){return this.data.numDataIds()},t.prototype.read=function(e){return qw(this,void 0,void 0,function(){return Gw(this,function(i){return[2,this.readSync(e)]})})},t.prototype.readSync=function(e){var i=this.data.get(e),r=i.dtype,a=i.complexTensorInfos;if(r==="complex64"){var s=this.readSync(a.real.dataId),o=this.readSync(a.imag.dataId);return T.backend_util.mergeRealAndImagArrays(s,o)}return this.data.get(e).values},t.prototype.bufferSync=function(e){var i=this.readSync(e.dataId),r=i;if(e.dtype==="string")try{r=i.map(function(a){return T.util.decodeString(a)})}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return T.buffer(e.shape,e.dtype,r)},t.prototype.makeOutput=function(e,i,r){var a=this.write(e,i,r);return T.engine().makeTensorFromDataId(a,i,r,this)},t.prototype.disposeData=function(e){if(this.data.has(e)){var i=this.data.get(e).complexTensorInfos;i!=null&&(this.disposeData(i.real.dataId),this.disposeData(i.imag.dataId)),this.data.delete(e)}},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.data.has(i)){var r=this.data.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.time=function(e){return qw(this,void 0,void 0,function(){var i,r;return Gw(this,function(a){return i=T.util.now(),e(),r=T.util.now()-i,[2,{kernelMs:r}]})})},t.prototype.memory=function(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}},t.prototype.stridedSlice=function(e,i,r,a){ae(e,"stridedSlice");var s=T.slice_util.computeOutShape(i,r,a);if(s.some(function(p){return p===0}))return T.tensor([],s);for(var o=T.buffer(s,e.dtype),l=this.bufferSync(e),u=0;u<o.size;u++){for(var c=o.indexToLoc(u),h=new Array(c.length),d=0;d<h.length;d++)h[d]=c[d]*a[d]+i[d];o.set.apply(o,[l.get.apply(l,h)].concat(c))}return o.toTensor()},t.prototype.diag=function(e){for(var i=this.readSync(e.dataId),r=T.buffer([e.size,e.size],e.dtype),a=r.values,s=0;s<i.length;s++)a[s*e.size+s]=i[s];return r.toTensor()},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=T.slice(e,l,u).reshape(a);return c},t.prototype.reverse=function(e,i){ae(e,"reverse");for(var r=T.buffer(e.shape,e.dtype),a=this.bufferSync(e),s=function(l){var u=r.indexToLoc(l),c=u.slice();i.forEach(function(h){return c[h]=e.shape[h]-1-c[h]}),r.set.apply(r,[a.get.apply(a,c)].concat(u))},o=0;o<r.size;o++)s(o);return r.toTensor()},t.prototype.neg=function(e){return ae(e,"neg"),T.mul(T.scalar(-1),e)},t.prototype.addN=function(e){var i=this;ae(e,"addN");for(var r=e.map(function(c){return i.readSync(c.dataId)}),a=T.buffer(e[0].shape,e[0].dtype),s=a.values,o=0;o<e.length;o++)for(var l=r[o],u=0;u<s.length;u++)s[u]+=l[u];return a.toTensor()},t.prototype.softmax=function(e,i){var r=T.util.parseAxisParam([i],e.shape),a=T.max(e,r),s=T.backend_util.expandShapeToKeepDim(a.shape,r),o=T.sub(e,a.reshape(s)),l=T.exp(o),u=this.sum(l,r).reshape(s);return T.div(l,u)},t.prototype.pow=function(e,i){return ae([e,i],"pow"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.pow(r,a)})},t.prototype.batchMatMul=function(e,i,r,a){ae([e,i],"matMul");for(var s=r?e.shape[1]:e.shape[2],o=r?e.shape[2]:e.shape[1],l=a?i.shape[1]:i.shape[2],u=e.shape[0],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=r?[e.strides[0],1,e.strides[1]]:[e.strides[0],e.strides[1],1],p=d[0],f=d[1],m=d[2],g=a?[1,i.strides[1],i.strides[0]]:[i.strides[1],1,i.strides[0]],v=g[0],b=g[1],w=g[2],S=o*l,L=T.buffer([u,o,l],e.dtype),x=L.values,C=this.blockSize,R=0;R<u;R++)for(var D=0;D<o;D+=C)for(var k=0;k<l;k+=C)for(var W=0;W<s;W+=C)for(var F=Math.min(D+C,o),P=Math.min(k+C,l),H=Math.min(W+C,s),_=D;_<F;_++)for(var K=k;K<P;K++){for(var j=0,q=W;q<H;q++)j+=c[R*p+_*f+q*m]*h[q*v+K*b+R*w];x[R*S+(_*l+K)]+=j}return L.toTensor()},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=this.batchMatMul(i,r,a,s);return o&&(c=T.add(c,o)),l&&(c=wd(this,c,l,u)),c},t.prototype.floorDiv=function(e,i){ae([e,i],"floorDiv");var r=function(s,o){return Math.floor(s/o)},a="int32";return this.broadcastedBinaryOp(e,i,a,r)},t.prototype.sum=function(e,i){ae(e,"sum"),T.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=0,m=0;m<u;++m)f+=h[p+m];c[d]=f}return l},t.prototype.prod=function(e,i){ae(e,"sum");for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=1,m=0;m<u;++m)f*=h[p+m];c[d]=f}return l},t.prototype.unsortedSegmentSum=function(e,i,r){ae(e,"unsortedSegmentSum");for(var a=[],s=e.rank-i.rank,o=0;o<s;++o)i=i.expandDims(o+1);for(var o=0;o<r;++o){var l=T.scalar(o,"int32"),u=T.equal(l,i).asType("float32"),c=u.mul(e).sum(0);a.push(c)}return T.stack(a)},t.prototype.argMin=function(e,i){ae(e,"argMin");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMin",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v<f&&(f=v,m=g)}c[d]=m}return l},t.prototype.argMax=function(e,i){ae(e,"argMax");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMax",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v>f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;d<u.length;d+=c)for(var p=0;p<c;p++){var f=h(d,p);if(p===0)l[f]=r?0:u[f];else{var m=h(d,p-1);l[f]=r?u[m]+l[m]:u[f]+l[m]}}return o},t.prototype.equal=function(e,i){return ae([e,i],"equal"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r===a?1:0})},t.prototype.notEqual=function(e,i){return ae([e,i],"notEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r!==a?1:0})},t.prototype.less=function(e,i){return ae([e,i],"less"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<a?1:0})},t.prototype.lessEqual=function(e,i){return ae([e,i],"lessEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<=a?1:0})},t.prototype.greater=function(e,i){return ae([e,i],"greater"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>a?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d<a.length;d++)for(var p=0;p<h;p++)a[d]===1?u[c++]=s[d]:u[c++]=o[d];return l},t.prototype.where=function(e){ae([e],"where");var i=this.readSync(e.dataId);return dB(e.shape,i)},t.prototype.topk=function(e,i,r){ae(e,"topk");var a=this.readSync(e.dataId);return hB(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){ae(e,"min"),T.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];m<p&&(p=m)}u[h]=p}return o},t.prototype.minimum=function(e,i){return ae([e,i],"minimum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.min(r,a)})},t.prototype.mod=function(e,i){return ae([e,i],"mod"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r%a;return r<0&&a<0||r>=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p&&m}u[h]=p}return o},t.prototype.any=function(e,i){ae(e,"any"),T.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p||m}u[h]=p}return o},t.prototype.squaredDifference=function(e,i){return ae([e,i],"squaredDifference"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r-a;return s*s})},t.prototype.linear=function(e){return e},t.prototype.relu=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.max(0,a[s]);return i},t.prototype.relu6=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.min(Math.max(0,a[s]),6);return i},t.prototype.prelu=function(e,i){return ae([e,i],"prelu"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return r<0?a*r:r})},t.prototype.eluDer=function(e,i){ae([e,i],"eluDer");for(var r=new Float32Array(i.size),a=this.readSync(i.dataId),s=this.readSync(e.dataId),o=0;o<a.length;++o){var l=a[o];l>=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=wd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),x=this.readSync(i.dataId),C=d.values,R=0;R<r.batchSize;++R)for(var D=R*p,k=R*v,W=0;W<r.outHeight;++W)for(var F=k+W*b,P=W*r.strideHeight-c,H=0;H<a;H++){var _=P+H*o;if(_<0||_>=r.inHeight)continue;for(var K=H*i.strides[0],j=D+_*f,q=0;q<r.outWidth;++q)for(var G=F+q*w,Z=q*r.strideWidth-u,X=0;X<s;X++){var ee=Z+X*l;if(ee<0||ee>=r.inWidth)continue;for(var ne=K+X*i.strides[1],ie=j+ee*m,te=ne,re=0;re<r.inChannels;++re){for(var le=L[ie+re*g],he=0;he<r.outChannels;++he)C[G+he*S]+=le*x[te+he];te+=r.outChannels}}}return d.toTensor()},t.prototype.conv3d=function(e,i,r){for(var a=r.filterDepth,s=r.filterHeight,o=r.filterWidth,l=r.dilationDepth,u=r.dilationHeight,c=r.dilationWidth,h=r.padInfo.front,d=r.padInfo.left,p=r.padInfo.top,f=T.buffer(r.outShape,e.dtype),m=this.readSync(e.dataId),g=this.readSync(i.dataId),v=f.values,b=0;b<r.batchSize;++b)for(var w=b*e.strides[0],S=b*f.strides[0],L=0;L<r.outDepth;++L)for(var x=S+L*f.strides[1],C=L*r.strideDepth-h,R=0;R<a;R++){var D=C+R*l;if(D<0||D>=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F<r.outHeight;++F)for(var P=x+F*f.strides[2],H=F*r.strideHeight-p,_=0;_<s;_++){var K=H+_*u;if(K<0||K>=r.inHeight)continue;for(var j=k+_*i.strides[1],q=W+K*e.strides[2],G=0;G<r.outWidth;++G)for(var Z=P+G*r.outChannels,X=G*r.strideWidth-d,ee=0;ee<o;ee++){var ne=X+ee*c;if(ne<0||ne>=r.inWidth)continue;for(var ie=j+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le<r.inChannels;++le){for(var he=m[te+le],ye=0;ye<r.outChannels;++ye)v[Z+ye]+=he*g[re+ye];re+=r.outChannels}}}}return f.toTensor()},t.prototype.conv2dDerInput=function(e,i,r){ae([e,i],"conv2dDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=this.readSync(e.dataId),l=this.readSync(i.dataId),u=i.strides,c=u[0],h=u[1],d=u[2],p=r.batchSize,f=r.filterHeight,m=r.filterWidth,g=r.inChannels,v=r.inHeight,b=r.inWidth,w=r.outChannels,S=r.outHeight,L=r.outWidth,x=r.strideHeight,C=r.strideWidth,R=r.dataFormat,D=f-1-r.padInfo.top,k=m-1-r.padInfo.left,W=R==="channelsLast",F=a.strides[0],P=W?a.strides[1]:a.strides[2],H=W?a.strides[2]:1,_=W?1:a.strides[1],K=e.strides[0],j=W?e.strides[1]:e.strides[2],q=W?e.strides[2]:1,G=W?1:e.strides[1],Z=0;Z<p;++Z)for(var X=0;X<g;++X)for(var ee=0;ee<v;++ee)for(var ne=ee-D,ie=Math.max(0,Math.ceil(ne/x)),te=Math.min(S,(f+ne)/x),re=0;re<b;++re){for(var le=re-k,he=Math.max(0,Math.ceil(le/C)),ye=Math.min(L,(m+le)/C),Oe=0,Fe=ie;Fe<te;++Fe)for(var _e=Fe*x-ne,Me=he;Me<ye;++Me)for(var ht=Me*C-le,St=K*Z+j*Fe+q*Me,Ge=c*(f-1-_e)+h*(m-1-ht)+d*X,Ue=0;Ue<w;++Ue){var me=o[St+G*Ue],Dt=l[Ge+Ue];Oe+=me*Dt}var lt=F*Z+P*ee+H*re+_*X;s[lt]=Oe}return a.toTensor()},t.prototype.conv3dDerInput=function(e,i,r){for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=o[3],d=this.readSync(e.dataId),p=e.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=r.batchSize,D=r.filterDepth,k=r.filterHeight,W=r.filterWidth,F=r.inChannels,P=r.inDepth,H=r.inHeight,_=r.inWidth,K=r.outChannels,j=r.outDepth,q=r.outHeight,G=r.outWidth,Z=r.strideDepth,X=r.strideHeight,ee=r.strideWidth,ne=D-1-r.padInfo.front,ie=k-1-r.padInfo.top,te=W-1-r.padInfo.left,re=0;re<R;++re)for(var le=0;le<F;++le)for(var he=0;he<P;++he)for(var ye=he-ne,Oe=Math.max(0,Math.ceil(ye/Z)),Fe=Math.min(j,(D+ye)/Z),_e=0;_e<H;++_e)for(var Me=_e-ie,ht=Math.max(0,Math.ceil(Me/X)),St=Math.min(q,(k+Me)/X),Ge=0;Ge<_;++Ge){for(var Ue=Ge-te,me=Math.max(0,Math.ceil(Ue/ee)),Dt=Math.min(G,(W+Ue)/ee),lt=0,dt=Oe;dt<Fe;++dt)for(var xn=dt*Z-ye,Rt=ht;Rt<St;++Rt)for(var Lt=Rt*X-Me,Ce=me;Ce<Dt;++Ce)for(var kt=Ce*ee-Ue,pt=f*re+m*dt+g*Rt+v*Ce,it=S*(D-1-xn)+L*(k-1-Lt)+x*(W-1-kt)+C*le,on=0;on<K;++on){var hs=d[pt+on],Bt=b[it+on];lt+=hs*Bt}s[l*re+u*he+c*_e+h*Ge+le]=lt}return a.toTensor()},t.prototype.conv2dDerFilter=function(e,i,r){ae([e,i],"conv2dDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=r.dataFormat==="channelsLast",c=T.buffer(r.filterShape,"float32"),h=r.padInfo.left,d=r.padInfo.top,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((d-m)/a)),v=Math.min(r.outHeight,(r.inHeight+d-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((h-b)/s)),S=Math.min(r.outWidth,(r.inWidth+h-b)/s),L=0;L<r.inChannels;++L)for(var x=0;x<r.outChannels;++x){for(var C=0,R=0;R<r.batchSize;++R)for(var D=g;D<v;++D)for(var k=m+D*a-d,W=w;W<S;++W){var F=b+W*s-h;u?C+=p.get(R,k,F,L)*f.get(R,D,W,x):C+=p.get(R,L,k,F)*f.get(R,x,D,W)}c.set(C,m,b,L,x)}return c.toTensor()},t.prototype.conv3dDerFilter=function(e,i,r){for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=T.buffer(r.filterShape,"float32"),d=h.values,p=h.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=this.readSync(e.dataId),D=e.strides,k=D[0],W=D[1],F=D[2],P=D[3],H=r.padInfo.front,_=r.padInfo.left,K=r.padInfo.top,j=0;j<l;++j)for(var q=Math.max(0,Math.ceil((H-j)/a)),G=Math.min(r.outDepth,(r.inDepth+H-j)/a),Z=j*f,X=0;X<u;++X)for(var ee=Math.max(0,Math.ceil((K-X)/s)),ne=Math.min(r.outHeight,(r.inHeight+K-X)/s),ie=X*m+Z,te=0;te<c;++te)for(var re=Math.max(0,Math.ceil((_-te)/o)),le=Math.min(r.outWidth,(r.inWidth+_-te)/o),he=te*g+ie,ye=0;ye<r.inChannels;++ye)for(var Oe=ye*v+he,Fe=0;Fe<r.outChannels;++Fe){for(var _e=0,Me=0;Me<r.batchSize;++Me)for(var ht=Me*k,St=Me*S,Ge=q;Ge<G;++Ge)for(var Ue=j+Ge*a-H,me=Ue*W+ht,Dt=Ge*L+St,lt=ee;lt<ne;++lt)for(var dt=X+lt*s-K,xn=dt*F+me,Rt=lt*x+Dt,Lt=re;Lt<le;++Lt){var Ce=te+Lt*o-_,kt=Ce*P+xn,pt=Lt*C+Rt;_e+=R[kt+ye]*b[pt+Fe]}d[Oe+Fe]=_e}return h.toTensor()},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.depthwiseConv2D(i,r,a);return s&&(u=T.add(u,s)),o&&(u=wd(this,u,o,l)),u},t.prototype.depthwiseConv2D=function(e,i,r){ae([e,i],"depthwiseConv2D");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.outChannels/r.inChannels,d=T.buffer(r.outShape,e.dtype),p=this.readSync(e.dataId),f=this.readSync(i.dataId),m=d.values,g=0;g<r.batchSize;++g)for(var v=g*e.strides[0],b=g*d.strides[0],w=0;w<r.outHeight;++w)for(var S=b+w*d.strides[1],L=w*r.strideHeight-u,x=0;x<a;++x){var C=L+x*o;if(C<0||C>=r.inHeight)continue;for(var R=x*i.strides[0],D=v+C*e.strides[1],k=0;k<r.outWidth;++k)for(var W=S+k*d.strides[2],F=k*r.strideWidth-c,P=0;P<s;++P){var H=F+P*l;if(H<0||H>=r.inWidth)continue;for(var _=R+P*i.strides[1],K=D+H*r.inChannels,j=W,q=_,G=0;G<r.inChannels;++G){for(var Z=p[K+G],X=0;X<h;++X)m[j+X]+=Z*f[q+X];j+=h,q+=h}}}return d.toTensor()},t.prototype.depthwiseConv2DDerInput=function(e,i,r){ae([e,i],"depthwiseConv2DDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=this.readSync(e.dataId),d=e.strides,p=d[0],f=d[1],m=d[2],g=this.readSync(i.dataId),v=i.strides,b=v[0],w=v[1],S=v[2],L=r.batchSize,x=r.filterHeight,C=r.filterWidth,R=r.inChannels,D=r.inHeight,k=r.inWidth,W=r.outChannels,F=r.outHeight,P=r.outWidth,H=r.strideHeight,_=r.strideWidth,K=x-1-r.padInfo.top,j=C-1-r.padInfo.left,q=W/R,G=0;G<L;++G)for(var Z=0;Z<R;++Z)for(var X=0;X<D;++X)for(var ee=X-K,ne=Math.max(0,Math.ceil(ee/H)),ie=Math.min(F,(x+ee)/H),te=0;te<k;++te){for(var re=te-j,le=Math.max(0,Math.ceil(re/_)),he=Math.min(P,(C+re)/_),ye=0,Oe=ne;Oe<ie;++Oe)for(var Fe=Oe*H-ee,_e=le;_e<he;++_e)for(var Me=_e*_-re,ht=p*G+f*Oe+m*_e,St=b*(x-1-Fe)+w*(C-1-Me)+S*Z,Ge=0;Ge<q;++Ge){var Ue=Z*q+Ge,me=h[ht+Ue],Dt=g[St+Ge];ye+=me*Dt}s[l*G+u*X+c*te+Z]=ye}return a.toTensor()},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){ae([e,i],"depthwiseConv2DDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=T.buffer(r.filterShape,"float32"),c=r.padInfo.left,h=r.padInfo.top,d=r.outChannels/r.inChannels,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((h-m)/a)),v=Math.min(r.outHeight,(r.inHeight+h-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((c-b)/s)),S=Math.min(r.outWidth,(r.inWidth+c-b)/s),L=0;L<r.outChannels;++L){for(var x=Math.trunc(L/d),C=L%d,R=0,D=0;D<r.batchSize;++D)for(var k=g;k<v;++k)for(var W=m+k*a-h,F=w;F<S;++F){var P=b+F*s-c;R+=p.get(D,W,P,x)*f.get(D,k,F,L)}u.set(R,m,b,x,C)}return u.toTensor()},t.prototype.tile=function(e,i){return ae(e,"tile"),cB(this.bufferSync(e),i)},t.prototype.gather=function(e,i,r){ae([e,i],"gather");var a=e.shape.slice(),s=this.readSync(i.dataId);a[r]=s.length;for(var o=T.buffer(a,e.dtype),l=this.bufferSync(e),u=0;u<o.size;++u){var c=o.indexToLoc(u),h=c.slice();h[r]=s[c[r]];var d=l.locToIndex(h);o.values[u]=l.values[d]}return o.toTensor()},t.prototype.batchToSpaceND=function(e,i,r){ae([e],"batchToSpaceND");var a=i.reduce(function(h,d){return h*d}),s=T.backend_util.getReshaped(e.shape,i,a),o=T.backend_util.getPermuted(s.length,i.length),l=T.backend_util.getReshapedPermuted(e.shape,i,a),u=T.backend_util.getSliceBeginCoords(r,i.length),c=T.backend_util.getSliceSize(l,r,i.length);return T.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.pool3d=function(e,i,r){ae(e,"pool3d");for(var a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=r==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,b=this.readSync(e.dataId),w=T.buffer(i.outShape,e.dtype),S=w.values,L=i.outShape[1]*i.outShape[2]*i.outShape[3]*i.outShape[4],x=i.outShape[2]*i.outShape[3]*i.outShape[4],C=i.outShape[3]*i.outShape[4],R=i.outShape[4],D=0;D<i.batchSize;++D)for(var k=D*L,W=D*e.strides[0],F=0;F<i.inChannels;++F)for(var P=0;P<i.outDepth;++P){for(var H=P*a-f,_=H;_<0;)_+=l;for(var K=Math.min(i.inDepth,h+H),j=k+P*x,q=0;q<i.outHeight;++q){for(var G=q*s-m,Z=G;Z<0;)Z+=u;for(var X=Math.min(i.inHeight,d+G),ee=j+q*C,ne=0;ne<i.outWidth;++ne){for(var ie=ne*o-g,te=ie;te<0;)te+=c;for(var re=Math.min(i.inWidth,p+ie),le=ee+ne*R,he=v,ye=0,Oe=0,Fe=_;Fe<K;Fe+=l){for(var _e=W+Fe*e.strides[1],Me=Z;Me<X;Me+=u){for(var ht=_e+Me*e.strides[2],St=te;St<re;St+=c){var Ge=ht+St*e.strides[3],Ue=b[Ge+F];if(r==="max"&&Ue>he?he=Ue:r==="avg"&&(ye+=Ue,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),x=this.bufferSync(e),C=0;C<r.batchSize;++C)for(var R=0;R<r.inChannels;++R)for(var D=0;D<r.inDepth;++D)for(var k=0;k<r.inHeight;++k)for(var W=0;W<r.inWidth;++W){for(var F=D-v,P=k-w,H=W-b,_=0,K=0;K<f;K+=h){var j=(F+K)/a;if(j<0||j>=r.outDepth||Math.floor(j)!==j)continue;for(var q=0;q<m;q+=d){var G=(P+q)/s;if(G<0||G>=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z<g;Z+=p){var X=(H+Z)/o;if(X<0||X>=r.outWidth||Math.floor(X)!==X)continue;var ee=x.get(C,j,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b<i.batchSize;++b)for(var w=0;w<i.inChannels;++w)for(var S=0;S<i.outDepth;++S){for(var L=S*a-f,x=L;x<0;)x+=l;for(var C=Math.min(i.inDepth,h+L),R=0;R<i.outHeight;++R){for(var D=R*s-m,k=D;k<0;)k+=u;for(var W=Math.min(i.inHeight,d+D),F=0;F<i.outWidth;++F){for(var P=F*o-g,H=P;H<0;)H+=c;for(var _=Math.min(i.inWidth,p+P),K=Number.NEGATIVE_INFINITY,j=-1,q=x;q<C;q+=l)for(var G=q-L,Z=k;Z<W;Z+=u)for(var X=Z-D,ee=H;ee<_;ee+=c){var ne=ee-P,ie=v.get(b,q,Z,ee,w);ie>=K&&(K=ie,j=G*d*p+X*d+ne)}r.set(j,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),x=0;x<a.batchSize;++x)for(var C=0;C<a.inChannels;++C)for(var R=0;R<a.inDepth;++R)for(var D=0;D<a.inHeight;++D)for(var k=0;k<a.inWidth;++k){for(var W=R-g,F=D-b,P=k-v,H=0,_=0;_<p;_+=c){var K=(W+_)/o;if(K<0||K>=a.outDepth||Math.floor(K)!==K)continue;for(var j=0;j<f;j+=h){var q=(F+j)/l;if(q<0||q>=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<m;G+=d){var Z=(P+G)/u;if(Z<0||Z>=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(x,K,q,Z,C),ee=_*f*m+j*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(x,K,q,Z,C);H+=ie*ne}}}w.set(H,x,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b<o;b++)for(var w=0;w<i;w++)for(var S=g*w,L=Math.floor(S),x=S-L,C=Math.min(l-1,Math.ceil(S)),R=b*e.strides[0]+L*e.strides[1],D=b*e.strides[0]+C*e.strides[1],k=0;k<r;k++)for(var W=v*k,F=Math.floor(W),P=W-F,H=Math.min(u-1,Math.ceil(W)),_=R+F*e.strides[2],K=D+F*e.strides[2],j=R+H*e.strides[2],q=D+H*e.strides[2],G=0;G<c;G++){var Z=h[_+G],X=h[K+G],ee=h[j+G],ne=h[q+G],ie=Z+(ee-Z)*P,te=X+(ne-X)*P,re=ie+(te-ie)*x;d[m++]=re}return T.tensor(d,[o,i,r,c])},t.prototype.resizeBilinearBackprop=function(e,i,r){ae([e,i],"resizeBilinearBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=[r&&h>1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S<s;S++)for(var L=S*i.strides[0],x=0;x<h;x++)for(var C=x*g,R=Math.floor(C),D=Math.min(Math.ceil(C),o-1),k=L+R*i.strides[1],W=L+D*i.strides[1],F=C-R,P=1-F,H=0;H<d;H++)for(var _=H*v,K=Math.floor(_),j=Math.min(Math.ceil(_),l-1),q=_-K,G=1-q,Z=k+K*i.strides[2],X=k+j*i.strides[2],ee=W+K*i.strides[2],ne=W+j*i.strides[2],ie=P*G,te=P*q,re=F*G,le=F*q,he=0;he<u;he++){var ye=b[w++];p[Z+he]+=ye*ie,p[X+he]+=ye*te,p[ee+he]+=ye*re,p[ne+he]+=ye*le}return T.tensor4d(p,[s,l,o,u],i.dtype)},t.prototype.resizeNearestNeighbor=function(e,i,r,a){ae(e,"resizeNearestNeighbor");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(o*i*r*c),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b<o;b++)for(var w=b*e.strides[0],S=0;S<i;S++)for(var L=m*S,x=Math.min(l-1,a?Math.round(L):Math.floor(L)),C=w+x*e.strides[1],R=0;R<r;R++)for(var D=g*R,k=Math.min(u-1,a?Math.round(D):Math.floor(D)),W=C+k*e.strides[2],F=0;F<c;F++){var P=h[W+F];d[v++]=P}return T.tensor(d,[o,i,r,c],e.dtype)},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){ae([e,i],"resizeNearestNeighborBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=this.readSync(e.dataId),m=[r&&h>1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,x=Math.ceil(S)*2+2,C=0;C<s;C++)for(var R=C*i.strides[0],D=0;D<o;D++)for(var k=R+D*i.strides[1],W=Math.floor(D*w),F=Math.floor(W-L/2),P=0;P<l;P++)for(var H=k+P*i.strides[2],_=Math.floor(P*S),K=Math.floor(_-x/2),j=0;j<u;j++){for(var q=0,G=0;G<L;G++){var Z=G+F;if(Z<0||Z>=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie<x;ie++){var te=ie+K;if(te<0||te>=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+j])}}p[H+j]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p<c;p++){var f=d(p),m=u[p]*Math.pow(r+a*f,-s);h[p]=m}return T.tensor4d(h,e.shape)},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){ae(e,"LRNGrad");for(var u=e.shape[3],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=this.readSync(r.dataId),p=new Float32Array(e.size),f=e.size,m=0;m<f;m++){for(var g=m%u,v=m-g+Math.max(0,g-a),b=m-g+Math.min(u,g+a+1),w=0,S=v;S<b;S++)w+=Math.pow(h[S],2);w=o*w+s;for(var S=v;S<b;S++){var L=-2*o*l*h[S]*d[m]/w;m===S&&(L+=Math.pow(w,-l)),L*=c[m],p[S]+=L}}return T.tensor4d(p,e.shape)},t.prototype.multinomial=function(e,i,r,a){ae(e,"multinomial");for(var s=i?e:T.softmax(e),o=s.shape[0],l=s.shape[1],u=T.zeros([o,r],"int32"),c=this.readSync(u.dataId),h=this.readSync(s.dataId),d=0;d<o;++d){var p=d*l,f=new Float32Array(l-1);f[0]=h[p];for(var m=1;m<f.length;++m)f[m]=f[m-1]+h[p+m];for(var g=sB.alea(a.toString()),v=d*r,b=0;b<r;++b){var w=g();c[v+b]=f.length;for(var S=0;S<f.length;S++)if(w<f[S]){c[v+b]=S;break}}}return u},t.prototype.oneHot=function(e,i,r,a){ae(e,"oneHot");var s=new Float32Array(e.size*i);s.fill(a);for(var o=this.readSync(e.dataId),l=0;l<e.size;++l)o[l]>=0&&o[l]<i&&(s[l*i+o[l]]=r);return T.tensor2d(s,[e.size,i],"int32")},t.prototype.nonMaxSuppression=function(e,i,r,a,s){ae(e,"nonMaxSuppression");var o=this.readSync(e.dataId),l=this.readSync(i.dataId);return lB(o,l,r,a,s)},t.prototype.depthToSpace=function(e,i,r){T.util.assert(r==="NHWC",function(){return"Only NHWC dataFormat supported on CPU for depthToSpace. Got "+r}),T.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m<a;++m)for(var g=0;g<u;++g)for(var v=Math.floor(g/i),b=g%i,w=0;w<c;++w)for(var S=Math.floor(w/i),L=w%i,x=(b*i+L)*h,C=0;C<h;++C){var R=C+x,D=R+l*(S+o*(v+s*m));p[f++]=d[D]}return T.tensor4d(p,[a,u,c,h])},t.prototype.broadcastedBinaryOp=function(e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(e.shape,i.shape),o=T.buffer(s,r),l=this.readSync(e.dataId),u=this.readSync(i.dataId),c=T.backend_util.getBroadcastDims(e.shape,s),h=T.backend_util.getBroadcastDims(i.shape,s),d=o.values;if(c.length+h.length===0)for(var p=0;p<d.length;++p)d[p]=a(l[p%l.length],u[p%u.length]);else for(var f=this.bufferSync(e),m=this.bufferSync(i),g=function(v){var b=o.indexToLoc(v),w=b.slice(-e.rank);c.forEach(function(C){return w[C]=0});var S=f.locToIndex(w),L=b.slice(-i.rank);h.forEach(function(C){return L[C]=0});var x=m.locToIndex(L);d[v]=a(l[S],u[x])},p=0;p<d.length;++p)g(p);return o.toTensor()},t.prototype.split=function(e,i,r){return uB(e,i,r)},t.prototype.dispose=function(){},t.prototype.floatPrecision=function(){return 32},t.prototype.epsilon=function(){return n.prototype.epsilon.call(this)},t.prototype.cropAndResize=function(e,i,r,a,s,o){for(var l=e.shape,u=l[0],c=l[1],h=l[2],d=l[3],p=i.shape[0],f=a[0],m=a[1],g=T.buffer([p,f,m,d],"float32"),v=this.readSync(i.dataId),b=this.readSync(r.dataId),w=this.readSync(e.dataId),S=e.strides,L=g.strides,x=0;x<p;x++){var C=x*4,R=v[C],D=v[C+1],k=v[C+2],W=v[C+3],F=b[x];if(F>=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_<f;_++){var K=f>1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(K<0||K>c-1){for(var j=0;j<m;j++)for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}if(s==="bilinear")for(var Z=Math.floor(K),X=Math.ceil(K),ee=K-Z,j=0;j<m;j++){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var ie=Math.floor(ne),te=Math.ceil(ne),re=ne-ie,q=0;q<d;q++){var G=q+ie*S[2]+Z*S[1]+F*S[0],le=w[G];G=q+te*S[2]+Z*S[1]+F*S[0];var he=w[G];G=q+ie*S[2]+X*S[1]+F*S[0];var ye=w[G];G=q+te*S[2]+X*S[1]+F*S[0];var Oe=w[G],Fe=le+(he-le)*re,_e=ye+(Oe-ye)*re;G=q+j*L[2]+_*L[1]+x*L[0],g.values[G]=Fe+(_e-Fe)*ee}}else for(var j=0;j<m;++j){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var Me=Math.round(ne),ht=Math.round(K),q=0;q<d;q++){var St=q+Me*S[2]+ht*S[1]+F*S[0],Ge=q+j*L[2]+_*L[1]+x*L[0];g.values[Ge]=w[St]}}}}return g.toTensor()},t.prototype.sparseToDense=function(e,i,r,a){var s=T.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.sliceSize,c=s.strides,h=s.outputSize,d=!1;return this.scatter(e,i,r,h,u,l,o,c,a,d)},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=T.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3];if(l===0)return T.tensor([],o,e.dtype);for(var h=new T.TensorBuffer([l,u],e.dtype),d=this.readSync(i.dataId),p=this.readSync(e.dataId),f=0;f<l;f++){for(var m=[],g=0,v=0;v<a;v++){var b=d[f*a+v];g+=b*c[v],m.push(b)}if(g<0||g>=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w<u;w++)h.values[f*u+w]=p[g*u+w]}return h.toTensor().reshape(o)},t.prototype.scatterND=function(e,i,r){var a=T.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=T.scalar(0),d=!0;return this.scatter(e,i,r,c,l,o,s,u,h,d)},t.prototype.fill=function(e,i,r){r=r||T.util.inferDtype(i);var a=T.util.getArrayFromDType(r,T.util.sizeFromShape(e));return a.fill(i),T.engine().makeTensor(a,e,r,this)},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported for string tensors");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){var i=T.util.getArrayFromDType(e.dtype,T.util.sizeFromShape(e.shape));return this.makeOutput(i,e.shape,e.dtype)},t.prototype.linspace=function(e,i,r){return T.backend_util.linspaceImpl(e,i,r)},t.prototype.scatter=function(e,i,r,a,s,o,l,u,c,h){var d=[a/s,s],p=this.readSync(e.dataId),f=this.readSync(i.dataId);if(a===0)return T.tensor([],r,i.dtype);var m=new T.TensorBuffer(d,i.dtype);m.values.fill(this.readSync(c.dataId)[0]);for(var g=0;g<o;g++){for(var v=[],b=0,w=0;w<l;w++){var S=p[g*l+w];v.push(S),b+=S*u[w]}if(b<0||b>=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;L<s;L++)h?m.values[b*s+L]+=f[g*s+L]:m.values[b*s+L]=i.rank===0?f[0]:f[g*s+L]}return m.toTensor().reshape(r)},t}(T.KernelBackend);function Kw(n){for(var t=new Float32Array(n.length),e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}var pB=function(n){var t=n.inputs.x,e=n.backend,i=new Float32Array(T.util.sizeFromShape(t.shape));if(t.dtype!=="complex64"){var r=e.data.get(t.dataId).values;i=Kw(r)}else for(var a=e.data.get(t.dataId),s=a.complexTensorInfos.real,o=a.complexTensorInfos.imag,l=e.data.get(s.dataId).values,u=e.data.get(o.dataId).values,c=0;c<l.length;c++){var h=l[c],d=u[c];i[c]=Math.hypot(h,d)}return e.makeOutput(i,t.shape,"float32")},fB={kernelName:T.Abs,backendName:"cpu",kernelFunc:pB};function lr(n){return function(t,e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=T.util.computeStrides(s),u=T.util.sizeFromShape(s),c=T.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=T.util.computeStrides(t),f=T.util.computeStrides(e),m=T.backend_util.getBroadcastDims(t,s),g=T.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(var v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(var b=function(w){var S=T.util.indexToLoc(w,o,l),L=S.slice(-h);m.forEach(function(D){return L[D]=0});var x=T.util.locToIndex(L,h,p),C=S.slice(-d);g.forEach(function(D){return C[D]=0});var R=T.util.locToIndex(C,d,f);c[w]=n(i[x],r[R])},v=0;v<c.length;++v)b(v);return[c,s]}}function In(n){var t=n.inputs,e=n.backend,i=t.real,r=t.imag,a=e.data.get(i.dataId).values,s=e.data.get(r.dataId).values,o=e.makeTensorInfo(i.shape,"complex64"),l=e.data.get(o.dataId);return l.complexTensorInfos={real:e.makeTensorInfo(i.shape,"float32",a),imag:e.makeTensorInfo(r.shape,"float32",s)},o}var mB={kernelName:T.Complex,backendName:"cpu",kernelFunc:In};function Xr(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var gB={kernelName:T.Identity,backendName:"cpu",kernelFunc:Xr};function $a(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.real,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var vB={kernelName:T.Real,backendName:"cpu",kernelFunc:$a};function Xa(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.dtype;if(a==="complex64"){if(r.dtype==="complex64")return Xr({inputs:{x:r},backend:e});var s=T.zeros(r.shape),o=Xa({inputs:{x:r},backend:e,attrs:{dtype:"float32"}}),l=In({inputs:{real:o,imag:s},backend:e});return s.dispose(),e.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){var u=$a({inputs:{input:r},backend:e}),l=Xa({inputs:{x:u},backend:e,attrs:{dtype:a}});return e.disposeIntermediateTensorInfo(u),l}if(!T.util.hasEncodingLoss(r.dtype,a)){var l=Xr({inputs:{x:r},backend:e});return{dataId:l.dataId,shape:l.shape,dtype:a}}if(a==="int32"){var c=e.data.get(r.dataId).values,h=Int32Array.from(c);return e.makeTensorInfo(r.shape,"int32",h)}if(a==="bool"){var d=e.data.get(r.dataId).values,p=T.util.toTypedArray([0],r.dtype),f=lr(function(v,b){return v!==b?1:0})(r.shape,[],d,p,"bool"),m=f[0],g=f[1];return e.makeTensorInfo(g,"bool",m)}throw new Error("Error in Cast: failed to cast "+r.dtype+" to "+a)}var yB={kernelName:T.Cast,backendName:"cpu",kernelFunc:Xa};function Jr(n,t,e,i){return e==null?function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;ae([l,u],n);var h=c.data.get(l.dataId).values,d=c.data.get(u.dataId).values,p=i||l.dtype,f=t(l.shape,u.shape,h,d,p),m=f[0],g=f[1];return c.makeTensorInfo(g,p,m)}:function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;if(l.dtype==="complex64"||u.dtype==="complex64"){var h=Xa({inputs:{x:l},backend:c,attrs:{dtype:"complex64"}}),d=c.data.get(h.dataId),p=d.complexTensorInfos.real,f=d.complexTensorInfos.imag,m=c.data.get(p.dataId).values,g=c.data.get(f.dataId).values,v=Xa({inputs:{x:u},backend:c,attrs:{dtype:"complex64"}}),b=c.data.get(v.dataId),w=b.complexTensorInfos.real,S=b.complexTensorInfos.imag,L=c.data.get(w.dataId).values,x=c.data.get(S.dataId).values,C=e(l.shape,u.shape,m,g,L,x),R=C[0],D=C[1],k=C[2],W=c.makeTensorInfo(k,"float32",R),F=c.makeTensorInfo(k,"float32",D),P=In({inputs:{real:W,imag:F},backend:c});return c.disposeIntermediateTensorInfo(h),c.disposeIntermediateTensorInfo(v),c.disposeIntermediateTensorInfo(W),c.disposeIntermediateTensorInfo(F),P}else{var H=c.data.get(l.dataId).values,_=c.data.get(u.dataId).values,K=i||l.dtype,j=t(l.shape,u.shape,H,_,K),q=j[0],k=j[1];return c.makeTensorInfo(k,K,q)}}}function Sd(n){return function(t,e,i,r,a,s){var o=T.backend_util.assertAndGetBroadcastShape(t,e),l=T.util.sizeFromShape(o),u=o.length,c=T.util.computeStrides(o),h=T.util.getTypedArrayFromDType("float32",l),d=T.util.getTypedArrayFromDType("float32",l),p=T.backend_util.getBroadcastDims(t,o),f=T.backend_util.getBroadcastDims(e,o),m=T.backend_util.mergeRealAndImagArrays(i,r),g=T.backend_util.mergeRealAndImagArrays(a,s),v=t.length,b=T.util.computeStrides(t),w=e.length,S=T.util.computeStrides(e);if(p.length+f.length===0)for(var L=0;L<h.length;L++){var x=L%m.length,C=L%g.length,R=n(m[x*2],m[x*2+1],g[C*2],g[C*2+1]);h[L]=R.real,d[L]=R.imag}else for(var D=function(k){var W=T.util.indexToLoc(k,u,c),F=W.slice(-v);p.forEach(function(j){return F[j]=0});var P=T.util.locToIndex(F,v,b),H=W.slice(-w);f.forEach(function(j){return H[j]=0});var _=T.util.locToIndex(H,w,S),K=n(m[P*2],m[P*2+1],g[_*2],g[_*2+1]);h[k]=K.real,d[k]=K.imag},L=0;L<h.length;L++)D(L);return[h,d,o]}}var jw=lr(function(n,t){return n+t}),bB=Sd(function(n,t,e,i){return{real:n+e,imag:t+i}}),$w=Jr(T.Add,jw,bB),wB={kernelName:T.Add,backendName:"cpu",kernelFunc:$w};function Zr(n){return function(t,e,i){for(var r=T.util.getTypedArrayFromDType(e,t.length),a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}function $e(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");for(var l=s,u=l.data.get(o.dataId).values,c=T.util.sizeFromShape(o.shape),h=e||o.dtype,d=T.util.getArrayFromDType(h,c),p=0;p<c;++p)d[p]=t(u[p],a);return l.makeTensorInfo(o.shape,h,d)}}function Qr(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");var l=s,u=l.data.get(o.dataId).values,c=e||o.dtype,h=t(u,c,a);return l.makeTensorInfo(o.shape,c,h)}}var Xw=Zr(function(n){return Math.ceil(n)}),SB=Qr(T.Ceil,Xw),LB={kernelName:T.Ceil,backendName:"cpu",kernelFunc:SB};var Jw=Zr(function(n){return Math.exp(n)}),IB=Qr(T.Exp,Jw),AB={kernelName:T.Exp,backendName:"cpu",kernelFunc:IB};var Zw=Zr(function(n){return Math.expm1(n)}),TB=Qr(T.Expm1,Zw),NB={kernelName:T.Expm1,backendName:"cpu",kernelFunc:TB};var Qw=Zr(function(n){return Math.floor(n)}),xB=Qr(T.Floor,Qw),CB={kernelName:T.Floor,backendName:"cpu",kernelFunc:xB};var e0=Zr(function(n){return Math.log(n)}),RB=Qr(T.Log,e0),OB={kernelName:T.Log,backendName:"cpu",kernelFunc:RB};function t0(n,t,e,i){for(var r=T.util.getTypedArrayFromDType(i,T.util.sizeFromShape(e)),a=0;a<r.length;++a){for(var s=a*t,o=n[s],l=0;l<t;++l){var u=n[s+l];u>o&&(o=u)}r[a]=o}return r}var n0=lr(function(n,t){return n*t}),EB=Sd(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),i0=Jr(T.Multiply,n0,EB),DB={kernelName:T.Multiply,backendName:"cpu",kernelFunc:i0};var r0=Zr(function(n){return 1/Math.sqrt(n)}),kB=Qr(T.Rsqrt,r0),FB={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:kB};function a0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;c<s;++c){var h=e.length,d=T.util.computeStrides(e),p=T.util.indexToLoc(c,h,d),f=p.map(function(g,v){return g+t[v]}),m=T.util.locToIndex(f,i.length,o);u[c]=n[m]}return u}function Ld(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.begin,s=i.size;ae(r,"slice");var o=T.slice_util.parseSliceParams(r,a,s),l=o[0],u=o[1];T.slice_util.assertParamsValid(r,l,u);var c=e.data.get(r.dataId).values,h=a0(c,l,u,r.shape,r.dtype);return e.makeTensorInfo(u,r.dtype,h)}var WB={kernelName:T.Slice,backendName:"cpu",kernelFunc:Ld};var s0=lr(function(n,t){return n-t}),UB=Sd(function(n,t,e,i){return{real:n-e,imag:t-i}}),o0=Jr(T.Sub,s0,UB),BB={kernelName:T.Sub,backendName:"cpu",kernelFunc:o0};function Id(n,t,e,i,r){for(var a=t.length,s=T.util.sizeFromShape(t),o=T.util.computeStrides(t),l=T.util.computeStrides(r),u=T.util.getTypedArrayFromDType(e,T.util.sizeFromShape(r)),c=0;c<s;++c){for(var h=T.util.indexToLoc(c,a,o),d=new Array(h.length),p=0;p<d.length;p++)d[p]=h[i[p]];var f=T.util.locToIndex(d,a,l);u[f]=n[c]}return u}function l0(n,t,e,i){for(var r=T.util.parseAxisParam(t,e)[0],a=[1,e[0],1],s=0;s<r;s++)a[0]*=e[s];a[1]=e[r];for(var s=r+1;s<e.length;s++)a[2]*=e[s];for(var o={},l=new Int32Array(e[r]),u=new T.TensorBuffer(a,i,n),c=[],h=a[0]===1&&a[2]===1,s=0;s<e[r];s++){var d=void 0;if(h)d=n[s].toString();else{for(var p=[],f=0;f<a[0];f++)for(var m=0;m<a[2];m++)p.push(u.get(f,s,m));d=p.join(",")}if(o[d]!==void 0)l[s]=o[d];else{var g=Object.keys(o).length;o[d]=g,l[s]=g,c.push(s)}}var v=a.slice();v[1]=Object.keys(o).length;var b=new T.TensorBuffer(v,i);c.forEach(function(S,L){for(var x=0;x<a[0];x++)for(var C=0;C<a[2];C++)b.set(u.get(x,S,C),x,L,C)});var w=e.slice();return w[r]=v[1],{outputValues:b.values,outputShape:w,indices:l}}var zB={__proto__:null,simpleAbsImpl:Kw,addImpl:jw,ceilImpl:Xw,expImpl:Jw,expm1Impl:Zw,floorImpl:Qw,logImpl:e0,maxImpl:t0,multiplyImpl:n0,rsqrtImpl:r0,sliceImpl:a0,subImpl:s0,transposeImpl:Id,uniqueImpl:l0};var PB="2.6.0";T.registerBackend("cpu",function(){return new Yw},1);var _B=$e(T.Acos,function(n){return Math.acos(n)}),MB={kernelName:T.Acos,backendName:"cpu",kernelFunc:_B};var HB=$e(T.Acosh,function(n){return Math.acosh(n)}),VB={kernelName:T.Acosh,backendName:"cpu",kernelFunc:HB};var qB=$e(T.Asin,function(n){return Math.asin(n)}),GB={kernelName:T.Asin,backendName:"cpu",kernelFunc:qB};var YB=$e(T.Asinh,function(n){return Math.asinh(n)}),KB={kernelName:T.Asinh,backendName:"cpu",kernelFunc:YB};var jB=$e(T.Atan,function(n){return Math.atan(n)}),$B={kernelName:T.Atan,backendName:"cpu",kernelFunc:jB};var XB=$e(T.Atanh,function(n){return Math.atanh(n)}),JB={kernelName:T.Atanh,backendName:"cpu",kernelFunc:XB};function Ad(n,t,e,i,r,a){for(var s=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,u=r.dilationWidth,c=r.effectiveFilterHeight,h=r.effectiveFilterWidth,d=r.padInfo.top,p=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=T.buffer(r.outShape,e),g=m.values,v=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],w=r.outShape[3],S=0;S<r.batchSize;++S)for(var L=S*v,x=S*i[0],C=0;C<r.inChannels;++C)for(var R=0;R<r.outHeight;++R)for(var D=R*s-d,k=Math.max(0,D),W=Math.min(r.inHeight,c+D),F=L+R*b,P=0;P<r.outWidth;++P){for(var H=P*o-p,_=Math.max(0,H),K=Math.min(r.inWidth,h+H),j=f,q=0,G=0,Z=k;Z<W;Z+=l){for(var X=x+Z*i[1],ee=_;ee<K;ee+=u){var ne=X+ee*i[2],ie=n[ne+C];a==="max"&&ie>j?j=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(j))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:j}return m}function u0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;g<i.batchSize;++g)for(var v=0;v<i.inChannels;++v)for(var b=0;b<i.outHeight;++b){for(var w=b*o-p,S=w;S<0;)S+=u;for(var L=Math.min(i.inHeight,h+w),x=0;x<i.outWidth;++x){for(var C=x*l-f,R=C;R<0;)R+=c;for(var D=Math.min(i.inWidth,d+C),k=Number.NEGATIVE_INFINITY,W=-1,F=S;F<L;F+=u)for(var P=F-w,H=R;H<D;H+=c){var _=H-C,K=m.get(g,F,H,v);K>k&&(k=K,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,x,v)}}return s}function ZB(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Ad(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var QB={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:ZB};function ez(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),x=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D<c.batchSize;++D)for(var k=0;k<c.inChannels;++k)for(var W=0;W<c.inHeight;++W)for(var F=0;F<c.inWidth;++F){for(var P=W-S,H=F-w,_=0,K=0;K<v;K+=m){var j=(P+K)/h;if(j<0||j>=c.outHeight||Math.floor(j)!==j)continue;for(var q=0;q<b;q+=g){var G=(H+q)/d;if(G<0||G>=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,j,G,k);_+=Z}}L.set(_*x,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var tz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:ez};function nz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,x=0,C=0,R=0;R<c.length;++R)m[R]=f[S++]+(c[R]-h[L++])*p[x++]/Math.sqrt(d[C++]+u),S>=g&&(S=0),L>=w&&(L=0),x>=v&&(x=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var iz={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:nz};var rz=$e(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n<e.clipValueMin?e.clipValueMin:n}),az={kernelName:T.ClipByValue,backendName:"cpu",kernelFunc:rz};function Eo(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.imag,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var sz={kernelName:T.Imag,backendName:"cpu",kernelFunc:Eo};function Ai(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=T.util.sizeFromShape(r.shape),o=T.util.inferFromImplicitShape(a,s),l=T.util.sizeFromShape(o);T.util.assert(s===l,function(){return"The new shape ("+o+") has "+l+" elements and the old "+("shape ("+r.shape+") has "+s+" elements. The new shape and old ")+"shape must have the same number of elements."}),e.incRef(r.dataId);var u=e.data.get(r.dataId);if(u.complexTensorInfos!=null){var c=u.complexTensorInfos.real,h=u.complexTensorInfos.imag;c.shape=o,h.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var oz={kernelName:T.Reshape,backendName:"cpu",kernelFunc:Ai};function Ja(n){var t=n.inputs,e=n.backend,i=n.attrs,r=i.axis,a=T.util.parseAxisParam(r,t[0].shape)[0],s=T.backend_util.computeOutShape(t.map(function(S){return S.shape}),a);if(T.util.sizeFromShape(s)===0)return e.makeTensorInfo(s,t[0].dtype,[]);var o=t.filter(function(S){return T.util.sizeFromShape(S.shape)>0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return $a({inputs:{input:S},backend:e})}),c=o.map(function(S){return Eo({inputs:{input:S},backend:e})}),h=Ja({inputs:u,backend:e,attrs:{axis:r}}),d=Ja({inputs:c,backend:e,attrs:{axis:r}}),p=In({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),x=[-1,L];return Ai({inputs:{x:S},backend:e,attrs:{shape:x}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,x=T.util.sizeFromShape(S.shape);m.set(L,g),g+=x})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,x=0,C=0;C<S.shape[0];++C)for(var R=C*s[1]+v,D=0;D<S.shape[1];++D)m[R+D]=L[x++];v+=S.shape[1]})}var b=T.backend_util.computeOutShape(o.map(function(S){return S.shape}),a),w=e.makeTensorInfo(b,t[0].dtype,m);return f.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),w}var lz={kernelName:T.Concat,backendName:"cpu",kernelFunc:Ja};var uz=$e(T.Cos,function(n){return Math.cos(n)}),cz={kernelName:T.Cos,backendName:"cpu",kernelFunc:uz};var hz=$e(T.Cosh,function(n){return Math.cosh(n)}),dz={kernelName:T.Cosh,backendName:"cpu",kernelFunc:hz};var pz={kernelName:T.Dilation2D,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=i,l=o.strides,u=o.pad,c=o.dilations,h=e,d=h.data.get(a.dataId).values,p=a.shape.length,f=h.data.get(s.dataId).values,m=s.shape.length,g=T.backend_util.computeDilation2DInfo(a.shape,s.shape,l,u,"NHWC",c),v=g.batchSize,b=g.inHeight,w=g.inWidth,S=g.inChannels,L=g.outHeight,x=g.outWidth,C=g.padInfo,R=g.strideHeight,D=g.strideWidth,k=g.filterHeight,W=g.filterWidth,F=g.dilationHeight,P=g.dilationWidth,H=g.outShape,_=T.util.sizeFromShape(H),K=H.length,j=T.util.getArrayFromDType(a.dtype,_),q=0;q<v;++q)for(var G=0;G<L;++G)for(var Z=G*R-C.top,X=0;X<x;++X)for(var ee=X*D-C.left,ne=0;ne<S;++ne){for(var ie=Number.MIN_SAFE_INTEGER,te=0;te<k;++te){var re=Z+te*F;if(re>=0&&re<b)for(var le=0;le<W;++le){var he=ee+le*P;if(he>=0&&he<w){var ye=T.util.locToIndex([q,re,he,ne],p,T.util.computeStrides(a.shape)),Oe=T.util.locToIndex([te,le,ne],m,T.util.computeStrides(s.shape)),Fe=d[ye]+f[Oe];Fe>ie&&(ie=Fe)}}}var _e=T.util.locToIndex([q,G,X,ne],K,T.util.computeStrides(H));j[_e]=ie}var Me=h.write(T.util.toTypedArray(j,a.dtype),H,a.dtype);return{dataId:Me,shape:H,dtype:a.dtype}}};var fz={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=0,ie=0,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var mz={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=q<0?0:q,ie=Z<0?0:Z,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=re,ie=he)}}}_[K][ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var gz=lr(function(n,t){return n/t}),vz=Jr(T.Div,gz),Td={kernelName:T.Div,backendName:"cpu",kernelFunc:vz};var yz=$e(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),bz={kernelName:T.Elu,backendName:"cpu",kernelFunc:yz};var wz=T.backend_util.ERF_P,Sz=T.backend_util.ERF_A1,Lz=T.backend_util.ERF_A2,Iz=T.backend_util.ERF_A3,Az=T.backend_util.ERF_A4,Tz=T.backend_util.ERF_A5,Nz=$e(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+wz*e);return t*(1-((((Tz*i+Az)*i+Iz)*i+Lz)*i+Sz)*i*Math.exp(-e*e))}),xz={kernelName:T.Erf,backendName:"cpu",kernelFunc:Nz};function c0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p<r;p++){for(var f=Ld({inputs:{x:o},backend:e,attrs:{begin:[p,0],size:[1,a]}}),m=Ld({inputs:{x:l},backend:e,attrs:{begin:[p,0],size:[1,a]}}),g=In({inputs:{real:f,imag:m},backend:e}),v=Cz(g,t,e),b=v.real,w=v.imag,S=T.backend_util.mergeRealAndImagArrays(b,w),L=0;L<a;L++){var x=T.backend_util.getComplexWithIndex(S,L);h[p*a+L]=x.real,d[p*a+L]=x.imag}e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(g)}var C=e.makeTensorInfo(u,"float32",h),R=e.makeTensorInfo(u,"float32",d),D=In({inputs:{real:C,imag:R},backend:e});return e.disposeIntermediateTensorInfo(C),e.disposeIntermediateTensorInfo(R),D}function Cz(n,t,e){var i=T.util.sizeFromShape(n.shape),r=e.data.get(n.dataId),a=e.data.get(r.complexTensorInfos.real.dataId).values,s=e.data.get(r.complexTensorInfos.imag.dataId).values;if(Rz(i)){var o=Nd(a,s,i,t,e),l=[n.shape[0],n.shape[1]];if(t){var u=e.makeTensorInfo(l,"float32",o.real),c=e.makeTensorInfo(l,"float32",o.imag),h=e.makeTensorInfo([],"float32",T.util.createScalarValue(i,"float32")),d=Xr({inputs:{x:h},backend:e}),p=Td.kernelFunc({inputs:{a:u,b:h},backend:e}),f=Td.kernelFunc({inputs:{a:c,b:d},backend:e}),m=e.data.get(p.dataId).values,g=e.data.get(f.dataId).values;return e.disposeIntermediateTensorInfo(u),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return o}else{var v=T.backend_util.mergeRealAndImagArrays(a,s),b=Oz(v,i,t);return T.backend_util.splitRealAndImagArrays(b)}}function Rz(n){return(n&n-1)===0}function Nd(n,t,e,i,r){if(e===1)return{real:n,imag:t};var a=T.backend_util.mergeRealAndImagArrays(n,t),s=e/2,o=T.backend_util.complexWithEvenIndex(a),l=o.real,u=o.imag,c=[l.length],h=r.makeTensorInfo(c,"float32",l),d=r.makeTensorInfo(c,"float32",u),p=In({inputs:{real:h,imag:d},backend:r}),f=T.backend_util.complexWithOddIndex(a),m=f.real,g=f.imag,v=[m.length],b=r.makeTensorInfo(v,"float32",m),w=r.makeTensorInfo(v,"float32",g),S=In({inputs:{real:b,imag:w},backend:r}),L=Nd(l,u,s,i,r),x=L.real,C=L.imag,R=[x.length],D=r.makeTensorInfo(R,"float32",x),k=r.makeTensorInfo(R,"float32",C),W=In({inputs:{real:D,imag:k},backend:r}),F=Nd(m,g,s,i,r),P=F.real,H=F.imag,_=[P.length],K=r.makeTensorInfo(_,"float32",P),j=r.makeTensorInfo(_,"float32",H),q=In({inputs:{real:K,imag:j},backend:r}),G=T.backend_util.exponents(e,i),Z=[G.real.length],X=r.makeTensorInfo(Z,"float32",G.real),ee=r.makeTensorInfo(Z,"float32",G.imag),ne=In({inputs:{real:X,imag:ee},backend:r}),ie=i0({inputs:{a:ne,b:q},backend:r}),te=$w({inputs:{a:W,b:ie},backend:r}),re=o0({inputs:{a:W,b:ie},backend:r}),le=$a({inputs:{input:te},backend:r}),he=$a({inputs:{input:re},backend:r}),ye=Eo({inputs:{input:te},backend:r}),Oe=Eo({inputs:{input:re},backend:r}),Fe=Ja({inputs:[le,he],backend:r,attrs:{axis:0}}),_e=Ja({inputs:[ye,Oe],backend:r,attrs:{axis:0}}),Me=r.data.get(Fe.dataId).values,ht=r.data.get(_e.dataId).values;return r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(w),r.disposeIntermediateTensorInfo(S),r.disposeIntermediateTensorInfo(D),r.disposeIntermediateTensorInfo(k),r.disposeIntermediateTensorInfo(W),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(q),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(ie),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(re),r.disposeIntermediateTensorInfo(le),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(he),r.disposeIntermediateTensorInfo(Oe),r.disposeIntermediateTensorInfo(Fe),r.disposeIntermediateTensorInfo(_e),{real:Me,imag:ht}}function Oz(n,t,e){for(var i=new Float32Array(t*2),r=0;r<t;r++){for(var a=0,s=0,o=0;o<t;o++){var l=T.backend_util.exponent(r*o,t,e),u=T.backend_util.getComplexWithIndex(n,o);a+=u.real*l.real-u.imag*l.imag,s+=u.real*l.imag+u.imag*l.real}e&&(a/=t,s/=t),T.backend_util.assignToTypedArray(i,a,s,r)}return i}function Ez(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!1,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Dz={kernelName:T.FFT,backendName:"cpu",kernelFunc:Ez};var kz={kernelName:T.FlipLeftRight,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=i,s=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),o=r.shape,l=o[0],u=o[1],c=o[2],h=o[3],d=a.data.get(r.dataId).values,p=0;p<l;p++)for(var f=p*c*u*h,m=0;m<u;m++)for(var g=m*(c*h),v=0;v<c;v++)for(var b=v*h,w=0;w<h;w++){var S=[l,m,v,w],L=S[2],x=Math.round(c-L),C=f+g+b+w,R=d[C];if(x>=0&&x<c){var D=x*h,k=f+g+D+w;R=d[k]}s[C]=R}var W=a.write(s,r.shape,r.dtype);return{dataId:W,shape:r.shape,dtype:r.dtype}}};function Fz(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!0,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Wz={kernelName:T.IFFT,backendName:"cpu",kernelFunc:Fz};var Uz=$e(T.IsFinite,function(n){return Number.isFinite(n)?1:0},"bool"),Bz={kernelName:T.IsFinite,backendName:"cpu",kernelFunc:Uz};var zz=$e(T.IsInf,function(n){return Math.abs(n)===Infinity?1:0},"bool"),Pz={kernelName:T.IsInf,backendName:"cpu",kernelFunc:zz};var _z=$e(T.IsNan,function(n){return Number.isNaN(n)?1:0},"bool"),Mz={kernelName:T.IsNan,backendName:"cpu",kernelFunc:_z};var Hz=$e(T.Log1p,function(n){return Math.log1p(n)}),Vz={kernelName:T.Log1p,backendName:"cpu",kernelFunc:Hz};var qz=$e(T.LogicalNot,function(n){return n?0:1},"bool"),Gz={kernelName:T.LogicalNot,backendName:"cpu",kernelFunc:qz};var Yz={kernelName:T.Max,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape,c=u.length,h=T.util.parseAxisParam(s,u),d=h,p=T.backend_util.getAxesPermutation(d,c),f=l.data.get(r.dataId).values;if(p!=null){for(var m=new Array(c),g=0;g<m.length;g++)m[g]=u[p[g]];f=Id(f,u,r.dtype,p,m),d=T.backend_util.getInnerMostAxes(d.length,c),u=m}ae(r,"max"),T.backend_util.assertAxesAreInnerMostDims("max",d,c);var v=T.backend_util.computeOutAndReduceShapes(u,d),b=v[0],w=v[1],S=T.util.sizeFromShape(w),L=t0(f,S,b,r.dtype),x=l.write(L,b,r.dtype),C=b;if(o){var m=T.backend_util.expandShapeToKeepDim(b,h);C=m}return{dataId:x,shape:C,dtype:r.dtype}}};function Kz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Ad(d,r.shape,r.dtype,p,c,"max");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var jz={kernelName:T.MaxPool,backendName:"cpu",kernelFunc:Kz};function $z(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ae([a,s],"maxPoolBackprop");for(var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=T.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=e.data.get(o.dataId).values,f=T.buffer(d.outShape,o.dtype,u0(p,o.shape,o.dtype,d).values),m=d.strideHeight,g=d.strideWidth,v=d.dilationHeight,b=d.dilationWidth,w=d.effectiveFilterHeight,S=d.effectiveFilterWidth,L=S-1-d.padInfo.left,x=w-1-d.padInfo.top,C=T.buffer(o.shape,"float32"),R=e.data.get(r.dataId).values,D=T.buffer(r.shape,"float32",R),k=0;k<d.batchSize;++k)for(var W=0;W<d.inChannels;++W)for(var F=0;F<d.inHeight;++F)for(var P=0;P<d.inWidth;++P){for(var H=F-x,_=P-L,K=0,j=0;j<w;j+=v){var q=(H+j)/m;if(q<0||q>=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<S;G+=b){var Z=(_+G)/g;if(Z<0||Z>=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=j*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);K+=ie*ne}}C.set(K,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var Xz={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:$z};function Jz(n,t,e,i,r){var a=T.util.computeStrides(t),s=Ad(n,t,e,a,r,"max"),o=u0(n,t,e,r,!0,i);return[s.values,o.values]}var Zz={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=Jz(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var Qz=T.kernel_impls.nonMaxSuppressionV4Impl,eP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=Qz(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var tP=T.kernel_impls.nonMaxSuppressionV5Impl,nP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=tP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var iP=lr(function(n,t){return n!==t?1:0}),rP=Jr(T.NotEqual,iP,null,"bool"),aP={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:rP};function sP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(x,C){return x[0]+r.shape[C]+x[1]}),l=a.map(function(x){return x[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v<c;v++){var b=T.util.indexToLoc(v,h,d),w=b.map(function(x,C){return x+l[C]}),S=T.util.locToIndex(w,f,m);g[S]=u[v]}var L=e.write(g,o,r.dtype);return{dataId:L,shape:o,dtype:r.dtype}}var h0={kernelName:T.PadV2,backendName:"cpu",kernelFunc:sP};var oP=$e(T.Reciprocal,function(n){return 1/n}),lP={kernelName:T.Reciprocal,backendName:"cpu",kernelFunc:oP};var uP={kernelName:T.RotateWithOffset,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),h=r.shape,d=h[0],p=h[1],f=h[2],m=h[3],g=T.backend_util.getImageCenter(l,p,f),v=g[0],b=g[1],w=255,S=Math.sin(s),L=Math.cos(s),x=u.data.get(r.dataId).values,C=0;C<d;C++)for(var R=C*f*p*m,D=0;D<p;D++)for(var k=D*(f*m),W=0;W<f;W++)for(var F=W*m,P=0;P<m;P++){var H=[d,D,W,P],_=H[2],K=H[1],j=(_-v)*L-(K-b)*S,q=(_-v)*S+(K-b)*L;j=Math.round(j+v),q=Math.round(q+b);var G=o;if(typeof o!="number"&&(P===3?G=w:G=o[P]),j>=0&&j<f&&q>=0&&q<p){var Z=q*(f*m),X=j*m,ee=R+Z+X+P;G=x[ee]}var ne=R+k+F+P;c[ne]=G}var ie=u.write(c,r.shape,r.dtype);return{dataId:ie,shape:r.shape,dtype:r.dtype}}};var cP=$e(T.Round,function(n){var t=Math.floor(n);return n-t<.5?Math.floor(n):n-t>.5?Math.ceil(n):t%2===0?t:t+1}),hP={kernelName:T.Round,backendName:"cpu",kernelFunc:cP};var dP=T.backend_util.SELU_SCALEALPHA,pP=T.backend_util.SELU_SCALE,fP=$e(T.Selu,function(n){return n>=0?pP*n:dP*(Math.exp(n)-1)}),mP={kernelName:T.Selu,backendName:"cpu",kernelFunc:fP};var gP=$e(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),vP={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:gP};var yP=$e(T.Sign,function(n){return n<0?-1:n>0?1:0}),bP={kernelName:T.Sign,backendName:"cpu",kernelFunc:yP};var wP=$e(T.Sin,function(n){return Math.sin(n)}),SP={kernelName:T.Sin,backendName:"cpu",kernelFunc:wP};var LP=$e(T.Sinh,function(n){return Math.sinh(n)}),IP={kernelName:T.Sinh,backendName:"cpu",kernelFunc:LP};var AP=11920928955078125e-23,d0=Math.log(AP)+2,TP=$e(T.Softplus,function(n){var t=n>-d0,e=n<d0,i=Math.exp(n),r;return e?r=i:t?r=n:r=Math.log(1+i),r}),NP={kernelName:T.Softplus,backendName:"cpu",kernelFunc:TP};function p0(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm;ae(r,"transpose");for(var s=r.shape.length,o=new Array(s),l=0;l<o.length;l++)o[l]=r.shape[a[l]];var u=i.data.get(r.dataId).values,c=Id(u,r.shape,r.dtype,a,o),h=i.write(c,o,r.dtype);return{dataId:h,shape:o,dtype:r.dtype}}var xP={kernelName:T.Transpose,backendName:"cpu",kernelFunc:p0};function CP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.blockShape,s=i.paddings;ae([r],"spaceToBatchND");var o=T.util.sizeFromShape(a),l=[[0,0]];l.push.apply(l,s);for(var u=1+a.length;u<r.shape.length;++u)l.push([0,0]);var c=h0.kernelFunc({inputs:{x:r},backend:e,attrs:{paddings:l,constantValue:0}}),h=T.backend_util.getReshaped(c.shape,a,o,!1),d=T.backend_util.getPermuted(h.length,a.length,!1),p=T.backend_util.getReshapedPermuted(c.shape,a,o,!1),f={x:c},m={shape:h},g=Ai({inputs:f,backend:e,attrs:m}),v={x:g},b={perm:d},w=p0({inputs:v,backend:e,attrs:b}),S={x:w},L={shape:p},x=Ai({inputs:S,backend:e,attrs:L});return e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(w),x}var RP={kernelName:T.SpaceToBatchND,backendName:"cpu",kernelFunc:CP};var OP=$e(T.Sqrt,function(n){return Math.sqrt(n)}),EP={kernelName:T.Sqrt,backendName:"cpu",kernelFunc:OP};var DP={kernelName:T.Square,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.x,r=e;ae(i,"square");for(var a=r.data.get(i.dataId).values,s=new Float32Array(a.length),o=0;o<a.length;++o){var l=a[o];s[o]=l*l}var u=r.write(s,i.shape,i.dtype);return{dataId:u,shape:i.shape,dtype:i.dtype}}};var kP=lr(function(n,t){var e=n-t;return e*e}),FP=Jr(T.SquaredDifference,kP),WP={kernelName:T.SquaredDifference,backendName:"cpu",kernelFunc:FP};var UP=$e(T.Step,function(n,t){var e=t;return isNaN(n)?NaN:n>0?1:e.alpha}),BP={kernelName:T.Step,backendName:"cpu",kernelFunc:UP};var zP=$e(T.Tan,function(n){return Math.tan(n)}),PP={kernelName:T.Tan,backendName:"cpu",kernelFunc:zP};var _P=$e(T.Tanh,function(n){return Math.tanh(n)}),MP={kernelName:T.Tanh,backendName:"cpu",kernelFunc:_P};function HP(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=l0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var VP={kernelName:T.Unique,backendName:"cpu",kernelFunc:HP};var qP=[fB,MB,VB,wB,GB,KB,$B,JB,QB,tz,iz,yB,LB,az,mB,lz,cz,dz,pz,mz,fz,Td,bz,xz,AB,NB,Dz,kz,CB,gB,Wz,sz,Bz,Pz,Mz,OB,Vz,Gz,jz,Xz,Zz,Yz,DB,eP,nP,aP,h0,vB,lP,oz,uP,hP,FB,mP,vP,bP,SP,IP,WB,NP,RP,EP,DP,WP,BP,BB,PP,MP,xP,VP];for(var xd=0,f0=qP;xd<f0.length;xd++){var GP=f0[xd];T.registerKernel(GP)}ja.MathBackendCPU=Yw;ja.shared=zB;ja.version_cpu=PB});var JS=be(Pn=>{"use strict";Object.defineProperty(Pn,"__esModule",{value:!0});var N=Xi();var Cd=function(n,t){return Cd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Cd(n,t)};function YP(n,t){Cd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Do(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function ko(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var ur={},Rd={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function g0(n,t){ur[n]=t}function _n(n){if(!(n in ur)){var t=KP(n);if(t!==null)ur[n]=t;else return console.log("Could not get context for WebGL version",n),null}var e=ur[n];return e.isContextLost()?(delete ur[n],_n(n)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),ur[n])}function jP(n){if(typeof OffscreenCanvas!="undefined"&&n===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function KP(n){if(n!==1&&n!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");var t=jP(n);return t.addEventListener("webglcontextlost",function(e){e.preventDefault(),delete ur[n]},!1),n===1?t.getContext("webgl",Rd)||t.getContext("experimental-webgl",Rd):t.getContext("webgl2",Rd)}var Za;(function(n){n[n.DENSE=0]="DENSE",n[n.SHARED_BATCH=1]="SHARED_BATCH"})(Za||(Za={}));var rn;(function(n){n[n.RENDER=0]="RENDER",n[n.UPLOAD=1]="UPLOAD",n[n.PIXELS=2]="PIXELS",n[n.DOWNLOAD=3]="DOWNLOAD"})(rn||(rn={}));var Ct;(function(n){n[n.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",n[n.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",n[n.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",n[n.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",n[n.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Ct||(Ct={}));function Qa(n,t){return[t,n]}function $P(n,t){return n*t}function es(n){var t=N.util.sizeFromShape(n),e=Math.ceil(t/4);return N.util.sizeToSquarishShape(e)}function ea(n,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(n/2))]}function XP(n,t){var e=ea(n,t),i=e[0],r=e[1];return i*r*4}function Od(n,t){var e=n,i,r,a,s,o,l,u,c,h,d;return N.env().getNumber("WEBGL_VERSION")===2?(i=e.R32F,r=e.R16F,a=e.RGBA16F,s=e.RGBA32F,o=e.RED,u=4,c=1,h=e.HALF_FLOAT,d=e.FLOAT):(i=n.RGBA,r=n.RGBA,a=n.RGBA,s=e.RGBA,o=n.RGBA,u=4,c=4,h=t!=null?t.HALF_FLOAT_OES:null,d=n.FLOAT),l=n.RGBA,{internalFormatFloat:i,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:s,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:u,defaultNumChannels:c,textureTypeHalfFloat:h,textureTypeFloat:d}}function ce(n,t){var e=t();return N.env().getBool("DEBUG")&&JP(n),e}function JP(n){var t=n.getError();if(t!==n.NO_ERROR)throw new Error("WebGL Error: "+v0(n,t))}var ZP=596e-10,QP=65504;function y0(n){return!!(N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||n===0||ZP<Math.abs(n)&&Math.abs(n)<QP)}function v0(n,t){switch(t){case n.NO_ERROR:return"NO_ERROR";case n.INVALID_ENUM:return"INVALID_ENUM";case n.INVALID_VALUE:return"INVALID_VALUE";case n.INVALID_OPERATION:return"INVALID_OPERATION";case n.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case n.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case n.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return"Unknown error code "+t}}function ts(n,t){return ei(n,function(){return n.getExtension(t)},'Extension "'+t+'" not supported on this browser.')}function b0(n,t){var e=ei(n,function(){return n.createShader(n.VERTEX_SHADER)},"Unable to create vertex WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw console.log(n.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function w0(n,t){var e=ei(n,function(){return n.createShader(n.FRAGMENT_SHADER)},"Unable to create fragment WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw e_(t,n.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var t_=/ERROR: [0-9]+:([0-9]+):/g;function e_(n,t){var e=t_.exec(t);if(e==null){console.log("Couldn't parse line number in error: "+t),console.log(n);return}for(var i=+e[1],r=n.split(`
+============================`));var a={};return this.data.set(a,{values:e,dtype:r,refCount:1}),a},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return{dataId:a,shape:e,dtype:i}},t.prototype.incRef=function(e){var i=this.data.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.data.has(e)){var i=this.data.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){this.data.set(e,{values:i,dtype:a,refCount:1})},t.prototype.numDataIds=function(){return this.data.numDataIds()},t.prototype.read=function(e){return qw(this,void 0,void 0,function(){return Gw(this,function(i){return[2,this.readSync(e)]})})},t.prototype.readSync=function(e){var i=this.data.get(e),r=i.dtype,a=i.complexTensorInfos;if(r==="complex64"){var s=this.readSync(a.real.dataId),o=this.readSync(a.imag.dataId);return T.backend_util.mergeRealAndImagArrays(s,o)}return this.data.get(e).values},t.prototype.bufferSync=function(e){var i=this.readSync(e.dataId),r=i;if(e.dtype==="string")try{r=i.map(function(a){return T.util.decodeString(a)})}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return T.buffer(e.shape,e.dtype,r)},t.prototype.makeOutput=function(e,i,r){var a=this.write(e,i,r);return T.engine().makeTensorFromDataId(a,i,r,this)},t.prototype.disposeData=function(e){if(this.data.has(e)){var i=this.data.get(e).complexTensorInfos;i!=null&&(this.disposeData(i.real.dataId),this.disposeData(i.imag.dataId)),this.data.delete(e)}},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.data.has(i)){var r=this.data.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.time=function(e){return qw(this,void 0,void 0,function(){var i,r;return Gw(this,function(a){return i=T.util.now(),e(),r=T.util.now()-i,[2,{kernelMs:r}]})})},t.prototype.memory=function(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}},t.prototype.stridedSlice=function(e,i,r,a){ae(e,"stridedSlice");var s=T.slice_util.computeOutShape(i,r,a);if(s.some(function(p){return p===0}))return T.tensor([],s);for(var o=T.buffer(s,e.dtype),l=this.bufferSync(e),u=0;u<o.size;u++){for(var c=o.indexToLoc(u),h=new Array(c.length),d=0;d<h.length;d++)h[d]=c[d]*a[d]+i[d];o.set.apply(o,[l.get.apply(l,h)].concat(c))}return o.toTensor()},t.prototype.diag=function(e){for(var i=this.readSync(e.dataId),r=T.buffer([e.size,e.size],e.dtype),a=r.values,s=0;s<i.length;s++)a[s*e.size+s]=i[s];return r.toTensor()},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=T.slice(e,l,u).reshape(a);return c},t.prototype.reverse=function(e,i){ae(e,"reverse");for(var r=T.buffer(e.shape,e.dtype),a=this.bufferSync(e),s=function(l){var u=r.indexToLoc(l),c=u.slice();i.forEach(function(h){return c[h]=e.shape[h]-1-c[h]}),r.set.apply(r,[a.get.apply(a,c)].concat(u))},o=0;o<r.size;o++)s(o);return r.toTensor()},t.prototype.neg=function(e){return ae(e,"neg"),T.mul(T.scalar(-1),e)},t.prototype.addN=function(e){var i=this;ae(e,"addN");for(var r=e.map(function(c){return i.readSync(c.dataId)}),a=T.buffer(e[0].shape,e[0].dtype),s=a.values,o=0;o<e.length;o++)for(var l=r[o],u=0;u<s.length;u++)s[u]+=l[u];return a.toTensor()},t.prototype.softmax=function(e,i){var r=T.util.parseAxisParam([i],e.shape),a=T.max(e,r),s=T.backend_util.expandShapeToKeepDim(a.shape,r),o=T.sub(e,a.reshape(s)),l=T.exp(o),u=this.sum(l,r).reshape(s);return T.div(l,u)},t.prototype.pow=function(e,i){return ae([e,i],"pow"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.pow(r,a)})},t.prototype.batchMatMul=function(e,i,r,a){ae([e,i],"matMul");for(var s=r?e.shape[1]:e.shape[2],o=r?e.shape[2]:e.shape[1],l=a?i.shape[1]:i.shape[2],u=e.shape[0],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=r?[e.strides[0],1,e.strides[1]]:[e.strides[0],e.strides[1],1],p=d[0],f=d[1],m=d[2],g=a?[1,i.strides[1],i.strides[0]]:[i.strides[1],1,i.strides[0]],v=g[0],b=g[1],w=g[2],S=o*l,L=T.buffer([u,o,l],e.dtype),x=L.values,C=this.blockSize,R=0;R<u;R++)for(var D=0;D<o;D+=C)for(var k=0;k<l;k+=C)for(var W=0;W<s;W+=C)for(var F=Math.min(D+C,o),P=Math.min(k+C,l),H=Math.min(W+C,s),_=D;_<F;_++)for(var K=k;K<P;K++){for(var j=0,q=W;q<H;q++)j+=c[R*p+_*f+q*m]*h[q*v+K*b+R*w];x[R*S+(_*l+K)]+=j}return L.toTensor()},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=this.batchMatMul(i,r,a,s);return o&&(c=T.add(c,o)),l&&(c=Sd(this,c,l,u)),c},t.prototype.floorDiv=function(e,i){ae([e,i],"floorDiv");var r=function(s,o){return Math.floor(s/o)},a="int32";return this.broadcastedBinaryOp(e,i,a,r)},t.prototype.sum=function(e,i){ae(e,"sum"),T.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=0,m=0;m<u;++m)f+=h[p+m];c[d]=f}return l},t.prototype.prod=function(e,i){ae(e,"sum");for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=1,m=0;m<u;++m)f*=h[p+m];c[d]=f}return l},t.prototype.unsortedSegmentSum=function(e,i,r){ae(e,"unsortedSegmentSum");for(var a=[],s=e.rank-i.rank,o=0;o<s;++o)i=i.expandDims(o+1);for(var o=0;o<r;++o){var l=T.scalar(o,"int32"),u=T.equal(l,i).asType("float32"),c=u.mul(e).sum(0);a.push(c)}return T.stack(a)},t.prototype.argMin=function(e,i){ae(e,"argMin");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMin",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v<f&&(f=v,m=g)}c[d]=m}return l},t.prototype.argMax=function(e,i){ae(e,"argMax");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMax",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v>f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;d<u.length;d+=c)for(var p=0;p<c;p++){var f=h(d,p);if(p===0)l[f]=r?0:u[f];else{var m=h(d,p-1);l[f]=r?u[m]+l[m]:u[f]+l[m]}}return o},t.prototype.equal=function(e,i){return ae([e,i],"equal"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r===a?1:0})},t.prototype.notEqual=function(e,i){return ae([e,i],"notEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r!==a?1:0})},t.prototype.less=function(e,i){return ae([e,i],"less"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<a?1:0})},t.prototype.lessEqual=function(e,i){return ae([e,i],"lessEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<=a?1:0})},t.prototype.greater=function(e,i){return ae([e,i],"greater"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>a?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d<a.length;d++)for(var p=0;p<h;p++)a[d]===1?u[c++]=s[d]:u[c++]=o[d];return l},t.prototype.where=function(e){ae([e],"where");var i=this.readSync(e.dataId);return dB(e.shape,i)},t.prototype.topk=function(e,i,r){ae(e,"topk");var a=this.readSync(e.dataId);return hB(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){ae(e,"min"),T.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];m<p&&(p=m)}u[h]=p}return o},t.prototype.minimum=function(e,i){return ae([e,i],"minimum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.min(r,a)})},t.prototype.mod=function(e,i){return ae([e,i],"mod"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r%a;return r<0&&a<0||r>=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p&&m}u[h]=p}return o},t.prototype.any=function(e,i){ae(e,"any"),T.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p||m}u[h]=p}return o},t.prototype.squaredDifference=function(e,i){return ae([e,i],"squaredDifference"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r-a;return s*s})},t.prototype.linear=function(e){return e},t.prototype.relu=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.max(0,a[s]);return i},t.prototype.relu6=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.min(Math.max(0,a[s]),6);return i},t.prototype.prelu=function(e,i){return ae([e,i],"prelu"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return r<0?a*r:r})},t.prototype.eluDer=function(e,i){ae([e,i],"eluDer");for(var r=new Float32Array(i.size),a=this.readSync(i.dataId),s=this.readSync(e.dataId),o=0;o<a.length;++o){var l=a[o];l>=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Sd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),x=this.readSync(i.dataId),C=d.values,R=0;R<r.batchSize;++R)for(var D=R*p,k=R*v,W=0;W<r.outHeight;++W)for(var F=k+W*b,P=W*r.strideHeight-c,H=0;H<a;H++){var _=P+H*o;if(_<0||_>=r.inHeight)continue;for(var K=H*i.strides[0],j=D+_*f,q=0;q<r.outWidth;++q)for(var G=F+q*w,Z=q*r.strideWidth-u,X=0;X<s;X++){var ee=Z+X*l;if(ee<0||ee>=r.inWidth)continue;for(var ne=K+X*i.strides[1],ie=j+ee*m,te=ne,re=0;re<r.inChannels;++re){for(var le=L[ie+re*g],he=0;he<r.outChannels;++he)C[G+he*S]+=le*x[te+he];te+=r.outChannels}}}return d.toTensor()},t.prototype.conv3d=function(e,i,r){for(var a=r.filterDepth,s=r.filterHeight,o=r.filterWidth,l=r.dilationDepth,u=r.dilationHeight,c=r.dilationWidth,h=r.padInfo.front,d=r.padInfo.left,p=r.padInfo.top,f=T.buffer(r.outShape,e.dtype),m=this.readSync(e.dataId),g=this.readSync(i.dataId),v=f.values,b=0;b<r.batchSize;++b)for(var w=b*e.strides[0],S=b*f.strides[0],L=0;L<r.outDepth;++L)for(var x=S+L*f.strides[1],C=L*r.strideDepth-h,R=0;R<a;R++){var D=C+R*l;if(D<0||D>=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F<r.outHeight;++F)for(var P=x+F*f.strides[2],H=F*r.strideHeight-p,_=0;_<s;_++){var K=H+_*u;if(K<0||K>=r.inHeight)continue;for(var j=k+_*i.strides[1],q=W+K*e.strides[2],G=0;G<r.outWidth;++G)for(var Z=P+G*r.outChannels,X=G*r.strideWidth-d,ee=0;ee<o;ee++){var ne=X+ee*c;if(ne<0||ne>=r.inWidth)continue;for(var ie=j+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le<r.inChannels;++le){for(var he=m[te+le],ye=0;ye<r.outChannels;++ye)v[Z+ye]+=he*g[re+ye];re+=r.outChannels}}}}return f.toTensor()},t.prototype.conv2dDerInput=function(e,i,r){ae([e,i],"conv2dDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=this.readSync(e.dataId),l=this.readSync(i.dataId),u=i.strides,c=u[0],h=u[1],d=u[2],p=r.batchSize,f=r.filterHeight,m=r.filterWidth,g=r.inChannels,v=r.inHeight,b=r.inWidth,w=r.outChannels,S=r.outHeight,L=r.outWidth,x=r.strideHeight,C=r.strideWidth,R=r.dataFormat,D=f-1-r.padInfo.top,k=m-1-r.padInfo.left,W=R==="channelsLast",F=a.strides[0],P=W?a.strides[1]:a.strides[2],H=W?a.strides[2]:1,_=W?1:a.strides[1],K=e.strides[0],j=W?e.strides[1]:e.strides[2],q=W?e.strides[2]:1,G=W?1:e.strides[1],Z=0;Z<p;++Z)for(var X=0;X<g;++X)for(var ee=0;ee<v;++ee)for(var ne=ee-D,ie=Math.max(0,Math.ceil(ne/x)),te=Math.min(S,(f+ne)/x),re=0;re<b;++re){for(var le=re-k,he=Math.max(0,Math.ceil(le/C)),ye=Math.min(L,(m+le)/C),Oe=0,Fe=ie;Fe<te;++Fe)for(var _e=Fe*x-ne,Me=he;Me<ye;++Me)for(var ht=Me*C-le,St=K*Z+j*Fe+q*Me,Ge=c*(f-1-_e)+h*(m-1-ht)+d*X,Ue=0;Ue<w;++Ue){var me=o[St+G*Ue],Dt=l[Ge+Ue];Oe+=me*Dt}var lt=F*Z+P*ee+H*re+_*X;s[lt]=Oe}return a.toTensor()},t.prototype.conv3dDerInput=function(e,i,r){for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=o[3],d=this.readSync(e.dataId),p=e.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=r.batchSize,D=r.filterDepth,k=r.filterHeight,W=r.filterWidth,F=r.inChannels,P=r.inDepth,H=r.inHeight,_=r.inWidth,K=r.outChannels,j=r.outDepth,q=r.outHeight,G=r.outWidth,Z=r.strideDepth,X=r.strideHeight,ee=r.strideWidth,ne=D-1-r.padInfo.front,ie=k-1-r.padInfo.top,te=W-1-r.padInfo.left,re=0;re<R;++re)for(var le=0;le<F;++le)for(var he=0;he<P;++he)for(var ye=he-ne,Oe=Math.max(0,Math.ceil(ye/Z)),Fe=Math.min(j,(D+ye)/Z),_e=0;_e<H;++_e)for(var Me=_e-ie,ht=Math.max(0,Math.ceil(Me/X)),St=Math.min(q,(k+Me)/X),Ge=0;Ge<_;++Ge){for(var Ue=Ge-te,me=Math.max(0,Math.ceil(Ue/ee)),Dt=Math.min(G,(W+Ue)/ee),lt=0,dt=Oe;dt<Fe;++dt)for(var xn=dt*Z-ye,Rt=ht;Rt<St;++Rt)for(var Lt=Rt*X-Me,Ce=me;Ce<Dt;++Ce)for(var kt=Ce*ee-Ue,pt=f*re+m*dt+g*Rt+v*Ce,it=S*(D-1-xn)+L*(k-1-Lt)+x*(W-1-kt)+C*le,sn=0;sn<K;++sn){var hs=d[pt+sn],Bt=b[it+sn];lt+=hs*Bt}s[l*re+u*he+c*_e+h*Ge+le]=lt}return a.toTensor()},t.prototype.conv2dDerFilter=function(e,i,r){ae([e,i],"conv2dDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=r.dataFormat==="channelsLast",c=T.buffer(r.filterShape,"float32"),h=r.padInfo.left,d=r.padInfo.top,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((d-m)/a)),v=Math.min(r.outHeight,(r.inHeight+d-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((h-b)/s)),S=Math.min(r.outWidth,(r.inWidth+h-b)/s),L=0;L<r.inChannels;++L)for(var x=0;x<r.outChannels;++x){for(var C=0,R=0;R<r.batchSize;++R)for(var D=g;D<v;++D)for(var k=m+D*a-d,W=w;W<S;++W){var F=b+W*s-h;u?C+=p.get(R,k,F,L)*f.get(R,D,W,x):C+=p.get(R,L,k,F)*f.get(R,x,D,W)}c.set(C,m,b,L,x)}return c.toTensor()},t.prototype.conv3dDerFilter=function(e,i,r){for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=T.buffer(r.filterShape,"float32"),d=h.values,p=h.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=this.readSync(e.dataId),D=e.strides,k=D[0],W=D[1],F=D[2],P=D[3],H=r.padInfo.front,_=r.padInfo.left,K=r.padInfo.top,j=0;j<l;++j)for(var q=Math.max(0,Math.ceil((H-j)/a)),G=Math.min(r.outDepth,(r.inDepth+H-j)/a),Z=j*f,X=0;X<u;++X)for(var ee=Math.max(0,Math.ceil((K-X)/s)),ne=Math.min(r.outHeight,(r.inHeight+K-X)/s),ie=X*m+Z,te=0;te<c;++te)for(var re=Math.max(0,Math.ceil((_-te)/o)),le=Math.min(r.outWidth,(r.inWidth+_-te)/o),he=te*g+ie,ye=0;ye<r.inChannels;++ye)for(var Oe=ye*v+he,Fe=0;Fe<r.outChannels;++Fe){for(var _e=0,Me=0;Me<r.batchSize;++Me)for(var ht=Me*k,St=Me*S,Ge=q;Ge<G;++Ge)for(var Ue=j+Ge*a-H,me=Ue*W+ht,Dt=Ge*L+St,lt=ee;lt<ne;++lt)for(var dt=X+lt*s-K,xn=dt*F+me,Rt=lt*x+Dt,Lt=re;Lt<le;++Lt){var Ce=te+Lt*o-_,kt=Ce*P+xn,pt=Lt*C+Rt;_e+=R[kt+ye]*b[pt+Fe]}d[Oe+Fe]=_e}return h.toTensor()},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.depthwiseConv2D(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Sd(this,u,o,l)),u},t.prototype.depthwiseConv2D=function(e,i,r){ae([e,i],"depthwiseConv2D");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.outChannels/r.inChannels,d=T.buffer(r.outShape,e.dtype),p=this.readSync(e.dataId),f=this.readSync(i.dataId),m=d.values,g=0;g<r.batchSize;++g)for(var v=g*e.strides[0],b=g*d.strides[0],w=0;w<r.outHeight;++w)for(var S=b+w*d.strides[1],L=w*r.strideHeight-u,x=0;x<a;++x){var C=L+x*o;if(C<0||C>=r.inHeight)continue;for(var R=x*i.strides[0],D=v+C*e.strides[1],k=0;k<r.outWidth;++k)for(var W=S+k*d.strides[2],F=k*r.strideWidth-c,P=0;P<s;++P){var H=F+P*l;if(H<0||H>=r.inWidth)continue;for(var _=R+P*i.strides[1],K=D+H*r.inChannels,j=W,q=_,G=0;G<r.inChannels;++G){for(var Z=p[K+G],X=0;X<h;++X)m[j+X]+=Z*f[q+X];j+=h,q+=h}}}return d.toTensor()},t.prototype.depthwiseConv2DDerInput=function(e,i,r){ae([e,i],"depthwiseConv2DDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=this.readSync(e.dataId),d=e.strides,p=d[0],f=d[1],m=d[2],g=this.readSync(i.dataId),v=i.strides,b=v[0],w=v[1],S=v[2],L=r.batchSize,x=r.filterHeight,C=r.filterWidth,R=r.inChannels,D=r.inHeight,k=r.inWidth,W=r.outChannels,F=r.outHeight,P=r.outWidth,H=r.strideHeight,_=r.strideWidth,K=x-1-r.padInfo.top,j=C-1-r.padInfo.left,q=W/R,G=0;G<L;++G)for(var Z=0;Z<R;++Z)for(var X=0;X<D;++X)for(var ee=X-K,ne=Math.max(0,Math.ceil(ee/H)),ie=Math.min(F,(x+ee)/H),te=0;te<k;++te){for(var re=te-j,le=Math.max(0,Math.ceil(re/_)),he=Math.min(P,(C+re)/_),ye=0,Oe=ne;Oe<ie;++Oe)for(var Fe=Oe*H-ee,_e=le;_e<he;++_e)for(var Me=_e*_-re,ht=p*G+f*Oe+m*_e,St=b*(x-1-Fe)+w*(C-1-Me)+S*Z,Ge=0;Ge<q;++Ge){var Ue=Z*q+Ge,me=h[ht+Ue],Dt=g[St+Ge];ye+=me*Dt}s[l*G+u*X+c*te+Z]=ye}return a.toTensor()},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){ae([e,i],"depthwiseConv2DDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=T.buffer(r.filterShape,"float32"),c=r.padInfo.left,h=r.padInfo.top,d=r.outChannels/r.inChannels,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((h-m)/a)),v=Math.min(r.outHeight,(r.inHeight+h-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((c-b)/s)),S=Math.min(r.outWidth,(r.inWidth+c-b)/s),L=0;L<r.outChannels;++L){for(var x=Math.trunc(L/d),C=L%d,R=0,D=0;D<r.batchSize;++D)for(var k=g;k<v;++k)for(var W=m+k*a-h,F=w;F<S;++F){var P=b+F*s-c;R+=p.get(D,W,P,x)*f.get(D,k,F,L)}u.set(R,m,b,x,C)}return u.toTensor()},t.prototype.tile=function(e,i){return ae(e,"tile"),cB(this.bufferSync(e),i)},t.prototype.gather=function(e,i,r){ae([e,i],"gather");var a=e.shape.slice(),s=this.readSync(i.dataId);a[r]=s.length;for(var o=T.buffer(a,e.dtype),l=this.bufferSync(e),u=0;u<o.size;++u){var c=o.indexToLoc(u),h=c.slice();h[r]=s[c[r]];var d=l.locToIndex(h);o.values[u]=l.values[d]}return o.toTensor()},t.prototype.batchToSpaceND=function(e,i,r){ae([e],"batchToSpaceND");var a=i.reduce(function(h,d){return h*d}),s=T.backend_util.getReshaped(e.shape,i,a),o=T.backend_util.getPermuted(s.length,i.length),l=T.backend_util.getReshapedPermuted(e.shape,i,a),u=T.backend_util.getSliceBeginCoords(r,i.length),c=T.backend_util.getSliceSize(l,r,i.length);return T.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.pool3d=function(e,i,r){ae(e,"pool3d");for(var a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=r==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,b=this.readSync(e.dataId),w=T.buffer(i.outShape,e.dtype),S=w.values,L=i.outShape[1]*i.outShape[2]*i.outShape[3]*i.outShape[4],x=i.outShape[2]*i.outShape[3]*i.outShape[4],C=i.outShape[3]*i.outShape[4],R=i.outShape[4],D=0;D<i.batchSize;++D)for(var k=D*L,W=D*e.strides[0],F=0;F<i.inChannels;++F)for(var P=0;P<i.outDepth;++P){for(var H=P*a-f,_=H;_<0;)_+=l;for(var K=Math.min(i.inDepth,h+H),j=k+P*x,q=0;q<i.outHeight;++q){for(var G=q*s-m,Z=G;Z<0;)Z+=u;for(var X=Math.min(i.inHeight,d+G),ee=j+q*C,ne=0;ne<i.outWidth;++ne){for(var ie=ne*o-g,te=ie;te<0;)te+=c;for(var re=Math.min(i.inWidth,p+ie),le=ee+ne*R,he=v,ye=0,Oe=0,Fe=_;Fe<K;Fe+=l){for(var _e=W+Fe*e.strides[1],Me=Z;Me<X;Me+=u){for(var ht=_e+Me*e.strides[2],St=te;St<re;St+=c){var Ge=ht+St*e.strides[3],Ue=b[Ge+F];if(r==="max"&&Ue>he?he=Ue:r==="avg"&&(ye+=Ue,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),x=this.bufferSync(e),C=0;C<r.batchSize;++C)for(var R=0;R<r.inChannels;++R)for(var D=0;D<r.inDepth;++D)for(var k=0;k<r.inHeight;++k)for(var W=0;W<r.inWidth;++W){for(var F=D-v,P=k-w,H=W-b,_=0,K=0;K<f;K+=h){var j=(F+K)/a;if(j<0||j>=r.outDepth||Math.floor(j)!==j)continue;for(var q=0;q<m;q+=d){var G=(P+q)/s;if(G<0||G>=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z<g;Z+=p){var X=(H+Z)/o;if(X<0||X>=r.outWidth||Math.floor(X)!==X)continue;var ee=x.get(C,j,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b<i.batchSize;++b)for(var w=0;w<i.inChannels;++w)for(var S=0;S<i.outDepth;++S){for(var L=S*a-f,x=L;x<0;)x+=l;for(var C=Math.min(i.inDepth,h+L),R=0;R<i.outHeight;++R){for(var D=R*s-m,k=D;k<0;)k+=u;for(var W=Math.min(i.inHeight,d+D),F=0;F<i.outWidth;++F){for(var P=F*o-g,H=P;H<0;)H+=c;for(var _=Math.min(i.inWidth,p+P),K=Number.NEGATIVE_INFINITY,j=-1,q=x;q<C;q+=l)for(var G=q-L,Z=k;Z<W;Z+=u)for(var X=Z-D,ee=H;ee<_;ee+=c){var ne=ee-P,ie=v.get(b,q,Z,ee,w);ie>=K&&(K=ie,j=G*d*p+X*d+ne)}r.set(j,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),x=0;x<a.batchSize;++x)for(var C=0;C<a.inChannels;++C)for(var R=0;R<a.inDepth;++R)for(var D=0;D<a.inHeight;++D)for(var k=0;k<a.inWidth;++k){for(var W=R-g,F=D-b,P=k-v,H=0,_=0;_<p;_+=c){var K=(W+_)/o;if(K<0||K>=a.outDepth||Math.floor(K)!==K)continue;for(var j=0;j<f;j+=h){var q=(F+j)/l;if(q<0||q>=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<m;G+=d){var Z=(P+G)/u;if(Z<0||Z>=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(x,K,q,Z,C),ee=_*f*m+j*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(x,K,q,Z,C);H+=ie*ne}}}w.set(H,x,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b<o;b++)for(var w=0;w<i;w++)for(var S=g*w,L=Math.floor(S),x=S-L,C=Math.min(l-1,Math.ceil(S)),R=b*e.strides[0]+L*e.strides[1],D=b*e.strides[0]+C*e.strides[1],k=0;k<r;k++)for(var W=v*k,F=Math.floor(W),P=W-F,H=Math.min(u-1,Math.ceil(W)),_=R+F*e.strides[2],K=D+F*e.strides[2],j=R+H*e.strides[2],q=D+H*e.strides[2],G=0;G<c;G++){var Z=h[_+G],X=h[K+G],ee=h[j+G],ne=h[q+G],ie=Z+(ee-Z)*P,te=X+(ne-X)*P,re=ie+(te-ie)*x;d[m++]=re}return T.tensor(d,[o,i,r,c])},t.prototype.resizeBilinearBackprop=function(e,i,r){ae([e,i],"resizeBilinearBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=[r&&h>1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S<s;S++)for(var L=S*i.strides[0],x=0;x<h;x++)for(var C=x*g,R=Math.floor(C),D=Math.min(Math.ceil(C),o-1),k=L+R*i.strides[1],W=L+D*i.strides[1],F=C-R,P=1-F,H=0;H<d;H++)for(var _=H*v,K=Math.floor(_),j=Math.min(Math.ceil(_),l-1),q=_-K,G=1-q,Z=k+K*i.strides[2],X=k+j*i.strides[2],ee=W+K*i.strides[2],ne=W+j*i.strides[2],ie=P*G,te=P*q,re=F*G,le=F*q,he=0;he<u;he++){var ye=b[w++];p[Z+he]+=ye*ie,p[X+he]+=ye*te,p[ee+he]+=ye*re,p[ne+he]+=ye*le}return T.tensor4d(p,[s,l,o,u],i.dtype)},t.prototype.resizeNearestNeighbor=function(e,i,r,a){ae(e,"resizeNearestNeighbor");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(o*i*r*c),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b<o;b++)for(var w=b*e.strides[0],S=0;S<i;S++)for(var L=m*S,x=Math.min(l-1,a?Math.round(L):Math.floor(L)),C=w+x*e.strides[1],R=0;R<r;R++)for(var D=g*R,k=Math.min(u-1,a?Math.round(D):Math.floor(D)),W=C+k*e.strides[2],F=0;F<c;F++){var P=h[W+F];d[v++]=P}return T.tensor(d,[o,i,r,c],e.dtype)},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){ae([e,i],"resizeNearestNeighborBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=this.readSync(e.dataId),m=[r&&h>1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,x=Math.ceil(S)*2+2,C=0;C<s;C++)for(var R=C*i.strides[0],D=0;D<o;D++)for(var k=R+D*i.strides[1],W=Math.floor(D*w),F=Math.floor(W-L/2),P=0;P<l;P++)for(var H=k+P*i.strides[2],_=Math.floor(P*S),K=Math.floor(_-x/2),j=0;j<u;j++){for(var q=0,G=0;G<L;G++){var Z=G+F;if(Z<0||Z>=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie<x;ie++){var te=ie+K;if(te<0||te>=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+j])}}p[H+j]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p<c;p++){var f=d(p),m=u[p]*Math.pow(r+a*f,-s);h[p]=m}return T.tensor4d(h,e.shape)},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){ae(e,"LRNGrad");for(var u=e.shape[3],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=this.readSync(r.dataId),p=new Float32Array(e.size),f=e.size,m=0;m<f;m++){for(var g=m%u,v=m-g+Math.max(0,g-a),b=m-g+Math.min(u,g+a+1),w=0,S=v;S<b;S++)w+=Math.pow(h[S],2);w=o*w+s;for(var S=v;S<b;S++){var L=-2*o*l*h[S]*d[m]/w;m===S&&(L+=Math.pow(w,-l)),L*=c[m],p[S]+=L}}return T.tensor4d(p,e.shape)},t.prototype.multinomial=function(e,i,r,a){ae(e,"multinomial");for(var s=i?e:T.softmax(e),o=s.shape[0],l=s.shape[1],u=T.zeros([o,r],"int32"),c=this.readSync(u.dataId),h=this.readSync(s.dataId),d=0;d<o;++d){var p=d*l,f=new Float32Array(l-1);f[0]=h[p];for(var m=1;m<f.length;++m)f[m]=f[m-1]+h[p+m];for(var g=sB.alea(a.toString()),v=d*r,b=0;b<r;++b){var w=g();c[v+b]=f.length;for(var S=0;S<f.length;S++)if(w<f[S]){c[v+b]=S;break}}}return u},t.prototype.oneHot=function(e,i,r,a){ae(e,"oneHot");var s=new Float32Array(e.size*i);s.fill(a);for(var o=this.readSync(e.dataId),l=0;l<e.size;++l)o[l]>=0&&o[l]<i&&(s[l*i+o[l]]=r);return T.tensor2d(s,[e.size,i],"int32")},t.prototype.nonMaxSuppression=function(e,i,r,a,s){ae(e,"nonMaxSuppression");var o=this.readSync(e.dataId),l=this.readSync(i.dataId);return lB(o,l,r,a,s)},t.prototype.depthToSpace=function(e,i,r){T.util.assert(r==="NHWC",function(){return"Only NHWC dataFormat supported on CPU for depthToSpace. Got "+r}),T.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m<a;++m)for(var g=0;g<u;++g)for(var v=Math.floor(g/i),b=g%i,w=0;w<c;++w)for(var S=Math.floor(w/i),L=w%i,x=(b*i+L)*h,C=0;C<h;++C){var R=C+x,D=R+l*(S+o*(v+s*m));p[f++]=d[D]}return T.tensor4d(p,[a,u,c,h])},t.prototype.broadcastedBinaryOp=function(e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(e.shape,i.shape),o=T.buffer(s,r),l=this.readSync(e.dataId),u=this.readSync(i.dataId),c=T.backend_util.getBroadcastDims(e.shape,s),h=T.backend_util.getBroadcastDims(i.shape,s),d=o.values;if(c.length+h.length===0)for(var p=0;p<d.length;++p)d[p]=a(l[p%l.length],u[p%u.length]);else for(var f=this.bufferSync(e),m=this.bufferSync(i),g=function(v){var b=o.indexToLoc(v),w=b.slice(-e.rank);c.forEach(function(C){return w[C]=0});var S=f.locToIndex(w),L=b.slice(-i.rank);h.forEach(function(C){return L[C]=0});var x=m.locToIndex(L);d[v]=a(l[S],u[x])},p=0;p<d.length;++p)g(p);return o.toTensor()},t.prototype.split=function(e,i,r){return uB(e,i,r)},t.prototype.dispose=function(){},t.prototype.floatPrecision=function(){return 32},t.prototype.epsilon=function(){return n.prototype.epsilon.call(this)},t.prototype.cropAndResize=function(e,i,r,a,s,o){for(var l=e.shape,u=l[0],c=l[1],h=l[2],d=l[3],p=i.shape[0],f=a[0],m=a[1],g=T.buffer([p,f,m,d],"float32"),v=this.readSync(i.dataId),b=this.readSync(r.dataId),w=this.readSync(e.dataId),S=e.strides,L=g.strides,x=0;x<p;x++){var C=x*4,R=v[C],D=v[C+1],k=v[C+2],W=v[C+3],F=b[x];if(F>=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_<f;_++){var K=f>1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(K<0||K>c-1){for(var j=0;j<m;j++)for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}if(s==="bilinear")for(var Z=Math.floor(K),X=Math.ceil(K),ee=K-Z,j=0;j<m;j++){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var ie=Math.floor(ne),te=Math.ceil(ne),re=ne-ie,q=0;q<d;q++){var G=q+ie*S[2]+Z*S[1]+F*S[0],le=w[G];G=q+te*S[2]+Z*S[1]+F*S[0];var he=w[G];G=q+ie*S[2]+X*S[1]+F*S[0];var ye=w[G];G=q+te*S[2]+X*S[1]+F*S[0];var Oe=w[G],Fe=le+(he-le)*re,_e=ye+(Oe-ye)*re;G=q+j*L[2]+_*L[1]+x*L[0],g.values[G]=Fe+(_e-Fe)*ee}}else for(var j=0;j<m;++j){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var Me=Math.round(ne),ht=Math.round(K),q=0;q<d;q++){var St=q+Me*S[2]+ht*S[1]+F*S[0],Ge=q+j*L[2]+_*L[1]+x*L[0];g.values[Ge]=w[St]}}}}return g.toTensor()},t.prototype.sparseToDense=function(e,i,r,a){var s=T.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.sliceSize,c=s.strides,h=s.outputSize,d=!1;return this.scatter(e,i,r,h,u,l,o,c,a,d)},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=T.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3];if(l===0)return T.tensor([],o,e.dtype);for(var h=new T.TensorBuffer([l,u],e.dtype),d=this.readSync(i.dataId),p=this.readSync(e.dataId),f=0;f<l;f++){for(var m=[],g=0,v=0;v<a;v++){var b=d[f*a+v];g+=b*c[v],m.push(b)}if(g<0||g>=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w<u;w++)h.values[f*u+w]=p[g*u+w]}return h.toTensor().reshape(o)},t.prototype.scatterND=function(e,i,r){var a=T.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=T.scalar(0),d=!0;return this.scatter(e,i,r,c,l,o,s,u,h,d)},t.prototype.fill=function(e,i,r){r=r||T.util.inferDtype(i);var a=T.util.getArrayFromDType(r,T.util.sizeFromShape(e));return a.fill(i),T.engine().makeTensor(a,e,r,this)},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported for string tensors");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){var i=T.util.getArrayFromDType(e.dtype,T.util.sizeFromShape(e.shape));return this.makeOutput(i,e.shape,e.dtype)},t.prototype.linspace=function(e,i,r){return T.backend_util.linspaceImpl(e,i,r)},t.prototype.scatter=function(e,i,r,a,s,o,l,u,c,h){var d=[a/s,s],p=this.readSync(e.dataId),f=this.readSync(i.dataId);if(a===0)return T.tensor([],r,i.dtype);var m=new T.TensorBuffer(d,i.dtype);m.values.fill(this.readSync(c.dataId)[0]);for(var g=0;g<o;g++){for(var v=[],b=0,w=0;w<l;w++){var S=p[g*l+w];v.push(S),b+=S*u[w]}if(b<0||b>=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;L<s;L++)h?m.values[b*s+L]+=f[g*s+L]:m.values[b*s+L]=i.rank===0?f[0]:f[g*s+L]}return m.toTensor().reshape(r)},t}(T.KernelBackend);function Kw(n){for(var t=new Float32Array(n.length),e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}var pB=function(n){var t=n.inputs.x,e=n.backend,i=new Float32Array(T.util.sizeFromShape(t.shape));if(t.dtype!=="complex64"){var r=e.data.get(t.dataId).values;i=Kw(r)}else for(var a=e.data.get(t.dataId),s=a.complexTensorInfos.real,o=a.complexTensorInfos.imag,l=e.data.get(s.dataId).values,u=e.data.get(o.dataId).values,c=0;c<l.length;c++){var h=l[c],d=u[c];i[c]=Math.hypot(h,d)}return e.makeOutput(i,t.shape,"float32")},fB={kernelName:T.Abs,backendName:"cpu",kernelFunc:pB};function lr(n){return function(t,e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=T.util.computeStrides(s),u=T.util.sizeFromShape(s),c=T.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=T.util.computeStrides(t),f=T.util.computeStrides(e),m=T.backend_util.getBroadcastDims(t,s),g=T.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(var v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(var b=function(w){var S=T.util.indexToLoc(w,o,l),L=S.slice(-h);m.forEach(function(D){return L[D]=0});var x=T.util.locToIndex(L,h,p),C=S.slice(-d);g.forEach(function(D){return C[D]=0});var R=T.util.locToIndex(C,d,f);c[w]=n(i[x],r[R])},v=0;v<c.length;++v)b(v);return[c,s]}}function Ln(n){var t=n.inputs,e=n.backend,i=t.real,r=t.imag,a=e.data.get(i.dataId).values,s=e.data.get(r.dataId).values,o=e.makeTensorInfo(i.shape,"complex64"),l=e.data.get(o.dataId);return l.complexTensorInfos={real:e.makeTensorInfo(i.shape,"float32",a),imag:e.makeTensorInfo(r.shape,"float32",s)},o}var mB={kernelName:T.Complex,backendName:"cpu",kernelFunc:Ln};function Xr(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var gB={kernelName:T.Identity,backendName:"cpu",kernelFunc:Xr};function $a(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.real,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var vB={kernelName:T.Real,backendName:"cpu",kernelFunc:$a};function Xa(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.dtype;if(a==="complex64"){if(r.dtype==="complex64")return Xr({inputs:{x:r},backend:e});var s=T.zeros(r.shape),o=Xa({inputs:{x:r},backend:e,attrs:{dtype:"float32"}}),l=Ln({inputs:{real:o,imag:s},backend:e});return s.dispose(),e.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){var u=$a({inputs:{input:r},backend:e}),l=Xa({inputs:{x:u},backend:e,attrs:{dtype:a}});return e.disposeIntermediateTensorInfo(u),l}if(!T.util.hasEncodingLoss(r.dtype,a)){var l=Xr({inputs:{x:r},backend:e});return{dataId:l.dataId,shape:l.shape,dtype:a}}if(a==="int32"){var c=e.data.get(r.dataId).values,h=Int32Array.from(c);return e.makeTensorInfo(r.shape,"int32",h)}if(a==="bool"){var d=e.data.get(r.dataId).values,p=T.util.toTypedArray([0],r.dtype),f=lr(function(v,b){return v!==b?1:0})(r.shape,[],d,p,"bool"),m=f[0],g=f[1];return e.makeTensorInfo(g,"bool",m)}throw new Error("Error in Cast: failed to cast "+r.dtype+" to "+a)}var yB={kernelName:T.Cast,backendName:"cpu",kernelFunc:Xa};function Jr(n,t,e,i){return e==null?function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;ae([l,u],n);var h=c.data.get(l.dataId).values,d=c.data.get(u.dataId).values,p=i||l.dtype,f=t(l.shape,u.shape,h,d,p),m=f[0],g=f[1];return c.makeTensorInfo(g,p,m)}:function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;if(l.dtype==="complex64"||u.dtype==="complex64"){var h=Xa({inputs:{x:l},backend:c,attrs:{dtype:"complex64"}}),d=c.data.get(h.dataId),p=d.complexTensorInfos.real,f=d.complexTensorInfos.imag,m=c.data.get(p.dataId).values,g=c.data.get(f.dataId).values,v=Xa({inputs:{x:u},backend:c,attrs:{dtype:"complex64"}}),b=c.data.get(v.dataId),w=b.complexTensorInfos.real,S=b.complexTensorInfos.imag,L=c.data.get(w.dataId).values,x=c.data.get(S.dataId).values,C=e(l.shape,u.shape,m,g,L,x),R=C[0],D=C[1],k=C[2],W=c.makeTensorInfo(k,"float32",R),F=c.makeTensorInfo(k,"float32",D),P=Ln({inputs:{real:W,imag:F},backend:c});return c.disposeIntermediateTensorInfo(h),c.disposeIntermediateTensorInfo(v),c.disposeIntermediateTensorInfo(W),c.disposeIntermediateTensorInfo(F),P}else{var H=c.data.get(l.dataId).values,_=c.data.get(u.dataId).values,K=i||l.dtype,j=t(l.shape,u.shape,H,_,K),q=j[0],k=j[1];return c.makeTensorInfo(k,K,q)}}}function Ld(n){return function(t,e,i,r,a,s){var o=T.backend_util.assertAndGetBroadcastShape(t,e),l=T.util.sizeFromShape(o),u=o.length,c=T.util.computeStrides(o),h=T.util.getTypedArrayFromDType("float32",l),d=T.util.getTypedArrayFromDType("float32",l),p=T.backend_util.getBroadcastDims(t,o),f=T.backend_util.getBroadcastDims(e,o),m=T.backend_util.mergeRealAndImagArrays(i,r),g=T.backend_util.mergeRealAndImagArrays(a,s),v=t.length,b=T.util.computeStrides(t),w=e.length,S=T.util.computeStrides(e);if(p.length+f.length===0)for(var L=0;L<h.length;L++){var x=L%m.length,C=L%g.length,R=n(m[x*2],m[x*2+1],g[C*2],g[C*2+1]);h[L]=R.real,d[L]=R.imag}else for(var D=function(k){var W=T.util.indexToLoc(k,u,c),F=W.slice(-v);p.forEach(function(j){return F[j]=0});var P=T.util.locToIndex(F,v,b),H=W.slice(-w);f.forEach(function(j){return H[j]=0});var _=T.util.locToIndex(H,w,S),K=n(m[P*2],m[P*2+1],g[_*2],g[_*2+1]);h[k]=K.real,d[k]=K.imag},L=0;L<h.length;L++)D(L);return[h,d,o]}}var jw=lr(function(n,t){return n+t}),bB=Ld(function(n,t,e,i){return{real:n+e,imag:t+i}}),$w=Jr(T.Add,jw,bB),wB={kernelName:T.Add,backendName:"cpu",kernelFunc:$w};function Zr(n){return function(t,e,i){for(var r=T.util.getTypedArrayFromDType(e,t.length),a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}function $e(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");for(var l=s,u=l.data.get(o.dataId).values,c=T.util.sizeFromShape(o.shape),h=e||o.dtype,d=T.util.getArrayFromDType(h,c),p=0;p<c;++p)d[p]=t(u[p],a);return l.makeTensorInfo(o.shape,h,d)}}function Qr(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");var l=s,u=l.data.get(o.dataId).values,c=e||o.dtype,h=t(u,c,a);return l.makeTensorInfo(o.shape,c,h)}}var Xw=Zr(function(n){return Math.ceil(n)}),SB=Qr(T.Ceil,Xw),LB={kernelName:T.Ceil,backendName:"cpu",kernelFunc:SB};var Jw=Zr(function(n){return Math.exp(n)}),IB=Qr(T.Exp,Jw),AB={kernelName:T.Exp,backendName:"cpu",kernelFunc:IB};var Zw=Zr(function(n){return Math.expm1(n)}),TB=Qr(T.Expm1,Zw),NB={kernelName:T.Expm1,backendName:"cpu",kernelFunc:TB};var Qw=Zr(function(n){return Math.floor(n)}),xB=Qr(T.Floor,Qw),CB={kernelName:T.Floor,backendName:"cpu",kernelFunc:xB};var e0=Zr(function(n){return Math.log(n)}),RB=Qr(T.Log,e0),OB={kernelName:T.Log,backendName:"cpu",kernelFunc:RB};function t0(n,t,e,i){for(var r=T.util.getTypedArrayFromDType(i,T.util.sizeFromShape(e)),a=0;a<r.length;++a){for(var s=a*t,o=n[s],l=0;l<t;++l){var u=n[s+l];u>o&&(o=u)}r[a]=o}return r}var n0=lr(function(n,t){return n*t}),EB=Ld(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),i0=Jr(T.Multiply,n0,EB),DB={kernelName:T.Multiply,backendName:"cpu",kernelFunc:i0};var r0=Zr(function(n){return 1/Math.sqrt(n)}),kB=Qr(T.Rsqrt,r0),FB={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:kB};function a0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;c<s;++c){var h=e.length,d=T.util.computeStrides(e),p=T.util.indexToLoc(c,h,d),f=p.map(function(g,v){return g+t[v]}),m=T.util.locToIndex(f,i.length,o);u[c]=n[m]}return u}function Id(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.begin,s=i.size;ae(r,"slice");var o=T.slice_util.parseSliceParams(r,a,s),l=o[0],u=o[1];T.slice_util.assertParamsValid(r,l,u);var c=e.data.get(r.dataId).values,h=a0(c,l,u,r.shape,r.dtype);return e.makeTensorInfo(u,r.dtype,h)}var WB={kernelName:T.Slice,backendName:"cpu",kernelFunc:Id};var s0=lr(function(n,t){return n-t}),UB=Ld(function(n,t,e,i){return{real:n-e,imag:t-i}}),o0=Jr(T.Sub,s0,UB),BB={kernelName:T.Sub,backendName:"cpu",kernelFunc:o0};function Ad(n,t,e,i,r){for(var a=t.length,s=T.util.sizeFromShape(t),o=T.util.computeStrides(t),l=T.util.computeStrides(r),u=T.util.getTypedArrayFromDType(e,T.util.sizeFromShape(r)),c=0;c<s;++c){for(var h=T.util.indexToLoc(c,a,o),d=new Array(h.length),p=0;p<d.length;p++)d[p]=h[i[p]];var f=T.util.locToIndex(d,a,l);u[f]=n[c]}return u}function l0(n,t,e,i){for(var r=T.util.parseAxisParam(t,e)[0],a=[1,e[0],1],s=0;s<r;s++)a[0]*=e[s];a[1]=e[r];for(var s=r+1;s<e.length;s++)a[2]*=e[s];for(var o={},l=new Int32Array(e[r]),u=new T.TensorBuffer(a,i,n),c=[],h=a[0]===1&&a[2]===1,s=0;s<e[r];s++){var d=void 0;if(h)d=n[s].toString();else{for(var p=[],f=0;f<a[0];f++)for(var m=0;m<a[2];m++)p.push(u.get(f,s,m));d=p.join(",")}if(o[d]!==void 0)l[s]=o[d];else{var g=Object.keys(o).length;o[d]=g,l[s]=g,c.push(s)}}var v=a.slice();v[1]=Object.keys(o).length;var b=new T.TensorBuffer(v,i);c.forEach(function(S,L){for(var x=0;x<a[0];x++)for(var C=0;C<a[2];C++)b.set(u.get(x,S,C),x,L,C)});var w=e.slice();return w[r]=v[1],{outputValues:b.values,outputShape:w,indices:l}}var zB={__proto__:null,simpleAbsImpl:Kw,addImpl:jw,ceilImpl:Xw,expImpl:Jw,expm1Impl:Zw,floorImpl:Qw,logImpl:e0,maxImpl:t0,multiplyImpl:n0,rsqrtImpl:r0,sliceImpl:a0,subImpl:s0,transposeImpl:Ad,uniqueImpl:l0};var PB="2.6.0";T.registerBackend("cpu",function(){return new Yw},1);var _B=$e(T.Acos,function(n){return Math.acos(n)}),MB={kernelName:T.Acos,backendName:"cpu",kernelFunc:_B};var HB=$e(T.Acosh,function(n){return Math.acosh(n)}),VB={kernelName:T.Acosh,backendName:"cpu",kernelFunc:HB};var qB=$e(T.Asin,function(n){return Math.asin(n)}),GB={kernelName:T.Asin,backendName:"cpu",kernelFunc:qB};var YB=$e(T.Asinh,function(n){return Math.asinh(n)}),KB={kernelName:T.Asinh,backendName:"cpu",kernelFunc:YB};var jB=$e(T.Atan,function(n){return Math.atan(n)}),$B={kernelName:T.Atan,backendName:"cpu",kernelFunc:jB};var XB=$e(T.Atanh,function(n){return Math.atanh(n)}),JB={kernelName:T.Atanh,backendName:"cpu",kernelFunc:XB};function Td(n,t,e,i,r,a){for(var s=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,u=r.dilationWidth,c=r.effectiveFilterHeight,h=r.effectiveFilterWidth,d=r.padInfo.top,p=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=T.buffer(r.outShape,e),g=m.values,v=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],w=r.outShape[3],S=0;S<r.batchSize;++S)for(var L=S*v,x=S*i[0],C=0;C<r.inChannels;++C)for(var R=0;R<r.outHeight;++R)for(var D=R*s-d,k=Math.max(0,D),W=Math.min(r.inHeight,c+D),F=L+R*b,P=0;P<r.outWidth;++P){for(var H=P*o-p,_=Math.max(0,H),K=Math.min(r.inWidth,h+H),j=f,q=0,G=0,Z=k;Z<W;Z+=l){for(var X=x+Z*i[1],ee=_;ee<K;ee+=u){var ne=X+ee*i[2],ie=n[ne+C];a==="max"&&ie>j?j=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(j))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:j}return m}function u0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;g<i.batchSize;++g)for(var v=0;v<i.inChannels;++v)for(var b=0;b<i.outHeight;++b){for(var w=b*o-p,S=w;S<0;)S+=u;for(var L=Math.min(i.inHeight,h+w),x=0;x<i.outWidth;++x){for(var C=x*l-f,R=C;R<0;)R+=c;for(var D=Math.min(i.inWidth,d+C),k=Number.NEGATIVE_INFINITY,W=-1,F=S;F<L;F+=u)for(var P=F-w,H=R;H<D;H+=c){var _=H-C,K=m.get(g,F,H,v);K>k&&(k=K,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,x,v)}}return s}function ZB(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Td(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var QB={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:ZB};function ez(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),x=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D<c.batchSize;++D)for(var k=0;k<c.inChannels;++k)for(var W=0;W<c.inHeight;++W)for(var F=0;F<c.inWidth;++F){for(var P=W-S,H=F-w,_=0,K=0;K<v;K+=m){var j=(P+K)/h;if(j<0||j>=c.outHeight||Math.floor(j)!==j)continue;for(var q=0;q<b;q+=g){var G=(H+q)/d;if(G<0||G>=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,j,G,k);_+=Z}}L.set(_*x,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var tz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:ez};function nz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,x=0,C=0,R=0;R<c.length;++R)m[R]=f[S++]+(c[R]-h[L++])*p[x++]/Math.sqrt(d[C++]+u),S>=g&&(S=0),L>=w&&(L=0),x>=v&&(x=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var iz={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:nz};var rz=$e(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n<e.clipValueMin?e.clipValueMin:n}),az={kernelName:T.ClipByValue,backendName:"cpu",kernelFunc:rz};function Eo(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.imag,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var sz={kernelName:T.Imag,backendName:"cpu",kernelFunc:Eo};function Ai(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=T.util.sizeFromShape(r.shape),o=T.util.inferFromImplicitShape(a,s),l=T.util.sizeFromShape(o);T.util.assert(s===l,function(){return"The new shape ("+o+") has "+l+" elements and the old "+("shape ("+r.shape+") has "+s+" elements. The new shape and old ")+"shape must have the same number of elements."}),e.incRef(r.dataId);var u=e.data.get(r.dataId);if(u.complexTensorInfos!=null){var c=u.complexTensorInfos.real,h=u.complexTensorInfos.imag;c.shape=o,h.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var oz={kernelName:T.Reshape,backendName:"cpu",kernelFunc:Ai};function Ja(n){var t=n.inputs,e=n.backend,i=n.attrs,r=i.axis,a=T.util.parseAxisParam(r,t[0].shape)[0],s=T.backend_util.computeOutShape(t.map(function(S){return S.shape}),a);if(T.util.sizeFromShape(s)===0)return e.makeTensorInfo(s,t[0].dtype,[]);var o=t.filter(function(S){return T.util.sizeFromShape(S.shape)>0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return $a({inputs:{input:S},backend:e})}),c=o.map(function(S){return Eo({inputs:{input:S},backend:e})}),h=Ja({inputs:u,backend:e,attrs:{axis:r}}),d=Ja({inputs:c,backend:e,attrs:{axis:r}}),p=Ln({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),x=[-1,L];return Ai({inputs:{x:S},backend:e,attrs:{shape:x}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,x=T.util.sizeFromShape(S.shape);m.set(L,g),g+=x})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,x=0,C=0;C<S.shape[0];++C)for(var R=C*s[1]+v,D=0;D<S.shape[1];++D)m[R+D]=L[x++];v+=S.shape[1]})}var b=T.backend_util.computeOutShape(o.map(function(S){return S.shape}),a),w=e.makeTensorInfo(b,t[0].dtype,m);return f.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),w}var lz={kernelName:T.Concat,backendName:"cpu",kernelFunc:Ja};var uz=$e(T.Cos,function(n){return Math.cos(n)}),cz={kernelName:T.Cos,backendName:"cpu",kernelFunc:uz};var hz=$e(T.Cosh,function(n){return Math.cosh(n)}),dz={kernelName:T.Cosh,backendName:"cpu",kernelFunc:hz};var pz={kernelName:T.Dilation2D,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=i,l=o.strides,u=o.pad,c=o.dilations,h=e,d=h.data.get(a.dataId).values,p=a.shape.length,f=h.data.get(s.dataId).values,m=s.shape.length,g=T.backend_util.computeDilation2DInfo(a.shape,s.shape,l,u,"NHWC",c),v=g.batchSize,b=g.inHeight,w=g.inWidth,S=g.inChannels,L=g.outHeight,x=g.outWidth,C=g.padInfo,R=g.strideHeight,D=g.strideWidth,k=g.filterHeight,W=g.filterWidth,F=g.dilationHeight,P=g.dilationWidth,H=g.outShape,_=T.util.sizeFromShape(H),K=H.length,j=T.util.getArrayFromDType(a.dtype,_),q=0;q<v;++q)for(var G=0;G<L;++G)for(var Z=G*R-C.top,X=0;X<x;++X)for(var ee=X*D-C.left,ne=0;ne<S;++ne){for(var ie=Number.MIN_SAFE_INTEGER,te=0;te<k;++te){var re=Z+te*F;if(re>=0&&re<b)for(var le=0;le<W;++le){var he=ee+le*P;if(he>=0&&he<w){var ye=T.util.locToIndex([q,re,he,ne],p,T.util.computeStrides(a.shape)),Oe=T.util.locToIndex([te,le,ne],m,T.util.computeStrides(s.shape)),Fe=d[ye]+f[Oe];Fe>ie&&(ie=Fe)}}}var _e=T.util.locToIndex([q,G,X,ne],K,T.util.computeStrides(H));j[_e]=ie}var Me=h.write(T.util.toTypedArray(j,a.dtype),H,a.dtype);return{dataId:Me,shape:H,dtype:a.dtype}}};var fz={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=0,ie=0,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var mz={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=q<0?0:q,ie=Z<0?0:Z,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=re,ie=he)}}}_[K][ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var gz=lr(function(n,t){return n/t}),vz=Jr(T.Div,gz),Nd={kernelName:T.Div,backendName:"cpu",kernelFunc:vz};var yz=$e(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),bz={kernelName:T.Elu,backendName:"cpu",kernelFunc:yz};var wz=T.backend_util.ERF_P,Sz=T.backend_util.ERF_A1,Lz=T.backend_util.ERF_A2,Iz=T.backend_util.ERF_A3,Az=T.backend_util.ERF_A4,Tz=T.backend_util.ERF_A5,Nz=$e(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+wz*e);return t*(1-((((Tz*i+Az)*i+Iz)*i+Lz)*i+Sz)*i*Math.exp(-e*e))}),xz={kernelName:T.Erf,backendName:"cpu",kernelFunc:Nz};function c0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p<r;p++){for(var f=Id({inputs:{x:o},backend:e,attrs:{begin:[p,0],size:[1,a]}}),m=Id({inputs:{x:l},backend:e,attrs:{begin:[p,0],size:[1,a]}}),g=Ln({inputs:{real:f,imag:m},backend:e}),v=Cz(g,t,e),b=v.real,w=v.imag,S=T.backend_util.mergeRealAndImagArrays(b,w),L=0;L<a;L++){var x=T.backend_util.getComplexWithIndex(S,L);h[p*a+L]=x.real,d[p*a+L]=x.imag}e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(g)}var C=e.makeTensorInfo(u,"float32",h),R=e.makeTensorInfo(u,"float32",d),D=Ln({inputs:{real:C,imag:R},backend:e});return e.disposeIntermediateTensorInfo(C),e.disposeIntermediateTensorInfo(R),D}function Cz(n,t,e){var i=T.util.sizeFromShape(n.shape),r=e.data.get(n.dataId),a=e.data.get(r.complexTensorInfos.real.dataId).values,s=e.data.get(r.complexTensorInfos.imag.dataId).values;if(Rz(i)){var o=xd(a,s,i,t,e),l=[n.shape[0],n.shape[1]];if(t){var u=e.makeTensorInfo(l,"float32",o.real),c=e.makeTensorInfo(l,"float32",o.imag),h=e.makeTensorInfo([],"float32",T.util.createScalarValue(i,"float32")),d=Xr({inputs:{x:h},backend:e}),p=Nd.kernelFunc({inputs:{a:u,b:h},backend:e}),f=Nd.kernelFunc({inputs:{a:c,b:d},backend:e}),m=e.data.get(p.dataId).values,g=e.data.get(f.dataId).values;return e.disposeIntermediateTensorInfo(u),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return o}else{var v=T.backend_util.mergeRealAndImagArrays(a,s),b=Oz(v,i,t);return T.backend_util.splitRealAndImagArrays(b)}}function Rz(n){return(n&n-1)===0}function xd(n,t,e,i,r){if(e===1)return{real:n,imag:t};var a=T.backend_util.mergeRealAndImagArrays(n,t),s=e/2,o=T.backend_util.complexWithEvenIndex(a),l=o.real,u=o.imag,c=[l.length],h=r.makeTensorInfo(c,"float32",l),d=r.makeTensorInfo(c,"float32",u),p=Ln({inputs:{real:h,imag:d},backend:r}),f=T.backend_util.complexWithOddIndex(a),m=f.real,g=f.imag,v=[m.length],b=r.makeTensorInfo(v,"float32",m),w=r.makeTensorInfo(v,"float32",g),S=Ln({inputs:{real:b,imag:w},backend:r}),L=xd(l,u,s,i,r),x=L.real,C=L.imag,R=[x.length],D=r.makeTensorInfo(R,"float32",x),k=r.makeTensorInfo(R,"float32",C),W=Ln({inputs:{real:D,imag:k},backend:r}),F=xd(m,g,s,i,r),P=F.real,H=F.imag,_=[P.length],K=r.makeTensorInfo(_,"float32",P),j=r.makeTensorInfo(_,"float32",H),q=Ln({inputs:{real:K,imag:j},backend:r}),G=T.backend_util.exponents(e,i),Z=[G.real.length],X=r.makeTensorInfo(Z,"float32",G.real),ee=r.makeTensorInfo(Z,"float32",G.imag),ne=Ln({inputs:{real:X,imag:ee},backend:r}),ie=i0({inputs:{a:ne,b:q},backend:r}),te=$w({inputs:{a:W,b:ie},backend:r}),re=o0({inputs:{a:W,b:ie},backend:r}),le=$a({inputs:{input:te},backend:r}),he=$a({inputs:{input:re},backend:r}),ye=Eo({inputs:{input:te},backend:r}),Oe=Eo({inputs:{input:re},backend:r}),Fe=Ja({inputs:[le,he],backend:r,attrs:{axis:0}}),_e=Ja({inputs:[ye,Oe],backend:r,attrs:{axis:0}}),Me=r.data.get(Fe.dataId).values,ht=r.data.get(_e.dataId).values;return r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(w),r.disposeIntermediateTensorInfo(S),r.disposeIntermediateTensorInfo(D),r.disposeIntermediateTensorInfo(k),r.disposeIntermediateTensorInfo(W),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(q),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(ie),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(re),r.disposeIntermediateTensorInfo(le),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(he),r.disposeIntermediateTensorInfo(Oe),r.disposeIntermediateTensorInfo(Fe),r.disposeIntermediateTensorInfo(_e),{real:Me,imag:ht}}function Oz(n,t,e){for(var i=new Float32Array(t*2),r=0;r<t;r++){for(var a=0,s=0,o=0;o<t;o++){var l=T.backend_util.exponent(r*o,t,e),u=T.backend_util.getComplexWithIndex(n,o);a+=u.real*l.real-u.imag*l.imag,s+=u.real*l.imag+u.imag*l.real}e&&(a/=t,s/=t),T.backend_util.assignToTypedArray(i,a,s,r)}return i}function Ez(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!1,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Dz={kernelName:T.FFT,backendName:"cpu",kernelFunc:Ez};var kz={kernelName:T.FlipLeftRight,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=i,s=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),o=r.shape,l=o[0],u=o[1],c=o[2],h=o[3],d=a.data.get(r.dataId).values,p=0;p<l;p++)for(var f=p*c*u*h,m=0;m<u;m++)for(var g=m*(c*h),v=0;v<c;v++)for(var b=v*h,w=0;w<h;w++){var S=[l,m,v,w],L=S[2],x=Math.round(c-L),C=f+g+b+w,R=d[C];if(x>=0&&x<c){var D=x*h,k=f+g+D+w;R=d[k]}s[C]=R}var W=a.write(s,r.shape,r.dtype);return{dataId:W,shape:r.shape,dtype:r.dtype}}};function Fz(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!0,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Wz={kernelName:T.IFFT,backendName:"cpu",kernelFunc:Fz};var Uz=$e(T.IsFinite,function(n){return Number.isFinite(n)?1:0},"bool"),Bz={kernelName:T.IsFinite,backendName:"cpu",kernelFunc:Uz};var zz=$e(T.IsInf,function(n){return Math.abs(n)===Infinity?1:0},"bool"),Pz={kernelName:T.IsInf,backendName:"cpu",kernelFunc:zz};var _z=$e(T.IsNan,function(n){return Number.isNaN(n)?1:0},"bool"),Mz={kernelName:T.IsNan,backendName:"cpu",kernelFunc:_z};var Hz=$e(T.Log1p,function(n){return Math.log1p(n)}),Vz={kernelName:T.Log1p,backendName:"cpu",kernelFunc:Hz};var qz=$e(T.LogicalNot,function(n){return n?0:1},"bool"),Gz={kernelName:T.LogicalNot,backendName:"cpu",kernelFunc:qz};var Yz={kernelName:T.Max,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape,c=u.length,h=T.util.parseAxisParam(s,u),d=h,p=T.backend_util.getAxesPermutation(d,c),f=l.data.get(r.dataId).values;if(p!=null){for(var m=new Array(c),g=0;g<m.length;g++)m[g]=u[p[g]];f=Ad(f,u,r.dtype,p,m),d=T.backend_util.getInnerMostAxes(d.length,c),u=m}ae(r,"max"),T.backend_util.assertAxesAreInnerMostDims("max",d,c);var v=T.backend_util.computeOutAndReduceShapes(u,d),b=v[0],w=v[1],S=T.util.sizeFromShape(w),L=t0(f,S,b,r.dtype),x=l.write(L,b,r.dtype),C=b;if(o){var m=T.backend_util.expandShapeToKeepDim(b,h);C=m}return{dataId:x,shape:C,dtype:r.dtype}}};function Kz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Td(d,r.shape,r.dtype,p,c,"max");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var jz={kernelName:T.MaxPool,backendName:"cpu",kernelFunc:Kz};function $z(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ae([a,s],"maxPoolBackprop");for(var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=T.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=e.data.get(o.dataId).values,f=T.buffer(d.outShape,o.dtype,u0(p,o.shape,o.dtype,d).values),m=d.strideHeight,g=d.strideWidth,v=d.dilationHeight,b=d.dilationWidth,w=d.effectiveFilterHeight,S=d.effectiveFilterWidth,L=S-1-d.padInfo.left,x=w-1-d.padInfo.top,C=T.buffer(o.shape,"float32"),R=e.data.get(r.dataId).values,D=T.buffer(r.shape,"float32",R),k=0;k<d.batchSize;++k)for(var W=0;W<d.inChannels;++W)for(var F=0;F<d.inHeight;++F)for(var P=0;P<d.inWidth;++P){for(var H=F-x,_=P-L,K=0,j=0;j<w;j+=v){var q=(H+j)/m;if(q<0||q>=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<S;G+=b){var Z=(_+G)/g;if(Z<0||Z>=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=j*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);K+=ie*ne}}C.set(K,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var Xz={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:$z};function Jz(n,t,e,i,r){var a=T.util.computeStrides(t),s=Td(n,t,e,a,r,"max"),o=u0(n,t,e,r,!0,i);return[s.values,o.values]}var Zz={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=Jz(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var Qz=T.kernel_impls.nonMaxSuppressionV4Impl,eP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=Qz(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var tP=T.kernel_impls.nonMaxSuppressionV5Impl,nP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=tP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var iP=lr(function(n,t){return n!==t?1:0}),rP=Jr(T.NotEqual,iP,null,"bool"),aP={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:rP};function sP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(x,C){return x[0]+r.shape[C]+x[1]}),l=a.map(function(x){return x[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v<c;v++){var b=T.util.indexToLoc(v,h,d),w=b.map(function(x,C){return x+l[C]}),S=T.util.locToIndex(w,f,m);g[S]=u[v]}var L=e.write(g,o,r.dtype);return{dataId:L,shape:o,dtype:r.dtype}}var h0={kernelName:T.PadV2,backendName:"cpu",kernelFunc:sP};var oP=$e(T.Reciprocal,function(n){return 1/n}),lP={kernelName:T.Reciprocal,backendName:"cpu",kernelFunc:oP};var uP={kernelName:T.RotateWithOffset,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),h=r.shape,d=h[0],p=h[1],f=h[2],m=h[3],g=T.backend_util.getImageCenter(l,p,f),v=g[0],b=g[1],w=255,S=Math.sin(s),L=Math.cos(s),x=u.data.get(r.dataId).values,C=0;C<d;C++)for(var R=C*f*p*m,D=0;D<p;D++)for(var k=D*(f*m),W=0;W<f;W++)for(var F=W*m,P=0;P<m;P++){var H=[d,D,W,P],_=H[2],K=H[1],j=(_-v)*L-(K-b)*S,q=(_-v)*S+(K-b)*L;j=Math.round(j+v),q=Math.round(q+b);var G=o;if(typeof o!="number"&&(P===3?G=w:G=o[P]),j>=0&&j<f&&q>=0&&q<p){var Z=q*(f*m),X=j*m,ee=R+Z+X+P;G=x[ee]}var ne=R+k+F+P;c[ne]=G}var ie=u.write(c,r.shape,r.dtype);return{dataId:ie,shape:r.shape,dtype:r.dtype}}};var cP=$e(T.Round,function(n){var t=Math.floor(n);return n-t<.5?Math.floor(n):n-t>.5?Math.ceil(n):t%2===0?t:t+1}),hP={kernelName:T.Round,backendName:"cpu",kernelFunc:cP};var dP=T.backend_util.SELU_SCALEALPHA,pP=T.backend_util.SELU_SCALE,fP=$e(T.Selu,function(n){return n>=0?pP*n:dP*(Math.exp(n)-1)}),mP={kernelName:T.Selu,backendName:"cpu",kernelFunc:fP};var gP=$e(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),vP={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:gP};var yP=$e(T.Sign,function(n){return n<0?-1:n>0?1:0}),bP={kernelName:T.Sign,backendName:"cpu",kernelFunc:yP};var wP=$e(T.Sin,function(n){return Math.sin(n)}),SP={kernelName:T.Sin,backendName:"cpu",kernelFunc:wP};var LP=$e(T.Sinh,function(n){return Math.sinh(n)}),IP={kernelName:T.Sinh,backendName:"cpu",kernelFunc:LP};var AP=11920928955078125e-23,d0=Math.log(AP)+2,TP=$e(T.Softplus,function(n){var t=n>-d0,e=n<d0,i=Math.exp(n),r;return e?r=i:t?r=n:r=Math.log(1+i),r}),NP={kernelName:T.Softplus,backendName:"cpu",kernelFunc:TP};function p0(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm;ae(r,"transpose");for(var s=r.shape.length,o=new Array(s),l=0;l<o.length;l++)o[l]=r.shape[a[l]];var u=i.data.get(r.dataId).values,c=Ad(u,r.shape,r.dtype,a,o),h=i.write(c,o,r.dtype);return{dataId:h,shape:o,dtype:r.dtype}}var xP={kernelName:T.Transpose,backendName:"cpu",kernelFunc:p0};function CP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.blockShape,s=i.paddings;ae([r],"spaceToBatchND");var o=T.util.sizeFromShape(a),l=[[0,0]];l.push.apply(l,s);for(var u=1+a.length;u<r.shape.length;++u)l.push([0,0]);var c=h0.kernelFunc({inputs:{x:r},backend:e,attrs:{paddings:l,constantValue:0}}),h=T.backend_util.getReshaped(c.shape,a,o,!1),d=T.backend_util.getPermuted(h.length,a.length,!1),p=T.backend_util.getReshapedPermuted(c.shape,a,o,!1),f={x:c},m={shape:h},g=Ai({inputs:f,backend:e,attrs:m}),v={x:g},b={perm:d},w=p0({inputs:v,backend:e,attrs:b}),S={x:w},L={shape:p},x=Ai({inputs:S,backend:e,attrs:L});return e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(w),x}var RP={kernelName:T.SpaceToBatchND,backendName:"cpu",kernelFunc:CP};var OP=$e(T.Sqrt,function(n){return Math.sqrt(n)}),EP={kernelName:T.Sqrt,backendName:"cpu",kernelFunc:OP};var DP={kernelName:T.Square,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.x,r=e;ae(i,"square");for(var a=r.data.get(i.dataId).values,s=new Float32Array(a.length),o=0;o<a.length;++o){var l=a[o];s[o]=l*l}var u=r.write(s,i.shape,i.dtype);return{dataId:u,shape:i.shape,dtype:i.dtype}}};var kP=lr(function(n,t){var e=n-t;return e*e}),FP=Jr(T.SquaredDifference,kP),WP={kernelName:T.SquaredDifference,backendName:"cpu",kernelFunc:FP};var UP=$e(T.Step,function(n,t){var e=t;return isNaN(n)?NaN:n>0?1:e.alpha}),BP={kernelName:T.Step,backendName:"cpu",kernelFunc:UP};var zP=$e(T.Tan,function(n){return Math.tan(n)}),PP={kernelName:T.Tan,backendName:"cpu",kernelFunc:zP};var _P=$e(T.Tanh,function(n){return Math.tanh(n)}),MP={kernelName:T.Tanh,backendName:"cpu",kernelFunc:_P};function HP(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=l0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var VP={kernelName:T.Unique,backendName:"cpu",kernelFunc:HP};var qP=[fB,MB,VB,wB,GB,KB,$B,JB,QB,tz,iz,yB,LB,az,mB,lz,cz,dz,pz,mz,fz,Nd,bz,xz,AB,NB,Dz,kz,CB,gB,Wz,sz,Bz,Pz,Mz,OB,Vz,Gz,jz,Xz,Zz,Yz,DB,eP,nP,aP,h0,vB,lP,oz,uP,hP,FB,mP,vP,bP,SP,IP,WB,NP,RP,EP,DP,WP,BP,BB,PP,MP,xP,VP];for(var Cd=0,f0=qP;Cd<f0.length;Cd++){var GP=f0[Cd];T.registerKernel(GP)}ja.MathBackendCPU=Yw;ja.shared=zB;ja.version_cpu=PB});var JS=be(Pn=>{"use strict";Object.defineProperty(Pn,"__esModule",{value:!0});var N=Xi();var Rd=function(n,t){return Rd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Rd(n,t)};function YP(n,t){Rd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Do(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function ko(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var ur={},Od={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function g0(n,t){ur[n]=t}function _n(n){if(!(n in ur)){var t=KP(n);if(t!==null)ur[n]=t;else return console.log("Could not get context for WebGL version",n),null}var e=ur[n];return e.isContextLost()?(delete ur[n],_n(n)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),ur[n])}function jP(n){if(typeof OffscreenCanvas!="undefined"&&n===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function KP(n){if(n!==1&&n!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");var t=jP(n);return t.addEventListener("webglcontextlost",function(e){e.preventDefault(),delete ur[n]},!1),n===1?t.getContext("webgl",Od)||t.getContext("experimental-webgl",Od):t.getContext("webgl2",Od)}var Za;(function(n){n[n.DENSE=0]="DENSE",n[n.SHARED_BATCH=1]="SHARED_BATCH"})(Za||(Za={}));var nn;(function(n){n[n.RENDER=0]="RENDER",n[n.UPLOAD=1]="UPLOAD",n[n.PIXELS=2]="PIXELS",n[n.DOWNLOAD=3]="DOWNLOAD"})(nn||(nn={}));var Ct;(function(n){n[n.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",n[n.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",n[n.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",n[n.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",n[n.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Ct||(Ct={}));function Qa(n,t){return[t,n]}function $P(n,t){return n*t}function es(n){var t=N.util.sizeFromShape(n),e=Math.ceil(t/4);return N.util.sizeToSquarishShape(e)}function ea(n,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(n/2))]}function XP(n,t){var e=ea(n,t),i=e[0],r=e[1];return i*r*4}function Ed(n,t){var e=n,i,r,a,s,o,l,u,c,h,d;return N.env().getNumber("WEBGL_VERSION")===2?(i=e.R32F,r=e.R16F,a=e.RGBA16F,s=e.RGBA32F,o=e.RED,u=4,c=1,h=e.HALF_FLOAT,d=e.FLOAT):(i=n.RGBA,r=n.RGBA,a=n.RGBA,s=e.RGBA,o=n.RGBA,u=4,c=4,h=t!=null?t.HALF_FLOAT_OES:null,d=n.FLOAT),l=n.RGBA,{internalFormatFloat:i,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:s,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:u,defaultNumChannels:c,textureTypeHalfFloat:h,textureTypeFloat:d}}function ce(n,t){var e=t();return N.env().getBool("DEBUG")&&JP(n),e}function JP(n){var t=n.getError();if(t!==n.NO_ERROR)throw new Error("WebGL Error: "+v0(n,t))}var ZP=596e-10,QP=65504;function y0(n){return!!(N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||n===0||ZP<Math.abs(n)&&Math.abs(n)<QP)}function v0(n,t){switch(t){case n.NO_ERROR:return"NO_ERROR";case n.INVALID_ENUM:return"INVALID_ENUM";case n.INVALID_VALUE:return"INVALID_VALUE";case n.INVALID_OPERATION:return"INVALID_OPERATION";case n.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case n.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case n.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return"Unknown error code "+t}}function ts(n,t){return ei(n,function(){return n.getExtension(t)},'Extension "'+t+'" not supported on this browser.')}function b0(n,t){var e=ei(n,function(){return n.createShader(n.VERTEX_SHADER)},"Unable to create vertex WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw console.log(n.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function w0(n,t){var e=ei(n,function(){return n.createShader(n.FRAGMENT_SHADER)},"Unable to create fragment WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw e_(t,n.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var t_=/ERROR: [0-9]+:([0-9]+):/g;function e_(n,t){var e=t_.exec(t);if(e==null){console.log("Couldn't parse line number in error: "+t),console.log(n);return}for(var i=+e[1],r=n.split(`
 `),a=r.length.toString().length+2,s=r.map(function(d,p){return N.util.rightPad((p+1).toString(),a)+d}),o=0,l=0;l<s.length;l++)o=Math.max(s[l].length,o);var u=s.slice(0,i-1),c=s.slice(i-1,i),h=s.slice(i);console.log(u.join(`
 `)),console.log(t.split(`
 `)[0]),console.log("%c "+N.util.rightPad(c[0],o),"border:1px solid red; background-color:#e3d2d2; color:#a61717"),console.log(h.join(`
-`))}function S0(n){return ei(n,function(){return n.createProgram()},"Unable to create WebGLProgram.")}function L0(n,t){if(ce(n,function(){return n.linkProgram(t)}),n.getProgramParameter(t,n.LINK_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function Fo(n,t){if(ce(n,function(){return n.validateProgram(t)}),n.getProgramParameter(t,n.VALIDATE_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function I0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function A0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ELEMENT_ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ELEMENT_ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function n_(){return N.env().getNumber("WEBGL_VERSION")===2?1:4}function T0(n){return ei(n,function(){return n.createTexture()},"Unable to create WebGLTexture.")}function N0(n,t){var e=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(n<=0||t<=0){var i="["+n+"x"+t+"]";throw new Error("Requested texture size "+i+" is invalid.")}if(n>e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function x0(n){return ei(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function Ed(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function R0(n,t,e){C0(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function i_(n,t){C0(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function O0(n,t,e){return ei(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function E0(n,t,e){return n.getUniformLocation(t,e)}function D0(n,t,e,i){ce(n,function(){return R0(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function r_(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function Wo(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function Dd(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function ns(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+k0(n,t))}function k0(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function ei(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function C0(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(i<n.TEXTURE0||i>e){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function cr(n,t){return t===void 0&&(t=2),N.util.sizeFromShape(n.slice(0,n.length-t))}function hr(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function Uo(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[cr(n)].concat(hr(n))),t}function F0(n,t){var e;t===void 0&&(t=!1);var i=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?N.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=N.util.squeezeShape(n);n=r.newShape}var a=N.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=cr(n),o=2,l=2;return n.length&&(e=hr(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),N.util.sizeToSquarishShape(a).map(function(u){return u*2})}return N.util.sizeToSquarishShape(a)}function Bo(n){return n%2===0}function is(n,t){if(n=n.slice(-2),t=t.slice(-2),N.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(Bo(e)&&Bo(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&Bo(n[0])&&Bo(t[0])}var zo,Po;function W0(n){if(zo==null){var t=_n(n);zo=t.getParameter(t.MAX_TEXTURE_SIZE)}return zo}function a_(){zo=null}function s_(){Po=null}function U0(n){if(Po==null){var t=_n(n);Po=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,Po)}function B0(n){if(n===0)return 0;var t,e=_n(n);return an(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:an(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function an(n,t){var e=n.getExtension(t);return e!=null}function kd(n){try{var t=_n(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function z0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!an(t,"OES_texture_float"))return!1}else if(!an(t,"EXT_color_buffer_float"))return!1;var e=Fd(t);return e}function P0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!an(t,"OES_texture_float"))return!1;if(!an(t,"WEBGL_color_buffer_float"))return!1}else{if(an(t,"EXT_color_buffer_float"))return Fd(t);var e="EXT_color_buffer_half_float";if(an(t,e)){var i=t.getExtension(e);return o_(t,i)}return!1}var r=Fd(t);return r}function Fd(n){var t=Od(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function o_(n,t){var e=Od(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function _0(n){if(n!==2)return!1;var t=_n(n),e=t.fenceSync!=null;return e}function ta(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&N.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var l_={__proto__:null,callAndCheck:ce,canBeRepresented:y0,getWebGLErrorMessage:v0,getExtensionOrThrow:ts,createVertexShader:b0,createFragmentShader:w0,createProgram:S0,linkProgram:L0,validateProgram:Fo,createStaticVertexBuffer:I0,createStaticIndexBuffer:A0,getNumChannels:n_,createTexture:T0,validateTextureSize:N0,createFramebuffer:x0,bindVertexBufferToProgramAttribute:Ed,bindTextureUnit:R0,unbindTextureUnit:i_,getProgramUniformLocationOrThrow:O0,getProgramUniformLocation:E0,bindTextureToProgramUniformSampler:D0,bindCanvasToFramebuffer:r_,bindColorTextureToFramebuffer:Wo,unbindColorTextureFromFramebuffer:Dd,validateFramebuffer:ns,getFramebufferErrorMessage:k0,getBatchDim:cr,getRowsCols:hr,getShapeAs3D:Uo,getTextureShapeFromLogicalShape:F0,isReshapeFree:is,getWebGLMaxTextureSize:W0,resetMaxTextureSize:a_,resetMaxTexturesInShader:s_,getMaxTexturesInShader:U0,getWebGLDisjointQueryTimerVersion:B0,hasExtension:an,isWebGLVersionEnabled:kd,isCapableOfRenderingToFloatTexture:z0,isDownloadFloatTextureEnabled:P0,isWebGLFenceEnabled:_0,assertNotComplex:ta};var Se=N.env();Se.registerFlag("HAS_WEBGL",function(){return Se.getNumber("WEBGL_VERSION")>0});Se.registerFlag("WEBGL_VERSION",function(){return kd(2)?2:kd(1)?1:0});Se.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});Se.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return Se.get("WEBGL_VERSION")===2});Se.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});Se.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});Se.registerFlag("WEBGL_PACK",function(){return Se.getBool("HAS_WEBGL")});Se.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_CLIP",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});Se.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_REDUCE",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_LAZILY_UNPACK",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_CONV_IM2COL",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return W0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return U0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=Se.getNumber("WEBGL_VERSION");return n===0?0:B0(n)});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return Se.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!N.device_util.isMobile()});Se.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return z0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return Se.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:Se.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});Se.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return P0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return _0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=Se.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});Se.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function u_(n){const t=new Float32Array(n.length);for(let e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}function Wd(n){return(t,e,i,r,a)=>{const s=N.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=N.util.computeStrides(s),u=N.util.sizeFromShape(s),c=N.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=N.util.computeStrides(t),f=N.util.computeStrides(e),m=N.backend_util.getBroadcastDims(t,s),g=N.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(let v=0;v<c.length;++v){const b=N.util.indexToLoc(v,o,l),w=b.slice(-h);m.forEach(C=>w[C]=0);const S=N.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const x=N.util.locToIndex(L,d,f);c[v]=n(i[S],r[x])}return[c,s]}}const c_=Wd((n,t)=>n+t);function na(n){return(t,e,i)=>{const r=N.util.getTypedArrayFromDType(e,t.length);for(let a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}const h_=na(n=>Math.ceil(n));const d_=na(n=>Math.exp(n));const p_=na(n=>Math.expm1(n));const f_=na(n=>Math.floor(n));const m_=na(n=>Math.log(n));function g_(n,t,e,i){const r=N.util.getTypedArrayFromDType(i,N.util.sizeFromShape(e));for(let a=0;a<r.length;++a){const s=a*t;let o=n[s];for(let l=0;l<t;++l){const u=n[s+l];u>o&&(o=u)}r[a]=o}return r}const v_=Wd((n,t)=>n*t);const y_=na(n=>1/Math.sqrt(n));function b_(n,t,e,i,r){const a=N.slice_util.isSliceContinous(i,t,e),s=N.util.sizeFromShape(e),o=N.util.computeStrides(i);if(a){const u=N.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=N.util.getTypedArrayFromDType(r,s);for(let u=0;u<s;++u){const c=e.length,h=N.util.computeStrides(e),d=N.util.indexToLoc(u,c,h),p=d.map((m,g)=>m+t[g]),f=N.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const w_=Wd((n,t)=>n-t);function S_(n,t,e,i,r){const a=t.length,s=N.util.sizeFromShape(t),o=N.util.computeStrides(t),l=N.util.computeStrides(r),u=N.util.getTypedArrayFromDType(e,N.util.sizeFromShape(r));for(let c=0;c<s;++c){const h=N.util.indexToLoc(c,a,o),d=new Array(h.length);for(let f=0;f<d.length;f++)d[f]=h[i[f]];const p=N.util.locToIndex(d,a,l);u[p]=n[c]}return u}function L_(n,t,e,i){const r=N.util.parseAxisParam(t,e)[0],a=[1,e[0],1];for(let f=0;f<r;f++)a[0]*=e[f];a[1]=e[r];for(let f=r+1;f<e.length;f++)a[2]*=e[f];const s={},o=new Int32Array(e[r]),l=new N.TensorBuffer(a,i,n),u=[],c=a[0]===1&&a[2]===1;for(let f=0;f<e[r];f++){let m;if(c)m=n[f].toString();else{const g=[];for(let v=0;v<a[0];v++)for(let b=0;b<a[2];b++)g.push(l.get(v,f,b));m=g.join(",")}if(s[m]!==void 0)o[f]=s[m];else{const g=Object.keys(s).length;s[m]=g,o[f]=g,u.push(f)}}const h=a.slice();h[1]=Object.keys(s).length;const d=new N.TensorBuffer(h,i);u.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let v=0;v<a[2];v++)d.set(l.get(g,f,v),g,m,v)});const p=e.slice();return p[r]=h[1],{outputValues:d.values,outputShape:p,indices:o}}var I_=u_,A_=c_,T_=h_,N_=d_,x_=p_,C_=f_,R_=m_,O_=g_,E_=v_,D_=y_,k_=b_,F_=w_,M0=S_,W_=L_;var U_=function(){function n(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map(function(a,s){return"T"+s});var i=[];this.variableNames.forEach(function(a){i.push("float v"+a+" = get"+a+"AtOutCoords();")});var r=this.variableNames.map(function(a){return"v"+a}).join(" + ");this.userCode=`
+`))}function S0(n){return ei(n,function(){return n.createProgram()},"Unable to create WebGLProgram.")}function L0(n,t){if(ce(n,function(){return n.linkProgram(t)}),n.getProgramParameter(t,n.LINK_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function Fo(n,t){if(ce(n,function(){return n.validateProgram(t)}),n.getProgramParameter(t,n.VALIDATE_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function I0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function A0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ELEMENT_ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ELEMENT_ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function n_(){return N.env().getNumber("WEBGL_VERSION")===2?1:4}function T0(n){return ei(n,function(){return n.createTexture()},"Unable to create WebGLTexture.")}function N0(n,t){var e=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(n<=0||t<=0){var i="["+n+"x"+t+"]";throw new Error("Requested texture size "+i+" is invalid.")}if(n>e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function x0(n){return ei(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function Dd(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function R0(n,t,e){C0(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function i_(n,t){C0(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function O0(n,t,e){return ei(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function E0(n,t,e){return n.getUniformLocation(t,e)}function D0(n,t,e,i){ce(n,function(){return R0(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function r_(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function Wo(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function kd(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function ns(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+k0(n,t))}function k0(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function ei(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function C0(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(i<n.TEXTURE0||i>e){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function cr(n,t){return t===void 0&&(t=2),N.util.sizeFromShape(n.slice(0,n.length-t))}function hr(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function Uo(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[cr(n)].concat(hr(n))),t}function F0(n,t){var e;t===void 0&&(t=!1);var i=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?N.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=N.util.squeezeShape(n);n=r.newShape}var a=N.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=cr(n),o=2,l=2;return n.length&&(e=hr(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),N.util.sizeToSquarishShape(a).map(function(u){return u*2})}return N.util.sizeToSquarishShape(a)}function Bo(n){return n%2===0}function is(n,t){if(n=n.slice(-2),t=t.slice(-2),N.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(Bo(e)&&Bo(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&Bo(n[0])&&Bo(t[0])}var zo,Po;function W0(n){if(zo==null){var t=_n(n);zo=t.getParameter(t.MAX_TEXTURE_SIZE)}return zo}function a_(){zo=null}function s_(){Po=null}function U0(n){if(Po==null){var t=_n(n);Po=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,Po)}function B0(n){if(n===0)return 0;var t,e=_n(n);return rn(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:rn(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function rn(n,t){var e=n.getExtension(t);return e!=null}function Fd(n){try{var t=_n(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function z0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!rn(t,"OES_texture_float"))return!1}else if(!rn(t,"EXT_color_buffer_float"))return!1;var e=Wd(t);return e}function P0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!rn(t,"OES_texture_float"))return!1;if(!rn(t,"WEBGL_color_buffer_float"))return!1}else{if(rn(t,"EXT_color_buffer_float"))return Wd(t);var e="EXT_color_buffer_half_float";if(rn(t,e)){var i=t.getExtension(e);return o_(t,i)}return!1}var r=Wd(t);return r}function Wd(n){var t=Ed(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function o_(n,t){var e=Ed(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function _0(n){if(n!==2)return!1;var t=_n(n),e=t.fenceSync!=null;return e}function ta(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&N.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var l_={__proto__:null,callAndCheck:ce,canBeRepresented:y0,getWebGLErrorMessage:v0,getExtensionOrThrow:ts,createVertexShader:b0,createFragmentShader:w0,createProgram:S0,linkProgram:L0,validateProgram:Fo,createStaticVertexBuffer:I0,createStaticIndexBuffer:A0,getNumChannels:n_,createTexture:T0,validateTextureSize:N0,createFramebuffer:x0,bindVertexBufferToProgramAttribute:Dd,bindTextureUnit:R0,unbindTextureUnit:i_,getProgramUniformLocationOrThrow:O0,getProgramUniformLocation:E0,bindTextureToProgramUniformSampler:D0,bindCanvasToFramebuffer:r_,bindColorTextureToFramebuffer:Wo,unbindColorTextureFromFramebuffer:kd,validateFramebuffer:ns,getFramebufferErrorMessage:k0,getBatchDim:cr,getRowsCols:hr,getShapeAs3D:Uo,getTextureShapeFromLogicalShape:F0,isReshapeFree:is,getWebGLMaxTextureSize:W0,resetMaxTextureSize:a_,resetMaxTexturesInShader:s_,getMaxTexturesInShader:U0,getWebGLDisjointQueryTimerVersion:B0,hasExtension:rn,isWebGLVersionEnabled:Fd,isCapableOfRenderingToFloatTexture:z0,isDownloadFloatTextureEnabled:P0,isWebGLFenceEnabled:_0,assertNotComplex:ta};var Se=N.env();Se.registerFlag("HAS_WEBGL",function(){return Se.getNumber("WEBGL_VERSION")>0});Se.registerFlag("WEBGL_VERSION",function(){return Fd(2)?2:Fd(1)?1:0});Se.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});Se.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return Se.get("WEBGL_VERSION")===2});Se.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});Se.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});Se.registerFlag("WEBGL_PACK",function(){return Se.getBool("HAS_WEBGL")});Se.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_CLIP",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});Se.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_REDUCE",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_LAZILY_UNPACK",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_CONV_IM2COL",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return W0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return U0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=Se.getNumber("WEBGL_VERSION");return n===0?0:B0(n)});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return Se.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!N.device_util.isMobile()});Se.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return z0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return Se.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:Se.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});Se.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return P0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return _0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=Se.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});Se.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function u_(n){const t=new Float32Array(n.length);for(let e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}function Ud(n){return(t,e,i,r,a)=>{const s=N.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=N.util.computeStrides(s),u=N.util.sizeFromShape(s),c=N.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=N.util.computeStrides(t),f=N.util.computeStrides(e),m=N.backend_util.getBroadcastDims(t,s),g=N.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(let v=0;v<c.length;++v){const b=N.util.indexToLoc(v,o,l),w=b.slice(-h);m.forEach(C=>w[C]=0);const S=N.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const x=N.util.locToIndex(L,d,f);c[v]=n(i[S],r[x])}return[c,s]}}const c_=Ud((n,t)=>n+t);function na(n){return(t,e,i)=>{const r=N.util.getTypedArrayFromDType(e,t.length);for(let a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}const h_=na(n=>Math.ceil(n));const d_=na(n=>Math.exp(n));const p_=na(n=>Math.expm1(n));const f_=na(n=>Math.floor(n));const m_=na(n=>Math.log(n));function g_(n,t,e,i){const r=N.util.getTypedArrayFromDType(i,N.util.sizeFromShape(e));for(let a=0;a<r.length;++a){const s=a*t;let o=n[s];for(let l=0;l<t;++l){const u=n[s+l];u>o&&(o=u)}r[a]=o}return r}const v_=Ud((n,t)=>n*t);const y_=na(n=>1/Math.sqrt(n));function b_(n,t,e,i,r){const a=N.slice_util.isSliceContinous(i,t,e),s=N.util.sizeFromShape(e),o=N.util.computeStrides(i);if(a){const u=N.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=N.util.getTypedArrayFromDType(r,s);for(let u=0;u<s;++u){const c=e.length,h=N.util.computeStrides(e),d=N.util.indexToLoc(u,c,h),p=d.map((m,g)=>m+t[g]),f=N.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const w_=Ud((n,t)=>n-t);function S_(n,t,e,i,r){const a=t.length,s=N.util.sizeFromShape(t),o=N.util.computeStrides(t),l=N.util.computeStrides(r),u=N.util.getTypedArrayFromDType(e,N.util.sizeFromShape(r));for(let c=0;c<s;++c){const h=N.util.indexToLoc(c,a,o),d=new Array(h.length);for(let f=0;f<d.length;f++)d[f]=h[i[f]];const p=N.util.locToIndex(d,a,l);u[p]=n[c]}return u}function L_(n,t,e,i){const r=N.util.parseAxisParam(t,e)[0],a=[1,e[0],1];for(let f=0;f<r;f++)a[0]*=e[f];a[1]=e[r];for(let f=r+1;f<e.length;f++)a[2]*=e[f];const s={},o=new Int32Array(e[r]),l=new N.TensorBuffer(a,i,n),u=[],c=a[0]===1&&a[2]===1;for(let f=0;f<e[r];f++){let m;if(c)m=n[f].toString();else{const g=[];for(let v=0;v<a[0];v++)for(let b=0;b<a[2];b++)g.push(l.get(v,f,b));m=g.join(",")}if(s[m]!==void 0)o[f]=s[m];else{const g=Object.keys(s).length;s[m]=g,o[f]=g,u.push(f)}}const h=a.slice();h[1]=Object.keys(s).length;const d=new N.TensorBuffer(h,i);u.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let v=0;v<a[2];v++)d.set(l.get(g,f,v),g,m,v)});const p=e.slice();return p[r]=h[1],{outputValues:d.values,outputShape:p,indices:o}}var I_=u_,A_=c_,T_=h_,N_=d_,x_=p_,C_=f_,R_=m_,O_=g_,E_=v_,D_=y_,k_=b_,F_=w_,M0=S_,W_=L_;var U_=function(){function n(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map(function(a,s){return"T"+s});var i=[];this.variableNames.forEach(function(a){i.push("float v"+a+" = get"+a+"AtOutCoords();")});var r=this.variableNames.map(function(a){return"v"+a}).join(" + ");this.userCode=`
       void main() {
         `+i.join(`
         `)+`
@@ -69,7 +69,7 @@ Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed thi
         }
         setOutput(float(bestIndex));
       }
-    `}return n}();function H0(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function Jt(n,t){return t===1?[n]:H0(n,t)}function P_(n,t){if(n===1)return"rc";for(var e="",i=0;i<n;i++)e+=t[i],i<n-1&&(e+=",");return e}function Ft(){var n,t,e,i,r,a,s,o,l,u;return N.env().getNumber("WEBGL_VERSION")===2?(n="#version 300 es",t="in",e="out",i="in",r="texture",a="outputColor",s="out vec4 outputColor;",o=`
+    `}return n}();function H0(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function Xt(n,t){return t===1?[n]:H0(n,t)}function P_(n,t){if(n===1)return"rc";for(var e="",i=0;i<n;i++)e+=t[i],i<n-1&&(e+=",");return e}function Ft(){var n,t,e,i,r,a,s,o,l,u;return N.env().getNumber("WEBGL_VERSION")===2?(n="#version 300 es",t="in",e="out",i="in",r="texture",a="outputColor",s="out vec4 outputColor;",o=`
       bool isnan_custom(float val) {
         return (val > 0.0 || val < 0.0) ? false : val != 0.0;
       }
@@ -114,7 +114,7 @@ Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed thi
       ivec4 round(vec4 value) {
         return ivec4(floor(value + vec4(0.5)));
       }
-    `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function dr(n,t,e){e===void 0&&(e="index");var i=N.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function Ud(n){var t=N.util.computeStrides(n).map(function(e){return e.toString()});return`
+    `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function dr(n,t,e){e===void 0&&(e="index");var i=N.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function Bd(n){var t=N.util.computeStrides(n).map(function(e){return e.toString()});return`
   int getFlatIndex(ivec3 coords) {
     return coords.x * `+t[0]+" + coords.y * "+t[1]+` + coords.z;
   }
@@ -715,7 +715,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
       `+d+`
       return get`+i+"("+f+`);
     }
-  `}function Ze(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function aa(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function sa(n,t){return t.map(function(e){return n[e]}).join(", ")}var LM=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,N.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Ze(l),c=Jt("coords",l),h,d;if(s===1){d=l+1;var p=Ze(d);h=`
+  `}function Ze(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function aa(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function sa(n,t){return t.map(function(e){return n[e]}).join(", ")}var LM=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,N.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Ze(l),c=Xt("coords",l),h,d;if(s===1){d=l+1;var p=Ze(d);h=`
         `+p+" sourceLocR = "+p+"("+c.join()+`, 0);
         ++`+c[l-1]+`;
         `+p+" sourceLocG = "+p+"("+c.join()+`, 0);
@@ -731,7 +731,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
         `+u+` sourceLocA = coords;
         --`+c[l-1]+`;
         `+u+` sourceLocB = coords;
-        --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=Jt("sourceLocR",d-1).concat("inIdx.r"),b=Jt("sourceLocG",d-1).concat("inIdx.g"),w=Jt("sourceLocB",d-1).concat("inIdx.b"),S=Jt("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",x=r?"":`
+        --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=Xt("sourceLocR",d-1).concat("inIdx.r"),b=Xt("sourceLocG",d-1).concat("inIdx.g"),w=Xt("sourceLocB",d-1).concat("inIdx.b"),S=Xt("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",x=r?"":`
           inIdx = round(vec4(getBestIndicesAChannel(`+v.join()+`),
                              getBestIndicesAChannel(`+b.join()+`),
                              getBestIndicesAChannel(`+w.join()+`),
@@ -891,7 +891,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
     `}return n}();var $0=`
   if (isnan(a)) return a;
   if (isnan(b)) return b;
-`,Bd="return a + b;",zd="return a - b;",X0="return a * b;",TM=`
+`,zd="return a + b;",Pd="return a - b;",X0="return a * b;",TM=`
   float s = sign(a) * sign(b);
   int ia = round(a);
   int ib = round(b);
@@ -1019,7 +1019,7 @@ return (round(mod(b, 2.0)) != 1) ?
             result.y = (coords + 1) >= `+this.outputShape[0]+` ? 0. : result.y;
             result.z = 0.;
             result.w = 0.;
-          `;else{var l=Jt("coords",a);s+=`
+          `;else{var l=Xt("coords",a);s+=`
             bool nextRowOutOfBounds =
               (`+l[a-2]+" + 1) >= "+this.outputShape[a-2]+`;
             bool nextColOutOfBounds =
@@ -1090,7 +1090,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+r.join(`
         `)+`
       }
-    `}return n}();var i5=function(){function n(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=N.backend_util.computeOutShape(t,e);var i=this.outputShape,r=i.length,a=Ze(r),s=Jt("coords",r),o=["x","y","z","w","u","v"].slice(0,r);this.variableNames=t.map(function(v,b){return"T"+b});var l=new Array(t.length-1);l[0]=t[0][e];for(var u=1;u<l.length;u++)l[u]=l[u-1]+t[u][e];for(var c=o[e],h=o.slice(-2),d=o.join(),p="if ("+c+" < "+l[0]+`) {
+    `}return n}();var i5=function(){function n(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=N.backend_util.computeOutShape(t,e);var i=this.outputShape,r=i.length,a=Ze(r),s=Xt("coords",r),o=["x","y","z","w","u","v"].slice(0,r);this.variableNames=t.map(function(v,b){return"T"+b});var l=new Array(t.length-1);l[0]=t[0][e];for(var u=1;u<l.length;u++)l[u]=l[u-1]+t[u][e];for(var c=o[e],h=o.slice(-2),d=o.join(),p="if ("+c+" < "+l[0]+`) {
         return getChannel(
             getT0(`+d+"), vec2("+h.join()+`));
         }`,u=1;u<l.length;u++){var f=l[u-1];p+=`
@@ -1965,14 +1965,14 @@ return (round(mod(b, 2.0)) != 1) ?
           float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;
           setOutput(val);
       }
-    `}return n}();var g5=function(){function n(t){this.variableNames=["A"],this.outTexUsage=rn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
+    `}return n}();var g5=function(){function n(t){this.variableNames=["A"],this.outTexUsage=nn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
       `+V0+`
 
       void main() {
         float x = getAAtOutCoords();
         `+e.output+` = encode_float(x);
       }
-    `}return n}();var v5=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=rn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
+    `}return n}();var v5=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=nn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
       `+V0+`
 
       void main() {
@@ -1981,7 +1981,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+e.output+` = encode_float(x);
       }
     `}return n}();var y5=function(){function n(t,e,i){i===void 0&&(i=!1),this.variableNames=["A"];var r=Ft(),a=e[0],s=e[1];this.outputShape=t;var o="result";i&&(o="floor(result * 255. + 0.5)"),this.userCode=`
-      `+Ud(t)+`
+      `+Bd(t)+`
 
       void main() {
         ivec3 coords = getOutputCoords();
@@ -2039,7 +2039,7 @@ return (round(mod(b, 2.0)) != 1) ?
             }
           }
         `}this.userCode=`
-      `+Ud(t)+`
+      `+Bd(t)+`
 
       void main() {
         ivec3 coords = getOutputCoords();
@@ -2118,7 +2118,7 @@ return (round(mod(b, 2.0)) != 1) ?
     void main() {
       gl_Position = vec4(clipSpacePos, 1);
       resultUV = uv;
-    }`;return b0(n,e)}function lS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return I0(n,t)}function uS(n){var t=new Uint16Array([0,1,2,2,1,3]);return A0(n,t)}function rs(n,t,e,i,r,a){N0(t,e);var s=T0(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function Pd(n){return n.internalFormatFloat}function cS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Pd(i),i.textureFormatFloat,n.FLOAT)}function _d(n){return n.internalFormatHalfFloat}function hS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,_d(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function Md(n){return n.downloadTextureFormat}function dS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Md(i),n.RGBA,n.UNSIGNED_BYTE)}function Hd(n){return n.internalFormatPackedFloat}function pS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,Hd(i),n.RGBA,n.FLOAT)}function Vd(n){return n.internalFormatPackedHalfFloat}function fS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,Vd(i),n.RGBA,i.textureTypeHalfFloat)}function mS(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=Ed(n,t,"clipSpacePos",e,3,a,i);return s&&Ed(n,t,"uv",e,2,a,r)}function gS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function vS(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function yS(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function bS(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function wS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array($P(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function SS(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(XP(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function LS(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var A5={__proto__:null,createVertexShader:oS,createVertexBuffer:lS,createIndexBuffer:uS,getInternalFormatForFloat32MatrixTexture:Pd,createFloat32MatrixTexture:cS,getInternalFormatForFloat16MatrixTexture:_d,createFloat16MatrixTexture:hS,getInternalFormatForUnsignedBytesMatrixTexture:Md,createUnsignedBytesMatrixTexture:dS,getInternalFormatForPackedMatrixTexture:Hd,createPackedMatrixTexture:pS,getInternalFormatForFloat16PackedMatrixTexture:Vd,createFloat16PackedMatrixTexture:fS,bindVertexProgramAttributeStreams:mS,uploadDenseMatrixToTexture:gS,uploadPixelDataToTexture:vS,createBufferFromOutputTexture:yS,downloadFloat32MatrixFromBuffer:bS,downloadByteEncodedFloatMatrixFromOutputTexture:wS,downloadPackedMatrixFromBuffer:SS,downloadMatrixFromPackedOutputTexture:LS};var IS=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=N.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,g0(e,t)):this.gl=_n(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(N.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ts(this.gl,a),an(this.gl,s))this.textureHalfFloatExtension=ts(this.gl,s);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),an(this.gl,r))this.colorBufferHalfFloatExtension=ts(this.gl,r);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",an(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(an(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=lS(this.gl),this.indexBuffer=uS(this.gl),this.framebuffer=x0(this.gl),this.textureConfig=Od(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return N.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture.  This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),cS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),hS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),dS(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),vS(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),gS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),fS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),pS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(Dd(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return wS(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return SS(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return bS(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=yS(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(N.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return LS(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=w0(e,t),r=oS(e),a=S0(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),L0(e,a),this.debug&&Fo(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=mS(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&Fo(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?O0(this.gl,t,e):E0(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),D0(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=ea(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&Fo(this.gl,this.program),ns(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ts(this.gl,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Do(this,void 0,void 0,function(){var e=this;return ko(this,function(i){switch(i.label){case 0:return[4,N.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=T5(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;N.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),Wo(this.gl,t,this.framebuffer),this.debug&&ns(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(Wo(this.gl,this.outputTexture,this.framebuffer),this.debug&&ns(this.gl)):Dd(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;Wo(r,t,this.framebuffer),this.debug&&ns(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function T5(n){for(var t=0;t<n.length;++t){var e=n[t]();if(!e)break}return t-1}function N5(n,t,e,i){var r=t.userCode,a=e.map(function(g,v){var b={logicalShape:g.shape,texShape:g.isUniform?null:g.texData.texShape,isUniform:g.isUniform,isPacked:g.isUniform?!1:g.texData.isPacked,flatOffset:null};return g.texData!=null&&g.texData.slice!=null&&g.texData.slice.flatOffset>0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=j_(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);N.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p<t.variableNames.length;p++){var f=t.variableNames[p],m=!1;d[f]=n.getUniformLocation(u,f,m),d["offset"+f]=n.getUniformLocation(u,"offset"+f,m)}return{program:t,source:l,webGLProgram:u,uniformLocations:d,inShapeInfos:s,outShapeInfo:o,infLoc:c,nanLoc:h}}function AS(n,t){if(n.length!==t.length)throw Error("Binary was compiled with "+n.length+" inputs, but "+("was executed with "+t.length+" inputs"));n.forEach(function(e,i){var r=e.logicalShape,a=t[i],s=a.shape;if(!N.util.arraysEqual(r,s))throw Error("Binary was compiled with different shapes than "+("the current args. Shapes "+r+" and "+s+" must match"));if(e.isUniform&&a.isUniform)return;var o=e.texShape,l=a.isUniform?null:a.texData.texShape;if(!N.util.arraysEqual(o,l))throw Error("Binary was compiled with different texture shapes than the"+(" current args. Shape "+o+" and "+l+" must match"))})}function x5(n,t,e,i,r){AS(t.inShapeInfos,e),AS([t.outShapeInfo],[i]);var a=i.texData.texture,s=i.texData.texShape;i.texData.isPacked?n.setOutputPackedMatrixTexture(a,s[0],s[1]):n.setOutputMatrixTexture(a,s[0],s[1]),n.setProgram(t.webGLProgram),N.env().getNumber("WEBGL_VERSION")===1&&(t.infLoc!==null&&n.gl.uniform1f(t.infLoc,Infinity)),t.nanLoc!==null&&n.gl.uniform1f(t.nanLoc,NaN),e.forEach(function(o,l){var u=t.program.variableNames[l],c=t.uniformLocations[u],h=t.uniformLocations["offset"+u];if(c==null)return;if(o.isUniform){if(N.util.sizeFromShape(o.shape)<2)n.gl.uniform1f(c,o.uniformValues[0]);else{var d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),n.gl.uniform1fv(c,d)}return}o.texData.slice!=null&&h!=null&&n.gl.uniform1i(h,o.texData.slice.flatOffset),n.setInputMatrixTexture(o.texData.texture,c,l)}),r!=null&&r(n,t.webGLProgram),n.executeProgram()}function C5(n,t,e){var i="";t.concat(e).forEach(function(s){var o=s.texData!=null&&s.texData.slice!=null&&s.texData.slice.flatOffset>0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var R5=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Ft(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var x=0;x<=1;x++)S+=`
+    }`;return b0(n,e)}function lS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return I0(n,t)}function uS(n){var t=new Uint16Array([0,1,2,2,1,3]);return A0(n,t)}function rs(n,t,e,i,r,a){N0(t,e);var s=T0(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function _d(n){return n.internalFormatFloat}function cS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,_d(i),i.textureFormatFloat,n.FLOAT)}function Md(n){return n.internalFormatHalfFloat}function hS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Md(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function Hd(n){return n.downloadTextureFormat}function dS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Hd(i),n.RGBA,n.UNSIGNED_BYTE)}function Vd(n){return n.internalFormatPackedFloat}function pS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,Vd(i),n.RGBA,n.FLOAT)}function qd(n){return n.internalFormatPackedHalfFloat}function fS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,qd(i),n.RGBA,i.textureTypeHalfFloat)}function mS(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=Dd(n,t,"clipSpacePos",e,3,a,i);return s&&Dd(n,t,"uv",e,2,a,r)}function gS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function vS(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function yS(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function bS(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function wS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array($P(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function SS(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(XP(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function LS(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var A5={__proto__:null,createVertexShader:oS,createVertexBuffer:lS,createIndexBuffer:uS,getInternalFormatForFloat32MatrixTexture:_d,createFloat32MatrixTexture:cS,getInternalFormatForFloat16MatrixTexture:Md,createFloat16MatrixTexture:hS,getInternalFormatForUnsignedBytesMatrixTexture:Hd,createUnsignedBytesMatrixTexture:dS,getInternalFormatForPackedMatrixTexture:Vd,createPackedMatrixTexture:pS,getInternalFormatForFloat16PackedMatrixTexture:qd,createFloat16PackedMatrixTexture:fS,bindVertexProgramAttributeStreams:mS,uploadDenseMatrixToTexture:gS,uploadPixelDataToTexture:vS,createBufferFromOutputTexture:yS,downloadFloat32MatrixFromBuffer:bS,downloadByteEncodedFloatMatrixFromOutputTexture:wS,downloadPackedMatrixFromBuffer:SS,downloadMatrixFromPackedOutputTexture:LS};var IS=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=N.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,g0(e,t)):this.gl=_n(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(N.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ts(this.gl,a),rn(this.gl,s))this.textureHalfFloatExtension=ts(this.gl,s);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),rn(this.gl,r))this.colorBufferHalfFloatExtension=ts(this.gl,r);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",rn(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(rn(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=lS(this.gl),this.indexBuffer=uS(this.gl),this.framebuffer=x0(this.gl),this.textureConfig=Ed(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return N.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture.  This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),cS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),hS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),dS(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),vS(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),gS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),fS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),pS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(kd(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return wS(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return SS(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return bS(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=yS(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(N.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return LS(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=w0(e,t),r=oS(e),a=S0(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),L0(e,a),this.debug&&Fo(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=mS(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&Fo(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?O0(this.gl,t,e):E0(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),D0(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=ea(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&Fo(this.gl,this.program),ns(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ts(this.gl,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Do(this,void 0,void 0,function(){var e=this;return ko(this,function(i){switch(i.label){case 0:return[4,N.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=T5(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;N.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),Wo(this.gl,t,this.framebuffer),this.debug&&ns(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(Wo(this.gl,this.outputTexture,this.framebuffer),this.debug&&ns(this.gl)):kd(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;Wo(r,t,this.framebuffer),this.debug&&ns(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function T5(n){for(var t=0;t<n.length;++t){var e=n[t]();if(!e)break}return t-1}function N5(n,t,e,i){var r=t.userCode,a=e.map(function(g,v){var b={logicalShape:g.shape,texShape:g.isUniform?null:g.texData.texShape,isUniform:g.isUniform,isPacked:g.isUniform?!1:g.texData.isPacked,flatOffset:null};return g.texData!=null&&g.texData.slice!=null&&g.texData.slice.flatOffset>0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=j_(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);N.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p<t.variableNames.length;p++){var f=t.variableNames[p],m=!1;d[f]=n.getUniformLocation(u,f,m),d["offset"+f]=n.getUniformLocation(u,"offset"+f,m)}return{program:t,source:l,webGLProgram:u,uniformLocations:d,inShapeInfos:s,outShapeInfo:o,infLoc:c,nanLoc:h}}function AS(n,t){if(n.length!==t.length)throw Error("Binary was compiled with "+n.length+" inputs, but "+("was executed with "+t.length+" inputs"));n.forEach(function(e,i){var r=e.logicalShape,a=t[i],s=a.shape;if(!N.util.arraysEqual(r,s))throw Error("Binary was compiled with different shapes than "+("the current args. Shapes "+r+" and "+s+" must match"));if(e.isUniform&&a.isUniform)return;var o=e.texShape,l=a.isUniform?null:a.texData.texShape;if(!N.util.arraysEqual(o,l))throw Error("Binary was compiled with different texture shapes than the"+(" current args. Shape "+o+" and "+l+" must match"))})}function x5(n,t,e,i,r){AS(t.inShapeInfos,e),AS([t.outShapeInfo],[i]);var a=i.texData.texture,s=i.texData.texShape;i.texData.isPacked?n.setOutputPackedMatrixTexture(a,s[0],s[1]):n.setOutputMatrixTexture(a,s[0],s[1]),n.setProgram(t.webGLProgram),N.env().getNumber("WEBGL_VERSION")===1&&(t.infLoc!==null&&n.gl.uniform1f(t.infLoc,Infinity)),t.nanLoc!==null&&n.gl.uniform1f(t.nanLoc,NaN),e.forEach(function(o,l){var u=t.program.variableNames[l],c=t.uniformLocations[u],h=t.uniformLocations["offset"+u];if(c==null)return;if(o.isUniform){if(N.util.sizeFromShape(o.shape)<2)n.gl.uniform1f(c,o.uniformValues[0]);else{var d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),n.gl.uniform1fv(c,d)}return}o.texData.slice!=null&&h!=null&&n.gl.uniform1i(h,o.texData.slice.flatOffset),n.setInputMatrixTexture(o.texData.texture,c,l)}),r!=null&&r(n,t.webGLProgram),n.executeProgram()}function C5(n,t,e){var i="";t.concat(e).forEach(function(s){var o=s.texData!=null&&s.texData.slice!=null&&s.texData.slice.flatOffset>0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var R5=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Ft(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var x=0;x<=1;x++)S+=`
           blockIndex = rc.y + `+x+`;
           pos = rc.x + `+L+`;
 
@@ -2408,7 +2408,7 @@ return (round(mod(b, 2.0)) != 1) ?
         }
         setOutput(dotProd);
       }
-    `}return n}();var qd=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) {
+    `}return n}();var Gd=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) {
           vec4 b = getPreluActivationWeightsAtOutCoords();
           `+s+`
         }`:f=`vec4 activation(vec4 x) {
@@ -2475,7 +2475,7 @@ return (round(mod(b, 2.0)) != 1) ?
         void main() {
           setOutput(vec4(getA(), 0., 0., 0.));
         }
-      `;else{var i=Jt("rc",e),r=Ze(e),a=B5(e,t,i),s=z5(e,t[t.length-1],t[t.length-2],i),o=P5(t,i);this.userCode=`
+      `;else{var i=Xt("rc",e),r=Ze(e),a=B5(e,t,i),s=z5(e,t[t.length-1],t[t.length-2],i),o=P5(t,i);this.userCode=`
         void main() {
           `+r+` rc = getOutputCoords();
 
@@ -2525,7 +2525,7 @@ return (round(mod(b, 2.0)) != 1) ?
           setOutput(getX(`+l+`));
         }
       }
-    `}return n}();var V5=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Ze(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=Jt("rc",r),u=Jt("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1;
+    `}return n}();var V5=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Ze(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=Xt("rc",r),u=Xt("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1;
        if(`+c+`) {
       `,r===1?"":`}
        rc = outputLoc;
@@ -2690,7 +2690,7 @@ return (round(mod(b, 2.0)) != 1) ?
         }
         setOutput(`+L+`);
       }
-    `}return n}(),Gd=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=`
+    `}return n}(),Yd=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=`
         const ivec3 strides =
             ivec3(`+o+", "+l+", "+u+`);
         const ivec3 pads = ivec3(`+g+", "+v+", "+b+`);
@@ -2951,7 +2951,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+(r>0?"}":"")+`
       `}this.userCode=`
       `+q5(e)+`
-      `+Ud(t)+`
+      `+Bd(t)+`
 
       void main() {
         ivec3 rc = getOutputCoords();
@@ -3265,7 +3265,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+s+` coords = getOutputCoords();
         setOutput(getX(`+a+`));
       }
-    `}return n}();var J5=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=Jt("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Ze(i);i===1?this.userCode=`
+    `}return n}();var J5=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=Xt("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Ze(i);i===1?this.userCode=`
         void main(){
           int rc = getOutputCoords();
           vec4 result = vec4(0.);
@@ -3429,7 +3429,7 @@ return (round(mod(b, 2.0)) != 1) ?
           setOutput(getB(`+a+`));
         }
       }
-    `}return n}();var t9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i="uniform int start["+this.rank+"];",r=e9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+Yd[l]+" = start["+l+"] + coords."+Yd[l]+";"});a=`
+    `}return n}();var t9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i="uniform int start["+this.rank+"];",r=e9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+Kd[l]+" = start["+l+"] + coords."+Kd[l]+";"});a=`
         `+e+` sourceLoc;
         `+e+` coords = getOutputCoords();
         `+s.join(`
@@ -3440,7 +3440,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+a+`
         setOutput(getSource(`+r+`));
       }
-    `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),Yd=["x","y","z","w","u","v"];function e9(n){if(n===1)return"sourceLoc";if(n<=6)return Yd.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var n9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i=Jt("coords",this.rank),r=Jt("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=`
+    `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),Kd=["x","y","z","w","u","v"];function e9(n){if(n===1)return"sourceLoc";if(n<=6)return Kd.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var n9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i=Xt("coords",this.rank),r=Xt("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=`
       result.x = `+s+`;
       if (++`+i[this.rank-1]+" < "+t[this.rank-1]+`) {
         ++`+r[this.rank-1]+`;
@@ -3478,7 +3478,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+s+` coords = getOutputCoords();
         setOutput(getX(`+o+`));
       }
-    `}return n}();var r9=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=RS(e,i),a=OS(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=CS(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Ct.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Ct.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Ct.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=RS(i,r),s=OS(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=CS(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=N.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function a9(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function CS(n,t,e,i,r){var a=s9(t,i),s;if(r){var o=ea(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=Qa(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=a9(e,a);return s*p}function s9(n,t){switch(n){case Ct.PACKED_2X2_FLOAT32:return Hd(t);case Ct.PACKED_2X2_FLOAT16:return Vd(t);case Ct.UNPACKED_FLOAT32:return Pd(t);case Ct.UNPACKED_FLOAT16:return _d(t);case Ct.PACKED_4X1_UNSIGNED_BYTE:return Md(t);default:throw new Error("Unknown physical texture type "+n)}}function o9(n){return N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Ct.PACKED_2X2_FLOAT32:Ct.UNPACKED_FLOAT32:n?Ct.PACKED_2X2_FLOAT16:Ct.UNPACKED_FLOAT16}function RS(n,t){if(n===rn.UPLOAD)return Ct.PACKED_2X2_FLOAT32;if(n===rn.RENDER||n==null)return o9(t);if(n===rn.DOWNLOAD||n===rn.PIXELS)return Ct.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function OS(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var u9=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[r]*e[r];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=l9(t);this.userCode=`
+    `}return n}();var r9=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=RS(e,i),a=OS(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=CS(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Ct.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Ct.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Ct.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=RS(i,r),s=OS(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=CS(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=N.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function a9(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function CS(n,t,e,i,r){var a=s9(t,i),s;if(r){var o=ea(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=Qa(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=a9(e,a);return s*p}function s9(n,t){switch(n){case Ct.PACKED_2X2_FLOAT32:return Vd(t);case Ct.PACKED_2X2_FLOAT16:return qd(t);case Ct.UNPACKED_FLOAT32:return _d(t);case Ct.UNPACKED_FLOAT16:return Md(t);case Ct.PACKED_4X1_UNSIGNED_BYTE:return Hd(t);default:throw new Error("Unknown physical texture type "+n)}}function o9(n){return N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Ct.PACKED_2X2_FLOAT32:Ct.UNPACKED_FLOAT32:n?Ct.PACKED_2X2_FLOAT16:Ct.UNPACKED_FLOAT16}function RS(n,t){if(n===nn.UPLOAD)return Ct.PACKED_2X2_FLOAT32;if(n===nn.RENDER||n==null)return o9(t);if(n===nn.DOWNLOAD||n===nn.PIXELS)return Ct.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function OS(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var u9=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[r]*e[r];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=l9(t);this.userCode=`
       void main() {
         `+a+` resRC = getOutputCoords();
         setOutput(getA(`+s+`));
@@ -3634,14 +3634,14 @@ return (round(mod(b, 2.0)) != 1) ?
 
         setOutput(y);
       }
-    `}return n}();var P9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=Jt("rc",e),r=Ze(e),a=P_(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=`
+    `}return n}();var P9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=Xt("rc",e),r=Ze(e),a=P_(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=`
       void main() {
         `+r+` rc = getOutputCoords();
         vec4 packedInput = getA(`+a+`);
 
         setOutput(getChannel(packedInput, `+o+`));
       }
-    `}return n}();var VS=N.backend_util.segment_util,_9=N.kernel_impls.split,M9=N.kernel_impls.tile,H9=N.kernel_impls.topkImpl,V9=N.kernel_impls.whereImpl,q9=1e-7,G9=1e-4,Vo={};function Y9(n){return n in Vo||(Vo[n]={}),Vo[n]}function qo(n,t){if(t===void 0&&(t=!1),n==="linear")return t?B9:c9;if(n==="relu")return t?_S:DS;if(n==="elu")return t?HS:FS;if(n==="relu6")return t?MS:kS;if(n==="prelu")return t?Z0:J0;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var K9=128,j9=600;function $9(){return N.env().global.screen==null?1024:N.env().global.screen.height*N.env().global.screen.width*window.devicePixelRatio*j9/1024/1024}var qS=1e3,GS=function(n){YP(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!N.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=_n(N.env().getNumber("WEBGL_VERSION"));i.binaryCache=Y9(N.env().getNumber("WEBGL_VERSION")),i.gpgpu=new IS(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new r9(i.gpgpu),i.numMBBeforeWarning=$9(),i.texData=new N.DataStorage(i,N.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((N.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||N.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:rn.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(N.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:rn.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new ss(l,Ho):c=new Re(l,Ho);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=N.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=N.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=N.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R;return ko(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new ss(s,Ho):h=new Re(s,Ho),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&N.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&N.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(es(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=N.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=N.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),x=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(x)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,x]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i<e.length;i++){var r=e[i];if(!y0(r))throw N.env().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error("The value "+r+" cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'"):Error("The value "+r+" cannot be represented on this device.")}},t.prototype.getValuesFromTexture=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.isPacked,l=N.util.sizeFromShape(a);if(N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){var u=this.decode(e),c=this.texData.get(u.dataId),h=(i=this.gpgpu).downloadMatrixFromPackedTexture.apply(i,[c.texture].concat(es(a))).subarray(0,l);return this.disposeIntermediateTensorInfo(u),h}var d=N.env().getBool("WEBGL_PACK")&&o===!0,p=d?Uo(a):a,f=d?new v5(p):new g5(p),m=this.runWebGLProgram(f,[{shape:p,dtype:s,dataId:e}],"float32"),g=this.texData.get(m.dataId),v=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(g.texture,g.texShape[0],g.texShape[1]).subarray(0,l);return this.disposeIntermediateTensorInfo(m),v},t.prototype.time=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u;return ko(this,function(c){switch(c.label){case 0:return i=this.activeTimers,r=[],a=!1,this.programTimersStack==null?(this.programTimersStack=r,a=!0):this.activeTimers.push(r),this.activeTimers=r,e(),s=N.util.flatten(this.activeTimers.map(function(h){return h.query})).filter(function(h){return h!=null}),o=N.util.flatten(this.activeTimers.map(function(h){return h.name})).filter(function(h){return h!=null}),this.activeTimers=i,a&&(this.programTimersStack=null),l={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=N.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:N.util.now(),endMs:null}},t.prototype.endTimer=function(e){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=N.util.now(),e)},t.prototype.getQueryTime=function(e){return Do(this,void 0,void 0,function(){var i;return ko(this,function(r){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return N.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=N.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=K9);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&N.util.sizeFromShape(s.shape)<i})},t.prototype.getGPGPUContext=function(){return this.gpgpu},t.prototype.complex=function(e,i){var r=this.makeOutput(e.shape,"complex64"),a=this.texData.get(r.dataId);return a.complexTensors={real:N.engine().keep(e.clone()),imag:N.engine().keep(i.clone())},r},t.prototype.real=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.real.clone()},t.prototype.imag=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.imag.clone()},t.prototype.slice=function(e,i,r){if(this.shouldExecuteOnCPU([e])){var a=k_(this.texData.get(e.dataId).values,i,r,e.shape,e.dtype);return this.makeOutput(r,e.dtype,a)}if(N.util.sizeFromShape(r)===0)return N.tensor([],r,e.dtype);var s=this.texData.get(e.dataId).isPacked,o=N.slice_util.isSliceContinous(e.shape,i,r);if(s||!o){var l=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new n9(r):new t9(r),u=l.getCustomSetupFunc(i);return this.compileAndRun(l,[e],null,u)}return this.uploadToGPU(e.dataId),this.shallowSlice(e,i,r)},t.prototype.shallowSlice=function(e,i,r){var a=this.texData.get(e.dataId),s=this.makeOutput(r,e.dtype),o=this.texData.get(s.dataId);Object.assign(o,a),o.shape=r,o.dtype=e.dtype;var l=N.slice_util.computeFlatOffset(i,e.strides);a.slice&&(l+=a.slice.flatOffset),o.slice={flatOffset:l,origDataId:a.slice&&a.slice.origDataId||e.dataId};var u=this.dataRefCount.get(o.slice.origDataId)||1;return this.dataRefCount.set(o.slice.origDataId,u+1),s},t.prototype.stridedSlice=function(e,i,r,a){var s=this,o=this.tryRunOnCpuOrThrow([e],function(){return s.cpuBackend.stridedSlice(e,i,r,a)});if(o)return o;var l=N.slice_util.computeOutShape(i,r,a);if(l.some(function(c){return c===0}))return N.tensor([],l);var u=new i9(i,a,l);return this.compileAndRun(u,[e])},t.prototype.reverse=function(e,i){var r=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new J5(e.shape,i):new X5(e.shape,i);return this.compileAndRun(r,[e])},t.prototype.concat=function(e,i){if(e[0].dtype==="complex64"){var r=e.map(function(f){return N.real(f)}),a=e.map(function(f){return N.imag(f)});return N.complex(this.concat(r,i),this.concat(a,i))}if(e.length===1)return e[0];if(e.length>N.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new i5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=N.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,N.util.sizeFromShape(f.shape.slice(i)))}),d=new n5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,WS,e.dtype);var a=new Re(e.shape,WS);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>qS){r&&(e=N.transpose(e,[0,2,1])),a&&(i=N.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=N.upcastType(e.dtype,i.dtype),m=new qd(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=N.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?qo(l,!0):null,b=new qd(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new j0(K0.REAL,e.shape,i.shape),o=new j0(K0.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=E_(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,X0,e.dtype);var g=new wt(X0,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=N.env().getBool("WEBGL_PACK_NORMALIZATION")?new D5(e.shape,i,r,a,s):new O5(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new E5(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return N.util.decodeString(l)}),s=N.buffer(e.shape,e.dtype,a);return M9(s,i)}var o=new u9(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new V5(e.shape,i,r):new H5(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new L5(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=N.backend_util.getReshaped(e.shape,i,a),o=N.backend_util.getPermuted(s.length,i.length),l=N.backend_util.getReshapedPermuted(e.shape,i,a),u=N.backend_util.getSliceBeginCoords(r,i.length),c=N.backend_util.getSliceSize(l,r,i.length);return N.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;o<e.shape.length;++o)s.push([0,0]);var l=e.pad(s),u=N.backend_util.getReshaped(l.shape,i,a,!1),c=N.backend_util.getPermuted(u.length,i.length,!1),h=N.backend_util.getReshapedPermuted(l.shape,i,a,!1),d=N.transpose(l.reshape(u),c);return N.reshape(d,h)},t.prototype.reduce=function(e,i,r){var a=e.shape[0],s=e.shape[1],o=N.backend_util.computeOptimalWindowSize(s),l=Math.ceil(s/o),u={windowSize:o,inSize:s,batchSize:a,outSize:l},c=new TS(u,i),h=this.compileAndRun(c,[e],r);return h.shape[1]===1?h:this.reduce(h,i,r)},t.prototype.argReduce=function(e,i,r){r===void 0&&(r=null);var a=e.shape[0],s=e.shape[1];r!=null&&(a=r.shape[0],s=r.shape[1]);var o=N.backend_util.computeOptimalWindowSize(s),l={windowSize:o,inSize:s,batchSize:a,outSize:Math.ceil(s/o)},u=new z_(l,i,r==null),c=[e];r!=null&&c.push(r);var h=this.compileAndRun(u,c,"int32");return h.shape[1]===1?h:this.argReduce(e,i,h)},t.prototype.argReducePacked=function(e,i,r){r===void 0&&(r=null);var a=r!=null?r.shape:e.shape,s=a[a.length-1],o=N.backend_util.computeOptimalWindowSize(s),l=new LM(a,o,i,r==null),u=r==null?[e]:[e,r],c=this.compileAndRun(l,u,"int32");return c.rank===e.rank?this.argReducePacked(e,i,c):c},t.prototype.sum=function(e,i){N.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o),u=N.sumOutType(e.dtype);return this.reduce(l,"sum",u).reshape(a)},t.prototype.prod=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e],function(){return r.cpuBackend.prod(e,i)});if(a)return a;var s=N.backend_util.computeOutAndReduceShapes(e.shape,i),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u),h=N.sumOutType(e.dtype);return this.reduce(c,"prod",h).reshape(o)},t.prototype.unsortedSegmentSum=function(e,i,r){var a=0,s=N.backend_util.getAxesPermutation([a],e.rank),o=e;s!=null&&(o=N.transpose(e,s),a=N.backend_util.getInnerMostAxes(1,e.rank)[0]);var l=VS.computeOutShape(o.shape,a,r),u=N.util.sizeFromShape([o.shape[a]]),c=o.as2D(-1,u),h=N.sumOutType(e.dtype),d=this.segOpCompute(c,"unsortedSegmentSum",i,h,r).reshape(l);return s!=null&&(d=N.transpose(d,N.backend_util.getUndoAxesPermutation(s))),d},t.prototype.segOpCompute=function(e,i,r,a,s){var o=e.shape[0],l=e.shape[1],u=VS.segOpComputeOptimalWindowSize(l,s),c={windowSize:u,inSize:l,batchSize:o,numSegments:s},h=new Z5(c,i),d=this.compileAndRun(h,[e,r],a);return d.shape[1]===s?d:(r=N.range(0,s).tile([l/u]),this.segOpCompute(d,i,r,a,s))},t.prototype.argMinMaxReduce=function(e,i,r){var a=[i];if(N.backend_util.assertAxesAreInnerMostDims("arg"+r.charAt(0).toUpperCase()+r.slice(1),a,e.rank),!N.env().getBool("WEBGL_PACK_REDUCE")||e.rank<=2){var s=N.backend_util.computeOutAndReduceShapes(e.shape,a),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u);return this.argReduce(c,r).reshape(o)}return this.argReducePacked(e,r)},t.prototype.argMin=function(e,i){return this.argMinMaxReduce(e,i,"min")},t.prototype.argMax=function(e,i){return this.argMinMaxReduce(e,i,"max")},t.prototype.cumsum=function(e,i,r,a){if(i!==e.rank-1)throw new Error("WebGL cumsum shader expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=e.shape[i],o=e,l=0;l<=Math.ceil(Math.log2(s))-1;l++){var u=new rS(e.shape,!1,a),c=u.getCustomSetupFunc(l),h=o;o=this.compileAndRun(u,[o],o.dtype,c),h.dispose()}if(r){var u=new rS(e.shape,r,a),h=o;o=this.compileAndRun(u,[o]),h.dispose()}return o},t.prototype.equal=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,HM,"bool");var r=new wt(xM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.notEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,VM,"bool");var r=new wt(CM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.less=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.less(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,qM,"bool");var s=new wt(RM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.lessEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,GM,"bool");var r=new wt(OM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.greater=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.greater(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,YM,"bool");var s=new wt(EM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.greaterEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,KM,"bool");var r=new wt(DM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalNot=function(e){var i=new Re(e.shape,W9);return this.compileAndRun(i,[e])},t.prototype.logicalAnd=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,jM,"bool");var r=new wt(kM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalOr=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,$M,"bool");var r=new wt(FM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.select=function(e,i,r){var a=new Q5(e.rank,i.shape,i.rank);return this.compileAndRun(a,[e,i,r],N.upcastType(i.dtype,r.dtype))},t.prototype.where=function(e){N.backend_util.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");var i=e.dataSync();return V9(e.shape,i)},t.prototype.topk=function(e,i,r){var a=e.dataSync();return H9(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){N.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"min",l.dtype).reshape(a)},t.prototype.minimum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.minimum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(JM,e.shape,i.shape):new wt(UM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.mod=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(ZM,e.shape,i.shape):new wt(BM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.maximum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.maximum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(XM,e.shape,i.shape):new wt(WM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.all=function(e,i){N.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"all",l.dtype).reshape(a)},t.prototype.any=function(e,i){N.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"any",l.dtype).reshape(a)},t.prototype.floorDiv=function(e,i){var r=TM,a="int32";if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,PM,a);var s=new wt(r,e.shape,i.shape);return this.compileAndRun(s,[e,i],a)},t.prototype.add=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,Bd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=A_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,Bd,r);var c=new wt(Bd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.packedUnaryOp=function(e,i,r){var a=new ss(e.shape,i);return this.compileAndRun(a,[e],r)},t.prototype.packedBinaryOp=function(e,i,r,a,s){s===void 0&&(s=!1);var o=new Ti(r,e.shape,i.shape,s);return this.compileAndRun(o,[e,i],a)},t.prototype.complexSeparableBinaryOp=function(e,i,r){var a=this,s=this.texData.get(e.dataId),o=this.texData.get(i.dataId),l=[[s.complexTensors.real,o.complexTensors.real],[s.complexTensors.imag,o.complexTensors.imag]].map(function(d){var p=d[0],f=d[1],m=a.makeComplexComponentTensorInfo(e,p),g=a.makeComplexComponentTensorInfo(i,f),v=new wt(r,e.shape,i.shape);return a.compileAndRun(v,[m,g],N.upcastType(p.dtype,f.dtype))}),u=l[0],c=l[1],h=this.complex(u,c);return u.dispose(),c.dispose(),h},t.prototype.makeComplexComponentTensorInfo=function(e,i){return{dataId:i.dataId,dtype:i.dtype,shape:e.shape}},t.prototype.addN=function(e){if(e.length===1)return e[0];if(e.length>N.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return N.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=N.env().getBool("WEBGL_PACK"),u=l?new B_(e[0].shape,o):new U_(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,zd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=F_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,zd,e.dtype);var c=new wt(zd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new Ti(_M,e.shape,i.shape):new wt(NM,e.shape,i.shape),s=N.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=T_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,US,e.dtype);var r=new Re(e.shape,US);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=C_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,BS,e.dtype);var r=new Re(e.shape,BS);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,p9);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,f9);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,m9);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,g9);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,v9);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=N_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,zS,e.dtype);var r=new Re(e.shape,zS);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=x_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,PS,e.dtype);var r=new Re(e.shape,PS);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=N.util.parseAxisParam([i],e.shape),a=N.max(e,r),s=N.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return N.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=R_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,z9,e.dtype);var r=new Re(e.shape,y9);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,b9);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,w9);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=D_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,S9);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,F9);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,_S):i=new Re(e.shape,DS),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,MS):i=new Re(e.shape,kS),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(Z0,e.shape,i.shape):new wt(J0,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,HS,e.dtype);var i=new Re(e.shape,FS);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(MM,e.shape,i.shape):new wt(zM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,h9);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,U9);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;N.env().getBool("WEBGL_PACK_CLIP")?a=new e5(e.shape):a=new QM(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=I_(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,ES,e.dtype);var r=new Re(e.shape,ES);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new t5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,L9);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,I9);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,A9);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,T9);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,N9);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,x9);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,C9);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,R9);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,O9);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,E9);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,D9);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,k9);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,d9(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>qS,v=l[2]%2!==0&&!!u.isPacked;if(g||!N.env().getBool("WEBGL_LAZILY_UNPACK")||!N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=N.reshape(e,[1,b,r.inChannels]),S=N.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return N.reshape(L,r.outShape)}var x=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,x,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,N.util.assert(is(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=N.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return N.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,N.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),x=new R5(v,S.shape,r),C=this.compileAndRun(x,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?qo(s,!0):null,W=new qd(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(N.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?qo(o,!1):null,d=new Q0(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(N.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new Q0(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new a5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new r5(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?qo(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new tS(a,d,c,p),this.compileAndRun(f,h)):(f=new eS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new tS(r),this.compileAndRun(a,[e,i])):(a=new eS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new u5(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new l5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new c5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new o5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new s5(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return N.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=this.slice(e,l,u).reshape(a);return c},t.prototype.avgPool3d=function(e,i){var r=new Gd(i,"avg",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.avgPool3dBackprop=function(e,i,r){var a=new AM(r);return this.compileAndRun(a,[e],i.dtype)},t.prototype.maxPool3d=function(e,i){var r=new Gd(i,"max",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.maxPool3dBackprop=function(e,i,r,a){var s=!0,o=new Gd(a,"max",s),l=this.compileAndRun(o,[i]),u=new F5(a),c=this.compileAndRun(u,[e,l],i.dtype);return l.dispose(),c},t.prototype.resizeBilinear=function(e,i,r,a){var s=N.env().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new K5(e.shape,i,r,a):new Y5(e.shape,i,r,a);return this.compileAndRun(s,[e],"float32")},t.prototype.resizeBilinearBackprop=function(e,i,r){var a=new G5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.resizeNearestNeighbor=function(e,i,r,a){var s=new $5(e.shape,i,r,a);return this.compileAndRun(s,[e])},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){var a=new j5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.multinomial=function(e,i,r,a){var s=i?e:N.softmax(e),o=s.shape[0],l=s.shape[1],u=new W5(o,l,r),c=u.getCustomSetupFunc(a);return this.compileAndRun(u,[s],"int32",c)},t.prototype.oneHot=function(e,i,r,a){var s=new U5(e.size,i,r,a);return this.compileAndRun(s,[e])},t.prototype.diag=function(e){var i=new m5(e.size);return this.compileAndRun(i,[e])},t.prototype.cropAndResize=function(e,i,r,a,s,o){var l=new h5(e.shape,i.shape,a,s,o);return this.compileAndRun(l,[e,i,r],"float32")},t.prototype.depthToSpace=function(e,i,r){N.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new f5(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return _9(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=N.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return N.backend_util.reshapeTensor(N.tensor([]),r);var f=N.scalar(0),m=new xS(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=N.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new xS(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new sS(aS.REAL,e.shape,i),s=new sS(aS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=N.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new I5(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||N.util.inferDtype(i),r==="string"){var a=N.util.getArrayFromDType(r,N.util.sizeFromShape(e));return a.fill(i),N.engine().makeTensor(a,e,r,this)}else{var s=new w5(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return N.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return N.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new P9(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new _5(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[cr(e.shape)].concat(hr(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[cr(i)].concat(hr(i)),o=new NS(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=Uo(a),l;r?l=new p5(o):l=new d5(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===Za.DENSE){var c=es(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),N.util.sizeFromShape(l.shape)===0)return u.values=N.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&N.util.sizeFromShape(w.shape)<=N.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!is(S.shape,w.shape)){var L=w,x=w.shape;w.shape=S.shape,w=o.packedReshape(w,x),h.push(w),S=o.texData.get(w.dataId),L.shape=x}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=C5(e,d,p),m=this.getAndSaveBinary(f,function(){return N5(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),x5(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!N.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return N.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!N.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=N.tidy(function(){if(!N.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=N.env().getBool("DEBUG");N.env().set("DEBUG",!1);var r=e.abs(N.scalar(1e-8)).dataSync()[0];if(N.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?q9:G9},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=N.util.now());var p=r.texShape;if(p==null&&(p=F0(a,c),r.texShape=p),o!=null){var f=Uo(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=ea(p[0],p[1]),g=i[0],v=i[1],m=new b5(f,[v,g],b)):m=new y5(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=rn.PIXELS:this.texData.get(w.dataId).usage=rn.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),x=this.texData.get(L.dataId);r.texture=x.texture,r.texShape=x.texShape,r.isPacked=x.isPacked,r.usage=x.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=N.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=X9(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*N.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(N.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(N.KernelBackend);function X9(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i<e.length;++i)e[i]=Math.round(n[i]);return e}else throw new Error("Unknown dtype "+t)}var J9="2.6.0";function YS(){N.env().set("WEBGL_FORCE_F16_TEXTURES",!0)}N.device_util.isBrowser()&&N.registerBackend("webgl",function(){return new GS},2);var Z9={forceHalfFloat:YS},KS="if (isnan(x)) return x;",Q9=`
+    `}return n}();var VS=N.backend_util.segment_util,_9=N.kernel_impls.split,M9=N.kernel_impls.tile,H9=N.kernel_impls.topkImpl,V9=N.kernel_impls.whereImpl,q9=1e-7,G9=1e-4,Vo={};function Y9(n){return n in Vo||(Vo[n]={}),Vo[n]}function qo(n,t){if(t===void 0&&(t=!1),n==="linear")return t?B9:c9;if(n==="relu")return t?_S:DS;if(n==="elu")return t?HS:FS;if(n==="relu6")return t?MS:kS;if(n==="prelu")return t?Z0:J0;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var K9=128,j9=600;function $9(){return N.env().global.screen==null?1024:N.env().global.screen.height*N.env().global.screen.width*window.devicePixelRatio*j9/1024/1024}var qS=1e3,GS=function(n){YP(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!N.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=_n(N.env().getNumber("WEBGL_VERSION"));i.binaryCache=Y9(N.env().getNumber("WEBGL_VERSION")),i.gpgpu=new IS(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new r9(i.gpgpu),i.numMBBeforeWarning=$9(),i.texData=new N.DataStorage(i,N.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((N.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||N.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:nn.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(N.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:nn.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new ss(l,Ho):c=new Re(l,Ho);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=N.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=N.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=N.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R;return ko(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new ss(s,Ho):h=new Re(s,Ho),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&N.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&N.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(es(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=N.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=N.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),x=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(x)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,x]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i<e.length;i++){var r=e[i];if(!y0(r))throw N.env().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error("The value "+r+" cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'"):Error("The value "+r+" cannot be represented on this device.")}},t.prototype.getValuesFromTexture=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.isPacked,l=N.util.sizeFromShape(a);if(N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){var u=this.decode(e),c=this.texData.get(u.dataId),h=(i=this.gpgpu).downloadMatrixFromPackedTexture.apply(i,[c.texture].concat(es(a))).subarray(0,l);return this.disposeIntermediateTensorInfo(u),h}var d=N.env().getBool("WEBGL_PACK")&&o===!0,p=d?Uo(a):a,f=d?new v5(p):new g5(p),m=this.runWebGLProgram(f,[{shape:p,dtype:s,dataId:e}],"float32"),g=this.texData.get(m.dataId),v=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(g.texture,g.texShape[0],g.texShape[1]).subarray(0,l);return this.disposeIntermediateTensorInfo(m),v},t.prototype.time=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u;return ko(this,function(c){switch(c.label){case 0:return i=this.activeTimers,r=[],a=!1,this.programTimersStack==null?(this.programTimersStack=r,a=!0):this.activeTimers.push(r),this.activeTimers=r,e(),s=N.util.flatten(this.activeTimers.map(function(h){return h.query})).filter(function(h){return h!=null}),o=N.util.flatten(this.activeTimers.map(function(h){return h.name})).filter(function(h){return h!=null}),this.activeTimers=i,a&&(this.programTimersStack=null),l={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=N.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:N.util.now(),endMs:null}},t.prototype.endTimer=function(e){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=N.util.now(),e)},t.prototype.getQueryTime=function(e){return Do(this,void 0,void 0,function(){var i;return ko(this,function(r){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return N.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=N.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=K9);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&N.util.sizeFromShape(s.shape)<i})},t.prototype.getGPGPUContext=function(){return this.gpgpu},t.prototype.complex=function(e,i){var r=this.makeOutput(e.shape,"complex64"),a=this.texData.get(r.dataId);return a.complexTensors={real:N.engine().keep(e.clone()),imag:N.engine().keep(i.clone())},r},t.prototype.real=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.real.clone()},t.prototype.imag=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.imag.clone()},t.prototype.slice=function(e,i,r){if(this.shouldExecuteOnCPU([e])){var a=k_(this.texData.get(e.dataId).values,i,r,e.shape,e.dtype);return this.makeOutput(r,e.dtype,a)}if(N.util.sizeFromShape(r)===0)return N.tensor([],r,e.dtype);var s=this.texData.get(e.dataId).isPacked,o=N.slice_util.isSliceContinous(e.shape,i,r);if(s||!o){var l=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new n9(r):new t9(r),u=l.getCustomSetupFunc(i);return this.compileAndRun(l,[e],null,u)}return this.uploadToGPU(e.dataId),this.shallowSlice(e,i,r)},t.prototype.shallowSlice=function(e,i,r){var a=this.texData.get(e.dataId),s=this.makeOutput(r,e.dtype),o=this.texData.get(s.dataId);Object.assign(o,a),o.shape=r,o.dtype=e.dtype;var l=N.slice_util.computeFlatOffset(i,e.strides);a.slice&&(l+=a.slice.flatOffset),o.slice={flatOffset:l,origDataId:a.slice&&a.slice.origDataId||e.dataId};var u=this.dataRefCount.get(o.slice.origDataId)||1;return this.dataRefCount.set(o.slice.origDataId,u+1),s},t.prototype.stridedSlice=function(e,i,r,a){var s=this,o=this.tryRunOnCpuOrThrow([e],function(){return s.cpuBackend.stridedSlice(e,i,r,a)});if(o)return o;var l=N.slice_util.computeOutShape(i,r,a);if(l.some(function(c){return c===0}))return N.tensor([],l);var u=new i9(i,a,l);return this.compileAndRun(u,[e])},t.prototype.reverse=function(e,i){var r=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new J5(e.shape,i):new X5(e.shape,i);return this.compileAndRun(r,[e])},t.prototype.concat=function(e,i){if(e[0].dtype==="complex64"){var r=e.map(function(f){return N.real(f)}),a=e.map(function(f){return N.imag(f)});return N.complex(this.concat(r,i),this.concat(a,i))}if(e.length===1)return e[0];if(e.length>N.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new i5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=N.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,N.util.sizeFromShape(f.shape.slice(i)))}),d=new n5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,WS,e.dtype);var a=new Re(e.shape,WS);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>qS){r&&(e=N.transpose(e,[0,2,1])),a&&(i=N.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=N.upcastType(e.dtype,i.dtype),m=new Gd(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=N.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?qo(l,!0):null,b=new Gd(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new j0(K0.REAL,e.shape,i.shape),o=new j0(K0.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=E_(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,X0,e.dtype);var g=new wt(X0,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=N.env().getBool("WEBGL_PACK_NORMALIZATION")?new D5(e.shape,i,r,a,s):new O5(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new E5(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return N.util.decodeString(l)}),s=N.buffer(e.shape,e.dtype,a);return M9(s,i)}var o=new u9(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new V5(e.shape,i,r):new H5(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new L5(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=N.backend_util.getReshaped(e.shape,i,a),o=N.backend_util.getPermuted(s.length,i.length),l=N.backend_util.getReshapedPermuted(e.shape,i,a),u=N.backend_util.getSliceBeginCoords(r,i.length),c=N.backend_util.getSliceSize(l,r,i.length);return N.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;o<e.shape.length;++o)s.push([0,0]);var l=e.pad(s),u=N.backend_util.getReshaped(l.shape,i,a,!1),c=N.backend_util.getPermuted(u.length,i.length,!1),h=N.backend_util.getReshapedPermuted(l.shape,i,a,!1),d=N.transpose(l.reshape(u),c);return N.reshape(d,h)},t.prototype.reduce=function(e,i,r){var a=e.shape[0],s=e.shape[1],o=N.backend_util.computeOptimalWindowSize(s),l=Math.ceil(s/o),u={windowSize:o,inSize:s,batchSize:a,outSize:l},c=new TS(u,i),h=this.compileAndRun(c,[e],r);return h.shape[1]===1?h:this.reduce(h,i,r)},t.prototype.argReduce=function(e,i,r){r===void 0&&(r=null);var a=e.shape[0],s=e.shape[1];r!=null&&(a=r.shape[0],s=r.shape[1]);var o=N.backend_util.computeOptimalWindowSize(s),l={windowSize:o,inSize:s,batchSize:a,outSize:Math.ceil(s/o)},u=new z_(l,i,r==null),c=[e];r!=null&&c.push(r);var h=this.compileAndRun(u,c,"int32");return h.shape[1]===1?h:this.argReduce(e,i,h)},t.prototype.argReducePacked=function(e,i,r){r===void 0&&(r=null);var a=r!=null?r.shape:e.shape,s=a[a.length-1],o=N.backend_util.computeOptimalWindowSize(s),l=new LM(a,o,i,r==null),u=r==null?[e]:[e,r],c=this.compileAndRun(l,u,"int32");return c.rank===e.rank?this.argReducePacked(e,i,c):c},t.prototype.sum=function(e,i){N.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o),u=N.sumOutType(e.dtype);return this.reduce(l,"sum",u).reshape(a)},t.prototype.prod=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e],function(){return r.cpuBackend.prod(e,i)});if(a)return a;var s=N.backend_util.computeOutAndReduceShapes(e.shape,i),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u),h=N.sumOutType(e.dtype);return this.reduce(c,"prod",h).reshape(o)},t.prototype.unsortedSegmentSum=function(e,i,r){var a=0,s=N.backend_util.getAxesPermutation([a],e.rank),o=e;s!=null&&(o=N.transpose(e,s),a=N.backend_util.getInnerMostAxes(1,e.rank)[0]);var l=VS.computeOutShape(o.shape,a,r),u=N.util.sizeFromShape([o.shape[a]]),c=o.as2D(-1,u),h=N.sumOutType(e.dtype),d=this.segOpCompute(c,"unsortedSegmentSum",i,h,r).reshape(l);return s!=null&&(d=N.transpose(d,N.backend_util.getUndoAxesPermutation(s))),d},t.prototype.segOpCompute=function(e,i,r,a,s){var o=e.shape[0],l=e.shape[1],u=VS.segOpComputeOptimalWindowSize(l,s),c={windowSize:u,inSize:l,batchSize:o,numSegments:s},h=new Z5(c,i),d=this.compileAndRun(h,[e,r],a);return d.shape[1]===s?d:(r=N.range(0,s).tile([l/u]),this.segOpCompute(d,i,r,a,s))},t.prototype.argMinMaxReduce=function(e,i,r){var a=[i];if(N.backend_util.assertAxesAreInnerMostDims("arg"+r.charAt(0).toUpperCase()+r.slice(1),a,e.rank),!N.env().getBool("WEBGL_PACK_REDUCE")||e.rank<=2){var s=N.backend_util.computeOutAndReduceShapes(e.shape,a),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u);return this.argReduce(c,r).reshape(o)}return this.argReducePacked(e,r)},t.prototype.argMin=function(e,i){return this.argMinMaxReduce(e,i,"min")},t.prototype.argMax=function(e,i){return this.argMinMaxReduce(e,i,"max")},t.prototype.cumsum=function(e,i,r,a){if(i!==e.rank-1)throw new Error("WebGL cumsum shader expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=e.shape[i],o=e,l=0;l<=Math.ceil(Math.log2(s))-1;l++){var u=new rS(e.shape,!1,a),c=u.getCustomSetupFunc(l),h=o;o=this.compileAndRun(u,[o],o.dtype,c),h.dispose()}if(r){var u=new rS(e.shape,r,a),h=o;o=this.compileAndRun(u,[o]),h.dispose()}return o},t.prototype.equal=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,HM,"bool");var r=new wt(xM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.notEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,VM,"bool");var r=new wt(CM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.less=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.less(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,qM,"bool");var s=new wt(RM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.lessEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,GM,"bool");var r=new wt(OM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.greater=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.greater(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,YM,"bool");var s=new wt(EM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.greaterEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,KM,"bool");var r=new wt(DM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalNot=function(e){var i=new Re(e.shape,W9);return this.compileAndRun(i,[e])},t.prototype.logicalAnd=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,jM,"bool");var r=new wt(kM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalOr=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,$M,"bool");var r=new wt(FM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.select=function(e,i,r){var a=new Q5(e.rank,i.shape,i.rank);return this.compileAndRun(a,[e,i,r],N.upcastType(i.dtype,r.dtype))},t.prototype.where=function(e){N.backend_util.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");var i=e.dataSync();return V9(e.shape,i)},t.prototype.topk=function(e,i,r){var a=e.dataSync();return H9(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){N.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"min",l.dtype).reshape(a)},t.prototype.minimum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.minimum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(JM,e.shape,i.shape):new wt(UM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.mod=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(ZM,e.shape,i.shape):new wt(BM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.maximum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.maximum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(XM,e.shape,i.shape):new wt(WM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.all=function(e,i){N.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"all",l.dtype).reshape(a)},t.prototype.any=function(e,i){N.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"any",l.dtype).reshape(a)},t.prototype.floorDiv=function(e,i){var r=TM,a="int32";if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,PM,a);var s=new wt(r,e.shape,i.shape);return this.compileAndRun(s,[e,i],a)},t.prototype.add=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,zd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=A_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,zd,r);var c=new wt(zd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.packedUnaryOp=function(e,i,r){var a=new ss(e.shape,i);return this.compileAndRun(a,[e],r)},t.prototype.packedBinaryOp=function(e,i,r,a,s){s===void 0&&(s=!1);var o=new Ti(r,e.shape,i.shape,s);return this.compileAndRun(o,[e,i],a)},t.prototype.complexSeparableBinaryOp=function(e,i,r){var a=this,s=this.texData.get(e.dataId),o=this.texData.get(i.dataId),l=[[s.complexTensors.real,o.complexTensors.real],[s.complexTensors.imag,o.complexTensors.imag]].map(function(d){var p=d[0],f=d[1],m=a.makeComplexComponentTensorInfo(e,p),g=a.makeComplexComponentTensorInfo(i,f),v=new wt(r,e.shape,i.shape);return a.compileAndRun(v,[m,g],N.upcastType(p.dtype,f.dtype))}),u=l[0],c=l[1],h=this.complex(u,c);return u.dispose(),c.dispose(),h},t.prototype.makeComplexComponentTensorInfo=function(e,i){return{dataId:i.dataId,dtype:i.dtype,shape:e.shape}},t.prototype.addN=function(e){if(e.length===1)return e[0];if(e.length>N.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return N.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=N.env().getBool("WEBGL_PACK"),u=l?new B_(e[0].shape,o):new U_(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,Pd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=F_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,Pd,e.dtype);var c=new wt(Pd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new Ti(_M,e.shape,i.shape):new wt(NM,e.shape,i.shape),s=N.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=T_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,US,e.dtype);var r=new Re(e.shape,US);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=C_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,BS,e.dtype);var r=new Re(e.shape,BS);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,p9);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,f9);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,m9);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,g9);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,v9);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=N_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,zS,e.dtype);var r=new Re(e.shape,zS);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=x_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,PS,e.dtype);var r=new Re(e.shape,PS);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=N.util.parseAxisParam([i],e.shape),a=N.max(e,r),s=N.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return N.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=R_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,z9,e.dtype);var r=new Re(e.shape,y9);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,b9);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,w9);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=D_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,S9);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,F9);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,_S):i=new Re(e.shape,DS),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,MS):i=new Re(e.shape,kS),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(Z0,e.shape,i.shape):new wt(J0,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,HS,e.dtype);var i=new Re(e.shape,FS);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(MM,e.shape,i.shape):new wt(zM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,h9);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,U9);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;N.env().getBool("WEBGL_PACK_CLIP")?a=new e5(e.shape):a=new QM(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=I_(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,ES,e.dtype);var r=new Re(e.shape,ES);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new t5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,L9);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,I9);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,A9);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,T9);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,N9);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,x9);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,C9);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,R9);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,O9);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,E9);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,D9);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,k9);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,d9(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>qS,v=l[2]%2!==0&&!!u.isPacked;if(g||!N.env().getBool("WEBGL_LAZILY_UNPACK")||!N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=N.reshape(e,[1,b,r.inChannels]),S=N.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return N.reshape(L,r.outShape)}var x=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,x,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,N.util.assert(is(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=N.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return N.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,N.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),x=new R5(v,S.shape,r),C=this.compileAndRun(x,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?qo(s,!0):null,W=new Gd(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(N.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?qo(o,!1):null,d=new Q0(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(N.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new Q0(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new a5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new r5(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?qo(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new tS(a,d,c,p),this.compileAndRun(f,h)):(f=new eS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new tS(r),this.compileAndRun(a,[e,i])):(a=new eS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new u5(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new l5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new c5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new o5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new s5(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return N.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=this.slice(e,l,u).reshape(a);return c},t.prototype.avgPool3d=function(e,i){var r=new Yd(i,"avg",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.avgPool3dBackprop=function(e,i,r){var a=new AM(r);return this.compileAndRun(a,[e],i.dtype)},t.prototype.maxPool3d=function(e,i){var r=new Yd(i,"max",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.maxPool3dBackprop=function(e,i,r,a){var s=!0,o=new Yd(a,"max",s),l=this.compileAndRun(o,[i]),u=new F5(a),c=this.compileAndRun(u,[e,l],i.dtype);return l.dispose(),c},t.prototype.resizeBilinear=function(e,i,r,a){var s=N.env().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new K5(e.shape,i,r,a):new Y5(e.shape,i,r,a);return this.compileAndRun(s,[e],"float32")},t.prototype.resizeBilinearBackprop=function(e,i,r){var a=new G5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.resizeNearestNeighbor=function(e,i,r,a){var s=new $5(e.shape,i,r,a);return this.compileAndRun(s,[e])},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){var a=new j5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.multinomial=function(e,i,r,a){var s=i?e:N.softmax(e),o=s.shape[0],l=s.shape[1],u=new W5(o,l,r),c=u.getCustomSetupFunc(a);return this.compileAndRun(u,[s],"int32",c)},t.prototype.oneHot=function(e,i,r,a){var s=new U5(e.size,i,r,a);return this.compileAndRun(s,[e])},t.prototype.diag=function(e){var i=new m5(e.size);return this.compileAndRun(i,[e])},t.prototype.cropAndResize=function(e,i,r,a,s,o){var l=new h5(e.shape,i.shape,a,s,o);return this.compileAndRun(l,[e,i,r],"float32")},t.prototype.depthToSpace=function(e,i,r){N.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new f5(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return _9(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=N.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return N.backend_util.reshapeTensor(N.tensor([]),r);var f=N.scalar(0),m=new xS(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=N.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new xS(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new sS(aS.REAL,e.shape,i),s=new sS(aS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=N.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new I5(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||N.util.inferDtype(i),r==="string"){var a=N.util.getArrayFromDType(r,N.util.sizeFromShape(e));return a.fill(i),N.engine().makeTensor(a,e,r,this)}else{var s=new w5(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return N.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return N.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new P9(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new _5(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[cr(e.shape)].concat(hr(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[cr(i)].concat(hr(i)),o=new NS(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=Uo(a),l;r?l=new p5(o):l=new d5(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===Za.DENSE){var c=es(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),N.util.sizeFromShape(l.shape)===0)return u.values=N.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&N.util.sizeFromShape(w.shape)<=N.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!is(S.shape,w.shape)){var L=w,x=w.shape;w.shape=S.shape,w=o.packedReshape(w,x),h.push(w),S=o.texData.get(w.dataId),L.shape=x}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=C5(e,d,p),m=this.getAndSaveBinary(f,function(){return N5(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),x5(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!N.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return N.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!N.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=N.tidy(function(){if(!N.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=N.env().getBool("DEBUG");N.env().set("DEBUG",!1);var r=e.abs(N.scalar(1e-8)).dataSync()[0];if(N.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?q9:G9},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=N.util.now());var p=r.texShape;if(p==null&&(p=F0(a,c),r.texShape=p),o!=null){var f=Uo(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=ea(p[0],p[1]),g=i[0],v=i[1],m=new b5(f,[v,g],b)):m=new y5(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=nn.PIXELS:this.texData.get(w.dataId).usage=nn.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),x=this.texData.get(L.dataId);r.texture=x.texture,r.texShape=x.texShape,r.isPacked=x.isPacked,r.usage=x.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=N.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=X9(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*N.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(N.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(N.KernelBackend);function X9(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i<e.length;++i)e[i]=Math.round(n[i]);return e}else throw new Error("Unknown dtype "+t)}var J9="2.6.0";function YS(){N.env().set("WEBGL_FORCE_F16_TEXTURES",!0)}N.device_util.isBrowser()&&N.registerBackend("webgl",function(){return new GS},2);var Z9={forceHalfFloat:YS},KS="if (isnan(x)) return x;",Q9=`
   if (isnan(a)) return a;
   if (isnan(b)) return b;
 `,e6=`
@@ -3649,14 +3649,14 @@ return (round(mod(b, 2.0)) != 1) ?
   result.g = isNaN.g > 0. ? NAN : result.g;
   result.b = isNaN.b > 0. ? NAN : result.b;
   result.a = isNaN.a > 0. ? NAN : result.a;
-`;function Go(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function Kd(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(t,l.shape,u.shape,!!e):new wt(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var t6=Q9+`
+`;function Go(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function jd(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(t,l.shape,u.shape,!!e):new wt(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var t6=Q9+`
   return atan(a, b);
 `,n6=`
   vec4 result = atan(a, b);
   vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
   `+e6+`
   return result;
-`,i6=Kd(t6,n6),r6={kernelName:N.Atan2,backendName:"webgl",kernelFunc:i6};function jd(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var a6={kernelName:N.Identity,backendName:"webgl",kernelFunc:jd};function s6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return jd({inputs:{x:r},backend:e});var h=new as(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var o6={kernelName:N.AvgPool,backendName:"webgl",kernelFunc:s6};function l6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ta([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=N.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new IM(c);return e.runWebGLProgram(h,[r],s.dtype)}var u6={kernelName:N.AvgPoolBackprop,backendName:"webgl",kernelFunc:l6};var c6=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],N.backend_util.assertAndGetBroadcastShape(t,e),N.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(N.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(N.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
+`,i6=jd(t6,n6),r6={kernelName:N.Atan2,backendName:"webgl",kernelFunc:i6};function $d(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var a6={kernelName:N.Identity,backendName:"webgl",kernelFunc:$d};function s6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return $d({inputs:{x:r},backend:e});var h=new as(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var o6={kernelName:N.AvgPool,backendName:"webgl",kernelFunc:s6};function l6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ta([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=N.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new IM(c);return e.runWebGLProgram(h,[r],s.dtype)}var u6={kernelName:N.AvgPoolBackprop,backendName:"webgl",kernelFunc:l6};var c6=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],N.backend_util.assertAndGetBroadcastShape(t,e),N.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(N.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(N.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
       void main() {
         float x = getXAtOutCoords();
         float mean = getMeanAtOutCoords();
@@ -3703,7 +3703,7 @@ return a / b;`,y6=`
   }
 
   return result;
-`,b6=Kd(v6,y6,!0),w6={kernelName:N.Div,backendName:"webgl",kernelFunc:b6};var S6=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=`
+`,b6=jd(v6,y6,!0),w6={kernelName:N.Div,backendName:"webgl",kernelFunc:b6};var S6=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=`
         void main() {
           ivec4 coords = getOutputCoords();
           int x = coords[2];
@@ -3773,7 +3773,7 @@ return a / b;`,y6=`
 
         `+e.output+` = result;
       }
-    `}return n}();var N6={kernelName:N.FromPixels,backendName:"webgl",kernelFunc:T6},oa;function T6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(oa==null&&(oa=document.createElement("canvas").getContext("2d")),oa.canvas.width=u,oa.canvas.height=c,oa.drawImage(r,0,0,u,c),r=oa.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=rn.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=N.env().getBool("WEBGL_PACK")?new A6(d):new I6(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function x6(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=N.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function C6(n,t,e,i){for(var r=x6(n.shape),a=n,s=0;s<r.length;s++){var o=r[s],l=o.inSize,u=o.windowSize,c=o.outSize,h=new TS({windowSize:u,inSize:l,batchSize:n.shape[0],outSize:c},e),d=a;a=i.runWebGLProgram(h,[a],t),d.dataId!==n.dataId&&i.disposeData(d.dataId)}return a}function R6(n,t,e){var i=[cr(n.shape)].concat(hr(n.shape)),r={dtype:n.dtype,shape:i,dataId:n.dataId},a=[cr(t)].concat(hr(t)),s=new NS(a,i),o=!0,l=e.runWebGLProgram(s,[r],n.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function $d(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=e,o=N.util.sizeFromShape(r.shape),l=N.util.inferFromImplicitShape(a,o),u=N.util.sizeFromShape(l);N.util.assert(o===u,function(){return"The new shape ("+l+") has "+u+" elements and the old "+("shape ("+r.shape+") has "+o+" elements. The new shape and old ")+"shape must have the same number of elements."});var c=s.texData.get(r.dataId);return c.isPacked&&!is(r.shape,l)&&!(c.texture!==null&&is(c.shape,l))?R6(r,l,s):(s.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var O6={kernelName:N.Reshape,backendName:"webgl",kernelFunc:$d};function E6(n,t,e,i){var r=N.util.sizeFromShape(t),a=N.util.sizeFromShape(n.shape),s=a/r,o=$d({inputs:{x:n},attrs:{shape:[s,r]},backend:i}),l=C6(o,n.dtype,"max",i),u=$d({inputs:{x:l},attrs:{shape:e},backend:i});return i.disposeIntermediateTensorInfo(o),i.disposeIntermediateTensorInfo(l),u}var k6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[e[r]];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=D6(e);this.userCode=`
+    `}return n}();var N6={kernelName:N.FromPixels,backendName:"webgl",kernelFunc:T6},oa;function T6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(oa==null&&(oa=document.createElement("canvas").getContext("2d")),oa.canvas.width=u,oa.canvas.height=c,oa.drawImage(r,0,0,u,c),r=oa.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=nn.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=N.env().getBool("WEBGL_PACK")?new A6(d):new I6(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function x6(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=N.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function C6(n,t,e,i){for(var r=x6(n.shape),a=n,s=0;s<r.length;s++){var o=r[s],l=o.inSize,u=o.windowSize,c=o.outSize,h=new TS({windowSize:u,inSize:l,batchSize:n.shape[0],outSize:c},e),d=a;a=i.runWebGLProgram(h,[a],t),d.dataId!==n.dataId&&i.disposeData(d.dataId)}return a}function R6(n,t,e){var i=[cr(n.shape)].concat(hr(n.shape)),r={dtype:n.dtype,shape:i,dataId:n.dataId},a=[cr(t)].concat(hr(t)),s=new NS(a,i),o=!0,l=e.runWebGLProgram(s,[r],n.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function Xd(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=e,o=N.util.sizeFromShape(r.shape),l=N.util.inferFromImplicitShape(a,o),u=N.util.sizeFromShape(l);N.util.assert(o===u,function(){return"The new shape ("+l+") has "+u+" elements and the old "+("shape ("+r.shape+") has "+o+" elements. The new shape and old ")+"shape must have the same number of elements."});var c=s.texData.get(r.dataId);return c.isPacked&&!is(r.shape,l)&&!(c.texture!==null&&is(c.shape,l))?R6(r,l,s):(s.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var O6={kernelName:N.Reshape,backendName:"webgl",kernelFunc:Xd};function E6(n,t,e,i){var r=N.util.sizeFromShape(t),a=N.util.sizeFromShape(n.shape),s=a/r,o=Xd({inputs:{x:n},attrs:{shape:[s,r]},backend:i}),l=C6(o,n.dtype,"max",i),u=Xd({inputs:{x:l},attrs:{shape:e},backend:i});return i.disposeIntermediateTensorInfo(o),i.disposeIntermediateTensorInfo(l),u}var k6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[e[r]];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=D6(e);this.userCode=`
     void main() {
       `+a+` resRC = getOutputCoords();
       setOutput(getA(`+s+`));
@@ -3795,7 +3795,7 @@ return a / b;`,y6=`
       }
       setOutput(result);
     }
-    `}return n}();function jS(n,t,e){var i=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new F6(n.shape,t):new k6(n.shape,t);return e.runWebGLProgram(i,[n],n.dtype)}var W6={kernelName:N.Max,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape.length,c=N.util.parseAxisParam(s,r.shape),h=c,d=N.backend_util.getAxesPermutation(h,u),p=d!=null,f=l.shouldExecuteOnCPU([r]),m=r;if(p){if(f){for(var g=l.texData.get(m.dataId),v=g.values,b=new Array(u),w=0;w<b.length;w++)b[w]=r.shape[d[w]];var S=M0(v,r.shape,r.dtype,d,b);m=l.makeTensorInfo(b,r.dtype);var L=l.texData.get(m.dataId);L.values=S}else m=jS(r,d,l);h=N.backend_util.getInnerMostAxes(h.length,u)}N.backend_util.assertAxesAreInnerMostDims("max",h,u);var x=N.backend_util.computeOutAndReduceShapes(m.shape,h),C=x[0],R=x[1],D=C;o&&(D=N.backend_util.expandShapeToKeepDim(C,c));var k;if(f){var g=l.texData.get(m.dataId),v=g.values,W=O_(v,N.util.sizeFromShape(R),D,r.dtype);k=l.makeTensorInfo(D,r.dtype);var F=l.texData.get(k.dataId);F.values=W}else k=E6(m,R,D,l);return p&&l.disposeIntermediateTensorInfo(m),k}};function U6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return jd({inputs:{x:r},backend:e});var h=new as(c,"max",!1);return e.runWebGLProgram(h,[r],r.dtype)}var B6={kernelName:N.MaxPool,backendName:"webgl",kernelFunc:U6};function z6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ta([a,s],"maxPoolBackprop");var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=N.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=!0,f=new as(d,"max",p),m=e.runWebGLProgram(f,[o],o.dtype),g=new k5(d),v=e.runWebGLProgram(g,[r,m],o.dtype);return e.disposeIntermediateTensorInfo(m),v}var P6={kernelName:N.MaxPoolBackprop,backendName:"webgl",kernelFunc:z6};function _6(n,t,e,i){var r=new as(e,"max",!1),a=i.runWebGLProgram(r,[n],"float32");r=new as(e,"max",!0,!0,t);var s=i.runWebGLProgram(r,[n],"float32");return[a,s]}var M6={kernelName:N.MaxPoolWithArgmax,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;N.util.assert(r.shape.length===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+r.shape.length+"."});var h=[1,1];N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(o,h),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+o+" and dilations '"+h+"'")});var d=N.backend_util.computePool2DInfo(r.shape,s,o,h,l),p=_6(r,u,d,c),f=p[0],m=p[1];return[f,m]}};var H6={kernelName:N.NonMaxSuppressionV3,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=e,d=h.readSync(a.dataId),p=h.readSync(s.dataId),f=l,m=u,g=c;return N.kernel_impls.nonMaxSuppressionV3Impl(d,p,f,m,g)}};var V6=N.kernel_impls.nonMaxSuppressionV4Impl,q6={kernelName:N.NonMaxSuppressionV4,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=V6(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var G6=N.kernel_impls.nonMaxSuppressionV5Impl,Y6={kernelName:N.NonMaxSuppressionV5,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=l,g=u,v=c,b=h,w=G6(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var K6=function(){function n(t,e,i,r){this.variableNames=["Image"],this.outputShape=[];var a=t[1],s=t[2],o=Math.sin(e).toFixed(3),l=Math.cos(e).toFixed(3);this.outputShape=t;var u=N.backend_util.getImageCenter(r,a,s),c=u[0],h=u[1],d=c.toFixed(3),p=h.toFixed(3),f="";typeof i=="number"?f="float outputValue = "+i.toFixed(2)+";":f=`
+    `}return n}();function jS(n,t,e){var i=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new F6(n.shape,t):new k6(n.shape,t);return e.runWebGLProgram(i,[n],n.dtype)}var W6={kernelName:N.Max,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape.length,c=N.util.parseAxisParam(s,r.shape),h=c,d=N.backend_util.getAxesPermutation(h,u),p=d!=null,f=l.shouldExecuteOnCPU([r]),m=r;if(p){if(f){for(var g=l.texData.get(m.dataId),v=g.values,b=new Array(u),w=0;w<b.length;w++)b[w]=r.shape[d[w]];var S=M0(v,r.shape,r.dtype,d,b);m=l.makeTensorInfo(b,r.dtype);var L=l.texData.get(m.dataId);L.values=S}else m=jS(r,d,l);h=N.backend_util.getInnerMostAxes(h.length,u)}N.backend_util.assertAxesAreInnerMostDims("max",h,u);var x=N.backend_util.computeOutAndReduceShapes(m.shape,h),C=x[0],R=x[1],D=C;o&&(D=N.backend_util.expandShapeToKeepDim(C,c));var k;if(f){var g=l.texData.get(m.dataId),v=g.values,W=O_(v,N.util.sizeFromShape(R),D,r.dtype);k=l.makeTensorInfo(D,r.dtype);var F=l.texData.get(k.dataId);F.values=W}else k=E6(m,R,D,l);return p&&l.disposeIntermediateTensorInfo(m),k}};function U6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return $d({inputs:{x:r},backend:e});var h=new as(c,"max",!1);return e.runWebGLProgram(h,[r],r.dtype)}var B6={kernelName:N.MaxPool,backendName:"webgl",kernelFunc:U6};function z6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ta([a,s],"maxPoolBackprop");var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=N.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=!0,f=new as(d,"max",p),m=e.runWebGLProgram(f,[o],o.dtype),g=new k5(d),v=e.runWebGLProgram(g,[r,m],o.dtype);return e.disposeIntermediateTensorInfo(m),v}var P6={kernelName:N.MaxPoolBackprop,backendName:"webgl",kernelFunc:z6};function _6(n,t,e,i){var r=new as(e,"max",!1),a=i.runWebGLProgram(r,[n],"float32");r=new as(e,"max",!0,!0,t);var s=i.runWebGLProgram(r,[n],"float32");return[a,s]}var M6={kernelName:N.MaxPoolWithArgmax,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;N.util.assert(r.shape.length===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+r.shape.length+"."});var h=[1,1];N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(o,h),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+o+" and dilations '"+h+"'")});var d=N.backend_util.computePool2DInfo(r.shape,s,o,h,l),p=_6(r,u,d,c),f=p[0],m=p[1];return[f,m]}};var H6={kernelName:N.NonMaxSuppressionV3,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=e,d=h.readSync(a.dataId),p=h.readSync(s.dataId),f=l,m=u,g=c;return N.kernel_impls.nonMaxSuppressionV3Impl(d,p,f,m,g)}};var V6=N.kernel_impls.nonMaxSuppressionV4Impl,q6={kernelName:N.NonMaxSuppressionV4,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=V6(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var G6=N.kernel_impls.nonMaxSuppressionV5Impl,Y6={kernelName:N.NonMaxSuppressionV5,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=l,g=u,v=c,b=h,w=G6(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var K6=function(){function n(t,e,i,r){this.variableNames=["Image"],this.outputShape=[];var a=t[1],s=t[2],o=Math.sin(e).toFixed(3),l=Math.cos(e).toFixed(3);this.outputShape=t;var u=N.backend_util.getImageCenter(r,a,s),c=u[0],h=u[1],d=c.toFixed(3),p=h.toFixed(3),f="";typeof i=="number"?f="float outputValue = "+i.toFixed(2)+";":f=`
         vec3 fill = vec3(`+i.join(",")+`);
         float outputValue = fill[coords[3]];`,this.userCode=`
         void main() {
@@ -3814,7 +3814,7 @@ return a / b;`,y6=`
         }
     `}return n}();var j6={kernelName:N.RotateWithOffset,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=new K6(r.shape,s,o,l),h=u.runWebGLProgram(c,[r],r.dtype);return h}};var $6=KS+`
   return sin(x);
-`,X6=Go($6),J6={kernelName:N.Sin,backendName:"webgl",kernelFunc:X6};var Z6="return x * x;",Q6=Go(Z6),e8={kernelName:N.Square,backendName:"webgl",kernelFunc:Q6};var $S="return (a - b) * (a - b);",t8=Kd($S,$S),n8={kernelName:N.SquaredDifference,backendName:"webgl",kernelFunc:t8};var i8="return tan(x);",r8=Go(i8),a8={kernelName:N.Tan,backendName:"webgl",kernelFunc:r8};var s8={kernelName:N.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u<l.length;u++)l[u]=r.shape[a[u]];var c;if(s.shouldExecuteOnCPU([r])){var h=s.texData.get(r.dataId),d=h.values,p=M0(d,r.shape,r.dtype,a,l);c=s.makeTensorInfo(l,r.dtype);var f=s.texData.get(c.dataId);f.values=p}else c=jS(r,a,s);return c}};function o8(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ta(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");var s=i.readSync(a.dataId),o=W_(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var l8={kernelName:N.Unique,backendName:"webgl",kernelFunc:o8};var u8=[r6,o6,u6,p6,g6,w6,L6,N6,a6,W6,B6,P6,M6,H6,q6,Y6,O6,j6,J6,e8,n8,a8,s8,l8];for(var Xd=0,XS=u8;Xd<XS.length;Xd++){var c8=XS[Xd];N.registerKernel(c8)}Pn.GPGPUContext=IS;Pn.MathBackendWebGL=GS;Pn.forceHalfFloat=YS;Pn.gpgpu_util=A5;Pn.setWebGLContext=g0;Pn.version_webgl=J9;Pn.webgl=Z9;Pn.webgl_util=l_});var Wt=be(fr=>{"use strict";Object.defineProperty(fr,"__esModule",{value:!0});var Jd=Xi(),Zd=Yb(),Qd=ow(),ZS=Cw(),h8=m0(),d8=JS();var p8="2.6.0";var f8={"tfjs-core":Jd.version_core,"tfjs-backend-cpu":h8.version_cpu,"tfjs-backend-webgl":d8.version_webgl,"tfjs-data":ZS.version_data,"tfjs-layers":Zd.version_layers,"tfjs-converter":Qd.version_converter,tfjs:p8};Object.keys(Jd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Jd[n]}})});Object.keys(Zd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Zd[n]}})});Object.keys(Qd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Qd[n]}})});fr.data=ZS;fr.version=f8});var iL=be(Yo=>{const ke=Wt(),QS=6;function m8(n){const t={strides:[n/16,n/8],anchors:[2,6]},e=[];for(let i=0;i<t.strides.length;i++){const r=t.strides[i],a=Math.floor((n+r-1)/r),s=Math.floor((n+r-1)/r),o=t.anchors[i];for(let l=0;l<a;l++){const u=r*(l+.5);for(let c=0;c<s;c++){const h=r*(c+.5);for(let d=0;d<o;d++)e.push([h,u])}}}return e}const eL=n=>{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},tL=n=>({startEndTensor:n,startPoint:ke.slice(n,[0,0],[-1,2]),endPoint:ke.slice(n,[0,2],[-1,2])}),g8=(n,t)=>{const e=ke.mul(n.startPoint,t),i=ke.mul(n.endPoint,t),r=ke.concat2d([e,i],1);return tL(r)};function v8(n,t,e){const i=ke.slice(n,[0,1],[-1,2]),r=ke.add(i,t),a=ke.slice(n,[0,3],[-1,2]),s=ke.div(a,e),o=ke.div(r,e),l=ke.div(s,2),u=ke.sub(o,l),c=ke.add(o,l),h=ke.mul(u,e),d=ke.mul(c,e),p=1;return ke.concat2d([h,d],p)}function y8(n,t){return ke.tidy(()=>{const e=n.box?n.box:n;return g8(e,t).startEndTensor.squeeze()})}class nL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=m8(t.detector.inputSize),this.anchors=ke.tensor2d(this.anchorsData),this.inputSize=ke.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){if(!n||n.isDisposedInternal||n.shape.length!==4||n.shape[1]<1||n.shape[2]<1)return null;const[t,e,i]=ke.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=ke.mul(ke.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u);let h;if(Array.isArray(c)){const m=c.sort((w,S)=>w.size-S.size),g=ke.concat([m[0],m[2]],2),v=ke.concat([m[1],m[3]],2),b=ke.concat([v,g],1);h=b.squeeze(0)}else h=c.squeeze();const d=v8(h,this.anchors,this.inputSize),p=ke.slice(h,[0,0],[-1,1]),f=ke.sigmoid(p).squeeze();return[h,d,f]}),r=await ke.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>ke.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l<s.length;l++){const u=s[l],c=ke.tidy(()=>{const h=tL(u),d=a[l],p=this.anchorsData[d],f=ke.slice(t,[d,QS-1],[1,-1]).squeeze().reshape([QS,-1]),m=ke.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.width,n.shape[1]/this.height]}}async estimateFaces(n){const t=ke.tidy(()=>(n instanceof ke.Tensor||(n=ke.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:e,scaleFactor:i}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(e.map(async r=>{const a=y8(r,i),[s,o,l]=await Promise.all([r.landmarks,a,r.probability].map(async f=>f.array())),u=r.anchor,[c,h]=i,d=s.map(f=>[(f[0]+u[0])*c,(f[1]+u[1])*h]),p={topLeft:o.slice(0,2),bottomRight:o.slice(2),landmarks:d,probability:l};return eL(r.box),r.landmarks.dispose(),r.probability.dispose(),a.dispose(),p}))}}async function b8(n){const t=await ke.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new nL(t,n);return e}Yo.load=b8;Yo.BlazeFaceModel=nL;Yo.disposeBox=eL});var tp=be(ep=>{ep.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};ep.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var rL=be(mr=>{const w8=Wt();function S8(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}mr.scaleBoxCoordinates=S8;function np(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}mr.getBoxSize=np;function ip(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}mr.getBoxCenter=ip;function L8(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return w8.image.cropAndResize(t,a,[0],e)}mr.cutBoxFromImageAndResize=L8;function I8(n,t=1.5){const e=ip(n),i=np(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.enlargeBox=I8;function A8(n){const t=ip(n),e=np(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.squarifyBox=A8});var uL=be(An=>{An.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function aL(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}An.normalizeRadians=aL;function T8(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return aL(e)}An.computeRotation=T8;function N8(n){return n*180/Math.PI}An.radToDegrees=N8;function sL(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function la(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}An.dot=la;function oL(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}An.getColumnFrom2DArr=oL;function lL(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(la(n[r],oL(t,a)))}return e}function x8(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=sL(t[0],t[1]),s=lL(a,r),o=sL(-t[0],-t[1]);return lL(s,o)}An.buildRotationMatrix=x8;function C8(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-la(t[0],e),-la(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}An.invertTransformMatrix=C8;function R8(n,t){return[la(n,t[0]),la(n,t[1])]}An.rotatePoint=R8;function O8(n,t){return Math.sqrt((n[0]-t[0])**2+(n[1]-t[1])**2)}An.xyDistanceBetweenPoints=O8});var hL=be(cL=>{const Ni=Wt(),Mn=rL(),xi=tp(),Ci=uL(),E8=468,D8=.25,k8=13,F8=[k8,xi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],W8=3,U8=2,B8=[W8,U8],rp=xi.MESH_ANNOTATIONS.leftEyeLower0,ap=[rp[0],rp[rp.length-1]],sp=xi.MESH_ANNOTATIONS.rightEyeLower0,op=[sp[0],sp[sp.length-1]],z8=3,P8=4,_8=71,lp=76;function Ko(n,t,e,i){for(let r=0;r<xi.MESH_TO_IRIS_INDICES_MAP.length;r++){const{key:a,indices:s}=xi.MESH_TO_IRIS_INDICES_MAP[r],o=xi.MESH_ANNOTATIONS[`${e}${a}`],l=i==null;if(l||i.includes(a))for(let u=0;u<s.length;u++){const c=s[u];n[o[u]]=[t[c][0],t[c][1],(t[c][2]+n[o[u]][2])/2]}}}class M8{constructor(n,t,e,i){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.irisModel=e,this.meshWidth=i.mesh.inputSize,this.meshHeight=i.mesh.inputSize,this.irisSize=i.iris.inputSize,this.irisEnlarge=i.iris.enlargeFactor}transformRawCoords(n,t,e,i){const r=Mn.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Ci.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Ci.rotatePoint(d,o),d[2]]),u=Ci.invertTransformMatrix(i),c=[...Mn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Ci.dot(c,u[0]),Ci.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[ap[0]][2],e=n[op[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Mn.squarifyBox(Mn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Mn.getBoxSize(a);let o=Ni.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Ni.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a<lp;a++){const s=n[a*3],o=n[a*3+1],l=n[a*3+2];r.push([(i?1-s/this.irisSize:s/this.irisSize)*e[0]+t.startPoint[0],o/this.irisSize*e[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(_8)}}getAdjustedIrisCoords(n,t,e){const i=n[xi.MESH_ANNOTATIONS[`${e}EyeUpper0`][z8]][2],r=n[xi.MESH_ANNOTATIONS[`${e}EyeLower0`][P8]][2],a=(i+r)/2;return t.map((s,o)=>{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Mn.scaleBoxCoordinates(o,r),u=Mn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Ni.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=E8;let[o,l]=F8;s===!1&&([o,l]=B8),a=Ci.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Mn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Ci.IDENTITY_MATRIX;a!==0&&(h=Ni.image.rotateWithOffset(n,a,0,c),d=Ci.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Mn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Ni.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:x,boxSize:C,crop:R}=this.getEyeBox(b,f,ap[0],ap[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,op[0],op[1]),F=this.irisModel.predict(Ni.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,lp*3),{rawCoords:_,iris:K}=this.getEyeCoords(H,x,C,!0),j=P.slice(lp*3),{rawCoords:q,iris:G}=this.getEyeCoords(j,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(Ko(b,_,"left"),Ko(b,q,"right")):Z<1?Ko(b,_,"left",["EyeUpper0","EyeLower0"]):Ko(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,K,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Ni.dispose(b);const S=Mn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const x=Ni.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:x.arraySync()};const C={coords:x,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t<n.length;t++){const e=n[t],i=this.regionsOfInterest[t];let r=0;if(i&&i.startPoint){const[a,s]=e.startPoint,[o,l]=e.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}r<D8&&(this.regionsOfInterest[t]=e)}this.regionsOfInterest=this.regionsOfInterest.slice(0,n.length)}clearRegionOfInterest(n){this.regionsOfInterest[n]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,n),...this.regionsOfInterest.slice(n+1)])}shouldUpdateRegionsOfInterest(){const n=this.regionsOfInterest.length,t=n===0;return this.maxFaces===1||t?t:n!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}cL.Pipeline=M8});var pL=be(dL=>{dL.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var fL=be(H8=>{xp(H8,{default:()=>V8});var V8=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var vL=be(os=>{const gr=Wt(),q8=iL(),mL=tp(),G8=hL(),Y8=pL(),K8=fL().default;class gL{constructor(n,t,e,i){this.pipeline=new G8.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=gr.tidy(()=>(n instanceof gr.Tensor||(n=gr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);gr.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in mL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=mL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?gr.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function j8(n){const t=await Promise.all([q8.load(n),gr.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),gr.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new gL(t[0],t[1],t[2],n);return e}os.load=j8;os.MediaPipeFaceMesh=gL;os.uv_coords=Y8;os.triangulation=K8});var bL=be(jo=>{const sn=Wt(),Ri={};let yL={age:0,gender:""},up=0;async function $8(n,t){const e=sn.browser.fromPixels(n),i=sn.image.resizeBilinear(e,[t,t]),r=sn.cast(sn.expandDims(i,0),"float32");return r}async function X8(n){return Ri.age||(Ri.age=await sn.loadGraphModel(n.face.age.modelPath)),Ri.age}async function J8(n){return Ri.gender||(Ri.gender=await sn.loadGraphModel(n.face.gender.modelPath)),Ri.gender}async function Z8(n,t){if(up+=1,up>=t.face.age.skipFrames)return up=0,yL;let e;if(n instanceof sn.Tensor){const r=sn.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=sn.mul(r,[255]),sn.dispose(r)}else e=await $8(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await Ri.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,sn.dispose(r)}if(t.face.gender.enabled){const r=await Ri.gender.predict(e),a=await r.data();i.gender=Math.trunc(100*a[0])<50?"female":"male",sn.dispose(r)}return sn.dispose(e),yL=i,i}jo.predict=Z8;jo.loadAge=X8;jo.loadGender=J8});var LL=be(cp=>{const Ut=Wt(),Q8=["angry","discust","fear","happy","sad","surpise","neutral"],$o={};let wL=[],hp=0;const SL=1.5;function e7(n,t){const e=Ut.tidy(()=>{const i=Ut.browser.fromPixels(n,1),r=Ut.image.resizeBilinear(i,[t,t]),a=Ut.cast(Ut.expandDims(r,0),"float32");return a});return e}async function t7(n){return $o.emotion||($o.emotion=await Ut.loadGraphModel(n.face.emotion.modelPath)),$o.emotion}async function n7(n,t){if(hp+=1,hp>=t.face.emotion.skipFrames)return hp=0,wL;const e=Ut.tidy(()=>{if(n instanceof Ut.Tensor){const r=Ut.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Ut.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Ut.mul(a,[.2989]),u=Ut.mul(s,[.587]),c=Ut.mul(o,[.114]),h=Ut.addN([l,u,c]);return h}return s}return e7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await $o.emotion.predict(e),a=await r.data();for(let s=0;s<a.length;s++)SL*a[s]>t.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*SL*a[s])/100),emotion:Q8[s]});i.sort((s,o)=>o.score-s.score),Ut.dispose(r)}return Ut.dispose(e),wL=i,i}cp.predict=n7;cp.load=t7});var TL=be(IL=>{const AL=Wt();class i7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;AL.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return AL.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}IL.BaseModel=i7});var dp=be(NL=>{const xL=Wt(),r7=TL();class a7 extends r7.BaseModel{preprocessInput(n){return xL.tidy(()=>xL.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}NL.MobileNet=a7});var RL=be(CL=>{function pp(n){return Math.floor(n/2)}class s7{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(pp(n),n);)this.exchange(n,pp(n)),n=pp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(n,t))break;this.exchange(n,t),n=t}}getValueAt(n){return this.getElementValue(this.priorityQueue[n])}less(n,t){return this.getValueAt(n)<this.getValueAt(t)}exchange(n,t){const e=this.priorityQueue[n];this.priorityQueue[n]=this.priorityQueue[t],this.priorityQueue[t]=e}}CL.MaxHeap=s7});var EL=be(OL=>{const o7=RL();function l7(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;h<c;++h){const d=Math.max(i-r,0),p=Math.min(i+r+1,o);for(let f=d;f<p;++f)if(a.get(h,f,n)>t){l=!1;break}if(!l)break}return l}function u7(n,t,e){const[i,r,a]=e.shape,s=new o7.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o<i;++o)for(let l=0;l<r;++l)for(let u=0;u<a;++u){const c=e.get(o,l,u);if(c<n)continue;l7(u,c,o,l,t,e)&&s.enqueue({score:c,part:{heatmapY:o,heatmapX:l,id:u}})}return s}OL.buildPartWithScoreQueue=u7});var ls=be(Tn=>{Tn.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];Tn.NUM_KEYPOINTS=Tn.partNames.length;Tn.partIds=Tn.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const c7=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];Tn.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];Tn.connectedPartIndices=c7.map(([n,t])=>[Tn.partIds[n],Tn.partIds[t]]);Tn.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var mp=be(Oi=>{const h7=ls();function DL(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+h7.NUM_KEYPOINTS)}}Oi.getOffsetPoint=DL;function d7(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=DL(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Oi.getImageCoords=d7;function p7(n,t){const e=new Array(t);for(let i=0;i<t;i++)e[i]=n;return e}Oi.fillArray=p7;function fp(n,t,e){return n<t?t:n>e?e:n}Oi.clamp=fp;function f7(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Oi.squaredDistance=f7;function m7(n,t){return{x:n.x+t.x,y:n.y+t.y}}Oi.addVectors=m7;function g7(n,t,e){return{y:fp(n.y,t,e),x:fp(n.x,t,e)}}Oi.clampVector=g7});var BL=be(kL=>{const us=ls(),ua=mp(),FL=us.poseChain.map(([n,t])=>[us.partIds[n],us.partIds[t]]),gp=FL.map(([,n])=>n),WL=FL.map(([n])=>n);function v7(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function vp(n,t,e,i){return{y:ua.clamp(Math.round(n.y/t),0,e-1),x:ua.clamp(Math.round(n.x/t),0,i-1)}}function UL(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=vp(t.position,a,l,u),h=v7(n,c,s),d=ua.addVectors(t.position,h);let p=d;for(let g=0;g<o;g++){const v=vp(p,a,l,u),b=ua.getOffsetPoint(v.y,v.x,e,r);p=ua.addVectors({x:v.x*a,y:v.y*a},{x:b.x,y:b.y})}const f=vp(p,a,l,u),m=i.get(f.y,f.x,e);return{position:p,part:us.partNames[e],score:m}}function y7(n,t,e,i,r,a){const s=t.shape[2],o=gp.length,l=new Array(s),{part:u,score:c}=n,h=ua.getImageCoords(u,i,e);l[u.id]={score:c,part:us.partNames[u.id],position:h};for(let d=o-1;d>=0;--d){const p=gp[d],f=WL[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,a))}for(let d=0;d<o;++d){const p=WL[d],f=gp[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,r))}return l}kL.decodePose=y7});var yp=be(zL=>{const b7=EL(),w7=BL(),PL=mp();function _L(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return PL.squaredDistance(i,e,s.y,s.x)<=t})}function S7(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(_L(n,t,a,o)||(r+=s),r),0);return i/e.length}const L7=1;function I7(n,t,e,i,r,a,s=.5,o=20){const l=[],u=b7.buildPartWithScoreQueue(s,L7,n),c=o*o;for(;l.length<a&&!u.empty();){const h=u.dequeue(),d=PL.getImageCoords(h.part,r,t);if(_L(l,c,d,h.part.id))continue;const p=w7.decodePose(h,n,t,r,e,i),f=S7(l,c,p);l.push({keypoints:p,score:f})}return l}zL.decodeMultiplePoses=I7});var wp=be(dn=>{const ca=Wt(),A7=ls();function T7(n,t,e){return n<e||t<e}function N7(n,t){return A7.connectedPartIndices.reduce((e,[i,r])=>(T7(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}dn.getAdjacentKeyPoints=N7;const{NEGATIVE_INFINITY:ML,POSITIVE_INFINITY:HL}=Number;function VL(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:ML,maxY:ML,minX:HL,minY:HL})}dn.getBoundingBox=VL;function x7(n){const{minX:t,minY:e,maxX:i,maxY:r}=VL(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}dn.getBoundingBoxPoints=x7;async function C7(n){return Promise.all(n.map(t=>t.buffer()))}dn.toTensorBuffers3D=C7;function qL(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}dn.scalePose=qL;function GL(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>qL(a,t,e,i,r))}dn.scalePoses=GL;function YL(n){return n instanceof ca.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}dn.getInputTensorDimensions=YL;function bp(n){return n instanceof ca.Tensor?n:ca.browser.fromPixels(n)}dn.toInputTensor=bp;function R7(n,t,e){return ca.tidy(()=>{const i=bp(n);return i.resizeBilinear([t,e])})}dn.toResizedInputTensor=R7;function O7(n,[t,e]){const[i,r]=YL(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s<a?(o=0,l=0,u=Math.round(.5*(a*i-r)),c=Math.round(.5*(a*i-r))):(o=Math.round(.5*(1/a*r-i)),l=Math.round(.5*(1/a*r-i)),u=0,c=0);const h=ca.tidy(()=>{let d=bp(n);return d=ca.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}dn.padAndResizeTo=O7;function E7(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=GL(n,s,o,-a.top,-a.left);return l}dn.scaleAndFlipPoses=E7});var jL=be(Sp=>{const D7=Wt(),k7=dp(),F7=yp(),Xo=wp();class KL{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=Xo.getInputTensorDimensions(n),{resized:a,padding:s}=Xo.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await Xo.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await F7.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=Xo.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}Sp.PoseNet=KL;async function W7(n){const t=await D7.loadGraphModel(n.modelPath),e=new k7.MobileNet(t,n.outputStride);return new KL(e)}async function U7(n){return W7(n)}Sp.load=U7});var XL=be(Zt=>{const B7=dp(),$L=jL(),z7=yp(),Jo=ls(),cs=wp();Zt.load=$L.load;Zt.PoseNet=$L.PoseNet;Zt.MobileNet=B7.MobileNet;Zt.decodeMultiplePoses=z7.decodeMultiplePoses;Zt.partChannels=Jo.partChannels;Zt.partIds=Jo.partIds;Zt.partNames=Jo.partNames;Zt.poseChain=Jo.poseChain;Zt.getAdjacentKeyPoints=cs.getAdjacentKeyPoints;Zt.getBoundingBox=cs.getBoundingBox;Zt.getBoundingBoxPoints=cs.getBoundingBoxPoints;Zt.scaleAndFlipPoses=cs.scaleAndFlipPoses;Zt.scalePose=cs.scalePose});var Ap=be(Ei=>{const P7=Wt();function Lp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Ei.getBoxSize=Lp;function Ip(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Ei.getBoxCenter=Ip;function _7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return P7.image.cropAndResize(t,a,[0],e)}Ei.cutBoxFromImageAndResize=_7;function M7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}Ei.scaleBoxCoordinates=M7;function H7(n,t=1.5){const e=Ip(n),i=Lp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.enlargeBox=H7;function V7(n){const t=Ip(n),e=Lp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.squarifyBox=V7;function q7(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}Ei.shiftBox=q7});var ZL=be(JL=>{const qe=Wt(),G7=Ap();class Y7{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=qe.tensor2d(this.anchors),this.inputSizeTensor=qe.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=qe.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return qe.tidy(()=>{const t=qe.slice(n,[0,0],[-1,2]),e=qe.slice(n,[0,2],[-1,2]),i=qe.add(qe.div(t,this.inputSizeTensor),this.anchorsTensor),r=qe.div(e,this.doubleInputSizeTensor),a=qe.mul(qe.sub(i,r),this.inputSizeTensor),s=qe.mul(qe.add(i,r),this.inputSizeTensor);return qe.concat2d([a,s],1)})}normalizeLandmarks(n,t){return qe.tidy(()=>{const e=qe.add(qe.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return qe.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=qe.tidy(()=>qe.mul(qe.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=qe.tidy(()=>qe.sigmoid(qe.slice(i,[0,0],[-1,1])).squeeze()),a=qe.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await qe.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=qe.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=qe.slice(s,[p,0],[1,-1]),m=qe.slice(i,[p,5],[1,14]),g=qe.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=qe.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(G7.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}JL.HandDetector=Y7});var eI=be(QL=>{QL.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var aI=be(Di=>{function tI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Di.normalizeRadians=tI;function K7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return tI(e)}Di.computeRotation=K7;const nI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function ha(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Di.dot=ha;function iI(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Di.getColumnFrom2DArr=iI;function rI(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(ha(n[r],iI(t,a)))}return e}function j7(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=nI(t[0],t[1]),s=rI(a,r),o=nI(-t[0],-t[1]);return rI(s,o)}Di.buildRotationMatrix=j7;function $7(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-ha(t[0],e),-ha(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Di.invertTransformMatrix=$7;function X7(n,t){return[ha(n,t[0]),ha(n,t[1])]}Di.rotatePoint=X7});var uI=be(sI=>{const oI=Wt(),Hn=Ap(),ki=aI(),J7=.8,Z7=[0,-.4],Q7=[0,-.1],eH=1.65,lI=[0,5,9,13,17,1,2],tH=0,nH=2;class iH{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return ki.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(i,Z7)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(t,Q7)),eH),i=[];for(let r=0;r<lI.length;r++)i.push(n[lI[r]].slice(0,2));return e.palmLandmarks=i,e}transformRawCoords(n,t,e,i){const r=Hn.getBoxSize(t),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=ki.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=ki.rotatePoint(d,o);return[...p,d[2]]}),u=ki.invertTransformMatrix(i),c=[...Hn.getBoxCenter(t),1],h=[ki.dot(c,u[0]),ki.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=ki.computeRotation(a.palmLandmarks[tH],a.palmLandmarks[nH]),o=Hn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=oI.image.rotateWithOffset(n,s,0,l),c=ki.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=Hn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),v<t.minConfidence)return g.dispose(),this.regionsOfInterest[r]=[],i;const b=oI.reshape(g,[-1,3]),w=await b.array();g.dispose(),b.dispose();const S=this.transformRawCoords(w,h,s,c),L=this.getBoxForHandLandmarks(S);this.updateRegionsOfInterest(L,!1,r);const x={landmarks:S,confidence:v,box:{topLeft:L.startPoint,bottomRight:L.endPoint}};i.push(x)}return i}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>J7?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}sI.HandPipeline=iH});var dI=be(Tp=>{const vr=Wt(),rH=ZL(),cI=eI(),aH=uI();class hI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=vr.tidy(()=>(n instanceof vr.Tensor||(n=vr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(cI.MESH_ANNOTATIONS))s[o]=cI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}Tp.HandPose=hI;async function sH(n){if(vr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return vr.util.fetch(n).then(t=>t.json())}async function oH(n){const[t,e,i]=await Promise.all([sH(n.detector.anchors),vr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),vr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new rH.HandDetector(e,t,n),a=new aH.HandPipeline(r,i,n),s=new hI(a);return s}Tp.load=oH});var pI=be(lH=>{xp(lH,{default:()=>uH});var uH={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:5,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:2,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var mI=be((oV,fI)=>{fI.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});var LI=be(Nn=>{const qt=Wt(),gI=vL(),Zo=bL(),vI=LL(),yI=XL(),bI=dI(),wI=pI().default,cH=mI();let Pe;const Et={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},Vn=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),Qo=(...n)=>{Pe.console&&console.log(...n)};function SI(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=SI(a,s):e[r]=s}),e),{})}function hH(n){if(!n)return"input is not defined";const t=n.naturalWidth||n.videoWidth||n.width||n.shape&&n.shape[1]>0;if(!t||t===0)return"input is empty";if(n.readyState&&n.readyState<=2)return"input is not ready";try{qt.getBackend()}catch{return"backend not loaded"}return null}async function dH(n,t){Pe=SI(wI,t);const e=hH(n);return e?(Qo(e,n),{error:e}):new Promise(async i=>{const r=Object.values(Et).filter(h=>h).length;r===0&&Qo("Human library starting"),qt.getBackend()!==Pe.backend&&(Qo("Human library setting backend:",Pe.backend),await qt.setBackend(Pe.backend),await qt.ready());let a;qt.getBackend()==="webgl"&&(a=qt.env().get("WEBGL_PACK_DEPTHWISECONV"),qt.env().set("WEBGL_PACK_DEPTHWISECONV",!0)),Pe.face.enabled&&!Et.facemesh&&(Et.facemesh=await gI.load(Pe.face)),Pe.body.enabled&&!Et.posenet&&(Et.posenet=await yI.load(Pe.body)),Pe.hand.enabled&&!Et.handpose&&(Et.handpose=await bI.load(Pe.hand)),Pe.face.enabled&&Pe.face.age.enabled&&!Et.age&&(Et.age=await Zo.loadAge(Pe)),Pe.face.enabled&&Pe.face.gender.enabled&&!Et.gender&&(Et.gender=await Zo.loadGender(Pe)),Pe.face.enabled&&Pe.face.emotion.enabled&&!Et.emotion&&(Et.emotion=await vI.load(Pe));const s={};let o;o=Vn(),qt.engine().startScope();const l=Pe.body.enabled?await Et.posenet.estimatePoses(n,Pe.body):[];qt.engine().endScope(),s.body=Math.trunc(Vn()-o),o=Vn(),qt.engine().startScope();const u=Pe.hand.enabled?await Et.handpose.estimateHands(n,Pe.hand):[];qt.engine().endScope(),s.hand=Math.trunc(Vn()-o);const c=[];if(Pe.face.enabled){o=Vn(),qt.engine().startScope();const h=await Et.facemesh.estimateFaces(n,Pe.face);s.face=Math.trunc(Vn()-o);for(const d of h){if(!d.image||d.image.isDisposedInternal){Qo("face object is disposed:",d.image);continue}o=Vn();const p=Pe.face.age.enabled||Pe.face.gender.enabled?await Zo.predict(d.image,Pe):{};s.agegender=Math.trunc(Vn()-o),o=Vn();const f=Pe.face.emotion.enabled?await vI.predict(d.image,Pe):{};s.emotion=Math.trunc(Vn()-o),d.image.dispose();const m=d.annotations.leftEyeIris&&d.annotations.rightEyeIris?Math.max(d.annotations.leftEyeIris[3][0]-d.annotations.leftEyeIris[1][0],d.annotations.rightEyeIris[3][0]-d.annotations.rightEyeIris[1][0]):0;c.push({confidence:d.confidence,box:d.box,mesh:d.mesh,annotations:d.annotations,age:p.age,gender:p.gender,emotion:f,iris:m!==0?Math.trunc(100*11.7/m)/100:0})}qt.engine().endScope()}qt.env().set("WEBGL_PACK_DEPTHWISECONV",a),s.total=Object.values(s).reduce((h,d)=>h+d),i({face:c,body:l,hand:u,performance:s})})}Nn.detect=dH;Nn.defaults=wI;Nn.config=Pe;Nn.models=Et;Nn.facemesh=gI;Nn.ssrnet=Zo;Nn.posenet=yI;Nn.handpose=bI;Nn.tf=qt;Nn.version=cH.version});export default LI();
+`,X6=Go($6),J6={kernelName:N.Sin,backendName:"webgl",kernelFunc:X6};var Z6="return x * x;",Q6=Go(Z6),e8={kernelName:N.Square,backendName:"webgl",kernelFunc:Q6};var $S="return (a - b) * (a - b);",t8=jd($S,$S),n8={kernelName:N.SquaredDifference,backendName:"webgl",kernelFunc:t8};var i8="return tan(x);",r8=Go(i8),a8={kernelName:N.Tan,backendName:"webgl",kernelFunc:r8};var s8={kernelName:N.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u<l.length;u++)l[u]=r.shape[a[u]];var c;if(s.shouldExecuteOnCPU([r])){var h=s.texData.get(r.dataId),d=h.values,p=M0(d,r.shape,r.dtype,a,l);c=s.makeTensorInfo(l,r.dtype);var f=s.texData.get(c.dataId);f.values=p}else c=jS(r,a,s);return c}};function o8(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ta(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");var s=i.readSync(a.dataId),o=W_(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var l8={kernelName:N.Unique,backendName:"webgl",kernelFunc:o8};var u8=[r6,o6,u6,p6,g6,w6,L6,N6,a6,W6,B6,P6,M6,H6,q6,Y6,O6,j6,J6,e8,n8,a8,s8,l8];for(var Jd=0,XS=u8;Jd<XS.length;Jd++){var c8=XS[Jd];N.registerKernel(c8)}Pn.GPGPUContext=IS;Pn.MathBackendWebGL=GS;Pn.forceHalfFloat=YS;Pn.gpgpu_util=A5;Pn.setWebGLContext=g0;Pn.version_webgl=J9;Pn.webgl=Z9;Pn.webgl_util=l_});var Wt=be(fr=>{"use strict";Object.defineProperty(fr,"__esModule",{value:!0});var Zd=Xi(),Qd=Yb(),ep=ow(),ZS=Cw(),h8=m0(),d8=JS();var p8="2.6.0";var f8={"tfjs-core":Zd.version_core,"tfjs-backend-cpu":h8.version_cpu,"tfjs-backend-webgl":d8.version_webgl,"tfjs-data":ZS.version_data,"tfjs-layers":Qd.version_layers,"tfjs-converter":ep.version_converter,tfjs:p8};Object.keys(Zd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Zd[n]}})});Object.keys(Qd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Qd[n]}})});Object.keys(ep).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return ep[n]}})});fr.data=ZS;fr.version=f8});var iL=be(Yo=>{const ke=Wt(),QS=6;function m8(n){const t={strides:[n/16,n/8],anchors:[2,6]},e=[];for(let i=0;i<t.strides.length;i++){const r=t.strides[i],a=Math.floor((n+r-1)/r),s=Math.floor((n+r-1)/r),o=t.anchors[i];for(let l=0;l<a;l++){const u=r*(l+.5);for(let c=0;c<s;c++){const h=r*(c+.5);for(let d=0;d<o;d++)e.push([h,u])}}}return e}const eL=n=>{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},tL=n=>({startEndTensor:n,startPoint:ke.slice(n,[0,0],[-1,2]),endPoint:ke.slice(n,[0,2],[-1,2])}),g8=(n,t)=>{const e=ke.mul(n.startPoint,t),i=ke.mul(n.endPoint,t),r=ke.concat2d([e,i],1);return tL(r)};function v8(n,t,e){const i=ke.slice(n,[0,1],[-1,2]),r=ke.add(i,t),a=ke.slice(n,[0,3],[-1,2]),s=ke.div(a,e),o=ke.div(r,e),l=ke.div(s,2),u=ke.sub(o,l),c=ke.add(o,l),h=ke.mul(u,e),d=ke.mul(c,e),p=1;return ke.concat2d([h,d],p)}function y8(n,t){return ke.tidy(()=>{const e=n.box?n.box:n;return g8(e,t).startEndTensor.squeeze()})}class nL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=m8(t.detector.inputSize),this.anchors=ke.tensor2d(this.anchorsData),this.inputSize=ke.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){if(!n||n.isDisposedInternal||n.shape.length!==4||n.shape[1]<1||n.shape[2]<1)return null;const[t,e,i]=ke.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=ke.mul(ke.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u);let h;if(Array.isArray(c)){const m=c.sort((w,S)=>w.size-S.size),g=ke.concat([m[0],m[2]],2),v=ke.concat([m[1],m[3]],2),b=ke.concat([v,g],1);h=b.squeeze(0)}else h=c.squeeze();const d=v8(h,this.anchors,this.inputSize),p=ke.slice(h,[0,0],[-1,1]),f=ke.sigmoid(p).squeeze();return[h,d,f]}),r=await ke.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>ke.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l<s.length;l++){const u=s[l],c=ke.tidy(()=>{const h=tL(u),d=a[l],p=this.anchorsData[d],f=ke.slice(t,[d,QS-1],[1,-1]).squeeze().reshape([QS,-1]),m=ke.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.width,n.shape[1]/this.height]}}async estimateFaces(n){const t=ke.tidy(()=>(n instanceof ke.Tensor||(n=ke.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:e,scaleFactor:i}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(e.map(async r=>{const a=y8(r,i),[s,o,l]=await Promise.all([r.landmarks,a,r.probability].map(async f=>f.array())),u=r.anchor,[c,h]=i,d=s.map(f=>[(f[0]+u[0])*c,(f[1]+u[1])*h]),p={topLeft:o.slice(0,2),bottomRight:o.slice(2),landmarks:d,probability:l};return eL(r.box),r.landmarks.dispose(),r.probability.dispose(),a.dispose(),p}))}}async function b8(n){const t=await ke.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new nL(t,n);return e}Yo.load=b8;Yo.BlazeFaceModel=nL;Yo.disposeBox=eL});var np=be(tp=>{tp.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};tp.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var rL=be(mr=>{const w8=Wt();function S8(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}mr.scaleBoxCoordinates=S8;function ip(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}mr.getBoxSize=ip;function rp(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}mr.getBoxCenter=rp;function L8(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return w8.image.cropAndResize(t,a,[0],e)}mr.cutBoxFromImageAndResize=L8;function I8(n,t=1.5){const e=rp(n),i=ip(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.enlargeBox=I8;function A8(n){const t=rp(n),e=ip(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.squarifyBox=A8});var uL=be(In=>{In.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function aL(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}In.normalizeRadians=aL;function T8(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return aL(e)}In.computeRotation=T8;function N8(n){return n*180/Math.PI}In.radToDegrees=N8;function sL(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function la(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}In.dot=la;function oL(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}In.getColumnFrom2DArr=oL;function lL(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(la(n[r],oL(t,a)))}return e}function x8(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=sL(t[0],t[1]),s=lL(a,r),o=sL(-t[0],-t[1]);return lL(s,o)}In.buildRotationMatrix=x8;function C8(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-la(t[0],e),-la(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}In.invertTransformMatrix=C8;function R8(n,t){return[la(n,t[0]),la(n,t[1])]}In.rotatePoint=R8;function O8(n,t){return Math.sqrt((n[0]-t[0])**2+(n[1]-t[1])**2)}In.xyDistanceBetweenPoints=O8});var hL=be(cL=>{const Ni=Wt(),Mn=rL(),xi=np(),Ci=uL(),E8=468,D8=.25,k8=13,F8=[k8,xi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],W8=3,U8=2,B8=[W8,U8],ap=xi.MESH_ANNOTATIONS.leftEyeLower0,sp=[ap[0],ap[ap.length-1]],op=xi.MESH_ANNOTATIONS.rightEyeLower0,lp=[op[0],op[op.length-1]],z8=3,P8=4,_8=71,up=76;function Ko(n,t,e,i){for(let r=0;r<xi.MESH_TO_IRIS_INDICES_MAP.length;r++){const{key:a,indices:s}=xi.MESH_TO_IRIS_INDICES_MAP[r],o=xi.MESH_ANNOTATIONS[`${e}${a}`],l=i==null;if(l||i.includes(a))for(let u=0;u<s.length;u++){const c=s[u];n[o[u]]=[t[c][0],t[c][1],(t[c][2]+n[o[u]][2])/2]}}}class M8{constructor(n,t,e,i){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.irisModel=e,this.meshWidth=i.mesh.inputSize,this.meshHeight=i.mesh.inputSize,this.irisSize=i.iris.inputSize,this.irisEnlarge=i.iris.enlargeFactor}transformRawCoords(n,t,e,i){const r=Mn.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Ci.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Ci.rotatePoint(d,o),d[2]]),u=Ci.invertTransformMatrix(i),c=[...Mn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Ci.dot(c,u[0]),Ci.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[sp[0]][2],e=n[lp[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Mn.squarifyBox(Mn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Mn.getBoxSize(a);let o=Ni.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Ni.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a<up;a++){const s=n[a*3],o=n[a*3+1],l=n[a*3+2];r.push([(i?1-s/this.irisSize:s/this.irisSize)*e[0]+t.startPoint[0],o/this.irisSize*e[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(_8)}}getAdjustedIrisCoords(n,t,e){const i=n[xi.MESH_ANNOTATIONS[`${e}EyeUpper0`][z8]][2],r=n[xi.MESH_ANNOTATIONS[`${e}EyeLower0`][P8]][2],a=(i+r)/2;return t.map((s,o)=>{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Mn.scaleBoxCoordinates(o,r),u=Mn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Ni.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=E8;let[o,l]=F8;s===!1&&([o,l]=B8),a=Ci.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Mn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Ci.IDENTITY_MATRIX;a!==0&&(h=Ni.image.rotateWithOffset(n,a,0,c),d=Ci.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Mn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Ni.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:x,boxSize:C,crop:R}=this.getEyeBox(b,f,sp[0],sp[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,lp[0],lp[1]),F=this.irisModel.predict(Ni.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,up*3),{rawCoords:_,iris:K}=this.getEyeCoords(H,x,C,!0),j=P.slice(up*3),{rawCoords:q,iris:G}=this.getEyeCoords(j,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(Ko(b,_,"left"),Ko(b,q,"right")):Z<1?Ko(b,_,"left",["EyeUpper0","EyeLower0"]):Ko(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,K,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Ni.dispose(b);const S=Mn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const x=Ni.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:x.arraySync()};const C={coords:x,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t<n.length;t++){const e=n[t],i=this.regionsOfInterest[t];let r=0;if(i&&i.startPoint){const[a,s]=e.startPoint,[o,l]=e.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}r<D8&&(this.regionsOfInterest[t]=e)}this.regionsOfInterest=this.regionsOfInterest.slice(0,n.length)}clearRegionOfInterest(n){this.regionsOfInterest[n]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,n),...this.regionsOfInterest.slice(n+1)])}shouldUpdateRegionsOfInterest(){const n=this.regionsOfInterest.length,t=n===0;return this.maxFaces===1||t?t:n!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}cL.Pipeline=M8});var pL=be(dL=>{dL.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var fL=be(H8=>{xp(H8,{default:()=>V8});var V8=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var vL=be(os=>{const gr=Wt(),q8=iL(),mL=np(),G8=hL(),Y8=pL(),K8=fL().default;class gL{constructor(n,t,e,i){this.pipeline=new G8.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=gr.tidy(()=>(n instanceof gr.Tensor||(n=gr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);gr.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in mL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=mL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?gr.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function j8(n){const t=await Promise.all([q8.load(n),gr.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),gr.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new gL(t[0],t[1],t[2],n);return e}os.load=j8;os.MediaPipeFaceMesh=gL;os.uv_coords=Y8;os.triangulation=K8});var bL=be(jo=>{const an=Wt(),Ri={};let yL={age:0,gender:""},$o=0;async function $8(n,t){const e=an.browser.fromPixels(n),i=an.image.resizeBilinear(e,[t,t]),r=an.cast(an.expandDims(i,0),"float32");return r}async function X8(n){return Ri.age||(Ri.age=await an.loadGraphModel(n.face.age.modelPath)),Ri.age}async function J8(n){return Ri.gender||(Ri.gender=await an.loadGraphModel(n.face.gender.modelPath)),Ri.gender}async function Z8(n,t){if($o>t.face.age.skipFrames?$o=0:$o+=1,$o===0)return yL;let e;if(n instanceof an.Tensor){const r=an.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=an.mul(r,[255]),an.dispose(r)}else e=await $8(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await Ri.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,an.dispose(r)}if(t.face.gender.enabled){const r=await Ri.gender.predict(e),a=await r.data(),s=Math.trunc(Math.abs(1.9*100*(a[0]-.5)))/100;s>t.face.gender.minConfidence&&(i.gender=a[0]<=.5?"female":"male",i.confidence=s),an.dispose(r)}return an.dispose(e),yL=i,i}jo.predict=Z8;jo.loadAge=X8;jo.loadGender=J8});var LL=be(cp=>{const Ut=Wt(),Q8=["angry","discust","fear","happy","sad","surpise","neutral"],Xo={};let wL=[],hp=0;const SL=1.5;function e7(n,t){const e=Ut.tidy(()=>{const i=Ut.browser.fromPixels(n,1),r=Ut.image.resizeBilinear(i,[t,t]),a=Ut.cast(Ut.expandDims(r,0),"float32");return a});return e}async function t7(n){return Xo.emotion||(Xo.emotion=await Ut.loadGraphModel(n.face.emotion.modelPath)),Xo.emotion}async function n7(n,t){if(hp+=1,hp>=t.face.emotion.skipFrames)return hp=0,wL;const e=Ut.tidy(()=>{if(n instanceof Ut.Tensor){const r=Ut.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Ut.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Ut.mul(a,[.2989]),u=Ut.mul(s,[.587]),c=Ut.mul(o,[.114]),h=Ut.addN([l,u,c]);return h}return s}return e7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await Xo.emotion.predict(e),a=await r.data();for(let s=0;s<a.length;s++)SL*a[s]>t.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*SL*a[s])/100),emotion:Q8[s]});i.sort((s,o)=>o.score-s.score),Ut.dispose(r)}return Ut.dispose(e),wL=i,i}cp.predict=n7;cp.load=t7});var TL=be(IL=>{const AL=Wt();class i7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;AL.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return AL.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}IL.BaseModel=i7});var dp=be(NL=>{const xL=Wt(),r7=TL();class a7 extends r7.BaseModel{preprocessInput(n){return xL.tidy(()=>xL.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}NL.MobileNet=a7});var RL=be(CL=>{function pp(n){return Math.floor(n/2)}class s7{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(pp(n),n);)this.exchange(n,pp(n)),n=pp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(n,t))break;this.exchange(n,t),n=t}}getValueAt(n){return this.getElementValue(this.priorityQueue[n])}less(n,t){return this.getValueAt(n)<this.getValueAt(t)}exchange(n,t){const e=this.priorityQueue[n];this.priorityQueue[n]=this.priorityQueue[t],this.priorityQueue[t]=e}}CL.MaxHeap=s7});var EL=be(OL=>{const o7=RL();function l7(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;h<c;++h){const d=Math.max(i-r,0),p=Math.min(i+r+1,o);for(let f=d;f<p;++f)if(a.get(h,f,n)>t){l=!1;break}if(!l)break}return l}function u7(n,t,e){const[i,r,a]=e.shape,s=new o7.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o<i;++o)for(let l=0;l<r;++l)for(let u=0;u<a;++u){const c=e.get(o,l,u);if(c<n)continue;l7(u,c,o,l,t,e)&&s.enqueue({score:c,part:{heatmapY:o,heatmapX:l,id:u}})}return s}OL.buildPartWithScoreQueue=u7});var ls=be(An=>{An.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];An.NUM_KEYPOINTS=An.partNames.length;An.partIds=An.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const c7=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];An.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];An.connectedPartIndices=c7.map(([n,t])=>[An.partIds[n],An.partIds[t]]);An.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var mp=be(Oi=>{const h7=ls();function DL(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+h7.NUM_KEYPOINTS)}}Oi.getOffsetPoint=DL;function d7(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=DL(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Oi.getImageCoords=d7;function p7(n,t){const e=new Array(t);for(let i=0;i<t;i++)e[i]=n;return e}Oi.fillArray=p7;function fp(n,t,e){return n<t?t:n>e?e:n}Oi.clamp=fp;function f7(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Oi.squaredDistance=f7;function m7(n,t){return{x:n.x+t.x,y:n.y+t.y}}Oi.addVectors=m7;function g7(n,t,e){return{y:fp(n.y,t,e),x:fp(n.x,t,e)}}Oi.clampVector=g7});var BL=be(kL=>{const us=ls(),ua=mp(),FL=us.poseChain.map(([n,t])=>[us.partIds[n],us.partIds[t]]),gp=FL.map(([,n])=>n),WL=FL.map(([n])=>n);function v7(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function vp(n,t,e,i){return{y:ua.clamp(Math.round(n.y/t),0,e-1),x:ua.clamp(Math.round(n.x/t),0,i-1)}}function UL(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=vp(t.position,a,l,u),h=v7(n,c,s),d=ua.addVectors(t.position,h);let p=d;for(let g=0;g<o;g++){const v=vp(p,a,l,u),b=ua.getOffsetPoint(v.y,v.x,e,r);p=ua.addVectors({x:v.x*a,y:v.y*a},{x:b.x,y:b.y})}const f=vp(p,a,l,u),m=i.get(f.y,f.x,e);return{position:p,part:us.partNames[e],score:m}}function y7(n,t,e,i,r,a){const s=t.shape[2],o=gp.length,l=new Array(s),{part:u,score:c}=n,h=ua.getImageCoords(u,i,e);l[u.id]={score:c,part:us.partNames[u.id],position:h};for(let d=o-1;d>=0;--d){const p=gp[d],f=WL[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,a))}for(let d=0;d<o;++d){const p=WL[d],f=gp[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,r))}return l}kL.decodePose=y7});var yp=be(zL=>{const b7=EL(),w7=BL(),PL=mp();function _L(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return PL.squaredDistance(i,e,s.y,s.x)<=t})}function S7(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(_L(n,t,a,o)||(r+=s),r),0);return i/e.length}const L7=1;function I7(n,t,e,i,r,a,s=.5,o=20){const l=[],u=b7.buildPartWithScoreQueue(s,L7,n),c=o*o;for(;l.length<a&&!u.empty();){const h=u.dequeue(),d=PL.getImageCoords(h.part,r,t);if(_L(l,c,d,h.part.id))continue;const p=w7.decodePose(h,n,t,r,e,i),f=S7(l,c,p);l.push({keypoints:p,score:f})}return l}zL.decodeMultiplePoses=I7});var wp=be(hn=>{const ca=Wt(),A7=ls();function T7(n,t,e){return n<e||t<e}function N7(n,t){return A7.connectedPartIndices.reduce((e,[i,r])=>(T7(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}hn.getAdjacentKeyPoints=N7;const{NEGATIVE_INFINITY:ML,POSITIVE_INFINITY:HL}=Number;function VL(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:ML,maxY:ML,minX:HL,minY:HL})}hn.getBoundingBox=VL;function x7(n){const{minX:t,minY:e,maxX:i,maxY:r}=VL(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}hn.getBoundingBoxPoints=x7;async function C7(n){return Promise.all(n.map(t=>t.buffer()))}hn.toTensorBuffers3D=C7;function qL(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}hn.scalePose=qL;function GL(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>qL(a,t,e,i,r))}hn.scalePoses=GL;function YL(n){return n instanceof ca.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}hn.getInputTensorDimensions=YL;function bp(n){return n instanceof ca.Tensor?n:ca.browser.fromPixels(n)}hn.toInputTensor=bp;function R7(n,t,e){return ca.tidy(()=>{const i=bp(n);return i.resizeBilinear([t,e])})}hn.toResizedInputTensor=R7;function O7(n,[t,e]){const[i,r]=YL(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s<a?(o=0,l=0,u=Math.round(.5*(a*i-r)),c=Math.round(.5*(a*i-r))):(o=Math.round(.5*(1/a*r-i)),l=Math.round(.5*(1/a*r-i)),u=0,c=0);const h=ca.tidy(()=>{let d=bp(n);return d=ca.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}hn.padAndResizeTo=O7;function E7(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=GL(n,s,o,-a.top,-a.left);return l}hn.scaleAndFlipPoses=E7});var jL=be(Sp=>{const D7=Wt(),k7=dp(),F7=yp(),Jo=wp();class KL{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=Jo.getInputTensorDimensions(n),{resized:a,padding:s}=Jo.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await Jo.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await F7.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=Jo.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}Sp.PoseNet=KL;async function W7(n){const t=await D7.loadGraphModel(n.modelPath),e=new k7.MobileNet(t,n.outputStride);return new KL(e)}async function U7(n){return W7(n)}Sp.load=U7});var XL=be(Jt=>{const B7=dp(),$L=jL(),z7=yp(),Zo=ls(),cs=wp();Jt.load=$L.load;Jt.PoseNet=$L.PoseNet;Jt.MobileNet=B7.MobileNet;Jt.decodeMultiplePoses=z7.decodeMultiplePoses;Jt.partChannels=Zo.partChannels;Jt.partIds=Zo.partIds;Jt.partNames=Zo.partNames;Jt.poseChain=Zo.poseChain;Jt.getAdjacentKeyPoints=cs.getAdjacentKeyPoints;Jt.getBoundingBox=cs.getBoundingBox;Jt.getBoundingBoxPoints=cs.getBoundingBoxPoints;Jt.scaleAndFlipPoses=cs.scaleAndFlipPoses;Jt.scalePose=cs.scalePose});var Ap=be(Ei=>{const P7=Wt();function Lp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Ei.getBoxSize=Lp;function Ip(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Ei.getBoxCenter=Ip;function _7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return P7.image.cropAndResize(t,a,[0],e)}Ei.cutBoxFromImageAndResize=_7;function M7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}Ei.scaleBoxCoordinates=M7;function H7(n,t=1.5){const e=Ip(n),i=Lp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.enlargeBox=H7;function V7(n){const t=Ip(n),e=Lp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.squarifyBox=V7;function q7(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}Ei.shiftBox=q7});var ZL=be(JL=>{const qe=Wt(),G7=Ap();class Y7{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=qe.tensor2d(this.anchors),this.inputSizeTensor=qe.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=qe.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return qe.tidy(()=>{const t=qe.slice(n,[0,0],[-1,2]),e=qe.slice(n,[0,2],[-1,2]),i=qe.add(qe.div(t,this.inputSizeTensor),this.anchorsTensor),r=qe.div(e,this.doubleInputSizeTensor),a=qe.mul(qe.sub(i,r),this.inputSizeTensor),s=qe.mul(qe.add(i,r),this.inputSizeTensor);return qe.concat2d([a,s],1)})}normalizeLandmarks(n,t){return qe.tidy(()=>{const e=qe.add(qe.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return qe.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=qe.tidy(()=>qe.mul(qe.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=qe.tidy(()=>qe.sigmoid(qe.slice(i,[0,0],[-1,1])).squeeze()),a=qe.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await qe.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=qe.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=qe.slice(s,[p,0],[1,-1]),m=qe.slice(i,[p,5],[1,14]),g=qe.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=qe.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(G7.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}JL.HandDetector=Y7});var eI=be(QL=>{QL.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var aI=be(Di=>{function tI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Di.normalizeRadians=tI;function K7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return tI(e)}Di.computeRotation=K7;const nI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function ha(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Di.dot=ha;function iI(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Di.getColumnFrom2DArr=iI;function rI(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(ha(n[r],iI(t,a)))}return e}function j7(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=nI(t[0],t[1]),s=rI(a,r),o=nI(-t[0],-t[1]);return rI(s,o)}Di.buildRotationMatrix=j7;function $7(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-ha(t[0],e),-ha(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Di.invertTransformMatrix=$7;function X7(n,t){return[ha(n,t[0]),ha(n,t[1])]}Di.rotatePoint=X7});var uI=be(sI=>{const oI=Wt(),Hn=Ap(),ki=aI(),J7=.8,Z7=[0,-.4],Q7=[0,-.1],eH=1.65,lI=[0,5,9,13,17,1,2],tH=0,nH=2;class iH{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return ki.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(i,Z7)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(t,Q7)),eH),i=[];for(let r=0;r<lI.length;r++)i.push(n[lI[r]].slice(0,2));return e.palmLandmarks=i,e}transformRawCoords(n,t,e,i){const r=Hn.getBoxSize(t),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=ki.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=ki.rotatePoint(d,o);return[...p,d[2]]}),u=ki.invertTransformMatrix(i),c=[...Hn.getBoxCenter(t),1],h=[ki.dot(c,u[0]),ki.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=ki.computeRotation(a.palmLandmarks[tH],a.palmLandmarks[nH]),o=Hn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=oI.image.rotateWithOffset(n,s,0,l),c=ki.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=Hn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),v<t.minConfidence)return g.dispose(),this.regionsOfInterest[r]=[],i;const b=oI.reshape(g,[-1,3]),w=await b.array();g.dispose(),b.dispose();const S=this.transformRawCoords(w,h,s,c),L=this.getBoxForHandLandmarks(S);this.updateRegionsOfInterest(L,!1,r);const x={landmarks:S,confidence:v,box:{topLeft:L.startPoint,bottomRight:L.endPoint}};i.push(x)}return i}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>J7?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}sI.HandPipeline=iH});var dI=be(Tp=>{const vr=Wt(),rH=ZL(),cI=eI(),aH=uI();class hI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=vr.tidy(()=>(n instanceof vr.Tensor||(n=vr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(cI.MESH_ANNOTATIONS))s[o]=cI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}Tp.HandPose=hI;async function sH(n){if(vr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return vr.util.fetch(n).then(t=>t.json())}async function oH(n){const[t,e,i]=await Promise.all([sH(n.detector.anchors),vr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),vr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new rH.HandDetector(e,t,n),a=new aH.HandPipeline(r,i,n),s=new hI(a);return s}Tp.load=oH});var pI=be(lH=>{xp(lH,{default:()=>uH});var uH={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,minConfidence:.8,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:10,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:10,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var mI=be((oV,fI)=>{fI.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});var LI=be(Tn=>{const Nn=Wt(),gI=vL(),Qo=bL(),vI=LL(),yI=XL(),bI=dI(),wI=pI().default,cH=mI();let Pe;const Et={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},Vn=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),el=(...n)=>{Pe.console&&console.log(...n)};function SI(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=SI(a,s):e[r]=s}),e),{})}function hH(n){if(!n)return"input is not defined";const t=n.naturalWidth||n.videoWidth||n.width||n.shape&&n.shape[1]>0;if(!t||t===0)return"input is empty";if(n.readyState&&n.readyState<=2)return"input is not ready";try{Nn.getBackend()}catch{return"backend not loaded"}return null}async function dH(n,t){Pe=SI(wI,t);const e=hH(n);return e?(el(e,n),{error:e}):new Promise(async i=>{const r=Object.values(Et).filter(c=>c).length;r===0&&el("Human library starting"),Nn.getBackend()!==Pe.backend&&(el("Human library setting backend:",Pe.backend),await Nn.setBackend(Pe.backend),await Nn.ready()),Pe.face.enabled&&!Et.facemesh&&(Et.facemesh=await gI.load(Pe.face)),Pe.body.enabled&&!Et.posenet&&(Et.posenet=await yI.load(Pe.body)),Pe.hand.enabled&&!Et.handpose&&(Et.handpose=await bI.load(Pe.hand)),Pe.face.enabled&&Pe.face.age.enabled&&!Et.age&&(Et.age=await Qo.loadAge(Pe)),Pe.face.enabled&&Pe.face.gender.enabled&&!Et.gender&&(Et.gender=await Qo.loadGender(Pe)),Pe.face.enabled&&Pe.face.emotion.enabled&&!Et.emotion&&(Et.emotion=await vI.load(Pe));const a={};let s;s=Vn(),Nn.engine().startScope();const o=Pe.body.enabled?await Et.posenet.estimatePoses(n,Pe.body):[];Nn.engine().endScope(),a.body=Math.trunc(Vn()-s),s=Vn(),Nn.engine().startScope();const l=Pe.hand.enabled?await Et.handpose.estimateHands(n,Pe.hand):[];Nn.engine().endScope(),a.hand=Math.trunc(Vn()-s);const u=[];if(Pe.face.enabled){s=Vn(),Nn.engine().startScope();const c=await Et.facemesh.estimateFaces(n,Pe.face);a.face=Math.trunc(Vn()-s);for(const h of c){if(!h.image||h.image.isDisposedInternal){el("face object is disposed:",h.image);continue}s=Vn();const d=Pe.face.age.enabled||Pe.face.gender.enabled?await Qo.predict(h.image,Pe):{};a.agegender=Math.trunc(Vn()-s),s=Vn();const p=Pe.face.emotion.enabled?await vI.predict(h.image,Pe):{};a.emotion=Math.trunc(Vn()-s),h.image.dispose();const f=h.annotations.leftEyeIris&&h.annotations.rightEyeIris?Math.max(h.annotations.leftEyeIris[3][0]-h.annotations.leftEyeIris[1][0],h.annotations.rightEyeIris[3][0]-h.annotations.rightEyeIris[1][0]):0;u.push({confidence:h.confidence,box:h.box,mesh:h.mesh,annotations:h.annotations,age:d.age,gender:d.gender,agConfidence:d.confidence,emotion:p,iris:f!==0?Math.trunc(100*11.7/f)/100:0})}Nn.engine().endScope()}a.total=Object.values(a).reduce((c,h)=>c+h),i({face:u,body:o,hand:l,performance:a})})}Tn.detect=dH;Tn.defaults=wI;Tn.config=Pe;Tn.models=Et;Tn.facemesh=gI;Tn.ssrnet=Qo;Tn.posenet=yI;Tn.handpose=bI;Tn.tf=Nn;Tn.version=cH.version});export default LI();
 /*! *****************************************************************************
 Copyright (c) Microsoft Corporation. All rights reserved.
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use
diff --git a/dist/human.esm.js.map b/dist/human.esm.js.map
index 3e61fbab..77fb9fe9 100644
--- a/dist/human.esm.js.map
+++ b/dist/human.esm.js.map
@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["empty:/home/vlado/dev/human/node_modules/node-fetch/browser.js", "empty:util", "empty:crypto", "../node_modules/@tensorflow/tfjs-core/src/backends/backend.ts", "../node_modules/@tensorflow/tfjs-core/src/environment.ts", "../node_modules/@tensorflow/tfjs-core/src/global_util.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_names.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/util.ts", "../node_modules/@tensorflow/tfjs-core/src/profiler.ts", "../node_modules/@tensorflow/tfjs-core/src/tape.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_format.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/types.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util.ts", "../node_modules/@tensorflow/tfjs-core/src/engine.ts", "../node_modules/@tensorflow/tfjs-core/src/device_util.ts", "../node_modules/@tensorflow/tfjs-core/src/flags.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util_env.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/operation.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/complex.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/io/types.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/io/router_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/io/indexed_db.ts", "../node_modules/@tensorflow/tfjs-core/src/io/local_storage.ts", "../node_modules/@tensorflow/tfjs-core/src/io/model_management.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_browser.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_node.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/buffer.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clone.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/print.ts", "../node_modules/@tensorflow/tfjs-core/src/base_side_effects.ts", "../node_modules/@tensorflow/tfjs-core/src/io/browser_files.ts", "../node_modules/@tensorflow/tfjs-core/src/io/progress.ts", "../node_modules/@tensorflow/tfjs-core/src/io/weights_loader.ts", "../node_modules/@tensorflow/tfjs-core/src/io/http.ts", "../node_modules/@tensorflow/tfjs-core/src/io/passthrough.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/confusion_matrix.ts", "../node_modules/@tensorflow/tfjs-core/src/math.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/browser.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice_util.ts", "../node_modules/@tensorflow/tfjs-core/src/serialization.ts", "../node_modules/@tensorflow/tfjs-core/src/test_util.ts", "../node_modules/@tensorflow/tfjs-core/src/version.ts", "../node_modules/@tensorflow/tfjs-core/src/globals.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add_n.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/axis_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/basic_lstm_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/diag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/eye.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fill.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reduce_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/segment_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/imag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linspace.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_with_argmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moments.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multi_rnn_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multinomial.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/real.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/outer_product.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rand.ts", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-core/src/ops/rand_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_gamma.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_uniform.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/range.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/setdiff1d_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor5d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor6d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/truncated_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/variable.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/where_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/boolean_mask.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/compare.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/binary_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moving_average.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/in_top_k.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hamming_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hann_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/frame.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/stft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/crop_and_resize.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/flip_left_right.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/rotate_with_offset.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/nonmax_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/array_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/non_max_suppression_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/band_part.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/gram_schmidt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/qr.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/loss_ops_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/compute_weighted_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/absolute_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/cosine_distance.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/hinge_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/huber_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/log_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/mean_squared_error.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/sigmoid_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/softmax_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ops.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adadelta_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adagrad_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adam_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adamax_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/sgd_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/momentum_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/rmsprop_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer_constructors.ts", "../node_modules/@tensorflow/tfjs-core/src/train.ts", "../node_modules/@tensorflow/tfjs-core/src/browser_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rotate_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/array_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf_util.ts", "../node_modules/@tensorflow/tfjs-core/src/log.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/complex_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/backend_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/split_shared.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/tile_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/topk_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/kernel_impls.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Abs_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Add_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AddN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchMatMul_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchToSpaceND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BroadcastTo_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cast_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Ceil_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ClipByValue_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Concat_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2DBackpropInput_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cumsum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/DepthwiseConv2dNative_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Dilation2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Div_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Elu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Erf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Exp_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Expm1_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Floor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FloorDiv_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FusedBatchNorm_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GatherV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GreaterEqual_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Identity_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsFinite_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsInf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsNan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log1p_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LogSoftmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LRN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/min_max_grad_util.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Max_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Maximum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Min_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Minimum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Mod_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Multiply_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Negate_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OneHot_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OnesLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/PadV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Pow_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Prelu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reciprocal_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu6_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reshape_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeBilinear_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeNearestNeighbor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reverse_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Round_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Rsqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SelectV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Selu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sigmoid_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sign_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Slice_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softplus_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SpaceToBatchND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SplitV_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Square_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SquaredDifference_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Step_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sub_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tile_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Transpose_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Unpack_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/UnsortedSegmentSum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ZerosLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/register_all_gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_type.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as3d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as4d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as5d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2D_deprecated.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/flatten.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape_as.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_bool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_float.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_int.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/errors.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/generic_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/math_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/tfjs_backend.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/initializer_config.ts", "../node_modules/@tensorflow/tfjs-layers/src/initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/state.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/types_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/variable_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/variables.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/topology.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/input_layer.ts", "../node_modules/@tensorflow/tfjs-layers/src/logs.ts", "../node_modules/@tensorflow/tfjs-layers/src/base_callbacks.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/serialization.ts", "../node_modules/@tensorflow/tfjs-layers/src/losses.ts", "../node_modules/@tensorflow/tfjs-layers/src/metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/optimizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/user_defined_metadata.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/layer_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/serialization_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/version.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/executor.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/container.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_dataset.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_tensors.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training.ts", "../node_modules/@tensorflow/tfjs-layers/src/models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports.ts", "../node_modules/@tensorflow/tfjs-layers/src/activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/advanced_activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/conv_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_depthwise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/core.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/embeddings.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/merge.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/noise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/normalization.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/padding.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/pooling.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/wrappers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_layers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/callbacks.ts", "../node_modules/@tensorflow/tfjs-converter/src/data/compiled_api.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/register.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/arithmetic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/basic_math.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/control.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/convolution.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/creation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/dynamic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/evaluation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/graph.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/image.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/logical.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/matrices.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/normalization.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/reduction.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/slice_join.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/spectral.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/transformation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_mapper.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/node_value_impl.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/arithmetic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/basic_math_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_array.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_list.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/control_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/convolution_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/creation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/dynamic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/evaluation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/image_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/logical_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/matrices_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/normalization_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/reduction_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/slice_join_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/spectral_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/transformation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/execution_context.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/model_analysis.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_model.ts", "../node_modules/@tensorflow/tfjs-converter/src/version.ts", "empty:/home/vlado/dev/human/node_modules/string_decoder/lib/string_decoder.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-data/src/util/deep_map.ts", "../node_modules/@tensorflow/tfjs-data/src/util/deep_clone.ts", "../node_modules/@tensorflow/tfjs-data/src/util/ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/util/growing_ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/lazy_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/text_line_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/csv_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/microphone_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/webcam_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/datasource.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/string_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/byte_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/file_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/url_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/util/source_util.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/file_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/url_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/readers.ts", "../node_modules/@tensorflow/tfjs-data/src/version.ts", "../node_modules/seedrandom/lib/alea.js", "../node_modules/seedrandom/lib/xor128.js", "../node_modules/seedrandom/lib/xorwow.js", "../node_modules/seedrandom/lib/xorshift7.js", "../node_modules/seedrandom/lib/xor4096.js", "../node_modules/seedrandom/lib/tychei.js", "../node_modules/seedrandom/seedrandom.js", "../node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-backend-cpu/src/cpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/backend_cpu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Abs.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/binary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Complex.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Real.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cast.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/kernel_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Add.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Ceil.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Exp.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Expm1.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Floor.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Multiply.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Rsqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Slice.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sub.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/shared.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/pool_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Clip.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Imag.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Concat.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2D.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropFilter.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropInput.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Elu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Erf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/fft_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IFFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsFinite.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsInf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsNaN.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log1p.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/LogicalNot.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NotEqual.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/PadV2.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reciprocal.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Round.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Selu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sigmoid.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sign.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Softplus.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SpaceToBatchND.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Step.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/canvas_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tex_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flags_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Abs.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/binary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Add.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/unary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Ceil.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Exp.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Expm1.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Floor.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Log.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Max_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Multiply.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Rsqrt.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Slice.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Sub.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Transpose_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Unique_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/shared.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/packing_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/glsl_version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/avg_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_complex_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/complex_abs_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_packed_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/crop_and_resize_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/cumsum_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/depth_to_space_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/diag_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fft_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fill_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_nd_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_context.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_math.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/im2col_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_grad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/max_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/mulmat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/multinomial_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/onehot_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pool_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reduce_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reshape_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/scatter_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/segment_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/select_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/strided_slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/texture_manager.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tile_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unpack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/backend_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/kernel_funcs_utils.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Atan2.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flip_left_right_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reduce.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV3.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/rotate_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs/src/version.ts", "../node_modules/@tensorflow/tfjs/src/index.ts", "../src/facemesh/blazeface.js", "../src/facemesh/keypoints.js", "../src/facemesh/box.js", "../src/facemesh/util.js", "../src/facemesh/pipeline.js", "../src/facemesh/uvcoords.js", "../src/facemesh/triangulation.js", "../src/facemesh/facemesh.js", "../src/ssrnet/ssrnet.js", "../src/emotion/emotion.js", "../src/posenet/modelBase.js", "../src/posenet/modelMobileNet.js", "../src/posenet/heapSort.js", "../src/posenet/buildParts.js", "../src/posenet/keypoints.js", "../src/posenet/vectors.js", "../src/posenet/decodePose.js", "../src/posenet/decodeMultiple.js", "../src/posenet/util.js", "../src/posenet/modelPoseNet.js", "../src/posenet/posenet.js", "../src/handpose/box.js", "../src/handpose/handdetector.js", "../src/handpose/keypoints.js", "../src/handpose/util.js", "../src/handpose/pipeline.js", "../src/handpose/handpose.js", "../src/config.js", "../src/index.js"],
-  "sourcesContent": ["", "", "", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n  kernelMs: number|{error: string};\n  getExtraProfileInfo?(): string;  // a field for additional timing information\n                                   // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n  disposeData(dataId: DataId): void;\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void;\n  memory(): {unreliable: boolean;};  // Backend-specific information.\n  /** Returns number of data ids currently in the storage. */\n  numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage<T> {\n  private data = new WeakMap<DataId, T>();\n  private dataIdsCount = 0;\n\n  constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n  get(dataId: DataId) {\n    if (!this.data.has(dataId)) {\n      this.dataMover.moveData(this.backend, dataId);\n    }\n    return this.data.get(dataId);\n  }\n\n  set(dataId: DataId, value: T): void {\n    this.dataIdsCount++;\n    this.data.set(dataId, value);\n  }\n\n  has(dataId: DataId): boolean {\n    return this.data.has(dataId);\n  }\n\n  delete(dataId: DataId): boolean {\n    this.dataIdsCount--;\n    return this.data.delete(dataId);\n  }\n\n  numDataIds(): number {\n    return this.dataIdsCount;\n  }\n}\n\nexport interface DataMover {\n  /**\n   * To be called by backends whenever they see a dataId that they don't own.\n   * Upon calling this method, the mover will fetch the tensor from another\n   * backend and register it with the current active backend.\n   */\n  moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo>;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo> {\n    return notYetImplemented('time');\n  }\n  read(dataId: object): Promise<BackendValues> {\n    return notYetImplemented('read');\n  }\n  readSync(dataId: object): BackendValues {\n    return notYetImplemented('readSync');\n  }\n  numDataIds(): number {\n    return notYetImplemented('numDataIds');\n  }\n  disposeData(dataId: object): void {\n    return notYetImplemented('disposeData');\n  }\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    return notYetImplemented('write');\n  }\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    return notYetImplemented('move');\n  }\n  memory(): {unreliable: boolean; reasons?: string[]} {\n    return notYetImplemented('memory');\n  }\n  /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n  floatPrecision(): 16|32 {\n    return notYetImplemented('floatPrecision');\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    return notYetImplemented('batchMatMul');\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          FusedBatchMatMulConfig): Tensor3D {\n    return notYetImplemented('fusedBatchMatMul');\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    return notYetImplemented('slice');\n  }\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    return notYetImplemented('stridedSlice');\n  }\n  unstack(x: Tensor, axis: number): Tensor[] {\n    return notYetImplemented('unstack');\n  }\n  reverse<T extends Tensor>(a: T, axis: number[]): T {\n    return notYetImplemented('reverse');\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    return notYetImplemented('concat');\n  }\n\n  neg<T extends Tensor>(a: T): T {\n    return notYetImplemented('neg');\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('add');\n  }\n  addN<T extends Tensor>(tensors: T[]): T {\n    return notYetImplemented('addN');\n  }\n  subtract(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('subtract');\n  }\n  multiply(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('multiply');\n  }\n  realDivide(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('realDivide');\n  }\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('floorDiv');\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('sum');\n  }\n  prod(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('prod');\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    return notYetImplemented('unsortedSegmentSum');\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMin');\n  }\n  argMax(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMax');\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('equal');\n  }\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('notEqual');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('less');\n  }\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('lessEqual');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greater');\n  }\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greaterEqual');\n  }\n\n  logicalNot<T extends Tensor>(a: T): T {\n    return notYetImplemented('logicalNot');\n  }\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalAnd');\n  }\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalOr');\n  }\n\n  where(condition: Tensor): Tensor2D {\n    return notYetImplemented('where');\n  }\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('select');\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    return notYetImplemented('topk');\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('min');\n  }\n  minimum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('minimum');\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('mod');\n  }\n\n  max(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('max');\n  }\n  maximum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('maximum');\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('all');\n  }\n  any(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('any');\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('squaredDifference');\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    return notYetImplemented('ceil');\n  }\n  floor<T extends Tensor>(x: T): T {\n    return notYetImplemented('floor');\n  }\n  round<T extends Tensor>(x: T): T {\n    return notYetImplemented('round');\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    return notYetImplemented('sign');\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    return notYetImplemented('isNaN');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    return notYetImplemented('isInf');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    return notYetImplemented('isFinite');\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    return notYetImplemented('pow');\n  }\n  exp<T extends Tensor>(x: T): T {\n    return notYetImplemented('exp');\n  }\n  expm1<T extends Tensor>(x: T): T {\n    return notYetImplemented('expm1');\n  }\n  softmax<T extends Tensor>(x: T, dim: number): T {\n    return notYetImplemented('softmax');\n  }\n  log<T extends Tensor>(x: T): T {\n    return notYetImplemented('log');\n  }\n  log1p<T extends Tensor>(x: T): T {\n    return notYetImplemented('log1p');\n  }\n  sqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('sqrt');\n  }\n  rsqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('rsqrt');\n  }\n  square<T extends Tensor>(x: T): T {\n    return notYetImplemented('square');\n  }\n  reciprocal<T extends Tensor>(x: T): T {\n    return notYetImplemented('reciprocal');\n  }\n  relu<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu');\n  }\n  relu6<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu6');\n  }\n  prelu<T extends Tensor>(x: T, a: T): T {\n    return notYetImplemented('prelu');\n  }\n  elu<T extends Tensor>(x: T): T {\n    return notYetImplemented('elu');\n  }\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    return notYetImplemented('eluDer');\n  }\n  selu<T extends Tensor>(x: T): T {\n    return notYetImplemented('selu');\n  }\n  int<T extends Tensor>(x: T): T {\n    return notYetImplemented('int');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    return notYetImplemented('clip');\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    return notYetImplemented('abs');\n  }\n  complexAbs<T extends Tensor>(x: T): T {\n    return notYetImplemented('complexAbs');\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    return notYetImplemented('sigmoid');\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    return notYetImplemented('softplus');\n  }\n\n  sin<T extends Tensor>(x: T): T {\n    return notYetImplemented('sin');\n  }\n  cos<T extends Tensor>(x: T): T {\n    return notYetImplemented('cos');\n  }\n  tan<T extends Tensor>(x: T): T {\n    return notYetImplemented('tan');\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    return notYetImplemented('asin');\n  }\n  acos<T extends Tensor>(x: T): T {\n    return notYetImplemented('acos');\n  }\n  atan<T extends Tensor>(x: T): T {\n    return notYetImplemented('atan');\n  }\n  atan2<T extends Tensor>(a: T, b: T): T {\n    return notYetImplemented('atan2');\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('sinh');\n  }\n  cosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('cosh');\n  }\n  tanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('tanh');\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('asinh');\n  }\n  acosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('acosh');\n  }\n  atanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('atanh');\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    return notYetImplemented('erf');\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    return notYetImplemented('step');\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedConv2d');\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2d');\n  }\n  conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('conv2dDerInput');\n  }\n  conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2dDerFilter');\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedDepthwiseConv2D');\n  }\n\n  depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2D');\n  }\n  depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerInput');\n  }\n  depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerFilter');\n  }\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3d');\n  }\n  conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n      Tensor5D {\n    return notYetImplemented('conv3dDerInput');\n  }\n  conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3dDerFilter');\n  }\n  maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('maxPool');\n  }\n  maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('maxPoolBackprop');\n  }\n  avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPool');\n  }\n  avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPoolBackprop');\n  }\n  avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3d');\n  }\n  avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3dBackprop');\n  }\n  maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3d');\n  }\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3dBackprop');\n  }\n\n  reshape<T extends Tensor, R extends Rank>(x: T, shape: ShapeMap[R]):\n      Tensor<R> {\n    return notYetImplemented('reshape');\n  }\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return notYetImplemented('cast');\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    return notYetImplemented('tile');\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    return notYetImplemented('pad');\n  }\n\n  transpose<T extends Tensor>(x: T, perm: number[]): T {\n    return notYetImplemented('transpose');\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    return notYetImplemented('gather');\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    return notYetImplemented('gatherND');\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    return notYetImplemented('scatterND');\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    return notYetImplemented('batchToSpaceND');\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: number[][]): T {\n    return notYetImplemented('spaceToBatchND');\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeBilinear');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    return notYetImplemented('resizeBilinearBackprop');\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHEight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighbor');\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighborBackprop');\n  }\n\n  batchNorm(\n      x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n      offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n      varianceEpsilon?: number): Tensor4D {\n    return notYetImplemented('batchNorm');\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    return notYetImplemented('localResponseNormalization4D');\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n      bias: number, alpha: number, beta: number): Tensor4D {\n    return notYetImplemented('LRNGrad');\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    return notYetImplemented('multinomial');\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    return notYetImplemented('oneHot');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    return notYetImplemented('cumsum');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold?: number): Tensor1D {\n    return notYetImplemented('nonMaxSuppression');\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('fft');\n  }\n  ifft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('ifft');\n  }\n  complex<T extends Tensor>(real: T, imag: T): T {\n    return notYetImplemented('complex');\n  }\n  real<T extends Tensor>(input: T): T {\n    return notYetImplemented('real');\n  }\n  imag<T extends Tensor>(input: T): T {\n    return notYetImplemented('imag');\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    return notYetImplemented('cropAndResize');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n    return notYetImplemented('depthToSpace');\n  }\n\n  // Aligns with the \"SplitV\" kernel in TensorFlow.\n  split<T extends Tensor>(value: T, sizeSplits: number[], axis: number): T[] {\n    return notYetImplemented('split');\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    return notYetImplemented('sparseToDense');\n  }\n\n  diag(x: Tensor): Tensor {\n    return notYetImplemented('diag');\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    return notYetImplemented('fill');\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('onesLike');\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('zerosLike');\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return notYetImplemented('linspace');\n  }\n\n  dispose(): void {\n    return notYetImplemented('dispose');\n  }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n  throw new Error(\n      `'${kernelName}' not yet implemented or not found in the registry. ` +\n      `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise<FlagValue>);\nexport type Flags = {\n  [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n  evaluationFn: FlagEvaluationFn;\n  setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n  private flags: Flags = {};\n  private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n  private urlFlags: Flags = {};\n\n  platformName: string;\n  platform: Platform;\n\n  // tslint:disable-next-line: no-any\n  constructor(public global: any) {\n    this.populateURLFlags();\n  }\n\n  setPlatform(platformName: string, platform: Platform) {\n    if (this.platform != null) {\n      console.warn(\n          `Platform ${this.platformName} has already been set. ` +\n          `Overwriting the platform with ${platform}.`);\n    }\n    this.platformName = platformName;\n    this.platform = platform;\n  }\n\n  registerFlag(\n      flagName: string, evaluationFn: FlagEvaluationFn,\n      setHook?: (value: FlagValue) => void) {\n    this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n    // Override the flag value from the URL. This has to happen here because the\n    // environment is initialized before flags get registered.\n    if (this.urlFlags[flagName] != null) {\n      const flagValue = this.urlFlags[flagName];\n      console.warn(\n          `Setting feature override from URL ${flagName}: ${flagValue}.`);\n      this.set(flagName, flagValue);\n    }\n  }\n\n  async getAsync(flagName: string): Promise<FlagValue> {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    this.flags[flagName] = await this.evaluateFlag(flagName);\n    return this.flags[flagName];\n  }\n\n  get(flagName: string): FlagValue {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    const flagValue = this.evaluateFlag(flagName);\n    if (flagValue instanceof Promise) {\n      throw new Error(\n          `Flag ${flagName} cannot be synchronously evaluated. ` +\n          `Please use getAsync() instead.`);\n    }\n\n    this.flags[flagName] = flagValue;\n\n    return this.flags[flagName];\n  }\n\n  getNumber(flagName: string): number {\n    return this.get(flagName) as number;\n  }\n\n  getBool(flagName: string): boolean {\n    return this.get(flagName) as boolean;\n  }\n\n  getFlags(): Flags {\n    return this.flags;\n  }\n  // For backwards compatibility.\n  get features(): Flags {\n    return this.flags;\n  }\n\n  set(flagName: string, value: FlagValue): void {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot set flag ${flagName} as it has not been registered.`);\n    }\n    this.flags[flagName] = value;\n    if (this.flagRegistry[flagName].setHook != null) {\n      this.flagRegistry[flagName].setHook(value);\n    }\n  }\n\n  private evaluateFlag(flagName: string): FlagValue|Promise<FlagValue> {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n    }\n    return this.flagRegistry[flagName].evaluationFn();\n  }\n\n  setFlags(flags: Flags) {\n    this.flags = Object.assign({}, flags);\n  }\n\n  reset() {\n    this.flags = {};\n    this.urlFlags = {};\n    this.populateURLFlags();\n  }\n\n  private populateURLFlags(): void {\n    if (typeof this.global === 'undefined' ||\n        typeof this.global.location === 'undefined' ||\n        typeof this.global.location.search === 'undefined') {\n      return;\n    }\n\n    const urlParams = getQueryParams(this.global.location.search);\n    if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n      const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n      keyValues.forEach(keyValue => {\n        const [key, value] = keyValue.split(':') as [string, string];\n        this.urlFlags[key] = parseValue(key, value);\n      });\n    }\n  }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n  const params = {};\n  queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n    decodeParam(params, t[0], t[1]);\n    return t.join('=');\n  });\n  return params;\n}\n\nfunction decodeParam(\n    params: {[key: string]: string}, name: string, value?: string) {\n  params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n  value = value.toLowerCase();\n  if (value === 'true' || value === 'false') {\n    return value === 'true';\n  } else if (`${+ value}` === value) {\n    return +value;\n  }\n  throw new Error(\n      `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n  return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n  ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map<string, any>};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map<string, any>} {\n  if (globalNameSpace == null) {\n    // tslint:disable-next-line:no-any\n    let ns: any;\n    if (typeof (window) !== 'undefined') {\n      ns = window;\n    } else if (typeof (global) !== 'undefined') {\n      ns = global;\n    } else if (typeof (process) !== 'undefined') {\n      ns = process;\n    } else if (typeof (self) !== 'undefined') {\n      ns = self;\n    } else {\n      throw new Error('Could not find a global object');\n    }\n    globalNameSpace = ns;\n  }\n  return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map<string, any> {\n  const ns = getGlobalNamespace();\n  if (ns._tfGlobals == null) {\n    ns._tfGlobals = new Map();\n  }\n  return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n *             the first time it is fetched.\n */\nexport function getGlobal<T>(key: string, init: () => T): T {\n  const globalMap = getGlobalMap();\n  if (globalMap.has(key)) {\n    return globalMap.get(key);\n  } else {\n    const singleton = init();\n    globalMap.set(key, singleton);\n    return globalMap.get(key);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AllAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AnyAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMaxAttrs {\n  axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMinAttrs {\n  axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\nexport interface BatchMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BatchToSpaceNDAttrs {\n  blockShape: number[];\n  crops: number[][];\n}\n\nexport type BinaryInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BroadCastToAttrs {\n  shape: number[];\n  inputShape: number[];  // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n  dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n  clipValueMin: number;\n  clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick<NamedTensorInfoMap, 'real'|'imag'>;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n  axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'dy'>;\nexport interface Conv2DBackpropFilterAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'|'filter'>;\nexport interface Conv2DBackpropInputAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv3DAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'y'>;\n\nexport interface Conv3DBackpropFilterAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'>;\nexport interface Conv3DBackpropInputAttrs {\n  pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface CumsumAttrs {\n  axis: number;\n  exclusive: boolean;\n  reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n    Pick<NamedTensorInfoMap, 'image'|'boxes'|'boxInd'>;\nexport interface CropAndResizeAttrs {\n  cropSize: [number, number];\n  method: 'bilinear'|'nearest';\n  extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface DepthToSpaceAttrs {\n  blockSize: number;\n  dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface DepthwiseConv2dNativeAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n    'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'dy'>;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n    'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'dy'>;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Dilation2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick<NamedTensorInfoMap, 'dy'|'y'>;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n  shape: number[];\n  value: number|string;\n  dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick<NamedTensorInfoMap, 'image'>;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n    Pick<NamedTensorInfoMap, 'x'|'scale'|'offset'|'mean'|'variance'>;\nexport interface FusedBatchNormAttrs {\n  varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick<NamedTensorInfoMap, 'x'|'indices'>;\nexport interface GatherV2Attrs {\n  axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick<NamedTensorInfoMap, 'params'|'indices'>;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n  start: number;\n  stop: number;\n  num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface LogSoftmaxAttrs {\n  axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface LRNAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick<NamedTensorInfoMap, 'x'|'y'|'dy'>;\nexport interface LRNBackpropAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxAttrs {\n  reductionIndices: number|number[];\n  keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolWithArgmaxAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MeanAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MinAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV3Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV4Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV5Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick<NamedTensorInfoMap, 'indices'>;\nexport interface OneHotAttrs {\n  depth: number;\n  onValue: number;\n  offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface PadV2Attrs {\n  paddings: Array<[number, number]>;\n  constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick<NamedTensorInfoMap, 'x'|'alpha'>;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ProdAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n  start: number;\n  stop: number;\n  step: number;\n  dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReshapeAttrs {\n  shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeNearestNeighborAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n    Pick<NamedTensorInfoMap, 'images'>;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeBilinearAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick<NamedTensorInfoMap, 'images'>;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReverseAttrs {\n  dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick<NamedTensorInfoMap, 'indices'|'updates'>;\nexport interface ScatterNdAttrs {\n  shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick<NamedTensorInfoMap, 'condition'|'t'|'e'>;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SliceAttrs {\n  begin: number|number[];\n  size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SumAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SpaceToBatchNDAttrs {\n  blockShape: number[];\n  paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SplitVAttrs {\n  numOrSizeSplits: number[]|number;\n  axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface SoftmaxAttrs {\n  dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n    Pick<NamedTensorInfoMap, 'sparseIndices'|'sparseValues'|'defaultValue'>;\nexport interface SparseToDenseAttrs {\n  outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface StridedSliceAttrs {\n  begin: number[];\n  end: number[];\n  strides: number[];\n  beginMask: number;\n  endMask: number;\n  ellipsisMask: number;\n  newAxisMask: number;\n  shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TileAttrs {\n  reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TopKAttrs {\n  k: number;\n  sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TransposeAttrs {\n  perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface UniqueAttrs {\n  axis: number;\n}\n\nexport type UnaryInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick<NamedTensorInfoMap, 'value'>;\nexport interface UnpackAttrs {\n  axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n    Pick<NamedTensorInfoMap, 'x'|'segmentIds'>;\nexport interface UnsortedSegmentSumAttrs {\n  numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n  alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n  pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n      HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n  numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick<NamedTensorInfoMap, 'image'>;\nexport interface RotateWithOffsetAttrs {\n  radians: number;\n  fillValue: number|[number, number, number];\n  center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n  a: TensorInfo;\n  b: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n  activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n    getGlobal('kernelRegistry', () => new Map<string, KernelConfig>());\nconst gradRegistry =\n    getGlobal('gradRegistry', () => new Map<string, GradConfig>());\n\nexport type DataId = object;\n\ntype AttributeValue =\n    number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray<AttributeValue>;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n  inputs: NamedTensorInfoMap,\n  backend: {},\n  attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n    (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n        NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n  kernelName: string;\n  backendName: string;\n  kernelFunc: KernelFunc;\n  setupFunc?: KernelSetupFunc;\n  disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n  kernelName: string;\n  inputsToSave?: string[];\n  // When saveAllInputs is true, all inputs will be saved. Only use this flag\n  // if inputs is an array of Tensors.\n  saveAllInputs?: boolean;\n  outputsToSave?: boolean[];\n  gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n  dataId: DataId;\n  shape: number[];\n  dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n  [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n  [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n    kernelName: string, backendName: string): KernelConfig {\n  const key = makeKey(kernelName, backendName);\n  return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n  return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n  const it = kernelRegistry.entries();\n  const result: KernelConfig[] = [];\n\n  while (true) {\n    const {done, value} = it.next();\n    if (done) {\n      break;\n    }\n    const [key, config] = value;\n    const [backend, ] = key.split('_');\n    if (backend === backendName) {\n      result.push(config);\n    }\n  }\n  return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n  const {kernelName, backendName} = config;\n  const key = makeKey(kernelName, backendName);\n  if (kernelRegistry.has(key)) {\n    console.warn(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is already registered`);\n  }\n  kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n  const {kernelName} = config;\n\n  if (gradRegistry.has(kernelName)) {\n    // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n    // to debug mode.\n    if (env().getBool('DEBUG')) {\n      console.warn(`Overriding the gradient for '${kernelName}'`);\n    }\n  }\n  gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n    kernelName: string, backendName: string): void {\n  const key = makeKey(kernelName, backendName);\n  if (!kernelRegistry.has(key)) {\n    throw new Error(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is not registered`);\n  }\n  kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n  if (!gradRegistry.has(kernelName)) {\n    throw new Error(\n        `The gradient '${kernelName}' for backend is not registered`);\n  }\n  gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n    registeredBackendName: string, newBackendName: string): void {\n  const kernels = getKernelsForBackend(registeredBackendName);\n  kernels.forEach(kernelConfig => {\n    const newKernelConfig =\n        Object.assign({}, kernelConfig, {backendName: newBackendName});\n    registerKernel(newKernelConfig);\n  });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n  return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n                        Float32Array): void {\n  let counter = array.length;\n  let temp = 0;\n  let index = 0;\n  // While there are elements in the array\n  while (counter > 0) {\n    // Pick a random index\n    index = (Math.random() * counter) | 0;\n    // Decrease counter by 1\n    counter--;\n    // And swap the last element with it\n    temp = array[counter];\n    array[counter] = array[index];\n    array[index] = temp;\n  }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n  return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n  return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n  let sum = 0;\n  for (let i = 0; i < arr.length; i++) {\n    sum += arr[i];\n  }\n  return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n  const r = Math.random();\n  return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n  let result = 0;\n  for (let i = 0; i < a.length; i++) {\n    const diff = Number(a[i]) - Number(b[i]);\n    result += diff * diff;\n  }\n  return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n *     error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n  if (!expr) {\n    throw new Error(typeof msg === 'string' ? msg : msg());\n  }\n}\n\nexport function assertShapesMatch(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  assert(\n      arraysEqual(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n  assert(\n      a != null,\n      () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten<number>().\n/**\n *  Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n *  @param arr The nested array to flatten.\n *  @param result The destination array which holds the elements.\n *  @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n *      to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten<T extends number|boolean|string|Promise<number>|TypedArray>(\n    arr: T|RecursiveArray<T>, result: T[] = [], skipTypedArray = false): T[] {\n  if (result == null) {\n    result = [];\n  }\n  if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n    for (let i = 0; i < arr.length; ++i) {\n      flatten(arr[i], result, skipTypedArray);\n    }\n  } else {\n    result.push(arr as T);\n  }\n  return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n  if (shape.length === 0) {\n    // Scalar.\n    return 1;\n  }\n  let size = shape[0];\n  for (let i = 1; i < shape.length; i++) {\n    size *= shape[i];\n  }\n  return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n  return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n  if (n1 === n2) {\n    return true;\n  }\n  if (n1 == null || n2 == null) {\n    return false;\n  }\n\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function isInt(a: number): boolean {\n  return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n  // tslint:disable-next-line:no-any\n  if ((Math as any).tanh != null) {\n    // tslint:disable-next-line:no-any\n    return (Math as any).tanh(x);\n  }\n  if (x === Infinity) {\n    return 1;\n  } else if (x === -Infinity) {\n    return -1;\n  } else {\n    const e2x = Math.exp(2 * x);\n    return (e2x - 1) / (e2x + 1);\n  }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n  const width = Math.ceil(Math.sqrt(size));\n  return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n  const shuffledIndices = new Uint32Array(n);\n  for (let i = 0; i < n; ++i) {\n    shuffledIndices[i] = i;\n  }\n  shuffle(shuffledIndices);\n  return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n  if (size <= a.length) {\n    return a;\n  }\n  return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n    checkFn: () => boolean, delayFn = (counter: number) => 0,\n    maxCounter?: number): Promise<void> {\n  return new Promise<void>((resolve, reject) => {\n    let tryCount = 0;\n\n    const tryFn = () => {\n      if (checkFn()) {\n        resolve();\n        return;\n      }\n\n      tryCount++;\n\n      const nextBackoff = delayFn(tryCount);\n\n      if (maxCounter != null && tryCount >= maxCounter) {\n        reject();\n        return;\n      }\n      setTimeout(tryFn, nextBackoff);\n    };\n\n    tryFn();\n  });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n    shape: number[], size: number): number[] {\n  let shapeProd = 1;\n  let implicitIdx = -1;\n\n  for (let i = 0; i < shape.length; ++i) {\n    if (shape[i] >= 0) {\n      shapeProd *= shape[i];\n    } else if (shape[i] === -1) {\n      if (implicitIdx !== -1) {\n        throw Error(\n            `Shapes can only have 1 implicit size. ` +\n            `Found -1 at dim ${implicitIdx} and dim ${i}`);\n      }\n      implicitIdx = i;\n    } else if (shape[i] < 0) {\n      throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n    }\n  }\n\n  if (implicitIdx === -1) {\n    if (size > 0 && size !== shapeProd) {\n      throw Error(`Size(${size}) must match the product of shape ${shape}`);\n    }\n    return shape;\n  }\n\n  if (shapeProd === 0) {\n    throw Error(\n        `Cannot infer the missing size in [${shape}] when ` +\n        `there are 0 elements`);\n  }\n  if (size % shapeProd !== 0) {\n    throw Error(\n        `The implicit shape can't be a fractional number. ` +\n        `Got ${size} / ${shapeProd}`);\n  }\n\n  const newShape = shape.slice();\n  newShape[implicitIdx] = size / shapeProd;\n  return newShape;\n}\n\nexport function parseAxisParam(\n    axis: number|number[], shape: number[]): number[] {\n  const rank = shape.length;\n\n  // Normalize input\n  axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n  // Check for valid range\n  assert(\n      axis.every(ax => ax >= -rank && ax < rank),\n      () =>\n          `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n          `got axis ${axis}`);\n\n  // Check for only integers\n  assert(\n      axis.every(ax => isInt(ax)),\n      () => `All values in axis param must be integers but ` +\n          `got axis ${axis}`);\n\n  // Handle negative axis.\n  return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n    {newShape: number[], keptDims: number[]} {\n  const newShape: number[] = [];\n  const keptDims: number[] = [];\n  const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n  const axes = (axis == null || isEmptyArray) ?\n      null :\n      parseAxisParam(axis, shape).sort();\n  let j = 0;\n  for (let i = 0; i < shape.length; ++i) {\n    if (axes != null) {\n      if (axes[j] === i && shape[i] !== 1) {\n        throw new Error(\n            `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n      }\n      if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n        newShape.push(shape[i]);\n        keptDims.push(i);\n      }\n      if (axes[j] <= i) {\n        j++;\n      }\n    }\n    if (shape[i] !== 1) {\n      newShape.push(shape[i]);\n      keptDims.push(i);\n    }\n  }\n  return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType<D extends NumericDataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType<D extends DataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else if (dtype === 'string') {\n    values = new Array<'string'>(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors<D extends DataType>(\n    vals: DataTypeMap[D]|number[], dtype: D): void {\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n    }\n  }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n  return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n      dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n  if (newType === 'complex64') {\n    return false;\n  }\n  if (newType === 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'bool' && oldType === 'bool') {\n    return false;\n  }\n  return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n  return a instanceof Float32Array || a instanceof Int32Array ||\n      a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n  if (dtype === 'float32' || dtype === 'int32') {\n    return 4;\n  } else if (dtype === 'complex64') {\n    return 8;\n  } else if (dtype === 'bool') {\n    return 1;\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n  if (arr == null) {\n    return 0;\n  }\n  let bytes = 0;\n  arr.forEach(x => bytes += x.length);\n  return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n  return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n  return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n  return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n  if (Array.isArray(values)) {\n    return inferDtype(values[0]);\n  }\n  if (values instanceof Float32Array) {\n    return 'float32';\n  } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n    return 'int32';\n  } else if (isNumber(values)) {\n    return 'float32';\n  } else if (isString(values)) {\n    return 'string';\n  } else if (isBoolean(values)) {\n    return 'bool';\n  }\n  return 'float32';\n}\n\nexport function isFunction(f: Function) {\n  return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n  for (let i = start; i < size; ++i) {\n    if (size % i === 0) {\n      return i;\n    }\n  }\n  return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n  const rank = shape.length;\n  if (rank < 2) {\n    return [];\n  }\n\n  // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n  // strides.\n  const strides = new Array(rank - 1);\n  strides[rank - 2] = shape[rank - 1];\n  for (let i = rank - 3; i >= 0; --i) {\n    strides[i] = strides[i + 1] * shape[i + 1];\n  }\n  return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n    value: DataType, dtype: DataType): BackendValues {\n  if (dtype === 'string') {\n    return encodeString(value);\n  }\n\n  return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n  if (dtype === 'string') {\n    throw new Error('Cannot convert a string[] to a TypedArray');\n  }\n  if (Array.isArray(a)) {\n    a = flatten(a);\n  }\n\n  if (env().getBool('DEBUG')) {\n    checkConversionForErrors(a as number[], dtype);\n  }\n  if (noConversionNeeded(a, dtype)) {\n    return a as TypedArray;\n  }\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(a as number[]);\n  } else if (dtype === 'int32') {\n    return new Int32Array(a as number[]);\n  } else if (dtype === 'bool') {\n    const bool = new Uint8Array((a as number[]).length);\n    for (let i = 0; i < bool.length; ++i) {\n      if (Math.round((a as number[])[i]) !== 0) {\n        bool[i] = 1;\n      }\n    }\n    return bool;\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n  const ret = new Array();\n  if (shape.length === 1) {\n    const d = shape[0];\n    for (let i = 0; i < d; i++) {\n      ret[i] = a[offset + i];\n    }\n  } else {\n    const d = shape[0];\n    const rest = shape.slice(1);\n    const len = rest.reduce((acc, c) => acc * c);\n    for (let i = 0; i < d; i++) {\n      ret[i] = createNestedArray(offset + i * len, rest, a);\n    }\n  }\n  return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n  if (shape.length === 0) {\n    // Scalar type should return a single number.\n    return a[0];\n  }\n  const size = shape.reduce((acc, c) => acc * c);\n  if (size === 0) {\n    // A tensor with shape zero should be turned into empty list.\n    return [];\n  }\n  if (size !== a.length) {\n    throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n  }\n\n  return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n  return (a instanceof Float32Array && dtype === 'float32') ||\n      (a instanceof Int32Array && dtype === 'int32') ||\n      (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  const array = makeZerosTypedArray(size, dtype);\n  for (let i = 0; i < array.length; i++) {\n    array[i] = 1;\n  }\n  return array;\n}\n\nexport function makeZerosTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'int32') {\n    return new Int32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'bool') {\n    return new Uint8Array(size) as DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray<D extends DataType>(\n    shape: number[], dtype: D) {\n  const size = shape.reduce((prev, curr) => prev * curr, 1);\n  if (dtype == null || dtype === 'float32') {\n    return toNestedArray(shape, new Float32Array(size));\n  } else if (dtype === 'int32') {\n    return toNestedArray(shape, new Int32Array(size));\n  } else if (dtype === 'bool') {\n    return toNestedArray(shape, new Uint8Array(size));\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n  return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n  shape.forEach(dimSize => {\n    assert(\n        Number.isInteger(dimSize) && dimSize >= 0,\n        () =>\n            `Tensor must have a shape comprised of positive integers but got ` +\n            `shape [${shape}].`);\n  });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n    path: string, requestInits?: RequestInit): Promise<Response> {\n  return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n  encoding = encoding || 'utf-8';\n  return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n  encoding = encoding || 'utf-8';\n  return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n    locs: number[], rank: number, strides: number[]): number {\n  if (rank === 0) {\n    return 0;\n  } else if (rank === 1) {\n    return locs[0];\n  }\n  let index = locs[locs.length - 1];\n  for (let i = 0; i < locs.length - 1; ++i) {\n    index += strides[i] * locs[i];\n  }\n  return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n    index: number, rank: number, strides: number[]): number[] {\n  if (rank === 0) {\n    return [];\n  } else if (rank === 1) {\n    return [index];\n  }\n  const locs: number[] = new Array(rank);\n  for (let i = 0; i < locs.length - 1; ++i) {\n    locs[i] = Math.floor(index / strides[i]);\n    index -= locs[i] * strides[i];\n  }\n  locs[locs.length - 1] = index;\n  return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n  kernelName: string,\n  outputs: Tensor[],\n  inputs: NamedTensorMap,\n  timeMs: Promise<number|{error: string}>,\n  extraInfo: Promise<string>\n};\n\nexport class Profiler {\n  constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n    if (logger == null) {\n      this.logger = new Logger();\n    }\n  }\n\n  profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n      KernelProfile {\n    let outputs: Tensor[];\n    const holdResultWrapperFn = () => {\n      outputs = f();\n    };\n    const timer = this.backendTimer.time(holdResultWrapperFn);\n\n    for (let i = 0; i < outputs.length; i++) {\n      const output = outputs[i];\n      // Dangling promise here because we don't want to propagate up\n      // asynchronicity.\n      output.data().then(tensorVals => {\n        checkComputationForErrors(tensorVals, output.dtype, kernelName);\n      });\n    }\n\n    const kernelProfile = {\n      kernelName,\n      outputs,\n      inputs,\n      timeMs: timer.then(timing => timing.kernelMs),\n      extraInfo: timer.then(\n          timing => timing.getExtraProfileInfo != null ?\n              timing.getExtraProfileInfo() :\n              '')\n    };\n    return kernelProfile;\n  }\n\n  logKernelProfile(kernelProfile: KernelProfile): void {\n    const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n    outputs.forEach(result => {\n      Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n        this.logger.logKernelProfile(\n            kernelName, result, valueContainer[0], valueContainer[1], inputs,\n            valueContainer[2]);\n      });\n    });\n  }\n}\n\nexport function checkComputationForErrors<D extends DataType>(\n    vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n  if (dtype !== 'float32') {\n    // Only floating point computations will generate NaN values\n    return false;\n  }\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      // Throwing custom exception so behavior is testable.\n      console.warn(`Found ${num} in the result of '${kernelName}'`);\n      return true;\n    }\n  }\n  return false;\n}\n\nexport class Logger {\n  logKernelProfile(\n      name: string, result: Tensor, vals: TypedArray,\n      timeMs: number|{error: string}, inputs: NamedTensorMap,\n      extraInfo?: string) {\n    const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n                                              timeMs['error'];\n    const paddedName = util.rightPad(name, 25);\n    const rank = result.rank;\n    const size = result.size;\n    const shape = util.rightPad(result.shape.toString(), 14);\n    let inputShapesDescription = '';\n\n    for (const name in inputs) {\n      const input = inputs[name];\n      if (input != null) {\n        // The input might be a non-tensor (e.g HTMLImageElement), in which case\n        // we claim the output shape as input shape.\n        const inputShape = input.shape || result.shape;\n        const inputRank = inputShape.length;\n        inputShapesDescription +=\n            `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n      }\n    }\n\n    console.log(\n        `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n            inputShapesDescription}\\t%c${extraInfo}`,\n        'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n        'color: green', 'color: steelblue');\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n  id: number;\n  kernelName: string;\n  outputs: Tensor[];\n  inputs: NamedTensorMap;\n  // Optional params, defined only for ops with gradient impl.\n  gradient?: (dys: Tensor[]) => NamedGradientMap;\n  saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n  [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n    tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n  // Forward pass to compute all the nodes and Tensors that are transitively a\n  // function of x.\n  const tensorsFromX: {[tensorId: number]: boolean} = {};\n  const nodesFromX: {[nodeId: number]: boolean} = {};\n  for (let i = 0; i < xs.length; i++) {\n    tensorsFromX[xs[i].id] = true;\n  }\n\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n    for (const inputName in nodeInputs) {\n      const input = nodeInputs[inputName];\n\n      let anyInputFromX = false;\n      for (let j = 0; j < xs.length; j++) {\n        if (tensorsFromX[input.id]) {\n          node.outputs.forEach(output => tensorsFromX[output.id] = true);\n          anyInputFromX = true;\n          nodesFromX[node.id] = true;\n          break;\n        }\n      }\n\n      if (anyInputFromX) {\n        break;\n      }\n    }\n  }\n\n  // Backward pass to find all of the nodes and Tensors that lead to y.\n  const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n  tensorsLeadToY[y.id] = true;\n  const nodesToY: {[nodeId: number]: boolean} = {};\n\n  for (let i = tape.length - 1; i >= 0; i--) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n\n    // If any of the outputs lead to y, mark all of the inputs as leading to y.\n    for (let j = 0; j < node.outputs.length; j++) {\n      if (tensorsLeadToY[node.outputs[j].id]) {\n        for (const inputName in nodeInputs) {\n          tensorsLeadToY[nodeInputs[inputName].id] = true;\n          nodesToY[node.id] = true;\n        }\n        break;\n      }\n    }\n  }\n\n  // Return the paths that come from x and lead to y.\n  const filteredTape: TapeNode[] = [];\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n\n    if (nodesFromX[node.id] && nodesToY[node.id]) {\n      // Prune the inputs from the node that aren't a function of x.\n      const prunedInputs: {[inputName: string]: Tensor} = {};\n      for (const inputName in node.inputs) {\n        const nodeInput = node.inputs[inputName];\n        if (tensorsFromX[nodeInput.id]) {\n          prunedInputs[inputName] = nodeInput;\n        }\n      }\n\n      // Copy the node and overwrite inputsAndArgs to the pruned version.\n      const prunedNode = Object.assign({}, node);\n      prunedNode.inputs = prunedInputs;\n      prunedNode.outputs = node.outputs;\n\n      filteredTape.push(prunedNode);\n    }\n  }\n\n  return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n    tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n    filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n    add: (a: Tensor, b: Tensor) => Tensor) {\n  // Walk the tape backward and keep a map of Tensor to its gradient.\n  for (let i = filteredTape.length - 1; i >= 0; i--) {\n    const node = filteredTape[i];\n\n    const dys: Tensor[] = [];\n    node.outputs.forEach(o => {\n      const gradTensor = tensorAccumulatedGradientMap[o.id];\n      if (gradTensor != null) {\n        dys.push(gradTensor);\n      } else {\n        // This particular output is not in the back-propagation subgraph, so it\n        // does not affect the final output, thus we put null for its dy.\n        dys.push(null);\n      }\n    });\n\n    if (node.gradient == null) {\n      throw new Error(\n          `Cannot compute gradient: gradient function not found ` +\n          `for ${node.kernelName}.`);\n    }\n\n    // Backprop dy through this node and accumulate gradients over the inputs.\n    const inputGradients = node.gradient(dys);\n\n    for (const inputName in node.inputs) {\n      if (!(inputName in inputGradients)) {\n        throw new Error(\n            `Cannot backprop through input ${inputName}. ` +\n            `Available gradients found: ${Object.keys(inputGradients)}.`);\n      }\n\n      // Call the gradient function.\n      const dx = tidy(() => inputGradients[inputName]());\n      if (dx.dtype !== 'float32') {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n      }\n      const x = node.inputs[inputName];\n      if (!util.arraysEqual(dx.shape, x.shape)) {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `'${inputName}' has shape '${dx.shape}', which does not match ` +\n            `the shape of the input '${x.shape}'`);\n      }\n\n      if (tensorAccumulatedGradientMap[x.id] == null) {\n        tensorAccumulatedGradientMap[x.id] = dx;\n      } else {\n        const curGradient = tensorAccumulatedGradientMap[x.id];\n        tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n        curGradient.dispose();\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    verbose: boolean) {\n  const strides = computeStrides(shape);\n  const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n  const rank = shape.length;\n  const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n  const lines = ['Tensor'];\n  if (verbose) {\n    lines.push(`  dtype: ${dtype}`);\n    lines.push(`  rank: ${rank}`);\n    lines.push(`  shape: [${shape}]`);\n    lines.push(`  values:`);\n  }\n  lines.push(valsLines.map(l => '    ' + l).join('\\n'));\n  return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[]): number[] {\n  const n = sizeFromShape(shape);\n  const numCols = strides[strides.length - 1];\n  const padPerCol = new Array(numCols).fill(0);\n  const rank = shape.length;\n  const valuesOrTuples =\n      dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n  if (rank > 1) {\n    for (let row = 0; row < n / numCols; row++) {\n      const offset = row * numCols;\n      for (let j = 0; j < numCols; j++) {\n        padPerCol[j] = Math.max(\n            padPerCol[j],\n            valToString(valuesOrTuples[offset + j], 0, dtype).length);\n      }\n    }\n  }\n  return padPerCol;\n}\n\nfunction valToString(\n    val: number|string|[number, number], pad: number, dtype: DataType) {\n  let valStr: string;\n  if (Array.isArray(val)) {\n    valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n        `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n  } else if (isString(val)) {\n    valStr = `'${val}'`;\n  } else if (dtype === 'bool') {\n    valStr = boolNumToString(val);\n  } else {\n    valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n  }\n\n  return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n  return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[], padPerCol: number[], isLast = true): string[] {\n  const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n  const size = shape[0];\n  const rank = shape.length;\n  if (rank === 0) {\n    if (dtype === 'complex64') {\n      const complexTuple = createComplexTuples(vals);\n      return [valToString(complexTuple[0], 0, dtype)];\n    }\n    if (dtype === 'bool') {\n      return [boolNumToString(vals[0] as number)];\n    }\n    return [vals[0].toString()];\n  }\n\n  if (rank === 1) {\n    if (size > FORMAT_LIMIT_NUM_VALS) {\n      const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n      let firstVals = Array.from<number|string|[number, number]>(\n          vals.slice(0, firstValsSize));\n      let lastVals = Array.from<number|string|[number, number]>(vals.slice(\n          (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n          size * storagePerElement));\n      if (dtype === 'complex64') {\n        firstVals = createComplexTuples(firstVals);\n        lastVals = createComplexTuples(lastVals);\n      }\n      return [\n        '[' +\n        firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n            .join(', ') +\n        ', ..., ' +\n        lastVals\n            .map(\n                (x, i) => valToString(\n                    x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n            .join(', ') +\n        ']'\n      ];\n    }\n    const displayVals: Array<number|string|[number, number]> =\n        dtype === 'complex64' ? createComplexTuples(vals) :\n                                Array.from<number|string>(vals);\n\n    return [\n      '[' +\n      displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n          .join(', ') +\n      ']'\n    ];\n  }\n\n  // The array is rank 2 or more.\n  const subshape = shape.slice(1);\n  const substrides = strides.slice(1);\n  const stride = strides[0] * storagePerElement;\n  const lines: string[] = [];\n  if (size > FORMAT_LIMIT_NUM_VALS) {\n    for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          false /* isLast */));\n    }\n    lines.push('...');\n    for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  } else {\n    for (let i = 0; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  }\n  const sep = rank === 2 ? ',' : '';\n  lines[0] = '[' + lines[0] + sep;\n  for (let i = 1; i < lines.length - 1; i++) {\n    lines[i] = ' ' + lines[i] + sep;\n  }\n  let newLineSep = ',\\n';\n  for (let i = 2; i < rank; i++) {\n    newLineSep += '\\n';\n  }\n  lines[lines.length - 1] =\n      ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n  return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n                             TypedArray): Array<[number, number]> {\n  const complexTuples: Array<[number, number]> = [];\n  for (let i = 0; i < vals.length; i += 2) {\n    complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n  }\n  return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData<D extends DataType> {\n  dataId?: DataId;\n  values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer<R extends Rank, D extends DataType = 'float32'> {\n  size: number;\n  shape: ShapeMap[R];\n  strides: number[];\n  values: DataTypeMap[D];\n\n  constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.size = util.sizeFromShape(shape);\n\n    if (values != null) {\n      const n = values.length;\n      util.assert(\n          n === this.size,\n          () => `Length of values '${n}' does not match the size ` +\n              `inferred by the shape '${this.size}'.`);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `complex64 dtype TensorBuffers are not supported. Please create ` +\n          `a TensorBuffer for the real and imaginary parts separately and ` +\n          `call tf.complex(real, imag).`);\n    }\n    this.values = values || util.getArrayFromDType(dtype, this.size);\n    this.strides = computeStrides(shape);\n  }\n\n  /**\n   * Sets a value in the buffer at a given location.\n   *\n   * @param value The value to set.\n   * @param locs  The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  set(value: SingleValueMap[D], ...locs: number[]): void {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    util.assert(\n        locs.length === this.rank,\n        () => `The number of provided coordinates (${locs.length}) must ` +\n            `match the rank (${this.rank})`);\n\n    const index = this.locToIndex(locs);\n    this.values[index] = value as number;\n  }\n\n  /**\n   * Returns the value in the buffer at the provided location.\n   *\n   * @param locs The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  get(...locs: number[]): SingleValueMap[D] {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    let i = 0;\n    for (const loc of locs) {\n      if (loc < 0 || loc >= this.shape[i]) {\n        const msg = `Requested out of range element at ${locs}. ` +\n            `  Buffer shape=${this.shape}`;\n        throw new Error(msg);\n      }\n      i++;\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return this.values[index] as SingleValueMap[D];\n  }\n\n  locToIndex(locs: number[]): number {\n    if (this.rank === 0) {\n      return 0;\n    } else if (this.rank === 1) {\n      return locs[0];\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return index;\n  }\n\n  indexToLoc(index: number): number[] {\n    if (this.rank === 0) {\n      return [];\n    } else if (this.rank === 1) {\n      return [index];\n    }\n    const locs: number[] = new Array(this.shape.length);\n    for (let i = 0; i < locs.length - 1; ++i) {\n      locs[i] = Math.floor(index / this.strides[i]);\n      index -= locs[i] * this.strides[i];\n    }\n    locs[locs.length - 1] = index;\n    return locs;\n  }\n\n  get rank() {\n    return this.shape.length;\n  }\n\n  /**\n   * Creates an immutable `tf.Tensor` object from the buffer.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  toTensor(): Tensor<R> {\n    return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n        Tensor<R>;\n  }\n}\n\nexport interface TensorTracker {\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: Backend): Tensor;\n  makeVariable(\n      initialValue: Tensor, trainable?: boolean, name?: string,\n      dtype?: DataType): Variable;\n  incRef(a: Tensor, backend: Backend): void;\n  disposeTensor(t: Tensor): void;\n  disposeVariable(v: Variable): void;\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n  cast<T extends Tensor>(x: T, dtype: DataType): T;\n  buffer<R extends Rank, D extends DataType>(\n      shape: ShapeMap[R], dtype: D,\n      values?: DataTypeMap[D]): TensorBuffer<R, D>;\n  print<T extends Tensor>(x: T, verbose: boolean): void;\n  clone<T extends Tensor>(x: T): T;\n  // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n  trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n  opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n  deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object;  // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor<R extends Rank = Rank> {\n  /** Unique id of this tensor. */\n  readonly id: number;\n  /**\n   * Id of the bucket holding the data for this tensor. Multiple arrays can\n   * point to the same bucket (e.g. when calling array.reshape()).\n   */\n  dataId: DataId;\n  /** The shape of the tensor. */\n  readonly shape: ShapeMap[R];\n  /** Number of elements in the tensor. */\n  readonly size: number;\n  /** The data type for the array. */\n  readonly dtype: DataType;\n  /** The rank type for the array (see `Rank` enum). */\n  readonly rankType: R;\n\n  /** Whether this tensor has been globally kept. */\n  kept = false;\n  /** The id of the scope this tensor is being tracked in. */\n  scopeId: number;\n\n  /**\n   * Number of elements to skip in each dimension when indexing. See\n   * https://docs.scipy.org/doc/numpy/reference/generated/\\\n   * numpy.ndarray.strides.html\n   */\n  readonly strides: number[];\n\n  constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.dtype = dtype || 'float32';\n    this.size = util.sizeFromShape(shape);\n    this.strides = computeStrides(shape);\n    this.dataId = dataId;\n    this.id = id;\n    this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n  }\n\n  get rank(): number {\n    return this.shape.length;\n  }\n\n  /**\n   * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async buffer<D extends DataType = 'float32'>(): Promise<TensorBuffer<R, D>> {\n    const vals = await this.data<D>();\n    return opHandler.buffer(this.shape, this.dtype as D, vals);\n  }\n\n  /**\n   * Returns a `tf.TensorBuffer` that holds the underlying data.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  bufferSync<D extends DataType = 'float32'>(): TensorBuffer<R, D> {\n    return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * asynchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async array(): Promise<ArrayMap[R]> {\n    const vals = await this.data();\n    return toNestedArray(this.shape, vals) as ArrayMap[R];\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * synchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  arraySync(): ArrayMap[R] {\n    return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n  }\n\n  /**\n   * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n   * promise of `TypedArray` that resolves when the computation has finished.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]> {\n    this.throwIfDisposed();\n    const data = trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      const bytes = await data as Uint8Array[];\n      try {\n        return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as Promise<DataTypeMap[D]>;\n  }\n\n  /**\n   * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n   * UI thread until the values are ready, which can cause performance issues.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D] {\n    this.throwIfDisposed();\n    const data = trackerFn().readSync(this.dataId);\n    if (this.dtype === 'string') {\n      try {\n        return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n            DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as DataTypeMap[D];\n  }\n\n  /** Returns the underlying bytes of the tensor's data. */\n  async bytes(): Promise<Uint8Array[]|Uint8Array> {\n    this.throwIfDisposed();\n    const data = await trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      return data as Uint8Array[];\n    } else {\n      return new Uint8Array((data as TypedArray).buffer);\n    }\n  }\n\n  /**\n   * Disposes `tf.Tensor` from memory.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dispose(): void {\n    if (this.isDisposed) {\n      return;\n    }\n    trackerFn().disposeTensor(this);\n    this.isDisposedInternal = true;\n  }\n\n  protected isDisposedInternal = false;\n  get isDisposed(): boolean {\n    return this.isDisposedInternal;\n  }\n\n  throwIfDisposed() {\n    if (this.isDisposed) {\n      throw new Error(`Tensor is disposed.`);\n    }\n  }\n\n  /**\n   * Prints the `tf.Tensor`. See `tf.print` for details.\n   *\n   * @param verbose Whether to print verbose information about the tensor,\n   *    including dtype and size.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  print(verbose = false): void {\n    return opHandler.print(this, verbose);\n  }\n\n  /**\n   * Returns a copy of the tensor. See `tf.clone` for details.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  clone<T extends Tensor>(this: T): T {\n    this.throwIfDisposed();\n    return opHandler.clone(this);\n  }\n\n  /**\n   * Returns a human-readable description of the tensor. Useful for logging.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  toString(verbose = false): string {\n    const vals = this.dataSync();\n    return tensorToString(vals, this.shape, this.dtype, verbose);\n  }\n\n  cast<T extends this>(dtype: DataType): T {\n    this.throwIfDisposed();\n    return opHandler.cast(this as T, dtype);\n  }\n  variable(trainable = true, name?: string, dtype?: DataType): Variable<R> {\n    this.throwIfDisposed();\n    return trackerFn().makeVariable(this, trainable, name, dtype) as\n        Variable<R>;\n  }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n  value: (instance: Tensor) => {\n    // Implementation note: we should use properties of the object that will be\n    // defined before the constructor body has finished executing (methods).\n    // This is because when this code is transpiled by babel, babel will call\n    // classCallCheck before the constructor body is run.\n    // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n    return !!instance && instance.data != null && instance.dataSync != null &&\n        instance.throwIfDisposed != null;\n  }\n});\n\nexport interface NumericTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: NumericDataType;\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D];\n  data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]>;\n}\n\nexport interface StringTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: 'string';\n  dataSync<D extends DataType = 'string'>(): DataTypeMap[D];\n  data<D extends DataType = 'string'>(): Promise<DataTypeMap[D]>;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor<Rank.R0>;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor<Rank.R1>;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor<Rank.R2>;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor<Rank.R3>;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor<Rank.R4>;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor<Rank.R5>;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor<Rank.R6>;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable<R extends Rank = Rank> extends Tensor<R> {\n  name: string;\n\n  constructor(\n      initialValue: Tensor<R>, public trainable: boolean, name: string,\n      tensorId: number) {\n    super(\n        initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n    this.name = name;\n  }\n\n  /**\n   * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n   * the same shape and dtype as the old `tf.Tensor`.\n   *\n   * @param newValue New tensor to be assigned to this variable.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  assign(newValue: Tensor<R>): void {\n    if (newValue.dtype !== this.dtype) {\n      throw new Error(\n          `dtype of the new value (${newValue.dtype}) and ` +\n          `previous value (${this.dtype}) must match`);\n    }\n    if (!util.arraysEqual(newValue.shape, this.shape)) {\n      throw new Error(\n          `shape of the new value (${newValue.shape}) and ` +\n          `previous value (${this.shape}) must match`);\n    }\n    trackerFn().disposeTensor(this);\n    this.dataId = newValue.dataId;\n    trackerFn().incRef(this, null /* backend */);\n  }\n\n  dispose(): void {\n    trackerFn().disposeVariable(this);\n    this.isDisposedInternal = true;\n  }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n  value: (instance: Variable) => {\n    return instance instanceof Tensor && instance.assign != null &&\n        instance.assign instanceof Function;\n  }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n  R0: number[];\n  R1: [number];\n  R2: [number, number];\n  R3: [number, number, number];\n  R4: [number, number, number, number];\n  R5: [number, number, number, number, number];\n  R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n  R0: number;\n  R1: number[];\n  R2: number[][];\n  R3: number[][][];\n  R4: number[][][][];\n  R5: number[][][][][];\n  R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n  float32: Float32Array;\n  int32: Int32Array;\n  bool: Uint8Array;\n  complex64: Float32Array;\n  string: string[];\n}\n\nexport interface SingleValueMap {\n  bool: boolean;\n  int32: number;\n  float32: number;\n  complex64: number;\n  string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n  R0 = 'R0',\n  R1 = 'R1',\n  R2 = 'R2',\n  R3 = 'R3',\n  R4 = 'R4',\n  R5 = 'R5',\n  R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray<T> =\n    T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray<T extends any> {\n  [index: number]: T|RecursiveArray<T>;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'int32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'bool',\n  'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n  'float32' = 'float32',\n  'int32' = 'float32',\n  'bool' = 'float32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n  'float32' = 'complex64',\n  'int32' = 'complex64',\n  'bool' = 'complex64',\n  'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n  'float32': UpcastFloat32AndMap,\n  'int32': UpcastInt32AndMap,\n  'bool': UpcastBoolAndMap,\n  'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n  if (typeA === 'string' || typeB === 'string') {\n    if (typeA === 'string' && typeB === 'string') {\n      return 'string';\n    }\n    throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n  }\n  return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n  return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n    TypedArray|number|boolean|string|RecursiveArray<number|number[]|TypedArray>|\n    RecursiveArray<boolean>|RecursiveArray<string>|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n    string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n    boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n    boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n    TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n    string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n    TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n    string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n  width: number;\n  height: number;\n  data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch<T extends Tensor>(a: T, b: T): [T, T] {\n  if (a.dtype === b.dtype) {\n    return [a, b];\n  }\n  const dtype = upcastType(a.dtype, b.dtype);\n  return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n  assert(\n      a.dtype === b.dtype,\n      () => `The dtypes of the first(${a.dtype}) and` +\n          ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n  return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n *   `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n *   is safe to pass any object here, except that `Promise`s are not\n *   supported.\n * @returns An array of `Tensors` found within the passed object. If the\n *   argument is simply a `Tensor', a list containing that `Tensor` is\n *   returned. If the object is not a `Tensor` or does not\n *   contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n  const list: Tensor[] = [];\n  const seen = new Set<{}|void>();\n  walkTensorContainer(result, list, seen);\n  return list;\n}\n\nfunction walkTensorContainer(\n    container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n  if (container == null) {\n    return;\n  }\n  if (container instanceof Tensor) {\n    list.push(container);\n    return;\n  }\n  if (!isIterable(container)) {\n    return;\n  }\n  // Iteration over keys works also for arrays.\n  const iterable = container as TensorContainerArray;\n  for (const k in iterable) {\n    const val = iterable[k];\n    if (!seen.has(val)) {\n      seen.add(val);\n      walkTensorContainer(val, list, seen);\n    }\n  }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n  return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc<T> = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n *   value: Tensor,\n *   gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc<T extends Tensor> =\n    (...inputs: Array<Tensor|GradSaveFunc>) => {\n      value: T;\n      gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n    };\n\nexport type MemoryInfo = {\n  numTensors: number; numDataBuffers: number; numBytes: number;\n  unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n  name: string; bytesAdded: number; totalBytesSnapshot: number;\n  tensorsAdded: number;\n  totalTensorsSnapshot: number;\n  inputShapes: number[][];\n  outputShapes: number[][];\n  kernelTimeMs: number | {error: string} | Promise<number|{error: string}>;\n  extraInfo: string | Promise<string>;\n};\n\nexport type ProfileInfo = {\n  newBytes: number; newTensors: number; peakBytes: number;\n  kernels: KernelInfo[];\n  result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n  wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn<T extends TensorContainer> = () => T;\n\ninterface ScopeState {\n  track: Tensor[];\n  name: string;\n  id: number;\n}\n\nclass EngineState {\n  // Public since optimizers will use it.\n  registeredVariables: NamedVariableMap = {};\n\n  nextTapeNodeId = 0;\n  numBytes = 0;\n  numTensors = 0;\n  numStringTensors = 0;\n  numDataBuffers = 0;\n\n  activeTape: TapeNode[];\n  // Number of nested tf.grad() statements when computing higher-order\n  // gradients. E.g. `1` for first-order gradients and `2` for second-order\n  // gradients. Used to track if the tape should be removed after a backprop.\n  gradientDepth = 0;\n  // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n  // off the tape.\n  kernelDepth = 0;\n\n  // Keep Tensors that parallel the tapes.\n  activeScope: ScopeState;\n  scopeStack: ScopeState[] = [];\n  /**\n   * Keeps track of the number of data moves during a kernel execution. We\n   * maintain a stack since kernels can call other kernels, recursively.\n   */\n  numDataMovesStack: number[] = [];\n  nextScopeId = 0;\n\n  tensorInfo = new WeakMap<DataId, {\n    backend: KernelBackend,\n    bytes: number,\n    dtype: DataType,\n    shape: number[],\n    refCount: number\n  }>();\n\n  profiling = false;\n  activeProfile: ProfileInfo =\n      {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n  dispose() {\n    for (const variableName in this.registeredVariables) {\n      this.registeredVariables[variableName].dispose();\n    }\n  }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n  state: EngineState;\n  backendName: string;\n  registry: {[id: string]: KernelBackend} = {};\n  registryFactory: {\n    [id: string]: {\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority: number\n    }\n  } = {};\n\n  private profiler: Profiler;\n  private backendInstance: KernelBackend;\n  private pendingBackendInit: Promise<boolean>;\n  private pendingBackendInitId = 0;\n\n  constructor(public ENV: Environment) {\n    this.state = new EngineState();\n  }\n\n  async ready(): Promise<void> {\n    if (this.pendingBackendInit != null) {\n      return this.pendingBackendInit.then(() => {});\n    }\n    if (this.backendInstance != null) {\n      return;\n    }\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const success = await this.initializeBackend(backendName).success;\n      if (success) {\n        await this.setBackend(backendName);\n        return;\n      }\n    }\n\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  get backend(): KernelBackend {\n    if (this.pendingBackendInit != null) {\n      throw new Error(\n          `Backend '${this.backendName}' has not yet been initialized. Make ` +\n          `sure to await tf.ready() or await tf.setBackend() before calling ` +\n          `other methods`);\n    }\n    if (this.backendInstance == null) {\n      const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n      if (asyncInit) {\n        throw new Error(\n            `The highest priority backend '${name}' has not yet been ` +\n            `initialized. Make sure to await tf.ready() or ` +\n            `await tf.setBackend() before calling other methods`);\n      }\n      this.setBackend(name);\n    }\n    return this.backendInstance;\n  }\n\n  backendNames(): string[] {\n    return Object.keys(this.registryFactory);\n  }\n\n  findBackend(backendName: string): KernelBackend {\n    if (!(backendName in this.registry)) {\n      // If the backend hasn't been initialized but we have a registry entry for\n      // it, initialize it and return it.\n      if (backendName in this.registryFactory) {\n        const {asyncInit} = this.initializeBackend(backendName);\n        if (asyncInit) {\n          // Backend is not ready yet.\n          return null;\n        }\n      } else {\n        return null;\n      }\n    }\n    return this.registry[backendName];\n  }\n\n  findBackendFactory(backendName: string):\n      () => KernelBackend | Promise<KernelBackend> {\n    if (!(backendName in this.registryFactory)) {\n      return null;\n    }\n    return this.registryFactory[backendName].factory;\n  }\n\n  registerBackend(\n      backendName: string,\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority = 1): boolean {\n    if (backendName in this.registryFactory) {\n      console.warn(\n          `${backendName} backend was already registered. ` +\n          `Reusing existing backend factory.`);\n      return false;\n    }\n    this.registryFactory[backendName] = {factory, priority};\n    return true;\n  }\n\n  async setBackend(backendName: string): Promise<boolean> {\n    if (this.registryFactory[backendName] == null) {\n      throw new Error(`Backend name '${backendName}' not found in registry`);\n    }\n    this.backendName = backendName;\n    if (this.registry[backendName] == null) {\n      this.backendInstance = null;\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      const result = asyncInit ? await success : success;\n      if (!result) {\n        return false;\n      }\n    }\n    this.backendInstance = this.registry[backendName];\n    this.setupRegisteredKernels();\n    // Reset the profiler.\n    this.profiler = new Profiler(this.backendInstance);\n\n    return true;\n  }\n\n  private setupRegisteredKernels(): void {\n    const kernels = getKernelsForBackend(this.backendName);\n    kernels.forEach(kernel => {\n      if (kernel.setupFunc != null) {\n        kernel.setupFunc(this.backendInstance);\n      }\n    });\n  }\n\n  private disposeRegisteredKernels(backendName: string): void {\n    const kernels = getKernelsForBackend(backendName);\n    kernels.forEach(kernel => {\n      if (kernel.disposeFunc != null) {\n        kernel.disposeFunc(this.registry[backendName]);\n      }\n    });\n  }\n\n  /**\n   * Initializes a backend by looking up the backend name in the factory\n   * registry and calling the factory method. Returns a boolean representing\n   * whether the initialization of the backend suceeded. Throws an error if\n   * there is no backend in the factory registry.\n   */\n  private initializeBackend(backendName: string):\n      {success: boolean|Promise<boolean>, asyncInit: boolean} {\n    const registryFactoryEntry = this.registryFactory[backendName];\n    if (registryFactoryEntry == null) {\n      throw new Error(\n          `Cannot initialize backend ${backendName}, no registration found.`);\n    }\n\n    try {\n      const backend = registryFactoryEntry.factory();\n      /* Test if the factory returns a promise.\n      Done in a more liberal way than\n      previous 'Promise.resolve(backend)===backend'\n      as we needed to account for custom Promise\n      implementations (e.g. Angular) */\n      if (backend && !(backend instanceof KernelBackend)\n          && typeof backend.then === 'function') {\n        const promiseId = ++this.pendingBackendInitId;\n        const success =\n            backend\n                .then(backendInstance => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.registry[backendName] = backendInstance;\n                  this.pendingBackendInit = null;\n                  return true;\n                })\n                .catch(err => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.pendingBackendInit = null;\n                  console.warn(\n                      `Initialization of backend ${backendName} failed`);\n                  console.warn(err.stack || err.message);\n                  return false;\n                });\n        this.pendingBackendInit = success;\n        return {success, asyncInit: true};\n      } else {\n        this.registry[backendName] = backend as KernelBackend;\n        return {success: true, asyncInit: false};\n      }\n    } catch (err) {\n      console.warn(`Initialization of backend ${backendName} failed`);\n      console.warn(err.stack || err.message);\n      return {success: false, asyncInit: false};\n    }\n  }\n\n  removeBackend(backendName: string): void {\n    if (!(backendName in this.registryFactory)) {\n      throw new Error(`${backendName} backend not found in registry`);\n    }\n    if (this.backendName === backendName && this.pendingBackendInit != null) {\n      // There is a pending promise of the backend we want to remove. Make it\n      // obsolete.\n      this.pendingBackendInitId++;\n    }\n\n    if (backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n\n    delete this.registryFactory[backendName];\n\n    // Unset the backend if it is active.\n    if (this.backendName === backendName) {\n      this.pendingBackendInit = null;\n      this.backendName = null;\n      this.backendInstance = null;\n    }\n  }\n\n  private getSortedBackends(): string[] {\n    if (Object.keys(this.registryFactory).length === 0) {\n      throw new Error('No backend found in registry.');\n    }\n    return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n      // Highest priority comes first.\n      return this.registryFactory[b].priority -\n          this.registryFactory[a].priority;\n    });\n  }\n\n  private initializeBackendsAndReturnBest():\n      {name: string, asyncInit: boolean} {\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      if (asyncInit || success) {\n        return {name: backendName, asyncInit};\n      }\n    }\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  moveData(backend: KernelBackend, dataId: DataId) {\n    const info = this.state.tensorInfo.get(dataId);\n    const srcBackend = info.backend;\n    const values = this.readSync(dataId);\n    // Delete the tensor from the old backend and move it to the new\n    // backend.\n    srcBackend.disposeData(dataId);\n    info.backend = backend;\n    backend.move(dataId, values, info.shape, info.dtype);\n    if (this.shouldCheckForMemLeaks()) {\n      // Track the number of moves during a kernel execution to correctly\n      // detect memory leaks.\n      this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n    }\n  }\n\n  tidy<T extends TensorContainer>(nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>):\n      T {\n    let name: string = null;\n    if (fn == null) {\n      // Called with only 1 argument.\n      if (typeof nameOrFn !== 'function') {\n        throw new Error('Please provide a function to tidy()');\n      }\n      fn = nameOrFn;\n    } else {\n      // Called with 2 arguments.\n      if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n        throw new Error(\n            'When calling with two arguments, the first argument ' +\n            'to tidy() must be a string');\n      }\n      if (typeof fn !== 'function') {\n        throw new Error(\n            'When calling with two arguments, the 2nd argument ' +\n            'to tidy() must be a function');\n      }\n      name = nameOrFn as string;\n      // TODO(nsthorat,smilkov): Do operation logging and performance\n      // profiling.\n    }\n    let result: T;\n    return this.scopedRun(\n        () => this.startScope(name), () => this.endScope(result), () => {\n          result = fn();\n          if (result instanceof Promise) {\n            console.error('Cannot return a Promise inside of tidy.');\n          }\n          return result;\n        });\n  }\n\n  private scopedRun<T>(start: () => void, end: () => void, f: () => T): T {\n    start();\n    try {\n      const res = f();\n      end();\n      return res;\n    } catch (ex) {\n      end();\n      throw ex;\n    }\n  }\n\n  private static nextTensorId = 0;\n  private nextTensorId(): number {\n    return Engine.nextTensorId++;\n  }\n\n  private static nextVariableId = 0;\n  private nextVariableId(): number {\n    return Engine.nextVariableId++;\n  }\n\n  /**\n   * This method is called instead of the public-facing tensor.clone() when\n   * saving a tensor for backwards pass. It makes sure to add the clone\n   * operation to the tape regardless of being called inside a kernel\n   * execution.\n   *\n   * This method will go away once all kernels are modularized since we won't\n   * need to turn off the tape inside runKernel().\n   */\n  private clone(x: Tensor): Tensor {\n    const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n    const inputs = {x};\n    const grad = (dy: Tensor) => ({\n      x: () => {\n        const dtype = 'float32';\n        const gradInputs = {x: dy};\n        const attrs = {dtype};\n\n        return ENGINE.runKernelFunc(\n            backend => backend.cast(dy, dtype),\n            gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n            attrs as {} as NamedAttrMap);\n      }\n    });\n    const saved: Tensor[] = [];\n    this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n    return y;\n  }\n\n  /**\n   * Execute a kernel with the given name and return the output tensor.\n   *\n   * @param kernelName The name of the kernel to execute.\n   * @param inputs A map of input names to tensors.\n   * @param attrs A map of attribute names to their values. An attribute is a\n   *     primitive (non-tensor) input to the kernel.\n   * @param inputsToSave A list of tensors, inputs to save for the backprop\n   *     computation.\n   * @param outputsToSave A list of booleans, specifying which output to save\n   *     for the backprop computation. These are booleans since the output\n   * tensors are not visible to the user.\n   */\n  runKernel(\n      kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n      inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n    const forwardFunc: null = null;\n    const backwardsFunc: null = null;\n    // Call runKernel as a stop-gap until we modularize all kernels.\n    // Once we modularize all kernels, we will remove the existing\n    // `runKernelFunc`.\n    return this.runKernelFunc(\n        forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n        outputsToSave);\n  }\n\n  private shouldCheckForMemLeaks(): boolean {\n    return this.ENV.getBool('IS_TEST');\n  }\n\n  private checkKernelForMemLeak(\n      kernelName: string, numDataIdsBefore: number,\n      outInfos: TensorInfo[]): void {\n    const numDataIdsAfter = this.backend.numDataIds();\n\n    // Count the number of data ids associated with the result of the kernel.\n    let numOutputDataIds = 0;\n    outInfos.forEach(info => {\n      // Complex numbers allocate 3 data ids, one for 'real', one for\n      // 'imaginary', and one for the container that holds the former two.\n      numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n    });\n\n    // Account for the number of moves during kernel execution. A \"data move\"\n    // can happen in the middle of a kernel execution, placing a new (key,value)\n    // pair in the data storage. Since data moves have net zero effect (we\n    // always remove the data from the old backend), we have to cancel them out\n    // when detecting memory leaks.\n    const numMoves =\n        this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n    const dataIdsLeaked =\n        numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n    if (dataIdsLeaked > 0) {\n      throw new Error(\n          `Backend '${this.backendName}' has an internal memory leak ` +\n          `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n    }\n  }\n\n  /**\n   * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n   *     only for kernels that are not yet fully modularized.\n   */\n  runKernelFunc<T extends Tensor|Tensor[], I extends NamedTensorMap>(\n      forwardFunc: ForwardFunc<T>, inputs: I,\n      backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n      kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n      outputsToSave?: boolean[]): T {\n    let outputs: Tensor[];\n    let saved: Tensor[] = [];\n    const isTapeOn = this.isTapeOn();\n    if (kernelName == null) {\n      kernelName =\n          this.state.activeScope != null ? this.state.activeScope.name : '';\n    }\n\n    const startingBytecount = this.state.numBytes;\n    const startingNumTensors = this.state.numTensors;\n\n    if (this.shouldCheckForMemLeaks()) {\n      this.state.numDataMovesStack.push(0);\n    }\n\n    let kernelFunc: () => Tensor[];\n    const kernel = getKernel(kernelName, this.backendName);\n    let out: TensorInfo|TensorInfo[];\n    if (kernel != null) {\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n        const outInfos = Array.isArray(out) ? out : [out];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n        }\n        const outTensors = outInfos.map(\n            ({dataId, shape, dtype}) =>\n                this.makeTensorFromDataId(dataId, shape, dtype));\n\n        // Save the inputs and outputs.\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (isTapeOn) {\n          let tensorsToSave =\n              this.getTensorsForGradient(kernelName, inputs, outTensors);\n          if (tensorsToSave == null) {\n            // Fallback for ops that call runKernelFunc and pass in\n            // inputsToSave and outputsToSave. Currently this is the set of ops\n            // with kernel support in the WASM backend. Once those ops and\n            // respective gradients are modularised we can remove this path.\n            if (outputsToSave == null) {\n              outputsToSave = [];\n            }\n            const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n            tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n          }\n          saved = this.saveTensorsForBackwardMode(tensorsToSave);\n        }\n        return outTensors;\n      };\n    } else {\n      const saveFunc: GradSaveFunc = (tensors) => {\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (!isTapeOn) {\n          return;\n        }\n        saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n      };\n\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n        const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n        }\n        return outs;\n      };\n    }\n\n    // Stop recording to a tape when running a kernel.\n    let kernelProfile: KernelProfile;\n    this.scopedRun(\n        () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n          if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n            outputs = kernelFunc();\n          } else {\n            kernelProfile = this.profiler.profileKernel(\n                kernelName, inputs, () => kernelFunc());\n            if (this.ENV.getBool('DEBUG')) {\n              this.profiler.logKernelProfile(kernelProfile);\n            }\n            outputs = kernelProfile.outputs;\n          }\n        });\n\n    if (isTapeOn) {\n      this.addTapeNode(\n          kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n    }\n\n    if (this.state.profiling) {\n      this.state.activeProfile.kernels.push({\n        name: kernelName,\n        bytesAdded: this.state.numBytes - startingBytecount,\n        totalBytesSnapshot: this.state.numBytes,\n        tensorsAdded: this.state.numTensors - startingNumTensors,\n        totalTensorsSnapshot: this.state.numTensors,\n        inputShapes: Object.keys(inputs).map(\n            key => inputs[key] != null ? inputs[key].shape : null),\n        outputShapes: outputs.map(item => item.shape),\n        kernelTimeMs: kernelProfile.timeMs,\n        extraInfo: kernelProfile.extraInfo\n      });\n    }\n    return (Array.isArray(out) ? outputs : outputs[0]) as T;\n  }\n\n  /**\n   * Saves tensors used in forward mode for use in backward mode.\n   *\n   * @param tensors the list of tensors to save.\n   */\n  private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n    const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n    return saved;\n  }\n\n  /**\n   * Returns a list of tensors to save for a given gradient calculation.\n   *\n   * Returns undefined if their is no registered gradient for this kernel in the\n   * gradient registry.\n   *\n   * @param kernelName name of kernel to look up gradient for.\n   * @param inputs a map of input tensors.\n   * @param outputs an array of output tensors from forward mode of kernel.\n   */\n  private getTensorsForGradient(\n      kernelName: string, inputs: NamedTensorMap,\n      outputs: Tensor[]): Tensor[]|null {\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      const inputsToSave: string[] = gradConfig.inputsToSave || [];\n      const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n      // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n      // specified in inputsToSave will be saved.\n      let inputTensorsToSave: Tensor[];\n      if (gradConfig.saveAllInputs) {\n        util.assert(\n            Array.isArray(inputs),\n            () => 'saveAllInputs is true, expected inputs to be an array.');\n\n        inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n      } else {\n        inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n      }\n\n      const outputTensorsToSave: Tensor[] =\n          outputs.filter((_, i) => outputsToSave[i]);\n\n      return inputTensorsToSave.concat(outputTensorsToSave);\n    }\n    // TODO(yassogba) throw exception here once all runkernelFunc calls with\n    // inputsToSave/outputsToSave are removed\n    return null;\n  }\n\n  /**\n   * Internal method used by public APIs for tensor creation. Makes a new\n   * tensor with the provided shape, dtype and values. It always\n   * creates a new data id and writes the values to the underlying backend.\n   */\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    if (values == null) {\n      throw new Error('Values passed to engine.makeTensor() are null');\n    }\n    dtype = dtype || 'float32';\n    backend = backend || this.backend;\n    let backendVals = values as BackendValues;\n    if (dtype === 'string' && util.isString(values[0])) {\n      backendVals = (values as string[]).map(d => util.encodeString(d));\n    }\n    const dataId = backend.write(backendVals, shape, dtype);\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n\n    // Count bytes for string tensors.\n    if (dtype === 'string') {\n      const info = this.state.tensorInfo.get(dataId);\n      const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n      this.state.numBytes += newBytes - info.bytes;\n      info.bytes = newBytes;\n    }\n    return t;\n  }\n\n  /**\n   * Internal method used by backends. Makes a new tensor\n   * that is a wrapper around an existing data id. It doesn't create\n   * a new data id, only increments the ref count used in memory tracking.\n   */\n  makeTensorFromDataId(\n      dataId: DataId, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    dtype = dtype || 'float32';\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n    return t;\n  }\n\n  makeVariable(\n      initialValue: Tensor, trainable = true, name?: string,\n      dtype?: DataType): Variable {\n    name = name || this.nextVariableId().toString();\n    if (dtype != null && dtype !== initialValue.dtype) {\n      initialValue = initialValue.cast(dtype);\n    }\n    const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n    if (this.state.registeredVariables[v.name] != null) {\n      throw new Error(`Variable with name ${v.name} was already registered`);\n    }\n    this.state.registeredVariables[v.name] = v;\n    this.incRef(v, this.backend);\n    return v;\n  }\n\n  incRef(a: Tensor, backend: KernelBackend): void {\n    const refCount = this.state.tensorInfo.has(a.dataId) ?\n        this.state.tensorInfo.get(a.dataId).refCount :\n        0;\n    this.state.numTensors++;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors++;\n    }\n    if (refCount === 0) {\n      this.state.numDataBuffers++;\n\n      // Bytes for complex numbers are counted by their components. Bytes for\n      // string tensors are counted when writing values.\n      let bytes = 0;\n      if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n        bytes = a.size * util.bytesPerElement(a.dtype);\n      }\n      this.state.tensorInfo.set(a.dataId, {\n        backend: backend || this.backend,\n        dtype: a.dtype,\n        shape: a.shape,\n        bytes,\n        refCount: 0\n      });\n      this.state.numBytes += bytes;\n    }\n\n    this.state.tensorInfo.get(a.dataId).refCount++;\n\n    if (!(a instanceof Variable)) {\n      this.track(a);\n    }\n  }\n\n  disposeTensor(a: Tensor): void {\n    if (!this.state.tensorInfo.has(a.dataId)) {\n      return;\n    }\n\n    this.state.numTensors--;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors--;\n    }\n    const info = this.state.tensorInfo.get(a.dataId);\n    const refCount = info.refCount;\n\n    if (refCount <= 1) {\n      // Don't count bytes for complex numbers as they are counted by their\n      // components.\n      if (a.dtype !== 'complex64') {\n        this.state.numBytes -= info.bytes;\n      }\n      this.state.numDataBuffers--;\n\n      info.backend.disposeData(a.dataId);\n      this.state.tensorInfo.delete(a.dataId);\n    } else {\n      this.state.tensorInfo.get(a.dataId).refCount--;\n    }\n    // TODO(nsthorat): Construct an error and save the stack trace for\n    // debugging when in debug mode. Creating a stack trace is too expensive\n    // to do unconditionally.\n  }\n\n  disposeVariables(): void {\n    for (const varName in this.state.registeredVariables) {\n      const v = this.state.registeredVariables[varName];\n      this.disposeVariable(v);\n    }\n  }\n\n  disposeVariable(v: Variable): void {\n    this.disposeTensor(v);\n    if (this.state.registeredVariables[v.name] != null) {\n      delete this.state.registeredVariables[v.name];\n    }\n  }\n\n  memory(): MemoryInfo {\n    const info = this.backend.memory() as MemoryInfo;\n    info.numTensors = this.state.numTensors;\n    info.numDataBuffers = this.state.numDataBuffers;\n    info.numBytes = this.state.numBytes;\n    if (this.state.numStringTensors > 0) {\n      info.unreliable = true;\n      if (info.reasons == null) {\n        info.reasons = [];\n      }\n      info.reasons.push(\n          'Memory usage by string tensors is approximate ' +\n          '(2 bytes per character)');\n    }\n    return info;\n  }\n\n  async profile(query: () => (TensorContainer | Promise<TensorContainer>)):\n      Promise<ProfileInfo> {\n    this.state.profiling = true;\n\n    const startBytes = this.state.numBytes;\n    const startNumTensors = this.state.numTensors;\n\n    this.state.activeProfile.kernels = [];\n    this.state.activeProfile.result = await query();\n\n    this.state.profiling = false;\n\n    this.state.activeProfile.peakBytes = Math.max(\n        ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n    this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n    this.state.activeProfile.newTensors =\n        this.state.numTensors - startNumTensors;\n    for (const kernel of this.state.activeProfile.kernels) {\n      kernel.kernelTimeMs = await kernel.kernelTimeMs;\n      kernel.extraInfo = await kernel.extraInfo;\n    }\n    return this.state.activeProfile;\n  }\n\n  isTapeOn(): boolean {\n    return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n  }\n\n  private addTapeNode(\n      kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n      gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n    const tapeNode: TapeNode =\n        {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      gradientsFunc = gradConfig.gradFunc;\n    }\n    if (gradientsFunc != null) {\n      tapeNode.gradient = (dys: Tensor[]) => {\n        // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n        // the backprop graph to the user as null instead of zeros\n        dys = dys.map((dy, i) => {\n          if (dy == null) {\n            const output = outputs[i];\n            const vals = util.makeZerosTypedArray(output.size, output.dtype);\n            return this.makeTensor(vals, output.shape, output.dtype);\n          }\n          return dy;\n        });\n        // Grad functions of ops with single outputs expect a dy, while ops\n        // with multiple outputs expect dys (array of dy).\n        return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n      };\n    }\n    this.state.activeTape.push(tapeNode);\n  }\n\n  keep<T extends Tensor>(result: T): T {\n    result.kept = true;\n    return result;\n  }\n\n  private startTape() {\n    if (this.state.gradientDepth === 0) {\n      this.state.activeTape = [];\n    }\n    this.state.gradientDepth++;\n  }\n\n  private endTape() {\n    this.state.gradientDepth--;\n  }\n\n  /**\n   * Start a scope. Use this with endScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  startScope(name?: string) {\n    const scopeInfo: ScopeState = {\n      track: [],\n      name: 'unnamed scope',\n      id: this.state.nextScopeId++\n    };\n    if (name) {\n      scopeInfo.name = name;\n    }\n    this.state.scopeStack.push(scopeInfo);\n    this.state.activeScope = scopeInfo;\n  }\n\n  /**\n   * End a scope. Use this with startScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  endScope(result?: TensorContainer) {\n    const tensorsToTrackInParent = getTensorsInContainer(result);\n    const tensorsToTrackInParentSet =\n        new Set(tensorsToTrackInParent.map(t => t.id));\n\n    // Dispose the arrays tracked in this scope.\n    for (let i = 0; i < this.state.activeScope.track.length; i++) {\n      const tensor = this.state.activeScope.track[i];\n      if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n        tensor.dispose();\n      }\n    }\n\n    const oldScope = this.state.scopeStack.pop();\n    this.state.activeScope = this.state.scopeStack.length === 0 ?\n        null :\n        this.state.scopeStack[this.state.scopeStack.length - 1];\n\n    // Track the current result in the parent scope.\n    tensorsToTrackInParent.forEach(tensor => {\n      // Only track the tensor if was allocated in the inner scope and is not\n      // globally kept.\n      if (!tensor.kept && tensor.scopeId === oldScope.id) {\n        this.track(tensor);\n      }\n    });\n  }\n\n  /**\n   * Returns gradients of `f` with respect to each of the `xs`. The gradients\n   * returned are of the same length as `xs`, but some might be null if `f`\n   * was not a function of that `x`. It also takes optional dy to multiply the\n   * gradient, which defaults to `1`.\n   */\n  gradients<T extends Tensor>(\n      f: () => T, xs: Tensor[], dy?: T,\n      allowNoGradients = false): {value: T, grads: Tensor[]} {\n    util.assert(\n        xs.length > 0, () => 'gradients() received an empty list of xs.');\n    if (dy != null && dy.dtype !== 'float32') {\n      throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n    }\n\n    const y = this.scopedRun(\n        () => this.startTape(), () => this.endTape(),\n        () => this.tidy('forward', f));\n\n    util.assert(\n        y instanceof Tensor,\n        () => 'The result y returned by f() must be a tensor.');\n    // Filter out the nodes that don't connect x => y.\n    const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n    if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n      throw new Error(\n          'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n          'that the f you passed encloses all operations that lead from x ' +\n          'to y.');\n    }\n\n    return this.tidy('backward', () => {\n      const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n      accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n      // Backprop gradients through the filtered nodes.\n      backpropagateGradients(\n          accumulatedGradientMap, filteredTape,\n          // Pass the tidy function to avoid circular dep with `tape.ts`.\n          f => this.tidy(f as ScopeFn<Tensor>),\n          // Pass an add function to avoide a circular dep with `tape.ts`.\n          add);\n      const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n      if (this.state.gradientDepth === 0) {\n        // This means that we are not computing higher-order gradients\n        // and can clean up the tape.\n        this.state.activeTape.forEach(node => {\n          for (const tensor of node.saved) {\n            tensor.dispose();\n          }\n        });\n        this.state.activeTape = null;\n      }\n      return {value: y, grads};\n    });\n  }\n\n  customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n      (...args: Array<Tensor|GradSaveFunc>) => T {\n    util.assert(\n        util.isFunction(f),\n        () => 'The f passed in customGrad(f) must be a function.');\n    return (...inputs: Tensor[]): T => {\n      util.assert(\n          inputs.every(t => t instanceof Tensor),\n          () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n              'tensors');\n\n      let res: {\n        value: T,\n        gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n      };\n      const inputMap: NamedTensorMap = {};\n      inputs.forEach((input, i) => {\n        inputMap[i] = input;\n      });\n      return this.runKernelFunc(\n          (_, save) => {\n            res = f(...[...inputs, save]);\n            util.assert(\n                res.value instanceof Tensor,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.value` is a tensor');\n            util.assert(\n                util.isFunction(res.gradFunc),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function.');\n            return res.value;\n          },\n          inputMap,\n          (dy: T, saved: Tensor[]) => {\n            const gradRes = res.gradFunc(dy, saved);\n            const grads: Tensor[] =\n                Array.isArray(gradRes) ? gradRes : [gradRes];\n            util.assert(\n                grads.length === inputs.length,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'the same number of tensors as inputs passed to f(...).');\n            util.assert(\n                grads.every(t => t instanceof Tensor),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'a list of only tensors.');\n            const gradMap: {[key: string]: () => Tensor} = {};\n            grads.forEach((grad, i) => {\n              gradMap[i] = () => grad;\n            });\n            return gradMap;\n          });\n    };\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.readSync(dataId);\n  }\n  read(dataId: DataId): Promise<BackendValues> {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.read(dataId);\n  }\n\n  async time(query: () => void): Promise<TimingInfo> {\n    const start = now();\n    const timingInfo = await this.backend.time(query) as TimingInfo;\n    timingInfo.wallMs = now() - start;\n    return timingInfo;\n  }\n\n  /**\n   * Tracks a Tensor in the current scope to be automatically cleaned up\n   * when the current scope ends, and returns the value.\n   *\n   * @param result The Tensor to track in the current scope.\n   */\n  private track<T extends Tensor>(result: T): T {\n    if (this.state.activeScope != null) {\n      result.scopeId = this.state.activeScope.id;\n      this.state.activeScope.track.push(result);\n    }\n\n    return result;\n  }\n\n  get registeredVariables(): NamedVariableMap {\n    return this.state.registeredVariables;\n  }\n\n  /**\n   * Resets the engine state. Removes all backends but does not remove\n   * registered backend factories.\n   */\n  reset(): void {\n    // Make any pending promise obsolete.\n    this.pendingBackendInitId++;\n\n    this.state.dispose();\n    this.ENV.reset();\n    this.state = new EngineState();\n\n    for (const backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n    this.backendName = null;\n    this.backendInstance = null;\n    this.pendingBackendInit = null;\n  }\n}\n\nfunction ones(shape: number[]): Tensor {\n  const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n  return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n  const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n  if (ns._tfengine == null) {\n    const environment = new Environment(ns);\n    ns._tfengine = new Engine(environment);\n  }\n  setEnvironmentGlobal(ns._tfengine.ENV);\n\n  // Tell the current tensor interface that the global engine is responsible\n  // for tracking.\n  setTensorTracker(() => ns._tfengine);\n  return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n  // We duplicate Add here to avoid a circular dependency with add.ts.\n  const inputs = {a, b};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.add(a, b);\n    save([a, b]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n  return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n  if (_isNavigatorDefined()) {\n    // tslint:disable-next-line:no-any\n    const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n    // tslint:disable-next-line:max-line-length\n    return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n               .test(a) ||\n        // tslint:disable-next-line:max-line-length\n        /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n            .test(a.substr(0, 4));\n  }\n  return false;\n}\n\nexport function isBrowser(): boolean {\n  return (typeof window !== 'undefined' && window.document != null) ||\n      //@ts-ignore\n      (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n  if (debugValue) {\n    console.warn(\n        'Debugging mode is ON. The output of every math call will ' +\n        'be downloaded to CPU and checked for NaNs. ' +\n        'This significantly impacts performance.');\n  }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n    'IS_NODE',\n    () => (typeof process !== 'undefined') &&\n        (typeof process.versions !== 'undefined') &&\n        (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n    'IS_CHROME',\n    () => typeof navigator !== 'undefined' && navigator != null &&\n        navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n        /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n    'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n  let firstElem: typeof val = val;\n\n  if (isTypedArray(val)) {\n    return dtype === 'string' ? [] : [val.length];\n  }\n  if (!Array.isArray(val)) {\n    return [];  // Scalar.\n  }\n  const shape: number[] = [];\n\n  while (Array.isArray(firstElem) ||\n         isTypedArray(firstElem) && dtype !== 'string') {\n    shape.push(firstElem.length);\n    firstElem = firstElem[0];\n  }\n  if (Array.isArray(val) &&\n      env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n    deepAssertShapeConsistency(val, shape, []);\n  }\n\n  return shape;\n}\n\nfunction deepAssertShapeConsistency(\n    val: TensorLike, shape: number[], indices: number[]) {\n  indices = indices || [];\n  if (!(Array.isArray(val)) && !isTypedArray(val)) {\n    assert(\n        shape.length === 0,\n        () => `Element arr[${indices.join('][')}] is a primitive, ` +\n            `but should be an array/TypedArray of ${shape[0]} elements`);\n    return;\n  }\n  assert(\n      shape.length > 0,\n      () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n          `but is an array of ${val.length} elements`);\n  assert(\n      val.length === shape[0],\n      () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n          `elements, but has ${val.length} elements`);\n  const subShape = shape.slice(1);\n  for (let i = 0; i < val.length; ++i) {\n    deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n  }\n}\n\nfunction assertDtype(\n    expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n    functionName: string) {\n  if (expectedDtype == null) {\n    return;\n  }\n  if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n      expectedDtype === 'numeric' && actualDType === 'string') {\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must ` +\n        `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n  }\n}\n\nexport function convertToTensor<T extends Tensor>(\n    x: T|TensorLike, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T {\n  if (x instanceof Tensor) {\n    assertDtype(parseAsDtype, x.dtype, argName, functionName);\n    return x;\n  }\n  let inferredDtype = inferDtype(x);\n  // If the user expects a bool/int/float, use that info to update the\n  // inferredDtype when it is not a string.\n  if (inferredDtype !== 'string' &&\n      ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n    inferredDtype = parseAsDtype as DataType;\n  }\n  assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n  if ((x == null) ||\n      (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n       typeof x !== 'boolean' && typeof x !== 'string')) {\n    const type = x == null ? 'null' : (x as {}).constructor.name;\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must be a ` +\n        `Tensor or TensorLike, but got '${type}'`);\n  }\n  const inferredShape = inferShape(x, inferredDtype);\n  if (!isTypedArray(x) && !Array.isArray(x)) {\n    x = [x] as number[];\n  }\n  const skipTypedArray = true;\n  const values = inferredDtype !== 'string' ?\n      toTypedArray(x, inferredDtype as DataType) :\n      flatten(x as string[], [], skipTypedArray) as string[];\n  return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray<T extends Tensor>(\n    arg: Array<T|TensorLike>, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n  if (!Array.isArray(arg)) {\n    throw new Error(\n        `Argument ${argName} passed to ${functionName} must be a ` +\n        '`Tensor[]` or `TensorLike[]`');\n  }\n  const tensors = arg as T[];\n  return tensors.map(\n      (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n      parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op<T extends Function>(f: {[name: string]: T}): T {\n  const keys = Object.keys(f);\n  if (keys.length !== 1) {\n    throw new Error(\n        `Please provide an object with a single key ` +\n        `(operation name) mapping to a function. Got an object with ` +\n        `${keys.length} keys.`);\n  }\n\n  let opName = keys[0];\n  const fn = f[opName];\n\n  // Strip the underscore from the end of the function name.\n  if (opName.endsWith('_')) {\n    opName = opName.substring(0, opName.length - 1);\n  }\n\n  // add an __op suffix to distinguish ops from kernels in tf.profile\n  opName = opName + OP_SCOPE_SUFFIX;\n\n  // tslint:disable-next-line:no-any\n  const f2 = (...args: any[]) => {\n    ENGINE.startScope(opName);\n    try {\n      const result = fn(...args);\n      if (result instanceof Promise) {\n        console.error('Cannot return a Promise inside of tidy.');\n      }\n      ENGINE.endScope(result);\n      return result;\n    } catch (ex) {\n      ENGINE.endScope(null);\n      throw ex;\n    }\n  };\n  Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n  // tslint:disable-next-line:no-any\n  return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_<T extends Tensor>(real: T|TensorLike, imag: T|TensorLike): T {\n  const $real = convertToTensor(real, 'real', 'complex');\n  const $imag = convertToTensor(imag, 'imag', 'complex');\n  util.assertShapesMatch(\n      $real.shape, $imag.shape,\n      `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n          `must match in call to tf.complex().`);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.complex($real, $imag);\n  };\n  const inputs: ComplexInputs = {real: $real, imag: $imag};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n    values: TensorLike, shape: number[], inferredShape: number[],\n    dtype?: DataType): Tensor {\n  if (dtype == null) {\n    dtype = inferDtype(values);\n  }\n  if (dtype === 'complex64') {\n    throw new Error(\n        `Cannot construct a complex64 tensor directly. ` +\n        `Please use tf.complex(real, imag).`);\n  }\n  if (!isTypedArray(values) && !Array.isArray(values) &&\n      typeof values !== 'number' && typeof values !== 'boolean' &&\n      typeof values !== 'string') {\n    throw new Error(\n        'values passed to tensor(values) must be a number/boolean/string or ' +\n        'an array of numbers/booleans/strings, or a TypedArray');\n  }\n  if (shape != null) {\n    assertNonNegativeIntegerDimensions(shape);\n\n    const providedSize = sizeFromShape(shape);\n    const inferredSize = sizeFromShape(inferredShape);\n    assert(\n        providedSize === inferredSize,\n        () =>\n            `Based on the provided shape, [${shape}], the tensor should have ` +\n            `${providedSize} values but has ${inferredSize}`);\n\n    for (let i = 0; i < inferredShape.length; ++i) {\n      const inferred = inferredShape[i];\n      const flatDimsDontMatch = i === inferredShape.length - 1 ?\n          inferred !== sizeFromShape(shape.slice(i)) :\n          true;\n      assert(\n          inferredShape[i] === shape[i] || !flatDimsDontMatch,\n          () => `Error creating a new Tensor. Inferred shape ` +\n              `(${inferredShape}) does not match the provided ` +\n              `shape (${shape}). `);\n    }\n  }\n\n  if (!isTypedArray(values) && !Array.isArray(values)) {\n    values = [values] as number[];\n  }\n\n  shape = shape || inferredShape;\n  values = dtype !== 'string' ?\n      toTypedArray(values, dtype) :\n      flatten(values as string[], [], true) as string[];\n  return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`. If the values are strings,\n *     they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor<R extends Rank>(\n    values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor<R> {\n  const inferredShape = inferShape(values, dtype);\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n  'float32': 4,\n  'float16': 2,\n  'int32': 4,\n  'uint16': 2,\n  'uint8': 1,\n  'bool': 1,\n  'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n  /**\n   * An ordered list of paths.\n   *\n   * Paths are intentionally abstract in order to be general. For example, they\n   * can be relative URL paths or relative paths on the file system.\n   */\n  paths: string[];\n\n  /**\n   * Specifications of the weights stored in the paths.\n   */\n  weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n  /**\n   * Name of the weight, e.g., 'Dense_1/bias'\n   */\n  name: string;\n\n  /**\n   * Shape of the weight.\n   */\n  shape: number[];\n\n  /**\n   * Data type of the weight.\n   */\n  dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n  /**\n   * Type of the weight.\n   *\n   * Optional.\n   *\n   * The value 'optimizer' indicates the weight belongs to an optimizer\n   * (i.e., used only during model training and not during inference).\n   */\n  group?: WeightGroup;\n\n  /**\n   * Information for dequantization of the weight.\n   */\n  quantization?: {\n    scale?: number,  // The scaling constant to multiply by.\n    min?: number,    // The (possibly nudged) minimum weight to add.\n       dtype: 'uint16'|'uint8'|'float16'  // The dtype of the quantized weights.\n  };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n  /**\n   * Whether to save only the trainable weights of the model, ignoring the\n   * non-trainable ones.\n   */\n  trainableOnly?: boolean;\n\n  /**\n   * Whether the optimizer will be saved (if exists).\n   *\n   * Default: `false`.\n   */\n  includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n  /**\n   * Information about the model artifacts saved.\n   */\n  modelArtifactsInfo: ModelArtifactsInfo;\n\n  /**\n   * HTTP responses from the server that handled the model-saving request (if\n   * any). This is applicable only to server-based saving routes.\n   */\n  responses?: Response[];\n\n  /**\n   * Error messages and related data (if any).\n   */\n  errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n  /**\n   * Timestamp for when the model is saved.\n   */\n  dateSaved: Date;\n\n  /**\n   * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n   * come in a JSON format and none of our IOHandlers support a non json\n   * format. We could conder replacing this with 'Binary' if we want to\n   * allow future handlers to save to non json formats (though they will\n   * probably want more information than 'Binary').\n   * Type of the model topology\n   *\n   * Type of the model topology\n   *\n   * Possible values:\n   *   - JSON: JSON config (human-readable, e.g., Keras JSON).\n   *   - GraphDef: TensorFlow\n   *     [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n   *     protocol buffer (binary).\n   */\n  modelTopologyType: 'JSON'|'GraphDef';\n\n  /**\n   * Size of model topology (Keras JSON or GraphDef), in bytes.\n   */\n  modelTopologyBytes?: number;\n\n  /**\n   * Size of weight specification or manifest, in bytes.\n   */\n  weightSpecsBytes?: number;\n\n  /**\n   * Size of weight value data, in bytes.\n   */\n  weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  // See\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n  /** Optimizer used for the model training. */\n  optimizer_config: {};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Loss function(s) for the model's output(s). */\n  loss: string|string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Metric function(s) for the model's output(s). */\n  metrics?: string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  weighted_metrics?: string[];\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  sample_weight_mode?: string;\n\n  loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers.  For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology?: {}|ArrayBuffer;\n\n  /**\n   * Serialized configuration for the model's training.\n   */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weight specifications.\n   *\n   * This corresponds to the weightsData below.\n   */\n  weightSpecs?: WeightsManifestEntry[];\n\n  /**\n   * Binary buffer for all weight values concatenated in the order specified\n   * by `weightSpecs`.\n   */\n  weightData?: ArrayBuffer;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology: {};\n\n  /** Model training configuration. */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weights manifest.\n   *\n   * The weights manifest consists of an ordered list of weight-manifest\n   * groups. Each weight-manifest group consists of a number of weight values\n   * stored in a number of paths. See the documentation of\n   * `WeightsManifestConfig` for more details.\n   */\n  weightsManifest: WeightsManifestConfig;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise<ModelArtifacts>;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n    Promise<SaveResult>;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n  save?: SaveHandler;\n  load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n  /**\n   * List all models in the model store.\n   *\n   * @returns A dictionary mapping paths of existing models to their\n   *   model artifacts info. Model artifacts info include type of the model's\n   *   topology, byte sizes of the topology, weights, etc.\n   */\n  listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n  /**\n   * Remove a model specified by `path`.\n   *\n   * @param path\n   * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n   *   is successful).\n   * @throws Error if deletion fails, e.g., if no model exists at `path`.\n   */\n  removeModel(path: string): Promise<ModelArtifactsInfo>;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n  /**\n   * RequestInit (options) for HTTP requests.\n   *\n   * For detailed information on the supported fields, see\n   * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n   *     https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n   */\n  requestInit?: RequestInit;\n\n  /**\n   * Progress callback.\n   */\n  onProgress?: OnProgressCallback;\n\n  /**\n   * A function used to override the `window.fetch` function.\n   */\n  fetchFunc?: Function;\n\n  /**\n   * Strict loading model: whether extraneous weights or missing\n   * weights should trigger an `Error`.\n   *\n   * If `true`, require that the provided weights exactly match those\n   * required by the layers. `false` means that both extra weights\n   * and missing weights will be silently ignored.\n   *\n   * Default: `true`.\n   */\n  strict?: boolean;\n\n  /**\n   * Path prefix for weight files, by default this is calculated from the\n   * path of the model JSON file.\n   *\n   * For instance, if the path to the model JSON file is\n   * `http://localhost/foo/model.json`, then the default path prefix will be\n   * `http://localhost/foo/`. If a weight file has the path value\n   * `group1-shard1of2` in the weight manifest, then the weight file will be\n   * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n   * if you provide a `weightPathPrefix` value of\n   * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n   * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n   */\n  weightPathPrefix?: string;\n\n  /**\n   * Whether the module or model is to be loaded from TF Hub.\n   *\n   * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n   * standard model file name and the query parameters.\n   *\n   * Default: `false`.\n   */\n  fromTFHub?: boolean;\n\n  /**\n   * An async function to convert weight file name to URL. The weight file\n   * names are stored in model.json's weightsManifest.paths field. By default we\n   * consider weight files are colocated with the model.json file. For example:\n   *     model.json URL: https://www.google.com/models/1/model.json\n   *     group1-shard1of1.bin url:\n   *        https://www.google.com/models/1/group1-shard1of1.bin\n   *\n   * With this func you can convert the weight file name to any URL.\n   */\n  weightUrlConverter?: (weightFileName: string) => Promise<string>;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n  /**\n   * Is this request for a binary file (as opposed to a json file)\n   */\n  isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n *     concatenated.\n *   - An `Array` of `WeightManifestEntry`s, carrying information including\n *     tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n    tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n    Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n  // TODO(adarob, cais): Support quantization.\n  const specs: WeightsManifestEntry[] = [];\n  const dataPromises: Array<Promise<TypedArray>> = [];\n\n  const names: string[] = Array.isArray(tensors) ?\n      tensors.map(tensor => tensor.name) :\n      Object.keys(tensors);\n\n  for (let i = 0; i < names.length; ++i) {\n    const name = names[i];\n    const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n    if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n        t.dtype !== 'string' && t.dtype !== 'complex64') {\n      throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n    }\n    const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n    if (t.dtype === 'string') {\n      const utf8bytes = new Promise<TypedArray>(async resolve => {\n        const vals = await t.bytes() as Uint8Array[];\n        const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n            NUM_BYTES_STRING_LENGTH * vals.length;\n        const bytes = new Uint8Array(totalNumBytes);\n        let offset = 0;\n        for (let i = 0; i < vals.length; i++) {\n          const val = vals[i];\n          const bytesOfLength =\n              new Uint8Array(new Uint32Array([val.length]).buffer);\n          bytes.set(bytesOfLength, offset);\n          offset += NUM_BYTES_STRING_LENGTH;\n          bytes.set(val, offset);\n          offset += val.length;\n        }\n        resolve(bytes);\n      });\n      dataPromises.push(utf8bytes);\n    } else {\n      dataPromises.push(t.data());\n    }\n    if (group != null) {\n      spec.group = group;\n    }\n    specs.push(spec);\n  }\n\n  const tensorValues = await Promise.all(dataPromises);\n  return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n *   concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n *   whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n *   to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n    buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n  // TODO(adarob, cais): Support quantization.\n  const out: NamedTensorMap = {};\n  let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n  let offset = 0;\n  for (const spec of specs) {\n    const name = spec.name;\n    const dtype = spec.dtype;\n    const shape = spec.shape;\n    const size = sizeFromShape(shape);\n    let values: TypedArray|string[]|Uint8Array[];\n\n    if ('quantization' in spec) {\n      const quantization = spec.quantization;\n      if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n        if (!('min' in quantization && 'scale' in quantization)) {\n          throw new Error(\n              `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n              `doesn't have corresponding metadata min and scale.`);\n        }\n      } else if (quantization.dtype === 'float16') {\n        if (dtype !== 'float32') {\n          throw new Error(\n              `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n              `which only supports weights of type float32 not ${dtype}.`);\n        }\n      } else {\n        throw new Error(\n            `Weight ${spec.name} has unknown ` +\n            `quantization dtype ${quantization.dtype}. ` +\n            `Supported quantization dtypes are: ` +\n            `'uint8', 'uint16', and 'float16'.`);\n      }\n      const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n      const byteBuffer =\n          buffer.slice(offset, offset + size * quantizationSizeFactor);\n      const quantizedArray = (quantization.dtype === 'uint8') ?\n          new Uint8Array(byteBuffer) :\n          new Uint16Array(byteBuffer);\n      if (dtype === 'float32') {\n        if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n          values = new Float32Array(quantizedArray.length);\n          for (let i = 0; i < quantizedArray.length; i++) {\n            const v = quantizedArray[i];\n            values[i] = v * quantization.scale + quantization.min;\n          }\n        } else if (quantization.dtype === 'float16') {\n          if (float16Decode === undefined) {\n            float16Decode = getFloat16Decoder();\n          }\n          values = float16Decode(quantizedArray as Uint16Array);\n        } else {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type float32.`);\n        }\n      } else if (dtype === 'int32') {\n        if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type int32.`);\n        }\n        values = new Int32Array(quantizedArray.length);\n        for (let i = 0; i < quantizedArray.length; i++) {\n          const v = quantizedArray[i];\n          values[i] = Math.round(v * quantization.scale + quantization.min);\n        }\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * quantizationSizeFactor;\n    } else if (dtype === 'string') {\n      const size = sizeFromShape(spec.shape);\n      values = [];\n      for (let i = 0; i < size; i++) {\n        const byteLength = new Uint32Array(\n            buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n        offset += NUM_BYTES_STRING_LENGTH;\n        const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n        (values as Uint8Array[]).push(bytes);\n        offset += byteLength;\n      }\n    } else {\n      const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n      const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n      if (dtype === 'float32') {\n        values = new Float32Array(byteBuffer);\n      } else if (dtype === 'int32') {\n        values = new Int32Array(byteBuffer);\n      } else if (dtype === 'bool') {\n        values = new Uint8Array(byteBuffer);\n      } else if (dtype === 'complex64') {\n        values = new Float32Array(byteBuffer);\n        const real = new Float32Array(values.length / 2);\n        const image = new Float32Array(values.length / 2);\n        for (let i = 0; i < real.length; i++) {\n          real[i] = values[i * 2];\n          image[i] = values[i * 2 + 1];\n        }\n        const realTensor = tensor(real, shape, 'float32');\n        const imageTensor = tensor(image, shape, 'float32');\n        out[name] = complex(realTensor, imageTensor);\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * dtypeFactor;\n    }\n    if (dtype !== 'complex64') {\n      out[name] = tensor(values, shape, dtype);\n    }\n  }\n  return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n  // TODO(adarob, cais): Support quantization.\n  if (xs === null) {\n    throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n  }\n\n  let totalByteLength = 0;\n\n  // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n  // can have a different byte length from that of the `TypedArray` itself,\n  // for example, when the `TypedArray` is created from an offset in an\n  // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n  // the `TypedArray` in byte length. If an element of `xs` does not show\n  // this property, a new `TypedArray` that satisfy this property will be\n  // constructed and pushed into `normalizedXs`.\n  const normalizedXs: TypedArray[] = [];\n  xs.forEach((x: TypedArray) => {\n    totalByteLength += x.byteLength;\n    // tslint:disable:no-any\n    normalizedXs.push(\n        x.byteLength === x.buffer.byteLength ? x :\n                                               new (x.constructor as any)(x));\n    if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n          x as any instanceof Uint8Array)) {\n      throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n    }\n    // tslint:enable:no-any\n  });\n\n  const y = new Uint8Array(totalByteLength);\n  let offset = 0;\n  normalizedXs.forEach((x: TypedArray) => {\n    y.set(new Uint8Array(x.buffer), offset);\n    offset += x.byteLength;\n  });\n\n  return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n    (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n     typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n  if (useNodeBuffer) {\n    return Buffer.byteLength(str);\n  }\n  return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n  if (useNodeBuffer) {\n    return Buffer.from(buffer).toString('base64');\n  }\n  const buf = new Uint8Array(buffer);\n  let s = '';\n  for (let i = 0, l = buf.length; i < l; i++) {\n    s += String.fromCharCode(buf[i]);\n  }\n  return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n  if (useNodeBuffer) {\n    const buf = Buffer.from(str, 'base64');\n    return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n  }\n  const s = atob(str);\n  const buffer = new Uint8Array(s.length);\n  for (let i = 0; i < s.length; ++i) {\n    buffer.set([s.charCodeAt(i)], i);\n  }\n  return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n  if (buffers.length === 1) {\n    return buffers[0];\n  }\n\n  let totalByteLength = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    totalByteLength += buffer.byteLength;\n  });\n\n  const temp = new Uint8Array(totalByteLength);\n  let offset = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    temp.set(new Uint8Array(buffer), offset);\n    offset += buffer.byteLength;\n  });\n  return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n  const SEPARATOR = '/';\n  path = path.trim();\n  while (path.endsWith(SEPARATOR)) {\n    path = path.slice(0, path.length - 1);\n  }\n  const items = path.split(SEPARATOR);\n  return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n    ModelArtifactsInfo {\n  if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n    throw new Error('Expected JSON model topology, received ArrayBuffer.');\n  }\n\n  return {\n    dateSaved: new Date(),\n    modelTopologyType: 'JSON',\n    modelTopologyBytes: modelArtifacts.modelTopology == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n    weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n    weightDataBytes: modelArtifacts.weightData == null ?\n        0 :\n        modelArtifacts.weightData.byteLength,\n  };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n  const convertMantissa = (i: number): number => {\n    let m = i << 13;\n    let e = 0;\n\n    while ((m & 0x00800000) === 0) {\n      e -= 0x00800000;\n      m <<= 1;\n    }\n    m &= ~0x00800000;\n    e += 0x38800000;\n\n    return m | e;\n  };\n\n  const mantisaTable = new Uint32Array(2048);\n\n  mantisaTable[0] = 0;\n  for (let i = 1; i < 1024; i++) {\n    mantisaTable[i] = convertMantissa(i);\n  }\n  for (let i = 1024; i < 2048; i++) {\n    mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n  }\n\n  return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n  const exponentTable = new Uint32Array(64);\n\n  exponentTable[0] = 0;\n  exponentTable[31] = 0x47800000;\n  exponentTable[32] = 0x80000000;\n  exponentTable[63] = 0xc7800000;\n  for (let i = 1; i < 31; i++) {\n    exponentTable[i] = i << 23;\n  }\n  for (let i = 33; i < 63; i++) {\n    exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n  }\n\n  return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n  const offsetTable = new Uint32Array(64);\n\n  for (let i = 0; i < 64; i++) {\n    offsetTable[i] = 1024;\n  }\n  offsetTable[0] = offsetTable[32] = 0;\n\n  return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n *          the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n  // Algorithm is based off of\n  // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n  // Cache lookup tables\n  const mantisaTable = computeFloat16MantisaTable();\n  const exponentTable = computeFloat16ExponentTable();\n  const offsetTable = computeFloat16OffsetTable();\n\n  return (quantizedArray: Uint16Array) => {\n    const buffer = new ArrayBuffer(4 * quantizedArray.length);\n    const bufferUint32View = new Uint32Array(buffer);\n    for (let index = 0; index < quantizedArray.length; index++) {\n      const float16Bits = quantizedArray[index];\n      const float32Bits =\n          mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n          exponentTable[float16Bits >> 10];\n      bufferUint32View[index] = float32Bits;\n    }\n    return new Float32Array(buffer);\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n    IOHandler;\n\nexport class IORouterRegistry {\n  // Singleton instance.\n  private static instance: IORouterRegistry;\n\n  private saveRouters: IORouter[];\n  private loadRouters: IORouter[];\n\n  private constructor() {\n    this.saveRouters = [];\n    this.loadRouters = [];\n  }\n\n  private static getInstance(): IORouterRegistry {\n    if (IORouterRegistry.instance == null) {\n      IORouterRegistry.instance = new IORouterRegistry();\n    }\n    return IORouterRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerSaveRouter(saveRouter: IORouter) {\n    IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n  }\n\n  /**\n   * Register a load-handler router.\n   *\n   * @param loadRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `load` method defined or `null`.\n   */\n  static registerLoadRouter(loadRouter: IORouter) {\n    IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n  }\n\n  /**\n   * Look up IOHandler for saving, given a URL-like string.\n   *\n   * @param url\n   * @returns If only one match is found, an instance of IOHandler with the\n   * `save` method defined. If no match is found, `null`.\n   * @throws Error, if more than one match is found.\n   */\n  static getSaveHandlers(url: string|string[]): IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'save');\n  }\n\n  /**\n   * Look up IOHandler for loading, given a URL-like string.\n   *\n   * @param url\n   * @param loadOptions Optional, custom load options.\n   * @returns All valid handlers for `url`, given the currently registered\n   *   handler routers.\n   */\n  static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n      IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n  }\n\n  private static getHandlers(\n      url: string|string[], handlerType: 'save'|'load',\n      loadOptions?: LoadOptions): IOHandler[] {\n    const validHandlers: IOHandler[] = [];\n    const routers = handlerType === 'load' ?\n        IORouterRegistry.getInstance().loadRouters :\n        IORouterRegistry.getInstance().saveRouters;\n    routers.forEach(router => {\n      const handler = router(url, loadOptions);\n      if (handler !== null) {\n        validHandlers.push(handler);\n      }\n    });\n    return validHandlers;\n  }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n    IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n    (url: string|string[], loadOptions?: LoadOptions) =>\n        IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n//    as the type of topology (JSON vs binary), byte size of the topology, byte\n//    size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise<void> {\n  const idbFactory = getIndexedDBFactory();\n\n  return new Promise<void>((resolve, reject) => {\n    const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n    deleteRequest.onsuccess = () => resolve();\n    deleteRequest.onerror = error => reject(error);\n  });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n  if (!env().getBool('IS_BROWSER')) {\n    // TODO(cais): Add more info about what IOHandler subtypes are available.\n    //   Maybe point to a doc page on the web and/or automatically determine\n    //   the available IOHandlers and print them in the error message.\n    throw new Error(\n        'Failed to obtain IndexedDB factory because the current environment' +\n        'is not a web browser.');\n  }\n  // tslint:disable-next-line:no-any\n  const theWindow: any = typeof window === 'undefined' ? self : window;\n  const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n      theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n      theWindow.shimIndexedDB;\n  if (factory == null) {\n    throw new Error(\n        'The current browser does not appear to support IndexedDB.');\n  }\n  return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n  const db = openRequest.result as IDBDatabase;\n  db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n  db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n  protected readonly indexedDB: IDBFactory;\n  protected readonly modelPath: string;\n\n  static readonly URL_SCHEME = 'indexeddb://';\n\n  constructor(modelPath: string) {\n    this.indexedDB = getIndexedDBFactory();\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For IndexedDB, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    // TODO(cais): Support saving GraphDef models.\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    return this.databaseAction(this.modelPath, modelArtifacts) as\n        Promise<SaveResult>;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    return this.databaseAction(this.modelPath) as Promise<ModelArtifacts>;\n  }\n\n  /**\n   * Perform database action to put model artifacts into or read model artifacts\n   * from IndexedDB object store.\n   *\n   * Whether the action is put or get depends on whether `modelArtifacts` is\n   * specified. If it is specified, the action will be put; otherwise the action\n   * will be get.\n   *\n   * @param modelPath A unique string path for the model.\n   * @param modelArtifacts If specified, it will be the model artifacts to be\n   *   stored in IndexedDB.\n   * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n   *   of `ModelArtifacts`, if the action is get.\n   */\n  private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n      Promise<ModelArtifacts|SaveResult> {\n    return new Promise<ModelArtifacts|SaveResult>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n\n        if (modelArtifacts == null) {\n          // Read model out from object store.\n          const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n          const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n          const getRequest = modelStore.get(this.modelPath);\n          getRequest.onsuccess = () => {\n            if (getRequest.result == null) {\n              db.close();\n              return reject(new Error(\n                  `Cannot find model with path '${this.modelPath}' ` +\n                  `in IndexedDB.`));\n            } else {\n              resolve(getRequest.result.modelArtifacts);\n            }\n          };\n          getRequest.onerror = error => {\n            db.close();\n            return reject(getRequest.error);\n          };\n          modelTx.oncomplete = () => db.close();\n        } else {\n          // Put model into object store.\n          const modelArtifactsInfo: ModelArtifactsInfo =\n              getModelArtifactsInfoForJSON(modelArtifacts);\n          // First, put ModelArtifactsInfo into info store.\n          const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n          let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n          const putInfoRequest =\n              infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n          let modelTx: IDBTransaction;\n          putInfoRequest.onsuccess = () => {\n            // Second, put model data into model store.\n            modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n            const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n            const putModelRequest = modelStore.put({\n              modelPath: this.modelPath,\n              modelArtifacts,\n              modelArtifactsInfo\n            });\n            putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n            putModelRequest.onerror = error => {\n              // If the put-model request fails, roll back the info entry as\n              // well.\n              infoStore = infoTx.objectStore(INFO_STORE_NAME);\n              const deleteInfoRequest = infoStore.delete(this.modelPath);\n              deleteInfoRequest.onsuccess = () => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n              deleteInfoRequest.onerror = error => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n            };\n          };\n          putInfoRequest.onerror = error => {\n            db.close();\n            return reject(putInfoRequest.error);\n          };\n          infoTx.oncomplete = () => {\n            if (modelTx == null) {\n              db.close();\n            } else {\n              modelTx.oncomplete = () => db.close();\n            }\n          };\n        }\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n      return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n *   which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n  return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n      key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n      key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n  private indexedDB: IDBFactory;\n\n  constructor() {\n    this.indexedDB = getIndexedDBFactory();\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    return new Promise<{[path: string]: ModelArtifactsInfo}>(\n        (resolve, reject) => {\n          const openRequest =\n              this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n          openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n          openRequest.onsuccess = () => {\n            const db = openRequest.result;\n            const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n            const store = tx.objectStore(INFO_STORE_NAME);\n            // tslint:disable:max-line-length\n            // Need to cast `store` as `any` here because TypeScript's DOM\n            // library does not have the `getAll()` method even though the\n            // method is supported in the latest version of most mainstream\n            // browsers:\n            // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n            // tslint:enable:max-line-length\n            // tslint:disable-next-line:no-any\n            const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n            getAllInfoRequest.onsuccess = () => {\n              const out: {[path: string]: ModelArtifactsInfo} = {};\n              for (const item of getAllInfoRequest.result) {\n                out[item.modelPath] = item.modelArtifactsInfo;\n              }\n              resolve(out);\n            };\n            getAllInfoRequest.onerror = error => {\n              db.close();\n              return reject(getAllInfoRequest.error);\n            };\n            tx.oncomplete = () => db.close();\n          };\n          openRequest.onerror = error => reject(openRequest.error);\n        });\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    return new Promise<ModelArtifactsInfo>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n        const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n        const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n        const getInfoRequest = infoStore.get(path);\n        let modelTx: IDBTransaction;\n        getInfoRequest.onsuccess = () => {\n          if (getInfoRequest.result == null) {\n            db.close();\n            return reject(new Error(\n                `Cannot find model with path '${path}' ` +\n                `in IndexedDB.`));\n          } else {\n            // First, delete the entry in the info store.\n            const deleteInfoRequest = infoStore.delete(path);\n            const deleteModelData = () => {\n              // Second, delete the entry in the model store.\n              modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n              const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n              const deleteModelRequest = modelStore.delete(path);\n              deleteModelRequest.onsuccess = () =>\n                  resolve(getInfoRequest.result.modelArtifactsInfo);\n              deleteModelRequest.onerror = error =>\n                  reject(getInfoRequest.error);\n            };\n            // Proceed with deleting model data regardless of whether deletion\n            // of info data succeeds or not.\n            deleteInfoRequest.onsuccess = deleteModelData;\n            deleteInfoRequest.onerror = error => {\n              deleteModelData();\n              db.close();\n              return reject(getInfoRequest.error);\n            };\n          }\n        };\n        getInfoRequest.onerror = error => {\n          db.close();\n          return reject(getInfoRequest.error);\n        };\n\n        infoTx.oncomplete = () => {\n          if (modelTx == null) {\n            db.close();\n          } else {\n            modelTx.oncomplete = () => db.close();\n          }\n        };\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n  if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n      typeof window.localStorage === 'undefined') {\n    throw new Error(\n        'purgeLocalStorageModels() cannot proceed because local storage is ' +\n        'unavailable in the current environment.');\n  }\n  const LS = window.localStorage;\n  const purgedModelPaths: string[] = [];\n  for (let i = 0; i < LS.length; ++i) {\n    const key = LS.key(i);\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    if (key.startsWith(prefix) && key.length > prefix.length) {\n      LS.removeItem(key);\n      const modelName = getModelPathFromKey(key);\n      if (purgedModelPaths.indexOf(modelName) === -1) {\n        purgedModelPaths.push(modelName);\n      }\n    }\n  }\n  return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n} {\n  return {\n    info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n    topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n    weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n    weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n    modelMetadata:\n        [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n  };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n  const items = key.split(PATH_SEPARATOR);\n  if (items.length < 3) {\n    throw new Error(`Invalid key format: ${key}`);\n  }\n  return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n      key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n      key;\n}\n\ndeclare type LocalStorageKeys = {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n  protected readonly LS: Storage;\n  protected readonly modelPath: string;\n  protected readonly keys: LocalStorageKeys;\n\n  static readonly URL_SCHEME = 'localstorage://';\n\n  constructor(modelPath: string) {\n    if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n        typeof window.localStorage === 'undefined') {\n      // TODO(cais): Add more info about what IOHandler subtypes are\n      // available.\n      //   Maybe point to a doc page on the web and/or automatically determine\n      //   the available IOHandlers and print them in the error message.\n      throw new Error(\n          'The current environment does not support local storage.');\n    }\n    this.LS = window.localStorage;\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For local storage, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n    this.keys = getModelKeys(this.modelPath);\n  }\n\n  /**\n   * Save model artifacts to browser local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @param modelArtifacts The model artifacts to be stored.\n   * @returns An instance of SaveResult.\n   */\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const topology = JSON.stringify(modelArtifacts.modelTopology);\n      const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n      const modelArtifactsInfo: ModelArtifactsInfo =\n          getModelArtifactsInfoForJSON(modelArtifacts);\n\n      try {\n        this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n        this.LS.setItem(this.keys.topology, topology);\n        this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n        this.LS.setItem(\n            this.keys.weightData,\n            arrayBufferToBase64String(modelArtifacts.weightData));\n        this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n          format: modelArtifacts.format,\n          generatedBy: modelArtifacts.generatedBy,\n          convertedBy: modelArtifacts.convertedBy,\n          userDefinedMetadata: modelArtifacts.userDefinedMetadata\n        }));\n\n        return {modelArtifactsInfo};\n      } catch (err) {\n        // If saving failed, clean up all items saved so far.\n        this.LS.removeItem(this.keys.info);\n        this.LS.removeItem(this.keys.topology);\n        this.LS.removeItem(this.keys.weightSpecs);\n        this.LS.removeItem(this.keys.weightData);\n        this.LS.removeItem(this.keys.modelMetadata);\n\n        throw new Error(\n            `Failed to save model '${this.modelPath}' to local storage: ` +\n            `size quota being exceeded is a possible cause of this failure: ` +\n            `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n            `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n            `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n      }\n    }\n  }\n\n  /**\n   * Load a model from local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const info =\n        JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n    if (info == null) {\n      throw new Error(\n          `In local storage, there is no model with name '${this.modelPath}'`);\n    }\n\n    if (info.modelTopologyType !== 'JSON') {\n      throw new Error(\n          'BrowserLocalStorage does not support loading non-JSON model ' +\n          'topology yet.');\n    }\n\n    const out: ModelArtifacts = {};\n\n    // Load topology.\n    const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n    if (topology == null) {\n      throw new Error(\n          `In local storage, the topology of model '${this.modelPath}' ` +\n          `is missing.`);\n    }\n    out.modelTopology = topology;\n\n    // Load weight specs.\n    const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n    if (weightSpecs == null) {\n      throw new Error(\n          `In local storage, the weight specs of model '${this.modelPath}' ` +\n          `are missing.`);\n    }\n    out.weightSpecs = weightSpecs;\n\n    // Load meta-data fields.\n    const metadataString = this.LS.getItem(this.keys.modelMetadata);\n    if (metadataString != null) {\n      const metadata = JSON.parse(metadataString) as ModelArtifacts;\n      out.format = metadata['format'];\n      out.generatedBy = metadata['generatedBy'];\n      out.convertedBy = metadata['convertedBy'];\n      out.userDefinedMetadata = metadata['userDefinedMetadata'];\n    }\n\n    // Load weight data.\n    const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n    if (weightDataBase64 == null) {\n      throw new Error(\n          `In local storage, the binary weight values of model ` +\n          `'${this.modelPath}' are missing.`);\n    }\n    out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n    return out;\n  }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n      return browserLocalStorage(\n          url.slice(BrowserLocalStorage.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n *   - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n *     model, such as date saved, type of the topology, size in bytes, etc.\n *   - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n *     style models, this is a stringized JSON.\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n *     model, can be used to decode the saved binary weight values (see\n *     item below).\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n *     weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n *   `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n  return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n  private readonly LS: Storage;\n\n  constructor() {\n    assert(\n        env().getBool('IS_BROWSER'),\n        () => 'Current environment is not a web browser');\n    assert(\n        typeof window === 'undefined' ||\n            typeof window.localStorage !== 'undefined',\n        () => 'Current browser does not appear to support localStorage');\n    this.LS = window.localStorage;\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    const out: {[path: string]: ModelArtifactsInfo} = {};\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n    for (let i = 0; i < this.LS.length; ++i) {\n      const key = this.LS.key(i);\n      if (key.startsWith(prefix) && key.endsWith(suffix)) {\n        const modelPath = getModelPathFromKey(key);\n        out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n      }\n    }\n    return out;\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    const keys = getModelKeys(path);\n    if (this.LS.getItem(keys.info) == null) {\n      throw new Error(`Cannot find model at path '${path}'`);\n    }\n    const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n    this.LS.removeItem(keys.info);\n    this.LS.removeItem(keys.topology);\n    this.LS.removeItem(keys.weightSpecs);\n    this.LS.removeItem(keys.weightData);\n    return info;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n *   string.\n * - Moving or copying model from one path to another in the same medium or from\n *   one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n  // Singleton instance.\n  private static instance: ModelStoreManagerRegistry;\n\n  private managers: {[scheme: string]: ModelStoreManager};\n\n  private constructor() {\n    this.managers = {};\n  }\n\n  private static getInstance(): ModelStoreManagerRegistry {\n    if (ModelStoreManagerRegistry.instance == null) {\n      ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n    }\n    return ModelStoreManagerRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerManager(scheme: string, manager: ModelStoreManager) {\n    assert(scheme != null, () => 'scheme must not be undefined or null.');\n    if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n      scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n    }\n    assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n    const registry = ModelStoreManagerRegistry.getInstance();\n    assert(\n        registry.managers[scheme] == null,\n        () => `A model store manager is already registered for scheme '${\n            scheme}'.`);\n    registry.managers[scheme] = manager;\n  }\n\n  static getManager(scheme: string): ModelStoreManager {\n    const manager = this.getInstance().managers[scheme];\n    if (manager == null) {\n      throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n    }\n    return manager;\n  }\n\n  static getSchemes(): string[] {\n    return Object.keys(this.getInstance().managers);\n  }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n *   Scheme: e.g., 'localstorage' in the example above.\n *   Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n  if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n    throw new Error(\n        `The url string provided does not contain a scheme. ` +\n        `Supported schemes are: ` +\n        `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n  }\n  return {\n    scheme: url.split(URL_SCHEME_SUFFIX)[0],\n    path: url.split(URL_SCHEME_SUFFIX)[1],\n  };\n}\n\nasync function cloneModelInternal(\n    sourceURL: string, destURL: string,\n    deleteSource = false): Promise<ModelArtifactsInfo> {\n  assert(\n      sourceURL !== destURL,\n      () => `Old path and new path are the same: '${sourceURL}'`);\n\n  const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n  assert(\n      loadHandlers.length > 0,\n      () => `Copying failed because no load handler is found for source URL ${\n          sourceURL}.`);\n  assert(\n      loadHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `load handlers for source URL ${sourceURL}.`);\n  const loadHandler = loadHandlers[0];\n\n  const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n  assert(\n      saveHandlers.length > 0,\n      () => `Copying failed because no save handler is found for destination ` +\n          `URL ${destURL}.`);\n  assert(\n      saveHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `save handlers for destination URL ${destURL}.`);\n  const saveHandler = saveHandlers[0];\n\n  const sourceScheme = parseURL(sourceURL).scheme;\n  const sourcePath = parseURL(sourceURL).path;\n  const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n  const modelArtifacts = await loadHandler.load();\n\n  // If moving within the same storage medium, remove the old model as soon as\n  // the loading is done. Without doing this, it is possible that the combined\n  // size of the two models will cause the cloning to fail.\n  if (deleteSource && sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  const saveResult = await saveHandler.save(modelArtifacts);\n\n  // If moving between mediums, the deletion is done after the save succeeds.\n  // This guards against the case in which saving to the destination medium\n  // fails.\n  if (deleteSource && !sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n *   'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n  const schemes = ModelStoreManagerRegistry.getSchemes();\n  const out: {[url: string]: ModelArtifactsInfo} = {};\n  for (const scheme of schemes) {\n    const schemeOut =\n        await ModelStoreManagerRegistry.getManager(scheme).listModels();\n    for (const path in schemeOut) {\n      const url = scheme + URL_SCHEME_SUFFIX + path;\n      out[url] = schemeOut[path];\n    }\n  }\n  return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n *   'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n *   is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise<ModelArtifactsInfo> {\n  const schemeAndPath = parseURL(url);\n  const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n  return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function copyModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = false;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function moveModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = true;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n  // According to the spec, the built-in encoder can do only UTF-8 encoding.\n  // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n  private textEncoder: TextEncoder;\n\n  fetch(path: string, init?: RequestInit): Promise<Response> {\n    return fetch(path, init);\n  }\n\n  now(): number {\n    return performance.now();\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Browser's encoder only supports utf-8, but got ${encoding}`);\n    }\n    if (this.textEncoder == null) {\n      this.textEncoder = new TextEncoder();\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    return new TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_BROWSER')) {\n  env().setPlatform('browser', new PlatformBrowser());\n\n  // Register LocalStorage IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n  } catch (err) {\n  }\n\n  // Register IndexedDB IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n  } catch (err) {\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n  // tslint:disable-next-line:no-require-imports\n  importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise<Response>;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n  systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n  systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n  return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n  private textEncoder: TextEncoder;\n  // tslint:disable-next-line:no-any\n  util: any;\n\n  constructor() {\n    // tslint:disable-next-line:no-require-imports\n    this.util = require('util');\n    // According to the spec, the built-in encoder can do only UTF-8 encoding.\n    // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n    this.textEncoder = new this.util.TextEncoder();\n  }\n\n  fetch(path: string, requestInits?: RequestInit): Promise<Response> {\n    if (env().global.fetch != null) {\n      return env().global.fetch(path, requestInits);\n    }\n\n    if (systemFetch == null) {\n      systemFetch = getNodeFetch.importFetch();\n    }\n    return systemFetch(path, requestInits);\n  }\n\n  now(): number {\n    const time = process.hrtime();\n    return time[0] * 1000 + time[1] / 1000000;\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Node built-in encoder only supports utf-8, but got ${encoding}`);\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    if (bytes.length === 0) {\n      return '';\n    }\n    return new this.util.TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_NODE')) {\n  env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer<R extends Rank, D extends DataType = 'float32'>(\n    shape: ShapeMap[R], dtype: D = 'float32' as D,\n    values?: DataTypeMap[D]): TensorBuffer<R, D> {\n  dtype = dtype || 'float32' as D;\n  util.assertNonNegativeIntegerDimensions(shape);\n  return new TensorBuffer<R, D>(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_<T extends Tensor>(x: T|TensorLike, dtype: DataType): T {\n  const $x = convertToTensor(x, 'x', 'cast');\n\n  // Sanity checks.\n  if (!util.isValidDtype(dtype)) {\n    throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n  }\n  if (dtype === 'string' && $x.dtype !== 'string' ||\n      dtype !== 'string' && $x.dtype === 'string') {\n    throw new Error('Only strings can be casted to strings');\n  }\n\n  const inputs: CastInputs = {x: $x};\n  const attrs: CastAttrs = {dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n      null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'clone', null);\n  const forward = () =>\n      ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n  const inputs: IdentityInputs = {x: $x};\n\n  // Note this op is called tf.identity in python. Hence the kernel name used\n  // here.\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print<T extends Tensor>(x: T, verbose = false): void {\n  console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n  buffer,\n  cast,\n  clone,\n  print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer<T>(f: () => T): Promise<T> {\n  return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n  private readonly modelTopologyFileName: string;\n  private readonly weightDataFileName: string;\n  private readonly jsonAnchor: HTMLAnchorElement;\n  private readonly weightDataAnchor: HTMLAnchorElement;\n\n  static readonly URL_SCHEME = 'downloads://';\n\n  constructor(fileNamePrefix?: string) {\n    if (!env().getBool('IS_BROWSER')) {\n      // TODO(cais): Provide info on what IOHandlers are available under the\n      //   current environment.\n      throw new Error(\n          'browserDownloads() cannot proceed because the current environment ' +\n          'is not a browser.');\n    }\n\n    if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n      fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n    }\n    if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n      fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n    }\n\n    this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n    this.weightDataFileName =\n        fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (typeof (document) === 'undefined') {\n      throw new Error(\n          'Browser downloads are not supported in ' +\n          'this environment since `document` is not present');\n    }\n    const weightsURL = window.URL.createObjectURL(new Blob(\n        [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserDownloads.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const weightsManifest: WeightsManifestConfig = [{\n        paths: ['./' + this.weightDataFileName],\n        weights: modelArtifacts.weightSpecs\n      }];\n      const modelTopologyAndWeightManifest: ModelJSON = {\n        modelTopology: modelArtifacts.modelTopology,\n        format: modelArtifacts.format,\n        generatedBy: modelArtifacts.generatedBy,\n        convertedBy: modelArtifacts.convertedBy,\n        weightsManifest\n      };\n      const modelTopologyAndWeightManifestURL =\n          window.URL.createObjectURL(new Blob(\n              [JSON.stringify(modelTopologyAndWeightManifest)],\n              {type: 'application/json'}));\n\n      // If anchor elements are not provided, create them without attaching them\n      // to parents, so that the downloaded file names can be controlled.\n      const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n                                                   this.jsonAnchor;\n      jsonAnchor.download = this.modelTopologyFileName;\n      jsonAnchor.href = modelTopologyAndWeightManifestURL;\n      // Trigger downloads by evoking a click event on the download anchors.\n      // When multiple downloads are started synchronously, Firefox will only\n      // save the last one.\n      await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n      if (modelArtifacts.weightData != null) {\n        const weightDataAnchor = this.weightDataAnchor == null ?\n            document.createElement('a') :\n            this.weightDataAnchor;\n        weightDataAnchor.download = this.weightDataFileName;\n        weightDataAnchor.href = weightsURL;\n        await defer(\n            () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n      }\n\n      return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n    }\n  }\n}\n\nclass BrowserFiles implements IOHandler {\n  private readonly files: File[];\n\n  constructor(files: File[]) {\n    if (files == null || files.length < 1) {\n      throw new Error(\n          `When calling browserFiles, at least 1 file is required, ` +\n          `but received ${files}`);\n    }\n    this.files = files;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    const jsonFile = this.files[0];\n    const weightFiles = this.files.slice(1);\n\n    return new Promise<ModelArtifacts>((resolve, reject) => {\n      const jsonReader = new FileReader();\n      jsonReader.onload = (event: Event) => {\n        // tslint:disable-next-line:no-any\n        const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n        const modelTopology = modelJSON.modelTopology;\n        if (modelTopology == null) {\n          reject(new Error(\n              `modelTopology field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        if (weightFiles.length === 0) {\n          resolve({modelTopology});\n        }\n\n        const weightsManifest = modelJSON.weightsManifest;\n        if (weightsManifest == null) {\n          reject(new Error(\n              `weightManifest field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        let pathToFile: {[path: string]: File};\n        try {\n          pathToFile =\n              this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n        } catch (err) {\n          reject(err);\n          return;\n        }\n\n        const weightSpecs: WeightsManifestEntry[] = [];\n        const paths: string[] = [];\n        const perFileBuffers: ArrayBuffer[] = [];\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            paths.push(path);\n            perFileBuffers.push(null);\n          });\n          weightSpecs.push(...weightsGroup.weights);\n        });\n\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            const weightFileReader = new FileReader();\n            weightFileReader.onload = (event: Event) => {\n              // tslint:disable-next-line:no-any\n              const weightData = (event.target as any).result as ArrayBuffer;\n              const index = paths.indexOf(path);\n              perFileBuffers[index] = weightData;\n              if (perFileBuffers.indexOf(null) === -1) {\n                resolve({\n                  modelTopology,\n                  weightSpecs,\n                  weightData: concatenateArrayBuffers(perFileBuffers),\n                  format: modelJSON.format,\n                  generatedBy: modelJSON.generatedBy,\n                  convertedBy: modelJSON.convertedBy,\n                  userDefinedMetadata: modelJSON.userDefinedMetadata\n                });\n              }\n            };\n            weightFileReader.onerror = error =>\n                reject(`Failed to weights data from file of path '${path}'.`);\n            weightFileReader.readAsArrayBuffer(pathToFile[path]);\n          });\n        });\n      };\n      jsonReader.onerror = error => reject(\n          `Failed to read model topology and weights manifest JSON ` +\n          `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n          `Keras-style tf.Model artifacts only.`);\n      jsonReader.readAsText(jsonFile);\n    });\n  }\n\n  /**\n   * Check the compatibility between weights manifest and weight files.\n   */\n  private checkManifestAndWeightFiles(\n      manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n    const basenames: string[] = [];\n    const fileNames = files.map(file => basename(file.name));\n    const pathToFile: {[path: string]: File} = {};\n    for (const group of manifest) {\n      group.paths.forEach(path => {\n        const pathBasename = basename(path);\n        if (basenames.indexOf(pathBasename) !== -1) {\n          throw new Error(\n              `Duplicate file basename found in weights manifest: ` +\n              `'${pathBasename}'`);\n        }\n        basenames.push(pathBasename);\n        if (fileNames.indexOf(pathBasename) === -1) {\n          throw new Error(\n              `Weight file with basename '${pathBasename}' is not provided.`);\n        } else {\n          pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n        }\n      });\n    }\n\n    if (basenames.length !== files.length) {\n      throw new Error(\n          `Mismatch in the number of files in weights manifest ` +\n          `(${basenames.length}) and the number of weight files provided ` +\n          `(${files.length}).`);\n    }\n    return pathToFile;\n  }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n      return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * //   'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n *   `tf.Model`, `fileNamePrefix` should follow either of the following two\n *   formats:\n *   1. `null` or `undefined`, in which case the default file\n *      names will be used:\n *      - 'model.json' for the JSON file containing the model topology and\n *        weights manifest.\n *      - 'model.weights.bin' for the binary file containing the binary weight\n *        values.\n *   2. A single string or an Array of a single string, as the file name prefix.\n *      For example, if `'foo'` is provided, the downloaded JSON\n *      file and binary weights file will be named 'foo.json' and\n *      'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n  return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * //   elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (`<input type=\"file\" ...>`)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n *   which an `Array` of `File`s is expected (in that order):\n *   - A JSON file containing the model topology and weight manifest.\n *   - Optionally, One or more binary files containing the binary weights.\n *     These files must have names that match the paths in the `weightsManifest`\n *     contained by the aforementioned JSON file, or errors will be thrown\n *     during loading. These weights files have the same format as the ones\n *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n *     Python PIP package. If no weights files are provided, only the model\n *     topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n  return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n    promises: Array<Promise<{}|void>>, onProgress: OnProgressCallback,\n    startFraction?: number, endFraction?: number) {\n  checkPromises(promises);\n  startFraction = startFraction == null ? 0 : startFraction;\n  endFraction = endFraction == null ? 1 : endFraction;\n  checkFraction(startFraction, endFraction);\n  let resolvedPromise = 0;\n\n  const registerMonitor = (promise: Promise<{}>) => {\n    promise.then(value => {\n      const fraction = startFraction +\n          ++resolvedPromise / promises.length * (endFraction - startFraction);\n      // pass fraction as parameter to callback function.\n      onProgress(fraction);\n      return value;\n    });\n    return promise;\n  };\n\n  function checkPromises(promises: Array<Promise<{}|void>>): void {\n    assert(\n        promises != null && Array.isArray(promises) && promises.length > 0,\n        () => 'promises must be a none empty array');\n  }\n\n  function checkFraction(startFraction: number, endFraction: number): void {\n    assert(\n        startFraction >= 0 && startFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got startFraction ${startFraction}`);\n    assert(\n        endFraction >= 0 && endFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got endFraction ${endFraction}`);\n    assert(\n        endFraction >= startFraction,\n        () => `startFraction must be no more than endFraction, but ` +\n            `got startFraction ${startFraction} and endFraction ` +\n            `${endFraction}`);\n  }\n\n  return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n *   before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n *   length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n    fetchURLs: string[], loadOptions?: LoadOptions): Promise<ArrayBuffer[]> {\n  if (loadOptions == null) {\n    loadOptions = {};\n  }\n\n  const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n                                                    loadOptions.fetchFunc;\n\n  // Create the requests for all of the weights in parallel.\n  const requests = fetchURLs.map(\n      fetchURL =>\n          fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n  const fetchStartFraction = 0;\n  const fetchEndFraction = 0.5;\n\n  const responses = loadOptions.onProgress == null ?\n      await Promise.all(requests) :\n      await monitorPromisesProgress(\n          requests, loadOptions.onProgress, fetchStartFraction,\n          fetchEndFraction);\n\n  const bufferPromises = responses.map(response => response.arrayBuffer());\n\n  const bufferStartFraction = 0.5;\n  const bufferEndFraction = 1;\n\n  const buffers = loadOptions.onProgress == null ?\n      await Promise.all(bufferPromises) :\n      await monitorPromisesProgress(\n          bufferPromises, loadOptions.onProgress, bufferStartFraction,\n          bufferEndFraction);\n  return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n *     Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n    manifest: WeightsManifestConfig, filePathPrefix = '',\n    weightNames?: string[],\n    requestInit?: RequestInit): Promise<NamedTensorMap> {\n  // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n  // single weight from a group, the whole group will be fetched. At a future\n  // date, we should support fetching only the individual shards within a\n  // group that are needed to reconstruct the requested weight.\n  // TODO(cais): Use `decodeWeights` for implementation.\n\n  const fetchWeights = (fetchUrls: string[]) =>\n      loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n  const loadWeights = weightsLoaderFactory(fetchWeights);\n\n  return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n *   filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n *   fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n    fetchWeightsFunction: (fetchUrls: string[]) => Promise<ArrayBuffer[]>):\n    (manifest: WeightsManifestConfig, filePathPrefix?: string,\n     weightNames?: string[]) => Promise<NamedTensorMap> {\n  return async(\n             manifest: WeightsManifestConfig, filePathPrefix = '',\n             weightNames?: string[]): Promise<NamedTensorMap> => {\n    // Collect all the groups, weights, and their relative offsets to be\n    // fetched.\n    const groupIndicesToFetchMap = manifest.map(() => false);\n    const groupWeightsToFetch: {\n      [group: number]: Array<{\n        manifestEntry: WeightsManifestEntry; groupOffset: number;\n        sizeBytes: number;\n      }>\n    } = {};\n    const weightsFound =\n        weightNames != null ? weightNames.map(() => false) : [];\n    const allManifestWeightNames: string[] = [];\n    manifest.forEach((manifestGroupConfig, groupIndex) => {\n      let groupOffset = 0;\n      manifestGroupConfig.weights.forEach(weightsEntry => {\n        const rawDtype = ('quantization' in weightsEntry) ?\n            weightsEntry.quantization.dtype :\n            weightsEntry.dtype;\n\n        const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n            util.sizeFromShape(weightsEntry.shape);\n\n        const enqueueWeightsForFetchingFn = () => {\n          groupIndicesToFetchMap[groupIndex] = true;\n          if (groupWeightsToFetch[groupIndex] == null) {\n            groupWeightsToFetch[groupIndex] = [];\n          }\n\n          groupWeightsToFetch[groupIndex].push({\n            manifestEntry: weightsEntry,\n            groupOffset,\n            sizeBytes: weightsBytes\n          });\n        };\n\n        if (weightNames != null) {\n          weightNames.forEach((weightName, weightIndex) => {\n            if (weightName === weightsEntry.name) {\n              enqueueWeightsForFetchingFn();\n              weightsFound[weightIndex] = true;\n            }\n          });\n        } else {\n          enqueueWeightsForFetchingFn();\n        }\n\n        allManifestWeightNames.push(weightsEntry.name);\n        groupOffset += weightsBytes;\n      });\n    });\n\n    if (!weightsFound.every(found => found)) {\n      const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n      throw new Error(\n          `Could not find weights in manifest with names: ` +\n          `${weightsNotFound.join(', ')}. \\n` +\n          `Manifest JSON has weights with names: ` +\n          `${allManifestWeightNames.join(', ')}.`);\n    }\n\n    // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n    // IDs.\n    const groupIndicesToFetch =\n        groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n          if (shouldFetch) {\n            accumulator.push(i);\n          }\n          return accumulator;\n        }, []);\n\n    const fetchUrls: string[] = [];\n    groupIndicesToFetch.forEach(i => {\n      manifest[i].paths.forEach(filepath => {\n        const fetchUrl = filePathPrefix +\n            (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n        fetchUrls.push(fetchUrl);\n      });\n    });\n    const buffers = await fetchWeightsFunction(fetchUrls);\n\n    const weightsTensorMap: NamedTensorMap = {};\n    let bufferIndexOffset = 0;\n    groupIndicesToFetch.forEach(i => {\n      const numBuffers = manifest[i].paths.length;\n\n      let groupBytes = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        groupBytes += buffers[bufferIndexOffset + i].byteLength;\n      }\n\n      // Create a buffer for the whole group.\n      const groupBuffer = new ArrayBuffer(groupBytes);\n      const groupByteBuffer = new Uint8Array(groupBuffer);\n      let groupBufferOffset = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n        groupByteBuffer.set(buffer, groupBufferOffset);\n        groupBufferOffset += buffer.byteLength;\n      }\n\n      const weightsEntries = groupWeightsToFetch[i];\n      weightsEntries.forEach(weightsEntry => {\n        const byteBuffer = groupBuffer.slice(\n            weightsEntry.groupOffset,\n            weightsEntry.groupOffset + weightsEntry.sizeBytes);\n        const nameToTensorMap =\n            decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n        for (const name in nameToTensorMap) {\n          weightsTensorMap[name] = nameToTensorMap[name];\n        }\n      });\n\n      bufferIndexOffset += numBuffers;\n    });\n\n    return weightsTensorMap;\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n  protected readonly path: string;\n  protected readonly requestInit: RequestInit;\n\n  private readonly fetch: Function;\n  private readonly weightUrlConverter: (weightName: string) => Promise<string>;\n\n  readonly DEFAULT_METHOD = 'POST';\n\n  static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n  private readonly weightPathPrefix: string;\n  private readonly onProgress: OnProgressCallback;\n\n  constructor(path: string, loadOptions?: LoadOptions) {\n    if (loadOptions == null) {\n      loadOptions = {};\n    }\n    this.weightPathPrefix = loadOptions.weightPathPrefix;\n    this.onProgress = loadOptions.onProgress;\n    this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n    if (loadOptions.fetchFunc != null) {\n      assert(\n          typeof loadOptions.fetchFunc === 'function',\n          () => 'Must pass a function that matches the signature of ' +\n              '`fetch` (see ' +\n              'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n      this.fetch = loadOptions.fetchFunc;\n    } else {\n      this.fetch = env().platform.fetch;\n    }\n\n    assert(\n        path != null && path.length > 0,\n        () => 'URL path for http must not be null, undefined or ' +\n            'empty.');\n\n    if (Array.isArray(path)) {\n      assert(\n          path.length === 2,\n          () => 'URL paths for http must have a length of 2, ' +\n              `(actual length is ${path.length}).`);\n    }\n    this.path = path;\n\n    if (loadOptions.requestInit != null &&\n        loadOptions.requestInit.body != null) {\n      throw new Error(\n          'requestInit is expected to have no pre-existing body, but has one.');\n    }\n    this.requestInit = loadOptions.requestInit || {};\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserHTTPRequest.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n    init.body = new FormData();\n\n    const weightsManifest: WeightsManifestConfig = [{\n      paths: ['./model.weights.bin'],\n      weights: modelArtifacts.weightSpecs,\n    }];\n    const modelTopologyAndWeightManifest: ModelJSON = {\n      modelTopology: modelArtifacts.modelTopology,\n      format: modelArtifacts.format,\n      generatedBy: modelArtifacts.generatedBy,\n      convertedBy: modelArtifacts.convertedBy,\n      userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n      weightsManifest\n    };\n\n    init.body.append(\n        'model.json',\n        new Blob(\n            [JSON.stringify(modelTopologyAndWeightManifest)],\n            {type: JSON_TYPE}),\n        'model.json');\n\n    if (modelArtifacts.weightData != null) {\n      init.body.append(\n          'model.weights.bin',\n          new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n          'model.weights.bin');\n    }\n\n    const response = await this.fetch(this.path, init);\n\n    if (response.ok) {\n      return {\n        modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n        responses: [response],\n      };\n    } else {\n      throw new Error(\n          `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n          `${response.status}.`);\n    }\n  }\n\n  /**\n   * Load model artifacts via HTTP request(s).\n   *\n   * See the documentation to `tf.io.http` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model artifacts (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n    if (!modelConfigRequest.ok) {\n      throw new Error(\n          `Request to ${this.path} failed with status code ` +\n          `${modelConfigRequest.status}. Please verify this URL points to ` +\n          `the model JSON of the model to load.`);\n    }\n    let modelConfig: ModelJSON;\n    try {\n      modelConfig = await modelConfigRequest.json();\n    } catch (e) {\n      let message = `Failed to parse model JSON of response from ${this.path}.`;\n      // TODO(nsthorat): Remove this after some time when we're comfortable that\n      // .pb files are mostly gone.\n      if (this.path.endsWith('.pb')) {\n        message += ' Your path contains a .pb file extension. ' +\n            'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n            'in favor of .json models. You can re-convert your Python ' +\n            'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n            'or you can convert your.pb models with the \\'pb2json\\'' +\n            'NPM script in the tensorflow/tfjs-converter repository.';\n      } else {\n        message += ' Please make sure the server is serving valid ' +\n            'JSON for this request.';\n      }\n      throw new Error(message);\n    }\n    const modelTopology = modelConfig.modelTopology;\n    const weightsManifest = modelConfig.weightsManifest;\n    const generatedBy = modelConfig.generatedBy;\n    const convertedBy = modelConfig.convertedBy;\n    const format = modelConfig.format;\n    const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n    // We do not allow both modelTopology and weightsManifest to be missing.\n    if (modelTopology == null && weightsManifest == null) {\n      throw new Error(\n          `The JSON from HTTP path ${this.path} contains neither model ` +\n          `topology or manifest for weights.`);\n    }\n\n    let weightSpecs: WeightsManifestEntry[];\n    let weightData: ArrayBuffer;\n    if (weightsManifest != null) {\n      const results = await this.loadWeights(weightsManifest);\n      [weightSpecs, weightData] = results;\n    }\n\n    const artifacts: ModelArtifacts = {\n      modelTopology,\n      weightSpecs,\n      weightData,\n      userDefinedMetadata,\n      generatedBy,\n      convertedBy,\n      format\n    };\n\n    const initializer = modelConfig.modelInitializer;\n    if (initializer) {\n      artifacts.modelInitializer = initializer;\n    }\n\n    return artifacts;\n  }\n\n  private async loadWeights(weightsManifest: WeightsManifestConfig):\n      Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n    const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n    const [prefix, suffix] = parseUrl(weightPath);\n    const pathPrefix = this.weightPathPrefix || prefix;\n\n    const weightSpecs = [];\n    for (const entry of weightsManifest) {\n      weightSpecs.push(...entry.weights);\n    }\n\n    const fetchURLs: string[] = [];\n    const urlPromises: Array<Promise<string>> = [];\n    for (const weightsGroup of weightsManifest) {\n      for (const path of weightsGroup.paths) {\n        if (this.weightUrlConverter != null) {\n          urlPromises.push(this.weightUrlConverter(path));\n        } else {\n          fetchURLs.push(pathPrefix + path + suffix);\n        }\n      }\n    }\n\n    if (this.weightUrlConverter) {\n      fetchURLs.push(...await Promise.all(urlPromises));\n    }\n\n    const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n      requestInit: this.requestInit,\n      fetchFunc: this.fetch,\n      onProgress: this.onProgress\n    });\n    return [weightSpecs, concatenateArrayBuffers(buffers)];\n  }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n  const lastSlash = url.lastIndexOf('/');\n  const lastSearchParam = url.lastIndexOf('?');\n  const prefix = url.substring(0, lastSlash);\n  const suffix =\n      lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n  return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n  return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n    (url: string, loadOptions?: LoadOptions) => {\n      if (typeof fetch === 'undefined' &&\n          (loadOptions == null || loadOptions.fetchFunc == null)) {\n        // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n        // an environment that is not the browser or node they have to setup a\n        // global fetch polyfill.\n        return null;\n      } else {\n        let isHTTP = true;\n        if (Array.isArray(url)) {\n          isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n        } else {\n          isHTTP = isHTTPScheme(url);\n        }\n        if (isHTTP) {\n          return http(url, loadOptions);\n        }\n      }\n      return null;\n    };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n *   - A JSON file consisting of `modelTopology` and `weightsManifest`.\n *   - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n *     'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n *   Can be an absolute HTTP path (e.g.,\n *   'http://localhost:8000/model-upload)') or a relative path (e.g.,\n *   './model-upload').\n * @param requestInit Request configurations to be used when sending\n *    HTTP request to server using `fetch`. It can contain fields such as\n *    `method`, `credentials`, `headers`, `mode`, etc. See\n *    https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n *    for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n *   following fields:\n *   - weightPathPrefix Optional, this specifies the path prefix for weight\n *     files, by default this is calculated from the path param.\n *   - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n *     the `fetch` from node-fetch can be used here.\n *   - onProgress Optional, progress callback function, fired periodically\n *     before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n  return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n    path: string, loadOptions?: LoadOptions): IOHandler {\n  return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n  constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n  async load(): Promise<ModelArtifacts> {\n    return this.modelArtifacts;\n  }\n}\n\nclass PassthroughSaver implements IOHandler {\n  constructor(\n      private readonly saveHandler:\n          (artifacts: ModelArtifacts) => Promise<SaveResult>) {}\n\n  async save(modelArtifacts: ModelArtifacts) {\n    return this.saveHandler(modelArtifacts);\n  }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n *     modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n *   the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n *   names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n *   concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n    modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n    weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n  if (arguments.length === 1) {\n    const isModelArtifacts =\n        (modelArtifacts as ModelArtifacts).modelTopology != null ||\n        (modelArtifacts as ModelArtifacts).weightSpecs != null;\n    if (isModelArtifacts) {\n      return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n    } else {\n      // Legacy support: with only modelTopology.\n      // TODO(cais): Remove this deprecated API.\n      console.warn(\n          'Please call tf.io.fromMemory() with only one argument. ' +\n          'The argument should be of type ModelArtifacts. ' +\n          'The multi-argument signature of tf.io.fromMemory() has been ' +\n          'deprecated and will be removed in a future release.');\n      return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n    }\n  } else {\n    // Legacy support.\n    // TODO(cais): Remove this deprecated API.\n    console.warn(\n        'Please call tf.io.fromMemory() with only one argument. ' +\n        'The argument should be of type ModelArtifacts. ' +\n        'The multi-argument signature of tf.io.fromMemory() has been ' +\n        'deprecated and will be removed in a future release.');\n    return new PassthroughLoader({\n      modelTopology: modelArtifacts as {},\n      weightSpecs,\n      weightData,\n      trainingConfig\n    });\n  }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n *   // ... do something with the artifacts ...\n *   return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n *     `SaveResult`.\n */\nexport function withSaveHandler(\n    saveHandler: (artifacts: ModelArtifacts) =>\n        Promise<SaveResult>): IOHandler {\n  return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n  browserFiles,\n  browserHTTPRequest,\n  concatenateArrayBuffers,\n  decodeWeights,\n  encodeWeights,\n  fromMemory,\n  getLoadHandlers,\n  getModelArtifactsInfoForJSON,\n  getSaveHandlers,\n  http,\n  IOHandler,\n  isHTTPScheme,\n  LoadHandler,\n  LoadOptions,\n  loadWeights,\n  ModelArtifacts,\n  ModelArtifactsInfo,\n  ModelJSON,\n  ModelStoreManager,\n  OnProgressCallback,\n  registerLoadRouter,\n  registerSaveRouter,\n  RequestDetails,\n  SaveConfig,\n  SaveHandler,\n  SaveResult,\n  WeightGroup,\n  weightsLoaderFactory,\n  WeightsManifestConfig,\n  WeightsManifestEntry,\n  withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  const $x = convertToTensor(x, 'x', 'reshape', null);\n\n  const inputs: ReshapeInputs = {x: $x};\n  const attrs: ReshapeAttrs = {shape};\n  const forward: ForwardFunc<\n      Tensor<R>> = (backend: KernelBackend, save: GradSaveFunc) => {\n    shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n    util.assert(\n        $x.size === util.sizeFromShape(shape),\n        () => 'new shape and old shape must have the same number of elements.');\n    save([$x]);\n    return backend.reshape($x, shape);\n  };\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n      attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print();  // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike, transposeA = false,\n    transposeB = false): T {\n  let $a = convertToTensor(a, 'a', 'matMul');\n  let $b = convertToTensor(b, 'b', 'matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n          `got ranks ${$a.rank} and ${$b.rank}.`);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n                           reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n                           reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([a3D, b3D]);\n\n    return backend.batchMatMul(\n        a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n  };\n\n  const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n  const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n      attrs as {} as NamedAttrMap);\n\n  return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n *     not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n    indices: Tensor|TensorLike, depth: number, onValue = 1,\n    offValue = 0): Tensor {\n  if (depth < 2) {\n    throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n  }\n  const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n  const outShape = [...$indices.shape, depth];\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$indices]);\n    return reshape(\n        backend.oneHot(\n            reshape($indices, [$indices.size]), depth, onValue, offValue),\n        outShape);\n  };\n\n  const inputs: OneHotInputs = {indices: $indices};\n  const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print();  // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_<T extends Tensor>(x: T|TensorLike, perm?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'transpose');\n\n  if (perm == null) {\n    perm = $x.shape.map((s, i) => i).reverse();\n  }\n  util.assert(\n      $x.rank === perm.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of perm ${perm}.`);\n  perm.forEach(axis => {\n    util.assert(\n        axis >= 0 && axis < $x.rank,\n        () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n            ` but got ${perm}`);\n  });\n\n  if ($x.rank <= 1) {\n    return $x.clone();\n  }\n\n  const inputs: TransposeInputs = {x: $x};\n  const attrs: TransposeAttrs = {perm};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n      null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * //  [0, 1, 1],\n * //  [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n *   for the classes. The shape is `[numExamples]`, where\n *   `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n *   0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n *   Its value must be larger than the largest element in `labels` and\n *   `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n *   row `r` and column `c` is the number of times examples of actual class\n *   `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n    labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n    numClasses: number): Tensor2D {\n  const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n  util.assert(\n      numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n      () => `If provided, numClasses must be a positive integer, ` +\n          `but got ${numClasses}`);\n  util.assert(\n      $labels.rank === 1,\n      () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n  util.assert(\n      $predictions.rank === 1,\n      () => `Expected the rank of predictions to be 1, ` +\n          `but got ${$predictions.rank}`);\n  util.assert(\n      $labels.shape[0] === $predictions.shape[0],\n      () => `Mismatch in the number of examples: ` +\n          `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n          `Labels and predictions should have the same number of elements.`);\n  util.assert(\n      numClasses > 0 && Number.isInteger(numClasses),\n      () => `numClasses is required to be a positive integer, but got ` +\n          `${numClasses}`);\n  // TODO(cais): In the future, if oneHot supports tensors inputs for\n  //   `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n  const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n  const oneHotPredictions =\n      oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n  const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n  return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided,  it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n    values: TensorLike3D, shape?: [number, number, number],\n    dtype?: DataType): Tensor3D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 3) {\n    throw new Error('tensor3d() requires shape to have three numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor3d() requires values to be number[][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor3d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement,\n    numChannels = 3): Tensor3D {\n  // Sanity checks.\n  if (numChannels > 4) {\n    throw new Error(\n        'Cannot construct Tensor with more than 4 channels from pixels.');\n  }\n  if (pixels == null) {\n    throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n  }\n  let isPixelData = false;\n  let isImageData = false;\n  let isVideo = false;\n  let isImage = false;\n  let isCanvasLike = false;\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    isPixelData = true;\n  } else if (\n      typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n    isImageData = true;\n  } else if (\n      typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement) {\n    isVideo = true;\n  } else if (\n      typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement) {\n    isImage = true;\n    // tslint:disable-next-line: no-any\n  } else if ((pixels as any).getContext != null) {\n    isCanvasLike = true;\n  } else {\n    throw new Error(\n        'pixels passed to tf.browser.fromPixels() must be either an ' +\n        `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n        `in browser, or OffscreenCanvas, ImageData in webworker` +\n        ` or {data: Uint32Array, width: number, height: number}, ` +\n        `but was ${(pixels as {}).constructor.name}`);\n  }\n  if (isVideo) {\n    const HAVE_CURRENT_DATA_READY_STATE = 2;\n    if (isVideo &&\n        (pixels as HTMLVideoElement).readyState <\n            HAVE_CURRENT_DATA_READY_STATE) {\n      throw new Error(\n          'The video element has not loaded data yet. Please wait for ' +\n          '`loadeddata` event on the <video> element.');\n    }\n  }\n  // If the current backend has 'FromPixels' registered, it has a more\n  // efficient way of handling pixel uploads, so we call that.\n  const kernel = getKernel(FromPixels, ENGINE.backendName);\n  if (kernel != null) {\n    const inputs: FromPixelsInputs = {pixels};\n    const attrs: FromPixelsAttrs = {numChannels};\n    return ENGINE.runKernel(\n               FromPixels, inputs as {} as NamedTensorMap,\n               attrs as {} as NamedAttrMap) as Tensor3D;\n  }\n\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n  let vals: Uint8ClampedArray|Uint8Array;\n\n  if (isCanvasLike) {\n    vals =\n        // tslint:disable-next-line:no-any\n        (pixels as any).getContext('2d').getImageData(0, 0, width, height).data;\n  } else if (isImageData || isPixelData) {\n    vals = (pixels as PixelData | ImageData).data;\n  } else if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement, 0, 0, width, height);\n    vals = fromPixels2DContext.getImageData(0, 0, width, height).data;\n  }\n  let values: Int32Array;\n  if (numChannels === 4) {\n    values = new Int32Array(vals);\n  } else {\n    const numPixels = width * height;\n    values = new Int32Array(numPixels * numChannels);\n    for (let i = 0; i < numPixels; i++) {\n      for (let channel = 0; channel < numChannels; ++channel) {\n        values[i * numChannels + channel] = vals[i * 4 + channel];\n      }\n    }\n  }\n  const outShape: [number, number, number] = [height, width, numChannels];\n  return tensor3d(values, outShape, 'int32');\n}\n\n/**\n * Draws a `tf.Tensor` of pixel values to a byte array or optionally a\n * canvas.\n *\n * When the dtype of the input is 'float32', we assume values in the range\n * [0-1]. Otherwise, when input is 'int32', we assume values in the range\n * [0-255].\n *\n * Returns a promise that resolves when the canvas has been drawn to.\n *\n * @param img A rank-2 or rank-3 tensor. If rank-2, draws grayscale. If\n *     rank-3, must have depth of 1, 3 or 4. When depth of 1, draws\n * grayscale. When depth of 3, we draw with the first three components of\n * the depth dimension corresponding to r, g, b and alpha = 1. When depth of\n * 4, all four components of the depth dimension correspond to r, g, b, a.\n * @param canvas The canvas to draw to.\n *\n * @doc {heading: 'Browser', namespace: 'browser'}\n */\nexport async function toPixels(\n    img: Tensor2D|Tensor3D|TensorLike,\n    canvas?: HTMLCanvasElement): Promise<Uint8ClampedArray> {\n  let $img = convertToTensor(img, 'img', 'toPixels');\n  if (!(img instanceof Tensor)) {\n    // Assume int32 if user passed a native array.\n    const originalImgTensor = $img;\n    $img = cast(originalImgTensor, 'int32');\n    originalImgTensor.dispose();\n  }\n  if ($img.rank !== 2 && $img.rank !== 3) {\n    throw new Error(\n        `toPixels only supports rank 2 or 3 tensors, got rank ${$img.rank}.`);\n  }\n  const [height, width] = $img.shape.slice(0, 2);\n  const depth = $img.rank === 2 ? 1 : $img.shape[2];\n\n  if (depth > 4 || depth === 2) {\n    throw new Error(\n        `toPixels only supports depth of size ` +\n        `1, 3 or 4 but got ${depth}`);\n  }\n\n  if ($img.dtype !== 'float32' && $img.dtype !== 'int32') {\n    throw new Error(\n        `Unsupported type for toPixels: ${$img.dtype}.` +\n        ` Please use float32 or int32 tensors.`);\n  }\n\n  const data = await $img.data();\n  const multiplier = $img.dtype === 'float32' ? 255 : 1;\n  const bytes = new Uint8ClampedArray(width * height * 4);\n\n  for (let i = 0; i < height * width; ++i) {\n    const rgba = [0, 0, 0, 255];\n\n    for (let d = 0; d < depth; d++) {\n      const value = data[i * depth + d];\n\n      if ($img.dtype === 'float32') {\n        if (value < 0 || value > 1) {\n          throw new Error(\n              `Tensor values for a float32 Tensor must be in the ` +\n              `range [0 - 1] but encountered ${value}.`);\n        }\n      } else if ($img.dtype === 'int32') {\n        if (value < 0 || value > 255) {\n          throw new Error(\n              `Tensor values for a int32 Tensor must be in the ` +\n              `range [0 - 255] but encountered ${value}.`);\n        }\n      }\n\n      if (depth === 1) {\n        rgba[0] = value * multiplier;\n        rgba[1] = value * multiplier;\n        rgba[2] = value * multiplier;\n      } else {\n        rgba[d] = value * multiplier;\n      }\n    }\n\n    const j = i * 4;\n    bytes[j + 0] = Math.round(rgba[0]);\n    bytes[j + 1] = Math.round(rgba[1]);\n    bytes[j + 2] = Math.round(rgba[2]);\n    bytes[j + 3] = Math.round(rgba[3]);\n  }\n\n  if (canvas != null) {\n    canvas.width = width;\n    canvas.height = height;\n    const ctx = canvas.getContext('2d');\n    const imageData = new ImageData(bytes, width, height);\n    ctx.putImageData(imageData, 0, 0);\n  }\n  if ($img !== img) {\n    $img.dispose();\n  }\n  return bytes;\n}\n\nexport const fromPixels = op({fromPixels_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\nimport {computeStrides} from '../util';\n\n/**\n * Validate gather nd inputs.\n *\n * @param tensor The tensor contains the source values.\n * @param indices The tensor contains the indices to slice the source.\n *\n * @returns [resultShape, numUpdates, sliceSize, strides]\n */\nexport function prepareAndValidate(\n    tensor: Tensor, indices: Tensor): [number[], number, number, number[]] {\n  if (tensor.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the input to be rank 1 or higher,' +\n        ` but the rank was ${tensor.rank}.`);\n  }\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(\n        'tf.gatherND() expects the indices to be int32 type,' +\n        ` but the dtype was ${indices.dtype}.`);\n  }\n  if (indices.shape[indices.rank - 1] > tensor.rank) {\n    throw new Error(\n        'index innermost dimension length must be <= tensor rank; saw: ' +\n        `${indices.shape[indices.rank - 1]} vs. ${tensor.rank}`);\n  }\n\n  if (tensor.size === 0) {\n    throw new Error(\n        'Requested more than 0 entries, but input is empty.' +\n        ` Input shape: ${tensor.shape}.`);\n  }\n\n  const indicesShape = indices.shape;\n  const sliceRank = indicesShape[indicesShape.length - 1];\n\n  // The result shape is\n  //   indices.shape[:-1] + params.shape[indices.shape[-1]:]\n  let nResult = 1;\n  for (let i = 0; i < indicesShape.length - 1; ++i) {\n    nResult *= indicesShape[i];\n  }\n\n  const inputShape = tensor.shape;\n\n  const resultShape = indicesShape.slice();\n  resultShape.pop();\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < tensor.rank; ++i) {\n    sliceSize *= inputShape[i];\n    resultShape.push(inputShape[i]);\n  }\n\n  const strides =\n      [...computeStrides(tensor.shape).map(stride => stride / sliceSize),\n       1].slice(0, sliceRank);\n\n  return [resultShape, nResult, sliceSize, strides];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {computeStrides, sizeFromShape} from '../util';\n\n/**\n * Check whether updates.shape = indices.shape[:batchDim] +\n * shape[sliceDim:]\n *\n * @param x The input tensor.\n */\nexport function validateUpdateShape(\n    shape: number[], indices: Tensor, updates: Tensor) {\n  const sliceDim = (indices.rank > 1) ? indices.shape[indices.rank - 1] : 1;\n  const batchDim = (indices.rank > 1) ? indices.rank - 1 : 1;\n\n  const shapeError = 'Must have updates.shape = indices.shape[:batchDim] + ' +\n      `shape[sliceDim:], got updates.shape: ${updates.shape}` +\n      `, indices.shape: ${indices.shape}, shape: ${shape}` +\n      `, sliceDim: ${sliceDim}, and batchDim: ${batchDim}.`;\n\n  if (updates.rank < batchDim) {\n    throw new Error(shapeError + ` update.rank < ${batchDim}. `);\n  }\n  if (shape.length < sliceDim + (updates.rank - batchDim)) {\n    throw new Error(\n        shapeError +\n        ` Output shape length < ${sliceDim + (updates.rank - batchDim)}`);\n  }\n  if (updates.rank !== batchDim + shape.length - sliceDim) {\n    throw new Error(\n        shapeError + ` update.rank != ${batchDim + shape.length - sliceDim}`);\n  }\n  for (let d = 0; d < batchDim; ++d) {\n    if (updates.shape[d] !== indices.shape[d]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d}] (${updates.shape[d]}) != indices.shape[${d}] (${\n              indices.shape[d]}).`);\n    }\n  }\n  for (let d = 0; d < updates.rank - batchDim; ++d) {\n    if (updates.shape[d + batchDim] !== shape[d + sliceDim]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d + batchDim}] (${\n              updates.shape[d + batchDim]}) != shape[${d + batchDim}] (${\n              shape[d + batchDim]})`);\n    }\n  }\n}\n\nexport interface ScatterShapeInfo {\n  sliceRank: number;\n  numUpdates: number;\n  sliceSize: number;\n  strides: number[];\n  outputSize: number;\n}\n/**\n * Validate scatter nd inputs.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n */\nexport function validateInput(\n    updates: Tensor, indices: Tensor, shape: number[]) {\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (updates.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the updates to be rank 1 or higher,' +\n        ` but the rank was ${updates.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${\n        indices.dtype}`);\n  }\n  if (shape.length < 1) {\n    throw new Error(\n        `Output rank must be greater or equal to 1, but got shape: ${shape}`);\n  }\n\n  if (shape.length === 0) {\n    if (indices.size === 0) {\n      throw new Error(`Indices specified for empty output. indices shape: ${\n          indices.shape}`);\n    }\n    if (updates.size === 0) {\n      throw new Error(`Updates specified for empty output. updates shape: ${\n          updates.shape}`);\n    }\n  }\n\n  validateUpdateShape(shape, indices, updates);\n}\n\n/**\n * Calculate the shape information for the output.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n *\n * @returns ScatterShapeInfo\n */\nexport function calculateShapes(\n    updates: TensorInfo, indices: TensorInfo,\n    shape: number[]): ScatterShapeInfo {\n  // Calculate the number of dimensions in indices\n  const indicesRank = indices.shape.length;\n  const sliceRank = (indicesRank > 1) ? indices.shape[indicesRank - 1] : 1;\n\n  // Calculate the number of elements that make up each slice of our updated\n  // tensor. This allows us to work with flattened tensors and copy over whole\n  // slices at a time.\n  const totalNd = shape.length;\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < totalNd; ++i) {\n    sliceSize *= shape[i];\n  }\n\n  const safeSliceDim = (sliceRank < 1) ? 1 : sliceRank;\n  const numUpdates = sizeFromShape(indices.shape) / safeSliceDim;\n\n  const strides = [...computeStrides(shape.slice(0, sliceRank)), 1];\n  const outputSize = sizeFromShape(shape);\n  return {sliceRank, numUpdates, sliceSize, strides, outputSize};\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '../kernel_registry';\nimport * as util from '../util';\n\nexport function assertParamsValid(\n    input: TensorInfo, begin: number[], size: number[]): void {\n  const inputRank = input.shape.length;\n  util.assert(\n      inputRank === begin.length,\n      () => `Error in slice${inputRank}D: Length of begin ${begin} must ` +\n          `match the rank of the array (${inputRank}).`);\n  util.assert(\n      inputRank === size.length,\n      () => `Error in slice${inputRank}D: Length of size ${size} must ` +\n          `match the rank of the array (${inputRank}).`);\n\n  for (let i = 0; i < inputRank; ++i) {\n    util.assert(\n        begin[i] + size[i] <= input.shape[i],\n        () => `Error in slice${inputRank}D: begin[${i}] + size[${i}] ` +\n            `(${begin[i] + size[i]}) would overflow input.shape[${i}] (${\n                  input.shape[i]})`);\n  }\n}\n\n/** Converts a binary mask to an array of axes. Used in stridedSlice(). */\nexport function maskToAxes(mask: number): number[] {\n  const axes = [];\n  let axis = 0;\n  while (mask > 0) {\n    if (mask & 1) {\n      axes.push(axis);\n    }\n    mask /= 2;\n    axis++;\n  }\n  return axes;\n}\n\n/** Computes the output shape given the strided slice params. */\nexport function computeOutShape(\n    begin: number[], end: number[], strides: number[]): number[] {\n  const size = [];\n  for (let axis = 0; axis < begin.length; axis++) {\n    size[axis] = Math.ceil((end[axis] - begin[axis]) / strides[axis]);\n  }\n  return size;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stride value. Otherwise, insert.\nexport function stridesWithElidedDims(\n    strides: number[], ellipsisInsertionIndex: number, numElidedAxes: number,\n    inputShape: number[]): number[] {\n  const newStrides = [...strides];\n  for (let i = newStrides.length; i < inputShape.length; i++) {\n    newStrides.push(1);\n  }\n  for (let i = 0; i < numElidedAxes; i++) {\n    if (i === 0) {\n      newStrides[ellipsisInsertionIndex] = 1;\n    } else {\n      newStrides.splice(\n          ellipsisInsertionIndex, 0 /* num elements to delete */,\n          1 /* element to add */);\n      newStrides.pop();\n    }\n  }\n  return newStrides;\n}\n\nfunction unnormalizeAxis(\n    ellipsisInsertionIndex: number, numElidedAxes: number,\n    normalizedAxis: number): number {\n  if (normalizedAxis <= ellipsisInsertionIndex) {\n    return normalizedAxis;\n  }\n\n  return normalizedAxis - (numElidedAxes - 1);\n}\n\nfunction getElidedAxes(numElidedAxes: number, ellipsisInsertionIndex: number) {\n  const elidedAxes = [];\n  for (let i = 0; i < numElidedAxes; i++) {\n    elidedAxes.push(ellipsisInsertionIndex + i);\n  }\n  return elidedAxes;\n}\n\n// Normalize the start, end and strides.\nexport function getNormalizedAxes(\n    inputShape: number[], ellipsisAxes: number[], numInterpolatedAxes: number,\n    begin: number[], end: number[], strides: number[], beginMask: number,\n    endMask: number,\n    ellipsisMask: number): {begin: number[], end: number[], strides: number[]} {\n  const inputRank = inputShape.length;\n  let normalizedBegin = new Array(inputRank),\n      normalizedEnd = new Array(inputRank),\n      normalizedStrides = new Array(inputRank);\n  if (ellipsisAxes.length && numInterpolatedAxes > 0) {\n    const fullIndex = ellipsisAxes[0];\n\n    // The ellipsis applies to the masked index as well as any dimensions\n    // that are interpolated.\n    const numElidedAxes = numInterpolatedAxes + 1;\n    normalizedBegin = startIndicesWithElidedDims(\n        beginMask, fullIndex, numElidedAxes, begin, inputShape);\n    normalizedEnd = stopIndicesWithElidedDims(\n        endMask, fullIndex, numElidedAxes, end, inputShape);\n    normalizedStrides =\n        stridesWithElidedDims(strides, fullIndex, numElidedAxes, inputShape);\n  } else {\n    for (let axis = 0; axis < inputRank; axis++) {\n      normalizedBegin[axis] = startForAxis(\n          beginMask, begin, strides, inputShape, axis, ellipsisMask);\n      normalizedEnd[axis] =\n          stopForAxis(endMask, end, strides, inputShape, axis, ellipsisMask);\n      normalizedStrides[axis] = stridesForAxis(strides, axis, ellipsisMask);\n    }\n  }\n\n  return {\n    begin: normalizedBegin,\n    end: normalizedEnd,\n    strides: normalizedStrides\n  };\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current start value. Otherwise, insert.\nexport function startIndicesWithElidedDims(\n    beginMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalBegin: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = 0;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalBegin[originalAxis];\n      if (beginMask & 1 << originalAxis) {\n        originalValue = 0;\n      }\n\n      newIndices[axis] = originalValue;\n    }\n  }\n  return newIndices;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stop value. Otherwise, insert.\nexport function stopIndicesWithElidedDims(\n    endMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalEnd: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = Number.MAX_SAFE_INTEGER;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalEnd[originalAxis];\n      if (endMask & 1 << originalAxis) {\n        originalValue = Number.MAX_SAFE_INTEGER;\n      }\n      newIndices[axis] = originalValue;\n    }\n  }\n\n  for (let i = 0; i < newIndices.length; i++) {\n    // Handle negative indices\n    const axisSize = inputShape[i];\n    if (newIndices[i] < 0) {\n      newIndices[i] += axisSize;\n    }\n    newIndices[i] = util.clamp(0, newIndices[i], inputShape[i]);\n  }\n  return newIndices;\n}\n\nexport function stridesForAxis(\n    strides: number[], axis: number, ellipsisMask: number): number {\n  let stride = strides[axis];\n  if (ellipsisMask & (1 << axis) || stride == null) {\n    stride = 1;\n  }\n\n  return stride;\n}\n\nexport function startForAxis(\n    beginMask: number, startIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let start = startIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or the begin index is not set\n  // for the axis.\n  if (beginMask & 1 << axis || ellipsisMask & 1 << axis || start == null) {\n    if (stride > 0) {\n      // Forward iteration - use the first element. These values will get\n      // clamped below (Note: We could have set them to 0 and axis_size-1, but\n      // use lowest() and max() to maintain symmetry with StopForAxis())\n      start = Number.MIN_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the last element.\n      start = Number.MAX_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (start < 0) {\n    start += axisSize;\n  }\n\n  // Clamping\n  start = util.clamp(0, start, axisSize - 1);\n\n  return start;\n}\n\nexport function stopForAxis(\n    endMask: number, stopIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let stop = stopIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or if the stop index is not\n  // set for this axis.\n  if (endMask & (1 << axis) || ellipsisMask & (1 << axis) || stop == null) {\n    if (stride > 0) {\n      // Forward iteration - use the last element. These values will get\n      // clamped below\n      stop = Number.MAX_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the first element.\n      stop = Number.MIN_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (stop < 0) {\n    stop += axisSize;\n  }\n\n  // Clamping\n  // Because the end index points one past the last element, we need slightly\n  // different clamping ranges depending on the direction.\n  if (stride > 0) {\n    // Forward iteration\n    stop = util.clamp(0, stop, axisSize);\n  } else {\n    // Backward iteration\n    stop = util.clamp(-1, stop, axisSize - 1);\n  }\n\n  return stop;\n}\n\n/**\n * Returns true if the slice occupies a continous set of elements in the\n * 'flat' space.\n */\nexport function isSliceContinous(\n    shape: number[], begin: number[], size: number[]) {\n  // Index of the first axis that has size > 1.\n  let firstNonOneAxis = size.length;\n  for (let i = 0; i < size.length; i++) {\n    if (size[i] > 1) {\n      firstNonOneAxis = i;\n      break;\n    }\n  }\n\n  for (let i = firstNonOneAxis + 1; i < size.length; i++) {\n    if (begin[i] > 0 || size[i] !== shape[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function computeFlatOffset(begin: number[], strides: number[]): number {\n  let flatOffset = begin.length > 0 ? begin[begin.length - 1] : 1;\n  for (let i = 0; i < begin.length - 1; i++) {\n    flatOffset += begin[i] * strides[i];\n  }\n  return flatOffset;\n}\n\nexport function parseSliceParams(\n    x: TensorInfo, begin: number|number[], size?: number|number[]) {\n  // The following logic allows for more ergonomic calls.\n  let begin_: number[];\n  const xRank = x.shape.length;\n  if (typeof begin === 'number') {\n    begin_ = [begin, ...new Array(xRank - 1).fill(0)];\n  } else if (begin.length < xRank) {\n    begin_ = begin.concat(new Array(xRank - begin.length).fill(0));\n  } else {\n    begin_ = begin.slice();\n  }\n  begin_.forEach(d => {\n    util.assert(\n        d !== -1, () => 'slice() does not support negative begin indexing.');\n  });\n  let size_: number[];\n  if (size == null) {\n    size_ = new Array(xRank).fill(-1);\n  } else if (typeof size === 'number') {\n    size_ = [size, ...new Array(xRank - 1).fill(-1)];\n  } else if (size.length < xRank) {\n    size_ = size.concat(new Array(xRank - size.length).fill(-1));\n  } else {\n    size_ = size;\n  }\n  size_ = size_.map((d, i) => {\n    if (d >= 0) {\n      return d;\n    } else {\n      util.assert(\n          d === -1,\n          () => `Negative size values should be exactly -1 but got ` +\n              `${d} for the slice() size at index ${i}.`);\n      return x.shape[i] - begin_[i];\n    }\n  });\n  return [begin_, size_];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from './util';\n\n/**\n * Types to support JSON-esque data structures internally.\n *\n * Internally ConfigDict's use camelCase keys and values where the\n * values are class names to be instantiated.  On the python side, these\n * will be snake_case.  Internally we allow Enums into the values for better\n * type safety, but these need to be converted to raw primitives (usually\n * strings) for round-tripping with python.\n *\n * toConfig returns the TS-friendly representation. model.toJSON() returns\n * the pythonic version as that's the portable format.  If you need to\n * python-ify a non-model level toConfig output, you'll need to use a\n * convertTsToPythonic from serialization_utils in -Layers.\n *\n */\nexport declare type ConfigDictValue =\n    boolean | number | string | null | ConfigDictArray | ConfigDict;\nexport declare interface ConfigDict {\n  [key: string]: ConfigDictValue;\n}\nexport declare interface ConfigDictArray extends Array<ConfigDictValue> {}\n\n/**\n * Type to represent the class-type of Serializable objects.\n *\n * Ie the class prototype with access to the constructor and any\n * static members/methods. Instance methods are not listed here.\n *\n * Source for this idea: https://stackoverflow.com/a/43607255\n */\nexport declare type SerializableConstructor<T extends Serializable> = {\n  // tslint:disable-next-line:no-any\n  new (...args: any[]): T; className: string; fromConfig: FromConfigMethod<T>;\n};\nexport declare type FromConfigMethod<T extends Serializable> =\n    (cls: SerializableConstructor<T>, config: ConfigDict) => T;\n\n/**\n * Serializable defines the serialization contract.\n *\n * TFJS requires serializable classes to return their className when asked\n * to avoid issues with minification.\n */\nexport abstract class Serializable {\n  /**\n   * Return the class name for this class to use in serialization contexts.\n   *\n   * Generally speaking this will be the same thing that constructor.name\n   * would have returned.  However, the class name needs to be robust\n   * against minification for serialization/deserialization to work properly.\n   *\n   * There's also places such as initializers.VarianceScaling, where\n   * implementation details between different languages led to different\n   * class hierarchies and a non-leaf node is used for serialization purposes.\n   */\n  getClassName(): string {\n    return (this.constructor as SerializableConstructor<Serializable>)\n        .className;\n  }\n\n  /**\n   * Return all the non-weight state needed to serialize this object.\n   */\n  abstract getConfig(): ConfigDict;\n\n  /**\n   * Creates an instance of T from a ConfigDict.\n   *\n   * This works for most descendants of serializable.  A few need to\n   * provide special handling.\n   * @param cls A Constructor for the class to instantiate.\n   * @param config The Configuration for the object.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config);\n  }\n}\n\n/**\n * Maps string keys to class constructors.\n *\n * Used during (de)serialization from the cross-language JSON format, which\n * requires the class name in the serialization format matches the class\n * names as used in Python, should it exist.\n */\nexport class SerializationMap {\n  private static instance: SerializationMap;\n  classNameMap: {\n    [className: string]:\n        [SerializableConstructor<Serializable>, FromConfigMethod<Serializable>]\n  };\n\n  private constructor() {\n    this.classNameMap = {};\n  }\n\n  /**\n   * Returns the singleton instance of the map.\n   */\n  static getMap(): SerializationMap {\n    if (SerializationMap.instance == null) {\n      SerializationMap.instance = new SerializationMap();\n    }\n    return SerializationMap.instance;\n  }\n\n  /**\n   * Registers the class as serializable.\n   */\n  static register<T extends Serializable>(cls: SerializableConstructor<T>) {\n    SerializationMap.getMap().classNameMap[cls.className] =\n        [cls, cls.fromConfig];\n  }\n}\n\n/**\n * Register a class with the serialization map of TensorFlow.js.\n *\n * This is often used for registering custom Layers, so they can be\n * serialized and deserialized.\n *\n * Example:\n *\n * ```js\n * class MyCustomLayer extends tf.layers.Layer {\n *   static className = 'MyCustomLayer';\n *\n *   constructor(config) {\n *     super(config);\n *   }\n * }\n * tf.serialization.registerClass(MyCustomLayer);\n * ```\n *\n * @param cls The class to be registered. It must have a public static member\n *   called `className` defined and the value must be a non-empty string.\n *\n * @doc {heading: 'Models', subheading: 'Serialization', ignoreCI: true}\n */\nexport function registerClass<T extends Serializable>(\n    cls: SerializableConstructor<T>) {\n  assert(\n      cls.className != null,\n      () => `Class being registered does not have the static className ` +\n          `property defined.`);\n  assert(\n      typeof cls.className === 'string',\n      () => `className is required to be a string, but got type ` +\n          typeof cls.className);\n  assert(\n      cls.className.length > 0,\n      () => `Class being registered has an empty-string as its className, ` +\n          `which is disallowed.`);\n\n  SerializationMap.register(cls);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {inferShape} from './tensor_util_env';\nimport {RecursiveArray, TensorLike, TypedArray} from './types';\nimport {arraysEqual, flatten, isString, isTypedArray} from './util';\n\nconst TEST_EPSILON_FLOAT32 = 1e-3;\nexport const TEST_EPSILON_FLOAT16 = 1e-1;\n\nexport function expectArraysClose(\n    actual: TypedArray|number|RecursiveArray<number>,\n    expected: TypedArray|number|RecursiveArray<number>, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, epsilon));\n}\n\nexport function testEpsilon() {\n  return ENGINE.backend.floatPrecision() === 32 ? TEST_EPSILON_FLOAT32 :\n                                                  TEST_EPSILON_FLOAT16;\n}\n\nfunction expectArraysPredicate(\n    actual: TensorLike, expected: TensorLike,\n    predicate: (a: {}, b: {}) => boolean) {\n  let checkClassType = true;\n  if (isTypedArray(actual) || isTypedArray(expected)) {\n    checkClassType = false;\n  }\n  if (isTypedArray(actual) && isTypedArray(expected)) {\n    checkClassType = true;\n  }\n  if (checkClassType) {\n    const aType = actual.constructor.name;\n    const bType = expected.constructor.name;\n\n    if (aType !== bType) {\n      throw new Error(\n          `Arrays are of different type. Actual: ${aType}. ` +\n          `Expected: ${bType}`);\n    }\n  }\n\n  if (Array.isArray(actual) && Array.isArray(expected)) {\n    const actualShape = inferShape(actual);\n    const expectedShape = inferShape(expected);\n    if (!arraysEqual(actualShape, expectedShape)) {\n      throw new Error(\n          `Arrays have different shapes. ` +\n          `Actual: [${actualShape}]. Expected: [${expectedShape}]`);\n    }\n  }\n\n  const actualFlat =\n      isTypedArray(actual) ? actual : flatten(actual as RecursiveArray<number>);\n  const expectedFlat = isTypedArray(expected) ?\n      expected :\n      flatten(expected as RecursiveArray<number>);\n\n  if (actualFlat.length !== expectedFlat.length) {\n    throw new Error(\n        `Arrays have different lengths actual: ${actualFlat.length} vs ` +\n        `expected: ${expectedFlat.length}.\\n` +\n        `Actual:   ${actualFlat}.\\n` +\n        `Expected: ${expectedFlat}.`);\n  }\n  for (let i = 0; i < expectedFlat.length; ++i) {\n    const a = actualFlat[i];\n    const e = expectedFlat[i];\n\n    if (!predicate(a, e)) {\n      throw new Error(\n          `Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${e}.\\n` +\n          `Actual:   ${actualFlat}.\\n` +\n          `Expected: ${expectedFlat}.`);\n    }\n  }\n}\n\nexport interface DoneFn {\n  (): void;\n  fail: (message?: Error|string) => void;\n}\n\nexport function expectPromiseToFail(fn: () => Promise<{}>, done: DoneFn): void {\n  fn().then(() => done.fail(), () => done());\n}\n\nexport function expectArraysEqual(actual: TensorLike, expected: TensorLike) {\n  const exp = typeof expected === 'string' || typeof expected === 'number' ||\n          typeof expected === 'boolean' ?\n      [expected] as number[] :\n      expected as number[];\n  if (isString(actual) || isString((actual as string[])[0]) ||\n      isString(expected) || isString((expected as string[])[0])) {\n    // tslint:disable-next-line: triple-equals\n    return expectArraysPredicate(actual, exp, (a, b) => a == b);\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, 0));\n}\n\nexport function expectNumbersClose(a: number, e: number, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  if (!areClose(a, e, epsilon)) {\n    throw new Error(`Numbers differ: actual === ${a}, expected === ${e}`);\n  }\n}\n\nfunction areClose(a: number, e: number, epsilon: number): boolean {\n  if (!isFinite(a) && !isFinite(e)) {\n    return true;\n  }\n  if (isNaN(a) || isNaN(e) || Math.abs(a - e) > epsilon) {\n    return false;\n  }\n  return true;\n}\n\nexport function expectValuesInRange(\n    actual: TypedArray|number[], low: number, high: number) {\n  for (let i = 0; i < actual.length; i++) {\n    if (actual[i] < low || actual[i] > high) {\n      throw new Error(\n          `Value out of range:${actual[i]} low: ${low}, high: ${high}`);\n    }\n  }\n}\n\nexport function expectArrayBuffersEqual(\n    actual: ArrayBuffer, expected: ArrayBuffer) {\n  // Safari & Jasmine don't like comparing ArrayBuffers directly. Wrapping in\n  // a Float32Array solves this issue.\n  expect(new Float32Array(actual)).toEqual(new Float32Array(expected));\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from './backends/backend';\nimport {ENGINE, Engine, MemoryInfo, ProfileInfo, ScopeFn, TimingInfo} from './engine';\nimport {env} from './environment';\n\nimport {Platform} from './platforms/platform';\nimport {setDeprecationWarningFn, Tensor} from './tensor';\nimport {TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\n\n/**\n * Enables production mode which disables correctness checks in favor of\n * performance.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableProdMode(): void {\n  env().set('PROD', true);\n}\n\n/**\n * Enables debug mode which will log information about all executed kernels:\n * the elapsed time of the kernel execution, as well as the rank, shape, and\n * size of the output tensor.\n *\n * Debug mode will significantly slow down your application as it will\n * download the result of every operation to the CPU. This should not be used in\n * production. Debug mode does not affect the timing information of the kernel\n * execution as we do not measure download time in the kernel execution time.\n *\n * See also: `tf.profile`, `tf.memory`.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableDebugMode(): void {\n  env().set('DEBUG', true);\n}\n\n/** Globally disables deprecation warnings */\nexport function disableDeprecationWarnings(): void {\n  env().set('DEPRECATION_WARNINGS_ENABLED', false);\n  console.warn(`TensorFlow.js deprecation warnings have been disabled.`);\n}\n\n/** Warn users about deprecated functionality. */\nexport function deprecationWarn(msg: string) {\n  if (env().getBool('DEPRECATION_WARNINGS_ENABLED')) {\n    console.warn(\n        msg + ' You can disable deprecation warnings with ' +\n        'tf.disableDeprecationWarnings().');\n  }\n}\nsetDeprecationWarningFn(deprecationWarn);\n\n/**\n * Dispose all variables kept in backend engine.\n *\n * @doc {heading: 'Environment'}\n */\nexport function disposeVariables(): void {\n  ENGINE.disposeVariables();\n}\n\n/**\n * It returns the global engine that keeps track of all tensors and backends.\n *\n * @doc {heading: 'Environment'}\n */\nexport function engine(): Engine {\n  return ENGINE;\n}\n\n/**\n * Returns memory info at the current time in the program. The result is an\n * object with the following properties:\n *\n * - `numBytes`: Number of bytes allocated (undisposed) at this time.\n * - `numTensors`: Number of unique tensors allocated.\n * - `numDataBuffers`: Number of unique data buffers allocated\n *   (undisposed) at this time, which is \u2264 the number of tensors\n *   (e.g. `a.reshape(newShape)` makes a new Tensor that shares the same\n *   data buffer with `a`).\n * - `unreliable`: True if the memory usage is unreliable. See `reasons` when\n *    `unreliable` is true.\n * - `reasons`: `string[]`, reasons why the memory is unreliable, present if\n *    `unreliable` is true.\n *\n * WebGL Properties:\n * - `numBytesInGPU`: Number of bytes allocated (undisposed) in the GPU only at\n *     this time.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function memory(): MemoryInfo {\n  return ENGINE.memory();\n}\n\n/**\n * Executes the provided function `f()` and returns a promise that resolves\n * with information about the function's memory use:\n * - `newBytes`: the number of new bytes allocated\n * - `newTensors`: the number of new tensors created\n * - `peakBytes`: the peak number of bytes allocated\n * - `kernels`: an array of objects for each kernel involved that reports\n * their input and output shapes, number of bytes used, and number of new\n * tensors created.\n *\n * ```js\n * const profile = await tf.profile(() => {\n *   const x = tf.tensor1d([1, 2, 3]);\n *   let x2 = x.square();\n *   x2.dispose();\n *   x2 = x.square();\n *   x2.dispose();\n *   return x;\n * });\n *\n * console.log(`newBytes: ${profile.newBytes}`);\n * console.log(`newTensors: ${profile.newTensors}`);\n * console.log(`byte usage over all kernels: ${profile.kernels.map(k =>\n * k.totalBytesSnapshot)}`);\n * ```\n *\n *\n * @doc {heading: 'Performance', subheading: 'Profile'}\n */\nexport function profile(f: () => (TensorContainer | Promise<TensorContainer>)):\n    Promise<ProfileInfo> {\n  return ENGINE.profile(f);\n}\n\n/**\n * Executes the provided function `fn` and after it is executed, cleans up all\n * intermediate tensors allocated by `fn` except those returned by `fn`.\n * `fn` must not return a Promise (async functions not allowed). The returned\n * result can be a complex object.\n *\n * Using this method helps avoid memory leaks. In general, wrap calls to\n * operations in `tf.tidy` for automatic memory cleanup.\n *\n * NOTE: Variables do *not* get cleaned up when inside a tidy(). If you want to\n * dispose variables, please use `tf.disposeVariables` or call dispose()\n * directly on variables.\n *\n * ```js\n * // y = 2 ^ 2 + 1\n * const y = tf.tidy(() => {\n *   // a, b, and one will be cleaned up when the tidy ends.\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *   const b = a.square();\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * y.print();\n * ```\n *\n * @param nameOrFn The name of the closure, or the function to execute.\n *     If a name is provided, the 2nd argument should be the function.\n *     If debug mode is on, the timing and the memory usage of the function\n *     will be tracked and displayed on the console using the provided name.\n * @param fn The function to execute.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function tidy<T extends TensorContainer>(\n    nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>): T {\n  return ENGINE.tidy(nameOrFn, fn);\n}\n\n/**\n * Disposes any `tf.Tensor`s found within the provided object.\n *\n * @param container an object that may be a `tf.Tensor` or may directly\n *     contain `tf.Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. If\n *     the object is not a `tf.Tensor` or does not contain `Tensors`, nothing\n *     happens. In general it is safe to pass any object here, except that\n *     `Promise`s are not supported.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function dispose(container: TensorContainer) {\n  const tensors = getTensorsInContainer(container);\n  tensors.forEach(tensor => tensor.dispose());\n}\n\n/**\n * Keeps a `tf.Tensor` generated inside a `tf.tidy` from being disposed\n * automatically.\n *\n * ```js\n * let b;\n * const y = tf.tidy(() => {\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *\n *   // b will not be cleaned up by the tidy. a and one will be cleaned up\n *   // when the tidy ends.\n *   b = tf.keep(a.square());\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * console.log('y:');\n * y.print();\n * console.log('b:');\n * b.print();\n * ```\n *\n * @param result The tensor to keep from being disposed.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function keep<T extends Tensor>(result: T): T {\n  return ENGINE.keep(result);\n}\n\n/**\n * Executes `f()` and returns a promise that resolves with timing\n * information.\n *\n * The result is an object with the following properties:\n *\n * - `wallMs`: Wall execution time.\n * - `kernelMs`: Kernel execution time, ignoring data transfer. If using the\n * WebGL backend and the query timer extension is not available, this will\n * return an error object.\n * - On `WebGL` The following additional properties exist:\n *   - `uploadWaitMs`: CPU blocking time on texture uploads.\n *   - `downloadWaitMs`: CPU blocking time on texture downloads (readPixels).\n *\n * ```js\n * const x = tf.randomNormal([20, 20]);\n * const time = await tf.time(() => x.matMul(x));\n *\n * console.log(`kernelMs: ${time.kernelMs}, wallTimeMs: ${time.wallMs}`);\n * ```\n *\n * @param f The function to execute and time.\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nexport function time(f: () => void): Promise<TimingInfo> {\n  return ENGINE.time(f);\n}\n\n/**\n * Sets the backend (cpu, webgl, wasm, etc) responsible for creating tensors and\n * executing operations on those tensors. Returns a promise that resolves\n * to a boolean if the backend initialization was successful.\n *\n * Note this disposes the current backend, if any, as well as any tensors\n * associated with it. A new backend is initialized, even if it is of the\n * same type as the previous one.\n *\n * @param backendName The name of the backend. Currently supports\n *     `'webgl'|'cpu'` in the browser, `'tensorflow'` under node.js\n *     (requires tfjs-node), and `'wasm'` (requires tfjs-backend-wasm).\n *\n * @doc {heading: 'Backends'}\n */\nexport function setBackend(backendName: string): Promise<boolean> {\n  return ENGINE.setBackend(backendName);\n}\n\n/**\n * Returns a promise that resolves when the currently selected backend (or the\n * highest priority one) has initialized. Await this promise when you are using\n * a backend that has async initialization.\n *\n * @doc {heading: 'Backends'}\n */\nexport function ready(): Promise<void> {\n  return ENGINE.ready();\n}\n\n/**\n * Returns the current backend name (cpu, webgl, etc). The backend is\n * responsible for creating tensors and executing operations on those tensors.\n *\n * @doc {heading: 'Backends'}\n */\nexport function getBackend(): string {\n  return ENGINE.backendName;\n}\n\n/**\n * Removes a backend and the registered factory.\n *\n * @doc {heading: 'Backends'}\n */\nexport function removeBackend(name: string): void {\n  ENGINE.removeBackend(name);\n}\n\n/**\n * Finds the backend registered under the provided name. Returns null if the\n * name is not in the registry, or the registration hasn't finished yet.\n */\nexport function findBackend(name: string): KernelBackend {\n  return ENGINE.findBackend(name);\n}\n\n/**\n * Finds the backend factory registered under the provided name. Returns a\n * function that produces a new backend when called. Returns null if the name\n * is not in the registry.\n */\nexport function findBackendFactory(name: string): () =>\n    KernelBackend | Promise<KernelBackend> {\n  return ENGINE.findBackendFactory(name);\n}\n\n/**\n * Registers a global backend. The registration should happen when importing\n * a module file (e.g. when importing `backend_webgl.ts`), and is used for\n * modular builds (e.g. custom tfjs bundle with only webgl support).\n *\n * @param factory The backend factory function. When called, it should\n * return a backend instance, or a promise of an instance.\n * @param priority The priority of the backend (higher = more important).\n *     In case multiple backends are registered, the priority is used to find\n *     the best backend. Defaults to 1.\n * @return False if there is already a registered backend under this name, true\n *     if not.\n *\n * @doc {heading: 'Backends'}\n */\nexport function registerBackend(\n    name: string, factory: () => KernelBackend | Promise<KernelBackend>,\n    priority = 1): boolean {\n  return ENGINE.registerBackend(name, factory, priority);\n}\n\n/**\n * Gets the current backend. If no backends have been initialized, this will\n * attempt to initialize the best backend. Will throw an error if the highest\n * priority backend has async initialization, in which case, you should call\n * 'await tf.ready()' before running other code.\n *\n * @doc {heading: 'Backends'}\n */\nexport function backend(): KernelBackend {\n  return ENGINE.backend;\n}\n\n/**\n * Sets the global platform.\n *\n * @param platformName The name of this platform.\n * @param platform A platform implementation.\n */\nexport function setPlatform(platformName: string, platform: Platform) {\n  env().setPlatform(platformName, platform);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Add, AddInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n *\n * ```js\n * // Broadcast add a with b.\n * const a = tf.scalar(5);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n * @param a The first `tf.Tensor` to add.\n * @param b The second `tf.Tensor` to add. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction add_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'add');\n  let $b = convertToTensor(b, 'b', 'add');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.add($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: AddInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Add) as T;\n}\n\nexport const add = op({add_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FloorDiv, FloorDivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n * The result is rounded with floor function.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.floorDiv(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.floorDiv(b).print();  // or tf.floorDiv(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction floorDiv_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'floorDiv');\n  let $b = convertToTensor(b, 'b', 'floorDiv');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.floorDiv($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: FloorDivInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             FloorDiv) as T;\n}\n\nexport const floorDiv = op({floorDiv_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Div, DivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {floorDiv} from './floorDiv';\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction div_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'int32' && $b.dtype === 'int32') {\n    return floorDiv($a, $b);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.realDivide($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: DivInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Div,\n             attrs) as T;\n}\n\nexport const div = op({div_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Multiply, MultiplyInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Multiplies two `tf.Tensor`s element-wise, A * B. Supports broadcasting.\n *\n * We also expose `tf.mulStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([2, 3, 4, 5]);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n *\n * ```js\n * // Broadcast mul a with b.\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.scalar(5);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n * @param a The first tensor to multiply.\n * @param b The second tensor to multiply. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mul_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mul');\n  let $b = convertToTensor(b, 'b', 'mul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.multiply($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MultiplyInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Multiply) as T;\n}\nexport const mul = op({mul_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Abs, AbsInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes absolute value element-wise: `abs(x)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.abs().print();  // or tf.abs(x)\n * ```\n * @param x The input `tf.Tensor`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction abs_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'abs');\n\n  const inputs: AbsInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    if ($x.dtype === 'complex64') {\n      return backend.complexAbs($x);\n    }\n\n    return backend.abs($x);\n  }, inputs as {} as NamedTensorMap, null /* grad */, Abs);\n}\n\nexport const abs = op({abs_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Acos, AcosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes acos of the input `tf.Tensor` element-wise: `acos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.acos().print();  // or tf.acos(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acos');\n  const inputs: AcosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acos);\n}\nexport const acos = op({acos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Acosh, AcoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the inverse hyperbolic cos of the input `tf.Tensor` element-wise:\n * `acosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([10, 1, 3, 5.7]);\n *\n * x.acosh().print();  // or tf.acosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acosh');\n  const inputs: AcoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acosh);\n}\nexport const acosh = op({acosh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AddN, AddNInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Adds a list of `tf.Tensor`s element-wise, each with the same shape and dtype.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n *\n * tf.addN([a, b, c]).print();\n * ```\n * @param tensors A list of tensors with the same shape and dtype.\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction addN_<T extends Tensor>(tensors: Array<T|TensorLike>): T {\n  util.assert(\n      Array.isArray(tensors),\n      () => 'The argument passed to tf.addN() must be a list of tensors');\n  util.assert(\n      tensors.length >= 1,\n      () => `Must pass at least one tensor to tf.addN(), but got ` +\n          `${tensors.length}`);\n\n  const $tensors =\n      tensors.map((t, i) => convertToTensor(t, `tensors${i}`, 'addN'));\n\n  const firstTensor = $tensors[0];\n  $tensors.forEach(t => {\n    if (t.dtype !== firstTensor.dtype) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same dtype');\n    }\n  });\n\n  $tensors.forEach(t => {\n    if (!util.arraysEqual(t.shape, firstTensor.shape)) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same shape');\n    }\n  });\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.addN($tensors);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: AddNInputs = $tensors;\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, AddN) as\n      T;\n}\n\nexport const addN = op({addN_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\n/**\n * Returns true if the axis specifies the inner most dimensions of the\n * array.\n */\nexport function axesAreInnerMostDims(axes: number[], rank: number): boolean {\n  for (let i = 0; i < axes.length; ++i) {\n    if (axes[axes.length - i - 1] !== rank - 1 - i) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function combineLocations(\n    outputLoc: number[], reduceLoc: number[], axes: number[]): number[] {\n  const rank = outputLoc.length + reduceLoc.length;\n  const loc = [];\n  let outIdx = 0;\n  let reduceIdx = 0;\n  \u00A0 for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      loc.push(outputLoc[outIdx++]);\n    } else {\n      loc.push(reduceLoc[reduceIdx++]);\n    }\n  }\n  return loc;\n}\n\nexport function computeOutAndReduceShapes(\n    aShape: number[], axes: number[]): [number[], number[]] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      outShape.push(aShape[dim]);\n    }\n  }\n  const reduceShape = axes.map(dim => aShape[dim]);\n  return [outShape, reduceShape];\n}\n\nexport function expandShapeToKeepDim(\n    shape: number[], axes: number[]): number[] {\n  const reduceSubShape = axes.map(x => 1);\n  return combineLocations(shape, reduceSubShape, axes);\n}\n\nexport function assertAxesAreInnerMostDims(\n    msg: string, axes: number[], rank: number): void {\n  util.assert(\n      axesAreInnerMostDims(axes, rank),\n      () => `${msg} supports only inner-most axes for now. ` +\n          `Got axes ${axes} and rank-${rank} input.`);\n}\n\n/**\n * Returns the axes permutation to be used with `tf.transpose`, if such\n * permutation is necessary. Otherwise it returns null. This method is used by\n * operations that operate only on inner-most axes.\n */\nexport function getAxesPermutation(axes: number[], rank: number): number[]|\n    null {\n  if (axesAreInnerMostDims(axes, rank)) {\n    return null;\n  }\n  const result: number[] = [];\n  for (let i = 0; i < rank; ++i) {\n    if (axes.indexOf(i) === -1) {\n      result.push(i);\n    }\n  }\n  axes.forEach(axis => result.push(axis));\n  return result;\n}\n\n/** Returns the axes permutation that undoes the original permutation. */\nexport function getUndoAxesPermutation(axes: number[]): number[] {\n  return axes.map((axis, i) => [i, axis])\n      .sort((a, b) => a[1] - b[1])\n      .map(x => x[0]);\n}\n\nexport function getInnerMostAxes(numAxes: number, rank: number): number[] {\n  const res: number[] = [];\n  for (let i = rank - numAxes; i < rank; ++i) {\n    res.push(i);\n  }\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {All, AllAttrs, AllInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical and of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.all().print();  // or tf.all(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.all(axis).print();  // or tf.all(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction all_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'all', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.all($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AllInputs = {x: $x};\n  const attrs: AllAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, All,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const all = op({all_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Any, AnyAttrs, AnyInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical or of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.any().print();  // or tf.any(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.any(axis).print();  // or tf.any(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction any_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'any', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.any($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AnyInputs = {x: $x};\n  const attrs: AnyAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Any,\n             attrs as {} as NamedAttrMap) as T;\n}\n\n// tslint:disable-next-line:variable-name\nexport const any = op({any_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMax, ArgMaxAttrs, ArgMaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the maximum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMax().print();  // or tf.argMax(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMax(axis).print();  // or tf.argMax(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMax_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMax');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMax($x, axes[0]);\n  };\n\n  const inputs: ArgMaxInputs = {x: $x};\n  const attrs: ArgMaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMax,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMax = op({argMax_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMin, ArgMinAttrs, ArgMinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the minimum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMin().print();  // or tf.argMin(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMin(axis).print();  // or tf.argMin(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMin_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMin');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if (axis == null) {\n      axis = 0;\n    }\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMin($x, axes[0]);\n  };\n\n  const inputs: ArgMinInputs = {x: $x};\n  const attrs: ArgMinAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMin,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMin = op({argMin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asin, AsinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes asin of the input `tf.Tensor` element-wise: `asin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asin().print();  // or tf.asin(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asin');\n  const inputs: AsinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asin);\n}\nexport const asin = op({asin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asinh, AsinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic sin of the input `tf.Tensor` element-wise:\n * `asinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asinh().print();  // or tf.asinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asinh');\n\n  const inputs: AsinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asinh);\n}\nexport const asinh = op({asinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atan, AtanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes atan of the input `tf.Tensor` element-wise: `atan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.atan().print();  // or tf.atan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atan');\n\n  const inputs: AtanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atan);\n}\nexport const atan = op({atan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Atan2, Atan2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes arctangent of `tf.Tensor`s a / b element-wise: `atan2(a, b)`.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1.0, 1.0, -1.0, .7]);\n * const b = tf.tensor1d([2.0, 13.0, 3.5, .21]);\n *\n * tf.atan2(a, b).print()\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan2_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'atan2');\n  let $b = convertToTensor(b, 'b', 'atan2');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.atan2($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: Atan2Inputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Atan2) as T;\n}\n\nexport const atan2 = op({atan2_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atanh, AtanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic tan of the input `tf.Tensor` element-wise:\n * `atanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.atanh().print();  // or tf.atanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atanh');\n\n  const inputs: AtanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atanh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atanh);\n}\nexport const atanh = op({atanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\ntype PadType = 'SAME'|'VALID'|'NUMBER'|'EXPLICIT';\n\n// For NHWC should be in the following form:\n//  [[0, 0], [pad_top,pad_bottom], [pad_left, pad_right], [0, 0]]\n// For NCHW should be in the following form:\n//  [[0, 0], [0, 0], [pad_top,pad_bottom], [pad_left, pad_right]]\n// Reference: https://www.tensorflow.org/api_docs/python/tf/nn/conv2d\nexport type ExplicitPadding =\n    [[number, number], [number, number], [number, number], [number, number]];\n\nexport type PadInfo = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  type: PadType\n};\n\nexport type PadInfo3D = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  front: number,\n  back: number,\n  type: PadType\n};\n\n/**\n * Information about the forward pass of a convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv2DInfo = {\n  batchSize: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideHeight: number,\n  strideWidth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo,\n  inShape: [number, number, number, number],\n  outShape: [number, number, number, number],\n  filterShape: [number, number, number, number]\n};\n\n/**\n *\n * @param inputShape Input tensor shape is of the following dimensions:\n *     `[batch, height, width, inChannels]`.\n * @param filterShape The filter shape is of the following dimensions:\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat The data format of the input and output data.\n *     Defaults to 'NHWC'.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.\n *     Defaults to `[1, 1]`. If `dilations` is a single number, then\n *     `dilationHeight == dilationWidth`.\n */\nexport function computeDilation2DInfo(\n    inputShape: [number, number, number, number],\n    filterShape: [number, number, number], strides: number|[number, number],\n    pad: 'same'|'valid'|number, dataFormat: 'NHWC' = 'NHWC',\n    dilations: number|[number, number]) {\n  // `computerConv2DInfo` require filterShape to be in the dimension of:\n  // `[filterHeight, filterWidth, depth, outDepth]`, dilation2d doesn't have\n  // outDepth, it should have the same depth as the input.\n  // Input shape: [batch, height, width, inChannels]\n  const inputChannels = inputShape[3];\n  const $filterShape =\n      [...filterShape, inputChannels] as [number, number, number, number];\n  const $dataFormat = convertConv2DDataFormat(dataFormat);\n\n  return computeConv2DInfo(\n      inputShape, $filterShape, strides, dilations, pad,\n      null /* roundingMode */, null /* depthWise */, $dataFormat);\n}\n\nexport function computePool2DInfo(\n    inShape: [number, number, number, number],\n    filterSize: [number, number]|number, strides: number|[number, number],\n    dilations: number|[number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  const [filterHeight, filterWidth] = parseTupleParam(filterSize);\n\n  let filterShape: [number, number, number, number];\n  if (dataFormat === 'channelsLast') {\n    filterShape = [filterHeight, filterWidth, inShape[3], inShape[3]];\n  } else if (dataFormat === 'channelsFirst') {\n    filterShape = [filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv2DInfo(\n      inShape, filterShape, strides, dilations, pad, roundingMode, false,\n      dataFormat);\n}\n\n/**\n * Computes the information for a forward pass of a pooling3D operation.\n */\nexport function computePool3DInfo(\n    inShape: [number, number, number, number, number],\n    filterSize: number|[number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC'): Conv3DInfo {\n  const [filterDepth, filterHeight, filterWidth] = parse3TupleParam(filterSize);\n\n  let filterShape: [number, number, number, number, number];\n  let $dataFormat: 'channelsFirst'|'channelsLast';\n  if (dataFormat === 'NDHWC') {\n    $dataFormat = 'channelsLast';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[4], inShape[4]];\n  } else if (dataFormat === 'NCDHW') {\n    $dataFormat = 'channelsFirst';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv3DInfo(\n      inShape, filterShape, strides, dilations, pad, false, $dataFormat,\n      roundingMode);\n}\n\n/**\n * Computes the information for a forward pass of a convolution/pooling\n * operation.\n */\nexport function computeConv2DInfo(\n    inShape: [number, number, number, number],\n    filterShape: [number, number, number, number],\n    strides: number|[number, number], dilations: number|[number, number],\n    pad: 'same'|'valid'|number|ExplicitPadding,\n    roundingMode?: 'floor'|'round'|'ceil', depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  let [batchSize, inHeight, inWidth, inChannels] = [-1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterHeight, filterWidth, , filterChannels] = filterShape;\n  const [strideHeight, strideWidth] = parseTupleParam(strides);\n  const [dilationHeight, dilationWidth] = parseTupleParam(dilations);\n\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outHeight, outWidth} = getPadAndOutInfo(\n      pad, inHeight, inWidth, strideHeight, strideWidth, effectiveFilterHeight,\n      effectiveFilterWidth, roundingMode, dataFormat);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inHeight,\n    inWidth,\n    inChannels,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideHeight,\n    strideWidth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\n/**\n * Information about the forward pass of a 3D convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv3DInfo = {\n  batchSize: number,\n  inDepth: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideDepth: number,\n  strideHeight: number,\n  strideWidth: number,\n  dilationDepth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterDepth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterDepth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo3D,\n  inShape: [number, number, number, number, number],\n  outShape: [number, number, number, number, number],\n  filterShape: [number, number, number, number, number]\n};\n\n/**\n * Computes the information for a forward pass of a 3D convolution/pooling\n * operation.\n */\nexport function computeConv3DInfo(\n    inShape: [number, number, number, number, number],\n    filterShape: [number, number, number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast',\n    roundingMode?: 'floor'|'round'|'ceil'): Conv3DInfo {\n  let [batchSize, inDepth, inHeight, inWidth, inChannels] =\n      [-1, -1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inDepth, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inDepth, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterDepth, filterHeight, filterWidth, , filterChannels] =\n      filterShape;\n  const [strideDepth, strideHeight, strideWidth] = parse3TupleParam(strides);\n  const [dilationDepth, dilationHeight, dilationWidth] =\n      parse3TupleParam(dilations);\n\n  const effectiveFilterDepth =\n      getEffectiveFilterSize(filterDepth, dilationDepth);\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outDepth, outHeight, outWidth} = get3DPadAndOutInfo(\n      pad, inDepth, inHeight, inWidth, strideDepth, strideHeight, strideWidth,\n      effectiveFilterDepth, effectiveFilterHeight, effectiveFilterWidth,\n      roundingMode);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outDepth, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outDepth, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inDepth,\n    inHeight,\n    inWidth,\n    inChannels,\n    outDepth,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideDepth,\n    strideHeight,\n    strideWidth,\n    filterDepth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterDepth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationDepth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\nfunction computeOutputShape2D(\n    inShape: [number, number], fieldSize: number, stride: number,\n    zeroPad?: number, roundingMode?: 'floor'|'round'|'ceil'): [number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputRows = inShape[0];\n  const inputCols = inShape[1];\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputRows, outputCols];\n}\n\nfunction computeOutputShape4D(\n    inShape: [number, number, number, number], fieldSize: number,\n    outChannels: number, stride: number, zeroPad?: number,\n    roundingMode?: 'floor'|'round'|'ceil'): [number, number, number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputDepth = inShape[0];\n  const inputRows = inShape[1];\n  const inputCols = inShape[2];\n\n  const outputDepths = conditionalRound(\n      (inputDepth - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputDepths),\n      () => `The output # of depths (${outputDepths}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputDepths, outputRows, outputCols, outChannels];\n}\n\nexport function computeDefaultPad(\n    inputShape: [number, number]|[number, number, number, number],\n    fieldSize: number, stride: number, dilation = 1): number {\n  const effectiveFieldSize = getEffectiveFilterSize(fieldSize, dilation);\n  return Math.floor(\n      (inputShape[0] * (stride - 1) - stride + effectiveFieldSize) / 2);\n}\n\nfunction parseTupleParam(param: number|number[]): [number, number, number] {\n  if (typeof param === 'number') {\n    return [param, param, param];\n  }\n  if (param.length === 2) {\n    return [param[0], param[1], 1];\n  }\n  return param as [number, number, number];\n}\n\nfunction parse3TupleParam(param: number|[number, number, number]):\n    [number, number, number] {\n  return typeof param === 'number' ? [param, param, param] : param;\n}\n\n/* See https://www.tensorflow.org/api_docs/python/tf/nn/atrous_conv2d\n * Atrous convolution is equivalent to standard convolution with upsampled\n * filters with effective_filter_height =\n * filter_height + (filter_height - 1) * (dilation - 1)\n * and effective_filter_width =\n * filter_width + (filter_width - 1) * (dilation - 1),\n * produced by inserting dilation - 1 zeros along consecutive elements across\n * the filters' spatial dimensions.\n * When there is a dilation, this converts a filter dimension to the\n * effective filter dimension, so it can be used in a standard convolution.\n */\nfunction getEffectiveFilterSize(filterSize: number, dilation: number) {\n  if (dilation <= 1) {\n    return filterSize;\n  }\n\n  return filterSize + (filterSize - 1) * (dilation - 1);\n}\n\nfunction getPadAndOutInfo(\n    pad: 'same'|'valid'|number|ExplicitPadding, inHeight: number,\n    inWidth: number, strideHeight: number, strideWidth: number,\n    filterHeight: number, filterWidth: number,\n    roundingMode: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|\n    'channelsLast'): {padInfo: PadInfo, outHeight: number, outWidth: number} {\n  let padInfo: PadInfo;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {top: pad, bottom: pad, left: pad, right: pad, type: padType};\n    const outShape = computeOutputShape2D(\n        [inHeight, inWidth], filterHeight, strideHeight, pad, roundingMode);\n    outHeight = outShape[0];\n    outWidth = outShape[1];\n  } else if (pad === 'same') {\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongHeight =\n        Math.max(0, (outHeight - 1) * strideHeight + filterHeight - inHeight);\n    const padAlongWidth =\n        Math.max(0, (outWidth - 1) * strideWidth + filterWidth - inWidth);\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n    padInfo = {top, bottom, left, right, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {top: 0, bottom: 0, left: 0, right: 0, type: 'VALID'};\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else if (typeof pad === 'object') {\n    const top = dataFormat === 'channelsLast' ? pad[1][0] : pad[2][0];\n    const bottom = dataFormat === 'channelsLast' ? pad[1][1] : pad[2][1];\n    const left = dataFormat === 'channelsLast' ? pad[2][0] : pad[3][0];\n    const right = dataFormat === 'channelsLast' ? pad[2][1] : pad[3][1];\n    const padType = (top === 0 && bottom === 0 && left === 0 && right === 0) ?\n        'VALID' :\n        'EXPLICIT';\n    padInfo = {top, bottom, left, right, type: padType};\n    outHeight = conditionalRound(\n        (inHeight - filterHeight + top + bottom) / strideHeight + 1,\n        roundingMode);\n    outWidth = conditionalRound(\n        (inWidth - filterWidth + left + right) / strideWidth + 1, roundingMode);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outHeight, outWidth};\n}\n\nfunction get3DPadAndOutInfo(\n    pad: 'same'|'valid'|number, inDepth: number, inHeight: number,\n    inWidth: number, strideDepth: number, strideHeight: number,\n    strideWidth: number, filterDepth: number, filterHeight: number,\n    filterWidth: number, roundingMode?: 'floor'|'round'|'ceil'): {\n  padInfo: PadInfo3D,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number\n} {\n  let padInfo: PadInfo3D;\n  let outDepth: number;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {\n      top: pad,\n      bottom: pad,\n      left: pad,\n      right: pad,\n      front: pad,\n      back: pad,\n      type: padType\n    };\n    const outShape = computeOutputShape4D(\n        [inDepth, inHeight, inWidth, 1], filterDepth, 1, strideDepth, pad,\n        roundingMode);\n    outDepth = outShape[0];\n    outHeight = outShape[1];\n    outWidth = outShape[2];\n  } else if (pad === 'same') {\n    outDepth = Math.ceil(inDepth / strideDepth);\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongDepth = (outDepth - 1) * strideDepth + filterDepth - inDepth;\n    const padAlongHeight =\n        (outHeight - 1) * strideHeight + filterHeight - inHeight;\n    const padAlongWidth = (outWidth - 1) * strideWidth + filterWidth - inWidth;\n    const front = Math.floor(padAlongDepth / 2);\n    const back = padAlongDepth - front;\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n\n    padInfo = {top, bottom, left, right, front, back, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {\n      top: 0,\n      bottom: 0,\n      left: 0,\n      right: 0,\n      front: 0,\n      back: 0,\n      type: 'VALID'\n    };\n    outDepth = Math.ceil((inDepth - filterDepth + 1) / strideDepth);\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outDepth, outHeight, outWidth};\n}\n\n/**\n * Rounds a value depending on the rounding mode\n * @param value\n * @param roundingMode\n */\nfunction conditionalRound(\n    value: number, roundingMode?: 'floor'|'round'|'ceil') {\n  if (!roundingMode) {\n    return value;\n  }\n  switch (roundingMode) {\n    case 'round':\n      // used for Caffe Conv\n      return Math.round(value);\n    case 'ceil':\n      // used for Caffe Pool\n      return Math.ceil(value);\n    case 'floor':\n      return Math.floor(value);\n    default:\n      throw new Error(`Unknown roundingMode ${roundingMode}`);\n  }\n}\n\nexport function tupleValuesAreOne(param: number|number[]): boolean {\n  const [dimA, dimB, dimC] = parseTupleParam(param);\n  return dimA === 1 && dimB === 1 && dimC === 1;\n}\n\nexport function eitherStridesOrDilationsAreOne(\n    strides: number|number[], dilations: number|number[]): boolean {\n  return tupleValuesAreOne(strides) || tupleValuesAreOne(dilations);\n}\n\n/**\n * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to\n *    'channelsLast'|'channelsFirst'\n * @param dataFormat in 'NHWC'|'NCHW' mode\n * @return dataFormat in 'channelsLast'|'channelsFirst' mode\n * @throws unknown dataFormat\n */\nexport function convertConv2DDataFormat(dataFormat: 'NHWC'|'NCHW'):\n    'channelsLast'|'channelsFirst' {\n  if (dataFormat === 'NHWC') {\n    return 'channelsLast';\n  } else if (dataFormat === 'NCHW') {\n    return 'channelsFirst';\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D average pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction avgPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'avgPool', 'float32');\n  const dilations = 1;\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in avgPool: x must be rank 4 but got rank ${x4D.rank}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    save([x4D]);\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      return x4D.clone();\n    }\n\n    return backend.avgPool(x4D, convInfo);\n  };\n\n  const inputs: AvgPoolInputs = {x: x4D};\n\n  const attrs: AvgPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, $x.dtype);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool = op({avgPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {AvgPool3D, AvgPool3DAttrs, AvgPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D average pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.avgPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates:\n *     `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction avgPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'avgPool3d', 'float32');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in avgPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in avgPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n\n    save([x5D]);\n\n    return backend.avgPool3d(x5D, convInfo);\n  };\n\n  const inputs: AvgPool3DInputs = {x: x5D};\n\n  const attrs: AvgPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool3D,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, x5D.dtype);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3d = op({avgPool3d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\nexport function assertParamsConsistent(shapes: number[][], axis: number) {\n  const rank = shapes[0].length;\n  shapes.forEach((shape, i) => {\n    util.assert(\n        shape.length === rank,\n        () =>\n            `Error in concat${rank}D: rank of tensors[${i}] must be the same ` +\n            `as the rank of the rest (${rank})`);\n  });\n\n  util.assert(\n      axis >= 0 && axis < rank,\n      () => `Error in concat${rank}D: axis must be between 0 and ${rank - 1}.`);\n\n  const firstShape = shapes[0];\n  shapes.forEach((shape, i) => {\n    for (let r = 0; r < rank; r++) {\n      util.assert(\n          (r === axis) || (shape[r] === firstShape[r]),\n          () => `Error in concat${rank}D: Shape of tensors[${i}] (${shape}) ` +\n              `does not match the shape of the rest (${firstShape}) ` +\n              `along the non-concatenated axis ${i}.`);\n    }\n  });\n}\n\nexport function computeOutShape(shapes: number[][], axis: number): number[] {\n  const outputShape = shapes[0].slice();\n  for (let i = 1; i < shapes.length; i++) {\n    outputShape[axis] += shapes[i][axis];\n  }\n  return outputShape;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Concat, ConcatAttrs, ConcatInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, parseAxisParam, sizeFromShape} from '../util';\n\nimport {assertParamsConsistent, computeOutShape} from './concat_util';\nimport {op} from './operation';\nimport {tensor} from './tensor';\n\n/**\n * Concatenates a list of `tf.Tensor`s along a given axis.\n *\n * The tensors ranks and types must match, and their sizes must match in all\n * dimensions except `axis`.\n *\n * Also available are stricter rank-specific methods that assert that\n * `tensors` are of the given rank:\n *   - `tf.concat1d`\n *   - `tf.concat2d`\n *   - `tf.concat3d`\n *   - `tf.concat4d`\n *\n * Except `tf.concat1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * a.concat(b).print();  // or a.concat(b)\n * ```\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.concat([a, b, c]).print();\n * ```\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [10, 20]]);\n * const b = tf.tensor2d([[3, 4], [30, 40]]);\n * const axis = 1;\n * tf.concat([a, b], axis).print();\n * ```\n * @param tensors A list of tensors to concatenate.\n * @param axis The axis to concate along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction concat_<T extends Tensor>(tensors: Array<T|TensorLike>, axis = 0): T {\n  assert(tensors.length >= 1, () => 'Pass at least one tensor to concat');\n\n  let $tensors = convertToTensorArray(tensors, 'tensors', 'concat');\n\n  if ($tensors[0].dtype === 'complex64') {\n    $tensors.forEach(tensor => {\n      if (tensor.dtype !== 'complex64') {\n        throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${tensor.dtype}. `);\n      }\n    });\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $axis = parseAxisParam(axis, $tensors[0].shape)[0];\n    const outShape = computeOutShape($tensors.map(t => t.shape), $axis);\n    if (sizeFromShape(outShape) === 0) {\n      return tensor([], outShape) as T;\n    }\n    // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n    $tensors = $tensors.filter(t => t.size > 0);\n    if ($tensors.length === 1) {\n      return $tensors[0];\n    }\n\n    const shapes = $tensors.map(t => t.shape);\n    assertParamsConsistent(shapes, $axis);\n\n    const res = backend.concat($tensors, $axis);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: ConcatInputs = $tensors;\n  const attr: ConcatAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Concat,\n             attr as {} as NamedAttrMap) as T;\n}\n\nexport const concat = op({concat_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sigmoid, SigmoidInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sigmoid element-wise, `1 / (1 + exp(-x))`\n *\n * ```js\n * const x = tf.tensor1d([0, -1, 2, -3]);\n *\n * x.sigmoid().print();  // or tf.sigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sigmoid');\n\n  const inputs: SigmoidInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sigmoid($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sigmoid);\n}\nexport const sigmoid = op({sigmoid_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Slice, SliceAttrs, SliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as slice_util from './slice_util';\n\n/**\n * Extracts a slice from a `tf.Tensor` starting at coordinates `begin`\n * and is of size `size`.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `x` is of the given rank:\n *   - `tf.slice1d`\n *   - `tf.slice2d`\n *   - `tf.slice3d`\n *   - `tf.slice4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.slice([1], [2]).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * x.slice([1, 0], [1, 2]).print();\n * ```\n * @param x The input `tf.Tensor` to slice from.\n * @param begin The coordinates to start the slice from. The length can be\n *     less than the rank of x - the rest of the axes will have implicit 0 as\n *     start. Can also be a single number, in which case it specifies the\n *     first axis.\n * @param size The size of the slice. The length can be less than the rank of\n *     x - the rest of the axes will have implicit -1. A value of -1 requests\n *     the rest of the dimensions in the axis. Can also be a single number,\n *     in which case it specifies the size of the first axis.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction slice_<R extends Rank, T extends Tensor<R>>(\n    x: T|TensorLike, begin: number|number[], size?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'slice');\n\n  if ($x.rank === 0) {\n    throw new Error('Slicing scalar is not possible');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const [begin_, size_] = slice_util.parseSliceParams($x, begin, size);\n    slice_util.assertParamsValid($x, begin_, size_);\n    save([$x]);\n    return backend.slice($x, begin_, size_);\n  };\n\n  const inputs: SliceInputs = {x: $x};\n  const attrs: SliceAttrs = {begin, size};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Slice,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const slice = op({slice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tanh, TanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic tangent of the input `tf.Tensor` element-wise: `tanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, 70]);\n *\n * x.tanh().print();  // or tf.tanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tanh');\n\n  const inputs: TanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const y = backend.tanh($x);\n    save([y]);\n    return y;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tanh);\n}\nexport const tanh = op({tanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {add} from './add';\nimport {concat} from './concat';\nimport {matMul} from './mat_mul';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {slice} from './slice';\nimport {tanh} from './tanh';\n\n/**\n * Computes the next state and output of a BasicLSTMCell.\n *\n * Returns `[newC, newH]`.\n *\n * Derived from tf.contrib.rnn.BasicLSTMCell.\n *\n * @param forgetBias Forget bias for the cell.\n * @param lstmKernel The weights for the cell.\n * @param lstmBias The bias for the cell.\n * @param data The input to the cell.\n * @param c Previous cell state.\n * @param h Previous cell output.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction basicLSTMCell_(\n    forgetBias: Scalar|TensorLike, lstmKernel: Tensor2D|TensorLike,\n    lstmBias: Tensor1D|TensorLike, data: Tensor2D|TensorLike,\n    c: Tensor2D|TensorLike, h: Tensor2D|TensorLike): [Tensor2D, Tensor2D] {\n  const $forgetBias =\n      convertToTensor(forgetBias, 'forgetBias', 'basicLSTMCell');\n  const $lstmKernel =\n      convertToTensor(lstmKernel, 'lstmKernel', 'basicLSTMCell');\n  const $lstmBias = convertToTensor(lstmBias, 'lstmBias', 'basicLSTMCell');\n  const $data = convertToTensor(data, 'data', 'basicLSTMCell');\n  const $c = convertToTensor(c, 'c', 'basicLSTMCell');\n  const $h = convertToTensor(h, 'h', 'basicLSTMCell');\n\n  const combined = concat([$data, $h], 1);\n  const weighted = matMul(combined, $lstmKernel);\n  const res: Tensor2D = add(weighted, $lstmBias);\n\n  // i = input_gate, j = new_input, f = forget_gate, o = output_gate\n  const batchSize = res.shape[0];\n  const sliceCols = res.shape[1] / 4;\n  const sliceSize: [number, number] = [batchSize, sliceCols];\n  const i = slice(res, [0, 0], sliceSize);\n  const j = slice(res, [0, sliceCols], sliceSize);\n  const f = slice(res, [0, sliceCols * 2], sliceSize);\n  const o = slice(res, [0, sliceCols * 3], sliceSize);\n\n  const newC: Tensor2D =\n      add(mul(sigmoid(i), tanh(j)),\n          mul($c, sigmoid(add($forgetBias, f)) as Tensor2D));\n  const newH: Tensor2D = mul(tanh(newC), sigmoid(o));\n  return [newC, newH];\n}\n\nexport const basicLSTMCell = op({basicLSTMCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchToSpaceND, BatchToSpaceNDAttrs, BatchToSpaceNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation reshapes the \"batch\" dimension 0 into `M + 1` dimensions of\n * shape `blockShape + [batch]`, interleaves these blocks back into the grid\n * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with\n * the same rank as the input. The spatial dimensions of this intermediate\n * result are then optionally cropped according to `crops` to produce the\n * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise\n * description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);\n * const blockShape = [2, 2];\n * const crops = [[0, 0], [0, 0]];\n *\n * x.batchToSpaceND(blockShape, crops).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param crops A 2-D array.  Must have shape `[M, 2]`, all values must be >= 0.\n * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input\n * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required\n * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,\n * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,\n * x.shape[N-1]]`\n *\n * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /\n * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`\n * according to `crops` to produce the output of shape: `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],\n * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -\n * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction batchToSpaceND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], crops: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'batchToSpaceND');\n  const prod = blockShape.reduce((a, b) => a * b);\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank is ${$x.rank} but should be > than blockShape.length ${\n          blockShape.length}`);\n\n  util.assert(\n      crops.length === blockShape.length,\n      () => `crops.length is ${\n          crops.length} but should be equal to blockShape.length  ${\n          blockShape.length}`);\n\n  util.assert(\n      $x.shape[0] % prod === 0,\n      () => `input tensor batch is ${\n                $x.shape[0]} but is not divisible by the product of ` +\n          `the elements of blockShape ${blockShape.join(' * ')} === ${prod}`);\n\n  const forward: ForwardFunc<T> = backend => {\n    return backend.batchToSpaceND($x, blockShape, crops);\n  };\n\n  const inputs: BatchToSpaceNDInputs = {x: $x};\n  const attrs: BatchToSpaceNDAttrs = {blockShape, crops};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      BatchToSpaceND, attrs as {} as NamedAttrMap);\n}\n\nexport const batchToSpaceND = op({batchToSpaceND_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, Tensor4D} from '../tensor';\nimport {Rank} from '../types';\nimport {reshape} from './reshape';\n\nexport function xAs4D<R extends Rank>(x: Tensor<R>) {\n  let x4D: Tensor4D;\n  if (x.rank === 0 || x.rank === 1) {\n    x4D = reshape(x, [1, 1, 1, x.size]);\n  } else if (x.rank === 2) {\n    x4D = reshape(x, [1, 1, x.shape[0], x.shape[1]]);\n  } else if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  } else {\n    x4D = x as Tensor4D;\n  }\n\n  return x4D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {xAs4D} from './batchnorm_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Batch normalization.\n *\n * As described in\n * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).\n *\n * Mean, variance, scale, and offset can be of two shapes:\n *   - The same shape as the input.\n *   - In the common case, the depth dimension is the last dimension of x, so\n *     the values would be an `tf.Tensor1D` of shape [depth].\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that parameters passed are of given rank\n *   - `tf.batchNorm2d`\n *   - `tf.batchNorm3d`\n *   - `tf.batchNorm4d`\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction batchNorm_<R extends Rank>(\n    x: Tensor<R>|TensorLike, mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor<R>|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor<R>|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n\n  util.assert(\n      $mean.rank === $variance.rank,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      $offset == null || $mean.rank === $offset.rank,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      $scale == null || $mean.rank === $scale.rank,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  const x4D: Tensor4D = xAs4D($x);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([x4D, $mean, $variance, $scale]);\n\n    return backend.batchNorm(\n        x4D, as1DOr4D($mean), as1DOr4D($variance), as1DOr4D($offset),\n        as1DOr4D($scale), varianceEpsilon);\n  };\n\n  const inputs: FusedBatchNormInputs = {\n    x: x4D,\n    scale: $scale,\n    offset: $offset,\n    mean: $mean,\n    variance: $variance\n  };\n\n  const attrs: FusedBatchNormAttrs = {varianceEpsilon};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      FusedBatchNorm, attrs as {} as NamedAttrMap);\n\n  return reshape(res, $x.shape);\n}\n\nfunction as1DOr4D(x: Tensor): Tensor4D|Tensor1D {\n  if (x == null) {\n    return null;\n  }\n  if (x.rank === 0) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [x.size]) as Tensor1D;\n  } else if (x.rank === 1) {\n    return x as Tensor1D;\n  } else if (x.rank === 2) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, 1, x.shape[0], x.shape[1]]) as Tensor4D;\n  } else if (x.rank === 3) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]) as Tensor4D;\n  }\n  return x as Tensor4D;\n}\n\nexport const batchNorm = op({batchNorm_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 2D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm2d_(\n    x: Tensor2D|TensorLike, mean: Tensor2D|Tensor1D|TensorLike,\n    variance: Tensor2D|Tensor1D|TensorLike,\n    offset?: Tensor2D|Tensor1D|TensorLike, scale?: Tensor2D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor2D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor2D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 2,\n      () => `Error in batchNorm2D: x must be rank 2 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 2 || $mean.rank === 1,\n      () => `Error in batchNorm2D: mean must be rank 2 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 2 || $variance.rank === 1,\n      () => `Error in batchNorm2D: variance must be rank 2 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 2 || $scale.rank === 1,\n        () => `Error in batchNorm2D: scale must be rank 2 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 2 || $offset.rank === 1,\n        () => `Error in batchNorm2D: offset must be rank 2 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm2d = op({batchNorm2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 3D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm3d_(\n    x: Tensor3D|TensorLike, mean: Tensor3D|Tensor1D|TensorLike,\n    variance: Tensor3D|Tensor1D|TensorLike,\n    offset?: Tensor3D|Tensor1D|TensorLike, scale?: Tensor3D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor3D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor3D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 3,\n      () => `Error in batchNorm3D: x must be rank 3 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 3 || $mean.rank === 1,\n      () => `Error in batchNorm3D: mean must be rank 3 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 3 || $variance.rank === 1,\n      () => `Error in batchNorm3D: variance must be rank 3 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 3 || $scale.rank === 1,\n        () => `Error in batchNorm3D: scale must be rank 3 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 3 || $offset.rank === 1,\n        () => `Error in batchNorm3D: offset must be rank 3 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm3d = op({batchNorm3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 4D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm4d_(\n    x: Tensor4D|TensorLike, mean: Tensor4D|Tensor1D|TensorLike,\n    variance: Tensor4D|Tensor1D|TensorLike,\n    offset?: Tensor4D|Tensor1D|TensorLike, scale?: Tensor4D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor4D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor4D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 4,\n      () => `Error in batchNorm4D: x must be rank 4 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 4 || $mean.rank === 1,\n      () => `Error in batchNorm4D: mean must be rank 4 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 4 || $variance.rank === 1,\n      () => `Error in batchNorm4D: variance must be rank 4 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 4 || $scale.rank === 1,\n        () => `Error in batchNorm4D: scale must be rank 4 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 4 || $offset.rank === 1,\n        () => `Error in batchNorm4D: offset must be rank 4 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm4d = op({batchNorm4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE} from '../engine';\nimport {BroadcastTo, BroadCastToAttrs, BroadcastToInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Broadcast an array to a compatible shape NumPy-style.\n *\n * The tensor's shape is compared to the broadcast shape from end to beginning.\n * Ones are prepended to the tensor's shape until is has the same length as\n * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is\n * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then\n * the input tensor is tiled N times along that axis (using tf.tile).\n *\n * @param input The tensor that is to be broadcasted.\n * @param shape The input is to be broadcast to this shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction broadcastTo_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  let input = convertToTensor(x, 'broadcastTo', 'x');\n  const xShape = input.shape;\n\n  if (shape.some(d => !(d > 0) || d % 1 !== 0)) {\n    throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);\n  }\n\n  if (shape.length < input.rank) {\n    throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${\n        input.rank}.`);\n  }\n\n  if (shape.length > input.rank) {\n    const newShape = input.shape.slice();\n    while (newShape.length < shape.length) {\n      newShape.unshift(1);\n    }\n    input = reshape(input, newShape);\n  }\n\n  const inputShape = input.shape;\n  const reps: number[] = Array.from(shape);\n  for (let i = shape.length - 1; i >= 0; i--) {\n    if (inputShape[i] === shape[i]) {\n      reps[i] = 1;\n    } else if (input.shape[i] !== 1) {\n      throw new Error(\n          `broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);\n    }\n  }\n  const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);\n\n  if (axes.length === 0) {\n    return clone(input) as Tensor<R>;\n  }\n\n  const forward = (backend: KernelBackend) => backend.tile(input, reps);\n\n  const inputs: BroadcastToInputs = {x: input};\n  const attrs: BroadCastToAttrs = {shape, inputShape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             BroadcastTo, attrs as unknown as NamedAttrMap) as Tensor<R>;\n}\n\nexport const broadcastTo = op({broadcastTo_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Ceil, CeilInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes ceiling of input `tf.Tensor` element-wise: `ceil(x)`\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.ceil().print();  // or tf.ceil(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction ceil_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'ceil');\n\n  const inputs: CeilInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.ceil($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Ceil);\n}\nexport const ceil = op({ceil_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {ClipByValue, ClipByValueAttrs, ClipByValueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)\n * ```\n * @param x The input tensor.\n * @param clipValueMin Lower-bound of range to be clipped to.\n * @param clipValueMax Upper-bound of range to be clipped to.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction clipByValue_<T extends Tensor>(\n    x: T|TensorLike, clipValueMin: number, clipValueMax: number): T {\n  const $x = convertToTensor(x, 'x', 'clipByValue');\n  util.assert(\n      (clipValueMin <= clipValueMax),\n      () => `Error in clip: min (${clipValueMin}) must be ` +\n          `less than or equal to max (${clipValueMax}).`);\n\n  const inputs: ClipByValueInputs = {x: $x};\n  const attrs: ClipByValueAttrs = {clipValueMin, clipValueMax};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const res = backend.clip($x, clipValueMin, clipValueMax);\n        save([$x]);\n        return res;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, ClipByValue,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const clipByValue = op({clipByValue_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor1D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(3) = |r1, g1, b1|\n * B: shape(2) = |r2, g2|\n * C = tf.concat1d([A, B]) == |r1, g1, b1, r2, g2|\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @return The concatenated array.\n */\nfunction concat1d_(tensors: Array<Tensor1D|TensorLike>): Tensor1D {\n  return concat(tensors, 0 /* axis */);\n}\n\nexport const concat1d = op({concat1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor2D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *\n * B: shape(2, 3) = | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * C = tf.concat2d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *                  | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * if axis = 1:\n * C = shape(2, 6) = | r1, g1, b1, r3, g3, b3 |\n *                   | r2, g2, b2, r4, g4, b4 |\n *\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concatenate along.\n * @return The concatenated array.\n */\nfunction concat2d_(\n    tensors: Array<Tensor2D|TensorLike>, axis: number): Tensor2D {\n  return concat(tensors, axis);\n}\n\nexport const concat2d = op({concat2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor3D`s along an axis.\n * See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *\n * B: shape(2, 1, 3) = | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * C = tf.concat3d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *                     | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * if axis = 1:\n * C: shape(2, 2, 3) = | r1, g1, b1, r3, g3, b3 |\n *                     | r2, g2, b2, r4, g4, b4 |\n *\n * if axis = 2:\n * C = shape(2, 1, 6) = | r1, g1, b1, r3, g3, b3 |\n *                      | r2, g2, b2, r4, g4, b4 |\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat3d_(\n    tensors: Array<Tensor3D|TensorLike>, axis: number): Tensor3D {\n  return concat(tensors, axis);\n}\n\nexport const concat3d = op({concat3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor4D`s along an axis.\n * See `concat` for details.\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat4d_(\n    tensors: Array<Tensor4D|TensorLike>, axis: number): Tensor4D {\n  return concat(tensors, axis);\n}\n\nexport const concat4d = op({concat4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2D, Conv2DAttrs, Conv2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 2D convolution over the input x.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2d: input must be rank 4, but got rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  util.assert(\n      inDepth === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${inDepth}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        false, $dataFormat);\n    const res = backend.conv2d(x4D, $filter, convInfo);\n\n    save([x4D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DInputs = {x: x4D, filter: $filter};\n  const attrs:\n      Conv2DAttrs = {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv2D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2d = op({conv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D, Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 1D convolution over the input x.\n *\n * @param x The input tensor, of rank 3 or rank 2, of shape\n *     `[batch, width, inChannels]`. If rank 2, batch of 1 is assumed.\n * @param filter The filter, rank 3, of shape\n *     `[filterWidth, inDepth, outDepth]`.\n * @param stride The number of entries by which the filter is moved right at\n *     each step.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from \"NWC\", \"NCW\". Defaults to \"NWC\",\n *     the data is stored in the order of [batch, in_width, in_channels]. Only\n *     \"NWC\" is currently supported.\n * @param dilation The dilation rate in which we sample input values in\n *     atrous convolution. Defaults to `1`. If it is greater than 1, then\n *     stride must be `1`.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv1d_<T extends Tensor2D|Tensor3D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike, stride: number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv1d');\n  const $filter = convertToTensor(filter, 'filter', 'conv1d');\n\n  let x3D = $x as Tensor3D;\n  let reshapedTo3D = false;\n  if ($x.rank === 2) {\n    reshapedTo3D = true;\n    x3D = reshape($x, [1, $x.shape[0], $x.shape[1]]);\n  }\n\n  util.assert(\n      x3D.rank === 3,\n      () => `Error in conv1d: input must be rank 3, but got rank ${x3D.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in conv1d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv1d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x3D.shape[2] === $filter.shape[1],\n      () => `Error in conv1d: depth of input (${x3D.shape[2]}) must match ` +\n          `input depth for filter ${$filter.shape[1]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(stride, dilation),\n      () => 'Error in conv1D: Either stride or dilation must be 1. ' +\n          `Got stride ${stride} and dilation '${dilation}'`);\n  util.assert(\n      dataFormat === 'NWC',\n      () => `Error in conv1d: got dataFormat of ${\n          dataFormat} but only NWC is currently supported.`);\n\n  const filter4D = reshape(\n      $filter, [1, $filter.shape[0], $filter.shape[1], $filter.shape[2]]);\n  const input4D = reshape(x3D, [x3D.shape[0], 1, x3D.shape[1], x3D.shape[2]]);\n  const strides: [number, number] = [1, stride];\n  const dilations: [number, number] = [1, dilation];\n\n  const conv2dDataFormat = 'NHWC';\n\n  const res = conv2d(\n      (input4D as Tensor4D), (filter4D as Tensor4D), strides, pad,\n      conv2dDataFormat, dilations, dimRoundingMode);\n\n  if (reshapedTo3D) {\n    return reshape(res, [res.shape[2], res.shape[3]]) as T;\n  }\n\n  return reshape(res, [res.shape[0], res.shape[2], res.shape[3]]) as T;\n}\n\nexport const conv1d = op({conv1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs, Conv2DBackpropInputInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 2D convolution.\n *\n * @param xShape The shape of the input: [batch, height, width, inDepth].\n * If length of 3, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 4 or rank 3 of shape\n *   `[batch, outHeight, outWidth, outDepth]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction conv2DBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape4D = xShape as [number, number, number, number];\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n    xShape4D = [1, xShape[0], xShape[1], xShape[2]];\n  }\n\n  util.assert(\n      xShape4D.length === 4,\n      () =>\n          `Error in conv2dDerInput: inShape must be length 4, but got length ` +\n          `${xShape4D.length}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerInput: dy must be rank 4, but got ` +\n          `rank ${dy4D.rank}`);\n  util.assert(\n      filter.rank === 4,\n      () => `Error in conv2dDerInput: filter must be rank 4, but got ` +\n          `rank ${filter.rank}`);\n  const inDepth = dataFormat === 'NHWC' ? xShape4D[3] : xShape4D[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filter.shape[2],\n      () => `Error in conv2dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[2]}.`);\n  util.assert(\n      outDepth === filter.shape[3],\n      () => `Error in conv2dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[3]}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerInput: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const dilations = 1;\n\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        xShape4D, filter.shape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    const res = backend.conv2dDerInput(dy4D, filter, convInfo);\n\n    save([dy4D, filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DBackpropInputInputs = {dy: dy4D, filter};\n\n  const attrs: Conv2DBackpropInputAttrs =\n      {strides, pad, dataFormat, dimRoundingMode, inputShape: xShape4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      Conv2DBackpropInput, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2DBackpropInput = op({conv2DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv2DBackpropInput} from './conv2d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 2D convolution of an image, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 4 or rank 3, of shape\n *   `[batch, height, width, inDepth]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 4 or rank 3:\n *     `[batch, height, width, outDepth]`. If rank 3, batch of 1 is assumed.\n * @param strides The strides of the original convolution:\n *     `[strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n * @param dimRoundingMode The rounding mode used when computing output\n *    dimensions if pad is a number. If none is provided, it will not round\n *    and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2dTranspose_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv2dTranspose');\n\n  return conv2DBackpropInput(\n      outputShape, $x, $filter, strides, pad, 'NHWC', dimRoundingMode);\n}\n\nexport const conv2dTranspose = op({conv2dTranspose_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3D, Conv3DAttrs, Conv3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {eitherStridesOrDilationsAreOne} from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 3D convolution over the input x.\n *\n * @param x The input tensor, of rank 5 or rank 4, of shape\n *     `[batch, depth, height, width, channels]`. If rank 4,\n * batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inChannels, outChannels]`.\n *      inChannels must match between input and filter.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationDepth, dilationHeight,\n *     dilationWidth]` in which we sample input values across the height\n *     and width dimensions in atrous convolution. Defaults to `[1, 1, 1]`.\n *     If `dilations` is a single number, then\n *     `dilationDepth == dilationHeight == dilationWidth`. If it is greater\n *     than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    strides: [number, number, number]|number, pad: 'valid'|'same',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations: [number, number, number]|number = [1, 1, 1]): T {\n  const $x = convertToTensor(x, 'x', 'conv3d');\n  const $filter = convertToTensor(filter, 'filter', 'conv3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3d: input must be rank 5, but got rank ${x5D.rank}.`);\n  util.assert(\n      $filter.rank === 5,\n      () => `Error in conv3d: filter must be rank 5, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x5D.shape[4] === $filter.shape[3],\n      () => `Error in conv3d: depth of input (${x5D.shape[4]}) must match ` +\n          `input depth for filter ${$filter.shape[3]}.`);\n  util.assert(\n      eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv3D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in conv3d: got dataFormat of ${\n          dataFormat} but only NDHWC is currently supported.`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, $filter.shape, strides, dilations, pad);\n    const res = backend.conv3d(x5D, $filter, convInfo);\n\n    save([x5D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv3DInputs = {x: x5D, filter: $filter};\n\n  const attrs: Conv3DAttrs = {strides, pad, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3d = op({conv3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropInputAttrs, Conv3DBackpropInputInputs, Conv3DBackpropInputV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 3D convolution.\n *\n * @param xShape The shape of the input: [batch, depth, height, width,\n * in_channels]. If length of 4, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 5 or rank 4 of shape\n *   `[batch, outDepth, outHeight, outWidth, in_channels]`.\n * If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n */\nfunction conv3DBackpropInput_<T extends Tensor4D|Tensor5D>(\n    xShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    dy: T, filter: Tensor5D, strides: [number, number, number]|number,\n    pad: 'valid'|'same'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape5D = xShape as [number, number, number, number, number];\n  let dy5D = dy as Tensor5D;\n  let reshapedTo5D = false;\n  if (dy.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n    xShape5D = [1, xShape[0], xShape[1], xShape[2], xShape[3]];\n  }\n\n  const inDepth = xShape5D[4];\n  const outDepth = dy5D.shape[4];\n  util.assert(\n      xShape5D.length === 5,\n      () =>\n          `Error in conv3dDerInput: inShape must be length 5, but got length ` +\n          `${xShape5D.length}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerInput: dy must be rank 5, but got ` +\n          `rank ${dy5D.rank}`);\n  util.assert(\n      filter.rank === 5,\n      () => `Error in conv3dDerInput: filter must be rank 5, but got ` +\n          `rank ${filter.rank}`);\n  util.assert(\n      inDepth === filter.shape[3],\n      () => `Error in conv3dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[3]}.`);\n  util.assert(\n      outDepth === filter.shape[4],\n      () => `Error in conv3dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[4]}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        xShape5D, filter.shape, strides, dilations, pad);\n\n    return backend.conv3dDerInput(dy5D, filter, convInfo);\n  };\n\n  const inputs: Conv3DBackpropInputInputs = {dy: dy5D};\n\n  const attrs: Conv3DBackpropInputAttrs = {pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, Conv3DBackpropInputV2,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3DBackpropInput = op({conv3DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D, Tensor5D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv3DBackpropInput} from './conv3d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 3D convolution of a volume, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 5 or rank 4, of shape\n *   `[batch, depth, height, width, inDepth]`. If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[depth, filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 5 or rank 4:\n *     `[batch, depth, height, width, outDepth]`. If rank 3, batch of 1 is\n *    assumed.\n * @param strides The strides of the original convolution:\n *     `[strideDepth, strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3dTranspose_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    outputShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): T {\n  const $x = convertToTensor(x, 'x', 'conv3dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv3dTranspose');\n\n  return conv3DBackpropInput(outputShape, $x, $filter, strides, pad);\n}\n\nexport const conv3dTranspose = op({conv3dTranspose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cos, CosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes cos of the input `tf.Tensor` element-wise: `cos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.cos().print();  // or tf.cos(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cos');\n\n  const inputs: CosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cos);\n}\nexport const cos = op({cos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cosh, CoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic cos of the input `tf.Tensor` element-wise: `cosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.cosh().print();  // or tf.cosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cosh');\n  const inputs: CoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cosh);\n}\nexport const cosh = op({cosh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Cumsum, CumsumAttrs, CumsumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {getAxesPermutation, getInnerMostAxes, getUndoAxesPermutation} from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the cumulative sum of a `tf.Tensor` along `axis`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4]);\n * x.cumsum().print();\n * ```\n * ```js\n * const x = tf.tensor([[1, 2], [3, 4]]);\n * x.cumsum().print();\n * ```\n *\n * @param x The input tensor to be summed.\n * @param axis The axis along which to sum. Optional. Defaults to 0.\n * @param exclusive Whether to perform exclusive cumulative sum. Optional.\n *     Defaults to false. If set to true then the sum of each tensor entry\n *     does not include its own value, but only the values previous to it\n *     along the specified axis.\n * @param reverse Whether to sum in the opposite direction. Optional.\n *     Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Scan'}\n */\nfunction cumsum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis = 0, exclusive = false, reverse = false): T {\n  const $x = convertToTensor(x, 'x', 'cumsum');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const permutation = getAxesPermutation([axis], $x.rank);\n        let permutedX = $x;\n        if (permutation != null) {\n          permutedX = transpose($x, permutation);\n        }\n        const permutedAxis = getInnerMostAxes(1, $x.rank)[0];\n        let value = backend.cumsum(permutedX, permutedAxis, exclusive, reverse);\n        save([$x]);\n\n        if (permutation != null) {\n          const reversePermutation = getUndoAxesPermutation(permutation);\n          value = transpose(value, reversePermutation);\n        }\n        return value;\n      };\n\n  const inputs: CumsumInputs = {x: $x};\n  const attrs: CumsumAttrs = {axis, exclusive, reverse};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Cumsum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const cumsum = op({cumsum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthToSpace, DepthToSpaceAttrs, DepthToSpaceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Rearranges data from depth into blocks of spatial data. More specifically,\n * this op outputs a copy of the input tensor where values from the `depth`\n * dimension are moved in spatial blocks to the `height` and `width` dimensions.\n * The attr `blockSize` indicates the input block size and how the data is\n * moved.\n *\n *  - Chunks of data of size `blockSize * blockSize` from depth are rearranged\n * into non-overlapping blocks of size `blockSize x blockSize`\n *\n *  - The width the output tensor is `inputWidth * blockSize`, whereas the\n * height is `inputHeight * blockSize`\n *\n *  - The Y, X coordinates within each block of the output image are determined\n * by the high order component of the input channel index\n *\n *  - The depth of the input tensor must be divisible by `blockSize *\n * blockSize`\n *\n * The `dataFormat` attr specifies the layout of the input and output tensors\n * with the following options: \"NHWC\": [ `batch, height, width, channels` ]\n * \"NCHW\": [ `batch, channels, height, width` ]\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 1, 1, 4]);\n * const blockSize = 2;\n * const dataFormat = \"NHWC\";\n *\n * tf.depthToSpace(x, blockSize, dataFormat).print();\n * ```\n *\n * @param x The input tensor of rank 4\n * @param blockSIze  An `int` that is `>= 2`. The size of the spatial block\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\"\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction depthToSpace_(\n    x: Tensor4D|TensorLike4D, blockSize: number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'depthToSpace') as Tensor4D;\n\n  const inputHeight = (dataFormat === 'NHWC') ? $x.shape[1] : $x.shape[2];\n  const inputWidth = (dataFormat === 'NHWC') ? $x.shape[2] : $x.shape[3];\n  const inputDepth = (dataFormat === 'NHWC') ? $x.shape[3] : $x.shape[1];\n\n  util.assert(\n      inputHeight * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputHeight} and ${blockSize}  for depthToSpace with input shape\n    ${$x.shape}`);\n\n  util.assert(\n      inputWidth * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputWidth} and ${blockSize} for depthToSpace with input shape\n        ${$x.shape}`);\n\n  util.assert(\n      (inputDepth % (blockSize * blockSize) === 0),\n      () => `Dimension size must be evenly divisible by ${\n          blockSize * blockSize} but is ${\n          inputDepth} for depthToSpace with input shape ${$x.shape}`);\n\n  const forward: ForwardFunc<Tensor4D> = backend =>\n      backend.depthToSpace($x, blockSize, dataFormat);\n\n  const inputs: DepthToSpaceInputs = {x: $x};\n  const attrs: DepthToSpaceAttrs = {blockSize, dataFormat};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      DepthToSpace, attrs as {} as NamedAttrMap);\n}\n\nexport const depthToSpace = op({depthToSpace_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs, DepthwiseConv2dNativeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Depthwise 2D convolution.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction depthwiseConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in depthwiseConv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in depthwiseConv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1];\n    }\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in depthwiseConv2d: Either strides or dilations must be ' +\n            `1. Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        true /* depthwise */);\n    const res = backend.depthwiseConv2D(x4D, $filter, convInfo);\n    save([x4D, $filter]);\n    return res;\n  };\n\n  const inputs: DepthwiseConv2dNativeInputs = {x: x4D, filter: $filter};\n  const attrs: DepthwiseConv2dNativeAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      DepthwiseConv2dNative, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2d = op({depthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Diag, DiagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a diagonal tensor with a given diagonal values.\n *\n * Given a diagonal, this operation returns a tensor with the diagonal and\n * everything else padded with zeros.\n *\n * Assume the input has dimensions `[D1,..., Dk]`, then the output is a tensor\n * of rank 2k with dimensions `[D1,..., Dk, D1,..., Dk]`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * tf.diag(x).print()\n * ```\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4, 5, 6, 6, 8], [4, 2])\n *\n * tf.diag(x).print()\n * ```\n * @param x The input tensor.\n */\nfunction diag_(x: Tensor): Tensor {\n  const $x = convertToTensor(x, 'x', 'diag');\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const flat = reshape($x, [$x.size]);\n    const result = backend.diag(flat);\n    const outShape = [...x.shape, ...x.shape];\n\n    return reshape(result, outShape);\n  };\n\n  const inputs: DiagInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Diag);\n}\n\nexport const diag = op({diag_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DAttrs, Dilation2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the grayscale dilation over the input `x`.\n *\n * @param x The input tensor, rank 3 or rank 4 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter tensor, rank 3, of shape\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat Specify the data format of the input and output data.\n *      Defaults to 'NHWC'. Only 'NHWC' is currently supported. With the\n *      default format \"NHWC\", the data is stored in the order of: [batch,\n *      height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     for atrous morphological dilation. Defaults to `[1, 1]`. If `dilations`\n *     is a single number, then `dilationHeight == dilationWidth`. If it is\n *     greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction dilation2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same',\n    dilations: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'dilation2d');\n  const $filter = convertToTensor(filter, 'filter', 'dilation2d');\n\n  util.assert(\n      $x.rank === 3 || $x.rank === 4,\n      () => `Error in dilation2d: input must be rank 3 or 4, but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in dilation2d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in dilation2d: Only NHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n    reshapedTo4D = true;\n  }\n\n  const inputs: Dilation2DInputs = {x: x4D, filter: $filter};\n  const attrs: Dilation2DAttrs = {strides, pad, dilations};\n\n  const res = ENGINE.runKernel(\n                  Dilation2D, inputs as {} as NamedTensorMap,\n                  attrs as {} as NamedAttrMap) as T;\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res;\n}\n\nexport const dilation2d = op({dilation2d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Returns the dimensions in the input shape that are broadcasted to\n * produce the provided output shape.\n *\n * The returned dimensions are 0-indexed and sorted. An example:\n * inShape = [4, 1, 3]\n * outShape = [5, 4, 3, 3]\n * result = [1]. Dimension 1 (2nd dimension of input) gets broadcasted 1 => 3.\n */\nexport function getBroadcastDims(\n    inShape: number[], outShape: number[]): number[] {\n  const inRank = inShape.length;\n  const dims: number[] = [];\n  for (let i = 0; i < inRank; i++) {\n    const dim = inRank - 1 - i;\n    const a = inShape[dim] || 1;\n    const b = outShape[outShape.length - 1 - i] || 1;\n    if (b > 1 && a === 1) {\n      dims.unshift(dim);\n    }\n  }\n  return dims;\n}\n\n/**\n * Returns the axes in the output space that should be reduced to produce\n * the input space.\n */\nexport function getReductionAxes(\n    inShape: number[], outShape: number[]): number[] {\n  const result: number[] = [];\n  for (let i = 0; i < outShape.length; i++) {\n    const inDim = inShape[inShape.length - i - 1];\n    const outAxis = outShape.length - i - 1;\n    const outDim = outShape[outAxis];\n    if (inDim == null || (inDim === 1 && outDim > 1)) {\n      result.unshift(outAxis);\n    }\n  }\n  return result;\n}\n\nexport function assertAndGetBroadcastShape(\n    shapeA: number[], shapeB: number[]): number[] {\n  const result: number[] = [];\n  const l = Math.max(shapeA.length, shapeB.length);\n\n  for (let i = 0; i < l; i++) {\n    let a = shapeA[shapeA.length - i - 1];\n    if (a == null) {\n      a = 1;\n    }\n    let b = shapeB[shapeB.length - i - 1];\n    if (b == null) {\n      b = 1;\n    }\n    if (a === 1) {\n      result.unshift(b);\n    } else if (b === 1) {\n      result.unshift(a);\n    } else if (a !== b) {\n      const errMsg = `Operands could not be broadcast together with shapes ` +\n          `${shapeA} and ${shapeB}.`;\n      throw Error(errMsg);\n    } else {\n      result.unshift(a);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Equal, EqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a == b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.equal(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction equal_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'equal');\n  let $b = convertToTensor(b, 'b', 'equal');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.equal($a, $b);\n\n  const inputs: EqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, Equal) as T;\n}\n\nexport const equal = op({equal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SelectV2, SelectV2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch} from '../util';\n\nimport {broadcastTo} from './broadcast_to';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the elements, either `a` or `b` depending on the `condition`.\n *\n * If the condition is true, select from `a`, otherwise select from `b`.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const a = tf.tensor1d([1 , 2, 3]);\n * const b = tf.tensor1d([-1, -2, -3]);\n *\n * a.where(cond, b).print();\n * ```\n *\n * @param condition The input condition. Must be of dtype bool.\n * @param a If `condition` is rank 1, `a` may have a higher rank but\n *     its first dimension must match the size of `condition`.\n * @param b A tensor with the same dtype as `a` and with shape that is\n *     compatible with `a`.\n * @return A tensor with same dtype as `a` and `b`, and shape that is\n *     broadcastable from `a` and `b`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction where_<T extends Tensor>(\n    condition: Tensor|TensorLike, a: T|TensorLike, b: T|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'where');\n  const $b = convertToTensor(b, 'b', 'where');\n  const $condition = convertToTensor(condition, 'condition', 'where', 'bool');\n  // TODO: move this logic to forward function when the broadcastTo op is\n  // implemented in WASM.\n  // Find the broadcastable shape for $a and $b.\n  const broadcastShape = assertAndGetBroadcastShape($a.shape, $b.shape);\n  const $broadcastedA = broadcastTo($a, broadcastShape);\n  const $broadcastedB = broadcastTo($b, broadcastShape);\n  if ($condition.rank === 1) {\n    // If condition rank is 1, then the first dimension must match the size of\n    // condition.\n    assert(\n        $condition.shape[0] === $a.shape[0],\n        () => 'The first dimension of `a` must match the size of `condition`.');\n  }\n\n  if ($condition.rank !== 1) {\n    // A must have the same shape as condition.\n    assertShapesMatch(\n        $condition.shape, $broadcastedB.shape, 'Error in where: ');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.select($condition, $broadcastedA, $broadcastedB);\n    save([$condition]);\n    return res;\n  };\n  const inputs: SelectV2Inputs = {\n    condition: $condition,\n    t: $broadcastedA,\n    e: $broadcastedB\n  };\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* gradient */,\n             SelectV2) as T;\n}\n\nexport const where = op({where_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {ZerosLike, ZerosLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.zerosLike(x).print();\n * ```\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction zerosLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'zerosLike');\n  const inputs: ZerosLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             backend => backend.zerosLike($x), inputs as {} as NamedTensorMap,\n             null /* grad */, ZerosLike) as T;\n}\nexport const zerosLike = op({zerosLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {div} from './div';\nimport {equal} from './equal';\nimport {op} from './operation';\nimport {where} from './where';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting. Return 0\n * if denominator is 0.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n * const c = tf.tensor1d([0, 0, 0, 0]);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n * const c = tf.scalar(0);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction divNoNan_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  // TODO: Make this into its own kernel.\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const divResult = div($a, $b);\n  const zeros = zerosLike(divResult);\n  const bEqualsZero = equal($b, zeros);\n  return where(bEqualsZero, zeros, divResult) as T;\n}\n\nexport const divNoNan = op({divNoNan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor,} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices and/or vectors, `t1` and `t2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor2d([[1, 2], [3, 4]]);\n * const c = tf.tensor2d([[1, 2, 3], [4, 5, 6]]);\n *\n * a.dot(b).print();  // or tf.dot(a, b)\n * b.dot(a).print();\n * b.dot(c).print();\n * ```\n * @param t1 The first tensor in the dot operation.\n * @param t2 The second tensor in the dot operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction dot_(t1: Tensor|TensorLike, t2: Tensor|TensorLike): Tensor {\n  const $t1 = convertToTensor(t1, 't1', 'dot');\n  const $t2 = convertToTensor(t2, 't2', 'dot');\n\n  util.assert(\n      ($t1.rank === 1 || $t1.rank === 2) && ($t2.rank === 1 || $t2.rank === 2),\n      () => `Error in dot: inputs must all be rank 1 or 2, but got ranks ` +\n          `${$t1.rank} and ${$t2.rank}.`);\n\n  const t1Inner = ($t1.rank === 1 ? $t1.size : $t1.shape[1]);\n  const t2Inner = ($t2.rank === 1 ? $t2.size : $t2.shape[0]);\n\n  util.assert(\n      t1Inner === t2Inner,\n      () => `Error in dot: inner dimensions of inputs must match, but got ` +\n          `${t1Inner} and ${t2Inner}.`);\n\n  if ($t1.rank === 1 && $t2.rank === 1) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, []);\n  } else if ($t1.rank === 1 && $t2.rank === 2) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else if ($t1.rank === 2 && $t2.rank === 1) {\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul($t1, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else {\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul($t1, t22D);\n    return t1t2;\n  }\n}\n\nexport const dot = op({dot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential linear element-wise: `x > 0 ? e ^ x - 1 : 0`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 1, -3, 2]);\n *\n * x.elu().print();  // or tf.elu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction elu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'elu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.elu($x);\n    save([y]);\n    return y;\n  };\n\n  const inputs: EluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Elu) as\n      T;\n}\n\nexport const elu = op({elu_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Erf, ErfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes gause error function of the input `tf.Tensor` element-wise:\n * `erf(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.erf().print(); // or tf.erf(x);\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction erf_<T extends Tensor>(x: T|TensorLike): T {\n  let $x = convertToTensor(x, 'x', 'erf');\n  util.assert(\n      $x.dtype === 'int32' || $x.dtype === 'float32',\n      () => 'Input dtype must be `int32` or `float32`.');\n\n  if ($x.dtype === 'int32') {\n    $x = cast($x, 'float32');\n  }\n\n  const inputs: ErfInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.erf($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Erf);\n}\nexport const erf = op({erf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Exp, ExpInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.exp().print();  // or tf.exp(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction exp_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'exp');\n\n  const inputs: ExpInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.exp($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Exp);\n}\nexport const exp = op({exp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension\n * into the tensor's shape.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const axis = 1;\n * x.expandDims(axis).print();\n * ```\n *\n * @param x The input tensor whose dimensions to be expanded.\n * @param axis The dimension index at which to insert shape of `1`. Defaults\n *     to 0 (the first dimension).\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction expandDims_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'expandDims', parseAs);\n\n  util.assert(axis <= $x.rank, () => 'Axis must be <= rank of the tensor');\n  const newShape = $x.shape.slice();\n  if (axis < 0) {\n    // Negative value is counted from the tail of rank.\n    util.assert(\n        -($x.rank + 1) <= axis,\n        () => `Axis must be in the interval [${- ($x.rank + 1)}, ${$x.rank}]`);\n    axis = $x.rank + axis + 1;\n  }\n  newShape.splice(axis, 0, 1);\n  return reshape($x, newShape) as T;\n}\n\nexport const expandDims = op({expandDims_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Expm1, Expm1Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` minus one element-wise.\n * `e ^ x - 1`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.expm1().print();  // or tf.expm1(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction expm1_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'expm1');\n\n  const inputs: Expm1Inputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.expm1($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Expm1);\n}\nexport const expm1 = op({expm1_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Tile, TileAttrs, TileInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Construct a tensor by repeating it the number of times given by reps.\n *\n * This operation creates a new tensor by replicating `input` `reps`\n * times. The output tensor's i'th dimension has `input.shape[i] *\n * reps[i]` elements, and the values of `input` are replicated\n * `reps[i]` times along the i'th dimension. For example, tiling\n * `[a, b, c, d]` by `[2]` produces `[a, b, c, d, a, b, c, d]`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n *\n * a.tile([2]).print();    // or a.tile([2])\n * ```\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.tile([1, 2]).print();  // or a.tile([1, 2])\n * ```\n * @param x The tensor to tile.\n * @param reps Determines the number of replications per dimension.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction tile_<T extends Tensor>(x: T|TensorLike, reps: number[]): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'tile', parseAs);\n  util.assert(\n      $x.rank === reps.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of reps ${reps}.`);\n\n  const forward: ForwardFunc<T> = (backend, save) => {\n    const res = backend.tile($x, reps);\n    save([$x]);\n    return res;\n  };\n\n  const inputsToSave = [$x];\n  const inputs: TileInputs = {x: $x};\n  const attrs: TileAttrs = {reps};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, Tile,\n      attrs as unknown as NamedAttrMap, inputsToSave);\n}\n\nexport const tile = op({tile_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {DataType} from '../types';\n\nimport {buffer} from './buffer';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {tile} from './tile';\n\n/**\n * Create an identity matrix.\n *\n * @param numRows Number of rows.\n * @param numColumns Number of columns. Defaults to `numRows`.\n * @param batchShape If provided, will add the batch shape to the beginning\n *   of the shape of the returned `tf.Tensor` by repeating the identity\n *   matrix.\n * @param dtype Data type.\n * @returns Identity matrix of the specified size and data type, possibly\n *   with batch repetition if `batchShape` is specified.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction eye_(\n    numRows: number, numColumns?: number,\n    batchShape?:\n        [\n          number\n        ]|[number,\n           number]|[number, number, number]|[number, number, number, number],\n    dtype: DataType = 'float32'): Tensor2D {\n  if (numColumns == null) {\n    numColumns = numRows;\n  }\n  const buff = buffer([numRows, numColumns], dtype);\n  const n = numRows <= numColumns ? numRows : numColumns;\n  for (let i = 0; i < n; ++i) {\n    buff.set(1, i, i);\n  }\n  const out: Tensor2D = reshape(buff.toTensor(), [numRows, numColumns]);\n  if (batchShape == null) {\n    return out;\n  } else {\n    if (batchShape.length === 1) {\n      return tile(expandDims(out, 0), [batchShape[0], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 2) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(\n                 expandDims(expandDims(out, 0), 0),\n                 [batchShape[0], batchShape[1], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 3) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(expandDims(expandDims(expandDims(out, 0), 0), 0), [\n               batchShape[0], batchShape[1], batchShape[2], 1, 1\n             ]) as Tensor2D;\n    } else {\n      throw new Error(\n          `eye() currently supports only 1D and 2D ` +\n          // tslint:disable-next-line:no-any\n          `batchShapes, but received ${(batchShape as any).length}D.`);\n    }\n  }\n}\n\nexport const eye = op({eye_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Fill, FillAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\n/**\n * Creates a `tf.Tensor` filled with a scalar value.\n *\n * ```js\n * tf.fill([2, 2], 4).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param value The scalar value to fill the tensor with.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n * 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction fill<R extends Rank>(\n    shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n  const attrs: FillAttrs = {shape, value, dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.fill(shape, value, dtype), {}, null, Fill,\n      attrs as {} as NamedAttrMap);\n}\n\nexport {fill};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Floor, FloorInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes floor of input `tf.Tensor` element-wise: `floor(x)`.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.floor().print();  // or tf.floor(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction floor_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'floor');\n\n  const inputs: FloorInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.floor($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Floor);\n}\nexport const floor = op({floor_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inputs of size above this threshold will be parallelized by calling multiple\n * shader programs.\n */\nimport {nearestDivisor} from '../util';\n\nexport const PARALLELIZE_THRESHOLD = 30;\n\nexport interface ReduceInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  outSize: number;\n}\n\nexport function computeOptimalWindowSize(inSize: number): number {\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    return inSize;\n  }\n  return nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {nearestDivisor} from '../util';\n\nimport {PARALLELIZE_THRESHOLD} from './reduce_util';\n\nexport interface SegOpInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  numSegments: number;\n}\n\nexport function segOpComputeOptimalWindowSize(\n    inSize: number, numSegments: number): number {\n  let done = false;\n  let res;\n\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    res = inSize;\n    done = true;\n  } else {\n    res = nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n  }\n\n  while (!done) {\n    if (res > numSegments || res === inSize) {\n      done = true;\n    } else {\n      res = nearestDivisor(inSize, res + 1);\n    }\n  }\n  return res;\n}\n\nexport function computeOutShape(\n    aShape: number[], axis: number, numSegments: number): number[] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (dim !== axis) {\n      outShape.push(aShape[dim]);\n    } else {\n      outShape.push(numSegments);\n    }\n  }\n  return outShape;\n}\n\nexport interface GatherOpShapeInfo {\n  batchSize: number;\n  sliceSize: number;\n  dimSize: number;\n  outputShape: number[];\n}\nexport function collectGatherOpShapeInfo(\n    x: Tensor, indices: Tensor, axis: number): GatherOpShapeInfo {\n  const dimSize = x.shape[axis];\n\n  const outputShape: number[] = [];\n  let batchSize = 1;\n  let sliceSize = 1;\n  for (let i = 0; i < axis; i++) {\n    outputShape.push(x.shape[i]);\n    batchSize *= x.shape[i];\n  }\n\n  for (let i = 0; i < indices.rank; i++) {\n    outputShape.push(indices.shape[i]);\n  }\n\n  for (let i = axis + 1; i < x.rank; i++) {\n    outputShape.push(x.shape[i]);\n    sliceSize *= x.shape[i];\n  }\n\n  return {batchSize, sliceSize, dimSize, outputShape};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherV2, GatherV2Attrs, GatherV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {collectGatherOpShapeInfo} from './segment_util';\n\n/**\n * Gather slices from tensor `x`'s axis `axis` according to `indices`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const indices = tf.tensor1d([1, 3, 3], 'int32');\n *\n * x.gather(indices).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const indices = tf.tensor1d([1, 1, 0], 'int32');\n *\n * x.gather(indices).print();\n * ```\n * @param x The input tensor whose slices to be gathered.\n * @param indices The indices of the values to extract.\n * @param axis The axis over which to select values. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction gather_<T extends Tensor>(\n    x: T|TensorLike, indices: Tensor|TensorLike, axis = 0): T {\n  const $x = convertToTensor(x, 'x', 'gather');\n  const $indices = convertToTensor(indices, 'indices', 'gather', 'int32');\n\n  const inputs: GatherV2Inputs = {x: $x, indices: $indices};\n  const attrs: GatherV2Attrs = {axis};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const parsedAxis = parseAxisParam(axis, $x.shape)[0];\n    const shapeInfo = collectGatherOpShapeInfo($x, $indices, parsedAxis);\n\n    const res =\n        backend.gather($x, reshape($indices, [$indices.size]), parsedAxis);\n    save([$x, $indices]);\n\n    return reshape(res, shapeInfo.outputShape);\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, GatherV2,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const gather = op({gather_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Greater, GreaterInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a > b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greater(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greater_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greater');\n  let $b = convertToTensor(b, 'b', 'greater');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.greater($a, $b);\n\n  const inputs: GreaterInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             Greater) as T;\n}\n\nexport const greater = op({greater_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GreaterEqual, GreaterEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a >= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greaterEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greaterEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greaterEqual');\n  let $b = convertToTensor(b, 'b', 'greaterEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.greaterEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: GreaterEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             GreaterEqual) as T;\n}\n\nexport const greaterEqual = op({greaterEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Imag, ImagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n/**\n * Returns the imaginary part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the imaginary part of each element in input considered as a complex number.\n * If input is real, a tensor of all zeros is returned.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.imag(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction imag_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'imag');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.imag($input);\n  };\n\n  const inputs: ImagInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Imag) as T;\n}\n\nexport const imag = op({imag_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsFinite, IsFiniteInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are finite.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isFinite().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isFinite_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isFinite');\n\n  const inputs: IsFiniteInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isFinite($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsFinite);\n}\nexport const isFinite = op({isFinite_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsInf, IsInfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are Infinity or -Infinity.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isInf().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isInf_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isInf');\n\n  const inputs: IsInfInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isInf($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsInf);\n}\nexport const isInf = op({isInf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsNan, IsNanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * RReturns which elements of x are NaN.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isNaN().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isNaN_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isNaN');\n  const inputs: IsNanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.isNaN($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsNan);\n}\nexport const isNaN = op({isNaN_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Maximum, MaximumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the max of a and b (`a > b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `tf.maximumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * ```js\n * // Broadcast maximum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction maximum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'maximum');\n  let $b = convertToTensor(b, 'b', 'maximum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.maximum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: MaximumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Maximum) as T;\n}\n\nexport const maximum = op({maximum_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar} from '../tensor';\nimport {DataType} from '../types';\nimport {isTypedArray} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-0 `tf.Tensor` (scalar) with the provided value and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.scalar` as it makes the code more readable.\n *\n * ```js\n * tf.scalar(3.14).print();\n * ```\n *\n * @param value The value of the scalar.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function scalar(\n    value: number|boolean|string|Uint8Array, dtype?: DataType): Scalar {\n  if (((isTypedArray(value) && dtype !== 'string') || Array.isArray(value)) &&\n      dtype !== 'complex64') {\n    throw new Error(\n        'Error creating a new Scalar: value must be a primitive ' +\n        '(number|boolean|string)');\n  }\n  if (dtype === 'string' && isTypedArray(value) &&\n      !(value instanceof Uint8Array)) {\n    throw new Error(\n        'When making a scalar from encoded string, ' +\n        'the value must be `Uint8Array`.');\n  }\n  const shape: number[] = [];\n  const inferredShape: number[] = [];\n  return makeTensor(value, shape, inferredShape, dtype) as Scalar;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {maximum} from './maximum';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {scalar} from './scalar';\n\n/**\n * Computes leaky rectified linear element-wise.\n *\n * See\n * [http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf](\n *     http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf)\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.leakyRelu(0.1).print();  // or tf.leakyRelu(x, 0.1)\n * ```\n * @param x The input tensor.\n * @param alpha The scaling factor for negative values, defaults to 0.2.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction leakyRelu_<T extends Tensor>(x: T|TensorLike, alpha = 0.2): T {\n  const $x = convertToTensor(x, 'x', 'leakyRelu');\n  return maximum(mul(scalar(alpha), $x), $x);\n}\n\nexport const leakyRelu = op({leakyRelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Less, LessInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a < b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.less(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction less_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'less');\n  let $b = convertToTensor(b, 'b', 'less');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.less($a, $b);\n\n  const inputs: LessInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Less) as\n      T;\n}\n\nexport const less = op({less_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LessEqual, LessEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a <= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.lessEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction lessEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'lessEqual');\n  let $b = convertToTensor(b, 'b', 'lessEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.lessEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: LessEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LessEqual) as T;\n}\n\nexport const lessEqual = op({lessEqual_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LinSpace, LinSpaceAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor1D} from '../tensor';\n\n/**\n * Return an evenly spaced sequence of numbers over the given interval.\n *\n * ```js\n * tf.linspace(0, 9, 10).print();\n * ```\n * @param start The start value of the sequence.\n * @param stop The end value of the sequence.\n * @param num The number of values to generate.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function linspace(start: number, stop: number, num: number): Tensor1D {\n  if (num <= 0) {\n    throw new Error('The number of values should be positive.');\n  }\n\n  const attrs: LinSpaceAttrs = {start, stop, num};\n  return ENGINE.runKernelFunc(\n      backend => backend.linspace(start, stop, num), {} /* inputs */,\n      null /* grad */, LinSpace, attrs as {} as NamedAttrMap);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRN, LRNAttrs, LRNInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Normalizes the activation of a local neighborhood across or within\n * channels.\n *\n * @param x The input tensor. The 4-D input tensor is treated as a 3-D array\n *     of 1D vectors (along the last dimension), and each vector is\n *     normalized independently.\n * @param depthRadius The number of adjacent channels in the 1D normalization\n *     window.\n * @param bias A constant bias term for the basis.\n * @param alpha A scale factor, usually positive.\n * @param beta An exponent.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction localResponseNormalization_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const $x = convertToTensor(x, 'x', 'localResponseNormalization');\n  util.assert(\n      $x.rank === 4 || $x.rank === 3,\n      () => `Error in localResponseNormalization: x must be rank 3 or 4 but got\n               rank ${$x.rank}.`);\n  util.assert(\n      util.isInt(depthRadius),\n      () => `Error in localResponseNormalization: depthRadius must be an ` +\n          `integer but got depthRadius ${depthRadius}.`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.localResponseNormalization4D(\n        x4D, depthRadius, bias, alpha, beta);\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: LRNInputs = {x: x4D};\n\n  const attrs: LRNAttrs = {depthRadius, bias, alpha, beta};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, LRN,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  } else {\n    return res as T;\n  }\n}\n\nexport const localResponseNormalization = op({localResponseNormalization_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log, LogInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` element-wise: `ln(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E]);\n *\n * x.log().print();  // or tf.log(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log');\n\n  const inputs: LogInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log);\n}\nexport const log = op({log_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log1p, Log1pInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` plus one\n * element-wise: `ln(1 + x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E - 1]);\n *\n * x.log1p().print();  // or tf.log1p(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log1p_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log1p');\n\n  const inputs: Log1pInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log1p($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log1p);\n}\nexport const log1p = op({log1p_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {CustomGradientFunc, ENGINE} from './engine';\nimport {Scalar, Tensor, Variable} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {convertToTensor, convertToTensorArray} from './tensor_util_env';\nimport {TensorLike} from './types';\nimport * as util from './util';\n\n/**\n * Provided `f(x)`, returns another function `g(x, dy?)`, which gives the\n * gradient of `f(x)` with respect to `x`.\n *\n * If `dy` is provided, the gradient of `f(x).mul(dy).sum()` with respect to\n * `x` is computed instead. `f(x)` must take a single tensor `x` and return a\n * single tensor `y`. If `f()` takes multiple inputs, use `tf.grads` instead.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.grad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * g(x).print();\n * ```\n *\n * ```js\n * // f(x) = x ^ 3\n * const f = x => x.pow(tf.scalar(3, 'int32'));\n * // f'(x) = 3x ^ 2\n * const g = tf.grad(f);\n * // f''(x) = 6x\n * const gg = tf.grad(g);\n *\n * const x = tf.tensor1d([2, 3]);\n * gg(x).print();\n * ```\n *\n * @param f The function f(x), to compute gradient for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grad(f: (x: Tensor) => Tensor): (\n    x: TensorLike|Tensor, dy?: TensorLike|Tensor) => Tensor {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grad(f) must be a function');\n  return (x: TensorLike|Tensor, dy?: TensorLike|Tensor): Tensor => {\n    // x can be of any dtype, thus null as the last argument.\n    const $x = convertToTensor(x, 'x', 'tf.grad', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grad') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f($x), [$x], $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grad(f)(x, dy) must match the shape ' +\n                'returned by f(x)');\n      }\n      checkGrads(grads);\n      return grads[0];\n    });\n  };\n}\n\n/**\n * Provided `f(x1, x2,...)`, returns another function `g([x1, x2,...], dy?)`,\n * which gives an array of gradients of `f()` with respect to each input\n * [`x1`,`x2`,...].\n *\n * If `dy` is passed when calling `g()`, the gradient of\n * `f(x1,...).mul(dy).sum()` with respect to each input is computed instead.\n * The provided `f` must take one or more tensors and return a single tensor\n * `y`. If `f()` takes a single input, we recommend using `tf.grad` instead.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df / da = b, df / db = a\n * const g = tf.grads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const [da, db] = g([a, b]);\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @param f The function `f(x1, x2,...)` to compute gradients for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grads(f: (...args: Tensor[]) => Tensor): (\n    args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike) => Tensor[] {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grads(f) must be a function');\n  return (args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike): Tensor[] => {\n    util.assert(\n        Array.isArray(args),\n        () => 'The args passed in grads(f)(args) must be an array ' +\n            'of `Tensor`s or `TensorLike`s');\n    // args can be of any dtype, thus null as the last argument.\n    const $args = convertToTensorArray(args, 'args', 'tf.grads', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grads') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f(...$args), $args, $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grads(f)([x1,...], dy) must ' +\n                'match the shape returned by f([x1,...])');\n      }\n      checkGrads(grads);\n      return grads;\n    });\n  };\n}\n\n/**\n * Like `tf.grad`, but also returns the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grad: The gradient of `f(x)` w.r.t `x` (result of `tf.grad`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.valueAndGrad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * const {value, grad} = g(x);\n *\n * console.log('value');\n * value.print();\n * console.log('grad');\n * grad.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrad<I extends Tensor, O extends Tensor>(f: (x: I) => O): (\n    x: I, dy?: O) => {\n  value: O;\n  grad: I;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrad(f) must be a function');\n  return (x: I, dy?: O) => {\n    util.assert(\n        x instanceof Tensor,\n        () => 'The x passed in valueAndGrad(f)(x) must be a tensor');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrad(f)(x, dy) must be a tensor');\n    const {grads, value} = ENGINE.gradients(() => f(x), [x], dy);\n    checkGrads(grads);\n    return {grad: grads[0] as I, value};\n  };\n}\n\n/**\n * Like `tf.grads`, but returns also the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grads: The gradients of `f()` w.r.t each input (result of `tf.grads`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df/da = b, df/db = a\n * const g = tf.valueAndGrads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const {value, grads} = g([a, b]);\n *\n * const [da, db] = grads;\n *\n * console.log('value');\n * value.print();\n *\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrads<O extends Tensor>(f: (...args: Tensor[]) => O): (\n    args: Tensor[], dy?: O) => {\n  grads: Tensor[];\n  value: O;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrads(f) must be a function');\n  return (args: Tensor[], dy?: O) => {\n    util.assert(\n        Array.isArray(args) && args.every(arg => arg instanceof Tensor),\n        () => 'The args passed in valueAndGrads(f)(args) must be array of ' +\n            'tensors');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrads(f)(args, dy) must be a tensor');\n    const res = ENGINE.gradients(() => f(...args), args, dy);\n    if (dy != null) {\n      util.assertShapesMatch(\n          res.value.shape, dy.shape,\n          'The shape of dy passed in valueAndGrads(f)([x1,...], dy) must ' +\n              'match the shape returned by f([x1,...])');\n    }\n    checkGrads(res.grads);\n    return res;\n  };\n}\n\n/**\n * Computes and returns the gradient of f(x) with respect to the list of\n * trainable variables provided by `varList`. If no list is provided, it\n * defaults to all trainable variables.\n *\n * ```js\n * const a = tf.variable(tf.tensor1d([3, 4]));\n * const b = tf.variable(tf.tensor1d([5, 6]));\n * const x = tf.tensor1d([1, 2]);\n *\n * // f(a, b) = a * x ^ 2 + b * x\n * const f = () => a.mul(x.square()).add(b.mul(x)).sum();\n * // df/da = x ^ 2, df/db = x\n * const {value, grads} = tf.variableGrads(f);\n *\n * Object.keys(grads).forEach(varName => grads[varName].print());\n * ```\n *\n * @param f The function to execute. f() should return a scalar.\n * @param varList The list of variables to compute the gradients with respect\n *     to. Defaults to all trainable variables.\n * @returns An object with the following keys and values:\n *   - `value`: The value of the function `f`.\n *   - `grads`: A map from the names of the variables to the gradients.\n *     If the `varList` argument is provided explicitly and contains a subset of\n *     non-trainable variables, this map in the return value will contain keys\n *     that map the names of the non-trainable variables to `null`.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction variableGrads(f: () => Scalar, varList?: Variable[]):\n    {value: Scalar, grads: NamedTensorMap} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in variableGrads(f) must be a function');\n  util.assert(\n      varList == null ||\n          Array.isArray(varList) && varList.every(v => v instanceof Variable),\n      () =>\n          'The varList passed in variableGrads(f, varList) must be an array ' +\n          'of variables');\n\n  const specifiedVarList = varList != null;\n  if (!specifiedVarList) {\n    // Get all of the trainable variables.\n    varList = [];\n    for (const varName in ENGINE.registeredVariables) {\n      varList.push(ENGINE.registeredVariables[varName]);\n    }\n  }\n\n  const specifiedNonTrainable: Variable[] =\n      specifiedVarList ? varList.filter(variable => !variable.trainable) : null;\n\n  // Prune non-trainable variables.\n  const originalVarCount = varList.length;\n  varList = varList.filter(variable => variable.trainable);\n  util.assert(\n      varList.length > 0,\n      () => `variableGrads() expects at least one of the input variables to ` +\n          `be trainable, but none of the ${originalVarCount} variables is ` +\n          `trainable.`);\n\n  const allowNoGradients = true;\n  const {value, grads} = ENGINE.gradients(f, varList, null, allowNoGradients);\n\n  util.assert(\n      grads.some(g => g != null),\n      () => 'Cannot find a connection between any variable and the result of ' +\n          'the loss function y=f(x). Please make sure the operations that ' +\n          'use variables are inside the function f passed to minimize().');\n  util.assert(\n      value.rank === 0,\n      () => `The f passed in variableGrads(f) must return a scalar, but it ` +\n          `returned a rank-${value.rank} tensor`);\n\n  const namedGrads: NamedTensorMap = {};\n  varList.forEach((v, i) => {\n    if (grads[i] != null) {\n      namedGrads[v.name] = grads[i];\n    }\n  });\n  if (specifiedNonTrainable != null) {\n    // If varList is explicitly provided and contains non-trainable values,\n    // add them to the returned gradients with `null` values.\n    specifiedNonTrainable.forEach(v => namedGrads[v.name] = null);\n  }\n  return {value, grads: namedGrads};\n}\n\n/**\n * Overrides the gradient computation of a function `f`.\n *\n * Takes a function\n * `f(...inputs, save) => {value: Tensor, gradFunc: (dy, saved) => Tensor[]}`\n * and returns another function `g(...inputs)` which takes the same inputs as\n * `f`. When called, `g` returns `f().value`. In backward mode, custom gradients\n * with respect to each input of `f` are computed using `f().gradFunc`.\n *\n * The `save` function passsed to `f` should be used for saving tensors needed\n * in the gradient. And the `saved` passed to the `gradFunc` is a\n * `NamedTensorMap`, which contains those saved tensor.\n *\n * ```js\n * const customOp = tf.customGrad((x, save) => {\n *   // Save x to make sure it's available later for the gradient.\n *   save([x]);\n *   // Override gradient of our custom x ^ 2 op to be dy * abs(x);\n *   return {\n *     value: x.square(),\n *     // Note `saved.x` which points to the `x` we saved earlier.\n *     gradFunc: (dy, saved) => [dy.mul(saved[0].abs())]\n *   };\n * });\n *\n * const x = tf.tensor1d([-1, -2, 3]);\n * const dx = tf.grad(x => customOp(x));\n *\n * console.log(`f(x):`);\n * customOp(x).print();\n * console.log(`f'(x):`);\n * dx(x).print();\n * ```\n *\n * @param f The function to evaluate in forward mode, which should return\n *     `{value: Tensor, gradFunc: (dy, saved) => Tensor[]}`, where `gradFunc`\n *     returns the custom gradients of `f` with respect to its inputs.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n    (...args: Tensor[]) => T {\n  return ENGINE.customGrad(f);\n}\n\nfunction checkGrads(grads: Tensor[]) {\n  const numNullGradients = grads.filter(g => g == null).length;\n  if (numNullGradients > 0) {\n    throw new Error(\n        `Cannot compute gradient of y=f(x) with respect to x. Make sure that\n    the f you passed encloses all operations that lead from x to y.`);\n  }\n}\n\nexport {\n  customGrad,\n  variableGrads,\n  valueAndGrad,\n  valueAndGrads,\n  grad,\n  grads,\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Negate, NegateInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes `-1 * x` element-wise.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, -2, 0], [2, 2]);\n *\n * x.neg().print();  // or tf.neg(x)\n * ```\n *\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction neg_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'neg');\n\n  const inputs: NegateInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.neg($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Negate);\n}\nexport const neg = op({neg_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softplus, SoftplusInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes softplus of the input `tf.Tensor` element-wise: `log(exp(x) + 1)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.softplus().print();  // or tf.softplus(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction softplus_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'softplus');\n\n  const inputs: SoftplusInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.softplus($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Softplus);\n}\nexport const softplus = op({softplus_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {mul} from './mul';\nimport {neg} from './neg';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {softplus} from './softplus';\n\n/**\n * Computes log sigmoid of the input `tf.Tensor` element-wise:\n * `logSigmoid(x)`. For numerical stability, we use `-tf.softplus(-x)`.\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.logSigmoid().print();  // or tf.logSigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction logSigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logSigmoid');\n\n  // Use a custom gradient to maintain previous implementation.\n  // There is no LogSigmoid kernel in TF so we can't use engine.runKernel\n  // directly\n  const customOp = customGrad((x: Tensor) => {\n    // TODO(yassogba) we can remove the chained softplus call here only\n    // after backends have modualrized softplus at which point we can call\n    // engine runKernel(..., Sotfplus, ...) directly.\n    const value = neg(softplus(neg(x)));\n\n    const gradFunc = (dy: T) => {\n      const derX = mul(dy, sigmoid(neg(x)));\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\nexport const logSigmoid = op({logSigmoid_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Max, MaxAttrs, MaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the maximum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.max().print();  // or tf.max(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.max(axis).print();  // or tf.max(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction max_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'max');\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = util.parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let maxInput = $x;\n        if (permutedAxes != null) {\n          maxInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, maxInput.rank);\n        }\n\n        const y = backend.max(maxInput, axes);\n        if (permutedAxes != null) {\n          maxInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape = axis_util.expandShapeToKeepDim(\n              res.shape, util.parseAxisParam(axis, $x.shape));\n          res = reshape(res, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MaxInputs = {x: $x};\n  const attrs: MaxAttrs = {reductionIndices: axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Max,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const max = op({max_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sub, SubInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Subtracts two `tf.Tensor`s element-wise, A - B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n *\n * ```js\n * // Broadcast subtract a with b.\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.scalar(5);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n * @param a The first `tf.Tensor` to subtract from.\n * @param b The second `tf.Tensor` to be subtracted. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction sub_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'sub');\n  let $b = convertToTensor(b, 'b', 'sub');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.subtract($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SubInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sub) as\n      T;\n}\n\nexport const sub = op({sub_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sum, SumAttrs, SumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the sum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If axes has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.sum().print();  // or tf.sum(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.sum(axis).print();  // or tf.sum(x, axis)\n * ```\n *\n * @param x The input tensor to compute the sum over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction sum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'sum');\n  if ($x.dtype === 'bool') {\n    $x = cast($x, 'int32');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.sum(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n    return value;\n  };\n\n  const inputs: SumInputs = {x: $x};\n  const attrs: SumAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const sum = op({sum_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LogSoftmax, LogSoftmaxAttrs, LogSoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log softmax.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param axis The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction logSoftmax_<T extends Tensor>(logits: T|TensorLike, axis = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'logSoftmax');\n\n  if (axis === -1) {\n    axis = $logits.rank - 1;\n  }\n  if (axis !== $logits.rank - 1) {\n    throw Error(\n        'Log Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and axis was ${axis}`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const keepDims = true;\n    const xMax = max(logits, axis, true);\n    const shifted = sub(logits, xMax);\n    const value =\n        sub(cast(shifted, 'float32'), log(sum(exp(shifted), axis, keepDims)));\n    save([value]);\n    return value;\n  };\n\n  const inputs: LogSoftmaxInputs = {logits: $logits};\n  const attrs: LogSoftmaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LogSoftmax, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const logSoftmax = op({logSoftmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {add} from './add';\nimport {expandShapeToKeepDim} from './axis_util';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log(sum(exp(elements across the reduction dimensions)).\n *\n * Reduces the input along the dimensions given in `axis`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.logSumExp().print();  // or tf.logSumExp(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.logSumExp(axis).print();  // or tf.logSumExp(a, axis)\n * ```\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. If null (the default),\n *     reduces all dimensions.\n * @param keepDims If true, retains reduced dimensions with length\n *     of 1. Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction logSumExp_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'logSumExp');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const xMax = max($x, axes, true /* keepDims */);\n  const a = sub($x, xMax);\n  const b = exp(a);\n  const c = sum(b, axes);\n  const d = log(c);\n  const res = add(reshape(xMax, d.shape), d);\n\n  if (keepDims) {\n    const newShape = expandShapeToKeepDim(res.shape, axes);\n    return reshape(res, newShape) as T;\n  }\n  return res as T;\n}\n\nexport const logSumExp = op({logSumExp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {LogicalAnd, LogicalAndInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a AND b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalAnd(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalAnd_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalAnd', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalAnd', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalAndInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalAnd($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalAnd) as T;\n}\n\nexport const logicalAnd = op({logicalAnd_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalNot, LogicalNotInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `NOT x` element-wise.\n *\n * ```js\n * const a = tf.tensor1d([false, true], 'bool');\n *\n * a.logicalNot().print();\n * ```\n *\n * @param x The input tensor. Must be of dtype 'bool'.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalNot_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logicalNot', 'bool');\n  const inputs: LogicalNotInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.logicalNot($x), inputs as {} as NamedTensorMap,\n      null /* grad */, LogicalNot);\n}\n\nexport const logicalNot = op({logicalNot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalOr, LogicalOrInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a OR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalOr(b).print();\n * ```\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalOr_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalOr', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalOr', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalOrInputs = {a: $a, b: $b};\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalOr($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalOr) as T;\n}\nexport const logicalOr = op({logicalOr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {logicalAnd} from './logical_and';\nimport {logicalNot} from './logical_not';\nimport {logicalOr} from './logical_or';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a XOR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalXor(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalXor_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalXor', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalXor', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  // x ^ y = (x | y) & ~(x & y)\n  return logicalAnd(logicalOr(a, b), logicalNot(logicalAnd(a, b)));\n}\n\nexport const logicalXor = op({logicalXor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool, MaxPoolAttrs, MaxPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D max pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction maxPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'maxPool');\n  const dilations = 1;\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in maxPool: input must be rank 4 but got rank ${x4D.rank}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    let y;\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      y = x4D.clone();\n    } else {\n      y = backend.maxPool(x4D, convInfo);\n    }\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: MaxPoolInputs = {x: x4D};\n\n  const attrs: MaxPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const maxPool = op({maxPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {MaxPool3D, MaxPool3DAttrs, MaxPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D max pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.maxPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number = [1, 1, 1],\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'maxPool3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in maxPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in maxPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n    const y = backend.maxPool3d(x5D, convInfo);\n    save([x5D, y]);\n    return y;\n  };\n\n  const inputs: MaxPool3DInputs = {x: x5D};\n\n  const attrs: MaxPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3d = op({maxPool3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the 2D max pooling of an image with Argmax index.\n * The indices in argmax are flattened, so that a maximum value at position `[b,\n * y, x, c]` becomes flattened index: `(y * width + x) * channels + c` if\n * include_batch_in_index is False; `((b * height + y) * width + x) * channels\n * +c` if include_batch_in_index is True.\n *\n * The indices returned are always in `[0, height) x [0, width)` before\n * flattening.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param includeBatchIndex Defaults to False. Whether to include batch\n *    dimension in flattened index of argmax.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPoolWithArgmax_<T extends Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    includeBatchInIndex = false): NamedTensorMap {\n  const $x = convertToTensor(x, 'x', 'maxPoolWithArgmax');\n\n  const inputs: MaxPoolWithArgmaxInputs = {x: $x};\n  const attrs:\n      MaxPoolWithArgmaxAttrs = {filterSize, strides, pad, includeBatchInIndex};\n\n  const result = ENGINE.runKernel(\n                     MaxPoolWithArgmax, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {result: result[0], indexes: result[1]};\n}\n\nexport const maxPoolWithArgmax = op({maxPoolWithArgmax_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeZerosTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0.\n *\n * ```js\n * tf.zeros([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Can\n *     be 'float32', 'int32' or 'bool'. Defaults to 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function zeros<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = zeros(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeZerosTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeOnesTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\nimport {zeros} from './zeros';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1.\n *\n * ```js\n * tf.ones([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n *     'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function ones<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = ones(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeOnesTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam, sizeFromShape} from '../util';\n\nimport {computeOutAndReduceShapes} from './axis_util';\nimport {cast} from './cast';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {ones} from './ones';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sum} from './sum';\n\n/**\n * Computes the mean of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces `x` along the dimensions given in `axis`. Unless `keepDims` is\n * true, the rank of the `tf.Tensor` is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and a `tf.Tensor` with\n * a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.mean().print();  // or tf.mean(a)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.mean(axis).print();  // or tf.mean(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction mean_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'mean');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const shapes = computeOutAndReduceShapes($x.shape, axes);\n  const reduceShape = shapes[1];\n  const reduceSize = sizeFromShape(reduceShape);\n\n  // Use a custom gradient to bypass 2 gradient backprops since mean is used\n  // extremely often.\n  const customOp = customGrad((x: Tensor) => {\n    const reduceSizeScalar = scalar(reduceSize);\n    // Cast if needed.\n    const xReduce = reduceSizeScalar.dtype === x.dtype ?\n        x :\n        cast(x, reduceSizeScalar.dtype);\n    const res = div(xReduce, reduceSizeScalar);\n    const value = sum(res, axis, keepDims);\n\n    const gradFunc = (dy: Tensor) => {\n      const expandedDyShape = x.shape.slice();\n      axes.forEach(axis => {\n        expandedDyShape[axis] = 1;\n      });\n      const expandedDy = reshape(dy, expandedDyShape);\n      const derX = div(mul(expandedDy, ones(x.shape, 'float32')), reduceSize);\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\n\nexport const mean = op({mean_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Min, MinAttrs, MinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the minimum value from the input.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axes`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axes` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.min().print();  // or tf.min(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.min(axis).print();  // or tf.min(x, axis)\n * ```\n *\n * @param x The input Tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction min_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'min');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let minInput = $x;\n        if (permutedAxes != null) {\n          minInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n        }\n\n        const y = backend.min(minInput, axes);\n        if (permutedAxes != null) {\n          minInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape =\n              axis_util.expandShapeToKeepDim(res.shape, origAxes);\n          res = reshape(y, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MinInputs = {x: $x};\n  const attrs: MinAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Min,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const min = op({min_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Minimum, MinimumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the min of a and b (`a < b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `minimumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * ```js\n * // Broadcast minimum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction minimum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'minimum');\n  let $b = convertToTensor(b, 'b', 'minimum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.minimum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MinimumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Minimum) as T;\n}\n\nexport const minimum = op({minimum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Mod, ModInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns the mod of a and b element-wise.\n * `floor(x / y) * y + mod(x, y) = x`\n * Supports broadcasting.\n *\n * We also expose `tf.modStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * ```js\n * // Broadcast a mod b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mod_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mod');\n  let $b = convertToTensor(b, 'b', 'mod');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.mod($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: ModInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Mod) as T;\n}\n\nexport const mod = op({mod_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Computes square of `x` element-wise: `x ^ 2`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.sqrt(2), -1]);\n *\n * x.square().print();  // or tf.square(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction square_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'square');\n  const attrs = {};\n  const inputsToSave = [$x];\n  const outputsToSave: boolean[] = [];\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    return backend.square($x);\n  }, {x: $x}, null /* grad */, 'Square', attrs, inputsToSave, outputsToSave);\n}\n\nexport const square = op({square_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim} from './axis_util';\nimport {cast} from './cast';\nimport {mean} from './mean';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {square} from './square';\nimport {sub} from './sub';\n\n/**\n * Calculates the mean and variance of `x`. The mean and variance are\n * calculated by aggregating the contents of `x` across `axes`. If `x` is\n * 1-D and `axes = [0]` this is just the mean and variance of a vector.\n *\n * @param x The input tensor.\n * @param axis The dimension(s) along with to compute mean and\n *     variance. By default it reduces all dimensions.\n * @param keepDims If true, the moments have the same dimensionality as the\n *     input.\n * @return An object with two keys: `mean` and `variance`.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction moments_(\n    x: Tensor|TensorLike, axis: number|number[] = null,\n    keepDims = false): {mean: Tensor, variance: Tensor} {\n  x = convertToTensor(x, 'x', 'moments');\n  const axes = parseAxisParam(axis, x.shape);\n  const xMean = mean(x, axes, keepDims);\n  let keepDimsShape = xMean.shape;\n  if (!keepDims) {\n    keepDimsShape = expandShapeToKeepDim(xMean.shape, axes);\n  }\n  const devSquared =\n      square(sub(cast(x, 'float32'), reshape(xMean, keepDimsShape)));\n  const variance = mean(devSquared, axes, keepDims);\n  return {mean: xMean, variance};\n}\n\nexport const moments = op({moments_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor, convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * @docalias (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D]\n */\nexport type LSTMCellFunc = {\n  (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D];\n};\n\n/**\n * Computes the next states and outputs of a stack of LSTMCells.\n *\n * Each cell output is used as input to the next cell.\n *\n * Returns `[cellState, cellOutput]`.\n *\n * Derived from tf.contrib.rn.MultiRNNCell.\n *\n * @param lstmCells Array of LSTMCell functions.\n * @param data The input to the cell.\n * @param c Array of previous cell states.\n * @param h Array of previous cell outputs.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction multiRNNCell_(\n    lstmCells: LSTMCellFunc[], data: Tensor2D|TensorLike,\n    c: Array<Tensor2D|TensorLike>,\n    h: Array<Tensor2D|TensorLike>): [Tensor2D[], Tensor2D[]] {\n  const $data = convertToTensor(data, 'data', 'multiRNNCell');\n  const $c = convertToTensorArray(c, 'c', 'multiRNNCell');\n  const $h = convertToTensorArray(h, 'h', 'multiRNNCell');\n\n  let input = $data;\n  const newStates = [];\n  for (let i = 0; i < lstmCells.length; i++) {\n    const output = lstmCells[i](input, $c[i], $h[i]);\n    newStates.push(output[0]);\n    newStates.push(output[1]);\n    input = output[1];\n  }\n  const newC: Tensor2D[] = [];\n  const newH: Tensor2D[] = [];\n  for (let i = 0; i < newStates.length; i += 2) {\n    newC.push(newStates[i]);\n    newH.push(newStates[i + 1]);\n  }\n  return [newC, newH];\n}\nexport const multiRNNCell = op({multiRNNCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a `tf.Tensor` with values drawn from a multinomial distribution.\n *\n * ```js\n * const probs = tf.tensor([.75, .25]);\n * tf.multinomial(probs, 3).print();\n * ```\n *\n * @param logits 1D array with unnormalized log-probabilities, or\n *     2D array of shape `[batchSize, numOutcomes]`. See the `normalized`\n *     parameter.\n * @param numSamples Number of samples to draw for each row slice.\n * @param seed The seed number.\n * @param normalized Whether the provided `logits` are normalized true\n *     probabilities (sum to 1). Defaults to false.\n * @return 1D array of shape `[numSamples]`, or 2D array of shape\n *     `[batchSize, numSamples]`, depending on the rank of the input.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction multinomial_(\n    logits: Tensor1D|Tensor2D|TensorLike, numSamples: number, seed?: number,\n    normalized = false): Tensor1D|Tensor2D {\n  const $logits = convertToTensor(logits, 'logits', 'multinomial');\n  const numOutcomes = $logits.size;\n  const origRank = $logits.rank;\n  if (numOutcomes < 2) {\n    throw new Error(\n        `Error in multinomial: you need at least 2 outcomes, but got ` +\n        `${numOutcomes}.`);\n  }\n  if (origRank > 2) {\n    throw new Error(`Rank of probabilities must be 1 or 2, but is ${origRank}`);\n  }\n  seed = seed || Math.random();\n  const logits2D: Tensor2D =\n      origRank === 1 ? reshape($logits, [1, -1]) : $logits as Tensor2D;\n  const res = ENGINE.runKernelFunc(\n      backend => backend.multinomial(logits2D, normalized, numSamples, seed),\n      {logits2D});\n  // tslint:disable-next-line:no-unnecessary-type-assertion\n  return origRank === 1 ? reshape(res, [res.size]) as Tensor1D : res;\n}\n\nexport const multinomial = op({multinomial_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {NotEqual, NotEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a != b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([0, 2, 3]);\n *\n * a.notEqual(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction notEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'notEqual');\n  let $b = convertToTensor(b, 'b', 'notEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => backend.notEqual($a, $b);\n\n  const inputs: NotEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             NotEqual) as T;\n}\n\nexport const notEqual = op({notEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Real, RealInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the real part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the real part of each element in input considered as a complex number.\n *\n * If the input is real, it simply makes a clone.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.real(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction real_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'real');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.real($input);\n  };\n\n  const inputs: RealInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Real) as T;\n}\n\nexport const real = op({real_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OnesLike, OnesLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {complex} from './complex';\nimport {imag} from './imag';\nimport {op} from './operation';\nimport {real} from './real';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.onesLike(x).print();\n * ```\n * @param x A tensor.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction onesLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'onesLike');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if ($x.dtype === 'complex64') {\n      const r = onesLike(real($x));\n      const i = zerosLike(imag($x));\n      return complex(r, i);\n    }\n\n    return backend.onesLike($x);\n  };\n\n  const inputs: OnesLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             OnesLike) as T;\n}\n\nexport const onesLike = op({onesLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the outer product of two vectors, `v1` and `v2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([3, 4, 5]);\n *\n * tf.outerProduct(a, b).print();\n * ```\n * @param v1 The first vector in the outer product operation.\n * @param v2 The second vector in the outer product operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction outerProduct_(\n    v1: Tensor1D|TensorLike, v2: Tensor1D|TensorLike): Tensor2D {\n  const $v1 = convertToTensor(v1, 'v1', 'outerProduct');\n  const $v2 = convertToTensor(v2, 'v2', 'outerProduct');\n\n  util.assert(\n      $v1.rank === 1 && $v2.rank === 1,\n      () => `Error in outerProduct: inputs must be rank 1, but got ranks ` +\n          `${$v1.rank} and ${$v2.rank}.`);\n\n  const v12D = reshape($v1, [-1, 1]);\n  const v22D = reshape($v2, [1, -1]);\n  return matMul(v12D, v22D) as Tensor2D;\n}\n\nexport const outerProduct = op({outerProduct_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {PadV2, PadV2Attrs, PadV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Pads a `tf.Tensor` with a given value and paddings.\n *\n * This operation currently only implements the `CONSTANT` mode.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `paddings` is of given length.\n *   - `tf.pad1d`\n *   - `tf.pad2d`\n *   - `tf.pad3d`\n *   - `tf.pad4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.pad([[1, 2]]).print();\n * ```\n * @param x The tensor to pad.\n * @param paddings An array of length `R` (the rank of the tensor), where\n * each element is a length-2 tuple of ints `[padBefore, padAfter]`,\n * specifying how much to pad along each dimension of the tensor.\n * @param constantValue The pad value to use. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction pad_<T extends Tensor>(\n    x: T|TensorLike, paddings: Array<[number, number]>, constantValue = 0): T {\n  const $x = convertToTensor(x, 'x', 'pad');\n  if ($x.rank === 0) {\n    throw new Error('pad(scalar) is not defined. Pass non-scalar to pad');\n  }\n  const forward: ForwardFunc<T> = (backend, save) => {\n    save([$x]);\n    return backend.pad($x, paddings, constantValue);\n  };\n\n  const attrs: PadV2Attrs = {paddings, constantValue};\n  const inputs: PadV2Inputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, PadV2,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const pad = op({pad_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor1D` with a given value and paddings. See `pad` for details.\n */\nfunction pad1d_(\n    x: Tensor1D|TensorLike, paddings: [number, number],\n    constantValue = 0): Tensor1D {\n  assert(\n      paddings.length === 2,\n      () => 'Invalid number of paddings. Must be length of 2.');\n  return pad(x, [paddings], constantValue);\n}\n\nexport const pad1d = op({pad1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor2D` with a given value and paddings. See `pad` for details.\n */\nfunction pad2d_(\n    x: Tensor2D|TensorLike, paddings: [[number, number], [number, number]],\n    constantValue = 0): Tensor2D {\n  assert(\n      paddings.length === 2 && paddings[0].length === 2 &&\n          paddings[1].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad2d = op({pad2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor3D` with a given value and paddings. See `pad` for details.\n */\nfunction pad3d_(\n    x: Tensor3D|TensorLike,\n    paddings: [[number, number], [number, number], [number, number]],\n    constantValue = 0): Tensor3D {\n  assert(\n      paddings.length === 3 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad3d = op({pad3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor4D` with a given value and paddings. See `pad` for details.\n */\nfunction pad4d_(\n    x: Tensor4D|TensorLike,\n    paddings:\n        [\n          [number, number], [number, number], [number, number], [number, number]\n        ],\n    constantValue = 0): Tensor4D {\n  assert(\n      paddings.length === 4 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2 &&\n          paddings[3].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad4d = op({pad4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation divides \"spatial\" dimensions `[1, ..., M]` of the input into\n * a grid of blocks of shape `blockShape`, and interleaves these blocks with\n * the \"batch\" dimension (0) such that in the output, the spatial\n * dimensions `[1, ..., M]` correspond to the position within the grid,\n * and the batch dimension combines both the position within a spatial block\n * and the original batch position. Prior to division into blocks,\n * the spatial dimensions of the input are optionally zero padded\n * according to `paddings`. See below for a precise description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);\n * const blockShape = [2, 2];\n * const paddings = [[0, 0], [0, 0]];\n *\n * x.spaceToBatchND(blockShape, paddings).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param paddings A 2-D array. Must have shape `[M, 2]`, all values must be >=\n *     0. `paddings[i] = [padStart, padEnd]` specifies the amount to zero-pad\n * from input dimension `i + 1`, which corresponds to spatial dimension `i`. It\n * is required that\n * `(inputShape[i + 1] + padStart + padEnd) % blockShape[i] === 0`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Zero-pad the start and end of dimensions `[1, ..., M]` of the input\n * according to `paddings` to produce `padded` of shape paddedShape.\n *\n * 2. Reshape `padded` to `reshapedPadded` of shape:\n * `[batch] + [paddedShape[1] / blockShape[0], blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1], blockShape[M-1]] + remainingShape`\n *\n * 3. Permute dimensions of `reshapedPadded` to produce `permutedReshapedPadded`\n * of shape: `blockShape + [batch] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * 4. Reshape `permutedReshapedPadded` to flatten `blockShape` into the\n * batch dimension, producing an output tensor of shape:\n * `[batch * prod(blockShape)] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction spaceToBatchND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], paddings: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'spaceToBatchND');\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank ${$x.rank} should be > than [blockShape] ${\n          blockShape.length}`);\n\n  util.assert(\n      paddings.length === blockShape.length,\n      () => `paddings.shape[0] ${\n          paddings.length} must be equal to [blockShape] ${blockShape.length}`);\n\n  util.assert(\n      $x.shape.reduce(\n          (a, b, i) => {\n            if (i > 0 && i <= blockShape.length) {\n              return a &&\n                  ((b + paddings[i - 1][0] + paddings[i - 1][1]) %\n                       blockShape[i - 1] ===\n                   0);\n            }\n            return a;\n          },\n          true),\n      () => `input spatial dimensions ${$x.shape.slice(1)} with paddings ${\n          paddings.toString()} must be divisible by blockShapes ${\n          blockShape.toString()}`);\n\n  const forward: ForwardFunc<T> = backend =>\n      backend.spaceToBatchND($x, blockShape, paddings);\n\n  const inputs: SpaceToBatchNDInputs = {x: $x};\n  const attrs: SpaceToBatchNDAttrs = {blockShape, paddings};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      SpaceToBatchND, attrs as {} as NamedAttrMap);\n}\n\nexport const spaceToBatchND = op({spaceToBatchND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {avgPool} from './avg_pool';\nimport {batchToSpaceND} from './batch_to_space_nd';\nimport * as conv_util from './conv_util';\nimport {maxPool} from './max_pool';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {spaceToBatchND} from './space_to_batch_nd';\n\n/**\n * Performs an N-D pooling operation\n *\n * @param input The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param windowShape The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param poolingType The type of pooling, either 'max' or 'avg'.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilationRate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction pool_<T extends Tensor3D|Tensor4D>(\n    input: T|TensorLike, windowShape: [number, number]|number,\n    poolingType: 'avg'|'max', pad: 'valid'|'same'|number,\n    dilations?: [number, number]|number, strides?: [number, number]|number) {\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  if (strides == null) {\n    strides = 1;\n  }\n  if (pad === 0) {\n    pad = 'valid';\n  }\n\n  const $x = convertToTensor(input, 'x', 'maxPool');\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in pool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = conv_util.computePool2DInfo(\n      x4D.shape, windowShape, strides, dilations, pad);\n  const dilation: [number, number] =\n      [convInfo.dilationHeight, convInfo.dilationWidth];\n\n  // The following implementation does batchToSpace(pool(spaceToBatch(x)))\n  // whenever dilation > 1 since the TF kernels do not support dilation > 1.\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L1037\n\n  let basePadding: number[][];\n  if (pad === 'same') {\n    basePadding = withSpaceToBatchBasePaddings(\n        [convInfo.filterHeight, convInfo.filterWidth], dilation);\n  } else {\n    basePadding = [[0, 0], [0, 0]];\n  }\n\n  const isDilationOne = dilation[0] === 1 && dilation[1] === 1;\n  const [adjustedPadding, adjustedCrops] = requiredSpaceToBatchPaddings(\n      [convInfo.inHeight, convInfo.inWidth], dilation, basePadding);\n  const convertedPad = isDilationOne ? pad : 'valid';\n  const convertedX =\n      isDilationOne ? x4D : spaceToBatchND(x4D, dilation, adjustedPadding);\n\n  const forwardOp = poolingType === 'avg' ?\n      () => avgPool(convertedX, windowShape, strides, convertedPad) :\n      () => maxPool(convertedX, windowShape, strides, convertedPad);\n  const y = forwardOp();\n\n  const res = isDilationOne ? y : batchToSpaceND(y, dilation, adjustedCrops);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\n// Helper function to compute crops and paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/array_ops.py#L2184\nfunction requiredSpaceToBatchPaddings(\n    inputShape: [number, number], blockShape: [number, number],\n    basePadding: number[][]) {\n  const padStart = basePadding.map(b => b[0]);\n  const origPadEnd = basePadding.map(b => b[1]);\n  const fullInputShape = inputShape.concat(padStart, origPadEnd);\n  const padEndExtra = blockShape.map((b, i) => (b - fullInputShape[i] % b) % b);\n  const padEnd = origPadEnd.map((s, i) => s + padEndExtra[i]);\n  const paddings = blockShape.map((_, i) => [padStart[i], padEnd[i]]);\n  const crops = blockShape.map((_, i) => [0, padEndExtra[i]]);\n  return [paddings, crops];\n}\n\n// Helper function to compute base paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L524\nfunction withSpaceToBatchBasePaddings(\n    filterShape: [number, number], dilation: [number, number]) {\n  // Spatial dimensions of the filters and the upsampled filters in which we\n  // introduce (rate - 1) zeros between consecutive filter values.\n  const dilatedFilterShape = filterShape.map((s, i) => {\n    return s + (s - 1) * (dilation[i] - 1);\n  });\n  const padExtraShape = dilatedFilterShape.map(s => s - 1);\n\n  // When padding is odd, we pad more at end, following the same\n  // convention as conv2d.\n  const padExtraStart = padExtraShape.map(s => Math.floor(s / 2));\n  const padExtraEnd = padExtraShape.map((s, i) => s - padExtraStart[i]);\n  return padExtraShape.map((_, i) => {\n    return [padExtraStart[i], padExtraEnd[i]];\n  });\n}\n\nexport const pool = op({pool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Pow, PowInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the power of one `tf.Tensor` to another. Supports broadcasting.\n *\n * Given a `tf.Tensor` x and a `tf.Tensor` y, this operation computes x^y for\n * corresponding elements in x and y. The result's dtype will be the upcasted\n * type of the `base` and `exp` dtypes.\n *\n * ```js\n * const a = tf.tensor([[2, 3], [4, 5]])\n * const b = tf.tensor([[1, 2], [3, 0]]).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n *\n * ```js\n * const a = tf.tensor([[1, 2], [3, 4]])\n * const b = tf.tensor(2).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n * We also expose `powStrict` which has the same signature as this op and\n * asserts that `base` and `exp` are the same shape (does not broadcast).\n *\n * @param base The base `tf.Tensor` to pow element-wise.\n * @param exp The exponent `tf.Tensor` to pow element-wise.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction pow_<T extends Tensor>(\n    base: Tensor|TensorLike, exp: Tensor|TensorLike): T {\n  let $base = convertToTensor(base, 'base', 'pow');\n  let $exp = convertToTensor(exp, 'exp', 'pow');\n  [$base, $exp] = makeTypesMatch($base, $exp);\n\n  const inputs: PowInputs = {a: $base, b: $exp};\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.pow($base, $exp);\n    save([$base, $exp, y]);\n    return y;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Pow) as T;\n}\n\nexport const pow = op({pow_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prelu, PreluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes leaky rectified linear element-wise with parametric alphas.\n *\n * `x < 0 ? alpha * x : f(x) = x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n * const alpha = tf.scalar(0.1);\n *\n * x.prelu(alpha).print();  // or tf.prelu(x, alpha)\n * ```\n * @param x The input tensor.\n * @param alpha Scaling factor for negative values.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction prelu_<T extends Tensor>(x: T|TensorLike, alpha: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'prelu');\n  const $alpha = convertToTensor(alpha, 'alpha', 'prelu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.prelu($x, $alpha);\n    save([$x, $alpha]);\n    return res;\n  };\n\n  const inputs: PreluInputs = {x: $x, alpha: $alpha};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prelu) as\n      T;\n}\n\nexport const prelu = op({prelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prod, ProdAttrs, ProdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the product of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.prod().print();  // or tf.prod(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.prod(axis).print();  // or tf.prod(x, axis)\n * ```\n *\n * @param x The input tensor to compute the product over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction prod_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'prod');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if ($x.dtype === 'bool') {\n      $x = cast($x, 'int32');\n    }\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.prod(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n\n    return value as T;\n  };\n\n  const inputs: ProdInputs = {x: $x};\n  const attrs: ProdAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prod,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const prod = op({prod_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a random number generator\n * function defined by the user.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param randFunction A random number generator function which is called\n * for each element in the output tensor.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n */\nfunction rand_<R extends Rank>(\n    shape: ShapeMap[R], randFunction: () => number,\n    dtype?: DataType): Tensor<R> {\n  const size = sizeFromShape(shape);\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  for (let i = 0; i < size; i++) {\n    values[i] = randFunction();\n  }\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n\nexport const rand = op({rand_});\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as seedrandom from 'seedrandom';\n\nimport {expectNumbersClose, testEpsilon} from '../test_util';\nimport {TypedArray} from '../types';\n\nexport interface RandomBase {\n  nextValue(): number;\n}\n\nexport interface RandomGamma {\n  nextValue(): number;\n}\n\nexport interface RandNormalDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\nexport interface RandGammaDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\n// https://en.wikipedia.org/wiki/Marsaglia_polar_method\nexport class MPRandGauss implements RandomBase {\n  private mean: number;\n  private stdDev: number;\n  private nextVal: number;\n  private dtype?: keyof RandNormalDataTypes;\n  private truncated?: boolean;\n  private upper?: number;\n  private lower?: number;\n  private random: seedrandom.prng;\n\n  constructor(\n      mean: number, stdDeviation: number, dtype?: keyof RandNormalDataTypes,\n      truncated?: boolean, seed?: number) {\n    this.mean = mean;\n    this.stdDev = stdDeviation;\n    this.dtype = dtype;\n    this.nextVal = NaN;\n    this.truncated = truncated;\n    if (this.truncated) {\n      this.upper = this.mean + this.stdDev * 2;\n      this.lower = this.mean - this.stdDev * 2;\n    }\n    const seedValue = seed ? seed : Math.random();\n    this.random = seedrandom.alea(seedValue.toString());\n  }\n\n  /** Returns next sample from a Gaussian distribution. */\n  public nextValue(): number {\n    if (!isNaN(this.nextVal)) {\n      const value = this.nextVal;\n      this.nextVal = NaN;\n      return value;\n    }\n\n    let resultX: number, resultY: number;\n    let isValid = false;\n    while (!isValid) {\n      let v1: number, v2: number, s: number;\n      do {\n        v1 = 2 * this.random() - 1;\n        v2 = 2 * this.random() - 1;\n        s = v1 * v1 + v2 * v2;\n      } while (s >= 1 || s === 0);\n\n      const mul = Math.sqrt(-2.0 * Math.log(s) / s);\n      resultX = this.mean + this.stdDev * v1 * mul;\n      resultY = this.mean + this.stdDev * v2 * mul;\n\n      if (!this.truncated || this.isValidTruncated(resultX)) {\n        isValid = true;\n      }\n    }\n\n    if (!this.truncated || this.isValidTruncated(resultY)) {\n      this.nextVal = this.convertValue(resultY);\n    }\n    return this.convertValue(resultX);\n  }\n\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype == null || this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  /** Returns true if less than 2-standard-deviations from the mean. */\n  private isValidTruncated(value: number): boolean {\n    return value <= this.upper && value >= this.lower;\n  }\n}\n\n// Marsaglia, George, and Wai Wan Tsang. 2000. \"A Simple Method for Generating\n// Gamma Variables.\"\nexport class RandGamma implements RandomGamma {\n  private alpha: number;\n  private beta: number;\n  private d: number;\n  private c: number;\n  private dtype?: keyof RandGammaDataTypes;\n  private randu: seedrandom.prng;\n  private randn: MPRandGauss;\n\n  constructor(\n      alpha: number, beta: number, dtype: keyof RandGammaDataTypes,\n      seed?: number) {\n    this.alpha = alpha;\n    this.beta = 1 / beta;  // convert rate to scale parameter\n    this.dtype = dtype;\n\n    const seedValue = seed ? seed : Math.random();\n    this.randu = seedrandom.alea(seedValue.toString());\n    this.randn = new MPRandGauss(0, 1, dtype, false, this.randu());\n\n    if (alpha < 1) {\n      this.d = alpha + (2 / 3);\n    } else {\n      this.d = alpha - (1 / 3);\n    }\n    this.c = 1 / Math.sqrt(9 * this.d);\n  }\n\n  /** Returns next sample from a gamma distribution. */\n  public nextValue(): number {\n    let x2: number, v0: number, v1: number, x: number, u: number, v: number;\n    while (true) {\n      do {\n        x = this.randn.nextValue();\n        v = 1 + (this.c * x);\n      } while (v <= 0);\n      v *= v * v;\n      x2 = x * x;\n      v0 = 1 - (0.331 * x2 * x2);\n      v1 = (0.5 * x2) + (this.d * (1 - v + Math.log(v)));\n      u = this.randu();\n      if (u < v0 || Math.log(u) < v1) {\n        break;\n      }\n    }\n    v = (1 / this.beta) * this.d * v;\n    if (this.alpha < 1) {\n      v *= Math.pow(this.randu(), 1 / this.alpha);\n    }\n    return this.convertValue(v);\n  }\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n}\n\nexport class UniformRandom implements RandomBase {\n  private min: number;\n  private range: number;\n  private random: seedrandom.prng;\n  private dtype?: keyof RandNormalDataTypes;\n\n  constructor(\n      min = 0, max = 1, dtype?: keyof RandNormalDataTypes,\n      seed?: string|number) {\n    this.min = min;\n    this.range = max - min;\n    this.dtype = dtype;\n    if (seed == null) {\n      seed = Math.random();\n    }\n    if (typeof seed === 'number') {\n      seed = seed.toString();\n    }\n\n    if (!this.canReturnFloat() && this.range <= 1) {\n      throw new Error(\n          `The difference between ${min} - ${max} <= 1 and dtype is not float`);\n    }\n    this.random = seedrandom.alea(seed);\n  }\n\n  /** Handles proper rounding for non floating point numbers. */\n  private canReturnFloat = () =>\n      (this.dtype == null || this.dtype === 'float32');\n\n  private convertValue(value: number): number {\n    if (this.canReturnFloat()) {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  nextValue() {\n    return this.convertValue(this.min + this.range * this.random());\n  }\n}\n\nexport function jarqueBeraNormalityTest(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test\n  const n = values.length;\n  const s = skewness(values);\n  const k = kurtosis(values);\n  const jb = n / 6 * (Math.pow(s, 2) + 0.25 * Math.pow(k - 3, 2));\n  // JB test requires 2-degress of freedom from Chi-Square @ 0.95:\n  // http://www.itl.nist.gov/div898/handbook/eda/section3/eda3674.htm\n  const CHI_SQUARE_2DEG = 5.991;\n  if (jb > CHI_SQUARE_2DEG) {\n    throw new Error(`Invalid p-value for JB: ${jb}`);\n  }\n}\n\nexport function expectArrayInMeanStdRange(\n    actual: TypedArray|number[], expectedMean: number, expectedStdDev: number,\n    epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  const actualMean = mean(actual);\n  expectNumbersClose(actualMean, expectedMean, epsilon);\n  expectNumbersClose(\n      standardDeviation(actual, actualMean), expectedStdDev, epsilon);\n}\n\nfunction mean(values: TypedArray|number[]) {\n  let sum = 0;\n  for (let i = 0; i < values.length; i++) {\n    sum += values[i];\n  }\n  return sum / values.length;\n}\n\nfunction standardDeviation(values: TypedArray|number[], mean: number) {\n  let squareDiffSum = 0;\n  for (let i = 0; i < values.length; i++) {\n    const diff = values[i] - mean;\n    squareDiffSum += diff * diff;\n  }\n  return Math.sqrt(squareDiffSum / values.length);\n}\n\nfunction kurtosis(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Kurtosis\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum4 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum4 += Math.pow(v, 4);\n  }\n  return (1 / n) * sum4 / Math.pow((1 / n) * sum2, 2);\n}\n\nfunction skewness(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Skewness\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum3 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum3 += Math.pow(v, 3);\n  }\n  return (1 / n) * sum3 / Math.pow((1 / (n - 1)) * sum2, 3 / 2);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {RandGamma} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a gamma distribution.\n *\n * ```js\n * tf.randomGamma([2, 2], 1).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param alpha The shape parameter of the gamma distribution.\n * @param beta The inverse scale parameter of the gamma distribution. Defaults\n *     to 1.\n * @param dtype The data type of the output. Defaults to float32.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomGamma_<R extends Rank>(\n    shape: ShapeMap[R], alpha: number, beta = 1,\n    dtype: 'float32'|'int32' = 'float32', seed?: number): Tensor<R> {\n  if (beta == null) {\n    beta = 1;\n  }\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const rgamma = new RandGamma(alpha, beta, dtype, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = rgamma.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomGamma = op({randomGamma_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a normal distribution.\n *\n * ```js\n * tf.randomNormal([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, false /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomNormal = op({randomNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {UniformRandom} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a uniform distribution.\n *\n * The generated values follow a uniform distribution in the range [minval,\n * maxval). The lower bound minval is included in the range, while the upper\n * bound maxval is excluded.\n *\n * ```js\n * tf.randomUniform([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param minval The lower bound on the range of random values to generate.\n *   Defaults to 0.\n * @param maxval The upper bound on the range of random values to generate.\n *   Defaults to 1.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomUniform_<R extends Rank>(\n    shape: ShapeMap[R], minval = 0, maxval = 1, dtype: DataType = 'float32',\n    seed?: number|string): Tensor<R> {\n  const res = buffer(shape, dtype);\n  const random = new UniformRandom(minval, maxval, null, seed);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = random.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomUniform = op({randomUniform_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike1D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-1 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor1d` as it makes the code more readable.\n *\n * ```js\n * tf.tensor1d([1, 2, 3]).print();\n * ```\n *\n * @param values The values of the tensor. Can be array of numbers,\n *     or a `TypedArray`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor1d(values: TensorLike1D, dtype?: DataType): Tensor1D {\n  assertNonNull(values);\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 1) {\n    throw new Error('tensor1d() requires values to be a flat/TypedArray');\n  }\n  const shape: number[] = null;\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Range, RangeAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {makeZerosTypedArray} from '../util';\n\nimport {tensor1d} from './tensor1d';\nimport {zeros} from './zeros';\n\n/**\n * Creates a new `tf.Tensor1D` filled with the numbers in the range provided.\n *\n * The tensor is a is half-open interval meaning it includes start, but\n * excludes stop. Decrementing ranges and negative step values are also\n * supported.sv\n *\n *\n * ```js\n * tf.range(0, 9, 2).print();\n * ```\n *\n * @param start An integer start value\n * @param stop An integer stop value\n * @param step An integer increment (will default to 1 or -1)\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function range(\n    start: number, stop: number, step = 1,\n    dtype: 'float32'|'int32' = 'float32'): Tensor1D {\n  if (step === 0) {\n    throw new Error('Cannot have a step of zero');\n  }\n\n  const forward: ForwardFunc<Tensor> = () => {\n    const sameStartStop = start === stop;\n    const increasingRangeNegativeStep = start < stop && step < 0;\n    const decreasingRangePositiveStep = stop < start && step > 1;\n\n    if (sameStartStop || increasingRangeNegativeStep ||\n        decreasingRangePositiveStep) {\n      return zeros([0], dtype);\n    }\n\n    const numElements = Math.abs(Math.ceil((stop - start) / step));\n    const values = makeZerosTypedArray(numElements, dtype);\n\n    if (stop < start && step === 1) {\n      // Auto adjust the step's sign if it hasn't been set\n      // (or was set to 1)\n      step = -1;\n    }\n\n    values[0] = start;\n    for (let i = 1; i < values.length; i++) {\n      values[i] = values[i - 1] + step;\n    }\n\n    return tensor1d(values, dtype);\n  };\n\n  const attrs: RangeAttrs = {start, stop, step, dtype};\n\n  return ENGINE.runKernelFunc(\n             forward, {} /* inputs */, null /* grad */, Range,\n             attrs as {} as NamedAttrMap) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Reciprocal, ReciprocalInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of x element-wise: `1 / x`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, 2]);\n *\n * x.reciprocal().print();  // or tf.reciprocal(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction reciprocal_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'reciprocal');\n\n  const inputs: ReciprocalInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.reciprocal($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Reciprocal);\n}\nexport const reciprocal = op({reciprocal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu, ReluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport { cast } from './cast';\n\n/**\n * Computes rectified linear element-wise: `max(x, 0)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.relu().print();  // or tf.relu(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu($x);\n  };\n\n  const inputs: ReluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu) as\n      T;\n}\n\nexport const relu = op({relu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu6, Relu6Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes rectified linear 6 element-wise: `min(max(x, 0), 6)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 8]);\n *\n * x.relu6().print();  // or tf.relu6(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu6_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu6');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu6($x);\n  };\n\n  const inputs: Relu6Inputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu6) as\n      T;\n}\n\nexport const relu6 = op({relu6_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reverse, ReverseAttrs, ReverseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Reverses a `tf.Tensor` along a specified axis.\n *\n * Also available are stricter rank-specific methods that assert that `x` is\n * of the given rank:\n *   - `tf.reverse1d`\n *   - `tf.reverse2d`\n *   - `tf.reverse3d`\n *   - `tf.reverse4d`\n *\n * Except `tf.reverse1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.reverse().print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.reverse(axis).print();\n * ```\n * @param x The input tensor to be reversed.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction reverse_<T extends Tensor>(\n    x: T|TensorLike, axis?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'reverse');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const axes = parseAxisParam(axis, $x.shape);\n    if ($x.rank === 0) {\n      return clone($x);\n    }\n    const res = backend.reverse($x, axes);\n    return reshape(res, $x.shape);\n  };\n\n  const inputs: ReverseInputs = {x: $x};\n  const attrs: ReverseAttrs = {dims: axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Reverse, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const reverse = op({reverse_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor1D`.\n *\n * @param x The input tensor.\n */\nfunction reverse1d_(x: Tensor1D|TensorLike): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 1,\n      () => `Error in reverse1D: x must be rank 1 but got rank ${$x.rank}.`);\n  return reverse($x, 0);\n}\n\nexport const reverse1d = op({reverse1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor2D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse2d_(x: Tensor2D|TensorLike, axis?: number|number[]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 2,\n      () => `Error in reverse2D: x must be rank 2 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse2d = op({reverse2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor3D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse3d_(x: Tensor3D|TensorLike, axis?: number|number[]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 3,\n      () => `Error in reverse3D: x must be rank 3 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse3d = op({reverse3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor4D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse4d_(x: Tensor4D|TensorLike, axis?: number|number[]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 4,\n      () => `Error in reverse4D: x must be rank 4 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse4d = op({reverse4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Round, RoundInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes round of input `tf.Tensor` element-wise: `round(x)`.\n * It implements banker's rounding.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.round().print();  // or tf.round(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction round_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'round');\n  const inputs: RoundInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.round($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Round);\n}\n\nexport const round = op({round_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Rsqrt, RsqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of square root of the input `tf.Tensor` element-wise:\n * `y = 1 / sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.rsqrt().print();  // or tf.rsqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction rsqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'rsqrt');\n\n  const inputs: RsqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.rsqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Rsqrt);\n}\nexport const rsqrt = op({rsqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Selu, SeluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes scaled exponential linear element-wise.\n *\n * `x < 0 ? scale * alpha * (exp(x) - 1) : x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.selu().print();  // or tf.selu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction selu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'selu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.selu($x);\n    save([$x]);\n    return res;\n  };\n\n  const inputs: SeluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Selu) as\n      T;\n}\n\nexport const selu = op({selu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport {depthwiseConv2d} from './depthwise_conv2d';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * 2-D convolution with separable filters.\n *\n * Performs a depthwise convolution that acts separately on channels followed\n * by a pointwise convolution that mixes channels. Note that this is\n * separability between dimensions [1, 2] and 3, not spatial separability\n * between dimensions 1 and 2.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param depthwiseFilter The depthwise filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`. This is\n *     the filter used in the first step.\n * @param pointwiseFilter The pointwise filter tensor, rank 4, of shape\n *     `[1, 1, inChannels * channelMultiplier, outChannels]`. This is\n *     the filter used in the second step.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction separableConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthwiseFilter: Tensor4D|TensorLike,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'separableConv2d');\n  const $depthwiseFilter =\n      convertToTensor(depthwiseFilter, 'depthwiseFilter', 'separableConv2d');\n  const $pointwiseFilter =\n      convertToTensor(pointwiseFilter, 'pointwiseFilter', 'separableConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  if (dataFormat === 'NCHW') {\n    throw new Error(\n        'separableConv2d currently does not support dataFormat NCHW; only ' +\n        'NHWC is supported');\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in separableConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $depthwiseFilter.rank === 4,\n      () => `Error in separableConv2d: depthwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.rank === 4,\n      () => `Error in separableConv2d: pointwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.shape[0] === 1,\n      () =>\n          `Error in separableConv2d: the first dimension of pointwise filter ` +\n          ` must be 1, but got ${$pointwiseFilter.shape[0]}.`);\n  util.assert(\n      $pointwiseFilter.shape[1] === 1,\n      () => `Error in separableConv2d: the second dimension of pointwise ` +\n          `filter must be 1, but got ${$pointwiseFilter.shape[1]}.`);\n\n  const inChannels = $depthwiseFilter.shape[2];\n  const channelMultiplier = $depthwiseFilter.shape[3];\n  util.assert(\n      $pointwiseFilter.shape[2] === inChannels * channelMultiplier,\n      () =>\n          `Error in separableConv2d: the third dimension of pointwise filter ` +\n          `must be ${inChannels * channelMultiplier}, ` +\n          `but got ${$pointwiseFilter.shape[2]}.`);\n\n  const depthwise = depthwiseConv2d(\n      x4D, $depthwiseFilter, strides, pad, dataFormat, dilation);\n  const pointwiseStride = 1;\n  const res =\n      conv2d(depthwise, $pointwiseFilter, pointwiseStride, 'valid', dataFormat);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const separableConv2d = op({separableConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\n/**\n * Computes the difference between two lists of numbers.\n *\n * Given a Tensor `x` and a Tensor `y`, this operation returns a Tensor `out`\n * that represents all values that are in `x` but not in `y`. The returned\n * Tensor `out` is sorted in the same order that the numbers appear in `x`\n * (duplicates are preserved). This operation also returns a Tensor indices that\n * represents the position of each out element in `x`. In other words:\n *\n * `out[i] = x[idx[i]] for i in [0, 1, ..., out.length - 1]`\n *\n * ```js\n * const x = [1, 2, 3, 4, 5, 6];\n * const y = [1, 3, 5];\n *\n * const [out, indices] = await tf.setdiff1dAsync(x, y);\n * out.print(); // [2, 4, 6]\n * indices.print(); // [1, 3, 5]\n * ```\n *\n * @param x 1-D Tensor. Values to keep.\n * @param y 1-D Tensor. Must have the same type as x. Values to exclude in the\n *     output.\n * @returns Promise of Tensor tuple [out, indices].\n *  out: Tensor with the same type as x.\n *  indices: A Tensor of type int32.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nasync function setdiff1dAsync_(\n    x: Tensor|TensorLike, y: Tensor|TensorLike): Promise<[Tensor, Tensor]> {\n  const $x = convertToTensor(x, 'x', 'setdiff1d');\n  const $y = convertToTensor(y, 'y', 'setdiff1d');\n\n  util.assert(\n      $x.dtype === $y.dtype,\n      () => `x and y should have the same dtype, but got x (${\n          $x.dtype}) and y (${$y.dtype}).`);\n\n  util.assert(\n      $x.rank === 1, () => `x should be 1D tensor, but got x (${$x.shape}).`);\n\n  util.assert(\n      $y.rank === 1, () => `y should be 1D tensor, but got y (${$y.shape}).`);\n\n  const xVals = await $x.data();\n  const yVals = await $y.data();\n  const ySet = new Set(yVals);\n\n  let outputSize = 0;\n  for (let i = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      outputSize++;\n    }\n  }\n\n  const buffer = new TensorBuffer([outputSize], $x.dtype);\n  const indices = new TensorBuffer([outputSize], 'int32');\n  for (let i = 0, p = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      buffer.values[p] = xVals[i];\n      indices.values[p] = i;\n      p++;\n    }\n  }\n  return [buffer.toTensor(), indices.toTensor()];\n}\nexport const setdiff1dAsync = setdiff1dAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sign, SignInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns an element-wise indication of the sign of a number.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3, NaN, 0]);\n *\n * x.sign().print();  // or tf.sign(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sign_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sign');\n  const inputs: SignInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.sign($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Sign);\n}\nexport const sign = op({sign_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sin, SinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sin of the input Tensor element-wise: `sin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.sin().print();  // or tf.sin(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sin');\n\n  const inputs: SinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sin);\n}\nexport const sin = op({sin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sinh, SinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic sin of the input `tf.Tensor` element-wise: `sinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.sinh().print();  // or tf.sinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sinh');\n  const inputs: SinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sinh);\n}\nexport const sinh = op({sinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 1D slice from 1D array starting at coordinates `begin` and is\n * of length `size`. See `slice` for details.\n */\nfunction slice1d_(\n    x: Tensor1D|TensorLike, begin: number, size: number): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'slice1d');\n  util.assert(\n      $x.rank === 1,\n      () =>\n          `slice1d expects a rank-1 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, [begin], [size]);\n}\nexport const slice1d = op({slice1d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 2D slice from a 2D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice2d_(\n    x: Tensor2D|TensorLike, begin: [number, number],\n    size: [number, number]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'slice2d');\n  util.assert(\n      $x.rank === 2,\n      () =>\n          `slice2d expects a rank-2 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice2d = op({slice2d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 3D slice from a 3D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice3d_(\n    x: Tensor3D|TensorLike, begin: [number, number, number],\n    size: [number, number, number]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'slice3d');\n  util.assert(\n      $x.rank === 3,\n      () =>\n          `slice3d expects a rank-3 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice3d = op({slice3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 4D slice from a 4D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice4d_(\n    x: Tensor4D|TensorLike, begin: [number, number, number, number],\n    size: [number, number, number, number]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'slice4d');\n  util.assert(\n      $x.rank === 4,\n      () =>\n          `slice4d expects a rank-4 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice4d = op({slice4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softmax, SoftmaxAttrs, SoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the softmax normalized vector given the logits.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction softmax_<T extends Tensor>(logits: T|TensorLike, dim = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'softmax', 'float32');\n\n  if (dim === -1) {\n    dim = $logits.rank - 1;\n  }\n  if (dim !== $logits.rank - 1) {\n    throw Error(\n        'Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and dim was ${dim}`);\n  }\n\n  const inputs: SoftmaxInputs = {logits: $logits};\n  const attrs: SoftmaxAttrs = {dim};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const y = backend.softmax($logits, dim);\n        save([y]);\n        return y;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, Softmax,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const softmax = op({softmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FFT, FFTInputs} from '../../kernel_names';\nimport {Tensor} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\n\n/**\n * Fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the inner-most\n * dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.fft().print();  // tf.spectral.fft(x).print();\n * ```\n * @param input The complex input to compute an fft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction fft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.fft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: FFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D = input.as2D(batch, innerDimensionSize);\n    const result = backend.fft(input2D);\n    return result.reshape(input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, FFT);\n}\n\nexport const fft = op({fft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {IFFT, IFFTInputs} from '../../kernel_names';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Inverse fast Fourier transform.\n *\n * Computes the inverse 1-dimensional discrete Fourier transform over the\n * inner-most dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.ifft().print();  // tf.spectral.ifft(x).print();\n * ```\n * @param input The complex input to compute an ifft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction ifft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.ifft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: IFFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D: Tensor2D = reshape(input, [batch, innerDimensionSize]);\n    const result = backend.ifft(input2D);\n    return reshape(result, input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, IFFT);\n}\n\nexport const ifft = op({ifft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {reverse} from '../reverse';\nimport {scalar} from '../scalar';\nimport {slice} from '../slice';\n\nimport {ifft} from './ifft';\n\n/**\n * Inversed real value input fast Fourier transform.\n *\n * Computes the 1-dimensional inversed discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([0, 0, 0]);\n * const x = tf.complex(real, imag);\n *\n * x.irfft().print();\n * ```\n * @param input The real value input to compute an irfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction irfft_(input: Tensor): Tensor {\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n  let ret: Tensor;\n  if (innerDimensionSize <= 2) {\n    const complexInput = reshape(input, [batch, innerDimensionSize]);\n    ret = ifft(complexInput);\n  } else {\n    // The length of unique components of the DFT of a real-valued signal\n    // is 2 * (input_len - 1)\n    const outputShape = [batch, 2 * (innerDimensionSize - 1)];\n    const realInput = reshape(real(input), [batch, innerDimensionSize]);\n    const imagInput = reshape(imag(input), [batch, innerDimensionSize]);\n\n    const realConjugate =\n        reverse(slice(realInput, [0, 1], [batch, innerDimensionSize - 2]), 1);\n    const imagConjugate: Tensor2D = mul(\n        reverse(slice(imagInput, [0, 1], [batch, innerDimensionSize - 2]), 1),\n        scalar(-1));\n\n    const r = concat([realInput, realConjugate], 1);\n    const i = concat([imagInput, imagConjugate], 1);\n    const complexInput =\n        reshape(complex(r, i), [outputShape[0], outputShape[1]]);\n    ret = ifft(complexInput);\n  }\n  ret = real(ret);\n  // reshape the result if the input is 3D tensor.\n  if (input.rank === 3 && input.shape[0] !== 0) {\n    const temp = ret;\n    const batch = input.shape[0];\n    ret = reshape(ret, [batch, ret.shape[0] / batch, ret.shape[1]]);\n    temp.dispose();\n  }\n  return ret;\n}\n\nexport const irfft = op({irfft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {assert} from '../util';\n\n/**\n * Prepare the split size array. When the input is a number, the axis is evenly\n * divided among the split size. When the input contains the negative value, the\n * rest of the axis is allocated toward that.\n */\nexport function prepareSplitSize(\n    x: Tensor|TensorInfo, numOrSizeSplits: number[]|number,\n    axis = 0): number[] {\n  let splitSizes = [];\n  if (typeof (numOrSizeSplits) === 'number') {\n    assert(\n        x.shape[axis] % numOrSizeSplits === 0,\n        () => 'Number of splits must evenly divide the axis.');\n    splitSizes =\n        new Array(numOrSizeSplits).fill(x.shape[axis] / numOrSizeSplits);\n  } else {\n    const numOfNegs = numOrSizeSplits.reduce((count, value) => {\n      if (value === -1) {\n        count += 1;\n      }\n      return count;\n    }, 0);\n    assert(\n        numOfNegs <= 1,\n        () => 'There should be only one negative value in split array.');\n    const negIndex = numOrSizeSplits.indexOf(-1);\n    // Allow the number of split array to be -1, which indicates the rest\n    // of dimension is allocated to that split.\n    if (negIndex !== -1) {\n      const total = numOrSizeSplits.reduce((a, b) => b > 0 ? a + b : a);\n      numOrSizeSplits[negIndex] = x.shape[axis] - total;\n    }\n    assert(\n        x.shape[axis] === numOrSizeSplits.reduce((a, b) => a + b),\n        () => 'The sum of sizes must match the size of the axis dimension.');\n    splitSizes = numOrSizeSplits;\n  }\n\n  return splitSizes;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SplitV, SplitVAttrs, SplitVInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {prepareSplitSize} from './split_util';\n\n/**\n * Splits a `tf.Tensor` into sub tensors.\n *\n * If `numOrSizeSplits` is a number, splits `x` along dimension `axis`\n * into `numOrSizeSplits` smaller tensors.\n * Requires that `numOrSizeSplits` evenly divides `x.shape[axis]`.\n *\n * If `numOrSizeSplits` is a number array, splits `x` into\n * `numOrSizeSplits.length` pieces. The shape of the `i`-th piece has the\n * same size as `x` except along dimension `axis` where the size is\n * `numOrSizeSplits[i]`.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4, 5, 6, 7, 8], [2, 4]);\n * const [a, b] = tf.split(x, 2, 1);\n * a.print();\n * b.print();\n *\n * const [c, d, e] = tf.split(x, [1, 2, 1], 1);\n * c.print();\n * d.print();\n * e.print();\n * ```\n *\n * @param x The input tensor to split.\n * @param numOrSizeSplits Either an integer indicating the number of\n * splits along the axis or an array of integers containing the sizes of\n * each output tensor along the axis. If a number then it must evenly divide\n * `x.shape[axis]`; otherwise the sum of sizes must match `x.shape[axis]`.\n * Can contain one -1 indicating that dimension is to be inferred.\n * @param axis The dimension along which to split. Defaults to 0 (the first\n * dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction split_<T extends Tensor>(\n    x: Tensor|TensorLike, numOrSizeSplits: number[]|number, axis = 0): T[] {\n  const $x = convertToTensor(x, 'x', 'split');\n\n  const forward: ForwardFunc<Tensor> = (backend, _) => {\n    const $axis = parseAxisParam(axis, $x.shape)[0];\n    const splitSizes = prepareSplitSize($x, numOrSizeSplits, $axis);\n    return backend.split($x, splitSizes, $axis) as {} as T;\n  };\n\n  const inputs: SplitVInputs = {x: $x};\n  const attr: SplitVAttrs = {numOrSizeSplits, axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, SplitV,\n             attr as {} as NamedAttrMap) as {} as T[];\n}\n\nexport const split = op({split_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {assert} from '../../util';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {split} from '../split';\nimport {zeros} from '../zeros';\nimport {zerosLike} from '../zeros_like';\n\nimport {fft} from './fft';\n\n/**\n * Real value input fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n *\n * real.rfft().print();\n * ```\n * @param input The real value input to compute an rfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction rfft_(input: Tensor, fftLength?: number): Tensor {\n  assert(\n      input.dtype === 'float32',\n      () => `The dtype for rfft() must be real value but got ${input.dtype}`);\n\n  let innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n\n  let adjustedInput: Tensor;\n  if (fftLength != null && fftLength < innerDimensionSize) {\n    // Need to crop\n    const begin = input.shape.map(v => 0);\n    const size = input.shape.map(v => v);\n    size[input.shape.length - 1] = fftLength;\n    adjustedInput = slice(input, begin, size);\n    innerDimensionSize = fftLength;\n  } else if (fftLength != null && fftLength > innerDimensionSize) {\n    // Need to pad with zeros\n    const zerosShape = input.shape.map(v => v);\n    zerosShape[input.shape.length - 1] = fftLength - innerDimensionSize;\n    adjustedInput = concat([input, zeros(zerosShape)], input.shape.length - 1);\n    innerDimensionSize = fftLength;\n  } else {\n    adjustedInput = input;\n  }\n\n  // Complement the input with zero imaginary numbers.\n  const zerosInput = zerosLike(adjustedInput);\n  const complexInput =\n      reshape(complex(adjustedInput, zerosInput), [batch, innerDimensionSize]);\n\n  const ret = fft(complexInput);\n\n  // Exclude complex conjugations. These conjugations are put symmetrically.\n  const half = Math.floor(innerDimensionSize / 2) + 1;\n  const realValues = real(ret);\n  const imagValues = imag(ret);\n  const realComplexConjugate = split(\n      realValues, [half, innerDimensionSize - half],\n      realValues.shape.length - 1);\n  const imagComplexConjugate = split(\n      imagValues, [half, innerDimensionSize - half],\n      imagValues.shape.length - 1);\n\n  const outputShape = adjustedInput.shape.slice();\n  outputShape[adjustedInput.shape.length - 1] = half;\n\n  return reshape(\n      complex(realComplexConjugate[0], imagComplexConjugate[0]), outputShape);\n}\n\nexport const rfft = op({rfft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sqrt, SqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes square root of the input `tf.Tensor` element-wise: `y = sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.sqrt().print();  // or tf.sqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sqrt');\n\n  const inputs: SqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sqrt);\n}\nexport const sqrt = op({sqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SquaredDifference, SquaredDifferenceInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns (a - b) * (a - b) element-wise.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * ```js\n * // Broadcast squared difference  a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction squaredDifference_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'squaredDifference');\n  let $b = convertToTensor(b, 'b', 'squaredDifference');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.squaredDifference($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SquaredDifferenceInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             SquaredDifference, attrs) as T;\n}\n\nexport const squaredDifference = op({squaredDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {squeezeShape} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Removes dimensions of size 1 from the shape of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4], [1, 1, 4]);\n * x.squeeze().print();\n * ```\n *\n * @param x The input tensor to be squeezed.\n * @param axis An optional list of numbers. If specified, only\n *     squeezes the dimensions listed. The dimension index starts at 0. It\n * is an error to squeeze a dimension that is not 1.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction squeeze_<T extends Tensor>(x: Tensor|TensorLike, axis?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'squeeze');\n  return reshape($x, squeezeShape($x.shape, axis).newShape) as T;\n}\n\nexport const squeeze = op({squeeze_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {concat} from './concat';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\n\n/**\n * Stacks a list of rank-`R` `tf.Tensor`s into one rank-`(R+1)` `tf.Tensor`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.stack([a, b, c]).print();\n * ```\n *\n * @param tensors A list of tensor objects with the same shape and dtype.\n * @param axis The axis to stack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction stack_<T extends Tensor>(\n    tensors: Array<T|TensorLike>, axis = 0): Tensor {\n  const $tensors = convertToTensorArray(tensors, 'tensors', 'stack');\n\n  util.assert(\n      $tensors.length >= 1, () => 'Pass at least one tensor to tf.stack');\n\n  if ($tensors.length === 1) {\n    return expandDims($tensors[0], axis);\n  }\n\n  const rank = $tensors[0].rank;\n  const shape = $tensors[0].shape;\n  const dtype = $tensors[0].dtype;\n\n  util.assert(axis <= rank, () => 'Axis must be <= rank of the tensor');\n\n  $tensors.forEach(t => {\n    util.assertShapesMatch(\n        shape, t.shape,\n        'All tensors passed to stack must have matching shapes');\n    util.assert(\n        dtype === t.dtype,\n        () => 'All tensors passed to stack must have matching dtypes');\n  });\n\n  const expandedTensors = $tensors.map(t => expandDims(t, axis));\n  // Stack exists in the TensorFlow C++ API\n  // (https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/stack) but not\n  // in\n  // https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/core/ops/ops.pbtxt.\n  // Therefore we are treating it like a high-level op rather than\n  // creating a dedicated stack kernel.\n  return concat(expandedTensors, axis);\n}\n\nexport const stack = op({stack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Step, StepAttrs, StepInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes step of the input `tf.Tensor` element-wise: `x > 0 ? 1 : alpha * x`\n *\n * ```js\n * const x = tf.tensor1d([0, 2, -1, -3]);\n *\n * x.step(.5).print();  // or tf.step(x, .5)\n * ```\n * @param x The input tensor.\n * @param alpha The gradient when input is negative.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction step_<T extends Tensor>(x: T|TensorLike, alpha = 0.0): T {\n  const $x = convertToTensor(x, 'x', 'step');\n\n  const inputs: StepInputs = {x: $x};\n  const attrs: StepAttrs = {alpha};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.step($x, alpha), inputs as {} as NamedTensorMap,\n      null /* grad */, Step, attrs as {} as NamedAttrMap);\n}\nexport const step = op({step_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {StridedSlice, StridedSliceAttrs, StridedSliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {slice} from './slice';\nimport {computeOutShape, getNormalizedAxes, maskToAxes} from './slice_util';\n\n/**\n * Extracts a strided slice of a tensor.\n *\n * Roughly speaking, this op extracts a slice of size (end-begin)/stride from\n * the given input tensor (x). Starting at the location specified by begin the\n * slice continues by adding stride to the index until all dimensions are not\n * less than end. Note that a stride can be negative, which causes a reverse\n * slice.\n *\n * ```js\n * const t = tf.tensor3d([1, 1, 1 ,2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6],\n *    [3, 2, 3]);\n * t.stridedSlice([1, 0, 0], [2, 1, 3], [1, 1, 1]).print()  // [[[3, 3, 3]]]\n * t.stridedSlice([1, 0, 0], [2, 2, 3], [1, 1, 1]).print()  // [[[3, 3, 3],\n *                                                     // [4, 4, 4]]]\n * t.stridedSlice([1, -1, 0], [2, -3, 3], [1, -1, 1]).print() // [[[4, 4, 4],\n *                                                     // [3, 3, 3]]]\n * ```\n *\n * @param x The tensor to stride slice.\n * @param begin The coordinates to start the slice from.\n * @param end: The coordinates to end the slice at.\n * @param strides: The size of the slice.\n * @param beginMask: If the ith bit of beginMask is set, begin[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param endMask: If the ith bit of endMask is set, end[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param shrinkAxisMask: a bitmask where bit i implies that\n * the ith specification should shrink the dimensionality. begin and end must\n * imply a slice of size 1 in the dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction stridedSlice_(\n    x: Tensor|TensorLike, begin: number[], end: number[], strides?: number[],\n    beginMask = 0, endMask = 0, ellipsisMask = 0, newAxisMask = 0,\n    shrinkAxisMask = 0): Tensor {\n  let $x = convertToTensor(x, 'x', 'stridedSlice');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if (strides == null) {\n      strides = new Array(begin.length);\n    }\n\n    const ellipsisAxes = maskToAxes(ellipsisMask);\n    if (ellipsisAxes.length > 1) {\n      throw new Error('Multiple ellipses in slice is not allowed.');\n    }\n\n    if (ellipsisMask !== 0 && newAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and newAxisMask is not yet supported.');\n    }\n\n    if (ellipsisMask !== 0 && shrinkAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and shrinkAxisMask is not yet supported.');\n    }\n\n    const numInterpolatedAxes = $x.rank - begin.length;\n\n    // Expand the dims of x based on the newAxisMask.\n    const expandAxes = maskToAxes(newAxisMask);\n    const newShape = $x.shape.slice();\n    expandAxes.forEach(axis => {\n      begin[axis] = 0;\n      end[axis] = 1;\n      newShape.splice(axis, 0, 1);\n    });\n    $x = reshape($x, newShape);\n\n    const {\n      begin: normalizedBegin,\n      end: normalizedEnd,\n      strides: normalizedStrides\n    } =\n        getNormalizedAxes(\n            $x.shape, ellipsisAxes, numInterpolatedAxes, begin, end, strides,\n            beginMask, endMask, ellipsisMask);\n    begin = normalizedBegin;\n    end = normalizedEnd;\n    strides = normalizedStrides;\n\n    const shrinkAxes = maskToAxes(shrinkAxisMask);\n    // Adjust the ends based on the shrink mask.\n    shrinkAxes.forEach(axis => {\n      end[axis] = begin[axis] + 1;\n      strides[axis] = 1;\n    });\n\n    // Figure out the output shape.\n    const size = computeOutShape(begin, end, strides);\n    // Remove the axes based on shrinkMask.\n    const outShape = size.filter((_, axis) => shrinkAxes.indexOf(axis) === -1);\n\n    const nonStrided = strides.every(v => v === 1);\n    if (nonStrided) {\n      return reshape(slice($x, begin, size), outShape);\n    }\n\n    const res = backend.stridedSlice($x, begin, end, strides);\n    return reshape(res, outShape);\n  };\n\n  const inputs: StridedSliceInputs = {x: $x};\n  const attrs: StridedSliceAttrs = {\n    begin,\n    end,\n    strides,\n    beginMask,\n    endMask,\n    ellipsisMask,\n    newAxisMask,\n    shrinkAxisMask\n  };\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, StridedSlice,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const stridedSlice = op({stridedSlice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tan, TanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes tan of the input `tf.Tensor` element-wise, `tan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.tan().print();  // or tf.tan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tan');\n\n  const inputs: TanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.tan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tan);\n}\nexport const tan = op({tan_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike2D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-2 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor2d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor2d([[1, 2], [3, 4]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor2d(\n    values: TensorLike2D, shape?: [number, number],\n    dtype?: DataType): Tensor2D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 2) {\n    throw new Error('tensor2d() requires shape to have two numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 2 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor2d() requires values to be number[][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor2d() requires shape to be provided when `values` ' +\n        'are a flat/TypedArray');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-4 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor4d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor4d([[[[1], [2]], [[3], [4]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor4d(\n    values: TensorLike4D, shape?: [number, number, number, number],\n    dtype?: DataType): Tensor4D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 4) {\n    throw new Error('tensor4d() requires shape to have four numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 4 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor4d() requires values to be number[][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor4d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor4D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor5D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike5D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-5 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor5d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor5d([[[[[1], [2]], [[3], [4]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor5d(\n    values: TensorLike5D, shape?: [number, number, number, number, number],\n    dtype?: DataType): Tensor5D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 5) {\n    throw new Error('tensor5d() requires shape to have five numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 5 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor5d() requires values to be ' +\n        'number[][][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor5d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor5D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor6D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike6D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-6 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor6d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor6d([[[[[[1],[2]],[[3],[4]]],[[[5],[6]],[[7],[8]]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor6d([1, 2, 3, 4, 5, 6, 7, 8], [1, 1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor6d(\n    values: TensorLike6D,\n    shape?: [number, number, number, number, number, number],\n    dtype?: DataType): Tensor6D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 6) {\n    throw new Error('tensor6d() requires shape to have six numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 6 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor6d() requires values to be number[][][][][][] or ' +\n        'flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor6d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  shape = shape ||\n      inferredShape as [number, number, number, number, number, number];\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor6D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {TopK, TopKAttrs, TopKInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {NumericTensor, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Finds the values and indices of the `k` largest entries along the last\n * dimension.\n *\n * If the input is a vector (rank=1), finds the k largest entries in the vector\n * and outputs their values and indices as vectors. Thus values[j] is the j-th\n * largest entry in input, and its index is indices[j].\n * For higher rank inputs, computes the top k entries along the last dimension.\n *\n * If two elements are equal, the lower-index element appears first.\n *\n * ```js\n * const a = tf.tensor2d([[1, 5], [4, 3]]);\n * const {values, indices} = tf.topk(a);\n * values.print();\n * indices.print();\n * ```\n * @param x 1-D or higher `tf.Tensor` with last dimension being at least `k`.\n * @param k Number of top elements to look for along the last dimension.\n * @param sorted If true, the resulting `k` elements will be sorted by the\n *     values in descending order.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction topk_<T extends Tensor>(\n    x: T|TensorLike, k = 1, sorted = true): {values: T, indices: T} {\n  const $x = convertToTensor(x, 'x', 'topk');\n  if ($x.rank === 0) {\n    throw new Error('topk() expects the input to be of rank 1 or higher');\n  }\n  const lastDim = $x.shape[$x.shape.length - 1];\n  if (k > lastDim) {\n    throw new Error(\n        `'k' passed to topk() must be <= the last dimension (${lastDim}) ` +\n        `but got ${k}`);\n  }\n\n  const inputs: TopKInputs = {x: $x};\n  const attrs: TopKAttrs = {k, sorted};\n\n  const [values, indices] = ENGINE.runKernelFunc(\n      b => b.topk($x as NumericTensor, k, sorted),\n      inputs as {} as NamedTensorMap, null /* grad */, TopK,\n      attrs as {} as NamedAttrMap);\n\n  return {values, indices} as {values: T, indices: T};\n}\n\nexport const topk = op({topk_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a truncated normal\n * distribution.\n *\n * ```js\n * tf.truncatedNormal([2, 2]).print();\n * ```\n *\n * The generated values follow a normal distribution with specified mean and\n * standard deviation, except that values whose magnitude is more than 2\n * standard deviations from the mean are dropped and re-picked.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output tensor.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction truncatedNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type $ { dtype }`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, true /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const truncatedNormal = op({truncatedNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Unique, UniqueAttrs, UniqueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Finds unique elements along an axis of a tensor.\n *\n * It returns a tensor `values` containing all of the unique elements along the\n * `axis` of the given tensor `x` in the same order that they occur along the\n * `axis` in `x`; `x` does not need to be sorted. It also returns a tensor\n * `indices` the same size as the number of the elements in `x` along the `axis`\n * dimension. It contains the index in the unique output `values`.\n *\n * ```js\n * // A 1-D tensor\n * const a = tf.tensor1d([1, 1, 2, 4, 4, 4, 7, 8, 8]);\n * const {values, indices} = tf.unique(a);\n * values.print();   // [1, 2, 4, 7, 8,]\n * indices.print();  // [0, 0, 1, 2, 2, 2, 3, 4, 4]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=0\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 0)\n * values.print();   // [[1, 0, 0],\n *                   //  [2, 0, 0]]\n * indices.print();  // [0, 0, 1]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=1\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 1)\n * values.print();   // [[1, 0],\n *                   //  [1, 0],\n *                   //  [2, 0]]\n * indices.print();  // [0, 1, 1]\n * ```\n * @param x A tensor (int32, string, bool).\n * @param axis The axis of the tensor to find the unique elements.\n * @returns [uniqueElements, indices] (see above for details)\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction unique_<T extends Tensor>(\n    x: T|TensorLike, axis = 0): {values: T, indices: Tensor1D} {\n  // x can be of any dtype, thus null as the last argument.\n  const $x = convertToTensor(x, 'x', 'unique', null);\n  assert($x.rank > 0, () => 'The input tensor must be at least 1D');\n\n  const inputs: UniqueInputs = {x: $x};\n  const attrs: UniqueAttrs = {axis};\n  const [values, indices] = ENGINE.runKernel(\n                                Unique, inputs as {} as NamedTensorMap,\n                                attrs as {} as NamedAttrMap) as [T, Tensor1D];\n  return {values, indices};\n}\n\nexport const unique = op({unique_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {UnsortedSegmentSum, UnsortedSegmentSumAttrs, UnsortedSegmentSumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, isInt} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Computes the sum along segments of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const segmentIds = tf.tensor1d([1, 2, 0, 1], 'int32');\n * const numSegments = 3;\n *\n * x.unsortedSegmentSum(segmentIds, numSegments).print()\n * //or tf.unsortedSegmentSum(x, segmentIds, numSegments)\n * ```\n * @param x The `tf.Tensor` that will be summed along its segments.\n * @param segmentIds A `tf.Tensor1D` whose rank is equal to the rank of `x`'s\n * dimension along the `axis`.  Maps each element of `x` to a segment.\n * @param numSegments The number of distinct `segmentIds`.\n *\n * @doc {heading: 'Operations', subheading: 'Segment'}\n */\nfunction unsortedSegmentSum_<T extends Tensor>(\n    x: T|TensorLike, segmentIds: Tensor1D|TensorLike, numSegments: number): T {\n  const $x = convertToTensor(x, 'x', 'unsortedSegmentSum');\n  const $segmentIds =\n      convertToTensor(segmentIds, 'segmentIds', 'unsortedSegmentSum', 'int32');\n  assert(isInt(numSegments), () => 'numSegments must be of dtype int');\n\n  const inputs: UnsortedSegmentSumInputs = {x: $x, segmentIds: $segmentIds};\n  const attrs: UnsortedSegmentSumAttrs = {numSegments};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.unsortedSegmentSum($x, $segmentIds, numSegments);\n    save([$segmentIds]);\n    return res;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             UnsortedSegmentSum, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const unsortedSegmentSum = op({unsortedSegmentSum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Unpack, UnpackAttrs, UnpackInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Unstacks a `tf.Tensor` of rank-`R` into a list of rank-`(R-1)` `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * tf.unstack(a).forEach(tensor => tensor.print());\n * ```\n *\n * @param x A tensor object.\n * @param axis The axis to unstack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction unstack_(x: Tensor|TensorLike, axis = 0): Tensor[] {\n  const $x = convertToTensor(x, 'x', 'unstack');\n  util.assert(\n      axis >= -$x.shape.length && axis < $x.shape.length,\n      () =>\n          `Axis = ${axis} is not in [-${$x.shape.length}, ${$x.shape.length})`);\n  if (axis < 0) {\n    axis += $x.shape.length;\n  }\n  const inputs: UnpackInputs = {value: $x};\n  const attrs: UnpackAttrs = {axis};\n  const forward: ForwardFunc<Tensor[]> = (backend: KernelBackend) =>\n      backend.unstack($x, axis);\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Unpack,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const unstack = op({unstack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor, Variable} from '../tensor';\nimport {DataType, Rank} from '../types';\n\n/**\n * Creates a new variable with the provided initial value.\n * ```js\n * const x = tf.variable(tf.tensor([1, 2, 3]));\n * x.assign(tf.tensor([4, 5, 6]));\n *\n * x.print();\n * ```\n *\n * @param initialValue Initial value for the tensor.\n * @param trainable If true, optimizers are allowed to update it.\n * @param name Name of the variable. Defaults to a unique id.\n * @param dtype If set, initialValue will be converted to the given type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function variable<R extends Rank>(\n    initialValue: Tensor<R>, trainable = true, name?: string,\n    dtype?: DataType): Variable<R> {\n  return ENGINE.makeVariable(initialValue, trainable, name, dtype) as\n      Variable<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the Where kernel shared between cpu and webgl */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor2D} from '../tensor';\nimport {TypedArray} from '../types';\n\nexport function whereImpl(condShape: number[], condVals: TypedArray): Tensor2D {\n  const indices = [];\n  for (let i = 0; i < condVals.length; i++) {\n    if (condVals[i]) {\n      indices.push(i);\n    }\n  }\n\n  const inBuffer = buffer(condShape, 'int32');\n\n  const out = buffer([indices.length, condShape.length], 'int32');\n  for (let i = 0; i < indices.length; i++) {\n    const loc = inBuffer.indexToLoc(indices[i]);\n    const offset = i * condShape.length;\n    out.values.set(loc, offset);\n  }\n  return out.toTensor() as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {whereImpl} from '../backends/where_impl';\nimport {Tensor, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\n/**\n * Returns the coordinates of true elements of condition.\n *\n * The coordinates are returned in a 2-D tensor where the first dimension (rows)\n * represents the number of true elements, and the second dimension (columns)\n * represents the coordinates of the true elements. Keep in mind, the shape of\n * the output tensor can vary depending on how many true values there are in\n * input. Indices are output in row-major order. The resulting tensor has the\n * shape `[numTrueElems, condition.rank]`.\n *\n * This is analogous to calling the python `tf.where(cond)` without an x or y.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const result = await tf.whereAsync(cond);\n * result.print();\n * ```\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nasync function whereAsync_(condition: Tensor|TensorLike): Promise<Tensor2D> {\n  const $condition =\n      convertToTensor(condition, 'condition', 'whereAsync', 'bool');\n  const vals = await $condition.data();\n  const res = whereImpl($condition.shape, vals);\n  if (condition !== $condition) {\n    $condition.dispose();\n  }\n  return res;\n}\n\nexport const whereAsync = whereAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {gather} from './gather';\nimport {reshape} from './reshape';\nimport {squeeze} from './squeeze';\nimport {whereAsync} from './where_async';\n\n/**\n * Apply boolean mask to tensor.\n *\n * ```js\n * const tensor = tf.tensor2d([1, 2, 3, 4, 5, 6], [3, 2]);\n * const mask = tf.tensor1d([1, 0, 1], 'bool');\n * const result = await tf.booleanMaskAsync(tensor, mask);\n * result.print();\n * ```\n *\n * @param tensor N-D tensor.\n * @param mask K-D boolean tensor, K <= N and K must be known statically.\n * @param axis A 0-D int Tensor representing the axis in tensor to mask from.\n *     By default, axis is 0 which will mask from the first dimension.\n *     Otherwise K + axis <= N.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nasync function booleanMaskAsync_(\n    tensor: Tensor|TensorLike, mask: Tensor|TensorLike,\n    axis?: number): Promise<Tensor> {\n  const $tensor = convertToTensor(tensor, 'tensor', 'boolMask');\n  const $mask = convertToTensor(mask, 'mask', 'boolMask', 'bool');\n\n  const axisFrom = axis == null ? 0 : axis;\n  const maskDim = $mask.rank;\n  const tensorShape = $tensor.shape;\n\n  util.assert(maskDim > 0, () => 'mask cannot be scalar');\n  util.assertShapesMatch(\n      tensorShape.slice(axisFrom, axisFrom + maskDim), $mask.shape,\n      `mask's shape must match the first K dimensions of tensor's shape,`);\n\n  let leadingSize = 1;\n  for (let i = axisFrom; i < axisFrom + maskDim; i++) {\n    leadingSize *= tensorShape[i];\n  }\n  const targetTensorShape =\n      tensorShape.slice(0, axisFrom)\n          .concat([leadingSize], tensorShape.slice(axisFrom + maskDim));\n  const reshapedTensor = reshape($tensor, targetTensorShape);\n  const reshapedMask = reshape($mask, [-1]);\n  const positivePositions = await whereAsync(reshapedMask);\n  const indices = squeeze(positivePositions, [1]);\n\n  const res = gather(reshapedTensor, indices, axisFrom);\n\n  // Ensure no memory leak.\n  if (tensor !== $tensor) {\n    $tensor.dispose();\n  }\n  if (mask !== $mask) {\n    $mask.dispose();\n  }\n  indices.dispose();\n  reshapedTensor.dispose();\n  reshapedMask.dispose();\n  positivePositions.dispose();\n\n  return res;\n}\n\nexport const booleanMaskAsync = booleanMaskAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertShapesMatch} from '../util';\n\nimport {equal} from './equal';\nimport {greater} from './greater';\nimport {greaterEqual} from './greater_equal';\nimport {less} from './less';\nimport {lessEqual} from './less_equal';\nimport {notEqual} from './not_equal';\nimport {op} from './operation';\n\n/**\n * @deprecated\n * Strict version of `tf.notEqual` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction notEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'notEqualStrict');\n  const $b = convertToTensor(b, 'b', 'notEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in notEqualStrict: ');\n  return notEqual($a, $b);\n}\n\n/**\n * @deprecated\n * Strict version of `tf.less` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction lessStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessStrict');\n  const $b = convertToTensor(b, 'b', 'lessStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessStrict: ');\n  return less($a, $b);\n}\n\nfunction equalStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'equalStrict');\n  const $b = convertToTensor(b, 'b', 'equalStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in equalStrict: ');\n  return equal($a, $b);\n}\n\nfunction lessEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessEqualStrict');\n  const $b = convertToTensor(b, 'b', 'lessEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessEqualStrict: ');\n  return lessEqual($a, $b);\n}\n\nfunction greaterStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterStrict');\n  const $b = convertToTensor(b, 'b', 'greaterStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterStrict: ');\n  return greater($a, $b);\n}\n\nfunction greaterEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterEqualStrict');\n  const $b = convertToTensor(b, 'b', 'greaterEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterEqualStrict: ');\n  return greaterEqual($a, $b);\n}\n\nexport const equalStrict = op({equalStrict_});\nexport const greaterEqualStrict = op({greaterEqualStrict_});\nexport const greaterStrict = op({greaterStrict_});\nexport const lessEqualStrict = op({lessEqualStrict_});\nexport const lessStrict = op({lessStrict_});\nexport const notEqualStrict = op({notEqualStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {maximum} from './maximum';\nimport {minimum} from './minimum';\nimport {mod} from './mod';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {squaredDifference} from './squared_difference';\nimport {sub} from './sub';\n\n/**\n * @deprecated\n * Adds two `tf.Tensor`s element-wise, A + B.\n *\n * Inputs must be the same shape. For broadcasting support, use add() instead.\n *\n * @param a The first Tensor to add element-wise.\n * @param b The second Tensor to add element-wise.\n */\nfunction addStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'addStrict');\n  const $b = convertToTensor(b, 'b', 'addStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in addStrict: ');\n  return add($a, $b);\n}\n\n/**\n * @deprecated\n * Subtracts two `tf.Tensor`s element-wise, A - B. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.sub` instead.\n *\n * @param a The first Tensor to subtract element-wise.\n * @param b The second Tensor to subtract element-wise.\n */\nfunction subStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'subStrict');\n  const $b = convertToTensor(b, 'b', 'subStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in subStrict: ');\n  return sub($a, $b);\n}\n\n/**\n * @deprecated\n * Computes the power of one `tf.Tensor` to another. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.pow` instead.\n *\n * @param base The base tensor to pow element-wise.\n * @param exp The exponent tensor to pow element-wise.\n */\nfunction powStrict_<T extends Tensor>(base: T, exp: Tensor): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  util.assertShapesMatch(base.shape, exp.shape, 'Error in powStrict: ');\n  return pow(base, exp);\n}\n\n/**\n * @deprecated\n * Multiplies two `tf.Tensor`s element-wise, A * B.\n *\n * Inputs must be the same shape. For broadcasting support, use `tf.mul`.\n *\n * @param a The first tensor to multiply.\n * @param b The first tensor to multiply. Must have the same\n *    dtype as `a`.\n */\nfunction mulStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'mul');\n  const $b = convertToTensor(b, 'b', 'mul');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in multiplyStrict: ');\n  return mul($a, $b);\n}\n\n/**\n * @deprecated\n * Divides two `tf.Tensor`s element-wise, A / B. Inputs must\n * be the same shape.\n *\n * @param a The first tensor as the numerator for element-wise division.\n * @param b The second tensor as the denominator for element-wise division.\n */\nfunction divStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'div');\n  const $b = convertToTensor(b, 'b', 'div');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in divideStrict: ');\n  return div($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the mod of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use mod().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction modStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'modStrict');\n  const $b = convertToTensor(b, 'b', 'modStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in modStrict: ');\n  return mod($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the min of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use minimum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction minimumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'minimumStrict');\n  const $b = convertToTensor(b, 'b', 'minimumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in minimumStrict: ');\n  return minimum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the max of a and b (`a > b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use maximum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction maximumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'maximumStrict');\n  const $b = convertToTensor(b, 'b', 'maximumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in maximumStrict: ');\n  return maximum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns (a - b) * (a - b) element-wise.\n *\n * Inputs must be the same shape. For broadcasting support, use\n * `tf.squaredDifference` instead.\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n */\nfunction squaredDifferenceStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'squaredDifferenceStrict');\n  const $b = convertToTensor(b, 'b', 'squaredDifferenceStrict');\n  util.assertShapesMatch(\n      $a.shape, $b.shape, 'Error in squaredDifferenceStrict: ');\n  return squaredDifference($a, $b);\n}\n\nexport const addStrict = op({addStrict_});\nexport const divStrict = op({divStrict_});\nexport const maximumStrict = op({maximumStrict_});\nexport const minimumStrict = op({minimumStrict_});\nexport const modStrict = op({modStrict_});\nexport const mulStrict = op({mulStrict_});\nexport const powStrict = op({powStrict_});\nexport const squaredDifferenceStrict = op({squaredDifferenceStrict_});\nexport const subStrict = op({subStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {abs} from './abs';\nimport * as axis_util from './axis_util';\nimport {max} from './max';\nimport {min} from './min';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sqrt} from './sqrt';\nimport {square} from './square';\nimport {sum} from './sum';\n\n/**\n * Computes the norm of scalar, vectors, and matrices.\n * This function can compute several different vector norms (the 1-norm, the\n * Euclidean or 2-norm, the inf-norm, and in general the p-norm for p > 0)\n * and matrix norms (Frobenius, 1-norm, and inf-norm).\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.norm().print();  // or tf.norm(x)\n * ```\n *\n * @param x The input array.\n * @param ord Optional. Order of the norm. Supported norm types are\n * following:\n *\n *  | ord        | norm for matrices         | norm for vectors\n *  |------------|---------------------------|---------------------\n *  |'euclidean' |Frobenius norm             |2-norm\n *  |'fro'       |Frobenius norm\t           |\n *  |Infinity    |max(sum(abs(x), axis=1))   |max(abs(x))\n *  |-Infinity   |min(sum(abs(x), axis=1))   |min(abs(x))\n *  |1           |max(sum(abs(x), axis=0))   |sum(abs(x))\n *  |2           |                           |sum(abs(x)^2)^1/2*\n *\n * @param axis Optional. If axis is null (the default), the input is\n * considered a vector and a single vector norm is computed over the entire\n * set of values in the Tensor, i.e. norm(x, ord) is equivalent\n * to norm(x.reshape([-1]), ord). If axis is a integer, the input\n * is considered a batch of vectors, and axis determines the axis in x\n * over which to compute vector norms. If axis is a 2-tuple of integer it is\n * considered a batch of matrices and axis determines the axes in NDArray\n * over which to compute a matrix norm.\n * @param keepDims Optional. If true, the norm have the same dimensionality\n * as the input.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction norm_(\n    x: Tensor|TensorLike, ord: number|'euclidean'|'fro' = 'euclidean',\n    axis: number|number[] = null, keepDims = false): Tensor {\n  x = convertToTensor(x, 'x', 'norm');\n\n  const norm = normImpl(x, ord, axis);\n  let keepDimsShape = norm.shape;\n  if (keepDims) {\n    const axes = parseAxisParam(axis, x.shape);\n    keepDimsShape = axis_util.expandShapeToKeepDim(norm.shape, axes);\n  }\n  return reshape(norm, keepDimsShape);\n}\n\nfunction normImpl(\n    x: Tensor, p: number|string, axis: number|number[] = null): Tensor {\n  if (x.rank === 0) {\n    return abs(x);\n  }\n\n  // consider vector when no axis is specified\n  if (x.rank !== 1 && axis === null) {\n    return normImpl(reshape(x, [-1]), p, axis);\n  }\n\n  // vector\n  if (x.rank === 1 || typeof axis === 'number' ||\n      Array.isArray(axis) && axis.length === 1) {\n    if (p === 1) {\n      return sum(abs(x), axis);\n    }\n    if (p === Infinity) {\n      return max(abs(x), axis);\n    }\n    if (p === -Infinity) {\n      return min(abs(x), axis);\n    }\n    if (p === 'euclidean' || p === 2) {\n      // norm(x, 2) = sum(abs(xi) ^ 2) ^ 1/2\n      return sqrt(sum(pow(abs(x), scalar(2, 'int32')), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  // matrix (assumption axis[0] < axis[1])\n  if (Array.isArray(axis) && axis.length === 2) {\n    if (p === 1) {\n      return max(sum(abs(x), axis[0]), axis[1] - 1);\n    }\n    if (p === Infinity) {\n      return max(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === -Infinity) {\n      return min(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === 'fro' || p === 'euclidean') {\n      // norm(x) = sqrt(sum(pow(x, 2)))\n      return sqrt(sum(square(x), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  throw new Error(`Error in norm: invalid axis: ${axis}`);\n}\n\nexport const norm = op({norm_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor} from '../tensor';\nimport {assertTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {scalar} from './scalar';\nimport {sub} from './sub';\n\n/**\n * Compute the moving average of a variable.\n *\n * Without zeroDebias, the moving average operation is defined by:\n *   `v += delta`\n * where\n *   `delta = (1 - decay) * (x - v)`\n *\n * With zeroDebias (default), the `delta` term is scaled to debias the\n * effect of the (assumed) zero-initialization of `v`.\n *   `delta /= (1 - decay ^ step)`\n *\n * For more details on the zero-debiasing algorithm, see:\n *   https://arxiv.org/abs/1412.6980\n *\n * Note that this function is completely stateless and does not keep track of\n * step count. The step count needs to be maintained by the caller and passed\n * in as `step`.\n *\n * @param v The current moving average value.\n * @param x New input value, must have the same shape and dtype as `v`.\n * @param decay The decay factor. Typical values are 0.95 and 0.99.\n * @param step Step count.\n * @param zeroDebias: Whether zeroDebias is to be performed (default: `true`).\n * @returns The new moving average value.\n *\n * @doc {heading: 'Operations', subheading: 'Moving Average'}\n */\nfunction movingAverage_<T extends Tensor>(\n    v: T|TensorLike, x: T|TensorLike, decay: number|Scalar,\n    step?: number|Scalar, zeroDebias = true): T {\n  const $v = convertToTensor(v, 'v', 'movingAverage');\n  const $x = convertToTensor(x, 'x', 'movingAverage');\n  const $decay = convertToTensor(decay, 'decay', 'movingAverage');\n\n  assertTypesMatch($v, $x);\n  util.assert(\n      util.arraysEqual($v.shape, $x.shape), () => 'Shape mismatch in v and x');\n\n  const one = scalar(1);\n  const oneMinusDecay = sub(one, $decay);\n\n  let update = mul(sub($x, $v), oneMinusDecay);\n  if (zeroDebias) {\n    util.assert(\n        step != null, () => 'When using zeroDebias: true, step is required.');\n    const $step = convertToTensor(step, 'step', 'movingAverage');\n    update = div(update, sub(one, pow($decay, $step)));\n  }\n  return add($v, update);\n}\n\nexport const movingAverage = op({movingAverage_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ScatterNd, ScatterNdAttrs, ScatterNdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as scatter_nd_util from './scatter_nd_util';\n\n/**\n * Creates a new tensor by applying sparse updates to individual\n * values or slices within a zero tensor of the given shape tensor according to\n * indices. This operator is the inverse of the `tf.gatherND` operator which\n * extracts values or slices from a given tensor.\n *\n * ```js\n * const indices = tf.tensor2d([4, 3, 1, 7], [4, 1], 'int32');\n * const updates = tf.tensor1d([9, 10, 11, 12]);\n * const shape = [8];\n * tf.scatterND(indices, updates, shape).print() //[0, 11, 0, 10, 9, 0, 0, 12]\n * ```\n *\n * @param indices The tensor contains the indices into the output tensor.\n * @param updates The tensor contains the value for the indices.\n * @param shape: The shape of the output tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction scatterND_<R extends Rank>(\n    indices: Tensor|TensorLike, updates: Tensor|TensorLike,\n    shape: ShapeMap[R]): Tensor<R> {\n  const $indices = convertToTensor(indices, 'indices', 'scatterND', 'int32');\n  const $updates = convertToTensor(updates, 'updates', 'scatterND');\n  scatter_nd_util.validateInput($updates, $indices, shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.scatterND($indices, $updates, shape);\n  };\n\n  const inputs: ScatterNdInputs = {indices: $indices, updates: $updates};\n  const attrs: ScatterNdAttrs = {shape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             ScatterNd, attrs as {} as NamedAttrMap) as Tensor<R>;\n}\n\nexport const scatterND = op({scatterND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\n\n/**\n * Validate sparseToDense inputs.\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape number[]. Shape of the dense output tensor.\n * @param validateIndices boolean. indice validation is not supported, error\n * will be thrown if it is set.\n */\nexport function validateInput(\n    sparseIndices: Tensor, sparseValues: Tensor, outputShape: number[],\n    defaultValues: Tensor) {\n  if (sparseIndices.dtype !== 'int32') {\n    throw new Error(\n        'tf.sparseToDense() expects the indices to be int32 type,' +\n        ` but the dtype was ${sparseIndices.dtype}.`);\n  }\n  if (sparseIndices.rank > 2) {\n    throw new Error(\n        'sparseIndices should be a scalar, vector, or matrix,' +\n        ` but got shape ${sparseIndices.shape}.`);\n  }\n\n  const numElems = sparseIndices.rank > 0 ? sparseIndices.shape[0] : 1;\n  const numDims = sparseIndices.rank > 1 ? sparseIndices.shape[1] : 1;\n\n  if (outputShape.length !== numDims) {\n    throw new Error(\n        'outputShape has incorrect number of elements:,' +\n        ` ${outputShape.length}, should be: ${numDims}.`);\n  }\n\n  const numValues = sparseValues.size;\n  if (!(sparseValues.rank === 0 ||\n        sparseValues.rank === 1 && numValues === numElems)) {\n    throw new Error(\n        'sparseValues has incorrect shape ' +\n        `${sparseValues.shape}, should be [] or [${numElems}]`);\n  }\n\n  if (sparseValues.dtype !== defaultValues.dtype) {\n    throw new Error('sparseValues.dtype must match defaultValues.dtype');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {SparseToDense, SparseToDenseAttrs, SparseToDenseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport * as sparse_to_dense from '../ops/sparse_to_dense_util';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ScalarLike, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Converts a sparse representation into a dense tensor.\n *\n * Builds an array dense with shape outputShape such that:\n *\n * // If sparseIndices is scalar\n * dense[i] = (i == sparseIndices ? sparseValues : defaultValue)\n *\n * // If sparseIndices is a vector, then for each i\n * dense[sparseIndices[i]] = sparseValues[i]\n *\n * // If sparseIndices is an n by d matrix, then for each i in [0, n)\n * dense[sparseIndices[i][0], ..., sparseIndices[i][d-1]] = sparseValues[i]\n * All other values in dense are set to defaultValue. If sparseValues is a\n * scalar, all sparse indices are set to this single value.\n *\n * If indices are repeated the final value is summed over all values for those\n * indices.\n *\n * ```js\n * const indices = tf.tensor1d([4, 5, 6, 1, 2, 3], 'int32');\n * const values = tf.tensor1d([10, 11, 12, 13, 14, 15], 'float32');\n * const shape = [8];\n * tf.sparseToDense(indices, values, shape).print();\n * ```\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape Shape of the dense output tensor. the type is inferred.\n * @param defaultValue Scalar. Value to set for indices not specified in\n * sparseIndices. Defaults to zero.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction sparseToDense_<R extends Rank>(\n    sparseIndices: Tensor|TensorLike, sparseValues: Tensor|TensorLike,\n    outputShape: ShapeMap[R], defaultValue: Scalar|ScalarLike = 0): Tensor<R> {\n  const $sparseIndices =\n      convertToTensor(sparseIndices, 'sparseIndices', 'sparseToDense', 'int32');\n  const $sparseValues =\n      convertToTensor(sparseValues, 'sparseValues', 'sparseToDense');\n  const $defaultValue = convertToTensor(\n      defaultValue, 'defaultValue', 'sparseToDense', $sparseValues.dtype);\n\n  sparse_to_dense.validateInput(\n      $sparseIndices, $sparseValues, outputShape, $defaultValue);\n\n  const inputs: SparseToDenseInputs = {\n    sparseIndices: $sparseIndices,\n    sparseValues: $sparseValues,\n    defaultValue: $defaultValue\n  };\n\n  const attrs: SparseToDenseAttrs = {outputShape};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.sparseToDense(\n          $sparseIndices, $sparseValues, outputShape, $defaultValue),\n      inputs as {} as NamedTensorMap, null /* grad */, SparseToDense,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const sparseToDense = op({sparseToDense_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherNd, GatherNdInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Gather slices from input tensor into a Tensor with shape specified by\n * `indices`.\n *\n * `indices` is an K-dimensional integer tensor, best thought of as a\n * (K-1)-dimensional tensor of indices into input, where each element defines a\n * slice of input:\n * output[\\\\(i_0, ..., i_{K-2}\\\\)] = input[indices[\\\\(i_0, ..., i_{K-2}\\\\)]]\n *\n * Whereas in `tf.gather`, `indices` defines slices into the first dimension of\n * input, in `tf.gatherND`, `indices` defines slices into the first N dimensions\n * of input, where N = indices.shape[-1].\n *\n * The last dimension of indices can be at most the rank of input:\n * indices.shape[-1] <= input.rank\n *\n * The last dimension of `indices` corresponds to elements\n * (if indices.shape[-1] == input.rank) or slices\n * (if indices.shape[-1] < input.rank) along dimension indices.shape[-1] of\n * input.\n * The output tensor has shape\n * indices.shape[:-1] + input.shape[indices.shape[-1]:]\n *\n * Note that on CPU, if an out of bound index is found, an error is returned. On\n * GPU, if an out of bound index is found, a 0 is stored in the corresponding\n * output value.\n *\n * ```js\n * const indices = tf.tensor2d([0, 1, 1, 0], [2,2], 'int32');\n * const input = tf.tensor2d([9, 10, 11, 12], [2, 2]);\n * tf.gatherND(input, indices).print() // [10, 11]\n * ```\n *\n * @param x The tensor from which to gather values.\n * @param indices Index tensor, must be of type int32.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction gatherND_(x: Tensor|TensorLike, indices: Tensor|TensorLike): Tensor {\n  const $indices = convertToTensor(indices, 'indices', 'gatherND', 'int32');\n  const $x = convertToTensor(x, 'x', 'gatherND');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.gatherND($x, $indices);\n  };\n\n  const inputs: GatherNdInputs = {params: $x, indices: $indices};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */, GatherNd);\n}\n\nexport const gatherND = op({gatherND_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\n/**\n * Normalize noise shape based on provided tensor and noise shape.\n *\n * @param x Tensor.\n * @param noiseShape The shape for the randomly generated keep/drop flags, as\n *   an array of numbers. Optional.\n * @returns Normalized noise shape.\n */\nexport function getNoiseShape(x: Tensor, noiseShape?: number[]): number[] {\n  if (noiseShape == null) {\n    return x.shape.slice();\n  }\n  if (util.arraysEqual(x.shape, noiseShape)) {\n    return noiseShape;\n  }\n  if (x.shape.length === noiseShape.length) {\n    const newDimension: number[] = [];\n    for (let i = 0; i < x.shape.length; i++) {\n      if (noiseShape[i] == null && x.shape[i] != null) {\n        newDimension.push(x.shape[i]);\n      } else {\n        newDimension.push(noiseShape[i]);\n      }\n    }\n    return newDimension;\n  }\n\n  return noiseShape;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {getNoiseShape} from './dropout_util';\nimport {floor} from './floor';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {randomUniform} from './random_uniform';\n\n/**\n * Computes dropout.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 2, 1]);\n * const rate = 0.75;\n * const output = tf.dropout(x, rate);\n * output.print();\n * ```\n *\n * @param x A floating point Tensor or TensorLike.\n * @param rate A float in the range [0, 1). The probability that each element\n *   of x is discarded.\n * @param noiseShape An array of numbers of type int32, representing the\n * shape for randomly generated keep/drop flags. If the noiseShape has null\n * value, it will be automatically replaced with the x's relative dimension\n * size. Optional.\n * @param seed Used to create random seeds. Optional.\n * @returns A Tensor of the same shape of x.\n *\n * @doc {heading: 'Operations', subheading: 'Dropout'}\n */\nfunction dropout_(\n    x: Tensor|TensorLike, rate: number, noiseShape?: number[],\n    seed?: number|string): Tensor {\n  const $x = convertToTensor(x, 'x', 'dropout');\n\n  util.assert(\n      $x.dtype === 'float32',\n      () => `x has to be a floating point tensor since it's going to be ` +\n          `scaled, but got a ${$x.dtype} tensor instead.`);\n  util.assert(\n      rate >= 0 && rate < 1,\n      () => `rate must be a float in the range [0, 1), but got ${rate}.`);\n\n  if (rate === 0) {\n    return x instanceof Tensor ? $x.clone() : $x;\n  }\n\n  const $noiseShape = getNoiseShape($x, noiseShape);\n  const keepProb = 1 - rate;\n  const multiplier = div(\n      floor(add(randomUniform($noiseShape, 0, 1, 'float32', seed), keepProb)),\n      keepProb);\n\n  return mul($x, multiplier);\n}\n\nexport const dropout = op({dropout_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {tensor1d} from './tensor1d';\n\nexport function enclosingPowerOfTwo(value: number) {\n  // Return 2**N for integer N such that 2**N >= value.\n  return Math.floor(Math.pow(2, Math.ceil(Math.log(value) / Math.log(2.0))));\n}\n\nexport function cosineWindow(\n    windowLength: number, a: number, b: number): Tensor1D {\n  const even = 1 - windowLength % 2;\n  const newValues = new Float32Array(windowLength);\n  for (let i = 0; i < windowLength; ++i) {\n    const cosArg = (2.0 * Math.PI * i) / (windowLength + even - 1);\n    newValues[i] = a - b * Math.cos(cosArg);\n  }\n  return tensor1d(newValues, 'float32');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch, getTypedArrayFromDType} from '../util';\nimport {tensor} from './tensor';\n\n/**\n * Returns whether the targets are in the top K predictions.\n *\n * ```js\n * const predictions = tf.tensor2d([[20, 10, 40, 30], [30, 50, -20, 10]]);\n * const targets = tf.tensor1d([2, 0]);\n * const precision = await tf.inTopKAsync(predictions, targets);\n * precision.print();\n * ```\n * @param predictions 2-D or higher `tf.Tensor` with last dimension being\n *     at least `k`.\n * @param targets 1-D or higher `tf.Tensor`.\n * @param k Optional Number of top elements to look at for computing precision,\n *     default to 1.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nasync function inTopKAsync_<T extends Tensor, U extends Tensor>(\n    predictions: T|TensorLike, targets: U|TensorLike, k = 1): Promise<U> {\n  const $predictions = convertToTensor(predictions, 'predictions', 'inTopK');\n  const $targets = convertToTensor(targets, 'targets', 'inTopK');\n\n  assert(\n      $predictions.rank > 1,\n      () => 'inTopK() expects the predictions to be of rank 2 or higher, ' +\n          `but got ${$predictions.rank}`);\n  assert(\n      $predictions.rank - 1 === $targets.rank,\n      () => `predictions rank should be 1 larger than ` +\n          `targets rank, but got predictions rank ` +\n          `${$predictions.rank} and targets rank ${$targets.rank}`);\n  assertShapesMatch(\n      $predictions.shape.slice(0, $predictions.shape.length - 1),\n      $targets.shape,\n      `predictions's shape should be align with the targets' shape, ` +\n          'except the last dimension.');\n  const lastDim = $predictions.shape[$predictions.shape.length - 1];\n  assert(\n      k > 0 && k <= lastDim,\n      () => `'k' passed to inTopK() must be > 0 && <= the predictions last ` +\n          `dimension (${lastDim}), but got ${k}`);\n\n  const predictionsVals = await $predictions.data();\n  const targetsVals = await $targets.data();\n\n  // Reshape predictionsVals into a 2d tensor [batch, lastDim]\n  // and look up topK along lastDim.\n  const [batch, size] = [predictionsVals.length / lastDim, lastDim];\n  const precision = getTypedArrayFromDType('bool', batch);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = predictionsVals.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    precision[b] = 0;\n    for (let i = 0; i < k; i++) {\n      if (valAndInd[i].index === targetsVals[b]) {\n        precision[b] = 1;\n        break;\n      }\n    }\n  }\n\n  if (predictions !== $predictions) {\n    $predictions.dispose();\n  }\n  if (targets !== $targets) {\n    $targets.dispose();\n  }\n\n  // Output precision has the same shape as targets.\n  return tensor(precision, $targets.shape, 'bool') as U;\n}\n\nexport const inTopKAsync = inTopKAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropFilter, Conv2DBackpropFilterAttrs, Conv2DBackpropFilterInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 2D convolution.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     [batch, height, width, inChannels]. If rank 3, batch of 1 is assumed.\n * @param dy The dy image, of rank 4 or rank 3, of shape\n *     [batch, height, width, outDepth]. If rank 3, batch of 1 is assumed.\n * @param filterShape The shape of the filter, length 4,\n *     [filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction conv2DBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2dDerFilter: input must be rank 4, but got shape ` +\n          `${x4D.shape}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerFilter: dy must be rank 4, but got shape ` +\n          `${dy4D.shape}.`);\n  util.assert(\n      filterShape.length === 4,\n      () => `Error in conv2dDerFilter: filterShape must be length 4, but got ` +\n          `${filterShape}.`);\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filterShape[2],\n      () => `Error in conv2dDerFilter: depth of input ${inDepth}) must ` +\n          `match input depth in filter (${filterShape[2]}.`);\n  util.assert(\n      outDepth === filterShape[3],\n      () => `Error in conv2dDerFilter: depth of dy (${outDepth}) must ` +\n          `match output depth for filter (${filterShape[3]}).`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerFilter: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, filterShape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    return backend.conv2dDerFilter(x4D, dy4D, convInfo);\n  };\n\n  const inputs: Conv2DBackpropFilterInputs = {x: x4D, dy: dy4D};\n  const attrs:\n      Conv2DBackpropFilterAttrs = {strides, pad, dataFormat, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv2DBackpropFilter, attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const conv2DBackpropFilter = op({conv2DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\nimport * as broadcast_util from './broadcast_util';\nimport {elu} from './elu';\nimport {Activation} from './fused_types';\nimport {mul} from './mul';\nimport {prelu} from './prelu';\nimport {relu} from './relu';\nimport {relu6} from './relu6';\nimport {reshape} from './reshape';\nimport {step} from './step';\nimport {sum} from './sum';\n\n// Returns gradient for fused activation.\nexport function getFusedDyActivation(\n    dy: Tensor, y: Tensor, activation: Activation): Tensor {\n  if (activation == null || activation === 'linear') {\n    return dy;\n  }\n  if (activation === 'relu') {\n    return mul(dy, step(y));\n  }\n  throw new Error(\n      `Cannot compute gradient for fused activation ${activation}.`);\n}\n\n// Returns gradient for fused bias.\nexport function getFusedBiasGradient(\n    bias: Tensor, dyActivation: Tensor): Tensor {\n  let res = dyActivation;\n  const reduceAxes =\n      broadcast_util.getReductionAxes(bias.shape, dyActivation.shape);\n  if (reduceAxes.length > 0) {\n    res = sum(res, reduceAxes);\n  }\n  return reshape(res, bias.shape);\n}\n\nexport function applyActivation(\n    x: Tensor, activation: Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return x;\n  } else if (activation === 'relu') {\n    return relu(x);\n  } else if (activation === 'elu') {\n    return elu(x);\n  } else if (activation === 'relu6') {\n    return relu6(x);\n  } else if (activation === 'prelu') {\n    return prelu(x, preluActivationWeights);\n  }\n  throw new Error(`Unknown fused activation ${activation}.`);\n}\n\n// Whether we should call fused ops.\nexport const shouldFuse = (gradientDepth: number, activation: Activation) => {\n  const gradientMode = gradientDepth > 0;\n  return !gradientMode || activation === 'linear';\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedConv2D, FusedConv2DAttrs, FusedConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {conv2d as unfusedConv2d} from '../conv2d';\nimport {conv2DBackpropFilter} from '../conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../conv2d_backprop_input';\nimport * as conv_util from '../conv_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes a 2D convolution over the input x, optionally fused with adding a\n * bias and applying an activation.\n *\n * ```js\n * const inputDepth = 2;\n * const inShape = [2, 2, 2, inputDepth];\n * const outputDepth = 2;\n * const fSize = 1;\n * const pad = 0;\n * const strides = 1;\n *\n * const x = tf.tensor4d( [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,\n * 16], inShape);\n * const w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth,\n * outputDepth]);\n *\n * tf.fused.conv2d({ x, filter: w, strides, pad, dataFormat: 'NHWC',\n * dilations: [1, 1], bias: tf.scalar(5), activation: 'relu' }).print();\n * ```\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid` output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`) to be\n *     applied\n *      after biasAdd.\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  activation = activation || 'linear';\n\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused conv2d: input must be rank 4, but got rank ` +\n          `${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${x4D.shape[3]}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in conv2d: got dataFormat of ${\n          dataFormat} but only NHWC is currently supported.`);\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused conv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    const [$filter, x4D, y, $bias] =\n        saved as [Tensor4D, Tensor4D, Tensor4D, Tensor];\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused conv2D: ' +\n            `dilation rates greater than 1 ` +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const xDer =\n        conv2DBackpropInput(x4D.shape, dyActivation, $filter, strides, pad);\n    const filterDer =\n        conv2DBackpropFilter(x4D, dyActivation, $filter.shape, strides, pad);\n    const der: Tensor[] = [xDer, filterDer];\n\n    if ($bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      der.push(biasDer);\n    }\n    return der;\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedConv2d({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n\n  const attrs: FusedConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const conv2d = op({fusedConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropFilterInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    convInfo: conv_util.Conv2DInfo): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerFilter(x4D, dy4D, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropFilterInputs = {x: x4D, dy: dy4D};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             DepthwiseConv2dNativeBackpropFilter) as Tensor4D;\n}\n\nexport const depthwiseConv2dNativeBackpropFilter =\n    op({depthwiseConv2dNativeBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropInput, DepthwiseConv2dNativeBackpropInputInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, convInfo: conv_util.Conv2DInfo): T {\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerInput(dy4D, filter, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropInputInputs = {dy: dy4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null,\n      DepthwiseConv2dNativeBackpropInput);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2dNativeBackpropInput =\n    op({depthwiseConv2dNativeBackpropInput_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedDepthwiseConv2D, FusedDepthwiseConv2DAttrs, FusedDepthwiseConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport * as conv_util from '../conv_util';\nimport {depthwiseConv2d as unfusedDepthwiseConv2d} from '../depthwise_conv2d';\nimport {depthwiseConv2dNativeBackpropFilter} from '../depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../depthwise_conv2d_native_backprop_input';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes depthwise 2D convolution, optionally fused with adding a\n * bias and applying an activation.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`).\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedDepthwiseConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedDepthwiseConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused depthwiseConv2d: filter must be rank 4, ` +\n          `but got rank ${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in fused depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () =>\n          'Error in fused depthwiseConv2d: Either strides or dilations must ' +\n          `be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused depthwiseConv2d: pad must be an integer when ` +\n            `using dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n      true /* depthwise */);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused depthwiseConv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused depthwiseConv2d: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${dilations}'`);\n    const [$filter, x4D, y, bias] = saved;\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    const xDer = depthwiseConv2dNativeBackpropInput(\n        (x4D as Tensor4D).shape, dyActivation, $filter as Tensor4D, convInfo);\n    const filterDer = depthwiseConv2dNativeBackpropFilter(\n        x4D as Tensor4D, dyActivation, ($filter as Tensor4D).shape, convInfo);\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [xDer, filterDer, biasDer];\n    }\n    return [xDer, filterDer];\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedDepthwiseConv2D({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedDepthwiseConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: FusedDepthwiseConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const depthwiseConv2d = op({fusedDepthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {_FusedMatMul, _FusedMatMulAttrs, _FusedMatMulInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {matMul as unfusedMatMul} from '../mat_mul';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes the dot product of two matrices with optional activation and bias.\n *\n * ```js\n * const a = tf.tensor2d([-1, -2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const bias = tf.tensor2d([1, 2], [1, 2]);\n *\n * tf.fused.matMul({a, b, bias, activation: 'relu'}).print();\n * ```\n *\n * @param obj An object with the following properties:\n * - `a` First matrix in dot product operation.\n * - `b` Second matrix in dot product operation.\n * - `transposeA` If true, `a` is transposed before multiplication.\n * - `transposeB` If true, `b` is transposed before multiplication.\n * - `bias` Matrix to be added to the result.\n * - `activation` Name of activation kernel (defaults to `linear`).\n * - `preluActivationWeights` Tensor of prelu weights.\n */\nfunction fusedMatMul_<T extends Tensor>({\n  a,\n  b,\n  transposeA = false,\n  transposeB = false,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  a: T|TensorLike,\n  b: T|TensorLike,\n  transposeA?: boolean,\n  transposeB?: boolean,\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedMatMul(a, b, transposeA, transposeB);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  let $a = convertToTensor(a, 'a', 'fused matMul');\n  let $b = convertToTensor(b, 'b', 'fused matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () =>\n          `Error in fused matMul: inputs must have the same rank of at least ` +\n          `2, got ranks ${$a.rank} and ${$b.rank}.`);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in fused matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in fused matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D: Tensor3D = transposeA ?\n      reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n      reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D: Tensor3D = transposeB ?\n      reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n      reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused matMul');\n    [$bias] = makeTypesMatch($bias, $a);\n\n    broadcast_util.assertAndGetBroadcastShape(outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused matMul');\n  }\n\n  const grad = (dy: Tensor3D, saved: Tensor[]) => {\n    const [a3D, b3D, y, $bias] = saved;\n    // we reshape dy because the result of the forward is not\n    // necessarily going to be a 3d tensor due to a reshape done at the end of\n    // the customOp.\n    const dyActivation =\n        getFusedDyActivation(reshape(dy, y.shape), y, activation);\n    let aDer: Tensor;\n    let bDer: Tensor;\n\n    if (!transposeA && !transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, true, false);\n    } else if (!transposeA && transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, false);\n      bDer = unfusedMatMul(dyActivation, a3D, true, false);\n    } else if (transposeA && !transposeB) {\n      aDer = unfusedMatMul(b3D, dyActivation, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, false, false);\n    } else {\n      aDer = unfusedMatMul(b3D, dyActivation, true, true);\n      bDer = unfusedMatMul(dyActivation, a3D, true, true);\n    }\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [aDer, bDer, biasDer];\n    } else {\n      return [aDer, bDer];\n    }\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const y = backend.fusedBatchMatMul({\n      a: a3D,\n      b: b3D,\n      transposeA,\n      transposeB,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return y;\n  };\n\n  const inputs: _FusedMatMulInputs = {\n    a: a3D,\n    b: b3D,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: _FusedMatMulAttrs = {transposeA, transposeB, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((a3D: Tensor3D, b3D: Tensor3D, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n    return customOp(a3D, b3D) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (a3D: Tensor3D, b3D: Tensor3D, $bias: Tensor, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res, $bias]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n\n    return customOpWithBias(a3D, b3D, $bias) as T;\n  }\n}\n\nexport const matMul = op({fusedMatMul_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {conv2d} from './fused/conv2d';\nimport {depthwiseConv2d} from './fused/depthwise_conv2d';\nimport {matMul} from './fused/mat_mul';\nimport {Activation} from './fused_types';\n\nexport {Activation, conv2d, depthwiseConv2d, matMul};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a hamming window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hammingWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hammingWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.54, 0.46);\n}\nexport const hammingWindow = op({hammingWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a Hann window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hannWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hannWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.5, 0.5);\n}\n\nexport const hannWindow = op({hannWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {fill} from '../fill';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {tensor2d} from '../tensor2d';\n\n/**\n * Expands input into frames of frameLength.\n * Slides a window size with frameStep.\n *\n * ```js\n * tf.signal.frame([1, 2, 3], 2, 1).print();\n * ```\n * @param signal The input tensor to be expanded\n * @param frameLength Length of each frame\n * @param frameStep The frame hop size in samples.\n * @param padEnd Whether to pad the end of signal with padValue.\n * @param padValue An number to use where the input signal does\n *     not exist when padEnd is True.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction frame_(\n    signal: Tensor1D, frameLength: number, frameStep: number, padEnd = false,\n    padValue = 0): Tensor {\n  let start = 0;\n  const output: Tensor[] = [];\n  while (start + frameLength <= signal.size) {\n    output.push(slice(signal, start, frameLength));\n    start += frameStep;\n  }\n\n  if (padEnd) {\n    while (start < signal.size) {\n      const padLen = (start + frameLength) - signal.size;\n      const pad = concat([\n        slice(signal, start, frameLength - padLen), fill([padLen], padValue)\n      ]);\n      output.push(pad);\n      start += frameStep;\n    }\n  }\n\n  if (output.length === 0) {\n    return tensor2d([], [0, frameLength]);\n  }\n\n  return reshape(concat(output), [output.length, frameLength]);\n}\nexport const frame = op({frame_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {enclosingPowerOfTwo} from '../signal_ops_util';\nimport {slice} from '../slice';\nimport {rfft} from '../spectral/rfft';\n\nimport {frame} from './frame';\nimport {hannWindow} from './hann_window';\n\n/**\n * Computes the Short-time Fourier Transform of signals\n * See: https://en.wikipedia.org/wiki/Short-time_Fourier_transform\n *\n * ```js\n * const input = tf.tensor1d([1, 1, 1, 1, 1])\n * tf.signal.stft(input, 3, 1).print();\n * ```\n * @param signal 1-dimensional real value tensor.\n * @param frameLength The window length of samples.\n * @param frameStep The number of samples to step.\n * @param fftLength The size of the FFT to apply.\n * @param windowFn A callable that takes a window length and returns 1-d tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction stft_(\n    signal: Tensor1D, frameLength: number, frameStep: number,\n    fftLength?: number,\n    windowFn: (length: number) => Tensor1D = hannWindow): Tensor {\n  if (fftLength == null) {\n    fftLength = enclosingPowerOfTwo(frameLength);\n  }\n  const framedSignal = frame(signal, frameLength, frameStep);\n  const windowedSignal = mul(framedSignal, windowFn(frameLength));\n  const output: Tensor[] = [];\n  for (let i = 0; i < framedSignal.shape[0]; i++) {\n    output.push(\n        rfft(slice(windowedSignal, [i, 0], [1, frameLength]), fftLength));\n  }\n  return concat(output);\n}\nexport const stft = op({stft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {CropAndResize, CropAndResizeAttrs, CropAndResizeInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor1D, Tensor2D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Extracts crops from the input image tensor and resizes them using bilinear\n * sampling or nearest neighbor sampling (possibly with aspect ratio change)\n * to a common output size specified by crop_size.\n *\n * @param image 4d tensor of shape `[batch,imageHeight,imageWidth, depth]`,\n *     where imageHeight and imageWidth must be positive, specifying the\n *     batch of images from which to take crops\n * @param boxes 2d float32 tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the normalized\n *     coordinates of the box in the boxInd[i]'th image in the batch\n * @param boxInd 1d int32 tensor of shape `[numBoxes]` with values in range\n *     `[0, batch)` that specifies the image that the `i`-th box refers to.\n * @param cropSize 1d int32 tensor of 2 elements `[cropHeigh, cropWidth]`\n *     specifying the size to which all crops are resized to.\n * @param method Optional string from `'bilinear' | 'nearest'`,\n *     defaults to bilinear, which specifies the sampling method for resizing\n * @param extrapolationValue A threshold for deciding when to remove boxes based\n *     on score. Defaults to 0.\n * @return A 4D tensor of the shape `[numBoxes,cropHeight,cropWidth,depth]`\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction cropAndResize_(\n    image: Tensor4D|TensorLike,\n    boxes: Tensor2D|TensorLike,\n    boxInd: Tensor1D|TensorLike,\n    cropSize: [number, number],\n    method?: 'bilinear'|'nearest',\n    extrapolationValue?: number,\n    ): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'cropAndResize');\n  const $boxes = convertToTensor(boxes, 'boxes', 'cropAndResize', 'float32');\n  const $boxInd = convertToTensor(boxInd, 'boxInd', 'cropAndResize', 'int32');\n  method = method || 'bilinear';\n  extrapolationValue = extrapolationValue || 0;\n\n  const numBoxes = $boxes.shape[0];\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in cropAndResize: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n  util.assert(\n      $boxes.rank === 2 && $boxes.shape[1] === 4,\n      () => `Error in cropAndResize: boxes must be have size [${numBoxes},4] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      $boxInd.rank === 1 && $boxInd.shape[0] === numBoxes,\n      () => `Error in cropAndResize: boxInd must be have size [${numBoxes}] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      cropSize.length === 2,\n      () => `Error in cropAndResize: cropSize must be of length 2, but got ` +\n          `length ${cropSize.length}.`);\n  util.assert(\n      cropSize[0] >= 1 && cropSize[1] >= 1,\n      () => `cropSize must be atleast [1,1], but was ${cropSize}`);\n  util.assert(\n      method === 'bilinear' || method === 'nearest',\n      () => `method must be bilinear or nearest, but was ${method}`);\n\n  const forward: ForwardFunc<Tensor4D> = (backend) => backend.cropAndResize(\n      $image, $boxes, $boxInd, cropSize, method, extrapolationValue);\n\n  const inputs:\n      CropAndResizeInputs = {image: $image, boxes: $boxes, boxInd: $boxInd};\n  const attrs: CropAndResizeAttrs = {method, extrapolationValue, cropSize};\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, CropAndResize,\n      attrs as {} as NamedAttrMap);\n  return res;\n}\n\nexport const cropAndResize = op({cropAndResize_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FlipLeftRight, FlipLeftRightInputs} from '../../kernel_names';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {op} from '../operation';\n\n/**\n * Flips the image left to right. Currently available in the CPU, WebGL, and\n * WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n */\n/** @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'} */\nfunction flipLeftRight_(image: Tensor4D|TensorLike): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'flipLeftRight', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in flipLeftRight: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: FlipLeftRightInputs = {image: $image};\n  const res =\n      ENGINE.runKernel(FlipLeftRight, inputs as {} as NamedTensorMap, {});\n  return res as Tensor4D;\n}\n\nexport const flipLeftRight = op({flipLeftRight_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Rotates the input image tensor counter-clockwise with an optional offset\n * center of rotation. Currently available in the CPU, WebGL, and WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n * @param radians The amount of rotation.\n * @param fillValue The value to fill in the empty space leftover\n *     after rotation. Can be either a single grayscale value (0-255), or an\n *     array of three numbers `[red, green, blue]` specifying the red, green,\n *     and blue channels. Defaults to `0` (black).\n * @param center The center of rotation. Can be either a single value (0-1), or\n *     an array of two numbers `[centerX, centerY]`. Defaults to `0.5` (rotates\n *     the image around its center).\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction rotateWithOffset_(\n    image: Tensor4D|TensorLike, radians: number,\n    fillValue: number|[number, number, number] = 0,\n    center: number|[number, number] = 0.5): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'rotateWithOffset', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in rotateWithOffset: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: RotateWithOffsetInputs = {image: $image};\n  const attrs: RotateWithOffsetAttrs = {radians, fillValue, center};\n  const res = ENGINE.runKernel(\n      RotateWithOffset, inputs as {} as NamedTensorMap,\n      attrs as {} as NamedAttrMap);\n  return res as Tensor4D;\n}\n\nexport const rotateWithOffset = op({rotateWithOffset_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport * as util from '../util';\n\nfunction nonMaxSuppSanityCheck(\n    boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma?: number): {\n  maxOutputSize: number,\n  iouThreshold: number,\n  scoreThreshold: number,\n  softNmsSigma: number\n} {\n  if (iouThreshold == null) {\n    iouThreshold = 0.5;\n  }\n  if (scoreThreshold == null) {\n    scoreThreshold = Number.NEGATIVE_INFINITY;\n  }\n  if (softNmsSigma == null) {\n    softNmsSigma = 0.0;\n  }\n\n  const numBoxes = boxes.shape[0];\n  maxOutputSize = Math.min(maxOutputSize, numBoxes);\n\n  util.assert(\n      0 <= iouThreshold && iouThreshold <= 1,\n      () => `iouThreshold must be in [0, 1], but was '${iouThreshold}'`);\n  util.assert(\n      boxes.rank === 2,\n      () => `boxes must be a 2D tensor, but was of rank '${boxes.rank}'`);\n  util.assert(\n      boxes.shape[1] === 4,\n      () =>\n          `boxes must have 4 columns, but 2nd dimension was ${boxes.shape[1]}`);\n  util.assert(scores.rank === 1, () => 'scores must be a 1D tensor');\n  util.assert(\n      scores.shape[0] === numBoxes,\n      () => `scores has incompatible shape with boxes. Expected ${numBoxes}, ` +\n          `but was ${scores.shape[0]}`);\n  util.assert(\n      0 <= softNmsSigma && softNmsSigma <= 1,\n      () => `softNmsSigma must be in [0, 1], but was '${softNmsSigma}'`);\n  return {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n}\n\nexport {nonMaxSuppSanityCheck};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV3} from '../../kernel_names';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\nfunction nonMaxSuppression_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Tensor1D {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const attrs = {maxOutputSize, iouThreshold, scoreThreshold};\n  return ENGINE.runKernelFunc(\n      b => b.nonMaxSuppression(\n          $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold),\n      {boxes: $boxes, scores: $scores}, null /* grad */, NonMaxSuppressionV3,\n      attrs);\n}\n\nexport const nonMaxSuppression = op({nonMaxSuppression_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inserts a value into a sorted array. This method allows duplicate, meaning it\n * allows inserting duplicate value, in which case, the element will be inserted\n * at the lowest index of the value.\n * @param arr The array to modify.\n * @param element The element to insert.\n * @param comparator Optional. If no comparator is specified, elements are\n * compared using array_util.defaultComparator, which is suitable for Strings\n * and Numbers in ascending arrays. If the array contains multiple instances of\n * the target value, the left-most instance will be returned. To provide a\n * comparator, it should take 2 arguments to compare and return a negative,\n * zero, or a positive number.\n */\nexport function binaryInsert<T>(\n    arr: T[], element: T, comparator?: (a: T, b: T) => number) {\n  const index = binarySearch(arr, element, comparator);\n  const insertionPoint = index < 0 ? -(index + 1) : index;\n  arr.splice(insertionPoint, 0, element);\n}\n\n/**\n * Searches the array for the target using binary search, returns the index\n * of the found element, or position to insert if element not found. If no\n * comparator is specified, elements are compared using array_\n * util.defaultComparator, which is suitable for Strings and Numbers in\n * ascending arrays. If the array contains multiple instances of the target\n * value, the left-most instance will be returned.\n * @param arr The array to be searched in.\n * @param target The target to be searched for.\n * @param comparator Should take 2 arguments to compare and return a negative,\n *    zero, or a positive number.\n * @return Lowest index of the target value if found, otherwise the insertion\n *    point where the target should be inserted, in the form of\n *    (-insertionPoint - 1).\n */\nexport function binarySearch<T>(\n    arr: T[], target: T, comparator?: (a: T, b: T) => number) {\n  return binarySearch_(arr, target, comparator || defaultComparator);\n}\n\n/**\n * Compares its two arguments for order.\n * @param a The first element to be compared.\n * @param b The second element to be compared.\n * @return A negative number, zero, or a positive number as the first\n *     argument is less than, equal to, or greater than the second.\n */\nfunction defaultComparator<T>(a: T, b: T): number {\n  return a > b ? 1 : a < b ? -1 : 0;\n}\n\nfunction binarySearch_<T>(\n    arr: T[], target: T, comparator: (a: T, b: T) => number) {\n  let left = 0;\n  let right = arr.length;\n  let middle = 0;\n  let found = false;\n  while (left < right) {\n    middle = left + ((right - left) >>> 1);\n    const compareResult = comparator(target, arr[middle]);\n    if (compareResult > 0) {\n      left = middle + 1;\n    } else {\n      right = middle;\n      // If compareResult is 0, the value is found. We record it is found,\n      // and then keep looking because there may be duplicate.\n      found = !compareResult;\n    }\n  }\n\n  return found ? left : -left - 1;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Implementation of the NonMaxSuppression kernel shared between webgl and cpu.\n */\n\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\n\nimport {binaryInsert} from './array_util';\n\ninterface Candidate {\n  score: number;\n  boxIndex: number;\n  suppressBeginIndex: number;\n}\n\nexport function nonMaxSuppressionV3Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number): Tensor1D {\n  return nonMaxSuppressionImpl_(\n             boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n             0 /* softNmsSigma */)\n             .selectedIndices as Tensor1D;\n}\n\nexport function nonMaxSuppressionV4Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    padToMaxOutputSize: boolean): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n      0 /* softNmsSigma */, false /* returnScoresTensor */,\n      padToMaxOutputSize /* padToMaxOutputSize */, true\n      /* returnValidOutputs */);\n}\n\nexport function nonMaxSuppressionV5Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    softNmsSigma: number): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma,\n      true /* returnScoresTensor */);\n}\n\nfunction nonMaxSuppressionImpl_(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma: number,\n    returnScoresTensor = false, padToMaxOutputSize = false,\n    returnValidOutputs = false): NamedTensorMap {\n  // The list is sorted in ascending order, so that we can always pop the\n  // candidate with the largest score in O(1) time.\n  const candidates = [];\n\n  for (let i = 0; i < scores.length; i++) {\n    if (scores[i] > scoreThreshold) {\n      candidates.push({score: scores[i], boxIndex: i, suppressBeginIndex: 0});\n    }\n  }\n\n  candidates.sort(ascendingComparator);\n\n  // If softNmsSigma is 0, the outcome of this algorithm is exactly same as\n  // before.\n  const scale = softNmsSigma > 0 ? (-0.5 / softNmsSigma) : 0.0;\n\n  const selectedIndices: number[] = [];\n  const selectedScores: number[] = [];\n\n  while (selectedIndices.length < maxOutputSize && candidates.length > 0) {\n    const candidate = candidates.pop();\n    const {score: originalScore, boxIndex, suppressBeginIndex} = candidate;\n\n    if (originalScore < scoreThreshold) {\n      break;\n    }\n\n    // Overlapping boxes are likely to have similar scores, therefore we\n    // iterate through the previously selected boxes backwards in order to\n    // see if candidate's score should be suppressed. We use\n    // suppressBeginIndex to track and ensure a candidate can be suppressed\n    // by a selected box no more than once. Also, if the overlap exceeds\n    // iouThreshold, we simply ignore the candidate.\n    let ignoreCandidate = false;\n    for (let j = selectedIndices.length - 1; j >= suppressBeginIndex; --j) {\n      const iou = intersectionOverUnion(boxes, boxIndex, selectedIndices[j]);\n\n      if (iou >= iouThreshold) {\n        ignoreCandidate = true;\n        break;\n      }\n\n      candidate.score =\n          candidate.score * suppressWeight(iouThreshold, scale, iou);\n\n      if (candidate.score <= scoreThreshold) {\n        break;\n      }\n    }\n\n    // At this point, if `candidate.score` has not dropped below\n    // `scoreThreshold`, then we know that we went through all of the\n    // previous selections and can safely update `suppressBeginIndex` to the\n    // end of the selected array. Then we can re-insert the candidate with\n    // the updated score and suppressBeginIndex back in the candidate list.\n    // If on the other hand, `candidate.score` has dropped below the score\n    // threshold, we will not add it back to the candidates list.\n    candidate.suppressBeginIndex = selectedIndices.length;\n\n    if (!ignoreCandidate) {\n      // Candidate has passed all the tests, and is not suppressed, so\n      // select the candidate.\n      if (candidate.score === originalScore) {\n        selectedIndices.push(boxIndex);\n        selectedScores.push(candidate.score);\n      } else if (candidate.score > scoreThreshold) {\n        // Candidate's score is suppressed but is still high enough to be\n        // considered, so add back to the candidates list.\n        binaryInsert(candidates, candidate, ascendingComparator);\n      }\n    }\n  }\n\n  // NonMaxSuppressionV4 feature: padding output to maxOutputSize.\n  const validOutputs = selectedIndices.length;\n  const elemsToPad = maxOutputSize - validOutputs;\n\n  if (padToMaxOutputSize && elemsToPad > 0) {\n    selectedIndices.push(...new Array(elemsToPad).fill(0));\n    selectedScores.push(...new Array(elemsToPad).fill(0.0));\n  }\n\n  const result:\n      NamedTensorMap = {selectedIndices: tensor1d(selectedIndices, 'int32')};\n\n  if (returnScoresTensor) {\n    result['selectedScores'] = tensor1d(selectedScores, 'float32');\n  }\n\n  if (returnValidOutputs) {\n    result['validOutputs'] = scalar(validOutputs, 'int32');\n  }\n\n  return result;\n}\n\nfunction intersectionOverUnion(boxes: TypedArray, i: number, j: number) {\n  const iCoord = boxes.subarray(i * 4, i * 4 + 4);\n  const jCoord = boxes.subarray(j * 4, j * 4 + 4);\n  const yminI = Math.min(iCoord[0], iCoord[2]);\n  const xminI = Math.min(iCoord[1], iCoord[3]);\n  const ymaxI = Math.max(iCoord[0], iCoord[2]);\n  const xmaxI = Math.max(iCoord[1], iCoord[3]);\n  const yminJ = Math.min(jCoord[0], jCoord[2]);\n  const xminJ = Math.min(jCoord[1], jCoord[3]);\n  const ymaxJ = Math.max(jCoord[0], jCoord[2]);\n  const xmaxJ = Math.max(jCoord[1], jCoord[3]);\n  const areaI = (ymaxI - yminI) * (xmaxI - xminI);\n  const areaJ = (ymaxJ - yminJ) * (xmaxJ - xminJ);\n  if (areaI <= 0 || areaJ <= 0) {\n    return 0.0;\n  }\n  const intersectionYmin = Math.max(yminI, yminJ);\n  const intersectionXmin = Math.max(xminI, xminJ);\n  const intersectionYmax = Math.min(ymaxI, ymaxJ);\n  const intersectionXmax = Math.min(xmaxI, xmaxJ);\n  const intersectionArea = Math.max(intersectionYmax - intersectionYmin, 0.0) *\n      Math.max(intersectionXmax - intersectionXmin, 0.0);\n  return intersectionArea / (areaI + areaJ - intersectionArea);\n}\n\n// A Gaussian penalty function, this method always returns values in [0, 1].\n// The weight is a function of similarity, the more overlap two boxes are, the\n// smaller the weight is, meaning highly overlapping boxe will be significantly\n// penalized. On the other hand, a non-overlapping box will not be penalized.\nfunction suppressWeight(iouThreshold: number, scale: number, iou: number) {\n  const weight = Math.exp(scale * iou * iou);\n  return iou <= iouThreshold ? weight : 0.0;\n}\n\nfunction ascendingComparator(c1: Candidate, c2: Candidate) {\n  // For objects with same scores, we make the object with the larger index go\n  // first. In an array that pops from the end, this means that the object with\n  // the smaller index will be popped first. This ensures the same output as\n  // the TensorFlow python version.\n  return (c1.score - c2.score) ||\n      ((c1.score === c2.score) && (c2.boxIndex - c1.boxIndex));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {nonMaxSuppressionV3Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This is the async version of `nonMaxSuppression`\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @return A 1D tensor with the selected box indices.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Promise<Tensor1D> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV3Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionAsync = nonMaxSuppressionAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionWithScore_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const inputs: NonMaxSuppressionV5Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV5Attrs =\n      {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV5, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], selectedScores: result[1]};\n}\n\nexport const nonMaxSuppressionWithScore = op({nonMaxSuppressionWithScore_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV5Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionWithScoreAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV5Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionWithScoreAsync = nonMaxSuppressionWithScoreAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionPadded_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const inputs: NonMaxSuppressionV4Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV4Attrs = {\n    maxOutputSize: $maxOutputSize,\n    iouThreshold: $iouThreshold,\n    scoreThreshold: $scoreThreshold,\n    padToMaxOutputSize\n  };\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV4, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], validOutputs: result[1]};\n}\n\nexport const nonMaxSuppressionPadded = op({nonMaxSuppressionPadded_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV4Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionPaddedAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const [boxesVals, scoresVals] =\n      await Promise.all([$boxes.data(), $scores.data()]);\n\n  // We call a cpu based impl directly with the typedarray data here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV4Impl(\n      boxesVals, scoresVals, $maxOutputSize, $iouThreshold, $scoreThreshold,\n      padToMaxOutputSize);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionPaddedAsync = nonMaxSuppressionPaddedAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Bilinear resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeBilinear_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeBilinear');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeBilinear: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () => `Error in resizeBilinear: new shape must 2D, but got shape ` +\n          `${size}.`);\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n\n  const [newHeight, newWidth] = size;\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeBilinear(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const inputs: ResizeBilinearInputs = {images: batchImages};\n  const attrs: ResizeBilinearAttrs = {alignCorners, size};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeBilinear, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeBilinear = op({resizeBilinear_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * NearestNeighbor resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeNearestNeighbor_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeNearestNeighbor');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeNearestNeighbor: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () =>\n          `Error in resizeNearestNeighbor: new shape must 2D, but got shape ` +\n          `${size}.`);\n  util.assert(\n      $images.dtype === 'float32' || $images.dtype === 'int32',\n      () => '`images` must have `int32` or `float32` as dtype');\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n  const [newHeight, newWidth] = size;\n\n  const inputs: ResizeNearestNeighborInputs = {images: batchImages};\n  const attrs: ResizeNearestNeighborAttrs = {alignCorners, size};\n\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeNearestNeighbor(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeNearestNeighbor, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeNearestNeighbor = op({resizeNearestNeighbor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assert} from '../../util';\n\nimport {greaterEqual} from '../greater_equal';\nimport {lessEqual} from '../less_equal';\nimport {logicalAnd} from '../logical_and';\nimport {op} from '../operation';\nimport {range} from '../range';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\nimport {zeros} from '../zeros';\n\n/**\n * Copy a tensor setting everything outside a central band in each innermost\n * matrix to zero.\n *\n * The band part is computed as follows: Assume input has `k` dimensions\n * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where\n * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`.\n * The indicator function\n * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))`\n * `&& (num_upper < 0 || (n-m) <= num_upper)`\n *\n * ```js\n * const x = tf.tensor2d([[ 0,  1,  2, 3],\n *                        [-1,  0,  1, 2],\n *                        [-2, -1,  0, 1],\n *                        [-3, -2, -1, 0]]);\n * let y = tf.linalg.bandPart(x, 1, -1);\n * y.print(); // [[ 0,  1,  2, 3],\n *            //  [-1,  0,  1, 2],\n *            //  [ 0, -1,  0, 1],\n *            //  [ 0, 0 , -1, 0]]\n * let z = tf.linalg.bandPart(x, 2, 1);\n * z.print(); // [[ 0,  1,  0, 0],\n *            //  [-1,  0,  1, 0],\n *            //  [-2, -1,  0, 1],\n *            //  [ 0, -2, -1, 0]]\n * ```\n *\n * @param x Rank `k` tensor\n * @param numLower Number of subdiagonals to keep.\n *   If negative, keep entire lower triangle.\n * @param numUpper Number of subdiagonals to keep.\n *   If negative, keep entire upper triangle.\n * @returns Rank `k` tensor of the same shape as input.\n *   The extracted banded tensor.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction bandPart_<T extends Tensor>(\n    a: T|TensorLike, numLower: number, numUpper: number): T {\n  assert(\n      numLower % 1 === 0,\n      () => `bandPart(): numLower must be an integer, got ${numLower}.`);\n  assert(\n      numUpper % 1 === 0,\n      () => `bandPart(): numUpper must be an integer, got ${numUpper}.`);\n\n  const $a = convertToTensor(a, 'a', 'bandPart');\n\n  assert(\n      $a.rank >= 2,\n      () => `bandPart(): Rank must be at least 2, got ${$a.rank}.`);\n\n  const shape = $a.shape;\n  const [M, N] = $a.shape.slice(-2);\n\n  if (!(numLower <= M)) {\n    throw new Error(\n        `bandPart(): numLower (${numLower})` +\n        ` must not be greater than the number of rows (${M}).`);\n  }\n  if (!(numUpper <= N)) {\n    throw new Error(\n        `bandPart(): numUpper (${numUpper})` +\n        ` must not be greater than the number of columns (${N}).`);\n  }\n\n  if (numLower < 0) {\n    numLower = M;\n  }\n  if (numUpper < 0) {\n    numUpper = N;\n  }\n\n  const i = reshape(range(0, M, 1, 'int32'), [-1, 1]);\n  const j = range(0, N, 1, 'int32');\n  const ij = sub(i, j);\n\n  const inBand = logicalAnd(\n      lessEqual(ij, scalar(+numLower, 'int32')),\n      greaterEqual(ij, scalar(-numUpper, 'int32')));\n\n  const zero = zeros([M, N], $a.dtype);\n\n  return reshape(\n             stack(unstack(reshape($a, [-1, M, N]))\n                       .map(mat => where(inBand, mat, zero))),\n             shape) as T;\n}\n\nexport const bandPart = op({bandPart_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {div} from '../div';\nimport {mul} from '../mul';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {split} from '../split';\nimport {squeeze} from '../squeeze';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\n/**\n * Gram-Schmidt orthogonalization.\n *\n * ```js\n * const x = tf.tensor2d([[1, 2], [3, 4]]);\n * let y = tf.linalg.gramSchmidt(x);\n * y.print();\n * console.log('Othogonalized:');\n * y.dot(y.transpose()).print();  // should be nearly the identity matrix.\n * console.log('First row direction maintained:');\n * const data = await y.array();\n * console.log(data[0][1] / data[0][0]);  // should be nearly 2.\n * ```\n *\n * @param xs The vectors to be orthogonalized, in one of the two following\n *   formats:\n *   - An Array of `tf.Tensor1D`.\n *   - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows\n *     of `xs`.\n *   In each case, all the vectors must have the same length and the length\n *   must be greater than or equal to the number of vectors.\n * @returns The orthogonalized and normalized vectors or matrix.\n *   Orthogonalization means that the vectors or the rows of the matrix\n *   are orthogonal (zero inner products). Normalization means that each\n *   vector or each row of the matrix has an L2 norm that equals `1`.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D {\n  let inputIsTensor2D: boolean;\n  if (Array.isArray(xs)) {\n    inputIsTensor2D = false;\n    assert(\n        xs != null && xs.length > 0,\n        () => 'Gram-Schmidt process: input must not be null, undefined, or ' +\n            'empty');\n    const dim = xs[0].shape[0];\n    for (let i = 1; i < xs.length; ++i) {\n      assert(\n          xs[i].shape[0] === dim,\n          () =>\n              'Gram-Schmidt: Non-unique lengths found in the input vectors: ' +\n              `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`);\n    }\n  } else {\n    inputIsTensor2D = true;\n    xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0]));\n  }\n\n  assert(\n      xs.length <= xs[0].shape[0],\n      () => `Gram-Schmidt: Number of vectors (${\n                (xs as Tensor1D[]).length}) exceeds ` +\n          `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`);\n\n  const ys: Tensor1D[] = [];\n  const xs1d = xs;\n  for (let i = 0; i < xs.length; ++i) {\n    ys.push(ENGINE.tidy(() => {\n      let x = xs1d[i];\n      if (i > 0) {\n        for (let j = 0; j < i; ++j) {\n          const proj = mul(sum(mul(ys[j], x)), ys[j]);\n          x = sub(x, proj);\n        }\n      }\n      return div(x, norm(x, 'euclidean'));\n    }));\n  }\n\n  if (inputIsTensor2D) {\n    return stack(ys, 0) as Tensor2D;\n  } else {\n    return ys;\n  }\n}\n\nexport const gramSchmidt = op({gramSchmidt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../../engine';\nimport {dispose} from '../../globals';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {clone} from '../clone';\nimport {concat} from '../concat';\nimport {div} from '../div';\nimport {eye} from '../eye';\nimport {greater} from '../greater';\nimport {matMul} from '../mat_mul';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {tensor2d} from '../tensor2d';\nimport {transpose} from '../transpose';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\n\n/**\n * Compute QR decomposition of m-by-n matrix using Householder transformation.\n *\n * Implementation based on\n *   [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf]\n * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf)\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [3, 4]]);\n * let [q, r] = tf.linalg.qr(a);\n * console.log('Q');\n * q.print();\n * console.log('R');\n * r.print();\n * console.log('Orthogonalized');\n * q.dot(q.transpose()).print()  // should be nearly the identity matrix.\n * console.log('Reconstructed');\n * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]];\n * ```\n *\n * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose\n *   it has the shape `[..., M, N]`.\n * @param fullMatrices An optional boolean parameter. Defaults to `false`.\n *   If `true`, compute full-sized `Q`. If `false` (the default),\n *   compute only the leading N columns of `Q` and `R`.\n * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix,\n *   i.e., its columns all have unit norm and are mutually orthogonal.\n *   If `M >= N`,\n *     If `fullMatrices` is `false` (default),\n *       - `Q` has a shape of `[..., M, N]`,\n *       - `R` has a shape of `[..., N, N]`.\n *     If `fullMatrices` is `true` (default),\n *       - `Q` has a shape of `[..., M, M]`,\n *       - `R` has a shape of `[..., M, N]`.\n *   If `M < N`,\n *     - `Q` has a shape of `[..., M, M]`,\n *     - `R` has a shape of `[..., M, N]`.\n * @throws If the rank of `x` is less than 2.\n *\n * @doc {heading:'Operations',\n *       subheading:'Linear Algebra',\n *       namespace:'linalg'}\n */\nfunction qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] {\n  assert(\n      x.rank >= 2,\n      () => `qr() requires input tensor to have a rank >= 2, but got rank ${\n          x.rank}`);\n\n  if (x.rank === 2) {\n    return qr2d(x as Tensor2D, fullMatrices);\n  } else {\n    // Rank > 2.\n    // TODO(cais): Below we split the input into individual 2D tensors,\n    //   perform QR decomposition on them and then stack the results back\n    //   together. We should explore whether this can be parallelized.\n    const outerDimsProd = x.shape.slice(0, x.shape.length - 2)\n                              .reduce((value, prev) => value * prev);\n    const x2ds = unstack(\n        reshape(\n            x,\n            [\n              outerDimsProd, x.shape[x.shape.length - 2],\n              x.shape[x.shape.length - 1]\n            ]),\n        0);\n    const q2ds: Tensor2D[] = [];\n    const r2ds: Tensor2D[] = [];\n    x2ds.forEach(x2d => {\n      const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices);\n      q2ds.push(q2d);\n      r2ds.push(r2d);\n    });\n    const q = reshape(stack(q2ds, 0), x.shape);\n    const r = reshape(stack(r2ds, 0), x.shape);\n    return [q, r];\n  }\n}\n\nfunction qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] {\n  return ENGINE.tidy(() => {\n    assert(\n        x.shape.length === 2,\n        () => `qr2d() requires a 2D Tensor, but got a ${\n            x.shape.length}D Tensor.`);\n\n    const m = x.shape[0];\n    const n = x.shape[1];\n\n    let q = eye(m);    // Orthogonal transform so far.\n    let r = clone(x);  // Transformed matrix so far.\n\n    const one2D = tensor2d([[1]], [1, 1]);\n    let w: Tensor2D = clone(one2D);\n\n    const iters = m >= n ? n : m;\n    for (let j = 0; j < iters; ++j) {\n      // This tidy within the for-loop ensures we clean up temporary\n      // tensors as soon as they are no longer needed.\n      const rTemp = r;\n      const wTemp = w;\n      const qTemp = q;\n      [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => {\n        // Find H = I - tau * w * w', to put zeros below R(j, j).\n        const rjEnd1 = slice(r, [j, j], [m - j, 1]);\n        const normX = norm(rjEnd1);\n        const rjj = slice(r, [j, j], [1, 1]);\n\n        // The sign() function returns 0 on 0, which causes division by zero.\n        const s = where(greater(rjj, 0), tensor2d([[-1]]), tensor2d([[1]]));\n\n        const u1 = sub(rjj, mul(s, normX));\n        const wPre = div(rjEnd1, u1);\n        if (wPre.shape[0] === 1) {\n          w = clone(one2D);\n        } else {\n          w = concat(\n              [\n                one2D,\n                slice(wPre, [1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as\n                    Tensor2D\n              ],\n              0);\n        }\n        const tau = neg(div(matMul(s, u1), normX)) as Tensor2D;\n\n        // -- R := HR, Q := QH.\n        const rjEndAll = slice(r, [j, 0], [m - j, n]);\n        const tauTimesW: Tensor2D = mul(tau, w);\n        const wT: Tensor2D = transpose(w);\n        if (j === 0) {\n          r = sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n        } else {\n          const rTimesTau: Tensor2D =\n              sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n          r = concat([slice(r, [0, 0], [j, n]), rTimesTau], 0);\n        }\n        const tawTimesWT: Tensor2D = transpose(tauTimesW);\n        const qAllJEnd = slice(q, [0, j], [m, q.shape[1] - j]);\n        if (j === 0) {\n          q = sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n        } else {\n          const qTimesTau: Tensor2D =\n              sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n          q = concat([slice(q, [0, 0], [m, j]), qTimesTau], 1);\n        }\n        return [w, r, q];\n      });\n      dispose([rTemp, wTemp, qTemp]);\n    }\n\n    if (!fullMatrices && m > n) {\n      q = slice(q, [0, 0], [m, n]);\n      r = slice(r, [0, 0], [n, n]);\n    }\n\n    return [q, r];\n  }) as [Tensor2D, Tensor2D];\n}\n\nexport const qr = op({qr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport enum Reduction {\n  NONE,\n  MEAN,\n  SUM,\n  SUM_BY_NONZERO_WEIGHTS\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {Reduction} from '../loss_ops_utils';\nimport {mean} from '../mean';\nimport {mul} from '../mul';\nimport {notEqual} from '../not_equal';\nimport {ones} from '../ones';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sum} from '../sum';\n\n/**\n * Computes the weighted loss between two tensors.\n *\n * @param losses Tensor of shape `[batch_size, d1, ... dN]`.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `losses`, and must be broadcastable to `losses` (i.e., all\n *    dimensions must be either `1`, or the same as the corresponding\n *    `losses` dimension).\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction computeWeightedLoss_<T extends Tensor, O extends Tensor>(\n    losses: T|TensorLike, weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $losses = convertToTensor(losses, 'losses', 'computeWeightedLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'computeWeightedLoss');\n  }\n\n  const weightedLoss = ($weights == null) ? $losses : mul($losses, $weights);\n\n  if (reduction === Reduction.NONE) {\n    return weightedLoss as O;\n  }\n  if (reduction === Reduction.SUM) {\n    return sum(weightedLoss);\n  }\n  if (reduction === Reduction.MEAN) {\n    if ($weights == null) {\n      return mean(weightedLoss);\n    } else {\n      const broadcastFactor = $losses.size / $weights.size;\n      const result = div(sum(weightedLoss), sum($weights));\n      return broadcastFactor > 1 ? div(result, scalar(broadcastFactor)) :\n                                   result as O;\n    }\n  }\n  if (reduction === Reduction.SUM_BY_NONZERO_WEIGHTS) {\n    if ($weights == null) {\n      return div(sum(weightedLoss), scalar($losses.size));\n    } else {\n      const broadcastedWeights = mul($weights, ones($losses.shape));\n\n      const numNonZeros =\n          cast(sum(notEqual(broadcastedWeights, scalar(0))), 'float32');\n      return div(sum(weightedLoss), numNonZeros);\n    }\n  }\n\n  throw Error(`Unknown reduction: ${reduction}`);\n}\nexport const computeWeightedLoss = op({computeWeightedLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the absolute difference loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction absoluteDifference_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'absoluteDifference');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'absoluteDifference');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'absoluteDifference');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in absoluteDifference: ');\n\n  const losses = abs(sub($labels, $predictions));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const absoluteDifference = op({absoluteDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the cosine distance loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param axis The dimension along which the cosine distance is computed.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction cosineDistance_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike, axis: number,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'cosineDistance');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'cosineDistance');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'cosineDistance');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in cosineDistance: ');\n\n  const one = scalar(1);\n  const losses = sub(one, sum(mul($labels, $predictions), axis, true));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const cosineDistance = op({cosineDistance_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the Hinge loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction hingeLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $labels = convertToTensor(labels, 'labels', 'hingeLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'hingeLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'hingeLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in hingeLoss: ');\n\n  const one = scalar(1);\n  // Convert binary labels to (-1, 1)\n  $labels = sub(mul(scalar(2), $labels), one);\n  const losses = relu(sub(one, mul($labels, $predictions)));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const hingeLoss = op({hingeLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {Reduction} from '../loss_ops_utils';\nimport {minimum} from '../minimum';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {square} from '../square';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the huber loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param delta Point where huber loss changes from quadratic to linear.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`.\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction huberLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, delta = 1.0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'huberLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'huberLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'huberLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in huberLoss: ');\n\n  const deltaScalar = scalar(delta);\n  const error = abs(sub($predictions, $labels));\n  const quadratic = minimum(error, deltaScalar);\n  const linear = sub(error, quadratic);\n\n  const losses =\n      add(mul(scalar(0.5), square(quadratic)), mul(deltaScalar, linear));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const huberLoss = op({huberLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {log} from '../log';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the log loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param epsilon A small increment to avoid taking log of zero\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction logLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, epsilon = 1e-7,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'logLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'logLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'logLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in logLoss: ');\n\n  const one = scalar(1);\n  const epsilonScalar = scalar(epsilon);\n\n  const l1 = neg(mul($labels, log(add($predictions, epsilonScalar))));\n  const l2 =\n      mul(sub(one, $labels), log(add(sub(one, $predictions), epsilonScalar)));\n  const losses = sub(l1, l2);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const logLoss = op({logLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {squaredDifference} from '../squared_difference';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the mean squared error between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction meanSquaredError_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'meanSquaredError');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'meanSquaredError');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'meanSquaredError');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in meanSquaredError: ');\n\n  const losses = squaredDifference($labels, $predictions);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const meanSquaredError = op({meanSquaredError_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {exp} from '../exp';\nimport {log1p} from '../log1p';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\nfunction sigmoidCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, logits: T|TensorLike): O {\n  const $labels =\n      convertToTensor(labels, 'labels', 'sigmoidCrossEntropyWithLogits');\n  const $logits =\n      convertToTensor(logits, 'logits', 'sigmoidCrossEntropyWithLogits');\n  assertShapesMatch(\n      $labels.shape, $logits.shape, 'Error in sigmoidCrossEntropyWithLogits: ');\n\n  /**\n   * Implementation Details:\n   *\n   * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n   *     z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n   *   = (1 - z) * x + log(1 + exp(-x))\n   *   = x - x * z + log(1 + exp(-x))\n   *\n   *   For x < 0, to avoid overflow in exp(-x), we reformulate the above\n   *     x - x * z + log(1 + exp(-x))\n   *   = log(exp(x)) - x * z + log(1 + exp(-x))\n   *   = - x * z + log(1 + exp(x))\n   *\n   * Hence, to ensure stability and avoid overflow, the implementation uses\n   * this equivalent formulation:\n   *     max(x, 0) - x * z + log(1 + exp(-abs(x)))\n   */\n  const maxOutput = relu($logits);\n  const outputXTarget = mul($logits, $labels);\n  const sigmoidOutput = log1p(exp(neg(abs($logits))));\n\n  return add(sub(maxOutput, outputXTarget), sigmoidOutput);\n}\n\n/**\n * Computes the sigmoid cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newMulticlassLabels = multiclassLabels * (1 - labelSmoothing)\n *                         + 0.5 * labelSmoothing\n *\n * @param multiClassLabels The ground truth output tensor of shape\n * [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction sigmoidCrossEntropy_<T extends Tensor, O extends Tensor>(\n    multiClassLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $multiClassLabels = convertToTensor(\n      multiClassLabels, 'multiClassLabels', 'sigmoidCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'sigmoidCrossEntropy');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'sigmoidCrossEntropy');\n  }\n  assertShapesMatch(\n      $multiClassLabels.shape, $logits.shape, 'Error in sigmoidCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const half = scalar(0.5);\n\n    $multiClassLabels =\n        add(mul($multiClassLabels, sub(one, labelSmoothingScalar)),\n            mul(half, labelSmoothingScalar));\n  }\n  const losses = sigmoidCrossEntropyWithLogits_($multiClassLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const sigmoidCrossEntropy = op({sigmoidCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../../gradients';\nimport {Tensor} from '../../tensor';\nimport {GradSaveFunc} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {expandShapeToKeepDim} from '../axis_util';\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {exp} from '../exp';\nimport {logSumExp} from '../log_sum_exp';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes softmax cross entropy between logits and labels.\n *\n * Measures the probability error in discrete classification tasks in which\n * the classes are mutually exclusive (each entry is in exactly one class).\n * For example, each CIFAR-10 image is labeled with one and only one label: an\n * image can be a dog or a truck, but not both.\n *\n * `NOTE`: While the classes are mutually exclusive, their probabilities need\n * not be. All that is required is that each row of labels is a valid\n * probability distribution. If they are not, the computation of the gradient\n * will be incorrect.\n *\n * `WARNING`: This op expects unscaled logits, since it performs a softmax on\n * logits internally for efficiency. Do not call this op with the output of\n * softmax, as it will produce incorrect results.\n *\n * logits and labels must have the same shape, e.g. [batch_size, num_classes]\n * and the same dtype.\n * @param labels The labels array.\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n */\nfunction softmaxCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T, logits: T, dim = -1): O {\n  if (dim === -1) {\n    dim = logits.rank - 1;\n  }\n\n  if (dim !== logits.rank - 1) {\n    throw Error(\n        `Softmax cross entropy along a non-last dimension is not yet ` +\n        `supported. Labels / logits was rank ${logits.rank} ` +\n        `and dim was ${dim}`);\n  }\n  // Use a custom gradient for numerical stability.\n  const customOp =\n      customGrad((labels: Tensor, logits: Tensor, save: GradSaveFunc) => {\n        // Reference:\n        //   1. http://cs231n.github.io/linear-classify/#softmax\n        //   2. https://blog.feedly.com/tricks-of-the-trade-logsumexp/\n        const keepDims = true;\n        const lse = logSumExp(logits, [dim], keepDims);\n        const logResult = sub(cast(logits, 'float32'), lse);\n        save([labels, logResult]);\n\n        const costVector = neg(mul(logResult, labels));\n        const value: O = sum(costVector, [dim]);\n\n        const gradFunc = (dy: O, saved: Tensor[]) => {\n          const [labels, logResult] = saved;\n          const dyShape = expandShapeToKeepDim(dy.shape, [dim]);\n          return [\n            mul(reshape(dy, dyShape),\n                sub(cast(labels, 'float32'), exp(logResult))),\n            mul(reshape(dy, dyShape),\n                sub(exp(logResult), cast(labels, 'float32'))),\n          ];\n        };\n        return {value, gradFunc};\n      });\n\n  return customOp(labels, logits);\n}\n\n/**\n * Computes the softmax cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newOnehotLabels = onehotLabels * (1 - labelSmoothing)\n *                         + labelSmoothing / numClasses\n *\n * @param onehotLabels One hot encoded labels\n *    [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or 1, and must be\n *    broadcastable to `loss`  of shape [batch_size]\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction softmaxCrossEntropy_<T extends Tensor, O extends Tensor>(\n    onehotLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $onehotLabels =\n      convertToTensor(onehotLabels, 'onehotLabels', 'softmaxCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'softmaxCrossEntropy');\n  let $weights: Tensor = null;\n\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'softmaxCrossEntropy');\n  }\n\n  assertShapesMatch(\n      $onehotLabels.shape, $logits.shape, 'Error in softmaxCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const numClasses = scalar($onehotLabels.shape[1]);\n\n    $onehotLabels =\n        add(mul($onehotLabels, sub(one, labelSmoothingScalar)),\n            div(labelSmoothingScalar, numClasses));\n  }\n\n  const losses = softmaxCrossEntropyWithLogits_($onehotLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const softmaxCrossEntropy = op({softmaxCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Modularized ops.\nexport {abs} from './abs';\nexport {acos} from './acos';\nexport {acosh} from './acosh';\nexport {add} from './add';\nexport {addN} from './add_n';\nexport {all} from './all';\nexport {any} from './any';\nexport {argMax} from './arg_max';\nexport {argMin} from './arg_min';\nexport {asin} from './asin';\nexport {asinh} from './asinh';\nexport {atan} from './atan';\nexport {atan2} from './atan2';\nexport {atanh} from './atanh';\nexport {avgPool} from './avg_pool';\nexport {avgPool3d} from './avg_pool_3d';\nexport {basicLSTMCell} from './basic_lstm_cell';\nexport {batchToSpaceND} from './batch_to_space_nd';\nexport {batchNorm} from './batchnorm';\nexport {batchNorm2d} from './batchnorm2d';\nexport {batchNorm3d} from './batchnorm3d';\nexport {batchNorm4d} from './batchnorm4d';\nexport {broadcastTo} from './broadcast_to';\nexport {buffer} from './buffer';\nexport {cast} from './cast';\nexport {ceil} from './ceil';\nexport {clipByValue} from './clip_by_value';\nexport {clone} from './clone';\nexport {complex} from './complex';\nexport {concat} from './concat';\nexport {concat1d} from './concat_1d';\nexport {concat2d} from './concat_2d';\nexport {concat3d} from './concat_3d';\nexport {concat4d} from './concat_4d';\nexport {conv1d} from './conv1d';\nexport {conv2d} from './conv2d';\nexport {conv2dTranspose} from './conv2d_transpose';\nexport {conv3d} from './conv3d';\nexport {conv3dTranspose} from './conv3d_transpose';\nexport {cos} from './cos';\nexport {cosh} from './cosh';\nexport {cumsum} from './cumsum';\nexport {depthToSpace} from './depth_to_space';\nexport {depthwiseConv2d} from './depthwise_conv2d';\nexport {diag} from './diag';\nexport {dilation2d} from './dilation2d';\nexport {div} from './div';\nexport {divNoNan} from './div_no_nan';\nexport {dot} from './dot';\nexport {elu} from './elu';\nexport {equal} from './equal';\nexport {erf} from './erf';\nexport {exp} from './exp';\nexport {expandDims} from './expand_dims';\nexport {expm1} from './expm1';\nexport {eye} from './eye';\nexport {fill} from './fill';\nexport {floor} from './floor';\nexport {floorDiv} from './floorDiv';\nexport {gather} from './gather';\nexport {greater} from './greater';\nexport {greaterEqual} from './greater_equal';\nexport {imag} from './imag';\nexport {isFinite} from './is_finite';\nexport {isInf} from './is_inf';\nexport {isNaN} from './is_nan';\nexport {leakyRelu} from './leaky_relu';\nexport {less} from './less';\nexport {lessEqual} from './less_equal';\nexport {linspace} from './linspace';\nexport {localResponseNormalization} from './local_response_normalization';\nexport {log} from './log';\nexport {log1p} from './log1p';\nexport {logSigmoid} from './log_sigmoid';\nexport {logSoftmax} from './log_softmax';\nexport {logSumExp} from './log_sum_exp';\nexport {logicalAnd} from './logical_and';\nexport {logicalNot} from './logical_not';\nexport {logicalOr} from './logical_or';\nexport {logicalXor} from './logical_xor';\nexport {matMul} from './mat_mul';\nexport {max} from './max';\nexport {maxPool} from './max_pool';\nexport {maxPool3d} from './max_pool_3d';\nexport {maxPoolWithArgmax} from './max_pool_with_argmax';\nexport {maximum} from './maximum';\nexport {mean} from './mean';\nexport {min} from './min';\nexport {minimum} from './minimum';\nexport {mod} from './mod';\nexport {moments} from './moments';\nexport {mul} from './mul';\nexport {LSTMCellFunc, multiRNNCell} from './multi_rnn_cell';\nexport {multinomial} from './multinomial';\nexport {neg} from './neg';\nexport {notEqual} from './not_equal';\nexport {oneHot} from './one_hot';\nexport {ones} from './ones';\nexport {onesLike} from './ones_like';\nexport {outerProduct} from './outer_product';\nexport {pad} from './pad';\nexport {pad1d} from './pad1d';\nexport {pad2d} from './pad2d';\nexport {pad3d} from './pad3d';\nexport {pad4d} from './pad4d';\nexport {pool} from './pool';\nexport {pow} from './pow';\nexport {prelu} from './prelu';\nexport {print} from './print';\nexport {prod} from './prod';\nexport {rand} from './rand';\nexport {randomGamma} from './random_gamma';\nexport {randomNormal} from './random_normal';\nexport {randomUniform} from './random_uniform';\nexport {range} from './range';\nexport {real} from './real';\nexport {reciprocal} from './reciprocal';\nexport {relu} from './relu';\nexport {relu6} from './relu6';\nexport {reshape} from './reshape';\nexport {reverse} from './reverse';\nexport {reverse1d} from './reverse_1d';\nexport {reverse2d} from './reverse_2d';\nexport {reverse3d} from './reverse_3d';\nexport {reverse4d} from './reverse_4d';\nexport {round} from './round';\nexport {rsqrt} from './rsqrt';\nexport {scalar} from './scalar';\nexport {selu} from './selu';\nexport {separableConv2d} from './separable_conv2d';\nexport {setdiff1dAsync} from './setdiff1d_async';\nexport {sigmoid} from './sigmoid';\nexport {sign} from './sign';\nexport {sin} from './sin';\nexport {sinh} from './sinh';\nexport {slice} from './slice';\nexport {slice1d} from './slice1d';\nexport {slice2d} from './slice2d';\nexport {slice3d} from './slice3d';\nexport {slice4d} from './slice4d';\nexport {softmax} from './softmax';\nexport {softplus} from './softplus';\nexport {spaceToBatchND} from './space_to_batch_nd';\nexport {fft} from './spectral/fft';\nexport {ifft} from './spectral/ifft';\nexport {irfft} from './spectral/irfft';\nexport {rfft} from './spectral/rfft';\nexport {split} from './split';\nexport {sqrt} from './sqrt';\nexport {square} from './square';\nexport {squaredDifference} from './squared_difference';\nexport {squeeze} from './squeeze';\nexport {stack} from './stack';\nexport {step} from './step';\nexport {stridedSlice} from './strided_slice';\nexport {sub} from './sub';\nexport {sum} from './sum';\nexport {tan} from './tan';\nexport {tanh} from './tanh';\nexport {tensor} from './tensor';\nexport {tensor1d} from './tensor1d';\nexport {tensor2d} from './tensor2d';\nexport {tensor3d} from './tensor3d';\nexport {tensor4d} from './tensor4d';\nexport {tensor5d} from './tensor5d';\nexport {tensor6d} from './tensor6d';\nexport {tile} from './tile';\nexport {topk} from './topk';\nexport {truncatedNormal} from './truncated_normal';\nexport {unique} from './unique';\nexport {unsortedSegmentSum} from './unsorted_segment_sum';\nexport {unstack} from './unstack';\nexport {variable} from './variable';\nexport {where} from './where';\nexport {whereAsync} from './where_async';\nexport {zeros} from './zeros';\nexport {zerosLike} from './zeros_like';\n\nexport * from './boolean_mask';\nexport * from './compare';\nexport * from './binary_ops';\nexport * from './transpose';\nexport * from './norm';\nexport * from './moving_average';\nexport * from './scatter_nd';\nexport * from './sparse_to_dense';\nexport * from './gather_nd';\nexport * from './dropout';\nexport * from './signal_ops_util';\nexport * from './in_top_k';\n\nexport {op, OP_SCOPE_SUFFIX} from './operation';\n\nimport {rfft} from './spectral/rfft';\nimport {fft} from './spectral/fft';\nimport {ifft} from './spectral/ifft';\nimport {irfft} from './spectral/irfft';\nconst spectral = {\n  fft,\n  ifft,\n  rfft,\n  irfft\n};\n\nimport * as fused from './fused_ops';\n\nimport {hammingWindow} from './signal/hamming_window';\nimport {hannWindow} from './signal/hann_window';\nimport {frame} from './signal/frame';\nimport {stft} from './signal/stft';\nconst signal = {\n  hammingWindow,\n  hannWindow,\n  frame,\n  stft,\n};\n\n// Image Ops namespace\nimport {cropAndResize} from './image/crop_and_resize';\nimport {flipLeftRight} from './image/flip_left_right';\nimport {rotateWithOffset} from './image/rotate_with_offset';\nimport {nonMaxSuppression} from './image/non_max_suppression';\nimport {nonMaxSuppressionAsync} from './image/non_max_suppression_async';\nimport {nonMaxSuppressionWithScore} from './image/non_max_suppression_with_score';\nimport {nonMaxSuppressionWithScoreAsync} from './image/non_max_suppression_with_score_async';\nimport {nonMaxSuppressionPadded} from './image/non_max_suppression_padded';\nimport {nonMaxSuppressionPaddedAsync} from './image/non_max_suppression_padded_async';\nimport {resizeBilinear} from './image/resize_bilinear';\nimport {resizeNearestNeighbor} from './image/resize_nearest_neighbor';\nconst image = {\n  flipLeftRight,\n  resizeNearestNeighbor,\n  resizeBilinear,\n  rotateWithOffset,\n  cropAndResize,\n  nonMaxSuppression,\n  nonMaxSuppressionAsync,\n  nonMaxSuppressionWithScore,\n  nonMaxSuppressionWithScoreAsync,\n  nonMaxSuppressionPadded,\n  nonMaxSuppressionPaddedAsync\n};\n\n// linalg namespace\nimport {bandPart} from './linalg/band_part';\nimport {gramSchmidt} from './linalg/gram_schmidt';\nimport {qr} from './linalg/qr';\nconst linalg = {\n  bandPart,\n  gramSchmidt,\n  qr\n};\n\n// losses namespace;\nimport {absoluteDifference} from './losses/absolute_difference';\nimport {computeWeightedLoss} from './losses/compute_weighted_loss';\nimport {cosineDistance} from './losses/cosine_distance';\nimport {hingeLoss} from './losses/hinge_loss';\nimport {huberLoss} from './losses/huber_loss';\nimport {logLoss} from './losses/log_loss';\nimport {meanSquaredError} from './losses/mean_squared_error';\nimport {sigmoidCrossEntropy} from './losses/sigmoid_cross_entropy';\nimport {softmaxCrossEntropy} from './losses/softmax_cross_entropy';\nconst losses = {\n  absoluteDifference,\n  computeWeightedLoss,\n  cosineDistance,\n  hingeLoss,\n  huberLoss,\n  logLoss,\n  meanSquaredError,\n  sigmoidCrossEntropy,\n  softmaxCrossEntropy\n};\n\n// Second level exports.\nexport {image, linalg, losses, spectral, fused, signal};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {dispose} from '../globals';\nimport {variableGrads} from '../gradients';\nimport {scalar} from '../ops/ops';\nimport {Serializable} from '../serialization';\nimport {Scalar, Variable} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\n/**\n * A variable that belongs to an optimizer.\n *\n * The `originalName` field is required for keeping track of the canonical\n * name of the variable, which is usually the name of the model weight that\n * the variable is related to plus a suffix, e.g., 'dense1/kernel/momentum'.\n * The name of the `Variable` object itself cannot be used directly due to\n * possible deduplication: Every `Variable` must have a unique name but more\n * than one optimizer objects of the same type may be created for the same model\n * or the same `Variable`.\n */\nexport interface OptimizerVariable {\n  originalName: string;\n  variable: Variable;\n}\n\n/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */\nexport abstract class Optimizer extends Serializable {\n  protected iterations_: number;\n\n  /**\n   * Executes `f()` and minimizes the scalar output of `f()` by computing\n   * gradients of y with respect to the list of trainable variables provided by\n   * `varList`. If no list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to minimize.\n   * @param returnCost Whether to return the scalar cost value produced by\n   * executing `f()`.\n   * @param varList An optional list of variables to update. If specified, only\n   * the trainable variables in varList will be updated by minimize. Defaults to\n   * all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  minimize(f: () => Scalar, returnCost = false, varList?: Variable[]): Scalar\n      |null {\n    const {value, grads} = this.computeGradients(f, varList);\n\n    if (varList != null) {\n      const gradArray: NamedTensor[] =\n          varList.map(v => ({name: v.name, tensor: grads[v.name]}));\n      this.applyGradients(gradArray);\n    } else {\n      this.applyGradients(grads);\n    }\n\n    // Dispose gradients.\n    dispose(grads);\n\n    if (returnCost) {\n      return value;\n    } else {\n      value.dispose();\n      return null;\n    }\n  }\n\n  /**\n   * The number of iterations that this optimizer instance has been invoked for.\n   */\n  get iterations(): number {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return this.iterations_;\n  }\n\n  protected incrementIterations() {\n    this.iterations_ = this.iterations + 1;\n  }\n\n  /**\n   * Executes f() and computes the gradient of the scalar output of f() with\n   * respect to the list of trainable variables provided by `varList`. If no\n   * list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to use for computing\n   * gradients with respect to variables.\n   * @param varList An optional list of variables to compute gradients with\n   * respect to. If specified, only the trainable variables in varList will have\n   * gradients computed with respect to. Defaults to all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  computeGradients(f: () => Scalar, varList?: Variable[]):\n      {value: Scalar, grads: NamedTensorMap} {\n    return variableGrads(f, varList);\n  }\n\n  /**\n   * Updates variables by using the computed gradients.\n   *\n   * @param variableGradients A mapping of variable name to its gradient value.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  abstract applyGradients(variableGradients: NamedTensorMap|\n                          NamedTensor[]): void;\n\n  /**\n   * Dispose the variables (if any) owned by this optimizer instance.\n   */\n  dispose(): void {\n    if (this.iterations_ != null) {\n      dispose(this.iterations_);\n    }\n  }\n\n  async saveIterations(): Promise<NamedTensor> {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return {\n      name: 'iter',  // Named for Python compatibility.\n      // TODO(cais): Use 'int64' type when available.\n      tensor: scalar(this.iterations_, 'int32')\n    };\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for this optimizer yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error(\n        `setWeights() is not implemented for this optimizer class ` +\n        `${this.getClassName()}`);\n  }\n\n  /**\n   * Extract the first element of the weight values and set it\n   * as the iterations counter variable of this instance of optimizer.\n   *\n   * @param weightValues\n   * @returns Weight values with the first element consumed and excluded.\n   */\n  protected async extractIterations(weightValues: NamedTensor[]):\n      Promise<NamedTensor[]> {\n    this.iterations_ = (await weightValues[0].tensor.data())[0];\n    return weightValues.slice(1);\n  }\n}\n\nObject.defineProperty(Optimizer, Symbol.hasInstance, {\n  value: (instance: Optimizer) => {\n    return instance.minimize != null && instance.computeGradients != null &&\n        instance.applyGradients != null;\n  }\n});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/ops';\nimport {square} from '../ops/square';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdadeltaOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adadelta';  // Name matters for Python compatibility.\n  private accumulatedGrads: OptimizerVariable[] = [];\n  private accumulatedUpdates: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected rho: number,\n      protected epsilon: number = null) {\n    super();\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedGrads[i] == null) {\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accum_grad`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedUpdates[i] == null) {\n        this.accumulatedUpdates[i] = {\n          originalName: `${name}/accum_var`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n      const accumulatedUpdate = this.accumulatedUpdates[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad =\n            add(mul(accumulatedGrad, this.rho),\n                mul(square(gradient), 1 - this.rho));\n\n        const updates =\n            mul(div(sqrt(add(accumulatedUpdate, this.epsilon)),\n                    sqrt(add(accumulatedGrad, this.epsilon))),\n                gradient);\n\n        const newAccumulatedUpdate =\n            add(mul(accumulatedUpdate, this.rho),\n                mul(square(updates), 1 - this.rho));\n\n        accumulatedGrad.assign(newAccumulatedGrad);\n        accumulatedUpdate.assign(newAccumulatedUpdate);\n\n        const newValue = add(mul(updates, -this.learningRate), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedUpdates != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n      dispose(this.accumulatedUpdates.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedGrads, ...this.accumulatedUpdates];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedGrads =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedUpdates =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'rho': this.rho,\n      'epsilon': this.epsilon\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['rho'], config['epsilon']);\n  }\n}\nregisterClass(AdadeltaOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {fill} from '../ops/fill';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdagradOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adagrad';  // Note: Name matters for Python compatibility.\n\n  private accumulatedGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private initialAccumulatorValue = 0.1) {\n    super();\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulatedGrads[i] == null) {\n        const trainable = false;\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accumulator`,\n          variable: tidy(\n              () => fill(value.shape, this.initialAccumulatorValue)\n                        .variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad = add(accumulatedGrad, square(gradient));\n        accumulatedGrad.assign(newAccumulatedGrad);\n\n        const newValue = add(\n            mul(div(gradient,\n                    sqrt(add(newAccumulatedGrad, ENGINE.backend.epsilon()))),\n                -this.learningRate),\n            value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedGrads != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulatedGrads.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulatedGrads = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'initialAccumulatorValue': this.initialAccumulatorValue,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['initialAccumulatorValue']);\n  }\n}\nregisterClass(AdagradOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adam';  // Note: Name matters for Python compatibility.\n  private accBeta1: Variable;\n  private accBeta2: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedSecondMoment: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null) {\n    super();\n    tidy(() => {\n      // accB* will be updated by batch.\n      this.accBeta1 = scalar(beta1).variable();\n      this.accBeta2 = scalar(beta2).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const oneMinusAccBeta2 = sub(1, this.accBeta2);\n\n      varNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n        if (this.accumulatedSecondMoment[i] == null) {\n          this.accumulatedSecondMoment[i] = {\n            originalName: `${name}/v`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const secondMoment = this.accumulatedSecondMoment[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n        const newSecondMoment =\n            add(mul(secondMoment, this.beta2),\n                mul(square(gradient), 1 - this.beta2));\n\n        const biasCorrectedFirstMoment = div(newFirstMoment, oneMinusAccBeta1);\n        const biasCorrectedSecondMoment =\n            div(newSecondMoment, oneMinusAccBeta2);\n\n        firstMoment.assign(newFirstMoment);\n        secondMoment.assign(newSecondMoment);\n\n        const newValue =\n            add(mul(div(biasCorrectedFirstMoment,\n                        add(sqrt(biasCorrectedSecondMoment), this.epsilon)),\n                    -this.learningRate),\n                value);\n        value.assign(newValue);\n      });\n\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n      this.accBeta2.assign(mul(this.accBeta2, this.beta2));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.accBeta2.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedSecondMoment != null) {\n      dispose(this.accumulatedSecondMoment.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedFirstMoment, ...this.accumulatedSecondMoment];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    tidy(() => {\n      this.accBeta1.assign(pow(this.beta1, this.iterations_ + 1));\n      this.accBeta2.assign(pow(this.beta2, this.iterations_ + 1));\n    });\n\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedFirstMoment =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedSecondMoment =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon']);\n  }\n}\nregisterClass(AdamOptimizer);\n", "\uFEFF/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {abs} from '../ops/abs';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {maximum} from '../ops/maximum';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamaxOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adamax';  // Note: Name matters for Python compatbility.\n  private accBeta1: Variable;\n  private iteration: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null,\n      protected decay = 0.0) {\n    super();\n\n    tidy(() => {\n      this.iteration = scalar(0).variable();\n      this.accBeta1 = scalar(beta1).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const lr =\n          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));\n\n      variableNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n        if (this.accumulatedWeightedInfNorm[i] == null) {\n          this.accumulatedWeightedInfNorm[i] = {\n            originalName: `${name}/v`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n\n        const ut0 = mul(weightedInfNorm, this.beta2);\n        const ut1 = abs(gradient);\n\n        const newWeightedInfNorm = maximum(ut0, ut1);\n\n        firstMoment.assign(newFirstMoment);\n        weightedInfNorm.assign(newWeightedInfNorm);\n\n        const newValue =\n            add(mul(div(lr, oneMinusAccBeta1),\n                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),\n                value);\n\n        value.assign(newValue);\n      });\n\n      this.iteration.assign(add(this.iteration, 1));\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.iteration.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedWeightedInfNorm != null) {\n      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for Adamax yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error('setWeights() is not implemented for Adamax yet.');\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n      'decay': this.decay\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon'], config['decay']);\n  }\n}\nregisterClass(AdamaxOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {keep, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer} from './optimizer';\n\n/** @doclink Optimizer */\nexport class SGDOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'SGD';  // Note: Name matters for Python compatibility.\n  protected c: Scalar;\n\n  constructor(protected learningRate: number) {\n    super();\n    this.setLearningRate(learningRate);\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    varNames.forEach((name, i) => {\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n      const value = ENGINE.registeredVariables[name];\n      tidy(() => {\n        const newValue = add(mul(this.c, gradient), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  /**\n   * Sets the learning rate of the optimizer.\n   */\n  setLearningRate(learningRate: number) {\n    this.learningRate = learningRate;\n    if (this.c != null) {\n      this.c.dispose();\n    }\n    this.c = keep(scalar(-learningRate));\n  }\n\n  dispose() {\n    this.c.dispose();\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    return [await this.saveIterations()];\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    if (weightValues.length !== 0) {\n      throw new Error('SGD optimizer does not have settable weights.');\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {'learningRate': this.learningRate};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate']);\n  }\n}\nregisterClass(SGDOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {OptimizerVariable} from './optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\n/** @doclink Optimizer */\nexport class MomentumOptimizer extends SGDOptimizer {\n  /** @nocollapse */\n  static className = 'Momentum';  // Name matters for Python compatibility.\n  private m: Scalar;\n  private accumulations: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private momentum: number,\n      private useNesterov = false) {\n    super(learningRate);\n    this.m = scalar(this.momentum);\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulations[i] == null) {\n        const trainable = false;\n        this.accumulations[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const accumulation = this.accumulations[i].variable;\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      tidy(() => {\n        let newValue: Tensor;\n        const newAccumulation = add(mul(this.m, accumulation), gradient);\n        if (this.useNesterov) {\n          newValue = add(\n              mul(this.c, add(gradient, mul(newAccumulation, this.m))), value);\n        } else {\n          newValue = add(mul(this.c, newAccumulation), value);\n        }\n        accumulation.assign(newAccumulation);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.m.dispose();\n    if (this.accumulations != null) {\n      dispose(this.accumulations.map(v => v.variable));\n    }\n  }\n\n  /**\n   * Sets the momentum of the optimizer.\n   *\n   * @param momentum\n   */\n  setMomentum(momentum: number) {\n    this.momentum = momentum;\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulations.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulations = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'momentum': this.momentum,\n      'useNesterov': this.useNesterov\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['momentum'], config['useNesterov']);\n  }\n}\nregisterClass(MomentumOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class RMSPropOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'RMSProp';  // Note: Name matters for Python compatibility.\n  private centered: boolean;\n\n  private accumulatedMeanSquares: OptimizerVariable[] = [];\n  private accumulatedMoments: OptimizerVariable[] = [];\n  private accumulatedMeanGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected decay = 0.9,\n      protected momentum = 0.0, protected epsilon: number = null,\n      centered = false) {\n    super();\n\n    this.centered = centered;\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n    if (learningRate == null) {\n      throw new Error(`learningRate for RMSPropOptimizer must be defined.`);\n    }\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedMeanSquares[i] == null) {\n        this.accumulatedMeanSquares[i] = {\n          originalName: `${name}/rms`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMoments[i] == null) {\n        this.accumulatedMoments[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMeanGrads[i] == null && this.centered) {\n        this.accumulatedMeanGrads[i] = {\n          originalName: `${name}/mg`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedMeanSquare = this.accumulatedMeanSquares[i].variable;\n      const accumulatedMoments = this.accumulatedMoments[i].variable;\n      tidy(() => {\n        const newAccumulatedMeanSquare =\n            add(mul(accumulatedMeanSquare, this.decay),\n                mul(square(gradient), 1 - this.decay));\n\n        if (this.centered) {\n          const accumulatedMeanGrad = this.accumulatedMeanGrads[i].variable;\n          // Centered gradient\n          const newAccumulatedMeanGrad =\n              add(mul(accumulatedMeanGrad, this.decay),\n                  mul(gradient, 1 - this.decay));\n\n          const gradContribution =\n              div(mul(gradient, this.learningRate),\n                  sqrt(\n                      sub(newAccumulatedMeanSquare,\n                          add(square(newAccumulatedMeanGrad), this.epsilon))));\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum), gradContribution);\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMeanGrad.assign(newAccumulatedMeanGrad);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        } else {\n          // Plain gradient\n          const newAccumulatedMeanSquare =\n              add(mul(accumulatedMeanSquare, this.decay),\n                  mul(square(gradient), 1 - this.decay));\n\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum),\n                  div(mul(gradient, this.learningRate),\n                      sqrt(add(newAccumulatedMeanSquare, this.epsilon))));\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        }\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedMeanSquares != null) {\n      dispose(this.accumulatedMeanSquares.map(v => v.variable));\n    }\n    if (this.accumulatedMeanGrads != null && this.centered) {\n      dispose(this.accumulatedMeanGrads.map(v => v.variable));\n    }\n    if (this.accumulatedMoments != null) {\n      dispose(this.accumulatedMoments.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedMeanSquares, ...this.accumulatedMoments];\n    if (this.centered) {\n      variables.push(...this.accumulatedMeanGrads);\n    }\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount =\n        this.centered ? weightValues.length / 3 : weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedMeanSquares =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedMoments =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n    if (this.centered) {\n      this.accumulatedMeanGrads =\n          weightValues.slice(variableCount * 2, variableCount * 3)\n              .map(v => ({\n                     originalName: v.name,\n                     variable: v.tensor.variable(trainable)\n                   }));\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'decay': this.decay,\n      'momentum': this.momentum,\n      'epsilon': this.epsilon,\n      'centered': this.centered\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['decay'], config['momentum'],\n        config['epsilon'], config['centered']);\n  }\n}\nregisterClass(RMSPropOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AdadeltaOptimizer} from './adadelta_optimizer';\nimport {AdagradOptimizer} from './adagrad_optimizer';\nimport {AdamOptimizer} from './adam_optimizer';\nimport {AdamaxOptimizer} from './adamax_optimizer';\nimport {MomentumOptimizer} from './momentum_optimizer';\nimport {RMSPropOptimizer} from './rmsprop_optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\nexport class OptimizerConstructors {\n  /**\n   * Constructs a `tf.SGDOptimizer` that uses stochastic gradient descent.\n   *\n   * ```js\n   * // Fit a quadratic function by learning the coefficients a, b, c.\n   * const xs = tf.tensor1d([0, 1, 2, 3]);\n   * const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);\n   *\n   * const a = tf.scalar(Math.random()).variable();\n   * const b = tf.scalar(Math.random()).variable();\n   * const c = tf.scalar(Math.random()).variable();\n   *\n   * // y = a * x^2 + b * x + c.\n   * const f = x => a.mul(x.square()).add(b.mul(x)).add(c);\n   * const loss = (pred, label) => pred.sub(label).square().mean();\n   *\n   * const learningRate = 0.01;\n   * const optimizer = tf.train.sgd(learningRate);\n   *\n   * // Train the model.\n   * for (let i = 0; i < 10; i++) {\n   *   optimizer.minimize(() => loss(f(xs), ys));\n   * }\n   *\n   * // Make predictions.\n   * console.log(\n   *     `a: ${a.dataSync()}, b: ${b.dataSync()}, c: ${c.dataSync()}`);\n   * const preds = f(xs).dataSync();\n   * preds.forEach((pred, i) => {\n   *   console.log(`x: ${i}, pred: ${pred}`);\n   * });\n   * ```\n   *\n   * @param learningRate The learning rate to use for the SGD algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static sgd(learningRate: number): SGDOptimizer {\n    return new SGDOptimizer(learningRate);\n  }\n\n  /**\n   * Constructs a `tf.MomentumOptimizer` that uses momentum gradient\n   * descent.\n   *\n   * See\n   * [http://proceedings.mlr.press/v28/sutskever13.pdf](\n   * http://proceedings.mlr.press/v28/sutskever13.pdf)\n   *\n   * @param learningRate The learning rate to use for the Momentum gradient\n   * descent algorithm.\n   * @param momentum The momentum to use for the momentum gradient descent\n   * algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static momentum(learningRate: number, momentum: number, useNesterov = false):\n      MomentumOptimizer {\n    return new MomentumOptimizer(learningRate, momentum, useNesterov);\n  }\n\n  /**\n   * Constructs a `tf.RMSPropOptimizer` that uses RMSProp gradient\n   * descent. This implementation uses plain momentum and is not centered\n   * version of RMSProp.\n   *\n   * See\n   * [http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf](\n   * http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)\n   *\n   * @param learningRate The learning rate to use for the RMSProp gradient\n   * descent algorithm.\n   * @param decay The discounting factor for the history/coming gradient.\n   * @param momentum The momentum to use for the RMSProp gradient descent\n   * algorithm.\n   * @param epsilon Small value to avoid zero denominator.\n   * @param centered If true, gradients are normalized by the estimated\n   * variance of the gradient.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static rmsprop(\n      learningRate: number, decay = .9, momentum = 0.0, epsilon: number = null,\n      centered = false): RMSPropOptimizer {\n    return new RMSPropOptimizer(\n        learningRate, decay, momentum, epsilon, centered);\n  }\n\n  /**\n   * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adam gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adam(\n      learningRate = 0.001, beta1 = 0.9, beta2 = 0.999,\n      epsilon: number = null): AdamOptimizer {\n    return new AdamOptimizer(learningRate, beta1, beta2, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.\n   * See [https://arxiv.org/abs/1212.5701](https://arxiv.org/abs/1212.5701)\n   *\n   * @param learningRate The learning rate to use for the Adadelta gradient\n   * descent algorithm.\n   * @param rho The learning rate decay over each update.\n   * @param epsilon A constant epsilon used to better condition the grad\n   * update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adadelta(learningRate = .001, rho = .95, epsilon: number = null):\n      AdadeltaOptimizer {\n    return new AdadeltaOptimizer(learningRate, rho, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdamaxOptimizer` that uses the Adamax algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adamax gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   * @param decay The learning rate decay over each update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adamax(\n      learningRate = 0.002, beta1 = 0.9, beta2 = 0.999, epsilon: number = null,\n      decay = 0.0): AdamaxOptimizer {\n    return new AdamaxOptimizer(learningRate, beta1, beta2, epsilon, decay);\n  }\n\n  /**\n   * Constructs a `tf.AdagradOptimizer` that uses the Adagrad algorithm.\n   * See\n   * [http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf](\n   * http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\n   * or\n   * [http://ruder.io/optimizing-gradient-descent/index.html#adagrad](\n   * http://ruder.io/optimizing-gradient-descent/index.html#adagrad)\n   *\n   * @param learningRate The learning rate to use for the Adagrad gradient\n   * descent algorithm.\n   * @param initialAccumulatorValue Starting value for the accumulators, must be\n   * positive.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adagrad(learningRate: number, initialAccumulatorValue = 0.1):\n      AdagradOptimizer {\n    return new AdagradOptimizer(learningRate, initialAccumulatorValue);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// So typings can propagate.\nimport {AdadeltaOptimizer} from './optimizers/adadelta_optimizer';\nimport {AdagradOptimizer} from './optimizers/adagrad_optimizer';\nimport {AdamOptimizer} from './optimizers/adam_optimizer';\nimport {AdamaxOptimizer} from './optimizers/adamax_optimizer';\nimport {MomentumOptimizer} from './optimizers/momentum_optimizer';\nimport {OptimizerConstructors} from './optimizers/optimizer_constructors';\nimport {RMSPropOptimizer} from './optimizers/rmsprop_optimizer';\nimport {SGDOptimizer} from './optimizers/sgd_optimizer';\n\n// tslint:disable-next-line:no-unused-expression\n[MomentumOptimizer, SGDOptimizer, AdadeltaOptimizer, AdagradOptimizer,\n RMSPropOptimizer, AdamaxOptimizer, AdamOptimizer];\n\nexport const train = {\n  sgd: OptimizerConstructors.sgd,\n  momentum: OptimizerConstructors.momentum,\n  adadelta: OptimizerConstructors.adadelta,\n  adagrad: OptimizerConstructors.adagrad,\n  rmsprop: OptimizerConstructors.rmsprop,\n  adamax: OptimizerConstructors.adamax,\n  adam: OptimizerConstructors.adam\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst delayCallback: Function = (() => {\n  if (typeof requestAnimationFrame !== 'undefined') {\n    return requestAnimationFrame;\n  } else if (typeof setImmediate !== 'undefined') {\n    return setImmediate;\n  }\n  return (f: Function) => f();  // no delays\n})();\n\n/**\n * Returns a promise that resolve when a requestAnimationFrame has completed.\n *\n * On Node.js this uses setImmediate instead of requestAnimationFrame.\n *\n * This is simply a sugar method so that users can do the following:\n * `await tf.nextFrame();`\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nfunction nextFrame(): Promise<void> {\n  return new Promise<void>(resolve => delayCallback(() => resolve()));\n}\n\nexport {nextFrame};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Returns the image center in pixels.\nexport function getImageCenter(\n    center: number|[number, number], imageHeight: number,\n    imageWidth: number): [number, number] {\n  const centerX =\n      imageWidth * (typeof center === 'number' ? center : center[0]);\n  const centerY =\n      imageHeight * (typeof center === 'number' ? center : center[1]);\n  return [centerX, centerY];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Gets the new shape of the input Tensor after it's been reshaped\n * to:\n * [blockShape[0], ..., blockShape[M-1], batch / prod(blockShape),\n * inputShape[1], ..., inputShape[N-1]]\n *\n * See step 1: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshaped(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  let reshaped: number[] = [];\n  if (batchToSpace) {\n    reshaped = reshaped.concat(blockShape.slice(0));\n    reshaped.push(inputShape[0] / prod);\n    reshaped = reshaped.concat(inputShape.slice(1));\n  } else {\n    reshaped = reshaped.concat(inputShape[0]);\n    const spatialLength = blockShape.length;\n    for (let i = 0; i < spatialLength; ++i) {\n      reshaped =\n          reshaped.concat([inputShape[i + 1] / blockShape[i], blockShape[i]]);\n    }\n    reshaped = reshaped.concat(inputShape.slice(spatialLength + 1));\n  }\n  return reshaped;\n}\n\n/**\n * Gets the permutation that will transpose the dimensions of the\n * reshaped tensor to shape:\n *\n * [batch / prod(block_shape),inputShape[1], blockShape[0], ...,\n * inputShape[M], blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * see step 2: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getPermuted(\n    reshapedRank: number, blockShapeRank: number,\n    batchToSpace = true): number[] {\n  const permuted = [];\n  if (batchToSpace) {\n    permuted.push(blockShapeRank);\n    for (let i = blockShapeRank + 1; i < reshapedRank; ++i) {\n      if (i <= 2 * blockShapeRank) {\n        permuted.push(i);\n        permuted.push(i - (blockShapeRank + 1));\n      } else {\n        permuted.push(i);\n      }\n    }\n  } else {\n    const permutedBeforeBatch = [];\n    const permutedAfterBatch = [];\n    for (let i = 1; i < reshapedRank; ++i) {\n      if (i >= blockShapeRank * 2 + 1 || i % 2 === 1) {\n        permutedAfterBatch.push(i);\n      } else {\n        permutedBeforeBatch.push(i);\n      }\n    }\n    permuted.push(...permutedBeforeBatch);\n    permuted.push(0);\n    permuted.push(...permutedAfterBatch);\n  }\n  return permuted;\n}\n\n/**\n * Gets the shape of the reshaped and permuted input Tensor before any cropping\n * is applied.  The new shape will be:\n *\n * [batch / prod(blockShape),inputShape[1] * blockShape[0], ...,\n * inputShape[M] * blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 3: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshapedPermuted(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  const reshapedPermuted = [];\n\n  if (batchToSpace) {\n    reshapedPermuted.push(inputShape[0] / prod);\n  } else {\n    reshapedPermuted.push(inputShape[0] * prod);\n  }\n\n  for (let i = 1; i < inputShape.length; ++i) {\n    if (i <= blockShape.length) {\n      if (batchToSpace) {\n        reshapedPermuted.push(blockShape[i - 1] * inputShape[i]);\n      } else {\n        reshapedPermuted.push(inputShape[i] / blockShape[i - 1]);\n      }\n    } else {\n      reshapedPermuted.push(inputShape[i]);\n    }\n  }\n\n  return reshapedPermuted;\n}\n\n/**\n * Converts the crops argument into the beginning coordinates of a slice\n * operation.\n */\nexport function getSliceBeginCoords(\n    crops: number[][], blockShape: number): number[] {\n  const sliceBeginCoords = [0];\n  for (let i = 0; i < blockShape; ++i) {\n    sliceBeginCoords.push(crops[i][0]);\n  }\n  return sliceBeginCoords;\n}\n\n/**\n * Converts the crops argument into the size of a slice operation.  When\n * combined with getSliceBeginCoords this function allows the reshaped and\n * permuted Tensor to be cropped to its final output shape of:\n *\n * inputShape[1] * blockShape[0] - crops[0,0] - crops[0,1], ...,\n * inputShape[M] * blockShape[M-1] -crops[M-1,0] -\n * crops[M-1,1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 4: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getSliceSize(\n    uncroppedShape: number[], crops: number[][], blockShape: number): number[] {\n  const sliceSize = uncroppedShape.slice(0, 1);\n  for (let i = 0; i < blockShape; ++i) {\n    sliceSize.push(uncroppedShape[i + 1] - crops[i][0] - crops[i][1]);\n  }\n\n  return sliceSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const SELU_SCALEALPHA = 1.7580993408473768599402175208123;\nexport const SELU_SCALE = 1.0507009873554804934193349852946;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const ERF_P = 0.3275911;\nexport const ERF_A1 = 0.254829592;\nexport const ERF_A2 = -0.284496736;\nexport const ERF_A3 = 1.421413741;\nexport const ERF_A4 = -1.453152027;\nexport const ERF_A5 = 1.061405429;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\n\nexport function warn(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.warn(...msg);\n  }\n}\n\nexport function log(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.log(...msg);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TypedArray} from '../types';\n/**\n * Merges real and imaginary Float32Arrays into a single complex Float32Array.\n *\n * The memory layout is interleaved as follows:\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n * complex: [r0, i0, r1, i1, r2, i2]\n *\n * This is the inverse of splitRealAndImagArrays.\n *\n * @param real The real values of the complex tensor values.\n * @param imag The imag values of the complex tensor values.\n * @returns A complex tensor as a Float32Array with merged values.\n */\nexport function mergeRealAndImagArrays(\n    real: Float32Array, imag: Float32Array): Float32Array {\n  if (real.length !== imag.length) {\n    throw new Error(\n        `Cannot merge real and imag arrays of different lengths. real:` +\n        `${real.length}, imag: ${imag.length}.`);\n  }\n  const result = new Float32Array(real.length * 2);\n  for (let i = 0; i < result.length; i += 2) {\n    result[i] = real[i / 2];\n    result[i + 1] = imag[i / 2];\n  }\n  return result;\n}\n\n/**\n * Splits a complex Float32Array into real and imag parts.\n *\n * The memory layout is interleaved as follows:\n * complex: [r0, i0, r1, i1, r2, i2]\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n *\n * This is the inverse of mergeRealAndImagArrays.\n *\n * @param complex The complex tensor values.\n * @returns An object with real and imag Float32Array components of the complex\n *     tensor.\n */\nexport function splitRealAndImagArrays(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(complex.length / 2);\n  const imag = new Float32Array(complex.length / 2);\n  for (let i = 0; i < complex.length; i += 2) {\n    real[i / 2] = complex[i];\n    imag[i / 2] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts even indexed complex values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithEvenIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.ceil(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 0; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts odd indexed comple values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithOddIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.floor(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 2; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Get the map representing a complex value in the given array.\n * @param complex The complex tensor values.\n * @param index An index of the target complex value.\n */\nexport function getComplexWithIndex(\n    complex: Float32Array, index: number): {real: number, imag: number} {\n  const real = complex[index * 2];\n  const imag = complex[index * 2 + 1];\n  return {real, imag};\n}\n\n/**\n * Insert a given complex value into the TypedArray.\n * @param data The array in which the complex value is inserted.\n * @param c The complex value to be inserted.\n * @param index An index of the target complex value.\n */\nexport function assignToTypedArray(\n    data: TypedArray, real: number, imag: number, index: number) {\n  data[index * 2] = real;\n  data[index * 2 + 1] = imag;\n}\n\n/**\n * Make the list of exponent terms used by FFT.\n */\nexport function exponents(\n    n: number, inverse: boolean): {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(n / 2);\n  const imag = new Float32Array(n / 2);\n  for (let i = 0; i < Math.ceil(n / 2); i++) {\n    const x = (inverse ? 2 : -2) * Math.PI * (i / n);\n    real[i] = Math.cos(x);\n    imag[i] = Math.sin(x);\n  }\n  return {real, imag};\n}\n\n/**\n * Make the exponent term used by FFT.\n */\nexport function exponent(\n    k: number, n: number, inverse: boolean): {real: number, imag: number} {\n  const x = (inverse ? 2 : -2) * Math.PI * (k / n);\n  const real = Math.cos(x);\n  const imag = Math.sin(x);\n  return {real, imag};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {cast} from '../ops/cast';\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\nimport {Rank} from '../types';\nimport {DataType, ShapeMap} from '../types';\nimport {hasEncodingLoss, makeZerosTypedArray} from '../util';\n\nimport {KernelBackend} from './backend';\n\n// Utilities needed by backend consumers of tf-core.\nexport * from '../ops/axis_util';\nexport * from '../ops/broadcast_util';\nexport * from '../ops/concat_util';\nexport * from '../ops/conv_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/fused_types';\nexport * from '../ops/reduce_util';\n\nimport * as slice_util from '../ops/slice_util';\nexport {slice_util};\n\nexport {BackendValues, TypedArray, upcastType, PixelData} from '../types';\nexport {MemoryInfo, TimingInfo} from '../engine';\nexport * from '../ops/rotate_util';\nexport * from '../ops/array_ops_util';\nexport * from '../ops/gather_nd_util';\nexport * from '../ops/scatter_nd_util';\nexport * from '../ops/selu_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/erf_util';\nexport * from '../log';\nexport * from '../backends/complex_util';\nexport * from '../ops/split_util';\n\nimport * as segment_util from '../ops/segment_util';\nexport {segment_util};\n\nexport function castTensor<T extends Tensor>(\n    x: T, dtype: DataType, backend: KernelBackend): T {\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return x.clone();\n    }\n    const zerosTensor = zeros(x.shape);\n    const floatX = cast(x, 'float32');\n    const result = backend.complex(floatX, zerosTensor);\n    zerosTensor.dispose();\n    floatX.dispose();\n    return result as T;\n  }\n\n  if (!hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    return ENGINE.makeTensorFromDataId(x.dataId, x.shape, dtype) as T;\n  }\n  if (x.dtype === 'complex64') {\n    const real = backend.real(x);\n    const result = cast(real, dtype);\n    real.dispose();\n    return result;\n  }\n  if (dtype === 'int32') {\n    return backend.int(x);\n  } else if (dtype === 'bool') {\n    const zero = scalar(0, x.dtype);\n    const result = backend.notEqual(x, zero) as T;\n    zero.dispose();\n    return result;\n  } else {\n    throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n  }\n}\n\nexport function reshapeTensor<T extends Tensor, R extends Rank>(\n    x: T, shape: ShapeMap[R]): Tensor<R> {\n  return ENGINE.makeTensorFromDataId(x.dataId, shape, x.dtype) as Tensor<R>;\n}\n\nexport function linspaceImpl(start: number, stop: number, num: number) {\n  const step = (stop - start) / (num - 1);\n\n  const values = makeZerosTypedArray(num, 'float32');\n  values[0] = start;\n  for (let i = 1; i < values.length; i++) {\n    values[i] = values[i - 1] + step;\n  }\n\n  return tensor1d(values, 'float32');\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\n// TODO(annxingyuan): Use this helper in WASM Split kernel once intermediate\n// kernels have been modularized in WebGL and CPU\n// https://github.com/tensorflow/tfjs/issues/2822.\n/** Shared implementation of the split kernel across WebGL and CPU. */\nexport function split<T extends Tensor>(\n    x: T, sizeSplits: number[], axis: number): T[] {\n  const begin = new Array(x.rank).fill(0);\n  const size = x.shape.slice();\n  return sizeSplits.map(s => {\n    const sliceSize = [...size];\n    sliceSize[axis] = s;\n    const sliceT = slice(x, begin, sliceSize);\n    begin[axis] += s;\n    return sliceT;\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * An implementation of the tile kernel shared between webgl and cpu for string\n * tensors only.\n */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {DataType, Rank} from '../types';\n\nexport function tile<R extends Rank>(\n    xBuf: TensorBuffer<R, DataType>, reps: number[]): Tensor<R> {\n  const newShape: number[] = new Array(xBuf.rank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = xBuf.shape[i] * reps[i];\n  }\n  const result = buffer(newShape, xBuf.dtype);\n  for (let i = 0; i < result.values.length; ++i) {\n    const newLoc = result.indexToLoc(i);\n\n    const originalLoc: number[] = new Array(xBuf.rank);\n    for (let j = 0; j < originalLoc.length; j++) {\n      originalLoc[j] = newLoc[j] % xBuf.shape[j];\n    }\n\n    const originalIndex = xBuf.locToIndex(originalLoc);\n\n    result.values[i] = xBuf.values[originalIndex];\n  }\n  return result.toTensor() as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the TopK kernel shared between webgl and cpu. */\n\nimport {tensor} from '../ops/tensor';\nimport {Tensor} from '../tensor';\nimport {NumericDataType, TypedArray} from '../types';\nimport {getTypedArrayFromDType} from '../util';\n\nexport function topkImpl<T extends Tensor>(\n    x: TypedArray, xShape: number[], xDtype: NumericDataType, k: number,\n    sorted: boolean): [T, T] {\n  // Reshape into a 2d tensor [batch, lastDim] and compute topk along lastDim.\n  const lastDim = xShape[xShape.length - 1];\n  const [batch, size] = [x.length / lastDim, lastDim];\n  const allTopKVals = getTypedArrayFromDType(xDtype, batch * k);\n  const allTopKIndices = getTypedArrayFromDType('int32', batch * k);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = x.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    const outOffset = b * k;\n    const topKVals = allTopKVals.subarray(outOffset, outOffset + k);\n    const topKIndices = allTopKIndices.subarray(outOffset, outOffset + k);\n    for (let i = 0; i < k; i++) {\n      topKVals[i] = valAndInd[i].value;\n      topKIndices[i] = valAndInd[i].index;\n    }\n  }\n  // Reshape back to the original input shape, except that the last\n  // dimension is k.\n  const outputShape = xShape.slice();\n  outputShape[outputShape.length - 1] = k;\n  return [\n    tensor(allTopKVals, outputShape, xDtype) as T,\n    tensor(allTopKIndices, outputShape, 'int32') as T\n  ];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport {nonMaxSuppressionV3Impl, nonMaxSuppressionV4Impl, nonMaxSuppressionV5Impl} from './non_max_suppression_impl';\nexport {split} from './split_shared';\nexport {tile} from './tile_impl';\nexport {topkImpl} from './topk_impl';\nexport {whereImpl} from './where_impl';\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const absGradConfig: GradConfig = {\n  kernelName: Abs,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, step(cast(x, 'float32'), -1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acosGradConfig: GradConfig = {\n  kernelName: Acos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = square(cast(x, 'float32'));\n        const b = sqrt(sub(scalar(1), a));\n        return neg(div(dy, b));\n      }\n\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acoshGradConfig: GradConfig = {\n  kernelName: Acosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(sub(square(cast(x, 'float32')), 1));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Add} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const addGradConfig: GradConfig = {\n  kernelName: Add,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AddN} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const addNGradConfig: GradConfig = {\n  kernelName: AddN,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const ders: {[key: string]: () => Tensor} = {};\n    saved.forEach((_, i) => {\n      ders[i] = () => dy.clone();\n    });\n    return ders;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMax} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMaxGradConfig: GradConfig = {\n  kernelName: ArgMax,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMinGradConfig: GradConfig = {\n  kernelName: ArgMin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const asinGradConfig: GradConfig = {\n  kernelName: Asin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, sqrt(sub(scalar(1), square(cast(x, 'float32')))))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const asinhGradConfig: GradConfig = {\n  kernelName: Asinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(add(scalar(1), square(cast(x, 'float32'))));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const atan2GradConfig: GradConfig = {\n  kernelName: Atan2,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const d = add(square(a), square(b));\n      let res = mul(dy, div(b, d));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      const d = add(square(a), square(b));\n      let res = neg(mul(dy, div(a, d)));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const atanGradConfig: GradConfig = {\n  kernelName: Atan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, add(square(cast(x, 'float32')), 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {scalar} from '../ops/scalar';\nimport {Tensor} from '../tensor';\n\nexport const atanhGradConfig: GradConfig = {\n  kernelName: Atanh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, sub(scalar(1), square(cast(x, 'float32'))))};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool3DBackprop, AvgPool3DBackpropAttrs, AvgPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d avg pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0. The dilation\n *     rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction avgPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in avgPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in avgPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.avgPool3dBackprop(dy5D, input5D, convInfo);\n  };\n\n  const inputs: AvgPool3DBackpropInputs = {dy: dy5D, input: input5D};\n\n  const attrs: AvgPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      AvgPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3dBackprop = op({avgPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool3D, AvgPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPool3dBackprop} from '../ops/avg_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const avgPool3DGradConfig: GradConfig = {\n  kernelName: AvgPool3D,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as AvgPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => avgPool3dBackprop(\n          dy as Tensor5D, x, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of an 2D avg pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The input image, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction avgPoolBackprop_<T extends Tensor3D|Tensor4D>(\n    dy: T|TensorLike, input: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy (${\n          $dy.rank})`);\n\n  let input4D = $input as Tensor4D;\n  let dy4D = $dy as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($input.rank === 3) {\n    reshapedTo4D = true;\n    input4D =\n        reshape($input, [1, $input.shape[0], $input.shape[1], $input.shape[2]]);\n    dy4D = reshape($dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2]]);\n  }\n\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in avgPoolBackprop: dy must be rank 4 but got rank ` +\n          `${dy4D.rank}.`);\n  util.assert(\n      input4D.rank === 4,\n      () => `Error in avgPoolBackprop: input must be rank 4 but got rank ` +\n          `${input4D.rank}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        input4D.shape, filterSize, strides, 1 /* dilations */, pad);\n\n    return backend.avgPoolBackprop(dy4D, input4D, convInfo);\n  };\n\n  const inputs: AvgPoolBackpropInputs = {dy: dy4D, input: input4D};\n\n  const attrs: AvgPoolBackpropAttrs = {filterSize, strides, pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, AvgPoolBackprop,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const avgPoolBackprop = op({avgPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool, AvgPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPoolBackprop} from '../ops/avg_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const avgPoolGradConfig: GradConfig = {\n  kernelName: AvgPool,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as AvgPoolAttrs;\n    return {\n      x: () => avgPoolBackprop(dy as Tensor4D, x, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchMatMul, BatchMatMulAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {matMul} from '../ops/mat_mul';\nimport {Tensor, Tensor3D} from '../tensor';\n\nexport const batchMatMulGradConfig: GradConfig = {\n  kernelName: BatchMatMul,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor3D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [a, b] = saved as Tensor3D[];\n\n    const {transposeA, transposeB} = attrs as {} as BatchMatMulAttrs;\n\n    if (!transposeA && !transposeB) {\n      return {\n        a: () => matMul(dy, b, false, true),\n        b: () => matMul(a, dy, true, false)\n      };\n    } else if (!transposeA && transposeB) {\n      return {\n        a: () => matMul(dy, b, false, false),\n        b: () => matMul(dy, a, true, false)\n      };\n    } else if (transposeA && !transposeB) {\n      return {\n        a: () => matMul(b, dy, false, true),\n        b: () => matMul(a, dy, false, false)\n      };\n    } else {\n      return {\n        a: () => matMul(b, dy, true, true),\n        b: () => matMul(dy, a, true, true)\n      };\n    }\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchToSpaceND, BatchToSpaceNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {spaceToBatchND} from '../ops/space_to_batch_nd';\nimport {Tensor} from '../tensor';\n\nexport const batchToSpaceNDGradConfig: GradConfig = {\n  kernelName: BatchToSpaceND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, crops} = attrs as {} as BatchToSpaceNDAttrs;\n    return {x: () => spaceToBatchND(dy, blockShape, crops)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BroadcastTo, BroadCastToAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const broadcastToGradConfig: GradConfig = {\n  kernelName: BroadcastTo,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const broadCastToAttrs: BroadCastToAttrs =\n        attrs as unknown as BroadCastToAttrs;\n\n    const inputShape = broadCastToAttrs.inputShape;\n    const outputShape = broadCastToAttrs.shape;\n\n    const reps: number[] = Array.from(outputShape);\n    for (let i = inputShape.length - 1; i >= 0; i--) {\n      if (inputShape[i] === outputShape[i]) {\n        reps[i] = 1;\n      } else if (inputShape[i] !== 1) {\n        throw new Error(`broadcastTo(): [${\n            inputShape}] cannot be broadcast to [${outputShape}].`);\n      }\n    }\n    const axes: number[] = [];\n    for (let i = 0; i < reps.length; i++) {\n      if (reps[i] > 1) {\n        axes.push(i);\n      }\n    }\n\n    return {x: () => sum(dy, axes, true /* keepDims */)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cast} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const castGradConfig: GradConfig = {\n  kernelName: Cast,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => dy.clone()};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const ceilGradConfig: GradConfig = {\n  kernelName: Ceil,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {lessEqual} from '../ops/less_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const clipByValueGradConfig: GradConfig = {\n  kernelName: ClipByValue,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {clipValueMin, clipValueMax} = attrs as {} as ClipByValueAttrs;\n    return {\n      x: () => where(\n          logicalAnd(greaterEqual(x, clipValueMin), lessEqual(x, clipValueMax)),\n          dy, zerosLike(dy)),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Concat, ConcatAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {split} from '../ops/split';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const concatGradConfig: GradConfig = {\n  kernelName: Concat,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const shapes = saved.map(t => t.shape);\n    const {axis} = attrs as {} as ConcatAttrs;\n    const $axis = parseAxisParam(axis, saved[0].shape)[0];\n    const sizeSplits = shapes.map(s => s[$axis]);\n    const derTensors = split(dy, sizeSplits, $axis);\n    return derTensors.map(t => () => t) as {};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2D, Conv2DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../ops/conv2d_backprop_input';\nimport * as conv_util from '../ops/conv_util';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv2DGradConfig: GradConfig = {\n  kernelName: Conv2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x4D, $filter] = saved as [Tensor4D, Tensor4D];\n    const {dilations, strides, pad, dataFormat} = attrs as {} as Conv2DAttrs;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of conv2D: dilation rates greater than 1 ' +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    return {\n      x: () =>\n          conv2DBackpropInput(x4D.shape, dy, $filter, strides, pad, dataFormat),\n      filter: () =>\n          conv2DBackpropFilter(x4D, dy, $filter.shape, strides, pad, dataFormat)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2d} from '../ops/conv2d';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const conv2DBackpropInputGradConfig: GradConfig = {\n  kernelName: Conv2DBackpropInput,\n  inputsToSave: ['dy', 'filter'],\n  gradFunc: (ddx: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [dy, filter] = saved as [Tensor4D, Tensor4D];\n\n    const {strides, pad, dataFormat, dimRoundingMode} =\n        attrs as {} as Conv2DBackpropInputAttrs;\n\n    return {\n      dy: () => conv2d(\n          ddx, filter, strides, pad, dataFormat, 1 /* dilations */,\n          dimRoundingMode),\n      filter: () => conv2DBackpropFilter(\n          ddx, dy, filter.shape, strides, pad, dataFormat, dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropFilterAttrs, Conv3DBackpropFilterInputs, Conv3DBackpropFilterV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 3D convolution.\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     [batch, depth, height, width, inChannels]. If rank 4, batch of 1 is\n *     assumed.\n * @param dy The dy image, of rank 5 or rank 4, of shape\n *     [batch, depth, height, width, outDepth]. If rank 4, batch of 1 is\n *     assumed.\n * @param filterShape The shape of the filter, length 5,\n *     [filterDepth, filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideDepth, strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction conv3DBackpropFilter_<T extends Tensor4D|Tensor5D>(\n    x: T, dy: T, filterShape: [number, number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): Tensor5D {\n  let x5D = x as Tensor5D;\n  if (x.rank === 4) {\n    x5D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2], x.shape[3]]);\n  }\n  let dy5D = dy as Tensor5D;\n  if (dy5D.rank === 4) {\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3dDerFilter: input must be rank 5, but got shape ` +\n          `${x5D.shape}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerFilter: dy must be rank 5, but got shape ` +\n          `${dy5D.shape}.`);\n  util.assert(\n      filterShape.length === 5,\n      () => `Error in conv3dDerFilter: filterShape must be length 5, but got ` +\n          `${filterShape}.`);\n  util.assert(\n      x5D.shape[4] === filterShape[3],\n      () => `Error in conv3dDerFilter: depth of input ${x5D.shape[4]}) must ` +\n          `match input depth in filter (${filterShape[3]}.`);\n  util.assert(\n      dy5D.shape[4] === filterShape[4],\n      () => `Error in conv3dDerFilter: depth of dy (${dy5D.shape[4]}) must ` +\n          `match output depth for filter (${filterShape[4]}).`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, filterShape, strides, dilations, pad);\n\n    return backend.conv3dDerFilter(x5D, dy5D, convInfo);\n  };\n\n  const inputs: Conv3DBackpropFilterInputs = {x: x5D, y: dy5D};\n\n  const attrs: Conv3DBackpropFilterAttrs = {strides, pad};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv3DBackpropFilterV2, attrs as {} as NamedAttrMap) as Tensor5D;\n}\n\nexport const conv3DBackpropFilter = op({conv3DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv3D, Conv3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv3DBackpropFilter} from '../ops/conv3d_backprop_filter';\nimport {conv3DBackpropInput} from '../ops/conv3d_backprop_input';\nimport {tupleValuesAreOne} from '../ops/conv_util';\nimport {Tensor, Tensor5D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv3DGradConfig: GradConfig = {\n  kernelName: Conv3D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor5D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad} = attrs as {} as Conv3DAttrs;\n    util.assert(\n        tupleValuesAreOne(dilations),\n        () =>\n            'Error in gradient of conv3D: dilation rates greater than 1 are ' +\n            `not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const [x5D, $filter] = saved;\n\n    return {\n      x: () => conv3DBackpropInput(\n          (x5D as Tensor5D).shape, dy, $filter as Tensor5D, strides, pad),\n      filter: () => conv3DBackpropFilter(\n          x5D as Tensor5D, dy, ($filter as Tensor5D).shape, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {sin} from '../ops/sin';\nimport {Tensor} from '../tensor';\n\nexport const cosGradConfig: GradConfig = {\n  kernelName: Cos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(neg(sin(cast(x, 'float32'))), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {sinh} from '../ops/sinh';\nimport {Tensor} from '../tensor';\n\nexport const coshGradConfig: GradConfig = {\n  kernelName: Cosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(sinh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cumsum, CumsumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getAxesPermutation} from '../ops/axis_util';\nimport {cumsum} from '../ops/cumsum';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const cumsumGradConfig: GradConfig = {\n  kernelName: Cumsum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {axis, exclusive, reverse}: CumsumAttrs = attrs as {} as CumsumAttrs;\n\n    return {\n      x: () => {\n        const permutation = getAxesPermutation([axis], x.rank);\n\n        let out = cumsum(dy, axis, exclusive, !reverse);\n\n        if (permutation != null) {\n          out = transpose(out, permutation);\n        }\n\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as conv_util from '../ops/conv_util';\nimport {depthwiseConv2dNativeBackpropFilter} from '../ops/depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../ops/depthwise_conv2d_native_backprop_input';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const depthwiseConv2dNativeGradConfig: GradConfig = {\n  kernelName: DepthwiseConv2dNative,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad, dimRoundingMode} =\n        attrs as {} as DepthwiseConv2dNativeAttrs;\n\n    const $dilations = dilations == null ? [1, 1] : dilations;\n\n    util.assert(\n        conv_util.tupleValuesAreOne($dilations),\n        () => 'Error in gradient of depthwiseConv2dNative: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${$dilations}'`);\n\n    const [x, filter] = saved as [Tensor4D, Tensor4D];\n\n    util.assert(\n        x.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: input must be ` +\n            `rank 4, but got rank ${x.rank}.`);\n    util.assert(\n        filter.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: filter must be ` +\n            `rank 4, but got rank ${filter.rank}.`);\n    util.assert(\n        x.shape[3] === filter.shape[2],\n        () => `Error in gradient of depthwiseConv2d: number of input ` +\n            `channels (${x.shape[3]}) must match the inChannels dimension ` +\n            `in filter ${filter.shape[2]}.`);\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, $dilations),\n        () => 'Error in gradient of depthwiseConv2d: Either strides or ' +\n            `dilations must be  1. Got strides ${strides} and dilations ` +\n            `'${$dilations}'.`);\n\n    if (dimRoundingMode != null) {\n      util.assert(\n          util.isInt(pad as number),\n          () =>\n              `Error in depthwiseConv2d: pad must be an integer when using, ` +\n              `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n    }\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x.shape, filter.shape, strides, $dilations as number | [number, number],\n        pad, dimRoundingMode, true /* depthwise */);\n\n    return {\n      x: () =>\n          depthwiseConv2dNativeBackpropInput(x.shape, dy, filter, convInfo),\n      filter: () =>\n          depthwiseConv2dNativeBackpropFilter(x, dy, filter.shape, convInfo),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DBackpropFilter, Dilation2DBackpropFilterInputs, Dilation2DBackpropInput, Dilation2DBackpropInputInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const dilation2dGradConfig: GradConfig = {\n  kernelName: Dilation2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, filter] = saved as [Tensor4D, Tensor3D];\n\n    const inputInputs: Dilation2DBackpropInputInputs = {x, filter, dy};\n    const filterInputs: Dilation2DBackpropFilterInputs = {x, filter, dy};\n\n    return {\n      x: () => ENGINE.runKernel(\n                   Dilation2DBackpropInput, inputInputs as {} as NamedTensorMap,\n                   attrs) as Tensor,\n      filter: () => ENGINE.runKernel(\n                        Dilation2DBackpropFilter,\n                        filterInputs as {} as NamedTensorMap, attrs) as Tensor\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const divGradConfig: GradConfig = {\n  kernelName: Div,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluGrad, EluGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const eluGradConfig: GradConfig = {\n  kernelName: Elu,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      return backend.eluDer(dy, y);\n    };\n\n    const inputs: EluGradInputs = {dy, y};\n\n    return {\n      x: () => ENGINE.runKernelFunc(\n          backPropKernelFunc, inputs as {} as NamedTensorMap, null /* grad */,\n          EluGrad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Erf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const erfGradConfig: GradConfig = {\n  kernelName: Erf,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const a = mul(exp(neg(square(x))), 2 / Math.sqrt(Math.PI));\n    return {x: () => mul(dy, a)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expGradConfig: GradConfig = {\n  kernelName: Exp,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n    return {x: () => mul(dy, y)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expm1GradConfig: GradConfig = {\n  kernelName: Expm1,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, exp(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const floorGradConfig: GradConfig = {\n  kernelName: Floor,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FloorDiv} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const floorDivGradConfig: GradConfig = {\n  kernelName: FloorDiv,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {FusedBatchNorm, FusedBatchNormAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {rsqrt} from '../ops/rsqrt';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {tile} from '../ops/tile';\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nexport const fusedBatchNormGradConfig: GradConfig = {\n  kernelName: FusedBatchNorm,\n  inputsToSave: ['x', 'mean', 'variance', 'scale'],\n  gradFunc: <R extends Rank>(\n      dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {varianceEpsilon} = attrs as {} as FusedBatchNormAttrs;\n    const [x, mean, variance, scale] = saved;\n\n    const scaleValue = scale == null ? scalar(1) : scale;\n    const reductionAxes = getReductionAxes(mean.shape, x.shape);\n    const tileShape: number[] = [];\n    if (mean.rank === 1) {\n      for (let i = 0; i < x.shape.length - 1; ++i) {\n        tileShape.push(x.shape[i]);\n      }\n      tileShape.push(1);\n    }\n\n    const xMinusMean = sub(x, mean);\n    const dyTimesScaleValue = mul(dy, scaleValue);\n    const oneOverSqrtVariance = rsqrt(add(variance, scalar(varianceEpsilon)));\n    const minusHalfRCube = mul(\n        mul(mul(oneOverSqrtVariance, oneOverSqrtVariance), oneOverSqrtVariance),\n        scalar(-0.5));\n\n    const derX = () => {\n      if (mean.rank === 1) {\n        return reshape(\n            mul(mul(dy,\n                    tile(\n                        reshape(oneOverSqrtVariance, [1, 1, 1, mean.shape[0]]),\n                        tileShape)),\n                scaleValue),\n            x.shape);\n      } else {\n        return reshape(mul(mul(dy, oneOverSqrtVariance), scaleValue), x.shape);\n      }\n    };\n    const derMean = () => {\n      let meanDer =\n          mul(mul(oneOverSqrtVariance, scalar(-1)), dyTimesScaleValue);\n      if (mean.rank === 1) {\n        meanDer = sum(meanDer, reductionAxes);\n      }\n      return reshape(meanDer, mean.shape as ShapeMap[R]);\n    };\n    const derVariance = () => {\n      let varianceDer = mul(mul(minusHalfRCube, xMinusMean), dyTimesScaleValue);\n\n      if (mean.rank === 1) {\n        varianceDer = sum(varianceDer, reductionAxes);\n      }\n      return reshape(varianceDer, mean.shape as ShapeMap[R]);\n    };\n    const derScale = () => {\n      const xMinusMean2TimesRsqrt = mul(xMinusMean, oneOverSqrtVariance);\n\n      let scaleDer = mul(dy, xMinusMean2TimesRsqrt);\n      if (mean.rank === 1) {\n        scaleDer = sum(scaleDer, reductionAxes);\n      }\n      return reshape(scaleDer, mean.shape as ShapeMap[R]);\n    };\n    const derOffset = () => {\n      let offsetDer = dy;\n      if (mean.rank === 1) {\n        offsetDer = sum(offsetDer, reductionAxes);\n      }\n      return reshape(offsetDer, mean.shape as ShapeMap[R]);\n    };\n\n    return {\n      x: derX,\n      mean: derMean,\n      variance: derVariance,\n      scale: derScale,\n      offset: derOffset\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GatherV2, GatherV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getUndoAxesPermutation} from '../ops/axis_util';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {unsortedSegmentSum} from '../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const gatherGradConfig: GradConfig = {\n  kernelName: GatherV2,\n  inputsToSave: ['x', 'indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, indices] = saved;\n    const {axis} = attrs as {} as GatherV2Attrs;\n\n    const parsedAxis = parseAxisParam(axis, x.shape)[0];\n\n    const derX = () => {\n      const paramsShape = x.shape;\n      const indicesSize = indices.size;\n\n      const outerShape = paramsShape.slice(0, parsedAxis);\n      const outerDims = outerShape.length;\n      const innerShape = paramsShape.slice(axis, paramsShape.length).slice(1);\n      const innerDims = innerShape.length;\n\n      const outerAxesIndices = arrayRange(0, outerDims);\n      const innerAxesIndices =\n          arrayRange(outerDims + 1, outerDims + 1 + innerDims);\n\n      const valuesShape = arrayConcat([outerShape, [indicesSize], innerShape]);\n\n      const values = reshape(dy, valuesShape);\n      const reshapedIndices = reshape(indices, [indicesSize]);\n\n      const transposeDims =\n          arrayConcat([[outerDims], outerAxesIndices, innerAxesIndices]);\n      const valuesTranspose = transpose(values, transposeDims);\n      let paramsGrad = unsortedSegmentSum(\n          valuesTranspose, reshapedIndices as Tensor1D, x.shape[parsedAxis]);\n\n      const invertTransposeDims = getUndoAxesPermutation(transposeDims);\n      paramsGrad = transpose(paramsGrad, invertTransposeDims);\n\n      return paramsGrad;\n    };\n    return {x: derX, indices: () => indices};\n  }\n};\n\nfunction arrayRange(start: number, stop: number): number[] {\n  const result = [];\n  for (let i = start; i < stop; ++i) {\n    result.push(i);\n  }\n  return result;\n}\n\nfunction arrayConcat(arrays: number[][]): number[] {\n  const result = [];\n  for (let i = 0; i < arrays.length; ++i) {\n    for (let j = 0; j < arrays[i].length; ++j) {\n      result.push(arrays[i][j]);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GreaterEqual} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const greaterEqualGradConfig: GradConfig = {\n  kernelName: GreaterEqual,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    return {a: () => zerosLike(a), b: () => zerosLike(b)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {Tensor} from '../tensor';\n\nexport const identityGradConfig: GradConfig = {\n  kernelName: Identity,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => cast(dy, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isFiniteGradConfig: GradConfig = {\n  kernelName: IsFinite,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isInfGradConfig: GradConfig = {\n  kernelName: IsInf,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isNanGradConfig: GradConfig = {\n  kernelName: IsNan,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log1p} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const log1pGradConfig: GradConfig = {\n  kernelName: Log1p,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, add(x, 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const logGradConfig: GradConfig = {\n  kernelName: Log,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, cast(x, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {LogSoftmax, LogSoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const logSoftmaxGradConfig: GradConfig = {\n  kernelName: LogSoftmax,\n  inputsToSave: [],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [value] = saved;\n    const {axis} = attrs as {} as LogSoftmaxAttrs;\n    return {\n      logits: () => {\n        const keepDims = true;\n        const softmax = exp(value);\n        return sub(dy, mul(sum(dy, axis, keepDims), softmax));\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRNBackprop, LRNBackpropAttrs, LRNBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport {op} from './operation';\n\nfunction localResponseNormalizationBackprop_<T extends Tensor4D>(\n    x: T, y: T, dy: T, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.LRNGrad(dy, x, y, depthRadius, bias, alpha, beta);\n\n  const inputs: LRNBackpropInputs = {x, y, dy};\n\n  const attrs: LRNBackpropAttrs = {depthRadius, bias, alpha, beta};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LRNBackprop, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const localResponseNormalizationBackprop =\n    op({localResponseNormalizationBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {LRN, LRNAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {localResponseNormalizationBackprop} from '../ops/local_response_normalization_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const lrnGradConfig: GradConfig = {\n  kernelName: LRN,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {depthRadius, bias, alpha, beta} = attrs as {} as LRNAttrs;\n\n    return {\n      x: () => localResponseNormalizationBackprop(\n          x, y, dy, depthRadius, bias, alpha, beta)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as axis_util from '../ops/axis_util';\nimport {cast} from '../ops/cast';\nimport {equal} from '../ops/equal';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\n/**\n * Gradient helper function for the min and max operations.\n */\nexport function gradForMinAndMax<T extends Tensor>(\n    dy: T, y: T, xOrig: Tensor, origAxes: number[], permutedAxes: number[]) {\n  if (y.rank < xOrig.rank) {\n    y = reshape(y, axis_util.expandShapeToKeepDim(y.shape, origAxes)) as T;\n  }\n  if (dy.rank < xOrig.rank) {\n    dy = reshape(dy, axis_util.expandShapeToKeepDim(dy.shape, origAxes)) as T;\n  }\n  return {\n    x: () => {\n      const dx = mul(dy, cast(equal(xOrig, y), dy.dtype));\n      return permutedAxes == null ? dx : transpose(dx, permutedAxes);\n    }\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const maxGradConfig: GradConfig = {\n  kernelName: Max,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const maxAttrs: MaxAttrs = attrs as {} as MaxAttrs;\n    const {reductionIndices} = maxAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const maxGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = maxGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Maximum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {less} from '../ops/less';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const maximumGradConfig: GradConfig = {\n  kernelName: Maximum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(greaterEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(less(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool3DBackprop, MaxPool3DBackpropAttrs, MaxPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d max pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank 4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param output The original output image, of rank 5 of shape\n *     [batchSize, outDepth, outHeight, outWidth, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike, output: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'maxPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPool3dBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let output5D = $output as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n    output5D = reshape($output, [\n      1, $output.shape[0], $output.shape[1], $output.shape[2], $output.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in maxPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in maxPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n  util.assert(\n      output5D.rank === 5,\n      () => `Error in maxPool3dBackprop: output must be rank 5 but got rank ` +\n          `${output5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.maxPool3dBackprop(dy5D, input5D, output5D, convInfo);\n  };\n\n  const inputs:\n      MaxPool3DBackpropInputs = {dy: dy5D, input: input5D, output: output5D};\n\n  const attrs: MaxPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      MaxPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3dBackprop = op({maxPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool3D, MaxPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPool3dBackprop} from '../ops/max_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const maxPool3DGradConfig: GradConfig = {\n  kernelName: MaxPool3D,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor5D, Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as MaxPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => maxPool3dBackprop(\n          dy as Tensor5D, x, y, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\n\n/**\n * Computes the backprop of a 2D max pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The original input image, of rank 4, of shape\n *     [batchSize, height, width, channels].\n * @param output The original output image, of rank 4, of shape\n *     [batchSize, outHeight, outWidth, channels].\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPoolBackprop_(\n    dy: Tensor4D|TensorLike, input: Tensor4D|TensorLike,\n    output: Tensor4D|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  const $dy = convertToTensor(dy, 'dy', 'maxPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPoolBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy ` +\n          `(${$dy.rank})`);\n\n  util.assert(\n      $dy.rank === 4,\n      () => `Error in maxPoolBackprop: dy must be rank 4 but got rank ` +\n          `${$dy.rank}.`);\n  util.assert(\n      $input.rank === 4,\n      () => `Error in maxPoolBackprop: input must be rank 4 but got rank ` +\n          `${$input.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPoolBackprop: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        $input.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    return backend.maxPoolBackprop($dy, $input, $output, convInfo);\n  };\n\n  const inputs:\n      MaxPoolBackpropInputs = {dy: $dy, input: $input, output: $output};\n\n  const attrs:\n      MaxPoolBackpropAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, MaxPoolBackprop,\n             attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const maxPoolBackprop = op({maxPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool, MaxPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPoolBackprop} from '../ops/max_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const maxPoolGradConfig: GradConfig = {\n  kernelName: MaxPool,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as MaxPoolAttrs;\n\n    return {\n      x: () => maxPoolBackprop(dy as Tensor4D, x, y, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Min, MinAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const minGradConfig: GradConfig = {\n  kernelName: Min,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const minAttrs: MinAttrs = attrs as {} as MinAttrs;\n    const {axis} = minAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(axis, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const minGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = minGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Minimum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const minimumGradConfig: GradConfig = {\n  kernelName: Minimum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(lessEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(greater(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Mod} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {floor} from '../ops/floor';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const modGradConfig: GradConfig = {\n  kernelName: Mod,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(dy, reduceAxes), a.shape);\n      }\n      return dy;\n    };\n    const derB = () => {\n      const res = mul(dy, neg(floor(div(a, b))));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Multiply} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const multiplyGradConfig: GradConfig = {\n  kernelName: Multiply,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = mul(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      const res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Negate} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {neg} from '../ops/neg';\nimport {Tensor} from '../tensor';\n\nexport const negateGradConfig: GradConfig = {\n  kernelName: Negate,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => neg(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OneHot} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\n\nexport const oneHotGradConfig: GradConfig = {\n  kernelName: OneHot,\n  inputsToSave: ['indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const indices = saved[0];\n    return {indices: () => zeros(indices.shape, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OnesLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const onesLikeGradConfig: GradConfig = {\n  kernelName: OnesLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PadV2, PadV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\nexport const padV2GradConfig: GradConfig = {\n  kernelName: PadV2,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    // Pad introduces values around the original tensor, so the gradient\n    // slices the original shape out of the gradient.\n    const x = saved[0];\n    const {paddings} = attrs as unknown as PadV2Attrs;\n    const begin = paddings.map(p => p[0]);\n    return {x: () => slice(dy, begin, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Pow} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {log} from '../ops/log';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {reshape} from '../ops/reshape';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const powGradConfig: GradConfig = {\n  kernelName: Pow,\n  inputsToSave: ['a', 'b'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b, y] = saved;\n    const base = a;\n    const exp = b;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(base.shape, exp.shape);\n\n    const derBase = () => {\n      const expFloat = cast(exp, 'float32');\n      let res = mul(dy, mul(expFloat, pow(base, sub(expFloat, scalar(1)))));\n      const reduceAxes = broadcast_util.getReductionAxes(base.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, base.shape);\n    };\n    const derExp = () => {\n      const condition = greater(base, 0);\n      const logBase = where(condition, log(base), zerosLike(base));\n      let res = mul(dy, mul(y, logBase));\n      const reduceAxes = broadcast_util.getReductionAxes(exp.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, exp.shape);\n    };\n    return {a: derBase, b: derExp};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Prelu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const preluGradConfig: GradConfig = {\n  kernelName: Prelu,\n  inputsToSave: ['x', 'alpha'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x, alpha] = saved;\n    const mask = greater(x, 0);\n\n    return {\n      x: () => where(mask, dy, mul(dy, alpha)),\n      alpha: () => {\n        let res = where(mask, zerosLike(dy), mul(dy, x));\n        const reduceAxes = getReductionAxes(alpha.shape, dy.shape);\n        if (reduceAxes.length > 0) {\n          res = sum(res, reduceAxes);\n        }\n        return reshape(res, alpha.shape);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Reciprocal} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const reciprocalGradConfig: GradConfig = {\n  kernelName: Reciprocal,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, neg(square(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu6} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const relu6GradConfig: GradConfig = {\n  kernelName: Relu6,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const mask = mul(lessEqual(x, 6), step(x));\n\n    return {x: () => mul(dy, cast(mask, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const reluGradConfig: GradConfig = {\n  kernelName: Relu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, cast(step(x), 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reshape} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\n\nexport const reshapeGradConfig: GradConfig = {\n  kernelName: Reshape,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => reshape(dy, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearGrad, ResizeBilinearGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeBilinearGradConfig: GradConfig = {\n  kernelName: ResizeBilinear,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeBilinearAttrs;\n      return backend.resizeBilinearBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeBilinearGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeBilinearGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborGrad, ResizeNearestNeighborGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeNearestNeighborGradConfig: GradConfig = {\n  kernelName: ResizeNearestNeighbor,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeNearestNeighborAttrs;\n      return backend.resizeNearestNeighborBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeNearestNeighborGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeNearestNeighborGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reverse, ReverseAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {reverse} from '../ops/reverse';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const reverseGradConfig: GradConfig = {\n  kernelName: Reverse,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dims} = attrs as {} as ReverseAttrs;\n    const axes = parseAxisParam(dims, dy.shape);\n    return {x: () => reverse(dy, axes)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Round} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const roundGradConfig: GradConfig = {\n  kernelName: Round,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rsqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {pow} from '../ops/pow';\nimport {Tensor} from '../tensor';\n\nexport const rsqrtGradConfig: GradConfig = {\n  kernelName: Rsqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => neg(div(dy, mul(pow(x, 1.5), 2)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SelectV2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {logicalNot} from '../ops/logical_not';\nimport {mul} from '../ops/mul';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const selectV2PoolGradConfig: GradConfig = {\n  kernelName: SelectV2,\n  inputsToSave: ['condition'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [condition] = saved;\n    return {\n      // TODO(julianoks): Return null for condition gradient\n      // when backprop supports it.\n      condition: () => cast(zerosLike(condition), 'float32'),\n      t: () => mul(dy, cast(condition, dy.dtype)),\n      e: () => mul(dy, cast(logicalNot(condition), dy.dtype))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Selu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {exp} from '../ops/exp';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {SELU_SCALE, SELU_SCALEALPHA} from '../ops/selu_util';\nimport {where} from '../ops/where';\nimport {Tensor} from '../tensor';\n\nexport const seluGradConfig: GradConfig = {\n  kernelName: Selu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {\n      x: () => {\n        const mask = greater(x, scalar(0));\n\n        const scaleAlpha = scalar(SELU_SCALEALPHA);\n        const scale = scalar(SELU_SCALE);\n\n        const greaterThanZeroDer = mul(dy, scale);\n        const lessEqualZeroDer =\n            mul(mul(dy, scaleAlpha), exp(cast(x, 'float32')));\n\n        return where(mask, greaterThanZeroDer, lessEqualZeroDer);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sigmoid} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const sigmoidGradConfig: GradConfig = {\n  kernelName: Sigmoid,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(dy, mul(y, sub(scalar(1), y)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sign} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const signGradConfig: GradConfig = {\n  kernelName: Sign,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cos} from '../ops/cos';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinGradConfig: GradConfig = {\n  kernelName: Sin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(cos(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cosh} from '../ops/cosh';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinhGradConfig: GradConfig = {\n  kernelName: Sinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(cosh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Slice, SliceAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {pad} from '../ops/pad';\nimport {parseSliceParams} from '../ops/slice_util';\nimport {Tensor} from '../tensor';\n\nexport const sliceGradConfig: GradConfig = {\n  kernelName: Slice,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {begin, size} = attrs as {} as SliceAttrs;\n\n    const inputShape = x.shape;\n    const [begin_, size_] = parseSliceParams(x, begin, size);\n\n    // Create an Nx2 padding where the first column represents how many\n    // zeros are prepended (at start) for each dimension, and the second\n    // column indicates how many zeros are appended (at end).\n\n    // The number of zeros to append is the shape of the input\n    // elementwise-subtracted by both the begin vector and sizes vector.\n    const paddings: Array<[number, number]> = [];\n    for (let i = 0; i < dy.rank; i++) {\n      paddings.push([begin_[i], inputShape[i] - begin_[i] - size_[i]]);\n    }\n    return {x: () => pad(dy, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softmax, SoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const softmaxGradConfig: GradConfig = {\n  kernelName: Softmax,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [y] = saved;\n    const {dim} = attrs as {} as SoftmaxAttrs;\n    const keepDims = true;\n\n    const dyTimesY = mul(dy, y);\n    return {\n      logits: () => sub(dyTimesY, mul(sum(dyTimesY, [dim], keepDims), y))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softplus} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sigmoid} from '../ops/sigmoid';\nimport {Tensor} from '../tensor';\n\nexport const softplusGradConfig: GradConfig = {\n  kernelName: Softplus,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(dy, sigmoid(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SpaceToBatchND, SpaceToBatchNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {batchToSpaceND} from '../ops/batch_to_space_nd';\nimport {Tensor} from '../tensor';\n\nexport const spaceToBatchNDGradConfig: GradConfig = {\n  kernelName: SpaceToBatchND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, paddings} = attrs as {} as SpaceToBatchNDAttrs;\n    return {x: () => batchToSpaceND(dy, blockShape, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {SplitV, SplitVAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {concat} from '../ops/concat';\nimport {Tensor} from '../tensor';\n\nexport const splitVGradConfig: GradConfig = {\n  kernelName: SplitV,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const {axis} = attrs as {} as SplitVAttrs;\n\n    return {x: () => concat(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {Tensor} from '../tensor';\n\nexport const sqrtGradConfig: GradConfig = {\n  kernelName: Sqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, mul(sqrt(cast(x, 'float32')), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const squareGradConfig: GradConfig = {\n  kernelName: Square,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, mul(cast(x, 'float32'), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SquaredDifference} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const squaredDifferenceGradConfig: GradConfig = {\n  kernelName: SquaredDifference,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const two = scalar(2);\n    const derA = () => mul(dy, mul(two, sub(a, b)));\n    const derB = () => mul(dy, mul(two, sub(b, a)));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Step} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const stepGradConfig: GradConfig = {\n  kernelName: Step,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports\n    // it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Sub} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const subGradConfig: GradConfig = {\n  kernelName: Sub,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(neg(res), b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sum, SumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {ones} from '../ops/ones';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const sumGradConfig: GradConfig = {\n  kernelName: Sum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const expandedDyShape = x.shape.slice();\n    const {axis} = attrs as {} as SumAttrs;\n\n    const axes = parseAxisParam(axis, x.shape);\n    axes.forEach(axis => {\n      expandedDyShape[axis] = 1;\n    });\n    const expandedDy = reshape(dy, expandedDyShape);\n    const derX = mul(expandedDy, ones(x.shape, 'float32'));\n\n    return {x: () => derX};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cos} from '../ops/cos';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const tanGradConfig: GradConfig = {\n  kernelName: Tan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, square(cos(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const tanhGradConfig: GradConfig = {\n  kernelName: Tanh,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(sub(scalar(1), square(y)), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tile, TileAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {slice} from '../ops/slice';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const tileGradConfig: GradConfig = {\n  kernelName: Tile,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {reps} = attrs as unknown as TileAttrs;\n\n    const derX = () => {\n      let xGrad = zerosLike(x);\n      // TODO(cais): Maybe reduce memory footprint by avoiding repeated\n      // slicing.\n      if (x.rank === 1) {\n        for (let i = 0; i < reps[0]; ++i) {\n          xGrad = add(xGrad, slice(dy, [i * x.shape[0]], [x.shape[0]]));\n        }\n      } else if (x.rank === 2) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            xGrad = add(xGrad, slice(dy, [i * x.shape[0], j * x.shape[1]], [\n                          x.shape[0], x.shape[1]\n                        ]));\n          }\n        }\n      } else if (x.rank === 3) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              xGrad =\n                  add(xGrad,\n                      slice(\n                          dy, [i * x.shape[0], j * x.shape[1], k * x.shape[2]],\n                          [x.shape[0], x.shape[1], x.shape[2]]));\n            }\n          }\n        }\n      } else if (x.rank === 4) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              for (let l = 0; l < reps[3]; ++l) {\n                xGrad =\n                    add(xGrad,\n                        slice(\n                            dy,\n                            [\n                              i * x.shape[0], j * x.shape[1], k * x.shape[2],\n                              l * x.shape[3]\n                            ],\n                            [x.shape[0], x.shape[1], x.shape[2], x.shape[3]]));\n              }\n            }\n          }\n        }\n      } else {\n        throw new Error(\n            `Gradient for tile operation is not implemented for rank-` +\n            `${x.rank} tensors yet.`);\n      }\n      return xGrad;\n    };\n    return {x: derX};\n  },\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Transpose, TransposeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const transposeGradConfig: GradConfig = {\n  kernelName: Transpose,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const transposeAttrs: TransposeAttrs = attrs as {} as TransposeAttrs;\n    const {perm} = transposeAttrs;\n    const undoPerm = axis_util.getUndoAxesPermutation(perm);\n    return {x: () => transpose(dy, undoPerm)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Unpack, UnpackAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {stack} from '../ops/stack';\nimport {Tensor} from '../tensor';\n\nexport const unpackGradConfig: GradConfig = {\n  kernelName: Unpack,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const unpackAttrs: UnpackAttrs = attrs as {} as UnpackAttrs;\n    const {axis} = unpackAttrs;\n    return {value: () => stack(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {UnsortedSegmentSum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {expandDims} from '../ops/expand_dims';\nimport {gather} from '../ops/gather';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {maximum} from '../ops/maximum';\nimport {ones} from '../ops/ones';\nimport {scalar} from '../ops/scalar';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor, Tensor1D} from '../tensor';\n\nexport const unsortedSegmentSumGradConfig: GradConfig = {\n  kernelName: UnsortedSegmentSum,\n  inputsToSave: ['segmentIds'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [segmentIds] = saved;\n\n    const derX = () => {\n      return gatherDropNegatives(dy, segmentIds as Tensor1D);\n    };\n    return {x: derX};\n  }\n};\n\nfunction gatherDropNegatives<T extends Tensor>(x: T, indices: Tensor1D) {\n  // Helper function for unsorted segment ops. Gathers params for\n  // positive segment ids and gathers 0 for inputs with negative segment id.\n  // Mirrors _GatherDropNegatives from tensorflow/python/ops/math_grad.py\n  const zeroClippedIndices = maximum(indices, zerosLike(indices));\n  const gathered = gather(x, zeroClippedIndices as Tensor1D);\n  let isPositive = greaterEqual(indices, scalar(0, 'int32'));\n  const numIters = gathered.rank - isPositive.rank;\n  for (let i = 0; i < numIters; ++i) {\n    isPositive = expandDims(isPositive, i + 1);\n  }\n  isPositive = logicalAnd(isPositive, ones(gathered.shape, 'bool'));\n  const zeroSlice = zerosLike(gathered);\n  return where(isPositive, gathered, zeroSlice);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ZerosLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const zerosLikeGradConfig: GradConfig = {\n  kernelName: ZerosLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {absGradConfig} from './gradients/Abs_grad';\nimport {acosGradConfig} from './gradients/Acos_grad';\nimport {acoshGradConfig} from './gradients/Acosh_grad';\nimport {addGradConfig} from './gradients/Add_grad';\nimport {addNGradConfig} from './gradients/AddN_grad';\nimport {argMaxGradConfig} from './gradients/ArgMax_grad';\nimport {argMinGradConfig} from './gradients/ArgMin_grad';\nimport {asinGradConfig} from './gradients/Asin_grad';\nimport {asinhGradConfig} from './gradients/Asinh_grad';\nimport {atan2GradConfig} from './gradients/Atan2_grad';\nimport {atanGradConfig} from './gradients/Atan_grad';\nimport {atanhGradConfig} from './gradients/Atanh_grad';\nimport {avgPool3DGradConfig} from './gradients/AvgPool3D_grad';\nimport {avgPoolGradConfig} from './gradients/AvgPool_grad';\nimport {batchMatMulGradConfig} from './gradients/BatchMatMul_grad';\nimport {batchToSpaceNDGradConfig} from './gradients/BatchToSpaceND_grad';\nimport {broadcastToGradConfig} from './gradients/BroadcastTo_grad';\nimport {castGradConfig} from './gradients/Cast_grad';\nimport {ceilGradConfig} from './gradients/Ceil_grad';\nimport {clipByValueGradConfig} from './gradients/ClipByValue_grad';\nimport {concatGradConfig} from './gradients/Concat_grad';\nimport {conv2DGradConfig} from './gradients/Conv2D_grad';\nimport {conv2DBackpropInputGradConfig} from './gradients/Conv2DBackpropInput_grad';\nimport {conv3DGradConfig} from './gradients/Conv3D_grad';\nimport {cosGradConfig} from './gradients/Cos_grad';\nimport {coshGradConfig} from './gradients/Cosh_grad';\nimport {cumsumGradConfig} from './gradients/Cumsum_grad';\nimport {depthwiseConv2dNativeGradConfig} from './gradients/DepthwiseConv2dNative_grad';\nimport {dilation2dGradConfig} from './gradients/Dilation2D_grad';\nimport {divGradConfig} from './gradients/Div_grad';\nimport {eluGradConfig} from './gradients/Elu_grad';\nimport {erfGradConfig} from './gradients/Erf_grad';\nimport {expGradConfig} from './gradients/Exp_grad';\nimport {expm1GradConfig} from './gradients/Expm1_grad';\nimport {floorGradConfig} from './gradients/Floor_grad';\nimport {floorDivGradConfig} from './gradients/FloorDiv_grad';\nimport {fusedBatchNormGradConfig} from './gradients/FusedBatchNorm_grad';\nimport {gatherGradConfig} from './gradients/GatherV2_grad';\nimport {greaterEqualGradConfig} from './gradients/GreaterEqual_grad';\nimport {identityGradConfig} from './gradients/Identity_grad';\nimport {isFiniteGradConfig} from './gradients/IsFinite_grad';\nimport {isInfGradConfig} from './gradients/IsInf_grad';\nimport {isNanGradConfig} from './gradients/IsNan_grad';\nimport {log1pGradConfig} from './gradients/Log1p_grad';\nimport {logGradConfig} from './gradients/Log_grad';\nimport {logSoftmaxGradConfig} from './gradients/LogSoftmax_grad';\nimport {lrnGradConfig} from './gradients/LRN_grad';\nimport {maxGradConfig} from './gradients/Max_grad';\nimport {maximumGradConfig} from './gradients/Maximum_grad';\nimport {maxPool3DGradConfig} from './gradients/MaxPool3D_grad';\nimport {maxPoolGradConfig} from './gradients/MaxPool_grad';\nimport {minGradConfig} from './gradients/Min_grad';\nimport {minimumGradConfig} from './gradients/Minimum_grad';\nimport {modGradConfig} from './gradients/Mod_grad';\nimport {multiplyGradConfig} from './gradients/Multiply_grad';\nimport {negateGradConfig} from './gradients/Negate_grad';\nimport {oneHotGradConfig} from './gradients/OneHot_grad';\nimport {onesLikeGradConfig} from './gradients/OnesLike_grad';\nimport {padV2GradConfig} from './gradients/PadV2_grad';\nimport {powGradConfig} from './gradients/Pow_grad';\nimport {preluGradConfig} from './gradients/Prelu_grad';\nimport {reciprocalGradConfig} from './gradients/Reciprocal_grad';\nimport {relu6GradConfig} from './gradients/Relu6_grad';\nimport {reluGradConfig} from './gradients/Relu_grad';\nimport {reshapeGradConfig} from './gradients/Reshape_grad';\nimport {resizeBilinearGradConfig} from './gradients/ResizeBilinear_grad';\nimport {resizeNearestNeighborGradConfig} from './gradients/ResizeNearestNeighbor_grad';\nimport {reverseGradConfig} from './gradients/Reverse_grad';\nimport {roundGradConfig} from './gradients/Round_grad';\nimport {rsqrtGradConfig} from './gradients/Rsqrt_grad';\nimport {selectV2PoolGradConfig} from './gradients/SelectV2_grad';\nimport {seluGradConfig} from './gradients/Selu_grad';\nimport {sigmoidGradConfig} from './gradients/Sigmoid_grad';\nimport {signGradConfig} from './gradients/Sign_grad';\nimport {sinGradConfig} from './gradients/Sin_grad';\nimport {sinhGradConfig} from './gradients/Sinh_grad';\nimport {sliceGradConfig} from './gradients/Slice_grad';\nimport {softmaxGradConfig} from './gradients/Softmax_grad';\nimport {softplusGradConfig} from './gradients/Softplus_grad';\nimport {spaceToBatchNDGradConfig} from './gradients/SpaceToBatchND_grad';\nimport {splitVGradConfig} from './gradients/SplitV_grad';\nimport {sqrtGradConfig} from './gradients/Sqrt_grad';\nimport {squareGradConfig} from './gradients/Square_grad';\nimport {squaredDifferenceGradConfig} from './gradients/SquaredDifference_grad';\nimport {stepGradConfig} from './gradients/Step_grad';\nimport {subGradConfig} from './gradients/Sub_grad';\nimport {sumGradConfig} from './gradients/Sum_grad';\nimport {tanGradConfig} from './gradients/Tan_grad';\nimport {tanhGradConfig} from './gradients/Tanh_grad';\nimport {tileGradConfig} from './gradients/Tile_grad';\nimport {transposeGradConfig} from './gradients/Transpose_grad';\nimport {unpackGradConfig} from './gradients/Unpack_grad';\nimport {unsortedSegmentSumGradConfig} from './gradients/UnsortedSegmentSum_grad';\nimport {zerosLikeGradConfig} from './gradients/ZerosLike_grad';\nimport {GradConfig} from './kernel_registry';\nimport {registerGradient} from './kernel_registry';\n\n// Export all kernel configs here so that the package can auto register them\nconst gradConfigs: GradConfig[] = [\n  absGradConfig,\n  acosGradConfig,\n  acoshGradConfig,\n  addGradConfig,\n  addNGradConfig,\n  argMaxGradConfig,\n  argMinGradConfig,\n  asinGradConfig,\n  asinhGradConfig,\n  atan2GradConfig,\n  atanGradConfig,\n  atanhGradConfig,\n  avgPool3DGradConfig,\n  avgPoolGradConfig,\n  batchMatMulGradConfig,\n  batchToSpaceNDGradConfig,\n  broadcastToGradConfig,\n  castGradConfig,\n  ceilGradConfig,\n  clipByValueGradConfig,\n  concatGradConfig,\n  conv2DBackpropInputGradConfig,\n  conv2DGradConfig,\n  conv3DGradConfig,\n  cosGradConfig,\n  coshGradConfig,\n  cumsumGradConfig,\n  depthwiseConv2dNativeGradConfig,\n  dilation2dGradConfig,\n  divGradConfig,\n  eluGradConfig,\n  erfGradConfig,\n  expGradConfig,\n  expm1GradConfig,\n  floorDivGradConfig,\n  floorGradConfig,\n  fusedBatchNormGradConfig,\n  gatherGradConfig,\n  greaterEqualGradConfig,\n  identityGradConfig,\n  isFiniteGradConfig,\n  isInfGradConfig,\n  isNanGradConfig,\n  log1pGradConfig,\n  logGradConfig,\n  logSoftmaxGradConfig,\n  lrnGradConfig,\n  maxGradConfig,\n  maxGradConfig,\n  maximumGradConfig,\n  maxPool3DGradConfig,\n  maxPoolGradConfig,\n  minGradConfig,\n  minimumGradConfig,\n  modGradConfig,\n  multiplyGradConfig,\n  negateGradConfig,\n  oneHotGradConfig,\n  onesLikeGradConfig,\n  padV2GradConfig,\n  padV2GradConfig,\n  powGradConfig,\n  preluGradConfig,\n  reciprocalGradConfig,\n  relu6GradConfig,\n  reluGradConfig,\n  reshapeGradConfig,\n  resizeBilinearGradConfig,\n  resizeNearestNeighborGradConfig,\n  reverseGradConfig,\n  roundGradConfig,\n  rsqrtGradConfig,\n  selectV2PoolGradConfig,\n  seluGradConfig,\n  sigmoidGradConfig,\n  signGradConfig,\n  sinGradConfig,\n  sinhGradConfig,\n  sliceGradConfig,\n  softmaxGradConfig,\n  softplusGradConfig,\n  spaceToBatchNDGradConfig,\n  spaceToBatchNDGradConfig,\n  splitVGradConfig,\n  splitVGradConfig,\n  sqrtGradConfig,\n  squaredDifferenceGradConfig,\n  squareGradConfig,\n  stepGradConfig,\n  subGradConfig,\n  sumGradConfig,\n  tanGradConfig,\n  tanhGradConfig,\n  tileGradConfig,\n  transposeGradConfig,\n  unpackGradConfig,\n  unsortedSegmentSumGradConfig,\n  zerosLikeGradConfig\n];\n\nfor (const gradientConfig of gradConfigs) {\n  registerGradient(gradientConfig);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {abs} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    abs<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.abs = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return abs(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acos = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acosh = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {addStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    addStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.addStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return addStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {add} from '../../ops/add';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    add<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.add = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return add(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {all} from '../../ops/all';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    all<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.all = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return all(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {any} from '../../ops/any';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    any<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.any = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return any(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMax} from '../../ops/arg_max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMax<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMax = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return argMax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMin} from '../../ops/arg_min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMin<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMin = function<T extends Tensor>(axis: number): T {\n  this.throwIfDisposed();\n  return argMin(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Scalar, Tensor} from '../../tensor';\nimport {Rank} from '../../types';\nimport {assert} from '../../util';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asScalar<T extends Tensor>(): Scalar;\n  }\n}\n\n/** Converts a size-1 `tf.Tensor` to a `tf.Scalar`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asScalar = function<T extends Tensor>(this: T): Scalar {\n  this.throwIfDisposed();\n  assert(this.size === 1, () => 'The array must have only 1 element.');\n  return reshape(this, []);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asType<T extends Tensor>(this: T, dtype: DataType): T;\n  }\n}\n\n/**\n * Casts a `tf.Tensor` to a specified dtype.\n *\n * @param dtype Data-type to cast the tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asType = function<T extends Tensor>(\n    this: T, dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast<T>(this, dtype);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as1D<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Converts a `tf.Tensor` to a `tf.Tensor1D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as1D = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as2D<T extends Tensor>(rows: number, columns: number): Tensor2D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor2D`.\n *\n * @param rows Number of rows in `tf.Tensor2D`.\n * @param columns Number of columns in `tf.Tensor2D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as2D = function<T extends Tensor>(\n    rows: number, columns: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as3D<T extends Tensor>(rows: number, columns: number, depth: number):\n        Tensor3D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor3D`.\n *\n * @param rows Number of rows in `tf.Tensor3D`.\n * @param columns Number of columns in `tf.Tensor3D`.\n * @param depth Depth of `tf.Tensor3D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as3D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as4D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number): Tensor4D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor4D`.\n *\n * @param rows Number of rows in `tf.Tensor4D`.\n * @param columns Number of columns in `tf.Tensor4D`.\n * @param depth Depth of `tf.Tensor4D`.\n * @param depth2 4th dimension of `tf.Tensor4D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as4D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as5D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number,\n        depth3: number): Tensor5D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor5D`.\n *\n * @param rows Number of rows in `tf.Tensor5D`.\n * @param columns Number of columns in `tf.Tensor5D`.\n * @param depth Depth of `tf.Tensor5D`.\n * @param depth2 4th dimension of `tf.Tensor5D`.\n * @param depth3 5th dimension of 'tf.Tensor5D'\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as5D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number,\n    depth3: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2, depth3]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {atan2} from '../../ops/atan2';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan2<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.atan2 = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return atan2(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {avgPool} from '../../ops/avg_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    avgPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.avgPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return avgPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchToSpaceND} from '../../ops/batch_to_space_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchToSpaceND<R extends Rank>(blockShape: number[], crops: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.batchToSpaceND = function<R extends Rank>(\n    blockShape: number[], crops: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return batchToSpaceND(this, blockShape, crops);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchNorm} from '../../ops/batchnorm';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchNorm<T extends Tensor>(\n        mean: Tensor<R>|Tensor1D|TensorLike,\n        variance: Tensor<R>|Tensor1D|TensorLike,\n        offset?: Tensor<R>|Tensor1D|TensorLike,\n        scale?: Tensor<R>|Tensor1D|TensorLike,\n        varianceEpsilon?: number): Tensor<R>;\n  }\n}\n\nTensor.prototype.batchNorm = function<R extends Rank>(\n    mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  this.throwIfDisposed();\n  return batchNorm(this, mean, variance, offset, scale, varianceEpsilon);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {broadcastTo} from '../../ops/broadcast_to';\nimport {Tensor} from '../../tensor';\nimport {Rank, ShapeMap} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    broadcastTo<R extends Rank>(shape: ShapeMap[R]): Tensor<R>;\n  }\n}\n\nTensor.prototype.broadcastTo = function<R extends Rank>(shape: ShapeMap[R]):\n    Tensor<R> {\n  this.throwIfDisposed();\n  return broadcastTo(this, shape);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cast<T extends Tensor>(dtype: DataType): T;\n  }\n}\n\nTensor.prototype.cast = function<T extends Tensor>(dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast(this, dtype) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ceil} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ceil<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.ceil = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return ceil(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {clipByValue} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    clipByValue<T extends Tensor>(min: number, max: number): Tensor;\n  }\n}\n\nTensor.prototype.clipByValue = function<T extends Tensor>(\n    min: number, max: number): T {\n  this.throwIfDisposed();\n  return clipByValue(this, min, max) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {concat} from '../../ops/concat';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    concat<T extends Tensor>(tensors: T|Array<T|TensorLike>, axis?: number): T;\n  }\n}\n\nTensor.prototype.concat = function<T extends Tensor>(\n    x: T|Array<T|TensorLike>, axis?: number): T {\n  this.throwIfDisposed();\n  if (x instanceof Tensor) {\n    x = [x];\n  }\n  return concat([this, ...x], axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv1d} from '../../ops/conv1d';\nimport {ExplicitPadding} from '../../ops/conv_util';\nimport {Tensor, Tensor2D, Tensor3D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv1d<T extends Tensor2D|Tensor3D>(\n        filter: Tensor3D|TensorLike3D, stride: number,\n        pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n        dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv1d = function<T extends Tensor2D|Tensor3D>(\n    filter: Tensor3D|TensorLike3D, stride: number,\n    pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n    dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv1d(\n             this, filter, stride, pad, dataFormat, dilation,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2dTranspose} from '../../ops/conv2d_transpose';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2dTranspose<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D,\n        outputShape: [number, number, number, number]|[number, number, number],\n        strides: [number, number]|number, pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2dTranspose = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2dTranspose(\n             this, filter, outputShape, strides, pad, dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2d} from '../../ops/conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cos = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cosh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {cumsum} from '../../ops/cumsum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cumsum<R extends Rank>(\n        axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R>;\n  }\n}\n\nTensor.prototype.cumsum = function<R extends Rank>(\n    axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R> {\n  this.throwIfDisposed();\n  return cumsum(this, axis, exclusive, reverse);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthToSpace} from '../../ops/depth_to_space';\nimport {Tensor, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthToSpace<T extends Tensor4D>(\n        blockSize: number, dataFormat: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.depthToSpace = function<T extends Tensor4D>(\n    blockSize: number, dataFormat: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return depthToSpace(this, blockSize, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../../globals';\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2D<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\n/**\n * @deprecated Use `depthwiseConv2d` instead.\n */\nTensor.prototype.depthwiseConv2D = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  deprecationWarn('depthwiseConv2D is deprecated, use depthwiseConv2d instead');\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.depthwiseConv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dilation2d} from '../../ops/dilation2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dilation2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n        pad: 'valid'|'same', dilations?: [number, number]|number,\n        dataFormat?: 'NHWC'): T;\n  }\n}\n\nTensor.prototype.dilation2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n    pad: 'valid'|'same', dilations?: [number, number]|number,\n    dataFormat?: 'NHWC'): T {\n  this.throwIfDisposed();\n  return dilation2d(this, filter, strides, pad, dilations, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {divNoNan} from '../../ops/div_no_nan';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divNoNan<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.divNoNan = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return divNoNan(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {divStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.divStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return divStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {div} from '../../ops/div';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    div<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.div = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return div(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dot} from '../../ops/dot';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dot<T extends Tensor>(b: Tensor|TensorLike): Tensor;\n  }\n}\n\nTensor.prototype.dot = function<T extends Tensor>(b: T|TensorLike): Tensor {\n  this.throwIfDisposed();\n  return dot(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {elu} from '../../ops/elu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    elu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.elu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return elu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {equalStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equalStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.equalStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return equalStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {equal} from '../../ops/equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equal<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.equal = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return equal(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {erf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    erf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.erf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return erf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {exp} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    exp<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.exp = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return exp(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {expandDims} from '../../ops/expand_dims';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expandDims<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.expandDims = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return expandDims(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {expm1} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expm1<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.expm1 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return expm1(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {fft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    fft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.fft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return fft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    flatten<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Flatten a Tensor to a 1D array.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.flatten = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {floor} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floor<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.floor = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return floor(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {floorDiv} from '../../ops/floorDiv';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floorDiv<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.floorDiv = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return floorDiv(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {gather} from '../../ops/gather';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    gather<T extends Tensor>(\n        this: T, indices: Tensor|TensorLike, axis?: number): T;\n  }\n}\n\nTensor.prototype.gather = function<T extends Tensor>(\n    this: T, indices: Tensor|TensorLike, axis?: number): T {\n  this.throwIfDisposed();\n  return gather(this, indices, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return greaterEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greaterEqual} from '../../ops/greater_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greaterEqual = function<T extends Tensor>(b: Tensor|\n                                                           TensorLike): T {\n  this.throwIfDisposed();\n  return greaterEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return greaterStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greater} from '../../ops/greater';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greater<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greater = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return greater(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ifft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ifft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.ifft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return ifft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {irfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    irfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.irfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return irfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isFinite} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isFinite<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isFinite = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isFinite(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isInf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isInf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isInf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isInf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isNaN} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isNaN<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isNaN = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isNaN(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {leakyRelu} from '../../ops/leaky_relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    leakyRelu<T extends Tensor>(alpha: number): T;\n  }\n}\n\nTensor.prototype.leakyRelu = function<T extends Tensor>(\n    this: T, alpha: number): T {\n  this.throwIfDisposed();\n  return leakyRelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.lessEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {lessEqual} from '../../ops/less_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessEqual = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return lessEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {less} from '../../ops/less';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    less<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.less = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return less(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {localResponseNormalization} from '../../ops/local_response_normalization';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    localResponseNormalization<T extends Tensor>(\n        depthRadius?: number, bias?: number, alpha?: number, beta?: number): T;\n  }\n}\n\nTensor.prototype.localResponseNormalization = function<T extends Tensor>(\n    depthRadius?: number, bias?: number, alpha?: number, beta?: number): T {\n  this.throwIfDisposed();\n  return localResponseNormalization(this, depthRadius, bias, alpha, beta) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.logSigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return logSigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSoftmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSoftmax<T extends Tensor>(this: T, axis?: number): T;\n  }\n}\n\nTensor.prototype.logSoftmax = function<T extends Tensor>(\n    this: T, axis?: number): T {\n  this.throwIfDisposed();\n  return logSoftmax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logSumExp} from '../../ops/log_sum_exp';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSumExp<T extends Tensor>(\n        this: T, axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.logSumExp = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return logSumExp(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log1p} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log1p<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log1p = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log1p(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalAnd} from '../../ops/logical_and';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalAnd<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalAnd = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalAnd(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalNot} from '../../ops/logical_not';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalNot<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.logicalNot = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return logicalNot(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalOr} from '../../ops/logical_or';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalOr<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalOr = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return logicalOr(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalXor} from '../../ops/logical_xor';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalXor<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalXor = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalXor(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {matMul} from '../../ops/mat_mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    matMul<T extends Tensor>(\n        b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T;\n  }\n}\n\nTensor.prototype.matMul = function<T extends Tensor>(\n    this: T, b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T {\n  this.throwIfDisposed();\n  return matMul(this, b, transposeA, transposeB);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maxPool} from '../../ops/max_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maxPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.maxPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return maxPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {max} from '../../ops/max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    max<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.max = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return max(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {maximumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.maximumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return maximumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maximum} from '../../ops/maximum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.maximum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return maximum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mean} from '../../ops/mean';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mean<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.mean = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return mean(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {min} from '../../ops/min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    min<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.min = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return min(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {minimumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.minimumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return minimumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {minimum} from '../../ops/minimum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.minimum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return minimum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {modStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    modStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.modStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return modStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mod} from '../../ops/mod';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mod<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mod = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mod(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {mulStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mulStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.mulStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return mulStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mul} from '../../ops/mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mul<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mul = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mul(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {neg} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    neg<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.neg = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return neg(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {norm} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    norm<T extends Tensor>(\n        ord?: number|'euclidean'|'fro', axis?: number|number[],\n        keepDims?: boolean): Tensor;\n  }\n}\n\nTensor.prototype.norm = function<T extends Tensor>(\n    ord?: number|'euclidean'|'fro', axis?: number|number[],\n    keepDims?: boolean) {\n  this.throwIfDisposed();\n  return norm(this, ord, axis, keepDims) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {notEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.notEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return notEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {notEqual} from '../../ops/not_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.notEqual = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return notEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {oneHot} from '../../ops/one_hot';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    oneHot(depth: number, onValue: number, offValue: number): Tensor;\n  }\n}\n\nTensor.prototype.oneHot = function(\n    depth: number, onValue = 1, offValue = 0): Tensor {\n  this.throwIfDisposed();\n  return oneHot(this, depth, onValue, offValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {onesLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    onesLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.onesLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return onesLike(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {pad} from '../../ops/pad';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pad<T extends Tensor>(\n        paddings: Array<[number, number]>, constantValue?: number): T;\n  }\n}\n\nTensor.prototype.pad = function<T extends Tensor>(\n    this: T, paddings: Array<[number, number]>, constantValue: number): T {\n  this.throwIfDisposed();\n  return pad(this, paddings, constantValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pool} from '../../ops/pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pool<T extends Tensor3D|Tensor4D>(\n        windowShape: [number, number]|number, poolingType: 'avg'|'max',\n        padding: 'valid'|'same'|number, diationRate?: [number, number]|number,\n        strides?: [number, number]|number): T;\n  }\n}\n\nTensor.prototype.pool = function<T extends Tensor3D|Tensor4D>(\n    this: T, windowShape: [number, number]|number, poolingType: 'max'|'avg',\n    padding: 'valid'|'same'|number, dilationRate?: [number, number]|number,\n    strides?: [number, number]|number): T {\n  this.throwIfDisposed();\n  return pool(this, windowShape, poolingType, padding, dilationRate, strides);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {powStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    powStrict<T extends Tensor>(exp: Tensor): Tensor;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.powStrict = function<T extends Tensor>(exp: Tensor): T {\n  this.throwIfDisposed();\n  return powStrict(this, exp) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pow} from '../../ops/pow';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pow<T extends Tensor>(exp: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.pow = function<T extends Tensor>(exp: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return pow(this, exp);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prelu} from '../../ops/prelu';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prelu<T extends Tensor>(alpha: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.prelu = function<T extends Tensor>(\n    this: T, alpha: T|TensorLike): T {\n  this.throwIfDisposed();\n  return prelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prod} from '../../ops/prod';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prod<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.prod = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return prod(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {reciprocal} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reciprocal<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.reciprocal = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return reciprocal(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu} from '../../ops/relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu6} from '../../ops/relu6';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu6<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu6 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu6(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshapeAs<T extends Tensor>(x: T): T;\n  }\n}\n\n/**\n * Reshapes the tensor into the shape of the provided tensor.\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.reshapeAs = function<T extends Tensor>(x: T): T {\n  this.throwIfDisposed();\n  return reshape(this, x.shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshape<T extends Tensor>(shape: number[]): T;\n  }\n}\n\nTensor.prototype.reshape = function<T extends Tensor>(shape: number[]): T {\n  this.throwIfDisposed();\n  return reshape(this, shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeBilinear} from '../../ops/image/resize_bilinear';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeBilinear<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeBilinear = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeBilinear(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeNearestNeighbor} from '../../ops/image/resize_nearest_neighbor';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeNearestNeighbor<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeNearestNeighbor = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeNearestNeighbor(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reverse} from '../../ops/reverse';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reverse<T extends Tensor>(this: T, axis?: number|number[]): T;\n  }\n}\n\nTensor.prototype.reverse = function<T extends Tensor>(\n    this: T, axis?: number|number[]): T {\n  this.throwIfDisposed();\n  return reverse(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.rfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return rfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {round} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    round<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.round = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return round(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rsqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rsqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.rsqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return rsqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {selu} from '../../ops/selu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    selu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.selu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return selu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {separableConv2d} from '../../ops/separable_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    separableConv2d<T extends Tensor3D|Tensor4D>(\n        depthwiseFilter: Tensor4D|TensorLike4D,\n        pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n        pad: 'valid'|'same', dilation?: [number, number]|number,\n        dataFormat?: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.separableConv2d = function<T extends Tensor3D|Tensor4D>(\n    depthwiseFilter: Tensor4D|TensorLike4D,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation?: [number, number]|number,\n    dataFormat?: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return separableConv2d(\n             this, depthwiseFilter, pointwiseFilter, strides, pad, dilation,\n             dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sign} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sign<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sign = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sign(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {slice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    slice<T extends Tensor>(\n        this: T, begin: number|number[], size?: number|number[]): T;\n  }\n}\n\nTensor.prototype.slice = function<T extends Tensor>(\n    this: T, begin: number|number[], size?: number|number[]): T {\n  this.throwIfDisposed();\n  return slice(this, begin, size);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softmax<T extends Tensor>(this: T, dim?: number): T;\n  }\n}\n\nTensor.prototype.softmax = function<T extends Tensor>(this: T, dim: number): T {\n  this.throwIfDisposed();\n  return softmax(this, dim);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softplus} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softplus<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.softplus = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return softplus(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {spaceToBatchND} from '../../ops/space_to_batch_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    spaceToBatchND<R extends Rank>(blockShape: number[], paddings: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.spaceToBatchND = function<R extends Rank>(\n    blockShape: number[], paddings: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return spaceToBatchND(this, blockShape, paddings);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {split} from '../../ops/split';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    split<T extends Tensor>(numOrSizeSplits: number[]|number, axis?: number):\n        T[];\n  }\n}\n\nTensor.prototype.split = function<T extends Tensor>(\n    numOrSizeSplits: number[]|number, axis?: number): T[] {\n  this.throwIfDisposed();\n  return split(this, numOrSizeSplits, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {square} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    square<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.square = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return square(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {squaredDifference} from '../../ops/squared_difference';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifference<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.squaredDifference = function<T extends Tensor>(b: Tensor|\n                                                                TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifference(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {squaredDifferenceStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifferenceStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.squaredDifferenceStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifferenceStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {squeeze} from '../../ops/squeeze';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squeeze<T extends Tensor>(axis?: number[]): T;\n  }\n}\n\nTensor.prototype.squeeze = function<T extends Tensor>(axis?: number[]): T {\n  this.throwIfDisposed();\n  return squeeze(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {stack} from '../../ops/stack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stack<T extends Tensor>(x: Tensor|Tensor[], axis?: number): T;\n  }\n}\n\nTensor.prototype.stack = function<T extends Tensor>(\n    x: Tensor|Tensor[], axis?: number): T {\n  this.throwIfDisposed();\n  const tensorsToBeStacked = x instanceof Tensor ? [this, x] : [this, ...x];\n  return stack(tensorsToBeStacked, axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {step} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    step<T extends Tensor>(this: T, alpha?: number): T;\n  }\n}\n\nTensor.prototype.step = function<T extends Tensor>(this: T, alpha?: number) {\n  this.throwIfDisposed();\n  return step(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {stridedSlice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stridedSlice<T extends Tensor>(\n        this: Tensor, begin: number[], end: number[], strides: number[],\n        beginMask?: number, endMask?: number, ellipsisMask?: number,\n        newAxisMask?: number, shrinkAxisMask?: number): Tensor;\n  }\n}\n\nTensor.prototype.stridedSlice = function<T extends Tensor>(\n    this: Tensor, begin: number[], end: number[], strides: number[],\n    beginMask?: number, endMask?: number, ellipsisMask?: number,\n    newAxisMask?: number, shrinkAxisMask?: number): T {\n  this.throwIfDisposed();\n  return stridedSlice(\n             this, begin, end, strides, beginMask, endMask, ellipsisMask,\n             newAxisMask, shrinkAxisMask) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {subStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    subStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.subStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return subStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sub} from '../../ops/sub';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sub<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.sub = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return sub(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sum} from '../../ops/sum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sum<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.sum = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return sum(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tile} from '../../ops/tile';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tile<T extends Tensor>(b: number[]): T;\n  }\n}\n\nTensor.prototype.tile = function<T extends Tensor>(reps: number[]): T {\n  this.throwIfDisposed();\n  return tile(this, reps) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toBool<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `bool`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toBool = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'bool');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toFloat<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `float32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toFloat = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'float32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toInt<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `int32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toInt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'int32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {topk} from '../../ops/topk';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    topk<T extends Tensor>(this: T, k?: number, sorted?: boolean):\n        {values: T, indices: T};\n  }\n}\n\nTensor.prototype.topk = function<T extends Tensor>(\n    this: T, k?: number, sorted?: boolean): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return topk(this, k, sorted) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {transpose} from '../../ops/transpose';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    transpose<T extends Tensor>(perm?: number[]): T;\n  }\n}\n\nTensor.prototype.transpose = function<T extends Tensor>(\n    this: T, perm?: number[]): T {\n  this.throwIfDisposed();\n  return transpose(this, perm);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unique} from '../../ops/unique';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unique<T extends Tensor>(this: T, axis?: number): {values: T, indices: T};\n  }\n}\n\nTensor.prototype.unique = function<T extends Tensor>(\n    this: T, axis?: number): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return unique(this, axis) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unsortedSegmentSum} from '../../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike1D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unsortedSegmentSum<T extends Tensor>(\n        this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T;\n  }\n}\n\nTensor.prototype.unsortedSegmentSum = function<T extends Tensor>(\n    this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T {\n  this.throwIfDisposed();\n  return unsortedSegmentSum(this, segmentIds, numSegments);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {unstack} from '../../ops/unstack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unstack<T extends Tensor>(axis?: number): T[];\n  }\n}\n\nTensor.prototype.unstack = function<T extends Tensor>(axis?: number): T[] {\n  this.throwIfDisposed();\n  return unstack(this, axis) as T[];\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {where} from '../../ops/where';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    where<T extends Tensor>(condition: Tensor|TensorLike, x: Tensor|TensorLike):\n        T;\n  }\n}\n\nTensor.prototype.where = function<T extends Tensor>(\n    condition: Tensor|TensorLike, x: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return where(condition, this, x) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {zerosLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    zerosLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.zerosLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return zerosLike(this);\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {backend} from '@tensorflow/tfjs-core';\nimport {DataFormat} from '../keras_format/common';\n\nlet _epsilon: number;\n\n/**\n * Returns the value of the fuzz factor used in numeric expressions.\n */\nexport function epsilon() {\n  if (_epsilon == null) {\n    _epsilon = backend().epsilon();\n  }\n  return _epsilon;\n}\n\n/**\n * Sets the value of the fuzz factor used in numeric expressions.\n * @param e New value of epsilon.\n */\nexport function setEpsilon(e: number) {\n  _epsilon = e;\n}\n\n/**\n * Returns the default image data format convention.\n */\nexport function imageDataFormat(): DataFormat {\n  return 'channelsLast';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Explicit error types.\n *\n * See the following link for more information about why the code includes\n * calls to setPrototypeOf:\n *\n * https://github.com/Microsoft/TypeScript-wiki/blob/master/Breaking-Changes.md#extending-built-ins-like-error-array-and-map-may-no-longer-work\n */\n// tslint:enable\n\n/**\n * Equivalent of Python's AttributeError.\n */\nexport class AttributeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AttributeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's RuntimeError.\n */\nexport class RuntimeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, RuntimeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's ValueError.\n */\nexport class ValueError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, ValueError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's NotImplementedError.\n */\nexport class NotImplementedError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, NotImplementedError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's AssertionError.\n */\nexport class AssertionError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AssertionError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's IndexError.\n */\nexport class IndexError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, IndexError.prototype);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {DataType, fused, serialization, util} from '@tensorflow/tfjs-core';\n\nimport {AssertionError, ValueError} from '../errors';\n\n// tslint:enable\n\n/**\n * If `value` is an Array, equivalent to Python's `value * numValues`.\n * If `value` is not an Array, equivalent to Python's `[value] * numValues`\n */\n// tslint:disable-next-line:no-any\nexport function pyListRepeat(value: any, numValues: number): any[] {\n  if (Array.isArray(value)) {\n    // tslint:disable-next-line:no-any\n    let newArray: any[] = [];\n    for (let i = 0; i < numValues; i++) {\n      newArray = newArray.concat(value);\n    }\n    return newArray;\n  } else {\n    const newArray = new Array(numValues);\n    newArray.fill(value);\n    return newArray;\n  }\n}\n\nexport function assert(val: boolean, message?: string): void {\n  if (!val) {\n    throw new AssertionError(message);\n  }\n}\n\n/**\n * Count the number of elements of the `array` that are equal to `reference`.\n */\nexport function count<T>(array: T[], refernce: T) {\n  let counter = 0;\n  for (const item of array) {\n    if (item === refernce) {\n      counter++;\n    }\n  }\n  return counter;\n}\n\n/**\n * If an array is of length 1, just return the first element. Otherwise, return\n * the full array.\n * @param tensors\n */\nexport function singletonOrArray<T>(xs: T[]): T|T[] {\n  if (xs.length === 1) {\n    return xs[0];\n  }\n  return xs;\n}\n\n/**\n * Normalizes a list/tensor into a list.\n *\n * If a tensor is passed, we return\n * a list of size 1 containing the tensor.\n *\n * @param x target object to be normalized.\n */\n// tslint:disable-next-line:no-any\nexport function toList(x: any): any[] {\n  if (Array.isArray(x)) {\n    return x;\n  }\n  return [x];\n}\n\n/**\n * Generate a UID for a list\n */\n// tslint:disable-next-line:no-any\nexport function objectListUid(objs: any|any[]): string {\n  const objectList = toList(objs);\n  let retVal = '';\n  for (const obj of objectList) {\n    if (obj.id == null) {\n      throw new ValueError(\n          `Object ${obj} passed to objectListUid without an id`);\n    }\n    if (retVal !== '') {\n      retVal = retVal + ', ';\n    }\n    retVal = `${retVal}${Math.abs(obj.id)}`;\n  }\n  return retVal;\n}\n/**\n * Converts string to snake-case.\n * @param name\n */\nexport function toSnakeCase(name: string): string {\n  const intermediate = name.replace(/(.)([A-Z][a-z0-9]+)/g, '$1_$2');\n  const insecure =\n      intermediate.replace(/([a-z])([A-Z])/g, '$1_$2').toLowerCase();\n  /*\n   If the class is private the name starts with \"_\" which is not secure\n   for creating scopes. We prefix the name with \"private\" in this case.\n   */\n  if (insecure[0] !== '_') {\n    return insecure;\n  }\n  return 'private' + insecure;\n}\n\nexport function toCamelCase(identifier: string): string {\n  // quick return for empty string or single character strings\n  if (identifier.length <= 1) {\n    return identifier;\n  }\n  // Check for the underscore indicating snake_case\n  if (identifier.indexOf('_') === -1) {\n    return identifier;\n  }\n  return identifier.replace(/[_]+(\\w|$)/g, (m, p1) => p1.toUpperCase());\n}\n\n// tslint:disable-next-line:no-any\nlet _GLOBAL_CUSTOM_OBJECTS = {} as {[objName: string]: any};\n\nexport function serializeKerasObject(instance: serialization.Serializable):\n    serialization.ConfigDictValue {\n  if (instance === null || instance === undefined) {\n    return null;\n  }\n  const dict: serialization.ConfigDictValue = {};\n  dict['className'] = instance.getClassName();\n  dict['config'] = instance.getConfig();\n  return dict;\n}\n\n/**\n * Replace ndarray-style scalar objects in serialization objects with numbers.\n *\n * Background: In some versions of tf.keras, certain scalar values in the HDF5\n * model save file can be serialized as: `{'type': 'ndarray', 'value': num}`,\n * where in `num` is a plain number. This method converts such serialization\n * to a `number`.\n *\n * @param config The keras-format serialization object to be processed\n *   (in place).\n */\nfunction convertNDArrayScalarsInConfig(config: serialization.ConfigDictValue):\n    void {\n  if (config == null || typeof config !== 'object') {\n    return;\n  } else if (Array.isArray(config)) {\n    config.forEach(configItem => convertNDArrayScalarsInConfig(configItem));\n  } else {\n    const fields = Object.keys(config);\n    for (const field of fields) {\n      const value = config[field];\n      if (value != null && typeof value === 'object') {\n        if (!Array.isArray(value) && value['type'] === 'ndarray' &&\n            typeof value['value'] === 'number') {\n          config[field] = value['value'];\n        } else {\n          convertNDArrayScalarsInConfig(value as serialization.ConfigDict);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Deserialize a saved Keras Object\n * @param identifier either a string ID or a saved Keras dictionary\n * @param moduleObjects a list of Python class names to object constructors\n * @param customObjects a list of Python class names to object constructors\n * @param printableModuleName debug text for the object being reconstituted\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns a TensorFlow.js Layers object\n */\n// tslint:disable:no-any\nexport function deserializeKerasObject(\n    identifier: string|serialization.ConfigDict,\n    moduleObjects = {} as {[objName: string]: any},\n    customObjects = {} as {[objName: string]: any},\n    printableModuleName = 'object', fastWeightInit = false): any {\n  // tslint:enable\n  if (typeof identifier === 'string') {\n    const functionName = identifier;\n    let fn;\n    if (functionName in customObjects) {\n      fn = customObjects[functionName];\n    } else if (functionName in _GLOBAL_CUSTOM_OBJECTS) {\n      fn = _GLOBAL_CUSTOM_OBJECTS[functionName];\n    } else {\n      fn = moduleObjects[functionName];\n      if (fn == null) {\n        throw new ValueError(\n            `Unknown ${printableModuleName}: ${identifier}. ` +\n            `This may be due to one of the following reasons:\\n` +\n            `1. The ${printableModuleName} is defined in Python, in which ` +\n            `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n            `code.\\n` +\n            `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n            `but is not registered properly with ` +\n            `tf.serialization.registerClass().`);\n        // TODO(cais): Add link to tutorial page on custom layers.\n      }\n    }\n    return fn;\n  } else {\n    // In this case we are dealing with a Keras config dictionary.\n    const config = identifier;\n    if (config['className'] == null || config['config'] == null) {\n      throw new ValueError(\n          `${printableModuleName}: Improper config format: ` +\n          `${JSON.stringify(config)}.\\n` +\n          `'className' and 'config' must set.`);\n    }\n    const className = config['className'] as string;\n    let cls, fromConfig;\n    if (className in customObjects) {\n      [cls, fromConfig] = customObjects[className];\n    } else if (className in _GLOBAL_CUSTOM_OBJECTS) {\n      [cls, fromConfig] = _GLOBAL_CUSTOM_OBJECTS['className'];\n    } else if (className in moduleObjects) {\n      [cls, fromConfig] = moduleObjects[className];\n    }\n    if (cls == null) {\n      throw new ValueError(\n          `Unknown ${printableModuleName}: ${className}. ` +\n          `This may be due to one of the following reasons:\\n` +\n          `1. The ${printableModuleName} is defined in Python, in which ` +\n          `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n          `code.\\n` +\n          `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n          `but is not registered properly with ` +\n          `tf.serialization.registerClass().`);\n      // TODO(cais): Add link to tutorial page on custom layers.\n    }\n    if (fromConfig != null) {\n      // Porting notes: Instead of checking to see whether fromConfig accepts\n      // customObjects, we create a customObjects dictionary and tack it on to\n      // config['config'] as config['config'].customObjects. Objects can use it,\n      // if they want.\n\n      // tslint:disable-next-line:no-any\n      const customObjectsCombined = {} as {[objName: string]: any};\n      for (const key of Object.keys(_GLOBAL_CUSTOM_OBJECTS)) {\n        customObjectsCombined[key] = _GLOBAL_CUSTOM_OBJECTS[key];\n      }\n      for (const key of Object.keys(customObjects)) {\n        customObjectsCombined[key] = customObjects[key];\n      }\n      // Add the customObjects to config\n      const nestedConfig = config['config'] as serialization.ConfigDict;\n      nestedConfig['customObjects'] = customObjectsCombined;\n\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      convertNDArrayScalarsInConfig(config['config']);\n      const returnObj =\n          fromConfig(cls, config['config'], customObjects, fastWeightInit);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n\n      return returnObj;\n    } else {\n      // Then `cls` may be a function returning a class.\n      // In this case by convention `config` holds\n      // the kwargs of the function.\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      // In python this is **config['config'], for tfjs-layers we require\n      // classes that use this fall-through construction method to take\n      // a config interface that mimics the expansion of named parameters.\n      const returnObj = new cls(config['config']);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n      return returnObj;\n    }\n  }\n}\n\n/**\n * Compares two numbers for sorting.\n * @param a\n * @param b\n */\nexport function numberCompare(a: number, b: number) {\n  return (a < b) ? -1 : ((a > b) ? 1 : 0);\n}\n\n/**\n * Comparison of two numbers for reverse sorting.\n * @param a\n * @param b\n */\nexport function reverseNumberCompare(a: number, b: number) {\n  return -1 * numberCompare(a, b);\n}\n\n/**\n * Convert a string into the corresponding DType.\n * @param dtype\n * @returns An instance of DType.\n */\nexport function stringToDType(dtype: string): DataType {\n  switch (dtype) {\n    case 'float32':\n      return 'float32';\n    default:\n      throw new ValueError(`Invalid dtype: ${dtype}`);\n  }\n}\n\n/**\n * Test the element-by-element equality of two Arrays of strings.\n * @param xs First array of strings.\n * @param ys Second array of strings.\n * @returns Wether the two arrays are all equal, element by element.\n */\nexport function stringsEqual(xs: string[], ys: string[]): boolean {\n  if (xs == null || ys == null) {\n    return xs === ys;\n  }\n  if (xs.length !== ys.length) {\n    return false;\n  }\n  for (let i = 0; i < xs.length; ++i) {\n    if (xs[i] !== ys[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Get the unique elements of an array.\n * @param xs Array.\n * @returns An Array consisting of the unique elements in `xs`.\n */\nexport function unique<T>(xs: T[]): T[] {\n  if (xs == null) {\n    return xs;\n  }\n  const out: T[] = [];\n  // TODO(cais): Maybe improve performance by sorting.\n  for (const x of xs) {\n    if (out.indexOf(x) === -1) {\n      out.push(x);\n    }\n  }\n  return out;\n}\n\n/**\n * Determine if an Object is empty (i.e., does not have own properties).\n * @param obj Object\n * @returns Whether the Object is empty.\n * @throws ValueError: If object is `null` or `undefined`.\n */\nexport function isObjectEmpty(obj: {}): boolean {\n  if (obj == null) {\n    throw new ValueError(`Invalid value in obj: ${JSON.stringify(obj)}`);\n  }\n  for (const key in obj) {\n    if (obj.hasOwnProperty(key)) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Helper function used to build type union/enum run-time checkers.\n * @param values The list of allowed values.\n * @param label A string name for the type\n * @param value The value to test.\n * @throws ValueError: If the value is not in values nor `undefined`/`null`.\n */\nexport function checkStringTypeUnionValue(\n    values: string[], label: string, value: string): void {\n  if (value == null) {\n    return;\n  }\n  if (values.indexOf(value) < 0) {\n    throw new ValueError(`${value} is not a valid ${label}.  Valid values are ${\n        values} or null/undefined.`);\n  }\n}\n\n/**\n * Helper function for verifying the types of inputs.\n *\n * Ensures that the elements of `x` are all of type `expectedType`.\n * Also verifies that the length of `x` is within bounds.\n *\n * @param x Object to test.\n * @param expectedType The string expected type of all of the elements in the\n * Array.\n * @param minLength Return false if x.length is less than this.\n * @param maxLength Return false if x.length is greater than this.\n * @returns true if and only if `x` is an `Array<expectedType>` with\n * length >= `minLength` and <= `maxLength`.\n */\n// tslint:disable:no-any\nexport function checkArrayTypeAndLength(\n    x: any, expectedType: string, minLength = 0,\n    maxLength = Infinity): boolean {\n  assert(minLength >= 0);\n  assert(maxLength >= minLength);\n  return (\n      Array.isArray(x) && x.length >= minLength && x.length <= maxLength &&\n      x.every(e => typeof e === expectedType));\n}\n// tslint:enable:no-any\n\n/**\n * Assert that a value or an array of value are positive integer.\n *\n * @param value The value being asserted on. May be a single number or an array\n *   of numbers.\n * @param name Name of the value, used to make the error message.\n */\nexport function assertPositiveInteger(value: number|number[], name: string) {\n  if (Array.isArray(value)) {\n    util.assert(\n        value.length > 0, () => `${name} is unexpectedly an empty array.`);\n    value.forEach(\n        (v, i) => assertPositiveInteger(v, `element ${i + 1} of ${name}`));\n  } else {\n    util.assert(\n        Number.isInteger(value) && value > 0,\n        () => `Expected ${name} to be a positive integer, but got ` +\n            `${formatAsFriendlyString(value)}.`);\n  }\n}\n\n/**\n * Format a value into a display-friendly, human-readable fashion.\n *\n * - `null` is formatted as `'null'`\n * - Strings are formated with flanking pair of quotes.\n * - Arrays are formatted with flanking pair of square brackets.\n *\n * @param value The value to display.\n * @return Formatted string.\n */\n// tslint:disable-next-line:no-any\nexport function formatAsFriendlyString(value: any): string {\n  if (value === null) {\n    return 'null';\n  } else if (Array.isArray(value)) {\n    return '[' + value.map(v => formatAsFriendlyString(v)).join(',') + ']';\n  } else if (typeof value === 'string') {\n    return `\"${value}\"`;\n  } else {\n    return `${value}`;\n  }\n}\n\n/**\n * Returns a function `f2` (decorator) which wraps the original function\n * `f`. `f2` guarantees that `f` can be called at most once\n * every `waitMs` ms. If `f2` is called more often, it will return\n * the last returned result of `f`.\n *\n * @param f The original function `f` to wrap.\n * @param waitMs The time between two consecutive calls to `f` in ms.\n */\nexport function debounce<T>(\n    f: (...args: Array<{}>) => T, waitMs: number): (...args: Array<{}>) => T {\n  let lastTime = util.now();\n  let lastResult: T;\n  const f2 = (...args: Array<{}>) => {\n    const now = util.now();\n    if (now - lastTime < waitMs) {\n      return lastResult;\n    }\n    lastTime = now;\n    lastResult = f(...args);\n    return lastResult;\n  };\n  return f2;\n}\n\n/**\n * Returns the fusable activation given a layers identifier.\n *\n * @param activationName The layers identifier string.\n * @return The name of the fusable activation.\n */\nexport function mapActivationToFusedKernel(activationName: string):\n    fused.Activation {\n  if (activationName === 'relu') {\n    return 'relu';\n  }\n  if (activationName === 'linear') {\n    return 'linear';\n  }\n  if (activationName === 'elu') {\n    return 'elu';\n  }\n  return null;\n}\n\ntype PossibleValues = Array<Array<boolean|string|number>>;\n\n/**\n * Returns the cartesian product of sets of values.\n * This works the same as itertools.product in Python.\n *\n * Example:\n *\n * filters = [128, 256, 512]\n * paddings = ['same', 'valid']\n *\n * product = [ [128, 'same'], [128, 'valid'], [256, 'same'], [256, 'valid'],\n * [512, 'same'], [512, 'valid']]\n *\n * @param arrayOfValues List/array of values.\n * @return The cartesian product.\n */\nexport function getCartesianProductOfValues(...arrayOfValues: PossibleValues):\n    PossibleValues {\n  assert(arrayOfValues.length > 0, 'arrayOfValues is empty');\n\n  for (const values of arrayOfValues) {\n    assert(Array.isArray(values), 'one of the values is not an array');\n    assert(values.length > 0, 'one of the values is empty');\n  }\n\n  return arrayOfValues.reduce((products, values) => {\n    if (products.length === 0) {\n      return values.map(value => [value]);\n    }\n\n    return values\n        .map(value => {\n          return products.map((prevValue) => [...prevValue, value]);\n        })\n        .reduce((flattenedProduct, unflattenedProduct) => {\n          return flattenedProduct.concat(unflattenedProduct);\n        }, []);\n  }, [] as PossibleValues);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/contraints.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport {epsilon} from './backend/common';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\n/**\n * Helper function used by many of the Constraints to find the L2Norms.\n */\nfunction calcL2Norms(w: Tensor, axis: number): Tensor {\n  return tidy(() => tfc.sqrt(tfc.sum(tfc.mul(w, w), axis, true)));\n}\n\n/**\n * Base class for functions that impose constraints on weight values\n *\n * @doc {\n *   heading: 'Constraints',\n *   subheading: 'Classes',\n *   namespace: 'constraints'\n * }\n */\nexport abstract class Constraint extends serialization.Serializable {\n  /* Porting note: was __call__, apply chosen to match other similar choices */\n  abstract apply(w: Tensor): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport interface MaxNormArgs {\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   *\n   *  For instance, in a `Dense` layer the weight matrix\n   *  has shape `[inputDim, outputDim]`,\n   *  set `axis` to `0` to constrain each weight vector\n   *  of length `[inputDim,]`.\n   *  In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   *  the weight tensor has shape\n   *  `[rows, cols, inputDepth, outputDepth]`,\n   *  set `axis` to `[0, 1, 2]`\n   *  to constrain the weights of each filter tensor of size\n   *  `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class MaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MaxNorm';\n  private maxValue: number;\n  private axis: number;\n  private readonly defaultMaxValue = 2;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MaxNormArgs) {\n    super();\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.clipByValue(norms, 0, this.maxValue);\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {maxValue: this.maxValue, axis: this.axis};\n  }\n}\nserialization.registerClass(MaxNorm);\n\nexport interface UnitNormArgs {\n  /**\n   * Axis along which to calculate norms.\n   *\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * [rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class UnitNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'UnitNorm';\n  private axis: number;\n  private readonly defaultAxis = 0;\n  constructor(args: UnitNormArgs) {\n    super();\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(\n        () => tfc.div(w, tfc.add(epsilon(), calcL2Norms(w, this.axis))));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {axis: this.axis};\n  }\n}\nserialization.registerClass(UnitNorm);\n\nexport class NonNeg extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'NonNeg';\n\n  apply(w: Tensor): Tensor {\n    return tfc.relu(w);\n  }\n}\nserialization.registerClass(NonNeg);\n\nexport interface MinMaxNormArgs {\n  /**\n   * Minimum norm for incoming weights\n   */\n  minValue?: number;\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * `[rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n  /**\n   * Rate for enforcing the constraint: weights will be rescaled to yield:\n   * `(1 - rate) * norm + rate * norm.clip(minValue, maxValue)`.\n   * Effectively, this means that rate=1.0 stands for strict\n   * enforcement of the constraint, while rate<1.0 means that\n   * weights will be rescaled at each step to slowly move\n   * towards a value inside the desired interval.\n   */\n  rate?: number;\n}\n\nexport class MinMaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MinMaxNorm';\n  private minValue: number;\n  private maxValue: number;\n  private rate: number;\n  private axis: number;\n  private readonly defaultMinValue = 0.0;\n  private readonly defaultMaxValue = 1.0;\n  private readonly defaultRate = 1.0;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MinMaxNormArgs) {\n    super();\n    this.minValue =\n        args.minValue != null ? args.minValue : this.defaultMinValue;\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.rate = args.rate != null ? args.rate : this.defaultRate;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.add(\n          tfc.mul(\n              this.rate, tfc.clipByValue(norms, this.minValue, this.maxValue)),\n          tfc.mul(1.0 - this.rate, norms));\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      minValue: this.minValue,\n      maxValue: this.maxValue,\n      rate: this.rate,\n      axis: this.axis\n    };\n  }\n}\nserialization.registerClass(MinMaxNorm);\n\n/** @docinline */\nexport type ConstraintIdentifier =\n    'maxNorm'|'minMaxNorm'|'nonNeg'|'unitNorm'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in ConstraintIdentifier]: string} = {\n      'maxNorm': 'MaxNorm',\n      'minMaxNorm': 'MinMaxNorm',\n      'nonNeg': 'NonNeg',\n      'unitNorm': 'UnitNorm'\n    };\n\nexport function serializeConstraint(constraint: Constraint):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeConstraint(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Constraint {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'constraint');\n}\n\nexport function getConstraint(identifier: ConstraintIdentifier|\n                              serialization.ConfigDict|Constraint): Constraint {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeConstraint(config);\n  } else if (identifier instanceof Constraint) {\n    return identifier;\n  } else {\n    return deserializeConstraint(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constraint, MaxNorm, MaxNormArgs, MinMaxNorm, MinMaxNormArgs, NonNeg, UnitNorm, UnitNormArgs} from './constraints';\n\n/**\n * MaxNorm weight constraint.\n *\n * Constrains the weights incident to each hidden unit\n * to have a norm less than or equal to a desired value.\n *\n * References\n *       - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting\n * Srivastava, Hinton, et al.\n * 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Constraints',namespace: 'constraints'}\n */\nexport function maxNorm(args: MaxNormArgs): Constraint {\n  return new MaxNorm(args);\n}\n\n/**\n * Constrains the weights incident to each hidden unit to have unit norm.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function unitNorm(args: UnitNormArgs): Constraint {\n  return new UnitNorm(args);\n}\n\n/**\n * Constains the weight to be non-negative.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function nonNeg(): Constraint {\n  return new NonNeg();\n}\n\n/** @doc {heading: 'Constraints', namespace: 'constraints'} */\nexport function minMaxNorm(config: MinMaxNormArgs): Constraint {\n  return new MinMaxNorm(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// TODO(huan): add layer-specific input shape types (see: https://github.com/tensorflow/tfjs-layers/pull/492)\n/** @docalias (null | number)[] */\nexport type Shape = Array<null | number>;\n\n// The tfjs-core version of DataType must stay synced with this.\nexport type DataType = 'float32'|'int32'|'bool'|'complex64'|'string';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type DataFormat = 'channelsFirst'|'channelsLast';\nexport const VALID_DATA_FORMAT_VALUES = ['channelsFirst', 'channelsLast'];\n\n// These constants have a snake vs. camel distinction.\nexport type DataFormatSerialization = 'channels_first'|'channels_last';\n\n/** @docinline */\nexport type PaddingMode = 'valid'|'same'|'causal';\nexport const VALID_PADDING_MODE_VALUES = ['valid', 'same', 'causal'];\n\n/** @docinline */\nexport type PoolMode = 'max'|'avg';\nexport const VALID_POOL_MODE_VALUES = ['max', 'avg'];\n\n/** @docinline */\nexport type BidirectionalMergeMode = 'sum'|'mul'|'concat'|'ave';\nexport const VALID_BIDIRECTIONAL_MERGE_MODES = ['sum', 'mul', 'concat', 'ave'];\n\n/** @docinline */\nexport type SampleWeightMode = 'temporal';\nexport const VALID_SAMPLE_WEIGHT_MODES = ['temporal'];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Common functions for TensorFlow.js Layers.\n */\nimport {VALID_DATA_FORMAT_VALUES, VALID_PADDING_MODE_VALUES, VALID_POOL_MODE_VALUES} from './keras_format/common';\nimport {checkStringTypeUnionValue} from './utils/generic_utils';\n\n// A map from the requested scoped name of a Tensor to the number of Tensors\n// wanting that name so far.  This allows enforcing name uniqueness by appending\n// an incrementing index, e.g. scope/name, scope/name_1, scope/name_2, etc.\nconst nameMap: Map<string, number> = new Map<string, number>();\n\nexport function checkDataFormat(value?: string): void {\n  checkStringTypeUnionValue(VALID_DATA_FORMAT_VALUES, 'DataFormat', value);\n}\n\nexport function checkPaddingMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_PADDING_MODE_VALUES, 'PaddingMode', value);\n}\n\nexport function checkPoolMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_POOL_MODE_VALUES, 'PoolMode', value);\n}\n\nconst _nameScopeStack: string[] = [];\nconst _nameScopeDivider = '/';\n\n/**\n * Enter namescope, which can be nested.\n */\nexport function nameScope<T>(name: string, fn: () => T): T {\n  _nameScopeStack.push(name);\n  try {\n    const val: T = fn();\n    _nameScopeStack.pop();\n    return val;\n  } catch (e) {\n    _nameScopeStack.pop();\n    throw e;\n  }\n}\n\n/**\n * Get the current namescope as a flat, concatenated string.\n */\nfunction currentNameScopePrefix(): string {\n  if (_nameScopeStack.length === 0) {\n    return '';\n  } else {\n    return _nameScopeStack.join(_nameScopeDivider) + _nameScopeDivider;\n  }\n}\n\n/**\n * Get the name a Tensor (or Variable) would have if not uniqueified.\n * @param tensorName\n * @return Scoped name string.\n */\nexport function getScopedTensorName(tensorName: string): string {\n  if (!isValidTensorName(tensorName)) {\n    throw new Error('Not a valid tensor name: \\'' + tensorName + '\\'');\n  }\n  return currentNameScopePrefix() + tensorName;\n}\n\n/**\n * Get unique names for Tensors and Variables.\n * @param scopedName The fully-qualified name of the Tensor, i.e. as produced by\n *  `getScopedTensorName()`.\n * @return A unique version of the given fully scoped name.\n *   If this is the first time that the scoped name is seen in this session,\n *   then the given `scopedName` is returned unaltered.  If the same name is\n *   seen again (producing a collision), an incrementing suffix is added to the\n *   end of the name, so it takes the form 'scope/name_1', 'scope/name_2', etc.\n */\nexport function getUniqueTensorName(scopedName: string): string {\n  if (!isValidTensorName(scopedName)) {\n    throw new Error('Not a valid tensor name: \\'' + scopedName + '\\'');\n  }\n  if (!nameMap.has(scopedName)) {\n    nameMap.set(scopedName, 0);\n  }\n  const index = nameMap.get(scopedName);\n  nameMap.set(scopedName, nameMap.get(scopedName) + 1);\n\n  if (index > 0) {\n    const result = `${scopedName}_${index}`;\n    // Mark the composed name as used in case someone wants\n    // to call getUniqueTensorName(\"name_1\").\n    nameMap.set(result, 1);\n    return result;\n  } else {\n    return scopedName;\n  }\n}\n\nconst tensorNameRegex = new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\\._\\/]*$/);\n\n/**\n * Determine whether a string is a valid tensor name.\n * @param name\n * @returns A Boolean indicating whether `name` is a valid tensor name.\n */\nexport function isValidTensorName(name: string): boolean {\n  return !!name.match(tensorNameRegex);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Math utility functions.\n *\n * This file contains some frequently used math function that operates on\n * number[] or Float32Array and return a number. Many of these functions are\n * not-so-thick wrappers around TF.js Core functions. But they offer the\n * convenience of\n * 1) not having to convert the inputs into Tensors,\n * 2) not having to convert the returned Tensors to numbers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar, Tensor1D, tensor1d} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\n\nexport type ArrayTypes = Uint8Array|Int32Array|Float32Array;\n\n/**\n * Determine if a number is an integer.\n */\nexport function isInteger(x: number): boolean {\n  return x === parseInt(x.toString(), 10);\n}\n\n/**\n * Calculate the product of an array of numbers.\n * @param array The array to calculate the product over.\n * @param begin Beginning index, inclusive.\n * @param end Ending index, exclusive.\n * @return The product.\n */\nexport function arrayProd(\n    array: number[]|ArrayTypes, begin?: number, end?: number): number {\n  if (begin == null) {\n    begin = 0;\n  }\n  if (end == null) {\n    end = array.length;\n  }\n\n  let prod = 1;\n  for (let i = begin; i < end; ++i) {\n    prod *= array[i];\n  }\n  return prod;\n}\n\n/**\n * A helper function transforms the two input types to an instance of Tensor1D,\n * so the return value can be fed directly into various TF.js Core functions.\n * @param array\n */\nfunction toArray1D(array: number[]|Float32Array): Tensor1D {\n  array = Array.isArray(array) ? new Float32Array(array) : array;\n  return tensor1d(array);\n}\n\n/**\n * Compute minimum value.\n * @param array\n * @return minimum value.\n */\nexport function min(array: number[]|Float32Array): number {\n  return tfc.min(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute maximum value.\n * @param array\n * @return maximum value\n */\nexport function max(array: number[]|Float32Array): number {\n  return tfc.max(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute sum of array.\n * @param array\n * @return The sum.\n */\nexport function sum(array: number[]|Float32Array): number {\n  return tfc.sum(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute mean of array.\n * @param array\n * @return The mean.\n */\nexport function mean(array: number[]|Float32Array): number {\n  return sum(array) / array.length;\n}\n\n/**\n * Compute variance of array.\n * @param array\n * @return The variance.\n */\nexport function variance(array: number[]|Float32Array): number {\n  const demeaned = tfc.sub(toArray1D(array), scalar(mean(array)));\n  const sumSquare = tfc.sum(tfc.mul(demeaned, demeaned)).dataSync()[0];\n  return sumSquare / array.length;\n}\n\n/**\n * Compute median of array.\n * @param array\n * @return The median value.\n */\nexport function median(array: number[]|Float32Array): number {\n  const arraySorted = array.slice().sort((a, b) => a - b);\n  const lowIdx = Math.floor((arraySorted.length - 1) / 2);\n  const highIdx = Math.ceil((arraySorted.length - 1) / 2);\n  if (lowIdx === highIdx) {\n    return arraySorted[lowIdx];\n  }\n  return (arraySorted[lowIdx] + arraySorted[highIdx]) / 2;\n}\n\n/**\n * Generate an array of integers in [begin, end).\n * @param begin Beginning integer, inclusive.\n * @param end Ending integer, exclusive.\n * @returns Range array.\n * @throws ValueError, iff `end` < `begin`.\n */\nexport function range(begin: number, end: number): number[] {\n  if (end < begin) {\n    throw new ValueError(`end (${end}) < begin (${begin}) is forbidden.`);\n  }\n  const out: number[] = [];\n  for (let i = begin; i < end; ++i) {\n    out.push(i);\n  }\n  return out;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * deeplearn.js backend.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {onesLike as coreOnesLike, scalar, Tensor, Tensor1D, tensor1d, Tensor2D, Tensor3D, Tensor4D, Tensor5D, tidy, where, zerosLike as coreZerosLike} from '@tensorflow/tfjs-core';\nimport {checkDataFormat} from '../common';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {HasShape} from '../types';\nimport * as math_utils from '../utils/math_utils';\n\nimport {imageDataFormat} from './common';\n\n// tslint:enable\n\n/* Setting and getting backend from deeplearn.js. */\n\n// Default deeplearn.js backend is WebGL (GPU).\nlet backend: 'cpu'|'webgl' = 'webgl';\n\nexport function setBackend(requestedBackend: 'cpu'|'webgl') {\n  tfc.setBackend(requestedBackend);\n  backend = requestedBackend;\n}\n\nexport function getBackend(): 'cpu'|'webgl' {\n  return backend;\n}\n\n/**\n * Indicates whether the backend is operating symbolically.\n *\n * This function will be used to determine how to interpret user code. If\n * it returns true, calls to the backend construct a symbolic graph; if\n * it returns false, calls to the backend execute immediately.\n */\nexport function isBackendSymbolic(): boolean {\n  return false;\n}\n\n/**\n * Get the number of elements in a Tensor.\n * @param x The Tensor.\n * @return Number of elements in `x`.\n */\nexport function countParams(x: HasShape): number {\n  const shape = x.shape;\n  if (shape.length > 0) {\n    return shape.reduce((a: number, b: number) => a * b);\n  } else {\n    // Scalar.\n    return 1;\n  }\n}\n\n/**\n * Casts a tensor to a different dtype and returns it.\n * @param x Input tensor.\n * @param dtype String: 'float32'|'int32'|'bool'.\n * @returns Tensor of the specified `dtype`.\n */\nexport function cast(x: Tensor, dtype: tfc.DataType): Tensor {\n  return x.asType(dtype);\n}\n\n/**\n * Adds a 1-sized dimension at index \"axis\".\n * @param x Input tensor.\n * @param axis Position where to add the new axis.\n * @returns Result of the dimension expansion.\n */\nexport function expandDims(x: Tensor, axis = -1): Tensor {\n  const outShape = x.shape.slice();\n  if (axis < 0) {\n    axis = outShape.length + axis + 1;\n  }\n  outShape.splice(axis, 0, 1);\n  return x.reshape(outShape);\n}\n\n/**\n * Repeats a 2D tensor.\n *\n * If `x` has shape `[samples, dim]` and `n` is 2, for example, the output\n * will have shape `[samples, 2, dim]`.\n *\n * @param x Input tensor.\n * @param n Integer, number of times to repeat.\n * @returns The result of the repeat operation.\n * @throws ValueError: If input tensor is not 2D.\n */\nexport function repeat(x: Tensor, n: number): Tensor {\n  return tidy(() => {\n    if (x.shape.length !== 2) {\n      throw new ValueError(\n          `repeat() expects a rank-2 tensor, but received a ` +\n          `rank-${x.shape.length} tensor.`);\n    }\n    const y = expandDims(x, 1);\n    return tile(y, [1, n, 1]);\n  });\n}\n\n/**\n * Flatten a Tensor into 1D.\n * @param x Input tensor.\n * @return The result of the flattening `x`.\n */\nexport function flatten(x: Tensor): Tensor {\n  const newShape = [math_utils.arrayProd(x.shape)];\n  return x.reshape(newShape);\n}\n\n/**\n * Turn a nD tensor into a 2D tensor with same 0th dimension.\n * In other words, it flattens each data samples of a batch.\n *\n * @param x The tensor to flatten. The rank of this tensor is required to be 2\n *   or higher.\n * @return The result of the flattening.\n */\nexport function batchFlatten(x: Tensor): Tensor {\n  if (x.rank <= 1) {\n    throw new ValueError(\n        `batchFlatten requires a minimum rank of 2. Got rank: ${x.rank}.`);\n  }\n  const newShape = [x.shape[0], math_utils.arrayProd(x.shape, 1)];\n  return x.reshape(newShape);\n}\n\n/**\n * Do slicing along the first axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the first axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongFirstAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [start, 0], [size, array.shape[1]]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [start, 0, 0],\n            [size, array.shape[1], array.shape[2]]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [start, 0, 0, 0],\n            [size, array.shape[1], array.shape[2], array.shape[3]]);\n      case 5:\n        return tfc.slice(array as Tensor5D, [start, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4]\n        ]);\n      case 6:\n        return tfc.slice(array, [start, 0, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4],\n          array.shape[5]\n        ]);\n      default:\n        throw new ValueError(\n            `sliceAlongFirstAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the last axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the last axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongLastAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [0, start], [array.shape[0], size]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [0, 0, start],\n            [array.shape[0], array.shape[1], size]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [0, 0, 0, start],\n            [array.shape[0], array.shape[1], array.shape[2], size]);\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the sepcified axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size of the slice along the chosen axis.\n * @param choose an axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongAxis(\n    array: Tensor, start: number, size: number, axis: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 3:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice3d(\n                array as Tensor3D, [0, start, 0],\n                [array.shape[0], size, array.shape[2]]);\n          case 3:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 4:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice4d(\n                array as Tensor4D, [0, start, 0, 0],\n                [array.shape[0], size, array.shape[2], array.shape[3]]);\n          case 3:\n            return tfc.slice4d(\n                array as Tensor4D, [0, 0, start, 0],\n                [array.shape[0], array.shape[1], size, array.shape[3]]);\n          case 4:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Concatenates a list of tensors alongside the specified axis.\n * @param tensors `Array` of tensors to concatenate.\n * @param axis Concatenation axis.\n * @returns The result of the concatenation.\n */\nexport function concatenate(tensors: Tensor[], axis = -1): Tensor {\n  let rank: number;\n  if (axis < 0) {\n    rank = tensors[0].rank;\n    if (rank !== 0) {\n      axis = rank;\n    } else {\n      axis = 0;\n    }\n  }\n  if (axis === tensors[0].rank) {\n    // Porting Note: This is necessary because tfc.concat() requires axis to be\n    //   in the interval [-rank, rank).\n    axis = -1;\n  }\n  // Porting Note: Sparse concat is not supported yet.\n  return tfc.concat(tensors, axis);\n}\n\n/**\n * Concatenate two arrays along the first dimension.\n * @param a The 1st `tf.Tensor` to concatenate.\n * @param b The 2nd `tf.Tensor` to concatenate.\n * @returns Result of the concatenation.\n * @throws ValueError: If `a` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function concatAlongFirstAxis(a: Tensor, b: Tensor): Tensor {\n  switch (a.rank) {\n    case 1:\n      return tfc.concat1d([a as Tensor1D, b as Tensor1D]);\n    case 2:\n      return tfc.concat2d([a as Tensor2D, b as Tensor2D], 0);\n    case 3:\n      return tfc.concat3d([a as Tensor3D, b as Tensor3D], 0);\n    case 4:\n      return tfc.concat4d([a as Tensor4D, b as Tensor4D], 0);\n    default:\n      throw new ValueError(\n          `concatAlongFirstAxis() received an unsupported ` +\n          `tensor rank: ${a.rank}`);\n  }\n}\n\n/**\n * Creates a tensor by tiling `x` by `n`.\n * @param x A tensor.\n * @param n An Array of integers or a single integer. If an Array, the length\n *   must be the same as the number of dimensions in `x`. If a single integer,\n *   it will be treated as an Array of length 1.\n */\nexport function tile(x: Tensor, n: number|number[]): Tensor {\n  if (!Array.isArray(n)) {\n    n = [n];\n  }\n  if (x.rank !== n.length) {\n    throw new ValueError(\n        `The length of input n (${n.length}) does not match ` +\n        `the number of dimensions in input x (${x.rank})`);\n  }\n  return tfc.tile(x, n);\n}\n\n/* Creation of random tensors. */\n\n/**\n * Get a tensor with normal distribution of values.\n *\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @return The normal tensor.\n */\nexport function randomNormal(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: 'float32'|'int32',\n    seed?: number): Tensor {\n  return tfc.randomNormal(shape, mean, stddev, dtype, seed);\n}\n\n/* Linear Algebra */\n\n/**\n * Multiply two tensors and returns the result as a tensor.\n *\n * For 2D tensors, this is equivalent to matrix multiplication (matMul).\n * For tensors of higher ranks, it follows the Theano behavior,\n * (e.g. `(2, 3) * (4, 3, 5) -> (2, 4, 5)`).  From the Theano documentation:\n *\n * For N dimensions it is a sum product over the last axis of x and the\n * second-to-last of y:\n *\n * @param a A tensor of at least rank 2.\n * @param b A tensor of at least rank 2.\n * @param activation (optional) A string identifying the activation\n *   function.\n * @return Result of the dot operation.\n */\nexport function dot(\n    a: Tensor, b: Tensor, activation?: tfc.fused.Activation,\n    bias?: Tensor): Tensor {\n  if ((a.rank < 2) || (b.rank < 2)) {\n    throw new NotImplementedError(\n        `dot requires both inputs to be rank >= 2` +\n        ` but got x shape = ${a.shape} and y shape = ${b.shape}`);\n  }\n  if (b.rank >= 3) {\n    const xLastDim = a.shape.slice(-1)[0];\n    const ySecondLastDim = b.shape.slice(-2)[0];\n    if (xLastDim !== ySecondLastDim) {\n      throw new NotImplementedError(\n          `If rank y >= 3, then the second last dim` +\n          ` of y must equal the last dim of x but got x shape = ${\n              a.shape} and ` +\n          ` y shape = ${b.shape}`);\n    }\n  }\n  // Handle basic 2D x 2D case.\n  if ((a.rank === 2) && (b.rank === 2)) {\n    const transposeA = false;\n    const transposeB = false;\n    // tfc.fused.matMul only fuses certain activation functions. Unsupported\n    // activation functions are treated as 'linear' activations, which is\n    // equivalent to a no-op.\n    return tfc.fused.matMul({\n      a,\n      b: b as Tensor2D,\n      transposeA,\n      transposeB,\n      bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n      activation\n    });\n  } else {\n    // Reshape x into the analogous 2D Tensor.\n    const aFirstDims = a.shape.slice();  // Holds all but the last dim of x.\n    const aLastDim = aFirstDims.pop();\n    a = a.reshape([-1, aLastDim]);\n\n    // Reshape y into the analogous 2D Tensor, and keep track of the\n    // required dimensions to reproduce the output shape.\n    const bShape = b.shape.slice();\n    const bLastDim = bShape.pop();\n    const ySecondLastDim = bShape.pop();\n    const yOtherDims = [...bShape, bLastDim];\n    // permutation should be like [r-2, 0, 1, 2, ... r-4, r-3, r-1]\n    // where r is the rank of y.\n    const perm = Array.from({length: b.rank}, (_, i) => {\n      if (i === 0) {\n        return b.rank - 2;\n      } else if (i <= b.rank - 2) {\n        return i - 1;\n      }\n      return i;\n    });\n    b = b.transpose(perm).reshape([ySecondLastDim, -1]);\n\n    // Multiply x and y as 2D Tensors, and then reshape back to original.\n    const outputShape = [...aFirstDims, ...yOtherDims];\n    const transposeA = false;\n    const transposeB = false;\n    return tfc.fused\n        .matMul({\n          a,\n          b,\n          transposeA,\n          transposeB,\n          bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n          activation\n        })\n        .reshape(outputShape);\n  }\n}\n\n/**\n * Compute the sign Tensor of an input Tensor.\n *\n * Elements of the input `tf.Tensor` that are === 0 are mapped to 0.\n * Elements of the input `tf.Tensor` that are > 0 are mapped to 1.\n * Elements of the input `tf.Tensor` that are < 0 are mapped to -1.\n *\n * @param x Input `tf.Tensor`.\n * @return The sign `tf.Tensor`.\n */\nexport function sign(x: Tensor): Tensor {\n  // TODO(cais): Move to the core.\n  return tidy(() => {\n    const zerosLikeX = coreZerosLike(x);\n    const onesLikeX = coreOnesLike(x);\n    return where(\n        tfc.equal(x, zerosLikeX), zerosLikeX,\n        where(\n            tfc.greater(x, coreZerosLike(x)), onesLikeX,\n            tfc.mul(-1, onesLikeX)));\n  });\n}\n\n/**\n * Computes the one-hot representation of an integer tensor.\n * @param indices nD integer tensor of shape\n *   `(batch_size, dim1, dim2, ... dim(n-1))`\n * @param numClasses Integer, number of classes to consider.\n * @returns (n + 1)D one hot representation of the input\n *   with shape `(batch_size, dim1, dim2, ... dim(n-1), num_classes)`\n */\nexport function oneHot(indices: Tensor, numClasses: number): Tensor {\n  return tidy(() => {\n    if (indices.rank !== 1) {\n      throw new Error(\n          'Only 1D one-hot tensors are supported in the ' +\n          'deeplearn backend, at present.');\n    }\n    indices = indices.toInt();\n    return tfc.oneHot(indices as Tensor1D, numClasses).toFloat();\n  });\n}\n\n/* Elementary math functions. */\n\n/**\n * Retrieves the elements of indices `indices` in the tensor `reference`.\n * @param reference A tensor.\n * @param indices An integer tensor of indices or an `Array` of integers.\n * @param axis Axis along which to perform the gather operation.\n * @returns The result of the gathering as a tensor.\n */\nexport function gather(\n    reference: Tensor, indices: number[]|Tensor1D, axis?: number): Tensor {\n  return tidy(() => {\n    if (Array.isArray(indices)) {\n      indices = tensor1d(indices, 'int32');\n    } else {\n      indices = indices.toInt();\n    }\n    return tfc.gather(reference, indices, axis);\n  });\n}\n\n/**\n * Element-wise square.\n * @param x Input tensor.\n * @return element-wise x^2\n */\nexport function square(x: Tensor): Tensor {\n  return tfc.mul(x, x);\n}\n\n/**\n * Element-wise exponentiation.\n *\n * Porting Note: In PyKeras, `a` (the exponent) is a Python integer, which\n *   takes advatnage of the backend's (e.g., TensorFlow's) automatic\n * conversion to tensor. Here we allow `a` to be either a number or a tensor.\n *\n * @param x The base tensor.\n * @param a The exponent, tensor or number. If a number, it is rounded to the\n *   nearest integer and converted to a tensor.\n * @returns A tensor of the same shape as `x`.\n */\nexport function pow(x: Tensor, a: Tensor|number): Tensor {\n  return tidy(() => {\n    if (typeof (a) === 'number') {\n      a = scalar(Math.round(a), 'int32');\n    }\n    if (a.dtype !== 'int32') {\n      throw new NotImplementedError(\n          `Non-int32 dtype (${a.dtype}) is not supported by pow() yet`);\n    }\n    return tfc.pow(x, a);\n  });\n}\n\n/**\n * Reshapes bias tensor according to rank of x.\n */\nfunction reshapeBias(xRank: number, bias: Tensor, dataFormat: string) {\n  const biasShape = bias.shape;\n\n  if (bias.rank !== 1 && bias.rank !== xRank) {\n    throw new ValueError(\n        `Unexpected bias dimensions: ${bias.rank}` +\n        `; expected it to be 1 or ${xRank}`);\n  }\n\n  if (xRank === 5) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1, 1]);\n      } else {\n        return bias.reshape(\n            [1, biasShape[3], biasShape[0], biasShape[1], biasShape[2]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 4) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1]);\n      } else {\n        return bias.reshape([1, biasShape[2], biasShape[0], biasShape[1]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 3) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1]);\n      } else {\n        return bias.reshape([1, biasShape[1], biasShape[0]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank < 3) {\n    return bias;\n  }\n  throw new ValueError(`Unsupported input rank by biasAdd: ${bias.rank}`);\n}\n\n/* Neural-network operations. */\n\n/**\n * Add a bias to a tensor.\n *\n * @param x The tensor to add the bias to.\n * @param bias The bias to add to `x`. Must be 1D or the same rank as `x`.\n * @return Result of the bias adding.\n * @throws ValueError: If the rank of `bias` is incorrect.\n */\nexport function biasAdd(\n    x: Tensor, bias: Tensor, dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n\n    return x.add(reshapeBias(x.rank, bias, dataFormat));\n  });\n}\n\n/**\n * Exponential linear unit (ELU).\n * @param x A tensor or variable to compute the activation function for.\n * @param alpha: A scalar, a scaling factor for the negative section.\n * @return Output of the ELU operation.\n */\nexport function elu(x: Tensor, alpha = 1): Tensor {\n  // TODO(cais): Add support for alpha values other than 1.\n  if (alpha !== 1) {\n    throw new NotImplementedError(\n        `Support for alpha values other than 1 (${alpha}) is not implemented ` +\n        `yet.`);\n  }\n  return tfc.elu(x);\n}\n\n/**\n * Softsign of a tensor.\n *\n * Defined as x / (abs(x) + 1), element-wise.\n *\n * @param x: Input.\n * @returns Output.\n */\nexport function softsign(x: Tensor): Tensor {\n  return tidy(() => tfc.div(x, tfc.abs(x).add(1)));\n}\n\n/**\n * Sets entries in `x` to zero at random, while scaling the entire tensor.\n *\n * @param x input tensor.\n * @param level fraction of the entries in the tensor that will be set to 0.\n * @param noiseShape shape of randomly generated keep/drop flags, must be\n *   broadcastable to the shape of `x`. Optional.\n * @param seed random seed to ensure determinism. Optional.\n * @returns Result of the dropout operation.\n */\nexport function dropout(\n    x: Tensor, level: number, noiseShape?: number[], seed?: number): Tensor {\n  return tidy(() => tfc.dropout(x, level, noiseShape, seed));\n}\n\n/**\n * Element-wise, segment-wise linear approximation of sigmoid.\n *\n * Returns `0.` if `x < -2.5`, `1.` if `x > 2.5`.\n * In `-2.5 <= x <= 2.5`, returns `0.2 * x + 0.5`.\n *\n * @param x Input tensor.\n * @returns Output tensor.\n */\nexport function hardSigmoid(x: Tensor): Tensor {\n  return tidy(() => {\n    const y = tfc.add(.5, tfc.mul(.2, x));\n    return tfc.clipByValue(y, 0, 1);\n  });\n}\n\n/**\n * Invoke `x` in the training phase, and `alt` otherwise.\n *\n * Porting Note: We do not create placeholder tensors for the `training`\n * boolean flag here, because there is no such thing in the TF.js imperative\n * backend.\n *\n * @param x The function to invoke iff `training` is `true`.\n * @param alt The function to invoke iff `training` is `false`.\n * @param training Boolean flag for whether training phase is active.\n * @returns The return value of `x()` if `training` is `true`, or the return\n *   value of `alt()` if `training` is `false`.\n */\nexport function inTrainPhase<T>(x: () => T, alt: () => T, training = false): T {\n  return training ? x() : alt();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {BaseSerialization} from './types';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type FanMode = 'fanIn'|'fanOut'|'fanAvg';\nexport const VALID_FAN_MODE_VALUES = ['fanIn', 'fanOut', 'fanAvg'];\n\n// These constants have a snake vs. camel distinction.\nexport type FanModeSerialization = 'fan_in'|'fan_out'|'fan_avg';\n\n/** @docinline */\nexport type Distribution = 'normal'|'uniform'|'truncatedNormal';\nexport const VALID_DISTRIBUTION_VALUES =\n    ['normal', 'uniform', 'truncatedNormal'];\n// These constants have a snake vs. camel distinction.\nexport type DistributionSerialization = 'normal'|'uniform'|'truncated_normal';\n\nexport type ZerosSerialization = BaseSerialization<'Zeros', {}>;\n\nexport type OnesSerialization = BaseSerialization<'Ones', {}>;\n\nexport type ConstantConfig = {\n  value: number;\n};\n\nexport type ConstantSerialization =\n    BaseSerialization<'Constant', ConstantConfig>;\n\nexport type RandomNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type RandomNormalSerialization =\n    BaseSerialization<'RandomNormal', RandomNormalConfig>;\n\nexport type RandomUniformConfig = {\n  minval?: number;\n  maxval?: number;\n  seed?: number;\n};\n\nexport type RandomUniformSerialization =\n    BaseSerialization<'RandomUniform', RandomUniformConfig>;\n\nexport type TruncatedNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type TruncatedNormalSerialization =\n    BaseSerialization<'TruncatedNormal', TruncatedNormalConfig>;\n\nexport type VarianceScalingConfig = {\n  scale?: number;\n\n  mode?: FanModeSerialization;\n  distribution?: DistributionSerialization;\n  seed?: number;\n};\n\nexport type VarianceScalingSerialization =\n    BaseSerialization<'VarianceScaling', VarianceScalingConfig>;\n\nexport type OrthogonalConfig = {\n  seed?: number;\n  gain?: number;\n};\n\nexport type OrthogonalSerialization =\n    BaseSerialization<'Orthogonal', OrthogonalConfig>;\n\nexport type IdentityConfig = {\n  gain?: number;\n};\n\nexport type IdentitySerialization =\n    BaseSerialization<'Identity', IdentityConfig>;\n\n// Update initializerClassNames below in concert with this.\nexport type InitializerSerialization = ZerosSerialization|OnesSerialization|\n    ConstantSerialization|RandomUniformSerialization|RandomNormalSerialization|\n    TruncatedNormalSerialization|IdentitySerialization|\n    VarianceScalingSerialization|OrthogonalSerialization;\n\nexport type InitializerClassName = InitializerSerialization['class_name'];\n\n// We can't easily extract a string[] from the string union type, but we can\n// recapitulate the list, enforcing at compile time that the values are valid\n// and that we have the right number of them.\n\n/**\n * A string array of valid Initializer class names.\n *\n * This is guaranteed to match the `InitializerClassName` union type.\n */\nexport const initializerClassNames: InitializerClassName[] = [\n  'Zeros', 'Ones', 'Constant', 'RandomNormal', 'RandomUniform',\n  'TruncatedNormal', 'VarianceScaling', 'Orthogonal', 'Identity'\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, eye, linalg, mul, ones, randomUniform, scalar, serialization, Tensor, Tensor2D, tidy, truncatedNormal, zeros} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {checkDataFormat} from './common';\nimport {NotImplementedError, ValueError} from './errors';\nimport {DataFormat, Shape} from './keras_format/common';\nimport {Distribution, FanMode, VALID_DISTRIBUTION_VALUES, VALID_FAN_MODE_VALUES} from './keras_format/initializer_config';\nimport {checkStringTypeUnionValue, deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\nimport {arrayProd} from './utils/math_utils';\n\nexport function checkFanMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_FAN_MODE_VALUES, 'FanMode', value);\n}\n\nexport function checkDistribution(value?: string): void {\n  checkStringTypeUnionValue(VALID_DISTRIBUTION_VALUES, 'Distribution', value);\n}\n\n/**\n * Initializer base class.\n *\n * @doc {\n *   heading: 'Initializers', subheading: 'Classes', namespace: 'initializers'}\n */\nexport abstract class Initializer extends serialization.Serializable {\n  public fromConfigUsesCustomObjects(): boolean {\n    return false;\n  }\n  /**\n   * Generate an initial value.\n   * @param shape\n   * @param dtype\n   * @return The init value.\n   */\n  abstract apply(shape: Shape, dtype?: DataType): Tensor;\n\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport class Zeros extends Initializer {\n  /** @nocollapse */\n  static className = 'Zeros';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return zeros(shape, dtype);\n  }\n}\nserialization.registerClass(Zeros);\n\nexport class Ones extends Initializer {\n  /** @nocollapse */\n  static className = 'Ones';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return ones(shape, dtype);\n  }\n}\nserialization.registerClass(Ones);\n\nexport interface ConstantArgs {\n  /** The value for each element in the variable. */\n  value: number;\n}\n\nexport class Constant extends Initializer {\n  /** @nocollapse */\n  static className = 'Constant';\n  private value: number;\n  constructor(args: ConstantArgs) {\n    super();\n    if (typeof args !== 'object') {\n      throw new ValueError(\n          `Expected argument of type ConstantConfig but got ${args}`);\n    }\n    if (args.value === undefined) {\n      throw new ValueError(`config must have value set but got ${args}`);\n    }\n    this.value = args.value;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => mul(scalar(this.value), ones(shape, dtype)));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      value: this.value,\n    };\n  }\n}\nserialization.registerClass(Constant);\n\nexport interface RandomUniformArgs {\n  /** Lower bound of the range of random values to generate. */\n  minval?: number;\n  /** Upper bound of the range of random values to generate. */\n  maxval?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomUniform extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomUniform';\n  readonly DEFAULT_MINVAL = -0.05;\n  readonly DEFAULT_MAXVAL = 0.05;\n  private minval: number;\n  private maxval: number;\n  private seed: number;\n\n  constructor(args: RandomUniformArgs) {\n    super();\n    this.minval = args.minval || this.DEFAULT_MINVAL;\n    this.maxval = args.maxval || this.DEFAULT_MAXVAL;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return randomUniform(shape, this.minval, this.maxval, dtype);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {minval: this.minval, maxval: this.maxval, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomUniform);\n\nexport interface RandomNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomNormal';\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: RandomNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `randomNormal does not support dType ${dtype}.`);\n    }\n\n    return K.randomNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomNormal);\n\nexport interface TruncatedNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class TruncatedNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'TruncatedNormal';\n\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: TruncatedNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `truncatedNormal does not support dType ${dtype}.`);\n    }\n    return truncatedNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(TruncatedNormal);\n\nexport interface IdentityArgs {\n  /**\n   * Multiplicative factor to apply to the identity matrix.\n   */\n  gain?: number;\n}\n\nexport class Identity extends Initializer {\n  /** @nocollapse */\n  static className = 'Identity';\n  private gain: number;\n  constructor(args: IdentityArgs) {\n    super();\n    this.gain = args.gain != null ? args.gain : 1.0;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length !== 2 || shape[0] !== shape[1]) {\n        throw new ValueError(\n            'Identity matrix initializer can only be used for' +\n            ' 2D square matrices.');\n      } else {\n        return mul(this.gain, eye(shape[0]));\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {gain: this.gain};\n  }\n}\nserialization.registerClass(Identity);\n\n/**\n * Computes the number of input and output units for a weight shape.\n * @param shape Shape of weight.\n * @param dataFormat data format to use for convolution kernels.\n *   Note that all kernels in Keras are standardized on the\n *   CHANNEL_LAST ordering (even when inputs are set to CHANNEL_FIRST).\n * @return An length-2 array: fanIn, fanOut.\n */\nfunction computeFans(\n    shape: Shape, dataFormat: DataFormat = 'channelsLast'): number[] {\n  let fanIn: number;\n  let fanOut: number;\n  checkDataFormat(dataFormat);\n  if (shape.length === 2) {\n    fanIn = shape[0];\n    fanOut = shape[1];\n  } else if ([3, 4, 5].indexOf(shape.length) !== -1) {\n    if (dataFormat === 'channelsFirst') {\n      const receptiveFieldSize = arrayProd(shape, 2);\n      fanIn = shape[1] * receptiveFieldSize;\n      fanOut = shape[0] * receptiveFieldSize;\n    } else if (dataFormat === 'channelsLast') {\n      const receptiveFieldSize = arrayProd(shape, 0, shape.length - 2);\n      fanIn = shape[shape.length - 2] * receptiveFieldSize;\n      fanOut = shape[shape.length - 1] * receptiveFieldSize;\n    }\n  } else {\n    const shapeProd = arrayProd(shape);\n    fanIn = Math.sqrt(shapeProd);\n    fanOut = Math.sqrt(shapeProd);\n  }\n\n  return [fanIn, fanOut];\n}\n\nexport interface VarianceScalingArgs {\n  /** Scaling factor (positive float). */\n  scale?: number;\n\n  /** Fanning mode for inputs and outputs. */\n  mode?: FanMode;\n\n  /** Probabilistic distribution of the values. */\n  distribution?: Distribution;\n\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class VarianceScaling extends Initializer {\n  /** @nocollapse */\n  static className = 'VarianceScaling';\n  private scale: number;\n  private mode: FanMode;\n  private distribution: Distribution;\n  private seed: number;\n\n  /**\n   * Constructor of VarianceScaling.\n   * @throws ValueError for invalid value in scale.\n   */\n  constructor(args: VarianceScalingArgs) {\n    super();\n    if (args.scale < 0.0) {\n      throw new ValueError(\n          `scale must be a positive float. Got: ${args.scale}`);\n    }\n    this.scale = args.scale == null ? 1.0 : args.scale;\n    this.mode = args.mode == null ? 'fanIn' : args.mode;\n    checkFanMode(this.mode);\n    this.distribution =\n        args.distribution == null ? 'normal' : args.distribution;\n    checkDistribution(this.distribution);\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    const fans = computeFans(shape);\n    const fanIn = fans[0];\n    const fanOut = fans[1];\n    let scale = this.scale;\n    if (this.mode === 'fanIn') {\n      scale /= Math.max(1, fanIn);\n    } else if (this.mode === 'fanOut') {\n      scale /= Math.max(1, fanOut);\n    } else {\n      scale /= Math.max(1, (fanIn + fanOut) / 2);\n    }\n\n    if (this.distribution === 'normal') {\n      const stddev = Math.sqrt(scale);\n      dtype = dtype || 'float32';\n      if (dtype !== 'float32' && dtype !== 'int32') {\n        throw new NotImplementedError(\n            `${this.getClassName()} does not support dType ${dtype}.`);\n      }\n      return truncatedNormal(shape, 0, stddev, dtype, this.seed);\n    } else {\n      const limit = Math.sqrt(3 * scale);\n      return randomUniform(shape, -limit, limit, dtype);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      scale: this.scale,\n      mode: this.mode,\n      distribution: this.distribution,\n      seed: this.seed\n    };\n  }\n}\nserialization.registerClass(VarianceScaling);\n\nexport interface SeedOnlyInitializerArgs {\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class GlorotUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotUniform';\n\n  /**\n   * Constructor of GlorotUniform\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotUniform);\n\nexport class GlorotNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotNormal';\n\n  /**\n   * Constructor of GlorotNormal.\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotNormal);\n\nexport class HeNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeNormal);\n\nexport class HeUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeUniform';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeUniform);\n\nexport class LeCunNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunNormal);\n\nexport class LeCunUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunUniform);\n\nexport interface OrthogonalArgs extends SeedOnlyInitializerArgs {\n  /**\n   * Multiplicative factor to apply to the orthogonal matrix. Defaults to 1.\n   */\n  gain?: number;\n}\n\nexport class Orthogonal extends Initializer {\n  /** @nocollapse */\n  static className = 'Orthogonal';\n  readonly DEFAULT_GAIN = 1;\n  protected readonly gain: number;\n  protected readonly seed: number;\n\n  constructor(args?: OrthogonalArgs) {\n    super();\n    this.gain = args.gain == null ? this.DEFAULT_GAIN : args.gain;\n    this.seed = args.seed;\n\n    if (this.seed != null) {\n      throw new NotImplementedError(\n          'Random seed is not implemented for Orthogonal Initializer yet.');\n    }\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length < 2) {\n        throw new NotImplementedError('Shape must be at least 2D.');\n      }\n      if (shape[0] * shape[1] > 2000) {\n        console.warn(\n            `Orthogonal initializer is being called on a matrix with more ` +\n            `than 2000 (${shape[0] * shape[1]}) elements: ` +\n            `Slowness may result.`);\n      }\n\n      // TODO(cais): Add seed support.\n      const normalizedShape =\n          shape[0] > shape[1] ? [shape[1], shape[0]] : shape;\n      const a = K.randomNormal(normalizedShape, 0, 1, 'float32') as Tensor2D;\n      let q = linalg.gramSchmidt(a) as Tensor2D;\n      if (shape[0] > shape[1]) {\n        q = q.transpose();\n      }\n      return mul(this.gain, q);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      gain: this.gain,\n      seed: this.seed,\n    };\n  }\n}\nserialization.registerClass(Orthogonal);\n\n/** @docinline */\nexport type InitializerIdentifier =\n    'constant'|'glorotNormal'|'glorotUniform'|'heNormal'|'heUniform'|'identity'|\n    'leCunNormal'|'leCunUniform'|'ones'|'orthogonal'|'randomNormal'|\n    'randomUniform'|'truncatedNormal'|'varianceScaling'|'zeros'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in InitializerIdentifier]: string} = {\n      'constant': 'Constant',\n      'glorotNormal': 'GlorotNormal',\n      'glorotUniform': 'GlorotUniform',\n      'heNormal': 'HeNormal',\n      'heUniform': 'HeUniform',\n      'identity': 'Identity',\n      'leCunNormal': 'LeCunNormal',\n      'leCunUniform': 'LeCunUniform',\n      'ones': 'Ones',\n      'orthogonal': 'Orthogonal',\n      'randomNormal': 'RandomNormal',\n      'randomUniform': 'RandomUniform',\n      'truncatedNormal': 'TruncatedNormal',\n      'varianceScaling': 'VarianceScaling',\n      'zeros': 'Zeros'\n    };\n\nfunction deserializeInitializer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Initializer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'initializer');\n}\n\nexport function serializeInitializer(initializer: Initializer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(initializer);\n}\n\nexport function getInitializer(identifier: InitializerIdentifier|Initializer|\n                               serialization.ConfigDict): Initializer {\n  if (typeof identifier === 'string') {\n    const className = identifier in INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    /* We have four 'helper' classes for common initializers that\n    all get serialized as 'VarianceScaling' and shouldn't go through\n    the deserializeInitializer pathway. */\n    if (className === 'GlorotNormal') {\n      return new GlorotNormal();\n    } else if (className === 'GlorotUniform') {\n      return new GlorotUniform();\n    } else if (className === 'HeNormal') {\n      return new HeNormal();\n    } else if (className === 'HeUniform') {\n      return new HeUniform();\n    } else if (className === 'LeCunNormal') {\n      return new LeCunNormal();\n    } else if (className === 'LeCunUniform') {\n      return new LeCunUniform();\n    } else {\n      const config: serialization.ConfigDict = {};\n      config['className'] = className;\n      config['config'] = {};\n      return deserializeInitializer(config);\n    }\n  } else if (identifier instanceof Initializer) {\n    return identifier;\n  } else {\n    return deserializeInitializer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constant, ConstantArgs, GlorotNormal, GlorotUniform, HeNormal, HeUniform, Identity, IdentityArgs, Initializer, LeCunNormal, LeCunUniform, Ones, Orthogonal, OrthogonalArgs, RandomNormal, RandomNormalArgs, RandomUniform, RandomUniformArgs, SeedOnlyInitializerArgs, TruncatedNormal, TruncatedNormalArgs, VarianceScaling, VarianceScalingArgs, Zeros} from './initializers';\n\n/**\n * Initializer that generates tensors initialized to 0.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function zeros(): Zeros {\n  return new Zeros();\n}\n\n/**\n * Initializer that generates tensors initialized to 1.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function ones(): Initializer {\n  return new Ones();\n}\n\n/**\n * Initializer that generates values initialized to some constant.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function constant(args: ConstantArgs): Initializer {\n  return new Constant(args);\n}\n\n/**\n * Initializer that generates random values initialized to a uniform\n * distribution.\n *\n * Values will be distributed uniformly between the configured minval and\n * maxval.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomUniform(args: RandomUniformArgs): Initializer {\n  return new RandomUniform(args);\n}\n\n/**\n * Initializer that generates random values initialized to a normal\n * distribution.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomNormal(args: RandomNormalArgs): Initializer {\n  return new RandomNormal(args);\n}\n\n/**\n * Initializer that generates random values initialized to a truncated normal.\n * distribution.\n *\n * These values are similar to values from a `RandomNormal` except that values\n * more than two standard deviations from the mean are discarded and re-drawn.\n * This is the recommended initializer for neural network weights and filters.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function truncatedNormal(args: TruncatedNormalArgs): Initializer {\n  return new TruncatedNormal(args);\n}\n\n/**\n * Initializer that generates the identity matrix.\n * Only use for square 2D matrices.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function identity(args: IdentityArgs): Initializer {\n  return new Identity(args);\n}\n\n/**\n * Initializer capable of adapting its scale to the shape of weights.\n * With distribution=NORMAL, samples are drawn from a truncated normal\n * distribution centered on zero, with `stddev = sqrt(scale / n)` where n is:\n *   - number of input units in the weight tensor, if mode = FAN_IN.\n *   - number of output units, if mode = FAN_OUT.\n *   - average of the numbers of input and output units, if mode = FAN_AVG.\n * With distribution=UNIFORM,\n * samples are drawn from a uniform distribution\n * within [-limit, limit], with `limit = sqrt(3 * scale / n)`.\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function varianceScaling(config: VarianceScalingArgs): Initializer {\n  return new VarianceScaling(config);\n}\n\n/**\n * Glorot uniform initializer, also called Xavier uniform initializer.\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotUniform(args);\n}\n\n/**\n * Glorot normal initializer, also called Xavier normal initializer.\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor.\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotNormal(args);\n}\n\n/**\n * He normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function heNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeNormal(args);\n}\n\n/**\n * He uniform initializer.\n *\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / fan_in)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function heUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeUniform(args);\n}\n\n/**\n * LeCun normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(1 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * References:\n *   [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *   [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunNormal(args);\n}\n\n/**\n * LeCun uniform initializer.\n *\n * It draws samples from a uniform distribution in the interval\n * `[-limit, limit]` with `limit = sqrt(3 / fanIn)`,\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunUniform(args);\n}\n\n/**\n * Initializer that generates a random orthogonal matrix.\n *\n * Reference:\n * [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function orthogonal(args: OrthogonalArgs): Initializer {\n  return new Orthogonal(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Utilities related to persistent state in the backend.\n */\n\n/**\n * An ID to track `tf.SymbolicTensor`s and derived classes.\n * Required in different places in engine/topology.ts to identify unique\n * tensors.\n */\nlet _nextUniqueTensorId = 0;\n\nexport function getNextUniqueTensorId(): number {\n  return _nextUniqueTensorId++;\n}\n\nconst _uidPrefixes: {[prefix: string]: number} = {};\n\n/**\n * Provides a unique UID given a string prefix.\n *\n * @param prefix\n */\nexport function getUid(prefix = ''): string {\n  if (!(prefix in _uidPrefixes)) {\n    _uidPrefixes[prefix] = 0;\n  }\n  _uidPrefixes[prefix] += 1;\n  return prefix + _uidPrefixes[prefix].toString();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\n// tslint:enable\n\n/**\n * Determine whether the input is an Array of Shapes.\n */\nexport function isArrayOfShapes(x: Shape|Shape[]): boolean {\n  return Array.isArray(x) && Array.isArray(x[0]);\n}\n\n/**\n * Special case of normalizing shapes to lists.\n *\n * @param x A shape or list of shapes to normalize into a list of Shapes.\n * @return A list of Shapes.\n */\nexport function normalizeShapeList(x: Shape|Shape[]): Shape[] {\n  if (x.length === 0) {\n    return [];\n  }\n  if (!Array.isArray(x[0])) {\n    return [x] as Shape[];\n  }\n  return x as Shape[];\n}\n\n/**\n * Helper function to obtain exactly one Tensor.\n * @param xs: A single `tf.Tensor` or an `Array` of `tf.Tensor`s.\n * @return A single `tf.Tensor`. If `xs` is an `Array`, return the first one.\n * @throws ValueError: If `xs` is an `Array` and its length is not 1.\n */\nexport function getExactlyOneTensor(xs: Tensor|Tensor[]): Tensor {\n  let x: Tensor;\n  if (Array.isArray(xs)) {\n    if (xs.length !== 1) {\n      throw new ValueError(`Expected Tensor length to be 1; got ${xs.length}`);\n    }\n    x = xs[0];\n  } else {\n    x = xs;\n  }\n  return x;\n}\n\n/**\n * Helper function to obtain exactly on instance of Shape.\n *\n * @param shapes Input single `Shape` or Array of `Shape`s.\n * @returns If input is a single `Shape`, return it unchanged. If the input is\n *   an `Array` containing exactly one instance of `Shape`, return the instance.\n *   Otherwise, throw a `ValueError`.\n * @throws ValueError: If input is an `Array` of `Shape`s, and its length is not\n *   1.\n */\nexport function getExactlyOneShape(shapes: Shape|Shape[]): Shape {\n  if (Array.isArray(shapes) && Array.isArray(shapes[0])) {\n    if (shapes.length === 1) {\n      shapes = shapes as Shape[];\n      return shapes[0];\n    } else {\n      throw new ValueError(`Expected exactly 1 Shape; got ${shapes.length}`);\n    }\n  } else {\n    return shapes as Shape;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {LayerVariable} from '../variables';\n\n/**\n * Count the elements in an Array of LayerVariables.\n *\n * @param weights: The LayerVariables of which the constituent numbers are to\n *   be counted.\n * @returns A count of the elements in all the LayerVariables\n */\nexport function countParamsInWeights(weights: LayerVariable[]): number {\n  let count = 0;\n  for (const weight of weights) {\n    if (weight.shape.length === 0) {\n      count += 1;\n    } else {\n      count += weight.shape.reduce((a, b) => a * b);\n    }\n  }\n  return count;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, Tensor, variableGrads} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId} from './backend/state';\nimport {getScopedTensorName, getUniqueTensorName} from './common';\nimport {Constraint} from './constraints';\nimport {NotImplementedError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {HasShape} from './types';\n\nconst DEFAULT_VARIABLE_NAME_PREFIX = 'Variable';\n\n/**\n * A `tf.layers.LayerVariable` is similar to a `tf.Tensor` in that it has a\n * dtype and shape, but its value is mutable.  The value is itself represented\n * as a`tf.Tensor`, and can be read with the `read()` method and updated with\n * the `write()` method.\n */\nexport class LayerVariable {\n  readonly dtype: DataType;\n  readonly shape: Shape;\n\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName: string;\n  private trainable_: boolean;\n\n  protected readonly val: tfc.Variable;\n  readonly constraint: Constraint;\n  /**\n   * Construct Variable from a `tf.Tensor`.\n   *\n   * If not explicitly named, the Variable will be given a name with the\n   * prefix 'Variable'. Variable names are unique. In the case of name\n   * collision, suffixies '_<num>' will be added to the name.\n   *\n   * @param val Initial value of the Variable.\n   * @param name Name of the variable. If `null` or `undefined` is provided, it\n   *   will default a name with the prefix 'Variable'.\n   * @param constraint Optional, projection function to be applied to the\n   * variable after optimize updates\n   * @throws ValueError if `name` is `null` or `undefined`.\n   */\n  constructor(\n      val: Tensor, dtype: DataType = 'float32',\n      name = DEFAULT_VARIABLE_NAME_PREFIX, trainable = true,\n      constraint: Constraint = null) {\n    this.dtype = dtype == null ? 'float32' : dtype;\n    this.shape = val.shape;\n    this.id = getNextUniqueTensorId();\n\n    name = name == null ? DEFAULT_VARIABLE_NAME_PREFIX : name;\n    this.originalName = getScopedTensorName(name);\n    this.name = getUniqueTensorName(this.originalName);\n\n    this.trainable_ = trainable;\n    this.constraint = constraint;\n\n    this.val = tfc.variable(val, this.trainable_, this.name, this.dtype);\n  }\n\n  /**\n   * Get a snapshot of the Variable's value.\n   *\n   * The returned value is a snapshot of the Variable's value at the time of\n   * the invocation. Future mutations in the value of the tensor will only\n   * be reflected by future calls to this method.\n   */\n  read(): Tensor {\n    this.assertNotDisposed();\n    return this.val;\n  }\n\n  /**\n   * Update the value of the Variable.\n   *\n   * @param newVal: The new value to update to. Must be consistent with the\n   *   dtype and shape of the Variable.\n   * @return This Variable.\n   */\n  write(newVal: Tensor) {\n    // TODO(cais): Once  TF.js Core supports Tensor.dtype, check dtype match.\n    this.assertNotDisposed();\n    checkShapesMatch(this.val, newVal);\n    // Skip updating if this is the exact same tensor.\n    if (this.val.id !== newVal.id) {\n      this.val.assign(newVal);\n      if (this.constraint != null) {\n        this.val.assign(this.constraint.apply(this.val));\n      }\n    }\n    return this;\n  }\n\n  /**\n   * Dispose this LayersVariable instance from memory.\n   */\n  dispose(): void {\n    this.assertNotDisposed();\n    this.val.dispose();\n  }\n\n  protected assertNotDisposed(): void {\n    if (this.val.isDisposed) {\n      throw new Error(`LayersVariable ${this.name} is already disposed.`);\n    }\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.trainable_ = trainable;\n    this.val.trainable = trainable;\n  }\n}\n\nfunction checkShapesMatch(x: HasShape, y: HasShape): void {\n  if (x.shape.toString() !== y.shape.toString()) {\n    throw new Error(\n        'Shape mismatch: ' + JSON.stringify(x.shape) + ' vs. ' +\n        JSON.stringify(y.shape));\n  }\n}\n\n/**\n * Create a Variable.\n * @param x The initial value of the `Variable`.\n * @param dtype optional, the type of the variable.\n * @param name optional, the name of the variable, default provided by\n * Variable.\n * @param constraint optional, a constraint to be applied after every update.\n * @return The newly instantiated `Variable`.\n */\nexport function variable(\n    x: Tensor, dtype?: DataType, name?: string,\n    constraint?: Constraint): LayerVariable {\n  return new LayerVariable(x, dtype, name, true, constraint);\n}\n\n/**\n * Instantiates an all-zeros Variable and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-zero Variable.\n */\nexport function zerosVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  return new LayerVariable(tfc.zeros(shape), dtype, name);\n}\n\n/**\n * Instantiates an all-zeros tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function zerosLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.zerosLike(x), dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-ones Variable.\n */\nexport function onesVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  const allocated = tfc.ones(shape);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function onesLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  const allocated = tfc.onesLike(x);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiate an identity matrix and returns it, as a Variable\n *\n * @param size Number of rows/columns.\n * @param dtype Data type of returned Variable.\n * @param name Name of returned Variable.\n * @return A Variable, an identity matrix.\n */\nexport function eyeVariable(\n    size: number, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.eye(size), dtype, name);\n}\n\n/**\n * Get a Variable with uniform distribution of values.\n * @param shape Shape of the tensor.\n * @param minval Lower bound of the uniform distribution.\n * @param maxval Upper bound of the uniform distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The uniform-random Variable.\n */\nexport function randomUniformVariable(\n    shape: Shape, minval: number, maxval: number, dtype?: DataType,\n    seed?: number, name = 'randomUniform'): LayerVariable {\n  return new LayerVariable(\n      tfc.randomUniform(shape, minval, maxval, dtype), dtype, name);\n}\n\n/**\n * Get a Variable with truncated-normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function truncatedNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'truncatedNormal'): LayerVariable {\n  // TODO(cais): Implement logic for dtype and seed once they are supported\n  // by deeplearn.js.\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormal does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.truncatedNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n/**\n * Get a Variable with normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function randomNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'randomNormal'): LayerVariable {\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormalVariable does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.randomNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n\n/**\n * Update the value of a Variable.\n * @param x The Variable to be updated.\n * @param xNew The new value to update to.\n * @return The Variable updated.\n */\nexport function update(x: LayerVariable, xNew: Tensor): LayerVariable {\n  return x.write(xNew);\n}\n\n/**\n * Update the value of a Variable by adding an increment.\n * @param x The Variable to be updated.\n * @param increment The incrment to add to `x`.\n * @return The Variable updated.\n */\nexport function updateAdd(x: LayerVariable, increment: Tensor): LayerVariable {\n  return x.write(tfc.add(x.read(), increment));\n}\n\n/**\n * Update the value of a Variable by subtracting a decrement.\n * @param x The Variable to be updated.\n * @param decrement The decrement to subtract from `x`.\n * @return The Variable updated.\n */\nexport function updateSub(x: LayerVariable, decrement: Tensor): LayerVariable {\n  return x.write(tfc.sub(x.read(), decrement));\n}\n\n/**\n * Get the values of an array of Variables.\n *\n * @param tensors An `Array` of `Variable`s to get the values of.\n * @return The values of the inputs, as an `Array` of`tf.Tensor`s.\n */\nexport function batchGetValue(xs: LayerVariable[]): Tensor[] {\n  return xs.map(x => x.read());\n}\n\n/**\n * Update the value of multiple Variables at once.\n *\n * @param variablesAndValues An `Array`, each element is of type\n *   [Variable, Tensor]. The first item is the\n *   `Variable` of which the value is to be updated. The second item\n *   carries the new value.\n */\nexport function batchSetValue(\n    variablesAndValues: Array<[LayerVariable, Tensor]>): void {\n  variablesAndValues.forEach(variableAndValue => {\n    const variable: LayerVariable = variableAndValue[0];\n    variable.write(variableAndValue[1]);\n  });\n}\n\n/**\n * Returns the gradients of `variables` w.r.t. the return value of `lossFn`.\n * @param lossFn A function which returns a Scalar to be used as the function\n *   value (i.e., numerator) for differentiation.\n * @param variables List of variables to be used as the independent variables\n *   (i.e., denominator) for differentiation.\n * @returns An Array of gradients tensors.\n */\nexport function gradients(\n    lossFn: () => tfc.Scalar, variables: LayerVariable[]): Tensor[] {\n  // TODO(cais): The return type signature can be simplified if deeplearn makes\n  //   the corresponding type public.\n  const variableList =\n      variables.map(variable => variable.read() as tfc.Variable);\n  const valudAndGrads = variableGrads(lossFn, variableList);\n  return variables.map(variable => valudAndGrads.grads[variable.name]);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {DataType, Scalar, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId, getUid} from '../backend/state';\nimport {getScopedTensorName, getUniqueTensorName, nameScope} from '../common';\nimport {Constraint} from '../constraints';\nimport {AttributeError, NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {getInitializer, Initializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs, RegularizerFn} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as types_utils from '../utils/types_utils';\nimport * as variable_utils from '../utils/variable_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\n\n// TODO(michaelterry): This is a stub until it's defined.\nexport type Op = (x: LayerVariable) => LayerVariable;\n\n/**\n * Constructor arguments for InputSpec.\n */\nexport interface InputSpecArgs {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n}\n\n/**\n * Specifies the ndim, dtype and shape of every input to a layer.\n *\n * Every layer should expose (if appropriate) an `inputSpec` attribute:\n * a list of instances of InputSpec (one per input tensor).\n *\n * A null entry in a shape is compatible with any dimension,\n * a null shape is compatible with any shape.\n */\nexport class InputSpec {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n\n  constructor(args: InputSpecArgs) {\n    this.dtype = args.dtype;\n    this.shape = args.shape;\n    /*\n      TODO(michaelterry): Could throw error if ndim and shape are both defined\n        (then backport).\n    */\n    if (args.shape != null) {\n      this.ndim = args.shape.length;\n    } else {\n      this.ndim = args.ndim;\n    }\n    this.maxNDim = args.maxNDim;\n    this.minNDim = args.minNDim;\n    this.axes = args.axes || {};\n  }\n}\n\n/**\n * `tf.SymbolicTensor` is a placeholder for a Tensor without any concrete value.\n *\n * They are most often encountered when building a graph of `Layer`s for a\n * a `tf.LayersModel` and the input data's shape, but not values are known.\n *\n * @doc {heading: 'Models', 'subheading': 'Classes'}\n */\nexport class SymbolicTensor {\n  /* A unique ID for the tensor to be able to differentiate tensors. */\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName?: string;\n  /**\n   * Rank/dimensionality of the tensor.\n   */\n  readonly rank: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  nodeIndex: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  tensorIndex: number;\n\n  /**\n   *\n   * @param dtype\n   * @param shape\n   * @param sourceLayer The Layer that produced this symbolic tensor.\n   * @param inputs The inputs passed to sourceLayer's __call__() method.\n   * @param nodeIndex\n   * @param tensorIndex\n   * @param callArgs The keyword arguments passed to the __call__() method.\n   * @param name\n   * @param outputTensorIndex The index of this tensor in the list of outputs\n   *   returned by apply().\n   */\n  constructor(\n      readonly dtype: DataType, readonly shape: Shape,\n      public sourceLayer: Layer, readonly inputs: SymbolicTensor[],\n      readonly callArgs: Kwargs, name?: string,\n      readonly outputTensorIndex?: number) {\n    this.id = getNextUniqueTensorId();\n    if (name != null) {\n      this.originalName = getScopedTensorName(name);\n      this.name = getUniqueTensorName(this.originalName);\n    }\n    this.rank = shape.length;\n  }\n}\n\n/**\n * Constructor arguments for Node.\n */\nexport interface NodeArgs {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`.\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n}\n\n/**\n * The type of the return value of Layer.dispose() and Container.dispose().\n */\nexport interface DisposeResult {\n  /**\n   * Reference count after the dispose call.\n   */\n  refCountAfterDispose: number;\n\n  /**\n   * Number of variables dispose in this dispose call.\n   */\n  numDisposedVariables: number;\n}\n\nlet _nextNodeID = 0;\n\n/**\n * A `Node` describes the connectivity between two layers.\n *\n * Each time a layer is connected to some new input,\n * a node is added to `layer.inboundNodes`.\n *\n * Each time the output of a layer is used by another layer,\n * a node is added to `layer.outboundNodes`.\n *\n * `nodeIndices` and `tensorIndices` are basically fine-grained coordinates\n * describing the origin of the `inputTensors`, verifying the following:\n *\n * `inputTensors[i] ==\n * inboundLayers[i].inboundNodes[nodeIndices[i]].outputTensors[\n *   tensorIndices[i]]`\n *\n * A node from layer A to layer B is added to:\n *     A.outboundNodes\n *     B.inboundNodes\n */\nexport class Node {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n\n  readonly id: number;\n\n  constructor(\n      args: NodeArgs,\n      // TODO(michaelterry): Define actual type for this.\n      public callArgs?: Kwargs) {\n    this.id = _nextNodeID++;\n    /*\n      Layer instance (NOT a list).\n      this is the layer that takes a list of input tensors\n      and turns them into a list of output tensors.\n      the current node will be added to\n      the inboundNodes of outboundLayer.\n    */\n    this.outboundLayer = args.outboundLayer;\n\n    /*\n        The following 3 properties describe where\n        the input tensors come from: which layers,\n        and for each layer, which node and which\n        tensor output of each node.\n    */\n\n    // List of layer instances.\n    this.inboundLayers = args.inboundLayers;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.nodeIndices = args.nodeIndices;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.tensorIndices = args.tensorIndices;\n\n    /*\n        Following 2 properties:\n        tensor inputs and outputs of outboundLayer.\n    */\n\n    // List of tensors. 1:1 mapping with inboundLayers.\n    this.inputTensors = args.inputTensors;\n    // List of tensors, created by outboundLayer.call().\n    this.outputTensors = args.outputTensors;\n\n    /*\n        Following 2 properties: input and output masks.\n        List of tensors, 1:1 mapping with inputTensor.\n    */\n    this.inputMasks = args.inputMasks;\n    // List of tensors, created by outboundLayer.computeMask().\n    this.outputMasks = args.outputMasks;\n\n    // Following 2 properties: input and output shapes.\n\n    // List of shape tuples, shapes of inputTensors.\n    this.inputShapes = args.inputShapes;\n    // List of shape tuples, shapes of outputTensors.\n    this.outputShapes = args.outputShapes;\n\n    // Add nodes to all layers involved.\n    for (const layer of args.inboundLayers) {\n      if (layer != null) {\n        layer.outboundNodes.push(this);\n      }\n    }\n    args.outboundLayer.inboundNodes.push(this);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const inboundNames: string[] = [];\n    for (const layer of this.inboundLayers) {\n      if (layer != null) {\n        inboundNames.push(layer.name);\n      } else {\n        inboundNames.push(null);\n      }\n    }\n    return {\n      outboundLayer: this.outboundLayer ? this.outboundLayer.name : null,\n      inboundLayers: inboundNames,\n      nodeIndices: this.nodeIndices,\n      tensorIndices: this.tensorIndices\n    };\n  }\n}\n\n/** Constructor arguments for Layer. */\nexport declare interface LayerArgs {\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  inputShape?: Shape;\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  batchInputShape?: Shape;\n  /**\n   * If `inputShape` is specified and `batchInputShape` is *not* specified,\n   * `batchSize` is used to construct the `batchInputShape`: `[batchSize,\n   * ...inputShape]`\n   */\n  batchSize?: number;\n  /**\n   * The data-type for this layer. Defaults to 'float32'.\n   * This argument is only applicable to input layers (the first layer of a\n   * model).\n   */\n  dtype?: DataType;\n  /** Name for this layer. */\n  name?: string;\n  /**\n   * Whether the weights of this layer are updatable by `fit`.\n   * Defaults to true.\n   */\n  trainable?: boolean;\n  /**\n   * Initial weight values of the layer.\n   */\n  weights?: Tensor[];\n  /** Legacy support. Do not use for new code. */\n  inputDType?: DataType;\n}\n\n// If necessary, add `output` arguments to the CallHook function.\n// This is currently used for testing only, but may be used for debugger-related\n// purposes in the future.\nexport type CallHook = (inputs: Tensor|Tensor[], kwargs: Kwargs) => void;\n\nlet _nextLayerID = 0;\n\n/**\n * A layer is a grouping of operations and weights that can be composed to\n * create a `tf.LayersModel`.\n *\n * Layers are constructed by using the functions under the\n * [tf.layers](#Layers-Basic) namespace.\n *\n * @doc {heading: 'Layers', subheading: 'Classes', namespace: 'layers'}\n */\nexport abstract class Layer extends serialization.Serializable {\n  /** Name for this layer. Must be unique within a model. */\n  name: string;\n  /**\n   * List of InputSpec class instances.\n   *\n   * Each entry describes one required input:\n   * - ndim\n   * - dtype\n   * A layer with `n` input tensors must have an `inputSpec` of length `n`.\n   */\n  inputSpec: InputSpec[];\n  supportsMasking: boolean;\n  /** Whether the layer weights will be updated during training. */\n  protected trainable_: boolean;\n  batchInputShape: Shape;\n  dtype: DataType;\n  initialWeights: Tensor[];\n\n  inboundNodes: Node[];\n  outboundNodes: Node[];\n\n  activityRegularizer: Regularizer;\n\n  protected _trainableWeights: LayerVariable[];\n  private _nonTrainableWeights: LayerVariable[];\n  private _losses: RegularizerFn[];\n  // TODO(cais): _updates is currently unused.\n  private _updates: Tensor[];\n  private _built: boolean;\n  private _callHook: CallHook = null;\n\n  private _addedWeightNames: string[] = [];\n\n  readonly id: number;\n\n  // Porting Notes: PyKeras does not have this property in this base Layer\n  //   class. Instead lets Layer subclass set it dynamically and checks the\n  //   value with `hasattr`. In tfjs-layers, we let this be a member of this\n  //   base class.\n  protected _stateful = false;\n\n  protected _refCount: number|null;\n\n  // A flag for whether fast (i.e., all-zero) weight initialization is to\n  // be used during `build()` call. This speeds up weight initialization\n  // by saving unnecessary calls to expensive initializers in cases where\n  // the initialized values will be overwritten by loaded weight values\n  // during model loading.\n  private fastWeightInitDuringBuild: boolean;\n\n  constructor(args: LayerArgs = {}) {\n    super();\n    this.id = _nextLayerID++;\n\n    this.activityRegularizer = null;\n\n    this.inputSpec = null;\n    this.supportsMasking = false;\n\n    // These properties will be set upon call of this.build()\n    this._trainableWeights = [];\n    this._nonTrainableWeights = [];\n    this._losses = [];\n    this._updates = [];\n    this._built = false;\n\n    /*\n      These lists will be filled via successive calls\n      to this.addInboundNode().\n     */\n    this.inboundNodes = [];\n    this.outboundNodes = [];\n\n    let name = args.name;\n    if (!name) {\n      const prefix = this.getClassName();\n      name = generic_utils.toSnakeCase(prefix) + '_' + getUid(prefix);\n    }\n    this.name = name;\n\n    this.trainable_ = args.trainable == null ? true : args.trainable;\n\n    if (args.inputShape != null || args.batchInputShape != null) {\n      /*\n        In this case we will later create an input layer\n        to insert before the current layer\n       */\n      let batchInputShape: Shape;\n      if (args.batchInputShape != null) {\n        batchInputShape = args.batchInputShape;\n      } else if (args.inputShape != null) {\n        let batchSize: number = null;\n        if (args.batchSize != null) {\n          batchSize = args.batchSize;\n        }\n        batchInputShape = [batchSize].concat(args.inputShape);\n      }\n      this.batchInputShape = batchInputShape;\n\n      // Set dtype.\n      let dtype = args.dtype;\n      if (dtype == null) {\n        dtype = args.inputDType;\n      }\n      if (dtype == null) {\n        dtype = 'float32';\n      }\n      this.dtype = dtype;\n    }\n\n    if (args.weights != null) {\n      this.initialWeights = args.weights;\n    } else {\n      this.initialWeights = null;\n    }\n\n    // The value of `_refCount` is initialized to null. When the layer is used\n    // in a symbolic way for the first time, it will be set to 1.\n    this._refCount = null;\n\n    this.fastWeightInitDuringBuild = false;\n  }\n\n  /**\n   * Converts a layer and its index to a unique (immutable type) name.\n   * This function is used internally with `this.containerNodes`.\n   * @param layer The layer.\n   * @param nodeIndex The layer's position (e.g. via enumerate) in a list of\n   *   nodes.\n   *\n   * @returns The unique name.\n   */\n  protected static nodeKey(layer: Layer, nodeIndex: number) {\n    return layer.name + '_ib-' + nodeIndex.toString();\n  }\n\n  /**\n   * Returns this.inboundNode at index nodeIndex.\n   *\n   * Porting note: This is a replacement for _get_node_attribute_at_index()\n   * @param nodeIndex\n   * @param attrName The name of the attribute related to request for this node.\n   */\n  private getNodeAtIndex(nodeIndex: number, attrName: string): Node {\n    if (this.inboundNodes.length === 0) {\n      throw new RuntimeError(\n          'The layer has never been called ' +\n          `and thus has no defined ${attrName}.`);\n    }\n    if (this.inboundNodes.length <= nodeIndex) {\n      throw new ValueError(\n          `Asked to get ${attrName} at node ${nodeIndex}, ` +\n          `but the layer has only ${this.inboundNodes.length} inbound nodes.`);\n    }\n    return this.inboundNodes[nodeIndex];\n  }\n\n  /**\n   * Retrieves the input tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple inputs).\n   */\n  getInputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple outputs).\n   */\n  getOutputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'output').outputTensors);\n  }\n\n  // Properties\n\n  /**\n   * Retrieves the input tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Input tensor or list of input tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get input(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer input\" ' +\n          'is ill-defined. ' +\n          'Use `getInputAt(nodeIndex)` instead.');\n    } else if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' is not connected, no input to return.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Output tensor or list of output tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get output(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has no inbound nodes.');\n    }\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer output\" ' +\n          'is ill-defined. ' +\n          'Use `getOutputAt(nodeIndex)` instead.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'output').outputTensors);\n  }\n\n  get losses(): RegularizerFn[] {\n    return this._losses;\n  }\n\n  /**\n   * Retrieves the Layer's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Layer.loss in PyKeras.\n    //   In PyKeras, Layer.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return this.losses.map(lossFn => lossFn());\n  }\n\n  get updates(): Tensor[] {\n    return this._updates;\n  }\n\n  get built(): boolean {\n    return this._built;\n  }\n\n  set built(built: boolean) {\n    this._built = built;\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this._trainableWeights.forEach(w => w.trainable = trainable);\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (this.trainable_) {\n      return this._trainableWeights.filter(w => w.trainable);\n    } else {\n      return [];\n    }\n  }\n\n  set trainableWeights(weights: LayerVariable[]) {\n    this._trainableWeights = weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    if (this.trainable) {\n      return this._trainableWeights.filter(w => !w.trainable)\n          .concat(this._nonTrainableWeights);\n    } else {\n      return this._trainableWeights.concat(this._nonTrainableWeights);\n    }\n  }\n\n  set nonTrainableWeights(weights: LayerVariable[]) {\n    this._nonTrainableWeights = weights;\n  }\n\n  /**\n   * The concatenation of the lists trainableWeights and nonTrainableWeights\n   * (in this order).\n   */\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  get stateful(): boolean {\n    return this._stateful;\n  }\n\n  /**\n   * Reset the states of the layer.\n   *\n   * This method of the base Layer class is essentially a no-op.\n   * Subclasses that are stateful (e.g., stateful RNNs) should override this\n   * method.\n   */\n  resetStates(): void {\n    if (!this.stateful) {\n      throw new Error(\n          'Cannot call the resetStates() method of a non-stateful Layer ' +\n          'object.');\n    }\n  }\n\n  /**\n   * Checks compatibility between the layer and provided inputs.\n   *\n   * This checks that the tensor(s) `input`\n   * verify the input assumptions of the layer\n   * (if any). If not, exceptions are raised.\n   *\n   * @param inputs Input tensor or list of input tensors.\n   *\n   * @exception ValueError in case of mismatch between\n   *   the provided inputs and the expectations of the layer.\n   */\n  protected assertInputCompatibility(inputs: Tensor|Tensor[]|SymbolicTensor|\n                                     SymbolicTensor[]): void {\n    inputs = generic_utils.toList(inputs);\n    if (this.inputSpec == null || this.inputSpec.length === 0) {\n      return;\n    }\n    const inputSpec = generic_utils.toList(this.inputSpec);\n    if (inputs.length !== inputSpec.length) {\n      throw new ValueError(\n          `Layer ${this.name} expects ${inputSpec.length} inputs, ` +\n          `but it received ${inputs.length} input tensors. ` +\n          `Input received: ${inputs}`);\n    }\n    for (let inputIndex = 0; inputIndex < inputs.length; inputIndex++) {\n      const x = inputs[inputIndex];\n      const spec: InputSpec = inputSpec[inputIndex];\n      if (spec == null) {\n        continue;\n      }\n\n      // Check ndim.\n      const ndim = x.rank;\n      if (spec.ndim != null) {\n        if (ndim !== spec.ndim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}: ` +\n              `expected ndim=${spec.ndim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.maxNDim != null) {\n        if (ndim > spec.maxNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected max_ndim=${spec.maxNDim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.minNDim != null) {\n        if (ndim < spec.minNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected min_ndim=${spec.minNDim}, found ndim=${ndim}.`);\n        }\n      }\n\n      // Check dtype.\n      if (spec.dtype != null) {\n        if (x.dtype !== spec.dtype) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name} ` +\n              `: expected dtype=${spec.dtype}, found dtype=${x.dtype}.`);\n        }\n      }\n\n      // Check specific shape axes.\n      if (spec.axes) {\n        const xShape = x.shape;\n        for (const key in spec.axes) {\n          const axis = Number(key);\n          const value = spec.axes[key];\n          // Perform Python-style slicing in case axis < 0;\n          // TODO(cais): Use https://github.com/alvivi/typescript-underscore to\n          // ensure type safety through Underscore calls.\n          const xShapeAtAxis =\n              axis >= 0 ? xShape[axis] : xShape[xShape.length + axis];\n          if (value != null && [value, null].indexOf(xShapeAtAxis) === -1) {\n            throw new ValueError(\n                `Input ${inputIndex} is incompatible with layer ` +\n                `${this.name}: expected axis ${axis} of input shape to ` +\n                `have value ${value} but got shape ${xShape}.`);\n          }\n        }\n      }\n\n      // Check shape.\n      if (spec.shape != null) {\n        for (let i = 0; i < spec.shape.length; ++i) {\n          const specDim = spec.shape[i];\n          const dim = x.shape[i];\n          if (specDim != null && dim != null) {\n            if (specDim !== dim) {\n              throw new ValueError(\n                  `Input ${inputIndex} is incompatible with layer ` +\n                  `${this.name}: expected shape=${spec.shape}, ` +\n                  `found shape=${x.shape}.`);\n            }\n          }\n        }\n      }\n    }\n  }\n\n  /**\n   * This is where the layer's logic lives.\n   *\n   * @param inputs Input tensor, or list/tuple of input tensors.\n   * @param kwargs Additional keyword arguments.\n   *\n   * @return A tensor or list/tuple of tensors.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return inputs;\n  }\n\n  protected invokeCallHook(inputs: Tensor|Tensor[], kwargs: Kwargs) {\n    if (this._callHook != null) {\n      this._callHook(inputs, kwargs);\n    }\n  }\n\n  /**\n   * Set call hook.\n   * This is currently used for testing only.\n   * @param callHook\n   */\n  setCallHook(callHook: CallHook) {\n    this._callHook = callHook;\n  }\n\n  /**\n   * Clear call hook.\n   * This is currently used for testing only.\n   */\n  clearCallHook() {\n    this._callHook = null;\n  }\n\n  /**\n   * Builds or executes a `Layer's logic.\n   *\n   * When called with `tf.Tensor`(s), execute the `Layer`s computation and\n   * return Tensor(s). For example:\n   *\n   * ```js\n   * const denseLayer = tf.layers.dense({\n   *   units: 1,\n   *   kernelInitializer: 'zeros',\n   *   useBias: false\n   * });\n   *\n   * // Invoke the layer's apply() method with a `tf.Tensor` (with concrete\n   * // numeric values).\n   * const input = tf.ones([2, 2]);\n   * const output = denseLayer.apply(input);\n   *\n   * // The output's value is expected to be [[0], [0]], due to the fact that\n   * // the dense layer has a kernel initialized to all-zeros and does not have\n   * // a bias.\n   * output.print();\n   * ```\n   *\n   * When called with `tf.SymbolicTensor`(s), this will prepare the layer for\n   * future execution.  This entails internal book-keeping on shapes of\n   * expected Tensors, wiring layers together, and initializing weights.\n   *\n   * Calling `apply` with `tf.SymbolicTensor`s are typically used during the\n   * building of non-`tf.Sequential` models. For example:\n   *\n   * ```js\n   * const flattenLayer = tf.layers.flatten();\n   * const denseLayer = tf.layers.dense({units: 1});\n   *\n   * // Use tf.layers.input() to obtain a SymbolicTensor as input to apply().\n   * const input = tf.input({shape: [2, 2]});\n   * const output1 = flattenLayer.apply(input);\n   *\n   * // output1.shape is [null, 4]. The first dimension is the undetermined\n   * // batch size. The second dimension comes from flattening the [2, 2]\n   * // shape.\n   * console.log(JSON.stringify(output1.shape));\n   *\n   * // The output SymbolicTensor of the flatten layer can be used to call\n   * // the apply() of the dense layer:\n   * const output2 = denseLayer.apply(output1);\n   *\n   * // output2.shape is [null, 1]. The first dimension is the undetermined\n   * // batch size. The second dimension matches the number of units of the\n   * // dense layer.\n   * console.log(JSON.stringify(output2.shape));\n   *\n   * // The input and output and be used to construct a model that consists\n   * // of the flatten and dense layers.\n   * const model = tf.model({inputs: input, outputs: output2});\n   * ```\n   *\n   * @param inputs a `tf.Tensor` or `tf.SymbolicTensor` or an Array of them.\n   * @param kwargs Additional keyword arguments to be passed to `call()`.\n   *\n   * @return Output of the layer's `call` method.\n   *\n   * @exception ValueError error in case the layer is missing shape information\n   *   for its `build` call.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  // Porting Note: This is a replacement for __call__() in Python.\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    kwargs = kwargs || {};\n\n    this.assertNotDisposed();\n\n    // Ensure inputs are all the same type.\n    const inputsList = generic_utils.toList(inputs);\n\n    let allAreSymbolic = true;\n    for (const input of inputsList) {\n      if (!(input instanceof SymbolicTensor)) {\n        allAreSymbolic = false;\n        break;\n      }\n    }\n    let noneAreSymbolic = true;\n    for (const input of inputsList) {\n      if (input instanceof SymbolicTensor) {\n        noneAreSymbolic = false;\n        break;\n      }\n    }\n\n    if (allAreSymbolic === noneAreSymbolic) {\n      throw new ValueError(\n          'Arguments to apply() must be all ' +\n          'SymbolicTensors or all Tensors');\n    }\n\n    // TODO(michaelterry): nameScope() may not be necessary.\n    return nameScope(this.name, () => {\n      // Handle laying building (weight creating, input spec locking).\n      if (!this.built) {\n        /*\n          Throw exceptions in case the input is not compatible\n          with the inputSpec specified in the layer constructor.\n         */\n        this.assertInputCompatibility(inputs);\n\n        // Collect input shapes to build layer.\n        const inputShapes: Shape[] = [];\n        for (const xElem of generic_utils.toList(inputs)) {\n          inputShapes.push(xElem.shape);\n        }\n        this.build(generic_utils.singletonOrArray(inputShapes));\n        this.built = true;\n\n        // Load weights that were specified at layer instantiation.\n        if (this.initialWeights) {\n          this.setWeights(this.initialWeights);\n        }\n\n        if (this._refCount === null && noneAreSymbolic) {\n          // The first use of this layer is a non-symbolic call, set ref count\n          // to 1 so the Layer can be properly disposed if its dispose() method\n          // is called.\n          this._refCount = 1;\n        }\n      }\n\n      /*\n        Throw exceptions in case the input is not compatible\n        with the inputSpec set at build time.\n      */\n      this.assertInputCompatibility(inputs);\n\n      // Handle mask propagation.\n      // TODO(michaelterry): Mask propagation not currently implemented.\n\n      // Actually call the layer, collecting output(s), mask(s), and shape(s).\n      if (noneAreSymbolic) {\n        let output = this.call(inputs as Tensor | Tensor[], kwargs);\n        // TODO(michaelterry): Compute the outputMask\n\n        // If the layer returns tensors from its inputs, unmodified,\n        // we copy them to avoid loss of tensor metadata.\n        const outputList: Tensor[] = generic_utils.toList(output);\n        const outputListCopy: Tensor[] = [];\n        // TODO(michaelterry): This copying may not be necessary given our eager\n        // backend.\n        for (let x of outputList) {\n          if (inputsList.indexOf(x) !== -1) {\n            x = x.clone();\n          }\n          outputListCopy.push(x);\n        }\n        output = generic_utils.singletonOrArray(outputListCopy);\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        // TODO(michaelterry): Call addInboundNode()?\n        return output;\n      } else {\n        const inputShape = collectInputShape(inputs);\n        const outputShape = this.computeOutputShape(inputShape);\n        let output: SymbolicTensor|SymbolicTensor[];\n        const outputDType = guessOutputDType(inputs);\n        this.warnOnIncompatibleInputShape(\n            Array.isArray(inputs) ? inputShape[0] as Shape :\n                                    inputShape as Shape);\n\n        if (outputShape != null && outputShape.length > 0 &&\n            Array.isArray(outputShape[0])) {\n          // We have multiple output shapes. Create multiple output tensors.\n          output = (outputShape as Shape[])\n                       .map(\n                           (shape, index) => new SymbolicTensor(\n                               outputDType, shape, this,\n                               generic_utils.toList(inputs), kwargs, this.name,\n                               index));\n        } else {\n          output = new SymbolicTensor(\n              outputDType, outputShape as Shape, this,\n              generic_utils.toList(inputs), kwargs, this.name);\n        }\n\n        /*\n          Add an inbound node to the layer, so that it keeps track\n          of the call and of all new variables created during the call.\n          This also updates the layer history of the output tensor(s).\n          If the input tensor(s) had no previous history,\n          this does nothing.\n        */\n        this.addInboundNode(\n            inputs as SymbolicTensor | SymbolicTensor[], output, null, null,\n            inputShape, outputShape, kwargs);\n        this._refCount++;\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        return output;\n      }\n    });\n  }\n\n  /**\n   * Check compatibility between input shape and this layer's batchInputShape.\n   *\n   * Print warning if any incompatibility is found.\n   *\n   * @param inputShape Input shape to be checked.\n   */\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {\n    if (this.batchInputShape == null) {\n      return;\n    } else if (inputShape.length !== this.batchInputShape.length) {\n      console.warn(\n          `The rank of the input tensor provided (shape: ` +\n          `${JSON.stringify(inputShape)}) does not match that of the ` +\n          `batchInputShape (${JSON.stringify(this.batchInputShape)}) ` +\n          `of the layer ${this.name}`);\n    } else {\n      let dimMismatch = false;\n      this.batchInputShape.forEach((dimension, i) => {\n        if (dimension != null && inputShape[i] != null &&\n            inputShape[i] !== dimension) {\n          dimMismatch = true;\n        }\n      });\n      if (dimMismatch) {\n        console.warn(\n            `The shape of the input tensor ` +\n            `(${JSON.stringify(inputShape)}) does not ` +\n            `match the expectation of layer ${this.name}: ` +\n            `${JSON.stringify(this.batchInputShape)}`);\n      }\n    }\n  }\n\n  /**\n   * Retrieves the output shape(s) of a layer.\n   *\n   * Only applicable if the layer has only one inbound node, or if all inbound\n   * nodes have the same output shape.\n   *\n   * @returns Output shape or shapes.\n   * @throws AttributeError: if the layer is connected to more than one incoming\n   *   nodes.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  get outputShape(): Shape|Shape[] {\n    if (this.inboundNodes == null || this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `The layer ${this.name} has never been called and thus has no ` +\n          `defined output shape.`);\n    }\n    const allOutputShapes: string[] = [];\n    for (const node of this.inboundNodes) {\n      const shapeString = JSON.stringify(node.outputShapes);\n      if (allOutputShapes.indexOf(shapeString) === -1) {\n        allOutputShapes.push(shapeString);\n      }\n    }\n    if (allOutputShapes.length === 1) {\n      const outputShapes = this.inboundNodes[0].outputShapes;\n      if (Array.isArray(outputShapes) && Array.isArray(outputShapes[0]) &&\n          outputShapes.length === 1) {\n        return (outputShapes as Shape[])[0];\n      } else {\n        return outputShapes;\n      }\n\n    } else {\n      throw new AttributeError(\n          `The layer ${this.name} has multiple inbound nodes with different ` +\n          `output shapes. Hence the notion of \"output shape\" is ill-defined ` +\n          `for the layer.`);\n      // TODO(cais): Implement getOutputShapeAt().\n    }\n  }\n\n  /**\n   * Counts the total number of numbers (e.g., float32, int32) in the\n   * weights.\n   *\n   * @returns An integer count.\n   * @throws RuntimeError: If the layer is not built yet (in which case its\n   *   weights are not defined yet.)\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  countParams(): number {\n    if (!this.built) {\n      throw new RuntimeError(\n          `You tried to call countParams() on ${this.name}, ` +\n          `but the layer is not built yet. Build it first by calling ` +\n          `build(batchInputShape).`);\n    }\n    return variable_utils.countParamsInWeights(this.weights);\n  }\n\n  /**\n   * Creates the layer weights.\n   *\n   * Must be implemented on all layers that have weights.\n   *\n   * Called when apply() is called to construct the weights.\n   *\n   * @param inputShape A `Shape` or array of `Shape` (unused).\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  build(inputShape: Shape|Shape[]) {\n    this.built = true;\n  }\n\n  /**\n   * Returns the current values of the weights of the layer.\n   *\n   * @param trainableOnly Whether to get the values of only trainable weights.\n   * @returns Weight values as an `Array` of `tf.Tensor`s.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getWeights(trainableOnly = false): Tensor[] {\n    return batchGetValue(trainableOnly ? this.trainableWeights : this.weights);\n  }\n\n  /**\n   * Sets the weights of the layer, from Tensors.\n   *\n   * @param weights a list of Tensors. The number of arrays and their shape\n   *   must match number of the dimensions of the weights of the layer (i.e.\n   *   it should match the output of `getWeights`).\n   *\n   * @exception ValueError If the provided weights list does not match the\n   *   layer's specifications.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  setWeights(weights: Tensor[]): void {\n    tidy(() => {\n      const params = this.weights;\n      if (params.length !== weights.length) {\n        // TODO(cais): Restore the following and use `providedWeights`, instead\n        // of `weights` in the error message, once the deeplearn.js bug is\n        // fixed: https://github.com/PAIR-code/deeplearnjs/issues/498 const\n        // providedWeights = JSON.stringify(weights).substr(0, 50);\n        throw new ValueError(\n            `You called setWeights(weights) on layer \"${this.name}\" ` +\n            `with a weight list of length ${weights.length}, ` +\n            `but the layer was expecting ${params.length} weights. ` +\n            `Provided weights: ${weights}...`);\n      }\n      if (params.length === 0) {\n        return;\n      }\n      const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n      const paramValues = batchGetValue(params);\n      for (let i = 0; i < paramValues.length; ++i) {\n        const pv = paramValues[i];\n        const p = params[i];\n        const w = weights[i];\n        if (!util.arraysEqual(pv.shape, w.shape)) {\n          throw new ValueError(\n              `Layer weight shape ${pv.shape} ` +\n              `not compatible with provided weight shape ${w.shape}`);\n        }\n        weightValueTuples.push([p, w]);\n      }\n      batchSetValue(weightValueTuples);\n    });\n  }\n\n  /**\n   * Adds a weight variable to the layer.\n   *\n   * @param name Name of the new weight variable.\n   * @param shape The shape of the weight.\n   * @param dtype The dtype of the weight.\n   * @param initializer An initializer instance.\n   * @param regularizer A regularizer instance.\n   * @param trainable Whether the weight should be trained via backprop or not\n   *   (assuming that the layer itself is also trainable).\n   * @param constraint An optional trainable.\n   * @return The created weight variable.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  protected addWeight(\n      name: string, shape: Shape, dtype?: DataType, initializer?: Initializer,\n      regularizer?: Regularizer, trainable?: boolean,\n      constraint?: Constraint): LayerVariable {\n    // Reject duplicate weight names.\n    if (this._addedWeightNames.indexOf(name) !== -1) {\n      throw new ValueError(\n          `Duplicate weight name ${name} for layer ${this.name}`);\n    }\n    this._addedWeightNames.push(name);\n\n    if (dtype == null) {\n      dtype = 'float32';\n    }\n\n    if (this.fastWeightInitDuringBuild) {\n      initializer = getInitializer('zeros');\n    }\n    const initValue = initializer.apply(shape, dtype);\n    const weight =\n        new LayerVariable(initValue, dtype, name, trainable, constraint);\n    initValue.dispose();\n    // Request backend not to dispose the weights of the model on scope() exit.\n    if (regularizer != null) {\n      this.addLoss(() => regularizer.apply(weight.read()));\n    }\n    if (trainable == null) {\n      trainable = true;\n    }\n    if (trainable) {\n      this._trainableWeights.push(weight);\n    } else {\n      this._nonTrainableWeights.push(weight);\n    }\n    return weight;\n  }\n\n  /**\n   * Set the fast-weight-initialization flag.\n   *\n   * In cases where the initialized weight values will be immediately\n   * overwritten by loaded weight values during model loading, setting\n   * the flag to `true` saves unnecessary calls to potentially expensive\n   * initializers and speeds up the loading process.\n   *\n   * @param value Target value of the flag.\n   */\n  setFastWeightInitDuringBuild(value: boolean) {\n    this.fastWeightInitDuringBuild = value;\n  }\n\n  /**\n   * Add losses to the layer.\n   *\n   * The loss may potentionally be conditional on some inputs tensors,\n   * for instance activity losses are conditional on the layer's inputs.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  addLoss(losses: RegularizerFn|RegularizerFn[]): void {\n    if (losses == null || Array.isArray(losses) && losses.length === 0) {\n      return;\n    }\n    // Update this.losses\n    losses = generic_utils.toList(losses);\n    if (this._losses !== undefined && this._losses !== null) {\n      this.losses.push(...losses);\n    }\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (!this.supportsMasking) {\n      if (mask != null) {\n        if (Array.isArray(mask)) {\n          mask.forEach(maskElement => {\n            if (maskElement != null) {\n              throw new TypeError(\n                  `Layer ${this.name} does not support masking, ` +\n                  'but was passed an inputMask.');\n            }\n          });\n        } else {\n          throw new TypeError(\n              `Layer ${this.name} does not support masking, ` +\n              'but was passed an inputMask.');\n        }\n      }\n      // masking not explicitly supported: return null as mask\n      return null;\n    }\n    // if masking is explictly supported, by default\n    // carry over the input mask\n    return mask;\n  }\n\n  /**\n   * Internal method to create an inbound node for the layer.\n   *\n   * @param inputTensors List of input tensors.\n   * @param outputTensors List of output tensors.\n   * @param inputMasks List of input masks (a mask can be a tensor, or null).\n   * @param outputMasks List of output masks (a mask can be a tensor, or null).\n   * @param inputShapes List of input shape tuples.\n   * @param outputShapes List of output shape tuples.\n   * @param kwargs Dictionary of keyword arguments that were passed to the\n   *   `call` method of the layer at the call that created the node.\n   */\n  private addInboundNode(\n      inputTensors: SymbolicTensor|SymbolicTensor[],\n      outputTensors: SymbolicTensor|SymbolicTensor[],\n      inputMasks: Tensor|Tensor[], outputMasks: Tensor|Tensor[],\n      inputShapes: Shape|Shape[], outputShapes: Shape|Shape[],\n      kwargs: {} = null): void {\n    const inputTensorList: SymbolicTensor[] =\n        generic_utils.toList(inputTensors);\n    outputTensors = generic_utils.toList(outputTensors);\n    inputMasks = generic_utils.toList(inputMasks);\n    outputMasks = generic_utils.toList(outputMasks);\n    inputShapes = types_utils.normalizeShapeList(inputShapes);\n    outputShapes = types_utils.normalizeShapeList(outputShapes);\n\n    // Collect input tensor(s) coordinates.\n    const inboundLayers: Layer[] = [];\n    const nodeIndices: number[] = [];\n    const tensorIndices: number[] = [];\n    for (const x of inputTensorList) {\n      /*\n       * TODO(michaelterry): Keras adds this value to tensors; it's not\n       * clear whether we'll use this or not.\n       */\n      inboundLayers.push(x.sourceLayer);\n      nodeIndices.push(x.nodeIndex);\n      tensorIndices.push(x.tensorIndex);\n    }\n\n    // Create node, add it to inbound nodes.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node(\n        {\n          outboundLayer: this,\n          inboundLayers,\n          nodeIndices,\n          tensorIndices,\n          inputTensors: inputTensorList,\n          outputTensors,\n          inputMasks,\n          outputMasks,\n          inputShapes,\n          outputShapes\n        },\n        kwargs);\n\n    // Update tensor history\n    for (let i = 0; i < outputTensors.length; i++) {\n      // TODO(michaelterry: _uses_learning_phase not tracked.\n      outputTensors[i].sourceLayer = this;\n      outputTensors[i].nodeIndex = this.inboundNodes.length - 1;\n      outputTensors[i].tensorIndex = i;\n    }\n  }\n\n  /**\n   * Returns the config of the layer.\n   *\n   * A layer config is a TS dictionary (serializable)\n   * containing the configuration of a layer.\n   * The same layer can be reinstantiated later\n   * (without its trained weights) from this configuration.\n   *\n   * The config of a layer does not include connectivity\n   * information, nor the layer class name.  These are handled\n   * by 'Container' (one layer of abstraction above).\n   *\n   * Porting Note: The TS dictionary follows TS naming standrds for\n   * keys, and uses tfjs-layers type-safe Enums.  Serialization methods\n   * should use a helper function to convert to the pythonic storage\n   * standard. (see serialization_utils.convertTsToPythonic)\n   *\n   * @returns TS dictionary of configuration.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getConfig(): serialization.ConfigDict {\n    const config:\n        serialization.ConfigDict = {name: this.name, trainable: this.trainable};\n    if (this.batchInputShape != null) {\n      config['batchInputShape'] = this.batchInputShape;\n    }\n    if (this.dtype != null) {\n      config['dtype'] = this.dtype;\n    }\n    return config;\n  }\n\n  /**\n   * Dispose the weight variables that this Layer instance holds.\n   *\n   * @returns {number} Number of disposed variables.\n   */\n  protected disposeWeights(): number {\n    this.weights.forEach(weight => weight.dispose());\n    return this.weights.length;\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Layer '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose layer's weights.\n   *\n   * This method decrease the reference count of the Layer object by 1.\n   *\n   * A Layer is reference-counted. Its reference count is incremented by 1\n   * the first item its `apply()` method is called and when it becomes a part\n   * of a new `Node` (through calling the `apply()`) method on a\n   * `tf.SymbolicTensor`).\n   *\n   * If the reference count of a Layer becomes 0, all the weights will be\n   * disposed and the underlying memory (e.g., the textures allocated in WebGL)\n   * will be freed.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * weights of the Layer will *not* be disposed.\n   *\n   * After a Layer is disposed, it cannot be used in calls such as `apply()`,\n   * `getWeights()` or `setWeights()` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the Container after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the layer has already\n   *   been disposed.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  dispose(): DisposeResult {\n    if (!this.built) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been ` +\n          `built yet.`);\n    }\n\n    if (this._refCount === null) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been used ` +\n          `yet.`);\n    }\n\n    this.assertNotDisposed();\n\n    let numDisposedVariables = 0;\n    if (--this._refCount === 0) {\n      numDisposedVariables = this.disposeWeights();\n    }\n\n    return {refCountAfterDispose: this._refCount, numDisposedVariables};\n  }\n}\n\n/**\n * Collects the input shape(s) of a list of `tf.Tensor`s or\n * `tf.SymbolicTensor`s.\n *\n * TODO(michaelterry): Update PyKeras docs (backport).\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return List of shape tuples (or single tuple), one tuple per input.\n */\nfunction collectInputShape(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                           Tensor[]): Shape|Shape[] {\n  inputTensors =\n      generic_utils.toList(inputTensors) as SymbolicTensor[] | Tensor[];\n  const shapes: Shape[] = [];\n  for (const x of inputTensors) {\n    shapes.push(x.shape);\n  }\n  return generic_utils.singletonOrArray(shapes);\n}\n\n/**\n * Guesses output dtype based on inputs.\n *\n * At present, just returns 'float32' for any input.\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return The guessed DType. At present, always returns 'float32'.\n */\nfunction guessOutputDType(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                          Tensor[]): DataType {\n  return 'float32';\n}\n\n/**\n * Returns the list of input tensors necessary to compute `tensor`.\n *\n * Output will always be a list of tensors (potentially with 1 element).\n *\n * @param tensor The tensor to start from.\n * @param layer Origin layer of the tensor.\n * @param nodeIndex Origin node index of the tensor.\n *\n * @return Array of input tensors.\n */\nexport function getSourceInputs(\n    tensor: SymbolicTensor, layer?: Layer,\n    nodeIndex?: number): SymbolicTensor[] {\n  if (layer == null || (nodeIndex != null && nodeIndex > 0)) {\n    layer = tensor.sourceLayer;\n    nodeIndex = tensor.nodeIndex;\n  }\n  if (layer.inboundNodes.length === 0) {\n    return [tensor];\n  } else {\n    const node = layer.inboundNodes[nodeIndex];\n    if (node.inboundLayers.length === 0) {\n      return node.inputTensors;\n    } else {\n      const sourceTensors: SymbolicTensor[] = [];\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const x = node.inputTensors[i];\n        const layer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const previousSources = getSourceInputs(x, layer, nodeIndex);\n        // Avoid input redundancy.\n        for (const x of previousSources) {\n          if (sourceTensors.indexOf(x) === -1) {\n            sourceTensors.push(x);\n          }\n        }\n      }\n      return sourceTensors;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\n\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/**\n * Constructor arguments for InputLayer.\n *\n * Note: You should provide only inputShape or batchInputShape (not both).\n * If only inputShape is provided, then the batchInputShape is determined by\n * the batchSize argument and the inputShape: [batchSize].concat(inputShape).\n */\nexport declare interface InputLayerArgs {\n  /** Input shape, not including the batch axis. */\n  inputShape?: Shape;\n  /** Optional input batch size (integer or null). */\n  batchSize?: number;\n  /** Batch input shape, including the batch axis. */\n  batchInputShape?: Shape;\n  /** Datatype of the input.  */\n  dtype?: DataType;\n  /**\n   * Whether the placeholder created is meant to be sparse.\n   */\n  sparse?: boolean;  // TODO(michaelterry): Not clear whether we'll need this.\n\n  /** Name of the layer. */\n  name?: string;\n}\n\nexport class InputLayer extends Layer {\n  /** @nocollapse */\n  static readonly className = 'InputLayer';\n  sparse: boolean;\n  constructor(args: InputLayerArgs) {\n    super({\n      dtype: args.dtype,\n      name: args.name != null ? args.name : getUid('input').toString()\n    });\n    // Normalize config.batchSize and config.sparse\n    if (args.batchSize == null) {\n      args.batchSize = null;\n    }\n    if (args.sparse == null) {\n      args.sparse = false;\n    }\n\n    this.trainable = false;\n    this.built = true;\n    this.sparse = args.sparse;\n\n    if (args.inputShape != null && args.batchInputShape != null) {\n      throw new ValueError(\n          'Only provide the inputShape OR ' +\n          'batchInputShape argument to inputLayer, not both at the same time.');\n    }\n    let batchInputShape = args.batchInputShape;\n    if (batchInputShape == null) {\n      if (args.inputShape == null) {\n        throw new ValueError(\n            'An InputLayer should be passed either a ' +\n            '`batchInputShape` or an `inputShape`.');\n      } else {\n        batchInputShape = [args.batchSize].concat(args.inputShape);\n      }\n    } else {\n      // TODO(michaelterry): Backport to PyKeras\n      if (args.batchSize != null) {\n        throw new ValueError(\n            'Cannot specify batchSize if batchInputShape is ' +\n            'specified when creating an InputLayer.');\n      }\n    }\n\n    const dtype = args.dtype || 'float32';\n\n    this.batchInputShape = batchInputShape;\n    this.dtype = dtype;\n    // TODO(michaelterry): Backport this to PyKeras?\n    this.inputSpec = [{shape: batchInputShape}];\n\n    const inputTensor = new SymbolicTensor(\n        this.dtype, this.batchInputShape, this, [], {}, this.name);\n    inputTensor.nodeIndex = 0;\n    inputTensor.tensorIndex = 0;\n\n    // Create an input node to add to this.outboundNode.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: [inputTensor],\n      outputTensors: [inputTensor],\n      inputMasks: [null],\n      outputMasks: [null],\n      inputShapes: [batchInputShape],\n      outputShapes: [batchInputShape]\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor {\n    throw new ValueError(\n        'Cannot pass any input to an ' +\n        `InputLayer's apply() method. InputLayer name: ${this.name}`);\n  }\n\n  dispose(): DisposeResult {\n    // dispose() for InputLayer is overridden as no-op.\n    return {refCountAfterDispose: this._refCount, numDisposedVariables: 0};\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      batchInputShape: this.batchInputShape,\n      dtype: this.dtype,\n      sparse: this.sparse,\n      name: this.name\n    };\n  }\n}\nserialization.registerClass(InputLayer);\n\n/**\n * Config for the Input function.\n *\n * Note: You should provide only shape or batchShape (not both).\n * If only shape is provided, then the batchShape becomes\n * [null].concat(inputShape).\n */\nexport interface InputConfig {\n  /**\n   * A shape, not including the batch size. For instance, `shape=[32]`\n   * indicates that the expected input will be batches of 32-dimensional\n   * vectors.\n   */\n  shape?: Shape;\n  /**\n   * A shape tuple (integer), including the batch size. For instance,\n   * `batchShape=[10, 32]` indicates that the expected input will be batches of\n   * 10 32-dimensional vectors. `batchShape=[null, 32]` indicates batches of an\n   * arbitrary number of 32-dimensional vectors.\n   */\n  batchShape?: Shape;\n  /**\n   * An optional name string for the layer. Should be unique in a model (do not\n   * reuse the same name twice). It will be autogenerated if it isn't provided.\n   */\n  name?: string;\n  dtype?: DataType;\n  /**\n   * A boolean specifying whether the placeholder to be created is sparse.\n   */\n  sparse?: boolean;\n}\n\nexport function Input(config: InputConfig): SymbolicTensor {\n  if (config.batchShape == null && config.shape == null) {\n    throw new Error(\n        'Please provide to Input either a `shape`' +\n        ' or a `batchShape` argument. Note that ' +\n        '`shape` does not include the batch ' +\n        'dimension.');\n  }\n  if (config.batchShape != null && config.shape != null) {\n    // TODO(michaelterry): Backport to PyKeras.\n    throw new ValueError(\n        'Please provide either a `shape` or `batchShape` ' +\n        'argument to Input, but not both.');\n  }\n  let batchShape = config.batchShape;\n  if (config.shape != null && batchShape == null) {\n    batchShape = [null].concat(config.shape);\n  }\n\n  let dtype = config.dtype;\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n\n  const inputLayer = new InputLayer({\n    batchInputShape: batchShape,\n    name: config.name,\n    dtype,\n    sparse: config.sparse\n  });\n\n  const outputs = inputLayer.inboundNodes[0].outputTensors;\n  return outputs[0];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, Scalar} from '@tensorflow/tfjs-core';\n\n/**\n * Logs in which values can be either numbers or Tensors (Scalars).\n *\n * Used internally.\n */\nexport type UnresolvedLogs = {\n  [key: string]: number|Scalar;\n};\n\n/**\n * Turn any Scalar values in a Logs object into actual number values.\n *\n * @param logs The `Logs` object to be resolved in place.\n */\nexport async function resolveScalarsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n  const keys: string[] = [];\n  const scalarsToDispose: Scalar[] = [];\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      const valueScalar = value;\n      promises.push(valueScalar.data());\n      keys.push(key);\n      scalarsToDispose.push(valueScalar);\n    }\n  }\n  if (promises.length > 0) {\n    const values = await Promise.all(promises);\n    for (let i = 0; i < values.length; ++i) {\n      logs[keys[i]] = values[i][0];\n    }\n    // Dispose the original scalar tensors.\n    dispose(scalarsToDispose);\n  }\n}\n\n/**\n * Dispose all Tensors in an UnresolvedLogs object.\n *\n * @param logs An `UnresolvedLogs` object potentially containing `tf.Tensor`s in\n *   places where the values can be `tf.Tensor` or `number`.\n */\nexport function disposeTensorsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      value.dispose();\n    }\n  }\n}\n\n/**\n * Logs in which values can only be numbers.\n *\n * Used when calling client-provided custom callbacks.\n */\nexport type Logs = {\n  [key: string]: number;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {add, div, keep, mul, nextFrame, Scalar, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Container} from './engine/container';\nimport {ValueError} from './errors';\nimport {Logs, resolveScalarsInLogs, UnresolvedLogs} from './logs';\nimport * as generic_utils from './utils/generic_utils';\n\n/** Verbosity logging level when fitting a model. */\nexport enum ModelLoggingVerbosity {\n  SILENT = 0,\n  VERBOSE = 1\n}\n\n/** How often to yield to the main thread when training (in ms). */\nexport const DEFAULT_YIELD_EVERY_MS = 125;\n\nexport type Params = {\n  [key: string]: number|string|boolean|number[]|string[]|boolean[];\n};\n\nexport type YieldEveryOptions = 'auto'|'batch'|'epoch'|'never'|number;\n\n/**\n * Abstract base class used to build new callbacks.\n *\n * The `logs` dictionary that callback methods take as argument will contain\n * keys for quantities relevant to the current batch or epoch.\n *\n * Currently, the `.fit()` method of the `Sequential` model class\n * will include the following quantities in the `logs` that\n * it passes to its callbacks:\n *\n * onEpochEnd: Logs include `acc` and `loss`, and optionally include `valLoss`\n *   (if validation is enabled in `fit`), and `valAcc` (if validation and\n *   accuracy monitoring are enabled).\n * onBatchBegin: Logs include `size`, the number of samples in the current\n *   batch.\n * onBatchEnd: Logs include `loss`, and optionally `acc` (if accuracy monitoring\n *   is enabled).\n */\nexport abstract class BaseCallback {\n  // TODO(michaelterry): This type is a best guess.\n  validationData: Tensor|Tensor[] = null;\n  /**\n   * Training parameters (eg. verbosity, batch size, number of epochs...).\n   */\n  params: Params;\n\n  setParams(params: Params): void {\n    this.params = params;\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {}\n\n  async onTrainBegin(logs?: UnresolvedLogs) {}\n\n  async onTrainEnd(logs?: UnresolvedLogs) {}\n\n  // LayersModel needs to call Callback.setModel(), but cannot actually depend\n  // on Callback because that creates a cyclic dependency.  Providing this no-op\n  // method on BaseCallback breaks the cycle: this way LayersModel can depend on\n  // BaseCallback but not on Callback.  The argument is typed as `Container`\n  // (the superclass of LayersModel) to avoid recapitulating the cycle. Callback\n  // overrides this method and enforces that the argument is really a\n  // LayersModel.\n  setModel(model: Container): void {\n    // Do nothing. Use Callback instead of BaseCallback to track the model.\n  }\n}\n\n/**\n * Container abstracting a list of callbacks.\n */\nexport class CallbackList {\n  callbacks: BaseCallback[];\n  queueLength: number;\n\n  // TODO(cais): When the need arises, uncomment the following lines and\n  // implement the queue for time values.\n  // private deltaTBatch: number;\n  // private deltaTsBatchBegin: Array<number>;\n  // private deltaTsBatchEnd: Array<number>;\n\n  /**\n   * Constructor of CallbackList.\n   * @param callbacks Array of `Callback` instances.\n   * @param queueLength Queue length for keeping running statistics over\n   *   callback execution time.\n   */\n  constructor(callbacks?: BaseCallback[], queueLength = 10) {\n    // TODO(cais): Make use of queueLength when implementing the queue for time\n    // values.\n    if (callbacks == null) {\n      callbacks = [];\n    }\n    this.callbacks = callbacks;\n    this.queueLength = queueLength;\n  }\n\n  append(callback: BaseCallback): void {\n    this.callbacks.push(callback);\n  }\n\n  setParams(params: Params): void {\n    for (const callback of this.callbacks) {\n      callback.setParams(params);\n    }\n  }\n\n  setModel(model: Container): void {\n    for (const callback of this.callbacks) {\n      callback.setModel(model);\n    }\n  }\n\n  /**\n   * Called at the start of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochBegin(epoch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochEnd(epoch, logs);\n    }\n  }\n\n  /**\n   * Called  right before processing a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchBegin(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchEnd(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the beginning of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainBegin(logs);\n    }\n  }\n\n  /**\n   * Called at the end of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainEnd(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainEnd(logs);\n    }\n  }\n}\n\n/**\n * Callback that accumulates epoch averages of metrics.\n *\n * This callback is automatically applied to every LayersModel.\n */\nexport class BaseLogger extends BaseCallback {\n  private seen: number;\n  private totals: UnresolvedLogs;\n\n  constructor() {\n    super();\n  }\n\n  async onEpochBegin(epoch: number) {\n    this.seen = 0;\n    this.totals = {};\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const batchSize = logs['size'] == null ? 0 : logs['size'] as number;\n    this.seen += batchSize;\n    for (const key in logs) {\n      const value = logs[key];\n      if (typeof value === 'number') {\n        if (!this.totals.hasOwnProperty(key)) {\n          this.totals[key] = 0;\n        }\n        this.totals[key] = this.totals[key] as number + value * batchSize;\n      } else {\n        let oldTotalsToDispose: Scalar;\n        if (key in this.totals) {\n          oldTotalsToDispose = this.totals[key] as Scalar;\n        } else {\n          this.totals[key] = 0;\n        }\n        const total: Scalar =\n            tidy(() => add((this.totals[key]), mul(value, batchSize)));\n        this.totals[key] = total;\n        if (oldTotalsToDispose != null) {\n          oldTotalsToDispose.dispose();\n        }\n      }\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs != null) {\n      for (const key of this.params['metrics'] as string[]) {\n        if (this.totals[key] == null) {\n          continue;\n        }\n        if (typeof this.totals[key] === 'number') {\n          logs[key] = this.totals[key] as number / this.seen;\n        } else {\n          tidy(() => {\n            const log: Scalar = mul(div(1, this.seen), this.totals[key]);\n            logs[key] = log;\n            (this.totals[key] as Tensor).dispose();\n            keep(logs[key] as Scalar);\n          });\n        }\n      }\n    }\n  }\n}\n\n/**\n * Callback that records events into a `History` object. This callback is\n * automatically applied to every TF.js Layers model. The `History` object\n * gets returned by the `fit` method of models.\n */\nexport class History extends BaseCallback {\n  epoch: number[];\n  history: {[key: string]: Array<number|Tensor>};\n\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    this.epoch = [];\n    this.history = {};\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    this.epoch.push(epoch);\n    for (const key in logs) {\n      if (this.history[key] == null) {\n        this.history[key] = [];\n      }\n      this.history[key].push(logs[key]);\n    }\n  }\n\n  /**\n   * Await the values of all losses and metrics.\n   */\n  async syncData() {\n    const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n    const keys: string[] = [];\n    const indices: number[] = [];\n    for (const key in this.history) {\n      const valueArray = this.history[key];\n      for (let i = 0; i < valueArray.length; ++i) {\n        if (typeof valueArray[i] !== 'number') {\n          const valueScalar = valueArray[i] as Tensor;\n          promises.push(valueScalar.data());\n          keys.push(key);\n          indices.push(i);\n        }\n      }\n    }\n    const values = await Promise.all(promises);\n    for (let n = 0; n < values.length; ++n) {\n      const tensorToDispose = this.history[keys[n]][indices[n]] as Tensor;\n      tensorToDispose.dispose();\n      this.history[keys[n]][indices[n]] = values[n][0];\n    }\n  }\n}\n\nexport interface CustomCallbackArgs {\n  onTrainBegin?: (logs?: Logs) => void | Promise<void>;\n  onTrainEnd?: (logs?: Logs) => void | Promise<void>;\n  onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n}\n\n/**\n * Custom callback for training.\n */\nexport class CustomCallback extends BaseCallback {\n  protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;\n  protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;\n  protected readonly epochBegin:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly epochEnd:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchBegin:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchEnd:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly yield:\n      (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n\n  private yieldEvery: YieldEveryOptions;\n  private currentEpoch = 0;\n\n  constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions) {\n    super();\n    this.yieldEvery = yieldEvery || 'auto';\n    if (this.yieldEvery === 'auto') {\n      this.yieldEvery = DEFAULT_YIELD_EVERY_MS;\n    }\n    if (this.yieldEvery === 'never' && args.onYield != null) {\n      throw new Error(\n          'yieldEvery is `never` but you provided an `onYield` callback. ' +\n          'Either change `yieldEvery` or remove the callback');\n    }\n    if (util.isNumber(this.yieldEvery)) {\n      // Decorate `maybeWait` so it will be called at most once every\n      // `yieldEvery` ms.\n      this.maybeWait = generic_utils.debounce(\n          this.maybeWait.bind(this), this.yieldEvery as number);\n    }\n    this.trainBegin = args.onTrainBegin;\n    this.trainEnd = args.onTrainEnd;\n    this.epochBegin = args.onEpochBegin;\n    this.epochEnd = args.onEpochEnd;\n    this.batchBegin = args.onBatchBegin;\n    this.batchEnd = args.onBatchEnd;\n    this.yield = args.onYield;\n  }\n\n  async maybeWait(epoch: number, batch: number, logs: UnresolvedLogs) {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.yield != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.yield(epoch, batch, logs as Logs));\n    }\n    ps.push(nextFrame());\n    await Promise.all(ps);\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    this.currentEpoch = epoch;\n    if (this.epochBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.epochBegin(epoch, logs as Logs);\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.epochEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.epochEnd(epoch, logs as Logs));\n    }\n    if (this.yieldEvery === 'epoch') {\n      ps.push(nextFrame());\n    }\n    await Promise.all(ps);\n  }\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    if (this.batchBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.batchBegin(batch, logs as Logs);\n    }\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.batchEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.batchEnd(batch, logs as Logs));\n    }\n    if (this.yieldEvery === 'batch') {\n      ps.push(nextFrame());\n    } else if (util.isNumber(this.yieldEvery)) {\n      ps.push(this.maybeWait(this.currentEpoch, batch, logs));\n    }\n    await Promise.all(ps);\n  }\n\n  async onTrainBegin(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainBegin(logs as Logs);\n    }\n  }\n\n  async onTrainEnd(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainEnd != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainEnd(logs as Logs);\n    }\n  }\n}\n\n/**\n * Standardize callbacks or configurations of them to an Array of callbacks.\n */\nexport function standardizeCallbacks(\n    callbacks: BaseCallback|BaseCallback[]|CustomCallbackArgs|\n    CustomCallbackArgs[],\n    yieldEvery: YieldEveryOptions): BaseCallback[] {\n  if (callbacks == null) {\n    callbacks = {} as BaseCallback;\n  }\n  if (callbacks instanceof BaseCallback) {\n    return [callbacks];\n  }\n  if (Array.isArray(callbacks) && callbacks[0] instanceof BaseCallback) {\n    return callbacks as BaseCallback[];\n  }\n  // Convert custom callback configs to custom callback objects.\n  const callbackConfigs =\n      generic_utils.toList(callbacks) as CustomCallbackArgs[];\n  return callbackConfigs.map(\n      callbackConfig => new CustomCallback(callbackConfig, yieldEvery));\n}\n\nexport declare type BaseCallbackConstructor = {\n  new (): BaseCallback\n};\n\n/**\n * A global registry for callback constructors to be used during\n * LayersModel.fit().\n */\nexport class CallbackConstructorRegistry {\n  private static constructors:\n      {[verbosityLevel: number]: BaseCallbackConstructor[]} = {};\n\n  /**\n   * Blocks public access to constructor.\n   */\n  private constructor() {}\n\n  /**\n   * Register a tf.LayersModel.fit() callback constructor.\n   *\n   * The registered callback constructor will be used to instantiate\n   * callbacks for every tf.LayersModel.fit() call afterwards.\n   *\n   * @param verbosityLevel Level of verbosity at which the `callbackConstructor`\n   *   is to be reigstered.\n   * @param callbackConstructor A no-arg constructor for `tf.Callback`.\n   * @throws Error, if the same callbackConstructor has been registered before,\n   *   either at the same or a different `verbosityLevel`.\n   */\n  static registerCallbackConstructor(\n      verbosityLevel: number, callbackConstructor: BaseCallbackConstructor) {\n    util.assert(\n        verbosityLevel >= 0 && Number.isInteger(verbosityLevel),\n        () => `Verbosity level is expected to be an integer >= 0, ` +\n            `but got ${verbosityLevel}`);\n    CallbackConstructorRegistry.checkForDuplicate(callbackConstructor);\n    if (CallbackConstructorRegistry.constructors[verbosityLevel] == null) {\n      CallbackConstructorRegistry.constructors[verbosityLevel] = [];\n    }\n    CallbackConstructorRegistry.constructors[verbosityLevel].push(\n        callbackConstructor);\n  }\n\n  private static checkForDuplicate(callbackConstructor:\n                                       BaseCallbackConstructor) {\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const constructors = CallbackConstructorRegistry.constructors[+levelName];\n      constructors.forEach(ctor => {\n        if (ctor === callbackConstructor) {\n          throw new ValueError('Duplicate callback constructor.');\n        }\n      });\n    }\n  }\n\n  /**\n   * Clear all registered callback constructors.\n   */\n  protected static clear() {\n    CallbackConstructorRegistry.constructors = {};\n  }\n\n  /**\n   * Create callbacks using the registered callback constructors.\n   *\n   * Given `verbosityLevel`, all constructors registered at that level or above\n   * will be called and the instantiated callbacks will be used.\n   *\n   * @param verbosityLevel: Level of verbosity.\n   */\n  static createCallbacks(verbosityLevel: number): BaseCallback[] {\n    const constructors: BaseCallbackConstructor[] = [];\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const level = +levelName;\n      if (verbosityLevel >= level) {\n        constructors.push(...CallbackConstructorRegistry.constructors[level]);\n      }\n    }\n    return constructors.map(ctor => new ctor());\n  }\n}\n\nexport function configureCallbacks(\n    callbacks: BaseCallback[], verbose: ModelLoggingVerbosity, epochs: number,\n    initialEpoch: number, numTrainSamples: number, stepsPerEpoch: number,\n    batchSize: number, doValidation: boolean,\n    callbackMetrics: string[]): {callbackList: CallbackList, history: History} {\n  const history = new History();\n  const actualCallbacks: BaseCallback[] = [\n    new BaseLogger(), ...CallbackConstructorRegistry.createCallbacks(verbose)\n  ];\n  if (callbacks != null) {\n    actualCallbacks.push(...callbacks);\n  }\n  actualCallbacks.push(history);\n  const callbackList = new CallbackList(actualCallbacks);\n\n  // TODO(cais): Figure out when this LayersModel instance can have a\n  // dynamically\n  //   set property called 'callback_model' as in PyKeras.\n\n  callbackList.setParams({\n    epochs,\n    initialEpoch,\n    samples: numTrainSamples,\n    steps: stepsPerEpoch,\n    batchSize,\n    verbose,\n    doValidation,\n    metrics: callbackMetrics,\n  });\n  return {callbackList, history};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source layers/__init__.py */\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {deserializeKerasObject} from '../utils/generic_utils';\n\n/**\n * Instantiate a layer from a config dictionary.\n * @param config dict of the form {class_name: str, config: dict}\n * @param customObjects dict mapping class names (or function names)\n *   of custom (non-Keras) objects to class/functions\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns Layer instance (may be LayersModel, Sequential, Layer...)\n */\nexport function deserialize(\n    config: serialization.ConfigDict,\n    customObjects = {} as serialization.ConfigDict,\n    fastWeightInit = false): serialization.Serializable {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'layer', fastWeightInit);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: losses.py */\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\nimport * as K from './backend/tfjs_backend';\nimport {ValueError} from './errors';\nimport {LossOrMetricFn} from './types';\n\n/**\n * Normalizes a tensor wrt the L2 norm alongside the specified axis.\n * @param x\n * @param axis Axis along which to perform normalization.\n */\nexport function l2Normalize(x: Tensor, axis?: number): Tensor {\n  return tidy(() => {\n    if (x.dtype !== 'float32') {\n      x = x.asType('float32');\n    }\n    const squareSum = tfc.sum(K.square(x), axis, true);\n    const epsilonTensor = tfc.fill(squareSum.shape, epsilon());\n    const norm = tfc.sqrt(tfc.maximum(squareSum, epsilonTensor));\n    return tfc.div(x, norm);\n  });\n}\n\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(K.square(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(tfc.abs(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const diff = tfc.sub(yTrue, yPred);\n    const clippedTrue =\n        tfc.clipByValue(tfc.abs(yTrue), epsilon(), Number.MAX_VALUE);\n    const absResult = tfc.abs(tfc.div(diff, clippedTrue));\n    return tfc.mul(100, tfc.mean(absResult, -1));\n  });\n}\n\nexport function meanSquaredLogarithmicError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), Number.MAX_VALUE);\n    const firstLog = tfc.log(tfc.add(1, clippedPred));\n\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), Number.MAX_VALUE);\n    const secondLog = tfc.log(tfc.add(1, clippedTrue));\n\n    return tfc.mean(K.square(tfc.sub(firstLog, secondLog)), -1);\n  });\n}\n\nexport function squaredHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(K.square(maxResult), -1);\n  });\n}\n\nexport function hinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(maxResult, -1);\n  });\n}\n\nexport function categoricalHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const pos = tfc.sum(tfc.mul(yTrue, yPred), -1);\n    const neg = tfc.max(tfc.mul(tfc.sub(1, yTrue), yPred), -1);\n    return tfc.maximum(0, tfc.add(1, tfc.sub(neg, pos)));\n  });\n}\n\n/**\n * Logarithm of the hyperbolic cosine of the prediction error.\n *\n * `log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and\n * to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly\n * like the mean squared error, but will not be so strongly affected by the\n * occasional wildly incorrect prediction.\n */\nexport function logcosh(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const log2 = Math.log(2);\n    const predictionDiff = tfc.sub(yPred, yTrue);\n    const logcoshResult = tfc.sub(\n        tfc.add(predictionDiff, tfc.softplus(tfc.mul(-2, predictionDiff))),\n        log2);\n    return tfc.mean(logcoshResult, -1);\n  });\n}\n\nexport function categoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    if (fromLogits) {\n      output = tfc.softmax(output);\n    } else {\n      // scale preds so that the class probabilities of each sample sum to 1.\n      const outputSum = tfc.sum(output, output.shape.length - 1, true);\n      output = tfc.div(output, outputSum);\n    }\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    return tfc.neg(tfc.sum(\n        tfc.mul(target.toFloat(), tfc.log(output)), output.shape.length - 1));\n  });\n}\n\n/**\n * Categorical crossentropy with integer targets.\n *\n * @param target An integer tensor.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n */\nexport function sparseCategoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    const flatTarget = tfc.floor(K.flatten(target)).toInt() as Tensor1D;\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    const outputShape = output.shape;\n    const oneHotTarget =\n        tfc.oneHot(flatTarget, outputShape[outputShape.length - 1])\n            .reshape(outputShape);\n    return categoricalCrossentropy(oneHotTarget, output, fromLogits);\n  });\n}\n\n/**\n * From TensorFlow's implementation in nn_impl.py:\n *\n * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n *      z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n *    = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n *    = (1 - z) * x + log(1 + exp(-x))\n *    = x - x * z + log(1 + exp(-x))\n * For x < 0, to avoid overflow in exp(-x), we reformulate the above\n *      x - x * z + log(1 + exp(-x))\n *    = log(exp(x)) - x * z + log(1 + exp(-x))\n *    = - x * z + log(1 + exp(x))\n * Hence, to ensure stability and avoid overflow, the implementation uses this\n * equivalent formulation\n *    max(x, 0) - x * z + log(1 + exp(-abs(x)))\n *\n * @param labels The labels.\n * @param logits The logits.\n */\nexport function sigmoidCrossEntropyWithLogits(\n    labels: Tensor, logits: Tensor): Tensor {\n  if (!util.arraysEqual(labels.shape, logits.shape)) {\n    throw new ValueError(\n        `logits and labels must have the same shape, but got shapes ` +\n        `${JSON.stringify(labels.shape)} and ${JSON.stringify(logits.shape)}`);\n  }\n  return tidy(() => {\n    // The logistic loss formula from above is\n    //   x - x * z + log(1 + exp(-x))\n    // For x < 0, a more numerically stable formula is\n    //   -x * z + log(1 + exp(x))\n    // Note that these two expressions can be combined into the following:\n    //   max(x, 0) - x * z + log(1 + exp(-abs(x)))\n    const reluLogits = logits.relu();\n    const negAbsLogits = logits.abs().neg();\n    return reluLogits.sub(logits.mul(labels)).add(negAbsLogits.exp().log1p());\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    let y: Tensor;\n    y = tfc.clipByValue(yPred, epsilon(), 1 - epsilon());\n    y = tfc.log(tfc.div(y, tfc.sub(1, y)));\n    return tfc.mean(sigmoidCrossEntropyWithLogits(yTrue, y), -1);\n  });\n}\n\nexport function kullbackLeiblerDivergence(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), 1);\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), 1);\n    return tfc.sum(\n        tfc.mul(yTrue, tfc.log(tfc.div(clippedTrue, clippedPred))), -1);\n  });\n}\n\nexport function poisson(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const logPred = tfc.log(tfc.add(epsilon(), yPred));\n    return tfc.mean(tfc.sub(yPred, tfc.mul(yTrue, logPred)), -1);\n  });\n}\n\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const trueNormalized = l2Normalize(yTrue, -1);\n    const predNormalized = l2Normalize(yPred, -1);\n    const trueXPred = tfc.mul(trueNormalized, predNormalized);\n    return tfc.neg(tfc.sum(trueXPred, -1));\n  });\n}\n\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const msle = meanSquaredLogarithmicError;\nexport const MSLE = meanSquaredLogarithmicError;\nexport const kld = kullbackLeiblerDivergence;\nexport const KLD = kullbackLeiblerDivergence;\nexport const cosine = cosineProximity;\n\n// TODO(michaelterry): Add deserialize() function.\n\nexport const lossesMap: {[functionName: string]: LossOrMetricFn} = {\n  meanSquaredError,\n  meanAbsoluteError,\n  meanAbsolutePercentageError,\n  meanSquaredLogarithmicError,\n  squaredHinge,\n  hinge,\n  categoricalHinge,\n  logcosh,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  binaryCrossentropy,\n  kullbackLeiblerDivergence,\n  poisson,\n  cosineProximity\n};\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function get(identifierOrFn: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifierOrFn === 'string') {\n    if (identifierOrFn in lossesMap) {\n      return lossesMap[identifierOrFn];\n    }\n    let errMsg = `Unknown loss ${identifierOrFn}`;\n    if (identifierOrFn.toLowerCase().includes('softmaxcrossentropy')) {\n      errMsg = `Unknown loss ${identifierOrFn}. ` +\n          'Use \"categoricalCrossentropy\" as the string name for ' +\n          'tf.losses.softmaxCrossEntropy';\n    }\n    throw new ValueError(errMsg);\n  } else {\n    return identifierOrFn;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Built-in metrics.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {NotImplementedError, ValueError} from './errors';\nimport {categoricalCrossentropy as categoricalCrossentropyLoss, cosineProximity, meanAbsoluteError, meanAbsolutePercentageError, meanSquaredError, sparseCategoricalCrossentropy as sparseCategoricalCrossentropyLoss} from './losses';\nimport {binaryCrossentropy as lossBinaryCrossentropy} from './losses';\nimport {lossesMap} from './losses';\nimport {LossOrMetricFn} from './types';\nimport * as util from './utils/generic_utils';\n\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const threshold = tfc.mul(.5, tfc.onesLike(yPred));\n    const yPredThresholded = K.cast(tfc.greater(yPred, threshold), yTrue.dtype);\n    return tfc.mean(tfc.equal(yTrue, yPredThresholded), -1);\n  });\n}\n\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(\n      () => K.cast(\n          tfc.equal(tfc.argMax(yTrue, -1), tfc.argMax(yPred, -1)), 'float32'));\n}\n\nfunction truePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nfunction falseNegatives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(0)).sum().cast('float32');\n  });\n}\n\nfunction falsePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(0), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fp = falsePositives(yTrue, yPred);\n\n    const denominator = tp.add(fp);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fn = falseNegatives(yTrue, yPred);\n\n    const denominator = tp.add(fn);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return lossBinaryCrossentropy(yTrue, yPred);\n}\n\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  if (yTrue.rank === yPred.rank) {\n    yTrue = yTrue.squeeze([yTrue.rank - 1]);\n  }\n  yPred = yPred.argMax(-1);\n  if (yPred.dtype !== yTrue.dtype) {\n    yPred = yPred.asType(yTrue.dtype);\n  }\n  return tfc.equal(yTrue, yPred).asType('float32');\n}\n\nexport function topKCategoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\nexport function sparseTopKCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\n// Aliases.\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const categoricalCrossentropy = categoricalCrossentropyLoss;\nexport const cosine = cosineProximity;\nexport const sparseCategoricalCrossentropy = sparseCategoricalCrossentropyLoss;\n\n// TODO(cais, nielsene): Add serialize().\n\nexport const metricsMap: {[functionName: string]: LossOrMetricFn} = {\n  binaryAccuracy,\n  categoricalAccuracy,\n  precision,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  mse,\n  MSE,\n  mae,\n  MAE,\n  mape,\n  MAPE,\n  cosine\n};\n\nexport function get(identifier: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifier === 'string' && identifier in metricsMap) {\n    return metricsMap[identifier];\n  } else if (typeof identifier !== 'string' && identifier != null) {\n    return identifier;\n  } else {\n    throw new ValueError(`Unknown metric ${identifier}`);\n  }\n}\n\n/**\n * Get the shortcut function name.\n *\n * If the fn name is a string,\n *   directly return the string name.\n * If the function is included in metricsMap or lossesMap,\n *   return key of the map.\n *   - If the function relative to multiple keys,\n *     return the first found key as the function name.\n *   - If the function exists in both lossesMap and metricsMap,\n *     search lossesMap first.\n * If the function is not included in metricsMap or lossesMap,\n *   return the function name.\n *\n * @param fn loss function, metric function, or short cut name.\n * @returns Loss or Metric name in string.\n */\nexport function getLossOrMetricName(fn: string|LossOrMetricFn): string {\n  util.assert(fn !== null, `Unknown LossOrMetricFn ${fn}`);\n  if (typeof fn === 'string') {\n    return fn;\n  } else {\n    let fnName;\n    for (const key of Object.keys(lossesMap)) {\n      if (lossesMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    for (const key of Object.keys(metricsMap)) {\n      if (metricsMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    return (fn as Function).name;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Optimizers.\n */\n\nimport {Optimizer, train} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\n\nimport {ValueError} from './errors';\n\n// Add (de)serialize()\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function getOptimizer(identifier: string): Optimizer {\n  const optimizerMap: {[optimizerName: string]: () => Optimizer} = {\n    'Adagrad': () => train.adagrad(0.01),\n    'Adadelta': () => train.adadelta(1, 0.95, epsilon()),\n    'Adam': () => train.adam(0.001, 0.9, 0.999, epsilon()),\n    'Adamax': () => train.adamax(0.002, 0.9, 0.999, epsilon(), 0),\n    'RMSProp': () => train.rmsprop(0.001, 0.9, 0, epsilon()),\n    'SGD': () => train.sgd(0.01)\n  };\n  optimizerMap['adagrad'] = optimizerMap['Adagrad'];\n  optimizerMap['adadelta'] = optimizerMap['Adadelta'];\n  optimizerMap['adam'] = optimizerMap['Adam'];\n  optimizerMap['adamax'] = optimizerMap['Adamax'];\n  optimizerMap['rmsprop'] = optimizerMap['RMSProp'];\n  optimizerMap['sgd'] = optimizerMap['SGD'];\n\n  if (identifier in optimizerMap) {\n    return optimizerMap[identifier]();\n  }\n  throw new ValueError(`Unknown Optimizer ${identifier}`);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/** Utility functions related to user-defined metadata. */\n\n// Maximum recommended serialized size for user-defined metadata.\n// Beyond this limit, a warning message will be printed during model loading and\n// saving.\nexport const MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH = 1 * 1024 * 1024;\n\n/**\n * Check validity of user-defined metadata.\n *\n * @param userDefinedMetadata\n * @param modelName Name of the model that the user-defined metadata belongs to.\n *   Used during construction of error messages.\n * @param checkSize Whether to check the size of the metadata is under\n *   recommended limit. Default: `false`. If `true`, will try stringify the\n *   JSON object and print a console warning if the serialzied size is above the\n *   limit.\n * @throws Error if `userDefinedMetadata` is not a plain JSON object.\n */\nexport function checkUserDefinedMetadata(\n    userDefinedMetadata: {}, modelName: string, checkSize = false): void {\n  if (userDefinedMetadata == null ||\n      typeof userDefinedMetadata !== 'object' ||\n      Object.getPrototypeOf(userDefinedMetadata) !== Object.prototype ||\n      !plainObjectCheck(userDefinedMetadata)) {\n    throw new Error(\n        'User-defined metadata is expected to be a JSON object, but is not.');\n  }\n\n  if (checkSize) {\n    const out = JSON.stringify(userDefinedMetadata);\n    if (out.length > MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH) {\n      console.warn(\n          `User-defined metadata of model \"${modelName}\" is too large in ` +\n          `size (length=${out.length} when serialized). It is not ` +\n          `recommended to store such large objects in user-defined metadata. ` +\n          `Please make sure its serialized length is <= ` +\n          `${MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH}.`);\n    }\n  }\n}\n\n/**\n * Check if an input is plain JSON object or any valid subfield of it.\n *\n * @param x The input to be checked.\n * @param assertObject Whether to assert `x` is a JSON object, i.e., reject\n *   cases of arrays and primitives.\n * @return Returns `true` if and only if `x` is a plain JSON object,\n *   a JSON-valid primitive including string, number, boolean and null,\n *   or an array of the said types.\n */\n// tslint:disable-next-line:no-any\nexport function plainObjectCheck(x: any): boolean {\n  if (x === null) {\n    // Note: typeof `null` is 'object', and `null` is valid in JSON.\n    return true;\n  } else if (typeof x === 'object') {\n    if (Object.getPrototypeOf(x) === Object.prototype) {\n      // `x` is a JavaScript object and its prototype is Object.\n      const keys = Object.keys(x);\n      for (const key of keys) {\n        if (typeof key !== 'string') {\n          // JSON keys must be strings.\n          return false;\n        }\n        if (!plainObjectCheck(x[key])) {  // Recursive call.\n          return false;\n        }\n      }\n      return true;\n    } else {\n      // `x` is a JavaScript object but its prototype is not Object.\n      if (Array.isArray(x)) {\n        // `x` is a JavaScript array.\n        for (const item of x) {\n          if (!plainObjectCheck(item)) {  // Recursive call.\n            return false;\n          }\n        }\n        return true;\n      } else {\n        // `x` is a JavaScript object and its prototype is not Object,\n        // and it's not an Array. I.e., it's a complex object such as\n        // `Error` and `Date`.\n        return false;\n      }\n    }\n  } else {\n    // `x` is not a JavaScript object or `null`.\n    const xType = typeof x;\n    return xType === 'string' || xType === 'number' || xType === 'boolean';\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {Container} from '../engine/container';\nimport {Layer, Node} from '../engine/topology';\nimport {countParamsInWeights} from './variable_utils';\n\n/**\n * Print the summary of a LayersModel object.\n *\n * @param model tf.LayersModel instance.\n * @param lineLength Total length of printed lines. Set this to adapt to the\n *   display to different terminal or console sizes.\n * @param positions Relative or absolute positions of log elements in each\n *   line. Each number corresponds to right-most (i.e., ending) position of a\n *   column.\n *   If not provided, defaults to `[0.45, 0.85, 1]` for sequential-like\n *   models and `[0.33, 0.55, 0.67, 1]` for non-sequential like models.\n * @param printFn Print function to use.\n *   It will be called on each line of the summary. You can provide a custom\n *   function in order to capture the string summary. Defaults to `console.log`.\n */\nexport function printSummary(\n    model: Container, lineLength?: number, positions?: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void =\n        console.log): void {\n  const sequentialLike = isModelSequentialLike(model);\n\n  // Header names for different log elements.\n  const toDisplay: string[] = ['Layer (type)', 'Output shape', 'Param #'];\n  if (sequentialLike) {\n    lineLength = lineLength || 65;\n    positions = positions || [0.45, 0.85, 1];\n  } else {\n    lineLength = lineLength || 98;\n    positions = positions || [0.33, 0.55, 0.67, 1];\n    // Header names for different log elements.\n  }\n\n  if (positions[positions.length - 1] <= 1) {\n    // `positions` is relative. Convert it to absolute positioning.\n    positions = positions.map(p => Math.floor(lineLength * p));\n  }\n\n  let relevantNodes: Node[];\n  if (!sequentialLike) {\n    toDisplay.push('Receives inputs');\n    relevantNodes = [];\n    for (const depth in model.nodesByDepth) {\n      relevantNodes.push(...model.nodesByDepth[depth]);\n    }\n  }\n\n  printFn('_'.repeat(lineLength));\n  printRow(toDisplay, positions, printFn);\n  printFn('='.repeat(lineLength));\n\n  const layers = model.layers;\n  for (let i = 0; i < layers.length; ++i) {\n    if (sequentialLike) {\n      printLayerSummary(layers[i], positions, printFn);\n    } else {\n      printLayerSummaryWithConnections(\n          layers[i], positions, relevantNodes, printFn);\n    }\n    printFn((i === layers.length - 1 ? '=' : '_').repeat(lineLength));\n  }\n\n  // tslint:disable-next-line:no-any\n  (model as any).checkTrainableWeightsConsistency();\n\n  const trainableCount = countTrainableParams(model);\n  const nonTrainableCount = countParamsInWeights(model.nonTrainableWeights);\n\n  printFn(`Total params: ${trainableCount + nonTrainableCount}`);\n  printFn(`Trainable params: ${trainableCount}`);\n  printFn(`Non-trainable params: ${nonTrainableCount}`);\n  printFn('_'.repeat(lineLength));\n}\n\nfunction countTrainableParams(model: Container): number {\n  let trainableCount: number;\n  // tslint:disable:no-any\n  if ((model as any).collectedTrainableWeights != null) {\n    trainableCount =\n        countParamsInWeights((model as any).collectedTrainableWeights);\n  } else {\n    trainableCount = countParamsInWeights(model.trainableWeights);\n  }\n  // tslint:enable:no-any\n  return trainableCount;\n}\n\nfunction isModelSequentialLike(model: Container): boolean {\n  let sequentialLike = true;\n  const nodesByDepth: Node[][] = [];\n  const nodes: Node[] = [];\n  for (const depth in model.nodesByDepth) {\n    nodesByDepth.push(model.nodesByDepth[depth]);\n  }\n  for (const depthNodes of nodesByDepth) {\n    if (depthNodes.length > 1 ||\n        depthNodes.length === 1 && depthNodes[0].inboundLayers.length > 1) {\n      sequentialLike = false;\n      break;\n    }\n    nodes.push(...depthNodes);\n  }\n  if (sequentialLike) {\n    // Search for shared layers.\n    for (const layer of model.layers) {\n      let flag = false;\n      for (const node of layer.inboundNodes) {\n        if (nodes.indexOf(node) !== -1) {\n          if (flag) {\n            sequentialLike = false;\n            break;\n          } else {\n            flag = true;\n          }\n        }\n      }\n      if (!sequentialLike) {\n        break;\n      }\n    }\n  }\n  return sequentialLike;\n}\n\nfunction printRow(\n    fields: string[], positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void = console.log) {\n  let line = '';\n  for (let i = 0; i < fields.length; ++i) {\n    if (i > 0) {\n      line = line.slice(0, line.length - 1) + ' ';\n    }\n    line += fields[i];\n    line = line.slice(0, positions[i]);\n    line += ' '.repeat(positions[i] - line.length);\n  }\n  printFn(line);\n}\n\n/**\n * Prints a summary for a single Layer, without connectivity information.\n *\n * @param layer: Layer instance to print.\n */\nfunction printLayerSummary(\n    layer: Layer, positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const name = layer.name;\n  const className = layer.getClassName();\n  const fields: string[] =\n      [`${name} (${className})`, outputShape, layer.countParams().toString()];\n  printRow(fields, positions, printFn);\n}\n\n/**\n * Prints a summary for a single Layer, with connectivity information.\n */\nfunction printLayerSummaryWithConnections(\n    layer: Layer, positions: number[], relevantNodes: Node[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const connections: string[] = [];\n  for (const node of layer.inboundNodes) {\n    if (relevantNodes != null && relevantNodes.length > 0 &&\n        relevantNodes.indexOf(node) === -1) {\n      continue;\n    }\n    for (let i = 0; i < node.inboundLayers.length; ++i) {\n      const inboundLayer = node.inboundLayers[i].name;\n      const inboundLayerIndex = node.nodeIndices[i];\n      const inboundTensorIndex = node.tensorIndices[i];\n      connections.push(\n          `${inboundLayer}[${inboundLayerIndex}][${inboundTensorIndex}]`);\n    }\n  }\n  const name = layer.name;\n  const className = layer.getClassName();\n  const firstConnection = connections.length === 0 ? '' : connections[0];\n  const fields: string[] = [\n    `${name} (${className})`, outputShape, layer.countParams().toString(),\n    firstConnection\n  ];\n\n  printRow(fields, positions, printFn);\n  for (let i = 1; i < connections.length; ++i) {\n    printRow(['', '', '', connections[i]], positions, printFn);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Porting note: This file doesn't exist in PyKeras.\n// Its purpose here is to centralize the boundary layer between\n// tfjs-layers's internal Config TS-Centric format and PyKeras's\n// serialized Python Config format.\n\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {PyJsonValue} from '../keras_format/types';\nimport * as generic_utils from '../utils/generic_utils';\n// tslint:enable\n\n/**\n * Test whether a value in an array is the name of a LayersModel or Layer.\n * @param key The key name that the value is found under. Note that the key\n *   may not be at the level immediately above the value, if the value is in a\n *   nested array.\n * @param index Index of the value in the Array that it is found in.\n * @param value The value object.\n * @returns A boolean indicating whether value is a name.\n */\nfunction isArrayItemInputOrOutputName<T>(\n    key: string, index: number, value: T): boolean {\n  return (key === 'inboundNodes' || key === 'outputLayers' ||\n          key === 'inputLayers') &&\n      index === 0 && typeof value === 'string';\n}\n\n/**\n * Convert a Pythonic config object to TypeScript config object.\n * @param pythonicConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertPythonicToTs(\n    pythonicConfig: PyJsonValue, key?: string): serialization.ConfigDictValue {\n  if (pythonicConfig === null) {\n    return null;\n  } else if (typeof pythonicConfig === 'string') {\n    return generic_utils.toCamelCase(pythonicConfig);\n  } else if (\n      (typeof pythonicConfig === 'number') ||\n      (typeof pythonicConfig === 'boolean')) {\n    return pythonicConfig;\n  } else if (pythonicConfig instanceof Array) {\n    const tsArray = [];\n    const arrayLength = pythonicConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = pythonicConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        tsArray.push(item);\n      } else {\n        tsArray.push(convertPythonicToTs(item, key));\n      }\n    }\n    return tsArray;\n  } else {\n    const tsDict: serialization.ConfigDict = {};\n    for (const pythonicKey of Object.keys(pythonicConfig)) {\n      const pythonicValue = pythonicConfig[pythonicKey];\n      if (pythonicKey === 'name' && typeof pythonicValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // camel-case conversion.\n        tsDict[pythonicKey] = pythonicValue;\n      } else {\n        const tsKey = generic_utils.toCamelCase(pythonicKey);\n        tsDict[tsKey] = convertPythonicToTs(pythonicValue, tsKey);\n      }\n    }\n    return tsDict;\n  }\n}\n\n/**\n * Convert a TypeScript config object to Python config object.\n * @param tsConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertTsToPythonic(\n    tsConfig: serialization.ConfigDictValue, key?: string): PyJsonValue {\n  if (tsConfig === null || tsConfig === undefined) {\n    return null;\n  } else if (typeof tsConfig === 'string') {\n    return generic_utils.toSnakeCase(tsConfig);\n  } else if (\n      (typeof tsConfig === 'number') || (typeof tsConfig === 'boolean')) {\n    return tsConfig;\n  } else if (tsConfig instanceof Array) {\n    const pyArray = [];\n    const arrayLength = tsConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = tsConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        pyArray.push(item);\n      } else {\n        pyArray.push(convertTsToPythonic(item, key));\n      }\n    }\n    return pyArray;\n  } else {\n    const pyDict: serialization.ConfigDict = {};\n    for (const tsKey of Object.keys(tsConfig)) {\n      const tsValue = tsConfig[tsKey];\n      const pyKey = generic_utils.toSnakeCase(tsKey);\n      if ((tsKey === 'name' || tsKey === 'className') &&\n          typeof tsValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // snake-case conversion.\n        pyDict[pyKey] = tsValue;\n      } else {\n        pyDict[pyKey] = convertTsToPythonic(tsValue, tsKey);\n      }\n    }\n    return pyDict;\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Executor: Evaluates SymbolicTensor based on feeds.\n */\n\nimport {cast, dispose, memory, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {ValueError} from '../errors';\nimport {Kwargs} from '../types';\nimport {toList} from '../utils/generic_utils';\n\nimport {InputLayer} from './input_layer';\nimport {SymbolicTensor} from './topology';\n\n/**\n * Helper function to check the dtype and shape compatibility of a feed value.\n */\nfunction assertFeedCompatibility(key: SymbolicTensor, val: Tensor): Tensor {\n  // Check dtype compatibility.\n  if (key.dtype == null || key.dtype === val.dtype) {\n    //  a.  If types match, return val tensor as is.\n    return val;\n  }\n  try {\n    //  b. Attempt to convert to expected type.\n    return cast(val, key.dtype);\n  } catch (err) {\n    //  c. If conversion fails, return helpful error.\n    throw new ValueError(\n        `The dtype of the feed (${val.dtype}) can not be cast to the dtype ` +\n        `of the key '${key.name}' (${key.dtype}).`);\n  }\n}\n\n/**\n * A concrete Tensor value for a symbolic tensor as the key.\n */\nexport interface Feed {\n  key: SymbolicTensor;\n  value: Tensor;\n}\n\n/**\n * FeedDict: A mapping from unique SymbolicTensors to feed values for them.\n * A feed value is a concrete value represented as an `Tensor`.\n */\nexport class FeedDict {\n  private id2Value: {[id: number]: Tensor} = {};\n  private id2Mask: {[id: number]: Tensor} = {};\n  private name2Id: {[name: string]: number} = {};\n\n  /**\n   * Constructor, optionally does copy-construction.\n   * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case\n   *   copy-construction will be performed.\n   */\n  constructor(feeds?: Feed[]|FeedDict) {\n    if (feeds instanceof FeedDict) {\n      for (const id in feeds.id2Value) {\n        this.id2Value[id] = feeds.id2Value[id];\n        if (id in feeds.id2Mask) {\n          this.id2Mask[id] = feeds.id2Mask[id];\n        }\n      }\n    } else {\n      if (feeds == null) {\n        return;\n      }\n      for (const feed of feeds) {\n        this.add(feed.key, feed.value);\n      }\n    }\n  }\n\n  /**\n   * Add a key-value pair to the FeedDict.\n   *\n   * @param key The key of the feed.\n   * @param value The value of the tensor feed.\n   * @param mask The value of the mask feed (optional).\n   * @returns This `FeedDict`.\n   * @throws ValueError: If the key `SymbolicTensor` already exists in the\n   *   `FeedDict`.\n   */\n  add(key: SymbolicTensor, value: Tensor, mask?: Tensor): FeedDict {\n    if (this.id2Value[key.id] == null) {\n      this.id2Value[key.id] = assertFeedCompatibility(key, value);\n      this.name2Id[key.name] = key.id;\n      if (mask != null) {\n        this.id2Mask[key.id] = mask;\n      }\n    } else {\n      throw new ValueError(`Duplicate key: name=${key.name}, id=${key.id}`);\n    }\n    return this;\n  }\n\n  /**\n   * Add a Feed to the FeedDict.\n   * @param feed The new `Feed` to add.\n   * @returns This `FeedDict`.\n   */\n  addFeed(feed: Feed) {\n    this.add(feed.key, feed.value);\n  }\n\n  /**\n   * Probe whether a key already exists in the FeedDict.\n   * @param key\n   */\n  hasKey(key: SymbolicTensor): boolean {\n    return this.id2Value[key.id] != null;\n  }\n\n  /**\n   * Get all the SymbolicTensor available in this FeedDict.\n   */\n  names(): string[] {\n    return Object.keys(this.name2Id);\n  }\n\n  /**\n   * Get the feed value for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed value.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getValue(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Value[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Value[id];\n    }\n  }\n\n  /**\n   * Get the feed mask for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed mask.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getMask(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Mask[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Mask[id];\n    }\n  }\n\n  /** Dispose all mask Tensors held by this object. */\n  disposeMasks() {\n    if (this.id2Mask != null) {\n      dispose(this.id2Mask);\n    }\n  }\n}\n\n// Cache for topologically sorted SymbolicTensors for given execution\n// targets (i.e., fetches).\nconst cachedSorted: {[concatFetchNames: string]: SymbolicTensor[]} = {};\n\n// Cache for recipient count maps for given execution targets (i.e., fetches).\nconst cachedRecipientCounts:\n    {[concatFetchNames: string]: {[fetchName: string]: number}} = {};\n\n/**\n * Interface for the optional object used for probing the memory\n * usage and other statistics during execution.\n */\nexport interface ExecutionProbe {\n  /**\n   * Maximum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  maxNumTensors?: number;\n\n  /**\n   * Minimum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  minNumTensors?: number;\n}\n\n/**\n * Execute a SymbolicTensor by using concrete feed values.\n *\n * A `SymbolicTensor` object is a node in a computation graph of TF.js\n * Layers. The object is backed by a source layer and input\n * `SymbolicTensor`s to the source layer. This method evaluates\n * the `call()` method of the source layer, using concrete values of the\n * inputs obtained from either\n * * `feedDict`, if the input key exists in `feedDict`, or else,\n * * a recursive call to `execute()` itself.\n *\n * @param x: The `SymbolicTensor` to execute.\n * @param feedDict: The feed values, as base condition of the recursion.\n *   execution.\n * @param kwargs: Optional keyword arguments.\n * @param probe: A probe object (of interface `ExecutionProbe`) used for\n *   testing memory footprint of `execute` calls.\n * @returns Result of the execution.\n * @throws ValueError: If any `SymbolicTensor`s from `InputLayer`s\n *   encountered during the execution lacks a feed value in `feedDict`.\n */\nexport function execute(\n    fetches: SymbolicTensor|SymbolicTensor[], feedDict: FeedDict,\n    kwargs?: Kwargs, probe?: ExecutionProbe): Tensor|\n    Tensor[]|[Tensor | Tensor[]] {\n  const training: boolean = kwargs == null ? false : kwargs['training'];\n\n  const arrayFetches = Array.isArray(fetches);\n  const fetchArray: SymbolicTensor[] =\n      arrayFetches ? fetches as SymbolicTensor[] : [fetches as SymbolicTensor];\n\n  const outputNames = fetchArray.map(t => t.name);\n  const finalOutputs: Tensor[] = [];\n  const feedNames = feedDict.names();\n  for (const outputName of outputNames) {\n    if (feedNames.indexOf(outputName) !== -1) {\n      finalOutputs.push(feedDict.getValue(outputName));\n    } else {\n      finalOutputs.push(null);\n    }\n  }\n\n  if (probe != null) {\n    // For optional probing of memory footprint during execution.\n    probe.maxNumTensors = -Infinity;\n    probe.minNumTensors = Infinity;\n  }\n\n  // Check cache.\n  const fetchAndFeedKey =\n      outputNames.join(',') + '|' + feedDict.names().join(',');\n  let sorted: SymbolicTensor[];\n  let recipientCounts: {[fetchName: string]: number};\n  if (cachedSorted[fetchAndFeedKey] == null) {\n    // Cache doesn't contain the desired combination of fetches. Compute\n    // topological sort for the combination for the first time.\n    const out = getTopologicalSortAndRecipientCounts(fetchArray, feedDict);\n    sorted = out.sorted;\n    recipientCounts = out.recipientCounts;\n\n    // Store results in cache for future use.\n    cachedSorted[fetchAndFeedKey] = sorted;\n    cachedRecipientCounts[fetchAndFeedKey] = recipientCounts;\n  }\n  sorted = cachedSorted[fetchAndFeedKey];\n  recipientCounts = {};\n  if (!training) {\n    Object.assign(recipientCounts, cachedRecipientCounts[fetchAndFeedKey]);\n  }\n\n  const internalFeedDict = new FeedDict(feedDict);\n\n  // Start iterative execution on the topologically-sorted SymbolicTensors.\n  for (let i = 0; i < sorted.length; ++i) {\n    if (probe != null) {\n      // For optional probing of memory usage during execution.\n      const numTensors = memory().numTensors;\n      if (numTensors > probe.maxNumTensors) {\n        probe.maxNumTensors = numTensors;\n      }\n      if (numTensors < probe.minNumTensors) {\n        probe.minNumTensors = numTensors;\n      }\n    }\n\n    const symbolic = sorted[i];\n    const srcLayer = symbolic.sourceLayer;\n    if (srcLayer instanceof InputLayer) {\n      continue;\n    }\n    const inputValues: Tensor[] = [];\n    const inputMasks: Tensor[] = [];\n    const tensorsToDispose: Tensor[] = [];\n\n    let maskExists = false;\n    for (const input of symbolic.inputs) {\n      const value = internalFeedDict.getValue(input);\n      const mask = internalFeedDict.getMask(input);\n      inputValues.push(value);\n      inputMasks.push(mask);\n      if (mask != null) {\n        maskExists = true;\n      }\n      if (!training) {\n        recipientCounts[input.name]--;\n        if (recipientCounts[input.name] === 0 && !feedDict.hasKey(input) &&\n            outputNames.indexOf(input.name) === -1 && !value.isDisposed &&\n            input.sourceLayer.stateful !== true) {\n          tensorsToDispose.push(value);\n        }\n      }\n    }\n\n    if (maskExists) {\n      kwargs = kwargs || {};\n      kwargs['mask'] = inputMasks[0];\n    }\n    const outputTensors =\n        toList(srcLayer.apply(inputValues, kwargs)) as Tensor[];\n    let outputMask: Tensor|Tensor[] = null;\n    if (srcLayer.supportsMasking) {\n      outputMask = srcLayer.computeMask(inputValues, inputMasks);\n    }\n    const layerOutputs = getNodeOutputs(symbolic);\n    const outputSymbolicTensors =\n        Array.isArray(layerOutputs) ? layerOutputs : [layerOutputs];\n    for (let i = 0; i < outputSymbolicTensors.length; ++i) {\n      if (!internalFeedDict.hasKey(outputSymbolicTensors[i])) {\n        internalFeedDict.add(\n            outputSymbolicTensors[i], outputTensors[i],\n            Array.isArray(outputMask) ? outputMask[0] : outputMask);\n      }\n      const index = outputNames.indexOf(outputSymbolicTensors[i].name);\n      if (index !== -1) {\n        finalOutputs[index] = outputTensors[i];\n      }\n    }\n\n    if (!training) {\n      // Clean up Tensors that are no longer needed.\n      dispose(tensorsToDispose);\n    }\n  }\n  // NOTE(cais): Unlike intermediate tensors, we don't discard mask\n  // tensors as we go, because these tensors are sometimes passed over a\n  // series of mutliple layers, i.e., not obeying the immediate input\n  // relations in the graph. If this becomes a memory-usage concern,\n  // we can improve this in the future.\n  internalFeedDict.disposeMasks();\n\n  return arrayFetches ? finalOutputs : finalOutputs[0];\n}\n\ntype RecipientCounts = {\n  [fetchName: string]: number\n};\n\nexport type RecipientMap = {\n  [fetchName: string]: Set<string>;\n};\n\n/**\n * Sort the `SymbolicTensor`s topologically, for an array of fetches.\n *\n * This function calls getTopologicalSortAndRecipientCountsForOneFetch and\n * merges their results.\n *\n * @param fetch The array of fetches requested. Must be a non-empty array.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientCounts: Recipient counts for all SymbolicTensors in `sorted`.\n */\nfunction getTopologicalSortAndRecipientCounts(\n    fetches: SymbolicTensor[], feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientCounts: RecipientCounts} {\n  util.assert(\n      fetches != null && fetches.length > 0,\n      () => `Expected at least one fetch, got none`);\n\n  let finalSorted: SymbolicTensor[] = [];\n  let finalRecipientMap: RecipientMap = {};\n  if (fetches.length === 1) {\n    // Special-casing 1 fetch for efficiency.\n    const out =\n        getTopologicalSortAndRecipientCountsForOneFetch(fetches[0], feedDict);\n    finalSorted = out.sorted;\n    finalRecipientMap = out.recipientMap;\n  } else {\n    const visited = new Set<string>();\n    for (const fetch of fetches) {\n      const {sorted, recipientMap} =\n          getTopologicalSortAndRecipientCountsForOneFetch(fetch, feedDict);\n\n      // Merge sorted SymbolicTensor Arrays.\n      for (const symbolicTensor of sorted) {\n        if (!visited.has(symbolicTensor.name)) {\n          finalSorted.push(symbolicTensor);\n          visited.add(symbolicTensor.name);\n        }\n      }\n\n      // Merge recipient maps.\n      for (const name in recipientMap) {\n        if (finalRecipientMap[name] == null) {\n          finalRecipientMap[name] = new Set<string>();\n        }\n        recipientMap[name].forEach(\n            recipient => finalRecipientMap[name].add(recipient));\n      }\n    }\n  }\n  return {\n    sorted: finalSorted,\n    recipientCounts: recipientMap2Counts(finalRecipientMap)\n  };\n}\n\nfunction recipientMap2Counts(recipientMap: RecipientMap): RecipientCounts {\n  const recipientCounts: RecipientCounts = {};\n  for (const name in recipientMap) {\n    recipientCounts[name] = recipientMap[name].size;\n  }\n  return recipientCounts;\n}\n\n/**\n * Sort the `SymbolicTensor`s topologically, for a single fetch.\n *\n * This helper function processes the upstream SymbolicTensors of a single\n * fetch.\n *\n * @param fetch The single fetch requested.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientMap: Recipient names for all SymbolicTensors in `sorted`.\n */\nexport function getTopologicalSortAndRecipientCountsForOneFetch(\n    fetch: SymbolicTensor, feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientMap: RecipientMap} {\n  const visited = new Set<string>();\n  const sorted: SymbolicTensor[] = [];\n  const recipientMap: RecipientMap = {};\n\n  // Put keys of the feedDict into visited first, so they don't have to be\n  // walked. This is needed in case where there are feeds for intermediate\n  // SymbolicTensors of the graph.\n  for (const key of feedDict.names()) {\n    visited.add(key);\n  }\n\n  const stack: SymbolicTensor[] = [];\n  const marks: number[] = [];\n\n  // Initial population of stack and marks.\n  stack.push(fetch);\n\n  while (stack.length > 0) {\n    const top = stack[stack.length - 1];\n    if (visited.has(top.name)) {\n      stack.pop();\n      continue;\n    }\n    const topIsMarked = marks[marks.length - 1] === stack.length - 1;\n    if (top.inputs.length === 0 || topIsMarked) {\n      // Input SymbolicTensor or all children have been visited.\n      stack.pop();\n      sorted.push(top);\n      visited.add(top.name);\n      if (topIsMarked) {\n        marks.pop();\n      }\n    } else {\n      // A non-input SymbolicTensor whose upstream SymbolicTensors haven't\n      // been visited yet. Push them onto the stack.\n      marks.push(stack.length - 1);\n      for (const input of top.inputs) {\n        // Increment the recipient count. Note that this needs to happen\n        // regardless of whether the SymbolicTensor has been visited before.\n        if (recipientMap[input.name] == null) {\n          recipientMap[input.name] = new Set<string>();\n        }\n        recipientMap[input.name].add(top.name);\n\n        if (visited.has(input.name)) {\n          continue;  // Avoid repeated visits to the same SymbolicTensor.\n        }\n        stack.push(input);\n      }\n    }\n  }\n  return {sorted, recipientMap};\n}\n\n/**\n * Get the symbolic output tensors of the node to which a given fetch belongs.\n * @param fetch The fetched symbolic tensor.\n * @returns The Array of symbolic tensors output by the node to which `fetch`\n *   belongs.\n */\nfunction getNodeOutputs(fetch: SymbolicTensor): SymbolicTensor|\n    SymbolicTensor[] {\n  let layerOutputs: SymbolicTensor|SymbolicTensor[];\n  if (fetch.sourceLayer.inboundNodes.length === 1) {\n    layerOutputs = fetch.sourceLayer.output;\n  } else {\n    let nodeIndex: number = null;\n    for (let i = 0; i < fetch.sourceLayer.inboundNodes.length; ++i) {\n      for (const outputTensor of fetch.sourceLayer.inboundNodes[i]\n               .outputTensors) {\n        if (outputTensor.id === fetch.id) {\n          nodeIndex = i;\n          break;\n        }\n      }\n    }\n    layerOutputs = fetch.sourceLayer.getOutputAt(nodeIndex);\n  }\n  return layerOutputs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {NamedTensorMap, Scalar, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {TensorKeyWithArgsArray} from '../keras_format/node_config';\nimport {PyJsonDict} from '../keras_format/types';\nimport {deserialize as deserializeLayer} from '../layers/serialization';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {convertTsToPythonic} from '../utils/serialization_utils';\nimport * as types_utils from '../utils/types_utils';\nimport {batchSetValue, LayerVariable} from '../variables';\nimport {version as layersVersion} from '../version';\n\nimport {execute, FeedDict} from './executor';\nimport {InputLayer} from './input_layer';\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/** Constructor config for Container. */\nexport interface ContainerArgs {\n  inputs: SymbolicTensor|SymbolicTensor[];\n  outputs: SymbolicTensor|SymbolicTensor[];\n  name?: string;\n}\n\n/**\n * A Container is a directed acyclic graph of layers.\n *\n * It is the topological form of a \"model\". A LayersModel\n * is simply a Container with added training routines.\n *\n */\nexport abstract class Container extends Layer {\n  inputs: SymbolicTensor[];\n  outputs: SymbolicTensor[];\n\n  inputLayers: Layer[];\n  inputLayersNodeIndices: number[];\n  inputLayersTensorIndices: number[];\n\n  outputLayers: Layer[];\n  outputLayersNodeIndices: number[];\n  outputLayersTensorIndices: number[];\n\n  layers: Layer[];\n  layersByDepth: {[depth: string]: Layer[]};\n  nodesByDepth: {[depth: string]: Node[]};\n\n  internalContainerRefs: Container[];\n\n  containerNodes = new Set<string>();\n\n  // TODO(michaelterry): Add cache support\n  // private outputMaskCache: any;\n  // private outputTensorCache: any;\n  // private outputShapeCache: any;\n\n  inputNames: string[];\n  outputNames: string[];\n  feedInputShapes: Shape[];\n\n  protected internalInputShapes: Shape[];\n  protected internalOutputShapes: Shape[];\n  // TODO(cais): Maybe 'feed' should not in the names of these variables,\n  //   due to the fact that our backend is not symbolic.\n  protected feedInputNames: string[];\n  protected feedOutputNames: string[];\n\n  constructor(args: ContainerArgs) {\n    // No args passed to super's constructor.\n    super({});\n    this.name = args.name;\n    if (this.name == null) {\n      const prefix = this.getClassName().toLowerCase();\n      this.name = getUid(prefix);\n    }\n\n    this.supportsMasking = false;\n    this.trainable_ = true;\n\n    // TODO(michaelterry): Initialize perInputLosses/Updates here.\n\n    // Container-specific properties.\n    if (Array.isArray(args.inputs)) {\n      this.inputs = args.inputs.slice();\n    } else {\n      this.inputs = [args.inputs];\n    }\n    if (Array.isArray(args.outputs)) {\n      this.outputs = args.outputs.slice();\n    } else {\n      this.outputs = [args.outputs];\n    }\n\n    // Check for redundancy in inputs.\n    if (generic_utils.unique(this.inputs).length !== this.inputs.length) {\n      throw new ValueError(\n          'The list of inputs passed to the model is ' +\n          'redundant. All inputs should only appear once. Found: ' +\n          `${this.inputs.map(x => x.name)}`);\n    }\n\n    // Check for redundancy in outputs.\n    if (generic_utils.unique(this.outputs).length !== this.outputs.length) {\n      console.warn(\n          'The list of outputs passed to the model is redundant. ' +\n          'All outputs should only appear once. Found: ' +\n          `${this.outputs.map(x => x.name)}`);\n    }\n\n    /*\n      List of initial layers (1 to 1 mapping with this.inputs, hence the same\n      layer might appear twice)\n    */\n    this.inputLayers = [];\n    this.inputLayersNodeIndices = [];\n    this.inputLayersTensorIndices = [];\n    /*\n      List of layers (1 to 1 mapping with this.outputs, hence the same layer\n      might appear twice)\n    */\n    this.outputLayers = [];\n    this.outputLayersNodeIndices = [];\n    this.outputLayersTensorIndices = [];\n    /*\n      All layers in order of horizontal graph traversal. Entries are unique.\n      Includes input and output layers.\n    */\n    this.layers = [];\n\n    /*\n      References to container layers that were constructed internally. We need\n      these to properly dispose of tensors from nested containers.\n    */\n    this.internalContainerRefs = [];\n\n    // TODO(michaelterry): Determine if caching still needed with eager\n    // backend.\n    /*\n      This is for performance optimization when calling the Container on new\n      inputs. Every time the Container is called on a set on input tensors,\n      we compute the output tensors, output masks and output shapes in one pass,\n      then cache them here. When one of these outputs is queried later,\n      we retrieve it from there instead of recomputing it.\n    */\n    // this.outputTensorCache = {};\n    // this.outputShapeCache = {};\n\n    // Build this.outputLayers:\n    for (const x of this.outputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      this.outputLayers.push(layer);\n      this.outputLayersNodeIndices.push(nodeIndex);\n      this.outputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // TODO(michaelterry): Add output mask cache code.\n\n    // Build this.inputLayers:\n    for (const x of this.inputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      /*\n        It's supposed to be an input layer, so only one node\n        and one tensor output.\n      */\n      generic_utils.assert(nodeIndex === 0, 'input layer has >1 nodes');\n      generic_utils.assert(tensorIndex === 0, 'input layer has >1 tensors');\n      this.inputLayers.push(layer);\n      this.inputLayersNodeIndices.push(nodeIndex);\n      this.inputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // Build this.inputNames and this.outputNames.\n    this.inputNames = [];\n    this.outputNames = [];\n    this.feedInputShapes = [];\n    this.feedInputNames = [];\n    this.feedOutputNames = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      // Check that layer is an InputLayer.\n      if (!(layer instanceof InputLayer)) {\n        throw new TypeError(\n            'Input layers to a LayersModel must be InputLayer objects. ' +\n            `Received inputs: ${args.inputs}. ` +\n            `Input ${i} (0-based) originates ` +\n            `from layer type ${layer.getClassName()}.`);\n      }\n      this.inputNames.push(layer.name);\n      this.feedInputShapes.push(layer.batchInputShape);\n\n      this.feedInputNames.push(layer.name);\n    }\n    for (const layer of this.outputLayers) {\n      this.outputNames.push(layer.name);\n    }\n\n    this.internalInputShapes = this.inputs.map(x => x.shape);\n    this.internalOutputShapes = this.outputs.map(x => x.shape);\n\n    /*\n      Container_nodes: set of nodes included in the graph (not all nodes\n      included in the layers are relevant to the current graph).\n    */\n    // ids of all nodes relevant to the Container:\n    const nodesDepths: {[nodeID: string]: number} = {};\n    // To recover nodes from their ID.\n    const nodeIDToNode: {[nodeID: string]: Node} = {};\n    const layersDepths: {[layerID: string]: number} = {};\n    // To layers from their ID.\n    const layerIDToLayer: {[layerID: string]: Layer} = {};\n    const layerIndices: {[layerID: string]: number} = {};\n    const nodesInDecreasingDepth: Node[] = [];\n\n    /**\n     * Builds a map of the graph of layers.\n     *\n     * This recursively updates the map `layerIndices`,\n     * the list `nodesInDecreasingDepth` and the set `containerNodes`.\n     *\n     * @param tensor Some tensor in a graph.\n     * @param finishedNodes Set of nodes whose subgraphs have been traversed\n     *         completely. Useful to prevent duplicated work.\n     * @param nodesInProgress Set of nodes that are currently active on the\n     *         recursion stack. Useful to detect cycles.\n     * @param layer Layer from which `tensor` comes from. If not provided,\n     *   will be obtained from tensor.sourceLayer.\n     * @param nodeIndex Node index from which `tensor` comes from.\n     * @param tensorIndex TensorIndex from which `tensor` comes from.\n     *\n     * @exception RuntimeError if a cycle is detected.\n     */\n    const buildMapOfGraph =\n        (tensor: SymbolicTensor, finishedNodes: Node[], nodesInProgress: Node[],\n         layer?: Layer, nodeIndex?: number, tensorIndex?: number) => {\n          if (layer == null || nodeIndex == null || tensorIndex == null) {\n            layer = tensor.sourceLayer;\n            nodeIndex = tensor.nodeIndex;\n            tensorIndex = tensor.tensorIndex;\n          }\n          const node = layer.inboundNodes[nodeIndex];\n\n          // Prevent cycles.\n          if (nodesInProgress.indexOf(node) !== -1) {\n            throw new RuntimeError(\n                `The tensor ${tensor.name} at layer \"${layer.name}\" ` +\n                'is part of a cycle.');\n          }\n\n          // Don't repeat work for shared subgraphs\n          if (finishedNodes.indexOf(node) !== -1) {\n            return;\n          }\n\n          // Update containerNodes.\n          this.containerNodes.add(Container.nodeKey(layer, nodeIndex));\n\n          // Store the traversal order for layer sorting.\n          if (!(layer.id in layerIndices)) {\n            layerIndices[layer.id] = Object.keys(layerIndices).length;\n          }\n\n          if (nodesInProgress.indexOf(node) === -1) {\n            nodesInProgress.push(node);\n          }\n\n          // Propagate to all previous tensors connected to this node.\n          const numInboundLayers = node.inboundLayers.length;\n          for (let i = 0; i < numInboundLayers; i++) {\n            const x = node.inputTensors[i];\n            const layer = node.inboundLayers[i];\n            const nodeIndex = node.nodeIndices[i];\n            const tensorIndex = node.tensorIndices[i];\n            buildMapOfGraph(\n                x, finishedNodes, nodesInProgress, layer, nodeIndex,\n                tensorIndex);\n          }\n          finishedNodes.push(node);\n          while (nodesInProgress.indexOf(node) >= 0) {\n            nodesInProgress.splice(nodesInProgress.indexOf(node), 1);\n          }\n          nodesInDecreasingDepth.push(node);\n        };\n\n    const finishedNodes: Node[] = [];\n    const nodesInProgress: Node[] = [];\n    for (const x of this.outputs) {\n      buildMapOfGraph(x, finishedNodes, nodesInProgress);\n    }\n\n    const reversedNodesInDecreasingDepth =\n        nodesInDecreasingDepth.slice().reverse();\n    for (const node of reversedNodesInDecreasingDepth) {\n      nodeIDToNode[node.id] = node;\n      // If the depth is not set, the node has no outbound nodes (depth 0).\n      if (!(node.id in nodesDepths)) {\n        nodesDepths[node.id] = 0;\n      }\n      let depth = nodesDepths[node.id];\n\n      // Update the depth of the corresponding layer\n      const previousDepth =\n          (layersDepths[node.outboundLayer.id] == null ?\n               0 :\n               layersDepths[node.outboundLayer.id]);\n\n      /*\n        If we've seen this layer before at a higher depth, we should use that\n        depth instead of the node depth.  This is necessary for shared layers\n        that have inputs at different depth levels in the graph.\n      */\n      depth = Math.max(depth, previousDepth);\n      layersDepths[node.outboundLayer.id] = depth;\n      layerIDToLayer[node.outboundLayer.id] = node.outboundLayer;\n      nodesDepths[node.id] = depth;\n\n      // Update the depth of inbound nodes.\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const inboundLayer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const inboundNode = inboundLayer.inboundNodes[nodeIndex];\n        const previousDepth =\n            (nodesDepths[inboundNode.id] == null ? 0 :\n                                                   nodesDepths[inboundNode.id]);\n        nodesDepths[inboundNode.id] = Math.max(depth + 1, previousDepth);\n        nodeIDToNode[inboundNode.id] = inboundNode;\n      }\n    }\n\n    // Build a dict {depth: list of nodes with this depth}\n    const nodesByDepth: {[depth: string]: Node[]} = {};\n    for (const nodeID in nodesDepths) {\n      const depth = nodesDepths[nodeID];\n      if (!(depth in nodesByDepth)) {\n        nodesByDepth[depth] = [];\n      }\n      nodesByDepth[depth].push(nodeIDToNode[nodeID]);\n    }\n\n    // Build a dict {depth: list of layers with this depth}\n    const layersByDepth: {[depth: string]: Layer[]} = {};\n    for (const layerID in layersDepths) {\n      const depth = layersDepths[layerID];\n      if (!(depth in layersByDepth)) {\n        layersByDepth[depth] = [];\n      }\n      layersByDepth[depth].push(layerIDToLayer[layerID]);\n    }\n\n    // Get sorted list of layer depths.\n    let depthKeys = Object.keys(layersByDepth)\n                        .map(x => parseInt(x, 10))\n                        .sort(generic_utils.reverseNumberCompare);\n\n    // Set this.layers and this.layersByDepth.\n    this.layers = [];\n    for (const depth of depthKeys) {\n      const layersForDepth = layersByDepth[depth];\n      // Container.layers needs to have a deterministic order:\n      // here we order them by traversal order.\n      layersForDepth.sort((a, b) => {\n        const aIndex = layerIndices[a.id];\n        const bIndex = layerIndices[b.id];\n        if (aIndex < bIndex) {\n          return -1;\n        }\n        if (aIndex > bIndex) {\n          return 1;\n        }\n        return 0;\n      });\n      for (const layer of layersForDepth) {\n        if (layer instanceof Container) {\n          this.internalContainerRefs.push(layer);\n        }\n        this.layers.push(layer);\n      }\n    }\n    this.layersByDepth = layersByDepth;\n\n    // Get sorted list of node depths;\n    depthKeys = Object.keys(nodesByDepth)\n                    .map(x => parseInt(x, 10))\n                    .sort(generic_utils.reverseNumberCompare);\n\n    // Check that all tensors required are computable.\n    // computable_tensors: all tensors in the graph\n    // that can be computed from the inputs provided.\n    const computableTensors = this.inputs.slice();\n\n    // To provide a better error msg.\n    const layersWithCompleteInput: string[] = [];\n    for (const depth of depthKeys) {\n      for (const node of nodesByDepth[depth]) {\n        const layer = node.outboundLayer;\n        if (layer != null) {\n          for (const x of node.inputTensors) {\n            if (computableTensors.indexOf(x) === -1) {\n              throw new RuntimeError(\n                  `Graph disconnected: cannot obtain value for tensor ${x}` +\n                  ` at layer \"${layer.name}\". ` +\n                  'The following previous layers were accessed without ' +\n                  `issue: ${layersWithCompleteInput}`);\n            }\n          }\n          for (const x of node.outputTensors) {\n            computableTensors.push(x);\n          }\n          layersWithCompleteInput.push(layer.name);\n        }\n      }\n    }\n\n    // Set this.containerNodes and this.nodesByDepth.\n    this.nodesByDepth = nodesByDepth;\n\n    // Ensure name unicity, which will be crucial for serialization\n    // (since serialized nodes refer to layers by their name).\n    const allNames = this.layers.map(x => x.name);\n    for (const name of allNames) {\n      const numOccurrences = allNames.filter(x => x === name).length;\n      if (numOccurrences !== 1) {\n        throw new RuntimeError(\n            `The name \"${name}\" is used ${numOccurrences} times ` +\n            'in the model. All layer names should be unique. Layer names: ' +\n            JSON.stringify(allNames));\n      }\n    }\n\n    // Layer parameters.\n    // The new container starts with a single inbound node\n    // for its inputs, and no outbound nodes.\n    // Will be appended to by future calls to apply().\n    this.outboundNodes = [];\n    // Will be appended to below, and by future calls to apply().\n    this.inboundNodes = [];\n\n    // Create the node linking internal inputs to internal outputs.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: this.inputs,\n      outputTensors: this.outputs,\n      inputMasks: this.inputs.map(x => null),\n      outputMasks: this.outputs.map(x => null),\n      inputShapes: this.inputs.map(x => x.shape),\n      outputShapes: this.outputs.map(x => x.shape)\n    });\n    this.built = true;\n    this._refCount = 1;  // The ref count of a container always start at 1.\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Container '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose a LayersModel's weights.\n   *\n   * This method decrease the reference count of the LayersModel object by 1.\n   *\n   * A LayersModel is reference-counted. Its reference count is incremented by 1\n   * when it is first constructed and when it is used as a Layer of another\n   * LayersModel.\n   *\n   * If the reference count of a LayersModel becomes 0, the `dispose` method of\n   * all its constituent `Layer`s will be called.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * `dispose` method of its constituent `Layer`s will *not* be called.\n   *\n   * After a LayersModel is disposed, it cannot be used in calls such as\n   * 'predict`, `evaluate` or `fit` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the LayersModel after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the LayersModel has\n   *   already been disposed.\n   */\n  dispose(): DisposeResult {\n    this.assertNotDisposed();\n    const result:\n        DisposeResult = {refCountAfterDispose: null, numDisposedVariables: 0};\n    if (--this._refCount === 0) {\n      for (const layer of this.layers) {\n        result.numDisposedVariables += layer.dispose().numDisposedVariables;\n      }\n\n      // Call dispose on each internally created container layer again to ensure\n      // their refCounts hit zero and their tensors are subsequently deleted.\n      for (const container of this.internalContainerRefs) {\n        result.numDisposedVariables += container.dispose().numDisposedVariables;\n      }\n    }\n    result.refCountAfterDispose = this._refCount;\n    return result;\n  }\n\n  get trainable() {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.layers.forEach(layer => {\n      // tslint:disable-next-line:no-any\n      ((layer as any)._trainableWeights as LayerVariable[])\n          .forEach(w => w.trainable = trainable);\n    });\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    // Porting Note: This check below is to prevent errors where the\n    //   _trainableWeights inherited from the parent class (Layer) gets\n    //   inadvertently used.\n    if (this._trainableWeights.length > 0) {\n      throw new ValueError(\n          'Container instance unexpectedly contains _trainableWeights.' +\n          'The trainable weights of a Container are a union of the ' +\n          'trainable weights of its consituent Layers. Its own ' +\n          '_trainableWeights must remain an empty Array.');\n    }\n\n    if (!this.trainable) {\n      return [];\n    }\n    let weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights = weights.concat(layer.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights.push(...layer.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const layer of this.layers) {\n        trainableWeights.push(...layer.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  /**\n   * Loads all layer weights from a JSON object.\n   *\n   * Porting Note: HDF5 weight files cannot be directly loaded in JavaScript /\n   *   TypeScript. The utility script at `scripts/pykeras.py` offers means\n   *   to convert them into JSON strings compatible with this method.\n   * Porting Note: TensorFlow.js Layers supports only loading by name currently.\n   *\n   * @param weights A JSON mapping weight names to weight values as nested\n   *   arrays of numbers, or a `NamedTensorMap`, i.e., a JSON mapping weight\n   *   names to `tf.Tensor` objects.\n   * @param strict Require that the provided weights exactly match those\n   *   required by the container.  Default: `true`.  Passing `false` means that\n   *   extra weights and missing weights will be silently ignored.\n   */\n  loadWeights(weights: NamedTensorMap, strict = true) {\n    const nameToWeight: {[name: string]: LayerVariable} = {};\n    let totalWeightsCount = 0;\n    for (const layer of this.layers) {\n      for (const weight of layer.weights) {\n        if (nameToWeight[weight.originalName] != null) {\n          throw new ValueError(`Duplicate weight name: ${weight.originalName}`);\n        }\n        nameToWeight[weight.originalName] = weight;\n        totalWeightsCount++;\n      }\n    }\n\n    const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n    for (const name in weights) {\n      // TF 2.2.0 added cell name to the weight name in the format of\n      // layer_name/cell_name/weight_name, we need to remove\n      // the inner cell name.\n      let validatedName = name;\n      if (nameToWeight[name] == null) {\n        const tokens = name.split('/');\n        const shortenNameArray =\n            tokens.slice(0, -2).concat([tokens[tokens.length - 1]]);\n        validatedName = shortenNameArray.join('/');\n      }\n      if (nameToWeight[validatedName] != null) {\n        weightValueTuples.push([nameToWeight[validatedName], weights[name]]);\n      } else if (strict) {\n        throw new ValueError(\n            `Provided weight data has no target variable: ${name}`);\n      }\n      delete nameToWeight[validatedName];\n    }\n\n    if (strict) {\n      // Check that all weights are set.\n      const unsetNames: string[] = [];\n      for (const name in nameToWeight) {\n        unsetNames.push(name);\n      }\n      if (unsetNames.length > 0) {\n        throw new ValueError(\n            `${unsetNames.length} of ${\n                totalWeightsCount} weights are not set: ` +\n            `${unsetNames}`);\n      }\n    }\n\n    batchSetValue(weightValueTuples);\n  }\n\n  /**\n   * Util shared between different serialization methods.\n   * @returns LayersModel config with Keras version information added.\n   */\n  protected updatedConfig(): serialization.ConfigDict {\n    const theConfig = this.getConfig();\n    const modelConfig: serialization.ConfigDict = {};\n    modelConfig['className'] = this.getClassName();\n    modelConfig['config'] = theConfig;\n    modelConfig['kerasVersion'] = `tfjs-layers ${layersVersion}`;\n    // TODO(nielsene): Replace something like K.backend() once\n    // possible.\n    modelConfig['backend'] = 'TensorFlow.js';\n    return modelConfig;\n  }\n\n  /**\n   * Returns a JSON string containing the network configuration.\n   *\n   * To load a network from a JSON save file, use\n   * models.modelFromJSON(jsonString);\n   * @param extraJsonArgs Unused in tfjs-layers, maintained for PyKeras\n   * @param returnString Whether the return value should be stringified\n   *    (default: `true`).\n   * @returns a JSON string if `returnString` (default), or a JSON object if\n   *   `!returnString`.\n   */\n  // tslint:disable-next-line:no-any\n  toJSON(unused?: any, returnString = true): string|PyJsonDict {\n    const modelConfig = convertTsToPythonic(this.updatedConfig()) as PyJsonDict;\n    return returnString ? JSON.stringify(modelConfig) : modelConfig;\n  }\n\n  /**\n   * Call the model on new inputs.\n   *\n   * In this case `call` just reapplies all ops in the graph to the new inputs\n   * (e.g. build a new computational graph from the provided inputs).\n   *\n   * @param inputs A tensor or list of tensors.\n   * @param mask A mask or list of masks. A mask can be either a tensor or null\n   *   (no mask).\n   *\n   * @return A tensor if there is a single output, or a list of tensors if there\n   *   are more than one outputs.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      const feedDict = new FeedDict();\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n      return execute(this.outputs, feedDict, kwargs) as Tensor | Tensor[];\n    });\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      let masks: Tensor[];\n      if (mask == null) {\n        masks = generic_utils.pyListRepeat(null, inputs.length);\n      } else {\n        masks = generic_utils.toList(mask);\n      }\n      // TODO(michaelterry): Add support for mask caching.\n      return this.runInternalGraph(inputs, masks)[1];\n    });\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    const inputShapes = types_utils.normalizeShapeList(inputShape);\n    if (inputShapes.length !== this.inputLayers.length) {\n      throw new ValueError(\n          `Invalid inputShape argument ${inputShape}: ` +\n          `model has ${this.inputLayers.length} tensor inputs.`);\n    }\n\n    // TODO(michaelterry): Add caching\n    const layersToOutputShapes: {[shapeKey: string]: Shape} = {};\n    for (let i = 0; i < inputShapes.length; i++) {\n      const layer = this.inputLayers[i];\n      const inputShape = inputShapes[i];\n      // It's an input layer: computeOutputShape is identity,\n      // and there is only one node and one tensor output.\n      const shapeKey = layer.name + '_0_0';\n      layersToOutputShapes[shapeKey] = inputShape;\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    // Iterate over nodes, by depth level.\n    if (depthKeys.length > 1) {\n      for (const depth of depthKeys) {\n        const nodes = this.nodesByDepth[depth];\n        for (const node of nodes) {\n          // This is always a single layer, never a list.\n          const layer = node.outboundLayer;\n          if (this.inputLayers.map(x => x.id).indexOf(layer.id) !== -1) {\n            // We've already covered the input layers a few lines above.\n            continue;\n          }\n          // Potentially redundant list, same size of node.inputTensors.\n          const inputShapes: Shape[] = [];\n          for (let j = 0; j < node.inboundLayers.length; j++) {\n            const inboundLayer = node.inboundLayers[j];\n            const nodeIndex = node.nodeIndices[j];\n            const tensorIndex = node.tensorIndices[j];\n            const shapeKey = `${inboundLayer.name}_${nodeIndex}_${tensorIndex}`;\n            const inputShape = layersToOutputShapes[shapeKey];\n            inputShapes.push(inputShape);\n          }\n\n          const outputShape = layer.computeOutputShape(\n              generic_utils.singletonOrArray(inputShapes));\n\n          const outputShapes = types_utils.normalizeShapeList(outputShape);\n          const nodeIndex = layer.inboundNodes.indexOf(node);\n          for (let j = 0; j < outputShapes.length; j++) {\n            const shapeKey = `${layer.name}_${nodeIndex}_${j}`;\n            layersToOutputShapes[shapeKey] = outputShapes[j];\n          }\n        }\n      }\n    }\n\n    // Read final output shapes from layersToOutputShapes.\n    const outputShapes: Shape[] = [];\n    const outputShapeKeys: string[] = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      const shapeKey = `${layer.name}_${nodeIndex}_${tensorIndex}`;\n      outputShapeKeys.push(shapeKey);\n    }\n\n    for (let i = 0; i < outputShapeKeys.length; i++) {\n      const key = outputShapeKeys[i];\n      generic_utils.assert(key in layersToOutputShapes);\n      outputShapes.push(layersToOutputShapes[key]);\n    }\n\n    // TODO(michaelterry): Update cache\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  /**\n   * Computes output tensors for new inputs.\n   *\n   * Note:\n   *   - Expects `inputs` to be a list (potentially with 1 element).\n   *\n   * @param inputs List of tensors\n   * @param masks List of masks (tensors or null).\n   * @return Three lists: outputTensors, outputMasks, outputShapes\n   */\n  protected runInternalGraph(inputs: Tensor[], masks?: Tensor[]):\n      [Tensor[], Tensor[], Shape[]] {\n    if (masks == null) {\n      masks = generic_utils.pyListRepeat(null, inputs.length);\n    }\n\n    // Dictionary mapping reference tensors to tuples\n    // (computed tensor, compute mask)\n    // we assume a 1:1 mapping from tensor to mask\n    // TODO: raise exception when a `.computeMask()` call\n    // does not return a list the same size as `call`\n    const tensorMap: {[tensorID: string]: [Tensor, Tensor]} = {};\n    for (let i = 0; i < this.inputs.length; ++i) {\n      const x = this.inputs[i];\n      const y = inputs[i];\n      const mask = masks[i];\n      tensorMap[x.id] = [y, mask];\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    for (const depth of depthKeys) {\n      const nodes = this.nodesByDepth[depth];\n      for (const node of nodes) {\n        // This is always a single layer, never a list.\n        const layer = node.outboundLayer;\n        const referenceInputTensors = node.inputTensors;\n        const referenceOutputTensors = node.outputTensors;\n\n        // If all previous input tensors are available in tensorMap,\n        // then call node.inboundLayer on them.\n        // List of tuples [input, mask]:\n        const computedData = new Array<[Tensor, Tensor]>();\n        for (const x of referenceInputTensors) {\n          if (x.id in tensorMap) {\n            computedData.push(tensorMap[x.id]);\n          }\n        }\n        if (computedData.length === referenceInputTensors.length) {\n          // TODO(michaelterry): Add K.name_scope here, if we need it.\n          let kwargs: Kwargs = {};\n          let computedTensors: Tensor[];\n          let computedMasks: Tensor[];\n          let outputTensors: Tensor[];\n          let outputMasks: Tensor[];\n          // call layer\n          if (node.callArgs != null) {\n            kwargs = node.callArgs;\n          }\n          if (computedData.length === 1) {\n            const [computedTensor, computedMask] = computedData[0];\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMask;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensor, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensor, computedMask));\n            computedTensors = [computedTensor];\n            computedMasks = [computedMask];\n          } else {\n            computedTensors = computedData.map(x => x[0]);\n            computedMasks = computedData.map(x => x[1]);\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMasks;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensors, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensors, computedMasks));\n          }\n\n          if (layer.activityRegularizer) {\n            throw new NotImplementedError(\n                'LayersModel invocation with concrete Tensor value(s) in the ' +\n                'presence of activity regularizer(s) is not supported yet.');\n          }\n          // TODO(michaelterry): Add model updates and losses\n\n          // Update tensor map.\n          for (let i = 0; i < referenceOutputTensors.length; ++i) {\n            const x = referenceOutputTensors[i];\n            const y = outputTensors[i];\n            const mask = outputMasks[i];\n            tensorMap[x.id] = [y, mask];\n          }\n        }\n      }\n    }\n\n    const outputTensors: Tensor[] = [];\n    const outputMasks: Tensor[] = [];\n    const outputShapes: Shape[] = [];\n    for (const x of this.outputs) {\n      generic_utils.assert(\n          x.id in tensorMap, `Could not compute output ${x.name} : ${x.id}`);\n      const [tensor, mask] = tensorMap[x.id];\n      outputShapes.push(tensor.shape);\n      outputTensors.push(tensor);\n      outputMasks.push(mask);\n    }\n\n    // TODO(michaelterry): Add support for caches.\n    return [outputTensors, outputMasks, outputShapes];\n  }\n\n  /**\n   * Builds a map of internal node keys to node ordering.\n   * Used in serializaion a node orderings may change as unused nodes are\n   * dropped. Porting Note:  This helper method was pulled out of getConfig to\n   * improve readability.\n   * @param layers An array of Layers in the model.\n   * @returns Map of Node Keys to index order within the layer.\n   */\n  private buildNodeConversionMap(layers: Layer[]): {[nodeKey: string]: number} {\n    const nodeConversionMap: {[nodeKey: string]: number} = {};\n    let keptNodes: number;\n    for (const layer of this.layers) {\n      keptNodes = layer instanceof Container ? 1 : 0;\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        if (this.containerNodes.has(nodeKey)) {\n          // i.e. we mark it to be saved\n          nodeConversionMap[nodeKey] = keptNodes;\n          keptNodes += 1;\n        }\n      }\n    }\n    return nodeConversionMap;\n  }\n\n  /**\n   * Retrieves a layer based on either its name (unique) or index.\n   *\n   * Indices are based on order of horizontal graph traversal (bottom-up).\n   *\n   * If both `name` and `index` are specified, `index` takes precedence.\n   *\n   * @param name Name of layer.\n   * @param index Index of layer.\n   * @returns A Layer instance.\n   * @throws ValueError: In case of invalid layer name or index.\n   *\n   * @doc {\n   *    heading: 'Layers',\n   *    subheading: 'Classes',\n   *    namespace: 'layers',\n   *    subclasses: ['LayersModel']\n   * }\n   */\n  getLayer(name?: string, index?: number): Layer {\n    if (index != null) {\n      if (this.layers.length <= index) {\n        throw new ValueError(\n            `Was asked to retrieve layer at index ${index}, but model only ` +\n            `has ${this.layers.length} layer(s).`);\n      } else {\n        return this.layers[index];\n      }\n    } else {\n      if (name == null) {\n        throw new ValueError('Provide either a layer name or layer index');\n      }\n    }\n\n    for (const layer of this.layers) {\n      if (layer.name === name) {\n        return layer;\n      }\n    }\n    throw new ValueError(`No such layer: ${name}`);\n  }\n\n  /**\n   * Retrieves the Container's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Container.loss in PyKeras.\n    //   In PyKeras, Container.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return tidy(() => {\n      const losses: Scalar[] = [];\n      for (const layer of this.layers) {\n        for (let nodeIndex = 0; nodeIndex < layer.inboundNodes.length;\n             ++nodeIndex) {\n          const nodeKey = Container.nodeKey(layer, nodeIndex);\n          if (this.containerNodes.has(nodeKey)) {\n            losses.push(...layer.calculateLosses());\n          }\n        }\n      }\n      // TODO(cais): Add any unconditional model-level losses?\n      return losses;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {name: this.name};\n\n    // Build a map from layer unique name (self._node_key)\n    // to the index of the nodes that are saved in the config.\n    // Only nodes in container_nodes are saved.\n    const nodeConversionMap: {[nodeKey: string]: number} =\n        this.buildNodeConversionMap(this.layers);\n\n    // Serialize and save the layers in layerConfigs\n    const layerConfigs = [];\n    for (const layer of this.layers) {\n      const layerClassName = layer.getClassName();\n      const layerConfig = layer.getConfig();\n      const filteredInboundNodes = [];\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const node = layer.inboundNodes[originalNodeIndex];\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        let kwargs = {};\n        if (this.containerNodes.has(nodeKey)) {\n          // The node is relevant to the model:\n          // add to filteredInboundNodes.\n          if (node.callArgs) {\n            try {\n              JSON.stringify(node.callArgs);\n              kwargs = node.callArgs;\n            } catch (err) {\n              console.warn(\n                  `Layer ${layer.name} was passed ` +\n                  `non-serializable keyword arguments: ` +\n                  `${node.callArgs}. They will not be included ` +\n                  `in the serialized model (and thus will be ` +\n                  `missing at deserialization time).`);\n              kwargs = {};\n            }\n          }\n          if (node.inboundLayers.length > 0) {\n            const nodeData = [];\n            for (let i = 0; i < node.inboundLayers.length; i++) {\n              const inboundLayer = node.inboundLayers[i];\n              const nodeIndex = node.nodeIndices[i];\n              const tensorIndex = node.tensorIndices[i];\n              const nodeKey = Container.nodeKey(inboundLayer, nodeIndex);\n              let newNodeIndex = nodeConversionMap[nodeKey];\n              if (newNodeIndex == null) {\n                newNodeIndex = 0;\n              }\n              nodeData.push(\n                  [inboundLayer.name, newNodeIndex, tensorIndex, kwargs]);\n            }\n            filteredInboundNodes.push(nodeData);\n          }\n        }\n      }\n      const dict: serialization.ConfigDict = {};\n      dict['name'] = layer.name;\n      dict['className'] = layerClassName;\n      dict['config'] = layerConfig;\n      dict['inboundNodes'] = filteredInboundNodes;\n      layerConfigs.push(dict);\n    }\n    config['layers'] = layerConfigs;\n    // Gather info about inputs and outputs\n    const modelInputs = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      const nodeIndex = this.inputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.inputLayersTensorIndices[i];\n      modelInputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['inputLayers'] = modelInputs;\n\n    const modelOutputs = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      modelOutputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['outputLayers'] = modelOutputs;\n    return config;\n  }\n\n  /**\n   * Instantiates a LayersModel from its config (output of `get_config()`).\n   * @param cls the class to create\n   * @param config LayersModel config dictionary.\n   * @param customObjects An optional dictionary of custom objects.\n   * @param fastWeightInit Optional flag to use fast weight initialization\n   *   during deserialization. This is applicable to cases in which\n   *   the initialization will be immediately overwritten by loaded weight\n   *   values. Default: `false`.\n   * @returns A LayersModel instance.\n   * @throws ValueError: In case of improperly formatted config dict.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    // Layer instances created during\n    // the graph reconstruction process\n    const createdLayers: {[layerName: string]: Layer} = {};\n\n    // Dictionary mapping layer instances to\n    // node data that specifies a layer call.\n    // It acts as a queue that maintains any unprocessed\n    // layer call until it becomes possible to process it\n    // (i.e. until the input tensors to the call all exist).\n    const unprocessedNodes: {[layer: string]: TensorKeyWithArgsArray[][]} = {};\n    function addUnprocessedNode(\n        layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      if (!(layer.name in unprocessedNodes)) {\n        unprocessedNodes[layer.name] = [nodeData];\n      } else {\n        unprocessedNodes[layer.name].push(nodeData);\n      }\n    }\n\n    function processNode(layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      const inputTensors: SymbolicTensor[] = [];\n      let kwargs;\n      for (const inputData of nodeData) {\n        const inboundLayerName = inputData[0];\n        const inboundNodeIndex = inputData[1];\n        const inboundTensorIndex = inputData[2];\n\n        kwargs = inputData[3] == null ?\n            {} :\n            inputData[3] as serialization.ConfigDict;\n        if (!(inboundLayerName in createdLayers)) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundLayer = createdLayers[inboundLayerName];\n        if (inboundLayer.inboundNodes.length <= inboundNodeIndex) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundNode = inboundLayer.inboundNodes[inboundNodeIndex];\n        inputTensors.push(inboundNode.outputTensors[inboundTensorIndex]);\n      }\n      // Call layer on its inputs, thus creating the node\n      // and building the layer if needed.\n      // Note: This has Eager vs Graph Implications.\n      if (inputTensors.length > 0) {\n        layer.apply(\n            generic_utils.singletonOrArray(inputTensors),\n            kwargs);  // was ** kwargs\n      }\n    }\n\n    /**\n     * Deserialize a layer, then call it on appropriate inputs.\n     * @param layerData: layer config dict.\n     * @throws ValueError: In case of improperly formatted `layer_data`\n     * dict.\n     */\n    function processLayer(layerData: serialization.ConfigDict|null) {\n      const layerName = layerData['name'] as string;\n      // Instantiate layer.\n      const layer =\n          deserializeLayer(\n              layerData,\n              config['customObjects'] != null ?\n                  config['customObjects'] as serialization.ConfigDict :\n                  {}) as Layer;\n      layer.setFastWeightInitDuringBuild(fastWeightInit);\n      createdLayers[layerName] = layer;\n      // Gather layer inputs.\n      const inboundNodesData =\n          layerData['inboundNodes'] as TensorKeyWithArgsArray[][];\n      inboundNodesData.forEach(nodeData => {\n        if (!(nodeData instanceof Array)) {\n          throw new ValueError(\n              `Corrupted configuration, expected array for nodeData: ${\n                  nodeData}`);\n        }\n        // We don't process nodes (i.e. make layer calls)\n        // on the fly because the inbound node may not yet exist,\n        // in case of layer shared at different topological depths\n        // (e.g.a model such as A(B(A(B(x)))))\n        addUnprocessedNode(layer, nodeData);\n      });\n    }\n\n    // First, we create all layers and enqueue nodes to be processed.\n    const name = config['name'];\n    const layersFromConfig = config['layers'] as serialization.ConfigDict[];\n    for (const layerData of layersFromConfig) {\n      processLayer(layerData);\n    }\n\n    // Then we process nodes in order of layer depth.\n    // Nodes that cannot yet be processed(if the inbound node\n    // does not yet exist) are re - enqueued, and the process\n    // is repeated until all nodes are processed.\n    while (!generic_utils.isObjectEmpty(unprocessedNodes)) {\n      for (const layerData of layersFromConfig) {\n        const layer = createdLayers[layerData['name'] as string];\n        if (layer.name in unprocessedNodes) {\n          const currentUnprocessedNodesForLayer = unprocessedNodes[layer.name];\n          delete unprocessedNodes[layer.name];\n          for (const nodeData of currentUnprocessedNodesForLayer) {\n            processNode(layer, nodeData);\n          }\n        }\n      }\n    }\n\n    const inputTensors: SymbolicTensor[] = [];\n    const outputTensors: SymbolicTensor[] = [];\n    const inputLayersFromConfig =\n        config['inputLayers'] as serialization.ConfigDict[];\n    for (const layerData of inputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      inputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    const outputLayersFromConfig =\n        config['outputLayers'] as serialization.ConfigDict[];\n    for (const layerData of outputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      outputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    return new cls({inputs: inputTensors, outputs: outputTensors, name});\n  }\n\n  /**\n   * Determine whether the container is stateful.\n   *\n   * Porting Note: this is the equivalent of the stateful @property of\n   *   the Container class in PyKeras.\n   */\n  get stateful(): boolean {\n    // Porting Note: This check is to prevent inadvertent setting of the\n    //   _stateful property of the Container instance.\n    if (this._stateful) {\n      throw new ValueError(\n          'Container instance unexpectedly has _stateful = true. The ' +\n          'statefulness of a Container is determined by the Layers it ' +\n          'contains. Its _stateful property must remain the default false.');\n    }\n    for (const layer of this.layers) {\n      if (layer.stateful) {\n        return true;\n      }\n    }\n    return false;\n  }\n\n  /**\n   * Reset the state of all stateful constituent layers (if any).\n   *\n   * Examples of stateful layers include RNN layers whose `stateful` property\n   * is set as `true`.\n   */\n  resetStates() {\n    tidy(() => {\n      this.layers.forEach(layer => {\n        // tslint:disable:no-any\n        if (layer.stateful) {\n          layer.resetStates();\n        }\n        // tslint:enable:no-any\n      });\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, mul, Tensor, Tensor1D, tensor1d, tidy} from '@tensorflow/tfjs-core';\n\n/**\n * For multi-class classification problems, this object is designed to store a\n * mapping from class index to the \"weight\" of the class, where higher weighted\n * classes have larger impact on loss, accuracy, and other metrics.\n *\n * This is useful for cases in which you want the model to \"pay more attention\"\n * to examples from an under-represented class, e.g., in unbalanced datasets.\n */\nexport type ClassWeight = {\n  [classIndex: number]: number\n};\n\n/**\n * Class weighting for a model with multiple outputs.\n *\n * This object maps each output name to a class-weighting object.\n */\nexport type ClassWeightMap = {\n  [outputName: string]: ClassWeight\n};\n\nfunction standardizeSampleOrClassWeights(\n    xWeight: ClassWeight|ClassWeight[]|ClassWeightMap, outputNames: string[],\n    weightType: 'sampleWeight'|'classWeight'): ClassWeight[] {\n  const numOutputs = outputNames.length;\n  if (xWeight == null || (Array.isArray(xWeight) && xWeight.length === 0)) {\n    return outputNames.map(name => null);\n  }\n  if (numOutputs === 1) {\n    if (Array.isArray(xWeight) && xWeight.length === 1) {\n      return xWeight;\n    } else if (typeof xWeight === 'object' && outputNames[0] in xWeight) {\n      return [(xWeight as ClassWeightMap)[outputNames[0]]];\n    } else {\n      return [xWeight as ClassWeight];\n    }\n  }\n  if (Array.isArray(xWeight)) {\n    if (xWeight.length !== numOutputs) {\n      throw new Error(\n          `Provided ${weightType} is an array of ${xWeight.length} ` +\n          `element(s), but the model has ${numOutputs} outputs. ` +\n          `Make sure a set of weights is provided for each model output.`);\n    }\n    return xWeight;\n  } else if (\n      typeof xWeight === 'object' && Object.keys(xWeight).length > 0 &&\n      typeof (xWeight as ClassWeightMap)[Object.keys(xWeight)[0]] ===\n          'object') {\n    const output: ClassWeight[] = [];\n    outputNames.forEach(outputName => {\n      if (outputName in xWeight) {\n        output.push((xWeight as ClassWeightMap)[outputName]);\n      } else {\n        output.push(null);\n      }\n    });\n    return output;\n  } else {\n    throw new Error(\n        `The model has multiple (${numOutputs}) outputs, ` +\n        `so ${weightType} must be either an array with ` +\n        `${numOutputs} elements or an object with ${outputNames} keys. ` +\n        `Provided ${weightType} not understood: ${JSON.stringify(xWeight)}`);\n  }\n}\n\n/**\n * Standardize class weighting objects.\n *\n * This function takes a single class-weighting object, an array of them,\n * or a map from output name to class-weighting object. It compares it to the\n * output name(s) of the model, base on which it outputs an array of\n * class-weighting objects of which the length matches the number of outputs.\n *\n * @param classWeight Input class-weighting object(s).\n * @param outputNames All output name(s) of the model.\n * @return An array of class-weighting objects. The length of the array matches\n *   the model's number of outputs.\n */\nexport function standardizeClassWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'classWeight');\n}\n\nexport function standardizeSampleWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'sampleWeight');\n}\n\n/**\n * Standardize by-sample and/or by-class weights for training.\n *\n * Note that this function operates on one model output at a time. For a model\n * with multiple outputs, you must call this function multiple times.\n *\n * @param y The target tensor that the by-sample and/or by-class weight is for.\n *     The values of y are assumed to encode the classes, either directly\n *     as an integer index, or as one-hot encoding.\n * @param sampleWeight By-sample weights.\n * @param classWeight By-class weights: an object mapping class indices\n *     (integers) to a weight (float) to apply to the model's loss for the\n *     samples from this class during training. This can be useful to tell the\n *     model to \"pay more attention\" to samples from an under-represented class.\n * @param sampleWeightMode The mode for the sample weights.\n * @return A Promise of weight tensor, of which the size of the first dimension\n *     matches that of `y`.\n */\nexport async function standardizeWeights(\n    y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight,\n    sampleWeightMode?: 'temporal'): Promise<Tensor> {\n  if (sampleWeight != null || sampleWeightMode != null) {\n    // TODO(cais): Once 'temporal' mode is implemented, document it in the doc\n    // string.\n    throw new Error('Support sampleWeight is not implemented yet');\n  }\n\n  if (classWeight != null) {\n    // Apply class weights per sample.\n    const yClasses: Tensor1D = tidy(() => {\n      if (y.shape.length === 1) {\n        // Assume class indices.\n        return y.clone() as Tensor1D;\n      } else if (y.shape.length === 2) {\n        if (y.shape[1] > 1) {\n          // Assume one-hot encoding of classes.\n          const axis = 1;\n          return y.argMax(axis);\n        } else if (y.shape[1] === 1) {\n          // Class index.\n          return y.reshape([y.shape[0]]);\n        } else {\n          throw new Error(\n              `Encountered unexpected last-dimension size (${y.shape[1]}) ` +\n              `during handling of class weights. The size is expected to be ` +\n              `>= 1.`);\n        }\n      } else {\n        throw new Error(\n            `Unexpected rank of target (y) tensor (${y.rank}) during ` +\n            `handling of class weights. The rank is expected to be 1 or 2.`);\n      }\n    });\n\n    const yClassIndices = Array.from(await yClasses.data());\n    dispose(yClasses);\n    const classSampleWeight: number[] = [];\n    yClassIndices.forEach(classIndex => {\n      if (classWeight[classIndex] == null) {\n        throw new Error(\n            `classWeight must contain all classes in the training data. ` +\n            `The class ${classIndex} exists in the data but not in ` +\n            `classWeight`);\n      } else {\n        classSampleWeight.push(classWeight[classIndex]);\n      }\n    });\n\n    return tensor1d(classSampleWeight, 'float32');\n  } else {\n    return null;\n  }\n}\n\n/**\n * Apply per-sample weights on the loss values from a number of samples.\n *\n * @param losses Loss tensor of shape `[batchSize]`.\n * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.\n * @returns Tensor of the same shape as`losses`.\n */\nexport function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor) {\n  return mul(losses, sampleWeights);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using TensorFlow.js datasets.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar} from '@tensorflow/tfjs-core';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {TensorOrArrayOrMap} from '../types';\nimport {singletonOrArray, toList} from '../utils/generic_utils';\n\nimport {Dataset, LazyIterator} from './dataset_stub';\nimport {ClassWeight, ClassWeightMap, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Interface for configuring model training based on a dataset object.\n */\nexport interface ModelFitDatasetArgs<T> {\n  /**\n   * (Optional) Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. It should\n   * typically be equal to the number of samples of your dataset divided by\n   * the batch size, so that `fitDataset`() call can utilize the entire dataset.\n   * If it is not provided, use `done` return value in `iterator.next()` as\n   * signal to finish an epoch.\n   */\n  batchesPerEpoch?: number;\n\n  /**\n   * Integer number of times to iterate over the training dataset.\n   */\n  epochs: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be any of the following:\n   *\n   *   - An array `[xVal, yVal]`, where the two values may be `tf.Tensor`,\n   *     an array of Tensors, or a map of string to Tensor.\n   *   - Similarly, an array ` [xVal, yVal, valSampleWeights]`\n   *     (not implemented yet).\n   *   - a `Dataset` object with elements of the form `{xs: xVal, ys: yVal}`,\n   *     where `xs` and `ys` are the feature and label tensors, respectively.\n   *\n   * If `validationData` is an Array of Tensor objects, each `tf.Tensor` will be\n   * sliced into batches during validation, using the parameter\n   * `validationBatchSize` (which defaults to 32). The entirety of the\n   * `tf.Tensor` objects will be used in the validation.\n   *\n   * If `validationData` is a dataset object, and the `validationBatches`\n   * parameter is specified, the validation will use `validationBatches` batches\n   * drawn from the dataset object. If `validationBatches` parameter is not\n   * specified, the validation will stop when the dataset is exhausted.\n   *\n   * The model will not be trained on this data.\n   */\n  validationData?: [\n    TensorOrArrayOrMap, TensorOrArrayOrMap\n  ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|Dataset<T>;\n\n  /**\n   * Optional batch size for validation.\n   *\n   * Used only if `validationData` is an array of `tf.Tensor` objects, i.e., not\n   * a dataset object.\n   *\n   * If not specified, its value defaults to 32.\n   */\n  validationBatchSize?: number;\n\n  /**\n   * (Optional) Only relevant if `validationData` is specified and is a dataset\n   * object.\n   *\n   * Total number of batches of samples to draw from `validationData` for\n   * validation purpose before stopping at the end of every epoch. If not\n   * specified, `evaluateDataset` will use `iterator.next().done` as signal to\n   * stop validation.\n   */\n  validationBatches?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - a `number`: Will yield every `number` milliseconds.\n   *   - `'never'`: never yield. (But yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n}\n\nexport interface FitDatasetElement {\n  xs: TensorOrArrayOrMap;\n  ys: TensorOrArrayOrMap;\n}\n\n/**\n * Interface for configuring model evaluation based on a dataset object.\n */\nexport interface ModelEvaluateDatasetArgs {\n  /**\n   * Number of batches to draw from the dataset object before ending the\n   * evaluation.\n   */\n  batches?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n}\n\n// Default batch size used during tensor-based validation.\nconst DEFAULT_VALIDATION_BATCH_SIZE = 32;\n\n/**\n * Standardize the output of a dataset iterator for use by\n * LayersModel.fitDataset().\n *\n * @param model: A `tf.LayersModel` object.\n * @param iteratorOut The output of a dataset iterator. It is required to be\n *   an object of the form `{xs: TensorOrArrayOrMap, ys:\n * TensorOrArrayOrMap}`, where `TensorOrArrayOrMap` is a single `tf.Tensor`,\n * a `tf.Tensor[]`, or a flat map from string names to `tf.Tensor`s.\n * @returns A flat array of `tf.Tensor` objects: the input `tf.Tensor`s\n *   followed by the target `tf.Tensor`s.  When `tf.Tensor`s are provided\n *   as a map, the order in the resulting array is taken from the `inputNames`\n *   and `outputNames` of the model.\n */\nfunction standardizeDataIteratorOutput(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, iteratorOut: {}): {xs: tfc.Tensor[], ys: tfc.Tensor[]} {\n  let xs: TensorOrArrayOrMap;\n  let ys: TensorOrArrayOrMap;\n\n  const iteratorOutObj = iteratorOut as FitDatasetElement;\n  xs = iteratorOutObj['xs'];\n  ys = iteratorOutObj['ys'];\n  tfc.util.assert(\n      xs != null && ys != null,\n      () => 'A Dataset iterator for fitDataset() is expected to generate ' +\n          'objects of the form `{xs: xVal, ys: yVal}`, where the two ' +\n          'values may be `tf.Tensor`, an array of Tensors, or a map of ' +\n          'string to Tensor.  The provided Dataset instead generates ' +\n          `${iteratorOut}`);\n\n  const flattenedXs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('input', model.inputNames, xs);\n  const flattenedYs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('output', model.outputNames, ys);\n\n  const batchSize: number = flattenedXs[0].shape[0];\n\n  tfc.util.assert(\n      flattenedXs.length === model.inputs.length,\n      () => `LayersModel has ${model.inputs.length} inputs, but the dataset ` +\n          `provides ${flattenedXs.length} inputs.  (Expected input keys: ` +\n          `${JSON.stringify(model.inputNames)})`);\n\n  tfc.util.assert(\n      flattenedYs.length === model.outputs.length,\n      () =>\n          `LayersModel has ${model.outputs.length} outputs, but the dataset ` +\n          `provides ${flattenedYs.length} outputs.  (Expected output keys: ` +\n          `${JSON.stringify(model.outputNames)})`);\n\n  for (let xIndex = 0; xIndex < flattenedXs.length; xIndex++) {\n    tfc.util.assert(\n        flattenedXs[xIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: input ` +\n            `${model.inputNames[xIndex]} has ${\n                  flattenedXs[xIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  for (let yIndex = 0; yIndex < flattenedYs.length; yIndex++) {\n    tfc.util.assert(\n        flattenedYs[yIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: output ` +\n            `${model.outputNames[yIndex]} has ${\n                  flattenedYs[yIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  return {xs: flattenedXs, ys: flattenedYs};\n}\n\nfunction flattenTensorOrArrayOrMap(\n    inputOrOutput: string, names: string[], values: TensorOrArrayOrMap) {\n  if (values instanceof tfc.Tensor) {\n    return [values];\n  } else if (Array.isArray(values)) {\n    tfc.util.assert(\n        values.length === names.length,\n        () => `Received an array of ${values.length} Tensors, but expected ${\n            names.length} to match the ${inputOrOutput} keys ${names}.`);\n    return values;\n  } else {\n    const result: tfc.Tensor[] = [];\n    // Check that all the required keys are available.\n    for (const name of names) {\n      if (values[name] == null) {\n        throw new ValueError(\n            `The feature data generated by the dataset lacks the required ` +\n            `${inputOrOutput} key '${name}'.`);\n      }\n      result.push(values[name]);\n    }\n    return result;\n  }\n}\n\nfunction standardizeTensorValidationData<T>(\n    data:\n        [\n          tfc.Tensor|tfc.Tensor[], tfc.Tensor|tfc.Tensor[]\n        ]|[tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n           tfc.Tensor | tfc.Tensor[]]):\n    {xs: tfc.Tensor|tfc.Tensor[], ys: tfc.Tensor|tfc.Tensor[]} {\n  if (data.length === 3) {\n    throw new NotImplementedError(\n        'Validation with sample weights is not implemented yet.');\n  }\n  return {xs: data[0], ys: data[1]};\n}\n\nexport async function fitDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>,\n    args: ModelFitDatasetArgs<T>): Promise<History> {\n  const hasBatchesPerEpoch = args.batchesPerEpoch != null;\n  tfc.util.assert(\n      model.optimizer != null,\n      () => 'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileConfig).');\n\n  tfc.util.assert(\n      args != null,\n      () => `For fitDataset(), the 2nd argument (config) is required, ` +\n          `but it is not provided in this call.`);\n  tfc.util.assert(\n      args.epochs != null && args.epochs > 0 && Number.isInteger(args.epochs),\n      () => `For fitDataset(), config.epochs is expected to be a positive ` +\n          `integer, but got ${args.epochs}`);\n  tfc.util.assert(\n      !hasBatchesPerEpoch ||\n          (args.batchesPerEpoch > 0 && Number.isInteger(args.batchesPerEpoch)),\n      () => `For fitDataset(), config.batchesPerEpoch is expected to be a ` +\n          `positive integer if specified, but got ${args.batchesPerEpoch}`);\n  tfc.util.assert(\n      // tslint:disable-next-line:no-any\n      (args as any)['validationSplit'] == null,\n      () => '`validationSplit` is not supported by `fitDataset()`. ' +\n          'Use validationData instead.');\n\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n\n  try {\n    const doValidation = args.validationData != null;\n    let valXs: tfc.Tensor|tfc.Tensor[];\n    let valYs: tfc.Tensor|tfc.Tensor[];\n    if (doValidation) {\n      if (isDatasetObject(args.validationData)) {\n        tfc.util.assert(\n            args.validationBatches == null ||\n                (args.validationBatches > 0 &&\n                 Number.isInteger(args.validationBatches)),\n            () => `For fitDataset() with dataset-based validation, ` +\n                `config.validationBatches is expected not to be provided, ` +\n                `or to be a positive integer, ` +\n                `but got ${args.validationBatches}`);\n      } else {\n        const validationData = standardizeTensorValidationData(\n            args.validationData as\n                    [tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[]] |\n            [\n              tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n              tfc.Tensor | tfc.Tensor[]\n            ]);\n        valXs = validationData.xs;\n        valYs = validationData.ys;\n      }\n    }\n\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let callbackMetrics: string[];\n    if (doValidation) {\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const verbose = args.verbose == null ? 1 : args.verbose;\n    const {callbackList, history} = configureCallbacks(\n        callbacks, verbose, args.epochs, null, null,\n        getStepsPerEpoch(dataset, args),\n        null,  // Batch size determined by the dataset itself.\n        doValidation, callbackMetrics);\n    callbackList.setModel(model);\n    model.history = history;\n\n    await callbackList.onTrainBegin();\n    model.stopTraining_ = false;\n    let epoch = args.initialEpoch == null ? 0 : args.initialEpoch;\n\n    let dataIterator = await dataset.iterator();\n    while (epoch < args.epochs) {\n      const epochLogs: UnresolvedLogs = {};\n      await callbackList.onEpochBegin(epoch);\n      let stepsDone = 0;\n      let batchIndex = 0;\n      if (!hasBatchesPerEpoch) {\n        dataIterator = await dataset.iterator();\n      }\n      while (hasBatchesPerEpoch ? stepsDone < args.batchesPerEpoch : true) {\n        const iteratorOut = await dataIterator.next();\n\n        // If `batchesPerEpoch` is specified, the dataset should not be\n        // exhausted until all epoches are done.\n        if (hasBatchesPerEpoch && iteratorOut.done) {\n          console.warn(\n              'You provided `batchesPerEpoch` as ' +\n              `${args.batchesPerEpoch}, ` +\n              'but your dataset iterator ran out of data after ' +\n              `${stepsDone} batches; ` +\n              'interrupting training. Make sure that your ' +\n              'dataset can generate at least `batchesPerEpoch * epochs` ' +\n              'batches (in this case, ' +\n              `${args.batchesPerEpoch * args.epochs} batches). ` +\n              'You may need to use the repeat() function when building ' +\n              'your dataset.');\n          break;\n        }\n\n        if (iteratorOut.value != null) {\n          const {xs, ys} =\n              standardizeDataIteratorOutput(model, iteratorOut.value);\n          const batchLogs: UnresolvedLogs = {};\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = xs[0].shape[0];\n\n          await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n          const sampleWeights: tfc.Tensor[] = [];\n          if (args.classWeight != null) {\n            const standardClassWeights =\n                standardizeClassWeights(args.classWeight, model.outputNames);\n            for (let i = 0; i < standardClassWeights.length; ++i) {\n              sampleWeights.push(await standardizeWeights(\n                  ys[i], null, standardClassWeights[i]));\n            }\n          }\n\n          // Train on batch.\n          const ins = xs.concat(ys).concat(sampleWeights);\n          const outs = trainFunction(ins);\n          tfc.dispose(ins);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n          }\n\n          await callbackList.onBatchEnd(batchIndex, batchLogs);\n          disposeTensorsInLogs(batchLogs);\n\n          batchIndex++;\n          stepsDone++;\n        }\n\n        if (hasBatchesPerEpoch ? stepsDone >= args.batchesPerEpoch :\n                                 iteratorOut.done) {\n          // Epoch finished. Perform validation.\n          if (doValidation) {\n            let valOuts: tfc.Scalar[];\n            if (isDatasetObject(args.validationData)) {\n              valOuts = toList(await model.evaluateDataset(\n                  args.validationData, {batches: args.validationBatches}));\n            } else {\n              valOuts = toList(model.evaluate(valXs, valYs, {\n                batchSize: args.validationBatchSize == null ?\n                    DEFAULT_VALIDATION_BATCH_SIZE :\n                    args.validationBatchSize,\n                verbose: 0\n              }));\n            }\n            for (let i = 0; i < model.metricsNames.length; ++i) {\n              epochLogs[`val_${model.metricsNames[i]}`] = valOuts[i];\n            }\n          }\n          // Call `break` to exit one epoch lopp after validation is done. If\n          // config.batchesPerEpoch is specified, an epoch while loop will\n          // stop when `stepsDone >= config.batchesPerEpoch`. When\n          // config.batchesPerEpoch is not provided, the following `break` is\n          // required to exit the while lopp after dataset is exhausted.\n          break;\n        }\n\n        if (model.stopTraining_) {\n          break;\n        }\n      }\n      await callbackList.onEpochEnd(epoch, epochLogs);\n      epoch++;\n      if (model.stopTraining_) {\n        break;\n      }\n    }\n    await callbackList.onTrainEnd();\n    await model.history.syncData();\n    return model.history;\n  } finally {\n    model.isTraining = false;\n  }\n}\n\n/** Helper function that determines number of steps (batches) per epoch. */\nfunction getStepsPerEpoch<T>(\n    dataset: Dataset<T>, args: ModelFitDatasetArgs<T>): number {\n  // Attempt to determine # of batches in an epoch.\n  let stepsPerEpoch: number = null;\n  if (args.batchesPerEpoch != null) {\n    stepsPerEpoch = args.batchesPerEpoch;\n  } else if (Number.isFinite(dataset.size)) {\n    stepsPerEpoch = dataset.size;\n  }\n  return stepsPerEpoch;\n}\n\n// Check if provided object is a Dataset object by checking its .iterator\n// element.\nfunction isDatasetObject<T>(\n    dataset:\n        [\n          TensorOrArrayOrMap, TensorOrArrayOrMap\n        ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|\n    Dataset<T>): boolean {\n  return (typeof (dataset as Dataset<T>).iterator === 'function');\n}\n\n// Check if provided object is a LazyIterator object by checking it's .next\n// element.\nfunction isLazyIteratorObject<T>(iterator: Dataset<T>|\n                                 LazyIterator<T>): boolean {\n  return (typeof (iterator as LazyIterator<T>).next === 'function');\n}\n\nexport async function evaluateDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>|LazyIterator<T>,\n    args: ModelEvaluateDatasetArgs): Promise<tfc.Scalar|tfc.Scalar[]> {\n  args = args || {};\n  const hasBatches = args.batches != null;\n  const f = model.testFunction;\n  let outs: tfc.Scalar[] = [];\n  if (args.verbose > 0) {\n    throw new NotImplementedError('Verbose mode is not implemented yet.');\n  }\n\n  tfc.util.assert(\n      !hasBatches || (args.batches > 0 && Number.isInteger(args.batches)),\n      () => 'Test loop expects `batches` to be a positive integer, but ' +\n          `received ${JSON.stringify(args.batches)}`);\n  const dataIterator = isLazyIteratorObject(dataset) ?\n      dataset as LazyIterator<T>:\n      await (dataset as Dataset<T>).iterator();\n  // Keeps track of number of examples used in this evaluation.\n  let numExamples = 0;\n  let batch = 0;\n\n  while (hasBatches ? batch < args.batches : true) {\n    const iteratorOut = await dataIterator.next();\n    outs = tfc.tidy(() => {\n      if (iteratorOut.value) {\n        // TODO(cais): Once real dataset is available, use\n        //   `map(x => standardizeDataIteratorOutput(model, x).map(f)`.\n        const {xs, ys} =\n            standardizeDataIteratorOutput(model, iteratorOut.value);\n        const xsAndYs = xs.concat(ys);\n        const batchOuts = tfc.tidy(() => f(xsAndYs));\n        tfc.dispose(xsAndYs);\n\n        if (batch === 0) {\n          for (let i = 0; i < batchOuts.length; ++i) {\n            outs.push(scalar(0));\n          }\n        }\n\n        const batchSize = xsAndYs[0].shape[0];\n        for (let i = 0; i < batchOuts.length; ++i) {\n          const batchOut = batchOuts[i];\n          const oldScalar = outs[i];\n          outs[i] =\n              tfc.tidy(() => tfc.add(outs[i], tfc.mul(batchSize, batchOut)));\n          if (batch > 0) {\n            tfc.dispose(oldScalar);\n          }\n        }\n        tfc.dispose(batchOuts);\n        numExamples += batchSize;\n\n        ++batch;\n      }\n      return outs;\n    });\n\n    if (iteratorOut.done) {\n      if (hasBatches) {\n        console.warn(\n            'Your dataset iterator ran out of data during evaluateDataset(). ' +\n            'Interrupting evalution. Make sure that your ' +\n            'dataset can generate at least `batches` ' +\n            `batches (in this case, ${args.batches} batches). ` +\n            'You may need to use the repeat() function when building ' +\n            'your dataset.');\n      }\n      break;\n    }\n  }\n\n  for (let i = 0; i < outs.length; ++i) {\n    const oldScalar = outs[i];\n    outs[i] = tfc.div(outs[i], numExamples);\n    tfc.dispose(oldScalar);\n  }\n\n  return singletonOrArray(outs);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using tf.Tensor objects.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Scalar, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport {expandDims, gather, sliceAlongFirstAxis} from '../backend/tfjs_backend';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {range} from '../utils/math_utils';\nimport {ClassWeight, ClassWeightMap} from './training_utils';\n\n/**\n * Interface configuration model training based on data as `tf.Tensor`s.\n */\nexport interface ModelFitArgs {\n  /**\n   * Number of samples per gradient update. If unspecified, it\n   * will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Integer number of times to iterate over the training data arrays.\n   */\n  epochs?: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Float between 0 and 1: fraction of the training data\n   * to be used as validation data. The model will set apart this fraction of\n   * the training data, will not train on it, and will evaluate the loss and\n   * any model metrics on this data at the end of each epoch.\n   * The validation data is selected from the last samples in the `x` and `y`\n   * data provided, before shuffling.\n   */\n  validationSplit?: number;\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal,\n   * valSampleWeights]. The model will not be trained on this data.\n   * `validationData` will override `validationSplit`.\n   */\n  validationData?: [\n    Tensor|Tensor[], Tensor|Tensor[]\n  ]|[Tensor | Tensor[], Tensor|Tensor[], Tensor|Tensor[]];\n\n  /**\n   * Whether to shuffle the training data before each epoch. Has\n   * no effect when `stepsPerEpoch` is not `null`.\n   */\n  shuffle?: boolean;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n\n  /**\n   * Optional array of the same length as x, containing\n   * weights to apply to the model's loss for each sample. In the case of\n   * temporal data, you can pass a 2D array with shape (samples,\n   * sequenceLength), to apply a different weight to every timestep of every\n   * sample. In this case you should make sure to specify\n   * sampleWeightMode=\"temporal\" in compile().\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. When training\n   * with Input Tensors such as TensorFlow data tensors, the default `null` is\n   * equal to the number of unique samples in your dataset divided by the\n   * batch size, or 1 if that cannot be determined.\n   */\n  stepsPerEpoch?: number;\n\n  /**\n   * Only relevant if `stepsPerEpoch` is specified. Total number of steps\n   * (batches of samples) to validate before stopping.\n   */\n  validationSteps?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - any `number`: yield every `number` milliseconds.\n   *   - `'never'`: never yield. (yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n}\n\nexport function checkBatchSize(batchSize: number) {\n  tfc.util.assert(\n      batchSize > 0 && Number.isInteger(batchSize),\n      () => `batchSize is required to be a positive integer, but got ${\n          batchSize}`);\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by start and stop indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArraysByIndices()` together.\n *\n * @param arrays: the input.\n * @param start: the starting index (inclusive).\n * @param stop: the stopping index (exclusive).\n * @returns The result of the slicing. If `arrays` is an `Array` of\n *   `tf.Tensor`s, the slicing will be applied to all elements of the `Array`\n *   in the same way.\n */\nexport function sliceArrays(\n    arrays: Tensor|Tensor[], start: number, stop: number): Tensor|Tensor[] {\n  if (arrays == null) {\n    return [null];\n  } else if (Array.isArray(arrays)) {\n    return arrays.map(array => sliceAlongFirstAxis(array, start, stop - start));\n  } else {  // Tensor.\n    return sliceAlongFirstAxis(arrays, start, stop - start);\n  }\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by random-order indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArrays()` together.\n *\n * @param arrays The input `tf.Tensor` or `Array` of `tf.Tensor`s to slice.\n *   If an `Array` of `tf.Tensor`s, all `tf.Tensor`s will be sliced in the\n *   same fashion.\n * @param indices The indices to use for slicing along the first (batch)\n *   dimension.\n * @returns Result(s) of the slicing.\n */\nexport function sliceArraysByIndices(\n    arrays: Tensor|Tensor[], indices: Tensor1D): Tensor|Tensor[] {\n  return tfc.tidy(() => {\n    if (arrays == null) {\n      return null;\n    } else if (Array.isArray(arrays)) {\n      return arrays.map(\n          array => (sliceArraysByIndices(array, indices) as Tensor));\n    } else {\n      // TODO(cais): indices should be a pre-constructed Tensor1D to avoid\n      //   tensor1d() calls.\n      return gather(\n          arrays, indices.dtype === 'int32' ? indices : indices.toInt());\n    }\n  });\n}\n\n/**\n * Returns a list of batch indices (tuples of indices).\n * @param size: Integer, total size of the data to slice into batches.\n * @param batchSize: Integer, batch size.\n * @returns An Array of [batchStart, batchEnd] tuples. batchStart is\n *   inclusive; batchEnd is exclusive. I.e., each batch consists of indices x\n *   that satisfy batchStart <= x < batchEnd.\n */\nexport function makeBatches(\n    size: number, batchSize: number): Array<[number, number]> {\n  const output: Array<[number, number]> = [];\n  let batchStart = 0;\n  let batchEnd: number = null;\n  while (batchStart < size) {\n    batchEnd = batchStart + batchSize;\n    if (batchEnd >= size) {\n      batchEnd = size;\n    }\n    output.push([batchStart, batchEnd]);\n    batchStart = batchEnd;\n  }\n  return output;\n}\n\n/**\n * Abstract fit function for `f(ins)`.\n * @param f A Function returning a list of tensors. For training, this\n *   function is expected to perform the updates to the variables.\n * @param ins List of tensors to be fed to `f`.\n * @param outLabels List of strings, display names of the outputs of `f`.\n * @param batchSize Integer batch size or `== null` if unknown. Default : 32.\n * @param epochs Number of times to iterate over the data. Default : 1.\n * @param verbose Verbosity mode: 0, 1, or 2. Default: 1.\n * @param callbacks List of callbacks to be called during training.\n * @param valF Function to call for validation.\n * @param valIns List of tensors to be fed to `valF`.\n * @param shuffle Whether to shuffle the data at the beginning of every\n * epoch. Default : true.\n * @param callbackMetrics List of strings, the display names of the metrics\n *   passed to the callbacks. They should be the concatenation of the\n *   display names of the outputs of `f` and the list of display names\n *   of the outputs of `valF`.\n * @param initialEpoch Epoch at which to start training (useful for\n *   resuming a previous training run). Default : 0.\n * @param stepsPerEpoch Total number of steps (batches on samples) before\n *   declaring one epoch finished and starting the next epoch. Ignored with\n *   the default value of `undefined` or `null`.\n * @param validationSteps Number of steps to run validation for (only if\n *   doing validation from data tensors). Not applicable for tfjs-layers.\n * @returns A `History` object.\n */\nasync function fitLoop(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, f: (data: Tensor[]) => Scalar[], ins: Tensor[],\n    outLabels?: string[], batchSize?: number, epochs?: number, verbose?: number,\n    callbacks?: BaseCallback[], valF?: (data: Tensor[]) => Scalar[],\n    valIns?: Tensor[], shuffle?: boolean|string, callbackMetrics?: string[],\n    initialEpoch?: number, stepsPerEpoch?: number,\n    validationSteps?: number): Promise<History> {\n  if (batchSize == null) {\n    batchSize = 32;\n  }\n  if (epochs == null) {\n    epochs = 1;\n  }\n  if (shuffle == null) {\n    shuffle = true;\n  }\n  if (initialEpoch == null) {\n    initialEpoch = 0;\n  }\n\n  // TODO(cais): Change const to let below when implementing validation.\n  let doValidation = false;\n  if (valF != null && valIns != null) {\n    doValidation = true;\n    // TODO(cais): verbose message.\n  }\n  if (validationSteps != null) {\n    doValidation = true;\n    if (stepsPerEpoch == null) {\n      throw new ValueError(\n          'Can only use `validationSteps` when doing step-wise training, ' +\n          'i.e., `stepsPerEpoch` must be set.');\n    }\n  }\n\n  const numTrainSamples =\n      model.checkNumSamples(ins, batchSize, stepsPerEpoch, 'steps_per_epoch');\n  let indexArray: number[];\n  if (numTrainSamples != null) {\n    indexArray = range(0, numTrainSamples);\n  }\n\n  if (verbose == null) {\n    verbose = 1;\n  }\n\n  const {callbackList, history} = configureCallbacks(\n      callbacks, verbose, epochs, initialEpoch, numTrainSamples, stepsPerEpoch,\n      batchSize, doValidation, callbackMetrics);\n  callbackList.setModel(model);\n  model.history = history;\n  await callbackList.onTrainBegin();\n  model.stopTraining_ = false;\n  // TODO(cais): Take care of callbacks.validation_data as in PyKeras.\n  // TODO(cais): Pre-convert feeds for performance as in PyKeras.\n\n  for (let epoch = initialEpoch; epoch < epochs; ++epoch) {\n    await callbackList.onEpochBegin(epoch);\n    const epochLogs: UnresolvedLogs = {};\n    if (stepsPerEpoch != null) {\n      throw new NotImplementedError(\n          'stepsPerEpoch mode is not implemented yet.');\n    } else {\n      if (shuffle === 'batch') {\n        throw new NotImplementedError('batch shuffling is not implemneted yet');\n      } else if (shuffle) {\n        util.shuffle(indexArray);\n      }\n      // Convert the potentially shuffled indices to Tensor1D, to avoid the\n      // cost of repeated creation of Array1Ds later on.\n      const epochIndexArray1D = tensor1d(indexArray);\n\n      const batches = makeBatches(numTrainSamples, batchSize);\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchLogs: UnresolvedLogs = {};\n        await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n        tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds = sliceAlongFirstAxis(\n                               epochIndexArray1D, batchStart,\n                               batchEnd - batchStart) as Tensor1D;\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = batchEnd - batchStart;\n\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Tensor[];\n          const outs = f(insBatch);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n            // TODO(cais): Use scope() to avoid ownership.\n          }\n\n          if (batchIndex === batches.length - 1) {  // Last batch.\n            if (doValidation) {\n              const valOuts = model.testLoop(valF, valIns, batchSize);\n              // Porting Notes: In tfjs-layers, valOuts is always an Array.\n              for (let i = 0; i < outLabels.length; ++i) {\n                const label = outLabels[i];\n                const out = valOuts[i];\n                tfc.keep(out);\n                // TODO(cais): Use scope() to avoid ownership.\n                epochLogs['val_' + label] = out;\n              }\n            }\n          }\n        });\n\n        await callbackList.onBatchEnd(batchIndex, batchLogs);\n        disposeTensorsInLogs(batchLogs);\n\n        if (model.stopTraining_) {\n          break;\n        }\n        // TODO(cais): return outs as list of Tensor.\n      }\n\n      epochIndexArray1D.dispose();\n    }\n    // TODO(cais): Run validation at the end of the epoch.\n    await callbackList.onEpochEnd(epoch, epochLogs);\n    if (model.stopTraining_) {\n      break;\n    }\n  }\n  await callbackList.onTrainEnd();\n\n  await model.history.syncData();\n  return model.history;\n}\n\nexport async function fitTensors(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    args: ModelFitArgs = {}): Promise<History> {\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n  let inputs: Tensor[];\n  let targets: Tensor[];\n  let inputValX: Tensor|Tensor[];\n  let inputValY: Tensor|Tensor[];\n  let valX: Tensor|Tensor[];\n  let valY: Tensor|Tensor[];\n  let sampleWeights: Tensor[];\n  try {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // Validate user data.\n    // TODO(cais): Support sampleWeight.\n    const checkBatchAxis = false;\n    const standardizedOuts =\n        await model.standardizeUserData(\n            x, y, args.sampleWeight, args.classWeight, checkBatchAxis,\n            batchSize) as [Tensor[], Tensor[], Tensor[]];\n    inputs = standardizedOuts[0];\n    targets = standardizedOuts[1];\n    sampleWeights = standardizedOuts[2];\n\n    // Prepare validation data.\n    let doValidation = false;\n    let valIns: Tensor[];\n    if (args.validationData != null && args.validationData.length > 0) {\n      doValidation = true;\n      if (args.validationData.length === 2) {\n        // config.validationData consists of valX and valY.\n        inputValX = args.validationData[0];\n        inputValY = args.validationData[1];\n      } else if (args.validationData.length === 3) {\n        throw new NotImplementedError(\n            'validationData including sample weights is not supported yet.');\n      } else {\n        throw new ValueError(\n            `When passing validation data, it must contain 2 (valX, valY) ` +\n            `or 3 (valX, valY, valSampleWeight) items; ` +\n            `${args.validationData} is invalid.`);\n      }\n\n      const checkBatchAxis = true;\n      const valStandardized =\n          await model.standardizeUserData(\n              inputValX, inputValY, null, /** Unused sample weights. */\n              null,                       /** Unused class weights. */\n              checkBatchAxis, batchSize) as [Tensor[], Tensor[], Tensor[]];\n      valX = valStandardized[0];\n      valY = valStandardized[1];\n      valIns = valX.concat(valY);\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (\n        args.validationSplit != null && args.validationSplit > 0 &&\n        args.validationSplit < 1) {\n      doValidation = true;\n      // Porting Note: In tfjs-layers, inputs[0] is always a Tensor.\n      const splitAt =\n          Math.floor(inputs[0].shape[0] * (1 - args.validationSplit));\n      const originalBatchSize = inputs[0].shape[0];\n      valX = sliceArrays(inputs, splitAt, originalBatchSize) as Tensor[];\n      inputs = sliceArrays(inputs, 0, splitAt) as Tensor[];\n      valY = sliceArrays(targets, splitAt, originalBatchSize) as Tensor[];\n      targets = sliceArrays(targets, 0, splitAt) as Tensor[];\n      // TODO(cais): Once sampleWeights becomes available, slice it to get\n      //   valSampleWeights.\n      valIns = valX.concat(valY);\n\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (args.validationSteps != null) {\n      doValidation = true;\n      // TODO(cais): Add useLearningPhase.\n    }\n\n    const ins = inputs.concat(targets).concat(sampleWeights);\n\n    model.checkTrainableWeightsConsistency();\n\n    // TODO(cais): Handle use_learning_phase and learning_phase?\n\n    // Porting Note: Here we see a key deviation of tfjs-layers from\n    // Keras.\n    //  Due to the imperative nature of tfjs-layers' backend (tfjs-core),\n    //  we do not construct symbolic computation graphs to embody the\n    //  training process. Instead, we define a function that performs the\n    //  training action. In PyKeras, the data (inputs and targets) are fed\n    //  through graph placeholders. In tfjs-layers, the data are fed as\n    //  function arguments. Since the function are defined below in the\n    //  scope, we don't have equivalents of PyKeras's\n    //  `_make_train_funciton`.\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let valFunction: (data: Tensor[]) => Scalar[];\n    let callbackMetrics: string[];\n    if (doValidation) {\n      model.makeTestFunction();\n      valFunction = model.testFunction;\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      valFunction = null;\n      valIns = [];\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const out = await fitLoop(\n        model, trainFunction, ins, outLabels, batchSize, args.epochs,\n        args.verbose, callbacks, valFunction, valIns, args.shuffle,\n        callbackMetrics, args.initialEpoch, null, null);\n    return out;\n  } finally {\n    model.isTraining = false;\n    // Memory clean up.\n    disposeNewTensors(inputs, x);\n    disposeNewTensors(targets, y);\n    disposeNewTensors(valX as Tensor[], inputValX);\n    disposeNewTensors(valY as Tensor[], inputValY);\n    if (sampleWeights != null) {\n      tfc.dispose(sampleWeights);\n    }\n  }\n  // TODO(cais): Add value to outLabels.\n}\n\n/**\n * Ensure tensors all have a rank of at least 2.\n *\n * If a tensor has a rank of 1, it is dimension-expanded to rank 2.\n * If any tensor has a rank of 0 (i.e., is a scalar), an error will be thrown.\n */\nexport function ensureTensorsRank2OrHigher(tensors: Tensor|Tensor[]): Tensor[] {\n  const outs: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    tensors = [tensors];\n  }\n\n  // Make Tensors at least 2D.\n  for (let i = 0; i < tensors.length; ++i) {\n    const tensor = tensors[i];\n    if (tensor.rank === 1) {\n      outs.push(expandDims(tensor, 1));\n    } else if (tensor.rank === 0) {\n      throw new Error(\n          'Expected tensor to be at least 1D, but received a 0D tensor ' +\n          '(scalar).');\n    } else {\n      outs.push(tensor);\n    }\n  }\n  return outs;\n}\n\n/**\n * Compare a set of tensors with a reference (old) set, discard the ones\n * in the new set that are not present in the reference set.\n *\n * This method is used for memory clenaup during calls such as\n * LayersModel.fit().\n *\n * @param tensors New set which may contain Tensors not present in\n *   `refTensors`.\n * @param refTensors Reference Tensor set.\n */\n// TODO(cais, kangyizhang): Deduplicate with tfjs-data.\nexport function disposeNewTensors(\n    tensors: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    refTensors: Tensor|Tensor[]|{[inputName: string]: Tensor}): void {\n  if (tensors == null) {\n    return;\n  }\n  const oldTensorIds: number[] = [];\n  if (refTensors instanceof Tensor) {\n    oldTensorIds.push(refTensors.id);\n  } else if (Array.isArray(refTensors)) {\n    refTensors.forEach(t => oldTensorIds.push(t.id));\n  } else if (refTensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in refTensors) {\n      const oldTensor = refTensors[name];\n      oldTensorIds.push(oldTensor.id);\n    }\n  }\n\n  const tensorsToDispose: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    if (oldTensorIds.indexOf(tensors.id) === -1) {\n      tensorsToDispose.push(tensors);\n    }\n  } else if (Array.isArray(tensors)) {\n    tensors.forEach(t => {\n      if (oldTensorIds.indexOf(t.id) === -1) {\n        tensorsToDispose.push(t);\n      }\n    });\n  } else if (tensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in tensors) {\n      const tensor = tensors[name];\n      if (oldTensorIds.indexOf(tensor.id) === -1) {\n        tensorsToDispose.push(tensor);\n      }\n    }\n  }\n\n  tensorsToDispose.forEach(t => {\n    if (!t.isDisposed) {\n      t.dispose();\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: engine/training.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {io, ModelPredictConfig as ModelPredictArgs, NamedTensorMap, Optimizer, Scalar, scalar, serialization, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {History, ModelLoggingVerbosity} from '../base_callbacks';\nimport {nameScope} from '../common';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {LossIdentifier} from '../keras_format/loss_config';\nimport {OptimizerSerialization} from '../keras_format/optimizer_config';\nimport {MetricsIdentifier, TrainingConfig} from '../keras_format/training_config';\nimport {deserialize} from '../layers/serialization';\nimport * as losses from '../losses';\nimport * as Metrics from '../metrics';\nimport * as optimizers from '../optimizers';\nimport {LossOrMetricFn, NamedTensor} from '../types';\nimport {checkUserDefinedMetadata} from '../user_defined_metadata';\nimport {count, pyListRepeat, singletonOrArray, toCamelCase, toSnakeCase, unique} from '../utils/generic_utils';\nimport {printSummary} from '../utils/layer_utils';\nimport {range} from '../utils/math_utils';\nimport {convertPythonicToTs} from '../utils/serialization_utils';\nimport {LayerVariable} from '../variables';\nimport {version} from '../version';\n\nimport {Container, ContainerArgs} from './container';\nimport {Dataset} from './dataset_stub';\nimport {execute, FeedDict} from './executor';\nimport {DisposeResult, SymbolicTensor} from './topology';\nimport {evaluateDataset, fitDataset, ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './training_dataset';\nimport {checkBatchSize, disposeNewTensors, ensureTensorsRank2OrHigher, fitTensors, makeBatches, ModelFitArgs, sliceArrays, sliceArraysByIndices} from './training_tensors';\nimport {ClassWeight, ClassWeightMap, computeWeightedLoss, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Helper function for polymorphic input data: 1. singleton Tensor.\n */\nexport function isDataTensor(x: Tensor|Tensor[]|{[inputName: string]: Tensor}|\n                             {[inputName: string]: Tensor[]}): boolean {\n  return x instanceof Tensor;\n}\n\n/**\n * Helper function for polymorphic input data: 2. Array of Tensor.\n */\nexport function isDataArray(x: Tensor|Tensor[]|\n                            {[inputName: string]: Tensor}): boolean {\n  return Array.isArray(x);\n}\n\n/**\n * Helper function for polymorphic input data: 3. \"dict\" of Tensor.\n */\nexport function isDataDict(x: Tensor|Tensor[]|\n                           {[inputName: string]: Tensor}): boolean {\n  return !isDataTensor(x) && !isDataArray(x);\n}\n\n/**\n * Normalizes inputs and targets provided by users.\n * @param data User-provided input data (polymorphic).\n * @param names An Array of expected Tensor names.\n * @param shapes Optional Array of expected Tensor shapes.\n * @param checkBatchAxis Whether to check that the batch axis of the arrays\n *   match  the expected value found in `shapes`.\n * @param exceptionPrefix String prefix used for exception formatting.\n * @returns List of standardized input Tensors (one Tensor per model input).\n * @throws ValueError: in case of improperly formatted user data.\n */\nexport function standardizeInputData(\n    data: Tensor|Tensor[]|{[inputName: string]: Tensor}, names: string[],\n    shapes?: Shape[], checkBatchAxis = true, exceptionPrefix = ''): Tensor[] {\n  if (names == null || names.length === 0) {\n    // Check for the case where the model expected no data, but some data got\n    // sent.\n    if (data != null) {\n      let gotUnexpectedData = false;\n      if (isDataArray(data) && (data as Tensor[]).length > 0) {\n        gotUnexpectedData = true;\n      } else if (isDataDict(data)) {\n        for (const key in data) {\n          if (data.hasOwnProperty(key)) {\n            gotUnexpectedData = true;\n            break;\n          }\n        }\n      } else {\n        // `data` is a singleton Tensor in this case.\n        gotUnexpectedData = true;\n      }\n      if (gotUnexpectedData) {\n        throw new ValueError(\n            `Error when checking model ${exceptionPrefix} expected no data, ` +\n            `but got ${data}`);\n      }\n    }\n    return [];\n  }\n  if (data == null) {\n    return names.map(name => null);\n  }\n\n  let arrays: Tensor[];\n  if (isDataDict(data)) {\n    data = data as {[inputName: string]: Tensor};\n    arrays = [];\n    for (const name of names) {\n      if (data[name] == null) {\n        throw new ValueError(\n            `No data provided for \"${name}\". Need data for each key in: ` +\n            `${names}`);\n      }\n      arrays.push(data[name]);\n    }\n  } else if (isDataArray(data)) {\n    data = data as Tensor[];\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `model expected. Expected to see ${names.length} Tensor(s), but ` +\n          `instead got the following list of Tensor(s): ${data}`);\n    }\n    arrays = data;\n  } else {\n    data = data as Tensor;\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model ${exceptionPrefix} expects ${names.length} Tensor(s), ` +\n          `but only received one Tensor. Found: Tensor with shape ${\n              data.shape}`);\n    }\n    arrays = [data];\n  }\n\n  arrays = ensureTensorsRank2OrHigher(arrays);\n\n  // Check shape compatibility.\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s). but got array with ` +\n            `shape ${array.shape}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          // Skip the first (batch) axis.\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null && refDim >= 0 && dim !== refDim) {\n          throw new ValueError(\n              `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n              `to have shape [${shapes[i]}], but got array with shape ` +\n              `[${array.shape}].`);\n        }\n      }\n    }\n  }\n  return arrays;\n}\n\n/**\n * User input validation for Tensors.\n * @param inputs `Array` of `tf.Tensor`s for inputs.\n * @param targets `Array` of `tf.Tensor`s for targets.\n * @param weights Optional `Array` of `tf.Tensor`s for sample weights.\n * @throws ValueError: in case of incorrectly formatted data.\n */\nexport function checkArrayLengths(\n    inputs: Tensor[], targets: Tensor[], weights?: Tensor[]) {\n  const setX = unique(inputs.map(input => input.shape[0]));\n  setX.sort();\n  const setY = unique(targets.map(target => target.shape[0]));\n  setY.sort();\n  // TODO(cais): Check `weights` as well.\n  if (setX.length > 1) {\n    throw new ValueError(\n        `All input Tensors (x) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(inputs.map(input => input.shape))}`);\n  }\n  if (setY.length > 1) {\n    throw new ValueError(\n        `All target Tensors (y) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(targets.map(target => target.shape))}`);\n  }\n  if (setX.length > 0 && setY.length > 0 && !util.arraysEqual(setX, setY)) {\n    throw new ValueError(\n        `Input Tensors should have the same number of samples as target ` +\n        `Tensors. Found ${setX[0]} input sample(s) and ${setY[0]} target ` +\n        `sample(s).`);\n  }\n}\n\n/**\n * Validation on the compatibility of targes and loss functions.\n *\n * This helps prevent users from using loss functions incorrectly.\n *\n * @param targets `Array` of `tf.Tensor`s of targets.\n * @param lossFns `Array` of loss functions.\n * @param outputShapes `Array` of shapes of model outputs.\n */\nfunction checkLossAndTargetCompatibility(\n    targets: Tensor[], lossFns: LossOrMetricFn[], outputShapes: Shape[]) {\n  // TODO(cais): Dedicated test coverage?\n  const keyLosses = [\n    losses.meanSquaredError, losses.binaryCrossentropy,\n    losses.categoricalCrossentropy\n  ];\n  for (let i = 0; i < targets.length; ++i) {\n    const y = targets[i];\n    const loss = lossFns[i];\n    const shape = outputShapes[i];\n    if (loss == null) {\n      continue;\n    }\n    if (loss === losses.categoricalCrossentropy) {\n      if (y.shape[y.shape.length - 1] === 1) {\n        throw new ValueError(\n            `You are passing a target array of shape ${y.shape} while using ` +\n            `a loss 'categorical_crossentropy'. 'categorical_crossentropy'` +\n            `expects targets to be binary matrices (1s and 0s) of shape ` +\n            `[samples, classes].`);\n        // TODO(cais): Example code in error message.\n      }\n    }\n    if (keyLosses.indexOf(loss) !== -1) {\n      const slicedYShape = y.shape.slice(1);\n      const slicedShape = shape.slice(1);\n      for (let j = 0; j < slicedYShape.length; ++j) {\n        const targetDim = slicedYShape[j];\n        const outDim = slicedShape[j];\n        if (outDim != null && targetDim !== outDim) {\n          throw new ValueError(\n              `A target Tensor with shape ${y.shape} was passed for an ` +\n              `output of shape ${shape}, while using a loss function that ` +\n              `expects targets to have the same shape as the output.`);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Check inputs provided by the user.\n *\n * Porting Note: This corresponds to _standardize_input_data() in Python\n *   Keras. Because of the strong typing in TF.js, we do not need to convert\n *   the data. Specifically:\n *   1) in PyKeras, `data` can be `DataFrame` instances from pandas, for\n *      example. We don't need to worry about that here because there is no\n *      widely popular javascript/typesdcript equivalent of pandas (so far).\n *      If one becomes available in the future, we can add support.\n *   2) in PyKeras, inputs can be Python dict. But here we are stipulating\n * that the data is either a single `tf.Tensor` or an Array of `tf.Tensor`s. We\n * may add support for `Object` data inputs in the future when the need\n * arises.\n *\n * Instead, we perform basic checks for number of parameters and shapes.\n *\n * @param data: The input data.\n * @param names: Name for the inputs, from the model.\n * @param shapes: Expected shapes for the input data, from the model.\n * @param checkBatchAxis: Whether the size along the batch axis (i.e., the\n *   first dimension) will be checked for matching.\n * @param exceptionPrefix: Execption prefix message, used in generating error\n *   messages.\n * @throws ValueError: on incorrect number of inputs or mismatches in shapes.\n */\nfunction checkInputData(\n    data: Tensor|Tensor[], names: string[], shapes?: Shape[],\n    checkBatchAxis = true, exceptionPrefix = '') {\n  let arrays: Tensor[];\n  if (Array.isArray(data)) {\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `the model expected. Expected to see ${names.length} Tensor(s),` +\n          ` but instead got ${data.length} Tensors(s).`);\n    }\n    arrays = data;\n  } else {\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model expects ${names.length} ${exceptionPrefix} Tensors, ` +\n          `but only received one Tensor. Found: array with shape ` +\n          `${JSON.stringify(data.shape)}.`);\n    }\n    arrays = [data];\n  }\n\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s), but got array with ` +\n            `shape ${JSON.stringify(array.shape)}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null) {\n          if (refDim !== dim) {\n            throw new ValueError(\n                `Error when checking ${exceptionPrefix}: expected ` +\n                `${names[i]} to have shape ${JSON.stringify(shapes[i])} but ` +\n                `got array with shape ${JSON.stringify(array.shape)}.`);\n          }\n        }\n      }\n    }\n  }\n}\n\n/**\n * Maps metric functions to model outputs.\n * @param metrics An shortcut strings name, metric function, `Array` or dict\n *   (`Object`) of metric functions.\n * @param outputNames An `Array` of the names of model outputs.\n * @returns An `Array` (one entry per model output) of `Array` of metric\n *   functions. For instance, if the model has 2 outputs, and for the first\n *   output we want to compute `binaryAccuracy` and `binaryCrossentropy`,\n *   and just `binaryAccuracy` for the second output, the `Array` would look\n *   like:\n *     `[[binaryAccuracy, binaryCrossentropy],  [binaryAccuracy]]`\n * @throws TypeError: incompatible metrics format.\n */\nexport function collectMetrics(\n    metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n    {[outputName: string]: string | LossOrMetricFn},\n    outputNames: string[]): Array<Array<string|LossOrMetricFn>> {\n  if (metrics == null || Array.isArray(metrics) && metrics.length === 0) {\n    return outputNames.map(name => []);\n  }\n\n  let wrappedMetrics: Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  if (typeof metrics === 'string' || typeof metrics === 'function') {\n    wrappedMetrics = [metrics];\n  } else if (Array.isArray(metrics) || typeof metrics === 'object') {\n    wrappedMetrics = metrics as Array<string|LossOrMetricFn>|\n        {[outputName: string]: string} | {[outputName: string]: LossOrMetricFn};\n  } else {\n    throw new TypeError(\n        'Type of metrics argument not understood. Expected an string,' +\n        `function, Array, or Object, found: ${metrics}`);\n  }\n\n  if (Array.isArray(wrappedMetrics)) {\n    // We then apply all metrics to all outputs.\n    return outputNames.map(\n        name => wrappedMetrics as Array<string|LossOrMetricFn>);\n  } else {\n    // In this case, metrics is a dict.\n    const nestedMetrics: Array<Array<string|LossOrMetricFn>> = [];\n    for (const name of outputNames) {\n      let outputMetrics: string|LossOrMetricFn|Array<string|LossOrMetricFn> =\n          wrappedMetrics.hasOwnProperty(name) ? wrappedMetrics[name] : [];\n      if (!Array.isArray(outputMetrics)) {\n        outputMetrics = [outputMetrics];\n      }\n      nestedMetrics.push(outputMetrics);\n    }\n    return nestedMetrics;\n  }\n}\n\nexport interface ModelEvaluateArgs {\n  /**\n   * Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * Tensor of weights to weight the contribution of different samples to the\n   * loss and metrics.\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * integer: total number of steps (batches of samples)\n   * before declaring the evaluation round finished. Ignored with the default\n   * value of `undefined`.\n   */\n  steps?: number;\n}\n\n/**\n * Configuration for calls to `LayersModel.compile()`.\n */\nexport interface ModelCompileArgs {\n  /**\n   * An instance of `tf.train.Optimizer` or a string name for an Optimizer.\n   */\n  optimizer: string|Optimizer;\n\n  /**\n   * Object function(s) or name(s) of object function(s).\n   * If the model has multiple outputs, you can use a different loss\n   * on each output by passing a dictionary or an Array of losses.\n   * The loss value that will be minimized by the model will then be the sum\n   * of all individual losses.\n   */\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n\n  /**\n   * List of metrics to be evaluated by the model during training and testing.\n   * Typically you will use `metrics=['accuracy']`.\n   * To specify different metrics for different outputs of a multi-output\n   * model, you could also pass a dictionary.\n   */\n  metrics?: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n\n  // TODO(cais): Add lossWeights, sampleWeightMode, weightedMetrics, and\n  //   targetTensors.\n}\n\nconst LAYERS_MODEL_FORMAT_NAME = 'layers-model';\n\n/**\n * A `tf.LayersModel` is a directed, acyclic graph of `tf.Layer`s plus methods\n * for training, evaluation, prediction and saving.\n *\n * `tf.LayersModel` is the basic unit of training, inference and evaluation in\n * TensorFlow.js. To create a `tf.LayersModel`, use `tf.LayersModel`.\n *\n * See also:\n *   `tf.Sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class LayersModel extends Container implements tfc.InferenceModel {\n  // The class name is 'Model' rather than 'LayersModel' for backwards\n  // compatibility since this class name shows up in the serialization format.\n  /** @nocollapse */\n  static className = 'Model';\n  protected optimizer_: Optimizer;\n  // Whether the model instance owns the optimizer: `true` if and only if\n  // `optimizer` is created from a string parameter during `compile()` call.\n  protected isOptimizerOwned: boolean;\n\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n  lossFunctions: LossOrMetricFn[];\n\n  // TODO(cais): These private variables should probably not have the string\n  //   'feed' in their names, because we are not dealing with a symbolic\n  //   backend.\n  private feedOutputShapes: Shape[];\n  private feedLossFns: LossOrMetricFn[];\n  private collectedTrainableWeights: LayerVariable[];\n  private testFunction: (data: Tensor[]) => Scalar[];\n  history: History;\n\n  // A public property that can be set by Callbacks to order early stopping\n  // during `fit()` calls.\n  protected stopTraining_: boolean;\n  protected isTraining: boolean;\n\n  metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  metricsNames: string[];\n  // Porting Note: `metrics_tensors` in PyKeras is a symbolic tensor. But given\n  //   the imperative nature of tfjs-core, `metricsTensors` is a\n  //   TypeScript function here.\n  //   Also note that due to the imperative nature of tfjs-core, `metricsTensor`\n  //   here needs an output index to keep track of which output of the\n  //   LayersModel a metric belongs to. This is unlike `metrics_tensors` in\n  //   PyKeras, which is a `list` of symbolic tensors, each of which has\n  //   implicit \"knowledge\" of the outputs it depends on.\n  metricsTensors: Array<[LossOrMetricFn, number]>;\n\n  // User defind metadata (if any).\n  private userDefinedMetadata: {};\n\n  constructor(args: ContainerArgs) {\n    super(args);\n    this.isTraining = false;\n  }\n\n  /**\n   * Print a text summary of the model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - If the model has non-sequential-like topology, the inputs each layer\n   *   receives\n   * - The total number of trainable and non-trainable parameters of the model.\n   *\n   * ```js\n   * const input1 = tf.input({shape: [10]});\n   * const input2 = tf.input({shape: [20]});\n   * const dense1 = tf.layers.dense({units: 4}).apply(input1);\n   * const dense2 = tf.layers.dense({units: 8}).apply(input2);\n   * const concat = tf.layers.concatenate().apply([dense1, dense2]);\n   * const output =\n   *     tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat);\n   *\n   * const model = tf.model({inputs: [input1, input2], outputs: output});\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      throw new ValueError(\n          `This model has never been called, thus its weights have not been ` +\n          `created yet. So no summary can be displayed. Build the model ` +\n          `first (e.g., by calling it on some test data).`);\n    }\n    printSummary(this, lineLength, positions, printFn);\n  }\n\n  /**\n   * Configures and prepares the model for training and evaluation.  Compiling\n   * outfits the model with an optimizer, loss, and/or metrics.  Calling `fit`\n   * or `evaluate` on an un-compiled model will throw an error.\n   *\n   * @param args a `ModelCompileArgs` specifying the loss, optimizer, and\n   * metrics to be used for fitting and evaluating this model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  compile(args: ModelCompileArgs): void {\n    if (args.loss == null) {\n      args.loss = [];\n    }\n    this.loss = args.loss;\n\n    if (typeof args.optimizer === 'string') {\n      this.optimizer_ = optimizers.getOptimizer(args.optimizer);\n      this.isOptimizerOwned = true;\n    } else {\n      if (!(args.optimizer instanceof Optimizer)) {\n        throw new ValueError(\n            `User-defined optimizer must be an instance of tf.Optimizer.`);\n      }\n      this.optimizer_ = args.optimizer;\n      this.isOptimizerOwned = false;\n    }\n\n    // TODO(cais): Add lossWeights.\n    // TODO(cais): Add sampleWeightMode.\n\n    // Prepare loss functions.\n    let lossFunctions: LossOrMetricFn[] = [];\n    if (!Array.isArray(args.loss) && typeof args.loss !== 'string' &&\n        typeof args.loss !== 'function') {\n      args.loss = args.loss as {[outputName: string]: string};\n      for (const name in args.loss) {\n        if (this.outputNames.indexOf(name) === -1) {\n          throw new ValueError(\n              `Unknown entry in loss dictionary: \"${name}\". ` +\n              `Only expected the following keys: ${this.outputNames}`);\n        }\n      }\n      for (const name of this.outputNames) {\n        if (args.loss[name] == null) {\n          console.warn(\n              `Output \"${name}\" is missing from loss dictionary. We assume ` +\n              `this was done on purpose, and we will not be expecting data ` +\n              `to be passed to ${name} during training`);\n        }\n        lossFunctions.push(losses.get(args.loss[name]));\n      }\n    } else if (Array.isArray(args.loss)) {\n      if (args.loss.length !== this.outputs.length) {\n        throw new ValueError(\n            `When passing an Array as loss, it should have one entry per ` +\n            `model output. The model has ${this.outputs.length} output(s), ` +\n            `but you passed loss=${args.loss}.`);\n      }\n      const theLosses = args.loss as Array<string|LossOrMetricFn>;\n      lossFunctions = theLosses.map(l => losses.get(l));\n    } else {\n      const lossFunction = losses.get(args.loss);\n      this.outputs.forEach(_ => {\n        lossFunctions.push(lossFunction);\n      });\n    }\n\n    this.lossFunctions = lossFunctions;\n\n    this.feedOutputNames = [];\n    this.feedOutputShapes = [];\n    this.feedLossFns = [];\n    for (let i = 0; i < this.outputs.length; ++i) {\n      // TODO(cais): Logic for skipping target(s).\n      const shape = this.internalOutputShapes[i];\n      const name = this.outputNames[i];\n      this.feedOutputNames.push(name);\n      this.feedOutputShapes.push(shape);\n      this.feedLossFns.push(this.lossFunctions[i]);\n    }\n\n    // TODO(cais): Add logic for output masks.\n    // TODO(cais): Add logic for sample weights.\n    const skipTargetIndices: number[] = [];\n\n    // Prepare metrics.\n    this.metrics = args.metrics;\n    // TODO(cais): Add weightedMetrics.\n    this.metricsNames = ['loss'];\n    this.metricsTensors = [];\n\n    // Compute total loss.\n    // Porting Note: In PyKeras, metrics_tensors are symbolic tensor objects.\n    //   Here, metricsTensors are TypeScript functions. This difference is due\n    //   to the difference in symbolic/imperative property of the backends.\n    nameScope('loss', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        // TODO(cais): Add weightedLoss, sampleWeight and mask.\n        //   The following line should be weightedLoss\n        const weightedLoss = this.lossFunctions[i];\n        if (this.outputs.length > 1) {\n          this.metricsTensors.push([weightedLoss, i]);\n          this.metricsNames.push(this.outputNames[i] + '_loss');\n        }\n      }\n\n      // Porting Note: Due to the imperative nature of the backend, we calculate\n      //   the regularizer penalties in the totalLossFunction, instead of here.\n    });\n\n    const nestedMetrics = collectMetrics(args.metrics, this.outputNames);\n    // TODO(cais): Add nestedWeightedMetrics.\n\n    /**\n     * Helper function used in loop below.\n     */\n    const appendMetric =\n        (outputIndex: number, metricName: string,\n         metricTensor: LossOrMetricFn) => {\n          if (this.outputNames.length > 1) {\n            metricName = this.outputNames[outputIndex] + '_' + metricName;\n          }\n          this.metricsNames.push(metricName);\n          this.metricsTensors.push([metricTensor, outputIndex]);\n        };\n\n    nameScope('metric', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        const outputMetrics = nestedMetrics[i];\n        // TODO(cais): Add weights and outputWeightedMetrics.\n\n        // TODO(cais): Add optional arg `weights` to the following function.\n        const handleMetrics = (metrics: Array<string|LossOrMetricFn>) => {\n          const metricNamePrefix = '';\n          let metricName: string;\n          let accFn: LossOrMetricFn;\n          let weightedMetricFn: LossOrMetricFn;\n          //  TODO(cais): Use 'weights_' for weighted metrics.\n\n          for (const metric of metrics) {\n            if (typeof metric === 'string' &&\n                ['accuracy', 'acc', 'crossentropy', 'ce'].indexOf(metric) !==\n                    -1) {\n              const outputShape = this.internalOutputShapes[i];\n\n              if (outputShape[outputShape.length - 1] === 1 ||\n                  this.lossFunctions[i] === losses.binaryCrossentropy) {\n                // case: binary accuracy/crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryCrossentropy;\n                }\n              } else if (\n                  this.lossFunctions[i] ===\n                  losses.sparseCategoricalCrossentropy) {\n                // case: categorical accuracy / crossentropy with sparse\n                // targets.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalCrossentropy;\n                }\n              } else {\n                // case: categorical accuracy / crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalCrossentropy;\n                }\n              }\n              let suffix: string;\n              if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                suffix = 'acc';\n              } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                suffix = 'ce';\n              }\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = accFn;\n              metricName = metricNamePrefix + suffix;\n            } else {\n              const metricFn = Metrics.get(metric);\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = metricFn;\n              metricName =\n                  metricNamePrefix + Metrics.getLossOrMetricName(metric);\n            }\n\n            // TODO(cais): Add weighting and masking to metricResult.\n            let metricResult: LossOrMetricFn;\n            nameScope(metricName, () => {\n              metricResult = weightedMetricFn;\n            });\n            appendMetric(i, metricName, metricResult);\n          }\n        };\n\n        handleMetrics(outputMetrics);\n        // TODO(cais): Call handleMetrics with weights.\n      }\n    });\n\n    // Porting Notes: Given the imperative backend of tfjs-core,\n    //   there is no need for constructing the symbolic graph and placeholders.\n    this.collectedTrainableWeights = this.trainableWeights;\n  }\n\n  /**\n   * Check trainable weights count consistency.\n   *\n   * This will raise a warning if `this.trainableWeights` and\n   * `this.collectedTrainableWeights` are inconsistent (i.e., have different\n   * numbers of parameters).\n   * Inconsistency will typically arise when one modifies `model.trainable`\n   * without calling `model.compile()` again.\n   */\n  protected checkTrainableWeightsConsistency(): void {\n    if (this.collectedTrainableWeights == null) {\n      return;\n    }\n    if (this.trainableWeights.length !==\n        this.collectedTrainableWeights.length) {\n      console.warn(\n          'Discrepancy between trainableweights and collected trainable ' +\n          'weights. Did you set `model.trainable` without calling ' +\n          '`model.compile()` afterwards?');\n    }\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(\n   *     tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4});\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateArgs`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // TODO(cais): Standardize `config.sampleWeights` as well.\n    // Validate user data.\n    const checkBatchAxis = true;\n    const standardizedOuts =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    try {\n      // TODO(cais): If uses `useLearningPhase`, set the corresponding element\n      // of the input to 0.\n      const ins = standardizedOuts[0].concat(standardizedOuts[1]);\n      this.makeTestFunction();\n      const f = this.testFunction;\n      const testOuts =\n          this.testLoop(f, ins, batchSize, args.verbose, args.steps);\n      return singletonOrArray(testOuts);\n    } finally {\n      disposeNewTensors(standardizedOuts[0], x);\n      disposeNewTensors(standardizedOuts[1], y);\n    }\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args?: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    this.makeTestFunction();\n    return evaluateDataset(this, dataset, args);\n  }\n\n  /**\n   * Get number of samples provided for training, evaluation or prediction.\n   *\n   * @param ins Input `tf.Tensor`.\n   * @param batchSize Integer batch size, optional.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring loop finished. Optional.\n   * @param stepsName The public API's parameter name for `steps`.\n   * @returns Number of samples provided.\n   */\n  private checkNumSamples(\n      ins: Tensor|Tensor[], batchSize?: number, steps?: number,\n      stepsName = 'steps'): number {\n    let numSamples: number;\n    if (steps != null) {\n      numSamples = null;\n      if (batchSize != null) {\n        throw new ValueError(\n            `If ${stepsName} is set, batchSize must be null or undefined.` +\n            `Got batchSize = ${batchSize}`);\n      }\n    } else if (ins != null) {\n      if (Array.isArray(ins)) {\n        numSamples = ins[0].shape[0];\n      } else {\n        numSamples = ins.shape[0];\n      }\n    } else {\n      throw new ValueError(\n          `Either the input data should have a defined shape, or ` +\n          `${stepsName} shoud be specified.`);\n    }\n    return numSamples;\n  }\n\n  /**\n   * Execute internal tensors of the model with input data feed.\n   * @param inputs Input data feed. Must match the inputs of the model.\n   * @param outputs Names of the output tensors to be fetched. Must match\n   *   names of the SymbolicTensors that belong to the graph.\n   * @returns Fetched values for `outputs`.\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs: string|string[]):\n      Tensor|Tensor[] {\n    if (Array.isArray(outputs) && outputs.length === 0) {\n      throw new ValueError(\n          '`outputs` is an empty Array, which is not allowed.');\n    }\n\n    const outputsIsArray = Array.isArray(outputs);\n    const outputNames =\n        (outputsIsArray ? outputs as string[] : [outputs as string]);\n    const outputSymbolicTensors = this.retrieveSymbolicTensors(outputNames);\n\n    // Format the input into a FeedDict.\n    const feedDict = new FeedDict();\n    if (inputs instanceof Tensor) {\n      inputs = [inputs];\n    }\n    if (Array.isArray(inputs)) {\n      if (inputs.length !== this.inputs.length) {\n        throw new ValueError(\n            `The number of inputs provided (${inputs.length}) ` +\n            `does not match the number of inputs of this model ` +\n            `(${this.inputs.length}).`);\n      }\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n    } else {\n      for (const input of this.inputs) {\n        const tensorValue = inputs[input.name];\n        if (tensorValue == null) {\n          throw new ValueError(\n              `No value is provided for the model's input ${input.name}`);\n        }\n        feedDict.add(input, tensorValue);\n      }\n    }\n\n    // Run execution.\n    const executeOutputs = execute(outputSymbolicTensors, feedDict) as Tensor[];\n    return outputsIsArray ? executeOutputs : executeOutputs[0];\n  }\n\n  /**\n   * Retrieve the model's internal symbolic tensors from symbolic-tensor names.\n   */\n  private retrieveSymbolicTensors(symbolicTensorNames: string[]):\n      SymbolicTensor[] {\n    const outputSymbolicTensors: SymbolicTensor[] =\n        pyListRepeat(null, symbolicTensorNames.length);\n    let outputsRemaining = symbolicTensorNames.length;\n    for (const layer of this.layers) {\n      const layerOutputs: SymbolicTensor[] =\n          Array.isArray(layer.output) ? layer.output : [layer.output];\n      const layerOutputNames = layerOutputs.map(output => output.name);\n      for (let i = 0; i < symbolicTensorNames.length; ++i) {\n        const index = layerOutputNames.indexOf(symbolicTensorNames[i]);\n        if (index !== -1) {\n          outputSymbolicTensors[i] = layerOutputs[index];\n          outputsRemaining--;\n        }\n        if (outputsRemaining === 0) {\n          break;\n        }\n      }\n      if (outputsRemaining === 0) {\n        break;\n      }\n    }\n\n    if (outputsRemaining > 0) {\n      const remainingNames: string[] = [];\n      outputSymbolicTensors.forEach((tensor, i) => {\n        if (tensor == null) {\n          remainingNames.push(symbolicTensorNames[i]);\n        }\n      });\n      throw new ValueError(\n          `Cannot find SymbolicTensors for output name(s): ` +\n          `${JSON.stringify(remainingNames)}`);\n    }\n    return outputSymbolicTensors;\n  }\n\n  /**\n   * Helper method to loop over some data in batches.\n   *\n   * Porting Note: Not using the functional approach in the Python equivalent\n   *   due to the imperative backend.\n   * Porting Note: Does not support step mode currently.\n   *\n   * @param ins: input data\n   * @param batchSize: integer batch size.\n   * @param verbose: verbosity model\n   * @returns: Predictions as `tf.Tensor` (if a single output) or an `Array` of\n   *   `tf.Tensor` (if multipe outputs).\n   */\n  private predictLoop(ins: Tensor|Tensor[], batchSize = 32, verbose = false):\n      Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins);\n      if (verbose) {\n        throw new NotImplementedError(\n            'Verbose predictLoop() is not implemented yet.');\n      }\n\n      // Sample-based predictions.\n      // Porting Note: Tensor currently does not support sliced assignments as\n      //   in numpy, e.g., x[1:3] = y. Therefore we use concatenation while\n      //   iterating over the batches.\n\n      const batches = makeBatches(numSamples, batchSize);\n      const outsBatches: Tensor[][] = this.outputs.map(output => []);\n\n      // TODO(cais): Can the scope() be pushed down inside the for loop?\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchOuts = tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          // TODO(cais): Take care of the case of the last element is a flag for\n          //   training/test.\n          const insBatch = sliceArrays(ins, batchStart, batchEnd);\n\n          // Construct the feeds for execute();\n          const feeds = [];\n          if (Array.isArray(insBatch)) {\n            for (let i = 0; i < insBatch.length; ++i) {\n              feeds.push({key: this.inputs[i], value: insBatch[i]});\n            }\n          } else {\n            feeds.push({key: this.inputs[0], value: insBatch});\n          }\n          const feedDict = new FeedDict(feeds);\n          return execute(this.outputs, feedDict) as Tensor[];\n        });\n        batchOuts.forEach((batchOut, i) => outsBatches[i].push(batchOut));\n      }\n      return singletonOrArray(\n          outsBatches.map(batches => tfc.concat(batches, 0)));\n    });\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFlow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([8, 10]), {batchSize: 4}).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param args A `ModelPredictArgs` object containing optional fields.\n   *\n   * @return Prediction results as a `tf.Tensor`(s).\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    const xsRank2OrHigher = ensureTensorsRank2OrHigher(x);\n    checkInputData(\n        xsRank2OrHigher, this.inputNames, this.feedInputShapes, false);\n    try {\n      // TODO(cais): Take care of stateful models.\n      //   if (this.stateful) ...\n      // TODO(cais): Take care of the learning_phase boolean flag.\n      //   if (this.useLearningPhase) ...\n      const batchSize = args.batchSize == null ? 32 : args.batchSize;\n      checkBatchSize(batchSize);\n      return this.predictLoop(xsRank2OrHigher, batchSize);\n    } finally {\n      disposeNewTensors(xsRank2OrHigher, x);\n    }\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predictOnBatch(tf.ones([8, 10])).print();\n   * ```\n   * @param x: Input samples, as a Tensor (for models with exactly one\n   *   input) or an array of Tensors (for models with more than one input).\n   * @return Tensor(s) of predictions\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predictOnBatch(x: Tensor|Tensor[]): Tensor|Tensor[] {\n    checkInputData(x, this.inputNames, this.feedInputShapes, true);\n    // TODO(cais): Take care of the learning_phase boolean flag.\n    //   if (this.useLearningPhase) ...\n    const batchSize = (Array.isArray(x) ? x[0] : x).shape[0];\n    return this.predictLoop(x, batchSize);\n  }\n\n  protected standardizeUserDataXY(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor}, checkBatchAxis = true,\n      batchSize?: number): [Tensor[], Tensor[]] {\n    // TODO(cais): Add sampleWeight, classWeight\n    if (this.optimizer_ == null) {\n      throw new RuntimeError(\n          'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileArgs).');\n    }\n    const outputShapes: Shape[] = [];\n    for (let i = 0; i < this.feedOutputShapes.length; ++i) {\n      const outputShape = this.feedOutputShapes[i];\n      const lossFn = this.feedLossFns[i];\n      if (lossFn === losses.sparseCategoricalCrossentropy) {\n        outputShapes.push(\n            outputShape.slice(0, outputShape.length - 1).concat([1]));\n      } else {\n        // Porting Note: Because of strong typing `lossFn` must be a function.\n        outputShapes.push(outputShape);\n      }\n    }\n    x = standardizeInputData(\n        x, this.feedInputNames, this.feedInputShapes, false, 'input');\n    y = standardizeInputData(\n        y, this.feedOutputNames, outputShapes, false, 'target');\n    // TODO(cais): Standardize sampleWeights & classWeights.\n    checkArrayLengths(x, y, null);\n    // TODO(cais): Check sampleWeights as well.\n    checkLossAndTargetCompatibility(y, this.feedLossFns, this.feedOutputShapes);\n    if (this.stateful && batchSize != null && batchSize > 0) {\n      if (x[0].shape[0] % batchSize !== 0) {\n        throw new ValueError(\n            `In a stateful network, you should only pass inputs with a ` +\n            `number of samples that is divisible by the batch size ` +\n            `${batchSize}. Found: ${x[0].shape[0]} sample(s).`);\n      }\n    }\n    return [x, y];\n  }\n\n  protected async standardizeUserData(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      sampleWeight?: Tensor|Tensor[]|{[outputName: string]: Tensor},\n      classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap,\n      checkBatchAxis = true,\n      batchSize?: number): Promise<[Tensor[], Tensor[], Tensor[]]> {\n    const [standardXs, standardYs] =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    // TODO(cais): Handle sampleWeights.\n    if (sampleWeight != null) {\n      throw new Error('sample weight is not supported yet.');\n    }\n\n    let standardSampleWeights: Tensor[] = null;\n    if (classWeight != null) {\n      const classWeights =\n          standardizeClassWeights(classWeight, this.outputNames);\n      standardSampleWeights = [];\n      for (let i = 0; i < classWeights.length; ++i) {\n        standardSampleWeights.push(\n            await standardizeWeights(standardYs[i], null, classWeights[i]));\n      }\n    }\n\n    // TODO(cais): Deal with the case of model.stateful == true.\n    return [standardXs, standardYs, standardSampleWeights];\n  }\n\n  /**\n   * Loop over some test data in batches.\n   * @param f A Function returning a list of tensors.\n   * @param ins Array of tensors to be fed to `f`.\n   * @param batchSize Integer batch size or `null` / `undefined`.\n   * @param verbose verbosity mode.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring test finished. Ignored with the default value of `null` /\n   * `undefined`.\n   * @returns Array of Scalars.\n   */\n  private testLoop(\n      f: (data: Tensor[]) => Scalar[], ins: Tensor[], batchSize?: number,\n      verbose = 0, steps?: number): Scalar[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins, batchSize, steps, 'steps');\n      const outs: Scalar[] = [];\n      if (verbose > 0) {\n        throw new NotImplementedError('Verbose mode is not implemented yet.');\n      }\n      // TODO(cais): Use `indicesForConversionToDense' to prevent slow down.\n      if (steps != null) {\n        throw new NotImplementedError(\n            'steps mode in testLoop() is not implemented yet');\n      } else {\n        const batches = makeBatches(numSamples, batchSize);\n        const indexArray = tensor1d(range(0, numSamples));\n        for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds =\n              K.sliceAlongFirstAxis(\n                  indexArray, batchStart, batchEnd - batchStart) as Tensor1D;\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Scalar[];\n          const batchOuts = f(insBatch);\n          if (batchIndex === 0) {\n            for (let i = 0; i < batchOuts.length; ++i) {\n              outs.push(scalar(0));\n            }\n          }\n          for (let i = 0; i < batchOuts.length; ++i) {\n            const batchOut = batchOuts[i];\n            outs[i] =\n                tfc.add(outs[i], tfc.mul(batchEnd - batchStart, batchOut));\n          }\n        }\n        for (let i = 0; i < outs.length; ++i) {\n          outs[i] = tfc.div(outs[i], numSamples);\n        }\n      }\n      return outs;\n    });\n  }\n\n  protected getDedupedMetricsNames(): string[] {\n    const outLabels = this.metricsNames;\n    // Rename duplicated metrics names (can happen with an output layer\n    // shared among multiple dataflows).\n    const dedupedOutLabels = [];\n    for (let i = 0; i < outLabels.length; ++i) {\n      const label = outLabels[i];\n      let newLabel = label;\n      if (count(outLabels, label) > 1) {\n        const dupIndex = count(outLabels.slice(0, i), label);\n        newLabel += `_${dupIndex}`;\n      }\n      dedupedOutLabels.push(newLabel);\n    }\n    return dedupedOutLabels;\n  }\n\n  /**\n   * Creates a function that performs the following actions:\n   *\n   * 1. computes the losses\n   * 2. sums them to get the total loss\n   * 3. call the optimizer computes the gradients of the LayersModel's\n   *    trainable weights w.r.t. the total loss and update the variables\n   * 4. calculates the metrics\n   * 5. returns the values of the losses and metrics.\n   */\n  protected makeTrainFunction(): (data: Tensor[]) => Scalar[] {\n    return (data: Tensor[]) => {\n      const lossValues: Scalar[] = [];\n\n      const inputs = data.slice(0, this.inputs.length);\n      const targets = data.slice(\n          this.inputs.length, this.inputs.length + this.outputs.length);\n      const sampleWeights = data.slice(\n          this.inputs.length + this.outputs.length,\n          this.inputs.length + this.outputs.length * 2);\n\n      const metricsValues: Scalar[] = [];\n\n      // Create a function that computes the total loss based on the\n      // inputs. This function is used for obtaining gradients through\n      // backprop.\n      const totalLossFunction = () => {\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs =\n            execute(this.outputs, feedDict, {'training': true}) as Tensor[];\n        // TODO(cais): Take care of the case of multiple outputs from a\n        //   single layer?\n\n        let totalLoss: Tensor;\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          let loss = lossFunction(targets[i], outputs[i]);\n          if (sampleWeights[i] != null) {\n            loss = computeWeightedLoss(loss, sampleWeights[i]);\n          }\n\n          // TODO(cais): push Scalar instead.\n          const meanLoss: Scalar = tfc.mean(loss);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          lossValues.push(meanLoss);\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n        }\n\n        // Compute the metrics.\n        // TODO(cais): These should probably be calculated outside\n        //   totalLossFunction to benefit speed?\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          let weightedMetric: Scalar;\n\n          if (this.outputs.length > 1 && i < this.outputs.length) {\n            weightedMetric = lossValues[i];\n          } else {\n            const metric = this.metricsTensors[i][0];\n            const outputIndex = this.metricsTensors[i][1];\n            weightedMetric =\n                tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          }\n\n          tfc.keep(weightedMetric);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          metricsValues.push(weightedMetric);\n        }\n\n        totalLoss = tfc.mean(totalLoss);\n\n        // Add regularizer penalties.\n        this.calculateLosses().forEach(regularizerLoss => {\n          totalLoss = tfc.add(totalLoss, regularizerLoss);\n        });\n\n        return totalLoss as Scalar;\n      };\n\n      const variables = this.collectedTrainableWeights.map(\n          param => param.read() as tfc.Variable);\n      const returnCost = true;\n      const totalLossValue =\n          this.optimizer_.minimize(totalLossFunction, returnCost, variables);\n\n      return [totalLossValue].concat(metricsValues);\n    };\n  }\n\n  /**\n   * Create a function which, when invoked with an array of `tf.Tensor`s as a\n   * batch of inputs, returns the prespecified loss and metrics of the model\n   * under the batch of input data.\n   */\n  private makeTestFunction() {\n    this.testFunction = (data: Tensor[]) => {\n      return tfc.tidy(() => {\n        const valOutputs: Scalar[] = [];\n        let totalLoss: Scalar;\n        const inputs = data.slice(0, this.inputs.length);\n        const targets = data.slice(\n            this.inputs.length, this.inputs.length + this.outputs.length);\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs = execute(this.outputs, feedDict) as Tensor[];\n        // Compute total loss.\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          // TODO(cais): Add sample weighting and replace the simple\n          // averaging.\n          const loss: Scalar = tfc.mean(lossFunction(targets[i], outputs[i]));\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n          valOutputs.push(totalLoss);\n        }\n        // Compute the metrics.\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          const metric = this.metricsTensors[i][0];\n          const outputIndex = this.metricsTensors[i][1];\n          // TODO(cais): Replace K.mean() with a proper weighting function.\n          const meanMetric =\n              tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          valOutputs.push(meanMetric as Scalar);\n        }\n        return valOutputs;\n      });\n    };\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a\n   * dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *     layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * for (let i = 1; i < 5 ; ++i) {\n   *   const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *       batchSize: 4,\n   *       epochs: 3\n   *   });\n   *   console.log(\"Loss after Epoch \" + i + \" : \" + h.history.loss[0]);\n   * }\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you\n   * can also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named,\n   * you can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args A `ModelFitArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input\n   * data and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    return fitTensors(this, x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below when it's possible to instantiate\n  //   actual dataset objects.\n  /**\n   * Trains the model using a dataset object.\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for training. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs.\n   *   Of the two items in the array, the first is the input feature(s) and\n   *   the second is the output target(s).\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    return fitDataset(this, dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    // TODO(cais): Support sampleWeight and classWeight.\n    // TODO(cais): Support Dataset objects.\n    const standardizeOut = await this.standardizeUserData(x, y);\n    const inputs = standardizeOut[0];\n    const targets = standardizeOut[1];\n    const trainFunction = this.makeTrainFunction();\n    const losses = trainFunction(inputs.concat(targets));\n    const lossValues: number[] = [];\n    for (const loss of losses) {\n      const v = await loss.data();\n      lossValues.push(v[0]);\n    }\n    tfc.dispose(losses);\n    return singletonOrArray(lossValues);\n  }\n\n  /**\n   * Extract weight values of the model.\n   *\n   * @param config: An instance of `io.SaveConfig`, which specifies\n   * model-saving options such as whether only trainable weights are to be\n   * saved.\n   * @returns A `NamedTensorMap` mapping original weight names (i.e.,\n   *   non-uniqueified weight names) to their values.\n   */\n  protected getNamedWeights(config?: io.SaveConfig): NamedTensor[] {\n    const namedWeights: NamedTensor[] = [];\n\n    const trainableOnly = config != null && config.trainableOnly;\n    const weights = trainableOnly ? this.trainableWeights : this.weights;\n    const weightValues = this.getWeights(trainableOnly);\n    for (let i = 0; i < weights.length; ++i) {\n      if (trainableOnly && !weights[i].trainable) {\n        // Optionally skip non-trainable weights.\n        continue;\n      }\n      namedWeights.push(\n          {name: weights[i].originalName, tensor: weightValues[i]});\n    }\n    return namedWeights;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const input = tf.input({shape: [10]});\n   * const output = tf.layers.dense({units: 1}).apply(input);\n   * const model = tf.model({inputs: [input], outputs: [output]});\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10\n   * values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    this.stopTraining_ = stop;\n  }\n\n  get stopTraining(): boolean {\n    return this.stopTraining_;\n  }\n\n  get optimizer(): Optimizer {\n    return this.optimizer_;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    if (this.optimizer_ !== optimizer) {\n      this.optimizer_ = optimizer;\n      this.isOptimizerOwned = false;\n    }\n  }\n\n  dispose(): DisposeResult {\n    const result = super.dispose();\n    if (result.refCountAfterDispose === 0 && this.optimizer != null &&\n        this.isOptimizerOwned) {\n      const numTensorsBeforeOptmizerDisposal = tfc.memory().numTensors;\n      this.optimizer_.dispose();\n      result.numDisposedVariables +=\n          numTensorsBeforeOptmizerDisposal - tfc.memory().numTensors;\n    }\n    return result;\n  }\n\n  private getLossIdentifiers(): LossIdentifier|LossIdentifier[]|\n      {[outputName: string]: LossIdentifier} {\n    let lossNames: LossIdentifier|LossIdentifier[]|\n        {[outputName: string]: LossIdentifier};\n    if (typeof this.loss === 'string') {\n      lossNames = toSnakeCase(this.loss) as LossIdentifier;\n    } else if (Array.isArray(this.loss)) {\n      for (const loss of this.loss) {\n        if (typeof loss !== 'string') {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n      lossNames = (this.loss as string[]).map(name => toSnakeCase(name)) as\n          LossIdentifier[];\n    } else {\n      const outputNames = Object.keys(this.loss);\n      lossNames = {} as {[outputName: string]: LossIdentifier};\n      const losses =\n          this.loss as {[outputName: string]: LossOrMetricFn | string};\n      for (const outputName of outputNames) {\n        if (typeof losses[outputName] === 'string') {\n          lossNames[outputName] =\n              toSnakeCase(losses[outputName] as string) as LossIdentifier;\n        } else {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n    }\n    return lossNames;\n  }\n\n  private getMetricIdentifiers(): MetricsIdentifier[]|\n      {[key: string]: MetricsIdentifier} {\n    if (typeof this.metrics === 'string' ||\n        typeof this.metrics === 'function') {\n      return [toSnakeCase(Metrics.getLossOrMetricName(this.metrics))];\n    } else if (Array.isArray(this.metrics)) {\n      return this.metrics.map(\n          metric => toSnakeCase(Metrics.getLossOrMetricName(metric)));\n    } else {\n      const metricsIdentifiers: {[key: string]: MetricsIdentifier} = {};\n      for (const key in this.metrics) {\n        metricsIdentifiers[key] =\n            toSnakeCase(Metrics.getLossOrMetricName(this.metrics[key]));\n      }\n      return metricsIdentifiers;\n    }\n  }\n\n  protected getTrainingConfig(): TrainingConfig {\n    return {\n      loss: this.getLossIdentifiers(),\n      metrics: this.getMetricIdentifiers(),\n      optimizer_config: {\n        class_name: this.optimizer.getClassName(),\n        config: this.optimizer.getConfig()\n      } as OptimizerSerialization\n    };\n    // TODO(cais): Add weight_metrics when they are supported.\n    // TODO(cais): Add sample_weight_mode when it's supported.\n    // TODO(cais): Add loss_weights when it's supported.\n  }\n\n  loadTrainingConfig(trainingConfig: TrainingConfig) {\n    if (trainingConfig.weighted_metrics != null) {\n      throw new Error('Loading weight_metrics is not supported yet.');\n    }\n    if (trainingConfig.loss_weights != null) {\n      throw new Error('Loading loss_weights is not supported yet.');\n    }\n    if (trainingConfig.sample_weight_mode != null) {\n      throw new Error('Loading sample_weight_mode is not supported yet.');\n    }\n\n    const tsConfig = convertPythonicToTs(trainingConfig.optimizer_config) as\n        serialization.ConfigDict;\n    const optimizer = deserialize(tsConfig) as Optimizer;\n\n    let loss;\n    if (typeof trainingConfig.loss === 'string') {\n      loss = toCamelCase(trainingConfig.loss);\n    } else if (Array.isArray(trainingConfig.loss)) {\n      loss = trainingConfig.loss.map(lossEntry => toCamelCase(lossEntry));\n    } else if (trainingConfig.loss != null) {\n      loss = {} as {[outputName: string]: LossIdentifier};\n      for (const key in trainingConfig.loss) {\n        loss[key] = toCamelCase(trainingConfig.loss[key]) as LossIdentifier;\n      }\n    }\n\n    let metrics;\n    if (Array.isArray(trainingConfig.metrics)) {\n      metrics = trainingConfig.metrics.map(metric => toCamelCase(metric));\n    } else if (trainingConfig.metrics != null) {\n      metrics = {} as {[outputName: string]: MetricsIdentifier};\n      for (const key in trainingConfig.metrics) {\n        metrics[key] = toCamelCase(trainingConfig.metrics[key]);\n      }\n    }\n\n    this.compile({loss, metrics, optimizer});\n  }\n\n  /**\n   * Save the configuration and/or weights of the LayersModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 2. Saving `model`'s topology and weights to browser\n   * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('indexeddb://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 3. Saving `model`'s topology and weights as two files\n   * (`my-model-1.json` and `my-model-1.weights.bin`) downloaded from\n   * browser.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('downloads://my-model-1');\n   * ```\n   *\n   * Example 4. Send  `model`'s topology and weights to an HTTP server.\n   * See the documentation of `tf.io.http` for more details\n   * including specifying request parameters and implementation of the\n   * server.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('http://my-server/model/upload');\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new ValueError(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new ValueError(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new ValueError(\n          'LayersModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    const weightDataAndSpecs =\n        await io.encodeWeights(this.getNamedWeights(config));\n\n    const returnString = false;\n    const unusedArg: {} = null;\n    const modelConfig = this.toJSON(unusedArg, returnString);\n    const modelArtifacts: io.ModelArtifacts = {\n      modelTopology: modelConfig,\n      format: LAYERS_MODEL_FORMAT_NAME,\n      generatedBy: `TensorFlow.js tfjs-layers v${version}`,\n      convertedBy: null,\n    };\n\n    const includeOptimizer = config == null ? false : config.includeOptimizer;\n    if (includeOptimizer && this.optimizer != null) {\n      modelArtifacts.trainingConfig = this.getTrainingConfig();\n      const weightType = 'optimizer';\n      const {data: optimizerWeightData, specs: optimizerWeightSpecs} =\n          await io.encodeWeights(await this.optimizer.getWeights(), weightType);\n      weightDataAndSpecs.specs.push(...optimizerWeightSpecs);\n      weightDataAndSpecs.data = io.concatenateArrayBuffers(\n          [weightDataAndSpecs.data, optimizerWeightData]);\n    }\n\n    if (this.userDefinedMetadata != null) {\n      // Check serialized size of user-defined metadata.\n      const checkSize = true;\n      checkUserDefinedMetadata(this.userDefinedMetadata, this.name, checkSize);\n      modelArtifacts.userDefinedMetadata = this.userDefinedMetadata;\n    }\n\n    modelArtifacts.weightData = weightDataAndSpecs.data;\n    modelArtifacts.weightSpecs = weightDataAndSpecs.specs;\n    return handlerOrURL.save(modelArtifacts);\n  }\n\n  /**\n   * Set user-defined metadata.\n   *\n   * The set metadata will be serialized together with the topology\n   * and weights of the model during `save()` calls.\n   *\n   * @param setUserDefinedMetadata\n   */\n  setUserDefinedMetadata(userDefinedMetadata: {}): void {\n    checkUserDefinedMetadata(userDefinedMetadata, this.name);\n    this.userDefinedMetadata = userDefinedMetadata;\n  }\n\n  /**\n   * Get user-defined metadata.\n   *\n   * The metadata is supplied via one of the two routes:\n   *   1. By calling `setUserDefinedMetadata()`.\n   *   2. Loaded during model loading (if the model is constructed\n   *      via `tf.loadLayersModel()`.)\n   *\n   * If no user-defined metadata is available from either of the\n   * two routes, this function will return `undefined`.\n   */\n  getUserDefinedMetadata(): {} {\n    return this.userDefinedMetadata;\n  }\n}\nserialization.registerClass(LayersModel);\n\n/**\n * A `tf.Functional` is an alias to `tf.LayersModel`.\n *\n * See also:\n *   `tf.LayersModel`, `tf.Sequential`, `tf.loadLayersModel`.\n */\n/** @doc {heading: 'Models', subheading: 'Classes'} */\nexport class Functional extends LayersModel {\n  static className = 'Functional';\n}\nserialization.registerClass(Functional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source keras/models.py */\n\nimport {dispose, io, NamedTensorMap, Optimizer, Scalar, serialization, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {getUid} from './backend/state';\nimport {History} from './base_callbacks';\nimport {Dataset} from './engine/dataset_stub';\nimport {Input} from './engine/input_layer';\nimport {getSourceInputs, Layer, Node, SymbolicTensor} from './engine/topology';\nimport {LayersModel, ModelCompileArgs, ModelEvaluateArgs} from './engine/training';\nimport {ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './engine/training_dataset';\nimport {ModelFitArgs} from './engine/training_tensors';\nimport {NotImplementedError, RuntimeError, ValueError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {TrainingConfig} from './keras_format/training_config';\nimport {PyJsonDict} from './keras_format/types';\nimport {deserialize} from './layers/serialization';\nimport {Kwargs, NamedTensor} from './types';\nimport * as generic_utils from './utils/generic_utils';\nimport {convertPythonicToTs} from './utils/serialization_utils';\nimport {getExactlyOneShape} from './utils/types_utils';\n\n/**\n * Parses a JSON model configuration file and returns a model instance.\n *\n * ```js\n * // This example shows how to serialize a model using `toJSON()` and\n * // deserialize it as another model using `tf.models.modelFromJSON()`.\n * // Note: this example serializes and deserializes only the topology\n * // of the model; the weights of the loaded model will be different\n * // from those of the the original model, due to random weight\n * // initialization.\n * // To load the topology and weights of a model, use `tf.loadLayersModel()`.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.repeatVector({inputShape: [2], n: 4}));\n * // Serialize `model1` as a JSON object.\n * const model1JSON = model1.toJSON(null, false);\n * model1.summary();\n *\n * const model2 = await tf.models.modelFromJSON(model1JSON);\n * model2.summary();\n * ```\n *\n *  @param modelAndWeightsConfig JSON object or string encoding a model and\n *       weights configuration. It can also be only the topology JSON of the\n *       model, in which case the weights will not be loaded.\n *  @param custom_objects Optional dictionary mapping names\n *       (strings) to custom classes or functions to be\n *       considered during deserialization.\n * @returns A TensorFlow.js Layers `tf.LayersModel` instance (uncompiled).\n */\nexport async function modelFromJSON(\n    modelAndWeightsConfig: ModelAndWeightsConfig|PyJsonDict,\n    customObjects?: serialization.ConfigDict): Promise<LayersModel> {\n  if (!('modelTopology' in modelAndWeightsConfig)) {\n    modelAndWeightsConfig = {modelTopology: modelAndWeightsConfig};\n  }\n  modelAndWeightsConfig = modelAndWeightsConfig as ModelAndWeightsConfig;\n\n  let modelTopology = modelAndWeightsConfig.modelTopology;\n  if (modelTopology['model_config'] != null) {\n    // If the model-topology JSON contains a 'model_config' field, then it is\n    // a full model JSON (e.g., from `keras.Model.save()`), which contains\n    // not only the model's architecture in its 'model_config' field, but\n    // additional information such as the model's optimizer. We use only the\n    // 'model_config' field currently.\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n  const tsConfig =\n      convertPythonicToTs(modelTopology) as serialization.ConfigDict;\n  const model = deserialize(tsConfig, customObjects) as LayersModel;\n\n  if (modelAndWeightsConfig.weightsManifest != null) {\n    // Load the weight values keyed by the original tensor names in the model\n    // file that was loaded.  These should match the keys of the weight\n    // manifest.\n    const weightValues = await io.loadWeights(\n        modelAndWeightsConfig.weightsManifest, modelAndWeightsConfig.pathPrefix,\n        model.weights.map(weight => weight.originalName));\n\n    // Map the weights to the unique tensor names generated during model loading\n    const uniqueWeightValues: NamedTensorMap = {};\n    for (const weight of model.weights) {\n      uniqueWeightValues[weight.originalName] =\n          weightValues[weight.originalName];\n    }\n\n    model.loadWeights(uniqueWeightValues);\n    // Dispose temporary weight values.\n    dispose(weightValues);\n  }\n  return model;\n}\n\n/**\n * Options for loading a saved mode in TensorFlow.js format.\n */\nexport interface ModelAndWeightsConfig {\n  /**\n   * A JSON object or JSON string containing the model config.\n   *\n   * This can be either of the following two formats:\n   *   - A model archiecture-only config,  i.e., a format consistent with the\n   *     return value of`keras.Model.to_json()`.\n   *   - A full model config, containing not only model architecture, but also\n   *     training options and state, i.e., a format consistent with the return\n   *     value of `keras.models.save_model()`.\n   */\n  modelTopology: PyJsonDict;\n\n  /**\n   * A weights manifest in TensorFlow.js format.\n   */\n  weightsManifest?: io.WeightsManifestConfig;\n\n  /**\n   * Path to prepend to the paths in `weightManifest` before fetching.\n   *\n   * The path may optionally end in a slash ('/').\n   */\n  pathPrefix?: string;\n}\n\n// TODO(nielsene): Remove after: https://github.com/tensorflow/tfjs/issues/400\nexport interface ModelPredictArgs {\n  /**\n   * Optional. Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Optional. Verbosity mode. Defaults to false.\n   */\n  verbose?: boolean;\n}\n\n/**\n * Load a model, including its topology and optionally weights.  See the\n * Tutorial named \"How to import a Keras Model\" for usage examples.\n *\n * Example 1: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 2. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * Example 4. Load a model from an HTTP server.\n *\n * ```js\n * const model = await\n *     tf.loadLayersModel('https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. This path will be\n *      interpreted as a relative HTTP path, to which `fetch` will be used to\n *      request the model topology and weight manifest JSON.\n *      The content of the JSON file is assumed to be a JSON object with the\n *      following fields and values:\n *      - 'modelTopology': A JSON object that can be either of:\n *        1. a model architecture JSON consistent with the format of the return\n *            value of `keras.Model.to_json()`\n *        2. a full model JSON in the format of `keras.models.save_model()`.\n *      - 'weightsManifest': A TensorFlow.js weights manifest.\n *      See the Python converter function `save_model()` for more details.\n *      It is also assumed that model weights can be accessed from relative\n *      paths described by the `paths` fields in weights manifest.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A progress callback of the form:\n *     `(fraction: number) => void`. This callback can be used to monitor the\n *     model-loading process.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n */\nexport async function loadLayersModelInternal(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (typeof pathOrIOHandler === 'string') {\n    const handlers = io.getLoadHandlers(pathOrIOHandler, options);\n    if (handlers.length === 0) {\n      // For backward compatibility: if no load handler can be found,\n      // assume it is a relative http path.\n      // TODO(cais): Reformat the args into a single `LoadOptions` once the core\n      // is refactored.\n      handlers.push(io.browserHTTPRequest(pathOrIOHandler, options));\n    } else if (handlers.length > 1) {\n      throw new ValueError(\n          `Found more than one (${handlers.length}) load handlers for ` +\n          `URL '${pathOrIOHandler}'`);\n    }\n    pathOrIOHandler = handlers[0];\n  }\n  return loadLayersModelFromIOHandler(pathOrIOHandler, undefined, options);\n}\n\n/**\n * Load a model and optionally its weights, using an IOHandler object.\n *\n * @param handler The instance of `IOHandler` to be used during the model\n *   loading.\n * @param customObjects Any optional custom objects to be used during model\n *   loading.\n * @param strict Whether the weight loading will be done in strict mode.\n *   Default: `true`.\n */\nexport async function loadLayersModelFromIOHandler(\n    handler: io.IOHandler, customObjects?: serialization.ConfigDict,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (handler.load == null) {\n    throw new ValueError(\n        'Cannot proceed with model loading because the IOHandler provided ' +\n        'does not have the `load` method implemented.');\n  }\n  const artifacts = await handler.load();\n  let modelTopology = artifacts.modelTopology as PyJsonDict;\n  if (modelTopology['model_config'] != null) {\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n\n  const strict = options.strict == null ? true : options.strict;\n  // If weights are provided and the weight-loading mode is strict, use\n  // fast weight initialization. This skips costly initializers such as\n  // 'orthogonal' and saves unnecessary computation in cases where\n  // the initialized weight values will immediately be overwritten by\n  // loaded weight values.\n  const fastWeightInit =\n      artifacts.weightData != null && artifacts.weightSpecs != null && strict;\n  const model =\n      deserialize(\n          convertPythonicToTs(modelTopology) as serialization.ConfigDict,\n          customObjects, fastWeightInit) as LayersModel;\n\n  const trainingConfig = artifacts.trainingConfig as TrainingConfig;\n  if (trainingConfig != null) {\n    model.loadTrainingConfig(trainingConfig);\n  }\n  if (artifacts.userDefinedMetadata != null) {\n    model.setUserDefinedMetadata(artifacts.userDefinedMetadata);\n  }\n\n  // If weightData is present, load the weights into the model.\n  if (artifacts.weightData != null) {\n    // Loading weights requires weightSpecs.\n    if (artifacts.weightSpecs == null) {\n      throw new ValueError(\n          'LayersModel artifacts contains weight data, but not weight specs. ' +\n          'Therefore loading of weights cannot proceed.');\n    }\n\n    const {modelWeights, optimizerWeights} = decodeModelAndOptimizerWeights(\n        artifacts.weightData, artifacts.weightSpecs);\n    model.loadWeights(modelWeights, strict);\n\n    if (model.optimizer != null && optimizerWeights.length > 0) {\n      await model.optimizer.setWeights(optimizerWeights);\n    }\n\n    // Dispose temporary weight values.\n    dispose(modelWeights);\n    dispose(optimizerWeights.map(w => w.tensor));\n  }\n  return model;\n}\n\nfunction decodeModelAndOptimizerWeights(\n    buffer: ArrayBuffer, specs: io.WeightsManifestEntry[]):\n    {modelWeights: NamedTensorMap, optimizerWeights: NamedTensor[]} {\n  const name2Tensor = io.decodeWeights(buffer, specs);\n  const modelWeights: NamedTensorMap = {};\n  const optimizerWeights: NamedTensor[] = [];\n  specs.forEach(spec => {\n    if (spec.group === 'optimizer') {\n      optimizerWeights.push({name: spec.name, tensor: name2Tensor[spec.name]});\n    } else {\n      modelWeights[spec.name] = name2Tensor[spec.name];\n    }\n  });\n  return {modelWeights, optimizerWeights};\n}\n\n/**\n * Configuration for a Sequential model.\n */\nexport interface SequentialArgs {\n  /** Stack of layers for the model. */\n  layers?: Layer[];\n\n  /** The name of this model. */\n  name?: string;\n}\n\n/**\n * A model with a stack of layers, feeding linearly from one to the next.\n *\n * `tf.sequential` is a factory function that creates an instance of\n * `tf.Sequential`.\n *\n * ```js\n *  // Define a model for linear regression.\n *  const model = tf.sequential();\n *  model.add(tf.layers.dense({units: 1, inputShape: [1]}));\n *\n *  // Prepare the model for training: Specify the loss and the optimizer.\n *  model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n *\n *  // Generate some synthetic data for training.\n *  const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);\n *  const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);\n *\n *  // Train the model using the data then do inference on a data point the\n *  // model hasn't seen:\n *  await model.fit(xs, ys);\n *  model.predict(tf.tensor2d([5], [1, 1])).print();\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class Sequential extends LayersModel {\n  /** @nocollapse */\n  static className = 'Sequential';\n  private model: LayersModel;\n  constructor(args?: SequentialArgs) {\n    super({inputs: [], outputs: []});\n    args = args || {};\n\n    this.trainable = true;\n    this.built = false;\n\n    // Set model name.\n    this.name = (args.name != null) ? args.name : getUid('sequential_');\n\n    // Add to the model any layers passed to the constructor.\n    if (args.layers != null) {\n      for (const layer of args.layers) {\n        this.add(layer);\n      }\n    }\n  }\n\n  // Helper function to Sequential.add  Throws if the new output shape will be\n  // invalid.\n  private checkShape(layer: Layer) {\n    const shape = layer.inboundNodes[0].outputTensors[0].shape;\n    if (shape.some(x => x < 0)) {\n      throw new ValueError(\n          'Negative dimension size caused by adding layer ' +\n          `${layer.name} with input shape [` +\n          `${layer.inboundNodes[0].inputTensors[0].shape}]`);\n    }\n  }\n\n  /**\n   * Adds a layer instance on top of the layer stack.\n   *\n   * ```js\n   *  const model = tf.sequential();\n   *  model.add(tf.layers.dense({units: 8, inputShape: [1]}));\n   *  model.add(tf.layers.dense({units: 4, activation: 'relu6'}));\n   *  model.add(tf.layers.dense({units: 1, activation: 'relu6'}));\n   *  // Note that the untrained model is random at this point.\n   *  model.predict(tf.randomNormal([10, 1])).print();\n   * ```\n   * @param layer Layer instance.\n   *\n   * @exception ValueError In case the `layer` argument does not know its\n   * input shape.\n   * @exception ValueError In case the `layer` argument has multiple output\n   *   tensors, or is already connected somewhere else (forbidden in\n   *   `Sequential` models).\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  add(layer: Layer): void {\n    const isLayerModelInstance =\n        layer instanceof Sequential || layer instanceof LayersModel;\n    let modelLayer: LayersModel;\n    if (isLayerModelInstance) {\n      modelLayer = layer as LayersModel;\n      if (modelLayer.outputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      if (modelLayer.inputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single input tensor. ' +\n            'For multi-input layers, ' +\n            'use the functional API.');\n      }\n    }\n\n    if (this.outputs.length === 0) {\n      // first layer in model: check that it is an input layer\n      if (layer.inboundNodes.length === 0) {\n        // create an input layer\n        if (layer.batchInputShape == null) {\n          throw new ValueError(\n              'The first layer in a Sequential model must ' +\n              'get an `inputShape` or `batchInputShape` argument.');\n        }\n        // Instantiate the input layer.\n        const x = Input({\n          batchShape: layer.batchInputShape,\n          dtype: layer.dtype,\n          name: layer.name + '_input'\n        });\n        // This will build the current layer and create the node connecting\n        // the current layer to the input layer we just created.\n        layer.apply(x);\n      }\n\n      if (isLayerModelInstance) {\n        this.outputs = modelLayer.outputs;\n        this.inputs = modelLayer.inputs;\n      } else {\n        if (layer.inboundNodes.length !== 1) {\n          throw new ValueError(\n              'A layer added to a Sequential model must not already be ' +\n              `connected somewhere else. LayersModel received layer ${\n                  layer.name} ` +\n              `which has ${layer.inboundNodes.length} pre-existing inbound ` +\n              'connections.');\n        }\n\n        if (layer.inboundNodes[0].outputTensors.length !== 1) {\n          throw new ValueError(\n              'All layers in a Sequential model ' +\n              'should have a single output tensor. ' +\n              'For multi-output layers, ' +\n              'use the functional API.');\n        }\n        this.checkShape(layer);\n        this.outputs = [layer.inboundNodes[0].outputTensors[0]];\n        this.inputs = getSourceInputs(this.outputs[0]);\n      }\n\n      this.inboundNodes = [];\n      // We create an input node, which we will keep updated\n      // as we add more layers.\n      // (This call has side effects.)\n      // tslint:disable-next-line:no-unused-expression\n      new Node({\n        outboundLayer: this,\n        inboundLayers: [],\n        nodeIndices: [],\n        tensorIndices: [],\n        inputTensors: this.inputs,\n        outputTensors: this.outputs,\n        // no model-level masking for now\n        inputMasks: generic_utils.pyListRepeat(null, this.inputs.length),\n        outputMasks: [null],\n        inputShapes: this.inputs.map(x => x.shape),\n        outputShapes: this.outputs[0].shape\n      });\n    } else {\n      const outputTensor = layer.apply(this.outputs[0]);\n      if (Array.isArray(outputTensor)) {\n        throw new TypeError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      this.checkShape(layer);\n      this.outputs = [outputTensor as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n\n    this.layers.push(layer);\n    this.built = false;\n  }\n\n  /**\n   * Removes the last layer in the model.\n   *\n   * @exception TypeError if there are no layers in the model.\n   */\n  pop(): void {\n    if (this.layers.length === 0) {\n      throw new TypeError('There are no layers in the model.');\n    }\n\n    this.layers.pop();\n    if (this.layers.length === 0) {\n      this.outputs = [];\n      this.inboundNodes = [];\n      this.outboundNodes = [];\n    } else {\n      const lastLayerIndex = this.layers.length - 1;\n      this.layers[lastLayerIndex].outboundNodes = [];\n      this.outputs = [this.layers[lastLayerIndex].output as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.call(inputs, kwargs);\n  }\n\n  build(inputShape?: Shape|Shape[]) {\n    // Call `getExactlyOneShape` without using its return value,\n    // to verify that exactly one input shape is provided.\n    getExactlyOneShape(inputShape);\n\n    if (this.inputs.length === 0 || this.outputs.length === 0) {\n      throw new TypeError(\n          'Sequential model cannot be built: model is empty.' +\n          ' Add some layers first.');\n    }\n    // actually create the model\n    this.model = new LayersModel({\n      inputs: this.inputs,\n      outputs: this.outputs[0],\n      name: this.name + '_model'\n    });\n    this.model.trainable = this.trainable;\n\n    // mirror model attributes\n    this.supportsMasking = this.model.supportsMasking;\n    // TODO(michaelterry): Add caches\n    this.inputLayers = this.model.inputLayers;\n    this.inputLayersNodeIndices = this.model.inputLayersNodeIndices;\n    this.inputLayersTensorIndices = this.model.inputLayersTensorIndices;\n    this.outputLayers = this.model.outputLayers;\n    this.outputLayersNodeIndices = this.model.outputLayersNodeIndices;\n    this.outputLayersTensorIndices = this.model.outputLayersTensorIndices;\n    this.nodesByDepth = this.model.nodesByDepth;\n    this.containerNodes = this.model.containerNodes;\n    this.outputNames = this.model.outputNames;\n    this.inputNames = this.model.inputNames;\n    // TODO(michaelterry): Add feedInputNames, feedInputs, if needed.\n    // TODO(michaelterry): Add callbackModel if needed.\n    this.built = true;\n  }\n\n  countParams(): number {\n    if (!this.built) {\n      this.build();\n    }\n    return super.countParams();\n  }\n\n  /**\n   * Print a text summary of the Sequential model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - The total number of trainable and non-trainable parameters of the\n   * model.\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(\n   *     tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'}));\n   * model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n   *\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      this.build();\n    }\n    super.summary(lineLength, positions, printFn);\n  }\n\n  /**\n   * Sets the weights of the model.\n   *\n   * @param weights Should be a list of Tensors with shapes and types matching\n   *   the output of `model.getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    if (this.model == null) {\n      this.build();\n    }\n    this.model.setWeights(weights);\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   * });\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateConfig`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluate(x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluateDataset(dataset, args);\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([2, 10])).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param conifg A `ModelPredictConfig` object containing optional fields.\n   *\n   * @return `tf.Tensor`(s) of predictions.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predict(x, args);\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * @param x: Input samples, as a Tensor, or list of Tensors (if the model\n   *   has multiple inputs).\n   * @return Tensor(s) of predictions\n   */\n  predictOnBatch(x: Tensor): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predictOnBatch(x);\n  }\n\n  /**\n   * See `LayersModel.compile`.\n   *\n   * @param args\n   */\n  compile(args: ModelCompileArgs): void {\n    this.build();\n    this.model.compile(args);\n    this.optimizer_ = this.model.optimizer;\n    // tslint:disable-next-line:no-any\n    this.isOptimizerOwned = (this.model as any).isOptimizerOwned;\n    this.loss = this.model.loss;\n    this.metrics = this.model.metrics;\n    // TODO(cais): Add this.lossWeights, this.sampleWeightMode,\n    //   this.weightedMetrics, this.targets.\n    this.metricsTensors = this.model.metricsTensors;\n    this.metricsNames = this.model.metricsNames;\n    // TODO(cais): Add sampleWeights.\n  }\n\n  get optimizer(): Optimizer {\n    return this.model == null ? undefined : this.model.optimizer;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    this.model.optimizer = optimizer;\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   *   epochs: 3\n   * });\n   * console.log(history.history.loss[0]);\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you can\n   * also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named, you\n   *  can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args  A `ModelFitConfig`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fit(x, y, args);\n  }\n\n  /**\n   * Trains the model using a dataset object.\n   *\n   * ```js\n   * const xArray = [\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   * ];\n   * const yArray = [1, 1, 1, 1];\n   * // Create a dataset from the JavaScript array.\n   * const xDataset = tf.data.array(xArray);\n   * const yDataset = tf.data.array(yArray);\n   * // Zip combines the `x` and `y` Datasets into a single Dataset, the\n   * // iterator of which will return an object containing of two tensors,\n   * // corresponding to `x` and `y`.  The call to `batch(4)` will bundle\n   * // four such samples into a single object, with the same keys now pointing\n   * // to tensors that hold 4 examples, organized along the batch dimension.\n   * // The call to `shuffle(4)` causes each iteration through the dataset to\n   * // happen in a different order.  The size of the shuffle window is 4.\n   * const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset})\n   *     .batch(4)\n   *     .shuffle(4);\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [9]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fitDataset(xyDataset, {\n   *   epochs: 4,\n   *   callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)}\n   * });\n   * ```\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected to\n   *   generate a dataset iterator object, the `next()` method of which is\n   *   expected to produce data batches for evaluation. The return value of the\n   *   `next()` call ought to contain a boolean `done` field and a `value`\n   *   field.\n   *\n   *   The `value` field is expected to be an object of with fields\n   *   `xs` and `ys`, which point to the feature tensor and the target tensor,\n   *   respectively. This case is for models with exactly one input and one\n   *   output (e.g.. a sequential model). For example:\n   *   ```js\n   *   {value: {xs: xsTensor, ys: ysTensor}, done: false}\n   *   ```\n   *\n   *   If the model has multiple inputs, the `xs` field of `value` should\n   *   be an object mapping input names to their respective feature tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: {\n   *         input_1: xsTensor1,\n   *         input_2: xsTensor2\n   *       },\n   *       ys: ysTensor\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   *   If the model has multiple outputs, the `ys` field of `value` should\n   *   be an object mapping output names to their respective target tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: xsTensor,\n   *       ys: {\n   *         output_1: ysTensor1,\n   *         output_2: ysTensor2\n   *       },\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fitDataset(dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    return this.model.trainOnBatch(x, y);\n  }\n\n  /* See parent class for JsDoc */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    let configArray: serialization.ConfigDictArray;\n    let extraModelConfig: serialization.ConfigDict = {};\n    if (config instanceof Array) {\n      if (!(config[0].className != null) ||\n          config[0]['className'] === 'Merge') {\n        throw new ValueError('Legacy serialization format not supported yet.');\n      }\n      configArray = config;\n    } else {\n      util.assert(\n          config['layers'] != null,\n          () =>\n              `When the config data for a Sequential model is not an Array, ` +\n              `it must be an Object that contains the 'layers' field.`);\n      configArray = config['layers'] as serialization.ConfigDictArray;\n      delete config['layers'];\n      extraModelConfig = config;\n    }\n\n    const model = new cls(extraModelConfig);\n    if (!(model instanceof Sequential)) {\n      throw new NotImplementedError(\n          `Sequential.fromConfig called on non-Sequential input: ${model}`);\n    }\n    for (const conf of configArray) {\n      const customObjects: serialization.ConfigDict = undefined;\n      const layer = deserialize(\n                        conf as serialization.ConfigDict, customObjects,\n                        fastWeightInit) as Layer;\n      if (fastWeightInit) {\n        layer.setFastWeightInitDuringBuild(true);\n      }\n      model.add(layer);\n    }\n    return model;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(tf.layers.dense({units: 1, inputShape: [10]}));\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10 values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    // TODO(cais): When refactoring to remove the composition pattern happens,\n    // remove this method overriding.\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot set the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    this.model.stopTraining = stop;\n  }\n\n  get stopTraining(): boolean {\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot get the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    return this.model.stopTraining;\n  }\n\n  // TODO(cais): Override get trainableWeights() here\n\n  // tslint:disable-next-line:no-any\n  getConfig(): any {\n    // NOTE(cais): We override the return type of getConfig() to `any` here,\n    //   because the `Sequential` class is a special case among `Container`\n    //   subtypes in that its getConfig() method returns an Array (not a\n    //   dict).\n    const layers: serialization.ConfigDict[] = [];\n    for (const layer of this.layers) {\n      const dict: serialization.ConfigDict = {};\n      dict['className'] = layer.getClassName();\n      dict['config'] = layer.getConfig();\n      layers.push(dict);\n    }\n    return {name: this.name, layers};\n  }\n}\nserialization.registerClass(Sequential);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Exported functions.\n */\n\nimport {io} from '@tensorflow/tfjs-core';\n\nimport {BaseCallbackConstructor, CallbackConstructorRegistry} from './base_callbacks';\nimport {ContainerArgs} from './engine/container';\nimport {Input, InputConfig,} from './engine/input_layer';\nimport {SymbolicTensor} from './engine/topology';\nimport {LayersModel} from './engine/training';\nimport {loadLayersModelInternal, Sequential, SequentialArgs} from './models';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// LayersModel and related factory methods.\n\n/**\n * A model is a data structure that consists of `Layers` and defines inputs\n * and outputs.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.model` is more generic, supporting an arbitrary graph (without\n * cycles) of layers. `tf.sequential` is less generic and supports only a linear\n * stack of layers.\n *\n * When creating a `tf.LayersModel`, specify its input(s) and output(s). Layers\n * are used to wire input(s) to output(s).\n *\n * For example, the following code snippet defines a model consisting of\n * two `dense` layers, with 10 and 4 units, respectively.\n *\n * ```js\n * // Define input, which has a size of 5 (not including batch dimension).\n * const input = tf.input({shape: [5]});\n *\n * // First dense layer uses relu activation.\n * const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});\n * // Second dense layer uses softmax activation.\n * const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});\n *\n * // Obtain the output symbolic tensor by applying the layers on the input.\n * const output = denseLayer2.apply(denseLayer1.apply(input));\n *\n * // Create the model based on the inputs.\n * const model = tf.model({inputs: input, outputs: output});\n *\n * // The model can be used for training, evaluation and prediction.\n * // For example, the following line runs prediction with the model on\n * // some fake data.\n * model.predict(tf.ones([2, 5])).print();\n * ```\n * See also:\n *   `tf.sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function model(args: ContainerArgs): LayersModel {\n  return new LayersModel(args);\n}\n\n/**\n * Creates a `tf.Sequential` model.  A sequential model is any model where the\n * outputs of one layer are the inputs to the next layer, i.e. the model\n * topology is a simple 'stack' of layers, with no branching or skipping.\n *\n * This means that the first layer passed to a `tf.Sequential` model should have\n * a defined input shape. What that means is that it should have received an\n * `inputShape` or `batchInputShape` argument, or for some type of layers\n * (recurrent, Dense...) an `inputDim` argument.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.sequential` is less generic, supporting only a linear stack of layers.\n * `tf.model` is more generic and supports an arbitrary graph (without\n * cycles) of layers.\n *\n * Examples:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have an input shape defined.\n * model.add(tf.layers.dense({units: 32, inputShape: [50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output, which equals\n * // `[null, 4]`. The 1st dimension is the undetermined batch dimension; the\n * // 2nd is the output size of the model's last layer.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * It is also possible to specify a batch size (with potentially undetermined\n * batch dimension, denoted by \"null\") for the first layer using the\n * `batchInputShape` key. The following example is equivalent to the above:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have a defined input shape\n * model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * You can also use an `Array` of already-constructed `Layer`s to create\n * a `tf.Sequential` model:\n *\n * ```js\n * const model = tf.sequential({\n *   layers: [tf.layers.dense({units: 32, inputShape: [50]}),\n *            tf.layers.dense({units: 4})]\n * });\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function sequential(config?: SequentialArgs): Sequential {\n  return new Sequential(config);\n}\n\n/**\n * Load a model composed of Layer objects, including its topology and optionally\n * weights. See the Tutorial named \"How to import a Keras Model\" for usage\n * examples.\n *\n * This method is applicable to:\n *\n * 1. Models created with the `tf.layers.*`, `tf.sequential`, and\n * `tf.model` APIs of TensorFlow.js and later saved with the\n * `tf.LayersModel.save` method.\n * 2. Models converted from Keras or TensorFlow tf.keras using the\n * [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).\n *\n * This mode is *not* applicable to TensorFlow `SavedModel`s or their converted\n * forms. For those models, use `tf.loadGraphModel`.\n *\n * Example 1. Load a model from an HTTP server.\n *\n * ```js\n * const model = await tf.loadLayersModel(\n *     'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * Example 2: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 4. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. For file://\n *      (tfjs-node-only), http:// and https:// schemas, the path can be\n *      either absolute or relative.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A function of the signature `(fraction: number) => void',\n *     that can be used as the progress callback for the model loading.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport function loadLayersModel(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  return loadLayersModelInternal(pathOrIOHandler, options);\n}\n\n/**\n * Used to instantiate an input to a model as a `tf.SymbolicTensor`.\n *\n * Users should call the `input` factory function for\n * consistency with other generator functions.\n *\n * Example:\n *\n * ```js\n * // Defines a simple logistic regression model with 32 dimensional input\n * // and 3 dimensional output.\n * const x = tf.input({shape: [32]});\n * const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x);\n * const model = tf.model({inputs: x, outputs: y});\n * model.predict(tf.ones([2, 32])).print();\n * ```\n *\n * Note: `input` is only necessary when using `model`. When using\n * `sequential`, specify `inputShape` for the first layer or use `inputLayer`\n * as the first layer.\n *\n * @doc {heading: 'Models', subheading: 'Inputs'}\n */\nexport function input(config: InputConfig): SymbolicTensor {\n  return Input(config);\n}\n\nexport function registerCallbackConstructor(\n    verbosityLevel: number,\n    callbackConstructor: BaseCallbackConstructor): void {\n  CallbackConstructorRegistry.registerCallbackConstructor(\n      verbosityLevel, callbackConstructor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Layer activation functions\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {ActivationIdentifier} from './keras_format/activation_config';\nimport {deserializeKerasObject} from './utils/generic_utils';\n\n/**\n * Base class for Activations.\n *\n * Special note: due to cross-language compatibility reasons, the\n * static readonly className field in this family of classes must be set to\n * the initialLowerCamelCase name of the activation.\n */\nexport abstract class Activation extends serialization.Serializable {\n  abstract apply(tensor: Tensor, axis?: number): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\n/**\n * Exponential linear unit (ELU).\n * Reference: https://arxiv.org/abs/1511.07289\n */\nexport class Elu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'elu';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x: Input.\n   * @param alpha: Scaling factor the negative section.\n   * @return Output of the ELU activation.\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return K.elu(x, alpha);\n  }\n}\nserialization.registerClass(Elu);\n\n/**\n * Scaled Exponential Linear Unit. (Klambauer et al., 2017).\n * Reference: Self-Normalizing Neural Networks, https://arxiv.org/abs/1706.02515\n * Notes:\n *   - To be used together with the initialization \"lecunNormal\".\n *   - To be used together with the dropout variant \"AlphaDropout\".\n */\nexport class Selu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'selu';\n  apply(x: Tensor): Tensor {\n    return tfc.selu(x);\n  }\n}\nserialization.registerClass(Selu);\n\n/**\n *  Rectified linear unit\n */\nexport class Relu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu';\n  apply(x: Tensor): Tensor {\n    return tfc.relu(x);\n  }\n}\nserialization.registerClass(Relu);\n\n/**\n * Rectified linear unit activation maxing out at 6.0.\n */\nexport class Relu6 extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu6';\n  apply(x: Tensor): Tensor {\n    return tidy(() => tfc.minimum(6.0, tfc.relu(x)));\n  }\n}\nserialization.registerClass(Relu6);\n\n//* Linear activation (no-op) */\nexport class Linear extends Activation {\n  /** @nocollapse */\n  static readonly className = 'linear';\n  apply(x: Tensor): Tensor {\n    return x;\n  }\n}\nserialization.registerClass(Linear);\n\n/**\n * Sigmoid activation function.\n */\nexport class Sigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'sigmoid';\n  apply(x: Tensor): Tensor {\n    return tfc.sigmoid(x);\n  }\n}\nserialization.registerClass(Sigmoid);\n\n/**\n * Segment-wise linear approximation of sigmoid.\n */\nexport class HardSigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'hardSigmoid';\n  apply(x: Tensor): Tensor {\n    return K.hardSigmoid(x);\n  }\n}\nserialization.registerClass(HardSigmoid);\n\n/**\n * Softplus activation function.\n */\nexport class Softplus extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softplus';\n  apply(x: Tensor): Tensor {\n    return tfc.softplus(x);\n  }\n}\nserialization.registerClass(Softplus);\n\n/**\n * Softsign activation function.\n */\nexport class Softsign extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softsign';\n  apply(x: Tensor): Tensor {\n    return K.softsign(x);\n  }\n}\nserialization.registerClass(Softsign);\n\n/**\n * Hyperbolic tangent function.\n */\nexport class Tanh extends Activation {\n  /** @nocollapse */\n  static readonly className = 'tanh';\n  apply(x: Tensor): Tensor {\n    return tfc.tanh(x);\n  }\n}\nserialization.registerClass(Tanh);\n\n/**\n * Softmax activation function\n */\nexport class Softmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softmax';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.softmax(x, axis);\n  }\n}\nserialization.registerClass(Softmax);\n\n/**\n * Log softmax activation function\n */\nexport class LogSoftmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'logSoftmax';\n  /**\n   * Calculate the activation function of log softmax:\n   * log( exp(x_i) / sum(exp(x)) )\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.logSoftmax(x, axis);\n  }\n}\nserialization.registerClass(LogSoftmax);\n\n/**\n * Swish activation function\n */\nexport class Swish extends Activation {\n  /** @nocollapse */\n  static readonly className = 'swish';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param alpha Scaling factor for the sigmoid function.\n   * @returns a Tensor of the same shape as x\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return tidy(() => tfc.sigmoid(x.mul(alpha)).mul(x));\n  }\n}\nserialization.registerClass(Swish);\n\nexport function serializeActivation(activation: Activation): string {\n  return activation.getClassName();\n}\n\nexport function deserializeActivation(\n   config: serialization.ConfigDict,\n   customObjects: serialization.ConfigDict = {}): Activation {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'activation');\n}\n\nexport function getActivation(identifier: ActivationIdentifier|\n                              serialization.ConfigDict|Activation): Activation {\n  if (identifier == null) {\n    const config: serialization.ConfigDict = {};\n    config['className'] = 'linear';\n    config['config'] = {};\n    return deserializeActivation(config);\n  }\n  if (typeof identifier === 'string') {\n    const config: serialization.ConfigDict = {};\n    config['className'] = identifier;\n    config['config'] = {};\n    return deserializeActivation(config);\n  } else if (identifier instanceof Activation) {\n    return identifier;\n  } else {\n    return deserializeActivation(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* original source: keras/regularizers.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {abs, add, Scalar, serialization, sum, Tensor, tidy, zeros} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\nfunction assertObjectArgs(args: L1Args | L2Args | L1L2Args): void {\n  if (args != null && typeof args !== 'object') {\n    throw new Error(\n        `Argument to L1L2 regularizer's constructor is expected to be an ` +\n        `object, but received: ${args}`);\n  }\n}\n\n/**\n * Regularizer base class.\n */\nexport abstract class Regularizer extends serialization.Serializable {\n  abstract apply(x: Tensor): Scalar;\n}\n\nexport interface L1L2Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1?: number;\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2?: number;\n}\n\nexport interface L1Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1: number;\n}\n\nexport interface L2Args {\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2: number;\n}\n\nexport class L1L2 extends Regularizer {\n  /** @nocollapse */\n  static className = 'L1L2';\n\n  private readonly l1: number;\n  private readonly l2: number;\n  private readonly hasL1: boolean;\n  private readonly hasL2: boolean;\n  constructor(args?: L1L2Args) {\n    super();\n\n    assertObjectArgs(args);\n\n    this.l1 = args == null || args.l1 == null ? 0.01 : args.l1;\n    this.l2 = args == null || args.l2 == null ? 0.01 : args.l2;\n    this.hasL1 = this.l1 !== 0;\n    this.hasL2 = this.l2 !== 0;\n  }\n\n  /**\n   * Porting note: Renamed from __call__.\n   * @param x Variable of which to calculate the regularization score.\n   */\n  apply(x: Tensor): Scalar {\n    return tidy(() => {\n      let regularization: Tensor = zeros([1]);\n      if (this.hasL1) {\n        regularization = add(regularization, sum(tfc.mul(this.l1, abs(x))));\n      }\n      if (this.hasL2) {\n        regularization =\n            add(regularization, sum(tfc.mul(this.l2, K.square(x))));\n      }\n      return regularization.asScalar();\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {'l1': this.l1, 'l2': this.l2};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls({l1: config['l1'] as number, l2: config['l2'] as number});\n  }\n}\nserialization.registerClass(L1L2);\n\nexport function l1(args?: L1Args) {\n  assertObjectArgs(args);\n  return new L1L2({l1: args != null ? args.l1 : null, l2: 0});\n}\n\nexport function l2(args: L2Args) {\n  assertObjectArgs(args);\n  return new L1L2({l2: args != null ? args.l2 : null, l1: 0});\n}\n\n/** @docinline */\nexport type RegularizerIdentifier = 'l1l2'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding keras symbols.\nexport const REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in RegularizerIdentifier]: string} = {\n      'l1l2': 'L1L2'\n    };\n\nexport function serializeRegularizer(constraint: Regularizer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeRegularizer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Regularizer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'regularizer');\n}\n\nexport function getRegularizer(identifier: RegularizerIdentifier|\n                               serialization.ConfigDict|\n                               Regularizer): Regularizer {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeRegularizer(config);\n  } else if (identifier instanceof Regularizer) {\n    return identifier;\n  } else {\n    return deserializeRegularizer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n *  Advanced activation layers.\n */\n\nimport {clipByValue, elu, leakyRelu, prelu, relu, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {Softmax as softmaxActivation} from '../activations';\nimport {cast} from '../backend/tfjs_backend';\nimport {Constraint, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface ReLULayerArgs extends LayerArgs {\n  /**\n   * Float, the maximum output value.\n   */\n  maxValue?: number;\n}\n\nexport class ReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ReLU';\n  maxValue: number;\n\n  constructor(args?: ReLULayerArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maxValue = args.maxValue;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    let output = relu(inputs);\n    if (this.maxValue != null) {\n      output = clipByValue(output, 0, this.maxValue);\n    }\n    return output;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {maxValue: this.maxValue};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ReLU);\n\nexport declare interface LeakyReLULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `0.3`.\n   */\n  alpha?: number;\n}\n\nexport class LeakyReLU extends Layer {\n  /** @nocollapse */\n  static className = 'LeakyReLU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 0.3;\n\n  constructor(args?: LeakyReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return leakyRelu(x, this.alpha);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LeakyReLU);\n\nexport declare interface PReLULayerArgs extends LayerArgs {\n  /**\n   * Initializer for the learnable alpha.\n   */\n  alphaInitializer?: Initializer|InitializerIdentifier;\n\n  /**\n   * Regularizer for the learnable alpha.\n   */\n  alphaRegularizer?: Regularizer;\n\n  /**\n   * Constraint for the learnable alpha.\n   */\n  alphaConstraint?: Constraint;\n\n  /**\n   * The axes along which to share learnable parameters for the activation\n   * function. For example, if the incoming feature maps are from a 2D\n   * convolution with output shape `[numExamples, height, width, channels]`,\n   * and you wish to share parameters across space (height and width) so that\n   * each filter channels has only one set of parameters, set\n   * `shared_axes: [1, 2]`.\n   */\n  sharedAxes?: number|number[];\n}\n\nexport class PReLU extends Layer {\n  /** @nocollapse */\n  static className = 'PReLU';\n  private readonly alphaInitializer: Initializer;\n  private readonly alphaRegularizer: Regularizer;\n  private readonly alphaConstraint: Constraint;\n  private readonly sharedAxes: number[];\n  private alpha: LayerVariable;\n\n  readonly DEFAULT_ALPHA_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args?: PReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.supportsMasking = true;\n    this.alphaInitializer =\n        getInitializer(args.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER);\n    this.alphaRegularizer = getRegularizer(args.alphaRegularizer);\n    this.alphaConstraint = getConstraint(args.alphaConstraint);\n    if (args.sharedAxes == null) {\n      this.sharedAxes = null;\n    } else if (Array.isArray(args.sharedAxes)) {\n      this.sharedAxes = args.sharedAxes;\n    } else if (typeof args.sharedAxes === 'number') {\n      this.sharedAxes = [args.sharedAxes];\n    } else {\n      throw new ValueError(\n          `Expected sharedAxes to be a number or an array of numbers, ` +\n          `but got ${args.sharedAxes}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]) {\n    inputShape = getExactlyOneShape(inputShape);\n    const paramShape: Shape = inputShape.slice(1);\n    if (this.sharedAxes != null) {\n      for (const i of this.sharedAxes) {\n        paramShape[i - 1] = 1;\n      }\n    }\n    this.alpha = this.addWeight(\n        'alpha', paramShape, 'float32', this.alphaInitializer,\n        this.alphaRegularizer, true, this.alphaConstraint);\n    // Set input spec.\n    const axes: {[axis: number]: number} = {};\n    if (this.sharedAxes != null) {\n      for (let i = 1; i < inputShape.length; ++i) {\n        axes[i] = inputShape[i];\n      }\n    }\n    this.inputSpec = [new InputSpec({\n      ndim: inputShape.length,\n      axes,\n    })];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    return prelu(inputs, this.alpha.read());\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      alphaInitializer: serializeInitializer(this.alphaInitializer),\n      alphaRegularizer: serializeRegularizer(this.alphaRegularizer),\n      alphaConstraint: serializeConstraint(this.alphaConstraint),\n      sharedAxes: this.sharedAxes\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(PReLU);\n\nexport declare interface ELULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `1.0`.\n   */\n  alpha?: number;\n}\n\nexport class ELU extends Layer {\n  /** @nocollapse */\n  static className = 'ELU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 1.0;\n\n  constructor(args?: ELULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    if (args.alpha != null && args.alpha !== this.DEFAULT_ALPHA) {\n      throw new NotImplementedError(\n          `Non-default alpha value (${args.alpha}) is not supported by the ` +\n          `ELU layer yet.`);\n    }\n\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return elu(x);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ELU);\n\nexport declare interface ThresholdedReLULayerArgs extends LayerArgs {\n  /**\n   * Float >= 0. Threshold location of activation.\n   */\n  theta?: number;\n}\n\nexport class ThresholdedReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ThresholdedReLU';\n  readonly theta: number;\n\n  readonly DEFAULT_THETA = 1.0;\n\n  constructor(args?: ThresholdedReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.theta = args.theta == null ? this.DEFAULT_THETA : args.theta;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return x.mul(cast(x.greater(this.theta), 'float32'));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {theta: this.theta};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ThresholdedReLU);\n\nexport declare interface SoftmaxLayerArgs extends LayerArgs {\n  /**\n   * Integer, axis along which the softmax normalization is applied.\n   * Defaults to `-1` (i.e., the last axis).\n   */\n  axis?: number;\n}\n\nexport class Softmax extends Layer {\n  /** @nocollapse */\n  static className = 'Softmax';\n  readonly axis: number;\n  readonly softmax: (t: Tensor, a?: number) => Tensor;\n  readonly DEFAULT_AXIS = 1.0;\n\n  constructor(args?: SoftmaxLayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.softmax = new softmaxActivation().apply;\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return this.softmax(x, this.axis);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {axis: this.axis};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Softmax);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {ValueError} from '../errors';\nimport {PaddingMode} from '../keras_format/common';\n\nimport {pyListRepeat} from './generic_utils';\nimport {isInteger, max} from './math_utils';\n\n/**\n * Transforms a single number of array of numbers into an array of numbers.\n * @param value\n * @param n: The size of the tuple to be returned.\n * @param name: Name of the parameter, used for generating error messages.\n * @returns An array of numbers.\n */\nexport function normalizeArray(\n    value: number|number[], n: number, name: string): number[] {\n  if (typeof value === 'number') {\n    return pyListRepeat(value, n);\n  } else {\n    if (value.length !== n) {\n      throw new ValueError(\n          `The ${name} argument must be an integer or tuple of ${n} integers.` +\n          ` Received: ${value.length} elements.`);\n    }\n    for (let i = 0; i < n; ++i) {\n      const singleValue = value[i];\n      if (!isInteger(singleValue)) {\n        throw new ValueError(\n            `The ${name} argument must be an integer or tuple of ${n}` +\n            ` integers. Received: ${JSON.stringify(value)} including a` +\n            ` non-integer number ${singleValue}`);\n      }\n    }\n    return value;\n  }\n}\n\n/**\n * Determines output length of a convolution given input length.\n * @param inputLength\n * @param filterSize\n * @param padding\n * @param stride\n * @param dilation: dilation rate.\n */\nexport function convOutputLength(\n    inputLength: number, filterSize: number, padding: PaddingMode,\n    stride: number, dilation = 1): number {\n  if (inputLength == null) {\n    return inputLength;\n  }\n  const dilatedFilterSize = filterSize + (filterSize - 1) * (dilation - 1);\n  let outputLength: number;\n  if (padding === 'same') {\n    outputLength = inputLength;\n  } else {  // VALID\n    outputLength = inputLength - dilatedFilterSize + 1;\n  }\n  return Math.floor((outputLength + stride - 1) / stride);\n}\n\nexport function deconvLength(\n    dimSize: number, strideSize: number, kernelSize: number,\n    padding: PaddingMode): number {\n  if (dimSize == null) {\n    return null;\n  }\n\n  if (padding === 'valid') {\n    dimSize = dimSize * strideSize + max([kernelSize - strideSize, 0]);\n  } else if (padding === 'same') {\n    dimSize = dimSize * strideSize;\n  } else {\n    throw new ValueError(`Unsupport padding mode: ${padding}.`);\n  }\n  return dimSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {fused, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, PaddingMode, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, deconvLength, normalizeArray} from '../utils/conv_utils';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Transpose and cast the input before the conv2d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv2DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  // TODO(cais): Cast type to float32 if not.\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 1]);  // NCHW -> NHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * Transpose and cast the input before the conv3d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv3DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 4, 1]);  // NCDHW -> NDHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * 1D-convolution with bias added.\n *\n * Porting Note: This function does not exist in the Python Keras backend.\n *   It is exactly the same as `conv2d`, except the added `bias`.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.\n * @param bias Bias, rank-3, of shape `[outDepth]`.\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    // Check the ranks of x, kernel and bias.\n    if (x.shape.length !== 3) {\n      throw new ValueError(\n          `The input of a conv1dWithBias operation should be 3, but is ` +\n          `${x.shape.length} instead.`);\n    }\n    if (kernel.shape.length !== 3) {\n      throw new ValueError(\n          `The kernel for a conv1dWithBias operation should be 3, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    if (bias != null && bias.shape.length !== 1) {\n      throw new ValueError(\n          `The bias for a conv1dWithBias operation should be 1, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    // TODO(cais): Support CAUSAL padding mode.\n    if (dataFormat === 'channelsFirst') {\n      x = tfc.transpose(x, [0, 2, 1]);  // NCW -> NWC.\n    }\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    let y: Tensor = tfc.conv1d(\n        x as Tensor2D | Tensor3D, kernel as Tensor3D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias);\n    }\n    return y;\n  });\n}\n\n/**\n * 1D-convolution.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.s\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1d(\n    x: Tensor, kernel: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv1dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 2D pooling.\n */\nexport function conv2d(\n    x: Tensor, kernel: Tensor, strides = [1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv2dWithBiasActivation(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution with an added bias and optional activation.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv2d`, except the added `bias`.\n */\nexport function conv2dWithBiasActivation(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat, dilationRate?: [number, number],\n    activation: fused.Activation = null): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 3 && x.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects input to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    if (kernel.rank !== 3 && kernel.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects kernel to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.fused.conv2d({\n      x: y as Tensor3D | Tensor4D,\n      filter: kernel as Tensor4D,\n      strides: strides as [number, number],\n      pad: padding === 'same' ? 'same' : 'valid',\n      dilations: dilationRate,\n      dataFormat: 'NHWC',\n      bias,\n      activation\n    });\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\n/**\n * 3D Convolution.\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 3D convolution.\n */\nexport function conv3d(\n    x: Tensor, kernel: Tensor, strides = [1, 1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv3dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 3D Convolution with an added bias.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv3d`, except the added `bias`.\n */\nexport function conv3dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 4 && x.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects input to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    if (kernel.rank !== 4 && kernel.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects kernel to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    let y = preprocessConv3DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv3dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.conv3d(\n        y as Tensor4D | tfc.Tensor<tfc.Rank.R5>,\n        kernel as tfc.Tensor<tfc.Rank.R5>, strides as [number, number, number],\n        padding === 'same' ? 'same' : 'valid', 'NDHWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias as Tensor1D);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);\n    }\n    return y;\n  });\n}\n\n/**\n * Base LayerConfig for depthwise and non-depthwise convolutional layers.\n */\nexport declare interface BaseConvLayerArgs extends LayerArgs {\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number]|[number, number, number];\n\n  /**\n   * Activation function of the layer.\n   *\n   * If you don't specify the activation, none is applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector. Defaults to `true`.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the convolutional kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the convolutional kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\n/**\n * LayerConfig for non-depthwise convolutional layers.\n * Applies to non-depthwise convolution of all ranks (e.g, Conv1D, Conv2D,\n * Conv3D).\n */\nexport declare interface ConvLayerArgs extends BaseConvLayerArgs {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n}\n\n/**\n * Abstract convolution layer.\n */\nexport abstract class BaseConv extends Layer {\n  protected readonly rank: number;\n  protected readonly kernelSize: number[];\n  protected readonly strides: number[];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n  protected readonly activation: Activation;\n  protected readonly useBias: boolean;\n  protected readonly dilationRate: number[];\n\n  // Bias-related members are here because all convolution subclasses use the\n  // same configuration parmeters to control bias.  Kernel-related members\n  // are in subclass `Conv` because some subclasses use different parameters to\n  // control kernel properties, for instance, `DepthwiseConv2D` uses\n  // `depthwiseInitializer` instead of `kernelInitializer`.\n  protected readonly biasInitializer?: Initializer;\n  protected readonly biasConstraint?: Constraint;\n  protected readonly biasRegularizer?: Regularizer;\n\n  protected bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(rank: number, args: BaseConvLayerArgs) {\n    super(args as LayerArgs);\n    BaseConv.verifyArgs(args);\n    this.rank = rank;\n    generic_utils.assertPositiveInteger(this.rank, 'rank');\n    if (this.rank !== 1 && this.rank !== 2 && this.rank !== 3) {\n      throw new NotImplementedError(\n          `Convolution layer for rank other than 1, 2, or 3 (${\n              this.rank}) is ` +\n          `not implemented yet.`);\n    }\n    this.kernelSize = normalizeArray(args.kernelSize, rank, 'kernelSize');\n    this.strides = normalizeArray(\n        args.strides == null ? 1 : args.strides, rank, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.activation = getActivation(args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.dilationRate = normalizeArray(\n        args.dilationRate == null ? 1 : args.dilationRate, rank,\n        'dilationRate');\n    if (this.rank === 1 &&\n        (Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)) {\n      throw new ValueError(\n          `dilationRate must be a number or an array of a single number ` +\n          `for 1D convolution, but received ` +\n          `${JSON.stringify(this.dilationRate)}`);\n    } else if (this.rank === 2) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate = [this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 2) {\n        throw new ValueError(\n            `dilationRate must be a number or array of two numbers for 2D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    } else if (this.rank === 3) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate =\n            [this.dilationRate, this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 3) {\n        throw new ValueError(\n            `dilationRate must be a number or array of three numbers for 3D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    }\n  }\n\n  protected static verifyArgs(args: BaseConvLayerArgs) {\n    // Check config.kernelSize type and shape.\n    generic_utils.assert(\n        'kernelSize' in args, `required key 'kernelSize' not in config`);\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 3)) {\n      throw new ValueError(\n          `BaseConv expects config.kernelSize to be number or number[] with ` +\n          `length 1, 2, or 3, but received ${\n              JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      kernelSize: this.kernelSize,\n      strides: this.strides,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\n/**\n * Abstract nD convolution layer.  Ancestor of convolution layers which reduce\n * across channels, i.e., Conv1D and Conv2D, but not DepthwiseConv2D.\n */\nexport abstract class Conv extends BaseConv {\n  protected readonly filters: number;\n\n  protected kernel: LayerVariable = null;\n\n  // Bias-related properties are stored in the superclass `BaseConv` because all\n  // convolution subclasses use the same configuration parameters to control\n  // bias. Kernel-related properties are defined here rather than in the\n  // superclass because some convolution subclasses use different names and\n  // configuration parameters for their internal kernel state.\n  protected readonly kernelInitializer?: Initializer;\n  protected readonly kernelConstraint?: Constraint;\n  protected readonly kernelRegularizer?: Regularizer;\n\n  constructor(rank: number, args: ConvLayerArgs) {\n    super(rank, args as BaseConvLayerArgs);\n    Conv.verifyArgs(args);\n    this.filters = args.filters;\n    generic_utils.assertPositiveInteger(this.filters, 'filters');\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n    const inputDim = inputShape[channelAxis];\n\n    const kernelShape = this.kernelSize.concat([inputDim, this.filters]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.inputSpec = [{ndim: this.rank + 2, axes: {[channelAxis]: inputDim}}];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs: Tensor;\n      const biasValue = this.bias == null ? null : this.bias.read();\n      const fusedActivationName = generic_utils.mapActivationToFusedKernel(\n                                            this.activation.getClassName());\n\n      if (fusedActivationName != null && this.rank === 2) {\n        outputs = conv2dWithBiasActivation(\n            inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n            this.dataFormat, this.dilationRate as [number, number],\n            fusedActivationName);\n      } else {\n        if (this.rank === 1) {\n          outputs = conv1dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides[0],\n              this.padding, this.dataFormat, this.dilationRate[0]);\n        } else if (this.rank === 2) {\n          // TODO(cais): Move up to constructor.\n          outputs = conv2dWithBiasActivation(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number]);\n        } else if (this.rank === 3) {\n          outputs = conv3dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number, number]);\n        } else {\n          throw new NotImplementedError(\n              'convolutions greater than 3D are not implemented yet.');\n        }\n\n        if (this.activation != null) {\n          outputs = this.activation.apply(outputs);\n        }\n      }\n\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const newSpace: number[] = [];\n    const space = (this.dataFormat === 'channelsLast') ?\n        inputShape.slice(1, inputShape.length - 1) :\n        inputShape.slice(2);\n    for (let i = 0; i < space.length; ++i) {\n      const newDim = convOutputLength(\n          space[i], this.kernelSize[i], this.padding, this.strides[i],\n          typeof this.dilationRate === 'number' ? this.dilationRate :\n                                                  this.dilationRate[i]);\n      newSpace.push(newDim);\n    }\n\n    let outputShape = [inputShape[0]];\n    if (this.dataFormat === 'channelsLast') {\n      outputShape = outputShape.concat(newSpace);\n      outputShape.push(this.filters);\n    } else {\n      outputShape.push(this.filters);\n      outputShape = outputShape.concat(newSpace);\n    }\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      filters: this.filters,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // Check config.filters type, shape, and value.\n    if (!('filters' in args) || typeof args.filters !== 'number' ||\n        args.filters < 1) {\n      throw new ValueError(\n          `Convolution layer expected config.filters to be a 'number' > 0 ` +\n          `but got ${JSON.stringify(args.filters)}`);\n    }\n  }\n}\n\nexport class Conv2D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv2D';\n  constructor(args: ConvLayerArgs) {\n    super(2, args);\n    Conv2D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if ((typeof args.kernelSize !== 'number') &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 2)) {\n      throw new ValueError(\n          `Conv2D expects config.kernelSize to be number or number[] with ` +\n          `length 1 or 2, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv2D);\n\nexport class Conv3D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv3D';\n  constructor(args: ConvLayerArgs) {\n    super(3, args);\n    Conv3D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number') {\n      if (!(Array.isArray(args.kernelSize) &&\n            (args.kernelSize.length === 1 || args.kernelSize.length === 3))) {\n        throw new ValueError(\n            `Conv3D expects config.kernelSize to be number or` +\n            ` [number, number, number], but received ${\n                JSON.stringify(args.kernelSize)}.`);\n      }\n    }\n  }\n}\nserialization.registerClass(Conv3D);\n\nexport class Conv2DTranspose extends Conv2D {\n  /** @nocollapse */\n  static className = 'Conv2DTranspose';\n  inputSpec: InputSpec[];\n\n  constructor(args: ConvLayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n\n    if (this.padding !== 'same' && this.padding !== 'valid') {\n      throw new ValueError(\n          `Conv2DTranspose currently supports only padding modes 'same' ` +\n          `and 'valid', but received padding mode ${this.padding}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    if (inputShape.length !== 4) {\n      throw new ValueError(\n          'Input should have rank 4; Received input shape: ' +\n          JSON.stringify(inputShape));\n    }\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          'The channel dimension of the inputs should be defined. ' +\n          'Found `None`.');\n    }\n    const inputDim = inputShape[channelAxis];\n    const kernelShape = this.kernelSize.concat([this.filters, inputDim]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, 'float32', this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    // Set input spec.\n    this.inputSpec =\n        [new InputSpec({ndim: 4, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs);\n      if (input.shape.length !== 4) {\n        throw new ValueError(\n            `Conv2DTranspose.call() expects input tensor to be rank-4, but ` +\n            `received a tensor of rank-${input.shape.length}`);\n      }\n\n      const inputShape = input.shape;\n      const batchSize = inputShape[0];\n\n      let hAxis: number;\n      let wAxis: number;\n      if (this.dataFormat === 'channelsFirst') {\n        hAxis = 2;\n        wAxis = 3;\n      } else {\n        hAxis = 1;\n        wAxis = 2;\n      }\n\n      const height = inputShape[hAxis];\n      const width = inputShape[wAxis];\n      const kernelH = this.kernelSize[0];\n      const kernelW = this.kernelSize[1];\n      const strideH = this.strides[0];\n      const strideW = this.strides[1];\n\n      // Infer the dynamic output shape.\n      const outHeight = deconvLength(height, strideH, kernelH, this.padding);\n      const outWidth = deconvLength(width, strideW, kernelW, this.padding);\n\n      // Porting Note: We don't branch based on `this.dataFormat` here,\n      // because\n      //   the tjfs-core function `conv2dTranspose` called below always\n      //   assumes channelsLast.\n      const outputShape: [number, number, number, number] =\n          [batchSize, outHeight, outWidth, this.filters];\n\n      if (this.dataFormat !== 'channelsLast') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n      }\n      let outputs = tfc.conv2dTranspose(\n          input as Tensor4D, this.kernel.read() as Tensor4D, outputShape,\n          this.strides as [number, number], this.padding as 'same' | 'valid');\n      if (this.dataFormat !== 'channelsLast') {\n        outputs = tfc.transpose(outputs, [0, 3, 1, 2]);\n      }\n\n      if (this.bias != null) {\n        outputs =\n            K.biasAdd(outputs, this.bias.read(), this.dataFormat) as Tensor4D;\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs) as Tensor4D;\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n\n    let channelAxis: number;\n    let heightAxis: number;\n    let widthAxis: number;\n    if (this.dataFormat === 'channelsFirst') {\n      channelAxis = 1;\n      heightAxis = 2;\n      widthAxis = 3;\n    } else {\n      channelAxis = 3;\n      heightAxis = 1;\n      widthAxis = 2;\n    }\n\n    const kernelH = this.kernelSize[0];\n    const kernelW = this.kernelSize[1];\n    const strideH = this.strides[0];\n    const strideW = this.strides[1];\n\n    outputShape[channelAxis] = this.filters;\n    outputShape[heightAxis] =\n        deconvLength(outputShape[heightAxis], strideH, kernelH, this.padding);\n    outputShape[widthAxis] =\n        deconvLength(outputShape[widthAxis], strideW, kernelW, this.padding);\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['dilationRate'];\n    return config;\n  }\n}\nserialization.registerClass(Conv2DTranspose);\n\nexport declare interface SeparableConvLayerArgs extends ConvLayerArgs {\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal\n   * to `filtersIn * depthMultiplier`. Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the pointwise kernel matrix.\n   */\n  pointwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the pointwise kernel matrix.\n   */\n  pointwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the pointwise kernel matrix.\n   */\n  pointwiseConstraint?: ConstraintIdentifier|Constraint;\n}\n\nexport class SeparableConv extends Conv {\n  /** @nocollapse */\n  static className = 'SeparableConv';\n\n  readonly depthMultiplier: number;\n\n  protected readonly depthwiseInitializer?: Initializer;\n  protected readonly depthwiseRegularizer?: Regularizer;\n  protected readonly depthwiseConstraint?: Constraint;\n  protected readonly pointwiseInitializer?: Initializer;\n  protected readonly pointwiseRegularizer?: Regularizer;\n  protected readonly pointwiseConstraint?: Constraint;\n\n  readonly DEFAULT_DEPTHWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n  readonly DEFAULT_POINTWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n\n  protected depthwiseKernel: LayerVariable = null;\n  protected pointwiseKernel: LayerVariable = null;\n\n  constructor(rank: number, config?: SeparableConvLayerArgs) {\n    super(rank, config);\n\n    if (config.filters == null) {\n      throw new ValueError(\n          'The `filters` configuration field is required by SeparableConv, ' +\n          'but is unspecified.');\n    }\n    if (config.kernelInitializer != null || config.kernelRegularizer != null ||\n        config.kernelConstraint != null) {\n      throw new ValueError(\n          'Fields kernelInitializer, kernelRegularizer and kernelConstraint ' +\n          'are invalid for SeparableConv2D. Use depthwiseInitializer, ' +\n          'depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, ' +\n          'pointwiseRegularizer and pointwiseConstraint instead.');\n    }\n    if (config.padding != null && config.padding !== 'same' &&\n        config.padding !== 'valid') {\n      throw new ValueError(\n          `SeparableConv${this.rank}D supports only padding modes: ` +\n          `'same' and 'valid', but received ${JSON.stringify(config.padding)}`);\n    }\n\n    this.depthMultiplier =\n        config.depthMultiplier == null ? 1 : config.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER);\n    this.depthwiseRegularizer = getRegularizer(config.depthwiseRegularizer);\n    this.depthwiseConstraint = getConstraint(config.depthwiseConstraint);\n    this.pointwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER);\n    this.pointwiseRegularizer = getRegularizer(config.pointwiseRegularizer);\n    this.pointwiseConstraint = getConstraint(config.pointwiseConstraint);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < this.rank + 2) {\n      throw new ValueError(\n          `Inputs to SeparableConv${this.rank}D should have rank ` +\n          `${this.rank + 2}, but received input shape: ` +\n          `${JSON.stringify(inputShape)}`);\n    }\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          `The channel dimension of the inputs should be defined, ` +\n          `but found ${JSON.stringify(inputShape[channelAxis])}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape =\n        this.kernelSize.concat([inputDim, this.depthMultiplier]);\n    const pointwiseKernelShape = [];\n    for (let i = 0; i < this.rank; ++i) {\n      pointwiseKernelShape.push(1);\n    }\n    pointwiseKernelShape.push(inputDim * this.depthMultiplier, this.filters);\n\n    const trainable = true;\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, 'float32',\n        this.depthwiseInitializer, this.depthwiseRegularizer, trainable,\n        this.depthwiseConstraint);\n    this.pointwiseKernel = this.addWeight(\n        'pointwise_kernel', pointwiseKernelShape, 'float32',\n        this.pointwiseInitializer, this.pointwiseRegularizer, trainable,\n        this.pointwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, trainable, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n\n    this.inputSpec =\n        [new InputSpec({ndim: this.rank + 2, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      let output: Tensor;\n      if (this.rank === 1) {\n        throw new NotImplementedError(\n            '1D separable convolution is not implemented yet.');\n      } else if (this.rank === 2) {\n        if (this.dataFormat === 'channelsFirst') {\n          inputs = tfc.transpose(inputs, [0, 2, 3, 1]);  // NCHW -> NHWC.\n        }\n\n        output = tfc.separableConv2d(\n            inputs as Tensor4D, this.depthwiseKernel.read() as Tensor4D,\n            this.pointwiseKernel.read() as Tensor4D,\n            this.strides as [number, number], this.padding as 'same' | 'valid',\n            this.dilationRate as [number, number], 'NHWC');\n      }\n\n      if (this.useBias) {\n        output = K.biasAdd(output, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      if (this.dataFormat === 'channelsFirst') {\n        output = tfc.transpose(output, [0, 3, 1, 2]);  // NHWC -> NCHW.\n      }\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['kernelInitializer'];\n    delete config['kernelRegularizer'];\n    delete config['kernelConstraint'];\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['pointwiseInitializer'] =\n        serializeInitializer(this.pointwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['pointwiseRegularizer'] =\n        serializeRegularizer(this.pointwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseConstraint);\n    config['pointwiseConstraint'] =\n        serializeConstraint(this.pointwiseConstraint);\n    return config;\n  }\n}\n\nexport class SeparableConv2D extends SeparableConv {\n  /** @nocollapse */\n  static className = 'SeparableConv2D';\n  constructor(args?: SeparableConvLayerArgs) {\n    super(2, args);\n  }\n}\nserialization.registerClass(SeparableConv2D);\n\nexport class Conv1D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv1D';\n  constructor(args: ConvLayerArgs) {\n    super(1, args);\n    Conv1D.verifyArgs(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['dataFormat'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 1)) {\n      throw new ValueError(\n          `Conv1D expects config.kernelSize to be number or number[] with ` +\n          `length 1, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv1D);\n\nexport declare interface Cropping2DLayerArgs extends LayerArgs {\n  /**\n   * Dimension of the cropping along the width and the height.\n   * - If integer: the same symmetric cropping\n   *  is applied to width and height.\n   * - If list of 2 integers:\n   *   interpreted as two different\n   *   symmetric cropping values for height and width:\n   *   `[symmetric_height_crop, symmetric_width_crop]`.\n   * - If a list of 2 list of 2 integers:\n   *   interpreted as\n   *   `[[top_crop, bottom_crop], [left_crop, right_crop]]`\n   */\n  cropping: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   * `channels_first` corresponds to inputs with shape\n   *   `(batch, channels, ...)`\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class Cropping2D extends Layer {\n  /** @nocollapse */\n  static className = 'Cropping2D';\n  protected readonly cropping: [[number, number], [number, number]];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Cropping2DLayerArgs) {\n    super(args);\n    if (typeof args.cropping === 'number') {\n      this.cropping =\n          [[args.cropping, args.cropping], [args.cropping, args.cropping]];\n    } else if (typeof args.cropping[0] === 'number') {\n      this.cropping = [\n        [args.cropping[0], args.cropping[0]],\n        [args.cropping[1] as number, args.cropping[1] as number]\n      ];\n    } else {\n      this.cropping = args.cropping as [[number, number], [number, number]];\n    }\n    this.dataFormat =\n        args.dataFormat === undefined ? 'channelsLast' : args.dataFormat;\n    this.inputSpec = [{ndim: 4}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      return [\n        inputShape[0], inputShape[1],\n        inputShape[2] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[3] - this.cropping[1][0] - this.cropping[1][1]\n      ];\n    } else {\n      return [\n        inputShape[0],\n        inputShape[1] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[2] - this.cropping[1][0] - this.cropping[1][1], inputShape[3]\n      ];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      if (this.dataFormat === 'channelsLast') {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[1] - this.cropping[0][0] - this.cropping[0][1], 2);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[2] - this.cropping[1][1] - this.cropping[1][0], 3);\n      } else {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[2] - this.cropping[0][0] - this.cropping[0][1], 3);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[3] - this.cropping[1][1] - this.cropping[1][0], 4);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {cropping: this.cropping, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Cropping2D);\n\nexport declare interface UpSampling2DLayerArgs extends LayerArgs {\n  /**\n   * The upsampling factors for rows and columns.\n   *\n   * Defaults to `[2, 2]`.\n   */\n  size?: number[];\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `\"channelsLast\"` corresponds to inputs with shape\n   *   `[batch, ..., channels]`\n   *\n   *  `\"channelsFirst\"` corresponds to inputs with shape `[batch, channels,\n   * ...]`.\n   *\n   * Defaults to `\"channelsLast\"`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class UpSampling2D extends Layer {\n  /** @nocollapse */\n  static className = 'UpSampling2D';\n  protected readonly DEFAULT_SIZE = [2, 2];\n  protected readonly size: number[];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: UpSampling2DLayerArgs) {\n    super(args);\n    this.inputSpec = [{ndim: 4}];\n    this.size = args.size == null ? this.DEFAULT_SIZE : args.size;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      const height =\n          inputShape[2] == null ? null : this.size[0] * inputShape[2];\n      const width = inputShape[3] == null ? null : this.size[1] * inputShape[3];\n      return [inputShape[0], inputShape[1], height, width];\n    } else {\n      const height =\n          inputShape[1] == null ? null : this.size[0] * inputShape[1];\n      const width = inputShape[2] == null ? null : this.size[1] * inputShape[2];\n      return [inputShape[0], height, width, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs) as Tensor4D;\n      const inputShape = input.shape;\n\n      if (this.dataFormat === 'channelsFirst') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n        const height = this.size[0] * inputShape[2];\n        const width = this.size[1] * inputShape[3];\n        const resized = input.resizeNearestNeighbor([height, width]);\n        return tfc.transpose(resized, [0, 3, 1, 2]);\n      } else {\n        const height = this.size[0] * inputShape[1];\n        const width = this.size[1] * inputShape[2];\n        return input.resizeNearestNeighbor([height, width]);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {size: this.size, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(UpSampling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Depthwise Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {BaseConv, BaseConvLayerArgs, ConvLayerArgs, preprocessConv2DInput} from './convolutional';\n\n/**\n * 2D convolution with separable filters.\n * @param x Input tensor.\n * @param depthwiseKernel Convolution kernel for depthwise convolution.\n * @param strides Strides (Array of two integers).\n * @param padding Padding model.\n * @param dataFormat Data format.\n * @param dilationRate Array of two integers, dilation rates for the separable\n *   convolution.\n * @returns Output tensor.\n * @throws ValueError If depthwiseKernel is not a 4D array.\n */\nexport function depthwiseConv2d(\n    x: Tensor, depthwiseKernel: Tensor, strides: [number, number] = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `Input for depthwiseConv2d is required to be 4-D, but is instead ` +\n          `${x.rank}-D`);\n    }\n    if (depthwiseKernel.rank !== 4) {\n      throw new ValueError(\n          `depthwiseKernel is required to be 4-D, but is instead ` +\n          `${depthwiseKernel.rank}-D`);\n    }\n    y = tfc.depthwiseConv2d(\n        y as Tensor4D, depthwiseKernel as Tensor4D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NHWC', dilationRate);\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\nexport declare interface DepthwiseConv2DLayerArgs extends BaseConvLayerArgs {\n  /**\n   * An integer or Array of 2 integers, specifying the width and height of the\n   * 2D convolution window. Can be a single integer to specify the same value\n   * for all spatial dimensions.\n   */\n  kernelSize: number|[number, number];\n\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal to\n   * `filtersIn * depthMultiplier`.\n   * Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   * Default: GlorotNormal.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regulzarizer function for the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class DepthwiseConv2D extends BaseConv {\n  /** @nocollapse */\n  static className = 'DepthwiseConv2D';\n  private readonly depthMultiplier: number;\n  private readonly depthwiseInitializer: Initializer;\n  private readonly depthwiseConstraint: Constraint;\n  private readonly depthwiseRegularizer: Regularizer;\n\n  private depthwiseKernel: LayerVariable = null;\n\n  constructor(args: DepthwiseConv2DLayerArgs) {\n    super(2, args as ConvLayerArgs);\n    this.depthMultiplier =\n        args.depthMultiplier == null ? 1 : args.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        args.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.depthwiseConstraint = getConstraint(args.depthwiseConstraint);\n    this.depthwiseRegularizer = getRegularizer(args.depthwiseRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 4) {\n      throw new ValueError(\n          `Inputs to DepthwiseConv2D should have rank 4. ` +\n          `Received input shape: ${JSON.stringify(inputShape)}.`);\n    }\n    const channelAxis = this.dataFormat === 'channelsFirst' ? 1 : 3;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          'The channel dimension of the inputs to DepthwiseConv2D should ' +\n          `be defined, but is not (${inputShape[channelAxis]}).`);\n    }\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape: Shape = [\n      this.kernelSize[0], this.kernelSize[1], inputDim, this.depthMultiplier\n    ];\n\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, null,\n        this.depthwiseInitializer, this.depthwiseRegularizer, true,\n        this.depthwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [inputDim * this.depthMultiplier], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs = depthwiseConv2d(\n          inputs, this.depthwiseKernel.read(), this.strides as [number, number],\n          this.padding, this.dataFormat, null);\n      // TODO(cais): Add support for dilation.\n      if (this.useBias) {\n        outputs = K.biasAdd(outputs, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs);\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    const cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    const outFilters = this.dataFormat === 'channelsFirst' ?\n        inputShape[1] * this.depthMultiplier :\n        inputShape[3] * this.depthMultiplier;\n    const outRows = convOutputLength(\n        rows, this.kernelSize[0], this.padding, this.strides[0]);\n    const outCols = convOutputLength(\n        cols, this.kernelSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], outFilters, outRows, outCols];\n    } else {\n      // In this case, assume 'channelsLast'.\n      return [inputShape[0], outRows, outCols, outFilters];\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    config['depthMultiplier'] = this.depthMultiplier;\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseRegularizer);\n    return config;\n  }\n}\nserialization.registerClass(DepthwiseConv2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Recurrent Neural Network Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, SymbolicTensor} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, Ones, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs, RnnStepFunction} from '../types';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor, isArrayOfShapes} from '../utils/types_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\nimport {deserialize} from './serialization';\n\n/**\n * Standardize `apply()` args to a single list of tensor inputs.\n *\n * When running a model loaded from file, the input tensors `initialState` and\n * `constants` are passed to `RNN.apply()` as part of `inputs` instead of the\n * dedicated kwargs fields. `inputs` consists of\n * `[inputs, initialState0, initialState1, ..., constant0, constant1]` in this\n * case.\n * This method makes sure that arguments are\n * separated and that `initialState` and `constants` are `Array`s of tensors\n * (or None).\n *\n * @param inputs Tensor or `Array` of  tensors.\n * @param initialState Tensor or `Array` of tensors or `null`/`undefined`.\n * @param constants Tensor or `Array` of tensors or `null`/`undefined`.\n * @returns An object consisting of\n *   inputs: A tensor.\n *   initialState: `Array` of tensors or `null`.\n *   constants: `Array` of tensors or `null`.\n * @throws ValueError, if `inputs` is an `Array` but either `initialState` or\n *   `constants` is provided.\n */\nexport function standardizeArgs(\n    inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    initialState: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    constants: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    numConstants?: number): {\n  inputs: Tensor|SymbolicTensor,\n  initialState: Tensor[]|SymbolicTensor[],\n  constants: Tensor[]|SymbolicTensor[]\n} {\n  if (Array.isArray(inputs)) {\n    if (initialState != null || constants != null) {\n      throw new ValueError(\n          'When inputs is an array, neither initialState or constants ' +\n          'should be provided');\n    }\n    if (numConstants != null) {\n      constants = inputs.slice(inputs.length - numConstants, inputs.length);\n      inputs = inputs.slice(0, inputs.length - numConstants);\n    }\n    if (inputs.length > 1) {\n      initialState = inputs.slice(1, inputs.length);\n    }\n    inputs = inputs[0];\n  }\n\n  function toListOrNull(x: Tensor|Tensor[]|SymbolicTensor|\n                        SymbolicTensor[]): Tensor[]|SymbolicTensor[] {\n    if (x == null || Array.isArray(x)) {\n      return x as Tensor[] | SymbolicTensor[];\n    } else {\n      return [x] as Tensor[] | SymbolicTensor[];\n    }\n  }\n\n  initialState = toListOrNull(initialState);\n  constants = toListOrNull(constants);\n\n  return {inputs, initialState, constants};\n}\n\n/**\n * Iterates over the time dimension of a tensor.\n *\n * @param stepFunction RNN step function.\n *   Parameters:\n *     inputs: tensor with shape `[samples, ...]` (no time dimension),\n *       representing input for the batch of samples at a certain time step.\n *     states: an Array of tensors.\n *   Returns:\n *     outputs: tensor with shape `[samples, outputDim]` (no time dimension).\n *     newStates: list of tensors, same length and shapes as `states`. The first\n *       state in the list must be the output tensor at the previous timestep.\n * @param inputs Tensor of temporal data of shape `[samples, time, ...]` (at\n *   least 3D).\n * @param initialStates Tensor with shape `[samples, outputDim]` (no time\n *   dimension), containing the initial values of the states used in the step\n *   function.\n * @param goBackwards If `true`, do the iteration over the time dimension in\n *   reverse order and return the reversed sequence.\n * @param mask Binary tensor with shape `[sample, time, 1]`, with a zero for\n *   every element that is masked.\n * @param constants An Array of constant values passed at each step.\n * @param unroll Whether to unroll the RNN or to use a symbolic loop. *Not*\n *   applicable to this imperative deeplearn.js backend. Its value is ignored.\n * @param needPerStepOutputs Whether the per-step outputs are to be\n *   concatenated into a single tensor and returned (as the second return\n *   value). Default: `false`. This arg is included so that the relatively\n *   expensive concatenation of the stepwise outputs can be omitted unless\n *   the stepwise outputs need to be kept (e.g., for an LSTM layer of which\n *   `returnSequence` is `true`.)\n * @returns An Array: `[lastOutput, outputs, newStates]`.\n *   lastOutput: the lastest output of the RNN, of shape `[samples, ...]`.\n *   outputs: tensor with shape `[samples, time, ...]` where each entry\n *     `output[s, t]` is the output of the step function at time `t` for sample\n *     `s`. This return value is provided if and only if the\n *     `needPerStepOutputs` is set as `true`. If it is set as `false`, this\n *     return value will be `undefined`.\n *   newStates: Array of tensors, latest states returned by the step function,\n *      of shape `(samples, ...)`.\n * @throws ValueError If input dimension is less than 3.\n *\n * TODO(nielsene): This needs to be tidy-ed.\n */\nexport function rnn(\n    stepFunction: RnnStepFunction, inputs: Tensor, initialStates: Tensor[],\n    goBackwards = false, mask?: Tensor, constants?: Tensor[], unroll = false,\n    needPerStepOutputs = false): [Tensor, Tensor, Tensor[]] {\n  return tfc.tidy(() => {\n    const ndim = inputs.shape.length;\n    if (ndim < 3) {\n      throw new ValueError(`Input should be at least 3D, but is ${ndim}D.`);\n    }\n\n    // Transpose to time-major, i.e., from [batch, time, ...] to [time, batch,\n    // ...].\n    const axes = [1, 0].concat(math_utils.range(2, ndim));\n    inputs = tfc.transpose(inputs, axes);\n\n    if (constants != null) {\n      throw new NotImplementedError(\n          'The rnn() functoin of the deeplearn.js backend does not support ' +\n          'constants yet.');\n    }\n\n    // Porting Note: the unroll option is ignored by the imperative backend.\n    if (unroll) {\n      console.warn(\n          'Backend rnn(): the unroll = true option is not applicable to the ' +\n          'imperative deeplearn.js backend.');\n    }\n\n    if (mask != null) {\n      mask = mask.asType('bool').asType('float32');\n      if (mask.rank === ndim - 1) {\n        mask = tfc.expandDims(mask, -1);\n      }\n      mask = tfc.transpose(mask, axes);\n    }\n\n    if (goBackwards) {\n      inputs = tfc.reverse(inputs, 0);\n      if (mask != null) {\n        mask = tfc.reverse(mask, 0);\n      }\n    }\n\n    // Porting Note: PyKeras with TensorFlow backend uses a symbolic loop\n    //   (tf.while_loop). But for the imperative deeplearn.js backend, we just\n    //   use the usual TypeScript control flow to iterate over the time steps in\n    //   the inputs.\n    // Porting Note: PyKeras patches a \"_use_learning_phase\" attribute to\n    // outputs.\n    //   This is not idiomatic in TypeScript. The info regarding whether we are\n    //   in a learning (i.e., training) phase for RNN is passed in a different\n    //   way.\n\n    const perStepOutputs: Tensor[] = [];\n    let lastOutput: Tensor;\n    let states = initialStates;\n    const timeSteps = inputs.shape[0];\n    const perStepInputs = tfc.unstack(inputs);\n    let perStepMasks: Tensor[];\n    if (mask != null) {\n      perStepMasks = tfc.unstack(mask);\n    }\n\n    for (let t = 0; t < timeSteps; ++t) {\n      const currentInput = perStepInputs[t];\n      const stepOutputs = tfc.tidy(() => stepFunction(currentInput, states));\n\n      if (mask == null) {\n        lastOutput = stepOutputs[0];\n        states = stepOutputs[1];\n      } else {\n        const maskedOutputs = tfc.tidy(() => {\n          const stepMask = perStepMasks[t];\n          const negStepMask = tfc.onesLike(stepMask).sub(stepMask);\n          // TODO(cais): Would tfc.where() be better for performance?\n          const output =\n              stepOutputs[0].mul(stepMask).add(states[0].mul(negStepMask));\n          const newStates = states.map((state, i) => {\n            return stepOutputs[1][i].mul(stepMask).add(state.mul(negStepMask));\n          });\n          return {output, newStates};\n        });\n        lastOutput = maskedOutputs.output;\n        states = maskedOutputs.newStates;\n      }\n\n      if (needPerStepOutputs) {\n        perStepOutputs.push(lastOutput);\n      }\n    }\n    let outputs: Tensor;\n    if (needPerStepOutputs) {\n      const axis = 1;\n      outputs = tfc.stack(perStepOutputs, axis);\n    }\n    return [lastOutput, outputs, states] as [Tensor, Tensor, Tensor[]];\n  });\n}\n\nexport declare interface BaseRNNLayerArgs extends LayerArgs {\n  /**\n   * A RNN cell instance. A RNN cell is a class that has:\n   *   - a `call()` method, which takes `[Tensor, Tensor]` as the\n   *     first input argument. The first item is the input at time t, and\n   *     second item is the cell state at time t.\n   *     The `call()` method returns `[outputAtT, statesAtTPlus1]`.\n   *     The `call()` method of the cell can also take the argument `constants`,\n   *     see section \"Note on passing external constants\" below.\n   *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,\n   *       which are Layer subtypes, to accept two arguments. tfjs-layers does\n   *       not do such overriding. Instead we preseve the `call()` signature,\n   *       which due to its `Tensor|Tensor[]` argument and return value, is\n   *       flexible enough to handle the inputs and states.\n   *   - a `stateSize` attribute. This can be a single integer (single state)\n   *     in which case it is the size of the recurrent state (which should be\n   *     the same as the size of the cell output). This can also be an Array of\n   *     integers (one size per state). In this case, the first entry\n   *     (`stateSize[0]`) should be the same as the size of the cell output.\n   * It is also possible for `cell` to be a list of RNN cell instances, in which\n   * case the cells get stacked on after the other in the RNN, implementing an\n   * efficient stacked RNN.\n   */\n  cell?: RNNCell|RNNCell[];\n\n  /**\n   * Whether to return the last output in the output sequence, or the full\n   * sequence.\n   */\n  returnSequences?: boolean;\n\n  /**\n   * Whether to return the last state in addition to the output.\n   */\n  returnState?: boolean;\n\n  /**\n   * If `true`, process the input sequence backwards and return the reversed\n   * sequence (default: `false`).\n   */\n  goBackwards?: boolean;\n\n  /**\n   * If `true`, the last state for each sample at index i in a batch will be\n   * used as initial state of the sample of index i in the following batch\n   * (default: `false`).\n   *\n   * You can set RNN layers to be \"stateful\", which means that the states\n   * computed for the samples in one batch will be reused as initial states\n   * for the samples in the next batch. This assumes a one-to-one mapping\n   * between samples in different successive batches.\n   *\n   * To enable \"statefulness\":\n   *   - specify `stateful: true` in the layer constructor.\n   *   - specify a fixed batch size for your model, by passing\n   *     - if sequential model:\n   *       `batchInputShape: [...]` to the first layer in your model.\n   *     - else for functional model with 1 or more Input layers:\n   *       `batchShape: [...]` to all the first layers in your model.\n   *     This is the expected shape of your inputs\n   *     *including the batch size*.\n   *     It should be a tuple of integers, e.g., `[32, 10, 100]`.\n   *   - specify `shuffle: false` when calling `LayersModel.fit()`.\n   *\n   * To reset the state of your model, call `resetStates()` on either the\n   * specific layer or on the entire model.\n   */\n  stateful?: boolean;\n  // TODO(cais): Explore whether we can warn users when they fail to set\n  //   `shuffle: false` when training a model consisting of stateful RNNs\n  //   and any stateful Layers in general.\n\n  /**\n   * If `true`, the network will be unrolled, else a symbolic loop will be\n   * used. Unrolling can speed-up a RNN, although it tends to be more memory-\n   * intensive. Unrolling is only suitable for short sequences (default:\n   * `false`).\n   * Porting Note: tfjs-layers has an imperative backend. RNNs are executed with\n   *   normal TypeScript control flow. Hence this property is inapplicable and\n   *   ignored in tfjs-layers.\n   */\n  unroll?: boolean;\n\n  /**\n   * Dimensionality of the input (integer).\n   *   This option (or alternatively, the option `inputShape`) is required when\n   *   this layer is used as the first layer in a model.\n   */\n  inputDim?: number;\n\n  /**\n   * Length of the input sequences, to be specified when it is constant.\n   * This argument is required if you are going to connect `Flatten` then\n   * `Dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed). Note that if the recurrent layer is not the first layer in\n   * your model, you would need to specify the input length at the level of the\n   * first layer (e.g., via the `inputShape` option).\n   */\n  inputLength?: number;\n}\n\nexport class RNN extends Layer {\n  /** @nocollapse */\n  static className = 'RNN';\n  public readonly cell: RNNCell;\n  public readonly returnSequences: boolean;\n  public readonly returnState: boolean;\n  public readonly goBackwards: boolean;\n  public readonly unroll: boolean;\n\n  public stateSpec: InputSpec[];\n  protected states_: Tensor[];\n\n  // NOTE(cais): For stateful RNNs, the old states cannot be disposed right\n  // away when new states are set, because the old states may need to be used\n  // later for backpropagation through time (BPTT) and other purposes. So we\n  // keep them here for final disposal when the state is reset completely\n  // (i.e., through no-arg call to `resetStates()`).\n  protected keptStates: Tensor[][];\n\n  private numConstants: number;\n\n  constructor(args: RNNLayerArgs) {\n    super(args);\n    let cell: RNNCell;\n    if (args.cell == null) {\n      throw new ValueError(\n          'cell property is missing for the constructor of RNN.');\n    } else if (Array.isArray(args.cell)) {\n      cell = new StackedRNNCells({cells: args.cell});\n    } else {\n      cell = args.cell;\n    }\n    if (cell.stateSize == null) {\n      throw new ValueError(\n          'The RNN cell should have an attribute `stateSize` (tuple of ' +\n          'integers, one integer per RNN state).');\n    }\n    this.cell = cell;\n    this.returnSequences =\n        args.returnSequences == null ? false : args.returnSequences;\n    this.returnState = args.returnState == null ? false : args.returnState;\n    this.goBackwards = args.goBackwards == null ? false : args.goBackwards;\n    this._stateful = args.stateful == null ? false : args.stateful;\n    this.unroll = args.unroll == null ? false : args.unroll;\n\n    this.supportsMasking = true;\n    this.inputSpec = [new InputSpec({ndim: 3})];\n    this.stateSpec = null;\n    this.states_ = null;\n    // TODO(cais): Add constantsSpec and numConstants.\n    this.numConstants = null;\n    // TODO(cais): Look into the use of initial_state in the kwargs of the\n    //   constructor.\n\n    this.keptStates = [];\n  }\n\n  // Porting Note: This is the equivalent of `RNN.states` property getter in\n  //   PyKeras.\n  getStates(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      return math_utils.range(0, numStates).map(x => null);\n    } else {\n      return this.states_;\n    }\n  }\n\n  // Porting Note: This is the equivalent of the `RNN.states` property setter in\n  //   PyKeras.\n  setStates(states: Tensor[]): void {\n    this.states_ = states;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    // TODO(cais): Remove the casting once stacked RNN cells become supported.\n    let stateSize = this.cell.stateSize;\n    if (!Array.isArray(stateSize)) {\n      stateSize = [stateSize];\n    }\n    const outputDim = stateSize[0];\n    let outputShape: Shape|Shape[];\n    if (this.returnSequences) {\n      outputShape = [inputShape[0], inputShape[1], outputDim];\n    } else {\n      outputShape = [inputShape[0], outputDim];\n    }\n\n    if (this.returnState) {\n      const stateShape: Shape[] = [];\n      for (const dim of stateSize) {\n        stateShape.push([inputShape[0], dim]);\n      }\n      return [outputShape].concat(stateShape);\n    } else {\n      return outputShape;\n    }\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tfc.tidy(() => {\n      if (Array.isArray(mask)) {\n        mask = mask[0];\n      }\n      const outputMask = this.returnSequences ? mask : null;\n\n      if (this.returnState) {\n        const stateMask = this.states.map(s => null);\n        return [outputMask].concat(stateMask);\n      } else {\n        return outputMask;\n      }\n    });\n  }\n\n  /**\n   * Get the current state tensors of the RNN.\n   *\n   * If the state hasn't been set, return an array of `null`s of the correct\n   * length.\n   */\n  get states(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      const output: Tensor[] = [];\n      for (let i = 0; i < numStates; ++i) {\n        output.push(null);\n      }\n      return output;\n    } else {\n      return this.states_;\n    }\n  }\n\n  set states(s: Tensor[]) {\n    this.states_ = s;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    // Note inputShape will be an Array of Shapes of initial states and\n    // constants if these are passed in apply().\n    const constantShape: Shape[] = null;\n    if (this.numConstants != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    }\n\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    const batchSize: number = this.stateful ? inputShape[0] : null;\n    const inputDim = inputShape.slice(2);\n    this.inputSpec[0] = new InputSpec({shape: [batchSize, null, ...inputDim]});\n\n    // Allow cell (if RNNCell Layer) to build before we set or validate\n    // stateSpec.\n    const stepInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (constantShape != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    } else {\n      this.cell.build(stepInputShape);\n    }\n\n    // Set or validate stateSpec.\n    let stateSize: number[];\n    if (Array.isArray(this.cell.stateSize)) {\n      stateSize = this.cell.stateSize;\n    } else {\n      stateSize = [this.cell.stateSize];\n    }\n\n    if (this.stateSpec != null) {\n      if (!util.arraysEqual(\n              this.stateSpec.map(spec => spec.shape[spec.shape.length - 1]),\n              stateSize)) {\n        throw new ValueError(\n            `An initialState was passed that is not compatible with ` +\n            `cell.stateSize. Received stateSpec=${this.stateSpec}; ` +\n            `However cell.stateSize is ${this.cell.stateSize}`);\n      }\n    } else {\n      this.stateSpec =\n          stateSize.map(dim => new InputSpec({shape: [null, dim]}));\n    }\n    if (this.stateful) {\n      this.resetStates();\n    }\n  }\n\n  /**\n   * Reset the state tensors of the RNN.\n   *\n   * If the `states` argument is `undefined` or `null`, will set the\n   * state tensor(s) of the RNN to all-zero tensors of the appropriate\n   * shape(s).\n   *\n   * If `states` is provided, will set the state tensors of the RNN to its\n   * value.\n   *\n   * @param states Optional externally-provided initial states.\n   * @param training Whether this call is done during training. For stateful\n   *   RNNs, this affects whether the old states are kept or discarded. In\n   *   particular, if `training` is `true`, the old states will be kept so\n   *   that subsequent backpropgataion through time (BPTT) may work properly.\n   *   Else, the old states will be discarded.\n   */\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n      const batchSize = this.inputSpec[0].shape[0];\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n      // Initialize state if null.\n      if (this.states_ == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_ = [tfc.zeros([batchSize, this.cell.stateSize])];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_[0] = tfc.zeros([batchSize, this.cell.stateSize]);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training === true) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n          const dim = Array.isArray(this.cell.stateSize) ?\n              this.cell.stateSize[index] :\n              this.cell.stateSize;\n          const expectedShape = [batchSize, dim];\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n          this.states_[index] = value;\n        }\n      }\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    // TODO(cais): Figure out whether initialState is in kwargs or inputs.\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    // If any of `initial_state` or `constants` are specified and are\n    // `tf.SymbolicTensor`s, then add them to the inputs and temporarily modify\n    // the input_spec to include them.\n\n    let additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    let additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      kwargs['initialState'] = initialState;\n      additionalInputs = additionalInputs.concat(initialState);\n      this.stateSpec = [];\n      for (const state of initialState) {\n        this.stateSpec.push(new InputSpec({shape: state.shape}));\n      }\n      // TODO(cais): Use the following instead.\n      // this.stateSpec = initialState.map(state => new InputSpec({shape:\n      // state.shape}));\n      additionalSpecs = additionalSpecs.concat(this.stateSpec);\n    }\n    if (constants != null) {\n      kwargs['constants'] = constants;\n      additionalInputs = additionalInputs.concat(constants);\n      // TODO(cais): Add this.constantsSpec.\n      this.numConstants = constants.length;\n    }\n\n    const isTensor = additionalInputs[0] instanceof SymbolicTensor;\n    if (isTensor) {\n      // Compute full input spec, including state and constants.\n      const fullInput =\n          [inputs].concat(additionalInputs) as Tensor[] | SymbolicTensor[];\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call with temporarily replaced inputSpec.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output = super.apply(fullInput, kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  // tslint:disable-next-line:no-any\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    // Input shape: `[samples, time (padded with zeros), input_dim]`.\n    // Note that the .build() method of subclasses **must** define\n    // this.inputSpec and this.stateSpec owith complete input shapes.\n    return tidy(() => {\n      const mask = kwargs == null ? null : kwargs['mask'] as Tensor;\n      const training = kwargs == null ? null : kwargs['training'];\n      let initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      inputs = getExactlyOneTensor(inputs);\n      if (initialState == null) {\n        if (this.stateful) {\n          initialState = this.states_;\n        } else {\n          initialState = this.getInitialState(inputs);\n        }\n      }\n\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      if (initialState.length !== numStates) {\n        throw new ValueError(\n            `RNN Layer has ${numStates} state(s) but was passed ` +\n            `${initialState.length} initial state(s).`);\n      }\n      if (this.unroll) {\n        console.warn(\n            'Ignoring unroll = true for RNN layer, due to imperative backend.');\n      }\n\n      const cellCallKwargs: Kwargs = {training};\n\n      // TODO(cais): Add support for constants.\n      const step = (inputs: Tensor, states: Tensor[]) => {\n        // `inputs` and `states` are concatenated to form a single `Array` of\n        // `tf.Tensor`s as the input to `cell.call()`.\n        const outputs =\n            this.cell.call([inputs].concat(states), cellCallKwargs) as Tensor[];\n        // Marshall the return value into output and new states.\n        return [outputs[0], outputs.slice(1)] as [Tensor, Tensor[]];\n      };\n\n      // TODO(cais): Add support for constants.\n\n      const rnnOutputs =\n          rnn(step, inputs, initialState, this.goBackwards, mask, null,\n              this.unroll, this.returnSequences);\n      const lastOutput = rnnOutputs[0];\n      const outputs = rnnOutputs[1];\n      const states = rnnOutputs[2];\n\n      if (this.stateful) {\n        this.resetStates(states, training);\n      }\n\n      const output = this.returnSequences ? outputs : lastOutput;\n\n      // TODO(cais): Porperty set learning phase flag.\n\n      if (this.returnState) {\n        return [output].concat(states);\n      } else {\n        return output;\n      }\n    });\n  }\n\n  getInitialState(inputs: Tensor): Tensor[] {\n    return tidy(() => {\n      // Build an all-zero tensor of shape [samples, outputDim].\n      // [Samples, timeSteps, inputDim].\n      let initialState = tfc.zeros(inputs.shape);\n      // [Samples].\n      initialState = tfc.sum(initialState, [1, 2]);\n      initialState = K.expandDims(initialState);  // [Samples, 1].\n\n      if (Array.isArray(this.cell.stateSize)) {\n        return this.cell.stateSize.map(\n            dim => dim > 1 ? K.tile(initialState, [1, dim]) : initialState);\n      } else {\n        return this.cell.stateSize > 1 ?\n            [K.tile(initialState, [1, this.cell.stateSize])] :\n            [initialState];\n      }\n    });\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    return this.cell.trainableWeights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    if (!this.trainable) {\n      return this.cell.weights;\n    }\n    return this.cell.nonTrainableWeights;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.cell != null) {\n      this.cell.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      returnSequences: this.returnSequences,\n      returnState: this.returnState,\n      goBackwards: this.goBackwards,\n      stateful: this.stateful,\n      unroll: this.unroll,\n    };\n\n    if (this.numConstants != null) {\n      config['numConstants'] = this.numConstants;\n    }\n\n    const cellConfig = this.cell.getConfig();\n\n    if (this.getClassName() === RNN.className) {\n      config['cell'] = {\n        'className': this.cell.getClassName(),\n        'config': cellConfig,\n      } as serialization.ConfigDictValue;\n    }\n\n    // this order is necessary, to prevent cell name from replacing layer name\n    return {...cellConfig, ...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cellConfig = config['cell'] as serialization.ConfigDict;\n    const cell = deserialize(cellConfig, customObjects) as RNNCell;\n    return new cls(Object.assign(config, {cell}));\n  }\n}\nserialization.registerClass(RNN);\n\n// Porting Note: This is a common parent class for RNN cells. There is no\n// equivalent of this in PyKeras. Having a common parent class forgoes the\n//  need for `has_attr(cell, ...)` checks or its TypeScript equivalent.\n/**\n * An RNNCell layer.\n *\n * @doc {heading: 'Layers', subheading: 'Classes'}\n */\nexport abstract class RNNCell extends Layer {\n  /**\n   * Size(s) of the states.\n   * For RNN cells with only a single state, this is a single integer.\n   */\n  public stateSize: number|number[];\n  public dropoutMask: Tensor|Tensor[];\n  public recurrentDropoutMask: Tensor|Tensor[];\n}\n\nexport declare interface SimpleRNNCellLayerArgs extends LayerArgs {\n  /**\n   * units: Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   * Default: hyperbolic tangent ('tanh').\n   * If you pass `null`,  'linear' activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the `kernel` weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the `recurrent_kernel` weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the `kernel` weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the `recurrentKernel` weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraintfunction applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\nexport class SimpleRNNCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'SimpleRNNCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args: SimpleRNNCellLayerArgs) {\n    super(args);\n    this.units = args.units;\n    assertPositiveInteger(this.units, `units`);\n    this.activation = getActivation(\n        args.activation == null ? this.DEFAULT_ACTIVATION : args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    // TODO(cais): Use regularizer.\n    this.kernel = this.addWeight(\n        'kernel', [inputShape[inputShape.length - 1], this.units], null,\n        this.kernelInitializer, this.kernelRegularizer, true,\n        this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  // Porting Note: PyKeras' equivalent of this method takes two tensor inputs:\n  //   `inputs` and `states`. Here, the two tensors are combined into an\n  //   `Tensor[]` Array as the first input argument.\n  //   Similarly, PyKeras' equivalent of this method returns two values:\n  //    `output` and `[output]`. Here the two are combined into one length-2\n  //    `Tensor[]`, consisting of `output` repeated.\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `SimpleRNNCell expects 2 input Tensors, got ${inputs.length}.`);\n      }\n      let prevOutput = inputs[1];\n      inputs = inputs[0];\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training\n                           }) as Tensor;\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(prevOutput),\n                                      rate: this.recurrentDropout,\n                                      training\n                                    }) as Tensor;\n      }\n      let h: Tensor;\n      const dpMask: Tensor = this.dropoutMask as Tensor;\n      const recDpMask: Tensor = this.recurrentDropoutMask as Tensor;\n      if (dpMask != null) {\n        h = K.dot(tfc.mul(inputs, dpMask), this.kernel.read());\n      } else {\n        h = K.dot(inputs, this.kernel.read());\n      }\n      if (this.bias != null) {\n        h = K.biasAdd(h, this.bias.read());\n      }\n      if (recDpMask != null) {\n        prevOutput = tfc.mul(prevOutput, recDpMask);\n      }\n      let output = tfc.add(h, K.dot(prevOutput, this.recurrentKernel.read()));\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      // TODO(cais): Properly set learning phase on output tensor?\n      return [output, output];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(SimpleRNNCell);\n\nexport declare interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {\n  /**\n   * Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   *\n   * Defaults to  hyperbolic tangent (`tanh`)\n   *\n   * If you pass `null`, no activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the recurrentKernel weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the kernel weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the recurrentKernel weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\n/**\n * RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the\n * `cell` property required. This interface is to be used with constructors\n * of concrete RNN layer subtypes.\n */\nexport declare interface RNNLayerArgs extends BaseRNNLayerArgs {\n  cell: RNNCell|RNNCell[];\n}\n\nexport class SimpleRNN extends RNN {\n  /** @nocollapse */\n  static className = 'SimpleRNN';\n  constructor(args: SimpleRNNLayerArgs) {\n    args.cell = new SimpleRNNCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\nserialization.registerClass(SimpleRNN);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n\n  /**\n   * GRU convention (whether to apply reset gate after or before matrix\n   * multiplication). false = \"before\", true = \"after\" (only false is\n   * supported).\n   */\n  resetAfter?: boolean;\n}\n\nexport class GRUCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'GRUCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION: ActivationIdentifier = 'hardSigmoid';\n\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: GRUCellLayerArgs) {\n    super(args);\n    if (args.resetAfter) {\n      throw new ValueError(\n          `GRUCell does not support reset_after parameter set to true.`);\n    }\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 3], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 3], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units * 3], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `GRUCell expects 2 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      let hTMinus1 = inputs[1];  // Previous memory state.\n      inputs = inputs[0];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2, regardless of the actual value of\n      // config.implementation.\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 3\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 3\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor];\n      const recDpMask = this.recurrentDropoutMask as [Tensor, Tensor, Tensor];\n      let z: Tensor;\n      let r: Tensor;\n      let hh: Tensor;\n\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let matrixX = K.dot(inputs, this.kernel.read());\n      if (this.useBias) {\n        matrixX = K.biasAdd(matrixX, this.bias.read());\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n\n      const recurrentKernelValue = this.recurrentKernel.read();\n      const [rk1, rk2] = tfc.split(\n          recurrentKernelValue, [2 * this.units, this.units],\n          recurrentKernelValue.rank - 1);\n      const matrixInner = K.dot(hTMinus1, rk1);\n\n      const [xZ, xR, xH] = tfc.split(matrixX, 3, matrixX.rank - 1);\n      const [recurrentZ, recurrentR] =\n          tfc.split(matrixInner, 2, matrixInner.rank - 1);\n      z = this.recurrentActivation.apply(tfc.add(xZ, recurrentZ));\n      r = this.recurrentActivation.apply(tfc.add(xR, recurrentR));\n\n      const recurrentH = K.dot(tfc.mul(r, hTMinus1), rk2);\n      hh = this.activation.apply(tfc.add(xH, recurrentH));\n\n      const h =\n          tfc.add(tfc.mul(z, hTMinus1), tfc.mul(tfc.add(1, tfc.neg(z)), hh));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n      resetAfter: false\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(GRUCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface GRULayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   * smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class GRU extends RNN {\n  /** @nocollapse */\n  static className = 'GRU';\n  constructor(args: GRULayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new GRUCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(GRU);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class LSTMCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'LSTMCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n  readonly unitForgetBias: boolean;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number[];\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION = 'hardSigmoid';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n\n  readonly DEFAULT_BIAS_INITIALIZER = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: LSTMCellLayerArgs) {\n    super(args);\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.unitForgetBias = args.unitForgetBias;\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = [this.units, this.units];\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 4], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 4], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    let biasInitializer: Initializer;\n    if (this.useBias) {\n      if (this.unitForgetBias) {\n        const capturedBiasInit = this.biasInitializer;\n        const capturedUnits = this.units;\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): Tensor {\n            // TODO(cais): More informative variable names?\n            const bI = capturedBiasInit.apply([capturedUnits]);\n            const bF = (new Ones()).apply([capturedUnits]);\n            const bCAndH = capturedBiasInit.apply([capturedUnits * 2]);\n            return K.concatAlongFirstAxis(\n                K.concatAlongFirstAxis(bI, bF), bCAndH);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n      this.bias = this.addWeight(\n          'bias', [this.units * 4], null, biasInitializer, this.biasRegularizer,\n          true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `LSTMCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n      let hTMinus1 = inputs[1];    // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n      inputs = inputs[0];\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 4\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 4\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor, Tensor];\n      const recDpMask =\n          this.recurrentDropoutMask as [Tensor, Tensor, Tensor, Tensor];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2 regardless of the actual value of\n      // config.implementation.\n      let i: Tensor;\n      let f: Tensor;\n      let c: Tensor;\n      let o: Tensor;\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let z = K.dot(inputs, this.kernel.read());\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n      z = tfc.add(z, K.dot(hTMinus1, this.recurrentKernel.read()));\n      if (this.useBias) {\n        z = K.biasAdd(z, this.bias.read());\n      }\n\n      const [z0, z1, z2, z3] = tfc.split(z, 4, z.rank - 1);\n\n      i = this.recurrentActivation.apply(z0);\n      f = this.recurrentActivation.apply(z1);\n      c = tfc.add(tfc.mul(f, cTMinus1), tfc.mul(i, this.activation.apply(z2)));\n      o = this.recurrentActivation.apply(z3);\n\n      const h = tfc.mul(o, this.activation.apply(c));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      unitForgetBias: this.unitForgetBias,\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(LSTMCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface LSTMLayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *   Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions, whereas mode 2 will\n   *   batch them into fewer, larger operations. These modes will\n   *   have different performance profiles on different hardware and\n   *   for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this config field.\n   */\n  implementation?: number;\n}\n\nexport class LSTM extends RNN {\n  /** @nocollapse */\n  static className = 'LSTM';\n  constructor(args: LSTMLayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new LSTMCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(LSTM);\n\nexport declare interface StackedRNNCellsArgs extends LayerArgs {\n  /**\n   * A `Array` of `RNNCell` instances.\n   */\n  cells: RNNCell[];\n}\n\nexport class StackedRNNCells extends RNNCell {\n  /** @nocollapse */\n  static className = 'StackedRNNCells';\n  protected cells: RNNCell[];\n\n  constructor(args: StackedRNNCellsArgs) {\n    super(args);\n    this.cells = args.cells;\n  }\n\n  get stateSize(): number[] {\n    // States are a flat list in reverse order of the cell stack.\n    // This allows perserving the requirement `stack.statesize[0] ===\n    // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,\n    // assuming one LSTM has states `[h, c]`.\n    const stateSize: number[] = [];\n    for (const cell of this.cells.slice().reverse()) {\n      if (Array.isArray(cell.stateSize)) {\n        stateSize.push(...cell.stateSize);\n      } else {\n        stateSize.push(cell.stateSize);\n      }\n    }\n    return stateSize;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      let states = inputs.slice(1);\n\n      // Recover per-cell states.\n      const nestedStates: Tensor[][] = [];\n      for (const cell of this.cells.slice().reverse()) {\n        if (Array.isArray(cell.stateSize)) {\n          nestedStates.push(states.splice(0, cell.stateSize.length));\n        } else {\n          nestedStates.push(states.splice(0, 1));\n        }\n      }\n      nestedStates.reverse();\n\n      // Call the cells in order and store the returned states.\n      const newNestedStates: Tensor[][] = [];\n      let callInputs: Tensor[];\n      for (let i = 0; i < this.cells.length; ++i) {\n        const cell = this.cells[i];\n        states = nestedStates[i];\n        // TODO(cais): Take care of constants.\n        if (i === 0) {\n          callInputs = [inputs[0]].concat(states);\n        } else {\n          callInputs = [callInputs[0]].concat(states);\n        }\n        callInputs = cell.call(callInputs, kwargs) as Tensor[];\n        newNestedStates.push(callInputs.slice(1));\n      }\n\n      // Format the new states as a flat list in reverse cell order.\n      states = [];\n      for (const cellStates of newNestedStates.slice().reverse()) {\n        states.push(...cellStates);\n      }\n      return [callInputs[0]].concat(states);\n    });\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    if (isArrayOfShapes(inputShape)) {\n      // TODO(cais): Take care of input constants.\n      // const constantShape = inputShape.slice(1);\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n    let outputDim: number;\n    this.cells.forEach((cell, i) => {\n      nameScope(`RNNCell_${i}`, () => {\n        // TODO(cais): Take care of input constants.\n\n        cell.build(inputShape);\n        if (Array.isArray(cell.stateSize)) {\n          outputDim = cell.stateSize[0];\n        } else {\n          outputDim = cell.stateSize;\n        }\n        inputShape = [inputShape[0], outputDim] as Shape;\n      });\n    });\n    this.built = true;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const getCellConfig = (cell: RNNCell) => {\n      return {\n        'className': cell.getClassName(),\n        'config': cell.getConfig(),\n      };\n    };\n\n    const cellConfigs = this.cells.map(getCellConfig);\n\n    const config = {'cells': cellConfigs};\n\n    return {...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cells: RNNCell[] = [];\n    for (const cellConfig of (config['cells'] as serialization.ConfigDict[])) {\n      cells.push(deserialize(cellConfig, customObjects) as RNNCell);\n    }\n    return new cls({cells});\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const cell of this.cells) {\n        trainableWeights.push(...cell.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  /**\n   * Retrieve the weights of a the model.\n   *\n   * @returns A flat `Array` of `tf.Tensor`s.\n   */\n  getWeights(): Tensor[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.weights);\n    }\n    return batchGetValue(weights);\n  }\n\n  /**\n   * Set the weights of the model.\n   *\n   * @param weights An `Array` of `tf.Tensor`s with shapes and types matching\n   *     the output of `getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    const tuples: Array<[LayerVariable, Tensor]> = [];\n    for (const cell of this.cells) {\n      const numParams = cell.weights.length;\n      const inputWeights = weights.splice(numParams);\n      for (let i = 0; i < cell.weights.length; ++i) {\n        tuples.push([cell.weights[i], inputWeights[i]]);\n      }\n    }\n    batchSetValue(tuples);\n  }\n\n  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.\n}\nserialization.registerClass(StackedRNNCells);\n\nexport function generateDropoutMask(args: {\n  ones: () => tfc.Tensor,\n  rate: number,\n  training?: boolean,\n  count?: number,\n}): tfc.Tensor|tfc.Tensor[] {\n  const {ones, rate, training = false, count = 1} = args;\n\n  const droppedInputs = () => K.dropout(ones(), rate);\n\n  const createMask = () => K.inTrainPhase(droppedInputs, ones, training);\n\n  // just in case count is provided with null or undefined\n  if (!count || count <= 1) {\n    return tfc.keep(createMask().clone());\n  }\n\n  const masks = Array(count).fill(undefined).map(createMask);\n\n  return masks.map(m => tfc.keep(m.clone()));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {Activation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint} from '../constraints';\nimport {InputSpec} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {Initializer} from '../initializers';\nimport {DataFormat, DataType, PaddingMode, Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, normalizeArray} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\nimport {BaseRNNLayerArgs, generateDropoutMask, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNNCellLayerArgs} from './recurrent';\n\ndeclare interface ConvRNN2DCellArgs extends\n    Omit<SimpleRNNCellLayerArgs, 'units'> {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number];\n}\n\nabstract class ConvRNN2DCell extends RNNCell {\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n}\n\ndeclare interface ConvRNN2DLayerArgs extends BaseRNNLayerArgs,\n                                             ConvRNN2DCellArgs {}\n\n/**\n * Base class for convolutional-recurrent layers.\n */\nclass ConvRNN2D extends RNN {\n  /** @nocollapse */\n  static className = 'ConvRNN2D';\n\n  readonly cell: ConvRNN2DCell;\n\n  constructor(args: ConvRNN2DLayerArgs) {\n    if (args.unroll) {\n      throw new NotImplementedError(\n          'Unrolling is not possible with convolutional RNNs.');\n    }\n\n    if (Array.isArray(args.cell)) {\n      throw new NotImplementedError(\n          'It is not possible at the moment to stack convolutional cells.');\n    }\n\n    super(args as RNNLayerArgs);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n\n        this.cell.dropoutMask = null;\n      }\n\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n\n        this.cell.recurrentDropoutMask = null;\n      }\n\n      if (kwargs && kwargs['constants']) {\n        throw new ValueError('ConvRNN2D cell does not support constants');\n      }\n\n      const mask = kwargs == null ? null : kwargs['mask'];\n\n      const training = kwargs == null ? null : kwargs['training'];\n\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  computeOutputShape(inputShape: Shape): Shape|Shape[] {\n    let outShape: Shape = this.computeSingleOutputShape(inputShape);\n\n    if (!this.returnSequences) {\n      outShape = [outShape[0], ...outShape.slice(2)];\n    }\n\n    if (this.returnState) {\n      outShape =\n          [outShape, ...Array(2).fill([inputShape[0], ...outShape.slice(-3)])];\n    }\n\n    return outShape;\n  }\n\n  getInitialState(inputs: tfc.Tensor): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      const {stateSize} = this.cell;\n\n      const inputShape = inputs.shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const initialState = tfc.zeros(stateShape);\n\n      if (Array.isArray(stateSize)) {\n        return Array(stateSize.length).fill(initialState);\n      }\n\n      return [initialState];\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tfc.tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n\n      const inputShape = this.inputSpec[0].shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const batchSize = inputShape[0];\n\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n\n      // Initialize state if null.\n      if (this.getStates() == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_ = [tfc.zeros(stateShape)];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_[0] = tfc.zeros(stateShape);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n\n          const expectedShape = stateShape;\n\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n\n          this.states_[index] = value;\n        }\n      }\n\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  protected computeSingleOutputShape(inputShape: Shape): Shape {\n    const {dataFormat, filters, kernelSize, padding, strides, dilationRate} =\n        this.cell;\n\n    const isChannelsFirst = dataFormat === 'channelsFirst';\n\n    const h = inputShape[isChannelsFirst ? 3 : 2];\n    const w = inputShape[isChannelsFirst ? 4 : 3];\n\n    const hOut = convOutputLength(\n        h, kernelSize[0], padding, strides[0], dilationRate[0]);\n    const wOut = convOutputLength(\n        w, kernelSize[1], padding, strides[1], dilationRate[1]);\n\n    const outShape: Shape = [\n      ...inputShape.slice(0, 2),\n      ...(isChannelsFirst ? [filters, hOut, wOut] : [hOut, wOut, filters])\n    ];\n\n    return outShape;\n  }\n}\n\nexport declare interface ConvLSTM2DCellArgs extends\n    Omit<LSTMCellLayerArgs, 'units'>, ConvRNN2DCellArgs {}\n\nexport class ConvLSTM2DCell extends LSTMCell implements ConvRNN2DCell {\n  /** @nocollapse */\n  static className = 'ConvLSTM2DCell';\n\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  constructor(args: ConvLSTM2DCellArgs) {\n    const {\n      filters,\n      kernelSize,\n      strides,\n      padding,\n      dataFormat,\n      dilationRate,\n    } = args;\n\n    super({...args, units: filters});\n\n    this.filters = filters;\n    assertPositiveInteger(this.filters, 'filters');\n\n    this.kernelSize = normalizeArray(kernelSize, 2, 'kernelSize');\n    this.kernelSize.forEach(size => assertPositiveInteger(size, 'kernelSize'));\n\n    this.strides = normalizeArray(strides || 1, 2, 'strides');\n    this.strides.forEach(stride => assertPositiveInteger(stride, 'strides'));\n\n    this.padding = padding || 'valid';\n    checkPaddingMode(this.padding);\n\n    this.dataFormat = dataFormat || 'channelsLast';\n    checkDataFormat(this.dataFormat);\n\n    this.dilationRate = normalizeArray(dilationRate || 1, 2, 'dilationRate');\n    this.dilationRate.forEach(\n        rate => assertPositiveInteger(rate, 'dilationRate'));\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n\n    const numOfKernels = 4;\n\n    const kernelShape =\n        this.kernelSize.concat([inputDim, this.filters * numOfKernels]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n\n    const recurrentKernelShape =\n        this.kernelSize.concat([this.filters, this.filters * numOfKernels]);\n\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', recurrentKernelShape, null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n\n    if (this.useBias) {\n      let biasInitializer: Initializer;\n\n      if (this.unitForgetBias) {\n        const init = this.biasInitializer;\n\n        const filters = this.filters;\n\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): tfc.Tensor {\n            const biasI = init.apply([filters]);\n            const biasF = tfc.ones([filters]);\n            const biasCAndO = init.apply([filters * 2]);\n            return K.concatenate([biasI, biasF, biasCAndO]);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n\n      this.bias = this.addWeight(\n          'bias', [this.filters * numOfKernels], null, biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: tfc.Tensor[], kwargs: Kwargs): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] || false;\n\n      const x = inputs[0];         // Current input\n      const hTMinus1 = inputs[1];  // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n\n      const numOfKernels = 4;\n\n      type DropoutMasks = [tfc.Tensor, tfc.Tensor, tfc.Tensor, tfc.Tensor];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(x),\n                             rate: this.dropout,\n                             training,\n                             count: numOfKernels\n                           }) as tfc.Tensor[];\n      }\n\n      const dropoutMask = this.dropoutMask as DropoutMasks;\n\n      const applyDropout =\n          (x: tfc.Tensor, mask: tfc.Tensor[], index: number) => {\n            if (!mask || !mask[index]) {\n              return x;\n            }\n\n            return tfc.mul(mask[index], x);\n          };\n\n      let xI = applyDropout(x, dropoutMask, 0);\n      let xF = applyDropout(x, dropoutMask, 1);\n      let xC = applyDropout(x, dropoutMask, 2);\n      let xO = applyDropout(x, dropoutMask, 3);\n\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: numOfKernels\n                                    }) as tfc.Tensor[];\n      }\n\n      const recDropoutMask = this.recurrentDropoutMask as DropoutMasks;\n\n      let hI = applyDropout(hTMinus1, recDropoutMask, 0);\n      let hF = applyDropout(hTMinus1, recDropoutMask, 1);\n      let hC = applyDropout(hTMinus1, recDropoutMask, 2);\n      let hO = applyDropout(hTMinus1, recDropoutMask, 3);\n\n      const kernelChannelAxis = 3;\n\n      const [kernelI, kernelF, kernelC, kernelO]: tfc.Tensor[] =\n          tfc.split(this.kernel.read(), numOfKernels, kernelChannelAxis);\n\n      const [biasI, biasF, biasC, biasO]: tfc.Tensor[] = this.useBias ?\n          tfc.split(this.bias.read(), numOfKernels) :\n          [null, null, null, null];\n\n      xI = this.inputConv(xI, kernelI, biasI, this.padding);\n      xF = this.inputConv(xF, kernelF, biasF, this.padding);\n      xC = this.inputConv(xC, kernelC, biasC, this.padding);\n      xO = this.inputConv(xO, kernelO, biasO, this.padding);\n\n      const [recKernelI, recKernelF, recKernelC, recKernelO]: tfc.Tensor[] =\n          tfc.split(\n              this.recurrentKernel.read(), numOfKernels, kernelChannelAxis);\n\n      hI = this.recurrentConv(hI, recKernelI);\n      hF = this.recurrentConv(hF, recKernelF);\n      hC = this.recurrentConv(hC, recKernelC);\n      hO = this.recurrentConv(hO, recKernelO);\n\n      const i = this.recurrentActivation.apply(tfc.add(xI, hI));\n      const f = this.recurrentActivation.apply(tfc.add(xF, hF));\n      const c = tfc.add(\n          tfc.mul(f, cTMinus1),\n          tfc.mul(i, this.activation.apply(tfc.add(xC, hC))));\n      const h = tfc.mul(\n          this.recurrentActivation.apply(tfc.add(xO, hO)),\n          this.activation.apply(c));\n\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): tfc.serialization.ConfigDict {\n    const {'units': _, ...baseConfig} = super.getConfig();\n\n    const config: tfc.serialization.ConfigDict = {\n      filters: this.filters,\n      kernelSize: this.kernelSize,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      strides: this.strides,\n    };\n\n    return {...baseConfig, ...config};\n  }\n\n  inputConv(x: Tensor, w: Tensor, b?: Tensor, padding?: PaddingMode) {\n    const out = tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, this.strides as [number, number],\n        (padding || 'valid') as 'same' | 'valid',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC',\n        this.dilationRate as [number, number]);\n\n    if (b) {\n      return K.biasAdd(out, b, this.dataFormat) as tfc.Tensor3D;\n    }\n\n    return out;\n  }\n\n  recurrentConv(x: Tensor, w: Tensor) {\n    const strides = 1;\n\n    return tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, strides, 'same',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC');\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2DCell);\n\nexport declare interface ConvLSTM2DArgs extends\n    Omit<LSTMLayerArgs, 'units'|'cell'>, ConvRNN2DLayerArgs {}\n\nexport class ConvLSTM2D extends ConvRNN2D {\n  /** @nocollapse */\n  static className = 'ConvLSTM2D';\n\n  constructor(args: ConvLSTM2DArgs) {\n    const cell = new ConvLSTM2DCell(args);\n\n    super({...args, cell} as ConvRNN2DLayerArgs);\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends tfc.serialization.Serializable>(\n      cls: tfc.serialization.SerializableConstructor<T>,\n      config: tfc.serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Basic Layers.\n */\n\nimport {any, notEqual, serialization, Tensor, tidy, transpose, util} from '@tensorflow/tfjs-core';\n\nimport {Activation as ActivationFn, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {DisposeResult, InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {LayerConfig} from '../keras_format/topology_config';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {assertPositiveInteger, mapActivationToFusedKernel} from '../utils/generic_utils';\nimport {arrayProd, range} from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface DropoutLayerArgs extends LayerArgs {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /**\n   * Integer array representing the shape of the binary dropout mask that will\n   * be multiplied with the input.\n   *\n   * For instance, if your inputs have shape `(batchSize, timesteps, features)`\n   * and you want the dropout mask to be the same for all timesteps, you can use\n   * `noise_shape=(batch_size, 1, features)`.\n   */\n  noiseShape?: number[];\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class Dropout extends Layer {\n  /** @nocollapse */\n  static className = 'Dropout';\n  private readonly rate: number;\n  private readonly noiseShape: number[];\n  private readonly seed: number;\n\n  constructor(args: DropoutLayerArgs) {\n    super(args);\n    this.rate = Math.max(Math.min(args.rate, 1), 0);\n    // So that the scalar doesn't get tidied up between executions.\n    this.noiseShape = args.noiseShape;\n    this.seed = args.seed;\n    this.supportsMasking = true;\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    if (this.noiseShape == null) {\n      return this.noiseShape;\n    }\n    const inputShape = input.shape;\n    const noiseShape: Shape = [];\n    for (let i = 0; i < this.noiseShape.length; ++i) {\n      noiseShape.push(\n          this.noiseShape[i] == null ? inputShape[i] : this.noiseShape[i]);\n    }\n    return noiseShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      if (0 < this.rate && this.rate < 1) {\n        const training =\n            kwargs['training'] == null ? false : kwargs['training'];\n        const noiseShape = this.getNoiseShape(input);\n        const output = K.inTrainPhase(\n            () => K.dropout(input, this.rate, noiseShape, this.seed),\n            () => input, training);\n        return output;\n      }\n      return inputs;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      rate: this.rate,\n      noiseShape: this.noiseShape,\n      seed: this.seed,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  dispose(): DisposeResult {\n    return super.dispose();\n  }\n}\nserialization.registerClass(Dropout);\n\nexport declare interface DenseLayerArgs extends LayerArgs {\n  /** Positive integer, dimensionality of the output space. */\n  units: number;\n  /**\n   * Activation function to use.\n   *\n   * If unspecified, no activation is applied.\n   */\n  activation?: ActivationIdentifier;\n  /** Whether to apply a bias. */\n  useBias?: boolean;\n  /**\n   * Initializer for the dense kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * If specified, defines inputShape as `[inputDim]`.\n   */\n  inputDim?: number;\n\n  /**\n   * Constraint for the kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the dense kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport interface SpatialDropout1DLayerConfig extends LayerConfig {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class SpatialDropout1D extends Dropout {\n  /** @nocollapse */\n  static className = 'SpatialDropout1D';\n\n  constructor(args: SpatialDropout1DLayerConfig) {\n    super(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    const inputShape = input.shape;\n    return [inputShape[0], 1, inputShape[2]];\n  }\n}\nserialization.registerClass(SpatialDropout1D);\n\nexport class Dense extends Layer {\n  /** @nocollapse */\n  static className = 'Dense';\n  private units: number;\n  // Default activation: Linear (none).\n  private activation: ActivationFn = null;\n  private useBias = true;\n  private kernelInitializer: Initializer;\n  private biasInitializer: Initializer;\n  private kernel: LayerVariable = null;\n  private bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n  private readonly kernelConstraint?: Constraint;\n  private readonly biasConstraint?: Constraint;\n  private readonly kernelRegularizer?: Regularizer;\n  private readonly biasRegularizer?: Regularizer;\n\n  constructor(args: DenseLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null &&\n        args.inputDim != null) {\n      // This logic is copied from Layer's constructor, since we can't\n      // do exactly what the Python constructor does for Dense().\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      this.batchInputShape = [batchSize, args.inputDim];\n    }\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(args.activation);\n    if (args.useBias != null) {\n      this.useBias = args.useBias;\n    }\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.supportsMasking = true;\n\n    this.inputSpec = [{minNDim: 2}];\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputLastDim = inputShape[inputShape.length - 1];\n    if (this.kernel == null) {\n      this.kernel = this.addWeight(\n          'kernel', [inputLastDim, this.units], null, this.kernelInitializer,\n          this.kernelRegularizer, true, this.kernelConstraint);\n      if (this.useBias) {\n        this.bias = this.addWeight(\n            'bias', [this.units], null, this.biasInitializer,\n            this.biasRegularizer, true, this.biasConstraint);\n      }\n    }\n\n    this.inputSpec = [{minNDim: 2, axes: {[-1]: inputLastDim}}];\n    this.built = true;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    outputShape[outputShape.length - 1] = this.units;\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Dense layer accepts only a single input.\n      const input = getExactlyOneTensor(inputs);\n      const fusedActivationName =\n          mapActivationToFusedKernel(this.activation.getClassName());\n      let output: Tensor;\n\n      if (fusedActivationName != null) {\n        output = K.dot(\n            input, this.kernel.read(), fusedActivationName,\n            this.bias ? this.bias.read() : null);\n      } else {\n        output = K.dot(input, this.kernel.read());\n        if (this.bias != null) {\n          output = K.biasAdd(output, this.bias.read());\n        }\n        if (this.activation != null) {\n          output = this.activation.apply(output);\n        }\n      }\n\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dense);\n\nexport declare interface FlattenLayerArgs extends LayerArgs {\n  /** Image data format: channeLast (default) or channelFirst. */\n  dataFormat?: DataFormat;\n}\n\nexport class Flatten extends Layer {\n  private dataFormat: DataFormat;\n\n  /** @nocollapse */\n  static className = 'Flatten';\n  constructor(args?: FlattenLayerArgs) {\n    args = args || {};\n    super(args);\n    this.inputSpec = [{minNDim: 3}];\n    this.dataFormat = args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    for (const dim of inputShape.slice(1)) {\n      if (dim == null) {\n        throw new ValueError(\n            `The shape of the input to \"Flatten\" is not fully defined ` +\n            `(got ${inputShape.slice(1)}). Make sure to pass a complete ` +\n            `\"input_shape\" or \"batch_input_shape\" argument to the first ` +\n            `layer in your model.`);\n      }\n    }\n    return [inputShape[0], arrayProd(inputShape, 1)];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n\n      let input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsFirst' && input.rank > 1) {\n        const permutation: number[] = [0];\n        for (let i = 2; i < input.rank; ++i) {\n          permutation.push(i);\n        }\n        permutation.push(1);\n        input = input.transpose(permutation);\n      }\n\n      return K.batchFlatten(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {};\n    if (this.dataFormat != null) {\n      config['dataFormat'] = this.dataFormat;\n    }\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Flatten);\n\nexport declare interface ActivationLayerArgs extends LayerArgs {\n  /**\n   * Name of the activation function to use.\n   */\n  activation: ActivationIdentifier;\n}\n\nexport class Activation extends Layer {\n  /** @nocollapse */\n  static className = 'Activation';\n  activation: ActivationFn;\n\n  constructor(args: ActivationLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n    this.activation = getActivation(args.activation);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      return this.activation.apply(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {activation: serializeActivation(this.activation)};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Activation);\n\nexport declare interface ReshapeLayerArgs extends LayerArgs {\n  /** The target shape. Does not include the batch axis. */\n  targetShape: Shape;\n}\n\nexport declare interface RepeatVectorLayerArgs extends LayerArgs {\n  /**\n   * The integer number of times to repeat the input.\n   */\n  n: number;\n}\n\nexport class RepeatVector extends Layer {\n  /** @nocollapse */\n  static className = 'RepeatVector';\n  readonly n: number;\n\n  constructor(args: RepeatVectorLayerArgs) {\n    super(args);\n    this.n = args.n;\n    this.inputSpec = [{ndim: 2}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], this.n, inputShape[1]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      return K.repeat(inputs, this.n);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      n: this.n,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(RepeatVector);\n\nexport class Reshape extends Layer {\n  /** @nocollapse */\n  static className = 'Reshape';\n  private targetShape: Shape;\n\n  constructor(args: ReshapeLayerArgs) {\n    super(args);\n    this.targetShape = args.targetShape;\n\n    // Make sure that all unknown dimensions are represented as `null`.\n    for (let i = 0; i < this.targetShape.length; ++i) {\n      if (this.isUnknown(this.targetShape[i])) {\n        this.targetShape[i] = null;\n      }\n    }\n  }\n\n  private isUnknown(dim: number): boolean {\n    return dim < 0 || dim == null;\n  }\n\n  /**\n   * Finds and replaces a missing dimension in output shape.\n   *\n   * This is a near direct port of the internal Numpy function\n   * `_fix_unknown_dimension` in `numpy/core/src/multiarray/shape.c`.\n   *\n   * @param inputShape: Original shape of array begin reshape.\n   * @param outputShape: Target shape of the array, with at most a single\n   * `null` or negative number, which indicates an underdetermined dimension\n   * that should be derived from `inputShape` and the known dimensions of\n   *   `outputShape`.\n   * @returns: The output shape with `null` replaced with its computed value.\n   * @throws: ValueError: If `inputShape` and `outputShape` do not match.\n   */\n  private fixUnknownDimension(inputShape: Shape, outputShape: Shape): Shape {\n    const errorMsg = 'Total size of new array must be unchanged.';\n    const finalShape = outputShape.slice();\n    let known = 1;\n    let unknown = null;\n    for (let i = 0; i < finalShape.length; ++i) {\n      const dim = finalShape[i];\n      if (this.isUnknown(dim)) {\n        if (unknown === null) {\n          unknown = i;\n        } else {\n          throw new ValueError('Can only specifiy one unknown dimension.');\n        }\n      } else {\n        known *= dim;\n      }\n    }\n\n    const originalSize = arrayProd(inputShape);\n    if (unknown !== null) {\n      if (known === 0 || originalSize % known !== 0) {\n        throw new ValueError(errorMsg);\n      }\n      finalShape[unknown] = originalSize / known;\n    } else if (originalSize !== known) {\n      throw new ValueError(errorMsg);\n    }\n\n    return finalShape;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    let anyUnknownDims = false;\n    for (let i = 0; i < inputShape.length; ++i) {\n      if (this.isUnknown(inputShape[i])) {\n        anyUnknownDims = true;\n        break;\n      }\n    }\n\n    if (anyUnknownDims) {\n      return inputShape.slice(0, 1).concat(this.targetShape);\n    } else {\n      return inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const outputShape = inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n      return input.reshape(outputShape);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      targetShape: this.targetShape,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Reshape);\n\nexport declare interface PermuteLayerArgs extends LayerArgs {\n  /**\n   * Array of integers. Permutation pattern. Does not include the\n   * sample (batch) dimension. Index starts at 1.\n   * For instance, `[2, 1]` permutes the first and second dimensions\n   * of the input.\n   */\n  dims: number[];\n}\n\nexport class Permute extends Layer {\n  /** @nocollapse */\n  static className = 'Permute';\n  readonly dims: number[];\n  private readonly dimsIncludingBatch: number[];\n\n  constructor(args: PermuteLayerArgs) {\n    super(args);\n    if (args.dims == null) {\n      throw new Error(\n          'Required configuration field `dims` is missing during Permute ' +\n          'constructor call.');\n    }\n    if (!Array.isArray(args.dims)) {\n      throw new Error(\n          'Permute constructor requires `dims` to be an Array, but received ' +\n          `${args.dims} instead.`);\n    }\n\n    // Check the validity of the permutation indices.\n    const expectedSortedIndices = range(1, args.dims.length + 1);\n    if (!util.arraysEqual(args.dims.slice().sort(), expectedSortedIndices)) {\n      throw new Error(\n          'Invalid permutation `dims`: ' + JSON.stringify(args.dims) +\n          ' `dims` must contain consecutive integers starting from 1.');\n    }\n\n    this.dims = args.dims;\n    this.dimsIncludingBatch = [0].concat(this.dims);\n    this.inputSpec = [new InputSpec({ndim: this.dims.length + 1})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    this.dims.forEach((dim: number, i: number) => {\n      outputShape[i + 1] = (inputShape as Shape)[dim];\n    });\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return transpose(getExactlyOneTensor(inputs), this.dimsIncludingBatch);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      dims: this.dims,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Permute);\n\nexport declare interface MaskingArgs extends LayerArgs {\n  /**\n   * Masking Value. Defaults to `0.0`.\n   */\n  maskValue?: number;\n}\n\nexport class Masking extends Layer {\n  /** @nocollapse */\n  static className = 'Masking';\n  maskValue: number;\n\n  constructor(args?: MaskingArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maskValue = args.maskValue == null ? 0 : args.maskValue;\n    } else {\n      this.maskValue = 0;\n    }\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig() {\n    const baseConfig = super.getConfig();\n    const config = {maskValue: this.maskValue};\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    const input = getExactlyOneTensor(inputs);\n    const axis = -1;\n    return any(notEqual(input, this.maskValue), axis);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const axis = -1;\n      const keepDims = true;\n      const booleanMask = any(notEqual(input, this.maskValue), axis, keepDims);\n      const output = input.mul(booleanMask.asType(input.dtype));\n      return output;\n    });\n  }\n}\nserialization.registerClass(Masking);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Embedding Layer.\n *\n * Original source: keras/constraints.py\n */\nimport {notEqual, serialization, Tensor, tidy, zerosLike} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface EmbeddingLayerArgs extends LayerArgs {\n  /**\n   * Integer > 0. Size of the vocabulary, i.e. maximum integer index + 1.\n   */\n  inputDim: number;\n  /**\n   * Integer >= 0. Dimension of the dense embedding.\n   */\n  outputDim: number;\n  /**\n   * Initializer for the `embeddings` matrix.\n   */\n  embeddingsInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Regularizer function applied to the `embeddings` matrix.\n   */\n  embeddingsRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Constraint function applied to the `embeddings` matrix.\n   */\n  embeddingsConstraint?: ConstraintIdentifier|Constraint;\n  /**\n   * Whether the input value 0 is a special \"padding\" value that should be\n   * masked out. This is useful when using recurrent layers which may take\n   * variable length input.\n   *\n   * If this is `True` then all subsequent layers in the model need to support\n   * masking or an exception will be raised. If maskZero is set to `True`, as a\n   * consequence, index 0 cannot be used in the vocabulary (inputDim should\n   * equal size of vocabulary + 1).\n   */\n  maskZero?: boolean;\n  /**\n   * Length of input sequences, when it is constant.\n   *\n   * This argument is required if you are going to connect `flatten` then\n   * `dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed).\n   */\n  inputLength?: number|number[];\n}\n\nexport class Embedding extends Layer {\n  /** @nocollapse */\n  static className = 'Embedding';\n  private inputDim: number;\n  private outputDim: number;\n  private embeddingsInitializer: Initializer;\n  private maskZero: boolean;\n  private inputLength: number|number[];\n\n  private embeddings: LayerVariable = null;\n\n  readonly DEFAULT_EMBEDDINGS_INITIALIZER: InitializerIdentifier =\n      'randomUniform';\n  private readonly embeddingsRegularizer?: Regularizer;\n  private readonly embeddingsConstraint?: Constraint;\n\n  constructor(args: EmbeddingLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null) {\n      // Porting Note: This logic is copied from Layer's constructor, since we\n      // can't do exactly what the Python constructor does for Embedding().\n      // Specifically, the super constructor can not be called after the\n      // mutation of the `config` argument.\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      if (args.inputLength == null) {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (None, )\n        this.batchInputShape = [batchSize, null];\n      } else {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (config.inputLength, )\n        this.batchInputShape =\n            [batchSize].concat(generic_utils.toList(args.inputLength));\n      }\n    }\n    this.inputDim = args.inputDim;\n    generic_utils.assertPositiveInteger(this.inputDim, 'inputDim');\n    this.outputDim = args.outputDim;\n    generic_utils.assertPositiveInteger(this.outputDim, 'outputDim');\n    this.embeddingsInitializer = getInitializer(\n        args.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER);\n    this.embeddingsRegularizer = getRegularizer(args.embeddingsRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.embeddingsConstraint = getConstraint(args.embeddingsConstraint);\n    this.maskZero = args.maskZero;\n    this.supportsMasking = args.maskZero;\n    this.inputLength = args.inputLength;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    this.embeddings = this.addWeight(\n        'embeddings', [this.inputDim, this.outputDim], this.dtype,\n        this.embeddingsInitializer, this.embeddingsRegularizer, true,\n        this.embeddingsConstraint);\n    this.built = true;\n  }\n\n  // Override warnOnIncompatibleInputShape because an embedding layer allows\n  // the input to have varying ranks.\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {}\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tidy(() => {\n      if (!this.maskZero) {\n        return null;\n      } else {\n        inputs = getExactlyOneTensor(inputs);\n        return notEqual(inputs, zerosLike(inputs));\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    if (this.inputLength == null) {\n      return [...inputShape, this.outputDim];\n    }\n    // inputLength can be an array if input is 3D or higher.\n    const inLens: number[] = generic_utils.toList(this.inputLength);\n    if (inLens.length !== inputShape.length - 1) {\n      throw new ValueError(\n          `\"inputLength\" is ${this.inputLength}, but received ` +\n          `input shape has shape ${inputShape}`);\n    } else {\n      let i = 0;\n      for (let k = 0; k < inLens.length; ++k) {\n        const s1 = inLens[k];\n        const s2 = inputShape[k + 1];\n        if ((s1 != null) && (s2 != null) && (s1 !== s2)) {\n          throw new ValueError(\n              `\"inputLength\" is ${this.inputLength}, but received ` +\n              `input shape has shape ${inputShape}`);\n        } else if (s1 == null) {\n          inLens[i] = s2;\n        }\n        i++;\n      }\n    }\n    return [inputShape[0], ...inLens, this.outputDim];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Embedding layer accepts only a single input.\n      let input = getExactlyOneTensor(inputs);\n      if (input.dtype !== 'int32') {\n        input = K.cast(input, 'int32');\n      }\n      const output = K.gather(this.embeddings.read(), input.as1D());\n      return output.reshape(\n          getExactlyOneShape(this.computeOutputShape(input.shape)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      inputDim: this.inputDim,\n      outputDim: this.outputDim,\n      embeddingsInitializer: serializeInitializer(this.embeddingsInitializer),\n      embeddingsRegularizer: serializeRegularizer(this.embeddingsRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      embeddingsConstraint: serializeConstraint(this.embeddingsConstraint),\n      maskZero: this.maskZero,\n      inputLength: this.inputLength\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Embedding);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Merge Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {l2Normalize} from '../losses';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as mathUtils from '../utils/math_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\n/**\n * Generic Merge layer for element-wise merge functions.\n *\n * Used to implement `Sum`, `Average`, `Concatenate`, etc.\n */\nexport abstract class Merge extends Layer {\n  protected reshapeRequired: boolean;\n\n  constructor(args?: LayerArgs) {\n    super(args || {});\n    this.supportsMasking = true;\n  }\n\n  /**\n   * Logic for merging multiple tensors, to be overridden by subclasses.\n   * @param inputs\n   */\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    throw new NotImplementedError();\n  }\n\n  /**\n   * Computes the shape of the result of an elementwise operation.\n   *\n   * @param shape1: Shape of the first tensor.\n   * @param shape2: Shape of the second tensor.\n   * @returns Expected output shape when an elementwise operation is carried\n   *   out on 2 tensors with shapes `shape1` and `shape2`.\n   * @throws ValueError: If `shape1` and `shape2` are not compatible for\n   *   element-wise operations.\n   */\n  private computeElementwiseOpOutputShape(shape1: Shape, shape2: Shape): Shape {\n    if (shape1 == null || shape2 == null) {\n      return null;\n    } else if (shape1.length < shape2.length) {\n      return this.computeElementwiseOpOutputShape(shape2, shape1);\n    } else if (shape2.length === 0) {\n      return shape1;\n    }\n    const outputShape: Shape = shape1.slice(0, shape1.length - shape2.length);\n    for (let k = 0; k < shape2.length; ++k) {\n      const i = shape1[shape1.length - shape2.length + k];\n      const j = shape2[k];\n      if (i == null || j == null || i < 0 || j < 0) {\n        outputShape.push(null);\n      } else if (i === 1) {\n        outputShape.push(j);\n      } else if (j === 1) {\n        outputShape.push(i);\n      } else {\n        if (i !== j) {\n          throw new ValueError(\n              'Operands could not be broadcast together with shapes ' +\n              JSON.stringify(shape1) + ' ' + JSON.stringify(shape2));\n        }\n        outputShape.push(i);\n      }\n    }\n    return outputShape;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.\n    if (Array.isArray(inputShape) && !Array.isArray(inputShape[0])) {\n      // Make sure that inputShape is an Array of shape.\n      inputShape = [getExactlyOneShape(inputShape)];\n    }\n    inputShape = inputShape as Shape[];\n    if (inputShape.length < 2) {\n      throw new ValueError(\n          'A merge layer should be called on an Array of at least 2 inputs.' +\n          ` Got ${inputShape.length} input(s).`);\n    }\n\n    // Make sure that there is at most one unique batch size among the input\n    // shapes.\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length > 1) {\n      throw new ValueError(\n          `Can not merge tensors with different batch sizes. ` +\n          `Got tensors with shapes: ${JSON.stringify(inputShape)}.`);\n    }\n\n    let outputShape: Shape =\n        inputShape[0] == null ? null : inputShape[0].slice(1);\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n    // If the inputs have different ranks, we have to reshape them to make them\n    // broadcastable.\n    const allRanks = inputShape.map(shape => shape.length);\n    if (inputShape.indexOf(null) === -1 &&\n        generic_utils.unique(allRanks).length === 1) {\n      this.reshapeRequired = false;\n    } else {\n      this.reshapeRequired = true;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (this.reshapeRequired) {\n        const reshapedInputs: Tensor[] = [];\n        const inputDims = inputs.map(input => input.rank);\n        if (inputDims.indexOf(null) === -1) {\n          // If ranks of all inputs are available, we simply expand each of them\n          // at axis=1 until all of them have the same rank.\n          const maxNDim = mathUtils.max(inputDims);\n          for (let x of inputs) {\n            const xNDim = x.rank;\n            for (let k = 0; k < maxNDim - xNDim; ++k) {\n              x = K.expandDims(x, 1);\n            }\n            reshapedInputs.push(x);\n          }\n          return this.mergeFunction(reshapedInputs);\n        } else {\n          // Transpose all inputs so that batch size is the last dimension.\n          // [batchSize, dim1, dim2, ...] -> [dim1, dim2, ..., batchSize]\n          let transposed = false;\n          for (const x of inputs) {\n            const xNDim = x.rank;\n            if (xNDim == null) {\n              const xShape = x.shape;\n              const batchSize = xShape[0];\n              const newShape = xShape.slice(1).concat([batchSize]);\n              let xTransposed = x.reshape(\n                  [batchSize].concat(mathUtils.arrayProd(xShape.slice(1))));\n              xTransposed = tfc.transpose(xTransposed, [1, 0]);\n              xTransposed = xTransposed.reshape(newShape);\n              reshapedInputs.push(xTransposed);\n              transposed = true;\n            } else if (xNDim > 1) {\n              const dims = mathUtils.range(1, xNDim).concat([0]);\n              reshapedInputs.push(tfc.transpose(x, dims));\n              transposed = true;\n            } else {\n              // We don't transpose inputs if they are 1D vectors or scalars.\n              reshapedInputs.push(x);\n            }\n          }\n          let y = this.mergeFunction(reshapedInputs);\n          const yNDim = y.rank;\n          if (transposed) {\n            // If inputs have been transposed, we have to transpose the output\n            // too.\n            if (yNDim == null) {\n              const yShape = y.shape;\n              const yNDim = yShape.length;\n              const batchSize = yShape[yNDim - 1];\n              const newShape =\n                  [batchSize].concat(yShape.slice(0, yShape.length - 1));\n              y = tfc.transpose(y.reshape([-1, batchSize]), [1, 0])\n                      .reshape(newShape);\n            } else if (yNDim > 1) {\n              const dims = [yNDim - 1].concat(mathUtils.range(0, yNDim - 1));\n              y = tfc.transpose(y, dims);\n            }\n          }\n          return y;\n        }\n      } else {\n        return this.mergeFunction(inputs);\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape[];\n    let outputShape: Shape;\n    if (inputShape[0] == null) {\n      outputShape = null;\n    } else {\n      outputShape = inputShape[0].slice(1);\n    }\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length === 1) {\n      outputShape = batchSizes.concat(outputShape);\n    } else {\n      outputShape = [null].concat(outputShape);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tfc.tidy(() => {\n      if (mask == null) {\n        return null;\n      }\n      if (!Array.isArray(mask)) {\n        throw new ValueError('`mask` should be an Array');\n      }\n      if (!Array.isArray(inputs)) {\n        throw new ValueError('`inputs` should be an Array');\n      }\n      if (mask.length !== inputs.length) {\n        throw new ValueError(\n            `The Array 'inputs' and 'mask' are expected to have the same ` +\n            `length, but have different lengths ` +\n            `(${inputs.length} vs ${mask.length})`);\n      }\n      if (mask.every(m => m == null)) {\n        return null;\n      }\n      mask = mask.map(m => m == null ? m : tfc.expandDims(m, 0));\n      let output = mask[0];\n      for (let i = 1; i < mask.length - 1; ++i) {\n        output = tfc.logicalAnd(output, mask[i]);\n      }\n      return output;\n    });\n  }\n}\n\nexport class Add extends Merge {\n  /** @nocollapse */\n  static className = 'Add';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Add);\n\n/**\n * Calculate the element-wise sum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Add` layer, by using no input argument\n *    or a single configuration argument. The resultant `Add` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const addLayer = tf.layers.add();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = addLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.add([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.add([input1, input2]).print();\n * // Gives [[11, 22], [33, 44]].\n *\n */\nexport function add(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Add({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Add(config);\n  }\n}\n\nexport class Multiply extends Merge {\n  /** @nocollapse */\n  static className = 'Multiply';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.mul(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Multiply);\n\n/**\n * Calculate the element-wise product of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Multiply` layer, by using no input argument\n *    or a single configuration argument. The resultant `Multiply` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const multiplyLayer = tf.layers.multiply();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = multiplyLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.multiply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.multiply([input1, input2]).print();\n * // Gives [[10, 40], [90, 160]].\n *\n */\nexport function multiply(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Multiply({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Multiply(config);\n  }\n}\n\nexport class Average extends Merge {\n  /** @nocollapse */\n  static className = 'Average';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return tfc.mul(1 / inputs.length, output);\n    });\n  }\n}\nserialization.registerClass(Average);\n\n/**\n * Calculate the element-wise arithmetic mean of inputs, which all have the same\n * shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Average` layer, by using no input argument\n *    or a single configuration argument. The resultant `Average` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const averageLayer = tf.layers.average();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = averageLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.average([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.average([input1, input2]).print();\n * // Gives [[5.5, 11], [16.5, 22]].\n *\n */\nexport function average(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Average({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Average(config);\n  }\n}\n\nexport class Maximum extends Merge {\n  /** @nocollapse */\n  static className = 'Maximum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.maximum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Maximum);\n\n/**\n * Calculate the element-wise maximum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Maximum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Maximum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const maximumLayer = tf.layers.maximum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = maximumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.maximum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.maximum([input1, input2]).print();\n * // Gives [[10, 20], [30, 40]].\n *\n */\nexport function maximum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Maximum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Maximum(config);\n  }\n}\n\nexport class Minimum extends Merge {\n  /** @nocollapse */\n  static className = 'Minimum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.minimum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Minimum);\n\n/**\n * Calculate the element-wise minimum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Minimum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Minimum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const minimumLayer = tf.layers.minimum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = minimumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.minimum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.minimum([input1, input2]).print();\n * // Gives [[1, 2], [3, 4]].\n *\n */\nexport function minimum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Minimum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Minimum(config);\n  }\n}\n\nexport declare interface ConcatenateLayerArgs extends LayerArgs {\n  /**\n   * Axis along which to concatenate.\n   */\n  axis?: number;\n}\n\nexport class Concatenate extends Merge {\n  /** @nocollapse */\n  static className = 'Concatenate';\n  readonly DEFAULT_AXIS = -1;\n  private readonly axis: number;\n\n  constructor(args?: ConcatenateLayerArgs) {\n    super(args);\n    if (args == null) {\n      args = {};\n    }\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.]\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0])) ||\n        inputShape.length === 1) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of at least 2 ' +\n          'inputs');\n    }\n    inputShape = inputShape as Shape[];\n\n    let allNoneShape = true;\n    for (const shape of inputShape) {\n      if (shape != null) {\n        allNoneShape = false;\n        break;\n      }\n    }\n    if (allNoneShape) {\n      return;\n    }\n\n    const shapeSet: Shape[] = [];\n    for (let i = 0; i < inputShape.length; ++i) {\n      const shapeWithoutConcatAxis = inputShape[i].slice();\n      shapeWithoutConcatAxis.splice(this.axis, 1);\n      let exists = false;\n      for (const shape of shapeSet) {\n        if (util.arraysEqual(shape, shapeWithoutConcatAxis)) {\n          exists = true;\n          break;\n        }\n      }\n      if (!exists) {\n        shapeSet.push(shapeWithoutConcatAxis);\n      }\n    }\n    if (shapeSet.length > 1) {\n      throw new ValueError(\n          'A `Concatenate` layer requires inputs with matching shapes ' +\n          'except for the concat axis. Got input shapes: ' +\n          JSON.stringify(inputShape));\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      return K.concatenate(inputs, this.axis);\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0]))) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of inputs.');\n    }\n    const inputShapes = inputShape as Shape[];\n    const outputShape = inputShapes[0].slice();\n    const axis = this.axis < 0 ? outputShape.length + this.axis : this.axis;\n    // Porting Note: the line above is because TypeScript doesn't support\n    //   negative indices.\n    for (const shape of inputShapes.slice(1)) {\n      if (outputShape[axis] == null || shape[axis] == null) {\n        outputShape[axis] = null;\n        break;\n      }\n      outputShape[axis] += shape[axis];\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    if (mask == null) {\n      return null;\n    }\n    if (!Array.isArray(mask)) {\n      throw new ValueError('`mask` should be an array for Concatenate');\n    }\n    if (!Array.isArray(inputs)) {\n      throw new ValueError('`inputs` should be an array for Concatenate');\n    }\n    if (mask.length !== inputs.length) {\n      throw new ValueError(\n          `Mismatch in the length of mask (${mask.length}) ` +\n          `and the legnth of inputs (${inputs.length})`);\n    }\n    return tfc.tidy(() => {\n      let allNullMasks = true;\n      mask.forEach(m => {\n        if (m != null) {\n          allNullMasks = false;\n          return;\n        }\n      });\n      if (allNullMasks) {\n        return null;\n      }\n      const outputMasks: Tensor[] = [];\n      for (let i = 0; i < inputs.length; ++i) {\n        if (mask[i] == null) {\n          // Input is unmasked. Append all 1's to masks.\n          outputMasks.push(tfc.onesLike(inputs[i]).asType('bool'));\n        } else if (mask[i].rank < inputs[i].rank) {\n          // Mask is smaller than the input, expand it.\n          outputMasks.push(tfc.expandDims(mask[i], -1));\n        } else {\n          outputMasks.push(mask[i]);\n        }\n      }\n      const concatenatedMasks = tfc.concat(outputMasks, this.axis);\n      return tfc.all(concatenatedMasks, -1, false);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axis': this.axis,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Concatenate);\n\n/**\n * Concatenate an `Array` of inputs.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Concatenate` layer, by using no input argument\n *    or a single configuration argument. The resultant `Concatenate` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const concatLayer = tf.layers.concatenate();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = concatLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 7], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = tf.layers.concatenate([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([[1, 2], [3, 4]], [2, 2]);\n * const input2 = tf.tensor2d([[10, 20], [30, 40]], [2, 2]);\n * tf.layers.concatenate([input1, input2]).print();\n * // Gives [[1, 2, 10, 20], [3, 4, 30, 40]].\n *\n */\nexport function concatenate(config?: SymbolicTensor[]|Tensor[]|\n                            ConcatenateLayerArgs): Layer|SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Concatenate({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Concatenate(config);\n  }\n}\n\nexport declare interface DotLayerArgs extends LayerArgs {\n  /**\n   * Axis or axes along which the dot product will be taken.\n   *\n   * Integer or an Array of integers.\n   */\n  axes: number|[number, number];\n\n  /**\n   * Whether to L2-normalize samples along the dot product axis\n   * before taking the dot product.\n   *\n   * If set to `true`, the output of the dot product isthe cosine\n   * proximity between the two samples.\n   */\n  normalize?: boolean;\n}\n\n/**\n * Interpretable potentially negative axis index.\n *\n * For example, given axis = -1, and dim = 3, this function will return 2.\n *\n * @param axis The axis index, may be a positive, zero or negative integer.\n * @param dim Total number of dimensions, a positive integer.\n * @returns A non-negative axis index equivalent to the input `axis`.\n */\nfunction interpretAxis(axis: number, dim: number): number {\n  while (axis < 0) {\n    axis += dim;\n  }\n  return axis;\n}\n\nfunction batchDot(x: Tensor, y: Tensor, axes: number|[number, number]): Tensor {\n  if (x.shape.length > 3 || y.shape.length > 3) {\n    throw new NotImplementedError(\n        'batchDot is not implemented for tensors of 4D or higher rank yet');\n  }\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of x to be >= 2, ` +\n          `but got ${x.shape.length}`);\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of y to be >= 2, ` +\n          `but got ${y.shape.length}`);\n\n  if (typeof axes === 'number') {\n    axes = [axes, axes];\n  }\n\n  if (x.dtype === 'complex64' || y.dtype === 'complex64') {\n    throw new NotImplementedError(\n        'batchDot is not implemented for complex64-type Tensors yet.');\n  }\n\n  const xNDim = x.shape.length;\n  const yNDim = y.shape.length;\n  if (axes == null) {\n    // Behave like batchMatmul by default.\n    axes = [xNDim - 1, yNDim - 2];\n  }\n  const axesArray = axes as [number, number];\n\n  return tfc.tidy(() => {\n    let diff: number;\n    if (xNDim > yNDim) {\n      diff = xNDim - yNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      y = y.reshape(y.shape.concat(diffShape));\n    } else if (yNDim > xNDim) {\n      diff = yNDim - xNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      x = x.reshape(x.shape.concat(diffShape));\n    } else {\n      diff = 0;\n    }\n\n    let out: Tensor;\n    if (x.shape.length === 2 && y.shape.length === 2) {\n      if (axesArray[0] === axesArray[1]) {\n        out = x.mul(y).sum(axesArray[0]);\n      } else {\n        out = x.transpose([1, 0]).mul(y).sum(axesArray[1]);\n      }\n    } else {\n      const adjX = axesArray[0] !== x.shape.length - 1;\n      const adjY = axesArray[1] === y.shape.length - 1;\n      out = x.matMul(y, adjX, adjY);\n    }\n\n    if (diff > 0) {\n      let idx: number;\n      if (xNDim > yNDim) {\n        idx = xNDim + yNDim - 3;\n      } else {\n        idx = xNDim - 1;\n      }\n      const squeezeAxes: number[] = [];\n      for (let i = idx; i < idx + diff; ++i) {\n        squeezeAxes.push(i);\n      }\n      out = out.squeeze(squeezeAxes);\n    }\n    if (out.shape.length === 1) {\n      out = out.expandDims(1);\n    }\n    return out;\n  });\n}\n\nexport class Dot extends Merge {\n  /** @nocollapse */\n  static className = 'Dot';\n\n  private axes: number|[number, number];\n  private normalize: boolean;\n\n  constructor(args: DotLayerArgs) {\n    super(args);\n    this.axes = args.axes;\n    this.normalize = args.normalize == null ? false : args.normalize;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = inputShape[0] as Shape;\n    const shape2 = inputShape[1] as Shape;\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    if (shape1[axes[0]] !== shape2[axes[1]]) {\n      throw new ValueError(\n          `Dimension incompatibility: ` +\n          `${shape1[axes[0]]} !== ${shape2[axes[1]]}`);\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    if (inputs.length !== 2) {\n      throw new ValueError(\n          'A `Dot` layer must be called on exactly 2 inputs, ' +\n          `but received ${inputs.length} input(s).`);\n    }\n\n    let x1 = inputs[0];\n    let x2 = inputs[1];\n    let axes: [number, number];\n    if (!Array.isArray(this.axes)) {\n      axes = [\n        interpretAxis(this.axes, x1.shape.length),\n        interpretAxis(this.axes, x2.shape.length)\n      ];\n    } else {\n      axes = this.axes.map(\n                 (axis, i) => interpretAxis(\n                     axis, inputs[i].shape.length)) as [number, number];\n    }\n    if (this.normalize) {\n      x1 = l2Normalize(x1, axes[0]);\n      x2 = l2Normalize(x2, axes[1]);\n    }\n    return batchDot(x1, x2, axes);\n  }\n\n  private interpretAxes(shape1: Shape, shape2: Shape): number[] {\n    let axes: number[];\n    if (!Array.isArray(this.axes)) {\n      // `this.axes` is a single integer.\n      axes = [\n        interpretAxis(this.axes, shape1.length),\n        interpretAxis(this.axes, shape2.length)\n      ];\n    } else {\n      // `this.axes` is an Array of integers.\n      axes = this.axes;\n    }\n    return axes;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = (inputShape[0] as Shape).slice();\n    const shape2 = (inputShape[1] as Shape).slice();\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    shape1.splice(axes[0], 1);\n    shape2.splice(axes[1], 1);\n    shape2.splice(0, 1);\n    const outputShape = shape1.concat(shape2);\n    if (outputShape.length === 1) {\n      outputShape.push(1);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return null;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axes': this.axes,\n      'normalize': this.normalize\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dot);\n\n// TODO(cais): Add functional interfaces for the merge layers.\n", "/**\r\n * @license\r\n * Copyright 2018 Google LLC\r\n *\r\n * Use of this source code is governed by an MIT-style\r\n * license that can be found in the LICENSE file or at\r\n * https://opensource.org/licenses/MIT.\r\n * =============================================================================\r\n */\r\n\r\n/**\r\n * TensorFlow.js Layers: Noise Layers.\r\n */\r\n\r\nimport {greaterEqual, randomUniform, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\r\n\r\nimport * as K from '../backend/tfjs_backend';\r\nimport {Layer, LayerArgs} from '../engine/topology';\r\nimport {Shape} from '../keras_format/common';\r\nimport {Kwargs} from '../types';\r\nimport {getExactlyOneTensor} from '../utils/types_utils';\r\n\r\nexport declare interface GaussianNoiseArgs extends LayerArgs {\r\n  /** Standard Deviation.  */\r\n  stddev: number;\r\n}\r\n\r\nexport class GaussianNoise extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianNoise';\r\n  readonly stddev: number;\r\n\r\n  constructor(args: GaussianNoiseArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.stddev = args.stddev;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {stddev: this.stddev};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      const noised = () =>\r\n          K.randomNormal(input.shape, 0, this.stddev).add(input);\r\n      const output =\r\n          K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      return output;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianNoise);\r\n\r\nexport declare interface GaussianDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n}\r\n\r\nexport class GaussianDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianDropout';\r\n  readonly rate: number;\r\n\r\n  constructor(args: GaussianDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      if (this.rate > 0 && this.rate < 1) {\r\n        const noised = () => {\r\n          const stddev = Math.sqrt(this.rate / (1 - this.rate));\r\n          return input.mul(K.randomNormal(input.shape, 1, stddev));\r\n        };\r\n        return K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      }\r\n      return input;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianDropout);\r\n\r\nexport declare interface AlphaDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n  /**\r\n   * A 1-D `Tensor` of type `int32`, representing the\r\n   * shape for randomly generated keep/drop flags.\r\n   */\r\n  noiseShape?: Shape;\r\n}\r\n\r\n/**\r\n * Applies Alpha Dropout to the input.\r\n *\r\n * As it is a regularization layer, it is only active at training time.\r\n *\r\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\r\n * to their original values, in order to ensure the self-normalizing property\r\n * even after this dropout.\r\n * Alpha Dropout fits well to Scaled Exponential Linear Units\r\n * by randomly setting activations to the negative saturation value.\r\n *\r\n * Arguments:\r\n *   - `rate`: float, drop probability (as with `Dropout`).\r\n *     The multiplicative noise will have\r\n *     standard deviation `sqrt(rate / (1 - rate))`.\r\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\r\n *     shape for randomly generated keep/drop flags.\r\n *\r\n * Input shape:\r\n *   Arbitrary. Use the keyword argument `inputShape`\r\n *   (tuple of integers, does not include the samples axis)\r\n *   when using this layer as the first layer in a model.\r\n *\r\n * Output shape:\r\n *   Same shape as input.\r\n *\r\n * References:\r\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\r\n */\r\nexport class AlphaDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'AlphaDropout';\r\n  readonly rate: number;\r\n  readonly noiseShape: Shape;\r\n\r\n  constructor(args: AlphaDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n    this.noiseShape = args.noiseShape;\r\n  }\r\n\r\n  _getNoiseShape(inputs: Tensor|Tensor[]) {\r\n    return this.noiseShape || getExactlyOneTensor(inputs).shape;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      if (this.rate < 1 && this.rate > 0) {\r\n        const noiseShape = this._getNoiseShape(inputs);\r\n\r\n        const droppedInputs = () => {\r\n          const input = getExactlyOneTensor(inputs);\r\n\r\n          const alpha = 1.6732632423543772848170429916717;\r\n          const scale = 1.0507009873554804934193349852946;\r\n\r\n          const alphaP = -alpha * scale;\r\n\r\n          let keptIdx = greaterEqual(randomUniform(noiseShape), this.rate);\r\n\r\n          keptIdx = K.cast(keptIdx, 'float32');  // get default dtype.\r\n\r\n          // Get affine transformation params.\r\n          const a = ((1 - this.rate) * (1 + this.rate * alphaP ** 2)) ** -0.5;\r\n          const b = -a * alphaP * this.rate;\r\n\r\n          // Apply mask.\r\n          const x = input.mul(keptIdx).add(keptIdx.add(-1).mul(alphaP));\r\n\r\n          return x.mul(a).add(b);\r\n        };\r\n        return K.inTrainPhase(\r\n            droppedInputs, () => getExactlyOneTensor(inputs),\r\n            kwargs['training'] || false);\r\n      }\r\n      return inputs;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(AlphaDropout);\r\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Normalization layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {moments, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Applies batch normalization on x given mean, var, beta and gamma.\n *\n * I.e. returns:\n *   `output = (x - mean) / (sqrt(var) + epsilon) * gamma + beta`\n *\n * @param x Input tensor.\n * @param mean Mean of batch.\n * @param variance Variance of batch.\n * @param beta Tensor with which to center the input.\n * @param gamma Tensor by which to scale the input.\n * @param epsilon Fuzz factor.\n * @returns The result of the batch normalization.\n */\nexport function batchNormalization(\n    x: Tensor, mean: Tensor, variance: Tensor, beta?: Tensor, gamma?: Tensor,\n    epsilon = 1e-3): Tensor {\n  let out: Tensor;\n  if (x.rank === 2) {\n    out = tfc.batchNorm2d(\n        x as Tensor2D, mean as Tensor2D | Tensor1D,\n        variance as Tensor2D | Tensor1D, beta as Tensor2D | Tensor1D,\n        gamma as Tensor2D | Tensor1D, epsilon);\n  } else if (x.rank === 3) {\n    // TODO(cais): Check rank; give proper error message.\n    out = tfc.batchNorm3d(\n        x as Tensor3D, mean as Tensor3D | Tensor1D,\n        variance as Tensor3D | Tensor1D, beta as Tensor3D | Tensor1D,\n        gamma as Tensor3D | Tensor1D, epsilon);\n  } else if (x.rank === 4) {\n    out = tfc.batchNorm4d(\n        x as Tensor4D, mean as Tensor4D | Tensor1D,\n        variance as Tensor4D | Tensor1D, beta as Tensor4D | Tensor1D,\n        gamma as Tensor4D | Tensor1D, epsilon);\n  } else {\n    throw new NotImplementedError(\n        `batchNormalization is not implemented for array of rank ${x.rank} ` +\n        `yet`);\n  }\n  return out;\n}\n\n/**\n * Non-broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction regularNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const normed =\n               batchNormalization(x, mean, variance, beta, gamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction broadcastNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const targetShape: number[] = [];\n           for (const axis of math_utils.range(0, x.rank)) {\n             if (reductionAxes.indexOf(axis) !== -1) {\n               targetShape.push(1);\n             } else {\n               targetShape.push(x.shape[axis]);\n             }\n           }\n           const broadcastMean = mean.reshape(targetShape);\n           const broadcastVariance = variance.reshape(targetShape);\n           const broadcastGamma =\n               gamma == null ? null : gamma.reshape(targetShape);\n           const broadcastBeta =\n               beta == null ? null : beta.reshape(targetShape);\n           const normed = batchNormalization(\n               x, broadcastMean, broadcastVariance, broadcastBeta,\n               broadcastGamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Batch normalization for use in training (not inference).\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nexport function normalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  if (util.arraysEqual(\n          reductionAxes.slice().sort(), math_utils.range(0, x.rank - 1))) {\n    return regularNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  } else {\n    return broadcastNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  }\n}\n\nexport declare interface BatchNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The integer axis that should be normalized (typically the features axis).\n   * Defaults to -1.\n   *\n   * For instance, after a `Conv2D` layer with `data_format=\"channels_first\"`,\n   * set `axis=1` in `batchNormalization`.\n   */\n  axis?: number;\n\n  /**\n   * Momentum of the moving average. Defaults to 0.99.\n   */\n  momentum?: number;\n\n  /**\n   * Small float added to the variance to avoid dividing by zero. Defaults to\n   * 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Defaults to `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear (also e.g. `nn.relu`),\n   * this can be disabled since the scaling will be done by the next layer.\n   * Defaults to `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   *  Defaults to 'zeros'.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   *  Defaults to `ones`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving mean.\n   * Defaults to `zeros`\n   */\n  movingMeanInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving variance.\n   *  Defaults to 'Ones'.\n   */\n  movingVarianceInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the beta weight.\n   */\n  betaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for gamma weight.\n   */\n  gammaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer for the beta weight.\n   */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer for the gamma weight.\n   */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class BatchNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'BatchNormalization';\n  private readonly axis: number;\n  private readonly momentum: number;\n  private readonly epsilon: number;\n  private readonly center: boolean;\n  private readonly scale: boolean;\n  private readonly betaInitializer: Initializer;\n  private readonly gammaInitializer: Initializer;\n  private readonly movingMeanInitializer: Initializer;\n  private readonly movingVarianceInitializer: Initializer;\n  private readonly betaConstraint: Constraint;\n  private readonly gammaConstraint: Constraint;\n  private readonly betaRegularizer: Regularizer;\n  private readonly gammaRegularizer: Regularizer;\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n  private movingMean: LayerVariable;\n  private movingVariance: LayerVariable;\n\n  constructor(args?: BatchNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.supportsMasking = true;\n    this.axis = args.axis == null ? -1 : args.axis;\n    this.momentum = args.momentum == null ? 0.99 : args.momentum;\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.movingMeanInitializer =\n        getInitializer(args.movingMeanInitializer || 'zeros');\n    this.movingVarianceInitializer =\n        getInitializer(args.movingVarianceInitializer || 'ones');\n    this.betaConstraint = getConstraint(args.betaConstraint);\n    this.gammaConstraint = getConstraint(args.gammaConstraint);\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const axis = this.axis >= 0 ? this.axis : (this.axis + inputShape.length);\n    const dim = inputShape[axis];\n    if (dim == null) {\n      throw new ValueError(\n          `Axis ${axis} of input tensor should have a defined dimension but ` +\n          `the layer received an input with shape ` +\n          `${JSON.stringify(inputShape)}.`);\n    }\n    this.inputSpec =\n        [new InputSpec({ndim: inputShape.length, axes: {[axis]: dim}})];\n    const shape = [dim];\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', shape, null, this.gammaInitializer, this.gammaRegularizer,\n          true, this.gammaConstraint);\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', shape, null, this.betaInitializer, this.betaRegularizer, true,\n          this.betaConstraint);\n    }\n    this.movingMean = this.addWeight(\n        'moving_mean', shape, null, this.movingMeanInitializer, null, false);\n    this.movingVariance = this.addWeight(\n        'moving_variance', shape, null, this.movingVarianceInitializer, null,\n        false);\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const ndim = inputShape.length;\n      const reductionAxes = math_utils.range(0, ndim);\n      const axis = this.axis >= 0 ? this.axis : (this.axis + ndim);\n      reductionAxes.splice(axis, 1);\n      const broadcastShape = generic_utils.pyListRepeat(1, ndim);\n      broadcastShape[axis] = inputShape[axis];\n\n      const sortedReductionAxes = reductionAxes.slice();\n      sortedReductionAxes.sort();\n      const needsBroadcasting = !util.arraysEqual(\n          sortedReductionAxes, math_utils.range(0, ndim).slice(0, ndim - 1));\n\n      const normalizeInference: () => Tensor = () => {\n        if (needsBroadcasting) {\n          const broadcastMovingMean =\n              this.movingMean.read().reshape(broadcastShape);\n          const broadcastMovingVariance =\n              this.movingVariance.read().reshape(broadcastShape);\n          const broadcastBeta =\n              this.center ? this.beta.read().reshape(broadcastShape) : null;\n          const broadcastGamma =\n              this.scale ? this.gamma.read().reshape(broadcastShape) : null;\n          return batchNormalization(\n              input, broadcastMovingMean, broadcastMovingVariance,\n              broadcastBeta, broadcastGamma, this.epsilon);\n        } else {\n          return batchNormalization(\n              input, this.movingMean.read(), this.movingVariance.read(),\n              this.beta == null ? null : this.beta.read(),\n              this.gamma == null ? null : this.gamma.read(), this.epsilon);\n        }\n      };\n\n      if (!training) {\n        return normalizeInference();\n      }\n\n      const [normedTraining, mean, variance] = normalizeBatchInTraining(\n          input, this.gamma.read(), this.beta.read(), reductionAxes,\n          this.epsilon);\n\n      const doMovingAverage =\n          (variable: LayerVariable, value: Tensor, momentum: number): void => {\n            tfc.tidy(() => {\n              const decay = 1 - momentum;\n              const origValue = variable.read();\n              const updateDelta = origValue.sub(value).mul(decay);\n              variable.write(origValue.sub(updateDelta));\n            });\n          };\n\n      // Perform updates to moving mean and moving variance for training.\n      // Porting Note: In PyKeras, these updates to `movingMean` and\n      //   `movingAverage` are done as a deferred Graph, added to the `Layer`'s\n      //   `update`s using the `add_update()` method. Here we do it imperatively\n      //   and encapsulate the updates in a function that is invoked\n      //   immediately.\n      const updateMovingMeanAndVariance = () => {\n        doMovingAverage(this.movingMean, mean, this.momentum);\n        doMovingAverage(this.movingVariance, variance, this.momentum);\n      };\n      updateMovingMeanAndVariance();\n\n      return normedTraining;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      momentum: this.momentum,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      movingMeanInitializer: serializeInitializer(this.movingMeanInitializer),\n      movingVarianceInitializer:\n          serializeInitializer(this.movingVarianceInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer),\n      betaConstraint: serializeConstraint(this.betaConstraint),\n      gammaConstraint: serializeConstraint(this.gammaConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(BatchNormalization);\n\nexport interface LayerNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The axis or axes that should be normalized (typically, the feature axis.)\n   * Defaults to -1 (the last axis.)\n   */\n  axis?: number|number[];\n\n  /**\n   * A small positive float added to variance to avoid divison by zero.\n   * Defaults to 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Default: `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply output by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear, this can be disabled since scaling will\n   * be done by the next layer.\n   * Default: `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   * Default: `'zeros'`.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   * Default: `'ones'`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /** Regularizer for the beta weight. */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /** Regularizer for the gamma weight. */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class LayerNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'LayerNormalization';\n\n  private axis: number|number[];\n  readonly epsilon: number;\n  readonly center: boolean;\n  readonly scale: boolean;\n  readonly betaInitializer: Initializer;\n  readonly gammaInitializer: Initializer;\n  readonly betaRegularizer: Regularizer;\n  readonly gammaRegularizer: Regularizer;\n\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n\n  constructor(args?: LayerNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.axis = args.axis == null ? -1 : args.axis;\n    if (typeof this.axis === 'number') {\n      if (!Number.isInteger(this.axis)) {\n        throw new Error(\n            `Expected axis to be an integer, but received ${this.axis}`);\n      }\n    } else if (Array.isArray(this.axis)) {\n      for (const axis of this.axis) {\n        if (!Number.isInteger(axis)) {\n          throw new Error(\n              `Expected axis to be an array of integers, ` +\n              `but received ${JSON.stringify(this.axis)}`);\n        }\n      }\n    } else {\n      throw new Error(\n          `Expected axis to be an integer or an array of integers, ` +\n          `but received ${JSON.stringify(this.axis)}`);\n    }\n\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n\n    this.supportsMasking = true;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const nDims = inputShape.length;\n\n    // Convert axis to array and resolve negatives.\n    if (typeof this.axis === 'number') {\n      this.axis = [this.axis];\n    }\n    for (let i = 0; i < this.axis.length; ++i) {\n      if (this.axis[i] < 0) {\n        this.axis[i] += nDims;\n      }\n    }\n\n    // Further validate axes.\n    for (const axis of this.axis) {\n      if (axis < 0 || axis >= nDims) {\n        throw new Error(`Invalid axis: ${axis}`);\n      }\n    }\n    if (this.axis.length !== generic_utils.unique(this.axis).length) {\n      throw new Error(`Found duplicate axes in: ${this.axis}`);\n    }\n\n    const paramShape = this.axis.map(axis => inputShape[axis]) as number[];\n\n    const trainable = true;\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', paramShape, 'float32', this.gammaInitializer,\n          this.gammaRegularizer, trainable);\n    } else {\n      this.gamma = null;\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', paramShape, 'float32', this.betaInitializer,\n          this.betaRegularizer, trainable);\n    } else {\n      this.beta = null;\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const input = getExactlyOneTensor(inputs);\n    const inputShape = input.shape;\n    const nDims = inputShape.length;\n\n    return tidy(() => {\n      const keepDims = true;\n      let {mean, variance} = moments(input, this.axis, keepDims);\n      const broadcastShape = generic_utils.pyListRepeat(1, nDims);\n      for (const dim of this.axis as number[]) {\n        broadcastShape[dim] = inputShape[dim];\n      }\n\n      const broadcast = (v: Tensor) => {\n        if (v != null && v.shape.length !== nDims &&\n            this.axis !== [nDims - 1]) {\n          return v.reshape(broadcastShape);\n        } else {\n          return v;\n        }\n      };\n\n      let scale = broadcast(this.gamma.read());\n      let offset = broadcast(this.beta.read());\n\n      // TODO(https://github.com/tensorflow/tfjs/issues/2120): The tiling below\n      // is a workaround for the limitation of core's batchNormalization?d don't\n      // support broadcasting in their gradients. In addition, the tiling is\n      // necessary to ensure correctness on the browser CPU backend regardless\n      // of forward or backward computation. Remove this workaround once the\n      // limitation is addressed. See .\n      const momentsTiling: number[] = [];\n      const scaleOffsetTiling: number[] = [];\n      for (let i = 0; i < nDims; ++i) {\n        if ((this.axis as number[]).indexOf(i) !== -1) {\n          momentsTiling.push(inputShape[i]);\n          scaleOffsetTiling.push(1);\n        } else {\n          momentsTiling.push(1);\n          scaleOffsetTiling.push(inputShape[i]);\n        }\n      }\n      mean = mean.tile(momentsTiling);\n      variance = variance.tile(momentsTiling);\n      scale = scale.tile(scaleOffsetTiling);\n      offset = offset.tile(scaleOffsetTiling);\n\n      return batchNormalization(\n          input, mean, variance, offset, scale, this.epsilon);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LayerNormalization);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Padding Layers.\n */\n\n// Porting Note: In Python Keras, the padding layers are in convolutional.py,\n//   but we decided to put them in a separate file (padding.ts) for clarity.\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\n/**\n * Pads the middle dimension of a 3D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of 2 integers, how many zeros to add at the start and\n *   end of the middle dimension (i.e., dimension 1).\n * @return A padded 3D `tf.Tensor`.\n */\nexport function temporalPadding(x: Tensor, padding?: [number, number]): Tensor {\n  return tidy(() => {\n    if (x.rank !== 3) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 3-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [1, 1];\n    }\n    if (padding.length !== 2) {\n      throw new ValueError(\n          `temporalPadding expects input padding pattern to be a length-2 ` +\n          `array, but received a length-${padding.length} array.`);\n    }\n\n    const pattern: Array<[number, number]> = [[0, 0], padding, [0, 0]];\n    return tfc.pad(x, pattern);\n  });\n}\n\n/**\n * Pads the 2nd and 3rd dimensions of a 4D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of two `Array`s, each of which is an `Array` of two\n *   integers. The amount of padding at the beginning and end of the 2nd and 3rd\n *   dimensions, respectively.\n * @param dataFormat 'channelsLast' (default) or 'channelsFirst'.\n * @return Padded 4D `tf.Tensor`.\n */\nexport function spatial2dPadding(\n    x: Tensor, padding?: [[number, number], [number, number]],\n    dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 4-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [[1, 1], [1, 1]];\n    }\n    if (padding.length !== 2 || padding[0].length !== 2 ||\n        padding[1].length !== 2) {\n      throw new ValueError(\n          'spatial2dPadding expects `padding` to be an Array of two Arrays, ' +\n          'each of which is an Array of two integers.');\n    }\n\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (dataFormat !== 'channelsLast' && dataFormat !== 'channelsFirst') {\n      throw new ValueError(\n          `Unknown data format: ${dataFormat}. ` +\n          `Supported data formats are 'channelsLast' and 'channelsFirst.`);\n    }\n\n    let pattern: Array<[number, number]>;\n    if (dataFormat === 'channelsFirst') {\n      pattern = [[0, 0], [0, 0], padding[0], padding[1]];\n    } else {\n      pattern = [[0, 0], padding[0], padding[1], [0, 0]];\n    }\n\n    return tfc.pad(x, pattern);\n  });\n}\n\nexport declare interface ZeroPadding2DLayerArgs extends LayerArgs {\n  /**\n   * Integer, or `Array` of 2 integers, or `Array` of 2 `Array`s, each of\n   * which is an `Array` of 2 integers.\n   * - If integer, the same symmetric padding is applied to width and height.\n   * - If Array` of 2 integers, interpreted as two different symmetric values\n   *   for height and width:\n   *   `[symmetricHeightPad, symmetricWidthPad]`.\n   * - If `Array` of 2 `Array`s, interpreted as:\n   *   `[[topPad, bottomPad], [leftPad, rightPad]]`.\n   */\n  padding?: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * One of `'channelsLast'` (default) and `'channelsFirst'`.\n   *\n   * The ordering of the dimensions in the inputs.\n   * `channelsLast` corresponds to inputs with shape\n   * `[batch, height, width, channels]` while `channelsFirst`\n   * corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class ZeroPadding2D extends Layer {\n  /** @nocollapse */\n  static className = 'ZeroPadding2D';\n  readonly dataFormat: DataFormat;\n  readonly padding: [[number, number], [number, number]];\n\n  constructor(args?: ZeroPadding2DLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.dataFormat =\n        args.dataFormat == null ? imageDataFormat() : args.dataFormat;\n    // TODO(cais): Maybe refactor the following logic surrounding `padding`\n    //   into a helper method.\n    if (args.padding == null) {\n      this.padding = [[1, 1], [1, 1]];\n    } else if (typeof args.padding === 'number') {\n      this.padding =\n          [[args.padding, args.padding], [args.padding, args.padding]];\n    } else {\n      args.padding = args.padding;\n      if (args.padding.length !== 2) {\n        throw new ValueError(\n            `ZeroPadding2D expects padding to be a length-2 array, but ` +\n            `received a length-${args.padding.length} array.`);\n      }\n\n      let heightPadding: [number, number];\n      let widthPadding: [number, number];\n      if (typeof args.padding[0] === 'number') {\n        heightPadding = [args.padding[0], args.padding[0]];\n        widthPadding = [args.padding[1] as number, args.padding[1] as number];\n      } else {\n        args.padding = args.padding as [[number, number], [number, number]];\n\n        if (args.padding[0].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects height padding to be a length-2 array, ` +\n              `but received a length-${args.padding[0].length} array.`);\n        }\n        heightPadding = args.padding[0] as [number, number];\n\n        if (args.padding[1].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects width padding to be a length-2 array, ` +\n              `but received a length-${args.padding[1].length} array.`);\n        }\n        widthPadding = args.padding[1] as [number, number];\n      }\n      this.padding = [heightPadding, widthPadding];\n    }\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n\n    let rows: number;\n    let cols: number;\n    if (this.dataFormat === 'channelsFirst') {\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        rows = inputShape[2] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[3] != null && inputShape[3] >= 0) {\n        cols = inputShape[3] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      if (inputShape[1] != null && inputShape[1] >= 0) {\n        rows = inputShape[1] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        cols = inputShape[2] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(\n        () => spatial2dPadding(\n            getExactlyOneTensor(inputs), this.padding, this.dataFormat));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ZeroPadding2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Pooling Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor3D, Tensor4D, Tensor5D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode, checkPoolMode} from '../common';\nimport {InputSpec} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, PaddingMode, PoolMode, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\nimport {preprocessConv2DInput, preprocessConv3DInput} from './convolutional';\n\n/**\n * 2D pooling.\n * @param x\n * @param poolSize\n * @param stridesdes strides. Defaults to [1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 2D pooling.\n */\nexport function pool2d(\n    x: Tensor, poolSize: [number, number], strides?: [number, number],\n    padding?: PaddingMode, dataFormat?: DataFormat,\n    poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // TODO(cais): Remove the preprocessing step once deeplearn.js supports\n    // dataFormat as an input argument.\n    x = preprocessConv2DInput(x, dataFormat);  // x is NHWC after preprocessing.\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      // TODO(cais): Rank check?\n      y = tfc.maxPool(x as Tensor4D, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      // TODO(cais): Check the dtype and rank of x and give clear error message\n      //   if those are incorrect.\n      y = tfc.avgPool(\n          // TODO(cais): Rank check?\n          x as Tensor3D | Tensor4D, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);  // NHWC -> NCHW.\n    }\n    return y;\n  });\n}\n\n/**\n * 3D pooling.\n * @param x\n * @param poolSize. Default to [1, 1, 1].\n * @param strides strides. Defaults to [1, 1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 3D pooling.\n */\nexport function pool3d(\n    x: Tensor5D, poolSize: [number, number, number],\n    strides?: [number, number, number], padding?: PaddingMode,\n    dataFormat?: DataFormat, poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // x is NDHWC after preprocessing.\n    x = preprocessConv3DInput(x as Tensor, dataFormat) as Tensor5D;\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      y = tfc.maxPool3d(x, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      y = tfc.avgPool3d(x, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);  // NDHWC -> NCDHW.\n    }\n    return y;\n  });\n}\n\nexport declare interface Pooling1DLayerArgs extends LayerArgs {\n  /**\n   * Size of the window to pool over, should be an integer.\n   */\n  poolSize?: number|[number];\n  /**\n   * Period at which to sample the pooled values.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number];\n  /** How to fill in data that's not an integer multiple of poolSize. */\n  padding?: PaddingMode;\n}\n\n/**\n * Abstract class for different pooling 1D layers.\n */\nexport abstract class Pooling1D extends Layer {\n  protected readonly poolSize: [number];\n  protected readonly strides: [number];\n  protected readonly padding: PaddingMode;\n\n  /**\n   *\n   * @param args Parameters for the Pooling layer.\n   *\n   * config.poolSize defaults to 2.\n   */\n  constructor(args: Pooling1DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = 2;\n    }\n    super(args);\n    if (typeof args.poolSize === 'number') {\n      this.poolSize = [args.poolSize];\n    } else if (\n        Array.isArray(args.poolSize) &&\n        (args.poolSize as number[]).length === 1 &&\n        typeof (args.poolSize as number[])[0] === 'number') {\n      this.poolSize = args.poolSize;\n    } else {\n      throw new ValueError(\n          `poolSize for 1D convolutional layer must be a number or an ` +\n          `Array of a single number, but received ` +\n          `${JSON.stringify(args.poolSize)}`);\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else {\n      if (typeof args.strides === 'number') {\n        this.strides = [args.strides];\n      } else if (\n          Array.isArray(args.strides) &&\n          (args.strides as number[]).length === 1 &&\n          typeof (args.strides as number[])[0] === 'number') {\n        this.strides = args.strides;\n      } else {\n        throw new ValueError(\n            `strides for 1D convolutional layer must be a number or an ` +\n            `Array of a single number, but received ` +\n            `${JSON.stringify(args.strides)}`);\n      }\n    }\n    assertPositiveInteger(this.strides, 'strides');\n\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const length = convOutputLength(\n        inputShape[1], this.poolSize[0], this.padding, this.strides[0]);\n    return [inputShape[0], length, inputShape[2]];\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Add dummy last dimension.\n      inputs = K.expandDims(getExactlyOneTensor(inputs), 2);\n      const output = this.poolingFunction(\n          getExactlyOneTensor(inputs), [this.poolSize[0], 1],\n          [this.strides[0], 1], this.padding, 'channelsLast');\n      // Remove dummy last dimension.\n      return tfc.squeeze(output, [2]);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'MaxPooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling1D);\n\nexport class AveragePooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'AveragePooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling1D);\n\nexport declare interface Pooling2DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [vertical, horizontal].\n   * Expects an integer or an array of 2 integers.\n   *\n   * For example, `[2, 2]` will halve the input in both spatial dimension.\n   * If only one integer is specified, the same window length\n   * will be used for both dimensions.\n   */\n  poolSize?: number|[number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 2 integers. Integer, tuple of 2 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 2D layers.\n */\nexport abstract class Pooling2D extends Layer {\n  protected readonly poolSize: [number, number];\n  protected readonly strides: [number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling2DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 2) {\n        throw new ValueError(\n            `If the strides property of a 2D pooling layer is an Array, ` +\n            `it is expected to have a length of 2, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    rows =\n        convOutputLength(rows, this.poolSize[0], this.padding, this.strides[0]);\n    cols =\n        convOutputLength(cols, this.poolSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'MaxPooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling2D);\n\nexport class AveragePooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'AveragePooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling2D);\n\nexport declare interface Pooling3DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [depth, height, width].\n   * Expects an integer or an array of 3 integers.\n   *\n   * For example, `[2, 2, 2]` will halve the input in three dimensions.\n   * If only one integer is specified, the same window length\n   * will be used for all dimensions.\n   */\n  poolSize?: number|[number, number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 3 integers. Integer, tuple of 3 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 3D layers.\n */\nexport abstract class Pooling3D extends Layer {\n  protected readonly poolSize: [number, number, number];\n  protected readonly strides: [number, number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling3DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 3) {\n        throw new ValueError(\n            `If the strides property of a 3D pooling layer is an Array, ` +\n            `it is expected to have a length of 3, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let depths =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[4] : inputShape[3];\n    depths = convOutputLength(\n        depths, this.poolSize[0], this.padding, this.strides[0]);\n    rows =\n        convOutputLength(rows, this.poolSize[1], this.padding, this.strides[1]);\n    cols =\n        convOutputLength(cols, this.poolSize[2], this.padding, this.strides[2]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], depths, rows, cols];\n    } else {\n      return [inputShape[0], depths, rows, cols, inputShape[4]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'MaxPooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling3D);\n\nexport class AveragePooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'AveragePooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling3D);\n\n/**\n * Abstract class for different global pooling 1D layers.\n */\nexport abstract class GlobalPooling1D extends Layer {\n  constructor(args: LayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], inputShape[2]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n}\n\nexport class GlobalAveragePooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling1D';\n  constructor(args?: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.mean(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling1D);\n\nexport class GlobalMaxPooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling1D';\n  constructor(args: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.max(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling1D);\n\nexport declare interface GlobalPooling2DLayerArgs extends LayerArgs {\n  /**\n   * One of `CHANNEL_LAST` (default) or `CHANNEL_FIRST`.\n   *\n   * The ordering of the dimensions in the inputs. `CHANNEL_LAST` corresponds\n   * to inputs with shape `[batch, height, width, channels[` while\n   * `CHANNEL_FIRST` corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different global pooling 2D layers.\n */\nexport abstract class GlobalPooling2D extends Layer {\n  protected dataFormat: DataFormat;\n  constructor(args: GlobalPooling2DLayerArgs) {\n    super(args);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape;\n    if (this.dataFormat === 'channelsLast') {\n      return [inputShape[0], inputShape[3]];\n    } else {\n      return [inputShape[0], inputShape[1]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class GlobalAveragePooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.mean(input, [1, 2]);\n      } else {\n        return tfc.mean(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling2D);\n\nexport class GlobalMaxPooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.max(input, [1, 2]);\n      } else {\n        return tfc.max(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Layers that augment the functionality of a base layer.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {InputSpec, Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {BidirectionalMergeMode, Shape, VALID_BIDIRECTIONAL_MERGE_MODES} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {RegularizerFn, RnnStepFunction} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {rnn, RNN, standardizeArgs} from './recurrent';\nimport {deserialize} from './serialization';\n\nexport declare interface WrapperLayerArgs extends LayerArgs {\n  /**\n   * The layer to be wrapped.\n   */\n  layer: Layer;\n}\n\n/**\n * Abstract wrapper base class.\n *\n * Wrappers take another layer and augment it in various ways.\n * Do not use this class as a layer, it is only an abstract base class.\n * Two usable wrappers are the `TimeDistributed` and `Bidirectional` wrappers.\n */\nexport abstract class Wrapper extends Layer {\n  readonly layer: Layer;\n\n  constructor(args: WrapperLayerArgs) {\n    // Porting Note: In PyKeras, `self.layer` is set prior to the calling\n    //   `super()`. But we can't do that here due to TypeScript's restriction.\n    //   See: https://github.com/Microsoft/TypeScript/issues/8277\n    //   As a result, we have to add checks in `get trainable()` and\n    //   `set trainable()` below in order to prevent using `this.layer` when\n    //   its value is `undefined`. The super constructor does use the getter\n    //   and the setter of `this.layer`.\n    super(args);\n    this.layer = args.layer;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    this.built = true;\n  }\n\n  // TODO(cais): Implement activityRegularizer getter.\n\n  get trainable(): boolean {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      return this.layer.trainable;\n    } else {\n      return false;\n    }\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      this.layer.trainable = value;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.layer.trainableWeights;\n  }\n  // TODO(cais): Implement setter for trainableWeights.\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.layer.nonTrainableWeights;\n  }\n  // TODO(cais): Implement setter for nonTrainableWeights.\n\n  get updates(): Tensor[] {\n    // tslint:disable-next-line:no-any\n    return (this.layer as any)._updates;\n  }\n\n  // TODO(cais): Implement getUpdatesFor().\n\n  get losses(): RegularizerFn[] {\n    return this.layer.losses;\n  }\n\n  // TODO(cais): Implement getLossesFor().\n\n  getWeights(): Tensor[] {\n    return this.layer.getWeights();\n  }\n\n  setWeights(weights: Tensor[]): void {\n    this.layer.setWeights(weights);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'layer': {\n        'className': this.layer.getClassName(),\n        'config': this.layer.getConfig(),\n      }\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.layer != null) {\n      this.layer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const layerConfig = config['layer'] as serialization.ConfigDict;\n    const layer = deserialize(layerConfig, customObjects) as Layer;\n    delete config['layer'];\n    const newConfig = {layer};\n    Object.assign(newConfig, config);\n    return new cls(newConfig);\n  }\n}\n\nexport class TimeDistributed extends Wrapper {\n  /** @nocollapse */\n  static className = 'TimeDistributed';\n  constructor(args: WrapperLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 3) {\n      throw new ValueError(\n          `TimeDistributed layer expects an input shape >= 3D, but received ` +\n          `input shape ${JSON.stringify(inputShape)}`);\n    }\n    this.inputSpec = [{shape: inputShape}];\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (!this.layer.built) {\n      this.layer.build(childInputShape);\n      this.layer.built = true;\n    }\n    super.build(inputShape);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    const childOutputShape =\n        this.layer.computeOutputShape(childInputShape) as Shape;\n    const timesteps = inputShape[1];\n    return [childOutputShape[0], timesteps].concat(childOutputShape.slice(1));\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      // TODO(cais): Add 'training' and 'useLearningPhase' to kwargs.\n      inputs = getExactlyOneTensor(inputs);\n      // Porting Note: In tfjs-layers, `inputs` are always concrete tensor\n      // values. Hence the inputs can't have an undetermined first (batch)\n      // dimension, which is why we always use the K.rnn approach here.\n      const step: RnnStepFunction = (inputs: Tensor, states: Tensor[]) => {\n        // TODO(cais): Add useLearningPhase.\n        // NOTE(cais): `layer.call` may return a length-1 array of Tensor in\n        //   some cases (e.g., `layer` is a `Sequential` instance), which is\n        //   why `getExactlyOneTensor` is used below.\n        const output = getExactlyOneTensor(this.layer.call(inputs, kwargs));\n        return [output, []];\n      };\n      const rnnOutputs =\n          rnn(step, inputs, [], false /* goBackwards */, null /* mask */,\n              null /* constants */, false /* unroll */,\n              true /* needPerStepOutputs */);\n      const y = rnnOutputs[1];\n      // TODO(cais): Add activity regularization.\n      // TODO(cais): Add useLearningPhase.\n      return y;\n    });\n  }\n\n  // TODO(cais): Implement detailed computeMask() logic.\n}\nserialization.registerClass(TimeDistributed);\n\nexport function checkBidirectionalMergeMode(value?: string): void {\n  generic_utils.checkStringTypeUnionValue(\n      VALID_BIDIRECTIONAL_MERGE_MODES, 'BidirectionalMergeMode', value);\n}\n\nexport declare interface BidirectionalLayerArgs extends WrapperLayerArgs {\n  /**\n   * The instance of an `RNN` layer to be wrapped.\n   */\n  layer: RNN;\n\n  /**\n   * Mode by which outputs of the forward and backward RNNs are\n   * combined. If `null` or `undefined`, the output will not be\n   * combined, they will be returned as an `Array`.\n   *\n   * If `undefined` (i.e., not provided), defaults to `'concat'`.\n   */\n  mergeMode?: BidirectionalMergeMode;\n}\n\nconst DEFAULT_BIDIRECTIONAL_MERGE_MODE: BidirectionalMergeMode = 'concat';\n\nexport class Bidirectional extends Wrapper {\n  /** @nocollapse */\n  static className = 'Bidirectional';\n  mergeMode: BidirectionalMergeMode;\n  private forwardLayer: RNN;\n  private backwardLayer: RNN;\n  private returnSequences: boolean;\n  private returnState: boolean;\n  private numConstants?: number;\n  private _trainable: boolean;\n\n  constructor(args: BidirectionalLayerArgs) {\n    super(args);\n\n    // Note: When creating `this.forwardLayer`, the original Layer object\n    //   (`config.layer`) ought to be cloned. This is why we call\n    //   `getConfig()` followed by `deserialize()`. Without this cloning,\n    //   the layer names saved during serialization will incorrectly contain\n    //   the 'forward_' prefix. In Python Keras, this is done using\n    //   `copy.copy` (shallow copy), which does not have a simple equivalent\n    //   in JavaScript. JavaScript's `Object.assign()` does not copy\n    //   methods.\n    const layerConfig = args.layer.getConfig();\n    const forwDict: serialization.ConfigDict = {};\n    forwDict['className'] = args.layer.getClassName();\n    forwDict['config'] = layerConfig;\n    this.forwardLayer = deserialize(forwDict) as RNN;\n    layerConfig['goBackwards'] =\n        layerConfig['goBackwards'] === true ? false : true;\n    const backDict: serialization.ConfigDict = {};\n    backDict['className'] = args.layer.getClassName();\n    backDict['config'] = layerConfig;\n    this.backwardLayer = deserialize(backDict) as RNN;\n    this.forwardLayer.name = 'forward_' + this.forwardLayer.name;\n    this.backwardLayer.name = 'backward_' + this.backwardLayer.name;\n\n    this.mergeMode = args.mergeMode === undefined ?\n        DEFAULT_BIDIRECTIONAL_MERGE_MODE :\n        args.mergeMode;\n    checkBidirectionalMergeMode(this.mergeMode);\n    if (args.weights) {\n      throw new NotImplementedError(\n          'weights support is not implemented for Bidirectional layer yet.');\n    }\n    this._stateful = args.layer.stateful;\n    this.returnSequences = args.layer.returnSequences;\n    this.returnState = args.layer.returnState;\n    this.supportsMasking = true;\n    this._trainable = true;\n    this.inputSpec = args.layer.inputSpec;\n    this.numConstants = null;\n  }\n\n  get trainable(): boolean {\n    return this._trainable;\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    this._trainable = value;\n    if (this.forwardLayer != null) {\n      this.forwardLayer.trainable = value;\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.trainable = value;\n    }\n  }\n\n  getWeights(): Tensor[] {\n    return this.forwardLayer.getWeights().concat(\n        this.backwardLayer.getWeights());\n  }\n\n  setWeights(weights: Tensor[]): void {\n    const numWeights = weights.length;\n    const numeightsOver2 = Math.floor(numWeights / 2);\n    this.forwardLayer.setWeights(weights.slice(0, numeightsOver2));\n    this.backwardLayer.setWeights(weights.slice(numeightsOver2));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    let layerShapes: Shape|Shape[] =\n        this.forwardLayer.computeOutputShape(inputShape);\n    if (!(Array.isArray(layerShapes) && Array.isArray(layerShapes[0]))) {\n      layerShapes = [layerShapes as Shape];\n    }\n    layerShapes = layerShapes as Shape[];\n\n    let outputShape: Shape;\n    let outputShapes: Shape[];\n    let stateShape: Shape[];\n    if (this.returnState) {\n      stateShape = layerShapes.slice(1);\n      outputShape = layerShapes[0];\n    } else {\n      outputShape = layerShapes[0];\n    }\n    outputShape = outputShape;\n    if (this.mergeMode === 'concat') {\n      outputShape[outputShape.length - 1] *= 2;\n      outputShapes = [outputShape];\n    } else if (this.mergeMode == null) {\n      outputShapes = [outputShape, outputShape.slice()];\n    } else {\n      outputShapes = [outputShape];\n    }\n\n    if (this.returnState) {\n      if (this.mergeMode == null) {\n        return outputShapes.concat(stateShape).concat(stateShape.slice());\n      }\n      return [outputShape].concat(stateShape).concat(stateShape.slice());\n    }\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    if (Array.isArray(inputs)) {\n      initialState = (inputs as Tensor[] | SymbolicTensor[]).slice(1);\n      inputs = (inputs as Tensor[] | SymbolicTensor[])[0];\n    }\n\n    if ((initialState == null || initialState.length === 0) &&\n        constants == null) {\n      return super.apply(inputs, kwargs);\n    }\n    const additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    const additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      const numStates = initialState.length;\n      if (numStates % 2 > 0) {\n        throw new ValueError(\n            'When passing `initialState` to a Bidrectional RNN, ' +\n            'the state should be an Array containing the states of ' +\n            'the underlying RNNs.');\n      }\n      kwargs['initialState'] = initialState;\n      additionalInputs.push(...initialState);\n      const stateSpecs = (initialState as Array<Tensor|SymbolicTensor>)\n                             .map(state => new InputSpec({shape: state.shape}));\n      this.forwardLayer.stateSpec = stateSpecs.slice(0, numStates / 2);\n      this.backwardLayer.stateSpec = stateSpecs.slice(numStates / 2);\n      additionalSpecs.push(...stateSpecs);\n    }\n    if (constants != null) {\n      throw new NotImplementedError(\n          'Support for constants in Bidirectional layers is not ' +\n          'implemented yet.');\n    }\n\n    const isSymbolicTensor = additionalInputs[0] instanceof SymbolicTensor;\n    for (const tensor of additionalInputs) {\n      if (tensor instanceof SymbolicTensor !== isSymbolicTensor) {\n        throw new ValueError(\n            'The initial state of a Bidirectional layer cannot be ' +\n            'specified as a mix of symbolic and non-symbolic tensors');\n      }\n    }\n\n    if (isSymbolicTensor) {\n      // Compute the full input and specs, including the states.\n      const fullInput = [inputs].concat(additionalInputs);\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call temporarily and replace inputSpec.\n      // Note: with initial states symbolic calls and non-symbolic calls to\n      // this method differ in how the initial states are passed. For\n      // symbolic calls, the initial states are passed in the first arg, as\n      // an Array of SymbolicTensors; for non-symbolic calls, they are\n      // passed in the second arg as a part of the kwargs. Hence the need to\n      // temporarily modify inputSpec here.\n      // TODO(cais): Make refactoring so that this hacky code below is no\n      // longer needed.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output =\n          super.apply(fullInput as Tensor[] | SymbolicTensor[], kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const initialState = kwargs['initialState'];\n\n      let y: Tensor|Tensor[];\n      let yRev: Tensor|Tensor[];\n      if (initialState == null) {\n        y = this.forwardLayer.call(inputs, kwargs);\n        yRev = this.backwardLayer.call(inputs, kwargs);\n      } else {\n        const forwardState = initialState.slice(0, initialState.length / 2);\n        const backwardState = initialState.slice(initialState.length / 2);\n        y = this.forwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: forwardState}));\n        yRev = this.backwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: backwardState}));\n      }\n\n      let states: Tensor[];\n      if (this.returnState) {\n        if (Array.isArray(y)) {\n          states = y.slice(1).concat((yRev as Tensor[]).slice(1));\n        } else {\n        }\n        y = (y as Tensor[])[0];\n        yRev = (yRev as Tensor[])[0];\n      }\n\n      if (this.returnSequences) {\n        yRev = tfc.reverse(yRev as Tensor, 1);\n      }\n\n      let output: Tensor|Tensor[];\n      if (this.mergeMode === 'concat') {\n        output = K.concatenate([y as Tensor, yRev as Tensor]);\n      } else if (this.mergeMode === 'sum') {\n        output = tfc.add(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode === 'ave') {\n        output = tfc.mul(.5, tfc.add(y as Tensor, yRev as Tensor));\n      } else if (this.mergeMode === 'mul') {\n        output = tfc.mul(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode == null) {\n        output = [y as Tensor, yRev as Tensor];\n      }\n\n      // TODO(cais): Properly set learning phase.\n      if (this.returnState) {\n        if (this.mergeMode == null) {\n          return (output as Tensor[]).concat(states);\n        }\n        return [output as Tensor].concat(states);\n      }\n      return output;\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[]): void {\n    this.forwardLayer.resetStates();\n    this.backwardLayer.resetStates();\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    nameScope(this.forwardLayer.name, () => {\n      this.forwardLayer.build(inputShape);\n    });\n    nameScope(this.backwardLayer.name, () => {\n      this.backwardLayer.build(inputShape);\n    });\n    this.built = true;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (Array.isArray(mask)) {\n      mask = mask[0];\n    }\n    let outputMask: Tensor|Tensor[];\n    if (this.returnSequences) {\n      if (this.mergeMode == null) {\n        outputMask = [mask, mask];\n      } else {\n        outputMask = mask;\n      }\n    } else {\n      if (this.mergeMode == null) {\n        outputMask = [null, null];\n      } else {\n        outputMask = null;\n      }\n    }\n    if (this.returnState) {\n      const states = this.forwardLayer.states;\n      const stateMask: Tensor[] = states.map(state => null);\n      if (Array.isArray(outputMask)) {\n        return outputMask.concat(stateMask).concat(stateMask);\n      } else {\n        return [outputMask].concat(stateMask).concat(stateMask);\n      }\n    } else {\n      return outputMask;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.forwardLayer.trainableWeights.concat(\n        this.backwardLayer.trainableWeights);\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.forwardLayer.nonTrainableWeights.concat(\n        this.backwardLayer.nonTrainableWeights);\n  }\n\n  // TODO(cais): Implement constraints().\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.forwardLayer != null) {\n      this.forwardLayer.setFastWeightInitDuringBuild(value);\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'mergeMode': this.mergeMode,\n    };\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    const rnnLayer =\n        deserialize(config['layer'] as serialization.ConfigDict) as RNN;\n    delete config['layer'];\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    if (config['numConstants'] != null) {\n      throw new NotImplementedError(\n          `Deserialization of a Bidirectional layer with numConstants ` +\n          `present is not supported yet.`);\n    }\n    // tslint:disable-next-line:no-any\n    const newConfig: {[key: string]: any} = config;\n    newConfig['layer'] = rnnLayer;\n    return new cls(newConfig);\n  }\n}\nserialization.registerClass(Bidirectional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {InputLayer, InputLayerArgs} from './engine/input_layer';\nimport {Layer, LayerArgs} from './engine/topology';\nimport {input} from './exports';\nimport {ELU, ELULayerArgs, LeakyReLU, LeakyReLULayerArgs, PReLU, PReLULayerArgs, ReLU, ReLULayerArgs, Softmax, SoftmaxLayerArgs, ThresholdedReLU, ThresholdedReLULayerArgs} from './layers/advanced_activations';\nimport {Conv1D, Conv2D, Conv2DTranspose, Conv3D, ConvLayerArgs, Cropping2D, Cropping2DLayerArgs, SeparableConv2D, SeparableConvLayerArgs, UpSampling2D, UpSampling2DLayerArgs} from './layers/convolutional';\nimport {DepthwiseConv2D, DepthwiseConv2DLayerArgs} from './layers/convolutional_depthwise';\nimport {ConvLSTM2D, ConvLSTM2DArgs, ConvLSTM2DCell, ConvLSTM2DCellArgs} from './layers/convolutional_recurrent';\nimport {Activation, ActivationLayerArgs, Dense, DenseLayerArgs, Dropout, DropoutLayerArgs, Flatten, FlattenLayerArgs, Masking, MaskingArgs, Permute, PermuteLayerArgs, RepeatVector, RepeatVectorLayerArgs, Reshape, ReshapeLayerArgs, SpatialDropout1D, SpatialDropout1DLayerConfig} from './layers/core';\nimport {Embedding, EmbeddingLayerArgs} from './layers/embeddings';\nimport {Add, Average, Concatenate, ConcatenateLayerArgs, Dot, DotLayerArgs, Maximum, Minimum, Multiply} from './layers/merge';\nimport {AlphaDropout, AlphaDropoutArgs, GaussianDropout, GaussianDropoutArgs, GaussianNoise, GaussianNoiseArgs} from './layers/noise';\nimport {BatchNormalization, BatchNormalizationLayerArgs, LayerNormalization, LayerNormalizationLayerArgs} from './layers/normalization';\nimport {ZeroPadding2D, ZeroPadding2DLayerArgs} from './layers/padding';\nimport {AveragePooling1D, AveragePooling2D, AveragePooling3D, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalPooling2DLayerArgs, MaxPooling1D, MaxPooling2D, MaxPooling3D, Pooling1DLayerArgs, Pooling2DLayerArgs, Pooling3DLayerArgs} from './layers/pooling';\nimport {GRU, GRUCell, GRUCellLayerArgs, GRULayerArgs, LSTM, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNN, SimpleRNNCell, SimpleRNNCellLayerArgs, SimpleRNNLayerArgs, StackedRNNCells, StackedRNNCellsArgs} from './layers/recurrent';\nimport {Bidirectional, BidirectionalLayerArgs, TimeDistributed, WrapperLayerArgs} from './layers/wrappers';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// Input Layer.\n/**\n * An input layer is an entry point into a `tf.LayersModel`.\n *\n * `InputLayer` is generated automatically for `tf.Sequential`` models by\n * specifying the `inputshape` or `batchInputShape` for the first layer.  It\n * should not be specified explicitly. However, it can be useful sometimes,\n * e.g., when constructing a sequential model from a subset of another\n * sequential model's layers. Like the code snippet below shows.\n *\n * ```js\n * // Define a model which simply adds two inputs.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));\n * model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n * model1.summary();\n * model1.predict(tf.zeros([1, 4])).print();\n *\n * // Construct another model, reusing the second layer of `model1` while\n * // not using the first layer of `model1`. Note that you cannot add the second\n * // layer of `model` directly as the first layer of the new sequential model,\n * // because doing so will lead to an error related to the fact that the layer\n * // is not an input layer. Instead, you need to create an `inputLayer` and add\n * // it to the new sequential model before adding the reused layer.\n * const model2 = tf.sequential();\n * // Use an inputShape that matches the input shape of `model1`'s second\n * // layer.\n * model2.add(tf.layers.inputLayer({inputShape: [3]}));\n * model2.add(model1.layers[1]);\n * model2.summary();\n * model2.predict(tf.zeros([1, 3])).print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}\n */\nexport function inputLayer(args: InputLayerArgs): Layer {\n  return new InputLayer(args);\n}\n\n// Advanced Activation Layers.\n\n/**\n * Exponetial Linear Unit (ELU).\n *\n * It follows:\n * `f(x) =  alpha * (exp(x) - 1.) for x < 0`,\n * `f(x) = x for x >= 0`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Fast and Accurate Deep Network Learning by Exponential Linear Units\n * (ELUs)](https://arxiv.org/abs/1511.07289v1)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function elu(args?: ELULayerArgs): Layer {\n  return new ELU(args);\n}\n\n/**\n * Rectified Linear Unit activation function.\n *\n * Input shape:\n *   Arbitrary. Use the config field `inputShape` (Array of integers, does\n *   not include the sample axis) when using this layer as the first layer\n *   in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function reLU(args?: ReLULayerArgs): Layer {\n  return new ReLU(args);\n}\n\n/**\n * Leaky version of a rectified linear unit.\n *\n * It allows a small gradient when the unit is not active:\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function leakyReLU(args?: LeakyReLULayerArgs): Layer {\n  return new LeakyReLU(args);\n}\n\n/**\n * Parameterized version of a leaky rectified linear unit.\n *\n * It follows\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n * wherein `alpha` is a trainable weight.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function prelu(args?: PReLULayerArgs): Layer {\n  return new PReLU(args);\n}\n\n/**\n * Softmax activation layer.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function softmax(args?: SoftmaxLayerArgs): Layer {\n  return new Softmax(args);\n}\n\n/**\n * Thresholded Rectified Linear Unit.\n *\n * It follows:\n * `f(x) = x for x > theta`,\n * `f(x) = 0 otherwise`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Zero-Bias Autoencoders and the Benefits of Co-Adapting\n * Features](http://arxiv.org/abs/1402.3337)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function thresholdedReLU(args?: ThresholdedReLULayerArgs): Layer {\n  return new ThresholdedReLU(args);\n}\n\n// Convolutional Layers.\n\n/**\n * 1D convolution layer (e.g., temporal convolution).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input over a single spatial (or temporal) dimension\n * to produce a tensor of outputs.\n *\n * If `use_bias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model, provide an\n * `inputShape` argument `Array` or `null`.\n *\n * For example, `inputShape` would be:\n * - `[10, 128]` for sequences of 10 vectors of 128-dimensional vectors\n * - `[null, 128]` for variable-length sequences of 128-dimensional vectors.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional',  namespace: 'layers'}\n */\nexport function conv1d(args: ConvLayerArgs): Layer {\n  return new Conv1D(args);\n}\n\n/**\n * 2D convolution layer (e.g. spatial convolution over images).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 3]` for 128x128 RGB pictures\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2d(args: ConvLayerArgs): Layer {\n  return new Conv2D(args);\n}\n\n/**\n * Transposed convolutional layer (sometimes called Deconvolution).\n *\n * The need for transposed convolutions generally arises\n * from the desire to use a transformation going in the opposite direction of\n * a normal convolution, i.e., from something that has the shape of the output\n * of some convolution to something that has the shape of its input while\n * maintaining a connectivity pattern that is compatible with said\n * convolution.\n *\n * When using this layer as the first layer in a model, provide the\n * configuration `inputShape` (`Array` of integers, does not include the\n * sample axis), e.g., `inputShape: [128, 128, 3]` for 128x128 RGB pictures in\n * `dataFormat: 'channelsLast'`.\n *\n * Input shape:\n *   4D tensor with shape:\n *   `[batch, channels, rows, cols]` if `dataFormat` is `'channelsFirst'`.\n *   or 4D tensor with shape\n *   `[batch, rows, cols, channels]` if `dataFormat` is `'channelsLast`.\n *\n * Output shape:\n *   4D tensor with shape:\n *   `[batch, filters, newRows, newCols]` if `dataFormat` is\n * `'channelsFirst'`. or 4D tensor with shape:\n *   `[batch, newRows, newCols, filters]` if `dataFormat` is `'channelsLast'`.\n *\n * References:\n *   - [A guide to convolution arithmetic for deep\n * learning](https://arxiv.org/abs/1603.07285v1)\n *   - [Deconvolutional\n * Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2dTranspose(args: ConvLayerArgs): Layer {\n  return new Conv2DTranspose(args);\n}\n\n/**\n * 3D convolution layer (e.g. spatial convolution over volumes).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 128, 1]` for 128x128x128 grayscale volumes\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv3d(args: ConvLayerArgs): Layer {\n  return new Conv3D(args);\n}\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Separable convolution consists of first performing\n * a depthwise spatial convolution\n * (which acts on each input channel separately)\n * followed by a pointwise convolution which mixes together the resulting\n * output channels. The `depthMultiplier` argument controls how many\n * output channels are generated per input channel in the depthwise step.\n *\n * Intuitively, separable convolutions can be understood as\n * a way to factorize a convolution kernel into two smaller kernels,\n * or as an extreme version of an Inception block.\n *\n * Input shape:\n *   4D tensor with shape:\n *     `[batch, channels, rows, cols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, rows, cols, channels]` if data_format='channelsLast'.\n *\n * Output shape:\n *   4D tensor with shape:\n *     `[batch, filters, newRows, newCols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, newRows, newCols, filters]` if data_format='channelsLast'.\n *     `rows` and `cols` values might have changed due to padding.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function separableConv2d(args: SeparableConvLayerArgs): Layer {\n  return new SeparableConv2D(args);\n}\n\n/**\n * Cropping layer for 2D input (e.g., image).\n *\n * This layer can crop an input\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, croppedRows, croppedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, croppedRows, croppedCols]`.\n *\n * Examples\n * ```js\n *\n * const model = tf.sequential();\n * model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]],\n *                                inputShape: [128, 128, 3]}));\n * //now output shape is [batch, 124, 124, 3]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function cropping2D(args: Cropping2DLayerArgs): Layer {\n  return new Cropping2D(args);\n}\n\n/**\n * Upsampling layer for 2D inputs.\n *\n * Repeats the rows and columns of the data\n * by size[0] and size[1] respectively.\n *\n *\n * Input shape:\n *    4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *         `[batch, rows, cols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *        `[batch, channels, rows, cols]`\n *\n * Output shape:\n *     4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *        `[batch, upsampledRows, upsampledCols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *         `[batch, channels, upsampledRows, upsampledCols]`\n *\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function upSampling2d(args: UpSampling2DLayerArgs): Layer {\n  return new UpSampling2D(args);\n}\n\n// Convolutional(depthwise) Layers.\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Depthwise Separable convolutions consists in performing just the first step\n * in a depthwise spatial convolution (which acts on each input channel\n * separately). The `depthMultplier` argument controls how many output channels\n * are generated per input channel in the depthwise step.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function depthwiseConv2d(args: DepthwiseConv2DLayerArgs): Layer {\n  return new DepthwiseConv2D(args);\n}\n\n// Basic Layers.\n\n/**\n * Applies an activation function to an output.\n *\n * This layer applies element-wise activation function.  Other layers, notably\n * `dense` can also apply activation functions.  Use this isolated activation\n * function to extract the values before and after the\n * activation. For instance:\n *\n * ```js\n * const input = tf.input({shape: [5]});\n * const denseLayer = tf.layers.dense({units: 1});\n * const activationLayer = tf.layers.activation({activation: 'relu6'});\n *\n * // Obtain the output symbolic tensors by applying the layers in order.\n * const denseOutput = denseLayer.apply(input);\n * const activationOutput = activationLayer.apply(denseOutput);\n *\n * // Create the model based on the inputs.\n * const model = tf.model({\n *     inputs: input,\n *     outputs: [denseOutput, activationOutput]\n * });\n *\n * // Collect both outputs and print separately.\n * const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));\n * denseOut.print();\n * activationOut.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function activation(args: ActivationLayerArgs): Layer {\n  return new Activation(args);\n}\n\n/**\n * Creates a dense (fully connected) layer.\n *\n * This layer implements the operation:\n *   `output = activation(dot(input, kernel) + bias)`\n *\n * `activation` is the element-wise activation function\n *   passed as the `activation` argument.\n *\n * `kernel` is a weights matrix created by the layer.\n *\n * `bias` is a bias vector created by the layer (only applicable if `useBias`\n * is `true`).\n *\n * **Input shape:**\n *\n *   nD `tf.Tensor` with shape: `(batchSize, ..., inputDim)`.\n *\n *   The most common situation would be\n *   a 2D input with shape `(batchSize, inputDim)`.\n *\n * **Output shape:**\n *\n *   nD tensor with shape: `(batchSize, ..., units)`.\n *\n *   For instance, for a 2D input with shape `(batchSize, inputDim)`,\n *   the output would have shape `(batchSize, units)`.\n *\n * Note: if the input to the layer has a rank greater than 2, then it is\n * flattened prior to the initial dot product with the kernel.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dense(args: DenseLayerArgs): Layer {\n  return new Dense(args);\n}\n\n/**\n * Applies\n * [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to\n * the input.\n *\n * Dropout consists in randomly setting a fraction `rate` of input units to 0 at\n * each update during training time, which helps prevent overfitting.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dropout(args: DropoutLayerArgs): Layer {\n  return new Dropout(args);\n}\n\n/**\n * Spatial 1D version of Dropout.\n *\n * This Layer type performs the same function as the Dropout layer, but it drops\n * entire 1D feature maps instead of individual elements. For example, if an\n * input example consists of 3 timesteps and the feature map for each timestep\n * has a size of 4, a `spatialDropout1d` layer may zero out the feature maps\n * of the 1st timesteps and 2nd timesteps completely while sparing all feature\n * elements of the 3rd timestep.\n *\n * If adjacent frames (timesteps) are strongly correlated (as is normally the\n * case in early convolution layers), regular dropout will not regularize the\n * activation and will otherwise just result in merely an effective learning\n * rate decrease. In this case, `spatialDropout1d` will help promote\n * independence among feature maps and should be used instead.\n *\n * **Arguments:**\n *   rate: A floating-point number >=0 and <=1. Fraction of the input elements\n *     to drop.\n *\n * **Input shape:**\n *   3D tensor with shape `(samples, timesteps, channels)`.\n *\n * **Output shape:**\n *   Same as the input shape.\n *\n * References:\n *   - [Efficient Object Localization Using Convolutional\n *      Networks](https://arxiv.org/abs/1411.4280)\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function spatialDropout1d(args: SpatialDropout1DLayerConfig): Layer {\n  return new SpatialDropout1D(args);\n}\n\n/**\n * Flattens the input. Does not affect the batch size.\n *\n * A `Flatten` layer flattens each batch in its inputs to 1D (making the output\n * 2D).\n *\n * For example:\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const flattenLayer = tf.layers.flatten();\n * // Inspect the inferred output shape of the flatten layer, which\n * // equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the\n * // flattening. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(flattenLayer.apply(input).shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function flatten(args?: FlattenLayerArgs): Layer {\n  return new Flatten(args);\n}\n\n/**\n * Repeats the input n times in a new dimension.\n *\n * ```js\n *  const model = tf.sequential();\n *  model.add(tf.layers.repeatVector({n: 4, inputShape: [2]}));\n *  const x = tf.tensor2d([[10, 20]]);\n *  // Use the model to do inference on a data point the model hasn't see\n *  model.predict(x).print();\n *  // output shape is now [batch, 2, 4]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function repeatVector(args: RepeatVectorLayerArgs): Layer {\n  return new RepeatVector(args);\n}\n\n/**\n * Reshapes an input to a certain shape.\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const reshapeLayer = tf.layers.reshape({targetShape: [2, 6]});\n * // Inspect the inferred output shape of the Reshape layer, which\n * // equals `[null, 2, 6]`. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(reshapeLayer.apply(input).shape));\n * ```\n *\n * Input shape:\n *   Arbitrary, although all dimensions in the input shape must be fixed.\n *   Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n *\n * Output shape:\n *   [batchSize, targetShape[0], targetShape[1], ...,\n *    targetShape[targetShape.length - 1]].\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function reshape(args: ReshapeLayerArgs): Layer {\n  return new Reshape(args);\n}\n\n/**\n * Permutes the dimensions of the input according to a given pattern.\n *\n * Useful for, e.g., connecting RNNs and convnets together.\n *\n * Example:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.permute({\n *   dims: [2, 1],\n *   inputShape: [10, 64]\n * }));\n * console.log(model.outputShape);\n * // Now model's output shape is [null, 64, 10], where null is the\n * // unpermuted sample (batch) dimension.\n * ```\n *\n * Input shape:\n *   Arbitrary. Use the configuration field `inputShape` when using this\n *   layer as the first layer in a model.\n *\n * Output shape:\n *   Same rank as the input shape, but with the dimensions re-ordered (i.e.,\n *   permuted) according to the `dims` configuration of this layer.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function permute(args: PermuteLayerArgs): Layer {\n  return new Permute(args);\n}\n\n/**\n * Maps positive integers (indices) into dense vectors of fixed size.\n * eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]\n *\n * **Input shape:** 2D tensor with shape: `[batchSize, sequenceLength]`.\n *\n * **Output shape:** 3D tensor with shape: `[batchSize, sequenceLength,\n * outputDim]`.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function embedding(args: EmbeddingLayerArgs): Layer {\n  return new Embedding(args);\n}\n\n// Merge Layers.\n\n/**\n * Layer that performs element-wise addition on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). The inputs are specified as an\n * `Array` when the `apply` method of the `Add` layer instance is called. For\n * example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const addLayer = tf.layers.add();\n * const sum = addLayer.apply([input1, input2]);\n * console.log(JSON.stringify(sum.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function add(args?: LayerArgs): Layer {\n  return new Add(args);\n}\n\n/**\n * Layer that performs element-wise averaging on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const averageLayer = tf.layers.average();\n * const average = averageLayer.apply([input1, input2]);\n * console.log(JSON.stringify(average.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function average(args?: LayerArgs): Layer {\n  return new Average(args);\n}\n\n/**\n * Layer that concatenates an `Array` of inputs.\n *\n * It takes a list of tensors, all of the same shape except for the\n * concatenation axis, and returns a single tensor, the concatenation\n * of all inputs. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 3]});\n * const concatLayer = tf.layers.concatenate();\n * const output = concatLayer.apply([input1, input2]);\n * console.log(JSON.stringify(output.shape));\n * // You get [null, 2, 5], with the first dimension as the undetermined batch\n * // dimension. The last dimension (5) is the result of concatenating the\n * // last dimensions of the inputs (2 and 3).\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function concatenate(args?: ConcatenateLayerArgs): Layer {\n  return new Concatenate(args);\n}\n\n/**\n * Layer that computes the element-wise maximum an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const maxLayer = tf.layers.maximum();\n * const max = maxLayer.apply([input1, input2]);\n * console.log(JSON.stringify(max.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function maximum(args?: LayerArgs): Layer {\n  return new Maximum(args);\n}\n\n/**\n * Layer that computes the element-wise minimum of an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const minLayer = tf.layers.minimum();\n * const min = minLayer.apply([input1, input2]);\n * console.log(JSON.stringify(min.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function minimum(args?: LayerArgs): Layer {\n  return new Minimum(args);\n}\n\n/**\n * Layer that multiplies (element-wise) an `Array` of inputs.\n *\n * It takes as input an Array of tensors, all of the same\n * shape, and returns a single tensor (also of the same shape).\n * For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const input3 = tf.input({shape: [2, 2]});\n * const multiplyLayer = tf.layers.multiply();\n * const product = multiplyLayer.apply([input1, input2, input3]);\n * console.log(product.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function multiply(args?: LayerArgs): Layer {\n  return new Multiply(args);\n}\n\n/**\n * Layer that computes a dot product between samples in two tensors.\n *\n * E.g., if applied to a list of two tensors `a` and `b` both of shape\n * `[batchSize, n]`, the output will be a tensor of shape `[batchSize, 1]`,\n * where each entry at index `[i, 0]` will be the dot product between\n * `a[i, :]` and `b[i, :]`.\n *\n * Example:\n *\n * ```js\n * const dotLayer = tf.layers.dot({axes: -1});\n * const x1 = tf.tensor2d([[10, 20], [30, 40]]);\n * const x2 = tf.tensor2d([[-1, -2], [-3, -4]]);\n *\n * // Invoke the layer's apply() method in eager (imperative) mode.\n * const y = dotLayer.apply([x1, x2]);\n * y.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function dot(args: DotLayerArgs): Layer {\n  return new Dot(args);\n}\n\n// Normalization Layers.\n\n/**\n * Batch normalization layer (Ioffe and Szegedy, 2014).\n *\n * Normalize the activations of the previous layer at each batch,\n * i.e. applies a transformation that maintains the mean activation\n * close to 0 and the activation standard deviation close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape` (Array of integers, does\n *   not include the sample axis) when calling the constructor of this class,\n *   if this layer is used as a first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Batch Normalization: Accelerating Deep Network Training by Reducing\n * Internal Covariate Shift](https://arxiv.org/abs/1502.03167)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function batchNormalization(args?: BatchNormalizationLayerArgs): Layer {\n  return new BatchNormalization(args);\n}\n\n/**\n * Layer-normalization layer (Ba et al., 2016).\n *\n * Normalizes the activations of the previous layer for each given example in a\n * batch independently, instead of across a batch like in `batchNormalization`.\n * In other words, this layer applies a transformation that maintanis the mean\n * activation within each example close to0 and activation variance close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the argument `inputShape` when using this layer as the first\n *   layer in a model.\n *\n * Output shape:\n *   Same as input.\n *\n * References:\n *   - [Layer Normalization](https://arxiv.org/abs/1607.06450)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function layerNormalization(args?: LayerNormalizationLayerArgs): Layer {\n  return new LayerNormalization(args);\n}\n\n// Padding Layers.\n\n/**\n * Zero-padding layer for 2D input (e.g., image).\n *\n * This layer can add rows and columns of zeros\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, paddedRows, paddedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, paddedRows, paddedCols]`.\n *\n * @doc {heading: 'Layers', subheading: 'Padding', namespace: 'layers'}\n */\nexport function zeroPadding2d(args?: ZeroPadding2DLayerArgs): Layer {\n  return new ZeroPadding2D(args);\n}\n\n// Pooling Layers.\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape: `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * `tf.avgPool1d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling1d(args: Pooling1DLayerArgs): Layer {\n  return new AveragePooling1D(args);\n}\nexport function avgPool1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape:\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, rows, cols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, pooleRows, pooledCols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, pooleRows, pooledCols]`\n *\n * `tf.avgPool2d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling2d(args: Pooling2DLayerArgs): Layer {\n  return new AveragePooling2D(args);\n}\nexport function avgPool2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n\n/**\n * Average pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      4D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling3d(args: Pooling3DLayerArgs): Layer {\n  return new AveragePooling3D(args);\n}\nexport function avgPool3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n\n/**\n * Global average pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling1d(args?: LayerArgs): Layer {\n  return new GlobalAveragePooling1D(args);\n}\n\n/**\n * Global average pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalAveragePooling2D(args);\n}\n\n/**\n * Global max pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling1d(args?: LayerArgs): Layer {\n  return new GlobalMaxPooling1D(args);\n}\n\n/**\n * Global max pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalMaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for temporal data.\n *\n * Input shape:  `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling1d(args: Pooling1DLayerArgs): Layer {\n  return new MaxPooling1D(args);\n}\n\n/**\n * Max pooling operation for spatial data.\n *\n * Input shape\n *   - If `dataFormat === CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, rows, cols, channels]`\n *   - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *       `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, pooleRows, pooledCols, channels]`\n *   - If `dataFormat=CHANNEL_FIRST`:\n *       4D tensor with shape:\n *       `[batchSize, channels, pooleRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling2d(args: Pooling2DLayerArgs): Layer {\n  return new MaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      5D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling3d(args: Pooling3DLayerArgs): Layer {\n  return new MaxPooling3D(args);\n}\n\n// Recurrent Layers.\n\n/**\n * Gated Recurrent Unit - Cho et al. 2014.\n *\n * This is an `RNN` layer consisting of one `GRUCell`. However, unlike\n * the underlying `GRUCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.gru({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `GRUCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gru(args: GRULayerArgs): Layer {\n  return new GRU(args);\n}\n\n/**\n * Cell class for `GRU`.\n *\n * `GRUCell` is distinct from the `RNN` subclass `GRU` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `GRU` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.gruCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `GRUCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.gruCell({units: 4}),\n *   tf.layers.gruCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `gruCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `GRUCell`, use the\n * `tf.layers.gru`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gruCell(args: GRUCellLayerArgs): RNNCell {\n  return new GRUCell(args);\n}\n\n/**\n * Long-Short Term Memory layer - Hochreiter 1997.\n *\n * This is an `RNN` layer consisting of one `LSTMCell`. However, unlike\n * the underlying `LSTMCell`, the `apply` method of `LSTM` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const lstm = tf.layers.lstm({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = lstm.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `LSTMCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstm(args: LSTMLayerArgs): Layer {\n  return new LSTM(args);\n}\n\n/**\n * Cell class for `LSTM`.\n *\n * `LSTMCell` is distinct from the `RNN` subclass `LSTM` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `LSTM` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.lstmCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `LSTMCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.lstmCell({units: 4}),\n *   tf.layers.lstmCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `lstmCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `LSTMCell`, use the\n * `tf.layers.lstm`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstmCell(args: LSTMCellLayerArgs): RNNCell {\n  return new LSTMCell(args);\n}\n\n/**\n * Fully-connected RNN where the output is to be fed back to input.\n *\n * This is an `RNN` layer consisting of one `SimpleRNNCell`. However, unlike\n * the underlying `SimpleRNNCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.simpleRNN({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `SimpleRNNCell`'s number of units.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNN(args: SimpleRNNLayerArgs): Layer {\n  return new SimpleRNN(args);\n}\n\n/**\n * Cell class for `SimpleRNN`.\n *\n * `SimpleRNNCell` is distinct from the `RNN` subclass `SimpleRNN` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `SimpleRNN` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.simpleRNNCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `SimpleRNNCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.simpleRNNCell({units: 4}),\n *   tf.layers.simpleRNNCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `SimpleRNNCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `SimpleRNNCell`, use the\n * `tf.layers.simpleRNN`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNNCell(args: SimpleRNNCellLayerArgs): RNNCell {\n  return new SimpleRNNCell(args);\n}\n\n/**\n * Convolutional LSTM layer - Xingjian Shi 2015.\n *\n * This is an `ConvRNN2D` layer consisting of one `ConvLSTM2DCell`. However,\n * unlike the underlying `ConvLSTM2DCell`, the `apply` method of `ConvLSTM2D`\n * operates on a sequence of inputs. The shape of the input (not including the\n * first, batch dimension) needs to be 4-D, with the first dimension being time\n * steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const batchSize = 4;\n * const sequenceLength = 2;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [batchSize, sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const layer = tf.layers.convLstm2d({filters, kernelSize});\n *\n * const output = layer.apply(input);\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2d(args: ConvLSTM2DArgs): ConvLSTM2D {\n  return new ConvLSTM2D(args);\n}\n\n/**\n * Cell class for `ConvLSTM2D`.\n *\n * `ConvLSTM2DCell` is distinct from the `ConvRNN2D` subclass `ConvLSTM2D` in\n * that its `call` method takes the input data of only a single time step and\n * returns the cell's output at the time step, while `ConvLSTM2D` takes the\n * input data over a number of time steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const sequenceLength = 1;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const cell = tf.layers.convLstm2dCell({filters, kernelSize});\n *\n * cell.build(input.shape);\n *\n * const outputSize = size - kernelSize + 1;\n * const outShape = [sequenceLength, outputSize, outputSize, filters];\n *\n * const initialH = tf.zeros(outShape);\n * const initialC = tf.zeros(outShape);\n *\n * const [o, h, c] = cell.call([input, initialH, initialC], {});\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2dCell(args: ConvLSTM2DCellArgs): ConvLSTM2DCell {\n  return new ConvLSTM2DCell(args);\n}\n\n/**\n * Base class for recurrent layers.\n *\n * Input shape:\n *   3D tensor with shape `[batchSize, timeSteps, inputDim]`.\n *\n * Output shape:\n *   - if `returnState`, an Array of tensors (i.e., `tf.Tensor`s). The first\n *     tensor is the output. The remaining tensors are the states at the\n *     last time step, each with shape `[batchSize, units]`.\n *   - if `returnSequences`, the output will have shape\n *     `[batchSize, timeSteps, units]`.\n *   - else, the output will have shape `[batchSize, units]`.\n *\n * Masking:\n *   This layer supports masking for input data with a variable number\n *   of timesteps. To introduce masks to your data,\n *   use an embedding layer with the `mask_zero` parameter\n *   set to `True`.\n *\n * Notes on using statefulness in RNNs:\n *   You can set RNN layers to be 'stateful', which means that the states\n *   computed for the samples in one batch will be reused as initial states\n *   for the samples in the next batch. This assumes a one-to-one mapping\n *   between samples in different successive batches.\n *\n *   To enable statefulness:\n *     - specify `stateful: true` in the layer constructor.\n *     - specify a fixed batch size for your model, by passing\n *       if sequential model:\n *         `batchInputShape=[...]` to the first layer in your model.\n *       else for functional model with 1 or more Input layers:\n *         `batchShape=[...]` to all the first layers in your model.\n *       This is the expected shape of your inputs *including the batch size*.\n *       It should be a tuple of integers, e.g. `(32, 10, 100)`.\n *     - specify `shuffle=False` when calling fit().\n *\n *   To reset the states of your model, call `.resetStates()` on either\n *   a specific layer, or on your entire model.\n *\n * Note on specifying the initial state of RNNs\n *   You can specify the initial state of RNN layers symbolically by\n *   calling them with the option `initialState`. The value of\n *   `initialState` should be a tensor or list of tensors representing\n *   the initial state of the RNN layer.\n *\n *   You can specify the initial state of RNN layers numerically by\n *   calling `resetStates` with the keyword argument `states`. The value of\n *   `states` should be a numpy array or list of numpy arrays representing\n *   the initial state of the RNN layer.\n *\n * Note on passing external constants to RNNs\n *   You can pass \"external\" constants to the cell using the `constants`\n *   keyword argument of `RNN.call` method. This requires that the `cell.call`\n *   method accepts the same keyword argument `constants`. Such constants\n *   can be used to conditon the cell transformation on additional static inputs\n *   (not changing over time), a.k.a an attention mechanism.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function rnn(args: RNNLayerArgs): Layer {\n  return new RNN(args);\n}\n\n/**\n * Wrapper allowing a stack of RNN cells to behave as a single cell.\n *\n * Used to implement efficient stacked RNNs.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function stackedRNNCells(args: StackedRNNCellsArgs): RNNCell {\n  return new StackedRNNCells(args);\n}\n\n// Wrapper Layers.\n\n/** @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'} */\nexport function bidirectional(args: BidirectionalLayerArgs): Bidirectional {\n  return new Bidirectional(args);\n}\n\n/**\n * This wrapper applies a layer to every temporal slice of an input.\n *\n * The input should be at least 3D,  and the dimension of the index `1` will be\n * considered to be the temporal dimension.\n *\n * Consider a batch of 32 samples, where each sample is a sequence of 10 vectors\n * of 16 dimensions. The batch input shape of the layer is then `[32,  10,\n * 16]`, and the `inputShape`, not including the sample dimension, is\n * `[10, 16]`.\n *\n * You can then use `TimeDistributed` to apply a `Dense` layer to each of the 10\n * timesteps, independently:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.dense({units: 8}),\n *   inputShape: [10, 16],\n * }));\n *\n * // Now model.outputShape = [null, 10, 8].\n * // The output will then have shape `[32, 10, 8]`.\n *\n * // In subsequent layers, there is no need for `inputShape`:\n * model.add(tf.layers.timeDistributed({layer: tf.layers.dense({units: 32})}));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * // Now model.outputShape = [null, 10, 32].\n * ```\n *\n * The output will then have shape `[32, 10, 32]`.\n *\n * `TimeDistributed` can be used with arbitrary layers, not just `Dense`, for\n * instance a `Conv2D` layer.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.conv2d({filters: 64, kernelSize: [3, 3]}),\n *   inputShape: [10, 299, 299, 3],\n * }));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}\n */\nexport function timeDistributed(args: WrapperLayerArgs): Layer {\n  return new TimeDistributed(args);\n}\n\n// Aliases for pooling.\nexport const globalMaxPool1d = globalMaxPooling1d;\nexport const globalMaxPool2d = globalMaxPooling2d;\nexport const maxPool1d = maxPooling1d;\nexport const maxPool2d = maxPooling2d;\n\nexport {Layer, RNN, RNNCell, input /* alias for tf.input */};\n\n/**\n * Apply additive zero-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * This is useful to mitigate overfitting\n * (you could see it as a form of random data augmentation).\n * Gaussian Noise (GS) is a natural choice as corruption process\n * for real valued inputs.\n *\n * # Arguments\n *     stddev: float, standard deviation of the noise distribution.\n *\n * # Input shape\n *         Arbitrary. Use the keyword argument `input_shape`\n *         (tuple of integers, does not include the samples axis)\n *         when using this layer as the first layer in a model.\n *\n * # Output shape\n *         Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianNoise(args: GaussianNoiseArgs): GaussianNoise {\n  return new GaussianNoise(args);\n}\n\n/**\n * Apply multiplicative 1-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting](\n *      http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianDropout(args: GaussianDropoutArgs): GaussianDropout {\n  return new GaussianDropout(args);\n}\n\n/**\n * Applies Alpha Dropout to the input.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\n * to their original values, in order to ensure the self-normalizing property\n * even after this dropout.\n * Alpha Dropout fits well to Scaled Exponential Linear Units\n * by randomly setting activations to the negative saturation value.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\n *     shape for randomly generated keep/drop flags.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function alphaDropout(args: AlphaDropoutArgs): AlphaDropout {\n  return new AlphaDropout(args);\n}\n\n/**\n * Masks a sequence by using a mask value to skip timesteps.\n *\n * If all features for a given sample timestep are equal to `mask_value`,\n * then the sample timestep will be masked (skipped) in all downstream layers\n * (as long as they support masking).\n *\n * If any downstream layer does not support masking yet receives such\n * an input mask, an exception will be raised.\n *\n * Arguments:\n *   - `maskValue`: Either None or mask value to skip.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Mask', namespace: 'layers'}\n */\nexport function masking(args?: MaskingArgs): Layer {\n  return new Masking(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport * as losses from './losses';\nimport * as metrics from './metrics';\n\n/**\n * Binary accuracy metric function.\n *\n * `yTrue` and `yPred` can have 0-1 values. Example:\n * ```js\n * const x = tf.tensor2d([[1, 1, 1, 1], [0, 0, 0, 0]], [2, 4]);\n * const y = tf.tensor2d([[1, 0, 1, 0], [0, 0, 0, 1]], [2, 4]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * `yTrue` and `yPred` can also have floating-number values between 0 and 1, in\n * which case the values will be thresholded at 0.5 to yield 0-1 values (i.e.,\n * a value >= 0.5 and <= 1.0 is interpreted as 1.\n * )\n * Example:\n * ```js\n * const x = tf.tensor1d([1, 1, 1, 1, 0, 0, 0, 0]);\n * const y = tf.tensor1d([0.2, 0.4, 0.6, 0.8, 0.2, 0.3, 0.4, 0.7]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryAccuracy(yTrue, yPred);\n}\n\n/**\n * Binary crossentropy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0], [1], [1], [1]]);\n * const y = tf.tensor2d([[0], [0], [0.5], [1]]);\n * const crossentropy = tf.metrics.binaryCrossentropy(x, y);\n * crossentropy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction, probabilities for the `1` case.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryCrossentropy(yTrue, yPred);\n}\n\n/**\n * Sparse categorical accuracy metric function.\n *\n * Example:\n * ```js\n *\n * const yTrue = tf.tensor1d([1, 1, 2, 2, 0]);\n * const yPred = tf.tensor2d(\n *      [[0, 1, 0], [1, 0, 0], [0, 0.4, 0.6], [0, 0.6, 0.4], [0.7, 0.3, 0]]);\n * const crossentropy = tf.metrics.sparseCategoricalAccuracy(yTrue, yPred);\n * crossentropy.print();\n * ```\n *\n * @param yTrue True labels: indices.\n * @param yPred Predicted probabilities or logits.\n * @returns Accuracy tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.sparseCategoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical accuracy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0, 0, 0, 1], [0, 0, 0, 1]]);\n * const y = tf.tensor2d([[0.1, 0.8, 0.05, 0.05], [0.1, 0.05, 0.05, 0.8]]);\n * const accuracy = tf.metrics.categoricalAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth: one-hot encoding of categories.\n * @param yPred Binary Tensor of prediction: probabilities or logits for the\n *   same categories as in `yTrue`.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical crossentropy between an output tensor and a target tensor.\n *\n * @param target A tensor of the same shape as `output`.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalCrossentropy(yTrue, yPred);\n}\n\n/**\n * Computes the precision of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const precision = tf.metrics.precision(x, y);\n * precision.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Precision Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.precision(yTrue, yPred);\n}\n\n/**\n * Computes the recall of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const recall = tf.metrics.recall(x, y);\n * recall.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Recall Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.recall(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Cosine proximity.\n *\n * Mathematically, cosine proximity is defined as:\n *   `-sum(l2Normalize(yTrue) * l2Normalize(yPred))`,\n * wherein `l2Normalize()` normalizes the L2 norm of the input to 1 and `*`\n * represents element-wise multiplication.\n *\n * ```js\n * const yTrue = tf.tensor2d([[1, 0], [1, 0]]);\n * const yPred = tf.tensor2d([[1 / Math.sqrt(2), 1 / Math.sqrt(2)], [0, 1]]);\n * const proximity = tf.metrics.cosineProximity(yTrue, yPred);\n * proximity.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Cosine proximity Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.cosineProximity(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute error.\n *\n * Mathematically, mean absolute error is defined as:\n *   `mean(abs(yPred - yTrue))`,\n * wherein the `mean` is applied over feature dimensions.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [0, 0], [2, 3]]);\n * const yPred = tf.tensor2d([[0, 1], [0, 1], [-2, -3]]);\n * const mse = tf.metrics.meanAbsoluteError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsoluteError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute percentage error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [10, 20]]);\n * const yPred = tf.tensor2d([[0, 1], [11, 24]]);\n * const mse = tf.metrics.meanAbsolutePercentageError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MAPE`, `tf.metrics.mape`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute percentage error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function MAPE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function mape(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean squared error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [3, 4]]);\n * const yPred = tf.tensor2d([[0, 1], [-3, -4]]);\n * const mse = tf.metrics.meanSquaredError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MSE`, `tf.metrics.mse`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean squared error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function MSE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function mse(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nexport {modelFromJSON} from './models';\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport * as regularizers from './regularizers';\n// tslint:disable-next-line:max-line-length\nimport {L1Args, L1L2, L1L2Args, L2Args, Regularizer} from './regularizers';\n\n/**\n * Regularizer for L1 and L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x)) + sum(l2 * x^2)\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1l2(config?: L1L2Args): Regularizer {\n  return new L1L2(config);\n}\n\n/**\n * Regularizer for L1 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x))\n * @param args l1 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1(config?: L1Args): Regularizer {\n  return regularizers.l1(config);\n}\n\n/**\n * Regularizer for L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l2 * x^2)\n * @param args l2 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l2(config?: L2Args): Regularizer {\n  return regularizers.l2(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {BaseCallback} from './base_callbacks';\nimport {Container} from './engine/container';\nimport {LayersModel} from './engine/training';\nimport {NotImplementedError} from './errors';\nimport {Logs, resolveScalarsInLogs} from './logs';\n\nexport abstract class Callback extends BaseCallback {\n  /** Instance of `keras.models.Model`. Reference of the model being trained. */\n  model: LayersModel = null;\n\n  setModel(model: Container): void {\n    if (!(model instanceof LayersModel)) {\n      throw new Error('model must be a LayersModel, not some other Container');\n    }\n    this.model = model;\n  }\n}\n\nexport interface EarlyStoppingCallbackArgs {\n  /**\n   * Quantity to be monitored.\n   *\n   * Defaults to 'val_loss'.\n   */\n  monitor?: string;\n\n  /**\n   * Minimum change in the monitored quantity to qualify as improvement,\n   * i.e., an absolute change of less than `minDelta` will count as no\n   * improvement.\n   *\n   * Defaults to 0.\n   */\n  minDelta?: number;\n\n  /**\n   * Number of epochs with no improvement after which training will be stopped.\n   *\n   * Defaults to 0.\n   */\n  patience?: number;\n\n  /** Verbosity mode. */\n  verbose?: number;\n\n  /**\n   * Mode: one of 'min', 'max', and 'auto'.\n   * - In 'min' mode, training will be stopped when the quantity monitored has\n   *   stopped decreasing.\n   * - In 'max' mode, training will be stopped when the quantity monitored has\n   *   stopped increasing.\n   * - In 'auto' mode, the direction is inferred automatically from the name of\n   *   the monitored quantity.\n   *\n   * Defaults to 'auto'.\n   */\n  mode?: 'auto'|'min'|'max';\n\n  /**\n   * Baseline value of the monitored quantity.\n   *\n   * If specified, training will be stopped if the model doesn't show\n   * improvement over the baseline.\n   */\n  baseline?: number;\n\n  /**\n   * Whether to restore model weights from the epoch with the best value\n   * of the monitored quantity. If `False`, the model weights obtained at the\n   * at the last step of training are used.\n   *\n   * **`True` is not supported yet.**\n   */\n  restoreBestWeights?: boolean;\n}\n\nfunction less(currVal: number, prevVal: number) {\n  return currVal < prevVal;\n}\n\nfunction greater(currVal: number, prevVal: number) {\n  return currVal > prevVal;\n}\n\n/**\n * A Callback that stops training when a monitored quantity has stopped\n * improving.\n */\nexport class EarlyStopping extends Callback {\n  protected readonly monitor: string;\n  protected readonly minDelta: number;\n  protected readonly patience: number;\n  protected readonly baseline: number;\n  protected readonly verbose: number;\n  protected readonly mode: 'auto'|'min'|'max';\n\n  protected monitorFunc: (currVal: number, prevVal: number) => boolean;\n\n  private wait: number;\n  private stoppedEpoch: number;\n  private best: number;\n\n  constructor(args?: EarlyStoppingCallbackArgs) {\n    super();\n    if (args == null) {\n      args = {};\n    }\n    if (args.restoreBestWeights) {\n      throw new NotImplementedError(\n          'restoreBestWeights = True is not implemented in EarlyStopping yet.');\n    }\n\n    this.monitor = args.monitor || 'val_loss';\n    this.minDelta = Math.abs(args.minDelta || 0);\n    this.patience = args.patience || 0;\n    this.verbose = args.verbose || 0;\n    this.mode = args.mode || 'auto';\n    this.baseline = args.baseline;\n\n    if (['auto', 'min', 'max'].indexOf(this.mode) === -1) {\n      console.warn(\n          `EarlyStopping mode '${this.mode}' is invalid. ` +\n          `Falling back to mode 'auto'.`);\n      this.mode = 'auto';\n    }\n\n    if (this.mode === 'min') {\n      this.monitorFunc = less;\n    } else if (this.mode === 'max') {\n      this.monitorFunc = greater;\n    } else {\n      // For mode === 'auto'.\n      if (this.monitor.indexOf('acc') !== -1) {\n        this.monitorFunc = greater;\n      } else {\n        this.monitorFunc = less;\n      }\n    }\n\n    if (this.monitorFunc === less) {\n      this.minDelta *= -1;\n    }\n  }\n\n  async onTrainBegin(logs?: Logs) {\n    this.wait = 0;\n    this.stoppedEpoch = 0;\n    if (this.baseline != null) {\n      this.best = this.baseline;\n    } else {\n      this.best = this.monitorFunc === less ? Infinity : -Infinity;\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: Logs) {\n    await resolveScalarsInLogs(logs);\n    const current = this.getMonitorValue(logs);\n    if (current == null) {\n      return;\n    }\n\n    if (this.monitorFunc(current - this.minDelta, this.best)) {\n      this.best = current;\n      this.wait = 0;\n      // TODO(cais): Logic for restoreBestWeights.\n    } else {\n      this.wait++;\n      if (this.wait >= this.patience) {\n        this.stoppedEpoch = epoch;\n        this.model.stopTraining = true;\n      }\n      // TODO(cais): Logic for restoreBestWeights.\n    }\n  }\n\n  async onTrainEnd(logs?: Logs) {\n    if (this.stoppedEpoch > 0 && this.verbose) {\n      console.log(`Epoch ${this.stoppedEpoch}: early stopping.`);\n    }\n  }\n\n  private getMonitorValue(logs: Logs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const monitorValue = logs[this.monitor];\n    if (monitorValue == null) {\n      console.warn(\n          `Metric for EarlyStopping ${this.monitor} is not available. ` +\n          `Available metrics are: ${Object.keys(logs)}`);\n    }\n    return monitorValue;\n  }\n}\n\n/**\n * Factory function for a Callback that stops training when a monitored\n * quantity has stopped improving.\n *\n * Early stopping is a type of regularization, and protects model against\n * overfitting.\n *\n * The following example based on fake data illustrates how this callback\n * can be used during `tf.LayersModel.fit()`:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense({\n *   units: 3,\n *   activation: 'softmax',\n *   kernelInitializer: 'ones',\n *   inputShape: [2]\n * }));\n * const xs = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const ys = tf.tensor2d([[1, 0, 0], [0, 1, 0]], [2, 3]);\n * const xsVal = tf.tensor2d([4, 3, 2, 1], [2, 2]);\n * const ysVal = tf.tensor2d([[0, 0, 1], [0, 1, 0]], [2, 3]);\n * model.compile(\n *     {loss: 'categoricalCrossentropy', optimizer: 'sgd', metrics: ['acc']});\n *\n * // Without the EarlyStopping callback, the val_acc value would be:\n * //   0.5, 0.5, 0.5, 0.5, ...\n * // With val_acc being monitored, training should stop after the 2nd epoch.\n * const history = await model.fit(xs, ys, {\n *   epochs: 10,\n *   validationData: [xsVal, ysVal],\n *   callbacks: tf.callbacks.earlyStopping({monitor: 'val_acc'})\n * });\n *\n * // Expect to see a length-2 array.\n * console.log(history.history.val_acc);\n * ```\n *\n * @doc {\n *   heading: 'Callbacks',\n *   namespace: 'callbacks'\n * }\n */\nexport function earlyStopping(args?: EarlyStoppingCallbackArgs) {\n  return new EarlyStopping(args);\n}\n\nexport const callbacks = {earlyStopping};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/* tslint:disable */\n\n/** Properties of an Any. */\nexport declare interface IAny {\n  /** Any typeUrl */\n  typeUrl?: (string|null);\n\n  /** Any value */\n  value?: (Uint8Array|null);\n}\n\n/** DataType enum. */\nexport enum DataType {\n  'DT_INVALID' = 0,\n  'DT_FLOAT' = 1,\n  'DT_DOUBLE' = 2,\n  'DT_INT32' = 3,\n  'DT_UINT8' = 4,\n  'DT_INT16' = 5,\n  'DT_INT8' = 6,\n  'DT_STRING' = 7,\n  'DT_COMPLEX64' = 8,\n  'DT_INT64' = 9,\n  'DT_BOOL' = 10,\n  'DT_QINT8' = 11,\n  'DT_QUINT8' = 12,\n  'DT_QINT32' = 13,\n  'DT_BFLOAT16' = 14,\n  'DT_FLOAT_REF' = 101,\n  'DT_DOUBLE_REF' = 102,\n  'DT_INT32_REF' = 103,\n  'DT_UINT8_REF' = 104,\n  'DT_INT16_REF' = 105,\n  'DT_INT8_REF' = 106,\n  'DT_STRING_REF' = 107,\n  'DT_COMPLEX64_REF' = 108,\n  'DT_INT64_REF' = 109,\n  'DT_BOOL_REF' = 110,\n  'DT_QINT8_REF' = 111,\n  'DT_QUINT8_REF' = 112,\n  'DT_QINT32_REF' = 113,\n  'DT_BFLOAT16_REF' = 114\n}\n\n/** Properties of a TensorShape. */\nexport declare interface ITensorShape {\n  /** TensorShape dim */\n  dim?: (TensorShape.IDim[]|null);\n\n  /** TensorShape unknownRank */\n  unknownRank?: (boolean|null);\n}\n\nexport namespace TensorShape {\n  /** Properties of a Dim. */\n  export declare interface IDim {\n    /** Dim size */\n    size?: (number|string|null);\n\n    /** Dim name */\n    name?: (string|null);\n  }\n}\n\n/** Properties of a Tensor. */\nexport declare interface ITensor {\n  /** Tensor dtype */\n  dtype?: (DataType|null);\n\n  /** Tensor tensorShape */\n  tensorShape?: (ITensorShape|null);\n\n  /** Tensor versionNumber */\n  versionNumber?: (number|null);\n\n  /** Tensor tensorContent */\n  tensorContent?: (Uint8Array|null);\n\n  /** Tensor floatVal */\n  floatVal?: (number[]|null);\n\n  /** Tensor doubleVal */\n  doubleVal?: (number[]|null);\n\n  /** Tensor intVal */\n  intVal?: (number[]|null);\n\n  /** Tensor stringVal */\n  stringVal?: (Uint8Array[]|null);\n\n  /** Tensor scomplexVal */\n  scomplexVal?: (number[]|null);\n\n  /** Tensor int64Val */\n  int64Val?: ((number | string)[]|null);\n\n  /** Tensor boolVal */\n  boolVal?: (boolean[]|null);\n\n  /** Tensor uint32Val */\n  uint32Val?: (number[]|null);\n\n  /** Tensor uint64Val */\n  uint64Val?: ((number | string)[]|null);\n}\n\n/** Properties of an AttrValue. */\nexport declare interface IAttrValue {\n  /** AttrValue list */\n  list?: (AttrValue.IListValue|null);\n\n  /** AttrValue s */\n  s?: (string|null);\n\n  /** AttrValue i */\n  i?: (number|string|null);\n\n  /** AttrValue f */\n  f?: (number|null);\n\n  /** AttrValue b */\n  b?: (boolean|null);\n\n  /** AttrValue type */\n  type?: (DataType|null);\n\n  /** AttrValue shape */\n  shape?: (ITensorShape|null);\n\n  /** AttrValue tensor */\n  tensor?: (ITensor|null);\n\n  /** AttrValue placeholder */\n  placeholder?: (string|null);\n\n  /** AttrValue func */\n  func?: (INameAttrList|null);\n}\n\nexport namespace AttrValue {\n  /** Properties of a ListValue. */\n  export declare interface IListValue {\n    /** ListValue s */\n    s?: (string[]|null);\n\n    /** ListValue i */\n    i?: ((number | string)[]|null);\n\n    /** ListValue f */\n    f?: (number[]|null);\n\n    /** ListValue b */\n    b?: (boolean[]|null);\n\n    /** ListValue type */\n    type?: (DataType[]|null);\n\n    /** ListValue shape */\n    shape?: (ITensorShape[]|null);\n\n    /** ListValue tensor */\n    tensor?: (ITensor[]|null);\n\n    /** ListValue func */\n    func?: (INameAttrList[]|null);\n  }\n}\n\n/** Properties of a NameAttrList. */\nexport declare interface INameAttrList {\n  /** NameAttrList name */\n  name?: (string|null);\n\n  /** NameAttrList attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a NodeDef. */\nexport declare interface INodeDef {\n  /** NodeDef name */\n  name?: (string|null);\n\n  /** NodeDef op */\n  op?: (string|null);\n\n  /** NodeDef input */\n  input?: (string[]|null);\n\n  /** NodeDef device */\n  device?: (string|null);\n\n  /** NodeDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a VersionDef. */\nexport declare interface IVersionDef {\n  /** VersionDef producer */\n  producer?: (number|null);\n\n  /** VersionDef minConsumer */\n  minConsumer?: (number|null);\n\n  /** VersionDef badConsumers */\n  badConsumers?: (number[]|null);\n}\n\n/** Properties of a GraphDef. */\nexport declare interface IGraphDef {\n  /** GraphDef node */\n  node?: (INodeDef[]|null);\n\n  /** GraphDef versions */\n  versions?: (IVersionDef|null);\n\n  /** GraphDef library */\n  library?: (IFunctionDefLibrary|null);\n}\n\n/** Properties of a CollectionDef. */\nexport declare interface ICollectionDef {\n  /** CollectionDef nodeList */\n  nodeList?: (CollectionDef.INodeList|null);\n\n  /** CollectionDef bytesList */\n  bytesList?: (CollectionDef.IBytesList|null);\n\n  /** CollectionDef int64List */\n  int64List?: (CollectionDef.IInt64List|null);\n\n  /** CollectionDef floatList */\n  floatList?: (CollectionDef.IFloatList|null);\n\n  /** CollectionDef anyList */\n  anyList?: (CollectionDef.IAnyList|null);\n}\n\nexport namespace CollectionDef {\n  /** Properties of a NodeList. */\n  export declare interface INodeList {\n    /** NodeList value */\n    value?: (string[]|null);\n  }\n\n  /** Properties of a BytesList. */\n  export declare interface IBytesList {\n    /** BytesList value */\n    value?: (Uint8Array[]|null);\n  }\n\n  /** Properties of an Int64List. */\n  export declare interface IInt64List {\n    /** Int64List value */\n    value?: ((number | string)[]|null);\n  }\n\n  /** Properties of a FloatList. */\n  export declare interface IFloatList {\n    /** FloatList value */\n    value?: (number[]|null);\n  }\n\n  /** Properties of an AnyList. */\n  export declare interface IAnyList {\n    /** AnyList value */\n    value?: (IAny[]|null);\n  }\n}\n\n/** Properties of a SaverDef. */\nexport declare interface ISaverDef {\n  /** SaverDef filenameTensorName */\n  filenameTensorName?: (string|null);\n\n  /** SaverDef saveTensorName */\n  saveTensorName?: (string|null);\n\n  /** SaverDef restoreOpName */\n  restoreOpName?: (string|null);\n\n  /** SaverDef maxToKeep */\n  maxToKeep?: (number|null);\n\n  /** SaverDef sharded */\n  sharded?: (boolean|null);\n\n  /** SaverDef keepCheckpointEveryNHours */\n  keepCheckpointEveryNHours?: (number|null);\n\n  /** SaverDef version */\n  version?: (SaverDef.CheckpointFormatVersion|null);\n}\n\nexport namespace SaverDef {\n  /** CheckpointFormatVersion enum. */\n  export enum CheckpointFormatVersion {'LEGACY' = 0, 'V1' = 1, 'V2' = 2}\n}\n\n/** Properties of a TensorInfo. */\nexport declare interface ITensorInfo {\n  /** TensorInfo name */\n  name?: (string|null);\n\n  /** TensorInfo cooSparse */\n  cooSparse?: (TensorInfo.ICooSparse|null);\n\n  /** TensorInfo dtype */\n  dtype?: (DataType|null);\n\n  /** TensorInfo tensorShape */\n  tensorShape?: (ITensorShape|null);\n}\n\nexport namespace TensorInfo {\n  /** Properties of a CooSparse. */\n  export declare interface ICooSparse {\n    /** CooSparse valuesTensorName */\n    valuesTensorName?: (string|null);\n\n    /** CooSparse indicesTensorName */\n    indicesTensorName?: (string|null);\n\n    /** CooSparse denseShapeTensorName */\n    denseShapeTensorName?: (string|null);\n  }\n}\n\n/** Properties of a SignatureDef. */\nexport declare interface ISignatureDef {\n  /** SignatureDef inputs */\n  inputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef outputs */\n  outputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef methodName */\n  methodName?: (string|null);\n}\n\n/** Properties of an AssetFileDef. */\nexport declare interface IAssetFileDef {\n  /** AssetFileDef tensorInfo */\n  tensorInfo?: (ITensorInfo|null);\n\n  /** AssetFileDef filename */\n  filename?: (string|null);\n}\n\n/** Properties of an OpDef. */\nexport declare interface IOpDef {\n  /** OpDef name */\n  name?: (string|null);\n\n  /** OpDef inputArg */\n  inputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef outputArg */\n  outputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef attr */\n  attr?: (OpDef.IAttrDef[]|null);\n\n  /** OpDef deprecation */\n  deprecation?: (OpDef.IOpDeprecation|null);\n\n  /** OpDef summary */\n  summary?: (string|null);\n\n  /** OpDef description */\n  description?: (string|null);\n\n  /** OpDef isCommutative */\n  isCommutative?: (boolean|null);\n\n  /** OpDef isAggregate */\n  isAggregate?: (boolean|null);\n\n  /** OpDef isStateful */\n  isStateful?: (boolean|null);\n\n  /** OpDef allowsUninitializedInput */\n  allowsUninitializedInput?: (boolean|null);\n}\n\nexport namespace OpDef {\n  /** Properties of an ArgDef. */\n  export declare interface IArgDef {\n    /** ArgDef name */\n    name?: (string|null);\n\n    /** ArgDef description */\n    description?: (string|null);\n\n    /** ArgDef type */\n    type?: (DataType|null);\n\n    /** ArgDef typeAttr */\n    typeAttr?: (string|null);\n\n    /** ArgDef numberAttr */\n    numberAttr?: (string|null);\n\n    /** ArgDef typeListAttr */\n    typeListAttr?: (string|null);\n\n    /** ArgDef isRef */\n    isRef?: (boolean|null);\n  }\n\n  /** Properties of an AttrDef. */\n  export declare interface IAttrDef {\n    /** AttrDef name */\n    name?: (string|null);\n\n    /** AttrDef type */\n    type?: (string|null);\n\n    /** AttrDef defaultValue */\n    defaultValue?: (IAttrValue|null);\n\n    /** AttrDef description */\n    description?: (string|null);\n\n    /** AttrDef hasMinimum */\n    hasMinimum?: (boolean|null);\n\n    /** AttrDef minimum */\n    minimum?: (number|string|null);\n\n    /** AttrDef allowedValues */\n    allowedValues?: (IAttrValue|null);\n  }\n\n  /** Properties of an OpDeprecation. */\n  export declare interface IOpDeprecation {\n    /** OpDeprecation version */\n    version?: (number|null);\n\n    /** OpDeprecation explanation */\n    explanation?: (string|null);\n  }\n}\n\n/** Properties of an OpList. */\nexport declare interface IOpList {\n  /** OpList op */\n  op?: (IOpDef[]|null);\n}\n\n/** Properties of a MetaGraphDef. */\nexport declare interface IMetaGraphDef {\n  /** MetaGraphDef metaInfoDef */\n  metaInfoDef?: (MetaGraphDef.IMetaInfoDef|null);\n\n  /** MetaGraphDef graphDef */\n  graphDef?: (IGraphDef|null);\n\n  /** MetaGraphDef saverDef */\n  saverDef?: (ISaverDef|null);\n\n  /** MetaGraphDef collectionDef */\n  collectionDef?: ({[k: string]: ICollectionDef}|null);\n\n  /** MetaGraphDef signatureDef */\n  signatureDef?: ({[k: string]: ISignatureDef}|null);\n\n  /** MetaGraphDef assetFileDef */\n  assetFileDef?: (IAssetFileDef[]|null);\n}\n\nexport namespace MetaGraphDef {\n  /** Properties of a MetaInfoDef. */\n  export declare interface IMetaInfoDef {\n    /** MetaInfoDef metaGraphVersion */\n    metaGraphVersion?: (string|null);\n\n    /** MetaInfoDef strippedOpList */\n    strippedOpList?: (IOpList|null);\n\n    /** MetaInfoDef anyInfo */\n    anyInfo?: (IAny|null);\n\n    /** MetaInfoDef tags */\n    tags?: (string[]|null);\n\n    /** MetaInfoDef tensorflowVersion */\n    tensorflowVersion?: (string|null);\n\n    /** MetaInfoDef tensorflowGitVersion */\n    tensorflowGitVersion?: (string|null);\n  }\n}\n\n/** Properties of a SavedModel. */\nexport declare interface ISavedModel {\n  /** SavedModel savedModelSchemaVersion */\n  savedModelSchemaVersion?: (number|string|null);\n\n  /** SavedModel metaGraphs */\n  metaGraphs?: (IMetaGraphDef[]|null);\n}\n\n/** Properties of a FunctionDefLibrary. */\nexport declare interface IFunctionDefLibrary {\n  /** FunctionDefLibrary function */\n  'function'?: (IFunctionDef[]|null);\n\n  /** FunctionDefLibrary gradient */\n  gradient?: (IGradientDef[]|null);\n}\n\n/** Properties of a FunctionDef. */\nexport declare interface IFunctionDef {\n  /** FunctionDef signature */\n  signature?: (IOpDef|null);\n\n  /** FunctionDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n\n  /** FunctionDef nodeDef */\n  nodeDef?: (INodeDef[]|null);\n\n  /** FunctionDef ret */\n  ret?: ({[k: string]: string}|null);\n}\n\n/** Properties of a GradientDef. */\nexport declare interface IGradientDef {\n  /** GradientDef functionName */\n  functionName?: (string|null);\n\n  /** GradientDef gradientFunc */\n  gradientFunc?: (string|null);\n}\n", "\n/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpExecutor, OpMapper} from '../types';\n\nconst CUSTOM_OPS: {[key: string]: OpMapper} = {};\n\n/**\n * Register an Op for graph model executor. This allow you to register\n * TensorFlow custom op or override existing op.\n *\n * Here is an example of registering a new MatMul Op.\n * ```js\n * const customMatmul = (node) =>\n *    tf.matMul(\n *        node.inputs[0], node.inputs[1],\n *        node.attrs['transpose_a'], node.attrs['transpose_b']);\n *\n * tf.registerOp('MatMul', customMatmul);\n * ```\n * The inputs and attrs of the node object is based on the TensorFlow op\n * registry.\n *\n * @param name The Tensorflow Op name.\n * @param opFunc An op function which is called with the current graph node\n * during execution and needs to return a tensor or a list of tensors. The node\n * has the following attributes:\n *    - attr: A map from attribute name to its value\n *    - inputs: A list of input tensors\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function registerOp(name: string, opFunc: OpExecutor) {\n  const opMapper: OpMapper = {\n    tfOpName: name,\n    category: 'custom',\n    inputs: [],\n    attrs: [],\n    customExecutor: opFunc\n  };\n\n  CUSTOM_OPS[name] = opMapper;\n}\n\n/**\n * Retrieve the OpMapper object for the registered op.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function getRegisteredOp(name: string): OpMapper {\n  return CUSTOM_OPS[name];\n}\n\n/**\n * Deregister the Op for graph model executor.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function deregisterOp(name: string) {\n  delete CUSTOM_OPS[name];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {clone, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {Node, ValueType} from '../types';\n\nexport function getParamValue(\n    paramName: string, node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  const inputParam = node.inputParams[paramName];\n  if (inputParam && inputParam.inputIndexStart !== undefined) {\n    const start = inputParam.inputIndexStart;\n    const end = inputParam.inputIndexEnd === 0 ?\n        undefined :\n        (inputParam.inputIndexEnd === undefined ? start + 1 :\n                                                  inputParam.inputIndexEnd);\n    if (inputParam.type === 'tensor') {\n      return getTensor(\n          node.inputNames[inputParam.inputIndexStart], tensorMap, context);\n    }\n    if (inputParam.type === 'tensors') {\n      const inputs = node.inputNames.slice(start, end);\n\n      return inputs.map(name => getTensor(name, tensorMap, context));\n    }\n    const tensor =\n        getTensor(node.inputNames.slice(start)[0], tensorMap, context);\n    const data = tensor.dataSync();\n    return inputParam.type === 'number' ?\n        data[0] :\n        util.toNestedArray(tensor.shape, data);\n  }\n  const attrParam = node.attrParams[paramName];\n  return attrParam && attrParam.value;\n}\n\n/**\n * Retrieve the tensor based on input name by extracting the node name and\n * output index information.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensor(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor {\n  const [nodeName, index] = parseNodeName(name);\n  const contextId = context.currentContextIds.find(contextId => {\n    return !!tensorsMap[getNodeNameWithContextId(nodeName, contextId)];\n  });\n\n  return contextId !== undefined ?\n      tensorsMap[getNodeNameWithContextId(nodeName, contextId)][index] :\n      undefined;\n}\n\n/**\n * Retrieve the tensors based on input name for current context.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensorsForCurrentContenxt(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor[] {\n  return tensorsMap[getNodeNameWithContextId(name, context.currentContextId)];\n}\n\n/**\n * Returns the node name and index from the Node input name.\n * @param inputName The input name of the node, in format of\n * node_name:output_index, i.e. MatMul:0, if the output_index is not set, it is\n * default to 0.\n */\nexport function getNodeNameAndIndex(\n    inputName: string, context?: ExecutionContext): [string, number] {\n  const [nodeName, index] = parseNodeName(inputName);\n\n  return [\n    getNodeNameWithContextId(nodeName, context && context.currentContextId),\n    index\n  ];\n}\n\nfunction getNodeNameWithContextId(name: string, contextId?: string): string {\n  return !!contextId ? `${name}-${contextId}` : name;\n}\n\nexport function parseNodeName(name: string): [string, number] {\n  const parts = name.split(':');\n  if (parts.length === 1) {\n    return [name, 0];\n  }\n\n  const nodeName = parts[0];\n  return [nodeName, Number(parts[parts.length - 1])];\n}\n\nexport function split(arr: number[], size: number) {\n  const res = [];\n  for (let i = 0; i < arr.length; i += size) {\n    res.push(arr.slice(i, i + size));\n  }\n  return res;\n}\nexport function getPadding(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  let pad = getParamValue('pad', node, tensorMap, context);\n  if (pad === 'explicit') {\n    // This is 1d array, we need to convert it to 2d array\n    pad = getParamValue('explicitPaddings', node, tensorMap, context);\n    const explicitPadding: [\n      [number, number], [number, number], [number, number], [number, number]\n    ] = [[0, 0], [0, 0], [0, 0], [0, 0]];\n    for (let i = 0; i < 4; i++) {\n      explicitPadding[i][0] = (pad as number[])[i * 2];\n      explicitPadding[i][1] = (pad as number[])[i * 2 + 1];\n    }\n    return explicitPadding;\n  }\n  return pad;\n}\n\n/**\n *  Reuse the tensor if it is marked as keep, otherwise clone the tensor to\n *  avoid disposal. This is important for TensorArray and TensorList ops, since\n *  internally they use a tensor as the id for TensorArray and TensorList, and\n * to simplify lookup, they also use Tensor.id as the key to the internal map.\n * These id tensors have been marked as kept in the backend, we need avoid clone\n * them in order to create new Tensor.id.\n * @param tensor\n */\nexport function cloneTensor(tensor: Tensor): Tensor {\n  return tensor.kept ? tensor : clone(tensor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Add',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddV2',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddN',\n    'category': 'arithmetic',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'BiasAdd',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sub',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'RealDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Div',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'DivNoNan',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mul',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Maximum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Minimum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Pow',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SquaredDifference',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorMod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Abs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan2',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ceil',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ClipByValue',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'clip_value_min', 'name': 'clipValueMin', 'type': 'number'},\n      {'tfName': 'clip_value_max', 'name': 'clipValueMax', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Complex',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'real', 'type': 'tensor'},\n      {'start': 1, 'name': 'imag', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ComplexAbs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Elu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Exp',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Floor',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Imag',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Neg',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Real',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Prelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'alpha', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu6',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'clipValueMin',\n        'name': 'clipValueMin',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'clipValueMax',\n        'name': 'clipValueMax',\n        'type': 'number',\n        'defaultValue': 6\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Selu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sigmoid',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Rsqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Square',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sign',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Round',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Expm1',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log1p',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Reciprocal',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Softplus',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Erf',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axes', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'keep_dims',\n        'name': 'keepDims',\n        'type': 'bool',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LeakyRelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 0.2\n      },\n      {\n        'tfName': 'T',\n        'name': 'dtype',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'LoopCond',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'pred', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Switch',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'data', 'type': 'tensor'},\n      {'start': 1, 'name': 'pred', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Merge',\n    'category': 'control',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Enter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'frame_name', 'name': 'frameName', 'type': 'string'},\n      {'tfName': 'is_constant', 'name': 'isConstant', 'type': 'bool'}\n    ]\n  },\n  {\n    'tfOpName': 'Exit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NextIteration',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'size', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'dynamic_size', 'name': 'dynamicSize', 'type': 'bool'},\n      {'tfName': 'clear_after_read', 'name': 'clearAfterRead', 'type': 'bool'},\n      {\n        'tfName': 'identical_element_shapes',\n        'name': 'identicalElementShapes',\n        'type': 'bool'\n      },\n      {'tfName': 'tensor_array_name', 'name': 'name', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayWriteV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayReadV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{\n      'tfName': 'dtype',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  },\n  {\n    'tfOpName': 'TensorArrayGatherV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayScatterV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArrayConcatV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}, {\n        'tfName': 'element_shape_except0',\n        'name': 'elementShapeExcept0',\n        'type': 'shape',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'TensorArraySplitV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArraySizeV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayCloseV3',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'StatelessIf',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'If',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'StatelessWhile',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'While',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListScatter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListScatterV2',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 3, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGather',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListSetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListReserve',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 1, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListFromTensor',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListStack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs': [\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'},\n      {'tfName': 'num_elements', 'name': 'numElements', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListSplit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListConcat',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListPopBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListPushBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n];\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'AvgPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPoolWithArgmax',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'}, {\n        'tfName': 'include_batch_in_index',\n        'name': 'includeBatchInIndex',\n        'type': 'bool'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AvgPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Conv1D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'stride', 'name': 'stride', 'type': 'number'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NWC'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'dilation',\n        'name': 'dilation',\n        'type': 'number',\n        'defaultValue': 1\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'useCudnnOnGpu', 'name': 'useCudnnOnGpu', 'type': 'bool'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': '_FusedConv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'use_cudnn_on_gpu',\n        'name': 'useCudnnOnGpu',\n        'type': 'bool',\n        'defaultValue': true\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n    ]\n  },\n  {\n    'tfOpName': 'Conv2DBackpropInput',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 2, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 0, 'name': 'outputShape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2d',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'FusedDepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ],\n  },\n  {\n    'tfOpName': 'Dilation2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'rates', 'name': 'dilations', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Fill',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n      {'start': 1, 'name': 'value', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'LinSpace',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'num', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'OneHot',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'depth', 'type': 'number'},\n      {'start': 2, 'name': 'onValue', 'type': 'number', 'defaultValue': 1},\n      {'start': 3, 'name': 'offValue', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [\n      {\n        'tfName': 'axis',\n        'name': 'axis',\n        'type': 'number',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ones',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'OnesLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'RandomUniform',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'minval',\n        'name': 'minval',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'maxval',\n        'name': 'maxval',\n        'type': 'number',\n        'defaultValue': 1\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Range',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'step', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [{'tfName': 'Tidx', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TruncatedNormal',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'means',\n        'name': 'mean',\n        'type': 'number',\n        'defaultValue': 0.0\n      },\n      {\n        'tfName': 'stddev',\n        'name': 'stdDev',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Zeros',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'ZerosLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'Multinomial',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'logits', 'type': 'tensor'},\n      {'start': 1, 'name': 'numSamples', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'},\n      {'tfName': 'seed2', 'name': 'seed2', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'output_dtype', 'name': 'output_dtype', 'type': 'dtype'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'NonMaxSuppressionV2',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV3',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV4',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'T_threshold',\n        'name': 'threshold',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {\n        'tfName': 'pad_to_max_output_size',\n        'name': 'padToMaxOutputSize',\n        'type': 'bool'\n      }\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV5',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'},\n      {'start': 5, 'name': 'softNmsSigma', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Where',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ListDiff',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'TopKV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'k', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'sorted', 'name': 'sorted', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Unique',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n  },\n  {\n    'tfOpName': 'UniqueV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n  },\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'PlaceholderWithDefault',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'default', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'Placeholder',\n    'category': 'graph',\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {'tfOpName': 'Const', 'category': 'graph'}, {\n    'tfOpName': 'Identity',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IdentityN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Snapshot',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Rank',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Size',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Shape',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'ShapeN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Print',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'data', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'message', 'name': 'message', 'type': 'string'}, {\n        'tfName': 'first_n',\n        'name': 'firstN',\n        'type': 'number',\n        'notSupported': true\n      },\n      {\n        'tfName': 'summarize',\n        'name': 'summarize',\n        'type': 'number',\n        'defaultValue': 3\n      }\n    ]\n  },\n  {'tfOpName': 'NoOp', 'category': 'graph', 'inputs': []}, {\n    'tfOpName': 'StopGradient',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'FakeQuantWithMinMaxVars',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'min', 'name': 'min', 'type': 'number'},\n      {'tfName': 'max', 'name': 'max', 'type': 'number'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ResizeBilinear',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ResizeNearestNeighbor',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'CropAndResize',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'image', 'type': 'tensor'},\n      {'start': 1, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 2, 'name': 'boxInd', 'type': 'tensor'},\n      {'start': 3, 'name': 'cropSize', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'method', 'name': 'method', 'type': 'string'}, {\n        'tfName': 'extrapolation_value',\n        'name': 'extrapolationValue',\n        'type': 'number'\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Equal',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NotEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Greater',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'GreaterEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Less',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LessEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalAnd',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalNot',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalOr',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Select',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SelectV2',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': '_FusedMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'}, {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMulV2',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Transpose',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'perm', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FusedBatchNorm',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV2',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV3',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'LRN',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'depth_radius',\n        'name': 'radius',\n        'type': 'number',\n        'defaultValue': 5\n      },\n      {'tfName': 'bias', 'name': 'bias', 'type': 'number', 'defaultValue': 1.0},\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {\n        'tfName': 'beta',\n        'name': 'beta',\n        'type': 'number',\n        'defaultValue': 0.5\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Softmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'LogSoftmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': true,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Max',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Mean',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Min',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Sum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'All',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Any',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'ArgMax',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'ArgMin',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Cumsum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'exclusive', 'name': 'exclusive', 'type': 'bool'},\n      {'tfName': 'reverse', 'name': 'reverse', 'type': 'bool'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ConcatV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': -1, 'name': 'tensors', 'type': 'tensors'},\n      {'start': -1, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs':\n        [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n  },\n  {\n    'tfOpName': 'Concat',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 1, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n      {'start': 0, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs': [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n\n  },\n  {\n    'tfOpName': 'GatherV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Gather',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'validate_indices',\n        'name': 'validateIndices',\n        'type': 'bool',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reverse',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'dims', 'type': 'bool', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ReverseV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Slice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'size', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'StridedSlice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'end', 'type': 'number[]'},\n      {'start': 3, 'name': 'strides', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'begin_mask',\n        'name': 'beginMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'end_mask',\n        'name': 'endMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'new_axis_mask',\n        'name': 'newAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'ellipsis_mask',\n        'name': 'ellipsisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'shrink_axis_mask',\n        'name': 'shrinkAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Pack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Unpack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'num',\n        'name': 'num',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Tile',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'reps', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Split',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'axis', 'type': 'number', 'defaultValue': 0},\n      {'start': 1, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'num_split',\n      'name': 'numOrSizeSplits',\n      'type': 'number',\n      'defaultValue': 1\n    }]\n  },\n  {\n    'tfOpName': 'SplitV',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'numOrSizeSplits', 'type': 'number[]'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'ScatterNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'values', 'type': 'tensor'},\n      {'start': 2, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'GatherNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': false,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IFFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'RFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'IRFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Cast',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'SrcT',\n        'name': 'sdtype',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {'tfName': 'DstT', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'ExpandDims',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Pad',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'constant_value',\n      'name': 'constantValue',\n      'type': 'number',\n      'defaultValue': 0\n    }]\n  },\n  {\n    'tfOpName': 'PadV2',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'}, {\n        'start': 2,\n        'name': 'constantValue',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reshape',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Squeeze',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'axis',\n      'tfDeprecatedName': 'squeeze_dims',\n      'name': 'axis',\n      'type': 'number[]'\n    }]\n  },\n  {\n    'tfOpName': 'SpaceToBatchND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'paddings', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'BatchToSpaceND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'crops', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthToSpace',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'block_size', 'name': 'blockSize', 'type': 'number'},\n      {'tfName': 'data_format', 'name': 'dataFormat', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'BroadcastTo',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': []\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, env} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\n\nimport {getRegisteredOp} from './custom_op/register';\nimport {getNodeNameAndIndex} from './executors/utils';\nimport * as arithmetic from './op_list/arithmetic';\nimport * as basicMath from './op_list/basic_math';\nimport * as control from './op_list/control';\nimport * as convolution from './op_list/convolution';\nimport * as creation from './op_list/creation';\nimport * as dynamic from './op_list/dynamic';\nimport * as evaluation from './op_list/evaluation';\nimport * as graph from './op_list/graph';\nimport * as image from './op_list/image';\nimport * as logical from './op_list/logical';\nimport * as matrices from './op_list/matrices';\nimport * as normalization from './op_list/normalization';\nimport * as reduction from './op_list/reduction';\nimport * as sliceJoin from './op_list/slice_join';\nimport * as spectral from './op_list/spectral';\nimport * as transformation from './op_list/transformation';\nimport {Graph, InputParamValue, Node, OpMapper, ParamValue} from './types';\n\nexport class OperationMapper {\n  private static _instance: OperationMapper;\n\n  private opMappers: {[key: string]: OpMapper};\n\n  // Singleton instance for the mapper\n  public static get Instance() {\n    return this._instance || (this._instance = new this());\n  }\n\n  // Loads the op mapping from the JSON file.\n  private constructor() {\n    const ops = [\n      arithmetic, basicMath, control, convolution, creation, dynamic,\n      evaluation, logical, image, graph, matrices, normalization, reduction,\n      sliceJoin, spectral, transformation\n    ];\n    const mappersJson: OpMapper[] = [].concat(...ops.map(op => op.json));\n\n    this.opMappers = mappersJson.reduce<{[key: string]: OpMapper}>(\n        (map, mapper: OpMapper) => {\n          map[mapper.tfOpName] = mapper;\n          return map;\n        },\n        {});\n  }\n\n  // Converts the model inference graph from Tensorflow GraphDef to local\n  // representation for TensorFlow.js API\n  transformGraph(\n      graph: tensorflow.IGraphDef,\n      signature: tensorflow.ISignatureDef = {}): Graph {\n    const tfNodes = graph.node;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    const initNodes: Node[] = [];\n    const nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n      map[node.name] = this.mapNode(node);\n      if (node.op.startsWith('Placeholder')) {\n        placeholders.push(map[node.name]);\n      } else if (node.op === 'Const') {\n        weights.push(map[node.name]);\n      } else if (node.input == null || node.input.length === 0) {\n        initNodes.push(map[node.name]);\n      }\n      return map;\n    }, {});\n\n    let inputs: Node[] = [];\n    const outputs: Node[] = [];\n    let inputNodeNameToKey: {[key: string]: string} = {};\n    let outputNodeNameToKey: {[key: string]: string} = {};\n    if (signature != null) {\n      inputNodeNameToKey = this.mapSignatureEntries(signature.inputs);\n      outputNodeNameToKey = this.mapSignatureEntries(signature.outputs);\n    }\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    // if signature has not outputs set, add any node that does not have\n    // outputs.\n    if (Object.keys(outputNodeNameToKey).length === 0) {\n      allNodes.forEach(key => {\n        const node = nodes[key];\n        if (node.children.length === 0) {\n          outputs.push(node);\n        }\n      });\n    } else {\n      Object.keys(outputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node != null) {\n          node.signatureKey = outputNodeNameToKey[name];\n          outputs.push(node);\n        }\n      });\n    }\n\n    if (Object.keys(inputNodeNameToKey).length > 0) {\n      Object.keys(inputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node) {\n          node.signatureKey = inputNodeNameToKey[name];\n          inputs.push(node);\n        }\n      });\n    } else {\n      inputs = placeholders;\n    }\n\n    let functions = {};\n    if (graph.library != null && graph.library.function != null) {\n      functions = graph.library.function.reduce((functions, func) => {\n        functions[func.signature.name] = this.mapFunction(func);\n        return functions;\n      }, {} as {[key: string]: Graph});\n    }\n\n    const result: Graph =\n        {nodes, inputs, outputs, weights, placeholders, signature, functions};\n\n    if (initNodes.length > 0) {\n      result.initNodes = initNodes;\n    }\n\n    return result;\n  }\n\n  private mapSignatureEntries(entries: {[k: string]: tensorflow.ITensorInfo}) {\n    return Object.keys(entries || {})\n        .reduce<{[key: string]: string}>((prev, curr) => {\n          prev[entries[curr].name] = curr;\n          return prev;\n        }, {});\n  }\n\n  private mapNode(node: tensorflow.INodeDef): Node {\n    // Unsupported ops will cause an error at run-time (not parse time), since\n    // they may not be used by the actual execution subgraph.\n    const mapper =\n        getRegisteredOp(node.op) || this.opMappers[node.op] || {} as OpMapper;\n    if (node.attr == null) {\n      node.attr = {};\n    }\n\n    const newNode: Node = {\n      name: node.name,\n      op: node.op,\n      category: mapper.category,\n      inputNames:\n          (node.input ||\n           []).map(input => input.startsWith('^') ? input.substr(1) : input),\n      inputs: [],\n      children: [],\n      inputParams: {},\n      attrParams: {},\n      rawAttrs: node.attr\n    };\n\n    if (mapper.inputs != null) {\n      newNode.inputParams =\n          mapper.inputs.reduce<{[key: string]: InputParamValue}>(\n              (map, param) => {\n                map[param.name] = {\n                  type: param.type,\n                  inputIndexStart: param.start,\n                  inputIndexEnd: param.end\n                };\n                return map;\n              },\n              {});\n    }\n    if (mapper.attrs != null) {\n      newNode.attrParams =\n          mapper.attrs.reduce<{[key: string]: ParamValue}>((map, param) => {\n            const type = param.type;\n            let value = undefined;\n            switch (param.type) {\n              case 'string':\n                value = getStringParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'string[]':\n                value = getStringArrayParam(\n                    node.attr, param.tfName, param.defaultValue as string[]);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string[]);\n                }\n                break;\n              case 'number':\n                value = getNumberParam(\n                    node.attr, param.tfName,\n                    (param.defaultValue || 0) as number);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumberParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number);\n                }\n                break;\n              case 'number[]':\n                value = getNumericArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumericArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'bool':\n                value = getBoolParam(\n                    node.attr, param.tfName, param.defaultValue as boolean);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean);\n                }\n                break;\n              case 'bool[]':\n                value = getBoolArrayParam(\n                    node.attr, param.tfName, param.defaultValue as boolean[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean[]);\n                }\n                break;\n              case 'shape':\n                value = getTensorShapeParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'shape[]':\n                value = getTensorShapeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[][]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[][]);\n                }\n                break;\n              case 'dtype':\n                value = getDtypeParam(\n                    node.attr, param.tfName, param.defaultValue as DataType);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType);\n                }\n                break;\n              case 'dtype[]':\n                value = getDtypeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as DataType[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType[]);\n                }\n                break;\n              case 'func':\n                value = getFuncParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getFuncParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'tensor':\n              case 'tensors':\n                break;\n              default:\n                throw new Error(\n                    `Unsupported param type: ${param.type} for op: ${node.op}`);\n            }\n            map[param.name] = {value, type};\n            return map;\n          }, {});\n    }\n    return newNode;\n  }\n\n  // map the TFunctionDef to TFJS graph object\n  private mapFunction(functionDef: tensorflow.IFunctionDef): Graph {\n    const tfNodes = functionDef.nodeDef;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    let nodes: {[key: string]: Node} = {};\n    if (tfNodes != null) {\n      nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n        map[node.name] = this.mapNode(node);\n        if (node.op === 'Const') {\n          weights.push(map[node.name]);\n        }\n        return map;\n      }, {});\n    }\n    const inputs: Node[] = [];\n    const outputs: Node[] = [];\n\n    functionDef.signature.inputArg.forEach(arg => {\n      const [nodeName, ] = getNodeNameAndIndex(arg.name);\n      const node: Node = {\n        name: nodeName,\n        op: 'Placeholder',\n        inputs: [],\n        inputNames: [],\n        category: 'graph',\n        inputParams: {},\n        attrParams: {dtype: {value: parseDtypeParam(arg.type), type: 'dtype'}},\n        children: []\n      };\n      node.signatureKey = arg.name;\n      inputs.push(node);\n      nodes[nodeName] = node;\n    });\n\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    const returnNodeMap = functionDef.ret;\n\n    functionDef.signature.outputArg.forEach(output => {\n      const [nodeName, index] = getNodeNameAndIndex(returnNodeMap[output.name]);\n      const node = nodes[nodeName];\n      if (node != null) {\n        node.defaultOutput = index;\n        outputs.push(node);\n      }\n    });\n\n    const signature = this.mapArgsToSignature(functionDef);\n    return {nodes, inputs, outputs, weights, placeholders, signature};\n  }\n\n  private mapArgsToSignature(functionDef: tensorflow.IFunctionDef):\n      tensorflow.ISignatureDef {\n    return {\n      methodName: functionDef.signature.name,\n      inputs: functionDef.signature.inputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n      outputs: functionDef.signature.outputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg, functionDef.ret);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n    };\n  }\n\n  private mapArgToTensorInfo(\n      arg: tensorflow.OpDef.IArgDef,\n      nameMap?: {[key: string]: string}): tensorflow.ITensorInfo {\n    let name = arg.name;\n    if (nameMap != null) {\n      name = nameMap[name];\n    }\n    return {name, dtype: arg.type};\n  }\n}\n\nexport function decodeBase64(text: string): string {\n  const global = env().global;\n  if (typeof global.atob !== 'undefined') {\n    return global.atob(text);\n  } else if (typeof Buffer !== 'undefined') {\n    return new Buffer(text, 'base64').toString();\n  } else {\n    throw new Error(\n        'Unable to decode base64 in this environment. ' +\n        'Missing built-in atob() or Buffer()');\n  }\n}\n\nexport function parseStringParam(s: []|string, keepCase: boolean): string {\n  const value =\n      Array.isArray(s) ? String.fromCharCode.apply(null, s) : decodeBase64(s);\n  return keepCase ? value : value.toLowerCase();\n}\n\nexport function getStringParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string,\n    keepCase = false): string {\n  const param = attrs[name];\n  if (param != null) {\n    return parseStringParam(param.s, keepCase);\n  }\n  return def;\n}\n\nexport function getBoolParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean): boolean {\n  const param = attrs[name];\n  return param ? param.b : def;\n}\n\nexport function getNumberParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number): number {\n  const param = attrs[name] || {};\n  const value =\n      param['i'] != null ? param['i'] : (param['f'] != null ? param['f'] : def);\n  return (typeof value === 'number') ? value : parseInt(value, 10);\n}\n\nexport function parseDtypeParam(value: string|tensorflow.DataType): DataType {\n  if (typeof (value) === 'string') {\n    // tslint:disable-next-line:no-any\n    value = tensorflow.DataType[value as any];\n  }\n  switch (value) {\n    case tensorflow.DataType.DT_FLOAT:\n      return 'float32';\n    case tensorflow.DataType.DT_INT32:\n    case tensorflow.DataType.DT_INT64:\n    case tensorflow.DataType.DT_INT8:\n    case tensorflow.DataType.DT_UINT8:\n      return 'int32';\n    case tensorflow.DataType.DT_BOOL:\n      return 'bool';\n    case tensorflow.DataType.DT_DOUBLE:\n      return 'float32';\n    case tensorflow.DataType.DT_STRING:\n      return 'string';\n    default:\n      // Unknown dtype error will happen at runtime (instead of parse time),\n      // since these nodes might not be used by the actual subgraph execution.\n      return null;\n  }\n}\n\nexport function getFuncParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: string): string {\n  const param = attrs[name];\n  if (param && param.func) {\n    return param.func.name;\n  }\n  return def;\n}\n\nexport function getDtypeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType): DataType {\n  const param = attrs[name];\n  if (param && param.type) {\n    return parseDtypeParam(param.type);\n  }\n  return def;\n}\n\nexport function getDtypeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType[]): DataType[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.type) {\n    return param.list.type.map(v => parseDtypeParam(v));\n  }\n  return def;\n}\n\nexport function parseTensorShapeParam(shape: tensorflow.ITensorShape): number[]|\n    undefined {\n  if (shape.unknownRank) {\n    return undefined;\n  }\n  if (shape.dim != null) {\n    return shape.dim.map(\n        dim =>\n            (typeof dim.size === 'number') ? dim.size : parseInt(dim.size, 10));\n  }\n  return [];\n}\n\nexport function getTensorShapeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def?: number[]): number[]|undefined {\n  const param = attrs[name];\n  if (param && param.shape) {\n    return parseTensorShapeParam(param.shape);\n  }\n  return def;\n}\n\nexport function getNumericArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[]): number[] {\n  const param = attrs[name];\n  if (param) {\n    return ((param.list.f && param.list.f.length ? param.list.f :\n                                                   param.list.i) ||\n            [])\n        .map(v => (typeof v === 'number') ? v : parseInt(v, 10));\n  }\n  return def;\n}\n\nexport function getStringArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string[],\n    keepCase = false): string[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.s) {\n    return param.list.s.map((v) => {\n      return parseStringParam(v, keepCase);\n    });\n  }\n  return def;\n}\n\nexport function getTensorShapeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[][]): number[][] {\n  const param = attrs[name];\n  if (param && param.list && param.list.shape) {\n    return param.list.shape.map((v) => {\n      return parseTensorShapeParam(v);\n    });\n  }\n  return def;\n}\n\nexport function getBoolArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean[]): boolean[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.b) {\n    return param.list.b;\n  }\n  return def;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {getTensor} from '../executors/utils';\nimport {getBoolArrayParam, getBoolParam, getDtypeArrayParam, getDtypeParam, getNumberParam, getNumericArrayParam, getStringArrayParam, getStringParam, getTensorShapeArrayParam, getTensorShapeParam} from '../operation_mapper';\nimport {GraphNode, Node, ValueType} from '../types';\n\n/**\n * Helper class for lookup inputs and params for nodes in the model graph.\n */\nexport class NodeValueImpl implements GraphNode {\n  public readonly inputs: Tensor[] = [];\n  public readonly attrs: {[key: string]: ValueType} = {};\n  constructor(\n      private node: Node, private tensorMap: NamedTensorsMap,\n      private context: ExecutionContext) {\n    this.inputs = node.inputNames.map(name => this.getInput(name));\n    if (node.rawAttrs != null) {\n      this.attrs = Object.keys(node.rawAttrs)\n                       .reduce((attrs: {[key: string]: ValueType}, key) => {\n                         attrs[key] = this.getAttr(key);\n                         return attrs;\n                       }, {});\n    }\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getInput(name: string): Tensor {\n    return getTensor(name, this.tensorMap, this.context);\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getAttr(name: string, defaultValue?: ValueType): ValueType {\n    const value = this.node.rawAttrs[name];\n    if (value.tensor != null) {\n      return getTensor(name, this.tensorMap, this.context);\n    }\n    if (value.i != null || value.f != null) {\n      return getNumberParam(this.node.rawAttrs, name, defaultValue as number);\n    }\n    if (value.s != null) {\n      return getStringParam(this.node.rawAttrs, name, defaultValue as string);\n    }\n    if (value.b != null) {\n      return getBoolParam(this.node.rawAttrs, name, defaultValue as boolean);\n    }\n    if (value.shape != null) {\n      return getTensorShapeParam(\n          this.node.rawAttrs, name, defaultValue as number[]);\n    }\n    if (value.type != null) {\n      return getDtypeParam(this.node.rawAttrs, name, defaultValue as DataType);\n    }\n    if (value.list != null) {\n      if (value.list.i != null || value.list.f != null) {\n        return getNumericArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[]);\n      }\n      if (value.list.s != null) {\n        return getStringArrayParam(\n            this.node.rawAttrs, name, defaultValue as string[]);\n      }\n      if (value.list.shape != null) {\n        return getTensorShapeArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[][]);\n      }\n      if (value.list.b != null) {\n        return getBoolArrayParam(\n            this.node.rawAttrs, name, defaultValue as boolean[]);\n      }\n      if (value.list.type != null) {\n        return getDtypeArrayParam(\n            this.node.rawAttrs, name, defaultValue as DataType[]);\n      }\n    }\n\n    return defaultValue;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BiasAdd':\n        case 'AddV2':\n        case 'Add': {\n          return [tfOps.add(\n              (getParamValue('a', node, tensorMap, context) as Tensor),\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'AddN': {\n          return [tfOps.addN((\n              getParamValue('tensors', node, tensorMap, context) as Tensor[]))];\n        }\n        case 'FloorMod':\n        case 'Mod':\n          return [tfOps.mod(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'Mul':\n          return [tfOps.mul(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'RealDiv':\n        case 'Div': {\n          return [tfOps.div(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'DivNoNan': {\n          return [tfOps.divNoNan(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'FloorDiv': {\n          return [tfOps.floorDiv(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sub': {\n          return [tfOps.sub(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Minimum': {\n          return [tfOps.minimum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Maximum': {\n          return [tfOps.maximum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Pow': {\n          return [tfOps.pow(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'SquaredDifference': {\n          return [tfOps.squaredDifference(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'arithmetic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Abs':\n        case 'ComplexAbs':\n          return [tfOps.abs(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acos':\n          return [tfOps.acos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acosh':\n          return [tfOps.acosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asin':\n          return [tfOps.asin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asinh':\n          return [tfOps.asinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan':\n          return [tfOps.atan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan2':\n          return [tfOps.atan2(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('y', node, tensorMap, context) as Tensor)];\n        case 'Atanh':\n          return [tfOps.atanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Ceil':\n          return [tfOps.ceil(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Complex':\n          return [tfOps.complex(\n              getParamValue('real', node, tensorMap, context) as Tensor,\n              getParamValue('imag', node, tensorMap, context) as Tensor)];\n        case 'Cos':\n          return [tfOps.cos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Cosh':\n          return [tfOps.cosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Elu':\n          return [tfOps.elu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Erf':\n          return [tfOps.erf(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Exp':\n          return [tfOps.exp(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Expm1': {\n          return [tfOps.expm1(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Floor':\n          return [tfOps.floor(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log':\n          return [tfOps.log(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log1p': {\n          return [tfOps.log1p(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Imag':\n          return [tfOps.imag(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n\n        case 'Neg':\n          return [tfOps.neg(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Reciprocal': {\n          return [tfOps.reciprocal(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Real':\n          return [tfOps.real(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu':\n          return [tfOps.relu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Round': {\n          return [tfOps.round(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Selu':\n          return [tfOps.selu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sigmoid':\n          return [tfOps.sigmoid(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sin':\n          return [tfOps.sin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sign': {\n          return [tfOps.sign(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sinh': {\n          return [tfOps.sinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Softplus': {\n          return [tfOps.softplus(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sqrt': {\n          return [tfOps.sqrt(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Square': {\n          return [tfOps.square(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tanh': {\n          return [tfOps.tanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tan':\n          return [tfOps.tan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu6':\n        case 'ClipByValue':\n          return [tfOps.clipByValue(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('clipValueMin', node, tensorMap, context) as number,\n              getParamValue('clipValueMax', node, tensorMap, context) as\n                  number)];\n        case 'Rsqrt':\n          return [tfOps.rsqrt(\n              getTensor(node.inputNames[0], tensorMap, context))];\n        case 'Prod':\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('axes', node, tensorMap, context) as number[])];\n        case 'LeakyRelu':\n          return [tfOps.leakyRelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as number)];\n        case 'Prelu':\n          return [tfOps.prelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as Tensor)];\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'basic_math';\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n/**\n * This differs from util.assertShapesMatch in that it allows values of\n * negative one, an undefined size of a dimensinon, in a shape to match\n * anything.\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function assertShapesMatchAllowUndefinedSize(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  util.assert(\n      shapesEqualAllowUndefinedSize(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function shapesEqualAllowUndefinedSize(n1: number[], n2: number[]) {\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== -1 && n2[i] !== -1 && n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\nexport interface TensorWithState {\n  tensor?: Tensor;\n  written?: boolean;\n  read?: boolean;\n  cleared?: boolean;\n}\n/**\n * The TensorArray object keeps an array of Tensors.  It\n * allows reading from the array and writing to the array.\n */\nexport class TensorArray {\n  private tensors: TensorWithState[] = [];\n  private closed_ = false;\n  readonly idTensor: Tensor;\n  constructor(\n      readonly name: string, readonly dtype: DataType, private maxSize: number,\n      private elementShape: number[], readonly identicalElementShapes: boolean,\n      readonly dynamicSize: boolean, readonly clearAfterRead: boolean) {\n    this.idTensor = scalar(0);\n    keep(this.idTensor);\n  }\n\n  get id() {\n    return this.idTensor.id;\n  }\n\n  get closed() {\n    return this.closed_;\n  }\n\n  /**\n   * Dispose the tensors and idTensor and mark the TensoryArray as closed.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.tensor.id)) {\n        tensor.tensor.dispose();\n      }\n    });\n    this.tensors = [];\n    this.closed_ = true;\n    this.idTensor.dispose();\n  }\n\n  size(): number {\n    return this.tensors.length;\n  }\n\n  /**\n   * Read the value at location index in the TensorArray.\n   * @param index Number the index to read from.\n   */\n  read(index: number): Tensor {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || index >= this.size()) {\n      throw new Error(`Tried to read from index ${index}, but array size is: ${\n          this.size()}`);\n    }\n\n    const tensorWithState = this.tensors[index];\n    if (tensorWithState.cleared) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not read index ${\n              index} twice because it was cleared after a previous read ` +\n          `(perhaps try setting clear_after_read = false?).`);\n    }\n\n    if (this.clearAfterRead) {\n      tensorWithState.cleared = true;\n    }\n\n    tensorWithState.read = true;\n    return tensorWithState.tensor;\n  }\n\n  /**\n   * Helper method to read multiple tensors from the specified indices.\n   */\n  readMany(indices: number[]): Tensor[] {\n    return indices.map(index => this.read(index));\n  }\n\n  /**\n   * Write value into the index of the TensorArray.\n   * @param index number the index to write to.\n   * @param tensor\n   */\n  write(index: number, tensor: Tensor) {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || !this.dynamicSize && index >= this.maxSize) {\n      throw new Error(`Tried to write to index ${\n          index}, but array is not resizeable and size is: ${this.maxSize}`);\n    }\n\n    const t = this.tensors[index] || {};\n\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray ${\n          this.name}: Could not write to TensorArray index ${index},\n          because the value dtype is ${\n          tensor.dtype}, but TensorArray dtype is ${this.dtype}.`);\n    }\n\n    // Set the shape for the first time write to unknow shape tensor array\n    if (this.size() === 0 &&\n        (this.elementShape == null || this.elementShape.length === 0)) {\n      this.elementShape = tensor.shape;\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape,\n        `TensorArray ${this.name}: Could not write to TensorArray index ${\n            index}.`);\n\n    if (t.read) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been read.`);\n    }\n\n    if (t.written) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been written.`);\n    }\n\n    t.tensor = tensor;\n    keep(tensor);\n    t.written = true;\n\n    this.tensors[index] = t;\n  }\n\n  /**\n   * Helper method to write multiple tensors to the specified indices.\n   */\n  writeMany(indices: number[], tensors: Tensor[]) {\n    if (indices.length !== tensors.length) {\n      throw new Error(\n          `TensorArray ${this.name}: could not write multiple tensors,` +\n          `because the index size: ${\n              indices.length} is not the same as tensors size: ${\n              tensors.length}.`);\n    }\n\n    indices.forEach((i, index) => this.write(i, tensors[index]));\n  }\n\n  /**\n   * Return selected values in the TensorArray as a packed Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param [indices] number[] Optional. Taking values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size(). If not specified returns\n   *    all tensors in the original order.\n   * @param [dtype]\n   */\n  gather(indices?: number[], dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but gather requested dtype ${dtype}`);\n    }\n\n    if (!indices) {\n      indices = [];\n      for (let i = 0; i < this.size(); i++) {\n        indices.push(i);\n      }\n    } else {\n      indices = indices.slice(0, this.size());\n    }\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    // Read all the PersistentTensors into a vector to keep track of\n    // their memory.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape, 'TensorArray shape mismatch: ');\n\n    return stack(tensors, 0);\n  }\n\n  /**\n   * Return the values in the TensorArray as a concatenated Tensor.\n   */\n  concat(dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but concat requested dtype ${dtype}`);\n    }\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    const indices = [];\n    for (let i = 0; i < this.size(); i++) {\n      indices.push(i);\n    }\n    // Collect all the tensors from the tensors array.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape,\n        `TensorArray shape mismatch: tensor array shape (${\n            this.elementShape}) vs first tensor shape (${tensors[0].shape})`);\n\n    return concat(tensors, 0);\n  }\n\n  /**\n   * Scatter the values of a Tensor in specific indices of a TensorArray.\n   * @param indices nummber[] values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size().\n   * @param tensor Tensor input tensor.\n   */\n  scatter(indices: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n\n    if (indices.length !== tensor.shape[0]) {\n      throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n          indices.length} vs. ${tensor.shape[0]}`);\n    }\n\n    const maxIndex = Math.max(...indices);\n\n    if (!this.dynamicSize && maxIndex >= this.maxSize) {\n      throw new Error(\n          `Max index must be < array size (${maxIndex}  vs. ${this.maxSize})`);\n    }\n\n    this.writeMany(indices, unstack(tensor, 0));\n  }\n\n  /**\n   * Split the values of a Tensor into the TensorArray.\n   * @param length number[] with the lengths to use when splitting value along\n   *    its first dimension.\n   * @param tensor Tensor, the tensor to split.\n   */\n  split(length: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n    let totalLength = 0;\n    const cumulativeLengths = length.map(len => {\n      totalLength += len;\n      return totalLength;\n    });\n\n    if (totalLength !== tensor.shape[0]) {\n      throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n    }\n\n    if (!this.dynamicSize && length.length !== this.maxSize) {\n      throw new Error(\n          `TensorArray's size is not equal to the size of lengths (${\n              this.maxSize} vs. ${length.length}), ` +\n          'and the TensorArray is not marked as dynamically resizeable');\n    }\n\n    const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n    const tensors: Tensor[] = [];\n    tidy(() => {\n      tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n      for (let i = 0; i < length.length; ++i) {\n        const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n        const indices = [0, previousLength, 0];\n        const sizes = [1, length[i], elementPerRow];\n        tensors[i] = reshape(slice(tensor, indices, sizes), this.elementShape);\n      }\n      return tensors;\n    });\n    const indices = [];\n    for (let i = 0; i < length.length; i++) {\n      indices[i] = i;\n    }\n    this.writeMany(indices, tensors);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\n/**\n * TensorList stores a container of `tf.Tensor` objects, which are accessible\n * via tensors field.\n *\n * In order to get a copy of the underlying list, use the copy method:\n * ```\n *    TensorList b = a.copy();\n *    b.tensors().pushBack(t);  // This does not modify a.tensors().\n * ```\n *\n * Note that this is not a deep copy: the memory locations of the underlying\n * tensors will still point to the same locations of the corresponding tensors\n * in the original.\n */\n\nexport class TensorList {\n  readonly idTensor: Tensor;\n  maxNumElements: number;\n\n  get id() {\n    return this.idTensor.id;\n  }\n  /**\n   *\n   * @param tensors list of tensors\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param maxNumElements The maximum allowed size of `tensors`. Defaults to -1\n   *   meaning that the size of `tensors` is unbounded.\n   */\n  constructor(\n      readonly tensors: Tensor[], readonly elementShape: number[],\n      readonly elementDtype: DataType, maxNumElements = -1) {\n    if (tensors != null) {\n      tensors.forEach(tensor => {\n        if (elementDtype !== tensor.dtype) {\n          throw new Error(`Invalid data types; op elements ${\n              elementDtype}, but list elements ${tensor.dtype}`);\n        }\n        assertShapesMatchAllowUndefinedSize(\n            elementShape, tensor.shape, 'TensorList shape mismatch: ');\n\n        keep(tensor);\n      });\n    }\n    this.idTensor = scalar(0);\n    this.maxNumElements = maxNumElements;\n    keep(this.idTensor);\n  }\n\n  /**\n   * Get a new TensorList containing a copy of the underlying tensor container.\n   */\n  copy(): TensorList {\n    return new TensorList(\n        [...this.tensors], this.elementShape, this.elementDtype);\n  }\n\n  /**\n   * Dispose the tensors and idTensor and clear the tensor list.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.id)) {\n        tensor.dispose();\n      }\n    });\n    this.tensors.length = 0;\n    this.idTensor.dispose();\n  }\n  /**\n   * The size of the tensors in the tensor list.\n   */\n  size() {\n    return this.tensors.length;\n  }\n\n  /**\n   * Return a tensor that stacks a list of rank-R tf.Tensors into one rank-(R+1)\n   * tf.Tensor.\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param numElements the number of elements to stack\n   */\n  stack(elementShape: number[], elementDtype: DataType, numElements = -1):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (numElements !== -1 && this.tensors.length !== numElements) {\n      throw new Error(`Operation expected a list with ${\n          numElements} elements but got a list with ${\n          this.tensors.length} elements.`);\n    }\n    assertShapesMatchAllowUndefinedSize(\n        elementShape, this.elementShape, 'TensorList shape mismatch: ');\n    return tidy(() => {\n      const reshapedTensors =\n          this.tensors.map(tensor => reshape(tensor, elementShape));\n      return stack(reshapedTensors, 0);\n    });\n  }\n\n  /**\n   * Pop a tensor from the end of the list.\n   * @param elementShape shape of the tensor\n   * @param elementDtype data type of the tensor\n   */\n  popBack(elementShape: number[], elementDtype: DataType): Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (this.size() === 0) {\n      throw new Error('Trying to pop from an empty list.');\n    }\n\n    const tensor = this.tensors.pop();\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, elementShape, 'TensorList shape mismatch: ');\n    return reshape(tensor, elementShape);\n  }\n\n  /**\n   * Push a tensor to the end of the list.\n   * @param tensor Tensor to be pushed.\n   */\n  pushBack(tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, this.elementShape, 'TensorList shape mismatch: ');\n\n    if (this.maxNumElements === this.size()) {\n      throw new Error(`Trying to push element into a full list.`);\n    }\n    keep(tensor);\n    this.tensors.push(tensor);\n  }\n\n  /**\n   * Update the size of the list.\n   * @param size the new size of the list.\n   */\n  resize(size: number) {\n    if (size < 0) {\n      throw new Error(\n          `TensorListResize expects size to be non-negative. Got: ${size}`);\n    }\n\n    if (this.maxNumElements !== -1 && size > this.maxNumElements) {\n      throw new Error(`TensorListResize input size ${\n          size} is greater maxNumElement ${this.maxNumElements}.`);\n    }\n    this.tensors.length = size;\n  }\n\n  /**\n   * Retrieve the element at the provided index\n   * @param elementShape shape of the tensor\n   * @param elementDtype dtype of the tensor\n   * @param elementIndex index of the tensor\n   */\n  getItem(elementIndex: number, elementShape: number[], elementDtype: DataType):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (elementIndex < 0 || elementIndex > this.tensors.length) {\n      throw new Error(`Trying to access element ${\n          elementIndex} in a list with ${this.tensors.length} elements.`);\n    }\n\n    if (this.tensors[elementIndex] == null) {\n      throw new Error(`element at index ${elementIndex} is null.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.tensors[elementIndex].shape, elementShape,\n        'TensorList shape mismatch: ');\n\n    return this.tensors[elementIndex];\n  }\n\n  /**\n   * Set the tensor at the index\n   * @param elementIndex index of the tensor\n   * @param tensor the tensor to be inserted into the list\n   */\n  setItem(elementIndex: number, tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (elementIndex < 0 ||\n        this.maxNumElements !== -1 && elementIndex >= this.maxNumElements) {\n      throw new Error(`Trying to set element ${\n          elementIndex} in a list with max ${this.maxNumElements} elements.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape, 'TensorList shape mismatch: ');\n    keep(tensor);\n    this.tensors[elementIndex] = tensor;\n  }\n\n  /**\n   * Return selected values in the TensorList as a stacked Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param indices indices of tensors to gather\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  gather(indices: number[], elementDtype: DataType, elementShape: number[]):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    // When indices is greater than the size of the list, indices beyond the\n    // size of the list are ignored.\n    indices = indices.slice(0, this.size());\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = indices.map(i => reshape(this.tensors[i], elementShape));\n      return stack(tensors, 0);\n    });\n  }\n\n  /**\n   * Return the values in the TensorList as a concatenated Tensor.\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  concat(elementDtype: DataType, elementShape: number[]): Tensor {\n    if (!!elementDtype && elementDtype !== this.elementDtype) {\n      throw new Error(`TensorList dtype is ${\n          this.elementDtype} but concat requested dtype ${elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = this.tensors.map(t => reshape(t, elementShape));\n      return concat(tensors, 0);\n    });\n  }\n}\n\n/**\n * Creates a TensorList which, when stacked, has the value of tensor.\n * @param tensor from tensor\n * @param elementShape output tensor element shape\n */\nexport function fromTensor(\n    tensor: Tensor, elementShape: number[], elementDtype: DataType) {\n  const dtype = tensor.dtype;\n  if (tensor.shape.length < 1) {\n    throw new Error(\n        `Tensor must be at least a vector, but saw shape: ${tensor.shape}`);\n  }\n  if (tensor.dtype !== elementDtype) {\n    throw new Error(`Invalid data types; op elements ${\n        tensor.dtype}, but list elements ${elementDtype}`);\n  }\n  const outputShape = tensor.shape.slice(1);\n  assertShapesMatchAllowUndefinedSize(\n      outputShape, elementShape, 'TensorList shape mismatch: ');\n\n  const tensorList: Tensor[] = unstack(tensor);\n  return new TensorList(tensorList, elementShape, dtype);\n}\n\n/**\n * Return a TensorList of the given size with empty elements.\n * @param elementShape the shape of the future elements of the list\n * @param elementDtype the desired type of elements in the list\n * @param numElements the number of elements to reserve\n */\nexport function reserve(\n    elementShape: number[], elementDtype: DataType, numElements: number) {\n  return new TensorList([], elementShape, elementDtype, numElements);\n}\n\n/**\n * Put tensors at specific indices of a stacked tensor into a TensorList.\n * @param indices list of indices on how to scatter the tensor.\n * @param tensor input tensor.\n * @param elementShape the shape of the future elements of the list\n * @param numElements the number of elements to scatter\n */\nexport function scatter(\n    tensor: Tensor, indices: number[], elementShape: number[],\n    numElements?: number): TensorList {\n  if (indices.length !== tensor.shape[0]) {\n    throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n        indices.length} vs. ${tensor.shape[0]}`);\n  }\n\n  const maxIndex = Math.max(...indices);\n\n  if (numElements != null && numElements !== -1 && maxIndex >= numElements) {\n    throw new Error(\n        `Max index must be < array size (${maxIndex}  vs. ${numElements})`);\n  }\n\n  const list = new TensorList([], elementShape, tensor.dtype, numElements);\n  const tensors = unstack(tensor, 0);\n  indices.forEach((value, index) => {\n    list.setItem(value, tensors[index]);\n  });\n  return list;\n}\n\n/**\n * Split the values of a Tensor into a TensorList.\n * @param length the lengths to use when splitting value along\n *    its first dimension.\n * @param tensor the tensor to split.\n * @param elementShape the shape of the future elements of the list\n */\nexport function split(\n    tensor: Tensor, length: number[], elementShape: number[]) {\n  let totalLength = 0;\n  const cumulativeLengths = length.map(len => {\n    totalLength += len;\n    return totalLength;\n  });\n\n  if (totalLength !== tensor.shape[0]) {\n    throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n  }\n\n  const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n  const tensors: Tensor[] = tidy(() => {\n    const tensors = [];\n    tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n    for (let i = 0; i < length.length; ++i) {\n      const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n      const indices = [0, previousLength, 0];\n      const sizes = [1, length[i], elementPerRow];\n      tensors[i] = reshape(slice(tensor, indices, sizes), elementShape);\n    }\n    tensor.dispose();\n    return tensors;\n  });\n\n  const list = new TensorList([], elementShape, tensor.dtype, length.length);\n\n  for (let i = 0; i < tensors.length; i++) {\n    list.setItem(i, tensors[i]);\n  }\n  return list;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, scalar, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {TensorArray} from '../../executor/tensor_array';\nimport {fromTensor, reserve, scatter, split} from '../../executor/tensor_list';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'If':\n    case 'StatelessIf': {\n      const thenFunc =\n          getParamValue('thenBranch', node, tensorMap, context) as string;\n      const elseFunc =\n          getParamValue('elseBranch', node, tensorMap, context) as string;\n      const cond = getParamValue('cond', node, tensorMap, context) as Tensor;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n      const condValue = await cond.data();\n      if (condValue[0]) {\n        return context.functionMap[thenFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      } else {\n        return context.functionMap[elseFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      }\n    }\n    case 'While':\n    case 'StatelessWhile': {\n      const bodyFunc =\n          getParamValue('body', node, tensorMap, context) as string;\n      const condFunc =\n          getParamValue('cond', node, tensorMap, context) as string;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n\n      // Calculate the condition of the loop\n      const condResult =\n          (await context.functionMap[condFunc].executeFunctionAsync(\n              args, context.tensorArrayMap, context.tensorListMap));\n      const argIds = args.map(tensor => tensor.id);\n      let condValue = await condResult[0].data();\n      // Dispose the intermediate tensors for condition function\n      condResult.forEach(tensor => {\n        if (!tensor.kept && argIds.indexOf(tensor.id) === -1) {\n          tensor.dispose();\n        }\n      });\n\n      let result: Tensor[] = args;\n\n      while (condValue[0]) {\n        // Record the previous result for intermediate tensor tracking\n        const origResult = result;\n        // Execution the body of the loop\n        result = await context.functionMap[bodyFunc].executeFunctionAsync(\n            result, context.tensorArrayMap, context.tensorListMap);\n        const resultIds = result.map(tensor => tensor.id);\n\n        // Dispose the intermediate tensor for body function that is not global\n        // kept, not input/output of the body function\n        origResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n\n        // Recalcuate the condition of the loop using the latest results.\n        const condResult =\n            (await context.functionMap[condFunc].executeFunctionAsync(\n                result, context.tensorArrayMap, context.tensorListMap));\n        condValue = await condResult[0].data();\n        // Dispose the intermediate tensors for condition function\n        condResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n      }\n      return result;\n    }\n    case 'LoopCond': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      return [cloneTensor(pred)];\n    }\n    case 'Switch': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      let data = getParamValue('data', node, tensorMap, context) as Tensor;\n      if (!data.kept) {\n        data = cloneTensor(data);\n      }\n      // Outputs nodes :0 => false, :1 => true\n      return (await pred.data())[0] ? [undefined, data] : [data, undefined];\n    }\n    case 'Merge': {\n      const inputName = node.inputNames.find(\n          name => getTensor(name, tensorMap, context) !== undefined);\n      if (inputName) {\n        const data = getTensor(inputName, tensorMap, context);\n        return [cloneTensor(data)];\n      }\n      return undefined;\n    }\n    case 'Enter': {\n      const frameId =\n          getParamValue('frameName', node, tensorMap, context) as string;\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.enterFrame(frameId);\n      return [cloneTensor(data)];\n    }\n    case 'Exit': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.exitFrame();\n      return [cloneTensor(data)];\n    }\n    case 'NextIteration': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.nextIteration();\n      return [cloneTensor(data)];\n    }\n    case 'TensorArrayV3': {\n      const size = getParamValue('size', node, tensorMap, context) as number;\n      const dtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const dynamicSize =\n          getParamValue('dynamicSize', node, tensorMap, context) as boolean;\n      const clearAfterRead =\n          getParamValue('clearAfterRead', node, tensorMap, context) as boolean;\n      const identicalElementShapes =\n          getParamValue('identicalElementShapes', node, tensorMap, context) as\n          boolean;\n      const name = getParamValue('name', node, tensorMap, context) as string;\n      const tensorArray = new TensorArray(\n          name, dtype, size, elementShape, identicalElementShapes, dynamicSize,\n          clearAfterRead);\n      context.addTensorArray(tensorArray);\n      return [tensorArray.idTensor, scalar(1.0)];\n    }\n    case 'TensorArrayWriteV3': {\n      const id =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const writeTensorArray = context.getTensorArray(id.id);\n      writeTensorArray.write(index, writeTensor);\n      return [writeTensorArray.idTensor];\n    }\n    case 'TensorArrayReadV3': {\n      const readId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const readTensorArray = context.getTensorArray(readId.id);\n      return [readTensorArray.read(readIndex)];\n    }\n    case 'TensorArrayGatherV3': {\n      const gatherId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const gatherDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const gatherTensorArray = context.getTensorArray(gatherId.id);\n      return [gatherTensorArray.gather(gatherIndices, gatherDtype)];\n    }\n    case 'TensorArrayScatterV3': {\n      const scatterId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const scatterTensorArray = context.getTensorArray(scatterId.id);\n      scatterTensorArray.scatter(scatterIndices, scatterTensor);\n      return [scatterTensorArray.idTensor];\n    }\n    case 'TensorArrayConcatV3': {\n      const concatId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const concatTensorArray = context.getTensorArray(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      return [concatTensorArray.concat(concatDtype)];\n    }\n    case 'TensorArraySplitV3': {\n      const splitId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n      const splitTensorArray = context.getTensorArray(splitId.id);\n      splitTensorArray.split(lengths, splitTensor);\n      return [splitTensorArray.idTensor];\n    }\n    case 'TensorArraySizeV3': {\n      const sizeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const sizeTensorArray = context.getTensorArray(sizeId.id);\n      return [scalar(sizeTensorArray.size(), 'int32')];\n    }\n    case 'TensorArrayCloseV3': {\n      const closeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const closeTensorArray = context.getTensorArray(closeId.id);\n      closeTensorArray.clearAndClose();\n      return [closeTensorArray.idTensor];\n    }\n    case 'TensorListSetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.setItem(index, writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.getItem(readIndex, elementShape, elementDType)];\n    }\n    case 'TensorListScatterV2':\n    case 'TensorListScatter': {\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList =\n          scatter(scatterTensor, scatterIndices, elementShape, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListReserve': {\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = reserve(elementShape, elementDtype, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGather': {\n      const gatherId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(gatherId.id);\n      return [tensorList.gather(gatherIndices, elementDtype, elementShape)];\n    }\n    case 'TensorListStack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.stack(elementShape, elementDtype, numElements)];\n    }\n    case 'TensorListFromTensor': {\n      const tensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = fromTensor(tensor, elementShape, elementDtype);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListConcat': {\n      const concatId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      return [tensorList.concat(concatDtype, elementShape)];\n    }\n    case 'TensorListPushBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.pushBack(writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListPopBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.popBack(elementShape, elementDType)];\n    }\n    case 'TensorListSplit': {\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n\n      const tensorList = split(splitTensor, lengths, elementShape);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'control';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rank, Tensor, Tensor3D, Tensor4D, Tensor5D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getPadding, getParamValue} from './utils';\n\nfunction fusedConvAndDepthWiseParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const [extraOp, activationFunc] =\n      (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n  const isBiasAdd = extraOp === 'biasadd';\n  const isPrelu = activationFunc === 'prelu';\n  const isBatchNorm = extraOp === 'fusedbatchnorm';\n\n  const numArgs =\n      (getParamValue('numArgs', node, tensorMap, context) as number);\n  if (isBiasAdd) {\n    if (isPrelu && numArgs !== 2) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu ' +\n          'must have two extra arguments: bias and alpha.');\n    }\n    if (!isPrelu && numArgs !== 1) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd must have ' +\n          'one extra argument: bias.');\n    }\n  }\n  if (isBatchNorm) {\n    throw new Error(\n        'FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.');\n  }\n  const stride = getParamValue('strides', node, tensorMap, context) as number[];\n  const pad = getPadding(node, tensorMap, context);\n  const dataFormat =\n      (getParamValue('dataFormat', node, tensorMap, context) as string)\n          .toUpperCase();\n  const dilations =\n      getParamValue('dilations', node, tensorMap, context) as number[];\n  const [biasArg, preluArg] =\n      getParamValue('args', node, tensorMap, context) as Tensor[];\n\n  return {\n    stride,\n    pad,\n    dataFormat,\n    dilations,\n    biasArg,\n    preluArg,\n    activationFunc\n  };\n}\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Conv1D': {\n          const stride =\n              getParamValue('stride', node, tensorMap, context) as number;\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilation =\n              getParamValue('dilation', node, tensorMap, context) as number;\n          return [tfOps.conv1d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              stride, pad as 'valid' | 'same', dataFormat as 'NWC' | 'NCW',\n              dilation)];\n        }\n        case 'Conv2D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case '_FusedConv2D': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.conv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n\n        case 'FusedDepthwiseConv2dNative': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.depthwiseConv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n        case 'Conv2DBackpropInput':\n        case 'Conv2dTranspose': {\n          const shape = getParamValue(\n                            'outputShape', node, tensorMap,\n                            context) as [number, number, number] |\n              [number, number, number, number];\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          return [tfOps.conv2dTranspose(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              shape, [stride[1], stride[2]], pad as 'valid' | 'same')];\n        }\n        case 'DepthwiseConv2dNative':\n        case 'DepthwiseConv2d': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n\n          return [tfOps.depthwiseConv2d(\n              getParamValue('input', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case 'Conv3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv3d(\n              getParamValue('x', node, tensorMap, context) as Tensor4D |\n                  Tensor<Rank.R5>,\n              getParamValue('filter', node, tensorMap, context) as\n                  Tensor<Rank.R5>,\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same',\n              dataFormat as 'NDHWC' | 'NCDHW',\n              [dilations[1], dilations[2], dilations[3]])];\n        }\n        case 'AvgPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPoolWithArgmax': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n          const includeBatchInIndex =\n              getParamValue('includeBatchInIndex', node, tensorMap, context) as\n              boolean;\n          const {result, indexes} = tfOps.maxPoolWithArgmax(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same', includeBatchInIndex);\n          return [result, indexes];\n        }\n        case 'AvgPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'MaxPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'Dilation2D': {\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n\n          // strides: [1, stride_height, stride_width, 1].\n          const strideHeight = strides[1];\n          const strideWidth = strides[2];\n\n          // dilations: [1, dilation_height, dilation_width, 1].\n          const dilationHeight = dilations[1];\n          const dilationWidth = dilations[2];\n\n          return [tfOps.dilation2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              [strideHeight, strideWidth], pad as 'valid' | 'same',\n              [dilationHeight, dilationWidth], 'NHWC' /* dataFormat */)];\n        }\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'convolution';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor, Tensor1D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Fill': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const dtype =\n              getParamValue('dtype', node, tensorMap, context) as DataType;\n          const value =\n              getParamValue('value', node, tensorMap, context) as number;\n          return [tfOps.fill(shape, value, dtype)];\n        }\n        case 'LinSpace': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const num = getParamValue('num', node, tensorMap, context) as number;\n          return [tfOps.linspace(start, stop, num)];\n        }\n        case 'Multinomial': {\n          const logits =\n              getParamValue('logits', node, tensorMap, context) as Tensor1D;\n          const numSamples =\n              getParamValue('numSamples', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.multinomial(logits, numSamples, seed)];\n        }\n        case 'OneHot': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          const depth =\n              getParamValue('depth', node, tensorMap, context) as number;\n          const onValue =\n              getParamValue('onValue', node, tensorMap, context) as number;\n          const offValue =\n              getParamValue('offValue', node, tensorMap, context) as number;\n          return [tfOps.oneHot(indices, depth, onValue, offValue)];\n        }\n        case 'Ones': {\n          return [tfOps.ones(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'OnesLike': {\n          return [tfOps.onesLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'RandomUniform': {\n          return [tfOps.randomUniform(\n              // tslint:disable-next-line:no-any\n              getParamValue('shape', node, tensorMap, context) as any,\n              getParamValue('minval', node, tensorMap, context) as number,\n              getParamValue('maxval', node, tensorMap, context) as number,\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'Range': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const step =\n              getParamValue('step', node, tensorMap, context) as number;\n          return [tfOps.range(\n              start, stop, step,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32')];\n        }\n        case 'TruncatedNormal': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const mean =\n              getParamValue('mean', node, tensorMap, context) as number;\n          const stdDev =\n              getParamValue('stdDev', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.truncatedNormal(\n              shape, mean, stdDev,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32',\n              seed)];\n        }\n        case 'Zeros': {\n          return [tfOps.zeros(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'ZerosLike': {\n          return [tfOps.zerosLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'creation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nfunction nmsParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const boxes = getParamValue('boxes', node, tensorMap, context) as Tensor;\n  const scores = getParamValue('scores', node, tensorMap, context) as Tensor;\n  const maxOutputSize =\n      getParamValue('maxOutputSize', node, tensorMap, context) as number;\n  const iouThreshold =\n      getParamValue('iouThreshold', node, tensorMap, context) as number;\n  const scoreThreshold =\n      getParamValue('scoreThreshold', node, tensorMap, context) as number;\n  const softNmsSigma =\n      getParamValue('softNmsSigma', node, tensorMap, context) as number;\n\n  return {\n    boxes,\n    scores,\n    maxOutputSize,\n    iouThreshold,\n    scoreThreshold,\n    softNmsSigma\n  };\n}\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'NonMaxSuppressionV5': {\n      const {\n        boxes,\n        scores,\n        maxOutputSize,\n        iouThreshold,\n        scoreThreshold,\n        softNmsSigma\n      } = nmsParams(node, tensorMap, context);\n\n      const result = await tfOps.image.nonMaxSuppressionWithScoreAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, softNmsSigma);\n\n      return [result.selectedIndices, result.selectedScores];\n    }\n    case 'NonMaxSuppressionV4': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      const padToMaxOutputSize =\n          getParamValue('padToMaxOutputSize', node, tensorMap, context) as\n          boolean;\n\n      const result = await tfOps.image.nonMaxSuppressionPaddedAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, padToMaxOutputSize);\n\n      return [result.selectedIndices, result.validOutputs];\n    }\n    case 'NonMaxSuppressionV3':\n    case 'NonMaxSuppressionV2': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      return [await tfOps.image.nonMaxSuppressionAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold)];\n    }\n    case 'Where': {\n      const condition = tfOps.cast(\n          (getParamValue('condition', node, tensorMap, context) as Tensor),\n          'bool');\n      const result = [await tfOps.whereAsync(condition)];\n      condition.dispose();\n      return result;\n    }\n    case 'ListDiff': {\n      return tfOps.setdiff1dAsync(\n          getParamValue('x', node, tensorMap, context) as Tensor,\n          getParamValue('y', node, tensorMap, context) as Tensor);\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'dynamic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'TopKV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const k = getParamValue('k', node, tensorMap, context) as number;\n              const sorted =\n                  getParamValue('sorted', node, tensorMap, context) as boolean;\n              const result = tfOps.topk(x, k, sorted);\n              return [result.values, result.indices];\n            }\n            case 'Unique': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const result = tfOps.unique(x);\n              return [result.values, result.indices];\n            }\n            case 'UniqueV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const axis =\n                  getParamValue('axis', node, tensorMap, context) as number;\n              const result = tfOps.unique(x, axis);\n              return [result.values, result.indices];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'evaluation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Const': {\n          return tensorMap[node.name];\n        }\n        case 'PlaceholderWithDefault':\n          const def =\n              getParamValue('default', node, tensorMap, context) as Tensor;\n          return [getTensor(node.name, tensorMap, context) || def];\n        case 'Placeholder':\n          return [getTensor(node.name, tensorMap, context)];\n        case 'Identity':\n        case 'StopGradient':\n        case 'FakeQuantWithMinMaxVars': {  // This op is currently ignored.\n          const data = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [cloneTensor(data)];\n        }\n        case 'IdentityN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => cloneTensor(t));\n        case 'Snapshot':\n          const snapshot =\n              (getParamValue('x', node, tensorMap, context) as Tensor);\n          return [cloneTensor(snapshot)];\n        case 'Shape':\n          return [tfOps.tensor1d(\n              (getParamValue('x', node, tensorMap, context) as Tensor).shape,\n              'int32')];\n        case 'ShapeN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => tfOps.tensor1d(t.shape));\n        case 'Size':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).size,\n              'int32')];\n        case 'Rank':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).rank,\n              'int32')];\n        case 'NoOp':\n          return [tfOps.scalar(1)];\n        case 'Print':\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const data =\n              getParamValue('data', node, tensorMap, context) as Tensor[];\n          const message =\n              getParamValue('message', node, tensorMap, context) as string;\n          const summarize =\n              getParamValue('summarize', node, tensorMap, context) as number;\n          console.warn(\n              'The graph has a tf.print() operation,' +\n              'usually used for debugging, which slows down performance.');\n          console.log(message);\n          for (let i = 0; i < data.length; i++) {\n            console.log(Array.prototype.slice.call(data[i].dataSync())\n                            .slice(0, summarize));\n          }\n          return [input];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'graph';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ResizeBilinear': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeBilinear(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'ResizeNearestNeighbor': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeNearestNeighbor(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'CropAndResize': {\n          const image =\n              getParamValue('image', node, tensorMap, context) as Tensor;\n          const boxes =\n              getParamValue('boxes', node, tensorMap, context) as Tensor;\n          const boxInd =\n              getParamValue('boxInd', node, tensorMap, context) as Tensor;\n          const cropSize =\n              getParamValue('cropSize', node, tensorMap, context) as number[];\n          const method =\n              getParamValue('method', node, tensorMap, context) as string;\n          const extrapolationValue =\n              getParamValue('extrapolationValue', node, tensorMap, context) as\n              number;\n          return [tfOps.image.cropAndResize(\n              image as Tensor4D, boxes as Tensor2D, boxInd as Tensor1D,\n              cropSize as [number, number], method as 'bilinear' | 'nearest',\n              extrapolationValue)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'image';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Equal': {\n          return [tfOps.equal(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'NotEqual': {\n          return [tfOps.notEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Greater': {\n          return [tfOps.greater(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'GreaterEqual': {\n          return [tfOps.greaterEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Less': {\n          return [tfOps.less(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LessEqual': {\n          return [tfOps.lessEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalAnd': {\n          return [tfOps.logicalAnd(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalNot': {\n          return [tfOps.logicalNot(\n              getParamValue('a', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalOr': {\n          return [tfOps.logicalOr(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Select':\n        case 'SelectV2': {\n          return [tfOps.where(\n              getParamValue('condition', node, tensorMap, context) as Tensor,\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'logical';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BatchMatMul':\n        case 'BatchMatMulV2':\n        case 'MatMul':\n          return [tfOps.matMul(\n              getParamValue('a', node, tensorMap, context) as Tensor2D,\n              getParamValue('b', node, tensorMap, context) as Tensor2D,\n              getParamValue('transposeA', node, tensorMap, context) as boolean,\n              getParamValue('transposeB', node, tensorMap, context) as\n                  boolean)];\n\n        case 'Transpose':\n          return [tfOps.transpose(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('perm', node, tensorMap, context) as number[])];\n\n        case '_FusedMatMul':\n          const [extraOp, activationFunc] =\n              (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n          const isBiasAdd = extraOp === 'biasadd';\n          const isPrelu = activationFunc === 'prelu';\n\n          const numArgs =\n              (getParamValue('numArgs', node, tensorMap, context) as number);\n          if (isBiasAdd) {\n            if (isPrelu && numArgs !== 2) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd and Prelu must have two ' +\n                  'extra arguments: bias and alpha.');\n            }\n            if (!isPrelu && numArgs !== 1) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd must have one extra argument: bias.');\n            }\n          }\n          const [biasArg, preluArg] =\n              getParamValue('args', node, tensorMap, context) as Tensor[];\n          return [tfOps.fused.matMul({\n            a: getParamValue('a', node, tensorMap, context) as Tensor2D,\n            b: getParamValue('b', node, tensorMap, context) as Tensor2D,\n            transposeA: getParamValue('transposeA', node, tensorMap, context) as\n                boolean,\n            transposeB: getParamValue('transposeB', node, tensorMap, context) as\n                boolean,\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'matrices';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'FusedBatchNorm':\n        case 'FusedBatchNormV2': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'FusedBatchNormV3': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'LRN': {\n          return [tfOps.localResponseNormalization(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('radius', node, tensorMap, context) as number,\n              getParamValue('bias', node, tensorMap, context) as number,\n              getParamValue('alpha', node, tensorMap, context) as number,\n              getParamValue('beta', node, tensorMap, context) as number)];\n        }\n        case 'Softmax': {\n          return [tfOps.softmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogSoftmax': {\n          return [tfOps.logSoftmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'SparseToDense': {\n          return [tfOps.sparseToDense(\n              getParamValue('sparseIndices', node, tensorMap, context) as\n                  Tensor,\n              getParamValue('outputShape', node, tensorMap, context) as Tensor,\n              getParamValue('sparseValues', node, tensorMap, context) as\n                  number[],\n              getParamValue('defaultValue', node, tensorMap, context) as\n                  Scalar)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'normalization';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Max': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.max(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Mean': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.mean(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Min': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.min(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Sum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.sum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'All': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.all(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Any': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.any(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'ArgMax': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMax(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'ArgMin': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMin(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Prod': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Cumsum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const exclusive =\n              getParamValue('exclusive', node, tensorMap, context) as boolean;\n          const reverse =\n              getParamValue('reverse', node, tensorMap, context) as boolean;\n          return [tfOps.cumsum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              exclusive, reverse)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'reduction';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ConcatV2':\n        case 'Concat': {\n          const n = getParamValue('n', node, tensorMap, context) as number;\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          let inputs =\n              getParamValue('tensors', node, tensorMap, context) as Tensor[];\n          inputs = inputs.slice(0, n);\n          return [tfOps.concat(inputs, axis)];\n        }\n        case 'GatherV2':\n        case 'Gather': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          return [tfOps.gather(input, tfOps.cast(indices, 'int32'), axis)];\n        }\n        case 'ReverseV2':\n        case 'Reverse': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [tfOps.reverse(input, axis)];\n        }\n        case 'Slice': {\n          // tslint:disable-next-line:no-any\n          const begin = getParamValue('begin', node, tensorMap, context) as any;\n          // tslint:disable-next-line:no-any\n          const size = getParamValue('size', node, tensorMap, context) as any;\n          return [tfOps.slice(\n              getParamValue('x', node, tensorMap, context) as Tensor, begin,\n              size)];\n        }\n        case 'StridedSlice': {\n          const begin =\n              getParamValue('begin', node, tensorMap, context) as number[];\n          const end =\n              getParamValue('end', node, tensorMap, context) as number[];\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const beginMask =\n              getParamValue('beginMask', node, tensorMap, context) as number;\n          const endMask =\n              getParamValue('endMask', node, tensorMap, context) as number;\n          const ellipsisMask =\n              getParamValue('ellipsisMask', node, tensorMap, context) as number;\n          const newAxisMask =\n              getParamValue('newAxisMask', node, tensorMap, context) as number;\n          const shrinkAxisMask =\n              getParamValue('shrinkAxisMask', node, tensorMap, context) as\n              number;\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return [tfOps.stridedSlice(\n              tensor, begin, end, strides, beginMask, endMask, ellipsisMask,\n              newAxisMask, shrinkAxisMask)];\n        }\n        case 'Pack': {\n          return tidy(() => {\n            const axis =\n                getParamValue('axis', node, tensorMap, context) as number;\n            const tensors =\n                getParamValue('tensors', node, tensorMap, context) as Tensor[];\n            // Reshape the tensors to the first tensor's shape if they don't\n            // match.\n            const shape = tensors[0].shape;\n            const squeezedShape = tfOps.squeeze(tensors[0]).shape;\n            const mapped = tensors.map(tensor => {\n              const sameShape = util.arraysEqual(tensor.shape, shape);\n              if (!sameShape &&\n                  !util.arraysEqual(\n                      tfOps.squeeze(tensor).shape, squeezedShape)) {\n                throw new Error('the input tensors shape does not match');\n              }\n              return sameShape ? tensor : tfOps.reshape(tensor, shape);\n            });\n            return [tfOps.stack(mapped, axis)];\n          });\n        }\n        case 'Unpack': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const tensor =\n              getParamValue('tensor', node, tensorMap, context) as Tensor;\n          return tfOps.unstack(tensor, axis);\n        }\n        case 'Tile': {\n          const reps =\n              getParamValue('reps', node, tensorMap, context) as number[];\n          return [tfOps.tile(\n              getParamValue('x', node, tensorMap, context) as Tensor, reps)];\n        }\n        case 'Split':\n        case 'SplitV': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const numOrSizeSplits =\n              getParamValue('numOrSizeSplits', node, tensorMap, context) as\n                  number |\n              number[];\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return tfOps.split(tensor, numOrSizeSplits, axis);\n        }\n        case 'ScatterNd': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          const values =\n              getParamValue('values', node, tensorMap, context) as Tensor;\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          return [tfOps.scatterND(indices, values, shape)];\n        }\n        case 'GatherNd': {\n          const x = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          return [tfOps.gatherND(x, indices)];\n        }\n        case 'SparseToDense': {\n          const indices =\n              getParamValue('sparseIndices', node, tensorMap, context) as\n              Tensor;\n          const shape =\n              getParamValue('outputShape', node, tensorMap, context) as\n              number[];\n          const sparseValues =\n              getParamValue('sparseValues', node, tensorMap, context) as Tensor;\n          const defaultValue =\n              getParamValue('defaultValue', node, tensorMap, context) as Scalar;\n          return [tfOps.sparseToDense(\n              indices, sparseValues, shape,\n              sparseValues.dtype === defaultValue.dtype ?\n                  defaultValue :\n                  tfOps.cast(defaultValue, sparseValues.dtype))];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'slice_join';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'FFT': {\n              return [tfOps.fft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IFFT': {\n              return [tfOps.ifft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'RFFT': {\n              return [tfOps.rfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IRFFT': {\n              return [tfOps.irfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'spectral';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Cast': {\n          return [tfOps.cast(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('dtype', node, tensorMap, context) as 'int32' |\n                  'float32' | 'bool')];\n        }\n        case 'ExpandDims': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.expandDims(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Squeeze': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          return [tfOps.squeeze(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n\n        case 'Reshape': {\n          return [tfOps.reshape(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        case 'PadV2':\n        case 'Pad': {\n          return [tfOps.pad(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('padding', node, tensorMap, context) as\n                  Array<[number, number]>,\n              getParamValue('constantValue', node, tensorMap, context) as\n                  number)];\n        }\n        case 'SpaceToBatchND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const paddings =\n              getParamValue('paddings', node, tensorMap, context) as number[][];\n          return [tfOps.spaceToBatchND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, paddings)];\n        }\n        case 'BatchToSpaceND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const crops =\n              getParamValue('crops', node, tensorMap, context) as number[][];\n          return [tfOps.batchToSpaceND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, crops)];\n        }\n        case 'DepthToSpace': {\n          const blockSize =\n              getParamValue('blockSize', node, tensorMap, context) as number;\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as\n               string).toUpperCase() as 'NHWC' |\n              'NCHW';\n          return [tfOps.depthToSpace(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              blockSize, dataFormat)];\n        }\n        case 'BroadcastTo': {\n          return [tfOps.broadcastTo(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'transformation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {ExecutionContext} from '../executor/execution_context';\n\nimport {NodeValueImpl} from './custom_op/node_value_impl';\nimport {getRegisteredOp} from './custom_op/register';\nimport * as arithmetic from './executors/arithmetic_executor';\nimport * as basicMath from './executors/basic_math_executor';\nimport * as control from './executors/control_executor';\nimport * as convolution from './executors/convolution_executor';\nimport * as creation from './executors/creation_executor';\nimport * as dynamic from './executors/dynamic_executor';\nimport * as evaluation from './executors/evaluation_executor';\nimport * as graph from './executors/graph_executor';\nimport * as image from './executors/image_executor';\nimport * as logical from './executors/logical_executor';\nimport * as matrices from './executors/matrices_executor';\nimport * as normalization from './executors/normalization_executor';\nimport * as reduction from './executors/reduction_executor';\nimport * as sliceJoin from './executors/slice_join_executor';\nimport * as spectral from './executors/spectral_executor';\nimport * as transformation from './executors/transformation_executor';\nimport {Node} from './types';\n\n/**\n * Executes the op defined by the node object.\n * @param node\n * @param tensorMap contains tensors for executed nodes and weights\n */\nexport function executeOp(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): tfc.Tensor[]|Promise<tfc.Tensor[]> {\n  const value =\n      ((node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) => {\n        switch (node.category) {\n          case 'arithmetic':\n            return tfc.tidy(\n                () => arithmetic.executeOp(node, tensorMap, context));\n          case 'basic_math':\n            return tfc.tidy(\n                () => basicMath.executeOp(node, tensorMap, context));\n          case 'control':\n            return control.executeOp(node, tensorMap, context);\n          case 'convolution':\n            return tfc.tidy(\n                () => convolution.executeOp(node, tensorMap, context));\n          case 'creation':\n            return tfc.tidy(() => creation.executeOp(node, tensorMap, context));\n          case 'dynamic':\n            return dynamic.executeOp(node, tensorMap, context);\n          case 'evaluation':\n            return tfc.tidy(\n                () => evaluation.executeOp(node, tensorMap, context));\n          case 'image':\n            return tfc.tidy(() => image.executeOp(node, tensorMap, context));\n          case 'graph':\n            return tfc.tidy(() => graph.executeOp(node, tensorMap, context));\n          case 'logical':\n            return tfc.tidy(() => logical.executeOp(node, tensorMap, context));\n          case 'matrices':\n            return tfc.tidy(() => matrices.executeOp(node, tensorMap, context));\n          case 'normalization':\n            return tfc.tidy(\n                () => normalization.executeOp(node, tensorMap, context));\n          case 'reduction':\n            return tfc.tidy(\n                () => reduction.executeOp(node, tensorMap, context));\n          case 'slice_join':\n            return tfc.tidy(\n                () => sliceJoin.executeOp(node, tensorMap, context));\n          case 'spectral':\n            return tfc.tidy(() => spectral.executeOp(node, tensorMap, context));\n          case 'transformation':\n            return tfc.tidy(\n                () => transformation.executeOp(node, tensorMap, context));\n          case 'custom':\n            const opMapper = getRegisteredOp(node.op);\n            if (opMapper && opMapper.customExecutor) {\n              return opMapper.customExecutor(\n                  new NodeValueImpl(node, tensorMap, context));\n            } else {\n              throw TypeError(`Custom op ${node.op} is not registered.`);\n            }\n          default:\n            throw TypeError(\n                `Unknown op '${node.op}'. File an issue at ` +\n                `https://github.com/tensorflow/tfjs/issues so we can add it` +\n                `, or register a custom execution with tf.registerOp()`);\n        }\n      })(node, tensorMap, context);\n  if (value instanceof Promise) {\n    return value.then((data) => [].concat(data));\n  }\n  return [].concat(value);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap, TensorArrayMap, TensorListMap} from '../data/types';\n\nimport {TensorArray} from './tensor_array';\nimport {TensorList} from './tensor_list';\nimport {FunctionExecutor} from './types';\n\nexport interface ExecutionContextInfo {\n  id: number;           // the unique id of the context info\n  frameName: string;    // The frame name of the loop, this comes from\n                        // the TensorFlow NodeDef.\n  iterationId: number;  // The iteration id of the loop\n}\n\n/**\n * ExecutionContext captures the runtime environment of the node. It keeps\n * track of the current frame and iteration for the control flow ops.\n *\n * For example, typical Dynamic RNN model may contain loops, for which\n * TensorFlow will generate graphs with Enter/Exit nodes to control the\n * current execution frame, and NextIteration Nodes for iteration id increment.\n * For model with branch logic, TensorFLow will generate Switch/Merge ops.\n */\nexport class ExecutionContext {\n  private rootContext = {id: 0, frameName: '', iterationId: 0};\n  private contexts: ExecutionContextInfo[] = [this.rootContext];\n  private lastId = 0;\n  private _currentContextIds: string[];\n\n  constructor(\n      readonly weightMap: NamedTensorsMap = {},\n      readonly tensorArrayMap: TensorArrayMap = {},\n      readonly tensorListMap: TensorListMap = {},\n      readonly functionMap: {[key: string]: FunctionExecutor} = {}) {\n    this.generateCurrentContextIds();\n  }\n\n  private newFrame(id: number, frameName: string) {\n    return {id, frameName, iterationId: 0};\n  }\n\n  /**\n   * Set the current context\n   * @param contexts: ExecutionContextInfo[] the current path of execution\n   * frames\n   */\n  set currentContext(contexts: ExecutionContextInfo[]) {\n    if (this.contexts !== contexts) {\n      this.contexts = contexts;\n      this.generateCurrentContextIds();\n    }\n  }\n\n  get currentContext(): ExecutionContextInfo[] {\n    return this.contexts;\n  }\n\n  /**\n   * Returns the current context in string format.\n   */\n  get currentContextId(): string {\n    return this._currentContextIds[0];\n  }\n\n  /**\n   * Returns the current context and all parent contexts in string format.\n   * This allow access to the nodes in the current and parent frames.\n   */\n  get currentContextIds(): string[] {\n    return this._currentContextIds;\n  }\n\n  private generateCurrentContextIds() {\n    const names = [];\n    for (let i = 0; i < this.contexts.length - 1; i++) {\n      const contexts = this.contexts.slice(0, this.contexts.length - i);\n      names.push(this.contextIdforContexts(contexts));\n    }\n    names.push('');\n    this._currentContextIds = names;\n  }\n\n  private contextIdforContexts(contexts: ExecutionContextInfo[]) {\n    return contexts ?\n        contexts\n            .map(\n                context => (context.id === 0 && context.iterationId === 0) ?\n                    '' :\n                    `${context.frameName}-${context.iterationId}`)\n            .join('/') :\n        '';\n  }\n\n  /**\n   * Enter a new frame, a new context is pushed on the current context list.\n   * @param frameId new frame id\n   */\n  enterFrame(frameId: string) {\n    if (this.contexts) {\n      this.lastId++;\n      this.contexts = this.contexts.slice();\n      this.contexts.push(this.newFrame(this.lastId, frameId));\n      this._currentContextIds.unshift(this.contextIdforContexts(this.contexts));\n    }\n  }\n\n  /**\n   * Exit the current frame, the last context is removed from the current\n   * context list.\n   */\n  exitFrame() {\n    if (this.contexts && this.contexts.length > 1) {\n      this.contexts = this.contexts.slice();\n      this.contexts.splice(-1);\n      this.currentContextIds.shift();\n    } else {\n      throw new Error('Cannot exit frame, the context is empty');\n    }\n  }\n\n  /**\n   * Enter the next iteration of a loop, the iteration id of last context is\n   * increased.\n   */\n  nextIteration() {\n    if (this.contexts && this.contexts.length > 0) {\n      this.contexts = this.contexts.slice();\n      this.lastId++;\n      const context =\n          Object.assign({}, this.contexts[this.contexts.length - 1]);\n      context.iterationId += 1;\n      context.id = this.lastId;\n      this.contexts.splice(-1, 1, context);\n      this._currentContextIds.splice(\n          0, 1, this.contextIdforContexts(this.contexts));\n    } else {\n      throw new Error('Cannot increase frame iteration, the context is empty');\n    }\n  }\n\n  getWeight(name: string): Tensor[] {\n    return this.weightMap[name];\n  }\n\n  addTensorArray(tensorArray: TensorArray) {\n    this.tensorArrayMap[tensorArray.id] = tensorArray;\n  }\n\n  getTensorArray(id: number): TensorArray {\n    return this.tensorArrayMap[id];\n  }\n\n  addTensorList(tensorList: TensorList) {\n    this.tensorListMap[tensorList.id] = tensorList;\n  }\n\n  getTensorList(id: number): TensorList {\n    return this.tensorListMap[id];\n  }\n\n  dispose(keepIds: Set<number>) {\n    for (const key in this.tensorArrayMap) {\n      this.tensorArrayMap[key].clearAndClose(keepIds);\n    }\n\n    for (const key in this.tensorListMap) {\n      this.tensorListMap[key].clearAndClose(keepIds);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NamedTensorMap} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {parseNodeName} from '../operations/executors/utils';\nimport {Graph, Node} from '../operations/types';\n\nexport interface ExecutionInfo {\n  inputs: NamedTensorMap;\n  outputs: Node[];\n  usedNodes: Set<string>;\n  missingInputs: string[];\n  dynamicNode: Node;\n  syncInputs: string[];\n}\n\n/**\n * Given graph inputs and desired outputs, find the minimal set of nodes\n * to execute in order to compute the outputs. In addition return other useful\n * info such:\n * - Missing inputs needed to compute the output.\n * - Whether the subgraph contains dynamic ops (control flow, dynamic shape).\n * - Alternative inputs in order to avoid async (dynamic op) execution.\n */\nexport function getExecutionSubgraph(\n    inputs: NamedTensorMap, outputs: Node[], weightMap: NamedTensorsMap,\n    initNodes?: Node[]): ExecutionInfo {\n  const usedNodes = new Set<string>();\n  const missingInputs: string[] = [];\n  let dynamicNode: Node = null;\n  let syncInputs: string[] = null;\n\n  // Start with the outputs, going backwards and find all the nodes that are\n  // needed to compute those outputs.\n  const seen = new Set<string>();\n  const inputNodeNames =\n      Object.keys(inputs).map(name => parseNodeName(name)[0]);\n\n  let initNodeNames: string[] = [];\n  if (initNodes != null) {\n    initNodeNames = initNodes.map(node => parseNodeName(node.name)[0]);\n  }\n\n  const frontier = [...outputs];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    if (isControlFlow(node) || isDynamicShape(node)) {\n      if (dynamicNode == null) {\n        dynamicNode = node;\n        syncInputs = dynamicNode.children.map(child => child.name)\n                         .filter(name => usedNodes.has(name));\n      }\n    }\n    usedNodes.add(node.name);\n\n    // Weights are dead end since we already have their values.\n    if (weightMap[node.name] != null) {\n      continue;\n    }\n    // This node is a dead end since it's one of the user-provided inputs.\n    if (inputNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    // This node is a dead end since it doesn't have any inputs.\n    if (initNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    if (node.inputs.length === 0) {\n      missingInputs.push(node.name);\n      continue;\n    }\n    node.inputs.forEach(input => {\n      // Don't add to the frontier if it is already there.\n      if (seen.has(input.name)) {\n        return;\n      }\n      seen.add(input.name);\n      frontier.push(input);\n    });\n  }\n  return {inputs, outputs, usedNodes, missingInputs, dynamicNode, syncInputs};\n}\n\n/**\n * Given the execution info, return a list of nodes in topological order that\n * need to be executed to compute the output.\n */\nexport function getNodesInTopologicalOrder(\n    graph: Graph, weightMap: NamedTensorsMap,\n    executionInfo: ExecutionInfo): Node[] {\n  const {usedNodes, inputs} = executionInfo;\n  const frontier: Node[] = [];\n  const inputNodes = Object.keys(inputs)\n                         .map(name => parseNodeName(name)[0])\n                         .map(name => graph.nodes[name]);\n  const initNodes = graph.initNodes;\n\n  inputNodes.forEach(input => {\n    if (usedNodes.has(input.name)) {\n      frontier.push(input);\n    }\n  });\n  graph.weights.forEach(weight => {\n    if (usedNodes.has(weight.name)) {\n      frontier.push(weight);\n    }\n  });\n  if (initNodes != null) {\n    initNodes.forEach(node => {\n      if (usedNodes.has(node.name)) {\n        frontier.push(node);\n      }\n    });\n  }\n  const seen = new Set<string>();\n  const orderedNodes: Node[] = [];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    seen.add(node.name);\n    if (!weightMap[node.name]) {\n      orderedNodes.push(node);\n    }\n    node.children.forEach(child => {\n      if (!seen.has(child.name) && usedNodes.has(child.name) &&\n          child.inputs.every(input => seen.has(input.name))) {\n        frontier.push(child);\n      }\n    });\n  }\n  return orderedNodes;\n}\n\nconst CONTROL_FLOW_OPS = [\n  'Switch', 'Merge', 'Enter', 'Exit', 'NextIteration', 'StatelessIf',\n  'StatelessWhile', 'if', 'While'\n];\nconst DYNAMIC_SHAPE_OPS = [\n  'NonMaxSuppressionV2', 'NonMaxSuppressionV3', 'NonMaxSuppressionV5', 'Where'\n];\n\nexport function isControlFlow(node: Node) {\n  return CONTROL_FLOW_OPS.indexOf(node.op) >= 0;\n}\n\nexport function isDynamicShape(node: Node) {\n  return DYNAMIC_SHAPE_OPS.indexOf(node.op) >= 0;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NamedTensorMap, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {ISignatureDef} from '../data/compiled_api';\nimport {NamedTensorsMap, TensorArrayMap, TensorInfo, TensorListMap} from '../data/types';\nimport {getNodeNameAndIndex, getParamValue, getTensor, getTensorsForCurrentContenxt, parseNodeName} from '../operations/executors/utils';\nimport {executeOp} from '../operations/operation_executor';\nimport {Graph, Node} from '../operations/types';\n\nimport {ExecutionContext, ExecutionContextInfo} from './execution_context';\nimport {getExecutionSubgraph, getNodesInTopologicalOrder, isControlFlow} from './model_analysis';\nimport {FunctionExecutor} from './types';\n\ninterface NodeWithContexts {\n  contexts: ExecutionContextInfo[];\n  node: Node;\n}\n\nexport class GraphExecutor implements FunctionExecutor {\n  private compiledMap: Map<string, Node[]> = new Map();\n  private _weightMap: NamedTensorsMap = {};\n  private _weightIds: number[];\n  private _signature: ISignatureDef;\n  private _inputs: Node[];\n  private _outputs: Node[];\n  private _initNodes: Node[];\n  private SEPERATOR = ',';\n  private _functions: {[key: string]: Graph} = {};\n  private _functionExecutorMap: {[key: string]: FunctionExecutor} = {};\n\n  get weightIds(): number[] {\n    return this.parent ? this.parent.weightIds : this._weightIds;\n  }\n\n  get functionExecutorMap(): {[key: string]: FunctionExecutor} {\n    return this.parent ? this.parent.functionExecutorMap :\n                         this._functionExecutorMap;\n  }\n\n  get weightMap(): NamedTensorsMap {\n    return this.parent ? this.parent.weightMap : this._weightMap;\n  }\n\n  set weightMap(weightMap: NamedTensorsMap) {\n    const weightIds = Object.keys(weightMap).map(\n        key => weightMap[key].map(tensor => tensor.id));\n    this._weightIds = [].concat(...weightIds);\n    this._weightMap = weightMap;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this._inputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get outputs(): TensorInfo[] {\n    return this._outputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get inputNodes(): string[] {\n    return this._inputs.map(node => node.signatureKey || node.name);\n  }\n\n  get outputNodes(): string[] {\n    return this._outputs.map((node) => {\n      const name = node.signatureKey || node.name;\n      return node.defaultOutput ? (`${name}:${node.defaultOutput}`) : name;\n    });\n  }\n\n  get functions(): {[key: string]: ISignatureDef} {\n    return Object.keys(this._functions).reduce((map, key) => {\n      map[key] = this._functions[key].signature;\n      return map;\n    }, {} as {[key: string]: ISignatureDef});\n  }\n\n  /**\n   *\n   * @param graph Graph the model or function graph to be executed.\n   * @param parent When building function exector you need to set the parent\n   * executor. Since the weights and function executor maps are set at parant\n   * level, that function executor can access the function maps and weight maps\n   * through the parent.\n   */\n  constructor(private graph: Graph, private parent?: GraphExecutor) {\n    this._outputs = graph.outputs;\n    this._inputs = graph.inputs;\n    this._initNodes = graph.initNodes;\n    this._signature = graph.signature;\n    this._functions = graph.functions;\n    // create sub-graph executors\n    if (graph.functions != null) {\n      Object.keys(graph.functions).forEach(name => {\n        this._functionExecutorMap[name] =\n            new GraphExecutor(graph.functions[name], this);\n      });\n    }\n  }\n\n  private getCompilationKey(inputs: Node[], outputs: Node[]): string {\n    const sortedInputs = inputs.map(node => node.name).sort();\n    const sortedOutputs = outputs.map(node => node.name).sort();\n    return sortedInputs.join(this.SEPERATOR) + '--' +\n        sortedOutputs.join(this.SEPERATOR);\n  }\n\n  /**\n   * Compiles the inference graph and returns the minimal set of nodes that are\n   * required for execution, in the correct execution order.\n   */\n  private compile(inputs: NamedTensorMap, outputs: Node[]): Node[] {\n    const executionInfo =\n        getExecutionSubgraph(inputs, outputs, this.weightMap, this._initNodes);\n    const {missingInputs, dynamicNode, syncInputs} = executionInfo;\n    if (dynamicNode != null) {\n      throw new Error(\n          `This execution contains the node '${dynamicNode.name}', which has ` +\n          `the dynamic op '${dynamicNode.op}'. Please use ` +\n          `model.executeAsync() instead. Alternatively, to avoid the ` +\n          `dynamic ops, specify the inputs [${syncInputs}]`);\n    }\n\n    if (missingInputs.length > 0) {\n      const outNames = outputs.map(n => n.name);\n      const inNames = Object.keys(inputs);\n      throw new Error(\n          `Cannot compute the outputs [${outNames}] from the provided inputs ` +\n          `[${inNames}]. Missing the following inputs: [${missingInputs}]`);\n    }\n\n    return getNodesInTopologicalOrder(\n        this.graph, this.weightMap, executionInfo);\n  }\n\n  /**\n   * Executes the inference for given input tensors.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  execute(inputs: NamedTensorMap, outputs: string[]): Tensor[] {\n    inputs = this.mapInputs(inputs);\n    const names = Object.keys(inputs).sort();\n    this.checkInputs(inputs);\n    this.checkInputShapeAndType(inputs);\n    outputs = this.mapOutputs(outputs);\n    this.checkOutputs(outputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputs.map(name => parseNodeName(name)[0]);\n    let outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n\n    // If no outputs are specified, then use the default outputs of the model.\n    if (outputNodes.length === 0) {\n      outputNodes = this._outputs;\n    }\n\n    const compilationKey = this.getCompilationKey(inputNodes, outputNodes);\n\n    // Do nothing if the compiled graph cache contains the input.\n    let orderedNodes = this.compiledMap.get(compilationKey);\n    if (orderedNodes == null) {\n      orderedNodes = this.compile(inputs, outputNodes);\n      this.compiledMap.set(compilationKey, orderedNodes);\n    }\n\n    const tensorArrayMap: TensorArrayMap = {};\n    const tensorListMap: TensorListMap = {};\n    return tidy(() => {\n      const context = new ExecutionContext(\n          this.weightMap, tensorArrayMap, tensorListMap,\n          this.functionExecutorMap);\n      const tensorsMap: NamedTensorsMap = {...this.weightMap};\n\n      Object.keys(inputs).forEach(name => {\n        const [nodeName, index] = parseNodeName(name);\n        const tensors: Tensor[] = [];\n        tensors[index] = inputs[name];\n        tensorsMap[nodeName] = tensors;\n      });\n\n      const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n      const intermediateTensorConsumerCount: {[key: number]: number} = {};\n      for (let i = 0; i < orderedNodes.length; i++) {\n        const node = orderedNodes[i];\n        if (!tensorsMap[node.name]) {\n          const tensors = executeOp(node, tensorsMap, context) as Tensor[];\n          if (tensors instanceof Promise) {\n            throw new Error(\n                `The execution of the op '${node.op}' returned a promise. ` +\n                `Please use model.executeAsync() instead.`);\n          }\n          tensorsMap[node.name] = tensors;\n          this.checkTensorForDisposal(\n              node.name, node, tensorsMap, context, tensorsToKeep,\n              outputNodeNames, intermediateTensorConsumerCount);\n        }\n      }\n      // dispose the context for the root executor\n      if (this.parent == null) {\n        context.dispose(tensorsToKeep);\n      }\n      return outputs.map(name => getTensor(name, tensorsMap, context));\n    });\n  }\n\n  private getFrozenTensorIds(tensorMap: NamedTensorsMap): Set<number> {\n    const ids = [].concat.apply(\n        [],\n        Object.keys(tensorMap)\n            .map(key => tensorMap[key])\n            .map(tensors => tensors.map(tensor => tensor.id)));\n    return new Set(ids);\n  }\n  private checkTensorForDisposal(\n      nodeName: string, node: Node, tensorMap: NamedTensorsMap,\n      context: ExecutionContext, tensorsToKeep: Set<number>,\n      outputNames: string[],\n      intermediateTensorConsumerCount: {[key: string]: number}) {\n    // Skip output nodes and any control flow nodes, since its dependency is\n    // tricky to track correctly.\n    if (node.category === 'control' || outputNames.indexOf(nodeName) !== -1) {\n      return;\n    }\n\n    tensorMap[nodeName].forEach(tensor => {\n      if (tensor != null) {\n        intermediateTensorConsumerCount[tensor.id] =\n            (intermediateTensorConsumerCount[tensor.id] || 0) +\n            node.children.length;\n      }\n    });\n    node.inputs.forEach(input => {\n      // Skip any control flow nodes, since its dependency is tricky to track\n      // correctly.\n      if (input.category !== 'control') {\n        const tensors =\n            getTensorsForCurrentContenxt(input.name, tensorMap, context);\n        if (tensors != null) {\n          tensors.forEach(tensor => {\n            if (tensor && !tensorsToKeep.has(tensor.id)) {\n              const count = intermediateTensorConsumerCount[tensor.id];\n              if (count === 1) {\n                tensor.dispose();\n                delete intermediateTensorConsumerCount[tensor.id];\n              } else if (count != null) {\n                // only intermediate nodes has count set, inputs and weights are\n                // not.\n                intermediateTensorConsumerCount[tensor.id]--;\n              }\n            }\n          });\n        }\n      }\n    });\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  async executeAsync(inputs: NamedTensorMap, outputs: string[]):\n      Promise<Tensor[]> {\n    return this._executeAsync(inputs, outputs);\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   * @param isFunctionExecution Flag for executing a function.\n   * @param tensorArrayMap Optional, global TensorArray map by id. Used for\n   * function execution.\n   * @param tensorArrayMap Optinal global TensorList map by id. Used for\n   * function execution.\n   */\n  private async _executeAsync(\n      inputs: NamedTensorMap, outputs: string[], isFunctionExecution = false,\n      tensorArrayMap: TensorArrayMap = {},\n      tensorListMap: TensorListMap = {}): Promise<Tensor[]> {\n    if (!isFunctionExecution) {\n      inputs = this.mapInputs(inputs);\n      this.checkInputs(inputs);\n      this.checkInputShapeAndType(inputs);\n      outputs = this.mapOutputs(outputs);\n      this.checkOutputs(outputs);\n    }\n\n    const context = new ExecutionContext(\n        this.weightMap, tensorArrayMap, tensorListMap,\n        this.functionExecutorMap);\n\n    // Graph with control flow op requires runtime evaluation of the execution\n    // order, while without control flow the execution order is pre-determined\n    // in the compile method.\n    const tensorMap = await this.executeWithControlFlow(\n        inputs, context, outputs, isFunctionExecution);\n    const results = outputs.map(name => getTensor(name, tensorMap, context));\n\n    // dispose all the intermediate tensors\n    const outputIds = results.map(t => t.id);\n    const inputIds = Object.keys(inputs).map(name => inputs[name].id);\n    const keepIds =\n        new Set<number>([...outputIds, ...inputIds, ...this.weightIds]);\n    Object.keys(tensorMap).forEach(key => {\n      const tensorArray = tensorMap[key];\n      tensorArray.forEach(tensor => {\n        if (tensor && !tensor.isDisposed && !keepIds.has(tensor.id)) {\n          tensor.dispose();\n        }\n      });\n    });\n    // dispose the context for the root executor\n    if (this.parent == null) {\n      context.dispose(keepIds);\n    }\n\n    return results;\n  }\n\n  async executeFunctionAsync(\n      inputs: Tensor[], tensorArrayMap: TensorArrayMap,\n      tensorListMap: TensorListMap): Promise<Tensor[]> {\n    const mappedInputs = inputs.reduce((map, tensor, index) => {\n      map[this.inputs[index].name] = tensor;\n      return map;\n    }, {} as NamedTensorMap);\n\n    return this._executeAsync(\n        mappedInputs, this.outputNodes, true, tensorArrayMap, tensorListMap);\n  }\n  /**\n   * When there are control flow nodes in the graph, the graph execution use\n   * ExecutionContext to keep track of the frames and loop iterators.\n   * @param inputs placeholder tensors for the graph.\n   * @param context the execution context object for current execution.\n   * @param isFunctionExecution Flag for executing a function.\n   */\n  private async executeWithControlFlow(\n      inputs: NamedTensorMap, context: ExecutionContext, outputNames: string[],\n      isFunctionExecution: boolean): Promise<NamedTensorsMap> {\n    const names = Object.keys(inputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputNames.map(name => parseNodeName(name)[0]);\n    const outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n    const {usedNodes, missingInputs, dynamicNode, syncInputs} =\n        getExecutionSubgraph(inputs, outputNodes, this.weightMap);\n\n    const stack: NodeWithContexts[] =\n        [...inputNodes, ...this.graph.weights].map(node => {\n          return {node, contexts: context.currentContext};\n        });\n    const tensorsMap: NamedTensorsMap = {...this.weightMap};\n    Object.keys(inputs).forEach(name => {\n      const [nodeName, index] = parseNodeName(name);\n      const tensors: Tensor[] = [];\n      tensors[index] = inputs[name];\n      tensorsMap[nodeName] = tensors;\n    });\n    const intermediateTensorConsumerCount: {[key: number]: number} = {};\n    const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n    const added: {[key: string]: boolean} = {};\n    while (stack.length > 0) {\n      const promises = this.processStack(\n          inputNodes, stack, context, tensorsMap, added, tensorsToKeep,\n          outputNodeNames, intermediateTensorConsumerCount, usedNodes);\n      await Promise.all(promises);\n    }\n    if (dynamicNode == null && !isFunctionExecution) {\n      console.warn(\n          `This model execution did not contain any nodes with control flow ` +\n          `or dynamic output shapes. You can use model.execute() instead.`);\n    }\n    const missingOutputs =\n        outputNodes\n            .filter(\n                node => !isControlFlow(node) &&\n                    !getTensor(node.name, tensorsMap, context))\n            .map(node => node.name);\n    if (missingOutputs.length > 0) {\n      let alternativeMsg = '';\n      if (dynamicNode != null) {\n        alternativeMsg =\n            `Alternatively, to avoid the dynamic ops, use model.execute() ` +\n            `and specify the inputs [${syncInputs}]`;\n      }\n      throw new Error(\n          `Cannot compute the outputs [${missingOutputs}] from the provided ` +\n          `inputs [${names}]. Consider providing the following inputs: ` +\n          `[${missingInputs}]. ${alternativeMsg}`);\n    }\n    return tensorsMap;\n  }\n\n  private processStack(\n      inputNodes: Node[], stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      tensorsToKeep: Set<number>, outputNames: string[],\n      intermediateTensorConsumerCount: {[key: number]: number},\n      usedNodes: Set<string>) {\n    const promises: Array<Promise<Tensor[]>> = [];\n    while (stack.length > 0) {\n      const item = stack.pop();\n      context.currentContext = item.contexts;\n      let nodeName = '';\n      // The tensor of the Enter op with isConstant set should be set\n      // in the parent scope, so it will be available as constant for the\n      // whole loop.\n      if (item.node.op === 'Enter' &&\n          getParamValue('isConstant', item.node, tensorMap, context)) {\n        [nodeName] = getNodeNameAndIndex(item.node.name, context);\n      }\n\n      // only process nodes that are not provided as input nodes.\n      if (inputNodes.indexOf(item.node) === -1) {\n        const tensors = executeOp(item.node, tensorMap, context);\n        if (!nodeName) {\n          [nodeName] = getNodeNameAndIndex(item.node.name, context);\n        }\n        const currentContext = context.currentContext;\n        if (tensors instanceof Promise) {\n          promises.push(tensors.then(t => {\n            tensorMap[nodeName] = t;\n            context.currentContext = currentContext;\n            this.checkTensorForDisposal(\n                nodeName, item.node, tensorMap, context, tensorsToKeep,\n                outputNames, intermediateTensorConsumerCount);\n            this.processChildNodes(\n                item.node, stack, context, tensorMap, added, usedNodes);\n            return t;\n          }));\n        } else {\n          tensorMap[nodeName] = tensors;\n          this.checkTensorForDisposal(\n              nodeName, item.node, tensorMap, context, tensorsToKeep,\n              outputNames, intermediateTensorConsumerCount);\n          this.processChildNodes(\n              item.node, stack, context, tensorMap, added, usedNodes);\n        }\n      } else {\n        this.processChildNodes(\n            item.node, stack, context, tensorMap, added, usedNodes);\n      }\n    }\n    return promises;\n  }\n\n  private processChildNodes(\n      node: Node, stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      usedNodes: Set<string>) {\n    node.children.forEach((childNode) => {\n      const [nodeName, ] = getNodeNameAndIndex(childNode.name, context);\n      if (added[nodeName] || !usedNodes.has(childNode.name)) {\n        return;\n      }\n      // Merge op can be pushed if any of its inputs has value.\n      if (childNode.op === 'Merge') {\n        if (childNode.inputNames.some(name => {\n              return !!getTensor(name, tensorMap, context);\n            })) {\n          added[nodeName] = true;\n          stack.push({contexts: context.currentContext, node: childNode});\n        }\n      } else  // Otherwise all inputs must to have value.\n          if (childNode.inputNames.every(name => {\n                return !!getTensor(name, tensorMap, context);\n              })) {\n        added[nodeName] = true;\n        stack.push({contexts: context.currentContext, node: childNode});\n      }\n    });\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   */\n  dispose() {\n    Object.keys(this.weightMap)\n        .forEach(\n            key => this.weightMap[key].forEach(tensor => tensor.dispose()));\n  }\n\n  private checkInputShapeAndType(inputs: NamedTensorMap) {\n    Object.keys(inputs).forEach(name => {\n      const input = inputs[name];\n      const [nodeName, ] = parseNodeName(name);\n      const node = this.graph.nodes[nodeName];\n      if (node.attrParams['shape'] && node.attrParams['shape'].value) {\n        const shape = node.attrParams['shape'].value as number[];\n        const match = shape.length === input.shape.length &&\n            input.shape.every(\n                (dim, index) => shape[index] === -1 || shape[index] === dim);\n        util.assert(\n            match,\n            () => `The shape of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be [${shape}], but was ` +\n                `[${input.shape}]`);\n      }\n      if (node.attrParams['dtype'] && node.attrParams['dtype'].value) {\n        util.assert(\n            input.dtype === node.attrParams['dtype'].value as string,\n            () => `The dtype of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be ` +\n                `${node.attrParams['dtype'].value}, but was ${input.dtype}`);\n      }\n    });\n  }\n\n  private mapInputs(inputs: NamedTensorMap) {\n    const result: NamedTensorMap = {};\n    for (const inputName in inputs) {\n      if (this._signature != null && this._signature.inputs != null &&\n          this._signature.inputs[inputName] != null) {\n        const tensor = this._signature.inputs[inputName];\n        result[tensor.name] = inputs[inputName];\n      } else {\n        result[inputName] = inputs[inputName];\n      }\n    }\n    return result;\n  }\n\n  private checkInputs(inputs: NamedTensorMap) {\n    const notInGraph = Object.keys(inputs).filter(name => {\n      const [nodeName] = parseNodeName(name);\n      return this.graph.nodes[nodeName] == null;\n    });\n    if (notInGraph.length > 0) {\n      throw new Error(\n          `The dict provided in model.execute(dict) has ` +\n          `keys: [${notInGraph}] that are not part of graph`);\n    }\n  }\n\n  private mapOutputs(outputs: string[]) {\n    return outputs.map(name => {\n      if (this._signature != null && this._signature.outputs != null &&\n          this._signature.outputs[name] != null) {\n        const tensor = this._signature.outputs[name];\n        return tensor.name;\n      }\n      return name;\n    }, {});\n  }\n\n  private checkOutputs(outputs: string[]): void {\n    outputs.forEach(name => {\n      const [normalizedName] = parseNodeName(name);\n      if (!this.graph.nodes[normalizedName]) {\n        throw new Error(`The output '${name}' is not found in the graph`);\n      }\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {InferenceModel, io, ModelPredictConfig, NamedTensorMap, Tensor} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\nimport {NamedTensorsMap, TensorInfo} from '../data/types';\nimport {OperationMapper} from '../operations/operation_mapper';\n\nimport {GraphExecutor} from './graph_executor';\n\nexport const TFHUB_SEARCH_PARAM = '?tfjs-format=file';\nexport const DEFAULT_MODEL_NAME = 'model.json';\n/**\n * A `tf.GraphModel` is a directed, acyclic graph built from a\n * SavedModel GraphDef and allows inference execution.\n *\n * A `tf.GraphModel` can only be created by loading from a model converted from\n * a [TensorFlow SavedModel](https://www.tensorflow.org/guide/saved_model) using\n * the command line converter tool and loaded via `tf.loadGraphModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class GraphModel implements InferenceModel {\n  private executor: GraphExecutor;\n  private version = 'n/a';\n  private handler: io.IOHandler;\n  private artifacts: io.ModelArtifacts;\n  private initializer: GraphExecutor;\n\n  // Returns the version information for the tensorflow model GraphDef.\n  get modelVersion(): string {\n    return this.version;\n  }\n\n  get inputNodes(): string[] {\n    return this.executor.inputNodes;\n  }\n\n  get outputNodes(): string[] {\n    return this.executor.outputNodes;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this.executor.inputs;\n  }\n\n  get outputs(): TensorInfo[] {\n    return this.executor.outputs;\n  }\n\n  get weights(): NamedTensorsMap {\n    return this.executor.weightMap;\n  }\n\n  /**\n   * @param modelUrl url for the model, or an `io.IOHandler`.\n   * @param weightManifestUrl url for the weight file generated by\n   * scripts/convert.py script.\n   * @param requestOption options for Request, which allows to send credentials\n   * and custom headers.\n   * @param onProgress Optional, progress callback function, fired periodically\n   * before the load is completed.\n   */\n  constructor(\n      private modelUrl: string|io.IOHandler,\n      private loadOptions: io.LoadOptions = {}) {\n    if (loadOptions == null) {\n      this.loadOptions = {};\n    }\n  }\n\n  private findIOHandler() {\n    const path = this.modelUrl;\n    if ((path as io.IOHandler).load != null) {\n      // Path is an IO Handler.\n      this.handler = path as io.IOHandler;\n    } else if (this.loadOptions.requestInit != null) {\n      this.handler = io.browserHTTPRequest(path as string, this.loadOptions);\n    } else {\n      const handlers = io.getLoadHandlers(path as string, this.loadOptions);\n      if (handlers.length === 0) {\n        // For backward compatibility: if no load handler can be found,\n        // assume it is a relative http path.\n        handlers.push(io.browserHTTPRequest(path as string, this.loadOptions));\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) load handlers for ` +\n            `URL '${[path]}'`);\n      }\n      this.handler = handlers[0];\n    }\n  }\n\n  /**\n   * Loads the model and weight files, construct the in memory weight map and\n   * compile the inference graph.\n   */\n  async load(): Promise<boolean> {\n    this.findIOHandler();\n    if (this.handler.load == null) {\n      throw new Error(\n          'Cannot proceed with model loading because the IOHandler provided ' +\n          'does not have the `load` method implemented.');\n    }\n    const artifacts = await this.handler.load();\n\n    return this.loadSync(artifacts);\n  }\n\n  /**\n   * Synchronously construct the in memory weight map and\n   * compile the inference graph. Also initialize hashtable if any.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  loadSync(artifacts: io.ModelArtifacts) {\n    this.artifacts = artifacts;\n    const graph = this.artifacts.modelTopology as tensorflow.IGraphDef;\n    let signature = {};\n    if (this.artifacts.userDefinedMetadata != null) {\n      signature =  // tslint:disable-next-line:no-any\n          (this.artifacts.userDefinedMetadata as any).signature as\n          tensorflow.ISignatureDef;\n    }\n\n    this.version = `${graph.versions.producer}.${graph.versions.minConsumer}`;\n    const weightMap =\n        io.decodeWeights(this.artifacts.weightData, this.artifacts.weightSpecs);\n    this.executor = new GraphExecutor(\n        OperationMapper.Instance.transformGraph(graph, signature));\n    this.executor.weightMap = this.convertTensorMapToTensorsMap(weightMap);\n\n    if (artifacts.modelInitializer != null) {\n      const initializer =\n          OperationMapper.Instance.transformGraph(artifacts.modelInitializer);\n      this.initializer = new GraphExecutor(initializer);\n      this.initializer.weightMap = this.executor.weightMap;\n      this.initializer.execute({}, []);\n    }\n\n    return true;\n  }\n\n  /**\n   * Save the configuration and/or weights of the GraphModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const modelUrl =\n   *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n   * const model = await tf.loadGraphModel(modelUrl);\n   * const zeros = tf.zeros([1, 224, 224, 3]);\n   * model.predict(zeros).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * model.predict(zeros).print();\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new Error(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new Error(\n          'GraphModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    return handlerOrURL.save(this.artifacts);\n  }\n\n  /**\n   * Execute the inference for the input tensors.\n   *\n   * @param input The input tensors, when there is single input for the model,\n   * inputs param should be a `tf.Tensor`. For models with mutliple inputs,\n   * inputs params should be in either `tf.Tensor`[] if the input order is\n   * fixed, or otherwise NamedTensorMap format.\n   *\n   * For model with multiple inputs, we recommend you use NamedTensorMap as the\n   * input type, if you use `tf.Tensor`[], the order of the array needs to\n   * follow the\n   * order of inputNodes array. @see {@link GraphModel.inputNodes}\n   *\n   * You can also feed any intermediate nodes using the NamedTensorMap as the\n   * input type. For example, given the graph\n   *    InputNode => Intermediate => OutputNode,\n   * you can execute the subgraph Intermediate => OutputNode by calling\n   *    model.execute('IntermediateNode' : tf.tensor(...));\n   *\n   * This is useful for models that uses tf.dynamic_rnn, where the intermediate\n   * state needs to be fed manually.\n   *\n   * For batch inference execution, the tensors for each input need to be\n   * concatenated together. For example with mobilenet, the required input shape\n   * is [1, 244, 244, 3], which represents the [batch, height, width, channel].\n   * If we are provide a batched data of 100 images, the input tensor should be\n   * in the shape of [100, 244, 244, 3].\n   *\n   * @param config Prediction configuration for specifying the batch size and\n   * output node names. Currently the batch size option is ignored for graph\n   * model.\n   *\n   * @returns Inference result tensors. The output would be single `tf.Tensor`\n   * if model has single output node, otherwise Tensor[] or NamedTensorMap[]\n   * will be returned for model with multiple outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(inputs: Tensor|Tensor[]|NamedTensorMap, config?: ModelPredictConfig):\n      Tensor|Tensor[]|NamedTensorMap {\n    return this.execute(inputs, this.outputNodes);\n  }\n\n  private normalizeInputs(inputs: Tensor|Tensor[]|\n                          NamedTensorMap): NamedTensorMap {\n    if (!(inputs instanceof Tensor) && !Array.isArray(inputs)) {\n      // The input is already a NamedTensorMap.\n      return inputs;\n    }\n    inputs = Array.isArray(inputs) ? inputs : [inputs];\n    if (inputs.length !== this.inputNodes.length) {\n      throw new Error(\n          'Input tensor count mismatch,' +\n          `the graph model has ${this.inputNodes.length} placeholders, ` +\n          `while there are ${inputs.length} input tensors.`);\n    }\n    return this.inputNodes.reduce((map, inputName, i) => {\n      map[inputName] = (inputs as Tensor[])[i];\n      return map;\n    }, {} as NamedTensorMap);\n  }\n\n  private normalizeOutputs(outputs: string|string[]): string[] {\n    outputs = outputs || this.outputNodes;\n    return !Array.isArray(outputs) ? [outputs] : outputs;\n  }\n\n  /**\n   * Executes inference for the model for given input tensors.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no\n   * outputs are specified, the default outputs of the model would be used.\n   * You can inspect intermediate nodes of the model by adding them to the\n   * outputs array.\n   *\n   * @returns A single tensor if provided with a single output or no outputs\n   * are provided and there is only one default output, otherwise return a\n   * tensor array. The order of the tensor array is the same as the outputs\n   * if provided, otherwise the order of outputNodes attribute of the model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs?: string|string[]):\n      Tensor|Tensor[] {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = this.executor.execute(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n  /**\n   * Executes inference for the model for given input tensors in async\n   * fashion, use this method when your model contains control flow ops.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   *\n   * @returns A Promise of single tensor if provided with a single output or\n   * no outputs are provided and there is only one default output, otherwise\n   * return a tensor map.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async executeAsync(\n      inputs: Tensor|Tensor[]|NamedTensorMap,\n      outputs?: string|string[]): Promise<Tensor|Tensor[]> {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = await this.executor.executeAsync(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n\n  private convertTensorMapToTensorsMap(map: NamedTensorMap): NamedTensorsMap {\n    return Object.keys(map).reduce((newMap: NamedTensorsMap, key) => {\n      newMap[key] = [map[key]];\n      return newMap;\n    }, {});\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  dispose() {\n    this.executor.dispose();\n\n    if (this.initializer) {\n      this.initializer.dispose();\n    }\n  }\n}\n\n/**\n * Load a graph model given a URL to the model definition.\n *\n * Example of loading MobileNetV2 from a URL and making a prediction with a\n * zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n * const model = await tf.loadGraphModel(modelUrl);\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n *\n * Example of loading MobileNetV2 from a TF Hub URL and making a prediction with\n * a zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/2';\n * const model = await tf.loadGraphModel(modelUrl, {fromTFHub: true});\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n * @param modelUrl The url or an `io.IOHandler` that loads the model.\n * @param options Options for the HTTP request, which allows to send credentials\n *    and custom headers.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport async function loadGraphModel(\n    modelUrl: string|io.IOHandler,\n    options: io.LoadOptions = {}): Promise<GraphModel> {\n  if (modelUrl == null) {\n    throw new Error(\n        'modelUrl in loadGraphModel() cannot be null. Please provide a url ' +\n        'or an IOHandler that loads the model');\n  }\n  if (options == null) {\n    options = {};\n  }\n\n  if (options.fromTFHub) {\n    if ((modelUrl as io.IOHandler).load == null) {\n      if (!(modelUrl as string).endsWith('/')) {\n        modelUrl = (modelUrl as string) + '/';\n      }\n      modelUrl = `${modelUrl}${DEFAULT_MODEL_NAME}${TFHUB_SEARCH_PARAM}`;\n    }\n  }\n  const model = new GraphModel(modelUrl, options);\n  await model.load();\n  return model;\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\n\n// tslint:disable:no-any\n\n/**\n * A return value for a mapping function that can be applied via deepMap.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapResult = {\n  value: any,\n  recurse: boolean\n};\n\n/**\n * Apply a mapping function to a nested structure in a recursive manner.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapResult`.  The `DeepMapResult` either provides a\n *   replacement value for that node (i.e., replacing the subtree), or indicates\n *   that the node should be processed recursively.\n */\nexport function deepMap(input: any, mapFn: (x: any) => DeepMapResult): any|\n    any[] {\n  return deepMapInternal(input, mapFn);\n}\n\n/**\n * @param seen: A Map of known object mappings (i.e., memoized results of\n *   `mapFn()`)\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepMapInternal(\n    input: any, mapFn: (x: any) => DeepMapResult,\n    seen: Map<any, any> = new Map(), containedIn: Set<{}> = new Set()): any|\n    any[] {\n  if (input == null) {\n    return null;\n  }\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  if (seen.has(input)) {\n    return seen.get(input);\n  }\n  const result = mapFn(input);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep map function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    seen.set(input, result.value);\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const child = input[k];\n      const childResult = deepMapInternal(child, mapFn, seen, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// TODO(soergel, kangyizhang) Reconsider naming of deepZip() to avoid confusion\n// with zip()\n\n/**\n * Zip nested structures together in a recursive manner.\n *\n * This has the effect of transposing or pivoting data, e.g. converting it from\n * a row-major representation to a column-major representation.\n *\n * For example, `deepZip([{a: 1, b: 2}, {a: 3, b: 4}])` returns\n * `{a: [1, 3], b: [2, 4]}`.\n *\n * The inputs should all have the same nested structure (i.e., of arrays and\n * dicts).  The result is a single object with the same nested structure, where\n * the leaves are arrays collecting the values of the inputs at that location\n * (or, optionally, the result of a custom function applied to those arrays).\n *\n * @param inputs: An array of the objects to zip together.\n * @param zipFn: (optional) A function that expects an array of elements at a\n *   single node of the object tree, and returns a `DeepMapResult`.  The\n *   `DeepMapResult` either provides a result value for that node (i.e.,\n *   representing the subtree), or indicates that the node should be processed\n *   recursively.  The default zipFn recurses as far as possible and places\n *   arrays at the leaves.\n */\nexport function deepZip(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult = zipToList): any|any[] {\n  return deepZipInternal(inputs, zipFn);\n}\n\n/**\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepZipInternal(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult,\n    containedIn: Set<{}> = new Set()): any|any[] {\n  // The recursion follows the structure of input 0; it's assumed that all the\n  // other inputs have the same structure.\n  const input = inputs[0];\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  const result = zipFn(inputs);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep zip function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const children = inputs.map(x => x[k]);\n      const childResult = deepZipInternal(children, zipFn, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// tslint:disable-next-line:no-any\nexport function zipToList(x: any[]): DeepMapResult {\n  if (x === null) {\n    return null;\n  }\n  // TODO(soergel): validate array type?\n\n  if (isIterable(x[0])) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: x, recurse: false};\n  }\n}\n\n/**\n * A return value for an async map function for use with deepMapAndAwaitAll.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapAsyncResult = {\n  value: Promise<any>,\n  recurse: boolean\n};\n\n/**\n * Apply an async mapping function to a nested structure in a recursive manner.\n *\n * This first creates a nested structure of Promises, and then awaits all of\n * those, resulting in a single Promise for a resolved nested structure.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapAsyncResult`.  The `DeepMapAsyncResult` either provides\n *   a `Promise` for a replacement value for that node (i.e., replacing the\n *   subtree), or indicates that the node should be processed recursively.  Note\n *   that the decision whether or not to recurse must be made immediately; only\n *   the mapped value may be promised.\n */\nexport async function deepMapAndAwaitAll(\n    input: any, mapFn: (x: any) => DeepMapAsyncResult): Promise<any|any[]> {\n  const seen: Map<any, Promise<any>> = new Map();\n\n  // First do a normal deepMap, collecting Promises in 'seen' as a side effect.\n  deepMapInternal(input, mapFn, seen);\n\n  // Replace the Promises in 'seen' in place.\n  // Note TypeScript provides no async map iteration, and regular map iteration\n  // is broken too, so sadly we have to do Array.from() to make it work.\n  // (There's no advantage to Promise.all(), and that would be tricky anyway.)\n  for (const key of Array.from(seen.keys())) {\n    const value = seen.get(key);\n    if (value instanceof Promise) {\n      const mappedValue = await value;\n      seen.set(key, mappedValue);\n    }\n  }\n\n  // Normal deepMap again, this time filling in the resolved values.\n  // It's unfortunate that we have to do two passes.\n  // TODO(soergel): test performance and think harder about a fast solution.\n  const result = deepMapInternal(input, mapFn, seen);\n  return result;\n}\n\n/**\n * Determine whether the argument is iterable.\n *\n * @returns true if the argument is an array or any non-Tensor object.\n */\n// tslint:disable-next-line:no-any\nexport function isIterable(obj: any): boolean {\n  return obj != null && (!ArrayBuffer.isView(obj)) &&\n      (Array.isArray(obj) ||\n       (typeof obj === 'object' && !(obj instanceof tf.Tensor)));\n}\n\n/**\n * Determine whether the argument can be converted to Tensor.\n *\n * Tensors, primitives, arrays, and TypedArrays all qualify; anything else does\n * not.\n *\n * @returns true if the argument can be converted to Tensor.\n */\n// tslint:disable-next-line:no-any\nexport function canTensorify(obj: any): boolean {\n  return obj == null || isPrimitive(obj) || Array.isArray(obj) ||\n      (typeof obj === 'object' && (obj instanceof tf.Tensor)) ||\n      tf.util.isTypedArray(obj);\n}\n\n/**\n * Returns true if the given `value` is a primitive type. Otherwise returns\n * false. This is equivalant to node util.isPrimitive\n */\nfunction isPrimitive(value: any): boolean {\n  return (\n      value === null ||\n      (typeof value !== 'object' && typeof value !== 'function'));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {deepMap, DeepMapResult, isIterable} from './deep_map';\n\nexport function deepClone<T>(container: T): T {\n  return deepMap(container, cloneIfTensor);\n}\n\n// tslint:disable-next-line: no-any\nfunction cloneIfTensor(item: any): DeepMapResult {\n  if (item instanceof tf.Tensor) {\n    return ({value: item.clone(), recurse: false});\n  } else if (isIterable(item)) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: item, recurse: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/**\n * A ring buffer, providing O(1) FIFO, LIFO, and related operations.\n */\nexport class RingBuffer<T> {\n  // Note we store the indices in the range 0 <= index < 2*capacity.\n  // This allows us to distinguish the full from the empty case.\n  // See https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/\n  protected begin = 0;  // inclusive\n  protected end = 0;    // exclusive\n  protected doubledCapacity: number;\n\n  protected data: T[];\n\n  /**\n   * Constructs a `RingBuffer`.\n   * @param capacity The number of items that the buffer can accomodate.\n   */\n  constructor(public capacity: number) {\n    if (capacity == null) {\n      throw new RangeError('Can\\'t create a ring buffer of unknown capacity.');\n    }\n    if (capacity < 1) {\n      throw new RangeError('Can\\'t create ring buffer of capacity < 1.');\n    }\n    this.data = new Array<T>(capacity);\n    this.doubledCapacity = 2 * capacity;\n  }\n\n  /**\n   * Map any index into the range 0 <= index < 2*capacity.\n   */\n  protected wrap(index: number) {\n    // don't trust % on negative numbers\n    while (index < 0) {\n      index += this.doubledCapacity;\n    }\n    return index % this.doubledCapacity;\n  }\n\n  protected get(index: number) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t get item at a negative index.');\n    }\n    return this.data[index % this.capacity];\n  }\n\n  protected set(index: number, value: T) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t set item at a negative index.');\n    }\n    this.data[index % this.capacity] = value;\n  }\n\n  /**\n   * Returns the current number of items in the buffer.\n   */\n  length(): number {\n    let length = this.end - this.begin;\n    if (length < 0) {\n      length = this.doubledCapacity + length;\n    }\n    return length;\n  }\n\n  /**\n   * Reports whether the buffer is full.\n   * @returns true if the number of items in the buffer equals its capacity, and\n   *   false otherwise.\n   */\n  isFull() {\n    return this.length() === this.capacity;\n  }\n\n  /**\n   * Reports whether the buffer is empty.\n   * @returns true if the number of items in the buffer equals zero, and\n   *   false otherwise.\n   */\n  isEmpty() {\n    return this.length() === 0;\n  }\n\n  /**\n   * Adds an item to the end of the buffer.\n   */\n  push(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.set(this.end, value);\n    this.end = this.wrap(this.end + 1);\n  }\n\n  /**\n   * Adds many items to the end of the buffer, in order.\n   */\n  pushAll(values: T[]) {\n    for (const value of values) {\n      this.push(value);\n    }\n  }\n\n  /**\n   * Removes and returns the last item in the buffer.\n   */\n  pop(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    this.end = this.wrap(this.end - 1);\n    const result = this.get(this.end);\n    this.set(this.end, undefined);\n    return result;\n  }\n\n  /**\n   * Adds an item to the beginning of the buffer.\n   */\n  unshift(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.begin = this.wrap(this.begin - 1);\n    this.set(this.begin, value);\n  }\n\n  /**\n   * Removes and returns the first item in the buffer.\n   */\n  shift(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const result = this.get(this.begin);\n    this.set(this.begin, undefined);\n    this.begin = this.wrap(this.begin + 1);\n    return result;\n  }\n\n  /**\n   * Removes and returns a specific item in the buffer, and moves the last item\n   * to the vacated slot.  This is useful for implementing a shuffling stream.\n   * Note that this operation necessarily scrambles the original order.\n   *\n   * @param relativeIndex: the index of the item to remove, relative to the\n   *   first item in the buffer (e.g., hiding the ring nature of the underlying\n   *   storage).\n   */\n  shuffleExcise(relativeIndex: number): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const index = this.wrap(this.begin + relativeIndex);\n    const result = this.get(index);\n    this.set(index, this.pop());\n    return result;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {RingBuffer} from './ring_buffer';\n\nexport class GrowingRingBuffer<T> extends RingBuffer<T> {\n  private static INITIAL_CAPACITY = 32;\n\n  /**\n   * Constructs a `GrowingRingBuffer`.\n   */\n  constructor() {\n    super(GrowingRingBuffer.INITIAL_CAPACITY);\n  }\n\n  isFull() {\n    return false;\n  }\n\n  push(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.push(value);\n  }\n\n  unshift(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.unshift(value);\n  }\n\n  /**\n   * Doubles the capacity of the buffer.\n   */\n  private expand() {\n    const newCapacity = this.capacity * 2;\n    const newData = new Array<T>(newCapacity);\n    const len = this.length();\n\n    // Rotate the buffer to start at index 0 again, since we can't just\n    // allocate more space at the end.\n    for (let i = 0; i < len; i++) {\n      newData[i] = this.get(this.wrap(this.begin + i));\n    }\n\n    this.data = newData;\n    this.capacity = newCapacity;\n    this.doubledCapacity = 2 * this.capacity;\n    this.begin = 0;\n    this.end = len;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {Container} from '../types';\nimport {deepClone} from '../util/deep_clone';\nimport {deepMapAndAwaitAll, DeepMapAsyncResult, DeepMapResult, deepZip, zipToList} from '../util/deep_map';\nimport {GrowingRingBuffer} from '../util/growing_ring_buffer';\nimport {RingBuffer} from '../util/ring_buffer';\n\n/**\n * A nested structure of LazyIterators, used as the input to zip().\n */\nexport type IteratorContainer = Container<LazyIterator<tf.TensorContainer>>;\n\n// Here we implement a simple asynchronous iterator.\n// This lets us avoid using either third-party stream libraries or\n// recent TypeScript language support requiring polyfills.\n\n/**\n * Create a `LazyIterator` from an array of items.\n */\nexport function iteratorFromItems<T>(items: T[]): LazyIterator<T> {\n  return new ArrayIterator(items);\n}\n\n/**\n * Create a `LazyIterator` of incrementing integers.\n */\nexport function iteratorFromIncrementing(start: number): LazyIterator<number> {\n  let i = start;\n  return iteratorFromFunction(() => ({value: i++, done: false}));\n}\n\n/**\n * Create a `LazyIterator` from a function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * await iter.forEachAsync(e => console.log(e));\n * ```\n *\n * @param func A function that produces data on each call.\n */\nexport function iteratorFromFunction<T>(\n    func: () =>\n        IteratorResult<T>| Promise<IteratorResult<T>>): LazyIterator<T> {\n  return new FunctionCallIterator(func);\n}\n\n/**\n * Create a `LazyIterator` by concatenating underlying streams, which are\n * themselves provided as a stream.\n *\n * This can also be thought of as a \"stream flatten\" operation.\n *\n * @param baseIterators A stream of streams to be concatenated.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenated<T>(\n    baseIterators: LazyIterator<LazyIterator<T>>,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return new ChainedIterator(baseIterators, baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by concatenating streams produced by calling a\n * stream-generating function a given number of times.\n *\n * Since a `LazyIterator` is read-once, it cannot be repeated, but this\n * function can be used to achieve a similar effect:\n *\n *   LazyIterator.ofConcatenatedFunction(() => new MyIterator(), 6);\n *\n * @param iteratorFunc: A function that produces a new stream on each call.\n * @param count: The number of times to call the function.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenatedFunction<T>(\n    iteratorFunc: () => IteratorResult<LazyIterator<T>>, count: number,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return iteratorFromConcatenated(\n      iteratorFromFunction(iteratorFunc).take(count), baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by zipping together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nexport function iteratorFromZipped<O extends tf.TensorContainer>(\n    iterators: IteratorContainer,\n    mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL): LazyIterator<O> {\n  return new ZipIterator<O>(iterators, mismatchMode);\n}\n\n/**\n * An asynchronous iterator, providing lazy access to a potentially\n * unbounded stream of elements.\n *\n * Iterator can be obtained from a dataset:\n * `const iter = await dataset.iterator();`\n */\nexport abstract class LazyIterator<T> {\n  // This class implements AsyncIterator<T>, but we have not yet set the\n  // TypeScript --downlevelIteration flag to enable that.\n\n  abstract summary(): string;\n\n  /**\n   * Returns a `Promise` for the next element in the stream.\n   *\n   * When an item can be provided successfully, the return value is\n   * `{value:T, done:false}`.\n   *\n   * Calling next() on a closed stream returns `{value:null, done:true}`.\n   */\n  abstract async next(): Promise<IteratorResult<T>>;\n\n  /**\n   * Collect all remaining elements of a bounded stream into an array.\n   * Obviously this will succeed only for small streams that fit in memory.\n   * Useful for testing.\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArray(): Promise<T[]> {\n    const result: T[] = [];\n    let x = await this.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await this.next();\n    }\n    return result;\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArrayForTest(): Promise<T[]> {\n    const stream = this.prefetch(100);\n    const result: T[] = [];\n    let x = await stream.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await stream.next();\n    }\n    return result;\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveFully(): Promise<void> {\n    let x = await this.next();\n    while (!x.done) {\n      x = await this.next();\n    }\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted, or a predicate fails.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveWhile(predicate: (r: T) => boolean): Promise<void> {\n    let x = await this.next();\n    let shouldContinue = predicate(x.value);\n    while ((!x.done) && shouldContinue) {\n      x = await this.next();\n      shouldContinue = predicate(x.value);\n    }\n  }\n\n  /**\n   * Handles errors thrown on this stream using a provided handler function.\n   *\n   * @param handler A function that handles any `Error` thrown during a `next()`\n   *   call and returns true if the stream should continue (dropping the failed\n   *   call) or false if the stream should quietly terminate.  If the handler\n   *   itself throws (or rethrows) an `Error`, that will be propagated.\n   *\n   * @returns A `LazyIterator` of elements passed through from upstream,\n   *   possibly filtering or terminating on upstream `next()` calls that\n   *   throw an `Error`.\n   */\n  handleErrors(handler: (error: Error) => boolean): LazyIterator<T> {\n    return new ErrorHandlingLazyIterator(this, handler);\n  }\n\n  // TODO(soergel): Implement reduce() etc.\n\n  /**\n   * Filters this stream according to `predicate`.\n   *\n   * @param predicate A function mapping a stream element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `LazyIterator` of elements for which the predicate was true.\n   */\n  filter(predicate: (value: T) => boolean): LazyIterator<T> {\n    return new FilterIterator(this, predicate);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  map<O>(transform: (value: T) => O): LazyIterator<O> {\n    return new MapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through an async 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a `Promise` for a\n   *   transformed stream element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  mapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform, forcing serial execution.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  serialMapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform).serial();\n  }\n\n  /**\n   * Maps this stream through a 1-to-many transform.\n   *\n   * @param transform A function mapping a stream element to an array of\n   *   transformed elements.\n   *\n   * @returns A `DataStream` of transformed elements.\n   */\n  flatmap<O>(transform: (value: T) => O[]): LazyIterator<O> {\n    return new FlatmapIterator(this, transform);\n  }\n\n  /**\n   * Apply a function to every element of the stream.\n   *\n   * @param f A function to apply to each stream element.\n   */\n  async forEachAsync(f: (value: T) => void): Promise<void> {\n    return this.map(f).resolveFully();\n  }\n\n  /**\n   * Apply a function to every element of the stream, forcing serial execution.\n   *\n   * @param f A function to apply to each stream element.  Should return 'true'\n   *   to indicate that the stream should continue, or 'false' to cause it to\n   *   terminate.\n   */\n  async serialForEach(f: (value: T) => Promise<boolean>): Promise<void> {\n    return this.serialMapAsync(f).resolveWhile(x => (x === true));\n  }\n\n  /**\n   * Groups elements into batches, represented as arrays of elements.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"Row-major\" means that the resulting batch is simply a collection of\n   * rows: `[row1, row2, row3, ...]`.  This is contrast to the column-major\n   * form, which is needed for vectorized computation.\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `LazyIterator` of batches of elements, represented as arrays\n   *   of the original element type.\n   */\n  rowMajorBatch(batchSize: number, smallLastBatch = true): LazyIterator<T[]> {\n    return new RowMajorBatchIterator(this, batchSize, smallLastBatch);\n  }\n\n  /**\n   * Groups elements into batches, represented in column-major form.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"column-major\" means that the resulting batch is a (potentially\n   * nested) structure representing the columns.  Each column entry, then,\n   * contains a collection of the values found in that column for a range of\n   * input elements.  This representation allows for vectorized computation, in\n   * contrast to the row-major form.\n   *\n   * The inputs should all have the same nested structure (i.e., of arrays and\n   * dicts).  The result is a single object with the same nested structure,\n   * where the leaves are arrays collecting the values of the inputs at that\n   * location (or, optionally, the result of a custom function applied to those\n   * arrays).\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @param zipFn: (optional) A function that expects an array of elements at a\n   *   single node of the object tree, and returns a `DeepMapResult`.  The\n   *   `DeepMapResult` either provides a result value for that node (i.e.,\n   *   representing the subtree), or indicates that the node should be processed\n   *   recursively.  The default zipFn recurses as far as possible and places\n   *   arrays at the leaves.\n   * @returns A `LazyIterator` of batches of elements, represented as an object\n   *   with collections at the leaves.\n   */\n  columnMajorBatch(\n      batchSize: number, smallLastBatch = true,\n      // tslint:disable-next-line:no-any\n      zipFn: (xs: any[]) => DeepMapResult = zipToList):\n      LazyIterator<tf.TensorContainer> {\n    // First collect the desired number of input elements as a row-major batch.\n    const rowBatches = this.rowMajorBatch(batchSize, smallLastBatch);\n    // Now 'rotate' or 'pivot' the data, collecting all values from each column\n    // in the batch (i.e., for each key within the elements) into an array.\n    return rowBatches.map(x => deepZip(x, zipFn));\n  }\n\n  /**\n   * Concatenate this `LazyIterator` with another.\n   *\n   * @param iterator A `LazyIterator` to be concatenated onto this one.\n   * @param baseErrorHandler An optional function that can intercept `Error`s\n   *   raised during a `next()` call on the base stream.  This function can\n   *   decide whether the error should be propagated, whether the error should\n   *   be ignored, or whether the base stream should be terminated.\n   * @returns A `LazyIterator`.\n   */\n  concatenate(\n      iterator: LazyIterator<T>,\n      baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n    return new ChainedIterator(\n        iteratorFromItems([this, iterator]), baseErrorHandler);\n  }\n\n  /**\n   * Limits this stream to return at most `count` items.\n   *\n   * @param count The maximum number of items to provide from the stream. If\n   * a negative or undefined value is given, the entire stream is returned\n   *   unaltered.\n   */\n  take(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new TakeIterator(this, count);\n  }\n\n  /**\n   * Skips the first `count` items in this stream.\n   *\n   * @param count The number of items to skip.  If a negative or undefined\n   * value is given, the entire stream is returned unaltered.\n   */\n  skip(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new SkipIterator(this, count);\n  }\n\n  /**\n   * Prefetch the first `bufferSize` items in this stream.\n   *\n   * Note this prefetches Promises, but makes no guarantees about when those\n   * Promises resolve.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   */\n  prefetch(bufferSize: number): LazyIterator<T> {\n    return new PrefetchIterator(this, bufferSize);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  /**\n   * Randomly shuffles the elements of this stream.\n   *\n   * @param bufferSize: An integer specifying the number of elements from\n   * this stream from which the new stream will sample.\n   * @param seed: (Optional.) An integer specifying the random seed that\n   * will be used to create the distribution.\n   */\n  shuffle(windowSize: number, seed?: string): LazyIterator<T> {\n    return new ShuffleIterator(this, windowSize, seed);\n  }\n\n  /**\n   * Force an iterator to execute serially: each next() call will await the\n   * prior one, so that they cannot execute concurrently.\n   */\n  serial(): LazyIterator<T> {\n    return new SerialIterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on LazyIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// Iterators that just extend LazyIterator directly\n// ============================================================================\n\nclass ArrayIterator<T> extends LazyIterator<T> {\n  private trav = 0;\n  constructor(protected items: T[]) {\n    super();\n  }\n\n  summary() {\n    return `Array of ${this.items.length} items`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.trav >= this.items.length) {\n      return {value: null, done: true};\n    }\n    const item = this.items[this.trav];\n    this.trav++;\n    return {value: deepClone(item), done: false};\n  }\n}\n\nclass FunctionCallIterator<T> extends LazyIterator<T> {\n  constructor(\n      protected nextFn: () => IteratorResult<T>| Promise<IteratorResult<T>>) {\n    super();\n  }\n\n  summary() {\n    return `Function call`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    try {\n      return this.nextFn();\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message =\n          `Error thrown while iterating through a dataset: ${e.message}`;\n      throw e;\n    }\n  }\n}\n\nclass SerialIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(protected upstream: LazyIterator<T>) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Serial`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    return this.upstream.next();\n  }\n}\n\nclass SkipIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  count = 0;\n\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Skip`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider tradeoffs of reading in parallel, eg.\n    // collecting next() promises in an Array and then waiting for\n    // Promise.all() of those. Benefit: pseudo-parallel execution.  Drawback:\n    // maybe delayed GC.\n    while (this.count++ < this.maxCount) {\n      const skipped = await this.upstream.next();\n      // short-circuit if upstream is already empty\n      if (skipped.done) {\n        return skipped;\n      }\n      tf.dispose(skipped.value as {});\n    }\n    return this.upstream.next();\n  }\n}\n\nclass TakeIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Take`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.count++ >= this.maxCount) {\n      return {value: null, done: true};\n    }\n    return this.upstream.next();\n  }\n}\n\n// Note this batch just groups items into row-wise element arrays.\n// Rotating these to a column-wise representation happens only at the dataset\n// level.\nclass RowMajorBatchIterator<T> extends LazyIterator<T[]> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T[]>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected batchSize: number,\n      protected enableSmallLastBatch = true) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> RowMajorBatch`;\n  }\n\n  async next(): Promise<IteratorResult<T[]>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T[]>> {\n    const batch: T[] = [];\n    while (batch.length < this.batchSize) {\n      const item = await this.upstream.next();\n      if (item.done) {\n        if (this.enableSmallLastBatch && batch.length > 0) {\n          return {value: batch, done: false};\n        }\n        return {value: null, done: true};\n      }\n      batch.push(item.value);\n    }\n    return {value: batch, done: false};\n  }\n}\n\nclass FilterIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected predicate: (value: T) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Filter`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      const item = await this.upstream.next();\n      if (item.done || this.predicate(item.value)) {\n        return item;\n      }\n      tf.dispose(item.value as {});\n    }\n  }\n}\n\nclass MapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Map`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\nclass ErrorHandlingLazyIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected handler: (error: Error) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> handleErrors`;\n  }\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      try {\n        return await this.upstream.next();\n      } catch (e) {\n        if (!this.handler(e)) {\n          return {value: null, done: true};\n        }\n        // If the handler returns true, loop and fetch the next upstream item.\n\n        // If the upstream iterator throws an endless stream of errors, and if\n        // the handler says to ignore them, then we loop forever here.  That is\n        // the correct behavior-- it's up to the handler to decide when to stop.\n      }\n    }\n  }\n}\n\nclass AsyncMapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => Promise<O>) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> AsyncMap`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = await this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\n// Iterators that maintain a queue of pending items\n// ============================================================================\n\n/**\n * A base class for transforming streams that operate by maintaining an\n * output queue of elements that are ready to return via next().  This is\n * commonly required when the transformation is 1-to-many:  A call to next()\n * may trigger a call to the underlying stream, which will produce many\n * mapped elements of this stream-- of which we need to return only one, so\n * we have to queue the rest.\n */\nexport abstract class OneToManyIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  protected outputQueue: RingBuffer<T>;\n\n  constructor() {\n    super();\n    this.outputQueue = new GrowingRingBuffer<T>();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  /**\n   * Read one or more chunks from upstream and process them, possibly\n   * reading or writing a carryover, and adding processed items to the\n   * output queue.  Note it's possible that no items are added to the queue\n   * on a given pump() call, even if the upstream stream is not closed\n   * (e.g., because items are filtered).\n   *\n   * @return `true` if any action was taken, i.e. fetching items from the\n   *   upstream source OR adding items to the output queue.  `false` if the\n   *   upstream source is exhausted AND nothing was added to the queue\n   * (i.e., any remaining carryover).\n   */\n  protected abstract async pump(): Promise<boolean>;\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // Fetch so that the queue contains at least one item if possible.\n    // If the upstream source is exhausted, AND there are no items left in\n    // the output queue, then this stream is also exhausted.\n    while (this.outputQueue.length() === 0) {\n      // TODO(soergel): consider parallel reads.\n      if (!await this.pump()) {\n        return {value: null, done: true};\n      }\n    }\n    return {value: this.outputQueue.shift(), done: false};\n  }\n}\nclass FlatmapIterator<I, O> extends OneToManyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O[]) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Flatmap`;\n  }\n\n  async pump(): Promise<boolean> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return false;\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // that's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying any\n    // intermediate Tensors.  Here we are concerned only about the inputs.\n    const mappedArray = this.transform(item.value);\n    const outputTensors =\n        tf.tensor_util.getTensorsInContainer(mappedArray as {});\n    this.outputQueue.pushAll(mappedArray);\n\n    // TODO(soergel) faster intersection, and deduplicate outputTensors\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n\n    return true;\n  }\n}\n\n/**\n * Provides a `LazyIterator` that concatenates a stream of underlying\n * streams.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n */\nexport class ChainedIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>> = null;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private iterator: LazyIterator<T> = null;\n  private moreIterators: LazyIterator<LazyIterator<T>>;\n\n  constructor(\n      iterators: LazyIterator<LazyIterator<T>>,\n      private readonly baseErrorHandler?: (e: Error) => boolean) {\n    super();\n    this.moreIterators = iterators;\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of chained summaries';\n    return `${upstreamSummaries} -> Chained`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    this.lastRead = this.readFromChain(this.lastRead);\n    return this.lastRead;\n  }\n\n  private async readFromChain(lastRead: Promise<IteratorResult<T>>):\n      Promise<IteratorResult<T>> {\n    // Must await on the previous read since the previous read may have advanced\n    // the stream of streams, from which we need to read.\n    // This is unfortunate since we can't parallelize reads. Which means\n    // prefetching of chained streams is a no-op.\n    // One solution is to prefetch immediately upstream of this.\n    await lastRead;\n    if (this.iterator == null) {\n      const iteratorResult = await this.moreIterators.next();\n      if (iteratorResult.done) {\n        // No more streams to stream from.\n        return {value: null, done: true};\n      }\n      this.iterator = iteratorResult.value;\n      if (this.baseErrorHandler != null) {\n        this.iterator = this.iterator.handleErrors(this.baseErrorHandler);\n      }\n    }\n    const itemResult = await this.iterator.next();\n    if (itemResult.done) {\n      this.iterator = null;\n      return this.readFromChain(lastRead);\n    }\n    return itemResult;\n  }\n}\n\nexport enum ZipMismatchMode {\n  FAIL,      // require zipped streams to have the same length\n  SHORTEST,  // terminate zip when the first stream is exhausted\n  LONGEST    // use nulls for exhausted streams; use up the longest stream.\n}\n\n/**\n * Provides a `LazyIterator` that zips together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nclass ZipIterator<O extends tf.TensorContainer> extends LazyIterator<O> {\n  private count = 0;\n  private currentPromise: Promise<IteratorResult<O>> = null;\n\n  constructor(\n      protected readonly iterators: IteratorContainer,\n      protected readonly mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL) {\n    super();\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of zip summaries';\n    return `{${upstreamSummaries}} -> Zip`;\n  }\n\n  private async nextState(afterState: Promise<IteratorResult<O>>):\n      Promise<IteratorResult<O>> {\n    // This chaining ensures that the underlying next() are not even called\n    // before the previous ones have resolved.\n    await afterState;\n\n    // Collect underlying iterator \"done\" signals as a side effect in\n    // getNext()\n    let numIterators = 0;\n    let iteratorsDone = 0;\n\n    function getNext(container: IteratorContainer): DeepMapAsyncResult {\n      if (container instanceof LazyIterator) {\n        const result = container.next();\n        return {\n          value: result.then(x => {\n            numIterators++;\n            if (x.done) {\n              iteratorsDone++;\n            }\n            return x.value;\n          }),\n          recurse: false\n        };\n      } else {\n        return {value: null, recurse: true};\n      }\n    }\n\n    const mapped: O = await deepMapAndAwaitAll(this.iterators, getNext);\n\n    if (numIterators === iteratorsDone) {\n      // The streams have all ended.\n      return {value: null, done: true};\n    }\n    if (iteratorsDone > 0) {\n      switch (this.mismatchMode) {\n        case ZipMismatchMode.FAIL:\n          throw new Error(\n              'Zipped streams should have the same length. ' +\n              `Mismatched at element ${this.count}.`);\n        case ZipMismatchMode.SHORTEST:\n          return {value: null, done: true};\n        case ZipMismatchMode.LONGEST:\n        default:\n          // Continue.  The exhausted streams already produced value: null.\n      }\n    }\n\n    this.count++;\n    return {value: mapped, done: false};\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    this.currentPromise = this.nextState(this.currentPromise);\n    return this.currentPromise;\n  }\n}\n\n// Iterators that maintain a ring buffer of pending promises\n// ============================================================================\n\n/**\n * A stream that prefetches a given number of items from an upstream source,\n * returning them in FIFO order.\n *\n * Note this prefetches Promises, but makes no guarantees about when those\n * Promises resolve.\n */\nexport class PrefetchIterator<T> extends LazyIterator<T> {\n  protected buffer: RingBuffer<Promise<IteratorResult<T>>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected bufferSize: number) {\n    super();\n    this.buffer = new RingBuffer<Promise<IteratorResult<T>>>(bufferSize);\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Prefetch`;\n  }\n\n  /**\n   * Refill the prefetch buffer.  Returns only after the buffer is full, or\n   * the upstream source is exhausted.\n   */\n  protected refill() {\n    while (!this.buffer.isFull()) {\n      const v = this.upstream.next();\n      this.buffer.push(v);\n    }\n  }\n\n  next(): Promise<IteratorResult<T>> {\n    this.refill();\n    // This shift will never throw an error because the buffer is always\n    // full after a refill. If the stream is exhausted, the buffer will be\n    // full of Promises that will resolve to the end-of-stream signal.\n    return this.buffer.shift();\n  }\n}\n\n/**\n * A stream that performs a sliding-window random shuffle on an upstream\n * source. This is like a `PrefetchIterator` except that the items are\n * returned in randomized order.  Mixing naturally improves as the buffer\n * size increases.\n */\nexport class ShuffleIterator<T> extends PrefetchIterator<T> {\n  private readonly random: seedrandom.prng;\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private upstreamExhausted = false;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected windowSize: number,\n      seed?: string) {\n    super(upstream, windowSize);\n    this.random = seedrandom.alea(seed || tf.util.now().toString());\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private randomInt(max: number) {\n    return Math.floor(this.random() * max);\n  }\n\n  protected chooseIndex(): number {\n    return this.randomInt(this.buffer.length());\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider performance\n    if (!this.upstreamExhausted) {\n      this.refill();\n    }\n    while (!this.buffer.isEmpty()) {\n      const chosenIndex = this.chooseIndex();\n      const result = await this.buffer.shuffleExcise(chosenIndex);\n      if (result.done) {\n        this.upstreamExhausted = true;\n      } else {\n        this.refill();\n        return result;\n      }\n    }\n    return {value: null, done: true};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {TensorContainer, TensorLike} from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {iteratorFromConcatenated, iteratorFromFunction, iteratorFromItems, iteratorFromZipped, LazyIterator, ZipMismatchMode} from './iterators/lazy_iterator';\nimport {Container} from './types';\nimport {canTensorify, deepMapAndAwaitAll, DeepMapResult, isIterable} from './util/deep_map';\n\n/**\n * A nested structure of Datasets, used as the input to zip().\n */\nexport type DatasetContainer = Container<Dataset<TensorContainer>>;\n\n// TODO(soergel): consider vectorized operations within the pipeline.\n\n/**\n * Represents a potentially large list of independent data elements (typically\n * 'samples' or 'examples').\n *\n * A 'data example' may be a primitive, an array, a map from string keys to\n * values, or any nested structure of these.\n *\n * A `Dataset` represents an ordered collection of elements, together with a\n * chain of transformations to be performed on those elements. Each\n * transformation is a method of `Dataset` that returns another `Dataset`, so\n * these may be chained, e.g.\n * `const processedDataset = rawDataset.filter(...).map(...).batch(...)`.\n *\n * Data loading and transformation is done in a lazy, streaming fashion.  The\n * dataset may be iterated over multiple times; each iteration starts the data\n * loading anew and recapitulates the transformations.\n *\n * A `Dataset` is typically processed as a stream of unbatched examples --i.e.,\n * its transformations are applied one example at a time. Batching produces a\n * new `Dataset` where each element is a batch. Batching should usually come\n * last in a pipeline, because data transformations are easier to express on a\n * per-example basis than on a per-batch basis.\n *\n * The following code examples are calling `await dataset.forEachAsync(...)` to\n * iterate once over the entire dataset in order to print out the data.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport abstract class Dataset<T extends tf.TensorContainer> {\n  /*\n   * Provide a new stream of elements.  Note this will also start new streams\n   * from any underlying `Dataset`s.\n   *\n   * CAUTION: Any Tensors contained within the elements returned from\n   * this stream *must* be manually disposed to avoid a GPU memory leak.\n   * The tf.tidy() approach cannot be used in an asynchronous context.\n   */\n  abstract async iterator(): Promise<LazyIterator<T>>;\n\n  readonly size: number = null;\n\n  // TODO(soergel): Make Datasets report whether repeated iterator() calls\n  // produce the same result (e.g., reading from a file) or different results\n  // (e.g., from the webcam).  Currently we don't make this distinction but it\n  // could be important for the user to know.\n  // abstract isDeterministic(): boolean;\n\n  /**\n   * Groups elements into batches.\n   *\n   * It is assumed that each of the incoming dataset elements has the same\n   * structure-- i.e. the same set of keys at each location in an object\n   * hierarchy.  For each key, the resulting `Dataset` provides a batched\n   * element collecting all of the incoming values for that key.\n   *\n   *  * Incoming primitives are grouped into a 1-D Tensor.\n   *  * Incoming Tensors are grouped into a new Tensor where the 0'th axis is\n   *    the batch dimension.\n   *  * Incoming arrays are converted to Tensor and then batched.\n   *  * A nested array is interpreted as an n-D Tensor, so the batched result\n   *    has n+1 dimensions.\n   *  * An array that cannot be converted to Tensor produces an error.\n   *\n   * If an array should not be batched as a unit, it should first be converted\n   * to an object with integer keys.\n   *\n   * Here are a few examples:\n   *\n   * Batch a dataset of numbers:\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8]).batch(4);\n   * await a.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of arrays:\n   * ```js\n   * const b = tf.data.array([[1], [2], [3], [4], [5], [6], [7], [8]]).batch(4);\n   * await b.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of objects:\n   * ```js\n   * const c = tf.data.array([{a: 1, b: 11}, {a: 2, b: 12}, {a: 3, b: 13},\n   *   {a: 4, b: 14}, {a: 5, b: 15}, {a: 6, b: 16}, {a: 7, b: 17},\n   *   {a: 8, b: 18}]).batch(4);\n   * await c.forEachAsync(e => {\n   *   console.log('{');\n   *   for(var key in e) {\n   *     console.log(key+':');\n   *     e[key].print();\n   *   }\n   *   console.log('}');\n   * })\n   * ```\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `Dataset`, from which a stream of batches can be obtained.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  batch(batchSize: number, smallLastBatch = true): Dataset<tf.TensorContainer> {\n    const base = this;\n    tf.util.assert(\n        batchSize > 0, () => `batchSize needs to be positive, but it is\n      ${batchSize}`);\n    let size;\n    if (this.size === Infinity || this.size == null) {\n      // If the size of this dataset is infinity or null, the new size keeps the\n      // same.\n      size = this.size;\n    } else if (smallLastBatch) {\n      // If the size of this dataset is known and include small last batch, the\n      // new size is full batch count plus last batch.\n      size = Math.ceil(this.size / batchSize);\n    } else {\n      // If the size of this dataset is known and not include small last batch,\n      // the new size is full batch count.\n      size = Math.floor(this.size / batchSize);\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator())\n          .columnMajorBatch(batchSize, smallLastBatch, deepBatchConcat);\n    }, size);\n  }\n\n  /**\n   * Concatenates this `Dataset` with another.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * const b = tf.data.array([4, 5, 6]);\n   * const c = a.concatenate(b);\n   * await c.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param dataset A `Dataset` to be concatenated onto this one.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  concatenate(dataset: Dataset<T>): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity || dataset.size === Infinity) {\n      // If the size of any of these two dataset is infinity, new size is\n      // infinity.\n      size = Infinity;\n    } else if (this.size != null && dataset.size != null) {\n      // If the size of both datasets are known and not infinity, new size is\n      // sum the size of these two datasets.\n      size = this.size + dataset.size;\n    } else {\n      // If neither of these two datasets has infinite size and any of these two\n      // datasets' size is null, the new size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () =>\n            (await base.iterator()).concatenate(await dataset.iterator()),\n        size);\n  }\n\n  /**\n   * Filters this dataset according to `predicate`.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])\n   *   .filter(x => x%2 === 0);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param predicate A function mapping a dataset element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `Dataset` of elements for which the predicate was true.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  filter(predicate: (value: T) => boolean): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity) {\n      // If the size of this dataset is infinity, new size is infinity\n      size = Infinity;\n    } else {\n      // If this dataset has limited elements, new size is null because it might\n      // exhausted randomly.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).filter(x => tf.tidy(() => predicate(x)));\n    }, size);\n  }\n\n  /**\n   * Apply a function to every element of the dataset.\n   *\n   * After the function is applied to a dataset element, any Tensors contained\n   * within that element are disposed.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param f A function to apply to each dataset element.\n   * @returns A `Promise` that resolves after all elements have been processed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async forEachAsync(f: (input: T) => void): Promise<void> {\n    return (await this.iterator()).forEachAsync(f);\n  }\n\n  /**\n   * Maps this dataset through a 1-to-1 transform.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).map(x => x*x);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a transformed\n   *   dataset element.\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  map<O extends tf.TensorContainer>(transform: (value: T) => O): Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).map(x => tf.tidy(() => transform(x)));\n    }, this.size);\n  }\n\n  /**\n   * Maps this dataset through an async 1-to-1 transform.\n   *\n   * ```js\n   * const a =\n   *  tf.data.array([1, 2, 3]).mapAsync(x => new Promise(function(resolve){\n   *    setTimeout(() => {\n   *      resolve(x * x);\n   *    }, Math.random()*1000 + 500);\n   *  }));\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a `Promise` for a\n   *   transformed dataset element.  This transform is responsible for disposing\n   *   any intermediate `Tensor`s, i.e. by wrapping its computation in\n   *   `tf.tidy()`; that cannot be automated here (as it is in the synchronous\n   *   `map()` case).\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  mapAsync<O extends tf.TensorContainer>(transform: (value: T) => Promise<O>):\n      Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).mapAsync(transform);\n    }, this.size);\n  }\n\n  /**\n   *  Creates a `Dataset` that prefetches elements from this dataset.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  prefetch(bufferSize: number): Dataset<T> {\n    if (bufferSize == null) {\n      throw new RangeError(\n          '`Dataset.prefetch()` requires bufferSize to be specified.');\n    }\n\n    const base = this;\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).prefetch(bufferSize), this.size);\n  }\n\n  /**\n   * Repeats this dataset `count` times.\n   *\n   * NOTE: If this dataset is a function of global state (e.g. a random number\n   * generator), then different repetitions may produce different elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).repeat(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: (Optional) An integer, representing the number of times\n   *   the dataset should be repeated. The default behavior (if `count` is\n   *   `undefined` or negative) is for the dataset be repeated indefinitely.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  repeat(count?: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count > 0) {\n      // If this dataset has size and count is positive, new size is current\n      // size multiply count. This also covers the case that current size is\n      // infinity.\n      size = this.size * count;\n    } else if (count === 0) {\n      // If count is 0, new size is 0.\n      size = 0;\n    } else if (this.size != null && (count === undefined || count < 0)) {\n      // If this dataset has size and count is undefined or negative, the\n      // dataset will be repeated indefinitely and new size is infinity.\n      size = Infinity;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      const iteratorIterator = iteratorFromFunction(\n          async () => ({value: await base.iterator(), done: false}));\n      return iteratorFromConcatenated(iteratorIterator.take(count));\n    }, size);\n  }\n\n  /**\n   * Creates a `Dataset` that skips `count` initial elements from this dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).skip(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be skipped\n   *   to form the new dataset.  If `count` is greater than the size of this\n   *   dataset, the new dataset will contain no elements.  If `count`\n   *   is `undefined` or negative, skips the entire dataset.\n   *\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  skip(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count >= 0 && this.size >= count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is current size minus skipped size.This also covers the case that\n      // current size is infinity.\n      size = this.size - count;\n    } else if (\n        this.size != null &&\n        (this.size < count || count === undefined || count < 0)) {\n      // If the size of this dataset is smaller than count, or count is\n      // undefined or negative, skips the entire dataset and the new size is 0.\n      size = 0;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).skip(count), size);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  static readonly MAX_BUFFER_SIZE = 10000;\n\n  /**\n   * Pseudorandomly shuffles the elements of this dataset. This is done in a\n   * streaming manner, by sampling from a given number of prefetched elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).shuffle(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param bufferSize: An integer specifying the number of elements from this\n   *   dataset from which the new dataset will sample.\n   * @param seed: (Optional) An integer specifying the random seed that will\n   *   be used to create the distribution.\n   * @param reshuffleEachIteration: (Optional) A boolean, which if true\n   *   indicates that the dataset should be pseudorandomly reshuffled each time\n   *   it is iterated over. If false, elements will be returned in the same\n   *   shuffled order on each iteration. (Defaults to `true`.)\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  shuffle(bufferSize: number, seed?: string, reshuffleEachIteration = true):\n      Dataset<T> {\n    if (bufferSize == null || bufferSize < 0) {\n      if (this.size == null) {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.');\n      } else {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.  ' +\n            'If your data fits in main memory (for regular JS objects), ' +\n            'and/or GPU memory (for `tf.Tensor`s), consider setting ' +\n            `bufferSize to the dataset size (${this.size} elements)`);\n      }\n    }\n    const base = this;\n    const random = seedrandom.alea(seed || tf.util.now().toString());\n    return datasetFromIteratorFn(async () => {\n      let seed2 = random.int32();\n      if (reshuffleEachIteration) {\n        seed2 += random.int32();\n      }\n      return (await base.iterator()).shuffle(bufferSize, seed2.toString());\n    }, this.size);\n  }\n\n  /**\n   * Creates a `Dataset` with at most `count` initial elements from this\n   * dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).take(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be taken\n   *   to form the new dataset.  If `count` is `undefined` or negative, or if\n   *   `count` is greater than the size of this dataset, the new dataset will\n   *   contain all elements of this dataset.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  take(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && this.size > count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is count.\n      size = count;\n    } else if (this.size != null && this.size <= count) {\n      // If the size of this dataset is equal or smaller than count, the new\n      // dataset's size is the size of this dataset.\n      size = this.size;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).take(count), size);\n  }\n\n  /**\n   * Collect all elements of this dataset into an array.\n   *\n   * Obviously this will succeed only for small datasets that fit in memory.\n   * Useful for testing and generally should be avoided if possible.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]);\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async toArray() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArray();\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   */\n  async toArrayForTest() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArrayForTest();\n  }\n}\n\n/**\n * Create a `Dataset` defined by a provided iterator() function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * const ds = tf.data.datasetFromIteratorFn(iter);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n */\nexport function datasetFromIteratorFn<T extends tf.TensorContainer>(\n    iteratorFn: () => Promise<LazyIterator<T>>,\n    size: number = null): Dataset<T> {\n  return new class extends Dataset<T> {\n    size = size;\n\n    /*\n     * Provide a new stream of elements.  Note this will also start new streams\n     * from any underlying `Dataset`s.\n     */\n    async iterator(): Promise<LazyIterator<T>> {\n      return iteratorFn();\n    }\n  }\n  ();\n}\n\n/**\n * Create a `Dataset` from an array of elements.\n *\n * Create a Dataset from an array of objects:\n * ```js\n * const a = tf.data.array([{'item': 1}, {'item': 2}, {'item': 3}]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n *\n * Create a Dataset from an array of numbers:\n * ```js\n * const a = tf.data.array([4, 5, 6]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n * @param items An array of elements that will be parsed as items in a dataset.\n *\n * @doc {heading: 'Data', subheading: 'Creation', namespace: 'data'}\n */\nexport function array<T extends tf.TensorContainer>(items: T[]): Dataset<T> {\n  return datasetFromIteratorFn(\n      async () => iteratorFromItems(items), items.length);\n}\n\n/**\n * Create a `Dataset` by zipping together an array, dict, or nested\n * structure of `Dataset`s (and perhaps additional constants).\n * The underlying datasets must provide elements in a consistent order such that\n * they correspond.\n *\n * The number of elements in the resulting dataset is the same as the size of\n * the smallest dataset in datasets.\n *\n * The nested structure of the `datasets` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Note this means that, given an array of two datasets that produce dict\n * elements, the result is a dataset that produces elements that are arrays\n * of two dicts:\n *\n * Zip an array of datasets:\n * ```js\n * console.log('Zip two datasets of objects:');\n * const ds1 = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const ds2 = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const ds3 = tf.data.zip([ds1, ds2]);\n * await ds3.forEachAsync(e => console.log(JSON.stringify(e)));\n *\n * // If the goal is to merge the dicts in order to produce elements like\n * // {a: ..., b: ...}, this requires a second step such as:\n * console.log('Merge the objects:');\n * const ds4 = ds3.map(x => {return {a: x[0].a, b: x[1].b}});\n * await ds4.forEachAsync(e => console.log(e));\n * ```\n *\n * Zip a dict of datasets:\n * ```js\n * const a = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const b = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const c = tf.data.zip({c: a, d: b});\n * await c.forEachAsync(e => console.log(JSON.stringify(e)));\n * ```\n *\n * @doc {heading: 'Data', subheading: 'Operations', namespace: 'data'}\n */\nexport function zip<O extends tf.TensorContainer>(datasets: DatasetContainer):\n    Dataset<O> {\n  // manually type-check the argument for JS users\n  if (!isIterable(datasets)) {\n    throw new Error('The argument to zip() must be an object or array.');\n  }\n  let size;\n  if (Array.isArray(datasets)) {\n    for (let i = 0; i < datasets.length; i++) {\n      size = size == null ? (datasets[i] as Dataset<O>).size :\n                            Math.min(size, (datasets[i] as Dataset<O>).size);\n    }\n  } else if (datasets instanceof Object) {\n    for (const ds in datasets) {\n      size = size == null ? (datasets[ds] as Dataset<O>).size :\n                            Math.min(size, (datasets[ds] as Dataset<O>).size);\n    }\n  }\n  return datasetFromIteratorFn<O>(async () => {\n    const streams = await deepMapAndAwaitAll(datasets, d => {\n      if (d instanceof Dataset) {\n        return {value: d.iterator(), recurse: false};\n      } else if (isIterable(d)) {\n        return {value: null, recurse: true};\n      } else {\n        throw new Error(\n            'Leaves of the structure passed to zip() must be Datasets, ' +\n            'not primitives.');\n      }\n    });\n    return iteratorFromZipped<O>(streams, ZipMismatchMode.SHORTEST);\n  }, size);\n}\n\n/**\n * A zip function for use with deepZip, passed via the columnMajorBatch call.\n *\n * Accepts an array of identically-structured nested elements and either batches\n * them (if they are primitives, numeric arrays, or Tensors) or requests\n * recursion (if not).\n */\n// tslint:disable-next-line:no-any\nfunction deepBatchConcat(rows: any[]): DeepMapResult {\n  if (rows === null) {\n    return null;\n  }\n\n  // use the first item to decide whether to recurse or batch here.\n  const exampleRow = rows[0];\n\n  if (canTensorify(exampleRow)) {\n    // rows is an array of primitives, Tensors, or arrays.  Batch them.\n    const value = batchConcat(rows);\n    return {value, recurse: false};\n  }\n\n  // the example row is an object, so recurse into it.\n  return {value: null, recurse: true};\n}\n\n/**\n * Assembles a list of same-shaped numbers, number arrays, or Tensors\n * into a single new Tensor where axis 0 is the batch dimension.\n */\nfunction batchConcat<T extends(TensorLike | tf.Tensor)>(arrays: T[]):\n    tf.Tensor {\n  if (arrays.length === 0) {\n    // We can't return an empty Tensor because we don't know the element shape.\n    throw new Error('Can\\'t make a batch of zero elements.');\n  }\n\n  if (arrays[0] instanceof tf.Tensor) {\n    // Input is an array of Tensors\n    return tf.stack(arrays as tf.Tensor[]);\n  } else {\n    // Input is a possibly-nested array of numbers.\n    return tf.tensor(arrays as TensorLike);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\n\n/**\n * Represents a potentially large collection of text lines.\n *\n * The results are not batched.\n */\nexport class TextLineDataset extends Dataset<string> {\n  /**\n   * Create a `TextLineDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   */\n  constructor(protected readonly input: DataSource) {\n    super();\n  }\n\n  async iterator(): Promise<LazyIterator<string>> {\n    const inputIterator = await this.input.iterator();\n    const utf8Iterator = inputIterator.decodeUTF8();\n    const lineIterator = utf8Iterator.split('\\n').map(line => {\n      // Windows/DOS format text file has extra line breaker at the end of line.\n      if (line.endsWith('\\r')) {\n        line = line.slice(0, -1);\n      }\n      return line;\n    });\n    return lineIterator;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\nimport {ColumnConfig, CSVConfig} from '../types';\nimport {TextLineDataset} from './text_line_dataset';\n\nconst CODE_QUOTE = '\"';\nconst STATE_OUT = Symbol('out');\nconst STATE_FIELD = Symbol('field');\nconst STATE_QUOTE = Symbol('quote');\nconst STATE_QUOTE_AFTER_QUOTE = Symbol('quoteafterquote');\nconst STATE_WITHIN_QUOTE_IN_QUOTE = Symbol('quoteinquote');\n\n/**\n * Represents a potentially large collection of delimited text records.\n *\n * The produced `TensorContainer`s each contain one key-value pair for\n * every column of the table.  When a field is empty in the incoming data, the\n * resulting value is `undefined`, or throw error if it is required.  Values\n * that can be parsed as numbers are emitted as type `number`, other values\n * are parsed as `string`.\n *\n * The results are not batched.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport class CSVDataset extends Dataset<TensorContainer> {\n  base: TextLineDataset;\n  private hasHeader = true;\n  private fullColumnNames: string[] = null;\n  private columnNamesValidated = false;\n  private columnConfigs: {[key: string]: ColumnConfig} = null;\n  private configuredColumnsOnly = false;\n  private delimiter = ',';\n  private delimWhitespace = false;\n\n  /**\n   * Returns column names of the csv dataset. If `configuredColumnsOnly` is\n   * true, return column names in `columnConfigs`. If `configuredColumnsOnly` is\n   * false and `columnNames` is provided, `columnNames`. If\n   * `configuredColumnsOnly` is false and `columnNames` is not provided, return\n   * all column names parsed from the csv file. For example usage please go to\n   * `tf.data.csv`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async columnNames() {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    return this.configuredColumnsOnly ? Object.keys(this.columnConfigs) :\n                                        this.fullColumnNames;\n  }\n\n  /* 1) If `columnNames` is provided as string[], use this string[] as output\n   * keys in corresponding order. The length must match the number of inferred\n   * columns if `hasHeader` is true .\n   * 2) If `columnNames` is not provided, parse header line as `columnNames` if\n   * hasHeader is true. If `hasHeader` is false, throw an error.\n   * 3) If `columnConfigs` is provided, all the keys in `columnConfigs` must\n   * exist in parsed `columnNames`.\n   */\n  private async setColumnNames() {\n    const columnNamesFromFile = await this.maybeReadHeaderLine();\n    if (!this.fullColumnNames && !columnNamesFromFile) {\n      // Throw an error if columnNames is not provided and no header line.\n      throw new Error(\n          'Column names must be provided if there is no header line.');\n    } else if (this.fullColumnNames && columnNamesFromFile) {\n      // Check provided columnNames match header line.\n      util.assert(\n          columnNamesFromFile.length === this.fullColumnNames.length,\n          () => 'The length of provided columnNames (' +\n              this.fullColumnNames.length.toString() +\n              ') does not match the length of the header line read from ' +\n              'file (' + columnNamesFromFile.length.toString() + ').');\n    }\n    if (!this.fullColumnNames) {\n      this.fullColumnNames = columnNamesFromFile;\n    }\n    // Check if there are duplicate column names.\n    const counts: {[key: string]: number} = this.fullColumnNames.reduce(\n        (countAcc: {[key: string]: number}, name) => {\n          countAcc[name] = (countAcc[name] + 1) || 1;\n          return countAcc;\n        },\n        {});\n    const duplicateNames =\n        Object.keys(counts).filter((name) => (counts[name] > 1));\n    util.assert(\n        duplicateNames.length === 0,\n        () => 'Duplicate column names found: ' + duplicateNames.toString());\n    // Check if keys in columnConfigs match columnNames.\n    if (this.columnConfigs) {\n      for (const key of Object.keys(this.columnConfigs)) {\n        const index = this.fullColumnNames.indexOf(key);\n        if (index === -1) {\n          throw new Error(\n              'The key \"' + key +\n              '\" provided in columnConfigs does not match any of the column ' +\n              'names (' + this.fullColumnNames.toString() + ').');\n        }\n      }\n    }\n    this.columnNamesValidated = true;\n  }\n\n  private async maybeReadHeaderLine() {\n    if (this.hasHeader) {\n      const iter = await this.base.iterator();\n      const firstElement = await iter.next();\n      if (firstElement.done) {\n        throw new Error('No data was found for CSV parsing.');\n      }\n      const firstLine: string = firstElement.value;\n      const headers = this.parseRow(firstLine, false);\n      return headers;\n    } else {\n      return null;\n    }\n  }\n\n  /**\n   * Create a `CSVDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   * @param csvConfig (Optional) A CSVConfig object that contains configurations\n   *     of reading and decoding from CSV file(s).\n   *\n   *     hasHeader: (Optional) A boolean value that indicates whether the first\n   *     row of provided CSV file is a header line with column names, and should\n   *     not be included in the data. Defaults to `true`.\n   *\n   *     columnNames: (Optional) A list of strings that corresponds to\n   *     the CSV column names, in order. If provided, it ignores the column\n   *     names inferred from the header row. If not provided, infers the column\n   *     names from the first row of the records. If hasHeader is false and\n   *     columnNames is not provided, this method throws an error.\n   *\n   *     columnConfigs: (Optional) A dictionary whose key is column names, value\n   *     is an object stating if this column is required, column's data type,\n   *     default value, and if this column is label. If provided, keys must\n   *     correspond to names provided in columnNames or inferred from the file\n   *     header lines. If isLabel is true any column, returns an array of two\n   *     items: the first item is a dict of features key/value pairs, the second\n   *     item is a dict of labels key/value pairs. If no feature is marked as\n   *     label, returns a dict of features only.\n   *\n   *     configuredColumnsOnly (Optional) If true, only columns provided in\n   *     columnConfigs will be parsed and provided during iteration.\n   *\n   *     delimiter (Optional) The string used to parse each line of the input\n   *     file. Defaults to `,`.\n   */\n  constructor(protected readonly input: DataSource, csvConfig?: CSVConfig) {\n    super();\n    this.base = new TextLineDataset(input);\n    if (!csvConfig) {\n      csvConfig = {};\n    }\n    this.hasHeader = csvConfig.hasHeader === false ? false : true;\n    this.fullColumnNames = csvConfig.columnNames;\n    this.columnConfigs = csvConfig.columnConfigs;\n    this.configuredColumnsOnly = csvConfig.configuredColumnsOnly;\n    if (csvConfig.delimWhitespace) {\n      util.assert(\n          csvConfig.delimiter == null,\n          () =>\n              'Delimiter should not be provided when delimWhitespace is true.');\n      this.delimWhitespace = true;\n      this.delimiter = ' ';\n    } else {\n      this.delimiter = csvConfig.delimiter ? csvConfig.delimiter : ',';\n    }\n  }\n\n  async iterator(): Promise<LazyIterator<TensorContainer>> {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    let lines = await this.base.iterator();\n    if (this.hasHeader) {\n      // We previously read the first line to get the columnNames.\n      // Now that we're providing data, skip it.\n      lines = lines.skip(1);\n    }\n    return lines.map(x => this.makeDataElement(x));\n  }\n\n  makeDataElement(line: string): TensorContainer {\n    const values = this.parseRow(line);\n    const features: {[key: string]: TensorContainer} = {};\n    const labels: {[key: string]: TensorContainer} = {};\n\n    for (let i = 0; i < this.fullColumnNames.length; i++) {\n      const key = this.fullColumnNames[i];\n      const config = this.columnConfigs ? this.columnConfigs[key] : null;\n      if (this.configuredColumnsOnly && !config) {\n        // This column is not selected.\n        continue;\n      } else {\n        const value = values[i];\n        let parsedValue = null;\n        if (value === '') {\n          // If default value is provided, use it. If default value is not\n          // provided, set as undefined.\n          if (config && config.default !== undefined) {\n            parsedValue = config.default;\n          } else if (config && (config.required || config.isLabel)) {\n            throw new Error(\n                `Required column ${key} is empty in this line: ${line}`);\n          } else {\n            parsedValue = undefined;\n          }\n        } else {\n          // A value is present, so parse it based on type\n          const valueAsNum = Number(value);\n          if (isNaN(valueAsNum)) {\n            // The value is a string and this column is declared as boolean\n            // in config, parse it as boolean.\n            if (config && config.dtype === 'bool') {\n              parsedValue = this.getBoolean(value);\n            } else {\n              // Set value as string\n              parsedValue = value;\n            }\n          } else if (!config || !config.dtype) {\n            // If this value is a number and no type config is provided, return\n            // it as number.\n            parsedValue = valueAsNum;\n          } else {\n            // If this value is a number and data type is provided, parse it\n            // according to provided data type.\n            switch (config.dtype) {\n              case 'float32':\n                parsedValue = valueAsNum;\n                break;\n              case 'int32':\n                parsedValue = Math.floor(valueAsNum);\n                break;\n              case 'bool':\n                parsedValue = this.getBoolean(value);\n                break;\n              default:\n                parsedValue = valueAsNum;\n            }\n          }\n        }\n        // Check if this column is label.\n        (config && config.isLabel) ? labels[key] = parsedValue :\n                                     features[key] = parsedValue;\n      }\n    }\n    // If label exists, return an object of features and labels as {xs:features,\n    // ys:labels}, otherwise return features only.\n    if (Object.keys(labels).length === 0) {\n      return features;\n\n    } else {\n      return {xs: features, ys: labels};\n    }\n  }\n\n  private getBoolean(value: string): number {\n    if (value === '1' || value.toLowerCase() === 'true') {\n      return 1;\n    } else {\n      return 0;\n    }\n  }\n\n  // adapted from https://beta.observablehq.com/@mbostock/streaming-csv\n  private parseRow(line: string, validateElementCount = true): string[] {\n    const result: string[] = [];\n    let readOffset = 0;\n    const readLength = line.length;\n    let currentState = STATE_OUT;\n    // Goes through the line to parse quote.\n    for (let i = 0; i < readLength; i++) {\n      switch (currentState) {\n        // Before enter a new field\n        case STATE_OUT:\n          switch (line.charAt(i)) {\n            // Enter a quoted field\n            case CODE_QUOTE:\n              readOffset = i + 1;\n              currentState = STATE_QUOTE;\n              break;\n            // Read an empty field\n            case this.delimiter:\n              readOffset = i + 1;\n              // If delimiter is white space and configured to collapse\n              // multiple white spaces, ignore this white space.\n              if (this.delimiter === ' ' && this.delimWhitespace) {\n                break;\n              }\n              result.push('');\n              currentState = STATE_OUT;\n              break;\n            // Enter an unquoted field\n            default:\n              currentState = STATE_FIELD;\n              readOffset = i;\n              break;\n          }\n          break;\n        // In an unquoted field\n        case STATE_FIELD:\n          switch (line.charAt(i)) {\n            // Exit an unquoted field, add it to result\n            case this.delimiter:\n              result.push(line.substring(readOffset, i));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            default:\n          }\n          break;\n        // In a quoted field\n        case STATE_QUOTE:\n          switch (line.charAt(i)) {\n            // Read a quote after a quote\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE_AFTER_QUOTE;\n              break;\n            default:\n          }\n          break;\n        // This state means it's right after a second quote in a field\n        case STATE_QUOTE_AFTER_QUOTE:\n          switch (line.charAt(i)) {\n            // Finished a quoted field\n            case this.delimiter:\n              result.push(line.substring(readOffset, i - 1));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            // Finished a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            // In a quoted part in a quoted field\n            default:\n              currentState = STATE_WITHIN_QUOTE_IN_QUOTE;\n              break;\n          }\n          break;\n        case STATE_WITHIN_QUOTE_IN_QUOTE:\n          switch (line.charAt(i)) {\n            // Exit a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            default:\n          }\n          break;\n        default:\n      }\n    }\n    // Adds last item based on if it is quoted.\n    if (currentState === STATE_QUOTE_AFTER_QUOTE) {\n      result.push(line.substring(readOffset, readLength - 1));\n    } else {\n      result.push(line.substring(readOffset));\n    }\n    // Check if each row has the same number of elements as column names.\n    if (validateElementCount && result.length !== this.fullColumnNames.length) {\n      throw new Error(`Invalid row in csv file. Should have ${\n          this.fullColumnNames.length} elements in a row, but got ${result}`);\n    }\n    return result;\n  }\n}\n\n// TODO(soergel): add more basic datasets for parity with tf.data\n// tf.data.FixedLengthRecordDataset()\n// tf.data.TFRecordDataset()\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env, Tensor, tensor, Tensor2D, Tensor3D, TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {MicrophoneConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of tensors from microphone audio stream. The tensors are\n * representing audio data as frequency-domain spectrogram generated with\n * browser's native FFT. Tensors representing time-domain waveform is available\n * based on configuration. Only works in browser environment.\n */\nexport class MicrophoneIterator extends LazyIterator<TensorContainer> {\n  private isClosed = false;\n  private stream: MediaStream;\n  private readonly fftSize: number;\n  private readonly columnTruncateLength: number;\n  private freqData: Float32Array;\n  private timeData: Float32Array;\n  private readonly numFrames: number;\n  private analyser: AnalyserNode;\n  private audioContext: AudioContext;\n  private sampleRateHz: number;\n  private readonly audioTrackConstraints: MediaTrackConstraints;\n  private readonly smoothingTimeConstant: number;\n  private readonly includeSpectrogram: boolean;\n  private readonly includeWaveform: boolean;\n\n  private constructor(protected readonly microphoneConfig: MicrophoneConfig) {\n    super();\n    this.fftSize = microphoneConfig.fftSize || 1024;\n    const fftSizeLog2 = Math.log2(this.fftSize);\n    if (this.fftSize < 0 || fftSizeLog2 < 4 || fftSizeLog2 > 14 ||\n        !Number.isInteger(fftSizeLog2)) {\n      throw new Error(\n          `Invalid fftSize: it must be a power of 2 between ` +\n          `2 to 4 and 2 to 14, but got ${this.fftSize}`);\n    }\n\n    this.numFrames = microphoneConfig.numFramesPerSpectrogram || 43;\n    this.sampleRateHz = microphoneConfig.sampleRateHz;\n    this.columnTruncateLength =\n        microphoneConfig.columnTruncateLength || this.fftSize;\n    this.audioTrackConstraints = microphoneConfig.audioTrackConstraints;\n    this.smoothingTimeConstant = microphoneConfig.smoothingTimeConstant || 0;\n\n    this.includeSpectrogram =\n        microphoneConfig.includeSpectrogram === false ? false : true;\n    this.includeWaveform =\n        microphoneConfig.includeWaveform === true ? true : false;\n    if (!this.includeSpectrogram && !this.includeWaveform) {\n      throw new Error(\n          'Both includeSpectrogram and includeWaveform are false. ' +\n          'At least one type of data should be returned.');\n    }\n  }\n\n  summary() {\n    return `microphone`;\n  }\n\n  // Construct a MicrophoneIterator and start the audio stream.\n  static async create(microphoneConfig: MicrophoneConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'microphone API is only supported in browser environment.');\n    }\n\n    const microphoneIterator = new MicrophoneIterator(microphoneConfig);\n\n    // Call async function start() to initialize the audio stream.\n    await microphoneIterator.start();\n\n    return microphoneIterator;\n  }\n\n  // Start the audio stream and FFT.\n  async start(): Promise<void> {\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        audio: this.audioTrackConstraints == null ? true :\n                                                    this.audioTrackConstraints,\n        video: false\n      });\n    } catch (e) {\n      throw new Error(\n          `Error thrown while initializing video stream: ${e.message}`);\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain audio from microphone.');\n    }\n\n    const ctxConstructor =\n        // tslint:disable-next-line:no-any\n        (window as any).AudioContext || (window as any).webkitAudioContext;\n    this.audioContext = new ctxConstructor();\n\n    if (!this.sampleRateHz) {\n      // If sample rate is not provided, use the available sample rate on\n      // device.\n      this.sampleRateHz = this.audioContext.sampleRate;\n    } else if (this.audioContext.sampleRate !== this.sampleRateHz) {\n      throw new Error(\n          `Mismatch in sampling rate: ` +\n          `Expected: ${this.sampleRateHz}; ` +\n          `Actual: ${this.audioContext.sampleRate}`);\n    }\n\n    const streamSource = this.audioContext.createMediaStreamSource(this.stream);\n    this.analyser = this.audioContext.createAnalyser();\n    this.analyser.fftSize = this.fftSize * 2;\n    this.analyser.smoothingTimeConstant = this.smoothingTimeConstant;\n    streamSource.connect(this.analyser);\n    this.freqData = new Float32Array(this.fftSize);\n    this.timeData = new Float32Array(this.fftSize);\n    return;\n  }\n\n  async next(): Promise<IteratorResult<TensorContainer>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let spectrogramTensor: Tensor;\n    let waveformTensor: Tensor;\n\n    const audioDataQueue = await this.getAudioData();\n    if (this.includeSpectrogram) {\n      const freqData = this.flattenQueue(audioDataQueue.freqDataQueue);\n      spectrogramTensor = this.getTensorFromAudioDataArray(\n          freqData, [this.numFrames, this.columnTruncateLength, 1]);\n    }\n    if (this.includeWaveform) {\n      const timeData = this.flattenQueue(audioDataQueue.timeDataQueue);\n      waveformTensor = this.getTensorFromAudioDataArray(\n          timeData, [this.numFrames * this.fftSize, 1]);\n    }\n\n    return {\n      value: {'spectrogram': spectrogramTensor, 'waveform': waveformTensor},\n      done: false\n    };\n  }\n\n  // Capture one result from the audio stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<{spectrogram: Tensor3D, waveform: Tensor2D}> {\n    return (await this.next()).value as\n        {spectrogram: Tensor3D, waveform: Tensor2D};\n  }\n\n  private async getAudioData():\n      Promise<{freqDataQueue: Float32Array[], timeDataQueue: Float32Array[]}> {\n    const freqDataQueue: Float32Array[] = [];\n    const timeDataQueue: Float32Array[] = [];\n    let currentFrames = 0;\n    return new Promise(resolve => {\n      const intervalID = setInterval(() => {\n        if (this.includeSpectrogram) {\n          this.analyser.getFloatFrequencyData(this.freqData);\n          // If the audio stream is initializing, return empty queue.\n          if (this.freqData[0] === -Infinity) {\n            resolve({freqDataQueue, timeDataQueue});\n          }\n          freqDataQueue.push(this.freqData.slice(0, this.columnTruncateLength));\n        }\n        if (this.includeWaveform) {\n          this.analyser.getFloatTimeDomainData(this.timeData);\n          timeDataQueue.push(this.timeData.slice());\n        }\n\n        // Clean interval and return when all frames have been collected\n        if (++currentFrames === this.numFrames) {\n          clearInterval(intervalID);\n          resolve({freqDataQueue, timeDataQueue});\n        }\n      }, this.fftSize / this.sampleRateHz * 1e3);\n    });\n  }\n\n  // Stop the audio stream and pause the iterator.\n  stop(): void {\n    if (!this.isClosed) {\n      this.isClosed = true;\n      this.analyser.disconnect();\n      this.audioContext.close();\n      if (this.stream != null && this.stream.getTracks().length > 0) {\n        this.stream.getTracks()[0].stop();\n      }\n    }\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor[]> {\n    throw new Error('Can not convert infinite audio stream to array.');\n  }\n\n  // Return audio sampling rate in Hz\n  getSampleRate(): number {\n    return this.sampleRateHz;\n  }\n\n  private flattenQueue(queue: Float32Array[]): Float32Array {\n    const frameSize = queue[0].length;\n    const freqData = new Float32Array(queue.length * frameSize);\n    queue.forEach((data, i) => freqData.set(data, i * frameSize));\n    return freqData;\n  }\n\n  private getTensorFromAudioDataArray(freqData: Float32Array, shape: number[]):\n      Tensor {\n    const vals = new Float32Array(util.sizeFromShape(shape));\n    // If the data is less than the output shape, the rest is padded with zeros.\n    vals.set(freqData, vals.length - freqData.length);\n    return tensor(vals, shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {browser, env, image, tensor1d, Tensor1D, tensor2d, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\nimport {WebcamConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of image tensors from webcam video stream. Only works in\n * browser environment.\n */\nexport class WebcamIterator extends LazyIterator<Tensor3D> {\n  private isClosed = true;\n  private stream: MediaStream;\n  private resize = false;\n  private cropSize: [number, number];\n  private cropBox: Tensor2D;\n  private cropBoxInd: Tensor1D;\n\n  private constructor(\n      protected readonly webcamVideoElement: HTMLVideoElement,\n      protected readonly webcamConfig: WebcamConfig) {\n    super();\n    if (this.needToResize()) {\n      this.resize = true;\n      this.cropSize =\n          [this.webcamConfig.resizeHeight, this.webcamConfig.resizeWidth];\n      this.cropBoxInd = tensor1d([0], 'int32');\n      if (this.webcamConfig.centerCrop) {\n        // Calculate the box based on resizing shape.\n        const widthCroppingRatio =\n            this.webcamConfig.resizeWidth * 1.0 / this.webcamVideoElement.width;\n        const heightCroppingRatio = this.webcamConfig.resizeHeight * 1.0 /\n            this.webcamVideoElement.height;\n        const widthCropStart = (1 - widthCroppingRatio) / 2;\n        const heightCropStart = (1 - heightCroppingRatio) / 2;\n        const widthCropEnd = widthCropStart + widthCroppingRatio;\n        const heightCropEnd = heightCroppingRatio + heightCropStart;\n        this.cropBox = tensor2d(\n            [heightCropStart, widthCropStart, heightCropEnd, widthCropEnd],\n            [1, 4]);\n      } else {\n        this.cropBox = tensor2d([0, 0, 1, 1], [1, 4]);\n      }\n    }\n  }\n\n  summary() {\n    return `webcam`;\n  }\n\n  // Construct a WebcamIterator and start it's video stream.\n  static async create(\n      webcamVideoElement?: HTMLVideoElement, webcamConfig: WebcamConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'tf.data.webcam is only supported in browser environment.');\n    }\n\n    if (!webcamVideoElement) {\n      // If webcam video element is not provided, create a hidden video element\n      // with provided width and height.\n      webcamVideoElement = document.createElement('video');\n      if (!webcamConfig.resizeWidth || !webcamConfig.resizeHeight) {\n        throw new Error(\n            'Please provide webcam video element, or resizeWidth and ' +\n            'resizeHeight to create a hidden video element.');\n      }\n      webcamVideoElement.width = webcamConfig.resizeWidth;\n      webcamVideoElement.height = webcamConfig.resizeHeight;\n    }\n    const webcamIterator = new WebcamIterator(webcamVideoElement, webcamConfig);\n\n    // Call async function to initialize the video stream.\n    await webcamIterator.start();\n\n    return webcamIterator;\n  }\n\n  // Async function to start video stream.\n  async start(): Promise<void> {\n    if (this.webcamConfig.facingMode) {\n      util.assert(\n          (this.webcamConfig.facingMode === 'user') ||\n              (this.webcamConfig.facingMode === 'environment'),\n          () =>\n              `Invalid webcam facing mode: ${this.webcamConfig.facingMode}. ` +\n              `Please provide 'user' or 'environment'`);\n    }\n\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        video: {\n          deviceId: this.webcamConfig.deviceId,\n          facingMode: this.webcamConfig.facingMode ?\n              this.webcamConfig.facingMode :\n              'user',\n          width: this.webcamVideoElement.width,\n          height: this.webcamVideoElement.height\n        }\n      });\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message = `Error thrown while initializing video stream: ${e.message}`;\n      throw e;\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain video from webcam.');\n    }\n\n    // Older browsers may not have srcObject\n    try {\n      this.webcamVideoElement.srcObject = this.stream;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = window.URL.createObjectURL(this.stream);\n    }\n    // Start the webcam video stream\n    this.webcamVideoElement.play();\n\n    this.isClosed = false;\n\n    return new Promise<void>(resolve => {\n      // Add event listener to make sure the webcam has been fully initialized.\n      this.webcamVideoElement.onloadedmetadata = () => {\n        resolve();\n      };\n    });\n  }\n\n  async next(): Promise<IteratorResult<Tensor3D>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let img;\n    try {\n      img = browser.fromPixels(this.webcamVideoElement);\n    } catch (e) {\n      throw new Error(\n          `Error thrown converting video to pixels: ${JSON.stringify(e)}`);\n    }\n    if (this.resize) {\n      try {\n        return {value: this.cropAndResizeFrame(img), done: false};\n      } catch (e) {\n        throw new Error(`Error thrown cropping the video: ${e.message}`);\n      } finally {\n        img.dispose();\n      }\n    } else {\n      return {value: img, done: false};\n    }\n  }\n\n  private needToResize() {\n    // If resizeWidth and resizeHeight are provided, and different from the\n    // width and height of original HTMLVideoElement, then resizing and cropping\n    // is required.\n    if (this.webcamConfig.resizeWidth && this.webcamConfig.resizeHeight &&\n        (this.webcamVideoElement.width !== this.webcamConfig.resizeWidth ||\n         this.webcamVideoElement.height !== this.webcamConfig.resizeHeight)) {\n      return true;\n    }\n    return false;\n  }\n\n  // Cropping and resizing each frame based on config\n  cropAndResizeFrame(img: Tensor3D): Tensor3D {\n    return tidy(() => {\n      const expandedImage: Tensor4D = img.toFloat().expandDims(0);\n      let resizedImage;\n      resizedImage = image.cropAndResize(\n          expandedImage, this.cropBox, this.cropBoxInd, this.cropSize,\n          'bilinear');\n      // Extract image from batch cropping.\n      const shape = resizedImage.shape;\n      return resizedImage.reshape(shape.slice(1) as [number, number, number]);\n    });\n  }\n\n  // Capture one frame from the video stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<Tensor3D> {\n    return (await this.next()).value;\n  }\n\n  // Stop the video stream and pause webcam iterator.\n  stop(): void {\n    const tracks = this.stream.getTracks();\n\n    tracks.forEach(track => track.stop());\n\n    try {\n      this.webcamVideoElement.srcObject = null;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = null;\n    }\n    this.isClosed = true;\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor3D[]> {\n    throw new Error('Can not convert infinite video stream to array.');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {ByteChunkIterator} from './iterators/byte_chunk_iterator';\n\n/**\n * Represents a data source readable as a stream of binary data chunks.\n *\n * Because `Dataset`s can be read repeatedly (via `Dataset.iterator()`), this\n * provides a means to repeatedly create streams from the underlying data\n * sources.\n */\nexport abstract class DataSource {\n  /**\n   * Obtain a new stream of binary data chunks.\n   *\n   * Starts the new stream from the beginning of the data source, even if other\n   * streams have been obtained previously.\n   */\n  abstract async iterator(): Promise<ByteChunkIterator>;\n\n  // TODO(soergel): consider chainable Dataset construction here\n}\n\n// TODO(soergel): consider convenience factory functions here\n// in combination with chainable source->dataset above, e.g.:\n// tf.data.url(...).asCsvDataset().shuffle().batch()\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\n\nexport abstract class StringIterator extends LazyIterator<string> {\n  /**\n   * Splits a string stream on a given separator.\n   *\n   * It is assumed that the incoming chunk boundaries have no semantic meaning,\n   * so conceptually the incoming stream is treated simply as the concatenation\n   * of its elements.\n   *\n   * The outgoing stream provides chunks corresponding to the results of the\n   * standard string split() operation (even if such a chunk spanned incoming\n   * chunks).  The separators are not included.\n   *\n   * A typical usage is to split a text file (represented as a stream with\n   * arbitrary chunk boundaries) into lines.\n   *\n   * @param upstream A readable stream of strings that can be treated as\n   *   concatenated.\n   * @param separator A character to split on.\n   */\n  split(separator: string): StringIterator {\n    return new SplitIterator(this, separator);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on StringIterator.  Unfortunately they can't be placed in separate files, due\n// to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class SplitIterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass SplitIterator extends StringIterator {\n  private impl: SplitIteratorImpl;\n\n  constructor(protected upstream: LazyIterator<string>, separator: string) {\n    super();\n    this.impl = new SplitIteratorImpl(upstream, separator);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\nclass SplitIteratorImpl extends OneToManyIterator<string> {\n  // A partial string at the end of an upstream chunk\n  carryover = '';\n\n  constructor(\n      protected upstream: LazyIterator<string>, protected separator: string) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Split('${this.separator}')`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    if (chunkResult.done) {\n      if (this.carryover === '') {\n        return false;\n      }\n\n      // Pretend that the pump succeeded in order to emit the small last batch.\n      // The next pump() call will actually fail.\n      this.outputQueue.push(this.carryover);\n      this.carryover = '';\n      return true;\n    }\n    const lines = chunkResult.value.split(this.separator);\n    // Note the behavior: \" ab \".split(' ') === ['', 'ab', '']\n    // Thus the carryover may be '' if the separator falls on a chunk\n    // boundary; this produces the correct result.\n\n    lines[0] = this.carryover + lines[0];\n    for (const line of lines.slice(0, -1)) {\n      this.outputQueue.push(line);\n    }\n    this.carryover = lines[lines.length - 1];\n\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\nimport {StringIterator} from './string_iterator';\n\nexport abstract class ByteChunkIterator extends LazyIterator<Uint8Array> {\n  /**\n   * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n   *\n   * The byte arrays producetd from the ByteChunkIterator on which this is\n   * called will be interpreted as concatenated.  No assumptions are made about\n   * the boundaries of the incoming chunks, so a multi-byte UTF8 encoding of a\n   * character may span the boundary between chunks.  This naturally happens,\n   * for instance, when reading fixed-size byte arrays from a file.\n   */\n  decodeUTF8(): StringIterator {\n    return new Utf8Iterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on ByteChunkIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class Utf8Iterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass Utf8Iterator extends StringIterator {\n  private impl: Utf8IteratorImpl;\n\n  constructor(protected upstream: LazyIterator<Uint8Array>) {\n    super();\n    this.impl = new Utf8IteratorImpl(upstream);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\n/**\n * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n *\n * This is tricky because the incoming byte array boundaries may disrupt a\n * multi-byte UTF8 character. Thus any incomplete character data at the end of\n * a chunk must be carried over and prepended to the next chunk before\n * decoding. Luckily with native decoder, TextDecoder in browser and\n * string_decoder in node, byte array boundaries are handled automatically.\n *\n * In the context of an input pipeline for machine learning, UTF8 decoding is\n * needed to parse text files containing training examples or prediction\n * requests (e.g., formatted as CSV or JSON). We cannot use the built-in\n * decoding provided by FileReader.readAsText() because here we are in a\n * streaming context, which FileReader does not support.\n *\n * @param upstream A `LazyIterator` of `Uint8Arrays` containing UTF8-encoded\n *   text, which should be interpreted as concatenated.  No assumptions are\n *   made about the boundaries of the incoming chunks, so a multi-byte UTF8\n *   encoding of a character may span the boundary between chunks.  This\n *   naturally happens, for instance, when reading fixed-size byte arrays from a\n *   file.\n */\nclass Utf8IteratorImpl extends OneToManyIterator<string> {\n  // `decoder` as `any` here to dynamically assign value based on the\n  // environment.\n  // tslint:disable-next-line:no-any\n  decoder: any;\n\n  constructor(protected readonly upstream: LazyIterator<Uint8Array>) {\n    super();\n    if (env().get('IS_BROWSER')) {\n      this.decoder = new TextDecoder('utf-8');\n    } else {\n      // tslint:disable-next-line:no-require-imports\n      const {StringDecoder} = require('string_decoder');\n      this.decoder = new StringDecoder('utf8');\n    }\n  }\n  summary() {\n    return `${this.upstream.summary()} -> Utf8`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    let chunk;\n    if (chunkResult.done) {\n      return false;\n    } else {\n      chunk = chunkResult.value;\n    }\n\n    let text: string;\n    if (env().get('IS_BROWSER')) {\n      text = this.decoder.decode(chunk, {stream: true});\n    } else {\n      text = this.decoder.write(Buffer.from(chunk.buffer));\n    }\n    this.outputQueue.push(text);\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// inspired by https://github.com/maxogden/filereader-stream\nimport {env, util} from '@tensorflow/tfjs-core';\nimport {FileElement} from '../types';\nimport {ByteChunkIterator} from './byte_chunk_iterator';\n\nexport interface FileChunkIteratorOptions {\n  /** The byte offset at which to begin reading the File or Blob. Default 0. */\n  offset?: number;\n  /** The number of bytes to read at a time. Default 1MB. */\n  chunkSize?: number;\n}\n\n/**\n * Provide a stream of chunks from a File, Blob, or Uint8Array.\n * @param file The source File, Blob or Uint8Array.\n * @param options Optional settings controlling file reading.\n * @returns a lazy Iterator of Uint8Arrays containing sequential chunks of the\n *   input File, Blob or Uint8Array.\n */\nexport class FileChunkIterator extends ByteChunkIterator {\n  offset: number;\n  chunkSize: number;\n\n  constructor(\n      protected file: FileElement,\n      protected options: FileChunkIteratorOptions = {}) {\n    super();\n    util.assert(\n        (file instanceof Uint8Array) ||\n            (env().get('IS_BROWSER') ?\n                 (file instanceof File || file instanceof Blob) :\n                 false),\n        () => 'FileChunkIterator only supports File, Blob and Uint8Array ' +\n            'right now.');\n    this.offset = options.offset || 0;\n    // default 1MB chunk has tolerable perf on large files\n    this.chunkSize = options.chunkSize || 1024 * 1024;\n  }\n\n  summary() {\n    return `FileChunks ${this.file}`;\n  }\n\n  async next(): Promise<IteratorResult<Uint8Array>> {\n    if (this.offset >= ((this.file instanceof Uint8Array) ?\n                            this.file.byteLength :\n                            this.file.size)) {\n      return {value: null, done: true};\n    }\n    const chunk = new Promise<Uint8Array>((resolve, reject) => {\n      const end = this.offset + this.chunkSize;\n      if (this.file instanceof Uint8Array) {\n        // Note if end > this.uint8Array.byteLength, we just get a small last\n        // chunk.\n        resolve(new Uint8Array(this.file.slice(this.offset, end)));\n      } else {\n        // This branch assumes that this.file type is File or Blob, which\n        // means it is in the browser environment.\n\n        // TODO(soergel): is this a performance issue?\n        const fileReader = new FileReader();\n        fileReader.onload = (event) => {\n          let data: string|ArrayBuffer|Uint8Array = fileReader.result;\n          // Not sure we can trust the return type of\n          // FileReader.readAsArrayBuffer See e.g.\n          // https://github.com/node-file-api/FileReader/issues/2\n          if (data instanceof ArrayBuffer) {\n            data = new Uint8Array(data);\n          }\n          if (!(data instanceof Uint8Array)) {\n            return reject(new TypeError('FileReader returned unknown type.'));\n          }\n          resolve(data);\n        };\n        fileReader.onabort = (event) => {\n          return reject(new Error('Aborted'));\n        };\n        fileReader.onerror = (event) => {\n          return reject(new Error(event.type));\n        };\n        // TODO(soergel): better handle onabort, onerror\n        // Note if end > this.file.size, we just get a small last chunk.\n        const slice = this.file.slice(this.offset, end);\n        // We can't use readAsText here (even if we know the file is text)\n        // because the slice boundary may fall within a multi-byte character.\n        fileReader.readAsArrayBuffer(slice);\n      }\n      this.offset = end;\n    });\n    return {value: (await chunk), done: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\nimport {FileChunkIterator, FileChunkIteratorOptions} from './file_chunk_iterator';\n\n/**\n * Provide a stream of chunks from a URL.\n *\n * Note this class first downloads the entire file into memory before providing\n * the first element from the stream.  This is because the Fetch API does not\n * yet reliably provide a reader stream for the response body.\n */\nexport async function urlChunkIterator(\n    url: RequestInfo, options: FileChunkIteratorOptions = {}) {\n  let urlString;\n  let requestInit;\n  if ((typeof url) === 'string') {\n    urlString = url as string;\n  } else {\n    urlString = (url as Request).url;\n    requestInit = getRequestInitFromRequest(url as Request);\n  }\n  const response = await util.fetch(urlString, requestInit);\n  if (response.ok) {\n    const uint8Array = new Uint8Array(await response.arrayBuffer());\n    return new FileChunkIterator(uint8Array, options);\n  } else {\n    throw new Error(response.statusText);\n  }\n}\n\n// Generate RequestInit from Request to match tf.util.fetch signature.\nconst getRequestInitFromRequest = (request: Request) => {\n  const init = {\n    method: request.method,\n    headers: request.headers,\n    body: request.body,\n    mode: request.mode,\n    credentials: request.credentials,\n    cache: request.cache,\n    redirect: request.redirect,\n    referrer: request.referrer,\n    integrity: request.integrity,\n  };\n  return init;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// Skip tslint any type check cause this method is aiming to check type of\n// input.\n// tslint:disable-next-line:no-any\nexport function isLocalPath(source: any): boolean {\n  return (typeof source === 'string') && source.substr(0, 7) === 'file://';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIterator, FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {FileElement} from '../types';\nimport {isLocalPath} from '../util/source_util';\n\n/**\n * Represents a file, blob, or Uint8Array readable as a stream of binary data\n * chunks.\n */\nexport class FileDataSource extends DataSource {\n  /**\n   * Create a `FileDataSource`.\n   *\n   * @param input Local file path, or `File`/`Blob`/`Uint8Array` object to\n   *     read. Local file only works in node environment.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected input: FileElement|string,\n      protected readonly options: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.input) && env().get('IS_NODE')) {\n      // tslint:disable-next-line:no-require-imports\n      const fs = require('fs');\n      this.input = fs.readFileSync((this.input as string).substr(7));\n    }\n    // TODO(kangyizhang): Add LocalFileChunkIterator to split local streaming\n    // with file in browser.\n    return new FileChunkIterator(this.input as FileElement, this.options);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {urlChunkIterator} from '../iterators/url_chunk_iterator';\nimport {isLocalPath} from '../util/source_util';\nimport {FileDataSource} from './file_data_source';\n\n/*\n * Represents a URL readable as a stream of binary data chunks.\n */\nexport class URLDataSource extends DataSource {\n  /**\n   * Create a `URLDataSource`.\n   *\n   * @param url A source URL string, or a `Request` object.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected readonly url: RequestInfo,\n      protected readonly fileOptions: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  // TODO(soergel): provide appropriate caching options.  Currently this\n  // will download the URL anew for each call to iterator().  Since we have\n  // to treat the downloaded file as a blob/buffer anyway, we may as well retain\n  // it-- but that raises GC issues.  Also we may want a persistent disk cache.\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.url)) {\n      return (new FileDataSource(this.url as string, this.fileOptions))\n          .iterator();\n    } else {\n      return urlChunkIterator(this.url, this.fileOptions);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer} from '@tensorflow/tfjs-core';\nimport {Dataset, datasetFromIteratorFn} from './dataset';\nimport {CSVDataset} from './datasets/csv_dataset';\nimport {iteratorFromFunction} from './iterators/lazy_iterator';\nimport {MicrophoneIterator} from './iterators/microphone_iterator';\nimport {WebcamIterator} from './iterators/webcam_iterator';\nimport {URLDataSource} from './sources/url_data_source';\nimport {CSVConfig, MicrophoneConfig, WebcamConfig} from './types';\n\n/**\n * Create a `CSVDataset` by reading and decoding CSV file(s) from provided URL\n * or local path if it's in Node environment.\n *\n * Note: If isLabel in columnConfigs is `true` for at least one column, the\n * element in returned `CSVDataset` will be an object of\n * `{xs:features, ys:labels}`: xs is a dict of features key/value pairs, ys\n * is a dict of labels key/value pairs. If no column is marked as label,\n * returns a dict of features only.\n *\n * ```js\n * const csvUrl =\n * 'https://storage.googleapis.com/tfjs-examples/multivariate-linear-regression/data/boston-housing-train.csv';\n *\n * async function run() {\n *   // We want to predict the column \"medv\", which represents a median value of\n *   // a home (in $1000s), so we mark it as a label.\n *   const csvDataset = tf.data.csv(\n *     csvUrl, {\n *       columnConfigs: {\n *         medv: {\n *           isLabel: true\n *         }\n *       }\n *     });\n *\n *   // Number of features is the number of column names minus one for the label\n *   // column.\n *   const numOfFeatures = (await csvDataset.columnNames()).length - 1;\n *\n *   // Prepare the Dataset for training.\n *   const flattenedDataset =\n *     csvDataset\n *     .map(({xs, ys}) =>\n *       {\n *         // Convert xs(features) and ys(labels) from object form (keyed by\n *         // column name) to array form.\n *         return {xs:Object.values(xs), ys:Object.values(ys)};\n *       })\n *     .batch(10);\n *\n *   // Define the model.\n *   const model = tf.sequential();\n *   model.add(tf.layers.dense({\n *     inputShape: [numOfFeatures],\n *     units: 1\n *   }));\n *   model.compile({\n *     optimizer: tf.train.sgd(0.000001),\n *     loss: 'meanSquaredError'\n *   });\n *\n *   // Fit the model using the prepared Dataset\n *   return model.fitDataset(flattenedDataset, {\n *     epochs: 10,\n *     callbacks: {\n *       onEpochEnd: async (epoch, logs) => {\n *         console.log(epoch + ':' + logs.loss);\n *       }\n *     }\n *   });\n * }\n *\n * await run();\n * ```\n *\n * @param source URL or local path to get CSV file. If it's a local path, it\n * must have prefix `file://` and it only works in node environment.\n * @param csvConfig (Optional) A CSVConfig object that contains configurations\n *     of reading and decoding from CSV file(s).\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function csv(\n    source: RequestInfo, csvConfig: CSVConfig = {}): CSVDataset {\n  return new CSVDataset(new URLDataSource(source), csvConfig);\n}\n\n/**\n * Create a `Dataset` that produces each element by calling a provided function.\n *\n * Note that repeated iterations over this `Dataset` may produce different\n * results, because the function will be called anew for each element of each\n * iteration.\n *\n * Also, beware that the sequence of calls to this function may be out of order\n * in time with respect to the logical order of the Dataset. This is due to the\n * asynchronous lazy nature of stream processing, and depends on downstream\n * transformations (e.g. .shuffle()). If the provided function is pure, this is\n * no problem, but if it is a closure over a mutable state (e.g., a traversal\n * pointer), then the order of the produced elements may be scrambled.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const ds = tf.data.func(func);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param f A function that produces one data element on each call.\n */\nexport function func<T extends TensorContainer>(\n    f: () => IteratorResult<T>| Promise<IteratorResult<T>>): Dataset<T> {\n  const iter = iteratorFromFunction(f);\n  return datasetFromIteratorFn(async () => iter);\n}\n\n/**\n * Create a `Dataset` that produces each element from provided JavaScript\n * generator, which is a function*\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions),\n * or a function that returns an\n * iterator\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions).\n *\n * The returned iterator should have `.next()` function that returns element in\n * format of `{value: TensorContainer, done:boolean}`.\n *\n * Example of creating a dataset from an iterator factory:\n * ```js\n * function makeIterator() {\n *   const numElements = 10;\n *   let index = 0;\n *\n *   const iterator = {\n *     next: () => {\n *       let result;\n *       if (index < numElements) {\n *         result = {value: index, done: false};\n *         index++;\n *         return result;\n *       }\n *       return {value: index, done: true};\n *     }\n *   };\n *   return iterator;\n * }\n * const ds = tf.data.generator(makeIterator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * Example of creating a dataset from a generator:\n * ```js\n * function* dataGenerator() {\n *   const numElements = 10;\n *   let index = 0;\n *   while (index < numElements) {\n *     const x = index;\n *     index++;\n *     yield x;\n *   }\n * }\n *\n * const ds = tf.data.generator(dataGenerator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param generator A Javascript generator function that returns a JavaScript\n *     iterator.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function generator<T extends TensorContainer>(\n    generator: () => Iterator<T>| Promise<Iterator<T>>): Dataset<T> {\n  return datasetFromIteratorFn(async () => {\n    const gen = await generator();\n    return iteratorFromFunction(() => gen.next());\n  });\n}\n\n/**\n * Create an iterator that generate `Tensor`s from webcam video stream. This API\n * only works in Browser environment when the device has webcam.\n *\n * Note: this code snippet only works when the device has a webcam. It will\n * request permission to open the webcam when running.\n * ```js\n * const videoElement = document.createElement('video');\n * videoElement.width = 100;\n * videoElement.height = 100;\n * const cam = await tf.data.webcam(videoElement);\n * const img = await cam.capture();\n * img.print();\n * cam.stop();\n * ```\n *\n * @param webcamVideoElement A `HTMLVideoElement` used to play video from\n *     webcam. If this element is not provided, a hidden `HTMLVideoElement` will\n *     be created. In that case, `resizeWidth` and `resizeHeight` must be\n *     provided to set the generated tensor shape.\n * @param webcamConfig A `WebcamConfig` object that contains configurations of\n *     reading and manipulating data from webcam video stream.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function webcam(\n    webcamVideoElement?: HTMLVideoElement,\n    webcamConfig?: WebcamConfig): Promise<WebcamIterator> {\n  return WebcamIterator.create(webcamVideoElement, webcamConfig);\n}\n\n/**\n * Create an iterator that generate frequency-domain spectrogram `Tensor`s from\n * microphone audio stream with browser's native FFT. This API only works in\n * browser environment when the device has microphone.\n *\n * Note: this code snippet only works when the device has a microphone. It will\n * request permission to open the microphone when running.\n * ```js\n * const mic = await tf.data.microphone({\n *   fftSize: 1024,\n *   columnTruncateLength: 232,\n *   numFramesPerSpectrogram: 43,\n *   sampleRateHz:44100,\n *   includeSpectrogram: true,\n *   includeWaveform: true\n * });\n * const audioData = await mic.capture();\n * const spectrogramTensor = audioData.spectrogram;\n * spectrogramTensor.print();\n * const waveformTensor = audioData.waveform;\n * waveformTensor.print();\n * mic.stop();\n * ```\n *\n * @param microphoneConfig A `MicrophoneConfig` object that contains\n *     configurations of reading audio data from microphone.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function microphone(microphoneConfig?: MicrophoneConfig):\n    Promise<MicrophoneIterator> {\n  return MicrophoneIterator.create(microphoneConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${\n              opName} does not support complex64 tensors in the CPU backend.`);\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {backend_util, BackendTimingInfo, DataStorage, DataType, DataValues, engine, env, kernel_impls, KernelBackend, max, NumericDataType, Rank, Scalar, ShapeMap, slice_util, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, TensorInfo, TypedArray, upcastType, util} from '@tensorflow/tfjs-core';\n\nconst nonMaxSuppressionV3Impl = kernel_impls.nonMaxSuppressionV3Impl;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\nimport * as seedrandom from 'seedrandom';\nimport {assertNotComplex} from './cpu_util';\n\ninterface DataId {}\n\nfunction mapActivation(\n    backend: MathBackendCPU, x: Tensor, activation: backend_util.Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return backend.linear(x);\n  } else if (activation === 'relu') {\n    return backend.relu(x);\n  } else if (activation === 'elu') {\n    return tf.elu(x);\n  } else if (activation === 'relu6') {\n    return backend.relu6(x);\n  } else if (activation === 'prelu') {\n    return backend.prelu(x, preluActivationWeights);\n  }\n  throw new Error(\n      `Activation ${activation} has not been implemented for the CPU backend.`);\n}\n\nexport interface TensorData<D extends DataType> {\n  values?: backend_util.BackendValues;\n  dtype: D;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensorInfos field.\n  complexTensorInfos?: {real: TensorInfo, imag: TensorInfo};\n  // refCount keeps track of how many tensors reference it. Used for memory\n  // management.\n  refCount: number;\n}\n\nexport class MathBackendCPU extends KernelBackend {\n  public blockSize = 48;\n\n  data: DataStorage<TensorData<DataType>>;\n  private firstUse = true;\n\n  constructor() {\n    super();\n    this.data = new DataStorage(this, engine());\n  }\n\n  write(values: backend_util.BackendValues, shape: number[], dtype: DataType):\n      DataId {\n    if (this.firstUse) {\n      this.firstUse = false;\n      if (env().get('IS_NODE')) {\n        backend_util.warn(\n            '\\n============================\\n' +\n            'Hi there \uD83D\uDC4B. Looks like you are running TensorFlow.js in ' +\n            'Node.js. To speed things up dramatically, install our node ' +\n            'backend, which binds to TensorFlow C++, by running ' +\n            'npm i @tensorflow/tfjs-node, ' +\n            'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +\n            'Then call require(\\'@tensorflow/tfjs-node\\'); (-gpu ' +\n            'suffix for CUDA) at the start of your program. ' +\n            'Visit https://github.com/tensorflow/tfjs-node for more details.' +\n            '\\n============================');\n      }\n    }\n    const dataId = {};\n\n    this.data.set(dataId, {values, dtype, refCount: 1});\n\n    return dataId;\n  }\n\n  /**\n   * Create a data bucket in cpu backend.\n   * @param shape Shape of the `TensorInfo`.\n   * @param dtype DType of the `TensorInfo`.\n   * @param values The value of the `TensorInfo` stored as a flattened array.\n   */\n  makeTensorInfo(\n      shape: number[], dtype: DataType,\n      values?: backend_util.BackendValues): TensorInfo {\n    const outId = this.write(values, shape, dtype);\n\n    return {dataId: outId, shape, dtype};\n  }\n\n  /** Increase refCount of a `TensorData`. */\n  incRef(dataId: DataId): void {\n    const tensorData = this.data.get(dataId);\n    tensorData.refCount++;\n  }\n\n  /** Decrease refCount of a `TensorData`. */\n  decRef(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n      tensorData.refCount--;\n    }\n  }\n\n  move(\n      dataId: DataId, values: backend_util.BackendValues, shape: number[],\n      dtype: DataType): void {\n    this.data.set(dataId, {values, dtype, refCount: 1});\n  }\n\n  numDataIds(): number {\n    return this.data.numDataIds();\n  }\n\n  async read(dataId: DataId): Promise<backend_util.BackendValues> {\n    return this.readSync(dataId);\n  }\n  readSync(dataId: DataId): backend_util.BackendValues {\n    const {dtype, complexTensorInfos} = this.data.get(dataId);\n\n    if (dtype === 'complex64') {\n      const realValues =\n          this.readSync(complexTensorInfos.real.dataId) as Float32Array;\n      const imagValues =\n          this.readSync(complexTensorInfos.imag.dataId) as Float32Array;\n      return backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    }\n\n    return this.data.get(dataId).values;\n  }\n\n  private bufferSync<R extends Rank>(t: Tensor<R>): TensorBuffer<R> {\n    const data = this.readSync(t.dataId);\n    let decodedData = data as DataValues;\n    if (t.dtype === 'string') {\n      try {\n        // Decode the bytes into string.\n        decodedData = (data as Uint8Array[]).map(d => util.decodeString(d));\n      } catch {\n        throw new Error('Failed to decode encoded string bytes into utf-8');\n      }\n    }\n    return tf.buffer(t.shape, t.dtype, decodedData) as TensorBuffer<R>;\n  }\n\n  makeOutput<T extends Tensor>(\n      values: backend_util.BackendValues, shape: number[], dtype: DataType): T {\n    const dataId = this.write(values, shape, dtype);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  disposeData(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const {complexTensorInfos} = this.data.get(dataId);\n\n      if (complexTensorInfos != null) {\n        this.disposeData(complexTensorInfos.real.dataId);\n        this.disposeData(complexTensorInfos.imag.dataId);\n      }\n\n      this.data.delete(dataId);\n    }\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n\n      tensorData.refCount--;\n\n      if (tensorData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  async time(f: () => void): Promise<BackendTimingInfo> {\n    const start = util.now();\n    f();\n    const kernelMs = util.now() - start;\n    return {kernelMs};\n  }\n\n  memory() {\n    return {\n      // Unreliable due to automatic gc. The numbers above are cumulative.\n      unreliable: true,\n      reasons:\n          ['The reported memory is an upper bound. Due to automatic garbage ' +\n           'collection, the true allocated memory may be less.']\n    };\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    assertNotComplex(x, 'stridedSlice');\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tf.tensor([], outShape) as T;\n    }\n\n    const buffer = tf.buffer(outShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n    for (let i = 0; i < buffer.size; i++) {\n      const loc = buffer.indexToLoc(i);\n\n      const newLoc: number[] = new Array(loc.length);\n      for (let j = 0; j < newLoc.length; j++) {\n        newLoc[j] = loc[j] * strides[j] + begin[j];\n      }\n      buffer.set(xBuf.get(...newLoc), ...loc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  diag(x: Tensor): Tensor {\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const buffer = tf.buffer([x.size, x.size], x.dtype);\n    const vals = buffer.values;\n    for (let i = 0; i < xVals.length; i++) {\n      vals[i * x.size + i] = xVals[i];\n    }\n    return buffer.toTensor();\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = tf.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    assertNotComplex(x, 'reverse');\n\n    const buffer = tf.buffer(x.shape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < buffer.size; i++) {\n      const outLoc = buffer.indexToLoc(i);\n      const inLoc = outLoc.slice();\n      axis.forEach(ax => inLoc[ax] = x.shape[ax] - 1 - inLoc[ax]);\n      buffer.set(xBuf.get(...inLoc), ...outLoc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'neg');\n\n    // TODO(lina128): Use mul directly once neg is modularized.\n    return tf.mul(tf.scalar(-1), x);\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    assertNotComplex(tensors, 'addN');\n\n    const vals = tensors.map(t => this.readSync(t.dataId) as TypedArray);\n    const result = tf.buffer(tensors[0].shape, tensors[0].dtype as 'float32');\n    const resultVals = result.values;\n    for (let i = 0; i < tensors.length; i++) {\n      const currVals = vals[i];\n      for (let j = 0; j < resultVals.length; j++) {\n        resultVals[j] += currVals[j];\n      }\n    }\n    return result.toTensor() as T;\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n\n    // TODO(lina128): Use sub directly once softmax is modularized.\n    const a = tf.sub(logits, maxLogit.reshape(expandedShape));\n    const b = tf.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax\n    // kernel modularization.\n    return tf.div(b, sumExp);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    assertNotComplex([a, b], 'pow');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.pow(aValue, bValue)) as\n        T;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    assertNotComplex([a, b], 'matMul');\n\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const leftDim = transposeA ? a.shape[2] : a.shape[1];\n    const rightDim = transposeB ? b.shape[1] : b.shape[2];\n    const batchDim = a.shape[0];\n\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const [aBatch, aOuterStep, aInnerStep] = transposeA ?\n        [a.strides[0], 1, a.strides[1]] :\n        [a.strides[0], a.strides[1], 1];\n    const [bInnerStep, bOuterStep, bBatch] = transposeB ?\n        [1, b.strides[1], b.strides[0]] :\n        [b.strides[1], 1, b.strides[0]];\n\n    const size = leftDim * rightDim;\n    const result = tf.buffer([batchDim, leftDim, rightDim], a.dtype);\n    const resVals = result.values as TypedArray;\n    const blockSize = this.blockSize;\n\n    for (let b = 0; b < batchDim; b++) {\n      for (let i0 = 0; i0 < leftDim; i0 += blockSize) {\n        for (let j0 = 0; j0 < rightDim; j0 += blockSize) {\n          for (let k0 = 0; k0 < sharedDim; k0 += blockSize) {\n            // for when blockSize doesn't evenly divide the input\n            const iBlock = Math.min(i0 + blockSize, leftDim);\n            const jBlock = Math.min(j0 + blockSize, rightDim);\n            const kBlock = Math.min(k0 + blockSize, sharedDim);\n\n            for (let i = i0; i < iBlock; i++) {\n              for (let j = j0; j < jBlock; j++) {\n                let sum = 0.0;\n\n                for (let k = k0; k < kBlock; k++) {\n                  sum += aValues[b * aBatch + i * aOuterStep + k * aInnerStep] *\n                      bValues[k * bInnerStep + j * bOuterStep + b * bBatch];\n                }\n                resVals[b * size + (i * rightDim + j)] += sum;\n              }\n            }\n          }\n        }\n      }\n    }\n    return result.toTensor() as Tensor3D;\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    let result = this.batchMatMul(a, b, transposeA, transposeB);\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedBatchMatMul is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor3D;\n    }\n\n    return result;\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'floorDiv');\n\n    const op = (a: number, b: number) => Math.floor(a / b);\n    const outputDtype = 'int32';\n    return this.broadcastedBinaryOp(a, b, outputDtype, op);\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let sum = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        sum += aVals[offset + j];\n      }\n      vals[i] = sum;\n    }\n    return result;\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let prod = 1;\n      for (let j = 0; j < reduceSize; ++j) {\n        prod *= aVals[offset + j];\n      }\n      vals[i] = prod;\n    }\n    return result;\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    assertNotComplex(x, 'unsortedSegmentSum');\n\n    const res = [];\n\n    // Reshape the segment id's so that they can be broadcast with\n    // x. The new shape should be [segmentIds.shape, 1, ..., 1]\n    const numIters = x.rank - segmentIds.rank;\n    for (let i = 0; i < numIters; ++i) {\n      segmentIds = segmentIds.expandDims(i + 1);\n    }\n\n    for (let i = 0; i < numSegments; ++i) {\n      const segmentId = tf.scalar(i, 'int32');\n      const mask = tf.equal(segmentId, segmentIds).asType('float32');\n      const sum = mask.mul(x).sum(0);\n      res.push(sum);\n    }\n\n    return tf.stack(res);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMin');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMin', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      let minIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n          minIndex = j;\n        }\n      }\n      vals[i] = minIndex;\n    }\n    return result;\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMax');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMax', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let max = aVals[offset];\n      let maxIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value > max) {\n          max = value;\n          maxIndex = j;\n        }\n      }\n      vals[i] = maxIndex;\n    }\n    return result;\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    assertNotComplex(x, 'cumsum');\n\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `backend.cumsum in CPU expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(x.shape, resultDtype);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    const finalDim = x.shape[x.rank - 1];\n    const indexAdjuster = reverse ?\n        (i: number, j: number) => i + finalDim - j - 1 :\n        (i: number, j: number) => i + j;\n    for (let i = 0; i < aVals.length; i += finalDim) {\n      for (let j = 0; j < finalDim; j++) {\n        const idx = indexAdjuster(i, j);\n        if (j === 0) {\n          vals[idx] = exclusive ? 0 : aVals[idx];\n        } else {\n          const prevIdx = indexAdjuster(i, j - 1);\n          vals[idx] = exclusive ? aVals[prevIdx] + vals[prevIdx] :\n                                  aVals[idx] + vals[prevIdx];\n        }\n      }\n    }\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'equal');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal === bVal) ? 1 : 0;\n    });\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'notEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal !== bVal) ? 1 : 0;\n    });\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'less');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal < bVal) ? 1 : 0;\n    });\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'lessEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal <= bVal) ? 1 : 0;\n    });\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greater');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal > bVal) ? 1 : 0;\n    });\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greaterEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal >= bVal) ? 1 : 0;\n    });\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalAnd');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal && bVal;\n    });\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalOr');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal || bVal;\n    });\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([condition, a, b], 'select');\n\n    const values = this.readSync(condition.dataId) as TypedArray;\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const result = tf.zeros(a.shape, upcastType(a.dtype, b.dtype));\n    const newValues = this.readSync(result.dataId) as TypedArray;\n    let index = 0;\n    const offset = condition.rank === 0 || condition.rank > 1 || a.rank === 1 ?\n        1 :\n        util.sizeFromShape(a.shape.slice(1));\n\n    for (let i = 0; i < values.length; i++) {\n      for (let j = 0; j < offset; j++) {\n        if (values[i] === 1) {\n          newValues[index++] = aValues[i];\n        } else {\n          newValues[index++] = bValues[i];\n        }\n      }\n    }\n\n    return result;\n  }\n\n  where(condition: Tensor): Tensor2D {\n    assertNotComplex([condition], 'where');\n\n    const condVals = this.readSync(condition.dataId) as TypedArray;\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    assertNotComplex(x, 'topk');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'min');\n\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n        }\n      }\n      vals[i] = min;\n    }\n    return result;\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'minimum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.min(aVal, bVal));\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'mod');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const rem = aVal % bVal;\n      if ((aVal < 0 && bVal < 0) || (aVal >= 0 && bVal >= 0)) {\n        return rem;\n      } else {\n        return (rem + bVal) % bVal;\n      }\n    });\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'maximum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.max(aVal, bVal));\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'all');\n\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let all = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        all = all && value;\n      }\n      vals[i] = all;\n    }\n    return result;\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'any');\n\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let anyVal = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        anyVal = anyVal || value;\n      }\n      vals[i] = anyVal;\n    }\n    return result;\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'squaredDifference');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const diff = aVal - bVal;\n      return diff * diff;\n    });\n  }\n\n  linear<T extends Tensor>(x: T): T {\n    return x;\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.max(0, inVals[i]);\n    }\n    return res as T;\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.min(Math.max(0, inVals[i]), 6);\n    }\n    return res as T;\n  }\n\n  prelu<T extends Tensor>(x: T, a: T): T {\n    assertNotComplex([x, a], 'prelu');\n\n    return this.broadcastedBinaryOp(\n               x, a, x.dtype,\n               (xValue, aValue) => xValue < 0 ? aValue * xValue : xValue) as T;\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    assertNotComplex([dy, y], 'eluDer');\n\n    const resultValues = new Float32Array(y.size);\n    const values = this.readSync(y.dataId) as TypedArray;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    for (let i = 0; i < values.length; ++i) {\n      const v = values[i];\n      if (v >= 1) {\n        resultValues[i] = dyValues[i];\n      } else {\n        resultValues[i] = dyValues[i] * (v + 1);\n      }\n    }\n    return this.makeOutput(resultValues, y.shape, 'float32');\n  }\n\n  atan2<T extends Tensor>(a: T, b: T): T {\n    assertNotComplex([a, b], 'atan2');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.atan2(aValue, bValue)) as\n        T;\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.conv2d(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedConv2d is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, filter], 'conv2d');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n\n    const xBatchStride = x.strides[0];\n    const xRowStride = isChannelsLast ? x.strides[1] : x.strides[2];\n    const xColStride = isChannelsLast ? x.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : x.strides[1];\n    const yBatchStride = y.strides[0];\n    const yRowStride = isChannelsLast ? y.strides[1] : y.strides[2];\n    const yColStride = isChannelsLast ? y.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : y.strides[1];\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * xBatchStride;\n      const yOffset1 = b * yBatchStride;\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * yRowStride;\n        const xRCorner = yR * convInfo.strideHeight - padTop;\n        for (let wR = 0; wR < filterHeight; wR++) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * xRowStride;\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * yColStride;\n            const xCCorner = yC * convInfo.strideWidth - padLeft;\n            for (let wC = 0; wC < filterWidth; wC++) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * xColStride;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1 * xChannelStride];\n                for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                  yVals[yOffset3 + d2 * yChannelStride] +=\n                      xVal * wVals[wOffset3 + d2];\n                }\n                wOffset3 += convInfo.outChannels;\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor() as Tensor4D;\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padFront = convInfo.padInfo.front;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const y = tf.buffer<Rank.R5>(convInfo.outShape, x.dtype as 'float32');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yF = 0; yF < convInfo.outDepth; ++yF) {\n        const yOffset2 = yOffset1 + yF * y.strides[1];\n        const xFCorner = yF * convInfo.strideDepth - padFront;\n        for (let wF = 0; wF < filterDepth; wF++) {\n          const xF = xFCorner + wF * dilationDepth;\n          if (xF < 0 || xF >= convInfo.inDepth) {\n            continue;\n          }\n          const wOffset1 = wF * filter.strides[0];\n          const xOffset2 = xOffset1 + xF * x.strides[1];\n\n          for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n            const yOffset3 = yOffset2 + yR * y.strides[2];\n            const xRCorner = yR * convInfo.strideHeight - padTop;\n            for (let wR = 0; wR < filterHeight; wR++) {\n              const xR = xRCorner + wR * dilationHeight;\n              if (xR < 0 || xR >= convInfo.inHeight) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wR * filter.strides[1];\n              const xOffset3 = xOffset2 + xR * x.strides[2];\n              for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n                const yOffset4 = yOffset3 + yC * convInfo.outChannels;\n                const xCCorner = yC * convInfo.strideWidth - padLeft;\n                for (let wC = 0; wC < filterWidth; wC++) {\n                  const xC = xCCorner + wC * dilationWidth;\n                  if (xC < 0 || xC >= convInfo.inWidth) {\n                    continue;\n                  }\n                  const wOffset3 = wOffset2 + wC * filter.strides[2];\n                  const xOffset4 = xOffset3 + xC * convInfo.inChannels;\n                  let wOffset4 = wOffset3;\n                  for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                    const xVal = xVals[xOffset4 + d1];\n                    for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                      yVals[yOffset4 + d2] += xVal * wVals[wOffset4 + d2];\n                    }\n                    wOffset4 += convInfo.outChannels;\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor();\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'conv2dDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth,\n      dataFormat\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const xBatchStride = dx.strides[0];\n    const xRowStride = isChannelsLast ? dx.strides[1] : dx.strides[2];\n    const xColStride = isChannelsLast ? dx.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : dx.strides[1];\n    const yBatchStride = dy.strides[0];\n    const yRowStride = isChannelsLast ? dy.strides[1] : dy.strides[2];\n    const yColStride = isChannelsLast ? dy.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : dy.strides[1];\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset =\n                    yBatchStride * b + yRowStride * yR + yColStride * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let d2 = 0; d2 < outChannels; ++d2) {\n                  const pixel = dyValues[dyOffset + yChannelStride * d2];\n                  const weight = fltValues[fltOffset + d2];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            const dxOffset = xBatchStride * b + xRowStride * xR +\n                xColStride * xC + xChannelStride * d1;\n            dxValues[dxOffset] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const dx = tf.buffer<Rank.R5>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2, dxS3] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2, fltS3] = filter.strides;\n    const {\n      batchSize,\n      filterDepth,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inDepth,\n      inHeight,\n      inWidth,\n      outChannels,\n      outDepth,\n      outHeight,\n      outWidth,\n      strideDepth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const frontPad = filterDepth - 1 - convInfo.padInfo.front;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        // Frames of depth\n        for (let xF = 0; xF < inDepth; ++xF) {\n          const xFCorner = xF - frontPad;\n          const xFMin = Math.max(0, Math.ceil(xFCorner / strideDepth));\n          const yFMax =\n              Math.min(outDepth, (filterDepth + xFCorner) / strideDepth);\n\n          // Rows as per standard 2d matrix notation\n          for (let xR = 0; xR < inHeight; ++xR) {\n            const xRCorner = xR - topPad;\n            const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n            const yRMax =\n                Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n            // Columns as per standard 2d matrix notation\n            for (let xC = 0; xC < inWidth; ++xC) {\n              const xCCorner = xC - leftPad;\n              const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n              const yCMax =\n                  Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n              let dotProd = 0;\n              for (let yF = xFMin; yF < yFMax; ++yF) {\n                const wF = yF * strideDepth - xFCorner;\n\n                for (let yR = xRMin; yR < yRMax; ++yR) {\n                  const wR = yR * strideHeight - xRCorner;\n\n                  for (let yC = xCMin; yC < yCMax; ++yC) {\n                    const wC = yC * strideWidth - xCCorner;\n                    const dyOffset =\n                        dyS0 * b + dyS1 * yF + dyS2 * yR + dyS3 * yC;\n                    const fltOffset = fltS0 * (filterDepth - 1 - wF) +\n                        fltS1 * (filterHeight - 1 - wR) +\n                        fltS2 * (filterWidth - 1 - wC) + fltS3 * d1;\n\n                    for (let d2 = 0; d2 < outChannels; ++d2) {\n                      const pixel = dyValues[dyOffset + d2];\n                      const weight = fltValues[fltOffset + d2];\n                      dotProd += pixel * weight;\n                    }\n                  }\n                }\n              }\n              dxValues[dxS0 * b + dxS1 * xF + dxS2 * xR + dxS3 * xC + d1] =\n                  dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, dy], 'conv2dDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n          for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n            // Need to convolve.\n            let dotProd = 0;\n            for (let b = 0; b < convInfo.batchSize; ++b) {\n              for (let yR = yRMin; yR < yRMax; ++yR) {\n                const xR = wR + yR * strideHeight - topPad;\n                for (let yC = yCMin; yC < yCMax; ++yC) {\n                  const xC = wC + yC * strideWidth - leftPad;\n                  if (isChannelsLast) {\n                    dotProd +=\n                        xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n                  } else {\n                    dotProd +=\n                        xBuf.get(b, d1, xR, xC) * dyBuf.get(b, d2, yR, yC);\n                  }\n                }\n              }\n            }\n            dW.set(dotProd, wR, wC, d1, d2);\n          }\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const dw = tf.buffer<Rank.R5>(convInfo.filterShape, 'float32');\n    const dwValues = dw.values;\n    const [dwS0, dwS1, dwS2, dwS3] = dw.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const [xS0, xS1, xS2, xS3] = x.strides;\n\n    const frontPad = convInfo.padInfo.front;\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n\n    for (let wF = 0; wF < filterDepth; ++wF) {\n      const yFMin = Math.max(0, Math.ceil((frontPad - wF) / strideDepth));\n      const yFMax = Math.min(\n          convInfo.outDepth, (convInfo.inDepth + frontPad - wF) / strideDepth);\n      const wOffset1 = wF * dwS0;\n\n      for (let wR = 0; wR < filterHeight; ++wR) {\n        const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n        const yRMax = Math.min(\n            convInfo.outHeight,\n            (convInfo.inHeight + topPad - wR) / strideHeight);\n        const wOffset2 = wR * dwS1 + wOffset1;\n\n        for (let wC = 0; wC < filterWidth; ++wC) {\n          const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n          const yCMax = Math.min(\n              convInfo.outWidth,\n              (convInfo.inWidth + leftPad - wC) / strideWidth);\n          const wOffset3 = wC * dwS2 + wOffset2;\n\n          for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n            const wOffset4 = d1 * dwS3 + wOffset3;\n\n            for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n              let dotProd = 0;\n              for (let b = 0; b < convInfo.batchSize; ++b) {\n                const xOffset1 = b * xS0;\n                const yOffset1 = b * dyS0;\n\n                for (let yF = yFMin; yF < yFMax; ++yF) {\n                  const xF = wF + yF * strideDepth - frontPad;\n                  const xOffset2 = xF * xS1 + xOffset1;\n                  const yOffset2 = yF * dyS1 + yOffset1;\n\n                  for (let yR = yRMin; yR < yRMax; ++yR) {\n                    const xR = wR + yR * strideHeight - topPad;\n                    const xOffset3 = xR * xS2 + xOffset2;\n                    const yOffset3 = yR * dyS2 + yOffset2;\n\n                    for (let yC = yCMin; yC < yCMax; ++yC) {\n                      const xC = wC + yC * strideWidth - leftPad;\n                      const xOffset4 = xC * xS3 + xOffset3;\n                      const yOffset4 = yC * dyS3 + yOffset3;\n\n                      dotProd +=\n                          xValues[xOffset4 + d1] * dyValues[yOffset4 + d2];\n                    }\n                  }\n                }\n              }\n              dwValues[wOffset4 + d2] = dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dw.toTensor();\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.depthwiseConv2D(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedDepthwiseConv2D is\n      // modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, filter], 'depthwiseConv2D');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * y.strides[1];\n        const xRCorner = yR * convInfo.strideHeight - padLeft;\n        for (let wR = 0; wR < filterHeight; ++wR) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * x.strides[1];\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * y.strides[2];\n            const xCCorner = yC * convInfo.strideWidth - padTop;\n            for (let wC = 0; wC < filterWidth; ++wC) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * convInfo.inChannels;\n              let yOffset4 = yOffset3;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1];\n                for (let q = 0; q < chMul; ++q) {\n                  yVals[yOffset4 + q] += xVal * wVals[wOffset3 + q];\n                }\n                yOffset4 += chMul;\n                wOffset3 += chMul;\n              }\n            }\n          }\n        }\n      }\n    }\n\n    return y.toTensor() as Tensor4D;\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'depthwiseConv2DDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n    const chMul = outChannels / inChannels;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset = dyS0 * b + dyS1 * yR + dyS2 * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let dm = 0; dm < chMul; ++dm) {\n                  const d2 = d1 * chMul + dm;\n                  const pixel = dyValues[dyOffset + d2];\n                  const weight = fltValues[fltOffset + dm];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            dxValues[dxS0 * b + dxS1 * xR + dxS2 * xC + d1] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, dy], 'depthwiseConv2DDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n          const d1 = Math.trunc(d2 / chMul);\n          const dm = d2 % chMul;\n\n          let dotProd = 0;\n          for (let b = 0; b < convInfo.batchSize; ++b) {\n            for (let yR = yRMin; yR < yRMax; ++yR) {\n              const xR = wR + yR * strideHeight - topPad;\n              for (let yC = yCMin; yC < yCMax; ++yC) {\n                const xC = wC + yC * strideWidth - leftPad;\n                dotProd += xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n              }\n            }\n          }\n          dW.set(dotProd, wR, wC, d1, dm);\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    assertNotComplex(x, 'tile');\n    return tile(this.bufferSync(x), reps) as T;\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    assertNotComplex([x, indices], 'gather');\n\n    const newShape: number[] = x.shape.slice();\n    const indicesValues = this.readSync(indices.dataId) as TypedArray;\n    newShape[axis] = indicesValues.length;\n    const result = tf.buffer(newShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < result.size; ++i) {\n      const newLoc = result.indexToLoc(i);\n\n      const originalLoc: number[] = newLoc.slice();\n      originalLoc[axis] = indicesValues[newLoc[axis]];\n\n      const originalIndex = xBuf.locToIndex(originalLoc);\n      result.values[i] = xBuf.values[originalIndex];\n    }\n    return result.toTensor() as T;\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    assertNotComplex([x], 'batchToSpaceND');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return tf.transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  private pool3d(\n      x: Tensor5D, convInfo: backend_util.Conv3DInfo,\n      poolType: 'max'|'avg'): Tensor5D {\n    assertNotComplex(x, 'pool3d');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const initialValue =\n        (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                              Number.POSITIVE_INFINITY);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = tf.buffer(convInfo.outShape, x.dtype);\n    const outputVals = output.values;\n\n    const outputBatchStrides = convInfo.outShape[1] * convInfo.outShape[2] *\n        convInfo.outShape[3] * convInfo.outShape[4];\n    const outputDepthStrides =\n        convInfo.outShape[2] * convInfo.outShape[3] * convInfo.outShape[4];\n    const outputRowStrides = convInfo.outShape[3] * convInfo.outShape[4];\n    const outputColStrides = convInfo.outShape[4];\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      const outputBatchOffset = batch * outputBatchStrides;\n      const inputBatchOffset = batch * x.strides[0];\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          const outputDepthOffset =\n              outputBatchOffset + yDepth * outputDepthStrides;\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            const outputRowOffset = outputDepthOffset + yRow * outputRowStrides;\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n              // Shader code begins\n              const outputColOffset = outputRowOffset + yCol * outputColStrides;\n              let minMaxValue = initialValue;\n              let avgValue = 0;\n              let count = 0;\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const xDepthOffset = inputBatchOffset + xDepth * x.strides[1];\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const xRowOffset = xDepthOffset + xRow * x.strides[2];\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const xColOffset = xRowOffset + xCol * x.strides[3];\n                    const pixel = xValues[xColOffset + channel];\n                    if ((poolType === 'max' && pixel > minMaxValue)) {\n                      minMaxValue = pixel;\n                    } else if (poolType === 'avg') {\n                      avgValue += pixel;\n                      count++;\n                    }\n                    if (isNaN(minMaxValue)) {\n                      break;\n                    }\n                  }\n                  if (isNaN(minMaxValue)) {\n                    break;\n                  }\n                }\n                if (isNaN(minMaxValue)) {\n                  break;\n                }\n              }\n              const outputOffset = outputColOffset + channel;\n              outputVals[outputOffset] =\n                  poolType === 'avg' ? avgValue / count : minMaxValue;\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor5D;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'avgPool3d');\n\n    return this.pool3d(x, convInfo, 'avg').toFloat();\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([dy, x], 'avgPool3dBackprop');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins.\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel;\n                  }\n                }\n              }\n              dx.set(\n                  dotProd * avgMultiplier, batch, dxDepth, dxRow, dxCol,\n                  channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'maxPool3d');\n\n    return this.pool3d(x, convInfo, 'max').toFloat();\n  }\n\n  private maxPool3dPositions(x: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const maxPositions = tf.buffer(convInfo.outShape, 'int32');\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const xBuf = this.bufferSync(x);\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n\n              // Shader code begins\n              let maxValue = Number.NEGATIVE_INFINITY;\n              let maxPosition = -1;\n\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const wDepth = xDepth - xDepthCorner;\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const wRow = xRow - xRowCorner;\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const wCol = xCol - xColCorner;\n                    const pixel = xBuf.get(batch, xDepth, xRow, xCol, channel);\n                    if (pixel >= maxValue) {\n                      maxValue = pixel;\n                      maxPosition = wDepth * effectiveFilterHeight *\n                              effectiveFilterWidth +\n                          wRow * effectiveFilterHeight + wCol;\n                    }\n                  }\n                }\n              }\n\n              maxPositions.set(maxPosition, batch, yDepth, yRow, yCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return maxPositions.toTensor() as Tensor5D;\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([x, y], 'maxPool3dBackprop');\n\n    const maxPositions = this.maxPool3dPositions(x, convInfo);\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const maxPosBuf = this.bufferSync(maxPositions);\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const maxPos = effectiveFilterDepth *\n                            effectiveFilterHeight * effectiveFilterWidth -\n                        1 -\n                        maxPosBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    const curPos =\n                        wDepth * effectiveFilterHeight * effectiveFilterWidth +\n                        wRow * effectiveFilterWidth + wCol;\n\n                    const mask = maxPos === curPos ? 1 : 0;\n                    if (mask === 0) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel * mask;\n                  }\n                }\n              }\n              dx.set(dotProd, batch, dxDepth, dxRow, dxCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeBilinear');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result = new Float32Array(\n        util.sizeFromShape([batch, newHeight, newWidth, numChannels]));\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n    let outputIdx = 0;\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n    for (let b = 0; b < batch; b++) {\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceRowFloor = Math.floor(sourceFracRow);\n        const rowFrac = sourceFracRow - sourceRowFloor;\n        const sourceRowCeil = Math.min(oldHeight - 1, Math.ceil(sourceFracRow));\n        const topRowOffset = b * x.strides[0] + sourceRowFloor * x.strides[1];\n        const botRowOffset = b * x.strides[0] + sourceRowCeil * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceColFloor = Math.floor(sourceFracCol);\n          const colFrac = sourceFracCol - sourceColFloor;\n          const sourceColCeil =\n              Math.min(oldWidth - 1, Math.ceil(sourceFracCol));\n          const topLeftOffest = topRowOffset + sourceColFloor * x.strides[2];\n          const botLeftOffset = botRowOffset + sourceColFloor * x.strides[2];\n          const topRightOffset = topRowOffset + sourceColCeil * x.strides[2];\n          const botRightOffest = botRowOffset + sourceColCeil * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n\n            // Compute the fractional index of the source.\n            const topLeft = xValues[topLeftOffest + d];\n            const bottomLeft = xValues[botLeftOffset + d];\n            const topRight = xValues[topRightOffset + d];\n            const bottomRight = xValues[botRightOffest + d];\n\n            const top = topLeft + (topRight - topLeft) * colFrac;\n            const bottom = bottomLeft + (bottomRight - bottomLeft) * colFrac;\n            const newValue = top + (bottom - top) * rowFrac;\n\n            result[outputIdx++] = newValue;\n          }\n        }\n      }\n    }\n    return tf.tensor(result, [batch, newHeight, newWidth, numChannels]);\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeBilinearBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass and add the\n    // corresponding coefficient from dy to the gradient (with some\n    // interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/3039375c86a5bbc9610c7725dcaa95d635f87ba2/tensorflow/core/kernels/resize_bilinear_op.cc#L275\n\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    let offset = 0;\n    for (let b = 0; b < batch; b++) {\n      const bOffset = b * x.strides[0];\n      for (let r = 0; r < yHeight; r++) {\n        const dxR = r * heightScale;\n        const topDxRIndex = Math.floor(dxR);\n        const bottomDxRIndex = Math.min(Math.ceil(dxR), xHeight - 1);\n\n        const topDxROffset = bOffset + topDxRIndex * x.strides[1];\n        const bottomDxROffset = bOffset + bottomDxRIndex * x.strides[1];\n\n        const dxRLerp = dxR - topDxRIndex;\n        const inverseDxRLerp = 1.0 - dxRLerp;\n        for (let c = 0; c < yWidth; c++) {\n          const dxC = c * widthScale;\n          const leftDxCIndex = Math.floor(dxC);\n          const rightDxCIndex = Math.min(Math.ceil(dxC), xWidth - 1);\n          const dxCLerp = dxC - leftDxCIndex;\n          const inverseDxCLerp = 1.0 - dxCLerp;\n\n          const topLeftRCOffset = topDxROffset + leftDxCIndex * x.strides[2];\n          const topRightRCOffset = topDxROffset + rightDxCIndex * x.strides[2];\n          const bottomLeftRCOffset =\n              bottomDxROffset + leftDxCIndex * x.strides[2];\n          const bottomRightRCOffset =\n              bottomDxROffset + rightDxCIndex * x.strides[2];\n\n          const inverseDxRLerpTimesInverseDxCLerp =\n              inverseDxRLerp * inverseDxCLerp;\n          const inverseDxRLerpTimesDxCLerp = inverseDxRLerp * dxCLerp;\n          const dxRLerpTimesInverseDxCLerp = dxRLerp * inverseDxCLerp;\n          const dxRLerpTimesDxCLerp = dxRLerp * dxCLerp;\n          for (let d = 0; d < depth; d++) {\n            const dyVal = dyValues[offset++];\n            output[topLeftRCOffset + d] +=\n                dyVal * inverseDxRLerpTimesInverseDxCLerp;\n            output[topRightRCOffset + d] += dyVal * inverseDxRLerpTimesDxCLerp;\n            output[bottomLeftRCOffset + d] +=\n                dyVal * dxRLerpTimesInverseDxCLerp;\n            output[bottomRightRCOffset + d] += dyVal * dxRLerpTimesDxCLerp;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, [batch, xWidth, xHeight, depth], x.dtype);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeNearestNeighbor');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = new Float32Array(batch * newHeight * newWidth * numChannels);\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n\n    let outputOffset = 0;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceNearestRow = Math.min(\n            oldHeight - 1,\n            alignCorners ? Math.round(sourceFracRow) :\n                           Math.floor(sourceFracRow));\n        const rowOffset = batchOffset + sourceNearestRow * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceNearestCol = Math.min(\n              oldWidth - 1,\n              alignCorners ? Math.round(sourceFracCol) :\n                             Math.floor(sourceFracCol));\n          const colOffset = rowOffset + sourceNearestCol * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n            // Compute the fractional index of the source.\n            const newVal = xValues[colOffset + d];\n            output[outputOffset++] = newVal;\n          }\n        }\n      }\n    }\n    return tf.tensor(\n        output, [batch, newHeight, newWidth, numChannels], x.dtype);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeNearestNeighborBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    // Loop over the output space.\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < xHeight; r++) {\n        const rowOffset = batchOffset + r * x.strides[1];\n\n        // Compute bounds for where in dy we will look\n        const startRLerp = Math.floor(r * invHeightScale);\n        const startDyR = Math.floor(startRLerp - (winHeight / 2));\n        for (let c = 0; c < xWidth; c++) {\n          const colOffset = rowOffset + c * x.strides[2];\n\n          // Compute bounds for where in dy we will look\n          const startCLerp = Math.floor(c * invWidthScale);\n          const startDyC = Math.floor(startCLerp - (winWidth / 2));\n\n          for (let d = 0; d < depth; d++) {\n            let accum = 0;\n            // loop over dy\n\n            for (let dyRIndex = 0; dyRIndex < winHeight; dyRIndex++) {\n              const dyR = dyRIndex + startDyR;\n              // Guard against the window exceeding the bounds of dy\n              if (dyR < 0 || dyR >= yHeight) {\n                continue;\n              }\n\n              const dyROffset = batchOffset + dyR * dy.strides[1];\n              const sourceFracRow = dyR * heightScale;\n              const sourceNearestRow = Math.min(\n                  xHeight - 1,\n                  alignCorners ? Math.round(sourceFracRow) :\n                                 Math.floor(sourceFracRow));\n              if (r !== sourceNearestRow) {\n                continue;\n              }\n              for (let dyCIndex = 0; dyCIndex < winWidth; dyCIndex++) {\n                const dyC = dyCIndex + startDyC;\n                // Guard against the window exceeding the bounds of dy\n                if (dyC < 0 || dyC >= yWidth) {\n                  continue;\n                }\n\n                const dyCOffset = dyROffset + dyC * dy.strides[2];\n                const sourceFracCol = dyC * widthScale;\n                const sourceNearestCol = Math.min(\n                    xWidth - 1,\n                    alignCorners ? Math.round(sourceFracCol) :\n                                   Math.floor(sourceFracCol));\n\n                if (c === sourceNearestCol) {\n                  accum += dyValues[dyCOffset + d];\n                }\n              }\n            }\n            output[colOffset + d] = accum;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, x.shape, x.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(x, 'localResponseNormalization4D');\n\n    const channels = x.shape[3];\n    const maxD = channels - 1;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const size = x.size;\n    const result = new Float32Array(size);\n\n    function sumAcrossChannels(offset: number) {\n      const currentChannel = offset % channels;\n      let beginSumOffset =\n          offset - currentChannel + Math.max(0, currentChannel - depthRadius);\n      const endSumOffset = offset - currentChannel +\n          Math.min(currentChannel + depthRadius, maxD);\n\n      let sum = 0.0;\n      for (; beginSumOffset <= endSumOffset; beginSumOffset++) {\n        const z = xValues[beginSumOffset];\n        sum += z * z;\n      }\n      return sum;\n    }\n\n    for (let offset = 0; offset < size; offset++) {\n      const sum = sumAcrossChannels(offset);\n      const val = xValues[offset] * Math.pow(bias + alpha * sum, -beta);\n      result[offset] = val;\n    }\n\n    return tf.tensor4d(result, x.shape);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(dy, 'LRNGrad');\n    const channels = dy.shape[3];\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const inputImageValues = this.readSync(inputImage.dataId) as TypedArray;\n    const outputImageValues = this.readSync(outputImage.dataId) as TypedArray;\n    const result = new Float32Array(dy.size);\n    const size = dy.size;\n\n    for (let offset = 0; offset < size; offset++) {\n      const currentChannel = offset % channels;\n      const depthBegin =\n          (offset - currentChannel) + Math.max(0, currentChannel - depthRadius);\n      const depthEnd = (offset - currentChannel) +\n          Math.min(channels, currentChannel + depthRadius + 1);\n\n      let norm = 0;\n      for (let k = depthBegin; k < depthEnd; k++) {\n        norm += Math.pow(inputImageValues[k], 2);\n      }\n      norm = alpha * norm + bias;\n\n      for (let k = depthBegin; k < depthEnd; k++) {\n        let dyi = -2 * alpha * beta * inputImageValues[k] *\n            outputImageValues[offset] / norm;\n        if (offset === k) {\n          dyi += Math.pow(norm, -beta);\n        }\n        dyi *= dyValues[offset];\n        result[k] += dyi;\n      }\n    }\n    return tf.tensor4d(result, dy.shape);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    assertNotComplex(logits, 'multinomial');\n\n    const probabilities = normalized ? logits : tf.softmax(logits);\n    const batchSize = probabilities.shape[0];\n    const numEvents = probabilities.shape[1];\n    const res = tf.zeros<Rank.R2>([batchSize, numSamples], 'int32');\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const probVals = this.readSync(probabilities.dataId) as TypedArray;\n\n    for (let b = 0; b < batchSize; ++b) {\n      const offset = b * numEvents;\n      // The cdf won't include the last event. It will be implicit if no other\n      // event happened.\n      const cdf = new Float32Array(numEvents - 1);\n      cdf[0] = probVals[offset];\n      for (let event = 1; event < cdf.length; ++event) {\n        cdf[event] = cdf[event - 1] + probVals[offset + event];\n      }\n\n      const random = seedrandom.alea(seed.toString());\n      const outOffset = b * numSamples;\n      for (let sampleId = 0; sampleId < numSamples; ++sampleId) {\n        const r = random();\n\n        // Assume last event happened by default.\n        resVals[outOffset + sampleId] = cdf.length;\n\n        for (let event = 0; event < cdf.length; event++) {\n          if (r < cdf[event]) {\n            resVals[outOffset + sampleId] = event;\n            break;\n          }\n        }\n      }\n    }\n    return res;\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    assertNotComplex(indices, 'oneHot');\n\n    const res = new Float32Array(indices.size * depth);\n    res.fill(offValue);\n    const indicesVal = this.readSync(indices.dataId) as TypedArray;\n\n    for (let event = 0; event < indices.size; ++event) {\n      if (indicesVal[event] >= 0 && indicesVal[event] < depth) {\n        res[event * depth + indicesVal[event]] = onValue;\n      }\n    }\n    return tf.tensor2d(res, [indices.size, depth], 'int32');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold: number): Tensor1D {\n    assertNotComplex(boxes, 'nonMaxSuppression');\n\n    const boxesVals = this.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = this.readSync(scores.dataId) as TypedArray;\n    return nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        dataFormat === 'NHWC',\n        () => `Only NHWC dataFormat supported on CPU for depthToSpace. Got ${\n            dataFormat}`);\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = x.shape[1];\n    const inputWidth = x.shape[2];\n    const inputDepth = x.shape[3];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result =\n        new Float32Array(batchSize * outputHeight * outputWidth * outputDepth);\n\n    let outputIdx = 0;\n    for (let b = 0; b < batchSize; ++b) {\n      for (let h = 0; h < outputHeight; ++h) {\n        const inH = Math.floor(h / blockSize);\n        const offsetH = (h % blockSize);\n        for (let w = 0; w < outputWidth; ++w) {\n          const inW = Math.floor(w / blockSize);\n          const offsetW = (w % blockSize);\n          const offsetD = (offsetH * blockSize + offsetW) * outputDepth;\n          for (let d = 0; d < outputDepth; ++d) {\n            const inD = d + offsetD;\n            const inputIdx =\n                inD + inputDepth * (inW + inputWidth * (inH + inputHeight * b));\n            result[outputIdx++] = xValues[inputIdx];\n          }\n        }\n      }\n    }\n    return tf.tensor4d(\n        result, [batchSize, outputHeight, outputWidth, outputDepth]);\n  }\n\n  private broadcastedBinaryOp(\n      a: Tensor, b: Tensor, dtype: DataType,\n      op: (a: number, b: number) => number): Tensor {\n    const newShape = backend_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const result = tf.buffer(newShape, dtype);\n    const aVals = this.readSync(a.dataId) as TypedArray;\n    const bVals = this.readSync(b.dataId) as TypedArray;\n    const aBroadcastDims = backend_util.getBroadcastDims(a.shape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(b.shape, newShape);\n\n    const resVals = result.values;\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resVals.length; ++i) {\n        resVals[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      const aBuf = this.bufferSync(a);\n      const bBuf = this.bufferSync(b);\n      for (let i = 0; i < resVals.length; ++i) {\n        const loc = result.indexToLoc(i);\n\n        const aLoc = loc.slice(-a.rank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = aBuf.locToIndex(aLoc);\n\n        const bLoc = loc.slice(-b.rank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = bBuf.locToIndex(bLoc);\n\n        resVals[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n    return result.toTensor();\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  dispose() {}\n\n  floatPrecision(): 16|32 {\n    return 32;\n  }\n\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return super.epsilon();\n  }\n\n  cropAndResize(\n      images: Tensor4D,\n      boxes: Tensor2D,\n      boxIndex: Tensor1D,\n      cropSize: [number, number],\n      method: string,\n      extrapolationValue: number,\n  ) {\n    const [batch, imageHeight, imageWidth, numChannels] = images.shape;\n    const numBoxes = boxes.shape[0];\n\n    const [cropHeight, cropWidth] = cropSize;\n    const output =\n        tf.buffer([numBoxes, cropHeight, cropWidth, numChannels], 'float32');\n\n    const boxVals = this.readSync(boxes.dataId) as TypedArray;\n    const boxIndVals = this.readSync(boxIndex.dataId) as TypedArray;\n    const imageVals = this.readSync(images.dataId) as TypedArray;\n\n    const inStride = images.strides;   // to calculate flat indexes into image\n    const outStride = output.strides;  // to calculate flat indexes into output\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op.cc\n    for (let b = 0; b < numBoxes; b++) {\n      const startInd = b * 4;\n      const y1 = boxVals[startInd];\n      const x1 = boxVals[startInd + 1];\n      const y2 = boxVals[startInd + 2];\n      const x2 = boxVals[startInd + 3];\n\n      const bInd: number = boxIndVals[b];\n      if (bInd >= batch) {\n        continue;\n      }\n\n      const heightScale = (cropHeight > 1) ?\n          (y2 - y1) * (imageHeight - 1) / (cropHeight - 1) :\n          0;\n      const widthScale =\n          (cropWidth > 1) ? (x2 - x1) * (imageWidth - 1) / (cropWidth - 1) : 0;\n\n      for (let y = 0; y < cropHeight; y++) {\n        const yInd: number = (cropHeight > 1) ?\n            y1 * (imageHeight - 1) + y * (heightScale) :\n            0.5 * (y1 + y2) * (imageHeight - 1);\n\n        if (yInd < 0 || yInd > imageHeight - 1) {\n          for (let x = 0; x < cropWidth; x++) {\n            for (let c = 0; c < numChannels; c++) {\n              const ind =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = extrapolationValue;\n            }\n          }\n          continue;\n        }\n\n        if (method === 'bilinear') {\n          const topInd = Math.floor(yInd);\n          const bottomInd = Math.ceil(yInd);\n          const yLerp = yInd - topInd;\n\n          for (let x = 0; x < cropWidth; x++) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const leftInd = Math.floor(xInd);\n            const rightInd = Math.ceil(xInd);\n            const xLerp = xInd - leftInd;\n\n            for (let c = 0; c < numChannels; c++) {\n              let ind = c + leftInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topRight = imageVals[ind];\n\n              ind = c + leftInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomRight = imageVals[ind];\n\n              const top = topLeft + (topRight - topLeft) * xLerp;\n              const bottom = bottomLeft + (bottomRight - bottomLeft) * xLerp;\n\n              ind = c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = top + ((bottom - top) * yLerp);\n            }\n          }\n        } else {  // method == \"nearest\"\n          for (let x = 0; x < cropWidth; ++x) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const closestX = Math.round(xInd);\n            const closestY = Math.round(yInd);\n            for (let c = 0; c < numChannels; c++) {\n              const inInd = c + closestX * inStride[2] +\n                  closestY * inStride[1] + bInd * inStride[0];\n              const outInd =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[outInd] = imageVals[inInd];\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor4D;\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n    const sumDupeIndices = false;\n    return this.scatter(\n        sparseIndices, sparseValues, outputShape, outputSize, sliceSize,\n        numUpdates, sliceRank, strides, defaultValue, sumDupeIndices);\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n    if (numSlices === 0) {\n      return tf.tensor([], resultShape, x.dtype);\n    }\n\n    const buffer = new TensorBuffer([numSlices, sliceSize], x.dtype);\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const xData = this.readSync(x.dataId) as TypedArray;\n\n    for (let i = 0; i < numSlices; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        flattenIndex += dim * strides[j];\n        index.push(dim);\n      }\n      if (flattenIndex < 0 || flattenIndex >= x.size / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${x.shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        buffer.values[i * sliceSize + k] = xData[flattenIndex * sliceSize + k];\n      }\n    }\n    return buffer.toTensor().reshape(resultShape);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n    const defaultValue = tf.scalar(0);\n    const sumDupeIndices = true;\n    return this.scatter(\n        indices, updates, shape, outputSize, sliceSize, numUpdates, sliceRank,\n        strides, defaultValue, sumDupeIndices);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n    const values =\n        util.getArrayFromDType(dtype, util.sizeFromShape(shape)) as TypedArray;\n    values.fill(value as number);\n    return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported for string tensors');\n    } else {\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    const values = util.getArrayFromDType(\n                       x.dtype, util.sizeFromShape(x.shape)) as TypedArray;\n    return this.makeOutput(values, x.shape, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  private scatter<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R], outputSize: number,\n      sliceSize: number, numUpdates: number, sliceRank: number,\n      strides: number[], defaultValue: Scalar,\n      sumDupeIndices: boolean): Tensor<R> {\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const updatesData = this.readSync(updates.dataId) as TypedArray;\n\n    if (outputSize === 0) {\n      return tf.tensor([], shape, updates.dtype);\n    }\n\n    const buffer = new TensorBuffer(flattenShape, updates.dtype as 'float32');\n    buffer.values.fill((this.readSync(defaultValue.dataId) as TypedArray)[0]);\n\n    for (let i = 0; i < numUpdates; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        index.push(dim);\n        flattenIndex += dim * strides[j];\n      }\n\n      if (flattenIndex < 0 || flattenIndex >= outputSize / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        if (sumDupeIndices) {\n          buffer.values[flattenIndex * sliceSize + k] +=\n              updatesData[i * sliceSize + k];\n        } else {\n          buffer.values[flattenIndex * sliceSize + k] = updates.rank === 0 ?\n              updatesData[0] :\n              updatesData[i * sliceSize + k];\n        }\n      }\n    }\n    return buffer.toTensor().reshape(shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs, AbsInputs, KernelConfig, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function simpleAbsImpl(vals: TypedArray): Float32Array {\n  const resultValues = new Float32Array(vals.length);\n  for (let i = 0; i < vals.length; ++i) {\n    resultValues[i] = Math.abs(vals[i]);\n  }\n  return resultValues;\n}\n\nexport const absKernelFunc =\n    (args: {inputs: AbsInputs, backend: MathBackendCPU}) => {\n      const {x} = args.inputs;\n      const cpuBackend = args.backend;\n      let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n      if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n        resultValues = simpleAbsImpl(values);\n      } else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals =\n            cpuBackend.data.get(real.dataId).values as Float32Array;\n        const imagVals =\n            cpuBackend.data.get(imag.dataId).values as Float32Array;\n        for (let i = 0; i < realVals.length; i++) {\n          const real = realVals[i];\n          const imag = imagVals[i];\n          resultValues[i] = Math.hypot(real, imag);\n        }\n      }\n      return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n    };\n\nexport const absConfig: KernelConfig = {\n  kernelName: Abs,\n  backendName: 'cpu',\n  kernelFunc: absKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleBinaryKernelImpl, SimpleBinaryOperation} from './binary_types';\n\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op: SimpleBinaryOperation):\n    SimpleBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aVals: TypedArray,\n          bVals: TypedArray, dtype: DataType): [TypedArray, number[]] => {\n    const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    const resultRank = newShape.length;\n    const resultStrides = util.computeStrides(newShape);\n    const resultSize = util.sizeFromShape(newShape);\n\n    const result =\n        util.getTypedArrayFromDType(dtype as NumericDataType, resultSize);\n\n    const aRank = aShape.length;\n    const bRank = bShape.length;\n\n    const aStrides = util.computeStrides(aShape);\n    const bStrides = util.computeStrides(bShape);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < result.length; ++i) {\n        result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      for (let i = 0; i < result.length; ++i) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        result[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n\n    return [result, newShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Complex, ComplexInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function complex(args: {inputs: ComplexInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {real, imag} = inputs;\n\n  const realVals = backend.data.get(real.dataId).values as TypedArray;\n  const imagVals = backend.data.get(imag.dataId).values as TypedArray;\n\n  const complexInfo = backend.makeTensorInfo(real.shape, 'complex64');\n\n  const complex = backend.data.get(complexInfo.dataId);\n\n  // The complex tensor owns the underlying real and imag tensorInfos, only the\n  // complex tensor tracks refCount, when complexData is disposed the\n  // underlying tensorData will be disposed.\n  complex.complexTensorInfos = {\n    real: backend.makeTensorInfo(real.shape, 'float32', realVals),\n    imag: backend.makeTensorInfo(imag.shape, 'float32', imagVals)\n  };\n\n  return complexInfo;\n}\n\nexport const complexConfig: KernelConfig = {\n  kernelName: Complex,\n  backendName: 'cpu',\n  kernelFunc: complex as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendCPU}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'cpu',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Real, RealInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function real(args: {inputs: RealInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const real = backend.data.get(input.dataId).complexTensorInfos.real;\n  const realVal = backend.data.get(real.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the real value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(real.shape, real.dtype, realVal);\n}\n\nexport const realConfig: KernelConfig = {\n  kernelName: Real,\n  backendName: 'cpu',\n  kernelFunc: real as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport * as tf from '@tensorflow/tfjs-core';\nimport {Cast, CastAttrs, CastInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\n\nimport {complex} from './Complex';\nimport {identity} from './Identity';\nimport {real} from './Real';\n\nexport function cast(\n    args: {inputs: CastInputs, backend: MathBackendCPU, attrs: CastAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {dtype} = attrs;\n\n  // Casting to complex64.\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return identity({inputs: {x}, backend});\n    }\n\n    // TODO(lina128): Import kernel function once zeros is modularized.\n    const zerosTensor = tf.zeros(x.shape);\n    const floatX = cast({inputs: {x}, backend, attrs: {dtype: 'float32'}});\n\n    const result =\n        complex({inputs: {real: floatX, imag: zerosTensor}, backend});\n\n    zerosTensor.dispose();\n    backend.disposeIntermediateTensorInfo(floatX);\n\n    return result;\n  }\n\n  // Casting from complex64\n  if (x.dtype === 'complex64') {\n    const realPart = real({inputs: {input: x}, backend});\n    const result = cast({inputs: {x: realPart}, backend, attrs: {dtype}});\n\n    backend.disposeIntermediateTensorInfo(realPart);\n\n    return result;\n  }\n\n  if (!util.hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    const result = identity({inputs: {x}, backend});\n    return {dataId: result.dataId, shape: result.shape, dtype};\n  }\n\n  if (dtype === 'int32') {\n    const values = backend.data.get(x.dataId).values as TypedArray;\n    const resultValues = Int32Array.from(values);\n    return backend.makeTensorInfo(x.shape, 'int32', resultValues);\n  }\n\n  if (dtype === 'bool') {\n    // This is essentially the result of notEqual(x, 0). We avoid using\n    // kernel notEqual to avoid circular dependency, i.e. binary_utils ->\n    // cast -> notEqual -> binary_utils.\n    const xVals = backend.data.get(x.dataId).values as TypedArray;\n    const zero = util.toTypedArray([0], x.dtype);\n\n    const [resultData, resultShape] = createSimpleBinaryKernelImpl(\n        (a, b) => (a !== b) ? 1 : 0)(x.shape, [], xVals, zero, 'bool');\n\n    return backend.makeTensorInfo(resultShape, 'bool', resultData);\n  }\n\n  throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n}\n\nexport const castConfig: KernelConfig = {\n  kernelName: Cast,\n  backendName: 'cpu',\n  kernelFunc: cast as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, BinaryInputs, DataType, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {cast} from '../kernels/Cast';\nimport {complex} from '../kernels/Complex';\n\nimport {ComplexBinaryKernelImpl, ComplexBinaryOperation, SimpleBinaryKernelImpl} from './binary_types';\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param name Kernel name.\n * @param binaryKernelImpl A `SimpleBinaryKernelImpl` for the kernel.\n * @param binaryKernelComplexImpl Optional. If exists, represents a\n *     `ComplexBinaryKernelImpl` for the kernel, will be used when input dtype\n *     is `complex64`.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    name: string, simpleImpl: SimpleBinaryKernelImpl,\n    complexImpl?: ComplexBinaryKernelImpl, dtype?: DataType): KernelFunc {\n  if (complexImpl == null) {\n    return ({inputs, backend}) => {\n      const {a, b} = inputs as BinaryInputs;\n      const cpuBackend = backend as MathBackendCPU;\n\n      assertNotComplex([a, b], name);\n\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    };\n  }\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    if (a.dtype === 'complex64' || b.dtype === 'complex64') {\n      const $aComplex = cast(\n          {inputs: {x: a}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $aComplexVals = cpuBackend.data.get($aComplex.dataId);\n\n      const aReal = $aComplexVals.complexTensorInfos.real;\n      const aImag = $aComplexVals.complexTensorInfos.imag;\n\n      const aRealVals =\n          cpuBackend.data.get(aReal.dataId).values as Float32Array;\n      const aImagVals =\n          cpuBackend.data.get(aImag.dataId).values as Float32Array;\n\n      const $bComplex = cast(\n          {inputs: {x: b}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $bComplexVals = cpuBackend.data.get($bComplex.dataId);\n\n      const bReal = $bComplexVals.complexTensorInfos.real;\n      const bImag = $bComplexVals.complexTensorInfos.imag;\n\n      const bRealVals =\n          cpuBackend.data.get(bReal.dataId).values as Float32Array;\n      const bImagVals =\n          cpuBackend.data.get(bImag.dataId).values as Float32Array;\n\n      const [resultRealData, resultImagData, resultShape] = complexImpl(\n          a.shape, b.shape, aRealVals, aImagVals, bRealVals, bImagVals);\n\n      const resultReal =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultRealData);\n\n      const resultImag =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultImagData);\n\n      const result = complex(\n          {inputs: {real: resultReal, imag: resultImag}, backend: cpuBackend});\n\n      cpuBackend.disposeIntermediateTensorInfo($aComplex);\n      cpuBackend.disposeIntermediateTensorInfo($bComplex);\n      cpuBackend.disposeIntermediateTensorInfo(resultReal);\n      cpuBackend.disposeIntermediateTensorInfo(resultImag);\n\n      return result;\n    } else {\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    }\n  };\n}\n\n/**\n * Template that creates the complex type implementation for binary ops.\n * Supports broadcast.\n */\nexport function createComplexBinaryKernelImpl(op: ComplexBinaryOperation):\n    ComplexBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aRealVals: Float32Array,\n          aImagVals: Float32Array, bRealVals: Float32Array,\n          bImagVals: Float32Array): [TypedArray, TypedArray, number[]] => {\n    const resultShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const resultSize = util.sizeFromShape(resultShape);\n    const resultRank = resultShape.length;\n    const resultStrides = util.computeStrides(resultShape);\n\n    const resultRealVals = util.getTypedArrayFromDType('float32', resultSize);\n    const resultImagVals = util.getTypedArrayFromDType('float32', resultSize);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, resultShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, resultShape);\n\n    const aVals = backend_util.mergeRealAndImagArrays(aRealVals, aImagVals);\n    const bVals = backend_util.mergeRealAndImagArrays(bRealVals, bImagVals);\n\n    const aRank = aShape.length;\n    const aStrides = util.computeStrides(aShape);\n\n    const bRank = bShape.length;\n    const bStrides = util.computeStrides(bShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const aIdx = i % aVals.length;\n        const bIdx = i % bVals.length;\n\n        const result =\n            op(aVals[aIdx * 2], aVals[aIdx * 2 + 1], bVals[bIdx * 2],\n               bVals[bIdx * 2 + 1]);\n\n        resultRealVals[i] = result.real;\n        resultImagVals[i] = result.imag;\n      }\n    } else {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        const opResult =\n            op(aVals[aIndex * 2], aVals[aIndex * 2 + 1], bVals[bIndex * 2],\n               bVals[bIndex * 2 + 1]);\n\n        resultRealVals[i] = opResult.real;\n        resultImagVals[i] = opResult.imag;\n      }\n    }\n    return [resultRealVals, resultImagVals, resultShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Add, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal + bReal, imag: aImag + bImag};\n    }));\n\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\n\nexport const addConfig: KernelConfig = {\n  kernelName: Add,\n  backendName: 'cpu',\n  kernelFunc: add\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NumericDataType, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op: SimpleUnaryOperation):\n    SimpleUnaryImpl {\n  return (values, dtype, attrs) => {\n    const newValues =\n        util.getTypedArrayFromDType(dtype as NumericDataType, values.length);\n    for (let i = 0; i < values.length; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return newValues;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelFunc, TypedArray, UnaryInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param name Kernel name.\n * @param op A `SimpleUnaryOperation` for the kernel.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFunc(\n    name: string, op: SimpleUnaryOperation, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xSize = util.sizeFromShape(x.shape);\n    const $dtype = dtype || x.dtype;\n    const newValues = util.getArrayFromDType($dtype, xSize);\n    for (let i = 0; i < xSize; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for unary ops from the given\n * `SimpleUnaryImpl`..\n * @param name Kernel name.\n * @param unaryImpl A `SimpleUnaryImpl` that implements the op.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFuncFromImpl(\n    name: string, unaryImpl: SimpleUnaryImpl, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const $dtype = dtype || x.dtype;\n    const newValues = unaryImpl(values, $dtype, attrs);\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\n\nexport const ceilConfig: KernelConfig = {\n  kernelName: Ceil,\n  backendName: 'cpu',\n  kernelFunc: ceilKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\n\nexport const expConfig: KernelConfig = {\n  kernelName: Exp,\n  backendName: 'cpu',\n  kernelFunc: expKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\n\nexport const expm1Config: KernelConfig = {\n  kernelName: Expm1,\n  backendName: 'cpu',\n  kernelFunc: expm1KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\n\nexport const floorConfig: KernelConfig = {\n  kernelName: Floor,\n  backendName: 'cpu',\n  kernelFunc: floorKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\n\nexport const logConfig: KernelConfig = {\n  kernelName: Log,\n  backendName: 'cpu',\n  kernelFunc: logKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function maxImpl(\n    aVals: TypedArray, reduceSize: number, outShape: number[],\n    dtype: DataType): TypedArray {\n  const vals = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(outShape));\n\n  for (let i = 0; i < vals.length; ++i) {\n    const offset = i * reduceSize;\n    let max = aVals[offset];\n    for (let j = 0; j < reduceSize; ++j) {\n      const value = aVals[offset + j];\n      if (value > max) {\n        max = value;\n      }\n    }\n    vals[i] = max;\n  }\n  return vals;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Multiply} from '@tensorflow/tfjs-core';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const multiplyImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n      };\n    }));\n\nexport const multiply =\n    binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\n\nexport const multiplyConfig: KernelConfig = {\n  kernelName: Multiply,\n  backendName: 'cpu',\n  kernelFunc: multiply\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Rsqrt} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\n\nexport const rsqrtConfig: KernelConfig = {\n  kernelName: Rsqrt,\n  backendName: 'cpu',\n  kernelFunc: rsqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelConfig, KernelFunc, NumericDataType, Slice, slice_util, SliceAttrs, SliceInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function sliceImpl(\n    vals: TypedArray, begin: number[], size: number[], shape: number[],\n    dtype: DataType): TypedArray {\n  const isContinous = slice_util.isSliceContinous(shape, begin, size);\n  const length = util.sizeFromShape(size);\n  const xStrides = util.computeStrides(shape);\n\n  if (isContinous) {\n    const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n    return vals.subarray(flatOffset, flatOffset + length);\n  }\n\n  const outVals = util.getTypedArrayFromDType(dtype as NumericDataType, length);\n  for (let i = 0; i < length; ++i) {\n    const rank = size.length;\n    const strides = util.computeStrides(size);\n    const loc = util.indexToLoc(i, rank, strides);\n    const xLoc = loc.map((idx: number, j) => idx + begin[j]);\n    const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n    outVals[i] = vals[xIndex];\n  }\n  return outVals;\n}\n\nexport function slice(\n    args: {inputs: SliceInputs, backend: MathBackendCPU, attrs: SliceAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {begin, size} = attrs;\n\n  assertNotComplex(x, 'slice');\n\n  const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n  slice_util.assertParamsValid(x, $begin, $size);\n\n  const vals = backend.data.get(x.dataId).values as TypedArray;\n  const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n  return backend.makeTensorInfo($size, x.dtype, outVals);\n}\n\nexport const sliceConfig: KernelConfig = {\n  kernelName: Slice,\n  backendName: 'cpu',\n  kernelFunc: slice as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sub} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const subImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal - bReal, imag: aImag - bImag};\n    }));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\n\nexport const subConfig: KernelConfig = {\n  kernelName: Sub,\n  backendName: 'cpu',\n  kernelFunc: sub\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function transposeImpl(\n    xVals: TypedArray, xShape: number[], dtype: DataType, perm: number[],\n    newShape: number[]): TypedArray {\n  const xRank = xShape.length;\n  const xSize = util.sizeFromShape(xShape);\n  const xStrides = util.computeStrides(xShape);\n  const newStrides = util.computeStrides(newShape);\n\n  const result = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(newShape));\n\n  for (let i = 0; i < xSize; ++i) {\n    const loc = util.indexToLoc(i, xRank, xStrides);\n\n    // Permute location.\n    const newLoc: number[] = new Array(loc.length);\n    for (let i = 0; i < newLoc.length; i++) {\n      newLoc[i] = loc[perm[i]];\n    }\n\n    const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n    result[newIndex] = xVals[i];\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendValues, DataType, TensorBuffer, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function uniqueImpl(\n    values: BackendValues, axis: number, shape: number[], dtype: DataType): {\n  outputValues: BackendValues,\n  outputShape: number[],\n  indices: BackendValues\n} {\n  // Normalize and validate axis.\n  const $axis = util.parseAxisParam(axis, shape)[0];\n\n  // Calculate the new shape that is suitable for extracting data along the\n  // given axis.\n  //\n  // The rank is 3.\n  // The size of the 1st dimension is the size of all the axes < the given axis.\n  // The size of the 2nd dimension is the same as the size of the given axis.\n  // The size of the 3rd dimension is the size of all the axes > the given axis.\n  //\n  // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n  // newShape would be: [2*3, 5, 4].\n  //\n  // Note that this is not the final output shape. This will be the shape for an\n  // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n  // values along the given axis. To demonstrate how it works, consider the\n  // following example:\n  //\n  // Input: a 3D tensor, with shape [1, 2, 3]\n  // [\n  //   [\n  //      [1,2,3],\n  //      [4,5,6]\n  //   ]\n  // ]\n  // Axis: 2 (the last axis).\n  // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n  //\n  // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n  // 1*2. The re-shaped data would look like:\n  //\n  // [\n  //   [\n  //     [1], [2], [3]\n  //   ],\n  //   [\n  //     [4], [5], [6]\n  //   ]\n  // ]\n  //\n  // Then, we can construct a 3-level nested loop by the following dimension\n  // order to extract the values along the axis (dimension1):\n  // i: dimension1       // 0,1,2 (newShape[1])\n  //   m: dimension0     // 0,1   (newShape[0])\n  //     n: dimension2   // 0     (newShape[2])\n  //\n  //                       m, i, n\n  //                      ---------\n  // Iteration 0: data at [0, 0, 0] => \"1\"\n  // Iteration 1: data at [1, 0, 0] => \"4\"\n  // We got [1,4].\n  // Iteration 2: data at [0, 1, 0] => \"2\"\n  // Iteration 3: data at [1, 1, 0] => \"5\"\n  // We got [2,5].\n  // Iteration 4: data at [0, 2, 0] => \"3\"\n  // Iteration 5: data at [1, 2, 0] => \"6\"\n  // We got [3,6].\n  const newShape = [1, shape[0], 1];\n  for (let i = 0; i < $axis; i++) {\n    newShape[0] *= shape[i];\n  }\n  newShape[1] = shape[$axis];\n  for (let i = $axis + 1; i < shape.length; i++) {\n    newShape[2] *= shape[i];\n  }\n\n  // A map from unique elements (their string representations) to their values\n  // in \"indices\" (below).\n  const uniqueElements: {[key: string]: number} = {};\n  // The indices of each unique element in the original tensor along the given\n  // axis. It is 1D and has the same size as the given axis.\n  const indices = new Int32Array(shape[$axis]);\n  // Create a buffer so we can easily extract value at a given location.\n  const inputBuffer = new TensorBuffer(newShape, dtype, values as TypedArray);\n  // The indices along the given axis that have unique elements. This is a\n  // de-duped version of \"indices\" above.\n  const uniqueIndices: number[] = [];\n  const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n  for (let i = 0; i < shape[$axis]; i++) {\n    // Extract values along the axis.\n    let element: string;\n    if (is1DTensor) {\n      // Fast path for 1D tensor input.\n      element = values[i].toString();\n    } else {\n      const axisValues = [];\n      for (let m = 0; m < newShape[0]; m++) {\n        for (let n = 0; n < newShape[2]; n++) {\n          axisValues.push(inputBuffer.get(m, i, n));\n        }\n      }\n      element = axisValues.join(',');\n    }\n\n    // Dedup and update various indices.\n    if (uniqueElements[element] !== undefined) {\n      indices[i] = uniqueElements[element];\n    } else {\n      const uniqueIndex = Object.keys(uniqueElements).length;\n      uniqueElements[element] = uniqueIndex;\n      indices[i] = uniqueIndex;\n      uniqueIndices.push(i);\n    }\n  }\n\n  // Now we know where each of the unique elements are located along the axis\n  // (uniqueIndices). Extract them from input buffer and store them in the\n  // output buffer.\n  const outputTmpShape = newShape.slice();\n  outputTmpShape[1] = Object.keys(uniqueElements).length;\n  const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n  uniqueIndices.forEach((uniqueElementIndex, i) => {\n    for (let m = 0; m < newShape[0]; m++) {\n      for (let n = 0; n < newShape[2]; n++) {\n        outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n      }\n    }\n  });\n\n  // The output shape can be calculated from the input shape with the size of\n  // the given axis replaced by the number of unique elements along that axis.\n  const outputShape = shape.slice();\n  outputShape[$axis] = outputTmpShape[1];\n\n  return {\n    outputValues: outputBuffer.values as BackendValues,\n    outputShape,\n    indices,\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Shared kernel impls for use in other backends.\nexport {simpleAbsImpl} from './kernels/Abs';\nexport {addImpl} from './kernels/Add';\nexport {ceilImpl} from './kernels/Ceil';\nexport {expImpl} from './kernels/Exp';\nexport {expm1Impl} from './kernels/Expm1';\nexport {floorImpl} from './kernels/Floor';\nexport {logImpl} from './kernels/Log';\nexport {maxImpl} from './kernels/Max_impl';\nexport {multiplyImpl} from './kernels/Multiply';\nexport {rsqrtImpl} from './kernels/Rsqrt';\nexport {sliceImpl} from './kernels/Slice';\nexport {subImpl} from './kernels/Sub';\nexport {transposeImpl} from './kernels/Transpose_impl';\nexport {uniqueImpl} from './kernels/Unique_impl';\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/*\n * base.ts contains all the exports from tfjs-backend-cpu\n * without auto-kernel registration\n */\nimport {registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from './backend_cpu';\nimport * as shared from './shared';\n\nexport {MathBackendCPU} from './backend_cpu';\nexport {version as version_cpu} from './version';\nexport {shared};\n\n// Side effects for default initialization of MathBackendCPU\nregisterBackend('cpu', () => new MathBackendCPU(), 1 /* priority */);\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acosKernelFunc = unaryKernelFunc(Acos, (xi) => Math.acos(xi));\n\nexport const acosConfig: KernelConfig = {\n  kernelName: Acos,\n  backendName: 'cpu',\n  kernelFunc: acosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acoshKernelFunc = unaryKernelFunc(Acosh, (xi) => Math.acosh(xi));\n\nexport const acoshConfig: KernelConfig = {\n  kernelName: Acosh,\n  backendName: 'cpu',\n  kernelFunc: acoshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinKernelFunc = unaryKernelFunc(Asin, (xi) => Math.asin(xi));\n\nexport const asinConfig: KernelConfig = {\n  kernelName: Asin,\n  backendName: 'cpu',\n  kernelFunc: asinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinhKernelFunc = unaryKernelFunc(Asinh, (xi) => Math.asinh(xi));\n\nexport const asinhConfig: KernelConfig = {\n  kernelName: Asinh,\n  backendName: 'cpu',\n  kernelFunc: asinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanKernelFunc = unaryKernelFunc(Atan, (xi) => Math.atan(xi));\n\nexport const atanConfig: KernelConfig = {\n  kernelName: Atan,\n  backendName: 'cpu',\n  kernelFunc: atanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanhKernelFunc = unaryKernelFunc(Atanh, (xi) => Math.atanh(xi));\n\nexport const atanhConfig: KernelConfig = {\n  kernelName: Atanh,\n  backendName: 'cpu',\n  kernelFunc: atanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, buffer, DataType, Rank, TensorBuffer, TypedArray} from '@tensorflow/tfjs-core';\n\nexport function pool(\n    xValues: TypedArray, xShape: number[], dtype: DataType, strides: number[],\n    convInfo: backend_util.Conv2DInfo,\n    poolType: 'max'|'avg'): TensorBuffer<Rank, DataType> {\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const initialValue =\n      (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                            Number.POSITIVE_INFINITY);\n\n  const output = buffer(convInfo.outShape, dtype);\n  const outputVals = output.values;\n\n  const outputBatchStrides =\n      convInfo.outShape[1] * convInfo.outShape[2] * convInfo.outShape[3];\n  const outputRowStrides = convInfo.outShape[2] * convInfo.outShape[3];\n  const outputColStrides = convInfo.outShape[3];\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    const outputBatchOffset = b * outputBatchStrides;\n    const inputBatchOffset = b * strides[0];\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        const outputRowOffset = outputBatchOffset + yR * outputRowStrides;\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          const xCMin = Math.max(0, xCCorner);\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let minMaxValue = initialValue;\n          let avgValue = 0;\n          let count = 0;\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const xROffset = inputBatchOffset + xR * strides[1];\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const xCOffset = xROffset + xC * strides[2];\n              const pixel = xValues[xCOffset + d];\n              if ((poolType === 'max' && pixel > minMaxValue)) {\n                minMaxValue = pixel;\n              } else if (poolType === 'avg') {\n                avgValue += pixel;\n                count++;\n              }\n            }\n            if (isNaN(minMaxValue)) {\n              break;\n            }\n          }\n          const outputOffset = outputRowOffset + yC * outputColStrides + d;\n          outputVals[outputOffset] =\n              poolType === 'avg' ? avgValue / count : minMaxValue;\n        }\n      }\n    }\n  }\n  return output;\n}\n\nexport function maxPoolPositions(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    convInfo: backend_util.Conv2DInfo, flattenPositions = false,\n    includeBatchInIndex = false): TensorBuffer<Rank, 'int32'> {\n  const maxPositions = buffer(convInfo.outShape, 'int32');\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const xBuf = buffer(xShape, dtype, xValues);\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        let xRMin = xRCorner;\n        while (xRMin < 0) {\n          xRMin += dilationHeight;\n        }\n        // const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          let xCMin = xCCorner;\n          while (xCMin < 0) {\n            xCMin += dilationWidth;\n          }\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let maxValue = Number.NEGATIVE_INFINITY;\n          let maxPosition = -1;\n\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const wR = xR - xRCorner;\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const wC = xC - xCCorner;\n              const pixel = xBuf.get(b, xR, xC, d);\n              if (pixel > maxValue) {\n                maxValue = pixel as number;\n                if (flattenPositions) {\n                  maxPosition = includeBatchInIndex ?\n                      ((b * convInfo.inHeight + xR) * convInfo.inWidth + xC) *\n                              convInfo.inChannels +\n                          d :\n                      (xR * convInfo.inWidth + xC) * convInfo.inChannels + d;\n                } else {\n                  maxPosition = wR * effectiveFilterWidth + wC;\n                }\n              }\n            }\n          }\n          maxPositions.set(maxPosition, b, yR, yC, d);\n        }\n      }\n    }\n  }\n  return maxPositions;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function avgPool(\n    args:\n        {inputs: AvgPoolInputs, backend: MathBackendCPU, attrs: AvgPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'avg');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'cpu',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, buffer, KernelConfig, KernelFunc, Rank, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const filterHeight = convInfo.filterHeight;\n  const filterWidth = convInfo.filterWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel;\n            }\n          }\n          dx.set(dotProd * avgMultiplier, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function batchNormKernelFunc(args: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendCPU,\n  attrs: FusedBatchNormAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x, scale, offset, mean, variance} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  assertNotComplex([x, mean, variance, scale, offset], 'batchNorm');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const mVals = backend.data.get(mean.dataId).values as TypedArray;\n  const varVals = backend.data.get(variance.dataId).values as TypedArray;\n  const sVals = scale ? backend.data.get(scale.dataId).values as TypedArray :\n                        new Float32Array([1]);\n  const offVals = offset ?\n      backend.data.get(offset.dataId).values as TypedArray :\n      new Float32Array([0]);\n  const outVals = new Float32Array(xVals.length);\n\n  const offValsLength = offVals.length;\n  const sValsLength = sVals.length;\n  const varValsLength = varVals.length;\n  const mValsLength = mVals.length;\n\n  let offi = 0;\n  let mi = 0;\n  let si = 0;\n  let vi = 0;\n  for (let i = 0; i < xVals.length; ++i) {\n    outVals[i] = offVals[offi++] +\n        (xVals[i] - mVals[mi++]) * sVals[si++] /\n            Math.sqrt(varVals[vi++] + varianceEpsilon);\n    if (offi >= offValsLength) {\n      offi = 0;\n    }\n    if (mi >= mValsLength) {\n      mi = 0;\n    }\n    if (si >= sValsLength) {\n      si = 0;\n    }\n    if (vi >= varValsLength) {\n      vi = 0;\n    }\n  }\n  return backend.makeTensorInfo(x.shape, x.dtype, outVals);\n}\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'cpu',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const clipKernelFunc = unaryKernelFunc(ClipByValue, (xi, attrs) => {\n  const clipAttrs = attrs as {} as ClipByValueAttrs;\n  if (xi > clipAttrs.clipValueMax) {\n    return clipAttrs.clipValueMax;\n  }\n  return xi < clipAttrs.clipValueMin ? clipAttrs.clipValueMin : xi;\n});\n\nexport const clipConfig: KernelConfig = {\n  kernelName: ClipByValue,\n  backendName: 'cpu',\n  kernelFunc: clipKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Imag, ImagInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function imag(args: {inputs: ImagInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const imag = backend.data.get(input.dataId).complexTensorInfos.imag;\n  const imagVal = backend.data.get(imag.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the imag value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(imag.shape, imag.dtype, imagVal);\n}\n\nexport const imagConfig: KernelConfig = {\n  kernelName: Imag,\n  backendName: 'cpu',\n  kernelFunc: imag as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function reshape(\n    args:\n        {inputs: ReshapeInputs, backend: MathBackendCPU, attrs: ReshapeAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  backend.incRef(x.dataId);\n\n  const xData = backend.data.get(x.dataId);\n\n  if (xData.complexTensorInfos != null) {\n    const real = xData.complexTensorInfos.real;\n    const imag = xData.complexTensorInfos.imag;\n\n    real.shape = $shape;\n    imag.shape = $shape;\n  }\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'cpu',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Concat, ConcatAttrs, ConcatInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nimport {complex} from './Complex';\nimport {imag} from './Imag';\nimport {real} from './Real';\nimport {reshape} from './Reshape';\n\nexport function concat(\n    args: {inputs: ConcatInputs, backend: MathBackendCPU, attrs: ConcatAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {axis} = attrs;\n\n  const $axis = util.parseAxisParam(axis, inputs[0].shape)[0];\n  let outShape = backend_util.computeOutShape(inputs.map(t => t.shape), $axis);\n\n  if (util.sizeFromShape(outShape) === 0) {\n    return backend.makeTensorInfo(outShape, inputs[0].dtype, []);\n  }\n\n  // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n  const $inputs = inputs.filter(t => util.sizeFromShape(t.shape) > 0);\n  if ($inputs.length === 1) {\n    return $inputs[0];\n  }\n\n  const shapes = $inputs.map(t => t.shape);\n  backend_util.assertParamsConsistent(shapes, $axis);\n\n  if ($inputs[0].dtype === 'complex64') {\n    const reals = $inputs.map((t) => real({inputs: {input: t}, backend}));\n    const imags = $inputs.map((t) => imag({inputs: {input: t}, backend}));\n\n    const realConcated = concat({inputs: reals, backend, attrs: {axis}});\n    const imagConcated = concat({inputs: imags, backend, attrs: {axis}});\n\n    const result =\n        complex({inputs: {real: realConcated, imag: imagConcated}, backend});\n\n    reals.forEach(r => backend.disposeIntermediateTensorInfo(r));\n    imags.forEach(i => backend.disposeIntermediateTensorInfo(i));\n    backend.disposeIntermediateTensorInfo(realConcated);\n    backend.disposeIntermediateTensorInfo(imagConcated);\n\n    return result;\n  }\n\n  // Any concat of n-dimensional tensors across any axis can be reduced to\n  // a concatenation of two-dimensional tensors across the axis 1 by first\n  // partitioning the axes of the original tensors into those less than the\n  // axis to be concatenated and the rest. Then reshape the tensors\n  // into a two-dimensional tensor by collapsing these two sets of axes and\n  // concatenate the resulting matrices across the axis 1, finally reshaping\n  // the result to have the proper shape.\n  const inputs2D = $inputs.map(t => {\n    const innerSize = util.sizeFromShape(t.shape.slice($axis));\n    const shape = [-1, innerSize];\n    return reshape({inputs: {x: t}, backend, attrs: {shape}});\n  });\n\n  // Concats 2d tensors along axis=1.\n  outShape =\n      backend_util.computeOutShape(inputs2D.map(t => t.shape), 1 /* axis */);\n\n  const outVals = util.getTypedArrayFromDType(\n      $inputs[0].dtype as 'float32', util.sizeFromShape(outShape));\n\n  if (inputs2D[0].shape[0] === 1) {\n    // Use built-in TypedArray.set() method for speed.\n    let offset = 0;\n    inputs2D.forEach(t => {\n      const val = backend.data.get(t.dataId).values as TypedArray;\n      const size = util.sizeFromShape(t.shape);\n\n      outVals.set(val, offset);\n      offset += size;\n    });\n  } else {\n    let colOffset = 0;\n\n    inputs2D.forEach(t => {\n      const tVals = backend.data.get(t.dataId).values as TypedArray;\n\n      let tIdx = 0;\n\n      for (let row = 0; row < t.shape[0]; ++row) {\n        const resIdx = row * outShape[1] + colOffset;\n        for (let col = 0; col < t.shape[1]; ++col) {\n          outVals[resIdx + col] = tVals[tIdx++];\n        }\n      }\n\n      colOffset += t.shape[1];\n    });\n  }\n\n  const finalOutShape =\n      backend_util.computeOutShape($inputs.map(t => t.shape), $axis);\n\n  const outInfo =\n      backend.makeTensorInfo(finalOutShape, inputs[0].dtype, outVals);\n\n  inputs2D.forEach(t => backend.disposeIntermediateTensorInfo(t));\n\n  return outInfo;\n}\n\nexport const concatConfig: KernelConfig = {\n  kernelName: Concat,\n  backendName: 'cpu',\n  kernelFunc: concat as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const cosKernelFunc = unaryKernelFunc(Cos, (xi) => Math.cos(xi));\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'cpu',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const coshKernelFunc = unaryKernelFunc(Cosh, (xi) => Math.cosh(xi));\n\nexport const coshConfig: KernelConfig = {\n  kernelName: Cosh,\n  backendName: 'cpu',\n  kernelFunc: coshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2D, Dilation2DAttrs, Dilation2DInputs, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dConfig: KernelConfig = {\n  kernelName: Dilation2D,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter} = inputs as Dilation2DInputs;\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xRank = x.shape.length;\n\n    const filterVals = cpuBackend.data.get(filter.dataId).values as TypedArray;\n    const filterRank = filter.shape.length;\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    const outSize = util.sizeFromShape(outShape);\n    const outRank = outShape.length;\n    const outputVals = util.getArrayFromDType(x.dtype, outSize);\n\n    // Upsampling the input by fill in `dilation size - 1` values between each\n    // input value.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const xIndex = util.locToIndex(\n                        [b, hIn, wIn, d], xRank, util.computeStrides(x.shape));\n                    const filterIndex = util.locToIndex(\n                        [h, w, d], filterRank,\n                        util.computeStrides(filter.shape));\n                    const val = xVals[xIndex] + filterVals[filterIndex];\n                    if (val > curVal) {\n                      curVal = val;\n                    }\n                  }\n                }\n              }\n            }\n            const outputIndex = util.locToIndex(\n                [b, hOut, wOut, d], outRank, util.computeStrides(outShape));\n            outputVals[outputIndex] = curVal;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(outputVals, x.dtype), outShape, x.dtype);\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropFilter, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropFilterConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropFilter,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropFilter}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed filter gradients has the same dimensions as the filter:\n    // [filterHeight, filterWidth, depth]\n    const gradients = util.makeZerosNestedTypedArray(\n                          filter.shape, filter.dtype) as number[][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hMax = 0;\n            let wMax = 0;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hMax = h;\n                      wMax = w;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[hMax][wMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), filter.shape, filter.dtype);\n\n    return {dataId, shape: filter.shape, dtype: filter.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropInput, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropInputConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropInput,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropInput}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed gradients has the same dimensions as the input:\n    // [batch, inputHeight, inputCols, inChannel]\n    const gradients =\n        util.makeZerosNestedTypedArray(x.shape, x.dtype) as number[][][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hInMax = (hBeg < 0) ? 0 : hBeg;\n            let wInMax = (wBeg < 0) ? 0 : wBeg;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hInMax = hIn;\n                      wInMax = wIn;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[b][hInMax][wInMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), x.shape, x.dtype);\n\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const divImpl =\n    createSimpleBinaryKernelImpl((a: number, b: number) => a / b);\nexport const div = binaryKernelFunc(Div, divImpl);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'cpu',\n  kernelFunc: div\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Elu, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const eluKernelFunc =\n    unaryKernelFunc(Elu, (xi) => xi >= 0 ? xi : (Math.exp(xi) - 1));\n\nexport const eluConfig: KernelConfig = {\n  kernelName: Elu,\n  backendName: 'cpu',\n  kernelFunc: eluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Erf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst p = backend_util.ERF_P;\nconst a1 = backend_util.ERF_A1;\nconst a2 = backend_util.ERF_A2;\nconst a3 = backend_util.ERF_A3;\nconst a4 = backend_util.ERF_A4;\nconst a5 = backend_util.ERF_A5;\n\nexport const erfKernelFunc = unaryKernelFunc(\n    Erf,\n    (xi) => {\n      const sign = Math.sign(xi);\n      const v = Math.abs(xi);\n      const t = 1.0 / (1.0 + p * v);\n      return sign *\n          (1.0 -\n           (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t *\n               Math.exp(-v * v));\n    },\n);\n\nexport const erfConfig: KernelConfig = {\n  kernelName: Erf,\n  backendName: 'cpu',\n  kernelFunc: erfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Tensor, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {add} from '../kernels/Add';\nimport {complex} from '../kernels/Complex';\nimport {concat} from '../kernels/Concat';\nimport {divConfig} from '../kernels/Div';\nimport {identity} from '../kernels/Identity';\nimport {imag} from '../kernels/Imag';\nimport {multiply} from '../kernels/Multiply';\nimport {real} from '../kernels/Real';\nimport {slice} from '../kernels/Slice';\nimport {sub} from '../kernels/Sub';\n\n/**\n * Calculate FFT of inner most elements of batch tensor.\n */\nexport function fftBatch(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): TensorInfo {\n  const inputShape = input.shape;\n  const batch = inputShape[0];\n  const innerDim = inputShape[1];\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const real2D = inputVals.complexTensorInfos.real;\n  const imag2D = inputVals.complexTensorInfos.imag;\n\n  // Collects real and imaginary values separately.\n  const resultShape = [batch, innerDim];\n  const resultSize = util.sizeFromShape(resultShape);\n  const resultReal = util.getTypedArrayFromDType('float32', resultSize);\n  const resultImag = util.getTypedArrayFromDType('float32', resultSize);\n\n  for (let b = 0; b < batch; b++) {\n    // TODO: Support slice ops for complex type.\n    const r = slice({\n      inputs: {x: real2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n    const i = slice({\n      inputs: {x: imag2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n\n    const input = complex({inputs: {real: r, imag: i}, backend: cpuBackend});\n\n    // Run FFT by batch element.\n    const {real, imag} = fftImpl(input, inverse, cpuBackend);\n    const res = backend_util.mergeRealAndImagArrays(real, imag);\n\n    for (let d = 0; d < innerDim; d++) {\n      const c = backend_util.getComplexWithIndex(res, d);\n      resultReal[b * innerDim + d] = c.real;\n      resultImag[b * innerDim + d] = c.imag;\n    }\n\n    cpuBackend.disposeIntermediateTensorInfo(r);\n    cpuBackend.disposeIntermediateTensorInfo(i);\n    cpuBackend.disposeIntermediateTensorInfo(input);\n  }\n\n  const $realInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultReal);\n  const $imagInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultImag);\n\n  const result = complex(\n      {inputs: {real: $realInfo, imag: $imagInfo}, backend: cpuBackend});\n\n  cpuBackend.disposeIntermediateTensorInfo($realInfo);\n  cpuBackend.disposeIntermediateTensorInfo($imagInfo);\n\n  return result;\n}\n\nexport function fftImpl(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  const inputSize = util.sizeFromShape(input.shape);\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const realVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.real.dataId).values as\n      Float32Array;\n\n  const imagVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.imag.dataId).values as\n      Float32Array;\n\n  if (isExponentOf2(inputSize)) {\n    const result =\n        fftRadix2(realVals, imagVals, inputSize, inverse, cpuBackend);\n\n    const resultShape = [input.shape[0], input.shape[1]];\n\n    if (inverse) {\n      const realInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.real);\n      const imagInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.imag);\n\n      const sizeInfo: TensorInfo = cpuBackend.makeTensorInfo(\n          [], 'float32',\n          util.createScalarValue(inputSize as {} as 'float32', 'float32'));\n      const sizeInfoCopy =\n          identity({inputs: {x: sizeInfo}, backend: cpuBackend});\n\n      const divRealInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: realInfo, b: sizeInfo}, backend: cpuBackend}) as\n          TensorInfo;\n      const divImagInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: imagInfo, b: sizeInfoCopy}, backend: cpuBackend}) as\n          TensorInfo;\n\n      const divRealVals =\n          cpuBackend.data.get(divRealInfo.dataId).values as Float32Array;\n      const divImagVals =\n          cpuBackend.data.get(divImagInfo.dataId).values as Float32Array;\n\n      cpuBackend.disposeIntermediateTensorInfo(realInfo);\n      cpuBackend.disposeIntermediateTensorInfo(imagInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfoCopy);\n      cpuBackend.disposeIntermediateTensorInfo(divRealInfo);\n      cpuBackend.disposeIntermediateTensorInfo(divImagInfo);\n\n      return {real: divRealVals, imag: divImagVals};\n    }\n\n    return result;\n  } else {\n    const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n    const rawOutput =\n        fourierTransformByMatmul(data, inputSize, inverse) as Float32Array;\n\n    return backend_util.splitRealAndImagArrays(rawOutput);\n  }\n}\n\nfunction isExponentOf2(size: number): boolean {\n  return (size & size - 1) === 0;\n}\n\n// FFT using Cooley-Tukey algorithm on radix 2 dimensional input.\nfunction fftRadix2(\n    realVals: Float32Array, imagVals: Float32Array, size: number,\n    inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  if (size === 1) {\n    return {real: realVals, imag: imagVals};\n  }\n\n  const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n  const half = size / 2;\n\n  const evenComplex = backend_util.complexWithEvenIndex(data);\n\n  const evenRealVals = evenComplex.real;\n  const evenImagVals = evenComplex.imag;\n\n  const evenShape = [evenRealVals.length];\n\n  const evenRealInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenRealVals);\n  const evenImagInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenImagVals);\n\n  const evenTensorInfo = complex(\n      {inputs: {real: evenRealInfo, imag: evenImagInfo}, backend: cpuBackend});\n\n  const oddComplex = backend_util.complexWithOddIndex(data);\n\n  const oddRealVals = oddComplex.real;\n  const oddImagVals = oddComplex.imag;\n\n  const oddShape = [oddRealVals.length];\n\n  const oddRealInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddRealVals);\n  const oddImagInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddImagVals);\n\n  const oddTensorInfo = complex(\n      {inputs: {real: oddRealInfo, imag: oddImagInfo}, backend: cpuBackend});\n\n  // Recursive call for half part of original input.\n  const $evenComplex =\n      fftRadix2(evenRealVals, evenImagVals, half, inverse, cpuBackend);\n\n  const $evenRealVals = $evenComplex.real;\n  const $evenImagVals = $evenComplex.imag;\n\n  const $evenShape = [$evenRealVals.length];\n\n  const $evenRealInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenRealVals);\n  const $evenImagInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenImagVals);\n\n  const $evenTensorInfo = complex({\n    inputs: {real: $evenRealInfo, imag: $evenImagInfo},\n    backend: cpuBackend\n  });\n\n  const $oddComplex =\n      fftRadix2(oddRealVals, oddImagVals, half, inverse, cpuBackend);\n\n  const $oddRealVals = $oddComplex.real;\n  const $oddImagVals = $oddComplex.imag;\n\n  const $oddShape = [$oddRealVals.length];\n\n  const $oddRealInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddRealVals);\n  const $oddImagInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddImagVals);\n\n  const $oddTensorInfo = complex(\n      {inputs: {real: $oddRealInfo, imag: $oddImagInfo}, backend: cpuBackend});\n\n  const e = backend_util.exponents(size, inverse);\n  const eShape = [e.real.length];\n\n  const eRealInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.real);\n  const eImagInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.imag);\n\n  const complexInfo = complex(\n      {inputs: {real: eRealInfo, imag: eImagInfo}, backend: cpuBackend});\n\n  const exponentInfo =\n      multiply(\n          {inputs: {a: complexInfo, b: $oddTensorInfo}, backend: cpuBackend}) as\n      TensorInfo;\n\n  const addPart = add({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n  const subPart = sub({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n\n  const addPartReal = real({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartReal = real({inputs: {input: subPart}, backend: cpuBackend});\n\n  const addPartImag = imag({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartImag = imag({inputs: {input: subPart}, backend: cpuBackend});\n\n  const $real = concat({\n    inputs: [addPartReal as Tensor, subPartReal as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n  const $imag = concat({\n    inputs: [addPartImag as Tensor, subPartImag as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n\n  const $realVals = cpuBackend.data.get($real.dataId).values as Float32Array;\n  const $imagVals = cpuBackend.data.get($imag.dataId).values as Float32Array;\n\n  cpuBackend.disposeIntermediateTensorInfo(evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(complexInfo);\n  cpuBackend.disposeIntermediateTensorInfo(exponentInfo);\n  cpuBackend.disposeIntermediateTensorInfo(addPart);\n  cpuBackend.disposeIntermediateTensorInfo(subPart);\n  cpuBackend.disposeIntermediateTensorInfo(addPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(addPartImag);\n  cpuBackend.disposeIntermediateTensorInfo(subPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(subPartImag);\n  cpuBackend.disposeIntermediateTensorInfo($real);\n  cpuBackend.disposeIntermediateTensorInfo($imag);\n\n  return {real: $realVals, imag: $imagVals};\n}\n\n// Calculate fourier transform by multplying sinusoid matrix.\nfunction fourierTransformByMatmul(\n    data: TypedArray, size: number, inverse: boolean): TypedArray {\n  const ret = new Float32Array(size * 2);\n  // TODO: Use matmul instead once it supports complex64 type.\n  for (let r = 0; r < size; r++) {\n    let real = 0.0;\n    let imag = 0.0;\n    for (let c = 0; c < size; c++) {\n      const e = backend_util.exponent(r * c, size, inverse);\n      const term = backend_util.getComplexWithIndex(data as Float32Array, c);\n      real += term.real * e.real - term.imag * e.imag;\n      imag += term.real * e.imag + term.imag * e.real;\n    }\n    if (inverse) {\n      real /= size;\n      imag /= size;\n    }\n    backend_util.assignToTypedArray(ret, real, imag, r);\n  }\n  return ret;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FFT, FFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function fft(args: {inputs: FFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, false, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const fftConfig: KernelConfig = {\n  kernelName: FFT,\n  backendName: 'cpu',\n  kernelFunc: fft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n\n            const coordX = Math.round(imageWidth - x);\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n\n            let outputValue = imageVals[outIdx];\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth) {\n              // set the output to the image value at the coordinate position.\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n            output[outIdx] = outputValue;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IFFT, IFFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function ifft(args: {inputs: IFFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, true, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const ifftConfig: KernelConfig = {\n  kernelName: IFFT,\n  backendName: 'cpu',\n  kernelFunc: ifft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isFiniteKernelFunc =\n    unaryKernelFunc(IsFinite, (xi) => Number.isFinite(xi) ? 1 : 0, 'bool');\n\nexport const isFiniteConfig: KernelConfig = {\n  kernelName: IsFinite,\n  backendName: 'cpu',\n  kernelFunc: isFiniteKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isInfKernelFunc =\n    unaryKernelFunc(IsInf, (xi) => Math.abs(xi) === Infinity ? 1 : 0, 'bool');\n\nexport const isInfConfig: KernelConfig = {\n  kernelName: IsInf,\n  backendName: 'cpu',\n  kernelFunc: isInfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isNaNKernelFunc =\n    unaryKernelFunc(IsNan, (xi) => Number.isNaN(xi) ? 1 : 0, 'bool');\n\nexport const isNaNConfig: KernelConfig = {\n  kernelName: IsNan,\n  backendName: 'cpu',\n  kernelFunc: isNaNKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log1p} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const log1pKernelFunc = unaryKernelFunc(Log1p, (xi) => Math.log1p(xi));\n\nexport const log1pConfig: KernelConfig = {\n  kernelName: Log1p,\n  backendName: 'cpu',\n  kernelFunc: log1pKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, LogicalNot} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const logicalNotKernelFunc =\n    unaryKernelFunc(LogicalNot, (xi) => xi ? 0 : 1, 'bool');\n\nexport const logicalNotConfig: KernelConfig = {\n  kernelName: LogicalNot,\n  backendName: 'cpu',\n  kernelFunc: logicalNotKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig} from '@tensorflow/tfjs-core';\nimport {TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    let xShape = x.shape;\n    const xRank = xShape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, xShape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    let xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    if (permutedAxes != null) {\n      const newShape: number[] = new Array(xRank);\n      for (let i = 0; i < newShape.length; i++) {\n        newShape[i] = xShape[permutedAxes[i]];\n      }\n\n      xVals = transposeImpl(xVals, xShape, x.dtype, permutedAxes, newShape);\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n\n      xShape = newShape;\n    }\n\n    assertNotComplex(x, 'max');\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(xShape, axes);\n\n    const reduceSize = util.sizeFromShape(reduceShape);\n\n    const result = maxImpl(xVals, reduceSize, maxOutShape, x.dtype);\n    const dataId = cpuBackend.write(result, maxOutShape, x.dtype);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // reshape\n      const newShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n      outShape = newShape;\n    }\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function maxPool(\n    args:\n        {inputs: MaxPoolInputs, backend: MathBackendCPU, attrs: MaxPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'max');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'cpu',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, buffer, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, Rank, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {maxPoolPositions} from '../utils/pool_utils';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const xValues = backend.data.get(x.dataId).values as TypedArray;\n  const maxPosBuf = buffer(\n      convInfo.outShape, x.dtype,\n      maxPoolPositions(xValues, x.shape, x.dtype, convInfo).values);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n              const maxPos = effectiveFilterHeight * effectiveFilterWidth - 1 -\n                  (maxPosBuf.get(b, dyR, dyC, d) as number);\n              const curPos = wR * effectiveFilterWidth + wC;\n\n              const mask = maxPos === curPos ? 1 : 0;\n              if (mask === 0) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel * mask;\n            }\n          }\n          dx.set(dotProd, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, DataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {maxPoolPositions, pool} from '../utils/pool_utils';\nexport function maxPoolWithArgmaxImpl(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    includeBatchInIndex: boolean, convInfo: backend_util.Conv2DInfo) {\n  const strides = util.computeStrides(xShape);\n  const maxPools = pool(xValues, xShape, dtype, strides, convInfo, 'max');\n  const maxPositions = maxPoolPositions(\n      xValues, xShape, dtype, convInfo, true, includeBatchInIndex);\n\n  return [maxPools.values, maxPositions.values];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'MaxPoolWithArgmax');\n\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        [1, 1], pad);\n    const [pooled, indexes] = maxPoolWithArgmaxImpl(\n        values, x.shape, x.dtype, includeBatchInIndex, convInfo);\n\n    const pooledDataId =\n        cpuBackend.write(pooled as Float32Array, convInfo.outShape, x.dtype);\n    const indexesDataId =\n        cpuBackend.write(indexes as Int32Array, convInfo.outShape, x.dtype);\n    return [\n      {dataId: pooledDataId, shape: convInfo.outShape, dtype: x.dtype},\n      {dataId: indexesDataId, shape: convInfo.outShape, dtype: 'int32'}\n    ];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionPadded');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionWithScore');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NotEqual} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const notEqualImpl =\n    createSimpleBinaryKernelImpl(((a, b) => (a !== b) ? 1 : 0));\nexport const notEqual =\n    binaryKernelFunc(NotEqual, notEqualImpl, null /* complexOp */, 'bool');\n\nexport const notEqualConfig: KernelConfig = {\n  kernelName: NotEqual,\n  backendName: 'cpu',\n  kernelFunc: notEqual\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, NumericDataType, PadV2, PadV2Attrs, PadV2Inputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function padV2(\n    args: {inputs: PadV2Inputs, backend: MathBackendCPU, attrs: PadV2Attrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {paddings, constantValue} = attrs;\n\n  assertNotComplex(x, 'pad');\n\n  const outShape = paddings.map(\n      (p, i) => p[0] /* beforePad */ + x.shape[i] + p[1] /* afterPad */);\n\n  const start = paddings.map(p => p[0]);\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const xSize = util.sizeFromShape(x.shape);\n  const xRank = x.shape.length;\n  const xStrides = util.computeStrides(x.shape);\n\n  const resultSize = util.sizeFromShape(outShape);\n  const resultRank = outShape.length;\n  const resultStrides = util.computeStrides(outShape);\n  const resVals =\n      util.getTypedArrayFromDType(x.dtype as NumericDataType, resultSize);\n\n  if (constantValue !== 0) {\n    resVals.fill(constantValue);\n  }\n\n  for (let i = 0; i < xSize; i++) {\n    const coords = util.indexToLoc(i, xRank, xStrides);\n    const outCoords = coords.map((c, i) => c + start[i]);\n    const outIndex = util.locToIndex(outCoords, resultRank, resultStrides);\n\n    resVals[outIndex] = xVals[i];\n  }\n\n  const outId = backend.write(resVals, outShape, x.dtype);\n\n  return {dataId: outId, shape: outShape, dtype: x.dtype};\n}\n\nexport const padV2Config: KernelConfig = {\n  kernelName: PadV2,\n  backendName: 'cpu',\n  kernelFunc: padV2 as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Reciprocal} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const reciprocalKernelFunc = unaryKernelFunc(Reciprocal, (xi) => 1 / xi);\n\nexport const reciprocalConfig: KernelConfig = {\n  kernelName: Reciprocal,\n  backendName: 'cpu',\n  kernelFunc: reciprocalKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {backend_util, RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const fullOpacityValue = 255;\n\n    const sinFactor = Math.sin(radians);\n    const cosFactor = Math.cos(radians);\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n            const y = coords[1];\n\n            // coordX/coordY are the result of rotating and translating x/y.\n            let coordX = (x - centerX) * cosFactor - (y - centerY) * sinFactor;\n            let coordY = (x - centerX) * sinFactor + (y - centerY) * cosFactor;\n            coordX = Math.round(coordX + centerX);\n            coordY = Math.round(coordY + centerY);\n\n            let outputValue = fillValue;\n            if (typeof fillValue !== 'number') {\n              if (channel === 3) {\n                outputValue = fullOpacityValue;\n              } else {\n                outputValue = fillValue[channel];\n              }\n            }\n\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth && coordY >= 0 &&\n                coordY < imageHeight) {\n              // set the output to the image value at the coordinate position.\n              const rotatedRowOffset = coordY * (imageWidth * numChannels);\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rotatedRowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n            output[outIdx] = outputValue as number;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Round} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const roundKernelFunc = unaryKernelFunc(Round, (xi) => {\n  // The algorithm is based on banker's rounding.\n  const base = Math.floor(xi);\n  if (xi - base < 0.5) {\n    return Math.floor(xi);\n  } else if (xi - base > 0.5) {\n    return Math.ceil(xi);\n  } else {\n    if (base % 2.0 === 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n});\n\nexport const roundConfig: KernelConfig = {\n  kernelName: Round,\n  backendName: 'cpu',\n  kernelFunc: roundKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, Selu} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst scaleAlpha = backend_util.SELU_SCALEALPHA;\nconst scale = backend_util.SELU_SCALE;\n\nexport const seluKernelFunc = unaryKernelFunc(Selu, (xi) => {\n  if (xi >= 0) {\n    return scale * xi;\n  } else {\n    return scaleAlpha * (Math.exp(xi) - 1);\n  }\n});\n\nexport const seluConfig: KernelConfig = {\n  kernelName: Selu,\n  backendName: 'cpu',\n  kernelFunc: seluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sigmoid} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sigmoidKernelFunc =\n    unaryKernelFunc(Sigmoid, (xi) => 1 / (1 + Math.exp(-xi)));\n\nexport const sigmoidConfig: KernelConfig = {\n  kernelName: Sigmoid,\n  backendName: 'cpu',\n  kernelFunc: sigmoidKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sign} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const signKernelFunc = unaryKernelFunc(Sign, (xi) => {\n  if (xi < 0) {\n    return -1;\n  } else if (xi > 0) {\n    return 1;\n  } else {\n    return 0;\n  }\n});\n\nexport const signConfig: KernelConfig = {\n  kernelName: Sign,\n  backendName: 'cpu',\n  kernelFunc: signKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinKernelFunc = unaryKernelFunc(Sin, (xi) => Math.sin(xi));\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'cpu',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sinh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinhKernelFunc = unaryKernelFunc(Sinh, (xi) => Math.sinh(xi));\n\nexport const sinhConfig: KernelConfig = {\n  kernelName: Sinh,\n  backendName: 'cpu',\n  kernelFunc: sinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Softplus} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\n// mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n\n// epsilon is the difference between 1.0 and the next representable float.\n// For a single precision 32 bit float this should be 2^-23, see:\n// https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\nconst epsilon = 1.1920928955078125e-7;\nconst threshold = Math.log(epsilon) + 2.0;\n\nexport const softplusKernelFunc = unaryKernelFunc(Softplus, (xi) => {\n  // Value above which exp(x) may overflow, but softplus(x) == x\n  // is within machine epsilon.\n  const tooLarge = xi > -threshold;\n\n  // Value below which exp(x) may underflow, but softplus(x) == exp(x)\n  // is within machine epsilon.\n  const tooSmall = xi < threshold;\n\n  const expX = Math.exp(xi);\n  let result;\n\n  if (tooSmall) {\n    result = expX;\n  } else if (tooLarge) {\n    result = xi;\n  } else {\n    result = Math.log(1.0 + expX);\n  }\n  return result;\n});\n\nexport const softplusConfig: KernelConfig = {\n  kernelName: Softplus,\n  backendName: 'cpu',\n  kernelFunc: softplusKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {transposeImpl} from './Transpose_impl';\n\nexport function transpose(args: {\n  inputs: TransposeInputs,\n  attrs: TransposeAttrs,\n  backend: MathBackendCPU\n}): TensorInfo {\n  const {inputs, attrs, backend} = args;\n  const {x} = inputs;\n  const {perm} = attrs;\n\n  assertNotComplex(x, 'transpose');\n\n  const xRank = x.shape.length;\n\n  const newShape: number[] = new Array(xRank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = x.shape[perm[i]];\n  }\n\n  const values = backend.data.get(x.dataId).values as TypedArray;\n  const result = transposeImpl(values, x.shape, x.dtype, perm, newShape);\n\n  const dataId = backend.write(result, newShape, x.dtype);\n  return {dataId, shape: newShape, dtype: x.dtype};\n}\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'cpu',\n  kernelFunc: transpose as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, KernelFunc, ReshapeAttrs, ReshapeInputs, SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs, TensorInfo, TransposeAttrs, TransposeInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {padV2Config} from './PadV2';\nimport {reshape} from './Reshape';\nimport {transpose} from './Transpose';\n\nexport function spaceToBatchND(args: {\n  inputs: SpaceToBatchNDInputs,\n  backend: MathBackendCPU,\n  attrs: SpaceToBatchNDAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {blockShape, paddings} = attrs;\n\n  assertNotComplex([x], 'spaceToBatchND');\n\n  const prod = util.sizeFromShape(blockShape);\n\n  const completePaddings: Array<[number, number]> = [[0, 0]];\n  completePaddings.push(...(paddings as Array<[number, number]>));\n\n  for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n    completePaddings.push([0, 0]);\n  }\n\n  const paddedX = padV2Config.kernelFunc({\n    inputs: {x},\n    backend,\n    attrs: {paddings: completePaddings, constantValue: 0}\n  }) as TensorInfo;\n\n  const reshapedPaddedShape =\n      backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n  const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n      reshapedPaddedShape.length, blockShape.length, false);\n\n  const flattenShape =\n      backend_util.getReshapedPermuted(paddedX.shape, blockShape, prod, false);\n\n  const reshapeInputs: ReshapeInputs = {x: paddedX};\n  const reshapeAttrs: ReshapeAttrs = {shape: reshapedPaddedShape};\n  const paddedXReshaped =\n      reshape({inputs: reshapeInputs, backend, attrs: reshapeAttrs});\n\n  const transposeInputs: TransposeInputs = {x: paddedXReshaped};\n  const transposeAttrs:\n      TransposeAttrs = {perm: permutedReshapedPaddedPermutation};\n  const paddedXT =\n      transpose({inputs: transposeInputs, backend, attrs: transposeAttrs});\n\n  const resultReshapeInputs: ReshapeInputs = {x: paddedXT};\n  const resultReshapeAttrs: ReshapeAttrs = {shape: flattenShape};\n  const result = reshape(\n      {inputs: resultReshapeInputs, backend, attrs: resultReshapeAttrs});\n\n  backend.disposeIntermediateTensorInfo(paddedX);\n  backend.disposeIntermediateTensorInfo(paddedXReshaped);\n  backend.disposeIntermediateTensorInfo(paddedXT);\n\n  return result;\n}\n\nexport const spaceToBatchNDConfig: KernelConfig = {\n  kernelName: SpaceToBatchND,\n  backendName: 'cpu',\n  kernelFunc: spaceToBatchND as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sqrt} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sqrtKernelFunc = unaryKernelFunc(Sqrt, (xi) => Math.sqrt(xi));\n\nexport const sqrtConfig: KernelConfig = {\n  kernelName: Sqrt,\n  backendName: 'cpu',\n  kernelFunc: sqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square, SquareInputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend}) => {\n    const {x} = inputs as SquareInputs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'square');\n\n    const values = cpuBackend.data.get(x.dataId).values as Float32Array;\n    const newValues = new Float32Array(values.length);\n    for (let i = 0; i < values.length; ++i) {\n      const value = values[i];\n      newValues[i] = value * value;\n    }\n    const dataId = cpuBackend.write(newValues, x.shape, x.dtype);\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const squaredDifferenceImpl = createSimpleBinaryKernelImpl(((a, b) => {\n  const diff = a - b;\n  return diff * diff;\n}));\nexport const squaredDifference =\n    binaryKernelFunc(SquaredDifference, squaredDifferenceImpl);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'cpu',\n  kernelFunc: squaredDifference\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Step, StepAttrs} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const stepKernelFunc = unaryKernelFunc(Step, (xi, attrs) => {\n  const stepAttrs = attrs as {} as StepAttrs;\n  if (isNaN(xi)) {\n    return NaN;\n  } else {\n    return xi > 0 ? 1 : stepAttrs.alpha;\n  }\n});\n\nexport const stepConfig: KernelConfig = {\n  kernelName: Step,\n  backendName: 'cpu',\n  kernelFunc: stepKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanKernelFunc = unaryKernelFunc(Tan, (xi) => Math.tan(xi));\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'cpu',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tanh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanhKernelFunc = unaryKernelFunc(Tanh, (xi) => Math.tanh(xi));\n\nexport const tanhConfig: KernelConfig = {\n  kernelName: Tanh,\n  backendName: 'cpu',\n  kernelFunc: tanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {uniqueImpl} from './Unique_impl';\n\nexport function unique(\n    args: {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendCPU}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  const values = backend.data.get(x.dataId).values;\n  const {outputValues, outputShape, indices} =\n      uniqueImpl(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'cpu',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// We explicitly import the modular kernels so they get registered in the\n// global registry when we compile the library. A modular build would replace\n// the contents of this file and import only the kernels that are needed.\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {absConfig} from './kernels/Abs';\nimport {acosConfig} from './kernels/Acos';\nimport {acoshConfig} from './kernels/Acosh';\nimport {addConfig} from './kernels/Add';\nimport {asinConfig} from './kernels/Asin';\nimport {asinhConfig} from './kernels/Asinh';\nimport {atanConfig} from './kernels/Atan';\nimport {atanhConfig} from './kernels/Atanh';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {castConfig} from './kernels/Cast';\nimport {ceilConfig} from './kernels/Ceil';\nimport {clipConfig} from './kernels/Clip';\nimport {complexConfig} from './kernels/Complex';\nimport {concatConfig} from './kernels/Concat';\nimport {cosConfig} from './kernels/Cos';\nimport {coshConfig} from './kernels/Cosh';\nimport {dilation2dConfig} from './kernels/Dilation2D';\nimport {dilation2dBackpropFilterConfig} from './kernels/Dilation2DBackpropFilter';\nimport {dilation2dBackpropInputConfig} from './kernels/Dilation2DBackpropInput';\nimport {divConfig} from './kernels/Div';\nimport {eluConfig} from './kernels/Elu';\nimport {erfConfig} from './kernels/Erf';\nimport {expConfig} from './kernels/Exp';\nimport {expm1Config} from './kernels/Expm1';\nimport {fftConfig} from './kernels/FFT';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {floorConfig} from './kernels/Floor';\nimport {identityConfig} from './kernels/Identity';\nimport {ifftConfig} from './kernels/IFFT';\nimport {imagConfig} from './kernels/Imag';\nimport {isFiniteConfig} from './kernels/IsFinite';\nimport {isInfConfig} from './kernels/IsInf';\nimport {isNaNConfig} from './kernels/IsNaN';\nimport {logConfig} from './kernels/Log';\nimport {log1pConfig} from './kernels/Log1p';\nimport {logicalNotConfig} from './kernels/LogicalNot';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {multiplyConfig} from './kernels/Multiply';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {notEqualConfig} from './kernels/NotEqual';\nimport {padV2Config} from './kernels/PadV2';\nimport {realConfig} from './kernels/Real';\nimport {reciprocalConfig} from './kernels/Reciprocal';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {roundConfig} from './kernels/Round';\nimport {rsqrtConfig} from './kernels/Rsqrt';\nimport {seluConfig} from './kernels/Selu';\nimport {sigmoidConfig} from './kernels/Sigmoid';\nimport {signConfig} from './kernels/Sign';\nimport {sinConfig} from './kernels/Sin';\nimport {sinhConfig} from './kernels/Sinh';\nimport {sliceConfig} from './kernels/Slice';\nimport {softplusConfig} from './kernels/Softplus';\nimport {spaceToBatchNDConfig} from './kernels/SpaceToBatchND';\nimport {sqrtConfig} from './kernels/Sqrt';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {stepConfig} from './kernels/Step';\nimport {subConfig} from './kernels/Sub';\nimport {tanConfig} from './kernels/Tan';\nimport {tanhConfig} from './kernels/Tanh';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  absConfig,\n  acosConfig,\n  acoshConfig,\n  addConfig,\n  asinConfig,\n  asinhConfig,\n  atanConfig,\n  atanhConfig,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  castConfig,\n  ceilConfig,\n  clipConfig,\n  complexConfig,\n  concatConfig,\n  cosConfig,\n  coshConfig,\n  dilation2dConfig,\n  dilation2dBackpropInputConfig,\n  dilation2dBackpropFilterConfig,\n  divConfig,\n  eluConfig,\n  erfConfig,\n  expConfig,\n  expm1Config,\n  fftConfig,\n  flipLeftRightConfig,\n  floorConfig,\n  identityConfig,\n  ifftConfig,\n  imagConfig,\n  isFiniteConfig,\n  isInfConfig,\n  isNaNConfig,\n  logConfig,\n  log1pConfig,\n  logicalNotConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  maxConfig,\n  multiplyConfig,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  notEqualConfig,\n  padV2Config,\n  realConfig,\n  reciprocalConfig,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  roundConfig,\n  rsqrtConfig,\n  seluConfig,\n  sigmoidConfig,\n  signConfig,\n  sinConfig,\n  sinhConfig,\n  sliceConfig,\n  softplusConfig,\n  spaceToBatchNDConfig,\n  sqrtConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  stepConfig,\n  subConfig,\n  tanConfig,\n  tanhConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst contexts: {[key: string]: WebGLRenderingContext} = {};\n\nconst WEBGL_ATTRIBUTES: WebGLContextAttributes = {\n  alpha: false,\n  antialias: false,\n  premultipliedAlpha: false,\n  preserveDrawingBuffer: false,\n  depth: false,\n  stencil: false,\n  failIfMajorPerformanceCaveat: true\n};\n\nexport function setWebGLContext(\n    webGLVersion: number, gl: WebGLRenderingContext) {\n  contexts[webGLVersion] = gl;\n}\n\nexport function getWebGLContext(webGLVersion: number): WebGLRenderingContext {\n  if (!(webGLVersion in contexts)) {\n    const newCtx = getWebGLRenderingContext(webGLVersion);\n    if (newCtx !== null) {\n      contexts[webGLVersion] = newCtx;\n    } else {\n      console.log('Could not get context for WebGL version', webGLVersion);\n      return null;\n    }\n  }\n  const gl = contexts[webGLVersion];\n  if (gl.isContextLost()) {\n    delete contexts[webGLVersion];\n    return getWebGLContext(webGLVersion);\n  }\n\n  gl.disable(gl.DEPTH_TEST);\n  gl.disable(gl.STENCIL_TEST);\n  gl.disable(gl.BLEND);\n  gl.disable(gl.DITHER);\n  gl.disable(gl.POLYGON_OFFSET_FILL);\n  gl.disable(gl.SAMPLE_COVERAGE);\n  gl.enable(gl.SCISSOR_TEST);\n  gl.enable(gl.CULL_FACE);\n  gl.cullFace(gl.BACK);\n\n  return contexts[webGLVersion];\n}\n\nfunction createCanvas(webGLVersion: number) {\n  if (typeof OffscreenCanvas !== 'undefined' && webGLVersion === 2) {\n    return new OffscreenCanvas(300, 150);\n  } else if (typeof document !== 'undefined') {\n    return document.createElement('canvas');\n  } else {\n    throw new Error('Cannot create a canvas in this context');\n  }\n}\n\nfunction getWebGLRenderingContext(webGLVersion: number): WebGLRenderingContext {\n  if (webGLVersion !== 1 && webGLVersion !== 2) {\n    throw new Error('Cannot get WebGL rendering context, WebGL is disabled.');\n  }\n  const canvas = createCanvas(webGLVersion);\n\n  canvas.addEventListener('webglcontextlost', (ev: Event) => {\n    ev.preventDefault();\n    delete contexts[webGLVersion];\n  }, false);\n  if (webGLVersion === 1) {\n    return (canvas.getContext('webgl', WEBGL_ATTRIBUTES) ||\n            canvas.getContext('experimental-webgl', WEBGL_ATTRIBUTES)) as\n        WebGLRenderingContext;\n  }\n  return canvas.getContext('webgl2', WEBGL_ATTRIBUTES) as WebGLRenderingContext;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {backend_util, DataId, DataType, Tensor, util} from '@tensorflow/tfjs-core';\n\nexport enum PackingScheme {\n  /**\n   * All values in a single texel are densely packed without any constraints.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 4]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   010|011   020|021\n   * -------   -------   -------\n   * 002|003   012|013   022|023\n   *\n   * 100|101   110|111   120|121\n   * -------   -------   -------\n   * 102|103   112|113   122|123\n   *\n   */\n  DENSE,\n\n  /**\n   * Single texels contain only values from the same batch, and from adjacent\n   * rows and columns.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 5]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   002|003   004|xxx   020|021   022|023   024|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   * 100|101   102|103   104|xxx   120|121   122|123   124|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   */\n  SHARED_BATCH\n}\n\nexport enum TextureUsage {\n  RENDER,\n  UPLOAD,\n  PIXELS,\n  DOWNLOAD\n}\n\nexport enum PhysicalTextureType {\n  UNPACKED_FLOAT16,\n  UNPACKED_FLOAT32,\n  PACKED_4X1_UNSIGNED_BYTE,\n  PACKED_2X2_FLOAT32,\n  PACKED_2X2_FLOAT16\n}\n\nexport interface TextureData {\n  // Required.\n  shape: number[];\n  dtype: DataType;\n\n  // Optional.\n  values?: backend_util.BackendValues;\n  texture?: WebGLTexture;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensors field. When this is defined, texture will be null.\n  complexTensors?: {real: Tensor, imag: Tensor};\n  /** [rows, columns] shape of the texture. */\n  texShape?: [number, number];\n  usage?: TextureUsage;\n  isPacked?: boolean;\n\n  refCount: number;\n\n  // Available when the tensor has been sliced.\n  slice?: {\n    // Offset in the 'flat index' space.\n    flatOffset: number;\n    // Used for counting how many sliced tensors point to the same texture.\n    origDataId: DataId;\n  };\n}\n\nexport function getUnpackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns, rows];\n}\n\nexport function getUnpackedArraySizeFromMatrixSize(\n    matrixSize: number, channelsPerTexture: number): number {\n  return matrixSize * channelsPerTexture;\n}\n\nexport function getColorMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns * 4, rows];\n}\n\n/**\n * Get shape for densely packed RGBA texture.\n */\nexport function getDenseTexShape(shape: number[]): [number, number] {\n  const size = util.sizeFromShape(shape);\n  const texelsNeeded = Math.ceil(size / 4);\n  return util.sizeToSquarishShape(texelsNeeded);\n}\n\nexport function getMatrixSizeFromUnpackedArraySize(\n    unpackedSize: number, channelsPerTexture: number): number {\n  if (unpackedSize % channelsPerTexture !== 0) {\n    throw new Error(\n        `unpackedSize (${unpackedSize}) must be a multiple of ` +\n        `${channelsPerTexture}`);\n  }\n  return unpackedSize / channelsPerTexture;\n}\n\nexport function decodeMatrixFromUnpackedColorRGBAArray(\n    unpackedArray: Float32Array, matrix: Float32Array, channels: number) {\n  const requiredSize = unpackedArray.length * channels / 4;\n  if (matrix.length < requiredSize) {\n    throw new Error(\n        `matrix length (${matrix.length}) must be >= ${requiredSize}`);\n  }\n  let dst = 0;\n  for (let src = 0; src < unpackedArray.length; src += 4) {\n    for (let c = 0; c < channels; c++) {\n      matrix[dst++] = unpackedArray[src + c];\n    }\n  }\n}\n\nexport function getPackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [\n    Math.max(1, Math.ceil(columns / 2)), Math.max(1, Math.ceil(rows / 2))\n  ];\n}\n\nexport function getPackedRGBAArraySizeFromMatrixShape(\n    rows: number, columns: number): number {\n  const [w, h] = getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return w * h * 4;\n}\n\nexport interface TextureConfig {\n  internalFormatFloat: number;\n  textureFormatFloat: number;\n  internalFormatPackedHalfFloat: number;\n  internalFormatHalfFloat: number;\n  internalFormatPackedFloat: number;\n\n  // The format to use during a gl.readPixels call.\n  downloadTextureFormat: number;\n  // How many channels need to be unpacked after a gl.readPixels call.\n  downloadUnpackNumChannels: number;\n\n  defaultNumChannels: number;\n  textureTypeHalfFloat: number;\n  textureTypeFloat: number;\n}\n\nexport function getTextureConfig(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension?: any): TextureConfig {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n\n  let internalFormatFloat: number;\n  let internalFormatHalfFloat: number;\n  let internalFormatPackedHalfFloat: number;\n  let internalFormatPackedFloat: number;\n  let textureFormatFloat: number;\n\n  let downloadTextureFormat: number;\n  let downloadUnpackNumChannels: number;\n\n  let defaultNumChannels: number;\n  let textureTypeHalfFloat: number;\n  let textureTypeFloat: number;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    internalFormatFloat = glany.R32F;\n    internalFormatHalfFloat = glany.R16F;\n    internalFormatPackedHalfFloat = glany.RGBA16F;\n    internalFormatPackedFloat = glany.RGBA32F;\n    textureFormatFloat = glany.RED;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 1;\n    textureTypeHalfFloat = glany.HALF_FLOAT;\n    textureTypeFloat = glany.FLOAT;\n  } else {\n    internalFormatFloat = gl.RGBA;\n    internalFormatHalfFloat = gl.RGBA;\n    internalFormatPackedHalfFloat = gl.RGBA;\n    internalFormatPackedFloat = glany.RGBA;\n    textureFormatFloat = gl.RGBA;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 4;\n    textureTypeHalfFloat = textureHalfFloatExtension != null ?\n        textureHalfFloatExtension.HALF_FLOAT_OES :\n        null;\n    textureTypeFloat = gl.FLOAT;\n  }\n  downloadTextureFormat = gl.RGBA;\n\n  return {\n    internalFormatFloat,\n    internalFormatHalfFloat,\n    internalFormatPackedHalfFloat,\n    internalFormatPackedFloat,\n    textureFormatFloat,\n    downloadTextureFormat,\n    downloadUnpackNumChannels,\n    defaultNumChannels,\n    textureTypeHalfFloat,\n    textureTypeFloat\n  };\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext} from './canvas_util';\nimport {getTextureConfig} from './tex_util';\n\nexport function callAndCheck<T>(gl: WebGLRenderingContext, func: () => T): T {\n  const returnValue = func();\n  if (env().getBool('DEBUG')) {\n    checkWebGLError(gl);\n  }\n  return returnValue;\n}\n\nfunction checkWebGLError(gl: WebGLRenderingContext) {\n  const error = gl.getError();\n  if (error !== gl.NO_ERROR) {\n    throw new Error('WebGL Error: ' + getWebGLErrorMessage(gl, error));\n  }\n}\n\n// https://en.wikipedia.org/wiki/Half-precision_floating-point_format\nconst MIN_FLOAT16 = 5.96e-8;\nconst MAX_FLOAT16 = 65504;\n\nexport function canBeRepresented(num: number): boolean {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED') || num === 0 ||\n      (MIN_FLOAT16 < Math.abs(num) && Math.abs(num) < MAX_FLOAT16)) {\n    return true;\n  }\n  return false;\n}\n\nexport function getWebGLErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.NO_ERROR:\n      return 'NO_ERROR';\n    case gl.INVALID_ENUM:\n      return 'INVALID_ENUM';\n    case gl.INVALID_VALUE:\n      return 'INVALID_VALUE';\n    case gl.INVALID_OPERATION:\n      return 'INVALID_OPERATION';\n    case gl.INVALID_FRAMEBUFFER_OPERATION:\n      return 'INVALID_FRAMEBUFFER_OPERATION';\n    case gl.OUT_OF_MEMORY:\n      return 'OUT_OF_MEMORY';\n    case gl.CONTEXT_LOST_WEBGL:\n      return 'CONTEXT_LOST_WEBGL';\n    default:\n      return `Unknown error code ${status}`;\n  }\n}\n\nexport function getExtensionOrThrow(\n    gl: WebGLRenderingContext, extensionName: string): {} {\n  return throwIfNull<{}>(\n      gl, () => gl.getExtension(extensionName),\n      'Extension \"' + extensionName + '\" not supported on this browser.');\n}\n\nexport function createVertexShader(\n    gl: WebGLRenderingContext, vertexShaderSource: string): WebGLShader {\n  const vertexShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.VERTEX_SHADER),\n      'Unable to create vertex WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(vertexShader, vertexShaderSource));\n  callAndCheck(gl, () => gl.compileShader(vertexShader));\n  if (gl.getShaderParameter(vertexShader, gl.COMPILE_STATUS) === false) {\n    console.log(gl.getShaderInfoLog(vertexShader));\n    throw new Error('Failed to compile vertex shader.');\n  }\n  return vertexShader;\n}\n\nexport function createFragmentShader(\n    gl: WebGLRenderingContext, fragmentShaderSource: string): WebGLShader {\n  const fragmentShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.FRAGMENT_SHADER),\n      'Unable to create fragment WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(fragmentShader, fragmentShaderSource));\n  callAndCheck(gl, () => gl.compileShader(fragmentShader));\n  if (gl.getShaderParameter(fragmentShader, gl.COMPILE_STATUS) === false) {\n    logShaderSourceAndInfoLog(\n        fragmentShaderSource, gl.getShaderInfoLog(fragmentShader));\n    throw new Error('Failed to compile fragment shader.');\n  }\n  return fragmentShader;\n}\n\nconst lineNumberRegex = /ERROR: [0-9]+:([0-9]+):/g;\nfunction logShaderSourceAndInfoLog(\n    shaderSource: string, shaderInfoLog: string) {\n  const lineNumberRegexResult = lineNumberRegex.exec(shaderInfoLog);\n  if (lineNumberRegexResult == null) {\n    console.log(`Couldn't parse line number in error: ${shaderInfoLog}`);\n    console.log(shaderSource);\n    return;\n  }\n\n  const lineNumber = +lineNumberRegexResult[1];\n\n  const shaderLines = shaderSource.split('\\n');\n  const pad = shaderLines.length.toString().length + 2;\n  const linesWithLineNumbers = shaderLines.map(\n      (line, lineNumber) =>\n          util.rightPad((lineNumber + 1).toString(), pad) + line);\n  let maxLineLength = 0;\n  for (let i = 0; i < linesWithLineNumbers.length; i++) {\n    maxLineLength = Math.max(linesWithLineNumbers[i].length, maxLineLength);\n  }\n\n  const beforeErrorLines = linesWithLineNumbers.slice(0, lineNumber - 1);\n  const errorLine = linesWithLineNumbers.slice(lineNumber - 1, lineNumber);\n  const afterErrorLines = linesWithLineNumbers.slice(lineNumber);\n\n  console.log(beforeErrorLines.join('\\n'));\n  console.log(shaderInfoLog.split('\\n')[0]);\n  console.log(\n      `%c ${util.rightPad(errorLine[0], maxLineLength)}`,\n      'border:1px solid red; background-color:#e3d2d2; color:#a61717');\n  console.log(afterErrorLines.join('\\n'));\n}\n\nexport function createProgram(gl: WebGLRenderingContext): WebGLProgram {\n  return throwIfNull<WebGLProgram>(\n      gl, () => gl.createProgram(), 'Unable to create WebGLProgram.');\n}\n\nexport function linkProgram(gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.linkProgram(program));\n  if (gl.getProgramParameter(program, gl.LINK_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Failed to link vertex and fragment shaders.');\n  }\n}\n\nexport function validateProgram(\n    gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.validateProgram(program));\n  if (gl.getProgramParameter(program, gl.VALIDATE_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Shader program validation failed.');\n  }\n}\n\nexport function createStaticVertexBuffer(\n    gl: WebGLRenderingContext, data: Float32Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(gl, () => gl.bufferData(gl.ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function createStaticIndexBuffer(\n    gl: WebGLRenderingContext, data: Uint16Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl, () => gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function getNumChannels(): number {\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    return 1;\n  }\n  return 4;\n}\n\nexport function createTexture(gl: WebGLRenderingContext): WebGLTexture {\n  return throwIfNull<WebGLTexture>(\n      gl, () => gl.createTexture(), 'Unable to create WebGLTexture.');\n}\n\nexport function validateTextureSize(width: number, height: number) {\n  const maxTextureSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if ((width <= 0) || (height <= 0)) {\n    const requested = `[${width}x${height}]`;\n    throw new Error('Requested texture size ' + requested + ' is invalid.');\n  }\n  if ((width > maxTextureSize) || (height > maxTextureSize)) {\n    const requested = `[${width}x${height}]`;\n    const max = `[${maxTextureSize}x${maxTextureSize}]`;\n    throw new Error(\n        'Requested texture size ' + requested +\n        ' greater than WebGL maximum on this browser / GPU ' + max + '.');\n  }\n}\n\nexport function createFramebuffer(gl: WebGLRenderingContext): WebGLFramebuffer {\n  return throwIfNull<WebGLFramebuffer>(\n      gl, () => gl.createFramebuffer(), 'Unable to create WebGLFramebuffer.');\n}\n\nexport function bindVertexBufferToProgramAttribute(\n    gl: WebGLRenderingContext, program: WebGLProgram, attribute: string,\n    buffer: WebGLBuffer, arrayEntriesPerItem: number, itemStrideInBytes: number,\n    itemOffsetInBytes: number): boolean {\n  const loc = gl.getAttribLocation(program, attribute);\n  if (loc === -1) {\n    // The GPU compiler decided to strip out this attribute because it's unused,\n    // thus no need to bind.\n    return false;\n  }\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl,\n      () => gl.vertexAttribPointer(\n          loc, arrayEntriesPerItem, gl.FLOAT, false, itemStrideInBytes,\n          itemOffsetInBytes));\n  callAndCheck(gl, () => gl.enableVertexAttribArray(loc));\n  return true;\n}\n\nexport function bindTextureUnit(\n    gl: WebGLRenderingContext, texture: WebGLTexture, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n}\n\nexport function unbindTextureUnit(\n    gl: WebGLRenderingContext, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function getProgramUniformLocationOrThrow(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return throwIfNull<WebGLUniformLocation>(\n      gl, () => gl.getUniformLocation(program, uniformName),\n      'uniform \"' + uniformName + '\" not present in program.');\n}\n\nexport function getProgramUniformLocation(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return gl.getUniformLocation(program, uniformName);\n}\n\nexport function bindTextureToProgramUniformSampler(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    uniformSamplerLocation: WebGLUniformLocation, textureUnit: number) {\n  callAndCheck(gl, () => bindTextureUnit(gl, texture, textureUnit));\n  callAndCheck(gl, () => gl.uniform1i(uniformSamplerLocation, textureUnit));\n}\n\nexport function bindCanvasToFramebuffer(gl: WebGLRenderingContext) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n  callAndCheck(gl, () => gl.viewport(0, 0, gl.canvas.width, gl.canvas.height));\n  callAndCheck(gl, () => gl.scissor(0, 0, gl.canvas.width, gl.canvas.height));\n}\n\nexport function bindColorTextureToFramebuffer(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0));\n}\n\nexport function unbindColorTextureFromFramebuffer(\n    gl: WebGLRenderingContext, framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, null, 0));\n}\n\nexport function validateFramebuffer(gl: WebGLRenderingContext) {\n  const status = gl.checkFramebufferStatus(gl.FRAMEBUFFER);\n  if (status !== gl.FRAMEBUFFER_COMPLETE) {\n    throw new Error(\n        'Error binding framebuffer: ' + getFramebufferErrorMessage(gl, status));\n  }\n}\n\nexport function getFramebufferErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:\n      return 'FRAMEBUFFER_INCOMPLETE_DIMENSIONS';\n    case gl.FRAMEBUFFER_UNSUPPORTED:\n      return 'FRAMEBUFFER_UNSUPPORTED';\n    default:\n      return `unknown error ${status}`;\n  }\n}\n\nfunction throwIfNull<T>(\n    gl: WebGLRenderingContext, returnTOrNull: () => T | null,\n    failureMessage: string): T {\n  const tOrNull: T|null = callAndCheck(gl, () => returnTOrNull());\n  if (tOrNull == null) {\n    throw new Error(failureMessage);\n  }\n  return tOrNull;\n}\n\nfunction validateTextureUnit(gl: WebGLRenderingContext, textureUnit: number) {\n  const maxTextureUnit = gl.MAX_COMBINED_TEXTURE_IMAGE_UNITS - 1;\n  const glTextureUnit = textureUnit + gl.TEXTURE0;\n  if (glTextureUnit < gl.TEXTURE0 || glTextureUnit > maxTextureUnit) {\n    const textureUnitRange = `[gl.TEXTURE0, gl.TEXTURE${maxTextureUnit}]`;\n    throw new Error(`textureUnit must be in ${textureUnitRange}.`);\n  }\n}\n\nexport function getBatchDim(shape: number[], dimsToSkip = 2): number {\n  return util.sizeFromShape(shape.slice(0, shape.length - dimsToSkip));\n}\n\nexport function getRowsCols(shape: number[]): [number, number] {\n  if (shape.length === 0) {\n    throw Error('Cannot get rows and columns of an empty shape array.');\n  }\n\n  return [\n    shape.length > 1 ? shape[shape.length - 2] : 1, shape[shape.length - 1]\n  ];\n}\n\nexport function getShapeAs3D(shape: number[]): [number, number, number] {\n  let shapeAs3D: [number, number, number] = [1, 1, 1];\n  const isScalar = shape.length === 0 || (shape.length === 1 && shape[0] === 1);\n  if (!isScalar) {\n    shapeAs3D =\n        [getBatchDim(shape), ...getRowsCols(shape)] as [number, number, number];\n  }\n  return shapeAs3D;\n}\n\nexport function getTextureShapeFromLogicalShape(\n    logShape: number[], isPacked = false): [number, number] {\n  let maxTexSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if (isPacked) {\n    maxTexSize = maxTexSize * 2;\n\n    // This logic ensures we accurately count the number of packed texels needed\n    // to accommodate the tensor. We can only pack values in the same texel if\n    // they are from adjacent pairs of rows/cols within the same batch. So if a\n    // tensor has 3 rows, we pretend it has 4 rows in order to account for the\n    // fact that the texels containing the third row are half empty.\n    logShape = logShape.map(\n        (d, i) => i >= logShape.length - 2 ?\n            util.nearestLargerEven(logShape[i]) :\n            logShape[i]);\n\n    // Packed texture height is at least 2 (the channel height of a single\n    // texel).\n    if (logShape.length === 1) {\n      logShape = [2, logShape[0]];\n    }\n  }\n\n  // If logical shape is 2, we don't squeeze, since we want to match physical.\n  if (logShape.length !== 2) {\n    const squeezeResult = util.squeezeShape(logShape);\n    logShape = squeezeResult.newShape;\n  }\n\n  let size = util.sizeFromShape(logShape);\n  if (logShape.length <= 1 && size <= maxTexSize) {\n    return [1, size];\n  } else if (\n      logShape.length === 2 && logShape[0] <= maxTexSize &&\n      logShape[1] <= maxTexSize) {\n    return logShape as [number, number];\n  } else if (\n      logShape.length === 3 && logShape[0] * logShape[1] <= maxTexSize &&\n      logShape[2] <= maxTexSize) {\n    return [logShape[0] * logShape[1], logShape[2]];\n  } else if (\n      logShape.length === 3 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2]];\n  } else if (\n      logShape.length === 4 &&\n      logShape[0] * logShape[1] * logShape[2] <= maxTexSize &&\n      logShape[3] <= maxTexSize) {\n    return [logShape[0] * logShape[1] * logShape[2], logShape[3]];\n  } else if (\n      logShape.length === 4 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] * logShape[3] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2] * logShape[3]];\n  } else {\n    if (isPacked) {\n      // For packed textures size equals the number of channels required to\n      // accommodate the texture data. However in order to squarify such that\n      // inner dimensions stay even, we rewrite size to equal the number of\n      // texels. Then in the return statement we rehydrate the squarified\n      // dimensions to channel units.\n\n      const batchDim = getBatchDim(logShape);\n      let rows = 2, cols = 2;\n      if (logShape.length) {\n        [rows, cols] = getRowsCols(logShape);\n      }\n      size = batchDim * (rows / 2) * (cols / 2);\n      return util.sizeToSquarishShape(size).map(d => d * 2) as [number, number];\n    }\n    return util.sizeToSquarishShape(size);\n  }\n}\n\nfunction isEven(n: number): boolean {\n  return n % 2 === 0;\n}\n\n/**\n * This determines whether reshaping a packed texture requires rearranging\n * the data within the texture, assuming 2x2 packing.\n */\nexport function isReshapeFree(shape1: number[], shape2: number[]): boolean {\n  shape1 = shape1.slice(-2);\n  shape2 = shape2.slice(-2);\n\n  if (util.arraysEqual(shape1, shape2)) {\n    return true;\n  }\n\n  if (!shape1.length || !shape2.length) {  // One of the shapes is a scalar.\n    return true;\n  }\n\n  if (shape1[0] === 0 || shape1[1] === 0 || shape2[0] === 0 ||\n      shape2[1] === 0) {\n    return true;\n  }\n\n  if (shape1.length !== shape2.length) {  // One of the shapes is a vector.\n    const shape1Cols = shape1.slice(-1)[0];\n    const shape2Cols = shape2.slice(-1)[0];\n    if (shape1Cols === shape2Cols) {\n      return true;\n    }\n\n    if (isEven(shape1Cols) && isEven(shape2Cols) &&\n        (shape1[0] === 1 || shape2[0] === 1)) {\n      return true;\n    }\n  }\n  return shape1[1] === shape2[1] && isEven(shape1[0]) && isEven(shape2[0]);\n}\n\n// We cache webgl params because the environment gets reset between\n// unit tests and we don't want to constantly query the WebGLContext for\n// MAX_TEXTURE_SIZE.\nlet MAX_TEXTURE_SIZE: number;\nlet MAX_TEXTURES_IN_SHADER: number;\n\nexport function getWebGLMaxTextureSize(webGLVersion: number): number {\n  if (MAX_TEXTURE_SIZE == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURE_SIZE = gl.getParameter(gl.MAX_TEXTURE_SIZE);\n  }\n  return MAX_TEXTURE_SIZE;\n}\n\nexport function resetMaxTextureSize() {\n  MAX_TEXTURE_SIZE = null;\n}\nexport function resetMaxTexturesInShader() {\n  MAX_TEXTURES_IN_SHADER = null;\n}\n\nexport function getMaxTexturesInShader(webGLVersion: number): number {\n  if (MAX_TEXTURES_IN_SHADER == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURES_IN_SHADER = gl.getParameter(gl.MAX_TEXTURE_IMAGE_UNITS);\n  }\n  // We cap at 16 to avoid spurious runtime \"memory exhausted\" error.\n  return Math.min(16, MAX_TEXTURES_IN_SHADER);\n}\n\nexport function getWebGLDisjointQueryTimerVersion(webGLVersion: number):\n    number {\n  if (webGLVersion === 0) {\n    return 0;\n  }\n\n  let queryTimerVersion: number;\n  const gl = getWebGLContext(webGLVersion);\n\n  if (hasExtension(gl, 'EXT_disjoint_timer_query_webgl2') &&\n      webGLVersion === 2) {\n    queryTimerVersion = 2;\n  } else if (hasExtension(gl, 'EXT_disjoint_timer_query')) {\n    queryTimerVersion = 1;\n  } else {\n    queryTimerVersion = 0;\n  }\n  return queryTimerVersion;\n}\n\nexport function hasExtension(gl: WebGLRenderingContext, extensionName: string) {\n  const ext = gl.getExtension(extensionName);\n  return ext != null;\n}\n\nexport function isWebGLVersionEnabled(webGLVersion: 1|2) {\n  try {\n    const gl = getWebGLContext(webGLVersion);\n    if (gl != null) {\n      return true;\n    }\n  } catch (e) {\n    console.log('Error when getting WebGL context: ', e);\n    return false;\n  }\n  return false;\n}\n\nexport function isCapableOfRenderingToFloatTexture(webGLVersion: number):\n    boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n  } else {\n    if (!hasExtension(gl, 'EXT_color_buffer_float')) {\n      return false;\n    }\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\n/**\n * Check if we can download values from a float/half-float texture.\n *\n * Note that for performance reasons we use binding a texture to a framebuffer\n * as a proxy for ability to download float values later using readPixels. The\n * texture params of this texture will not match those in readPixels exactly\n * but if we are unable to bind some kind of float texture to the frameBuffer\n * then we definitely will not be able to read float values from it.\n */\nexport function isDownloadFloatTextureEnabled(webGLVersion: number): boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n    if (!hasExtension(gl, 'WEBGL_color_buffer_float')) {\n      return false;\n    }\n  } else {\n    if (hasExtension(gl, 'EXT_color_buffer_float')) {\n      return createFloatTextureAndBindToFramebuffer(gl);\n    }\n\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (hasExtension(gl, COLOR_BUFFER_HALF_FLOAT)) {\n      const textureHalfFloatExtension =\n          gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      return createHalfFloatTextureAndBindToFramebuffer(\n          gl, textureHalfFloatExtension);\n    }\n\n    return false;\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\nfunction createFloatTextureAndBindToFramebuffer(gl: WebGLRenderingContext):\n    boolean {\n  const texConfig = getTextureConfig(gl);\n\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nfunction createHalfFloatTextureAndBindToFramebuffer(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension: any): boolean {\n  const texConfig = getTextureConfig(gl, textureHalfFloatExtension);\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatHalfFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeHalfFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nexport function isWebGLFenceEnabled(webGLVersion: number) {\n  if (webGLVersion !== 2) {\n    return false;\n  }\n  const gl = getWebGLContext(webGLVersion);\n\n  // tslint:disable-next-line:no-any\n  const isEnabled = (gl as any).fenceSync != null;\n  return isEnabled;\n}\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${opName} does not support complex64 tensors ` +\n              'in the WebGL backend.');\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {device_util, env} from '@tensorflow/tfjs-core';\n\nimport {getMaxTexturesInShader, getWebGLDisjointQueryTimerVersion, getWebGLMaxTextureSize, isCapableOfRenderingToFloatTexture, isDownloadFloatTextureEnabled, isWebGLFenceEnabled, isWebGLVersionEnabled} from './webgl_util';\n\nconst ENV = env();\n\n/**\n * This file contains WebGL-specific flag registrations.\n */\n\n/**\n * True if WebGL is supported.\n */\nENV.registerFlag('HAS_WEBGL', () => ENV.getNumber('WEBGL_VERSION') > 0);\n\n/** 0: No WebGL, 1: WebGL 1.0, 2: WebGL 2.0. */\nENV.registerFlag('WEBGL_VERSION', () => {\n  if (isWebGLVersionEnabled(2)) {\n    return 2;\n  } else if (isWebGLVersionEnabled(1)) {\n    return 1;\n  }\n  return 0;\n});\n\n/** Whether to check for numerical representation problems. */\nENV.registerFlag('WEBGL_CHECK_NUMERICAL_PROBLEMS', () => false);\n\nENV.registerFlag(\n    'WEBGL_BUFFER_SUPPORTED', () => ENV.get('WEBGL_VERSION') === 2);\n\n/** Whether the WebGL backend will sometimes forward ops to the CPU. */\nENV.registerFlag('WEBGL_CPU_FORWARD', () => true);\n\n/** Whether the WebGL backend will always use f16 textures for rendering. */\nENV.registerFlag('WEBGL_FORCE_F16_TEXTURES', () => false);\n\n/** Whether to turn all packing related flags on. */\nENV.registerFlag('WEBGL_PACK', () => ENV.getBool('HAS_WEBGL'));\n\n/** Whether we will pack the batchnormalization op. */\nENV.registerFlag('WEBGL_PACK_NORMALIZATION', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the clip op. */\nENV.registerFlag('WEBGL_PACK_CLIP', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the depthwise conv op. */\n// TODO: https://github.com/tensorflow/tfjs/issues/1679\nENV.registerFlag('WEBGL_PACK_DEPTHWISECONV', () => false);\n\n/** Whether we will pack binary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_BINARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack unary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_UNARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack array ops. */\nENV.registerFlag(\n    'WEBGL_PACK_ARRAY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack image ops. */\nENV.registerFlag(\n    'WEBGL_PACK_IMAGE_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack reduce ops. */\nENV.registerFlag('WEBGL_PACK_REDUCE', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether packed WebGL kernels lazily unpack their outputs. */\nENV.registerFlag('WEBGL_LAZILY_UNPACK', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will use the im2col algorithm to speed up convolutions. */\nENV.registerFlag('WEBGL_CONV_IM2COL', () => ENV.getBool('WEBGL_PACK'));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURE_SIZE',\n    () => getWebGLMaxTextureSize(ENV.getNumber('WEBGL_VERSION')));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURES_IN_SHADER',\n    () => getMaxTexturesInShader(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * The disjoint_query_timer extension version.\n * 0: disabled, 1: EXT_disjoint_timer_query, 2:\n * EXT_disjoint_timer_query_webgl2.\n * In Firefox with WebGL 2.0,\n * EXT_disjoint_timer_query_webgl2 is not available, so we must use the\n * WebGL 1.0 extension.\n */\nENV.registerFlag('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION', () => {\n  const webGLVersion = ENV.getNumber('WEBGL_VERSION');\n\n  if (webGLVersion === 0) {\n    return 0;\n  }\n  return getWebGLDisjointQueryTimerVersion(webGLVersion);\n});\n\n/**\n * Whether the timer object from the disjoint_query_timer extension gives\n * timing information that is reliable.\n */\nENV.registerFlag(\n    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE',\n    () => ENV.getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0 &&\n        !device_util.isMobile());\n\n/**\n * Whether the device is physically capable of rendering to float32 textures.\n */\nENV.registerFlag(\n    'WEBGL_RENDER_FLOAT32_CAPABLE',\n    () => isCapableOfRenderingToFloatTexture(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Whether rendering to float32 textures is enabled. If disabled, renders to\n * float16 textures.\n */\nENV.registerFlag('WEBGL_RENDER_FLOAT32_ENABLED', () => {\n  return ENV.getBool('WEBGL_FORCE_F16_TEXTURES') ?\n      false :\n      ENV.getBool('WEBGL_RENDER_FLOAT32_CAPABLE');\n});\n\n/**\n * Whether downloading float textures is enabled (16 or 32 bit). If disabled,\n * uses IEEE 754 encoding of the float32 values to 4 uint8 when downloading.\n */\nENV.registerFlag(\n    'WEBGL_DOWNLOAD_FLOAT_ENABLED',\n    () => isDownloadFloatTextureEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/** Whether the fence API is available. */\nENV.registerFlag(\n    'WEBGL_FENCE_API_ENABLED',\n    () => isWebGLFenceEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Tensors with size <= than this will be uploaded as uniforms, not textures.\n */\nENV.registerFlag('WEBGL_SIZE_UPLOAD_UNIFORM', () => {\n  // Use uniform uploads only when 32bit floats are supported. In\n  // 16bit\n  // environments there are problems with comparing a 16bit texture value\n  // with a 32bit uniform value.\n  const useUniforms = ENV.getBool('WEBGL_RENDER_FLOAT32_ENABLED');\n  return useUniforms ? 4 : 0;\n});\n\n/**\n * If the total number of bytes allocated on the GPU is greater than this\n * number, we will aggressively delete textures upon disposal with\n * gl.deleteMatrixTexture, rather than making them available for reuse.\n *\n * Default value -1 indicates that we will never aggressively delete textures.\n */\nENV.registerFlag(\n    'WEBGL_DELETE_TEXTURE_THRESHOLD',\n    () => {\n      return -1;\n    },\n    threshold => {\n      if (threshold < 0 && threshold !== -1) {\n        throw new Error(\n            `WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never ` +\n            `delete) or at least 0, but got ${threshold}.`);\n      }\n    });\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Abs, util } from '@tensorflow/tfjs-core';\nexport function simpleAbsImpl(vals) {\n    const resultValues = new Float32Array(vals.length);\n    for (let i = 0; i < vals.length; ++i) {\n        resultValues[i] = Math.abs(vals[i]);\n    }\n    return resultValues;\n}\nexport const absKernelFunc = (args) => {\n    const { x } = args.inputs;\n    const cpuBackend = args.backend;\n    let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n    if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values;\n        resultValues = simpleAbsImpl(values);\n    }\n    else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals = cpuBackend.data.get(real.dataId).values;\n        const imagVals = cpuBackend.data.get(imag.dataId).values;\n        for (let i = 0; i < realVals.length; i++) {\n            const real = realVals[i];\n            const imag = imagVals[i];\n            resultValues[i] = Math.hypot(real, imag);\n        }\n    }\n    return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n};\nexport const absConfig = {\n    kernelName: Abs,\n    backendName: 'cpu',\n    kernelFunc: absKernelFunc,\n};\n//# sourceMappingURL=Abs.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { backend_util, util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op) {\n    return (aShape, bShape, aVals, bVals, dtype) => {\n        const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n        const resultRank = newShape.length;\n        const resultStrides = util.computeStrides(newShape);\n        const resultSize = util.sizeFromShape(newShape);\n        const result = util.getTypedArrayFromDType(dtype, resultSize);\n        const aRank = aShape.length;\n        const bRank = bShape.length;\n        const aStrides = util.computeStrides(aShape);\n        const bStrides = util.computeStrides(bShape);\n        const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n        const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n        if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n            for (let i = 0; i < result.length; ++i) {\n                result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n            }\n        }\n        else {\n            for (let i = 0; i < result.length; ++i) {\n                const loc = util.indexToLoc(i, resultRank, resultStrides);\n                const aLoc = loc.slice(-aRank);\n                aBroadcastDims.forEach(d => aLoc[d] = 0);\n                const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n                const bLoc = loc.slice(-bRank);\n                bBroadcastDims.forEach(d => bLoc[d] = 0);\n                const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n                result[i] = op(aVals[aIndex], bVals[bIndex]);\n            }\n        }\n        return [result, newShape];\n    };\n}\n//# sourceMappingURL=binary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Add } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal + bReal, imag: aImag + bImag };\n}));\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\nexport const addConfig = {\n    kernelName: Add,\n    backendName: 'cpu',\n    kernelFunc: add\n};\n//# sourceMappingURL=Add.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op) {\n    return (values, dtype, attrs) => {\n        const newValues = util.getTypedArrayFromDType(dtype, values.length);\n        for (let i = 0; i < values.length; ++i) {\n            newValues[i] = op(values[i], attrs);\n        }\n        return newValues;\n    };\n}\n//# sourceMappingURL=unary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Ceil } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\nexport const ceilConfig = {\n    kernelName: Ceil,\n    backendName: 'cpu',\n    kernelFunc: ceilKernelFunc,\n};\n//# sourceMappingURL=Ceil.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Exp } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\nexport const expConfig = {\n    kernelName: Exp,\n    backendName: 'cpu',\n    kernelFunc: expKernelFunc,\n};\n//# sourceMappingURL=Exp.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Expm1 } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\nexport const expm1Config = {\n    kernelName: Expm1,\n    backendName: 'cpu',\n    kernelFunc: expm1KernelFunc,\n};\n//# sourceMappingURL=Expm1.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Floor } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\nexport const floorConfig = {\n    kernelName: Floor,\n    backendName: 'cpu',\n    kernelFunc: floorKernelFunc,\n};\n//# sourceMappingURL=Floor.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Log } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\nexport const logConfig = {\n    kernelName: Log,\n    backendName: 'cpu',\n    kernelFunc: logKernelFunc,\n};\n//# sourceMappingURL=Log.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function maxImpl(aVals, reduceSize, outShape, dtype) {\n    const vals = util.getTypedArrayFromDType(dtype, util.sizeFromShape(outShape));\n    for (let i = 0; i < vals.length; ++i) {\n        const offset = i * reduceSize;\n        let max = aVals[offset];\n        for (let j = 0; j < reduceSize; ++j) {\n            const value = aVals[offset + j];\n            if (value > max) {\n                max = value;\n            }\n        }\n        vals[i] = max;\n    }\n    return vals;\n}\n//# sourceMappingURL=Max_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Multiply } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const multiplyImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n    };\n}));\nexport const multiply = binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\nexport const multiplyConfig = {\n    kernelName: Multiply,\n    backendName: 'cpu',\n    kernelFunc: multiply\n};\n//# sourceMappingURL=Multiply.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Rsqrt } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\nexport const rsqrtConfig = {\n    kernelName: Rsqrt,\n    backendName: 'cpu',\n    kernelFunc: rsqrtKernelFunc,\n};\n//# sourceMappingURL=Rsqrt.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Slice, slice_util, util } from '@tensorflow/tfjs-core';\nimport { assertNotComplex } from '../cpu_util';\nexport function sliceImpl(vals, begin, size, shape, dtype) {\n    const isContinous = slice_util.isSliceContinous(shape, begin, size);\n    const length = util.sizeFromShape(size);\n    const xStrides = util.computeStrides(shape);\n    if (isContinous) {\n        const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n        return vals.subarray(flatOffset, flatOffset + length);\n    }\n    const outVals = util.getTypedArrayFromDType(dtype, length);\n    for (let i = 0; i < length; ++i) {\n        const rank = size.length;\n        const strides = util.computeStrides(size);\n        const loc = util.indexToLoc(i, rank, strides);\n        const xLoc = loc.map((idx, j) => idx + begin[j]);\n        const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n        outVals[i] = vals[xIndex];\n    }\n    return outVals;\n}\nexport function slice(args) {\n    const { inputs, backend, attrs } = args;\n    const { x } = inputs;\n    const { begin, size } = attrs;\n    assertNotComplex(x, 'slice');\n    const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n    slice_util.assertParamsValid(x, $begin, $size);\n    const vals = backend.data.get(x.dataId).values;\n    const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n    return backend.makeTensorInfo($size, x.dtype, outVals);\n}\nexport const sliceConfig = {\n    kernelName: Slice,\n    backendName: 'cpu',\n    kernelFunc: slice\n};\n//# sourceMappingURL=Slice.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Sub } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const subImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal - bReal, imag: aImag - bImag };\n}));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\nexport const subConfig = {\n    kernelName: Sub,\n    backendName: 'cpu',\n    kernelFunc: sub\n};\n//# sourceMappingURL=Sub.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function transposeImpl(xVals, xShape, dtype, perm, newShape) {\n    const xRank = xShape.length;\n    const xSize = util.sizeFromShape(xShape);\n    const xStrides = util.computeStrides(xShape);\n    const newStrides = util.computeStrides(newShape);\n    const result = util.getTypedArrayFromDType(dtype, util.sizeFromShape(newShape));\n    for (let i = 0; i < xSize; ++i) {\n        const loc = util.indexToLoc(i, xRank, xStrides);\n        // Permute location.\n        const newLoc = new Array(loc.length);\n        for (let i = 0; i < newLoc.length; i++) {\n            newLoc[i] = loc[perm[i]];\n        }\n        const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n        result[newIndex] = xVals[i];\n    }\n    return result;\n}\n//# sourceMappingURL=Transpose_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { TensorBuffer, util } from '@tensorflow/tfjs-core';\nexport function uniqueImpl(values, axis, shape, dtype) {\n    // Normalize and validate axis.\n    const $axis = util.parseAxisParam(axis, shape)[0];\n    // Calculate the new shape that is suitable for extracting data along the\n    // given axis.\n    //\n    // The rank is 3.\n    // The size of the 1st dimension is the size of all the axes < the given axis.\n    // The size of the 2nd dimension is the same as the size of the given axis.\n    // The size of the 3rd dimension is the size of all the axes > the given axis.\n    //\n    // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n    // newShape would be: [2*3, 5, 4].\n    //\n    // Note that this is not the final output shape. This will be the shape for an\n    // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n    // values along the given axis. To demonstrate how it works, consider the\n    // following example:\n    //\n    // Input: a 3D tensor, with shape [1, 2, 3]\n    // [\n    //   [\n    //      [1,2,3],\n    //      [4,5,6]\n    //   ]\n    // ]\n    // Axis: 2 (the last axis).\n    // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n    //\n    // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n    // 1*2. The re-shaped data would look like:\n    //\n    // [\n    //   [\n    //     [1], [2], [3]\n    //   ],\n    //   [\n    //     [4], [5], [6]\n    //   ]\n    // ]\n    //\n    // Then, we can construct a 3-level nested loop by the following dimension\n    // order to extract the values along the axis (dimension1):\n    // i: dimension1       // 0,1,2 (newShape[1])\n    //   m: dimension0     // 0,1   (newShape[0])\n    //     n: dimension2   // 0     (newShape[2])\n    //\n    //                       m, i, n\n    //                      ---------\n    // Iteration 0: data at [0, 0, 0] => \"1\"\n    // Iteration 1: data at [1, 0, 0] => \"4\"\n    // We got [1,4].\n    // Iteration 2: data at [0, 1, 0] => \"2\"\n    // Iteration 3: data at [1, 1, 0] => \"5\"\n    // We got [2,5].\n    // Iteration 4: data at [0, 2, 0] => \"3\"\n    // Iteration 5: data at [1, 2, 0] => \"6\"\n    // We got [3,6].\n    const newShape = [1, shape[0], 1];\n    for (let i = 0; i < $axis; i++) {\n        newShape[0] *= shape[i];\n    }\n    newShape[1] = shape[$axis];\n    for (let i = $axis + 1; i < shape.length; i++) {\n        newShape[2] *= shape[i];\n    }\n    // A map from unique elements (their string representations) to their values\n    // in \"indices\" (below).\n    const uniqueElements = {};\n    // The indices of each unique element in the original tensor along the given\n    // axis. It is 1D and has the same size as the given axis.\n    const indices = new Int32Array(shape[$axis]);\n    // Create a buffer so we can easily extract value at a given location.\n    const inputBuffer = new TensorBuffer(newShape, dtype, values);\n    // The indices along the given axis that have unique elements. This is a\n    // de-duped version of \"indices\" above.\n    const uniqueIndices = [];\n    const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n    for (let i = 0; i < shape[$axis]; i++) {\n        // Extract values along the axis.\n        let element;\n        if (is1DTensor) {\n            // Fast path for 1D tensor input.\n            element = values[i].toString();\n        }\n        else {\n            const axisValues = [];\n            for (let m = 0; m < newShape[0]; m++) {\n                for (let n = 0; n < newShape[2]; n++) {\n                    axisValues.push(inputBuffer.get(m, i, n));\n                }\n            }\n            element = axisValues.join(',');\n        }\n        // Dedup and update various indices.\n        if (uniqueElements[element] !== undefined) {\n            indices[i] = uniqueElements[element];\n        }\n        else {\n            const uniqueIndex = Object.keys(uniqueElements).length;\n            uniqueElements[element] = uniqueIndex;\n            indices[i] = uniqueIndex;\n            uniqueIndices.push(i);\n        }\n    }\n    // Now we know where each of the unique elements are located along the axis\n    // (uniqueIndices). Extract them from input buffer and store them in the\n    // output buffer.\n    const outputTmpShape = newShape.slice();\n    outputTmpShape[1] = Object.keys(uniqueElements).length;\n    const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n    uniqueIndices.forEach((uniqueElementIndex, i) => {\n        for (let m = 0; m < newShape[0]; m++) {\n            for (let n = 0; n < newShape[2]; n++) {\n                outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n            }\n        }\n    });\n    // The output shape can be calculated from the input shape with the size of\n    // the given axis replaced by the number of unique elements along that axis.\n    const outputShape = shape.slice();\n    outputShape[$axis] = outputTmpShape[1];\n    return {\n        outputValues: outputBuffer.values,\n        outputShape,\n        indices,\n    };\n}\n//# sourceMappingURL=Unique_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import shared functionality from tfjs-backend-cpu without triggering\n// side effects.\n// tslint:disable-next-line: no-imports-from-dist\nimport * as shared from '@tensorflow/tfjs-backend-cpu/dist/shared';\n\nconst {\n  simpleAbsImpl: simpleAbsImplCPU,\n  addImpl: addImplCPU,\n  ceilImpl: ceilImplCPU,\n  expImpl: expImplCPU,\n  expm1Impl: expm1ImplCPU,\n  floorImpl: floorImplCPU,\n  logImpl: logImplCPU,\n  maxImpl: maxImplCPU,\n  multiplyImpl: multiplyImplCPU,\n  rsqrtImpl: rsqrtImplCPU,\n  sliceImpl: sliceImplCPU,\n  subImpl: subImplCPU,\n  transposeImpl: transposeImplCPU,\n  uniqueImpl: uniqueImplCPU,\n} = shared;\n\nexport {\n  simpleAbsImplCPU,\n  addImplCPU,\n  ceilImplCPU,\n  expImplCPU,\n  expm1ImplCPU,\n  logImplCPU,\n  multiplyImplCPU,\n  sliceImplCPU,\n  subImplCPU,\n  floorImplCPU,\n  maxImplCPU,\n  rsqrtImplCPU,\n  transposeImplCPU,\n  uniqueImplCPU,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`float v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        float result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`vec4 v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        vec4 result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ArgMinMaxProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo, op: 'max'|'min',\n      firstPass: boolean) {\n    const {windowSize, batchSize, outSize} = reduceInfo;\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    this.outputShape = [batchSize, outSize];\n    const compOp = (op === 'max') ? '>' : '<';\n    const indexSnippet = firstPass ?\n        'inOffset + i;' :\n        'round(getBestIndicesA(batch, inOffset + i));';\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        int bestIndex = inOffset;\n        float bestValue = getA(batch, bestIndex);\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          int inIdx = ${indexSnippet};\n          float candidate = getA(batch, inIdx);\n          if (candidate ${compOp} bestValue) {\n            bestValue = candidate;\n            bestIndex = inIdx;\n          }\n        }\n        setOutput(float(bestIndex));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport function getVecChannels(name: string, rank: number): string[] {\n  return ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank).map(d => `${name}.${d}`);\n}\n\nexport function getChannels(name: string, rank: number): string[] {\n  if (rank === 1) {\n    return [name];\n  }\n  return getVecChannels(name, rank);\n}\n\nexport function getSourceCoords(rank: number, dims: string[]): string {\n  if (rank === 1) {\n    return 'rc';\n  }\n\n  let coords = '';\n  for (let i = 0; i < rank; i++) {\n    coords += dims[i];\n    if (i < rank - 1) {\n      coords += ',';\n    }\n  }\n  return coords;\n}", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '@tensorflow/tfjs-core';\n\nexport type GLSL = {\n  version: string,\n  attribute: string,\n  varyingVs: string,\n  varyingFs: string,\n  texture2D: string,\n  output: string,\n  defineOutput: string,\n  defineSpecialNaN: string,\n  defineSpecialInf: string,\n  defineRound: string\n};\n\nexport function getGlslDifferences(): GLSL {\n  let version: string;\n  let attribute: string;\n  let varyingVs: string;\n  let varyingFs: string;\n  let texture2D: string;\n  let output: string;\n  let defineOutput: string;\n  let defineSpecialNaN: string;\n  let defineSpecialInf: string;\n  let defineRound: string;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    version = '#version 300 es';\n    attribute = 'in';\n    varyingVs = 'out';\n    varyingFs = 'in';\n    texture2D = 'texture';\n    output = 'outputColor';\n    defineOutput = 'out vec4 outputColor;';\n\n    // Use custom isnan definition to work across differences between\n    // implementations on various platforms. While this should happen in ANGLE\n    // we still see differences between android and windows (on chrome) when\n    // using isnan directly.\n    defineSpecialNaN = `\n      bool isnan_custom(float val) {\n        return (val > 0.0 || val < 0.0) ? false : val != 0.0;\n      }\n\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan_custom(val.x),\n          isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));\n      }\n\n      #define isnan(value) isnan_custom(value)\n    `;\n    // In webgl 2 we do not need to specify a custom isinf so there is no\n    // need for a special INFINITY constant.\n    defineSpecialInf = ``;\n    defineRound = `\n      #define round(value) newRound(value)\n      int newRound(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 newRound(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  } else {\n    version = '';\n    attribute = 'attribute';\n    varyingVs = 'varying';\n    varyingFs = 'varying';\n    texture2D = 'texture2D';\n    output = 'gl_FragColor';\n    defineOutput = '';\n    // WebGL1 has no built in isnan so we define one here.\n    defineSpecialNaN = `\n      #define isnan(value) isnan_custom(value)\n      bool isnan_custom(float val) {\n        return (val > 0. || val < 1. || val == 0.) ? false : true;\n      }\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));\n      }\n    `;\n    defineSpecialInf = `\n      uniform float INFINITY;\n\n      bool isinf(float val) {\n        return abs(val) == INFINITY;\n      }\n      bvec4 isinf(vec4 val) {\n        return equal(abs(val), vec4(INFINITY));\n      }\n    `;\n    defineRound = `\n      int round(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 round(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  }\n\n  return {\n    version,\n    attribute,\n    varyingVs,\n    varyingFs,\n    texture2D,\n    output,\n    defineOutput,\n    defineSpecialNaN,\n    defineSpecialInf,\n    defineRound\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\n/**\n * Produces GLSL code that derives logical coordinates from a flat\n * index. The code performs integer division with each stride and decrements\n * the index until the index equals the final dimension coordinate.\n */\nexport function getLogicalCoordinatesFromFlatIndex(\n    coords: string[], shape: number[], index = 'index'): string {\n  const strides = util.computeStrides(shape);\n  return strides\n      .map((stride, i) => {\n        const line1 = `int ${coords[i]} = ${index} / ${stride}`;\n        const line2 = i === strides.length - 1 ?\n            `int ${coords[i + 1]} = ${index} - ${coords[i]} * ${stride}` :\n            `index -= ${coords[i]} * ${stride}`;\n        return `${line1}; ${line2};`;\n      })\n      .join('');\n}\n\nfunction buildVec(x: string[]): string {\n  if (x.length === 1) {\n    return `${x[0]}`;\n  }\n  return `vec${x.length}(${x.join(',')})`;\n}\n\n/**\n * Produces GLSL code that computes the dot product of the input x and y\n * vectors. Handles splitting inputs into increments of vec4s when necessary.\n */\nexport function dotify(x: string[], y: string[]): string {\n  if (x.length !== y.length) {\n    throw new Error(\n        `Vectors to be dotted must be of the same length -` +\n        `got ${x.length} and ${y.length}`);\n  }\n\n  const slices: string[] = [];\n  const nearestVec4 = Math.floor(x.length / 4);\n  const nearestVec4Remainder = x.length % 4;\n\n  for (let i = 0; i < nearestVec4; i++) {\n    const xSlice = x.slice(i * 4, i * 4 + 4);\n    const ySlice = y.slice(i * 4, i * 4 + 4);\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  if (nearestVec4Remainder !== 0) {\n    let xSlice = x.slice(nearestVec4 * 4);\n    let ySlice = y.slice(nearestVec4 * 4);\n    if (xSlice.length === 1) {\n      xSlice = xSlice.map(d => `float(${d})`);\n      ySlice = ySlice.map(d => `float(${d})`);\n    }\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  return slices.map((d, i) => `dot(${d})`).join('+');\n}\n\n/**\n * Produces GLSL that computes the flat index from 3D coordinates.\n */\nexport function getFlatIndexFrom3D(shape: [number, number, number]): string {\n  const strides = util.computeStrides(shape).map(d => d.toString());\n\n  return `\n  int getFlatIndex(ivec3 coords) {\n    return coords.x * ${strides[0]} + coords.y * ${strides[1]} + coords.z;\n  }\n`;\n}\n\nexport const ENCODE_FLOAT_SNIPPET = `\n  const float FLOAT_MAX = 1.70141184e38;\n  const float FLOAT_MIN = 1.17549435e-38;\n\n  lowp vec4 encode_float(highp float v) {\n    if (isnan(v)) {\n      return vec4(255, 255, 255, 255);\n    }\n\n    highp float av = abs(v);\n\n    if(av < FLOAT_MIN) {\n      return vec4(0.0, 0.0, 0.0, 0.0);\n    } else if(v > FLOAT_MAX) {\n      return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;\n    } else if(v < -FLOAT_MAX) {\n      return vec4(0.0, 0.0,  128.0, 255.0) / 255.0;\n    }\n\n    highp vec4 c = vec4(0,0,0,0);\n\n    highp float e = floor(log2(av));\n    highp float m = exp2(fract(log2(av))) - 1.0;\n\n    c[2] = floor(128.0 * m);\n    m -= c[2] / 128.0;\n    c[1] = floor(32768.0 * m);\n    m -= c[1] / 32768.0;\n    c[0] = floor(8388608.0 * m);\n\n    highp float ebias = e + 127.0;\n    c[3] = floor(ebias / 2.0);\n    ebias -= c[3] * 2.0;\n    c[2] += floor(ebias) * 128.0;\n\n    c[3] += 128.0 * step(0.0, -v);\n\n    return c / 255.0;\n  }\n`;\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\nconst {getBroadcastDims} = backend_util;\nimport {getGlslDifferences, GLSL} from './glsl_version';\nimport * as shader_util from './shader_compiler_util';\n\nexport type ShapeInfo = {\n  logicalShape: number[],\n  texShape: [number, number],\n  isUniform: boolean,\n  isPacked: boolean,\n  flatOffset: number\n};\n\nexport type InputInfo = {\n  name: string,\n  shapeInfo: ShapeInfo\n};\n\nexport function makeShader(\n    inputsInfo: InputInfo[], outputShape: ShapeInfo, userCode: string,\n    usesPackedTextures: boolean): string {\n  const prefixSnippets: string[] = [];\n  inputsInfo.forEach(x => {\n    const size = util.sizeFromShape(x.shapeInfo.logicalShape);\n\n    // Snippet when we decided to upload the values as uniform.\n    if (x.shapeInfo.isUniform) {\n      prefixSnippets.push(\n          `uniform float ${x.name}${size > 1 ? `[${size}]` : ''};`);\n    } else {\n      prefixSnippets.push(`uniform sampler2D ${x.name};`);\n      prefixSnippets.push(`uniform int offset${x.name};`);\n    }\n  });\n  const inputPrefixSnippet = prefixSnippets.join('\\n');\n\n  const inputSamplingSnippet =\n      inputsInfo\n          .map(x => getInputSamplingSnippet(x, outputShape, usesPackedTextures))\n          .join('\\n');\n  const outTexShape = outputShape.texShape;\n  const glsl = getGlslDifferences();\n  const floatTextureSampleSnippet = getFloatTextureSampleSnippet(glsl);\n  let outputSamplingSnippet: string;\n  let floatTextureSetOutputSnippet: string;\n  let shaderPrefix = getShaderPrefix(glsl);\n\n  if (outputShape.isPacked) {\n    outputSamplingSnippet =\n        getPackedOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRGBASnippet(glsl);\n  } else {\n    outputSamplingSnippet =\n        getOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRSnippet(glsl);\n  }\n\n  if (usesPackedTextures) {\n    shaderPrefix += SHADER_PACKED_PREFIX;\n  }\n\n  const source = [\n    shaderPrefix, floatTextureSampleSnippet, floatTextureSetOutputSnippet,\n    inputPrefixSnippet, outputSamplingSnippet, inputSamplingSnippet, userCode\n  ].join('\\n');\n  return source;\n}\n\nfunction getSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getSamplerScalar(inInfo);\n    case 1:\n      return getSampler1D(inInfo);\n    case 2:\n      return getSampler2D(inInfo);\n    case 3:\n      return getSampler3D(inInfo);\n    case 4:\n      return getSampler4D(inInfo);\n    case 5:\n      return getSampler5D(inInfo);\n    case 6:\n      return getSampler6D(inInfo);\n    default:\n      throw new Error(\n          `${shape.length}-D input sampling` +\n          ` is not yet supported`);\n  }\n}\n\nfunction getPackedSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getPackedSamplerScalar(inInfo);\n    case 1:\n      return getPackedSampler1D(inInfo);\n    case 2:\n      return getPackedSampler2D(inInfo);\n    case 3:\n      return getPackedSampler3D(inInfo);\n    default:\n      return getPackedSamplerND(inInfo);\n  }\n}\n\nfunction getInputSamplingSnippet(\n    inInfo: InputInfo, outShapeInfo: ShapeInfo,\n    usesPackedTextures = false): string {\n  let res = '';\n  if (usesPackedTextures) {\n    res += getPackedSamplerFromInInfo(inInfo);\n  } else {\n    res += getSamplerFromInInfo(inInfo);\n  }\n\n  const inShape = inInfo.shapeInfo.logicalShape;\n  const outShape = outShapeInfo.logicalShape;\n  if (inShape.length <= outShape.length) {\n    if (usesPackedTextures) {\n      res += getPackedSamplerAtOutputCoords(inInfo, outShapeInfo);\n    } else {\n      res += getSamplerAtOutputCoords(inInfo, outShapeInfo);\n    }\n  }\n  return res;\n}\n\nfunction getPackedOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutputPacked1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutputPacked2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutputPacked3DCoords(\n          outShape as [number, number, number], outTexShape);\n    default:\n      return getOutputPackedNDCoords(outShape, outTexShape);\n  }\n}\n\nfunction getOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutput1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutput2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutput3DCoords(\n          outShape as [number, number, number], outTexShape);\n    case 4:\n      return getOutput4DCoords(\n          outShape as [number, number, number, number], outTexShape);\n    case 5:\n      return getOutput5DCoords(\n          outShape as [number, number, number, number, number], outTexShape);\n    case 6:\n      return getOutput6DCoords(\n          outShape as [number, number, number, number, number, number],\n          outTexShape);\n    default:\n      throw new Error(\n          `${outShape.length}-D output sampling is not yet supported`);\n  }\n}\n\nfunction getFloatTextureSampleSnippet(glsl: GLSL): string {\n  return `\n    float sampleTexture(sampler2D textureSampler, vec2 uv) {\n      return ${glsl.texture2D}(textureSampler, uv).r;\n    }\n  `;\n}\n\nfunction getFloatTextureSetRSnippet(glsl: GLSL): string {\n  return `\n    void setOutput(float val) {\n      ${glsl.output} = vec4(val, 0, 0, 0);\n    }\n  `;\n}\n\nfunction getFloatTextureSetRGBASnippet(glsl: GLSL): string {\n  return `\n    void setOutput(vec4 val) {\n      ${glsl.output} = val;\n    }\n  `;\n}\n\nfunction getShaderPrefix(glsl: GLSL): string {\n  const SHADER_PREFIX = `${glsl.version}\n    precision highp float;\n    precision highp int;\n    precision highp sampler2D;\n    ${glsl.varyingFs} vec2 resultUV;\n    ${glsl.defineOutput}\n    const vec2 halfCR = vec2(0.5, 0.5);\n\n    struct ivec5\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n    };\n\n    struct ivec6\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n      int v;\n    };\n\n    uniform float NAN;\n    ${glsl.defineSpecialNaN}\n    ${glsl.defineSpecialInf}\n    ${glsl.defineRound}\n\n    int imod(int x, int y) {\n      return x - y * (x / y);\n    }\n\n    int idiv(int a, int b, float sign) {\n      int res = a / b;\n      int mod = imod(a, b);\n      if (sign < 0. && mod != 0) {\n        res -= 1;\n      }\n      return res;\n    }\n\n    //Based on the work of Dave Hoskins\n    //https://www.shadertoy.com/view/4djSRW\n    #define HASHSCALE1 443.8975\n    float random(float seed){\n      vec2 p = resultUV * seed;\n      vec3 p3  = fract(vec3(p.xyx) * HASHSCALE1);\n      p3 += dot(p3, p3.yzx + 19.19);\n      return fract((p3.x + p3.y) * p3.z);\n    }\n\n    ${SAMPLE_1D_SNIPPET}\n    ${SAMPLE_2D_SNIPPET}\n    ${SAMPLE_3D_SNIPPET}\n  `;\n\n  return SHADER_PREFIX;\n}\n\nconst SAMPLE_1D_SNIPPET = `\nvec2 uvFromFlat(int texNumR, int texNumC, int index) {\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\nvec2 packedUVfrom1D(int texNumR, int texNumC, int index) {\n  int texelIndex = index / 2;\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_2D_SNIPPET = `\nvec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,\n  int texNumC, int row, int col) {\n  int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_3D_SNIPPET = `\nvec2 packedUVfrom3D(int texNumR, int texNumC,\n    int texelsInBatch, int texelsInLogicalRow, int b,\n    int row, int col) {\n  int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SHADER_PACKED_PREFIX = `\n  float getChannel(vec4 frag, vec2 innerDims) {\n    vec2 modCoord = mod(innerDims, 2.);\n    return modCoord.x == 0. ?\n      (modCoord.y == 0. ? frag.r : frag.g) :\n      (modCoord.y == 0. ? frag.b : frag.a);\n  }\n  float getChannel(vec4 frag, int dim) {\n    float modCoord = mod(float(dim), 2.);\n    return modCoord == 0. ? frag.r : frag.g;\n  }\n`;\n\nfunction getOutputScalarCoords() {\n  return `\n    int getOutputCoords() {\n      return 0;\n    }\n  `;\n}\n\nfunction getOutputPacked1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (packedTexShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.x * ${packedTexShape[1]}.0);\n      }\n    `;\n  }\n\n  if (packedTexShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.y * ${packedTexShape[0]}.0);\n      }\n    `;\n  }\n\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      return 2 * (resTexRC.x * ${packedTexShape[1]} + resTexRC.y);\n    }\n  `;\n}\n\nfunction getOutput1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  if (texShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.x * ${texShape[1]}.0);\n      }\n    `;\n  }\n  if (texShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.y * ${texShape[0]}.0);\n      }\n    `;\n  }\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      return resTexRC.x * ${texShape[1]} + resTexRC.y;\n    }\n  `;\n}\n\nfunction getOutputPacked3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texelsInLogicalRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = texelsInLogicalRow * Math.ceil(shape[1] / 2);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec3(b, r, c);\n    }\n  `;\n}\n\nfunction getOutput3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n\nfunction getOutputPackedNDCoords(\n    shape: number[], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  const texelsInLogicalRow = Math.ceil(shape[shape.length - 1] / 2);\n  const texelsInBatch =\n      texelsInLogicalRow * Math.ceil(shape[shape.length - 2] / 2);\n  let texelsInBatchN = texelsInBatch;\n  let batches = ``;\n  let coords = 'b, r, c';\n\n  for (let b = 2; b < shape.length - 1; b++) {\n    texelsInBatchN *= shape[shape.length - b - 1];\n    batches = `\n      int b${b} = index / ${texelsInBatchN};\n      index -= b${b} * ${texelsInBatchN};\n    ` + batches;\n    coords = `b${b}, ` + coords;\n  }\n\n  return `\n    ivec${shape.length} getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      ${batches}\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec${shape.length}(${coords});\n    }\n  `;\n}\n\nfunction getOutput4DCoords(\n    shape: [number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2'], shape);\n\n  return `\n    ivec4 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec4(r, c, d, d2);\n    }\n  `;\n}\n\nfunction getOutput5DCoords(\n    shape: [number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3'], shape);\n\n  return `\n    ivec5 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx * vec2(${texShape[0]},\n                             ${texShape[1]}));\n\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec5 outShape = ivec5(r, c, d, d2, d3);\n      return outShape;\n    }\n  `;\n}\n\nfunction getOutput6DCoords(\n    shape: [number, number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3', 'd4'], shape);\n\n  return `\n    ivec6 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec6 result = ivec6(r, c, d, d2, d3, d4);\n      return result;\n    }\n  `;\n}\n\nfunction getOutputPacked2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return 2 * ivec2(resultUV.yx * vec2(${packedTexShape[0]}, ${\n        packedTexShape[1]}));\n      }\n    `;\n  }\n\n  // texels needed to accommodate a logical row\n  const texelsInLogicalRow = Math.ceil(shape[1] / 2);\n\n  /**\n   * getOutputCoords\n   *\n   * resTexRC: The rows and columns of the texels. If you move over one\n   * texel to the right in the packed texture, you are moving over one column\n   * (not two).\n   *\n   * index: The texel index\n   */\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getOutput2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return ivec2(resultUV.yx * vec2(${texShape[0]}, ${texShape[1]}));\n      }\n    `;\n  }\n  if (shape[1] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(index, 0);\n      }\n    `;\n  }\n  if (shape[0] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(0, index);\n      }\n    `;\n  }\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      int r = index / ${shape[1]};\n      int c = index - r * ${shape[1]};\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getFlatOffsetUniformName(texName: string): string {\n  return `offset${texName}`;\n}\n\nfunction getPackedSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}() {\n      return ${glsl.texture2D}(${texName}, halfCR);\n    }\n  `;\n}\n\nfunction getSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  if (inputInfo.shapeInfo.isUniform) {\n    return `float ${funcName}() {return ${texName};}`;\n  }\n  const [texNumR, texNumC] = inputInfo.shapeInfo.texShape;\n  if (texNumR === 1 && texNumC === 1) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n\n  const [tNumR, tNumC] = inputInfo.shapeInfo.texShape;\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}() {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int index) {\n      vec2 uv = packedUVfrom1D(\n        ${packedTexShape[0]}, ${packedTexShape[1]}, index);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int index) {\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const tNumR = texShape[0];\n  const tNumC = texShape[1];\n\n  if (tNumC === 1 && tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  if (tNumC === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2(0.5, (float(index + ${offset}) + 0.5) / ${tNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2((float(index + ${offset}) + 0.5) / ${tNumC}.0, 0.5);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  return `\n    float ${funcName}(int index) {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const glsl = getGlslDifferences();\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    return `\n      vec4 ${funcName}(int row, int col) {\n        vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n\n        return ${glsl.texture2D}(${texName}, uv);\n      }\n    `;\n  }\n\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const valuesPerRow = Math.ceil(shape[1] / 2);\n\n  return `\n    vec4 ${funcName}(int row, int col) {\n      vec2 uv = packedUVfrom2D(${valuesPerRow}, ${packedTexShape[0]}, ${\n      packedTexShape[1]}, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    const texNumR = texShape[0];\n    const texNumC = texShape[1];\n    return `\n    float ${funcName}(int row, int col) {\n      vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col) {\n        int index = round(dot(vec2(row, col), vec2(${shape[1]}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const offset = getFlatOffsetUniformName(texName);\n  if (texNumC === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2(0.5, (index + 0.5) / ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n  if (texNumR === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2((index + 0.5) / ${texNumC}.0, 0.5);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  return `\n  float ${funcName}(int row, int col) {\n    // Explicitly use integer operations as dot() only works on floats.\n    int index = row * ${shape[1]} + col + ${offset};\n    vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n    return sampleTexture(${texName}, uv);\n  }\n`;\n}\n\nfunction getPackedSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  if (shape[0] === 1) {\n    const squeezedShape = shape.slice(1);\n    const keptDims = [1, 2];\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['b', 'row', 'col'];\n    return `\n        ${getPackedSamplerFromInInfo(newInputInfo)}\n        vec4 ${funcName}(int b, int row, int col) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = valuesPerRow * Math.ceil(shape[1] / 2);\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int b, int row, int col) {\n      vec2 uv = packedUVfrom3D(\n        ${texNumR}, ${texNumC}, ${texelsInBatch}, ${valuesPerRow}, b, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride0 = shape[1] * shape[2];\n  const stride1 = shape[2];\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col', 'depth'];\n    return `\n        ${getSamplerFromInInfo(newInputInfo)}\n        float ${funcName}(int row, int col, int depth) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth) {\n        int index = round(dot(vec3(row, col, depth),\n                          vec3(${stride0}, ${stride1}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n        float ${funcName}(int row, int col, int depth) {\n          float texR = float(row);\n          float texC = dot(vec2(col, depth), vec2(${stride1}, 1));\n          vec2 uv = (vec2(texC, texR) + halfCR) /\n                     vec2(${texNumC}.0, ${texNumR}.0);\n          return sampleTexture(${texName}, uv);\n        }\n      `;\n  }\n\n  if (texNumC === stride1 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col, int depth) {\n      float texR = dot(vec2(row, col), vec2(${shape[1]}, 1));\n      float texC = float(depth);\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n      float ${funcName}(int row, int col, int depth) {\n        // Explicitly use integer operations as dot() only works on floats.\n        int index = row * ${stride0} + col * ${stride1} + depth + ${offset};\n        vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n        return sampleTexture(${texName}, uv);\n      }\n  `;\n}\n\nfunction getPackedSamplerND(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const rank = shape.length;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[rank - 1] / 2);\n  let texelsInBatch = valuesPerRow * Math.ceil(shape[rank - 2] / 2);\n  let params = `int b, int row, int col`;\n  let index = `b * ${texelsInBatch} + (row / 2) * ${valuesPerRow} + (col / 2)`;\n  for (let b = 2; b < rank - 1; b++) {\n    params = `int b${b}, ` + params;\n    texelsInBatch *= shape[rank - b - 1];\n    index = `b${b} * ${texelsInBatch} + ` + index;\n  }\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}(${params}) {\n      int index = ${index};\n      int texR = index / ${texNumC};\n      int texC = index - texR * ${texNumC};\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}, ${texNumR});\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler4D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride2 = shape[3];\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        int index = round(dot(vec4(row, col, depth, depth2),\n                          vec4(${stride0}, ${stride1}, ${stride2}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = float(row);\n        float texC =\n            dot(vec3(col, depth, depth2),\n                vec3(${stride1}, ${stride2}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride2 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = dot(vec3(row, col, depth),\n                         vec3(${shape[1] * shape[2]}, ${shape[2]}, 1));\n        float texC = float(depth2);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} +\n          depth * ${stride2} + depth2;\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler5D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride3 = shape[4];\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float index = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          depth3;\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n                         vec4(${stride1}, ${stride2}, ${stride3}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  if (texNumC === stride3 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float texR = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3]},\n               ${shape[2] * shape[3]}, ${shape[3]}, 1));\n        int texC = depth3;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler6D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3', 'depth4'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  const stride4 = shape[5];\n  const stride3 = shape[4] * stride4;\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n        int index = round(dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          dot(\n            vec2(depth3, depth4),\n            vec2(${stride4}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n          vec4(${stride1}, ${stride2}, ${stride3}, ${stride4})) +\n               float(depth4);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride4 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        float texR = dot(vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3] * shape[4]},\n               ${shape[2] * shape[3] * shape[4]},\n               ${shape[3] * shape[4]},\n               ${shape[4]})) + float(depth3);\n        int texC = depth4;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 * ${stride4} + depth4 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getUniformSampler(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n\n  if (inSize < 2) {\n    return `return ${texName};`;\n  }\n  return `\n    for (int i = 0; i < ${inSize}; i++) {\n      if (i == index) {\n        return ${texName}[i];\n      }\n    }\n  `;\n}\n\nfunction getPackedSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n\n  const type = getCoordsDataType(outRank);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  let output = `return outputValue;`;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n  const isInputScalar = inSize === 1;\n  const outSize = util.sizeFromShape(outShapeInfo.logicalShape);\n  const isOutputScalar = outSize === 1;\n\n  if (inRank === 1 && !isInputScalar && !isOutputScalar) {\n    output = `\n      return vec4(outputValue.xy, outputValue.xy);\n    `;\n  } else if (isInputScalar && !isOutputScalar) {\n    if (outRank === 1) {\n      output = `\n        return vec4(outputValue.x, outputValue.x, 0., 0.);\n      `;\n    } else {\n      output = `\n        return vec4(outputValue.x);\n      `;\n    }\n  } else if (broadcastDims.length) {\n    const rows = inRank - 2;\n    const cols = inRank - 1;\n\n    if (broadcastDims.indexOf(rows) > -1 && broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.x);`;\n    } else if (broadcastDims.indexOf(rows) > -1) {\n      output = `return vec4(outputValue.x, outputValue.y, ` +\n          `outputValue.x, outputValue.y);`;\n    } else if (broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.xx, outputValue.zz);`;\n    }\n  }\n\n  return `\n    vec4 ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      vec4 outputValue = get${texFuncSnippet}(${unpackedCoordsSnippet});\n      ${output}\n    }\n  `;\n}\n\nfunction getSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const outTexShape = outShapeInfo.texShape;\n  const inTexShape = inputInfo.shapeInfo.texShape;\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  if (!inputInfo.shapeInfo.isUniform && inRank === outRank &&\n      inputInfo.shapeInfo.flatOffset == null &&\n      util.arraysEqual(inTexShape, outTexShape)) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, resultUV);\n      }\n    `;\n  }\n\n  const type = getCoordsDataType(outRank);\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  return `\n    float ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      return get${texFuncSnippet}(${unpackedCoordsSnippet});\n    }\n  `;\n}\n\nexport function getCoordsDataType(rank: number): string {\n  if (rank <= 1) {\n    return 'int';\n  } else if (rank === 2) {\n    return 'ivec2';\n  } else if (rank === 3) {\n    return 'ivec3';\n  } else if (rank === 4) {\n    return 'ivec4';\n  } else if (rank === 5) {\n    return 'ivec5';\n  } else if (rank === 6) {\n    return 'ivec6';\n  } else {\n    throw Error(`GPU for rank ${rank} is not yet supported`);\n  }\n}\n\n/** Returns a new input info (a copy) that has a squeezed logical shape. */\nfunction squeezeInputInfo(\n    inInfo: InputInfo, squeezedShape: number[]): InputInfo {\n  // Deep copy.\n  const newInputInfo: InputInfo = JSON.parse(JSON.stringify(inInfo));\n  newInputInfo.shapeInfo.logicalShape = squeezedShape;\n  return newInputInfo;\n}\n\nfunction getSqueezedParams(params: string[], keptDims: number[]): string {\n  return keptDims.map(d => params[d]).join(', ');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ArgMinMaxPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      shape: number[], windowSize: number, op: 'max'|'min',\n      firstPass: boolean) {\n    util.assert(\n        shape.length > 2,\n        () => `Packed arg${\n            op.charAt(0).toUpperCase() +\n            op.slice(1)} supports only inputs with rank above 2.`);\n    const inSize = shape[shape.length - 1];\n    const outSize = Math.ceil(inSize / windowSize);\n    this.outputShape = shape.slice(0, -1);\n    if (outSize > 1) {\n      this.outputShape.push(outSize);\n    }\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    const outShape = this.outputShape;\n    const rank = outShape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n\n    let sourceLocSetup;\n    let sourceRank;\n    if (outSize === 1) {\n      sourceRank = rank + 1;\n      const sourceLocDType = getCoordsDataType(sourceRank);\n      sourceLocSetup = `\n        ${sourceLocDType} sourceLocR = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 1]};\n        ${sourceLocDType} sourceLocG = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 2]};\n        ${sourceLocDType} sourceLocA = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 1]};\n        ${sourceLocDType} sourceLocB = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 2]};`;\n    } else {\n      sourceRank = rank;\n      sourceLocSetup = `\n        ${dtype} sourceLocR = coords;\n        ++${coords[rank - 1]};\n        ${dtype} sourceLocG = coords;\n        ++${coords[rank - 2]};\n        ${dtype} sourceLocA = coords;\n        --${coords[rank - 1]};\n        ${dtype} sourceLocB = coords;\n        --${coords[rank - 2]};`;\n    }\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, sourceRank);\n    const inChannel = '.' + channels[sourceRank - 1];  // e.g. \".b\" for rank 3.\n    const intChannels = channels.map(x => 'int ' + x);\n    const srcRCoords =\n        getChannels('sourceLocR', sourceRank - 1).concat('inIdx.r');\n    const srcGCoords =\n        getChannels('sourceLocG', sourceRank - 1).concat('inIdx.g');\n    const srcBCoords =\n        getChannels('sourceLocB', sourceRank - 1).concat('inIdx.b');\n    const srcACoords =\n        getChannels('sourceLocA', sourceRank - 1).concat('inIdx.a');\n\n    const compOp = (op === 'max') ? 'greaterThan' : 'lessThan';\n    const fetchCandidateIdx = firstPass ? '' : `\n          inIdx = round(vec4(getBestIndicesAChannel(${srcRCoords.join()}),\n                             getBestIndicesAChannel(${srcGCoords.join()}),\n                             getBestIndicesAChannel(${srcBCoords.join()}),\n                             getBestIndicesAChannel(${srcACoords.join()})));`;\n\n    const fetchValue = `vec4(\n            getAChannel(${srcRCoords.join()}),\n            hasNextCol ? getAChannel(${srcGCoords.join()}) : 0.,\n            hasNextRow ? getAChannel(${srcBCoords.join()}) : 0.,\n            hasNextRow && hasNextCol ? getAChannel(${srcACoords.join()}) : 0.)`;\n\n    const getBestIndicesAChannelSnippet = firstPass ? '' : `\n      float getBestIndicesAChannel(${intChannels.join()}) {\n        return getChannel(getBestIndicesA(${channels.join()}),\n                                          vec2(${channels.slice(-2).join()}));\n      }`;\n\n    this.userCode = `\n      float getAChannel(${intChannels.join()}) {\n        return getChannel(getA(${channels.join()}),\n                               vec2(${channels.slice(-2).join()}));\n      }\n      ${getBestIndicesAChannelSnippet}\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        bool hasNextCol = ${coords[rank - 1]} < ${outShape[rank - 1] - 1};\n        bool hasNextRow = ${coords[rank - 2]} < ${outShape[rank - 2] - 1};\n        ${sourceLocSetup}\n        ivec4 srcIdx = ivec4(sourceLocR${inChannel}, sourceLocG${inChannel},\n          sourceLocB${inChannel}, sourceLocA${inChannel}) * ${windowSize};\n        ivec4 inIdx = srcIdx;\n        vec4 bestIndex = vec4(inIdx);\n        vec4 bestValue = ${fetchValue};\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          inIdx = srcIdx;\n          ${fetchCandidateIdx}\n          vec4 candidate = ${fetchValue};\n          bvec4 nan = isnan(candidate);\n          bvec4 replace = bvec4(\n            vec4(${compOp}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));\n\n          bestValue = vec4(replace.x  ? candidate.x : bestValue.x,\n                           replace.y  ? candidate.y : bestValue.y,\n                           replace.z  ? candidate.z : bestValue.z,\n                           replace.w  ? candidate.w : bestValue.w);\n          bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));\n          srcIdx++;\n        }\n        setOutput(bestIndex);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AvgPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth};\n            wC+= ${dilationWidth}) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n\n            dotProd += dyValue * avgMultiplier;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class AvgPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n\n              dotProd += dyValue * avgMultiplier;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\n// (Ar + Ai)(Br + Bi) =\n// ArBr + ArBi + AiBr + AiBi = ArBr - AB + ArBi + AiBr\n// Yr = ArBr - AB\n// Yi = ArBi + AiBr\nexport const COMPLEX_MULTIPLY = {\n  REAL: 'return areal * breal - aimag * bimag;',\n  IMAG: 'return areal * bimag + aimag * breal;'\n};\n\nexport class BinaryOpComplexProgram implements GPGPUProgram {\n  variableNames = ['AReal', 'AImag', 'BReal', 'BImag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    this.userCode = `\n      float binaryOpComplex(\n          float areal, float aimag, float breal, float bimag) {\n        ${op}\n      }\n\n      void main() {\n        float areal = getARealAtOutCoords();\n        float aimag = getAImagAtOutCoords();\n        float breal = getBRealAtOutCoords();\n        float bimag = getBImagAtOutCoords();\n        setOutput(binaryOpComplex(areal, aimag, breal, bimag));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nconst CHECK_NAN_SNIPPET = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const ADD = 'return a + b;';\nexport const SUB = 'return a - b;';\nexport const MUL = 'return a * b;';\n\n// We use native integer division to deal with floating point imprecision. Since\n// we implement floor division and glsl implements truncated division, we\n// correct for this by subtracting 1 from result when the result is negative and\n// there is a remainder.\nexport const INT_DIV = `\n  float s = sign(a) * sign(b);\n  int ia = round(a);\n  int ib = round(b);\n  if (ib != 0) {\n    // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n    return float(idiv(ia, ib, s));\n  } else {\n    return NAN;\n  }\n`;\n\nexport const POW = `\nif(a < 0.0 && floor(b) < b){\n  return NAN;\n}\nif (b == 0.0) {\n  return 1.0;\n}\nreturn (round(mod(b, 2.0)) != 1) ?\n    pow(abs(a), b) : sign(a) * pow(abs(a), b);\n`;\nexport const SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const EQUAL = `return float(a == b);`;\n\nexport const NOT_EQUAL = `return float(a != b);`;\n\nexport const LESS = `return float(a < b);`;\n\nexport const LESS_EQUAL = `return float(a <= b);`;\n\nexport const GREATER = `return float(a > b);`;\n\nexport const GREATER_EQUAL = `return float(a >= b);`;\n\nexport const LOGICAL_AND = `return float(a >= 1.0 && b >= 1.0);`;\n\nexport const LOGICAL_OR = `return float(a >= 1.0 || b >= 1.0);`;\n\nexport const MAX = CHECK_NAN_SNIPPET + `\n  return max(a, b);\n`;\nexport const MIN = CHECK_NAN_SNIPPET + `\n  return min(a, b);\n`;\nexport const MOD = `if (b == 0.0) return NAN;\n  return mod(a, b);`;\n\nexport const ELU_DER = `return (b >= 1.0) ? a : a * (b + 1.0);`;\n\nexport const PRELU = `return (a < 0.) ? b * a : a;`;\n\nexport class BinaryOpProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    this.userCode = `\n      float binaryOperation(float a, float b) {\n        ${op}\n      }\n\n      void main() {\n        float a = getAAtOutCoords();\n        float b = getBAtOutCoords();\n        setOutput(binaryOperation(a, b));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nconst CHECK_NAN_SNIPPET = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\nexport const INT_DIV = `\n  ivec4 ia = round(a);\n  ivec4 ib = round(b);\n  bvec4 cond = notEqual(ib, ivec4(0));\n  ivec4 result = ivec4(0);\n  vec4 s = sign(a) * sign(b);\n\n  // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n  if (cond[0]) {\n    result[0] = idiv(ia[0], ib[0], s[0]);\n  }\n  if (cond[1]) {\n    result[1] = idiv(ia[1], ib[1], s[1]);\n  }\n  if (cond[2]) {\n    result[2] = idiv(ia[2], ib[2], s[2]);\n  }\n  if (cond[3]) {\n    result[3] = idiv(ia[3], ib[3], s[3]);\n  }\n  return vec4(result);\n`;\n\nexport const POW = `\n  // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.\n  vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));\n  vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);\n  vec4 result = multiplier * pow(abs(a), b);\n\n  // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS\n  bvec4 isExpZero = equal(b, vec4(0.0));\n  result.r = isExpZero.r ? 1.0 : result.r;\n  result.g = isExpZero.g ? 1.0 : result.g;\n  result.b = isExpZero.b ? 1.0 : result.b;\n  result.a = isExpZero.a ? 1.0 : result.a;\n\n  vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const PRELU = `\n  vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));\n  return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);\n`;\n\nexport const ELU_DER = `\n  vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));\n  return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));\n`;\n\nexport const EQUAL = `\n  return vec4(equal(a, b));\n`;\n\nexport const NOT_EQUAL = `\n  return vec4(notEqual(a, b));\n`;\n\nexport const LESS = `\n  return vec4(lessThan(a, b));\n`;\n\nexport const LESS_EQUAL = `\n  return vec4(lessThanEqual(a, b));\n`;\n\nexport const GREATER = `\n  return vec4(greaterThan(a, b));\n`;\n\nexport const GREATER_EQUAL = `\n  return vec4(greaterThanEqual(a, b));\n`;\n\nexport const LOGICAL_AND = `\n  return vec4(\n    vec4(greaterThanEqual(a, vec4(1.0))) *\n    vec4(greaterThanEqual(b, vec4(1.0))));\n`;\n\nexport const LOGICAL_OR = `\n  return min(\n    vec4(greaterThanEqual(a, vec4(1.0))) +\n    vec4(greaterThanEqual(b, vec4(1.0))),\n    vec4(1.0));\n`;\n\nexport const MAX = `\n  vec4 result = vec4(max(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MIN = `\n  vec4 result = vec4(min(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MOD = `\n  vec4 result = mod(a, b);\n  vec4 isNaN = vec4(equal(b, vec4(0.0)));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport class BinaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n  supportsBroadcasting = true;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      op: string, aShape: number[], bShape: number[],\n      checkOutOfBounds = false) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const rank = this.outputShape.length;\n    let checkOutOfBoundsString = '';\n    if (checkOutOfBounds) {\n      if (rank === 0 || util.sizeFromShape(this.outputShape) === 1) {\n        checkOutOfBoundsString = `\n          result.y = 0.;\n          result.z = 0.;\n          result.w = 0.;\n        `;\n      } else {\n        const dtype = getCoordsDataType(rank);\n        checkOutOfBoundsString = `\n          ${dtype} coords = getOutputCoords();\n        `;\n        if (rank === 1) {\n          checkOutOfBoundsString += `\n            result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;\n            result.z = 0.;\n            result.w = 0.;\n          `;\n        } else {\n          const channels = getChannels('coords', rank);\n          checkOutOfBoundsString += `\n            bool nextRowOutOfBounds =\n              (${channels[rank - 2]} + 1) >= ${this.outputShape[rank - 2]};\n            bool nextColOutOfBounds =\n              (${channels[rank - 1]} + 1) >= ${this.outputShape[rank - 1]};\n            result.y = nextColOutOfBounds ? 0. : result.y;\n            result.z = nextRowOutOfBounds ? 0. : result.z;\n            result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;\n          `;\n        }\n      }\n    }\n\n    this.userCode = `\n      vec4 binaryOperation(vec4 a, vec4 b) {\n        ${op}\n      }\n\n      void main() {\n        vec4 a = getAAtOutCoords();\n        vec4 b = getBAtOutCoords();\n\n        vec4 result = binaryOperation(a, b);\n        ${checkOutOfBoundsString}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        float value = getAAtOutCoords();\n        if (isnan(value)) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, minVal, maxVal));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        vec4 value = getAAtOutCoords();\n\n        if (any(isnan(value))) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, vec4(minVal), vec4(maxVal)));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ComplexAbsProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(shape: number[]) {\n    this.outputShape = shape;\n    this.userCode = `\n      void main() {\n        float re = abs(getRealAtOutCoords());\n        float im = abs(getImagAtOutCoords());\n        float mx = max(re, im);\n\n        // sadly the length function in glsl is not underflow-safe\n        // (at least not on Intel GPUs). So the safe solution is\n        // to ensure underflow-safety in all cases.\n        setOutput(\n          mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))\n        );\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ConcatProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  // Concats 2d tensors along axis=1. See comments in MathBackendWebGL.concat().\n  constructor(shapes: Array<[number, number]>) {\n    this.outputShape = backend_util.computeOutShape(shapes, 1 /* axis */);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][1];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][1];\n    }\n\n    const snippets = [`if (yC < ${offsets[0]}) setOutput(getT0(yR, yC));`];\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      snippets.push(\n          `else if (yC < ${offsets[i]}) ` +\n          `setOutput(getT${i}(yR, yC-${shift}));`);\n    }\n    const lastIndex = offsets.length;\n    const lastShift = offsets[offsets.length - 1];\n    snippets.push(`else setOutput(getT${lastIndex}(yR, yC-${lastShift}));`);\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int yR = coords.x;\n        int yC = coords.y;\n\n        ${snippets.join('\\n        ')}\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ConcatPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(shapes: number[][], axis: number) {\n    this.outputShape = backend_util.computeOutShape(shapes, axis);\n    const shape = this.outputShape;\n    const rank = shape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][axis];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][axis];\n    }\n\n    const channel = channels[axis];\n    const lastChannels = channels.slice(-2);\n    const allChannels = channels.join();\n\n    let getValueSnippet = `if (${channel} < ${offsets[0]}) {\n        return getChannel(\n            getT0(${allChannels}), vec2(${lastChannels.join()}));\n        }`;\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      // Note: the >= comparison below may seem unnecessary given the check\n      // above but is needed to workaround branch execution issues on some\n      // devices. It makes all the conditions exclusive without relying on\n      // execution order.\n      getValueSnippet += `\n        if (${channel} < ${offsets[i]}  && ${channel} >= ${offsets[i - 1]}) {\n          return getChannel(\n            getT${i}(${shiftedChannels(channels, channel, shift)}),\n            vec2(${shiftedChannels(lastChannels, channel, shift)}));\n        }`;\n    }\n    const lastIndex = offsets.length;\n    const shift = offsets[offsets.length - 1];\n    getValueSnippet += `\n        return getChannel(\n          getT${lastIndex}(${shiftedChannels(channels, channel, shift)}),\n          vec2(${shiftedChannels(lastChannels, channel, shift)}));`;\n\n    this.userCode = `\n      float getValue(${channels.map(x => 'int ' + x)}) {\n        ${getValueSnippet}\n      }\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        vec4 result = vec4(getValue(${coords}), 0., 0., 0.);\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} + 1;\n        if (${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.g = getValue(${coords});\n        }\n\n        ${coords[rank - 2]} = ${coords[rank - 2]} + 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]}) {\n          result.a = getValue(${coords});\n        }\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} - 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]} &&\n            ${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.b = getValue(${coords});\n        }\n        setOutput(result);\n      }\n    `;\n  }\n}\n\n/**\n * Return an expression for coordinates into a vector where a given channel\n * will be offset by [shift].\n *\n * @param channels the channels to consider\n * @param channel the channel we want shifted\n * @param shift  the amount to subtract from the channel.\n *\n * @returns a string of the form 'x, y-[shift], z' where any one channel can\n * have the shift applied.\n */\nfunction shiftedChannels(channels: string[], channel: string, shift: number) {\n  const channelIdx = channels.indexOf(channel);\n  const res = channels.map((c, idx) => {\n    if (idx === channelIdx) {\n      return `${c} - ${shift}`;\n    } else {\n      return c;\n    }\n  });\n  return res.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int d2 = coords.w;\n\n        // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              if (${isChannelsLast}) {\n                float dyValue = getDy(b, yR, yC, d2);\n                float xValue = getX(b, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              } else {\n                float dyValue = getDy(b, d2, yR, yC);\n                float xValue = getX(b, d1, xR, xC);\n                dotProd += (xValue * dyValue);\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[${channelDim}];\n\n        ivec2 dyCorner = ivec2(coords[${rowDim}], coords[${colDim}]) - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n\n              if (${isChannelsLast}) {\n                float xValue = getDy(batch, idyR, idyC, d2);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              } else {\n                float xValue = getDy(batch, d2, idyR, idyC);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    this.userCode = `\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int wF = coords.x;\n        int wR = coords.y;\n        int wC = coords.z;\n        int d1 = coords.w;\n        int d2 = coords.u;\n\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yF = 0; yF < ${convInfo.outDepth}; yF++) {\n            int xF = wF + yF * ${strideDepth} - ${padFront};\n\n            if (xF < 0 || xF >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n              int xR = wR + yR * ${strideHeight} - ${padTop};\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n                int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float dyValue = getDy(b, yF, yR, yC, d2);\n                float xValue = getX(b, xF, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padFront = filterDepth - 1 - convInfo.padInfo.front;\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d1 = coords.u;\n\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyFCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          float dyF = float(dyFCorner + wF) / ${strideDepth}.0;\n\n          if (dyF < 0.0 || dyF >= ${convInfo.outDepth}.0 || fract(dyF) > 0.0) {\n            continue;\n          }\n          int idyF = int(dyF);\n\n          int wFPerm = ${filterDepth} - 1 - wF;\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n              fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            int wRPerm = ${filterHeight} - 1 - wR;\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              int wCPerm = ${filterWidth} - 1 - wC;\n\n              for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n                float xValue = getDy(batch, idyF, idyR, idyC, d2);\n                float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int dm = coords.w;\n        int d2 = d1 * ${channelMul} + dm;\n\n        float dotProd = 0.0;\n\n        // TO DO: Vec4 over the batch size\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float dyValue = getDy(b, yR, yC, d2);\n              float xValue = getX(b, xR, xC, d1);\n              dotProd += (xValue * dyValue);\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class DepthwiseConv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[3];\n        ivec2 dyCorner = coords.yz - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        float dotProd = 0.0;\n\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            // TO DO: Vec4 over the channelMul\n            for (int dm = 0; dm < ${channelMul}; dm++) {\n              int d2 = d1 * ${channelMul} + dm;\n              float xValue = getDy(batch, idyR, idyC, d2);\n              float wValue = getW(wRPerm, wCPerm, d1, dm);\n              dotProd += xValue * wValue;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivationWeights = false) {\n    this.outputShape = convInfo.outShape;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivationWeights) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivationWeights) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d2 = coords[${channelDim}];\n\n        ivec2 xRCCorner =\n            ivec2(coords[${rowDim}], coords[${colDim}]) * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${convInfo.inWidth}) {\n              continue;\n            }\n\n            for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n              vec4 wValues = vec4(\n                getW(wR, wC, d1, d2),\n                getW(wR, wC, d1 + 1, d2),\n                getW(wR, wC, d1 + 2, d2),\n                getW(wR, wC, d1 + 3, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec4 xValues = vec4(\n                  getX(batch, xR, xC, d1),\n                  getX(batch, xR, xC, d1 + 1),\n                  getX(batch, xR, xC, d1 + 2),\n                  getX(batch, xR, xC, d1 + 3)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec4 xValues = vec4(\n                  getX(batch, d1, xR, xC),\n                  getX(batch, d1 + 1, xR, xC),\n                  getX(batch, d1 + 2, xR, xC),\n                  getX(batch, d1 + 3, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n\n            if (${inputDepthVec4Remainder === 1}) {\n\n              if (${isChannelsLast}) {\n                dotProd +=\n                    getX(batch, xR, xC, ${inputDepthNearestVec4}) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              } else {\n                dotProd +=\n                    getX(batch, ${inputDepthNearestVec4}, xR, xC) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              }\n\n            } else if (${inputDepthVec4Remainder === 2}) {\n              vec2 wValues = vec2(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec2 xValues = vec2(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec2 xValues = vec2(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            } else if (${inputDepthVec4Remainder === 3}) {\n              vec3 wValues = vec3(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 2, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec3 xValues = vec3(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec3 xValues = vec3(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 2, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            }\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nexport class Conv3DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.outShape;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n\n    this.userCode = `\n      const ivec3 strides = ivec3(${strideDepth}, ${strideHeight}, ${\n        strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d2 = coords.u;\n\n        ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xFCorner = xFRCCorner.x;\n        int xRCorner = xFRCCorner.y;\n        int xCCorner = xFRCCorner.z;\n\n        // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get\n        // y(yF, yR, yC, d2). ? = to be determined. : = across all\n        // values in that axis.\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          int xF = xFCorner + wF * ${dilationDepth};\n\n          if (xF < 0 || xF >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            int xR = xRCorner + wR * ${dilationHeight};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n                vec4 xValues = vec4(\n                  getX(batch, xF, xR, xC, d1),\n                  getX(batch, xF, xR, xC, d1 + 1),\n                  getX(batch, xF, xR, xC, d1 + 2),\n                  getX(batch, xF, xR, xC, d1 + 3)\n                );\n                vec4 wValues = vec4(\n                  getW(wF, wR, wC, d1, d2),\n                  getW(wF, wR, wC, d1 + 1, d2),\n                  getW(wF, wR, wC, d1 + 2, d2),\n                  getW(wF, wR, wC, d1 + 3, d2)\n                );\n\n                dotProd += dot(xValues, wValues);\n              }\n\n              if (${inputDepthVec4Remainder === 1}) {\n                dotProd +=\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}) *\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2);\n              } else if (${inputDepthVec4Remainder === 2}) {\n                vec2 xValues = vec2(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                vec2 wValues = vec2(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else if (${inputDepthVec4Remainder === 3}) {\n                vec3 xValues = vec3(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                vec3 wValues = vec3(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 2, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2 / ${channelMul};\n        int q = d2 - d1 * ${channelMul};\n\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${xNumRows}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${xNumCols}) {\n              continue;\n            }\n\n            float xVal = getX(batch, xR, xC, d1);\n            float wVal = getW(wR, wC, d1, q);\n            dotProd += xVal * wVal;\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConvPacked2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const texelsAcross = filterWidth;\n\n    let mainLoop = `int xR; int xC; int xCOffset;`;\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `\n          vec4 xTexelR${r}C${c * 2} = vec4(0.);\n          vec4 wR${r}C${c} = vec4(0.);\n          vec4 xR${r}C${c} = vec4(0.);`;\n      }\n    }\n\n    /**\n     * This vectorized implementation works by gathering the values needed for\n     * each output channel's dot product into vec4's and then multiplying them\n     * all together (this happens in the final double for-loop below). Most of\n     * the main loop consists of constructing these vec4's with the minimum\n     * number of texture2D calls, which means making use of all four returned\n     * values from a texture2D call at once.\n     */\n    for (let r = 0; r < filterHeight; r++) {\n      for (let texelC = 0; texelC < texelsAcross; texelC++) {\n        const c = texelC * 2;\n\n        mainLoop += `\n          xR = xRCorner + ${r * dilationHeight};\n          xC = xCCorner + ${c * dilationWidth};\n        `;\n\n        if (strideWidth === 1) {\n          if (c < filterWidth) {\n            // If padding is odd, the outer texels have to be composed.\n            if (padLeft % 2 === 1) {\n              // TODO: Ensure vec4 previous does not result in redundant sample,\n              // and avoid setting xTexelRC's that exceed the boundary in the\n              // first place rather than resetting them to vec4(0)).\n\n              // To compute xCOffset:\n              // - If padding is odd, we must add 1 to ensure we ask for an\n              // even-numbered row.\n              // - We subtract 2 to access the previous texel.\n\n              mainLoop += `\n                xCOffset = xC + 1;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    xTexelR${r}C${c}.zw = vec2(0.);\n                  }\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + 1 - 2;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  vec4 previous = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    previous.zw = vec2(0.);\n                  }\n\n                  xR${r}C${c} = vec4(previous.zw, xTexelR${r}C${c}.xy);\n                } else {\n                  xR${r}C${c} = vec4(0, 0, xTexelR${r}C${c}.xy);\n                }\n              `;\n            } else {\n              // Padding is even, so xRC corresponds to a single texel.\n              mainLoop += `\n                if(xR >= 0 && xR < ${xNumRows} && xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xR${r}C${c} = xTexelR${r}C${c};\n              `;\n            }\n\n            if (c + 1 < filterWidth) {\n              // If dilation is even, the second entry should match the first\n              // (either both are composed or both are single samples). But if\n              // dilation is odd, then the second entry should be the opposite\n              // of the first (if the first is composed, the second is a single\n              // sample, and vice versa.)\n\n              const nextTexelOffset = padLeft % 2 === 0 ?\n                  util.nearestLargerEven(dilationWidth) :\n                  dilationWidth;\n\n              if ((dilationWidth % 2 === 0 && padLeft % 2 === 1) ||\n                  (dilationWidth % 2 !== 0 && padLeft % 2 !== 1)) {\n                mainLoop += `\n                  xCOffset = xC + ${padLeft % 2} + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n                `;\n\n                // If dilation > 1 then the xRC's will not be able to share any\n                // values, so each xRC will require two unique calls to getX.\n                if (dilationWidth > 1) {\n                  mainLoop += `\n                    xCOffset -= 2;\n                    if(xR >= 0 && xR < ${xNumRows} &&\n                      xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                      xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                    } else {\n                      xTexelR${r}C${c} = vec4(0.);\n                    }\n                  `;\n                }\n\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.xy);\n                `;\n              } else {\n                mainLoop += `\n                  xCOffset = xC + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n\n                  xR${r}C${c + 1} = xTexelR${r}C${c + 2};\n                `;\n              }\n            }\n          }\n        } else {  // stride > 1\n          if (c < filterWidth) {\n            mainLoop += `\n              if(xR >= 0 && xR < ${xNumRows}) {\n            `;\n\n            // Depending on whether padLeft is even or odd, we want either the\n            // xy or zw channels from X texels for xR${r}C${c}. If padLeft is\n            // even, xR${r}C${c + 1} is simply the zw channels of texels we've\n            // already sampled. But if padLeft is odd, xR${r}C{$c + 1}.zw will\n            // need to come from the xy channels of a new texel, hence the `vec4\n            // final` initialized below.\n            if (padLeft % 2 === 1) {\n              mainLoop += `\n                xCOffset = xC + 1 - ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                if(xC + 1 >= 0 && xC + 1 < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xC + 1, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  vec4 final = vec4(0.);\n                  xCOffset = xC + 1 + ${strideWidth};\n                  if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    final = getX(batch, xR, xCOffset, d1);\n                  }\n                  xR${r}C${c + 1} = vec4(xTexelR${r}C${c + 2}.xy, final.xy);\n                `;\n              }\n            } else {\n              mainLoop += `\n                if(xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.xy, xTexelR${r}C${c + 2}.xy);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n                `;\n              }\n            }\n\n            mainLoop += `}`;\n          }\n        }\n\n        if (c < filterWidth) {\n          mainLoop += `\n            vec4 wTexelR${r}C${c} = getW(${r}, ${c}, d1, q);\n            wR${r}C${c} = vec4(wTexelR${r}C${c}.xz, wTexelR${r}C${c}.xz);\n          `;\n\n          if (c + 1 < filterWidth) {\n            mainLoop += `\n              vec4 wTexelR${r}C${c + 1} = getW(${r}, ${c + 1}, d1, q);\n              wR${r}C${c + 1} =\n                vec4(wTexelR${r}C${c + 1}.xz, wTexelR${r}C${c + 1}.xz);`;\n          }\n        }\n      }\n    }\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `dotProd += xR${r}C${c} * wR${r}C${c};`;\n      }\n    }\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2;\n        int q = 0;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        vec4 dotProd = vec4(0.);\n\n        ${mainLoop}\n\n        vec4 result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class CropAndResizeProgram implements GPGPUProgram {\n  variableNames = ['Image', 'Boxes', 'BoxInd'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], boxShape: [number, number],\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number) {\n    const [batch, imageHeight, imageWidth, depth] = imageShape;\n    const [numBoxes, ] = boxShape;\n    const [cropHeight, cropWidth] = cropSize;\n    this.outputShape = [numBoxes, cropHeight, cropWidth, depth];\n    const methodId = method === 'bilinear' ? 1 : 0;\n\n    const [inputHeightFloat, inputWidthFloat] =\n        [`${imageHeight - 1}.0`, `${imageWidth - 1}.0`];\n\n    const [heightRatio, heightScale, inY] = cropHeight > 1 ?\n        [\n          `${(imageHeight - 1) / (cropHeight - 1)}`,\n          '(y2-y1) * height_ratio',\n          `y1*${inputHeightFloat} + float(y)*(height_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (y1+y2) * ${inputHeightFloat}`,\n        ];\n    const [widthRatio, widthScale, inX] = cropWidth > 1 ?\n        [\n          `${(imageWidth - 1) / (cropWidth - 1)}`,\n          '(x2-x1) * width_ratio',\n          `x1*${inputWidthFloat} + float(x)*(width_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (x1+x2) * ${inputWidthFloat}`,\n        ];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc\n    this.userCode = `\n      const float height_ratio = float(${heightRatio});\n      const float width_ratio = float(${widthRatio});\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int y = coords[1];\n        int x = coords[2];\n        int d = coords[3];\n\n        // get box vals\n        float y1 = getBoxes(b,0);\n        float x1 = getBoxes(b,1);\n        float y2 = getBoxes(b,2);\n        float x2 = getBoxes(b,3);\n\n        // get image in batch index\n        int bInd = round(getBoxInd(b));\n        if(bInd < 0 || bInd >= ${batch}) {\n          return;\n        }\n\n        float height_scale = ${heightScale};\n        float width_scale = ${widthScale};\n\n        float in_y = ${inY};\n        if( in_y < 0.0 || in_y > ${inputHeightFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n        float in_x = ${inX};\n        if( in_x < 0.0 || in_x > ${inputWidthFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n\n        vec2 sourceFracIndexCR = vec2(in_x,in_y);\n        if(${methodId} == 1) {\n          // Compute the four integer indices.\n          ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);\n          ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));\n\n          float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);\n          float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);\n          float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);\n          float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);\n\n          vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);\n\n          float top = topLeft + (topRight - topLeft) * fracCR.x;\n          float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;\n          float newValue = top + (bottom - top) * fracCR.y;\n          setOutput(newValue);\n        } else {\n          // Compute the coordinators of nearest neighbor point.\n          ivec2 sourceNearestCR = ivec2(floor(\n            sourceFracIndexCR + vec2(0.5,0.5)));\n          float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);\n          setOutput(newValue);\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class CumSumProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  index: WebGLUniformLocation;\n\n  constructor(shape: number[], exclusive: boolean, reverse: boolean) {\n    this.outputShape = shape;\n    const rank = shape.length;\n    const val = exclusive ? '0.0' : `getX(${getCoords(rank, 'coords')})`;\n    const length = shape[shape.length - 1];\n    let condition = '';\n    let idxString = '';\n    // When exclusive is set, the cumsum op becomes roll op that copies the\n    // value from the previous index based on the direction specified by the\n    // reverse flag.\n    if (exclusive) {\n      condition = reverse ? `end != ${length - 1}` : 'end != 0';\n      idxString = reverse ? 'end + 1' : 'end - 1';\n    } else {\n      condition = reverse ? `end + pow2 < ${length}` : 'end >= pow2';\n      idxString = (reverse ? 'end + pow2' : 'end - pow2');\n    }\n\n    this.userCode = `\n      uniform float index;\n      void main() {\n        ${getCoordsDataType(rank)} coords = getOutputCoords();\n        int end = ${getFinalCoord(rank, 'coords')};\n        float val = ${val};\n        int pow2 = int(pow(2.0, index));\n        if (${condition}) {\n          int idx = ${idxString};\n          ${getFinalCoord(rank, 'coords')} = idx;\n          val += getX(${getCoords(rank, 'coords')});\n        }\n        setOutput(val);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(index: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.index == null) {\n        this.index = gpgpu.getUniformLocation(webGLProgram, 'index');\n      }\n      gpgpu.gl.uniform1f(this.index, index);\n    };\n  }\n}\n\nfunction getCoords(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.x, ${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.x, ${name}.y, ${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.x, ${name}.y, ${name}.z, ${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n\nfunction getFinalCoord(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: [number, number, number];\n  packedInputs = false;\n  packedOutput = true;\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getA(rc.x, rc.y, rc.z);\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: [number, number, number];\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthToSpaceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  blockSize: number;\n  dataFormat: string;\n\n  constructor(\n      outputShape: number[], blockSize: number, dataFormat: 'NHWC'|'NCHW') {\n    this.outputShape = outputShape;\n    this.blockSize = blockSize;\n    this.dataFormat = dataFormat;\n    this.userCode = `\n    void main() {\n      ivec4 coords = getOutputCoords();\n      int b = coords[0];\n      int h = ${this.getHeightCoordString()};\n      int w = ${this.getWidthCoordString()};\n      int d = ${this.getDepthCoordString()};\n\n      int in_h = h / ${blockSize};\n      int offset_h = imod(h, ${blockSize});\n      int in_w = w / ${blockSize};\n      int offset_w = imod(w, ${blockSize});\n      int offset_d = (offset_h * ${blockSize} + offset_w) *\n        ${this.getOutputDepthSize()};\n      int in_d = d + offset_d;\n\n      float result = ${this.getInputSamplingString()};\n      setOutput(result);\n    }\n  `;\n  }\n\n  private getHeightCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[1]`;\n    } else {\n      return `coords[2]`;\n    }\n  }\n\n  private getWidthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[2]`;\n    } else {\n      return `coords[3]`;\n    }\n  }\n\n  private getDepthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[3]`;\n    } else {\n      return `coords[1]`;\n    }\n  }\n\n  private getOutputDepthSize(): number {\n    if (this.dataFormat === 'NHWC') {\n      return this.outputShape[3];\n    } else {\n      return this.outputShape[1];\n    }\n  }\n\n  private getInputSamplingString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `getX(b, in_h, in_w, in_d)`;\n    } else {\n      return `getX(b, in_d, in_h, in_w)`;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DiagProgram implements GPGPUProgram {\n  variableNames = ['X'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(size: number) {\n    this.outputShape = [size, size];\n    this.userCode = `\n      void main() {\n          ivec2 coords = getOutputCoords();\n          float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;\n          setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        float x = getAAtOutCoords();\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = false;\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: [number, number, number]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n        float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class EncodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let output = `result`;\n    if (inputIsUnsignedByte) {\n      output = `floor(result * 255. + 0.5)`;\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        int flatIndex = getFlatIndex(coords);\n        int offset = imod(flatIndex, 4);\n\n        flatIndex = idiv(flatIndex, 4, 1.);\n\n        int r = flatIndex / ${width};\n        int c = imod(flatIndex, ${width});\n        vec2 uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n        vec4 values = ${glsl.texture2D}(A, uv);\n\n        float result;\n\n        if(offset == 0) {\n          result = values[0];\n        } else if(offset == 1) {\n          result = values[1];\n        } else if(offset == 2) {\n          result = values[2];\n        } else {\n          result = values[3];\n        }\n\n        ${glsl.output} = vec4(${output}, 0., 0., 0.);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\n/*\nThis is how the shader encodes a tensor with shape = [2, 3, 5]\n(indices are [batch, row, col]).\n\n000|001   002|003   004|xxx   020|021   022|023   024|xxx\n-------   -------   -------   -------   -------   -------\n010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\n100|101   102|103   104|xxx   120|121   122|123   124|xxx\n-------   -------   -------   -------   -------   -------\n110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\nSingle texels contain only values from the same batch, and from adjacent rows\nand columns.\n */\n\nexport class EncodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let mainLoop = '';\n    let output = 'result';\n    if (inputIsUnsignedByte) {\n      output = 'floor(result * 255. + 0.5)';\n    }\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        const channel = row * 2 + col;\n\n        mainLoop += `\n          localCoords = coords;\n          if(localCoords[2] + ${col} < ${outputShape[2]}) {\n            localCoords[2] += ${col};\n            if(localCoords[1] + ${row} < ${outputShape[1]}) {\n              localCoords[1] += ${row};\n\n              flatIndex = getFlatIndex(localCoords);\n              offset = imod(flatIndex, 4);\n\n              flatIndex = idiv(flatIndex, 4, 1.);\n\n              r = flatIndex / ${width};\n              c = imod(flatIndex, ${width});\n              uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n              values = ${glsl.texture2D}(A, uv);\n\n              if(offset == 0) {\n                result[${channel}] = values[0];\n              } else if(offset == 1) {\n                result[${channel}] = values[1];\n              } else if(offset == 2) {\n                result[${channel}] = values[2];\n              } else {\n                result[${channel}] = values[3];\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        vec4 result = vec4(0.);\n        int flatIndex, r, c, offset;\n        ivec3 localCoords;\n        vec2 uv;\n        vec4 values;\n\n        ${mainLoop}\n\n        ${glsl.output} = ${output};\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const COMPLEX_FFT = {\n  REAL: 'return real * expR - imag * expI;',\n  IMAG: 'return real * expI + imag * expR;'\n};\n\nexport class FFTProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, inputShape: [number, number], inverse: boolean) {\n    const innerDim = inputShape[1];\n    this.outputShape = inputShape;\n\n    const exponentMultiplierSnippet =\n        inverse ? `2.0 * ${Math.PI}` : `-2.0 * ${Math.PI}`;\n    const resultDenominator = inverse ? `${innerDim}.0` : '1.0';\n\n    this.userCode = `\n      const float exponentMultiplier = ${exponentMultiplierSnippet};\n\n      float unaryOpComplex(float real, float expR, float imag, float expI) {\n        ${op}\n      }\n\n      float mulMatDFT(int batch, int index) {\n        float indexRatio = float(index) / float(${innerDim});\n        float exponentMultiplierTimesIndexRatio =\n            exponentMultiplier * indexRatio;\n\n        float result = 0.0;\n\n        for (int i = 0; i < ${innerDim}; i++) {\n          // x = (-2|2 * PI / N) * index * i;\n          float x = exponentMultiplierTimesIndexRatio * float(i);\n          float expR = cos(x);\n          float expI = sin(x);\n          float real = getReal(batch, i);\n          float imag = getImag(batch, i);\n\n          result +=\n              unaryOpComplex(real, expR, imag, expI) / ${resultDenominator};\n        }\n\n        return result;\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        setOutput(mulMatDFT(coords[0], coords[1]));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FillProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  valueLoc: WebGLUniformLocation;\n\n  constructor(shape: number[], value: number) {\n    this.variableNames = ['x'];\n    this.outputShape = shape;\n\n    this.userCode = `\n      uniform float value;\n      void main() {\n        // Input can be obtained from uniform value.\n        setOutput(value);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(value: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.valueLoc == null) {\n        this.valueLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'value');\n      }\n      gpgpu.gl.uniform1f(this.valueLoc, value);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherProgram implements GPGPUProgram {\n  variableNames = ['A', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], indicesLength: number, axis: number) {\n    const outputShape: number[] = aShape.slice();\n    outputShape[axis] = indicesLength;\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape, axis);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[], axis: number): string {\n  const rank = aShape.length;\n  if (rank > 4) {\n    throw Error(`Gather for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `int(getIndices(resRC))`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    if (i === axis) {\n      sourceCoords.push(`int(getIndices(${currentCoords[i]}))`);\n    } else {\n      sourceCoords.push(`${currentCoords[i]}`);\n    }\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherNDProgram implements GPGPUProgram {\n  variableNames = ['x', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  constructor(\n      private sliceDim: number, private strides: number[], shape: number[]) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    const strideString = this.sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${this.strides});\n         void main() {\n          ${dtype} coords = getOutputCoords();\n          int flattenIndex = 0;\n          for (int j = 0; j < ${this.sliceDim}; j++) {\n            int index = round(getIndices(coords[0], j));\n            flattenIndex += index * ${strideString};\n          }\n          setOutput(getX(flattenIndex, coords[1]));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PixelData, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {getGlslDifferences} from './glsl_version';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport * as webgl_util from './webgl_util';\n\nexport function createVertexShader(gl: WebGLRenderingContext): WebGLShader {\n  const glsl = getGlslDifferences();\n  const vertexShaderSource = `${glsl.version}\n    precision highp float;\n    ${glsl.attribute} vec3 clipSpacePos;\n    ${glsl.attribute} vec2 uv;\n    ${glsl.varyingVs} vec2 resultUV;\n\n    void main() {\n      gl_Position = vec4(clipSpacePos, 1);\n      resultUV = uv;\n    }`;\n  return webgl_util.createVertexShader(gl, vertexShaderSource);\n}\n\nexport function createVertexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // [x y z u v] * [upper-left, lower-left, upper-right, lower-right]\n  const vertexArray = new Float32Array(\n      [-1, 1, 0, 0, 1, -1, -1, 0, 0, 0, 1, 1, 0, 1, 1, 1, -1, 0, 1, 0]);\n  return webgl_util.createStaticVertexBuffer(gl, vertexArray);\n}\n\nexport function createIndexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // OpenGL (and WebGL) have \"CCW == front\" winding\n  const triangleVertexIndices = new Uint16Array([0, 1, 2, 2, 1, 3]);\n  return webgl_util.createStaticIndexBuffer(gl, triangleVertexIndices);\n}\n\nfunction createAndConfigureTexture(\n    gl: WebGLRenderingContext, width: number, height: number,\n    internalFormat: number, textureFormat: number,\n    textureType: number): WebGLTexture {\n  webgl_util.validateTextureSize(width, height);\n  const texture = webgl_util.createTexture(gl);\n\n  const tex2d = gl.TEXTURE_2D;\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(tex2d, texture));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MIN_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MAG_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          tex2d, 0, internalFormat, width, height, 0, textureFormat,\n          textureType, null));\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n  return texture;\n}\n\nexport function getInternalFormatForFloat32MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatFloat;\n}\n\nexport function createFloat32MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat32MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatHalfFloat;\n}\n\nexport function createFloat16MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, textureConfig.textureTypeHalfFloat);\n}\n\nexport function getInternalFormatForUnsignedBytesMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.downloadTextureFormat;\n}\n\nexport function createUnsignedBytesMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForUnsignedBytesMatrixTexture(textureConfig), gl.RGBA,\n      gl.UNSIGNED_BYTE);\n}\n\nexport function getInternalFormatForPackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedFloat;\n}\n\nexport function createPackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height, getInternalFormatForPackedMatrixTexture(textureConfig),\n      gl.RGBA, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16PackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedHalfFloat;\n}\n\nexport function createFloat16PackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16PackedMatrixTexture(textureConfig), gl.RGBA,\n      textureConfig.textureTypeHalfFloat);\n}\n\nexport function bindVertexProgramAttributeStreams(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    vertexBuffer: WebGLBuffer): boolean {\n  const posOffset = 0;               // x is the first buffer element\n  const uvOffset = 3 * 4;            // uv comes after [x y z]\n  const stride = (3 * 4) + (2 * 4);  // xyz + uv, each entry is 4-byte float.\n  webgl_util.callAndCheck(\n      gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, vertexBuffer));\n  const success = webgl_util.bindVertexBufferToProgramAttribute(\n      gl, program, 'clipSpacePos', vertexBuffer, 3, stride, posOffset);\n  return success &&\n      webgl_util.bindVertexBufferToProgramAttribute(\n          gl, program, 'uv', vertexBuffer, 2, stride, uvOffset);\n}\n\nexport function uploadDenseMatrixToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture, width: number,\n    height: number, data: TypedArray, textureConfig: TextureConfig) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n\n  let dataForUpload: TypedArray, texelDataType: number, internalFormat: number;\n  if (data instanceof Uint8Array) {\n    dataForUpload = new Uint8Array(width * height * 4);\n    texelDataType = gl.UNSIGNED_BYTE;\n    internalFormat = gl.RGBA;\n  } else {\n    dataForUpload = new Float32Array(width * height * 4);\n    texelDataType = gl.FLOAT;\n    internalFormat = textureConfig.internalFormatPackedFloat;\n  }\n\n  dataForUpload.set(data);\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          gl.TEXTURE_2D, 0, internalFormat, width, height, 0, gl.RGBA,\n          texelDataType, dataForUpload));\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function uploadPixelDataToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, pixels.width, pixels.height, 0, gl.RGBA,\n            gl.UNSIGNED_BYTE, (pixels as PixelData).data));\n  } else {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, gl.RGBA, gl.UNSIGNED_BYTE,\n            pixels as ImageData | HTMLImageElement | HTMLCanvasElement |\n                HTMLVideoElement));\n  }\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function createBufferFromOutputTexture(\n    gl2: WebGL2RenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLBuffer {\n  // Create and bind the buffer.\n  const buffer = gl2.createBuffer();\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer));\n\n  // Initialize the buffer to the size of the texture in bytes.\n  const bytesPerFloat = 4;\n  const valuesPerTexel = 4;\n  const bufferSizeBytes = bytesPerFloat * valuesPerTexel * rows * columns;\n\n  webgl_util.callAndCheck(\n      gl2,\n      () => gl2.bufferData(\n          gl2.PIXEL_PACK_BUFFER, bufferSizeBytes, gl2.STREAM_READ));\n\n  // Enqueue a command on the GPU command queue to copy of texture into the\n  // buffer.\n  webgl_util.callAndCheck(\n      gl2, () => gl2.readPixels(0, 0, columns, rows, gl2.RGBA, gl2.FLOAT, 0));\n\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null));\n\n  return buffer;\n}\n\nexport function downloadFloat32MatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer,\n    size: number): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget = new Float32Array(size);\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadByteEncodedFloatMatrixFromOutputTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig) {\n  const [w, h] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n\n  const numChannels = 4;\n  const downloadTarget = new Uint8Array(\n      tex_util.getUnpackedArraySizeFromMatrixSize(rows * columns, numChannels));\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, w, h, textureConfig.downloadTextureFormat, gl.UNSIGNED_BYTE,\n          downloadTarget));\n\n  // By wrapping the buffer in a Float32Array, we use native browser IEEE 754\n  // decoding of the 4 bytes that back each 32 bit float.\n  return new Float32Array(downloadTarget.buffer);\n}\n\nexport function downloadPackedMatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer, batch: number, rows: number,\n    cols: number, physicalRows: number, physicalCols: number,\n    textureConfig: TextureConfig): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget =\n      new Float32Array(tex_util.getPackedRGBAArraySizeFromMatrixShape(\n          physicalRows, physicalCols));\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadMatrixFromPackedOutputTexture(\n    gl: WebGLRenderingContext, physicalRows: number,\n    physicalCols: number): Float32Array {\n  const packedRGBA = new Float32Array(physicalRows * physicalCols * 4);\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, physicalCols, physicalRows, gl.RGBA, gl.FLOAT, packedRGBA));\n\n  return packedRGBA;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, PixelData, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext, setWebGLContext} from './canvas_util';\nimport * as gpgpu_util from './gpgpu_util';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport {WebGL1DisjointQueryTimerExtension, WebGL2DisjointQueryTimerExtension} from './webgl_types';\nimport * as webgl_util from './webgl_util';\n\nexport interface FenceContext {\n  query: WebGLQuery|WebGLSync;\n  isFencePassed(): boolean;\n}\n\nexport class GPGPUContext {\n  gl: WebGLRenderingContext;\n  textureFloatExtension: {};\n  textureHalfFloatExtension: {};\n  colorBufferFloatExtension: {};\n  colorBufferHalfFloatExtension: {};\n  disjointQueryTimerExtension: WebGL2DisjointQueryTimerExtension|\n      WebGL1DisjointQueryTimerExtension;\n  vertexBuffer: WebGLBuffer;\n  indexBuffer: WebGLBuffer;\n  framebuffer: WebGLFramebuffer;\n  outputTexture: WebGLTexture|null = null;\n  program: WebGLProgram|null = null;\n  private disposed = false;\n  private disjoint: boolean;\n  textureConfig: TextureConfig;\n\n  constructor(gl?: WebGLRenderingContext) {\n    const glVersion = env().getNumber('WEBGL_VERSION');\n    if (gl != null) {\n      this.gl = gl;\n      setWebGLContext(glVersion, gl);\n    } else {\n      this.gl = getWebGLContext(glVersion);\n    }\n    // WebGL 2.0 enables texture floats without an extension.\n    let COLOR_BUFFER_FLOAT = 'WEBGL_color_buffer_float';\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (env().getNumber('WEBGL_VERSION') === 1) {\n      const TEXTURE_FLOAT = 'OES_texture_float';\n      const TEXTURE_HALF_FLOAT = 'OES_texture_half_float';\n\n      this.textureFloatExtension =\n          webgl_util.getExtensionOrThrow(this.gl, TEXTURE_FLOAT);\n      if (webgl_util.hasExtension(this.gl, TEXTURE_HALF_FLOAT)) {\n        this.textureHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, TEXTURE_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support half float textures, yet the ' +\n            'environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n\n      this.colorBufferFloatExtension = this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, COLOR_BUFFER_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support color renderable half floats, yet ' +\n            'the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n    } else {\n      COLOR_BUFFER_FLOAT = 'EXT_color_buffer_float';\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_FLOAT)) {\n        this.colorBufferFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      } else if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      } else {\n        throw new Error('GL context does not support color renderable floats');\n      }\n    }\n\n    this.vertexBuffer = gpgpu_util.createVertexBuffer(this.gl);\n    this.indexBuffer = gpgpu_util.createIndexBuffer(this.gl);\n    this.framebuffer = webgl_util.createFramebuffer(this.gl);\n\n    this.textureConfig =\n        tex_util.getTextureConfig(this.gl, this.textureHalfFloatExtension);\n  }\n\n  private get debug(): boolean {\n    return env().getBool('DEBUG');\n  }\n\n  public dispose() {\n    if (this.disposed) {\n      return;\n    }\n    if (this.program != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound WebGLProgram.' +\n          ' This is probably a resource leak, delete the program with ' +\n          'GPGPUContext.deleteProgram before disposing.');\n    }\n    if (this.outputTexture != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound output matrix ' +\n          'texture.  This is probably a resource leak, delete the output ' +\n          'matrix texture with GPGPUContext.deleteMatrixTexture before ' +\n          'disposing.');\n    }\n    const gl = this.gl;\n    webgl_util.callAndCheck(gl, () => gl.finish());\n    webgl_util.callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteFramebuffer(this.framebuffer));\n    webgl_util.callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(\n        gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteBuffer(this.indexBuffer));\n    this.disposed = true;\n  }\n\n  public createFloat32MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat32MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createFloat16MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createUnsignedBytesMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createUnsignedBytesMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public uploadPixelDataToTexture(\n      texture: WebGLTexture,\n      pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadPixelDataToTexture(this.gl, texture, pixels);\n  }\n\n  public uploadDenseMatrixToTexture(\n      texture: WebGLTexture, width: number, height: number, data: TypedArray) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadDenseMatrixToTexture(\n        this.gl, texture, width, height, data, this.textureConfig);\n  }\n\n  public createFloat16PackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16PackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createPackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createPackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public deleteMatrixTexture(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    if (this.outputTexture === texture) {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n      this.outputTexture = null;\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.deleteTexture(texture));\n  }\n\n  public downloadByteEncodedFloatMatrixFromOutputTexture(\n      texture: WebGLTexture, rows: number, columns: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadByteEncodedFloatMatrixFromOutputTexture(\n            this.gl, rows, columns, this.textureConfig));\n  }\n\n  public downloadPackedMatrixFromBuffer(\n      buffer: WebGLBuffer, batch: number, rows: number, columns: number,\n      physicalRows: number, physicalCols: number): Float32Array {\n    return gpgpu_util.downloadPackedMatrixFromBuffer(\n        this.gl, buffer, batch, rows, columns, physicalRows, physicalCols,\n        this.textureConfig);\n  }\n\n  public downloadFloat32MatrixFromBuffer(buffer: WebGLBuffer, size: number):\n      Float32Array {\n    return gpgpu_util.downloadFloat32MatrixFromBuffer(this.gl, buffer, size);\n  }\n\n  public createBufferFromTexture(\n      texture: WebGLTexture, rows: number, columns: number): WebGLBuffer {\n    this.bindTextureToFrameBuffer(texture);\n    const result = gpgpu_util.createBufferFromOutputTexture(\n        this.gl as WebGL2RenderingContext, rows, columns, this.textureConfig);\n    this.unbindTextureToFrameBuffer();\n    return result;\n  }\n\n  public createAndWaitForFence(): Promise<void> {\n    const fenceContext = this.createFence(this.gl);\n    return this.pollFence(fenceContext);\n  }\n\n  private createFence(gl: WebGLRenderingContext): FenceContext {\n    let query: WebGLQuery|WebGLSync;\n    let isFencePassed: () => boolean;\n\n    if (env().getBool('WEBGL_FENCE_API_ENABLED')) {\n      const gl2 = gl as WebGL2RenderingContext;\n\n      const sync = gl2.fenceSync(gl2.SYNC_GPU_COMMANDS_COMPLETE, 0);\n      gl.flush();\n\n      isFencePassed = () => {\n        const status = gl2.clientWaitSync(sync, 0, 0);\n        return status === gl2.ALREADY_SIGNALED ||\n            status === gl2.CONDITION_SATISFIED;\n      };\n\n      query = sync;\n    } else if (\n        env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0) {\n      query = this.beginQuery();\n      this.endQuery();\n      isFencePassed = () => this.isQueryAvailable(\n          query,\n          env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n    } else {\n      // If we have no way to fence, return true immediately. This will fire in\n      // WebGL 1.0 when there is no disjoint query timer. In this case, because\n      // the fence passes immediately, we'll immediately ask for a download of\n      // the texture, which will cause the UI thread to hang.\n      isFencePassed = () => true;\n    }\n\n    return {query, isFencePassed};\n  }\n\n  public downloadMatrixFromPackedTexture(\n      texture: WebGLTexture, physicalRows: number,\n      physicalCols: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadMatrixFromPackedOutputTexture(\n            this.gl, physicalRows, physicalCols));\n  }\n\n  private vertexAttrsAreBound = false;\n\n  public createProgram(fragmentShaderSource: string): WebGLProgram {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    const fragmentShader: WebGLShader =\n        webgl_util.createFragmentShader(gl, fragmentShaderSource);\n    const vertexShader: WebGLShader = gpgpu_util.createVertexShader(gl);\n    const program: WebGLProgram = webgl_util.createProgram(gl);\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, vertexShader));\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, fragmentShader));\n    webgl_util.linkProgram(gl, program);\n    if (this.debug) {\n      webgl_util.validateProgram(gl, program);\n    }\n    if (!this.vertexAttrsAreBound) {\n      this.setProgram(program);\n      this.vertexAttrsAreBound = gpgpu_util.bindVertexProgramAttributeStreams(\n          gl, this.program, this.vertexBuffer);\n    }\n    return program;\n  }\n\n  public deleteProgram(program: WebGLProgram) {\n    this.throwIfDisposed();\n    if (program === this.program) {\n      this.program = null;\n    }\n    if (program != null) {\n      webgl_util.callAndCheck(this.gl, () => this.gl.deleteProgram(program));\n    }\n  }\n\n  public setProgram(program: WebGLProgram|null) {\n    this.throwIfDisposed();\n    this.program = program;\n    if ((this.program != null) && this.debug) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.useProgram(program));\n  }\n\n  public getUniformLocation(\n      program: WebGLProgram, uniformName: string,\n      shouldThrow = true): WebGLUniformLocation {\n    this.throwIfDisposed();\n    if (shouldThrow) {\n      return webgl_util.getProgramUniformLocationOrThrow(\n          this.gl, program, uniformName);\n    } else {\n      return webgl_util.getProgramUniformLocation(\n          this.gl, program, uniformName);\n    }\n  }\n\n  public getAttributeLocation(program: WebGLProgram, attribute: string):\n      number {\n    this.throwIfDisposed();\n    return webgl_util.callAndCheck(\n        this.gl, () => this.gl.getAttribLocation(program, attribute));\n  }\n\n  public getUniformLocationNoThrow(program: WebGLProgram, uniformName: string):\n      WebGLUniformLocation {\n    this.throwIfDisposed();\n    return this.gl.getUniformLocation(program, uniformName);\n  }\n\n  public setInputMatrixTexture(\n      inputMatrixTexture: WebGLTexture, uniformLocation: WebGLUniformLocation,\n      textureUnit: number) {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    webgl_util.bindTextureToProgramUniformSampler(\n        this.gl, inputMatrixTexture, uniformLocation, textureUnit);\n  }\n\n  public setOutputMatrixTexture(\n      outputMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.setOutputMatrixTextureDriver(outputMatrixTexture, columns, rows);\n  }\n\n  public setOutputPackedMatrixTexture(\n      outputPackedMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.throwIfDisposed();\n    const [width, height] =\n        tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n    this.setOutputMatrixTextureDriver(outputPackedMatrixTexture, width, height);\n  }\n\n  public setOutputMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    this.setOutputMatrixWriteRegionDriver(\n        startColumn, startRow, numColumns, numRows);\n  }\n\n  public setOutputPackedMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    throw new Error('setOutputPackedMatrixWriteRegion not implemented.');\n  }\n\n  public debugValidate() {\n    if (this.program != null) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.validateFramebuffer(this.gl);\n  }\n\n  public executeProgram() {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    const gl = this.gl;\n    if (this.debug) {\n      this.debugValidate();\n    }\n    webgl_util.callAndCheck(\n        gl, () => gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_SHORT, 0));\n  }\n\n  public blockUntilAllProgramsCompleted() {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(this.gl, () => this.gl.finish());\n  }\n\n  private getQueryTimerExtension(): WebGL1DisjointQueryTimerExtension\n      |WebGL2DisjointQueryTimerExtension {\n    if (this.disjointQueryTimerExtension == null) {\n      this.disjointQueryTimerExtension =\n          webgl_util.getExtensionOrThrow(\n              this.gl,\n              env().getNumber(\n                  'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2 ?\n                  'EXT_disjoint_timer_query_webgl2' :\n                  'EXT_disjoint_timer_query') as\n              WebGL1DisjointQueryTimerExtension |\n          WebGL2DisjointQueryTimerExtension;\n    }\n    return this.disjointQueryTimerExtension;\n  }\n\n  private getQueryTimerExtensionWebGL2(): WebGL2DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension();\n  }\n\n  private getQueryTimerExtensionWebGL1(): WebGL1DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension() as WebGL1DisjointQueryTimerExtension;\n  }\n\n  beginQuery(): WebGLQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const query = gl2.createQuery();\n      gl2.beginQuery(ext.TIME_ELAPSED_EXT, query);\n      return query;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    const query = ext.createQueryEXT() as WebGLQuery;\n    ext.beginQueryEXT(ext.TIME_ELAPSED_EXT, query);\n    return query;\n  }\n\n  endQuery() {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n      gl2.endQuery(ext.TIME_ELAPSED_EXT);\n      return;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    ext.endQueryEXT(ext.TIME_ELAPSED_EXT);\n  }\n\n  public async waitForQueryAndGetTime(query: WebGLQuery): Promise<number> {\n    await util.repeatedTry(\n        () => this.disposed ||  // while testing contexts are created / disposed\n                                // in rapid succession, so without this check we\n                                // may poll for the query timer indefinitely\n            this.isQueryAvailable(\n                query,\n                env().getNumber(\n                    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION')));\n    return this.getQueryTime(\n        query, env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n  }\n\n  private getQueryTime(query: WebGLQuery, queryTimerVersion: number): number {\n    if (queryTimerVersion === 0) {\n      return null;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n\n      const timeElapsedNanos = gl2.getQueryParameter(query, gl2.QUERY_RESULT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const timeElapsedNanos =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_EXT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    }\n  }\n\n  private isQueryAvailable(query: WebGLQuery, queryTimerVersion: number):\n      boolean {\n    if (queryTimerVersion === 0) {\n      return true;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const available =\n          gl2.getQueryParameter(query, gl2.QUERY_RESULT_AVAILABLE);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const available =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_AVAILABLE_EXT);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    }\n  }\n\n  pollFence(fenceContext: FenceContext) {\n    return new Promise<void>(resolve => {\n      this.addItemToPoll(() => fenceContext.isFencePassed(), () => resolve());\n    });\n  }\n\n  private itemsToPoll: PollItem[] = [];\n\n  pollItems(): void {\n    // Find the last query that has finished.\n    const index = linearSearchLastTrue(this.itemsToPoll.map(x => x.isDoneFn));\n    for (let i = 0; i <= index; ++i) {\n      const {resolveFn} = this.itemsToPoll[i];\n      resolveFn();\n    }\n    this.itemsToPoll = this.itemsToPoll.slice(index + 1);\n  }\n\n  private addItemToPoll(isDoneFn: () => boolean, resolveFn: () => void) {\n    this.itemsToPoll.push({isDoneFn, resolveFn});\n    if (this.itemsToPoll.length > 1) {\n      // We already have a running loop that polls.\n      return;\n    }\n    // Start a new loop that polls.\n    util.repeatedTry(() => {\n      this.pollItems();\n      // End the loop if no more items to poll.\n      return this.itemsToPoll.length === 0;\n    });\n  }\n\n  private bindTextureToFrameBuffer(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    webgl_util.bindColorTextureToFramebuffer(\n        this.gl, texture, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(this.gl);\n    }\n  }\n\n  private unbindTextureToFrameBuffer() {\n    if (this.outputTexture != null) {\n      webgl_util.bindColorTextureToFramebuffer(\n          this.gl, this.outputTexture, this.framebuffer);\n      if (this.debug) {\n        webgl_util.validateFramebuffer(this.gl);\n      }\n    } else {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n    }\n  }\n\n  private downloadMatrixDriver(\n      texture: WebGLTexture,\n      downloadAndDecode: () => Float32Array): Float32Array {\n    this.bindTextureToFrameBuffer(texture);\n    const result = downloadAndDecode();\n    this.unbindTextureToFrameBuffer();\n\n    return result;\n  }\n\n  private setOutputMatrixTextureDriver(\n      outputMatrixTextureMaybePacked: WebGLTexture, width: number,\n      height: number) {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    webgl_util.bindColorTextureToFramebuffer(\n        gl, outputMatrixTextureMaybePacked, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(gl);\n    }\n    this.outputTexture = outputMatrixTextureMaybePacked;\n    webgl_util.callAndCheck(gl, () => gl.viewport(0, 0, width, height));\n    webgl_util.callAndCheck(gl, () => gl.scissor(0, 0, width, height));\n  }\n\n  private setOutputMatrixWriteRegionDriver(\n      x: number, y: number, width: number, height: number) {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(\n        this.gl, () => this.gl.scissor(x, y, width, height));\n  }\n\n  private throwIfDisposed() {\n    if (this.disposed) {\n      throw new Error('Attempted to use disposed GPGPUContext.');\n    }\n  }\n\n  private throwIfNoProgram() {\n    if (this.program == null) {\n      throw new Error('No GPU program is currently set.');\n    }\n  }\n}\n\ntype PollItem = {\n  isDoneFn: () => boolean,\n  resolveFn: () => void\n};\n\n/**\n * Finds the index of the last true element using linear search.\n * Note: We can't do binary search because Chrome expects us to explicitly\n * test all fences before download:\n * https://github.com/tensorflow/tfjs/issues/1145\n */\nexport function linearSearchLastTrue(arr: Array<() => boolean>): number {\n  let i = 0;\n  for (; i < arr.length; ++i) {\n    const isDone = arr[i]();\n    if (!isDone) {\n      break;\n    }\n  }\n  return i - 1;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, Tensor, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport * as shader_compiler from './shader_compiler';\nimport {InputInfo, ShapeInfo} from './shader_compiler';\nimport {PackingScheme, TextureData, TextureUsage} from './tex_util';\n\nexport interface GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  /** If true, this program expects packed input textures. Defaults to false. */\n  packedInputs?: boolean;\n  /** If true, this program produces a packed texture. Defaults to false. */\n  packedOutput?: boolean;\n  /**\n   * Affects what type of texture we allocate for the output. Defaults to\n   * `TextureUsage.RENDER`.\n   */\n  outTexUsage?: TextureUsage;\n  /**\n   * The type of scheme to use when packing texels for the output values.\n   * See `PackingScheme` for details. Defaults to `PackingScheme.SHARED_BATCH`.\n   */\n  outPackingScheme?: PackingScheme;\n}\n\nexport interface GPGPUBinary {\n  webGLProgram: WebGLProgram;\n  program: GPGPUProgram;\n  uniformLocations: {[name: string]: WebGLUniformLocation};\n  source: string;\n  inShapeInfos: ShapeInfo[];\n  outShapeInfo: ShapeInfo;\n  infLoc: WebGLUniformLocation;\n  nanLoc: WebGLUniformLocation;\n}\n\nexport interface TensorData {\n  shape: number[];\n  texData: TextureData;\n  isUniform: boolean;\n  // Available when we decide to upload as uniform instead of texture.\n  uniformValues?: TypedArray;\n}\n\nexport function compileProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, program: GPGPUProgram, inputs: TensorData[],\n    output: TensorData): GPGPUBinary {\n  const userCode = program.userCode;\n  const inputInfos: InputInfo[] = inputs.map((input, i) => {\n    const shapeInfo: ShapeInfo = {\n      logicalShape: input.shape,\n      texShape: input.isUniform ? null : input.texData.texShape,\n      isUniform: input.isUniform,\n      isPacked: input.isUniform ? false : input.texData.isPacked,\n      flatOffset: null\n    };\n    if (input.texData != null && input.texData.slice != null &&\n        input.texData.slice.flatOffset > 0) {\n      shapeInfo.flatOffset = input.texData.slice.flatOffset;\n    }\n    return {name: program.variableNames[i], shapeInfo};\n  });\n  const inShapeInfos = inputInfos.map(x => x.shapeInfo);\n  const outShapeInfo: ShapeInfo = {\n    logicalShape: output.shape,\n    texShape: output.texData.texShape,\n    isUniform: false,\n    isPacked: output.texData.isPacked,\n    flatOffset: null\n  };\n  const source = shader_compiler.makeShader(\n      inputInfos, outShapeInfo, userCode, program.packedInputs);\n\n  const webGLProgram = gpgpu.createProgram(source);\n\n  // Add special uniforms (NAN, INFINITY)\n  let infLoc: WebGLUniformLocation = null;\n  const nanLoc = gpgpu.getUniformLocation(webGLProgram, 'NAN', false);\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    infLoc = gpgpu.getUniformLocation(webGLProgram, 'INFINITY', false);\n  }\n\n  // Add user-defined uniforms\n  const uniformLocations: {[name: string]: WebGLUniformLocation} = {};\n  for (let i = 0; i < program.variableNames.length; i++) {\n    const varName = program.variableNames[i];\n    const shouldThrow = false;\n    uniformLocations[varName] =\n        gpgpu.getUniformLocation(webGLProgram, varName, shouldThrow);\n    uniformLocations[`offset${varName}`] =\n        gpgpu.getUniformLocation(webGLProgram, `offset${varName}`, shouldThrow);\n  }\n\n  return {\n    program,\n    source,\n    webGLProgram,\n    uniformLocations,\n    inShapeInfos,\n    outShapeInfo,\n    infLoc,\n    nanLoc,\n  };\n}\n\nfunction validateBinaryAndProgram(\n    shapeInfos: ShapeInfo[], inputs: TensorData[]) {\n  if (shapeInfos.length !== inputs.length) {\n    throw Error(\n        `Binary was compiled with ${shapeInfos.length} inputs, but ` +\n        `was executed with ${inputs.length} inputs`);\n  }\n\n  shapeInfos.forEach((s, i) => {\n    const shapeA = s.logicalShape;\n    const input = inputs[i];\n    const shapeB = input.shape;\n\n    if (!util.arraysEqual(shapeA, shapeB)) {\n      throw Error(\n          `Binary was compiled with different shapes than ` +\n          `the current args. Shapes ${shapeA} and ${shapeB} must match`);\n    }\n    // The input is uploaded as uniform.\n    if (s.isUniform && input.isUniform) {\n      return;\n    }\n\n    const texShapeA = s.texShape;\n    const texShapeB = input.isUniform ? null : input.texData.texShape;\n    if (!util.arraysEqual(texShapeA, texShapeB)) {\n      throw Error(\n          `Binary was compiled with different texture shapes than the` +\n          ` current args. Shape ${texShapeA} and ${texShapeB} must match`);\n    }\n  });\n}\n\nexport function runProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, binary: GPGPUBinary, inputs: TensorData[],\n    output: TensorData,\n    customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) =>\n        void): void {\n  validateBinaryAndProgram(binary.inShapeInfos, inputs);\n  validateBinaryAndProgram([binary.outShapeInfo], [output]);\n\n  const outTex = output.texData.texture;\n  const outTexShape = output.texData.texShape;\n  if (output.texData.isPacked) {\n    gpgpu.setOutputPackedMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  } else {\n    gpgpu.setOutputMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  }\n  gpgpu.setProgram(binary.webGLProgram);\n\n  // Set special uniforms (NAN, INFINITY)\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    if (binary.infLoc !== null) {\n      gpgpu.gl.uniform1f(binary.infLoc, Infinity);\n    }\n  }\n  if (binary.nanLoc !== null) {\n    gpgpu.gl.uniform1f(binary.nanLoc, NaN);\n  }\n\n  // Set user-defined inputs\n  inputs.forEach((input, i) => {\n    const varName = binary.program.variableNames[i];\n    const varLoc = binary.uniformLocations[varName];\n    const varOffsetLoc = binary.uniformLocations[`offset${varName}`];\n\n    if (varLoc == null) {\n      // The compiler inferred that this variable is not used in this shader.\n      return;\n    }\n\n    if (input.isUniform) {\n      // Upload the values of the tensor as uniform.\n      if (util.sizeFromShape(input.shape) < 2) {\n        gpgpu.gl.uniform1f(varLoc, input.uniformValues[0]);\n      } else {\n        let vals = input.uniformValues;\n        if (!(vals instanceof Float32Array)) {\n          vals = new Float32Array(vals);\n        }\n        gpgpu.gl.uniform1fv(varLoc, vals);\n      }\n      return;\n    }\n\n    // If the input was sliced, upload the flat offset index.\n    if (input.texData.slice != null && varOffsetLoc != null) {\n      gpgpu.gl.uniform1i(varOffsetLoc, input.texData.slice.flatOffset);\n    }\n\n    gpgpu.setInputMatrixTexture(input.texData.texture, varLoc, i);\n  });\n\n  if (customSetup != null) {\n    customSetup(gpgpu, binary.webGLProgram);\n  }\n  gpgpu.executeProgram();\n}\n\nexport function makeShaderKey(\n    program: GPGPUProgram, inputs: TensorData[], output: TensorData): string {\n  let keyInputs = '';\n  inputs.concat(output).forEach(x => {\n    const hasOffset = x.texData != null && x.texData.slice != null &&\n        x.texData.slice.flatOffset > 0;\n    const texShape = x.isUniform ? 'uniform' : x.texData.texShape;\n    keyInputs += `${x.shape}_${texShape}_${hasOffset}`;\n  });\n  const keyUserCode = program.userCode;\n  let key = program.constructor.name;\n  // Fast string concat. See https://jsperf.com/string-concatenation/14.\n  key += '_' + keyInputs + '_' + keyUserCode;\n  return key;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Im2ColPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      outputShape: number[], inputShape: number[],\n      convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = outputShape;\n\n    const {\n      filterWidth,\n      inChannels,\n      strideWidth,\n      strideHeight,\n      padInfo,\n      outWidth,\n      dilationWidth,\n      dilationHeight,\n      dataFormat\n    } = convInfo;\n    const {left, top} = padInfo;\n    const itemsPerBlockRow = inChannels * filterWidth;\n    const glsl = getGlslDifferences();\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const rowDim = isChannelsLast ? 0 : 1;\n    const colDim = isChannelsLast ? 1 : 2;\n\n    let unrolled = ``;\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        unrolled += `\n          blockIndex = rc.y + ${col};\n          pos = rc.x + ${row};\n\n          if(blockIndex < ${outputShape[1]} && pos < ${outputShape[0]}) {\n            offsetY = int(blockIndex / (${outWidth})) * ${strideHeight} - ${\n            top};\n            d0 = offsetY + ${dilationHeight} * (pos / ${itemsPerBlockRow});\n\n            if(d0 < ${inputShape[rowDim]} && d0 >= 0) {\n\n              offsetX = int(mod(float(blockIndex), ${outWidth}.) * ${\n            strideWidth}. - ${left}.);\n              d1 = offsetX + ${dilationWidth} * (int(mod(float(pos), ${\n            itemsPerBlockRow}.) / ${inChannels}.));\n\n              if(d1 < ${inputShape[colDim]} && d1 >= 0) {\n\n                ch = int(mod(float(pos), ${inChannels}.));\n\n                if (${isChannelsLast}) {\n                  innerDims = vec2(d1, ch);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(d0, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                } else {\n                  innerDims = vec2(d0, d1);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(ch, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                }\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      void main() {\n        ivec2 rc = getOutputCoords();\n\n        vec4 result = vec4(0);\n\n        int blockIndex, pos, offsetY, d0, offsetX, d1, ch;\n        vec2 innerDims;\n\n        ${unrolled}\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n        int d = coords[3];\n        float x = getX(b, r, c, d);\n        float sum = 0.0;\n        for (int j = -${rad}; j <= ${rad}; j++) {\n          int idx = d + j;\n          if (idx >= 0 && idx <=  ${maxD}) {\n            float z = getX(b, r, c, idx);\n            sum += z * z;\n          }\n        }\n        float val = x * ${powOperator};\n        setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNGradProgram implements GPGPUProgram {\n  variableNames = ['inputImage', 'outputImage', 'dy'];\n  outputShape: number[] = [];\n  userCode: string;\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n  depth: number;\n\n  constructor(\n      inputShape: number[], depthRadius: number, bias: number, alpha: number,\n      beta: number) {\n    this.outputShape = inputShape;\n    this.depth = inputShape[3];\n    this.depthRadius = depthRadius;\n    this.bias = bias;\n    this.alpha = alpha;\n    this.beta = beta;\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n\n        float result = 0.0;\n        for (int d = 0; d < ${this.depth}; ++d) {\n          int depthBegin = int(max(0.0, float(d - ${depthRadius})));\n          int depthEnd = int(min(float(${this.depth}),\n              float(d + ${depthRadius} + 1)));\n\n          const int MIN_DEPTH_BEGIN = 0;\n          const int MAX_DEPTH_END = ${this.depth};\n\n          float norm = 0.0;\n          for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd) {\n              norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);\n            }\n            else {\n              break;\n            }\n          }\n\n          norm = float(${alpha}) * norm + float(${bias});\n\n          for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd){\n              float dyi = -2.0 * float(${alpha})\n                * float(${beta})\n                * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)\n                / norm;\n              if (k == d) {\n                dyi += pow(norm, -1.0 * ${beta});\n              }\n              if (k == coords[3]) {\n                dyi *= getDy(b, r, c, d);\n                result += dyi;\n              }\n            }\n            else {\n              break;\n            }\n          }\n      }\n      setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords.x;\n        int r = coords.y;\n        int c = coords.z;\n        int d = coords.w;\n\n        bool hasNextCol = d < ${this.outputShape[3]};\n        bool hasNextRow = c < ${this.outputShape[2]};\n\n        vec4 sum = vec4(0.);\n        vec4 xFragAtOutputCoords = getX(b, r, c, d);\n\n        vec4 xAtOutputCoords = vec4(\n          getChannel(xFragAtOutputCoords, vec2(c, d)),\n          hasNextCol ?\n            getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,\n          hasNextRow ?\n            getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0\n        );\n\n        int firstChannel = d - ${rad};\n        vec2 cache = vec2(0.);\n        if(firstChannel >= 0){\n          vec4 firstChannelFrag = getX(b, r, c, firstChannel);\n          cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));\n            if(hasNextRow){\n              cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));\n            }\n        }\n\n        ivec2 depth = ivec2(d, d + 1);\n        for (int j = - ${rad}; j <= ${rad}; j++) {\n          ivec2 idx = depth + j;\n          bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));\n          bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${maxD}));\n\n          bool depthInRange = aboveLowerBound.x && belowUpperBound.x;\n          bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;\n\n          if(depthInRange || depthPlusOneInRange){\n            vec4 z = vec4(0.);\n            vec4 xFragAtCurrentDepth;\n            z.xz = cache.xy;\n            if(depthPlusOneInRange && hasNextCol){\n              xFragAtCurrentDepth = idx.y != d ?\n                getX(b, r, c, idx.y) : xFragAtOutputCoords;\n              z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));\n              if(hasNextRow){\n                z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));\n              }\n            }\n            cache.xy = z.yw;\n            sum += z * z;\n          }\n        }\n        vec4 result = xAtOutputCoords * ${powOperator};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MaxPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex = effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n          wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n            int maxPosValue = ${lastIndex} - int(getMaxPos(b, idyR, idyC, d));\n\n            // Get the current value, check it against the value from the\n            // position matrix.\n            int curPosValue = wR * ${effectiveFilterWidth} + wC;\n            float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n            dotProd += dyValue * mask;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class MaxPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex =\n        effectiveFilterDepth * effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n           wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n              int maxPosValue = ${lastIndex} -\n                  int(getMaxPos(batch, idyD, idyR, idyC, ch));\n\n              // Get the current value, check it against the value from the\n              // position matrix.\n              int curPosValue =\n                  wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                  wR * ${effectiveFilterWidth} + wC;\n              float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n              dotProd += dyValue * mask;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MatMulPackedProgram implements GPGPUProgram {\n  variableNames = ['matrixA', 'matrixB'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      aShape: [number, number, number], outputShape: [number, number, number],\n      transposeA = false, transposeB = false, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = outputShape;\n\n    const sharedDim = transposeA ? aShape[1] : aShape[2];\n    const sharedDimensionPacked = Math.ceil(sharedDim / 2);\n\n    const aSample = transposeA ? 'i * 2, rc.y' : 'rc.y, i * 2';\n    const bSample = transposeB ? 'rc.z, i * 2' : 'i * 2, rc.z';\n    const aSwizzle = transposeA ? ['a.xxyy', 'a.zzww'] : ['a.xxzz', 'a.yyww'];\n    const bSwizzle = transposeB ? ['b.xzxz', 'b.ywyw'] : ['b.xyxy', 'b.zwzw'];\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const float sharedDimension = ${sharedDimensionPacked}.0;\n\n      vec4 dot2x2ARowBCol(ivec3 rc) {\n        vec4 result = vec4(0);\n        for (int i = 0; i < ${sharedDimensionPacked}; i++) {\n          vec4 a = getMatrixA(rc.x, ${aSample});\n          vec4 b = getMatrixB(rc.x, ${bSample});\n\n          // These swizzled products need to be separately added.\n          // See: https://github.com/tensorflow/tfjs/issues/1735\n          result += (${aSwizzle[0]} * ${bSwizzle[0]});\n          result += (${aSwizzle[1]} * ${bSwizzle[1]});\n        }\n        return result;\n      }\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n        vec4 result = dot2x2ARowBCol(rc);\n\n        ${addBiasSnippet}\n\n        ${applyActivationSnippet}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MultinomialProgram implements GPGPUProgram {\n  variableNames = ['probs'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(batchSize: number, numOutcomes: number, numSamples: number) {\n    this.outputShape = [batchSize, numSamples];\n\n    this.userCode = `\n      uniform float seed;\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n\n        float r = random(seed);\n        float cdf = 0.0;\n\n        for (int i = 0; i < ${numOutcomes - 1}; i++) {\n          cdf += getProbs(batch, i);\n\n          if (r < cdf) {\n            setOutput(float(i));\n            return;\n          }\n        }\n\n        // If no other event happened, last event happened.\n        setOutput(float(${numOutcomes - 1}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(seed: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.seedLoc == null) {\n        this.seedLoc = gpgpu.getUniformLocation(webGLProgram, 'seed');\n      }\n      gpgpu.gl.uniform1f(this.seedLoc, seed);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class OneHotProgram implements GPGPUProgram {\n  variableNames = ['indices'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(\n      numIndices: number, depth: number, onValue: number, offValue: number) {\n    this.outputShape = [numIndices, depth];\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int index = round(getIndices(coords.x));\n        setOutput(mix(float(${offValue}), float(${onValue}),\n                      float(index == coords.y)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape:\n          number[]) {  // TODO(https://github.com/tensorflow/tfjs/issues/893):\n                       // Only input / output 3D tensors.\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    if (rank === 0) {\n      this.userCode = `\n        void main() {\n          setOutput(vec4(getA(), 0., 0., 0.));\n        }\n      `;\n    } else {\n      const channels = getChannels('rc', rank);\n      const dtype = getCoordsDataType(rank);\n      const outOfBoundsCondition =\n          getOutOfBoundsCondition(rank, outputShape, channels);\n      const setup = getSetup(\n          rank, outputShape[outputShape.length - 1],\n          outputShape[outputShape.length - 2], channels);\n      const output = getOutput(outputShape, channels);\n\n      this.userCode = `\n        void main() {\n          ${dtype} rc = getOutputCoords();\n\n          if(${outOfBoundsCondition}) {\n            setOutput(vec4(0));\n          } else {\n            ${setup}\n\n            setOutput(vec4(${output}));\n          }\n        }\n      `;\n    }\n  }\n}\n\nfunction getSourceCoordsArr(rank: number, dims: string[]): string[] {\n  const coords = [];\n\n  for (let row = 0; row <= 1; row++) {\n    for (let col = 0; col <= 1; col++) {\n      let coord = `${row === 0 ? 'r' : 'rp1'}, ${col === 0 ? 'c' : 'cp1'}`;\n\n      for (let d = 2; d < rank; d++) {\n        coord = `${dims[dims.length - 1 - d]},` + coord;\n      }\n\n      coords.push(coord);\n    }\n  }\n  return coords;\n}\n\nfunction getOutOfBoundsCondition(\n    rank: number, shape: number[], dims: string[]): string {\n  if (rank === 1) {\n    return `rc > ${shape[0]}`;\n  }\n\n  let cond = '';\n  for (let i = rank - 2; i < rank; i++) {\n    cond += `${dims[i]} >= ${shape[i]}`;\n    if (i < rank - 1) {\n      cond += '||';\n    }\n  }\n\n  return cond;\n}\n\nfunction getSetup(\n    rank: number, cols: number, rows: number, dims: string[]): string {\n  if (rank === 1) {\n    return '';\n  }\n\n  const innerDims = dims.slice(-2);\n\n  return `\n    int r = ${innerDims[0]};\n    int c = ${innerDims[1]};\n    int rp1 = r + 1;\n    int cp1 = c + 1;\n\n    bool cEdge = cp1 >= ${cols};\n    bool rEdge = rp1 >= ${rows};\n  `;\n}\n\nfunction getOutput(shape: number[], dims: string[]): string {\n  const rank = shape.length;\n  const sourceCoords = getSourceCoordsArr(rank, dims);\n  if (rank === 1) {\n    return `getA(rc),\n            rc + 1 >= ${shape[0]} ? 0. : getA(rc + 1),\n            0, 0`;\n  }\n\n  return `getA(${sourceCoords[0]}),\n          cEdge ? 0. : getA(${sourceCoords[1]}),\n          rEdge ? 0. : getA(${sourceCoords[2]}),\n          rEdge || cEdge ? 0. : getA(${sourceCoords[3]})`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const type = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const unpackedCoords =\n        ['coords[0]', 'coords[1]', 'coords[2]', 'coords[3]'].slice(0, rank);\n\n    if (rank === 1) {\n      this.userCode = `\n        int start = ${start};\n        int end = ${end};\n\n        void main() {\n          int outC = getOutputCoords();\n          if (outC < start || outC >= end) {\n            setOutput(float(${constantValue}));\n          } else {\n            setOutput(getX(outC - start));\n          }\n        }\n      `;\n      return;\n    }\n    this.userCode = `\n      ${type} start = ${type}(${start});\n      ${type} end = ${type}(${end});\n\n      void main() {\n        ${type} outC = getOutputCoords();\n        if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {\n          setOutput(float(${constantValue}));\n        } else {\n          ${type} coords = outC - start;\n          setOutput(getX(${unpackedCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const dtype = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const coords = getChannels('rc', rank);\n    const source = getChannels('source', rank);\n    const cLimit = `${coords[rank - 1]} < ${this.outputShape[rank - 1]}`;\n    const innerDims =\n        rank === 1 ? 'source' : `vec2(${source.slice(-2).join()})`;\n\n    const componentSetup = [\n      `${dtype} rc = outputLoc;`, `${coords[rank - 1]} += 1;\n       if(${cLimit}) {\n      `,\n      rank === 1 ? '' : `}\n       rc = outputLoc;\n       ${coords[rank - 2]} += 1;\n       if(${coords[rank - 2]} < ${this.outputShape[rank - 2]}) {`,\n      rank === 1 ? '' : `  ${coords[rank - 1]} += 1;\n         if(${cLimit}) {`\n    ];\n\n    const paddingArea = rank === 1 ?\n        'rc < start || rc >= end' :\n        'any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))';\n    let mainLoop = '';\n    for (let i = 0, j = rank === 1 ? 2 : 4; i < j; i++) {\n      mainLoop += `\n        ${componentSetup[i]}\n        if (${paddingArea}) {\n          result[${i}] = float(${constantValue});\n        } else {\n          ${dtype} source = rc - start;\n          result[${i}] = getChannel(getX(${source.join()}), ${innerDims});\n        }\n      `;\n    }\n    mainLoop += (rank === 1 ? `} ` : `}}`);\n\n    this.userCode = `\n      const ${dtype} start = ${dtype}(${start});\n      const ${dtype} end = ${dtype}(${end});\n\n      void main() {\n        ${dtype} outputLoc = getOutputCoords();\n        vec4 result = vec4(0.);\n        ${mainLoop}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Pool2DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n    const batchFlattenPositionStr = `((batch  * ${convInfo.inHeight} + xR) * ${\n        convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n    const flattenPositionStr =\n        `(xR * ${convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n        const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int batch = coords[0];\n          int d = coords[3];\n\n          ivec2 xRCCorner = coords.yz * strides - pads;\n          int xRCorner = xRCCorner.x;\n          int xCCorner = xRCCorner.y;\n\n          // max/min x(?, ?, d) to get y(yR, yC, d).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n          float avgValue = 0.0;\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              int xC = xCCorner + wC;\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float value = getX(batch, xR, xC, d);\n\n              // If a min / max value has already been found, use it. If not,\n              // use the current value.\n              float currMinMaxValue = mix(\n                  value, minMaxValue, minMaxValueFound);\n              if (value ${compareOp} currMinMaxValue) {\n                minMaxValue = value;\n                minMaxValueFound = 1.0;\n                minMaxPosition = ${\n          flattenPositions ? (includeBatchInIndex ? batchFlattenPositionStr :\n                                                    flattenPositionStr) :\n                             `wR * ${effectiveFilterWidth} + wC`};\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xR, int xC, int d) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xR, xC, d);\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d = coords[3];\n\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // max/min x(?, ?, d) to get y(yR, yC, d).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          int xR = xRCorner + wR;\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              getValue(batch, xR, xC + 3 * ${dilationWidth}, d)\n            );\n\n            ${updateSnippet}\n          }\n\n          int xC = xCCorner + ${filterWidthNearestVec4};\n          if (${filterWidthVec4Remainder === 1}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              initializationValue,\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 2}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 3}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              initializationValue\n            );\n\n            ${updateSnippet}\n          }\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n\nexport class Pool3DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv3DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec3 strides =\n            ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n        const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n        void main() {\n          ivec5 coords = getOutputCoords();\n          int batch = coords.x;\n          int ch = coords.u;\n\n          ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n          int xDCorner = xCorner.x;\n          int xRCorner = xCorner.y;\n          int xCCorner = xCorner.z;\n\n          // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n\n          for (int wD = 0; wD < ${effectiveFilterDepth};\n              wD += ${dilationDepth}) {\n            int xD = xDCorner + wD;\n\n            if (xD < 0 || xD >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int wR = 0; wR < ${effectiveFilterHeight};\n                wR += ${dilationHeight}) {\n              int xR = xRCorner + wR;\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int wC = 0; wC < ${effectiveFilterWidth};\n                  wC += ${dilationWidth}) {\n                int xC = xCCorner + wC;\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float value = getX(batch, xD, xR, xC, ch);\n\n                // If a min / max value has already been found, use it. If not,\n                // use the current value.\n                float currMinMaxValue = mix(\n                    value, minMaxValue, minMaxValueFound);\n                if (value ${compareOp} currMinMaxValue) {\n                  minMaxValue = value;\n                  minMaxValueFound = 1.0;\n                  minMaxPosition = ${\n          flattenPositions ?\n              (includeBatchInIndex ?\n                   `(((batch * ${convInfo.inDepth} + xD) * ${\n                       convInfo.inHeight} + xR) * ${convInfo.inWidth} + xC) * ${\n                       convInfo.inChannels} + ch` :\n                   `((xD * ${convInfo.inHeight} + xR) * ${\n                       convInfo.inWidth} + xC) * ${convInfo.inChannels} + ch`) :\n              `wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                      wR * ${effectiveFilterWidth} + wC`};\n                }\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec3 strides =\n        ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xD, int xR, int xC, int ch) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xD, xR, xC, ch);\n      }\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xDCorner = xCorner.x;\n        int xRCorner = xCorner.y;\n        int xCCorner = xCorner.z;\n\n        // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          int xD = xDCorner + wD;\n\n          if (xD < 0 || xD >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 3 * ${dilationWidth}, ch)\n              );\n\n              ${updateSnippet}\n            }\n\n            int xC = xCCorner + ${filterWidthNearestVec4};\n            if (${filterWidthVec4Remainder === 1}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                initializationValue,\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 2}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 3}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                initializationValue\n              );\n\n              ${updateSnippet}\n            }\n          }\n          setOutput(${returnValue});\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ReduceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo,\n      reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod') {\n    const {windowSize, batchSize, inSize, outSize} = reduceInfo;\n    this.outputShape = [batchSize, outSize];\n\n    let initializationValue = '0.0';\n    let compareOp = ``;\n\n    if (reduceType === 'prod') {\n      initializationValue = '1.0';\n    } else if (reduceType === 'min') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '1.0 / 1e-20';\n      compareOp = `min`;\n    } else if (reduceType === 'max') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n      compareOp = `max`;\n    }\n\n    let returnValue = `${reduceType}(${reduceType}(${reduceType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n\n    if (reduceType === 'sum') {\n      returnValue = `sumValue`;\n    } else if (reduceType === 'prod') {\n      returnValue = `prodValue`;\n    } else if (reduceType === 'all') {\n      returnValue = `allValue`;\n    } else if (reduceType === 'any') {\n      returnValue = `anyValue`;\n    }\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    let updateSnippet = `\n      if (${reduceType === 'sum'}) {\n        sumValue += dot(values, ones);\n      } else if (${reduceType === 'prod'}) {\n        vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);\n        prodValue *= tmp[0] * tmp[1];\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    let vecType = `vec4`;\n\n    if (reduceType === 'all') {\n      initializationValue = '1.0';\n      updateSnippet = `\n        bool reducedAllValue = all(values);\n        float floatedReducedAllValue = float(reducedAllValue);\n        allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    } else if (reduceType === 'any') {\n      initializationValue = '0.0';\n      updateSnippet = `\n        bool reducedAnyValue = any(values);\n        float floatedReducedAnyValue = float(reducedAnyValue);\n        anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    }\n\n    let checkOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float getValue(int batch, int inIdx) {\n        ${checkOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        vec4 minMaxValue = vec4(${initializationValue});\n        float prodValue = 1.0;\n        float sumValue = 0.0;\n        float allValue = 1.0;\n        float anyValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class ReshapePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: [number, number, number], inputShape: [\n    number, number, number\n  ]) {\n    this.outputShape = outputShape;\n\n    let mainLoop = ``;\n    for (let i = 0; i < 4; i++) {\n      let thisRC = `thisRC = rc;`;\n      if (i % 2 === 1) {\n        thisRC += `thisRC.z += 1;`;\n      }\n      if (i > 1) {\n        thisRC += `thisRC.y += 1;`;\n      }\n\n      mainLoop += `\n        ${thisRC}\n        ${i > 0 ? `if(thisRC.y < rows && thisRC.z < cols){` : ''}\n          int flatIndex = getFlatIndex(thisRC);\n\n          ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);\n          vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));\n\n          result[${i}] =\n            getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);\n        ${i > 0 ? '}' : ''}\n      `;\n    }\n\n    this.userCode = `\n      ${getReshapedInputCoords(inputShape)}\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n\n        vec4 result = vec4(0.);\n\n        ivec3 thisRC;\n        int rows = ${outputShape[1]};\n        int cols = ${outputShape[2]};\n\n        ${mainLoop}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nfunction getReshapedInputCoords(shape: [number, number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 inputCoordsFromReshapedOutCoords(int index) {\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(startRLerp - float(winHeight / 2));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(startCLerp - float(winWidth / 2));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float dxR = float(dyR) * heightScale;\n            int topDxRIndex = int(floor(dxR));\n            int bottomDxRIndex = int(min(ceil(dxR), ${xHeight - 1}.0));\n            float dxRLerp = dxR - float(topDxRIndex);\n            float inverseDxRLerp = 1.0 - dxRLerp;\n\n            float dxC = float(dyC) * widthScale;\n            int leftDxCIndex = int(floor(dxC));\n            int rightDxCIndex = int(min(ceil(dxC), ${xWidth - 1}.0));\n            float dxCLerp = dxC - float(leftDxCIndex);\n            float inverseDxCLerp = 1.0 - dxCLerp;\n\n            if (r == topDxRIndex && c == leftDxCIndex) {\n              // topLeft\n              accumulator +=\n                getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;\n            }\n\n            if (r == topDxRIndex && c == rightDxCIndex) {\n              // topRight\n              accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == leftDxCIndex) {\n              // bottomLeft\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == rightDxCIndex) {\n              // bottomRight\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec2 sourceFloorRC = ivec2(sourceFracIndexRC);\n        ivec2 sourceCeilRC = ivec2(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);\n        float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);\n        float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);\n        float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);\n\n        vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);\n\n        float top = topLeft + (topRight - topLeft) * fracRC.y;\n        float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;\n        float newValue = top + (bottom - top) * fracRC.x;\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec3 effectiveInputOverOutputRatioRC = vec3(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec3 inputShapeRC = vec3(${oldHeight}.0, ${oldWidth}.0,\n                                     ${oldWidth}.0);\n\n      float getAValue(int b, int r, int c, int d) {\n        return getChannel(getA(b, r, c, d), vec2(c, d));\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        // Calculate values for next column in yRC.z.\n        ivec3 yRC = coords.yzz + ivec3(0, 0, 1);\n\n        // Fractional source index.\n        vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec3 sourceFloorRC = ivec3(sourceFracIndexRC);\n        ivec3 sourceCeilRC = ivec3(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        // Should we calculate next column and row elements in 2x2 packed cell.\n        bool hasNextCol = d < ${depth - 1};\n        bool hasNextRow = coords.z < ${newWidth - 1};\n\n        // In parallel, construct four corners for all four components in\n        // packed 2x2 cell.\n        vec4 topLeft = vec4(\n          getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 bottomLeft = vec4(\n          getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 topRight = vec4(\n          getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec4 bottomRight = vec4(\n          getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);\n\n        vec4 top = mix(topLeft, topRight, fracRC.yyzz);\n        vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);\n        vec4 newValue = mix(top, bottom, fracRC.x);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeigborBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(floor(startRLerp - float(winHeight / 2)));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(floor(startCLerp - float(winWidth / 2)));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float sourceFracRow =\n              float(${effectiveXSize[0]}) *\n                (float(dyR) / float(${effectiveYSize[0]}));\n\n            float sourceFracCol =\n                float(${effectiveXSize[1]}) *\n                  (float(dyC) / float(${effectiveYSize[1]}));\n\n            int sourceNearestRow = int(min(\n                float(int(${xHeight}) - 1),\n                ${alignCorners} ? float(round(sourceFracRow)) :\n                                  float(floor(sourceFracRow))));\n\n            int sourceNearestCol = int(min(\n                float(int(${xWidth}) - 1),\n                ${alignCorners} ? float(round(sourceFracCol)) :\n                                  float(floor(sourceFracCol))));\n\n            if (r == sourceNearestRow && c == sourceNearestCol) {\n              accumulator += getDy(b, dyR, dyC, d);\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeighborProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    // When align corners is false, we rounds the value with floor.\n    const roundBase = alignCorners ? '0.5' : '0.0';\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the coordinators of nearest neighbor point.\n        ivec2 sourceNearestRC = ivec2(\n          min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${roundBase})));\n\n        float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReverseProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n\n    if (rank === 1) {\n      this.userCode = `\n        void main() {\n          int coord = getOutputCoords();\n          setOutput(getX(${xShape[0]} - coord - 1));\n        }\n      `;\n      return;\n    }\n    const getInCoord = (i: number) => {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - coords[${i}] - 1`;\n      }\n      return `coords[${i}]`;\n    };\n    const inCoords = xShape.map((_, i) => getInCoord(i)).join(',');\n    const type = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${type} coords = getOutputCoords();\n        setOutput(getX(${inCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReversePackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n    const channels = getChannels('rc', rank);\n    const nextColumn =\n        `${channels[rank - 1]} + 1 < ${this.outputShape[rank - 1]}`;\n    const nextRow = `${channels[rank - 2]} + 1 < ${this.outputShape[rank - 2]}`;\n    const type = getCoordsDataType(rank);\n    if (rank === 1) {\n      this.userCode = `\n        void main(){\n          int rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = getChannel(getX(${xShape[0]} - rc - 1),\n            ${xShape[0]} - rc - 1);\n          if(${nextColumn}){\n              result.g = getChannel(getX(${xShape[0]} - (rc  + 1) - 1),\n                ${xShape[0]} - (rc  + 1) - 1);\n          }\n          setOutput(result);\n        }\n      `;\n    } else {\n      this.userCode = `\n        void main() {\n          ${type} rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = ${getR(channels.slice())};\n          if(${nextColumn}){\n            result.g = ${getG(channels.slice())};\n          }\n          if(${nextRow}) {\n            result.b = ${getB(channels.slice())};\n            if(${nextColumn}) {\n              result.a = ${getA(channels.slice())};\n            }\n          }\n          setOutput(result);\n        }\n    `;\n    }\n\n    function getR(channels: string[]): string {\n      return getChannel(channels);\n    }\n\n    function getG(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getB(channels: string[]): string {\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getA(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getChannel(channels: string[]): string {\n      const inCoordsArray = xShape.map((_, i) => getInCoord(i, channels));\n      const inCoords = inCoordsArray.join(',');\n      const innerDims = inCoordsArray.slice(-2).join(',');\n      return `getChannel(getX(${inCoords}), vec2(${innerDims}))`;\n    }\n\n    function getInCoord(i: number, channels1: string[]): string {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - ${channels1[i]} - 1`;\n      } else {\n        return `${channels1[i]}`;\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ScatterProgram implements GPGPUProgram {\n  variableNames = ['updates', 'indices', 'defaultValue'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      updateSize: number, sliceDim: number, indicesRank: number,\n      updatesRank: number, strides: number[], shape: number[],\n      summingDupeIndex = true) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    let indicesString = '';\n    if (indicesRank === 1) {\n      indicesString = 'i';\n    } else if (indicesRank === 2) {\n      indicesString = 'i, j';\n    }\n    const indicesSnippet = `getIndices(${indicesString})`;\n\n    let updatesString = '';\n    if (updatesRank === 1) {\n      updatesString = 'i';\n    } else if (updatesRank === 2) {\n      updatesString = 'i, coords[1]';\n    }\n    const updatesSnippet = `getUpdates(${updatesString})`;\n\n    const strideString = sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${strides});\n\n        void main() {\n          ${dtype} coords = getOutputCoords();\n          float sum = 0.0;\n          bool found = false;\n          for (int i = 0; i < ${updateSize}; i++) {\n            int flattenedIndex = 0;\n            for (int j = 0; j < ${sliceDim}; j++) {\n              int index = round(${indicesSnippet});\n              flattenedIndex += index * ${strideString};\n            }\n            if (flattenedIndex == coords[0]) {\n              sum += ${updatesSnippet};\n              found = true;\n            }\n          }\n          setOutput(mix(getDefaultValue(), sum, float(found)));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class SegmentOpProgram implements GPGPUProgram {\n  variableNames = ['x', 'segmentIds'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      segOpInfo: backend_util.segment_util.SegOpInfo,\n      segOpType: 'unsortedSegmentSum') {\n    const windowSize = segOpInfo.windowSize;\n    const batchSize = segOpInfo.batchSize;\n    const inSize = segOpInfo.inSize;\n    const numSegments = segOpInfo.numSegments;\n    const outSize = numSegments * Math.ceil(inSize / windowSize);\n    this.outputShape = [batchSize, outSize];\n\n    const initializationValue = '0.0';\n    const returnValue = `sumValue`;\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    const updateSnippet = `\n        sumValue += dot(values, segFilter);\n    `;\n\n    let checkValueOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkValueOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n\n    let checkSegmentIdOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkSegmentIdOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return -1.0;\n        }\n      `;\n    }\n\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n\n      float getValue(int batch, int inIdx) {\n        ${checkValueOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      float getSegmentIdAtIndex(int inIdx) {\n        ${checkSegmentIdOutOfBounds}\n        return getSegmentIds(inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = int(floor(float(outIdx) / float(\n          ${numSegments})) * float(${windowSize}));\n        int currentSeg = int(mod(float(outIdx), float(${numSegments})));\n\n        float sumValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          int inIdxSeg = int(getSegmentIdAtIndex(inIdx));\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            0,\n            0,\n            0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n              0,\n              0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            0\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SelectProgram implements GPGPUProgram {\n  variableNames = ['c', 'a', 'b'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(cRank: number, shape: number[], rank: number) {\n    this.outputShape = shape;\n\n    let cCoords;\n    let abCoords;\n    if (rank > 4) {\n      throw Error(`Where for rank ${rank} is not yet supported`);\n    }\n\n    if (rank === 1) {\n      abCoords = `resRC`;\n      cCoords = `resRC`;\n    } else {\n      const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n      const cCoordVars = [];\n      const abCoordVars = [];\n      for (let i = 0; i < shape.length; i++) {\n        abCoordVars.push(`${currentCoords[i]}`);\n        if (i < cRank) {\n          cCoordVars.push(`${currentCoords[i]}`);\n        }\n      }\n      cCoords = cCoordVars.join();\n      abCoords = abCoordVars.join();\n    }\n\n    const dtype = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        float cVal = getC(${cCoords});\n        if (cVal >= 1.0) {\n          setOutput(getA(${abCoords}));\n        } else {\n          setOutput(getB(${abCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SliceProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const uniformPart = `uniform int start[${this.rank}];`;\n    const sourceCoords = getCoords(this.rank);\n\n    let body: string;\n    const coordSum = destSize.map((_, i) => {\n      return `sourceLoc.${coords[i]} = start[${i}] + coords.${coords[i]};`;\n    });\n    body = `\n        ${dtype} sourceLoc;\n        ${dtype} coords = getOutputCoords();\n        ${coordSum.join('\\n')}\n      `;\n    this.userCode = `\n      ${uniformPart}\n      void main() {\n        ${body}\n        setOutput(getSource(${sourceCoords}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n\nconst coords = ['x', 'y', 'z', 'w', 'u', 'v'];\n\nfunction getCoords(rank: number): string {\n  if (rank === 1) {\n    return 'sourceLoc';\n  } else if (rank <= 6) {\n    return coords.slice(0, rank).map(x => 'sourceLoc.' + x).join(',');\n  } else {\n    throw Error(`Slicing for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SlicePackedProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const coords = getChannels('coords', this.rank);\n    const sourceLoc = getChannels('sourceLoc', this.rank);\n\n    const innerDims =\n        this.rank === 1 ? 'sourceLoc' : `vec2(${sourceLoc.slice(-2).join()})`;\n    const getChannel =\n        `getChannel(getSource(${sourceLoc.join()}), ${innerDims})`;\n    const upperRow = `\n      result.x = ${getChannel};\n      if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n        ++${sourceLoc[this.rank - 1]};\n        result.y = ${getChannel};\n        --${sourceLoc[this.rank - 1]};\n      }\n    `;\n    const lowerRow = this.rank === 1 ? '' : `\n      --${coords[this.rank - 1]};\n      if (++${coords[this.rank - 2]} < ${destSize[this.rank - 2]}) {\n        ++${sourceLoc[this.rank - 2]};\n        result.z = ${getChannel};\n        if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n          ++${sourceLoc[this.rank - 1]};\n          result.w = ${getChannel};\n        }\n      }\n    `;\n\n    const sourceLocSetup = this.rank <= 4 ?\n        `sourceLoc = coords +\n            ${dtype}(${destSize.map((_, i) => `start[${i}]`).join()});` :\n        destSize.map((_, i) => `${sourceLoc[i]} = ${coords[i]} + start[${i}];`)\n            .join('\\n');\n    this.userCode = `\n      uniform int start[${this.rank}];\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        ${dtype} sourceLoc;\n        ${sourceLocSetup}\n        vec4 result = vec4(0.);\n        ${upperRow}\n        ${lowerRow}\n        setOutput(result);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class StridedSliceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(begin: number[], strides: number[], size: number[]) {\n    this.outputShape = size;\n    const rank = size.length;\n    const inputDtype = getCoordsDataType(size.length);\n    const dtype = getCoordsDataType(size.length);\n\n    let newCoords = '';\n    if (rank === 1) {\n      newCoords = 'coords * strides + begin';\n    } else {\n      let outputAxis = 0;\n      newCoords =\n          size.map((_, i) => {\n                outputAxis++;\n                return size.length === 1 ?\n                    `coords * strides[${i}] + begin[${i}]` :\n                    `coords[${outputAxis - 1}] * strides[${i}] + begin[${i}]`;\n              })\n              .join(',');\n    }\n\n    this.userCode = `\n      ${inputDtype} begin = ${inputDtype}(${begin});\n      ${inputDtype} strides = ${inputDtype}(${strides});\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        setOutput(getX(${newCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {getInternalFormatForFloat16MatrixTexture, getInternalFormatForFloat16PackedMatrixTexture, getInternalFormatForFloat32MatrixTexture, getInternalFormatForPackedMatrixTexture, getInternalFormatForUnsignedBytesMatrixTexture} from './gpgpu_util';\nimport {getPackedMatrixTextureShapeWidthHeight, getUnpackedMatrixTextureShapeWidthHeight, PhysicalTextureType, TextureConfig, TextureUsage} from './tex_util';\n\nexport class TextureManager {\n  private numUsedTextures = 0;\n  private numFreeTextures = 0;\n  private _numBytesAllocated = 0;\n  private _numBytesFree = 0;  // How many bytes that have been allocated\n                              // are available for reuse.\n  private freeTextures: {[shape: string]: WebGLTexture[]} = {};\n  private logEnabled = false;\n  private usedTextures: {[shape: string]: WebGLTexture[]} = {};\n\n  constructor(private gpgpu: GPGPUContext) {}\n\n  acquireTexture(\n      shapeRC: [number, number], usage: TextureUsage,\n      isPacked: boolean): WebGLTexture {\n    const physicalTexType = getPhysicalFromLogicalTextureType(usage, isPacked);\n\n    const shapeKey = getKeyFromTextureShape(shapeRC, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n    if (!(shapeKey in this.usedTextures)) {\n      this.usedTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shapeRC, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n\n    if (this.freeTextures[shapeKey].length > 0) {\n      this.numFreeTextures--;\n      this.numUsedTextures++;\n      this._numBytesFree -= texBytes;\n      this.log();\n      const newTexture = this.freeTextures[shapeKey].shift();\n      this.usedTextures[shapeKey].push(newTexture);\n      return newTexture;\n    }\n\n    let newTexture: WebGLTexture;\n    if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT32) {\n      newTexture = this.gpgpu.createPackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16PackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT32) {\n      newTexture =\n          this.gpgpu.createFloat32MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (\n        physicalTexType === PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE) {\n      newTexture =\n          this.gpgpu.createUnsignedBytesMatrixTexture(shapeRC[0], shapeRC[1]);\n    }\n    this.usedTextures[shapeKey].push(newTexture);\n\n    this.numUsedTextures++;\n    this._numBytesAllocated += texBytes;\n    this.log();\n\n    return newTexture;\n  }\n\n  releaseTexture(\n      texture: WebGLTexture, shape: [number, number],\n      logicalTexType: TextureUsage, isPacked: boolean): void {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    const physicalTexType =\n        getPhysicalFromLogicalTextureType(logicalTexType, isPacked);\n    const shapeKey = getKeyFromTextureShape(shape, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shape, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n    const deleteTexThreshold = env().get('WEBGL_DELETE_TEXTURE_THRESHOLD');\n    if (deleteTexThreshold !== -1 &&\n        this._numBytesAllocated > deleteTexThreshold) {\n      this.gpgpu.deleteMatrixTexture(texture);\n      this._numBytesAllocated -= texBytes;\n    } else {\n      this.freeTextures[shapeKey].push(texture);\n      this.numFreeTextures++;\n      this._numBytesFree += texBytes;\n    }\n\n    this.numUsedTextures--;\n\n    const texList = this.usedTextures[shapeKey];\n    const texIndex = texList.indexOf(texture);\n    if (texIndex < 0) {\n      throw new Error(\n          'Cannot release a texture that was never provided by this ' +\n          'texture manager');\n    }\n    texList.splice(texIndex, 1);\n    this.log();\n  }\n\n  private log() {\n    if (!this.logEnabled) {\n      return;\n    }\n    const total = this.numFreeTextures + this.numUsedTextures;\n    console.log(\n        'Free/Used', `${this.numFreeTextures} / ${this.numUsedTextures}`,\n        `(${total})`);\n    const freeRatio = this._numBytesFree / this._numBytesAllocated;\n    console.log(`Bytes allocated: ${this._numBytesAllocated}`);\n    console.log(`Bytes unused: ${this._numBytesFree} (${\n        Math.round(100 * freeRatio)}%)`);\n  }\n\n  get numBytesAllocated(): number {\n    return this._numBytesAllocated;\n  }\n\n  get numBytesFree(): number {\n    return this._numBytesFree;\n  }\n\n  getNumUsedTextures(): number {\n    return this.numUsedTextures;\n  }\n\n  getNumFreeTextures(): number {\n    return this.numFreeTextures;\n  }\n\n  dispose() {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    for (const texShape in this.freeTextures) {\n      this.freeTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    for (const texShape in this.usedTextures) {\n      this.usedTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    this.freeTextures = null;\n    this.usedTextures = null;\n    this.numUsedTextures = 0;\n    this.numFreeTextures = 0;\n    this._numBytesAllocated = 0;\n    this._numBytesFree = 0;\n  }\n}\n\nfunction numBytesForInternalFormat(\n    gl: WebGLRenderingContext, internalFormat: number): number {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n  if (internalFormat === glany.R32F) {\n    return 4;\n  } else if (internalFormat === glany.R16F) {\n    return 2;\n  } else if (internalFormat === glany.RGBA32F) {\n    return 16;\n  } else if (internalFormat === gl.RGBA) {\n    return 16;\n  } else if (internalFormat === glany.RGBA16F) {\n    return 8;\n  }\n  throw new Error(`Unknown internal format ${internalFormat}`);\n}\n\nexport function computeBytes(\n    shape: [number, number], physicalTexType: PhysicalTextureType,\n    gl: WebGLRenderingContext, textureConfig: TextureConfig,\n    isPacked: boolean): number {\n  // It is not possible to infer packed status from the texture type because\n  // depending on the textureConfig, different  texture types may resolve to the\n  // same internal format (e.g. in WebGL1, the internal format for\n  // UNPACKED_FLOAT16 textures is gl.RGBA). Therefore we pass in `isPacked`\n  // explicitly.\n  const internalFormat =\n      internalFormatForPhysicalTexType(physicalTexType, textureConfig);\n\n  let numElements: number;\n  if (isPacked) {\n    const [packedWidth, packedHeight] =\n        getPackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = packedWidth * packedHeight;\n\n  } else {\n    const [width, height] =\n        getUnpackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = width * height;\n  }\n\n  const bytesPerElement = numBytesForInternalFormat(gl, internalFormat);\n  return numElements * bytesPerElement;\n}\n\nfunction internalFormatForPhysicalTexType(\n    physicalTexType: PhysicalTextureType,\n    textureConfig: TextureConfig): number {\n  switch (physicalTexType) {\n    case PhysicalTextureType.PACKED_2X2_FLOAT32:\n      return getInternalFormatForPackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_2X2_FLOAT16:\n      return getInternalFormatForFloat16PackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT32:\n      return getInternalFormatForFloat32MatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT16:\n      return getInternalFormatForFloat16MatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE:\n      return getInternalFormatForUnsignedBytesMatrixTexture(textureConfig);\n    default:\n      throw new Error(`Unknown physical texture type ${physicalTexType}`);\n  }\n}\n\nfunction getPhysicalTextureForRendering(isPacked: boolean):\n    PhysicalTextureType {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED')) {\n    if (isPacked) {\n      return PhysicalTextureType.PACKED_2X2_FLOAT32;\n    }\n    return PhysicalTextureType.UNPACKED_FLOAT32;\n  }\n\n  if (isPacked) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT16;\n  }\n  return PhysicalTextureType.UNPACKED_FLOAT16;\n}\n\nfunction getPhysicalFromLogicalTextureType(\n    logicalTexType: TextureUsage, isPacked: boolean): PhysicalTextureType {\n  if (logicalTexType === TextureUsage.UPLOAD) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT32;\n  } else if (logicalTexType === TextureUsage.RENDER || logicalTexType == null) {\n    return getPhysicalTextureForRendering(isPacked);\n  } else if (\n      logicalTexType === TextureUsage.DOWNLOAD ||\n      logicalTexType === TextureUsage.PIXELS) {\n    return PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE;\n  }\n  throw new Error(`Unknown logical texture type ${logicalTexType}`);\n}\n\nfunction getKeyFromTextureShape(\n    shapeRowsCol: [number, number], physicalTexType: PhysicalTextureType,\n    isPacked: boolean): string {\n  return `${shapeRowsCol[0]}_${shapeRowsCol[1]}_${physicalTexType}_${isPacked}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TileProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], reps: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[i] * reps[i];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[]): string {\n  const rank = aShape.length;\n  if (rank > 5) {\n    throw Error(`Tile for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `imod(resRC, ${aShape[0]})`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    sourceCoords.push(`imod(${currentCoords[i]}, ${aShape[i]})`);\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class UnaryOpProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      float unaryOperation(float x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        float x = getAAtOutCoords();\n        float y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n\nconst CHECK_NAN_SNIPPET = `if (isnan(x)) return x;`;\n\nexport const LINEAR = `return x;`;\n\nexport const ABS = `return abs(x);`;\n\nexport const RELU = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : x;\n`;\n\nexport const RELU6 = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : min(6.0, x);\n`;\n\nexport const ELU = `return (x >= 0.0) ? x : (exp(x) - 1.0);`;\n\nexport const SELU = `\n  // Stable and Attracting Fixed Point (0, 1) for Normalized Weights.\n  // see: https://arxiv.org/abs/1706.02515\n  float scaleAlpha = ${backend_util.SELU_SCALEALPHA};\n  float scale = ${backend_util.SELU_SCALE};\n  return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);\n`;\n\nexport function STEP(alpha = 0.0) {\n  return CHECK_NAN_SNIPPET + `\n    return x > 0.0 ? 1.0 : float(${alpha});\n  `;\n}\n\nexport const NEG = `return -x;`;\n\nexport const CEIL = `return ceil(x);`;\n\nexport const FLOOR = `return floor(x);`;\n\nexport const SIGN = `\n  if (isnan(x)) { return 0.0; }\n  return sign(x);\n`;\n\nexport const IS_NAN = `return float(isnan(x));`;\n\nexport const IS_INF = `return float(isinf(x));`;\n\nexport const IS_FINITE = `return float(!isnan(x) && !isinf(x));`;\n\nexport const ROUND = `\n  // OpenGL ES does not support round function.\n  // The algorithm is based on banker's rounding.\n  float base = floor(x);\n  if ((x - base) < 0.5) {\n    return floor(x);\n  } else if ((x - base) > 0.5) {\n    return ceil(x);\n  } else {\n    if (mod(base, 2.0) == 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n`;\n\nexport const EXP = `return exp(x);`;\n\nexport const EXPM1 = `return exp(x) - 1.0;`;\n\nexport const LOG = `if (x < 0.0) return NAN;\n  return log(x);`;\n\nexport const LOG1P = `return log(1.0 + x);`;\n\nexport const SQRT = `return sqrt(x);`;\n\nexport const RSQRT = `return inversesqrt(x);`;\n\nexport const SIGMOID = `return 1.0 / (1.0 + exp(-1.0 * x));`;\n\n/**\n * mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n *\n * epsilon is the difference between 1.0 and the next representable\n * float. For a single precision 32 bit float this should be 2^-23, see:\n * https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\n *\n * too_large = (x > -threshold) is value above which exp(x) may overflow\n * but softplus(x) == x is within machine epsilon\n *\n * too_small = (x < threshold) is value below which exp(x) may underflow,\n * but softplus(x) == exp(x) is within machine epsilon.\n */\nexport const SOFTPLUS = `\n  float epsilon = 1.1920928955078125e-7;\n  float threshold = log(epsilon) + 2.0;\n\n  bool too_large = x > -threshold;\n  bool too_small = x < threshold;\n\n  float result;\n  float exp_x = exp(x);\n\n  if (too_large){\n    result = x;\n  }\n  else if (too_small){\n    result = exp_x;\n  }\n  else{\n    result = log(exp_x + 1.0);\n  }\n  return result;\n`;\n\nexport const ASIN = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return asin(x);\n`;\n\nexport const ACOS = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return acos(x);\n`;\n\nexport const ATAN = CHECK_NAN_SNIPPET + `\n  return atan(x);\n`;\n\nexport const SINH = `\n  float e2x = exp(x);\n  return (e2x - 1.0 / e2x) / 2.0;\n`;\n\nexport const COSH = `\n  float e2x = exp(-x);\n  return (e2x + 1.0 / e2x) / 2.0;\n`;\n\nexport const TANH = `\n  float e2x = exp(-2.0 * abs(x));\n  return sign(x) * (1.0 - e2x) / (1.0 + e2x);\n`;\n\nexport const ASINH = CHECK_NAN_SNIPPET + `return log(x + sqrt(x * x + 1.0));`;\n\nexport const ACOSH = CHECK_NAN_SNIPPET + `\n  if (x < 1.0) return NAN;\n  return log(x + sqrt(x * x - 1.0));`;\n\nexport const ATANH = CHECK_NAN_SNIPPET + `\n  if ((x < -1.0) || (x > 1.0)) return NAN;\n  return (log(1.0 + x) - log(1.0 - x)) / 2.0;`;\n\nexport const ERF = `\n  // Error function is calculated approximately with elementary function.\n  // See \"Handbook of Mathematical Functions with Formulas,\n  // Graphs, and Mathematical Tables\", Abramowitz and Stegun.\n  float p = ${backend_util.ERF_P};\n  float a1 = ${backend_util.ERF_A1};\n  float a2 = ${backend_util.ERF_A2};\n  float a3 = ${backend_util.ERF_A3};\n  float a4 = ${backend_util.ERF_A4};\n  float a5 = ${backend_util.ERF_A5};\n\n  float sign = sign(x);\n  x = abs(x);\n  float t = 1.0 / (1.0 + p * x);\n  return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));\n`;\n\nexport const RECIPROCAL = `return 1.0 / x;`;\n\nexport const LOGICAL_NOT = `return float(!(x >= 1.0));`;\n\nexport const TO_INT = `return float(int(x));`;\n\nexport const CLONE = 'return x;';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const LINEAR = `return x;`;\n\nexport const LOG = `\n  vec4 result = log(x);\n  vec4 isNaN = vec4(lessThan(x, vec4(0.0)));\n  result.r = isNaN.r == 1.0 ? NAN : result.r;\n  result.g = isNaN.g == 1.0 ? NAN : result.g;\n  result.b = isNaN.b == 1.0 ? NAN : result.b;\n  result.a = isNaN.a == 1.0 ? NAN : result.a;\n\n  return result;\n`;\n\nexport const RELU = `\n  vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const RELU6 = `\n  vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const ELU = `\n  vec4 result;\n\n  result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);\n  result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);\n  result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);\n  result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);\n\n  return result;\n`;\n\nexport class UnaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      vec4 unaryOperation(vec4 x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        vec4 x = getAAtOutCoords();\n        vec4 y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels, getSourceCoords} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class UnpackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = false;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: number[]) {\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    const channels = getChannels('rc', rank);\n    const dtype = getCoordsDataType(rank);\n    const sourceCoords = getSourceCoords(rank, channels);\n    const innerDims = channels.slice(-2);\n    const coords = rank <= 1 ? 'rc' : `vec2(${innerDims.join(',')})`;\n\n    this.userCode = `\n      void main() {\n        ${dtype} rc = getOutputCoords();\n        vec4 packedInput = getA(${sourceCoords});\n\n        setOutput(getChannel(packedInput, ${coords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import webgl flags.\nimport './flags_webgl';\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {complex, DataId, div, engine, env, imag, max, MemoryInfo, range, real, RecursiveArray, reshape, scalar, softmax, tensor, tidy, TimingInfo, transpose} from '@tensorflow/tfjs-core';\nimport {backend_util, buffer, kernel_impls, slice_util, util} from '@tensorflow/tfjs-core';\nimport {DataStorage, DataType, KernelBackend, NumericDataType, Rank, Scalar, ShapeMap, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorInfo, TypedArray, upcastType} from '@tensorflow/tfjs-core';\n\nimport {addImplCPU, ceilImplCPU, expImplCPU, expm1ImplCPU, floorImplCPU, logImplCPU, multiplyImplCPU, rsqrtImplCPU, simpleAbsImplCPU, sliceImplCPU, subImplCPU} from './kernel_utils/shared';\n\nconst {segment_util} = backend_util;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\n\nimport {AddNProgram} from './addn_gpu';\nimport {AddNPackedProgram} from './addn_packed_gpu';\nimport {ArgMinMaxProgram} from './argminmax_gpu';\nimport {ArgMinMaxPackedProgram} from './argminmax_packed_gpu';\nimport {AvgPool3DBackpropProgram} from './avg_pool_backprop_gpu';\nimport * as binaryop_complex_gpu from './binaryop_complex_gpu';\nimport {BinaryOpComplexProgram} from './binaryop_complex_gpu';\nimport * as binaryop_gpu from './binaryop_gpu';\nimport {BinaryOpProgram} from './binaryop_gpu';\nimport * as binaryop_packed_gpu from './binaryop_packed_gpu';\nimport {BinaryOpPackedProgram} from './binaryop_packed_gpu';\nimport {getWebGLContext} from './canvas_util';\nimport {ClipProgram} from './clip_gpu';\nimport {ClipPackedProgram} from './clip_packed_gpu';\nimport {ComplexAbsProgram} from './complex_abs_gpu';\nimport {ConcatProgram} from './concat_gpu';\nimport {ConcatPackedProgram} from './concat_packed_gpu';\nimport {Conv2DDerFilterProgram, Conv2DDerInputProgram, Conv3DDerFilterProgram, Conv3DDerInputProgram} from './conv_backprop_gpu';\nimport {DepthwiseConv2DDerFilterProgram, DepthwiseConv2DDerInputProgram} from './conv_backprop_gpu_depthwise';\nimport {Conv2DProgram, Conv3DProgram} from './conv_gpu';\nimport {DepthwiseConv2DProgram} from './conv_gpu_depthwise';\nimport {DepthwiseConvPacked2DProgram} from './conv_packed_gpu_depthwise';\nimport {CropAndResizeProgram} from './crop_and_resize_gpu';\nimport {CumSumProgram} from './cumsum_gpu';\nimport {DecodeMatrixProgram} from './decode_matrix_gpu';\nimport {DecodeMatrixPackedProgram} from './decode_matrix_packed_gpu';\nimport {DepthToSpaceProgram} from './depth_to_space_gpu';\nimport {DiagProgram} from './diag_gpu';\nimport {EncodeFloatProgram} from './encode_float_gpu';\nimport {EncodeFloatPackedProgram} from './encode_float_packed_gpu';\nimport {EncodeMatrixProgram} from './encode_matrix_gpu';\nimport {EncodeMatrixPackedProgram} from './encode_matrix_packed_gpu';\nimport * as fft_gpu from './fft_gpu';\nimport {FFTProgram} from './fft_gpu';\nimport {FillProgram} from './fill_gpu';\nimport {GatherProgram} from './gather_gpu';\nimport {GatherNDProgram} from './gather_nd_gpu';\nimport {GPGPUContext} from './gpgpu_context';\nimport * as gpgpu_math from './gpgpu_math';\nimport {GPGPUBinary, GPGPUProgram, TensorData} from './gpgpu_math';\nimport {Im2ColPackedProgram} from './im2col_packed_gpu';\nimport {LRNProgram} from './lrn_gpu';\nimport {LRNGradProgram} from './lrn_grad_gpu';\nimport {LRNPackedProgram} from './lrn_packed_gpu';\nimport {MaxPool3DBackpropProgram} from './max_pool_backprop_gpu';\nimport {MatMulPackedProgram} from './mulmat_packed_gpu';\nimport {MultinomialProgram} from './multinomial_gpu';\nimport {OneHotProgram} from './onehot_gpu';\nimport {PackProgram} from './pack_gpu';\nimport {PadProgram} from './pad_gpu';\nimport {PadPackedProgram} from './pad_packed_gpu';\nimport {Pool3DProgram} from './pool_gpu';\nimport {ReduceProgram} from './reduce_gpu';\nimport {ReshapePackedProgram} from './reshape_packed_gpu';\nimport {ResizeBilinearBackpropProgram} from './resize_bilinear_backprop_gpu';\nimport {ResizeBilinearProgram} from './resize_bilinear_gpu';\nimport {ResizeBilinearPackedProgram} from './resize_bilinear_packed_gpu';\nimport {ResizeNearestNeigborBackpropProgram} from './resize_nearest_neighbor_backprop_gpu';\nimport {ResizeNearestNeighborProgram} from './resize_nearest_neighbor_gpu';\nimport {ReverseProgram} from './reverse_gpu';\nimport {ReversePackedProgram} from './reverse_packed_gpu';\nimport {ScatterProgram} from './scatter_gpu';\nimport {SegmentOpProgram} from './segment_gpu';\nimport {SelectProgram} from './select_gpu';\nimport {SliceProgram} from './slice_gpu';\nimport {SlicePackedProgram} from './slice_packed_gpu';\nimport {StridedSliceProgram} from './strided_slice_gpu';\nimport * as tex_util from './tex_util';\nimport {TextureData, TextureUsage} from './tex_util';\nimport {TextureManager} from './texture_manager';\nimport {TileProgram} from './tile_gpu';\nimport * as unary_op from './unaryop_gpu';\nimport {UnaryOpProgram} from './unaryop_gpu';\nimport * as unary_packed_op from './unaryop_packed_gpu';\nimport {UnaryOpPackedProgram} from './unaryop_packed_gpu';\nimport {UnpackProgram} from './unpack_gpu';\nimport * as webgl_util from './webgl_util';\nimport {BackendValues} from '@tensorflow/tfjs-core';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\ntype KernelInfo = {\n  name: string; query: Promise<number>;\n};\n\nexport type TimerNode = RecursiveArray<KernelInfo>|KernelInfo;\nexport interface CPUTimerQuery {\n  startMs: number;\n  endMs?: number;\n}\n\nexport interface WebGLMemoryInfo extends MemoryInfo {\n  numBytesInGPU: number;\n  // Tracks the total number of bytes allocated on the GPU, accounting for the\n  // physical texture type.\n  numBytesInGPUAllocated: number;\n  // Tracks byte size of textures that were created and then made available for\n  // reuse (disposed).\n  numBytesInGPUFree: number;\n  unreliable: boolean;\n}\n\nexport interface WebGLTimingInfo extends TimingInfo {\n  uploadWaitMs: number;\n  downloadWaitMs: number;\n}\n\nconst binaryCaches: {[webGLVersion: string]: {[key: string]: GPGPUBinary}} = {};\n\nexport function getBinaryCache(webGLVersion: number) {\n  if (webGLVersion in binaryCaches) {\n    return binaryCaches[webGLVersion];\n  }\n  binaryCaches[webGLVersion] = {};\n  return binaryCaches[webGLVersion];\n}\n\nfunction mapActivationToShaderProgram(\n    activation: backend_util.Activation, packed = false): string {\n  if (activation === 'linear') {\n    if (packed) {\n      return unary_packed_op.LINEAR;\n    }\n    return unary_op.LINEAR;\n  } else if (activation === 'relu') {\n    if (packed) {\n      return unary_packed_op.RELU;\n    }\n    return unary_op.RELU;\n  } else if (activation === 'elu') {\n    if (packed) {\n      return unary_packed_op.ELU;\n    }\n    return unary_op.ELU;\n  } else if (activation === 'relu6') {\n    if (packed) {\n      return unary_packed_op.RELU6;\n    }\n    return unary_op.RELU6;\n  } else if (activation === 'prelu') {\n    if (packed) {\n      return binaryop_packed_gpu.PRELU;\n    }\n    return binaryop_gpu.PRELU;\n  }\n  throw new Error(`Activation ${\n      activation} has not been implemented for the WebGL backend.`);\n}\n\n// Empirically determined constant used to determine size threshold for handing\n// off execution to the CPU.\nconst CPU_HANDOFF_SIZE_THRESHOLD = 128;\n\n// Empirically determined constant used to decide the number of MB on GPU\n// before we warn about high memory use. The MB are this constant * screen area\n// * dpi / 1024 / 1024.\nconst BEFORE_PAGING_CONSTANT = 600;\nfunction numMBBeforeWarning(): number {\n  if (env().global.screen == null) {\n    return 1024;  // 1 GB.\n  }\n  return (env().global.screen.height * env().global.screen.width *\n          window.devicePixelRatio) *\n      BEFORE_PAGING_CONSTANT / 1024 / 1024;\n}\n\n// Empirically determined minimal shared dimension in matmul before we forward\n// to a.mul(b).sum() in order to take advantage of GPU parallelism. See\n// https://github.com/tensorflow/tfjs-core/pull/1379 for benchmarks.\nexport const MATMUL_SHARED_DIM_THRESHOLD = 1000;\n\nexport class MathBackendWebGL extends KernelBackend {\n  texData: DataStorage<TextureData>;\n  gpgpu: GPGPUContext;\n\n  // Maps data ids that have a pending read operation, to list of subscribers.\n  private pendingRead = new WeakMap<DataId, Array<(arr: TypedArray) => void>>();\n  // List of data ids that are scheduled for disposal, but are waiting on a\n  // pending read operation.\n  private pendingDisposal = new WeakSet<DataId>();\n  // Used to count the number of 'shallow' sliced tensors that point to the\n  // same data id.\n  private dataRefCount = new WeakMap<DataId, number>();\n  private numBytesInGPU = 0;\n\n  private canvas: HTMLCanvasElement|OffscreenCanvas;\n\n  private programTimersStack: TimerNode[];\n  private activeTimers: TimerNode[];\n  // Accumulated time spent (including blocking) in uploading data to webgl.\n  private uploadWaitMs = 0;\n  // Accumulated time spent (including blocking in downloading data from webgl.\n  private downloadWaitMs = 0;\n  private cpuBackend: KernelBackend;\n\n  // Number of bits of precision of this backend.\n  private floatPrecisionValue: 32|16;\n\n  private textureManager: TextureManager;\n  private binaryCache: {[key: string]: GPGPUBinary};\n  private gpgpuCreatedLocally: boolean;\n  private numMBBeforeWarning: number;\n  private warnedAboutMemory = false;\n  private warnedAboutCPUBackend = false;\n\n  constructor(gpgpu?: GPGPUContext) {\n    super();\n    if (!env().getBool('HAS_WEBGL')) {\n      throw new Error('WebGL is not supported on this device');\n    }\n\n    if (gpgpu == null) {\n      const gl = getWebGLContext(env().getNumber('WEBGL_VERSION'));\n      this.binaryCache = getBinaryCache(env().getNumber('WEBGL_VERSION'));\n      this.gpgpu = new GPGPUContext(gl);\n      this.canvas = gl.canvas;\n      this.gpgpuCreatedLocally = true;\n    } else {\n      this.gpgpu = gpgpu;\n      this.binaryCache = {};\n      this.gpgpuCreatedLocally = false;\n      this.canvas = gpgpu.gl.canvas;\n    }\n    this.textureManager = new TextureManager(this.gpgpu);\n    this.numMBBeforeWarning = numMBBeforeWarning();\n\n    this.texData = new DataStorage(this, engine());\n  }\n\n  numDataIds() {\n    return this.texData.numDataIds() +\n        (this.cpuBackend ? this.cpuBackend.numDataIds() : 0) -\n        this.pendingDeletes;\n  }\n\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    if (env().getBool('WEBGL_CHECK_NUMERICAL_PROBLEMS') ||\n        env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64' && values != null) {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    const dataId = {};\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n    return dataId;\n  }\n\n  /** Increase refCount of a `TextureData`. */\n  incRef(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    texData.refCount++;\n  }\n\n  /** Decrease refCount of a `TextureData`. */\n  decRef(dataId: DataId): void {\n    if (this.texData.has(dataId)) {\n      const texData = this.texData.get(dataId);\n      texData.refCount--;\n    }\n  }\n\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    if (env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.texData.has(dataId)) {\n      const textureData = this.texData.get(dataId);\n\n      textureData.refCount--;\n\n      if (textureData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    const texData = this.texData.get(dataId);\n    const {values, dtype, complexTensors, slice, shape, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.readSync(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n    if (dtype === 'string') {\n      return values;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let result: Float32Array;\n    if (dtype === 'complex64') {\n      const realValues = complexTensors.real.dataSync() as Float32Array;\n      const imagValues = complexTensors.imag.dataSync() as Float32Array;\n      result = backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    } else {\n      result = this.getValuesFromTexture(dataId);\n    }\n\n    if (shouldTimeProgram) {\n      this.downloadWaitMs += util.now() - start;\n    }\n    return this.convertAndCacheOnCPU(dataId, result);\n  }\n\n  async read(dataId: DataId): Promise<BackendValues> {\n    if (this.pendingRead.has(dataId)) {\n      const subscribers = this.pendingRead.get(dataId);\n      return new Promise<TypedArray>(resolve => subscribers.push(resolve));\n    }\n    const texData = this.texData.get(dataId);\n    const {values, shape, slice, dtype, complexTensors, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.read(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n\n    if (!env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED') &&\n        env().getNumber('WEBGL_VERSION') === 2) {\n      throw new Error(\n          `tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and ` +\n          `WEBGL_VERSION=2 not yet supported.`);\n    }\n\n    let buffer = null;\n    let tmpDownloadTarget: TensorInfo;\n\n    if (dtype !== 'complex64' && env().get('WEBGL_BUFFER_SUPPORTED')) {\n      // Possibly copy the texture into a buffer before inserting a fence.\n      tmpDownloadTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpDownloadTarget.dataId);\n\n      buffer = this.gpgpu.createBufferFromTexture(\n          tmpData.texture, ...tex_util.getDenseTexShape(shape));\n    }\n\n    this.pendingRead.set(dataId, []);\n\n    if (dtype !== 'complex64') {\n      // Create a fence and wait for it to resolve.\n      await this.gpgpu.createAndWaitForFence();\n    }\n\n    // Download the values from the GPU.\n    let vals: Float32Array;\n    if (dtype === 'complex64') {\n      const ps = await Promise.all(\n          [complexTensors.real.data(), complexTensors.imag.data()]);\n      const realValues = ps[0];\n      const imagValues = ps[1];\n      vals = backend_util.mergeRealAndImagArrays(\n          realValues as Float32Array, imagValues as Float32Array);\n    } else if (buffer == null) {\n      vals = this.getValuesFromTexture(dataId);\n    } else {\n      const size = util.sizeFromShape(shape);\n      vals = this.gpgpu.downloadFloat32MatrixFromBuffer(buffer, size);\n    }\n    if (tmpDownloadTarget != null) {\n      this.disposeIntermediateTensorInfo(tmpDownloadTarget);\n    }\n    const dTypeVals = this.convertAndCacheOnCPU(dataId, vals);\n\n    const subscribers = this.pendingRead.get(dataId);\n    this.pendingRead.delete(dataId);\n\n    // Notify all pending reads.\n    subscribers.forEach(resolve => resolve(dTypeVals));\n    if (this.pendingDisposal.has(dataId)) {\n      this.pendingDisposal.delete(dataId);\n      this.disposeData(dataId);\n      this.pendingDeletes--;\n    }\n    return dTypeVals;\n  }\n\n  private checkNumericalProblems(values: BackendValues): void {\n    if (values == null) {\n      return;\n    }\n    for (let i = 0; i < values.length; i++) {\n      const num = values[i] as number;\n      if (!webgl_util.canBeRepresented(num)) {\n        if (env().getBool('WEBGL_RENDER_FLOAT32_CAPABLE')) {\n          throw Error(\n              `The value ${num} cannot be represented with your ` +\n              `current settings. Consider enabling float32 rendering: ` +\n              `'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`);\n        }\n        throw Error(`The value ${num} cannot be represented on this device.`);\n      }\n    }\n  }\n\n  private getValuesFromTexture(dataId: DataId): Float32Array {\n    const {shape, dtype, isPacked} = this.texData.get(dataId);\n    const size = util.sizeFromShape(shape);\n    if (env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED')) {\n      const tmpTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpTarget.dataId);\n      const vals = this.gpgpu\n                       .downloadMatrixFromPackedTexture(\n                           tmpData.texture, ...tex_util.getDenseTexShape(shape))\n                       .subarray(0, size);\n\n      this.disposeIntermediateTensorInfo(tmpTarget);\n\n      return vals;\n    }\n\n    const shouldUsePackedProgram =\n        env().getBool('WEBGL_PACK') && isPacked === true;\n    const outputShape =\n        shouldUsePackedProgram ? webgl_util.getShapeAs3D(shape) : shape;\n    const program = shouldUsePackedProgram ?\n        new EncodeFloatPackedProgram(outputShape as [number, number, number]) :\n        new EncodeFloatProgram(outputShape);\n    const output = this.runWebGLProgram(\n        program, [{shape: outputShape, dtype, dataId}], 'float32');\n    const tmpData = this.texData.get(output.dataId);\n    const vals =\n        this.gpgpu\n            .downloadByteEncodedFloatMatrixFromOutputTexture(\n                tmpData.texture, tmpData.texShape[0], tmpData.texShape[1])\n            .subarray(0, size);\n    this.disposeIntermediateTensorInfo(output);\n\n    return vals;\n  }\n\n  async time(f: () => void): Promise<WebGLTimingInfo> {\n    const oldActiveTimers = this.activeTimers;\n    const newActiveTimers: TimerNode[] = [];\n\n    let outerMostTime = false;\n    if (this.programTimersStack == null) {\n      this.programTimersStack = newActiveTimers;\n      outerMostTime = true;\n    } else {\n      this.activeTimers.push(newActiveTimers);\n    }\n    this.activeTimers = newActiveTimers;\n\n    f();\n\n    // needing to split these up because util.flatten only accepts certain types\n    const flattenedActiveTimerQueries =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.query))\n            .filter(d => d != null);\n    const flattenedActiveTimerNames =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.name))\n            .filter(d => d != null);\n\n    this.activeTimers = oldActiveTimers;\n\n    if (outerMostTime) {\n      this.programTimersStack = null;\n    }\n\n    const res: WebGLTimingInfo = {\n      uploadWaitMs: this.uploadWaitMs,\n      downloadWaitMs: this.downloadWaitMs,\n      kernelMs: null,\n      wallMs: null  // will be filled by the engine\n    };\n\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      const kernelMs = await Promise.all(flattenedActiveTimerQueries);\n\n      res['kernelMs'] = util.sum(kernelMs);\n      res['getExtraProfileInfo'] = () =>\n          kernelMs.map((d, i) => ({name: flattenedActiveTimerNames[i], ms: d}))\n              .map(d => `${d.name}: ${d.ms}`)\n              .join(', ');\n    } else {\n      res['kernelMs'] = {\n        error: 'WebGL query timers are not supported in this environment.'\n      };\n    }\n\n    this.uploadWaitMs = 0;\n    this.downloadWaitMs = 0;\n    return res;\n  }\n  memory(): WebGLMemoryInfo {\n    return {\n      unreliable: false,\n      numBytesInGPU: this.numBytesInGPU,\n      numBytesInGPUAllocated: this.textureManager.numBytesAllocated,\n      numBytesInGPUFree: this.textureManager.numBytesFree\n    } as WebGLMemoryInfo;\n  }\n\n  private startTimer(): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.beginQuery();\n    }\n    return {startMs: util.now(), endMs: null};\n  }\n\n  private endTimer(query: WebGLQuery|CPUTimerQuery): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      this.gpgpu.endQuery();\n      return query;\n    }\n    (query as CPUTimerQuery).endMs = util.now();\n    return query;\n  }\n\n  private async getQueryTime(query: WebGLQuery|CPUTimerQuery): Promise<number> {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.waitForQueryAndGetTime(query as WebGLQuery);\n    }\n    const timerQuery = query as CPUTimerQuery;\n    return timerQuery.endMs - timerQuery.startMs;\n  }\n\n  private pendingDeletes = 0;\n\n  disposeData(dataId: DataId): void {\n    if (this.pendingDisposal.has(dataId)) {\n      return;\n    }\n    if (this.pendingRead.has(dataId)) {\n      this.pendingDisposal.add(dataId);\n      this.pendingDeletes++;\n      return;\n    }\n    // No-op if already disposed.\n    if (!this.texData.has(dataId)) {\n      return;\n    }\n\n    this.releaseGPUData(dataId);\n    const {complexTensors} = this.texData.get(dataId);\n    if (complexTensors != null) {\n      complexTensors.real.dispose();\n      complexTensors.imag.dispose();\n    }\n    this.texData.delete(dataId);\n  }\n\n  private releaseGPUData(dataId: DataId): void {\n    const {texture, dtype, texShape, usage, isPacked, slice} =\n        this.texData.get(dataId);\n    const key = slice && slice.origDataId || dataId;\n    const refCount = this.dataRefCount.get(key);\n    if (refCount > 1) {\n      this.dataRefCount.set(key, refCount - 1);\n    } else {\n      this.dataRefCount.delete(key);\n      if (texture != null) {\n        this.numBytesInGPU -= this.computeBytes(texShape, dtype);\n        this.textureManager.releaseTexture(texture, texShape, usage, isPacked);\n      }\n    }\n    const texData = this.texData.get(dataId);\n    texData.texture = null;\n    texData.texShape = null;\n    texData.isPacked = false;\n    texData.slice = null;\n  }\n\n  getTexture(dataId: DataId): WebGLTexture {\n    this.uploadToGPU(dataId);\n    return this.texData.get(dataId).texture;\n  }\n\n  /**\n   * Returns internal information for the specific data bucket. Used in unit\n   * tests.\n   */\n  getDataInfo(dataId: DataId): TextureData {\n    return this.texData.get(dataId);\n  }\n\n  private getCPUBackend(): KernelBackend|null {\n    if (!env().getBool('WEBGL_CPU_FORWARD')) {\n      return null;\n    }\n\n    if (this.cpuBackend == null) {\n      this.cpuBackend = engine().findBackend('cpu');\n    }\n\n    return this.cpuBackend;\n  }\n\n  /*\n  Tests whether all the inputs to an op are small and on the CPU. This heuristic\n  determines when it would be faster to execute a kernel on the CPU. WebGL\n  kernels opt into running this check and forwarding when appropriate.\n  TODO(https://github.com/tensorflow/tfjs/issues/872): Develop a more\n  sustainable strategy for optimizing backend execution of ops.\n   */\n  shouldExecuteOnCPU(\n      inputs: TensorInfo[],\n      sizeThreshold = CPU_HANDOFF_SIZE_THRESHOLD): boolean {\n    const cpuBackend = this.getCPUBackend();\n    if (!this.warnedAboutCPUBackend && cpuBackend == null) {\n      console.warn(\n          'Your application contains ops that are small enough to be ' +\n          'executed on the CPU backend, however the CPU backend cannot ' +\n          'be found. Consider importing the CPU backend ' +\n          '(@tensorflow/tfjs-backend-cpu) for better performance.');\n\n      this.warnedAboutCPUBackend = true;\n    }\n\n    return cpuBackend != null &&\n        inputs.every(\n            input => this.texData.get(input.dataId).texture == null &&\n                util.sizeFromShape(input.shape) < sizeThreshold);\n  }\n\n  getGPGPUContext(): GPGPUContext {\n    return this.gpgpu;\n  }\n\n  complex<T extends Tensor>(real: T, imag: T): T {\n    const result = this.makeOutput(real.shape, 'complex64');\n    const resultData = this.texData.get(result.dataId);\n    // The backend owns the reference to the underlying real and imaginary\n    // clones. These will explicitly get disposed when the complex tensor is\n    // disposed.\n    resultData.complexTensors = {\n      real: engine().keep(real.clone()),\n      imag: engine().keep(imag.clone())\n    };\n\n    return result as T;\n  }\n  real<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.real.clone() as T;\n  }\n  imag<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.imag.clone() as T;\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = sliceImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, begin, size, x.shape,\n          x.dtype);\n      return this.makeOutput(size, x.dtype, outValues);\n    }\n    // Short-circuit computation if the slice is zero-sized.\n    if (util.sizeFromShape(size) === 0) {\n      return tensor([], size, x.dtype) as T;\n    }\n    const {isPacked} = this.texData.get(x.dataId);\n    const isContinous = slice_util.isSliceContinous(x.shape, begin, size);\n    if (isPacked || !isContinous) {\n      const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n          new SlicePackedProgram(size) :\n          new SliceProgram(size);\n      const customSetup = program.getCustomSetupFunc(begin);\n      return this.compileAndRun(program, [x], null, customSetup);\n    }\n    this.uploadToGPU(x.dataId);\n    return this.shallowSlice(x, begin, size) as T;\n  }\n\n  private shallowSlice(x: Tensor, begin: number[], size: number[]): Tensor {\n    const xTexData = this.texData.get(x.dataId);\n    const t = this.makeOutput(size, x.dtype);\n    const newTexData = this.texData.get(t.dataId);\n    // Copy texture data from the original tensor.\n    Object.assign(newTexData, xTexData);\n    newTexData.shape = size;\n    newTexData.dtype = x.dtype;\n    let flatOffset = slice_util.computeFlatOffset(begin, x.strides);\n    if (xTexData.slice) {\n      // We are slicing an already sliced tensor, so we have to accumulate\n      // the offset.\n      flatOffset += xTexData.slice.flatOffset;\n    }\n    newTexData.slice = {\n      flatOffset,\n      // Point to the original dataId, which is used to do ref counting.\n      origDataId: xTexData.slice && xTexData.slice.origDataId || x.dataId\n    };\n\n    // Increase the ref count for that data bucket.\n    const refCount = this.dataRefCount.get(newTexData.slice.origDataId) || 1;\n    this.dataRefCount.set(newTexData.slice.origDataId, refCount + 1);\n\n    return t;\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x], () => this.cpuBackend.stridedSlice(x, begin, end, strides));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tensor([], outShape) as T;\n    }\n\n    const program = new StridedSliceProgram(begin, strides, outShape);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new ReversePackedProgram(x.shape, axis) :\n        new ReverseProgram(x.shape, axis);\n    return this.compileAndRun(program, [x]);\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    if (tensors[0].dtype === 'complex64') {\n      const reals = tensors.map((t) => real(t));\n      const imags = tensors.map((t) => imag(t));\n      return complex(this.concat(reals, axis), this.concat(imags, axis));\n    }\n\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n    if (tensors.length > env().getNumber('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.concat(tensors.slice(0, midIndex), axis);\n      const rightSide = this.concat(tensors.slice(midIndex), axis);\n      return this.concat([leftSide, rightSide], axis);\n    }\n    if (env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') && tensors[0].rank > 1) {\n      const program = new ConcatPackedProgram(tensors.map(t => t.shape), axis);\n      return this.compileAndRun(program, tensors);\n    }\n    // Any concat of n-dimensional tensors across any axis can be reduced to\n    // a concatenation of two-dimensional tensors across the axis 1 by first\n    // partitioning the axes of the original tensors into those less than the\n    // axis to be concatenated and the rest. Then reshape the tensors\n    // into a two-dimensional tensor by collapsing these two sets of axes and\n    // concatenate the resulting matrices across the axis 1, finally reshaping\n    // the result to have the proper shape.\n    const outShape =\n        backend_util.computeOutShape(tensors.map(t => t.shape), axis);\n    const tensors2D =\n        tensors.map(t => t.as2D(-1, util.sizeFromShape(t.shape.slice(axis))));\n    const program = new ConcatProgram(tensors2D.map(t => t.shape));\n    const res: Tensor = this.compileAndRun(program, tensors2D);\n    return res.reshape(outShape);\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    const cpuRes = this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.neg(x));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.NEG, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.NEG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const [batch, , ] = a.shape;\n\n    // Since the matrices are vectors, it is faster to call mul().sum()\n    // because sum() is O(sqrt(N)) due to divide-and-conquer.\n    if ((outerShapeA === 1 || outerShapeB === 1) &&\n        sharedDim > MATMUL_SHARED_DIM_THRESHOLD) {\n      if (transposeA) {\n        a = transpose(a, [0, 2, 1]);\n      }\n      if (transposeB) {\n        b = transpose(b, [0, 2, 1]);\n      }\n\n      const a3D = outerShapeB === 1 ? a : a.as3D(batch, sharedDim, 1);\n      const axis = outerShapeB === 1 ? 2 : 1;\n      const b3D = outerShapeB === 1 ? b.as3D(batch, 1, sharedDim) : b;\n      return this.multiply(a3D, b3D).sum(axis, true /* keepDims */);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB);\n    return this.compileAndRun<Tensor3D>(program, [a, b], dtype);\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const [batch, , ] = a.shape;\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB,\n        hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [a, b];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun<Tensor3D>(program, inputs, dtype);\n  }\n\n  multiply(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64') {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n\n      const realProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.REAL, a.shape, b.shape);\n      const imagProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.IMAG, a.shape, b.shape);\n\n      const inputs = [\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.imag),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.imag)\n      ];\n      const real = this.compileAndRun<Tensor>(realProgram, inputs);\n      const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n\n      const complex = this.complex(real, imag);\n      real.dispose();\n      imag.dispose();\n      return complex;\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = multiplyImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.MUL, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.MUL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], a.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n        new LRNPackedProgram(x.shape, radius, bias, alpha, beta) :\n        new LRNProgram(x.shape, radius, bias, alpha, beta);\n    return this.compileAndRun(program, [x]);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program =\n        new LRNGradProgram(inputImage.shape, depthRadius, bias, alpha, beta);\n    return this.compileAndRun(program, [inputImage, outputImage, dy]);\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    if (x.dtype === 'string') {\n      const data = this.readSync(x.dataId) as Uint8Array[];\n      const decodedData = data.map(d => util.decodeString(d));\n      const buf = buffer(x.shape, x.dtype, decodedData);\n      return tile(buf, reps) as T;\n    }\n    const program = new TileProgram(x.shape, reps);\n    return this.compileAndRun(program, [x]);\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new PadPackedProgram(x.shape, paddings, constantValue) :\n        new PadProgram(x.shape, paddings, constantValue);\n    return this.compileAndRun(program, [x]);\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x, indices], () => this.cpuBackend.gather(x, indices, axis));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = new GatherProgram(x.shape, indices.size, axis);\n    return this.compileAndRun(program, [x, indices]);\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'batchToSpaceND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: Array<[number, number]>): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'spaceToBatchND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const completePaddings: Array<[number, number]> = [[0, 0]];\n    completePaddings.push(...paddings);\n    for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n      completePaddings.push([0, 0]);\n    }\n\n    const paddedX = x.pad(completePaddings);\n\n    const reshapedPaddedShape =\n        backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n    const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n        reshapedPaddedShape.length, blockShape.length, false);\n\n    const flattenShape = backend_util.getReshapedPermuted(\n        paddedX.shape, blockShape, prod, false);\n\n    const paddedXT = transpose(\n        paddedX.reshape(reshapedPaddedShape),\n        permutedReshapedPaddedPermutation);\n    return reshape(paddedXT, flattenShape) as T;\n  }\n\n  private reduce(\n      x: Tensor2D, reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod',\n      dtype: DataType): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const outSize = Math.ceil(inSize / windowSize);\n    const reduceInfo = {windowSize, inSize, batchSize, outSize};\n    const program = new ReduceProgram(reduceInfo, reduceType);\n    const output = this.compileAndRun<Tensor2D>(program, [x], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.reduce(output, reduceType, dtype);\n  }\n\n  private argReduce(\n      x: Tensor2D, reduceType: 'max'|'min',\n      bestIndicesA: Tensor2D = null): Tensor2D {\n    let batchSize = x.shape[0];\n    let inSize = x.shape[1];\n    if (bestIndicesA != null) {\n      batchSize = bestIndicesA.shape[0];\n      inSize = bestIndicesA.shape[1];\n    }\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const reduceInfo = {\n      windowSize,\n      inSize,\n      batchSize,\n      outSize: Math.ceil(inSize / windowSize)\n    };\n    const program =\n        new ArgMinMaxProgram(reduceInfo, reduceType, bestIndicesA == null);\n    const inputs = [x];\n    if (bestIndicesA != null) {\n      inputs.push(bestIndicesA);\n    }\n    const output = this.compileAndRun<Tensor2D>(program, inputs, 'int32');\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.argReduce(x, reduceType, output);\n  }\n\n  private argReducePacked(\n      x: Tensor, reduceType: 'max'|'min', bestIndicesA: Tensor = null): Tensor {\n    const inShape = bestIndicesA != null ? bestIndicesA.shape : x.shape;\n    const inSize = inShape[inShape.length - 1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const program = new ArgMinMaxPackedProgram(\n        inShape, windowSize, reduceType, bestIndicesA == null);\n    const inputs = bestIndicesA == null ? [x] : [x, bestIndicesA];\n    const output = this.compileAndRun<Tensor>(program, inputs, 'int32');\n    if (output.rank === x.rank) {\n      return this.argReducePacked(x, reduceType, output);\n    }\n    return output;\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'sum', outputDType).reshape(outShape);\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.prod(x, axes));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'prod', outputDType).reshape(outShape);\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    let axis = 0;\n    const permutation = backend_util.getAxesPermutation([axis], x.rank);\n    let permutedX = x;\n    if (permutation != null) {\n      permutedX = transpose(x, permutation);\n      axis = backend_util.getInnerMostAxes(1, x.rank)[0];\n    }\n\n    const outShape =\n        segment_util.computeOutShape(permutedX.shape, axis, numSegments);\n    const inSize = util.sizeFromShape([permutedX.shape[axis]]);\n    const a2D = permutedX.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    let result =\n        this.segOpCompute(\n                a2D, 'unsortedSegmentSum', segmentIds, outputDType, numSegments)\n            .reshape(outShape);\n    if (permutation != null) {\n      result =\n          transpose(result, backend_util.getUndoAxesPermutation(permutation));\n    }\n    return result;\n  }\n\n  private segOpCompute(\n      x: Tensor2D, segOpType: 'unsortedSegmentSum', segmentIds: Tensor1D,\n      dtype: DataType, numSegments: number): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize =\n        segment_util.segOpComputeOptimalWindowSize(inSize, numSegments);\n    const segOpInfo = {windowSize, inSize, batchSize, numSegments};\n    const program = new SegmentOpProgram(segOpInfo, segOpType);\n    const output =\n        this.compileAndRun<Tensor2D>(program, [x, segmentIds], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === numSegments) {\n      return output;\n    }\n    segmentIds = range(0, numSegments).tile([inSize / windowSize]);\n    return this.segOpCompute(output, segOpType, segmentIds, dtype, numSegments);\n  }\n\n  private argMinMaxReduce(x: Tensor, axis: number, reduceType: 'min'|'max'):\n      Tensor {\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims(\n        'arg' + reduceType.charAt(0).toUpperCase() + reduceType.slice(1), axes,\n        x.rank);\n    if (!env().getBool('WEBGL_PACK_REDUCE') || x.rank <= 2) {\n      const [outShape, reduceShape] =\n          backend_util.computeOutAndReduceShapes(x.shape, axes);\n      const inSize = util.sizeFromShape(reduceShape);\n      const a2D = x.as2D(-1, inSize);\n      return this.argReduce(a2D, reduceType).reshape(outShape);\n    }\n    return this.argReducePacked(x, reduceType);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'min');\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'max');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `WebGL cumsum shader expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const size = x.shape[axis];\n    let result = x;\n    // Use cumsum parallel algorithm, ref:\n    // https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-39-parallel-prefix-sum-scan-cuda\n    for (let i = 0; i <= Math.ceil(Math.log2(size)) - 1; i++) {\n      const program = new CumSumProgram(x.shape, false, reverse);\n      const customSetup = program.getCustomSetupFunc(i);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result], result.dtype, customSetup);\n      prevResult.dispose();\n    }\n    // For exclusive cumsum, shift the end result in the direction of sum and\n    // add 0 to the front index.\n    if (exclusive) {\n      const program = new CumSumProgram(x.shape, exclusive, reverse);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result]);\n      prevResult.dispose();\n    }\n\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.EQUAL, 'bool');\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.NOT_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.NOT_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.less(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.LESS, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LESS_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.greater(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.GREATER, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.GREATER, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.GREATER_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.GREATER_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalNot<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOGICAL_NOT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_AND, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_AND, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_OR, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_OR, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    const program = new SelectProgram(condition.rank, a.shape, a.rank);\n    return this.compileAndRun(\n        program, [condition, a, b], upcastType(a.dtype, b.dtype));\n  }\n\n  where(condition: Tensor): Tensor2D {\n    backend_util.warn(\n        'tf.where() in webgl locks the UI thread. ' +\n        'Call tf.whereAsync() instead');\n    const condVals = condition.dataSync();\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    const xVals = x.dataSync();\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'min', a2D.dtype).reshape(outShape);\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.minimum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MIN, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MIN, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MOD, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MOD, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.maximum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MAX, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MAX, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'all', a2D.dtype).reshape(outShape);\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'any', a2D.dtype).reshape(outShape);\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    const op = binaryop_gpu.INT_DIV;\n    const outputDtype = 'int32';\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.INT_DIV, outputDtype);\n    }\n    const program = new BinaryOpProgram(op, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], outputDtype);\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.ADD);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = addImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.ADD, dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.ADD, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  private packedUnaryOp(x: TensorInfo, op: string, dtype: DataType) {\n    const program = new UnaryOpPackedProgram(x.shape, op);\n    return this.compileAndRun<Tensor>(program, [x], dtype);\n  }\n\n  private packedBinaryOp(\n      a: TensorInfo, b: TensorInfo, op: string, dtype: DataType,\n      checkOutOfBounds = false) {\n    const program =\n        new BinaryOpPackedProgram(op, a.shape, b.shape, checkOutOfBounds);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  /**\n   * Computes a complex binary operation that can be decomposed into a simple\n   * binary operation on both the real and imagary parts.\n   */\n  private complexSeparableBinaryOp(a: Tensor, b: Tensor, op: string): Tensor {\n    const aData = this.texData.get(a.dataId);\n    const bData = this.texData.get(b.dataId);\n\n    const [real, imag] = [\n      [aData.complexTensors.real, bData.complexTensors.real],\n      [aData.complexTensors.imag, bData.complexTensors.imag]\n    ].map(complexParts => {\n      const [aPart, bPart] = complexParts;\n\n      const aHandle = this.makeComplexComponentTensorInfo(a, aPart);\n      const bHandle = this.makeComplexComponentTensorInfo(b, bPart);\n\n      const program = new BinaryOpProgram(op, a.shape, b.shape);\n      return this.compileAndRun<Tensor>(\n          program, [aHandle, bHandle], upcastType(aPart.dtype, bPart.dtype));\n    });\n\n    const complex = this.complex(real, imag);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  // Returns a TensorInfo with the complex shape and the dataId of the\n  // underlying part. We need to do this because a reshaped complex tensor is\n  // not reflected in its parts.\n  private makeComplexComponentTensorInfo(\n      complexTensor: Tensor, complexPart: Tensor): TensorInfo {\n    return {\n      dataId: complexPart.dataId,\n      dtype: complexPart.dtype,\n      shape: complexTensor.shape\n    };\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n\n    // Limit the number of uploaded textures for optimization.\n    if (tensors.length > env().get('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.addN(tensors.slice(0, midIndex));\n      const rightSide = this.addN(tensors.slice(midIndex));\n      return this.addN([leftSide, rightSide]);\n    }\n\n    const dtype =\n        tensors.map(t => t.dtype).reduce((d1, d2) => upcastType(d1, d2));\n    const shapes = tensors.map(t => t.shape);\n    // We can make sure shapes are identical in op level.\n    const usePackedOp = env().getBool('WEBGL_PACK');\n    const program = usePackedOp ?\n        new AddNPackedProgram(tensors[0].shape, shapes) :\n        new AddNProgram(tensors[0].shape, shapes);\n    return this.compileAndRun<T>(program, tensors, dtype);\n  }\n\n  subtract(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.SUB);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = subImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.SUB, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.SUB, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    const usePackedOp = env().getBool('WEBGL_PACK_BINARY_OPERATIONS');\n    const program = usePackedOp ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.POW, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.POW, a.shape, b.shape);\n    const dtype = upcastType(a.dtype, b.dtype);\n    return this.compileAndRun<T>(program, [a, b], dtype);\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          ceilImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.CEIL, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.CEIL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  floor<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = floorImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.FLOOR, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.FLOOR);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_NAN);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_INF);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_FINITE);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n\n  round<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ROUND);\n    return this.compileAndRun(program, [x]);\n  }\n\n  exp<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          expImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXP, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXP);\n    return this.compileAndRun(program, [x]);\n  }\n\n  expm1<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = expm1ImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXPM1, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXPM1);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n    const a = this.subtract(logits, maxLogit.reshape(expandedShape));\n    const b = this.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax kernel\n    // modularization.\n    return div(b, sumExp);\n  }\n\n  log<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          logImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.LOG, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  log1p<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG1P);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sqrt<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  rsqrt<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = rsqrtImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.RSQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reciprocal<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.RECIPROCAL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU6);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU6);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  prelu<T extends Tensor>(x: T, alpha: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.PRELU, x.shape, alpha.shape) :\n        new BinaryOpProgram(binaryop_gpu.PRELU, x.shape, alpha.shape);\n    return this.compileAndRun(program, [x, alpha]);\n  }\n\n  elu<T extends Tensor>(x: T): T {\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.ELU, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.ELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.ELU_DER, dy.shape, y.shape) :\n        new BinaryOpProgram(binaryop_gpu.ELU_DER, dy.shape, y.shape);\n    return this.compileAndRun(program, [dy, y]);\n  }\n\n  selu<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  int<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TO_INT);\n    return this.compileAndRun(program, [x], 'int32');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    let program;\n    if (env().getBool('WEBGL_PACK_CLIP')) {\n      program = new ClipPackedProgram(x.shape);\n    } else {\n      program = new ClipProgram(x.shape);\n    }\n    const customSetup = program.getCustomSetupFunc(min, max);\n    return this.compileAndRun(program, [x], null, customSetup);\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    // TODO: handle cases when x is complex.\n    if (this.shouldExecuteOnCPU([x]) && x.dtype !== 'complex64') {\n      const outValues =\n          simpleAbsImplCPU(this.texData.get(x.dataId).values as TypedArray);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.ABS, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.ABS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  complexAbs<T extends Tensor>(x: T): T {\n    const xData = this.texData.get(x.dataId);\n\n    const program = new ComplexAbsProgram(x.shape);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    return this.compileAndRun<Tensor>(program, inputs) as T;\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGMOID);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SOFTPLUS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASIN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acos<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atan<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATAN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.COSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  tanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ERF);\n    return this.compileAndRun(program, [x]);\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.STEP(alpha));\n    return this.compileAndRun(program, [x]);\n  }\n\n  private conv2dByMatMul(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Reshapes conv2D input to 2D tensors, uses matMul and then reshape the\n    // result from 2D to 4D.\n    const xShape = x.shape;\n    const xTexData = this.texData.get(x.dataId);\n    const sharedMatMulDim = convInfo.inChannels;\n    const outerShapeX = xShape[0] * xShape[1] * xShape[2];\n    const outerShapeFilter = convInfo.outChannels;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const transposeA = false;\n    const transposeB = false;\n\n    // TODO: Once reduction ops are packed, batchMatMul will always be packed\n    // and we can remove this condition.\n    const batchMatMulWillBeUnpacked =\n        (outerShapeX === 1 || outerShapeFilter === 1) &&\n        sharedMatMulDim > MATMUL_SHARED_DIM_THRESHOLD;\n    const reshapeWillBeExpensive = xShape[2] % 2 !== 0 && !!xTexData.isPacked;\n\n    if (batchMatMulWillBeUnpacked || !env().getBool('WEBGL_LAZILY_UNPACK') ||\n        !env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ||\n        !reshapeWillBeExpensive) {\n      const targetShape = isChannelsLast ? xShape[0] * xShape[1] * xShape[2] :\n                                           xShape[0] * xShape[2] * xShape[3];\n      const xReshaped = reshape(x, [1, targetShape, convInfo.inChannels]);\n      const filterReshaped =\n          reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n      const result = this.fusedBatchMatMul({\n        a: xReshaped as Tensor3D,\n        b: filterReshaped as Tensor3D,\n        transposeA,\n        transposeB,\n        bias,\n        activation,\n        preluActivationWeights\n      });\n      return reshape(result, convInfo.outShape);\n    }\n\n    // Following optimization is specific to packed |x| with odd row count\n    // (For example, in channelLast mode, 'row count' refers to x.shape[2]):\n    // we avoid expensive packed 2x2 reshape by padding row count to next,\n    // even number. When x.shape[2] is odd, the result of packed batchMatMul is\n    // the same (has the same texture layout and and values in the texture) as\n    // it is for even x.shape[2] + 1. We make the odd-rows tensor to look like\n    // even-rows tensor before the operation and, after the batchMatMul,\n    // fix the even-rows result to have odd number of rows.\n    const targetShape = isChannelsLast ?\n        xShape[0] * xShape[1] * (xShape[2] + 1) :\n        xShape[0] * xShape[2] * (xShape[3] + 1);\n    const xReshaped: TensorInfo = {\n      dataId: x.dataId,\n      shape: [1, targetShape, convInfo.inChannels],\n      dtype: x.dtype\n    };\n    // xTexData.shape gets referenced from GPGPUBinary.inShapeInfos.\n    // Decrementing row count, after batchMatMul->...->compileProgram leads to\n    // invalid row count within the reference in GPGPUBinary.inShapeInfos.\n    // Alternative fix would be to provide a copy to GPGPUBinary.inShapeInfos\n    // in compileProgram method, but that would affect compilation of all\n    // programs - instead, provide a copy here, with even row count, before\n    // calling batchMatMul->...->compileProgram and after that, the original\n    // xTexData.shape is restored.\n    const originalXTexDataShape = xTexData.shape;\n    xTexData.shape = xTexData.shape.slice();\n    xTexData.shape[xTexData.shape.length - 2]++;\n    util.assert(\n        webgl_util.isReshapeFree(xTexData.shape, xReshaped.shape),\n        () => `packed reshape ${xTexData.shape} to ${\n            xReshaped.shape} isn't free`);\n    const filterReshaped =\n        reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n    const pointwiseConv = this.fusedBatchMatMul({\n      a: xReshaped as Tensor3D,\n      b: filterReshaped as Tensor3D,\n      transposeA,\n      transposeB,\n      bias,\n      activation,\n      preluActivationWeights\n    });\n    const pointwiseConvTexData = this.texData.get(pointwiseConv.dataId);\n    util.assert(\n        pointwiseConvTexData.isPacked,\n        () => 'batchMatMul result is expected to be packed');\n    // Restore the input shape to original.\n    xTexData.shape = originalXTexDataShape;\n    // Set the output shape - there is no need for expensive reshape as data\n    // layout is already correct.\n    pointwiseConvTexData.shape = convInfo.outShape;\n    return engine().makeTensorFromDataId(\n               pointwiseConv.dataId, convInfo.outShape, pointwiseConv.dtype) as\n        Tensor4D;\n  }\n\n  private conv2dWithIm2Row(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Rearranges conv2d input so each block to be convolved over forms the\n    // column of a new matrix with shape [filterWidth * filterHeight *\n    // inChannels, outHeight * outWidth]. The filter is also rearranged so each\n    // output channel forms a row of a new matrix with shape [outChannels,\n    // filterWidth * filterHeight * inChannels]. The convolution is then\n    // computed by multiplying these matrices and reshaping the result.\n    const {\n      filterWidth,\n      filterHeight,\n      inChannels,\n      outWidth,\n      outHeight,\n      dataFormat\n    } = convInfo;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n\n    const sharedDim = filterWidth * filterHeight * inChannels;\n    const numCols = outHeight * outWidth;\n    const x2ColShape = [sharedDim, numCols];\n    const transposeA = true;\n    const transposeB = false;\n\n    const xSqueezed = x.squeeze([0]);\n    const w2Row = filter.reshape([1, sharedDim, -1]);\n\n    const im2ColProgram =\n        new Im2ColPackedProgram(x2ColShape, xSqueezed.shape, convInfo);\n    const im2Col: Tensor3D =\n        this.compileAndRun<Tensor2D>(im2ColProgram, [xSqueezed]).reshape([\n          1, x2ColShape[0], x2ColShape[1]\n        ]);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const matmulProgram = new MatMulPackedProgram(\n        im2Col.shape, [1, numCols, convInfo.outChannels], transposeA,\n        transposeB, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [im2Col, w2Row];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    const product = this.compileAndRun<Tensor4D>(matmulProgram, inputs);\n\n    if (isChannelsLast) {\n      return product.reshape([1, outHeight, outWidth, convInfo.outChannels]);\n    } else {\n      return product.reshape([1, convInfo.outChannels, outHeight, outWidth]);\n    }\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && input.shape[0] === 1) {\n      return this.conv2dWithIm2Row(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, false) : null;\n    const program = new Conv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [input, filter];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun(program, inputs);\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(x, filter, convInfo);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && x.shape[0] === 1) {\n      return this.conv2dWithIm2Row(x, filter, convInfo);\n    }\n    const program = new Conv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new Conv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    const program = new Conv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    const shouldPackDepthwiseConv = env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1;\n    const fusedActivation = activation ?\n        mapActivationToShaderProgram(activation, shouldPackDepthwiseConv) :\n        null;\n    const inputs: Tensor[] = [input, filter];\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    if (hasBias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (shouldPackDepthwiseConv) {\n      program = new DepthwiseConvPacked2DProgram(\n          convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n      return this.compileAndRun(program, inputs);\n    }\n\n    program = new DepthwiseConv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    return this.compileAndRun(program, inputs);\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1) {\n      program = new DepthwiseConvPacked2DProgram(convInfo);\n      return this.compileAndRun(program, [x, filter]);\n    }\n\n    program = new DepthwiseConv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Conv3DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return backend_util.castTensor(x, dtype, this);\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = this.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'avg', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const avgPool3dBackpropProgram = new AvgPool3DBackpropProgram(convInfo);\n    return this.compileAndRun(avgPool3dBackpropProgram, [dy], x.dtype);\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'max', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const getPositions = true;\n    const maxPool3dPositionsProgram =\n        new Pool3DProgram(convInfo, 'max', getPositions);\n    const maxPool3dPositions: Tensor5D =\n        this.compileAndRun(maxPool3dPositionsProgram, [x]);\n    const maxPool3dBackPropProgram = new MaxPool3DBackpropProgram(convInfo);\n    const result = this.compileAndRun(\n        maxPool3dBackPropProgram, [dy, maxPool3dPositions], x.dtype);\n    maxPool3dPositions.dispose();\n    return result as Tensor5D;\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_IMAGE_OPERATIONS') ?\n        new ResizeBilinearPackedProgram(\n            x.shape, newHeight, newWidth, alignCorners) :\n        new ResizeBilinearProgram(x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    const program = new ResizeBilinearBackpropProgram(dy, x, alignCorners);\n\n    return this.compileAndRun(program, [dy]);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = new ResizeNearestNeighborProgram(\n        x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x]);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    const program =\n        new ResizeNearestNeigborBackpropProgram(dy, x, alignCorners);\n    return this.compileAndRun(program, [dy]);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    const probs = normalized ? logits : softmax(logits);\n    const batchSize = probs.shape[0];\n    const numOutcomes = probs.shape[1];\n    const program = new MultinomialProgram(batchSize, numOutcomes, numSamples);\n    const customSetup = program.getCustomSetupFunc(seed);\n    return this.compileAndRun(program, [probs], 'int32', customSetup);\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    const program = new OneHotProgram(indices.size, depth, onValue, offValue);\n    return this.compileAndRun(program, [indices]);\n  }\n\n  diag(x: Tensor): Tensor {\n    const program = new DiagProgram(x.size);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    const program = new CropAndResizeProgram(\n        image.shape, boxes.shape, cropSize, method, extrapolationValue);\n    return this.compileAndRun(program, [image, boxes, boxIndex], 'float32');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = (dataFormat === 'NHWC') ? x.shape[1] : x.shape[2];\n    const inputWidth = (dataFormat === 'NHWC') ? x.shape[2] : x.shape[3];\n    const inputDepth = (dataFormat === 'NHWC') ? x.shape[3] : x.shape[1];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const outputShape = (dataFormat === 'NHWC') ?\n        [batchSize, outputHeight, outputWidth, outputDepth] :\n        [batchSize, outputDepth, outputHeight, outputWidth];\n\n    const program = new DepthToSpaceProgram(outputShape, blockSize, dataFormat);\n    return this.compileAndRun(program, [x]);\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n    const flattenIndices = indices.reshape([numUpdates, sliceRank]);\n    const flattenX = updates.reshape([numUpdates, sliceSize]);\n\n    if (outputSize === 0) {\n      return backend_util.reshapeTensor(tensor([]), shape);\n    }\n    const defaultValue = scalar(0);\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, flattenIndices.rank, flattenX.rank, strides,\n        flattenShape);\n    const res: Tensor =\n        this.compileAndRun(program, [flattenX, flattenIndices, defaultValue]);\n    return res.reshape(shape);\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n\n    const sumDupeIndices = false;\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, sparseIndices.rank, sparseValues.rank, strides,\n        [outputSize, 1], sumDupeIndices);\n    const res: Tensor = this.compileAndRun(\n        program, [sparseValues, sparseIndices, defaultValue]);\n    return res.reshape(outputShape);\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    const inverse = false;\n    return this.fftImpl(x, inverse);\n  }\n\n  ifft(x: Tensor2D): Tensor2D {\n    const inverse = true;\n    return this.fftImpl(x, inverse);\n  }\n\n  private fftImpl(x: Tensor2D, inverse: boolean): Tensor2D {\n    const xData = this.texData.get(x.dataId);\n\n    const realProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.REAL, x.shape, inverse);\n    const imagProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.IMAG, x.shape, inverse);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    const real = this.compileAndRun<Tensor>(realProgram, inputs);\n    const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n    const complex = this.complex(real, imag).as2D(x.shape[0], x.shape[1]);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n\n    const flattenIndices = indices.reshape([numSlices, sliceRank]);\n    const flattenX = x.reshape([x.size / sliceSize, sliceSize]);\n    const program =\n        new GatherNDProgram(sliceRank, strides, [numSlices, sliceSize]);\n    const res: Tensor = this.compileAndRun(program, [flattenX, flattenIndices]);\n    return res.reshape(resultShape);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n\n    if (dtype === 'string') {\n      // String type should be handled in CPU memory.\n      const values = util.getArrayFromDType(dtype, util.sizeFromShape(shape));\n      values.fill(value as string);\n      return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n    } else {\n      const program = new FillProgram(shape, value as number);\n      const customSetup = program.getCustomSetupFunc(value as number);\n      return this.compileAndRun(program, [], dtype, customSetup);\n    }\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported under string dtype');\n    } else {\n      // TODO(cais, smilkov): Add WebGL shader for onesLike:\n      //   https://github.com/tensorflow/tfjs/issues/1293\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return this.fill(x.shape, x.dtype === 'string' ? '' : 0, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    // TODO: Use CPU implementation due to the precision problem in Safari.\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  makeTensorInfo(shape: number[], dtype: DataType, values?: BackendValues):\n      TensorInfo {\n    const dataId = this.write(values, shape, dtype);\n    this.texData.get(dataId).usage = null;\n    return {dataId, shape, dtype};\n  }\n\n  private makeOutput<T extends Tensor>(\n      shape: number[], dtype: DataType, values?: BackendValues): T {\n    const {dataId} = this.makeTensorInfo(shape, dtype, values);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  private unpackTensor(input: TensorInfo): TensorInfo {\n    const program = new UnpackProgram(input.shape);\n    return this.runWebGLProgram(program, [input], input.dtype);\n  }\n\n  private packTensor(input: TensorInfo): TensorInfo {\n    const program = new PackProgram(input.shape);\n    const preventEagerUnpackingOutput = true;\n    return this.runWebGLProgram(\n        program, [input], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOutput);\n  }\n\n  private packedReshape(input: TensorInfo, afterShape: number[]): TensorInfo {\n    const input3DShape = [\n      webgl_util.getBatchDim(input.shape),\n      ...webgl_util.getRowsCols(input.shape)\n    ] as [number, number, number];\n    const input3D: TensorInfo = {\n      dtype: input.dtype,\n      shape: input3DShape,\n      dataId: input.dataId\n    };\n    const afterShapeAs3D = [\n      webgl_util.getBatchDim(afterShape), ...webgl_util.getRowsCols(afterShape)\n    ] as [number, number, number];\n\n    const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n    const preventEagerUnpackingOfOutput = true;\n    const output = this.runWebGLProgram(\n        program, [input3D], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOfOutput);\n    return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n  }\n\n  private decode(dataId: DataId): TensorInfo {\n    const texData = this.texData.get(dataId);\n    const {isPacked, shape, dtype} = texData;\n    const shapeAs3D =\n        webgl_util.getShapeAs3D(shape) as [number, number, number];\n    let program;\n    if (isPacked) {\n      program = new DecodeMatrixPackedProgram(shapeAs3D);\n    } else {\n      program = new DecodeMatrixProgram(shapeAs3D);\n    }\n    const preventEagerUnpackingOfOutput = true;\n    const out = this.runWebGLProgram(\n        program, [{shape: shapeAs3D, dtype, dataId}], dtype,\n        null /* customSetup */, preventEagerUnpackingOfOutput);\n    return {dtype, shape, dataId: out.dataId};\n  }\n\n  runWebGLProgram(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): TensorInfo {\n    const output = this.makeTensorInfo(program.outputShape, outputDtype);\n    const outData = this.texData.get(output.dataId);\n    if (program.packedOutput) {\n      outData.isPacked = true;\n    }\n    if (program.outPackingScheme === tex_util.PackingScheme.DENSE) {\n      const texelShape = tex_util.getDenseTexShape(program.outputShape);\n      // For a densely packed output, we explicitly set texShape\n      // so it doesn't get assigned later according to our typical packing\n      // scheme wherein a single texel can only contain values from adjacent\n      // rows/cols.\n      outData.texShape = texelShape.map(d => d * 2) as [number, number];\n    }\n    if (program.outTexUsage != null) {\n      outData.usage = program.outTexUsage;\n    }\n    if (util.sizeFromShape(output.shape) === 0) {\n      // Short-circuit the computation since the result is empty (has 0 in its\n      // shape).\n      outData.values =\n          util.getTypedArrayFromDType(output.dtype as 'float32', 0);\n      return output;\n    }\n\n    const dataToDispose: TensorInfo[] = [];\n    const inputsData: TensorData[] = inputs.map(input => {\n      if (input.dtype === 'complex64') {\n        throw new Error(\n            `GPGPUProgram does not support complex64 input. For complex64 ` +\n            `dtypes, please separate the program into real and imaginary ` +\n            `parts.`);\n      }\n\n      let texData = this.texData.get(input.dataId);\n\n      if (texData.texture == null) {\n        if (!program.packedInputs &&\n            util.sizeFromShape(input.shape) <=\n                env().getNumber('WEBGL_SIZE_UPLOAD_UNIFORM')) {\n          // Upload small tensors that live on the CPU as uniforms, not as\n          // textures. Do this only when the environment supports 32bit floats\n          // due to problems when comparing 16bit floats with 32bit floats.\n          // TODO(https://github.com/tensorflow/tfjs/issues/821): Make it\n          // possible for packed shaders to sample from uniforms.\n          return {\n            shape: input.shape,\n            texData: null,\n            isUniform: true,\n            uniformValues: texData.values as TypedArray\n          };\n        }\n\n        // This ensures that if a packed program's inputs have not yet been\n        // uploaded to the GPU, they get uploaded as packed right off the bat.\n        if (program.packedInputs) {\n          texData.isPacked = true;\n          texData.shape = input.shape;\n        }\n      } else if (!!texData.isPacked !== !!program.packedInputs) {\n        input = texData.isPacked ? this.unpackTensor(input) :\n                                   this.packTensor(input);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n      } else if (\n          texData.isPacked &&\n          !webgl_util.isReshapeFree(texData.shape, input.shape)) {\n        // This is a special case where a texture exists for a tensor\n        // but the shapes are incompatible (due to packing constraints) because\n        // the tensor did not have a chance to go through the packed reshape\n        // shader. This only happens when we reshape the *same* tensor to form\n        // *distinct* inputs to an op, e.g. dotting a vector with itself. This\n        // case will disappear once packed uploading is the default.\n\n        const savedInput = input;\n        const targetShape = input.shape;\n\n        input.shape = texData.shape;\n        input = this.packedReshape(input as Tensor, targetShape);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n\n        savedInput.shape = targetShape;\n      }\n\n      this.uploadToGPU(input.dataId);\n      return {shape: input.shape, texData, isUniform: false};\n    });\n\n    this.uploadToGPU(output.dataId);\n    const outputData:\n        TensorData = {shape: output.shape, texData: outData, isUniform: false};\n    const key = gpgpu_math.makeShaderKey(program, inputsData, outputData);\n    const binary = this.getAndSaveBinary(key, () => {\n      return gpgpu_math.compileProgram(\n          this.gpgpu, program, inputsData, outputData);\n    });\n    const shouldTimeProgram = this.activeTimers != null;\n    let query: WebGLQuery|CPUTimerQuery;\n    if (shouldTimeProgram) {\n      query = this.startTimer();\n    }\n\n    gpgpu_math.runProgram(\n        this.gpgpu, binary, inputsData, outputData, customSetup);\n\n    dataToDispose.forEach(info => this.disposeIntermediateTensorInfo(info));\n\n    if (shouldTimeProgram) {\n      query = this.endTimer(query);\n      this.activeTimers.push(\n          {name: program.constructor.name, query: this.getQueryTime(query)});\n    }\n\n    if (!env().getBool('WEBGL_LAZILY_UNPACK') && outData.isPacked &&\n        preventEagerUnpackingOfOutput === false) {\n      const unpacked = this.unpackTensor(output);\n      this.disposeIntermediateTensorInfo(output);\n      return unpacked;\n    }\n    return output;\n  }\n\n  compileAndRun<K extends TensorInfo>(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype?: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): K {\n    outputDtype = outputDtype || inputs[0].dtype;\n    const outInfo = this.runWebGLProgram(\n        program, inputs, outputDtype, customSetup,\n        preventEagerUnpackingOfOutput);\n    return engine().makeTensorFromDataId(\n               outInfo.dataId, outInfo.shape, outInfo.dtype) as {} as K;\n  }\n\n  private getAndSaveBinary(key: string, getBinary: () => GPGPUBinary):\n      GPGPUBinary {\n    if (!(key in this.binaryCache)) {\n      this.binaryCache[key] = getBinary();\n    }\n    return this.binaryCache[key];\n  }\n\n  getTextureManager(): TextureManager {\n    return this.textureManager;\n  }\n\n  private disposed = false;\n\n  dispose() {\n    if (this.disposed) {\n      return;\n    }\n    // Avoid disposing the compiled webgl programs during unit testing because\n    // it slows down test execution.\n    if (!env().getBool('IS_TEST')) {\n      const allKeys = Object.keys(this.binaryCache);\n      allKeys.forEach(key => {\n        this.gpgpu.deleteProgram(this.binaryCache[key].webGLProgram);\n        delete this.binaryCache[key];\n      });\n    }\n    this.textureManager.dispose();\n    if (this.canvas != null &&\n        (typeof (HTMLCanvasElement) !== 'undefined' &&\n         this.canvas instanceof HTMLCanvasElement)) {\n      this.canvas.remove();\n    } else {\n      this.canvas = null;\n    }\n    if (this.gpgpuCreatedLocally) {\n      this.gpgpu.program = null;\n      this.gpgpu.dispose();\n    }\n    this.disposed = true;\n  }\n\n  floatPrecision(): 16|32 {\n    if (this.floatPrecisionValue == null) {\n      this.floatPrecisionValue = tidy(() => {\n        if (!env().get('WEBGL_RENDER_FLOAT32_ENABLED')) {\n          // Momentarily switching DEBUG flag to false so we don't throw an\n          // error trying to upload a small value.\n          const debugFlag = env().getBool('DEBUG');\n          env().set('DEBUG', false);\n          const underflowCheckValue = this.abs(scalar(1e-8)).dataSync()[0];\n          env().set('DEBUG', debugFlag);\n\n          if (underflowCheckValue > 0) {\n            return 32;\n          }\n        }\n        return 16;\n      });\n    }\n    return this.floatPrecisionValue;\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  private uploadToGPU(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    const {shape, dtype, values, texture, usage, isPacked} = texData;\n\n    if (texture != null) {\n      // Array is already on GPU. No-op.\n      return;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let texShape = texData.texShape;\n    if (texShape == null) {\n      texShape = webgl_util.getTextureShapeFromLogicalShape(shape, isPacked);\n      texData.texShape = texShape;\n    }\n\n    if (values != null) {\n      const shapeAs3D = webgl_util.getShapeAs3D(shape);\n\n      let program;\n      let width = texShape[1], height = texShape[0];\n      const isByteArray = values instanceof Uint8Array;\n\n      if (isPacked) {\n        [width, height] = tex_util.getPackedMatrixTextureShapeWidthHeight(\n            texShape[0], texShape[1]);\n        program = new EncodeMatrixPackedProgram(\n            shapeAs3D, [height, width], isByteArray);\n      } else {\n        program =\n            new EncodeMatrixProgram(shapeAs3D, [height, width], isByteArray);\n      }\n\n      const tempDenseInputHandle = this.makeTensorInfo([height, width], dtype);\n      if (isByteArray) {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.PIXELS;\n      } else {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.UPLOAD;\n      }\n      this.gpgpu.uploadDenseMatrixToTexture(\n          this.getTexture(tempDenseInputHandle.dataId), width, height,\n          values as TypedArray);\n\n      // We want the output to remain packed regardless of the value of\n      // WEBGL_PACK.\n      const preventEagerUnpacking = true;\n      const encodedOutputTarget = this.runWebGLProgram(\n          program, [tempDenseInputHandle], dtype, null, preventEagerUnpacking);\n\n      // Have the original texture assume the identity of the encoded output.\n      const outputTexData = this.texData.get(encodedOutputTarget.dataId);\n      texData.texture = outputTexData.texture;\n      texData.texShape = outputTexData.texShape;\n      texData.isPacked = outputTexData.isPacked;\n      texData.usage = outputTexData.usage;\n\n      this.disposeIntermediateTensorInfo(tempDenseInputHandle);\n      this.texData.delete(encodedOutputTarget.dataId);\n\n      // Once uploaded, don't store the values on cpu.\n      texData.values = null;\n      if (shouldTimeProgram) {\n        this.uploadWaitMs += util.now() - start;\n      }\n    } else {\n      const newTexture = this.acquireTexture(texShape, usage, dtype, isPacked);\n      texData.texture = newTexture;\n    }\n  }\n\n  private convertAndCacheOnCPU(dataId: DataId, float32Values?: Float32Array):\n      TypedArray {\n    const texData = this.texData.get(dataId);\n    const {dtype} = texData;\n\n    this.releaseGPUData(dataId);\n\n    if (float32Values != null) {\n      texData.values = float32ToTypedArray(float32Values, dtype as 'float32');\n    }\n    return texData.values as TypedArray;\n  }\n\n  private acquireTexture(\n      texShape: [number, number], texType: TextureUsage, dtype: DataType,\n      isPacked: boolean): WebGLTexture {\n    this.numBytesInGPU += this.computeBytes(texShape, dtype);\n    if (!this.warnedAboutMemory &&\n        this.numBytesInGPU > this.numMBBeforeWarning * 1024 * 1024) {\n      const mb = (this.numBytesInGPU / 1024 / 1024).toFixed(2);\n      this.warnedAboutMemory = true;\n      console.warn(\n          `High memory usage in GPU: ${mb} MB, ` +\n          `most likely due to a memory leak`);\n    }\n    return this.textureManager.acquireTexture(texShape, texType, isPacked);\n  }\n\n  private computeBytes(shape: [number, number], dtype: DataType) {\n    return shape[0] * shape[1] * util.bytesPerElement(dtype);\n  }\n\n  private tryRunOnCpuOrThrow<T extends Tensor>(\n      inputs: TensorInfo[], fn: () => T): T|null {\n    if (this.shouldExecuteOnCPU(inputs)) {\n      try {\n        return fn();\n      } catch (e) {\n        if (env().getBool('IS_TEST')) {\n          throw new Error('CPU forwarding failed');\n        }\n      }\n    }\n    return null;\n  }\n}\n\nfunction float32ToTypedArray<D extends NumericDataType>(\n    a: Float32Array, dtype: D): tf.DataTypeMap[D] {\n  if (dtype === 'float32' || dtype === 'complex64') {\n    return a as tf.DataTypeMap[D];\n  } else if (dtype === 'int32' || dtype === 'bool') {\n    const result = (dtype === 'int32') ? new Int32Array(a.length) :\n                                         new Uint8Array(a.length);\n    for (let i = 0; i < result.length; ++i) {\n      result[i] = Math.round(a[i]);\n    }\n    return result as tf.DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport * as gpgpu_util from './gpgpu_util';\nimport * as webgl_util from './webgl_util';\n\nexport {MathBackendWebGL, WebGLMemoryInfo, WebGLTimingInfo} from './backend_webgl';\nexport {setWebGLContext} from './canvas_util';\nexport {GPGPUContext} from './gpgpu_context';\nexport {GPGPUProgram} from './gpgpu_math';\n// WebGL specific utils.\nexport {gpgpu_util, webgl_util};\n\n/**\n * Enforce use of half precision textures if available on the platform.\n *\n * @doc {heading: 'Environment', namespace: 'webgl'}\n */\nexport function forceHalfFloat(): void {\n  env().set('WEBGL_FORCE_F16_TEXTURES', true);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// base.ts is the webgl backend without auto kernel registration.\n\nimport {device_util, registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from './backend_webgl';\nexport {version as version_webgl} from './version';\n\nif (device_util.isBrowser()) {\n  registerBackend('webgl', () => new MathBackendWebGL(), 2 /* priority */);\n}\n\n// Export webgl utilities\nexport * from './webgl';\n\n// Export forceHalfFlost under webgl namespace for the union bundle.\nimport {forceHalfFloat} from './webgl';\nexport const webgl = {forceHalfFloat};\n", "import {BinaryInputs, DataType, env, KernelFunc, UnaryInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BinaryOpProgram} from '../binaryop_gpu';\nimport {BinaryOpPackedProgram} from '../binaryop_packed_gpu';\nimport {UnaryOpProgram} from '../unaryop_gpu';\n\nexport const CHECK_NAN_SNIPPET_UNARY = `if (isnan(x)) return x;`;\n\nexport const CHECK_NAN_SNIPPET_BINARY = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const CHECK_NAN_SNIPPET_BINARY_PACKED = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param opSnippets Op snippet to create `UnaryOpProgram`.\n */\nexport function unaryKernelFunc(opSnippet: string): KernelFunc {\n  return ({inputs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = new UnaryOpProgram(x.shape, opSnippet);\n    return webglBackend.runWebGLProgram(program, [x], x.dtype);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param opSnippet Op snippet to create `BinaryOpProgram`.\n * @param packedOpSnippet Op snippet to create `BinaryOpPackedProgram`.\n * @param checkOutOfBoundsForPackedProgram Whether to set checkOutOfBounds=true\n *     when creating BinaryOpPackedProgram.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    opSnippet: string, packedOpSnippet: string,\n    checkOutOfBoundsForPackedProgram?: boolean, dtype?: DataType): KernelFunc {\n  // TODO(jingjin): handle complex64.\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            packedOpSnippet, a.shape, b.shape,\n            !!checkOutOfBoundsForPackedProgram) :\n        new BinaryOpProgram(opSnippet, a.shape, b.shape);\n    const $dtype = dtype || a.dtype;\n    const output = webglBackend.runWebGLProgram(program, [a, b], $dtype);\n    return output;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc, CHECK_NAN_SNIPPET_BINARY, CHECK_NAN_SNIPPET_BINARY_PACKED} from '../kernel_utils/kernel_funcs_utils';\n\nconst ATAN2 = CHECK_NAN_SNIPPET_BINARY + `\n  return atan(a, b);\n`;\n\nconst ATAN2_PACKED = `\n  vec4 result = atan(a, b);\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET_BINARY_PACKED + `\n  return result;\n`;\n\nexport const atan2KernelFunc = binaryKernelFunc(ATAN2, ATAN2_PACKED);\n\nexport const atan2Config: KernelConfig = {\n  kernelName: Atan2,\n  backendName: 'webgl',\n  kernelFunc: atan2KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendWebGL}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'webgl',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function avgPool(args: {\n  inputs: AvgPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const avgPoolProgram = new Pool2DProgram(convInfo, 'avg', false);\n  return backend.runWebGLProgram(avgPoolProgram, [x], 'float32');\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'webgl',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {AvgPool2DBackpropProgram} from '../avg_pool_backprop_gpu';\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const avgPoolBackpropProgram = new AvgPool2DBackpropProgram(convInfo);\n  return backend.runWebGLProgram(avgPoolBackpropProgram, [dy], x.dtype);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = '0.0';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = '1.0';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        float x = getXAtOutCoords();\n        float mean = getMeanAtOutCoords();\n        float variance = getVarianceAtOutCoords();\n        float offset = ${offsetSnippet};\n        float scale = ${scaleSnippet};\n        float inv = scale * inversesqrt(variance + float(${varianceEpsilon}));\n        setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = 'vec4(0.0)';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = 'vec4(1.0)';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        vec4 offset = ${offsetSnippet};\n        vec4 scale = ${scaleSnippet};\n\n        vec4 x = getXAtOutCoords();\n        vec4 mean = getMeanAtOutCoords();\n        vec4 variance = getVarianceAtOutCoords();\n\n        vec4 inv = scale * inversesqrt(variance + vec4(${varianceEpsilon}));\n\n        setOutput((x - mean) * inv + offset);\n      }\n    `;\n  }\n}\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BatchNormProgram} from '../batchnorm_gpu';\nimport {BatchNormPackedProgram} from '../batchnorm_packed_gpu';\n\nexport const batchNormKernelFunc: (params: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendWebGL,\n  attrs: FusedBatchNormAttrs\n}) => TensorInfo = ({inputs, backend, attrs}) => {\n  const {x, mean, variance, offset, scale} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const finalInputs = [x, mean, variance];\n\n  let offsetShape = null;\n  if (offset != null) {\n    offsetShape = offset.shape;\n    finalInputs.push(offset);\n  }\n\n  let scaleShape = null;\n  if (scale != null) {\n    scaleShape = scale.shape;\n    finalInputs.push(scale);\n  }\n\n  const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n      new BatchNormPackedProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon) :\n      new BatchNormProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon);\n  const output =\n      backend.runWebGLProgram(program, finalInputs, finalInputs[0].dtype);\n\n  return output;\n};\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'webgl',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst COS = CHECK_NAN_SNIPPET_UNARY + `\n  return cos(x);\n`;\n\nexport const cosKernelFunc = unaryKernelFunc(COS);\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'webgl',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\n// Without the equality check div produces 0.9999 for a = b, which when\n// floored can cause errors.\nconst DIV = `\nif (a == b) {\n  return 1.0;\n};\nreturn a / b;`;\n\n// We do the same as in ./binaryop_gpu, with vec4 and ivec4.\n// On Linux, the vectorized implementation produces NaNs when a and b are 0.\nconst DIV_PACKED = `\n  // vec4 one = vec4(equal(a, b));\n  // return one + (vec4(1.0) - one) * a / b;\n  vec4 result = a / b;\n  if(a.x == b.x) {\n    result.x = 1.;\n  }\n  if(a.y == b.y) {\n    result.y = 1.;\n  }\n  if(a.z == b.z) {\n    result.z = 1.;\n  }\n  if(a.w == b.w) {\n    result.w = 1.;\n  }\n\n  return result;\n`;\n\nexport const divKernelFunc = binaryKernelFunc(\n    DIV, DIV_PACKED, true /* checkOutOfBoundsForPackedProgram */);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'webgl',\n  kernelFunc: divKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FlipLeftRightProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(imageShape: [number, number, number, number]) {\n    const imageWidth = imageShape[2];\n    this.outputShape = imageShape;\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n\n          int coordX = ${imageWidth} - x;\n          float outputValue;\n          if(coordX >= 0 && coordX < ${imageWidth}) {\n            outputValue = getImage(coords[0], coords[1], coordX, coords[3]);\n          } else {\n            outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {FlipLeftRightProgram} from '../flip_left_right_gpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new FlipLeftRightProgram((image as Tensor4D).shape);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n        vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${width}.0, ${height}.0);\n\n        vec4 values = ${glsl.texture2D}(A, uv);\n        float value;\n        if (depth == 0) {\n          value = values.r;\n        } else if (depth == 1) {\n          value = values.g;\n        } else if (depth == 2) {\n          value = values.b;\n        } else if (depth == 3) {\n          value = values.a;\n        }\n\n        setOutput(floor(value * 255.0 + 0.5));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n\n        vec4 result = vec4(0.);\n\n        for(int row=0; row<=1; row++) {\n          for(int col=0; col<=1; col++) {\n            texC = coords[1] + row;\n            depth = coords[2] + col;\n\n            vec2 uv = (vec2(texC, texR) + halfCR) /\n                       vec2(${width}.0, ${height}.0);\n            vec4 values = ${glsl.texture2D}(A, uv);\n            float value;\n            if (depth == 0) {\n              value = values.r;\n            } else if (depth == 1) {\n              value = values.g;\n            } else if (depth == 2) {\n              value = values.b;\n            } else if (depth == 3) {\n              value = values.a;\n            }\n\n            result[row * 2 + col] = floor(value * 255.0 + 0.5);\n          }\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {TextureUsage} from '../tex_util';\n\nimport {FromPixelsProgram} from './FromPixels_utils/from_pixels_gpu';\nimport {FromPixelsPackedProgram} from './FromPixels_utils/from_pixels_packed_gpu';\n\nexport const fromPixelsConfig: KernelConfig = {\n  kernelName: FromPixels,\n  backendName: 'webgl',\n  kernelFunc: fromPixels as {} as KernelFunc,\n};\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\nfunction fromPixels(args: {\n  inputs: FromPixelsInputs,\n  backend: MathBackendWebGL,\n  attrs: FromPixelsAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  let {pixels} = inputs;\n  const {numChannels} = attrs;\n\n  const isVideo = typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement;\n  const isImage = typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement;\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n\n  const texShape: [number, number] = [height, width];\n  const outShape = [height, width, numChannels];\n\n  if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement | HTMLImageElement, 0, 0, width, height);\n    pixels = fromPixels2DContext.canvas;\n  }\n\n  const tempPixelHandle = backend.makeTensorInfo(texShape, 'int32');\n  // This is a byte texture with pixels.\n  backend.texData.get(tempPixelHandle.dataId).usage = TextureUsage.PIXELS;\n  backend.gpgpu.uploadPixelDataToTexture(\n      backend.getTexture(tempPixelHandle.dataId), pixels as ImageData);\n  const program = env().getBool('WEBGL_PACK') ?\n      new FromPixelsPackedProgram(outShape) :\n      new FromPixelsProgram(outShape);\n  const res = backend.runWebGLProgram(program, [tempPixelHandle], 'int32');\n  backend.disposeData(tempPixelHandle.dataId);\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReduceProgram} from '../reduce_gpu';\n\ntype ReduceTypes = 'all'|'any'|'max'|'min'|'sum'|'prod';\n\n// Returns an array of configuration objects that describe each stage of the\n// reduction.\nfunction getReductionStages(inShape: number[]):\n    Array<{inSize: number, windowSize: number, outSize: number}> {\n  const stages = [];\n\n  while (stages.length === 0 || stages[stages.length - 1].outSize !== 1) {\n    const outSize: number =\n        stages.length ? stages[stages.length - 1].outSize : inShape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(outSize);\n    stages.push({\n      inSize: outSize,\n      windowSize,\n      outSize: Math.ceil(outSize / windowSize)\n    });\n  }\n\n  return stages;\n}\n\nexport function reduce(\n    x: TensorInfo, dtype: DataType, reductionType: ReduceTypes,\n    backend: MathBackendWebGL): TensorInfo {\n  const reductionStages = getReductionStages(x.shape);\n\n  let result = x;\n  for (let i = 0; i < reductionStages.length; i++) {\n    const {inSize, windowSize, outSize} = reductionStages[i];\n\n    const program = new ReduceProgram(\n        {windowSize, inSize, batchSize: x.shape[0], outSize}, reductionType);\n    const previousResult = result;\n    result = backend.runWebGLProgram(program, [result], dtype);\n\n    if (previousResult.dataId !== x.dataId) {\n      backend.disposeData(previousResult.dataId);\n    }\n  }\n\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReshapePackedProgram} from '../reshape_packed_gpu';\nimport {getBatchDim, getRowsCols} from '../webgl_util';\n\nexport function packedReshape(\n    input: TensorInfo, afterShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const input3DShape =\n      [getBatchDim(input.shape),\n       ...getRowsCols(input.shape)] as [number, number, number];\n  const input3D: TensorInfo = {\n    dtype: input.dtype,\n    shape: input3DShape,\n    dataId: input.dataId\n  };\n  const afterShapeAs3D =\n      [getBatchDim(afterShape),\n       ...getRowsCols(afterShape)] as [number, number, number];\n\n  const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n  const preventEagerUnpackingOfOutput = true;\n  const output = backend.runWebGLProgram(\n      program, [input3D], input.dtype, null /* customSetup */,\n      preventEagerUnpackingOfOutput);\n  return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {packedReshape} from '../kernel_utils/reshape';\nimport {isReshapeFree} from '../webgl_util';\n\nexport function reshape(args: {\n  inputs: ReshapeInputs,\n  backend: MathBackendWebGL,\n  attrs: ReshapeAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n  const webglBackend = backend;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  const xTexData = webglBackend.texData.get(x.dataId);\n  if (xTexData.isPacked && !isReshapeFree(x.shape, $shape) &&\n      !(xTexData.texture !== null && isReshapeFree(xTexData.shape, $shape))) {\n    return packedReshape(x, $shape, webglBackend);\n  }\n\n  webglBackend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'webgl',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {reduce} from '../kernel_utils/reduce';\nimport {reshape} from '../kernels/Reshape';\n\nexport function maxImpl(\n    x: TensorInfo, reduceShape: number[], outShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const inSize = util.sizeFromShape(reduceShape);\n  const xSize = util.sizeFromShape(x.shape);\n  const batchSize = xSize / inSize;\n  const reshapedInput =\n      reshape({inputs: {x}, attrs: {shape: [batchSize, inSize]}, backend});\n\n  const reduced = reduce(reshapedInput, x.dtype, 'max', backend);\n  const reshapedOutput =\n      reshape({inputs: {x: reduced}, attrs: {shape: outShape}, backend});\n\n  backend.disposeIntermediateTensorInfo(reshapedInput);\n  backend.disposeIntermediateTensorInfo(reduced);\n\n  return reshapedOutput;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposeProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const switched = getSwitchedCoords(newDim);\n\n    this.userCode = `\n    void main() {\n      ${dtype} resRC = getOutputCoords();\n      setOutput(getA(${switched}));\n    }\n    `;\n  }\n}\n\nfunction getSwitchedCoords(newDim: number[]): string {\n  const rank = newDim.length;\n  if (rank > 6) {\n    throw Error(`Transpose for rank ${rank} is not yet supported`);\n  }\n  const originalOrder =\n      ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u', 'resRC.v'];\n  const switchedCoords = new Array(rank);\n  for (let i = 0; i < newDim.length; i++) {\n    switchedCoords[newDim[i]] = originalOrder[i];\n  }\n  return switchedCoords.join();\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getVecChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    if (this.rank > 6) {\n      throw Error(\n          `Packed transpose for rank ${this.rank} is not yet supported.`);\n    }\n    const dtype = getCoordsDataType(this.rank);\n\n    const outputOrder = getVecChannels('rc', this.rank);\n    const switchedOrder = new Array(this.rank);\n    for (let i = 0; i < newDim.length; i++) {\n      switchedOrder[newDim[i]] = outputOrder[i];\n    }\n    const innerDims = `vec2(${switchedOrder.slice(-2).join()})`;\n    const nextColumn =\n        `++${outputOrder[this.rank - 1]} < ${outputShape[this.rank - 1]}`;\n    const getc = `getChannel(getA(${switchedOrder.join()}), ${innerDims})`;\n\n    this.userCode = `\n    void main() {\n      ${dtype} rc = getOutputCoords();\n      vec4 result = vec4(0.);\n      result[0] = ${getc};\n      if(${nextColumn}) {\n        result[1] = ${getc};\n      }\n      --${outputOrder[this.rank - 1]};\n      if(++${outputOrder[this.rank - 2]} < ${outputShape[this.rank - 2]}) {\n        result[2] = ${getc};\n        if(${nextColumn}) {\n          result[3] = ${getc};\n        }\n      }\n      setOutput(result);\n    }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {transposeImplCPU} from '../kernel_utils/shared';\nimport {TransposeProgram} from '../transpose_gpu';\nimport {TransposePackedProgram} from '../transpose_packed_gpu';\n\nexport function transposeImpl(\n    x: TensorInfo, perm: number[], backend: MathBackendWebGL): TensorInfo {\n  const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n      new TransposePackedProgram(x.shape, perm) :\n      new TransposeProgram(x.shape, perm);\n  return backend.runWebGLProgram(program, [x], x.dtype);\n}\n\nexport {transposeImplCPU};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {maxImplCPU} from '../kernel_utils/shared';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl, transposeImplCPU} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    const maxInputIsTransposed = permutedAxes != null;\n    const shouldExecuteOnCPU = webglBackend.shouldExecuteOnCPU([x]);\n\n    let maxInput = x;\n    if (maxInputIsTransposed) {\n      if (shouldExecuteOnCPU) {\n        const xTexData = webglBackend.texData.get(maxInput.dataId);\n        const values = xTexData.values as TypedArray;\n\n        const newShape: number[] = new Array(xRank);\n        for (let i = 0; i < newShape.length; i++) {\n          newShape[i] = x.shape[permutedAxes[i]];\n        }\n        const maxInputValues =\n            transposeImplCPU(values, x.shape, x.dtype, permutedAxes, newShape);\n\n        maxInput = webglBackend.makeTensorInfo(newShape, x.dtype);\n        const maxInputData = webglBackend.texData.get(maxInput.dataId);\n        maxInputData.values = maxInputValues;\n      } else {\n        maxInput = transposeImpl(x, permutedAxes, webglBackend);\n      }\n\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n    }\n\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(maxInput.shape, axes);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // rather than reshape at the end, set the target shape here.\n      outShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n    }\n\n    let out;\n    if (shouldExecuteOnCPU) {\n      const xTexData = webglBackend.texData.get(maxInput.dataId);\n      const values = xTexData.values as TypedArray;\n\n      const outValues = maxImplCPU(\n          values, util.sizeFromShape(reduceShape), outShape, x.dtype);\n\n      out = webglBackend.makeTensorInfo(outShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = maxImpl(maxInput, reduceShape, outShape, webglBackend);\n    }\n\n    if (maxInputIsTransposed) {\n      webglBackend.disposeIntermediateTensorInfo(maxInput);\n    }\n\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function maxPool(args: {\n  inputs: MaxPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const maxPoolProgram = new Pool2DProgram(convInfo, 'max', false);\n  return backend.runWebGLProgram(maxPoolProgram, [x], x.dtype);\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'webgl',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {MaxPool2DBackpropProgram} from '../max_pool_backprop_gpu';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const getPositions = true;\n  const maxPoolPositionsProgram =\n      new Pool2DProgram(convInfo, 'max', getPositions);\n  const maxPoolPositions: TensorInfo =\n      backend.runWebGLProgram(maxPoolPositionsProgram, [x], x.dtype);\n\n  const maxPoolBackPropProgram = new MaxPool2DBackpropProgram(convInfo);\n  const result = backend.runWebGLProgram(\n      maxPoolBackPropProgram, [dy, maxPoolPositions], x.dtype);\n  backend.disposeIntermediateTensorInfo(maxPoolPositions);\n  return result;\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\n\nexport function maxPoolWithArgmaxImpl(\n    x: TensorInfo, includeBatchInIndex: boolean,\n    convInfo: backend_util.Conv2DInfo,\n    backend: MathBackendWebGL): TensorInfo[] {\n  let program = new Pool2DProgram(convInfo, 'max', false);\n  const poolOutput = backend.runWebGLProgram(program, [x], 'float32');\n\n  program = new Pool2DProgram(convInfo, 'max', true, true, includeBatchInIndex);\n  const indexOutput = backend.runWebGLProgram(program, [x], 'float32');\n  return [poolOutput, indexOutput];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    util.assert(\n        x.shape.length === 4,\n        () => `Error in maxPool: input must be rank 4 but got rank ${\n            x.shape.length}.`);\n    const dilations: [number, number] = [1, 1];\n    util.assert(\n        backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n            `Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        dilations, pad);\n\n    const [result, indexes] =\n        maxPoolWithArgmaxImpl(x, includeBatchInIndex, convInfo, webglBackend);\n    return [result, indexes];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV3, NonMaxSuppressionV3Attrs, NonMaxSuppressionV3Inputs, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV3Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV3,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV3Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold} =\n        attrs as unknown as NonMaxSuppressionV3Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n\n    return kernel_impls.nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal);\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs, TypedArray} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class RotateProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], radians: number,\n      fillValue: number|[number, number, number],\n      center: number|[number, number]) {\n    const imageHeight = imageShape[1];\n    const imageWidth = imageShape[2];\n    const sinFactor = Math.sin(radians).toFixed(3);\n    const cosFactor = Math.cos(radians).toFixed(3);\n    this.outputShape = imageShape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const centerXString = centerX.toFixed(3);\n    const centerYString = centerY.toFixed(3);\n\n    let fillSnippet = '';\n    if (typeof fillValue === 'number') {\n      fillSnippet = `float outputValue = ${fillValue.toFixed(2)};`;\n    } else {\n      fillSnippet = `\n        vec3 fill = vec3(${fillValue.join(',')});\n        float outputValue = fill[coords[3]];`;\n    }\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n          int y = coords[1];\n          float coordXFloat = (float(x) - ${centerXString}) * ${\n        cosFactor} - (float(y) - ${centerYString}) * ${sinFactor};\n          float coordYFloat = (float(x) - ${centerXString}) * ${\n        sinFactor} + (float(y) - ${centerYString}) * ${cosFactor};\n          int coordX = int(round(coordXFloat + ${centerXString}));\n          int coordY = int(round(coordYFloat + ${centerYString}));\n          ${fillSnippet}\n          if(coordX >= 0 && coordX < ${imageWidth} && coordY >= 0 && coordY < ${\n        imageHeight}) {\n            outputValue = getImage(coords[0], coordY, coordX, coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {RotateProgram} from '../rotate_gpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new RotateProgram(\n        (image as Tensor4D).shape, radians, fillValue, center);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SIN = CHECK_NAN_SNIPPET_UNARY + `\n  return sin(x);\n`;\n\nexport const sinKernelFunc = unaryKernelFunc(SIN);\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'webgl',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Square} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARE = `return x * x;`;\n\nexport const squareKernelFunc = unaryKernelFunc(SQUARE);\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'webgl',\n  kernelFunc: squareKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const squaredDifferenceKernelFunc =\n    binaryKernelFunc(SQUARED_DIFFERENCE, SQUARED_DIFFERENCE);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'webgl',\n  kernelFunc: squaredDifferenceKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst TAN = `return tan(x);`;\n\nexport const tanKernelFunc = unaryKernelFunc(TAN);\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'webgl',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {transposeImpl} from './Transpose_impl';\nimport {transposeImplCPU as cpuTranspose} from './Transpose_impl';\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as TransposeInputs;\n    const {perm} = attrs as {} as TransposeAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const newShape: number[] = new Array(xRank);\n    for (let i = 0; i < newShape.length; i++) {\n      newShape[i] = x.shape[perm[i]];\n    }\n\n    let out: TensorInfo;\n    if (webglBackend.shouldExecuteOnCPU([x])) {\n      const xTexData = webglBackend.texData.get(x.dataId);\n      const values = xTexData.values as TypedArray;\n      const outValues = cpuTranspose(values, x.shape, x.dtype, perm, newShape);\n\n      out = webglBackend.makeTensorInfo(newShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = transposeImpl(x, perm, webglBackend);\n    }\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {uniqueImplCPU} from '../kernel_utils/shared';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function unique(\n    args:\n        {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendWebGL}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  // For now, always forward calculation to the CPU backend.\n  console.warn(\n      'WARNING: ',\n      'UI might be locked temporarily as data is being downloaded');\n  const values = backend.readSync(x.dataId);\n  const {outputValues, outputShape, indices} =\n      uniqueImplCPU(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'webgl',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {atan2Config} from './kernels/Atan2';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {cosConfig} from './kernels/Cos';\nimport {divConfig} from './kernels/Div';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {fromPixelsConfig} from './kernels/FromPixels';\nimport {identityConfig} from './kernels/Identity';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {nonMaxSuppressionV3Config} from './kernels/NonMaxSuppressionV3';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {sinConfig} from './kernels/Sin';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {tanConfig} from './kernels/Tan';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  atan2Config,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  cosConfig,\n  divConfig,\n  flipLeftRightConfig,\n  fromPixelsConfig,\n  identityConfig,\n  maxConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  nonMaxSuppressionV3Config,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  sinConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  tanConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport * from '@tensorflow/tfjs-core';\nexport * from '@tensorflow/tfjs-layers';\nexport * from '@tensorflow/tfjs-converter';\n\n// Export data api as tf.data\nimport * as data from '@tensorflow/tfjs-data';\nexport {data};\n\n// Import and register backends.\nimport '@tensorflow/tfjs-backend-cpu';\nimport '@tensorflow/tfjs-backend-webgl';\n\n// Import versions of all sub-packages.\nimport {version_core} from '@tensorflow/tfjs-core';\nimport {version_cpu} from '@tensorflow/tfjs-backend-cpu';\nimport {version_webgl} from '@tensorflow/tfjs-backend-webgl';\nimport {version_data} from '@tensorflow/tfjs-data';\nimport {version_layers} from '@tensorflow/tfjs-layers';\nimport {version_converter} from '@tensorflow/tfjs-converter';\nimport {version as version_union} from './version';\n\nexport const version = {\n  'tfjs-core': version_core,\n  'tfjs-backend-cpu': version_cpu,\n  'tfjs-backend-webgl': version_webgl,\n  'tfjs-data': version_data,\n  'tfjs-layers': version_layers,\n  'tfjs-converter': version_converter,\n  'tfjs': version_union\n};\n", "const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.age.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    obj.gender = Math.trunc(100 * data[0]) < 50 ? 'female' : 'male';\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "export default {\n  backend: 'webgl',\n  console: true,\n  face: {\n    enabled: true, // refers to detector, but since all other face modules rely on detector, it should be a global\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be blazeface-front or blazeface-back\n      inputSize: 256, // fixed value: 128 for front and tfhub and 256 for back\n      maxFaces: 10, // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10, // how many frames to go without running the bounding box detector\n      minConfidence: 0.5, // threshold for discarding a prediction\n      iouThreshold: 0.3, // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7, // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192, // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3, // empiric tuning\n      inputSize: 64, // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json',\n      inputSize: 64, // fixed value\n      skipFrames: 10,\n    },\n    gender: {\n      enabled: true,\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64, // fixed value\n      minConfidence: 0.5,\n      skipFrames: 10,\n      useGrayscale: true,\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257, // fixed value\n    outputStride: 16, // fixed value\n    maxDetections: 5,\n    scoreThreshold: 0.7,\n    nmsRadius: 20,\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256, // fixed value\n    skipFrames: 10,\n    minConfidence: 0.5,\n    iouThreshold: 0.3,\n    scoreThreshold: 0.7,\n    enlargeFactor: 1.65, // empiric tuning\n    maxHands: 2,\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    let savedWebglPackDepthwiseConvFlag;\n    if (tf.getBackend() === 'webgl') {\n      savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n      tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
-  "mappings": "oMAAA,kBCAA,kBCAA,0sDCAA,AAsBO,GAAM,IAAkB,KAClB,GAAkB,mBA0B7B,WAAoB,EAAgC,GAAhC,KAAA,QAAA,EAAgC,KAAA,UAAA,EAH5C,KAAA,KAAO,GAAI,SACX,KAAA,aAAe,EAIvB,SAAA,UAAA,IAAA,SAAI,GACF,MAAK,MAAK,KAAK,IAAI,IACjB,KAAK,UAAU,SAAS,KAAK,QAAS,GAEjC,KAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,IAAA,SAAI,EAAgB,GAClB,KAAK,eACL,KAAK,KAAK,IAAI,EAAQ,IAGxB,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,MAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,OAAA,SAAO,GACL,YAAK,eACE,KAAK,KAAK,OAAO,IAG1B,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,cAEhB,mBAqBA,cACE,SAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,WACE,MAAO,GAAkB,eAE3B,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,GAAkB,gBAE3B,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAC5C,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAE3D,MAAO,GAAkB,SAE3B,EAAA,UAAA,OAAA,WACE,MAAO,GAAkB,WAG3B,EAAA,UAAA,eAAA,WACE,MAAO,GAAkB,mBAG3B,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAG1D,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAEnD,MAAO,GAAkB,qBAG3B,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,MAAO,GAAkB,UAE3B,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,MAAO,GAAkB,iBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,GAAkB,WAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAG3B,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,uBAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAE3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAG3B,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,MAAO,GAAkB,iBAG3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,MAAO,GAAkB,WAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,MAAO,GAAkB,sBAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,GAAkB,WAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,MAAO,GAAkB,WAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAG3B,EAAA,UAAA,QAAA,SAA0B,GACxB,MAAO,GAAkB,YAG3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAG3B,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,yBAG3B,EAAA,UAAA,gBAAA,SAAgB,EAAiB,EAAkB,GAEjD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,wBAAA,SAAwB,EAAc,EAAkB,GAEtD,MAAO,GAAkB,4BAE3B,EAAA,UAAA,yBAAA,SAAyB,EAAa,EAAc,GAElD,MAAO,GAAkB,6BAE3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,EAAa,GAEtD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SAAkB,EAAc,EAAa,GAC3C,MAAO,GAAkB,sBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAAa,GAC1C,MAAO,GAAkB,sBAG3B,EAAA,UAAA,QAAA,SAA0C,EAAM,GAE9C,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,MAAO,GAAkB,QAG3B,EAAA,UAAA,UAAA,SAA4B,EAAM,GAChC,MAAO,GAAkB,cAG3B,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,MAAO,GAAkB,WAG3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GACpC,MAAO,GAAkB,cAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,mBAG3B,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,MAAO,GAAkB,2BAG3B,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,0BAG3B,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,MAAO,GAAkB,kCAG3B,EAAA,UAAA,UAAA,SACI,EAAa,EAAyB,EACtC,EAA4B,EAC5B,GACF,MAAO,GAAkB,cAG3B,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,MAAO,GAAkB,iCAG3B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EAAuB,EAC3D,EAAc,EAAe,GAC/B,MAAO,GAAkB,YAG3B,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,MAAO,GAAkB,WAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,MAAO,GAAkB,WAG3B,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,MAAO,GAAkB,sBAG3B,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAC3C,MAAO,GAAkB,iBAI3B,EAAA,UAAA,MAAA,SAAwB,EAAU,EAAsB,GACtD,MAAO,GAAkB,UAG3B,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,MAAO,GAAkB,SAG3B,EAAA,UAAA,SAAA,SAAyB,GACvB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,GAAkB,cAG3B,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAkB,aAG3B,EAAA,UAAA,QAAA,WACE,MAAO,GAAkB,YAE7B,KAEA,WAA2B,GACzB,KAAM,IAAI,OACN,IAAI,EAAU,4HCzpBpB,AAoBA,GAAM,IAA4B,0BA6BhC,WAAmB,GAAA,KAAA,OAAA,EATX,KAAA,MAAe,GACf,KAAA,aAAwD,GAExD,KAAA,SAAkB,GAOxB,KAAK,mBAGP,SAAA,UAAA,YAAA,SAAY,EAAsB,GAChC,AAAI,KAAK,UAAY,MACnB,QAAQ,KACJ,YAAY,KAAK,aAAY,0BAC7B,kCAAiC,EAAQ,MAE/C,KAAK,aAAe,EACpB,KAAK,SAAW,GAGlB,EAAA,UAAA,aAAA,SACI,EAAkB,EAClB,GAKF,GAJA,KAAK,aAAa,GAAY,CAAC,aAAY,EAAE,QAAO,GAIhD,KAAK,SAAS,IAAa,MAC7B,GAAM,GAAY,KAAK,SAAS,GAChC,QAAQ,KACJ,qCAAqC,EAAQ,KAAK,EAAS,KAC/D,KAAK,IAAI,EAAU,KAIjB,EAAA,UAAA,SAAN,SAAe,qGACb,MAAI,KAAY,MAAK,MACnB,CAAA,EAAO,KAAK,MAAM,IAGpB,GAAA,KAAK,MAAM,EAAA,EAAY,CAAA,EAAM,KAAK,aAAa,YAA/C,SAAA,GAAuB,EAAA,OACvB,CAAA,EAAO,KAAK,MAAM,UAGpB,EAAA,UAAA,IAAA,SAAI,GACF,GAAI,IAAY,MAAK,MACnB,MAAO,MAAK,MAAM,GAGpB,GAAM,GAAY,KAAK,aAAa,GACpC,GAAI,YAAqB,SACvB,KAAM,IAAI,OACN,QAAQ,EAAQ,sEAItB,YAAK,MAAM,GAAY,EAEhB,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,QAAA,SAAQ,GACN,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,OAGd,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,uCAGd,EAAA,UAAA,IAAA,SAAI,EAAkB,GACpB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,mBAAmB,EAAQ,mCAEjC,KAAK,MAAM,GAAY,EACvB,AAAI,KAAK,aAAa,GAAU,SAAW,MACzC,KAAK,aAAa,GAAU,QAAQ,IAIhC,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,yBAAyB,EAAQ,oCAEvC,MAAO,MAAK,aAAa,GAAU,gBAGrC,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,MAAQ,OAAO,OAAO,GAAI,IAGjC,EAAA,UAAA,MAAA,WACE,KAAK,MAAQ,GACb,KAAK,SAAW,GAChB,KAAK,oBAGC,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,GAAI,MAAO,MAAK,QAAW,aACvB,MAAO,MAAK,OAAO,UAAa,aAChC,MAAO,MAAK,OAAO,SAAS,QAAW,YACzC,OAGF,GAAM,GAAY,GAAe,KAAK,OAAO,SAAS,QACtD,GAAI,KAA6B,IAC/B,GAAM,GAAY,EAAU,IAA2B,MAAM,KAC7D,EAAU,QAAQ,SAAA,GACV,GAAA,GAAA,EAAA,MAAA,KAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,SAAS,GAAO,GAAW,EAAK,OAI7C,iBAE+B,GAC7B,GAAM,GAAS,GACf,SAAY,QAAQ,8BAA+B,SAAC,UAAG,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GACrD,UAAY,EAAQ,EAAE,GAAI,EAAE,IACrB,EAAE,KAAK,OAET,EAGT,YACI,EAAiC,EAAc,GACjD,EAAO,mBAAmB,IAAS,mBAAmB,GAAS,IAGjE,YAAoB,EAAkB,GAEpC,GADA,EAAQ,EAAM,cACV,IAAU,QAAU,IAAU,QAChC,MAAO,KAAU,OACZ,GAAI,GAAG,EAAE,IAAY,EAC1B,MAAO,CAAC,EAEV,KAAM,IAAI,OACN,oCAAoC,EAAK,aAAa,EAAQ,mBAYlE,MAAO,GAAA,IAGE,EAAA,IAAmB,iBACO,GACnC,EAAA,IAAM,EC9MR,AAqBA,GAAI,kBAGF,GAAI,IAAmB,MAErB,GAAI,GAAE,OACN,GAAI,MAAQ,SAAY,YACtB,EAAK,eACI,MAAQ,SAAY,YAC7B,EAAK,eACI,MAAQ,UAAa,YAC9B,EAAK,gBACI,MAAQ,OAAU,YAC3B,EAAK,SAEL,MAAM,IAAI,OAAM,kCAElB,GAAkB,EAEpB,MAAO,IAIT,cACE,GAAM,GAAK,KACX,MAAI,GAAG,YAAc,MACnB,GAAG,WAAa,GAAI,MAEf,EAAG,uBAUiB,EAAa,GACxC,GAAM,GAAY,KAClB,GAAI,EAAU,IAAI,GAChB,MAAO,GAAU,IAAI,GAErB,GAAM,GAAY,IAClB,SAAU,IAAI,EAAK,GACZ,EAAU,IAAI,MCzCZ,IAAM,MAGN,GAAO,OAGP,GAAQ,QAGR,GAAM,MAGN,GAAO,OAGP,GAAM,MAON,GAAM,MAON,GAAS,SAMT,GAAS,SAMT,GAAO,OAGP,GAAQ,QAGR,GAAO,OAGP,GAAQ,QAGR,GAAQ,QAGR,GAAU,UASV,GAAkB,kBAQlB,GAAY,YAWZ,GAAoB,oBAUpB,GAAc,cAOd,GAAiB,iBASjB,GAAc,cAOd,GAAO,OAMP,GAAO,OAGP,GAAc,cAOd,GAAU,UAGV,GAAS,SAMT,GAAS,SAUT,GAAuB,uBASvB,GAAsB,sBAUtB,GAAS,SAST,GAAyB,yBAQzB,GAAwB,wBAMxB,GAAM,MAGN,GAAO,OAGP,GAAS,SAQT,GAAgB,gBAShB,GAAe,eAOf,GAAwB,wBAWxB,GACT,sCAIS,GACT,qCAIS,GAAO,OAGP,GAAa,aAQb,GAA0B,0BAI1B,GAA2B,2BAI3B,GAAM,MAGN,GAAM,MAGN,GAAU,UAGV,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAO,OAOP,GAAgB,gBAGhB,GAAQ,QAGR,GAAW,WAGX,GAAiB,iBAOjB,GAAW,WAMX,GAAW,WAGX,GAAU,UAGV,GAAe,eAGf,GAAW,WAGX,GAAO,OAGP,GAAO,OAGP,GAAW,WAGX,GAAQ,QAGR,GAAQ,QAGR,GAAO,OAGP,GAAY,YAGZ,GAAW,WAMX,GAAM,MAGN,GAAQ,QAGR,GAAa,aAGb,GAAa,aAGb,GAAY,YAGZ,GAAa,aAMb,GAAM,MASN,GAAc,cASd,GAAM,MAON,GAAU,UAGV,GAAU,UASV,GAAkB,kBAUlB,GAAY,YAWZ,GAAoB,oBAWpB,GAAoB,oBASpB,GAAO,OAOP,GAAM,MAON,GAAU,UAGV,GAAM,MAGN,GAAW,WAGX,GAAS,SAGT,GAAW,WAGX,GAAsB,sBAStB,GAAsB,sBAUtB,GAAsB,sBAUtB,GAAW,WAGX,GAAS,SAQT,GAAQ,QAOR,GAAO,OAGP,GAAM,MAGN,GAAQ,QAGR,GAAO,OAOP,GAAQ,QAQR,GAAO,OAGP,GAAa,aAGb,GAAO,OAGP,GAAU,UAMV,GAAwB,wBAOxB,GAA4B,4BAI5B,GAAiB,iBAOjB,GAAqB,qBAGrB,GAAQ,QAGR,GAAU,UAMV,GAAQ,QAGR,GAAQ,QAGR,GAAY,YAMZ,GAAW,WAGX,GAAO,OAGP,GAAQ,QAMR,GAAM,MAGN,GAAO,OAGP,GAAO,OAGP,GAAU,UAGV,GAAW,WAGX,GAAO,OAGP,GAAM,MAON,GAAiB,iBAOjB,GAAS,SAOT,GAAU,UAMV,GAAoB,oBAGpB,GAAS,SAGT,GAAM,MAGN,GAAgB,gBAOhB,GAAe,eAaf,GAAM,MAGN,GAAO,OAGP,GAAO,OAMP,GAAO,OAOP,GAAY,YAMZ,GAAS,SAQT,GAAS,SAMT,GAAqB,qBAOrB,GAAY,YAMZ,GAAO,OAMP,GAAa,aASb,GAAmB,mBAQnB,GAAe,eAef,GAAc,cAgBd,GAAuB,uBC7zBpC,AAuBA,GAAM,IACF,GAAU,iBAAkB,WAAM,MAAA,IAAI,OACpC,GACF,GAAU,eAAgB,WAAM,MAAA,IAAI,mBAqEpC,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,MAAO,IAAe,IAAI,eAOA,GAC1B,MAAO,IAAa,IAAI,eAGW,GAInC,OAHM,GAAK,GAAe,UACpB,EAAyB,MAGvB,GAAA,GAAA,EAAA,OAAC,EAAA,EAAA,KAAM,EAAA,EAAA,MACb,GAAI,EACF,MAEK,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GACL,EAAA,EAAA,MAAA,KAAA,GACP,AAAI,IAAY,GACd,EAAO,KAAK,GAGhB,MAAO,eAcsB,GACtB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,YACb,EAAM,GAAQ,EAAY,GAChC,AAAI,GAAe,IAAI,IACrB,QAAQ,KACJ,eAAe,EAAU,iBACzB,KAAI,EAAW,4BAErB,GAAe,IAAI,EAAK,eAWO,GACxB,GAAA,GAAA,EAAA,WAEP,AAAI,GAAa,IAAI,IAGnB,CAAI,KAAM,QAAQ,UAChB,QAAQ,KAAK,gCAAgC,EAAU,MAG3D,GAAa,IAAI,EAAY,eAW3B,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,GAAI,CAAC,GAAe,IAAI,GACtB,KAAM,IAAI,OACN,eAAe,EAAU,iBACzB,KAAI,EAAW,wBAErB,GAAe,OAAO,eAIW,GACjC,GAAI,CAAC,GAAa,IAAI,GACpB,KAAM,IAAI,OACN,iBAAiB,EAAU,mCAEjC,GAAa,OAAO,eAUlB,EAA+B,GACjC,GAAM,GAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,GAAM,GACF,OAAO,OAAO,GAAI,EAAc,CAAC,YAAa,IAClD,GAAe,KAInB,YAAiB,EAAoB,GACnC,MAAU,GAAW,IAAI,ECrN3B,YAkCwB,GAMtB,OAJI,GAAU,EAAM,OAChB,EAAO,EACP,EAAQ,EAEL,EAAU,GAEf,EAAS,KAAK,SAAW,EAAW,EAEpC,IAEA,EAAO,EAAM,GACb,EAAM,GAAW,EAAM,GACvB,EAAM,GAAS,cAKG,EAAa,EAAW,GAC5C,MAAO,MAAK,IAAI,EAAK,KAAK,IAAI,EAAG,gBAGD,GAChC,MAAO,GAAM,IAAM,EAAI,EAAM,EAAM,cAGjB,GAElB,OADI,GAAM,EACD,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,GAAO,EAAI,GAEb,MAAO,eAUmB,EAAW,GACrC,GAAM,GAAI,KAAK,SACf,MAAQ,GAAI,EAAM,GAAI,GAAK,cAID,EAAe,GAEzC,OADI,GAAS,EACJ,EAAI,EAAG,EAAI,EAAE,OAAQ,KAC5B,GAAM,GAAO,OAAO,EAAE,IAAM,OAAO,EAAE,IACrC,GAAU,EAAO,EAEnB,MAAO,cAkBc,EAAe,GACpC,GAAI,CAAC,EACH,KAAM,IAAI,OAAM,MAAO,IAAQ,SAAW,EAAM,iBAKhD,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EACI,GAAY,EAAQ,GACpB,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlC,GAC5B,EACI,GAAK,KACL,WAAM,MAAA,8EAwBR,EAA0B,EAAkB,GAI9C,GAJ4B,AAAA,IAAA,QAAA,GAAA,IAAkB,AAAA,IAAA,QAAA,GAAA,IAC9C,AAAI,GAAU,MACZ,GAAS,IAEP,MAAM,QAAQ,IAAQ,GAAa,IAAQ,CAAC,EAC9C,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAAQ,EAAI,GAAI,EAAQ,OAG1B,GAAO,KAAK,GAEd,MAAO,eAcqB,GAC5B,GAAI,EAAM,SAAW,EAEnB,MAAO,GAGT,OADI,GAAO,EAAM,GACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,GAAQ,EAAM,GAEhB,MAAO,eAGqB,GAC5B,MAAO,GAAM,SAAW,cAGE,EAAgB,GAC1C,GAAI,IAAO,EACT,MAAO,GAET,GAAI,GAAM,MAAQ,GAAM,KACtB,MAAO,GAGT,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,EAAG,GACf,MAAO,GAGX,MAAO,eAGa,GACpB,MAAO,GAAI,IAAM,cAGE,GAEnB,GAAK,KAAa,MAAQ,KAExB,MAAQ,MAAa,KAAK,GAE5B,GAAI,IAAM,SACR,MAAO,GACF,GAAI,IAAM,UACf,MAAO,GAEP,GAAM,GAAM,KAAK,IAAI,EAAI,GACzB,MAAQ,GAAM,GAAM,GAAM,eAIM,GAClC,GAAM,GAAQ,KAAK,KAAK,KAAK,KAAK,IAClC,MAAO,CAAC,EAAO,KAAK,KAAK,EAAO,gBAeI,GAEpC,OADM,GAAkB,GAAI,aAAY,GAC/B,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAgB,GAAK,EAEvB,UAAQ,GACD,cAGgB,EAAW,GAClC,MAAI,IAAQ,EAAE,OACL,EAEF,EAAI,IAAI,OAAO,EAAO,EAAE,oBAI7B,EAAwB,EACxB,GADwB,MAAA,KAAA,QAAA,GAAA,SAAW,GAAoB,MAAA,KAElD,GAAI,SAAc,SAAC,EAAS,GACjC,GAAI,GAAW,EAET,EAAQ,WACZ,GAAI,KACF,IACA,OAGF,IAEA,GAAM,GAAc,EAAQ,GAE5B,GAAI,GAAc,MAAQ,GAAY,GACpC,IACA,OAEF,WAAW,EAAO,IAGpB,kBAcA,EAAiB,GAInB,OAHI,GAAY,EACZ,EAAc,GAET,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,GAAI,EAAM,IAAM,EACd,GAAa,EAAM,WACV,EAAM,KAAO,IACtB,GAAI,IAAgB,GAClB,KAAM,OACF,yCACA,oBAAmB,EAAW,YAAY,IAEhD,EAAc,UACL,EAAM,GAAK,EACpB,KAAM,OAAM,gCAAgC,EAAM,GAAE,WAAW,GAInE,GAAI,IAAgB,IAClB,GAAI,EAAO,GAAK,IAAS,EACvB,KAAM,OAAM,QAAQ,EAAI,qCAAqC,GAE/D,MAAO,GAGT,GAAI,IAAc,EAChB,KAAM,OACF,qCAAqC,EAAK,+BAGhD,GAAI,EAAO,IAAc,EACvB,KAAM,OACF,oDACA,QAAO,EAAI,MAAM,IAGvB,GAAM,GAAW,EAAM,QACvB,SAAS,GAAe,EAAO,EACxB,cAIL,EAAuB,GACzB,GAAM,GAAO,EAAM,OAGnB,SAAO,GAAQ,KAAO,EAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAK,GAAG,OAAO,GAGzD,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,CAAC,GAAQ,EAAK,IACrC,WACI,MAAA,+CAA+C,EAAI,KAAK,EAAI,SAC5D,aAAY,KAGpB,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,KACvB,WAAM,MAAA,iDACF,aAAY,KAGb,EAAK,IAAI,SAAA,GAAK,MAAA,GAAI,EAAI,EAAO,EAAI,gBAIb,EAAiB,GAS5C,OAPM,GAAqB,GACrB,EAAqB,GACrB,EAAe,GAAQ,MAAQ,MAAM,QAAQ,IAAS,EAAK,SAAW,EACtE,EAAQ,GAAQ,MAAQ,EAC1B,KACA,GAAe,EAAM,GAAO,OAC5B,EAAI,EACC,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,GAAQ,MACV,GAAI,EAAK,KAAO,GAAK,EAAM,KAAO,EAChC,KAAM,IAAI,OACN,sBAAsB,EAAC,mBAAmB,EAAM,GAAE,cAExD,AAAK,GAAK,IAAM,MAAQ,EAAK,GAAK,IAAM,EAAM,KAAO,GACnD,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAEhB,AAAI,EAAK,IAAM,GACb,IAGJ,AAAI,EAAM,KAAO,GACf,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAGlB,MAAO,CAAC,SAAQ,EAAE,SAAQ,eAIxB,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,SACnB,EAAS,GAAI,OAAgB,OAE7B,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAA+B,GACjC,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAC1B,KAAM,OAAM,oBAAoB,EAAK,4BAA4B,EAAG,kBAM7C,GAC3B,MAAO,KAAU,QAAU,IAAU,aAAe,IAAU,WAC1D,IAAU,SAAW,IAAU,qBAOL,EAAmB,GACjD,MAAI,KAAY,aAGZ,KAAY,WAAa,IAAY,aAGrC,IAAY,SAAW,IAAY,WAAa,IAAY,aALvD,GAQL,MAAY,QAAU,IAAY,oBAMX,GAC3B,MAAO,aAAa,eAAgB,YAAa,aAC7C,YAAa,wBAGa,GAC9B,GAAI,IAAU,WAAa,IAAU,QACnC,MAAO,GACF,GAAI,IAAU,YACnB,MAAO,GACF,GAAI,IAAU,OACnB,MAAO,GAEP,KAAM,IAAI,OAAM,iBAAiB,eAUA,GACnC,GAAI,GAAO,KACT,MAAO,GAET,GAAI,GAAQ,EACZ,SAAI,QAAQ,SAAA,GAAK,MAAA,IAAS,EAAE,SACrB,cAIgB,GACvB,MAAO,OAAO,IAAU,UAAY,YAAiB,oBAG7B,GACxB,MAAO,OAAO,IAAU,sBAGD,GACvB,MAAO,OAAO,IAAU,qBAGC,GACzB,MAAI,OAAM,QAAQ,GACT,GAAW,EAAO,IAEvB,YAAkB,cACb,UACE,YAAkB,aAAc,YAAkB,YACpD,QACE,GAAS,GACX,UACE,GAAS,GACX,SACE,GAAU,GACZ,OAEF,sBAGkB,GACzB,MAAO,CAAC,CAAE,IAAK,EAAE,aAAe,EAAE,MAAQ,EAAE,mBAGf,EAAc,GAC3C,OAAS,GAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,GAAI,EAAO,IAAM,EACf,MAAO,GAGX,MAAO,eAGsB,GAC7B,GAAM,GAAO,EAAM,OACnB,GAAI,EAAO,EACT,MAAO,GAKT,GAAM,GAAU,GAAI,OAAM,EAAO,GACjC,EAAQ,EAAO,GAAK,EAAM,EAAO,GACjC,OAAS,GAAI,EAAO,EAAG,GAAK,EAAG,EAAE,EAC/B,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAM,EAAI,GAE1C,MAAO,eAOL,EAAiB,GACnB,MAAI,KAAU,SACL,GAAa,GAGf,GAAa,CAAC,GAAQ,eAGF,EAAe,GAC1C,GAAI,IAAU,SACZ,KAAM,IAAI,OAAM,6CASlB,GAPA,AAAI,MAAM,QAAQ,IAChB,GAAI,GAAQ,IAGd,AAAI,KAAM,QAAQ,UAChB,GAAyB,EAAe,GAEtC,GAAmB,EAAG,GACxB,MAAO,GAET,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,QAEnB,OADM,GAAO,GAAI,YAAY,EAAe,QACnC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,AAAI,KAAK,MAAO,EAAe,MAAQ,GACrC,GAAK,GAAK,GAGd,MAAO,OAEP,MAAM,IAAI,OAAM,qBAAqB,GAIzC,YAA2B,EAAgB,EAAiB,GAC1D,GAAM,GAAM,GAAI,OAChB,GAAI,EAAM,SAAW,EAEnB,OADM,GAAI,EAAM,GACP,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,EAAE,EAAS,OAMtB,QAHM,GAAI,EAAM,GACV,EAAO,EAAM,MAAM,GACnB,EAAM,EAAK,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IACjC,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,GAAkB,EAAS,EAAI,EAAK,EAAM,GAGvD,MAAO,eAIqB,EAAiB,GAC7C,GAAI,EAAM,SAAW,EAEnB,MAAO,GAAE,GAEX,GAAM,GAAO,EAAM,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IAC5C,GAAI,IAAS,EAEX,MAAO,GAET,GAAI,IAAS,EAAE,OACb,KAAM,IAAI,OAAM,IAAI,EAAK,mCAAmC,EAAE,OAAM,KAGtE,MAAO,IAAkB,EAAG,EAAO,GAGrC,YAA4B,EAAe,GACzC,MAAQ,aAAa,eAAgB,IAAU,WAC1C,YAAa,aAAc,IAAU,SACrC,YAAa,aAAc,IAAU,mBAIxC,EAAc,GAEhB,OADM,GAAQ,GAAoB,EAAM,GAC/B,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAM,GAAK,EAEb,MAAO,eAIL,EAAc,GAChB,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,OACnB,MAAO,IAAI,YAAW,GAEtB,KAAM,IAAI,OAAM,qBAAqB,eAUrC,EAAiB,GACnB,GAAM,GAAO,EAAM,OAAO,SAAC,EAAM,GAAS,MAAA,GAAO,GAAM,GACvD,GAAI,GAAS,MAAQ,IAAU,UAC7B,MAAO,IAAc,EAAO,GAAI,cAAa,IACxC,GAAI,IAAU,QACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IACtC,GAAI,IAAU,OACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IAE3C,KAAM,IAAI,OAAM,qBAAqB,iBAgBvC,MAAO,MAAM,SAAS,kBAG2B,GACjD,EAAM,QAAQ,SAAA,GACZ,EACI,OAAO,UAAU,IAAY,GAAW,EACxC,WACI,MAAA,mEACA,WAAU,EAAK,sBAqBvB,EAAc,GAChB,MAAO,MAAM,SAAS,MAAM,EAAM,eAWP,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,SACtC,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAG,eAWL,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,SAC9C,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAO,eAYlC,EAAgB,EAAc,GAChC,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,EAAQ,GAAK,EAAK,GAE7B,MAAO,eAYL,EAAe,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,EAAQ,IACrC,GAAS,EAAK,GAAK,EAAQ,GAE7B,SAAK,EAAK,OAAS,GAAK,EACjB,22BC3xBT,AA+BA,GAAA,IAAA,WACE,WAAoB,EAAoC,GAApC,KAAA,aAAA,EAAoC,KAAA,OAAA,EACtD,AAAI,GAAU,MACZ,MAAK,OAAS,GAAI,KAItB,SAAA,UAAA,cAAA,SAAc,EAAoB,EAAwB,GAQxD,OANI,GACE,EAAsB,WAC1B,EAAU,KAEN,EAAQ,KAAK,aAAa,KAAK,cAE5B,GACP,GAAM,GAAS,EAAQ,GAGvB,EAAO,OAAO,KAAK,SAAA,GACjB,GAA0B,EAAY,EAAO,MAAO,MAL/C,EAAI,EAAG,EAAI,EAAQ,OAAQ,MAA3B,GAST,GAAM,GAAgB,CACpB,WAAU,EACV,QAAO,EACP,OAAM,EACN,OAAQ,EAAM,KAAK,SAAA,GAAU,MAAA,GAAO,WACpC,UAAW,EAAM,KACb,SAAA,GAAU,MAAA,GAAO,qBAAuB,KACpC,EAAO,sBACP,MAEV,MAAO,IAGT,EAAA,UAAA,iBAAA,SAAiB,GAAjB,GAAA,GAAA,KACS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,OAAQ,EAAA,EAAA,OAAQ,EAAA,EAAA,UAE5C,EAAQ,QAAQ,SAAA,GACd,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,IAAY,KAAK,SAAA,GACnD,EAAK,OAAO,iBACR,EAAY,EAAQ,EAAe,GAAI,EAAe,GAAI,EAC1D,EAAe,SAI3B,iBAGI,EAAsB,EAAU,GAClC,GAAI,IAAU,UAEZ,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAE1B,eAAQ,KAAK,SAAS,EAAG,sBAAsB,EAAU,KAClD,GAGX,MAAO,GAGT,GAAA,IAAA,WAAA,cACE,SAAA,UAAA,iBAAA,SACI,EAAc,EAAgB,EAC9B,EAAgC,EAChC,GACF,GAAM,GAAO,MAAO,IAAW,SAAW,GAAiB,EAAM,KAAM,GAC7B,EAAO,MAC3C,EAAa,GAAc,EAAM,IACjC,EAAO,EAAO,KACd,EAAO,EAAO,KACd,EAAQ,GAAc,EAAO,MAAM,WAAY,IACjD,EAAyB,GAE7B,OAAW,KAAQ,IACjB,GAAM,GAAQ,EAAO,GACrB,GAAI,GAAS,MAGX,GAAM,GAAa,EAAM,OAAS,EAAO,MACnC,EAAY,EAAW,OAC7B,GACO,EAAI,KAAK,EAAS,KAAK,GAAY,EAAI,EAAa,IAAE,KAIjE,QAAQ,IACJ,KAAK,EAAU,MAAO,EAAI,MAAO,EAAI,KAAK,EAAK,MAAO,EAAI,MACtD,EAAsB,MAAO,EACjC,mBAAoB,YAAa,aAAc,gBAC/C,eAAgB,qBAExB,KCjIA,YA4CI,EAAkB,EAAc,GAKlC,OAFM,GAA8C,GAC9C,EAA0C,GACvC,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,EAAa,EAAG,GAAG,IAAM,GAG3B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GACZ,EAAa,EAAK,OACxB,OAAW,KAAa,IAItB,OAHM,GAAQ,EAAW,GAErB,EAAgB,GACX,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAa,EAAM,KACrB,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAa,EAAO,IAAM,KACzD,EAAgB,GAChB,EAAW,EAAK,IAAM,GACtB,MAIJ,GAAI,EACF,OAMN,GAAM,GAAgD,GACtD,EAAe,EAAE,IAAM,GAGvB,OAFM,GAAwC,GAErC,EAAI,EAAK,OAAS,EAAG,GAAK,EAAG,IAKpC,OAJM,GAAO,EAAK,GACZ,EAAa,EAAK,OAGf,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,IACvC,GAAI,EAAe,EAAK,QAAQ,GAAG,KACjC,OAAW,KAAa,GACtB,EAAe,EAAW,GAAW,IAAM,GAC3C,EAAS,EAAK,IAAM,GAEtB,MAON,OADM,GAA2B,GACxB,EAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GAElB,GAAI,EAAW,EAAK,KAAO,EAAS,EAAK,KAEvC,GAAM,GAA8C,GACpD,OAAW,KAAa,GAAK,QAC3B,GAAM,GAAY,EAAK,OAAO,GAC9B,AAAI,EAAa,EAAU,KACzB,GAAa,GAAa,GAK9B,GAAM,GAAa,OAAO,OAAO,GAAI,GACrC,EAAW,OAAS,EACpB,EAAW,QAAU,EAAK,QAE1B,EAAa,KAAK,IAItB,MAAO,eAWL,EACA,EAA0B,EAC1B,GAEF,mBAAS,GACP,GAAM,GAAO,EAAa,GAEpB,EAAgB,GAYtB,GAXA,EAAK,QAAQ,QAAQ,SAAA,GACnB,GAAM,GAAa,EAA6B,EAAE,IAClD,AAAI,GAAc,KAChB,EAAI,KAAK,GAIT,EAAI,KAAK,QAIT,EAAK,UAAY,KACnB,KAAM,IAAI,OACN,wDACA,QAAO,EAAK,WAAU,MAI5B,GAAM,GAAiB,EAAK,SAAS,cAE1B,GACT,GAAI,CAAE,KAAa,IACjB,KAAM,IAAI,OACN,iCAAiC,EAAS,KAC1C,+BAA8B,OAAO,KAAK,GAAe,MAI/D,GAAM,GAAK,EAAK,WAAM,MAAA,GAAe,OACrC,GAAI,EAAG,QAAU,UACf,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BAChB,GAAS,wCAAwC,EAAG,MAAK,MAElE,GAAM,GAAI,EAAK,OAAO,GACtB,GAAI,CAAC,GAAiB,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BACnB,KAAI,EAAS,gBAAgB,EAAG,MAAK,4BACrC,4BAA2B,EAAE,MAAK,MAGxC,GAAI,EAA6B,EAAE,KAAO,KACxC,EAA6B,EAAE,IAAM,OAErC,GAAM,GAAc,EAA6B,EAAE,IACnD,EAA6B,EAAE,IAAM,EAAI,EAAa,GACtD,EAAY,YA7BhB,OAAW,KAAa,GAAK,SAAlB,IAxBJ,EAAI,EAAa,OAAS,EAAG,GAAK,EAAG,MAArC,GCvIX,AAqBA,GAAM,IAAwB,GAExB,GAA6B,EAE7B,GAAwB,cAG1B,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAU,GAAe,GACzB,EAAY,GAAwB,EAAM,EAAO,EAAO,GACxD,EAAO,EAAM,OACb,EAAY,GAAkB,EAAM,EAAO,EAAO,EAAS,GAC3D,EAAQ,CAAC,UACf,MAAI,IACF,GAAM,KAAK,YAAY,GACvB,EAAM,KAAK,WAAW,GACtB,EAAM,KAAK,aAAa,EAAK,KAC7B,EAAM,KAAK,cAEb,EAAM,KAAK,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,IAAG,KAAK;IACxC,EAAM,KAAK;GAGpB,YACI,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAI,GAAc,GAClB,EAAU,EAAQ,EAAQ,OAAS,GACnC,EAAY,GAAI,OAAM,GAAS,KAAK,GACpC,EAAO,EAAM,OACb,EACF,IAAU,YAAc,GAAoB,GAAQ,EAExD,GAAI,EAAO,EACT,OAAS,GAAM,EAAG,EAAM,EAAI,EAAS,IAEnC,OADM,GAAS,EAAM,EACZ,EAAI,EAAG,EAAI,EAAS,IAC3B,EAAU,GAAK,KAAK,IAChB,EAAU,GACV,GAAY,EAAe,EAAS,GAAI,EAAG,GAAO,QAI5D,MAAO,GAGT,YACI,EAAqC,EAAa,GACpD,GAAI,GACJ,MAAI,OAAM,QAAQ,GAChB,EAAY,WAAW,EAAI,GAAG,QAAQ,KAAuB,MACtD,YAAW,EAAI,GAAG,QAAQ,KAAuB,KACnD,AAAI,GAAS,GAClB,EAAS,IAAI,EAAG,IACX,AAAI,IAAU,OACnB,EAAS,GAAgB,GAEzB,EAAS,WAAW,EAAI,QAAQ,KAAwB,WAGnD,GAAS,EAAQ,GAG1B,YAAyB,GACvB,MAAO,KAAM,EAAI,QAAU,OAG7B,YACI,EAA2B,EAAiB,EAC5C,EAAmB,EAAqB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAM,GAAoB,IAAU,YAAc,EAAI,EAEhD,EAAO,EAAM,GACb,EAAO,EAAM,OACnB,GAAI,IAAS,GACX,GAAI,IAAU,aACZ,GAAM,GAAe,GAAoB,GACzC,MAAO,CAAC,GAAY,EAAa,GAAI,EAAG,IAE1C,MAAI,KAAU,OACL,CAAC,GAAgB,EAAK,KAExB,CAAC,EAAK,GAAG,YAGlB,GAAI,IAAS,GACX,GAAI,EAAO,IACT,GAAM,GAAgB,GAA6B,EAE/C,EAAY,MAAM,KAClB,EAAK,MAAM,EAAG,IACd,EAAW,MAAM,KAAqC,EAAK,MAC1D,GAAO,IAA8B,EACtC,EAAO,IACX,MAAI,KAAU,aACZ,GAAY,GAAoB,GAChC,EAAW,GAAoB,IAE1B,CACL,IACA,EAAU,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAChD,KAAK,MACV,UACA,EACK,IACG,SAAC,EAAG,GAAM,MAAA,IACN,EAAG,EAAU,EAAO,GAA6B,GAAI,KAC5D,KAAK,MACV,KAGJ,GAAM,GACF,IAAU,YAAc,GAAoB,GACpB,MAAM,KAAoB,GAEtD,MAAO,CACL,IACA,EAAY,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAClD,KAAK,MACV,KAKJ,GAAM,GAAW,EAAM,MAAM,GACvB,EAAa,EAAQ,MAAM,GAC3B,EAAS,EAAQ,GAAK,EACtB,EAAkB,GACxB,GAAI,EAAO,IACT,OAAS,GAAI,EAAG,EAAI,GAA4B,KAC9C,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,KAEN,EAAM,KAAK,OACX,OAAS,GAAI,EAAO,GAA4B,EAAI,EAAM,KACxD,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,SAGnB,QAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,IAGrB,GAAM,GAAM,IAAS,EAAI,IAAM,GAC/B,EAAM,GAAK,IAAM,EAAM,GAAK,EAC5B,OAAS,GAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,EAAM,GAAK,IAAM,EAAM,GAAK,EAG9B,OADI,GAAa;EACR,EAAI,EAAG,EAAI,EAAM,IACxB,GAAc;EAEhB,SAAM,EAAM,OAAS,GACjB,IAAM,EAAM,EAAM,OAAS,GAAK,IAAO,GAAS,GAAK,GAClD,EAGT,YAA6B,GAG3B,OADM,GAAyC,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,GAAK,EACpC,EAAc,KAAK,CAAC,EAAK,GAAI,EAAK,EAAI,KAExC,MAAO,GCnMT,kBA6CE,WAAY,EAA2B,EAAU,GAAjD,GAAA,GAAA,KAIE,GAJqC,KAAA,MAAA,EACrC,KAAK,MAAQ,EAAM,QACnB,KAAK,KAAO,GAAmB,GAE3B,GAAU,MACZ,GAAM,GAAI,EAAO,OACjB,EACI,IAAM,KAAK,KACX,WAAM,MAAA,qBAAqB,EAAC,6BACxB,2BAA0B,EAAK,KAAI,QAE7C,GAAI,IAAU,YACZ,KAAM,IAAI,OACN,8JAIN,KAAK,OAAS,GAAU,GAAuB,EAAO,KAAK,MAC3D,KAAK,QAAU,GAAe,GAWhC,SAAA,UAAA,IAAA,SAAI,UAAJ,GAAA,KAA8B,EAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GAC5B,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAEV,EACI,EAAK,SAAW,KAAK,KACrB,WAAM,MAAA,uCAAuC,EAAK,OAAM,UACpD,oBAAmB,EAAK,KAAI,OAEpC,GAAM,GAAQ,KAAK,WAAW,GAC9B,KAAK,OAAO,GAAS,GAUvB,EAAA,UAAA,IAAA,kBAAI,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACF,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAGV,OADI,GAAI,EACU,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,EAAM,GAAK,GAAO,KAAK,MAAM,IAC/B,GAAM,GAAM,qCAAqC,EAAI,KACjD,mBAAkB,KAAK,OAC3B,KAAM,IAAI,OAAM,GAElB,IAGF,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,MAAK,OAAO,IAGrB,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,IAGT,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,KAAK,MAAM,QACnC,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,KAAK,QAAQ,IAC1C,GAAS,EAAK,GAAK,KAAK,QAAQ,GAElC,SAAK,EAAK,OAAS,GAAK,EACjB,GAGT,OAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAY,WAAW,KAAK,OAAQ,KAAK,MAAO,KAAK,QAGhE,KA8BI,GAAiC,KAEjC,GAAuB,iBAaM,GAC/B,GAAY,cAQe,GAC3B,GAAY,mBA4DZ,WAAY,EAAoB,EAAiB,EAAgB,GAXjE,KAAA,KAAO,GAoIG,KAAA,mBAAqB,GAxH7B,KAAK,MAAQ,EAAM,QACnB,KAAK,MAAQ,GAAS,UACtB,KAAK,KAAO,GAAmB,GAC/B,KAAK,QAAU,GAAe,GAC9B,KAAK,OAAS,EACd,KAAK,GAAK,EACV,KAAK,SAAY,KAAK,KAAO,EAAI,KAAK,KAAK,WAAa,SAG1D,cAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQd,EAAA,UAAA,OAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAU,OAAO,KAAK,MAAO,KAAK,MAAY,UAOvD,EAAA,UAAA,WAAA,WACE,MAAO,IAAU,OAAO,KAAK,MAAO,KAAK,MAAY,KAAK,aAStD,EAAA,UAAA,MAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAc,KAAK,MAAO,UASnC,EAAA,UAAA,UAAA,WACE,MAAO,IAAc,KAAK,MAAO,KAAK,aASlC,EAAA,UAAA,KAAN,mHACE,MAAK,kBACC,EAAO,KAAY,KAAK,KAAK,QAC/B,KAAK,QAAU,SACH,CAAA,EAAM,GADlB,CAAA,EAAA,UACI,EAAQ,EAAA,OACd,IACE,MAAA,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,IAAkB,YACxC,GACA,KAAM,IAAI,OACN,kHAIR,MAAA,CAAA,EAAO,SAST,EAAA,UAAA,SAAA,WACE,KAAK,kBACL,GAAM,GAAO,KAAY,SAAS,KAAK,QACvC,GAAI,KAAK,QAAU,SACjB,IACE,MAAQ,GAAsB,IAAI,SAAA,GAAK,MAAA,IAAkB,WAEzD,GACA,KAAM,IAAI,OACN,iGAIR,MAAO,IAIH,EAAA,UAAA,MAAN,2GACE,YAAK,kBACQ,CAAA,EAAM,KAAY,KAAK,KAAK,gBACzC,MADM,GAAO,EAAA,OACT,KAAK,QAAU,SACjB,CAAA,EAAO,GAEP,CAAA,EAAO,GAAI,YAAY,EAAoB,eAS/C,EAAA,UAAA,QAAA,WACE,GAAI,KAAK,WACP,OAEF,KAAY,cAAc,MAC1B,KAAK,mBAAqB,IAI5B,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,oDAGd,EAAA,UAAA,gBAAA,WACE,GAAI,KAAK,WACP,KAAM,IAAI,OAAM,wBAYpB,EAAA,UAAA,MAAA,SAAM,GAAA,MAAA,KAAA,QAAA,GAAA,IACG,GAAU,MAAM,KAAM,IAO/B,EAAA,UAAA,MAAA,WACE,YAAK,kBACE,GAAU,MAAM,OAQzB,EAAA,UAAA,SAAA,SAAS,GAAA,AAAA,IAAA,QAAA,GAAA,IACP,GAAM,GAAO,KAAK,WAClB,MAAO,IAAe,EAAM,KAAK,MAAO,KAAK,MAAO,IAGtD,EAAA,UAAA,KAAA,SAAqB,GACnB,YAAK,kBACE,GAAU,KAAK,KAAW,IAEnC,EAAA,UAAA,SAAA,SAAS,EAAkB,EAAe,GAAjC,MAAA,KAAA,QAAA,GAAA,IACP,KAAK,kBACE,KAAY,aAAa,KAAM,EAAW,EAAM,IAG3D,KACA,OAAO,eAAe,EAAQ,OAAO,YAAa,CAChD,MAAO,SAAC,GAMN,MAAO,CAAC,CAAC,GAAY,EAAS,MAAQ,MAAQ,EAAS,UAAY,MAC/D,EAAS,iBAAmB,2BAoCiB,GAAA,EAAA,GAGnD,WACI,EAAgC,EAAoB,EACpD,GAFJ,GAAA,GAGE,EAAA,KAAA,KACI,EAAa,MAAO,EAAa,MAAO,EAAa,OAAQ,IAAS,KAHxC,SAAA,UAAA,EAIlC,EAAK,KAAO,IAWd,SAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAS,QAAU,KAAK,MAC1B,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,GAAI,CAAC,GAAiB,EAAS,MAAO,KAAK,OACzC,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,KAAY,cAAc,MAC1B,KAAK,OAAS,EAAS,OACvB,KAAY,OAAO,KAAM,OAG3B,EAAA,UAAA,QAAA,WACE,KAAY,gBAAgB,MAC5B,KAAK,mBAAqB,IAE9B,GAvCqD,GAyCrD,OAAO,eAAe,GAAU,OAAO,YAAa,CAClD,MAAO,SAAC,GACN,MAAO,aAAoB,IAAU,EAAS,QAAU,MACpD,EAAS,iBAAkB,aCnhBnC,AAgEA,AAAA,UAAY,GACV,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,OAPU,EAAA,MAAA,GAAA,KAAI,KAqBhB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,QACA,EAAA,UAAA,cAJG,IAAA,IAAiB,KAOtB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,OACA,EAAA,UAAA,cAJG,IAAA,IAAgB,KAOrB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,UACA,EAAA,KAAA,UACA,EAAA,UAAA,cAJG,IAAA,IAAmB,KAOxB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,YACA,EAAA,MAAA,YACA,EAAA,KAAA,YACA,EAAA,UAAA,cAJG,IAAA,IAAqB,KAO1B,GAAM,IAAgB,CACpB,QAAW,GACX,MAAS,GACT,KAAQ,GACR,UAAa,gBAGY,EAAiB,GAC1C,GAAI,IAAU,UAAY,IAAU,UAClC,GAAI,IAAU,UAAY,IAAU,SAClC,MAAO,SAET,KAAM,IAAI,OAAM,kBAAkB,EAAK,SAAS,GAElD,MAAO,IAAc,GAAO,eAIH,GACzB,MAAO,IAAW,EAAM,SCpI1B,YAsBiD,EAAM,GACrD,GAAI,EAAE,QAAU,EAAE,MAChB,MAAO,CAAC,EAAG,GAEb,GAAM,GAAQ,GAAW,EAAE,MAAO,EAAE,OACpC,MAAO,CAAC,EAAE,KAAK,GAAQ,EAAE,KAAK,gBAGC,EAAW,GAC1C,EACI,EAAE,QAAU,EAAE,MACd,WAAM,MAAA,2BAA2B,EAAE,MAAK,QACpC,YAAW,EAAE,MAAK,oCAGG,EAAgB,GAC7C,MAAO,GAAW,KAAK,SAAA,GAAK,MAAA,GAAE,KAAO,EAAO,iBAeR,GACpC,GAAM,GAAiB,GACjB,EAAO,GAAI,KACjB,UAAoB,EAAQ,EAAM,GAC3B,EAGT,YACI,EAA4B,EAAgB,GAC9C,GAAI,GAAa,KACf,OAEF,GAAI,YAAqB,IACvB,EAAK,KAAK,GACV,OAEF,GAAI,CAAC,GAAW,GACd,OAGF,GAAM,GAAW,EACjB,OAAW,KAAK,IACd,GAAM,GAAM,EAAS,GACrB,AAAK,EAAK,IAAI,IACZ,GAAK,IAAI,GACT,GAAoB,EAAK,EAAM,KAMrC,YAAoB,GAClB,MAAO,OAAM,QAAQ,IAAQ,MAAO,IAAQ,kHCrF9C,AAmFA,GAAA,IAAA,WAAA,aAEE,KAAA,oBAAwC,GAExC,KAAA,eAAiB,EACjB,KAAA,SAAW,EACX,KAAA,WAAa,EACb,KAAA,iBAAmB,EACnB,KAAA,eAAiB,EAMjB,KAAA,cAAgB,EAGhB,KAAA,YAAc,EAId,KAAA,WAA2B,GAK3B,KAAA,kBAA8B,GAC9B,KAAA,YAAc,EAEd,KAAA,WAAa,GAAI,SAQjB,KAAA,UAAY,GACZ,KAAA,cACI,CAAC,SAAU,EAAG,WAAY,EAAG,UAAW,EAAG,QAAS,GAAI,OAAQ,MAEpE,SAAA,UAAA,QAAA,WACE,OAAW,KAAgB,MAAK,oBAC9B,KAAK,oBAAoB,GAAc,WAG7C,KAEA,GAAA,WAgBE,WAAmB,GAAA,KAAA,IAAA,EAbnB,KAAA,SAA0C,GAC1C,KAAA,gBAKI,GAKI,KAAA,qBAAuB,EAG7B,KAAK,MAAQ,GAAI,IAGb,SAAA,UAAA,MAAN,iHACE,GAAI,KAAK,oBAAsB,KAC7B,MAAA,CAAA,EAAO,KAAK,mBAAmB,KAAK,eAEtC,GAAI,KAAK,iBAAmB,KAC1B,MAAA,CAAA,GAEI,EAAiB,KAAK,oBAEnB,EAAI,yBAAG,GAAI,EAAe,OAC3B,GAAc,EAAe,GACnB,CAAA,EAAM,KAAK,kBAAkB,GAAa,UAFnB,CAAA,EAAA,gBAEjC,GAAU,EAAA,OACZ,EACF,CAAA,EAAM,KAAK,WAAW,IADpB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,UALuC,wBAS3C,KAAM,IAAI,OACN,gFAIN,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,GAAI,KAAK,oBAAsB,KAC7B,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,uHAIlC,GAAI,KAAK,iBAAmB,MACpB,GAAA,GAAA,KAAA,kCAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UACb,GAAI,EACF,KAAM,IAAI,OACN,iCAAiC,EAAI,uHAI3C,KAAK,WAAW,GAElB,MAAO,MAAK,iDAGd,EAAA,UAAA,aAAA,WACE,MAAO,QAAO,KAAK,KAAK,kBAG1B,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,CAAE,KAAe,MAAK,UAGxB,GAAI,IAAe,MAAK,iBACf,GAAA,GAAA,KAAA,kBAAA,GAAA,UACP,GAAI,EAEF,MAAO,UAGT,OAAO,MAGX,MAAO,MAAK,SAAS,IAGvB,EAAA,UAAA,mBAAA,SAAmB,GAEjB,MAAM,KAAe,MAAK,gBAGnB,KAAK,gBAAgB,GAAa,QAFhC,MAKX,EAAA,UAAA,gBAAA,SACI,EACA,EACA,GACF,MADE,AAAA,KAAA,QAAA,GAAA,GACE,IAAe,MAAK,gBACtB,SAAQ,KACD,EAAW,sEAEX,IAET,MAAK,gBAAgB,GAAe,CAAC,QAAO,EAAE,SAAQ,GAC/C,KAGH,EAAA,UAAA,WAAN,SAAiB,2GACf,GAAI,KAAK,gBAAgB,IAAgB,KACvC,KAAM,IAAI,OAAM,iBAAiB,EAAW,iCAE9C,MAAK,YAAc,EACf,KAAK,SAAS,IAAgB,KAChC,MAAK,gBAAkB,KACjB,EAAuB,KAAK,kBAAkB,GAA7C,EAAO,EAAA,QAAE,EAAS,EAAA,UACV,EAAY,CAAA,EAAM,GAAlB,CAAA,EAAA,IAHb,CAAA,EAAA,UAGyB,SAAA,EAAA,oBAAgB,EAAA,mBAC3C,GADM,EAAM,EACR,CAAC,EACH,MAAA,CAAA,EAAO,qBAGX,YAAK,gBAAkB,KAAK,SAAS,GACrC,KAAK,yBAEL,KAAK,SAAW,GAAI,IAAS,KAAK,iBAElC,CAAA,EAAO,UAGD,EAAA,UAAA,uBAAR,WAAA,GAAA,GAAA,KACQ,EAAU,GAAqB,KAAK,aAC1C,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,WAAa,MACtB,EAAO,UAAU,EAAK,oBAKpB,EAAA,UAAA,yBAAR,SAAiC,GAAjC,GAAA,GAAA,KACQ,EAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,aAAe,MACxB,EAAO,YAAY,EAAK,SAAS,OAW/B,EAAA,UAAA,kBAAR,SAA0B,GAA1B,GAAA,GAAA,KAEQ,EAAuB,KAAK,gBAAgB,GAClD,GAAI,GAAwB,KAC1B,KAAM,IAAI,OACN,6BAA6B,EAAW,4BAG9C,IACE,GAAM,GAAU,EAAqB,UAMrC,GAAI,GAAW,CAAE,aAAmB,MAC7B,MAAO,GAAQ,MAAS,YAC7B,GAAM,GAAY,EAAE,KAAK,qBACnB,EACF,EACK,KAAK,SAAA,GAEJ,MAAI,GAAY,EAAK,qBACZ,GAET,GAAK,SAAS,GAAe,EAC7B,EAAK,mBAAqB,KACnB,MAER,MAAM,SAAA,GAEL,MAAI,GAAY,EAAK,sBAGrB,GAAK,mBAAqB,KAC1B,QAAQ,KACJ,6BAA6B,EAAW,WAC5C,QAAQ,KAAK,EAAI,OAAS,EAAI,UACvB,KAEjB,YAAK,mBAAqB,EACnB,CAAC,QAAO,EAAE,UAAW,QAE5B,aAAK,SAAS,GAAe,EACtB,CAAC,QAAS,GAAM,UAAW,UAE7B,GACP,eAAQ,KAAK,6BAA6B,EAAW,WACrD,QAAQ,KAAK,EAAI,OAAS,EAAI,SACvB,CAAC,QAAS,GAAO,UAAW,MAIvC,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAE,KAAe,MAAK,iBACxB,KAAM,IAAI,OAAS,EAAW,kCAEhC,AAAI,KAAK,cAAgB,GAAe,KAAK,oBAAsB,MAGjE,KAAK,uBAGP,AAAI,IAAe,MAAK,UACtB,MAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,IAGvB,MAAO,MAAK,gBAAgB,GAG5B,AAAI,KAAK,cAAgB,GACvB,MAAK,mBAAqB,KAC1B,KAAK,YAAc,KACnB,KAAK,gBAAkB,OAInB,EAAA,UAAA,kBAAR,WAAA,GAAA,GAAA,KACE,GAAI,OAAO,KAAK,KAAK,iBAAiB,SAAW,EAC/C,KAAM,IAAI,OAAM,iCAElB,MAAO,QAAO,KAAK,KAAK,iBAAiB,KAAK,SAAC,EAAW,GAExD,MAAO,GAAK,gBAAgB,GAAG,SAC3B,EAAK,gBAAgB,GAAG,YAIxB,EAAA,UAAA,gCAAR,WAIE,OAFM,GAAiB,KAAK,oBAEnB,EAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAc,EAAe,GAC7B,EAAA,KAAA,kBAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAChB,GAAI,GAAa,EACf,MAAO,CAAC,KAAM,EAAa,UAAS,GAGxC,KAAM,IAAI,OACN,2EAIN,EAAA,UAAA,SAAA,SAAS,EAAwB,GAC/B,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAa,EAAK,QAClB,EAAS,KAAK,SAAS,GAG7B,EAAW,YAAY,GACvB,EAAK,QAAU,EACf,EAAQ,KAAK,EAAQ,EAAQ,EAAK,MAAO,EAAK,OAC9C,AAAI,KAAK,0BAGP,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,MAIvE,EAAA,UAAA,KAAA,SAAgC,EAA6B,GAA7D,GAAA,GAAA,KAEM,EAAe,KACnB,GAAI,GAAM,MAER,GAAI,MAAO,IAAa,WACtB,KAAM,IAAI,OAAM,uCAElB,EAAK,OAGL,GAAI,MAAO,IAAa,UAAY,CAAE,aAAoB,SACxD,KAAM,IAAI,OACN,kFAGN,GAAI,MAAO,IAAO,WAChB,KAAM,IAAI,OACN,kFAGN,EAAO,EAIT,GAAI,GACJ,MAAO,MAAK,UACR,WAAM,MAAA,GAAK,WAAW,IAAO,WAAM,MAAA,GAAK,SAAS,IAAS,WACxD,SAAS,IACT,AAAI,YAAkB,UACpB,QAAQ,MAAM,2CAET,KAIP,EAAA,UAAA,UAAR,SAAqB,EAAmB,EAAiB,GACvD,IACA,IACE,GAAM,GAAM,IACZ,WACO,QACA,GACP,UACM,IAKF,EAAA,UAAA,aAAR,WACE,MAAO,GAAO,gBAIR,EAAA,UAAA,eAAR,WACE,MAAO,GAAO,kBAYR,EAAA,UAAA,MAAR,SAAc,GACZ,GAAM,GAAI,KAAK,qBAAqB,EAAE,OAAQ,EAAE,MAAO,EAAE,OACnD,EAAS,CAAC,EAAC,GACX,EAAO,SAAC,GAAe,MAAC,CAC5B,EAAG,WACD,GAAM,GAAQ,UACR,EAAa,CAAC,EAAG,GACjB,EAAQ,CAAC,MAAK,GAEpB,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAC5B,EAAoC,KAAiB,GACrD,MAGF,EAAkB,GACxB,YAAK,YAAY,KAAK,MAAM,YAAY,KAAM,EAAQ,CAAC,GAAI,EAAM,EAAO,IACjE,GAgBT,EAAA,UAAA,UAAA,SACI,EAAoB,EAAwB,EAC5C,EAAyB,GAC3B,GAAM,GAAoB,KACpB,EAAsB,KAI5B,MAAO,MAAK,cACR,EAAa,EAAQ,EAAe,EAAY,EAAO,EACvD,IAGE,EAAA,UAAA,uBAAR,WACE,MAAO,MAAK,IAAI,QAAQ,YAGlB,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAkB,KAAK,QAAQ,aAGjC,EAAmB,EACvB,EAAS,QAAQ,SAAA,GAGf,GAAqB,EAAK,QAAU,YAAc,EAAI,IAQxD,GAAM,GACF,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,GACjE,EACF,EAAkB,EAAmB,EAAmB,EAC5D,GAAI,EAAgB,EAClB,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,iCAC5B,KAAI,EAAa,6BAA6B,EAAU,OAQhE,EAAA,UAAA,cAAA,SACI,EAA6B,EAC7B,EACA,EAAqB,EAAsB,EAC3C,GAJJ,GAAA,GAAA,KAKM,EACA,EAAkB,GAChB,EAAW,KAAK,WACtB,AAAI,GAAc,MAChB,GACI,KAAK,MAAM,aAAe,KAAO,KAAK,MAAM,YAAY,KAAO,IAGrE,GAAM,GAAoB,KAAK,MAAM,SAC/B,EAAqB,KAAK,MAAM,WAEtC,AAAI,KAAK,0BACP,KAAK,MAAM,kBAAkB,KAAK,GAGpC,GAAI,GACE,EAAS,GAAU,EAAY,KAAK,aACtC,EACJ,GAAI,GAAU,KACZ,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAO,WAAW,CAAC,OAAM,EAAE,MAAK,EAAE,QAAS,EAAK,UACtD,GAAM,GAAW,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC7C,AAAI,EAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAE3D,GAAM,GAAa,EAAS,IACxB,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,MACb,MAAA,GAAK,qBAAqB,EAAQ,EAAO,KAMjD,GAAI,GACF,GAAI,GACA,EAAK,sBAAsB,EAAY,EAAQ,GACnD,GAAI,GAAiB,MAKnB,AAAI,GAAiB,MACnB,GAAgB,IAElB,GAAM,GAAa,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAC7D,EAAiB,IAAgB,IAAI,QAAQ,OAAO,GAEtD,EAAQ,EAAK,2BAA2B,GAE1C,MAAO,SAGT,GAAM,GAAyB,SAAC,GAI9B,GAAI,CAAC,EACH,OAEF,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,OAGrD,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAK,KAAK,WAAM,MAAA,GAAY,EAAK,QAAS,KAChD,GAAM,GAAQ,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC1C,MAAI,GAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAEpD,GAKX,GAAI,GACJ,YAAK,UACD,WAAM,MAAA,GAAK,MAAM,eAAe,WAAM,MAAA,GAAK,MAAM,eAAe,WAC9D,AAAI,CAAC,EAAK,IAAI,QAAQ,UAAY,CAAC,EAAK,MAAM,UAC5C,EAAU,IAEV,GAAgB,EAAK,SAAS,cAC1B,EAAY,EAAQ,WAAM,MAAA,OAC9B,AAAI,EAAK,IAAI,QAAQ,UACnB,EAAK,SAAS,iBAAiB,GAEjC,EAAU,EAAc,WAIhC,AAAI,GACF,KAAK,YACD,EAAY,EAAQ,EAAS,EAAe,EAAO,GAGzD,AAAI,KAAK,MAAM,WACb,KAAK,MAAM,cAAc,QAAQ,KAAK,CACpC,KAAM,EACN,WAAY,KAAK,MAAM,SAAW,EAClC,mBAAoB,KAAK,MAAM,SAC/B,aAAc,KAAK,MAAM,WAAa,EACtC,qBAAsB,KAAK,MAAM,WACjC,YAAa,OAAO,KAAK,GAAQ,IAC7B,SAAA,GAAO,MAAA,GAAO,IAAQ,KAAO,EAAO,GAAK,MAAQ,OACrD,aAAc,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,QACvC,aAAc,EAAc,OAC5B,UAAW,EAAc,YAGrB,MAAM,QAAQ,GAAO,EAAU,EAAQ,IAQzC,EAAA,UAAA,2BAAR,SAAmC,GAAnC,GAAA,GAAA,KACQ,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,MACzD,MAAO,IAaD,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAa,GAAY,GAC/B,GAAI,GAAc,MAChB,GAAM,GAAyB,EAAW,cAAgB,GACpD,EAA2B,EAAW,eAAiB,GAIzD,EAAkB,OACtB,AAAI,EAAW,cACb,GACI,MAAM,QAAQ,GACd,WAAM,MAAA,2DAEV,EAAqB,OAAO,KAAK,GAAQ,IAAI,SAAC,GAAQ,MAAA,GAAO,MAE7D,EAAqB,EAAa,IAAI,SAAC,GAAc,MAAA,GAAO,KAG9D,GAAM,GACF,EAAQ,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAE3C,MAAO,GAAmB,OAAO,GAInC,MAAO,OAQT,EAAA,UAAA,WAAA,SACI,EAAoB,EAAiB,EACrC,GACF,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,iDAElB,EAAQ,GAAS,UACjB,EAAU,GAAW,KAAK,QAC1B,GAAI,GAAc,EAClB,AAAI,IAAU,UAAY,GAAc,EAAO,KAC7C,GAAe,EAAoB,IAAI,SAAA,GAAK,MAAA,IAAkB,MAEhE,GAAM,GAAS,EAAQ,MAAM,EAAa,EAAO,GAC3C,EAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAIhD,GAHA,KAAK,OAAO,EAAG,GAGX,IAAU,UACZ,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAW,GAAqB,GACtC,KAAK,MAAM,UAAY,EAAW,EAAK,MACvC,EAAK,MAAQ,EAEf,MAAO,IAQT,EAAA,UAAA,qBAAA,SACI,EAAgB,EAAiB,EACjC,GACF,EAAQ,GAAS,UACjB,GAAM,GAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAChD,YAAK,OAAO,EAAG,GACR,GAGT,EAAA,UAAA,aAAA,SACI,EAAsB,EAAkB,EACxC,GADsB,AAAA,IAAA,QAAA,GAAA,IAExB,EAAO,GAAQ,KAAK,iBAAiB,WACrC,AAAI,GAAS,MAAQ,IAAU,EAAa,OAC1C,GAAe,EAAa,KAAK,IAEnC,GAAM,GAAI,GAAI,IAAS,EAAc,EAAW,EAAM,KAAK,gBAC3D,GAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,KAC5C,KAAM,IAAI,OAAM,sBAAsB,EAAE,KAAI,2BAE9C,YAAK,MAAM,oBAAoB,EAAE,MAAQ,EACzC,KAAK,OAAO,EAAG,KAAK,SACb,GAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAM,GAAW,KAAK,MAAM,WAAW,IAAI,EAAE,QACzC,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,SACpC,EAKJ,GAJA,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAET,IAAa,GACf,KAAK,MAAM,iBAIX,GAAI,GAAQ,EACZ,AAAI,EAAE,QAAU,aAAe,EAAE,QAAU,UACzC,GAAQ,EAAE,KAAO,GAAqB,EAAE,QAE1C,KAAK,MAAM,WAAW,IAAI,EAAE,OAAQ,CAClC,QAAS,GAAW,KAAK,QACzB,MAAO,EAAE,MACT,MAAO,EAAE,MACT,MAAK,EACL,SAAU,IAEZ,KAAK,MAAM,UAAY,EAGzB,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,WAEpC,AAAM,YAAa,KACjB,KAAK,MAAM,IAIf,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAC,KAAK,MAAM,WAAW,IAAI,EAAE,QAC/B,OAGF,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAEb,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,EAAE,QACnC,EAAW,EAAK,SAEtB,AAAI,GAAY,EAGd,CAAI,EAAE,QAAU,aACd,MAAK,MAAM,UAAY,EAAK,OAE9B,KAAK,MAAM,iBAEX,EAAK,QAAQ,YAAY,EAAE,QAC3B,KAAK,MAAM,WAAW,OAAO,EAAE,SAE/B,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,YAOxC,EAAA,UAAA,iBAAA,WACE,OAAW,KAAW,MAAK,MAAM,qBAC/B,GAAM,GAAI,KAAK,MAAM,oBAAoB,GACzC,KAAK,gBAAgB,KAIzB,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,cAAc,GACnB,AAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,MAC5C,MAAO,MAAK,MAAM,oBAAoB,EAAE,OAI5C,EAAA,UAAA,OAAA,WACE,GAAM,GAAO,KAAK,QAAQ,SAC1B,SAAK,WAAa,KAAK,MAAM,WAC7B,EAAK,eAAiB,KAAK,MAAM,eACjC,EAAK,SAAW,KAAK,MAAM,SAC3B,AAAI,KAAK,MAAM,iBAAmB,GAChC,GAAK,WAAa,GAClB,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,IAEjB,EAAK,QAAQ,KACT,0EAGC,GAGH,EAAA,UAAA,QAAN,SAAc,iHAEZ,YAAK,MAAM,UAAY,GAEjB,EAAa,KAAK,MAAM,SACxB,EAAkB,KAAK,MAAM,WAEnC,KAAK,MAAM,cAAc,QAAU,GACnC,EAAA,KAAK,MAAM,cAAuB,CAAA,EAAM,YAAxC,EAAyB,OAAS,EAAA,OAElC,KAAK,MAAM,UAAY,GAEvB,KAAK,MAAM,cAAc,UAAY,KAAK,IAAG,MAAR,KAC9B,KAAK,MAAM,cAAc,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,sBACnD,KAAK,MAAM,cAAc,SAAW,KAAK,MAAM,SAAW,EAC1D,KAAK,MAAM,cAAc,WACrB,KAAK,MAAM,WAAa,MACP,EAAA,KAAK,MAAM,cAAc,+BAAzB,GAAA,EAAA,OAAV,GAAM,EAAA,GACf,EAAA,EAAsB,CAAA,EAAM,EAAO,eADgB,CAAA,EAAA,UACnD,SAAO,aAAe,EAAA,OACtB,EAAA,EAAmB,CAAA,EAAM,EAAO,kBAAhC,EAAO,UAAY,EAAA,wBAFA,wBAIrB,MAAA,CAAA,EAAO,KAAK,MAAM,qBAGpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,MAAM,cAAgB,GAAK,KAAK,MAAM,cAAgB,GAG5D,EAAA,UAAA,YAAR,SACI,EAAoB,EAAwB,EAC5C,EAAyB,EAAiB,GAF9C,GAAA,GAAA,KAGQ,EACF,CAAC,GAAI,KAAK,MAAM,iBAAkB,WAAU,EAAE,OAAM,EAAE,QAAO,EAAE,MAAK,GAElE,EAAa,GAAY,GAC/B,AAAI,GAAc,MAChB,GAAgB,EAAW,UAE7B,AAAI,GAAiB,MACnB,GAAS,SAAW,SAAC,GAGnB,SAAM,EAAI,IAAI,SAAC,EAAI,GACjB,GAAI,GAAM,MACR,GAAM,GAAS,EAAQ,GACjB,EAAO,GAAyB,EAAO,KAAM,EAAO,OAC1D,MAAO,GAAK,WAAW,EAAM,EAAO,MAAO,EAAO,OAEpD,MAAO,KAIF,EAAc,EAAI,OAAS,EAAI,EAAM,EAAI,GAAI,EAAO,KAG/D,KAAK,MAAM,WAAW,KAAK,IAG7B,EAAA,UAAA,KAAA,SAAuB,GACrB,SAAO,KAAO,GACP,GAGD,EAAA,UAAA,UAAR,WACE,AAAI,KAAK,MAAM,gBAAkB,GAC/B,MAAK,MAAM,WAAa,IAE1B,KAAK,MAAM,iBAGL,EAAA,UAAA,QAAR,WACE,KAAK,MAAM,iBAOb,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAwB,CAC5B,MAAO,GACP,KAAM,gBACN,GAAI,KAAK,MAAM,eAEjB,AAAI,GACF,GAAU,KAAO,GAEnB,KAAK,MAAM,WAAW,KAAK,GAC3B,KAAK,MAAM,YAAc,GAO3B,EAAA,UAAA,SAAA,SAAS,GAMP,OANF,GAAA,KACQ,EAAyB,GAAsB,GAC/C,EACF,GAAI,KAAI,EAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,MAGrC,EAAI,EAAG,EAAI,KAAK,MAAM,YAAY,MAAM,OAAQ,KACvD,GAAM,GAAS,KAAK,MAAM,YAAY,MAAM,GAC5C,AAAI,CAAC,EAAO,MAAQ,CAAC,EAA0B,IAAI,EAAO,KACxD,EAAO,UAIX,GAAM,GAAW,KAAK,MAAM,WAAW,MACvC,KAAK,MAAM,YAAc,KAAK,MAAM,WAAW,SAAW,EACtD,KACA,KAAK,MAAM,WAAW,KAAK,MAAM,WAAW,OAAS,GAGzD,EAAuB,QAAQ,SAAA,GAG7B,AAAI,CAAC,EAAO,MAAQ,EAAO,UAAY,EAAS,IAC9C,EAAK,MAAM,MAWjB,EAAA,UAAA,UAAA,SACI,EAAY,EAAc,EAC1B,GAFJ,GAAA,GAAA,KAKE,GAHE,AAAA,IAAA,QAAA,GAAA,IACF,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,8CACrB,GAAM,MAAQ,EAAG,QAAU,UAC7B,KAAM,IAAI,OAAM,0CAA0C,EAAG,MAAK,KAGpE,GAAM,GAAI,KAAK,UACX,WAAM,MAAA,GAAK,aAAa,WAAM,MAAA,GAAK,WACnC,WAAM,MAAA,GAAK,KAAK,UAAW,KAE/B,EACI,YAAa,GACb,WAAM,MAAA,mDAEV,GAAM,GAAe,GAAqB,KAAK,MAAM,WAAY,EAAI,GACrE,GAAI,CAAC,GAAoB,EAAa,SAAW,GAAK,EAAG,OAAS,EAChE,KAAM,IAAI,OACN,uIAKN,MAAO,MAAK,KAAK,WAAY,WAC3B,GAAM,GAAuD,GAC7D,EAAuB,EAAE,IAAO,GAAc,GAAK,EAAE,OAGrD,GACI,EAAwB,EAExB,SAAA,GAAK,MAAA,GAAK,KAAK,IAEf,IACJ,GAAM,GAAQ,EAAG,IAAI,SAAA,GAAK,MAAA,GAAuB,EAAE,MAEnD,MAAI,GAAK,MAAM,gBAAkB,GAG/B,GAAK,MAAM,WAAW,QAAQ,SAAA,GAC5B,OAAqB,GAAA,EAAA,EAAA,EAAK,MAAL,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,EAAO,aAGX,EAAK,MAAM,WAAa,MAEnB,CAAC,MAAO,EAAG,MAAK,MAI3B,EAAA,UAAA,WAAA,SAA6B,GAA7B,GAAA,GAAA,KAEE,SACI,GAAgB,GAChB,WAAM,MAAA,sDACH,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACN,EACI,EAAO,MAAM,SAAA,GAAK,MAAA,aAAa,KAC/B,WAAM,MAAA,qEAGV,GAAI,GAIE,EAA2B,GACjC,SAAO,QAAQ,SAAC,EAAO,GACrB,EAAS,GAAK,IAET,EAAK,cACR,SAAC,EAAG,GACF,SAAM,EAAC,MAAA,OAAQ,EAAM,OAAA,CAAE,KACvB,EACI,EAAI,gBAAiB,GACrB,WAAM,MAAA,+FAEV,EACI,GAAgB,EAAI,UACpB,WAAM,MAAA,qGAEH,EAAI,OAEb,EACA,SAAC,EAAO,GACN,GAAM,GAAU,EAAI,SAAS,EAAI,GAC3B,EACF,MAAM,QAAQ,GAAW,EAAU,CAAC,GACxC,EACI,EAAM,SAAW,EAAO,OACxB,WAAM,MAAA,wKAGV,EACI,EAAM,MAAM,SAAA,GAAK,MAAA,aAAa,KAC9B,WAAM,MAAA,yIAGV,GAAM,GAAyC,GAC/C,SAAM,QAAQ,SAAC,EAAM,GACnB,EAAQ,GAAK,WAAM,MAAA,MAEd,MAKjB,EAAA,UAAA,SAAA,SAAS,GAEP,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,SAAS,IAE/B,EAAA,UAAA,KAAA,SAAK,GAEH,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,KAAK,IAGrB,EAAA,UAAA,KAAN,SAAW,qGACH,SAAQ,KACK,CAAA,EAAM,KAAK,QAAQ,KAAK,WAArC,SAAa,EAAA,OACnB,EAAW,OAAS,KAAQ,EAC5B,CAAA,EAAO,SASD,EAAA,UAAA,MAAR,SAAgC,GAC9B,MAAI,MAAK,MAAM,aAAe,MAC5B,GAAO,QAAU,KAAK,MAAM,YAAY,GACxC,KAAK,MAAM,YAAY,MAAM,KAAK,IAG7B,GAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAOpB,EAAA,UAAA,MAAA,WAEE,KAAK,uBAEL,KAAK,MAAM,UACX,KAAK,IAAI,QACT,KAAK,MAAQ,GAAI,IAEjB,OAAW,KAAe,MAAK,SAC7B,KAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,GAEvB,KAAK,YAAc,KACnB,KAAK,gBAAkB,KACvB,KAAK,mBAAqB,MA/sBb,EAAA,aAAe,EAKf,EAAA,eAAiB,EA4sBlC,KAEA,YAAc,GACZ,GAAM,GAAS,GAAmB,GAAc,GAAQ,WACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,yBAIxC,GAAM,GAAK,KACX,GAAI,EAAG,WAAa,MAClB,GAAM,GAAc,GAAI,IAAY,GACpC,EAAG,UAAY,GAAI,IAAO,GAE5B,UAAqB,EAAG,UAAU,KAIlC,GAAiB,WAAM,MAAA,GAAG,YACnB,EAAG,UAGL,GAAM,GAAS,iBAQF,EAAW,GAE7B,GAAM,GAAS,CAAC,EAAC,EAAE,EAAC,GACpB,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,EAAG,GAC3B,SAAK,CAAC,EAAG,IACF,GACN,EAAgC,KAAqB,ICzrC1D,AAkBA,cACE,MAAO,OAAO,YAAc,aAAe,WAAa,mBAIxD,GAAI,MAEF,GAAM,GAAI,UAAU,WAAa,UAAU,QAAW,OAAe,MAErE,MAAO,2TACK,KAAK,IAEb,0kDACK,KAAK,EAAE,OAAO,EAAG,IAE5B,MAAO,iBAIP,MAAQ,OAAO,SAAW,aAAe,OAAO,UAAY,MAEvD,MAAO,oBAAsB,6DCvCpC,AAqBA,GAAM,IAAM,KAOZ,GAAI,aAAa,QAAS,WAAM,MAAA,IAAO,SAAA,GACrC,AAAI,GACF,QAAQ,KACJ,iJAOR,GAAI,aAAa,aAAc,WAAM,MAAA,QAGrC,GAAI,aACA,UACA,WAAM,MAAC,OAAO,UAAY,aACrB,MAAO,SAAQ,UAAa,aAC5B,MAAO,SAAQ,SAAS,MAAS,cAG1C,GAAI,aACA,YACA,WAAM,MAAA,OAAO,YAAc,aAAe,WAAa,MACnD,UAAU,WAAa,MAAQ,SAAS,KAAK,UAAU,YACvD,aAAa,KAAK,UAAU,UAMpC,GAAI,aAAa,OAAQ,WAAM,MAAA,KAM/B,GAAI,aACA,qCAAsC,WAAM,MAAA,IAAI,QAAQ,WAG5D,GAAI,aAAa,+BAAgC,WAAM,MAAA,KAGvD,GAAI,aAAa,UAAW,WAAM,MAAA,KCvElC,YAuB2B,EAAiB,GAC1C,GAAI,GAAwB,EAE5B,GAAI,GAAa,GACf,MAAO,KAAU,SAAW,GAAK,CAAC,EAAI,QAExC,GAAI,CAAC,MAAM,QAAQ,GACjB,MAAO,GAIT,OAFM,GAAkB,GAEjB,MAAM,QAAQ,IACd,GAAa,IAAc,IAAU,UAC1C,EAAM,KAAK,EAAU,QACrB,EAAY,EAAU,GAExB,MAAI,OAAM,QAAQ,IACd,KAAM,QAAQ,uCAChB,GAA2B,EAAK,EAAO,IAGlC,EAGT,YACI,EAAiB,EAAiB,GAEpC,GADA,EAAU,GAAW,GACjB,CAAE,MAAM,QAAQ,IAAS,CAAC,GAAa,IACzC,EACI,EAAM,SAAW,EACjB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,qBACnC,yCAAwC,EAAM,GAAE,eACxD,OAEF,EACI,EAAM,OAAS,EACf,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,4BACnC,uBAAsB,EAAI,OAAM,eACxC,EACI,EAAI,SAAW,EAAM,GACrB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,iBAAiB,EAAM,GAAE,IAC5D,sBAAqB,EAAI,OAAM,eAEvC,OADM,GAAW,EAAM,MAAM,GACpB,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAA2B,EAAI,GAAI,EAAU,EAAQ,OAAO,IAIhE,YACI,EAAmC,EAAuB,EAC1D,GACF,GAAI,GAAiB,KACnB,OAEF,GAAI,IAAkB,WAAa,IAAkB,GACjD,IAAkB,WAAa,IAAgB,SACjD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,UAChD,OAAM,EAAa,oBAAoB,EAAW,uBAKtD,EAAiB,EAAiB,EAClC,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,YAAa,GACf,UAAY,EAAc,EAAE,MAAO,EAAS,GACrC,EAET,GAAI,GAAgB,GAAW,GAS/B,GANA,AAAI,IAAkB,UAClB,CAAC,OAAQ,QAAS,WAAW,QAAQ,IAAiB,GACxD,GAAgB,GAElB,GAAY,EAAc,EAAe,EAAS,GAE7C,GAAK,MACL,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IAAM,MAAO,IAAM,UACtD,MAAO,IAAM,WAAa,MAAO,IAAM,UAC1C,GAAM,GAAO,GAAK,KAAO,OAAU,EAAS,YAAY,KACxD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,eAChD,mCAAkC,EAAI,MAE5C,GAAM,GAAgB,GAAW,EAAG,GACpC,AAAI,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IACrC,GAAI,CAAC,IAEP,GAAM,GAAiB,GACjB,EAAS,IAAkB,SAC7B,GAAa,EAAG,GAChB,GAAQ,EAAe,GAAI,GAC/B,MAAO,GAAO,WAAW,EAAQ,EAAe,eAI9C,EAA0B,EAAiB,EAC3C,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,OACN,YAAY,EAAO,cAAc,EAAY,2CAGnD,GAAM,GAAU,EAChB,MAAO,GAAQ,IACX,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAM,EAAO,IAAI,EAAC,IAAK,IACjD,GCnIN,GAkBa,IAAkB,kBAOQ,GACrC,GAAM,GAAO,OAAO,KAAK,GACzB,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,OACN,yGAEG,GAAK,OAAM,WAGpB,GAAI,GAAS,EAAK,GACZ,EAAK,EAAE,GAGb,AAAI,EAAO,SAAS,MAClB,GAAS,EAAO,UAAU,EAAG,EAAO,OAAS,IAI/C,EAAS,EAAS,GAGlB,GAAM,GAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,EAAO,WAAW,GAClB,IACE,GAAM,GAAS,EAAE,MAAA,OAAI,GACrB,MAAI,aAAkB,UACpB,QAAQ,MAAM,2CAEhB,EAAO,SAAS,GACT,QACA,GACP,QAAO,SAAS,MACV,IAGV,cAAO,eAAe,EAAI,OAAQ,CAAC,MAAO,EAAQ,aAAc,KAGzD,EC/DT,AA8CA,YAAoC,EAAoB,GACtD,GAAM,GAAQ,EAAgB,EAAM,OAAQ,WACtC,EAAQ,EAAgB,EAAM,OAAQ,WAC5C,GACI,EAAM,MAAO,EAAM,MACnB,yBAAyB,EAAM,MAAK,QAAQ,EAAM,MAAK,yCAG3D,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,QAAQ,EAAO,IAE1B,EAAwB,CAAC,KAAM,EAAO,KAAM,GAClD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC/DnC,YAyBI,EAAoB,EAAiB,EACrC,GAIF,GAHA,AAAI,GAAS,MACX,GAAQ,GAAW,IAEjB,IAAU,YACZ,KAAM,IAAI,OACN,oFAGN,GAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IACxC,MAAO,IAAW,UAAY,MAAO,IAAW,WAChD,MAAO,IAAW,SACpB,KAAM,IAAI,OACN,4HAGN,GAAI,GAAS,MACX,GAAmC,GAEnC,GAAM,GAAe,GAAc,GAC7B,EAAe,GAAc,GACnC,EACI,IAAiB,EACjB,WACI,MAAA,iCAAiC,EAAK,6BACnC,GAAY,mBAAmB,KAE1C,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,EAAE,GAC1C,GAAM,GAAW,EAAc,GACzB,EAAoB,IAAM,EAAc,OAAS,EACnD,IAAa,GAAc,EAAM,MAAM,IACvC,GACJ,EACI,EAAc,KAAO,EAAM,IAAM,CAAC,EAClC,WAAM,MAAA,+CACF,KAAI,EAAa,kCACjB,WAAU,EAAK,UAI3B,MAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IAC1C,GAAS,CAAC,IAGZ,EAAQ,GAAS,EACjB,EAAS,IAAU,SACf,GAAa,EAAQ,GACrB,GAAQ,EAAoB,GAAI,IAC7B,EAAO,WAAW,EAAsB,EAAO,GC1ExD,YAqDI,EAAoB,EAAqB,GAC3C,GAAM,GAAgB,GAAW,EAAQ,GACzC,MAAO,IAAW,EAAQ,EAAO,EAAe,GCvDlD,AAsBO,GAAM,IAAkD,CAC7D,QAAW,EACX,QAAW,EACX,MAAS,EACT,OAAU,EACV,MAAS,EACT,KAAQ,EACR,UAAa,GC7Bf,AA2BA,GAAM,IAA0B,cAoB5B,EAAuC,oHAUzC,IAPM,EAAgC,GAChC,EAA2C,GAE3C,EAAkB,MAAM,QAAQ,GAClC,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,OAC7B,OAAO,KAAK,cAEP,GACP,GAAM,GAAO,EAAM,GACb,EAAI,MAAM,QAAQ,GAAW,EAAQ,GAAG,OAAS,EAAQ,GAC/D,GAAI,EAAE,QAAU,WAAa,EAAE,QAAU,SAAW,EAAE,QAAU,QAC5D,EAAE,QAAU,UAAY,EAAE,QAAU,YACtC,KAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,EAAE,OAE9D,GAAM,GAA6B,CAAC,KAAI,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,OACnE,GAAI,EAAE,QAAU,UACd,GAAM,GAAY,GAAI,SAAoB,SAAM,GAAO,MAAA,IAAA,EAAA,OAAA,OAAA,+EACxC,MAAA,CAAA,EAAM,EAAE,gBAKrB,IALM,EAAO,EAAA,OACP,EAAgB,EAAK,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAE,QAAQ,GACtD,GAA0B,EAAK,OAC7B,EAAQ,GAAI,YAAW,GACzB,EAAS,EACJ,EAAI,EAAG,EAAI,EAAK,OAAQ,IACzB,EAAM,EAAK,GACX,EACF,GAAI,YAAW,GAAI,aAAY,CAAC,EAAI,SAAS,QACjD,EAAM,IAAI,EAAe,GACzB,GAAU,GACV,EAAM,IAAI,EAAK,GACf,GAAU,EAAI,OAEhB,SAAQ,cAEV,EAAa,KAAK,OAElB,GAAa,KAAK,EAAE,QAEtB,AAAI,GAAS,MACX,GAAK,MAAQ,GAEf,EAAM,KAAK,IAjCJ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,IAA3B,GAoCY,MAAA,CAAA,EAAM,QAAQ,IAAI,WAAjC,SAAe,EAAA,OACrB,CAAA,EAAO,CAAC,KAAM,GAAuB,GAAe,MAAK,qBAmBvD,EAAqB,GAKvB,OAHM,GAAsB,GACxB,EACA,EAAS,EACM,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAO,EAAK,KACZ,EAAQ,EAAK,MACb,EAAQ,EAAK,MACb,EAAO,GAAc,GACvB,EAAM,OAEV,GAAI,gBAAkB,IACpB,GAAM,GAAe,EAAK,aAC1B,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,GAAI,CAAE,QAAS,IAAgB,SAAW,IACxC,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,+DAGxD,EAAa,QAAU,WAChC,GAAI,IAAU,UACZ,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,IAC3D,oDAAmD,EAAK,UAG9D,MAAM,IAAI,OACN,UAAU,EAAK,KAAI,gBACnB,uBAAsB,EAAa,MAAK,MACxC,wEAGN,GAAM,GAAyB,GAAqB,EAAa,OAC3D,EACF,EAAO,MAAM,EAAQ,EAAS,EAAO,GACnC,EAAkB,EAAa,QAAU,QAC3C,GAAI,YAAW,GACf,GAAI,aAAY,GACpB,GAAI,IAAU,UACZ,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,EAAS,GAAI,cAAa,EAAe,QACzC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,EAAI,EAAa,MAAQ,EAAa,aAE3C,EAAa,QAAU,UAChC,AAAI,IAAkB,QACpB,GAAgB,MAElB,EAAS,EAAc,OAEvB,MAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,qCAGhD,IAAU,SACnB,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,SAC3D,KAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,2BAGzD,EAAS,GAAI,YAAW,EAAe,QACvC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,KAAK,MAAM,EAAI,EAAa,MAAQ,EAAa,UAG/D,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,UACR,IAAU,UACnB,GAAM,GAAO,GAAc,EAAK,OAChC,EAAS,GACT,OAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAa,GAAI,aACnB,EAAO,MAAM,EAAQ,EAAS,KAA0B,GAC5D,GAAU,GACV,GAAM,GAAQ,GAAI,YAAW,EAAO,MAAM,EAAQ,EAAS,IAC1D,EAAwB,KAAK,GAC9B,GAAU,QAGZ,GAAM,GAAc,GAAqB,GACnC,EAAa,EAAO,MAAM,EAAQ,EAAS,EAAO,GAExD,GAAI,IAAU,UACZ,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,aACnB,EAAS,GAAI,cAAa,GAG1B,OAFM,GAAO,GAAI,cAAa,EAAO,OAAS,GACxC,EAAQ,GAAI,cAAa,EAAO,OAAS,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAK,GAAK,EAAO,EAAI,GACrB,EAAM,GAAK,EAAO,EAAI,EAAI,GAE5B,GAAM,GAAa,GAAO,EAAM,EAAO,WACjC,EAAc,GAAO,EAAO,EAAO,WACzC,EAAI,GAAQ,GAAQ,EAAY,OAEhC,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,EAEnB,AAAI,IAAU,aACZ,GAAI,GAAQ,GAAO,EAAQ,EAAO,IAGtC,MAAO,eAM8B,GAErC,GAAI,IAAO,KACT,KAAM,IAAI,OAAM,wBAAwB,KAAK,UAAU,IAGzD,GAAI,GAAkB,EAShB,EAA6B,GACnC,EAAG,QAAQ,SAAC,GAMV,GALA,GAAmB,EAAE,WAErB,EAAa,KACT,EAAE,aAAe,EAAE,OAAO,WAAa,EACA,GAAK,GAAE,YAAoB,IAClE,CAAE,aAAoB,eAAgB,YAAoB,aACxD,YAAoB,aACxB,KAAM,IAAI,OAAM,mCAAmC,EAAE,YAAY,QAKrE,GAAM,GAAI,GAAI,YAAW,GACrB,EAAS,EACb,SAAa,QAAQ,SAAC,GACpB,EAAE,IAAI,GAAI,YAAW,EAAE,QAAS,GAChC,GAAU,EAAE,aAGP,EAAE,OAIX,GAAM,IAAgB,MAAO,SAAW,aACnC,OAAO,OAAS,aAAe,MAAO,OAAS,aAC/C,MAAO,OAAS,yBAWY,GAC/B,MAAI,IACK,OAAO,WAAW,GAEpB,GAAI,MAAK,CAAC,IAAM,iBASiB,GACxC,GAAI,GACF,MAAO,QAAO,KAAK,GAAQ,SAAS,UAItC,OAFM,GAAM,GAAI,YAAW,GACvB,EAAI,GACC,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAI,EAAG,IACrC,GAAK,OAAO,aAAa,EAAI,IAE/B,MAAO,MAAK,eAS4B,GACxC,GAAI,IACF,GAAM,GAAM,OAAO,KAAK,EAAK,UAC7B,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YAI/D,OAFM,GAAI,KAAK,GACT,EAAS,GAAI,YAAW,EAAE,QACvB,EAAI,EAAG,EAAI,EAAE,OAAQ,EAAE,EAC9B,EAAO,IAAI,CAAC,EAAE,WAAW,IAAK,GAEhC,MAAO,GAAO,mBASwB,GACtC,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAI,GAAkB,EACtB,EAAQ,QAAQ,SAAC,GACf,GAAmB,EAAO,aAG5B,GAAM,GAAO,GAAI,YAAW,GACxB,EAAS,EACb,SAAQ,QAAQ,SAAC,GACf,EAAK,IAAI,GAAI,YAAW,GAAS,GACjC,GAAU,EAAO,aAEZ,EAAK,mBAUW,GACvB,GAAM,GAAY,IAElB,IADA,EAAO,EAAK,OACL,EAAK,SAAS,IACnB,EAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAErC,GAAM,GAAQ,EAAK,MAAM,GACzB,MAAO,GAAM,EAAM,OAAS,eAQe,GAE3C,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OAAM,uDAGlB,MAAO,CACL,UAAW,GAAI,MACf,kBAAmB,OACnB,mBAAoB,EAAe,eAAiB,KAChD,EACA,GAAiB,KAAK,UAAU,EAAe,gBACnD,iBAAkB,EAAe,aAAe,KAC5C,EACA,GAAiB,KAAK,UAAU,EAAe,cACnD,gBAAiB,EAAe,YAAc,KAC1C,EACA,EAAe,WAAW,YAUlC,cACE,GAAM,GAAkB,SAAC,GAIvB,OAHI,GAAI,GAAK,GACT,EAAI,EAEA,GAAI,WAAgB,GAC1B,GAAK,QACL,IAAM,EAER,UAAK,CAAC,QACN,GAAK,UAEE,EAAI,GAGP,EAAe,GAAI,aAAY,MAErC,EAAa,GAAK,EAClB,OAAS,GAAI,EAAG,EAAI,KAAM,IACxB,EAAa,GAAK,EAAgB,GAEpC,OAAS,GAAI,KAAM,EAAI,KAAM,IAC3B,EAAa,GAAK,UAAe,GAAI,MAAS,IAGhD,MAAO,GAST,cACE,GAAM,GAAgB,GAAI,aAAY,IAEtC,EAAc,GAAK,EACnB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,OAAS,GAAI,EAAG,EAAI,GAAI,IACtB,EAAc,GAAK,GAAK,GAE1B,OAAS,GAAI,GAAI,EAAI,GAAI,IACvB,EAAc,GAAK,WAAe,GAAI,IAAO,IAG/C,MAAO,GAST,cAGE,OAFM,GAAc,GAAI,aAAY,IAE3B,EAAI,EAAG,EAAI,GAAI,IACtB,EAAY,GAAK,KAEnB,SAAY,GAAK,EAAY,IAAM,EAE5B,gBAeP,GAAM,GAAe,KACf,EAAgB,KAChB,EAAc,KAEpB,MAAO,UAAC,GAGN,OAFM,GAAS,GAAI,aAAY,EAAI,EAAe,QAC5C,EAAmB,GAAI,aAAY,GAChC,EAAQ,EAAG,EAAQ,EAAe,OAAQ,KACjD,GAAM,GAAc,EAAe,GAC7B,EACF,EAAa,EAAY,GAAe,IAAO,GAAc,OAC7D,EAAc,GAAe,IACjC,EAAiB,GAAS,EAE5B,MAAO,IAAI,cAAa,IChf5B,AAsBA,GAAA,IAAA,WAOE,aACE,KAAK,YAAc,GACnB,KAAK,YAAc,GAGN,SAAA,YAAf,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UASnB,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAS3C,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAW3C,EAAA,gBAAP,SAAuB,GACrB,MAAO,GAAiB,YAAY,EAAK,SAWpC,EAAA,gBAAP,SAAuB,EAAsB,GAE3C,MAAO,GAAiB,YAAY,EAAK,OAAQ,IAGpC,EAAA,YAAf,SACI,EAAsB,EACtB,GACF,GAAM,GAA6B,GAC7B,EAAU,IAAgB,OAC5B,EAAiB,cAAc,YAC/B,EAAiB,cAAc,YACnC,SAAQ,QAAQ,SAAA,GACd,GAAM,GAAU,EAAO,EAAK,GAC5B,AAAI,IAAY,MACd,EAAc,KAAK,KAGhB,GAEX,KAEa,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAkB,SAAC,GAC5B,MAAA,IAAiB,gBAAgB,IACxB,GACT,SAAC,EAAsB,GACnB,MAAA,IAAiB,gBAAgB,EAAK,IC/G9C,AAyBA,GAAM,IAAgB,eAChB,GAAmB,EAKnB,GAAmB,eAInB,GAAkB,mBAexB,cACE,GAAI,CAAC,KAAM,QAAQ,cAIjB,KAAM,IAAI,OACN,2FAIN,GAAM,GAAiB,MAAO,SAAW,YAAc,KAAO,OACxD,EAAU,EAAU,WAAa,EAAU,cAC7C,EAAU,iBAAmB,EAAU,aACvC,EAAU,cACd,GAAI,GAAW,KACb,KAAM,IAAI,OACN,6DAEN,MAAO,GAGT,YAAuB,GACrB,GAAM,GAAK,EAAY,OACvB,EAAG,kBAAkB,GAAkB,CAAC,QAAS,cACjD,EAAG,kBAAkB,GAAiB,CAAC,QAAS,cAQlD,GAAA,IAAA,WAME,WAAY,GAGV,GAFA,KAAK,UAAY,KAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,kEAEN,KAAK,UAAY,EAGb,SAAA,UAAA,KAAN,SAAW,sEAET,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAIN,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,UAAW,SAIvC,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,iBAiB1B,EAAA,UAAA,eAAR,SAAuB,EAAmB,GAA1C,GAAA,GAAA,KAEE,MAAO,IAAI,SAAmC,SAAC,EAAS,GACtD,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OAEvB,GAAI,GAAkB,MAEpB,GAAM,GAAU,EAAG,YAAY,GAAkB,YAC3C,EAAa,EAAQ,YAAY,IACjC,EAAa,EAAW,IAAI,EAAK,WACvC,EAAW,UAAY,WACrB,GAAI,EAAW,QAAU,KACvB,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAK,UAAS,oBAGlD,EAAQ,EAAW,OAAO,iBAG9B,EAAW,QAAU,SAAA,GACnB,SAAG,QACI,EAAO,EAAW,QAE3B,EAAQ,WAAa,WAAM,MAAA,GAAG,cAG9B,GAAM,GACF,GAA6B,GAE3B,EAAS,EAAG,YAAY,GAAiB,aAC3C,EAAY,EAAO,YAAY,IAC7B,EACF,EAAU,IAAI,CAAC,UAAW,EAAK,UAAW,mBAAkB,IAC5D,EACJ,EAAe,UAAY,WAEzB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAkB,EAAW,IAAI,CACrC,UAAW,EAAK,UAChB,eAAc,EACd,mBAAkB,IAEpB,EAAgB,UAAY,WAAM,MAAA,GAAQ,CAAC,mBAAkB,KAC7D,EAAgB,QAAU,SAAA,GAGxB,EAAY,EAAO,YAAY,IAC/B,GAAM,GAAoB,EAAU,OAAO,EAAK,WAChD,EAAkB,UAAY,WAC5B,SAAG,QACI,EAAO,EAAgB,QAEhC,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAgB,UAIpC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAE/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,YAKtC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,WAvHtC,EAAA,WAAa,eA0H/B,KAEa,GAA4B,SAAC,GACxC,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAmBH,GAC/B,MAAO,IAAI,IAAiB,GAG9B,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAiB,YACnC,EAAI,MAAM,GAAiB,WAAW,QACtC,EAGN,GAAA,IAAA,WAGE,aACE,KAAK,UAAY,KAGb,SAAA,UAAA,WAAN,yFACE,MAAA,CAAA,EAAO,GAAI,SACP,SAAC,EAAS,GACR,GAAM,GACF,EAAK,UAAU,KAAK,GAAe,IACvC,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAK,EAAG,YAAY,GAAiB,YACrC,EAAQ,EAAG,YAAY,IASvB,EAAqB,EAAc,SACzC,EAAkB,UAAY,WAE5B,OADM,GAA4C,GAC/B,EAAA,EAAA,EAAA,EAAkB,OAAlB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAI,EAAK,WAAa,EAAK,mBAE7B,EAAQ,IAEV,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAkB,QAElC,EAAG,WAAa,WAAM,MAAA,GAAG,UAE3B,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAIpD,EAAA,UAAA,YAAN,SAAkB,iFAChB,SAAO,GAAiB,GACxB,CAAA,EAAO,GAAI,SAA4B,SAAC,EAAS,GAC/C,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAS,EAAG,YAAY,GAAiB,aACzC,EAAY,EAAO,YAAY,IAE/B,EAAiB,EAAU,IAAI,GACjC,EACJ,EAAe,UAAY,WACzB,GAAI,EAAe,QAAU,KAC3B,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAI,oBAIxC,GAAM,GAAoB,EAAU,OAAO,GACrC,EAAkB,WAEtB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAqB,EAAW,OAAO,GAC7C,EAAmB,UAAY,WAC3B,MAAA,GAAQ,EAAe,OAAO,qBAClC,EAAmB,QAAU,SAAA,GACzB,MAAA,GAAO,EAAe,SAI5B,EAAkB,UAAY,EAC9B,EAAkB,QAAU,SAAA,GAC1B,WACA,EAAG,QACI,EAAO,EAAe,SAInC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAG/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,WAIpC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAGxD,KCjWA,AAyBA,GAAM,IAAiB,IACjB,GAAc,sBACd,GAAc,OACd,GAAwB,iBACxB,GAAsB,eACtB,GAAqB,cACrB,GAAwB,iBA8B9B,YAAsB,GAOpB,MAAO,CACL,KAAM,CAAC,GAAa,EAAM,IAAa,KAAK,IAC5C,SAAU,CAAC,GAAa,EAAM,IAAuB,KAAK,IAC1D,YAAa,CAAC,GAAa,EAAM,IAAqB,KAAK,IAC3D,WAAY,CAAC,GAAa,EAAM,IAAoB,KAAK,IACzD,cACI,CAAC,GAAa,EAAM,IAAuB,KAAK,KAWxD,YAA6B,GAC3B,GAAM,GAAQ,EAAI,MAAM,IACxB,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OAAM,uBAAuB,GAEzC,MAAO,GAAM,MAAM,EAAG,EAAM,OAAS,GAAG,KAAK,IAG/C,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAoB,YACtC,EAAI,MAAM,GAAoB,WAAW,QACzC,EAgBN,GAAA,IAAA,WAOE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,eAAiB,MAAO,SAAW,aAClD,MAAO,QAAO,cAAiB,YAKjC,KAAM,IAAI,OACN,2DAIN,GAFA,KAAK,GAAK,OAAO,aAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,sEAEN,KAAK,UAAY,EACjB,KAAK,KAAO,GAAa,KAAK,WAY1B,SAAA,UAAA,KAAN,SAAW,gFACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAGE,EAAW,KAAK,UAAU,EAAe,eACzC,EAAc,KAAK,UAAU,EAAe,aAE5C,EACF,GAA6B,GAEjC,IACE,YAAK,GAAG,QAAQ,KAAK,KAAK,KAAM,KAAK,UAAU,IAC/C,KAAK,GAAG,QAAQ,KAAK,KAAK,SAAU,GACpC,KAAK,GAAG,QAAQ,KAAK,KAAK,YAAa,GACvC,KAAK,GAAG,QACJ,KAAK,KAAK,WACV,GAA0B,EAAe,aAC7C,KAAK,GAAG,QAAQ,KAAK,KAAK,cAAe,KAAK,UAAU,CACtD,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,uBAGtC,CAAA,EAAO,CAAC,mBAAkB,UACnB,GAEP,WAAK,GAAG,WAAW,KAAK,KAAK,MAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,UAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,aAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,YAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,eAEvB,GAAI,OACN,yBAAyB,KAAK,UAAS,sFAEvC,uBAAsB,EAAmB,mBAAkB,MAC3D,qBAAoB,EAAmB,iBAAgB,MACvD,oBAAmB,EAAmB,gBAAe,qBAazD,EAAA,UAAA,KAAN,gGAGE,GAFM,EACF,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,OACrC,GAAQ,KACV,KAAM,IAAI,OACN,kDAAkD,KAAK,UAAS,KAGtE,GAAI,EAAK,oBAAsB,OAC7B,KAAM,IAAI,OACN,6EAQN,GAJM,EAAsB,GAGtB,EAAW,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,WAClD,GAAY,KACd,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAS,iBAOhE,GAJA,EAAI,cAAgB,EAGd,EAAc,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,cACrD,GAAe,KACjB,KAAM,IAAI,OACN,gDAAgD,KAAK,UAAS,kBAiBpE,GAdA,EAAI,YAAc,EAGZ,EAAiB,KAAK,GAAG,QAAQ,KAAK,KAAK,eACjD,AAAI,GAAkB,MACd,GAAW,KAAK,MAAM,GAC5B,EAAI,OAAS,EAAS,OACtB,EAAI,YAAc,EAAS,YAC3B,EAAI,YAAc,EAAS,YAC3B,EAAI,oBAAsB,EAAS,qBAI/B,EAAmB,KAAK,GAAG,QAAQ,KAAK,KAAK,YAC/C,GAAoB,KACtB,KAAM,IAAI,OACN,uDACA,KAAI,KAAK,UAAS,mBAExB,SAAI,WAAa,GAA0B,GAE3C,CAAA,EAAO,QAzIO,EAAA,WAAa,kBA2I/B,KAEa,GAA+B,SAAC,GAC3C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAoB,aACrD,GACH,EAAI,MAAM,GAAoB,WAAW,SAJxC,MAUX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBA0BA,GAClC,MAAO,IAAI,IAAoB,GAGjC,GAAA,IAAA,WAGE,aACE,EACI,KAAM,QAAQ,cACd,WAAM,MAAA,6CACV,EACI,MAAO,SAAW,aACd,MAAO,QAAO,cAAiB,YACnC,WAAM,MAAA,4DACV,KAAK,GAAK,OAAO,aAGb,SAAA,UAAA,WAAN,8FAIE,IAHM,EAA4C,GAC5C,EAAS,GAAc,GACvB,EAAS,GAAiB,GACvB,EAAI,EAAG,EAAI,KAAK,GAAG,OAAQ,EAAE,EAC9B,EAAM,KAAK,GAAG,IAAI,GACxB,AAAI,EAAI,WAAW,IAAW,EAAI,SAAS,IACnC,GAAY,GAAoB,GACtC,EAAI,GAAa,KAAK,MAAM,KAAK,GAAG,QAAQ,KAGhD,MAAA,CAAA,EAAO,QAGH,EAAA,UAAA,YAAN,SAAkB,8EAGhB,GAFA,EAAO,GAAiB,GAClB,EAAO,GAAa,GACtB,KAAK,GAAG,QAAQ,EAAK,OAAS,KAChC,KAAM,IAAI,OAAM,8BAA8B,EAAI,KAE9C,SAAO,KAAK,MAAM,KAAK,GAAG,QAAQ,EAAK,OAE7C,KAAK,GAAG,WAAW,EAAK,MACxB,KAAK,GAAG,WAAW,EAAK,UACxB,KAAK,GAAG,WAAW,EAAK,aACxB,KAAK,GAAG,WAAW,EAAK,YACxB,CAAA,EAAO,QAEX,KCvVA,AAiCA,GAAM,IAAoB,MAE1B,GAAA,WAME,aACE,KAAK,SAAW,GAGH,SAAA,YAAf,WACE,MAAI,GAA0B,UAAY,MACxC,GAA0B,SAAW,GAAI,IAEpC,EAA0B,UAS5B,EAAA,gBAAP,SAAuB,EAAgB,GACrC,EAAO,GAAU,KAAM,WAAM,MAAA,0CAC7B,AAAI,EAAO,SAAS,KAClB,GAAS,EAAO,MAAM,EAAG,EAAO,QAAQ,MAE1C,EAAO,EAAO,OAAS,EAAG,WAAM,MAAA,wCAChC,GAAM,GAAW,EAA0B,cAC3C,EACI,EAAS,SAAS,IAAW,KAC7B,WAAM,MAAA,2DACF,EAAM,OACd,EAAS,SAAS,GAAU,GAGvB,EAAA,WAAP,SAAkB,GAChB,GAAM,GAAU,KAAK,cAAc,SAAS,GAC5C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,yCAAyC,EAAM,KAEjE,MAAO,IAGF,EAAA,WAAP,WACE,MAAO,QAAO,KAAK,KAAK,cAAc,WAE1C,KAUA,YAAkB,GAChB,GAAI,EAAI,QAAQ,MAAuB,GACrC,KAAM,IAAI,OACN,6EAEA,IAAG,GAA0B,aAAa,KAAK,OAErD,MAAO,CACL,OAAQ,EAAI,MAAM,IAAmB,GACrC,KAAM,EAAI,MAAM,IAAmB,IAIvC,YACI,EAAmB,EACnB,GAAA,MAAA,KAAA,QAAA,GAAA,6GACF,SACI,IAAc,EACd,WAAM,MAAA,wCAAwC,EAAS,MAErD,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,kEACF,EAAS,MACjB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,iCAAgC,EAAS,OAC3C,EAAc,EAAa,GAE3B,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,mEACF,QAAO,EAAO,OACtB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,sCAAqC,EAAO,OAC9C,EAAc,EAAa,GAE3B,EAAe,GAAS,GAAW,OACnC,EAAa,GAAS,GAAW,KACjC,EAAa,IAAiB,GAAS,GAAW,OAEjC,CAAA,EAAM,EAAY,qBAAnC,GAAiB,EAAA,OAKnB,GAAgB,EAClB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIiB,MAAA,CAAA,EAAM,EAAY,KAAK,iBAApC,GAAa,EAAA,OAKf,GAAgB,CAAC,EACnB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIF,MAAA,CAAA,EAAO,EAAW,yBAsCpB,4HACQ,EAAU,GAA0B,aACpC,EAA2C,OAC5B,EAAA,yBAAA,GAAA,EAAA,OAAV,GAAM,EAAA,GAEX,CAAA,EAAM,GAA0B,WAAW,GAAQ,eAF7B,CAAA,EAAA,UACpB,EACF,EAAA,OACJ,IAAW,IAAQ,GACX,EAAM,EAAS,GAAoB,EACzC,EAAI,GAAO,EAAU,oBALJ,wBAQrB,MAAA,CAAA,EAAO,QAoCT,YAA2B,8EACnB,SAAgB,GAAS,GACzB,EAAU,GAA0B,WAAW,EAAc,QACnE,CAAA,EAAO,EAAQ,YAAY,EAAc,WAkD3C,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QAiDhD,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QChWhD,AA0BA,GAAA,IAAA,WAAA,cAKE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,OAAM,EAAM,IAGrB,EAAA,UAAA,IAAA,WACE,MAAO,aAAY,OAGrB,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,kDAAkD,GAExD,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAAI,cAElB,KAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,IAAI,aAAY,GAAU,OAAO,IAE5C,KAEA,GAAI,KAAM,IAAI,eACZ,KAAM,YAAY,UAAW,GAAI,KAGjC,IACE,GAA0B,gBACtB,GAAoB,WAAY,GAAI,WACjC,IAIT,IACE,GAA0B,gBACtB,GAAiB,WAAY,GAAI,WAC9B,KCpEX,AAqBO,GAAM,IAAe,CAE1B,YAAa,WAAM,MAAA,QAIjB,GAaJ,GAAA,WAKE,aAEE,KAAK,KAAO,KAGZ,KAAK,YAAc,GAAI,MAAK,KAAK,YAGnC,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAI,MAAM,OAAO,OAAS,KACjB,KAAM,OAAO,MAAM,EAAM,GAGlC,CAAI,IAAe,MACjB,IAAc,GAAa,eAEtB,GAAY,EAAM,KAG3B,EAAA,UAAA,IAAA,WACE,GAAM,GAAO,QAAQ,SACrB,MAAO,GAAK,GAAK,IAAO,EAAK,GAAK,KAGpC,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,sDAAsD,GAE5D,MAAO,MAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAI,GAAM,SAAW,EACZ,GAEF,GAAI,MAAK,KAAK,YAAY,GAAU,OAAO,IAEtD,KAEA,AAAI,KAAM,IAAI,YACZ,KAAM,YAAY,OAAQ,GAAI,KCrFhC,YAgDI,EAAoB,EACpB,GADoB,MAAA,KAAA,QAAA,GAAW,WAEjC,EAAQ,GAAS,UACjB,GAAwC,GACjC,GAAI,IAAmB,EAAO,EAAO,GCpD9C,AAuCA,YAAiC,EAAiB,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAGnC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,mCAAmC,GAErD,GAAI,IAAU,UAAY,EAAG,QAAU,UACnC,IAAU,UAAY,EAAG,QAAU,SACrC,KAAM,IAAI,OAAM,yCAGlB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAGhB,IAAO,EAAG,CAAC,MAAK,KC3D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,QAAS,MACtC,EAAU,WACZ,MAAA,GAAO,qBAAqB,EAAG,OAAQ,EAAG,MAAO,EAAG,QAElD,EAAyB,CAAC,EAAG,GAInC,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAQ,EAAG,CAAC,OAAM,KCrD/B,YAgCwC,EAAM,GAAA,AAAA,IAAA,QAAA,GAAA,IAC5C,QAAQ,IAAI,EAAE,SAAS,ICjCzB,AAqBA,KAcA,GAAM,IAAuB,CAC3B,OAAM,GACN,KAAI,GACJ,MAAK,GACL,MAAK,IAEP,GAAa,ICzCb,AA6BA,GAAM,IAA2B,QAC3B,GAA8B,QAC9B,GAAqC,eAE3C,YAAkB,GAChB,MAAO,IAAI,SAAQ,SAAA,GAAW,MAAA,YAAW,KAAU,KAAK,GAG1D,GAAA,IAAA,WAQE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,cAGjB,KAAM,IAAI,OACN,uFAIN,AAAI,EAAe,WAAW,EAAiB,aAC7C,GAAiB,EAAe,MAAM,EAAiB,WAAW,SAEpE,AAAI,IAAkB,MAAQ,EAAe,SAAW,IACtD,GAAiB,IAGnB,KAAK,sBAAwB,EAAiB,GAC9C,KAAK,mBACD,EAAiB,GAGjB,SAAA,UAAA,KAAN,SAAW,6GACT,GAAI,MAAQ,WAAc,YACxB,KAAM,IAAI,OACN,8FAGA,EAAa,OAAO,IAAI,gBAAgB,GAAI,MAC9C,CAAC,EAAe,YAAa,CAAC,KAAM,+BAEpC,GAAe,wBAAyB,cAAxC,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,OACN,gGAGE,SAAyC,CAAC,CAC9C,MAAO,CAAC,KAAO,KAAK,oBACpB,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,gBAAe,GAEX,EACF,OAAO,IAAI,gBAAgB,GAAI,MAC3B,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,sBAIT,EAAa,KAAK,YAAc,KAAO,SAAS,cAAc,KACvB,KAAK,WAClD,EAAW,SAAW,KAAK,sBAC3B,EAAW,KAAO,EAIlB,CAAA,EAAM,GAAM,WAAM,MAAA,GAAW,cAAc,GAAI,YAAW,0BAA1D,GAAA,OAEI,EAAe,YAAc,KACzB,GAAmB,KAAK,kBAAoB,KAC9C,SAAS,cAAc,KACvB,KAAK,iBACT,EAAiB,SAAW,KAAK,mBACjC,EAAiB,KAAO,EACxB,CAAA,EAAM,GACF,WAAM,MAAA,GAAiB,cAAc,GAAI,YAAW,cAPtD,CAAA,EAAA,UAMF,EAAA,wBAIF,MAAA,CAAA,EAAO,CAAC,mBAAoB,GAA6B,WA1E7C,EAAA,WAAa,eA6E/B,KAEA,GAAA,WAGE,WAAY,GACV,GAAI,GAAS,MAAQ,EAAM,OAAS,EAClC,KAAM,IAAI,OACN,2DACA,iBAAgB,IAEtB,KAAK,MAAQ,EAGT,SAAA,UAAA,KAAN,6FACQ,SAAW,KAAK,MAAM,GACtB,EAAc,KAAK,MAAM,MAAM,GAErC,CAAA,EAAO,GAAI,SAAwB,SAAC,EAAS,GAC3C,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GAEnB,GAAM,GAAY,KAAK,MAAO,EAAM,OAAe,QAC7C,EAAgB,EAAU,cAChC,GAAI,GAAiB,MACnB,EAAO,GAAI,OACP,4CAA4C,EAAS,OACzD,OAGF,AAAI,EAAY,SAAW,GACzB,EAAQ,CAAC,cAAa,IAGxB,GAAM,GAAkB,EAAU,gBAClC,GAAI,GAAmB,MACrB,EAAO,GAAI,OACP,6CAA6C,EAAS,OAC1D,OAGF,GAAI,GACJ,IACE,EACI,EAAK,4BAA4B,EAAiB,SAC/C,GACP,EAAO,GACP,OAGF,GAAM,GAAsC,GACtC,EAAkB,GAClB,EAAgC,GACtC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,EAAM,KAAK,GACX,EAAe,KAAK,QAEtB,EAAY,KAAI,MAAhB,EAAoB,EAAa,WAGnC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,GAAM,GAAmB,GAAI,YAC7B,EAAiB,OAAS,SAAC,GAEzB,GAAM,GAAc,EAAM,OAAe,OACnC,EAAQ,EAAM,QAAQ,GAC5B,EAAe,GAAS,EACxB,AAAI,EAAe,QAAQ,QAAU,IACnC,EAAQ,CACN,cAAa,EACb,YAAW,EACX,WAAY,GAAwB,GACpC,OAAQ,EAAU,OAClB,YAAa,EAAU,YACvB,YAAa,EAAU,YACvB,oBAAqB,EAAU,uBAIrC,EAAiB,QAAU,SAAA,GACvB,MAAA,GAAO,6CAA6C,EAAI,OAC5D,EAAiB,kBAAkB,EAAW,SAIpD,EAAW,QAAU,SAAA,GAAS,MAAA,GAC1B,2DACA,eAAc,EAAS,KAAI,qCAC3B,yCACJ,EAAW,WAAW,WAOlB,EAAA,UAAA,4BAAR,SACI,EAAiC,GAInC,OAHM,GAAsB,GACtB,EAAY,EAAM,IAAI,SAAA,GAAQ,MAAA,IAAS,EAAK,QAC5C,EAAqC,GACvB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAM,MAAM,QAAQ,SAAA,GAClB,GAAM,GAAe,GAAS,GAC9B,GAAI,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,sDACA,KAAI,EAAY,MAGtB,GADA,EAAU,KAAK,GACX,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,8BAA8B,EAAY,sBAE9C,EAAW,GAAQ,EAAM,EAAU,QAAQ,MAKjD,GAAI,EAAU,SAAW,EAAM,OAC7B,KAAM,IAAI,OACN,uDACA,KAAI,EAAU,OAAM,8CACpB,KAAI,EAAM,OAAM,OAEtB,MAAO,IAEX,KAEa,GAAmC,SAAC,GAC/C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,gBAyCH,GAAA,MAAA,KAAA,QAAA,GAAA,SACxB,GAAI,IAAiB,eA2CD,GAC3B,MAAO,IAAI,IAAa,GC5V1B,YA8BI,EAAmC,EACnC,EAAwB,GAC1B,EAAc,GACd,EAAgB,GAAwB,EACxC,EAAc,GAAsB,EACpC,EAAc,EAAe,GAC7B,GAAI,GAAkB,EAEhB,EAAkB,SAAC,GACvB,SAAQ,KAAK,SAAA,GACX,GAAM,GAAW,EACb,GAAE,EAAkB,EAAS,OAAU,GAAc,GAEzD,SAAW,GACJ,IAEF,GAGT,WAAuB,GACrB,EACI,GAAY,MAAQ,MAAM,QAAQ,IAAa,EAAS,OAAS,EACjE,WAAM,MAAA,wCAGZ,WAAuB,EAAuB,GAC5C,EACI,GAAiB,GAAK,GAAiB,EACvC,WAAM,MAAA,kDACF,sBAAqB,KAC7B,EACI,GAAe,GAAK,GAAe,EACnC,WAAM,MAAA,kDACF,oBAAmB,KAC3B,EACI,GAAe,EACf,WAAM,MAAA,uDACF,sBAAqB,EAAa,qBAClC,IAAG,KAGb,MAAO,SAAQ,IAAI,EAAS,IAAI,ICvElC,YAqCI,EAAqB,6HACvB,AAAI,IAAe,MACjB,GAAc,IAGV,EAAY,EAAY,WAAa,KAAO,KAAM,SAAS,MACf,EAAY,UAGxD,EAAW,EAAU,IACvB,SAAA,GACI,MAAA,GAAU,EAAU,EAAY,YAAa,CAAC,SAAU,OAE1D,EAAqB,EACrB,EAAmB,GAEP,EAAY,YAAc,KACxC,CAAA,EAAM,QAAQ,IAAI,IADJ,CAAA,EAAA,UACd,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAU,EAAY,WAAY,EAClC,WAFJ,EAAA,EAAA,8BAFE,GAAS,EAMT,EAAiB,EAAU,IAAI,SAAA,GAAY,MAAA,GAAS,gBAEpD,EAAsB,GACtB,EAAoB,EAEV,EAAY,YAAc,KACtC,CAAA,EAAM,QAAQ,IAAI,IADN,CAAA,EAAA,UACZ,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAgB,EAAY,WAAY,EACxC,WAFJ,EAAA,EAAA,wBAFE,SAAO,EAKb,CAAA,EAAO,oBAaL,EAAiC,EACjC,EACA,GAFiC,MAAA,KAAA,QAAA,GAAA,wEAS7B,SAAe,SAAC,GAClB,MAAA,IAAyB,EAAW,CAAC,YAAW,KAC9C,EAAc,GAAqB,GAEzC,CAAA,EAAO,EAAY,EAAU,EAAgB,oBA4B3C,GADJ,GAAA,GAAA,KAIE,MAAO,UACI,EAAiC,EACjC,GADiC,MAAA,KAAA,QAAA,GAAA,4GAqD1C,GAjDM,EAAyB,EAAS,IAAI,WAAM,MAAA,KAC5C,EAKF,GACE,EACF,GAAe,KAAO,EAAY,IAAI,WAAM,MAAA,KAAS,GACnD,EAAmC,GACzC,EAAS,QAAQ,SAAC,EAAqB,GACrC,GAAI,GAAc,EAClB,EAAoB,QAAQ,QAAQ,SAAA,GAClC,GAAM,GAAY,gBAAkB,GAChC,EAAa,aAAa,MAC1B,EAAa,MAEX,EAAe,GAAqB,GACtC,GAAmB,EAAa,OAE9B,EAA8B,WAClC,EAAuB,GAAc,GACrC,AAAI,EAAoB,IAAe,MACrC,GAAoB,GAAc,IAGpC,EAAoB,GAAY,KAAK,CACnC,cAAe,EACf,YAAW,EACX,UAAW,KAIf,AAAI,GAAe,KACjB,EAAY,QAAQ,SAAC,EAAY,GAC/B,AAAI,IAAe,EAAa,MAC9B,KACA,EAAa,GAAe,MAIhC,IAGF,EAAuB,KAAK,EAAa,MACzC,GAAe,MAIf,CAAC,EAAa,MAAM,SAAA,GAAS,MAAA,KACzB,QAAkB,EAAY,OAAO,SAAC,EAAG,GAAM,MAAA,CAAC,EAAa,KAC7D,GAAI,OACN,kDACG,GAAgB,KAAK,MAAK;GAC7B,yCACG,GAAuB,KAAK,MAAK,MAKpC,SACF,EAAuB,OAAO,SAAC,EAAa,EAAa,GACvD,MAAI,IACF,EAAY,KAAK,GAEZ,GACN,IAED,EAAsB,GAC5B,EAAoB,QAAQ,SAAA,GAC1B,EAAS,GAAG,MAAM,QAAQ,SAAA,GACxB,GAAM,GAAW,EACZ,CAAC,EAAe,SAAS,KAAa,GAAN,KAAY,EACjD,EAAU,KAAK,OAGH,CAAA,EAAM,EAAqB,WAArC,SAAU,EAAA,OAEV,EAAmC,GACrC,EAAoB,EACxB,EAAoB,QAAQ,SAAA,GAI1B,OAHM,GAAa,EAAS,GAAG,MAAM,OAEjC,EAAa,EACR,EAAI,EAAG,EAAI,EAAY,IAC9B,GAAc,EAAQ,EAAoB,GAAG,WAO/C,OAHM,GAAc,GAAI,aAAY,GAC9B,EAAkB,GAAI,YAAW,GACnC,EAAoB,EACf,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAS,GAAI,YAAW,EAAQ,EAAoB,IAC1D,EAAgB,IAAI,EAAQ,GAC5B,GAAqB,EAAO,WAG9B,GAAM,GAAiB,EAAoB,GAC3C,EAAe,QAAQ,SAAA,GACrB,GAAM,GAAa,EAAY,MAC3B,EAAa,YACb,EAAa,YAAc,EAAa,WACtC,EACF,GAAc,EAAY,CAAC,EAAa,gBAC5C,OAAW,KAAQ,GACjB,EAAiB,GAAQ,EAAgB,KAI7C,GAAqB,IAGvB,CAAA,EAAO,SCpPX,AA+BA,GAAM,IAAyB,2BACzB,GAAY,mBAClB,GAAA,WAcE,WAAY,EAAc,GAgCxB,GAvCO,KAAA,eAAiB,OAQxB,AAAI,GAAe,MACjB,GAAc,IAEhB,KAAK,iBAAmB,EAAY,iBACpC,KAAK,WAAa,EAAY,WAC9B,KAAK,mBAAqB,EAAY,mBAEtC,AAAI,EAAY,WAAa,KAC3B,GACI,MAAO,GAAY,WAAc,WACjC,WAAM,MAAA,gIAGV,KAAK,MAAQ,EAAY,WAEzB,KAAK,MAAQ,KAAM,SAAS,MAG9B,EACI,GAAQ,MAAQ,EAAK,OAAS,EAC9B,WAAM,MAAA,4DAGV,AAAI,MAAM,QAAQ,IAChB,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,+CACF,sBAAqB,EAAK,OAAM,QAE1C,KAAK,KAAO,EAER,EAAY,aAAe,MAC3B,EAAY,YAAY,MAAQ,KAClC,KAAM,IAAI,OACN,sEAEN,KAAK,YAAc,EAAY,aAAe,GAG1C,SAAA,UAAA,KAAN,SAAW,yGACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,2FAIA,SAAO,OAAO,OAAO,CAAC,OAAQ,KAAK,gBAAiB,KAAK,aAC/D,EAAK,KAAO,GAAI,UAEV,EAAyC,CAAC,CAC9C,MAAO,CAAC,uBACR,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,oBACpC,gBAAe,GAGjB,EAAK,KAAK,OACN,aACA,GAAI,MACA,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,KACX,cAEJ,AAAI,EAAe,YAAc,MAC/B,EAAK,KAAK,OACN,oBACA,GAAI,MAAK,CAAC,EAAe,YAAa,CAAC,KAAM,KAC7C,qBAGW,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,WAE7C,GAFM,EAAW,EAAA,OAEb,EAAS,GACX,MAAA,CAAA,EAAO,CACL,mBAAoB,GAA6B,GACjD,UAAW,CAAC,KAGd,KAAM,IAAI,OACN,gEACG,GAAS,OAAM,YAYpB,EAAA,UAAA,KAAN,uIAC6B,MAAA,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,KAAK,qBAE5D,GAFM,EAAqB,EAAA,OAEvB,CAAC,EAAmB,GACtB,KAAM,IAAI,OACN,cAAc,KAAK,KAAI,4BACpB,GAAmB,OAAM,uCAC5B,sFAIU,CAAA,EAAM,EAAmB,eAAvC,SAAc,EAAA,qCAEV,EAAU,+CAA+C,KAAK,KAAI,IAGtE,AAAI,KAAK,KAAK,SAAS,OACrB,GAAW,+UAOX,GAAW,uEAGP,GAAI,OAAM,UAUlB,GARM,EAAgB,EAAY,cAC5B,EAAkB,EAAY,gBAC9B,EAAc,EAAY,YAC1B,EAAc,EAAY,YAC1B,EAAS,EAAY,OACrB,EAAsB,EAAY,oBAGpC,GAAiB,MAAQ,GAAmB,KAC9C,KAAM,IAAI,OACN,2BAA2B,KAAK,KAAI,mEAMtC,IAAmB,KACL,CAAA,EAAM,KAAK,YAAY,IADrC,CAAA,EAAA,UACI,EAAU,EAAA,OACf,EAAA,EAAA,GAAa,EAAA,EAAA,oBAGV,SAA4B,CAChC,cAAa,EACb,YAAW,EACX,WAAU,EACV,oBAAmB,EACnB,YAAW,EACX,YAAW,EACX,OAAM,GAGF,EAAc,EAAY,iBAChC,AAAI,GACF,GAAU,iBAAmB,GAG/B,CAAA,EAAO,SAGK,EAAA,UAAA,YAAd,SAA0B,2IAOxB,IALM,EAAa,MAAM,QAAQ,KAAK,MAAQ,KAAK,KAAK,GAAK,KAAK,KAC5D,EAAmB,GAAS,GAA3B,EAAM,EAAA,GAAE,EAAM,EAAA,GACf,EAAa,KAAK,kBAAoB,EAEtC,EAAc,GACpB,EAAA,EAAoB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAT,EAAK,EAAA,GACd,EAAY,KAAI,MAAhB,EAAoB,EAAM,SAK5B,IAFM,EAAsB,GACtB,EAAsC,GAC5C,EAAA,EAA2B,EAAA,EAAA,EAAA,EAAA,OAAA,IACzB,IADS,EAAY,EAAA,GACrB,EAAA,EAAmB,EAAA,EAAa,MAAb,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,AAAI,KAAK,oBAAsB,KAC7B,EAAY,KAAK,KAAK,mBAAmB,IAEzC,EAAU,KAAK,EAAa,EAAO,SAKrC,MAAK,sBACP,GAAA,EAAU,MAAI,SAAd,GAAkB,CAAA,EAAM,QAAQ,IAAI,KADlC,CAAA,EAAA,UACF,EAAA,MAAA,EAAA,EAAA,OAAA,CAAkB,EAAA,2BAGJ,MAAA,CAAA,EAAM,GAAyB,EAAW,CACxD,YAAa,KAAK,YAClB,UAAW,KAAK,MAChB,WAAY,KAAK,qBAHb,SAAU,EAAA,OAKhB,CAAA,EAAO,CAAC,EAAa,GAAwB,WA5M/B,EAAA,iBAAmB,eA8MrC,iBAayB,GACvB,GAAM,GAAY,EAAI,YAAY,KAC5B,EAAkB,EAAI,YAAY,KAClC,EAAS,EAAI,UAAU,EAAG,GAC1B,EACF,EAAkB,EAAY,EAAI,UAAU,GAAmB,GACnE,MAAO,CAAC,EAAS,IAAK,eAGK,GAC3B,MAAO,GAAI,MAAM,GAAY,mBAAqB,KAG7C,GAAM,IACT,SAAC,EAAa,GACZ,GAAI,MAAO,QAAU,aAChB,IAAe,MAAQ,EAAY,WAAa,MAInD,MAAO,MAEP,GAAI,GAAS,GAMb,MALA,AAAI,OAAM,QAAQ,GAChB,EAAS,EAAI,MAAM,SAAA,GAAW,MAAA,IAAa,KAE3C,EAAS,GAAa,GAEpB,EACK,GAAK,EAAK,GAGd,MAEb,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAwEf,EAAc,GACjC,MAAO,IAAI,IAAY,EAAM,eAS3B,EAAc,GAChB,MAAO,IAAK,EAAM,GC3XpB,AAuBA,GAAA,IAAA,WACE,WAA6B,GAAA,KAAA,eAAA,EAEvB,SAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,qBAEhB,KAEA,GAAA,WACE,WACqB,GAAA,KAAA,YAAA,EAGf,SAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,YAAY,SAE5B,iBAwBI,EAAmC,EACnC,EAA0B,GAC5B,GAAI,UAAU,SAAW,GACvB,GAAM,GACD,EAAkC,eAAiB,MACnD,EAAkC,aAAe,KACtD,MAAI,GACK,GAAI,IAAkB,GAI7B,SAAQ,KACJ,yNAIG,GAAI,IAAkB,CAAC,cAAe,SAK/C,gBAAQ,KACJ,yNAIG,GAAI,IAAkB,CAC3B,cAAe,EACf,YAAW,EACX,WAAU,EACV,eAAc,gBAqBhB,GAEF,MAAO,IAAI,IAAiB,GCpH9B,oYCAA,AAuDA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MAExC,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,MAAK,GAC5B,EACW,SAAC,EAAwB,GACxC,SAAQ,GAA4B,EAAO,EAAG,MAC9C,EACI,EAAG,OAAS,GAAmB,GAC/B,WAAM,MAAA,mEACV,EAAK,CAAC,IACC,EAAQ,QAAQ,EAAI,IAE7B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAEO,GAAU,EAAG,CAAC,SAAQ,KC1EnC,AA6CA,YACI,EAAiB,EAAiB,EAClC,SADkC,AAAA,IAAA,QAAA,GAAA,IAClC,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAK,EAAgB,EAAG,IAAK,UACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WAAM,MAAA,kEACF,cAAa,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE3C,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,sCAAsC,EAAU,UAC/C,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,kCAAkC,EAAW,UAC5C,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACxD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAExD,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,IAEJ,EAAQ,YACX,EAAiB,EAAiB,EAAY,IAG9C,EAA4B,CAAC,EAAG,EAAK,EAAG,GAExC,EAA0B,CAAC,WAAU,EAAE,WAAU,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAO,GAAQ,EAAK,MAGT,IAAS,EAAG,CAAC,QAAO,KC9GjC,AA+CA,YACI,EAA4B,EAAe,EAC3C,GACF,GAF6C,AAAA,IAAA,QAAA,GAAA,GAC3C,AAAA,IAAA,QAAA,GAAA,GACE,EAAQ,EACV,KAAM,IAAI,OAAM,iDAAiD,GAEnE,GAAM,GAAW,EAAgB,EAAS,UAAW,SAAU,SACzD,EAAe,EAAS,MAAK,OAAA,CAAE,IAE/B,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,IACC,EACH,EAAQ,OACJ,EAAQ,EAAU,CAAC,EAAS,OAAQ,EAAO,EAAS,GACxD,IAGA,EAAuB,CAAC,QAAS,GACjC,EAAqB,CAAC,MAAK,EAAE,QAAO,EAAE,SAAQ,GAEpD,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAS,EAAG,CAAC,QAAO,KCxEjC,AA+CA,YAAsC,EAAiB,GACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAgBnC,GAdA,AAAI,GAAQ,MACV,GAAO,EAAG,MAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAG,WAEnC,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OACzC,EAAK,QAAQ,SAAA,GACX,EACI,GAAQ,GAAK,EAAO,EAAG,KACvB,WAAM,MAAA,+CAA+C,GAAG,KAAO,GAC3D,aAAY,OAGlB,EAAG,MAAQ,EACb,MAAO,GAAG,QAGZ,GAAM,GAA0B,CAAC,EAAG,GAC9B,EAAwB,CAAC,KAAI,GAEnC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IAAO,EACxC,KAAqB,GAAW,MAGzB,IAAY,EAAG,CAAC,WAAU,KC5EvC,YA0DI,EAA6B,EAC7B,GACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,mBAC5C,EACF,EAAgB,EAAa,cAAe,mBAEhD,EACI,GAAc,MAAQ,EAAa,GAAK,OAAO,UAAU,GACzD,WAAM,MAAA,uDACF,YAAW,KACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,gDAAgD,EAAQ,OAClE,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,6CACF,YAAW,EAAa,QAChC,EACI,EAAQ,MAAM,KAAO,EAAa,MAAM,GACxC,WAAM,MAAA,uCACC,GAAQ,MAAM,GAAE,QAAQ,EAAa,MAAM,GAAE,MAChD,oEACR,EACI,EAAa,GAAK,OAAO,UAAU,GACnC,WAAM,MAAA,4DACF,IAAG,KAIX,GAAM,GAAe,GAAO,GAAK,EAAS,SAAU,GAC9C,EACF,GAAO,GAAK,EAAc,SAAU,GAClC,EAA0B,GAAU,GAC1C,MAAO,IAAK,GAAO,EAAe,GAAoB,SAGjD,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KC9FnD,2CCAA,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,mDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,oEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,AA6BA,GAAI,IAyBJ,YACI,EAEA,GAEF,GAFE,AAAA,IAAA,QAAA,GAAA,GAEE,EAAc,EAChB,KAAM,IAAI,OACN,kEAEN,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,4DAElB,GAAI,GAAc,GACd,EAAc,GACd,EAAU,GACV,EAAU,GACV,EAAe,GACnB,GAAK,EAAqB,eAAgB,YACxC,EAAc,WAEZ,MAAQ,YAAe,aAAe,YAAkB,WAC1D,EAAc,WAEZ,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAER,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAEA,EAAe,YAAc,KACvC,EAAe,OAEf,MAAM,IAAI,OACN,6OAIA,YAAY,EAAc,YAAY,OAE5C,GAAI,GACF,GAAM,GAAgC,EACtC,GAAI,GACC,EAA4B,WACzB,EACN,KAAM,IAAI,OACN,yGAMR,GAAM,GAAS,GAAU,GAAY,EAAO,aAC5C,GAAI,GAAU,MACZ,GAAM,GAA2B,CAAC,OAAM,GAClC,EAAyB,CAAC,YAAW,GAC3C,MAAO,GAAO,UACH,GAAY,EACZ,GAGP,GAAA,GAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAMV,EAEJ,AAAI,EACF,EAEK,EAAe,WAAW,MAAM,aAAa,EAAG,EAAG,EAAO,GAAQ,KAClE,AAAI,GAAe,EACxB,EAAQ,EAAiC,KACpC,AAAI,IAAW,IACpB,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAEpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA4B,EAAG,EAAG,EAAO,GAC7C,EAAO,GAAoB,aAAa,EAAG,EAAG,EAAO,GAAQ,MAE/D,GAAI,GACJ,GAAI,IAAgB,EAClB,EAAS,GAAI,YAAW,QAExB,GAAM,GAAY,EAAQ,EAC1B,EAAS,GAAI,YAAW,EAAY,GACpC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAU,EAAG,EAAU,EAAa,EAAE,EAC7C,EAAO,EAAI,EAAc,GAAW,EAAK,EAAI,EAAI,GAIvD,GAAM,GAAqC,CAAC,EAAQ,EAAO,GAC3D,MAAO,IAAS,EAAQ,EAAU,qBAuBhC,EACA,iIAQF,GAPI,EAAO,EAAgB,EAAK,MAAO,YACvC,AAAM,YAAe,IAEb,GAAoB,EAC1B,EAAO,GAAK,EAAmB,SAC/B,EAAkB,WAEhB,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,OACN,wDAAwD,EAAK,KAAI,KAKvE,GAHM,EAAkB,EAAK,MAAM,MAAM,EAAG,GAArC,EAAM,EAAA,GAAE,EAAK,EAAA,GACd,EAAQ,EAAK,OAAS,EAAI,EAAI,EAAK,MAAM,GAE3C,EAAQ,GAAK,IAAU,EACzB,KAAM,IAAI,OACN,wCACA,sBAAqB,IAG3B,GAAI,EAAK,QAAU,WAAa,EAAK,QAAU,QAC7C,KAAM,IAAI,OACN,kCAAkC,EAAK,MAAK,0CAIrC,MAAA,CAAA,EAAM,EAAK,eAIxB,IAJM,EAAO,EAAA,OACP,EAAa,EAAK,QAAU,UAAY,IAAM,EAC9C,EAAQ,GAAI,mBAAkB,EAAQ,EAAS,GAE5C,EAAI,EAAG,EAAI,EAAS,EAAO,EAAE,GAGpC,IAFM,EAAO,CAAC,EAAG,EAAG,EAAG,KAEd,EAAI,EAAG,EAAI,EAAO,KAGzB,GAFM,EAAQ,EAAK,EAAI,EAAQ,GAE3B,EAAK,QAAU,WACjB,GAAI,EAAQ,GAAK,EAAQ,EACvB,KAAM,IAAI,OACN,qDACA,kCAAiC,EAAK,cAEnC,EAAK,QAAU,SACpB,GAAQ,GAAK,EAAQ,KACvB,KAAM,IAAI,OACN,mDACA,oCAAmC,EAAK,MAIhD,AAAI,IAAU,EACZ,GAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,GAElB,EAAK,GAAK,EAAQ,EAIhB,EAAI,EAAI,EACd,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAGjC,MAAI,IAAU,MACZ,GAAO,MAAQ,EACf,EAAO,OAAS,EACV,EAAM,EAAO,WAAW,MACxB,EAAY,GAAI,WAAU,EAAO,EAAO,GAC9C,EAAI,aAAa,EAAW,EAAG,IAEjC,AAAI,IAAS,GACX,EAAK,UAEP,CAAA,EAAO,QAGF,GAAM,IAAa,EAAG,CAAC,YAAW,+DCrOrC,EAAgB,GAClB,GAAI,EAAO,KAAO,EAChB,KAAM,IAAI,OACN,0DACA,sBAAqB,EAAO,KAAI,MAEtC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,4DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OACN,sDACA,uBAAsB,EAAQ,MAAK,MAEzC,GAAI,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAAO,KAC3C,KAAM,IAAI,OACN,iEACG,GAAQ,MAAM,EAAQ,KAAO,GAAE,QAAQ,EAAO,OAGvD,GAAI,EAAO,OAAS,EAClB,KAAM,IAAI,OACN,qDACA,kBAAiB,EAAO,MAAK,MASnC,OANM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAIjD,EAAU,EACL,EAAI,EAAG,EAAI,EAAa,OAAS,EAAG,EAAE,EAC7C,GAAW,EAAa,GAG1B,GAAM,GAAa,EAAO,MAEpB,EAAc,EAAa,QACjC,EAAY,MAGZ,OADI,GAAY,EACP,EAAI,EAAW,EAAI,EAAO,KAAM,EAAE,EACzC,GAAa,EAAW,GACxB,EAAY,KAAK,EAAW,IAG9B,GAAM,GACE,GAAe,EAAO,OAAO,IAAI,SAAA,GAAU,MAAA,GAAS,IAAU,OAAA,CACjE,IAAG,MAAM,EAAG,GAEjB,MAAO,CAAC,EAAa,EAAS,EAAW,6DCtDvC,EAAiB,EAAiB,GACpC,GAAM,GAAY,EAAQ,KAAO,EAAK,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAClE,EAAY,EAAQ,KAAO,EAAK,EAAQ,KAAO,EAAI,EAEnD,EAAa,wDACf,yCAAwC,EAAQ,OAChD,qBAAoB,EAAQ,MAAK,YAAY,GAC7C,gBAAe,EAAQ,mBAAmB,EAAQ,KAEtD,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OAAM,EAAa,mBAAkB,EAAQ,OAEzD,GAAI,EAAM,OAAS,EAAY,GAAQ,KAAO,GAC5C,KAAM,IAAI,OACN,EACA,2BAA0B,GAAY,GAAQ,KAAO,MAE3D,GAAI,EAAQ,OAAS,EAAW,EAAM,OAAS,EAC7C,KAAM,IAAI,OACN,EAAa,oBAAmB,GAAW,EAAM,OAAS,KAEhE,OAAS,GAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,GAAI,EAAQ,MAAM,KAAO,EAAQ,MAAM,GACrC,KAAM,IAAI,OACN,EACA,mBAAkB,EAAC,MAAM,EAAQ,MAAM,GAAE,sBAAsB,EAAC,MAC5D,EAAQ,MAAM,GAAE,OAG5B,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAO,EAAU,EAAE,EAC7C,GAAI,EAAQ,MAAM,EAAI,KAAc,EAAM,EAAI,GAC5C,KAAM,IAAI,OACN,EACA,mBAAkB,GAAI,GAAQ,MAC1B,EAAQ,MAAM,EAAI,GAAS,cAAc,GAAI,GAAQ,MACrD,EAAM,EAAI,GAAS,kBAoB7B,EAAiB,EAAiB,GACpC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OAAM,0DACZ,EAAQ,OAEd,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OACN,6DAA6D,GAGnE,GAAI,EAAM,SAAW,GACnB,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAEd,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAIhB,GAAoB,EAAO,EAAS,eAalC,EAAqB,EACrB,GAWF,OATM,GAAc,EAAQ,MAAM,OAC5B,EAAa,EAAc,EAAK,EAAQ,MAAM,EAAc,GAAK,EAKjE,EAAU,EAAM,OAElB,EAAY,EACP,EAAI,EAAW,EAAI,EAAS,EAAE,EACrC,GAAa,EAAM,GAGrB,GAAM,GAAgB,EAAY,EAAK,EAAI,EACrC,EAAa,GAAc,EAAQ,OAAS,EAE5C,EAAc,GAAe,EAAM,MAAM,EAAG,IAAW,OAAA,CAAE,IACzD,EAAa,GAAc,GACjC,MAAO,CAAC,UAAS,EAAE,WAAU,EAAE,UAAS,EAAE,QAAO,EAAE,WAAU,sFCnJ/D,YAqBI,EAAmB,EAAiB,GACtC,GAAM,GAAY,EAAM,MAAM,OAC9B,EACI,IAAc,EAAM,OACpB,WAAM,MAAA,iBAAiB,EAAS,sBAAsB,EAAK,SACvD,iCAAgC,EAAS,QACjD,EACI,IAAc,EAAK,OACnB,WAAM,MAAA,iBAAiB,EAAS,qBAAqB,EAAI,SACrD,iCAAgC,EAAS,QAEjD,mBAAS,GACP,EACI,EAAM,GAAK,EAAK,IAAM,EAAM,MAAM,GAClC,WAAM,MAAA,iBAAiB,EAAS,YAAY,EAAC,YAAY,EAAC,KACtD,KAAI,GAAM,GAAK,EAAK,IAAE,gCAAgC,EAAC,MACjD,EAAM,MAAM,GAAE,QALrB,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,eAUgB,GAGzB,OAFM,GAAO,GACT,EAAO,EACJ,EAAO,GACZ,AAAI,EAAO,GACT,EAAK,KAAK,GAEZ,GAAQ,EACR,IAEF,MAAO,eAKL,EAAiB,EAAe,GAElC,OADM,GAAO,GACJ,EAAO,EAAG,EAAO,EAAM,OAAQ,IACtC,EAAK,GAAQ,KAAK,KAAM,GAAI,GAAQ,EAAM,IAAS,EAAQ,IAE7D,MAAO,eAML,EAAmB,EAAgC,EACnD,GAEF,OADM,GAAiB,EAAO,QACrB,EAAI,EAAW,OAAQ,EAAI,EAAW,OAAQ,IACrD,EAAW,KAAK,GAElB,OAAS,GAAI,EAAG,EAAI,EAAe,IACjC,AAAI,IAAM,EACR,EAAW,GAA0B,EAErC,GAAW,OACP,EAAwB,EACxB,GACJ,EAAW,OAGf,MAAO,GAGT,YACI,EAAgC,EAChC,GACF,MAAI,IAAkB,EACb,EAGF,EAAkB,GAAgB,GAG3C,YAAuB,EAAuB,GAE5C,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAe,IACjC,EAAW,KAAK,EAAyB,GAE3C,MAAO,eAKL,EAAsB,EAAwB,EAC9C,EAAiB,EAAe,EAAmB,EACnD,EACA,GACF,GAAM,GAAY,EAAW,OACzB,EAAkB,GAAI,OAAM,GAC5B,EAAgB,GAAI,OAAM,GAC1B,EAAoB,GAAI,OAAM,GAClC,GAAI,EAAa,QAAU,EAAsB,GAC/C,GAAM,GAAY,EAAa,GAIzB,EAAgB,EAAsB,EAC5C,EAAkB,GACd,EAAW,EAAW,EAAe,EAAO,GAChD,EAAgB,GACZ,EAAS,EAAW,EAAe,EAAK,GAC5C,EACI,GAAsB,EAAS,EAAW,EAAe,OAE7D,QAAS,GAAO,EAAG,EAAO,EAAW,IACnC,EAAgB,GAAQ,GACpB,EAAW,EAAO,EAAS,EAAY,EAAM,GACjD,EAAc,GACV,GAAY,EAAS,EAAK,EAAS,EAAY,EAAM,GACzD,EAAkB,GAAQ,GAAe,EAAS,EAAM,GAI5D,MAAO,CACL,MAAO,EACP,IAAK,EACL,QAAS,eAOT,EAAmB,EAAgC,EACnD,EAAyB,GAI3B,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAEnB,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAc,GAClC,AAAI,EAAY,GAAK,GACnB,GAAgB,GAGlB,EAAW,GAAQ,EAGvB,MAAO,eAML,EAAiB,EAAgC,EACjD,EAAuB,GAIzB,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAAO,sBAE1B,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAY,GAChC,AAAI,EAAU,GAAK,GACjB,GAAgB,OAAO,kBAEzB,EAAW,GAAQ,EAIvB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,KAErC,GAAM,GAAW,EAAW,GAC5B,AAAI,EAAW,GAAK,GAClB,GAAW,IAAM,GAEnB,EAAW,GAAK,GAAW,EAAG,EAAW,GAAI,EAAW,IAE1D,MAAO,eAIL,EAAmB,EAAc,GACnC,GAAI,GAAS,EAAQ,GACrB,MAAI,GAAgB,GAAK,GAAS,GAAU,OAC1C,GAAS,GAGJ,cAIL,EAAmB,EAAwB,EAC3C,EAAsB,EAAc,GAEtC,GAAI,GAAQ,EAAa,GACnB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAY,GAAK,GAAQ,EAAe,GAAK,GAAQ,GAAS,OAChE,CAAI,EAAS,EAIX,EAAQ,OAAO,iBAGf,EAAQ,OAAO,kBAKnB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAQ,GACV,IAAS,GAIX,EAAQ,GAAW,EAAG,EAAO,EAAW,GAEjC,cAIL,EAAiB,EAAuB,EACxC,EAAsB,EAAc,GAEtC,GAAI,GAAO,EAAY,GACjB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAW,GAAK,GAAS,EAAgB,GAAK,GAAS,GAAQ,OACjE,CAAI,EAAS,EAGX,EAAO,OAAO,iBAGd,EAAO,OAAO,kBAKlB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAO,GACT,IAAQ,GAMV,AAAI,EAAS,EAEX,EAAO,GAAW,EAAG,EAAM,GAG3B,EAAO,GAAW,GAAI,EAAM,EAAW,GAGlC,cAQL,EAAiB,EAAiB,GAGpC,OADI,GAAkB,EAAK,OAClB,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,GAAI,EAAK,GAAK,GACZ,EAAkB,EAClB,MAIJ,OAAS,GAAI,EAAkB,EAAG,EAAI,EAAK,OAAQ,IACjD,GAAI,EAAM,GAAK,GAAK,EAAK,KAAO,EAAM,GACpC,MAAO,GAGX,MAAO,eAGyB,EAAiB,GAEjD,OADI,GAAa,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EACrD,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAc,EAAM,GAAK,EAAQ,GAEnC,MAAO,eAIL,EAAe,EAAwB,GAEzC,GAAI,GACE,EAAQ,EAAE,MAAM,OACtB,AAAI,MAAO,IAAU,SACnB,EAAM,CAAI,GAAK,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,IACzC,AAAI,EAAM,OAAS,EACxB,EAAS,EAAM,OAAO,GAAI,OAAM,EAAQ,EAAM,QAAQ,KAAK,IAE3D,EAAS,EAAM,QAEjB,EAAO,QAAQ,SAAA,GACb,EACI,IAAM,GAAI,WAAM,MAAA,wDAEtB,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAAI,OAAM,GAAO,KAAK,IACzB,AAAI,MAAO,IAAS,SACzB,EAAK,CAAI,GAAI,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,KACvC,AAAI,EAAK,OAAS,EACvB,EAAQ,EAAK,OAAO,GAAI,OAAM,EAAQ,EAAK,QAAQ,KAAK,KAExD,EAAQ,EAEV,EAAQ,EAAM,IAAI,SAAC,EAAG,GACpB,MAAI,IAAK,EACA,EAEP,GACI,IAAM,GACN,WAAM,MAAA,qDACC,GAAC,kCAAkC,EAAC,OACxC,EAAE,MAAM,GAAK,EAAO,MAGxB,CAAC,EAAQ,wSCjWlB,AA8DA,GAAA,IAAA,WAAA,cAYE,SAAA,UAAA,aAAA,WACE,MAAQ,MAAK,YACR,WAiBA,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,IAEnB,KASA,GAAA,WAOE,aACE,KAAK,aAAe,GAMf,SAAA,OAAP,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UAMnB,EAAA,SAAP,SAAwC,GACtC,EAAiB,SAAS,aAAa,EAAI,WACvC,CAAC,EAAK,EAAI,aAElB,iBA2BI,GACF,EACI,EAAI,WAAa,KACjB,WAAM,MAAA,gFAEV,EACI,MAAO,GAAI,WAAc,SACzB,WAAM,MAAA,sDACF,MAAO,GAAI,YACnB,EACI,EAAI,UAAU,OAAS,EACvB,WAAM,MAAA,sFAGV,GAAiB,SAAS,gFC/K5B,AAsBA,GAAM,IAAuB,KAChB,GAAuB,eAGhC,EACA,EAAoD,GACtD,MAAI,IAAW,MACb,GAAU,MAEL,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,mBAInE,MAAO,GAAO,QAAQ,mBAAqB,GAAK,GACA,GAGlD,YACI,EAAoB,EACpB,GACF,GAAI,GAAiB,GAOrB,GANA,AAAI,IAAa,IAAW,GAAa,KACvC,GAAiB,IAEnB,AAAI,GAAa,IAAW,GAAa,IACvC,GAAiB,IAEf,GACF,GAAM,GAAQ,EAAO,YAAY,KAC3B,EAAQ,EAAS,YAAY,KAEnC,GAAI,IAAU,EACZ,KAAM,IAAI,OACN,yCAAyC,EAAK,KAC9C,cAAa,IAIrB,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,IACzC,GAAM,GAAc,GAAW,GACzB,EAAgB,GAAW,GACjC,GAAI,CAAC,GAAY,EAAa,GAC5B,KAAM,IAAI,OACN,iCACA,aAAY,EAAW,iBAAiB,EAAa,MAI7D,GAAM,GACF,GAAa,GAAU,EAAS,GAAQ,GACtC,EAAe,GAAa,GAC9B,EACA,GAAQ,GAEZ,GAAI,EAAW,SAAW,EAAa,OACrC,KAAM,IAAI,OACN,yCAAyC,EAAW,OAAM,OAC1D,cAAa,EAAa,OAAM;GAChC,cAAa,EAAU;GACvB,cAAa,EAAY,MAE/B,OAAS,GAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAI,EAAW,GACf,EAAI,EAAa,GAEvB,GAAI,CAAC,EAAU,EAAG,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAC,OAAO,EAAC,cAAc,EAAC,OAAO,EAAC;EACzD,cAAa,EAAU;GACvB,cAAa,EAAY,mBAUC,EAAuB,GACzD,IAAK,KAAK,WAAM,MAAA,GAAK,QAAQ,WAAM,MAAA,mBAGH,EAAoB,GACpD,GAAM,GAAM,MAAO,IAAa,UAAY,MAAO,IAAa,UACxD,MAAO,IAAa,UACxB,CAAC,GACD,EACJ,MAAI,IAAS,IAAW,GAAU,EAAoB,KAClD,GAAS,IAAa,GAAU,EAAsB,IAEjD,GAAsB,EAAQ,EAAK,SAAC,EAAG,GAAM,MAAA,IAAK,IAEpD,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,iBAGlC,EAAW,EAAW,GAIvD,GAHA,AAAI,GAAW,MACb,GAAU,MAER,CAAC,GAAS,EAAG,EAAG,GAClB,KAAM,IAAI,OAAM,8BAA8B,EAAC,kBAAkB,GAIrE,YAAkB,EAAW,EAAW,GACtC,MAAI,CAAC,SAAS,IAAM,CAAC,SAAS,GACrB,GAEL,QAAM,IAAM,MAAM,IAAM,KAAK,IAAI,EAAI,GAAK,eAO5C,EAA6B,EAAa,GAC5C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,GAAI,EAAO,GAAK,GAAO,EAAO,GAAK,EACjC,KAAM,IAAI,OACN,sBAAsB,EAAO,GAAE,SAAS,EAAG,WAAW,eAM5D,EAAqB,GAGvB,OAAO,GAAI,cAAa,IAAS,QAAQ,GAAI,cAAa,4MCzJ5D,GAGM,IAAU,QCHhB,cAiCE,KAAM,IAAI,OAAQ,kBAkBlB,KAAM,IAAI,QAAS,kBAKnB,KAAM,IAAI,+BAAgC,IAC1C,QAAQ,KAAK,sEAIiB,GAC9B,AAAI,KAAM,QAAQ,iCAChB,QAAQ,KACJ,EAAM,6FAYZ,EAAO,iCASP,MAAO,iBAyBP,MAAO,GAAO,qBAgCQ,GAEtB,MAAO,GAAO,QAAQ,eA4CpB,EAA6B,GAC/B,MAAO,GAAO,KAAK,EAAU,eAcP,GACtB,GAAM,GAAU,GAAsB,GACtC,EAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,wBAmCI,GACrC,MAAO,GAAO,KAAK,eA4BA,GACnB,MAAO,GAAO,KAAK,eAkBM,GACzB,MAAO,GAAO,WAAW,iBAWzB,MAAO,GAAO,sBAUd,MAAO,GAAO,wBAQc,GAC5B,EAAO,cAAc,eAOK,GAC1B,MAAO,GAAO,YAAY,eAQO,GAEjC,MAAO,GAAO,mBAAmB,eAmB/B,EAAc,EACd,GAAA,MAAA,KAAA,QAAA,GAAA,GACK,EAAO,gBAAgB,EAAM,EAAS,iBAY7C,MAAO,GAAO,oBASY,EAAsB,GAChD,KAAM,YAAY,EAAc,GC5XlC,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCnE3B,AAqDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCvErC,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OAGjC,GAFA,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAED,EAAG,QAAU,SAAW,EAAG,QAAU,QACvC,MAAO,IAAS,EAAI,GAGtB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,WAAW,EAAI,GACnC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAC/B,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KC3E3B,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAEA,GAAM,EAAG,CAAC,KAAI,KCpE3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GAEpC,MADA,GAAK,CAAC,IACF,EAAG,QAAU,YACR,EAAQ,WAAW,GAGrB,EAAQ,IAAI,IAClB,EAAgC,KAAiB,OAGzC,IAAM,EAAG,CAAC,KAAI,KCrD3B,AAoCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAuCA,YAAiC,GAC/B,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,+DACV,EACI,EAAQ,QAAU,EAClB,WAAM,MAAA,uDACF,IAAG,EAAQ,UAEnB,GAAM,GACF,EAAQ,IAAI,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAG,UAAU,EAAK,UAEtD,EAAc,EAAS,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAI,EAAE,QAAU,EAAY,MAC1B,KAAM,IAAI,OACN,8DAIR,EAAS,QAAQ,SAAA,GACf,GAAI,CAAC,GAAiB,EAAE,MAAO,EAAY,OACzC,KAAM,IAAI,OACN,8DAIR,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,GACE,GAGH,EAAqB,EAE3B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC/E7B,YAuBqC,EAAgB,GACnD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,GAAI,EAAK,EAAK,OAAS,EAAI,KAAO,EAAO,EAAI,EAC3C,MAAO,GAGX,MAAO,eAIL,EAAqB,EAAqB,GAK1C,OAJI,GAAO,EAAU,OAAS,EAAU,OACpC,EAAM,GACR,EAAS,EACT,EAAY,EACL,EAAM,EAAG,EAAM,EAAM,IAC9B,AAAI,EAAK,QAAQ,KAAS,GACxB,EAAI,KAAK,EAAU,MAEnB,EAAI,KAAK,EAAU,MAGvB,MAAO,eAIL,EAAkB,GAGpB,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,EAAK,QAAQ,KAAS,IACxB,EAAS,KAAK,EAAO,IAGzB,GAAM,GAAc,EAAK,IAAI,SAAA,GAAO,MAAA,GAAO,KAC3C,MAAO,CAAC,EAAU,eAIhB,EAAiB,GACnB,GAAM,GAAiB,EAAK,IAAI,SAAA,GAAK,MAAA,KACrC,MAAO,IAAiB,EAAO,EAAgB,eAI7C,EAAa,EAAgB,GAC/B,EACI,GAAqB,EAAM,GAC3B,WAAM,MAAG,GAAG,2CACR,aAAY,EAAI,aAAa,EAAI,yBAQR,EAAgB,GAEjD,GAAI,GAAqB,EAAM,GAC7B,MAAO,MAGT,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,AAAI,EAAK,QAAQ,KAAO,IACtB,EAAO,KAAK,GAGhB,SAAK,QAAQ,SAAA,GAAQ,MAAA,GAAO,KAAK,KAC1B,cAI8B,GACrC,MAAO,GAAK,IAAI,SAAC,EAAM,GAAM,MAAA,CAAC,EAAG,KAC5B,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAE,KACxB,IAAI,SAAA,GAAK,MAAA,GAAE,iBAGe,EAAiB,GAEhD,OADM,GAAgB,GACb,EAAI,EAAO,EAAS,EAAI,EAAM,EAAE,EACvC,EAAI,KAAK,GAEX,MAAO,GC3GT,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KCxF3B,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAIA,IAAM,EAAG,CAAC,KAAI,KCzF3B,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC7EjC,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,AAAI,GAAQ,MACV,GAAO,GAET,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KChFjC,AAqCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC/C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA2CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAEH,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAQ,EAAG,CAAC,OAAM,KC7D/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,YAoGI,EACA,EAAuC,EACvC,EAA4B,EAC5B,GAD4B,AAAA,IAAA,QAAA,GAAA,QAM9B,GAAM,GAAgB,EAAW,GAC3B,EACE,EAAW,OAAA,CAAE,IACf,EAAc,GAAwB,GAE5C,MAAO,IACH,EAAY,EAAc,EAAS,EAAW,EAC9C,KAAyB,KAAsB,eAIjD,EACA,EAAqC,EACrC,EAAoC,EACpC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,gBACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GAEjB,EACJ,GAAI,IAAe,eACjB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACrD,IAAe,gBACxB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAE9D,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,EAAc,GAC7D,eAOF,EACA,EACA,EACA,EAA4C,EAC5C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,SACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAE9B,EACA,EACJ,GAAI,IAAe,QACjB,EAAc,eACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACxD,IAAe,QACxB,EAAc,gBACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAEjE,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,GAAO,EACtD,eAQF,EACA,EACA,EAAkC,EAClC,EACA,EAAuC,EACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACvC,AAAA,IAAA,QAAA,GAAA,gBACE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GACnC,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WACtB,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAU,EAAA,EAAA,OAElC,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAC9B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GACf,EAAA,GAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAEjB,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,SAIrB,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAW,GAC1C,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAW,EAAU,IAGvC,CACL,UAAS,EACT,WAAU,EACV,SAAQ,EACR,QAAO,EACP,WAAU,EACV,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,aAAY,EACZ,YAAW,EACX,aAAY,EACZ,YAAW,EACX,sBAAqB,EACrB,qBAAoB,EACpB,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,eA2CX,EACA,EACA,EACA,EAA4C,EAC5C,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,gBAEE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GAE5C,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WAC/B,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,OAE3C,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAE3C,EAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAC5B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAe,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGhC,EACF,GAAuB,EAAa,GAClC,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,SAAU,EAAA,EAAA,UAAW,EAAA,EAAA,SAK/B,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAU,EAAW,GACpD,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAU,EAAW,EAAU,IAGjD,CACL,UAAS,EACT,WAAU,EACV,QAAO,EACP,SAAQ,EACR,QAAO,EACP,WAAU,EACV,SAAQ,EACR,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,YAAW,EACX,aAAY,EACZ,YAAW,EACX,YAAW,EACX,aAAY,EACZ,YAAW,EACX,qBAAoB,EACpB,sBAAqB,EACrB,qBAAoB,EACpB,cAAa,EACb,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,GAIf,YACI,EAA2B,EAAmB,EAC9C,EAAkB,GACpB,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAY,GAGtB,YACI,EAA2C,EAC3C,EAAqB,EAAgB,EACrC,GACF,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAa,EAAQ,GACrB,EAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAe,GAChB,GAAa,EAAY,EAAI,GAAW,EAAS,EAAG,GACzD,EACI,GAAW,GACX,WAAM,MAAA,2BAA2B,EAAY,uEAGjD,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAc,EAAY,EAAY,eAI5C,EACA,EAAmB,EAAgB,GAAA,AAAA,IAAA,QAAA,GAAA,GACrC,GAAM,GAAqB,GAAuB,EAAW,GAC7D,MAAO,MAAK,MACP,GAAW,GAAM,GAAS,GAAK,EAAS,GAAsB,GAGrE,YAAyB,GACvB,MAAI,OAAO,IAAU,SACZ,CAAC,EAAO,EAAO,GAEpB,EAAM,SAAW,EACZ,CAAC,EAAM,GAAI,EAAM,GAAI,GAEvB,EAGT,YAA0B,GAExB,MAAO,OAAO,IAAU,SAAW,CAAC,EAAO,EAAO,GAAS,EAc7D,YAAgC,EAAoB,GAClD,MAAI,IAAY,EACP,EAGF,EAAc,GAAa,GAAM,GAAW,GAGrD,YACI,EAA4C,EAC5C,EAAiB,EAAsB,EACvC,EAAsB,EACtB,EACA,GAEF,GAAI,GACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CAAC,IAAK,EAAK,OAAQ,EAAK,KAAM,EAAK,MAAO,EAAK,KAAM,GAC/D,GAAM,GAAW,GACb,CAAC,EAAU,GAAU,EAAc,EAAc,EAAK,GAC1D,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GACF,KAAK,IAAI,EAAI,GAAY,GAAK,EAAe,EAAe,GAC1D,EACF,KAAK,IAAI,EAAI,GAAW,GAAK,EAAc,EAAc,GACvD,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAC9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,gBAClC,IAAQ,QACjB,EAAU,CAAC,IAAK,EAAG,OAAQ,EAAG,KAAM,EAAG,MAAO,EAAG,KAAM,SACvD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,WAC1C,MAAO,IAAQ,UACxB,GAAM,GAAM,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GACzD,EAAS,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC5D,EAAO,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC1D,EAAQ,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC3D,EAAW,IAAQ,GAAK,IAAW,GAAK,IAAS,GAAK,IAAU,EAClE,QACA,WACJ,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,GAC3C,EAAY,GACP,GAAW,EAAe,EAAM,GAAU,EAAe,EAC1D,GACJ,EAAW,GACN,GAAU,EAAc,EAAO,GAAS,EAAc,EAAG,OAE9D,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,UAAS,EAAE,SAAQ,GAGtC,YACI,EAA4B,EAAiB,EAC7C,EAAiB,EAAqB,EACtC,EAAqB,EAAqB,EAC1C,EAAqB,GAMvB,GAAI,GACA,EACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,GAER,GAAM,GAAW,GACb,CAAC,EAAS,EAAU,EAAS,GAAI,EAAa,EAAG,EAAa,EAC9D,GACJ,EAAW,EAAS,GACpB,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAW,KAAK,KAAK,EAAU,GAC/B,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EACD,GAAY,GAAK,EAAe,EAAe,EAC9C,EAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EAAQ,KAAK,MAAM,EAAgB,GACnC,EAAO,EAAgB,EACvB,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAE9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,MAAK,EAAE,KAAI,EAAE,KAAM,gBAC/C,IAAQ,QACjB,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,SAER,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,GACnD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,OAEnD,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,SAAQ,EAAE,UAAS,EAAE,SAAQ,GAQhD,YACI,EAAe,GACjB,GAAI,CAAC,EACH,MAAO,GAET,OAAQ,OACD,QAEH,MAAO,MAAK,MAAM,OACf,OAEH,MAAO,MAAK,KAAK,OACd,QACH,MAAO,MAAK,MAAM,WAElB,KAAM,IAAI,OAAM,wBAAwB,gBAIZ,GAC1B,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,MAAO,KAAS,GAAK,IAAS,GAAK,IAAS,cAI1C,EAA0B,GAC5B,MAAO,IAAkB,IAAY,GAAkB,eAUjB,GAEtC,GAAI,IAAe,OACjB,MAAO,eACF,GAAI,IAAe,OACxB,MAAO,gBAEP,KAAM,IAAI,OAAM,sBAAsB,GC5mB1C,AAoDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,WACxC,EAAY,EAElB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,mDAAmD,EAAI,KAAI,MAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAIJ,MAFA,GAAK,CAAC,IAEF,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UACvC,EAAI,QAGN,EAAQ,QAAQ,EAAK,IAGxB,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAElE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAG,OAEf,EACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KClHnC,AA6EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,WAE5C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAGrB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GAEJ,SAAK,CAAC,IAEC,EAAQ,UAAU,EAAK,IAG1B,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAEjE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAI,OAEhB,EACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KCzJvC,YAmBuC,EAAoB,GACzD,GAAM,GAAO,EAAO,GAAG,OACvB,EAAO,QAAQ,SAAC,EAAO,GACrB,EACI,EAAM,SAAW,EACjB,WACI,MAAA,kBAAkB,EAAI,sBAAsB,EAAC,sBAC7C,6BAA4B,EAAI,SAG1C,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,kBAAkB,EAAI,iCAAiC,GAAO,GAAC,MAEzE,GAAM,GAAa,EAAO,GAC1B,EAAO,QAAQ,SAAC,EAAO,GACrB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EACK,IAAM,GAAU,EAAM,KAAO,EAAW,GACzC,WAAM,MAAA,kBAAkB,EAAI,uBAAuB,EAAC,MAAM,EAAK,KAC3D,0CAAyC,EAAU,MACnD,oCAAmC,EAAC,qBAKlB,EAAoB,GAElD,OADM,GAAc,EAAO,GAAG,QACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAY,IAAS,EAAO,GAAG,GAEjC,MAAO,GClDT,AAqEA,YAAmC,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/D,EAAO,EAAQ,QAAU,EAAG,WAAM,MAAA,uCAElC,GAAI,GAAW,GAAqB,EAAS,UAAW,UAExD,AAAI,EAAS,GAAG,QAAU,aACxB,EAAS,QAAQ,SAAA,GACf,GAAI,EAAO,QAAU,YACnB,KAAM,IAAI,OAAM;uBACD,EAAO,MAAK,QAKjC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAS,GAAG,OAAO,GAChD,EAAW,GAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAC7D,GAAI,GAAc,KAAc,EAC9B,MAAO,IAAO,GAAI,GAIpB,GADA,EAAW,EAAS,OAAO,SAAA,GAAK,MAAA,GAAE,KAAO,IACrC,EAAS,SAAW,EACtB,MAAO,GAAS,GAGlB,GAAM,GAAS,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QACnC,GAAuB,EAAQ,GAE/B,GAAM,GAAM,EAAQ,OAAO,EAAU,GACrC,SAAK,GACE,GAGH,EAAuB,EACvB,EAAoB,CAAC,KAAI,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC/GjC,AAsCA,YAAoC,GAClC,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAAwB,CAAC,EAAG,GAElC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,QAAQ,GAC5B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAU,EAAG,CAAC,SAAQ,KCjDnC,AA8DA,YACI,EAAiB,EAAwB,GAC3C,GAAM,GAAK,EAAgB,EAAG,IAAK,SAEnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,kCAGlB,GAAM,GAA+B,SAAC,EAAS,GACvC,GAAA,GAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,UAA6B,EAAI,EAAQ,GACzC,EAAK,CAAC,IACC,EAAQ,MAAM,EAAI,EAAQ,IAG7B,EAAsB,CAAC,EAAG,GAC1B,EAAoB,CAAC,MAAK,EAAE,KAAI,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCrF/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAI,EAAQ,KAAK,GACvB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA8CA,YACI,EAA+B,EAC/B,EAA+B,EAC/B,EAAwB,GAC1B,GAAM,GACF,EAAgB,EAAY,aAAc,iBACxC,EACF,EAAgB,EAAY,aAAc,iBACxC,EAAY,EAAgB,EAAU,WAAY,iBAClD,EAAQ,EAAgB,EAAM,OAAQ,iBACtC,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAE7B,EAAW,GAAO,CAAC,EAAO,GAAK,GAC/B,EAAW,GAAO,EAAU,GAC5B,EAAgB,GAAI,EAAU,GAG9B,EAAY,EAAI,MAAM,GACtB,EAAY,EAAI,MAAM,GAAK,EAC3B,EAA8B,CAAC,EAAW,GAC1C,EAAI,GAAM,EAAK,CAAC,EAAG,GAAI,GACvB,EAAI,GAAM,EAAK,CAAC,EAAG,GAAY,GAC/B,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GACnC,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GAEnC,EACF,GAAI,EAAI,GAAQ,GAAI,GAAK,IACrB,EAAI,EAAI,GAAQ,GAAI,EAAa,MACnC,EAAiB,EAAI,GAAK,GAAO,GAAQ,IAC/C,MAAO,CAAC,EAAM,MAGH,IAAgB,EAAG,CAAC,eAAc,KC/E/C,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAE7C,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,iBAAiB,EAAG,KAAI,2CAC1B,EAAW,SAEnB,EACI,EAAM,SAAW,EAAW,OAC5B,WAAM,MAAA,mBACF,EAAM,OAAM,8CACZ,EAAW,SAEnB,EACI,EAAG,MAAM,GAAK,IAAS,EACvB,WAAM,MAAA,yBACI,EAAG,MAAM,GAAE,2CACjB,+BAA8B,EAAW,KAAK,OAAM,QAAQ,KAEpE,GAAM,GAA0B,SAAA,GAC9B,MAAO,GAAQ,eAAe,EAAI,EAAY,IAG1C,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,MAAK,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,iBC1FX,GACpC,GAAI,GACJ,MAAI,GAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAG,EAAE,OACxB,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACvC,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAErD,EAAM,EAGD,EChCT,AAwDA,YACI,EAAyB,EACzB,EACA,EACA,EACA,GACF,AAAI,GAAmB,MACrB,GAAkB,MAEpB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,AAAI,GAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAG9C,EACI,EAAM,OAAS,EAAU,KACzB,WAAM,MAAA,iFAEV,EACI,GAAW,MAAQ,EAAM,OAAS,EAAQ,KAC1C,WAAM,MAAA,+EAEV,EACI,GAAU,MAAQ,EAAM,OAAS,EAAO,KACxC,WAAM,MAAA,8EAGV,GAAM,GAAgB,GAAM,GAEtB,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,EAAO,EAAW,IAEtB,EAAQ,UACX,EAAK,GAAS,GAAQ,GAAS,GAAY,GAAS,GACpD,GAAS,GAAS,IAGlB,EAA+B,CACnC,EAAG,EACH,MAAO,EACP,OAAQ,EACR,KAAM,EACN,SAAU,GAGN,EAA6B,CAAC,gBAAe,GAE7C,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAO,GAAQ,EAAK,EAAG,OAGzB,YAAkB,GAChB,MAAI,IAAK,KACA,KAEL,EAAE,OAAS,EAEN,EAAQ,EAAG,CAAC,EAAE,OACZ,EAAE,OAAS,EACb,EACE,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACpC,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAEjD,KAGI,IAAY,EAAG,CAAC,WAAU,KCrGvC,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAE/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC9E3C,AA4CA,YACI,EAAsB,GACxB,GAAI,GAAQ,EAAgB,EAAG,cAAe,KACxC,EAAS,EAAM,MAErB,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,CAAE,GAAI,IAAM,EAAI,IAAM,IACxC,KAAM,IAAI,OAAM,2CAA2C,EAAK,MAGlE,GAAI,EAAM,OAAS,EAAM,KACvB,KAAM,IAAI,OAAM,+BAA+B,EAAM,OAAM,iBACvD,EAAM,KAAI,KAGhB,GAAI,EAAM,OAAS,EAAM,MAEvB,OADM,GAAW,EAAM,MAAM,QACtB,EAAS,OAAS,EAAM,QAC7B,EAAS,QAAQ,GAEnB,EAAQ,EAAQ,EAAO,GAKzB,OAFM,GAAa,EAAM,MACnB,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAM,OAAS,EAAG,GAAK,EAAG,IACrC,GAAI,EAAW,KAAO,EAAM,GAC1B,EAAK,GAAK,UACD,EAAM,MAAM,KAAO,EAC5B,KAAM,IAAI,OACN,mBAAmB,EAAM,6BAA6B,EAAK,MAGnE,GAAM,GAAO,EAAK,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,KAAI,OAAO,SAAA,GAAK,MAAA,IAAK,IAEjE,GAAI,EAAK,SAAW,EAClB,MAAO,IAAM,GAGf,GAAM,GAAU,SAAC,GAA2B,MAAA,GAAQ,KAAK,EAAO,IAE1D,EAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,MAAK,EAAE,WAAU,GAElD,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAa,MAGb,IAAc,EAAG,CAAC,aAAY,KC5F3C,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAyCA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,eACnC,EACK,GAAgB,EACjB,WAAM,MAAA,uBAAuB,EAAY,aACrC,+BAA8B,EAAY,QAElD,GAAM,GAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,aAAY,EAAE,aAAY,GAE3D,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAM,EAAQ,KAAK,EAAI,EAAc,GAC3C,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAc,EAAG,CAAC,aAAY,KC7B3C,YAAmB,GACjB,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCYrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCDrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KC5BrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCnCrC,AA8DA,YACI,EAAiB,EACjB,EACA,EACA,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GACjE,EACI,IAAY,EAAQ,MAAM,GAC1B,WAAM,MAAA,oCAAoC,EAAO,gBAC7C,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,GAAO,GACL,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GACxC,EACY,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAS,EAAG,CAAC,QAAO,KC3EjC,YACI,EAAiB,EAA6B,EAC9C,EACA,EAAiC,EACjC,GADA,AAAA,IAAA,QAAA,GAAA,OAAiC,AAAA,IAAA,QAAA,GAAA,GAEnC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,MAG9C,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAQ,GACjD,WAAM,MAAA,yDACF,eAAc,EAAM,kBAAkB,EAAQ,OACtD,EACI,IAAe,MACf,WAAM,MAAA,sCACF,EAAU,0CAElB,GAAM,GAAW,EACb,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,KAC7D,EAAU,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAA4B,CAAC,EAAG,GAChC,EAA8B,CAAC,EAAG,GAElC,EAAmB,OAEnB,EAAM,GACP,EAAuB,EAAuB,EAAS,EACxD,EAAkB,EAAW,GAEjC,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAGxC,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,QAGhD,IAAS,EAAG,CAAC,QAAO,KCpHjC,AAoDA,YACI,EAAmE,EACnE,EAAkB,EAClB,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KAC1D,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,KAG9C,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,GAAM,GAAU,IAAe,OAAS,EAAS,GAAK,EAAS,GACzD,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAY,EAEZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEE,EAAM,EAAQ,eAAe,EAAM,EAAQ,GAEjD,SAAK,CAAC,EAAM,IAEL,GAGH,EAAoC,CAAC,GAAI,EAAM,OAAM,GAErD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,EAAE,WAAY,GAEtD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAqB,GAEzB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxF3D,YACI,EAAiB,EACjB,EACA,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IACH,EAAa,EAAI,EAAS,EAAS,EAAK,OAAQ,MAGzC,IAAkB,EAAG,CAAC,iBAAgB,KCxDnD,AA8DA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAExE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAA+B,EAAS,GACxC,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,QACf,WAAM,MAAA,sCACF,EAAU,4CAElB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,GAC5C,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GAExC,EAAqB,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,GAEzD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,KAGI,IAAS,EAAG,CAAC,QAAO,KC3HjC,AA6CA,YACI,EAGA,EAAO,EAAkB,EACzB,GACF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACvE,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,EAAO,KAGzD,GAAM,GAAU,EAAS,GACnB,EAAW,EAAK,MAAM,GAC5B,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,GAEhD,MAAO,GAAQ,eAAe,EAAM,EAAQ,IAGxC,EAAoC,CAAC,GAAI,GAEzC,EAAkC,CAAC,IAAG,GAEtC,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxE3D,YACI,EAAiB,EACjB,EAGA,EAA0C,GAC5C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IAAoB,EAAa,EAAI,EAAS,EAAS,MAGnD,IAAkB,EAAG,CAAC,iBAAgB,KCtDnD,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AAqDA,YACI,EAAsB,EAAU,EAAmB,GAA7B,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,IAAmB,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAG,MAC9C,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,IAE5B,GAAM,GAAe,GAAiB,EAAG,EAAG,MAAM,GAC9C,EAAQ,EAAQ,OAAO,EAAW,EAAc,EAAW,GAG/D,GAFA,EAAK,CAAC,IAEF,GAAe,MACjB,GAAM,GAAqB,GAAuB,GAClD,EAAQ,GAAU,EAAO,GAE3B,MAAO,IAGP,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,EAAE,UAAS,EAAE,QAAO,GAEpD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KCnFjC,AAiEA,YACI,EAA0B,EAC1B,GAAA,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,gBAE7B,EAAe,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC/D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC9D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAEpE,EACI,EAAc,GAAa,EAC3B,WAAM,MAAA;MACN,EAAW,QAAQ,EAAS;MAC5B,EAAG,QAEP,EACI,EAAa,GAAa,EAC1B,WAAM,MAAA;MACN,EAAU,QAAQ,EAAS;UACvB,EAAG,QAEX,EACK,EAAc,GAAY,KAAe,EAC1C,WAAM,MAAA,8CACF,EAAY,EAAS,WACrB,EAAU,sCAAsC,EAAG,QAE3D,GAAM,GAAiC,SAAA,GACnC,MAAA,GAAQ,aAAa,EAAI,EAAW,IAElC,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAAC,UAAS,EAAE,WAAU,GAEvD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAc,MAGP,IAAe,EAAG,CAAC,cAAa,KCvG7C,AA2EA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,sDACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAElC,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAGlB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,iEACF,mBAAkB,EAAO,mBAAmB,EAAS,OAE7D,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IACE,EAAM,EAAQ,gBAAgB,EAAK,EAAS,GAClD,SAAK,CAAC,EAAK,IACJ,GAGH,EAAsC,CAAC,EAAG,EAAK,OAAQ,GACvD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KC/InD,AA+CA,YAAe,GACb,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAA,GACnC,GAAM,GAAO,EAAQ,EAAI,CAAC,EAAG,OACvB,EAAS,EAAQ,KAAK,GACtB,EAAe,EAAE,MAAK,OAAK,EAAE,OAEnC,MAAO,GAAQ,EAAQ,IAGnB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAO,EAAG,CAAC,MAAK,KChE7B,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,GADA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IACzC,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAU,EAAgB,EAAQ,SAAU,cAElD,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA,gEACC,GAAG,KAAI,OAClB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,4DACC,GAAQ,KAAI,OACvB,EACI,IAAe,OACf,WAAM,MAAA,0DACF,0BAAyB,KAEjC,GAAI,GAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACzD,EAAe,IAGjB,GAAM,GAA2B,CAAC,EAAG,EAAK,OAAQ,GAC5C,EAAyB,CAAC,QAAO,EAAE,IAAG,EAAE,UAAS,GAEjD,EAAM,EAAO,UACH,GAAY,EACZ,GAEhB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAa,EAAG,CAAC,YAAW,KCvGzC,YA2BI,EAAmB,GAGrB,OAFM,GAAS,EAAQ,OACjB,EAAiB,GACd,EAAI,EAAG,EAAI,EAAQ,KAC1B,GAAM,GAAM,EAAS,EAAI,EACnB,EAAI,EAAQ,IAAQ,EACpB,EAAI,EAAS,EAAS,OAAS,EAAI,IAAM,EAC/C,AAAI,EAAI,GAAK,IAAM,GACjB,EAAK,QAAQ,GAGjB,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAQ,EAAQ,EAAQ,OAAS,EAAI,GACrC,EAAU,EAAS,OAAS,EAAI,EAChC,EAAS,EAAS,GACxB,AAAI,IAAS,MAAS,IAAU,GAAK,EAAS,IAC5C,EAAO,QAAQ,GAGnB,MAAO,eAIL,EAAkB,GAIpB,OAHM,GAAmB,GACnB,EAAI,KAAK,IAAI,EAAO,OAAQ,EAAO,QAEhC,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GACnC,AAAI,GAAK,MACP,GAAI,GAEN,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GAInC,GAHA,AAAI,GAAK,MACP,GAAI,GAEF,IAAM,EACR,EAAO,QAAQ,WACN,IAAM,EACf,EAAO,QAAQ,WACN,IAAM,GACf,GAAM,GAAS,wDACR,GAAM,QAAQ,EAAM,KAC3B,KAAM,OAAM,OAEZ,GAAO,QAAQ,GAGnB,MAAO,GCrFT,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,MAAM,EAAI,IAE5D,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,OAG/C,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AAoDA,YACI,EAA8B,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAa,EAAgB,EAAW,YAAa,QAAS,QAI9D,EAAiB,GAA2B,EAAG,MAAO,EAAG,OACzD,EAAgB,GAAY,EAAI,GAChC,EAAgB,GAAY,EAAI,GACtC,AAAI,EAAW,OAAS,GAGtB,EACI,EAAW,MAAM,KAAO,EAAG,MAAM,GACjC,WAAM,MAAA,mEAGZ,AAAI,EAAW,OAAS,GAEtB,GACI,EAAW,MAAO,EAAc,MAAO,oBAG7C,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,OAAO,EAAY,EAAe,GACtD,SAAK,CAAC,IACC,GAEH,EAAyB,CAC7B,UAAW,EACX,EAAG,EACH,EAAG,GAEL,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,OAGA,IAAQ,EAAG,CAAC,OAAM,KC5F/B,AAuCA,YAAsC,GACpC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAA0B,CAAC,EAAG,GACpC,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,IAAK,EAClC,KAAiB,OAEjB,IAAY,EAAG,CAAC,WAAU,KC9CvC,AA0DA,YACI,EAAsB,SAEpB,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAAY,GAAI,EAAI,GACpB,EAAQ,GAAU,GAClB,EAAc,GAAM,EAAI,GAC9B,MAAO,IAAM,EAAa,EAAO,MAGtB,IAAW,EAAG,CAAC,UAAS,KCvErC,AA2CA,YAAc,EAAuB,GACnC,GAAM,GAAM,EAAgB,EAAI,KAAM,OAChC,EAAM,EAAgB,EAAI,KAAM,OAEtC,EACK,GAAI,OAAS,GAAK,EAAI,OAAS,IAAO,GAAI,OAAS,GAAK,EAAI,OAAS,GACtE,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GACjD,EAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GAOvD,GALA,EACI,IAAY,EACZ,WAAM,MAAA,gEACC,GAAO,QAAQ,EAAO,OAE7B,EAAI,OAAS,GAAK,EAAI,OAAS,GACjC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,YACZ,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,CAAC,EAAK,eAClB,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAK,GACzB,MAAO,GAAQ,EAAM,CAAC,EAAK,YAE3B,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAK,GACzB,MAAO,OAIE,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,GACtB,SAAK,CAAC,IACC,GAGH,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCtD3B,AAyCA,YAAgC,GAC9B,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,EACI,EAAG,QAAU,SAAW,EAAG,QAAU,UACrC,WAAM,MAAA,8CAEV,AAAI,EAAG,QAAU,SACf,GAAK,GAAK,EAAI,YAGhB,GAAM,GAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KC1D3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAyCA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GAC3D,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,aAAc,GAEjD,EAAY,GAAQ,EAAG,KAAM,WAAM,MAAA,uCACnC,GAAM,GAAW,EAAG,MAAM,QAC1B,MAAI,GAAO,GAET,GACI,CAAE,GAAG,KAAO,IAAM,EAClB,WAAM,MAAA,iCAAiC,CAAG,GAAG,KAAO,GAAE,KAAK,EAAG,KAAI,MACtE,EAAO,EAAG,KAAO,EAAO,GAE1B,EAAS,OAAO,EAAM,EAAG,GAClB,EAAQ,EAAI,MAGR,IAAa,EAAG,CAAC,YAAW,KC1DzC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAqDA,YAAiC,EAAiB,GAChD,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,OAAQ,GAC3C,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OAEzC,GAAM,GAA0B,SAAC,EAAS,GACxC,GAAM,GAAM,EAAQ,KAAK,EAAI,GAC7B,SAAK,CAAC,IACC,GAGH,EAAe,CAAC,GAChB,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,GAE9B,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,EAAkC,MAG3B,IAAO,EAAG,CAAC,MAAK,KC5E7B,AAwCA,YACI,EAAiB,EACjB,EAKA,GAAA,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAc,MAChB,GAAa,GAIf,OAFM,GAAO,GAAO,CAAC,EAAS,GAAa,GACrC,EAAI,GAAW,EAAa,EAAU,EACnC,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAK,IAAI,EAAG,EAAG,GAEjB,GAAM,GAAgB,EAAQ,EAAK,WAAY,CAAC,EAAS,IACzD,GAAI,GAAc,KAChB,MAAO,GAEP,GAAI,EAAW,SAAW,EACxB,MAAO,IAAK,GAAW,EAAK,GAAI,CAAC,EAAW,GAAI,EAAG,IAC9C,GAAI,EAAW,SAAW,EAE/B,MAAO,IACI,GAAW,GAAW,EAAK,GAAI,GAC/B,CAAC,EAAW,GAAI,EAAW,GAAI,EAAG,IACxC,GAAI,EAAW,SAAW,EAE/B,MAAO,IAAK,GAAW,GAAW,GAAW,EAAK,GAAI,GAAI,GAAI,CACrD,EAAW,GAAI,EAAW,GAAI,EAAW,GAAI,EAAG,IAGzD,KAAM,IAAI,OACN,2CAEA,8BAA8B,EAAmB,OAAM,UAKpD,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAqCA,YACI,EAAoB,EAAsB,GAC5C,GAAM,GAAmB,CAAC,MAAK,EAAE,MAAK,EAAE,MAAK,GAE7C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAO,EAAO,IAAQ,GAAI,KAAM,GACxD,GC3CN,AAqCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC7C/B,AAuBO,GAAM,IAAwB,eASI,GACvC,MAAI,IAAU,GACL,EAEF,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KCpCrD,YA8BI,EAAgB,GAClB,GAAI,GAAO,GACP,EASJ,IAPA,AAAI,GAAU,GACZ,GAAM,EACN,EAAO,IAEP,EAAM,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KAG7C,CAAC,GACN,AAAI,EAAM,GAAe,IAAQ,EAC/B,EAAO,GAEP,EAAM,GAAe,EAAQ,EAAM,GAGvC,MAAO,eAIL,EAAkB,EAAc,GAGlC,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,IAAQ,EACV,EAAS,KAAK,EAAO,IAErB,EAAS,KAAK,GAGlB,MAAO,eAUL,EAAW,EAAiB,GAM9B,OALM,GAAU,EAAE,MAAM,GAElB,EAAwB,GAC1B,EAAY,EACZ,EAAY,EACP,EAAI,EAAG,EAAI,EAAM,IACxB,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAM,IAChC,EAAY,KAAK,EAAQ,MAAM,IAGjC,OAAS,GAAI,EAAO,EAAG,EAAI,EAAE,KAAM,IACjC,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,MAAO,CAAC,UAAS,EAAE,UAAS,EAAE,QAAO,EAAE,YAAW,2GC5FpD,AAoDA,YACI,EAAiB,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAW,EAAgB,EAAS,UAAW,SAAU,SAEzD,EAAyB,CAAC,EAAG,EAAI,QAAS,GAC1C,EAAuB,CAAC,KAAI,GAE5B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAa,GAAe,EAAM,EAAG,OAAO,GAC5C,EAAY,GAAyB,EAAI,EAAU,GAEnD,EACF,EAAQ,OAAO,EAAI,EAAQ,EAAU,CAAC,EAAS,OAAQ,GAC3D,SAAK,CAAC,EAAI,IAEH,EAAQ,EAAK,EAAU,cAGhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC5EjC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,QAAQ,EAAI,IAE9D,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC3DnC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,aAAa,EAAI,GACrC,SAAK,CAAC,EAAI,IACH,GAGH,EAA6B,CAAC,EAAG,EAAI,EAAG,GAE9C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAe,EAAG,CAAC,cAAa,KC/D7C,AAsCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCnD7B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GAEnC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,SAAS,IAAK,EACnC,KAAiB,OAEV,IAAW,EAAG,CAAC,UAAS,KC/CrC,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC/C/B,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC9C/B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAEhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAGH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,YAsCI,EAAyC,GAC3C,GAAM,IAAa,IAAU,IAAU,UAAa,MAAM,QAAQ,KAC9D,IAAU,YACZ,KAAM,IAAI,OACN,kFAGN,GAAI,IAAU,UAAY,GAAa,IACnC,CAAE,aAAiB,aACrB,KAAM,IAAI,OACN,6EAGN,GAAM,GAAkB,GAClB,EAA0B,GAChC,MAAO,IAAW,EAAO,EAAO,EAAe,GCrDjD,AA2CA,YAAsC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aACnC,MAAO,IAAQ,EAAI,GAAO,GAAQ,GAAK,MAG5B,IAAY,EAAG,CAAC,WAAU,KChDvC,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,QAC7B,EAAK,EAAgB,EAAG,IAAK,QACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAE3D,EAAqB,CAAC,EAAG,EAAI,EAAG,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC1D7B,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,UAAU,EAAI,GAClC,SAAK,CAAC,EAAI,IACH,GAGH,EAA0B,CAAC,EAAG,EAAI,EAAG,GAE3C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAY,EAAG,CAAC,WAAU,KC/DvC,YAkCyB,EAAe,EAAc,GACpD,GAAI,GAAO,EACT,KAAM,IAAI,OAAM,4CAGlB,GAAM,GAAuB,CAAC,MAAK,EAAE,KAAI,EAAE,IAAG,GAC9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,SAAS,EAAO,EAAM,IAAM,GAC/C,KAAiB,GAAU,GC1CjC,AA4CA,YACI,EAAiB,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IACzD,GAAM,GAAK,EAAgB,EAAG,IAAK,8BACnC,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA;sBACU,EAAG,KAAI,MAC3B,EACI,GAAW,GACX,WAAM,MAAA,+DACF,gCAA+B,EAAW,OAElD,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,6BACd,EAAK,EAAa,EAAM,EAAO,GAEnC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAoB,CAAC,EAAG,GAExB,EAAkB,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEpD,KAIE,IAA6B,EAAG,CAAC,4BAA2B,KCvFzE,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AA0DA,YAAc,GAEZ,SACI,GAAgB,GAAI,WAAM,MAAA,+CACvB,SAAC,EAAsB,GAE5B,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MACxC,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,WAAa,KAC1D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,kFAGN,GAAW,GACJ,EAAM,MAkCnB,YAAe,GAEb,SACI,GAAgB,GAAI,WAAM,MAAA,gDACvB,SAAC,EAAgC,GACtC,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,qFAGV,GAAM,GAAQ,GAAqB,EAAM,OAAQ,WAAY,MACvD,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,YAAc,KAC3D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,MAAA,OAAA,IAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,iGAGN,GAAW,GACJ,KA8Bb,YAA0D,GAKxD,SACI,GAAgB,GAChB,WAAM,MAAA,uDACH,SAAC,EAAM,GACZ,EACI,YAAa,GACb,WAAM,MAAA,wDACV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,6DACJ,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,UAAW,GACJ,CAAC,KAAM,EAAM,GAAS,MAAK,IAmCtC,YAAyC,GAKvC,SACI,GAAgB,GAChB,WAAM,MAAA,wDACH,SAAC,EAAgB,GACtB,EACI,MAAM,QAAQ,IAAS,EAAK,MAAM,SAAA,GAAO,MAAA,aAAe,KACxD,WAAM,MAAA,uEAEV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,iEACV,GAAM,GAAM,EAAO,UAAU,WAAM,MAAA,GAAC,MAAA,OAAI,IAAO,EAAM,GACrD,MAAI,IAAM,MACR,GACI,EAAI,MAAM,MAAO,EAAG,MACpB,yGAGN,GAAW,EAAI,OACR,GAkCX,YAAuB,EAAiB,GAEtC,EACI,GAAgB,GAChB,WAAM,MAAA,wDACV,EACI,GAAW,MACP,MAAM,QAAQ,IAAY,EAAQ,MAAM,SAAA,GAAK,MAAA,aAAa,MAC9D,WACI,MAAA,kFAGR,GAAM,GAAmB,GAAW,KACpC,GAAI,CAAC,GAEH,EAAU,GACV,OAAW,KAAW,GAAO,oBAC3B,EAAQ,KAAK,EAAO,oBAAoB,IAI5C,GAAM,GACF,EAAmB,EAAQ,OAAO,SAAA,GAAY,MAAA,CAAC,EAAS,YAAa,KAGnE,EAAmB,EAAQ,OACjC,EAAU,EAAQ,OAAO,SAAA,GAAY,MAAA,GAAS,YAC9C,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,kEACF,kCAAiC,EAAgB,kBACjD,eAER,GAAM,GAAmB,GACnB,EAAA,EAAA,UAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,EACI,EAAM,KAAK,SAAA,GAAK,MAAA,IAAK,OACrB,WAAM,MAAA,iMAGV,EACI,EAAM,OAAS,EACf,WAAM,MAAA,iEACF,oBAAmB,EAAM,KAAI,aAErC,GAAM,GAA6B,GACnC,SAAQ,QAAQ,SAAC,EAAG,GAClB,AAAI,EAAM,IAAM,MACd,GAAW,EAAE,MAAQ,EAAM,MAG/B,AAAI,GAAyB,MAG3B,EAAsB,QAAQ,SAAA,GAAK,MAAA,GAAW,EAAE,MAAQ,OAEnD,CAAC,MAAK,EAAE,MAAO,GA2CxB,YAAsC,GAEpC,MAAO,GAAO,WAAW,GAG3B,YAAoB,GAClB,GAAM,GAAmB,EAAM,OAAO,SAAA,GAAK,MAAA,IAAK,OAAM,OACtD,GAAI,EAAmB,EACrB,KAAM,IAAI,OACN;sEC7XR,AAuCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAuB,CAAC,EAAG,GACjC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,IAAI,IAAK,EAC5B,KAAiB,OAEV,IAAM,EAAG,CAAC,KAAI,KC/C3B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,SAAS,GAC7B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAW,EAAG,CAAC,UAAS,KChDrC,AAyCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAK7B,EAAW,GAAW,SAAC,GAI3B,GAAM,GAAQ,GAAI,GAAS,GAAI,KAEzB,EAAW,SAAC,GAChB,GAAM,GAAO,EAAI,EAAI,GAAQ,GAAI,KACjC,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAEL,IAAa,EAAG,CAAC,YAAW,KC9DzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAoB,EAAM,EAAG,OAC1C,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAS,OAG1D,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GAAgB,GAClB,EAAI,MAAO,GAAoB,EAAM,EAAG,QAC5C,EAAM,EAAQ,EAAK,GACnB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,iBAAkB,EAAM,SAAQ,GAEzD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCnE3B,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAGhB,GAAM,GAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IACN,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,IAAI,EAAW,GACnC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAEzB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KC/F3B,AAsDA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAAU,EAAgB,EAAQ,SAAU,cAKlD,GAHA,AAAI,IAAS,IACX,GAAO,EAAQ,KAAO,GAEpB,IAAS,EAAQ,KAAO,EAC1B,KAAM,OACF,gEACA,oBAAmB,EAAQ,KAAI,iBAAiB,IAGtD,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACX,EAAO,GAAI,EAAQ,EAAM,IACzB,EAAU,GAAI,EAAQ,GACtB,EACF,GAAI,GAAK,EAAS,WAAY,GAAI,GAAI,GAAI,GAAU,EAAM,KAC9D,SAAK,CAAC,IACC,GAGH,EAA2B,CAAC,OAAQ,GACpC,EAAyB,CAAC,KAAI,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAY,MAGZ,IAAa,EAAG,CAAC,YAAW,KCpFzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAO,GAAI,EAAI,EAAM,IACrB,EAAI,GAAI,EAAI,GACZ,EAAI,GAAI,GACR,EAAI,GAAI,EAAG,GACX,EAAI,GAAI,GACR,EAAM,GAAI,EAAQ,EAAM,EAAE,OAAQ,GAExC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,MAGI,IAAY,EAAG,CAAC,WAAU,KChFvC,AAyCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA2B,CAAC,EAAG,EAAI,EAAG,GAE5C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,WAAW,EAAI,IAClC,EAAgC,KAAiB,OAGjD,IAAa,EAAG,CAAC,YAAW,KCtDzC,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,WAAW,IAAK,EACnC,KAAiB,OAGV,IAAa,EAAG,CAAC,YAAW,KC9CzC,AAwCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,QAC1C,EAAK,EAAgB,EAAG,IAAK,YAAa,QAChD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA0B,CAAC,EAAG,EAAI,EAAG,GAC3C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IACjC,EAAgC,KAAiB,OAEjD,IAAY,EAAG,CAAC,WAAU,KCnDvC,AA0CA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,UAA2B,EAAG,MAAO,EAAG,OAGjC,GAAW,GAAU,EAAG,GAAI,GAAW,GAAW,EAAG,QAGjD,IAAa,EAAG,CAAC,YAAW,KCpDzC,AAwDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAC7B,EAAY,EAEd,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAEA,EAEJ,MAAI,GAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UAC9C,EAAI,EAAI,QAER,EAAI,EAAQ,QAAQ,EAAK,GAG3B,EAAK,CAAC,EAAK,IAEJ,GAGH,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAEhE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KCrHnC,AA2EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GAJiB,AAAA,IAAA,QAAA,GAAA,CAA+C,EAAG,EAAG,IAGtE,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE/B,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAErB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GACE,EAAI,EAAQ,UAAU,EAAK,GACjC,SAAK,CAAC,EAAK,IACJ,GAGH,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAE/D,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KClJvC,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAK,EAAgB,EAAG,IAAK,qBAE7B,EAAkC,CAAC,EAAG,GACtC,EACuB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,oBAAmB,GAErE,EAAS,EAAO,UACH,GAAmB,EACnB,GAEnB,MAAO,CAAC,OAAQ,EAAO,GAAI,QAAS,EAAO,OAGhC,IAAoB,EAAG,CAAC,mBAAkB,KC7EvD,YAsCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAM,EAAO,WACpB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAoB,GAAc,GAAQ,GACzD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC7C1C,YAuCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAK,EAAO,WACnB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAmB,GAAc,GAAQ,GACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC9C1C,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAS,GAA0B,EAAG,MAAO,GAC7C,EAAc,EAAO,GACrB,EAAa,GAAc,GAI3B,EAAW,GAAW,SAAC,GAC3B,GAAM,GAAmB,GAAO,GAE1B,EAAU,EAAiB,QAAU,EAAE,MACzC,EACA,GAAK,EAAG,EAAiB,OACvB,EAAM,GAAI,EAAS,GACnB,EAAQ,GAAI,EAAK,EAAM,GAEvB,EAAW,SAAC,GAChB,GAAM,GAAkB,EAAE,MAAM,QAChC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,GAAI,EAAI,EAAY,GAAK,EAAE,MAAO,YAAa,GAC5D,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAGL,IAAO,EAAG,CAAC,MAAK,KCpC7B,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAGpD,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GACF,GAA+B,EAAI,MAAO,GAC9C,EAAM,EAAQ,EAAG,GACjB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAGhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAEH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,AAuDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAEH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AAmCA,YAAmC,GACjC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAQ,GACR,EAAe,CAAC,GAChB,EAA2B,GACjC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,SAAK,CAAC,IACC,EAAQ,OAAO,IACrB,CAAC,EAAG,GAAK,KAAiB,SAAU,EAAO,EAAc,MAGjD,IAAS,EAAG,CAAC,QAAO,KC9CjC,AA4CA,YACI,EAAsB,EACtB,GADsB,AAAA,IAAA,QAAA,GAAA,MACtB,AAAA,IAAA,QAAA,GAAA,IACF,EAAI,EAAgB,EAAG,IAAK,WAC5B,GAAM,GAAO,GAAe,EAAM,EAAE,OAC9B,EAAQ,GAAK,EAAG,EAAM,GACxB,EAAgB,EAAM,MAC1B,AAAK,GACH,GAAgB,GAAqB,EAAM,MAAO,IAEpD,GAAM,GACF,GAAO,GAAI,GAAK,EAAG,WAAY,EAAQ,EAAO,KAC5C,EAAW,GAAK,EAAY,EAAM,GACxC,MAAO,CAAC,KAAM,EAAO,SAAQ,MAGlB,IAAU,EAAG,CAAC,SAAQ,KChBnC,YACI,EAA2B,EAC3B,EACA,GAOF,OANM,GAAQ,EAAgB,EAAM,OAAQ,gBACtC,EAAK,GAAqB,EAAG,IAAK,gBAClC,EAAK,GAAqB,EAAG,IAAK,gBAEpC,EAAQ,EACN,EAAY,GACT,EAAI,EAAG,EAAI,EAAU,OAAQ,KACpC,GAAM,GAAS,EAAU,GAAG,EAAO,EAAG,GAAI,EAAG,IAC7C,EAAU,KAAK,EAAO,IACtB,EAAU,KAAK,EAAO,IACtB,EAAQ,EAAO,GAIjB,OAFM,GAAmB,GACnB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,GAAK,EACzC,EAAK,KAAK,EAAU,IACpB,EAAK,KAAK,EAAU,EAAI,IAE1B,MAAO,CAAC,EAAM,MAEH,IAAe,EAAG,CAAC,cAAa,KCpE7C,AA4CA,YACI,EAAsC,EAAoB,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,eAC5C,EAAc,EAAQ,KACtB,EAAW,EAAQ,KACzB,GAAI,EAAc,EAChB,KAAM,IAAI,OACN,+DACG,GAAW,MAEpB,GAAI,EAAW,EACb,KAAM,IAAI,OAAM,gDAAgD,GAElE,EAAO,GAAQ,KAAK,SACpB,GAAM,GACF,IAAa,EAAI,EAAQ,EAAS,CAAC,EAAG,KAAO,EAC3C,EAAM,EAAO,cACf,SAAA,GAAW,MAAA,GAAQ,YAAY,EAAU,EAAY,EAAY,IACjE,CAAC,SAAQ,IAEb,MAAO,KAAa,EAAI,EAAQ,EAAK,CAAC,EAAI,OAAqB,KAGpD,IAAc,EAAG,CAAC,aAAY,KCpE3C,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,GAAY,MAAA,GAAQ,SAAS,EAAI,IAEjE,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KC1DrC,AAwCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCrD7B,AA0CA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAI,EAAG,QAAU,aACf,GAAM,GAAI,GAAS,GAAK,IAClB,EAAI,GAAU,GAAK,IACzB,MAAO,IAAQ,EAAG,GAGpB,MAAO,GAAQ,SAAS,IAGpB,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCtBrC,YACI,EAAyB,GAC3B,GAAM,GAAM,EAAgB,EAAI,KAAM,gBAChC,EAAM,EAAgB,EAAI,KAAM,gBAEtC,EACI,EAAI,OAAS,GAAK,EAAI,OAAS,EAC/B,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,EAAQ,EAAK,CAAC,EAAG,KAC9B,MAAO,IAAO,EAAM,MAGT,IAAe,EAAG,CAAC,cAAa,KCtD7C,AAmDA,YACI,EAAiB,EAAmC,GAAA,AAAA,IAAA,QAAA,GAAA,GACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAA0B,SAAC,EAAS,GACxC,SAAK,CAAC,IACC,EAAQ,IAAI,EAAI,EAAU,IAG7B,EAAoB,CAAC,SAAQ,EAAE,cAAa,GAC5C,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAM,EAAG,CAAC,KAAI,KC5C3B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,qDACH,GAAI,EAAG,CAAC,GAAW,MAGf,IAAQ,EAAG,CAAC,OAAM,KCT/B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCV/B,YACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,EACvD,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCX/B,YACI,EACA,EAIA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,GACnD,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCxC/B,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAEnC,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,cAAc,EAAG,KAAI,kCACvB,EAAW,SAEnB,EACI,EAAS,SAAW,EAAW,OAC/B,WAAM,MAAA,qBACF,EAAS,OAAM,kCAAkC,EAAW,SAEpE,EACI,EAAG,MAAM,OACL,SAAC,EAAG,EAAG,GACL,MAAI,GAAI,GAAK,GAAK,EAAW,OACpB,GACD,GAAI,EAAS,EAAI,GAAG,GAAK,EAAS,EAAI,GAAG,IACtC,EAAW,EAAI,KACnB,EAEA,GAET,IACJ,WAAM,MAAA,4BAA4B,EAAG,MAAM,MAAM,GAAE,kBAC/C,EAAS,WAAU,qCACnB,EAAW,aAEnB,GAAM,GAA0B,SAAA,GAC5B,MAAA,GAAQ,eAAe,EAAI,EAAY,IAErC,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,SAAQ,GAExD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,KCrHjD,AAwDA,YACI,EAAqB,EACrB,EAA0B,EAC1B,EAAqC,GACvC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,AAAI,GAAW,MACb,GAAU,GAEZ,AAAI,IAAQ,GACV,GAAM,SAGR,GAAM,GAAK,EAAgB,EAAO,IAAK,WACnC,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,yDACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAC1C,EACF,CAAC,EAAS,eAAgB,EAAS,eAOnC,EACJ,AAAI,IAAQ,OACV,EAAc,GACV,CAAC,EAAS,aAAc,EAAS,aAAc,GAEnD,EAAc,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,IAG7B,GAAM,GAAgB,EAAS,KAAO,GAAK,EAAS,KAAO,EACrD,EAAA,GAAA,CAAA,EAAA,SAAA,EAAA,SAAA,EAAA,GAAC,EAAA,EAAA,GAAiB,EAAA,EAAA,GAElB,EAAe,EAAgB,EAAM,QACrC,EACF,EAAgB,EAAM,GAAe,EAAK,EAAU,GAElD,EAAY,IAAgB,MAC9B,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAChD,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAC9C,EAAI,IAEJ,EAAM,EAAgB,EAAI,GAAe,EAAG,EAAU,GAE5D,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,EAMT,YACI,EAA8B,EAC9B,GACF,GAAM,GAAW,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,EAAa,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KACpC,EAAiB,EAAW,OAAO,EAAU,GAC7C,EAAc,EAAW,IAAI,SAAC,EAAG,GAAM,MAAC,GAAI,EAAe,GAAK,GAAK,IACrE,EAAS,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAY,KAClD,EAAW,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAS,GAAI,EAAO,MACzD,EAAQ,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAG,EAAY,MACvD,MAAO,CAAC,EAAU,GAMpB,YACI,EAA+B,GAGjC,GAAM,GAAqB,EAAY,IAAI,SAAC,EAAG,GAC7C,MAAO,GAAK,GAAI,GAAM,GAAS,GAAK,KAEhC,EAAgB,EAAmB,IAAI,SAAA,GAAK,MAAA,GAAI,IAIhD,EAAgB,EAAc,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAI,KACtD,EAAc,EAAc,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAc,KAClE,MAAO,GAAc,IAAI,SAAC,EAAG,GAC3B,MAAO,CAAC,EAAc,GAAI,EAAY,SAI7B,IAAO,EAAG,CAAC,MAAK,KChK7B,AAsDA,YACI,EAAyB,SACvB,EAAQ,EAAgB,EAAM,OAAQ,OACtC,EAAO,EAAgB,EAAK,MAAO,OACvC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,GAAM,GAAoB,CAAC,EAAG,EAAO,EAAG,GAClC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,EAAO,GAC7B,SAAK,CAAC,EAAO,EAAM,IACZ,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AA0CA,YAAkC,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAS,EAAgB,EAAO,QAAS,SAEzC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAGH,EAAsB,CAAC,EAAG,EAAI,MAAO,GAC3C,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AA8DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,QAE3B,EAA+B,SAAC,GACpC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAEhB,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,KAAK,EAAW,GACpC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAGzB,MAAO,IAGH,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,EAAE,SAAQ,GAExC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAO,EAAG,CAAC,MAAK,KChG7B,AAiCA,YACI,EAAoB,EACpB,GACF,GAAM,GAAO,GAAc,GACvB,EAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,IAEd,MAAO,GAAO,WAAW,EAAQ,EAAO,MAG7B,IAAO,EAAG,CAAC,MAAK,6OC1B7B,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,cAEJ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,AAyCA,GAAA,IAAA,WAUE,WACI,EAAc,EAAsB,EACpC,EAAqB,GACvB,KAAK,KAAO,EACZ,KAAK,OAAS,EACd,KAAK,MAAQ,EACb,KAAK,QAAU,IACf,KAAK,UAAY,EACjB,AAAI,KAAK,WACP,MAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,EACvC,KAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,GAEzC,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,OAAS,GAAgB,EAAU,YAInC,SAAA,UAAA,UAAP,WACE,GAAI,CAAC,MAAM,KAAK,UACd,GAAM,GAAQ,KAAK,QACnB,YAAK,QAAU,IACR,EAKT,OAFI,GAAiB,EACjB,EAAU,GACP,CAAC,IACN,GAAI,GAAE,OAAU,EAAE,OAAU,EAAC,OAC7B,EACE,GAAK,EAAI,KAAK,SAAW,EACzB,EAAK,EAAI,KAAK,SAAW,EACzB,EAAI,EAAK,EAAK,EAAK,QACZ,GAAK,GAAK,IAAM,GAEzB,GAAM,GAAM,KAAK,KAAK,GAAO,KAAK,IAAI,GAAK,GAC3C,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EACzC,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EAEzC,AAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,GAAU,IAId,MAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,MAAK,QAAU,KAAK,aAAa,IAE5B,KAAK,aAAa,IAInB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,OAAS,MAAQ,KAAK,QAAU,UAChC,EAEF,KAAK,MAAM,IAIZ,EAAA,UAAA,iBAAR,SAAyB,GACvB,MAAO,IAAS,KAAK,OAAS,GAAS,KAAK,OAEhD,KAIA,GAAA,WASE,WACI,EAAe,EAAc,EAC7B,GACF,KAAK,MAAQ,EACb,KAAK,KAAO,EAAI,EAChB,KAAK,MAAQ,EAEb,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,MAAQ,GAAgB,EAAU,YACvC,KAAK,MAAQ,GAAI,IAAY,EAAG,EAAG,EAAO,GAAO,KAAK,SAEtD,AAAI,EAAQ,EACV,KAAK,EAAI,EAAS,EAAI,EAEtB,KAAK,EAAI,EAAS,EAAI,EAExB,KAAK,EAAI,EAAI,KAAK,KAAK,EAAI,KAAK,GAI3B,SAAA,UAAA,UAAP,WAEE,OADI,GAAY,EAAY,EAAY,EAAW,EAAW,KAE5D,EACE,GAAI,KAAK,MAAM,YACf,EAAI,EAAK,KAAK,EAAI,QACX,GAAK,GAMd,GALA,GAAK,EAAI,EACT,EAAK,EAAI,EACT,EAAK,EAAK,KAAQ,EAAK,EACvB,EAAM,GAAM,EAAO,KAAK,EAAK,GAAI,EAAI,KAAK,IAAI,IAC9C,EAAI,KAAK,QACL,EAAI,GAAM,KAAK,IAAI,GAAK,EAC1B,MAGJ,SAAK,EAAI,KAAK,KAAQ,KAAK,EAAI,EAC/B,AAAI,KAAK,MAAQ,GACf,IAAK,KAAK,IAAI,KAAK,QAAS,EAAI,KAAK,QAEhC,KAAK,aAAa,IAGnB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,QAAU,UACV,EAEF,KAAK,MAAM,IAEtB,KAEA,GAAA,WAME,WACI,EAAS,EAAS,EAClB,GAFJ,GAAA,GAAA,KAaE,GAZE,AAAA,IAAA,QAAA,GAAA,GAAS,AAAA,IAAA,QAAA,GAAA,GAoBL,KAAA,eAAiB,WACrB,MAAC,GAAK,OAAS,MAAQ,EAAK,QAAU,WAnBxC,KAAK,IAAM,EACX,KAAK,MAAQ,EAAM,EACnB,KAAK,MAAQ,EACb,AAAI,GAAQ,MACV,GAAO,KAAK,UAEd,AAAI,MAAO,IAAS,UAClB,GAAO,EAAK,YAGV,CAAC,KAAK,kBAAoB,KAAK,OAAS,EAC1C,KAAM,IAAI,OACN,0BAA0B,EAAG,MAAM,EAAG,gCAE5C,KAAK,OAAS,GAAgB,GAOxB,SAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,iBACA,EAEF,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,WACE,MAAO,MAAK,aAAa,KAAK,IAAM,KAAK,MAAQ,KAAK,WAE1D,KCxNA,AAwCA,YACI,EAAoB,EAAe,EACnC,EAAsC,GAOxC,GARqC,AAAA,IAAA,QAAA,GAAA,GACnC,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAQ,MACV,GAAO,GAET,AAAI,GAAS,MACX,GAAQ,WAEN,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,OAAM,yBAAyB,GAI3C,OAFM,GAAS,GAAI,IAAU,EAAO,EAAM,EAAO,GAC3C,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAc,EAAG,CAAC,aAAY,KC5D3C,AAuCA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,yBAAyB,GAK3C,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAuB,GAC1D,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAe,EAAG,CAAC,cAAa,KCtD7C,AA4CA,YACI,EAAoB,EAAY,EAAY,EAC5C,GADoB,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,WAI9C,OAFM,GAAM,GAAO,EAAO,GACpB,EAAS,GAAI,IAAc,EAAQ,EAAQ,KAAM,GAC9C,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAgB,EAAG,CAAC,eAAc,KCvD/C,YAwCyB,EAAsB,GAC7C,GAAc,GACd,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,EAC3B,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAkB,KACxB,MAAO,IAAW,EAAQ,EAAO,EAAe,GC/ClD,YA8CI,EAAe,EAAc,EAC7B,GACF,GAF+B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAA,WACE,IAAS,EACX,KAAM,IAAI,OAAM,8BAGlB,GAAM,GAA+B,WACnC,GAAM,GAAgB,IAAU,EAC1B,EAA8B,EAAQ,GAAQ,EAAO,EACrD,EAA8B,EAAO,GAAS,EAAO,EAE3D,GAAI,GAAiB,GACjB,EACF,MAAO,IAAM,CAAC,GAAI,GAGpB,GAAM,GAAc,KAAK,IAAI,KAAK,KAAM,GAAO,GAAS,IAClD,EAAS,GAAoB,EAAa,GAEhD,AAAI,EAAO,GAAS,IAAS,GAG3B,GAAO,IAGT,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,IAGpB,EAAoB,CAAC,MAAK,EAAE,KAAI,EAAE,KAAI,EAAE,MAAK,GAEnD,MAAO,GAAO,cACH,EAAS,GAAiB,KAAiB,GAC3C,GCnFb,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAE7B,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,WAAW,GAC/B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAa,EAAG,CAAC,YAAW,KChDzC,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,KAAK,IAGhB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,MAAM,IAGjB,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC5D/B,AA6DA,YACI,EAAiB,GACnB,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAA+B,SAAC,GACpC,GAAM,GAAO,GAAe,EAAM,EAAG,OACrC,GAAI,EAAG,OAAS,EACd,MAAO,IAAM,GAEf,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,MAAO,GAAQ,EAAK,EAAG,QAGnB,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,KAAM,GAEnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAS,MAGT,IAAU,EAAG,CAAC,SAAQ,KClFnC,AA6BA,YAAoB,GAClB,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCrCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAGV,IAAQ,EAAG,CAAC,OAAM,KChD/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GAGH,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KCe7B,YACI,EAAiB,EACjB,EAAsC,EACtC,EAAqB,EACrB,GADqB,AAAA,IAAA,QAAA,GAAA,CAAqC,EAAG,IAC7D,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EACF,EAAgB,EAAiB,kBAAmB,mBAClD,EACF,EAAgB,EAAiB,kBAAmB,mBAEpD,EAAM,EACN,EAAe,GAMnB,GALA,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGvD,IAAe,OACjB,KAAM,IAAI,OACN,sFAIN,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,MAAM,KAAO,EAC9B,WACI,MAAA,qEACA,wBAAuB,EAAiB,MAAM,GAAE,OACxD,EACI,EAAiB,MAAM,KAAO,EAC9B,WAAM,MAAA,+DACF,8BAA6B,EAAiB,MAAM,GAAE,OAE9D,GAAM,GAAa,EAAiB,MAAM,GACpC,EAAoB,EAAiB,MAAM,GACjD,EACI,EAAiB,MAAM,KAAO,EAAa,EAC3C,WACI,MAAA,qEACA,YAAW,EAAa,EAAiB,MACzC,YAAW,EAAiB,MAAM,GAAE,OAE5C,GAAM,GAAY,GACd,EAAK,EAAkB,EAAS,EAAK,EAAY,GAC/C,EAAkB,EAClB,EACF,GAAO,EAAW,EAAkB,EAAiB,QAAS,GAElE,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KCxFnD,YACI,EAAsB,uHAClB,SAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aAEnC,EACI,EAAG,QAAU,EAAG,MAChB,WAAM,MAAA,kDACF,EAAG,MAAK,YAAY,EAAG,MAAK,OAEpC,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAEtE,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAExD,CAAA,EAAM,EAAG,eAAjB,SAAQ,EAAA,OACA,CAAA,EAAM,EAAG,eAIvB,IAJM,EAAQ,EAAA,OACR,EAAO,GAAI,KAAI,GAEjB,EAAa,EACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,AAAK,EAAK,IAAI,EAAM,KAClB,IAMJ,IAFM,EAAS,GAAI,IAAa,CAAC,GAAa,EAAG,OAC3C,EAAU,GAAI,IAAa,CAAC,GAAa,SACtC,EAAI,EAAG,EAAI,EAAG,EAAI,EAAM,OAAQ,IACvC,AAAK,EAAK,IAAI,EAAM,KAClB,GAAO,OAAO,GAAK,EAAM,GACzB,EAAQ,OAAO,GAAK,EACpB,KAGJ,MAAA,CAAA,EAAO,CAAC,EAAO,WAAY,EAAQ,qBAExB,IAAiB,GCxF9B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC7C7B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AA6BA,YACI,EAAwB,EAAe,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,CAAC,GAAQ,CAAC,OAEhB,IAAU,EAAG,CAAC,SAAQ,KCtCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AAgDA,YAAoC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxD,GAAM,GAAU,EAAgB,EAAQ,SAAU,UAAW,WAK7D,GAHA,AAAI,IAAQ,IACV,GAAM,EAAQ,KAAO,GAEnB,IAAQ,EAAQ,KAAO,EACzB,KAAM,OACF,4DACA,oBAAmB,EAAQ,KAAI,gBAAgB,IAGrD,GAAM,GAAwB,CAAC,OAAQ,GACjC,EAAsB,CAAC,IAAG,GAEhC,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAI,EAAQ,QAAQ,EAAS,GACnC,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAU,EAAG,CAAC,SAAQ,KCzEnC,AAyCA,YAAc,GACZ,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,qDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAoB,CAAC,MAAK,GAEhC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAU,EAAM,KAAK,EAAO,GAC5B,EAAS,EAAQ,IAAI,GAC3B,MAAO,GAAO,QAAQ,EAAM,QAC3B,EAAgC,KAAqB,OAG7C,IAAM,EAAG,CAAC,KAAI,KC5D3B,AA0CA,YAAe,GACb,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,sDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAqB,CAAC,MAAK,GAEjC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAoB,EAAQ,EAAO,CAAC,EAAO,IAC3C,EAAS,EAAQ,KAAK,GAC5B,MAAO,GAAQ,EAAQ,EAAM,QAC5B,EAAgC,KAAqB,OAG7C,IAAO,EAAG,CAAC,MAAK,KC7D7B,AAgDA,YAAgB,GACd,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EACvB,EACJ,GAAI,GAAsB,GACxB,GAAM,GAAe,EAAQ,EAAO,CAAC,EAAO,IAC5C,EAAM,GAAK,QAIX,GAAM,GAAc,CAAC,EAAO,EAAK,GAAqB,IAChD,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IACzC,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IAEzC,EACF,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACjE,EAA0B,EAC5B,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACnE,GAAO,KAEL,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EACF,EAAQ,GAAQ,EAAG,GAAI,CAAC,EAAY,GAAI,EAAY,KACxD,EAAM,GAAK,GAIb,GAFA,EAAM,GAAK,GAEP,EAAM,OAAS,GAAK,EAAM,MAAM,KAAO,GACzC,GAAM,GAAO,EACP,EAAQ,EAAM,MAAM,GAC1B,EAAM,EAAQ,EAAK,CAAC,EAAO,EAAI,MAAM,GAAK,EAAO,EAAI,MAAM,KAC3D,EAAK,UAEP,MAAO,MAGI,IAAQ,EAAG,CAAC,OAAM,iBC3D3B,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAa,GACjB,GAAI,MAAQ,IAAqB,SAC/B,EACI,EAAE,MAAM,GAAQ,IAAoB,EACpC,WAAM,MAAA,kDACV,EACI,GAAI,OAAM,GAAiB,KAAK,EAAE,MAAM,GAAQ,QAEpD,GAAM,GAAY,EAAgB,OAAO,SAAC,EAAO,GAC/C,MAAI,KAAU,IACZ,IAAS,GAEJ,GACN,GACH,EACI,GAAa,EACb,WAAM,MAAA,4DACV,GAAM,GAAW,EAAgB,QAAQ,IAGzC,GAAI,IAAa,IACf,GAAM,GAAQ,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,EAAI,IAC/D,EAAgB,GAAY,EAAE,MAAM,GAAQ,EAE9C,EACI,EAAE,MAAM,KAAU,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IACvD,WAAM,MAAA,gEACV,EAAa,EAGf,MAAO,GC1DT,AA+DA,YACI,EAAsB,EAAkC,GAAA,AAAA,IAAA,QAAA,GAAA,GAC1D,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAG,OAAO,GACvC,EAAa,GAAiB,EAAI,EAAiB,GACzD,MAAO,GAAQ,MAAM,EAAI,EAAY,IAGjC,EAAuB,CAAC,EAAG,GAC3B,EAAoB,CAAC,gBAAe,EAAE,KAAI,GAEhD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCjF/B,AA+CA,YAAe,EAAe,GAC5B,EACI,EAAM,QAAU,UAChB,WAAM,MAAA,mDAAmD,EAAM,QAEnE,GAAI,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACpD,EAAQ,EAAM,KAAO,EAEvB,EACJ,GAAI,GAAa,MAAQ,EAAY,GAEnC,GAAM,GAAQ,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAC7B,EAAO,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAClC,EAAK,EAAM,MAAM,OAAS,GAAK,EAC/B,EAAgB,GAAM,EAAO,EAAO,GACpC,EAAqB,UACZ,GAAa,MAAQ,EAAY,GAE1C,GAAM,GAAa,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KACxC,EAAW,EAAM,MAAM,OAAS,GAAK,EAAY,EACjD,EAAgB,GAAO,CAAC,EAAO,GAAM,IAAc,EAAM,MAAM,OAAS,GACxE,EAAqB,MAErB,GAAgB,EAIlB,GAAM,GAAa,GAAU,GACvB,EACF,EAAQ,GAAQ,EAAe,GAAa,CAAC,EAAO,IAElD,EAAM,GAAI,GAGV,EAAO,KAAK,MAAM,EAAqB,GAAK,EAC5C,EAAa,GAAK,GAClB,EAAa,GAAK,GAClB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GACxB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GAExB,EAAc,EAAc,MAAM,QACxC,SAAY,EAAc,MAAM,OAAS,GAAK,EAEvC,EACH,GAAQ,EAAqB,GAAI,EAAqB,IAAK,MAGpD,IAAO,EAAG,CAAC,MAAK,KClG7B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AAoDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,qBAC7B,EAAK,EAAgB,EAAG,IAAK,qBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,kBAAkB,EAAI,GAC1C,SAAK,CAAC,EAAI,IACH,GAGH,EAAkC,CAAC,EAAG,EAAI,EAAG,GAC7C,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAmB,MAGnB,IAAoB,EAAG,CAAC,mBAAkB,KC1EvD,AAwCA,YAAoC,EAAsB,GACxD,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,MAAO,GAAQ,EAAI,GAAa,EAAG,MAAO,GAAM,aAGrC,IAAU,EAAG,CAAC,SAAQ,KC7CnC,AAyCA,YACI,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAChC,GAAM,GAAW,GAAqB,EAAS,UAAW,SAK1D,GAHA,EACI,EAAS,QAAU,EAAG,WAAM,MAAA,yCAE5B,EAAS,SAAW,EACtB,MAAO,IAAW,EAAS,GAAI,GAGjC,GAAM,GAAO,EAAS,GAAG,KACnB,EAAQ,EAAS,GAAG,MACpB,EAAQ,EAAS,GAAG,MAE1B,EAAY,GAAQ,EAAM,WAAM,MAAA,uCAEhC,EAAS,QAAQ,SAAA,GACf,GACI,EAAO,EAAE,MACT,yDACJ,EACI,IAAU,EAAE,MACZ,WAAM,MAAA,4DAGZ,GAAM,GAAkB,EAAS,IAAI,SAAA,GAAK,MAAA,IAAW,EAAG,KAOxD,MAAO,IAAO,EAAiB,MAGpB,IAAQ,EAAG,CAAC,OAAM,KC7E/B,AAwCA,YAAiC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAEhB,IAAO,EAAG,CAAC,MAAK,KClD7B,AA+DA,YACI,EAAsB,EAAiB,EAAe,EACtD,EAAe,EAAa,EAAkB,EAC9C,GADA,AAAA,IAAA,QAAA,GAAA,GAAe,AAAA,IAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAAkB,AAAA,IAAA,QAAA,GAAA,GAC9C,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAE3B,EAA+B,SAAC,GACpC,AAAI,GAAW,MACb,GAAU,GAAI,OAAM,EAAM,SAG5B,GAAM,GAAe,GAAW,GAChC,GAAI,EAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAI,IAAiB,GAAK,IAAgB,EACxC,KAAM,IAAI,OACN,iEAGN,GAAI,IAAiB,GAAK,IAAmB,EAC3C,KAAM,IAAI,OACN,oEAGN,GAAM,GAAsB,EAAG,KAAO,EAAM,OAGtC,EAAa,GAAW,GACxB,EAAW,EAAG,MAAM,QAC1B,EAAW,QAAQ,SAAA,GACjB,EAAM,GAAQ,EACd,EAAI,GAAQ,EACZ,EAAS,OAAO,EAAM,EAAG,KAE3B,EAAK,EAAQ,EAAI,GAEX,GAAA,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GACJ,EAAA,EAAA,MACA,EAAA,EAAA,IACA,EAAA,EAAA,QAKF,EAAQ,EACR,EAAM,EACN,EAAU,EAEV,GAAM,GAAa,GAAW,GAE9B,EAAW,QAAQ,SAAA,GACjB,EAAI,GAAQ,EAAM,GAAQ,EAC1B,EAAQ,GAAQ,IAIlB,GAAM,GAAO,GAAgB,EAAO,EAAK,GAEnC,EAAW,EAAK,OAAO,SAAC,EAAG,GAAS,MAAA,GAAW,QAAQ,KAAU,KAEjE,EAAa,EAAQ,MAAM,SAAA,GAAK,MAAA,KAAM,IAC5C,GAAI,EACF,MAAO,GAAQ,GAAM,EAAI,EAAO,GAAO,GAGzC,GAAM,GAAM,EAAQ,aAAa,EAAI,EAAO,EAAK,GACjD,MAAO,GAAQ,EAAK,IAGhB,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAC/B,MAAK,EACL,IAAG,EACH,QAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,YAAW,EACX,eAAc,GAGhB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAe,EAAG,CAAC,cAAa,KCvJ7C,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,kEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,gFAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,sEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,wEAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GCjElD,YAgDI,EACA,EACA,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,0EAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,SAAQ,GACJ,EACG,GAAW,EAAQ,EAAO,EAAe,GCpElD,AAmDA,YACI,EAAiB,EAAO,GAAP,AAAA,IAAA,QAAA,GAAA,GAAO,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAK,EAAgB,EAAG,IAAK,QACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAU,EAAG,MAAM,EAAG,MAAM,OAAS,GAC3C,GAAI,EAAI,EACN,KAAM,IAAI,OACN,uDAAuD,EAAO,KAC9D,YAAW,IAGjB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,EAAC,EAAE,OAAM,GAE7B,EAAA,EAAA,cAAA,SAAA,GAAA,MAAA,GAAA,KAAA,EAAA,EAAA,IAAA,EAAA,KAAA,GAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAKf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAO,EAAG,CAAC,MAAK,KC3E7B,AA4CA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,qCAKlB,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAsB,GACzD,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAkB,EAAG,CAAC,iBAAgB,KC3DnD,AA6EA,YACI,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAEnB,GAAM,GAAK,EAAgB,EAAG,IAAK,SAAU,MAC7C,EAAO,EAAG,KAAO,EAAG,WAAM,MAAA,yCAE1B,GAAM,GAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAC1B,EAAA,EAAA,UAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAS,EAAG,CAAC,QAAO,KC3FjC,AA8CA,YACI,EAAiB,EAAiC,GACpD,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EACF,EAAgB,EAAY,aAAc,qBAAsB,SACpE,EAAO,GAAM,GAAc,WAAM,MAAA,qCAEjC,GAAM,GAAmC,CAAC,EAAG,EAAI,WAAY,GACvD,EAAiC,CAAC,YAAW,GAE7C,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,mBAAmB,EAAI,EAAa,GACxD,SAAK,CAAC,IACC,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAoB,MAGpB,IAAqB,EAAG,CAAC,oBAAmB,KCnEzD,AA2CA,YAAkB,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACtC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,EACI,GAAQ,CAAC,EAAG,MAAM,QAAU,EAAO,EAAG,MAAM,OAC5C,WACI,MAAA,UAAU,EAAI,gBAAgB,EAAG,MAAM,OAAM,KAAK,EAAG,MAAM,OAAM,MACzE,AAAI,EAAO,GACT,IAAQ,EAAG,MAAM,QAEnB,GAAM,GAAuB,CAAC,MAAO,GAC/B,EAAqB,CAAC,KAAI,GAC1B,EAAiC,SAAC,GACpC,MAAA,GAAQ,QAAQ,EAAI,IACxB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAU,EAAG,CAAC,SAAQ,KC7DnC,YAsCI,EAAyB,EAAkB,EAC3C,GADyB,MAAA,KAAA,QAAA,GAAA,IAEpB,EAAO,aAAa,EAAc,EAAW,EAAM,GCxC5D,YAuB0B,EAAqB,GAE7C,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,AAAI,EAAS,IACX,EAAQ,KAAK,GAOjB,OAHM,GAAW,GAAO,EAAW,SAE7B,EAAM,GAAO,CAAC,EAAQ,OAAQ,EAAU,QAAS,SAC9C,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAM,EAAS,WAAW,EAAQ,IAClC,EAAS,EAAI,EAAU,OAC7B,EAAI,OAAO,IAAI,EAAK,GAEtB,MAAO,GAAI,WCEb,YAA2B,uGACnB,SACF,EAAgB,EAAW,YAAa,aAAc,QAC7C,CAAA,EAAM,EAAW,eAAxB,SAAO,EAAA,OACP,EAAM,GAAU,EAAW,MAAO,GACxC,AAAI,IAAc,GAChB,EAAW,UAEb,CAAA,EAAO,WAGI,IAAa,GCpD1B,AA6CA,YACI,EAA2B,EAC3B,2HAcF,IAbM,EAAU,EAAgB,EAAQ,SAAU,YAC5C,EAAQ,EAAgB,EAAM,OAAQ,WAAY,QAElD,EAAW,GAAe,EAC1B,EAAU,EAAM,KAChB,EAAc,EAAQ,MAE5B,EAAY,EAAU,EAAG,WAAM,MAAA,0BAC/B,GACI,EAAY,MAAM,EAAU,EAAW,GAAU,EAAM,MACvD,qEAEA,EAAc,EACT,EAAI,EAAU,EAAI,EAAW,EAAS,IAC7C,GAAe,EAAY,GAEvB,SACF,EAAY,MAAM,EAAG,GAChB,OAAO,CAAC,GAAc,EAAY,MAAM,EAAW,IACtD,EAAiB,EAAQ,EAAS,GAClC,EAAe,EAAQ,EAAO,CAAC,KACX,CAAA,EAAM,GAAW,WAArC,SAAoB,EAAA,OACpB,EAAU,GAAQ,EAAmB,CAAC,IAEtC,EAAM,GAAO,EAAgB,EAAS,GAG5C,AAAI,IAAW,GACb,EAAQ,UAEV,AAAI,IAAS,GACX,EAAM,UAER,EAAQ,UACR,EAAe,UACf,EAAa,UACb,EAAkB,UAElB,CAAA,EAAO,WAGI,IAAmB,GCzFhC,AAuCA,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAK,EAAgB,EAAG,IAAK,kBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,6BAC/B,GAAS,EAAI,GAYtB,YAAuC,EAAiB,GACtD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAK,EAAgB,EAAG,IAAK,cACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,yBAC/B,GAAK,EAAI,GAGlB,YAAwC,EAAiB,GACvD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,eAC7B,EAAK,EAAgB,EAAG,IAAK,eACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,0BAC/B,GAAM,EAAI,GAGnB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAK,EAAgB,EAAG,IAAK,mBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,8BAC/B,GAAU,EAAI,GAGvB,YAA0C,EAAiB,GACzD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,4BAC/B,GAAQ,EAAI,GAGrB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EAAK,EAAgB,EAAG,IAAK,sBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,iCAC/B,GAAa,EAAI,MAGb,IAAc,EAAG,CAAC,aAAY,KAC9B,GAAqB,EAAG,CAAC,oBAAmB,KAC5C,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAkB,EAAG,CAAC,iBAAgB,KACtC,GAAa,EAAG,CAAC,YAAW,KAC5B,GAAiB,EAAG,CAAC,gBAAe,KCpHjD,AA2CA,YAAsC,EAAiB,GACrD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAS,GAC7C,UACI,6EAGJ,GAAuB,EAAK,MAAO,EAAI,MAAO,wBACvC,GAAI,EAAM,GAanB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,6BACpC,EAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,2BACpC,GAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAWjB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAWrB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAarB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,2BAC7B,EAAK,EAAgB,EAAG,IAAK,2BACnC,UACI,EAAG,MAAO,EAAG,MAAO,sCACjB,GAAkB,EAAI,MAGlB,IAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAA0B,EAAG,CAAC,yBAAwB,KACtD,GAAY,EAAG,CAAC,WAAU,KC5NvC,AAwEA,YACI,EAAsB,EACtB,EAA8B,GADR,AAAA,IAAA,QAAA,GAAA,aACtB,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IAChC,EAAI,EAAgB,EAAG,IAAK,QAE5B,GAAM,GAAO,GAAS,EAAG,EAAK,GAC1B,EAAgB,EAAK,MACzB,GAAI,GACF,GAAM,GAAO,GAAe,EAAM,EAAE,OACpC,EAAgB,GAA+B,EAAK,MAAO,GAE7D,MAAO,GAAQ,EAAM,GAGvB,YACI,EAAW,EAAkB,GAC/B,GAD+B,AAAA,IAAA,QAAA,GAAA,MAC3B,EAAE,OAAS,EACb,MAAO,IAAI,GAIb,GAAI,EAAE,OAAS,GAAK,IAAS,KAC3B,MAAO,IAAS,EAAQ,EAAG,CAAC,KAAM,EAAG,GAIvC,GAAI,EAAE,OAAS,GAAK,MAAO,IAAS,UAChC,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,aAAe,IAAM,EAE7B,MAAO,IAAK,GAAI,GAAI,GAAI,GAAI,GAAO,EAAG,UAAW,IAGnD,KAAM,IAAI,OAAM,qCAAqC,GAIvD,GAAI,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,GAAK,GAE7C,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,OAAS,IAAM,YAEvB,MAAO,IAAK,GAAI,GAAO,GAAI,IAG7B,KAAM,IAAI,OAAM,qCAAqC,GAGvD,KAAM,IAAI,OAAM,gCAAgC,MAGrC,IAAO,EAAG,CAAC,MAAK,KC3I7B,AA2DA,YACI,EAAiB,EAAiB,EAClC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAS,EAAgB,EAAO,QAAS,iBAE/C,GAAiB,EAAI,GACrB,EACI,GAAiB,EAAG,MAAO,EAAG,OAAQ,WAAM,MAAA,8BAEhD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAI,EAAK,GAE3B,EAAS,EAAI,GAAI,EAAI,GAAK,GAC9B,GAAI,GACF,EACI,GAAQ,KAAM,WAAM,MAAA,mDACxB,GAAM,GAAQ,EAAgB,EAAM,OAAQ,iBAC5C,EAAS,GAAI,EAAQ,GAAI,EAAK,GAAI,EAAQ,KAE5C,MAAO,IAAI,EAAI,MAGJ,IAAgB,EAAG,CAAC,eAAc,KCnF/C,AA+CA,YACI,EAA4B,EAC5B,GACF,GAAM,GAAW,EAAgB,EAAS,UAAW,YAAa,SAC5D,EAAW,EAAgB,EAAS,UAAW,aACrD,GAA8B,EAAU,EAAU,GAElD,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,UAAU,EAAU,EAAU,IAGzC,EAA0B,CAAC,QAAS,EAAU,QAAS,GACvD,EAAwB,CAAC,MAAK,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAW,MAGX,IAAY,EAAG,CAAC,WAAU,iBClCnC,EAAuB,EAAsB,EAC7C,GACF,GAAI,EAAc,QAAU,QAC1B,KAAM,IAAI,OACN,2DACA,uBAAsB,EAAc,MAAK,MAE/C,GAAI,EAAc,KAAO,EACvB,KAAM,IAAI,OACN,uDACA,mBAAkB,EAAc,MAAK,MAG3C,GAAM,GAAW,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAC7D,EAAU,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAElE,GAAI,EAAY,SAAW,EACzB,KAAM,IAAI,OACN,iDACA,KAAI,EAAY,OAAM,gBAAgB,EAAO,MAGnD,GAAM,GAAY,EAAa,KAC/B,GAAI,CAAE,GAAa,OAAS,GACtB,EAAa,OAAS,GAAK,IAAc,GAC7C,KAAM,IAAI,OACN,oCACG,GAAa,MAAK,sBAAsB,EAAQ,MAGzD,GAAI,EAAa,QAAU,EAAc,MACvC,KAAM,IAAI,OAAM,qDC/DpB,AAkEA,YACI,EAAkC,EAClC,EAA0B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC5B,GAAM,GACF,EAAgB,EAAe,gBAAiB,gBAAiB,SAC/D,EACF,EAAgB,EAAc,eAAgB,iBAC5C,EAAgB,EAClB,EAAc,eAAgB,gBAAiB,EAAc,OAEjE,GACI,EAAgB,EAAe,EAAa,GAEhD,GAAM,GAA8B,CAClC,cAAe,EACf,aAAc,EACd,aAAc,GAGV,EAA4B,CAAC,YAAW,GAE9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,cACf,EAAgB,EAAe,EAAa,IAChD,EAAgC,KAAiB,GACjD,MAGO,IAAgB,EAAG,CAAC,eAAc,KC9F/C,AA8DA,YAAmB,EAAsB,GACvC,GAAM,GAAW,EAAgB,EAAS,UAAW,WAAY,SAC3D,EAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,GACpC,MAAO,GAAQ,SAAS,EAAI,IAGxB,EAAyB,CAAC,OAAQ,EAAI,QAAS,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KAAqB,OAGvD,IAAW,EAAG,CAAC,UAAS,KC5ErC,YA4B8B,EAAW,GACvC,GAAI,GAAc,KAChB,MAAO,GAAE,MAAM,QAEjB,GAAI,GAAiB,EAAE,MAAO,GAC5B,MAAO,GAET,GAAI,EAAE,MAAM,SAAW,EAAW,QAEhC,OADM,GAAyB,GACtB,EAAI,EAAG,EAAI,EAAE,MAAM,OAAQ,IAClC,AAAI,EAAW,IAAM,MAAQ,EAAE,MAAM,IAAM,KACzC,EAAa,KAAK,EAAE,MAAM,IAE1B,EAAa,KAAK,EAAW,IAGjC,MAAO,GAGT,MAAO,GC/CT,AAoDA,YACI,EAAsB,EAAc,EACpC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAUnC,GARA,EACI,EAAG,QAAU,UACb,WAAM,MAAA,8DACF,sBAAqB,EAAG,MAAK,sBACrC,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,qDAAqD,EAAI,MAE/D,IAAS,EACX,MAAO,aAAa,GAAS,EAAG,QAAU,EAG5C,GAAM,GAAc,GAAc,EAAI,GAChC,EAAW,EAAI,EACf,EAAa,GACf,GAAM,GAAI,GAAc,EAAa,EAAG,EAAG,UAAW,GAAO,IAC7D,GAEJ,MAAO,GAAI,EAAI,MAGJ,IAAU,EAAG,CAAC,SAAQ,KC9EnC,YAoBoC,GAElC,MAAO,MAAK,MAAM,KAAK,IAAI,EAAG,KAAK,KAAK,KAAK,IAAI,GAAS,KAAK,IAAI,kBAIjE,EAAsB,EAAW,GAGnC,OAFM,GAAO,EAAI,EAAe,EAC1B,EAAY,GAAI,cAAa,GAC1B,EAAI,EAAG,EAAI,EAAc,EAAE,GAClC,GAAM,GAAU,EAAM,KAAK,GAAK,EAAM,GAAe,EAAO,GAC5D,EAAU,GAAK,EAAI,EAAI,KAAK,IAAI,GAElC,MAAO,IAAS,EAAW,WCjC7B,AAwCA,YACI,EAA2B,EAAuB,GAAA,MAAA,KAAA,QAAA,GAAA,wHAC9C,SAAe,EAAgB,EAAa,cAAe,UAC3D,EAAW,EAAgB,EAAS,UAAW,UAErD,EACI,EAAa,KAAO,EACpB,WAAM,MAAA,+DACF,YAAW,EAAa,QAChC,EACI,EAAa,KAAO,IAAM,EAAS,KACnC,WAAM,MAAA,mFAEC,GAAa,KAAI,qBAAqB,EAAS,QAC1D,GACI,EAAa,MAAM,MAAM,EAAG,EAAa,MAAM,OAAS,GACxD,EAAS,MACT,2FAEE,EAAU,EAAa,MAAM,EAAa,MAAM,OAAS,GAC/D,EACI,EAAI,GAAK,GAAK,EACd,WAAM,MAAA,iEACF,eAAc,EAAO,cAAc,KAEnB,CAAA,EAAM,EAAa,eAArC,SAAkB,EAAA,OACJ,CAAA,EAAM,EAAS,eAOnC,IAPM,EAAc,EAAA,OAId,EAAgB,CAAC,EAAgB,OAAS,EAAS,GAAlD,EAAK,EAAA,GAAE,EAAI,EAAA,GACZ,EAAY,GAAuB,OAAQ,GAExC,EAAI,EAAG,EAAI,EAAO,KAIzB,IAHM,EAAS,EAAI,EACb,EAAO,EAAgB,SAAS,EAAQ,EAAS,GACjD,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAKzC,IAHA,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAErC,EAAU,GAAK,EACN,EAAI,EAAG,EAAI,EAAG,IACrB,GAAI,EAAU,GAAG,QAAU,EAAY,IACrC,EAAU,GAAK,EACf,OAKN,MAAI,KAAgB,GAClB,EAAa,UAEf,AAAI,IAAY,GACd,EAAS,UAIX,CAAA,EAAO,GAAO,EAAW,EAAS,MAAO,iBAG9B,IAAc,GCtG3B,AAiDA,YACI,EAAM,EAAO,EACb,EACA,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE5D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GAC3D,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAY,GACxB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAY,GAAE,OACtD,EACI,IAAa,EAAY,GACzB,WAAM,MAAA,0CAA0C,EAAQ,UACpD,mCAAkC,EAAY,GAAE,QACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EACZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,GAAI,GAClD,EAC0B,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,GAE1E,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAsB,GAG5B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC/G7D,YAgCI,EAAY,EAAW,GACzB,GAAI,GAAc,MAAQ,IAAe,SACvC,MAAO,GAET,GAAI,IAAe,OACjB,MAAO,GAAI,EAAI,GAAK,IAEtB,KAAM,IAAI,OACN,gDAAgD,EAAU,iBAK5D,EAAc,GAChB,GAAI,GAAM,EACJ,EACF,GAAgC,EAAK,MAAO,EAAa,OAC7D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,mBAIvB,EAAW,EACX,GACF,GAAI,IAAe,SACjB,MAAO,GACF,GAAI,IAAe,OACxB,MAAO,IAAK,GACP,GAAI,IAAe,MACxB,MAAO,IAAI,GACN,GAAI,IAAe,QACxB,MAAO,IAAM,GACR,GAAI,IAAe,QACxB,MAAO,IAAM,EAAG,GAElB,KAAM,IAAI,OAAM,4BAA4B,EAAU,KAIjD,GAAM,IAAa,SAAC,EAAuB,GAChD,GAAM,GAAe,EAAgB,EACrC,MAAO,CAAC,GAAgB,IAAe,UC3EzC,AA8FA,YAAmD,MACjD,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAeA,GAFA,EAAa,GAAc,SAEvB,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,6DACC,GAAI,KAAI,OACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,8DACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,6DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,OACf,WAAM,MAAA,sCACF,EAAU,2CAElB,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,GAEnD,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACpB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAGlB,EAAe,GAAqB,EAAI,EAAG,GAEjD,EACI,GAA4B,GAC5B,WAAM,MAAA,oEAEF,uDAAsD,EAAS,OAEvE,GAAM,GACF,GAAoB,EAAI,MAAO,EAAc,EAAS,EAAS,GAC7D,EACF,GAAqB,EAAK,EAAc,EAAQ,MAAO,EAAS,GAC9D,GAAgB,CAAC,EAAM,GAE7B,GAAI,GAAS,MACX,GAAM,IAAU,GAAqB,EAAO,GAC5C,GAAI,KAAK,IAEX,MAAO,KAGH,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,YAAY,CAC9B,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA4B,CAChC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAGpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAS,EAAG,CAAC,aAAY,KC5QtC,AAyBA,YACI,EAAM,EAAO,EACb,GACF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,yBAAyB,EAAK,EAAM,IAE1C,EAAoD,CAAC,EAAG,EAAK,GAAI,GAEvE,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,IAGN,GAAM,IACT,EAAG,CAAC,qCAAoC,KChD5C,AAyBA,YACI,EAAmE,EACnE,EAAkB,GACpB,GAAI,GAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,wBAAwB,EAAM,EAAQ,IAE5C,EAAmD,CAAC,GAAI,GAExD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,IAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IACT,EAAG,CAAC,oCAAmC,KCnD3C,AAyFA,YAA4D,MAC1D,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAaA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,0DACF,iBAAgB,EAAQ,KAAI,OACpC,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,4DACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAClC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,EACI,GAAyC,EAAS,GAClD,WACI,MAAA,oEACA,sBAAqB,EAAO,mBAAmB,EAAS,OAEhE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,0BAAyB,EAAe,gBAAgB,EAAG,OAGrE,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IAEA,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,0BAG/C,GAAM,GAAO,SAAC,EAAc,GAC1B,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAS,OACd,GAAA,GAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAElB,EAAe,GAAqB,EAAI,EAAG,GAE3C,EAAO,GACR,EAAiB,MAAO,EAAc,EAAqB,GAC1D,EAAY,GACd,EAAiB,EAAe,EAAqB,MAAO,GAEhE,GAAI,GAAQ,MACV,GAAM,GAAU,GAAqB,EAAO,GAC5C,MAAO,CAAC,EAAM,EAAW,GAE3B,MAAO,CAAC,EAAM,IAGV,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,qBAAqB,CACvC,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAAqC,CACzC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAEpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAkB,EAAG,CAAC,sBAAqB,KClQxD,AAwDA,YAAwC,SACtC,EAAA,EAAA,EACA,EAAA,EAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAUA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GAAc,EAAG,EAAG,EAAY,GAC7C,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WACI,MAAA,qEACA,iBAAgB,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE9C,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAU,UACrD,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,wCAAwC,EAAW,UAClD,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACnC,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAErC,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAU,EAAM,QAG5D,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACnB,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAAK,GAAA,EAAA,GAAG,GAAA,EAAA,GAId,GACF,GAAqB,EAAQ,EAAI,GAAE,OAAQ,GAAG,GAC9C,GACA,GAgBJ,GAdA,AAAI,CAAC,GAAc,CAAC,EAClB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAM,KACzC,AAAI,CAAC,GAAc,EACxB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,GAAc,EAAK,GAAM,KACzC,AAAI,GAAc,CAAC,EACxB,IAAO,GAAc,EAAK,GAAc,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAO,KAE/C,IAAO,GAAc,EAAK,GAAc,GAAM,IAC9C,GAAO,GAAc,GAAc,EAAK,GAAM,KAG5C,GAAQ,MACV,GAAM,IAAU,GAAqB,GAAO,IAC5C,MAAO,CAAC,GAAM,GAAM,QAEpB,OAAO,CAAC,GAAM,KAIZ,EAA+B,SAAC,GACpC,GAAM,GAAI,EAAQ,iBAAiB,CACjC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA6B,CACjC,EAAG,EACH,EAAG,EACH,KAAM,EACN,uBAAwB,GAEpB,EAA2B,CAAC,WAAU,EAAE,WAAU,EAAE,WAAU,GAIpE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAe,GACxC,GAAM,GAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,IAET,CAAC,MAAO,EAAQ,EAAK,GAAW,SAAU,KAEvD,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAe,EAAe,GAC5C,GAAM,IAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,GAAK,IAEd,CAAC,MAAO,EAAQ,GAAK,GAAW,SAAU,KAGvD,MAAO,GAAiB,EAAK,EAAK,IAI/B,GAAM,IAAS,EAAG,CAAC,aAAY,KClOtC,+DCAA,AAiCA,YAAwB,GACtB,MAAO,IAAa,EAAc,IAAM,KAEnC,GAAM,IAAgB,EAAG,CAAC,eAAc,KCpC/C,AAiCA,YAAqB,GACnB,MAAO,IAAa,EAAc,GAAK,IAGlC,GAAM,IAAa,EAAG,CAAC,YAAW,KCrCzC,AAyCA,YACI,EAAkB,EAAqB,EAAmB,EAC1D,GAD0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,GAGF,OAFI,GAAQ,EACN,EAAmB,GAClB,EAAQ,GAAe,EAAO,MACnC,EAAO,KAAK,GAAM,EAAQ,EAAO,IACjC,GAAS,EAGX,GAAI,EACF,KAAO,EAAQ,EAAO,OACpB,GAAM,GAAU,EAAQ,EAAe,EAAO,KACxC,EAAM,GAAO,CACjB,GAAM,EAAQ,EAAO,EAAc,GAAS,GAAK,CAAC,GAAS,KAE7D,EAAO,KAAK,GACZ,GAAS,EAIb,MAAI,GAAO,SAAW,EACb,GAAS,GAAI,CAAC,EAAG,IAGnB,EAAQ,GAAO,GAAS,CAAC,EAAO,OAAQ,IAE1C,GAAM,IAAQ,EAAG,CAAC,OAAM,KCpE/B,AA4CA,YACI,EAAkB,EAAqB,EACvC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,AAAI,GAAa,MACf,GAAY,GAAoB,IAKlC,OAHM,GAAe,GAAM,EAAQ,EAAa,GAC1C,EAAiB,EAAI,EAAc,EAAS,IAC5C,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAa,MAAM,GAAI,IACzC,EAAO,KACH,GAAK,GAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,EAAG,IAAe,IAE5D,MAAO,IAAO,GAET,GAAM,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAmDA,YACI,EACA,EACA,EACA,EACA,EACA,GAEF,GAAM,GAAS,EAAgB,EAAO,QAAS,iBACzC,EAAS,EAAgB,EAAO,QAAS,gBAAiB,WAC1D,EAAU,EAAgB,EAAQ,SAAU,gBAAiB,SACnE,EAAS,GAAU,WACnB,EAAqB,GAAsB,EAE3C,GAAM,GAAW,EAAO,MAAM,GAE9B,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OACnC,EACI,EAAO,OAAS,GAAK,EAAO,MAAM,KAAO,EACzC,WAAM,MAAA,oDAAoD,EAAQ,OAC9D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAQ,OAAS,GAAK,EAAQ,MAAM,KAAO,EAC3C,WAAM,MAAA,qDAAqD,EAAQ,KAC/D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,iEACF,WAAU,EAAS,OAAM,OACjC,EACI,EAAS,IAAM,GAAK,EAAS,IAAM,EACnC,WAAM,MAAA,2CAA2C,IACrD,EACI,IAAW,YAAc,IAAW,UACpC,WAAM,MAAA,+CAA+C,IAEzD,GAAM,GAAiC,SAAC,GAAY,MAAA,GAAQ,cACxD,EAAQ,EAAQ,EAAS,EAAU,EAAQ,IAEzC,EACoB,CAAC,MAAO,EAAQ,MAAO,EAAQ,OAAQ,GAC3D,EAA4B,CAAC,OAAM,EAAE,mBAAkB,EAAE,SAAQ,GACjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GACJ,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KCtG/C,AAiCA,YAAwB,GACtB,GAAM,GAAS,EAAgB,EAAO,QAAS,gBAAiB,WAEhE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAA8B,CAAC,MAAO,GACtC,EACF,EAAO,UAAU,GAAe,EAAgC,IACpE,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KC/C/C,AA4CA,YACI,EAA4B,EAC5B,EACA,GADA,AAAA,IAAA,QAAA,GAAA,GACA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,mBAAoB,WAEnE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,mDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAAiC,CAAC,MAAO,GACzC,EAA+B,CAAC,QAAO,EAAE,UAAS,EAAE,OAAM,GAC1D,EAAM,EAAO,UACf,GAAkB,EAClB,GACJ,MAAO,GAGF,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC/DrD,AAoBA,YACI,EAAiB,EAAkB,EACnC,EAAsB,EAAwB,GAMhD,AAAI,GAAgB,MAClB,GAAe,IAEjB,AAAI,GAAkB,MACpB,GAAiB,OAAO,mBAE1B,AAAI,GAAgB,MAClB,GAAe,GAGjB,GAAM,GAAW,EAAM,MAAM,GAC7B,SAAgB,KAAK,IAAI,EAAe,GAExC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAClE,EACI,EAAM,OAAS,EACf,WAAM,MAAA,+CAA+C,EAAM,KAAI,MACnE,EACI,EAAM,MAAM,KAAO,EACnB,WACI,MAAA,oDAAoD,EAAM,MAAM,KACxE,EAAY,EAAO,OAAS,EAAG,WAAM,MAAA,+BACrC,EACI,EAAO,MAAM,KAAO,EACpB,WAAM,MAAA,sDAAsD,EAAQ,KAChE,YAAW,EAAO,MAAM,MAChC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAC3D,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GC3DnE,AA0BA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBAC1B,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAExB,GAAM,GAAQ,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,GAC1D,MAAO,GAAO,cACV,SAAA,GAAK,MAAA,GAAE,kBACH,EAAQ,EAAS,EAAe,EAAc,IAClD,CAAC,MAAO,EAAQ,OAAQ,GAAU,KAAiB,GACnD,GAGC,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC/CvD,YA+BI,EAAU,EAAY,GACxB,GAAM,GAAQ,GAAa,EAAK,EAAS,GACnC,EAAiB,EAAQ,EAAI,CAAE,GAAQ,GAAK,EAClD,EAAI,OAAO,EAAgB,EAAG,eAmB5B,EAAU,EAAW,GACvB,MAAO,IAAc,EAAK,EAAQ,GAAc,IAUlD,YAA8B,EAAM,GAClC,MAAO,GAAI,EAAI,EAAI,EAAI,EAAI,GAAK,EAGlC,YACI,EAAU,EAAW,GAKvB,OAJI,GAAO,EACP,EAAQ,EAAI,OACZ,EAAS,EACT,EAAQ,GACL,EAAO,IACZ,EAAS,EAAS,GAAQ,IAAU,GACpC,GAAM,GAAgB,EAAW,EAAQ,EAAI,IAC7C,AAAI,EAAgB,EAClB,EAAO,EAAS,EAEhB,GAAQ,EAGR,EAAQ,CAAC,GAIb,MAAO,GAAQ,EAAO,CAAC,EAAO,ECvFhC,YAoCI,EAAmB,EAAoB,EACvC,EAAsB,GACxB,MAAO,IACI,EAAO,EAAQ,EAAe,EAAc,EAC5C,GACC,4BAIV,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAC5C,EAAsB,GACtB,EAA6C,gBAK/C,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAAgB,EAC5D,IAGN,YACI,EAAmB,EAAoB,EACvC,EAAsB,EAAwB,EAC9C,EAA4B,EAC5B,GADA,AAAA,IAAA,QAAA,GAAA,IAA4B,AAAA,IAAA,QAAA,GAAA,IAC5B,AAAA,IAAA,QAAA,GAAA,IAKF,OAFM,GAAa,GAEV,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,EAAO,GAAK,GACd,EAAW,KAAK,CAAC,MAAO,EAAO,GAAI,SAAU,EAAG,mBAAoB,IAIxE,EAAW,KAAK,IAShB,OALM,GAAQ,EAAe,EAAK,IAAO,EAAgB,EAEnD,EAA4B,GAC5B,EAA2B,GAE1B,EAAgB,OAAS,GAAiB,EAAW,OAAS,IACnE,GAAM,GAAY,EAAW,MACtB,EAAA,EAAA,MAAsB,EAAA,EAAA,SAAU,EAAA,EAAA,mBAEvC,GAAI,EAAgB,EAClB,MAUF,OADI,GAAkB,GACb,EAAI,EAAgB,OAAS,EAAG,GAAK,EAAoB,EAAE,GAClE,GAAM,GAAM,GAAsB,EAAO,EAAU,EAAgB,IAEnE,GAAI,GAAO,GACT,EAAkB,GAClB,MAMF,GAHA,EAAU,MACN,EAAU,MAAQ,GAAe,EAAc,EAAO,GAEtD,EAAU,OAAS,EACrB,MAWJ,EAAU,mBAAqB,EAAgB,OAE/C,AAAK,GAGH,CAAI,EAAU,QAAU,EACtB,GAAgB,KAAK,GACrB,EAAe,KAAK,EAAU,QACzB,AAAI,EAAU,MAAQ,GAG3B,GAAa,EAAY,EAAW,KAM1C,GAAM,GAAe,EAAgB,OAC/B,EAAa,EAAgB,EAEnC,AAAI,GAAsB,EAAa,GACrC,GAAgB,KAAI,MAApB,EAAwB,GAAI,OAAM,GAAY,KAAK,IACnD,EAAe,KAAI,MAAnB,EAAuB,GAAI,OAAM,GAAY,KAAK,KAGpD,GAAM,GACe,CAAC,gBAAiB,GAAS,EAAiB,UAEjE,MAAI,IACF,GAAO,eAAoB,GAAS,EAAgB,YAGtD,AAAI,GACF,GAAO,aAAkB,GAAO,EAAc,UAGzC,EAGT,YAA+B,EAAmB,EAAW,GAC3D,GAAM,GAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAS,GAAQ,GAAU,GAAQ,GACnC,EAAS,GAAQ,GAAU,GAAQ,GACzC,GAAI,GAAS,GAAK,GAAS,EACzB,MAAO,GAET,GAAM,GAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAmB,EAAkB,GACnE,KAAK,IAAI,EAAmB,EAAkB,GAClD,MAAO,GAAoB,GAAQ,EAAQ,GAO7C,YAAwB,EAAsB,EAAe,GAC3D,GAAM,GAAS,KAAK,IAAI,EAAQ,EAAM,GACtC,MAAO,IAAO,EAAe,EAAS,EAGxC,YAA6B,EAAe,GAK1C,MAAQ,GAAG,MAAQ,EAAG,OAChB,EAAG,QAAU,EAAG,OAAW,EAAG,SAAW,EAAG,SC7MpD,AA2CA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,wHACpB,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAED,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,GACxD,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAyB,GC1EtC,AAyDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,GACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEtB,GAAM,GAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EACF,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GAExD,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,eAAgB,EAAO,IAGtD,GAAM,IAA6B,EAAG,CAAC,4BAA2B,KChCzE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,wGACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEC,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,EACpD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAkC,GCxF/C,AAmDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAEzB,EAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EAAkC,CACtC,cAAe,EACf,aAAc,EACd,eAAgB,EAChB,mBAAkB,GAGd,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,aAAc,EAAO,IAGpD,GAAM,IAA0B,EAAG,CAAC,yBAAwB,KCnCnE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,+GACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAG3B,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBADxC,SACF,EAAA,OADG,EAAS,EAAA,GAAE,EAAU,EAAA,GAMtB,EAAM,GACR,EAAW,EAAY,EAAgB,EAAe,EACtD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAA+B,GChF5C,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,2DACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,6DACC,GAAI,OAEf,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAG9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GACZ,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,eACX,EAAa,EAAW,EAAU,IAGlC,EAA+B,CAAC,OAAQ,GACxC,EAA6B,CAAC,aAAY,EAAE,KAAI,GAEhD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCpFjD,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,yBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,kEACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WACI,MAAA,oEACG,GAAI,OACf,EACI,EAAQ,QAAU,WAAa,EAAQ,QAAU,QACjD,WAAM,MAAA,qDAEV,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAE9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GAEZ,EAAsC,CAAC,OAAQ,GAC/C,EAAoC,CAAC,aAAY,EAAE,KAAI,GAEvD,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,sBACX,EAAa,EAAW,EAAU,IAGlC,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAwB,EAAG,CAAC,uBAAsB,KCxF/D,AAyEA,YACI,EAAiB,EAAkB,GACrC,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAClE,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAElE,GAAM,GAAK,EAAgB,EAAG,IAAK,YAEnC,EACI,EAAG,MAAQ,EACX,WAAM,MAAA,4CAA4C,EAAG,KAAI,MAE7D,GAAM,GAAQ,EAAG,MACX,EAAA,EAAA,MAAA,MAAA,IAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,kDAAiD,EAAC,OAExD,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,qDAAoD,EAAC,OAG3D,AAAI,EAAW,GACb,GAAW,GAEb,AAAI,EAAW,GACb,GAAW,GAGb,GAAM,GAAI,EAAQ,GAAM,EAAG,EAAG,EAAG,SAAU,CAAC,GAAI,IAC1C,EAAI,GAAM,EAAG,EAAG,EAAG,SACnB,EAAK,GAAI,EAAG,GAEZ,EAAS,GACX,GAAU,EAAI,GAAO,CAAC,EAAU,UAChC,GAAa,EAAI,GAAO,CAAC,EAAU,WAEjC,EAAO,GAAM,CAAC,EAAG,GAAI,EAAG,OAE9B,MAAO,GACI,GAAM,GAAQ,EAAQ,EAAI,CAAC,GAAI,EAAG,KACvB,IAAI,SAAA,GAAO,MAAA,IAAM,EAAQ,EAAK,MACzC,GAGN,GAAM,IAAW,EAAG,CAAC,UAAS,KC7HrC,AA2DA,YAAsB,GACpB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,EAAkB,GAClB,EACI,GAAM,MAAQ,EAAG,OAAS,EAC1B,WAAM,MAAA,sEAGV,OADM,GAAM,EAAG,GAAG,MAAM,cACf,GACP,EACI,EAAG,GAAG,MAAM,KAAO,EACnB,WACI,MAAA,gEACA,KAAK,EAAkB,GAAG,MAAM,GAAE,QAAQ,EAAG,QAL9C,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,OAQT,GAAkB,GAClB,EAAK,GAAM,EAAI,EAAG,MAAM,GAAI,GAAG,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,CAAC,MAGtD,EACI,EAAG,QAAU,EAAG,GAAG,MAAM,GACzB,WAAM,MAAA,oCACK,EAAkB,OAAM,aAC/B,0BAA0B,EAAkB,GAAG,MAAM,GAAE,QAI/D,OAFM,GAAiB,GACjB,EAAO,aACJ,GACP,EAAG,KAAK,EAAO,KAAK,WAClB,GAAI,GAAI,EAAK,GACb,GAAI,EAAI,EACN,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAO,EAAI,GAAI,EAAI,EAAG,GAAI,IAAK,EAAG,IACxC,EAAI,GAAI,EAAG,GAGf,MAAO,IAAI,EAAG,GAAK,EAAG,kBATjB,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,GAaT,MAAI,GACK,GAAM,EAAI,GAEV,EAIJ,GAAM,IAAc,EAAG,CAAC,aAAY,KC5G3C,AAmFA,YAAa,EAAW,GAMtB,GANsB,AAAA,IAAA,QAAA,GAAA,IACtB,EACI,EAAE,MAAQ,EACV,WAAM,MAAA,gEACF,EAAE,OAEN,EAAE,OAAS,EACb,MAAO,IAAK,EAAe,GAM3B,GAAM,GAAgB,EAAE,MAAM,MAAM,EAAG,EAAE,MAAM,OAAS,GAC7B,OAAO,SAAC,EAAO,GAAS,MAAA,GAAQ,IACrD,EAAO,GACT,EACI,EACA,CACE,EAAe,EAAE,MAAM,EAAE,MAAM,OAAS,GACxC,EAAE,MAAM,EAAE,MAAM,OAAS,KAE/B,GACE,EAAmB,GACnB,EAAmB,GACzB,EAAK,QAAQ,SAAA,GACL,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,KAAK,GACV,EAAK,KAAK,KAEZ,GAAM,GAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OAC9B,EAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OACpC,MAAO,CAAC,EAAG,GAIf,YAAc,EAAa,GAAA,MAAA,KAAA,QAAA,GAAA,IAClB,EAAO,KAAK,WACjB,EACI,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,0CACF,EAAE,MAAM,OAAM,cAYtB,OAVM,GAAI,EAAE,MAAM,GACZ,EAAI,EAAE,MAAM,GAEd,EAAI,GAAI,GACR,EAAI,GAAM,GAER,EAAQ,GAAS,CAAC,CAAC,IAAK,CAAC,EAAG,IAC9B,EAAc,GAAM,GAElB,EAAQ,GAAK,EAAI,EAAI,aAClB,SAGD,EAAQ,EACR,EAAQ,EACR,EAAQ,EACd,EAAA,EAAA,KAAA,wfAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GA8CP,GAAQ,CAAC,EAAO,EAAO,KApDhB,EAAI,EAAG,EAAI,EAAO,EAAE,IAApB,GAuDT,MAAI,CAAC,GAAgB,EAAI,GACvB,GAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,IACzB,EAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,KAGpB,CAAC,EAAG,KAIR,GAAM,IAAK,EAAG,CAAC,IAAG,KCxMzB,AAiBA,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,IAAA,GAAA,MACA,EAAA,EAAA,uBAAA,GAAA,2BAJU,EAAA,WAAA,GAAA,UAAS,KCyBrB,YACI,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GAAM,GAAgB,GAAY,KAAQ,EAAU,EAAI,EAAS,GAEjE,GAAI,IAAc,EAAA,UAAU,KAC1B,MAAO,GAET,GAAI,IAAc,EAAA,UAAU,IAC1B,MAAO,IAAI,GAEb,GAAI,IAAc,EAAA,UAAU,MAC1B,GAAI,GAAY,KACd,MAAO,IAAK,GAEZ,GAAM,GAAkB,EAAQ,KAAO,EAAS,KAC1C,EAAS,GAAI,GAAI,GAAe,GAAI,IAC1C,MAAO,GAAkB,EAAI,GAAI,EAAQ,GAAO,IACnB,EAGjC,GAAI,IAAc,EAAA,UAAU,wBAC1B,GAAI,GAAY,KACd,MAAO,IAAI,GAAI,GAAe,GAAO,EAAQ,OAE7C,GAAM,GAAqB,EAAI,EAAU,GAAK,EAAQ,QAEhD,EACF,GAAK,GAAI,GAAS,EAAoB,GAAO,KAAM,WACvD,MAAO,IAAI,GAAI,GAAe,GAIlC,KAAM,OAAM,sBAAsB,GAE7B,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCnF3D,AA2CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,sBAC5C,EACF,EAAgB,EAAa,cAAe,sBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,uBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,iCAEvC,GAAM,GAAS,GAAI,GAAI,EAAS,IAChC,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAqB,EAAG,CAAC,oBAAmB,KChBzD,YACI,EAAsB,EAA2B,EACjD,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAC5C,EACF,EAAgB,EAAa,cAAe,kBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,mBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,6BAEvC,GAAM,GAAM,GAAO,GACb,EAAS,GAAI,EAAK,GAAI,EAAI,EAAS,GAAe,EAAM,KAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCnBjD,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAU,EAAgB,EAAQ,SAAU,aAC1C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAM,GAAO,GAEnB,EAAU,GAAI,EAAI,GAAO,GAAI,GAAU,GACvC,GAAM,GAAS,GAAK,GAAI,EAAK,EAAI,EAAS,KAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KC9DvC,AAiDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,aAC5C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAc,GAAO,GACrB,EAAQ,GAAI,GAAI,EAAc,IAC9B,EAAY,GAAQ,EAAO,GAC3B,EAAS,GAAI,EAAO,GAEpB,EACF,GAAI,EAAI,GAAO,IAAM,GAAO,IAAa,EAAI,EAAa,IAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KCtEvC,AAgDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,MAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,WAC5C,EAAe,EAAgB,EAAa,cAAe,WAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,YAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,sBAErD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAO,GAEvB,EAAK,GAAI,EAAI,EAAS,GAAI,GAAI,EAAc,MAC5C,EACF,EAAI,GAAI,EAAK,GAAU,GAAI,GAAI,GAAI,EAAK,GAAe,KACrD,EAAS,GAAI,EAAI,GACvB,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAU,EAAG,CAAC,SAAQ,KCrEnC,AA0CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,oBAC5C,EACF,EAAgB,EAAa,cAAe,oBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,qBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,+BAEvC,GAAM,GAAS,GAAkB,EAAS,GAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC3DrD,AAmCA,YACI,EAAsB,GACxB,GAAM,GACF,EAAgB,EAAQ,SAAU,iCAChC,EACF,EAAgB,EAAQ,SAAU,iCACtC,GACI,EAAQ,MAAO,EAAQ,MAAO,4CAsBlC,GAAM,GAAY,GAAK,GACjB,EAAgB,EAAI,EAAS,GAC7B,EAAgB,GAAM,GAAI,GAAI,GAAI,MAExC,MAAO,IAAI,GAAI,EAAW,GAAgB,GAwB5C,YACI,EAAgC,EAChC,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAoB,EACpB,EAAkB,mBAAoB,uBACpC,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KAOvB,GANA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAEjD,GACI,EAAkB,MAAO,EAAQ,MAAO,kCAExC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAO,GAAO,IAEpB,EACI,GAAI,EAAI,EAAmB,GAAI,EAAK,IAChC,EAAI,EAAM,IAEpB,GAAM,GAAS,GAA+B,EAAmB,GAEjE,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxH3D,AA+DA,YACI,EAAW,EAAW,GAKxB,GALwB,AAAA,IAAA,QAAA,GAAA,IACxB,AAAI,IAAQ,IACV,GAAM,EAAO,KAAO,GAGlB,IAAQ,EAAO,KAAO,EACxB,KAAM,OACF,+DACA,wCAAuC,EAAO,KAAI,KAClD,gBAAe,IAGrB,GAAM,GACF,GAAW,SAAC,EAAgB,EAAgB,GAI1C,GAAM,GAAW,GACX,EAAM,GAAU,EAAQ,CAAC,GAAM,GAC/B,EAAY,GAAI,GAAK,EAAQ,WAAY,GAC/C,EAAK,CAAC,EAAQ,IAEd,GAAM,GAAa,GAAI,EAAI,EAAW,IAChC,EAAW,GAAI,EAAY,CAAC,IAE5B,EAAW,SAAC,EAAO,GAChB,GAAA,GAAA,EAAA,GAAQ,EAAA,EAAA,GACT,EAAU,GAAqB,EAAG,MAAO,CAAC,IAChD,MAAO,CACL,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAK,EAAQ,WAAY,GAAI,KACrC,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAI,GAAY,GAAK,EAAQ,eAGzC,MAAO,CAAC,MAAK,EAAE,SAAQ,KAG7B,MAAO,GAAS,EAAQ,GAsB1B,YACI,EAA4B,EAC5B,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GACA,EAAgB,EAAc,eAAgB,uBAC5C,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KASvB,GAPA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GACI,EAAc,MAAO,EAAQ,MAAO,kCAEpC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAa,GAAO,EAAc,MAAM,IAE9C,EACI,GAAI,EAAI,EAAe,GAAI,EAAK,IAC5B,GAAI,EAAsB,IAGpC,GAAM,GAAS,GAA+B,EAAe,GAE7D,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KC3J3D,GAuNM,IAAW,CACf,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,MAAK,IASD,GAAS,CACb,cAAa,GACb,WAAU,GACV,MAAK,GACL,KAAI,IAeA,GAAQ,CACZ,cAAa,GACb,sBAAqB,GACrB,eAAc,GACd,iBAAgB,GAChB,cAAa,GACb,kBAAiB,GACjB,uBAAsB,GACtB,2BAA0B,GAC1B,gCAA+B,GAC/B,wBAAuB,GACvB,6BAA4B,IAOxB,GAAS,CACb,SAAQ,GACR,YAAW,GACX,GAAE,IAaE,GAAS,CACb,mBAAkB,GAClB,oBAAmB,GACnB,eAAc,GACd,UAAS,GACT,UAAS,GACT,QAAO,GACP,iBAAgB,GAChB,oBAAmB,GACnB,oBAAmB,IClSrB,mBAyCwC,GAAA,EAAA,GAAxC,4DAiBE,SAAA,UAAA,SAAA,SAAS,EAAiB,EAAoB,GAApB,AAAA,IAAA,QAAA,GAAA,IAElB,GAAA,GAAA,KAAA,iBAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,GAAI,GAAW,MACb,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,KAAM,OAAQ,EAAM,EAAE,SACrD,KAAK,eAAe,OAEpB,MAAK,eAAe,GAMtB,MAFA,IAAQ,GAEJ,EACK,EAEP,GAAM,UACC,OAOX,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAEd,KAAK,6CAGJ,EAAA,UAAA,oBAAV,WACE,KAAK,YAAc,KAAK,WAAa,GAgBvC,EAAA,UAAA,iBAAA,SAAiB,EAAiB,GAEhC,MAAO,IAAc,EAAG,IAgB1B,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,aAAe,MACtB,GAAQ,KAAK,cAIX,EAAA,UAAA,eAAN,8EACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAErB,CAAA,EAAO,CACL,KAAM,OAEN,OAAQ,GAAO,KAAK,YAAa,gBAI/B,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,gEAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OACN,4DACA,IAAG,KAAK,sBAUE,EAAA,UAAA,kBAAhB,SAAkC,mGAEhC,SAAA,KAAoB,CAAA,EAAM,EAAa,GAAG,OAAO,eAAjD,SAAK,YAAe,EAAA,OAAqC,GACzD,CAAA,EAAO,EAAa,MAAM,UAE9B,GA5HwC,IA8HxC,OAAO,eAAe,GAAW,OAAO,YAAa,CACnD,MAAO,SAAC,GACN,MAAO,GAAS,UAAY,MAAQ,EAAS,kBAAoB,MAC7D,EAAS,gBAAkB,QC1KnC,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAAgC,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,MAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,IAAA,EAChC,EAAA,QAAA,EALN,EAAA,iBAAwC,GACxC,EAAA,mBAA0C,GAOhD,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,cACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,aACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAC3C,EAAoB,EAAK,mBAAmB,GAAG,SAErD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAiB,EAAK,KAC1B,EAAI,GAAO,GAAW,EAAI,EAAK,MAEjC,EACF,EAAI,GAAI,GAAK,GAAI,EAAmB,EAAK,UACjC,GAAK,GAAI,EAAiB,EAAK,WACnC,GAEF,EACF,GAAI,EAAI,EAAmB,EAAK,KAC5B,EAAI,GAAO,GAAU,EAAI,EAAK,MAEtC,EAAgB,OAAO,GACvB,EAAkB,OAAO,GAEzB,GAAM,GAAW,GAAI,EAAI,EAAS,CAAC,EAAK,cAAe,GACvD,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,oBAAsB,MAC7B,IAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,YACzC,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,cAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,iBAAgB,OAAK,KAAK,oBAC/B,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,iBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,IAAO,KAAK,IACZ,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,IAAQ,EAAO,UAjHxD,EAAA,UAAY,WAmHrB,GArHuC,IAsHvC,GAAc,ICrJd,mBA+BsC,GAAA,EAAA,GAMpC,WACc,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,IAD5C,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,aAAA,EAA8B,EAAA,wBAAA,EAHpC,EAAA,iBAAwC,KAOhD,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAM,GAAY,GAClB,EAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,eACrB,SAAU,GACN,WAAM,MAAA,IAAK,EAAM,MAAO,EAAK,yBAClB,SAAS,MAI5B,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAEjD,GAAK,WACH,GAAM,GAAqB,GAAI,EAAiB,GAAO,IACvD,EAAgB,OAAO,GAEvB,GAAM,GAAW,GACb,EAAI,GAAI,EACA,GAAK,GAAI,EAAoB,EAAO,QAAQ,aAChD,CAAC,EAAK,cACV,GACJ,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,kBAAoB,MAC3B,GAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIvC,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,iBAAiB,IAC9D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,iBAAmB,EAAa,IACjC,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,wBAA2B,KAAK,0BAK7B,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,0BA/EzC,EAAA,UAAY,UAiFrB,GAnFsC,IAoFtC,GAAc,ICnHd,mBAkCmC,GAAA,EAAA,GASjC,WACc,EAAgC,EAChC,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MAFvC,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EAL/B,EAAA,uBAA8C,GAC9C,EAAA,wBAA+C,GAMrD,GAAK,WAEH,EAAK,SAAW,GAAO,GAAO,WAC9B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EAAmB,GAAI,EAAG,EAAK,UAErC,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,wBAAwB,IAAM,MACrC,GAAK,wBAAwB,GAAK,CAChC,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAe,EAAK,wBAAwB,GAAG,SAE/C,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QACvD,EACF,GAAI,EAAI,EAAc,EAAK,OACvB,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EAA2B,GAAI,EAAgB,GAC/C,EACF,GAAI,EAAiB,GAEzB,EAAY,OAAO,GACnB,EAAa,OAAO,GAEpB,GAAM,GACF,GAAI,EAAI,GAAI,EACA,GAAI,GAAK,GAA4B,EAAK,UAC9C,CAAC,EAAK,cACV,GACR,EAAM,OAAO,KAGf,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,QAC7C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,SAAS,UAEd,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,yBAA2B,MAClC,GAAQ,KAAK,wBAAwB,IAAI,SAAA,GAAK,MAAA,GAAE,aAI9C,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,yBACrC,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,4GACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACf,GAAK,WACH,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,IACxD,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,MAGpD,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,wBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,UA5IN,EAAA,UAAY,OA8IrB,GAhJmC,IAiJnC,GAAc,ICnLb,mBAiCoC,GAAA,EAAA,GASnC,WACc,EAAgC,EAChC,EAAyB,EACzB,GADyB,AAAA,IAAA,QAAA,GAAA,MACzB,AAAA,IAAA,QAAA,GAAA,GAHd,GAAA,GAIE,EAAA,KAAA,OAAO,KAHK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EACzB,EAAA,MAAA,EANN,EAAA,uBAA8C,GAC9C,EAAA,2BAAkD,GAQxD,GAAK,WACH,EAAK,UAAY,GAAO,GAAG,WAC3B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EACF,GAAI,CAAC,EAAK,aAAc,GAAI,EAAI,EAAK,UAAW,EAAK,OAAQ,IAEjE,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAGxC,AAAI,EAAK,2BAA2B,IAAM,MACxC,GAAK,2BAA2B,GAAK,CACnC,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAIxC,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAkB,EAAK,2BAA2B,GAAG,SAErD,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QAEvD,EAAM,EAAI,EAAiB,EAAK,OAChC,EAAM,GAAI,GAEV,EAAqB,GAAQ,EAAK,GAExC,EAAY,OAAO,GACnB,EAAgB,OAAO,GAEvB,GAAM,GACF,GAAI,EAAI,GAAI,EAAI,GACR,GAAI,EAAgB,GAAI,EAAoB,EAAK,WACrD,GAER,EAAM,OAAO,KAGf,EAAK,UAAU,OAAO,GAAI,EAAK,UAAW,IAC1C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,UAAU,UAEf,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,4BAA8B,MACrC,GAAQ,KAAK,2BAA2B,IAAI,SAAA,GAAK,MAAA,GAAE,aAIjD,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,wDAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OAAM,wDAGlB,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,QAChB,MAAS,KAAK,QAKX,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,QAAY,EAAO,QAvHzB,EAAA,UAAY,SAyHrB,GA3HqC,IA4HrC,GAAc,IC7Jd,mBA6BkC,GAAA,EAAA,GAKhC,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,aAAA,EAEpB,EAAK,gBAAgB,KAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAEF,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAK,WACH,GAAM,GAAW,GAAI,EAAI,EAAK,EAAG,GAAW,GAC5C,EAAM,OAAO,OAGjB,KAAK,uBAMP,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,aAAe,EACpB,AAAI,KAAK,GAAK,MACZ,KAAK,EAAE,UAET,KAAK,EAAI,GAAK,GAAO,CAAC,KAGxB,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,WAGH,EAAA,UAAA,WAAN,qGACU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAA,CAAQ,EAAA,eAGJ,EAAA,UAAA,WAAN,SAAiB,6FACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAC5C,GADA,EAAe,EAAA,OACX,EAAa,SAAW,EAC1B,KAAM,IAAI,OAAM,iEAIpB,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,aAAgB,KAAK,eAIxB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,eA7DjB,EAAA,UAAY,MA+DrB,GAjEkC,IAkElC,GAAc,IC/Fd,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAA8B,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFZ,GAAA,GAGE,EAAA,KAAA,KAAM,IAAa,KAFP,SAAA,aAAA,EAA8B,EAAA,SAAA,EAChC,EAAA,YAAA,EAJJ,EAAA,cAAqC,GAM3C,EAAK,EAAI,GAAO,EAAK,YAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,cAAc,IAAM,MAC3B,GAAM,GAAY,GAClB,EAAK,cAAc,GAAK,CACtB,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,MAInD,GAAM,GAAe,EAAK,cAAc,GAAG,SACrC,EAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAK,WACH,GAAI,GACE,EAAkB,GAAI,EAAI,EAAK,EAAG,GAAe,GACvD,AAAI,EAAK,YACP,EAAW,GACP,EAAI,EAAK,EAAG,GAAI,EAAU,EAAI,EAAiB,EAAK,KAAM,GAE9D,EAAW,GAAI,EAAI,EAAK,EAAG,GAAkB,GAE/C,EAAa,OAAO,GACpB,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,UACP,AAAI,KAAK,eAAiB,MACxB,GAAQ,KAAK,cAAc,IAAI,SAAA,GAAK,MAAA,GAAE,aAS1C,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,SAAW,GAGZ,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,cAAc,IAC3D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,cAAgB,EAAa,IAC9B,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,SAAY,KAAK,SACjB,YAAe,KAAK,cAKjB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,SAAa,EAAO,cA3FlD,EAAA,UAAY,WA6FrB,GA/FuC,IAgGvC,GAAc,IC/Hd,mBAgCsC,GAAA,EAAA,GASpC,WACc,EAAgC,EAChC,EAA0B,EACpC,GAF0C,AAAA,IAAA,QAAA,GAAA,IAChC,AAAA,IAAA,QAAA,GAAA,GAA0B,AAAA,IAAA,QAAA,GAAA,MACpC,AAAA,IAAA,QAAA,GAAA,IAHJ,GAAA,GAIE,EAAA,KAAA,OAAO,KAOP,GAVY,EAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,SAAA,EAA0B,EAAA,QAAA,EANhC,EAAA,uBAA8C,GAC9C,EAAA,mBAA0C,GAC1C,EAAA,qBAA4C,GAQlD,EAAK,SAAW,EAEhB,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,WAE5B,GAAgB,KAClB,KAAM,IAAI,OAAM,+DAIpB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,OACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,qBAAqB,IAAM,MAAQ,EAAK,UAC/C,GAAK,qBAAqB,GAAK,CAC7B,aAAiB,EAAI,MACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAwB,EAAK,uBAAuB,GAAG,SACvD,EAAqB,EAAK,mBAAmB,GAAG,SACtD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEvC,GAAI,EAAK,UACP,GAAM,GAAsB,EAAK,qBAAqB,GAAG,SAEnD,EACF,GAAI,EAAI,EAAqB,EAAK,OAC9B,EAAI,EAAU,EAAI,EAAK,QAEzB,EACF,GAAI,EAAI,EAAU,EAAK,cACnB,GACI,GAAI,EACA,GAAI,GAAO,GAAyB,EAAK,YACnD,EACF,GAAI,EAAI,EAAoB,EAAK,UAAW,GAEhD,EAAsB,OAAO,GAC7B,EAAoB,OAAO,GAC3B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAGb,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EACF,GAAI,EAAI,EAAoB,EAAK,UAC7B,GAAI,EAAI,EAAU,EAAK,cACnB,GAAK,GAAI,EAA0B,EAAK,YAEpD,EAAsB,OAAO,GAC7B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAInB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,sBAAwB,MAAQ,KAAK,UAC5C,GAAQ,KAAK,qBAAqB,IAAI,SAAA,GAAK,MAAA,GAAE,YAE/C,AAAI,KAAK,oBAAsB,MAC7B,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,oBAC7C,AAAI,KAAK,UACP,EAAU,KAAI,MAAd,EAAkB,KAAK,sBAEjB,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EACF,KAAK,SAAW,EAAa,OAAS,EAAI,EAAa,OAAS,EAC9D,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,MAE3C,AAAI,KAAK,UACP,MAAK,qBACD,EAAa,MAAM,EAAgB,EAAG,EAAgB,GACjD,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,iBAK/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,SAAY,KAAK,SACjB,QAAW,KAAK,QAChB,SAAY,KAAK,WAKd,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,SAChD,EAAO,QAAY,EAAO,WA5KzB,EAAA,UAAY,UA8KrB,GAhLsC,IAiLtC,GAAc,ICjNd,AAyBA,GAAA,IAAA,WAAA,cAsCS,SAAA,IAAP,SAAW,GACT,MAAO,IAAI,IAAa,IAkBnB,EAAA,SAAP,SAAgB,EAAsB,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE/C,GAAI,IAAkB,EAAc,EAAU,IAuBhD,EAAA,QAAP,SACI,EAAsB,EAAY,EAAgB,EAClD,GADsB,MAAA,KAAA,QAAA,GAAA,IAAY,AAAA,IAAA,QAAA,GAAA,GAAgB,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,IACK,GAAI,IACP,EAAc,EAAO,EAAU,EAAS,IAevC,EAAA,KAAP,SACI,EAAsB,EAAa,EACnC,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MACnC,AAAA,IAAA,QAAA,GAAA,MACK,GAAI,IAAc,EAAc,EAAO,EAAO,IAehD,EAAA,SAAP,SAAgB,EAAqB,EAAW,GAAhC,MAAA,KAAA,QAAA,GAAA,MAAqB,AAAA,IAAA,QAAA,GAAA,KAAW,AAAA,IAAA,QAAA,GAAA,MAEvC,GAAI,IAAkB,EAAc,EAAK,IAgB3C,EAAA,OAAP,SACI,EAAsB,EAAa,EAAe,EAClD,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MAAe,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,GACK,GAAI,IAAgB,EAAc,EAAO,EAAO,EAAS,IAmB3D,EAAA,QAAP,SAAe,EAAsB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE5B,GAAI,IAAiB,EAAc,IAE9C,KC5LA,GA+Ba,IAAQ,CACnB,IAAK,GAAsB,IAC3B,SAAU,GAAsB,SAChC,SAAU,GAAsB,SAChC,QAAS,GAAsB,QAC/B,QAAS,GAAsB,QAC/B,OAAQ,GAAsB,OAC9B,KAAM,GAAsB,MCtC9B,AAiBA,GAAM,IAA2B,WAC/B,MAAI,OAAO,wBAA0B,YAC5B,sBACE,MAAO,eAAiB,YAC1B,aAEF,SAAC,GAAgB,MAAA,SAa1B,cACE,MAAO,IAAI,SAAc,SAAA,GAAW,MAAA,IAAc,WAAM,MAAA,SCrC1D,YAmBI,EAAiC,EACjC,GACF,GAAM,GACF,EAAc,OAAO,IAAW,SAAW,EAAS,EAAO,IACzD,EACF,EAAe,OAAO,IAAW,SAAW,EAAS,EAAO,IAChE,MAAO,CAAC,EAAS,GCzBnB,YA0BI,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAqB,GACzB,GAAI,EACF,EAAW,EAAS,OAAO,EAAW,MAAM,IAC5C,EAAS,KAAK,EAAW,GAAK,GAC9B,EAAW,EAAS,OAAO,EAAW,MAAM,SAE5C,EAAW,EAAS,OAAO,EAAW,IAEtC,OADM,GAAgB,EAAW,OACxB,EAAI,EAAG,EAAI,EAAe,EAAE,EACnC,EACI,EAAS,OAAO,CAAC,EAAW,EAAI,GAAK,EAAW,GAAI,EAAW,KAErE,EAAW,EAAS,OAAO,EAAW,MAAM,EAAgB,IAE9D,MAAO,eAaL,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAW,GACjB,GAAI,GACF,EAAS,KAAK,GACd,OAAS,GAAI,EAAiB,EAAG,EAAI,EAAc,EAAE,EACnD,AAAI,GAAK,EAAI,EACX,GAAS,KAAK,GACd,EAAS,KAAK,EAAK,GAAiB,KAEpC,EAAS,KAAK,QAMlB,OAFM,GAAsB,GACtB,EAAqB,GAClB,EAAI,EAAG,EAAI,EAAc,EAAE,EAClC,AAAI,GAAK,EAAiB,EAAI,GAAK,EAAI,IAAM,EAC3C,EAAmB,KAAK,GAExB,EAAoB,KAAK,GAG7B,EAAS,KAAI,MAAb,EAAiB,GACjB,EAAS,KAAK,GACd,EAAS,KAAI,MAAb,EAAiB,GAEnB,MAAO,eAaL,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAmB,GAEzB,AAAI,EACF,EAAiB,KAAK,EAAW,GAAK,GAEtC,EAAiB,KAAK,EAAW,GAAK,GAGxC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,GAAK,EAAW,OAClB,AAAI,EACF,EAAiB,KAAK,EAAW,EAAI,GAAK,EAAW,IAErD,EAAiB,KAAK,EAAW,GAAK,EAAW,EAAI,IAGvD,EAAiB,KAAK,EAAW,IAIrC,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,CAAC,GACjB,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAiB,KAAK,EAAM,GAAG,IAEjC,MAAO,eAeL,EAA0B,EAAmB,GAE/C,OADM,GAAY,EAAe,MAAM,EAAG,GACjC,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAU,KAAK,EAAe,EAAI,GAAK,EAAM,GAAG,GAAK,EAAM,GAAG,IAGhE,MAAO,GCvJT,AAiBO,GAAM,IAAkB,mBAClB,GAAa,mBClB1B,AAiBO,GAAM,IAAQ,SACR,GAAS,WACT,GAAS,YACT,GAAS,YACT,GAAS,aACT,GAAS,YCtBtB,qBAmBqB,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACnB,AAAK,KAAM,QAAQ,YACjB,QAAQ,KAAI,MAAZ,QAAgB,wBAIA,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GAClB,AAAK,KAAM,QAAQ,YACjB,QAAQ,IAAG,MAAX,QAAe,GC3BnB,YAiCI,EAAoB,GACtB,GAAI,EAAK,SAAW,EAAK,OACvB,KAAM,IAAI,OACN,gEACG,GAAK,OAAM,WAAW,EAAK,OAAM,MAG1C,OADM,GAAS,GAAI,cAAa,EAAK,OAAS,GACrC,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EACtC,EAAO,GAAK,EAAK,EAAI,GACrB,EAAO,EAAI,GAAK,EAAK,EAAI,GAE3B,MAAO,eAiB8B,GAIrC,OAFM,GAAO,GAAI,cAAa,EAAQ,OAAS,GACzC,EAAO,GAAI,cAAa,EAAQ,OAAS,GACtC,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,EAAI,GAAK,EAAQ,GACtB,EAAK,EAAI,GAAK,EAAQ,EAAI,GAE5B,MAAO,CAAC,KAAI,EAAE,KAAI,eAOiB,GAKnC,OAHM,GAAM,KAAK,KAAK,EAAQ,OAAS,GACjC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAOgB,GAKlC,OAHM,GAAM,KAAK,MAAM,EAAQ,OAAS,GAClC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAShB,EAAuB,GACzB,GAAM,GAAO,EAAQ,EAAQ,GACvB,EAAO,EAAQ,EAAQ,EAAI,GACjC,MAAO,CAAC,KAAI,EAAE,KAAI,eAUhB,EAAkB,EAAc,EAAc,GAChD,EAAK,EAAQ,GAAK,EAClB,EAAK,EAAQ,EAAI,GAAK,cAOpB,EAAW,GAGb,OAFM,GAAO,GAAI,cAAa,EAAI,GAC5B,EAAO,GAAI,cAAa,EAAI,GACzB,EAAI,EAAG,EAAI,KAAK,KAAK,EAAI,GAAI,KACpC,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GAC9C,EAAK,GAAK,KAAK,IAAI,GACnB,EAAK,GAAK,KAAK,IAAI,GAErB,MAAO,CAAC,KAAI,EAAE,KAAI,eAOhB,EAAW,EAAW,GACxB,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GACxC,EAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GACtB,MAAO,CAAC,KAAI,EAAE,KAAI,GCvJpB,YAyDI,EAAM,EAAiB,GACzB,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,GAAE,QAEX,GAAM,GAAc,GAAM,EAAE,OACtB,EAAS,GAAK,EAAG,WACjB,EAAS,EAAQ,QAAQ,EAAQ,GACvC,SAAY,UACZ,EAAO,UACA,EAGT,GAAI,CAAC,GAAgB,EAAE,MAAO,GAG5B,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAE,MAAO,GAExD,GAAI,EAAE,QAAU,aACd,GAAM,GAAO,EAAQ,KAAK,GACpB,EAAS,GAAK,EAAM,GAC1B,SAAK,UACE,EAET,GAAI,IAAU,QACZ,MAAO,GAAQ,IAAI,GACd,GAAI,IAAU,QACnB,GAAM,GAAO,GAAO,EAAG,EAAE,OACnB,EAAS,EAAQ,SAAS,EAAG,GACnC,SAAK,UACE,MAEP,MAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,eAK/D,EAAM,GACR,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAO,EAAE,mBAG3B,EAAe,EAAc,GACxD,GAAM,GAAQ,GAAO,GAAU,GAAM,GAE/B,EAAS,GAAoB,EAAK,WACxC,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,qvCC3G1B,YAyBI,EAAM,EAAsB,GAC9B,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,MAAO,GAAW,IAAI,SAAA,GACpB,GAAM,GAAgB,EAAI,QAC1B,EAAU,GAAQ,EAClB,GAAM,GAAS,GAAM,EAAG,EAAO,GAC/B,SAAM,IAAS,EACR,ICjCX,YA2BI,EAAiC,GAEnC,OADM,GAAqB,GAAI,OAAM,EAAK,MACjC,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAK,MAAM,GAAK,EAAK,GAGrC,OADM,GAAS,GAAO,EAAU,EAAK,OAC5B,EAAI,EAAG,EAAI,EAAO,OAAO,OAAQ,EAAE,GAI1C,OAHM,GAAS,EAAO,WAAW,GAE3B,EAAwB,GAAI,OAAM,EAAK,MACpC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,MAAM,GAG1C,GAAM,GAAgB,EAAK,WAAW,GAEtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,WC7ChB,YAyBI,EAAe,EAAkB,EAAyB,EAC1D,GAOF,OALM,GAAU,EAAO,EAAO,OAAS,GACjC,EAAA,CAAA,EAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAc,GAAuB,EAAQ,EAAQ,GACrD,EAAiB,GAAuB,QAAS,EAAQ,GAEtD,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHM,GAAS,EAAI,EACb,EAAO,EAAE,SAAS,EAAQ,EAAS,GACnC,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAEzC,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAKrC,OAHM,GAAY,EAAI,EAChB,EAAW,EAAY,SAAS,EAAW,EAAY,GACvD,EAAc,EAAe,SAAS,EAAW,EAAY,GAC1D,EAAI,EAAG,EAAI,EAAG,IACrB,EAAS,GAAK,EAAU,GAAG,MAC3B,EAAY,GAAK,EAAU,GAAG,MAKlC,GAAM,GAAc,EAAO,QAC3B,SAAY,EAAY,OAAS,GAAK,EAC/B,CACL,GAAO,EAAa,EAAa,GACjC,GAAO,EAAgB,EAAa,UCxDxC,mJCAA,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,WAAY,SC7BtD,AA4BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAO,GAAK,EAAG,YACnB,EAAI,GAAK,GAAI,GAAO,GAAI,IAC9B,MAAO,IAAI,GAAI,EAAI,QCtC3B,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAK,EAAG,YAAa,IAC/C,MAAO,IAAI,EAAI,OCnCvB,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAGxB,MAAO,CAAC,EAAS,EAAG,KChDxB,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAsC,GAC5C,SAAM,QAAQ,SAAC,EAAG,GAChB,EAAK,GAAK,WAAM,MAAA,GAAG,WAEd,IC7BX,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,mBChChE,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,cAC7C,MAAO,IAAI,EAAI,OCpCvB,AA6BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,EAAI,EAAI,GAAI,EAAG,IACnB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,GAAI,EAAI,EAAI,GAAI,EAAG,KACvB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAExB,MAAO,CAAC,EAAS,EAAG,KCtDxB,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAK,EAAG,YAAa,QC/B7D,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,kBC/B3D,AA2DA,YACI,EAAkB,EAClB,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBAE3C,EAAO,EACP,EAAU,EACV,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,MAIvE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OAEvB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,IAG5C,EAAkC,CAAC,GAAI,EAAM,MAAO,GAEpD,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KChIvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAY,EAAS,EAAY,EACpD,OCpCV,AA8CA,YACI,EAAkB,EAAqB,EACvC,EAAkC,GACpC,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBAE/C,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,gCAC/B,EAAI,KAAI,MAEhB,GAAI,GAAU,EACV,EAAO,EACP,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EACI,EAAQ,EAAQ,CAAC,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KACvE,EAAO,EAAQ,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGhE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,4DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,+DACC,GAAQ,KAAI,OAEvB,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAmB,GAE3D,MAAO,GAAQ,gBAAgB,EAAM,EAAS,IAG1C,EAAgC,CAAC,GAAI,EAAM,MAAO,GAElD,EAA8B,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,GAEvD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAC5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAY,EAAS,OC7BvE,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,WAEnB,MAAI,CAAC,GAAc,CAAC,EACX,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,MAEtB,CAAC,GAAc,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,MAEtB,GAAc,CAAC,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,MAGzB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,KAC7B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,QChDrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,MACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAQtC,OAPM,GACF,EAEE,EAAa,EAAiB,WAC9B,EAAc,EAAiB,MAE/B,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAW,OAAS,EAAG,GAAK,EAAG,IAC1C,GAAI,EAAW,KAAO,EAAY,GAChC,EAAK,GAAK,UACD,EAAW,KAAO,EAC3B,KAAM,IAAI,OAAM,mBACZ,EAAU,6BAA6B,EAAW,MAI1D,OADM,GAAiB,GACd,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,EAAK,GAAK,GACZ,EAAK,KAAK,GAId,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAM,QC/CnC,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,GAAG,YCxBxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAET,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AA0BO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,aAAc,EAAA,EAAA,aACrB,MAAO,CACL,EAAG,WAAM,MAAA,IACL,GAAW,GAAa,EAAG,GAAe,GAAU,EAAG,IACvD,EAAI,GAAU,QCnCxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAS,EAAM,IAAI,SAAA,GAAK,MAAA,GAAE,QACzB,EAAA,EAAA,KACD,EAAQ,GAAe,EAAM,EAAM,GAAG,OAAO,GAC7C,EAAa,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAa,GAAM,EAAI,EAAY,GACzC,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,YAAM,MAAA,QC/BrC,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACN,EAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAEhC,SACI,GAA4B,GAC5B,WAAM,MAAA,8DACF,uDAAsD,EAAS,OAEhE,CACL,EAAG,WACC,MAAA,IAAoB,EAAI,MAAO,EAAI,EAAS,EAAS,EAAK,IAC9D,OAAQ,WACJ,MAAA,IAAqB,EAAK,EAAI,EAAQ,MAAO,EAAS,EAAK,OCxCrE,AAsBO,GAAM,IAA4C,CACvD,WAAY,GACZ,aAAc,CAAC,KAAM,UACrB,SAAU,SAAC,EAAe,EAAiB,GACnC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAAY,EAAA,EAAA,gBAGjC,MAAO,CACL,GAAI,WAAM,MAAA,IACN,EAAK,EAAQ,EAAS,EAAK,EAAY,EACvC,IACJ,OAAQ,WAAM,MAAA,IACV,EAAK,EAAI,EAAO,MAAO,EAAS,EAAK,EAAY,OCpC3D,AA2CA,YACI,EAAM,EAAO,EACb,EAA0C,GAC5C,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEnE,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAEzE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,EACI,EAAI,MAAM,KAAO,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAI,MAAM,GAAE,UAC1D,iCAAgC,EAAY,GAAE,OACtD,EACI,EAAK,MAAM,KAAO,EAAY,GAC9B,WAAM,MAAA,0CAA0C,EAAK,MAAM,GAAE,UACzD,mCAAkC,EAAY,GAAE,QAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAEhD,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,EAAG,GAEjD,EAAmC,CAAC,QAAO,EAAE,IAAG,GAEtD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAwB,GAG9B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC7F7D,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAC3B,EACI,GAAkB,GAClB,WACI,MAAA,kEACA,mDAAkD,EAAS,OAE5D,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAEZ,MAAO,CACL,EAAG,WAAM,MAAA,IACJ,EAAiB,MAAO,EAAI,EAAqB,EAAS,IAC/D,OAAQ,WAAM,MAAA,IACV,EAAiB,EAAK,EAAqB,MAAO,EAAS,OCzCrE,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAI,GAAK,EAAG,aAAc,OC/BvD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UAAW,EAAA,EAAA,QAExB,MAAO,CACL,EAAG,WACD,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAE,MAE7C,EAAM,GAAO,EAAI,EAAM,EAAW,CAAC,GAEvC,MAAI,IAAe,MACjB,GAAM,GAAU,EAAK,IAGhB,MCzCf,AAwBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAG1B,EAAa,GAAoB,CAAC,EAAG,GAE3C,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAU,OAEhB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,EACI,EAAE,OAAS,EACX,WAAM,MAAA,6DACF,yBAAwB,EAAE,KAAI,OACtC,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,8DACF,yBAAwB,EAAO,KAAI,OAC3C,EACI,EAAE,MAAM,KAAO,EAAO,MAAM,GAC5B,WAAM,MAAA,yDACF,cAAa,EAAE,MAAM,GAAE,0CACvB,cAAa,EAAO,MAAM,GAAE,OAEpC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,sCAAqC,EAAO,mBAC5C,KAAI,EAAU,QAEtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WACI,MAAA,gEACA,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAW,GACb,EAAE,MAAO,EAAO,MAAO,EAAS,EAChC,EAAK,EAAiB,IAE1B,MAAO,CACL,EAAG,WACC,MAAA,IAAmC,EAAE,MAAO,EAAI,EAAQ,IAC5D,OAAQ,WACJ,MAAA,IAAoC,EAAG,EAAI,EAAO,MAAO,OC7EnE,AAuBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAA6C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAC3D,EAA+C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAEnE,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,UACH,GAAyB,EACzB,IACb,OAAQ,WAAM,MAAA,GAAO,UACH,GACA,EAAsC,OCtC9D,AA6BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OACnD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAgC,EAAE,MAAO,GAC5D,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GAC/C,MAAO,GAAQ,OAAO,EAAI,IAGtB,EAAwB,CAAC,GAAE,EAAE,EAAC,GAEpC,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,cACZ,EAAoB,EAAgC,KACpD,QCrCV,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAI,EAAI,GAAI,GAAI,GAAO,KAAM,EAAI,KAAK,KAAK,KAAK,KACtD,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC/B7B,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC3B7B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAI,QC5BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AA6BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAiB,EAAE,MAAO,GAC7C,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AA8BO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,IAAK,OAAQ,WAAY,SACxC,SAAU,SACN,EAAY,EAAiB,GACxB,GAAA,GAAA,EAAA,gBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GAAM,EAAA,EAAA,GAAU,EAAA,EAAA,GAEpB,EAAa,GAAgB,GAAO,GACpC,EAAgB,GAAiB,EAAK,MAAO,EAAE,OAC/C,EAAsB,GAC5B,GAAI,EAAK,OAAS,GAChB,OAAS,GAAI,EAAG,EAAI,EAAE,MAAM,OAAS,EAAG,EAAE,EACxC,EAAU,KAAK,EAAE,MAAM,IAEzB,EAAU,KAAK,GAGjB,GAAM,GAAa,GAAI,EAAG,GACpB,EAAoB,EAAI,EAAI,GAC5B,EAAsB,GAAM,GAAI,EAAU,GAAO,KACjD,EAAiB,EACnB,EAAI,EAAI,EAAqB,GAAsB,GACnD,GAAO,MAEL,EAAO,WACX,MAAI,GAAK,OAAS,EACT,EACH,EAAI,EAAI,EACA,GACI,EAAQ,EAAqB,CAAC,EAAG,EAAG,EAAG,EAAK,MAAM,KAClD,IACR,GACJ,EAAE,OAEC,EAAQ,EAAI,EAAI,EAAI,GAAsB,GAAa,EAAE,QAG9D,EAAU,WACd,GAAI,GACA,EAAI,EAAI,EAAqB,GAAO,KAAM,GAC9C,MAAI,GAAK,OAAS,GAChB,GAAU,GAAI,EAAS,IAElB,EAAQ,EAAS,EAAK,QAEzB,EAAc,WAClB,GAAI,GAAc,EAAI,EAAI,EAAgB,GAAa,GAEvD,MAAI,GAAK,OAAS,GAChB,GAAc,GAAI,EAAa,IAE1B,EAAQ,EAAa,EAAK,QAE7B,EAAW,WACf,GAAM,GAAwB,EAAI,EAAY,GAE1C,EAAW,EAAI,EAAI,GACvB,MAAI,GAAK,OAAS,GAChB,GAAW,GAAI,EAAU,IAEpB,EAAQ,EAAU,EAAK,QAE1B,EAAY,WAChB,GAAI,GAAY,EAChB,MAAI,GAAK,OAAS,GAChB,GAAY,GAAI,EAAW,IAEtB,EAAQ,EAAW,EAAK,QAGjC,MAAO,CACL,EAAG,EACH,KAAM,EACN,SAAU,EACV,MAAO,EACP,OAAQ,KC1Gd,AA0BO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,WACpB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACH,EAAA,EAAA,KAED,EAAa,GAAe,EAAM,EAAE,OAAO,GAE3C,EAAO,WACX,GAAM,GAAc,EAAE,MAChB,EAAc,EAAQ,KAEtB,EAAa,EAAY,MAAM,EAAG,GAClC,EAAY,EAAW,OACvB,EAAa,EAAY,MAAM,EAAM,EAAY,QAAQ,MAAM,GAC/D,EAAY,EAAW,OAEvB,EAAmB,GAAW,EAAG,GACjC,EACF,GAAW,EAAY,EAAG,EAAY,EAAI,GAExC,EAAc,GAAY,CAAC,EAAY,CAAC,GAAc,IAEtD,EAAS,EAAQ,EAAI,GACrB,EAAkB,EAAQ,EAAS,CAAC,IAEpC,EACF,GAAY,CAAC,CAAC,GAAY,EAAkB,IAC1C,EAAkB,GAAU,EAAQ,GACtC,EAAa,GACb,EAAiB,EAA6B,EAAE,MAAM,IAEpD,EAAsB,GAAuB,GACnD,SAAa,GAAU,EAAY,GAE5B,GAET,MAAO,CAAC,EAAG,EAAM,QAAS,WAAM,MAAA,OAIpC,YAAoB,EAAe,GAEjC,OADM,GAAS,GACN,EAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,EAAO,KAAK,GAEd,MAAO,GAGT,YAAqB,GAEnB,OADM,GAAS,GACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,EACtC,EAAO,KAAK,EAAO,GAAG,IAG1B,MAAO,GCnFT,AAqBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,IAAI,EAAG,WAAM,MAAA,IAAU,OC1BtD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAK,EAAI,eCzB9B,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC5B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,EAAG,QC5BpC,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,EAAG,gBC5BrC,AAyBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,GACd,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KACP,MAAO,CACL,OAAQ,WACN,GAAM,GAAW,GACX,EAAU,GAAI,GACpB,MAAO,IAAI,EAAI,EAAI,GAAI,EAAI,EAAM,GAAW,QCpCpD,AAyBA,YACI,EAAM,EAAM,EAAO,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,QAAQ,EAAI,EAAG,EAAG,EAAa,EAAM,EAAO,IAElD,EAA4B,CAAC,EAAC,EAAE,EAAC,EAAE,GAAE,GAErC,EAA0B,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAE/D,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAa,GAGnB,GAAM,IACT,EAAG,CAAC,oCAAmC,KCxC3C,AAqBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,YAAa,EAAA,EAAA,KAAM,EAAA,EAAA,MAAO,EAAA,EAAA,KAEjC,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAG,EAAG,EAAI,EAAa,EAAM,EAAO,OC/B9C,YA6BI,EAAO,EAAM,EAAe,EAAoB,GAClD,MAAI,GAAE,KAAO,EAAM,MACjB,GAAI,EAAQ,EAAG,GAA+B,EAAE,MAAO,KAEzD,AAAI,EAAG,KAAO,EAAM,MAClB,GAAK,EAAQ,EAAI,GAA+B,EAAG,MAAO,KAErD,CACL,EAAG,WACD,GAAM,GAAK,EAAI,EAAI,GAAK,GAAM,EAAO,GAAI,EAAG,QAC5C,MAAO,IAAgB,KAAO,EAAK,GAAU,EAAI,KCvCvD,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,iBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAkB,EAAE,OACnD,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAa,EAAG,GAAI,aAC9C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,GAAI,aAC5C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA6DA,YACI,EAAkB,EAAqB,EACvC,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE9C,EAAO,EACP,EAAU,EACV,EAAW,EACX,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KAErE,EAAW,EAAQ,EAAS,CAC1B,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAI3E,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAS,OAAS,EAClB,WAAM,MAAA,kEACC,GAAS,KAAI,OAExB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OACrE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,EAAU,IAGtD,EACwB,CAAC,GAAI,EAAM,MAAO,EAAS,OAAQ,GAE3D,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC1IvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAG,EAAY,EAAS,EAAY,EACvD,OCrCV,AAkDA,YACI,EAAyB,EACzB,EAA6B,EAC7B,EAAkC,EAClC,GACF,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBACzC,EAAU,EAAgB,EAAQ,SAAU,mBAElD,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,+BAC/B,KAAI,EAAI,KAAI,OAEpB,EACI,EAAI,OAAS,EACb,WAAM,MAAA,4DACC,GAAI,KAAI,OACnB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,+DACC,GAAO,KAAI,OACtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAO,MAAO,EAAY,EAAS,EAAmB,EACtD,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAQ,EAAS,IAGjD,EACsB,CAAC,GAAI,EAAK,MAAO,EAAQ,OAAQ,GAEvD,EACqB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAErE,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,GAC/C,GAGN,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAE5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAG,EAAY,EAAS,OC/B1E,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,KACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAM,EAAE,OACvC,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAU,EAAG,GAAI,aAC3C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAQ,EAAG,GAAI,aAC/C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA4BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAI,GAAa,EAAE,OAEjC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAI,GAAM,GAAI,EAAG,MAC/B,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KClDxB,AA0BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,OCzBzB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,WACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAU,EAAM,GACtB,MAAO,CAAC,QAAS,WAAM,MAAA,IAAM,EAAQ,MAAO,eC3BhD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAGtC,GAAM,GAAI,EAAM,GACT,EAAA,EAAA,SACD,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAM,EAAI,EAAO,EAAE,WC/BxC,AAgCO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GACP,EAAO,EACP,EAAM,EACN,EACF,GAA0C,EAAK,MAAO,EAAI,OAExD,EAAU,WACd,GAAM,GAAW,GAAK,EAAK,WACvB,EAAM,EAAI,EAAI,EAAI,EAAU,GAAI,EAAM,GAAI,EAAU,GAAO,OACzD,EAAa,GAAgC,EAAK,MAAO,GAC/D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,QAErB,EAAS,WACb,GAAM,GAAY,GAAQ,EAAM,GAC1B,EAAU,GAAM,EAAW,GAAI,GAAO,GAAU,IAClD,EAAM,EAAI,EAAI,EAAI,EAAG,IACnB,EAAa,GAAgC,EAAI,MAAO,GAC9D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAI,QAE1B,MAAO,CAAC,EAAG,EAAS,EAAG,KC9D3B,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,SACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,GAAQ,EAAG,GAExB,MAAO,CACL,EAAG,WAAM,MAAA,IAAM,EAAM,EAAI,EAAI,EAAI,KACjC,MAAO,WACL,GAAI,GAAM,GAAM,EAAM,GAAU,GAAK,EAAI,EAAI,IACvC,EAAa,GAAiB,EAAM,MAAO,EAAG,OACpD,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAM,WC1ClC,AAwBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,SC7BxC,AAwBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAO,EAAI,GAAU,EAAG,GAAI,GAAK,IAEvC,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAM,gBC/BxC,AAuBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,GAAI,gBC5B3C,AAqBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAQ,EAAI,EAAE,WC1BnC,AAuBO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,uBACX,EAAI,EAAoB,IAGxB,EAAmC,CAAC,OAAM,GAC1C,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAAoB,IAExB,MAAO,CAAC,OAAQ,KCxCpB,AAuBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,8BACX,EAAI,EAAoB,IAGxB,EAA0C,CAAC,OAAM,GACjD,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAA2B,IAE/B,MAAO,CAAC,OAAQ,KCxCpB,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,KACD,EAAO,GAAe,EAAM,EAAG,OACrC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAQ,EAAI,OC3BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAyBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,GAAI,EAAI,EAAI,GAAI,EAAG,KAAM,SC9BlD,AAyBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,aACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAGL,UAAW,WAAM,MAAA,IAAK,GAAU,GAAY,YAC5C,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAW,EAAG,SACpC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAW,GAAY,EAAG,YCnCtD,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CACL,EAAG,WACD,GAAM,GAAO,GAAQ,EAAG,GAAO,IAEzB,EAAa,GAAO,IACpB,EAAQ,GAAO,IAEf,EAAqB,EAAI,EAAI,GAC7B,EACF,EAAI,EAAI,EAAI,GAAa,GAAI,GAAK,EAAG,aAEzC,MAAO,IAAM,EAAM,EAAoB,OC3C/C,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,EAAG,GAAI,GAAO,GAAI,SC9BnD,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAK,EAAG,YAAa,OC7BlD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GActC,OAbO,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,KAER,EAAa,EAAE,MACf,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAQT,EAAoC,GACjC,EAAI,EAAG,EAAI,EAAG,KAAM,IAC3B,EAAS,KAAK,CAAC,EAAO,GAAI,EAAW,GAAK,EAAO,GAAK,EAAM,KAE9D,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,OC3C7B,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,IACD,EAAW,GAEX,EAAW,EAAI,EAAI,GACzB,MAAO,CACL,OAAQ,WAAM,MAAA,IAAI,EAAU,EAAI,GAAI,EAAU,CAAC,GAAM,GAAW,QClCtE,AAuBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAQ,QC7BrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,SACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,KAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAO,EAAI,OC1BhC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAI,GAAK,GAAK,EAAG,YAAa,QC/B3D,AAuBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,GAAK,EAAG,WAAY,QC5BrD,AAwBO,GAAM,IAA0C,CACrD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAM,GAAO,GACb,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MACrC,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MAC3C,MAAO,CAAC,EAAS,EAAG,KChCxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,GAAI,GAAM,EAAE,QAG7B,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAkB,EAAE,MAAM,QACzB,EAAA,EAAA,KAED,EAAO,GAAe,EAAM,EAAE,OACpC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,EAAI,EAAY,GAAK,EAAE,MAAO,YAE3C,MAAO,CAAC,EAAG,WAAM,MAAA,OCxCrB,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAO,GAAI,SC9BxC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAO,GAAI,GAAO,IAAK,OC/BpD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KAED,EAAO,WACX,GAAI,GAAQ,GAAU,GAGtB,GAAI,EAAE,OAAS,EACb,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,IAAK,CAAC,EAAE,MAAM,cAEjD,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IAAK,CACjD,EAAE,MAAM,GAAI,EAAE,MAAM,cAI7B,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IACjD,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,cAI1C,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EACA,CACE,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAC5C,EAAI,EAAE,MAAM,IAEd,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,UAMjE,MAAM,IAAI,OACN,2DACG,GAAE,KAAI,kBAEf,MAAO,IAET,MAAO,CAAC,EAAG,KCpFf,AAuBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAiC,EAChC,EAAA,EAAA,KACD,EAAW,GAAiC,GAClD,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,EAAI,OC7BnC,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACxC,GAAM,GAA2B,EAC1B,EAAA,EAAA,KACP,MAAO,CAAC,MAAO,WAAM,MAAA,IAAM,EAAI,OC1BnC,AA8BO,GAAM,IAA2C,CACtD,WAAY,GACZ,aAAc,CAAC,cACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAAO,WACX,MAAO,IAAoB,EAAI,IAEjC,MAAO,CAAC,EAAG,KAIf,YAA+C,EAAM,GAQnD,OAJM,GAAqB,GAAQ,EAAS,GAAU,IAChD,EAAW,GAAO,EAAG,GACvB,EAAa,GAAa,EAAS,GAAO,EAAG,UAC3C,EAAW,EAAS,KAAO,EAAW,KACnC,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,GAAW,EAAY,EAAI,GAE1C,EAAa,GAAW,EAAY,GAAK,EAAS,MAAO,SACzD,GAAM,GAAY,GAAU,GAC5B,MAAO,IAAM,EAAY,EAAU,GCxDrC,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAkHA,GAAM,IAA4B,CAChC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA6B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAxB,GAAM,IAAc,GAAA,IACvB,GAAiB,ICxNnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA+BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACL,EAAO,KAAK,OAAS,EAAG,WAAM,MAAA,wCACvB,EAAQ,KAAM,KClCvB,AAmCA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAQ,KAAM,ICtCvB,AA8BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AAkCA,EAAO,UAAU,KAAO,SACpB,EAAc,GAChB,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,KCrC9B,AAoCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,GACjC,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,KCvCvC,AAqCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,GAChD,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,KCxC9C,AAwCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,EAC9C,GACF,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,EAAQ,KC5CtD,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA+BA,EAAO,UAAU,UAAY,SACzB,EACA,EACA,EACA,EACA,GACF,YAAK,kBACE,GAAU,KAAM,EAAM,EAAU,EAAQ,EAAO,ICtCxD,AA2BA,EAAO,UAAU,YAAc,SAAyB,GAEtD,YAAK,kBACE,GAAY,KAAM,IC9B3B,AA4BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,YAAc,SAC3B,EAAa,GACf,YAAK,kBACE,GAAY,KAAM,EAAK,IC/BhC,AA0BA,EAAO,UAAU,OAAS,SACtB,EAA0B,GAC5B,YAAK,kBACL,AAAI,YAAa,IACf,GAAI,CAAC,IAEA,GAAM,CAAE,MAAI,OAAK,GAAI,IChC9B,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4C,EAC5C,EAAmB,GACrB,YAAK,kBACE,GACI,KAAM,EAAQ,EAAQ,EAAK,EAAY,EACvC,ICrCb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EACA,EAAkC,EAClC,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAa,EAAS,EAAK,ICrCtD,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAqB,GACtC,YAAK,kBACE,GAAO,KAAM,EAAM,EAAW,IC9BvC,AA2BA,EAAO,UAAU,aAAe,SAC5B,EAAmB,GACrB,YAAK,kBACE,GAAa,KAAM,EAAW,IC9BvC,AAkCA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,UAAgB,8DAChB,KAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,IC3Cb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA6BA,EAAO,UAAU,WAAa,SAC1B,EAA+B,EAC/B,EAAqB,EACrB,GACF,YAAK,kBACE,GAAW,KAAM,EAAQ,EAAS,EAAK,EAAW,IClC3D,AA2BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC9BxB,AA4BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC/BzB,AA2BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC7BnB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC5Bb,AA+BA,EAAO,UAAU,YAAc,SAClB,GACX,YAAK,kBACE,GAAY,KAAM,IClC3B,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA0BA,EAAO,UAAU,WAAa,SAA2B,GACvD,YAAK,kBACE,GAAW,KAAM,IC5B1B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA4BA,EAAO,UAAU,OAAS,SACb,EAA4B,GACvC,YAAK,kBACE,GAAO,KAAM,EAAS,IC/B/B,AA+BA,EAAO,UAAU,mBAAqB,SACzB,GACX,YAAK,kBACE,GAAmB,KAAM,IClClC,AA0BA,EAAO,UAAU,aAAe,SAA2B,GAEzD,YAAK,kBACE,GAAa,KAAM,IC7B5B,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA+BA,EAAO,UAAU,gBAAkB,SACtB,GACX,YAAK,kBACE,GAAgB,KAAM,IClC/B,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA0BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC5BpB,AA2BA,EAAO,UAAU,2BAA6B,SAC1C,EAAsB,EAAe,EAAgB,GACvD,YAAK,kBACE,GAA2B,KAAM,EAAa,EAAM,EAAO,IC9BpE,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA2BA,EAAO,UAAU,UAAY,SAChB,EAAwB,GACnC,YAAK,kBACE,GAAU,KAAM,EAAM,IC9B/B,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA0BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC5BpB,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA2BA,EAAO,UAAU,OAAS,SACb,EAAiB,EAAsB,GAClD,YAAK,kBACE,GAAO,KAAM,EAAG,EAAY,IC9BrC,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,KAAO,SACpB,EAAwB,GAC1B,YAAK,kBACE,GAAK,KAAM,EAAM,IC7B1B,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,EAAI,KAAM,IC5BnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,KAAO,SACpB,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAK,EAAM,IClC/B,AA+BA,EAAO,UAAU,eAAiB,SACrB,GACX,YAAK,kBACE,GAAe,KAAM,IClC9B,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAa,GAAb,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAC9B,KAAK,kBACE,GAAO,KAAM,EAAO,EAAS,IC9BtC,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,IAAM,SACV,EAAmC,GAC9C,YAAK,kBACE,GAAI,KAAM,EAAU,IC/B7B,AA6BA,EAAO,UAAU,KAAO,SACX,EAAsC,EAC/C,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAa,EAAa,EAAS,EAAc,IClCrE,AA+BA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,GAAU,KAAM,ICjCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,MAAQ,SACZ,GACX,YAAK,kBACE,GAAM,KAAM,IC7BrB,AA2BA,EAAO,UAAU,KAAO,SACX,EAAwB,GACnC,YAAK,kBACE,GAAK,KAAM,EAAM,IC9B1B,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA0BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC5Bf,AAkCA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,EAAQ,KAAM,EAAE,QCpCzB,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,EAAQ,KAAM,IC5BvB,AA2BA,EAAO,UAAU,eAAiB,SACrB,EAA8B,GACzC,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,sBAAwB,SAC5B,EAA8B,GACzC,YAAK,kBACE,GAAsB,KAAM,EAAY,IC9BjD,AA0BA,EAAO,UAAU,QAAU,SACd,GACX,YAAK,kBACE,GAAQ,KAAM,IC7BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EAAsC,EACtC,EAAqB,EACrB,GACF,YAAK,kBACE,GACI,KAAM,EAAiB,EAAiB,EAAS,EAAK,EACtD,ICtCb,AA4BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,OC9BjB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA6BA,EAAO,UAAU,MAAQ,SACZ,EAAwB,GACnC,YAAK,kBACE,GAAM,KAAM,EAAO,IChC5B,AA4BA,EAAO,UAAU,QAAU,SAAoC,GAC7D,YAAK,kBACE,GAAQ,KAAM,IC9BvB,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,MAAQ,SACrB,EAAkC,GACpC,YAAK,kBACE,GAAM,KAAM,EAAiB,IC9BtC,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAO,OC9BhB,AA2BA,EAAO,UAAU,kBAAoB,SAA2B,GAE9D,YAAK,kBACE,GAAkB,KAAM,IC9BjC,AA+BA,EAAO,UAAU,wBAA0B,SAC9B,GACX,YAAK,kBACE,GAAwB,KAAM,IClCvC,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,MAAQ,SACrB,EAAoB,GACtB,KAAK,kBACL,GAAM,GAAqB,YAAa,GAAS,CAAC,KAAM,GAAE,CAAI,MAAI,OAAK,GACvE,MAAO,IAAM,EAAoB,IC9BnC,AA4BA,EAAO,UAAU,KAAO,SAAoC,GAC1D,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA+BA,EAAO,UAAU,aAAe,SACd,EAAiB,EAAe,EAC9C,EAAoB,EAAkB,EACtC,EAAsB,GACxB,YAAK,kBACE,GACI,KAAM,EAAO,EAAK,EAAS,EAAW,EAAS,EAC/C,EAAa,ICtC1B,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC7BzB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC7BpB,AAgCA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAQ,KAAM,SClCvB,AAgCA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,KAAM,YClCvB,AAgCA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAQ,KAAM,UClCvB,AA4BA,EAAO,UAAU,KAAO,SACX,EAAY,GACvB,YAAK,kBACE,GAAK,KAAM,EAAG,IC/BvB,AA2BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC9BzB,AA2BA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAO,KAAM,IC9BtB,AA4BA,EAAO,UAAU,mBAAqB,SACzB,EAAmC,GAC9C,YAAK,kBACE,GAAmB,KAAM,EAAY,IC/B9C,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,MAAQ,SACrB,EAA8B,GAChC,YAAK,kBACE,GAAM,EAAW,KAAM,IC/BhC,AA4BA,EAAO,UAAU,UAAY,WAC3B,YAAK,kBACE,GAAU,4vQC9BnB,AAaA,GAAI,kBAMF,MAAI,KAAY,MACd,IAAW,EAAA,UAAU,WAEhB,iBAeP,MAAO,eCrCT,AAuBA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAChC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAa,aAE7C,MAAA,IANkC,OAWlC,EAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OAWhC,GAAA,SAAA,GAAyC,EAAA,EAAA,GACvC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAoB,aAEpD,MAAA,IANyC,OAWzC,GAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OC9EhC,YAuB6B,EAAY,GACvC,GAAI,MAAM,QAAQ,IAGhB,OADI,GAAkB,GACb,EAAI,EAAG,EAAI,EAAW,IAC7B,EAAW,EAAS,OAAO,GAE7B,MAAO,QAEP,GAAM,GAAW,GAAI,OAAM,GAC3B,SAAS,KAAK,GACP,eAIY,EAAc,GACnC,GAAI,CAAC,EACH,KAAM,IAAI,IAAe,eAOJ,EAAY,GAEnC,OADI,GAAU,EACK,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,IAAS,GACX,IAGJ,MAAO,eAQ2B,GAClC,MAAI,GAAG,SAAW,EACT,EAAG,GAEL,cAYc,GACrB,MAAI,OAAM,QAAQ,GACT,EAEF,CAAC,eA0BkB,GAC1B,GAAM,GAAe,EAAK,QAAQ,uBAAwB,SACpD,EACF,EAAa,QAAQ,kBAAmB,SAAS,cAKrD,MAAI,GAAS,KAAO,IACX,EAEF,UAAY,cAGO,GAE1B,MAAI,GAAW,QAAU,GAIrB,EAAW,QAAQ,OAAS,GAHvB,EAMF,EAAW,QAAQ,cAAe,SAAC,EAAG,GAAO,MAAA,GAAG,gBAIzD,GAAI,IAAyB,eAEQ,GAEnC,GAAI,GAAa,KACf,MAAO,MAET,GAAM,GAAsC,GAC5C,SAAK,UAAe,EAAS,eAC7B,EAAK,OAAY,EAAS,YACnB,EAcT,YAAuC,GAErC,GAAI,GAAU,MAAQ,MAAO,IAAW,SACtC,OACK,GAAI,MAAM,QAAQ,GACvB,EAAO,QAAQ,SAAA,GAAc,MAAA,IAA8B,SAG3D,QADM,GAAS,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAO,GACrB,AAAI,GAAS,MAAQ,MAAO,IAAU,UACpC,CAAI,CAAC,MAAM,QAAQ,IAAU,EAAM,OAAY,WAC3C,MAAO,GAAM,OAAa,SAC5B,EAAO,GAAS,EAAM,MAEtB,GAA8B,iBAqBpC,EACA,EACA,EACA,EAAgC,aAElC,GAJE,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,UAAgC,AAAA,IAAA,QAAA,GAAA,IAE9B,MAAO,IAAe,UACxB,GAAM,GAAe,EACjB,EAAE,OACN,GAAI,IAAgB,GAClB,EAAK,EAAc,WACV,IAAgB,IACzB,EAAK,GAAuB,WAE5B,EAAK,EAAc,GACf,GAAM,KACR,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAU;EAE7C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAKR,MAAO,QAGP,GAAM,GAAS,EACf,GAAI,EAAO,WAAgB,MAAQ,EAAO,QAAa,KACrD,KAAM,IAAI,GACH,EAAmB,6BACnB,MAAK,UAAU,GAAO;GACzB,sCAEN,GAAM,GAAY,EAAO,UACrB,EAAG,OAAE,EAAU,OAQnB,GAPA,AAAI,IAAa,GACf,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,GAAA,UAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IAEJ,GAAO,KACT,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAS;EAE5C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAIN,GAAI,GAAc,MAQhB,OADM,GAAwB,GACZ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,GAAuB,GAEtD,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,EAAc,GAG7C,GAAM,GAAe,EAAO,OAC5B,EAAa,cAAmB,EAGhC,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAE9C,GAA8B,EAAO,QACrC,GAAM,GACF,EAAW,EAAK,EAAO,OAAW,EAAe,GACrD,UAAsB,GAAA,GAAO,GAEtB,OAMP,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAK9C,GAAM,GAAY,GAAI,GAAI,EAAO,QACjC,UAAsB,GAAA,GAAO,GACtB,gBAUiB,EAAW,GACvC,MAAQ,GAAI,EAAK,GAAO,EAAI,EAAK,EAAI,cAQF,EAAW,GAC9C,MAAO,GAAK,GAAc,EAAG,eA2CL,GACxB,GAAI,GAAM,KACR,MAAO,GAIT,OAFM,GAAW,GAED,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAI,QAAQ,KAAO,IACrB,EAAI,KAAK,GAGb,MAAO,eASqB,GAC5B,GAAI,GAAO,KACT,KAAM,IAAI,GAAW,yBAAyB,KAAK,UAAU,IAE/D,OAAW,KAAO,GAChB,GAAI,EAAI,eAAe,GACrB,MAAO,GAGX,MAAO,eAWL,EAAkB,EAAe,GACnC,GAAI,GAAS,KACX,OAEF,GAAI,EAAO,QAAQ,GAAS,EAC1B,KAAM,IAAI,GAAc,EAAK,mBAAmB,EAAK,uBACjD,EAAM,mCAoBV,EAAQ,EAAsB,EAC9B,GAD8B,MAAA,KAAA,QAAA,GAAA,GAC9B,AAAA,IAAA,QAAA,GAAA,UACF,GAAO,GAAa,GACpB,GAAO,GAAa,GAEhB,MAAM,QAAQ,IAAM,EAAE,QAAU,GAAa,EAAE,QAAU,GACzD,EAAE,MAAM,SAAA,GAAK,MAAA,OAAO,KAAM,gBAWM,EAAwB,GAC5D,AAAI,MAAM,QAAQ,GAChB,GAAA,KAAK,OACD,EAAM,OAAS,EAAG,WAAM,MAAG,GAAI,qCACnC,EAAM,QACF,SAAC,EAAG,GAAM,MAAA,IAAsB,EAAG,WAAW,GAAI,GAAC,OAAO,MAE9D,EAAA,KAAK,OACD,OAAO,UAAU,IAAU,EAAQ,EACnC,WAAM,MAAA,YAAY,EAAI,sCACf,IAAuB,GAAM,mBAeL,GACrC,MAAI,KAAU,KACL,OACE,MAAM,QAAQ,GAChB,IAAM,EAAM,IAAI,SAAA,GAAK,MAAA,IAAuB,KAAI,KAAK,KAAO,IAC1D,MAAO,IAAU,SACnB,IAAI,EAAK,IAET,GAAG,cAcV,EAA8B,GAChC,GAAI,GAAW,EAAA,KAAK,MAChB,EACE,EAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,GAAM,GAAM,EAAA,KAAK,MACjB,MAAI,GAAM,EAAW,GAGrB,GAAW,EACX,EAAa,EAAC,MAAA,OAAI,IACX,GAET,MAAO,eASkC,GAEzC,MAAI,KAAmB,OACd,OAEL,IAAmB,SACd,SAEL,IAAmB,MACd,MAEF,KCvgBT,AAoBA,YAAqB,EAAW,GAC9B,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAI,EAAM,OAY1D,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAGE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GANyC,EAAA,cAAc,cA8BvD,GAAA,SAAA,GAA6B,EAAA,EAAA,GAQ3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAJQ,SAAA,gBAAkB,EAClB,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,YAAgB,EAAO,EAAG,EAAK,UAC/C,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,SAAU,KAAK,SAAU,KAAM,KAAK,OAtB9B,EAAA,UAAY,UAwB9B,GA1B6B,IA2B7B,EAAA,cAAc,cAAc,IAoB5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAK5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAFQ,SAAA,YAAc,EAG7B,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,KAAW,GAAY,EAAG,EAAK,WAG9D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OAdL,EAAA,UAAY,WAgB9B,GAlB8B,IAmB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAIE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAHF,EAAA,UAAY,SAK9B,GAP4B,IAQ5B,EAAA,cAAc,cAAc,IAoC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAY9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KANQ,SAAA,gBAAkB,EAClB,EAAA,gBAAkB,EAClB,EAAA,YAAc,EACd,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,YACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,IACZ,EAAA,IACI,EAAK,KAAM,EAAA,YAAgB,EAAO,EAAK,SAAU,EAAK,WAC1D,EAAA,IAAQ,EAAM,EAAK,KAAM,IAC7B,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,SAAU,KAAK,SACf,SAAU,KAAK,SACf,KAAM,KAAK,KACX,KAAM,KAAK,OApCC,EAAA,UAAY,aAuC9B,GAzCgC,IA0ChC,EAAA,cAAc,cAAc,IAQrB,GAAM,IACwC,CAC/C,QAAW,UACX,WAAc,aACd,OAAU,SACV,SAAY,wBAGkB,GAElC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA0C,GAC1C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCpQjC,YAyBwB,GACtB,MAAO,IAAI,IAAQ,eAQI,GACvB,MAAO,IAAI,IAAS,iBASpB,MAAO,IAAI,gBAIc,GACzB,MAAO,IAAI,IAAW,0ECjDxB,AAqBO,GAAM,IAA2B,CAAC,gBAAiB,gBAO7C,GAA4B,CAAC,QAAS,OAAQ,UAI9C,GAAyB,CAAC,MAAO,OAIjC,GAAkC,CAAC,MAAO,MAAO,SAAU,OCpCxE,AAmBA,GAAM,IAA+B,GAAI,iBAET,GAC9B,GAA0B,GAA0B,aAAc,eAGnC,GAC/B,GAA0B,GAA2B,cAAe,eAGxC,GAC5B,GAA0B,GAAwB,WAAY,GAGhE,GAAM,IAA4B,GAC5B,GAAoB,gBAKG,EAAc,GACzC,GAAgB,KAAK,GACrB,IACE,GAAM,GAAS,IACf,UAAgB,MACT,QACA,GACP,SAAgB,MACV,GAOV,cACE,MAAI,IAAgB,SAAW,EACtB,GAEA,GAAgB,KAAK,IAAqB,eASjB,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,MAAO,MAA2B,cAaA,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,AAAK,GAAQ,IAAI,IACf,GAAQ,IAAI,EAAY,GAE1B,GAAM,GAAQ,GAAQ,IAAI,GAG1B,GAFA,GAAQ,IAAI,EAAY,GAAQ,IAAI,GAAc,GAE9C,EAAQ,GACV,GAAM,GAAY,EAAU,IAAI,EAGhC,UAAQ,IAAI,EAAQ,GACb,MAEP,OAAO,GAIX,GAAM,IAAkB,GAAI,QAAO,+CAOD,GAChC,MAAO,CAAC,CAAC,EAAK,MAAM,ICjHtB,YA8B0B,GACxB,MAAO,KAAM,SAAS,EAAE,WAAY,gBAWlC,EAA4B,EAAgB,GAC9C,AAAI,GAAS,MACX,GAAQ,GAEV,AAAI,GAAO,MACT,GAAM,EAAM,QAId,OADI,GAAO,EACF,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,GAAQ,EAAM,GAEhB,MAAO,GAQT,YAAmB,GACjB,SAAQ,MAAM,QAAQ,GAAS,GAAI,cAAa,GAAS,EAClD,EAAA,SAAS,eAQE,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAQ1B,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAsDxB,EAAe,GACnC,GAAI,EAAM,EACR,KAAM,IAAI,GAAW,QAAQ,EAAG,cAAc,EAAK,mBAGrD,OADM,GAAgB,GACb,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,EAAI,KAAK,GAEX,MAAO,GChJT,YAwEqB,EAAW,GAC9B,MAAO,GAAE,OAAO,eASS,EAAW,GAAA,AAAA,IAAA,QAAA,GAAA,IACpC,GAAM,GAAW,EAAE,MAAM,QACzB,MAAI,GAAO,GACT,GAAO,EAAS,OAAS,EAAO,GAElC,EAAS,OAAO,EAAM,EAAG,GAClB,EAAE,QAAQ,eAcI,EAAW,GAChC,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,oDACA,SAAQ,EAAE,MAAM,OAAM,aAE5B,GAAM,GAAI,GAAW,EAAG,GACxB,MAAO,IAAK,EAAG,CAAC,EAAG,EAAG,kBASF,GACtB,GAAM,GAAW,CAAC,GAAqB,EAAE,QACzC,MAAO,GAAE,QAAQ,eAWU,GAC3B,GAAI,EAAE,MAAQ,EACZ,KAAM,IAAI,GACN,wDAAwD,EAAE,KAAI,KAEpE,GAAM,GAAW,CAAC,EAAE,MAAM,GAAI,GAAqB,EAAE,MAAO,IAC5D,MAAO,GAAE,QAAQ,eAYf,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,GAAI,CAAC,EAAM,EAAM,MAAM,SACnD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,GAC9B,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,EAAG,GACjC,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,MAAU,EAAmB,CAAC,EAAO,EAAG,EAAG,EAAG,GAAI,CACvD,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SAEjE,GACH,MAAO,GAAA,MAAU,EAAO,CAAC,EAAO,EAAG,EAAG,EAAG,EAAG,GAAI,CAC9C,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAClE,EAAM,MAAM,aAGd,KAAM,IAAI,GACN,8DACA,IAAG,EAAM,sBAcjB,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,GAAQ,CAAC,EAAM,MAAM,GAAI,QACjD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,GAC1B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,QAClC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAG,GAC7B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,YAErD,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAejB,EAAe,EAAe,EAAc,GAC9C,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,GAC9B,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,EAAG,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAO,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpD,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,YAGX,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAWO,EAAmB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC7C,GAAI,GACJ,MAAI,GAAO,GACT,GAAO,EAAQ,GAAG,KAClB,AAAI,IAAS,EACX,EAAO,EAEP,EAAO,GAGX,AAAI,IAAS,EAAQ,GAAG,MAGtB,GAAO,IAGF,EAAA,OAAW,EAAS,eAUQ,EAAW,GAC9C,OAAQ,EAAE,UACH,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,QACjC,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,WAEpD,KAAM,IAAI,GACN,kDACA,iBAAgB,EAAE,oBAWP,EAAW,GAI9B,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAI,CAAC,IAEH,EAAE,OAAS,EAAE,OACf,KAAM,IAAI,GACN,0BAA0B,EAAE,OAAM,oBAClC,yCAAwC,EAAE,KAAI,MAEpD,MAAO,GAAA,KAAS,EAAG,eAgBjB,EAAc,EAAY,EAAc,EACxC,GADc,MAAA,KAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAErB,EAAA,aAAiB,EAAO,EAAM,EAAQ,EAAO,eAsBlD,EAAW,EAAW,EACtB,GACF,GAAK,EAAE,KAAO,GAAO,EAAE,KAAO,EAC5B,KAAM,IAAI,IACN,2CACA,uBAAsB,EAAE,MAAK,kBAAkB,EAAE,QAEvD,GAAI,EAAE,MAAQ,GACZ,GAAM,GAAW,EAAE,MAAM,MAAM,IAAI,GAC7B,EAAiB,EAAE,MAAM,MAAM,IAAI,GACzC,GAAI,IAAa,EACf,KAAM,IAAI,IACN,2CACA,yDACI,EAAE,MAAK,SACX,eAAc,EAAE,QAIxB,GAAK,EAAE,OAAS,GAAO,EAAE,OAAS,GAChC,GAAM,GAAa,GACb,EAAa,GAInB,MAAO,GAAA,MAAU,OAAO,CACtB,EAAC,EACD,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,SAIZ,GAAM,GAAa,EAAE,MAAM,QACrB,EAAW,EAAW,MAC5B,EAAI,EAAE,QAAQ,CAAC,GAAI,IAInB,GAAM,GAAS,EAAE,MAAM,QACjB,EAAW,EAAO,MAClB,EAAiB,EAAO,MACxB,EAAiB,EAAM,OAAA,CAAE,IAGzB,EAAO,MAAM,KAAK,CAAC,OAAQ,EAAE,MAAO,SAAC,EAAG,GAC5C,MAAI,KAAM,EACD,EAAE,KAAO,EACP,GAAK,EAAE,KAAO,EAChB,EAAI,EAEN,IAET,EAAI,EAAE,UAAU,GAAM,QAAQ,CAAC,EAAgB,KAG/C,GAAM,GAAkB,EAAU,OAAK,GACjC,EAAa,GACb,EAAa,GACnB,MAAO,GAAA,MACF,OAAO,CACN,EAAC,EACD,EAAC,EACD,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,IAEX,QAAQ,gBAyDb,EAAmB,EAA4B,GACjD,MAAO,GAAA,KAAK,WACV,MAAI,OAAM,QAAQ,GAChB,EAAU,EAAA,SAAS,EAAS,SAE5B,EAAU,EAAQ,QAEb,EAAA,OAAW,EAAW,EAAS,iBASnB,GACrB,MAAO,GAAA,IAAQ,EAAG,GA+BpB,YAAqB,EAAe,EAAc,GAChD,GAAM,GAAY,EAAK,MAEvB,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,GACN,+BAA+B,EAAK,KACpC,6BAA4B,IAGlC,GAAI,IAAU,GACZ,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,EAAG,IAErC,EAAK,QACR,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,GAAI,EAAU,KAEzD,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAG,EAAU,KAEpC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,IAElC,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,KAE3D,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAU,KAEjC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,IAE/B,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,KAE7C,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAU,KAE9B,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,EAAQ,EACjB,MAAO,GAET,KAAM,IAAI,GAAW,sCAAsC,EAAK,kBAc9D,EAAW,EAAc,GAC3B,MAAO,GAAA,KAAK,WACV,MAAI,IAAc,MAChB,GAAa,MAEf,GAAgB,GAET,EAAE,IAAI,GAAY,EAAE,KAAM,EAAM,kBAUvB,EAAW,GAE7B,GAF6B,AAAA,IAAA,QAAA,GAAA,GAEzB,IAAU,EACZ,KAAM,IAAI,IACN,0CAA0C,EAAK,6BAGrD,MAAO,GAAA,IAAQ,eAWQ,GACvB,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,GAAG,IAAI,kBAc1C,EAAW,EAAe,EAAuB,GACnD,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAG,EAAO,EAAY,iBAY1B,GAC1B,MAAO,GAAA,KAAK,WACV,GAAM,GAAI,EAAA,IAAQ,GAAI,EAAA,IAAQ,GAAI,IAClC,MAAO,GAAA,YAAgB,EAAG,EAAG,iBAiBD,EAAY,EAAc,GAAA,MAAA,KAAA,QAAA,GAAA,IACjD,EAAW,IAAM,IC7sB1B,AAgBO,GAAM,IAAwB,CAAC,QAAS,SAAU,UAO5C,GACT,CAAC,SAAU,UAAW,mBCxB1B,YAoB6B,GAC3B,GAA0B,GAAuB,UAAW,eAG5B,GAChC,GAA0B,GAA2B,eAAgB,GASvE,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DACS,SAAA,UAAA,4BAAP,WACE,MAAO,IAUT,EAAA,UAAA,UAAA,WACE,MAAO,IAEX,GAf0C,EAAA,cAAc,cAiBxD,GAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,MAAM,EAAO,IAHf,EAAA,UAAY,QAKrB,GAP2B,IAQ3B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,KAAK,EAAO,IAHd,EAAA,UAAY,OAKrB,GAP0B,IAQ1B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,GACN,oDAAoD,GAE1D,GAAI,EAAK,QAAU,OACjB,KAAM,IAAI,GAAW,sCAAsC,GAE7D,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAI,EAAA,OAAO,EAAK,OAAQ,EAAA,KAAK,EAAO,OAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,QApBT,EAAA,UAAY,WAuBrB,GAzB8B,IA0B9B,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GASjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,eAAiB,KACjB,EAAA,eAAiB,IAOxB,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,cAAc,EAAO,KAAK,OAAQ,KAAK,OAAQ,IAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,OAAQ,KAAK,OAAQ,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAnBxD,EAAA,UAAY,gBAqBrB,GAvBmC,IAwBnC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAShC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,uCAAuC,EAAK,KAGlD,MAAO,IAAe,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGnE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,eA2BrB,GA7BkC,IA8BlC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,0CAA0C,EAAK,KAErD,MAAO,GAAA,gBAAgB,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGpE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,kBA2BrB,GA7BqC,IA8BrC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,IAG9C,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,SAAW,GAAK,EAAM,KAAO,EAAM,GAC3C,KAAM,IAAI,GACN,wEAGJ,MAAO,GAAA,IAAI,EAAK,KAAM,EAAA,IAAI,EAAM,QAKtC,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OApBd,EAAA,UAAY,WAsBrB,GAxB8B,IAyB9B,EAAA,cAAc,cAAc,IAU5B,YACI,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,gBAChB,GAAI,GACA,EAEJ,GADA,GAAgB,GACZ,EAAM,SAAW,EACnB,EAAQ,EAAM,GACd,EAAS,EAAM,WACN,CAAC,EAAG,EAAG,GAAG,QAAQ,EAAM,UAAY,IAC7C,GAAI,IAAe,iBACjB,GAAM,GAAqB,GAAU,EAAO,GAC5C,EAAQ,EAAM,GAAK,EACnB,EAAS,EAAM,GAAK,UACX,IAAe,gBACxB,GAAM,GAAqB,GAAU,EAAO,EAAG,EAAM,OAAS,GAC9D,EAAQ,EAAM,EAAM,OAAS,GAAK,EAClC,EAAS,EAAM,EAAM,OAAS,GAAK,QAGrC,GAAM,GAAY,GAAU,GAC5B,EAAQ,KAAK,KAAK,GAClB,EAAS,KAAK,KAAK,GAGrB,MAAO,CAAC,EAAO,GAiBjB,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAYnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,EAAK,MAAQ,EACf,KAAM,IAAI,GACN,wCAAwC,EAAK,OAEnD,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAM,EAAK,MAC7C,EAAK,KAAO,EAAK,MAAQ,KAAO,QAAU,EAAK,KAC/C,GAAa,EAAK,MAClB,EAAK,aACD,EAAK,cAAgB,KAAO,SAAW,EAAK,aAChD,GAAkB,EAAK,cACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAO,GAAY,GACnB,EAAQ,EAAK,GACb,EAAS,EAAK,GAChB,EAAQ,KAAK,MASjB,GARA,AAAI,KAAK,OAAS,QAChB,GAAS,KAAK,IAAI,EAAG,GAChB,AAAI,KAAK,OAAS,SACvB,GAAS,KAAK,IAAI,EAAG,GAErB,GAAS,KAAK,IAAI,EAAI,GAAQ,GAAU,GAGtC,KAAK,eAAiB,UACxB,GAAM,GAAS,KAAK,KAAK,GAEzB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACH,KAAK,eAAc,2BAA2B,EAAK,KAE5D,MAAO,GAAA,gBAAgB,EAAO,EAAG,EAAQ,EAAO,KAAK,WAErD,GAAM,GAAQ,KAAK,KAAK,EAAI,GAC5B,MAAO,GAAA,cAAc,EAAO,CAAC,EAAO,EAAO,KAI/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,MACZ,KAAM,KAAK,KACX,aAAc,KAAK,aACnB,KAAM,KAAK,OAzDR,EAAA,UAAY,kBA4DrB,GA9DqC,IA+DrC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,gBAwBrB,GA1BmC,IA2BnC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAWhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,eAwBrB,GA1BkC,IA2BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,WAiBrB,GAnB8B,IAoB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAI7B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,YAiBrB,GAnB+B,IAoB/B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAI/B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBiC,IAoBjC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAIhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBkC,IAoBlC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAO9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GATO,EAAA,aAAe,EAMtB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,KAAO,EAAK,KAEb,EAAK,MAAQ,KACf,KAAM,IAAI,IACN,2EAIR,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,IAAoB,8BAEhC,AAAI,EAAM,GAAK,EAAM,GAAK,KACxB,QAAQ,KACJ,gEACA,eAAc,EAAM,GAAK,EAAM,GAAE,gBACjC,wBAIN,GAAM,GACF,EAAM,GAAK,EAAM,GAAK,CAAC,EAAM,GAAI,EAAM,IAAM,EAC3C,EAAI,GAAe,EAAiB,EAAG,EAAG,WAC5C,EAAI,EAAA,OAAO,YAAY,GAC3B,MAAI,GAAM,GAAK,EAAM,IACnB,GAAI,EAAE,aAED,EAAA,IAAI,EAAK,KAAM,MAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CACL,KAAM,KAAK,KACX,KAAM,KAAK,OA3CR,EAAA,UAAY,aA8CrB,GAhDgC,IAiDhC,EAAA,cAAc,cAAc,IAUrB,GAAM,IACyC,CAChD,SAAY,WACZ,aAAgB,eAChB,cAAiB,gBACjB,SAAY,WACZ,UAAa,YACb,SAAY,WACZ,YAAe,cACf,aAAgB,eAChB,KAAQ,OACR,WAAc,aACd,aAAgB,eAChB,cAAiB,gBACjB,gBAAmB,kBACnB,gBAAmB,kBACnB,MAAS,SAGf,YACI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGgB,GAEnC,MAAO,IAAqB,eAGC,GAE7B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EAIJ,GAAI,IAAc,eAChB,MAAO,IAAI,IACN,GAAI,IAAc,gBACvB,MAAO,IAAI,IACN,GAAI,IAAc,WACvB,MAAO,IAAI,IACN,GAAI,IAAc,YACvB,MAAO,IAAI,IACN,GAAI,IAAc,cACvB,MAAO,IAAI,IACN,GAAI,IAAc,eACvB,MAAO,IAAI,IAEX,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAuB,OAE3B,OAAI,aAAsB,IACxB,EAEA,GAAuB,GCxoBlC,cAkBE,MAAO,IAAI,kBASX,MAAO,IAAI,gBAQY,GACvB,MAAO,IAAI,IAAS,eAYQ,GAC5B,MAAO,IAAI,IAAc,eASE,GAC3B,MAAO,IAAI,IAAa,eAaM,GAC9B,MAAO,IAAI,IAAgB,eASJ,GACvB,MAAO,IAAI,IAAS,eAgBU,GAC9B,MAAO,IAAI,IAAgB,eAgBC,GAC5B,MAAO,IAAI,IAAc,eAgBE,GAC3B,MAAO,IAAI,IAAa,eAeD,GACvB,MAAO,IAAI,IAAS,eAeI,GACxB,MAAO,IAAI,IAAU,eAgBK,GAC1B,MAAO,IAAI,IAAY,eAYI,GAC3B,MAAO,IAAI,IAAa,eAWC,GACzB,MAAO,IAAI,IAAW,kPChNxB,AAmBA,GAAI,IAAsB,gBAGxB,MAAO,MAGT,GAAM,IAA2C,eAO1B,GAAA,MAAA,KAAA,QAAA,GAAA,IACrB,AAAM,IAAU,KACd,IAAa,GAAU,GAEzB,GAAa,IAAW,EACjB,EAAS,GAAa,GAAQ,WCrCvC,YAoBgC,GAC9B,MAAO,OAAM,QAAQ,IAAM,MAAM,QAAQ,EAAE,gBASV,GACjC,MAAI,GAAE,SAAW,EACR,GAEJ,MAAM,QAAQ,EAAE,IAGd,EAFE,CAAC,eAWwB,GAClC,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAG,SAAW,EAChB,KAAM,IAAI,GAAW,uCAAuC,EAAG,QAEjE,EAAI,EAAG,OAEP,GAAI,EAEN,MAAO,eAa0B,GACjC,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,EAAO,KAChD,GAAI,EAAO,SAAW,EACpB,SAAS,EACF,EAAO,GAEd,KAAM,IAAI,GAAW,iCAAiC,EAAO,YAG/D,OAAO,GC9EX,YAmBqC,GAEnC,OADI,GAAQ,EACS,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,AAAI,EAAO,MAAM,SAAW,EAC1B,GAAS,EAET,GAAS,EAAO,MAAM,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAG/C,MAAO,GC5BT,AAoBA,GAAM,IAA+B,yBAqCnC,WACI,EAAa,EACb,EAAqC,EACrC,GAFa,AAAA,IAAA,QAAA,GAAA,WACb,AAAA,IAAA,QAAA,GAAA,IAAqC,AAAA,IAAA,QAAA,GAAA,IACrC,AAAA,IAAA,QAAA,GAAA,MACF,KAAK,MAAQ,GAAgB,UAC7B,KAAK,MAAQ,EAAI,MACjB,KAAK,GAAK,KAEV,EAAO,GAAe,GACtB,KAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,cAErC,KAAK,WAAa,EAClB,KAAK,WAAa,EAElB,KAAK,IAAM,EAAA,SAAa,EAAK,KAAK,WAAY,KAAK,KAAM,KAAK,OAUhE,SAAA,UAAA,KAAA,WACE,YAAK,oBACE,KAAK,KAUd,EAAA,UAAA,MAAA,SAAM,GAEJ,YAAK,oBACL,GAAiB,KAAK,IAAK,GAE3B,AAAI,KAAK,IAAI,KAAO,EAAO,IACzB,MAAK,IAAI,OAAO,GAChB,AAAI,KAAK,YAAc,MACrB,KAAK,IAAI,OAAO,KAAK,WAAW,MAAM,KAAK,OAGxC,MAMT,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,KAAK,IAAI,WAGD,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,IAAI,WACX,KAAM,IAAI,OAAM,kBAAkB,KAAK,KAAI,0BAI/C,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,WAAa,EAClB,KAAK,IAAI,UAAY,mCAEzB,KAEA,YAA0B,EAAa,GACrC,GAAI,EAAE,MAAM,aAAe,EAAE,MAAM,WACjC,KAAM,IAAI,OACN,mBAAqB,KAAK,UAAU,EAAE,OAAS,QAC/C,KAAK,UAAU,EAAE,oBA0LK,GAC5B,MAAO,GAAG,IAAI,SAAA,GAAK,MAAA,GAAE,qBAYnB,GACF,EAAmB,QAAQ,SAAA,GACzB,GAAM,GAA0B,EAAiB,GACjD,EAAS,MAAM,EAAiB,MClVpC,kBAuEE,WAAY,GACV,KAAK,MAAQ,EAAK,MAClB,KAAK,MAAQ,EAAK,MAKlB,AAAI,EAAK,OAAS,KAChB,KAAK,KAAO,EAAK,MAAM,OAEvB,KAAK,KAAO,EAAK,KAEnB,KAAK,QAAU,EAAK,QACpB,KAAK,QAAU,EAAK,QACpB,KAAK,KAAO,EAAK,MAAQ,GAE7B,MAAA,oBA6CE,WACa,EAA0B,EAC5B,EAA6B,EAC3B,EAAkB,EAClB,GAHA,KAAA,MAAA,EAA0B,KAAA,MAAA,EAC5B,KAAA,YAAA,EAA6B,KAAA,OAAA,EAC3B,KAAA,SAAA,EACA,KAAA,kBAAA,EACX,KAAK,GAAK,KACV,AAAI,GAAQ,MACV,MAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,eAEvC,KAAK,KAAO,EAAM,OAEtB,MAAA,MA2DI,GAAc,EAsBlB,GAAA,WAwCE,WACI,EAEO,GAAA,KAAA,SAAA,EACT,KAAK,GAAK,KAQV,KAAK,cAAgB,EAAK,cAU1B,KAAK,cAAgB,EAAK,cAE1B,KAAK,YAAc,EAAK,YAExB,KAAK,cAAgB,EAAK,cAQ1B,KAAK,aAAe,EAAK,aAEzB,KAAK,cAAgB,EAAK,cAM1B,KAAK,WAAa,EAAK,WAEvB,KAAK,YAAc,EAAK,YAKxB,KAAK,YAAc,EAAK,YAExB,KAAK,aAAe,EAAK,aAGzB,OAAoB,GAAA,EAAA,EAAA,EAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MACX,EAAM,cAAc,KAAK,MAG7B,EAAK,cAAc,aAAa,KAAK,MAGvC,SAAA,UAAA,UAAA,WAEE,OADM,GAAyB,GACX,EAAA,EAAA,EAAA,KAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,KACX,EAAa,KAAK,EAAM,MAExB,EAAa,KAAK,MAGtB,MAAO,CACL,cAAe,KAAK,cAAgB,KAAK,cAAc,KAAO,KAC9D,cAAe,EACf,YAAa,KAAK,YAClB,cAAe,KAAK,gBAG1B,KAkDI,GAAe,EAWnB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAmDlC,WAAY,GAAA,AAAA,IAAA,QAAA,GAAA,IAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAtBD,EAAA,UAAsB,KAEtB,EAAA,kBAA8B,GAQ5B,EAAA,UAAY,GAapB,EAAK,GAAK,KAEV,EAAK,oBAAsB,KAE3B,EAAK,UAAY,KACjB,EAAK,gBAAkB,GAGvB,EAAK,kBAAoB,GACzB,EAAK,qBAAuB,GAC5B,EAAK,QAAU,GACf,EAAK,SAAW,GAChB,EAAK,OAAS,GAMd,EAAK,aAAe,GACpB,EAAK,cAAgB,GAErB,GAAI,GAAO,EAAK,KAChB,GAAI,CAAC,GACH,GAAM,GAAS,EAAK,eACpB,EAAO,GAA0B,GAAU,IAAM,GAAO,GAM1D,GAJA,EAAK,KAAO,EAEZ,EAAK,WAAa,EAAK,WAAa,KAAO,GAAO,EAAK,UAEnD,EAAK,YAAc,MAAQ,EAAK,iBAAmB,MAKrD,GAAI,GAAe,OACnB,GAAI,EAAK,iBAAmB,KAC1B,EAAkB,EAAK,wBACd,EAAK,YAAc,MAC5B,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAkB,CAAC,GAAW,OAAO,EAAK,YAE5C,EAAK,gBAAkB,EAGvB,GAAI,GAAQ,EAAK,MACjB,AAAI,GAAS,MACX,GAAQ,EAAK,YAEf,AAAI,GAAS,MACX,GAAQ,WAEV,EAAK,MAAQ,EAGf,MAAI,GAAK,SAAW,KAClB,EAAK,eAAiB,EAAK,QAE3B,EAAK,eAAiB,KAKxB,EAAK,UAAY,KAEjB,EAAK,0BAA4B,KAYlB,SAAA,QAAjB,SAAyB,EAAc,GACrC,MAAO,GAAM,KAAO,OAAS,EAAU,YAUjC,EAAA,UAAA,eAAR,SAAuB,EAAmB,GACxC,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,mCACA,4BAA2B,EAAQ,MAEzC,GAAI,KAAK,aAAa,QAAU,EAC9B,KAAM,IAAI,GACN,gBAAgB,EAAQ,YAAY,EAAS,KAC7C,2BAA0B,KAAK,aAAa,OAAM,oBAExD,MAAO,MAAK,aAAa,IAY3B,EAAA,UAAA,WAAA,SAAW,GACT,MAAO,IACH,KAAK,eAAe,EAAW,SAAS,eAY9C,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,IACH,KAAK,eAAe,EAAW,UAAU,gBAgB/C,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,uHAIC,GAAI,KAAK,aAAa,SAAW,EACtC,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0CAEN,MAAO,IACH,KAAK,eAAe,EAAG,SAAS,+CActC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0BAEN,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,yHAKN,MAAO,IACH,KAAK,eAAe,EAAG,UAAU,gDAGvC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAQd,EAAA,UAAA,gBAAA,WAKE,MAAO,MAAK,OAAO,IAAI,SAAA,GAAU,MAAA,QAGnC,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,0CAGd,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,MAAO,MAAK,YAGd,SAAU,GACR,KAAK,OAAS,mCAGhB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,kBAAkB,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,IAClD,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAI,MAAK,WACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,GAAE,YAErC,QAIX,SAAqB,GACnB,KAAK,kBAAoB,mCAG3B,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAI,MAAK,UACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,CAAC,EAAE,YACxC,OAAO,KAAK,sBAEV,KAAK,kBAAkB,OAAO,KAAK,2BAI9C,SAAwB,GACtB,KAAK,qBAAuB,mCAO9B,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAG3C,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,2CAUd,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,SACR,KAAM,IAAI,OACN,yEAiBE,EAAA,UAAA,yBAAV,SAAmC,GAGjC,GADA,EAAS,GAAqB,GAC1B,KAAK,WAAa,MAAQ,KAAK,UAAU,SAAW,EACtD,OAEF,GAAM,GAAY,GAAqB,KAAK,WAC5C,GAAI,EAAO,SAAW,EAAU,OAC9B,KAAM,IAAI,GACN,SAAS,KAAK,KAAI,YAAY,EAAU,OAAM,YAC9C,oBAAmB,EAAO,OAAM,oBAChC,oBAAmB,IAEzB,OAAS,GAAa,EAAG,EAAa,EAAO,OAAQ,KACnD,GAAM,GAAI,EAAO,GACX,EAAkB,EAAU,GAClC,GAAI,GAAQ,KACV,SAIF,GAAM,GAAO,EAAE,KACf,GAAI,EAAK,MAAQ,MACX,IAAS,EAAK,KAChB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,KAC3D,kBAAiB,EAAK,KAAI,gBAAgB,IAGlD,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,IAG3D,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,EAAI,MAK/D,GAAI,EAAK,OAAS,MACZ,EAAE,QAAU,EAAK,MACnB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,IAC3D,qBAAoB,EAAK,MAAK,iBAAiB,EAAE,MAAK,MAK9D,GAAI,EAAK,MACP,GAAM,GAAS,EAAE,MACjB,OAAW,KAAO,GAAK,MACrB,GAAM,GAAO,OAAO,GACd,EAAQ,EAAK,KAAK,GAIlB,EACF,GAAQ,EAAI,EAAO,GAAQ,EAAO,EAAO,OAAS,GACtD,GAAI,GAAS,MAAQ,CAAC,EAAO,MAAM,QAAQ,KAAkB,GAC3D,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,mBAAmB,EAAI,uBACnC,eAAc,EAAK,kBAAkB,EAAM,OAMrD,GAAI,EAAK,OAAS,KAChB,OAAS,GAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAU,EAAK,MAAM,GACrB,EAAM,EAAE,MAAM,GACpB,GAAI,GAAW,MAAQ,GAAO,MACxB,IAAY,EACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,oBAAoB,EAAK,MAAK,MAC1C,gBAAe,EAAE,MAAK,SAgBtC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,IAGC,EAAA,UAAA,eAAV,SAAyB,EAAyB,GAChD,AAAI,KAAK,WAAa,MACpB,KAAK,UAAU,EAAQ,IAS3B,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,UAAY,GAOnB,EAAA,UAAA,cAAA,WACE,KAAK,UAAY,MAwEnB,EAAA,UAAA,MAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAGE,EAAS,GAAU,GAEnB,KAAK,oBAML,OAHM,GAAa,GAAqB,GAEpC,EAAiB,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,CAAE,aAAiB,MACrB,EAAiB,GACjB,OAIJ,OADI,GAAkB,GACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,YAAiB,KACnB,EAAkB,GAClB,OAIJ,GAAI,IAAmB,EACrB,KAAM,IAAI,GACN,mEAKN,MAAO,IAAU,KAAK,KAAM,WAE1B,GAAI,CAAC,EAAK,OAKR,EAAK,yBAAyB,GAI9B,OADM,GAAuB,GACT,EAAA,EAAA,EAAA,GAAqB,GAArB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,KAAK,EAAM,OAEzB,EAAK,MAAM,GAA+B,IAC1C,EAAK,MAAQ,GAGb,AAAI,EAAK,gBACP,EAAK,WAAW,EAAK,gBAGvB,AAAI,EAAK,YAAc,MAAQ,GAI7B,GAAK,UAAY,GAcrB,GANA,EAAK,yBAAyB,GAM1B,GAUF,OATI,GAAS,EAAK,KAAK,EAA6B,GAK9C,EAAuB,GAAqB,GAC5C,EAA2B,GAGnB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAT,GAAI,GAAC,EAAA,GACR,AAAI,EAAW,QAAQ,KAAO,IAC5B,GAAI,EAAE,SAER,EAAe,KAAK,GAItB,GAFA,EAAS,GAA+B,GAEpC,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAKN,MAAO,QAEP,GAAM,GAAa,GAAkB,GAC/B,EAAc,EAAK,mBAAmB,GACxC,EAAM,OACJ,EAAc,GAAiB,GAgCrC,GA/BA,EAAK,6BACD,MAAM,QAAQ,GAAU,EAAW,GACX,GAE5B,AAAI,GAAe,MAAQ,EAAY,OAAS,GAC5C,MAAM,QAAQ,EAAY,IAE5B,EAAU,EACI,IACG,SAAC,EAAO,GAAU,MAAA,IAAI,IAClB,EAAa,EAAO,EACpB,GAAqB,GAAS,EAAQ,EAAK,KAC3C,KAErB,EAAS,GAAI,IACT,EAAa,EAAsB,EACnC,GAAqB,GAAS,EAAQ,EAAK,MAUjD,EAAK,eACD,EAA6C,EAAQ,KAAM,KAC3D,EAAY,EAAa,GAC7B,EAAK,YAED,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAIN,MAAO,OAYH,EAAA,UAAA,6BAAV,SAAuC,GACrC,GAAI,KAAK,iBAAmB,KAC1B,OACK,GAAI,EAAW,SAAW,KAAK,gBAAgB,OACpD,QAAQ,KACJ,iDACG,MAAK,UAAU,GAAW,iCAC7B,qBAAoB,KAAK,UAAU,KAAK,iBAAgB,MACxD,iBAAgB,KAAK,YAEzB,GAAI,GAAc,GAClB,KAAK,gBAAgB,QAAQ,SAAC,EAAW,GACvC,AAAI,GAAa,MAAQ,EAAW,IAAM,MACtC,EAAW,KAAO,GACpB,GAAc,MAGlB,AAAI,GACF,QAAQ,KACJ,iCACA,KAAI,KAAK,UAAU,GAAW,eAC9B,mCAAkC,KAAK,KAAI,MAC3C,IAAG,KAAK,UAAU,KAAK,qBAiBjC,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,GAAI,KAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,EAC5D,KAAM,IAAI,IACN,aAAa,KAAK,KAAI,gEAI5B,OADM,GAA4B,GACf,EAAA,EAAA,EAAA,KAAK,aAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAc,KAAK,UAAU,EAAK,cACxC,AAAI,EAAgB,QAAQ,KAAiB,IAC3C,EAAgB,KAAK,GAGzB,GAAI,EAAgB,SAAW,GAC7B,GAAM,GAAe,KAAK,aAAa,GAAG,aAC1C,MAAI,OAAM,QAAQ,IAAiB,MAAM,QAAQ,EAAa,KAC1D,EAAa,SAAW,EAClB,EAAyB,GAE1B,MAIT,MAAM,IAAI,IACN,aAAa,KAAK,KAAI,+JAiB9B,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,sCAAsC,KAAK,KAAI,uFAIrD,MAAO,IAAoC,KAAK,UAclD,EAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAWf,EAAA,UAAA,WAAA,SAAW,GAAA,MAAA,KAAA,QAAA,GAAA,IACF,GAAc,EAAgB,KAAK,iBAAmB,KAAK,UAepE,EAAA,UAAA,WAAA,SAAW,GAAX,GAAA,GAAA,KACE,EAAA,KAAK,WACH,GAAM,GAAS,EAAK,QACpB,GAAI,EAAO,SAAW,EAAQ,OAK5B,KAAM,IAAI,GACN,4CAA4C,EAAK,KAAI,KACrD,iCAAgC,EAAQ,OAAM,MAC9C,gCAA+B,EAAO,OAAM,cAC5C,sBAAqB,EAAO,QAElC,GAAI,EAAO,SAAW,EACpB,OAIF,OAFM,GAAoD,GACpD,EAAc,GAAc,GACzB,EAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,GACxC,GAAM,GAAK,EAAY,GACjB,EAAI,EAAO,GACX,EAAI,EAAQ,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,GACN,sBAAsB,EAAG,MAAK,IAC9B,8CAA6C,EAAE,QAErD,EAAkB,KAAK,CAAC,EAAG,IAE7B,GAAc,MAmBR,EAAA,UAAA,UAAV,SACI,EAAc,EAAc,EAAkB,EAC9C,EAA2B,EAC3B,GAEF,GAAI,KAAK,kBAAkB,QAAQ,KAAU,GAC3C,KAAM,IAAI,GACN,yBAAyB,EAAI,cAAc,KAAK,MAEtD,KAAK,kBAAkB,KAAK,GAE5B,AAAI,GAAS,MACX,GAAQ,WAGV,AAAI,KAAK,2BACP,GAAc,GAAe,UAE/B,GAAM,GAAY,EAAY,MAAM,EAAO,GACrC,EACF,GAAI,IAAc,EAAW,EAAO,EAAM,EAAW,GACzD,SAAU,UAEV,AAAI,GAAe,MACjB,KAAK,QAAQ,WAAM,MAAA,GAAY,MAAM,EAAO,UAE9C,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,EACF,KAAK,kBAAkB,KAAK,GAE5B,KAAK,qBAAqB,KAAK,GAE1B,GAaT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,KAAK,0BAA4B,GAWnC,EAAA,UAAA,QAAA,SAAQ,SACN,GAAI,GAAU,MAAQ,MAAM,QAAQ,IAAW,EAAO,SAAW,EAC/D,OAGF,EAAS,GAAqB,GAC9B,AAAI,KAAK,UAAY,QAAa,KAAK,UAAY,MACjD,GAAA,KAAK,QAAO,KAAI,MAAA,EAAI,IAexB,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAYT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,GAAI,CAAC,KAAK,iBACR,GAAI,GAAQ,KACV,GAAI,MAAM,QAAQ,GAChB,EAAK,QAAQ,SAAA,GACX,GAAI,GAAe,KACjB,KAAM,IAAI,WACN,SAAS,EAAK,KAAI,iEAK1B,MAAM,IAAI,WACN,SAAS,KAAK,KAAI,2DAK1B,MAAO,MAIT,MAAO,IAeD,EAAA,UAAA,eAAR,SACI,EACA,EACA,EAA6B,EAC7B,EAA4B,EAC5B,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAM,GACF,GAAqB,GACzB,EAAgB,GAAqB,GACrC,EAAa,GAAqB,GAClC,EAAc,GAAqB,GACnC,EAAc,GAA+B,GAC7C,EAAe,GAA+B,GAM9C,OAHM,GAAyB,GACzB,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GAKV,EAAc,KAAK,EAAE,aACrB,EAAY,KAAK,EAAE,WACnB,EAAc,KAAK,EAAE,aAMvB,GAAI,IACA,CACE,cAAe,KACf,cAAa,EACb,YAAW,EACX,cAAa,EACb,aAAc,EACd,cAAa,EACb,WAAU,EACV,YAAW,EACX,YAAW,EACX,aAAY,GAEd,GAGJ,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,IAExC,EAAc,GAAG,YAAc,KAC/B,EAAc,GAAG,UAAY,KAAK,aAAa,OAAS,EACxD,EAAc,GAAG,YAAc,GAyBnC,EAAA,UAAA,UAAA,WACE,GAAM,GACyB,CAAC,KAAM,KAAK,KAAM,UAAW,KAAK,WACjE,MAAI,MAAK,iBAAmB,MAC1B,GAAO,gBAAqB,KAAK,iBAEnC,AAAI,KAAK,OAAS,MAChB,GAAO,MAAW,KAAK,OAElB,GAQC,EAAA,UAAA,eAAV,WACE,YAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,YAC/B,KAAK,QAAQ,QAGZ,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,UAAU,KAAK,KAAI,2BAkCvC,EAAA,UAAA,QAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,uCAIvC,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,sCAIvC,KAAK,oBAEL,GAAI,GAAuB,EAC3B,MAAI,EAAE,KAAK,YAAc,GACvB,GAAuB,KAAK,kBAGvB,CAAC,qBAAsB,KAAK,UAAW,qBAAoB,IAEtE,GAxoCoC,EAAA,cAAc,cAopClD,YAA2B,GAEzB,EACI,GAAqB,GAEzB,OADM,GAAkB,GACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAO,KAAK,EAAE,OAEhB,MAAO,IAA+B,GAYxC,YAA0B,GAExB,MAAO,sBAeL,EAAwB,EACxB,GAKF,GAJA,AAAI,IAAS,MAAS,GAAa,MAAQ,EAAY,IACrD,GAAQ,EAAO,YACf,EAAY,EAAO,WAEjB,EAAM,aAAa,SAAW,EAChC,MAAO,CAAC,GAER,GAAM,GAAO,EAAM,aAAa,GAChC,GAAI,EAAK,cAAc,SAAW,EAChC,MAAO,GAAK,aAGZ,OADM,GAAkC,GAC/B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,IAM7C,OALM,GAAI,EAAK,aAAa,GACtB,EAAQ,EAAK,cAAc,GAC3B,EAAY,EAAK,YAAY,GAC7B,EAAkB,GAAgB,EAAG,EAAO,GAElC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAc,QAAQ,KAAO,IAC/B,EAAc,KAAK,GAIzB,MAAO,GCvmDb,AA4CA,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CACJ,MAAO,EAAK,MACZ,KAAM,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAO,SAAS,cACtD,KAaF,GAXA,AAAI,EAAK,WAAa,MACpB,GAAK,UAAY,MAEnB,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,IAGhB,EAAK,UAAY,GACjB,EAAK,MAAQ,GACb,EAAK,OAAS,EAAK,OAEf,EAAK,YAAc,MAAQ,EAAK,iBAAmB,KACrD,KAAM,IAAI,GACN,qGAGN,GAAI,GAAkB,EAAK,gBAC3B,GAAI,GAAmB,MACrB,GAAI,EAAK,YAAc,KACrB,KAAM,IAAI,GACN,iFAGJ,EAAkB,CAAC,EAAK,WAAW,OAAO,EAAK,oBAI7C,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,yFAKR,GAAM,GAAQ,EAAK,OAAS,UAE5B,EAAK,gBAAkB,EACvB,EAAK,MAAQ,EAEb,EAAK,UAAY,CAAC,CAAC,MAAO,IAE1B,GAAM,GAAc,GAAI,IACpB,EAAK,MAAO,EAAK,gBAAiB,EAAM,GAAI,GAAI,EAAK,MACzD,SAAY,UAAY,EACxB,EAAY,YAAc,EAK1B,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,CAAC,GACf,cAAe,CAAC,GAChB,WAAY,CAAC,MACb,YAAa,CAAC,MACd,YAAa,CAAC,GACd,aAAc,CAAC,OAInB,SAAA,UAAA,MAAA,SACI,EACA,GACF,KAAM,IAAI,GACN,+BACA,kDAAiD,KAAK,QAG5D,EAAA,UAAA,QAAA,WAEE,MAAO,CAAC,qBAAsB,KAAK,UAAW,qBAAsB,IAGtE,EAAA,UAAA,UAAA,WACE,MAAO,CACL,gBAAiB,KAAK,gBACtB,MAAO,KAAK,MACZ,OAAQ,KAAK,OACb,KAAM,KAAK,OAzFC,EAAA,UAAY,aA4F9B,GA9FgC,IA+FhC,EAAA,cAAc,cAAc,gBAmCN,GACpB,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAC/C,KAAM,IAAI,OACN,gIAKN,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAE/C,KAAM,IAAI,GACN,oFAGN,GAAI,GAAa,EAAO,WACxB,AAAI,EAAO,OAAS,MAAQ,GAAc,MACxC,GAAa,CAAC,MAAM,OAAO,EAAO,QAGpC,GAAI,GAAQ,EAAO,MACnB,AAAI,GAAS,MACX,GAAQ,WAGV,GAAM,GAAa,GAAI,IAAW,CAChC,gBAAiB,EACjB,KAAM,EAAO,KACb,MAAK,EACL,OAAQ,EAAO,SAGX,EAAU,EAAW,aAAa,GAAG,cAC3C,MAAO,GAAQ,GC9MjB,YA0B2C,iHACzC,GAAI,GAAQ,KACV,MAAA,CAAA,GAEI,EAA+D,GAC/D,EAAiB,GACjB,EAA6B,GACnC,IAAW,IAAO,GACV,EAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACb,GAAc,EACpB,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAiB,KAAK,UAGtB,GAAS,OAAS,EACL,CAAA,EAAM,QAAQ,IAAI,IAD/B,CAAA,EAAA,UAEF,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAK,EAAK,IAAM,EAAO,GAAG,GAG5B,EAAA,QAAQ,+CAUyB,GACnC,GAAI,GAAQ,KACV,OAEF,OAAW,KAAO,IAChB,GAAM,GAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACnB,EAAM,WCjEZ,AAoBA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,QAAA,GAAA,YAFU,IAAA,IAAqB,KAM1B,GAAM,IAAyB,IA0BtC,GAAA,WAAA,aAEE,KAAA,eAAkC,KAMlC,SAAA,UAAA,UAAA,SAAU,GACR,KAAK,OAAS,GAGV,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,qFAEb,EAAA,UAAA,WAAN,SAAiB,qFASjB,EAAA,UAAA,SAAA,SAAS,KAGX,mBAqBE,WAAY,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,IAGtC,AAAI,GAAa,MACf,GAAY,IAEd,KAAK,UAAY,EACjB,KAAK,YAAc,EAGrB,SAAA,UAAA,OAAA,SAAO,GACL,KAAK,UAAU,KAAK,IAGtB,EAAA,UAAA,UAAA,SAAU,GACR,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,UAAU,KAIvB,EAAA,UAAA,SAAA,SAAS,GACP,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,SAAS,KAShB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,aAAN,SAAmB,uGACjB,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,KADO,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,WAAN,SAAiB,uGACf,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,KADS,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAI3B,KAOA,GAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,mBACE,GAAA,KAAA,OAAO,KAGH,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,OAAS,YAGV,EAAA,UAAA,WAAN,SAAiB,EAAe,yFAC9B,AAAI,GAAQ,MACV,GAAO,IAEH,EAAY,EAAK,MAAW,KAAO,EAAI,EAAK,KAClD,KAAK,MAAQ,aACF,GACT,GAAM,GAAQ,EAAK,GACnB,GAAI,MAAO,IAAU,SACnB,AAAK,EAAK,OAAO,eAAe,IAC9B,GAAK,OAAO,GAAO,GAErB,EAAK,OAAO,GAAO,EAAK,OAAO,GAAiB,EAAQ,OAExD,GAAI,GAAkB,OACtB,AAAI,IAAO,GAAK,OACd,EAAqB,EAAK,OAAO,GAEjC,EAAK,OAAO,GAAO,EAErB,GAAM,GACF,EAAA,KAAK,WAAM,MAAA,GAAA,IAAK,EAAK,OAAO,GAAO,EAAA,IAAI,EAAO,MAClD,EAAK,OAAO,GAAO,EACnB,AAAI,GAAsB,MACxB,EAAmB,mBAlBzB,IAAW,IAAO,KAAP,kBAwBP,EAAA,UAAA,WAAN,SAAiB,EAAe,2FAC9B,GAAI,GAAQ,KACV,eAAW,GACT,GAAI,EAAK,OAAO,IAAQ,sBAGxB,AAAI,MAAO,GAAK,OAAO,IAAS,SAC9B,EAAK,GAAO,EAAK,OAAO,GAAiB,EAAK,KAE9C,EAAA,KAAK,WACH,GAAM,GAAc,EAAA,IAAI,EAAA,IAAI,EAAG,EAAK,MAAO,EAAK,OAAO,IACvD,EAAK,GAAO,EACX,EAAK,OAAO,GAAgB,UAC7B,EAAA,KAAK,EAAK,cAXhB,EAAA,EAAkB,EAAA,KAAK,OAAO,QAAZ,EAAA,EAAA,OAAA,IAAP,EAAG,EAAA,KAAH,kBAiBjB,GA9DgC,mBAqEH,EAAA,EAAA,GAA7B,4DAIQ,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,MAAQ,GACb,KAAK,QAAU,YAGX,EAAA,UAAA,WAAN,SAAiB,EAAe,4EAC9B,AAAI,GAAQ,MACV,GAAO,IAET,KAAK,MAAM,KAAK,GAChB,IAAW,IAAO,GAChB,AAAI,KAAK,QAAQ,IAAQ,MACvB,MAAK,QAAQ,GAAO,IAEtB,KAAK,QAAQ,GAAK,KAAK,EAAK,mBAO1B,EAAA,UAAA,SAAN,6HACQ,EAA+D,GAC/D,EAAiB,GACjB,EAAoB,GAC1B,IAAW,IAAO,MAAK,QAErB,IADM,EAAa,KAAK,QAAQ,GACvB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,MAAO,GAAW,IAAO,UACrB,GAAc,EAAW,GAC/B,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAQ,KAAK,IAIJ,MAAA,CAAA,EAAM,QAAQ,IAAI,WACjC,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EAC7B,EAAkB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IACtD,EAAgB,UAChB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IAAM,EAAO,GAAG,mBAGpD,GA/C6B,mBA8DO,EAAA,EAAA,GAiBlC,WAAY,EAA0B,GAAtC,GAAA,GACE,EAAA,KAAA,OAAO,KAKP,GARM,EAAA,aAAe,EAIrB,EAAK,WAAa,GAAc,OAChC,AAAI,EAAK,aAAe,QACtB,GAAK,WAAa,IAEhB,EAAK,aAAe,SAAW,EAAK,SAAW,KACjD,KAAM,IAAI,OACN,mHAGN,MAAI,GAAA,KAAK,SAAS,EAAK,aAGrB,GAAK,UAAY,GACb,EAAK,UAAU,KAAK,GAAO,EAAK,aAEtC,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,MAAQ,EAAK,UAGd,SAAA,UAAA,UAAN,SAAgB,EAAe,EAAe,yGACtC,GAAgC,GAClC,KAAK,OAAS,KAChB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,MAAM,EAAO,EAAO,qBAEnC,SAAG,KAAK,EAAA,aACR,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAChC,MAAK,aAAe,EAChB,KAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,SACtB,EAAG,KAAK,EAAA,aAEV,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAC5B,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,QACtB,EAAG,KAAK,EAAA,aACH,AAAI,EAAA,KAAK,SAAS,KAAK,aAC5B,EAAG,KAAK,KAAK,UAAU,KAAK,aAAc,EAAO,IAEnD,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,mGACb,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,WAAtB,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,mGACX,MAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,SAAS,WAApB,EAAA,wCAGN,GA3GoC,gBAiHhC,EAEA,GAIF,GAHA,AAAI,GAAa,MACf,GAAY,IAEV,YAAqB,IACvB,MAAO,CAAC,GAEV,GAAI,MAAM,QAAQ,IAAc,EAAU,YAAc,IACtD,MAAO,GAGT,GAAM,GACF,GAAqB,GACzB,MAAO,GAAgB,IACnB,SAAA,GAAkB,MAAA,IAAI,IAAe,EAAgB,KAW3D,GAAA,IAAA,WAOE,cAcO,SAAA,4BAAP,SACI,EAAwB,GAC1B,EAAA,KAAK,OACD,GAAkB,GAAK,OAAO,UAAU,GACxC,WAAM,MAAA,sDACF,YAAW,KACnB,EAA4B,kBAAkB,GAC9C,AAAI,EAA4B,aAAa,IAAmB,MAC9D,GAA4B,aAAa,GAAkB,IAE7D,EAA4B,aAAa,GAAgB,KACrD,IAGS,EAAA,kBAAf,SAAiC,GAE/B,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAe,EAA4B,aAAa,CAAC,GAC/D,EAAa,QAAQ,SAAA,GACnB,GAAI,IAAS,EACX,KAAM,IAAI,GAAW,uCASZ,EAAA,MAAjB,WACE,EAA4B,aAAe,IAWtC,EAAA,gBAAP,SAAuB,GACrB,GAAM,GAA0C,GAChD,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAQ,CAAC,EACf,AAAI,GAAkB,GACpB,EAAa,KAAI,MAAjB,EAAqB,EAA4B,aAAa,IAGlE,MAAO,GAAa,IAAI,SAAA,GAAQ,MAAA,IAAI,MArEvB,EAAA,aAC6C,GAsE9D,iBAGI,EAA2B,EAAgC,EAC3D,EAAsB,EAAyB,EAC/C,EAAmB,EACnB,GACF,GAAM,GAAU,GAAI,IACd,EAAe,CACnB,GAAI,YAAiB,GAA4B,gBAAgB,IAEnE,AAAI,GAAa,MACf,EAAgB,KAAI,MAApB,EAAwB,GAE1B,EAAgB,KAAK,GACrB,GAAM,GAAe,GAAI,IAAa,GAMtC,SAAa,UAAU,CACrB,OAAM,EACN,aAAY,EACZ,QAAS,EACT,MAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,QAAS,IAEJ,CAAC,aAAY,EAAE,QAAO,GCnlB/B,YA2BI,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,QAAS,GChC9B,YAwB4B,EAAW,GACrC,MAAO,GAAA,KAAK,WACV,AAAI,EAAE,QAAU,WACd,GAAI,EAAE,OAAO,YAEf,GAAM,GAAY,EAAA,IAAQ,GAAS,GAAI,EAAM,IACvC,EAAgB,EAAA,KAAS,EAAU,MAAO,MAC1C,EAAO,EAAA,KAAS,EAAA,QAAY,EAAW,IAC7C,MAAO,GAAA,IAAQ,EAAG,iBAIW,EAAe,GAC9C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAO,IAAS,kBAG5B,EAAe,GAC/C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAO,IAAS,kBAIzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,EAAA,IAAQ,EAAO,GACtB,EACF,EAAA,YAAgB,EAAA,IAAQ,GAAQ,KAAW,OAAO,WAChD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAM,IACxC,MAAO,GAAA,IAAQ,IAAK,EAAA,KAAS,EAAW,mBAKxC,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAE9B,EAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAErC,MAAO,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAU,IAAa,kBAI/B,EAAe,GAC1C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,GAAS,GAAY,kBAInB,EAAe,GACnC,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,EAAW,kBAIE,EAAe,GAC9C,MAAO,GAAA,KAAK,WACV,GAAM,GAAM,EAAA,IAAQ,EAAA,IAAQ,EAAO,GAAQ,IACrC,EAAM,EAAA,IAAQ,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAQ,GAAQ,IACvD,MAAO,GAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAK,mBAY1B,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,KAAK,IAAI,GAChB,EAAiB,EAAA,IAAQ,EAAO,GAChC,EAAgB,EAAA,IAClB,EAAA,IAAQ,EAAgB,EAAA,SAAa,EAAA,IAAQ,GAAI,KACjD,GACJ,MAAO,GAAA,KAAS,EAAe,kBAK/B,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAI,EACF,EAAS,EAAA,QAAY,QAGrB,GAAM,GAAY,EAAA,IAAQ,EAAQ,EAAO,MAAM,OAAS,EAAG,IAC3D,EAAS,EAAA,IAAQ,EAAQ,GAE3B,SAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MACzC,EAAA,IAAQ,EAAA,IACX,EAAA,IAAQ,EAAO,UAAW,EAAA,IAAQ,IAAU,EAAO,MAAM,OAAS,kBActE,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAM,GAAa,EAAA,MAAU,GAAU,IAAS,QAChD,EAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MAChD,GAAM,GAAc,EAAO,MACrB,EACF,EAAA,OAAW,EAAY,EAAY,EAAY,OAAS,IACnD,QAAQ,GACjB,MAAO,IAAwB,EAAc,EAAQ,iBA0BrD,EAAgB,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAO,MAAO,EAAO,OACzC,KAAM,IAAI,GACN,8DACG,MAAK,UAAU,EAAO,OAAM,QAAQ,KAAK,UAAU,EAAO,SAEnE,MAAO,GAAA,KAAK,WAOV,GAAM,GAAa,EAAO,OACpB,EAAe,EAAO,MAAM,MAClC,MAAO,GAAW,IAAI,EAAO,IAAI,IAAS,IAAI,EAAa,MAAM,uBAIlC,EAAe,GAChD,MAAO,GAAA,KAAK,WACV,GAAI,GACJ,SAAI,EAAA,YAAgB,EAAO,KAAW,EAAI,MAC1C,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAG,KAC3B,EAAA,KAAS,GAA8B,EAAO,GAAI,kBAKzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,GAChD,EAAc,EAAA,YAAgB,EAAO,KAAW,GACtD,MAAO,GAAA,IACH,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAA,IAAQ,EAAa,KAAgB,kBAI5C,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,EAAA,IAAQ,EAAA,IAAQ,KAAW,IAC3C,MAAO,GAAA,KAAS,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAO,IAAW,kBAI7B,EAAe,GAC7C,MAAO,GAAA,KAAK,WACV,GAAM,GAAiB,GAAY,EAAO,IACpC,EAAiB,GAAY,EAAO,IACpC,EAAY,EAAA,IAAQ,EAAgB,GAC1C,MAAO,GAAA,IAAQ,EAAA,IAAQ,EAAW,OAkB/B,GAAM,IAAsD,CACjE,iBAAgB,GAChB,kBAAiB,GACjB,4BAA2B,GAC3B,4BAA2B,GAC3B,aAAY,GACZ,MAAK,GACL,iBAAgB,GAChB,QAAO,GACP,wBAAuB,GACvB,8BAA6B,GAC7B,mBAAkB,GAClB,0BAAyB,GACzB,QAAO,GACP,gBAAe,gBAKG,GAClB,GAAI,MAAO,IAAmB,UAC5B,GAAI,IAAkB,IACpB,MAAO,IAAU,GAEnB,GAAI,GAAS,gBAAgB,EAC7B,KAAI,GAAe,cAAc,SAAS,wBACxC,GAAS,gBAAgB,EAAc,wFAInC,GAAI,GAAW,OAErB,OAAO,GC7QX,YAyB+B,EAAe,GAC5C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,IAAQ,GAAI,EAAA,SAAa,IACrC,EAAmB,GAAO,EAAA,QAAY,EAAO,GAAY,EAAM,OACrE,MAAO,GAAA,KAAS,EAAA,MAAU,EAAO,GAAmB,kBAIpB,EAAe,GACjD,MAAO,GAAA,KACH,WAAM,MAAA,IACF,EAAA,MAAU,EAAA,OAAW,EAAO,IAAK,EAAA,OAAW,EAAO,KAAM,aAGnE,YAAuB,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,yBAI3C,EAAe,GACvC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIS,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIqB,EAAe,GAChD,MAAO,IAAuB,EAAO,eAInC,EAAe,GACjB,MAAI,GAAM,OAAS,EAAM,MACvB,GAAQ,EAAM,QAAQ,CAAC,EAAM,KAAO,KAEtC,EAAQ,EAAM,OAAO,IACrB,AAAI,EAAM,QAAU,EAAM,OACxB,GAAQ,EAAM,OAAO,EAAM,QAEtB,EAAA,MAAU,EAAO,GAAO,OAAO,WAajC,GAAM,IAAM,GACN,GAAM,GACN,GAAM,GACN,GAAM,GACN,GAAO,GACP,GAAO,GACP,GAA0B,GAC1B,GAAS,GACT,GAAgC,GAIhC,GAAuD,CAClE,eAAc,GACd,oBAAmB,GACnB,UAAS,GACT,wBAAuB,GACvB,8BAA6B,GAC7B,IAAG,GACH,IAAG,GACH,IAAG,GACH,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,OAAM,gBAGY,GAClB,GAAI,MAAO,IAAe,UAAY,IAAc,IAClD,MAAO,IAAW,GACb,GAAI,MAAO,IAAe,UAAY,GAAc,KACzD,MAAO,GAEP,KAAM,IAAI,GAAW,kBAAkB,eAqBP,GAElC,GADA,GAAY,IAAO,KAAM,0BAA0B,GAC/C,MAAO,IAAO,SAChB,MAAO,GAGP,OADI,GAAM,OACQ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAU,KAAS,GACrB,EAAS,EACT,OAGJ,GAAI,IAAW,OACb,MAAO,GAET,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAW,KAAS,GACtB,EAAS,EACT,OAGJ,MAAI,KAAW,OACN,EAED,EAAgB,KCzL5B,YAwB6B,GAC3B,GAAM,GAA2D,CAC/D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,MAC/B,SAAY,WAAM,MAAA,GAAA,MAAM,SAAS,EAAG,IAAM,OAC1C,KAAQ,WAAM,MAAA,GAAA,MAAM,KAAK,KAAO,GAAK,KAAO,OAC5C,OAAU,WAAM,MAAA,GAAA,MAAM,OAAO,KAAO,GAAK,KAAO,KAAW,IAC3D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,KAAO,GAAK,EAAG,OAC9C,IAAO,WAAM,MAAA,GAAA,MAAM,IAAI,OASzB,GAPA,EAAa,QAAa,EAAa,QACvC,EAAa,SAAc,EAAa,SACxC,EAAa,KAAU,EAAa,KACpC,EAAa,OAAY,EAAa,OACtC,EAAa,QAAa,EAAa,QACvC,EAAa,IAAS,EAAa,IAE/B,IAAc,GAChB,MAAO,GAAa,KAEtB,KAAM,IAAI,GAAW,qBAAqB,GC3C5C,AAeO,GAAM,IAA8C,EAAI,KAAO,iBAelE,EAAyB,EAAmB,GAC9C,GAD8C,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAuB,MACvB,MAAO,IAAwB,UAC/B,OAAO,eAAe,KAAyB,OAAO,WACtD,CAAC,GAAiB,GACpB,KAAM,IAAI,OACN,sEAGN,GAAI,GACF,GAAM,GAAM,KAAK,UAAU,GAC3B,AAAI,EAAI,OAAS,IACf,QAAQ,KACJ,mCAAmC,EAAS,qBAC5C,iBAAgB,EAAI,OAAM,iCAC1B,kHAEG,IAA2C,mBAgBvB,GAC/B,GAAI,IAAM,KAER,MAAO,GACF,GAAI,MAAO,IAAM,SACtB,GAAI,OAAO,eAAe,KAAO,OAAO,WAGtC,OADM,GAAO,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,MAAO,IAAQ,SAEjB,MAAO,GAET,GAAI,CAAC,GAAiB,EAAE,IACtB,MAAO,GAGX,MAAO,WAGH,MAAM,QAAQ,IAEhB,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,GAAiB,GACpB,MAAO,GAGX,MAAO,OAKP,OAAO,QAKX,GAAM,GAAQ,MAAO,GACrB,MAAO,KAAU,UAAY,IAAU,UAAY,IAAU,WCrGjE,YA8BI,EAAkB,EAAqB,EAEvC,GAAA,AAAA,IAAA,QAAA,GACI,QAAQ,KACd,GAAM,GAAiB,GAAsB,GAGvC,EAAsB,CAAC,eAAgB,eAAgB,WAC7D,AAAI,EACF,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAEtC,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAAM,IAI9C,AAAI,EAAU,EAAU,OAAS,IAAM,GAErC,GAAY,EAAU,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAa,MAGzD,GAAI,GACJ,GAAI,CAAC,GACH,EAAU,KAAK,mBACf,EAAgB,GAChB,OAAW,KAAS,GAAM,aACxB,EAAc,KAAI,MAAlB,EAAsB,EAAM,aAAa,IAI7C,EAAQ,IAAI,OAAO,IACnB,GAAS,EAAW,EAAW,GAC/B,EAAQ,IAAI,OAAO,IAGnB,OADM,GAAS,EAAM,OACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EACF,GAAkB,EAAO,GAAI,EAAW,GAExC,GACI,EAAO,GAAI,EAAW,EAAe,GAE3C,EAAS,KAAM,EAAO,OAAS,EAAI,IAAM,KAAK,OAAO,IAItD,EAAc,mCAEf,GAAM,GAAiB,GAAqB,GACtC,EAAoB,GAAqB,EAAM,qBAErD,EAAQ,iBAAiB,GAAiB,IAC1C,EAAQ,qBAAqB,GAC7B,EAAQ,yBAAyB,GACjC,EAAQ,IAAI,OAAO,IAGrB,YAA8B,GAC5B,GAAI,GAEJ,MAAK,GAAc,2BAA6B,KAC9C,EACI,GAAsB,EAAc,2BAExC,EAAiB,GAAqB,EAAM,kBAGvC,EAGT,YAA+B,GAC7B,GAAI,GAAiB,GACf,EAAyB,GACzB,EAAgB,GACtB,OAAW,KAAS,GAAM,aACxB,EAAa,KAAK,EAAM,aAAa,IAEvC,OAAyB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,EAAW,OAAS,GACpB,EAAW,SAAW,GAAK,EAAW,GAAG,cAAc,OAAS,GAClE,EAAiB,GACjB,MAEF,EAAM,KAAI,MAAV,EAAc,GAEhB,GAAI,EAEF,OAAoB,GAAA,EAAA,EAAA,EAAM,OAAN,EAAA,EAAA,OAAA,KAElB,OAFS,GAAK,EAAA,GACV,EAAO,GACQ,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAM,QAAQ,KAAU,GAC1B,GAAI,GACF,EAAiB,GACjB,UAEA,GAAO,GAIb,GAAI,CAAC,EACH,MAIN,MAAO,GAGT,YACI,EAAkB,EAElB,GAAA,AAAA,IAAA,QAAA,GAA6D,QAAQ,KAEvE,OADI,GAAO,GACF,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAI,GACN,GAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAAK,KAE1C,GAAQ,EAAO,GACf,EAAO,EAAK,MAAM,EAAG,EAAU,IAC/B,GAAQ,IAAI,OAAO,EAAU,GAAK,EAAK,QAEzC,EAAQ,GAQV,YACI,EAAc,EAEd,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAGhB,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EACF,CAAI,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,YAChE,GAAS,EAAQ,EAAW,GAM9B,YACI,EAAc,EAAqB,EAEnC,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAIhB,OADM,GAAwB,GACX,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,GAAiB,MAAQ,EAAc,OAAS,GAChD,EAAc,QAAQ,KAAU,GAClC,SAEF,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAAG,KACrC,EAAoB,EAAK,YAAY,GACrC,EAAqB,EAAK,cAAc,GAC9C,EAAY,KACL,EAAY,IAAI,EAAiB,KAAK,EAAkB,MAGnE,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EAAkB,EAAY,SAAW,EAAI,GAAK,EAAY,GAC9D,EAAmB,CACpB,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,WAC3D,GAGF,GAAS,EAAQ,EAAW,GAC5B,OAAS,GAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,EACxC,GAAS,CAAC,GAAI,GAAI,GAAI,EAAY,IAAK,EAAW,GCvNtD,AA8BA,YACI,EAAa,EAAe,GAC9B,MAAQ,KAAQ,gBAAkB,IAAQ,gBAClC,IAAQ,gBACZ,IAAU,GAAK,MAAO,IAAU,qBAUlC,EAA6B,GAC/B,GAAI,IAAmB,KACrB,MAAO,MACF,GAAI,MAAO,IAAmB,SACnC,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAmB,UAC1B,MAAO,IAAmB,UAC7B,MAAO,GACF,GAAI,YAA0B,QAGnC,OAFM,GAAU,GACV,EAAc,EAAe,OAC1B,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAe,GAC5B,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACf,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAArB,GAAM,GAAW,EAAA,GACd,EAAgB,EAAe,GACrC,GAAI,IAAgB,QAAU,MAAO,IAAkB,SAIrD,EAAO,GAAe,OAEtB,GAAM,GAAQ,GAA0B,GACxC,EAAO,GAAS,GAAoB,EAAe,IAGvD,MAAO,gBAWP,EAAyC,GAC3C,GAAI,GAAa,KACf,MAAO,MACF,GAAI,MAAO,IAAa,SAC7B,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAa,UAAc,MAAO,IAAa,UACzD,MAAO,GACF,GAAI,YAAoB,QAG7B,OAFM,GAAU,GACV,EAAc,EAAS,OACpB,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAS,GACtB,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACrB,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAU,EAAS,GACnB,EAAQ,GAA0B,GACxC,AAAK,KAAU,QAAU,IAAU,cAC/B,MAAO,IAAY,SAIrB,EAAO,GAAS,EAEhB,EAAO,GAAS,GAAoB,EAAS,GAGjD,MAAO,IC7HX,GAGM,IAAU,QCHhB,AA0BA,YAAiC,EAAqB,GAEpD,GAAI,EAAI,OAAS,MAAQ,EAAI,QAAU,EAAI,MAEzC,MAAO,GAET,IAEE,MAAO,GAAA,KAAK,EAAK,EAAI,aACd,GAEP,KAAM,IAAI,GACN,0BAA0B,EAAI,MAAK,kCACnC,gBAAe,EAAI,KAAI,MAAM,EAAI,MAAK,QAgB9C,GAAA,IAAA,WAUE,WAAY,GACV,GAVM,KAAA,SAAmC,GACnC,KAAA,QAAkC,GAClC,KAAA,QAAoC,GAQtC,YAAiB,GACnB,OAAW,KAAM,GAAM,SACrB,KAAK,SAAS,GAAM,EAAM,SAAS,GACnC,AAAI,IAAM,GAAM,SACd,MAAK,QAAQ,GAAM,EAAM,QAAQ,SAIrC,GAAI,GAAS,KACX,OAEF,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,KAAK,IAAI,EAAK,IAAK,EAAK,SAe9B,SAAA,UAAA,IAAA,SAAI,EAAqB,EAAe,GACtC,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAK,SAAS,EAAI,IAAM,GAAwB,EAAK,GACrD,KAAK,QAAQ,EAAI,MAAQ,EAAI,GAC7B,AAAI,GAAQ,MACV,MAAK,QAAQ,EAAI,IAAM,OAGzB,MAAM,IAAI,GAAW,uBAAuB,EAAI,KAAI,QAAQ,EAAI,IAElE,MAAO,OAQT,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,IAAI,EAAK,IAAK,EAAK,QAO1B,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,MAAK,SAAS,EAAI,KAAO,MAMlC,EAAA,UAAA,MAAA,WACE,MAAO,QAAO,KAAK,KAAK,UAU1B,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,SAAS,EAAI,SAG3B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,SAAS,KAWzB,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,QAAQ,EAAI,SAG1B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,QAAQ,KAKxB,EAAA,UAAA,aAAA,WACE,AAAI,KAAK,SAAW,MAClB,EAAA,QAAQ,KAAK,UAGnB,KAIM,GAA+D,GAG/D,GAC4D,eA4C9D,EAA0C,EAC1C,EAAiB,GAWnB,OATM,GAAoB,GAAU,KAAO,GAAQ,EAAO,SAEpD,EAAe,MAAM,QAAQ,GAC7B,EACF,EAAe,EAA8B,CAAC,GAE5C,EAAc,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,OACpC,EAAyB,GACzB,EAAY,EAAS,QACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,AAAI,EAAU,QAAQ,KAAgB,GACpC,EAAa,KAAK,EAAS,SAAS,IAEpC,EAAa,KAAK,MAItB,AAAI,GAAS,MAEX,GAAM,cAAgB,UACtB,EAAM,cAAgB,UAIxB,GAAM,GACF,EAAY,KAAK,KAAO,IAAM,EAAS,QAAQ,KAAK,KACpD,EACA,EACJ,GAAI,GAAa,IAAoB,MAGnC,GAAM,GAAM,GAAqC,EAAY,GAC7D,EAAS,EAAI,OACb,EAAkB,EAAI,gBAGtB,GAAa,GAAmB,EAChC,GAAsB,GAAmB,EAE3C,EAAS,GAAa,GACtB,EAAkB,GAClB,AAAK,GACH,OAAO,OAAO,EAAiB,GAAsB,IAMvD,OAHM,GAAmB,GAAI,IAAS,GAG7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAI,GAAS,MAEX,GAAM,GAAa,EAAA,SAAS,WAC5B,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAExB,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAI1B,GAAM,GAAW,EAAO,GAClB,EAAW,EAAS,YAC1B,GAAI,YAAoB,IACtB,SAOF,OALM,GAAwB,GACxB,EAAuB,GACvB,EAA6B,GAE/B,EAAa,GACG,EAAA,EAAA,EAAA,EAAS,OAAT,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAiB,SAAS,GAClC,EAAO,EAAiB,QAAQ,GACtC,EAAY,KAAK,GACjB,EAAW,KAAK,GAChB,AAAI,GAAQ,MACV,GAAa,IAEf,AAAK,GACH,GAAgB,EAAM,QACtB,AAAI,EAAgB,EAAM,QAAU,GAAK,CAAC,EAAS,OAAO,IACtD,EAAY,QAAQ,EAAM,QAAU,IAAM,CAAC,EAAM,YACjD,EAAM,YAAY,WAAa,IACjC,EAAiB,KAAK,IAK5B,AAAI,GACF,GAAS,GAAU,GACnB,EAAO,KAAU,EAAW,IAE9B,GAAM,GACF,GAAO,EAAS,MAAM,EAAa,IACnC,EAA8B,KAClC,AAAI,EAAS,iBACX,GAAa,EAAS,YAAY,EAAa,IAKjD,OAHM,GAAe,GAAe,GAC9B,EACF,MAAM,QAAQ,GAAgB,EAAe,CAAC,GACzC,EAAI,EAAG,EAAI,EAAsB,OAAQ,EAAE,GAClD,AAAK,EAAiB,OAAO,EAAsB,KACjD,EAAiB,IACb,EAAsB,GAAI,EAAc,GACxC,MAAM,QAAQ,GAAc,EAAW,GAAK,GAElD,GAAM,GAAQ,EAAY,QAAQ,EAAsB,GAAG,MAC3D,AAAI,IAAU,IACZ,GAAa,GAAS,EAAc,IAIxC,AAAK,GAEH,EAAA,QAAQ,GAQZ,SAAiB,eAEV,EAAe,EAAe,EAAa,GAsBpD,YACI,EAA2B,GAE7B,EAAA,KAAK,OACD,GAAW,MAAQ,EAAQ,OAAS,EACpC,WAAM,MAAA,0CAEV,GAAI,GAAgC,GAChC,EAAkC,GACtC,GAAI,EAAQ,SAAW,GAErB,GAAM,GACF,GAAgD,EAAQ,GAAI,GAChE,EAAc,EAAI,OAClB,EAAoB,EAAI,iBAGxB,QADM,GAAU,GAAI,KACA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAKlB,OALS,GAAK,EAAA,GACR,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,aAIc,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAxB,GAAM,GAAc,EAAA,GACvB,AAAK,EAAQ,IAAI,EAAe,OAC9B,GAAY,KAAK,GACjB,EAAQ,IAAI,EAAe,sBAKpB,GACT,AAAI,EAAkB,IAAS,MAC7B,GAAkB,GAAQ,GAAI,MAEhC,EAAa,GAAM,QACf,SAAA,GAAa,MAAA,GAAkB,GAAM,IAAI,MAL/C,OAAW,KAAQ,KAAR,GASf,MAAO,CACL,OAAQ,EACR,gBAAiB,GAAoB,IAIzC,YAA6B,GAC3B,GAAM,GAAmC,GACzC,OAAW,KAAQ,GACjB,EAAgB,GAAQ,EAAa,GAAM,KAE7C,MAAO,eAeL,EAAuB,GASzB,OAPM,GAAU,GAAI,KACd,EAA2B,GAC3B,EAA6B,GAKjB,EAAA,EAAA,EAAA,EAAS,QAAT,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAQ,IAAI,GAGd,GAAM,GAA0B,GAC1B,EAAkB,GAKxB,IAFA,EAAM,KAAK,GAEJ,EAAM,OAAS,IACpB,GAAM,GAAM,EAAM,EAAM,OAAS,GACjC,GAAI,EAAQ,IAAI,EAAI,OAClB,EAAM,MACN,SAEF,GAAM,GAAc,EAAM,EAAM,OAAS,KAAO,EAAM,OAAS,EAC/D,GAAI,EAAI,OAAO,SAAW,GAAK,EAE7B,EAAM,MACN,EAAO,KAAK,GACZ,EAAQ,IAAI,EAAI,MAChB,AAAI,GACF,EAAM,WAKR,EAAM,KAAK,EAAM,OAAS,GAC1B,OAAoB,GAAA,EAAA,EAAA,EAAI,OAAJ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GAQd,GALA,AAAI,EAAa,EAAM,OAAS,MAC9B,GAAa,EAAM,MAAQ,GAAI,MAEjC,EAAa,EAAM,MAAM,IAAI,EAAI,MAE7B,EAAQ,IAAI,EAAM,MACpB,SAEF,EAAM,KAAK,KAIjB,MAAO,CAAC,OAAM,EAAE,aAAY,GAS9B,YAAwB,GAEtB,GAAI,GACJ,GAAI,EAAM,YAAY,aAAa,SAAW,EAC5C,EAAe,EAAM,YAAY,YAGjC,OADI,GAAoB,KACf,EAAI,EAAG,EAAI,EAAM,YAAY,aAAa,OAAQ,EAAE,EAC3D,OAA2B,GAAA,EAAA,EAAA,EAAM,YAAY,aAAa,GAChD,cADiB,EAAA,EAAA,OAAA,KAAtB,GAAM,GAAY,EAAA,GAErB,GAAI,EAAa,KAAO,EAAM,IAC5B,EAAY,EACZ,OAIN,EAAe,EAAM,YAAY,YAAY,GAE/C,MAAO,GCjhBT,AA6CA,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAoCtC,WAAY,GAAZ,GAAA,GAEE,EAAA,KAAA,KAAM,KAAG,KAET,GAtBF,EAAA,eAAiB,GAAI,KAqBnB,EAAK,KAAO,EAAK,KACb,EAAK,MAAQ,MACf,GAAM,GAAS,EAAK,eAAe,cACnC,EAAK,KAAO,GAAO,GAqBrB,GAlBA,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAKlB,AAAI,MAAM,QAAQ,EAAK,QACrB,EAAK,OAAS,EAAK,OAAO,QAE1B,EAAK,OAAS,CAAC,EAAK,QAEtB,AAAI,MAAM,QAAQ,EAAK,SACrB,EAAK,QAAU,EAAK,QAAQ,QAE5B,EAAK,QAAU,CAAC,EAAK,SAInB,GAAqB,EAAK,QAAQ,SAAW,EAAK,OAAO,OAC3D,KAAM,IAAI,GACN,mGAEA,IAAG,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,SAIhC,AAAI,GAAqB,EAAK,SAAS,SAAW,EAAK,QAAQ,QAC7D,QAAQ,KACJ,qGAEA,IAAG,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,SAOjC,EAAK,YAAc,GACnB,EAAK,uBAAyB,GAC9B,EAAK,yBAA2B,GAKhC,EAAK,aAAe,GACpB,EAAK,wBAA0B,GAC/B,EAAK,0BAA4B,GAKjC,EAAK,OAAS,GAMd,EAAK,sBAAwB,GAe7B,OAAgB,GAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YACtB,EAAK,aAAa,KAAK,GACvB,EAAK,wBAAwB,KAAK,GAClC,EAAK,0BAA0B,KAAK,GAMtC,OAAgB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YAKtB,GAAqB,IAAc,EAAG,4BACtC,GAAqB,IAAgB,EAAG,8BACxC,EAAK,YAAY,KAAK,GACtB,EAAK,uBAAuB,KAAK,GACjC,EAAK,yBAAyB,KAAK,GAIrC,EAAK,WAAa,GAClB,EAAK,YAAc,GACnB,EAAK,gBAAkB,GACvB,EAAK,eAAiB,GACtB,EAAK,gBAAkB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,EAAK,YAAY,GAE/B,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,WACN,6DACA,qBAAoB,EAAK,OAAM,MAC/B,UAAS,EAAC,0BACV,oBAAmB,EAAM,eAAc,MAE7C,EAAK,WAAW,KAAK,EAAM,MAC3B,EAAK,gBAAgB,KAAK,EAAM,iBAEhC,EAAK,eAAe,KAAK,EAAM,MAEjC,OAAoB,GAAA,EAAA,EAAA,EAAK,aAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,YAAY,KAAK,EAAM,MAG9B,EAAK,oBAAsB,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QAClD,EAAK,qBAAuB,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,QAwFpD,OAjFM,GAA0C,GAE1C,EAAyC,GACzC,EAA4C,GAE5C,EAA6C,GAC7C,EAA4C,GAC5C,EAAiC,GAoBjC,EACF,SAAC,GAAwB,GAAuB,GAC/C,GAAe,GAAoB,IAClC,AAAI,KAAS,MAAQ,IAAa,MAAQ,IAAe,OACvD,IAAQ,GAAO,YACf,GAAY,GAAO,UACnB,GAAc,GAAO,aAEvB,GAAM,IAAO,GAAM,aAAa,IAGhC,GAAI,GAAgB,QAAQ,MAAU,GACpC,KAAM,IAAI,IACN,cAAc,GAAO,KAAI,cAAc,GAAM,KAAI,yBAKvD,GAAI,GAAc,QAAQ,MAAU,GAClC,OAIF,EAAK,eAAe,IAAI,EAAU,QAAQ,GAAO,KAGjD,AAAM,GAAM,KAAM,IAChB,GAAa,GAAM,IAAM,OAAO,KAAK,GAAc,QAGrD,AAAI,GAAgB,QAAQ,MAAU,IACpC,GAAgB,KAAK,IAKvB,OADM,IAAmB,GAAK,cAAc,OACnC,GAAI,EAAG,GAAI,GAAkB,MACpC,GAAM,IAAI,GAAK,aAAa,IACtB,GAAQ,GAAK,cAAc,IAC3B,GAAY,GAAK,YAAY,IAC7B,GAAc,GAAK,cAAc,IACvC,EACI,GAAG,GAAe,GAAiB,GAAO,GAC1C,IAGN,IADA,GAAc,KAAK,IACZ,GAAgB,QAAQ,KAAS,GACtC,GAAgB,OAAO,GAAgB,QAAQ,IAAO,GAExD,EAAuB,KAAK,KAG5B,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAgB,EAAG,EAAe,GAKpC,OAFM,GACF,EAAuB,QAAQ,UAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAa,EAAK,IAAM,EAExB,AAAM,EAAK,KAAM,IACf,GAAY,EAAK,IAAM,GAEzB,GAAI,GAAQ,EAAY,EAAK,IAGvB,EACD,EAAa,EAAK,cAAc,KAAO,KACnC,EACA,EAAa,EAAK,cAAc,IAOzC,EAAQ,KAAK,IAAI,EAAO,GACxB,EAAa,EAAK,cAAc,IAAM,EACtC,EAAe,EAAK,cAAc,IAAM,EAAK,cAC7C,EAAY,EAAK,IAAM,EAGvB,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAa,aAAa,GACxC,EACD,EAAY,EAAY,KAAO,KAAO,EACA,EAAY,EAAY,IACnE,EAAY,EAAY,IAAM,KAAK,IAAI,EAAQ,EAAG,GAClD,EAAa,EAAY,IAAM,GAKnC,GAAM,GAA0C,GAChD,OAAW,KAAU,IACnB,GAAM,GAAQ,EAAY,GAC1B,AAAM,IAAS,IACb,GAAa,GAAS,IAExB,EAAa,GAAO,KAAK,EAAa,IAIxC,GAAM,IAA4C,GAClD,OAAW,MAAW,IACpB,GAAM,GAAQ,EAAa,IAC3B,AAAM,IAAS,KACb,IAAc,GAAS,IAEzB,GAAc,GAAO,KAAK,EAAe,KAI3C,GAAI,IAAY,OAAO,KAAK,IACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAG1B,EAAK,OAAS,GACd,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACR,GAAiB,GAAc,GAGrC,GAAe,KAAK,SAAC,GAAG,IACtB,GAAM,IAAS,EAAa,GAAE,IACxB,GAAS,EAAa,GAAE,IAC9B,MAAI,IAAS,GACJ,GAEL,GAAS,GACJ,EAEF,IAET,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACd,AAAI,YAAiB,IACnB,EAAK,sBAAsB,KAAK,GAElC,EAAK,OAAO,KAAK,IAGrB,EAAK,cAAgB,GAGrB,GAAY,OAAO,KAAK,GACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAStB,OAJM,IAAoB,EAAK,OAAO,QAGhC,GAAoC,GACtB,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,KAClB,OADS,GAAK,GAAA,IACK,GAAA,EAAA,GAAA,EAAa,GAAb,GAAA,GAAA,OAAA,MAAd,GAAM,GAAI,GAAA,IACP,EAAQ,EAAK,cACnB,GAAI,GAAS,MACX,OAAgB,IAAA,EAAA,GAAA,EAAK,aAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAI,GAAkB,QAAQ,KAAO,GACnC,KAAM,IAAI,IACN,sDAAsD,EACtD,eAAc,EAAM,KAAI,OACxB,uDACA,WAAU,KAGlB,OAAgB,IAAA,EAAA,GAAA,EAAK,cAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAkB,KAAK,GAEzB,GAAwB,KAAK,EAAM,OAMzC,EAAK,aAAe,EAKpB,OADM,IAAW,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,mBAC7B,IACT,GAAM,IAAiB,GAAS,OAAO,SAAA,IAAK,MAAA,MAAM,KAAM,OACxD,GAAI,KAAmB,EACrB,KAAM,IAAI,IACN,aAAa,GAAI,aAAa,GAAc,uEAE5C,KAAK,UAAU,MANJ,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAd,GAAM,IAAI,GAAA,OAAJ,IAcX,SAAK,cAAgB,GAErB,EAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,EAAK,OACnB,cAAe,EAAK,QACpB,WAAY,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,QACjC,YAAa,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,QACnC,YAAa,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QACpC,aAAc,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,UAExC,EAAK,MAAQ,GACb,EAAK,UAAY,IAGT,SAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,cAAc,KAAK,KAAI,2BA8B3C,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,GAAM,GACc,CAAC,qBAAsB,KAAM,qBAAsB,GACvE,GAAI,EAAE,KAAK,YAAc,GACvB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAO,sBAAwB,EAAM,UAAU,qBAKjD,OAAwB,GAAA,EAAA,EAAA,KAAK,sBAAL,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAO,sBAAwB,EAAU,UAAU,sBAGvD,SAAO,qBAAuB,KAAK,UAC5B,GAGT,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,OAAO,QAAQ,SAAA,GAEhB,EAAc,kBACX,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,MAElC,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WAIE,GAAI,KAAK,kBAAkB,OAAS,EAClC,KAAM,IAAI,GACN,wNAMN,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADI,GAA2B,GACX,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAU,EAAQ,OAAO,EAAM,kBAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACb,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAQ,KAAI,MAAZ,EAAgB,EAAM,qBAExB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACtB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAiB,KAAI,MAArB,EAAyB,EAAM,kBAEjC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAkB3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,IAGnC,OAFM,GAAgD,GAClD,EAAoB,EACJ,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACO,EAAA,EAAA,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,EAAa,EAAO,eAAiB,KACvC,KAAM,IAAI,GAAW,0BAA0B,EAAO,cAExD,EAAa,EAAO,cAAgB,EACpC,IAIJ,GAAM,GAAoD,GAC1D,OAAW,KAAQ,IAIjB,GAAI,GAAgB,EACpB,GAAI,EAAa,IAAS,MACxB,GAAM,GAAS,EAAK,MAAM,KACpB,EACF,EAAO,MAAM,EAAG,IAAI,OAAO,CAAC,EAAO,EAAO,OAAS,KACvD,EAAgB,EAAiB,KAAK,KAExC,GAAI,EAAa,IAAkB,KACjC,EAAkB,KAAK,CAAC,EAAa,GAAgB,EAAQ,aACpD,EACT,KAAM,IAAI,GACN,gDAAgD,GAEtD,MAAO,GAAa,GAGtB,GAAI,GAEF,GAAM,GAAuB,GAC7B,OAAW,KAAQ,GACjB,EAAW,KAAK,GAElB,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,GACH,EAAW,OAAM,OAChB,EAAiB,yBACrB,IAAG,IAIX,GAAc,IAON,EAAA,UAAA,cAAV,WACE,GAAM,GAAY,KAAK,YACjB,EAAwC,GAC9C,SAAY,UAAe,KAAK,eAChC,EAAY,OAAY,EACxB,EAAY,aAAkB,eAAe,GAG7C,EAAY,QAAa,gBAClB,GAeT,EAAA,UAAA,OAAA,SAAO,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IACnB,GAAM,GAAc,GAAoB,KAAK,iBAC7C,MAAO,GAAe,KAAK,UAAU,GAAe,GAgBtD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAE9B,OADM,GAAW,GAAI,IACZ,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,EAAK,OAAO,GAAI,EAAO,IAEtC,MAAO,IAAQ,EAAK,QAAS,EAAU,MAa3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAC9B,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAA2B,KAAM,EAAO,QAEhD,EAAQ,GAAqB,GAGxB,EAAK,iBAAiB,EAAQ,GAAO,MAahD,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAM,GAAc,GAA+B,GACnD,GAAI,EAAY,SAAW,KAAK,YAAY,OAC1C,KAAM,IAAI,GACN,+BAA+B,EAAU,KACzC,cAAa,KAAK,YAAY,OAAM,oBAK1C,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,EAAY,OAAQ,KACtC,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAa,EAAY,GAGzB,EAAW,EAAM,KAAO,OAC9B,EAAqB,GAAY,EAGnC,GAAM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,GAAK,MAAA,UAAS,EAAG,MACrB,KAAK,IAE5B,GAAI,EAAU,OAAS,EACrB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GAEP,EAAQ,EAAK,cACnB,GAAI,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,QAAQ,EAAM,MAAQ,GAExD,SAIF,OADM,GAAuB,GACpB,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAc,EAAa,KAAI,IAAI,EAAS,IAAI,EAChD,EAAa,EAAqB,GACxC,EAAY,KAAK,GAQnB,OALM,GAAc,EAAM,mBACtB,GAA+B,IAE7B,EAAe,GAA+B,GAC9C,EAAY,EAAM,aAAa,QAAQ,GACpC,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAqB,GAAY,EAAa,IAStD,OAFM,GAAwB,GACxB,EAA4B,GACzB,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GACzC,EAAc,KAAK,0BAA0B,GAC7C,EAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAgB,KAAK,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAgB,OAAQ,KAC1C,GAAM,GAAM,EAAgB,GAC5B,GAAqB,IAAO,IAC5B,EAAa,KAAK,EAAqB,IAIzC,MAAO,IAA+B,IAa9B,EAAA,UAAA,iBAAV,SAA2B,EAAkB,GAE3C,AAAI,GAAS,MACX,GAAQ,GAA2B,KAAM,EAAO,SASlD,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,GACxC,GAAM,GAAI,KAAK,OAAO,GAChB,EAAI,EAAO,GACX,EAAO,EAAM,GACnB,EAAU,EAAE,IAAM,CAAC,EAAG,GAMxB,OAHM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAUjB,OAVS,GAAI,EAAA,GAEP,EAAQ,EAAK,cACb,EAAwB,EAAK,aAC7B,EAAyB,EAAK,cAK9B,EAAe,GAAI,OACT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAE,KAAM,IACV,EAAa,KAAK,EAAU,EAAE,KAGlC,GAAI,EAAa,SAAW,EAAsB,QAEhD,GAAI,GAAiB,GACjB,EAAe,OACf,EAAa,OACb,EAAa,OACb,EAAW,OAKf,GAHA,AAAI,EAAK,UAAY,MACnB,GAAS,EAAK,UAEZ,EAAa,SAAW,GACpB,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GACvB,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAgB,IACpD,EAAc,GACV,EAAM,YAAY,EAAgB,IACtC,EAAkB,CAAC,GACnB,EAAgB,CAAC,OAEjB,GAAkB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KAC1C,EAAgB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KACxC,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAiB,IACrD,EAAc,GACV,EAAM,YAAY,EAAiB,IAGzC,GAAI,EAAM,oBACR,KAAM,IAAI,IACN,yHAMN,OAAS,GAAI,EAAG,EAAI,EAAuB,OAAQ,EAAE,GACnD,GAAM,GAAI,EAAuB,GAC3B,EAAI,EAAc,GAClB,EAAO,EAAY,GACzB,EAAU,EAAE,IAAM,CAAC,EAAG,KAS9B,OAHM,GAA0B,GAC1B,EAAwB,GACxB,EAAwB,GACd,EAAA,EAAA,EAAA,KAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,GACI,EAAE,KAAM,GAAW,4BAA4B,EAAE,KAAI,MAAM,EAAE,IAC3D,GAAA,GAAA,EAAA,EAAA,IAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAa,KAAK,EAAO,OACzB,EAAc,KAAK,GACnB,EAAY,KAAK,GAInB,MAAO,CAAC,EAAe,EAAa,IAW9B,EAAA,UAAA,uBAAR,SAA+B,GAG7B,OAFM,GAAiD,GACnD,EACgB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,YAAiB,GAAY,EAAI,EAC7C,OAAS,GAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,KAAK,eAAe,IAAI,IAE1B,GAAkB,GAAW,EAC7B,GAAa,IAInB,MAAO,IAsBT,EAAA,UAAA,SAAA,SAAS,EAAe,GACtB,GAAI,GAAS,MACX,GAAI,KAAK,OAAO,QAAU,EACxB,KAAM,IAAI,GACN,wCAAwC,EAAK,oBAC7C,QAAO,KAAK,OAAO,OAAM,eAE7B,MAAO,MAAK,OAAO,WAGjB,GAAQ,KACV,KAAM,IAAI,GAAW,8CAIzB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,OAAS,EACjB,MAAO,GAGX,KAAM,IAAI,GAAW,kBAAkB,IAQzC,EAAA,UAAA,gBAAA,WAAA,GAAA,GAAA,KAKE,MAAO,GAAA,KAAK,WAEV,OADM,GAAmB,GACL,EAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACL,EAAY,EAAG,EAAY,EAAM,aAAa,OAClD,EAAE,GACL,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,EAAK,eAAe,IAAI,IAC1B,EAAO,KAAI,MAAX,EAAe,EAAM,mBAK3B,MAAO,MAIX,EAAA,UAAA,UAAA,WAWE,OAVM,GAAmC,CAAC,KAAM,KAAK,MAK/C,EACF,KAAK,uBAAuB,KAAK,QAG/B,EAAe,GACD,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EAAiB,EAAM,eACvB,EAAc,EAAM,YACpB,EAAuB,GACpB,EAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAO,EAAM,aAAa,GAC1B,EAAU,EAAU,QAAQ,EAAO,GACrC,EAAS,GACb,GAAI,KAAK,eAAe,IAAI,IAG1B,GAAI,EAAK,SACP,IACE,KAAK,UAAU,EAAK,UACpB,EAAS,EAAK,eACP,GACP,QAAQ,KACJ,SAAS,EAAM,KAAI,mDAEhB,GAAK,SAAQ,gCAChB,+EAEJ,EAAS,GAGb,GAAI,EAAK,cAAc,OAAS,GAE9B,OADM,GAAW,GACR,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAU,EAAU,QAAQ,EAAc,GAC5C,EAAe,EAAkB,GACrC,AAAI,GAAgB,MAClB,GAAe,GAEjB,EAAS,KACL,CAAC,EAAa,KAAM,EAAc,EAAa,IAErD,EAAqB,KAAK,KAIhC,GAAM,GAAiC,GACvC,EAAK,KAAU,EAAM,KACrB,EAAK,UAAe,EACpB,EAAK,OAAY,EACjB,EAAK,aAAkB,EACvB,EAAa,KAAK,GAEpB,EAAO,OAAY,EAGnB,OADM,GAAc,GACX,EAAI,EAAG,EAAI,KAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAY,KAAK,uBAAuB,GAExC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,yBAAyB,GAClD,EAAY,KAAK,CAAC,EAAM,KAAM,EAAc,IAE9C,EAAO,YAAiB,EAGxB,OADM,GAAe,GACZ,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GAEzC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,0BAA0B,GACnD,EAAa,KAAK,CAAC,EAAM,KAAM,EAAc,IAE/C,SAAO,aAAkB,EAClB,GAgBF,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAA8C,GAO9C,EAAkE,GACxE,WACI,EAAc,GAChB,AAAM,EAAM,OAAQ,GAGlB,EAAiB,EAAM,MAAM,KAAK,GAFlC,EAAiB,EAAM,MAAQ,CAAC,GAMpC,WAAqB,EAAc,GAGjC,OAFM,GAAiC,GACnC,GACoB,GAAA,EAAA,GAAA,EAAA,GAAA,GAAA,OAAA,MAAnB,GAAM,IAAS,GAAA,IACZ,GAAmB,GAAU,GAC7B,GAAmB,GAAU,GAC7B,GAAqB,GAAU,GAKrC,GAHA,GAAS,GAAU,IAAM,KACrB,GACA,GAAU,GACV,CAAE,MAAoB,KACxB,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAe,EAAc,IACnC,GAAI,GAAa,aAAa,QAAU,IACtC,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAc,GAAa,aAAa,IAC9C,EAAa,KAAK,GAAY,cAAc,KAK9C,AAAI,EAAa,OAAS,GACxB,EAAM,MACF,GAA+B,GAC/B,IAUR,WAAsB,GACpB,GAAM,GAAY,EAAU,KAEtB,EACF,GACI,EACA,EAAO,eAAoB,KACvB,EAAO,cACP,IACZ,EAAM,6BAA6B,GACnC,EAAc,GAAa,EAE3B,GAAM,IACF,EAAU,aACd,GAAiB,QAAQ,SAAA,IACvB,GAAI,CAAE,cAAoB,QACxB,KAAM,IAAI,GACN,yDACI,IAMV,EAAmB,EAAO,MAO9B,OAFM,GAAO,EAAO,KACd,EAAmB,EAAO,OACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAa,GAOf,KAAO,CAAC,GAA4B,IAClC,OAAwB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAQ,EAAc,EAAU,MACtC,GAAI,EAAM,OAAQ,IAChB,GAAM,GAAkC,EAAiB,EAAM,MAC/D,MAAO,GAAiB,EAAM,MAC9B,OAAuB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAY,EAAO,KAU3B,OAJM,GAAiC,GACjC,EAAkC,GAClC,EACF,EAAO,YACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAa,KAAK,EAAmB,IAIvC,OAFM,GACF,EAAO,aACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAc,KAAK,EAAmB,IAExC,MAAO,IAAI,GAAI,CAAC,OAAQ,EAAc,QAAS,EAAe,KAAI,KASpE,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WAGE,GAAI,KAAK,UACP,KAAM,IAAI,GACN,wLAIN,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,SACR,MAAO,GAGX,MAAO,oCAST,EAAA,UAAA,YAAA,WAAA,GAAA,GAAA,KACE,EAAA,KAAK,WACH,EAAK,OAAO,QAAQ,SAAA,GAElB,AAAI,EAAM,UACR,EAAM,mBAMhB,GA3vCwC,IC7CxC,AAiCA,YACI,EAAmD,EACnD,GACF,GAAM,GAAa,EAAY,OAC/B,GAAI,GAAW,MAAS,MAAM,QAAQ,IAAY,EAAQ,SAAW,EACnE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,QAEjC,GAAI,IAAe,EACjB,MAAI,OAAM,QAAQ,IAAY,EAAQ,SAAW,EACxC,EACE,MAAO,IAAY,UAAY,EAAY,IAAM,GACnD,CAAE,EAA2B,EAAY,KAEzC,CAAC,GAGZ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAQ,SAAW,EACrB,KAAM,IAAI,OACN,YAAY,EAAU,mBAAmB,EAAQ,OAAM,IACvD,kCAAiC,EAAU,cAC3C,iEAEN,MAAO,WAEL,MAAO,IAAY,UAAY,OAAO,KAAK,GAAS,OAAS,GAC7D,MAAQ,GAA2B,OAAO,KAAK,GAAS,KACpD,UACN,GAAM,GAAwB,GAC9B,SAAY,QAAQ,SAAA,GAClB,AAAI,IAAc,GAChB,EAAO,KAAM,EAA2B,IAExC,EAAO,KAAK,QAGT,MAEP,MAAM,IAAI,OACN,2BAA2B,EAAU,cACrC,OAAM,EAAU,kCACb,GAAU,+BAA+B,EAAW,WACvD,aAAY,EAAU,oBAAoB,KAAK,UAAU,iBAkB7D,EACA,GACF,MAAO,IACH,EAAa,EAAa,2BA6B5B,EAAW,EAAuB,EAClC,2GACF,GAAI,GAAgB,MAAQ,GAAoB,KAG9C,KAAM,IAAI,OAAM,qDAGd,IAAe,KAEX,GAAqB,EAAA,KAAK,WAC9B,GAAI,EAAE,MAAM,SAAW,EAErB,MAAO,GAAE,QACJ,GAAI,EAAE,MAAM,SAAW,EAC5B,GAAI,EAAE,MAAM,GAAK,GAEf,GAAM,GAAO,EACb,MAAO,GAAE,OAAO,QACX,GAAI,EAAE,MAAM,KAAO,EAExB,MAAO,GAAE,QAAQ,CAAC,EAAE,MAAM,KAE1B,KAAM,IAAI,OACN,+CAA+C,EAAE,MAAM,GAAE,4EAK/D,MAAM,IAAI,OACN,yCAAyC,EAAE,KAAI,4EAKjC,EAAA,GAAA,OAAM,KAAK,CAAA,EAAM,EAAS,SA3B9C,CAAA,EAAA,UA2BI,SAAgB,EAAA,MAAA,EAAA,CAAW,EAAA,SACjC,EAAA,QAAQ,GACF,EAA8B,GACpC,EAAc,QAAQ,SAAA,GACpB,GAAI,EAAY,IAAe,KAC7B,KAAM,IAAI,OACN,8DACA,cAAa,EAAU,mCACvB,eAEJ,EAAkB,KAAK,EAAY,MAIvC,CAAA,EAAO,EAAA,SAAS,EAAmB,mBAEnC,MAAA,CAAA,EAAO,uBAWyB,EAAgB,GAClD,MAAO,GAAA,IAAI,EAAQ,GC5LrB,AAyLA,GAAM,IAAgC,GAgBtC,YAII,EAAY,GACd,GAAI,GACA,EAEE,EAAiB,EACvB,EAAK,EAAe,GACpB,EAAK,EAAe,GACpB,EAAA,KAAS,OACL,GAAM,MAAQ,GAAM,KACpB,WAAM,MAAA,+OAIF,IAAG,KAEX,GAAM,GACF,GAA0B,QAAS,EAAM,WAAY,GACnD,EACF,GAA0B,SAAU,EAAM,YAAa,GAErD,EAAoB,EAAY,GAAG,MAAM,GAE/C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,OAAO,OACpC,WAAM,MAAA,mBAAmB,EAAM,OAAO,OAAM,4BACxC,aAAY,EAAY,OAAM,oCAC3B,MAAK,UAAU,EAAM,YAAW,OAE3C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,QAAQ,OACrC,WACI,MAAA,mBAAmB,EAAM,QAAQ,OAAM,6BACvC,aAAY,EAAY,OAAM,sCAC3B,MAAK,UAAU,EAAM,aAAY,OAE5C,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,8BACC,GAAM,WAAW,GAAO,QACrB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,+BACC,GAAM,YAAY,GAAO,QACtB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,MAAO,CAAC,GAAI,EAAa,GAAI,GAG/B,YACI,EAAuB,EAAiB,GAC1C,GAAI,YAAkB,GAAA,OACpB,MAAO,CAAC,GACH,GAAI,MAAM,QAAQ,GACvB,SAAA,KAAS,OACL,EAAO,SAAW,EAAM,OACxB,WAAM,MAAA,wBAAwB,EAAO,OAAM,0BACvC,EAAM,OAAM,iBAAiB,EAAa,SAAS,EAAK,MACzD,EAIP,OAFM,GAAuB,GAEV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,IAAS,KAClB,KAAM,IAAI,GACN,gEACG,GAAa,SAAS,EAAI,OAEnC,EAAO,KAAK,EAAO,IAErB,MAAO,GAIX,YACI,GAMF,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,IACN,0DAEN,MAAO,CAAC,GAAI,EAAK,GAAI,GAAI,EAAK,gBAO5B,EAAY,EACZ,yKA0BF,GAzBM,EAAqB,EAAK,iBAAmB,KACnD,EAAA,KAAS,OACL,EAAM,WAAa,KACnB,WAAM,MAAA,mGAGV,EAAA,KAAS,OACL,GAAQ,KACR,WAAM,MAAA,kGAEV,EAAA,KAAS,OACL,EAAK,QAAU,MAAQ,EAAK,OAAS,GAAK,OAAO,UAAU,EAAK,QAChE,WAAM,MAAA,gEACF,qBAAoB,EAAK,UACjC,EAAA,KAAS,OACL,CAAC,GACI,EAAK,gBAAkB,GAAK,OAAO,UAAU,EAAK,iBACvD,WAAM,MAAA,gEACF,2CAA0C,EAAK,mBACvD,EAAA,KAAS,OAEJ,EAAa,iBAAsB,KACpC,WAAM,MAAA,sFAGN,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,mDAGX,EAAe,EAAK,gBAAkB,KACxC,EAAK,OACL,EAAK,OACT,AAAI,GACF,CAAI,GAAgB,EAAK,gBACvB,EAAA,KAAS,OACL,EAAK,mBAAqB,MACrB,EAAK,kBAAoB,GACzB,OAAO,UAAU,EAAK,mBAC3B,WAAM,MAAA,yIAGF,YAAW,EAAK,qBAElB,GAAiB,GACnB,EAAK,gBAMT,EAAQ,EAAe,GACvB,EAAQ,EAAe,KAIrB,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAe,OACnB,AAAI,EACF,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,IAAK,MAAA,OAAS,MAEzD,EAAkB,EAAU,QAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YACtD,EAAU,EAAK,SAAW,KAAO,EAAI,EAAK,QAC1C,EAA0B,GAC5B,EAAW,EAAS,EAAK,OAAQ,KAAM,KACvC,GAAiB,EAAS,GAC1B,KACA,EAAc,GAJX,EAAY,EAAA,aAAE,EAAA,EAAA,QAKrB,EAAa,SAAS,GACtB,EAAM,QAAU,EAEhB,CAAA,EAAM,EAAa,uBAAnB,SAAA,OACA,EAAM,cAAgB,GAClB,EAAQ,EAAK,cAAgB,KAAO,EAAI,EAAK,aAE9B,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BACZ,GAAQ,EAAK,OACZ,GAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,KAFR,CAAA,EAAA,iBAExB,GAAA,OACI,EAAY,EACZ,EAAa,EACZ,EAAD,CAAA,EAAA,GACa,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BAEV,GAAqB,EAAY,EAAK,gBAAkB,IACzC,CAAA,EAAM,EAAa,QAD0B,CAAA,EAAA,WAKjE,MAJM,GAAc,EAAA,OAIhB,GAAsB,EAAY,KACpC,SAAQ,KACJ,qCACG,GAAK,gBAAe,MACvB,mDACG,GAAS,cACZ,8HAGG,GAAK,gBAAkB,EAAK,OAAM,eACrC,yEAEJ,CAAA,EAAA,KAGE,EAAY,OAAS,KACjB,GACF,GAA8B,EAAO,EAAY,OAD9C,EAAE,EAAA,GAAE,EAAE,EAAA,GAEP,EAA4B,GAClC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAG,GAAG,MAAM,GAEhC,CAAA,EAAM,EAAa,aAAa,EAAY,KAP1C,CAAA,EAAA,cAOF,EAAA,OAEM,EAA8B,IAChC,GAAK,aAAe,MAApB,MAAA,CAAA,EAAA,IACI,EACF,GAAwB,EAAK,YAAa,EAAM,aAC3C,EAAI,2BAAG,GAAI,EAAqB,OACvC,GAAA,GAAA,GAAc,KAAK,CAAA,EAAM,GACrB,EAAG,GAAI,KAAM,EAAqB,MAFO,CAAA,EAAA,YAC7C,EAAA,MAAA,EAAA,CAAmB,EAAA,4BAD4B,QAAE,iBAUrD,IAHM,EAAM,EAAG,OAAO,GAAI,OAAO,GAC3B,EAAO,EAAc,GAC3B,EAAA,QAAY,GACH,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EAChC,EAAQ,EAAU,GAClB,EAAM,EAAK,GACjB,EAAU,GAAS,EACnB,EAAA,KAAS,GAGX,MAAA,CAAA,EAAM,EAAa,WAAW,EAAY,YAA1C,EAAA,OACA,GAAqB,GAErB,IACA,6BAGE,GAAqB,GAAa,EAAK,gBAClB,EAAY,MAE/B,EACE,GAAO,OACP,GAAgB,EAAK,gBACb,GAAA,GAAO,CAAA,EAAM,EAAM,gBACzB,EAAK,eAAgB,CAAC,QAAS,EAAK,sBAFtC,CAAA,EAAA,KAFF,CAAA,EAAA,IAHF,CAAA,EAAA,YAME,SAAU,EAAA,MAAA,OAAA,CAAO,EAAA,wBAGjB,EAAU,GAAO,EAAM,SAAS,EAAO,EAAO,CAC5C,UAAW,EAAK,qBAAuB,KACnC,GACA,EAAK,oBACT,QAAS,wBAGb,IAAS,EAAI,EAAG,EAAI,EAAM,aAAa,OAAQ,EAAE,EAC/C,EAAU,OAAO,EAAM,aAAa,IAAQ,EAAQ,sBAQxD,MAAA,CAAA,EAAA,YAGF,MAAI,GAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,YAErC,MAFA,GAAA,OACA,IACI,EAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,sBAAnB,SAAA,OACA,CAAA,EAAM,EAAM,QAAQ,oBAApB,SAAA,OACA,CAAA,EAAO,EAAM,iBAEb,SAAM,WAAa,8BAKvB,YACI,EAAqB,GAEvB,GAAI,GAAwB,KAC5B,MAAI,GAAK,iBAAmB,KAC1B,EAAgB,EAAK,gBAChB,AAAI,OAAO,SAAS,EAAQ,OACjC,GAAgB,EAAQ,MAEnB,EAKT,YACI,GAKF,MAAQ,OAAQ,GAAuB,UAAa,WAKtD,YAAiC,GAE/B,MAAQ,OAAQ,GAA6B,MAAS,uBAOpD,EAAY,EACZ,uHAKF,GAJA,EAAO,GAAQ,GACT,EAAa,EAAK,SAAW,KAC7B,EAAI,EAAM,aACZ,EAAqB,GACrB,EAAK,QAAU,EACjB,KAAM,IAAI,IAAoB,8CAGhC,GAAA,KAAS,OACL,CAAC,GAAe,EAAK,QAAU,GAAK,OAAO,UAAU,EAAK,SAC1D,WAAM,MAAA,6DACF,aAAY,KAAK,UAAU,EAAK,YACnB,GAAqB,GACtC,GAAA,SADiB,CAAA,EAAA,UAEjB,MAAA,CAAA,EAAO,EAAuB,mBAA9B,EAAA,EAAA,wBAFE,EAAY,EAId,EAAc,EACd,EAAQ,uEAGU,MAAA,CAAA,EAAM,EAAa,eAmCvC,MAnCM,GAAc,EAAA,OACpB,EAAO,EAAA,KAAS,WACd,GAAI,EAAY,OAGR,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAU,EAAG,OAAO,GACpB,EAAY,EAAA,KAAS,WAAM,MAAA,GAAE,KAGnC,GAFA,EAAA,QAAY,GAER,IAAU,EACZ,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAKrB,OADM,GAAY,EAAQ,GAAG,MAAM,cAC1B,GACP,GAAM,GAAW,EAAU,GACrB,EAAY,EAAK,GACvB,EAAK,GACD,EAAA,KAAS,WAAM,MAAA,GAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,MACvD,AAAI,EAAQ,GACV,EAAA,QAAY,IANP,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,IAA/B,GAST,EAAA,QAAY,GACZ,GAAe,EAEf,EAAE,EAEJ,MAAO,KAGL,EAAY,KACd,CAAI,GACF,QAAQ,KACJ,uJAGA,2BAA0B,EAAK,QAAO,eACtC,qHA3CH,GAAa,EAAQ,EAAK,QAAU,YAAI,CAAA,EAAA,2DAkD/C,IAAS,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC3B,EAAY,EAAK,GACvB,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAC3B,EAAA,QAAY,GAGd,MAAA,CAAA,EAAO,GAAiB,SCrmB1B,YAiK+B,GAC7B,EAAA,KAAS,OACL,EAAY,GAAK,OAAO,UAAU,GAClC,WAAM,MAAA,2DACF,gBAiBN,EAAyB,EAAe,GAC1C,MAAI,IAAU,KACL,CAAC,MACC,MAAM,QAAQ,GAChB,EAAO,IAAI,SAAA,GAAS,MAAA,IAAoB,EAAO,EAAO,EAAO,KAE7D,GAAoB,EAAQ,EAAO,EAAO,eAkBjD,EAAyB,GAC3B,MAAO,GAAA,KAAS,WACd,MAAI,IAAU,KACL,KACE,MAAM,QAAQ,GAChB,EAAO,IACV,SAAA,GAAS,MAAC,IAAqB,EAAO,KAInC,GACH,EAAQ,EAAQ,QAAU,QAAU,EAAU,EAAQ,uBAc5D,EAAc,GAIhB,OAHM,GAAkC,GACpC,EAAa,EACb,EAAmB,KAChB,EAAa,GAClB,EAAW,EAAa,EACxB,AAAI,GAAY,GACd,GAAW,GAEb,EAAO,KAAK,CAAC,EAAY,IACzB,EAAa,EAEf,MAAO,GA8BT,YAGI,EAAY,EAAiC,EAC7C,EAAsB,EAAoB,EAAiB,EAC3D,EAA4B,EAC5B,EAAmB,EAA0B,EAC7C,EAAuB,EACvB,mHAoBF,GAnBA,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,GAAU,MACZ,GAAS,GAEX,AAAI,GAAW,MACb,GAAU,IAEZ,AAAI,GAAgB,MAClB,GAAe,GAIb,EAAe,GACnB,AAAI,GAAQ,MAAQ,GAAU,MAC5B,GAAe,IAGb,GAAmB,MACrB,GAAe,GACX,GAAiB,MACnB,KAAM,IAAI,GACN,oGAKF,SACF,EAAM,gBAAgB,EAAK,EAAW,EAAe,mBAEzD,AAAI,GAAmB,MACrB,GAAa,GAAM,EAAG,IAGxB,AAAI,GAAW,MACb,GAAU,GAGN,EAA0B,GAC5B,EAAW,EAAS,EAAQ,EAAc,EAAiB,EAC3D,EAAW,EAAc,GAFtB,EAAY,EAAA,aAAE,EAAO,EAAA,QAG5B,EAAa,SAAS,GACtB,EAAM,QAAU,EAChB,CAAA,EAAM,EAAa,uBAAnB,EAAA,OACA,EAAM,cAAgB,cAIb,qEACP,MAAA,CAAA,EAAM,EAAa,aAAa,cAAhC,EAAA,OACM,EAA4B,IAC9B,IAAiB,MAAjB,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,IACN,qDAEJ,GAAI,IAAY,QACd,KAAM,IAAI,IAAoB,0CACzB,AAAI,GACT,EAAA,KAAK,QAAQ,GAIT,EAAoB,EAAA,SAAS,GAE7B,EAAU,GAAY,EAAiB,cACpC,2DACD,SAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,EAAY,WAA5C,SAAA,OAEA,EAAA,KAAS,WACP,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,GAAW,GACI,EAAmB,EACnB,EAAW,GAChC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAW,EAM/B,OAFM,IAAW,GAAqB,EAAK,IACrC,GAAO,EAAE,IACN,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAK,IACjB,EAAU,IAAS,GACnB,EAAA,KAAS,IAIX,GAAI,IAAe,EAAQ,OAAS,GAC9B,EAGF,OAFM,IAAU,EAAM,SAAS,EAAM,EAAQ,GAEpC,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAQ,IACpB,EAAA,KAAS,IAET,EAAU,OAAS,IAAS,MAMpC,CAAA,EAAM,EAAa,WAAW,EAAY,WAG1C,MAHA,GAAA,OACA,GAAqB,GAEjB,EAAM,kCA3CH,EAAa,yBAAG,GAAa,EAAQ,YAArC,IAA2C,CAAA,EAAA,iEAAE,QAAE,eAiDxD,EAAkB,2BAGpB,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,WACrC,MADA,GAAA,OACI,EAAM,kCAtEH,EAAQ,yBAAc,GAAQ,OAA9B,IAAoC,CAAA,EAAA,iEAAE,QAAE,eA0EjD,MAAA,CAAA,EAAM,EAAa,qBAAnB,SAAA,OAEA,CAAA,EAAM,EAAM,QAAQ,mBAApB,SAAA,OACA,CAAA,EAAO,EAAM,0BAMX,EAAY,EACZ,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,yIACF,GAAI,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,iDASX,EAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIT,EAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAG,EAAG,EAAK,aAAc,EAAK,YAAa,EAC3C,cAHF,EACF,EAAA,OAGJ,EAAS,EAAiB,GAC1B,EAAU,EAAiB,GAC3B,EAAgB,EAAiB,GAG7B,EAAe,GACf,EAAM,QACN,GAAK,gBAAkB,MAAQ,EAAK,eAAe,OAAS,GAA5D,MAAA,CAAA,EAAA,GAEF,GADA,EAAe,GACX,EAAK,eAAe,SAAW,EAEjC,EAAY,EAAK,eAAe,GAChC,EAAY,EAAK,eAAe,OAC3B,MAAI,GAAK,eAAe,SAAW,EAClC,GAAI,IACN,iEAEE,GAAI,GACN,0GAEG,GAAK,eAAc,iBAGtB,SAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAW,EAAW,KACtB,KACA,EAAgB,WAJlB,SACF,EAAA,OAIJ,EAAO,EAAgB,GACvB,EAAO,EAAgB,GACvB,EAAS,EAAK,OAAO,gBAEhB,AACH,EAAK,iBAAmB,MAAQ,EAAK,gBAAkB,GACvD,EAAK,gBAAkB,EACzB,GAAe,GAET,EACF,KAAK,MAAM,EAAO,GAAG,MAAM,GAAM,GAAI,EAAK,kBACxC,EAAoB,EAAO,GAAG,MAAM,GAC1C,EAAO,GAAY,EAAQ,EAAS,GACpC,EAAS,GAAY,EAAQ,EAAG,GAChC,EAAO,GAAY,EAAS,EAAS,GACrC,EAAU,GAAY,EAAS,EAAG,GAGlC,EAAS,EAAK,OAAO,IAGhB,AAAI,EAAK,iBAAmB,MACjC,GAAe,qBAIX,SAAM,EAAO,OAAO,GAAS,OAAO,GAE1C,EAAM,mCAcA,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAW,OACX,EAAe,OACnB,AAAI,EACF,GAAM,mBACN,EAAc,EAAM,aACpB,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,MAEzD,GAAc,KACd,EAAS,GACT,EAAkB,EAAU,SAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YAChD,CAAA,EAAM,GACd,EAAO,EAAe,EAAK,EAAW,EAAW,EAAK,OACtD,EAAK,QAAS,EAAW,EAAa,EAAQ,EAAK,QACnD,EAAiB,EAAK,aAAc,KAAM,cAHxC,SAAM,EAAA,OAIZ,CAAA,EAAO,UAEP,SAAM,WAAa,GAEnB,GAAkB,EAAQ,GAC1B,GAAkB,EAAS,GAC3B,GAAkB,EAAkB,GACpC,GAAkB,EAAkB,GACpC,AAAI,GAAiB,MACnB,EAAA,QAAY,yCAYyB,GACzC,GAAM,GAAiB,GACvB,AAAI,YAAmB,GAAA,QACrB,GAAU,CAAC,IAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAS,EAAQ,GACvB,GAAI,EAAO,OAAS,EAClB,EAAK,KAAK,GAAW,EAAQ,SACxB,GAAI,EAAO,OAAS,EACzB,KAAM,IAAI,OACN,yEAGJ,EAAK,KAAK,IAGd,MAAO,eAgBL,EACA,GACF,GAAI,GAAW,KACb,OAEF,GAAM,GAAyB,GAC/B,GAAI,YAAsB,GAAA,OACxB,EAAa,KAAK,EAAW,YACpB,MAAM,QAAQ,GACvB,EAAW,QAAQ,SAAA,GAAK,MAAA,GAAa,KAAK,EAAE,cACnC,GAAc,KAEvB,OAAW,KAAQ,IACjB,GAAM,GAAY,EAAW,GAC7B,EAAa,KAAK,EAAU,IAIhC,GAAM,GAA6B,GACnC,GAAI,YAAmB,GAAA,OACrB,AAAI,EAAa,QAAQ,EAAQ,MAAQ,IACvC,EAAiB,KAAK,WAEf,MAAM,QAAQ,GACvB,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAa,QAAQ,EAAE,MAAQ,IACjC,EAAiB,KAAK,aAGjB,GAAW,KAEpB,OAAW,KAAQ,IACjB,GAAM,GAAS,EAAQ,GACvB,AAAI,EAAa,QAAQ,EAAO,MAAQ,IACtC,EAAiB,KAAK,GAK5B,EAAiB,QAAQ,SAAA,GACvB,AAAK,EAAE,YACL,EAAE,YCrnBR,YA+C6B,GAE3B,MAAO,aAAa,GAAA,mBAMM,GAE1B,MAAO,OAAM,QAAQ,eAMI,GAEzB,MAAO,CAAC,GAAa,IAAM,CAAC,GAAY,eAetC,EAAqD,EACrD,EAAkB,EAAuB,GAC3C,GADoB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACvC,GAAS,MAAQ,EAAM,SAAW,GAGpC,GAAI,GAAQ,MACV,GAAI,GAAoB,GACxB,GAAI,GAAY,IAAU,EAAkB,OAAS,EACnD,EAAoB,WACX,GAAW,IACpB,OAAW,KAAO,GAChB,GAAI,EAAK,eAAe,IACtB,EAAoB,GACpB,WAKJ,GAAoB,GAEtB,GAAI,EACF,KAAM,IAAI,GACN,6BAA6B,EAAe,sBAC5C,YAAW,IAGnB,MAAO,GAET,GAAI,GAAQ,KACV,MAAO,GAAM,IAAI,SAAA,GAAQ,MAAA,QAG3B,GAAI,GACJ,GAAI,GAAW,IACb,EAAO,EACP,EAAS,GACT,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAK,IAAS,KAChB,KAAM,IAAI,GACN,yBAAyB,EAAI,iCAC7B,IAAG,IAET,EAAO,KAAK,EAAK,aAEV,GAAY,IAErB,GADA,EAAO,EACH,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,oCAAmC,EAAM,OAAM,oBAC/C,iDAAgD,IAEtD,EAAS,OAGT,GADA,EAAO,EACH,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,aAAa,EAAe,YAAY,EAAM,OAAM,eACpD,2DACI,EAAK,QAEf,EAAS,CAAC,GAMZ,GAHA,EAAS,GAA2B,GAGhC,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,EAAM,QAErB,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EAEd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MAAQ,GAAU,GAAK,IAAQ,EAC3C,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,mBAAkB,EAAO,GAAE,gCAC3B,KAAI,EAAM,MAAK,QAK3B,MAAO,eAWL,EAAkB,EAAmB,GACvC,GAAM,GAAO,GAAO,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,MAAM,MACpD,EAAK,OACL,GAAM,GAAO,GAAO,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,MAAM,MAGvD,GAFA,EAAK,OAED,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,mFAEA,IAAG,KAAK,UAAU,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,WAElD,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,oFAEA,IAAG,KAAK,UAAU,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,WAErD,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,CAAC,EAAA,KAAK,YAAY,EAAM,GAChE,KAAM,IAAI,GACN,kEACA,mBAAkB,EAAK,GAAE,wBAAwB,EAAK,GAAE,YACxD,cAaR,YACI,EAAmB,EAA2B,GAMhD,OAJM,GAAY,CAChB,GAAyB,GACzB,IAEO,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAI,EAAQ,GACZ,EAAO,EAAQ,GACf,EAAQ,EAAa,GAC3B,GAAI,GAAQ,KACV,SAEF,GAAI,IAAS,IACP,EAAE,MAAM,EAAE,MAAM,OAAS,KAAO,EAClC,KAAM,IAAI,GACN,2CAA2C,EAAE,MAAK,4JAO1D,GAAI,EAAU,QAAQ,KAAU,GAG9B,OAFM,GAAe,EAAE,MAAM,MAAM,GAC7B,EAAc,EAAM,MAAM,GACvB,EAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAY,EAAa,GACzB,EAAS,EAAY,GAC3B,GAAI,GAAU,MAAQ,IAAc,EAClC,KAAM,IAAI,GACN,8BAA8B,EAAE,MAAK,sBACrC,oBAAmB,EAAK,uCACxB,2DAiCd,YACI,EAAuB,EAAiB,EACxC,EAAuB,GAAvB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACzB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,wCAAuC,EAAM,OAAM,eACnD,qBAAoB,EAAK,OAAM,iBAErC,EAAS,OAET,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,qBAAqB,EAAM,OAAM,IAAI,EAAe,mEAEjD,MAAK,UAAU,EAAK,OAAM,MAEnC,EAAS,CAAC,GAGZ,GAAI,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,KAAK,UAAU,EAAM,SAEpC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EACd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MACR,IAAW,EACb,KAAM,IAAI,GACN,uBAAuB,EAAe,cACnC,GAAM,GAAE,kBAAkB,KAAK,UAAU,EAAO,IAAG,SACtD,yBAAwB,KAAK,UAAU,EAAM,OAAM,oBAsB/D,EAEA,GACF,GAAI,GAAW,MAAQ,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAClE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,KAGjC,GAAI,GAEJ,GAAI,MAAO,IAAY,UAAY,MAAO,IAAY,WACpD,EAAiB,CAAC,WACT,MAAM,QAAQ,IAAY,MAAO,IAAY,SACtD,EAAiB,MAGjB,MAAM,IAAI,WACN,+DACA,uCAAsC,IAG5C,GAAI,MAAM,QAAQ,GAEhB,MAAO,GAAY,IACf,SAAA,GAAQ,MAAA,KAIZ,OADM,GAAqD,GACxC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACT,EACA,EAAe,eAAe,GAAQ,EAAe,GAAQ,GACjE,AAAK,MAAM,QAAQ,IACjB,GAAgB,CAAC,IAEnB,EAAc,KAAK,GAErB,MAAO,GA6DX,GAAM,IAA2B,8BAcA,EAAA,EAAA,GA4C/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WAAa,KAsCpB,SAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAGF,GAHE,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC1D,CAAC,KAAK,MACR,KAAM,IAAI,GACN,gLAIN,GAAa,KAAM,EAAY,EAAW,IAa5C,EAAA,UAAA,QAAA,SAAQ,GAAR,GAAA,GAAA,KAME,GALA,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAEd,KAAK,KAAO,EAAK,KAEb,MAAO,GAAK,WAAc,SAC5B,KAAK,WAAa,GAAwB,EAAK,WAC/C,KAAK,iBAAmB,QAExB,GAAI,CAAE,GAAK,oBAAqB,GAAA,WAC9B,KAAM,IAAI,GACN,+DAEN,KAAK,WAAa,EAAK,UACvB,KAAK,iBAAmB,GAO1B,GAAI,GAAkC,GACtC,GAAI,CAAC,MAAM,QAAQ,EAAK,OAAS,MAAO,GAAK,MAAS,UAClD,MAAO,GAAK,MAAS,YACvB,EAAK,KAAO,EAAK,KACjB,OAAW,KAAQ,GAAK,KACtB,GAAI,KAAK,YAAY,QAAQ,KAAU,GACrC,KAAM,IAAI,GACN,sCAAsC,EAAI,MAC1C,sCAAqC,KAAK,cAGlD,OAAmB,GAAA,EAAA,EAAA,KAAK,YAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAK,KAAK,IAAS,MACrB,QAAQ,KACJ,WAAW,EAAI,4GAEf,oBAAmB,EAAI,qBAE7B,EAAc,KAAK,GAAW,EAAK,KAAK,cAEjC,MAAM,QAAQ,EAAK,OAC5B,GAAI,EAAK,KAAK,SAAW,KAAK,QAAQ,OACpC,KAAM,IAAI,GACN,+DACA,gCAA+B,KAAK,QAAQ,OAAM,gBAClD,wBAAuB,EAAK,KAAI,MAEtC,GAAM,GAAY,EAAK,KACvB,EAAgB,EAAU,IAAI,SAAA,GAAK,MAAA,IAAW,UAE9C,GAAM,GAAe,GAAW,EAAK,MACrC,KAAK,QAAQ,QAAQ,SAAA,GACnB,EAAc,KAAK,KAIvB,KAAK,cAAgB,EAErB,KAAK,gBAAkB,GACvB,KAAK,iBAAmB,GACxB,KAAK,YAAc,GACnB,OAAS,GAAI,EAAG,EAAI,KAAK,QAAQ,OAAQ,EAAE,GAEzC,GAAM,GAAQ,KAAK,qBAAqB,GAClC,EAAO,KAAK,YAAY,GAC9B,KAAK,gBAAgB,KAAK,GAC1B,KAAK,iBAAiB,KAAK,GAC3B,KAAK,YAAY,KAAK,KAAK,cAAc,IAK3C,GAAM,GAA8B,GAGpC,KAAK,QAAU,EAAK,QAEpB,KAAK,aAAe,CAAC,QACrB,KAAK,eAAiB,GAMtB,GAAU,OAAQ,WAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,GACzC,GAAI,EAAkB,QAAQ,KAAO,GACnC,SAIF,GAAM,GAAe,EAAK,cAAc,GACxC,AAAI,EAAK,QAAQ,OAAS,GACxB,GAAK,eAAe,KAAK,CAAC,EAAc,IACxC,EAAK,aAAa,KAAK,EAAK,YAAY,GAAK,aAQnD,GAAM,GAAgB,GAAe,EAAK,QAAS,KAAK,aAMlD,EACF,SAAC,EAAqB,EACrB,GACC,AAAI,EAAK,YAAY,OAAS,GAC5B,GAAa,EAAK,YAAY,GAAe,IAAM,GAErD,EAAK,aAAa,KAAK,GACvB,EAAK,eAAe,KAAK,CAAC,EAAc,KAG9C,GAAU,SAAU,WAClB,mBAAS,GACP,GAAI,EAAkB,QAAQ,KAAO,oBAGrC,GAAM,GAAgB,EAAc,GAI9B,EAAgB,SAAC,GAOrB,OANM,GAAmB,GACrB,EACA,EACA,aAGO,GACT,GAAI,MAAO,IAAW,UAClB,CAAC,WAAY,MAAO,eAAgB,MAAM,QAAQ,KAC9C,IACN,GAAM,GAAc,EAAK,qBAAqB,GAE9C,AAAI,EAAY,EAAY,OAAS,KAAO,GACxC,EAAK,cAAc,KAAO,GAE5B,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAEL,AACH,EAAK,cAAc,KACnB,GAGF,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAIV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAGZ,GAAI,GAAM,OACV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAS,MACJ,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAS,MAGX,EAAmB,EACnB,EAAa,EAAmB,OAEhC,GAAM,GAAW,GAAY,GAE7B,EAAmB,EACnB,EACI,EAAmB,GAA4B,GAIrD,GAAI,GACJ,GAAU,EAAY,WACpB,EAAe,IAEjB,EAAa,EAAG,EAAY,IAtDT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,KAAN,KA0Db,EAAc,IAzEP,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,IAAlC,KAgFX,KAAK,0BAA4B,KAAK,kBAY9B,EAAA,UAAA,iCAAV,WACE,GAAI,KAAK,2BAA6B,KACpC,OAEF,AAAI,KAAK,iBAAiB,SACtB,KAAK,0BAA0B,QACjC,QAAQ,KACJ,sJAqCR,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIf,GAAM,GAAiB,GACjB,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GACrD,IAGE,GAAM,GAAM,EAAiB,GAAG,OAAO,EAAiB,IACxD,KAAK,mBACL,GAAM,GAAI,KAAK,aACT,EACF,KAAK,SAAS,EAAG,EAAK,EAAW,EAAK,QAAS,EAAK,OACxD,MAAO,IAAiB,WAExB,GAAkB,EAAiB,GAAI,GACvC,GAAkB,EAAiB,GAAI,KA0BrC,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,YAAK,mBACL,CAAA,EAAO,GAAgB,KAAM,EAAS,SAahC,EAAA,UAAA,gBAAR,SACI,EAAsB,EAAoB,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,SACF,GAAI,GACJ,GAAI,GAAS,MAEX,GADA,EAAa,KACT,GAAa,KACf,KAAM,IAAI,GACN,MAAM,EAAS,gDACf,oBAAmB,YAEhB,GAAO,KAChB,AAAI,MAAM,QAAQ,GAChB,EAAa,EAAI,GAAG,MAAM,GAE1B,EAAa,EAAI,MAAM,OAGzB,MAAM,IAAI,GACN,yDACG,GAAS,yBAElB,MAAO,IAUT,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,GAAI,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAC/C,KAAM,IAAI,GACN,sDAGN,GAAM,GAAiB,MAAM,QAAQ,GAC/B,EACD,EAAiB,EAAsB,CAAC,GACvC,EAAwB,KAAK,wBAAwB,GAGrD,EAAW,GAAI,IAIrB,GAHA,AAAI,YAAkB,GAAA,QACpB,GAAS,CAAC,IAER,MAAM,QAAQ,IAChB,GAAI,EAAO,SAAW,KAAK,OAAO,OAChC,KAAM,IAAI,GACN,kCAAkC,EAAO,OAAM,uDAE/C,KAAI,KAAK,OAAO,OAAM,OAE5B,OAAS,GAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,KAAK,OAAO,GAAI,EAAO,QAGtC,QAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAc,EAAO,EAAM,MACjC,GAAI,GAAe,KACjB,KAAM,IAAI,GACN,8CAA8C,EAAM,MAE1D,EAAS,IAAI,EAAO,GAKxB,GAAM,GAAiB,GAAQ,EAAuB,GACtD,MAAO,GAAiB,EAAiB,EAAe,IAMlD,EAAA,UAAA,wBAAR,SAAgC,GAK9B,OAHM,GACF,GAAa,KAAM,EAAoB,QACvC,EAAmB,EAAoB,OACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EACF,MAAM,QAAQ,EAAM,QAAU,EAAM,OAAS,CAAC,EAAM,QAClD,EAAmB,EAAa,IAAI,SAAA,GAAU,MAAA,GAAO,OAClD,EAAI,EAAG,EAAI,EAAoB,OAAQ,EAAE,GAChD,GAAM,GAAQ,EAAiB,QAAQ,EAAoB,IAK3D,GAJA,AAAI,IAAU,IACZ,GAAsB,GAAK,EAAa,GACxC,KAEE,IAAqB,EACvB,MAGJ,GAAI,IAAqB,EACvB,MAIJ,GAAI,EAAmB,GACrB,GAAM,GAA2B,GACjC,QAAsB,QAAQ,SAAC,EAAQ,GACrC,AAAI,GAAU,MACZ,EAAe,KAAK,EAAoB,MAGtC,GAAI,GACN,mDACA,IAAG,KAAK,UAAU,KAExB,MAAO,IAgBD,EAAA,UAAA,YAAR,SAAoB,EAAsB,EAAgB,GAA1D,GAAA,GAAA,KAA0C,MAAA,KAAA,QAAA,GAAA,IAAgB,AAAA,IAAA,QAAA,GAAA,IAEjD,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,GACxC,GAAI,EACF,KAAM,IAAI,IACN,iDAYN,OAJM,GAAU,GAAY,EAAY,GAClC,EAA0B,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,gBAGlD,GACP,GAAM,GAAY,EAAA,KAAS,WACzB,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAG/B,EAAW,GAAY,EAAK,EAAY,GAGxC,EAAQ,GACd,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,EAAE,EACrC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAS,SAGnD,GAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,IAE1C,GAAM,GAAW,GAAI,IAAS,GAC9B,MAAO,IAAQ,EAAK,QAAS,KAE/B,EAAU,QAAQ,SAAC,EAAU,GAAM,MAAA,GAAY,GAAG,KAAK,MApBhD,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,IAA/C,GAsBT,MAAO,IACH,EAAY,IAAI,SAAA,GAAW,MAAA,GAAA,OAAW,EAAS,SA+BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAkB,GAA2B,GACnD,GACI,EAAiB,KAAK,WAAY,KAAK,gBAAiB,IAC5D,IAKE,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,UAAe,GACR,KAAK,YAAY,EAAiB,WAEzC,GAAkB,EAAiB,KAmBvC,EAAA,UAAA,eAAA,SAAe,GACb,GAAe,EAAG,KAAK,WAAY,KAAK,gBAAiB,IAGzD,GAAM,GAAa,OAAM,QAAQ,GAAK,EAAE,GAAK,GAAG,MAAM,GACtD,MAAO,MAAK,YAAY,EAAG,IAGnB,EAAA,UAAA,sBAAV,SACI,EACA,EAAkD,EAClD,GAEF,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,IACN,gGAIN,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,KAAK,iBAAiB,OAAQ,EAAE,GAClD,GAAM,GAAc,KAAK,iBAAiB,GACpC,EAAS,KAAK,YAAY,GAChC,AAAI,IAAW,GACb,EAAa,KACT,EAAY,MAAM,EAAG,EAAY,OAAS,GAAG,OAAO,CAAC,KAGzD,EAAa,KAAK,GAWtB,GARA,EAAI,GACA,EAAG,KAAK,eAAgB,KAAK,gBAAiB,GAAO,SACzD,EAAI,GACA,EAAG,KAAK,gBAAiB,EAAc,GAAO,UAElD,GAAkB,EAAG,GAErB,GAAgC,EAAG,KAAK,YAAa,KAAK,kBACtD,KAAK,UAAY,GAAa,MAAQ,EAAY,GAChD,EAAE,GAAG,MAAM,GAAK,IAAc,EAChC,KAAM,IAAI,GACN,mHAEG,GAAS,YAAY,EAAE,GAAG,MAAM,GAAE,gBAG7C,MAAO,CAAC,EAAG,IAGG,EAAA,UAAA,oBAAhB,SACI,EACA,EACA,EACA,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAA,2GAKF,GAHM,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GAD9C,EAAU,EAAA,GAAE,EAAU,EAAA,GAGzB,GAAgB,KAClB,KAAM,IAAI,OAAM,0CAGd,EAAkC,MAClC,IAAe,MAAf,MAAA,CAAA,EAAA,GACI,EACF,GAAwB,EAAa,KAAK,aAC9C,EAAwB,GACf,EAAI,yBAAG,GAAI,EAAa,OAC/B,GAAA,GAAA,GAAsB,KAClB,CAAA,EAAM,GAAmB,EAAW,GAAI,KAAM,EAAa,MAF1B,CAAA,EAAA,UACrC,EAAA,MAAA,EAAA,CACI,EAAA,0BAFmC,QAAE,eAO7C,MAAA,CAAA,EAAO,CAAC,EAAY,EAAY,UAc1B,EAAA,UAAA,SAAR,SACI,EAAiC,EAAe,EAChD,EAAa,GAFjB,GAAA,GAAA,KAEI,MAAA,KAAA,QAAA,GAAA,GACK,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,EAAK,EAAW,EAAO,SACzD,EAAiB,GACvB,GAAI,EAAU,EACZ,KAAM,IAAI,IAAoB,wCAGhC,GAAI,GAAS,KACX,KAAM,IAAI,IACN,mDAIJ,OAFM,GAAU,GAAY,EAAY,GAClC,EAAa,EAAA,SAAS,GAAM,EAAG,IAC5B,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,GACtD,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,EACF,GACI,EAAY,EAAY,EAAW,GAGrC,EAAW,GAAqB,EAAK,GACrC,EAAY,EAAE,GACpB,GAAI,IAAe,EACjB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAGrB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAW,EAAU,GAC3B,EAAK,GACD,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,EAAY,KAGxD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAG/B,MAAO,MAID,EAAA,UAAA,uBAAV,WAKE,OAJM,GAAY,KAAK,aAGjB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAQ,EAAU,GACpB,EAAW,EACf,GAAI,GAAM,EAAW,GAAS,GAC5B,GAAM,GAAW,GAAM,EAAU,MAAM,EAAG,GAAI,GAC9C,GAAY,IAAI,EAElB,EAAiB,KAAK,GAExB,MAAO,IAaC,EAAA,UAAA,kBAAV,WAAA,GAAA,GAAA,KACE,MAAO,UAAC,GACN,GAAM,GAAuB,GAEvB,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAgB,EAAK,MACvB,EAAK,OAAO,OAAS,EAAK,QAAQ,OAClC,EAAK,OAAO,OAAS,EAAK,QAAQ,OAAS,GAEzC,EAA0B,GAK1B,EAAoB,WAExB,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KASjD,OAPM,GAAW,GAAI,IAAS,GACxB,EACF,GAAQ,EAAK,QAAS,EAAU,CAAC,SAAY,KAI7C,EACK,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GACpC,EAAO,EAAa,EAAQ,GAAI,EAAQ,IAC5C,AAAI,EAAc,IAAM,MACtB,GAAO,GAAoB,EAAM,EAAc,KAIjD,GAAM,GAAmB,EAAA,KAAS,GAElC,EAAW,KAAK,GAChB,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAOnC,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAI,GAAc,OAElB,GAAI,EAAK,QAAQ,OAAS,GAAK,EAAI,EAAK,QAAQ,OAC9C,EAAiB,EAAW,QAE5B,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAC3C,EACI,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAGpD,EAAA,KAAS,GAET,EAAc,KAAK,GAGrB,SAAY,EAAA,KAAS,GAGrB,EAAK,kBAAkB,QAAQ,SAAA,GAC7B,EAAY,EAAA,IAAQ,EAAW,KAG1B,GAGH,EAAY,EAAK,0BAA0B,IAC7C,SAAA,GAAS,MAAA,GAAM,SACb,EAAa,GACb,EACF,EAAK,WAAW,SAAS,EAAmB,EAAY,GAE5D,MAAO,CAAC,GAAgB,OAAO,KAS3B,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,KAAK,aAAe,SAAC,GACnB,MAAO,GAAA,KAAS,WAOd,OANM,GAAuB,GACzB,EACE,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KAKjD,OAHM,GAAW,GAAI,IAAS,GACxB,EAAU,GAAQ,EAAK,QAAS,GAE7B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAGlC,EAAe,EAAA,KAAS,EAAa,EAAQ,GAAI,EAAQ,KAC/D,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAEjC,EAAW,KAAK,GAGlB,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAErC,EACF,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAClD,EAAW,KAAK,GAElB,MAAO,OAuCP,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,MAAA,CAAA,EAAO,GAAW,KAAM,EAAG,EAAG,SA0B1B,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,MAAA,CAAA,EAAO,GAAW,KAAM,EAAS,SA0B7B,EAAA,UAAA,aAAN,SACI,EACA,qHAIqB,MAAA,CAAA,EAAM,KAAK,oBAAoB,EAAG,WAAnD,EAAiB,EAAA,OACjB,EAAS,EAAe,GACxB,EAAU,EAAe,GACzB,EAAgB,KAAK,oBACrB,EAAS,EAAc,EAAO,OAAO,IACrC,EAAuB,OACV,EAAA,yBAAA,GAAA,EAAA,OAAR,GAAI,EAAA,GACH,CAAA,EAAM,EAAK,SADE,CAAA,EAAA,UACjB,EAAI,EAAA,OACV,EAAW,KAAK,EAAE,qBAFD,wBAInB,SAAA,QAAY,GACZ,CAAA,EAAO,GAAiB,UAYhB,EAAA,UAAA,gBAAV,SAA0B,GAMxB,OALM,GAA8B,GAE9B,EAAgB,GAAU,MAAQ,EAAO,cACzC,EAAU,EAAgB,KAAK,iBAAmB,KAAK,QACvD,EAAe,KAAK,WAAW,GAC5B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAI,GAAiB,CAAC,EAAQ,GAAG,UAE/B,SAEF,EAAa,KACT,CAAC,KAAM,EAAQ,GAAG,aAAc,OAAQ,EAAa,KAE3D,MAAO,IAiCT,OAAA,eAAI,EAAA,UAAA,eAAY,KAIhB,WACE,MAAO,MAAK,mBALd,SAAiB,GACf,KAAK,cAAgB,mCAOvB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,AAAI,KAAK,aAAe,GACtB,MAAK,WAAa,EAClB,KAAK,iBAAmB,qCAI5B,EAAA,UAAA,QAAA,WACE,GAAM,GAAS,EAAA,UAAM,QAAO,KAAA,MAC5B,GAAI,EAAO,uBAAyB,GAAK,KAAK,WAAa,MACvD,KAAK,kBACP,GAAM,GAAmC,EAAA,SAAa,WACtD,KAAK,WAAW,UAChB,EAAO,sBACH,EAAmC,EAAA,SAAa,WAEtD,MAAO,IAGD,EAAA,UAAA,mBAAR,WAEE,GAAI,GAEJ,GAAI,MAAO,MAAK,MAAS,SACvB,EAAY,GAAY,KAAK,cACpB,MAAM,QAAQ,KAAK,OAC5B,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,OAAM,sDAGpB,EAAa,KAAK,KAAkB,IAAI,SAAA,GAAQ,MAAA,IAAY,UAG5D,GAAM,GAAc,OAAO,KAAK,KAAK,MACrC,EAAY,GAGZ,OAFM,GACF,KAAK,KACgB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,MAAO,GAAO,IAAgB,SAChC,EAAU,GACN,GAAY,EAAO,QAEvB,MAAM,IAAI,OAAM,uDAItB,MAAO,IAGD,EAAA,UAAA,qBAAR,WAEE,GAAI,MAAO,MAAK,SAAY,UACxB,MAAO,MAAK,SAAY,WAC1B,MAAO,CAAC,GAAY,GAA4B,KAAK,WAChD,GAAI,MAAM,QAAQ,KAAK,SAC5B,MAAO,MAAK,QAAQ,IAChB,SAAA,GAAU,MAAA,IAAY,GAA4B,MAEtD,GAAM,GAAyD,GAC/D,OAAW,KAAO,MAAK,QACrB,EAAmB,GACf,GAAY,GAA4B,KAAK,QAAQ,KAE3D,MAAO,IAID,EAAA,UAAA,kBAAV,WACE,MAAO,CACL,KAAM,KAAK,qBACX,QAAS,KAAK,uBACd,iBAAkB,CAChB,WAAY,KAAK,UAAU,eAC3B,OAAQ,KAAK,UAAU,eAQ7B,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,EAAe,kBAAoB,KACrC,KAAM,IAAI,OAAM,gDAElB,GAAI,EAAe,cAAgB,KACjC,KAAM,IAAI,OAAM,8CAElB,GAAI,EAAe,oBAAsB,KACvC,KAAM,IAAI,OAAM,oDAGlB,GAAM,GAAW,GAAoB,EAAe,kBAE9C,EAAY,GAAY,GAE1B,EACJ,GAAI,MAAO,GAAe,MAAS,SACjC,EAAO,GAAY,EAAe,cACzB,MAAM,QAAQ,EAAe,MACtC,EAAO,EAAe,KAAK,IAAI,SAAA,GAAa,MAAA,IAAY,aAC/C,EAAe,MAAQ,MAChC,EAAO,GACP,OAAW,KAAO,GAAe,KAC/B,EAAK,GAAO,GAAY,EAAe,KAAK,IAIhD,GAAI,GACJ,GAAI,MAAM,QAAQ,EAAe,SAC/B,EAAU,EAAe,QAAQ,IAAI,SAAA,GAAU,MAAA,IAAY,aAClD,EAAe,SAAW,MACnC,EAAU,GACV,OAAW,KAAO,GAAe,QAC/B,EAAQ,GAAO,GAAY,EAAe,QAAQ,IAItD,KAAK,QAAQ,CAAC,KAAI,EAAE,QAAO,EAAE,UAAS,KAoFlC,EAAA,UAAA,KAAN,SAAW,EAAmC,+HAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,GACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,GACN,gHAKF,MAAA,CAAA,EAAM,EAAA,GAAG,cAAc,KAAK,gBAAgB,kBAD1C,GACF,EAAA,OAEE,EAAe,GACf,EAAgB,KAChB,EAAc,KAAK,OAAO,EAAW,GACrC,EAAoC,CACxC,cAAe,EACf,OAAQ,GACR,YAAa,8BAA8B,GAC3C,YAAa,MAGT,EAAmB,GAAU,KAAO,GAAQ,EAAO,iBACrD,GAAoB,KAAK,WAAa,KACxC,GAAe,eAAiB,KAAK,oBAC/B,EAAa,YAET,EAAA,GAAA,EAAA,IAAG,cAAc,CAAA,EAAM,KAAK,UAAU,eAJ9C,CAAA,EAAA,UAIE,MAAA,CAAA,EAAM,EAAA,MAAA,EAAA,CAAiB,EAAA,OAAmC,YADxD,EACF,EAAA,OADS,EAAmB,EAAA,KAAS,EAAoB,EAAA,MAE7D,GAAA,EAAmB,OAAM,KAAI,MAAA,EAAI,GACjC,EAAmB,KAAO,EAAA,GAAG,wBACzB,CAAC,EAAmB,KAAM,qBAGhC,MAAI,MAAK,qBAAuB,MAExB,GAAY,GAClB,GAAyB,KAAK,oBAAqB,KAAK,KAAM,GAC9D,EAAe,oBAAsB,KAAK,qBAG5C,EAAe,WAAa,EAAmB,KAC/C,EAAe,YAAc,EAAmB,MAChD,CAAA,EAAO,EAAa,KAAK,UAW3B,EAAA,UAAA,uBAAA,SAAuB,GACrB,GAAyB,EAAqB,KAAK,MACnD,KAAK,oBAAsB,GAc7B,EAAA,UAAA,uBAAA,WACE,MAAO,MAAK,qBAv4CP,EAAA,UAAY,QAy4CrB,GA74CiC,IA84CjC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DACS,SAAA,UAAY,aACrB,GAFgC,IAGhC,EAAA,cAAc,cAAc,IC72D5B,YA8DI,EACA,uHACF,AAAM,iBAAmB,IACvB,GAAwB,CAAC,cAAe,IAE1C,EAAwB,EAEpB,EAAgB,EAAsB,cAC1C,AAAI,EAAc,cAAmB,MAMnC,GAAgB,EAAc,cAE1B,EACF,GAAoB,GAClB,EAAQ,GAAY,EAAU,GAEhC,EAAsB,iBAAmB,KAItB,CAAA,EAAM,EAAA,GAAG,YAC1B,EAAsB,gBAAiB,EAAsB,WAC7D,EAAM,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,iBANrC,CAAA,EAAA,UAUF,IANM,EAAe,EAAA,OAKf,EAAqC,GAC3C,EAAA,EAAqB,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,IAAV,EAAM,EAAA,GACf,EAAmB,EAAO,cACtB,EAAa,EAAO,cAG1B,EAAM,YAAY,GAElB,EAAA,QAAQ,oBAEV,MAAA,CAAA,EAAO,oBAsIL,EACA,4EAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,MAAO,IAAoB,UAE7B,GADM,EAAW,EAAA,GAAG,gBAAgB,EAAiB,GACjD,EAAS,SAAW,EAKtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAiB,YAC5C,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAe,MAE7B,EAAkB,EAAS,GAE7B,MAAA,CAAA,EAAO,GAA6B,EAAiB,OAAW,oBAc9D,EAAuB,EACvB,mHAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,EAAQ,MAAQ,KAClB,KAAM,IAAI,GACN,iHAGY,MAAA,CAAA,EAAM,EAAQ,kBAA1B,EAAY,EAAA,OACd,EAAgB,EAAU,cAC9B,AAAI,EAAc,cAAmB,MACnC,GAAgB,EAAc,cAG1B,EAAS,EAAQ,QAAU,KAAO,GAAO,EAAQ,OAMjD,EACF,EAAU,YAAc,MAAQ,EAAU,aAAe,MAAQ,EAC/D,EACF,GACI,GAAoB,GACpB,EAAe,GAEjB,EAAiB,EAAU,eACjC,AAAI,GAAkB,MACpB,EAAM,mBAAmB,GAE3B,AAAI,EAAU,qBAAuB,MACnC,EAAM,uBAAuB,EAAU,sBAIrC,GAAU,YAAc,MAAxB,MAAA,CAAA,EAAA,GAEF,GAAI,EAAU,aAAe,KAC3B,KAAM,IAAI,GACN,wHAIA,GAAmC,GACrC,EAAU,WAAY,EAAU,aAD7B,EAAY,EAAA,aAAE,EAAgB,EAAA,iBAErC,EAAM,YAAY,EAAc,GAE5B,EAAM,WAAa,MAAQ,EAAiB,OAAS,EACvD,CAAA,EAAM,EAAM,UAAU,WAAW,IAD/B,CAAA,EAAA,UACF,EAAA,wBAIF,EAAA,QAAQ,GACR,EAAA,QAAQ,EAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,2BAEtC,MAAA,CAAA,EAAO,QAGT,YACI,EAAqB,GAEvB,GAAM,GAAc,EAAA,GAAG,cAAc,EAAQ,GACvC,EAA+B,GAC/B,EAAkC,GACxC,SAAM,QAAQ,SAAA,GACZ,AAAI,EAAK,QAAU,YACjB,EAAiB,KAAK,CAAC,KAAM,EAAK,KAAM,OAAQ,EAAY,EAAK,QAEjE,EAAa,EAAK,MAAQ,EAAY,EAAK,QAGxC,CAAC,aAAY,EAAE,iBAAgB,sBAwCR,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CAAC,OAAQ,GAAI,QAAS,MAAI,KAUhC,GATA,EAAO,GAAQ,GAEf,EAAK,UAAY,GACjB,EAAK,MAAQ,GAGb,EAAK,KAAQ,EAAK,MAAQ,KAAQ,EAAK,KAAO,GAAO,eAGjD,EAAK,QAAU,KACjB,OAAoB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,IAAI,YAOP,SAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAQ,EAAM,aAAa,GAAG,cAAc,GAAG,MACrD,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,GAAI,IACtB,KAAM,IAAI,GACN,kDACG,GAAM,KAAI,uBACV,GAAM,aAAa,GAAG,aAAa,GAAG,MAAK,OAyBtD,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GACF,YAAiB,IAAc,YAAiB,IAChD,EACJ,GAAI,GAEF,GADA,EAAa,EACT,EAAW,QAAQ,SAAW,EAChC,KAAM,IAAI,GACN,yHAKN,GAAI,EAAW,OAAO,SAAW,EAC/B,KAAM,IAAI,GACN,uHAOR,GAAI,KAAK,QAAQ,SAAW,GAE1B,GAAI,EAAM,aAAa,SAAW,GAEhC,GAAI,EAAM,iBAAmB,KAC3B,KAAM,IAAI,GACN,iGAIN,GAAM,GAAI,GAAM,CACd,WAAY,EAAM,gBAClB,MAAO,EAAM,MACb,KAAM,EAAM,KAAO,WAIrB,EAAM,MAAM,GAGd,GAAI,EACF,KAAK,QAAU,EAAW,QAC1B,KAAK,OAAS,EAAW,YAEzB,GAAI,EAAM,aAAa,SAAW,EAChC,KAAM,IAAI,GACN,2DACA,yDACI,EAAM,KAAI,KACd,cAAa,EAAM,aAAa,OAAM,0BACtC,gBAGN,GAAI,EAAM,aAAa,GAAG,cAAc,SAAW,EACjD,KAAM,IAAI,GACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,EAAM,aAAa,GAAG,cAAc,IACpD,KAAK,OAAS,GAAgB,KAAK,QAAQ,IAG7C,KAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,KACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,KAAK,OACnB,cAAe,KAAK,QAEpB,WAAY,GAA2B,KAAM,KAAK,OAAO,QACzD,YAAa,CAAC,MACd,YAAa,KAAK,OAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QACpC,aAAc,KAAK,QAAQ,GAAG,aAGhC,GAAM,GAAe,EAAM,MAAM,KAAK,QAAQ,IAC9C,GAAI,MAAM,QAAQ,GAChB,KAAM,IAAI,WACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,GAEhB,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,OAGvD,KAAK,OAAO,KAAK,GACjB,KAAK,MAAQ,IAQf,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,OAAO,SAAW,EACzB,KAAM,IAAI,WAAU,qCAItB,GADA,KAAK,OAAO,MACR,KAAK,OAAO,SAAW,EACzB,KAAK,QAAU,GACf,KAAK,aAAe,GACpB,KAAK,cAAgB,QAErB,GAAM,GAAiB,KAAK,OAAO,OAAS,EAC5C,KAAK,OAAO,GAAgB,cAAgB,GAC5C,KAAK,QAAU,CAAC,KAAK,OAAO,GAAgB,QAE5C,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,SAIzD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,KAAK,EAAQ,IAGjC,EAAA,UAAA,MAAA,SAAM,GAKJ,GAFA,GAAmB,GAEf,KAAK,OAAO,SAAW,GAAK,KAAK,QAAQ,SAAW,EACtD,KAAM,IAAI,WACN,4EAIN,KAAK,MAAQ,GAAI,IAAY,CAC3B,OAAQ,KAAK,OACb,QAAS,KAAK,QAAQ,GACtB,KAAM,KAAK,KAAO,WAEpB,KAAK,MAAM,UAAY,KAAK,UAG5B,KAAK,gBAAkB,KAAK,MAAM,gBAElC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,uBAAyB,KAAK,MAAM,uBACzC,KAAK,yBAA2B,KAAK,MAAM,yBAC3C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,wBAA0B,KAAK,MAAM,wBAC1C,KAAK,0BAA4B,KAAK,MAAM,0BAC5C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,WAAa,KAAK,MAAM,WAG7B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,WACE,MAAK,MAAK,OACR,KAAK,QAEA,EAAA,UAAM,YAAW,KAAA,OAiC1B,EAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAAA,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC9D,AAAK,KAAK,OACR,KAAK,QAEP,EAAA,UAAM,QAAO,KAAA,KAAC,EAAY,EAAW,IASvC,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,OAAS,MAChB,KAAK,QAEP,KAAK,MAAM,WAAW,IAmCxB,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GACF,GADE,AAAA,IAAA,QAAA,GAAA,IACE,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAO,MAAK,MAAM,SAAS,EAAG,EAAG,IAyB7B,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAA,CAAA,EAAO,KAAK,MAAM,gBAAgB,EAAS,SA8B7C,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC1B,AAAI,KAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,QAAQ,EAAG,IAU/B,EAAA,UAAA,eAAA,SAAe,GACb,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,eAAe,IAQnC,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,QACL,KAAK,MAAM,QAAQ,GACnB,KAAK,WAAa,KAAK,MAAM,UAE7B,KAAK,iBAAoB,KAAK,MAAc,iBAC5C,KAAK,KAAO,KAAK,MAAM,KACvB,KAAK,QAAU,KAAK,MAAM,QAG1B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,aAAe,KAAK,MAAM,cAIjC,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAO,OAAY,KAAK,MAAM,eAGrD,SAAc,GACZ,KAAK,MAAM,UAAY,mCAkCnB,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,IAAI,EAAG,EAAG,SAwFxB,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,WAAW,EAAS,SA0BlC,EAAA,UAAA,aAAN,SACI,EACA,sEAEF,MAAA,CAAA,EAAO,KAAK,MAAM,aAAa,EAAG,SAK7B,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GACA,EAA6C,GACjD,GAAI,YAAkB,QACpB,GAAI,CAAE,GAAO,GAAG,WAAa,OACzB,EAAO,GAAG,YAAiB,QAC7B,KAAM,IAAI,GAAW,kDAEvB,EAAc,MAEd,GAAA,KAAK,OACD,EAAO,QAAa,KACpB,WACI,MAAA,wHAER,EAAc,EAAO,OACrB,MAAO,GAAO,OACd,EAAmB,EAGrB,GAAM,GAAQ,GAAI,GAAI,GACtB,GAAI,CAAE,aAAiB,IACrB,KAAM,IAAI,IACN,yDAAyD,GAE/D,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAA0C,OAC1C,EAAQ,GACI,EAAkC,EAClC,GAClB,AAAI,GACF,EAAM,6BAA6B,IAErC,EAAM,IAAI,GAEZ,MAAO,IA+BT,OAAA,eAAI,EAAA,UAAA,eAAY,KAWhB,WACE,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,MAAO,MAAK,MAAM,kBAjBpB,SAAiB,GAGf,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,KAAK,MAAM,aAAe,mCAe5B,EAAA,UAAA,UAAA,WAME,OADM,GAAqC,GACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAiC,GACvC,EAAK,UAAe,EAAM,eAC1B,EAAK,OAAY,EAAM,YACvB,EAAO,KAAK,GAEd,MAAO,CAAC,KAAM,KAAK,KAAM,OAAM,IAvsB1B,EAAA,UAAY,aAysBrB,GA3sBgC,IA4sBhC,EAAA,cAAc,cAAc,IC3kC5B,YAqEsB,GACpB,MAAO,IAAI,IAAY,eA+DE,GACzB,MAAO,IAAI,IAAW,eA+FpB,EACA,GACF,MAAI,IAAW,MACb,GAAU,IAEL,GAAwB,EAAiB,eA0B5B,GACpB,MAAO,IAAM,eAIX,EACA,GACF,GAA4B,4BACxB,EAAgB,GC5QtB,AAwBA,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAEE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GALyC,EAAA,cAAc,cAWvD,GAAA,SAAA,GAAyB,EAAA,EAAA,GAAzB,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,GAAM,EAAG,IATF,EAAA,UAAY,MAW9B,GAbyB,IAczB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAK,EAAA,KAAS,OAF9B,EAAA,UAAY,QAI9B,GAN2B,IAO3B,EAAA,cAAc,cAAc,IAG5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAFO,EAAA,UAAY,SAI9B,GAN4B,IAO5B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,QAAY,IAFL,EAAA,UAAY,UAI9B,GAN6B,IAO7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAAjC,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAc,IAFP,EAAA,UAAY,cAI9B,GANiC,IAOjC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,SAAa,IAFN,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAW,IAFJ,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAeE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,QAAY,EAAG,IAdR,EAAA,UAAY,UAgB9B,GAlB6B,IAmB7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DAgBE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,WAAe,EAAG,IAfX,EAAA,UAAY,aAiB9B,GAnBgC,IAoBhC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,EAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAE,IAAI,IAAQ,IAAI,MATlC,EAAA,UAAY,QAW9B,GAb2B,IAc3B,EAAA,cAAc,cAAc,gBAEQ,GAClC,MAAO,GAAW,2BAIjB,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACM,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,MAChB,GAAM,GAAmC,GACzC,SAAO,UAAe,SACtB,EAAO,OAAY,GACZ,GAAsB,GAE/B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCjQjC,AAiBA,YAA0B,GACxB,GAAI,GAAQ,MAAQ,MAAO,IAAS,SAClC,KAAM,IAAI,OACN,mEACA,0BAAyB,IAOjC,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DAEA,MAAA,IAF0C,EAAA,cAAc,cAqBxD,GAAA,SAAA,GAA0B,EAAA,EAAA,GAQxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAEP,UAAiB,GAEjB,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,MAAQ,EAAK,KAAO,EACzB,EAAK,MAAQ,EAAK,KAAO,IAO3B,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,GAAyB,EAAA,MAAM,CAAC,IACpC,MAAI,GAAK,OACP,GAAiB,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAI,OAEhE,AAAI,EAAK,OACP,GACI,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,GAAS,OAEjD,EAAe,cAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,GAAM,KAAK,GAAI,GAAM,KAAK,KAI7B,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,CAAC,GAAI,EAAO,GAAiB,GAAI,EAAO,MA3ClD,EAAA,UAAY,OA6CrB,GA/C0B,IAgD1B,EAAA,cAAc,cAAc,gBAET,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,gBAGvC,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,IAOnD,GAAM,IACyC,CAChD,KAAQ,oBAGuB,GAEnC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGU,GAG7B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAuB,OACzB,OAAI,aAAsB,IACxB,EAEA,GAAuB,GClJlC,AAmCA,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAKxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,MACV,GAAK,SAAW,EAAK,YAIzB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,EAAS,GAAoB,GAC7B,GAAI,GAAS,EAAA,KAAK,GAClB,MAAI,MAAK,UAAY,MACnB,GAAS,EAAA,YAAY,EAAQ,EAAG,KAAK,WAEhC,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,SAAU,KAAK,UACnD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA5BF,EAAA,UAAY,OA8BrB,GAhC0B,IAiC1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAO7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,GAIvB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,UAAU,EAAG,KAAK,QAG3B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,YA4BrB,GA9B+B,IA+B/B,EAAA,cAAc,cAAc,IA6B5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAWzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAU/B,GAbO,EAAA,0BAAmD,QAI1D,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,gBAAkB,GACvB,EAAK,iBACD,GAAe,EAAK,kBAAoB,EAAK,2BACjD,EAAK,iBAAmB,GAAe,EAAK,kBAC5C,EAAK,gBAAkB,GAAc,EAAK,iBACtC,EAAK,YAAc,KACrB,EAAK,WAAa,aACT,MAAM,QAAQ,EAAK,YAC5B,EAAK,WAAa,EAAK,mBACd,MAAO,GAAK,YAAe,SACpC,EAAK,WAAa,CAAC,EAAK,gBAExB,MAAM,IAAI,GACN,8DACA,YAAW,EAAK,sBAIxB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAChC,GAAM,GAAoB,EAAW,MAAM,GAC3C,GAAI,KAAK,YAAc,KACrB,OAAgB,GAAA,EAAA,EAAA,KAAK,WAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAW,EAAI,GAAK,EAGxB,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAAM,KAAK,iBAEtC,GAAM,GAAiC,GACvC,GAAI,KAAK,YAAc,KACrB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,EAAK,GAAK,EAAW,GAGzB,KAAK,UAAY,CAAC,GAAI,IAAU,CAC9B,KAAM,EAAW,OACjB,KAAI,KAEN,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,SAAS,GAAoB,GACtB,EAAA,MAAM,EAAQ,KAAK,MAAM,SAGlC,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,iBAAkB,GAAqB,KAAK,kBAC5C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAoB,KAAK,iBAC1C,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAxEF,EAAA,UAAY,QA0ErB,GA5E2B,IA6E3B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAK/B,GARO,EAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGL,EAAK,OAAS,MAAQ,EAAK,QAAU,EAAK,cAC5C,KAAM,IAAI,IACN,4BAA4B,EAAK,MAAK,4CAI5C,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,IAAI,IAGb,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjCF,EAAA,UAAY,MAmCrB,GArCyB,IAsCzB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAOnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAE,IAAI,GAAK,EAAE,QAAQ,KAAK,OAAQ,aAG3C,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,kBA6BrB,GA/BqC,IAgCrC,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,aAAe,EAItB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,QAAU,GAAI,MAAoB,MACvC,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,OAG3D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,MAAK,QAAQ,EAAG,KAAK,OAG9B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,KAAM,KAAK,MAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,UA6BrB,GA/B6B,IAgC7B,EAAA,cAAc,cAAc,ICxV5B,YAwBI,EAAwB,EAAW,GACrC,GAAI,MAAO,IAAU,SACnB,MAAO,IAAa,EAAO,GAE3B,GAAI,EAAM,SAAW,EACnB,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EAAC,aACxD,eAAc,EAAM,OAAM,eAEhC,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAc,EAAM,GAC1B,GAAI,CAAC,GAAU,GACb,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EACvD,yBAAwB,KAAK,UAAU,GAAM,gBAC7C,wBAAuB,IAG/B,MAAO,eAaP,EAAqB,EAAoB,EACzC,EAAgB,GAClB,GADkB,AAAA,IAAA,QAAA,GAAA,GACd,GAAe,KACjB,MAAO,GAET,GAAM,GAAoB,EAAc,GAAa,GAAM,GAAW,GAClE,EACJ,MAAI,KAAY,OACd,EAAe,EAEf,EAAe,EAAc,EAAoB,EAE5C,KAAK,MAAO,GAAe,EAAS,GAAK,eAI9C,EAAiB,EAAoB,EACrC,GACF,GAAI,GAAW,KACb,MAAO,MAGT,GAAI,IAAY,QACd,EAAU,EAAU,EAAa,GAAI,CAAC,EAAa,EAAY,YACtD,IAAY,OACrB,EAAU,EAAU,MAEpB,MAAM,IAAI,GAAW,2BAA2B,EAAO,KAEzD,MAAO,GCpFT,YAwCI,EAAW,GAEb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,IAE3B,gBAWT,EAAW,GACb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,IAE9B,gBAsBT,EAAW,EAAgB,EAAc,EAAa,EACtD,EAAyB,GADgB,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,SAC7B,AAAA,IAAA,QAAA,GAAA,GACpB,EAAA,KAAK,WAMV,GALA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAEZ,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,+DACG,GAAE,MAAM,OAAM,cAEvB,GAAI,EAAO,MAAM,SAAW,EAC1B,KAAM,IAAI,GACN,iEACG,GAAO,MAAM,OAAM,aAE5B,GAAI,GAAQ,MAAQ,EAAK,MAAM,SAAW,EACxC,KAAM,IAAI,GACN,+DACG,GAAO,MAAM,OAAM,aAM5B,GAHA,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,KAE1B,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,GAAI,GAAY,EAAA,OACZ,EAA0B,EAAoB,EAC9C,IAAY,OAAS,OAAS,QAAS,MAAO,GAClD,MAAI,IAAQ,MACV,GAAI,GAAU,EAAG,IAEZ,gBAoDP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EAAyB,EAC5C,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,IACvD,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,gEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,iEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,MAAU,OAAO,CACnB,EAAG,EACH,OAAQ,EACR,QAAS,EACT,IAAK,IAAY,OAAS,OAAS,QACnC,UAAW,EACX,WAAY,OACZ,KAAI,EACJ,WAAU,IAEZ,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBA8BP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EACnB,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,EAAG,IAC1D,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,MAEf,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,oEACG,GAAE,KAAI,MAEf,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,OACA,EACA,EAAmC,EACnC,IAAY,OAAS,OAAS,QAAS,QAAS,GACpD,AAAI,GAAQ,MACV,GAAI,GAAU,EAAG,IAEnB,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAmHX,GAAA,IAAA,SAAA,GAAuC,EAAA,EAAA,GAwBrC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,IAAkB,KAIxB,GAVQ,EAAA,KAAsB,KAEvB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAIzD,EAAS,WAAW,GACpB,EAAK,KAAO,EACZ,GAAoC,EAAK,KAAM,QAC3C,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,EAAK,OAAS,EACtD,KAAM,IAAI,IACN,qDACI,EAAK,KAAI,6BAqBnB,GAlBA,EAAK,WAAa,GAAe,EAAK,WAAY,EAAM,cACxD,EAAK,QAAU,GACX,EAAK,SAAW,KAAO,EAAI,EAAK,QAAS,EAAM,WACnD,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,WAAa,GAAc,EAAK,YACrC,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAClD,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,aAAe,GAChB,EAAK,cAAgB,KAAO,EAAI,EAAK,aAAc,EACnD,gBACA,EAAK,OAAS,GACb,MAAM,QAAQ,EAAK,eAAiB,EAAK,aAAa,SAAW,EACpE,KAAM,IAAI,GACN,iGAEA,IAAG,KAAK,UAAU,EAAK,gBACtB,GAAI,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aAAe,CAAC,EAAK,aAAc,EAAK,sBACpC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,gEACA,8BAA6B,KAAK,UAAU,EAAK,wBAE9C,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aACD,CAAC,EAAK,aAAc,EAAK,aAAc,EAAK,sBACvC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,kEACA,8BAA6B,KAAK,UAAU,EAAK,yBAK1C,SAAA,WAAjB,SAA4B,GAI1B,GAFA,GACI,cAAgB,GAAM,2CACtB,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,oEACA,oCACI,KAAK,UAAU,EAAK,YAAW,OAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,gBAAiB,GAAqB,KAAK,iBAC3C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA/GuC,IAqHvC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAcjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAA0B,KAZ9B,SAAA,OAAwB,KAahC,EAAK,WAAW,GAChB,EAAK,QAAU,EAAK,QACpB,GAAoC,EAAK,QAAS,WAClD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,kBAAoB,GAAe,EAAK,qBAG/C,SAAA,UAAA,MAAA,SAAM,SACJ,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAE1B,GAAM,GAAW,EAAW,GAEtB,EAAc,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,UAE3D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,KAAM,KAAK,gBACnC,KAAK,gBAAiB,GAAM,KAAK,iBAGvC,KAAK,UAAY,CAAC,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,KACtE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GACE,EAAY,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACjD,EAAsB,GACU,EAAK,WAAW,gBAEtD,GAAI,GAAuB,MAAQ,EAAK,OAAS,EAC/C,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,aACtB,QAEJ,GAAI,EAAK,OAAS,EAChB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAQ,GACpD,EAAK,QAAS,EAAK,WAAY,EAAK,aAAa,YAC5C,EAAK,OAAS,EAEvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,sBACjB,EAAK,OAAS,EACvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,kBAE1B,MAAM,IAAI,IACN,yDAGN,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAIpC,MAAO,MAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAKhC,OAJM,GAAqB,GACrB,EAAS,KAAK,aAAe,eAC/B,EAAW,MAAM,EAAG,EAAW,OAAS,GACxC,EAAW,MAAM,GACZ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAM,GAAS,GACX,EAAM,GAAI,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,GACzD,MAAO,MAAK,cAAiB,SAAW,KAAK,aACL,KAAK,aAAa,IAC9D,EAAS,KAAK,GAGhB,GAAI,GAAc,CAAC,EAAW,IAC9B,MAAI,MAAK,aAAe,eACtB,GAAc,EAAY,OAAO,GACjC,EAAY,KAAK,KAAK,UAEtB,GAAY,KAAK,KAAK,SACtB,EAAc,EAAY,OAAO,IAE5B,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,kBAAmB,GAAqB,KAAK,mBAC7C,iBAAkB,GAAoB,KAAK,mBAEvC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,CAAE,YAAa,KAAS,MAAO,GAAK,SAAY,UAChD,EAAK,QAAU,EACjB,KAAM,IAAI,GACN,kEACA,YAAW,KAAK,UAAU,EAAK,YAGzC,GA1ImC,IA4InC,GAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAK,MAAO,GAAK,YAAe,UAC5B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,gCAA+B,KAAK,UAAU,EAAK,YAAW,OAnB/D,EAAA,UAAY,SAsBrB,GAxB4B,IAyB5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UACzB,CAAE,OAAM,QAAQ,EAAK,aAClB,GAAK,WAAW,SAAW,GAAK,EAAK,WAAW,SAAW,IAChE,KAAM,IAAI,GACN,mDACA,4CACI,KAAK,UAAU,EAAK,YAAW,OApBtC,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAGX,GAFA,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KAEnC,EAAK,UAAY,QAAU,EAAK,UAAY,QAC9C,KAAM,IAAI,GACN,gEACA,2CAA0C,EAAK,mBAIvD,SAAA,UAAA,MAAA,SAAM,SAGJ,GAFA,EAAa,GAAmB,GAE5B,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,mDACA,KAAK,UAAU,IAGrB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,wEAGN,GAAM,GAAW,EAAW,GACtB,EAAc,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,IAE1D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,UAAW,KAAK,kBACvC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,GAAM,KAAK,iBAIvC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MAC3D,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAM,MAAM,SAAW,EACzB,KAAM,IAAI,GACN,iEACA,8BAA6B,EAAM,MAAM,SAG/C,GAAM,GAAa,EAAM,MACnB,EAAY,EAAW,GAEzB,EACA,EACJ,AAAI,EAAK,aAAe,gBACtB,GAAQ,EACR,EAAQ,GAER,GAAQ,EACR,EAAQ,GAGV,GAAM,GAAS,EAAW,GACpB,EAAQ,EAAW,GACnB,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,QAAQ,GACvB,EAAU,EAAK,QAAQ,GAGvB,EAAY,GAAa,EAAQ,EAAS,EAAS,EAAK,SACxD,EAAW,GAAa,EAAO,EAAS,EAAS,EAAK,SAMtD,EACF,CAAC,EAAW,EAAW,EAAU,EAAK,SAE1C,AAAI,EAAK,aAAe,gBACtB,GAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,KAEzC,GAAI,GAAU,EAAA,gBACV,EAAmB,EAAK,OAAO,OAAoB,EACnD,EAAK,QAA6B,EAAK,SAC3C,MAAI,GAAK,aAAe,gBACtB,GAAU,EAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,KAG7C,AAAI,EAAK,MAAQ,MACf,GACI,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEhD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAE3B,EACA,EACA,EACJ,AAAI,KAAK,aAAe,gBACtB,GAAc,EACd,EAAa,EACb,EAAY,GAEZ,GAAc,EACd,EAAa,EACb,EAAY,GAGd,GAAM,GAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,QAAQ,GACvB,EAAU,KAAK,QAAQ,GAE7B,SAAY,GAAe,KAAK,QAChC,EAAY,GACR,GAAa,EAAY,GAAa,EAAS,EAAS,KAAK,SACjE,EAAY,GACR,GAAa,EAAY,GAAY,EAAS,EAAS,KAAK,SACzD,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,aACP,GA9IF,EAAA,UAAY,kBAgJrB,GAlJqC,IAmJrC,EAAA,cAAc,cAAc,IA0C5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAqBjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAAO,KAEnB,GAXO,EAAA,8BACL,gBACK,EAAA,8BACL,gBAEM,EAAA,gBAAiC,KACjC,EAAA,gBAAiC,KAKrC,EAAO,SAAW,KACpB,KAAM,IAAI,GACN,uFAGN,GAAI,EAAO,mBAAqB,MAAQ,EAAO,mBAAqB,MAChE,EAAO,kBAAoB,KAC7B,KAAM,IAAI,GACN,sPAKN,GAAI,EAAO,SAAW,MAAQ,EAAO,UAAY,QAC7C,EAAO,UAAY,QACrB,KAAM,IAAI,GACN,gBAAgB,EAAK,KAAI,kCACzB,qCAAoC,KAAK,UAAU,EAAO,WAGhE,SAAK,gBACD,EAAO,iBAAmB,KAAO,EAAI,EAAO,gBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,qBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,uBAGlD,SAAA,UAAA,MAAA,SAAM,SAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,KAAK,KAAO,EAClC,KAAM,IAAI,GACN,0BAA0B,KAAK,KAAI,sBAChC,MAAK,KAAO,EAAC,gCAChB,IAAG,KAAK,UAAU,KAExB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,0DACA,cAAa,KAAK,UAAU,EAAW,MAO7C,OAJM,GAAW,EAAW,GACtB,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,kBACrC,EAAuB,GACpB,EAAI,EAAG,EAAI,KAAK,KAAM,EAAE,EAC/B,EAAqB,KAAK,GAE5B,EAAqB,KAAK,EAAW,KAAK,gBAAiB,KAAK,SAEhE,GAAM,GAAY,GAClB,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,EAAW,KAAK,gBAE1C,KAAK,KAAO,KAGd,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MACvE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAE7B,GAAI,GACJ,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,IACN,oDACC,MAAI,GAAK,OAAS,GACvB,CAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAG3C,EAAS,EAAA,gBACL,EAAoB,EAAK,gBAAgB,OACzC,EAAK,gBAAgB,OACrB,EAAK,QAA6B,EAAK,QACvC,EAAK,aAAkC,SAG7C,AAAI,EAAK,SACP,GAAS,GAAU,EAAQ,EAAK,KAAK,OAAQ,EAAK,aAEpD,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAGjC,AAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAEpC,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,kBACd,MAAO,GAAO,kBACd,MAAO,GAAO,iBACd,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,qBAC7B,EAAO,oBACH,GAAoB,KAAK,qBACtB,GAzJF,EAAA,UAAY,gBA2JrB,GA7JmC,IA+JnC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,SACV,GAAA,KAAA,KAAM,EAAG,IAAK,KAFT,SAAA,UAAY,kBAIrB,GANqC,IAOrC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,GAClB,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,WACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,2BAA0B,KAAK,UAAU,EAAK,YAAW,OArB1D,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAgC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAM9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,MAAI,OAAO,GAAK,UAAa,SAC3B,EAAK,SACD,CAAC,CAAC,EAAK,SAAU,EAAK,UAAW,CAAC,EAAK,SAAU,EAAK,WACrD,AAAI,MAAO,GAAK,SAAS,IAAO,SACrC,EAAK,SAAW,CACd,CAAC,EAAK,SAAS,GAAI,EAAK,SAAS,IACjC,CAAC,EAAK,SAAS,GAAc,EAAK,SAAS,KAG7C,EAAK,SAAW,EAAK,SAEvB,EAAK,WACD,EAAK,aAAe,OAAY,eAAiB,EAAK,WAC1D,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAI,MAAK,aAAe,gBACf,CACL,EAAW,GAAI,EAAW,GAC1B,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,IAGlD,CACL,EAAW,GACX,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GAAI,EAAW,KAK5E,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAFA,EAAS,GAAoB,GAEzB,EAAK,aAAe,gBACtB,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,QAEjE,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,OAKvE,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,SAAU,KAAK,SAAU,WAAY,KAAK,YACpD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAhEF,EAAA,UAAY,aAkErB,GApEgC,IAqEhC,EAAA,cAAc,cAAc,IAwB5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAOhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KALM,SAAA,aAAe,CAAC,EAAG,GAMpC,EAAK,UAAY,CAAC,CAAC,KAAM,IACzB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,aAGtD,SAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,KAAK,aAAe,iBACtB,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,QAE9C,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAO,EAAW,MAIrD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAC1B,EAAa,EAAM,MAEzB,GAAI,EAAK,aAAe,iBACtB,EAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,IACvC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GAClC,EAAU,EAAM,sBAAsB,CAAC,EAAQ,IACrD,MAAO,GAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,SAExC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GACxC,MAAO,GAAM,sBAAsB,CAAC,EAAQ,QAKlD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,KAAM,KAAK,KAAM,WAAY,KAAK,YAC5C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlDF,EAAA,UAAY,eAoDrB,GAtDkC,IAuDlC,EAAA,cAAc,cAAc,ICrvC5B,YA6CI,EAAW,EAAyB,EACpC,EAAmB,EACnB,GAFoC,MAAA,KAAA,QAAA,GAAA,CAA6B,EAAG,IACpE,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WACV,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAChB,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,OAEf,GAAI,EAAgB,OAAS,EAC3B,KAAM,IAAI,GACN,yDACG,GAAgB,KAAI,OAE7B,SAAI,EAAA,gBACA,EAAe,EAA6B,EAC5C,IAAY,OAAS,OAAS,QAAS,OAAQ,GACnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,IAsCX,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAsB,KAHzB,SAAA,gBAAiC,KAIvC,EAAK,gBACD,EAAK,iBAAmB,KAAO,EAAI,EAAK,gBAC5C,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,4BACtC,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,qBAAuB,GAAe,EAAK,wBAGlD,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,iDACA,0BAAyB,KAAK,UAAU,GAAW,MAEzD,GAAM,GAAc,KAAK,aAAe,gBAAkB,EAAI,EAC9D,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,iEACA,4BAA2B,EAAW,GAAY,OAExD,GAAM,GAAW,EAAW,GACtB,EAA8B,CAClC,KAAK,WAAW,GAAI,KAAK,WAAW,GAAI,EAAU,KAAK,iBAGzD,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,EAAW,KAAK,iBAAkB,KAAM,KAAK,gBACtD,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GAAU,GACV,EAAQ,EAAK,gBAAgB,OAAQ,EAAK,QAC1C,EAAK,QAAS,EAAK,WAAY,MAEnC,MAAI,GAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEtD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EAAa,KAAK,aAAe,gBACnC,EAAW,GAAK,KAAK,gBACrB,EAAW,GAAK,KAAK,gBACnB,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACnD,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,MAAI,MAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAY,EAAS,GAGrC,CAAC,EAAW,GAAI,EAAS,EAAS,IAI7C,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,SAAO,gBAAqB,KAAK,gBACjC,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,sBACtB,GAjGF,EAAA,UAAY,kBAmGrB,GArGqC,IAsGrC,EAAA,cAAc,cAAc,IClN5B,YA0DI,EACA,EACA,EACA,GAKF,GAAI,MAAM,QAAQ,IAChB,GAAI,GAAgB,MAAQ,GAAa,KACvC,KAAM,IAAI,GACN,iFAGN,AAAI,GAAgB,MAClB,GAAY,EAAO,MAAM,EAAO,OAAS,EAAc,EAAO,QAC9D,EAAS,EAAO,MAAM,EAAG,EAAO,OAAS,IAE3C,AAAI,EAAO,OAAS,GAClB,GAAe,EAAO,MAAM,EAAG,EAAO,SAExC,EAAS,EAAO,GAGlB,WAAsB,GAEpB,MAAI,IAAK,MAAQ,MAAM,QAAQ,GACtB,EAEA,CAAC,GAIZ,SAAe,EAAa,GAC5B,EAAY,EAAa,GAElB,CAAC,OAAM,EAAE,aAAY,EAAE,UAAS,eA+CrC,EAA+B,EAAgB,EAC/C,EAAqB,EAAe,EAAsB,EAC1D,GADA,MAAA,KAAA,QAAA,GAAA,IAA0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,IACK,EAAA,KAAS,WACd,GAAM,GAAO,EAAO,MAAM,OAC1B,GAAI,EAAO,EACT,KAAM,IAAI,GAAW,uCAAuC,EAAI,MAKlE,GAAM,GAAO,CAAC,EAAG,GAAG,OAAO,GAAiB,EAAG,IAG/C,GAFA,EAAS,EAAA,UAAc,EAAQ,GAE3B,GAAa,KACf,KAAM,IAAI,IACN,kFAKN,AAAI,GACF,QAAQ,KACJ,qGAIN,AAAI,GAAQ,MACV,GAAO,EAAK,OAAO,QAAQ,OAAO,WAClC,AAAI,EAAK,OAAS,EAAO,GACvB,GAAO,EAAA,WAAe,EAAM,KAE9B,EAAO,EAAA,UAAc,EAAM,IAG7B,AAAI,GACF,GAAS,EAAA,QAAY,EAAQ,GAC7B,AAAI,GAAQ,MACV,GAAO,EAAA,QAAY,EAAM,KAc7B,GAAM,GAA2B,GAC7B,EACA,EAAS,EACP,EAAY,EAAO,MAAM,GACzB,EAAgB,EAAA,QAAY,GAC9B,EACJ,AAAI,GAAQ,MACV,GAAe,EAAA,QAAY,IAG7B,mBAAS,GACP,GAAM,GAAe,EAAc,GAC7B,EAAc,EAAA,KAAS,WAAM,MAAA,GAAa,EAAc,KAE9D,GAAI,GAAQ,KACV,EAAa,EAAY,GACzB,EAAS,EAAY,QAErB,GAAM,GAAgB,EAAA,KAAS,WAC7B,GAAM,GAAW,EAAa,GACxB,EAAc,EAAA,SAAa,GAAU,IAAI,GAEzC,EACF,EAAY,GAAG,IAAI,GAAU,IAAI,EAAO,GAAG,IAAI,IAC7C,EAAY,EAAO,IAAI,SAAC,EAAO,GACnC,MAAO,GAAY,GAAG,GAAG,IAAI,GAAU,IAAI,EAAM,IAAI,MAEvD,MAAO,CAAC,OAAM,EAAE,UAAS,KAE3B,EAAa,EAAc,OAC3B,EAAS,EAAc,UAGzB,AAAI,GACF,EAAe,KAAK,IAxBf,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,GA2BT,GAAI,GACJ,GAAI,GACF,GAAM,GAAO,EACb,EAAU,EAAA,MAAU,EAAgB,GAEtC,MAAO,CAAC,EAAY,EAAS,wBAyGR,EAAA,EAAA,GAqBvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACP,EACJ,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,GACN,wDAMN,GALO,AAAI,MAAM,QAAQ,EAAK,MAC5B,EAAO,GAAI,IAAgB,CAAC,MAAO,EAAK,OAExC,EAAO,EAAK,KAEV,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,qGAGN,SAAK,KAAO,EACZ,EAAK,gBACD,EAAK,iBAAmB,KAAO,GAAQ,EAAK,gBAChD,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,UAAY,EAAK,UAAY,KAAO,GAAQ,EAAK,SACtD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAQ,EAAK,OAEjD,EAAK,gBAAkB,GACvB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KACvC,EAAK,UAAY,KACjB,EAAK,QAAU,KAEf,EAAK,aAAe,KAIpB,EAAK,WAAa,KAKpB,SAAA,UAAA,UAAA,WACE,GAAI,KAAK,SAAW,MAClB,GAAM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EACtE,MAAO,IAAiB,EAAG,GAAW,IAAI,SAAA,GAAK,MAAA,YAE/C,OAAO,MAAK,SAMhB,EAAA,UAAA,UAAA,SAAU,GACR,KAAK,QAAU,GAGjB,EAAA,UAAA,mBAAA,SAAmB,GACjB,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAGb,GAAI,GAAY,KAAK,KAAK,UAC1B,AAAK,MAAM,QAAQ,IACjB,GAAY,CAAC,IAEf,GAAM,GAAY,EAAU,GACxB,EAOJ,GANA,AAAI,KAAK,gBACP,EAAc,CAAC,EAAW,GAAI,EAAW,GAAI,GAE7C,EAAc,CAAC,EAAW,GAAI,GAG5B,KAAK,aAEP,OADM,GAAsB,GACV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAW,KAAK,CAAC,EAAW,GAAI,IAElC,MAAO,CAAC,GAAa,OAAO,OAE5B,OAAO,IAIX,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAS,WACd,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAM,GAAa,EAAK,gBAAkB,EAAO,KAEjD,GAAI,EAAK,aACP,GAAM,GAAY,EAAK,OAAO,IAAI,SAAA,GAAK,MAAA,QACvC,MAAO,CAAC,GAAY,OAAO,OAE3B,OAAO,MAWb,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,SAAW,MAIlB,OAHM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EAChE,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,EAAO,KAAK,MAEd,MAAO,OAEP,OAAO,MAAK,aAIhB,SAAW,GACT,KAAK,QAAU,mCAGV,EAAA,UAAA,MAAP,SAAa,GAIX,GAAI,KAAK,cAAgB,KACvB,KAAM,IAAI,IACN,oDAGN,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAEb,GAAM,GAAoB,KAAK,SAAW,EAAW,GAAK,KACpD,EAAW,EAAW,MAAM,GAClC,KAAK,UAAU,GAAK,GAAI,IAAU,CAAC,MAAK,CAAG,EAAW,MAAI,OAAK,KAI/D,GAAM,GAAiB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAK7D,KAAK,KAAK,MAAM,GAIlB,GAAI,GAOJ,GANA,AAAI,MAAM,QAAQ,KAAK,KAAK,WAC1B,EAAY,KAAK,KAAK,UAEtB,EAAY,CAAC,KAAK,KAAK,WAGrB,KAAK,WAAa,MACpB,GAAI,CAAC,EAAA,KAAK,YACF,KAAK,UAAU,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,EAAK,MAAM,OAAS,KAC1D,GACN,KAAM,IAAI,GACN,0DACA,uCAAsC,KAAK,UAAS,MACpD,8BAA6B,KAAK,KAAK,gBAG7C,MAAK,UACD,EAAU,IAAI,SAAA,GAAO,MAAA,IAAI,IAAU,CAAC,MAAO,CAAC,KAAM,OAExD,AAAI,KAAK,UACP,KAAK,eAqBT,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAK,WACH,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAEN,GAAM,GAAY,EAAK,UAAU,GAAG,MAAM,GAC1C,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUAQN,GAAI,EAAK,SAAW,KAClB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAU,CAAC,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,qBAEzC,GAAU,KAEnB,EAAA,QAAY,EAAK,SAEjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAQ,GAAK,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,iBAMpD,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAER,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,IAAa,GAKf,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GACf,EAAM,MAAM,QAAQ,EAAK,KAAK,WAChC,EAAK,KAAK,UAAU,GACpB,EAAK,KAAK,UACR,EAAgB,CAAC,EAAW,GAClC,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAEhB,EAAK,QAAQ,GAAS,GAG1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAI5D,EAAA,UAAA,MAAA,SACI,EACA,GAEF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAGX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAC1D,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAMzB,GAAI,GAAiD,GACjD,EAA+B,GACnC,GAAI,GAAgB,MAClB,EAAO,aAAkB,EACzB,EAAmB,EAAiB,OAAO,GAC3C,KAAK,UAAY,GACjB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,UAAU,KAAK,GAAI,IAAU,CAAC,MAAO,EAAM,SAKlD,EAAkB,EAAgB,OAAO,KAAK,WAEhD,AAAI,GAAa,MACf,GAAO,UAAe,EACtB,EAAmB,EAAiB,OAAO,GAE3C,KAAK,aAAe,EAAU,QAGhC,GAAM,GAAW,EAAiB,YAAc,IAChD,GAAI,GAEF,GAAM,GACF,CAAC,GAAQ,OAAO,GACd,EAAgB,KAAK,UAAU,OAAO,GAEtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GAAS,EAAA,UAAM,MAAK,KAAA,KAAC,EAAW,GACtC,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAK/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KAIE,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC5C,EACA,GAAU,KAAO,KAAO,EAAO,aAEnC,EAAS,GAAoB,GAC7B,AAAI,GAAgB,MAClB,CAAI,EAAK,SACP,EAAe,EAAK,QAEpB,EAAe,EAAK,gBAAgB,IAIxC,GAAM,GACF,MAAM,QAAQ,EAAK,KAAK,WAAa,EAAK,KAAK,UAAU,OAAS,EACtE,GAAI,EAAa,SAAW,EAC1B,KAAM,IAAI,GACN,iBAAiB,EAAS,4BACvB,GAAa,OAAM,uBAE5B,AAAI,EAAK,QACP,QAAQ,KACJ,oEAGN,GAAM,GAAyB,CAAC,SAAQ,GAGlC,EAAO,SAAC,EAAgB,GAG5B,GAAM,GACF,EAAK,KAAK,KAAK,CAAC,GAAQ,OAAO,GAAS,GAE5C,MAAO,CAAC,EAAQ,GAAI,EAAQ,MAAM,KAK9B,EACF,GAAI,EAAM,EAAQ,EAAc,EAAK,YAAa,EAAM,KACpD,EAAK,OAAQ,EAAK,iBACpB,EAAa,EAAW,GACxB,EAAU,EAAW,GACrB,EAAS,EAAW,GAE1B,AAAI,EAAK,UACP,EAAK,YAAY,EAAQ,GAG3B,GAAM,GAAS,EAAK,gBAAkB,EAAU,EAIhD,MAAI,GAAK,YACA,CAAC,GAAQ,OAAO,GAEhB,KAKb,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAAI,GAAe,EAAA,MAAU,EAAO,OAKpC,MAHA,GAAe,EAAA,IAAQ,EAAc,CAAC,EAAG,IACzC,EAAe,GAAa,GAExB,MAAM,QAAQ,EAAK,KAAK,WACnB,EAAK,KAAK,UAAU,IACvB,SAAA,GAAO,MAAA,GAAM,EAAI,GAAO,EAAc,CAAC,EAAG,IAAQ,IAE/C,EAAK,KAAK,UAAY,EACzB,CAAC,GAAO,EAAc,CAAC,EAAG,EAAK,KAAK,aACpC,CAAC,MAKX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAK,MAAK,UAIH,KAAK,KAAK,iBAHR,oCAMX,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,MAAK,MAAK,UAGH,KAAK,KAAK,oBAFR,KAAK,KAAK,yCAKrB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,MAAQ,MACf,KAAK,KAAK,6BAA6B,IAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,gBAAiB,KAAK,gBACtB,YAAa,KAAK,YAClB,YAAa,KAAK,YAClB,SAAU,KAAK,SACf,OAAQ,KAAK,QAGf,AAAI,KAAK,cAAgB,MACvB,GAAO,aAAkB,KAAK,cAGhC,GAAM,GAAa,KAAK,KAAK,YAE7B,MAAI,MAAK,iBAAmB,EAAI,WAC9B,GAAO,KAAU,CACf,UAAa,KAAK,KAAK,eACvB,OAAU,IAKd,GAAA,GAAW,EAAe,EAAe,IAIpC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAa,EAAO,KACpB,EAAO,GAAY,EAAY,GACrC,MAAO,IAAI,GAAI,OAAO,OAAO,EAAQ,CAAC,KAAI,MArfrC,EAAA,UAAY,MAufrB,GAzfyB,IA0fzB,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAAtC,4DAQA,MAAA,IARsC,IAwFtC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAiCjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KANJ,SAAA,mBAAqB,OACrB,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAIzD,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,YAAc,KAAO,EAAK,mBAAqB,EAAK,YAC7D,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAG9B,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAEhC,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAW,EAAW,OAAS,GAAI,KAAK,OAAQ,KAC3D,KAAK,kBAAmB,KAAK,kBAAmB,GAChD,KAAK,kBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,OAAQ,KAC9C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IASf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8CAA8C,EAAO,OAAM,KAEjE,GAAI,GAAa,EAAO,GACxB,EAAS,EAAO,GAChB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,KAG/B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,KAGxC,GAAI,GACE,EAAiB,EAAK,YACtB,EAAoB,EAAK,qBAC/B,AAAI,GAAU,KACZ,EAAI,GAAM,EAAA,IAAQ,EAAQ,GAAS,EAAK,OAAO,QAE/C,EAAI,GAAM,EAAQ,EAAK,OAAO,QAEhC,AAAI,EAAK,MAAQ,MACf,GAAI,GAAU,EAAG,EAAK,KAAK,SAE7B,AAAI,GAAa,MACf,GAAa,EAAA,IAAQ,EAAY,IAEnC,GAAI,GAAS,EAAA,IAAQ,EAAG,GAAM,EAAY,EAAK,gBAAgB,SAC/D,MAAI,GAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAI1B,CAAC,EAAQ,MAIpB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,kBAGzB,MAAA,IAAA,GAAW,EAAe,IArKrB,EAAA,UAAY,gBAuKrB,GAzKmC,IA0KnC,EAAA,cAAc,cAAc,IA2F5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAG7B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAK,KAAO,GAAI,IAAc,GAC9B,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IA7BV,EAAA,UAAY,YA+BrB,GAjC+B,IAkC/B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAqC3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAbO,EAAA,mBAAqB,OACrB,EAAA,6BAAqD,cAErD,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAQrD,EAAK,WACP,KAAM,IAAI,GACN,+DAEN,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,KAAK,gBACrC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,uDACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACzD,EAAW,EAAO,GACtB,EAAS,EAAO,GAKhB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EAAY,EAAK,qBACnB,EACA,EACA,EAEJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAU,GAAM,EAAQ,EAAK,OAAO,QACxC,AAAI,EAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,SAEzC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAGzC,GAAM,GAAuB,EAAK,gBAAgB,OAC5C,EAAA,EAAA,MAAA,EAAA,CAAA,EAAA,EAAA,MAAA,EAAA,OAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAGN,EAAc,GAAM,EAAU,GAE9B,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GACT,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAE/C,GAAM,GAAa,GAAM,EAAA,IAAQ,EAAG,GAAW,GAC/C,EAAK,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,IAEvC,GAAM,GACF,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,IAAK,IAElE,MAAO,CAAC,EAAG,MAIf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,eACrB,WAAY,IAGd,MAAA,IAAA,GAAW,EAAe,IAvMrB,EAAA,UAAY,UAyMrB,GA3M6B,IA4M7B,EAAA,cAAc,cAAc,IA8B5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAQ,GACxB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,MAuCrB,GAzCyB,IA0CzB,EAAA,cAAc,cAAc,IAuC5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAsC5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAZJ,SAAA,mBAAqB,OACrB,EAAA,6BAA+B,cAC/B,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAEhC,EAAA,yBAA2B,QASlC,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,EAAK,eAE3B,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,CAAC,EAAK,MAAO,EAAK,OACnC,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,GAAI,GACJ,GAAI,KAAK,SACP,GAAI,KAAK,gBACP,GAAM,GAAmB,KAAK,gBACxB,EAAgB,KAAK,MAC3B,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GAAM,GAAK,EAAiB,MAAM,CAAC,IAC7B,EAAM,GAAI,MAAQ,MAAM,CAAC,IACzB,EAAS,EAAiB,MAAM,CAAC,EAAgB,IACvD,MAAO,IACH,GAAuB,EAAI,GAAK,IAExC,GAZgD,IAEvC,EAAA,UAAY,oBAYrB,GAAkB,KAAK,gBAEzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,EAAiB,KAAK,gBACtD,GAAM,KAAK,oBAEf,MAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,wDACG,GAAO,OAAM,MAEtB,GAAI,GAAW,EAAO,GAChB,EAAW,EAAO,GACxB,EAAS,EAAO,GAChB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EACF,EAAK,qBAKL,EACA,EACA,EACA,EACJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAI,GAAM,EAAQ,EAAK,OAAO,QAClC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAEzC,EAAI,EAAA,IAAQ,EAAG,GAAM,EAAU,EAAK,gBAAgB,SACpD,AAAI,EAAK,SACP,GAAI,GAAU,EAAG,EAAK,KAAK,SAGvB,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,KACnE,EAAI,EAAK,oBAAoB,MAAM,GAEnC,GAAM,GAAI,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,IAE3C,MAAO,CAAC,EAAG,EAAG,MAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,eAAgB,KAAK,eACrB,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,gBAGvB,MAAA,IAAA,GAAW,EAAe,IAnNrB,EAAA,UAAY,WAqNrB,GAvN8B,IAwN9B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAGxB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAS,GACzB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,OAuCrB,GAzC0B,IA0C1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAME,OADM,GAAsB,GACT,EAAA,EAAA,EAAA,KAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAU,KAAI,MAAd,EAAkB,EAAK,WAEvB,EAAU,KAAK,EAAK,WAGxB,MAAO,oCAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,EAKT,OAJI,GAAS,EAAO,MAAM,GAGpB,EAA2B,GACd,EAAA,EAAA,EAAA,EAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAa,KAAK,EAAO,OAAO,EAAG,EAAK,UAAU,SAElD,EAAa,KAAK,EAAO,OAAO,EAAG,IAGvC,EAAa,UAKb,OAFM,GAA8B,GAChC,EACK,EAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAO,EAAK,MAAM,GACxB,EAAS,EAAa,GAEtB,AAAI,IAAM,EACR,EAAa,CAAC,EAAO,IAAI,OAAO,GAEhC,EAAa,CAAC,EAAW,IAAI,OAAO,GAEtC,EAAa,EAAK,KAAK,EAAY,GACnC,EAAgB,KAAK,EAAW,MAAM,IAIxC,EAAS,GACT,OAAyB,GAAA,EAAA,EAAA,EAAgB,QAAQ,UAAxB,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAO,KAAI,MAAX,EAAe,GAEjB,MAAO,CAAC,EAAW,IAAI,OAAO,MAI3B,EAAA,UAAA,MAAP,SAAa,GACX,AAAI,GAAgB,IAGlB,GAAc,EAAuB,IAEvC,EAAa,EACb,GAAI,GACJ,KAAK,MAAM,QAAQ,SAAC,EAAM,GACxB,GAAU,WAAW,EAAK,WAGxB,EAAK,MAAM,GACX,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAY,EAAK,UAAU,GAE3B,EAAY,EAAK,UAEnB,EAAa,CAAC,EAAW,GAAI,OAGjC,KAAK,MAAQ,IAGf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAgB,SAAC,GACrB,MAAO,CACL,UAAa,EAAK,eAClB,OAAU,EAAK,cAIb,EAAc,KAAK,MAAM,IAAI,GAE7B,EAAS,CAAC,MAAS,GAEzB,MAAA,IAAA,GAAW,EAAe,IAIrB,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAElB,OADM,GAAmB,GACA,EAAA,EAAA,EAAC,EAAO,MAAR,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAM,KAAK,GAAY,EAAY,IAErC,MAAO,IAAI,GAAI,CAAC,MAAK,KAGvB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,kBAEvB,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,qBAEvB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACvB,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAiB,KAAI,MAArB,EAAyB,EAAK,kBAEhC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAQT,EAAA,UAAA,WAAA,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,SAEvB,MAAO,IAAc,IASvB,EAAA,UAAA,WAAA,SAAW,GAET,OADM,GAAyC,GAC5B,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,IAGjB,OAHS,GAAI,EAAA,GACP,EAAY,EAAK,QAAQ,OACzB,EAAe,EAAQ,OAAO,GAC3B,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,EACzC,EAAO,KAAK,CAAC,EAAK,QAAQ,GAAI,EAAa,KAG/C,GAAc,IA5KT,EAAA,UAAY,kBAgLrB,GAlLqC,IAmLrC,EAAA,cAAc,cAAc,gBAEQ,GAM3B,GAAA,GAAA,EAAA,KAAM,EAAA,EAAA,KAAM,EAAA,EAAA,SAAA,EAAA,IAAA,OAAA,GAAA,EAAkB,EAAA,EAAA,MAAA,EAAA,IAAA,OAAA,EAAA,EAE/B,EAAgB,WAAM,MAAA,IAAU,IAAQ,IAExC,EAAa,WAAM,MAAA,IAAe,EAAe,EAAM,IAG7D,GAAI,CAAC,GAAS,GAAS,EACrB,MAAO,GAAA,KAAS,IAAa,SAG/B,GAAM,GAAQ,MAAM,GAAO,KAAK,QAAW,IAAI,GAE/C,MAAO,GAAM,IAAI,SAAA,GAAK,MAAA,GAAA,KAAS,EAAE,WC/iEnC,AAiFA,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAArC,4DAyBA,MAAA,IAzBqC,IAiCrC,GAAA,SAAA,GAAwB,EAAA,EAAA,GAMtB,WAAY,GAAZ,GAAA,GAAA,KACE,GAAI,EAAK,OACP,KAAM,IAAI,IACN,sDAGN,GAAI,MAAM,QAAQ,EAAK,MACrB,KAAM,IAAI,IACN,kEAGN,SAAA,EAAA,KAAA,KAAM,IAAqB,KAE3B,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WAad,GAZA,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aAEtB,EAAK,KAAK,YAAc,MAG1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBAEtB,EAAK,KAAK,qBAAuB,MAG/B,GAAU,EAAO,UACnB,KAAM,IAAI,GAAW,6CAGvB,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KAEtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAE1C,EACF,GAAU,KAAO,KAAO,EAAO,aAEnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAI3D,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GAAkB,KAAK,yBAAyB,GAEpD,MAAK,MAAK,iBACR,GAAQ,CAAI,EAAS,IAAE,OAAK,EAAS,MAAM,KAG7C,AAAI,KAAK,aACP,GAAQ,CACH,GAAQ,OAAK,MAAM,GAAG,KAAI,CAAE,EAAW,IAAE,OAAK,EAAS,MAAM,QAG7D,GAGT,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACP,GAAA,GAAA,EAAA,KAAA,UAED,EAAa,EAAO,MAEpB,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAe,EAAA,MAAU,GAE/B,MAAI,OAAM,QAAQ,GACT,MAAM,EAAU,QAAQ,KAAK,GAG/B,CAAC,MAIZ,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAS,WACP,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAGN,GAAM,GAAa,EAAK,UAAU,GAAG,MAE/B,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAY,EAAW,GAE7B,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUASN,GAAI,EAAK,aAAe,KACtB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAU,CAAC,EAAA,MAAU,YAEnB,GAAU,KAEnB,EAAA,QAAY,EAAK,SAGjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAQ,GAAK,EAAA,MAAU,QAO9B,GAJA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAGR,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,EAKF,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GAEf,EAAgB,EAEtB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAGhB,EAAK,QAAQ,GAAS,GAI1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAIlD,EAAA,UAAA,yBAAV,SAAmC,GAC3B,GAAA,GAAA,KAAA,KAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,QAAS,EAAA,EAAA,aAGpD,EAAkB,IAAe,gBAEjC,EAAI,EAAW,EAAkB,EAAI,GACrC,EAAI,EAAW,EAAkB,EAAI,GAErC,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAClD,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAElD,EACD,EAAW,MAAM,EAAG,GAAE,OACrB,EAAkB,CAAC,EAAS,EAAM,GAAQ,CAAC,EAAM,EAAM,IAG7D,MAAO,IAhMF,EAAA,UAAY,YAkMrB,GApMwB,IAyMxB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAWlC,WAAY,GAAZ,GAAA,GAAA,KAEI,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,aAGF,SAAA,EAAA,KAAA,KAAA,GAAA,GAAU,EAAI,CAAE,MAAO,MAAS,KAEhC,EAAK,QAAU,EACf,GAAsB,EAAK,QAAS,WAEpC,EAAK,WAAa,GAAe,EAAY,EAAG,cAChD,EAAK,WAAW,QAAQ,SAAA,GAAQ,MAAA,IAAsB,EAAM,gBAE5D,EAAK,QAAU,GAAe,GAAW,EAAG,EAAG,WAC/C,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,IAAsB,EAAQ,aAE7D,EAAK,QAAU,GAAW,QAC1B,GAAiB,EAAK,SAEtB,EAAK,WAAa,GAAc,eAChC,GAAgB,EAAK,YAErB,EAAK,aAAe,GAAe,GAAgB,EAAG,EAAG,gBACzD,EAAK,aAAa,QACd,SAAA,GAAQ,MAAA,IAAsB,EAAM,oBAGnC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAEhC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAElE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAG1B,GAAM,GAAW,EAAW,GAEtB,EAAe,EAEf,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,QAAU,IAErD,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBAEvC,GAAM,GACF,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,KAAK,QAAU,IAOzD,GALA,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBAEL,KAAK,SACP,GAAI,GAAe,OAEnB,GAAI,KAAK,gBACP,GAAM,GAAO,KAAK,gBAEZ,EAAU,KAAK,QAErB,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAQ,EAAK,MAAM,CAAC,IACpB,EAAQ,EAAA,KAAS,CAAC,IAClB,EAAY,EAAK,MAAM,CAAC,EAAU,IACxC,MAAO,IAAc,CAAC,EAAO,EAAO,KAExC,GAVgD,IAEvC,EAAA,UAAY,oBAUrB,GAAkB,KAAK,gBAGzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,QAAU,GAAe,KAAM,EAC7C,KAAK,gBAAiB,GAAM,KAAK,gBAGvC,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAsB,GAA3B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8DACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,GAEjC,EAAI,EAAO,GACX,EAAW,EAAO,GAClB,EAAW,EAAO,GAElB,EAAe,EAIrB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAI9B,GAAM,GAAc,EAAK,YAEnB,EACF,SAAC,GAAe,GAAoB,IAClC,MAAI,CAAC,IAAQ,CAAC,GAAK,IACV,GAGF,EAAA,IAAQ,GAAK,IAAQ,KAG9B,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAEtC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAIvC,GAAM,GAAiB,EAAK,qBAExB,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAE1C,EAAoB,EAEpB,EAAA,EAAA,MAAA,EAAA,OAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAG5B,EAAA,EAAA,uDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAI5B,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAEvC,GAAA,GAAA,EAAA,MAAA,EAAA,gBAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAI3C,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAE5B,GAAM,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAA,IACN,EAAA,IAAQ,GAAG,GACX,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,MAC3C,GAAI,EAAA,IACN,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC3C,EAAK,WAAW,MAAM,KAE1B,MAAO,CAAC,GAAG,GAAG,OAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAA,EAAA,UAAA,UAAA,KAAA,MAAC,EAAA,EAAA,MAAY,EAAA,GAAA,EAAA,CAAA,UAEb,EAAuC,CAC3C,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,QAAS,KAAK,SAGhB,MAAA,IAAA,GAAW,EAAe,IAG5B,EAAA,UAAA,UAAA,SAAU,EAAW,EAAW,EAAY,GAC1C,GAAM,GAAM,EAAA,OACR,EAAmB,EAAmB,KAAK,QAC1C,GAAW,QACZ,KAAK,aAAe,gBAAkB,OAAS,OAC/C,KAAK,cAET,MAAI,GACK,GAAU,EAAK,EAAG,KAAK,YAGzB,GAGT,EAAA,UAAA,cAAA,SAAc,EAAW,GACvB,GAAM,GAAU,EAEhB,MAAO,GAAA,OACH,EAAmB,EAAmB,EAAS,OAC/C,KAAK,aAAe,gBAAkB,OAAS,SAzO9C,EAAA,UAAY,iBA2OrB,GA7OoC,IA+OpC,EAAA,cAAkB,cAAc,IAKhC,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,GAAI,IAAe,GAEhC,SAAA,EAAA,KAAA,KAAM,GAAA,GAAI,EAAI,CAAE,KAAI,MAAwB,OAIvC,SAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IAZV,EAAA,UAAY,aAcrB,GAhBgC,IAkBhC,EAAA,cAAkB,cAAc,ICjkBhC,AAkDA,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,KAAK,IAAI,KAAK,IAAI,EAAK,KAAM,GAAI,GAE7C,EAAK,WAAa,EAAK,WACvB,EAAK,KAAO,EAAK,KACjB,EAAK,gBAAkB,KAGf,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,KAAK,YAAc,KACrB,MAAO,MAAK,WAId,OAFM,GAAa,EAAM,MACnB,EAAoB,GACjB,EAAI,EAAG,EAAI,KAAK,WAAW,OAAQ,EAAE,EAC5C,EAAW,KACP,KAAK,WAAW,IAAM,KAAO,EAAW,GAAK,KAAK,WAAW,IAEnE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAI,EAAK,MAAQ,EAAK,KAAO,GAC/B,GAAM,GACF,EAAO,UAAe,KAAO,GAAQ,EAAO,SAC1C,EAAa,EAAK,cAAc,GAChC,EAAS,GACX,WAAM,MAAA,IAAU,EAAO,EAAK,KAAM,EAAY,EAAK,OACnD,WAAM,MAAA,IAAO,GACjB,MAAO,GAET,MAAO,MAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,KACX,WAAY,KAAK,WACjB,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAxDf,EAAA,UAAY,UA0DrB,GA5D6B,IA6D7B,EAAA,cAAc,cAAc,IA4D5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAIpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,CAAC,KAAM,MAGjB,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAM,GAAa,EAAM,MACzB,MAAO,CAAC,EAAW,GAAI,EAAG,EAAW,KAThC,EAAA,UAAY,mBAWrB,GAbsC,IActC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAmBzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAhBM,EAAA,WAA2B,KAC3B,EAAA,QAAU,GAGV,EAAA,OAAwB,KACxB,EAAA,KAAsB,KAErB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAQrD,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MACnD,EAAK,UAAY,MAGnB,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAK,gBAAkB,CAAC,EAAW,EAAK,UAG1C,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GAAc,EAAK,YACrC,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAK,SAEtB,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,gBAAkB,GAEvB,EAAK,UAAY,CAAC,CAAC,QAAS,MAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAe,EAAW,EAAW,OAAS,GACpD,AAAI,KAAK,QAAU,MACjB,MAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAc,KAAK,OAAQ,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,kBAIzC,KAAK,UAAY,CAAC,CAAC,QAAS,EAAG,KAAI,GAAA,GAAG,EAAC,IAAK,EAAY,KACxD,KAAK,MAAQ,IAGf,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,SAAY,EAAY,OAAS,GAAK,KAAK,MACpC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAM,GAAQ,GAAoB,GAC5B,EACF,GAA2B,EAAK,WAAW,gBAC3C,EAEJ,MAAI,IAAuB,KACzB,EAAS,GACL,EAAO,EAAK,OAAO,OAAQ,EAC3B,EAAK,KAAO,EAAK,KAAK,OAAS,MAEnC,GAAS,GAAM,EAAO,EAAK,OAAO,QAClC,AAAI,EAAK,MAAQ,MACf,GAAS,GAAU,EAAQ,EAAK,KAAK,SAEvC,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,KAI5B,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArHF,EAAA,UAAY,QAuHrB,GAzH2B,IA0H3B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAO,GAAQ,GACf,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,UAAY,CAAC,CAAC,QAAS,IAC5B,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,OAAkB,GAAA,EAAA,EAAA,EAAW,MAAM,GAAjB,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAO,KACT,KAAM,IAAI,GACN,4DACA,SAAQ,EAAW,MAAM,GAAE,oCAC3B,mFAIR,MAAO,CAAC,EAAW,GAAI,GAAU,EAAY,KAG/C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAK,aAAe,iBAAmB,EAAM,KAAO,GAEtD,OADM,GAAwB,CAAC,GACtB,EAAI,EAAG,EAAI,EAAM,KAAM,EAAE,EAChC,EAAY,KAAK,GAEnB,EAAY,KAAK,GACjB,EAAQ,EAAM,UAAU,GAG1B,MAAO,IAAe,MAI1B,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,GACzC,AAAI,KAAK,YAAc,MACrB,GAAO,WAAgB,KAAK,YAE9B,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA/CF,EAAA,UAAY,UAiDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAK9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,WAAa,GAAc,EAAK,cAGvC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAK,WAAW,MAAM,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,GAAoB,KAAK,aAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArBF,EAAA,UAAY,aAuBrB,GAzBgC,IA0BhC,EAAA,cAAc,cAAc,IAc5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAKhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,EAAI,EAAK,EACd,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,KAAK,EAAG,EAAW,KAG5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAS,GAAoB,GACtB,GAAS,EAAQ,EAAK,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,EAAG,KAAK,GAEJ,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,eA4BrB,GA9BkC,IA+BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,YAAc,EAAK,YAGxB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,EAAE,EAC7C,AAAI,EAAK,UAAU,EAAK,YAAY,KAClC,GAAK,YAAY,GAAK,eAKpB,SAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,GAAM,GAAK,GAAO,MAiBnB,EAAA,UAAA,oBAAR,SAA4B,EAAmB,GAK7C,OAJM,GAAW,6CACX,EAAa,EAAY,QAC3B,EAAQ,EACR,EAAU,KACL,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAM,EAAW,GACvB,GAAI,KAAK,UAAU,GACjB,GAAI,IAAY,KACd,EAAU,MAEV,MAAM,IAAI,GAAW,gDAGvB,IAAS,EAIb,GAAM,GAAe,GAAU,GAC/B,GAAI,IAAY,MACd,GAAI,IAAU,GAAK,EAAe,IAAU,EAC1C,KAAM,IAAI,GAAW,GAEvB,EAAW,GAAW,EAAe,UAC5B,IAAiB,EAC1B,KAAM,IAAI,GAAW,GAGvB,MAAO,IAGT,EAAA,UAAA,mBAAA,SAAmB,GAEjB,OADI,GAAiB,GACZ,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,GAAI,KAAK,UAAU,EAAW,KAC5B,EAAiB,GACjB,MAIJ,MAAI,GACK,EAAW,MAAM,EAAG,GAAG,OAAO,KAAK,aAEnC,EAAW,MAAM,EAAG,GAAG,OAC1B,KAAK,oBAAoB,EAAW,MAAM,GAAI,KAAK,eAI3D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAc,EAAW,MAAM,EAAG,GAAG,OACvC,EAAK,oBAAoB,EAAW,MAAM,GAAI,EAAK,cACvD,MAAO,GAAM,QAAQ,MAIzB,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlGF,EAAA,UAAY,UAoGrB,GAtG6B,IAuG7B,EAAA,cAAc,cAAc,IAY5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAM3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,OACN,mFAGN,GAAI,CAAC,MAAM,QAAQ,EAAK,MACtB,KAAM,IAAI,OACN,oEACG,GAAK,KAAI,cAIlB,GAAM,GAAwB,GAAM,EAAG,EAAK,KAAK,OAAS,GAC1D,GAAI,CAAC,EAAA,KAAK,YAAY,EAAK,KAAK,QAAQ,OAAQ,GAC9C,KAAM,IAAI,OACN,+BAAiC,KAAK,UAAU,EAAK,MACrD,8DAGN,SAAK,KAAO,EAAK,KACjB,EAAK,mBAAqB,CAAC,GAAG,OAAO,EAAK,MAC1C,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,EAAK,KAAK,OAAS,OAG5D,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,YAAK,KAAK,QAAQ,SAAC,EAAa,GAC9B,EAAY,EAAI,GAAM,EAAqB,KAEtC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,UAAU,GAAoB,GAAS,KAAK,qBAGrD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjDF,EAAA,UAAY,UAmDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,KACV,EAAK,UAAY,EAAK,WAAa,KAAO,EAAI,EAAK,UAEnD,EAAK,UAAY,IAIrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,UAAW,KAAK,WAChC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACb,MAAO,GAAA,IAAI,EAAA,SAAS,EAAO,KAAK,WAAY,IAG9C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACP,EAAW,GACX,EAAc,EAAA,IAAI,EAAA,SAAS,EAAO,EAAK,WAAY,EAAM,GACzD,EAAS,EAAM,IAAI,EAAY,OAAO,EAAM,QAClD,MAAO,MAtCJ,EAAA,UAAY,UAyCrB,GA3C6B,IA4C7B,EAAA,cAAc,cAAc,IClqB5B,AA2EA,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAgB7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GATM,EAAA,WAA4B,KAE3B,EAAA,+BACL,gBAME,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MAKrD,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,AAAI,EAAK,aAAe,KAGtB,EAAK,gBAAkB,CAAC,EAAW,MAInC,EAAK,gBACD,CAAC,GAAW,OAAO,GAAqB,EAAK,cAGrD,SAAK,SAAW,EAAK,SACrB,GAAoC,EAAK,SAAU,YACnD,EAAK,UAAY,EAAK,UACtB,GAAoC,EAAK,UAAW,aACpD,EAAK,sBAAwB,GACzB,EAAK,uBAAyB,EAAK,gCACvC,EAAK,sBAAwB,GAAe,EAAK,uBACjD,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,qBAAuB,GAAc,EAAK,sBAC/C,EAAK,SAAW,EAAK,SACrB,EAAK,gBAAkB,EAAK,SAC5B,EAAK,YAAc,EAAK,cAGnB,SAAA,UAAA,MAAP,SAAa,GACX,KAAK,WAAa,KAAK,UACnB,aAAc,CAAC,KAAK,SAAU,KAAK,WAAY,KAAK,MACpD,KAAK,sBAAuB,KAAK,sBAAuB,GACxD,KAAK,sBACT,KAAK,MAAQ,IAKL,EAAA,UAAA,6BAAV,SAAuC,KAEvC,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAK,GAAK,SAGR,GAAS,GAAoB,GACtB,EAAA,SAAS,EAAQ,EAAA,UAAU,KAH3B,QAQb,EAAA,UAAA,mBAAA,SAAmB,GAEjB,GADA,EAAa,GAAmB,GAC5B,KAAK,aAAe,KACtB,MAAW,GAAU,OAAA,CAAE,KAAK,YAG9B,GAAM,GAAmB,GAAqB,KAAK,aACnD,GAAI,EAAO,SAAW,EAAW,OAAS,EACxC,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IAG7B,OADI,GAAI,EACC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAK,EAAO,GACZ,EAAK,EAAW,EAAI,GAC1B,GAAK,GAAM,MAAU,GAAM,MAAU,IAAO,EAC1C,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IACxB,AAAI,GAAM,MACf,GAAO,GAAK,GAEd,IAGJ,MAAA,CAAQ,EAAW,IAAE,OAAK,EAAM,CAAE,KAAK,aAGzC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,AAAI,EAAM,QAAU,SAClB,GAAQ,GAAO,EAAO,UAExB,GAAM,GAAS,GAAS,EAAK,WAAW,OAAQ,EAAM,QACtD,MAAO,GAAO,QACV,GAAmB,EAAK,mBAAmB,EAAM,YAIzD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,UAAW,KAAK,UAChB,sBAAuB,GAAqB,KAAK,uBACjD,sBAAuB,GAAqB,KAAK,uBACjD,oBAAqB,GAAqB,KAAK,qBAC/C,qBAAsB,GAAoB,KAAK,sBAC/C,SAAU,KAAK,SACf,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjIF,EAAA,UAAY,YAmIrB,GArI+B,IAsI/B,EAAA,cAAc,cAAc,ICjN5B,AA+BA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAGlC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAQ,KAAG,KACjB,SAAK,gBAAkB,KAOf,SAAA,UAAA,cAAV,SAAwB,GACtB,KAAM,IAAI,KAaJ,EAAA,UAAA,gCAAR,SAAwC,EAAe,GACrD,GAAI,GAAU,MAAQ,GAAU,KAC9B,MAAO,MACF,GAAI,EAAO,OAAS,EAAO,OAChC,MAAO,MAAK,gCAAgC,EAAQ,GAC/C,GAAI,EAAO,SAAW,EAC3B,MAAO,GAGT,OADM,GAAqB,EAAO,MAAM,EAAG,EAAO,OAAS,EAAO,QACzD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,EAAO,OAAS,EAAO,OAAS,GAC3C,EAAI,EAAO,GACjB,GAAI,GAAK,MAAQ,GAAK,MAAQ,EAAI,GAAK,EAAI,EACzC,EAAY,KAAK,cACR,IAAM,EACf,EAAY,KAAK,WACR,IAAM,EACf,EAAY,KAAK,QAEjB,GAAI,IAAM,EACR,KAAM,IAAI,GACN,wDACA,KAAK,UAAU,GAAU,IAAM,KAAK,UAAU,IAEpD,EAAY,KAAK,IAGrB,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GAOJ,GALA,AAAI,MAAM,QAAQ,IAAe,CAAC,MAAM,QAAQ,EAAW,KAEzD,GAAa,CAAC,GAAmB,KAEnC,EAAa,EACT,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,mEACA,SAAQ,EAAW,OAAM,eAM/B,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAI1B,GADA,EAAa,GAAqB,GAC9B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,qDACA,6BAA4B,KAAK,UAAU,GAAW,MAK5D,OAFI,GACA,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GAC9C,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,GAAM,GAAW,EAAW,IAAI,SAAA,GAAS,MAAA,GAAM,SAC/C,AAAI,EAAW,QAAQ,QAAU,IAC7B,GAAqB,GAAU,SAAW,EAC5C,KAAK,gBAAkB,GAEvB,KAAK,gBAAkB,IAI3B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAK,iBACP,GAAM,GAA2B,GAC3B,EAAY,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,OAC5C,GAAI,EAAU,QAAQ,QAAU,IAI9B,OADM,GAAU,GAAc,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAEZ,OAFO,GAAC,EAAA,GACF,EAAQ,EAAE,KACP,EAAI,EAAG,EAAI,EAAU,EAAO,EAAE,EACrC,EAAI,GAAa,EAAG,GAEtB,EAAe,KAAK,GAEtB,MAAO,GAAK,cAAc,QAK1B,OADI,GAAa,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,KAChB,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAY,EAAO,GACnB,EAAW,EAAO,MAAM,GAAG,OAAO,CAAC,IACrC,EAAc,EAAE,QAChB,CAAC,GAAW,OAAO,GAAoB,EAAO,MAAM,MACxD,EAAc,EAAA,UAAc,EAAa,CAAC,EAAG,IAC7C,EAAc,EAAY,QAAQ,GAClC,EAAe,KAAK,GACpB,EAAa,WACJ,EAAQ,GACjB,GAAM,GAAO,GAAgB,EAAG,GAAO,OAAO,CAAC,IAC/C,EAAe,KAAK,EAAA,UAAc,EAAG,IACrC,EAAa,OAGb,GAAe,KAAK,GAGxB,GAAI,GAAI,EAAK,cAAc,GACrB,EAAQ,EAAE,KAChB,GAAI,GAGF,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAQ,EAAO,OACf,EAAY,EAAO,EAAQ,GAC3B,EACF,CAAC,GAAW,OAAO,EAAO,MAAM,EAAG,EAAO,OAAS,IACvD,EAAI,EAAA,UAAc,EAAE,QAAQ,CAAC,GAAI,IAAa,CAAC,EAAG,IACzC,QAAQ,WACR,EAAQ,GACjB,GAAM,GAAO,CAAC,EAAQ,GAAG,OAAO,GAAgB,EAAG,EAAQ,IAC3D,EAAI,EAAA,UAAc,EAAG,IAGzB,MAAO,QAGT,OAAO,GAAK,cAAc,MAKhC,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,EACb,GAAI,GACJ,AAAI,EAAW,IAAM,KACnB,EAAc,KAEd,EAAc,EAAW,GAAG,MAAM,GAEpC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAG1B,SAAa,GAAqB,GAClC,AAAI,EAAW,SAAW,EACxB,EAAc,EAAW,OAAO,GAEhC,EAAc,CAAC,MAAM,OAAO,GAEvB,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6BAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+BAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,kGAEA,KAAI,EAAO,OAAM,OAAO,EAAK,OAAM,MAEzC,GAAI,EAAK,MAAM,SAAA,GAAK,MAAA,IAAK,OACvB,MAAO,MAET,EAAO,EAAK,IAAI,SAAA,GAAK,MAAA,IAAK,KAAO,EAAI,EAAA,WAAe,EAAG,KAEvD,OADI,GAAS,EAAK,GACT,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAS,EAAA,WAAe,EAAQ,EAAK,IAEvC,MAAO,MAGb,GAlOoC,IAoOpC,GAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,MAcrB,GAhByB,IAiBzB,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAG5B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,WAcrB,GAhB8B,IAiB9B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,GAAA,IAAQ,EAAI,EAAO,OAAQ,MAX/B,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA2D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IAiE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAM/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAJJ,SAAA,aAAe,GAKtB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,MACxD,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,yEAGN,EAAa,EAGb,OADI,GAAe,GACC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,GAAS,MACX,EAAe,GACf,OAGJ,GAAI,EACF,OAIF,OADM,GAAoB,GACjB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAyB,EAAW,GAAG,QAC7C,EAAuB,OAAO,KAAK,KAAM,GAEzC,OADI,GAAS,GACO,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAA,KAAK,YAAY,EAAO,IAC1B,EAAS,GACT,OAGJ,AAAK,GACH,EAAS,KAAK,GAGlB,GAAI,EAAS,OAAS,EACpB,KAAM,IAAI,GACN,4GAEA,KAAK,UAAU,KAIb,EAAA,UAAA,cAAV,SAAwB,GAAxB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAO,IAAc,EAAQ,EAAK,SAItC,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,KAC1D,KAAM,IAAI,GACN,+DAON,OALM,GAAc,EACd,EAAc,EAAY,GAAG,QAC7B,EAAO,KAAK,KAAO,EAAI,EAAY,OAAS,KAAK,KAAO,KAAK,KAG/C,EAAA,EAAA,EAAA,EAAY,MAAM,GAAlB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAY,IAAS,MAAQ,EAAM,IAAS,MAC9C,EAAY,GAAQ,KACpB,MAEF,EAAY,IAAS,EAAM,GAE7B,MAAO,IAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6CAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+CAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,mCAAmC,EAAK,OAAM,KAC9C,8BAA6B,EAAO,OAAM,MAEhD,MAAO,GAAA,KAAS,WACd,GAAI,GAAe,GAOnB,GANA,EAAK,QAAQ,SAAA,GACX,GAAI,GAAK,MACP,EAAe,GACf,UAGA,EACF,MAAO,MAGT,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAK,IAAM,KAEb,EAAY,KAAK,EAAA,SAAa,EAAO,IAAI,OAAO,SAC3C,AAAI,EAAK,GAAG,KAAO,EAAO,GAAG,KAElC,EAAY,KAAK,EAAA,WAAe,EAAK,GAAI,KAEzC,EAAY,KAAK,EAAK,IAG1B,GAAM,GAAoB,EAAA,OAAW,EAAa,EAAK,MACvD,MAAO,GAAA,IAAQ,EAAmB,GAAI,OAI1C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,MAET,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArIF,EAAA,UAAY,cAuIrB,GAzIiC,IA0IjC,EAAA,cAAc,cAAc,IAuF5B,YAAuB,EAAc,GACnC,KAAO,EAAO,GACZ,GAAQ,EAEV,MAAO,GAGT,YAAkB,EAAW,EAAW,GACtC,GAAI,EAAE,MAAM,OAAS,GAAK,EAAE,MAAM,OAAS,EACzC,KAAM,IAAI,IACN,oEAeN,GAbA,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAC3B,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAE3B,AAAI,MAAO,IAAS,UAClB,GAAO,CAAC,EAAM,IAGZ,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,KAAM,IAAI,IACN,+DAGN,GAAM,GAAQ,EAAE,MAAM,OAChB,EAAQ,EAAE,MAAM,OACtB,AAAI,GAAQ,MAEV,GAAO,CAAC,EAAQ,EAAG,EAAQ,IAE7B,GAAM,GAAY,EAElB,MAAO,GAAA,KAAS,WACd,GAAI,GACJ,GAAI,EAAQ,GACV,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,YACpB,EAAQ,GACjB,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,QAE7B,GAAO,EAGT,GAAI,GACJ,GAAI,EAAE,MAAM,SAAW,GAAK,EAAE,MAAM,SAAW,EAC7C,AAAI,EAAU,KAAO,EAAU,GAC7B,EAAM,EAAE,IAAI,GAAG,IAAI,EAAU,IAE7B,EAAM,EAAE,UAAU,CAAC,EAAG,IAAI,IAAI,GAAG,IAAI,EAAU,SAGjD,GAAM,GAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EACzC,EAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EAC/C,EAAM,EAAE,OAAO,EAAG,EAAM,GAG1B,GAAI,EAAO,GACT,GAAI,GAAG,OACP,AAAI,EAAQ,EACV,EAAM,EAAQ,EAAQ,EAEtB,EAAM,EAAQ,EAGhB,OADM,GAAwB,GACrB,EAAI,EAAK,EAAI,EAAM,EAAM,EAAE,EAClC,EAAY,KAAK,GAEnB,EAAM,EAAI,QAAQ,GAEpB,MAAI,GAAI,MAAM,SAAW,GACvB,GAAM,EAAI,WAAW,IAEhB,IAIX,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,EAAK,KACjB,EAAK,UAAY,EAAK,WAAa,KAAO,GAAQ,EAAK,UACvD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAS,EAAW,GACpB,EAAS,EAAW,GAC1B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,GAAI,EAAO,EAAK,MAAQ,EAAO,EAAK,IAClC,KAAM,IAAI,GACN,8BACG,GAAO,EAAK,IAAG,QAAQ,EAAO,EAAK,OAIpC,EAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,qDACA,iBAAgB,EAAO,OAAM,eAGnC,GAAI,GAAK,EAAO,GACZ,EAAK,EAAO,GACZ,EACJ,MAAK,OAAM,QAAQ,KAAK,MAMtB,EAAO,KAAK,KAAK,IACN,SAAC,EAAM,GAAM,MAAA,IACT,EAAM,EAAO,GAAG,MAAM,UAPrC,EAAO,CACL,GAAc,KAAK,KAAM,EAAG,MAAM,QAClC,GAAc,KAAK,KAAM,EAAG,MAAM,SAOtC,AAAI,KAAK,WACP,GAAK,GAAY,EAAI,EAAK,IAC1B,EAAK,GAAY,EAAI,EAAK,KAErB,GAAS,EAAI,EAAI,IAGlB,EAAA,UAAA,cAAR,SAAsB,EAAe,GACnC,GAAI,GACJ,MAAK,OAAM,QAAQ,KAAK,MAQtB,EAAO,KAAK,KANZ,EAAO,CACL,GAAc,KAAK,KAAM,EAAO,QAChC,GAAc,KAAK,KAAM,EAAO,SAM7B,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAU,EAAW,GAAa,QAClC,EAAU,EAAW,GAAa,QACxC,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAG,GACjB,GAAM,GAAc,EAAO,OAAO,GAClC,MAAI,GAAY,SAAW,GACzB,EAAY,KAAK,GAEZ,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,OAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,KACb,UAAa,KAAK,WAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA7GF,EAAA,UAAY,MA+GrB,GAjHyB,IAkHzB,EAAA,cAAc,cAAc,IChjC5B,AA2BA,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAKjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,OAAS,EAAK,SAGrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,OAAQ,KAAK,QAC7B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAS,WACX,MAAA,IAAe,EAAM,MAAO,EAAG,EAAK,QAAQ,IAAI,IAC9C,EACF,GAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAC9D,MAAO,MA5BJ,EAAA,UAAY,gBA+BrB,GAjCmC,IAkCnC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAS,WACb,GAAM,GAAS,KAAK,KAAK,EAAK,KAAQ,GAAI,EAAK,OAC/C,MAAO,GAAM,IAAI,GAAe,EAAM,MAAO,EAAG,KAElD,MAAO,IAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAEnE,MAAO,MA/BJ,EAAA,UAAY,kBAkCrB,GApCqC,IAqCrC,EAAA,cAAc,cAAc,IAyC5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAMhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,KACjB,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,YAAc,GAAoB,GAAQ,OAGxD,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAa,EAAK,eAAe,GAEjC,EAAgB,WACpB,GAAM,GAAQ,GAAoB,GAE5B,EAAQ,mBACR,EAAQ,mBAER,EAAS,CAAC,EAAQ,EAEpB,EAAU,EAAA,aAAa,EAAA,cAAc,GAAa,EAAK,MAE3D,EAAU,GAAO,EAAS,WAG1B,GAAM,GAAI,KAAA,IAAE,GAAI,EAAK,MAAS,GAAI,EAAK,KAAO,KAAA,IAAA,EAAU,IAAO,KACzD,EAAI,CAAC,EAAI,EAAS,EAAK,KAGvB,EAAI,EAAM,IAAI,GAAS,IAAI,EAAQ,IAAI,IAAI,IAAI,IAErD,MAAO,GAAE,IAAI,GAAG,IAAI,IAEtB,MAAO,IACH,EAAe,WAAM,MAAA,IAAoB,IACzC,EAAO,UAAe,IAE5B,MAAO,MAxDJ,EAAA,UAAY,eA2DrB,GA7DkC,IA8DlC,EAAA,cAAc,cAAc,IChN5B,YA4CI,EAAW,EAAc,EAAkB,EAAe,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GACJ,GAAI,EAAE,OAAS,EACb,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EAEpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EACpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,OAElC,MAAM,IAAI,IACN,2DAA2D,EAAE,KAAI,QAGvE,MAAO,GAoBT,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WACH,GAAM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EACF,GAAmB,EAAG,EAAM,EAAU,EAAM,EAAO,GACvD,MAAO,CAAC,EAAQ,EAAM,KAqBjC,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKH,OAJM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EAAwB,GACX,EAAA,EAAA,EAAA,GAAiB,EAAG,EAAE,MAAtB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAc,QAAQ,KAAU,GAClC,EAAY,KAAK,GAEjB,EAAY,KAAK,EAAE,MAAM,IAG7B,GAAM,GAAgB,EAAK,QAAQ,GAC7B,EAAoB,EAAS,QAAQ,GACrC,EACF,GAAS,KAAO,KAAO,EAAM,QAAQ,GACnC,EACF,GAAQ,KAAO,KAAO,EAAK,QAAQ,GACjC,EAAS,GACX,EAAG,EAAe,EAAmB,EACrC,EAAgB,GACpB,MAAO,CAAC,EAAQ,EAAM,iBAgB7B,EAAW,EAAe,EAAc,EACxC,GACF,MADE,AAAA,KAAA,QAAA,GAAA,MACE,EAAA,KAAK,YACD,EAAc,QAAQ,OAAQ,GAAiB,EAAG,EAAE,KAAO,IAC1D,GACH,EAAG,EAAO,EAAM,EAAe,GAE5B,GACH,EAAG,EAAO,EAAM,EAAe,GAsFvC,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAqBtC,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,IAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KAC1C,EAAK,SAAW,EAAK,UAAY,KAAO,IAAO,EAAK,SACpD,EAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,sBACD,GAAe,EAAK,uBAAyB,SACjD,EAAK,0BACD,GAAe,EAAK,2BAA6B,QACrD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAc,EAAK,iBAC1C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,oBAGvC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAO,KAAK,MAAQ,EAAI,KAAK,KAAQ,KAAK,KAAO,EAAW,OAC5D,EAAM,EAAW,GACvB,GAAI,GAAO,KACT,KAAM,IAAI,GACN,QAAQ,EAAI,+FAET,MAAK,UAAU,GAAW,MAEnC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAW,OAAQ,KAAI,GAAA,GAAG,EAAC,GAAO,EAAG,MAC/D,GAAM,GAAQ,CAAC,GACf,AAAI,KAAK,OACP,MAAK,MAAQ,KAAK,UACd,QAAS,EAAO,KAAM,KAAK,iBAAkB,KAAK,iBAClD,GAAM,KAAK,kBAEjB,AAAI,KAAK,QACP,MAAK,KAAO,KAAK,UACb,OAAQ,EAAO,KAAM,KAAK,gBAAiB,KAAK,gBAAiB,GACjE,KAAK,iBAEX,KAAK,WAAa,KAAK,UACnB,cAAe,EAAO,KAAM,KAAK,sBAAuB,KAAM,IAClE,KAAK,eAAiB,KAAK,UACvB,kBAAmB,EAAO,KAAM,KAAK,0BAA2B,KAChE,IACJ,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACvD,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAO,EAAW,OAClB,EAAgB,GAAiB,EAAG,GACpC,EAAO,EAAK,MAAQ,EAAI,EAAK,KAAQ,EAAK,KAAO,EACvD,EAAc,OAAO,EAAM,GAC3B,GAAM,GAAiB,GAA2B,EAAG,GACrD,EAAe,GAAQ,EAAW,GAElC,GAAM,GAAsB,EAAc,QAC1C,EAAoB,OACpB,GAAM,GAAoB,CAAC,EAAA,KAAK,YAC5B,EAAqB,GAAiB,EAAG,GAAM,MAAM,EAAG,EAAO,IAE7D,EAAmC,WACvC,GAAI,GACF,GAAM,GACF,EAAK,WAAW,OAAO,QAAQ,GAC7B,EACF,EAAK,eAAe,OAAO,QAAQ,GACjC,EACF,EAAK,OAAS,EAAK,KAAK,OAAO,QAAQ,GAAkB,KACvD,EACF,EAAK,MAAQ,EAAK,MAAM,OAAO,QAAQ,GAAkB,KAC7D,MAAO,IACH,EAAO,EAAqB,EAC5B,EAAe,EAAgB,EAAK,aAExC,OAAO,IACH,EAAO,EAAK,WAAW,OAAQ,EAAK,eAAe,OACnD,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACrC,EAAK,OAAS,KAAO,KAAO,EAAK,MAAM,OAAQ,EAAK,UAI5D,GAAI,CAAC,EACH,MAAO,KAGH,GAAA,GAAA,GAAA,EAAA,EAAA,MAAA,OAAA,EAAA,KAAA,OAAA,EAAA,EAAA,SAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAM,EAAA,EAAA,GAIvB,EACF,SAAC,EAAyB,EAAe,GACvC,EAAA,KAAS,WACP,GAAM,GAAQ,EAAI,EACZ,EAAY,EAAS,OACrB,EAAc,EAAU,IAAI,GAAO,IAAI,GAC7C,EAAS,MAAM,EAAU,IAAI,OAU/B,EAA8B,WAClC,EAAgB,EAAK,WAAY,EAAM,EAAK,UAC5C,EAAgB,EAAK,eAAgB,EAAU,EAAK,WAEtD,WAEO,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,SAAU,KAAK,SACf,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,sBAAuB,GAAqB,KAAK,uBACjD,0BACI,GAAqB,KAAK,2BAC9B,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,eAAgB,GAAoB,KAAK,gBACzC,gBAAiB,GAAoB,KAAK,kBAEtC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArKF,EAAA,UAAY,qBAuKrB,GAzKwC,IA0KxC,EAAA,cAAc,cAAc,IAkD5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAgBtC,WAAY,GAAZ,GAAA,GAAA,KAOE,GANA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KACtC,MAAO,GAAK,MAAS,UACvB,GAAI,CAAC,OAAO,UAAU,EAAK,MACzB,KAAM,IAAI,OACN,gDAAgD,EAAK,cAElD,MAAM,QAAQ,EAAK,MAC5B,OAAmB,GAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,6CACA,iBAAgB,KAAK,UAAU,EAAK,YAI5C,MAAM,IAAI,OACN,2DACA,iBAAgB,KAAK,UAAU,EAAK,QAG1C,SAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,kBAE5C,EAAK,gBAAkB,KAGlB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAQ,EAAW,OAGzB,AAAI,MAAO,MAAK,MAAS,UACvB,MAAK,KAAO,CAAC,KAAK,OAEpB,OAAS,GAAI,EAAG,EAAI,KAAK,KAAK,OAAQ,EAAE,EACtC,AAAI,KAAK,KAAK,GAAK,GACjB,MAAK,KAAK,IAAM,GAKpB,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,GAAK,GAAQ,EACtB,KAAM,IAAI,OAAM,iBAAiB,GAGrC,GAAI,KAAK,KAAK,SAAW,GAAqB,KAAK,MAAM,OACvD,KAAM,IAAI,OAAM,4BAA4B,KAAK,MAGnD,GAAM,GAAa,KAAK,KAAK,IAAI,SAAA,GAAQ,MAAA,GAAW,KAE9C,EAAY,GAClB,AAAI,KAAK,MACP,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAE3B,KAAK,MAAQ,KAEf,AAAI,KAAK,OACP,KAAK,KAAO,KAAK,UACb,OAAQ,EAAY,UAAW,KAAK,gBACpC,KAAK,gBAAiB,GAE1B,KAAK,KAAO,KAGd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACQ,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAQ,EAAW,OAEzB,MAAO,GAAA,KAAK,WAIV,OAHM,GAAW,GACb,EAAA,EAAA,QAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,SACL,EAAiB,GAA2B,EAAG,GACnC,EAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAe,GAAO,EAAW,GAuBnC,OApBM,GAAY,SAAC,GACjB,MAAI,IAAK,MAAQ,EAAE,MAAM,SAAW,GAChC,EAAK,OAAS,CAAC,EAAQ,GAClB,EAAE,QAAQ,GAEV,GAIP,EAAQ,EAAU,EAAK,MAAM,QAC7B,EAAS,EAAU,EAAK,KAAK,QAQ3B,EAA0B,GAC1B,EAA8B,GAC3B,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,AAAK,EAAK,KAAkB,QAAQ,KAAO,GACzC,GAAc,KAAK,EAAW,IAC9B,EAAkB,KAAK,IAEvB,GAAc,KAAK,GACnB,EAAkB,KAAK,EAAW,KAGtC,SAAO,EAAK,KAAK,GACjB,EAAW,EAAS,KAAK,GACzB,EAAQ,EAAM,KAAK,GACnB,EAAS,EAAO,KAAK,GAEd,GACH,EAAO,EAAM,EAAU,EAAQ,EAAO,EAAK,YAInD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,mBAExC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjKF,EAAA,UAAY,qBAmKrB,GArKwC,IAsKxC,EAAA,cAAc,cAAc,IC9nB5B,YAoEI,EAAW,EACX,GACF,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,kEACG,GAAE,KAAI,eAMf,GAHA,AAAI,GAAW,MACb,GAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,KAErB,EAAQ,SAAW,GAAK,EAAQ,GAAG,SAAW,GAC9C,EAAQ,GAAG,SAAW,EACxB,KAAM,IAAI,GACN,+GAON,GAHA,AAAI,GAAc,MAChB,GAAa,MAEX,IAAe,gBAAkB,IAAe,gBAClD,KAAM,IAAI,GACN,wBAAwB,EAAU,mEAIxC,GAAI,GACJ,MAAI,KAAe,gBACjB,EAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,IAE/C,EAAU,CAAC,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,GAAI,CAAC,EAAG,IAG1C,EAAA,IAAQ,EAAG,KA6BtB,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAMjC,WAAY,GAAZ,GAAA,GAAA,KAUE,GATA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,WACD,EAAK,YAAc,KAAO,KAAoB,EAAK,WAGnD,EAAK,SAAW,KAClB,EAAK,QAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,YACnB,MAAO,GAAK,SAAY,SACjC,EAAK,QACD,CAAC,CAAC,EAAK,QAAS,EAAK,SAAU,CAAC,EAAK,QAAS,EAAK,eAGvD,GADA,EAAK,QAAU,EAAK,QAChB,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,6DACA,sBAAqB,EAAK,QAAQ,OAAM,YAG9C,GAAI,GAAa,OACb,EAAY,OAChB,GAAI,MAAO,GAAK,QAAQ,IAAO,SAC7B,EAAgB,CAAC,EAAK,QAAQ,GAAI,EAAK,QAAQ,IAC/C,EAAe,CAAC,EAAK,QAAQ,GAAc,EAAK,QAAQ,SAIxD,GAFA,EAAK,QAAU,EAAK,QAEhB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,gEACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAIrD,GAFA,EAAgB,EAAK,QAAQ,GAEzB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,+DACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAErD,EAAe,EAAK,QAAQ,GAE9B,EAAK,QAAU,CAAC,EAAe,GAEjC,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAEhC,GAAI,GACA,EACJ,MAAI,MAAK,aAAe,gBACtB,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,IAE5C,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,MAIlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,IACF,GAAoB,GAAS,EAAK,QAAS,EAAK,eAG1D,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAnGF,EAAA,UAAY,gBAqGrB,GAvGmC,IAwGnC,EAAA,cAAc,cAAc,IC5O5B,YA0CI,EAAW,EAA4B,EACvC,EAAuB,EACvB,GACF,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,IAEhB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAKb,EAAI,GAAsB,EAAG,GAC7B,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MAEf,EAAI,EAAA,QAAY,EAAe,EAAU,EAAS,GAIlD,EAAI,EAAA,QAEA,EAA0B,EAAU,EAAS,GAEnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBAeP,EAAa,EACb,EAAoC,EACpC,EAAyB,GAC3B,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,EAAG,IAEnB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAIb,EAAI,GAAsB,EAAa,GACvC,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MACf,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAExC,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAE1C,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAsBX,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAWtC,WAAY,GAAZ,GAAA,GAAA,KAKE,GAJA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,GAElB,EAAA,EAAA,KAAA,KAAM,IAAK,KACP,MAAO,GAAK,UAAa,SAC3B,EAAK,SAAW,CAAC,EAAK,kBAEpB,MAAM,QAAQ,EAAK,WAClB,EAAK,SAAsB,SAAW,GACvC,MAAQ,GAAK,SAAsB,IAAO,SAC5C,EAAK,SAAW,EAAK,aAErB,MAAM,IAAI,GACN,qGAEA,IAAG,KAAK,UAAU,EAAK,YAG7B,GADA,GAAsB,EAAK,SAAU,YACjC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBAEhB,MAAO,GAAK,SAAY,SAC1B,EAAK,QAAU,CAAC,EAAK,iBAEnB,MAAM,QAAQ,EAAK,UAClB,EAAK,QAAqB,SAAW,GACtC,MAAQ,GAAK,QAAqB,IAAO,SAC3C,EAAK,QAAU,EAAK,YAEpB,MAAM,IAAI,GACN,oGAEA,IAAG,KAAK,UAAU,EAAK,WAG/B,UAAsB,EAAK,QAAS,WAEpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAS,GACX,EAAW,GAAI,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IAChE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAW,KAO5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,EAAS,GAAa,GAAoB,GAAS,GACnD,GAAM,GAAS,EAAK,gBAChB,GAAoB,GAAS,CAAC,EAAK,SAAS,GAAI,GAChD,CAAC,EAAK,QAAQ,GAAI,GAAI,EAAK,QAAS,gBAExC,MAAO,GAAA,QAAY,EAAQ,CAAC,OAIhC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,SAEV,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAxFwC,IA0FxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,IAEtB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,UACrB,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,SAErC,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAKrE,MAJA,GACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,GAErC,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,KAQlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAhFwC,IAkFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,EAAG,IAEzB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,SAAU,EAAK,UACpC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,QAAS,EAAK,SAEnD,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAOrE,MANA,GAAS,GACL,EAAQ,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,EAAM,GAE7C,CAAC,EAAW,GAAI,EAAQ,EAAM,EAAM,EAAW,KAS1D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GArFwC,IAuFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,eAcrB,GAhBkC,IAiBlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,mBAcrB,GAhBsC,IAiBtC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAC5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,EAAW,KAGpC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAEd,GAb8C,IAe9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAG1C,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,KAAS,EAAO,MARpB,EAAA,UAAY,yBAWrB,GAb4C,IAc5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAGtC,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,IAAQ,EAAO,MARnB,EAAA,UAAY,qBAWrB,GAbwC,IAcxC,EAAA,cAAc,cAAc,IAiB5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAE5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GAEjB,MADA,GAAa,EACT,KAAK,aAAe,eACf,CAAC,EAAW,GAAI,EAAW,IAE3B,CAAC,EAAW,GAAI,EAAW,KAItC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAGZ,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,KAAK,YAC3B,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA7B8C,IA+B9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAA5C,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,KAAS,EAAO,CAAC,EAAG,IAEpB,EAAA,KAAS,EAAO,CAAC,EAAG,OAR1B,EAAA,UAAY,yBAYrB,GAd4C,IAe5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAAxC,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,IAAQ,EAAO,CAAC,EAAG,IAEnB,EAAA,IAAQ,EAAO,CAAC,EAAG,OARzB,EAAA,UAAY,qBAYrB,GAdwC,IAexC,EAAA,cAAc,cAAc,IC5rB5B,AA4CA,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,GAAZ,GAAA,GAQE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAKf,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAIE,MAAI,MAAK,OAAS,KACT,KAAK,MAAM,UAEX,QAIX,SAAc,GAIZ,AAAI,KAAK,OAAS,MAChB,MAAK,MAAM,UAAY,oCAI3B,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,MAAM,kDAIpB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAIpB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WAEE,MAAQ,MAAK,MAAc,0CAK7B,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,MAAM,wCAKpB,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,MAAM,cAGpB,EAAA,UAAA,WAAA,SAAW,GACT,KAAK,MAAM,WAAW,IAGxB,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAS,CACP,UAAa,KAAK,MAAM,eACxB,OAAU,KAAK,MAAM,cAGnB,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,OAAS,MAChB,KAAK,MAAM,6BAA6B,IAKrC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAc,EAAO,MACrB,EAAQ,GAAY,EAAa,GACvC,MAAO,GAAO,MACd,GAAM,GAAY,CAAC,MAAK,GACxB,cAAO,OAAO,EAAW,GAClB,GAAI,GAAI,IAEnB,GAvGsC,IAyGtC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,oEACA,gBAAe,KAAK,UAAU,KAEpC,KAAK,UAAY,CAAC,CAAC,MAAO,IAC1B,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAChE,AAAK,KAAK,MAAM,OACd,MAAK,MAAM,MAAM,GACjB,KAAK,MAAM,MAAQ,IAErB,EAAA,UAAM,MAAK,KAAA,KAAC,IAGd,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAC1D,EACF,KAAK,MAAM,mBAAmB,GAC5B,EAAY,EAAW,GAC7B,MAAO,CAAC,EAAiB,GAAI,GAAW,OAAO,EAAiB,MAAM,KAGxE,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,EAAS,GAAoB,GAI7B,GAAM,GAAwB,SAAC,EAAgB,GAK7C,GAAM,GAAS,GAAoB,EAAK,MAAM,KAAK,EAAQ,IAC3D,MAAO,CAAC,EAAQ,KAEZ,EACF,GAAI,EAAM,EAAQ,GAAI,GAAyB,KAC3C,KAAsB,GACtB,IACF,EAAI,EAAW,GAGrB,MAAO,MArDJ,EAAA,UAAY,kBA0DrB,GA5DqC,IA6DrC,EAAA,cAAc,cAAc,gBAEgB,GAC1C,GACI,GAAiC,yBAA0B,GAmBjE,GAAM,IAA2D,SAEjE,GAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAUL,EAAc,EAAK,MAAM,YACzB,EAAqC,GAC3C,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,aAAe,GAAY,GAChC,EAAY,YACR,IAAY,cAAmB,IACnC,GAAM,GAAqC,GAW3C,GAVA,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,cAAgB,GAAY,GACjC,EAAK,aAAa,KAAO,WAAa,EAAK,aAAa,KACxD,EAAK,cAAc,KAAO,YAAc,EAAK,cAAc,KAE3D,EAAK,UAAY,EAAK,YAAc,OAChC,GACA,EAAK,UACT,GAA4B,EAAK,WAC7B,EAAK,QACP,KAAM,IAAI,IACN,mEAEN,SAAK,UAAY,EAAK,MAAM,SAC5B,EAAK,gBAAkB,EAAK,MAAM,gBAClC,EAAK,YAAc,EAAK,MAAM,YAC9B,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAClB,EAAK,UAAY,EAAK,MAAM,UAC5B,EAAK,aAAe,OAGtB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GAIZ,KAAK,WAAa,EAClB,AAAI,KAAK,cAAgB,MACvB,MAAK,aAAa,UAAY,GAEhC,AAAI,KAAK,eAAiB,MACxB,MAAK,cAAc,UAAY,oCAInC,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,aAAa,aAAa,OAClC,KAAK,cAAc,eAGzB,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAa,EAAQ,OACrB,EAAiB,KAAK,MAAM,EAAa,GAC/C,KAAK,aAAa,WAAW,EAAQ,MAAM,EAAG,IAC9C,KAAK,cAAc,WAAW,EAAQ,MAAM,KAG9C,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GACA,KAAK,aAAa,mBAAmB,GACzC,AAAM,MAAM,QAAQ,IAAgB,MAAM,QAAQ,EAAY,KAC5D,GAAc,CAAC,IAEjB,EAAc,EAEd,GAAI,GACA,EACA,EAiBJ,MAhBA,AAAI,MAAK,aACP,GAAa,EAAY,MAAM,IAC/B,EAAc,EAAY,GAI5B,EAAc,EACd,AAAI,KAAK,YAAc,SACrB,GAAY,EAAY,OAAS,IAAM,EACvC,EAAe,CAAC,IACX,AAAI,KAAK,WAAa,KAC3B,EAAe,CAAC,EAAa,EAAY,SAEzC,EAAe,CAAC,GAGd,KAAK,YACH,KAAK,WAAa,KACb,EAAa,OAAO,GAAY,OAAO,EAAW,SAEpD,CAAC,GAAa,OAAO,GAAY,OAAO,EAAW,SAErD,GAA+B,IAGxC,EAAA,UAAA,MAAA,SACI,EACA,GACF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAEX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAU1D,GATA,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAEzB,AAAI,MAAM,QAAQ,IAChB,GAAgB,EAAuC,MAAM,GAC7D,EAAU,EAAuC,IAG9C,IAAgB,MAAQ,EAAa,SAAW,IACjD,GAAa,KACf,MAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,GAE7B,GAAM,GAAiD,GACjD,EAA+B,GACrC,GAAI,GAAgB,MAClB,GAAM,GAAY,EAAa,OAC/B,GAAI,EAAY,EAAI,EAClB,KAAM,IAAI,GACN,iIAIN,EAAO,aAAkB,EACzB,EAAiB,KAAI,MAArB,EAAyB,GACzB,GAAM,GAAc,EACI,IAAI,SAAA,GAAS,MAAA,IAAI,IAAU,CAAC,MAAO,EAAM,UACjE,KAAK,aAAa,UAAY,EAAW,MAAM,EAAG,EAAY,GAC9D,KAAK,cAAc,UAAY,EAAW,MAAM,EAAY,GAC5D,EAAgB,KAAI,MAApB,EAAwB,GAE1B,GAAI,GAAa,KACf,KAAM,IAAI,IACN,yEAKN,OADM,GAAmB,EAAiB,YAAc,IACnC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,YAAkB,MAAmB,EACvC,KAAM,IAAI,GACN,gHAKR,GAAI,GAEF,GAAM,GAAY,CAAC,GAAQ,OAAO,GAC5B,EAAgB,KAAK,UAAU,OAAO,GAUtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GACF,EAAA,UAAM,MAAK,KAAA,KAAC,EAA0C,GAC1D,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAI/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAe,EAAO,aAExB,EACA,EACJ,GAAI,GAAgB,KAClB,EAAI,EAAK,aAAa,KAAK,EAAQ,GACnC,EAAO,EAAK,cAAc,KAAK,EAAQ,QAEvC,GAAM,GAAe,EAAa,MAAM,EAAG,EAAa,OAAS,GAC3D,EAAgB,EAAa,MAAM,EAAa,OAAS,GAC/D,EAAI,EAAK,aAAa,KAClB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KACjD,EAAO,EAAK,cAAc,KACtB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KAGnD,GAAI,GACJ,AAAI,EAAK,aACP,CAAI,MAAM,QAAQ,IAChB,GAAS,EAAE,MAAM,GAAG,OAAQ,EAAkB,MAAM,KAGtD,EAAK,EAAe,GACpB,EAAQ,EAAkB,IAG5B,AAAI,EAAK,iBACP,GAAO,EAAA,QAAY,EAAgB,IAGrC,GAAI,GAcJ,MAbA,AAAI,GAAK,YAAc,SACrB,EAAS,GAAc,CAAC,EAAa,IAChC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,GAAI,EAAA,IAAQ,EAAa,IACrC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,WAAa,MAC3B,GAAS,CAAC,EAAa,IAIrB,EAAK,YACH,EAAK,WAAa,KACZ,EAAoB,OAAO,GAE9B,CAAC,GAAkB,OAAO,GAE5B,KAIX,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,aAAa,cAClB,KAAK,cAAc,eAGrB,EAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,GAAU,KAAK,aAAa,KAAM,WAChC,EAAK,aAAa,MAAM,KAE1B,GAAU,KAAK,cAAc,KAAM,WACjC,EAAK,cAAc,MAAM,KAE3B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,SAAY,EAAyB,GAEnC,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAI,GAcJ,GAbA,AAAI,KAAK,gBACP,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,EAAM,GAEpB,EAAa,EAGf,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,KAAM,MAEpB,EAAa,KAGb,KAAK,aACP,GAAM,GAAS,KAAK,aAAa,OAC3B,EAAsB,EAAO,IAAI,SAAA,GAAS,MAAA,QAChD,MAAI,OAAM,QAAQ,GACT,EAAW,OAAO,GAAW,OAAO,GAEpC,CAAC,GAAY,OAAO,GAAW,OAAO,OAG/C,OAAO,IAIX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,aAAa,iBAAiB,OACtC,KAAK,cAAc,mDAGzB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,aAAa,oBAAoB,OACzC,KAAK,cAAc,sDAKzB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,cAAgB,MACvB,KAAK,aAAa,6BAA6B,GAEjD,AAAI,KAAK,eAAiB,MACxB,KAAK,cAAc,6BAA6B,IAIpD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,UAAa,KAAK,WAGd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAIF,EAAA,WAAP,SACI,EACA,GACF,GAAM,GACF,GAAY,EAAO,OAGvB,GAFA,MAAO,GAAO,MAEV,EAAO,cAAmB,KAC5B,KAAM,IAAI,IACN,4FAIN,GAAM,GAAkC,EACxC,SAAU,MAAW,EACd,GAAI,GAAI,IA7VV,EAAA,UAAY,gBA+VrB,GAjWmC,IAkWnC,EAAA,cAAc,cAAc,IC7kB5B,YAkE2B,GACzB,MAAO,IAAI,IAAW,eA6BJ,GAClB,MAAO,IAAI,IAAI,eAoBI,GACnB,MAAO,IAAI,IAAK,eAuBQ,GACxB,MAAO,IAAI,IAAU,eAwBD,GACpB,MAAO,IAAI,IAAM,eAmBK,GACtB,MAAO,IAAI,IAAQ,eA2BW,GAC9B,MAAO,IAAI,IAAgB,eAyBN,GACrB,MAAO,IAAI,IAAO,eAqBG,GACrB,MAAO,IAAI,IAAO,eAsCY,GAC9B,MAAO,IAAI,IAAgB,eAqBN,GACrB,MAAO,IAAI,IAAO,eAgCY,GAC9B,MAAO,IAAI,IAAgB,eAkCF,GACzB,MAAO,IAAI,IAAW,eA2BK,GAC3B,MAAO,IAAI,IAAa,eAeM,GAC9B,MAAO,IAAI,IAAgB,eAoCF,GACzB,MAAO,IAAI,IAAW,eAoCF,GACpB,MAAO,IAAI,IAAM,eAaK,GACtB,MAAO,IAAI,IAAQ,eAmCY,GAC/B,MAAO,IAAI,IAAiB,eAsBN,GACtB,MAAO,IAAI,IAAQ,eAiBQ,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,eA+BG,GACtB,MAAO,IAAI,IAAQ,eAcK,GACxB,MAAO,IAAI,IAAU,eAyBH,GAClB,MAAO,IAAI,IAAI,eAqBO,GACtB,MAAO,IAAI,IAAQ,eAuBO,GAC1B,MAAO,IAAI,IAAY,eAqBD,GACtB,MAAO,IAAI,IAAQ,eAqBG,GACtB,MAAO,IAAI,IAAQ,eAsBI,GACvB,MAAO,IAAI,IAAS,eAyBF,GAClB,MAAO,IAAI,IAAI,eA0BkB,GACjC,MAAO,IAAI,IAAmB,eAuBG,GACjC,MAAO,IAAI,IAAmB,eA2BF,GAC5B,MAAO,IAAI,IAAc,eAgBM,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eA0BO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAwBO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAYa,GACrC,MAAO,IAAI,IAAuB,eAiBG,GACrC,MAAO,IAAI,IAAuB,eAYD,GACjC,MAAO,IAAI,IAAmB,eAiBG,GACjC,MAAO,IAAI,IAAmB,eAYH,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eA4BN,GAClB,MAAO,IAAI,IAAI,eAgDO,GACtB,MAAO,IAAI,IAAQ,eA0BA,GACnB,MAAO,IAAI,IAAK,eAgDO,GACvB,MAAO,IAAI,IAAS,eA2BI,GACxB,MAAO,IAAI,IAAU,eAgDO,GAC5B,MAAO,IAAI,IAAc,eA8BA,GACzB,MAAO,IAAI,IAAW,eAoCO,GAC7B,MAAO,IAAI,IAAe,eA+DR,GAClB,MAAO,IAAI,IAAI,eAUe,GAC9B,MAAO,IAAI,IAAgB,eAMC,GAC5B,MAAO,IAAI,IAAc,eAiDK,GAC9B,MAAO,IAAI,IAAgB,GAItB,GAAM,IAAkB,GAClB,GAAkB,GAClB,GAAY,GACZ,GAAY,eA2BK,GAC5B,MAAO,IAAI,IAAc,eA2BK,GAC9B,MAAO,IAAI,IAAgB,eAkCA,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,glCClnDU,EAAe,GAC5C,MAAO,IAAuB,EAAO,eAoBJ,EAAe,GAChD,MAAO,IAA2B,EAAO,eAuBvC,EAAe,GACjB,MAAO,IAAkC,EAAO,eAqBd,EAAe,GACjD,MAAO,IAA4B,EAAO,eAcJ,EAAe,GACrD,MAAO,IAAgC,EAAO,eAsCtB,EAAe,GACvC,MAAO,IAAkB,EAAO,eAsCX,EAAe,GACpC,MAAO,IAAe,EAAO,eAwBC,EAAe,GAC7C,MAAO,IAAuB,EAAO,eAuBL,EAAe,GAC/C,MAAO,IAAyB,EAAO,eAsBrC,EAAe,GACjB,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAqBlB,EAAe,GAC9C,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,kSC1TxC,yCCAA,YAqBqB,GACnB,MAAO,IAAI,IAAK,eAYC,GACjB,MAAO,IAAgB,eAYN,GACjB,MAAO,IAAgB,+CChDzB,mBAkBuC,EAAA,EAAA,GAAvC,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KAEE,SAAA,MAAqB,OAErB,SAAA,UAAA,SAAA,SAAS,GACP,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,OAAM,yDAElB,KAAK,MAAQ,GAEjB,GAVuC,IAsEvC,YAAc,EAAiB,GAC7B,MAAO,GAAU,EAGnB,YAAiB,EAAiB,GAChC,MAAO,GAAU,qBAOgB,EAAA,EAAA,GAcjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GAHA,AAAI,GAAQ,MACV,GAAO,IAEL,EAAK,mBACP,KAAM,IAAI,IACN,sEAGN,SAAK,QAAU,EAAK,SAAW,WAC/B,EAAK,SAAW,KAAK,IAAI,EAAK,UAAY,GAC1C,EAAK,SAAW,EAAK,UAAY,EACjC,EAAK,QAAU,EAAK,SAAW,EAC/B,EAAK,KAAO,EAAK,MAAQ,OACzB,EAAK,SAAW,EAAK,SAErB,AAAI,CAAC,OAAQ,MAAO,OAAO,QAAQ,EAAK,QAAU,IAChD,SAAQ,KACJ,uBAAuB,EAAK,KAAI,8CAEpC,EAAK,KAAO,QAGd,AAAI,EAAK,OAAS,MAChB,EAAK,YAAc,GACd,AAAI,EAAK,OAAS,OAInB,EAAK,QAAQ,QAAQ,SAAW,GAHpC,EAAK,YAAc,GAMjB,EAAK,YAAc,GAIvB,AAAI,EAAK,cAAgB,IACvB,GAAK,UAAY,MAIf,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,aAAe,EACpB,AAAI,KAAK,UAAY,KACnB,KAAK,KAAO,KAAK,SAEjB,KAAK,KAAO,KAAK,cAAgB,GAAO,SAAW,mBAIjD,EAAA,UAAA,WAAN,SAAiB,EAAe,mGAC9B,MAAA,CAAA,EAAM,GAAqB,WAE3B,MAFA,GAAA,OACM,EAAU,KAAK,gBAAgB,GACjC,GAAW,KACb,CAAA,GAGF,CAAI,KAAK,YAAY,EAAU,KAAK,SAAU,KAAK,MACjD,MAAK,KAAO,EACZ,KAAK,KAAO,GAGZ,MAAK,OACL,AAAI,KAAK,MAAQ,KAAK,UACpB,MAAK,aAAe,EACpB,KAAK,MAAM,aAAe,gBAM1B,EAAA,UAAA,WAAN,SAAiB,sEACf,MAAI,MAAK,aAAe,GAAK,KAAK,SAChC,QAAQ,IAAI,SAAS,KAAK,aAAY,8BAIlC,EAAA,UAAA,gBAAR,SAAwB,GACtB,AAAI,GAAQ,MACV,GAAO,IAET,GAAM,GAAe,EAAK,KAAK,SAC/B,MAAI,IAAgB,MAClB,QAAQ,KACJ,4BAA4B,KAAK,QAAO,sBACxC,2BAA0B,OAAO,KAAK,KAErC,GAEX,GAzGmC,gBAsJL,GAC5B,MAAO,IAAI,IAAc,MAGd,IAAY,CAAC,cAAa,i8DC9PvC,AA8BA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,WAAA,GAAA,aACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,UACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,aAAA,GAAA,eACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,IAAA,UACA,EAAA,EAAA,SAAA,IAAA,WACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,YAAA,IAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,iBAAA,KAAA,mBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,gBAAA,KAAA,oBA7BU,IAAA,IAAQ,QAyRH,IAAjB,AAAA,UAAiB,GAEf,GAAY,GAAZ,AAAA,UAAY,GAAyB,EAAA,EAAA,OAAA,GAAA,SAAc,EAAA,EAAA,GAAA,GAAA,KAAU,EAAA,EAAA,GAAA,GAAA,OAAjD,EAAA,EAAA,yBAAA,GAAA,wBAAuB,OAFpB,IAAA,IAAQ,KCtTzB,AAmBA,GAAM,IAAwC,eA2BnB,EAAc,GACvC,GAAM,GAAqB,CACzB,SAAU,EACV,SAAU,SACV,OAAQ,GACR,MAAO,GACP,eAAgB,GAGlB,GAAW,GAAQ,cAUW,GAC9B,MAAO,IAAW,eAUS,GAC3B,MAAO,IAAW,GC9EpB,WAwBI,EAAmB,EAAY,EAC/B,GACF,GAAM,GAAa,EAAK,YAAY,GACpC,GAAI,GAAc,EAAW,kBAAoB,QAC/C,GAAM,GAAQ,EAAW,gBACnB,EAAM,EAAW,gBAAkB,EACrC,OACC,EAAW,gBAAkB,OAAY,EAAQ,EACR,EAAW,cACzD,GAAI,EAAW,OAAS,SACtB,MAAO,IACH,EAAK,WAAW,EAAW,iBAAkB,EAAW,GAE9D,GAAI,EAAW,OAAS,WACtB,GAAM,GAAS,EAAK,WAAW,MAAM,EAAO,GAE5C,MAAO,GAAO,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAEvD,GAAM,GACF,GAAU,EAAK,WAAW,MAAM,GAAO,GAAI,EAAW,GACpD,EAAO,EAAO,WACpB,MAAO,GAAW,OAAS,SACvB,EAAK,GACL,EAAA,KAAK,cAAc,EAAO,MAAO,GAEvC,GAAM,GAAY,EAAK,WAAW,GAClC,MAAO,IAAa,EAAU,kBAU5B,EAAc,EACd,GACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAY,EAAQ,kBAAkB,KAAK,SAAA,GAC/C,MAAO,CAAC,CAAC,EAAW,GAAyB,EAAU,MAGzD,MAAO,KAAc,OACjB,EAAW,GAAyB,EAAU,IAAY,GAC1D,mBASF,EAAc,EACd,GACF,MAAO,GAAW,GAAyB,EAAM,EAAQ,+BAUvD,EAAmB,GACf,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEjB,MAAO,CACL,GAAyB,EAAU,GAAW,EAAQ,kBACtD,GAIJ,YAAkC,EAAc,GAC9C,MAAO,AAAE,GAAe,EAAI,IAAI,EAAc,cAGlB,GAC5B,GAAM,GAAQ,EAAK,MAAM,KACzB,GAAI,EAAM,SAAW,EACnB,MAAO,CAAC,EAAM,GAGhB,GAAM,GAAW,EAAM,GACvB,MAAO,CAAC,EAAU,OAAO,EAAM,EAAM,OAAS,iBAW5C,EAAY,EACZ,GACF,GAAI,GAAM,EAAc,MAAO,EAAM,EAAW,GAChD,GAAI,IAAQ,YAEV,EAAM,EAAc,mBAAoB,EAAM,EAAW,GAIzD,OAHM,GAEF,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,IACxB,EAAI,EAAG,EAAI,EAAG,IACrB,EAAgB,GAAG,GAAM,EAAiB,EAAI,GAC9C,EAAgB,GAAG,GAAM,EAAiB,EAAI,EAAI,GAEpD,MAAO,GAET,MAAO,eAYmB,GAC1B,MAAO,GAAO,KAAO,EAAS,EAAA,MAAM,GCrJtC,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,mCCnLxE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,UAC7D,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,YAGjE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC7dxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAElD,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,QAC1D,CAAC,OAAU,mBAAoB,KAAQ,iBAAkB,KAAQ,QACjE,CACE,OAAU,2BACV,KAAQ,yBACR,KAAQ,QAEV,CAAC,OAAU,oBAAqB,KAAQ,OAAQ,KAAQ,YAG5D,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CACR,OAAU,QACV,KAAQ,QACR,KAAQ,QACR,aAAgB,MAGpB,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAAU,CACrD,OAAU,wBACV,KAAQ,sBACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,YAG3C,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,YAE3D,CACE,SAAY,cACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,KACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,iBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,WAG9D,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,wCCnWrE,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAAa,CAC7D,OAAU,yBACV,KAAQ,sBACR,KAAQ,QAEV,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAC/C,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,OAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,WACV,KAAQ,WACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,gBAAiB,KAAQ,gBAAiB,KAAQ,QAAS,CACpE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,eACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,mBACV,KAAQ,gBACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,QAItB,CACE,SAAY,sBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,aAE9C,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,kBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,wBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,6BACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,aACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,QAAS,KAAQ,YAAa,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,yCCpVnD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,WAEtC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,SAAU,aAAgB,GAClE,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,SAAU,aAAgB,IAErE,MAAS,CACP,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAEzD,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,IAEjE,MAAS,CAAC,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAExD,CACE,SAAY,kBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,OACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAAW,CACpD,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,WAE7C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAC3C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,UAC7C,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,SACzC,CAAC,OAAU,eAAgB,KAAQ,eAAgB,KAAQ,wCC3KjE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,YAGnD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,WAEjD,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,cACV,KAAQ,YACR,KAAQ,QACR,aAAgB,IAElB,CACE,OAAU,yBACV,KAAQ,qBACR,KAAQ,UAId,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,UAC/C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,WAE5C,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCChGtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAE3D,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,yCCvC3C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,yBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CACE,SAAY,cACZ,SAAY,QACZ,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CAAC,SAAY,QAAS,SAAY,SAAU,CAC1C,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,YACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,SACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAEvC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,UAAW,CAC1D,OAAU,UACV,KAAQ,SACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,SACR,aAAgB,KAItB,CAAC,SAAY,OAAQ,SAAY,QAAS,OAAU,IAAK,CACvD,SAAY,eACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,0BACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,UACzC,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,yCC3G/C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,wBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,aAE3C,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAAW,CACxD,OAAU,sBACV,KAAQ,qBACR,KAAQ,yCCvDhB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,eACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC9ItB,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,eACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAAW,CAC3D,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCCtItB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,eACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GACrE,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,aACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,gBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCCjJtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,QACrD,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,uCC1GvD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,GAAI,KAAQ,UAAW,KAAQ,WACnD,CAAC,MAAS,GAAI,KAAQ,OAAQ,KAAQ,WAExC,MACI,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAEtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,WAClD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAG3E,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,OAAQ,aAAgB,MAGjE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,eACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,YACpC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CACP,CACE,OAAU,aACV,KAAQ,YACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,WACV,KAAQ,UACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,cACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,mBACV,KAAQ,iBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,YAEpD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGzE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,MACV,KAAQ,MACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,MAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAC/D,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,YACV,KAAQ,kBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,kBAAmB,KAAQ,YAChD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAG5C,CACE,SAAY,gBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCC1NtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,mCChDxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,OACV,KAAQ,SACR,KAAQ,QACR,aAAgB,IAElB,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAGhD,CACE,SAAY,aACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,MACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CAAC,CACR,OAAU,iBACV,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,QACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAAa,CACnD,MAAS,EACT,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,OACV,iBAAoB,eACpB,KAAQ,OACR,KAAQ,cAGZ,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,cAG7C,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,eACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,YAG5D,CACE,SAAY,cACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,iCChIb,AAyCA,GAAA,IAAA,WAWE,aACE,GAAM,GAAM,CACV,GAAY,GAAW,GAAS,GAAa,GAAU,GACvD,GAAY,GAAS,GAAO,GAAO,GAAU,GAAe,GAC5D,GAAW,GAAU,IAEjB,EAA0B,GAAG,OAAM,MAAT,GAAa,EAAI,IAAI,SAAA,GAAM,MAAA,GAAG,QAE9D,KAAK,UAAY,EAAY,OACzB,SAAC,EAAK,GACJ,SAAI,EAAO,UAAY,EAChB,GAET,IAlBN,cAAA,eAAkB,EAAA,WAAQ,KAA1B,WACE,MAAO,MAAK,WAAc,MAAK,UAAY,GAAI,wCAsBjD,EAAA,UAAA,eAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAM,KAChB,EAAuB,GACvB,EAAkB,GAClB,EAAoB,GACpB,EAAQ,EAAQ,OAA8B,SAAC,EAAK,GACxD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,GAAG,WAAW,eACrB,EAAa,KAAK,EAAI,EAAK,OACtB,AAAI,EAAK,KAAO,QACrB,EAAQ,KAAK,EAAI,EAAK,OACjB,AAAI,GAAK,OAAS,MAAQ,EAAK,MAAM,SAAW,IACrD,EAAU,KAAK,EAAI,EAAK,OAEnB,GACN,IAEC,EAAiB,GACf,EAAkB,GACpB,EAA8C,GAC9C,EAA+C,GACnD,AAAI,GAAa,MACf,GAAqB,KAAK,oBAAoB,EAAU,QACxD,EAAsB,KAAK,oBAAoB,EAAU,UAE3D,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAMlC,AAAI,OAAO,KAAK,GAAqB,SAAW,EAC9C,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,AAAI,EAAK,SAAS,SAAW,GAC3B,EAAQ,KAAK,KAIjB,OAAO,KAAK,GAAqB,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,aAAe,EAAoB,GACxC,EAAQ,KAAK,MAKnB,AAAI,OAAO,KAAK,GAAoB,OAAS,EAC3C,OAAO,KAAK,GAAoB,QAAQ,SAAA,GAC/B,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GACF,GAAK,aAAe,EAAmB,GACvC,EAAO,KAAK,MAIhB,EAAS,EAGX,GAAI,GAAY,GAChB,AAAI,EAAM,SAAW,MAAQ,EAAM,QAAQ,UAAY,MACrD,GAAY,EAAM,QAAQ,SAAS,OAAO,SAAC,EAAW,GACpD,SAAU,EAAK,UAAU,MAAQ,EAAK,YAAY,GAC3C,GACN,KAGL,GAAM,GACF,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,EAAE,UAAS,GAExE,MAAI,GAAU,OAAS,GACrB,GAAO,UAAY,GAGd,GAGD,EAAA,UAAA,oBAAR,SAA4B,GAC1B,MAAO,QAAO,KAAK,GAAW,IACzB,OAAgC,SAAC,EAAM,GACtC,SAAK,EAAQ,GAAM,MAAQ,EACpB,GACN,KAGD,EAAA,UAAA,QAAR,SAAgB,GAGd,GAAM,GACF,GAAgB,EAAK,KAAO,KAAK,UAAU,EAAK,KAAO,GAC3D,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAGd,GAAM,GAAgB,CACpB,KAAM,EAAK,KACX,GAAI,EAAK,GACT,SAAU,EAAO,SACjB,WACK,GAAK,OACL,IAAI,IAAI,SAAA,GAAS,MAAA,GAAM,WAAW,KAAO,EAAM,OAAO,GAAK,IAChE,OAAQ,GACR,SAAU,GACV,YAAa,GACb,WAAY,GACZ,SAAU,EAAK,MAGjB,MAAI,GAAO,QAAU,MACnB,GAAQ,YACJ,EAAO,OAAO,OACV,SAAC,EAAK,GACJ,SAAI,EAAM,MAAQ,CAChB,KAAM,EAAM,KACZ,gBAAiB,EAAM,MACvB,cAAe,EAAM,KAEhB,GAET,KAEV,AAAI,EAAO,OAAS,MAClB,GAAQ,WACJ,EAAO,MAAM,OAAoC,SAAC,EAAK,GACrD,GAAM,GAAO,EAAM,KACf,EAAQ,OACZ,OAAQ,EAAM,UACP,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAChB,EAAM,cAAgB,GAC3B,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,aACA,UACH,cAEA,KAAM,IAAI,OACN,2BAA2B,EAAM,KAAI,YAAY,EAAK,IAE9D,SAAI,EAAM,MAAQ,CAAC,MAAK,EAAE,KAAI,GACvB,GACN,KAEF,GAID,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAU,EAAY,QACtB,EAAuB,GACvB,EAAkB,GACpB,EAA+B,GACnC,AAAI,GAAW,MACb,GAAQ,EAAQ,OAA8B,SAAC,EAAK,GAClD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,KAAO,SACd,EAAQ,KAAK,EAAI,EAAK,OAEjB,GACN,KAEL,GAAM,GAAiB,GACjB,EAAkB,GAExB,EAAY,UAAU,SAAS,QAAQ,SAAA,GAC9B,GAAA,GAAA,GAAA,EAAA,MAAA,GACD,EAAa,CACjB,KAAM,EACN,GAAI,cACJ,OAAQ,GACR,WAAY,GACZ,SAAU,QACV,YAAa,GACb,WAAY,CAAC,MAAO,CAAC,MAAO,GAAgB,EAAI,MAAO,KAAM,UAC7D,SAAU,IAEZ,EAAK,aAAe,EAAI,KACxB,EAAO,KAAK,GACZ,EAAM,GAAY,IAGpB,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAIlC,GAAM,GAAgB,EAAY,IAElC,EAAY,UAAU,UAAU,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,cAAgB,EACrB,EAAQ,KAAK,MAIjB,GAAM,GAAY,KAAK,mBAAmB,GAC1C,MAAO,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,IAG1D,EAAA,UAAA,mBAAR,SAA2B,GAA3B,GAAA,GAAA,KAEE,MAAO,CACL,WAAY,EAAY,UAAU,KAClC,OAAQ,EAAY,UAAU,SAAS,OACnC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,GACjC,GAET,IACJ,QAAS,EAAY,UAAU,UAAU,OACrC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,EAAK,EAAY,KAClD,GAET,MAIA,EAAA,UAAA,mBAAR,SACI,EACA,GACF,GAAI,GAAO,EAAI,KACf,MAAI,IAAW,MACb,GAAO,EAAQ,IAEV,CAAC,KAAI,EAAE,MAAO,EAAI,OAE7B,iBAE6B,GAC3B,GAAM,GAAS,EAAA,MAAM,OACrB,GAAI,MAAO,GAAO,MAAS,YACzB,MAAO,GAAO,KAAK,GACd,GAAI,MAAO,SAAW,YAC3B,MAAO,IAAI,QAAO,EAAM,UAAU,WAElC,KAAM,IAAI,OACN,gGAKyB,EAAc,GAC7C,GAAM,GACF,MAAM,QAAQ,GAAK,OAAO,aAAa,MAAM,KAAM,GAAK,GAAa,GACzE,MAAO,GAAW,EAAQ,EAAM,0BAI9B,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,KACJ,GAAiB,EAAM,EAAG,GAE5B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAO,GAAQ,EAAM,EAAI,cAIvB,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,IAAS,GACvB,EACF,EAAM,GAAQ,KAAO,EAAM,EAAQ,EAAM,GAAQ,KAAO,EAAM,EAAO,EACzE,MAAQ,OAAO,IAAU,SAAY,EAAQ,SAAS,EAAO,gBAG/B,GAC9B,AAAI,MAAQ,IAAW,UAErB,GAAQ,GAAoB,IAE9B,OAAQ,OACD,IAAoB,SACvB,MAAO,cACJ,IAAoB,aACpB,IAAoB,aACpB,IAAoB,YACpB,IAAoB,SACvB,MAAO,YACJ,IAAoB,QACvB,MAAO,WACJ,IAAoB,UACvB,MAAO,cACJ,IAAoB,UACvB,MAAO,iBAIP,MAAO,mBAKT,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,EAAM,KAAK,KAEb,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,GAAgB,EAAM,MAExB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,KAC7B,EAAM,KAAK,KAAK,IAAI,SAAA,GAAK,MAAA,IAAgB,KAE3C,cAG6B,GAEpC,MAAI,GAAM,YACR,OAEE,EAAM,KAAO,KACR,EAAM,IAAI,IACb,SAAA,GACI,MAAC,OAAO,GAAI,MAAS,SAAY,EAAI,KAAO,SAAS,EAAI,KAAM,MAElE,eAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MACV,GAAsB,EAAM,OAE9B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,GACO,IAAM,KAAK,GAAK,EAAM,KAAK,EAAE,OAAS,EAAM,KAAK,EACX,EAAM,KAAK,IAClD,IACH,IAAI,SAAA,GAAK,MAAC,OAAO,IAAM,SAAY,EAAI,SAAS,EAAG,MAEnD,cAIL,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAAE,IAAI,SAAC,GACvB,MAAO,IAAiB,EAAG,KAGxB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,MAC7B,EAAM,KAAK,MAAM,IAAI,SAAC,GAC3B,MAAO,IAAsB,KAG1B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAEb,ECtkBT,AA4BA,GAAA,IAAA,WAGE,WACY,EAAoB,EACpB,GAFZ,GAAA,GAAA,KACY,KAAA,KAAA,EAAoB,KAAA,UAAA,EACpB,KAAA,QAAA,EAJI,KAAA,OAAmB,GACnB,KAAA,MAAoC,GAIlD,KAAK,OAAS,EAAK,WAAW,IAAI,SAAA,GAAQ,MAAA,GAAK,SAAS,KACxD,AAAI,EAAK,UAAY,MACnB,MAAK,MAAQ,OAAO,KAAK,EAAK,UACZ,OAAO,SAAC,EAAmC,GAC1C,SAAM,GAAO,EAAK,QAAQ,GACnB,GACN,KAQhB,SAAA,UAAA,SAAR,SAAiB,GACf,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,UAOtC,EAAA,UAAA,QAAR,SAAgB,EAAc,GAC5B,GAAM,GAAQ,KAAK,KAAK,SAAS,GACjC,GAAI,EAAM,QAAU,KAClB,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,SAE9C,GAAI,EAAM,GAAK,MAAQ,EAAM,GAAK,KAChC,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAa,KAAK,KAAK,SAAU,EAAM,GAEhD,GAAI,EAAM,OAAS,KACjB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,MAAQ,KAChB,MAAO,IAAc,KAAK,KAAK,SAAU,EAAM,GAEjD,GAAI,EAAM,MAAQ,MAChB,GAAI,EAAM,KAAK,GAAK,MAAQ,EAAM,KAAK,GAAK,KAC1C,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,OAAS,KACtB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,MAAQ,KACrB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAIlC,MAAO,IAEX,KCtGA,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,cACA,YACA,MACH,MAAO,CAAC,EAAA,IACH,EAAc,IAAK,EAAM,EAAW,GACrC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,UAAW,EAAM,EAAW,SAE3C,eACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,cACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,oBACH,MAAO,CAAC,EAAA,kBACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC7F9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACA,aACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,OAAQ,EAAM,EAAW,SACxC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,YACA,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,eAAgB,EAAM,EAAW,GAC/C,EAAc,eAAgB,EAAM,EAAW,SAEhD,QACH,MAAO,CAAC,EAAA,MACJ,GAAU,EAAK,WAAW,GAAI,EAAW,SAC1C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SACxC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SACzC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAE5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC5K9C,YAyBI,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EAAA,KAAK,OACD,GAA8B,EAAQ,GACtC,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlB,EAAc,GAC1D,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,IAAM,EAAG,KAAO,IAAM,EAAG,KAAO,EAAG,GAC/C,MAAO,GAGX,MAAO,GCzCT,AA+BA,GAAA,IAAA,WAIE,WACa,EAAuB,EAAyB,EACjD,EAAiC,EAChC,EAA+B,GAF/B,KAAA,KAAA,EAAuB,KAAA,MAAA,EAAyB,KAAA,QAAA,EACjD,KAAA,aAAA,EAAiC,KAAA,uBAAA,EAChC,KAAA,YAAA,EAA+B,KAAA,eAAA,EANpC,KAAA,QAA6B,GAC7B,KAAA,QAAU,GAMhB,KAAK,SAAW,EAAA,OAAO,GACvB,EAAA,KAAK,KAAK,UAGZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAMd,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,OAAO,MAChD,EAAO,OAAO,YAGlB,KAAK,QAAU,GACf,KAAK,QAAU,GACf,KAAK,SAAS,WAGhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAOtB,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,GAAS,KAAK,OAC7B,KAAM,IAAI,OAAM,4BAA4B,EAAK,wBAC7C,KAAK,QAGX,GAAM,GAAkB,KAAK,QAAQ,GACrC,GAAI,EAAgB,QAClB,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0BACpB,EAAK,wGAIf,MAAI,MAAK,gBACP,GAAgB,QAAU,IAG5B,EAAgB,KAAO,GAChB,EAAgB,QAMzB,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GAAS,MAAA,GAAK,KAAK,MAQxC,EAAA,UAAA,MAAA,SAAM,EAAe,GACnB,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,CAAC,KAAK,aAAe,GAAS,KAAK,QAClD,KAAM,IAAI,OAAM,2BACZ,EAAK,8CAA8C,KAAK,SAG9D,GAAM,GAAI,KAAK,QAAQ,IAAU,GAEjC,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,eACZ,KAAK,KAAI,0CAA0C,EAAK;uCAExD,EAAO,MAAK,8BAA8B,KAAK,MAAK,KAc1D,GAVA,AAAI,KAAK,SAAW,GACf,MAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,IAC7D,MAAK,aAAe,EAAO,OAG7B,GACI,KAAK,aAAc,EAAO,MAC1B,eAAe,KAAK,KAAI,0CACpB,EAAK,KAET,EAAE,KACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,uCAGf,GAAI,EAAE,QACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,0CAGf,EAAE,OAAS,EACX,EAAA,KAAK,GACL,EAAE,QAAU,GAEZ,KAAK,QAAQ,GAAS,GAMxB,EAAA,UAAA,UAAA,SAAU,EAAmB,GAA7B,GAAA,GAAA,KACE,GAAI,EAAQ,SAAW,EAAQ,OAC7B,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,sCACxB,4BACI,EAAQ,OAAM,qCACd,EAAQ,OAAM,MAGxB,EAAQ,QAAQ,SAAC,EAAG,GAAU,MAAA,GAAK,MAAM,EAAG,EAAQ,OAWtD,EAAA,UAAA,OAAA,SAAO,EAAoB,GACzB,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAK,EAMH,EAAU,EAAQ,MAAM,EAAG,KAAK,aALhC,EAAU,GACV,OAAS,GAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAMjB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAKpC,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAAO,gCAElC,EAAA,MAAM,EAAS,IAMxB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAI,KAAK,SAAW,EAClB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAIpC,OADM,GAAU,GACP,EAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAGf,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAC9B,mDACI,KAAK,aAAY,4BAA4B,EAAQ,GAAG,MAAK,KAE9D,EAAA,OAAO,EAAS,IASzB,EAAA,UAAA,QAAA,SAAQ,EAAmB,GACzB,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAGhD,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,CAAC,KAAK,aAAe,GAAY,KAAK,QACxC,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,KAAK,QAAO,KAGtE,KAAK,UAAU,EAAS,EAAA,QAAQ,EAAQ,KAS1C,EAAA,UAAA,MAAA,SAAM,EAAkB,GAAxB,GAAA,GAAA,KACE,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAEhD,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEZ,EAAW,4BAA4B,EAAO,OAGpD,GAAI,CAAC,KAAK,aAAe,EAAO,SAAW,KAAK,QAC9C,KAAM,IAAI,OACN,2DACI,KAAK,QAAO,QAAQ,EAAO,OAAM,kEAI3C,GAAM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,GAC1B,EAAA,KAAK,WACH,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,EAAK,cAE3D,MAAO,KAGT,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAQ,GAAK,EAEf,KAAK,UAAU,EAAS,IAE5B,KC3TA,AAoCA,GAAA,IAAA,WAeE,WACa,EAA4B,EAC5B,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IADxB,KAAA,QAAA,EAA4B,KAAA,aAAA,EAC5B,KAAA,aAAA,EACX,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,IAAiB,EAAO,MAC1B,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,EAAO,OAEhD,GACI,EAAc,EAAO,MAAO,+BAEhC,EAAA,KAAK,KAGT,KAAK,SAAW,EAAA,OAAO,GACvB,KAAK,eAAiB,EACtB,EAAA,KAAK,KAAK,UA5BZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAiCvB,EAAA,UAAA,KAAA,WACE,MAAO,IAAI,GACH,KAAK,QAAO,QAAG,KAAK,aAAc,KAAK,eAMjD,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,MACzC,EAAO,YAGX,KAAK,QAAQ,OAAS,EACtB,KAAK,SAAS,WAKhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAUtB,EAAA,UAAA,MAAA,SAAM,EAAwB,EAAwB,GAAtD,GAAA,GAAA,KAEE,GAFoD,AAAA,IAAA,QAAA,GAAA,IAEhD,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,IAAgB,IAAM,KAAK,QAAQ,SAAW,EAChD,KAAM,IAAI,OAAM,kCACZ,EAAW,iCACX,KAAK,QAAQ,OAAM,cAEzB,UACI,EAAc,KAAK,aAAc,+BAC9B,EAAA,KAAK,WACV,GAAM,GACF,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAA,QAAQ,EAAQ,KAC/C,MAAO,GAAA,MAAM,EAAiB,MASlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAC9B,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAG9C,GAAI,KAAK,SAAW,EAClB,KAAM,IAAI,OAAM,qCAGlB,GAAM,GAAS,KAAK,QAAQ,MAC5B,UACI,EAAO,MAAO,EAAc,+BACzB,EAAA,QAAQ,EAAQ,IAOzB,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAM9C,GAHA,GACI,EAAO,MAAO,KAAK,aAAc,+BAEjC,KAAK,iBAAmB,KAAK,OAC/B,KAAM,IAAI,OAAM,4CAElB,EAAA,KAAK,GACL,KAAK,QAAQ,KAAK,IAOpB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAO,EACT,KAAM,IAAI,OACN,0DAA0D,GAGhE,GAAI,KAAK,iBAAmB,IAAM,EAAO,KAAK,eAC5C,KAAM,IAAI,OAAM,+BACZ,EAAI,6BAA6B,KAAK,eAAc,KAE1D,KAAK,QAAQ,OAAS,GASxB,EAAA,UAAA,QAAA,SAAQ,EAAsB,EAAwB,GAEpD,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,EAAe,GAAK,EAAe,KAAK,QAAQ,OAClD,KAAM,IAAI,OAAM,4BACZ,EAAY,mBAAmB,KAAK,QAAQ,OAAM,cAGxD,GAAI,KAAK,QAAQ,IAAiB,KAChC,KAAM,IAAI,OAAM,oBAAoB,EAAY,aAGlD,UACI,KAAK,QAAQ,GAAc,MAAO,EAClC,+BAEG,KAAK,QAAQ,IAQtB,EAAA,UAAA,QAAA,SAAQ,EAAsB,GAC5B,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAG9C,GAAI,EAAe,GACf,KAAK,iBAAmB,IAAM,GAAgB,KAAK,eACrD,KAAM,IAAI,OAAM,yBACZ,EAAY,uBAAuB,KAAK,eAAc,cAG5D,GACI,KAAK,aAAc,EAAO,MAAO,+BACrC,EAAA,KAAK,GACL,KAAK,QAAQ,GAAgB,GAU/B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAwB,GAAlD,GAAA,GAAA,KAEE,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAU9C,MAPA,IACI,KAAK,aAAc,EAAc,+BAIrC,EAAU,EAAQ,MAAM,EAAG,KAAK,QAE5B,EAAQ,SAAW,EACd,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAK,QAAQ,GAAI,KAC1D,MAAO,GAAA,MAAM,EAAS,MAS1B,EAAA,UAAA,OAAA,SAAO,EAAwB,GAA/B,GAAA,GAAA,KACE,GAAI,CAAC,CAAC,GAAgB,IAAiB,KAAK,aAC1C,KAAM,IAAI,OAAM,uBACZ,KAAK,aAAY,+BAA+B,GAMtD,MAHA,IACI,KAAK,aAAc,EAAc,+BAEjC,KAAK,SAAW,EACX,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAK,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAG,KACjD,MAAO,GAAA,OAAO,EAAS,MAG7B,iBAQI,EAAgB,EAAwB,GAC1C,GAAM,GAAQ,EAAO,MACrB,GAAI,EAAO,MAAM,OAAS,EACxB,KAAM,IAAI,OACN,oDAAoD,EAAO,OAEjE,GAAI,EAAO,QAAU,EACnB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,GAEzC,GAAM,GAAc,EAAO,MAAM,MAAM,GACvC,GACI,EAAa,EAAc,+BAE/B,GAAM,GAAuB,EAAA,QAAQ,GACrC,MAAO,IAAI,IAAW,EAAY,EAAc,eAU9C,EAAwB,EAAwB,GAClD,MAAO,IAAI,IAAW,GAAI,EAAc,EAAc,eAWpD,EAAgB,EAAmB,EACnC,GACF,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,GAAe,MAAQ,IAAgB,IAAM,GAAY,EAC3D,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,EAAW,KAGrE,GAAM,GAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,GACtD,EAAU,EAAA,QAAQ,EAAQ,GAChC,SAAQ,QAAQ,SAAC,EAAO,GACtB,EAAK,QAAQ,EAAO,EAAQ,MAEvB,cAWL,EAAgB,EAAkB,GACpC,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEV,EAAW,4BAA4B,EAAO,OAmBtD,OAhBM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,EAAA,KAAK,WAC7B,GAAM,GAAU,GAChB,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,GAEtD,SAAO,UACA,IAGH,EAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,EAAO,QAE1D,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAK,QAAQ,EAAG,EAAQ,IAE1B,MAAO,GC3YT,AAiBA,GAAA,IAAA,OAUa,GAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,kUACnB,EAAA,EAAK,iBACN,KAAA,MAAA,CAAA,EAAA,OACA,cAAA,MAAA,CAAA,EAAA,OAgBA,QAAA,MAAA,CAAA,EAAA,OACA,iBAAA,MAAA,CAAA,EAAA,OAsDA,WAAA,MAAA,CAAA,EAAA,OAIA,SAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QAOA,OAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAoBA,qBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAQA,sBAAA,MAAA,CAAA,EAAA,QAUA,uBAAA,MAAA,CAAA,EAAA,QAWA,sBAAA,MAAA,CAAA,EAAA,QAQA,qBAAA,MAAA,CAAA,EAAA,QAWA,oBAAA,MAAA,CAAA,EAAA,QAMA,qBAAA,MAAA,CAAA,EAAA,QAOA,oBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAaA,sBAAA,MAAA,CAAA,EAAA,QACA,oBAAA,MAAA,CAAA,EAAA,QAcA,oBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAYA,kBAAA,MAAA,CAAA,EAAA,QAYA,uBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAUA,qBAAA,MAAA,CAAA,EAAA,QASA,oBAAA,MAAA,CAAA,EAAA,QAUA,kBAAA,MAAA,CAAA,EAAA,wBAzTG,SACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClC,CAAA,EAAM,EAAK,eAC7B,MADM,GAAY,GAAA,OACd,EAAU,GACZ,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,gBAE1C,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAKtC,SACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAO,EAAc,OAAQ,EAAM,EAAW,GAI/C,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAFxC,SACD,GAAA,OAEC,EAAS,EAAK,IAAI,SAAA,IAAU,MAAA,IAAO,KACzB,CAAA,EAAM,EAAW,GAAG,eAAhC,EAAY,GAAA,OAEhB,EAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAChD,GAAO,YAIP,EAAmB,gFAIf,UAAa,EAEV,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACzC,EAAQ,EAAQ,eAAgB,EAAQ,uBAD5C,SAAS,GAAA,OAEH,GAAY,EAAO,IAAI,SAAA,IAAU,MAAA,IAAO,KAI9C,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,YAMN,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAQ,EAAQ,eAAgB,EAAQ,uBAF1C,UACD,GAAA,OAEO,CAAA,EAAM,GAAW,GAAG,eAAhC,SAAY,GAAA,OAEZ,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,4CA1BN,GAAU,WAAE,CAAA,EAAA,wCA8BnB,MAAA,CAAA,EAAO,UAGD,SAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAChD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClD,AAAK,EAAK,MACR,GAAO,GAAY,IAGb,CAAA,EAAM,EAAK,gBAAnB,MAAA,CAAA,EAAQ,GAAA,OAAmB,GAAK,CAAC,OAAW,GAAQ,CAAC,EAAM,iBAK3D,MAFM,GAAY,EAAK,WAAW,KAC9B,SAAA,IAAQ,MAAA,IAAU,GAAM,EAAW,KAAa,SAChD,EACI,GAAO,GAAU,EAAW,EAAW,GAC7C,CAAA,EAAO,CAAC,GAAY,MAEtB,CAAA,EAAO,gBAGD,SACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,WAAW,GACnB,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,YACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,gBACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,yBAA0B,EAAM,EAAW,GAEvD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAc,GAAI,IACpB,EAAM,EAAO,EAAM,EAAc,EAAwB,EACzD,GACJ,EAAQ,eAAe,GACvB,CAAA,EAAO,CAAC,EAAY,SAAU,EAAA,OAAO,aAG/B,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAmB,EAAQ,eAAe,EAAG,IACnD,EAAiB,MAAM,EAAO,GAC9B,CAAA,EAAO,CAAC,EAAiB,mBAGnB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAkB,EAAQ,eAAe,EAAO,IACtD,CAAA,EAAO,CAAC,EAAgB,KAAK,aAGvB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAoB,EAAQ,eAAe,EAAS,IAC1D,CAAA,EAAO,CAAC,EAAkB,OAAO,EAAe,aAG1C,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAqB,EAAQ,eAAe,GAAU,IAC5D,GAAmB,QAAQ,GAAgB,IAC3C,CAAA,EAAO,CAAC,GAAmB,mBAGrB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAoB,EAAQ,eAAe,GAAS,IACpD,GACF,EAAc,QAAS,EAAM,EAAW,GAC5C,CAAA,EAAO,CAAC,GAAkB,OAAO,cAG3B,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,MAAM,GAAS,IAChC,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAkB,EAAQ,eAAe,GAAO,IACtD,CAAA,EAAO,CAAC,EAAA,OAAO,GAAgB,OAAQ,mBAGjC,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,gBACjB,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,QAAQ,EAAO,GAC1B,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAE7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAW,EAAc,cAI9C,UACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GACF,GAAQ,GAAe,GAAgB,EAAc,IACzD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,GAAQ,EAAc,GAAc,IACvD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,EAAS,IAClD,CAAA,EAAO,CAAC,GAAW,OAAO,EAAe,GAAc,aAGjD,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,MAAM,EAAc,GAAc,cAG/C,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,GAAW,GAAQ,EAAc,IACpD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAC5C,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,CAAA,EAAO,CAAC,GAAW,OAAO,GAAa,aAGjC,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,SAAS,GACpB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAc,cAGnC,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,UAAW,EAAM,EAAW,GAExC,GAAa,GAAM,GAAa,GAAS,GAC/C,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGnB,KAAM,WAAU,aAAa,EAAK,GAAE,6BCvW1C,AA2BA,YACI,EAAY,EAA4B,GACpC,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAC7B,EAAc,IAAY,iBAE1B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,yGAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,oFAIR,GAAI,EACF,KAAM,IAAI,OACN,yEAEN,GAAM,GAAS,EAAc,UAAW,EAAM,EAAW,GACnD,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGhB,MAAO,CACL,OAAM,EACN,IAAG,EACH,WAAU,EACV,UAAS,EACT,QAAO,EACP,SAAQ,EACR,eAAc,GAIX,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAQ,EAAyB,EACjC,QAED,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,gBACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAIvB,8BACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,gBAAgB,CAClC,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAGvB,0BACA,mBACH,GAAM,GAAQ,EACI,cAAe,EAAM,EACrB,GAEZ,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GACxC,MAAO,CAAC,EAAA,gBACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAO,CAAC,EAAO,GAAI,EAAO,IAAK,QAEhC,4BACA,mBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cAET,MAAO,CAAC,EAAA,gBACJ,EAAc,QAAS,EAAM,EAAW,GAExC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GAEzC,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,EACnC,EACA,CAAC,EAAU,GAAI,EAAU,GAAI,EAAU,UAExC,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,qBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,sBAAuB,EAAM,EAAW,GAEpD,EAAA,EAAA,kBAAA,EAAA,IAAA,EAAA,EAAA,GAAA,CAAA,EAAA,GAAA,EAAA,IAAA,CAAA,EAAA,GAAA,EAAA,IAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,QAIf,MAAO,CAAC,EAAQ,OAEb,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,cACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,YAAa,EAAM,EAAW,GAG1C,EAAe,EAAQ,GACvB,EAAc,EAAQ,GAGtB,EAAiB,EAAU,GAC3B,EAAgB,EAAU,GAEhC,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAc,GAAc,EAC7B,CAAC,EAAgB,GAAgB,iBAIrC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC/S9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,KAAW,EAAO,EAAO,QAE9B,YACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAM,EAAc,MAAO,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,SAAe,EAAO,EAAM,QAEjC,eACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,YAAkB,EAAQ,EAAY,QAE3C,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OAAa,EAAS,EAAO,EAAS,QAE3C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cAEJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,SAEzC,SACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,MACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,SAGzC,mBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,gBACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,GAExC,QAED,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1H9C,AAiBA,GAAA,IAAA,OAUA,YACI,EAAY,EAA4B,GAC1C,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EAAS,EAAc,SAAU,EAAM,EAAW,GAClD,EACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CACL,MAAK,EACL,OAAM,EACN,cAAa,EACb,aAAY,EACZ,eAAc,EACd,aAAY,GAIT,GAAM,IAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,mHACnB,EAAA,EAAK,iBACN,sBAAA,MAAA,CAAA,EAAA,OAgBA,sBAAA,MAAA,CAAA,EAAA,OAcA,sBAAA,MAAA,CAAA,EAAA,OACA,sBAAA,MAAA,CAAA,EAAA,OAQA,QAAA,MAAA,CAAA,EAAA,OAQA,WAAA,MAAA,CAAA,EAAA,uBA9CG,SAOF,GAAU,EAAM,EAAW,GAN7B,EAAK,EAAA,MACL,EAAM,EAAA,OACN,EAAa,EAAA,cACb,EAAY,EAAA,aACZ,EAAc,EAAA,eACd,EAAY,EAAA,aAGC,CAAA,EAAM,EAAA,MAAY,gCAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,wBAGjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAG3D,EACF,EAAc,qBAAsB,EAAM,EAAW,GAG1C,CAAA,EAAM,EAAA,MAAY,6BAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,sBAIjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAGzD,CAAA,EAAM,EAAA,MAAY,uBACtB,EAAmB,EAAoB,EAAe,EACtD,WAFJ,MAAA,CAAA,EAAA,CAAQ,EAAA,gBAKF,SAAY,EAAA,KACb,EAAc,YAAa,EAAM,EAAW,GAC7C,QACY,CAAA,EAAM,EAAA,WAAiB,WAAjC,SAAM,CAAI,EAAA,QAChB,EAAU,UACV,CAAA,EAAO,UAGP,MAAA,CAAA,EAAO,EAAA,eACH,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,6BC3G1C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAS,EAAA,KAAW,EAAG,EAAG,GAChC,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAS,EAAA,OAAa,GAC5B,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAS,EAAA,OAAa,EAAG,GAC/B,MAAO,CAAC,EAAO,OAAQ,EAAO,iBAG9B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCpDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,GAAU,EAAK,UAEnB,yBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,IAAY,OACjD,cACH,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,QACrC,eACA,mBACA,2BACH,GAAM,GAAO,EAAc,IAAK,EAAM,EAAW,GACjD,MAAO,CAAC,GAAY,QAEjB,YACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,IAAY,SACjC,WACH,GAAM,GACD,EAAc,IAAK,EAAM,EAAW,GACzC,MAAO,CAAC,GAAY,QACjB,QACH,MAAO,CAAC,EAAA,SACH,EAAc,IAAK,EAAM,EAAW,GAAoB,MACzD,cACD,SACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,GAAA,SAAe,EAAE,aACtC,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OAAa,QAClB,QACH,GAAM,GAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,QAAQ,KACJ,kGAEJ,QAAQ,IAAI,GACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,QAAQ,IAAI,MAAM,UAAU,MAAM,KAAK,EAAK,GAAG,YAC9B,MAAM,EAAG,IAE5B,MAAO,CAAC,WAGR,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,eAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,yBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,sBAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,iBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,WAAY,EAAM,EAAW,GACzC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,qBAAsB,EAAM,EAAW,GAEzD,MAAO,CAAC,EAAA,MAAY,cAChB,EAAmB,EAAmB,EACtC,EAA8B,EAC9B,YAGJ,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzE9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,eACH,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACA,WACH,MAAO,CAAC,EAAA,MACJ,EAAc,YAAa,EAAM,EAAW,GAC5C,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACA,oBACA,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,aAAc,EAAM,EAAW,GAC7C,EAAc,aAAc,EAAM,EAAW,SAG9C,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SAExC,eACG,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAE7B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,sFAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,iEAGF,GAAA,GAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEhB,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,KAAM,EACN,WAAY,EACZ,uBAAwB,aAI1B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCjF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,qBACA,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,MACH,MAAO,CAAC,EAAA,2BACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,OAAQ,EAAM,EAAW,SAExC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cACJ,EAAc,gBAAiB,EAAM,EAAW,GAEhD,EAAc,cAAe,EAAM,EAAW,GAC9C,EAAc,eAAgB,EAAM,EAAW,GAE/C,EAAc,eAAgB,EAAM,EAAW,aAInD,KAAM,WAAU,aAAa,EAAK,GAAE,yBC9E9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,EAAW,YAGf,KAAM,WAAU,aAAa,EAAK,GAAE,yBCtH9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,eACA,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACvC,EACA,EAAc,UAAW,EAAM,EAAW,GAC9C,SAAS,EAAO,MAAM,EAAG,GAClB,CAAC,EAAA,OAAa,EAAQ,QAE1B,eACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OAAa,EAAO,EAAA,KAAW,EAAS,SAAU,QAEvD,gBACA,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,QAAc,EAAO,QAE1B,SAEH,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAEhD,EAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,gBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,MAAO,EAAM,EAAW,GACpC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAE/C,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,aACJ,EAAQ,EAAO,EAAK,EAAS,EAAW,EAAS,EACjD,EAAa,QAEd,OACH,MAAO,GAAA,KAAK,WACV,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GAGxC,EAAQ,EAAQ,GAAG,MACnB,EAAgB,EAAA,QAAc,EAAQ,IAAI,MAC1C,EAAS,EAAQ,IAAI,SAAA,GACzB,GAAM,GAAY,EAAA,KAAK,YAAY,EAAO,MAAO,GACjD,GAAI,CAAC,GACD,CAAC,EAAA,KAAK,YACF,EAAA,QAAc,GAAQ,MAAO,GACnC,KAAM,IAAI,OAAM,0CAElB,MAAO,GAAY,EAAS,EAAA,QAAc,EAAQ,KAEpD,MAAO,CAAC,EAAA,MAAY,EAAQ,UAG3B,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GAC7C,MAAO,GAAA,QAAc,EAAQ,OAE1B,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,YACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,kBAAmB,EAAM,EAAW,GAGhD,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,GAAA,MAAY,EAAQ,EAAiB,OAEzC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,UAAgB,EAAS,EAAQ,QAEtC,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,SAAe,EAAG,QAEvB,iBACH,GAAM,GACF,EAAc,gBAAiB,EAAM,EAAW,GAE9C,EACF,EAAc,cAAe,EAAM,EAAW,GAE5C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,MAAO,CAAC,EAAA,cACJ,EAAS,EAAc,EACvB,EAAa,QAAU,EAAa,MAChC,EACA,EAAA,KAAW,EAAc,EAAa,iBAG9C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1K9C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBChDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAGzC,cACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAGzD,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,UAAW,EAAM,EAAW,GAE1C,EAAc,gBAAiB,EAAM,EAAW,SAGjD,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,gBACH,GAAM,GACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACrC,cAEb,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAW,QAEZ,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAG5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnG9C,YAgDI,EAAY,EACZ,GACF,GAAM,GACD,SAAC,EAAY,EAA4B,GACxC,OAAQ,EAAK,cACN,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,cACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAsB,EAAM,EAAW,SAC9C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,UACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAkB,EAAM,EAAW,SACtD,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,gBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAwB,EAAM,EAAW,SAChD,YACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,iBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAyB,EAAM,EAAW,SACjD,SACH,GAAM,GAAW,GAAgB,EAAK,IACtC,GAAI,GAAY,EAAS,eACvB,MAAO,GAAS,eACZ,GAAI,IAAc,EAAM,EAAW,IAEvC,KAAM,WAAU,aAAa,EAAK,GAAE,+BAGtC,KAAM,WACF,eAAe,EAAK,GAAE,yIAI7B,EAAM,EAAW,GACxB,MAAI,aAAiB,SACZ,EAAM,KAAK,SAAC,GAAS,MAAA,GAAG,OAAO,KAEjC,GAAG,OAAO,GCvEnB,GAAA,IAAA,WAME,WACa,EACA,EACA,EACA,GAHA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IAHA,KAAA,UAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,YAAA,EATL,KAAA,YAAc,CAAC,GAAI,EAAG,UAAW,GAAI,YAAa,GAClD,KAAA,SAAmC,CAAC,KAAK,aACzC,KAAA,OAAS,EAQf,KAAK,4BAGC,SAAA,UAAA,SAAR,SAAiB,EAAY,GAC3B,MAAO,CAAC,GAAE,EAAE,UAAS,EAAE,YAAa,IAQtC,OAAA,eAAI,EAAA,UAAA,iBAAc,KAOlB,WACE,MAAO,MAAK,cARd,SAAmB,GACjB,AAAI,KAAK,WAAa,GACpB,MAAK,SAAW,EAChB,KAAK,8DAWT,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,mBAAmB,oCAOjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGN,EAAA,UAAA,0BAAR,WAEE,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,KAAK,SAAS,OAAS,EAAG,KAC5C,GAAM,GAAW,KAAK,SAAS,MAAM,EAAG,KAAK,SAAS,OAAS,GAC/D,EAAM,KAAK,KAAK,qBAAqB,IAEvC,EAAM,KAAK,IACX,KAAK,mBAAqB,GAGpB,EAAA,UAAA,qBAAR,SAA6B,GAC3B,MAAO,GACH,EACK,IACG,SAAA,GAAW,MAAC,GAAQ,KAAO,GAAK,EAAQ,cAAgB,EACpD,GACG,EAAQ,UAAS,IAAI,EAAQ,cACvC,KAAK,KACV,IAON,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,UACP,MAAK,SACL,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,KAAK,KAAK,SAAS,KAAK,OAAQ,IAC9C,KAAK,mBAAmB,QAAQ,KAAK,qBAAqB,KAAK,aAQnE,EAAA,UAAA,UAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,EAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,OAAO,IACrB,KAAK,kBAAkB,YAEvB,MAAM,IAAI,OAAM,4CAQpB,EAAA,UAAA,cAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,GAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SACL,GAAM,GACF,OAAO,OAAO,GAAI,KAAK,SAAS,KAAK,SAAS,OAAS,IAC3D,EAAQ,aAAe,EACvB,EAAQ,GAAK,KAAK,OAClB,KAAK,SAAS,OAAO,GAAI,EAAG,GAC5B,KAAK,mBAAmB,OACpB,EAAG,EAAG,KAAK,qBAAqB,KAAK,eAEzC,MAAM,IAAI,OAAM,0DAIpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,UAAU,IAGxB,EAAA,UAAA,eAAA,SAAe,GACb,KAAK,eAAe,EAAY,IAAM,GAGxC,EAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,eAAe,IAG7B,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,cAAc,EAAW,IAAM,GAGtC,EAAA,UAAA,cAAA,SAAc,GACZ,MAAO,MAAK,cAAc,IAG5B,EAAA,UAAA,QAAA,SAAQ,GACN,OAAW,KAAO,MAAK,eACrB,KAAK,eAAe,GAAK,cAAc,GAGzC,OAAW,KAAO,MAAK,cACrB,KAAK,cAAc,GAAK,cAAc,IAG5C,KC1LA,YAyCI,EAAwB,EAAiB,EACzC,GACF,GAAM,GAAY,GAAI,KAChB,EAA0B,GAC5B,EAAoB,KACpB,EAAuB,KAIrB,EAAO,GAAI,KACX,EACF,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAEpD,EAA0B,GAC9B,AAAI,GAAa,MACf,GAAgB,EAAU,IAAI,SAAA,GAAQ,MAAA,IAAc,EAAK,MAAM,MAIjE,OADM,GAAe,EAAO,QACrB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MAWtB,GAVA,AAAI,IAAc,IAAS,GAAe,KACxC,CAAI,GAAe,MACjB,GAAc,EACd,EAAa,EAAY,SAAS,IAAI,SAAA,GAAS,MAAA,GAAM,OACnC,OAAO,SAAA,GAAQ,MAAA,GAAU,IAAI,OAGnD,EAAU,IAAI,EAAK,MAGf,EAAU,EAAK,OAAS,KAC1B,SAGF,GAAI,EAAe,QAAQ,EAAK,QAAU,GACxC,SAGF,GAAI,EAAc,QAAQ,EAAK,QAAU,GACvC,SAEF,GAAI,EAAK,OAAO,SAAW,GACzB,EAAc,KAAK,EAAK,MACxB,SAEF,EAAK,OAAO,QAAQ,SAAA,GAElB,GAAI,EAAK,IAAI,EAAM,MACjB,OAEF,EAAK,IAAI,EAAM,MACf,EAAS,KAAK,KAGlB,MAAO,CAAC,OAAM,EAAE,QAAO,EAAE,UAAS,EAAE,cAAa,EAAE,YAAW,EAAE,WAAU,eAQxE,EAAc,EACd,GACK,GAAA,GAAA,EAAA,UAAW,EAAA,EAAA,OACZ,EAAmB,GACnB,EAAa,OAAO,KAAK,GACP,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAChC,IAAI,SAAA,GAAQ,MAAA,GAAM,MAAM,KAC1C,EAAY,EAAM,UAExB,EAAW,QAAQ,SAAA,GACjB,AAAI,EAAU,IAAI,EAAM,OACtB,EAAS,KAAK,KAGlB,EAAM,QAAQ,QAAQ,SAAA,GACpB,AAAI,EAAU,IAAI,EAAO,OACvB,EAAS,KAAK,KAGlB,AAAI,GAAa,MACf,EAAU,QAAQ,SAAA,GAChB,AAAI,EAAU,IAAI,EAAK,OACrB,EAAS,KAAK,KAMpB,OAFM,GAAO,GAAI,KACX,EAAuB,GACtB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MACtB,EAAK,IAAI,EAAK,MACd,AAAK,EAAU,EAAK,OAClB,EAAa,KAAK,GAEpB,EAAK,SAAS,QAAQ,SAAA,GACpB,AAAI,CAAC,EAAK,IAAI,EAAM,OAAS,EAAU,IAAI,EAAM,OAC7C,EAAM,OAAO,MAAM,SAAA,GAAS,MAAA,GAAK,IAAI,EAAM,SAC7C,EAAS,KAAK,KAIpB,MAAO,GAGT,GAAM,IAAmB,CACvB,SAAU,QAAS,QAAS,OAAQ,gBAAiB,cACrD,iBAAkB,KAAM,SAEpB,GAAoB,CACxB,sBAAuB,sBAAuB,sBAAuB,qBAGzC,GAC5B,MAAO,IAAiB,QAAQ,EAAK,KAAO,cAGf,GAC7B,MAAO,IAAkB,QAAQ,EAAK,KAAO,ECjK/C,AAkCA,GAAA,IAAA,WAsFE,WAAoB,EAAsB,GAA1C,GAAA,GAAA,KAAoB,KAAA,MAAA,EAAsB,KAAA,OAAA,EArFlC,KAAA,YAAmC,GAAI,KACvC,KAAA,WAA8B,GAM9B,KAAA,UAAY,IACZ,KAAA,WAAqC,GACrC,KAAA,qBAA0D,GA6EhE,KAAK,SAAW,EAAM,QACtB,KAAK,QAAU,EAAM,OACrB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UAExB,AAAI,EAAM,WAAa,MACrB,OAAO,KAAK,EAAM,WAAW,QAAQ,SAAA,GACnC,EAAK,qBAAqB,GACtB,GAAI,GAAc,EAAM,UAAU,GAAO,KApFnD,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,4CAGpD,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,oBACZ,KAAK,sDAG5B,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,gBAGpD,SAAc,GACZ,GAAM,GAAY,OAAO,KAAK,GAAW,IACrC,SAAA,GAAO,MAAA,GAAU,GAAK,IAAI,SAAA,GAAU,MAAA,GAAO,OAC/C,KAAK,WAAa,GAAG,OAAM,MAAT,GAAa,GAC/B,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GACtB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,IAAI,SAAA,GACvB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,cAAgB,EAAK,wCAG5D,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,IAAI,SAAC,GACxB,GAAM,GAAO,EAAK,cAAgB,EAAK,KACvC,MAAO,GAAK,cAAoB,EAAI,IAAI,EAAK,cAAmB,qCAIpE,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAAA,GAAA,GAAA,KACE,MAAO,QAAO,KAAK,KAAK,YAAY,OAAO,SAAC,EAAK,GAC/C,SAAI,GAAO,EAAK,WAAW,GAAK,UACzB,GACN,qCA0BG,EAAA,UAAA,kBAAR,SAA0B,EAAgB,GACxC,GAAM,GAAe,EAAO,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OAC7C,EAAgB,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OACrD,MAAO,GAAa,KAAK,KAAK,WAAa,KACvC,EAAc,KAAK,KAAK,YAOtB,EAAA,UAAA,QAAR,SAAgB,EAAwB,GACtC,GAAM,GACF,GAAqB,EAAQ,EAAS,KAAK,UAAW,KAAK,YACxD,EAAA,EAAA,cAAe,EAAA,EAAA,YAAa,EAAA,EAAA,WACnC,GAAI,GAAe,KACjB,KAAM,IAAI,OACN,qCAAqC,EAAY,KAAI,gBACrD,oBAAmB,EAAY,GAAE,kBACjC,6DACA,qCAAoC,EAAU,MAGpD,GAAI,EAAc,OAAS,GACzB,GAAM,GAAW,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,OAC9B,EAAU,OAAO,KAAK,GAC5B,KAAM,IAAI,OACN,+BAA+B,EAAQ,8BACvC,KAAI,EAAO,qCAAqC,EAAa,MAGnE,MAAO,IACH,KAAK,MAAO,KAAK,UAAW,IAYlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAAhC,GAAA,GAAA,KACE,EAAS,KAAK,UAAU,GACxB,GAAM,GAAQ,OAAO,KAAK,GAAQ,OAClC,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,GAClB,GAAM,GACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC5D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAG/D,AAAI,EAAY,SAAW,GACzB,GAAc,KAAK,UAGrB,GAAM,GAAiB,KAAK,kBAAkB,EAAY,GAGtD,EAAe,KAAK,YAAY,IAAI,GACxC,AAAI,GAAgB,MAClB,GAAe,KAAK,QAAQ,EAAQ,GACpC,KAAK,YAAY,IAAI,EAAgB,IAGvC,GAAM,GAAiC,GACjC,EAA+B,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,GAAI,IAChB,EAAK,UAAW,EAAgB,EAChC,EAAK,qBACH,EAAU,GAAA,GAAwB,EAAK,WAE7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAKzB,OAFM,GAAgB,EAAK,mBAAmB,GACxC,EAA2D,GACxD,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAO,EAAa,GAC1B,GAAI,CAAC,EAAW,EAAK,OACnB,GAAM,GAAU,GAAU,EAAM,EAAY,GAC5C,GAAI,YAAmB,SACrB,KAAM,IAAI,OACN,4BAA4B,EAAK,GAAE,kEAGzC,EAAW,EAAK,MAAQ,EACxB,EAAK,uBACD,EAAK,KAAM,EAAM,EAAY,EAAS,EACtC,EAAiB,IAIzB,MAAI,GAAK,QAAU,MACjB,EAAQ,QAAQ,GAEX,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAY,QAInD,EAAA,UAAA,mBAAR,SAA2B,GACzB,GAAM,GAAM,GAAG,OAAO,MAClB,GACA,OAAO,KAAK,GACP,IAAI,SAAA,GAAO,MAAA,GAAU,KACrB,IAAI,SAAA,GAAW,MAAA,GAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,QACrD,MAAO,IAAI,KAAI,IAET,EAAA,UAAA,uBAAR,SACI,EAAkB,EAAY,EAC9B,EAA2B,EAC3B,EACA,GAGF,GAAI,EAAK,WAAa,WAAa,EAAY,QAAQ,KAAc,GACnE,OAGF,EAAU,GAAU,QAAQ,SAAA,GAC1B,AAAI,GAAU,MACZ,GAAgC,EAAO,IAClC,GAAgC,EAAO,KAAO,GAC/C,EAAK,SAAS,UAGtB,EAAK,OAAO,QAAQ,SAAA,GAGlB,GAAI,EAAM,WAAa,WACrB,GAAM,GACF,GAA6B,EAAM,KAAM,EAAW,GACxD,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,GAAU,CAAC,EAAc,IAAI,EAAO,KACtC,GAAM,GAAQ,EAAgC,EAAO,IACrD,AAAI,IAAU,EACZ,GAAO,UACP,MAAO,GAAgC,EAAO,KACzC,AAAI,GAAS,MAGlB,EAAgC,EAAO,aAkB/C,EAAA,UAAA,aAAN,SAAmB,EAAwB,sEAEzC,MAAA,CAAA,EAAO,KAAK,cAAc,EAAQ,SAiBtB,EAAA,UAAA,cAAd,SACI,EAAwB,EAAmB,EAC3C,EACA,GAF2C,MAAA,KAAA,QAAA,GAAA,IAC3C,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,uGACF,MAAK,IACH,GAAS,KAAK,UAAU,GACxB,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,IAGd,EAAU,GAAI,IAChB,KAAK,UAAW,EAAgB,EAChC,KAAK,qBAKS,CAAA,EAAM,KAAK,uBACzB,EAAQ,EAAS,EAAS,WADxB,SAAY,EAAA,OAEZ,EAAU,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAGzD,EAAY,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAW,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,GAAO,GAAM,KACxD,EACF,GAAI,KAAgB,EAAS,OAAK,EAAa,KAAK,YACxD,OAAO,KAAK,GAAW,QAAQ,SAAA,GAC7B,GAAM,GAAc,EAAU,GAC9B,EAAY,QAAQ,SAAA,GAClB,AAAI,GAAU,CAAC,EAAO,YAAc,CAAC,EAAQ,IAAI,EAAO,KACtD,EAAO,cAKb,AAAI,KAAK,QAAU,MACjB,EAAQ,QAAQ,GAGlB,CAAA,EAAO,SAGH,EAAA,UAAA,qBAAN,SACI,EAAkB,EAClB,mFACI,SAAe,EAAO,OAAO,SAAC,EAAK,EAAQ,GAC/C,SAAI,EAAK,OAAO,GAAO,MAAQ,EACxB,GACN,IAEH,CAAA,EAAO,KAAK,cACR,EAAc,KAAK,YAAa,GAAM,EAAgB,SAS9C,EAAA,UAAA,uBAAd,SACI,EAAwB,EAA2B,EACnD,0IACI,EAAQ,OAAO,KAAK,GACpB,EACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAY,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC9D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAC3D,EACF,GAAqB,EAAQ,EAAa,KAAK,WAD5C,EAAS,EAAA,UAAE,EAAa,EAAA,cAAE,EAAW,EAAA,YAAE,EAAU,EAAA,WAGlD,EACE,EAAU,OAAK,KAAK,MAAM,SAAS,IAAI,SAAA,GACzC,MAAO,CAAC,KAAI,EAAE,SAAU,EAAQ,kBAEhC,EAAU,GAAA,GAAwB,KAAK,WAC7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAEnB,EAA2D,GAC3D,EAAgB,KAAK,mBAAmB,GACxC,EAAkC,0BACjC,GAAM,OAAS,EACd,GAAW,KAAK,aAClB,EAAY,EAAO,EAAS,EAAY,EAAO,EAC/C,EAAiB,EAAiC,GACtD,CAAA,EAAM,QAAQ,IAAI,KAJG,CAAA,EAAA,UAIrB,SAAA,oBAaF,GAXA,AAAI,GAAe,MAAQ,CAAC,GAC1B,QAAQ,KACJ,mIAGA,EACF,EACK,OACG,SAAA,GAAQ,MAAA,CAAC,GAAc,IACnB,CAAC,GAAU,EAAK,KAAM,EAAY,KACzC,IAAI,SAAA,GAAQ,MAAA,GAAK,OACtB,EAAe,OAAS,EACtB,QAAiB,GACrB,AAAI,GAAe,MACjB,GACI,gEACA,4BAA2B,EAAU,MAErC,GAAI,OACN,+BAA+B,EAAc,uBAC7C,YAAW,EAAK,gDAChB,KAAI,EAAa,MAAM,IAE7B,MAAA,CAAA,EAAO,SAGD,EAAA,UAAA,aAAR,SACI,EAAoB,EAA2B,EAC/C,EAA4B,EAC5B,EAA4B,EAC5B,EACA,GAEF,OAPF,GAAA,KAMQ,EAAqC,gBAEzC,GAAM,GAAO,EAAM,MACnB,EAAQ,eAAiB,EAAK,SAC9B,GAAI,GAAW,GAUf,GANA,AAAI,EAAK,KAAK,KAAO,SACjB,EAAc,aAAc,EAAK,KAAM,EAAW,IACnD,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAIC,EAAW,QAAQ,EAAK,QAAU,IACpC,GAAM,GAAU,GAAU,EAAK,KAAM,EAAW,GAChD,AAAK,GACF,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAEH,GAAM,GAAiB,EAAQ,eAC/B,AAAI,YAAmB,SACrB,EAAS,KAAK,EAAQ,KAAK,SAAA,GACzB,SAAU,GAAY,EACtB,EAAQ,eAAiB,EACzB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,GAC1C,KAGT,GAAU,GAAY,EACtB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,QAGnD,GAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,WAxC9C,EAAM,OAAS,OA2CtB,MAAO,IAGD,EAAA,UAAA,kBAAR,SACI,EAAY,EAA2B,EACvC,EAA4B,EAC5B,GACF,EAAK,SAAS,QAAQ,SAAC,GACd,GAAA,GAAA,GAAA,EAAA,KAAA,GAAA,GACP,GAAI,EAAM,IAAa,CAAC,EAAU,IAAI,EAAU,MAC9C,OAGF,AAAI,EAAU,KAAO,QACnB,AAAI,EAAU,WAAW,KAAK,SAAA,GACxB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAExC,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,KAGpD,AAAI,EAAU,WAAW,MAAM,SAAA,GACzB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAE5C,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,QAQ1D,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,OAAO,KAAK,KAAK,WACZ,QACG,SAAA,GAAO,MAAA,GAAK,UAAU,GAAK,QAAQ,SAAA,GAAU,MAAA,GAAO,eAGtD,EAAA,UAAA,uBAAR,SAA+B,GAA/B,GAAA,GAAA,KACE,OAAO,KAAK,GAAQ,QAAQ,SAAA,GAC1B,GAAM,GAAQ,EAAO,GACd,EAAA,GAAA,GAAA,GACD,EAAO,EAAK,MAAM,MAAM,GAC9B,GAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,GAAM,GAAQ,EAAK,WAAW,MAAS,MACjC,EAAQ,EAAM,SAAW,EAAM,MAAM,QACvC,EAAM,MAAM,MACR,SAAC,EAAK,GAAU,MAAA,GAAM,KAAW,IAAM,EAAM,KAAW,IAChE,EAAA,KAAK,OACD,EACA,WAAM,MAAA,sBAAsB,EAAK,KAAI,kBACjC,iCAAgC,EAAK,eACrC,KAAI,EAAM,MAAK,OAEzB,AAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,EAAA,KAAK,OACD,EAAM,QAAU,EAAK,WAAW,MAAS,MACzC,WAAM,MAAA,sBAAsB,EAAK,KAAI,8CAE9B,GAAK,WAAW,MAAS,MAAK,aAAa,EAAM,YAK1D,EAAA,UAAA,UAAR,SAAkB,GAChB,GAAM,GAAyB,GAC/B,OAAW,KAAa,GACtB,GAAI,KAAK,YAAc,MAAQ,KAAK,WAAW,QAAU,MACrD,KAAK,WAAW,OAAO,IAAc,MACvC,GAAM,GAAS,KAAK,WAAW,OAAO,GACtC,EAAO,EAAO,MAAQ,EAAO,OAE7B,GAAO,GAAa,EAAO,GAG/B,MAAO,IAGD,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAa,OAAO,KAAK,GAAQ,OAAO,SAAA,GACrC,GAAA,GAAA,GAAA,GAAA,GACP,MAAO,GAAK,MAAM,MAAM,IAAa,OAEvC,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,OACN,gDACA,WAAU,EAAU,kCAIpB,EAAA,UAAA,WAAR,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GACjB,GAAI,EAAK,YAAc,MAAQ,EAAK,WAAW,SAAW,MACtD,EAAK,WAAW,QAAQ,IAAS,MACnC,GAAM,GAAS,EAAK,WAAW,QAAQ,GACvC,MAAO,GAAO,KAEhB,MAAO,IACN,KAGG,EAAA,UAAA,aAAR,SAAqB,GAArB,GAAA,GAAA,KACE,EAAQ,QAAQ,SAAA,GACP,GAAA,GAAA,GAAA,GAAA,GACP,GAAI,CAAC,EAAK,MAAM,MAAM,GACpB,KAAM,IAAI,OAAM,eAAe,EAAI,kCAI3C,KC3lBA,AAyBO,GAAM,IAAqB,oBACrB,GAAqB,2BAoDhC,WACY,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IADA,KAAA,SAAA,EACA,KAAA,YAAA,EAzCJ,KAAA,QAAU,MA0ChB,AAAI,GAAe,MACjB,MAAK,YAAc,IArCvB,cAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,yCAGd,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,SAAS,4CAGvB,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,6CAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,SAAS,wCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,yCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,2CAoBf,EAAA,UAAA,cAAR,WACE,GAAM,GAAO,KAAK,SAClB,GAAK,EAAsB,MAAQ,KAEjC,KAAK,QAAU,UACN,KAAK,YAAY,aAAe,KACzC,KAAK,QAAU,EAAA,GAAG,mBAAmB,EAAgB,KAAK,kBAE1D,GAAM,GAAW,EAAA,GAAG,gBAAgB,EAAgB,KAAK,aACzD,GAAI,EAAS,SAAW,EAGtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAgB,KAAK,sBAChD,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,CAAC,GAAK,MAEpB,KAAK,QAAU,EAAS,KAQtB,EAAA,UAAA,KAAN,2GAEE,GADA,KAAK,gBACD,KAAK,QAAQ,MAAQ,KACvB,KAAM,IAAI,OACN,iHAGY,MAAA,CAAA,EAAM,KAAK,QAAQ,eAA/B,SAAY,EAAA,OAElB,CAAA,EAAO,KAAK,SAAS,UASvB,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,UAAY,EACjB,GAAM,GAAQ,KAAK,UAAU,cACzB,EAAY,GAChB,AAAI,KAAK,UAAU,qBAAuB,MACxC,GACK,KAAK,UAAU,oBAA4B,WAIlD,KAAK,QAAa,EAAM,SAAS,SAAQ,IAAI,EAAM,SAAS,YAC5D,GAAM,GACF,EAAA,GAAG,cAAc,KAAK,UAAU,WAAY,KAAK,UAAU,aAK/D,GAJA,KAAK,SAAW,GAAI,IAChB,GAAgB,SAAS,eAAe,EAAO,IACnD,KAAK,SAAS,UAAY,KAAK,6BAA6B,GAExD,EAAU,kBAAoB,MAChC,GAAM,GACF,GAAgB,SAAS,eAAe,EAAU,kBACtD,KAAK,YAAc,GAAI,IAAc,GACrC,KAAK,YAAY,UAAY,KAAK,SAAS,UAC3C,KAAK,YAAY,QAAQ,GAAI,IAG/B,MAAO,IA+CH,EAAA,UAAA,KAAN,SAAW,EAAmC,4EAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,OACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,OACN,+GAIN,MAAA,CAAA,EAAO,EAAa,KAAK,KAAK,iBAyChC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,MAAO,MAAK,QAAQ,EAAQ,KAAK,cAG3B,EAAA,UAAA,gBAAR,SAAwB,GAEtB,GAAI,CAAE,aAAkB,GAAA,SAAW,CAAC,MAAM,QAAQ,GAEhD,MAAO,GAGT,GADA,EAAS,MAAM,QAAQ,GAAU,EAAS,CAAC,GACvC,EAAO,SAAW,KAAK,WAAW,OACpC,KAAM,IAAI,OACN,+BACA,wBAAuB,KAAK,WAAW,OAAM,mBAC7C,oBAAmB,EAAO,OAAM,oBAEtC,MAAO,MAAK,WAAW,OAAO,SAAC,EAAK,EAAW,GAC7C,SAAI,GAAc,EAAoB,GAC/B,GACN,KAGG,EAAA,UAAA,iBAAR,SAAyB,GACvB,SAAU,GAAW,KAAK,YACnB,AAAC,MAAM,QAAQ,GAAuB,EAAZ,CAAC,IAmBpC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,EAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GAChC,GAAM,GAAS,KAAK,SAAS,QAAQ,EAAQ,GAC7C,MAAO,GAAO,OAAS,EAAI,EAAS,EAAO,IAkBvC,EAAA,UAAA,aAAN,SACI,EACA,mGACF,SAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GACjB,CAAA,EAAM,KAAK,SAAS,aAAa,EAAQ,WAAlD,SAAS,EAAA,OACf,CAAA,EAAO,EAAO,OAAS,EAAI,EAAS,EAAO,UAGrC,EAAA,UAAA,6BAAR,SAAqC,GACnC,MAAO,QAAO,KAAK,GAAK,OAAO,SAAC,EAAyB,GACvD,SAAO,GAAO,CAAC,EAAI,IACZ,GACN,KAQL,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UAEd,AAAI,KAAK,aACP,KAAK,YAAY,WAGvB,iBAiCI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,6FACF,GAAI,GAAY,KACd,KAAM,IAAI,OACN,0GAGN,MAAI,IAAW,MACb,GAAU,IAGZ,AAAI,EAAQ,WACV,CAAK,EAA0B,MAAQ,MACrC,CAAM,EAAoB,SAAS,MACjC,GAAY,EAAsB,KAEpC,EAAW,GAAG,EAAW,GAAqB,KAG5C,EAAQ,GAAI,IAAW,EAAU,GACvC,CAAA,EAAM,EAAM,eAAZ,SAAA,OACA,CAAA,EAAO,QC7ZT,GAGM,IAAU,4GCHhB,o8DC2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAEZ,GAAQ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,YAkDwB,EAAY,GAElC,MAAO,IAAgB,EAAO,GAShC,YACI,EAAY,EACZ,EAAiC,GAEnC,GAFE,AAAA,IAAA,QAAA,GAAA,GAA0B,MAAO,AAAA,IAAA,QAAA,GAAA,GAA2B,MAE1D,GAAS,KACX,MAAO,MAET,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAI,EAAK,IAAI,GACX,MAAO,GAAK,IAAI,GAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAGL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,GAChB,OAAW,KAAK,IACd,GAAM,GAAQ,EAAM,GACd,EAAc,GAAgB,EAAO,EAAO,EAAM,GACxD,EAAe,GAAK,EAEtB,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAdzD,UAAK,IAAI,EAAO,EAAO,OAChB,EAAO,kBA2Cd,EAAe,GAAA,MAAA,KAAA,QAAA,GAAA,IACV,GAAgB,EAAQ,GAOjC,YACI,EAAe,EACf,GAAA,AAAA,IAAA,QAAA,GAAA,GAA2B,MAG7B,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAEL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,kBACL,GACT,GAAM,GAAW,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC7B,EAAc,GAAgB,EAAU,EAAO,GACrD,EAAe,GAAK,GAHtB,OAAW,KAAK,KAAL,GAKX,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAbzD,OAAO,GAAO,kBAkBQ,GACxB,MAAI,KAAM,KACD,KAIL,GAAW,EAAE,IACR,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAG,QAAS,gBAsC3B,EAAY,+GACR,EAA+B,GAAI,KAGzC,GAAgB,EAAO,EAAO,OAMZ,EAAA,MAAM,KAAK,EAAK,+BAAhB,GAAA,EAAA,OAAP,GAAG,EAAA,GACN,EAAQ,EAAK,IAAI,GACnB,YAAiB,SACC,CAAA,EAAM,GADxB,CAAA,EAAA,IAFmC,CAAA,EAAA,UAG/B,EAAc,EAAA,OACpB,EAAK,IAAI,EAAK,oBAJA,wBAWZ,SAAS,GAAgB,EAAO,EAAO,GAC7C,CAAA,EAAO,oBASkB,GACzB,MAAO,IAAO,MAAS,CAAC,YAAY,OAAO,IACtC,OAAM,QAAQ,IACb,MAAO,IAAQ,UAAY,CAAE,aAAe,IAAA,qBAYvB,GAC3B,MAAO,IAAO,MAAQ,GAAY,IAAQ,MAAM,QAAQ,IACnD,MAAO,IAAQ,UAAa,YAAe,IAAA,QAC5C,GAAA,KAAQ,aAAa,GAO3B,YAAqB,GACnB,MACI,KAAU,MACT,MAAO,IAAU,UAAY,MAAO,IAAU,WCnRrD,YAqB6B,GAC3B,MAAO,IAAQ,EAAW,IAI5B,YAAuB,GACrB,MAAI,aAAgB,IAAA,OACV,CAAC,MAAO,EAAK,QAAS,QAAS,IAC9B,GAAW,GACb,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAM,QAAS,IChClC,AAqBA,GAAA,IAAA,WAcE,WAAmB,GACjB,GADiB,KAAA,SAAA,EAVT,KAAA,MAAQ,EACR,KAAA,IAAM,EAUV,GAAY,KACd,KAAM,IAAI,YAAW,mDAEvB,GAAI,EAAW,EACb,KAAM,IAAI,YAAW,6CAEvB,KAAK,KAAO,GAAI,OAAS,GACzB,KAAK,gBAAkB,EAAI,EAMnB,SAAA,UAAA,KAAV,SAAe,GAEb,KAAO,EAAQ,GACb,GAAS,KAAK,gBAEhB,MAAO,GAAQ,KAAK,iBAGZ,EAAA,UAAA,IAAV,SAAc,GACZ,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,MAAO,MAAK,KAAK,EAAQ,KAAK,WAGtB,EAAA,UAAA,IAAV,SAAc,EAAe,GAC3B,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,KAAK,KAAK,EAAQ,KAAK,UAAY,GAMrC,EAAA,UAAA,OAAA,WACE,GAAI,GAAS,KAAK,IAAM,KAAK,MAC7B,MAAI,GAAS,GACX,GAAS,KAAK,gBAAkB,GAE3B,GAQT,EAAA,UAAA,OAAA,WACE,MAAO,MAAK,WAAa,KAAK,UAQhC,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,WAAa,GAM3B,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,IAAI,KAAK,IAAK,GACnB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,IAMlC,EAAA,UAAA,QAAA,SAAQ,GACN,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,KAAK,KAOd,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,GAChC,GAAM,GAAS,KAAK,IAAI,KAAK,KAC7B,YAAK,IAAI,KAAK,IAAK,QACZ,GAMT,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GACpC,KAAK,IAAI,KAAK,MAAO,IAMvB,EAAA,UAAA,MAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAS,KAAK,IAAI,KAAK,OAC7B,YAAK,IAAI,KAAK,MAAO,QACrB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GAC7B,GAYT,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAQ,KAAK,KAAK,KAAK,MAAQ,GAC/B,EAAS,KAAK,IAAI,GACxB,YAAK,IAAI,EAAO,KAAK,OACd,GAEX,KC/KA,AAoBA,GAAA,IAAA,SAAA,GAA0C,GAAA,EAAA,GAMxC,mBACE,GAAA,KAAA,KAAM,EAAkB,mBAAiB,KAG3C,SAAA,UAAA,OAAA,WACE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,GACH,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,KAAI,KAAA,KAAC,IAGb,EAAA,UAAA,QAAA,SAAQ,GACN,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,QAAO,KAAA,KAAC,IAMR,EAAA,UAAA,OAAR,WAOE,OANM,GAAc,KAAK,SAAW,EAC9B,EAAU,GAAI,OAAS,GACvB,EAAM,KAAK,SAIR,EAAI,EAAG,EAAI,EAAK,IACvB,EAAQ,GAAK,KAAK,IAAI,KAAK,KAAK,KAAK,MAAQ,IAG/C,KAAK,KAAO,EACZ,KAAK,SAAW,EAChB,KAAK,gBAAkB,EAAI,KAAK,SAChC,KAAK,MAAQ,EACb,KAAK,IAAM,GA7CE,EAAA,iBAAmB,GA+CpC,GAhD0C,ICpB1C,YAuCqC,GACnC,MAAO,IAAI,IAAc,eAyBvB,GAEF,MAAO,IAAI,IAAqB,eAgB9B,EACA,GACF,MAAO,IAAI,IAAgB,EAAe,eAmDxC,EACA,GAAA,MAAA,KAAA,QAAA,GAAgC,GAAgB,MAC3C,GAAI,IAAe,EAAW,GAUvC,GAAA,IAAA,WAAA,cAwBQ,SAAA,UAAA,QAAN,6GACQ,SAAc,GACZ,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,KAAK,gBAAf,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAcH,EAAA,UAAA,eAAN,+GACQ,SAAS,KAAK,SAAS,KACvB,EAAc,GACZ,CAAA,EAAM,EAAO,eAAjB,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,EAAO,gBAAjB,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAUH,EAAA,UAAA,aAAN,2GACU,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACR,CAAA,EAAM,KAAK,eAAf,SAAI,EAAA,oCAWF,EAAA,UAAA,aAAN,SAAmB,qGACT,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,8BACzB,CAAC,EAAE,MAAS,EACd,CAAA,EAAM,KAAK,QADiB,CAAA,EAAA,UAChC,SAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,oCAgBjC,EAAA,UAAA,aAAA,SAAa,GACX,MAAO,IAAI,IAA0B,KAAM,IAa7C,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,IAAI,IAAe,KAAM,IAWlC,EAAA,UAAA,IAAA,SAAO,GACL,MAAO,IAAI,IAAY,KAAM,IAW/B,EAAA,UAAA,SAAA,SAAY,GACV,MAAO,IAAI,IAAiB,KAAM,IAWpC,EAAA,UAAA,eAAA,SAAkB,GAChB,MAAO,IAAI,IAAiB,KAAM,GAAW,UAW/C,EAAA,UAAA,QAAA,SAAW,GACT,MAAO,IAAI,IAAgB,KAAM,IAQ7B,EAAA,UAAA,aAAN,SAAmB,sEACjB,MAAA,CAAA,EAAO,KAAK,IAAI,GAAG,qBAUf,EAAA,UAAA,cAAN,SAAoB,sEAClB,MAAA,CAAA,EAAO,KAAK,eAAe,GAAG,aAAa,SAAA,GAAK,MAAC,KAAM,WAqBzD,EAAA,UAAA,cAAA,SAAc,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,IACxB,GAAI,IAAsB,KAAM,EAAW,IAmCpD,EAAA,UAAA,iBAAA,SACI,EAAmB,EAEnB,GAFmB,AAAA,IAAA,QAAA,GAAA,IAEnB,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAAa,KAAK,cAAc,EAAW,GAGjD,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,MAaxC,EAAA,UAAA,YAAA,SACI,EACA,GACF,MAAO,IAAI,IACP,GAAkB,CAAC,KAAM,IAAY,IAU3C,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAShC,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAYhC,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,IAAI,IAAiB,KAAM,IAapC,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAC1B,MAAO,IAAI,IAAgB,KAAM,EAAY,IAO/C,EAAA,UAAA,OAAA,WACE,MAAO,IAAI,IAAe,OAE9B,KAWA,GAAA,SAAA,GAA+B,GAAA,EAAA,GAE7B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,MAAA,EADd,EAAA,KAAO,IAKf,SAAA,UAAA,QAAA,WACE,MAAO,YAAY,KAAK,MAAM,OAAM,UAGhC,EAAA,UAAA,KAAN,oFACE,MAAI,MAAK,MAAQ,KAAK,MAAM,OAC1B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAO,KAAK,MAAM,KAAK,MAC7B,KAAK,OACL,CAAA,EAAO,CAAC,MAAO,GAAU,GAAO,KAAM,WAE1C,GAlB+B,IAoB/B,GAAA,SAAA,GAAsC,GAAA,EAAA,GACpC,WACc,GADd,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,OAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAO,iBAGH,EAAA,UAAA,KAAN,8EACE,IACE,MAAA,CAAA,EAAO,KAAK,gBACL,GAEP,QAAE,QACE,mDAAmD,EAAE,QACnD,iBAGZ,GApBsC,IAsBtC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,8EACE,MAAA,CAAA,EAAO,KAAK,SAAS,aAEzB,GA1BgC,IA4BhC,GAAA,SAAA,GAA8B,GAAA,EAAA,GAQ5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAF3D,EAAA,MAAQ,EAIN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,iHAKS,MAAK,QAAU,KAAK,SACT,CAAA,EAAM,KAAK,SAAS,QADH,CAAA,EAAA,UAGjC,MAFM,GAAU,EAAA,OAEZ,EAAQ,KACV,CAAA,EAAO,GAET,IAAA,QAAW,EAAQ,qBAErB,MAAA,CAAA,EAAO,KAAK,SAAS,cAEzB,GAzC8B,IA2C9B,GAAA,SAAA,GAA8B,GAAA,EAAA,GAE5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAD3D,EAAA,MAAQ,IAKR,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,8EACE,MAAI,MAAK,SAAW,KAAK,SACvB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,CAAA,EAAO,KAAK,SAAS,aAEzB,GAhB8B,IAqB9B,GAAA,SAAA,GAAuC,GAAA,EAAA,GAKrC,WACc,EAAqC,EACrC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EAAqC,EAAA,UAAA,EACrC,EAAA,qBAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,qBAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,6GACQ,EAAa,0BACZ,GAAM,OAAS,KAAK,UACZ,CAAA,EAAM,KAAK,SAAS,QADC,CAAA,EAAA,UAElC,MADM,GAAO,EAAA,OACT,EAAK,KACH,KAAK,sBAAwB,EAAM,OAAS,EAC9C,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,KAE9B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAM,KAAK,EAAK,qBAElB,MAAA,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,WAEhC,GAvCuC,IAyCvC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,2GAEiB,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,MAAQ,KAAK,UAAU,EAAK,OACnC,CAAA,EAAO,GAET,IAAA,QAAW,EAAK,qCAGtB,GAlCgC,IAoChC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAC9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,WAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAc7B,IAZM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAOzD,EAAS,KAAK,UAAU,EAAK,OAC7B,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCgC,IAqChC,GAAA,SAAA,GAA2C,GAAA,EAAA,GAEzC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,QAAA,EAHd,EAAA,MAAQ,EAKN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,oBAO7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGR,EAAA,UAAA,WAAN,yJAGa,CAAA,EAAM,KAAK,SAAS,eAA3B,MAAA,CAAA,EAAO,EAAA,eAEP,kBAAI,AAAC,KAAK,QAAQ,SAChB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,+CAUrC,GA1C2C,IA4C3C,GAAA,SAAA,GAAqC,GAAA,EAAA,GACnC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,KACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAe,GAAA,YAAe,sBAAsB,EAAK,OAOhD,CAAA,EAAM,KAAK,UAAU,EAAK,gBAKzC,IALM,EAAS,EAAA,OACT,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCqC,IAgDrC,GAAA,SAAA,GAAmD,GAAA,EAAA,GAQjD,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,YAAc,GAAI,IACvB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAiBR,EAAA,UAAA,WAAN,2GAIS,MAAK,YAAY,WAAa,EAE9B,CAAA,EAAM,KAAK,QAFoB,CAAA,EAAA,UAEpC,MAAK,GAAA,aACH,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,YAG/B,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,YAAY,QAAS,KAAM,WAEnD,GAjDmD,IAkDnD,GAAA,SAAA,GAAoC,GAAA,EAAA,GAClC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,eAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,IAeT,IAbM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAMzD,EAAc,KAAK,UAAU,EAAK,OAClC,EACF,GAAA,YAAe,sBAAsB,GACzC,KAAK,YAAY,QAAQ,GAIzB,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAIN,MAAA,CAAA,EAAO,UAEX,GArCoC,IAgDpC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAStC,WACI,EACiB,GAFrB,GAAA,GAGE,EAAA,KAAA,OAAO,KADY,SAAA,iBAAA,EARb,EAAA,SAAuC,KAGvC,EAAA,SAA4B,KAOlC,EAAK,cAAgB,IAGvB,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,8CAC1B,MAAU,GAAiB,eAGvB,EAAA,UAAA,KAAN,8EACE,YAAK,SAAW,KAAK,cAAc,KAAK,UACxC,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,cAAd,SAA4B,qGAO1B,MAAA,CAAA,EAAM,gBAAN,GAAA,OACI,KAAK,UAAY,KACI,CAAA,EAAM,KAAK,cAAc,QAD9C,CAAA,EAAA,UAEF,GADM,EAAiB,EAAA,OACnB,EAAe,KAEjB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,KAAK,SAAW,EAAe,MAC/B,AAAI,KAAK,kBAAoB,MAC3B,MAAK,SAAW,KAAK,SAAS,aAAa,KAAK,oCAGjC,MAAA,CAAA,EAAM,KAAK,SAAS,eACvC,MADM,GAAa,EAAA,OACf,EAAW,KACb,MAAK,SAAW,KAChB,CAAA,EAAO,KAAK,cAAc,KAE5B,CAAA,EAAO,SAEX,GApDwC,IAsD5B,GAAZ,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,YAHU,IAAA,IAAe,KAmC3B,GAAA,IAAA,SAAA,GAAwD,GAAA,EAAA,GAItD,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAgC,GAAgB,MAFvE,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,UAAA,EACA,EAAA,aAAA,EALf,EAAA,MAAQ,EACR,EAAA,eAA6C,OAQrD,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,0CAC1B,MAAO,IAAI,EAAiB,YAGhB,EAAA,UAAA,UAAd,SAAwB,2CAWtB,WAAiB,GACf,GAAI,YAAqB,KACvB,GAAM,GAAS,EAAU,OACzB,MAAO,CACL,MAAO,EAAO,KAAK,SAAA,GACjB,WACA,AAAI,EAAE,MACJ,IAEK,EAAE,QAEX,QAAS,QAGX,OAAO,CAAC,MAAO,KAAM,QAAS,gEArBlC,MAAA,CAAA,EAAM,UAAN,SAAA,OAII,EAAe,EACf,EAAgB,EAoBF,CAAA,EAAM,GAAmB,KAAK,UAAW,WAE3D,GAFM,EAAY,EAAA,OAEd,IAAiB,EAEnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAI,EAAgB,EAClB,OAAQ,KAAK,kBACN,IAAgB,KACnB,KAAM,IAAI,OACN,+CACA,0BAAyB,KAAK,MAAK,UACpC,IAAgB,SACnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,SACxB,IAAgB,SAMzB,YAAK,QACL,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAGzB,EAAA,UAAA,KAAN,8EACE,YAAK,eAAiB,KAAK,UAAU,KAAK,gBAC1C,CAAA,EAAO,KAAK,qBAEhB,GAxEwD,IAoFxD,GAAA,SAAA,GAAyC,GAAA,EAAA,GAGvC,WACc,EAAqC,GADnD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAAqC,EAAA,WAAA,EAEjD,EAAK,OAAS,GAAI,IAAuC,KAG3D,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAOzB,EAAA,UAAA,OAAV,WACE,KAAO,CAAC,KAAK,OAAO,WAClB,GAAM,GAAI,KAAK,SAAS,OACxB,KAAK,OAAO,KAAK,KAIrB,EAAA,UAAA,KAAA,WACE,YAAK,SAIE,KAAK,OAAO,SAEvB,GA/ByC,IAuCzC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAUtC,WACc,EAAqC,EAC/C,GAFJ,GAAA,GAGE,EAAA,KAAA,KAAM,EAAU,IAAW,KAFf,SAAA,SAAA,EAAqC,EAAA,WAAA,EAH3C,EAAA,kBAAoB,GAM1B,EAAK,OAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACpD,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGN,EAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,MAAK,MAAM,KAAK,SAAW,IAG1B,EAAA,UAAA,YAAV,WACE,MAAO,MAAK,UAAU,KAAK,OAAO,WAG9B,EAAA,UAAA,WAAN,6GAEE,AAAK,KAAK,mBACR,KAAK,gCAEC,MAAK,OAAO,UAAS,CAAA,EAAA,GACrB,GAAc,KAAK,cACV,CAAA,EAAM,KAAK,OAAO,cAAc,YAC/C,GADM,EAAS,EAAA,OACX,EAAO,KACT,KAAK,kBAAoB,OAEzB,aAAK,SACL,CAAA,EAAO,sBAGX,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,WAE/B,GApDwC,ICjnCxC,kBA6DA,aAWW,KAAA,KAAe,KA+DxB,SAAA,UAAA,MAAA,SAAM,EAAmB,GAAzB,GAAA,GAAA,KAAyB,AAAA,IAAA,QAAA,GAAA,IACvB,GAAM,GAAO,KACb,GAAA,KAAQ,OACJ,EAAY,EAAG,WAAM,MAAA;QACrB,IACJ,GAAI,GACJ,MAAI,MAAK,OAAS,UAAY,KAAK,MAAQ,KAGzC,EAAO,KAAK,KACP,AAAI,EAGT,EAAO,KAAK,KAAK,KAAK,KAAO,GAI7B,EAAO,KAAK,MAAM,KAAK,KAAO,GAEzB,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OACH,iBAAiB,EAAW,EAAgB,WAChD,IAkBL,EAAA,UAAA,YAAA,SAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,UAAY,EAAQ,OAAS,SAG7C,EAAO,SACF,AAAI,KAAK,MAAQ,MAAQ,EAAQ,MAAQ,KAG9C,EAAO,KAAK,KAAO,EAAQ,KAI3B,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,EAAA,MAAA,IAAA,KAAA,SAAA,0BACK,MAAA,CAAA,EAAM,EAAK,mBAAZ,SAAA,GAAC,EAAA,QAAuB,YAAY,CAAA,EAAM,EAAQ,mBAAlD,MAAA,CAAA,EAAA,EAAA,MAAA,EAAA,CAAoC,EAAA,gBACxC,IAmBN,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,SAEhB,EAAO,SAIP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,OAAO,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAClE,IAmBC,EAAA,UAAA,aAAN,SAAmB,6FACT,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,aAAa,UAkB9C,EAAA,UAAA,IAAA,SAAkC,GAAlC,GAAA,GAAA,KACQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,IAAI,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAC/D,KAAK,OA0BV,EAAA,UAAA,SAAA,SAAuC,GAAvC,GAAA,GAAA,KAEQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,SAAS,UACvC,KAAK,OAYV,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,GAAI,GAAc,KAChB,KAAM,IAAI,YACN,6DAGN,GAAM,GAAO,KACb,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,SAAS,UAAa,KAAK,OAqBrE,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,EAAQ,EAI/B,EAAO,KAAK,KAAO,EACd,AAAI,IAAU,EAEnB,EAAO,EACF,AAAI,KAAK,MAAQ,MAAS,KAAU,QAAa,EAAQ,GAG9D,EAAO,SAGP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mDACrB,SAAmB,GACrB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,MAAA,IAAA,KAAA,SAAA,sCAAqB,CAAA,EAAM,EAAK,mBAApB,MAAA,CAAA,EAAE,GAAA,MAAO,EAAA,OAAuB,EAAA,KAAM,GAAK,WAC3D,CAAA,EAAO,GAAyB,EAAiB,KAAK,UACrD,IAoBL,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,GAAS,GAAK,KAAK,MAAQ,EAIlD,EAAO,KAAK,KAAO,EACd,AACH,KAAK,MAAQ,MACZ,MAAK,KAAO,GAAS,IAAU,QAAa,EAAQ,GAGvD,EAAO,EAGP,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IA4BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,EAAe,GAA3C,GAAA,GAAA,KAEE,GAFyC,AAAA,IAAA,QAAA,GAAA,IAErC,GAAc,MAAQ,EAAa,EACrC,KAAI,MAAK,MAAQ,KACT,GAAI,YACN,4DAEE,GAAI,YACN,+KAGA,oCAAmC,KAAK,KAAI,eAGpD,GAAM,GAAO,KACP,EAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACrD,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACvB,SAAQ,EAAO,QACnB,AAAI,GACF,IAAS,EAAO,SAEV,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,QAAQ,EAAY,EAAM,mBACxD,KAAK,OAoBV,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,KAAK,KAAO,EAGnC,EAAO,EACF,AAAI,KAAK,MAAQ,MAAQ,KAAK,MAAQ,EAG3C,EAAO,KAAK,KAGZ,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IAmBjD,EAAA,UAAA,QAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,iBAc3B,EAAA,UAAA,eAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,wBA1HjB,EAAA,gBAAkB,IA4HpC,iBAeI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,GAAA,UAAA,GAAkB,GAAA,EAAA,GAAd,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KACT,SAAA,KAAO,IAMD,SAAA,UAAA,SAAN,8EACE,MAAA,CAAA,EAAO,UAEX,GAVyB,iBAgCyB,GAApD,GAAA,GAAA,KACE,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,GAAkB,SAAQ,EAAM,oBA4CA,GAAlD,GAAA,GAAA,KAGE,GAAI,CAAC,GAAW,GACd,KAAM,IAAI,OAAM,qDAElB,GAAI,GACJ,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAO,GAAQ,KAAQ,EAAS,GAAkB,KAC5B,KAAK,IAAI,EAAO,EAAS,GAAkB,cAE1D,YAAoB,QAC7B,OAAW,KAAM,GACf,EAAO,GAAQ,KAAQ,EAAS,GAAmB,KAC7B,KAAK,IAAI,EAAO,EAAS,GAAmB,MAGtE,MAAO,IAAyB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACd,MAAA,CAAA,EAAM,GAAmB,EAAU,SAAA,GACjD,GAAI,YAAa,IACf,MAAO,CAAC,MAAO,EAAE,WAAY,QAAS,IACjC,GAAI,GAAW,GACpB,MAAO,CAAC,MAAO,KAAM,QAAS,IAE9B,KAAM,IAAI,OACN,uFAPF,SAAU,EAAA,OAWhB,CAAA,EAAO,GAAsB,EAAS,GAAgB,iBACrD,GAWL,YAAyB,GACvB,GAAI,IAAS,KACX,MAAO,MAIT,GAAM,GAAa,EAAK,GAExB,GAAI,GAAa,IAEf,GAAM,GAAQ,GAAY,GAC1B,MAAO,CAAC,MAAK,EAAE,QAAS,IAI1B,MAAO,CAAC,MAAO,KAAM,QAAS,IAOhC,YAAwD,GAEtD,GAAI,EAAO,SAAW,EAEpB,KAAM,IAAI,OAAM,wCAGlB,MAAI,GAAO,YAAc,IAAA,OAEhB,GAAA,MAAS,GAGT,GAAA,OAAU,GC9rBrB,mBA2BqC,GAAA,EAAA,GAMnC,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,IAIzB,SAAA,UAAA,SAAN,+GACwB,MAAA,CAAA,EAAM,KAAK,MAAM,mBAAjC,SAAgB,EAAA,OAChB,EAAe,EAAc,aAC7B,EAAe,EAAa,MAAM;GAAM,IAAI,SAAA,GAEhD,MAAI,GAAK,SAAS,OAChB,GAAO,EAAK,MAAM,EAAG,KAEhB,IAET,CAAA,EAAO,SAEX,GAtBqC,IC3BrC,AAyBA,GAAM,IAAa,IACb,GAAY,OAAO,OACnB,GAAc,OAAO,SACrB,GAAc,OAAO,SACrB,GAA0B,OAAO,mBACjC,GAA8B,OAAO,+BAeX,GAAA,EAAA,GAgI9B,WAA+B,EAAmB,GAAlD,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,EA9HvB,EAAA,UAAY,GACZ,EAAA,gBAA4B,KAC5B,EAAA,qBAAuB,GACvB,EAAA,cAA+C,KAC/C,EAAA,sBAAwB,GACxB,EAAA,UAAY,IACZ,EAAA,gBAAkB,GA0HxB,EAAK,KAAO,GAAI,IAAgB,GAChC,AAAK,GACH,GAAY,IAEd,EAAK,UAAY,IAAU,YAAc,IACzC,EAAK,gBAAkB,EAAU,YACjC,EAAK,cAAgB,EAAU,cAC/B,EAAK,sBAAwB,EAAU,sBACvC,AAAI,EAAU,gBACZ,IAAA,KAAK,OACD,EAAU,WAAa,KACvB,WACI,MAAA,mEACR,EAAK,gBAAkB,GACvB,EAAK,UAAY,KAEjB,EAAK,UAAY,EAAU,UAAY,EAAU,UAAY,MA9H3D,SAAA,UAAA,YAAN,2GACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEF,MAAA,CAAA,EAAO,KAAK,sBAAwB,OAAO,KAAK,KAAK,eACjB,KAAK,uBAW7B,EAAA,UAAA,eAAd,8HAC8B,MAAA,CAAA,EAAM,KAAK,8BACvC,GADM,EAAsB,EAAA,OACxB,CAAC,KAAK,iBAAmB,CAAC,EAE5B,KAAM,IAAI,OACN,6DA0BN,GAzBO,AAAI,KAAK,iBAAmB,GAEjC,GAAA,KAAK,OACD,EAAoB,SAAW,KAAK,gBAAgB,OACpD,WAAM,MAAA,uCACF,EAAK,gBAAgB,OAAO,WAC5B,kEACW,EAAoB,OAAO,WAAa,OAE7D,AAAK,KAAK,iBACR,MAAK,gBAAkB,GAGnB,EAAkC,KAAK,gBAAgB,OACzD,SAAC,EAAmC,GAClC,SAAS,GAAS,EAAS,GAAQ,GAAM,EAClC,GAET,IACE,EACF,OAAO,KAAK,GAAQ,OAAO,SAAC,GAAS,MAAC,GAAO,GAAQ,IACzD,GAAA,KAAK,OACD,EAAe,SAAW,EAC1B,WAAM,MAAA,iCAAmC,EAAe,aAExD,KAAK,eACP,IAAA,EAAA,EAAkB,EAAA,OAAO,KAAK,KAAK,eAAjB,EAAA,EAAA,OAAA,IAEhB,GAFS,EAAG,EAAA,GACN,EAAQ,KAAK,gBAAgB,QAAQ,GACvC,IAAU,GACZ,KAAM,IAAI,OACN,YAAc,EACd,uEACY,KAAK,gBAAgB,WAAa,MAIxD,YAAK,qBAAuB,aAGhB,EAAA,UAAA,oBAAd,uHACM,MAAK,UACM,CAAA,EAAM,KAAK,KAAK,YAD3B,CAAA,EAAA,UACI,SAAO,EAAA,OACQ,CAAA,EAAM,EAAK,eAChC,GADM,EAAe,EAAA,OACjB,EAAa,KACf,KAAM,IAAI,OAAM,sCAEZ,SAAoB,EAAa,MACjC,EAAU,KAAK,SAAS,EAAW,IACzC,CAAA,EAAO,UAEP,MAAA,CAAA,EAAO,YA0DL,EAAA,UAAA,SAAN,wHACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEU,MAAA,CAAA,EAAM,KAAK,KAAK,mBAAxB,SAAQ,EAAA,OACZ,AAAI,KAAK,WAGP,GAAQ,EAAM,KAAK,IAErB,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,GAAK,gBAAgB,YAG7C,EAAA,UAAA,gBAAA,SAAgB,GAKd,OAJM,GAAS,KAAK,SAAS,GACvB,EAA6C,GAC7C,EAA2C,GAExC,EAAI,EAAG,EAAI,KAAK,gBAAgB,OAAQ,KAC/C,GAAM,GAAM,KAAK,gBAAgB,GAC3B,EAAS,KAAK,cAAgB,KAAK,cAAc,GAAO,KAC9D,GAAI,KAAK,uBAAyB,CAAC,EAEjC,SAEA,GAAM,GAAQ,EAAO,GACjB,EAAc,KAClB,GAAI,IAAU,GAGZ,GAAI,GAAU,EAAO,UAAY,OAC/B,EAAc,EAAO,aAChB,GAAI,GAAW,GAAO,UAAY,EAAO,SAC9C,KAAM,IAAI,OACN,mBAAmB,EAAG,2BAA2B,GAErD,EAAc,YAIhB,GAAM,GAAa,OAAO,GAC1B,GAAI,MAAM,GAGR,AAAI,GAAU,EAAO,QAAU,OAC7B,EAAc,KAAK,WAAW,GAG9B,EAAc,UAEP,CAAC,GAAU,CAAC,EAAO,MAG5B,EAAc,MAId,QAAQ,EAAO,WACR,UACH,EAAc,EACd,UACG,QACH,EAAc,KAAK,MAAM,GACzB,UACG,OACH,EAAc,KAAK,WAAW,GAC9B,cAEA,EAAc,GAKtB,AAAC,GAAU,EAAO,QAAW,EAAO,GAAO,EACd,EAAS,GAAO,EAKjD,MAAI,QAAO,KAAK,GAAQ,SAAW,EAC1B,EAGA,CAAC,GAAI,EAAU,GAAI,IAItB,EAAA,UAAA,WAAR,SAAmB,GACjB,MAAI,KAAU,KAAO,EAAM,gBAAkB,OACpC,EAEA,GAKH,EAAA,UAAA,SAAR,SAAiB,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IAM7B,OALM,GAAmB,GACrB,EAAa,EACX,EAAa,EAAK,OACpB,EAAe,GAEV,EAAI,EAAG,EAAI,EAAY,IAC9B,OAAQ,OAED,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAa,EAAI,EACjB,EAAe,GACf,UAEG,MAAK,UAIR,GAHA,EAAa,EAAI,EAGb,KAAK,YAAc,KAAO,KAAK,gBACjC,MAEF,EAAO,KAAK,IACZ,EAAe,GACf,cAGA,EAAe,GACf,EAAa,EACb,MAEJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,IACvC,EAAe,GACf,EAAa,EAAI,EACjB,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,EAAI,IAC3C,EAAe,GACf,EAAa,EAAI,EACjB,UAEG,IACH,EAAe,GACf,cAGA,EAAe,GACf,MAEJ,UACG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,MAWN,GANA,AAAI,IAAiB,GACnB,EAAO,KAAK,EAAK,UAAU,EAAY,EAAa,IAEpD,EAAO,KAAK,EAAK,UAAU,IAGzB,GAAwB,EAAO,SAAW,KAAK,gBAAgB,OACjE,KAAM,IAAI,OAAM,wCACZ,KAAK,gBAAgB,OAAM,+BAA+B,GAEhE,MAAO,IAEX,GA1VgC,IC7ChC,AA4BA,GAAA,IAAA,SAAA,GAAwC,GAAA,EAAA,GAgBtC,WAAuC,GAAvC,GAAA,GACE,EAAA,KAAA,OAAO,KAD8B,EAAA,iBAAA,EAf/B,EAAA,SAAW,GAiBjB,EAAK,QAAU,EAAiB,SAAW,KAC3C,GAAM,GAAc,KAAK,KAAK,EAAK,SACnC,GAAI,EAAK,QAAU,GAAK,EAAc,GAAK,EAAc,IACrD,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,oDACA,gCAA+B,EAAK,UAc1C,GAXA,EAAK,UAAY,EAAiB,yBAA2B,GAC7D,EAAK,aAAe,EAAiB,aACrC,EAAK,qBACD,EAAiB,sBAAwB,EAAK,QAClD,EAAK,sBAAwB,EAAiB,sBAC9C,EAAK,sBAAwB,EAAiB,uBAAyB,EAEvE,EAAK,mBACD,IAAiB,qBAAuB,IAC5C,EAAK,gBACD,EAAiB,kBAAoB,GACrC,CAAC,EAAK,oBAAsB,CAAC,EAAK,gBACpC,KAAM,IAAI,OACN,iHAKR,SAAA,UAAA,QAAA,WACE,MAAO,cAII,EAAA,OAAb,SAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,6FAClB,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGA,SAAqB,GAAI,GAAmB,GAGlD,CAAA,EAAM,EAAmB,gBAAzB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,8IAEI,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,KAAK,uBAAyB,KAAO,GACA,KAAK,sBACjD,MAAO,aAHT,SAAK,OAAS,EAAA,qCAMR,GAAI,OACN,iDAAiD,EAAE,gBAGzD,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,2CAQlB,GALM,EAED,OAAe,cAAiB,OAAe,mBACpD,KAAK,aAAe,GAAI,GAEpB,CAAC,KAAK,aAGR,KAAK,aAAe,KAAK,aAAa,mBAC7B,KAAK,aAAa,aAAe,KAAK,aAC/C,KAAM,IAAI,OACN,8BACA,cAAa,KAAK,aAAY,MAC9B,YAAW,KAAK,aAAa,aAG7B,SAAe,KAAK,aAAa,wBAAwB,KAAK,QACpE,KAAK,SAAW,KAAK,aAAa,iBAClC,KAAK,SAAS,QAAU,KAAK,QAAU,EACvC,KAAK,SAAS,sBAAwB,KAAK,sBAC3C,EAAa,QAAQ,KAAK,UAC1B,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,CAAA,SAGI,EAAA,UAAA,KAAN,mHACE,MAAI,MAAK,SACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAMN,CAAA,EAAM,KAAK,uBAA5B,SAAiB,EAAA,OACvB,AAAI,KAAK,oBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAoB,KAAK,4BACrB,EAAU,CAAC,KAAK,UAAW,KAAK,qBAAsB,KAE5D,AAAI,KAAK,iBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAiB,KAAK,4BAClB,EAAU,CAAC,KAAK,UAAY,KAAK,QAAS,KAGhD,CAAA,EAAO,CACL,MAAO,CAAC,YAAe,EAAmB,SAAY,GACtD,KAAM,WAMJ,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAIf,EAAA,UAAA,aAAd,+FAEQ,SAAgC,GAChC,EAAgC,GAClC,EAAgB,EACpB,CAAA,EAAO,GAAI,SAAQ,SAAA,GACjB,GAAM,GAAa,YAAY,WAC7B,AAAI,EAAK,oBACP,GAAK,SAAS,sBAAsB,EAAK,UAEzC,AAAI,EAAK,SAAS,KAAO,WACvB,EAAQ,CAAC,cAAa,EAAE,cAAa,IAEvC,EAAc,KAAK,EAAK,SAAS,MAAM,EAAG,EAAK,wBAEjD,AAAI,EAAK,iBACP,GAAK,SAAS,uBAAuB,EAAK,UAC1C,EAAc,KAAK,EAAK,SAAS,UAInC,AAAI,EAAE,IAAkB,EAAK,WAC3B,eAAc,GACd,EAAQ,CAAC,cAAa,EAAE,cAAa,MAEtC,EAAK,QAAU,EAAK,aAAe,aAK1C,EAAA,UAAA,KAAA,WACE,AAAK,KAAK,UACR,MAAK,SAAW,GAChB,KAAK,SAAS,aACd,KAAK,aAAa,QAClB,AAAI,KAAK,QAAU,MAAQ,KAAK,OAAO,YAAY,OAAS,GAC1D,KAAK,OAAO,YAAY,GAAG,SAMjC,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAIlB,EAAA,UAAA,cAAA,WACE,MAAO,MAAK,cAGN,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAY,EAAM,GAAG,OACrB,EAAW,GAAI,cAAa,EAAM,OAAS,GACjD,SAAM,QAAQ,SAAC,EAAM,GAAM,MAAA,GAAS,IAAI,EAAM,EAAI,KAC3C,GAGD,EAAA,UAAA,4BAAR,SAAoC,EAAwB,GAE1D,GAAM,GAAO,GAAI,cAAa,GAAA,KAAK,cAAc,IAEjD,SAAK,IAAI,EAAU,EAAK,OAAS,EAAS,QACnC,GAAA,OAAO,EAAM,IAExB,GA7MwC,IC5BxC,AA0BA,GAAA,IAAA,SAAA,GAAoC,GAAA,EAAA,GAQlC,WACuB,EACA,GAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KACP,GAHqB,EAAA,mBAAA,EACA,EAAA,aAAA,EATf,EAAA,SAAW,GAEX,EAAA,OAAS,GASX,EAAK,eAKP,GAJA,EAAK,OAAS,GACd,EAAK,SACD,CAAC,EAAK,aAAa,aAAc,EAAK,aAAa,aACvD,EAAK,WAAa,GAAA,SAAS,CAAC,GAAI,SAC5B,EAAK,aAAa,YAEpB,GAAM,GACF,EAAK,aAAa,YAAc,EAAM,EAAK,mBAAmB,MAC5D,EAAsB,EAAK,aAAa,aAAe,EACzD,EAAK,mBAAmB,OACtB,EAAkB,GAAI,GAAsB,EAC5C,EAAmB,GAAI,GAAuB,EAC9C,EAAe,EAAiB,EAChC,EAAgB,EAAsB,EAC5C,EAAK,QAAU,GAAA,SACX,CAAC,EAAiB,EAAgB,EAAe,GACjD,CAAC,EAAG,QAER,GAAK,QAAU,GAAA,SAAS,CAAC,EAAG,EAAG,EAAG,GAAI,CAAC,EAAG,aAKhD,SAAA,UAAA,QAAA,WACE,MAAO,UAII,EAAA,OAAb,SACI,EAAuC,GAAA,MAAA,KAAA,QAAA,GAAA,6FACzC,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGN,GAAI,CAAC,GAIH,GADA,EAAqB,SAAS,cAAc,SACxC,CAAC,EAAa,aAAe,CAAC,EAAa,aAC7C,KAAM,IAAI,OACN,0GAGN,EAAmB,MAAQ,EAAa,YACxC,EAAmB,OAAS,EAAa,aAErC,SAAiB,GAAI,GAAe,EAAoB,GAG9D,CAAA,EAAM,EAAe,gBAArB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,oHACE,AAAI,KAAK,aAAa,YACpB,GAAA,KAAK,OACA,KAAK,aAAa,aAAe,QAC7B,KAAK,aAAa,aAAe,cACtC,WACI,MAAA,+BAA+B,EAAK,aAAa,WAAU,2FAKnE,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,CACL,SAAU,KAAK,aAAa,SAC5B,WAAY,KAAK,aAAa,WAC1B,KAAK,aAAa,WAClB,OACJ,MAAO,KAAK,mBAAmB,MAC/B,OAAQ,KAAK,mBAAmB,kBAPpC,SAAK,OAAS,EAAA,qCAYd,EAAE,QAAU,iDAAiD,EAAE,QACzD,SAGR,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,uCAIlB,IACE,KAAK,mBAAmB,UAAY,KAAK,aAClC,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,OAAO,IAAI,gBAAgB,KAAK,QAGhE,YAAK,mBAAmB,OAExB,KAAK,SAAW,GAEhB,CAAA,EAAO,GAAI,SAAc,SAAA,GAEvB,EAAK,mBAAmB,iBAAmB,WACzC,cAKA,EAAA,UAAA,KAAN,oFACE,GAAI,KAAK,SACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAI7B,IACE,EAAM,GAAA,QAAQ,WAAW,KAAK,0BACvB,GACP,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAU,IAEjE,GAAI,KAAK,OACP,IACE,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,mBAAmB,GAAM,KAAM,WAC5C,GACP,KAAM,IAAI,OAAM,oCAAoC,EAAE,iBAEtD,EAAI,cAGN,OAAA,CAAA,EAAO,CAAC,MAAO,EAAK,KAAM,oBAItB,EAAA,UAAA,aAAR,WAIE,MAAI,QAAK,aAAa,aAAe,KAAK,aAAa,cAClD,MAAK,mBAAmB,QAAU,KAAK,aAAa,aACpD,KAAK,mBAAmB,SAAW,KAAK,aAAa,gBAO5D,EAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,IAAA,KAAK,WACV,GAAM,GAA0B,EAAI,UAAU,WAAW,GACrD,EACJ,EAAe,GAAA,MAAM,cACjB,EAAe,EAAK,QAAS,EAAK,WAAY,EAAK,SACnD,YAEJ,GAAM,GAAQ,EAAa,MAC3B,MAAO,GAAa,QAAQ,EAAM,MAAM,OAMtC,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAI7B,EAAA,UAAA,KAAA,WACE,GAAM,GAAS,KAAK,OAAO,YAE3B,EAAO,QAAQ,SAAA,GAAS,MAAA,GAAM,SAE9B,IACE,KAAK,mBAAmB,UAAY,WAC7B,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,KAEhC,KAAK,SAAW,IAIlB,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAEpB,GApMoC,IC1BpC,AA2BA,GAAA,IAAA,WAAA,cAUA,MAAA,MCrCA,AAoBA,GAAA,IAAA,SAAA,GAA6C,GAAA,EAAA,GAA7C,4DAmBE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAI,IAAc,KAAM,IAEnC,GAtB6C,IAmC7C,GAAA,SAAA,GAA4B,GAAA,EAAA,GAG1B,WAAsB,EAAgC,GAAtD,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAkB,EAAU,KAG9C,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf4B,IAiB5B,GAAA,SAAA,GAAgC,GAAA,EAAA,GAI9B,WACc,EAA0C,GADxD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAA0C,EAAA,UAAA,EAHxD,EAAA,UAAY,KAOZ,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAAc,KAAK,UAAS,MAGzD,EAAA,UAAA,KAAN,mHACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eACxC,GADM,EAAc,EAAA,OAChB,EAAY,KACd,MAAI,MAAK,YAAc,GACrB,CAAA,EAAO,IAKT,MAAK,YAAY,KAAK,KAAK,WAC3B,KAAK,UAAY,GACjB,CAAA,EAAO,KAQT,IANM,EAAQ,EAAY,MAAM,MAAM,KAAK,WAK3C,EAAM,GAAK,KAAK,UAAY,EAAM,GAClC,EAAA,EAAmB,EAAA,EAAM,MAAM,EAAG,IAAf,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,KAAK,YAAY,KAAK,GAExB,YAAK,UAAY,EAAM,EAAM,OAAS,GAEtC,CAAA,EAAO,UAEX,GAvCgC,ICxEhC,AAsBA,GAAA,IAAA,SAAA,GAAgD,GAAA,EAAA,GAAhD,4DAUE,SAAA,UAAA,WAAA,WACE,MAAO,IAAI,IAAa,OAE5B,GAbgD,IA0BhD,GAAA,SAAA,GAA2B,GAAA,EAAA,GAGzB,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAiB,KAGnC,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf2B,IAuC3B,GAAA,SAAA,GAA+B,GAAA,EAAA,GAM7B,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAF6B,EAAA,SAAA,EAEzB,GAAA,MAAM,IAAI,cACZ,EAAK,QAAU,GAAI,aAAY,cAGxB,GAAA,GAAA,KAAA,cACP,EAAK,QAAU,GAAI,GAAc,iBAGrC,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,+GACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eAExC,MAFM,GAAc,EAAA,OAEhB,EAAY,KACd,CAAA,EAAO,IAEP,GAAQ,EAAY,MAItB,AAAI,GAAA,MAAM,IAAI,cACZ,EAAO,KAAK,QAAQ,OAAO,EAAO,CAAC,OAAQ,KAE3C,EAAO,KAAK,QAAQ,MAAM,OAAO,KAAK,EAAM,SAE9C,KAAK,YAAY,KAAK,GACtB,CAAA,EAAO,WAEX,GAtC+B,ICvF/B,AAqCA,GAAA,IAAA,SAAA,GAAuC,GAAA,EAAA,GAIrC,WACc,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,KAAA,EACA,EAAA,QAAA,EAEZ,GAAA,KAAK,OACA,YAAgB,aACZ,IAAA,MAAM,IAAI,cACL,YAAgB,OAAQ,YAAgB,MACzC,IACT,WAAM,MAAA,yEAEV,EAAK,OAAS,EAAQ,QAAU,EAEhC,EAAK,UAAY,EAAQ,WAAa,KAAO,OAG/C,SAAA,UAAA,QAAA,WACE,MAAO,cAAc,KAAK,MAGtB,EAAA,UAAA,KAAN,oHACE,MAAI,MAAK,QAAY,MAAK,eAAgB,YAClB,KAAK,KAAK,WACV,KAAK,KAAK,MAChC,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAQ,GAAI,SAAoB,SAAC,EAAS,GAC9C,GAAM,GAAM,EAAK,OAAS,EAAK,UAC/B,GAAI,EAAK,eAAgB,YAGvB,EAAQ,GAAI,YAAW,EAAK,KAAK,MAAM,EAAK,OAAQ,UAMpD,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GACnB,GAAI,GAAsC,EAAW,OAOrD,GAHA,AAAI,YAAgB,cAClB,GAAO,GAAI,YAAW,IAEpB,CAAE,aAAgB,aACpB,MAAO,GAAO,GAAI,WAAU,sCAE9B,EAAQ,IAEV,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,aAE1B,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,EAAM,QAIhC,GAAM,GAAQ,EAAK,KAAK,MAAM,EAAK,OAAQ,GAG3C,EAAW,kBAAkB,GAE/B,EAAK,OAAS,SAEA,CAAA,EAAM,WAAtB,MAAA,CAAA,EAAQ,GAAA,MAAQ,EAAA,OAAc,EAAA,KAAM,GAAK,UAE7C,GAxEuC,ICrCvC,YA6BI,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,qGAGpB,MAAK,OAAO,IAAS,SACnB,EAAY,EAEZ,GAAa,EAAgB,IAC7B,EAAc,GAA0B,IAEzB,CAAA,EAAM,GAAA,KAAK,MAAM,EAAW,iBAAvC,GAAW,EAAA,OACb,EAAS,MACY,WAAU,KAAC,CAAA,EAAM,EAAS,gBAD/C,CAAA,EAAA,UACI,SAAa,GAAA,GAAA,MAAI,WAAU,CAAA,OAAC,EAAA,UAClC,CAAA,EAAO,GAAI,IAAkB,EAAY,WAEzC,KAAM,IAAI,OAAM,EAAS,iBAK7B,GAAM,IAA4B,SAAC,GACjC,GAAM,GAAO,CACX,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,KACd,KAAM,EAAQ,KACd,YAAa,EAAQ,YACrB,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,SAAU,EAAQ,SAClB,UAAW,EAAQ,WAErB,MAAO,IC5DT,YAqB4B,GAC1B,MAAQ,OAAO,IAAW,UAAa,EAAO,OAAO,EAAG,KAAO,UCtBjE,mBA6BoC,GAAA,EAAA,GASlC,WACc,EACS,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,MAAA,EACS,EAAA,QAAA,IAIjB,SAAA,UAAA,SAAN,oFACE,MAAI,IAAY,KAAK,QAAU,GAAA,MAAM,IAAI,YAEjC,GAAK,cACX,KAAK,MAAQ,EAAG,aAAc,KAAK,MAAiB,OAAO,KAI7D,CAAA,EAAO,GAAI,IAAkB,KAAK,MAAsB,KAAK,eAEjE,GAzBoC,IC7BpC,mBA4BmC,GAAA,EAAA,GAQjC,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,IAAA,EACA,EAAA,YAAA,IAQjB,SAAA,UAAA,SAAN,8EACE,MAAI,IAAY,KAAK,KACnB,CAAA,EAAQ,GAAI,IAAe,KAAK,IAAe,KAAK,aAC/C,YAEL,CAAA,EAAO,GAAiB,KAAK,IAAK,KAAK,mBAG7C,GA1BmC,IC5BnC,YA0GI,EAAqB,GAAA,MAAA,KAAA,QAAA,GAAA,IAChB,GAAI,IAAW,GAAI,IAAc,GAAS,eA4B/C,GADJ,GAAA,GAAA,KAEQ,EAAO,GAAqB,GAClC,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,qBAgEvC,GADJ,GAAA,GAAA,KAEE,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACf,MAAA,CAAA,EAAM,YAAZ,SAAM,EAAA,OACZ,CAAA,EAAO,GAAqB,WAAM,MAAA,GAAI,6BAmCtC,EACA,sEACF,MAAA,CAAA,EAAO,GAAe,OAAO,EAAoB,oBAqClB,sEAE/B,MAAA,CAAA,EAAO,GAAmB,OAAO,QCxRnC,GAGM,IAAU,oNCHhB,oBA2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9GnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC7EnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UClFnC,oBAKA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAER,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAG,EAC9B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9FnC,oBAyBA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UChJnC,oBAIA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UCnGnC,oBAwBA,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMjB,MAAO,KAAW,UAAY,GAAO,SACxC,GAAO,QAAU,EAEjB,IACE,EAAa,WACN,SACJ,AAAK,OAAO,SAAW,YAAc,OAAO,KACjD,OAAO,WAAa,MAAO,OAK3B,GACA,QCrPF,oBAYA,GAAI,IAAO,KAKP,GAAS,KAKT,GAAS,KAQT,GAAY,KASZ,GAAU,KAOV,GAAS,KAIT,GAAK,KAET,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAO,QAAU,6vDC3DjB,YAoBI,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MACF,GAAM,8DC7BpB,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBACvC,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAM/B,YACI,EAAyB,EAAW,EACpC,GACF,GAAI,IAAe,SACjB,MAAO,GAAQ,OAAO,GACjB,GAAI,IAAe,OACxB,MAAO,GAAQ,KAAK,GACf,GAAI,IAAe,MACxB,MAAO,GAAA,IAAO,GACT,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,GAChB,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,EAAG,GAE1B,KAAM,IAAI,OACN,cAAc,EAAU,qEAeM,GAAA,EAAA,GAMlC,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KANF,SAAA,UAAY,GAGX,EAAA,SAAW,GAIjB,EAAK,KAAO,GAAI,GAAA,YAAY,EAAM,EAAA,YAGpC,SAAA,UAAA,MAAA,SAAM,EAAoC,EAAiB,GAEzD,AAAI,KAAK,UACP,MAAK,SAAW,GAChB,AAAI,EAAA,MAAM,IAAI,YACZ,EAAA,aAAa,KACT;;;gCAYR,GAAM,GAAS,GAEf,YAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,IAEzC,GAST,EAAA,UAAA,eAAA,SACI,EAAiB,EACjB,GACF,GAAM,GAAQ,KAAK,MAAM,EAAQ,EAAO,GAExC,MAAO,CAAC,OAAQ,EAAO,MAAK,EAAE,MAAK,IAIrC,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,YAIb,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,aAIf,EAAA,UAAA,KAAA,SACI,EAAgB,EAAoC,EACpD,GACF,KAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,KAGlD,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,KAAK,cAGb,EAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,SAAS,SAEvB,EAAA,UAAA,SAAA,SAAS,GACD,GAAA,GAAA,KAAA,KAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,mBAEd,GAAI,IAAU,aACZ,GAAM,GACF,KAAK,SAAS,EAAmB,KAAK,QACpC,EACF,KAAK,SAAS,EAAmB,KAAK,QAC1C,MAAO,GAAA,aAAa,uBAAuB,EAAY,GAGzD,MAAO,MAAK,KAAK,IAAI,GAAQ,QAGvB,EAAA,UAAA,WAAR,SAAmC,GACjC,GAAM,GAAO,KAAK,SAAS,EAAE,QACzB,EAAc,EAClB,GAAI,EAAE,QAAU,SACd,IAEE,EAAe,EAAsB,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,WAChE,GACA,KAAM,IAAI,OAAM,oDAGpB,MAAO,GAAA,OAAU,EAAE,MAAO,EAAE,MAAO,IAGrC,EAAA,UAAA,WAAA,SACI,EAAoC,EAAiB,GACvD,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAG7D,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,KAAK,IAAI,IACT,GAAA,GAAA,KAAA,KAAA,IAAA,GAAA,mBAEP,AAAI,GAAsB,MACxB,MAAK,YAAY,EAAmB,KAAK,QACzC,KAAK,YAAY,EAAmB,KAAK,SAG3C,KAAK,KAAK,OAAO,KAIrB,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GAEjC,EAAW,WAEX,AAAI,EAAW,SAAW,GACxB,KAAK,YAAY,KAKjB,EAAA,UAAA,KAAN,SAAW,8EACH,SAAQ,EAAA,KAAK,MACnB,IACM,EAAW,EAAA,KAAK,MAAQ,EAC9B,CAAA,EAAO,CAAC,SAAQ,SAGlB,EAAA,UAAA,OAAA,WACE,MAAO,CAEL,WAAY,GACZ,QACI,CAAC,wHAKT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,GAAiB,EAAG,gBAEpB,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAU,GAAI,GAKvB,OAFM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GACpB,EAAI,EAAG,EAAI,EAAO,KAAM,KAI/B,OAHM,GAAM,EAAO,WAAW,GAExB,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,GAAK,EAAQ,GAAK,EAAM,GAE1C,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAM,OAAM,IAGrC,MAAO,GAAO,YAGhB,EAAA,UAAA,KAAA,SAAK,GAIH,OAHM,GAAQ,KAAK,SAAS,EAAE,QACxB,EAAS,EAAA,OAAU,CAAC,EAAE,KAAM,EAAE,MAAO,EAAE,OACvC,EAAO,EAAO,OACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAK,EAAI,EAAE,KAAO,GAAK,EAAM,GAE/B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,EAAA,MAAS,EAAG,EAAO,GAAM,QAAQ,GAE5C,MAAO,IAGT,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAiB,EAAG,WAKpB,OAHM,GAAS,EAAA,OAAU,EAAE,MAAO,EAAE,OAC9B,EAAO,KAAK,WAAW,cAEpB,GACP,GAAM,GAAS,EAAO,WAAW,GAC3B,EAAQ,EAAO,QACrB,EAAK,QAAQ,SAAA,GAAM,MAAA,GAAM,GAAM,EAAE,MAAM,GAAM,EAAI,EAAM,KACvD,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAK,OAAM,KAJ3B,EAAI,EAAG,EAAI,EAAO,KAAM,MAAxB,GAOT,MAAO,GAAO,YAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,UAAiB,EAAG,OAGb,EAAA,IAAO,EAAA,OAAU,IAAK,IAG/B,EAAA,UAAA,KAAA,SAAuB,GAAvB,GAAA,GAAA,KACE,GAAiB,EAAS,QAK1B,OAHM,GAAO,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAK,SAAS,EAAE,UACxC,EAAS,EAAA,OAAU,EAAQ,GAAG,MAAO,EAAQ,GAAG,OAChD,EAAa,EAAO,OACjB,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAElC,OADM,GAAW,EAAK,GACb,EAAI,EAAG,EAAI,EAAW,OAAQ,IACrC,EAAW,IAAM,EAAS,GAG9B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAGhD,EAAI,EAAA,IAAO,EAAQ,EAAS,QAAQ,IACpC,EAAI,EAAA,IAAO,GACX,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAO,EAAG,IAGnB,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,IAAI,EAAQ,MAIjE,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAiB,CAAC,EAAG,GAAI,UAqBzB,OAnBM,GAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAU,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5C,EAAW,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7C,EAAW,EAAE,MAAM,GAEnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,8DAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAAY,EAAA,EAAA,GAGrB,EAAA,8DAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAIzB,EAAO,EAAU,EACjB,EAAS,EAAA,OAAU,CAAC,EAAU,EAAS,GAAW,EAAE,OACpD,EAAU,EAAO,OACjB,EAAY,KAAK,UAEd,EAAI,EAAG,EAAI,EAAU,IAC5B,OAAS,GAAK,EAAG,EAAK,EAAS,GAAM,EACnC,OAAS,GAAK,EAAG,EAAK,EAAU,GAAM,EACpC,OAAS,GAAK,EAAG,EAAK,EAAW,GAAM,EAMrC,OAJM,GAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAE/B,EAAI,EAAI,EAAI,EAAQ,IAC3B,OAAS,GAAI,EAAI,EAAI,EAAQ,KAG3B,OAFI,GAAM,EAED,EAAI,EAAI,EAAI,EAAQ,IAC3B,GAAO,EAAQ,EAAI,EAAS,EAAI,EAAa,EAAI,GAC7C,EAAQ,EAAI,EAAa,EAAI,EAAa,EAAI,GAEpD,EAAQ,EAAI,EAAQ,GAAI,EAAW,KAAO,EAOtD,MAAO,GAAO,YAGhB,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE/C,EAAS,KAAK,YAAY,EAAG,EAAG,EAAY,GAChD,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAIvC,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAiB,CAAC,EAAG,GAAI,YAEzB,GAAM,GAAK,SAAC,EAAW,GAAc,MAAA,MAAK,MAAM,EAAI,IAC9C,EAAc,QACpB,MAAO,MAAK,oBAAoB,EAAG,EAAG,EAAa,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MASvD,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EACD,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAO,EAAM,EAAS,GAExB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,GAAiB,EAAG,OAUpB,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAO,EACF,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAQ,EAAM,EAAS,GAEzB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,EAAG,sBAOpB,OALM,GAAM,GAIN,EAAW,EAAE,KAAO,EAAW,KAC5B,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,EAAW,WAAW,EAAI,GAGzC,OAAS,GAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAY,EAAA,OAAU,EAAG,SACzB,EAAO,EAAA,MAAS,EAAW,GAAY,OAAO,WAC9C,EAAM,EAAK,IAAI,GAAG,IAAI,GAC5B,EAAI,KAAK,GAGX,MAAO,GAAA,MAAS,IAGlB,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAIlD,GAFA,GAAiB,EAAG,UAEhB,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,oDAAoD,GAAE,KAAO,GAAC,IAC9D,iBAAgB,IAWtB,OATM,GAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAE,MAAO,GAC3B,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAW,EAAE,MAAM,EAAE,KAAO,GAC5B,EAAgB,EAClB,SAAC,EAAW,GAAc,MAAA,GAAI,EAAW,EAAI,GAC7C,SAAC,EAAW,GAAc,MAAA,GAAI,GACzB,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACrC,OAAS,GAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAM,EAAc,EAAG,GAC7B,GAAI,IAAM,EACR,EAAK,GAAO,EAAY,EAAI,EAAM,QAElC,GAAM,GAAU,EAAc,EAAG,EAAI,GACrC,EAAK,GAAO,EAAY,EAAM,GAAW,EAAK,GACtB,EAAM,GAAO,EAAK,IAIhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,UAAiB,CAAC,EAAG,GAAI,YAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,UAAiB,CAAC,EAAG,GAAI,QAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,UAAiB,CAAC,EAAG,GAAI,gBAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,UAAiB,CAAC,EAAG,GAAI,cAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAiB,CAAC,EAAW,EAAG,GAAI,UAYpC,OAVM,GAAS,KAAK,SAAS,EAAU,QACjC,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,MAAS,EAAE,MAAO,EAAA,WAAW,EAAE,MAAO,EAAE,QACjD,EAAY,KAAK,SAAS,EAAO,QACnC,EAAQ,EACN,EAAS,EAAU,OAAS,GAAK,EAAU,KAAO,GAAK,EAAE,OAAS,EACpE,EACA,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAE5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,OAAS,GAAI,EAAG,EAAI,EAAQ,IAC1B,AAAI,EAAO,KAAO,EAChB,EAAU,KAAW,EAAQ,GAE7B,EAAU,KAAW,EAAQ,GAKnC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GACJ,GAAiB,CAAC,GAAY,SAE9B,GAAM,GAAW,KAAK,SAAS,EAAU,QACzC,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAiB,EAAG,QAEpB,GAAM,GAAQ,KAAK,SAAS,EAAE,QAC9B,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAM,EAAO,EACnB,MAAK,GAAO,GAAK,EAAO,GAAO,GAAQ,GAAK,GAAQ,EAC3C,EAEC,GAAM,GAAQ,KAK5B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAM,GAAO,EAEf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAS,EAAM,GACV,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAS,GAAU,EAErB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,UAAiB,CAAC,EAAG,GAAI,qBAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAO,EAAO,EACpB,MAAO,GAAO,KAIlB,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,IAGT,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,EAAG,EAAO,IAElC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,KAAK,IAAI,EAAG,EAAO,IAAK,GAEhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MACR,SAAC,EAAQ,GAAW,MAAA,GAAS,EAAI,EAAS,EAAS,KAGhE,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAiB,CAAC,EAAI,GAAI,UAK1B,OAHM,GAAe,GAAI,cAAa,EAAE,MAClC,EAAS,KAAK,SAAS,EAAE,QACzB,EAAW,KAAK,SAAS,EAAG,QACzB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,GACjB,AAAI,GAAK,EACP,EAAa,GAAK,EAAS,GAE3B,EAAa,GAAK,EAAS,GAAM,GAAI,GAGzC,MAAO,MAAK,WAAW,EAAc,EAAE,MAAO,YAGhD,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,MAAM,EAAQ,MAInE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,OAAO,EAAO,EAAQ,GAExC,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAiB,CAAC,EAAG,GAAS,UAyB9B,OAvBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAiB,EAAS,aAAe,eAEzC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OAEnC,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAChD,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAEhD,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EACf,EAAW,EAAI,EACZ,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EACxB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,IAAK,EAAW,EAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAC7B,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,GAAK,GAC1B,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,GAAK,IAClB,GAAO,EAAM,GAAW,IAE9B,IAAY,EAAS,cAOjC,MAAO,GAAE,YAGX,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAiBpC,OAfM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAI,EAAA,OAAmB,EAAS,SAAU,EAAE,OAE5C,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAKF,OAHM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAElC,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAS,YACpC,EAAW,EAAK,EAAS,YAAc,EACpC,GAAK,EAAG,GAAK,EAAa,MACjC,GAAM,IAAK,EAAW,GAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,GAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAAS,WACtC,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,IACrB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,KAAO,GAAO,EAAM,GAAW,IAElD,IAAY,EAAS,eASrC,MAAO,GAAE,YAGX,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,kBAkC/B,OAhCM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAiB,IAAe,eAChC,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GACjD,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GAE9C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,IAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAOjC,OANM,IAAK,GAAK,EAAc,GACxB,GACF,EAAe,EAAI,EAAa,GAAK,EAAa,GAChD,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,EAAiB,IAC7C,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,GAAM,IAAW,EAAe,EAAI,EAAa,GAC7C,EAAa,GAAK,EAAiB,EACvC,EAAS,IAAY,GAK7B,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GA6BF,OA5BM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAE1B,EAAA,EAAA,UACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,GAAA,EAAA,YAEI,GAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,GAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,GAAU,EAAc,EAAI,EAAS,QAAQ,KAE1C,GAAI,EAAG,GAAI,EAAW,EAAE,GAC/B,OAAS,IAAK,EAAG,GAAK,EAAY,EAAE,GAElC,OAAS,IAAK,EAAG,GAAK,EAAS,EAAE,GAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAGzC,GAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,KACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,IAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAc,GAErB,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAQjC,OAPM,IAAK,GAAK,GAAc,GACxB,GACF,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GACxC,GAAY,EAAS,GAAc,EAAI,IACzC,EAAS,GAAe,EAAI,IAC5B,EAAS,GAAc,EAAI,IAAM,EAAQ,GAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAK3B,EAAS,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GAAK,IACpD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAiB,CAAC,EAAG,GAAK,mBAa1B,OAXM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eACzC,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,EAC3C,OAAS,GAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAG5C,OADI,GAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,AAAI,EACF,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAEnD,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAK3D,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAKpC,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAqBzC,OAnBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAC9C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAEhB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAEvB,EAAK,EAAG,EAAK,EAAa,EAAE,EAMnC,OALM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAW,GAAM,IAChD,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAW,GAAM,GACtD,EAAW,EAAK,EAEb,EAAK,EAAG,EAAK,EAAc,EAAE,EAOpC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,GAAQ,KAAK,IACf,EAAS,UACR,GAAS,SAAW,EAAS,GAAM,GAClC,GAAW,EAAK,EAAO,EAEpB,GAAK,EAAG,GAAK,EAAa,EAAE,GAOnC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,IAAM,IAC/C,GAAQ,KAAK,IACf,EAAS,SACR,GAAS,QAAU,EAAU,IAAM,GAClC,GAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,GAG3C,OAFM,IAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,IAE5C,OADI,IAAU,EACL,GAAI,EAAG,GAAI,EAAS,UAAW,EAAE,GAIxC,OAHM,IAAW,GAAI,EACf,GAAW,GAAI,EAEZ,GAAK,EAAO,GAAK,EAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAe,EAC9B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,IACjC,GAAM,IAAK,GAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAE7B,IACI,EAAQ,GAAW,IAAM,EAAS,GAAW,IAKzD,EAAS,GAAW,IAAM,GAMpC,MAAO,GAAG,YAGZ,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,gBAAgB,EAAO,EAAQ,GAEjD,MAAI,IAGF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAiB,CAAC,EAAG,GAAS,mBAc9B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WACxC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OACnC,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,EAAE,GACpC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,EAAE,GACnC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAMF,OAJM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAS,WACtC,EAAW,EACX,EAAW,EACN,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,GAE3C,OADM,GAAO,EAAM,EAAW,GACrB,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAM,EAAW,IAAM,EAAO,EAAM,EAAW,GAEjD,GAAY,EACZ,GAAY,IAQxB,MAAO,GAAE,YAGX,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,2BA0B/B,OAxBM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAQ,EAAc,EAEnB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,GAAK,EAAG,EAAK,EAAU,EAAE,EAMhC,OALM,IAAW,EAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAMjC,OALM,IAAK,GAAK,EAAc,GACxB,GAAW,EAAO,EAAI,EAAO,GAAK,EAAO,GACzC,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAO,EAAE,IAC7B,GAAM,IAAK,EAAK,EAAQ,GAClB,GAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,EAAS,EAAO,EAAI,EAAO,EAAK,EAAO,GAAK,GAAM,GAK1D,MAAO,GAAG,YAGZ,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAiB,CAAC,EAAG,GAAK,4BAc1B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WAExC,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAK5C,OAJM,GAAK,KAAK,MAAM,EAAK,GACrB,EAAK,EAAK,EAEZ,EAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,GAAW,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAIhE,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAIlC,MAAO,GAAG,YAGZ,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,UAAiB,EAAG,QACb,GAAK,KAAK,WAAW,GAAI,IAGlC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,GAAiB,CAAC,EAAG,GAAU,UAE/B,GAAM,GAAqB,EAAE,MAAM,QAC7B,EAAgB,KAAK,SAAS,EAAQ,QAC5C,EAAS,GAAQ,EAAc,OAI/B,OAHM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GAEpB,EAAI,EAAG,EAAI,EAAO,KAAM,EAAE,GACjC,GAAM,GAAS,EAAO,WAAW,GAE3B,EAAwB,EAAO,QACrC,EAAY,GAAQ,EAAc,EAAO,IAEzC,GAAM,GAAgB,EAAK,WAAW,GACtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,YAGhB,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAa,EAAE,QAAQ,GAAW,GAC7B,QAAQ,GACR,MAAM,EAAkB,IAG9B,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAiB,EAAG,UA8BpB,OA5BM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,OAAU,EAAS,SAAU,EAAE,OACxC,EAAa,EAAO,OAEpB,EAAqB,EAAS,SAAS,GAAK,EAAS,SAAS,GAChE,EAAS,SAAS,GAAK,EAAS,SAAS,GACvC,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAGhD,OAFM,GAAoB,EAAQ,EAC5B,EAAmB,EAAQ,EAAE,QAAQ,GAClC,EAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAMf,OAJM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAChD,EACF,EAAoB,EAAS,EACxB,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAKb,OAHM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,GAAkB,EAAoB,EAAO,EAC1C,GAAO,EAAG,GAAO,EAAS,SAAU,EAAE,IAG7C,OAFM,IAAa,GAAO,EAAc,EACpC,GAAU,GACP,GAAU,GACf,IAAW,EASb,OAPM,IACF,KAAK,IAAI,EAAS,QAAS,EAAuB,IAEhD,GAAkB,GAAkB,GAAO,EAC7C,GAAc,EACd,GAAW,EACX,GAAQ,EACH,GAAS,EAAW,GAAS,EACjC,IAAU,GAEb,OADM,IAAe,EAAmB,GAAS,EAAE,QAAQ,GAClD,GAAO,EAAS,GAAO,EAC3B,IAAQ,GAEX,OADM,IAAa,GAAe,GAAO,EAAE,QAAQ,GAC1C,GAAO,GAAS,GAAO,GAC3B,IAAQ,GACX,GAAM,IAAa,GAAa,GAAO,EAAE,QAAQ,GAC3C,GAAQ,EAAQ,GAAa,GAOnC,GANA,AAAK,IAAa,OAAS,GAAQ,GACjC,GAAc,GACT,AAAI,IAAa,OACtB,KAAY,GACZ,MAEE,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAM,IAAe,GAAkB,EACvC,EAAW,IACP,IAAa,MAAQ,GAAW,GAAQ,KAMtD,MAAO,GAAO,YAGhB,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGzC,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,qBAuB1B,OArBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAgB,EAAK,GAAc,EAAe,GAElD,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,KAIjB,EAAG,IACC,EAAU,EAAe,EAAO,EAAS,EAAO,EAChD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGjC,EAAA,UAAA,mBAAR,SAA2B,EAAa,GAiBtC,OAfM,GAAe,EAAA,OAAU,EAAS,SAAU,SAC5C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,KAAK,WAAW,GACpB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAIf,OAFM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAC7C,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAIb,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAO,EAAG,EAAO,EAAS,SAAU,EAAE,GAG7C,OAFM,GAAa,EAAO,EAAc,EACpC,EAAU,EACP,EAAU,GACf,GAAW,EASb,OAPM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAGlD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAS,EAAW,EAAS,EACjC,GAAU,EAEb,OADM,GAAS,EAAS,EACf,EAAO,EAAS,EAAO,EAC3B,GAAQ,EAEX,OADM,GAAO,EAAO,EACX,GAAO,EAAS,GAAO,EAC3B,IAAQ,GACX,GAAM,IAAO,GAAO,EACd,GAAQ,EAAK,IAAI,EAAO,EAAQ,EAAM,GAAM,GAClD,AAAI,IAAS,GACX,GAAW,GACX,EAAc,EAAS,EACf,EACJ,EAAO,EAAwB,IAM3C,EAAa,IAAI,EAAa,EAAO,EAAQ,EAAM,EAAM,KAMnE,MAAO,GAAa,YAGtB,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAiB,CAAC,EAAG,GAAI,qBAoBzB,OAlBM,GAAe,KAAK,mBAAmB,EAAG,GAC1C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAY,KAAK,WAAW,GAC5B,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,GAAS,EACP,EAAwB,EAC5B,EACA,EAAU,IAAI,EAAO,EAAS,EAAO,EAAO,GAC1C,GACF,EAAS,EAAwB,EACjC,EAAO,EAAuB,EAE5B,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,GAAQ,KAIzB,EAAG,IAAI,EAAS,EAAO,EAAS,EAAO,EAAO,GAMxD,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,kBAqBpB,OAnBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cACf,EAAA,KAAK,cAAc,CAAC,EAAO,EAAW,EAAU,KAE9C,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAE9C,EAAY,EACV,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GACvC,EAAI,EAAG,EAAI,EAAO,IACzB,OAAS,GAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EAAgB,KAAK,IAAI,EAAY,EAAG,KAAK,KAAK,IAClD,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAiB,EAAE,QAAQ,GAC7D,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAgB,EAAE,QAAQ,GACzD,EAAI,EAAG,EAAI,EAAU,IAU5B,OATM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EACF,KAAK,IAAI,EAAW,EAAG,KAAK,KAAK,IAC/B,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GACvD,EAAI,EAAG,EAAI,EAAa,KAI/B,GAAM,GAAU,EAAQ,EAAgB,GAClC,EAAa,EAAQ,EAAgB,GACrC,GAAW,EAAQ,EAAiB,GACpC,GAAc,EAAQ,EAAiB,GAEvC,GAAM,EAAW,IAAW,GAAW,EACvC,GAAS,EAAc,IAAc,GAAc,EACnD,GAAW,GAAO,IAAS,IAAO,EAExC,EAAO,KAAe,GAK9B,MAAO,GAAA,OAAU,EAAQ,CAAC,EAAO,EAAW,EAAU,KAGxD,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAChD,GAAiB,CAAC,EAAI,GAAI,0BA+B1B,OA7BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GAOrD,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAMhD,EAAW,KAAK,SAAS,EAAG,QAC9B,EAAS,EACJ,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAU,EAAI,EAAE,QAAQ,GACrB,EAAI,EAAG,EAAI,EAAS,IAU3B,OATM,GAAM,EAAI,EACV,EAAc,KAAK,MAAM,GACzB,EAAiB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAU,GAEpD,EAAe,EAAU,EAAc,EAAE,QAAQ,GACjD,EAAkB,EAAU,EAAiB,EAAE,QAAQ,GAEvD,EAAU,EAAM,EAChB,EAAiB,EAAM,EACpB,EAAI,EAAG,EAAI,EAAQ,IAmB1B,OAlBM,GAAM,EAAI,EACV,EAAe,KAAK,MAAM,GAC1B,EAAgB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAS,GAClD,EAAU,EAAM,EAChB,EAAiB,EAAM,EAEvB,EAAkB,EAAe,EAAe,EAAE,QAAQ,GAC1D,EAAmB,EAAe,EAAgB,EAAE,QAAQ,GAC5D,GACF,EAAkB,EAAe,EAAE,QAAQ,GACzC,GACF,EAAkB,EAAgB,EAAE,QAAQ,GAE1C,GACF,EAAiB,EACf,GAA6B,EAAiB,EAC9C,GAA6B,EAAU,EACvC,GAAsB,EAAU,EAC7B,GAAI,EAAG,GAAI,EAAO,MACzB,GAAM,IAAQ,EAAS,KACvB,EAAO,EAAkB,KACrB,GAAQ,GACZ,EAAO,EAAmB,KAAM,GAAQ,GACxC,EAAO,GAAqB,KACxB,GAAQ,GACZ,EAAO,GAAsB,KAAM,GAAQ,GAKnD,MAAO,GAAA,SAAY,EAAQ,CAAC,EAAO,EAAQ,EAAS,GAAQ,EAAE,QAGhE,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,yBAsBpB,OApBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cAAa,EAAQ,EAAY,EAAW,GAEzD,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GAE5C,EAAe,EACV,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAY,EACZ,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAc,EAAmB,EAAE,QAAQ,GACpD,EAAI,EAAG,EAAI,EAAU,IAO5B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAW,EACX,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAY,EAAmB,EAAE,QAAQ,GAClD,EAAI,EAAG,EAAI,EAAa,KAG/B,GAAM,GAAS,EAAQ,EAAY,GACnC,EAAO,KAAkB,EAKjC,MAAO,GAAA,OACH,EAAQ,CAAC,EAAO,EAAW,EAAU,GAAc,EAAE,QAG3D,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,iCAiC1B,OA/BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GACrD,EAAW,KAAK,SAAS,EAAG,QAK5B,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAGzC,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAS,IAM3B,OALM,GAAY,EAAc,EAAI,EAAE,QAAQ,GAGxC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAY,GAC7C,EAAI,EAAG,EAAI,EAAQ,IAO1B,OANM,GAAY,EAAY,EAAI,EAAE,QAAQ,GAGtC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAW,GAE5C,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHI,GAAQ,EAGH,EAAW,EAAG,EAAW,EAAW,KAC3C,GAAM,GAAM,EAAW,EAEvB,GAAI,EAAM,GAAK,GAAO,EACpB,SAGF,GAAM,GAAY,EAAc,EAAM,EAAG,QAAQ,GAC3C,GAAgB,EAAM,EACtB,GAAmB,KAAK,IAC1B,EAAU,EACV,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAC9B,GAAI,IAAM,GACR,SAEF,OAAS,IAAW,EAAG,GAAW,EAAU,MAC1C,GAAM,IAAM,GAAW,EAEvB,GAAI,GAAM,GAAK,IAAO,EACpB,SAGF,GAAM,IAAY,EAAY,GAAM,EAAG,QAAQ,GACzC,GAAgB,GAAM,EACtB,GAAmB,KAAK,IAC1B,EAAS,EACT,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAE9B,AAAI,IAAM,IACR,IAAS,EAAS,GAAY,KAIpC,EAAO,EAAY,GAAK,EAKhC,MAAO,GAAA,SAAY,EAAQ,EAAE,MAAO,EAAE,QAGxC,EAAA,UAAA,6BAAA,SACI,EAAa,EAAqB,EAAc,EAChD,GACF,GAAiB,EAAG,gCAEpB,GAAM,GAAW,EAAE,MAAM,GACnB,EAAO,EAAW,EAClB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAO,EAAE,KACT,EAAS,GAAI,cAAa,GAEhC,WAA2B,GAQzB,OAPM,GAAiB,EAAS,EAC5B,EACA,EAAS,EAAiB,KAAK,IAAI,EAAG,EAAiB,GACrD,EAAe,EAAS,EAC1B,KAAK,IAAI,EAAiB,EAAa,GAEvC,EAAM,EACH,GAAkB,EAAc,KACrC,GAAM,GAAI,EAAQ,GAClB,GAAO,EAAI,EAEb,MAAO,GAGT,OAAS,GAAS,EAAG,EAAS,EAAM,KAClC,GAAM,GAAM,EAAkB,GACxB,EAAM,EAAQ,GAAU,KAAK,IAAI,EAAO,EAAQ,EAAK,CAAC,GAC5D,EAAO,GAAU,EAGnB,MAAO,GAAA,SAAY,EAAQ,EAAE,QAG/B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAiB,EAAI,WAQrB,OAPM,GAAW,EAAG,MAAM,GACpB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAmB,KAAK,SAAS,EAAW,QAC5C,EAAoB,KAAK,SAAS,EAAY,QAC9C,EAAS,GAAI,cAAa,EAAG,MAC7B,EAAO,EAAG,KAEP,EAAS,EAAG,EAAS,EAAM,KAQlC,OAPM,GAAiB,EAAS,EAC1B,EACD,EAAS,EAAkB,KAAK,IAAI,EAAG,EAAiB,GACvD,EAAY,EAAS,EACvB,KAAK,IAAI,EAAU,EAAiB,EAAc,GAElD,EAAO,EACF,EAAI,EAAY,EAAI,EAAU,IACrC,GAAQ,KAAK,IAAI,EAAiB,GAAI,GAExC,EAAO,EAAQ,EAAO,EAEtB,OAAS,GAAI,EAAY,EAAI,EAAU,KACrC,GAAI,GAAM,GAAK,EAAQ,EAAO,EAAiB,GAC3C,EAAkB,GAAU,EAChC,AAAI,IAAW,GACb,IAAO,KAAK,IAAI,EAAM,CAAC,IAEzB,GAAO,EAAS,GAChB,EAAO,IAAM,GAGjB,MAAO,GAAA,SAAY,EAAQ,EAAG,QAGhC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAiB,EAAQ,eASzB,OAPM,GAAgB,EAAa,EAAS,EAAA,QAAW,GACjD,EAAY,EAAc,MAAM,GAChC,EAAY,EAAc,MAAM,GAChC,EAAM,EAAA,MAAkB,CAAC,EAAW,GAAa,SACjD,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAW,KAAK,SAAS,EAAc,QAEpC,EAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,GAAS,EAAI,EAGb,EAAM,GAAI,cAAa,EAAY,GACzC,EAAI,GAAK,EAAS,GAClB,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,EAAE,EACxC,EAAI,GAAS,EAAI,EAAQ,GAAK,EAAS,EAAS,GAKlD,OAFM,GAAS,GAAA,KAAgB,EAAK,YAC9B,EAAY,EAAI,EACb,EAAW,EAAG,EAAW,EAAY,EAAE,GAC9C,GAAM,GAAI,IAGV,EAAQ,EAAY,GAAY,EAAI,OAEpC,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,IACtC,GAAI,EAAI,EAAI,IACV,EAAQ,EAAY,GAAY,EAChC,QAKR,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAiB,EAAS,UAE1B,GAAM,GAAM,GAAI,cAAa,EAAQ,KAAO,GAC5C,EAAI,KAAK,GAGT,OAFM,GAAa,KAAK,SAAS,EAAQ,QAEhC,EAAQ,EAAG,EAAQ,EAAQ,KAAM,EAAE,EAC1C,AAAI,EAAW,IAAU,GAAK,EAAW,GAAS,GAChD,GAAI,EAAQ,EAAQ,EAAW,IAAU,GAG7C,MAAO,GAAA,SAAY,EAAK,CAAC,EAAQ,KAAM,GAAQ,UAGjD,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,GAAiB,EAAO,qBAExB,GAAM,GAAY,KAAK,SAAS,EAAM,QAChC,EAAa,KAAK,SAAS,EAAO,QACxC,MAAO,IACH,EAAW,EAAY,EAAe,EAAc,IAG1D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,IAAe,OACf,WAAM,MAAA,+DACF,IACR,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAgB9D,OAdM,GAAY,EAAE,MAAM,GACpB,EAAc,EAAE,MAAM,GACtB,EAAa,EAAE,MAAM,GACrB,EAAa,EAAE,MAAM,GAErB,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAU,KAAK,SAAS,EAAE,QAC1B,EACF,GAAI,cAAa,EAAY,EAAe,EAAc,GAE1D,EAAY,EACP,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAI,EAAG,EAAI,EAAc,EAAE,EAGlC,OAFM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACZ,EAAI,EAAG,EAAI,EAAa,EAAE,EAIjC,OAHM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACf,EAAW,GAAU,EAAY,GAAW,EACzC,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAM,EAAI,EACV,EACF,EAAM,EAAc,GAAM,EAAc,GAAM,EAAc,IAChE,EAAO,KAAe,EAAQ,GAKtC,MAAO,GAAA,SACH,EAAQ,CAAC,EAAW,EAAc,EAAa,KAG7C,EAAA,UAAA,oBAAR,SACI,EAAW,EAAW,EACtB,GACF,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAE,MAAO,EAAE,OAC9D,EAAS,EAAA,OAAU,EAAU,GAC7B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GACxD,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GAExD,EAAU,EAAO,OACvB,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,EACpC,EAAQ,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAK3D,QAFM,GAAO,KAAK,WAAW,GACvB,EAAO,KAAK,WAAW,cACpB,GACP,GAAM,GAAM,EAAO,WAAW,GAExB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAEzB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAE/B,EAAQ,GAAK,EAAG,EAAM,GAAS,EAAM,KAX9B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,IAA7B,GAcX,MAAO,GAAO,YAGhB,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,QAAA,aAEA,EAAA,UAAA,eAAA,WACE,MAAO,KAIT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAGtB,EAAA,UAAA,cAAA,SACI,EACA,EACA,EACA,EACA,EACA,GAmBF,OAjBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GACjC,EAAW,EAAM,MAAM,GAEtB,EAAA,EAAA,GAAY,EAAA,EAAA,GACb,EACF,EAAA,OAAU,CAAC,EAAU,EAAY,EAAW,GAAc,WAExD,EAAU,KAAK,SAAS,EAAM,QAC9B,EAAa,KAAK,SAAS,EAAS,QACpC,EAAY,KAAK,SAAS,EAAO,QAEjC,EAAW,EAAO,QAClB,EAAY,EAAO,QAKhB,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAW,EAAI,EACf,EAAK,EAAQ,GACb,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GAExB,EAAe,EAAW,GAChC,GAAI,GAAQ,EACV,SASF,OANM,GAAe,EAAa,EAC7B,GAAK,GAAO,GAAc,GAAM,GAAa,GAC9C,EACE,EACD,EAAY,EAAM,GAAK,GAAO,GAAa,GAAM,GAAY,GAAK,EAE9D,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAgB,EAAa,EAC/B,EAAM,GAAc,GAAK,EAAK,EAC9B,GAAO,GAAK,GAAO,GAAc,GAErC,GAAI,EAAO,GAAK,EAAO,EAAc,GACnC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAGzB,SAGF,GAAI,IAAW,WAKb,OAJM,GAAS,KAAK,MAAM,GACpB,EAAY,KAAK,KAAK,GACtB,GAAQ,EAAO,EAEZ,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAOF,OAJM,IAAU,KAAK,MAAM,IACrB,GAAW,KAAK,KAAK,IACrB,GAAQ,GAAO,GAEZ,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAI,GAAM,EAAI,GAAU,EAAS,GAAK,EAAS,EAAS,GACpD,EAAO,EAAS,GACd,GAAU,EAAU,GAE1B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAS,EAAS,GACjD,EAAO,EAAS,GACpB,GAAM,IAAW,EAAU,GAE3B,EAAM,EAAI,GAAU,EAAS,GAAK,EAAY,EAAS,GACnD,EAAO,EAAS,GACpB,GAAM,IAAa,EAAU,GAE7B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAY,EAAS,GACpD,EAAO,EAAS,GACpB,GAAM,IAAc,EAAU,GAExB,GAAM,GAAW,IAAW,IAAW,GACvC,GAAS,GAAc,IAAc,IAAc,GAEzD,EAAM,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC9D,EAAO,OAAO,GAAO,GAAQ,IAAS,IAAO,QAIjD,QAAS,GAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAKF,OAFM,IAAW,KAAK,MAAM,IACtB,GAAW,KAAK,MAAM,GACnB,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,IAAQ,EAAI,GAAW,EAAS,GAClC,GAAW,EAAS,GAAK,EAAO,EAAS,GACvC,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,IAAU,EAAU,OAM5C,MAAO,GAAO,YAGhB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAiB,GACvB,MAAO,MAAK,QACR,EAAe,EAAc,EAAa,EAAY,EACtD,EAAY,EAAW,EAAS,EAAc,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAE1C,GAAI,IAAc,EAChB,MAAO,GAAA,OAAU,GAAI,EAAa,EAAE,OAOtC,OAJM,GAAS,GAAI,GAAA,aAAa,CAAC,EAAW,GAAY,EAAE,OACpD,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAQ,KAAK,SAAS,EAAE,QAErB,EAAI,EAAG,EAAI,EAAW,KAG7B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,GAAgB,EAAM,EAAQ,GAC9B,EAAM,KAAK,GAEb,GAAI,EAAe,GAAK,GAAgB,EAAE,KAAO,EAC/C,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,EAAE,OAGzD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,EAAO,OAAO,EAAI,EAAY,GAAK,EAAM,EAAe,EAAY,GAGxE,MAAO,GAAO,WAAW,QAAQ,IAGnC,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAe,EAAA,OAAU,GACzB,EAAiB,GACvB,MAAO,MAAK,QACR,EAAS,EAAS,EAAO,EAAY,EAAW,EAAY,EAC5D,EAAS,EAAc,IAG7B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,EAAQ,GAAS,EAAA,KAAK,WAAW,GACjC,GAAM,GACF,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IACrD,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,OAGnD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAEhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,GAAM,GAAS,EAAA,KAAK,kBACD,EAAE,MAAO,EAAA,KAAK,cAAc,EAAE,QACjD,MAAO,MAAK,WAAW,EAAQ,EAAE,MAAO,EAAE,QAG5C,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGxC,EAAA,UAAA,QAAR,SACI,EAAiB,EAAiB,EAAoB,EACtD,EAAmB,EAAoB,EACvC,EAAmB,EACnB,GACF,GAAM,GAAe,CAAC,EAAa,EAAW,GAExC,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAc,KAAK,SAAS,EAAQ,QAE1C,GAAI,IAAe,EACjB,MAAO,GAAA,OAAU,GAAI,EAAO,EAAQ,OAGtC,GAAM,GAAS,GAAI,GAAA,aAAa,EAAc,EAAQ,OACtD,EAAO,OAAO,KAAM,KAAK,SAAS,EAAa,QAAuB,IAEtE,OAAS,GAAI,EAAG,EAAI,EAAY,KAG9B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,EAAM,KAAK,GACX,GAAgB,EAAM,EAAQ,GAGhC,GAAI,EAAe,GAAK,GAAgB,EAAa,EACnD,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,GAGvD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,AAAI,EACF,EAAO,OAAO,EAAe,EAAY,IACrC,EAAY,EAAI,EAAY,GAEhC,EAAO,OAAO,EAAe,EAAY,GAAK,EAAQ,OAAS,EAC3D,EAAY,GACZ,EAAY,EAAI,EAAY,GAItC,MAAO,GAAO,WAAW,QAAQ,IAErC,GAzjFoC,EAAA,eC5DpC,YAqB8B,GAE5B,OADM,GAAe,GAAI,cAAa,EAAK,QAClC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAa,GAAK,KAAK,IAAI,EAAK,IAElC,MAAO,GAGF,GAAM,IACT,SAAC,GACQ,GAAA,GAAA,EAAA,OAAA,EACD,EAAa,EAAK,QACpB,EAAe,GAAI,cAAa,EAAA,KAAK,cAAc,EAAE,QACzD,GAAI,EAAE,QAAU,aACd,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC7C,EAAe,GAAc,OAS7B,QAPM,GAAc,EAAW,KAAK,IAAI,EAAE,QACpC,EAAO,EAAY,mBAAmB,KACtC,EAAO,EAAY,mBAAmB,KACtC,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC/B,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAO,EAAS,GAChB,EAAO,EAAS,GACtB,EAAa,GAAK,KAAK,MAAM,EAAM,GAGvC,MAAO,GAAW,WAAW,EAAc,EAAE,MAAO,YAG7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICzDd,YAwB6C,GAE3C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAmB,GACzB,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAE3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAEhC,EACF,EAAA,KAAK,uBAAuB,EAA0B,GAEpD,EAAQ,EAAO,OACf,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAE7D,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAG1D,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAE5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,KAX7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,IAA5B,GAeX,MAAO,CAAC,EAAQ,IClEpB,YAqBwB,GAEf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,KAAM,EAAA,EAAA,KAEP,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACzC,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAEzC,EAAc,EAAQ,eAAe,EAAK,MAAO,aAEjD,EAAU,EAAQ,KAAK,IAAI,EAAY,QAK7C,SAAQ,mBAAqB,CAC3B,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,GACpD,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,IAG/C,EAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC/Cd,YAsBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IClCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAGP,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAIvC,GAAM,GAAc,EAAA,MAAS,EAAE,OACzB,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,EAAE,MAAO,CAAC,MAAO,aAEpD,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAQ,KAAM,GAAc,QAAO,IAE/D,SAAY,UACZ,EAAQ,8BAA8B,GAE/B,EAIT,GAAI,EAAE,QAAU,aACd,GAAM,GAAW,GAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,IAC5C,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAO,EAAE,MAAO,CAAC,MAAK,KAElE,SAAQ,8BAA8B,GAE/B,EAGT,GAAI,CAAC,EAAA,KAAK,gBAAgB,EAAE,MAAO,IAGjC,GAAM,GAAS,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAC7C,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAO,MAAO,MAAK,GAG3D,GAAI,IAAU,SACZ,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAe,WAAW,KAAK,GACrC,MAAO,GAAQ,eAAe,EAAE,MAAO,QAAS,GAGlD,GAAI,IAAU,QAIZ,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAO,EAAA,KAAK,aAAa,CAAC,GAAI,EAAE,OAEhC,EAAA,GAAA,SAAA,EAAA,GAAA,MAAA,KAAA,EAAA,EAAA,IAAA,EAAA,MAAA,GAAA,EAAA,EAAA,QAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAQ,eAAe,EAAa,OAAQ,GAGrD,KAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,GAG1D,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC9Fd,YAsCI,EAAc,EACd,EAAuC,GACzC,MAAI,IAAe,KACV,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAiB,CAAC,EAAG,GAAI,GAEzB,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,IAInD,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,aACzC,GAAM,GAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGjC,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAY,KAAM,GAAa,QAAS,IAE5D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,OAEP,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,iBASd,GAE5C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAyB,EACzB,GACN,GAAM,GAAc,EAAA,aAAa,2BAA2B,EAAQ,GAC9D,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAY,OACzB,EAAgB,EAAA,KAAK,eAAe,GAEpC,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GACxD,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GAExD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAEvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GACvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GAEvD,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAO,EAAI,EAAM,OACjB,EAAO,EAAI,EAAM,OAEjB,EACF,EAAG,EAAM,EAAO,GAAI,EAAM,EAAO,EAAI,GAAI,EAAM,EAAO,GACnD,EAAM,EAAO,EAAI,IAExB,EAAe,GAAK,EAAO,KAC3B,EAAe,GAAK,EAAO,SAG7B,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EACF,EAAG,EAAM,EAAS,GAAI,EAAM,EAAS,EAAI,GAAI,EAAM,EAAS,GACzD,EAAM,EAAS,EAAI,IAE1B,EAAe,GAAK,EAAS,KAC7B,EAAe,GAAK,EAAS,MAhBtB,EAAI,EAAG,EAAI,EAAe,OAAQ,MAAlC,GAmBX,MAAO,CAAC,EAAgB,EAAgB,ICvL5C,AAsBO,GAAM,IAAU,GAA8B,SAAC,EAAG,GAAM,MAAA,GAAI,IACtD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAGpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAwBsC,GAEpC,MAAO,UAAC,EAAQ,EAAO,GAGrB,OAFM,GACF,EAAA,KAAK,uBAAuB,EAA0B,EAAO,QACxD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,IChCX,YAiCI,EAAc,EAA0B,GAC1C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAQlB,OALM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,GAAS,EAAE,MACpB,EAAY,EAAA,KAAK,kBAAkB,EAAQ,GACxC,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,gBAclD,EAAc,EAA4B,GAC5C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAGlB,GAAM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAS,GAAS,EAAE,MACpB,EAAY,EAAU,EAAQ,EAAQ,GAC5C,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,IC3EtD,AAsBO,GAAM,IAAW,GAAsB,SAAC,GAAO,MAAA,MAAK,KAAK,KACnD,GAAiB,GAAwB,EAAA,KAAM,IAE/C,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,YAoBI,EAAmB,EAAoB,EACvC,GAIF,OAHM,GAAO,EAAA,KAAK,uBACd,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,GCpCT,AAqBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CACL,KAAM,EAAQ,EAAQ,EAAQ,EAC9B,KAAM,EAAQ,EAAQ,EAAQ,KAIzB,GACT,GAAiB,EAAA,SAAU,GAAc,IAEhC,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICrCd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,GAAI,KAAK,KAAK,KACxD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,YAuBI,EAAkB,EAAiB,EAAgB,EACnD,GACF,GAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,GACF,GAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAIhD,OADM,GAAU,EAAA,KAAK,uBAAuB,EAA0B,GAC7D,EAAI,EAAG,EAAI,EAAQ,EAAE,GAC5B,GAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,SAAC,EAAa,GAAM,MAAA,GAAM,EAAM,KAC/C,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEpB,MAAO,eAIL,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAAO,EAAA,EAAA,KAEd,GAAiB,EAAG,SAEd,GAAA,GAAA,EAAA,WAAA,iBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAA,WAAW,kBAAkB,EAAG,EAAQ,GAExC,GAAM,GAAO,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAClC,EAAU,GAAU,EAAM,EAAQ,EAAO,EAAE,MAAO,EAAE,OAC1D,MAAO,GAAQ,eAAe,EAAO,EAAE,MAAO,GAGzC,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IClEd,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAEpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAqBI,EAAmB,EAAkB,EAAiB,EACtD,GASF,OARM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GAEjC,EAAS,EAAA,KAAK,uBAChB,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAO,EAAE,GAK3B,OAJM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAGhC,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,EAAK,IAGvB,GAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE3B,MAAO,GC3CT,YAoBI,EAAuB,EAAc,EAAiB,GAgExD,OA1DM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAyDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GACtB,EAAI,EAAG,EAAI,EAAO,IACzB,EAAS,IAAM,EAAM,GAEvB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACxC,EAAS,IAAM,EAAM,GAevB,OAVM,GAA0C,GAG1C,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAA0B,GAC1B,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EAC/C,EAAI,EAAG,EAAI,EAAM,GAAQ,KAEhC,GAAI,GAAO,OACX,GAAI,EAEF,EAAU,EAAO,GAAG,gBAGpB,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG1C,EAAU,EAAW,KAAK,KAI5B,GAAI,EAAe,KAAa,OAC9B,EAAQ,GAAK,EAAe,QAE5B,GAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAOvB,GAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,GAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,SAAC,EAAoB,GACzC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAOxE,GAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAE7B,CACL,aAAc,EAAa,OAC3B,YAAW,EACX,QAAO,GCzJX,+MCAA,GAGM,IAAU,QCHhB,AA8BA,EAAA,gBAAgB,MAAO,WAAM,MAAA,IAAI,KAAkB,GC9BnD,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,YAoBI,EAAqB,EAAkB,EAAiB,EACxD,EACA,GAsBF,OArBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAS,EAAA,OAAO,EAAS,SAAU,GACnC,EAAa,EAAO,OAEpB,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAoB,EAAI,EACxB,EAAmB,EAAI,EAAQ,GAC5B,EAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAM1C,OALM,GAAW,EAAK,EAAe,EAC/B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,EAAkB,EAAoB,EAAK,EACxC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAQzC,OAPM,GAAW,EAAK,EAAc,EAC9B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAc,EACd,EAAW,EACX,EAAQ,EACH,EAAK,EAAO,EAAK,EAAO,GAAM,GAErC,OADM,GAAW,EAAmB,EAAK,EAAQ,GACxC,GAAK,EAAO,GAAK,EAAO,IAAM,GACrC,GAAM,IAAW,EAAW,GAAK,EAAQ,GACnC,GAAQ,EAAQ,GAAW,GACjC,AAAK,IAAa,OAAS,GAAQ,EACjC,EAAc,GACT,AAAI,IAAa,OACtB,IAAY,GACZ,KAGJ,GAAI,MAAM,GACR,MAGJ,GAAM,IAAe,EAAkB,EAAK,EAAmB,EAC/D,EAAW,IACP,IAAa,MAAQ,EAAW,EAAQ,EAKpD,MAAO,eAIL,EAAqB,EAAkB,EACvC,EAAmC,EACnC,GADmC,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,IAYF,OAXM,GAAe,EAAA,OAAO,EAAS,SAAU,SACzC,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,EAAA,OAAO,EAAQ,EAAO,GAC1B,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,GAG1C,OAFM,GAAW,EAAK,EAAe,EACjC,EAAQ,EACL,EAAQ,GACb,GAAS,EAKX,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAGzC,OAFM,GAAW,EAAK,EAAc,EAChC,EAAQ,EACL,EAAQ,GACb,GAAS,EAOX,OALM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAK,EAAO,EAAK,EAAO,GAAM,EAErC,OADM,GAAK,EAAK,EACP,EAAK,EAAO,EAAK,EAAO,GAAM,GACrC,GAAM,GAAK,EAAK,EACV,EAAQ,EAAK,IAAI,EAAG,EAAI,EAAI,GAClC,AAAI,EAAQ,GACV,GAAW,EACX,AAAI,EACF,EAAc,EACR,IAAI,EAAS,SAAW,GAAM,EAAS,QAAU,GAC3C,EAAS,WACb,EACH,GAAK,EAAS,QAAU,GAAM,EAAS,WAAa,EAEzD,EAAc,EAAK,EAAuB,GAKlD,EAAa,IAAI,EAAa,EAAG,EAAI,EAAI,IAKjD,MAAO,GCrJT,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAqBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBAyB9B,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAgB,EAAK,GAAe,GAEpC,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAGF,GAAM,GAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAGf,EAAG,IAAI,EAAU,EAAe,EAAG,EAAK,EAAK,GAKrD,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,IC1Fd,YAsBoC,GAK3B,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EAAG,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,KAAM,EAAA,EAAA,SAE/B,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGV,GAAiB,CAAC,EAAG,EAAM,EAAU,EAAO,GAAS,aAEhD,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAsBpB,OAnBM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACtC,EAAU,EAAQ,KAAK,IAAI,EAAS,QAAQ,OAC5C,EAAQ,EAAQ,EAAQ,KAAK,IAAI,EAAM,QAAQ,OAC/B,GAAI,cAAa,CAAC,IAClC,EAAU,EACZ,EAAQ,KAAK,IAAI,EAAO,QAAQ,OAChC,GAAI,cAAa,CAAC,IAChB,EAAU,GAAI,cAAa,EAAM,QAEjC,EAAgB,EAAQ,OACxB,EAAc,EAAM,OACpB,EAAgB,EAAQ,OACxB,EAAc,EAAM,OAEtB,EAAO,EACP,EAAK,EACL,EAAK,EACL,EAAK,EACA,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,EAAQ,GAAK,EAAQ,KAChB,GAAM,GAAK,EAAM,MAAS,EAAM,KAC7B,KAAK,KAAK,EAAQ,KAAQ,GAClC,AAAI,GAAQ,GACV,GAAO,GAET,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAGT,MAAO,GAAQ,eAAe,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAM,IAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,IC5Fd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,YAAa,SAAC,EAAI,GAC9D,GAAM,GAAY,EAClB,MAAI,GAAK,EAAU,aACV,EAAU,aAEZ,EAAK,EAAU,aAAe,EAAU,aAAe,IAGnD,GAA2B,CACtC,WAAY,EAAA,YACZ,YAAa,MACb,WAAY,IChCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YAsBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAED,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,EAAQ,OAAO,EAAE,QAEjB,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAEjC,GAAI,EAAM,oBAAsB,MAC9B,GAAM,GAAO,EAAM,mBAAmB,KAChC,EAAO,EAAM,mBAAmB,KAEtC,EAAK,MAAQ,EACb,EAAK,MAAQ,EAGf,MAAO,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,OAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,ICzDd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,KAED,EAAQ,EAAA,KAAK,eAAe,EAAM,EAAO,GAAG,OAAO,GACrD,EAAW,EAAA,aAAa,gBAAgB,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtE,GAAI,EAAA,KAAK,cAAc,KAAc,EACnC,MAAO,GAAQ,eAAe,EAAU,EAAO,GAAG,MAAO,IAI3D,GAAM,GAAU,EAAO,OAAO,SAAA,GAAK,MAAA,GAAA,KAAK,cAAc,EAAE,OAAS,IACjE,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAM,GAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAGlC,GAFA,EAAA,aAAa,uBAAuB,EAAQ,GAExC,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAC5D,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAE5D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAC3D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAE3D,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAO,IAEtE,SAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAUT,GAAM,GAAW,EAAQ,IAAI,SAAA,GAC3B,GAAM,GAAY,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAC7C,EAAQ,CAAC,GAAI,GACnB,MAAO,IAAQ,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAO,EAAE,MAAO,CAAC,MAAK,OAIxD,EACI,EAAA,aAAa,gBAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAE7D,GAAM,GAAU,EAAA,KAAK,uBACjB,EAAQ,GAAG,MAAoB,EAAA,KAAK,cAAc,IAEtD,GAAI,EAAS,GAAG,MAAM,KAAO,GAE3B,GAAI,GAAS,EACb,EAAS,QAAQ,SAAA,GACf,GAAM,GAAM,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACjC,EAAO,EAAA,KAAK,cAAc,EAAE,OAElC,EAAQ,IAAI,EAAK,GACjB,GAAU,SAGZ,GAAI,GAAY,EAEhB,EAAS,QAAQ,SAAA,GAKf,OAJM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAErC,EAAO,EAEF,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EAEpC,OADM,GAAS,EAAM,EAAS,GAAK,EAC1B,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EACpC,EAAQ,EAAS,GAAO,EAAM,KAIlC,GAAa,EAAE,MAAM,KAIzB,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtD,EACF,EAAQ,eAAe,EAAe,EAAO,GAAG,MAAO,GAE3D,SAAS,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KAErD,EAGF,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,ICjId,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,SAAC,GAwCX,OAxCY,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OACJ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAE,MAAM,OAEhB,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAChD,EAAa,EAAO,MAAM,OAE1B,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOI,EAAU,EAAA,KAAK,cAAc,GAC7B,EAAU,EAAS,OACnB,EAAa,EAAA,KAAK,kBAAkB,EAAE,MAAO,GAM1C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,IAAO,EAAO,EAAc,EAAQ,KACjC,GAAI,EAAG,GAAI,EAAY,EAAE,IAEhC,OADI,IAAS,OAAO,iBACX,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,GAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAS,EAAA,KAAK,WAChB,CAAC,EAAG,GAAK,GAAK,IAAI,EAAO,EAAA,KAAK,eAAe,EAAE,QAC7C,GAAc,EAAA,KAAK,WACrB,CAAC,GAAG,GAAG,IAAI,EACX,EAAA,KAAK,eAAe,EAAO,QACzB,GAAM,EAAM,IAAU,EAAW,IACvC,AAAI,GAAM,IACR,IAAS,MAMnB,GAAM,IAAc,EAAA,KAAK,WACrB,CAAC,EAAG,EAAM,EAAM,IAAI,EAAS,EAAA,KAAK,eAAe,IACrD,EAAW,IAAe,GAMlC,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAY,EAAE,OAAQ,EAAU,EAAE,OAExD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAU,MAAO,EAAE,SCrG9C,AAsBO,GAAM,IAA+C,CAC1D,WAAY,EAAA,yBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,yBAAwB,QACtC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EAAY,EAAA,KAAK,0BACD,EAAO,MAAO,EAAO,OAOlC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAO,EACP,GAAO,EACF,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAO,GACP,GAAO,MAMjB,EAAU,IAAM,IAAM,IAAM,EAAI,GAAG,GAAM,GAAM,GAMvD,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAO,MAAO,EAAO,OAEhE,MAAO,CAAC,OAAM,GAAE,MAAO,EAAO,MAAO,MAAO,EAAO,SCrHvD,AAsBO,GAAM,IAA8C,CACzD,WAAY,EAAA,wBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,wBAAuB,QACrC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EACF,EAAA,KAAK,0BAA0B,EAAE,MAAO,EAAE,OAOrC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAU,EAAO,EAAK,EAAI,EAC1B,GAAU,EAAO,EAAK,EAAI,EACrB,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAS,GACT,GAAS,MAMnB,EAAU,GAAG,IAAQ,IAAQ,IAAM,EAAI,GAAG,GAAM,GAAM,GAM9D,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAE,MAAO,EAAE,OAEtD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrH7C,AAsBO,GAAM,IACT,GAA6B,SAAC,EAAW,GAAc,MAAA,GAAI,IAClD,GAAM,GAAiB,EAAA,IAAK,IAE5B,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC7Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,IAAM,EAAI,EAAM,KAAK,IAAI,GAAM,IAEnD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC3Bd,AAqBA,GAAM,IAAI,EAAA,aAAa,MACjB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAEX,GAAgB,GACzB,EAAA,IACA,SAAC,GACC,GAAM,GAAO,KAAK,KAAK,GACjB,EAAI,KAAK,IAAI,GACb,EAAI,EAAO,GAAM,GAAI,GAC3B,MAAO,GACF,GACK,OAAK,EAAI,IAAM,EAAK,IAAM,EAAI,IAAM,EAAI,IAAM,EAC/C,KAAK,IAAI,CAAC,EAAI,MAIhB,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Cd,YAmCI,EAAmB,EACnB,GAgBF,OAfM,GAAa,EAAM,MACnB,EAAQ,EAAW,GACnB,EAAW,EAAW,GAEtB,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EAAS,EAAU,mBAAmB,KACtC,EAAS,EAAU,mBAAmB,KAGtC,EAAc,CAAC,EAAO,GACtB,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAA,KAAK,uBAAuB,UAAW,GACpD,EAAa,EAAA,KAAK,uBAAuB,UAAW,GAEjD,EAAI,EAAG,EAAI,EAAO,KAmBzB,OAjBM,GAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAE7B,EAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAG7B,EAAQ,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAG,KAAM,GAAI,QAAS,IAGtD,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,KACP,EAAM,EAAA,aAAa,uBAAuB,EAAM,GAE7C,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAI,EAAA,aAAa,oBAAoB,EAAK,GAChD,EAAW,EAAI,EAAW,GAAK,EAAE,KACjC,EAAW,EAAI,EAAW,GAAK,EAAE,KAGnC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAG3C,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,GAChD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,GAAY,QAAS,IAE1D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,cAIL,EAAmB,EACnB,GACF,GAAM,GAAY,EAAA,KAAK,cAAc,EAAM,OAErC,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAG5D,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAGlE,GAAI,GAAc,IAChB,GAAM,GACF,GAAU,EAAU,EAAU,EAAW,EAAS,GAEhD,EAAc,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAEjD,GAAI,GACF,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,EAAO,MACvD,EACF,EAAW,eAAe,EAAa,UAAW,EAAO,MAEvD,EAAuB,EAAW,eACpC,GAAI,UACJ,EAAA,KAAK,kBAAkB,EAA8B,YACnD,EACF,GAAS,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAS,IAExC,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAW,QAAS,IAEhD,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAe,QAAS,IAGpD,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OACtC,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OAE5C,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,CAAC,KAAM,EAAa,KAAM,GAGnC,MAAO,QAEP,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EACF,GAAyB,EAAM,EAAW,GAE9C,MAAO,GAAA,aAAa,uBAAuB,IAI/C,YAAuB,GACrB,MAAQ,GAAO,EAAO,KAAO,EAI/B,YACI,EAAwB,EAAwB,EAChD,EACA,GACF,GAAI,IAAS,EACX,MAAO,CAAC,KAAM,EAAU,KAAM,GAGhC,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EAAO,EAAO,EAEd,EAAc,EAAA,aAAa,qBAAqB,GAEhD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAa,EAAA,aAAa,oBAAoB,GAE9C,EAAc,EAAW,KACzB,EAAc,EAAW,KAEzB,EAAW,CAAC,EAAY,QAExB,EACF,EAAW,eAAe,EAAU,UAAW,GAC7C,EACF,EAAW,eAAe,EAAU,UAAW,GAE7C,EAAgB,GAClB,CAAC,OAAQ,CAAC,KAAM,EAAa,KAAM,GAAc,QAAS,IAGxD,EACF,GAAU,EAAc,EAAc,EAAM,EAAS,GAEnD,EAAgB,EAAa,KAC7B,EAAgB,EAAa,KAE7B,EAAa,CAAC,EAAc,QAE5B,EACF,EAAW,eAAe,EAAY,UAAW,GAC/C,EACF,EAAW,eAAe,EAAY,UAAW,GAE/C,EAAkB,GAAQ,CAC9B,OAAQ,CAAC,KAAM,EAAe,KAAM,GACpC,QAAS,IAGL,EACF,GAAU,EAAa,EAAa,EAAM,EAAS,GAEjD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAI,EAAA,aAAa,UAAU,EAAM,GACjC,EAAS,CAAC,EAAE,KAAK,QAEjB,EAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAC3D,GAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAE3D,GAAc,GAChB,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,IAAY,QAAS,IAEpD,GACF,GACI,CAAC,OAAQ,CAAC,EAAG,GAAa,EAAG,GAAiB,QAAS,IAGzD,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAErB,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAGrB,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAEV,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAGV,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAC9C,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAEpD,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IAElC,CAAC,KAAM,GAAW,KAAM,IAIjC,YACI,EAAkB,EAAc,GAGlC,OAFM,GAAM,GAAI,cAAa,EAAO,GAE3B,EAAI,EAAG,EAAI,EAAM,KAGxB,OAFI,GAAO,EACP,EAAO,EACF,EAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAI,EAAA,aAAa,SAAS,EAAI,EAAG,EAAM,GACvC,EAAO,EAAA,aAAa,oBAAoB,EAAsB,GACpE,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAC3C,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAE7C,AAAI,GACF,IAAQ,EACR,GAAQ,GAEV,EAAA,aAAa,mBAAmB,EAAK,EAAM,EAAM,GAEnD,MAAO,GCjVT,YAuBoB,GAEX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAO,GAElC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICtDd,AAsBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,MACb,WAAY,SAAC,GAUX,OAVY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GAEX,EAAS,KAAK,MAAM,EAAa,GACjC,EAAS,EAAc,EAAY,EAAY,EAEjD,EAAc,EAAU,GAE5B,GAAI,GAAU,GAAK,EAAS,GAE1B,GAAM,GAAmB,EAAS,EAC5B,EACF,EAAc,EAAY,EAAmB,EACjD,EAAc,EAAU,GAE1B,EAAO,GAAU,EAMzB,GAAM,GAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,EAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SCpErD,YAuBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAM,GAEjC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtDd,AAqBO,GAAM,IACT,GAAgB,EAAA,SAAU,SAAC,GAAO,MAAA,QAAO,SAAS,GAAM,EAAI,GAAG,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,IAAI,KAAQ,SAAW,EAAI,GAAG,QAEzD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,QAAO,MAAM,GAAM,EAAI,GAAG,QAEhD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAK,EAAI,GAAG,QAEvC,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC3Bd,AA2BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAa,EACf,EAAS,EAAE,MACT,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,EAAkB,GACnD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACvD,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC1C,GAAI,GAAgB,MAElB,OADM,GAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAO,EAAa,IAGpC,EAAQ,GAAc,EAAO,EAAQ,EAAE,MAAO,EAAc,GAC5D,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAElD,EAAS,EAGX,GAAiB,EAAG,OACpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGd,EAAa,EAAA,KAAK,cAAc,GAEhC,EAAS,GAAQ,EAAO,EAAY,EAAa,EAAE,OACnD,EAAS,EAAW,MAAM,EAAQ,EAAa,EAAE,OAEnD,EAAW,EACf,GAAI,GAEF,GAAM,GAAW,EAAA,aAAa,qBAAqB,EAAa,GAChE,EAAW,EAGb,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,SCtE9C,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAyBlC,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAY,EAAA,OACd,EAAS,SAAU,EAAE,MACrB,GAAiB,EAAS,EAAE,MAAO,EAAE,MAAO,GAAU,QACpD,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,GAAM,GAAS,EAAwB,EAAuB,EACzD,EAAU,IAAI,EAAG,EAAK,EAAK,GAC1B,GAAS,EAAK,EAAuB,EAErC,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAAQ,IAGvB,EAAG,IAAI,EAAS,EAAG,EAAK,EAAK,GAKrC,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,ICnGd,YAoBI,EAAqB,EAAkB,EACvC,EAA8B,GAChC,GAAM,GAAU,EAAA,KAAK,eAAe,GAC9B,EAAW,GAAK,EAAS,EAAQ,EAAO,EAAS,EAAU,OAC3D,EAAe,GACjB,EAAS,EAAQ,EAAO,EAAU,GAAM,GAE5C,MAAO,CAAC,EAAS,OAAQ,EAAa,QC3BxC,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAa,EACnB,GAAiB,EAAG,qBAEpB,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,CAAC,EAAG,GAAI,GACN,EAAA,GAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGT,EACF,EAAW,MAAM,EAAwB,EAAS,SAAU,EAAE,OAC5D,EACF,EAAW,MAAM,EAAuB,EAAS,SAAU,EAAE,OACjE,MAAO,CACL,CAAC,OAAQ,EAAc,MAAO,EAAS,SAAU,MAAO,EAAE,OAC1D,CAAC,OAAQ,EAAe,MAAO,EAAS,SAAU,MAAO,YC/C/D,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEnB,GAAiB,EAAO,2BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC3C7B,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEnB,GAAiB,EAAO,8BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAG,GAAM,MAAC,KAAM,EAAK,EAAI,IAC/C,GACT,GAAiB,EAAA,SAAU,GAAc,KAAsB,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC9Bd,YAuBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,SAAU,EAAA,EAAA,cAEjB,GAAiB,EAAG,OAEpB,GAAM,GAAW,EAAS,IACtB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAE,MAAM,GAAK,EAAE,KAE9C,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAE5B,EAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAQ,EAAE,MAAM,OAChB,EAAW,EAAA,KAAK,eAAe,EAAE,OAEjC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EACF,EAAA,KAAK,uBAAuB,EAAE,MAA0B,GAE5D,AAAI,IAAkB,GACpB,EAAQ,KAAK,GAGf,OAAS,GAAI,EAAG,EAAI,EAAO,KACzB,GAAM,GAAS,EAAA,KAAK,WAAW,EAAG,EAAO,GACnC,EAAY,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAM,KAC3C,EAAW,EAAA,KAAK,WAAW,EAAW,EAAY,GAExD,EAAQ,GAAY,EAAM,GAG5B,GAAM,GAAQ,EAAQ,MAAM,EAAS,EAAU,EAAE,OAEjD,MAAO,CAAC,OAAQ,EAAO,MAAO,EAAU,MAAO,EAAE,OAG5C,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICnEd,AAqBO,GAAM,IAAuB,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAI,IAE/D,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,MACb,WAAY,SAAC,GAiBX,OAjBY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAmB,IAEnB,EAAY,KAAK,IAAI,GACrB,EAAY,KAAK,IAAI,GACrB,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GACX,EAAI,EAAO,GAGb,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACrD,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACzD,EAAS,KAAK,MAAM,EAAS,GAC7B,EAAS,KAAK,MAAM,EAAS,GAE7B,GAAI,GAAc,EAUlB,GATA,AAAI,MAAO,IAAc,UACvB,CAAI,IAAY,EACd,EAAc,EAEd,EAAc,EAAU,IAKxB,GAAU,GAAK,EAAS,GAAc,GAAU,GAChD,EAAS,GAEX,GAAM,GAAmB,EAAU,GAAa,GAC1C,EAAmB,EAAS,EAC5B,GACF,EAAc,EAAmB,EAAmB,EACxD,EAAc,EAAU,IAG1B,GAAM,IAAS,EAAc,EAAY,EAAY,EACrD,EAAO,IAAU,EAMzB,GAAM,IAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,GAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SC3FrD,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAErD,GAAM,GAAO,KAAK,MAAM,GACxB,MAAI,GAAK,EAAO,GACP,KAAK,MAAM,GACT,EAAK,EAAO,GACd,KAAK,KAAK,GAEb,EAAO,IAAQ,EACV,EAEA,EAAO,IAKP,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICxCd,AAqBA,GAAM,IAAa,EAAA,aAAa,gBAC1B,GAAQ,EAAA,aAAa,WAEd,GAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,IAAM,EACD,GAAQ,EAER,GAAc,MAAK,IAAI,GAAM,KAI3B,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICnCd,AAqBO,GAAM,IACT,GAAgB,EAAA,QAAS,SAAC,GAAO,MAAA,GAAK,GAAI,KAAK,IAAI,CAAC,MAE3C,GAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,GAAK,EACA,GACE,EAAK,EACP,EAEA,IAIE,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IClCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AA0BA,GAAM,IAAU,sBACV,GAAY,KAAK,IAAI,IAAW,EAEzB,GAAqB,GAAgB,EAAA,SAAU,SAAC,GAG3D,GAAM,GAAW,EAAK,CAAC,GAIjB,EAAW,EAAK,GAEhB,EAAO,KAAK,IAAI,GAClB,EAEJ,MAAI,GACF,EAAS,EACJ,AAAI,EACT,EAAS,EAET,EAAS,KAAK,IAAI,EAAM,GAEnB,IAGI,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICtDd,YAwB0B,GAKjB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EACA,EAAA,EAAA,KAEP,GAAiB,EAAG,aAKpB,OAHM,GAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAS,GAAc,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAEvD,EAAS,EAAQ,MAAM,EAAQ,EAAU,EAAE,OACjD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,OAGrC,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,MACb,WAAY,ICpDd,YA0B+B,GAKtB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,WAAY,EAAA,EAAA,SAEnB,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAA,KAAK,cAAc,GAE1B,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAA0B,GAE1B,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,GAAY,WAAW,CACrC,OAAQ,CAAC,EAAC,GACV,QAAO,EACP,MAAO,CAAC,SAAU,EAAkB,cAAe,KAG/C,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EACF,EAAA,aAAa,oBAAoB,EAAQ,MAAO,EAAY,EAAM,IAEhE,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,MAAO,GACrC,EACF,GAAQ,CAAC,OAAQ,EAAe,QAAO,EAAE,MAAO,IAE9C,EAAmC,CAAC,EAAG,GACvC,EACe,CAAC,KAAM,GACtB,EACF,GAAU,CAAC,OAAQ,EAAiB,QAAO,EAAE,MAAO,IAElD,EAAqC,CAAC,EAAG,GACzC,EAAmC,CAAC,MAAO,GAC3C,EAAS,GACX,CAAC,OAAQ,EAAqB,QAAO,EAAE,MAAO,IAElD,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAAqC,CAChD,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,ICvFd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,EACD,EAAa,EACnB,GAAiB,EAAG,UAIpB,OAFM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAY,GAAI,cAAa,EAAO,QACjC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAQ,EAAO,GACrB,EAAU,GAAK,EAAQ,EAEzB,GAAM,GAAS,EAAW,MAAM,EAAW,EAAE,MAAO,EAAE,OACtD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrC7C,AAsBO,GAAM,IAAwB,GAA8B,SAAC,EAAG,GACrE,GAAM,GAAO,EAAI,EACjB,MAAO,GAAO,IAEH,GACT,GAAiB,EAAA,kBAAmB,IAE3B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,IChCd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,EAAI,GACvD,GAAM,GAAY,EAClB,MAAI,OAAM,GACD,IAEA,EAAK,EAAI,EAAI,EAAU,QAIrB,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICjCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,YAyBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAEpB,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,IC5Cd,AA6FA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,0yDCvKjB,AAiBA,GAAM,IAAmD,GAEnD,GAA2C,CAC/C,MAAO,GACP,UAAW,GACX,mBAAoB,GACpB,sBAAuB,GACvB,MAAO,GACP,QAAS,GACT,6BAA8B,gBAI5B,EAAsB,GACxB,GAAS,GAAgB,cAGK,GAC9B,GAAI,CAAE,KAAgB,MACpB,GAAM,GAAS,GAAyB,GACxC,GAAI,IAAW,KACb,GAAS,GAAgB,MAEzB,gBAAQ,IAAI,0CAA2C,GAChD,KAGX,GAAM,GAAK,GAAS,GACpB,MAAI,GAAG,gBACL,OAAO,IAAS,GACT,GAAgB,IAGzB,GAAG,QAAQ,EAAG,YACd,EAAG,QAAQ,EAAG,cACd,EAAG,QAAQ,EAAG,OACd,EAAG,QAAQ,EAAG,QACd,EAAG,QAAQ,EAAG,qBACd,EAAG,QAAQ,EAAG,iBACd,EAAG,OAAO,EAAG,cACb,EAAG,OAAO,EAAG,WACb,EAAG,SAAS,EAAG,MAER,GAAS,IAGlB,YAAsB,GACpB,GAAI,MAAO,kBAAoB,aAAe,IAAiB,EAC7D,MAAO,IAAI,iBAAgB,IAAK,KAC3B,GAAI,MAAO,WAAa,YAC7B,MAAO,UAAS,cAAc,UAE9B,KAAM,IAAI,OAAM,0CAIpB,YAAkC,GAChC,GAAI,IAAiB,GAAK,IAAiB,EACzC,KAAM,IAAI,OAAM,0DAElB,GAAM,GAAS,GAAa,GAM5B,MAJA,GAAO,iBAAiB,mBAAoB,SAAC,GAC3C,EAAG,iBACH,MAAO,IAAS,IACf,IACC,IAAiB,EACX,EAAO,WAAW,QAAS,KAC3B,EAAO,WAAW,qBAAsB,IAG3C,EAAO,WAAW,SAAU,ICxFrC,AAoBA,GAAY,IAAZ,AAAA,UAAY,GAgBV,EAAA,EAAA,MAAA,GAAA,QAkBA,EAAA,EAAA,aAAA,GAAA,iBAlCU,IAAA,IAAa,KAqCzB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,SAAA,GAAA,aAJU,IAAA,IAAY,KAOxB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,yBAAA,GAAA,2BACA,EAAA,EAAA,mBAAA,GAAA,qBACA,EAAA,EAAA,mBAAA,GAAA,uBALU,IAAA,IAAmB,iBAqC3B,EAAc,GAChB,MAAO,CAAC,EAAS,eAIf,EAAoB,GACtB,MAAO,GAAa,cAWW,GAC/B,GAAM,GAAO,EAAA,KAAK,cAAc,GAC1B,EAAe,KAAK,KAAK,EAAO,GACtC,MAAO,GAAA,KAAK,oBAAoB,eA6B9B,EAAc,GAChB,MAAO,CACL,KAAK,IAAI,EAAG,KAAK,KAAK,EAAU,IAAK,KAAK,IAAI,EAAG,KAAK,KAAK,EAAO,iBAKlE,EAAc,GACV,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,GAAI,EAAI,cAsBb,EAA2B,GAE7B,GAAM,GAAQ,EAEV,EACA,EACA,EACA,EACA,EAEA,EACA,EAEA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAsB,EAAM,KAC5B,EAA0B,EAAM,KAChC,EAAgC,EAAM,QACtC,EAA4B,EAAM,QAClC,EAAqB,EAAM,IAC3B,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,EAAM,WAC7B,EAAmB,EAAM,OAEzB,GAAsB,EAAG,KACzB,EAA0B,EAAG,KAC7B,EAAgC,EAAG,KACnC,EAA4B,EAAM,KAClC,EAAqB,EAAG,KACxB,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,GAA6B,KAChD,EAA0B,eAC1B,KACJ,EAAmB,EAAG,OAExB,EAAwB,EAAG,KAEpB,CACL,oBAAmB,EACnB,wBAAuB,EACvB,8BAA6B,EAC7B,0BAAyB,EACzB,mBAAkB,EAClB,sBAAqB,EACrB,0BAAyB,EACzB,mBAAkB,EAClB,qBAAoB,EACpB,iBAAgB,GCzOpB,YAsBgC,EAA2B,GACzD,GAAM,GAAc,IACpB,MAAI,GAAA,MAAM,QAAQ,UAChB,GAAgB,GAEX,EAGT,YAAyB,GACvB,GAAM,GAAQ,EAAG,WACjB,GAAI,IAAU,EAAG,SACf,KAAM,IAAI,OAAM,gBAAkB,GAAqB,EAAI,IAK/D,GAAM,IAAc,QACd,GAAc,kBAEa,GAC/B,MAAI,KAAA,MAAM,QAAQ,iCAAmC,IAAQ,GACxD,GAAc,KAAK,IAAI,IAAQ,KAAK,IAAI,GAAO,gBAOlD,EAA2B,GAC7B,OAAQ,OACD,GAAG,SACN,MAAO,eACJ,GAAG,aACN,MAAO,mBACJ,GAAG,cACN,MAAO,oBACJ,GAAG,kBACN,MAAO,wBACJ,GAAG,8BACN,MAAO,oCACJ,GAAG,cACN,MAAO,oBACJ,GAAG,mBACN,MAAO,6BAEP,MAAO,sBAAsB,eAK/B,EAA2B,GAC7B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,aAAa,IAC1B,cAAgB,EAAgB,gDAIlC,EAA2B,GAC7B,GAAM,GAA4B,GAC9B,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,gBAC7B,wCAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAc,KACrD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAc,EAAG,kBAAoB,GAC7D,cAAQ,IAAI,EAAG,iBAAiB,IAC1B,GAAI,OAAM,oCAElB,MAAO,eAIL,EAA2B,GAC7B,GAAM,GAA8B,GAChC,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,kBAC7B,0CAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAgB,KACvD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAgB,EAAG,kBAAoB,GAC/D,SACI,EAAsB,EAAG,iBAAiB,IACxC,GAAI,OAAM,sCAElB,MAAO,GAGT,GAAM,IAAkB,2BACxB,YACI,EAAsB,GACxB,GAAM,GAAwB,GAAgB,KAAK,GACnD,GAAI,GAAyB,MAC3B,QAAQ,IAAI,wCAAwC,GACpD,QAAQ,IAAI,GACZ,OAWF,OARM,GAAa,CAAC,EAAsB,GAEpC,EAAc,EAAa,MAAM;GACjC,EAAM,EAAY,OAAO,WAAW,OAAS,EAC7C,EAAuB,EAAY,IACrC,SAAC,EAAM,GACH,MAAA,GAAA,KAAK,SAAU,GAAa,GAAG,WAAY,GAAO,IACtD,EAAgB,EACX,EAAI,EAAG,EAAI,EAAqB,OAAQ,IAC/C,EAAgB,KAAK,IAAI,EAAqB,GAAG,OAAQ,GAG3D,GAAM,GAAmB,EAAqB,MAAM,EAAG,EAAa,GAC9D,EAAY,EAAqB,MAAM,EAAa,EAAG,GACvD,EAAkB,EAAqB,MAAM,GAEnD,QAAQ,IAAI,EAAiB,KAAK;IAClC,QAAQ,IAAI,EAAc,MAAM;GAAM,IACtC,QAAQ,IACJ,MAAM,EAAA,KAAK,SAAS,EAAU,GAAI,GAClC,iEACJ,QAAQ,IAAI,EAAgB,KAAK;gBAGL,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGR,EAA2B,GAErD,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,KAClC,EAAG,oBAAoB,EAAS,EAAG,eAAiB,GACtD,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,2DAKhB,EAA2B,GAE7B,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,KACtC,EAAG,oBAAoB,EAAS,EAAG,mBAAqB,GAC1D,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,iDAKhB,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,EAAM,EAAG,eACxD,cAIL,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,KAC9D,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,EAAM,EAAG,eACvD,gBAIP,MAAI,GAAA,MAAM,UAAU,mBAAqB,EAChC,EAEF,cAGqB,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGA,EAAe,GACjD,GAAM,GAAiB,EAAA,MAAM,UAAU,0BACvC,GAAK,GAAS,GAAO,GAAU,GAC7B,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IACrC,KAAM,IAAI,OAAM,0BAA4B,EAAY,gBAE1D,GAAK,EAAQ,GAAoB,EAAS,GACxC,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IAC/B,EAAM,IAAI,EAAc,IAAI,EAAc,IAChD,KAAM,IAAI,OACN,0BAA4B,EAC5B,qDAAuD,EAAM,kBAInC,GAChC,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,qBAAqB,kDAIpC,EAA2B,EAAuB,EAClD,EAAqB,EAA6B,EAClD,GACF,GAAM,GAAM,EAAG,kBAAkB,EAAS,GAC1C,MAAI,KAAQ,GAGH,GAET,IAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GACI,EACA,WAAM,MAAA,GAAG,oBACL,EAAK,EAAqB,EAAG,MAAO,GAAO,EAC3C,KACR,GAAa,EAAI,WAAM,MAAA,GAAG,wBAAwB,KAC3C,gBAIL,EAA2B,EAAuB,GACpD,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,iBAInD,EAA2B,GAC7B,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAInD,EAA2B,EAC3B,GACF,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,mBAAmB,EAAS,IACzC,YAAc,EAAc,yCAI9B,EAA2B,EAC3B,GACF,MAAO,GAAG,mBAAmB,EAAS,eAIpC,EAA2B,EAC3B,EAA8C,GAChD,GAAa,EAAI,WAAM,MAAA,IAAgB,EAAI,EAAS,KACpD,GAAa,EAAI,WAAM,MAAA,GAAG,UAAU,EAAwB,iBAGtB,GACtC,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QAC1D,GAAa,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,UACpE,GAAa,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,sBAIjE,EAA2B,EAC3B,GACF,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,iBAIpE,EAA2B,GAC7B,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,KAAM,iBAGjC,GAClC,GAAM,GAAS,EAAG,uBAAuB,EAAG,aAC5C,GAAI,IAAW,EAAG,qBAChB,KAAM,IAAI,OACN,8BAAgC,GAA2B,EAAI,gBAKnE,EAA2B,GAC7B,OAAQ,OACD,GAAG,kCACN,MAAO,wCACJ,GAAG,0CACN,MAAO,gDACJ,GAAG,kCACN,MAAO,wCACJ,GAAG,wBACN,MAAO,kCAEP,MAAO,iBAAiB,GAI9B,YACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAa,EAAI,WAAM,MAAA,OAC/C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,GAElB,MAAO,GAGT,YAA6B,EAA2B,GACtD,GAAM,GAAiB,EAAG,iCAAmC,EACvD,EAAgB,EAAc,EAAG,SACvC,GAAI,EAAgB,EAAG,UAAY,EAAgB,GACjD,GAAM,GAAmB,2BAA2B,EAAc,IAClE,KAAM,IAAI,OAAM,0BAA0B,EAAgB,kBAIlC,EAAiB,GAAA,MAAA,KAAA,QAAA,GAAA,GACpC,EAAA,KAAK,cAAc,EAAM,MAAM,EAAG,EAAM,OAAS,gBAG9B,GAC1B,GAAI,EAAM,SAAW,EACnB,KAAM,OAAM,wDAGd,MAAO,CACL,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EAAG,EAAM,EAAM,OAAS,gBAI5C,GAC3B,GAAI,GAAsC,CAAC,EAAG,EAAG,GAC3C,EAAW,EAAM,SAAW,GAAM,EAAM,SAAW,GAAK,EAAM,KAAO,EAC3E,MAAK,IACH,GACI,CAAC,GAAY,IAAM,OAAK,GAAY,KAEnC,cAIL,EAAoB,SAAA,AAAA,IAAA,QAAA,GAAA,IACtB,GAAI,GAAa,EAAA,MAAM,UAAU,0BAsBjC,GArBA,AAAI,GACF,GAAa,EAAa,EAO1B,EAAW,EAAS,IAChB,SAAC,EAAG,GAAM,MAAA,IAAK,EAAS,OAAS,EAC7B,EAAA,KAAK,kBAAkB,EAAS,IAChC,EAAS,KAIjB,AAAI,EAAS,SAAW,GACtB,GAAW,CAAC,EAAG,EAAS,MAKxB,EAAS,SAAW,GACtB,GAAM,GAAgB,EAAA,KAAK,aAAa,GACxC,EAAW,EAAc,SAG3B,GAAI,GAAO,EAAA,KAAK,cAAc,GAC9B,GAAI,EAAS,QAAU,GAAK,GAAQ,EAClC,MAAO,CAAC,EAAG,GACN,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,IAAM,EACjB,MAAO,GACF,GACH,EAAS,SAAW,GAAK,EAAS,GAAK,EAAS,IAAM,GACtD,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAI,EAAS,IACvC,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,IAAM,EAC/B,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,IACvC,GACH,EAAS,SAAW,GACpB,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,GAC3C,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAK,EAAS,GAAI,EAAS,IACrD,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,EAC7C,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,GAAK,EAAS,IAE1D,GAAI,GAOF,GAAM,GAAW,GAAY,GACzB,EAAO,EAAG,EAAO,EACrB,MAAI,GAAS,QACX,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,IAET,EAAO,EAAY,GAAO,GAAM,GAAO,GAChC,EAAA,KAAK,oBAAoB,GAAM,IAAI,SAAA,GAAK,MAAA,GAAI,IAErD,MAAO,GAAA,KAAK,oBAAoB,GAIpC,YAAgB,GACd,MAAO,GAAI,IAAM,cAOW,EAAkB,GAI9C,GAHA,EAAS,EAAO,MAAM,IACtB,EAAS,EAAO,MAAM,IAElB,EAAA,KAAK,YAAY,EAAQ,GAC3B,MAAO,GAGT,GAAI,CAAC,EAAO,QAAU,CAAC,EAAO,OAC5B,MAAO,GAGT,GAAI,EAAO,KAAO,GAAK,EAAO,KAAO,GAAK,EAAO,KAAO,GACpD,EAAO,KAAO,EAChB,MAAO,GAGT,GAAI,EAAO,SAAW,EAAO,QAC3B,GAAM,GAAa,EAAO,MAAM,IAAI,GAC9B,EAAa,EAAO,MAAM,IAAI,GACpC,GAAI,IAAe,EACjB,MAAO,GAGT,GAAI,GAAO,IAAe,GAAO,IAC5B,GAAO,KAAO,GAAK,EAAO,KAAO,GACpC,MAAO,GAGX,MAAO,GAAO,KAAO,EAAO,IAAM,GAAO,EAAO,KAAO,GAAO,EAAO,IAMvE,GAAI,IACA,eAEmC,GACrC,GAAI,IAAoB,MACtB,GAAM,GAAK,GAAgB,GAC3B,GAAmB,EAAG,aAAa,EAAG,kBAExC,MAAO,kBAIP,GAAmB,mBAGnB,GAAyB,iBAGY,GACrC,GAAI,IAA0B,MAC5B,GAAM,GAAK,GAAgB,GAC3B,GAAyB,EAAG,aAAa,EAAG,yBAG9C,MAAO,MAAK,IAAI,GAAI,gBAG4B,GAEhD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAI,GACE,EAAK,GAAgB,GAE3B,MAAI,IAAa,EAAI,oCACjB,IAAiB,EACnB,EAAoB,EACf,AAAI,GAAa,EAAI,4BAC1B,EAAoB,EAEpB,EAAoB,EAEf,cAGoB,EAA2B,GACtD,GAAM,GAAM,EAAG,aAAa,GAC5B,MAAO,IAAO,iBAGsB,GACpC,IACE,GAAM,GAAK,GAAgB,GAC3B,GAAI,GAAM,KACR,MAAO,SAEF,GACP,eAAQ,IAAI,qCAAsC,GAC3C,GAET,MAAO,eAG0C,GAEjD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,WAGL,CAAC,GAAa,EAAI,0BACpB,MAAO,GAIX,GAAM,GAAwB,GAAuC,GACrE,MAAO,eAYqC,GAC5C,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,GAET,GAAI,CAAC,GAAa,EAAI,4BACpB,MAAO,QAGT,GAAI,GAAa,EAAI,0BACnB,MAAO,IAAuC,GAGhD,GAAM,GAA0B,8BAChC,GAAI,GAAa,EAAI,IACnB,GAAM,GACF,EAAG,aAAa,GACpB,MAAO,IACH,EAAI,GAGV,MAAO,GAGT,GAAM,GAAwB,GAAuC,GACrE,MAAO,GAGT,YAAgD,GAE9C,GAAM,GAAY,GAAiB,GAE7B,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,oBAAqB,EAAO,EAAQ,EAChE,EAAU,mBAAoB,EAAU,iBAAkB,MAE9D,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,EAGT,YAEI,EAA2B,GAC7B,GAAM,GAAY,GAAiB,EAAI,GACjC,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,wBAAyB,EAAO,EAAQ,EACpE,EAAU,mBAAoB,EAAU,qBAAsB,MAElE,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,cAG2B,GAClC,GAAI,IAAiB,EACnB,MAAO,GAET,GAAM,GAAK,GAAgB,GAGrB,EAAa,EAAW,WAAa,KAC3C,MAAO,eAIL,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MAAG,GAAM,smCC7qBzB,AAqBA,GAAM,IAAM,EAAA,MASZ,GAAI,aAAa,YAAa,WAAM,MAAA,IAAI,UAAU,iBAAmB,IAGrE,GAAI,aAAa,gBAAiB,WAChC,MAAI,IAAsB,GACjB,EACE,GAAsB,GACxB,EAEF,IAIT,GAAI,aAAa,iCAAkC,WAAM,MAAA,KAEzD,GAAI,aACA,yBAA0B,WAAM,MAAA,IAAI,IAAI,mBAAqB,IAGjE,GAAI,aAAa,oBAAqB,WAAM,MAAA,KAG5C,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aAAa,aAAc,WAAM,MAAA,IAAI,QAAQ,eAGjD,GAAI,aAAa,2BAA4B,WAAM,MAAA,IAAI,QAAQ,gBAG/D,GAAI,aAAa,kBAAmB,WAAM,MAAA,IAAI,QAAQ,gBAItD,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aACA,+BAAgC,WAAM,MAAA,IAAI,QAAQ,gBAGtD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aAAa,sBAAuB,WAAM,MAAA,IAAI,QAAQ,gBAG1D,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aACA,yBACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAG/C,GAAI,aACA,+BACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAU/C,GAAI,aAAa,+CAAgD,WAC/D,GAAM,GAAe,GAAI,UAAU,iBAEnC,MAAI,KAAiB,EACZ,EAEF,GAAkC,KAO3C,GAAI,aACA,gDACA,WAAM,MAAA,IAAI,UAAU,gDAAkD,GAClE,CAAC,EAAA,YAAY,aAKrB,GAAI,aACA,+BACA,WAAM,MAAA,IAAmC,GAAI,UAAU,oBAM3D,GAAI,aAAa,+BAAgC,WAC/C,MAAO,IAAI,QAAQ,4BACf,GACA,GAAI,QAAQ,kCAOlB,GAAI,aACA,+BACA,WAAM,MAAA,IAA8B,GAAI,UAAU,oBAGtD,GAAI,aACA,0BACA,WAAM,MAAA,IAAoB,GAAI,UAAU,oBAK5C,GAAI,aAAa,4BAA6B,WAK5C,GAAM,GAAc,GAAI,QAAQ,gCAChC,MAAO,GAAc,EAAI,IAU3B,GAAI,aACA,iCACA,WACE,MAAO,IAET,SAAA,GACE,GAAI,EAAY,GAAK,IAAc,GACjC,KAAM,IAAI,OACN,+DACA,mCAAkC,EAAS,QC1LvD,AAiBO,YAAuB,GAC1B,KAAM,GAAe,GAAI,cAAa,EAAK,QAC3C,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC/B,EAAa,GAAK,KAAK,IAAI,EAAK,IAEpC,MAAO,GCtBX,AAoBO,YAAsC,GACzC,MAAO,CAAC,EAAQ,EAAQ,EAAO,EAAO,KAClC,KAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAC3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAQ,EAAO,OACf,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAC7D,GAAI,EAAe,OAAS,EAAe,SAAW,EAClD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAI5D,QAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACjC,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GACrC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GACtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAC5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,IAG5C,MAAO,CAAC,EAAQ,IClDxB,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAG,IAAM,EAAI,GCnBnE,AAoBO,YAA+B,GAClC,MAAO,CAAC,EAAQ,EAAO,KACnB,KAAM,GAAY,EAAA,KAAK,uBAAuB,EAAO,EAAO,QAC5D,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAU,GAAK,EAAG,EAAO,GAAI,GAEjC,MAAO,IC1Bf,AAmBO,KAAM,IAAW,GAAsB,AAAC,GAAO,KAAK,KAAK,ICnBhE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAiBO,YAAiB,EAAO,EAAY,EAAU,GACjD,KAAM,GAAO,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACnE,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAC/B,KAAM,GAAS,EAAI,EACnB,GAAI,GAAM,EAAM,GAChB,OAAS,GAAI,EAAG,EAAI,EAAY,EAAE,GAC9B,KAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACR,GAAM,GAGd,EAAK,GAAK,EAEd,MAAO,GC9BX,AAmBO,KAAM,IAAe,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBvF,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,EAAI,KAAK,KAAK,ICnBrE,AAkBO,YAAmB,EAAM,EAAO,EAAM,EAAO,GAChD,KAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GACrC,GAAI,GACA,KAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAElD,KAAM,GAAU,EAAA,KAAK,uBAAuB,EAAO,GACnD,OAAS,GAAI,EAAG,EAAI,EAAQ,EAAE,GAC1B,KAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,CAAC,EAAK,IAAM,EAAM,EAAM,IACvC,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEtB,MAAO,GCnCX,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBlF,AAiBO,YAAuB,EAAO,EAAQ,EAAO,EAAM,GACtD,KAAM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GACjC,EAAS,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACrE,OAAS,GAAI,EAAG,EAAI,EAAO,EAAE,GACzB,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAEhC,EAAS,GAAI,OAAM,EAAI,QAC7B,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAO,GAAK,EAAI,EAAK,IAEzB,KAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE7B,MAAO,GCjCX,AAiBO,YAAoB,EAAQ,EAAM,EAAO,GAE5C,KAAM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAwDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GAC/B,OAAS,GAAI,EAAG,EAAI,EAAO,IACvB,EAAS,IAAM,EAAM,GAEzB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACtC,EAAS,IAAM,EAAM,GAIzB,KAAM,GAAiB,GAGjB,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAAgB,GAChB,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EACxD,OAAS,GAAI,EAAG,EAAI,EAAM,GAAQ,KAE9B,GAAI,GACJ,GAAI,EAEA,EAAU,EAAO,GAAG,gBAGpB,KAAM,GAAa,GACnB,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG9C,EAAU,EAAW,KAAK,KAG9B,GAAI,EAAe,KAAa,OAC5B,EAAQ,GAAK,EAAe,QAG5B,KAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAM3B,KAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,KAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,CAAC,EAAoB,KACvC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAM9E,KAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAC7B,CACH,aAAc,EAAa,OAC3B,cACA,WC9IR,AAuBE,GAAA,IAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GCpCF,AAmBA,GAAA,IAAA,WAKE,WAAY,EAAuB,GAHnC,KAAA,YAAwB,GAItB,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,UAAU,EAAQ,SAAS,EAAQ,oBAInD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;yBAEZ,EAAS;;;MAKlC,MAAA,MClDA,AAmBA,GAAA,IAAA,WAOE,WAAY,EAAuB,GALnC,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,SAAS,EAAQ,SAAS,EAAQ,oBAIlD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;wBAEb,EAAS;;;MAKjC,MAAA,MCpDA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAqC,EACrC,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAC9B,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,KAAK,YAAc,CAAC,EAAW,GAC/B,GAAM,GAAU,IAAO,MAAS,IAAM,IAChC,EAAe,EACjB,gBACA,+CAEJ,KAAK,SAAW;;;;;kCAKc,EAAU;;;;;8BAKd,EAAU;wBAChB,EAAY;;0BAEV,EAAM;;;;;;;MAShC,MAAA,MC5DA,YAiB+B,EAAc,GAC3C,MAAO,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAG,GAAI,IAAI,gBAG/C,EAAc,GACxC,MAAI,KAAS,EACJ,CAAC,GAEH,GAAe,EAAM,eAGE,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,KAIT,OADI,GAAS,GACJ,EAAI,EAAG,EAAI,EAAM,IACxB,GAAU,EAAK,GACf,AAAI,EAAI,EAAO,GACb,IAAU,KAGd,MAAO,GCxCT,cAgCE,GAAI,GACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAU,kBACV,EAAY,KACZ,EAAY,MACZ,EAAY,KACZ,EAAY,UACZ,EAAS,cACT,EAAe,wBAMf,EAAmB;;;;;;;;;;;MAcnB,EAAmB,GACnB,EAAc;;;;;;;;;OAWd,GAAU,GACV,EAAY,YACZ,EAAY,UACZ,EAAY,UACZ,EAAY,YACZ,EAAS,eACT,EAAe,GAEf,EAAmB;;;;;;;;MASnB,EAAmB;;;;;;;;;MAUnB,EAAc;;;;;;;;OAWT,CACL,QAAO,EACP,UAAS,EACT,UAAS,EACT,UAAS,EACT,UAAS,EACT,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,iBAAgB,EAChB,YAAW,GClIf,YAyBI,EAAkB,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,SACrC,GAAM,GAAU,EAAA,KAAK,eAAe,GACpC,MAAO,GACF,IAAI,SAAC,EAAQ,GACZ,GAAM,GAAQ,OAAO,EAAO,GAAE,MAAM,EAAK,MAAM,EACzC,EAAQ,IAAM,EAAQ,OAAS,EACjC,OAAO,EAAO,EAAI,GAAE,MAAM,EAAK,MAAM,EAAO,GAAE,MAAM,EACpD,YAAY,EAAO,GAAE,MAAM,EAC/B,MAAU,GAAK,KAAK,EAAK,MAE1B,KAAK,gBA+CuB,GACjC,GAAM,GAAU,EAAA,KAAK,eAAe,GAAO,IAAI,SAAA,GAAK,MAAA,GAAE,aAEtD,MAAO;;wBAEe,EAAQ,GAAE,iBAAiB,EAAQ,GAAE;;EAKtD,GAAM,IAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EC5FpC,AAkBO,GAAA,IAAA,EAAA,aAAA,6BAkBH,EAAyB,EAAwB,EACjD,GACF,GAAM,GAA2B,GACjC,EAAW,QAAQ,SAAA,GACjB,GAAM,GAAO,EAAA,KAAK,cAAc,EAAE,UAAU,cAG5C,AAAI,EAAE,UAAU,UACd,EAAe,KACX,iBAAiB,EAAE,KAAO,GAAO,EAAI,IAAI,EAAI,IAAM,IAAE,KAEzD,GAAe,KAAK,qBAAqB,EAAE,KAAI,KAC/C,EAAe,KAAK,qBAAqB,EAAE,KAAI,QAGnD,GAAM,GAAqB,EAAe,KAAK;GAEzC,EACF,EACK,IAAI,SAAA,GAAK,MAAA,IAAwB,EAAG,EAAa,KACjD,KAAK;GACR,EAAc,EAAY,SAC1B,EAAO,KACP,EAA4B,GAA6B,GAC3D,EACA,EACA,EAAe,GAAgB,GAEnC,AAAI,EAAY,SACd,GACI,GAA+B,EAAY,aAAc,GAC7D,EAA+B,GAA8B,IAE7D,GACI,GAAyB,EAAY,aAAc,GACvD,EAA+B,GAA2B,IAG5D,AAAI,GACF,IAAgB,IAGlB,GAAM,GAAS,CACb,EAAc,EAA2B,EACzC,EAAoB,EAAuB,EAAsB,GACjE,KAAK;GACP,MAAO,GAGT,YAA8B,GAC5B,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAiB,OACrB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,WAEpB,KAAM,IAAI,OACH,EAAM,OAAM,2CAKzB,YAAoC,GAClC,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAuB,OAC3B,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,WAE1B,MAAO,IAAmB,IAIhC,YACI,EAAmB,EACnB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAM,GACV,AAAI,EACF,GAAO,GAA2B,GAElC,GAAO,GAAqB,GAG9B,GAAM,GAAU,EAAO,UAAU,aAC3B,EAAW,EAAa,aAC9B,MAAI,GAAQ,QAAU,EAAS,QAC7B,CAAI,EACF,GAAO,GAA+B,EAAQ,GAE9C,GAAO,GAAyB,EAAQ,IAGrC,EAGT,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAwB,EAAsB,OAClD,GACH,MAAO,IAAwB,EAA8B,OAC1D,GACH,MAAO,IACH,EAAsC,WAE1C,MAAO,IAAwB,EAAU,IAI/C,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAkB,EAAsB,OAC5C,GACH,MAAO,IAAkB,EAA8B,OACpD,GACH,MAAO,IACH,EAAsC,OACvC,GACH,MAAO,IACH,EAA8C,OAC/C,GACH,MAAO,IACH,EAAsD,OACvD,GACH,MAAO,IACH,EACA,WAEJ,KAAM,IAAI,OACH,EAAS,OAAM,4CAI5B,YAAsC,GACpC,MAAO;;eAEM,EAAK,UAAS;;IAK7B,YAAoC,GAClC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAuC,GACrC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAyB,GACvB,GAAM,GAAmB,EAAK,QAAO;;;;MAIjC,EAAK,UAAS;MACd,EAAK,aAAY;;;;;;;;;;;;;;;;;;;;;;;MAuBjB,EAAK,iBAAgB;MACrB,EAAK,iBAAgB;MACrB,EAAK,YAAW;;;;;;;;;;;;;;;;;;;;;;;;;MAyBhB,GAAiB;MACjB,GAAiB;MACjB,GAAiB;IAGrB,MAAO,GAGT,GAAM,IAAoB;;;;;;;;;;;;EAcpB,GAAoB;;;;;;;;EAUpB,GAAoB;;;;;;;;;EAWpB,GAAuB;;;;;;;;;;;EAa7B,cACE,MAAO;;;;IAOT,YACI,EAAiB,GACnB,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,MAAI,GAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAKjD,EAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAK9C;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;IAKlD,YACI,EAAiB,GACnB,MAAI,GAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIvC,EAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIpC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;4BACnC,EAAS,GAAE;;IAKvC,YACI,EAAiC,GACnC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAqB,KAAK,KAAK,EAAM,GAAK,GAC1C,EAAgB,EAAqB,KAAK,KAAK,EAAM,GAAK,GAEhE,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;wBAE1B,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAiC,GACnC,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EAAiB,GAWnB,OAVM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEnD,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EACF,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EAAiB,EACjB,EAAU,GACV,EAAS,UAEJ,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAkB,EAAM,EAAM,OAAS,EAAI,GAC3C,EAAU;aACD,EAAC,cAAc,EAAc;kBACxB,EAAC,MAAM,EAAc;MAC/B,EACJ,EAAS,IAAI,EAAC,KAAO,EAGvB,MAAO;UACC,EAAM,OAAM;;oCAEc,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;QAE1C,EAAO;;wBAES,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;mBAE3B,EAAM,OAAM,IAAI,EAAM;;IAKzC,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,MAAO,GAE3B,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,MAAO,GAEjC,MAAO;;kDAEyC,EAAS,GAAE;+BAC9B,EAAS,GAAE;;iCAET,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,KAAM,MAAO,GAEvC,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EAAyB,GAC3B,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,GAAI,EAAA,KAAK,YAAY,EAAO,GAC1B,MAAO;;8CAEmC,EAAe,GAAE,KACvD,EAAe,GAAE;;MAMvB,GAAM,GAAqB,KAAK,KAAK,EAAM,GAAK,GAWhD,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;;iCAE1C,EAAe,GAAE;6BACrB,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAyB,GAC3B,MAAI,GAAA,KAAK,YAAY,EAAO,GACnB;;0CAE+B,EAAS,GAAE,KAAK,EAAS,GAAE;;MAI/D,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKxC,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKrC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;wBACpB,EAAM,GAAE;4BACJ,EAAM,GAAE;;;IAMpC,YAAkC,GAChC,MAAO,SAAS,EAGlB,YAAgC,GAC9B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAO,KACb,MAAO;WACE,EAAQ;eACJ,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAA0B,GACxB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACzE,GAAI,EAAU,UAAU,UACtB,MAAO,SAAS,EAAQ,cAAc,EAAO,KAEzC,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAChB,GAAI,IAAY,GAAK,IAAY,EAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAK9B,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,KAAK,EAAM;6BAC1B,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAe,GAAE,KAAK,EAAe,GAAE;eAClC,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEzE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;UACZ,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAQ,EAAS,GACjB,EAAQ,EAAS,GAEvB,GAAI,IAAU,GAAK,IAAU,EAC3B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAI,KAAU,EACL;cACG,EAAQ;6CACuB,EAAM,cAAc,EAAK;+BACvC,EAAO;;MAIhC,IAAU,EACL;cACG,EAAQ;wCACkB,EAAM,cAAc,EAAK;+BAClC,EAAO;;MAI7B;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,aAAa,EAAM;6BAClC,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAE/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAO,KACb,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,GAC9C,MAAO;aACE,EAAQ;qDACgC,EAAO,OAAO,EAAO;;iBAEzD,EAAK,UAAS,IAAI,EAAO;;MAKxC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAe,KAAK,KAAK,EAAM,GAAK,GAE1C,MAAO;WACE,EAAQ;iCACc,EAAY,KAAK,EAAe,GAAE,KAC7D,EAAe,GAAE;eACR,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAErC,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,IAC9C,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACzB,MAAO;YACC,EAAQ;mDAC+B,EAAO,OAAO,EAAO;6BAC3C,EAAO;;IAK5B,GAAA,GAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,OACvB,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;qDAC+B,EAAM,GAAE;UACnD,GAAkB,GAAU;;MAKpC,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAS,GAAyB,GACxC,MAAI,KAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;4CACtB,EAAO;6BACtB,EAAO;;IAI9B,IAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;uCAC3B,EAAO;6BACjB,EAAO;;IAK3B;UACC,EAAQ;;wBAEM,EAAM,GAAE,YAAY,EAAM;2BACvB,EAAO,KAAK,EAAO;2BACnB,EAAO;;EAKlC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEzD,GAAI,EAAM,KAAO,GACf,GAAM,GAAgB,EAAM,MAAM,GAC5B,EAAW,CAAC,EAAG,GACf,EAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,IAAK,MAAO,OAC5B,MAAO;UACD,GAA2B,GAAa;eACnC,EAAQ;mBACJ,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAM,GAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpC,EAAgB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAO,KAAK,EAAO,KAAK,EAAa,KAAK,EAAY;eACjD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAAK,EAAM,GAC3B,EAAU,EAAM,GAEhB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,SAC9B,MAAO;UACD,GAAqB,GAAa;gBAC5B,EAAQ;mBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO;UAC1C,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAa,EAAU,UAAU,WACvC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;gBACK,EAAQ;;oDAE4B,EAAO;;4BAE/B,EAAO,OAAO,EAAO;iCAChB,EAAO;;QAKtC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;YACC,EAAQ;8CAC0B,EAAM,GAAE;;qDAED,EAAO,OAAO,EAAO;6BAC7C,EAAO;;IAKlC,GAAM,GAAS,GAAyB,GACxC,MAAO;cACK,EAAQ;;4BAEM,EAAO,YAAY,EAAO,cAAc,EAAM;+BAC3C,EAAO,KAAK,EAAO;+BACnB,EAAO;;IAKtC,YAA4B,GAe1B,OAdM,GAAQ,EAAU,UAAU,aAC5B,EAAO,EAAM,OACb,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC7C,EAAgB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC3D,EAAS,0BACT,EAAQ,OAAO,EAAa,kBAAkB,EAAY,eACrD,EAAI,EAAG,EAAI,EAAO,EAAG,IAC5B,EAAS,QAAQ,EAAC,KAAO,EACzB,GAAiB,EAAM,EAAO,EAAI,GAClC,EAAQ,IAAI,EAAC,MAAM,EAAa,MAAQ,EAE1C,GAAM,GAAO,KACb,MAAO;WACE,EAAQ,IAAI,EAAM;oBACT,EAAK;2BACE,EAAO;kCACA,EAAO;qDACY,EAAO,KAAK,EAAO;eACzD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,UACvC,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO,KAAK,EAAO;UACtD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;uBAIC,EAAO,KAAK,EAAO;;0BAEhB,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;gCAEU,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGvC,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO;oBAChC,EAAO;6BACE,EAAO,KAAK,EAAO,aAAa,EAAM;6BACtC,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,UACjD,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;iBAGL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;UAElD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;gCAGU,EAAO,KAAK,EAAO,KAAK,EAAO;;0BAErC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAKpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGxB,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAM;6BACpB,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEnE,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,SAAU,UAC3D,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;;iBAEL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAM,GAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAE3B,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;mBAGzC,EAAO;UAChB,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;0BAGlC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBACzC,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE;iBACnB,EAAM,GAAE;;;yBAGA,EAAO,OAAO,EAAO;+BACf,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;;0BAGM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAO,eAAe,EAAM;6BAC1C,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAA2B,GACzB,GAAM,GAAU,EAAU,KACpB,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAEtD,MAAI,GAAS,EACJ,UAAU,EAAO,IAEnB;0BACiB,EAAM;;iBAEf,EAAO;;;IAMxB,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAEpC,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAE7C,EAAO,GAAkB,GACzB,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,AAAI,EAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAGpC,GAAI,GAAS,sBACP,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAChD,EAAgB,IAAW,EAC3B,EAAU,EAAA,KAAK,cAAc,EAAa,cAC1C,EAAiB,IAAY,EAEnC,GAAI,IAAW,GAAK,CAAC,GAAiB,CAAC,EACrC,EAAS;;cAGA,GAAiB,CAAC,EAC3B,AAAI,IAAY,EACd,EAAS;;QAIT,EAAS;;gBAIF,EAAc,QACvB,GAAM,GAAO,EAAS,EAChB,EAAO,EAAS,EAEtB,AAAI,EAAc,QAAQ,GAAQ,IAAM,EAAc,QAAQ,GAAQ,GACpE,EAAS,8BACJ,AAAI,EAAc,QAAQ,GAAQ,GACvC,EAAS,2EAEJ,AAAI,EAAc,QAAQ,GAAQ,IACvC,GAAS,gDAIb,MAAO;WACE,EAAQ;QACX,EAAI;QACJ,EAAa;8BACS,EAAc,IAAI,EAAqB;QAC7D,EAAM;;IAKd,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAc,EAAa,SAC3B,EAAa,EAAU,UAAU,SACjC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAE1C,GAAI,CAAC,EAAU,UAAU,WAAa,IAAW,GAC7C,EAAU,UAAU,YAAc,MAClC,EAAA,KAAK,YAAY,EAAY,GAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAKpC,GAAM,GAAO,GAAkB,GACzB,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAC7C,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,MAAI,GAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAG7B;YACG,EAAQ;QACZ,EAAI;QACJ,EAAa;kBACH,EAAc,IAAI,EAAqB;;gBAKvB,GAChC,GAAI,GAAQ,EACV,MAAO,MACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QAEP,KAAM,OAAM,gBAAgB,EAAI,yBAKpC,YACI,EAAmB,GAErB,GAAM,GAA0B,KAAK,MAAM,KAAK,UAAU,IAC1D,SAAa,UAAU,aAAe,EAC/B,EAGT,YAA2B,EAAkB,GAC3C,MAAO,GAAS,IAAI,SAAA,GAAK,MAAA,GAAO,KAAI,KAAK,MCt2C3C,AAuBA,GAAA,IAAA,WAOE,WACI,EAAiB,EAAoB,EACrC,GARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,EAAA,KAAK,OACD,EAAM,OAAS,EACf,WAAM,MAAA,aACF,GAAG,OAAO,GAAG,cACb,EAAG,MAAM,IAAE,6CACnB,GAAM,GAAS,EAAM,EAAM,OAAS,GAC9B,EAAU,KAAK,KAAK,EAAS,GACnC,KAAK,YAAc,EAAM,MAAM,EAAG,IAClC,AAAI,EAAU,GACZ,KAAK,YAAY,KAAK,GAExB,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,GAAM,GAAW,KAAK,YAChB,EAAO,EAAS,OAChB,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAEjC,EACA,EACJ,GAAI,IAAY,GACd,EAAa,EAAO,EACpB,GAAM,GAAiB,GAAkB,GACzC,EAAiB;UACb,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE,QAEtB,GAAa,EACb,EAAiB;UACb,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE,IAExB,GAAM,GAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACnD,EAAY,IAAM,EAAS,EAAa,GACxC,EAAc,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IACzC,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAE/C,EAAU,IAAO,MAAS,cAAgB,WAC1C,EAAoB,EAAY,GAAK;sDACO,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM,OAE7D,EAAa;0BACG,EAAW,OAAM;uCACJ,EAAW,OAAM;uCACjB,EAAW,OAAM;qDACH,EAAW,OAAM,UAE5D,EAAgC,EAAY,GAAK;qCACtB,EAAY,OAAM;4CACX,EAAS,OAAM;iDACV,EAAS,MAAM,IAAI,OAAM;SAGtE,KAAK,SAAW;0BACM,EAAY,OAAM;iCACX,EAAS,OAAM;sCACV,EAAS,MAAM,IAAI,OAAM;;QAEvD,EAA6B;;UAE3B,EAAK;4BACa,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;4BAC5C,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;UAC9D,EAAc;yCACiB,EAAS,eAAe,EAAS;sBACpD,EAAS,eAAe,EAAS,OAAO,EAAU;;;2BAG7C,EAAU;;8BAEP,EAAU;;YAE5B,EAAiB;6BACA,EAAU;;;mBAGpB,EAAM;;;;;;;;;;;MAazB,MAAA,MCjJA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAe,GAE1C,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;0CACT,EAAa;;;;;;;;;;;;;;gCAcvB,EAAqB;oBACjC,EAAc;gDACc,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;mBACnC,EAAa;kDACkB,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;;;;;;;MAevD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAc,EAAe,GAExD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAa;;;;;;;;;;;;;;;;;gCAiBvB,EAAoB;oBAChC,EAAa;gDACe,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;;;;;;;;MAgBzD,MAAA,MC3KA,AAwBO,GAAM,IAAmB,CAC9B,KAAM,wCACN,KAAM,yCAGR,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,QAAS,QAAS,QAAS,SAK1C,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GAEnE,KAAK,SAAW;;;UAGV,EAAE;;;;;;;;;;MAYZ,MAAA,MCpDA,AAqBA,GAAM,IAAoB;;;EAKb,GAAM,gBACN,GAAM,gBACN,GAAM,gBAMN,GAAU;;;;;;;;;;EAYV,GAAM;;;;;;;;;EAYN,GAAQ,wBAER,GAAY,wBAEZ,GAAO,uBAEP,GAAa,wBAEb,GAAU,uBAEV,GAAgB,wBAEhB,GAAc,sCAEd,GAAa,sCAEb,GAAM,GAAoB;;EAG1B,GAAM,GAAoB;;EAG1B,GAAM;qBAGN,GAAU,yCAEV,GAAQ,+BAErB,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;MAUZ,MAAA,MC1GA,AAuBA,GAAM,IAAoB;;;;;EAOb,GAAU;;;;;;;;;;;;;;;;;;;;;EAuBV,GAAM;;;;;;;;;;;;;;IAef,GAAoB;;EAIX,GAAQ;;;EAKR,GAAU;;;EAKV,GAAQ;;EAIR,GAAY;;EAIZ,GAAO;;EAIP,GAAa;;EAIb,GAAU;;EAIV,GAAgB;;EAIhB,GAAc;;;;EAMd,GAAa;;;;;EAOb,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIxB,GAAA,WAQE,WACI,EAAY,EAAkB,EAC9B,GAAA,AAAA,IAAA,QAAA,GAAA,IATJ,KAAA,cAAgB,CAAC,IAAK,KAGtB,KAAA,qBAAuB,GACvB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,GAAM,GAAO,KAAK,YAAY,OAC1B,EAAyB,GAC7B,GAAI,EACF,GAAI,IAAS,GAAK,EAAA,KAAK,cAAc,KAAK,eAAiB,EACzD,EAAyB;;;;eAMzB,GAAM,GAAQ,GAAkB,GAIhC,GAHA,EAAyB;YACrB,EAAK;UAEL,IAAS,EACX,GAA0B;yCACK,KAAK,YAAY,GAAE;;;iBAKlD,GAAM,GAAW,GAAY,SAAU,GACvC,GAA0B;;iBAEnB,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;iBAExD,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;;;aASrE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;UAQF,EAAsB;;;;MAMhC,MAAA,MC9MA,AAoBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,KASf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAgBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KCzDA,AAoBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;MAiBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KC5DA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,OAAQ,QAKvB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAepB,MAAA,MCzCA,AAoBA,GAAA,IAAA,WAME,WAAY,GAJZ,KAAA,YAAwB,GAKtB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAI1C,OADM,GAAW,CAAC,YAAY,EAAQ,GAAE,+BAC/B,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAC1B,EAAS,KACL,iBAAiB,EAAQ,GAAE,KAC3B,kBAAiB,EAAC,WAAW,EAAK,QAExC,GAAM,GAAY,EAAQ,OACpB,EAAY,EAAQ,EAAQ,OAAS,GAC3C,EAAS,KAAK,sBAAsB,EAAS,WAAW,EAAS,OAEjE,KAAK,SAAW;;;;;;UAMV,EAAS,KAAK;WAAa;;MAIrC,MAAA,MCzDA,AAuBA,GAAA,IAAA,WAOE,WAAY,EAAoB,GALhC,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,GAAM,GAAQ,KAAK,YACb,EAAO,EAAM,OACb,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAC/B,EAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACzD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAW1C,OARM,GAAU,EAAS,GACnB,EAAe,EAAS,MAAM,IAC9B,EAAc,EAAS,OAEzB,EAAkB,OAAO,EAAO,MAAM,EAAQ,GAAE;;oBAEpC,EAAW,WAAW,EAAa,OAAM;WAEhD,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAK1B,GAAmB;cACX,EAAO,MAAM,EAAQ,GAAE,QAAQ,EAAO,OAAO,EAAQ,EAAI,GAAE;;kBAEvD,EAAC,IAAI,GAAgB,EAAU,EAAS,GAAM;mBAC7C,GAAgB,EAAc,EAAS,GAAM;WAG5D,GAAM,GAAY,EAAQ,OACpB,EAAQ,EAAQ,EAAQ,OAAS,GACvC,GAAmB;;gBAEP,EAAS,IAAI,GAAgB,EAAU,EAAS,GAAM;iBACrD,GAAgB,EAAc,EAAS,GAAM,MAE1D,KAAK,SAAW;uBACG,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IAAE;UAC1C,EAAe;;;;UAIf,EAAK;sCACuB,EAAM;;UAElC,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;cACrC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;;MAMtC,MAAA,MAaA,YAAyB,EAAoB,EAAiB,GAC5D,GAAM,GAAa,EAAS,QAAQ,GAC9B,EAAM,EAAS,IAAI,SAAC,EAAG,GAC3B,MAAI,KAAQ,EACA,EAAC,MAAM,EAEV,IAGX,MAAO,GAAI,OC3Hb,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAiB,EAAS,aAAe,eAE/C,KAAK,SAAW;;;;;;;;;;;;8BAYU,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;oBAIhC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAExC,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;wCAEI,EAAM,aAAa,EAAM;;;;;;;gCAOjC,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;oCAEF,EAAS,YAAW;;oBAEpC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAEjC,KAAK,SAAW;;;;;;;;;;;8BAWU,EAAS,UAAS;kCACd,EAAS,SAAQ;iCAClB,EAAW,MAAM,EAAQ;;kCAExB,EAAS,QAAO;;;;oCAId,EAAS,UAAS;mCACnB,EAAY,MAAM,EAAM;;oCAEvB,EAAS,SAAQ;;;;sCAIf,EAAS,SAAQ;qCAClB,EAAW,MAAM,EAAO;;sCAEvB,EAAS,QAAO;;;;;;;;;;;;;MAetD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAEnD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAchC,EAAW;gDACK,EAAW;;oCAEvB,EAAS,SAAQ;;;;;yBAK5B,EAAW;;kCAEF,EAAY;kDACI,EAAY;;sCAExB,EAAS,UAAS;;;;;;2BAM7B,EAAY;;oCAEH,EAAW;oDACK,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;6BAM5B,EAAW;;sCAEF,EAAS,YAAW;;;;;;;;;;MAY1D,MAAA,MCzSA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;;;;;;;wBAOI,EAAU;;;;;8BAKJ,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;;;;;;;;;MAcpD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;gCAYnB,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;;oCAGF,EAAU;8BAChB,EAAU;;;;;;;;;MAWxC,MAAA,MCzIA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAChD,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAEpC,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;;2BAGT,EAAM,aAAa,EAAM;;;;;;;gCAOpB,EAAY;qCACP,EAAc;;gCAEnB,EAAS,SAAQ;;;;kCAIf,EAAW;uCACN,EAAa;;kCAElB,EAAS,QAAO;;;;oCAId,EAAqB;;;;;;;;oBAQrC,EAAc;;;;;;;;;;;;;;;;;;;kBAmBhB,KAA4B,GAAC;;oBAE3B,EAAc;;0CAEQ,EAAqB;mCAC5B,EAAqB;;;kCAGtB,EAAqB;mCACpB,EAAqB;;;yBAG/B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;;;;;yBAK5B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;gCACrB,EAAqB;;;;;;;;;;UAU3C,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAEtD,KAAK,SAAW;oCACgB,EAAW,KAAK,EAAY,KACxD,EAAW;iCACc,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;gCAgBhC,EAAW;qCACN,EAAa;;gCAElB,EAAS,QAAO;;;;kCAId,EAAY;uCACP,EAAc;;kCAEnB,EAAS,SAAQ;;;;oCAIf,EAAW;yCACN,EAAa;;oCAElB,EAAS,QAAO;;;;sCAId,EAAqB;;;;;;;;;;;;;;;;;oBAiBvC,KAA4B,GAAC;;4CAEL,EAAqB;qCAC5B,EAAqB;2BAC/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;;;2BAG/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;qCACrB,EAAqB;;;;;;;;;MAW1D,MAAA,MC3TA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAa,EAAS,YAAc,EAAS,WAE/C,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;wBAO3B,EAAU;4BACN,EAAU;;;;;;;;;gCASN,EAAY;qCACP,EAAc;;gCAEnB,EAAQ;;;;kCAIN,EAAW;uCACN,EAAa;;kCAElB,EAAQ;;;;;;;;;;;UAWhC,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MCrHA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAR/B,KAAA,cAAgB,CAAC,IAAK,KACtB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,SAgB5B,OAdM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAEjB,EAAW,gCAEN,EAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY;wBACI,EAAC,IAAI,EAAI,EAAC;mBACf,EAAC,IAAI,EAAC;mBACN,EAAC,IAAI,EAAC,eAYrB,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAS,EAAG,EAAS,EAAc,KAC1C,GAAM,GAAI,EAAS,EAOnB,GALA,GAAY;4BACQ,EAAI,EAAc;4BAClB,EAAI,EAAa;UAGjC,IAAgB,GAClB,GAAI,EAAI,GAEN,CAAI,EAAU,IAAM,EAUlB,GAAY;;qCAEW,EAAQ,mCAC3B,EAAQ;2BACC,EAAC,IAAI,EAAC;;;;uCAIM,EAAQ;6BAClB,EAAC,IAAI,EAAC;;;2BAGR,EAAC,IAAI,EAAC;;;;qCAII,EAAQ,mCAC3B,EAAQ;;;;;uCAKa,EAAQ;;;;sBAIzB,EAAC,IAAI,EAAC,+BAA+B,EAAC,IAAI,EAAC;;sBAE3C,EAAC,IAAI,EAAC,wBAAwB,EAAC,IAAI,EAAC;;gBAK5C,GAAY;qCACW,EAAQ,uBAAuB,EAAQ;2BACjD,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;oBAGb,EAAC,IAAI,EAAC,aAAa,EAAC,IAAI,EAAC;gBAI7B,EAAI,EAAI,IAOV,GAAM,GAAkB,EAAU,IAAM,EACpC,EAAA,KAAK,kBAAkB,GACvB,EAEJ,AAAK,EAAgB,IAAM,GAAK,EAAU,IAAM,GAC3C,EAAgB,IAAM,GAAK,EAAU,IAAM,EAC9C,IAAY;oCACQ,EAAU,EAAC,MAAM,EAAe;;uCAE7B,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;kBAMvB,AAAI,EAAgB,GAClB,IAAY;;yCAEW,EAAQ;oDACG,EAAQ;+BAC7B,EAAC,IAAI,EAAC;;+BAEN,EAAC,IAAI,EAAC;;qBAKrB,GAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;mBAG5C,GAAY;oCACQ,EAAe;;uCAEZ,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;;sBAGjB,EAAC,IAAI,GAAI,GAAC,aAAa,EAAC,IAAI,GAAI,GAAC;uBAM7C,AAAI,GAAI,GACN,IAAY;mCACW,EAAQ;cAS/B,AAAI,EAAU,IAAM,EAClB,IAAY;sCACY,EAAW;iDACA,EAAQ;2BAC9B,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;6CAGY,EAAQ;2BAC1B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;;wCAEY,EAAW;mDACA,EAAQ;;;sBAGrC,EAAC,IAAI,GAAI,GAAC,kBAAkB,EAAC,IAAI,GAAI,GAAC;oBAI9C,IAAY;qCACW,EAAQ;2BAClB,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;kCAGC,EAAW;iDACI,EAAQ;2BAC9B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;oBAKhD,GAAY,KAIhB,AAAI,EAAI,GACN,IAAY;0BACI,EAAC,IAAI,EAAC,WAAW,EAAC,KAAK,EAAC;gBAClC,EAAC,IAAI,EAAC,kBAAkB,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,EAAC;YAGzD,AAAI,EAAI,EAAI,GACV,IAAY;4BACI,EAAC,IAAI,GAAI,GAAC,WAAW,EAAC,KAAK,GAAI,GAAC;kBAC1C,EAAC,IAAI,GAAI,GAAC;8BACE,EAAC,IAAI,GAAI,GAAC,eAAe,EAAC,IAAI,GAAI,GAAC,UAM7D,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY,gBAAgB,EAAC,IAAI,EAAC,QAAQ,EAAC,IAAI,EAAC,IAIpD,GAAI,GAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;UAezC,EAAQ;;;UAGR,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MC3UA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAA4B,EAC5B,GAPJ,KAAA,cAAgB,CAAC,QAAS,QAAS,UACnC,KAAA,YAAwB,GAOf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAChC,EAAA,EAAA,GACA,EAAA,EAAA,GAAY,EAAA,EAAA,GACnB,KAAK,YAAc,CAAC,EAAU,EAAY,EAAW,GACrD,GAAM,GAAW,IAAW,WAAa,EAAI,EAEvC,EAAA,CAAA,EAAA,EAAA,KAAA,EAAA,EAAA,MAAC,EAAA,EAAA,GAAkB,EAAA,EAAA,GAGnB,EAAA,EAAA,oHAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAa,EAAA,EAAA,GAW3B,EAAA,EAAA,kHAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAe/B,KAAK,SAAW;yCACqB,EAAW;wCACZ,EAAU;;;;;;;;;;;;;;;;iCAgBjB,EAAK;;;;+BAIP,EAAW;8BACZ,EAAU;;uBAEjB,EAAG;mCACS,EAAgB;4BACvB,EAAkB;;;uBAGvB,EAAG;mCACS,EAAe;4BACtB,EAAkB;;;;;aAKjC,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;MA0BrB,MAAA,MC1GA,GAAA,WAQE,WAAY,EAAiB,EAAoB,GAPjD,KAAA,cAAgB,CAAC,KAQf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAM,OACb,EAAM,EAAY,MAAQ,QAAQ,GAAU,EAAM,UAAS,IAC3D,EAAS,EAAM,EAAM,OAAS,GAChC,EAAY,GACZ,EAAY,GAIhB,AAAI,EACF,GAAY,EAAU,UAAU,GAAS,GAAM,WAC/C,EAAY,EAAU,UAAY,WAElC,GAAY,EAAU,gBAAgB,EAAW,cACjD,EAAa,EAAU,aAAe,cAGxC,KAAK,SAAW;;;UAGV,GAAkB,GAAK;oBACb,GAAc,EAAM,UAAS;sBAC3B,EAAG;;cAEX,EAAS;sBACD,EAAS;YACnB,GAAc,EAAM,UAAS;wBACjB,GAAU,EAAM,UAAS;;;;MAO/C,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,OAAS,MAChB,GAAK,MAAQ,EAAM,mBAAmB,EAAc,UAEtD,EAAM,GAAG,UAAU,EAAK,MAAO,KAGrC,KAEA,YAAmB,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,KACpB,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,KAC/B,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,OAAO,EAAI,KAE/C,KAAM,OAAM,2BAA2B,EAAI,yBAI/C,YAAuB,EAAc,GACnC,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KAEd,KAAM,OAAM,2BAA2B,EAAI,yBCjG/C,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAEjB,KAAA,aAAe,GACf,KAAA,aAAe,GAEf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAuB,EAAmB,GAP9C,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAOtB,KAAK,YAAc,EACnB,KAAK,UAAY,EACjB,KAAK,WAAa,EAClB,KAAK,SAAW;;;;gBAIJ,KAAK,uBAAsB;gBAC3B,KAAK,sBAAqB;gBAC1B,KAAK,sBAAqB;;uBAEnB,EAAS;+BACD,EAAS;uBACjB,EAAS;+BACD,EAAS;mCACL,EAAS;UAClC,KAAK,qBAAoB;;;uBAGZ,KAAK,yBAAwB;;;IAM1C,SAAA,UAAA,qBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,mBAAR,WACE,MAAI,MAAK,aAAe,OACf,KAAK,YAAY,GAEjB,KAAK,YAAY,IAIpB,EAAA,UAAA,uBAAR,WACE,MAAI,MAAK,aAAe,OACf,4BAEA,6BAGb,KC5FA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,CAAC,EAAM,GAC1B,KAAK,SAAW;;;;;;MAQpB,MAAA,MClCA,AAsBA,GAAA,IAAA,WAME,WAAY,GALZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;UAIlB,EAAK,OAAM;;MAIrB,MAAA,MCxCA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;;UAKlB,EAAK,OAAM;;MAIrB,MAAA,MC3CA,AAqBA,GAAA,IAAA,WAKE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IANJ,KAAA,cAAgB,CAAC,KAOf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAS,SACb,AAAI,GACF,GAAS,8BAGX,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;8BAUrB,EAAK;kCACD,EAAK;iDACU,EAAK,OAAO,EAAM;wBAC3C,EAAK,UAAS;;;;;;;;;;;;;;UAc5B,EAAK,OAAM,WAAW,EAAM;;MAItC,MAAA,MCtEA,AAqCA,GAAA,IAAA,WAOE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAW,GACX,EAAS,SACb,AAAI,GACF,GAAS,8BAGX,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAC1B,GAAM,GAAU,EAAM,EAAI,EAE1B,GAAY;;gCAEY,EAAG,MAAM,EAAY,GAAE;gCACvB,EAAG;kCACD,EAAG,MAAM,EAAY,GAAE;kCACvB,EAAG;;;;;;;gCAOL,EAAK;oCACD,EAAK;kDACS,EAAK,OAAO,EAAM;yBAC3C,EAAK,UAAS;;;yBAGd,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;;;UAQ5B,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;;UAWzC,EAAQ;;UAER,EAAK,OAAM,MAAM,EAAM;;MAIjC,MAAA,MC/GA,AAmBO,GAAM,IAAc,CACzB,KAAM,oCACN,KAAM,qCAGR,GAAA,WAKE,WAAY,EAAY,EAA8B,GAJtD,KAAA,cAAgB,CAAC,OAAQ,QAKvB,GAAM,GAAW,EAAW,GAC5B,KAAK,YAAc,EAEnB,GAAM,GACF,EAAU,SAAS,KAAK,GAAO,UAAU,KAAK,GAC5C,EAAoB,EAAa,EAAQ,KAAO,MAEtD,KAAK,SAAW;yCACqB,EAAyB;;;UAGxD,EAAE;;;;kDAIsC,EAAQ;;;;;;8BAM5B,EAAQ;;;;;;;;;yDASmB,EAAiB;;;;;;;;;;MAY1E,MAAA,MCxEA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAiB,GAL7B,KAAA,YAAwB,GAMtB,KAAK,cAAgB,CAAC,KACtB,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;;MASlB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,UAEhE,EAAM,GAAG,UAAU,EAAK,SAAU,KAGxC,KChDA,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,EAAuB,GALrD,KAAA,cAAgB,CAAC,IAAK,WAMpB,GAAM,GAAwB,EAAO,QACrC,EAAY,GAAQ,EACpB,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,EAAQ,GAE7C,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,EAAkB,GACzC,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,mBAAmB,EAAI,yBAErC,GAAI,IAAS,EACX,MAAO,yBAMT,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAElD,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,IAAM,EACR,EAAa,KAAK,kBAAkB,EAAc,GAAE,MAEpD,EAAa,KAAK,GAAG,EAAc,IAGvC,MAAO,GAAa,OC3CtB,GAAA,IAAA,WAIE,WACY,EAA0B,EAAmB,GAA7C,KAAA,SAAA,EAA0B,KAAA,QAAA,EAJtC,KAAA,cAAgB,CAAC,IAAK,WAKpB,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAChC,EAAe,KAAK,SAAW,EAAI,aAAe,UACxD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,KAAK,QAAO;;YAElD,EAAK;;gCAEe,KAAK,SAAQ;;sCAEP,EAAY;;;;QAMlD,MAAA,MC1CA,YAwBmC,GACjC,GAAM,GAAO,KACP,EAAwB,EAAK,QAAO;;MAEtC,EAAK,UAAS;MACd,EAAK,UAAS;MACd,EAAK,UAAS;;;;;OAMlB,MAAO,IAA8B,EAAI,eAGR,GAEjC,GAAM,GAAc,GAAI,cACpB,CAAC,GAAI,EAAG,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,GAAI,EAAG,EAAG,IAClE,MAAO,IAAoC,EAAI,eAGf,GAEhC,GAAM,GAAwB,GAAI,aAAY,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,IAC9D,MAAO,IAAmC,EAAI,GAGhD,YACI,EAA2B,EAAe,EAC1C,EAAwB,EACxB,GACF,GAA+B,EAAO,GACtC,GAAM,GAAU,GAAyB,GAEnC,EAAQ,EAAG,WACjB,UAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAO,KACxD,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAO,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAC5C,EAAa,QACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,QACzD,cAIL,GACF,MAAO,GAAc,gCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAG,mBAIvC,GACF,MAAO,GAAc,oCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAc,kCAIlD,GACF,MAAO,GAAc,kCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAG,2BAIL,GACF,MAAO,GAAc,sCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EAAQ,GAAwC,GAC3D,EAAG,KAAM,EAAG,mBAId,GACF,MAAO,GAAc,0CAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAc,kCAIhB,EAA2B,EAC3B,GACF,GAAM,GAAY,EACZ,EAAW,EAAI,EACf,EAAU,EAAI,EAAM,EAAI,EAC9B,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KAC7C,GAAM,GAAU,GACZ,EAAI,EAAS,eAAgB,EAAc,EAAG,EAAQ,GAC1D,MAAO,IACH,GACI,EAAI,EAAS,KAAM,EAAc,EAAG,EAAQ,eAIlD,EAA2B,EAAuB,EAClD,EAAgB,EAAkB,GACpC,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAEhE,GAAI,GAA2B,EAAuB,EACtD,AAAI,YAAgB,YAClB,GAAgB,GAAI,YAAW,EAAQ,EAAS,GAChD,EAAgB,EAAG,cACnB,EAAiB,EAAG,MAEpB,GAAgB,GAAI,cAAa,EAAQ,EAAS,GAClD,EAAgB,EAAG,MACnB,EAAiB,EAAc,2BAGjC,EAAc,IAAI,GAElB,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAAG,KACvD,EAAe,KAEvB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA2B,EAC3B,GAEF,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAChE,AAAK,EAAqB,eAAgB,YACxC,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAO,MAAO,EAAO,OAAQ,EAAG,EAAG,KAC9D,EAAG,cAAgB,EAAqB,QAEhD,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAG,KAAM,EAAG,cACvC,KAIV,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA6B,EAAc,EAC3C,GAEF,GAAM,GAAS,EAAI,eACnB,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,KAGrD,GAAM,GAAgB,EAChB,EAAiB,EACjB,EAAkB,EAAgB,EAAiB,EAAO,EAEhE,UACI,EACA,WAAM,MAAA,GAAI,WACN,EAAI,kBAAmB,EAAiB,EAAI,eAIpD,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAG,EAAG,EAAS,EAAM,EAAI,KAAM,EAAI,MAAO,KAExE,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,QAE9C,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAM,EAEN,EAAiB,GAAI,cAAa,GAExC,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAGJ,EAAc,EACd,EAAiB,GAAI,YACvB,GAA4C,EAAO,EAAS,IAEhE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAG,EAAG,EAAc,sBAAuB,EAAG,cACpD,KAID,GAAI,cAAa,EAAe,oBAIrC,EAA2B,EAAqB,EAAe,EAC/D,EAAc,EAAsB,EACpC,GACF,GAAM,GAAM,EAEN,EACF,GAAI,cAAa,GACb,EAAc,IAEtB,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAa,GAAI,cAAa,EAAe,EAAe,GAClE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAc,EAAc,EAAG,KAAM,EAAG,MAAO,KAEtD,owBCxTT,kBAgDE,WAAY,GANZ,KAAA,cAAmC,KACnC,KAAA,QAA6B,KACrB,KAAA,SAAW,GAqOX,KAAA,oBAAsB,GAsPtB,KAAA,YAA0B,GAtdhC,GAAM,GAAY,EAAA,MAAM,UAAU,iBAClC,AAAI,GAAM,KACR,MAAK,GAAK,EACV,GAAgB,EAAW,IAE3B,KAAK,GAAK,GAAgB,GAG5B,GAAI,GAAqB,2BACnB,EAA0B,8BAChC,GAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAM,GAAgB,oBAChB,EAAqB,yBAI3B,GAFA,KAAK,sBACD,GAA+B,KAAK,GAAI,GACxC,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,sHAKN,GADA,KAAK,0BAA4B,KAAK,GAAG,aAAa,GAClD,GAAwB,KAAK,GAAI,GACnC,KAAK,8BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,uIAIN,EAAqB,yBACjB,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,KAAK,GAAG,aAAa,WAChB,GAAwB,KAAK,GAAI,GAC1C,KAAK,8BACD,KAAK,GAAG,aAAa,OAEzB,MAAM,IAAI,OAAM,uDAIpB,KAAK,aAAe,GAA8B,KAAK,IACvD,KAAK,YAAc,GAA6B,KAAK,IACrD,KAAK,YAAc,GAA6B,KAAK,IAErD,KAAK,cACD,GAA0B,KAAK,GAAI,KAAK,2BAG9C,cAAA,eAAY,EAAA,UAAA,QAAK,KAAjB,WACE,MAAO,GAAA,MAAM,QAAQ,0CAGhB,EAAA,UAAA,QAAP,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAEF,AAAI,KAAK,SAAW,MAClB,QAAQ,KACJ,wKAIN,AAAI,KAAK,eAAiB,MACxB,QAAQ,KACJ,sMAKN,GAAM,GAAK,KAAK,GAChB,GAAwB,EAAI,WAAM,MAAA,GAAG,WACrC,GAAwB,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QACrE,GAAwB,EAAI,WAAM,MAAA,GAAG,kBAAkB,EAAK,eAC5D,GAAwB,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,QACjE,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,QACrD,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAK,eACvD,KAAK,SAAW,IAGX,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,yBAAP,SACI,EACA,GACF,KAAK,kBACL,GAAoC,KAAK,GAAI,EAAS,IAGjD,EAAA,UAAA,2BAAP,SACI,EAAuB,EAAe,EAAgB,GACxD,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,EAAO,EAAQ,EAAM,KAAK,gBAG3C,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,0BAAP,SAAiC,EAAc,GAE7C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,oBAAP,SAA2B,GAA3B,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,KAAK,gBAAkB,GACzB,IAA6C,KAAK,GAAI,KAAK,aAC3D,KAAK,cAAgB,MAEvB,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAGxD,EAAA,UAAA,gDAAP,SACI,EAAuB,EAAc,GADzC,GAAA,GAAA,KAEE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAM,EAAS,EAAK,kBAGhC,EAAA,UAAA,+BAAP,SACI,EAAqB,EAAe,EAAc,EAClD,EAAsB,GACxB,MAAO,IACH,KAAK,GAAI,EAAQ,EAAO,EAAM,EAAS,EAAc,EACrD,KAAK,gBAGJ,EAAA,UAAA,gCAAP,SAAuC,EAAqB,GAE1D,MAAO,IAA2C,KAAK,GAAI,EAAQ,IAG9D,EAAA,UAAA,wBAAP,SACI,EAAuB,EAAc,GACvC,KAAK,yBAAyB,GAC9B,GAAM,GAAS,GACX,KAAK,GAA8B,EAAM,EAAS,KAAK,eAC3D,YAAK,6BACE,GAGF,EAAA,UAAA,sBAAP,WACE,GAAM,GAAe,KAAK,YAAY,KAAK,IAC3C,MAAO,MAAK,UAAU,IAGhB,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACM,EACA,EAEJ,GAAI,EAAA,MAAM,QAAQ,4BAChB,GAAM,GAAM,EAEN,EAAO,EAAI,UAAU,EAAI,2BAA4B,GAC3D,EAAG,QAEH,EAAgB,WACd,GAAM,GAAS,EAAI,eAAe,EAAM,EAAG,GAC3C,MAAO,KAAW,EAAI,kBAClB,IAAW,EAAI,qBAGrB,EAAQ,MACH,AACH,GAAA,MAAM,UAAU,gDAAkD,EACpE,GAAQ,KAAK,aACb,KAAK,WACL,EAAgB,WAAM,MAAA,GAAK,iBACvB,EACA,EAAA,MAAM,UAAU,mDAMpB,EAAgB,WAAM,MAAA,IAGxB,MAAO,CAAC,MAAK,EAAE,cAAa,IAGvB,EAAA,UAAA,gCAAP,SACI,EAAuB,EACvB,GAFJ,GAAA,GAAA,KAGE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAc,MAK1B,EAAA,UAAA,cAAP,SAAqB,GACnB,KAAK,kBACL,GAAM,GAAK,KAAK,GACV,EACF,GAAgC,EAAI,GAClC,EAA4B,GAA8B,GAC1D,EAAwB,GAAyB,GACvD,UAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAuB,EAAI,GAC3B,AAAI,KAAK,OACP,GAA2B,EAAI,GAEjC,AAAK,KAAK,qBACR,MAAK,WAAW,GAChB,KAAK,oBAAsB,GACvB,EAAI,KAAK,QAAS,KAAK,eAEtB,GAGF,EAAA,UAAA,cAAP,SAAqB,GAArB,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,IAAY,KAAK,SACnB,MAAK,QAAU,MAEjB,AAAI,GAAW,MACb,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAI1D,EAAA,UAAA,WAAP,SAAkB,GAAlB,GAAA,GAAA,KACE,KAAK,kBACL,KAAK,QAAU,EACf,AAAK,KAAK,SAAW,MAAS,KAAK,OACjC,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,WAAW,MAGrD,EAAA,UAAA,mBAAP,SACI,EAAuB,EACvB,GAEF,MAFE,AAAA,KAAA,QAAA,GAAA,IACF,KAAK,kBACD,EACK,GACH,KAAK,GAAI,EAAS,GAEf,GACH,KAAK,GAAI,EAAS,IAInB,EAAA,UAAA,qBAAP,SAA4B,EAAuB,GAAnD,GAAA,GAAA,KAEE,YAAK,kBACE,GACH,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,kBAAkB,EAAS,MAGjD,EAAA,UAAA,0BAAP,SAAiC,EAAuB,GAEtD,YAAK,kBACE,KAAK,GAAG,mBAAmB,EAAS,IAGtC,EAAA,UAAA,sBAAP,SACI,EAAkC,EAClC,GACF,KAAK,kBACL,KAAK,mBACL,GACI,KAAK,GAAI,EAAoB,EAAiB,IAG7C,EAAA,UAAA,uBAAP,SACI,EAAmC,EAAc,GACnD,KAAK,6BAA6B,EAAqB,EAAS,IAG3D,EAAA,UAAA,6BAAP,SACI,EAAyC,EAAc,GACzD,KAAK,kBACC,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,KAAK,6BAA6B,EAA2B,EAAO,IAG/D,EAAA,UAAA,2BAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAK,iCACD,EAAa,EAAU,EAAY,IAGlC,EAAA,UAAA,iCAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAM,IAAI,OAAM,sDAGX,EAAA,UAAA,cAAP,WACE,AAAI,KAAK,SAAW,MAClB,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAA+B,KAAK,KAG/B,EAAA,UAAA,eAAP,WACE,KAAK,kBACL,KAAK,mBACL,GAAM,GAAK,KAAK,GAChB,AAAI,KAAK,OACP,KAAK,gBAEP,GACI,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,UAAW,EAAG,EAAG,eAAgB,MAG7D,EAAA,UAAA,+BAAP,WAAA,GAAA,GAAA,KACE,KAAK,kBACL,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,YAGzC,EAAA,UAAA,uBAAR,WAEE,MAAI,MAAK,6BAA+B,MACtC,MAAK,4BACD,GACI,KAAK,GACL,EAAA,MAAM,UACF,kDAAoD,EACpD,kCACA,6BAIP,KAAK,6BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGd,EAAA,UAAA,WAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EAAQ,EAAI,cAClB,SAAI,WAAW,EAAI,iBAAkB,GAC9B,EAET,GAAM,GAAM,KAAK,+BACX,EAAQ,EAAI,iBAClB,SAAI,cAAc,EAAI,iBAAkB,GACjC,GAGT,EAAA,UAAA,SAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BACjB,EAAI,SAAS,EAAI,kBACjB,OAEF,GAAM,GAAM,KAAK,+BACjB,EAAI,YAAY,EAAI,mBAGT,EAAA,UAAA,uBAAb,SAAoC,wGAClC,MAAA,CAAA,EAAM,EAAA,KAAK,YACP,WAAM,MAAA,GAAK,UAGP,EAAK,iBACD,EACA,EAAA,MAAM,UACF,2DAPhB,SAAA,OAQA,CAAA,EAAO,KAAK,aACR,EAAO,EAAA,MAAM,UAAU,wDAGrB,EAAA,UAAA,aAAR,SAAqB,EAAmB,GACtC,GAAI,IAAsB,EACxB,MAAO,MAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GAEX,EAAmB,EAAI,kBAAkB,EAAO,EAAI,cAE1D,MAAO,GAAmB,SAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,kBAErC,MAAO,GAAmB,MAItB,EAAA,UAAA,iBAAR,SAAyB,EAAmB,GAE1C,GAAI,IAAsB,EACxB,MAAO,GAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,wBACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,cAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,4BACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,WAI9B,EAAA,UAAA,UAAA,SAAU,GAAV,GAAA,GAAA,KACE,MAAO,IAAI,SAAc,SAAA,GACvB,EAAK,cAAc,WAAM,MAAA,GAAa,iBAAiB,WAAM,MAAA,UAMjE,EAAA,UAAA,UAAA,WAGE,OADM,GAAQ,GAAqB,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,YACtD,EAAI,EAAG,GAAK,EAAO,EAAE,GACrB,GAAA,GAAA,KAAA,YAAA,GAAA,UACP,IAEF,KAAK,YAAc,KAAK,YAAY,MAAM,EAAQ,IAG5C,EAAA,UAAA,cAAR,SAAsB,EAAyB,GAA/C,GAAA,GAAA,KAEE,GADA,KAAK,YAAY,KAAK,CAAC,SAAQ,EAAE,UAAS,IACtC,KAAK,YAAY,OAAS,EAE5B,OAGF,EAAA,KAAK,YAAY,WACf,SAAK,YAEE,EAAK,YAAY,SAAW,KAI/B,EAAA,UAAA,yBAAR,SAAiC,GAC/B,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,KAAK,aAC3B,AAAI,KAAK,OACP,GAA+B,KAAK,KAIhC,EAAA,UAAA,2BAAR,WACE,AAAI,KAAK,eAAiB,KACxB,IACI,KAAK,GAAI,KAAK,cAAe,KAAK,aACtC,AAAI,KAAK,OACP,GAA+B,KAAK,KAGtC,GAA6C,KAAK,GAAI,KAAK,cAIvD,EAAA,UAAA,qBAAR,SACI,EACA,GACF,KAAK,yBAAyB,GAC9B,GAAM,GAAS,IACf,YAAK,6BAEE,GAGD,EAAA,UAAA,6BAAR,SACI,EAA8C,EAC9C,GACF,KAAK,kBACL,GAAM,GAAK,KAAK,GAChB,GACI,EAAI,EAAgC,KAAK,aAC7C,AAAI,KAAK,OACP,GAA+B,GAEjC,KAAK,cAAgB,EACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAO,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAO,MAGpD,EAAA,UAAA,iCAAR,SACI,EAAW,EAAW,EAAe,GADzC,GAAA,GAAA,KAEE,KAAK,kBACL,GACI,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,QAAQ,EAAG,EAAG,EAAO,MAG1C,EAAA,UAAA,gBAAR,WACE,GAAI,KAAK,SACP,KAAM,IAAI,OAAM,4CAIZ,EAAA,UAAA,iBAAR,WACE,GAAI,KAAK,SAAW,KAClB,KAAM,IAAI,OAAM,qCAGtB,iBAaqC,GAEnC,OADI,GAAI,EACD,EAAI,EAAI,OAAQ,EAAE,GACvB,GAAM,GAAS,EAAI,KACnB,GAAI,CAAC,EACH,MAGJ,MAAO,GAAI,ECtnBb,YAgEI,EAAqB,EAAuB,EAC5C,GACF,GAAM,GAAW,EAAQ,SACnB,EAA0B,EAAO,IAAI,SAAC,EAAO,GACjD,GAAM,GAAuB,CAC3B,aAAc,EAAM,MACpB,SAAU,EAAM,UAAY,KAAO,EAAM,QAAQ,SACjD,UAAW,EAAM,UACjB,SAAU,EAAM,UAAY,GAAQ,EAAM,QAAQ,SAClD,WAAY,MAEd,MAAI,GAAM,SAAW,MAAQ,EAAM,QAAQ,OAAS,MAChD,EAAM,QAAQ,MAAM,WAAa,GACnC,GAAU,WAAa,EAAM,QAAQ,MAAM,YAEtC,CAAC,KAAM,EAAQ,cAAc,GAAI,UAAS,KAE7C,EAAe,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,YACrC,EAA0B,CAC9B,aAAc,EAAO,MACrB,SAAU,EAAO,QAAQ,SACzB,UAAW,GACX,SAAU,EAAO,QAAQ,SACzB,WAAY,MAER,EAAS,GACX,EAAY,EAAc,EAAU,EAAQ,cAE1C,EAAe,EAAM,cAAc,GAGrC,EAA+B,KAC7B,EAAS,EAAM,mBAAmB,EAAc,MAAO,IAC7D,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAS,EAAM,mBAAmB,EAAc,WAAY,KAK9D,OADM,GAA2D,GACxD,EAAI,EAAG,EAAI,EAAQ,cAAc,OAAQ,KAChD,GAAM,GAAU,EAAQ,cAAc,GAChC,EAAc,GACpB,EAAiB,GACb,EAAM,mBAAmB,EAAc,EAAS,GACpD,EAAiB,SAAS,GACtB,EAAM,mBAAmB,EAAc,SAAS,EAAW,GAGjE,MAAO,CACL,QAAO,EACP,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,aAAY,EACZ,aAAY,EACZ,OAAM,EACN,OAAM,GAIV,YACI,EAAyB,GAC3B,GAAI,EAAW,SAAW,EAAO,OAC/B,KAAM,OACF,4BAA4B,EAAW,OAAM,gBAC7C,sBAAqB,EAAO,OAAM,YAGxC,EAAW,QAAQ,SAAC,EAAG,GACrB,GAAM,GAAS,EAAE,aACX,EAAQ,EAAO,GACf,EAAS,EAAM,MAErB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAQ,GAC5B,KAAM,OACF,kDACA,6BAA4B,EAAM,QAAQ,EAAM,gBAGtD,GAAI,EAAE,WAAa,EAAM,UACvB,OAGF,GAAM,GAAY,EAAE,SACd,EAAY,EAAM,UAAY,KAAO,EAAM,QAAQ,SACzD,GAAI,CAAC,EAAA,KAAK,YAAY,EAAW,GAC/B,KAAM,OACF,6DACA,yBAAwB,EAAS,QAAQ,EAAS,8BAMxD,EAAqB,EAAqB,EAC1C,EACA,GAEF,GAAyB,EAAO,aAAc,GAC9C,GAAyB,CAAC,EAAO,cAAe,CAAC,IAEjD,GAAM,GAAS,EAAO,QAAQ,QACxB,EAAc,EAAO,QAAQ,SACnC,AAAI,EAAO,QAAQ,SACjB,EAAM,6BAA6B,EAAQ,EAAY,GAAI,EAAY,IAEvE,EAAM,uBAAuB,EAAQ,EAAY,GAAI,EAAY,IAEnE,EAAM,WAAW,EAAO,cAGxB,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,CAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,WAGtC,AAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,KAIpC,EAAO,QAAQ,SAAC,EAAO,GACrB,GAAM,GAAU,EAAO,QAAQ,cAAc,GACvC,EAAS,EAAO,iBAAiB,GACjC,EAAe,EAAO,iBAAiB,SAAS,GAEtD,GAAI,GAAU,KAEZ,OAGF,GAAI,EAAM,WAER,GAAI,EAAA,KAAK,cAAc,EAAM,OAAS,EACpC,EAAM,GAAG,UAAU,EAAQ,EAAM,cAAc,SAE/C,GAAI,GAAO,EAAM,cACjB,AAAM,YAAgB,eACpB,GAAO,GAAI,cAAa,IAE1B,EAAM,GAAG,WAAW,EAAQ,GAE9B,OAIF,AAAI,EAAM,QAAQ,OAAS,MAAQ,GAAgB,MACjD,EAAM,GAAG,UAAU,EAAc,EAAM,QAAQ,MAAM,YAGvD,EAAM,sBAAsB,EAAM,QAAQ,QAAS,EAAQ,KAG7D,AAAI,GAAe,MACjB,EAAY,EAAO,EAAO,cAE5B,EAAM,6BAIJ,EAAuB,EAAsB,GAC/C,GAAI,GAAY,GAChB,EAAO,OAAO,GAAQ,QAAQ,SAAA,GAC5B,GAAM,GAAY,EAAE,SAAW,MAAQ,EAAE,QAAQ,OAAS,MACtD,EAAE,QAAQ,MAAM,WAAa,EAC3B,EAAW,EAAE,UAAY,UAAY,EAAE,QAAQ,SACrD,GAAgB,EAAE,MAAK,IAAI,EAAQ,IAAI,IAEzC,GAAM,GAAc,EAAQ,SACxB,EAAM,EAAQ,YAAY,KAE9B,UAAO,IAAM,EAAY,IAAM,EACxB,EC5OT,AAqBA,GAAA,IAAA,WAOE,WACI,EAAuB,EACvB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAsBnB,OAnBE,GAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,cACA,EAAA,EAAA,eACA,EAAA,EAAA,WAEK,EAAA,EAAA,KAAM,EAAA,EAAA,IACP,EAAmB,EAAa,EAChC,EAAO,KACP,EAAiB,IAAe,eAChC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAEhC,EAAW,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,GAAY;gCACY,EAAG;yBACV,EAAG;;4BAEA,EAAY,GAAE,aAAa,EAAY,GAAE;0CAC3B,EAAQ,QAAQ,EAAY,MAC1D,EAAG;6BACc,EAAc,aAAa,EAAgB;;sBAElD,EAAW,GAAO;;qDAEa,EAAQ,QACjD,EAAW,OAAO,EAAI;+BACH,EAAa,2BAChC,EAAgB,QAAQ,EAAU;;wBAEtB,EAAW,GAAO;;2CAEC,EAAU;;sBAE/B,EAAc;;2BAET,GAAM,EAAI,GAAG;;;;;2BAKb,GAAM,EAAI,GAAG;;;;;;;UAWpC,KAAK,SAAW;;;;;;;;;UASV,EAAQ;;UAER,EAAK,OAAM;;MAIrB,MAAA,MC5GA,AAmBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAgB,EAAc,EAChD,GANJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMtB,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;;wBASI,EAAG,UAAU,EAAG;;oCAEJ,EAAI;;;;;0BAKd,EAAW;;;MAKrC,MAAA,MClEA,AAmBA,GAAA,IAAA,WAUE,WACI,EAAsB,EAAqB,EAAc,EACzD,GAXJ,KAAA,cAAgB,CAAC,aAAc,cAAe,MAC9C,KAAA,YAAwB,GAWtB,KAAK,YAAc,EACnB,KAAK,MAAQ,EAAW,GACxB,KAAK,YAAc,EACnB,KAAK,KAAO,EACZ,KAAK,MAAQ,EACb,KAAK,KAAO,EACZ,KAAK,SAAW;;;;;;;;8BAQU,KAAK,MAAK;oDACY,EAAW;yCACtB,KAAK,MAAK;0BACzB,EAAW;;;sCAGC,KAAK,MAAK;;;;;;;;;;;;;;;yBAevB,EAAK,oBAAoB,EAAI;;;;;;;yCAOb,EAAK;0BACpB,EAAI;;;;0CAIY,EAAI;;;;;;;;;;;;;;MAgB9C,MAAA,MC/FA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAgB,EAAc,EAChD,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;gCAQY,KAAK,YAAY,GAAE;gCACnB,KAAK,YAAY,GAAE;;;;;;;;;;;;;;;iCAelB,EAAG;;;;;;;;;;;yBAWX,EAAG,UAAU,EAAG;;;6DAGoB,EAAI;;;;;;;;;;;;;;;;;;;;;0CAqBvB,EAAW;;;MAKrD,MAAA,MC/GA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAY,EAAwB,EAAuB,EACjE,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAcnB,EAAqB;kBACnC,EAAc;gDACgB,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;kDACJ,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;gCAOvB,EAAS;;;;qCAIJ,EAAoB;;;;;;;;MAUzD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EACF,EAAuB,EAAwB,EAAuB,EAC1E,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;gCAiBhC,EAAoB;mBACjC,EAAa;gDACgB,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;kCAOvB,EAAS;;;;;;yBAMlB,EAAqB,MAAM,EAAoB;yBAC/C,EAAoB;;;;;;;;;MAW7C,MAAA,MChLA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkC,EAClC,EAAoB,EAAoB,EACxC,EAA2B,GAD3B,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IACxC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAT/B,KAAA,cAAgB,CAAC,UAAW,WAC5B,KAAA,aAAe,GACf,KAAA,aAAe,GAQb,KAAK,YAAc,EAEnB,GAAM,GAAY,EAAa,EAAO,GAAK,EAAO,GAC5C,EAAwB,KAAK,KAAK,EAAY,GAE9C,EAAU,EAAa,cAAgB,cACvC,EAAU,EAAa,cAAgB,cACvC,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAC1D,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAE5D,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;sCAEa,EAAqB;;;;8BAI7B,EAAqB;sCACb,EAAO;sCACP,EAAO;;;;uBAItB,EAAS,GAAE,MAAM,EAAS,GAAE;uBAC5B,EAAS,GAAE,MAAM,EAAS,GAAE;;;;;;;;;UASzC,EAAc;;UAEd,EAAsB;;;;MAMhC,MAAA,MChGA,AAoBA,GAAA,IAAA,WAQE,WAAY,EAAmB,EAAqB,GAPpD,KAAA,cAAgB,CAAC,SAQf,KAAK,YAAc,CAAC,EAAW,GAE/B,KAAK,SAAW;;;;;;;;;;8BAUU,GAAc,GAAC;;;;;;;;;;0BAUnB,GAAc,GAAC;;MAKvC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAM,mBAAmB,EAAc,SAExD,EAAM,GAAG,UAAU,EAAK,QAAS,KAGvC,KChEA,AAmBA,GAAA,IAAA,WAQE,WACI,EAAoB,EAAe,EAAiB,GARxD,KAAA,cAAgB,CAAC,WASf,KAAK,YAAc,CAAC,EAAY,GAEhC,KAAK,SAAW;;;;8BAIU,EAAQ,YAAY,EAAO;;;MAKzD,MAAA,MCxCA,AAqBA,GAAA,IAAA,WAOE,WACI,GAPJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEzB,GAAI,IAAS,EACX,KAAK,SAAW;;;;aAMhB,GAAM,GAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EACF,GAAwB,EAAM,EAAa,GACzC,EAAQ,GACV,EAAM,EAAY,EAAY,OAAS,GACvC,EAAY,EAAY,OAAS,GAAI,GACnC,EAAS,GAAU,EAAa,GAEtC,KAAK,SAAW;;YAEV,EAAK;;eAEF,EAAoB;;;cAGrB,EAAK;;6BAEU,EAAM;;;SAMnC,MAAA,MAEA,YAA4B,EAAc,GAGxC,OAFM,GAAS,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAG1B,OAFI,GAAW,KAAQ,EAAI,IAAM,OAAK,KAAK,KAAQ,EAAI,IAAM,OAEpD,EAAI,EAAG,EAAI,EAAM,IACxB,EAAW,EAAK,EAAK,OAAS,EAAI,GAAE,IAAM,EAG5C,EAAO,KAAK,GAGhB,MAAO,GAGT,YACI,EAAc,EAAiB,GACjC,GAAI,IAAS,EACX,MAAO,QAAQ,EAAM,GAIvB,OADI,GAAO,GACF,EAAI,EAAO,EAAG,EAAI,EAAM,IAC/B,GAAW,EAAK,GAAE,OAAO,EAAM,GAC/B,AAAI,EAAI,EAAO,GACb,IAAQ,MAIZ,MAAO,GAGT,YACI,EAAc,EAAc,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,GAGT,GAAM,GAAY,EAAK,MAAM,IAE7B,MAAO;cACK,EAAU,GAAE;cACZ,EAAU,GAAE;;;;0BAIA,EAAI;0BACJ,EAAI;IAI9B,YAAmB,EAAiB,GAClC,GAAM,GAAO,EAAM,OACb,EAAe,GAAmB,EAAM,GAC9C,MAAI,KAAS,EACJ;wBACa,EAAM,GAAE;kBAIvB,QAAQ,EAAa,GAAE;8BACF,EAAa,GAAE;8BACf,EAAa,GAAE;uCACN,EAAa,GAAE,ICrItD,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAClB,GANJ,KAAA,cAAgB,CAAC,KAOf,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KACnD,GAAM,GAAO,EAAO,OACd,EAAO,GAAkB,GAEzB,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EACF,CAAC,YAAa,YAAa,YAAa,aAAa,MAAM,EAAG,GAElE,GAAI,IAAS,GACX,KAAK,SAAW;sBACA,EAAK;oBACP,EAAG;;;;;8BAKO,EAAa;;;;;QAMrC,OAEF,KAAK,SAAW;QACZ,EAAI,YAAY,EAAI,IAAI,EAAK;QAC7B,EAAI,UAAU,EAAI,IAAI,EAAG;;;UAGvB,EAAI;;4BAEc,EAAa;;YAE7B,EAAI;2BACW,EAAc;;;MAKzC,MAAA,MCrEA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAClB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KA4BnD,OA3BM,GAAO,EAAO,OACd,EAAQ,GAAkB,GAE1B,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EAAS,GAAY,KAAM,GAC3B,EAAS,GAAY,SAAU,GAC/B,EAAY,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAC1D,EACF,IAAS,EAAI,SAAW,QAAQ,EAAO,MAAM,IAAI,OAAM,IAErD,EAAiB,CAClB,EAAK,mBAAuB,EAAO,EAAO,GAAE;YACzC,EAAM;QAEZ,IAAS,EAAI,GAAK;;SAEf,EAAO,EAAO,GAAE;YACb,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAAE,MACtD,IAAS,EAAI,GAAK,KAAK,EAAO,EAAO,GAAE;cAC/B,EAAM,OAGV,EAAc,IAAS,EACzB,0BACA,6DACA,EAAW,GACN,EAAI,EAAG,EAAI,IAAS,EAAI,EAAI,EAAG,EAAI,EAAG,IAC7C,GAAY;UACR,EAAe,GAAE;cACb,EAAW;mBACN,EAAC,aAAa,EAAa;;YAElC,EAAK;mBACE,EAAC,uBAAuB,EAAO,OAAM,MAAM,EAAS;;QAInE,GAAa,IAAS,EAAI,KAAO,KAEjC,KAAK,SAAW;cACN,EAAK,YAAY,EAAK,IAAI,EAAK;cAC/B,EAAK,UAAU,EAAK,IAAI,EAAG;;;UAG/B,EAAK;;UAEL,EAAQ;;;MAKlB,MAAA,MCrFA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MACzB,EAA0B,cAAc,EAAS,SAAQ,YAC3D,EAAS,QAAO,YAAY,EAAS,WAAU,OAC7C,EACF,SAAS,EAAS,QAAO,YAAY,EAAS,WAAU,OAExD,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;sCACgB,EAAY,KAAK,EAAW;mCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBnB,EAAqB;sBACjC,EAAc;;;kCAGF,EAAS,SAAQ;;;;oCAIf,EAAoB;wBAChC,EAAa;;;oCAGD,EAAS,QAAO;;;;;;;;;;0BAU1B,EAAS;;;mCAIzB,GAAoB,EAAsB,EACA,EACvB,QAAQ,EAAoB,SAAO;;;;;;QAO1D,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;oCACgB,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;0CACT,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;kCAkBZ,EAAmB;;;;gCAIrB,EAAqB;oBACjC,EAAc;;;gCAGF,EAAS,SAAQ;;;;kCAIf,EAAsB;uCACjB,EAAa;;;;yCAIX,EAAa;6CACT,EAAa;6CACb,EAAa;;;cAG5C,EAAa;;;gCAGK,EAAsB;gBACtC,KAA6B,GAAC;;;;;;;;cAQhC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;;;;;cAKxC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;6CACT,EAAa;;;;cAI5C,EAAa;;;oBAGP,EAAW;;MAI/B,MAAA,MAEA,GAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MAE3B,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;;oBAEF,EAAW,KAAK,EAAY,KAAK,EAAW;mCAC7B,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBhC,EAAoB;sBAChC,EAAa;;;kCAGD,EAAS,QAAO;;;;oCAId,EAAqB;wBACjC,EAAc;;;oCAGF,EAAS,SAAQ;;;;sCAIf,EAAoB;0BAChC,EAAa;;;sCAGD,EAAS,QAAO;;;;;;;;;;4BAU1B,EAAS;;;qCAI3B,GACK,EACI,cAAc,EAAS,QAAO,YAC1B,EAAS,SAAQ,YAAY,EAAS,QAAO,YAC7C,EAAS,WAAU,QACvB,UAAU,EAAS,SAAQ,YACvB,EAAS,QAAO,YAAY,EAAS,WAAU,QACxD,QAAQ,EAAqB,MAAM,EAAoB;6BACxC,EAAoB,SAAO;;;;;;;QAQlD,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;;gBAEJ,EAAW,KAAK,EAAY,KAAK,EAAW;iCAC3B,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;;kCAmBZ,EAAmB;;;;gCAIrB,EAAoB;oBAChC,EAAa;;;gCAGD,EAAS,QAAO;;;;kCAId,EAAqB;oBACnC,EAAc;;;kCAGA,EAAS,SAAQ;;;;oCAIf,EAAsB;yCACjB,EAAa;;;;+CAIP,EAAa;mDACT,EAAa;mDACb,EAAa;;;gBAGhD,EAAa;;;kCAGK,EAAsB;kBACtC,KAA6B,GAAC;;;;;;;;gBAQhC,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;;;;;gBAK5C,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;mDACT,EAAa;;;;gBAIhD,EAAa;;;sBAGP,EAAW;;;MAKjC,MAAA,MCvcA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,OAAQ,EAAA,EAAA,QACtC,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAI,GAAsB,MACtB,EAAY,GAEhB,AAAI,IAAe,OACjB,EAAsB,MACjB,AAAI,IAAe,MAExB,GAAsB,cACtB,EAAY,OACP,AAAI,IAAe,OAExB,GAAsB,eACtB,EAAY,OAGd,GAAI,GAAiB,EAAU,IAAI,EAAU,IAAI,EAAU,qEAG3D,AAAI,IAAe,MACjB,EAAc,WACT,AAAI,IAAe,OACxB,EAAc,YACT,AAAI,IAAe,MACxB,EAAc,WACT,AAAI,IAAe,OACxB,GAAc,YAGhB,GAAM,GAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEzC,EAAgB;YACZ,KAAe,OAAK;;mBAEb,KAAe,QAAM;;;;wBAIhB,EAAS;;MAIzB,EAAU,OAEd,AAAI,IAAe,MACjB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SACL,AAAI,IAAe,OACxB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SAGZ,GAAI,GAAmB,GACvB,AAAI,EAAS,EAAa,GACxB,GAAmB;oCACW,EAAM;;;SAKtC,KAAK,SAAW;0CACsB,EAAmB;;;;UAInD,EAAgB;;;;;;;;kCAQQ,EAAU;;kCAEV,EAAmB;;;;;;8BAMvB,EAAqB;;YAEvC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;YAC/B,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCvKA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAuC,GANnD,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAGnB,OADI,GAAW,GACN,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAS,eACb,AAAI,EAAI,IAAM,GACZ,IAAU,kBAEZ,AAAI,EAAI,GACN,IAAU,kBAGZ,GAAY;UACR,EAAM;UACN,GAAI,EAAI,0CAA4C,IAAE;;;;;;mBAM7C,EAAC;;UAEV,GAAI,EAAI,IAAM,IAAE;QAItB,KAAK,SAAW;QACZ,GAAuB,GAAW;QAClC,GAA+B,GAAY;;;;;;;;qBAQ9B,EAAY,GAAE;qBACd,EAAY,GAAE;;UAEzB,EAAQ;;;;MAMlB,MAAA,MAEA,YAAgC,GAC9B,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;QAED,EAAsB;;;ICnF9B,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;;sDAMY,GAAU,GAAC;;;;;;qDAMZ,GAAS,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAgC/D,MAAA,MC1IA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MA+B/D,MAAA,MC3EA,AAmBA,GAAA,IAAA,WAOE,WACI,EAA8C,EAC9C,EAAkB,GARtB,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;uCACxB,EAAQ;;;;;;;;;;;;;;;;;;;;;;gCAsBf,GAAQ,GAAC;uCACF,GAAW,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAkDnD,MAAA,MCzHA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;sBAKpB,EAAe,GAAE;sCACD,EAAe,GAAE;;;wBAG/B,EAAe,GAAE;wCACD,EAAe,GAAE;;;4BAG7B,EAAO;kBACjB,EAAY;;;;4BAIF,EAAM;kBAChB,EAAY;;;;;;;;;;;;MAc9B,MAAA,MC/HA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAI5C,EAAY,EAAe,MAAQ,MAEzC,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;8DAaD,EAAS;;;;;;MAQvE,MAAA,MCpEA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAI5C,GAFA,KAAK,YAAc,EAEf,IAAS,GACX,KAAK,SAAW;;;2BAGK,EAAO,GAAE;;QAG9B,OAEF,GAAM,GAAa,SAAC,GAClB,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,aAAa,EAAC,QAE5B,UAAU,EAAC,KAEd,EAAW,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,KAAI,KAAK,KACpD,EAAO,GAAkB,GAE/B,KAAK,SAAW;;UAEV,EAAI;yBACW,EAAQ;;MAIjC,MAAA,MC1DA,AAqBA,GAAA,IAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAE5C,KAAK,YAAc,EACnB,GAAM,GAAW,GAAY,KAAM,GAC7B,EACC,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACrD,EAAa,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACjE,EAAO,GAAkB,GAC/B,AAAI,IAAS,EACX,KAAK,SAAW;;;;uCAIiB,EAAO,GAAE;cAClC,EAAO,GAAE;eACR,EAAU;2CACkB,EAAO,GAAE;kBAClC,EAAO,GAAE;;;;QAMrB,KAAK,SAAW;;YAEV,EAAI;;uBAEO,EAAK,EAAS,SAAQ;eAC9B,EAAU;yBACA,EAAK,EAAS,SAAQ;;eAEhC,EAAO;yBACG,EAAK,EAAS,SAAQ;iBAC9B,EAAU;2BACA,EAAK,EAAS,SAAQ;;;;;MAQ7C,WAAc,GACZ,MAAO,GAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QAChD,EAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAoB,GAClB,GAAM,GAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,EAAG,KACnD,EAAW,EAAc,KAAK,KAC9B,EAAY,EAAc,MAAM,IAAI,KAAK,KAC/C,MAAO,mBAAmB,EAAQ,WAAW,EAAS,KAGxD,WAAoB,EAAW,GAC7B,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,MAAM,EAAU,GAAE,OAE9B,GAAG,EAAU,IAI5B,MAAA,MC7GA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAoB,EAAkB,EACtC,EAAqB,EAAmB,EACxC,GAPJ,KAAA,cAAgB,CAAC,UAAW,UAAW,gBAQrC,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAClC,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,QAElB,GAAM,GAAiB,cAAc,EAAa,IAE9C,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,gBAElB,GAAM,GAAiB,cAAc,EAAa,IAE5C,EAAe,EAAW,EAAI,aAAe,UACnD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,EAAO;;;YAG7C,EAAK;;;gCAGe,EAAU;;kCAER,EAAQ;kCACR,EAAc;0CACN,EAAY;;;uBAG/B,EAAc;;;;;;QAQrC,MAAA,MCvEA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,IAAK,cAOpB,GAAM,GAAa,EAAU,WACvB,EAAY,EAAU,UACtB,EAAS,EAAU,OACnB,EAAc,EAAU,YACxB,EAAU,EAAc,KAAK,KAAK,EAAS,GACjD,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAM,GAAsB,MACtB,EAAc,WAEd,EAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEvC,EAAgB;;MAIlB,EAAwB,GAC5B,AAAI,EAAS,EAAa,GACxB,GAAwB;oCACM,EAAM;;;SAMtC,GAAI,GAA4B,GAChC,AAAI,EAAS,EAAa,GACxB,GAA4B;oCACE,EAAM;;;SAMtC,KAAK,SAAW;0CACsB,EAAmB;;;UAGnD,EAAqB;;;;;UAKrB,EAAyB;;;;;;;;;YASvB,EAAW,cAAc,EAAU;wDACS,EAAW;;;;8BAIrC,EAAqB;;;;;;;;;;;;;;;;YAgBvC,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;;;;;;;;;;;;;;;;;YAiB/B,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCjKA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAe,EAAiB,GAJ5C,KAAA,cAAgB,CAAC,IAAK,IAAK,KAKzB,KAAK,YAAc,EAEnB,GAAI,GACA,EACJ,GAAI,EAAO,EACT,KAAM,OAAM,kBAAkB,EAAI,yBAGpC,GAAI,IAAS,EACX,EAAW,QACX,EAAU,aAKV,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAClD,EAAa,GACb,EAAc,GACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAY,KAAK,GAAG,EAAc,IAClC,AAAI,EAAI,GACN,EAAW,KAAK,GAAG,EAAc,IAGrC,EAAU,EAAW,OACrB,EAAW,EAAY,OAGzB,GAAM,GAAQ,GAAkB,GAEhC,KAAK,SAAW;;UAEV,EAAK;4BACa,EAAO;;2BAER,EAAQ;;2BAER,EAAQ;;;MAKnC,MAAA,MCjEA,AAqBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,UASf,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAc,qBAAqB,KAAK,KAAI,KAC5C,EAAe,GAAU,KAAK,MAEhC,EACE,EAAW,EAAS,IAAI,SAAC,EAAG,GAChC,MAAO,aAAa,GAAO,GAAE,YAAY,EAAC,cAAc,GAAO,GAAE,MAEnE,EAAO;UACD,EAAK;UACL,EAAK;UACL,EAAS,KAAK;GAAK;QAEzB,KAAK,SAAW;QACZ,EAAW;;UAET,EAAI;8BACgB,EAAY;;MAKxC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KAEM,GAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,YAAmB,GACjB,GAAI,IAAS,EACX,MAAO,YACF,GAAI,GAAQ,EACjB,MAAO,IAAO,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAA,aAAe,IAAG,KAAK,KAE7D,KAAM,OAAM,oBAAoB,EAAI,yBCpFxC,AAsBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,UACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAS,GAAY,SAAU,KAAK,MACpC,EAAY,GAAY,YAAa,KAAK,MAE1C,EACF,KAAK,OAAS,EAAI,YAAc,QAAQ,EAAU,MAAM,IAAI,OAAM,IAChE,EACF,wBAAwB,EAAU,OAAM,MAAM,EAAS,IACrD,EAAW;mBACF,EAAU;cACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;YACnB,EAAU,KAAK,KAAO,GAAE;;MAG1B,EAAW,KAAK,OAAS,EAAI,GAAK;UAClC,EAAO,KAAK,KAAO,GAAE;cACjB,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;gBACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;cACpD,EAAU,KAAK,KAAO,GAAE;uBACf,EAAU;;;MAKvB,EAAiB,KAAK,MAAQ,EAChC;cACM,EAAK,IAAI,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,SAAS,EAAC,MAAK,OAAM,KAC3D,EAAS,IAAI,SAAC,EAAG,GAAM,MAAG,GAAU,GAAE,MAAM,EAAO,GAAE,YAAY,EAAC,OAC7D,KAAK;GACd,KAAK,SAAW;0BACM,KAAK,KAAI;;UAEzB,EAAK;UACL,EAAK;UACL,EAAc;;UAEd,EAAQ;UACR,EAAQ;;;MAMhB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KCtGA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAiB,EAAmB,GAJhD,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAK,OACZ,EAAa,GAAkB,EAAK,QACpC,EAAQ,GAAkB,EAAK,QAEjC,EAAY,GAChB,GAAI,IAAS,EACX,EAAY,gCAEZ,GAAI,GAAa,EACjB,EACI,EAAK,IAAI,SAAC,EAAG,GACP,WACO,EAAK,SAAW,EACnB,oBAAoB,EAAC,aAAa,EAAC,IACnC,UAAU,GAAa,GAAC,eAAe,EAAC,aAAa,EAAC,MAE3D,KAAK,KAGhB,KAAK,SAAW;QACZ,EAAU,YAAY,EAAU,IAAI,EAAK;QACzC,EAAU,cAAc,EAAU,IAAI,EAAO;;;UAG3C,EAAK;yBACU,EAAS;;MAIlC,MAAA,MCxDA,AAuBA,GAAA,IAAA,WAUE,WAAoB,GAAA,KAAA,MAAA,EATZ,KAAA,gBAAkB,EAClB,KAAA,gBAAkB,EAClB,KAAA,mBAAqB,EACrB,KAAA,cAAgB,EAEhB,KAAA,aAAkD,GAClD,KAAA,WAAa,GACb,KAAA,aAAkD,GAI1D,SAAA,UAAA,eAAA,SACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAkC,EAAO,GAE3D,EAAW,GAAuB,EAAS,EAAiB,GAClE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAEhC,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAS,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cACpD,GAEJ,GAAI,KAAK,aAAa,GAAU,OAAS,GACvC,KAAK,kBACL,KAAK,kBACL,KAAK,eAAiB,EACtB,KAAK,MACL,GAAM,GAAa,KAAK,aAAa,GAAU,QAC/C,YAAK,aAAa,GAAU,KAAK,GAC1B,EAGT,GAAI,GACJ,MAAI,KAAoB,GAAoB,mBAC1C,EAAa,KAAK,MAAM,0BAA0B,EAAQ,GAAI,EAAQ,IACjE,AAAI,IAAoB,GAAoB,mBACjD,EACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,IAC/D,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AACH,IAAoB,GAAoB,0BAC1C,GACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,KAEtE,KAAK,aAAa,GAAU,KAAK,GAEjC,KAAK,kBACL,KAAK,oBAAsB,EAC3B,KAAK,MAEE,GAGT,EAAA,UAAA,eAAA,SACI,EAAuB,EACvB,EAA8B,GAChC,GAAI,KAAK,cAAgB,KAEvB,OAEF,GAAM,GACF,GAAkC,EAAgB,GAChD,EAAW,GAAuB,EAAO,EAAiB,GAChE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAO,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cAClD,GACE,EAAqB,EAAA,MAAM,IAAI,kCACrC,AAAI,IAAuB,IACvB,KAAK,mBAAqB,EAC5B,MAAK,MAAM,oBAAoB,GAC/B,KAAK,oBAAsB,GAE3B,MAAK,aAAa,GAAU,KAAK,GACjC,KAAK,kBACL,KAAK,eAAiB,GAGxB,KAAK,kBAEL,GAAM,GAAU,KAAK,aAAa,GAC5B,EAAW,EAAQ,QAAQ,GACjC,GAAI,EAAW,EACb,KAAM,IAAI,OACN,4EAGN,EAAQ,OAAO,EAAU,GACzB,KAAK,OAGC,EAAA,UAAA,IAAR,WACE,GAAI,CAAC,KAAK,WACR,OAEF,GAAM,GAAQ,KAAK,gBAAkB,KAAK,gBAC1C,QAAQ,IACJ,YAAgB,KAAK,gBAAe,MAAM,KAAK,gBAC/C,IAAI,EAAK,KACb,GAAM,GAAY,KAAK,cAAgB,KAAK,mBAC5C,QAAQ,IAAI,oBAAoB,KAAK,oBACrC,QAAQ,IAAI,iBAAiB,KAAK,cAAa,KAC3C,KAAK,MAAM,IAAM,GAAU,OAGjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGd,OAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,+CAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,cAAgB,KAEvB,OAEF,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,KAAK,aAAe,KACpB,KAAK,aAAe,KACpB,KAAK,gBAAkB,EACvB,KAAK,gBAAkB,EACvB,KAAK,mBAAqB,EAC1B,KAAK,cAAgB,GAEzB,KAEA,YACI,EAA2B,GAE7B,GAAM,GAAQ,EACd,GAAI,IAAmB,EAAM,KAC3B,MAAO,GACF,GAAI,IAAmB,EAAM,KAClC,MAAO,GACF,GAAI,IAAmB,EAAM,QAClC,MAAO,IACF,GAAI,IAAmB,EAAG,KAC/B,MAAO,IACF,GAAI,IAAmB,EAAM,QAClC,MAAO,GAET,KAAM,IAAI,OAAM,2BAA2B,eAIzC,EAAyB,EACzB,EAA2B,EAC3B,GAMF,GAAM,GACF,GAAiC,EAAiB,GAElD,EACJ,GAAI,GACI,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAEpB,EAAc,EAAc,OAGtB,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,EAAc,EAAQ,EAGxB,GAAM,GAAkB,GAA0B,EAAI,GACtD,MAAO,GAAc,EAGvB,YACI,EACA,GACF,OAAQ,OACD,IAAoB,mBACvB,MAAO,IAAwC,OAC5C,IAAoB,mBACvB,MAAO,IAA+C,OACnD,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,yBACvB,MAAO,IAA+C,WAEtD,KAAM,IAAI,OAAM,iCAAiC,IAIvD,YAAwC,GAEtC,MAAI,GAAA,MAAM,QAAQ,gCACZ,EACK,GAAoB,mBAEtB,GAAoB,iBAGzB,EACK,GAAoB,mBAEtB,GAAoB,iBAG7B,YACI,EAA8B,GAChC,GAAI,IAAmB,GAAa,OAClC,MAAO,IAAoB,mBACtB,GAAI,IAAmB,GAAa,QAAU,GAAkB,KACrE,MAAO,IAA+B,GACjC,GACH,IAAmB,GAAa,UAChC,IAAmB,GAAa,OAClC,MAAO,IAAoB,yBAE7B,KAAM,IAAI,OAAM,gCAAgC,GAGlD,YACI,EAAgC,EAChC,GACF,MAAU,GAAa,GAAE,IAAI,EAAa,GAAE,IAAI,EAAe,IAAI,ECxRrE,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,GAEpC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,GAErC,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,GACvB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,iBAAiB,EAAI,yBAEnC,GAAI,IAAS,EACX,MAAO,eAAe,EAAO,GAAE,IAMjC,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,UAAW,WAE7D,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAa,KAAK,QAAQ,EAAc,GAAE,KAAK,EAAO,GAAE,KAE1D,MAAO,GAAa,OC5DtB,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MAEM,GAAoB,0BAEb,GAAS,YAET,GAAM,iBAEN,GAAO,GAAoB;;EAI3B,GAAQ,GAAoB;;EAI5B,GAAM,0CAEN,GAAO;;;uBAGG,EAAA,aAAa,gBAAe;kBACjC,EAAA,aAAa,WAAU;;cAIpB,GAAA,MAAA,KAAA,QAAA,GAAA,GACZ,GAAoB;mCACM,EAAK;KAIjC,GAAM,IAAM,aAEN,GAAO,kBAEP,GAAQ,mBAER,GAAO;;;EAKP,GAAS,0BAET,GAAS,0BAET,GAAY,wCAEZ,GAAQ;;;;;;;;;;;;;;;EAiBR,GAAM,iBAEN,GAAQ,uBAER,GAAM;kBAGN,GAAQ,uBAER,GAAO,kBAEP,GAAQ,yBAER,GAAU,sCAeV,GAAW;;;;;;;;;;;;;;;;;;;;EAsBX,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;EAI3B,GAAO;;;EAKP,GAAO;;;EAKP,GAAO;;;EAKP,GAAQ,GAAoB,qCAE5B,GAAQ,GAAoB;;sCAI5B,GAAQ,GAAoB;;+CAI5B,GAAM;;;;cAIL,EAAA,aAAa,MAAK;eACjB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;;;;;;EAQrB,GAAa,kBAEb,GAAc,6BAEd,GAAS,wBAET,GAAQ,YC9NrB,AAmBO,GAAM,IAAS,YAET,GAAM;;;;;;;;;EAWN,GAAO;;;;;;;;;;EAYP,GAAQ;;;;;;;;;;EAYR,GAAM;;;;;;;;;EAWnB,GAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MCzFA,AAqBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEnB,EAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EAAe,GAAgB,EAAM,GACrC,EAAY,EAAS,MAAM,IAC3B,EAAS,GAAQ,EAAI,KAAO,QAAQ,EAAU,KAAK,KAAI,IAE7D,KAAK,SAAW;;UAEV,EAAK;kCACmB,EAAY;;4CAEF,EAAM;;MAIlD,MAAA,MC/CA,AA2BO,GAAA,IAAA,EAAA,aAAA,aACD,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAiFlB,GAAkB,KAClB,GAAkB,KA4BzB,GAAuE,eAE9C,GAC7B,MAAI,KAAgB,KAGpB,IAAa,GAAgB,IACtB,GAAa,GAGtB,YACI,EAAqC,GACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACnC,IAAe,SACjB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,OACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,MACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GAET,KAAM,IAAI,OAAM,cACZ,EAAU,oDAKhB,GAAM,IAA6B,IAK7B,GAAyB,IAC/B,cACE,MAAI,GAAA,MAAM,OAAO,QAAU,KAClB,KAED,EAAA,MAAM,OAAO,OAAO,OAAS,EAAA,MAAM,OAAO,OAAO,MACjD,OAAO,iBACX,GAAyB,KAAO,KAM/B,GAAM,IAA8B,mBAEL,GAAA,EAAA,GAkCpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GA/BM,EAAA,YAAc,GAAI,SAGlB,EAAA,gBAAkB,GAAI,SAGtB,EAAA,aAAe,GAAI,SACnB,EAAA,cAAgB,EAOhB,EAAA,aAAe,EAEf,EAAA,eAAiB,EAUjB,EAAA,kBAAoB,GACpB,EAAA,sBAAwB,GAgXxB,EAAA,eAAiB,EAq+DjB,EAAA,SAAW,GAj1Eb,CAAC,EAAA,MAAM,QAAQ,aACjB,KAAM,IAAI,OAAM,yCAGlB,GAAI,GAAS,MACX,GAAM,GAAK,GAAgB,EAAA,MAAM,UAAU,kBAC3C,EAAK,YAAc,GAAe,EAAA,MAAM,UAAU,kBAClD,EAAK,MAAQ,GAAI,IAAa,GAC9B,EAAK,OAAS,EAAG,OACjB,EAAK,oBAAsB,OAE3B,GAAK,MAAQ,EACb,EAAK,YAAc,GACnB,EAAK,oBAAsB,GAC3B,EAAK,OAAS,EAAM,GAAG,OAEzB,SAAK,eAAiB,GAAI,IAAe,EAAK,OAC9C,EAAK,mBAAqB,KAE1B,EAAK,QAAU,GAAI,GAAA,YAAY,EAAM,EAAA,YAGvC,SAAA,UAAA,WAAA,WACE,MAAO,MAAK,QAAQ,aACf,MAAK,WAAa,KAAK,WAAW,aAAe,GAClD,KAAK,gBAGX,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAK5C,GAJA,AAAI,GAAA,MAAM,QAAQ,mCACd,EAAA,MAAM,QAAQ,WAChB,KAAK,uBAAuB,GAE1B,IAAU,aAAe,GAAU,KACrC,KAAM,IAAI,OACN,yEAGN,GAAM,GAAS,GACf,YAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,IAC1D,GAIT,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,YAIV,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,aAIZ,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAK3D,GAHA,AAAI,EAAA,MAAM,QAAQ,UAChB,KAAK,uBAAuB,GAE1B,IAAU,YACZ,KAAM,IAAI,OACN,yEAGN,KAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,KAGnE,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAc,KAAK,QAAQ,IAAI,GAErC,EAAY,WAEZ,AAAI,EAAY,SAAW,GACzB,KAAK,YAAY,KAKvB,EAAA,UAAA,SAAA,SAAS,GACP,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,eAAgB,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SAKpD,GAAI,GAAS,MACX,GAAI,GAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEtC,GAAM,GACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,SAAS,EAAI,QAC/B,YAAK,8BAA8B,GAC5B,EAET,GAAI,GAAU,KACZ,MAAO,MAAK,qBAAqB,GAEnC,GAAI,IAAU,SACZ,MAAO,GAET,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GACJ,GAAI,IAAU,aACZ,GAAM,GAAa,EAAe,KAAK,WACjC,EAAa,EAAe,KAAK,WACvC,EAAS,EAAA,aAAa,uBAAuB,EAAY,OAEzD,GAAS,KAAK,qBAAqB,GAGrC,MAAI,IACF,MAAK,gBAAkB,EAAA,KAAK,MAAQ,GAE/B,KAAK,qBAAqB,EAAQ,IAGrC,EAAA,UAAA,KAAN,SAAW,6IACT,GAAI,KAAK,YAAY,IAAI,GACjB,SAAc,KAAK,YAAY,IAAI,GACzC,CAAA,EAAO,GAAI,SAAoB,SAAA,GAAW,MAAA,GAAY,KAAK,MAQ7D,GANM,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAyD,EAAO,OAAxD,EAAiD,EAAO,MAAjD,EAA0C,EAAO,MAA1C,EAAmC,EAAO,MAAnC,EAA4B,EAAO,eAAnB,EAAY,EAAO,SAKnE,GAAS,KACP,SAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEhC,EACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,KAAK,EAAI,QAC3B,KAAK,8BAA8B,GACnC,CAAA,EAAO,GAGT,GAAI,GAAU,KACZ,MAAA,CAAA,EAAO,KAAK,qBAAqB,IAGnC,GAAI,CAAC,EAAA,MAAM,QAAQ,iCACf,EAAA,MAAM,UAAU,mBAAqB,EACvC,KAAM,IAAI,OACN,sGAIF,GAAS,KAGb,AAAI,IAAU,aAAe,EAAA,MAAM,IAAI,2BAErC,GAAoB,KAAK,OAAO,GAC1B,EAAU,KAAK,QAAQ,IAAI,EAAkB,QAEnD,EAAS,GAAA,KAAK,OAAM,wBAAuB,MAAA,EAAA,CACvC,EAAQ,SAAO,OAAK,GAA0B,MAGpD,KAAK,YAAY,IAAI,EAAQ,IAEzB,IAAU,YAEZ,CAAA,EAAM,KAAK,MAAM,yBAFf,CAAA,EAAA,UAEF,EAAA,8BAKE,KAAU,YACD,CAAA,EAAM,QAAQ,IACrB,CAAC,EAAe,KAAK,OAAQ,EAAe,KAAK,UAFnD,CAAA,EAAA,UACI,SAAK,EAAA,OAEL,EAAa,EAAG,GAChB,EAAa,EAAG,GACtB,EAAO,EAAA,aAAa,uBAChB,EAA4B,gBAC3B,AAAI,GAAU,KACnB,EAAO,KAAK,qBAAqB,GAE3B,GAAO,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,MAAM,gCAAgC,EAAQ,qBAE5D,MAAI,IAAqB,MACvB,KAAK,8BAA8B,GAE/B,EAAY,KAAK,qBAAqB,EAAQ,GAE9C,EAAc,KAAK,YAAY,IAAI,GACzC,KAAK,YAAY,OAAO,GAGxB,EAAY,QAAQ,SAAA,GAAW,MAAA,GAAQ,KACvC,AAAI,KAAK,gBAAgB,IAAI,IAC3B,MAAK,gBAAgB,OAAO,GAC5B,KAAK,YAAY,GACjB,KAAK,kBAEP,CAAA,EAAO,SAGD,EAAA,UAAA,uBAAR,SAA+B,GAC7B,GAAI,GAAU,KACZ,OAEF,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,KACjC,GAAM,GAAM,EAAO,GACnB,GAAI,CAAC,GAA4B,GAC/B,KAAI,GAAA,MAAM,QAAQ,gCACV,MACF,aAAa,EAAG,iJAIhB,MAAM,aAAa,EAAG,4CAK1B,EAAA,UAAA,qBAAR,SAA6B,SACrB,EAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SACf,EAAO,EAAA,KAAK,cAAc,GAChC,GAAI,EAAA,MAAM,QAAQ,iCAChB,GAAM,GAAY,KAAK,OAAO,GACxB,EAAU,KAAK,QAAQ,IAAI,EAAU,QACrC,EAAO,GAAA,KAAK,OACA,gCAA+B,MAAA,EAAA,CAC5B,EAAQ,SAAO,OAAK,GAA0B,KACjD,SAAS,EAAG,GAE9B,YAAK,8BAA8B,GAE5B,EAGT,GAAM,GACF,EAAA,MAAM,QAAQ,eAAiB,IAAa,GAC1C,EACF,EAAyB,GAAwB,GAAS,EACxD,EAAU,EACZ,GAAI,IAAyB,GAC7B,GAAI,IAAmB,GACrB,EAAS,KAAK,gBAChB,EAAS,CAAC,CAAC,MAAO,EAAa,MAAK,EAAE,OAAM,IAAI,WAC9C,EAAU,KAAK,QAAQ,IAAI,EAAO,QAClC,EACF,KAAK,MACA,gDACG,EAAQ,QAAS,EAAQ,SAAS,GAAI,EAAQ,SAAS,IAC1D,SAAS,EAAG,GACrB,YAAK,8BAA8B,GAE5B,GAGH,EAAA,UAAA,KAAN,SAAW,qHACH,GAAkB,KAAK,aACvB,EAA+B,GAEjC,EAAgB,GACpB,AAAI,KAAK,oBAAsB,KAC7B,MAAK,mBAAqB,EAC1B,EAAgB,IAEhB,KAAK,aAAa,KAAK,GAEzB,KAAK,aAAe,EAEpB,IAGM,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,SACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OACpB,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,QACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OAE1B,KAAK,aAAe,EAEpB,AAAI,GACF,MAAK,mBAAqB,MAGtB,EAAuB,CAC3B,aAAc,KAAK,aACnB,eAAgB,KAAK,eACrB,SAAU,KACV,OAAQ,MAGN,EAAA,MAAM,UAAU,iDAAmD,EACpD,CAAA,EAAM,QAAQ,IAAI,IADjC,CAAA,EAAA,UACI,SAAW,EAAA,OAEjB,EAAI,SAAc,EAAA,KAAK,IAAI,GAC3B,EAAI,oBAAyB,WACzB,MAAA,GAAS,IAAI,SAAC,EAAG,GAAM,MAAC,CAAC,KAAM,EAA0B,GAAI,GAAI,KAC5D,IAAI,SAAA,GAAK,MAAG,GAAE,KAAI,KAAK,EAAE,KACzB,KAAK,oBAEd,EAAI,SAAc,CAChB,MAAO,8EAIX,YAAK,aAAe,EACpB,KAAK,eAAiB,EACtB,CAAA,EAAO,SAET,EAAA,UAAA,OAAA,WACE,MAAO,CACL,WAAY,GACZ,cAAe,KAAK,cACpB,uBAAwB,KAAK,eAAe,kBAC5C,kBAAmB,KAAK,eAAe,eAInC,EAAA,UAAA,WAAR,WACE,MAAI,GAAA,MAAM,UAAU,iDAAmD,EAC9D,KAAK,MAAM,aAEb,CAAC,QAAS,EAAA,KAAK,MAAO,MAAO,OAG9B,EAAA,UAAA,SAAR,SAAiB,GACf,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,MAAK,MAAM,WACJ,GAER,GAAwB,MAAQ,EAAA,KAAK,MAC/B,IAGK,EAAA,UAAA,aAAd,SAA2B,4EACzB,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,CAAA,EAAO,KAAK,MAAM,uBAAuB,IAErC,GAAa,EACnB,CAAA,EAAO,EAAW,MAAQ,EAAW,eAKvC,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,gBAAgB,IAAI,GAC3B,OAEF,GAAI,KAAK,YAAY,IAAI,IACvB,KAAK,gBAAgB,IAAI,GACzB,KAAK,iBACL,OAGF,GAAI,CAAC,KAAK,QAAQ,IAAI,GACpB,OAGF,KAAK,eAAe,GACb,GAAA,GAAA,KAAA,QAAA,IAAA,GAAA,eACP,AAAI,GAAkB,MACpB,GAAe,KAAK,UACpB,EAAe,KAAK,WAEtB,KAAK,QAAQ,OAAO,IAGd,EAAA,UAAA,eAAR,SAAuB,GACf,GAAA,GAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAE5C,EAAM,GAAS,EAAM,YAAc,EACnC,EAAW,KAAK,aAAa,IAAI,GACvC,AAAI,EAAW,EACb,KAAK,aAAa,IAAI,EAAK,EAAW,GAEtC,MAAK,aAAa,OAAO,GACzB,AAAI,GAAW,MACb,MAAK,eAAiB,KAAK,aAAa,EAAU,GAClD,KAAK,eAAe,eAAe,EAAS,EAAU,EAAO,KAGjE,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,QAAU,KAClB,EAAQ,SAAW,KACnB,EAAQ,SAAW,GACnB,EAAQ,MAAQ,MAGlB,EAAA,UAAA,WAAA,SAAW,GACT,YAAK,YAAY,GACV,KAAK,QAAQ,IAAI,GAAQ,SAOlC,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,MAAK,QAAQ,IAAI,IAGlB,EAAA,UAAA,cAAR,WACE,MAAK,GAAA,MAAM,QAAQ,qBAInB,CAAI,KAAK,YAAc,MACrB,MAAK,WAAa,EAAA,SAAS,YAAY,QAGlC,KAAK,YAPH,MAiBX,EAAA,UAAA,mBAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAa,KAAK,gBACxB,MAAI,CAAC,KAAK,uBAAyB,GAAc,MAC/C,SAAQ,KACJ,6NAKJ,KAAK,sBAAwB,IAGxB,GAAc,MACjB,EAAO,MACH,SAAA,GAAS,MAAA,GAAK,QAAQ,IAAI,EAAM,QAAQ,SAAW,MAC/C,EAAA,KAAK,cAAc,EAAM,OAAS,KAGhD,EAAA,UAAA,gBAAA,WACE,MAAO,MAAK,OAGd,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,GAAM,GAAS,KAAK,WAAW,EAAK,MAAO,aACrC,EAAa,KAAK,QAAQ,IAAI,EAAO,QAI3C,SAAW,eAAiB,CAC1B,KAAM,EAAA,SAAS,KAAK,EAAK,SACzB,KAAM,EAAA,SAAS,KAAK,EAAK,UAGpB,GAET,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAExC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAGxC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAO,EAAM,EAAE,MAChE,EAAE,OACN,MAAO,MAAK,WAAW,EAAM,EAAE,MAAO,GAGxC,GAAI,EAAA,KAAK,cAAc,KAAU,EAC/B,MAAO,GAAA,OAAO,GAAI,EAAM,EAAE,OAErB,GAAA,GAAA,KAAA,QAAA,IAAA,EAAA,QAAA,SACD,EAAc,EAAA,WAAW,iBAAiB,EAAE,MAAO,EAAO,GAChE,GAAI,GAAY,CAAC,GACf,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAmB,GACvB,GAAI,IAAa,GACf,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,GAEhD,YAAK,YAAY,EAAE,QACZ,KAAK,aAAa,EAAG,EAAO,IAG7B,EAAA,UAAA,aAAR,SAAqB,EAAW,EAAiB,GAC/C,GAAM,GAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAI,KAAK,WAAW,EAAM,EAAE,OAC5B,EAAa,KAAK,QAAQ,IAAI,EAAE,QAEtC,OAAO,OAAO,EAAY,GAC1B,EAAW,MAAQ,EACnB,EAAW,MAAQ,EAAE,MACrB,GAAI,GAAa,EAAA,WAAW,kBAAkB,EAAO,EAAE,SACvD,AAAI,EAAS,OAGX,IAAc,EAAS,MAAM,YAE/B,EAAW,MAAQ,CACjB,WAAU,EAEV,WAAY,EAAS,OAAS,EAAS,MAAM,YAAc,EAAE,QAI/D,GAAM,GAAW,KAAK,aAAa,IAAI,EAAW,MAAM,aAAe,EACvE,YAAK,aAAa,IAAI,EAAW,MAAM,WAAY,EAAW,GAEvD,GAGT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GAD1C,GAAA,GAAA,KAEQ,EAAS,KAAK,mBAChB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,aAAa,EAAG,EAAO,EAAK,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAO,GAAI,GAGpB,GAAM,GAAU,GAAI,IAAoB,EAAO,EAAS,GACxD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAqB,EAAE,MAAO,GAClC,GAAI,IAAe,EAAE,MAAO,GAChC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,GAAI,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KAChC,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KACtC,MAAO,GAAA,QAAQ,KAAK,OAAO,EAAO,GAAO,KAAK,OAAO,EAAO,IAG9D,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAEjB,GAAI,EAAQ,OAAS,EAAA,MAAM,UAAU,iCACnC,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,OAAO,EAAQ,MAAM,EAAG,GAAW,GACnD,EAAY,KAAK,OAAO,EAAQ,MAAM,GAAW,GACvD,MAAO,MAAK,OAAO,CAAC,EAAU,GAAY,GAE5C,GAAI,EAAA,MAAM,QAAQ,gCAAkC,EAAQ,GAAG,KAAO,GACpE,GAAM,GAAU,GAAI,IAAoB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACnE,MAAO,MAAK,cAAc,EAAS,GASrC,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACtD,EACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAAK,GAAI,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,OAC3D,EAAU,GAAI,IAAc,EAAU,IAAI,SAAA,GAAK,MAAA,GAAE,SACjD,EAAc,KAAK,cAAc,EAAS,GAChD,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAsB,GAAtB,GAAA,GAAA,KACQ,EAAS,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,IAAI,KACtE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAE/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAM,GAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAA,EAAA,MAAC,EAAA,EAAA,GAIP,GAAK,KAAgB,GAAK,IAAgB,IACtC,EAAY,IACd,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAE1B,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAG1B,GAAM,GAAM,IAAgB,EAAI,EAAI,EAAE,KAAK,EAAO,EAAW,GACvD,EAAO,IAAgB,EAAI,EAAI,EAC/B,EAAM,IAAgB,EAAI,EAAE,KAAK,EAAO,EAAG,GAAa,EAC9D,MAAO,MAAK,SAAS,EAAK,GAAK,IAAI,EAAM,IAG3C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,GAC5D,MAAO,MAAK,cAAwB,EAAS,CAAC,EAAG,GAAI,IAGvD,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE7C,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAA,EAAA,MAAC,EAAA,EAAA,GAED,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,EACxD,EAAS,EAAiB,GACxB,EAAuB,CAAC,EAAG,GACjC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAwB,EAAS,EAAQ,IAGvD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aACd,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OACrD,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OAErD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAExD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAE/C,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,EAGT,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,EAAE,QAG/C,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAQ,EAAM,EAAO,GACnD,GAAI,IAAW,EAAE,MAAO,EAAQ,EAAM,EAAO,GACjD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAM,GACF,GAAI,IAAe,EAAW,MAAO,EAAa,EAAM,EAAO,GACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAY,EAAa,KAG/D,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAI,EAAE,QAAU,UACd,GAAM,GAAO,KAAK,SAAS,EAAE,QACvB,EAAc,EAAK,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,KAC9C,EAAM,EAAA,OAAO,EAAE,MAAO,EAAE,MAAO,GACrC,MAAO,IAAK,EAAK,GAEnB,GAAM,GAAU,GAAI,IAAY,EAAE,MAAO,GACzC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAU,GACxC,GAAI,IAAW,EAAE,MAAO,EAAU,GACtC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAAlD,GAAA,GAAA,KACQ,EAAS,KAAK,mBAChB,CAAC,EAAG,GAAU,WAAM,MAAA,GAAK,WAAW,OAAO,EAAG,EAAS,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAQ,KAAM,GACzD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAEV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAU,EAAE,QAAQ,GAAW,GAC1B,QAAQ,GACR,MAAM,EAAkB,IAGtC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAGV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAAyB,GACzB,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,EAAE,IAAI,GAEhB,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EAAe,EAAA,aAAa,oBAC9B,EAAQ,MAAO,EAAY,EAAM,IAE/B,EAAW,EAAA,UACb,EAAQ,QAAQ,GAChB,GACJ,MAAO,GAAA,QAAQ,EAAU,IAGnB,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,KAAK,KAAK,EAAS,GAC7B,EAAa,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,QAAO,GACpD,EAAU,GAAI,IAAc,EAAY,GACxC,EAAS,KAAK,cAAwB,EAAS,CAAC,GAAI,GAE1D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,OAAO,EAAQ,EAAY,IAGjC,EAAA,UAAA,UAAR,SACI,EAAa,EACb,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACrB,AAAI,GAAgB,MAClB,GAAY,EAAa,MAAM,GAC/B,EAAS,EAAa,MAAM,IAE9B,GAAM,GAAa,EAAA,aAAa,yBAAyB,GACnD,EAAa,CACjB,WAAU,EACV,OAAM,EACN,UAAS,EACT,QAAS,KAAK,KAAK,EAAS,IAExB,EACF,GAAI,IAAiB,EAAY,EAAY,GAAgB,MAC3D,EAAS,CAAC,GAChB,AAAI,GAAgB,MAClB,EAAO,KAAK,GAEd,GAAM,GAAS,KAAK,cAAwB,EAAS,EAAQ,SAE7D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,UAAU,EAAG,EAAY,IAG/B,EAAA,UAAA,gBAAR,SACI,EAAW,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MACtC,GAAM,GAAU,GAAgB,KAAO,EAAa,MAAQ,EAAE,MACxD,EAAS,EAAQ,EAAQ,OAAS,GAClC,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,GAAI,IAChB,EAAS,EAAY,EAAY,GAAgB,MAC/C,EAAS,GAAgB,KAAO,CAAC,GAAK,CAAC,EAAG,GAC1C,EAAS,KAAK,cAAsB,EAAS,EAAQ,SAC3D,MAAI,GAAO,OAAS,EAAE,KACb,KAAK,gBAAgB,EAAG,EAAY,GAEtC,GAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,MAAO,GAAa,QAAQ,IAGtD,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAC/D,GAAI,EACF,MAAO,GAGH,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,OAAQ,GAAa,QAAQ,IAGvD,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAI,GAAO,EACL,EAAc,EAAA,aAAa,mBAAmB,CAAC,GAAO,EAAE,MAC1D,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,EAAA,UAAU,EAAG,GACzB,EAAO,EAAA,aAAa,iBAAiB,EAAG,EAAE,MAAM,IAGlD,GAAM,GACF,GAAa,gBAAgB,EAAU,MAAO,EAAM,GAClD,EAAS,EAAA,KAAK,cAAc,CAAC,EAAU,MAAM,KAC7C,EAAM,EAAU,KAAK,GAAI,GACzB,EAAc,EAAA,WAAc,EAAE,OAChC,EACA,KAAK,aACG,EAAK,qBAAsB,EAAY,EAAa,GACvD,QAAQ,GACjB,MAAI,IAAe,MACjB,GACI,EAAA,UAAU,EAAQ,EAAA,aAAa,uBAAuB,KAErD,GAGD,EAAA,UAAA,aAAR,SACI,EAAa,EAAiC,EAC9C,EAAiB,GACnB,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EACF,GAAa,8BAA8B,EAAQ,GACjD,EAAY,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,YAAW,GACvD,EAAU,GAAI,IAAiB,EAAW,GAC1C,EACF,KAAK,cAAwB,EAAS,CAAC,EAAG,GAAa,GAE3D,MAAI,GAAO,MAAM,KAAO,EACf,EAET,GAAa,EAAA,MAAM,EAAG,GAAa,KAAK,CAAC,EAAS,IAC3C,KAAK,aAAa,EAAQ,EAAW,EAAY,EAAO,KAGzD,EAAA,UAAA,gBAAR,SAAwB,EAAW,EAAc,GAE/C,GAAM,GAAO,CAAC,GAId,GAHA,EAAA,aAAa,2BACT,MAAQ,EAAW,OAAO,GAAG,cAAgB,EAAW,MAAM,GAAI,EAClE,EAAE,MACF,CAAC,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAQ,GAC7C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,UAAU,EAAK,GAAY,QAAQ,GAEjD,MAAO,MAAK,gBAAgB,EAAG,IAGjC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,GAAI,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,kDAAkD,GAAE,KAAO,GAAC,IAC5D,iBAAgB,IAMtB,OAJM,GAAO,EAAE,MAAM,GACjB,EAAS,EAGJ,EAAI,EAAG,GAAK,KAAK,KAAK,KAAK,KAAK,IAAS,EAAG,KACnD,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,GAAO,GAC5C,EAAc,EAAQ,mBAAmB,GACzC,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,GAAS,EAAO,MAAO,GAC7D,EAAW,UAIb,GAAI,GACF,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAW,GAChD,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,IACtC,EAAW,UAGb,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA2B,QAE9D,GAAM,GAAU,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAE,OACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA+B,QAElE,GAAM,GACF,GAAI,IAAgB,GAAwB,EAAE,MAAO,EAAE,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAClE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA0B,QAG7D,GAAM,GAAU,GAAI,IAAgB,GAAmB,EAAE,MAAO,EAAE,OAClE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA6B,QAGhE,GAAM,GAAU,GAAI,IAAgB,GAAsB,EAAE,MAAO,EAAE,OACrE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAAmC,QAE/C,GAAM,GACF,GAAI,IAAgB,GAA4B,EAAE,MAAO,EAAE,OAC/D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAiC,QAEpE,GAAM,GACF,GAAI,IAAgB,GAA0B,EAAE,MAAO,EAAE,OAC7D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAM,GAAU,GAAI,IAAc,EAAU,KAAM,EAAE,MAAO,EAAE,MAC7D,MAAO,MAAK,cACR,EAAS,CAAC,EAAW,EAAG,GAAI,EAAA,WAAW,EAAE,MAAO,EAAE,SAGxD,EAAA,UAAA,MAAA,SAAM,GACJ,EAAA,aAAa,KACT,yEAEJ,GAAM,GAAW,EAAU,WAC3B,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAM,GAAQ,EAAE,WAChB,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAK,GACL,EAAc,QACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAA6B,GAEzC,GAAM,GAAU,GAAI,IAAgB,EAAI,EAAE,MAAO,EAAE,OACnD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAG1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,GAErD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAG7C,EAAA,UAAA,cAAR,SAAsB,EAAe,EAAY,GAC/C,GAAM,GAAU,GAAI,IAAqB,EAAE,MAAO,GAClD,MAAO,MAAK,cAAsB,EAAS,CAAC,GAAI,IAG1C,EAAA,UAAA,eAAR,SACI,EAAe,EAAe,EAAY,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GACF,GAAI,IAAsB,EAAI,EAAE,MAAO,EAAE,MAAO,GACpD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAO7C,EAAA,UAAA,yBAAR,SAAiC,EAAW,EAAW,GAAvD,GAAA,GAAA,KACQ,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAA,4SAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAcP,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,GAMD,EAAA,UAAA,+BAAR,SACI,EAAuB,GACzB,MAAO,CACL,OAAQ,EAAY,OACpB,MAAO,EAAY,MACnB,MAAO,EAAc,QAIzB,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAIjB,GAAI,EAAQ,OAAS,EAAA,MAAM,IAAI,iCAC7B,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,KAAK,EAAQ,MAAM,EAAG,IACtC,EAAY,KAAK,KAAK,EAAQ,MAAM,IAC1C,MAAO,MAAK,KAAK,CAAC,EAAU,IAG9B,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAO,OAAO,SAAC,EAAI,GAAO,MAAA,GAAA,WAAW,EAAI,KAC1D,EAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAE5B,EAAc,EAAA,MAAM,QAAQ,cAC5B,EAAU,EACZ,GAAI,IAAkB,EAAQ,GAAG,MAAO,GACxC,GAAI,IAAY,EAAQ,GAAG,MAAO,GACtC,MAAO,MAAK,cAAiB,EAAS,EAAS,IAGjD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,GAAM,GAAc,EAAA,MAAM,QAAQ,gCAC5B,EAAU,EACZ,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OAC/C,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,MAAO,MAAK,cAAiB,EAAS,CAAC,EAAG,GAAI,IAGhD,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAY,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACnE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAe,EAAE,OAGhD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAG1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAChD,EAAI,KAAK,SAAS,EAAQ,EAAS,QAAQ,IAC3C,EAAI,KAAK,IAAI,GACb,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAI,EAAG,IAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAGtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAE3C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA2B,EAAE,MAAO,EAAM,OAC9C,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAM,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAEtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA6B,EAAG,MAAO,EAAE,OAC7C,GAAI,IAAgB,GAAsB,EAAG,MAAO,EAAE,OAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,UAG1C,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,GAAI,GACJ,AAAI,EAAA,MAAM,QAAQ,mBAChB,EAAU,GAAI,IAAkB,EAAE,OAElC,EAAU,GAAI,IAAY,EAAE,OAE9B,GAAM,GAAc,EAAQ,mBAAmB,EAAK,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,IAGhD,EAAA,UAAA,IAAA,SAAsB,GAEpB,GAAI,KAAK,mBAAmB,CAAC,KAAO,EAAE,QAAU,aAC9C,GAAM,GACF,GAAiB,KAAK,QAAQ,IAAI,EAAE,QAAQ,QAChD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAU,GAAI,IAAkB,EAAE,OAClC,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAG9D,MAAO,MAAK,cAAsB,EAAS,IAG7C,EAAA,UAAA,QAAA,SAA0B,GACxB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,GAAc,IAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,KAG9B,EAAA,UAAA,eAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAGF,GAAM,GAAS,EAAE,MACX,EAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAkB,EAAS,WAC3B,EAAc,EAAO,GAAK,EAAO,GAAK,EAAO,GAC7C,EAAmB,EAAS,YAC5B,EAAiB,EAAS,aAAe,eACzC,EAAa,GACb,EAAa,GAIb,EACD,KAAgB,GAAK,IAAqB,IAC3C,EAAkB,GAChB,EAAyB,EAAO,GAAK,IAAM,GAAK,CAAC,CAAC,EAAS,SAEjE,GAAI,GAA6B,CAAC,EAAA,MAAM,QAAQ,wBAC5C,CAAC,EAAA,MAAM,QAAQ,iCACf,CAAC,GACH,GAAM,GAAc,EAAiB,EAAO,GAAK,EAAO,GAAK,EAAO,GAC/B,EAAO,GAAK,EAAO,GAAK,EAAO,GAC9D,EAAY,EAAA,QAAQ,EAAG,CAAC,EAAG,EAAa,EAAS,aACjD,EACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAS,KAAK,iBAAiB,CACnC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAExB,MAAO,GAAA,QAAQ,EAAQ,EAAS,UAWlC,GAAM,GAAc,EAChB,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACrC,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACnC,EAAwB,CAC5B,OAAQ,EAAE,OACV,MAAO,CAAC,EAAG,EAAa,EAAS,YACjC,MAAO,EAAE,OAUL,EAAwB,EAAS,MACvC,EAAS,MAAQ,EAAS,MAAM,QAChC,EAAS,MAAM,EAAS,MAAM,OAAS,KACvC,EAAA,KAAK,OACD,GAAyB,EAAS,MAAO,EAAU,OACnD,WAAM,MAAA,kBAAkB,EAAS,MAAK,OAClC,EAAU,MAAK,gBACvB,GAAM,GACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAgB,KAAK,iBAAiB,CAC1C,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAElB,EAAuB,KAAK,QAAQ,IAAI,EAAc,QAC5D,SAAA,KAAK,OACD,EAAqB,SACrB,WAAM,MAAA,gDAEV,EAAS,MAAQ,EAGjB,EAAqB,MAAQ,EAAS,SAC/B,EAAA,SAAS,qBACL,EAAc,OAAQ,EAAS,SAAU,EAAc,QAI5D,EAAA,UAAA,iBAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAQA,GAAA,GAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,WAGI,EAAiB,IAAe,eAEhC,EAAY,EAAc,EAAe,EACzC,EAAU,EAAY,EACtB,EAAa,CAAC,EAAW,GACzB,EAAa,GACb,EAAa,GAEb,EAAY,EAAE,QAAQ,CAAC,IACvB,EAAQ,EAAO,QAAQ,CAAC,EAAG,EAAW,KAEtC,EACF,GAAI,IAAoB,EAAY,EAAU,MAAO,GACnD,EACF,KAAK,cAAwB,EAAe,CAAC,IAAY,QAAQ,CAC/D,EAAG,EAAW,GAAI,EAAW,KAG7B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAgB,GAAI,IACtB,EAAO,MAAO,CAAC,EAAG,EAAS,EAAS,aAAc,EAClD,EAAY,EAAS,EAAiB,GACpC,EAAuB,CAAC,EAAQ,GACtC,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEd,GAAM,GAAU,KAAK,cAAwB,EAAe,GAE5D,MAAI,GACK,EAAQ,QAAQ,CAAC,EAAG,EAAW,EAAU,EAAS,cAElD,EAAQ,QAAQ,CAAC,EAAG,EAAS,YAAa,EAAW,KAIhE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAEjD,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAM,MAAM,KAAO,EAC3D,MAAO,MAAK,iBACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAGjD,GAAM,GAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAS,KAC7D,EAAU,GAAI,IAChB,EAAU,EAAS,EAAiB,GAClC,EAAuB,CAAC,EAAO,GACrC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAc,EAAS,IAGrC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eAAe,EAAG,EAAQ,GAExC,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAM,KAAO,EACvD,MAAO,MAAK,iBAAiB,EAAG,EAAQ,GAE1C,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAExC,EAA0B,EAAA,MAAM,QAAQ,6BAC1C,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EAC7C,EAAkB,EACpB,GAA6B,EAAY,GACzC,KACE,EAAmB,CAAC,EAAO,GAE3B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KAC5D,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAGd,GAAI,GACJ,MAAI,GACF,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,IAGrC,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,KAGrC,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,6BACd,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EACjD,GAAU,GAAI,IAA6B,GACpC,KAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,GAAU,GAAI,IAAuB,GAC9B,KAAK,cAAc,EAAS,CAAC,EAAG,MAGzC,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAA+B,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAM,GAAU,GAAI,IAAgC,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAA,aAAa,WAAW,EAAG,EAAO,OAG3C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,KAAK,MAAM,EAAG,EAAO,GAAM,QAAQ,GAE9C,MAAO,IAGT,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GAA2B,GAAI,IAAyB,GAC9D,MAAO,MAAK,cAAc,EAA0B,CAAC,GAAK,EAAE,QAG9D,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAM,GAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,KAAK,cAAc,EAA2B,CAAC,IAC7C,EAA2B,GAAI,IAAyB,GACxD,EAAS,KAAK,cAChB,EAA0B,CAAC,EAAI,GAAqB,EAAE,OAC1D,SAAmB,UACZ,GAGT,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IACA,EAAE,MAAO,EAAW,EAAU,GAClC,GAAI,IAAsB,EAAE,MAAO,EAAW,EAAU,GAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,GAAM,GAAU,GAAI,IAA8B,EAAI,EAAG,GAEzD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,GAAI,IAChB,EAAE,MAAO,EAAW,EAAU,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GACF,GAAI,IAAoC,EAAI,EAAG,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAM,GAAQ,EAAa,EAAS,EAAA,QAAQ,GACtC,EAAY,EAAM,MAAM,GACxB,EAAc,EAAM,MAAM,GAC1B,EAAU,GAAI,IAAmB,EAAW,EAAa,GACzD,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAQ,QAAS,IAGvD,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAM,GAAU,GAAI,IAAc,EAAQ,KAAM,EAAO,EAAS,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAAI,IAAY,EAAE,MAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,GAAM,GAAU,GAAI,IAChB,EAAM,MAAO,EAAM,MAAO,EAAU,EAAQ,GAChD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAO,EAAO,GAAW,YAG/D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAE9D,GAAM,GAAY,EAAE,MAAM,GACpB,EAAe,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAE5D,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAe,IAAe,OAChC,CAAC,EAAW,EAAc,EAAa,GACvC,CAAC,EAAW,EAAa,EAAc,GAErC,EAAU,GAAI,IAAoB,EAAa,EAAW,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAG5C,EAAe,CAAC,EAAa,EAAW,GACxC,EAAiB,EAAQ,QAAQ,CAAC,EAAY,IAC9C,EAAW,EAAQ,QAAQ,CAAC,EAAY,IAE9C,GAAI,IAAe,EACjB,MAAO,GAAA,aAAa,cAAc,EAAA,OAAO,IAAK,GAEhD,GAAM,GAAe,EAAA,OAAO,GACtB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAe,KAAM,EAAS,KAAM,EAC3D,GACE,EACF,KAAK,cAAc,EAAS,CAAC,EAAU,EAAgB,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAGjC,EAAiB,GACjB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAc,KAAM,EAAa,KAAM,EAC9D,CAAC,EAAY,GAAI,GACf,EAAc,KAAK,cACrB,EAAS,CAAC,EAAc,EAAe,IAC3C,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGzB,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGjB,EAAA,UAAA,QAAR,SAAgB,EAAa,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAGxD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAU,KAAK,QAAQ,EAAM,GAAM,KAAK,EAAE,MAAM,GAAI,EAAE,MAAM,IAClE,SAAK,UACL,EAAK,UACE,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAGpC,EAAiB,EAAQ,QAAQ,CAAC,EAAW,IAC7C,EAAW,EAAE,QAAQ,CAAC,EAAE,KAAO,EAAW,IAC1C,EACF,GAAI,IAAgB,EAAW,EAAS,CAAC,EAAW,IAClD,EAAc,KAAK,cAAc,EAAS,CAAC,EAAU,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAG5C,GAFA,EAAQ,GAAS,EAAA,KAAK,WAAW,GAE7B,IAAU,UAEZ,GAAM,GAAS,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IAChE,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,WAEjD,GAAM,GAAU,GAAI,IAAY,EAAO,GACjC,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,GAAI,EAAO,KAIlD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAIhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAE,QAAU,SAAW,GAAK,EAAG,EAAE,QAG7D,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GAEpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGhD,EAAA,UAAA,eAAA,SAAe,EAAiB,EAAiB,GAE/C,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,YAAK,QAAQ,IAAI,GAAQ,MAAQ,KAC1B,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAGtB,EAAA,UAAA,WAAR,SACI,EAAiB,EAAiB,GAC7B,GAAA,GAAA,KAAA,eAAA,EAAA,EAAA,GAAA,OACP,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAGrD,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAU,GAAI,IAAc,EAAM,OACxC,MAAO,MAAK,gBAAgB,EAAS,CAAC,GAAQ,EAAM,QAG9C,EAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAU,GAAI,IAAY,EAAM,OAChC,EAA8B,GACpC,MAAO,MAAK,gBACR,EAAS,CAAC,GAAQ,EAAM,MAAO,KAC/B,IAGE,EAAA,UAAA,cAAR,SAAsB,EAAmB,GACvC,GAAM,GAAe,CACnB,GAAuB,EAAM,eAC1B,GAAuB,EAAM,QAE5B,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EAAiB,CACrB,GAAuB,WAAgB,GAAuB,IAG1D,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,KAAK,gBAChB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,QAG1D,EAAA,UAAA,OAAR,SAAe,GACb,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,MAClB,EACF,GAAwB,GACxB,EACJ,AAAI,EACF,EAAU,GAAI,IAA0B,GAExC,EAAU,GAAI,IAAoB,GAEpC,GAAM,GAAgC,GAChC,EAAM,KAAK,gBACb,EAAS,CAAC,CAAC,MAAO,EAAW,MAAK,EAAE,OAAM,IAAI,EAC9C,KAAwB,GAC5B,MAAO,CAAC,MAAK,EAAE,MAAK,EAAE,OAAQ,EAAI,SAGpC,EAAA,UAAA,gBAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAHJ,GAAA,GAAA,KAGI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,KAAK,eAAe,EAAQ,YAAa,GAClD,EAAU,KAAK,QAAQ,IAAI,EAAO,QAIxC,GAHA,AAAI,EAAQ,cACV,GAAQ,SAAW,IAEjB,EAAQ,mBAAqB,GAAuB,OACtD,GAAM,GAAa,GAA0B,EAAQ,aAKrD,EAAQ,SAAW,EAAW,IAAI,SAAA,GAAK,MAAA,GAAI,IAK7C,GAHA,AAAI,EAAQ,aAAe,MACzB,GAAQ,MAAQ,EAAQ,aAEtB,EAAA,KAAK,cAAc,EAAO,SAAW,EAGvC,SAAQ,OACJ,EAAA,KAAK,uBAAuB,EAAO,MAAoB,GACpD,EAGT,GAAM,GAA8B,GAC9B,EAA2B,EAAO,IAAI,SAAA,GAC1C,GAAI,EAAM,QAAU,YAClB,KAAM,IAAI,OACN,mIAKN,GAAI,GAAU,EAAK,QAAQ,IAAI,EAAM,QAErC,GAAI,EAAQ,SAAW,MACrB,GAAI,CAAC,EAAQ,cACT,EAAA,KAAK,cAAc,EAAM,QACrB,EAAA,MAAM,UAAU,6BAMtB,MAAO,CACL,MAAO,EAAM,MACb,QAAS,KACT,UAAW,GACX,cAAe,EAAQ,QAM3B,AAAI,EAAQ,cACV,GAAQ,SAAW,GACnB,EAAQ,MAAQ,EAAM,eAEf,CAAC,CAAC,EAAQ,WAAa,CAAC,CAAC,EAAQ,aAC1C,EAAQ,EAAQ,SAAW,EAAK,aAAa,GAClB,EAAK,WAAW,GAC3C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,gBAE/B,EAAQ,UACR,CAAC,GAAyB,EAAQ,MAAO,EAAM,QAQjD,GAAM,GAAa,EACb,EAAc,EAAM,MAE1B,EAAM,MAAQ,EAAQ,MACtB,EAAQ,EAAK,cAAc,EAAiB,GAC5C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,QAEjC,EAAW,MAAQ,EAGrB,SAAK,YAAY,EAAM,QAChB,CAAC,MAAO,EAAM,MAAO,QAAO,EAAE,UAAW,MAGlD,KAAK,YAAY,EAAO,QACxB,GAAM,GACW,CAAC,MAAO,EAAO,MAAO,QAAS,EAAS,UAAW,IAC9D,EAAM,GAAyB,EAAS,EAAY,GACpD,EAAS,KAAK,iBAAiB,EAAK,WACxC,MAAO,IACH,EAAK,MAAO,EAAS,EAAY,KAEjC,EAAoB,KAAK,cAAgB,KAC3C,EAgBJ,GAfA,AAAI,GACF,GAAQ,KAAK,cAGf,GACI,KAAK,MAAO,EAAQ,EAAY,EAAY,GAEhD,EAAc,QAAQ,SAAA,GAAQ,MAAA,GAAK,8BAA8B,KAEjE,AAAI,GACF,GAAQ,KAAK,SAAS,GACtB,KAAK,aAAa,KACd,CAAC,KAAM,EAAQ,YAAY,KAAM,MAAO,KAAK,aAAa,MAG5D,CAAC,EAAA,MAAM,QAAQ,wBAA0B,EAAQ,UACjD,IAAkC,IACpC,GAAM,GAAW,KAAK,aAAa,GACnC,YAAK,8BAA8B,GAC5B,EAET,MAAO,IAGT,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,EAAc,GAAe,EAAO,GAAG,MACvC,GAAM,GAAU,KAAK,gBACjB,EAAS,EAAQ,EAAa,EAC9B,GACJ,MAAO,GAAA,SAAS,qBACL,EAAQ,OAAQ,EAAQ,MAAO,EAAQ,QAG5C,EAAA,UAAA,iBAAR,SAAyB,EAAa,GAEpC,MAAM,KAAO,MAAK,aAChB,MAAK,YAAY,GAAO,KAEnB,KAAK,YAAY,IAG1B,EAAA,UAAA,kBAAA,WACE,MAAO,MAAK,gBAKd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAIF,GAAI,CAAC,EAAA,MAAM,QAAQ,YACjB,GAAM,GAAU,OAAO,KAAK,KAAK,aACjC,EAAQ,QAAQ,SAAA,GACd,EAAK,MAAM,cAAc,EAAK,YAAY,GAAK,cAC/C,MAAO,GAAK,YAAY,KAG5B,KAAK,eAAe,UACpB,AAAI,KAAK,QAAU,MACd,MAAQ,oBAAuB,aAC/B,KAAK,iBAAkB,mBAC1B,KAAK,OAAO,SAEZ,KAAK,OAAS,KAEhB,AAAI,KAAK,qBACP,MAAK,MAAM,QAAU,KACrB,KAAK,MAAM,WAEb,KAAK,SAAW,IAGlB,EAAA,UAAA,eAAA,WAAA,GAAA,GAAA,KACE,MAAI,MAAK,qBAAuB,MAC9B,MAAK,oBAAsB,EAAA,KAAK,WAC9B,GAAI,CAAC,EAAA,MAAM,IAAI,iCAGb,GAAM,GAAY,EAAA,MAAM,QAAQ,SAChC,EAAA,MAAM,IAAI,QAAS,IACnB,GAAM,GAAsB,EAAK,IAAI,EAAA,OAAO,OAAO,WAAW,GAG9D,GAFA,EAAA,MAAM,IAAI,QAAS,GAEf,EAAsB,EACxB,MAAO,IAGX,MAAO,OAGJ,KAAK,qBAGd,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAGlD,EAAA,UAAA,YAAR,SAAoB,SACZ,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAE7C,GAAI,GAAW,KAEb,OAEF,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GAAW,EAAQ,SAMvB,GALA,AAAI,GAAY,MACd,GAAW,GAA2C,EAAO,GAC7D,EAAQ,SAAW,GAGjB,GAAU,MACZ,GAAM,GAAY,GAAwB,GAEtC,EAAO,OACP,EAAQ,EAAS,GAAI,EAAS,EAAS,GACrC,EAAc,YAAkB,YAEtC,AAAI,EACF,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,EAAU,GAAI,IACV,EAAW,CAAC,EAAQ,GAAQ,IAEhC,EACI,GAAI,IAAoB,EAAW,CAAC,EAAQ,GAAQ,GAG1D,GAAM,GAAuB,KAAK,eAAe,CAAC,EAAQ,GAAQ,GAClE,AAAI,EACF,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEjB,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEnB,KAAK,MAAM,2BACP,KAAK,WAAW,EAAqB,QAAS,EAAO,EACrD,GAIJ,GAAM,GAAwB,GACxB,EAAsB,KAAK,gBAC7B,EAAS,CAAC,GAAuB,EAAO,KAAM,GAG5C,EAAgB,KAAK,QAAQ,IAAI,EAAoB,QAC3D,EAAQ,QAAU,EAAc,QAChC,EAAQ,SAAW,EAAc,SACjC,EAAQ,SAAW,EAAc,SACjC,EAAQ,MAAQ,EAAc,MAE9B,KAAK,8BAA8B,GACnC,KAAK,QAAQ,OAAO,EAAoB,QAGxC,EAAQ,OAAS,KACjB,AAAI,GACF,MAAK,cAAgB,EAAA,KAAK,MAAQ,QAGpC,GAAM,GAAa,KAAK,eAAe,EAAU,EAAO,EAAO,GAC/D,EAAQ,QAAU,IAId,EAAA,UAAA,qBAAR,SAA6B,EAAgB,GAE3C,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAEP,YAAK,eAAe,GAEpB,AAAI,GAAiB,MACnB,GAAQ,OAAS,GAAoB,EAAe,IAE/C,EAAQ,QAGT,EAAA,UAAA,eAAR,SACI,EAA4B,EAAuB,EACnD,GAEF,GADA,KAAK,eAAiB,KAAK,aAAa,EAAU,GAC9C,CAAC,KAAK,mBACN,KAAK,cAAgB,KAAK,mBAAqB,KAAO,MACxD,GAAM,GAAM,MAAK,cAAgB,KAAO,MAAM,QAAQ,GACtD,KAAK,kBAAoB,GACzB,QAAQ,KACJ,6BAA6B,EAAE,yCAGrC,MAAO,MAAK,eAAe,eAAe,EAAU,EAAS,IAGvD,EAAA,UAAA,aAAR,SAAqB,EAAyB,GAC5C,MAAO,GAAM,GAAK,EAAM,GAAK,EAAA,KAAK,gBAAgB,IAG5C,EAAA,UAAA,mBAAR,SACI,EAAsB,GACxB,GAAI,KAAK,mBAAmB,GAC1B,IACE,MAAO,WACA,GACP,GAAI,EAAA,MAAM,QAAQ,WAChB,KAAM,IAAI,OAAM,yBAItB,MAAO,OAEX,GAriFsC,EAAA,eAuiFtC,YACI,EAAiB,GACnB,GAAI,IAAU,WAAa,IAAU,YACnC,MAAO,GACF,GAAI,IAAU,SAAW,IAAU,QAGxC,OAFM,GAAU,IAAU,QAAW,GAAI,YAAW,EAAE,QACjB,GAAI,YAAW,EAAE,QAC7C,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,KAAK,MAAM,EAAE,IAE3B,MAAO,OAEP,MAAM,IAAI,OAAM,iBAAiB,GChwFrC,GAGM,IAAU,QCHhB,cAmCE,EAAA,MAAM,IAAI,2BAA4B,ICnCxC,AAuBA,AAAI,EAAA,YAAY,aACd,EAAA,gBAAgB,QAAS,WAAM,MAAA,IAAI,KAAoB,MAQ5C,IAAQ,CAAC,eAAc,ICzBvB,GAA0B,0BAE1B,GAA2B;;;EAK3B,GAAkC;;;;;cAWf,GAC9B,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EACD,EAAe,EACf,EAAU,GAAI,IAAe,EAAE,MAAO,GAC5C,MAAO,GAAa,gBAAgB,EAAS,CAAC,GAAI,EAAE,oBAepD,EAAmB,EACnB,EAA4C,GAG9C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAe,EACf,EAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,EAAiB,EAAE,MAAO,EAAE,MAC5B,CAAC,CAAC,GACN,GAAI,IAAgB,EAAW,EAAE,MAAO,EAAE,OACxC,EAAS,GAAS,EAAE,MACpB,EAAS,EAAa,gBAAgB,EAAS,CAAC,EAAG,GAAI,GAC7D,MAAO,IC3DX,AAsBA,GAAM,IAAQ,GAA2B;;EAInC,GAAe;;;IAIjB,GAAkC;;EAIzB,GAAkB,GAAiB,GAAO,IAE1C,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,QACb,WAAY,ICvCd,YAqBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,QACb,WAAY,ICjCd,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,WAG/C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBACvB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAyB,GAAI,IAAyB,GAC5D,MAAO,GAAQ,gBAAgB,EAAwB,CAAC,GAAK,EAAE,OAG1D,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,IC3Cd,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,YAAwB,GAOtB,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,MACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,MACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;yBAKK,EAAa;wBACd,EAAY;2DACuB,EAAe;;;MAK1E,MAAA,MC5DA,AAoBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,YACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,YACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;wBAEI,EAAa;uBACd,EAAY;;;;;;yDAMsB,EAAe;;;;MAMxE,MAAA,MChEA,AAuBO,GAAM,IAIM,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,EAAA,EAAG,EAAA,EAAA,KAAM,EAAA,EAAA,SAAU,EAAA,EAAA,OAAQ,EAAA,EAAA,MAElC,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGL,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAGpB,GAAM,GAAc,CAAC,EAAG,EAAM,GAE1B,EAAc,KAClB,AAAI,GAAU,MACZ,GAAc,EAAO,MACrB,EAAY,KAAK,IAGnB,GAAI,GAAa,KACjB,AAAI,GAAS,MACX,GAAa,EAAM,MACnB,EAAY,KAAK,IAGnB,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACJ,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACF,EACF,EAAQ,gBAAgB,EAAS,EAAa,EAAY,GAAG,OAEjE,MAAO,IAGI,GAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,QACb,WAAY,IC/Ed,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAuBA,GAAM,IAAM;;;;eAQN,GAAa;;;;;;;;;;;;;;;;;;EAoBN,GAAgB,GACzB,GAAK,GAAY,IAER,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,ICzDd,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAItB,GAAM,GAAa,EAAW,GAC9B,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;yBAKK,EAAU;;uCAEI,EAAU;;;;;;;MASjD,MAAA,MC5CA,AAuBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,MACD,EAAe,EAEf,EAAU,GAAI,IAAsB,EAAmB,OACvD,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KChCX,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;uDAMmC,EAAK,OAAO,EAAM;;wBAEjD,EAAK,UAAS;;;;;;;;;;;;;;MAgBtC,MAAA,MCrDA,AAoBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;;8BAeU,EAAK,OAAO,EAAM;4BACpB,EAAK,UAAS;;;;;;;;;;;;;;;;UAgBhC,EAAK,OAAM;;MAIrB,MAAA,MCnEA,AA0BO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,QACb,WAAY,IAGV,GAEJ,YAAoB,GAKX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACnB,EAAA,EAAA,OACE,EAAA,EAAA,YAED,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAOR,EAA6B,CAAC,EAAQ,GACtC,EAAW,CAAC,EAAQ,EAAO,GAEjC,AAAI,IAAW,IACb,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAGpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA+C,EAAG,EAAG,EAAO,GAChE,EAAS,GAAoB,QAG/B,GAAM,GAAkB,EAAQ,eAAe,EAAU,SAEzD,EAAQ,QAAQ,IAAI,EAAgB,QAAQ,MAAQ,GAAa,OACjE,EAAQ,MAAM,yBACV,EAAQ,WAAW,EAAgB,QAAS,GAChD,GAAM,GAAU,EAAA,MAAM,QAAQ,cAC1B,GAAI,IAAwB,GAC5B,GAAI,IAAkB,GACpB,EAAM,EAAQ,gBAAgB,EAAS,CAAC,GAAkB,SAChE,SAAQ,YAAY,EAAgB,QAC7B,EC/ET,AA0BA,YAA4B,GAI1B,OAFM,GAAS,GAER,EAAO,SAAW,GAAK,EAAO,EAAO,OAAS,GAAG,UAAY,IAClE,GAAM,GACF,EAAO,OAAS,EAAO,EAAO,OAAS,GAAG,QAAU,EAAQ,GAC1D,EAAa,EAAA,aAAa,yBAAyB,GACzD,EAAO,KAAK,CACV,OAAQ,EACR,WAAU,EACV,QAAS,KAAK,KAAK,EAAU,KAIjC,MAAO,eAIL,EAAe,EAAiB,EAChC,GAIF,OAHM,GAAkB,GAAmB,EAAE,OAEzC,EAAS,EACJ,EAAI,EAAG,EAAI,EAAgB,OAAQ,KACpC,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,WAAY,EAAA,EAAA,QAErB,EAAU,GAAI,IAChB,CAAC,WAAU,EAAE,OAAM,EAAE,UAAW,EAAE,MAAM,GAAI,QAAO,GAAG,GACpD,EAAiB,EACvB,EAAS,EAAQ,gBAAgB,EAAS,CAAC,GAAS,GAEpD,AAAI,EAAe,SAAW,EAAE,QAC9B,EAAQ,YAAY,EAAe,QAIvC,MAAO,GC/DT,YAwBI,EAAmB,EACnB,GACF,GAAM,GACF,CAAC,GAAY,EAAM,QAAM,OACrB,GAAY,EAAM,QACpB,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EACF,CAAC,GAAY,IAAW,OACpB,GAAY,IAEd,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,EAAQ,gBACnB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,OC3ClE,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MACD,EAAe,EAEf,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QAC5C,MAAI,GAAS,UAAY,CAAC,GAAc,EAAE,MAAO,IAC7C,CAAE,GAAS,UAAY,MAAQ,GAAc,EAAS,MAAO,IACxD,GAAc,EAAG,EAAQ,GAGlC,GAAa,OAAO,EAAE,QAEf,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,QAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICzDd,YAwBI,EAAe,EAAuB,EACtC,GACF,GAAM,GAAS,EAAA,KAAK,cAAc,GAC5B,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAY,EAAQ,EACpB,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAC,GAAG,MAAO,CAAC,MAAO,CAAC,EAAW,IAAU,QAAO,IAEhE,EAAU,GAAO,EAAe,EAAE,MAAO,MAAO,GAChD,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAU,MAAO,CAAC,MAAO,GAAW,QAAO,IAEpE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,ECvCT,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAEjC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAW,GAAkB,GAEnC,KAAK,SAAW;;QAEZ,EAAK;uBACU,EAAQ;;MAI/B,MAAA,MAEA,YAA2B,GACzB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,sBAAsB,EAAI,yBAKxC,OAHM,GACF,CAAC,UAAW,UAAW,UAAW,UAAW,UAAW,WACtD,EAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAe,EAAO,IAAM,EAAc,GAE5C,MAAO,GAAe,OCxDxB,AAqBA,GAAA,IAAA,WAQE,WAAY,EAAkB,GAP9B,KAAA,cAAgB,CAAC,KAIjB,KAAA,aAAe,GACf,KAAA,aAAe,GAIb,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAIjC,GAFA,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACpB,KAAK,KAAO,EACd,KAAM,OACF,6BAA6B,KAAK,KAAI,0BAM5C,OAJM,GAAQ,GAAkB,KAAK,MAE/B,EAAc,GAAe,KAAM,KAAK,MACxC,EAAgB,GAAI,OAAM,KAAK,MAC5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAc,EAAO,IAAM,EAAY,GAEzC,GAAM,GAAY,QAAQ,EAAc,MAAM,IAAI,OAAM,IAClD,EACF,KAAK,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAC3D,EAAO,mBAAmB,EAAc,OAAM,MAAM,EAAS,IAEnE,KAAK,SAAW;;QAEZ,EAAK;;oBAEO,EAAI;WACb,EAAU;sBACC,EAAI;;UAEhB,EAAY,KAAK,KAAO,GAAE;aACvB,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAAE;sBACjD,EAAI;aACb,EAAU;wBACC,EAAI;;;;;MAO5B,MAAA,MCvEA,YAyBI,EAAe,EAAgB,GACjC,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAuB,EAAE,MAAO,GACpC,GAAI,IAAiB,EAAE,MAAO,GAClC,MAAO,GAAQ,gBAAgB,EAAS,CAAC,GAAI,EAAE,OC7BjD,AA0BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAW,EAAA,KAAK,eAAe,EAAkB,EAAE,OACrD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACrD,EAAuB,GAAgB,KACvC,EAAqB,EAAa,mBAAmB,CAAC,IAExD,EAAW,EACf,GAAI,GACF,GAAI,GAKF,OAJM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAa,IAErC,GAAM,GACF,GAAiB,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAc,GAE7D,EAAW,EAAa,eAAe,EAAU,EAAE,OACnD,GAAM,GAAe,EAAa,QAAQ,IAAI,EAAS,QACvD,EAAa,OAAS,MAEtB,GAAW,GAAc,EAAG,EAAc,GAG5C,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAGpD,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGhB,EAAW,EACf,AAAI,GAEF,GAAW,EAAA,aAAa,qBAAqB,EAAa,IAG5D,GAAI,GACJ,GAAI,GACF,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAY,GACd,EAAQ,EAAA,KAAK,cAAc,GAAc,EAAU,EAAE,OAEzD,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAQ,EAAU,EAAa,EAAU,GAGjD,MAAI,IACF,EAAa,8BAA8B,GAGtC,IC9FX,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,EAAE,OAGjD,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAuBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAC3B,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,EAAQ,gBAAgB,EAAyB,CAAC,GAAI,EAAE,OAEtD,EAAyB,GAAI,IAAyB,GACtD,EAAS,EAAQ,gBACnB,EAAwB,CAAC,EAAI,GAAmB,EAAE,OACtD,SAAQ,8BAA8B,GAC/B,EAGF,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,ICrDd,YAuBI,EAAe,EACf,EACA,GACF,GAAI,GAAU,GAAI,IAAc,EAAU,MAAO,IAC3C,EAAa,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAEzD,EAAU,GAAI,IAAc,EAAU,MAAO,GAAM,GAAM,GACzD,GAAM,GAAc,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAC1D,MAAO,CAAC,EAAY,GC/BtB,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAe,EAErB,EAAA,KAAK,OACD,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,uDACF,EAAE,MAAM,OAAM,MACtB,GAAM,GAA8B,CAAC,EAAG,GACxC,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,GAET,EAAA,GAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAEf,MAAO,CAAC,EAAQ,KCjDpB,AAoBO,GAAM,IAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAG9B,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EAE1B,MAAO,GAAA,aAAa,wBAChB,EAAW,EAAY,EAAkB,EACzC,KC3CR,AAkBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC1C7B,AAmBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAqBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EACA,GAPJ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAOtB,GAAM,GAAc,EAAW,GACzB,EAAa,EAAW,GACxB,EAAY,KAAK,IAAI,GAAS,QAAQ,GACtC,EAAY,KAAK,IAAI,GAAS,QAAQ,GAC5C,KAAK,YAAc,EAEb,GAAA,GAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAgB,EAAQ,QAAQ,GAChC,EAAgB,EAAQ,QAAQ,GAElC,EAAc,GAClB,AAAI,MAAO,IAAc,SACvB,EAAc,uBAAuB,EAAU,QAAQ,GAAE,IAEzD,EAAc;2BACO,EAAU,KAAK,KAAI;8CAI1C,KAAK,SAAW;;;;;4CAKwB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;4CACpB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;iDACf,EAAa;iDACb,EAAa;YAClD,EAAW;uCACgB,EAAU,+BACzC,EAAW;;;;;MAOnB,MAAA,MCtEA,AAuBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAe,EAEf,EAAU,GAAI,IACf,EAAmB,MAAO,EAAS,EAAW,GAC7C,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KClCX,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAqBA,GAAM,IAAS,gBAEF,GAAmB,GAAgB,IAEnC,GAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IC5Bd,AAqBA,GAAM,IAAqB,4BAEd,GACT,GAAiB,GAAoB,IAE5B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,IC7Bd,AAqBA,GAAM,IAAM,iBAEC,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC5Bd,AAwBO,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,QACb,WAAY,SAAC,GAQX,OARY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACA,EAAA,EAAA,KACD,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAI,GACJ,GAAI,EAAa,mBAAmB,CAAC,KACnC,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QACtC,EAAS,EAAS,OAClB,EAAY,GAAa,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAE/D,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAc,EAAG,EAAM,GAE/B,MAAO,KCnDX,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAGpB,QAAQ,KACJ,YACA,8DACJ,GAAM,GAAS,EAAQ,SAAS,EAAE,QAC5B,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IChDd,AA4CA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,+RCxEjB,AAGA,GAAM,IAAU,QCHhB,GAsCa,IAAU,CACrB,YAAa,GAAA,aACb,mBAAoB,GAAA,YACpB,qBAAsB,GAAA,cACtB,YAAa,GAAA,aACb,cAAe,GAAA,eACf,iBAAkB,GAAA,kBAClB,KAAQ,yZC7CV,oBAAM,IAAK,KAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,GAAG,IAAI,EAAI,WAAY,GAChC,EAAO,GAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,GAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,GAAG,IAAI,EAAW,GAC5B,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,GAAG,IAAI,EAAU,GACtC,EAAoB,GAAG,IAAI,EAAS,GACpC,EAAc,GAAG,IAAI,EAAoB,GACzC,EAAS,GAAG,IAAI,EAAmB,GACnC,EAAO,GAAG,IAAI,EAAmB,GACjC,EAAkB,GAAG,IAAI,EAAQ,GACjC,EAAgB,GAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,IAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,IAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,GAAG,SAAS,KAAK,aAChC,KAAK,UAAY,GAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,GAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,GAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,GAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,GAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,GACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,GAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,IAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,kBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,oBAAM,IAAK,KAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,GAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,YAAc,KClDtB,kBAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,GAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,GAAQ,wBAA0B,KCvFlC,eACA,KAAM,IAAK,KACL,GAAW,KACX,GAAY,KACZ,GAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,GAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,GAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,GAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,GAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,GAAU,yBAAyB,GACtD,EAAkB,GAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,GAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,GAAS,YAAY,GAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,GAAS,WAAW,GACpC,GAAI,GAAO,GAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,GAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,GAAS,oBAAoB,EAAe,GACxD,EAAc,GAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,GAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,GAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,GAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,GAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,GAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,EAAgB,KAAM,GAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,EAAa,QAAS,EAAiB,KAAM,GAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,EAAkB,KAAK,UAAU,QAAQ,GAAG,OAAO,CAAC,EAAa,KACjE,EAAqB,EAAe,WAC1C,EAAe,UACf,KAAM,GAAc,EAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,EAAkB,KAAM,GAAsB,KAAK,aAAa,EAAa,EAAY,EAAgB,IACtH,EAAe,EAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,EAAmB,KAAM,GAAuB,KAAK,aAAa,EAAc,EAAa,GAC1G,EAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,GAAiC,GAC5C,IAAsB,EAAW,EAAkB,QACnD,GAAsB,EAAW,EAAmB,UAE/C,AAAI,EAAgC,EACzC,GAAsB,EAAW,EAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,EAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,GAAyB,KAAK,sBAAsB,EAAW,EAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,EAAoB,SAC1F,EAAY,EAAU,OAAO,GAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,GAAG,QAAQ,GACX,KAAM,GAAe,GAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,GAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,GAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,GAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,kBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,0CAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,oBAAM,IAAK,KACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QAAS,GAAQ,GAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,GAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,GAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,oBAAM,IAAK,KAEL,GAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,GAAG,QAAQ,WAAW,GAC/B,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,IAAO,KAAK,IAAO,IAAM,KAAM,IAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,GAAO,IAGhB,kBAA0B,GACxB,MAAK,IAAO,QAAQ,IAAO,OAAS,KAAM,IAAG,eAAe,EAAO,KAAK,OAAO,YACxE,GAAO,OAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,IAAI,WAC3B,UAAQ,EACD,GAET,GAAI,GACJ,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,GAAG,IAAI,EAAQ,CAAC,MAC1B,GAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,IAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,GAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,IAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OAC3B,EAAI,OAAS,KAAK,MAAM,IAAM,EAAK,IAAM,GAAK,SAAW,OACzD,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KCzDrB,oBAAM,IAAK,KAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,GAAG,KAAK,KACrB,KAAM,GAAS,GAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,IAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,GAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,GAAG,IAAI,EAAG,CAAC,QAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAY,GAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,oBAAM,IAAK,KAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,oBAAM,IAAK,KACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,eACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,oBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,kBAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,GAAQ,cAAgB,GAAQ,UAAU,OAC1C,GAAQ,QAAU,GAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,GAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,GAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,GAAQ,QAAQ,GAAa,GAAQ,QAAQ,KACnI,GAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,oBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,GAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,GAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,GAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,GAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,GAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,GAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,GAAQ,YAAc,KCnDtB,oBAAM,IAAY,KACZ,GAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,GAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,GAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,GAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,GAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,oBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,oBAAM,IAAK,KACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,GAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,GAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,GAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,GAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,GAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,GAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,IAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,GAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,IAAG,OAAS,EAAQ,GAAG,QAAQ,WAAW,GAEpE,GAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,IAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,GAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,GAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,GAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,GAAQ,kBAAoB,KCrH5B,oBAAM,IAAK,KACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,oBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,GAAQ,KAAO,GAAa,KAC5B,GAAQ,QAAU,GAAa,QAE/B,GAAQ,UAAY,GAAe,UACnC,GAAQ,oBAAsB,GAAe,oBAC7C,GAAQ,aAAe,GAAU,aACjC,GAAQ,QAAU,GAAU,QAC5B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,eAAiB,GAAK,eAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,kBAAoB,GAAK,kBACjC,GAAQ,UAAY,GAAK,YCnBzB,oBAAM,IAAK,KAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,GAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,SAAW,KCtEnB,oBAAM,IAAK,KACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,GAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,GAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,GAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,IAAG,KAAK,KACb,KAAM,GAAa,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,GAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,IAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,IAAG,KAAK,KACb,KAAM,GAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,IAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,GAAG,KAAK,IAAM,GAAG,IAAI,GAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,GAAG,KAAK,IAAM,GAAG,QAAQ,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,GAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,GAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,GAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,kBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,2BAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,KCzEtB,oBAAM,IAAK,KACL,GAAW,KACX,GAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,IAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,IAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,GAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,GAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,GAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,GAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,GAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,oBAAM,IAAK,KACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,GAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,IAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,0CAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GACT,SAAU,CACR,UAAW,sCACX,UAAW,IACX,SAAU,GACV,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCACX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,EACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,EACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,6vECnEjB,oBAAM,IAAK,KACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,IAGJ,KAAM,IAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,GAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,GAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,GAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,GAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,IAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,GAAG,eAAiB,GAAO,SAC7B,IAAI,iCAAkC,GAAO,SAC7C,KAAM,IAAG,WAAW,GAAO,SAC3B,KAAM,IAAG,SAGX,GAAI,GACJ,AAAI,GAAG,eAAiB,SACtB,GAAkC,GAAG,MAAM,IAAI,4BAC/C,GAAG,MAAM,IAAI,2BAA4B,KAI3C,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,GAAO,OACvF,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,IAAI,SAAW,CAAC,GAAO,KAAK,IAAO,IAAM,KAAM,IAAO,QAAQ,KACrG,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,OAAO,SAAW,CAAC,GAAO,QAAQ,IAAO,OAAS,KAAM,IAAO,WAAW,KACjH,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,QAAQ,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,KAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,QAAQ,cAAc,EAAO,GAAO,MAAQ,GAC/F,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAAQ,GAChG,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,GAAO,KAAK,SACd,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAChE,EAAK,KAAO,KAAK,MAAM,KAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,KACZ,KAAM,GAAW,GAAO,KAAK,IAAI,SAAW,GAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,IAAU,GACrH,EAAK,UAAY,KAAK,MAAM,KAAQ,GAEpC,EAAY,KACZ,KAAM,GAAc,GAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,IAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,KAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,GAAG,SAAS,WAId,GAAG,MAAM,IAAI,2BAA4B,GAGzC,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,QAAU,GAClB,GAAQ,SAAW,GACnB,GAAQ,GAAK,GACb,GAAQ,QAAU,GAAI",
+  "sourcesContent": ["", "", "", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n  kernelMs: number|{error: string};\n  getExtraProfileInfo?(): string;  // a field for additional timing information\n                                   // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n  disposeData(dataId: DataId): void;\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void;\n  memory(): {unreliable: boolean;};  // Backend-specific information.\n  /** Returns number of data ids currently in the storage. */\n  numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage<T> {\n  private data = new WeakMap<DataId, T>();\n  private dataIdsCount = 0;\n\n  constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n  get(dataId: DataId) {\n    if (!this.data.has(dataId)) {\n      this.dataMover.moveData(this.backend, dataId);\n    }\n    return this.data.get(dataId);\n  }\n\n  set(dataId: DataId, value: T): void {\n    this.dataIdsCount++;\n    this.data.set(dataId, value);\n  }\n\n  has(dataId: DataId): boolean {\n    return this.data.has(dataId);\n  }\n\n  delete(dataId: DataId): boolean {\n    this.dataIdsCount--;\n    return this.data.delete(dataId);\n  }\n\n  numDataIds(): number {\n    return this.dataIdsCount;\n  }\n}\n\nexport interface DataMover {\n  /**\n   * To be called by backends whenever they see a dataId that they don't own.\n   * Upon calling this method, the mover will fetch the tensor from another\n   * backend and register it with the current active backend.\n   */\n  moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo>;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo> {\n    return notYetImplemented('time');\n  }\n  read(dataId: object): Promise<BackendValues> {\n    return notYetImplemented('read');\n  }\n  readSync(dataId: object): BackendValues {\n    return notYetImplemented('readSync');\n  }\n  numDataIds(): number {\n    return notYetImplemented('numDataIds');\n  }\n  disposeData(dataId: object): void {\n    return notYetImplemented('disposeData');\n  }\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    return notYetImplemented('write');\n  }\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    return notYetImplemented('move');\n  }\n  memory(): {unreliable: boolean; reasons?: string[]} {\n    return notYetImplemented('memory');\n  }\n  /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n  floatPrecision(): 16|32 {\n    return notYetImplemented('floatPrecision');\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    return notYetImplemented('batchMatMul');\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          FusedBatchMatMulConfig): Tensor3D {\n    return notYetImplemented('fusedBatchMatMul');\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    return notYetImplemented('slice');\n  }\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    return notYetImplemented('stridedSlice');\n  }\n  unstack(x: Tensor, axis: number): Tensor[] {\n    return notYetImplemented('unstack');\n  }\n  reverse<T extends Tensor>(a: T, axis: number[]): T {\n    return notYetImplemented('reverse');\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    return notYetImplemented('concat');\n  }\n\n  neg<T extends Tensor>(a: T): T {\n    return notYetImplemented('neg');\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('add');\n  }\n  addN<T extends Tensor>(tensors: T[]): T {\n    return notYetImplemented('addN');\n  }\n  subtract(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('subtract');\n  }\n  multiply(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('multiply');\n  }\n  realDivide(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('realDivide');\n  }\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('floorDiv');\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('sum');\n  }\n  prod(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('prod');\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    return notYetImplemented('unsortedSegmentSum');\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMin');\n  }\n  argMax(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMax');\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('equal');\n  }\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('notEqual');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('less');\n  }\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('lessEqual');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greater');\n  }\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greaterEqual');\n  }\n\n  logicalNot<T extends Tensor>(a: T): T {\n    return notYetImplemented('logicalNot');\n  }\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalAnd');\n  }\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalOr');\n  }\n\n  where(condition: Tensor): Tensor2D {\n    return notYetImplemented('where');\n  }\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('select');\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    return notYetImplemented('topk');\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('min');\n  }\n  minimum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('minimum');\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('mod');\n  }\n\n  max(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('max');\n  }\n  maximum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('maximum');\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('all');\n  }\n  any(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('any');\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('squaredDifference');\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    return notYetImplemented('ceil');\n  }\n  floor<T extends Tensor>(x: T): T {\n    return notYetImplemented('floor');\n  }\n  round<T extends Tensor>(x: T): T {\n    return notYetImplemented('round');\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    return notYetImplemented('sign');\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    return notYetImplemented('isNaN');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    return notYetImplemented('isInf');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    return notYetImplemented('isFinite');\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    return notYetImplemented('pow');\n  }\n  exp<T extends Tensor>(x: T): T {\n    return notYetImplemented('exp');\n  }\n  expm1<T extends Tensor>(x: T): T {\n    return notYetImplemented('expm1');\n  }\n  softmax<T extends Tensor>(x: T, dim: number): T {\n    return notYetImplemented('softmax');\n  }\n  log<T extends Tensor>(x: T): T {\n    return notYetImplemented('log');\n  }\n  log1p<T extends Tensor>(x: T): T {\n    return notYetImplemented('log1p');\n  }\n  sqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('sqrt');\n  }\n  rsqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('rsqrt');\n  }\n  square<T extends Tensor>(x: T): T {\n    return notYetImplemented('square');\n  }\n  reciprocal<T extends Tensor>(x: T): T {\n    return notYetImplemented('reciprocal');\n  }\n  relu<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu');\n  }\n  relu6<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu6');\n  }\n  prelu<T extends Tensor>(x: T, a: T): T {\n    return notYetImplemented('prelu');\n  }\n  elu<T extends Tensor>(x: T): T {\n    return notYetImplemented('elu');\n  }\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    return notYetImplemented('eluDer');\n  }\n  selu<T extends Tensor>(x: T): T {\n    return notYetImplemented('selu');\n  }\n  int<T extends Tensor>(x: T): T {\n    return notYetImplemented('int');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    return notYetImplemented('clip');\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    return notYetImplemented('abs');\n  }\n  complexAbs<T extends Tensor>(x: T): T {\n    return notYetImplemented('complexAbs');\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    return notYetImplemented('sigmoid');\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    return notYetImplemented('softplus');\n  }\n\n  sin<T extends Tensor>(x: T): T {\n    return notYetImplemented('sin');\n  }\n  cos<T extends Tensor>(x: T): T {\n    return notYetImplemented('cos');\n  }\n  tan<T extends Tensor>(x: T): T {\n    return notYetImplemented('tan');\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    return notYetImplemented('asin');\n  }\n  acos<T extends Tensor>(x: T): T {\n    return notYetImplemented('acos');\n  }\n  atan<T extends Tensor>(x: T): T {\n    return notYetImplemented('atan');\n  }\n  atan2<T extends Tensor>(a: T, b: T): T {\n    return notYetImplemented('atan2');\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('sinh');\n  }\n  cosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('cosh');\n  }\n  tanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('tanh');\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('asinh');\n  }\n  acosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('acosh');\n  }\n  atanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('atanh');\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    return notYetImplemented('erf');\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    return notYetImplemented('step');\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedConv2d');\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2d');\n  }\n  conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('conv2dDerInput');\n  }\n  conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2dDerFilter');\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedDepthwiseConv2D');\n  }\n\n  depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2D');\n  }\n  depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerInput');\n  }\n  depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerFilter');\n  }\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3d');\n  }\n  conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n      Tensor5D {\n    return notYetImplemented('conv3dDerInput');\n  }\n  conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3dDerFilter');\n  }\n  maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('maxPool');\n  }\n  maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('maxPoolBackprop');\n  }\n  avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPool');\n  }\n  avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPoolBackprop');\n  }\n  avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3d');\n  }\n  avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3dBackprop');\n  }\n  maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3d');\n  }\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3dBackprop');\n  }\n\n  reshape<T extends Tensor, R extends Rank>(x: T, shape: ShapeMap[R]):\n      Tensor<R> {\n    return notYetImplemented('reshape');\n  }\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return notYetImplemented('cast');\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    return notYetImplemented('tile');\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    return notYetImplemented('pad');\n  }\n\n  transpose<T extends Tensor>(x: T, perm: number[]): T {\n    return notYetImplemented('transpose');\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    return notYetImplemented('gather');\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    return notYetImplemented('gatherND');\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    return notYetImplemented('scatterND');\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    return notYetImplemented('batchToSpaceND');\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: number[][]): T {\n    return notYetImplemented('spaceToBatchND');\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeBilinear');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    return notYetImplemented('resizeBilinearBackprop');\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHEight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighbor');\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighborBackprop');\n  }\n\n  batchNorm(\n      x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n      offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n      varianceEpsilon?: number): Tensor4D {\n    return notYetImplemented('batchNorm');\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    return notYetImplemented('localResponseNormalization4D');\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n      bias: number, alpha: number, beta: number): Tensor4D {\n    return notYetImplemented('LRNGrad');\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    return notYetImplemented('multinomial');\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    return notYetImplemented('oneHot');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    return notYetImplemented('cumsum');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold?: number): Tensor1D {\n    return notYetImplemented('nonMaxSuppression');\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('fft');\n  }\n  ifft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('ifft');\n  }\n  complex<T extends Tensor>(real: T, imag: T): T {\n    return notYetImplemented('complex');\n  }\n  real<T extends Tensor>(input: T): T {\n    return notYetImplemented('real');\n  }\n  imag<T extends Tensor>(input: T): T {\n    return notYetImplemented('imag');\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    return notYetImplemented('cropAndResize');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n    return notYetImplemented('depthToSpace');\n  }\n\n  // Aligns with the \"SplitV\" kernel in TensorFlow.\n  split<T extends Tensor>(value: T, sizeSplits: number[], axis: number): T[] {\n    return notYetImplemented('split');\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    return notYetImplemented('sparseToDense');\n  }\n\n  diag(x: Tensor): Tensor {\n    return notYetImplemented('diag');\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    return notYetImplemented('fill');\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('onesLike');\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('zerosLike');\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return notYetImplemented('linspace');\n  }\n\n  dispose(): void {\n    return notYetImplemented('dispose');\n  }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n  throw new Error(\n      `'${kernelName}' not yet implemented or not found in the registry. ` +\n      `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise<FlagValue>);\nexport type Flags = {\n  [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n  evaluationFn: FlagEvaluationFn;\n  setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n  private flags: Flags = {};\n  private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n  private urlFlags: Flags = {};\n\n  platformName: string;\n  platform: Platform;\n\n  // tslint:disable-next-line: no-any\n  constructor(public global: any) {\n    this.populateURLFlags();\n  }\n\n  setPlatform(platformName: string, platform: Platform) {\n    if (this.platform != null) {\n      console.warn(\n          `Platform ${this.platformName} has already been set. ` +\n          `Overwriting the platform with ${platform}.`);\n    }\n    this.platformName = platformName;\n    this.platform = platform;\n  }\n\n  registerFlag(\n      flagName: string, evaluationFn: FlagEvaluationFn,\n      setHook?: (value: FlagValue) => void) {\n    this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n    // Override the flag value from the URL. This has to happen here because the\n    // environment is initialized before flags get registered.\n    if (this.urlFlags[flagName] != null) {\n      const flagValue = this.urlFlags[flagName];\n      console.warn(\n          `Setting feature override from URL ${flagName}: ${flagValue}.`);\n      this.set(flagName, flagValue);\n    }\n  }\n\n  async getAsync(flagName: string): Promise<FlagValue> {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    this.flags[flagName] = await this.evaluateFlag(flagName);\n    return this.flags[flagName];\n  }\n\n  get(flagName: string): FlagValue {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    const flagValue = this.evaluateFlag(flagName);\n    if (flagValue instanceof Promise) {\n      throw new Error(\n          `Flag ${flagName} cannot be synchronously evaluated. ` +\n          `Please use getAsync() instead.`);\n    }\n\n    this.flags[flagName] = flagValue;\n\n    return this.flags[flagName];\n  }\n\n  getNumber(flagName: string): number {\n    return this.get(flagName) as number;\n  }\n\n  getBool(flagName: string): boolean {\n    return this.get(flagName) as boolean;\n  }\n\n  getFlags(): Flags {\n    return this.flags;\n  }\n  // For backwards compatibility.\n  get features(): Flags {\n    return this.flags;\n  }\n\n  set(flagName: string, value: FlagValue): void {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot set flag ${flagName} as it has not been registered.`);\n    }\n    this.flags[flagName] = value;\n    if (this.flagRegistry[flagName].setHook != null) {\n      this.flagRegistry[flagName].setHook(value);\n    }\n  }\n\n  private evaluateFlag(flagName: string): FlagValue|Promise<FlagValue> {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n    }\n    return this.flagRegistry[flagName].evaluationFn();\n  }\n\n  setFlags(flags: Flags) {\n    this.flags = Object.assign({}, flags);\n  }\n\n  reset() {\n    this.flags = {};\n    this.urlFlags = {};\n    this.populateURLFlags();\n  }\n\n  private populateURLFlags(): void {\n    if (typeof this.global === 'undefined' ||\n        typeof this.global.location === 'undefined' ||\n        typeof this.global.location.search === 'undefined') {\n      return;\n    }\n\n    const urlParams = getQueryParams(this.global.location.search);\n    if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n      const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n      keyValues.forEach(keyValue => {\n        const [key, value] = keyValue.split(':') as [string, string];\n        this.urlFlags[key] = parseValue(key, value);\n      });\n    }\n  }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n  const params = {};\n  queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n    decodeParam(params, t[0], t[1]);\n    return t.join('=');\n  });\n  return params;\n}\n\nfunction decodeParam(\n    params: {[key: string]: string}, name: string, value?: string) {\n  params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n  value = value.toLowerCase();\n  if (value === 'true' || value === 'false') {\n    return value === 'true';\n  } else if (`${+ value}` === value) {\n    return +value;\n  }\n  throw new Error(\n      `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n  return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n  ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map<string, any>};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map<string, any>} {\n  if (globalNameSpace == null) {\n    // tslint:disable-next-line:no-any\n    let ns: any;\n    if (typeof (window) !== 'undefined') {\n      ns = window;\n    } else if (typeof (global) !== 'undefined') {\n      ns = global;\n    } else if (typeof (process) !== 'undefined') {\n      ns = process;\n    } else if (typeof (self) !== 'undefined') {\n      ns = self;\n    } else {\n      throw new Error('Could not find a global object');\n    }\n    globalNameSpace = ns;\n  }\n  return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map<string, any> {\n  const ns = getGlobalNamespace();\n  if (ns._tfGlobals == null) {\n    ns._tfGlobals = new Map();\n  }\n  return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n *             the first time it is fetched.\n */\nexport function getGlobal<T>(key: string, init: () => T): T {\n  const globalMap = getGlobalMap();\n  if (globalMap.has(key)) {\n    return globalMap.get(key);\n  } else {\n    const singleton = init();\n    globalMap.set(key, singleton);\n    return globalMap.get(key);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AllAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AnyAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMaxAttrs {\n  axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMinAttrs {\n  axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\nexport interface BatchMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BatchToSpaceNDAttrs {\n  blockShape: number[];\n  crops: number[][];\n}\n\nexport type BinaryInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BroadCastToAttrs {\n  shape: number[];\n  inputShape: number[];  // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n  dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n  clipValueMin: number;\n  clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick<NamedTensorInfoMap, 'real'|'imag'>;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n  axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'dy'>;\nexport interface Conv2DBackpropFilterAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'|'filter'>;\nexport interface Conv2DBackpropInputAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv3DAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'y'>;\n\nexport interface Conv3DBackpropFilterAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'>;\nexport interface Conv3DBackpropInputAttrs {\n  pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface CumsumAttrs {\n  axis: number;\n  exclusive: boolean;\n  reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n    Pick<NamedTensorInfoMap, 'image'|'boxes'|'boxInd'>;\nexport interface CropAndResizeAttrs {\n  cropSize: [number, number];\n  method: 'bilinear'|'nearest';\n  extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface DepthToSpaceAttrs {\n  blockSize: number;\n  dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface DepthwiseConv2dNativeAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n    'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'dy'>;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n    'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'dy'>;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Dilation2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick<NamedTensorInfoMap, 'dy'|'y'>;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n  shape: number[];\n  value: number|string;\n  dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick<NamedTensorInfoMap, 'image'>;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n    Pick<NamedTensorInfoMap, 'x'|'scale'|'offset'|'mean'|'variance'>;\nexport interface FusedBatchNormAttrs {\n  varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick<NamedTensorInfoMap, 'x'|'indices'>;\nexport interface GatherV2Attrs {\n  axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick<NamedTensorInfoMap, 'params'|'indices'>;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n  start: number;\n  stop: number;\n  num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface LogSoftmaxAttrs {\n  axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface LRNAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick<NamedTensorInfoMap, 'x'|'y'|'dy'>;\nexport interface LRNBackpropAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxAttrs {\n  reductionIndices: number|number[];\n  keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolWithArgmaxAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MeanAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MinAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV3Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV4Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV5Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick<NamedTensorInfoMap, 'indices'>;\nexport interface OneHotAttrs {\n  depth: number;\n  onValue: number;\n  offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface PadV2Attrs {\n  paddings: Array<[number, number]>;\n  constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick<NamedTensorInfoMap, 'x'|'alpha'>;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ProdAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n  start: number;\n  stop: number;\n  step: number;\n  dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReshapeAttrs {\n  shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeNearestNeighborAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n    Pick<NamedTensorInfoMap, 'images'>;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeBilinearAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick<NamedTensorInfoMap, 'images'>;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReverseAttrs {\n  dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick<NamedTensorInfoMap, 'indices'|'updates'>;\nexport interface ScatterNdAttrs {\n  shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick<NamedTensorInfoMap, 'condition'|'t'|'e'>;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SliceAttrs {\n  begin: number|number[];\n  size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SumAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SpaceToBatchNDAttrs {\n  blockShape: number[];\n  paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SplitVAttrs {\n  numOrSizeSplits: number[]|number;\n  axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface SoftmaxAttrs {\n  dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n    Pick<NamedTensorInfoMap, 'sparseIndices'|'sparseValues'|'defaultValue'>;\nexport interface SparseToDenseAttrs {\n  outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface StridedSliceAttrs {\n  begin: number[];\n  end: number[];\n  strides: number[];\n  beginMask: number;\n  endMask: number;\n  ellipsisMask: number;\n  newAxisMask: number;\n  shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TileAttrs {\n  reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TopKAttrs {\n  k: number;\n  sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TransposeAttrs {\n  perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface UniqueAttrs {\n  axis: number;\n}\n\nexport type UnaryInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick<NamedTensorInfoMap, 'value'>;\nexport interface UnpackAttrs {\n  axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n    Pick<NamedTensorInfoMap, 'x'|'segmentIds'>;\nexport interface UnsortedSegmentSumAttrs {\n  numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n  alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n  pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n      HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n  numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick<NamedTensorInfoMap, 'image'>;\nexport interface RotateWithOffsetAttrs {\n  radians: number;\n  fillValue: number|[number, number, number];\n  center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n  a: TensorInfo;\n  b: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n  activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n    getGlobal('kernelRegistry', () => new Map<string, KernelConfig>());\nconst gradRegistry =\n    getGlobal('gradRegistry', () => new Map<string, GradConfig>());\n\nexport type DataId = object;\n\ntype AttributeValue =\n    number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray<AttributeValue>;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n  inputs: NamedTensorInfoMap,\n  backend: {},\n  attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n    (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n        NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n  kernelName: string;\n  backendName: string;\n  kernelFunc: KernelFunc;\n  setupFunc?: KernelSetupFunc;\n  disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n  kernelName: string;\n  inputsToSave?: string[];\n  // When saveAllInputs is true, all inputs will be saved. Only use this flag\n  // if inputs is an array of Tensors.\n  saveAllInputs?: boolean;\n  outputsToSave?: boolean[];\n  gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n  dataId: DataId;\n  shape: number[];\n  dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n  [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n  [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n    kernelName: string, backendName: string): KernelConfig {\n  const key = makeKey(kernelName, backendName);\n  return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n  return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n  const it = kernelRegistry.entries();\n  const result: KernelConfig[] = [];\n\n  while (true) {\n    const {done, value} = it.next();\n    if (done) {\n      break;\n    }\n    const [key, config] = value;\n    const [backend, ] = key.split('_');\n    if (backend === backendName) {\n      result.push(config);\n    }\n  }\n  return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n  const {kernelName, backendName} = config;\n  const key = makeKey(kernelName, backendName);\n  if (kernelRegistry.has(key)) {\n    console.warn(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is already registered`);\n  }\n  kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n  const {kernelName} = config;\n\n  if (gradRegistry.has(kernelName)) {\n    // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n    // to debug mode.\n    if (env().getBool('DEBUG')) {\n      console.warn(`Overriding the gradient for '${kernelName}'`);\n    }\n  }\n  gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n    kernelName: string, backendName: string): void {\n  const key = makeKey(kernelName, backendName);\n  if (!kernelRegistry.has(key)) {\n    throw new Error(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is not registered`);\n  }\n  kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n  if (!gradRegistry.has(kernelName)) {\n    throw new Error(\n        `The gradient '${kernelName}' for backend is not registered`);\n  }\n  gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n    registeredBackendName: string, newBackendName: string): void {\n  const kernels = getKernelsForBackend(registeredBackendName);\n  kernels.forEach(kernelConfig => {\n    const newKernelConfig =\n        Object.assign({}, kernelConfig, {backendName: newBackendName});\n    registerKernel(newKernelConfig);\n  });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n  return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n                        Float32Array): void {\n  let counter = array.length;\n  let temp = 0;\n  let index = 0;\n  // While there are elements in the array\n  while (counter > 0) {\n    // Pick a random index\n    index = (Math.random() * counter) | 0;\n    // Decrease counter by 1\n    counter--;\n    // And swap the last element with it\n    temp = array[counter];\n    array[counter] = array[index];\n    array[index] = temp;\n  }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n  return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n  return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n  let sum = 0;\n  for (let i = 0; i < arr.length; i++) {\n    sum += arr[i];\n  }\n  return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n  const r = Math.random();\n  return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n  let result = 0;\n  for (let i = 0; i < a.length; i++) {\n    const diff = Number(a[i]) - Number(b[i]);\n    result += diff * diff;\n  }\n  return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n *     error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n  if (!expr) {\n    throw new Error(typeof msg === 'string' ? msg : msg());\n  }\n}\n\nexport function assertShapesMatch(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  assert(\n      arraysEqual(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n  assert(\n      a != null,\n      () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten<number>().\n/**\n *  Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n *  @param arr The nested array to flatten.\n *  @param result The destination array which holds the elements.\n *  @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n *      to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten<T extends number|boolean|string|Promise<number>|TypedArray>(\n    arr: T|RecursiveArray<T>, result: T[] = [], skipTypedArray = false): T[] {\n  if (result == null) {\n    result = [];\n  }\n  if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n    for (let i = 0; i < arr.length; ++i) {\n      flatten(arr[i], result, skipTypedArray);\n    }\n  } else {\n    result.push(arr as T);\n  }\n  return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n  if (shape.length === 0) {\n    // Scalar.\n    return 1;\n  }\n  let size = shape[0];\n  for (let i = 1; i < shape.length; i++) {\n    size *= shape[i];\n  }\n  return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n  return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n  if (n1 === n2) {\n    return true;\n  }\n  if (n1 == null || n2 == null) {\n    return false;\n  }\n\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function isInt(a: number): boolean {\n  return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n  // tslint:disable-next-line:no-any\n  if ((Math as any).tanh != null) {\n    // tslint:disable-next-line:no-any\n    return (Math as any).tanh(x);\n  }\n  if (x === Infinity) {\n    return 1;\n  } else if (x === -Infinity) {\n    return -1;\n  } else {\n    const e2x = Math.exp(2 * x);\n    return (e2x - 1) / (e2x + 1);\n  }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n  const width = Math.ceil(Math.sqrt(size));\n  return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n  const shuffledIndices = new Uint32Array(n);\n  for (let i = 0; i < n; ++i) {\n    shuffledIndices[i] = i;\n  }\n  shuffle(shuffledIndices);\n  return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n  if (size <= a.length) {\n    return a;\n  }\n  return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n    checkFn: () => boolean, delayFn = (counter: number) => 0,\n    maxCounter?: number): Promise<void> {\n  return new Promise<void>((resolve, reject) => {\n    let tryCount = 0;\n\n    const tryFn = () => {\n      if (checkFn()) {\n        resolve();\n        return;\n      }\n\n      tryCount++;\n\n      const nextBackoff = delayFn(tryCount);\n\n      if (maxCounter != null && tryCount >= maxCounter) {\n        reject();\n        return;\n      }\n      setTimeout(tryFn, nextBackoff);\n    };\n\n    tryFn();\n  });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n    shape: number[], size: number): number[] {\n  let shapeProd = 1;\n  let implicitIdx = -1;\n\n  for (let i = 0; i < shape.length; ++i) {\n    if (shape[i] >= 0) {\n      shapeProd *= shape[i];\n    } else if (shape[i] === -1) {\n      if (implicitIdx !== -1) {\n        throw Error(\n            `Shapes can only have 1 implicit size. ` +\n            `Found -1 at dim ${implicitIdx} and dim ${i}`);\n      }\n      implicitIdx = i;\n    } else if (shape[i] < 0) {\n      throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n    }\n  }\n\n  if (implicitIdx === -1) {\n    if (size > 0 && size !== shapeProd) {\n      throw Error(`Size(${size}) must match the product of shape ${shape}`);\n    }\n    return shape;\n  }\n\n  if (shapeProd === 0) {\n    throw Error(\n        `Cannot infer the missing size in [${shape}] when ` +\n        `there are 0 elements`);\n  }\n  if (size % shapeProd !== 0) {\n    throw Error(\n        `The implicit shape can't be a fractional number. ` +\n        `Got ${size} / ${shapeProd}`);\n  }\n\n  const newShape = shape.slice();\n  newShape[implicitIdx] = size / shapeProd;\n  return newShape;\n}\n\nexport function parseAxisParam(\n    axis: number|number[], shape: number[]): number[] {\n  const rank = shape.length;\n\n  // Normalize input\n  axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n  // Check for valid range\n  assert(\n      axis.every(ax => ax >= -rank && ax < rank),\n      () =>\n          `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n          `got axis ${axis}`);\n\n  // Check for only integers\n  assert(\n      axis.every(ax => isInt(ax)),\n      () => `All values in axis param must be integers but ` +\n          `got axis ${axis}`);\n\n  // Handle negative axis.\n  return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n    {newShape: number[], keptDims: number[]} {\n  const newShape: number[] = [];\n  const keptDims: number[] = [];\n  const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n  const axes = (axis == null || isEmptyArray) ?\n      null :\n      parseAxisParam(axis, shape).sort();\n  let j = 0;\n  for (let i = 0; i < shape.length; ++i) {\n    if (axes != null) {\n      if (axes[j] === i && shape[i] !== 1) {\n        throw new Error(\n            `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n      }\n      if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n        newShape.push(shape[i]);\n        keptDims.push(i);\n      }\n      if (axes[j] <= i) {\n        j++;\n      }\n    }\n    if (shape[i] !== 1) {\n      newShape.push(shape[i]);\n      keptDims.push(i);\n    }\n  }\n  return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType<D extends NumericDataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType<D extends DataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else if (dtype === 'string') {\n    values = new Array<'string'>(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors<D extends DataType>(\n    vals: DataTypeMap[D]|number[], dtype: D): void {\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n    }\n  }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n  return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n      dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n  if (newType === 'complex64') {\n    return false;\n  }\n  if (newType === 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'bool' && oldType === 'bool') {\n    return false;\n  }\n  return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n  return a instanceof Float32Array || a instanceof Int32Array ||\n      a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n  if (dtype === 'float32' || dtype === 'int32') {\n    return 4;\n  } else if (dtype === 'complex64') {\n    return 8;\n  } else if (dtype === 'bool') {\n    return 1;\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n  if (arr == null) {\n    return 0;\n  }\n  let bytes = 0;\n  arr.forEach(x => bytes += x.length);\n  return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n  return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n  return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n  return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n  if (Array.isArray(values)) {\n    return inferDtype(values[0]);\n  }\n  if (values instanceof Float32Array) {\n    return 'float32';\n  } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n    return 'int32';\n  } else if (isNumber(values)) {\n    return 'float32';\n  } else if (isString(values)) {\n    return 'string';\n  } else if (isBoolean(values)) {\n    return 'bool';\n  }\n  return 'float32';\n}\n\nexport function isFunction(f: Function) {\n  return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n  for (let i = start; i < size; ++i) {\n    if (size % i === 0) {\n      return i;\n    }\n  }\n  return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n  const rank = shape.length;\n  if (rank < 2) {\n    return [];\n  }\n\n  // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n  // strides.\n  const strides = new Array(rank - 1);\n  strides[rank - 2] = shape[rank - 1];\n  for (let i = rank - 3; i >= 0; --i) {\n    strides[i] = strides[i + 1] * shape[i + 1];\n  }\n  return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n    value: DataType, dtype: DataType): BackendValues {\n  if (dtype === 'string') {\n    return encodeString(value);\n  }\n\n  return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n  if (dtype === 'string') {\n    throw new Error('Cannot convert a string[] to a TypedArray');\n  }\n  if (Array.isArray(a)) {\n    a = flatten(a);\n  }\n\n  if (env().getBool('DEBUG')) {\n    checkConversionForErrors(a as number[], dtype);\n  }\n  if (noConversionNeeded(a, dtype)) {\n    return a as TypedArray;\n  }\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(a as number[]);\n  } else if (dtype === 'int32') {\n    return new Int32Array(a as number[]);\n  } else if (dtype === 'bool') {\n    const bool = new Uint8Array((a as number[]).length);\n    for (let i = 0; i < bool.length; ++i) {\n      if (Math.round((a as number[])[i]) !== 0) {\n        bool[i] = 1;\n      }\n    }\n    return bool;\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n  const ret = new Array();\n  if (shape.length === 1) {\n    const d = shape[0];\n    for (let i = 0; i < d; i++) {\n      ret[i] = a[offset + i];\n    }\n  } else {\n    const d = shape[0];\n    const rest = shape.slice(1);\n    const len = rest.reduce((acc, c) => acc * c);\n    for (let i = 0; i < d; i++) {\n      ret[i] = createNestedArray(offset + i * len, rest, a);\n    }\n  }\n  return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n  if (shape.length === 0) {\n    // Scalar type should return a single number.\n    return a[0];\n  }\n  const size = shape.reduce((acc, c) => acc * c);\n  if (size === 0) {\n    // A tensor with shape zero should be turned into empty list.\n    return [];\n  }\n  if (size !== a.length) {\n    throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n  }\n\n  return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n  return (a instanceof Float32Array && dtype === 'float32') ||\n      (a instanceof Int32Array && dtype === 'int32') ||\n      (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  const array = makeZerosTypedArray(size, dtype);\n  for (let i = 0; i < array.length; i++) {\n    array[i] = 1;\n  }\n  return array;\n}\n\nexport function makeZerosTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'int32') {\n    return new Int32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'bool') {\n    return new Uint8Array(size) as DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray<D extends DataType>(\n    shape: number[], dtype: D) {\n  const size = shape.reduce((prev, curr) => prev * curr, 1);\n  if (dtype == null || dtype === 'float32') {\n    return toNestedArray(shape, new Float32Array(size));\n  } else if (dtype === 'int32') {\n    return toNestedArray(shape, new Int32Array(size));\n  } else if (dtype === 'bool') {\n    return toNestedArray(shape, new Uint8Array(size));\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n  return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n  shape.forEach(dimSize => {\n    assert(\n        Number.isInteger(dimSize) && dimSize >= 0,\n        () =>\n            `Tensor must have a shape comprised of positive integers but got ` +\n            `shape [${shape}].`);\n  });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n    path: string, requestInits?: RequestInit): Promise<Response> {\n  return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n  encoding = encoding || 'utf-8';\n  return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n  encoding = encoding || 'utf-8';\n  return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n    locs: number[], rank: number, strides: number[]): number {\n  if (rank === 0) {\n    return 0;\n  } else if (rank === 1) {\n    return locs[0];\n  }\n  let index = locs[locs.length - 1];\n  for (let i = 0; i < locs.length - 1; ++i) {\n    index += strides[i] * locs[i];\n  }\n  return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n    index: number, rank: number, strides: number[]): number[] {\n  if (rank === 0) {\n    return [];\n  } else if (rank === 1) {\n    return [index];\n  }\n  const locs: number[] = new Array(rank);\n  for (let i = 0; i < locs.length - 1; ++i) {\n    locs[i] = Math.floor(index / strides[i]);\n    index -= locs[i] * strides[i];\n  }\n  locs[locs.length - 1] = index;\n  return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n  kernelName: string,\n  outputs: Tensor[],\n  inputs: NamedTensorMap,\n  timeMs: Promise<number|{error: string}>,\n  extraInfo: Promise<string>\n};\n\nexport class Profiler {\n  constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n    if (logger == null) {\n      this.logger = new Logger();\n    }\n  }\n\n  profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n      KernelProfile {\n    let outputs: Tensor[];\n    const holdResultWrapperFn = () => {\n      outputs = f();\n    };\n    const timer = this.backendTimer.time(holdResultWrapperFn);\n\n    for (let i = 0; i < outputs.length; i++) {\n      const output = outputs[i];\n      // Dangling promise here because we don't want to propagate up\n      // asynchronicity.\n      output.data().then(tensorVals => {\n        checkComputationForErrors(tensorVals, output.dtype, kernelName);\n      });\n    }\n\n    const kernelProfile = {\n      kernelName,\n      outputs,\n      inputs,\n      timeMs: timer.then(timing => timing.kernelMs),\n      extraInfo: timer.then(\n          timing => timing.getExtraProfileInfo != null ?\n              timing.getExtraProfileInfo() :\n              '')\n    };\n    return kernelProfile;\n  }\n\n  logKernelProfile(kernelProfile: KernelProfile): void {\n    const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n    outputs.forEach(result => {\n      Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n        this.logger.logKernelProfile(\n            kernelName, result, valueContainer[0], valueContainer[1], inputs,\n            valueContainer[2]);\n      });\n    });\n  }\n}\n\nexport function checkComputationForErrors<D extends DataType>(\n    vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n  if (dtype !== 'float32') {\n    // Only floating point computations will generate NaN values\n    return false;\n  }\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      // Throwing custom exception so behavior is testable.\n      console.warn(`Found ${num} in the result of '${kernelName}'`);\n      return true;\n    }\n  }\n  return false;\n}\n\nexport class Logger {\n  logKernelProfile(\n      name: string, result: Tensor, vals: TypedArray,\n      timeMs: number|{error: string}, inputs: NamedTensorMap,\n      extraInfo?: string) {\n    const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n                                              timeMs['error'];\n    const paddedName = util.rightPad(name, 25);\n    const rank = result.rank;\n    const size = result.size;\n    const shape = util.rightPad(result.shape.toString(), 14);\n    let inputShapesDescription = '';\n\n    for (const name in inputs) {\n      const input = inputs[name];\n      if (input != null) {\n        // The input might be a non-tensor (e.g HTMLImageElement), in which case\n        // we claim the output shape as input shape.\n        const inputShape = input.shape || result.shape;\n        const inputRank = inputShape.length;\n        inputShapesDescription +=\n            `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n      }\n    }\n\n    console.log(\n        `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n            inputShapesDescription}\\t%c${extraInfo}`,\n        'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n        'color: green', 'color: steelblue');\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n  id: number;\n  kernelName: string;\n  outputs: Tensor[];\n  inputs: NamedTensorMap;\n  // Optional params, defined only for ops with gradient impl.\n  gradient?: (dys: Tensor[]) => NamedGradientMap;\n  saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n  [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n    tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n  // Forward pass to compute all the nodes and Tensors that are transitively a\n  // function of x.\n  const tensorsFromX: {[tensorId: number]: boolean} = {};\n  const nodesFromX: {[nodeId: number]: boolean} = {};\n  for (let i = 0; i < xs.length; i++) {\n    tensorsFromX[xs[i].id] = true;\n  }\n\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n    for (const inputName in nodeInputs) {\n      const input = nodeInputs[inputName];\n\n      let anyInputFromX = false;\n      for (let j = 0; j < xs.length; j++) {\n        if (tensorsFromX[input.id]) {\n          node.outputs.forEach(output => tensorsFromX[output.id] = true);\n          anyInputFromX = true;\n          nodesFromX[node.id] = true;\n          break;\n        }\n      }\n\n      if (anyInputFromX) {\n        break;\n      }\n    }\n  }\n\n  // Backward pass to find all of the nodes and Tensors that lead to y.\n  const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n  tensorsLeadToY[y.id] = true;\n  const nodesToY: {[nodeId: number]: boolean} = {};\n\n  for (let i = tape.length - 1; i >= 0; i--) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n\n    // If any of the outputs lead to y, mark all of the inputs as leading to y.\n    for (let j = 0; j < node.outputs.length; j++) {\n      if (tensorsLeadToY[node.outputs[j].id]) {\n        for (const inputName in nodeInputs) {\n          tensorsLeadToY[nodeInputs[inputName].id] = true;\n          nodesToY[node.id] = true;\n        }\n        break;\n      }\n    }\n  }\n\n  // Return the paths that come from x and lead to y.\n  const filteredTape: TapeNode[] = [];\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n\n    if (nodesFromX[node.id] && nodesToY[node.id]) {\n      // Prune the inputs from the node that aren't a function of x.\n      const prunedInputs: {[inputName: string]: Tensor} = {};\n      for (const inputName in node.inputs) {\n        const nodeInput = node.inputs[inputName];\n        if (tensorsFromX[nodeInput.id]) {\n          prunedInputs[inputName] = nodeInput;\n        }\n      }\n\n      // Copy the node and overwrite inputsAndArgs to the pruned version.\n      const prunedNode = Object.assign({}, node);\n      prunedNode.inputs = prunedInputs;\n      prunedNode.outputs = node.outputs;\n\n      filteredTape.push(prunedNode);\n    }\n  }\n\n  return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n    tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n    filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n    add: (a: Tensor, b: Tensor) => Tensor) {\n  // Walk the tape backward and keep a map of Tensor to its gradient.\n  for (let i = filteredTape.length - 1; i >= 0; i--) {\n    const node = filteredTape[i];\n\n    const dys: Tensor[] = [];\n    node.outputs.forEach(o => {\n      const gradTensor = tensorAccumulatedGradientMap[o.id];\n      if (gradTensor != null) {\n        dys.push(gradTensor);\n      } else {\n        // This particular output is not in the back-propagation subgraph, so it\n        // does not affect the final output, thus we put null for its dy.\n        dys.push(null);\n      }\n    });\n\n    if (node.gradient == null) {\n      throw new Error(\n          `Cannot compute gradient: gradient function not found ` +\n          `for ${node.kernelName}.`);\n    }\n\n    // Backprop dy through this node and accumulate gradients over the inputs.\n    const inputGradients = node.gradient(dys);\n\n    for (const inputName in node.inputs) {\n      if (!(inputName in inputGradients)) {\n        throw new Error(\n            `Cannot backprop through input ${inputName}. ` +\n            `Available gradients found: ${Object.keys(inputGradients)}.`);\n      }\n\n      // Call the gradient function.\n      const dx = tidy(() => inputGradients[inputName]());\n      if (dx.dtype !== 'float32') {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n      }\n      const x = node.inputs[inputName];\n      if (!util.arraysEqual(dx.shape, x.shape)) {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `'${inputName}' has shape '${dx.shape}', which does not match ` +\n            `the shape of the input '${x.shape}'`);\n      }\n\n      if (tensorAccumulatedGradientMap[x.id] == null) {\n        tensorAccumulatedGradientMap[x.id] = dx;\n      } else {\n        const curGradient = tensorAccumulatedGradientMap[x.id];\n        tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n        curGradient.dispose();\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    verbose: boolean) {\n  const strides = computeStrides(shape);\n  const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n  const rank = shape.length;\n  const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n  const lines = ['Tensor'];\n  if (verbose) {\n    lines.push(`  dtype: ${dtype}`);\n    lines.push(`  rank: ${rank}`);\n    lines.push(`  shape: [${shape}]`);\n    lines.push(`  values:`);\n  }\n  lines.push(valsLines.map(l => '    ' + l).join('\\n'));\n  return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[]): number[] {\n  const n = sizeFromShape(shape);\n  const numCols = strides[strides.length - 1];\n  const padPerCol = new Array(numCols).fill(0);\n  const rank = shape.length;\n  const valuesOrTuples =\n      dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n  if (rank > 1) {\n    for (let row = 0; row < n / numCols; row++) {\n      const offset = row * numCols;\n      for (let j = 0; j < numCols; j++) {\n        padPerCol[j] = Math.max(\n            padPerCol[j],\n            valToString(valuesOrTuples[offset + j], 0, dtype).length);\n      }\n    }\n  }\n  return padPerCol;\n}\n\nfunction valToString(\n    val: number|string|[number, number], pad: number, dtype: DataType) {\n  let valStr: string;\n  if (Array.isArray(val)) {\n    valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n        `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n  } else if (isString(val)) {\n    valStr = `'${val}'`;\n  } else if (dtype === 'bool') {\n    valStr = boolNumToString(val);\n  } else {\n    valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n  }\n\n  return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n  return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[], padPerCol: number[], isLast = true): string[] {\n  const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n  const size = shape[0];\n  const rank = shape.length;\n  if (rank === 0) {\n    if (dtype === 'complex64') {\n      const complexTuple = createComplexTuples(vals);\n      return [valToString(complexTuple[0], 0, dtype)];\n    }\n    if (dtype === 'bool') {\n      return [boolNumToString(vals[0] as number)];\n    }\n    return [vals[0].toString()];\n  }\n\n  if (rank === 1) {\n    if (size > FORMAT_LIMIT_NUM_VALS) {\n      const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n      let firstVals = Array.from<number|string|[number, number]>(\n          vals.slice(0, firstValsSize));\n      let lastVals = Array.from<number|string|[number, number]>(vals.slice(\n          (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n          size * storagePerElement));\n      if (dtype === 'complex64') {\n        firstVals = createComplexTuples(firstVals);\n        lastVals = createComplexTuples(lastVals);\n      }\n      return [\n        '[' +\n        firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n            .join(', ') +\n        ', ..., ' +\n        lastVals\n            .map(\n                (x, i) => valToString(\n                    x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n            .join(', ') +\n        ']'\n      ];\n    }\n    const displayVals: Array<number|string|[number, number]> =\n        dtype === 'complex64' ? createComplexTuples(vals) :\n                                Array.from<number|string>(vals);\n\n    return [\n      '[' +\n      displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n          .join(', ') +\n      ']'\n    ];\n  }\n\n  // The array is rank 2 or more.\n  const subshape = shape.slice(1);\n  const substrides = strides.slice(1);\n  const stride = strides[0] * storagePerElement;\n  const lines: string[] = [];\n  if (size > FORMAT_LIMIT_NUM_VALS) {\n    for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          false /* isLast */));\n    }\n    lines.push('...');\n    for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  } else {\n    for (let i = 0; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  }\n  const sep = rank === 2 ? ',' : '';\n  lines[0] = '[' + lines[0] + sep;\n  for (let i = 1; i < lines.length - 1; i++) {\n    lines[i] = ' ' + lines[i] + sep;\n  }\n  let newLineSep = ',\\n';\n  for (let i = 2; i < rank; i++) {\n    newLineSep += '\\n';\n  }\n  lines[lines.length - 1] =\n      ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n  return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n                             TypedArray): Array<[number, number]> {\n  const complexTuples: Array<[number, number]> = [];\n  for (let i = 0; i < vals.length; i += 2) {\n    complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n  }\n  return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData<D extends DataType> {\n  dataId?: DataId;\n  values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer<R extends Rank, D extends DataType = 'float32'> {\n  size: number;\n  shape: ShapeMap[R];\n  strides: number[];\n  values: DataTypeMap[D];\n\n  constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.size = util.sizeFromShape(shape);\n\n    if (values != null) {\n      const n = values.length;\n      util.assert(\n          n === this.size,\n          () => `Length of values '${n}' does not match the size ` +\n              `inferred by the shape '${this.size}'.`);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `complex64 dtype TensorBuffers are not supported. Please create ` +\n          `a TensorBuffer for the real and imaginary parts separately and ` +\n          `call tf.complex(real, imag).`);\n    }\n    this.values = values || util.getArrayFromDType(dtype, this.size);\n    this.strides = computeStrides(shape);\n  }\n\n  /**\n   * Sets a value in the buffer at a given location.\n   *\n   * @param value The value to set.\n   * @param locs  The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  set(value: SingleValueMap[D], ...locs: number[]): void {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    util.assert(\n        locs.length === this.rank,\n        () => `The number of provided coordinates (${locs.length}) must ` +\n            `match the rank (${this.rank})`);\n\n    const index = this.locToIndex(locs);\n    this.values[index] = value as number;\n  }\n\n  /**\n   * Returns the value in the buffer at the provided location.\n   *\n   * @param locs The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  get(...locs: number[]): SingleValueMap[D] {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    let i = 0;\n    for (const loc of locs) {\n      if (loc < 0 || loc >= this.shape[i]) {\n        const msg = `Requested out of range element at ${locs}. ` +\n            `  Buffer shape=${this.shape}`;\n        throw new Error(msg);\n      }\n      i++;\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return this.values[index] as SingleValueMap[D];\n  }\n\n  locToIndex(locs: number[]): number {\n    if (this.rank === 0) {\n      return 0;\n    } else if (this.rank === 1) {\n      return locs[0];\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return index;\n  }\n\n  indexToLoc(index: number): number[] {\n    if (this.rank === 0) {\n      return [];\n    } else if (this.rank === 1) {\n      return [index];\n    }\n    const locs: number[] = new Array(this.shape.length);\n    for (let i = 0; i < locs.length - 1; ++i) {\n      locs[i] = Math.floor(index / this.strides[i]);\n      index -= locs[i] * this.strides[i];\n    }\n    locs[locs.length - 1] = index;\n    return locs;\n  }\n\n  get rank() {\n    return this.shape.length;\n  }\n\n  /**\n   * Creates an immutable `tf.Tensor` object from the buffer.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  toTensor(): Tensor<R> {\n    return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n        Tensor<R>;\n  }\n}\n\nexport interface TensorTracker {\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: Backend): Tensor;\n  makeVariable(\n      initialValue: Tensor, trainable?: boolean, name?: string,\n      dtype?: DataType): Variable;\n  incRef(a: Tensor, backend: Backend): void;\n  disposeTensor(t: Tensor): void;\n  disposeVariable(v: Variable): void;\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n  cast<T extends Tensor>(x: T, dtype: DataType): T;\n  buffer<R extends Rank, D extends DataType>(\n      shape: ShapeMap[R], dtype: D,\n      values?: DataTypeMap[D]): TensorBuffer<R, D>;\n  print<T extends Tensor>(x: T, verbose: boolean): void;\n  clone<T extends Tensor>(x: T): T;\n  // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n  trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n  opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n  deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object;  // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor<R extends Rank = Rank> {\n  /** Unique id of this tensor. */\n  readonly id: number;\n  /**\n   * Id of the bucket holding the data for this tensor. Multiple arrays can\n   * point to the same bucket (e.g. when calling array.reshape()).\n   */\n  dataId: DataId;\n  /** The shape of the tensor. */\n  readonly shape: ShapeMap[R];\n  /** Number of elements in the tensor. */\n  readonly size: number;\n  /** The data type for the array. */\n  readonly dtype: DataType;\n  /** The rank type for the array (see `Rank` enum). */\n  readonly rankType: R;\n\n  /** Whether this tensor has been globally kept. */\n  kept = false;\n  /** The id of the scope this tensor is being tracked in. */\n  scopeId: number;\n\n  /**\n   * Number of elements to skip in each dimension when indexing. See\n   * https://docs.scipy.org/doc/numpy/reference/generated/\\\n   * numpy.ndarray.strides.html\n   */\n  readonly strides: number[];\n\n  constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.dtype = dtype || 'float32';\n    this.size = util.sizeFromShape(shape);\n    this.strides = computeStrides(shape);\n    this.dataId = dataId;\n    this.id = id;\n    this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n  }\n\n  get rank(): number {\n    return this.shape.length;\n  }\n\n  /**\n   * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async buffer<D extends DataType = 'float32'>(): Promise<TensorBuffer<R, D>> {\n    const vals = await this.data<D>();\n    return opHandler.buffer(this.shape, this.dtype as D, vals);\n  }\n\n  /**\n   * Returns a `tf.TensorBuffer` that holds the underlying data.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  bufferSync<D extends DataType = 'float32'>(): TensorBuffer<R, D> {\n    return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * asynchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async array(): Promise<ArrayMap[R]> {\n    const vals = await this.data();\n    return toNestedArray(this.shape, vals) as ArrayMap[R];\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * synchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  arraySync(): ArrayMap[R] {\n    return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n  }\n\n  /**\n   * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n   * promise of `TypedArray` that resolves when the computation has finished.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]> {\n    this.throwIfDisposed();\n    const data = trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      const bytes = await data as Uint8Array[];\n      try {\n        return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as Promise<DataTypeMap[D]>;\n  }\n\n  /**\n   * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n   * UI thread until the values are ready, which can cause performance issues.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D] {\n    this.throwIfDisposed();\n    const data = trackerFn().readSync(this.dataId);\n    if (this.dtype === 'string') {\n      try {\n        return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n            DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as DataTypeMap[D];\n  }\n\n  /** Returns the underlying bytes of the tensor's data. */\n  async bytes(): Promise<Uint8Array[]|Uint8Array> {\n    this.throwIfDisposed();\n    const data = await trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      return data as Uint8Array[];\n    } else {\n      return new Uint8Array((data as TypedArray).buffer);\n    }\n  }\n\n  /**\n   * Disposes `tf.Tensor` from memory.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dispose(): void {\n    if (this.isDisposed) {\n      return;\n    }\n    trackerFn().disposeTensor(this);\n    this.isDisposedInternal = true;\n  }\n\n  protected isDisposedInternal = false;\n  get isDisposed(): boolean {\n    return this.isDisposedInternal;\n  }\n\n  throwIfDisposed() {\n    if (this.isDisposed) {\n      throw new Error(`Tensor is disposed.`);\n    }\n  }\n\n  /**\n   * Prints the `tf.Tensor`. See `tf.print` for details.\n   *\n   * @param verbose Whether to print verbose information about the tensor,\n   *    including dtype and size.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  print(verbose = false): void {\n    return opHandler.print(this, verbose);\n  }\n\n  /**\n   * Returns a copy of the tensor. See `tf.clone` for details.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  clone<T extends Tensor>(this: T): T {\n    this.throwIfDisposed();\n    return opHandler.clone(this);\n  }\n\n  /**\n   * Returns a human-readable description of the tensor. Useful for logging.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  toString(verbose = false): string {\n    const vals = this.dataSync();\n    return tensorToString(vals, this.shape, this.dtype, verbose);\n  }\n\n  cast<T extends this>(dtype: DataType): T {\n    this.throwIfDisposed();\n    return opHandler.cast(this as T, dtype);\n  }\n  variable(trainable = true, name?: string, dtype?: DataType): Variable<R> {\n    this.throwIfDisposed();\n    return trackerFn().makeVariable(this, trainable, name, dtype) as\n        Variable<R>;\n  }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n  value: (instance: Tensor) => {\n    // Implementation note: we should use properties of the object that will be\n    // defined before the constructor body has finished executing (methods).\n    // This is because when this code is transpiled by babel, babel will call\n    // classCallCheck before the constructor body is run.\n    // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n    return !!instance && instance.data != null && instance.dataSync != null &&\n        instance.throwIfDisposed != null;\n  }\n});\n\nexport interface NumericTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: NumericDataType;\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D];\n  data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]>;\n}\n\nexport interface StringTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: 'string';\n  dataSync<D extends DataType = 'string'>(): DataTypeMap[D];\n  data<D extends DataType = 'string'>(): Promise<DataTypeMap[D]>;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor<Rank.R0>;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor<Rank.R1>;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor<Rank.R2>;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor<Rank.R3>;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor<Rank.R4>;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor<Rank.R5>;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor<Rank.R6>;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable<R extends Rank = Rank> extends Tensor<R> {\n  name: string;\n\n  constructor(\n      initialValue: Tensor<R>, public trainable: boolean, name: string,\n      tensorId: number) {\n    super(\n        initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n    this.name = name;\n  }\n\n  /**\n   * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n   * the same shape and dtype as the old `tf.Tensor`.\n   *\n   * @param newValue New tensor to be assigned to this variable.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  assign(newValue: Tensor<R>): void {\n    if (newValue.dtype !== this.dtype) {\n      throw new Error(\n          `dtype of the new value (${newValue.dtype}) and ` +\n          `previous value (${this.dtype}) must match`);\n    }\n    if (!util.arraysEqual(newValue.shape, this.shape)) {\n      throw new Error(\n          `shape of the new value (${newValue.shape}) and ` +\n          `previous value (${this.shape}) must match`);\n    }\n    trackerFn().disposeTensor(this);\n    this.dataId = newValue.dataId;\n    trackerFn().incRef(this, null /* backend */);\n  }\n\n  dispose(): void {\n    trackerFn().disposeVariable(this);\n    this.isDisposedInternal = true;\n  }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n  value: (instance: Variable) => {\n    return instance instanceof Tensor && instance.assign != null &&\n        instance.assign instanceof Function;\n  }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n  R0: number[];\n  R1: [number];\n  R2: [number, number];\n  R3: [number, number, number];\n  R4: [number, number, number, number];\n  R5: [number, number, number, number, number];\n  R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n  R0: number;\n  R1: number[];\n  R2: number[][];\n  R3: number[][][];\n  R4: number[][][][];\n  R5: number[][][][][];\n  R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n  float32: Float32Array;\n  int32: Int32Array;\n  bool: Uint8Array;\n  complex64: Float32Array;\n  string: string[];\n}\n\nexport interface SingleValueMap {\n  bool: boolean;\n  int32: number;\n  float32: number;\n  complex64: number;\n  string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n  R0 = 'R0',\n  R1 = 'R1',\n  R2 = 'R2',\n  R3 = 'R3',\n  R4 = 'R4',\n  R5 = 'R5',\n  R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray<T> =\n    T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray<T extends any> {\n  [index: number]: T|RecursiveArray<T>;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'int32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'bool',\n  'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n  'float32' = 'float32',\n  'int32' = 'float32',\n  'bool' = 'float32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n  'float32' = 'complex64',\n  'int32' = 'complex64',\n  'bool' = 'complex64',\n  'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n  'float32': UpcastFloat32AndMap,\n  'int32': UpcastInt32AndMap,\n  'bool': UpcastBoolAndMap,\n  'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n  if (typeA === 'string' || typeB === 'string') {\n    if (typeA === 'string' && typeB === 'string') {\n      return 'string';\n    }\n    throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n  }\n  return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n  return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n    TypedArray|number|boolean|string|RecursiveArray<number|number[]|TypedArray>|\n    RecursiveArray<boolean>|RecursiveArray<string>|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n    string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n    boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n    boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n    TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n    string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n    TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n    string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n  width: number;\n  height: number;\n  data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch<T extends Tensor>(a: T, b: T): [T, T] {\n  if (a.dtype === b.dtype) {\n    return [a, b];\n  }\n  const dtype = upcastType(a.dtype, b.dtype);\n  return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n  assert(\n      a.dtype === b.dtype,\n      () => `The dtypes of the first(${a.dtype}) and` +\n          ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n  return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n *   `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n *   is safe to pass any object here, except that `Promise`s are not\n *   supported.\n * @returns An array of `Tensors` found within the passed object. If the\n *   argument is simply a `Tensor', a list containing that `Tensor` is\n *   returned. If the object is not a `Tensor` or does not\n *   contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n  const list: Tensor[] = [];\n  const seen = new Set<{}|void>();\n  walkTensorContainer(result, list, seen);\n  return list;\n}\n\nfunction walkTensorContainer(\n    container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n  if (container == null) {\n    return;\n  }\n  if (container instanceof Tensor) {\n    list.push(container);\n    return;\n  }\n  if (!isIterable(container)) {\n    return;\n  }\n  // Iteration over keys works also for arrays.\n  const iterable = container as TensorContainerArray;\n  for (const k in iterable) {\n    const val = iterable[k];\n    if (!seen.has(val)) {\n      seen.add(val);\n      walkTensorContainer(val, list, seen);\n    }\n  }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n  return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc<T> = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n *   value: Tensor,\n *   gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc<T extends Tensor> =\n    (...inputs: Array<Tensor|GradSaveFunc>) => {\n      value: T;\n      gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n    };\n\nexport type MemoryInfo = {\n  numTensors: number; numDataBuffers: number; numBytes: number;\n  unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n  name: string; bytesAdded: number; totalBytesSnapshot: number;\n  tensorsAdded: number;\n  totalTensorsSnapshot: number;\n  inputShapes: number[][];\n  outputShapes: number[][];\n  kernelTimeMs: number | {error: string} | Promise<number|{error: string}>;\n  extraInfo: string | Promise<string>;\n};\n\nexport type ProfileInfo = {\n  newBytes: number; newTensors: number; peakBytes: number;\n  kernels: KernelInfo[];\n  result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n  wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn<T extends TensorContainer> = () => T;\n\ninterface ScopeState {\n  track: Tensor[];\n  name: string;\n  id: number;\n}\n\nclass EngineState {\n  // Public since optimizers will use it.\n  registeredVariables: NamedVariableMap = {};\n\n  nextTapeNodeId = 0;\n  numBytes = 0;\n  numTensors = 0;\n  numStringTensors = 0;\n  numDataBuffers = 0;\n\n  activeTape: TapeNode[];\n  // Number of nested tf.grad() statements when computing higher-order\n  // gradients. E.g. `1` for first-order gradients and `2` for second-order\n  // gradients. Used to track if the tape should be removed after a backprop.\n  gradientDepth = 0;\n  // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n  // off the tape.\n  kernelDepth = 0;\n\n  // Keep Tensors that parallel the tapes.\n  activeScope: ScopeState;\n  scopeStack: ScopeState[] = [];\n  /**\n   * Keeps track of the number of data moves during a kernel execution. We\n   * maintain a stack since kernels can call other kernels, recursively.\n   */\n  numDataMovesStack: number[] = [];\n  nextScopeId = 0;\n\n  tensorInfo = new WeakMap<DataId, {\n    backend: KernelBackend,\n    bytes: number,\n    dtype: DataType,\n    shape: number[],\n    refCount: number\n  }>();\n\n  profiling = false;\n  activeProfile: ProfileInfo =\n      {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n  dispose() {\n    for (const variableName in this.registeredVariables) {\n      this.registeredVariables[variableName].dispose();\n    }\n  }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n  state: EngineState;\n  backendName: string;\n  registry: {[id: string]: KernelBackend} = {};\n  registryFactory: {\n    [id: string]: {\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority: number\n    }\n  } = {};\n\n  private profiler: Profiler;\n  private backendInstance: KernelBackend;\n  private pendingBackendInit: Promise<boolean>;\n  private pendingBackendInitId = 0;\n\n  constructor(public ENV: Environment) {\n    this.state = new EngineState();\n  }\n\n  async ready(): Promise<void> {\n    if (this.pendingBackendInit != null) {\n      return this.pendingBackendInit.then(() => {});\n    }\n    if (this.backendInstance != null) {\n      return;\n    }\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const success = await this.initializeBackend(backendName).success;\n      if (success) {\n        await this.setBackend(backendName);\n        return;\n      }\n    }\n\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  get backend(): KernelBackend {\n    if (this.pendingBackendInit != null) {\n      throw new Error(\n          `Backend '${this.backendName}' has not yet been initialized. Make ` +\n          `sure to await tf.ready() or await tf.setBackend() before calling ` +\n          `other methods`);\n    }\n    if (this.backendInstance == null) {\n      const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n      if (asyncInit) {\n        throw new Error(\n            `The highest priority backend '${name}' has not yet been ` +\n            `initialized. Make sure to await tf.ready() or ` +\n            `await tf.setBackend() before calling other methods`);\n      }\n      this.setBackend(name);\n    }\n    return this.backendInstance;\n  }\n\n  backendNames(): string[] {\n    return Object.keys(this.registryFactory);\n  }\n\n  findBackend(backendName: string): KernelBackend {\n    if (!(backendName in this.registry)) {\n      // If the backend hasn't been initialized but we have a registry entry for\n      // it, initialize it and return it.\n      if (backendName in this.registryFactory) {\n        const {asyncInit} = this.initializeBackend(backendName);\n        if (asyncInit) {\n          // Backend is not ready yet.\n          return null;\n        }\n      } else {\n        return null;\n      }\n    }\n    return this.registry[backendName];\n  }\n\n  findBackendFactory(backendName: string):\n      () => KernelBackend | Promise<KernelBackend> {\n    if (!(backendName in this.registryFactory)) {\n      return null;\n    }\n    return this.registryFactory[backendName].factory;\n  }\n\n  registerBackend(\n      backendName: string,\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority = 1): boolean {\n    if (backendName in this.registryFactory) {\n      console.warn(\n          `${backendName} backend was already registered. ` +\n          `Reusing existing backend factory.`);\n      return false;\n    }\n    this.registryFactory[backendName] = {factory, priority};\n    return true;\n  }\n\n  async setBackend(backendName: string): Promise<boolean> {\n    if (this.registryFactory[backendName] == null) {\n      throw new Error(`Backend name '${backendName}' not found in registry`);\n    }\n    this.backendName = backendName;\n    if (this.registry[backendName] == null) {\n      this.backendInstance = null;\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      const result = asyncInit ? await success : success;\n      if (!result) {\n        return false;\n      }\n    }\n    this.backendInstance = this.registry[backendName];\n    this.setupRegisteredKernels();\n    // Reset the profiler.\n    this.profiler = new Profiler(this.backendInstance);\n\n    return true;\n  }\n\n  private setupRegisteredKernels(): void {\n    const kernels = getKernelsForBackend(this.backendName);\n    kernels.forEach(kernel => {\n      if (kernel.setupFunc != null) {\n        kernel.setupFunc(this.backendInstance);\n      }\n    });\n  }\n\n  private disposeRegisteredKernels(backendName: string): void {\n    const kernels = getKernelsForBackend(backendName);\n    kernels.forEach(kernel => {\n      if (kernel.disposeFunc != null) {\n        kernel.disposeFunc(this.registry[backendName]);\n      }\n    });\n  }\n\n  /**\n   * Initializes a backend by looking up the backend name in the factory\n   * registry and calling the factory method. Returns a boolean representing\n   * whether the initialization of the backend suceeded. Throws an error if\n   * there is no backend in the factory registry.\n   */\n  private initializeBackend(backendName: string):\n      {success: boolean|Promise<boolean>, asyncInit: boolean} {\n    const registryFactoryEntry = this.registryFactory[backendName];\n    if (registryFactoryEntry == null) {\n      throw new Error(\n          `Cannot initialize backend ${backendName}, no registration found.`);\n    }\n\n    try {\n      const backend = registryFactoryEntry.factory();\n      /* Test if the factory returns a promise.\n      Done in a more liberal way than\n      previous 'Promise.resolve(backend)===backend'\n      as we needed to account for custom Promise\n      implementations (e.g. Angular) */\n      if (backend && !(backend instanceof KernelBackend)\n          && typeof backend.then === 'function') {\n        const promiseId = ++this.pendingBackendInitId;\n        const success =\n            backend\n                .then(backendInstance => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.registry[backendName] = backendInstance;\n                  this.pendingBackendInit = null;\n                  return true;\n                })\n                .catch(err => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.pendingBackendInit = null;\n                  console.warn(\n                      `Initialization of backend ${backendName} failed`);\n                  console.warn(err.stack || err.message);\n                  return false;\n                });\n        this.pendingBackendInit = success;\n        return {success, asyncInit: true};\n      } else {\n        this.registry[backendName] = backend as KernelBackend;\n        return {success: true, asyncInit: false};\n      }\n    } catch (err) {\n      console.warn(`Initialization of backend ${backendName} failed`);\n      console.warn(err.stack || err.message);\n      return {success: false, asyncInit: false};\n    }\n  }\n\n  removeBackend(backendName: string): void {\n    if (!(backendName in this.registryFactory)) {\n      throw new Error(`${backendName} backend not found in registry`);\n    }\n    if (this.backendName === backendName && this.pendingBackendInit != null) {\n      // There is a pending promise of the backend we want to remove. Make it\n      // obsolete.\n      this.pendingBackendInitId++;\n    }\n\n    if (backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n\n    delete this.registryFactory[backendName];\n\n    // Unset the backend if it is active.\n    if (this.backendName === backendName) {\n      this.pendingBackendInit = null;\n      this.backendName = null;\n      this.backendInstance = null;\n    }\n  }\n\n  private getSortedBackends(): string[] {\n    if (Object.keys(this.registryFactory).length === 0) {\n      throw new Error('No backend found in registry.');\n    }\n    return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n      // Highest priority comes first.\n      return this.registryFactory[b].priority -\n          this.registryFactory[a].priority;\n    });\n  }\n\n  private initializeBackendsAndReturnBest():\n      {name: string, asyncInit: boolean} {\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      if (asyncInit || success) {\n        return {name: backendName, asyncInit};\n      }\n    }\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  moveData(backend: KernelBackend, dataId: DataId) {\n    const info = this.state.tensorInfo.get(dataId);\n    const srcBackend = info.backend;\n    const values = this.readSync(dataId);\n    // Delete the tensor from the old backend and move it to the new\n    // backend.\n    srcBackend.disposeData(dataId);\n    info.backend = backend;\n    backend.move(dataId, values, info.shape, info.dtype);\n    if (this.shouldCheckForMemLeaks()) {\n      // Track the number of moves during a kernel execution to correctly\n      // detect memory leaks.\n      this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n    }\n  }\n\n  tidy<T extends TensorContainer>(nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>):\n      T {\n    let name: string = null;\n    if (fn == null) {\n      // Called with only 1 argument.\n      if (typeof nameOrFn !== 'function') {\n        throw new Error('Please provide a function to tidy()');\n      }\n      fn = nameOrFn;\n    } else {\n      // Called with 2 arguments.\n      if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n        throw new Error(\n            'When calling with two arguments, the first argument ' +\n            'to tidy() must be a string');\n      }\n      if (typeof fn !== 'function') {\n        throw new Error(\n            'When calling with two arguments, the 2nd argument ' +\n            'to tidy() must be a function');\n      }\n      name = nameOrFn as string;\n      // TODO(nsthorat,smilkov): Do operation logging and performance\n      // profiling.\n    }\n    let result: T;\n    return this.scopedRun(\n        () => this.startScope(name), () => this.endScope(result), () => {\n          result = fn();\n          if (result instanceof Promise) {\n            console.error('Cannot return a Promise inside of tidy.');\n          }\n          return result;\n        });\n  }\n\n  private scopedRun<T>(start: () => void, end: () => void, f: () => T): T {\n    start();\n    try {\n      const res = f();\n      end();\n      return res;\n    } catch (ex) {\n      end();\n      throw ex;\n    }\n  }\n\n  private static nextTensorId = 0;\n  private nextTensorId(): number {\n    return Engine.nextTensorId++;\n  }\n\n  private static nextVariableId = 0;\n  private nextVariableId(): number {\n    return Engine.nextVariableId++;\n  }\n\n  /**\n   * This method is called instead of the public-facing tensor.clone() when\n   * saving a tensor for backwards pass. It makes sure to add the clone\n   * operation to the tape regardless of being called inside a kernel\n   * execution.\n   *\n   * This method will go away once all kernels are modularized since we won't\n   * need to turn off the tape inside runKernel().\n   */\n  private clone(x: Tensor): Tensor {\n    const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n    const inputs = {x};\n    const grad = (dy: Tensor) => ({\n      x: () => {\n        const dtype = 'float32';\n        const gradInputs = {x: dy};\n        const attrs = {dtype};\n\n        return ENGINE.runKernelFunc(\n            backend => backend.cast(dy, dtype),\n            gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n            attrs as {} as NamedAttrMap);\n      }\n    });\n    const saved: Tensor[] = [];\n    this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n    return y;\n  }\n\n  /**\n   * Execute a kernel with the given name and return the output tensor.\n   *\n   * @param kernelName The name of the kernel to execute.\n   * @param inputs A map of input names to tensors.\n   * @param attrs A map of attribute names to their values. An attribute is a\n   *     primitive (non-tensor) input to the kernel.\n   * @param inputsToSave A list of tensors, inputs to save for the backprop\n   *     computation.\n   * @param outputsToSave A list of booleans, specifying which output to save\n   *     for the backprop computation. These are booleans since the output\n   * tensors are not visible to the user.\n   */\n  runKernel(\n      kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n      inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n    const forwardFunc: null = null;\n    const backwardsFunc: null = null;\n    // Call runKernel as a stop-gap until we modularize all kernels.\n    // Once we modularize all kernels, we will remove the existing\n    // `runKernelFunc`.\n    return this.runKernelFunc(\n        forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n        outputsToSave);\n  }\n\n  private shouldCheckForMemLeaks(): boolean {\n    return this.ENV.getBool('IS_TEST');\n  }\n\n  private checkKernelForMemLeak(\n      kernelName: string, numDataIdsBefore: number,\n      outInfos: TensorInfo[]): void {\n    const numDataIdsAfter = this.backend.numDataIds();\n\n    // Count the number of data ids associated with the result of the kernel.\n    let numOutputDataIds = 0;\n    outInfos.forEach(info => {\n      // Complex numbers allocate 3 data ids, one for 'real', one for\n      // 'imaginary', and one for the container that holds the former two.\n      numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n    });\n\n    // Account for the number of moves during kernel execution. A \"data move\"\n    // can happen in the middle of a kernel execution, placing a new (key,value)\n    // pair in the data storage. Since data moves have net zero effect (we\n    // always remove the data from the old backend), we have to cancel them out\n    // when detecting memory leaks.\n    const numMoves =\n        this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n    const dataIdsLeaked =\n        numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n    if (dataIdsLeaked > 0) {\n      throw new Error(\n          `Backend '${this.backendName}' has an internal memory leak ` +\n          `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n    }\n  }\n\n  /**\n   * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n   *     only for kernels that are not yet fully modularized.\n   */\n  runKernelFunc<T extends Tensor|Tensor[], I extends NamedTensorMap>(\n      forwardFunc: ForwardFunc<T>, inputs: I,\n      backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n      kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n      outputsToSave?: boolean[]): T {\n    let outputs: Tensor[];\n    let saved: Tensor[] = [];\n    const isTapeOn = this.isTapeOn();\n    if (kernelName == null) {\n      kernelName =\n          this.state.activeScope != null ? this.state.activeScope.name : '';\n    }\n\n    const startingBytecount = this.state.numBytes;\n    const startingNumTensors = this.state.numTensors;\n\n    if (this.shouldCheckForMemLeaks()) {\n      this.state.numDataMovesStack.push(0);\n    }\n\n    let kernelFunc: () => Tensor[];\n    const kernel = getKernel(kernelName, this.backendName);\n    let out: TensorInfo|TensorInfo[];\n    if (kernel != null) {\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n        const outInfos = Array.isArray(out) ? out : [out];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n        }\n        const outTensors = outInfos.map(\n            ({dataId, shape, dtype}) =>\n                this.makeTensorFromDataId(dataId, shape, dtype));\n\n        // Save the inputs and outputs.\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (isTapeOn) {\n          let tensorsToSave =\n              this.getTensorsForGradient(kernelName, inputs, outTensors);\n          if (tensorsToSave == null) {\n            // Fallback for ops that call runKernelFunc and pass in\n            // inputsToSave and outputsToSave. Currently this is the set of ops\n            // with kernel support in the WASM backend. Once those ops and\n            // respective gradients are modularised we can remove this path.\n            if (outputsToSave == null) {\n              outputsToSave = [];\n            }\n            const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n            tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n          }\n          saved = this.saveTensorsForBackwardMode(tensorsToSave);\n        }\n        return outTensors;\n      };\n    } else {\n      const saveFunc: GradSaveFunc = (tensors) => {\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (!isTapeOn) {\n          return;\n        }\n        saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n      };\n\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n        const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n        }\n        return outs;\n      };\n    }\n\n    // Stop recording to a tape when running a kernel.\n    let kernelProfile: KernelProfile;\n    this.scopedRun(\n        () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n          if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n            outputs = kernelFunc();\n          } else {\n            kernelProfile = this.profiler.profileKernel(\n                kernelName, inputs, () => kernelFunc());\n            if (this.ENV.getBool('DEBUG')) {\n              this.profiler.logKernelProfile(kernelProfile);\n            }\n            outputs = kernelProfile.outputs;\n          }\n        });\n\n    if (isTapeOn) {\n      this.addTapeNode(\n          kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n    }\n\n    if (this.state.profiling) {\n      this.state.activeProfile.kernels.push({\n        name: kernelName,\n        bytesAdded: this.state.numBytes - startingBytecount,\n        totalBytesSnapshot: this.state.numBytes,\n        tensorsAdded: this.state.numTensors - startingNumTensors,\n        totalTensorsSnapshot: this.state.numTensors,\n        inputShapes: Object.keys(inputs).map(\n            key => inputs[key] != null ? inputs[key].shape : null),\n        outputShapes: outputs.map(item => item.shape),\n        kernelTimeMs: kernelProfile.timeMs,\n        extraInfo: kernelProfile.extraInfo\n      });\n    }\n    return (Array.isArray(out) ? outputs : outputs[0]) as T;\n  }\n\n  /**\n   * Saves tensors used in forward mode for use in backward mode.\n   *\n   * @param tensors the list of tensors to save.\n   */\n  private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n    const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n    return saved;\n  }\n\n  /**\n   * Returns a list of tensors to save for a given gradient calculation.\n   *\n   * Returns undefined if their is no registered gradient for this kernel in the\n   * gradient registry.\n   *\n   * @param kernelName name of kernel to look up gradient for.\n   * @param inputs a map of input tensors.\n   * @param outputs an array of output tensors from forward mode of kernel.\n   */\n  private getTensorsForGradient(\n      kernelName: string, inputs: NamedTensorMap,\n      outputs: Tensor[]): Tensor[]|null {\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      const inputsToSave: string[] = gradConfig.inputsToSave || [];\n      const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n      // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n      // specified in inputsToSave will be saved.\n      let inputTensorsToSave: Tensor[];\n      if (gradConfig.saveAllInputs) {\n        util.assert(\n            Array.isArray(inputs),\n            () => 'saveAllInputs is true, expected inputs to be an array.');\n\n        inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n      } else {\n        inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n      }\n\n      const outputTensorsToSave: Tensor[] =\n          outputs.filter((_, i) => outputsToSave[i]);\n\n      return inputTensorsToSave.concat(outputTensorsToSave);\n    }\n    // TODO(yassogba) throw exception here once all runkernelFunc calls with\n    // inputsToSave/outputsToSave are removed\n    return null;\n  }\n\n  /**\n   * Internal method used by public APIs for tensor creation. Makes a new\n   * tensor with the provided shape, dtype and values. It always\n   * creates a new data id and writes the values to the underlying backend.\n   */\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    if (values == null) {\n      throw new Error('Values passed to engine.makeTensor() are null');\n    }\n    dtype = dtype || 'float32';\n    backend = backend || this.backend;\n    let backendVals = values as BackendValues;\n    if (dtype === 'string' && util.isString(values[0])) {\n      backendVals = (values as string[]).map(d => util.encodeString(d));\n    }\n    const dataId = backend.write(backendVals, shape, dtype);\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n\n    // Count bytes for string tensors.\n    if (dtype === 'string') {\n      const info = this.state.tensorInfo.get(dataId);\n      const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n      this.state.numBytes += newBytes - info.bytes;\n      info.bytes = newBytes;\n    }\n    return t;\n  }\n\n  /**\n   * Internal method used by backends. Makes a new tensor\n   * that is a wrapper around an existing data id. It doesn't create\n   * a new data id, only increments the ref count used in memory tracking.\n   */\n  makeTensorFromDataId(\n      dataId: DataId, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    dtype = dtype || 'float32';\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n    return t;\n  }\n\n  makeVariable(\n      initialValue: Tensor, trainable = true, name?: string,\n      dtype?: DataType): Variable {\n    name = name || this.nextVariableId().toString();\n    if (dtype != null && dtype !== initialValue.dtype) {\n      initialValue = initialValue.cast(dtype);\n    }\n    const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n    if (this.state.registeredVariables[v.name] != null) {\n      throw new Error(`Variable with name ${v.name} was already registered`);\n    }\n    this.state.registeredVariables[v.name] = v;\n    this.incRef(v, this.backend);\n    return v;\n  }\n\n  incRef(a: Tensor, backend: KernelBackend): void {\n    const refCount = this.state.tensorInfo.has(a.dataId) ?\n        this.state.tensorInfo.get(a.dataId).refCount :\n        0;\n    this.state.numTensors++;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors++;\n    }\n    if (refCount === 0) {\n      this.state.numDataBuffers++;\n\n      // Bytes for complex numbers are counted by their components. Bytes for\n      // string tensors are counted when writing values.\n      let bytes = 0;\n      if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n        bytes = a.size * util.bytesPerElement(a.dtype);\n      }\n      this.state.tensorInfo.set(a.dataId, {\n        backend: backend || this.backend,\n        dtype: a.dtype,\n        shape: a.shape,\n        bytes,\n        refCount: 0\n      });\n      this.state.numBytes += bytes;\n    }\n\n    this.state.tensorInfo.get(a.dataId).refCount++;\n\n    if (!(a instanceof Variable)) {\n      this.track(a);\n    }\n  }\n\n  disposeTensor(a: Tensor): void {\n    if (!this.state.tensorInfo.has(a.dataId)) {\n      return;\n    }\n\n    this.state.numTensors--;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors--;\n    }\n    const info = this.state.tensorInfo.get(a.dataId);\n    const refCount = info.refCount;\n\n    if (refCount <= 1) {\n      // Don't count bytes for complex numbers as they are counted by their\n      // components.\n      if (a.dtype !== 'complex64') {\n        this.state.numBytes -= info.bytes;\n      }\n      this.state.numDataBuffers--;\n\n      info.backend.disposeData(a.dataId);\n      this.state.tensorInfo.delete(a.dataId);\n    } else {\n      this.state.tensorInfo.get(a.dataId).refCount--;\n    }\n    // TODO(nsthorat): Construct an error and save the stack trace for\n    // debugging when in debug mode. Creating a stack trace is too expensive\n    // to do unconditionally.\n  }\n\n  disposeVariables(): void {\n    for (const varName in this.state.registeredVariables) {\n      const v = this.state.registeredVariables[varName];\n      this.disposeVariable(v);\n    }\n  }\n\n  disposeVariable(v: Variable): void {\n    this.disposeTensor(v);\n    if (this.state.registeredVariables[v.name] != null) {\n      delete this.state.registeredVariables[v.name];\n    }\n  }\n\n  memory(): MemoryInfo {\n    const info = this.backend.memory() as MemoryInfo;\n    info.numTensors = this.state.numTensors;\n    info.numDataBuffers = this.state.numDataBuffers;\n    info.numBytes = this.state.numBytes;\n    if (this.state.numStringTensors > 0) {\n      info.unreliable = true;\n      if (info.reasons == null) {\n        info.reasons = [];\n      }\n      info.reasons.push(\n          'Memory usage by string tensors is approximate ' +\n          '(2 bytes per character)');\n    }\n    return info;\n  }\n\n  async profile(query: () => (TensorContainer | Promise<TensorContainer>)):\n      Promise<ProfileInfo> {\n    this.state.profiling = true;\n\n    const startBytes = this.state.numBytes;\n    const startNumTensors = this.state.numTensors;\n\n    this.state.activeProfile.kernels = [];\n    this.state.activeProfile.result = await query();\n\n    this.state.profiling = false;\n\n    this.state.activeProfile.peakBytes = Math.max(\n        ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n    this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n    this.state.activeProfile.newTensors =\n        this.state.numTensors - startNumTensors;\n    for (const kernel of this.state.activeProfile.kernels) {\n      kernel.kernelTimeMs = await kernel.kernelTimeMs;\n      kernel.extraInfo = await kernel.extraInfo;\n    }\n    return this.state.activeProfile;\n  }\n\n  isTapeOn(): boolean {\n    return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n  }\n\n  private addTapeNode(\n      kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n      gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n    const tapeNode: TapeNode =\n        {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      gradientsFunc = gradConfig.gradFunc;\n    }\n    if (gradientsFunc != null) {\n      tapeNode.gradient = (dys: Tensor[]) => {\n        // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n        // the backprop graph to the user as null instead of zeros\n        dys = dys.map((dy, i) => {\n          if (dy == null) {\n            const output = outputs[i];\n            const vals = util.makeZerosTypedArray(output.size, output.dtype);\n            return this.makeTensor(vals, output.shape, output.dtype);\n          }\n          return dy;\n        });\n        // Grad functions of ops with single outputs expect a dy, while ops\n        // with multiple outputs expect dys (array of dy).\n        return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n      };\n    }\n    this.state.activeTape.push(tapeNode);\n  }\n\n  keep<T extends Tensor>(result: T): T {\n    result.kept = true;\n    return result;\n  }\n\n  private startTape() {\n    if (this.state.gradientDepth === 0) {\n      this.state.activeTape = [];\n    }\n    this.state.gradientDepth++;\n  }\n\n  private endTape() {\n    this.state.gradientDepth--;\n  }\n\n  /**\n   * Start a scope. Use this with endScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  startScope(name?: string) {\n    const scopeInfo: ScopeState = {\n      track: [],\n      name: 'unnamed scope',\n      id: this.state.nextScopeId++\n    };\n    if (name) {\n      scopeInfo.name = name;\n    }\n    this.state.scopeStack.push(scopeInfo);\n    this.state.activeScope = scopeInfo;\n  }\n\n  /**\n   * End a scope. Use this with startScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  endScope(result?: TensorContainer) {\n    const tensorsToTrackInParent = getTensorsInContainer(result);\n    const tensorsToTrackInParentSet =\n        new Set(tensorsToTrackInParent.map(t => t.id));\n\n    // Dispose the arrays tracked in this scope.\n    for (let i = 0; i < this.state.activeScope.track.length; i++) {\n      const tensor = this.state.activeScope.track[i];\n      if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n        tensor.dispose();\n      }\n    }\n\n    const oldScope = this.state.scopeStack.pop();\n    this.state.activeScope = this.state.scopeStack.length === 0 ?\n        null :\n        this.state.scopeStack[this.state.scopeStack.length - 1];\n\n    // Track the current result in the parent scope.\n    tensorsToTrackInParent.forEach(tensor => {\n      // Only track the tensor if was allocated in the inner scope and is not\n      // globally kept.\n      if (!tensor.kept && tensor.scopeId === oldScope.id) {\n        this.track(tensor);\n      }\n    });\n  }\n\n  /**\n   * Returns gradients of `f` with respect to each of the `xs`. The gradients\n   * returned are of the same length as `xs`, but some might be null if `f`\n   * was not a function of that `x`. It also takes optional dy to multiply the\n   * gradient, which defaults to `1`.\n   */\n  gradients<T extends Tensor>(\n      f: () => T, xs: Tensor[], dy?: T,\n      allowNoGradients = false): {value: T, grads: Tensor[]} {\n    util.assert(\n        xs.length > 0, () => 'gradients() received an empty list of xs.');\n    if (dy != null && dy.dtype !== 'float32') {\n      throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n    }\n\n    const y = this.scopedRun(\n        () => this.startTape(), () => this.endTape(),\n        () => this.tidy('forward', f));\n\n    util.assert(\n        y instanceof Tensor,\n        () => 'The result y returned by f() must be a tensor.');\n    // Filter out the nodes that don't connect x => y.\n    const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n    if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n      throw new Error(\n          'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n          'that the f you passed encloses all operations that lead from x ' +\n          'to y.');\n    }\n\n    return this.tidy('backward', () => {\n      const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n      accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n      // Backprop gradients through the filtered nodes.\n      backpropagateGradients(\n          accumulatedGradientMap, filteredTape,\n          // Pass the tidy function to avoid circular dep with `tape.ts`.\n          f => this.tidy(f as ScopeFn<Tensor>),\n          // Pass an add function to avoide a circular dep with `tape.ts`.\n          add);\n      const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n      if (this.state.gradientDepth === 0) {\n        // This means that we are not computing higher-order gradients\n        // and can clean up the tape.\n        this.state.activeTape.forEach(node => {\n          for (const tensor of node.saved) {\n            tensor.dispose();\n          }\n        });\n        this.state.activeTape = null;\n      }\n      return {value: y, grads};\n    });\n  }\n\n  customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n      (...args: Array<Tensor|GradSaveFunc>) => T {\n    util.assert(\n        util.isFunction(f),\n        () => 'The f passed in customGrad(f) must be a function.');\n    return (...inputs: Tensor[]): T => {\n      util.assert(\n          inputs.every(t => t instanceof Tensor),\n          () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n              'tensors');\n\n      let res: {\n        value: T,\n        gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n      };\n      const inputMap: NamedTensorMap = {};\n      inputs.forEach((input, i) => {\n        inputMap[i] = input;\n      });\n      return this.runKernelFunc(\n          (_, save) => {\n            res = f(...[...inputs, save]);\n            util.assert(\n                res.value instanceof Tensor,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.value` is a tensor');\n            util.assert(\n                util.isFunction(res.gradFunc),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function.');\n            return res.value;\n          },\n          inputMap,\n          (dy: T, saved: Tensor[]) => {\n            const gradRes = res.gradFunc(dy, saved);\n            const grads: Tensor[] =\n                Array.isArray(gradRes) ? gradRes : [gradRes];\n            util.assert(\n                grads.length === inputs.length,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'the same number of tensors as inputs passed to f(...).');\n            util.assert(\n                grads.every(t => t instanceof Tensor),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'a list of only tensors.');\n            const gradMap: {[key: string]: () => Tensor} = {};\n            grads.forEach((grad, i) => {\n              gradMap[i] = () => grad;\n            });\n            return gradMap;\n          });\n    };\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.readSync(dataId);\n  }\n  read(dataId: DataId): Promise<BackendValues> {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.read(dataId);\n  }\n\n  async time(query: () => void): Promise<TimingInfo> {\n    const start = now();\n    const timingInfo = await this.backend.time(query) as TimingInfo;\n    timingInfo.wallMs = now() - start;\n    return timingInfo;\n  }\n\n  /**\n   * Tracks a Tensor in the current scope to be automatically cleaned up\n   * when the current scope ends, and returns the value.\n   *\n   * @param result The Tensor to track in the current scope.\n   */\n  private track<T extends Tensor>(result: T): T {\n    if (this.state.activeScope != null) {\n      result.scopeId = this.state.activeScope.id;\n      this.state.activeScope.track.push(result);\n    }\n\n    return result;\n  }\n\n  get registeredVariables(): NamedVariableMap {\n    return this.state.registeredVariables;\n  }\n\n  /**\n   * Resets the engine state. Removes all backends but does not remove\n   * registered backend factories.\n   */\n  reset(): void {\n    // Make any pending promise obsolete.\n    this.pendingBackendInitId++;\n\n    this.state.dispose();\n    this.ENV.reset();\n    this.state = new EngineState();\n\n    for (const backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n    this.backendName = null;\n    this.backendInstance = null;\n    this.pendingBackendInit = null;\n  }\n}\n\nfunction ones(shape: number[]): Tensor {\n  const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n  return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n  const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n  if (ns._tfengine == null) {\n    const environment = new Environment(ns);\n    ns._tfengine = new Engine(environment);\n  }\n  setEnvironmentGlobal(ns._tfengine.ENV);\n\n  // Tell the current tensor interface that the global engine is responsible\n  // for tracking.\n  setTensorTracker(() => ns._tfengine);\n  return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n  // We duplicate Add here to avoid a circular dependency with add.ts.\n  const inputs = {a, b};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.add(a, b);\n    save([a, b]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n  return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n  if (_isNavigatorDefined()) {\n    // tslint:disable-next-line:no-any\n    const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n    // tslint:disable-next-line:max-line-length\n    return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n               .test(a) ||\n        // tslint:disable-next-line:max-line-length\n        /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n            .test(a.substr(0, 4));\n  }\n  return false;\n}\n\nexport function isBrowser(): boolean {\n  return (typeof window !== 'undefined' && window.document != null) ||\n      //@ts-ignore\n      (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n  if (debugValue) {\n    console.warn(\n        'Debugging mode is ON. The output of every math call will ' +\n        'be downloaded to CPU and checked for NaNs. ' +\n        'This significantly impacts performance.');\n  }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n    'IS_NODE',\n    () => (typeof process !== 'undefined') &&\n        (typeof process.versions !== 'undefined') &&\n        (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n    'IS_CHROME',\n    () => typeof navigator !== 'undefined' && navigator != null &&\n        navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n        /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n    'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n  let firstElem: typeof val = val;\n\n  if (isTypedArray(val)) {\n    return dtype === 'string' ? [] : [val.length];\n  }\n  if (!Array.isArray(val)) {\n    return [];  // Scalar.\n  }\n  const shape: number[] = [];\n\n  while (Array.isArray(firstElem) ||\n         isTypedArray(firstElem) && dtype !== 'string') {\n    shape.push(firstElem.length);\n    firstElem = firstElem[0];\n  }\n  if (Array.isArray(val) &&\n      env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n    deepAssertShapeConsistency(val, shape, []);\n  }\n\n  return shape;\n}\n\nfunction deepAssertShapeConsistency(\n    val: TensorLike, shape: number[], indices: number[]) {\n  indices = indices || [];\n  if (!(Array.isArray(val)) && !isTypedArray(val)) {\n    assert(\n        shape.length === 0,\n        () => `Element arr[${indices.join('][')}] is a primitive, ` +\n            `but should be an array/TypedArray of ${shape[0]} elements`);\n    return;\n  }\n  assert(\n      shape.length > 0,\n      () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n          `but is an array of ${val.length} elements`);\n  assert(\n      val.length === shape[0],\n      () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n          `elements, but has ${val.length} elements`);\n  const subShape = shape.slice(1);\n  for (let i = 0; i < val.length; ++i) {\n    deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n  }\n}\n\nfunction assertDtype(\n    expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n    functionName: string) {\n  if (expectedDtype == null) {\n    return;\n  }\n  if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n      expectedDtype === 'numeric' && actualDType === 'string') {\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must ` +\n        `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n  }\n}\n\nexport function convertToTensor<T extends Tensor>(\n    x: T|TensorLike, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T {\n  if (x instanceof Tensor) {\n    assertDtype(parseAsDtype, x.dtype, argName, functionName);\n    return x;\n  }\n  let inferredDtype = inferDtype(x);\n  // If the user expects a bool/int/float, use that info to update the\n  // inferredDtype when it is not a string.\n  if (inferredDtype !== 'string' &&\n      ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n    inferredDtype = parseAsDtype as DataType;\n  }\n  assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n  if ((x == null) ||\n      (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n       typeof x !== 'boolean' && typeof x !== 'string')) {\n    const type = x == null ? 'null' : (x as {}).constructor.name;\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must be a ` +\n        `Tensor or TensorLike, but got '${type}'`);\n  }\n  const inferredShape = inferShape(x, inferredDtype);\n  if (!isTypedArray(x) && !Array.isArray(x)) {\n    x = [x] as number[];\n  }\n  const skipTypedArray = true;\n  const values = inferredDtype !== 'string' ?\n      toTypedArray(x, inferredDtype as DataType) :\n      flatten(x as string[], [], skipTypedArray) as string[];\n  return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray<T extends Tensor>(\n    arg: Array<T|TensorLike>, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n  if (!Array.isArray(arg)) {\n    throw new Error(\n        `Argument ${argName} passed to ${functionName} must be a ` +\n        '`Tensor[]` or `TensorLike[]`');\n  }\n  const tensors = arg as T[];\n  return tensors.map(\n      (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n      parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op<T extends Function>(f: {[name: string]: T}): T {\n  const keys = Object.keys(f);\n  if (keys.length !== 1) {\n    throw new Error(\n        `Please provide an object with a single key ` +\n        `(operation name) mapping to a function. Got an object with ` +\n        `${keys.length} keys.`);\n  }\n\n  let opName = keys[0];\n  const fn = f[opName];\n\n  // Strip the underscore from the end of the function name.\n  if (opName.endsWith('_')) {\n    opName = opName.substring(0, opName.length - 1);\n  }\n\n  // add an __op suffix to distinguish ops from kernels in tf.profile\n  opName = opName + OP_SCOPE_SUFFIX;\n\n  // tslint:disable-next-line:no-any\n  const f2 = (...args: any[]) => {\n    ENGINE.startScope(opName);\n    try {\n      const result = fn(...args);\n      if (result instanceof Promise) {\n        console.error('Cannot return a Promise inside of tidy.');\n      }\n      ENGINE.endScope(result);\n      return result;\n    } catch (ex) {\n      ENGINE.endScope(null);\n      throw ex;\n    }\n  };\n  Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n  // tslint:disable-next-line:no-any\n  return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_<T extends Tensor>(real: T|TensorLike, imag: T|TensorLike): T {\n  const $real = convertToTensor(real, 'real', 'complex');\n  const $imag = convertToTensor(imag, 'imag', 'complex');\n  util.assertShapesMatch(\n      $real.shape, $imag.shape,\n      `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n          `must match in call to tf.complex().`);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.complex($real, $imag);\n  };\n  const inputs: ComplexInputs = {real: $real, imag: $imag};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n    values: TensorLike, shape: number[], inferredShape: number[],\n    dtype?: DataType): Tensor {\n  if (dtype == null) {\n    dtype = inferDtype(values);\n  }\n  if (dtype === 'complex64') {\n    throw new Error(\n        `Cannot construct a complex64 tensor directly. ` +\n        `Please use tf.complex(real, imag).`);\n  }\n  if (!isTypedArray(values) && !Array.isArray(values) &&\n      typeof values !== 'number' && typeof values !== 'boolean' &&\n      typeof values !== 'string') {\n    throw new Error(\n        'values passed to tensor(values) must be a number/boolean/string or ' +\n        'an array of numbers/booleans/strings, or a TypedArray');\n  }\n  if (shape != null) {\n    assertNonNegativeIntegerDimensions(shape);\n\n    const providedSize = sizeFromShape(shape);\n    const inferredSize = sizeFromShape(inferredShape);\n    assert(\n        providedSize === inferredSize,\n        () =>\n            `Based on the provided shape, [${shape}], the tensor should have ` +\n            `${providedSize} values but has ${inferredSize}`);\n\n    for (let i = 0; i < inferredShape.length; ++i) {\n      const inferred = inferredShape[i];\n      const flatDimsDontMatch = i === inferredShape.length - 1 ?\n          inferred !== sizeFromShape(shape.slice(i)) :\n          true;\n      assert(\n          inferredShape[i] === shape[i] || !flatDimsDontMatch,\n          () => `Error creating a new Tensor. Inferred shape ` +\n              `(${inferredShape}) does not match the provided ` +\n              `shape (${shape}). `);\n    }\n  }\n\n  if (!isTypedArray(values) && !Array.isArray(values)) {\n    values = [values] as number[];\n  }\n\n  shape = shape || inferredShape;\n  values = dtype !== 'string' ?\n      toTypedArray(values, dtype) :\n      flatten(values as string[], [], true) as string[];\n  return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`. If the values are strings,\n *     they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor<R extends Rank>(\n    values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor<R> {\n  const inferredShape = inferShape(values, dtype);\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n  'float32': 4,\n  'float16': 2,\n  'int32': 4,\n  'uint16': 2,\n  'uint8': 1,\n  'bool': 1,\n  'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n  /**\n   * An ordered list of paths.\n   *\n   * Paths are intentionally abstract in order to be general. For example, they\n   * can be relative URL paths or relative paths on the file system.\n   */\n  paths: string[];\n\n  /**\n   * Specifications of the weights stored in the paths.\n   */\n  weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n  /**\n   * Name of the weight, e.g., 'Dense_1/bias'\n   */\n  name: string;\n\n  /**\n   * Shape of the weight.\n   */\n  shape: number[];\n\n  /**\n   * Data type of the weight.\n   */\n  dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n  /**\n   * Type of the weight.\n   *\n   * Optional.\n   *\n   * The value 'optimizer' indicates the weight belongs to an optimizer\n   * (i.e., used only during model training and not during inference).\n   */\n  group?: WeightGroup;\n\n  /**\n   * Information for dequantization of the weight.\n   */\n  quantization?: {\n    scale?: number,  // The scaling constant to multiply by.\n    min?: number,    // The (possibly nudged) minimum weight to add.\n       dtype: 'uint16'|'uint8'|'float16'  // The dtype of the quantized weights.\n  };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n  /**\n   * Whether to save only the trainable weights of the model, ignoring the\n   * non-trainable ones.\n   */\n  trainableOnly?: boolean;\n\n  /**\n   * Whether the optimizer will be saved (if exists).\n   *\n   * Default: `false`.\n   */\n  includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n  /**\n   * Information about the model artifacts saved.\n   */\n  modelArtifactsInfo: ModelArtifactsInfo;\n\n  /**\n   * HTTP responses from the server that handled the model-saving request (if\n   * any). This is applicable only to server-based saving routes.\n   */\n  responses?: Response[];\n\n  /**\n   * Error messages and related data (if any).\n   */\n  errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n  /**\n   * Timestamp for when the model is saved.\n   */\n  dateSaved: Date;\n\n  /**\n   * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n   * come in a JSON format and none of our IOHandlers support a non json\n   * format. We could conder replacing this with 'Binary' if we want to\n   * allow future handlers to save to non json formats (though they will\n   * probably want more information than 'Binary').\n   * Type of the model topology\n   *\n   * Type of the model topology\n   *\n   * Possible values:\n   *   - JSON: JSON config (human-readable, e.g., Keras JSON).\n   *   - GraphDef: TensorFlow\n   *     [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n   *     protocol buffer (binary).\n   */\n  modelTopologyType: 'JSON'|'GraphDef';\n\n  /**\n   * Size of model topology (Keras JSON or GraphDef), in bytes.\n   */\n  modelTopologyBytes?: number;\n\n  /**\n   * Size of weight specification or manifest, in bytes.\n   */\n  weightSpecsBytes?: number;\n\n  /**\n   * Size of weight value data, in bytes.\n   */\n  weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  // See\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n  /** Optimizer used for the model training. */\n  optimizer_config: {};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Loss function(s) for the model's output(s). */\n  loss: string|string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Metric function(s) for the model's output(s). */\n  metrics?: string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  weighted_metrics?: string[];\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  sample_weight_mode?: string;\n\n  loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers.  For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology?: {}|ArrayBuffer;\n\n  /**\n   * Serialized configuration for the model's training.\n   */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weight specifications.\n   *\n   * This corresponds to the weightsData below.\n   */\n  weightSpecs?: WeightsManifestEntry[];\n\n  /**\n   * Binary buffer for all weight values concatenated in the order specified\n   * by `weightSpecs`.\n   */\n  weightData?: ArrayBuffer;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology: {};\n\n  /** Model training configuration. */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weights manifest.\n   *\n   * The weights manifest consists of an ordered list of weight-manifest\n   * groups. Each weight-manifest group consists of a number of weight values\n   * stored in a number of paths. See the documentation of\n   * `WeightsManifestConfig` for more details.\n   */\n  weightsManifest: WeightsManifestConfig;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise<ModelArtifacts>;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n    Promise<SaveResult>;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n  save?: SaveHandler;\n  load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n  /**\n   * List all models in the model store.\n   *\n   * @returns A dictionary mapping paths of existing models to their\n   *   model artifacts info. Model artifacts info include type of the model's\n   *   topology, byte sizes of the topology, weights, etc.\n   */\n  listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n  /**\n   * Remove a model specified by `path`.\n   *\n   * @param path\n   * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n   *   is successful).\n   * @throws Error if deletion fails, e.g., if no model exists at `path`.\n   */\n  removeModel(path: string): Promise<ModelArtifactsInfo>;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n  /**\n   * RequestInit (options) for HTTP requests.\n   *\n   * For detailed information on the supported fields, see\n   * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n   *     https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n   */\n  requestInit?: RequestInit;\n\n  /**\n   * Progress callback.\n   */\n  onProgress?: OnProgressCallback;\n\n  /**\n   * A function used to override the `window.fetch` function.\n   */\n  fetchFunc?: Function;\n\n  /**\n   * Strict loading model: whether extraneous weights or missing\n   * weights should trigger an `Error`.\n   *\n   * If `true`, require that the provided weights exactly match those\n   * required by the layers. `false` means that both extra weights\n   * and missing weights will be silently ignored.\n   *\n   * Default: `true`.\n   */\n  strict?: boolean;\n\n  /**\n   * Path prefix for weight files, by default this is calculated from the\n   * path of the model JSON file.\n   *\n   * For instance, if the path to the model JSON file is\n   * `http://localhost/foo/model.json`, then the default path prefix will be\n   * `http://localhost/foo/`. If a weight file has the path value\n   * `group1-shard1of2` in the weight manifest, then the weight file will be\n   * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n   * if you provide a `weightPathPrefix` value of\n   * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n   * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n   */\n  weightPathPrefix?: string;\n\n  /**\n   * Whether the module or model is to be loaded from TF Hub.\n   *\n   * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n   * standard model file name and the query parameters.\n   *\n   * Default: `false`.\n   */\n  fromTFHub?: boolean;\n\n  /**\n   * An async function to convert weight file name to URL. The weight file\n   * names are stored in model.json's weightsManifest.paths field. By default we\n   * consider weight files are colocated with the model.json file. For example:\n   *     model.json URL: https://www.google.com/models/1/model.json\n   *     group1-shard1of1.bin url:\n   *        https://www.google.com/models/1/group1-shard1of1.bin\n   *\n   * With this func you can convert the weight file name to any URL.\n   */\n  weightUrlConverter?: (weightFileName: string) => Promise<string>;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n  /**\n   * Is this request for a binary file (as opposed to a json file)\n   */\n  isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n *     concatenated.\n *   - An `Array` of `WeightManifestEntry`s, carrying information including\n *     tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n    tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n    Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n  // TODO(adarob, cais): Support quantization.\n  const specs: WeightsManifestEntry[] = [];\n  const dataPromises: Array<Promise<TypedArray>> = [];\n\n  const names: string[] = Array.isArray(tensors) ?\n      tensors.map(tensor => tensor.name) :\n      Object.keys(tensors);\n\n  for (let i = 0; i < names.length; ++i) {\n    const name = names[i];\n    const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n    if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n        t.dtype !== 'string' && t.dtype !== 'complex64') {\n      throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n    }\n    const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n    if (t.dtype === 'string') {\n      const utf8bytes = new Promise<TypedArray>(async resolve => {\n        const vals = await t.bytes() as Uint8Array[];\n        const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n            NUM_BYTES_STRING_LENGTH * vals.length;\n        const bytes = new Uint8Array(totalNumBytes);\n        let offset = 0;\n        for (let i = 0; i < vals.length; i++) {\n          const val = vals[i];\n          const bytesOfLength =\n              new Uint8Array(new Uint32Array([val.length]).buffer);\n          bytes.set(bytesOfLength, offset);\n          offset += NUM_BYTES_STRING_LENGTH;\n          bytes.set(val, offset);\n          offset += val.length;\n        }\n        resolve(bytes);\n      });\n      dataPromises.push(utf8bytes);\n    } else {\n      dataPromises.push(t.data());\n    }\n    if (group != null) {\n      spec.group = group;\n    }\n    specs.push(spec);\n  }\n\n  const tensorValues = await Promise.all(dataPromises);\n  return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n *   concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n *   whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n *   to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n    buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n  // TODO(adarob, cais): Support quantization.\n  const out: NamedTensorMap = {};\n  let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n  let offset = 0;\n  for (const spec of specs) {\n    const name = spec.name;\n    const dtype = spec.dtype;\n    const shape = spec.shape;\n    const size = sizeFromShape(shape);\n    let values: TypedArray|string[]|Uint8Array[];\n\n    if ('quantization' in spec) {\n      const quantization = spec.quantization;\n      if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n        if (!('min' in quantization && 'scale' in quantization)) {\n          throw new Error(\n              `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n              `doesn't have corresponding metadata min and scale.`);\n        }\n      } else if (quantization.dtype === 'float16') {\n        if (dtype !== 'float32') {\n          throw new Error(\n              `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n              `which only supports weights of type float32 not ${dtype}.`);\n        }\n      } else {\n        throw new Error(\n            `Weight ${spec.name} has unknown ` +\n            `quantization dtype ${quantization.dtype}. ` +\n            `Supported quantization dtypes are: ` +\n            `'uint8', 'uint16', and 'float16'.`);\n      }\n      const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n      const byteBuffer =\n          buffer.slice(offset, offset + size * quantizationSizeFactor);\n      const quantizedArray = (quantization.dtype === 'uint8') ?\n          new Uint8Array(byteBuffer) :\n          new Uint16Array(byteBuffer);\n      if (dtype === 'float32') {\n        if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n          values = new Float32Array(quantizedArray.length);\n          for (let i = 0; i < quantizedArray.length; i++) {\n            const v = quantizedArray[i];\n            values[i] = v * quantization.scale + quantization.min;\n          }\n        } else if (quantization.dtype === 'float16') {\n          if (float16Decode === undefined) {\n            float16Decode = getFloat16Decoder();\n          }\n          values = float16Decode(quantizedArray as Uint16Array);\n        } else {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type float32.`);\n        }\n      } else if (dtype === 'int32') {\n        if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type int32.`);\n        }\n        values = new Int32Array(quantizedArray.length);\n        for (let i = 0; i < quantizedArray.length; i++) {\n          const v = quantizedArray[i];\n          values[i] = Math.round(v * quantization.scale + quantization.min);\n        }\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * quantizationSizeFactor;\n    } else if (dtype === 'string') {\n      const size = sizeFromShape(spec.shape);\n      values = [];\n      for (let i = 0; i < size; i++) {\n        const byteLength = new Uint32Array(\n            buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n        offset += NUM_BYTES_STRING_LENGTH;\n        const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n        (values as Uint8Array[]).push(bytes);\n        offset += byteLength;\n      }\n    } else {\n      const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n      const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n      if (dtype === 'float32') {\n        values = new Float32Array(byteBuffer);\n      } else if (dtype === 'int32') {\n        values = new Int32Array(byteBuffer);\n      } else if (dtype === 'bool') {\n        values = new Uint8Array(byteBuffer);\n      } else if (dtype === 'complex64') {\n        values = new Float32Array(byteBuffer);\n        const real = new Float32Array(values.length / 2);\n        const image = new Float32Array(values.length / 2);\n        for (let i = 0; i < real.length; i++) {\n          real[i] = values[i * 2];\n          image[i] = values[i * 2 + 1];\n        }\n        const realTensor = tensor(real, shape, 'float32');\n        const imageTensor = tensor(image, shape, 'float32');\n        out[name] = complex(realTensor, imageTensor);\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * dtypeFactor;\n    }\n    if (dtype !== 'complex64') {\n      out[name] = tensor(values, shape, dtype);\n    }\n  }\n  return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n  // TODO(adarob, cais): Support quantization.\n  if (xs === null) {\n    throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n  }\n\n  let totalByteLength = 0;\n\n  // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n  // can have a different byte length from that of the `TypedArray` itself,\n  // for example, when the `TypedArray` is created from an offset in an\n  // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n  // the `TypedArray` in byte length. If an element of `xs` does not show\n  // this property, a new `TypedArray` that satisfy this property will be\n  // constructed and pushed into `normalizedXs`.\n  const normalizedXs: TypedArray[] = [];\n  xs.forEach((x: TypedArray) => {\n    totalByteLength += x.byteLength;\n    // tslint:disable:no-any\n    normalizedXs.push(\n        x.byteLength === x.buffer.byteLength ? x :\n                                               new (x.constructor as any)(x));\n    if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n          x as any instanceof Uint8Array)) {\n      throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n    }\n    // tslint:enable:no-any\n  });\n\n  const y = new Uint8Array(totalByteLength);\n  let offset = 0;\n  normalizedXs.forEach((x: TypedArray) => {\n    y.set(new Uint8Array(x.buffer), offset);\n    offset += x.byteLength;\n  });\n\n  return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n    (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n     typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n  if (useNodeBuffer) {\n    return Buffer.byteLength(str);\n  }\n  return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n  if (useNodeBuffer) {\n    return Buffer.from(buffer).toString('base64');\n  }\n  const buf = new Uint8Array(buffer);\n  let s = '';\n  for (let i = 0, l = buf.length; i < l; i++) {\n    s += String.fromCharCode(buf[i]);\n  }\n  return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n  if (useNodeBuffer) {\n    const buf = Buffer.from(str, 'base64');\n    return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n  }\n  const s = atob(str);\n  const buffer = new Uint8Array(s.length);\n  for (let i = 0; i < s.length; ++i) {\n    buffer.set([s.charCodeAt(i)], i);\n  }\n  return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n  if (buffers.length === 1) {\n    return buffers[0];\n  }\n\n  let totalByteLength = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    totalByteLength += buffer.byteLength;\n  });\n\n  const temp = new Uint8Array(totalByteLength);\n  let offset = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    temp.set(new Uint8Array(buffer), offset);\n    offset += buffer.byteLength;\n  });\n  return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n  const SEPARATOR = '/';\n  path = path.trim();\n  while (path.endsWith(SEPARATOR)) {\n    path = path.slice(0, path.length - 1);\n  }\n  const items = path.split(SEPARATOR);\n  return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n    ModelArtifactsInfo {\n  if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n    throw new Error('Expected JSON model topology, received ArrayBuffer.');\n  }\n\n  return {\n    dateSaved: new Date(),\n    modelTopologyType: 'JSON',\n    modelTopologyBytes: modelArtifacts.modelTopology == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n    weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n    weightDataBytes: modelArtifacts.weightData == null ?\n        0 :\n        modelArtifacts.weightData.byteLength,\n  };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n  const convertMantissa = (i: number): number => {\n    let m = i << 13;\n    let e = 0;\n\n    while ((m & 0x00800000) === 0) {\n      e -= 0x00800000;\n      m <<= 1;\n    }\n    m &= ~0x00800000;\n    e += 0x38800000;\n\n    return m | e;\n  };\n\n  const mantisaTable = new Uint32Array(2048);\n\n  mantisaTable[0] = 0;\n  for (let i = 1; i < 1024; i++) {\n    mantisaTable[i] = convertMantissa(i);\n  }\n  for (let i = 1024; i < 2048; i++) {\n    mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n  }\n\n  return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n  const exponentTable = new Uint32Array(64);\n\n  exponentTable[0] = 0;\n  exponentTable[31] = 0x47800000;\n  exponentTable[32] = 0x80000000;\n  exponentTable[63] = 0xc7800000;\n  for (let i = 1; i < 31; i++) {\n    exponentTable[i] = i << 23;\n  }\n  for (let i = 33; i < 63; i++) {\n    exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n  }\n\n  return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n  const offsetTable = new Uint32Array(64);\n\n  for (let i = 0; i < 64; i++) {\n    offsetTable[i] = 1024;\n  }\n  offsetTable[0] = offsetTable[32] = 0;\n\n  return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n *          the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n  // Algorithm is based off of\n  // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n  // Cache lookup tables\n  const mantisaTable = computeFloat16MantisaTable();\n  const exponentTable = computeFloat16ExponentTable();\n  const offsetTable = computeFloat16OffsetTable();\n\n  return (quantizedArray: Uint16Array) => {\n    const buffer = new ArrayBuffer(4 * quantizedArray.length);\n    const bufferUint32View = new Uint32Array(buffer);\n    for (let index = 0; index < quantizedArray.length; index++) {\n      const float16Bits = quantizedArray[index];\n      const float32Bits =\n          mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n          exponentTable[float16Bits >> 10];\n      bufferUint32View[index] = float32Bits;\n    }\n    return new Float32Array(buffer);\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n    IOHandler;\n\nexport class IORouterRegistry {\n  // Singleton instance.\n  private static instance: IORouterRegistry;\n\n  private saveRouters: IORouter[];\n  private loadRouters: IORouter[];\n\n  private constructor() {\n    this.saveRouters = [];\n    this.loadRouters = [];\n  }\n\n  private static getInstance(): IORouterRegistry {\n    if (IORouterRegistry.instance == null) {\n      IORouterRegistry.instance = new IORouterRegistry();\n    }\n    return IORouterRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerSaveRouter(saveRouter: IORouter) {\n    IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n  }\n\n  /**\n   * Register a load-handler router.\n   *\n   * @param loadRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `load` method defined or `null`.\n   */\n  static registerLoadRouter(loadRouter: IORouter) {\n    IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n  }\n\n  /**\n   * Look up IOHandler for saving, given a URL-like string.\n   *\n   * @param url\n   * @returns If only one match is found, an instance of IOHandler with the\n   * `save` method defined. If no match is found, `null`.\n   * @throws Error, if more than one match is found.\n   */\n  static getSaveHandlers(url: string|string[]): IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'save');\n  }\n\n  /**\n   * Look up IOHandler for loading, given a URL-like string.\n   *\n   * @param url\n   * @param loadOptions Optional, custom load options.\n   * @returns All valid handlers for `url`, given the currently registered\n   *   handler routers.\n   */\n  static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n      IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n  }\n\n  private static getHandlers(\n      url: string|string[], handlerType: 'save'|'load',\n      loadOptions?: LoadOptions): IOHandler[] {\n    const validHandlers: IOHandler[] = [];\n    const routers = handlerType === 'load' ?\n        IORouterRegistry.getInstance().loadRouters :\n        IORouterRegistry.getInstance().saveRouters;\n    routers.forEach(router => {\n      const handler = router(url, loadOptions);\n      if (handler !== null) {\n        validHandlers.push(handler);\n      }\n    });\n    return validHandlers;\n  }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n    IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n    (url: string|string[], loadOptions?: LoadOptions) =>\n        IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n//    as the type of topology (JSON vs binary), byte size of the topology, byte\n//    size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise<void> {\n  const idbFactory = getIndexedDBFactory();\n\n  return new Promise<void>((resolve, reject) => {\n    const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n    deleteRequest.onsuccess = () => resolve();\n    deleteRequest.onerror = error => reject(error);\n  });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n  if (!env().getBool('IS_BROWSER')) {\n    // TODO(cais): Add more info about what IOHandler subtypes are available.\n    //   Maybe point to a doc page on the web and/or automatically determine\n    //   the available IOHandlers and print them in the error message.\n    throw new Error(\n        'Failed to obtain IndexedDB factory because the current environment' +\n        'is not a web browser.');\n  }\n  // tslint:disable-next-line:no-any\n  const theWindow: any = typeof window === 'undefined' ? self : window;\n  const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n      theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n      theWindow.shimIndexedDB;\n  if (factory == null) {\n    throw new Error(\n        'The current browser does not appear to support IndexedDB.');\n  }\n  return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n  const db = openRequest.result as IDBDatabase;\n  db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n  db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n  protected readonly indexedDB: IDBFactory;\n  protected readonly modelPath: string;\n\n  static readonly URL_SCHEME = 'indexeddb://';\n\n  constructor(modelPath: string) {\n    this.indexedDB = getIndexedDBFactory();\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For IndexedDB, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    // TODO(cais): Support saving GraphDef models.\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    return this.databaseAction(this.modelPath, modelArtifacts) as\n        Promise<SaveResult>;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    return this.databaseAction(this.modelPath) as Promise<ModelArtifacts>;\n  }\n\n  /**\n   * Perform database action to put model artifacts into or read model artifacts\n   * from IndexedDB object store.\n   *\n   * Whether the action is put or get depends on whether `modelArtifacts` is\n   * specified. If it is specified, the action will be put; otherwise the action\n   * will be get.\n   *\n   * @param modelPath A unique string path for the model.\n   * @param modelArtifacts If specified, it will be the model artifacts to be\n   *   stored in IndexedDB.\n   * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n   *   of `ModelArtifacts`, if the action is get.\n   */\n  private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n      Promise<ModelArtifacts|SaveResult> {\n    return new Promise<ModelArtifacts|SaveResult>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n\n        if (modelArtifacts == null) {\n          // Read model out from object store.\n          const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n          const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n          const getRequest = modelStore.get(this.modelPath);\n          getRequest.onsuccess = () => {\n            if (getRequest.result == null) {\n              db.close();\n              return reject(new Error(\n                  `Cannot find model with path '${this.modelPath}' ` +\n                  `in IndexedDB.`));\n            } else {\n              resolve(getRequest.result.modelArtifacts);\n            }\n          };\n          getRequest.onerror = error => {\n            db.close();\n            return reject(getRequest.error);\n          };\n          modelTx.oncomplete = () => db.close();\n        } else {\n          // Put model into object store.\n          const modelArtifactsInfo: ModelArtifactsInfo =\n              getModelArtifactsInfoForJSON(modelArtifacts);\n          // First, put ModelArtifactsInfo into info store.\n          const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n          let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n          const putInfoRequest =\n              infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n          let modelTx: IDBTransaction;\n          putInfoRequest.onsuccess = () => {\n            // Second, put model data into model store.\n            modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n            const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n            const putModelRequest = modelStore.put({\n              modelPath: this.modelPath,\n              modelArtifacts,\n              modelArtifactsInfo\n            });\n            putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n            putModelRequest.onerror = error => {\n              // If the put-model request fails, roll back the info entry as\n              // well.\n              infoStore = infoTx.objectStore(INFO_STORE_NAME);\n              const deleteInfoRequest = infoStore.delete(this.modelPath);\n              deleteInfoRequest.onsuccess = () => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n              deleteInfoRequest.onerror = error => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n            };\n          };\n          putInfoRequest.onerror = error => {\n            db.close();\n            return reject(putInfoRequest.error);\n          };\n          infoTx.oncomplete = () => {\n            if (modelTx == null) {\n              db.close();\n            } else {\n              modelTx.oncomplete = () => db.close();\n            }\n          };\n        }\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n      return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n *   which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n  return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n      key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n      key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n  private indexedDB: IDBFactory;\n\n  constructor() {\n    this.indexedDB = getIndexedDBFactory();\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    return new Promise<{[path: string]: ModelArtifactsInfo}>(\n        (resolve, reject) => {\n          const openRequest =\n              this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n          openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n          openRequest.onsuccess = () => {\n            const db = openRequest.result;\n            const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n            const store = tx.objectStore(INFO_STORE_NAME);\n            // tslint:disable:max-line-length\n            // Need to cast `store` as `any` here because TypeScript's DOM\n            // library does not have the `getAll()` method even though the\n            // method is supported in the latest version of most mainstream\n            // browsers:\n            // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n            // tslint:enable:max-line-length\n            // tslint:disable-next-line:no-any\n            const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n            getAllInfoRequest.onsuccess = () => {\n              const out: {[path: string]: ModelArtifactsInfo} = {};\n              for (const item of getAllInfoRequest.result) {\n                out[item.modelPath] = item.modelArtifactsInfo;\n              }\n              resolve(out);\n            };\n            getAllInfoRequest.onerror = error => {\n              db.close();\n              return reject(getAllInfoRequest.error);\n            };\n            tx.oncomplete = () => db.close();\n          };\n          openRequest.onerror = error => reject(openRequest.error);\n        });\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    return new Promise<ModelArtifactsInfo>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n        const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n        const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n        const getInfoRequest = infoStore.get(path);\n        let modelTx: IDBTransaction;\n        getInfoRequest.onsuccess = () => {\n          if (getInfoRequest.result == null) {\n            db.close();\n            return reject(new Error(\n                `Cannot find model with path '${path}' ` +\n                `in IndexedDB.`));\n          } else {\n            // First, delete the entry in the info store.\n            const deleteInfoRequest = infoStore.delete(path);\n            const deleteModelData = () => {\n              // Second, delete the entry in the model store.\n              modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n              const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n              const deleteModelRequest = modelStore.delete(path);\n              deleteModelRequest.onsuccess = () =>\n                  resolve(getInfoRequest.result.modelArtifactsInfo);\n              deleteModelRequest.onerror = error =>\n                  reject(getInfoRequest.error);\n            };\n            // Proceed with deleting model data regardless of whether deletion\n            // of info data succeeds or not.\n            deleteInfoRequest.onsuccess = deleteModelData;\n            deleteInfoRequest.onerror = error => {\n              deleteModelData();\n              db.close();\n              return reject(getInfoRequest.error);\n            };\n          }\n        };\n        getInfoRequest.onerror = error => {\n          db.close();\n          return reject(getInfoRequest.error);\n        };\n\n        infoTx.oncomplete = () => {\n          if (modelTx == null) {\n            db.close();\n          } else {\n            modelTx.oncomplete = () => db.close();\n          }\n        };\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n  if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n      typeof window.localStorage === 'undefined') {\n    throw new Error(\n        'purgeLocalStorageModels() cannot proceed because local storage is ' +\n        'unavailable in the current environment.');\n  }\n  const LS = window.localStorage;\n  const purgedModelPaths: string[] = [];\n  for (let i = 0; i < LS.length; ++i) {\n    const key = LS.key(i);\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    if (key.startsWith(prefix) && key.length > prefix.length) {\n      LS.removeItem(key);\n      const modelName = getModelPathFromKey(key);\n      if (purgedModelPaths.indexOf(modelName) === -1) {\n        purgedModelPaths.push(modelName);\n      }\n    }\n  }\n  return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n} {\n  return {\n    info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n    topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n    weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n    weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n    modelMetadata:\n        [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n  };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n  const items = key.split(PATH_SEPARATOR);\n  if (items.length < 3) {\n    throw new Error(`Invalid key format: ${key}`);\n  }\n  return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n      key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n      key;\n}\n\ndeclare type LocalStorageKeys = {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n  protected readonly LS: Storage;\n  protected readonly modelPath: string;\n  protected readonly keys: LocalStorageKeys;\n\n  static readonly URL_SCHEME = 'localstorage://';\n\n  constructor(modelPath: string) {\n    if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n        typeof window.localStorage === 'undefined') {\n      // TODO(cais): Add more info about what IOHandler subtypes are\n      // available.\n      //   Maybe point to a doc page on the web and/or automatically determine\n      //   the available IOHandlers and print them in the error message.\n      throw new Error(\n          'The current environment does not support local storage.');\n    }\n    this.LS = window.localStorage;\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For local storage, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n    this.keys = getModelKeys(this.modelPath);\n  }\n\n  /**\n   * Save model artifacts to browser local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @param modelArtifacts The model artifacts to be stored.\n   * @returns An instance of SaveResult.\n   */\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const topology = JSON.stringify(modelArtifacts.modelTopology);\n      const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n      const modelArtifactsInfo: ModelArtifactsInfo =\n          getModelArtifactsInfoForJSON(modelArtifacts);\n\n      try {\n        this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n        this.LS.setItem(this.keys.topology, topology);\n        this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n        this.LS.setItem(\n            this.keys.weightData,\n            arrayBufferToBase64String(modelArtifacts.weightData));\n        this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n          format: modelArtifacts.format,\n          generatedBy: modelArtifacts.generatedBy,\n          convertedBy: modelArtifacts.convertedBy,\n          userDefinedMetadata: modelArtifacts.userDefinedMetadata\n        }));\n\n        return {modelArtifactsInfo};\n      } catch (err) {\n        // If saving failed, clean up all items saved so far.\n        this.LS.removeItem(this.keys.info);\n        this.LS.removeItem(this.keys.topology);\n        this.LS.removeItem(this.keys.weightSpecs);\n        this.LS.removeItem(this.keys.weightData);\n        this.LS.removeItem(this.keys.modelMetadata);\n\n        throw new Error(\n            `Failed to save model '${this.modelPath}' to local storage: ` +\n            `size quota being exceeded is a possible cause of this failure: ` +\n            `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n            `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n            `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n      }\n    }\n  }\n\n  /**\n   * Load a model from local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const info =\n        JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n    if (info == null) {\n      throw new Error(\n          `In local storage, there is no model with name '${this.modelPath}'`);\n    }\n\n    if (info.modelTopologyType !== 'JSON') {\n      throw new Error(\n          'BrowserLocalStorage does not support loading non-JSON model ' +\n          'topology yet.');\n    }\n\n    const out: ModelArtifacts = {};\n\n    // Load topology.\n    const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n    if (topology == null) {\n      throw new Error(\n          `In local storage, the topology of model '${this.modelPath}' ` +\n          `is missing.`);\n    }\n    out.modelTopology = topology;\n\n    // Load weight specs.\n    const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n    if (weightSpecs == null) {\n      throw new Error(\n          `In local storage, the weight specs of model '${this.modelPath}' ` +\n          `are missing.`);\n    }\n    out.weightSpecs = weightSpecs;\n\n    // Load meta-data fields.\n    const metadataString = this.LS.getItem(this.keys.modelMetadata);\n    if (metadataString != null) {\n      const metadata = JSON.parse(metadataString) as ModelArtifacts;\n      out.format = metadata['format'];\n      out.generatedBy = metadata['generatedBy'];\n      out.convertedBy = metadata['convertedBy'];\n      out.userDefinedMetadata = metadata['userDefinedMetadata'];\n    }\n\n    // Load weight data.\n    const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n    if (weightDataBase64 == null) {\n      throw new Error(\n          `In local storage, the binary weight values of model ` +\n          `'${this.modelPath}' are missing.`);\n    }\n    out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n    return out;\n  }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n      return browserLocalStorage(\n          url.slice(BrowserLocalStorage.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n *   - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n *     model, such as date saved, type of the topology, size in bytes, etc.\n *   - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n *     style models, this is a stringized JSON.\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n *     model, can be used to decode the saved binary weight values (see\n *     item below).\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n *     weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n *   `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n  return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n  private readonly LS: Storage;\n\n  constructor() {\n    assert(\n        env().getBool('IS_BROWSER'),\n        () => 'Current environment is not a web browser');\n    assert(\n        typeof window === 'undefined' ||\n            typeof window.localStorage !== 'undefined',\n        () => 'Current browser does not appear to support localStorage');\n    this.LS = window.localStorage;\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    const out: {[path: string]: ModelArtifactsInfo} = {};\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n    for (let i = 0; i < this.LS.length; ++i) {\n      const key = this.LS.key(i);\n      if (key.startsWith(prefix) && key.endsWith(suffix)) {\n        const modelPath = getModelPathFromKey(key);\n        out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n      }\n    }\n    return out;\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    const keys = getModelKeys(path);\n    if (this.LS.getItem(keys.info) == null) {\n      throw new Error(`Cannot find model at path '${path}'`);\n    }\n    const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n    this.LS.removeItem(keys.info);\n    this.LS.removeItem(keys.topology);\n    this.LS.removeItem(keys.weightSpecs);\n    this.LS.removeItem(keys.weightData);\n    return info;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n *   string.\n * - Moving or copying model from one path to another in the same medium or from\n *   one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n  // Singleton instance.\n  private static instance: ModelStoreManagerRegistry;\n\n  private managers: {[scheme: string]: ModelStoreManager};\n\n  private constructor() {\n    this.managers = {};\n  }\n\n  private static getInstance(): ModelStoreManagerRegistry {\n    if (ModelStoreManagerRegistry.instance == null) {\n      ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n    }\n    return ModelStoreManagerRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerManager(scheme: string, manager: ModelStoreManager) {\n    assert(scheme != null, () => 'scheme must not be undefined or null.');\n    if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n      scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n    }\n    assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n    const registry = ModelStoreManagerRegistry.getInstance();\n    assert(\n        registry.managers[scheme] == null,\n        () => `A model store manager is already registered for scheme '${\n            scheme}'.`);\n    registry.managers[scheme] = manager;\n  }\n\n  static getManager(scheme: string): ModelStoreManager {\n    const manager = this.getInstance().managers[scheme];\n    if (manager == null) {\n      throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n    }\n    return manager;\n  }\n\n  static getSchemes(): string[] {\n    return Object.keys(this.getInstance().managers);\n  }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n *   Scheme: e.g., 'localstorage' in the example above.\n *   Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n  if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n    throw new Error(\n        `The url string provided does not contain a scheme. ` +\n        `Supported schemes are: ` +\n        `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n  }\n  return {\n    scheme: url.split(URL_SCHEME_SUFFIX)[0],\n    path: url.split(URL_SCHEME_SUFFIX)[1],\n  };\n}\n\nasync function cloneModelInternal(\n    sourceURL: string, destURL: string,\n    deleteSource = false): Promise<ModelArtifactsInfo> {\n  assert(\n      sourceURL !== destURL,\n      () => `Old path and new path are the same: '${sourceURL}'`);\n\n  const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n  assert(\n      loadHandlers.length > 0,\n      () => `Copying failed because no load handler is found for source URL ${\n          sourceURL}.`);\n  assert(\n      loadHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `load handlers for source URL ${sourceURL}.`);\n  const loadHandler = loadHandlers[0];\n\n  const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n  assert(\n      saveHandlers.length > 0,\n      () => `Copying failed because no save handler is found for destination ` +\n          `URL ${destURL}.`);\n  assert(\n      saveHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `save handlers for destination URL ${destURL}.`);\n  const saveHandler = saveHandlers[0];\n\n  const sourceScheme = parseURL(sourceURL).scheme;\n  const sourcePath = parseURL(sourceURL).path;\n  const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n  const modelArtifacts = await loadHandler.load();\n\n  // If moving within the same storage medium, remove the old model as soon as\n  // the loading is done. Without doing this, it is possible that the combined\n  // size of the two models will cause the cloning to fail.\n  if (deleteSource && sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  const saveResult = await saveHandler.save(modelArtifacts);\n\n  // If moving between mediums, the deletion is done after the save succeeds.\n  // This guards against the case in which saving to the destination medium\n  // fails.\n  if (deleteSource && !sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n *   'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n  const schemes = ModelStoreManagerRegistry.getSchemes();\n  const out: {[url: string]: ModelArtifactsInfo} = {};\n  for (const scheme of schemes) {\n    const schemeOut =\n        await ModelStoreManagerRegistry.getManager(scheme).listModels();\n    for (const path in schemeOut) {\n      const url = scheme + URL_SCHEME_SUFFIX + path;\n      out[url] = schemeOut[path];\n    }\n  }\n  return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n *   'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n *   is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise<ModelArtifactsInfo> {\n  const schemeAndPath = parseURL(url);\n  const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n  return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function copyModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = false;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function moveModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = true;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n  // According to the spec, the built-in encoder can do only UTF-8 encoding.\n  // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n  private textEncoder: TextEncoder;\n\n  fetch(path: string, init?: RequestInit): Promise<Response> {\n    return fetch(path, init);\n  }\n\n  now(): number {\n    return performance.now();\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Browser's encoder only supports utf-8, but got ${encoding}`);\n    }\n    if (this.textEncoder == null) {\n      this.textEncoder = new TextEncoder();\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    return new TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_BROWSER')) {\n  env().setPlatform('browser', new PlatformBrowser());\n\n  // Register LocalStorage IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n  } catch (err) {\n  }\n\n  // Register IndexedDB IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n  } catch (err) {\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n  // tslint:disable-next-line:no-require-imports\n  importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise<Response>;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n  systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n  systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n  return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n  private textEncoder: TextEncoder;\n  // tslint:disable-next-line:no-any\n  util: any;\n\n  constructor() {\n    // tslint:disable-next-line:no-require-imports\n    this.util = require('util');\n    // According to the spec, the built-in encoder can do only UTF-8 encoding.\n    // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n    this.textEncoder = new this.util.TextEncoder();\n  }\n\n  fetch(path: string, requestInits?: RequestInit): Promise<Response> {\n    if (env().global.fetch != null) {\n      return env().global.fetch(path, requestInits);\n    }\n\n    if (systemFetch == null) {\n      systemFetch = getNodeFetch.importFetch();\n    }\n    return systemFetch(path, requestInits);\n  }\n\n  now(): number {\n    const time = process.hrtime();\n    return time[0] * 1000 + time[1] / 1000000;\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Node built-in encoder only supports utf-8, but got ${encoding}`);\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    if (bytes.length === 0) {\n      return '';\n    }\n    return new this.util.TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_NODE')) {\n  env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer<R extends Rank, D extends DataType = 'float32'>(\n    shape: ShapeMap[R], dtype: D = 'float32' as D,\n    values?: DataTypeMap[D]): TensorBuffer<R, D> {\n  dtype = dtype || 'float32' as D;\n  util.assertNonNegativeIntegerDimensions(shape);\n  return new TensorBuffer<R, D>(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_<T extends Tensor>(x: T|TensorLike, dtype: DataType): T {\n  const $x = convertToTensor(x, 'x', 'cast');\n\n  // Sanity checks.\n  if (!util.isValidDtype(dtype)) {\n    throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n  }\n  if (dtype === 'string' && $x.dtype !== 'string' ||\n      dtype !== 'string' && $x.dtype === 'string') {\n    throw new Error('Only strings can be casted to strings');\n  }\n\n  const inputs: CastInputs = {x: $x};\n  const attrs: CastAttrs = {dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n      null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'clone', null);\n  const forward = () =>\n      ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n  const inputs: IdentityInputs = {x: $x};\n\n  // Note this op is called tf.identity in python. Hence the kernel name used\n  // here.\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print<T extends Tensor>(x: T, verbose = false): void {\n  console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n  buffer,\n  cast,\n  clone,\n  print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer<T>(f: () => T): Promise<T> {\n  return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n  private readonly modelTopologyFileName: string;\n  private readonly weightDataFileName: string;\n  private readonly jsonAnchor: HTMLAnchorElement;\n  private readonly weightDataAnchor: HTMLAnchorElement;\n\n  static readonly URL_SCHEME = 'downloads://';\n\n  constructor(fileNamePrefix?: string) {\n    if (!env().getBool('IS_BROWSER')) {\n      // TODO(cais): Provide info on what IOHandlers are available under the\n      //   current environment.\n      throw new Error(\n          'browserDownloads() cannot proceed because the current environment ' +\n          'is not a browser.');\n    }\n\n    if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n      fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n    }\n    if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n      fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n    }\n\n    this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n    this.weightDataFileName =\n        fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (typeof (document) === 'undefined') {\n      throw new Error(\n          'Browser downloads are not supported in ' +\n          'this environment since `document` is not present');\n    }\n    const weightsURL = window.URL.createObjectURL(new Blob(\n        [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserDownloads.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const weightsManifest: WeightsManifestConfig = [{\n        paths: ['./' + this.weightDataFileName],\n        weights: modelArtifacts.weightSpecs\n      }];\n      const modelTopologyAndWeightManifest: ModelJSON = {\n        modelTopology: modelArtifacts.modelTopology,\n        format: modelArtifacts.format,\n        generatedBy: modelArtifacts.generatedBy,\n        convertedBy: modelArtifacts.convertedBy,\n        weightsManifest\n      };\n      const modelTopologyAndWeightManifestURL =\n          window.URL.createObjectURL(new Blob(\n              [JSON.stringify(modelTopologyAndWeightManifest)],\n              {type: 'application/json'}));\n\n      // If anchor elements are not provided, create them without attaching them\n      // to parents, so that the downloaded file names can be controlled.\n      const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n                                                   this.jsonAnchor;\n      jsonAnchor.download = this.modelTopologyFileName;\n      jsonAnchor.href = modelTopologyAndWeightManifestURL;\n      // Trigger downloads by evoking a click event on the download anchors.\n      // When multiple downloads are started synchronously, Firefox will only\n      // save the last one.\n      await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n      if (modelArtifacts.weightData != null) {\n        const weightDataAnchor = this.weightDataAnchor == null ?\n            document.createElement('a') :\n            this.weightDataAnchor;\n        weightDataAnchor.download = this.weightDataFileName;\n        weightDataAnchor.href = weightsURL;\n        await defer(\n            () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n      }\n\n      return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n    }\n  }\n}\n\nclass BrowserFiles implements IOHandler {\n  private readonly files: File[];\n\n  constructor(files: File[]) {\n    if (files == null || files.length < 1) {\n      throw new Error(\n          `When calling browserFiles, at least 1 file is required, ` +\n          `but received ${files}`);\n    }\n    this.files = files;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    const jsonFile = this.files[0];\n    const weightFiles = this.files.slice(1);\n\n    return new Promise<ModelArtifacts>((resolve, reject) => {\n      const jsonReader = new FileReader();\n      jsonReader.onload = (event: Event) => {\n        // tslint:disable-next-line:no-any\n        const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n        const modelTopology = modelJSON.modelTopology;\n        if (modelTopology == null) {\n          reject(new Error(\n              `modelTopology field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        if (weightFiles.length === 0) {\n          resolve({modelTopology});\n        }\n\n        const weightsManifest = modelJSON.weightsManifest;\n        if (weightsManifest == null) {\n          reject(new Error(\n              `weightManifest field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        let pathToFile: {[path: string]: File};\n        try {\n          pathToFile =\n              this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n        } catch (err) {\n          reject(err);\n          return;\n        }\n\n        const weightSpecs: WeightsManifestEntry[] = [];\n        const paths: string[] = [];\n        const perFileBuffers: ArrayBuffer[] = [];\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            paths.push(path);\n            perFileBuffers.push(null);\n          });\n          weightSpecs.push(...weightsGroup.weights);\n        });\n\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            const weightFileReader = new FileReader();\n            weightFileReader.onload = (event: Event) => {\n              // tslint:disable-next-line:no-any\n              const weightData = (event.target as any).result as ArrayBuffer;\n              const index = paths.indexOf(path);\n              perFileBuffers[index] = weightData;\n              if (perFileBuffers.indexOf(null) === -1) {\n                resolve({\n                  modelTopology,\n                  weightSpecs,\n                  weightData: concatenateArrayBuffers(perFileBuffers),\n                  format: modelJSON.format,\n                  generatedBy: modelJSON.generatedBy,\n                  convertedBy: modelJSON.convertedBy,\n                  userDefinedMetadata: modelJSON.userDefinedMetadata\n                });\n              }\n            };\n            weightFileReader.onerror = error =>\n                reject(`Failed to weights data from file of path '${path}'.`);\n            weightFileReader.readAsArrayBuffer(pathToFile[path]);\n          });\n        });\n      };\n      jsonReader.onerror = error => reject(\n          `Failed to read model topology and weights manifest JSON ` +\n          `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n          `Keras-style tf.Model artifacts only.`);\n      jsonReader.readAsText(jsonFile);\n    });\n  }\n\n  /**\n   * Check the compatibility between weights manifest and weight files.\n   */\n  private checkManifestAndWeightFiles(\n      manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n    const basenames: string[] = [];\n    const fileNames = files.map(file => basename(file.name));\n    const pathToFile: {[path: string]: File} = {};\n    for (const group of manifest) {\n      group.paths.forEach(path => {\n        const pathBasename = basename(path);\n        if (basenames.indexOf(pathBasename) !== -1) {\n          throw new Error(\n              `Duplicate file basename found in weights manifest: ` +\n              `'${pathBasename}'`);\n        }\n        basenames.push(pathBasename);\n        if (fileNames.indexOf(pathBasename) === -1) {\n          throw new Error(\n              `Weight file with basename '${pathBasename}' is not provided.`);\n        } else {\n          pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n        }\n      });\n    }\n\n    if (basenames.length !== files.length) {\n      throw new Error(\n          `Mismatch in the number of files in weights manifest ` +\n          `(${basenames.length}) and the number of weight files provided ` +\n          `(${files.length}).`);\n    }\n    return pathToFile;\n  }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n      return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * //   'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n *   `tf.Model`, `fileNamePrefix` should follow either of the following two\n *   formats:\n *   1. `null` or `undefined`, in which case the default file\n *      names will be used:\n *      - 'model.json' for the JSON file containing the model topology and\n *        weights manifest.\n *      - 'model.weights.bin' for the binary file containing the binary weight\n *        values.\n *   2. A single string or an Array of a single string, as the file name prefix.\n *      For example, if `'foo'` is provided, the downloaded JSON\n *      file and binary weights file will be named 'foo.json' and\n *      'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n  return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * //   elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (`<input type=\"file\" ...>`)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n *   which an `Array` of `File`s is expected (in that order):\n *   - A JSON file containing the model topology and weight manifest.\n *   - Optionally, One or more binary files containing the binary weights.\n *     These files must have names that match the paths in the `weightsManifest`\n *     contained by the aforementioned JSON file, or errors will be thrown\n *     during loading. These weights files have the same format as the ones\n *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n *     Python PIP package. If no weights files are provided, only the model\n *     topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n  return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n    promises: Array<Promise<{}|void>>, onProgress: OnProgressCallback,\n    startFraction?: number, endFraction?: number) {\n  checkPromises(promises);\n  startFraction = startFraction == null ? 0 : startFraction;\n  endFraction = endFraction == null ? 1 : endFraction;\n  checkFraction(startFraction, endFraction);\n  let resolvedPromise = 0;\n\n  const registerMonitor = (promise: Promise<{}>) => {\n    promise.then(value => {\n      const fraction = startFraction +\n          ++resolvedPromise / promises.length * (endFraction - startFraction);\n      // pass fraction as parameter to callback function.\n      onProgress(fraction);\n      return value;\n    });\n    return promise;\n  };\n\n  function checkPromises(promises: Array<Promise<{}|void>>): void {\n    assert(\n        promises != null && Array.isArray(promises) && promises.length > 0,\n        () => 'promises must be a none empty array');\n  }\n\n  function checkFraction(startFraction: number, endFraction: number): void {\n    assert(\n        startFraction >= 0 && startFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got startFraction ${startFraction}`);\n    assert(\n        endFraction >= 0 && endFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got endFraction ${endFraction}`);\n    assert(\n        endFraction >= startFraction,\n        () => `startFraction must be no more than endFraction, but ` +\n            `got startFraction ${startFraction} and endFraction ` +\n            `${endFraction}`);\n  }\n\n  return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n *   before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n *   length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n    fetchURLs: string[], loadOptions?: LoadOptions): Promise<ArrayBuffer[]> {\n  if (loadOptions == null) {\n    loadOptions = {};\n  }\n\n  const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n                                                    loadOptions.fetchFunc;\n\n  // Create the requests for all of the weights in parallel.\n  const requests = fetchURLs.map(\n      fetchURL =>\n          fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n  const fetchStartFraction = 0;\n  const fetchEndFraction = 0.5;\n\n  const responses = loadOptions.onProgress == null ?\n      await Promise.all(requests) :\n      await monitorPromisesProgress(\n          requests, loadOptions.onProgress, fetchStartFraction,\n          fetchEndFraction);\n\n  const bufferPromises = responses.map(response => response.arrayBuffer());\n\n  const bufferStartFraction = 0.5;\n  const bufferEndFraction = 1;\n\n  const buffers = loadOptions.onProgress == null ?\n      await Promise.all(bufferPromises) :\n      await monitorPromisesProgress(\n          bufferPromises, loadOptions.onProgress, bufferStartFraction,\n          bufferEndFraction);\n  return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n *     Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n    manifest: WeightsManifestConfig, filePathPrefix = '',\n    weightNames?: string[],\n    requestInit?: RequestInit): Promise<NamedTensorMap> {\n  // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n  // single weight from a group, the whole group will be fetched. At a future\n  // date, we should support fetching only the individual shards within a\n  // group that are needed to reconstruct the requested weight.\n  // TODO(cais): Use `decodeWeights` for implementation.\n\n  const fetchWeights = (fetchUrls: string[]) =>\n      loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n  const loadWeights = weightsLoaderFactory(fetchWeights);\n\n  return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n *   filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n *   fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n    fetchWeightsFunction: (fetchUrls: string[]) => Promise<ArrayBuffer[]>):\n    (manifest: WeightsManifestConfig, filePathPrefix?: string,\n     weightNames?: string[]) => Promise<NamedTensorMap> {\n  return async(\n             manifest: WeightsManifestConfig, filePathPrefix = '',\n             weightNames?: string[]): Promise<NamedTensorMap> => {\n    // Collect all the groups, weights, and their relative offsets to be\n    // fetched.\n    const groupIndicesToFetchMap = manifest.map(() => false);\n    const groupWeightsToFetch: {\n      [group: number]: Array<{\n        manifestEntry: WeightsManifestEntry; groupOffset: number;\n        sizeBytes: number;\n      }>\n    } = {};\n    const weightsFound =\n        weightNames != null ? weightNames.map(() => false) : [];\n    const allManifestWeightNames: string[] = [];\n    manifest.forEach((manifestGroupConfig, groupIndex) => {\n      let groupOffset = 0;\n      manifestGroupConfig.weights.forEach(weightsEntry => {\n        const rawDtype = ('quantization' in weightsEntry) ?\n            weightsEntry.quantization.dtype :\n            weightsEntry.dtype;\n\n        const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n            util.sizeFromShape(weightsEntry.shape);\n\n        const enqueueWeightsForFetchingFn = () => {\n          groupIndicesToFetchMap[groupIndex] = true;\n          if (groupWeightsToFetch[groupIndex] == null) {\n            groupWeightsToFetch[groupIndex] = [];\n          }\n\n          groupWeightsToFetch[groupIndex].push({\n            manifestEntry: weightsEntry,\n            groupOffset,\n            sizeBytes: weightsBytes\n          });\n        };\n\n        if (weightNames != null) {\n          weightNames.forEach((weightName, weightIndex) => {\n            if (weightName === weightsEntry.name) {\n              enqueueWeightsForFetchingFn();\n              weightsFound[weightIndex] = true;\n            }\n          });\n        } else {\n          enqueueWeightsForFetchingFn();\n        }\n\n        allManifestWeightNames.push(weightsEntry.name);\n        groupOffset += weightsBytes;\n      });\n    });\n\n    if (!weightsFound.every(found => found)) {\n      const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n      throw new Error(\n          `Could not find weights in manifest with names: ` +\n          `${weightsNotFound.join(', ')}. \\n` +\n          `Manifest JSON has weights with names: ` +\n          `${allManifestWeightNames.join(', ')}.`);\n    }\n\n    // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n    // IDs.\n    const groupIndicesToFetch =\n        groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n          if (shouldFetch) {\n            accumulator.push(i);\n          }\n          return accumulator;\n        }, []);\n\n    const fetchUrls: string[] = [];\n    groupIndicesToFetch.forEach(i => {\n      manifest[i].paths.forEach(filepath => {\n        const fetchUrl = filePathPrefix +\n            (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n        fetchUrls.push(fetchUrl);\n      });\n    });\n    const buffers = await fetchWeightsFunction(fetchUrls);\n\n    const weightsTensorMap: NamedTensorMap = {};\n    let bufferIndexOffset = 0;\n    groupIndicesToFetch.forEach(i => {\n      const numBuffers = manifest[i].paths.length;\n\n      let groupBytes = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        groupBytes += buffers[bufferIndexOffset + i].byteLength;\n      }\n\n      // Create a buffer for the whole group.\n      const groupBuffer = new ArrayBuffer(groupBytes);\n      const groupByteBuffer = new Uint8Array(groupBuffer);\n      let groupBufferOffset = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n        groupByteBuffer.set(buffer, groupBufferOffset);\n        groupBufferOffset += buffer.byteLength;\n      }\n\n      const weightsEntries = groupWeightsToFetch[i];\n      weightsEntries.forEach(weightsEntry => {\n        const byteBuffer = groupBuffer.slice(\n            weightsEntry.groupOffset,\n            weightsEntry.groupOffset + weightsEntry.sizeBytes);\n        const nameToTensorMap =\n            decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n        for (const name in nameToTensorMap) {\n          weightsTensorMap[name] = nameToTensorMap[name];\n        }\n      });\n\n      bufferIndexOffset += numBuffers;\n    });\n\n    return weightsTensorMap;\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n  protected readonly path: string;\n  protected readonly requestInit: RequestInit;\n\n  private readonly fetch: Function;\n  private readonly weightUrlConverter: (weightName: string) => Promise<string>;\n\n  readonly DEFAULT_METHOD = 'POST';\n\n  static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n  private readonly weightPathPrefix: string;\n  private readonly onProgress: OnProgressCallback;\n\n  constructor(path: string, loadOptions?: LoadOptions) {\n    if (loadOptions == null) {\n      loadOptions = {};\n    }\n    this.weightPathPrefix = loadOptions.weightPathPrefix;\n    this.onProgress = loadOptions.onProgress;\n    this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n    if (loadOptions.fetchFunc != null) {\n      assert(\n          typeof loadOptions.fetchFunc === 'function',\n          () => 'Must pass a function that matches the signature of ' +\n              '`fetch` (see ' +\n              'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n      this.fetch = loadOptions.fetchFunc;\n    } else {\n      this.fetch = env().platform.fetch;\n    }\n\n    assert(\n        path != null && path.length > 0,\n        () => 'URL path for http must not be null, undefined or ' +\n            'empty.');\n\n    if (Array.isArray(path)) {\n      assert(\n          path.length === 2,\n          () => 'URL paths for http must have a length of 2, ' +\n              `(actual length is ${path.length}).`);\n    }\n    this.path = path;\n\n    if (loadOptions.requestInit != null &&\n        loadOptions.requestInit.body != null) {\n      throw new Error(\n          'requestInit is expected to have no pre-existing body, but has one.');\n    }\n    this.requestInit = loadOptions.requestInit || {};\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserHTTPRequest.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n    init.body = new FormData();\n\n    const weightsManifest: WeightsManifestConfig = [{\n      paths: ['./model.weights.bin'],\n      weights: modelArtifacts.weightSpecs,\n    }];\n    const modelTopologyAndWeightManifest: ModelJSON = {\n      modelTopology: modelArtifacts.modelTopology,\n      format: modelArtifacts.format,\n      generatedBy: modelArtifacts.generatedBy,\n      convertedBy: modelArtifacts.convertedBy,\n      userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n      weightsManifest\n    };\n\n    init.body.append(\n        'model.json',\n        new Blob(\n            [JSON.stringify(modelTopologyAndWeightManifest)],\n            {type: JSON_TYPE}),\n        'model.json');\n\n    if (modelArtifacts.weightData != null) {\n      init.body.append(\n          'model.weights.bin',\n          new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n          'model.weights.bin');\n    }\n\n    const response = await this.fetch(this.path, init);\n\n    if (response.ok) {\n      return {\n        modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n        responses: [response],\n      };\n    } else {\n      throw new Error(\n          `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n          `${response.status}.`);\n    }\n  }\n\n  /**\n   * Load model artifacts via HTTP request(s).\n   *\n   * See the documentation to `tf.io.http` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model artifacts (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n    if (!modelConfigRequest.ok) {\n      throw new Error(\n          `Request to ${this.path} failed with status code ` +\n          `${modelConfigRequest.status}. Please verify this URL points to ` +\n          `the model JSON of the model to load.`);\n    }\n    let modelConfig: ModelJSON;\n    try {\n      modelConfig = await modelConfigRequest.json();\n    } catch (e) {\n      let message = `Failed to parse model JSON of response from ${this.path}.`;\n      // TODO(nsthorat): Remove this after some time when we're comfortable that\n      // .pb files are mostly gone.\n      if (this.path.endsWith('.pb')) {\n        message += ' Your path contains a .pb file extension. ' +\n            'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n            'in favor of .json models. You can re-convert your Python ' +\n            'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n            'or you can convert your.pb models with the \\'pb2json\\'' +\n            'NPM script in the tensorflow/tfjs-converter repository.';\n      } else {\n        message += ' Please make sure the server is serving valid ' +\n            'JSON for this request.';\n      }\n      throw new Error(message);\n    }\n    const modelTopology = modelConfig.modelTopology;\n    const weightsManifest = modelConfig.weightsManifest;\n    const generatedBy = modelConfig.generatedBy;\n    const convertedBy = modelConfig.convertedBy;\n    const format = modelConfig.format;\n    const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n    // We do not allow both modelTopology and weightsManifest to be missing.\n    if (modelTopology == null && weightsManifest == null) {\n      throw new Error(\n          `The JSON from HTTP path ${this.path} contains neither model ` +\n          `topology or manifest for weights.`);\n    }\n\n    let weightSpecs: WeightsManifestEntry[];\n    let weightData: ArrayBuffer;\n    if (weightsManifest != null) {\n      const results = await this.loadWeights(weightsManifest);\n      [weightSpecs, weightData] = results;\n    }\n\n    const artifacts: ModelArtifacts = {\n      modelTopology,\n      weightSpecs,\n      weightData,\n      userDefinedMetadata,\n      generatedBy,\n      convertedBy,\n      format\n    };\n\n    const initializer = modelConfig.modelInitializer;\n    if (initializer) {\n      artifacts.modelInitializer = initializer;\n    }\n\n    return artifacts;\n  }\n\n  private async loadWeights(weightsManifest: WeightsManifestConfig):\n      Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n    const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n    const [prefix, suffix] = parseUrl(weightPath);\n    const pathPrefix = this.weightPathPrefix || prefix;\n\n    const weightSpecs = [];\n    for (const entry of weightsManifest) {\n      weightSpecs.push(...entry.weights);\n    }\n\n    const fetchURLs: string[] = [];\n    const urlPromises: Array<Promise<string>> = [];\n    for (const weightsGroup of weightsManifest) {\n      for (const path of weightsGroup.paths) {\n        if (this.weightUrlConverter != null) {\n          urlPromises.push(this.weightUrlConverter(path));\n        } else {\n          fetchURLs.push(pathPrefix + path + suffix);\n        }\n      }\n    }\n\n    if (this.weightUrlConverter) {\n      fetchURLs.push(...await Promise.all(urlPromises));\n    }\n\n    const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n      requestInit: this.requestInit,\n      fetchFunc: this.fetch,\n      onProgress: this.onProgress\n    });\n    return [weightSpecs, concatenateArrayBuffers(buffers)];\n  }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n  const lastSlash = url.lastIndexOf('/');\n  const lastSearchParam = url.lastIndexOf('?');\n  const prefix = url.substring(0, lastSlash);\n  const suffix =\n      lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n  return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n  return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n    (url: string, loadOptions?: LoadOptions) => {\n      if (typeof fetch === 'undefined' &&\n          (loadOptions == null || loadOptions.fetchFunc == null)) {\n        // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n        // an environment that is not the browser or node they have to setup a\n        // global fetch polyfill.\n        return null;\n      } else {\n        let isHTTP = true;\n        if (Array.isArray(url)) {\n          isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n        } else {\n          isHTTP = isHTTPScheme(url);\n        }\n        if (isHTTP) {\n          return http(url, loadOptions);\n        }\n      }\n      return null;\n    };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n *   - A JSON file consisting of `modelTopology` and `weightsManifest`.\n *   - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n *     'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n *   Can be an absolute HTTP path (e.g.,\n *   'http://localhost:8000/model-upload)') or a relative path (e.g.,\n *   './model-upload').\n * @param requestInit Request configurations to be used when sending\n *    HTTP request to server using `fetch`. It can contain fields such as\n *    `method`, `credentials`, `headers`, `mode`, etc. See\n *    https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n *    for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n *   following fields:\n *   - weightPathPrefix Optional, this specifies the path prefix for weight\n *     files, by default this is calculated from the path param.\n *   - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n *     the `fetch` from node-fetch can be used here.\n *   - onProgress Optional, progress callback function, fired periodically\n *     before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n  return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n    path: string, loadOptions?: LoadOptions): IOHandler {\n  return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n  constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n  async load(): Promise<ModelArtifacts> {\n    return this.modelArtifacts;\n  }\n}\n\nclass PassthroughSaver implements IOHandler {\n  constructor(\n      private readonly saveHandler:\n          (artifacts: ModelArtifacts) => Promise<SaveResult>) {}\n\n  async save(modelArtifacts: ModelArtifacts) {\n    return this.saveHandler(modelArtifacts);\n  }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n *     modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n *   the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n *   names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n *   concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n    modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n    weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n  if (arguments.length === 1) {\n    const isModelArtifacts =\n        (modelArtifacts as ModelArtifacts).modelTopology != null ||\n        (modelArtifacts as ModelArtifacts).weightSpecs != null;\n    if (isModelArtifacts) {\n      return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n    } else {\n      // Legacy support: with only modelTopology.\n      // TODO(cais): Remove this deprecated API.\n      console.warn(\n          'Please call tf.io.fromMemory() with only one argument. ' +\n          'The argument should be of type ModelArtifacts. ' +\n          'The multi-argument signature of tf.io.fromMemory() has been ' +\n          'deprecated and will be removed in a future release.');\n      return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n    }\n  } else {\n    // Legacy support.\n    // TODO(cais): Remove this deprecated API.\n    console.warn(\n        'Please call tf.io.fromMemory() with only one argument. ' +\n        'The argument should be of type ModelArtifacts. ' +\n        'The multi-argument signature of tf.io.fromMemory() has been ' +\n        'deprecated and will be removed in a future release.');\n    return new PassthroughLoader({\n      modelTopology: modelArtifacts as {},\n      weightSpecs,\n      weightData,\n      trainingConfig\n    });\n  }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n *   // ... do something with the artifacts ...\n *   return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n *     `SaveResult`.\n */\nexport function withSaveHandler(\n    saveHandler: (artifacts: ModelArtifacts) =>\n        Promise<SaveResult>): IOHandler {\n  return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n  browserFiles,\n  browserHTTPRequest,\n  concatenateArrayBuffers,\n  decodeWeights,\n  encodeWeights,\n  fromMemory,\n  getLoadHandlers,\n  getModelArtifactsInfoForJSON,\n  getSaveHandlers,\n  http,\n  IOHandler,\n  isHTTPScheme,\n  LoadHandler,\n  LoadOptions,\n  loadWeights,\n  ModelArtifacts,\n  ModelArtifactsInfo,\n  ModelJSON,\n  ModelStoreManager,\n  OnProgressCallback,\n  registerLoadRouter,\n  registerSaveRouter,\n  RequestDetails,\n  SaveConfig,\n  SaveHandler,\n  SaveResult,\n  WeightGroup,\n  weightsLoaderFactory,\n  WeightsManifestConfig,\n  WeightsManifestEntry,\n  withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  const $x = convertToTensor(x, 'x', 'reshape', null);\n\n  const inputs: ReshapeInputs = {x: $x};\n  const attrs: ReshapeAttrs = {shape};\n  const forward: ForwardFunc<\n      Tensor<R>> = (backend: KernelBackend, save: GradSaveFunc) => {\n    shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n    util.assert(\n        $x.size === util.sizeFromShape(shape),\n        () => 'new shape and old shape must have the same number of elements.');\n    save([$x]);\n    return backend.reshape($x, shape);\n  };\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n      attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print();  // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike, transposeA = false,\n    transposeB = false): T {\n  let $a = convertToTensor(a, 'a', 'matMul');\n  let $b = convertToTensor(b, 'b', 'matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n          `got ranks ${$a.rank} and ${$b.rank}.`);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n                           reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n                           reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([a3D, b3D]);\n\n    return backend.batchMatMul(\n        a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n  };\n\n  const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n  const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n      attrs as {} as NamedAttrMap);\n\n  return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n *     not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n    indices: Tensor|TensorLike, depth: number, onValue = 1,\n    offValue = 0): Tensor {\n  if (depth < 2) {\n    throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n  }\n  const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n  const outShape = [...$indices.shape, depth];\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$indices]);\n    return reshape(\n        backend.oneHot(\n            reshape($indices, [$indices.size]), depth, onValue, offValue),\n        outShape);\n  };\n\n  const inputs: OneHotInputs = {indices: $indices};\n  const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print();  // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_<T extends Tensor>(x: T|TensorLike, perm?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'transpose');\n\n  if (perm == null) {\n    perm = $x.shape.map((s, i) => i).reverse();\n  }\n  util.assert(\n      $x.rank === perm.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of perm ${perm}.`);\n  perm.forEach(axis => {\n    util.assert(\n        axis >= 0 && axis < $x.rank,\n        () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n            ` but got ${perm}`);\n  });\n\n  if ($x.rank <= 1) {\n    return $x.clone();\n  }\n\n  const inputs: TransposeInputs = {x: $x};\n  const attrs: TransposeAttrs = {perm};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n      null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * //  [0, 1, 1],\n * //  [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n *   for the classes. The shape is `[numExamples]`, where\n *   `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n *   0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n *   Its value must be larger than the largest element in `labels` and\n *   `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n *   row `r` and column `c` is the number of times examples of actual class\n *   `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n    labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n    numClasses: number): Tensor2D {\n  const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n  util.assert(\n      numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n      () => `If provided, numClasses must be a positive integer, ` +\n          `but got ${numClasses}`);\n  util.assert(\n      $labels.rank === 1,\n      () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n  util.assert(\n      $predictions.rank === 1,\n      () => `Expected the rank of predictions to be 1, ` +\n          `but got ${$predictions.rank}`);\n  util.assert(\n      $labels.shape[0] === $predictions.shape[0],\n      () => `Mismatch in the number of examples: ` +\n          `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n          `Labels and predictions should have the same number of elements.`);\n  util.assert(\n      numClasses > 0 && Number.isInteger(numClasses),\n      () => `numClasses is required to be a positive integer, but got ` +\n          `${numClasses}`);\n  // TODO(cais): In the future, if oneHot supports tensors inputs for\n  //   `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n  const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n  const oneHotPredictions =\n      oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n  const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n  return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided,  it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n    values: TensorLike3D, shape?: [number, number, number],\n    dtype?: DataType): Tensor3D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 3) {\n    throw new Error('tensor3d() requires shape to have three numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor3d() requires values to be number[][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor3d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement,\n    numChannels = 3): Tensor3D {\n  // Sanity checks.\n  if (numChannels > 4) {\n    throw new Error(\n        'Cannot construct Tensor with more than 4 channels from pixels.');\n  }\n  if (pixels == null) {\n    throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n  }\n  let isPixelData = false;\n  let isImageData = false;\n  let isVideo = false;\n  let isImage = false;\n  let isCanvasLike = false;\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    isPixelData = true;\n  } else if (\n      typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n    isImageData = true;\n  } else if (\n      typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement) {\n    isVideo = true;\n  } else if (\n      typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement) {\n    isImage = true;\n    // tslint:disable-next-line: no-any\n  } else if ((pixels as any).getContext != null) {\n    isCanvasLike = true;\n  } else {\n    throw new Error(\n        'pixels passed to tf.browser.fromPixels() must be either an ' +\n        `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n        `in browser, or OffscreenCanvas, ImageData in webworker` +\n        ` or {data: Uint32Array, width: number, height: number}, ` +\n        `but was ${(pixels as {}).constructor.name}`);\n  }\n  if (isVideo) {\n    const HAVE_CURRENT_DATA_READY_STATE = 2;\n    if (isVideo &&\n        (pixels as HTMLVideoElement).readyState <\n            HAVE_CURRENT_DATA_READY_STATE) {\n      throw new Error(\n          'The video element has not loaded data yet. Please wait for ' +\n          '`loadeddata` event on the <video> element.');\n    }\n  }\n  // If the current backend has 'FromPixels' registered, it has a more\n  // efficient way of handling pixel uploads, so we call that.\n  const kernel = getKernel(FromPixels, ENGINE.backendName);\n  if (kernel != null) {\n    const inputs: FromPixelsInputs = {pixels};\n    const attrs: FromPixelsAttrs = {numChannels};\n    return ENGINE.runKernel(\n               FromPixels, inputs as {} as NamedTensorMap,\n               attrs as {} as NamedAttrMap) as Tensor3D;\n  }\n\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n  let vals: Uint8ClampedArray|Uint8Array;\n\n  if (isCanvasLike) {\n    vals =\n        // tslint:disable-next-line:no-any\n        (pixels as any).getContext('2d').getImageData(0, 0, width, height).data;\n  } else if (isImageData || isPixelData) {\n    vals = (pixels as PixelData | ImageData).data;\n  } else if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement, 0, 0, width, height);\n    vals = fromPixels2DContext.getImageData(0, 0, width, height).data;\n  }\n  let values: Int32Array;\n  if (numChannels === 4) {\n    values = new Int32Array(vals);\n  } else {\n    const numPixels = width * height;\n    values = new Int32Array(numPixels * numChannels);\n    for (let i = 0; i < numPixels; i++) {\n      for (let channel = 0; channel < numChannels; ++channel) {\n        values[i * numChannels + channel] = vals[i * 4 + channel];\n      }\n    }\n  }\n  const outShape: [number, number, number] = [height, width, numChannels];\n  return tensor3d(values, outShape, 'int32');\n}\n\n/**\n * Draws a `tf.Tensor` of pixel values to a byte array or optionally a\n * canvas.\n *\n * When the dtype of the input is 'float32', we assume values in the range\n * [0-1]. Otherwise, when input is 'int32', we assume values in the range\n * [0-255].\n *\n * Returns a promise that resolves when the canvas has been drawn to.\n *\n * @param img A rank-2 or rank-3 tensor. If rank-2, draws grayscale. If\n *     rank-3, must have depth of 1, 3 or 4. When depth of 1, draws\n * grayscale. When depth of 3, we draw with the first three components of\n * the depth dimension corresponding to r, g, b and alpha = 1. When depth of\n * 4, all four components of the depth dimension correspond to r, g, b, a.\n * @param canvas The canvas to draw to.\n *\n * @doc {heading: 'Browser', namespace: 'browser'}\n */\nexport async function toPixels(\n    img: Tensor2D|Tensor3D|TensorLike,\n    canvas?: HTMLCanvasElement): Promise<Uint8ClampedArray> {\n  let $img = convertToTensor(img, 'img', 'toPixels');\n  if (!(img instanceof Tensor)) {\n    // Assume int32 if user passed a native array.\n    const originalImgTensor = $img;\n    $img = cast(originalImgTensor, 'int32');\n    originalImgTensor.dispose();\n  }\n  if ($img.rank !== 2 && $img.rank !== 3) {\n    throw new Error(\n        `toPixels only supports rank 2 or 3 tensors, got rank ${$img.rank}.`);\n  }\n  const [height, width] = $img.shape.slice(0, 2);\n  const depth = $img.rank === 2 ? 1 : $img.shape[2];\n\n  if (depth > 4 || depth === 2) {\n    throw new Error(\n        `toPixels only supports depth of size ` +\n        `1, 3 or 4 but got ${depth}`);\n  }\n\n  if ($img.dtype !== 'float32' && $img.dtype !== 'int32') {\n    throw new Error(\n        `Unsupported type for toPixels: ${$img.dtype}.` +\n        ` Please use float32 or int32 tensors.`);\n  }\n\n  const data = await $img.data();\n  const multiplier = $img.dtype === 'float32' ? 255 : 1;\n  const bytes = new Uint8ClampedArray(width * height * 4);\n\n  for (let i = 0; i < height * width; ++i) {\n    const rgba = [0, 0, 0, 255];\n\n    for (let d = 0; d < depth; d++) {\n      const value = data[i * depth + d];\n\n      if ($img.dtype === 'float32') {\n        if (value < 0 || value > 1) {\n          throw new Error(\n              `Tensor values for a float32 Tensor must be in the ` +\n              `range [0 - 1] but encountered ${value}.`);\n        }\n      } else if ($img.dtype === 'int32') {\n        if (value < 0 || value > 255) {\n          throw new Error(\n              `Tensor values for a int32 Tensor must be in the ` +\n              `range [0 - 255] but encountered ${value}.`);\n        }\n      }\n\n      if (depth === 1) {\n        rgba[0] = value * multiplier;\n        rgba[1] = value * multiplier;\n        rgba[2] = value * multiplier;\n      } else {\n        rgba[d] = value * multiplier;\n      }\n    }\n\n    const j = i * 4;\n    bytes[j + 0] = Math.round(rgba[0]);\n    bytes[j + 1] = Math.round(rgba[1]);\n    bytes[j + 2] = Math.round(rgba[2]);\n    bytes[j + 3] = Math.round(rgba[3]);\n  }\n\n  if (canvas != null) {\n    canvas.width = width;\n    canvas.height = height;\n    const ctx = canvas.getContext('2d');\n    const imageData = new ImageData(bytes, width, height);\n    ctx.putImageData(imageData, 0, 0);\n  }\n  if ($img !== img) {\n    $img.dispose();\n  }\n  return bytes;\n}\n\nexport const fromPixels = op({fromPixels_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\nimport {computeStrides} from '../util';\n\n/**\n * Validate gather nd inputs.\n *\n * @param tensor The tensor contains the source values.\n * @param indices The tensor contains the indices to slice the source.\n *\n * @returns [resultShape, numUpdates, sliceSize, strides]\n */\nexport function prepareAndValidate(\n    tensor: Tensor, indices: Tensor): [number[], number, number, number[]] {\n  if (tensor.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the input to be rank 1 or higher,' +\n        ` but the rank was ${tensor.rank}.`);\n  }\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(\n        'tf.gatherND() expects the indices to be int32 type,' +\n        ` but the dtype was ${indices.dtype}.`);\n  }\n  if (indices.shape[indices.rank - 1] > tensor.rank) {\n    throw new Error(\n        'index innermost dimension length must be <= tensor rank; saw: ' +\n        `${indices.shape[indices.rank - 1]} vs. ${tensor.rank}`);\n  }\n\n  if (tensor.size === 0) {\n    throw new Error(\n        'Requested more than 0 entries, but input is empty.' +\n        ` Input shape: ${tensor.shape}.`);\n  }\n\n  const indicesShape = indices.shape;\n  const sliceRank = indicesShape[indicesShape.length - 1];\n\n  // The result shape is\n  //   indices.shape[:-1] + params.shape[indices.shape[-1]:]\n  let nResult = 1;\n  for (let i = 0; i < indicesShape.length - 1; ++i) {\n    nResult *= indicesShape[i];\n  }\n\n  const inputShape = tensor.shape;\n\n  const resultShape = indicesShape.slice();\n  resultShape.pop();\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < tensor.rank; ++i) {\n    sliceSize *= inputShape[i];\n    resultShape.push(inputShape[i]);\n  }\n\n  const strides =\n      [...computeStrides(tensor.shape).map(stride => stride / sliceSize),\n       1].slice(0, sliceRank);\n\n  return [resultShape, nResult, sliceSize, strides];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {computeStrides, sizeFromShape} from '../util';\n\n/**\n * Check whether updates.shape = indices.shape[:batchDim] +\n * shape[sliceDim:]\n *\n * @param x The input tensor.\n */\nexport function validateUpdateShape(\n    shape: number[], indices: Tensor, updates: Tensor) {\n  const sliceDim = (indices.rank > 1) ? indices.shape[indices.rank - 1] : 1;\n  const batchDim = (indices.rank > 1) ? indices.rank - 1 : 1;\n\n  const shapeError = 'Must have updates.shape = indices.shape[:batchDim] + ' +\n      `shape[sliceDim:], got updates.shape: ${updates.shape}` +\n      `, indices.shape: ${indices.shape}, shape: ${shape}` +\n      `, sliceDim: ${sliceDim}, and batchDim: ${batchDim}.`;\n\n  if (updates.rank < batchDim) {\n    throw new Error(shapeError + ` update.rank < ${batchDim}. `);\n  }\n  if (shape.length < sliceDim + (updates.rank - batchDim)) {\n    throw new Error(\n        shapeError +\n        ` Output shape length < ${sliceDim + (updates.rank - batchDim)}`);\n  }\n  if (updates.rank !== batchDim + shape.length - sliceDim) {\n    throw new Error(\n        shapeError + ` update.rank != ${batchDim + shape.length - sliceDim}`);\n  }\n  for (let d = 0; d < batchDim; ++d) {\n    if (updates.shape[d] !== indices.shape[d]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d}] (${updates.shape[d]}) != indices.shape[${d}] (${\n              indices.shape[d]}).`);\n    }\n  }\n  for (let d = 0; d < updates.rank - batchDim; ++d) {\n    if (updates.shape[d + batchDim] !== shape[d + sliceDim]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d + batchDim}] (${\n              updates.shape[d + batchDim]}) != shape[${d + batchDim}] (${\n              shape[d + batchDim]})`);\n    }\n  }\n}\n\nexport interface ScatterShapeInfo {\n  sliceRank: number;\n  numUpdates: number;\n  sliceSize: number;\n  strides: number[];\n  outputSize: number;\n}\n/**\n * Validate scatter nd inputs.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n */\nexport function validateInput(\n    updates: Tensor, indices: Tensor, shape: number[]) {\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (updates.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the updates to be rank 1 or higher,' +\n        ` but the rank was ${updates.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${\n        indices.dtype}`);\n  }\n  if (shape.length < 1) {\n    throw new Error(\n        `Output rank must be greater or equal to 1, but got shape: ${shape}`);\n  }\n\n  if (shape.length === 0) {\n    if (indices.size === 0) {\n      throw new Error(`Indices specified for empty output. indices shape: ${\n          indices.shape}`);\n    }\n    if (updates.size === 0) {\n      throw new Error(`Updates specified for empty output. updates shape: ${\n          updates.shape}`);\n    }\n  }\n\n  validateUpdateShape(shape, indices, updates);\n}\n\n/**\n * Calculate the shape information for the output.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n *\n * @returns ScatterShapeInfo\n */\nexport function calculateShapes(\n    updates: TensorInfo, indices: TensorInfo,\n    shape: number[]): ScatterShapeInfo {\n  // Calculate the number of dimensions in indices\n  const indicesRank = indices.shape.length;\n  const sliceRank = (indicesRank > 1) ? indices.shape[indicesRank - 1] : 1;\n\n  // Calculate the number of elements that make up each slice of our updated\n  // tensor. This allows us to work with flattened tensors and copy over whole\n  // slices at a time.\n  const totalNd = shape.length;\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < totalNd; ++i) {\n    sliceSize *= shape[i];\n  }\n\n  const safeSliceDim = (sliceRank < 1) ? 1 : sliceRank;\n  const numUpdates = sizeFromShape(indices.shape) / safeSliceDim;\n\n  const strides = [...computeStrides(shape.slice(0, sliceRank)), 1];\n  const outputSize = sizeFromShape(shape);\n  return {sliceRank, numUpdates, sliceSize, strides, outputSize};\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '../kernel_registry';\nimport * as util from '../util';\n\nexport function assertParamsValid(\n    input: TensorInfo, begin: number[], size: number[]): void {\n  const inputRank = input.shape.length;\n  util.assert(\n      inputRank === begin.length,\n      () => `Error in slice${inputRank}D: Length of begin ${begin} must ` +\n          `match the rank of the array (${inputRank}).`);\n  util.assert(\n      inputRank === size.length,\n      () => `Error in slice${inputRank}D: Length of size ${size} must ` +\n          `match the rank of the array (${inputRank}).`);\n\n  for (let i = 0; i < inputRank; ++i) {\n    util.assert(\n        begin[i] + size[i] <= input.shape[i],\n        () => `Error in slice${inputRank}D: begin[${i}] + size[${i}] ` +\n            `(${begin[i] + size[i]}) would overflow input.shape[${i}] (${\n                  input.shape[i]})`);\n  }\n}\n\n/** Converts a binary mask to an array of axes. Used in stridedSlice(). */\nexport function maskToAxes(mask: number): number[] {\n  const axes = [];\n  let axis = 0;\n  while (mask > 0) {\n    if (mask & 1) {\n      axes.push(axis);\n    }\n    mask /= 2;\n    axis++;\n  }\n  return axes;\n}\n\n/** Computes the output shape given the strided slice params. */\nexport function computeOutShape(\n    begin: number[], end: number[], strides: number[]): number[] {\n  const size = [];\n  for (let axis = 0; axis < begin.length; axis++) {\n    size[axis] = Math.ceil((end[axis] - begin[axis]) / strides[axis]);\n  }\n  return size;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stride value. Otherwise, insert.\nexport function stridesWithElidedDims(\n    strides: number[], ellipsisInsertionIndex: number, numElidedAxes: number,\n    inputShape: number[]): number[] {\n  const newStrides = [...strides];\n  for (let i = newStrides.length; i < inputShape.length; i++) {\n    newStrides.push(1);\n  }\n  for (let i = 0; i < numElidedAxes; i++) {\n    if (i === 0) {\n      newStrides[ellipsisInsertionIndex] = 1;\n    } else {\n      newStrides.splice(\n          ellipsisInsertionIndex, 0 /* num elements to delete */,\n          1 /* element to add */);\n      newStrides.pop();\n    }\n  }\n  return newStrides;\n}\n\nfunction unnormalizeAxis(\n    ellipsisInsertionIndex: number, numElidedAxes: number,\n    normalizedAxis: number): number {\n  if (normalizedAxis <= ellipsisInsertionIndex) {\n    return normalizedAxis;\n  }\n\n  return normalizedAxis - (numElidedAxes - 1);\n}\n\nfunction getElidedAxes(numElidedAxes: number, ellipsisInsertionIndex: number) {\n  const elidedAxes = [];\n  for (let i = 0; i < numElidedAxes; i++) {\n    elidedAxes.push(ellipsisInsertionIndex + i);\n  }\n  return elidedAxes;\n}\n\n// Normalize the start, end and strides.\nexport function getNormalizedAxes(\n    inputShape: number[], ellipsisAxes: number[], numInterpolatedAxes: number,\n    begin: number[], end: number[], strides: number[], beginMask: number,\n    endMask: number,\n    ellipsisMask: number): {begin: number[], end: number[], strides: number[]} {\n  const inputRank = inputShape.length;\n  let normalizedBegin = new Array(inputRank),\n      normalizedEnd = new Array(inputRank),\n      normalizedStrides = new Array(inputRank);\n  if (ellipsisAxes.length && numInterpolatedAxes > 0) {\n    const fullIndex = ellipsisAxes[0];\n\n    // The ellipsis applies to the masked index as well as any dimensions\n    // that are interpolated.\n    const numElidedAxes = numInterpolatedAxes + 1;\n    normalizedBegin = startIndicesWithElidedDims(\n        beginMask, fullIndex, numElidedAxes, begin, inputShape);\n    normalizedEnd = stopIndicesWithElidedDims(\n        endMask, fullIndex, numElidedAxes, end, inputShape);\n    normalizedStrides =\n        stridesWithElidedDims(strides, fullIndex, numElidedAxes, inputShape);\n  } else {\n    for (let axis = 0; axis < inputRank; axis++) {\n      normalizedBegin[axis] = startForAxis(\n          beginMask, begin, strides, inputShape, axis, ellipsisMask);\n      normalizedEnd[axis] =\n          stopForAxis(endMask, end, strides, inputShape, axis, ellipsisMask);\n      normalizedStrides[axis] = stridesForAxis(strides, axis, ellipsisMask);\n    }\n  }\n\n  return {\n    begin: normalizedBegin,\n    end: normalizedEnd,\n    strides: normalizedStrides\n  };\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current start value. Otherwise, insert.\nexport function startIndicesWithElidedDims(\n    beginMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalBegin: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = 0;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalBegin[originalAxis];\n      if (beginMask & 1 << originalAxis) {\n        originalValue = 0;\n      }\n\n      newIndices[axis] = originalValue;\n    }\n  }\n  return newIndices;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stop value. Otherwise, insert.\nexport function stopIndicesWithElidedDims(\n    endMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalEnd: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = Number.MAX_SAFE_INTEGER;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalEnd[originalAxis];\n      if (endMask & 1 << originalAxis) {\n        originalValue = Number.MAX_SAFE_INTEGER;\n      }\n      newIndices[axis] = originalValue;\n    }\n  }\n\n  for (let i = 0; i < newIndices.length; i++) {\n    // Handle negative indices\n    const axisSize = inputShape[i];\n    if (newIndices[i] < 0) {\n      newIndices[i] += axisSize;\n    }\n    newIndices[i] = util.clamp(0, newIndices[i], inputShape[i]);\n  }\n  return newIndices;\n}\n\nexport function stridesForAxis(\n    strides: number[], axis: number, ellipsisMask: number): number {\n  let stride = strides[axis];\n  if (ellipsisMask & (1 << axis) || stride == null) {\n    stride = 1;\n  }\n\n  return stride;\n}\n\nexport function startForAxis(\n    beginMask: number, startIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let start = startIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or the begin index is not set\n  // for the axis.\n  if (beginMask & 1 << axis || ellipsisMask & 1 << axis || start == null) {\n    if (stride > 0) {\n      // Forward iteration - use the first element. These values will get\n      // clamped below (Note: We could have set them to 0 and axis_size-1, but\n      // use lowest() and max() to maintain symmetry with StopForAxis())\n      start = Number.MIN_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the last element.\n      start = Number.MAX_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (start < 0) {\n    start += axisSize;\n  }\n\n  // Clamping\n  start = util.clamp(0, start, axisSize - 1);\n\n  return start;\n}\n\nexport function stopForAxis(\n    endMask: number, stopIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let stop = stopIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or if the stop index is not\n  // set for this axis.\n  if (endMask & (1 << axis) || ellipsisMask & (1 << axis) || stop == null) {\n    if (stride > 0) {\n      // Forward iteration - use the last element. These values will get\n      // clamped below\n      stop = Number.MAX_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the first element.\n      stop = Number.MIN_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (stop < 0) {\n    stop += axisSize;\n  }\n\n  // Clamping\n  // Because the end index points one past the last element, we need slightly\n  // different clamping ranges depending on the direction.\n  if (stride > 0) {\n    // Forward iteration\n    stop = util.clamp(0, stop, axisSize);\n  } else {\n    // Backward iteration\n    stop = util.clamp(-1, stop, axisSize - 1);\n  }\n\n  return stop;\n}\n\n/**\n * Returns true if the slice occupies a continous set of elements in the\n * 'flat' space.\n */\nexport function isSliceContinous(\n    shape: number[], begin: number[], size: number[]) {\n  // Index of the first axis that has size > 1.\n  let firstNonOneAxis = size.length;\n  for (let i = 0; i < size.length; i++) {\n    if (size[i] > 1) {\n      firstNonOneAxis = i;\n      break;\n    }\n  }\n\n  for (let i = firstNonOneAxis + 1; i < size.length; i++) {\n    if (begin[i] > 0 || size[i] !== shape[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function computeFlatOffset(begin: number[], strides: number[]): number {\n  let flatOffset = begin.length > 0 ? begin[begin.length - 1] : 1;\n  for (let i = 0; i < begin.length - 1; i++) {\n    flatOffset += begin[i] * strides[i];\n  }\n  return flatOffset;\n}\n\nexport function parseSliceParams(\n    x: TensorInfo, begin: number|number[], size?: number|number[]) {\n  // The following logic allows for more ergonomic calls.\n  let begin_: number[];\n  const xRank = x.shape.length;\n  if (typeof begin === 'number') {\n    begin_ = [begin, ...new Array(xRank - 1).fill(0)];\n  } else if (begin.length < xRank) {\n    begin_ = begin.concat(new Array(xRank - begin.length).fill(0));\n  } else {\n    begin_ = begin.slice();\n  }\n  begin_.forEach(d => {\n    util.assert(\n        d !== -1, () => 'slice() does not support negative begin indexing.');\n  });\n  let size_: number[];\n  if (size == null) {\n    size_ = new Array(xRank).fill(-1);\n  } else if (typeof size === 'number') {\n    size_ = [size, ...new Array(xRank - 1).fill(-1)];\n  } else if (size.length < xRank) {\n    size_ = size.concat(new Array(xRank - size.length).fill(-1));\n  } else {\n    size_ = size;\n  }\n  size_ = size_.map((d, i) => {\n    if (d >= 0) {\n      return d;\n    } else {\n      util.assert(\n          d === -1,\n          () => `Negative size values should be exactly -1 but got ` +\n              `${d} for the slice() size at index ${i}.`);\n      return x.shape[i] - begin_[i];\n    }\n  });\n  return [begin_, size_];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from './util';\n\n/**\n * Types to support JSON-esque data structures internally.\n *\n * Internally ConfigDict's use camelCase keys and values where the\n * values are class names to be instantiated.  On the python side, these\n * will be snake_case.  Internally we allow Enums into the values for better\n * type safety, but these need to be converted to raw primitives (usually\n * strings) for round-tripping with python.\n *\n * toConfig returns the TS-friendly representation. model.toJSON() returns\n * the pythonic version as that's the portable format.  If you need to\n * python-ify a non-model level toConfig output, you'll need to use a\n * convertTsToPythonic from serialization_utils in -Layers.\n *\n */\nexport declare type ConfigDictValue =\n    boolean | number | string | null | ConfigDictArray | ConfigDict;\nexport declare interface ConfigDict {\n  [key: string]: ConfigDictValue;\n}\nexport declare interface ConfigDictArray extends Array<ConfigDictValue> {}\n\n/**\n * Type to represent the class-type of Serializable objects.\n *\n * Ie the class prototype with access to the constructor and any\n * static members/methods. Instance methods are not listed here.\n *\n * Source for this idea: https://stackoverflow.com/a/43607255\n */\nexport declare type SerializableConstructor<T extends Serializable> = {\n  // tslint:disable-next-line:no-any\n  new (...args: any[]): T; className: string; fromConfig: FromConfigMethod<T>;\n};\nexport declare type FromConfigMethod<T extends Serializable> =\n    (cls: SerializableConstructor<T>, config: ConfigDict) => T;\n\n/**\n * Serializable defines the serialization contract.\n *\n * TFJS requires serializable classes to return their className when asked\n * to avoid issues with minification.\n */\nexport abstract class Serializable {\n  /**\n   * Return the class name for this class to use in serialization contexts.\n   *\n   * Generally speaking this will be the same thing that constructor.name\n   * would have returned.  However, the class name needs to be robust\n   * against minification for serialization/deserialization to work properly.\n   *\n   * There's also places such as initializers.VarianceScaling, where\n   * implementation details between different languages led to different\n   * class hierarchies and a non-leaf node is used for serialization purposes.\n   */\n  getClassName(): string {\n    return (this.constructor as SerializableConstructor<Serializable>)\n        .className;\n  }\n\n  /**\n   * Return all the non-weight state needed to serialize this object.\n   */\n  abstract getConfig(): ConfigDict;\n\n  /**\n   * Creates an instance of T from a ConfigDict.\n   *\n   * This works for most descendants of serializable.  A few need to\n   * provide special handling.\n   * @param cls A Constructor for the class to instantiate.\n   * @param config The Configuration for the object.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config);\n  }\n}\n\n/**\n * Maps string keys to class constructors.\n *\n * Used during (de)serialization from the cross-language JSON format, which\n * requires the class name in the serialization format matches the class\n * names as used in Python, should it exist.\n */\nexport class SerializationMap {\n  private static instance: SerializationMap;\n  classNameMap: {\n    [className: string]:\n        [SerializableConstructor<Serializable>, FromConfigMethod<Serializable>]\n  };\n\n  private constructor() {\n    this.classNameMap = {};\n  }\n\n  /**\n   * Returns the singleton instance of the map.\n   */\n  static getMap(): SerializationMap {\n    if (SerializationMap.instance == null) {\n      SerializationMap.instance = new SerializationMap();\n    }\n    return SerializationMap.instance;\n  }\n\n  /**\n   * Registers the class as serializable.\n   */\n  static register<T extends Serializable>(cls: SerializableConstructor<T>) {\n    SerializationMap.getMap().classNameMap[cls.className] =\n        [cls, cls.fromConfig];\n  }\n}\n\n/**\n * Register a class with the serialization map of TensorFlow.js.\n *\n * This is often used for registering custom Layers, so they can be\n * serialized and deserialized.\n *\n * Example:\n *\n * ```js\n * class MyCustomLayer extends tf.layers.Layer {\n *   static className = 'MyCustomLayer';\n *\n *   constructor(config) {\n *     super(config);\n *   }\n * }\n * tf.serialization.registerClass(MyCustomLayer);\n * ```\n *\n * @param cls The class to be registered. It must have a public static member\n *   called `className` defined and the value must be a non-empty string.\n *\n * @doc {heading: 'Models', subheading: 'Serialization', ignoreCI: true}\n */\nexport function registerClass<T extends Serializable>(\n    cls: SerializableConstructor<T>) {\n  assert(\n      cls.className != null,\n      () => `Class being registered does not have the static className ` +\n          `property defined.`);\n  assert(\n      typeof cls.className === 'string',\n      () => `className is required to be a string, but got type ` +\n          typeof cls.className);\n  assert(\n      cls.className.length > 0,\n      () => `Class being registered has an empty-string as its className, ` +\n          `which is disallowed.`);\n\n  SerializationMap.register(cls);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {inferShape} from './tensor_util_env';\nimport {RecursiveArray, TensorLike, TypedArray} from './types';\nimport {arraysEqual, flatten, isString, isTypedArray} from './util';\n\nconst TEST_EPSILON_FLOAT32 = 1e-3;\nexport const TEST_EPSILON_FLOAT16 = 1e-1;\n\nexport function expectArraysClose(\n    actual: TypedArray|number|RecursiveArray<number>,\n    expected: TypedArray|number|RecursiveArray<number>, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, epsilon));\n}\n\nexport function testEpsilon() {\n  return ENGINE.backend.floatPrecision() === 32 ? TEST_EPSILON_FLOAT32 :\n                                                  TEST_EPSILON_FLOAT16;\n}\n\nfunction expectArraysPredicate(\n    actual: TensorLike, expected: TensorLike,\n    predicate: (a: {}, b: {}) => boolean) {\n  let checkClassType = true;\n  if (isTypedArray(actual) || isTypedArray(expected)) {\n    checkClassType = false;\n  }\n  if (isTypedArray(actual) && isTypedArray(expected)) {\n    checkClassType = true;\n  }\n  if (checkClassType) {\n    const aType = actual.constructor.name;\n    const bType = expected.constructor.name;\n\n    if (aType !== bType) {\n      throw new Error(\n          `Arrays are of different type. Actual: ${aType}. ` +\n          `Expected: ${bType}`);\n    }\n  }\n\n  if (Array.isArray(actual) && Array.isArray(expected)) {\n    const actualShape = inferShape(actual);\n    const expectedShape = inferShape(expected);\n    if (!arraysEqual(actualShape, expectedShape)) {\n      throw new Error(\n          `Arrays have different shapes. ` +\n          `Actual: [${actualShape}]. Expected: [${expectedShape}]`);\n    }\n  }\n\n  const actualFlat =\n      isTypedArray(actual) ? actual : flatten(actual as RecursiveArray<number>);\n  const expectedFlat = isTypedArray(expected) ?\n      expected :\n      flatten(expected as RecursiveArray<number>);\n\n  if (actualFlat.length !== expectedFlat.length) {\n    throw new Error(\n        `Arrays have different lengths actual: ${actualFlat.length} vs ` +\n        `expected: ${expectedFlat.length}.\\n` +\n        `Actual:   ${actualFlat}.\\n` +\n        `Expected: ${expectedFlat}.`);\n  }\n  for (let i = 0; i < expectedFlat.length; ++i) {\n    const a = actualFlat[i];\n    const e = expectedFlat[i];\n\n    if (!predicate(a, e)) {\n      throw new Error(\n          `Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${e}.\\n` +\n          `Actual:   ${actualFlat}.\\n` +\n          `Expected: ${expectedFlat}.`);\n    }\n  }\n}\n\nexport interface DoneFn {\n  (): void;\n  fail: (message?: Error|string) => void;\n}\n\nexport function expectPromiseToFail(fn: () => Promise<{}>, done: DoneFn): void {\n  fn().then(() => done.fail(), () => done());\n}\n\nexport function expectArraysEqual(actual: TensorLike, expected: TensorLike) {\n  const exp = typeof expected === 'string' || typeof expected === 'number' ||\n          typeof expected === 'boolean' ?\n      [expected] as number[] :\n      expected as number[];\n  if (isString(actual) || isString((actual as string[])[0]) ||\n      isString(expected) || isString((expected as string[])[0])) {\n    // tslint:disable-next-line: triple-equals\n    return expectArraysPredicate(actual, exp, (a, b) => a == b);\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, 0));\n}\n\nexport function expectNumbersClose(a: number, e: number, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  if (!areClose(a, e, epsilon)) {\n    throw new Error(`Numbers differ: actual === ${a}, expected === ${e}`);\n  }\n}\n\nfunction areClose(a: number, e: number, epsilon: number): boolean {\n  if (!isFinite(a) && !isFinite(e)) {\n    return true;\n  }\n  if (isNaN(a) || isNaN(e) || Math.abs(a - e) > epsilon) {\n    return false;\n  }\n  return true;\n}\n\nexport function expectValuesInRange(\n    actual: TypedArray|number[], low: number, high: number) {\n  for (let i = 0; i < actual.length; i++) {\n    if (actual[i] < low || actual[i] > high) {\n      throw new Error(\n          `Value out of range:${actual[i]} low: ${low}, high: ${high}`);\n    }\n  }\n}\n\nexport function expectArrayBuffersEqual(\n    actual: ArrayBuffer, expected: ArrayBuffer) {\n  // Safari & Jasmine don't like comparing ArrayBuffers directly. Wrapping in\n  // a Float32Array solves this issue.\n  expect(new Float32Array(actual)).toEqual(new Float32Array(expected));\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from './backends/backend';\nimport {ENGINE, Engine, MemoryInfo, ProfileInfo, ScopeFn, TimingInfo} from './engine';\nimport {env} from './environment';\n\nimport {Platform} from './platforms/platform';\nimport {setDeprecationWarningFn, Tensor} from './tensor';\nimport {TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\n\n/**\n * Enables production mode which disables correctness checks in favor of\n * performance.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableProdMode(): void {\n  env().set('PROD', true);\n}\n\n/**\n * Enables debug mode which will log information about all executed kernels:\n * the elapsed time of the kernel execution, as well as the rank, shape, and\n * size of the output tensor.\n *\n * Debug mode will significantly slow down your application as it will\n * download the result of every operation to the CPU. This should not be used in\n * production. Debug mode does not affect the timing information of the kernel\n * execution as we do not measure download time in the kernel execution time.\n *\n * See also: `tf.profile`, `tf.memory`.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableDebugMode(): void {\n  env().set('DEBUG', true);\n}\n\n/** Globally disables deprecation warnings */\nexport function disableDeprecationWarnings(): void {\n  env().set('DEPRECATION_WARNINGS_ENABLED', false);\n  console.warn(`TensorFlow.js deprecation warnings have been disabled.`);\n}\n\n/** Warn users about deprecated functionality. */\nexport function deprecationWarn(msg: string) {\n  if (env().getBool('DEPRECATION_WARNINGS_ENABLED')) {\n    console.warn(\n        msg + ' You can disable deprecation warnings with ' +\n        'tf.disableDeprecationWarnings().');\n  }\n}\nsetDeprecationWarningFn(deprecationWarn);\n\n/**\n * Dispose all variables kept in backend engine.\n *\n * @doc {heading: 'Environment'}\n */\nexport function disposeVariables(): void {\n  ENGINE.disposeVariables();\n}\n\n/**\n * It returns the global engine that keeps track of all tensors and backends.\n *\n * @doc {heading: 'Environment'}\n */\nexport function engine(): Engine {\n  return ENGINE;\n}\n\n/**\n * Returns memory info at the current time in the program. The result is an\n * object with the following properties:\n *\n * - `numBytes`: Number of bytes allocated (undisposed) at this time.\n * - `numTensors`: Number of unique tensors allocated.\n * - `numDataBuffers`: Number of unique data buffers allocated\n *   (undisposed) at this time, which is \u2264 the number of tensors\n *   (e.g. `a.reshape(newShape)` makes a new Tensor that shares the same\n *   data buffer with `a`).\n * - `unreliable`: True if the memory usage is unreliable. See `reasons` when\n *    `unreliable` is true.\n * - `reasons`: `string[]`, reasons why the memory is unreliable, present if\n *    `unreliable` is true.\n *\n * WebGL Properties:\n * - `numBytesInGPU`: Number of bytes allocated (undisposed) in the GPU only at\n *     this time.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function memory(): MemoryInfo {\n  return ENGINE.memory();\n}\n\n/**\n * Executes the provided function `f()` and returns a promise that resolves\n * with information about the function's memory use:\n * - `newBytes`: the number of new bytes allocated\n * - `newTensors`: the number of new tensors created\n * - `peakBytes`: the peak number of bytes allocated\n * - `kernels`: an array of objects for each kernel involved that reports\n * their input and output shapes, number of bytes used, and number of new\n * tensors created.\n *\n * ```js\n * const profile = await tf.profile(() => {\n *   const x = tf.tensor1d([1, 2, 3]);\n *   let x2 = x.square();\n *   x2.dispose();\n *   x2 = x.square();\n *   x2.dispose();\n *   return x;\n * });\n *\n * console.log(`newBytes: ${profile.newBytes}`);\n * console.log(`newTensors: ${profile.newTensors}`);\n * console.log(`byte usage over all kernels: ${profile.kernels.map(k =>\n * k.totalBytesSnapshot)}`);\n * ```\n *\n *\n * @doc {heading: 'Performance', subheading: 'Profile'}\n */\nexport function profile(f: () => (TensorContainer | Promise<TensorContainer>)):\n    Promise<ProfileInfo> {\n  return ENGINE.profile(f);\n}\n\n/**\n * Executes the provided function `fn` and after it is executed, cleans up all\n * intermediate tensors allocated by `fn` except those returned by `fn`.\n * `fn` must not return a Promise (async functions not allowed). The returned\n * result can be a complex object.\n *\n * Using this method helps avoid memory leaks. In general, wrap calls to\n * operations in `tf.tidy` for automatic memory cleanup.\n *\n * NOTE: Variables do *not* get cleaned up when inside a tidy(). If you want to\n * dispose variables, please use `tf.disposeVariables` or call dispose()\n * directly on variables.\n *\n * ```js\n * // y = 2 ^ 2 + 1\n * const y = tf.tidy(() => {\n *   // a, b, and one will be cleaned up when the tidy ends.\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *   const b = a.square();\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * y.print();\n * ```\n *\n * @param nameOrFn The name of the closure, or the function to execute.\n *     If a name is provided, the 2nd argument should be the function.\n *     If debug mode is on, the timing and the memory usage of the function\n *     will be tracked and displayed on the console using the provided name.\n * @param fn The function to execute.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function tidy<T extends TensorContainer>(\n    nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>): T {\n  return ENGINE.tidy(nameOrFn, fn);\n}\n\n/**\n * Disposes any `tf.Tensor`s found within the provided object.\n *\n * @param container an object that may be a `tf.Tensor` or may directly\n *     contain `tf.Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. If\n *     the object is not a `tf.Tensor` or does not contain `Tensors`, nothing\n *     happens. In general it is safe to pass any object here, except that\n *     `Promise`s are not supported.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function dispose(container: TensorContainer) {\n  const tensors = getTensorsInContainer(container);\n  tensors.forEach(tensor => tensor.dispose());\n}\n\n/**\n * Keeps a `tf.Tensor` generated inside a `tf.tidy` from being disposed\n * automatically.\n *\n * ```js\n * let b;\n * const y = tf.tidy(() => {\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *\n *   // b will not be cleaned up by the tidy. a and one will be cleaned up\n *   // when the tidy ends.\n *   b = tf.keep(a.square());\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * console.log('y:');\n * y.print();\n * console.log('b:');\n * b.print();\n * ```\n *\n * @param result The tensor to keep from being disposed.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function keep<T extends Tensor>(result: T): T {\n  return ENGINE.keep(result);\n}\n\n/**\n * Executes `f()` and returns a promise that resolves with timing\n * information.\n *\n * The result is an object with the following properties:\n *\n * - `wallMs`: Wall execution time.\n * - `kernelMs`: Kernel execution time, ignoring data transfer. If using the\n * WebGL backend and the query timer extension is not available, this will\n * return an error object.\n * - On `WebGL` The following additional properties exist:\n *   - `uploadWaitMs`: CPU blocking time on texture uploads.\n *   - `downloadWaitMs`: CPU blocking time on texture downloads (readPixels).\n *\n * ```js\n * const x = tf.randomNormal([20, 20]);\n * const time = await tf.time(() => x.matMul(x));\n *\n * console.log(`kernelMs: ${time.kernelMs}, wallTimeMs: ${time.wallMs}`);\n * ```\n *\n * @param f The function to execute and time.\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nexport function time(f: () => void): Promise<TimingInfo> {\n  return ENGINE.time(f);\n}\n\n/**\n * Sets the backend (cpu, webgl, wasm, etc) responsible for creating tensors and\n * executing operations on those tensors. Returns a promise that resolves\n * to a boolean if the backend initialization was successful.\n *\n * Note this disposes the current backend, if any, as well as any tensors\n * associated with it. A new backend is initialized, even if it is of the\n * same type as the previous one.\n *\n * @param backendName The name of the backend. Currently supports\n *     `'webgl'|'cpu'` in the browser, `'tensorflow'` under node.js\n *     (requires tfjs-node), and `'wasm'` (requires tfjs-backend-wasm).\n *\n * @doc {heading: 'Backends'}\n */\nexport function setBackend(backendName: string): Promise<boolean> {\n  return ENGINE.setBackend(backendName);\n}\n\n/**\n * Returns a promise that resolves when the currently selected backend (or the\n * highest priority one) has initialized. Await this promise when you are using\n * a backend that has async initialization.\n *\n * @doc {heading: 'Backends'}\n */\nexport function ready(): Promise<void> {\n  return ENGINE.ready();\n}\n\n/**\n * Returns the current backend name (cpu, webgl, etc). The backend is\n * responsible for creating tensors and executing operations on those tensors.\n *\n * @doc {heading: 'Backends'}\n */\nexport function getBackend(): string {\n  return ENGINE.backendName;\n}\n\n/**\n * Removes a backend and the registered factory.\n *\n * @doc {heading: 'Backends'}\n */\nexport function removeBackend(name: string): void {\n  ENGINE.removeBackend(name);\n}\n\n/**\n * Finds the backend registered under the provided name. Returns null if the\n * name is not in the registry, or the registration hasn't finished yet.\n */\nexport function findBackend(name: string): KernelBackend {\n  return ENGINE.findBackend(name);\n}\n\n/**\n * Finds the backend factory registered under the provided name. Returns a\n * function that produces a new backend when called. Returns null if the name\n * is not in the registry.\n */\nexport function findBackendFactory(name: string): () =>\n    KernelBackend | Promise<KernelBackend> {\n  return ENGINE.findBackendFactory(name);\n}\n\n/**\n * Registers a global backend. The registration should happen when importing\n * a module file (e.g. when importing `backend_webgl.ts`), and is used for\n * modular builds (e.g. custom tfjs bundle with only webgl support).\n *\n * @param factory The backend factory function. When called, it should\n * return a backend instance, or a promise of an instance.\n * @param priority The priority of the backend (higher = more important).\n *     In case multiple backends are registered, the priority is used to find\n *     the best backend. Defaults to 1.\n * @return False if there is already a registered backend under this name, true\n *     if not.\n *\n * @doc {heading: 'Backends'}\n */\nexport function registerBackend(\n    name: string, factory: () => KernelBackend | Promise<KernelBackend>,\n    priority = 1): boolean {\n  return ENGINE.registerBackend(name, factory, priority);\n}\n\n/**\n * Gets the current backend. If no backends have been initialized, this will\n * attempt to initialize the best backend. Will throw an error if the highest\n * priority backend has async initialization, in which case, you should call\n * 'await tf.ready()' before running other code.\n *\n * @doc {heading: 'Backends'}\n */\nexport function backend(): KernelBackend {\n  return ENGINE.backend;\n}\n\n/**\n * Sets the global platform.\n *\n * @param platformName The name of this platform.\n * @param platform A platform implementation.\n */\nexport function setPlatform(platformName: string, platform: Platform) {\n  env().setPlatform(platformName, platform);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Add, AddInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n *\n * ```js\n * // Broadcast add a with b.\n * const a = tf.scalar(5);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n * @param a The first `tf.Tensor` to add.\n * @param b The second `tf.Tensor` to add. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction add_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'add');\n  let $b = convertToTensor(b, 'b', 'add');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.add($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: AddInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Add) as T;\n}\n\nexport const add = op({add_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FloorDiv, FloorDivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n * The result is rounded with floor function.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.floorDiv(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.floorDiv(b).print();  // or tf.floorDiv(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction floorDiv_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'floorDiv');\n  let $b = convertToTensor(b, 'b', 'floorDiv');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.floorDiv($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: FloorDivInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             FloorDiv) as T;\n}\n\nexport const floorDiv = op({floorDiv_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Div, DivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {floorDiv} from './floorDiv';\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction div_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'int32' && $b.dtype === 'int32') {\n    return floorDiv($a, $b);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.realDivide($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: DivInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Div,\n             attrs) as T;\n}\n\nexport const div = op({div_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Multiply, MultiplyInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Multiplies two `tf.Tensor`s element-wise, A * B. Supports broadcasting.\n *\n * We also expose `tf.mulStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([2, 3, 4, 5]);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n *\n * ```js\n * // Broadcast mul a with b.\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.scalar(5);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n * @param a The first tensor to multiply.\n * @param b The second tensor to multiply. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mul_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mul');\n  let $b = convertToTensor(b, 'b', 'mul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.multiply($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MultiplyInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Multiply) as T;\n}\nexport const mul = op({mul_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Abs, AbsInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes absolute value element-wise: `abs(x)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.abs().print();  // or tf.abs(x)\n * ```\n * @param x The input `tf.Tensor`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction abs_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'abs');\n\n  const inputs: AbsInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    if ($x.dtype === 'complex64') {\n      return backend.complexAbs($x);\n    }\n\n    return backend.abs($x);\n  }, inputs as {} as NamedTensorMap, null /* grad */, Abs);\n}\n\nexport const abs = op({abs_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Acos, AcosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes acos of the input `tf.Tensor` element-wise: `acos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.acos().print();  // or tf.acos(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acos');\n  const inputs: AcosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acos);\n}\nexport const acos = op({acos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Acosh, AcoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the inverse hyperbolic cos of the input `tf.Tensor` element-wise:\n * `acosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([10, 1, 3, 5.7]);\n *\n * x.acosh().print();  // or tf.acosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acosh');\n  const inputs: AcoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acosh);\n}\nexport const acosh = op({acosh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AddN, AddNInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Adds a list of `tf.Tensor`s element-wise, each with the same shape and dtype.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n *\n * tf.addN([a, b, c]).print();\n * ```\n * @param tensors A list of tensors with the same shape and dtype.\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction addN_<T extends Tensor>(tensors: Array<T|TensorLike>): T {\n  util.assert(\n      Array.isArray(tensors),\n      () => 'The argument passed to tf.addN() must be a list of tensors');\n  util.assert(\n      tensors.length >= 1,\n      () => `Must pass at least one tensor to tf.addN(), but got ` +\n          `${tensors.length}`);\n\n  const $tensors =\n      tensors.map((t, i) => convertToTensor(t, `tensors${i}`, 'addN'));\n\n  const firstTensor = $tensors[0];\n  $tensors.forEach(t => {\n    if (t.dtype !== firstTensor.dtype) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same dtype');\n    }\n  });\n\n  $tensors.forEach(t => {\n    if (!util.arraysEqual(t.shape, firstTensor.shape)) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same shape');\n    }\n  });\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.addN($tensors);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: AddNInputs = $tensors;\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, AddN) as\n      T;\n}\n\nexport const addN = op({addN_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\n/**\n * Returns true if the axis specifies the inner most dimensions of the\n * array.\n */\nexport function axesAreInnerMostDims(axes: number[], rank: number): boolean {\n  for (let i = 0; i < axes.length; ++i) {\n    if (axes[axes.length - i - 1] !== rank - 1 - i) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function combineLocations(\n    outputLoc: number[], reduceLoc: number[], axes: number[]): number[] {\n  const rank = outputLoc.length + reduceLoc.length;\n  const loc = [];\n  let outIdx = 0;\n  let reduceIdx = 0;\n  \u00A0 for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      loc.push(outputLoc[outIdx++]);\n    } else {\n      loc.push(reduceLoc[reduceIdx++]);\n    }\n  }\n  return loc;\n}\n\nexport function computeOutAndReduceShapes(\n    aShape: number[], axes: number[]): [number[], number[]] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      outShape.push(aShape[dim]);\n    }\n  }\n  const reduceShape = axes.map(dim => aShape[dim]);\n  return [outShape, reduceShape];\n}\n\nexport function expandShapeToKeepDim(\n    shape: number[], axes: number[]): number[] {\n  const reduceSubShape = axes.map(x => 1);\n  return combineLocations(shape, reduceSubShape, axes);\n}\n\nexport function assertAxesAreInnerMostDims(\n    msg: string, axes: number[], rank: number): void {\n  util.assert(\n      axesAreInnerMostDims(axes, rank),\n      () => `${msg} supports only inner-most axes for now. ` +\n          `Got axes ${axes} and rank-${rank} input.`);\n}\n\n/**\n * Returns the axes permutation to be used with `tf.transpose`, if such\n * permutation is necessary. Otherwise it returns null. This method is used by\n * operations that operate only on inner-most axes.\n */\nexport function getAxesPermutation(axes: number[], rank: number): number[]|\n    null {\n  if (axesAreInnerMostDims(axes, rank)) {\n    return null;\n  }\n  const result: number[] = [];\n  for (let i = 0; i < rank; ++i) {\n    if (axes.indexOf(i) === -1) {\n      result.push(i);\n    }\n  }\n  axes.forEach(axis => result.push(axis));\n  return result;\n}\n\n/** Returns the axes permutation that undoes the original permutation. */\nexport function getUndoAxesPermutation(axes: number[]): number[] {\n  return axes.map((axis, i) => [i, axis])\n      .sort((a, b) => a[1] - b[1])\n      .map(x => x[0]);\n}\n\nexport function getInnerMostAxes(numAxes: number, rank: number): number[] {\n  const res: number[] = [];\n  for (let i = rank - numAxes; i < rank; ++i) {\n    res.push(i);\n  }\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {All, AllAttrs, AllInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical and of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.all().print();  // or tf.all(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.all(axis).print();  // or tf.all(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction all_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'all', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.all($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AllInputs = {x: $x};\n  const attrs: AllAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, All,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const all = op({all_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Any, AnyAttrs, AnyInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical or of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.any().print();  // or tf.any(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.any(axis).print();  // or tf.any(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction any_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'any', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.any($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AnyInputs = {x: $x};\n  const attrs: AnyAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Any,\n             attrs as {} as NamedAttrMap) as T;\n}\n\n// tslint:disable-next-line:variable-name\nexport const any = op({any_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMax, ArgMaxAttrs, ArgMaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the maximum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMax().print();  // or tf.argMax(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMax(axis).print();  // or tf.argMax(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMax_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMax');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMax($x, axes[0]);\n  };\n\n  const inputs: ArgMaxInputs = {x: $x};\n  const attrs: ArgMaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMax,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMax = op({argMax_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMin, ArgMinAttrs, ArgMinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the minimum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMin().print();  // or tf.argMin(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMin(axis).print();  // or tf.argMin(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMin_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMin');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if (axis == null) {\n      axis = 0;\n    }\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMin($x, axes[0]);\n  };\n\n  const inputs: ArgMinInputs = {x: $x};\n  const attrs: ArgMinAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMin,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMin = op({argMin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asin, AsinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes asin of the input `tf.Tensor` element-wise: `asin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asin().print();  // or tf.asin(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asin');\n  const inputs: AsinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asin);\n}\nexport const asin = op({asin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asinh, AsinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic sin of the input `tf.Tensor` element-wise:\n * `asinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asinh().print();  // or tf.asinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asinh');\n\n  const inputs: AsinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asinh);\n}\nexport const asinh = op({asinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atan, AtanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes atan of the input `tf.Tensor` element-wise: `atan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.atan().print();  // or tf.atan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atan');\n\n  const inputs: AtanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atan);\n}\nexport const atan = op({atan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Atan2, Atan2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes arctangent of `tf.Tensor`s a / b element-wise: `atan2(a, b)`.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1.0, 1.0, -1.0, .7]);\n * const b = tf.tensor1d([2.0, 13.0, 3.5, .21]);\n *\n * tf.atan2(a, b).print()\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan2_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'atan2');\n  let $b = convertToTensor(b, 'b', 'atan2');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.atan2($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: Atan2Inputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Atan2) as T;\n}\n\nexport const atan2 = op({atan2_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atanh, AtanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic tan of the input `tf.Tensor` element-wise:\n * `atanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.atanh().print();  // or tf.atanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atanh');\n\n  const inputs: AtanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atanh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atanh);\n}\nexport const atanh = op({atanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\ntype PadType = 'SAME'|'VALID'|'NUMBER'|'EXPLICIT';\n\n// For NHWC should be in the following form:\n//  [[0, 0], [pad_top,pad_bottom], [pad_left, pad_right], [0, 0]]\n// For NCHW should be in the following form:\n//  [[0, 0], [0, 0], [pad_top,pad_bottom], [pad_left, pad_right]]\n// Reference: https://www.tensorflow.org/api_docs/python/tf/nn/conv2d\nexport type ExplicitPadding =\n    [[number, number], [number, number], [number, number], [number, number]];\n\nexport type PadInfo = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  type: PadType\n};\n\nexport type PadInfo3D = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  front: number,\n  back: number,\n  type: PadType\n};\n\n/**\n * Information about the forward pass of a convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv2DInfo = {\n  batchSize: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideHeight: number,\n  strideWidth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo,\n  inShape: [number, number, number, number],\n  outShape: [number, number, number, number],\n  filterShape: [number, number, number, number]\n};\n\n/**\n *\n * @param inputShape Input tensor shape is of the following dimensions:\n *     `[batch, height, width, inChannels]`.\n * @param filterShape The filter shape is of the following dimensions:\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat The data format of the input and output data.\n *     Defaults to 'NHWC'.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.\n *     Defaults to `[1, 1]`. If `dilations` is a single number, then\n *     `dilationHeight == dilationWidth`.\n */\nexport function computeDilation2DInfo(\n    inputShape: [number, number, number, number],\n    filterShape: [number, number, number], strides: number|[number, number],\n    pad: 'same'|'valid'|number, dataFormat: 'NHWC' = 'NHWC',\n    dilations: number|[number, number]) {\n  // `computerConv2DInfo` require filterShape to be in the dimension of:\n  // `[filterHeight, filterWidth, depth, outDepth]`, dilation2d doesn't have\n  // outDepth, it should have the same depth as the input.\n  // Input shape: [batch, height, width, inChannels]\n  const inputChannels = inputShape[3];\n  const $filterShape =\n      [...filterShape, inputChannels] as [number, number, number, number];\n  const $dataFormat = convertConv2DDataFormat(dataFormat);\n\n  return computeConv2DInfo(\n      inputShape, $filterShape, strides, dilations, pad,\n      null /* roundingMode */, null /* depthWise */, $dataFormat);\n}\n\nexport function computePool2DInfo(\n    inShape: [number, number, number, number],\n    filterSize: [number, number]|number, strides: number|[number, number],\n    dilations: number|[number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  const [filterHeight, filterWidth] = parseTupleParam(filterSize);\n\n  let filterShape: [number, number, number, number];\n  if (dataFormat === 'channelsLast') {\n    filterShape = [filterHeight, filterWidth, inShape[3], inShape[3]];\n  } else if (dataFormat === 'channelsFirst') {\n    filterShape = [filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv2DInfo(\n      inShape, filterShape, strides, dilations, pad, roundingMode, false,\n      dataFormat);\n}\n\n/**\n * Computes the information for a forward pass of a pooling3D operation.\n */\nexport function computePool3DInfo(\n    inShape: [number, number, number, number, number],\n    filterSize: number|[number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC'): Conv3DInfo {\n  const [filterDepth, filterHeight, filterWidth] = parse3TupleParam(filterSize);\n\n  let filterShape: [number, number, number, number, number];\n  let $dataFormat: 'channelsFirst'|'channelsLast';\n  if (dataFormat === 'NDHWC') {\n    $dataFormat = 'channelsLast';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[4], inShape[4]];\n  } else if (dataFormat === 'NCDHW') {\n    $dataFormat = 'channelsFirst';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv3DInfo(\n      inShape, filterShape, strides, dilations, pad, false, $dataFormat,\n      roundingMode);\n}\n\n/**\n * Computes the information for a forward pass of a convolution/pooling\n * operation.\n */\nexport function computeConv2DInfo(\n    inShape: [number, number, number, number],\n    filterShape: [number, number, number, number],\n    strides: number|[number, number], dilations: number|[number, number],\n    pad: 'same'|'valid'|number|ExplicitPadding,\n    roundingMode?: 'floor'|'round'|'ceil', depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  let [batchSize, inHeight, inWidth, inChannels] = [-1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterHeight, filterWidth, , filterChannels] = filterShape;\n  const [strideHeight, strideWidth] = parseTupleParam(strides);\n  const [dilationHeight, dilationWidth] = parseTupleParam(dilations);\n\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outHeight, outWidth} = getPadAndOutInfo(\n      pad, inHeight, inWidth, strideHeight, strideWidth, effectiveFilterHeight,\n      effectiveFilterWidth, roundingMode, dataFormat);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inHeight,\n    inWidth,\n    inChannels,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideHeight,\n    strideWidth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\n/**\n * Information about the forward pass of a 3D convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv3DInfo = {\n  batchSize: number,\n  inDepth: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideDepth: number,\n  strideHeight: number,\n  strideWidth: number,\n  dilationDepth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterDepth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterDepth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo3D,\n  inShape: [number, number, number, number, number],\n  outShape: [number, number, number, number, number],\n  filterShape: [number, number, number, number, number]\n};\n\n/**\n * Computes the information for a forward pass of a 3D convolution/pooling\n * operation.\n */\nexport function computeConv3DInfo(\n    inShape: [number, number, number, number, number],\n    filterShape: [number, number, number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast',\n    roundingMode?: 'floor'|'round'|'ceil'): Conv3DInfo {\n  let [batchSize, inDepth, inHeight, inWidth, inChannels] =\n      [-1, -1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inDepth, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inDepth, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterDepth, filterHeight, filterWidth, , filterChannels] =\n      filterShape;\n  const [strideDepth, strideHeight, strideWidth] = parse3TupleParam(strides);\n  const [dilationDepth, dilationHeight, dilationWidth] =\n      parse3TupleParam(dilations);\n\n  const effectiveFilterDepth =\n      getEffectiveFilterSize(filterDepth, dilationDepth);\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outDepth, outHeight, outWidth} = get3DPadAndOutInfo(\n      pad, inDepth, inHeight, inWidth, strideDepth, strideHeight, strideWidth,\n      effectiveFilterDepth, effectiveFilterHeight, effectiveFilterWidth,\n      roundingMode);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outDepth, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outDepth, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inDepth,\n    inHeight,\n    inWidth,\n    inChannels,\n    outDepth,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideDepth,\n    strideHeight,\n    strideWidth,\n    filterDepth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterDepth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationDepth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\nfunction computeOutputShape2D(\n    inShape: [number, number], fieldSize: number, stride: number,\n    zeroPad?: number, roundingMode?: 'floor'|'round'|'ceil'): [number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputRows = inShape[0];\n  const inputCols = inShape[1];\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputRows, outputCols];\n}\n\nfunction computeOutputShape4D(\n    inShape: [number, number, number, number], fieldSize: number,\n    outChannels: number, stride: number, zeroPad?: number,\n    roundingMode?: 'floor'|'round'|'ceil'): [number, number, number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputDepth = inShape[0];\n  const inputRows = inShape[1];\n  const inputCols = inShape[2];\n\n  const outputDepths = conditionalRound(\n      (inputDepth - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputDepths),\n      () => `The output # of depths (${outputDepths}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputDepths, outputRows, outputCols, outChannels];\n}\n\nexport function computeDefaultPad(\n    inputShape: [number, number]|[number, number, number, number],\n    fieldSize: number, stride: number, dilation = 1): number {\n  const effectiveFieldSize = getEffectiveFilterSize(fieldSize, dilation);\n  return Math.floor(\n      (inputShape[0] * (stride - 1) - stride + effectiveFieldSize) / 2);\n}\n\nfunction parseTupleParam(param: number|number[]): [number, number, number] {\n  if (typeof param === 'number') {\n    return [param, param, param];\n  }\n  if (param.length === 2) {\n    return [param[0], param[1], 1];\n  }\n  return param as [number, number, number];\n}\n\nfunction parse3TupleParam(param: number|[number, number, number]):\n    [number, number, number] {\n  return typeof param === 'number' ? [param, param, param] : param;\n}\n\n/* See https://www.tensorflow.org/api_docs/python/tf/nn/atrous_conv2d\n * Atrous convolution is equivalent to standard convolution with upsampled\n * filters with effective_filter_height =\n * filter_height + (filter_height - 1) * (dilation - 1)\n * and effective_filter_width =\n * filter_width + (filter_width - 1) * (dilation - 1),\n * produced by inserting dilation - 1 zeros along consecutive elements across\n * the filters' spatial dimensions.\n * When there is a dilation, this converts a filter dimension to the\n * effective filter dimension, so it can be used in a standard convolution.\n */\nfunction getEffectiveFilterSize(filterSize: number, dilation: number) {\n  if (dilation <= 1) {\n    return filterSize;\n  }\n\n  return filterSize + (filterSize - 1) * (dilation - 1);\n}\n\nfunction getPadAndOutInfo(\n    pad: 'same'|'valid'|number|ExplicitPadding, inHeight: number,\n    inWidth: number, strideHeight: number, strideWidth: number,\n    filterHeight: number, filterWidth: number,\n    roundingMode: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|\n    'channelsLast'): {padInfo: PadInfo, outHeight: number, outWidth: number} {\n  let padInfo: PadInfo;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {top: pad, bottom: pad, left: pad, right: pad, type: padType};\n    const outShape = computeOutputShape2D(\n        [inHeight, inWidth], filterHeight, strideHeight, pad, roundingMode);\n    outHeight = outShape[0];\n    outWidth = outShape[1];\n  } else if (pad === 'same') {\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongHeight =\n        Math.max(0, (outHeight - 1) * strideHeight + filterHeight - inHeight);\n    const padAlongWidth =\n        Math.max(0, (outWidth - 1) * strideWidth + filterWidth - inWidth);\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n    padInfo = {top, bottom, left, right, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {top: 0, bottom: 0, left: 0, right: 0, type: 'VALID'};\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else if (typeof pad === 'object') {\n    const top = dataFormat === 'channelsLast' ? pad[1][0] : pad[2][0];\n    const bottom = dataFormat === 'channelsLast' ? pad[1][1] : pad[2][1];\n    const left = dataFormat === 'channelsLast' ? pad[2][0] : pad[3][0];\n    const right = dataFormat === 'channelsLast' ? pad[2][1] : pad[3][1];\n    const padType = (top === 0 && bottom === 0 && left === 0 && right === 0) ?\n        'VALID' :\n        'EXPLICIT';\n    padInfo = {top, bottom, left, right, type: padType};\n    outHeight = conditionalRound(\n        (inHeight - filterHeight + top + bottom) / strideHeight + 1,\n        roundingMode);\n    outWidth = conditionalRound(\n        (inWidth - filterWidth + left + right) / strideWidth + 1, roundingMode);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outHeight, outWidth};\n}\n\nfunction get3DPadAndOutInfo(\n    pad: 'same'|'valid'|number, inDepth: number, inHeight: number,\n    inWidth: number, strideDepth: number, strideHeight: number,\n    strideWidth: number, filterDepth: number, filterHeight: number,\n    filterWidth: number, roundingMode?: 'floor'|'round'|'ceil'): {\n  padInfo: PadInfo3D,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number\n} {\n  let padInfo: PadInfo3D;\n  let outDepth: number;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {\n      top: pad,\n      bottom: pad,\n      left: pad,\n      right: pad,\n      front: pad,\n      back: pad,\n      type: padType\n    };\n    const outShape = computeOutputShape4D(\n        [inDepth, inHeight, inWidth, 1], filterDepth, 1, strideDepth, pad,\n        roundingMode);\n    outDepth = outShape[0];\n    outHeight = outShape[1];\n    outWidth = outShape[2];\n  } else if (pad === 'same') {\n    outDepth = Math.ceil(inDepth / strideDepth);\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongDepth = (outDepth - 1) * strideDepth + filterDepth - inDepth;\n    const padAlongHeight =\n        (outHeight - 1) * strideHeight + filterHeight - inHeight;\n    const padAlongWidth = (outWidth - 1) * strideWidth + filterWidth - inWidth;\n    const front = Math.floor(padAlongDepth / 2);\n    const back = padAlongDepth - front;\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n\n    padInfo = {top, bottom, left, right, front, back, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {\n      top: 0,\n      bottom: 0,\n      left: 0,\n      right: 0,\n      front: 0,\n      back: 0,\n      type: 'VALID'\n    };\n    outDepth = Math.ceil((inDepth - filterDepth + 1) / strideDepth);\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outDepth, outHeight, outWidth};\n}\n\n/**\n * Rounds a value depending on the rounding mode\n * @param value\n * @param roundingMode\n */\nfunction conditionalRound(\n    value: number, roundingMode?: 'floor'|'round'|'ceil') {\n  if (!roundingMode) {\n    return value;\n  }\n  switch (roundingMode) {\n    case 'round':\n      // used for Caffe Conv\n      return Math.round(value);\n    case 'ceil':\n      // used for Caffe Pool\n      return Math.ceil(value);\n    case 'floor':\n      return Math.floor(value);\n    default:\n      throw new Error(`Unknown roundingMode ${roundingMode}`);\n  }\n}\n\nexport function tupleValuesAreOne(param: number|number[]): boolean {\n  const [dimA, dimB, dimC] = parseTupleParam(param);\n  return dimA === 1 && dimB === 1 && dimC === 1;\n}\n\nexport function eitherStridesOrDilationsAreOne(\n    strides: number|number[], dilations: number|number[]): boolean {\n  return tupleValuesAreOne(strides) || tupleValuesAreOne(dilations);\n}\n\n/**\n * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to\n *    'channelsLast'|'channelsFirst'\n * @param dataFormat in 'NHWC'|'NCHW' mode\n * @return dataFormat in 'channelsLast'|'channelsFirst' mode\n * @throws unknown dataFormat\n */\nexport function convertConv2DDataFormat(dataFormat: 'NHWC'|'NCHW'):\n    'channelsLast'|'channelsFirst' {\n  if (dataFormat === 'NHWC') {\n    return 'channelsLast';\n  } else if (dataFormat === 'NCHW') {\n    return 'channelsFirst';\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D average pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction avgPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'avgPool', 'float32');\n  const dilations = 1;\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in avgPool: x must be rank 4 but got rank ${x4D.rank}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    save([x4D]);\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      return x4D.clone();\n    }\n\n    return backend.avgPool(x4D, convInfo);\n  };\n\n  const inputs: AvgPoolInputs = {x: x4D};\n\n  const attrs: AvgPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, $x.dtype);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool = op({avgPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {AvgPool3D, AvgPool3DAttrs, AvgPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D average pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.avgPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates:\n *     `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction avgPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'avgPool3d', 'float32');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in avgPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in avgPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n\n    save([x5D]);\n\n    return backend.avgPool3d(x5D, convInfo);\n  };\n\n  const inputs: AvgPool3DInputs = {x: x5D};\n\n  const attrs: AvgPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool3D,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, x5D.dtype);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3d = op({avgPool3d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\nexport function assertParamsConsistent(shapes: number[][], axis: number) {\n  const rank = shapes[0].length;\n  shapes.forEach((shape, i) => {\n    util.assert(\n        shape.length === rank,\n        () =>\n            `Error in concat${rank}D: rank of tensors[${i}] must be the same ` +\n            `as the rank of the rest (${rank})`);\n  });\n\n  util.assert(\n      axis >= 0 && axis < rank,\n      () => `Error in concat${rank}D: axis must be between 0 and ${rank - 1}.`);\n\n  const firstShape = shapes[0];\n  shapes.forEach((shape, i) => {\n    for (let r = 0; r < rank; r++) {\n      util.assert(\n          (r === axis) || (shape[r] === firstShape[r]),\n          () => `Error in concat${rank}D: Shape of tensors[${i}] (${shape}) ` +\n              `does not match the shape of the rest (${firstShape}) ` +\n              `along the non-concatenated axis ${i}.`);\n    }\n  });\n}\n\nexport function computeOutShape(shapes: number[][], axis: number): number[] {\n  const outputShape = shapes[0].slice();\n  for (let i = 1; i < shapes.length; i++) {\n    outputShape[axis] += shapes[i][axis];\n  }\n  return outputShape;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Concat, ConcatAttrs, ConcatInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, parseAxisParam, sizeFromShape} from '../util';\n\nimport {assertParamsConsistent, computeOutShape} from './concat_util';\nimport {op} from './operation';\nimport {tensor} from './tensor';\n\n/**\n * Concatenates a list of `tf.Tensor`s along a given axis.\n *\n * The tensors ranks and types must match, and their sizes must match in all\n * dimensions except `axis`.\n *\n * Also available are stricter rank-specific methods that assert that\n * `tensors` are of the given rank:\n *   - `tf.concat1d`\n *   - `tf.concat2d`\n *   - `tf.concat3d`\n *   - `tf.concat4d`\n *\n * Except `tf.concat1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * a.concat(b).print();  // or a.concat(b)\n * ```\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.concat([a, b, c]).print();\n * ```\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [10, 20]]);\n * const b = tf.tensor2d([[3, 4], [30, 40]]);\n * const axis = 1;\n * tf.concat([a, b], axis).print();\n * ```\n * @param tensors A list of tensors to concatenate.\n * @param axis The axis to concate along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction concat_<T extends Tensor>(tensors: Array<T|TensorLike>, axis = 0): T {\n  assert(tensors.length >= 1, () => 'Pass at least one tensor to concat');\n\n  let $tensors = convertToTensorArray(tensors, 'tensors', 'concat');\n\n  if ($tensors[0].dtype === 'complex64') {\n    $tensors.forEach(tensor => {\n      if (tensor.dtype !== 'complex64') {\n        throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${tensor.dtype}. `);\n      }\n    });\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $axis = parseAxisParam(axis, $tensors[0].shape)[0];\n    const outShape = computeOutShape($tensors.map(t => t.shape), $axis);\n    if (sizeFromShape(outShape) === 0) {\n      return tensor([], outShape) as T;\n    }\n    // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n    $tensors = $tensors.filter(t => t.size > 0);\n    if ($tensors.length === 1) {\n      return $tensors[0];\n    }\n\n    const shapes = $tensors.map(t => t.shape);\n    assertParamsConsistent(shapes, $axis);\n\n    const res = backend.concat($tensors, $axis);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: ConcatInputs = $tensors;\n  const attr: ConcatAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Concat,\n             attr as {} as NamedAttrMap) as T;\n}\n\nexport const concat = op({concat_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sigmoid, SigmoidInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sigmoid element-wise, `1 / (1 + exp(-x))`\n *\n * ```js\n * const x = tf.tensor1d([0, -1, 2, -3]);\n *\n * x.sigmoid().print();  // or tf.sigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sigmoid');\n\n  const inputs: SigmoidInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sigmoid($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sigmoid);\n}\nexport const sigmoid = op({sigmoid_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Slice, SliceAttrs, SliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as slice_util from './slice_util';\n\n/**\n * Extracts a slice from a `tf.Tensor` starting at coordinates `begin`\n * and is of size `size`.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `x` is of the given rank:\n *   - `tf.slice1d`\n *   - `tf.slice2d`\n *   - `tf.slice3d`\n *   - `tf.slice4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.slice([1], [2]).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * x.slice([1, 0], [1, 2]).print();\n * ```\n * @param x The input `tf.Tensor` to slice from.\n * @param begin The coordinates to start the slice from. The length can be\n *     less than the rank of x - the rest of the axes will have implicit 0 as\n *     start. Can also be a single number, in which case it specifies the\n *     first axis.\n * @param size The size of the slice. The length can be less than the rank of\n *     x - the rest of the axes will have implicit -1. A value of -1 requests\n *     the rest of the dimensions in the axis. Can also be a single number,\n *     in which case it specifies the size of the first axis.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction slice_<R extends Rank, T extends Tensor<R>>(\n    x: T|TensorLike, begin: number|number[], size?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'slice');\n\n  if ($x.rank === 0) {\n    throw new Error('Slicing scalar is not possible');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const [begin_, size_] = slice_util.parseSliceParams($x, begin, size);\n    slice_util.assertParamsValid($x, begin_, size_);\n    save([$x]);\n    return backend.slice($x, begin_, size_);\n  };\n\n  const inputs: SliceInputs = {x: $x};\n  const attrs: SliceAttrs = {begin, size};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Slice,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const slice = op({slice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tanh, TanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic tangent of the input `tf.Tensor` element-wise: `tanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, 70]);\n *\n * x.tanh().print();  // or tf.tanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tanh');\n\n  const inputs: TanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const y = backend.tanh($x);\n    save([y]);\n    return y;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tanh);\n}\nexport const tanh = op({tanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {add} from './add';\nimport {concat} from './concat';\nimport {matMul} from './mat_mul';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {slice} from './slice';\nimport {tanh} from './tanh';\n\n/**\n * Computes the next state and output of a BasicLSTMCell.\n *\n * Returns `[newC, newH]`.\n *\n * Derived from tf.contrib.rnn.BasicLSTMCell.\n *\n * @param forgetBias Forget bias for the cell.\n * @param lstmKernel The weights for the cell.\n * @param lstmBias The bias for the cell.\n * @param data The input to the cell.\n * @param c Previous cell state.\n * @param h Previous cell output.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction basicLSTMCell_(\n    forgetBias: Scalar|TensorLike, lstmKernel: Tensor2D|TensorLike,\n    lstmBias: Tensor1D|TensorLike, data: Tensor2D|TensorLike,\n    c: Tensor2D|TensorLike, h: Tensor2D|TensorLike): [Tensor2D, Tensor2D] {\n  const $forgetBias =\n      convertToTensor(forgetBias, 'forgetBias', 'basicLSTMCell');\n  const $lstmKernel =\n      convertToTensor(lstmKernel, 'lstmKernel', 'basicLSTMCell');\n  const $lstmBias = convertToTensor(lstmBias, 'lstmBias', 'basicLSTMCell');\n  const $data = convertToTensor(data, 'data', 'basicLSTMCell');\n  const $c = convertToTensor(c, 'c', 'basicLSTMCell');\n  const $h = convertToTensor(h, 'h', 'basicLSTMCell');\n\n  const combined = concat([$data, $h], 1);\n  const weighted = matMul(combined, $lstmKernel);\n  const res: Tensor2D = add(weighted, $lstmBias);\n\n  // i = input_gate, j = new_input, f = forget_gate, o = output_gate\n  const batchSize = res.shape[0];\n  const sliceCols = res.shape[1] / 4;\n  const sliceSize: [number, number] = [batchSize, sliceCols];\n  const i = slice(res, [0, 0], sliceSize);\n  const j = slice(res, [0, sliceCols], sliceSize);\n  const f = slice(res, [0, sliceCols * 2], sliceSize);\n  const o = slice(res, [0, sliceCols * 3], sliceSize);\n\n  const newC: Tensor2D =\n      add(mul(sigmoid(i), tanh(j)),\n          mul($c, sigmoid(add($forgetBias, f)) as Tensor2D));\n  const newH: Tensor2D = mul(tanh(newC), sigmoid(o));\n  return [newC, newH];\n}\n\nexport const basicLSTMCell = op({basicLSTMCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchToSpaceND, BatchToSpaceNDAttrs, BatchToSpaceNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation reshapes the \"batch\" dimension 0 into `M + 1` dimensions of\n * shape `blockShape + [batch]`, interleaves these blocks back into the grid\n * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with\n * the same rank as the input. The spatial dimensions of this intermediate\n * result are then optionally cropped according to `crops` to produce the\n * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise\n * description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);\n * const blockShape = [2, 2];\n * const crops = [[0, 0], [0, 0]];\n *\n * x.batchToSpaceND(blockShape, crops).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param crops A 2-D array.  Must have shape `[M, 2]`, all values must be >= 0.\n * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input\n * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required\n * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,\n * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,\n * x.shape[N-1]]`\n *\n * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /\n * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`\n * according to `crops` to produce the output of shape: `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],\n * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -\n * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction batchToSpaceND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], crops: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'batchToSpaceND');\n  const prod = blockShape.reduce((a, b) => a * b);\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank is ${$x.rank} but should be > than blockShape.length ${\n          blockShape.length}`);\n\n  util.assert(\n      crops.length === blockShape.length,\n      () => `crops.length is ${\n          crops.length} but should be equal to blockShape.length  ${\n          blockShape.length}`);\n\n  util.assert(\n      $x.shape[0] % prod === 0,\n      () => `input tensor batch is ${\n                $x.shape[0]} but is not divisible by the product of ` +\n          `the elements of blockShape ${blockShape.join(' * ')} === ${prod}`);\n\n  const forward: ForwardFunc<T> = backend => {\n    return backend.batchToSpaceND($x, blockShape, crops);\n  };\n\n  const inputs: BatchToSpaceNDInputs = {x: $x};\n  const attrs: BatchToSpaceNDAttrs = {blockShape, crops};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      BatchToSpaceND, attrs as {} as NamedAttrMap);\n}\n\nexport const batchToSpaceND = op({batchToSpaceND_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, Tensor4D} from '../tensor';\nimport {Rank} from '../types';\nimport {reshape} from './reshape';\n\nexport function xAs4D<R extends Rank>(x: Tensor<R>) {\n  let x4D: Tensor4D;\n  if (x.rank === 0 || x.rank === 1) {\n    x4D = reshape(x, [1, 1, 1, x.size]);\n  } else if (x.rank === 2) {\n    x4D = reshape(x, [1, 1, x.shape[0], x.shape[1]]);\n  } else if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  } else {\n    x4D = x as Tensor4D;\n  }\n\n  return x4D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {xAs4D} from './batchnorm_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Batch normalization.\n *\n * As described in\n * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).\n *\n * Mean, variance, scale, and offset can be of two shapes:\n *   - The same shape as the input.\n *   - In the common case, the depth dimension is the last dimension of x, so\n *     the values would be an `tf.Tensor1D` of shape [depth].\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that parameters passed are of given rank\n *   - `tf.batchNorm2d`\n *   - `tf.batchNorm3d`\n *   - `tf.batchNorm4d`\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction batchNorm_<R extends Rank>(\n    x: Tensor<R>|TensorLike, mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor<R>|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor<R>|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n\n  util.assert(\n      $mean.rank === $variance.rank,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      $offset == null || $mean.rank === $offset.rank,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      $scale == null || $mean.rank === $scale.rank,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  const x4D: Tensor4D = xAs4D($x);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([x4D, $mean, $variance, $scale]);\n\n    return backend.batchNorm(\n        x4D, as1DOr4D($mean), as1DOr4D($variance), as1DOr4D($offset),\n        as1DOr4D($scale), varianceEpsilon);\n  };\n\n  const inputs: FusedBatchNormInputs = {\n    x: x4D,\n    scale: $scale,\n    offset: $offset,\n    mean: $mean,\n    variance: $variance\n  };\n\n  const attrs: FusedBatchNormAttrs = {varianceEpsilon};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      FusedBatchNorm, attrs as {} as NamedAttrMap);\n\n  return reshape(res, $x.shape);\n}\n\nfunction as1DOr4D(x: Tensor): Tensor4D|Tensor1D {\n  if (x == null) {\n    return null;\n  }\n  if (x.rank === 0) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [x.size]) as Tensor1D;\n  } else if (x.rank === 1) {\n    return x as Tensor1D;\n  } else if (x.rank === 2) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, 1, x.shape[0], x.shape[1]]) as Tensor4D;\n  } else if (x.rank === 3) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]) as Tensor4D;\n  }\n  return x as Tensor4D;\n}\n\nexport const batchNorm = op({batchNorm_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 2D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm2d_(\n    x: Tensor2D|TensorLike, mean: Tensor2D|Tensor1D|TensorLike,\n    variance: Tensor2D|Tensor1D|TensorLike,\n    offset?: Tensor2D|Tensor1D|TensorLike, scale?: Tensor2D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor2D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor2D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 2,\n      () => `Error in batchNorm2D: x must be rank 2 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 2 || $mean.rank === 1,\n      () => `Error in batchNorm2D: mean must be rank 2 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 2 || $variance.rank === 1,\n      () => `Error in batchNorm2D: variance must be rank 2 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 2 || $scale.rank === 1,\n        () => `Error in batchNorm2D: scale must be rank 2 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 2 || $offset.rank === 1,\n        () => `Error in batchNorm2D: offset must be rank 2 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm2d = op({batchNorm2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 3D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm3d_(\n    x: Tensor3D|TensorLike, mean: Tensor3D|Tensor1D|TensorLike,\n    variance: Tensor3D|Tensor1D|TensorLike,\n    offset?: Tensor3D|Tensor1D|TensorLike, scale?: Tensor3D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor3D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor3D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 3,\n      () => `Error in batchNorm3D: x must be rank 3 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 3 || $mean.rank === 1,\n      () => `Error in batchNorm3D: mean must be rank 3 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 3 || $variance.rank === 1,\n      () => `Error in batchNorm3D: variance must be rank 3 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 3 || $scale.rank === 1,\n        () => `Error in batchNorm3D: scale must be rank 3 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 3 || $offset.rank === 1,\n        () => `Error in batchNorm3D: offset must be rank 3 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm3d = op({batchNorm3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 4D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm4d_(\n    x: Tensor4D|TensorLike, mean: Tensor4D|Tensor1D|TensorLike,\n    variance: Tensor4D|Tensor1D|TensorLike,\n    offset?: Tensor4D|Tensor1D|TensorLike, scale?: Tensor4D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor4D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor4D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 4,\n      () => `Error in batchNorm4D: x must be rank 4 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 4 || $mean.rank === 1,\n      () => `Error in batchNorm4D: mean must be rank 4 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 4 || $variance.rank === 1,\n      () => `Error in batchNorm4D: variance must be rank 4 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 4 || $scale.rank === 1,\n        () => `Error in batchNorm4D: scale must be rank 4 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 4 || $offset.rank === 1,\n        () => `Error in batchNorm4D: offset must be rank 4 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm4d = op({batchNorm4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE} from '../engine';\nimport {BroadcastTo, BroadCastToAttrs, BroadcastToInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Broadcast an array to a compatible shape NumPy-style.\n *\n * The tensor's shape is compared to the broadcast shape from end to beginning.\n * Ones are prepended to the tensor's shape until is has the same length as\n * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is\n * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then\n * the input tensor is tiled N times along that axis (using tf.tile).\n *\n * @param input The tensor that is to be broadcasted.\n * @param shape The input is to be broadcast to this shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction broadcastTo_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  let input = convertToTensor(x, 'broadcastTo', 'x');\n  const xShape = input.shape;\n\n  if (shape.some(d => !(d > 0) || d % 1 !== 0)) {\n    throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);\n  }\n\n  if (shape.length < input.rank) {\n    throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${\n        input.rank}.`);\n  }\n\n  if (shape.length > input.rank) {\n    const newShape = input.shape.slice();\n    while (newShape.length < shape.length) {\n      newShape.unshift(1);\n    }\n    input = reshape(input, newShape);\n  }\n\n  const inputShape = input.shape;\n  const reps: number[] = Array.from(shape);\n  for (let i = shape.length - 1; i >= 0; i--) {\n    if (inputShape[i] === shape[i]) {\n      reps[i] = 1;\n    } else if (input.shape[i] !== 1) {\n      throw new Error(\n          `broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);\n    }\n  }\n  const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);\n\n  if (axes.length === 0) {\n    return clone(input) as Tensor<R>;\n  }\n\n  const forward = (backend: KernelBackend) => backend.tile(input, reps);\n\n  const inputs: BroadcastToInputs = {x: input};\n  const attrs: BroadCastToAttrs = {shape, inputShape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             BroadcastTo, attrs as unknown as NamedAttrMap) as Tensor<R>;\n}\n\nexport const broadcastTo = op({broadcastTo_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Ceil, CeilInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes ceiling of input `tf.Tensor` element-wise: `ceil(x)`\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.ceil().print();  // or tf.ceil(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction ceil_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'ceil');\n\n  const inputs: CeilInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.ceil($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Ceil);\n}\nexport const ceil = op({ceil_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {ClipByValue, ClipByValueAttrs, ClipByValueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)\n * ```\n * @param x The input tensor.\n * @param clipValueMin Lower-bound of range to be clipped to.\n * @param clipValueMax Upper-bound of range to be clipped to.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction clipByValue_<T extends Tensor>(\n    x: T|TensorLike, clipValueMin: number, clipValueMax: number): T {\n  const $x = convertToTensor(x, 'x', 'clipByValue');\n  util.assert(\n      (clipValueMin <= clipValueMax),\n      () => `Error in clip: min (${clipValueMin}) must be ` +\n          `less than or equal to max (${clipValueMax}).`);\n\n  const inputs: ClipByValueInputs = {x: $x};\n  const attrs: ClipByValueAttrs = {clipValueMin, clipValueMax};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const res = backend.clip($x, clipValueMin, clipValueMax);\n        save([$x]);\n        return res;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, ClipByValue,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const clipByValue = op({clipByValue_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor1D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(3) = |r1, g1, b1|\n * B: shape(2) = |r2, g2|\n * C = tf.concat1d([A, B]) == |r1, g1, b1, r2, g2|\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @return The concatenated array.\n */\nfunction concat1d_(tensors: Array<Tensor1D|TensorLike>): Tensor1D {\n  return concat(tensors, 0 /* axis */);\n}\n\nexport const concat1d = op({concat1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor2D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *\n * B: shape(2, 3) = | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * C = tf.concat2d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *                  | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * if axis = 1:\n * C = shape(2, 6) = | r1, g1, b1, r3, g3, b3 |\n *                   | r2, g2, b2, r4, g4, b4 |\n *\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concatenate along.\n * @return The concatenated array.\n */\nfunction concat2d_(\n    tensors: Array<Tensor2D|TensorLike>, axis: number): Tensor2D {\n  return concat(tensors, axis);\n}\n\nexport const concat2d = op({concat2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor3D`s along an axis.\n * See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *\n * B: shape(2, 1, 3) = | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * C = tf.concat3d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *                     | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * if axis = 1:\n * C: shape(2, 2, 3) = | r1, g1, b1, r3, g3, b3 |\n *                     | r2, g2, b2, r4, g4, b4 |\n *\n * if axis = 2:\n * C = shape(2, 1, 6) = | r1, g1, b1, r3, g3, b3 |\n *                      | r2, g2, b2, r4, g4, b4 |\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat3d_(\n    tensors: Array<Tensor3D|TensorLike>, axis: number): Tensor3D {\n  return concat(tensors, axis);\n}\n\nexport const concat3d = op({concat3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor4D`s along an axis.\n * See `concat` for details.\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat4d_(\n    tensors: Array<Tensor4D|TensorLike>, axis: number): Tensor4D {\n  return concat(tensors, axis);\n}\n\nexport const concat4d = op({concat4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2D, Conv2DAttrs, Conv2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 2D convolution over the input x.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2d: input must be rank 4, but got rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  util.assert(\n      inDepth === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${inDepth}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        false, $dataFormat);\n    const res = backend.conv2d(x4D, $filter, convInfo);\n\n    save([x4D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DInputs = {x: x4D, filter: $filter};\n  const attrs:\n      Conv2DAttrs = {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv2D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2d = op({conv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D, Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 1D convolution over the input x.\n *\n * @param x The input tensor, of rank 3 or rank 2, of shape\n *     `[batch, width, inChannels]`. If rank 2, batch of 1 is assumed.\n * @param filter The filter, rank 3, of shape\n *     `[filterWidth, inDepth, outDepth]`.\n * @param stride The number of entries by which the filter is moved right at\n *     each step.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from \"NWC\", \"NCW\". Defaults to \"NWC\",\n *     the data is stored in the order of [batch, in_width, in_channels]. Only\n *     \"NWC\" is currently supported.\n * @param dilation The dilation rate in which we sample input values in\n *     atrous convolution. Defaults to `1`. If it is greater than 1, then\n *     stride must be `1`.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv1d_<T extends Tensor2D|Tensor3D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike, stride: number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv1d');\n  const $filter = convertToTensor(filter, 'filter', 'conv1d');\n\n  let x3D = $x as Tensor3D;\n  let reshapedTo3D = false;\n  if ($x.rank === 2) {\n    reshapedTo3D = true;\n    x3D = reshape($x, [1, $x.shape[0], $x.shape[1]]);\n  }\n\n  util.assert(\n      x3D.rank === 3,\n      () => `Error in conv1d: input must be rank 3, but got rank ${x3D.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in conv1d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv1d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x3D.shape[2] === $filter.shape[1],\n      () => `Error in conv1d: depth of input (${x3D.shape[2]}) must match ` +\n          `input depth for filter ${$filter.shape[1]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(stride, dilation),\n      () => 'Error in conv1D: Either stride or dilation must be 1. ' +\n          `Got stride ${stride} and dilation '${dilation}'`);\n  util.assert(\n      dataFormat === 'NWC',\n      () => `Error in conv1d: got dataFormat of ${\n          dataFormat} but only NWC is currently supported.`);\n\n  const filter4D = reshape(\n      $filter, [1, $filter.shape[0], $filter.shape[1], $filter.shape[2]]);\n  const input4D = reshape(x3D, [x3D.shape[0], 1, x3D.shape[1], x3D.shape[2]]);\n  const strides: [number, number] = [1, stride];\n  const dilations: [number, number] = [1, dilation];\n\n  const conv2dDataFormat = 'NHWC';\n\n  const res = conv2d(\n      (input4D as Tensor4D), (filter4D as Tensor4D), strides, pad,\n      conv2dDataFormat, dilations, dimRoundingMode);\n\n  if (reshapedTo3D) {\n    return reshape(res, [res.shape[2], res.shape[3]]) as T;\n  }\n\n  return reshape(res, [res.shape[0], res.shape[2], res.shape[3]]) as T;\n}\n\nexport const conv1d = op({conv1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs, Conv2DBackpropInputInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 2D convolution.\n *\n * @param xShape The shape of the input: [batch, height, width, inDepth].\n * If length of 3, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 4 or rank 3 of shape\n *   `[batch, outHeight, outWidth, outDepth]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction conv2DBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape4D = xShape as [number, number, number, number];\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n    xShape4D = [1, xShape[0], xShape[1], xShape[2]];\n  }\n\n  util.assert(\n      xShape4D.length === 4,\n      () =>\n          `Error in conv2dDerInput: inShape must be length 4, but got length ` +\n          `${xShape4D.length}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerInput: dy must be rank 4, but got ` +\n          `rank ${dy4D.rank}`);\n  util.assert(\n      filter.rank === 4,\n      () => `Error in conv2dDerInput: filter must be rank 4, but got ` +\n          `rank ${filter.rank}`);\n  const inDepth = dataFormat === 'NHWC' ? xShape4D[3] : xShape4D[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filter.shape[2],\n      () => `Error in conv2dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[2]}.`);\n  util.assert(\n      outDepth === filter.shape[3],\n      () => `Error in conv2dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[3]}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerInput: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const dilations = 1;\n\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        xShape4D, filter.shape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    const res = backend.conv2dDerInput(dy4D, filter, convInfo);\n\n    save([dy4D, filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DBackpropInputInputs = {dy: dy4D, filter};\n\n  const attrs: Conv2DBackpropInputAttrs =\n      {strides, pad, dataFormat, dimRoundingMode, inputShape: xShape4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      Conv2DBackpropInput, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2DBackpropInput = op({conv2DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv2DBackpropInput} from './conv2d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 2D convolution of an image, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 4 or rank 3, of shape\n *   `[batch, height, width, inDepth]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 4 or rank 3:\n *     `[batch, height, width, outDepth]`. If rank 3, batch of 1 is assumed.\n * @param strides The strides of the original convolution:\n *     `[strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n * @param dimRoundingMode The rounding mode used when computing output\n *    dimensions if pad is a number. If none is provided, it will not round\n *    and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2dTranspose_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv2dTranspose');\n\n  return conv2DBackpropInput(\n      outputShape, $x, $filter, strides, pad, 'NHWC', dimRoundingMode);\n}\n\nexport const conv2dTranspose = op({conv2dTranspose_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3D, Conv3DAttrs, Conv3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {eitherStridesOrDilationsAreOne} from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 3D convolution over the input x.\n *\n * @param x The input tensor, of rank 5 or rank 4, of shape\n *     `[batch, depth, height, width, channels]`. If rank 4,\n * batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inChannels, outChannels]`.\n *      inChannels must match between input and filter.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationDepth, dilationHeight,\n *     dilationWidth]` in which we sample input values across the height\n *     and width dimensions in atrous convolution. Defaults to `[1, 1, 1]`.\n *     If `dilations` is a single number, then\n *     `dilationDepth == dilationHeight == dilationWidth`. If it is greater\n *     than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    strides: [number, number, number]|number, pad: 'valid'|'same',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations: [number, number, number]|number = [1, 1, 1]): T {\n  const $x = convertToTensor(x, 'x', 'conv3d');\n  const $filter = convertToTensor(filter, 'filter', 'conv3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3d: input must be rank 5, but got rank ${x5D.rank}.`);\n  util.assert(\n      $filter.rank === 5,\n      () => `Error in conv3d: filter must be rank 5, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x5D.shape[4] === $filter.shape[3],\n      () => `Error in conv3d: depth of input (${x5D.shape[4]}) must match ` +\n          `input depth for filter ${$filter.shape[3]}.`);\n  util.assert(\n      eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv3D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in conv3d: got dataFormat of ${\n          dataFormat} but only NDHWC is currently supported.`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, $filter.shape, strides, dilations, pad);\n    const res = backend.conv3d(x5D, $filter, convInfo);\n\n    save([x5D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv3DInputs = {x: x5D, filter: $filter};\n\n  const attrs: Conv3DAttrs = {strides, pad, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3d = op({conv3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropInputAttrs, Conv3DBackpropInputInputs, Conv3DBackpropInputV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 3D convolution.\n *\n * @param xShape The shape of the input: [batch, depth, height, width,\n * in_channels]. If length of 4, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 5 or rank 4 of shape\n *   `[batch, outDepth, outHeight, outWidth, in_channels]`.\n * If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n */\nfunction conv3DBackpropInput_<T extends Tensor4D|Tensor5D>(\n    xShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    dy: T, filter: Tensor5D, strides: [number, number, number]|number,\n    pad: 'valid'|'same'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape5D = xShape as [number, number, number, number, number];\n  let dy5D = dy as Tensor5D;\n  let reshapedTo5D = false;\n  if (dy.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n    xShape5D = [1, xShape[0], xShape[1], xShape[2], xShape[3]];\n  }\n\n  const inDepth = xShape5D[4];\n  const outDepth = dy5D.shape[4];\n  util.assert(\n      xShape5D.length === 5,\n      () =>\n          `Error in conv3dDerInput: inShape must be length 5, but got length ` +\n          `${xShape5D.length}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerInput: dy must be rank 5, but got ` +\n          `rank ${dy5D.rank}`);\n  util.assert(\n      filter.rank === 5,\n      () => `Error in conv3dDerInput: filter must be rank 5, but got ` +\n          `rank ${filter.rank}`);\n  util.assert(\n      inDepth === filter.shape[3],\n      () => `Error in conv3dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[3]}.`);\n  util.assert(\n      outDepth === filter.shape[4],\n      () => `Error in conv3dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[4]}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        xShape5D, filter.shape, strides, dilations, pad);\n\n    return backend.conv3dDerInput(dy5D, filter, convInfo);\n  };\n\n  const inputs: Conv3DBackpropInputInputs = {dy: dy5D};\n\n  const attrs: Conv3DBackpropInputAttrs = {pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, Conv3DBackpropInputV2,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3DBackpropInput = op({conv3DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D, Tensor5D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv3DBackpropInput} from './conv3d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 3D convolution of a volume, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 5 or rank 4, of shape\n *   `[batch, depth, height, width, inDepth]`. If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[depth, filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 5 or rank 4:\n *     `[batch, depth, height, width, outDepth]`. If rank 3, batch of 1 is\n *    assumed.\n * @param strides The strides of the original convolution:\n *     `[strideDepth, strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3dTranspose_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    outputShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): T {\n  const $x = convertToTensor(x, 'x', 'conv3dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv3dTranspose');\n\n  return conv3DBackpropInput(outputShape, $x, $filter, strides, pad);\n}\n\nexport const conv3dTranspose = op({conv3dTranspose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cos, CosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes cos of the input `tf.Tensor` element-wise: `cos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.cos().print();  // or tf.cos(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cos');\n\n  const inputs: CosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cos);\n}\nexport const cos = op({cos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cosh, CoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic cos of the input `tf.Tensor` element-wise: `cosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.cosh().print();  // or tf.cosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cosh');\n  const inputs: CoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cosh);\n}\nexport const cosh = op({cosh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Cumsum, CumsumAttrs, CumsumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {getAxesPermutation, getInnerMostAxes, getUndoAxesPermutation} from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the cumulative sum of a `tf.Tensor` along `axis`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4]);\n * x.cumsum().print();\n * ```\n * ```js\n * const x = tf.tensor([[1, 2], [3, 4]]);\n * x.cumsum().print();\n * ```\n *\n * @param x The input tensor to be summed.\n * @param axis The axis along which to sum. Optional. Defaults to 0.\n * @param exclusive Whether to perform exclusive cumulative sum. Optional.\n *     Defaults to false. If set to true then the sum of each tensor entry\n *     does not include its own value, but only the values previous to it\n *     along the specified axis.\n * @param reverse Whether to sum in the opposite direction. Optional.\n *     Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Scan'}\n */\nfunction cumsum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis = 0, exclusive = false, reverse = false): T {\n  const $x = convertToTensor(x, 'x', 'cumsum');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const permutation = getAxesPermutation([axis], $x.rank);\n        let permutedX = $x;\n        if (permutation != null) {\n          permutedX = transpose($x, permutation);\n        }\n        const permutedAxis = getInnerMostAxes(1, $x.rank)[0];\n        let value = backend.cumsum(permutedX, permutedAxis, exclusive, reverse);\n        save([$x]);\n\n        if (permutation != null) {\n          const reversePermutation = getUndoAxesPermutation(permutation);\n          value = transpose(value, reversePermutation);\n        }\n        return value;\n      };\n\n  const inputs: CumsumInputs = {x: $x};\n  const attrs: CumsumAttrs = {axis, exclusive, reverse};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Cumsum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const cumsum = op({cumsum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthToSpace, DepthToSpaceAttrs, DepthToSpaceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Rearranges data from depth into blocks of spatial data. More specifically,\n * this op outputs a copy of the input tensor where values from the `depth`\n * dimension are moved in spatial blocks to the `height` and `width` dimensions.\n * The attr `blockSize` indicates the input block size and how the data is\n * moved.\n *\n *  - Chunks of data of size `blockSize * blockSize` from depth are rearranged\n * into non-overlapping blocks of size `blockSize x blockSize`\n *\n *  - The width the output tensor is `inputWidth * blockSize`, whereas the\n * height is `inputHeight * blockSize`\n *\n *  - The Y, X coordinates within each block of the output image are determined\n * by the high order component of the input channel index\n *\n *  - The depth of the input tensor must be divisible by `blockSize *\n * blockSize`\n *\n * The `dataFormat` attr specifies the layout of the input and output tensors\n * with the following options: \"NHWC\": [ `batch, height, width, channels` ]\n * \"NCHW\": [ `batch, channels, height, width` ]\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 1, 1, 4]);\n * const blockSize = 2;\n * const dataFormat = \"NHWC\";\n *\n * tf.depthToSpace(x, blockSize, dataFormat).print();\n * ```\n *\n * @param x The input tensor of rank 4\n * @param blockSIze  An `int` that is `>= 2`. The size of the spatial block\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\"\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction depthToSpace_(\n    x: Tensor4D|TensorLike4D, blockSize: number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'depthToSpace') as Tensor4D;\n\n  const inputHeight = (dataFormat === 'NHWC') ? $x.shape[1] : $x.shape[2];\n  const inputWidth = (dataFormat === 'NHWC') ? $x.shape[2] : $x.shape[3];\n  const inputDepth = (dataFormat === 'NHWC') ? $x.shape[3] : $x.shape[1];\n\n  util.assert(\n      inputHeight * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputHeight} and ${blockSize}  for depthToSpace with input shape\n    ${$x.shape}`);\n\n  util.assert(\n      inputWidth * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputWidth} and ${blockSize} for depthToSpace with input shape\n        ${$x.shape}`);\n\n  util.assert(\n      (inputDepth % (blockSize * blockSize) === 0),\n      () => `Dimension size must be evenly divisible by ${\n          blockSize * blockSize} but is ${\n          inputDepth} for depthToSpace with input shape ${$x.shape}`);\n\n  const forward: ForwardFunc<Tensor4D> = backend =>\n      backend.depthToSpace($x, blockSize, dataFormat);\n\n  const inputs: DepthToSpaceInputs = {x: $x};\n  const attrs: DepthToSpaceAttrs = {blockSize, dataFormat};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      DepthToSpace, attrs as {} as NamedAttrMap);\n}\n\nexport const depthToSpace = op({depthToSpace_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs, DepthwiseConv2dNativeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Depthwise 2D convolution.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction depthwiseConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in depthwiseConv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in depthwiseConv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1];\n    }\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in depthwiseConv2d: Either strides or dilations must be ' +\n            `1. Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        true /* depthwise */);\n    const res = backend.depthwiseConv2D(x4D, $filter, convInfo);\n    save([x4D, $filter]);\n    return res;\n  };\n\n  const inputs: DepthwiseConv2dNativeInputs = {x: x4D, filter: $filter};\n  const attrs: DepthwiseConv2dNativeAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      DepthwiseConv2dNative, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2d = op({depthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Diag, DiagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a diagonal tensor with a given diagonal values.\n *\n * Given a diagonal, this operation returns a tensor with the diagonal and\n * everything else padded with zeros.\n *\n * Assume the input has dimensions `[D1,..., Dk]`, then the output is a tensor\n * of rank 2k with dimensions `[D1,..., Dk, D1,..., Dk]`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * tf.diag(x).print()\n * ```\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4, 5, 6, 6, 8], [4, 2])\n *\n * tf.diag(x).print()\n * ```\n * @param x The input tensor.\n */\nfunction diag_(x: Tensor): Tensor {\n  const $x = convertToTensor(x, 'x', 'diag');\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const flat = reshape($x, [$x.size]);\n    const result = backend.diag(flat);\n    const outShape = [...x.shape, ...x.shape];\n\n    return reshape(result, outShape);\n  };\n\n  const inputs: DiagInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Diag);\n}\n\nexport const diag = op({diag_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DAttrs, Dilation2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the grayscale dilation over the input `x`.\n *\n * @param x The input tensor, rank 3 or rank 4 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter tensor, rank 3, of shape\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat Specify the data format of the input and output data.\n *      Defaults to 'NHWC'. Only 'NHWC' is currently supported. With the\n *      default format \"NHWC\", the data is stored in the order of: [batch,\n *      height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     for atrous morphological dilation. Defaults to `[1, 1]`. If `dilations`\n *     is a single number, then `dilationHeight == dilationWidth`. If it is\n *     greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction dilation2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same',\n    dilations: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'dilation2d');\n  const $filter = convertToTensor(filter, 'filter', 'dilation2d');\n\n  util.assert(\n      $x.rank === 3 || $x.rank === 4,\n      () => `Error in dilation2d: input must be rank 3 or 4, but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in dilation2d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in dilation2d: Only NHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n    reshapedTo4D = true;\n  }\n\n  const inputs: Dilation2DInputs = {x: x4D, filter: $filter};\n  const attrs: Dilation2DAttrs = {strides, pad, dilations};\n\n  const res = ENGINE.runKernel(\n                  Dilation2D, inputs as {} as NamedTensorMap,\n                  attrs as {} as NamedAttrMap) as T;\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res;\n}\n\nexport const dilation2d = op({dilation2d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Returns the dimensions in the input shape that are broadcasted to\n * produce the provided output shape.\n *\n * The returned dimensions are 0-indexed and sorted. An example:\n * inShape = [4, 1, 3]\n * outShape = [5, 4, 3, 3]\n * result = [1]. Dimension 1 (2nd dimension of input) gets broadcasted 1 => 3.\n */\nexport function getBroadcastDims(\n    inShape: number[], outShape: number[]): number[] {\n  const inRank = inShape.length;\n  const dims: number[] = [];\n  for (let i = 0; i < inRank; i++) {\n    const dim = inRank - 1 - i;\n    const a = inShape[dim] || 1;\n    const b = outShape[outShape.length - 1 - i] || 1;\n    if (b > 1 && a === 1) {\n      dims.unshift(dim);\n    }\n  }\n  return dims;\n}\n\n/**\n * Returns the axes in the output space that should be reduced to produce\n * the input space.\n */\nexport function getReductionAxes(\n    inShape: number[], outShape: number[]): number[] {\n  const result: number[] = [];\n  for (let i = 0; i < outShape.length; i++) {\n    const inDim = inShape[inShape.length - i - 1];\n    const outAxis = outShape.length - i - 1;\n    const outDim = outShape[outAxis];\n    if (inDim == null || (inDim === 1 && outDim > 1)) {\n      result.unshift(outAxis);\n    }\n  }\n  return result;\n}\n\nexport function assertAndGetBroadcastShape(\n    shapeA: number[], shapeB: number[]): number[] {\n  const result: number[] = [];\n  const l = Math.max(shapeA.length, shapeB.length);\n\n  for (let i = 0; i < l; i++) {\n    let a = shapeA[shapeA.length - i - 1];\n    if (a == null) {\n      a = 1;\n    }\n    let b = shapeB[shapeB.length - i - 1];\n    if (b == null) {\n      b = 1;\n    }\n    if (a === 1) {\n      result.unshift(b);\n    } else if (b === 1) {\n      result.unshift(a);\n    } else if (a !== b) {\n      const errMsg = `Operands could not be broadcast together with shapes ` +\n          `${shapeA} and ${shapeB}.`;\n      throw Error(errMsg);\n    } else {\n      result.unshift(a);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Equal, EqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a == b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.equal(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction equal_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'equal');\n  let $b = convertToTensor(b, 'b', 'equal');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.equal($a, $b);\n\n  const inputs: EqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, Equal) as T;\n}\n\nexport const equal = op({equal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SelectV2, SelectV2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch} from '../util';\n\nimport {broadcastTo} from './broadcast_to';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the elements, either `a` or `b` depending on the `condition`.\n *\n * If the condition is true, select from `a`, otherwise select from `b`.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const a = tf.tensor1d([1 , 2, 3]);\n * const b = tf.tensor1d([-1, -2, -3]);\n *\n * a.where(cond, b).print();\n * ```\n *\n * @param condition The input condition. Must be of dtype bool.\n * @param a If `condition` is rank 1, `a` may have a higher rank but\n *     its first dimension must match the size of `condition`.\n * @param b A tensor with the same dtype as `a` and with shape that is\n *     compatible with `a`.\n * @return A tensor with same dtype as `a` and `b`, and shape that is\n *     broadcastable from `a` and `b`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction where_<T extends Tensor>(\n    condition: Tensor|TensorLike, a: T|TensorLike, b: T|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'where');\n  const $b = convertToTensor(b, 'b', 'where');\n  const $condition = convertToTensor(condition, 'condition', 'where', 'bool');\n  // TODO: move this logic to forward function when the broadcastTo op is\n  // implemented in WASM.\n  // Find the broadcastable shape for $a and $b.\n  const broadcastShape = assertAndGetBroadcastShape($a.shape, $b.shape);\n  const $broadcastedA = broadcastTo($a, broadcastShape);\n  const $broadcastedB = broadcastTo($b, broadcastShape);\n  if ($condition.rank === 1) {\n    // If condition rank is 1, then the first dimension must match the size of\n    // condition.\n    assert(\n        $condition.shape[0] === $a.shape[0],\n        () => 'The first dimension of `a` must match the size of `condition`.');\n  }\n\n  if ($condition.rank !== 1) {\n    // A must have the same shape as condition.\n    assertShapesMatch(\n        $condition.shape, $broadcastedB.shape, 'Error in where: ');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.select($condition, $broadcastedA, $broadcastedB);\n    save([$condition]);\n    return res;\n  };\n  const inputs: SelectV2Inputs = {\n    condition: $condition,\n    t: $broadcastedA,\n    e: $broadcastedB\n  };\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* gradient */,\n             SelectV2) as T;\n}\n\nexport const where = op({where_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {ZerosLike, ZerosLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.zerosLike(x).print();\n * ```\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction zerosLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'zerosLike');\n  const inputs: ZerosLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             backend => backend.zerosLike($x), inputs as {} as NamedTensorMap,\n             null /* grad */, ZerosLike) as T;\n}\nexport const zerosLike = op({zerosLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {div} from './div';\nimport {equal} from './equal';\nimport {op} from './operation';\nimport {where} from './where';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting. Return 0\n * if denominator is 0.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n * const c = tf.tensor1d([0, 0, 0, 0]);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n * const c = tf.scalar(0);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction divNoNan_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  // TODO: Make this into its own kernel.\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const divResult = div($a, $b);\n  const zeros = zerosLike(divResult);\n  const bEqualsZero = equal($b, zeros);\n  return where(bEqualsZero, zeros, divResult) as T;\n}\n\nexport const divNoNan = op({divNoNan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor,} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices and/or vectors, `t1` and `t2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor2d([[1, 2], [3, 4]]);\n * const c = tf.tensor2d([[1, 2, 3], [4, 5, 6]]);\n *\n * a.dot(b).print();  // or tf.dot(a, b)\n * b.dot(a).print();\n * b.dot(c).print();\n * ```\n * @param t1 The first tensor in the dot operation.\n * @param t2 The second tensor in the dot operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction dot_(t1: Tensor|TensorLike, t2: Tensor|TensorLike): Tensor {\n  const $t1 = convertToTensor(t1, 't1', 'dot');\n  const $t2 = convertToTensor(t2, 't2', 'dot');\n\n  util.assert(\n      ($t1.rank === 1 || $t1.rank === 2) && ($t2.rank === 1 || $t2.rank === 2),\n      () => `Error in dot: inputs must all be rank 1 or 2, but got ranks ` +\n          `${$t1.rank} and ${$t2.rank}.`);\n\n  const t1Inner = ($t1.rank === 1 ? $t1.size : $t1.shape[1]);\n  const t2Inner = ($t2.rank === 1 ? $t2.size : $t2.shape[0]);\n\n  util.assert(\n      t1Inner === t2Inner,\n      () => `Error in dot: inner dimensions of inputs must match, but got ` +\n          `${t1Inner} and ${t2Inner}.`);\n\n  if ($t1.rank === 1 && $t2.rank === 1) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, []);\n  } else if ($t1.rank === 1 && $t2.rank === 2) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else if ($t1.rank === 2 && $t2.rank === 1) {\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul($t1, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else {\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul($t1, t22D);\n    return t1t2;\n  }\n}\n\nexport const dot = op({dot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential linear element-wise: `x > 0 ? e ^ x - 1 : 0`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 1, -3, 2]);\n *\n * x.elu().print();  // or tf.elu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction elu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'elu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.elu($x);\n    save([y]);\n    return y;\n  };\n\n  const inputs: EluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Elu) as\n      T;\n}\n\nexport const elu = op({elu_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Erf, ErfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes gause error function of the input `tf.Tensor` element-wise:\n * `erf(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.erf().print(); // or tf.erf(x);\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction erf_<T extends Tensor>(x: T|TensorLike): T {\n  let $x = convertToTensor(x, 'x', 'erf');\n  util.assert(\n      $x.dtype === 'int32' || $x.dtype === 'float32',\n      () => 'Input dtype must be `int32` or `float32`.');\n\n  if ($x.dtype === 'int32') {\n    $x = cast($x, 'float32');\n  }\n\n  const inputs: ErfInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.erf($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Erf);\n}\nexport const erf = op({erf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Exp, ExpInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.exp().print();  // or tf.exp(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction exp_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'exp');\n\n  const inputs: ExpInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.exp($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Exp);\n}\nexport const exp = op({exp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension\n * into the tensor's shape.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const axis = 1;\n * x.expandDims(axis).print();\n * ```\n *\n * @param x The input tensor whose dimensions to be expanded.\n * @param axis The dimension index at which to insert shape of `1`. Defaults\n *     to 0 (the first dimension).\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction expandDims_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'expandDims', parseAs);\n\n  util.assert(axis <= $x.rank, () => 'Axis must be <= rank of the tensor');\n  const newShape = $x.shape.slice();\n  if (axis < 0) {\n    // Negative value is counted from the tail of rank.\n    util.assert(\n        -($x.rank + 1) <= axis,\n        () => `Axis must be in the interval [${- ($x.rank + 1)}, ${$x.rank}]`);\n    axis = $x.rank + axis + 1;\n  }\n  newShape.splice(axis, 0, 1);\n  return reshape($x, newShape) as T;\n}\n\nexport const expandDims = op({expandDims_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Expm1, Expm1Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` minus one element-wise.\n * `e ^ x - 1`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.expm1().print();  // or tf.expm1(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction expm1_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'expm1');\n\n  const inputs: Expm1Inputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.expm1($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Expm1);\n}\nexport const expm1 = op({expm1_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Tile, TileAttrs, TileInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Construct a tensor by repeating it the number of times given by reps.\n *\n * This operation creates a new tensor by replicating `input` `reps`\n * times. The output tensor's i'th dimension has `input.shape[i] *\n * reps[i]` elements, and the values of `input` are replicated\n * `reps[i]` times along the i'th dimension. For example, tiling\n * `[a, b, c, d]` by `[2]` produces `[a, b, c, d, a, b, c, d]`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n *\n * a.tile([2]).print();    // or a.tile([2])\n * ```\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.tile([1, 2]).print();  // or a.tile([1, 2])\n * ```\n * @param x The tensor to tile.\n * @param reps Determines the number of replications per dimension.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction tile_<T extends Tensor>(x: T|TensorLike, reps: number[]): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'tile', parseAs);\n  util.assert(\n      $x.rank === reps.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of reps ${reps}.`);\n\n  const forward: ForwardFunc<T> = (backend, save) => {\n    const res = backend.tile($x, reps);\n    save([$x]);\n    return res;\n  };\n\n  const inputsToSave = [$x];\n  const inputs: TileInputs = {x: $x};\n  const attrs: TileAttrs = {reps};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, Tile,\n      attrs as unknown as NamedAttrMap, inputsToSave);\n}\n\nexport const tile = op({tile_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {DataType} from '../types';\n\nimport {buffer} from './buffer';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {tile} from './tile';\n\n/**\n * Create an identity matrix.\n *\n * @param numRows Number of rows.\n * @param numColumns Number of columns. Defaults to `numRows`.\n * @param batchShape If provided, will add the batch shape to the beginning\n *   of the shape of the returned `tf.Tensor` by repeating the identity\n *   matrix.\n * @param dtype Data type.\n * @returns Identity matrix of the specified size and data type, possibly\n *   with batch repetition if `batchShape` is specified.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction eye_(\n    numRows: number, numColumns?: number,\n    batchShape?:\n        [\n          number\n        ]|[number,\n           number]|[number, number, number]|[number, number, number, number],\n    dtype: DataType = 'float32'): Tensor2D {\n  if (numColumns == null) {\n    numColumns = numRows;\n  }\n  const buff = buffer([numRows, numColumns], dtype);\n  const n = numRows <= numColumns ? numRows : numColumns;\n  for (let i = 0; i < n; ++i) {\n    buff.set(1, i, i);\n  }\n  const out: Tensor2D = reshape(buff.toTensor(), [numRows, numColumns]);\n  if (batchShape == null) {\n    return out;\n  } else {\n    if (batchShape.length === 1) {\n      return tile(expandDims(out, 0), [batchShape[0], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 2) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(\n                 expandDims(expandDims(out, 0), 0),\n                 [batchShape[0], batchShape[1], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 3) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(expandDims(expandDims(expandDims(out, 0), 0), 0), [\n               batchShape[0], batchShape[1], batchShape[2], 1, 1\n             ]) as Tensor2D;\n    } else {\n      throw new Error(\n          `eye() currently supports only 1D and 2D ` +\n          // tslint:disable-next-line:no-any\n          `batchShapes, but received ${(batchShape as any).length}D.`);\n    }\n  }\n}\n\nexport const eye = op({eye_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Fill, FillAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\n/**\n * Creates a `tf.Tensor` filled with a scalar value.\n *\n * ```js\n * tf.fill([2, 2], 4).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param value The scalar value to fill the tensor with.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n * 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction fill<R extends Rank>(\n    shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n  const attrs: FillAttrs = {shape, value, dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.fill(shape, value, dtype), {}, null, Fill,\n      attrs as {} as NamedAttrMap);\n}\n\nexport {fill};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Floor, FloorInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes floor of input `tf.Tensor` element-wise: `floor(x)`.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.floor().print();  // or tf.floor(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction floor_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'floor');\n\n  const inputs: FloorInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.floor($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Floor);\n}\nexport const floor = op({floor_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inputs of size above this threshold will be parallelized by calling multiple\n * shader programs.\n */\nimport {nearestDivisor} from '../util';\n\nexport const PARALLELIZE_THRESHOLD = 30;\n\nexport interface ReduceInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  outSize: number;\n}\n\nexport function computeOptimalWindowSize(inSize: number): number {\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    return inSize;\n  }\n  return nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {nearestDivisor} from '../util';\n\nimport {PARALLELIZE_THRESHOLD} from './reduce_util';\n\nexport interface SegOpInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  numSegments: number;\n}\n\nexport function segOpComputeOptimalWindowSize(\n    inSize: number, numSegments: number): number {\n  let done = false;\n  let res;\n\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    res = inSize;\n    done = true;\n  } else {\n    res = nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n  }\n\n  while (!done) {\n    if (res > numSegments || res === inSize) {\n      done = true;\n    } else {\n      res = nearestDivisor(inSize, res + 1);\n    }\n  }\n  return res;\n}\n\nexport function computeOutShape(\n    aShape: number[], axis: number, numSegments: number): number[] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (dim !== axis) {\n      outShape.push(aShape[dim]);\n    } else {\n      outShape.push(numSegments);\n    }\n  }\n  return outShape;\n}\n\nexport interface GatherOpShapeInfo {\n  batchSize: number;\n  sliceSize: number;\n  dimSize: number;\n  outputShape: number[];\n}\nexport function collectGatherOpShapeInfo(\n    x: Tensor, indices: Tensor, axis: number): GatherOpShapeInfo {\n  const dimSize = x.shape[axis];\n\n  const outputShape: number[] = [];\n  let batchSize = 1;\n  let sliceSize = 1;\n  for (let i = 0; i < axis; i++) {\n    outputShape.push(x.shape[i]);\n    batchSize *= x.shape[i];\n  }\n\n  for (let i = 0; i < indices.rank; i++) {\n    outputShape.push(indices.shape[i]);\n  }\n\n  for (let i = axis + 1; i < x.rank; i++) {\n    outputShape.push(x.shape[i]);\n    sliceSize *= x.shape[i];\n  }\n\n  return {batchSize, sliceSize, dimSize, outputShape};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherV2, GatherV2Attrs, GatherV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {collectGatherOpShapeInfo} from './segment_util';\n\n/**\n * Gather slices from tensor `x`'s axis `axis` according to `indices`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const indices = tf.tensor1d([1, 3, 3], 'int32');\n *\n * x.gather(indices).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const indices = tf.tensor1d([1, 1, 0], 'int32');\n *\n * x.gather(indices).print();\n * ```\n * @param x The input tensor whose slices to be gathered.\n * @param indices The indices of the values to extract.\n * @param axis The axis over which to select values. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction gather_<T extends Tensor>(\n    x: T|TensorLike, indices: Tensor|TensorLike, axis = 0): T {\n  const $x = convertToTensor(x, 'x', 'gather');\n  const $indices = convertToTensor(indices, 'indices', 'gather', 'int32');\n\n  const inputs: GatherV2Inputs = {x: $x, indices: $indices};\n  const attrs: GatherV2Attrs = {axis};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const parsedAxis = parseAxisParam(axis, $x.shape)[0];\n    const shapeInfo = collectGatherOpShapeInfo($x, $indices, parsedAxis);\n\n    const res =\n        backend.gather($x, reshape($indices, [$indices.size]), parsedAxis);\n    save([$x, $indices]);\n\n    return reshape(res, shapeInfo.outputShape);\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, GatherV2,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const gather = op({gather_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Greater, GreaterInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a > b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greater(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greater_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greater');\n  let $b = convertToTensor(b, 'b', 'greater');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.greater($a, $b);\n\n  const inputs: GreaterInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             Greater) as T;\n}\n\nexport const greater = op({greater_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GreaterEqual, GreaterEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a >= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greaterEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greaterEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greaterEqual');\n  let $b = convertToTensor(b, 'b', 'greaterEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.greaterEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: GreaterEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             GreaterEqual) as T;\n}\n\nexport const greaterEqual = op({greaterEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Imag, ImagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n/**\n * Returns the imaginary part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the imaginary part of each element in input considered as a complex number.\n * If input is real, a tensor of all zeros is returned.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.imag(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction imag_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'imag');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.imag($input);\n  };\n\n  const inputs: ImagInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Imag) as T;\n}\n\nexport const imag = op({imag_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsFinite, IsFiniteInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are finite.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isFinite().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isFinite_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isFinite');\n\n  const inputs: IsFiniteInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isFinite($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsFinite);\n}\nexport const isFinite = op({isFinite_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsInf, IsInfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are Infinity or -Infinity.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isInf().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isInf_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isInf');\n\n  const inputs: IsInfInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isInf($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsInf);\n}\nexport const isInf = op({isInf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsNan, IsNanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * RReturns which elements of x are NaN.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isNaN().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isNaN_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isNaN');\n  const inputs: IsNanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.isNaN($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsNan);\n}\nexport const isNaN = op({isNaN_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Maximum, MaximumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the max of a and b (`a > b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `tf.maximumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * ```js\n * // Broadcast maximum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction maximum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'maximum');\n  let $b = convertToTensor(b, 'b', 'maximum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.maximum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: MaximumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Maximum) as T;\n}\n\nexport const maximum = op({maximum_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar} from '../tensor';\nimport {DataType} from '../types';\nimport {isTypedArray} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-0 `tf.Tensor` (scalar) with the provided value and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.scalar` as it makes the code more readable.\n *\n * ```js\n * tf.scalar(3.14).print();\n * ```\n *\n * @param value The value of the scalar.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function scalar(\n    value: number|boolean|string|Uint8Array, dtype?: DataType): Scalar {\n  if (((isTypedArray(value) && dtype !== 'string') || Array.isArray(value)) &&\n      dtype !== 'complex64') {\n    throw new Error(\n        'Error creating a new Scalar: value must be a primitive ' +\n        '(number|boolean|string)');\n  }\n  if (dtype === 'string' && isTypedArray(value) &&\n      !(value instanceof Uint8Array)) {\n    throw new Error(\n        'When making a scalar from encoded string, ' +\n        'the value must be `Uint8Array`.');\n  }\n  const shape: number[] = [];\n  const inferredShape: number[] = [];\n  return makeTensor(value, shape, inferredShape, dtype) as Scalar;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {maximum} from './maximum';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {scalar} from './scalar';\n\n/**\n * Computes leaky rectified linear element-wise.\n *\n * See\n * [http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf](\n *     http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf)\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.leakyRelu(0.1).print();  // or tf.leakyRelu(x, 0.1)\n * ```\n * @param x The input tensor.\n * @param alpha The scaling factor for negative values, defaults to 0.2.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction leakyRelu_<T extends Tensor>(x: T|TensorLike, alpha = 0.2): T {\n  const $x = convertToTensor(x, 'x', 'leakyRelu');\n  return maximum(mul(scalar(alpha), $x), $x);\n}\n\nexport const leakyRelu = op({leakyRelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Less, LessInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a < b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.less(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction less_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'less');\n  let $b = convertToTensor(b, 'b', 'less');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.less($a, $b);\n\n  const inputs: LessInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Less) as\n      T;\n}\n\nexport const less = op({less_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LessEqual, LessEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a <= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.lessEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction lessEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'lessEqual');\n  let $b = convertToTensor(b, 'b', 'lessEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.lessEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: LessEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LessEqual) as T;\n}\n\nexport const lessEqual = op({lessEqual_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LinSpace, LinSpaceAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor1D} from '../tensor';\n\n/**\n * Return an evenly spaced sequence of numbers over the given interval.\n *\n * ```js\n * tf.linspace(0, 9, 10).print();\n * ```\n * @param start The start value of the sequence.\n * @param stop The end value of the sequence.\n * @param num The number of values to generate.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function linspace(start: number, stop: number, num: number): Tensor1D {\n  if (num <= 0) {\n    throw new Error('The number of values should be positive.');\n  }\n\n  const attrs: LinSpaceAttrs = {start, stop, num};\n  return ENGINE.runKernelFunc(\n      backend => backend.linspace(start, stop, num), {} /* inputs */,\n      null /* grad */, LinSpace, attrs as {} as NamedAttrMap);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRN, LRNAttrs, LRNInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Normalizes the activation of a local neighborhood across or within\n * channels.\n *\n * @param x The input tensor. The 4-D input tensor is treated as a 3-D array\n *     of 1D vectors (along the last dimension), and each vector is\n *     normalized independently.\n * @param depthRadius The number of adjacent channels in the 1D normalization\n *     window.\n * @param bias A constant bias term for the basis.\n * @param alpha A scale factor, usually positive.\n * @param beta An exponent.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction localResponseNormalization_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const $x = convertToTensor(x, 'x', 'localResponseNormalization');\n  util.assert(\n      $x.rank === 4 || $x.rank === 3,\n      () => `Error in localResponseNormalization: x must be rank 3 or 4 but got\n               rank ${$x.rank}.`);\n  util.assert(\n      util.isInt(depthRadius),\n      () => `Error in localResponseNormalization: depthRadius must be an ` +\n          `integer but got depthRadius ${depthRadius}.`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.localResponseNormalization4D(\n        x4D, depthRadius, bias, alpha, beta);\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: LRNInputs = {x: x4D};\n\n  const attrs: LRNAttrs = {depthRadius, bias, alpha, beta};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, LRN,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  } else {\n    return res as T;\n  }\n}\n\nexport const localResponseNormalization = op({localResponseNormalization_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log, LogInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` element-wise: `ln(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E]);\n *\n * x.log().print();  // or tf.log(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log');\n\n  const inputs: LogInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log);\n}\nexport const log = op({log_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log1p, Log1pInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` plus one\n * element-wise: `ln(1 + x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E - 1]);\n *\n * x.log1p().print();  // or tf.log1p(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log1p_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log1p');\n\n  const inputs: Log1pInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log1p($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log1p);\n}\nexport const log1p = op({log1p_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {CustomGradientFunc, ENGINE} from './engine';\nimport {Scalar, Tensor, Variable} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {convertToTensor, convertToTensorArray} from './tensor_util_env';\nimport {TensorLike} from './types';\nimport * as util from './util';\n\n/**\n * Provided `f(x)`, returns another function `g(x, dy?)`, which gives the\n * gradient of `f(x)` with respect to `x`.\n *\n * If `dy` is provided, the gradient of `f(x).mul(dy).sum()` with respect to\n * `x` is computed instead. `f(x)` must take a single tensor `x` and return a\n * single tensor `y`. If `f()` takes multiple inputs, use `tf.grads` instead.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.grad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * g(x).print();\n * ```\n *\n * ```js\n * // f(x) = x ^ 3\n * const f = x => x.pow(tf.scalar(3, 'int32'));\n * // f'(x) = 3x ^ 2\n * const g = tf.grad(f);\n * // f''(x) = 6x\n * const gg = tf.grad(g);\n *\n * const x = tf.tensor1d([2, 3]);\n * gg(x).print();\n * ```\n *\n * @param f The function f(x), to compute gradient for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grad(f: (x: Tensor) => Tensor): (\n    x: TensorLike|Tensor, dy?: TensorLike|Tensor) => Tensor {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grad(f) must be a function');\n  return (x: TensorLike|Tensor, dy?: TensorLike|Tensor): Tensor => {\n    // x can be of any dtype, thus null as the last argument.\n    const $x = convertToTensor(x, 'x', 'tf.grad', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grad') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f($x), [$x], $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grad(f)(x, dy) must match the shape ' +\n                'returned by f(x)');\n      }\n      checkGrads(grads);\n      return grads[0];\n    });\n  };\n}\n\n/**\n * Provided `f(x1, x2,...)`, returns another function `g([x1, x2,...], dy?)`,\n * which gives an array of gradients of `f()` with respect to each input\n * [`x1`,`x2`,...].\n *\n * If `dy` is passed when calling `g()`, the gradient of\n * `f(x1,...).mul(dy).sum()` with respect to each input is computed instead.\n * The provided `f` must take one or more tensors and return a single tensor\n * `y`. If `f()` takes a single input, we recommend using `tf.grad` instead.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df / da = b, df / db = a\n * const g = tf.grads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const [da, db] = g([a, b]);\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @param f The function `f(x1, x2,...)` to compute gradients for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grads(f: (...args: Tensor[]) => Tensor): (\n    args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike) => Tensor[] {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grads(f) must be a function');\n  return (args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike): Tensor[] => {\n    util.assert(\n        Array.isArray(args),\n        () => 'The args passed in grads(f)(args) must be an array ' +\n            'of `Tensor`s or `TensorLike`s');\n    // args can be of any dtype, thus null as the last argument.\n    const $args = convertToTensorArray(args, 'args', 'tf.grads', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grads') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f(...$args), $args, $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grads(f)([x1,...], dy) must ' +\n                'match the shape returned by f([x1,...])');\n      }\n      checkGrads(grads);\n      return grads;\n    });\n  };\n}\n\n/**\n * Like `tf.grad`, but also returns the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grad: The gradient of `f(x)` w.r.t `x` (result of `tf.grad`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.valueAndGrad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * const {value, grad} = g(x);\n *\n * console.log('value');\n * value.print();\n * console.log('grad');\n * grad.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrad<I extends Tensor, O extends Tensor>(f: (x: I) => O): (\n    x: I, dy?: O) => {\n  value: O;\n  grad: I;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrad(f) must be a function');\n  return (x: I, dy?: O) => {\n    util.assert(\n        x instanceof Tensor,\n        () => 'The x passed in valueAndGrad(f)(x) must be a tensor');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrad(f)(x, dy) must be a tensor');\n    const {grads, value} = ENGINE.gradients(() => f(x), [x], dy);\n    checkGrads(grads);\n    return {grad: grads[0] as I, value};\n  };\n}\n\n/**\n * Like `tf.grads`, but returns also the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grads: The gradients of `f()` w.r.t each input (result of `tf.grads`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df/da = b, df/db = a\n * const g = tf.valueAndGrads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const {value, grads} = g([a, b]);\n *\n * const [da, db] = grads;\n *\n * console.log('value');\n * value.print();\n *\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrads<O extends Tensor>(f: (...args: Tensor[]) => O): (\n    args: Tensor[], dy?: O) => {\n  grads: Tensor[];\n  value: O;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrads(f) must be a function');\n  return (args: Tensor[], dy?: O) => {\n    util.assert(\n        Array.isArray(args) && args.every(arg => arg instanceof Tensor),\n        () => 'The args passed in valueAndGrads(f)(args) must be array of ' +\n            'tensors');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrads(f)(args, dy) must be a tensor');\n    const res = ENGINE.gradients(() => f(...args), args, dy);\n    if (dy != null) {\n      util.assertShapesMatch(\n          res.value.shape, dy.shape,\n          'The shape of dy passed in valueAndGrads(f)([x1,...], dy) must ' +\n              'match the shape returned by f([x1,...])');\n    }\n    checkGrads(res.grads);\n    return res;\n  };\n}\n\n/**\n * Computes and returns the gradient of f(x) with respect to the list of\n * trainable variables provided by `varList`. If no list is provided, it\n * defaults to all trainable variables.\n *\n * ```js\n * const a = tf.variable(tf.tensor1d([3, 4]));\n * const b = tf.variable(tf.tensor1d([5, 6]));\n * const x = tf.tensor1d([1, 2]);\n *\n * // f(a, b) = a * x ^ 2 + b * x\n * const f = () => a.mul(x.square()).add(b.mul(x)).sum();\n * // df/da = x ^ 2, df/db = x\n * const {value, grads} = tf.variableGrads(f);\n *\n * Object.keys(grads).forEach(varName => grads[varName].print());\n * ```\n *\n * @param f The function to execute. f() should return a scalar.\n * @param varList The list of variables to compute the gradients with respect\n *     to. Defaults to all trainable variables.\n * @returns An object with the following keys and values:\n *   - `value`: The value of the function `f`.\n *   - `grads`: A map from the names of the variables to the gradients.\n *     If the `varList` argument is provided explicitly and contains a subset of\n *     non-trainable variables, this map in the return value will contain keys\n *     that map the names of the non-trainable variables to `null`.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction variableGrads(f: () => Scalar, varList?: Variable[]):\n    {value: Scalar, grads: NamedTensorMap} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in variableGrads(f) must be a function');\n  util.assert(\n      varList == null ||\n          Array.isArray(varList) && varList.every(v => v instanceof Variable),\n      () =>\n          'The varList passed in variableGrads(f, varList) must be an array ' +\n          'of variables');\n\n  const specifiedVarList = varList != null;\n  if (!specifiedVarList) {\n    // Get all of the trainable variables.\n    varList = [];\n    for (const varName in ENGINE.registeredVariables) {\n      varList.push(ENGINE.registeredVariables[varName]);\n    }\n  }\n\n  const specifiedNonTrainable: Variable[] =\n      specifiedVarList ? varList.filter(variable => !variable.trainable) : null;\n\n  // Prune non-trainable variables.\n  const originalVarCount = varList.length;\n  varList = varList.filter(variable => variable.trainable);\n  util.assert(\n      varList.length > 0,\n      () => `variableGrads() expects at least one of the input variables to ` +\n          `be trainable, but none of the ${originalVarCount} variables is ` +\n          `trainable.`);\n\n  const allowNoGradients = true;\n  const {value, grads} = ENGINE.gradients(f, varList, null, allowNoGradients);\n\n  util.assert(\n      grads.some(g => g != null),\n      () => 'Cannot find a connection between any variable and the result of ' +\n          'the loss function y=f(x). Please make sure the operations that ' +\n          'use variables are inside the function f passed to minimize().');\n  util.assert(\n      value.rank === 0,\n      () => `The f passed in variableGrads(f) must return a scalar, but it ` +\n          `returned a rank-${value.rank} tensor`);\n\n  const namedGrads: NamedTensorMap = {};\n  varList.forEach((v, i) => {\n    if (grads[i] != null) {\n      namedGrads[v.name] = grads[i];\n    }\n  });\n  if (specifiedNonTrainable != null) {\n    // If varList is explicitly provided and contains non-trainable values,\n    // add them to the returned gradients with `null` values.\n    specifiedNonTrainable.forEach(v => namedGrads[v.name] = null);\n  }\n  return {value, grads: namedGrads};\n}\n\n/**\n * Overrides the gradient computation of a function `f`.\n *\n * Takes a function\n * `f(...inputs, save) => {value: Tensor, gradFunc: (dy, saved) => Tensor[]}`\n * and returns another function `g(...inputs)` which takes the same inputs as\n * `f`. When called, `g` returns `f().value`. In backward mode, custom gradients\n * with respect to each input of `f` are computed using `f().gradFunc`.\n *\n * The `save` function passsed to `f` should be used for saving tensors needed\n * in the gradient. And the `saved` passed to the `gradFunc` is a\n * `NamedTensorMap`, which contains those saved tensor.\n *\n * ```js\n * const customOp = tf.customGrad((x, save) => {\n *   // Save x to make sure it's available later for the gradient.\n *   save([x]);\n *   // Override gradient of our custom x ^ 2 op to be dy * abs(x);\n *   return {\n *     value: x.square(),\n *     // Note `saved.x` which points to the `x` we saved earlier.\n *     gradFunc: (dy, saved) => [dy.mul(saved[0].abs())]\n *   };\n * });\n *\n * const x = tf.tensor1d([-1, -2, 3]);\n * const dx = tf.grad(x => customOp(x));\n *\n * console.log(`f(x):`);\n * customOp(x).print();\n * console.log(`f'(x):`);\n * dx(x).print();\n * ```\n *\n * @param f The function to evaluate in forward mode, which should return\n *     `{value: Tensor, gradFunc: (dy, saved) => Tensor[]}`, where `gradFunc`\n *     returns the custom gradients of `f` with respect to its inputs.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n    (...args: Tensor[]) => T {\n  return ENGINE.customGrad(f);\n}\n\nfunction checkGrads(grads: Tensor[]) {\n  const numNullGradients = grads.filter(g => g == null).length;\n  if (numNullGradients > 0) {\n    throw new Error(\n        `Cannot compute gradient of y=f(x) with respect to x. Make sure that\n    the f you passed encloses all operations that lead from x to y.`);\n  }\n}\n\nexport {\n  customGrad,\n  variableGrads,\n  valueAndGrad,\n  valueAndGrads,\n  grad,\n  grads,\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Negate, NegateInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes `-1 * x` element-wise.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, -2, 0], [2, 2]);\n *\n * x.neg().print();  // or tf.neg(x)\n * ```\n *\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction neg_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'neg');\n\n  const inputs: NegateInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.neg($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Negate);\n}\nexport const neg = op({neg_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softplus, SoftplusInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes softplus of the input `tf.Tensor` element-wise: `log(exp(x) + 1)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.softplus().print();  // or tf.softplus(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction softplus_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'softplus');\n\n  const inputs: SoftplusInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.softplus($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Softplus);\n}\nexport const softplus = op({softplus_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {mul} from './mul';\nimport {neg} from './neg';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {softplus} from './softplus';\n\n/**\n * Computes log sigmoid of the input `tf.Tensor` element-wise:\n * `logSigmoid(x)`. For numerical stability, we use `-tf.softplus(-x)`.\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.logSigmoid().print();  // or tf.logSigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction logSigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logSigmoid');\n\n  // Use a custom gradient to maintain previous implementation.\n  // There is no LogSigmoid kernel in TF so we can't use engine.runKernel\n  // directly\n  const customOp = customGrad((x: Tensor) => {\n    // TODO(yassogba) we can remove the chained softplus call here only\n    // after backends have modualrized softplus at which point we can call\n    // engine runKernel(..., Sotfplus, ...) directly.\n    const value = neg(softplus(neg(x)));\n\n    const gradFunc = (dy: T) => {\n      const derX = mul(dy, sigmoid(neg(x)));\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\nexport const logSigmoid = op({logSigmoid_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Max, MaxAttrs, MaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the maximum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.max().print();  // or tf.max(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.max(axis).print();  // or tf.max(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction max_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'max');\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = util.parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let maxInput = $x;\n        if (permutedAxes != null) {\n          maxInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, maxInput.rank);\n        }\n\n        const y = backend.max(maxInput, axes);\n        if (permutedAxes != null) {\n          maxInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape = axis_util.expandShapeToKeepDim(\n              res.shape, util.parseAxisParam(axis, $x.shape));\n          res = reshape(res, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MaxInputs = {x: $x};\n  const attrs: MaxAttrs = {reductionIndices: axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Max,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const max = op({max_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sub, SubInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Subtracts two `tf.Tensor`s element-wise, A - B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n *\n * ```js\n * // Broadcast subtract a with b.\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.scalar(5);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n * @param a The first `tf.Tensor` to subtract from.\n * @param b The second `tf.Tensor` to be subtracted. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction sub_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'sub');\n  let $b = convertToTensor(b, 'b', 'sub');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.subtract($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SubInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sub) as\n      T;\n}\n\nexport const sub = op({sub_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sum, SumAttrs, SumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the sum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If axes has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.sum().print();  // or tf.sum(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.sum(axis).print();  // or tf.sum(x, axis)\n * ```\n *\n * @param x The input tensor to compute the sum over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction sum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'sum');\n  if ($x.dtype === 'bool') {\n    $x = cast($x, 'int32');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.sum(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n    return value;\n  };\n\n  const inputs: SumInputs = {x: $x};\n  const attrs: SumAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const sum = op({sum_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LogSoftmax, LogSoftmaxAttrs, LogSoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log softmax.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param axis The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction logSoftmax_<T extends Tensor>(logits: T|TensorLike, axis = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'logSoftmax');\n\n  if (axis === -1) {\n    axis = $logits.rank - 1;\n  }\n  if (axis !== $logits.rank - 1) {\n    throw Error(\n        'Log Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and axis was ${axis}`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const keepDims = true;\n    const xMax = max(logits, axis, true);\n    const shifted = sub(logits, xMax);\n    const value =\n        sub(cast(shifted, 'float32'), log(sum(exp(shifted), axis, keepDims)));\n    save([value]);\n    return value;\n  };\n\n  const inputs: LogSoftmaxInputs = {logits: $logits};\n  const attrs: LogSoftmaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LogSoftmax, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const logSoftmax = op({logSoftmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {add} from './add';\nimport {expandShapeToKeepDim} from './axis_util';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log(sum(exp(elements across the reduction dimensions)).\n *\n * Reduces the input along the dimensions given in `axis`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.logSumExp().print();  // or tf.logSumExp(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.logSumExp(axis).print();  // or tf.logSumExp(a, axis)\n * ```\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. If null (the default),\n *     reduces all dimensions.\n * @param keepDims If true, retains reduced dimensions with length\n *     of 1. Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction logSumExp_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'logSumExp');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const xMax = max($x, axes, true /* keepDims */);\n  const a = sub($x, xMax);\n  const b = exp(a);\n  const c = sum(b, axes);\n  const d = log(c);\n  const res = add(reshape(xMax, d.shape), d);\n\n  if (keepDims) {\n    const newShape = expandShapeToKeepDim(res.shape, axes);\n    return reshape(res, newShape) as T;\n  }\n  return res as T;\n}\n\nexport const logSumExp = op({logSumExp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {LogicalAnd, LogicalAndInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a AND b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalAnd(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalAnd_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalAnd', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalAnd', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalAndInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalAnd($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalAnd) as T;\n}\n\nexport const logicalAnd = op({logicalAnd_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalNot, LogicalNotInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `NOT x` element-wise.\n *\n * ```js\n * const a = tf.tensor1d([false, true], 'bool');\n *\n * a.logicalNot().print();\n * ```\n *\n * @param x The input tensor. Must be of dtype 'bool'.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalNot_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logicalNot', 'bool');\n  const inputs: LogicalNotInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.logicalNot($x), inputs as {} as NamedTensorMap,\n      null /* grad */, LogicalNot);\n}\n\nexport const logicalNot = op({logicalNot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalOr, LogicalOrInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a OR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalOr(b).print();\n * ```\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalOr_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalOr', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalOr', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalOrInputs = {a: $a, b: $b};\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalOr($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalOr) as T;\n}\nexport const logicalOr = op({logicalOr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {logicalAnd} from './logical_and';\nimport {logicalNot} from './logical_not';\nimport {logicalOr} from './logical_or';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a XOR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalXor(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalXor_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalXor', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalXor', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  // x ^ y = (x | y) & ~(x & y)\n  return logicalAnd(logicalOr(a, b), logicalNot(logicalAnd(a, b)));\n}\n\nexport const logicalXor = op({logicalXor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool, MaxPoolAttrs, MaxPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D max pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction maxPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'maxPool');\n  const dilations = 1;\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in maxPool: input must be rank 4 but got rank ${x4D.rank}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    let y;\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      y = x4D.clone();\n    } else {\n      y = backend.maxPool(x4D, convInfo);\n    }\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: MaxPoolInputs = {x: x4D};\n\n  const attrs: MaxPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const maxPool = op({maxPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {MaxPool3D, MaxPool3DAttrs, MaxPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D max pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.maxPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number = [1, 1, 1],\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'maxPool3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in maxPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in maxPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n    const y = backend.maxPool3d(x5D, convInfo);\n    save([x5D, y]);\n    return y;\n  };\n\n  const inputs: MaxPool3DInputs = {x: x5D};\n\n  const attrs: MaxPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3d = op({maxPool3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the 2D max pooling of an image with Argmax index.\n * The indices in argmax are flattened, so that a maximum value at position `[b,\n * y, x, c]` becomes flattened index: `(y * width + x) * channels + c` if\n * include_batch_in_index is False; `((b * height + y) * width + x) * channels\n * +c` if include_batch_in_index is True.\n *\n * The indices returned are always in `[0, height) x [0, width)` before\n * flattening.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param includeBatchIndex Defaults to False. Whether to include batch\n *    dimension in flattened index of argmax.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPoolWithArgmax_<T extends Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    includeBatchInIndex = false): NamedTensorMap {\n  const $x = convertToTensor(x, 'x', 'maxPoolWithArgmax');\n\n  const inputs: MaxPoolWithArgmaxInputs = {x: $x};\n  const attrs:\n      MaxPoolWithArgmaxAttrs = {filterSize, strides, pad, includeBatchInIndex};\n\n  const result = ENGINE.runKernel(\n                     MaxPoolWithArgmax, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {result: result[0], indexes: result[1]};\n}\n\nexport const maxPoolWithArgmax = op({maxPoolWithArgmax_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeZerosTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0.\n *\n * ```js\n * tf.zeros([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Can\n *     be 'float32', 'int32' or 'bool'. Defaults to 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function zeros<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = zeros(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeZerosTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeOnesTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\nimport {zeros} from './zeros';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1.\n *\n * ```js\n * tf.ones([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n *     'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function ones<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = ones(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeOnesTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam, sizeFromShape} from '../util';\n\nimport {computeOutAndReduceShapes} from './axis_util';\nimport {cast} from './cast';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {ones} from './ones';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sum} from './sum';\n\n/**\n * Computes the mean of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces `x` along the dimensions given in `axis`. Unless `keepDims` is\n * true, the rank of the `tf.Tensor` is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and a `tf.Tensor` with\n * a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.mean().print();  // or tf.mean(a)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.mean(axis).print();  // or tf.mean(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction mean_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'mean');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const shapes = computeOutAndReduceShapes($x.shape, axes);\n  const reduceShape = shapes[1];\n  const reduceSize = sizeFromShape(reduceShape);\n\n  // Use a custom gradient to bypass 2 gradient backprops since mean is used\n  // extremely often.\n  const customOp = customGrad((x: Tensor) => {\n    const reduceSizeScalar = scalar(reduceSize);\n    // Cast if needed.\n    const xReduce = reduceSizeScalar.dtype === x.dtype ?\n        x :\n        cast(x, reduceSizeScalar.dtype);\n    const res = div(xReduce, reduceSizeScalar);\n    const value = sum(res, axis, keepDims);\n\n    const gradFunc = (dy: Tensor) => {\n      const expandedDyShape = x.shape.slice();\n      axes.forEach(axis => {\n        expandedDyShape[axis] = 1;\n      });\n      const expandedDy = reshape(dy, expandedDyShape);\n      const derX = div(mul(expandedDy, ones(x.shape, 'float32')), reduceSize);\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\n\nexport const mean = op({mean_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Min, MinAttrs, MinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the minimum value from the input.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axes`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axes` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.min().print();  // or tf.min(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.min(axis).print();  // or tf.min(x, axis)\n * ```\n *\n * @param x The input Tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction min_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'min');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let minInput = $x;\n        if (permutedAxes != null) {\n          minInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n        }\n\n        const y = backend.min(minInput, axes);\n        if (permutedAxes != null) {\n          minInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape =\n              axis_util.expandShapeToKeepDim(res.shape, origAxes);\n          res = reshape(y, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MinInputs = {x: $x};\n  const attrs: MinAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Min,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const min = op({min_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Minimum, MinimumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the min of a and b (`a < b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `minimumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * ```js\n * // Broadcast minimum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction minimum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'minimum');\n  let $b = convertToTensor(b, 'b', 'minimum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.minimum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MinimumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Minimum) as T;\n}\n\nexport const minimum = op({minimum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Mod, ModInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns the mod of a and b element-wise.\n * `floor(x / y) * y + mod(x, y) = x`\n * Supports broadcasting.\n *\n * We also expose `tf.modStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * ```js\n * // Broadcast a mod b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mod_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mod');\n  let $b = convertToTensor(b, 'b', 'mod');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.mod($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: ModInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Mod) as T;\n}\n\nexport const mod = op({mod_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Computes square of `x` element-wise: `x ^ 2`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.sqrt(2), -1]);\n *\n * x.square().print();  // or tf.square(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction square_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'square');\n  const attrs = {};\n  const inputsToSave = [$x];\n  const outputsToSave: boolean[] = [];\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    return backend.square($x);\n  }, {x: $x}, null /* grad */, 'Square', attrs, inputsToSave, outputsToSave);\n}\n\nexport const square = op({square_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim} from './axis_util';\nimport {cast} from './cast';\nimport {mean} from './mean';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {square} from './square';\nimport {sub} from './sub';\n\n/**\n * Calculates the mean and variance of `x`. The mean and variance are\n * calculated by aggregating the contents of `x` across `axes`. If `x` is\n * 1-D and `axes = [0]` this is just the mean and variance of a vector.\n *\n * @param x The input tensor.\n * @param axis The dimension(s) along with to compute mean and\n *     variance. By default it reduces all dimensions.\n * @param keepDims If true, the moments have the same dimensionality as the\n *     input.\n * @return An object with two keys: `mean` and `variance`.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction moments_(\n    x: Tensor|TensorLike, axis: number|number[] = null,\n    keepDims = false): {mean: Tensor, variance: Tensor} {\n  x = convertToTensor(x, 'x', 'moments');\n  const axes = parseAxisParam(axis, x.shape);\n  const xMean = mean(x, axes, keepDims);\n  let keepDimsShape = xMean.shape;\n  if (!keepDims) {\n    keepDimsShape = expandShapeToKeepDim(xMean.shape, axes);\n  }\n  const devSquared =\n      square(sub(cast(x, 'float32'), reshape(xMean, keepDimsShape)));\n  const variance = mean(devSquared, axes, keepDims);\n  return {mean: xMean, variance};\n}\n\nexport const moments = op({moments_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor, convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * @docalias (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D]\n */\nexport type LSTMCellFunc = {\n  (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D];\n};\n\n/**\n * Computes the next states and outputs of a stack of LSTMCells.\n *\n * Each cell output is used as input to the next cell.\n *\n * Returns `[cellState, cellOutput]`.\n *\n * Derived from tf.contrib.rn.MultiRNNCell.\n *\n * @param lstmCells Array of LSTMCell functions.\n * @param data The input to the cell.\n * @param c Array of previous cell states.\n * @param h Array of previous cell outputs.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction multiRNNCell_(\n    lstmCells: LSTMCellFunc[], data: Tensor2D|TensorLike,\n    c: Array<Tensor2D|TensorLike>,\n    h: Array<Tensor2D|TensorLike>): [Tensor2D[], Tensor2D[]] {\n  const $data = convertToTensor(data, 'data', 'multiRNNCell');\n  const $c = convertToTensorArray(c, 'c', 'multiRNNCell');\n  const $h = convertToTensorArray(h, 'h', 'multiRNNCell');\n\n  let input = $data;\n  const newStates = [];\n  for (let i = 0; i < lstmCells.length; i++) {\n    const output = lstmCells[i](input, $c[i], $h[i]);\n    newStates.push(output[0]);\n    newStates.push(output[1]);\n    input = output[1];\n  }\n  const newC: Tensor2D[] = [];\n  const newH: Tensor2D[] = [];\n  for (let i = 0; i < newStates.length; i += 2) {\n    newC.push(newStates[i]);\n    newH.push(newStates[i + 1]);\n  }\n  return [newC, newH];\n}\nexport const multiRNNCell = op({multiRNNCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a `tf.Tensor` with values drawn from a multinomial distribution.\n *\n * ```js\n * const probs = tf.tensor([.75, .25]);\n * tf.multinomial(probs, 3).print();\n * ```\n *\n * @param logits 1D array with unnormalized log-probabilities, or\n *     2D array of shape `[batchSize, numOutcomes]`. See the `normalized`\n *     parameter.\n * @param numSamples Number of samples to draw for each row slice.\n * @param seed The seed number.\n * @param normalized Whether the provided `logits` are normalized true\n *     probabilities (sum to 1). Defaults to false.\n * @return 1D array of shape `[numSamples]`, or 2D array of shape\n *     `[batchSize, numSamples]`, depending on the rank of the input.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction multinomial_(\n    logits: Tensor1D|Tensor2D|TensorLike, numSamples: number, seed?: number,\n    normalized = false): Tensor1D|Tensor2D {\n  const $logits = convertToTensor(logits, 'logits', 'multinomial');\n  const numOutcomes = $logits.size;\n  const origRank = $logits.rank;\n  if (numOutcomes < 2) {\n    throw new Error(\n        `Error in multinomial: you need at least 2 outcomes, but got ` +\n        `${numOutcomes}.`);\n  }\n  if (origRank > 2) {\n    throw new Error(`Rank of probabilities must be 1 or 2, but is ${origRank}`);\n  }\n  seed = seed || Math.random();\n  const logits2D: Tensor2D =\n      origRank === 1 ? reshape($logits, [1, -1]) : $logits as Tensor2D;\n  const res = ENGINE.runKernelFunc(\n      backend => backend.multinomial(logits2D, normalized, numSamples, seed),\n      {logits2D});\n  // tslint:disable-next-line:no-unnecessary-type-assertion\n  return origRank === 1 ? reshape(res, [res.size]) as Tensor1D : res;\n}\n\nexport const multinomial = op({multinomial_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {NotEqual, NotEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a != b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([0, 2, 3]);\n *\n * a.notEqual(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction notEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'notEqual');\n  let $b = convertToTensor(b, 'b', 'notEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => backend.notEqual($a, $b);\n\n  const inputs: NotEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             NotEqual) as T;\n}\n\nexport const notEqual = op({notEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Real, RealInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the real part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the real part of each element in input considered as a complex number.\n *\n * If the input is real, it simply makes a clone.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.real(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction real_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'real');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.real($input);\n  };\n\n  const inputs: RealInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Real) as T;\n}\n\nexport const real = op({real_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OnesLike, OnesLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {complex} from './complex';\nimport {imag} from './imag';\nimport {op} from './operation';\nimport {real} from './real';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.onesLike(x).print();\n * ```\n * @param x A tensor.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction onesLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'onesLike');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if ($x.dtype === 'complex64') {\n      const r = onesLike(real($x));\n      const i = zerosLike(imag($x));\n      return complex(r, i);\n    }\n\n    return backend.onesLike($x);\n  };\n\n  const inputs: OnesLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             OnesLike) as T;\n}\n\nexport const onesLike = op({onesLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the outer product of two vectors, `v1` and `v2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([3, 4, 5]);\n *\n * tf.outerProduct(a, b).print();\n * ```\n * @param v1 The first vector in the outer product operation.\n * @param v2 The second vector in the outer product operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction outerProduct_(\n    v1: Tensor1D|TensorLike, v2: Tensor1D|TensorLike): Tensor2D {\n  const $v1 = convertToTensor(v1, 'v1', 'outerProduct');\n  const $v2 = convertToTensor(v2, 'v2', 'outerProduct');\n\n  util.assert(\n      $v1.rank === 1 && $v2.rank === 1,\n      () => `Error in outerProduct: inputs must be rank 1, but got ranks ` +\n          `${$v1.rank} and ${$v2.rank}.`);\n\n  const v12D = reshape($v1, [-1, 1]);\n  const v22D = reshape($v2, [1, -1]);\n  return matMul(v12D, v22D) as Tensor2D;\n}\n\nexport const outerProduct = op({outerProduct_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {PadV2, PadV2Attrs, PadV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Pads a `tf.Tensor` with a given value and paddings.\n *\n * This operation currently only implements the `CONSTANT` mode.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `paddings` is of given length.\n *   - `tf.pad1d`\n *   - `tf.pad2d`\n *   - `tf.pad3d`\n *   - `tf.pad4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.pad([[1, 2]]).print();\n * ```\n * @param x The tensor to pad.\n * @param paddings An array of length `R` (the rank of the tensor), where\n * each element is a length-2 tuple of ints `[padBefore, padAfter]`,\n * specifying how much to pad along each dimension of the tensor.\n * @param constantValue The pad value to use. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction pad_<T extends Tensor>(\n    x: T|TensorLike, paddings: Array<[number, number]>, constantValue = 0): T {\n  const $x = convertToTensor(x, 'x', 'pad');\n  if ($x.rank === 0) {\n    throw new Error('pad(scalar) is not defined. Pass non-scalar to pad');\n  }\n  const forward: ForwardFunc<T> = (backend, save) => {\n    save([$x]);\n    return backend.pad($x, paddings, constantValue);\n  };\n\n  const attrs: PadV2Attrs = {paddings, constantValue};\n  const inputs: PadV2Inputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, PadV2,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const pad = op({pad_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor1D` with a given value and paddings. See `pad` for details.\n */\nfunction pad1d_(\n    x: Tensor1D|TensorLike, paddings: [number, number],\n    constantValue = 0): Tensor1D {\n  assert(\n      paddings.length === 2,\n      () => 'Invalid number of paddings. Must be length of 2.');\n  return pad(x, [paddings], constantValue);\n}\n\nexport const pad1d = op({pad1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor2D` with a given value and paddings. See `pad` for details.\n */\nfunction pad2d_(\n    x: Tensor2D|TensorLike, paddings: [[number, number], [number, number]],\n    constantValue = 0): Tensor2D {\n  assert(\n      paddings.length === 2 && paddings[0].length === 2 &&\n          paddings[1].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad2d = op({pad2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor3D` with a given value and paddings. See `pad` for details.\n */\nfunction pad3d_(\n    x: Tensor3D|TensorLike,\n    paddings: [[number, number], [number, number], [number, number]],\n    constantValue = 0): Tensor3D {\n  assert(\n      paddings.length === 3 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad3d = op({pad3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor4D` with a given value and paddings. See `pad` for details.\n */\nfunction pad4d_(\n    x: Tensor4D|TensorLike,\n    paddings:\n        [\n          [number, number], [number, number], [number, number], [number, number]\n        ],\n    constantValue = 0): Tensor4D {\n  assert(\n      paddings.length === 4 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2 &&\n          paddings[3].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad4d = op({pad4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation divides \"spatial\" dimensions `[1, ..., M]` of the input into\n * a grid of blocks of shape `blockShape`, and interleaves these blocks with\n * the \"batch\" dimension (0) such that in the output, the spatial\n * dimensions `[1, ..., M]` correspond to the position within the grid,\n * and the batch dimension combines both the position within a spatial block\n * and the original batch position. Prior to division into blocks,\n * the spatial dimensions of the input are optionally zero padded\n * according to `paddings`. See below for a precise description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);\n * const blockShape = [2, 2];\n * const paddings = [[0, 0], [0, 0]];\n *\n * x.spaceToBatchND(blockShape, paddings).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param paddings A 2-D array. Must have shape `[M, 2]`, all values must be >=\n *     0. `paddings[i] = [padStart, padEnd]` specifies the amount to zero-pad\n * from input dimension `i + 1`, which corresponds to spatial dimension `i`. It\n * is required that\n * `(inputShape[i + 1] + padStart + padEnd) % blockShape[i] === 0`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Zero-pad the start and end of dimensions `[1, ..., M]` of the input\n * according to `paddings` to produce `padded` of shape paddedShape.\n *\n * 2. Reshape `padded` to `reshapedPadded` of shape:\n * `[batch] + [paddedShape[1] / blockShape[0], blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1], blockShape[M-1]] + remainingShape`\n *\n * 3. Permute dimensions of `reshapedPadded` to produce `permutedReshapedPadded`\n * of shape: `blockShape + [batch] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * 4. Reshape `permutedReshapedPadded` to flatten `blockShape` into the\n * batch dimension, producing an output tensor of shape:\n * `[batch * prod(blockShape)] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction spaceToBatchND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], paddings: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'spaceToBatchND');\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank ${$x.rank} should be > than [blockShape] ${\n          blockShape.length}`);\n\n  util.assert(\n      paddings.length === blockShape.length,\n      () => `paddings.shape[0] ${\n          paddings.length} must be equal to [blockShape] ${blockShape.length}`);\n\n  util.assert(\n      $x.shape.reduce(\n          (a, b, i) => {\n            if (i > 0 && i <= blockShape.length) {\n              return a &&\n                  ((b + paddings[i - 1][0] + paddings[i - 1][1]) %\n                       blockShape[i - 1] ===\n                   0);\n            }\n            return a;\n          },\n          true),\n      () => `input spatial dimensions ${$x.shape.slice(1)} with paddings ${\n          paddings.toString()} must be divisible by blockShapes ${\n          blockShape.toString()}`);\n\n  const forward: ForwardFunc<T> = backend =>\n      backend.spaceToBatchND($x, blockShape, paddings);\n\n  const inputs: SpaceToBatchNDInputs = {x: $x};\n  const attrs: SpaceToBatchNDAttrs = {blockShape, paddings};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      SpaceToBatchND, attrs as {} as NamedAttrMap);\n}\n\nexport const spaceToBatchND = op({spaceToBatchND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {avgPool} from './avg_pool';\nimport {batchToSpaceND} from './batch_to_space_nd';\nimport * as conv_util from './conv_util';\nimport {maxPool} from './max_pool';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {spaceToBatchND} from './space_to_batch_nd';\n\n/**\n * Performs an N-D pooling operation\n *\n * @param input The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param windowShape The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param poolingType The type of pooling, either 'max' or 'avg'.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilationRate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction pool_<T extends Tensor3D|Tensor4D>(\n    input: T|TensorLike, windowShape: [number, number]|number,\n    poolingType: 'avg'|'max', pad: 'valid'|'same'|number,\n    dilations?: [number, number]|number, strides?: [number, number]|number) {\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  if (strides == null) {\n    strides = 1;\n  }\n  if (pad === 0) {\n    pad = 'valid';\n  }\n\n  const $x = convertToTensor(input, 'x', 'maxPool');\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in pool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = conv_util.computePool2DInfo(\n      x4D.shape, windowShape, strides, dilations, pad);\n  const dilation: [number, number] =\n      [convInfo.dilationHeight, convInfo.dilationWidth];\n\n  // The following implementation does batchToSpace(pool(spaceToBatch(x)))\n  // whenever dilation > 1 since the TF kernels do not support dilation > 1.\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L1037\n\n  let basePadding: number[][];\n  if (pad === 'same') {\n    basePadding = withSpaceToBatchBasePaddings(\n        [convInfo.filterHeight, convInfo.filterWidth], dilation);\n  } else {\n    basePadding = [[0, 0], [0, 0]];\n  }\n\n  const isDilationOne = dilation[0] === 1 && dilation[1] === 1;\n  const [adjustedPadding, adjustedCrops] = requiredSpaceToBatchPaddings(\n      [convInfo.inHeight, convInfo.inWidth], dilation, basePadding);\n  const convertedPad = isDilationOne ? pad : 'valid';\n  const convertedX =\n      isDilationOne ? x4D : spaceToBatchND(x4D, dilation, adjustedPadding);\n\n  const forwardOp = poolingType === 'avg' ?\n      () => avgPool(convertedX, windowShape, strides, convertedPad) :\n      () => maxPool(convertedX, windowShape, strides, convertedPad);\n  const y = forwardOp();\n\n  const res = isDilationOne ? y : batchToSpaceND(y, dilation, adjustedCrops);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\n// Helper function to compute crops and paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/array_ops.py#L2184\nfunction requiredSpaceToBatchPaddings(\n    inputShape: [number, number], blockShape: [number, number],\n    basePadding: number[][]) {\n  const padStart = basePadding.map(b => b[0]);\n  const origPadEnd = basePadding.map(b => b[1]);\n  const fullInputShape = inputShape.concat(padStart, origPadEnd);\n  const padEndExtra = blockShape.map((b, i) => (b - fullInputShape[i] % b) % b);\n  const padEnd = origPadEnd.map((s, i) => s + padEndExtra[i]);\n  const paddings = blockShape.map((_, i) => [padStart[i], padEnd[i]]);\n  const crops = blockShape.map((_, i) => [0, padEndExtra[i]]);\n  return [paddings, crops];\n}\n\n// Helper function to compute base paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L524\nfunction withSpaceToBatchBasePaddings(\n    filterShape: [number, number], dilation: [number, number]) {\n  // Spatial dimensions of the filters and the upsampled filters in which we\n  // introduce (rate - 1) zeros between consecutive filter values.\n  const dilatedFilterShape = filterShape.map((s, i) => {\n    return s + (s - 1) * (dilation[i] - 1);\n  });\n  const padExtraShape = dilatedFilterShape.map(s => s - 1);\n\n  // When padding is odd, we pad more at end, following the same\n  // convention as conv2d.\n  const padExtraStart = padExtraShape.map(s => Math.floor(s / 2));\n  const padExtraEnd = padExtraShape.map((s, i) => s - padExtraStart[i]);\n  return padExtraShape.map((_, i) => {\n    return [padExtraStart[i], padExtraEnd[i]];\n  });\n}\n\nexport const pool = op({pool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Pow, PowInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the power of one `tf.Tensor` to another. Supports broadcasting.\n *\n * Given a `tf.Tensor` x and a `tf.Tensor` y, this operation computes x^y for\n * corresponding elements in x and y. The result's dtype will be the upcasted\n * type of the `base` and `exp` dtypes.\n *\n * ```js\n * const a = tf.tensor([[2, 3], [4, 5]])\n * const b = tf.tensor([[1, 2], [3, 0]]).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n *\n * ```js\n * const a = tf.tensor([[1, 2], [3, 4]])\n * const b = tf.tensor(2).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n * We also expose `powStrict` which has the same signature as this op and\n * asserts that `base` and `exp` are the same shape (does not broadcast).\n *\n * @param base The base `tf.Tensor` to pow element-wise.\n * @param exp The exponent `tf.Tensor` to pow element-wise.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction pow_<T extends Tensor>(\n    base: Tensor|TensorLike, exp: Tensor|TensorLike): T {\n  let $base = convertToTensor(base, 'base', 'pow');\n  let $exp = convertToTensor(exp, 'exp', 'pow');\n  [$base, $exp] = makeTypesMatch($base, $exp);\n\n  const inputs: PowInputs = {a: $base, b: $exp};\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.pow($base, $exp);\n    save([$base, $exp, y]);\n    return y;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Pow) as T;\n}\n\nexport const pow = op({pow_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prelu, PreluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes leaky rectified linear element-wise with parametric alphas.\n *\n * `x < 0 ? alpha * x : f(x) = x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n * const alpha = tf.scalar(0.1);\n *\n * x.prelu(alpha).print();  // or tf.prelu(x, alpha)\n * ```\n * @param x The input tensor.\n * @param alpha Scaling factor for negative values.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction prelu_<T extends Tensor>(x: T|TensorLike, alpha: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'prelu');\n  const $alpha = convertToTensor(alpha, 'alpha', 'prelu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.prelu($x, $alpha);\n    save([$x, $alpha]);\n    return res;\n  };\n\n  const inputs: PreluInputs = {x: $x, alpha: $alpha};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prelu) as\n      T;\n}\n\nexport const prelu = op({prelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prod, ProdAttrs, ProdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the product of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.prod().print();  // or tf.prod(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.prod(axis).print();  // or tf.prod(x, axis)\n * ```\n *\n * @param x The input tensor to compute the product over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction prod_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'prod');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if ($x.dtype === 'bool') {\n      $x = cast($x, 'int32');\n    }\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.prod(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n\n    return value as T;\n  };\n\n  const inputs: ProdInputs = {x: $x};\n  const attrs: ProdAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prod,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const prod = op({prod_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a random number generator\n * function defined by the user.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param randFunction A random number generator function which is called\n * for each element in the output tensor.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n */\nfunction rand_<R extends Rank>(\n    shape: ShapeMap[R], randFunction: () => number,\n    dtype?: DataType): Tensor<R> {\n  const size = sizeFromShape(shape);\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  for (let i = 0; i < size; i++) {\n    values[i] = randFunction();\n  }\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n\nexport const rand = op({rand_});\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as seedrandom from 'seedrandom';\n\nimport {expectNumbersClose, testEpsilon} from '../test_util';\nimport {TypedArray} from '../types';\n\nexport interface RandomBase {\n  nextValue(): number;\n}\n\nexport interface RandomGamma {\n  nextValue(): number;\n}\n\nexport interface RandNormalDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\nexport interface RandGammaDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\n// https://en.wikipedia.org/wiki/Marsaglia_polar_method\nexport class MPRandGauss implements RandomBase {\n  private mean: number;\n  private stdDev: number;\n  private nextVal: number;\n  private dtype?: keyof RandNormalDataTypes;\n  private truncated?: boolean;\n  private upper?: number;\n  private lower?: number;\n  private random: seedrandom.prng;\n\n  constructor(\n      mean: number, stdDeviation: number, dtype?: keyof RandNormalDataTypes,\n      truncated?: boolean, seed?: number) {\n    this.mean = mean;\n    this.stdDev = stdDeviation;\n    this.dtype = dtype;\n    this.nextVal = NaN;\n    this.truncated = truncated;\n    if (this.truncated) {\n      this.upper = this.mean + this.stdDev * 2;\n      this.lower = this.mean - this.stdDev * 2;\n    }\n    const seedValue = seed ? seed : Math.random();\n    this.random = seedrandom.alea(seedValue.toString());\n  }\n\n  /** Returns next sample from a Gaussian distribution. */\n  public nextValue(): number {\n    if (!isNaN(this.nextVal)) {\n      const value = this.nextVal;\n      this.nextVal = NaN;\n      return value;\n    }\n\n    let resultX: number, resultY: number;\n    let isValid = false;\n    while (!isValid) {\n      let v1: number, v2: number, s: number;\n      do {\n        v1 = 2 * this.random() - 1;\n        v2 = 2 * this.random() - 1;\n        s = v1 * v1 + v2 * v2;\n      } while (s >= 1 || s === 0);\n\n      const mul = Math.sqrt(-2.0 * Math.log(s) / s);\n      resultX = this.mean + this.stdDev * v1 * mul;\n      resultY = this.mean + this.stdDev * v2 * mul;\n\n      if (!this.truncated || this.isValidTruncated(resultX)) {\n        isValid = true;\n      }\n    }\n\n    if (!this.truncated || this.isValidTruncated(resultY)) {\n      this.nextVal = this.convertValue(resultY);\n    }\n    return this.convertValue(resultX);\n  }\n\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype == null || this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  /** Returns true if less than 2-standard-deviations from the mean. */\n  private isValidTruncated(value: number): boolean {\n    return value <= this.upper && value >= this.lower;\n  }\n}\n\n// Marsaglia, George, and Wai Wan Tsang. 2000. \"A Simple Method for Generating\n// Gamma Variables.\"\nexport class RandGamma implements RandomGamma {\n  private alpha: number;\n  private beta: number;\n  private d: number;\n  private c: number;\n  private dtype?: keyof RandGammaDataTypes;\n  private randu: seedrandom.prng;\n  private randn: MPRandGauss;\n\n  constructor(\n      alpha: number, beta: number, dtype: keyof RandGammaDataTypes,\n      seed?: number) {\n    this.alpha = alpha;\n    this.beta = 1 / beta;  // convert rate to scale parameter\n    this.dtype = dtype;\n\n    const seedValue = seed ? seed : Math.random();\n    this.randu = seedrandom.alea(seedValue.toString());\n    this.randn = new MPRandGauss(0, 1, dtype, false, this.randu());\n\n    if (alpha < 1) {\n      this.d = alpha + (2 / 3);\n    } else {\n      this.d = alpha - (1 / 3);\n    }\n    this.c = 1 / Math.sqrt(9 * this.d);\n  }\n\n  /** Returns next sample from a gamma distribution. */\n  public nextValue(): number {\n    let x2: number, v0: number, v1: number, x: number, u: number, v: number;\n    while (true) {\n      do {\n        x = this.randn.nextValue();\n        v = 1 + (this.c * x);\n      } while (v <= 0);\n      v *= v * v;\n      x2 = x * x;\n      v0 = 1 - (0.331 * x2 * x2);\n      v1 = (0.5 * x2) + (this.d * (1 - v + Math.log(v)));\n      u = this.randu();\n      if (u < v0 || Math.log(u) < v1) {\n        break;\n      }\n    }\n    v = (1 / this.beta) * this.d * v;\n    if (this.alpha < 1) {\n      v *= Math.pow(this.randu(), 1 / this.alpha);\n    }\n    return this.convertValue(v);\n  }\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n}\n\nexport class UniformRandom implements RandomBase {\n  private min: number;\n  private range: number;\n  private random: seedrandom.prng;\n  private dtype?: keyof RandNormalDataTypes;\n\n  constructor(\n      min = 0, max = 1, dtype?: keyof RandNormalDataTypes,\n      seed?: string|number) {\n    this.min = min;\n    this.range = max - min;\n    this.dtype = dtype;\n    if (seed == null) {\n      seed = Math.random();\n    }\n    if (typeof seed === 'number') {\n      seed = seed.toString();\n    }\n\n    if (!this.canReturnFloat() && this.range <= 1) {\n      throw new Error(\n          `The difference between ${min} - ${max} <= 1 and dtype is not float`);\n    }\n    this.random = seedrandom.alea(seed);\n  }\n\n  /** Handles proper rounding for non floating point numbers. */\n  private canReturnFloat = () =>\n      (this.dtype == null || this.dtype === 'float32');\n\n  private convertValue(value: number): number {\n    if (this.canReturnFloat()) {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  nextValue() {\n    return this.convertValue(this.min + this.range * this.random());\n  }\n}\n\nexport function jarqueBeraNormalityTest(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test\n  const n = values.length;\n  const s = skewness(values);\n  const k = kurtosis(values);\n  const jb = n / 6 * (Math.pow(s, 2) + 0.25 * Math.pow(k - 3, 2));\n  // JB test requires 2-degress of freedom from Chi-Square @ 0.95:\n  // http://www.itl.nist.gov/div898/handbook/eda/section3/eda3674.htm\n  const CHI_SQUARE_2DEG = 5.991;\n  if (jb > CHI_SQUARE_2DEG) {\n    throw new Error(`Invalid p-value for JB: ${jb}`);\n  }\n}\n\nexport function expectArrayInMeanStdRange(\n    actual: TypedArray|number[], expectedMean: number, expectedStdDev: number,\n    epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  const actualMean = mean(actual);\n  expectNumbersClose(actualMean, expectedMean, epsilon);\n  expectNumbersClose(\n      standardDeviation(actual, actualMean), expectedStdDev, epsilon);\n}\n\nfunction mean(values: TypedArray|number[]) {\n  let sum = 0;\n  for (let i = 0; i < values.length; i++) {\n    sum += values[i];\n  }\n  return sum / values.length;\n}\n\nfunction standardDeviation(values: TypedArray|number[], mean: number) {\n  let squareDiffSum = 0;\n  for (let i = 0; i < values.length; i++) {\n    const diff = values[i] - mean;\n    squareDiffSum += diff * diff;\n  }\n  return Math.sqrt(squareDiffSum / values.length);\n}\n\nfunction kurtosis(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Kurtosis\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum4 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum4 += Math.pow(v, 4);\n  }\n  return (1 / n) * sum4 / Math.pow((1 / n) * sum2, 2);\n}\n\nfunction skewness(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Skewness\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum3 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum3 += Math.pow(v, 3);\n  }\n  return (1 / n) * sum3 / Math.pow((1 / (n - 1)) * sum2, 3 / 2);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {RandGamma} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a gamma distribution.\n *\n * ```js\n * tf.randomGamma([2, 2], 1).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param alpha The shape parameter of the gamma distribution.\n * @param beta The inverse scale parameter of the gamma distribution. Defaults\n *     to 1.\n * @param dtype The data type of the output. Defaults to float32.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomGamma_<R extends Rank>(\n    shape: ShapeMap[R], alpha: number, beta = 1,\n    dtype: 'float32'|'int32' = 'float32', seed?: number): Tensor<R> {\n  if (beta == null) {\n    beta = 1;\n  }\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const rgamma = new RandGamma(alpha, beta, dtype, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = rgamma.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomGamma = op({randomGamma_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a normal distribution.\n *\n * ```js\n * tf.randomNormal([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, false /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomNormal = op({randomNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {UniformRandom} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a uniform distribution.\n *\n * The generated values follow a uniform distribution in the range [minval,\n * maxval). The lower bound minval is included in the range, while the upper\n * bound maxval is excluded.\n *\n * ```js\n * tf.randomUniform([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param minval The lower bound on the range of random values to generate.\n *   Defaults to 0.\n * @param maxval The upper bound on the range of random values to generate.\n *   Defaults to 1.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomUniform_<R extends Rank>(\n    shape: ShapeMap[R], minval = 0, maxval = 1, dtype: DataType = 'float32',\n    seed?: number|string): Tensor<R> {\n  const res = buffer(shape, dtype);\n  const random = new UniformRandom(minval, maxval, null, seed);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = random.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomUniform = op({randomUniform_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike1D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-1 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor1d` as it makes the code more readable.\n *\n * ```js\n * tf.tensor1d([1, 2, 3]).print();\n * ```\n *\n * @param values The values of the tensor. Can be array of numbers,\n *     or a `TypedArray`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor1d(values: TensorLike1D, dtype?: DataType): Tensor1D {\n  assertNonNull(values);\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 1) {\n    throw new Error('tensor1d() requires values to be a flat/TypedArray');\n  }\n  const shape: number[] = null;\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Range, RangeAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {makeZerosTypedArray} from '../util';\n\nimport {tensor1d} from './tensor1d';\nimport {zeros} from './zeros';\n\n/**\n * Creates a new `tf.Tensor1D` filled with the numbers in the range provided.\n *\n * The tensor is a is half-open interval meaning it includes start, but\n * excludes stop. Decrementing ranges and negative step values are also\n * supported.sv\n *\n *\n * ```js\n * tf.range(0, 9, 2).print();\n * ```\n *\n * @param start An integer start value\n * @param stop An integer stop value\n * @param step An integer increment (will default to 1 or -1)\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function range(\n    start: number, stop: number, step = 1,\n    dtype: 'float32'|'int32' = 'float32'): Tensor1D {\n  if (step === 0) {\n    throw new Error('Cannot have a step of zero');\n  }\n\n  const forward: ForwardFunc<Tensor> = () => {\n    const sameStartStop = start === stop;\n    const increasingRangeNegativeStep = start < stop && step < 0;\n    const decreasingRangePositiveStep = stop < start && step > 1;\n\n    if (sameStartStop || increasingRangeNegativeStep ||\n        decreasingRangePositiveStep) {\n      return zeros([0], dtype);\n    }\n\n    const numElements = Math.abs(Math.ceil((stop - start) / step));\n    const values = makeZerosTypedArray(numElements, dtype);\n\n    if (stop < start && step === 1) {\n      // Auto adjust the step's sign if it hasn't been set\n      // (or was set to 1)\n      step = -1;\n    }\n\n    values[0] = start;\n    for (let i = 1; i < values.length; i++) {\n      values[i] = values[i - 1] + step;\n    }\n\n    return tensor1d(values, dtype);\n  };\n\n  const attrs: RangeAttrs = {start, stop, step, dtype};\n\n  return ENGINE.runKernelFunc(\n             forward, {} /* inputs */, null /* grad */, Range,\n             attrs as {} as NamedAttrMap) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Reciprocal, ReciprocalInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of x element-wise: `1 / x`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, 2]);\n *\n * x.reciprocal().print();  // or tf.reciprocal(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction reciprocal_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'reciprocal');\n\n  const inputs: ReciprocalInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.reciprocal($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Reciprocal);\n}\nexport const reciprocal = op({reciprocal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu, ReluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport { cast } from './cast';\n\n/**\n * Computes rectified linear element-wise: `max(x, 0)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.relu().print();  // or tf.relu(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu($x);\n  };\n\n  const inputs: ReluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu) as\n      T;\n}\n\nexport const relu = op({relu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu6, Relu6Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes rectified linear 6 element-wise: `min(max(x, 0), 6)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 8]);\n *\n * x.relu6().print();  // or tf.relu6(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu6_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu6');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu6($x);\n  };\n\n  const inputs: Relu6Inputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu6) as\n      T;\n}\n\nexport const relu6 = op({relu6_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reverse, ReverseAttrs, ReverseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Reverses a `tf.Tensor` along a specified axis.\n *\n * Also available are stricter rank-specific methods that assert that `x` is\n * of the given rank:\n *   - `tf.reverse1d`\n *   - `tf.reverse2d`\n *   - `tf.reverse3d`\n *   - `tf.reverse4d`\n *\n * Except `tf.reverse1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.reverse().print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.reverse(axis).print();\n * ```\n * @param x The input tensor to be reversed.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction reverse_<T extends Tensor>(\n    x: T|TensorLike, axis?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'reverse');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const axes = parseAxisParam(axis, $x.shape);\n    if ($x.rank === 0) {\n      return clone($x);\n    }\n    const res = backend.reverse($x, axes);\n    return reshape(res, $x.shape);\n  };\n\n  const inputs: ReverseInputs = {x: $x};\n  const attrs: ReverseAttrs = {dims: axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Reverse, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const reverse = op({reverse_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor1D`.\n *\n * @param x The input tensor.\n */\nfunction reverse1d_(x: Tensor1D|TensorLike): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 1,\n      () => `Error in reverse1D: x must be rank 1 but got rank ${$x.rank}.`);\n  return reverse($x, 0);\n}\n\nexport const reverse1d = op({reverse1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor2D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse2d_(x: Tensor2D|TensorLike, axis?: number|number[]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 2,\n      () => `Error in reverse2D: x must be rank 2 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse2d = op({reverse2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor3D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse3d_(x: Tensor3D|TensorLike, axis?: number|number[]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 3,\n      () => `Error in reverse3D: x must be rank 3 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse3d = op({reverse3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor4D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse4d_(x: Tensor4D|TensorLike, axis?: number|number[]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 4,\n      () => `Error in reverse4D: x must be rank 4 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse4d = op({reverse4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Round, RoundInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes round of input `tf.Tensor` element-wise: `round(x)`.\n * It implements banker's rounding.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.round().print();  // or tf.round(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction round_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'round');\n  const inputs: RoundInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.round($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Round);\n}\n\nexport const round = op({round_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Rsqrt, RsqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of square root of the input `tf.Tensor` element-wise:\n * `y = 1 / sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.rsqrt().print();  // or tf.rsqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction rsqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'rsqrt');\n\n  const inputs: RsqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.rsqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Rsqrt);\n}\nexport const rsqrt = op({rsqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Selu, SeluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes scaled exponential linear element-wise.\n *\n * `x < 0 ? scale * alpha * (exp(x) - 1) : x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.selu().print();  // or tf.selu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction selu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'selu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.selu($x);\n    save([$x]);\n    return res;\n  };\n\n  const inputs: SeluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Selu) as\n      T;\n}\n\nexport const selu = op({selu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport {depthwiseConv2d} from './depthwise_conv2d';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * 2-D convolution with separable filters.\n *\n * Performs a depthwise convolution that acts separately on channels followed\n * by a pointwise convolution that mixes channels. Note that this is\n * separability between dimensions [1, 2] and 3, not spatial separability\n * between dimensions 1 and 2.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param depthwiseFilter The depthwise filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`. This is\n *     the filter used in the first step.\n * @param pointwiseFilter The pointwise filter tensor, rank 4, of shape\n *     `[1, 1, inChannels * channelMultiplier, outChannels]`. This is\n *     the filter used in the second step.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction separableConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthwiseFilter: Tensor4D|TensorLike,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'separableConv2d');\n  const $depthwiseFilter =\n      convertToTensor(depthwiseFilter, 'depthwiseFilter', 'separableConv2d');\n  const $pointwiseFilter =\n      convertToTensor(pointwiseFilter, 'pointwiseFilter', 'separableConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  if (dataFormat === 'NCHW') {\n    throw new Error(\n        'separableConv2d currently does not support dataFormat NCHW; only ' +\n        'NHWC is supported');\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in separableConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $depthwiseFilter.rank === 4,\n      () => `Error in separableConv2d: depthwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.rank === 4,\n      () => `Error in separableConv2d: pointwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.shape[0] === 1,\n      () =>\n          `Error in separableConv2d: the first dimension of pointwise filter ` +\n          ` must be 1, but got ${$pointwiseFilter.shape[0]}.`);\n  util.assert(\n      $pointwiseFilter.shape[1] === 1,\n      () => `Error in separableConv2d: the second dimension of pointwise ` +\n          `filter must be 1, but got ${$pointwiseFilter.shape[1]}.`);\n\n  const inChannels = $depthwiseFilter.shape[2];\n  const channelMultiplier = $depthwiseFilter.shape[3];\n  util.assert(\n      $pointwiseFilter.shape[2] === inChannels * channelMultiplier,\n      () =>\n          `Error in separableConv2d: the third dimension of pointwise filter ` +\n          `must be ${inChannels * channelMultiplier}, ` +\n          `but got ${$pointwiseFilter.shape[2]}.`);\n\n  const depthwise = depthwiseConv2d(\n      x4D, $depthwiseFilter, strides, pad, dataFormat, dilation);\n  const pointwiseStride = 1;\n  const res =\n      conv2d(depthwise, $pointwiseFilter, pointwiseStride, 'valid', dataFormat);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const separableConv2d = op({separableConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\n/**\n * Computes the difference between two lists of numbers.\n *\n * Given a Tensor `x` and a Tensor `y`, this operation returns a Tensor `out`\n * that represents all values that are in `x` but not in `y`. The returned\n * Tensor `out` is sorted in the same order that the numbers appear in `x`\n * (duplicates are preserved). This operation also returns a Tensor indices that\n * represents the position of each out element in `x`. In other words:\n *\n * `out[i] = x[idx[i]] for i in [0, 1, ..., out.length - 1]`\n *\n * ```js\n * const x = [1, 2, 3, 4, 5, 6];\n * const y = [1, 3, 5];\n *\n * const [out, indices] = await tf.setdiff1dAsync(x, y);\n * out.print(); // [2, 4, 6]\n * indices.print(); // [1, 3, 5]\n * ```\n *\n * @param x 1-D Tensor. Values to keep.\n * @param y 1-D Tensor. Must have the same type as x. Values to exclude in the\n *     output.\n * @returns Promise of Tensor tuple [out, indices].\n *  out: Tensor with the same type as x.\n *  indices: A Tensor of type int32.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nasync function setdiff1dAsync_(\n    x: Tensor|TensorLike, y: Tensor|TensorLike): Promise<[Tensor, Tensor]> {\n  const $x = convertToTensor(x, 'x', 'setdiff1d');\n  const $y = convertToTensor(y, 'y', 'setdiff1d');\n\n  util.assert(\n      $x.dtype === $y.dtype,\n      () => `x and y should have the same dtype, but got x (${\n          $x.dtype}) and y (${$y.dtype}).`);\n\n  util.assert(\n      $x.rank === 1, () => `x should be 1D tensor, but got x (${$x.shape}).`);\n\n  util.assert(\n      $y.rank === 1, () => `y should be 1D tensor, but got y (${$y.shape}).`);\n\n  const xVals = await $x.data();\n  const yVals = await $y.data();\n  const ySet = new Set(yVals);\n\n  let outputSize = 0;\n  for (let i = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      outputSize++;\n    }\n  }\n\n  const buffer = new TensorBuffer([outputSize], $x.dtype);\n  const indices = new TensorBuffer([outputSize], 'int32');\n  for (let i = 0, p = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      buffer.values[p] = xVals[i];\n      indices.values[p] = i;\n      p++;\n    }\n  }\n  return [buffer.toTensor(), indices.toTensor()];\n}\nexport const setdiff1dAsync = setdiff1dAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sign, SignInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns an element-wise indication of the sign of a number.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3, NaN, 0]);\n *\n * x.sign().print();  // or tf.sign(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sign_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sign');\n  const inputs: SignInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.sign($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Sign);\n}\nexport const sign = op({sign_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sin, SinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sin of the input Tensor element-wise: `sin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.sin().print();  // or tf.sin(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sin');\n\n  const inputs: SinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sin);\n}\nexport const sin = op({sin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sinh, SinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic sin of the input `tf.Tensor` element-wise: `sinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.sinh().print();  // or tf.sinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sinh');\n  const inputs: SinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sinh);\n}\nexport const sinh = op({sinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 1D slice from 1D array starting at coordinates `begin` and is\n * of length `size`. See `slice` for details.\n */\nfunction slice1d_(\n    x: Tensor1D|TensorLike, begin: number, size: number): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'slice1d');\n  util.assert(\n      $x.rank === 1,\n      () =>\n          `slice1d expects a rank-1 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, [begin], [size]);\n}\nexport const slice1d = op({slice1d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 2D slice from a 2D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice2d_(\n    x: Tensor2D|TensorLike, begin: [number, number],\n    size: [number, number]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'slice2d');\n  util.assert(\n      $x.rank === 2,\n      () =>\n          `slice2d expects a rank-2 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice2d = op({slice2d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 3D slice from a 3D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice3d_(\n    x: Tensor3D|TensorLike, begin: [number, number, number],\n    size: [number, number, number]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'slice3d');\n  util.assert(\n      $x.rank === 3,\n      () =>\n          `slice3d expects a rank-3 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice3d = op({slice3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 4D slice from a 4D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice4d_(\n    x: Tensor4D|TensorLike, begin: [number, number, number, number],\n    size: [number, number, number, number]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'slice4d');\n  util.assert(\n      $x.rank === 4,\n      () =>\n          `slice4d expects a rank-4 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice4d = op({slice4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softmax, SoftmaxAttrs, SoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the softmax normalized vector given the logits.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction softmax_<T extends Tensor>(logits: T|TensorLike, dim = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'softmax', 'float32');\n\n  if (dim === -1) {\n    dim = $logits.rank - 1;\n  }\n  if (dim !== $logits.rank - 1) {\n    throw Error(\n        'Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and dim was ${dim}`);\n  }\n\n  const inputs: SoftmaxInputs = {logits: $logits};\n  const attrs: SoftmaxAttrs = {dim};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const y = backend.softmax($logits, dim);\n        save([y]);\n        return y;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, Softmax,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const softmax = op({softmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FFT, FFTInputs} from '../../kernel_names';\nimport {Tensor} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\n\n/**\n * Fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the inner-most\n * dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.fft().print();  // tf.spectral.fft(x).print();\n * ```\n * @param input The complex input to compute an fft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction fft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.fft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: FFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D = input.as2D(batch, innerDimensionSize);\n    const result = backend.fft(input2D);\n    return result.reshape(input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, FFT);\n}\n\nexport const fft = op({fft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {IFFT, IFFTInputs} from '../../kernel_names';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Inverse fast Fourier transform.\n *\n * Computes the inverse 1-dimensional discrete Fourier transform over the\n * inner-most dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.ifft().print();  // tf.spectral.ifft(x).print();\n * ```\n * @param input The complex input to compute an ifft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction ifft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.ifft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: IFFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D: Tensor2D = reshape(input, [batch, innerDimensionSize]);\n    const result = backend.ifft(input2D);\n    return reshape(result, input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, IFFT);\n}\n\nexport const ifft = op({ifft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {reverse} from '../reverse';\nimport {scalar} from '../scalar';\nimport {slice} from '../slice';\n\nimport {ifft} from './ifft';\n\n/**\n * Inversed real value input fast Fourier transform.\n *\n * Computes the 1-dimensional inversed discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([0, 0, 0]);\n * const x = tf.complex(real, imag);\n *\n * x.irfft().print();\n * ```\n * @param input The real value input to compute an irfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction irfft_(input: Tensor): Tensor {\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n  let ret: Tensor;\n  if (innerDimensionSize <= 2) {\n    const complexInput = reshape(input, [batch, innerDimensionSize]);\n    ret = ifft(complexInput);\n  } else {\n    // The length of unique components of the DFT of a real-valued signal\n    // is 2 * (input_len - 1)\n    const outputShape = [batch, 2 * (innerDimensionSize - 1)];\n    const realInput = reshape(real(input), [batch, innerDimensionSize]);\n    const imagInput = reshape(imag(input), [batch, innerDimensionSize]);\n\n    const realConjugate =\n        reverse(slice(realInput, [0, 1], [batch, innerDimensionSize - 2]), 1);\n    const imagConjugate: Tensor2D = mul(\n        reverse(slice(imagInput, [0, 1], [batch, innerDimensionSize - 2]), 1),\n        scalar(-1));\n\n    const r = concat([realInput, realConjugate], 1);\n    const i = concat([imagInput, imagConjugate], 1);\n    const complexInput =\n        reshape(complex(r, i), [outputShape[0], outputShape[1]]);\n    ret = ifft(complexInput);\n  }\n  ret = real(ret);\n  // reshape the result if the input is 3D tensor.\n  if (input.rank === 3 && input.shape[0] !== 0) {\n    const temp = ret;\n    const batch = input.shape[0];\n    ret = reshape(ret, [batch, ret.shape[0] / batch, ret.shape[1]]);\n    temp.dispose();\n  }\n  return ret;\n}\n\nexport const irfft = op({irfft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {assert} from '../util';\n\n/**\n * Prepare the split size array. When the input is a number, the axis is evenly\n * divided among the split size. When the input contains the negative value, the\n * rest of the axis is allocated toward that.\n */\nexport function prepareSplitSize(\n    x: Tensor|TensorInfo, numOrSizeSplits: number[]|number,\n    axis = 0): number[] {\n  let splitSizes = [];\n  if (typeof (numOrSizeSplits) === 'number') {\n    assert(\n        x.shape[axis] % numOrSizeSplits === 0,\n        () => 'Number of splits must evenly divide the axis.');\n    splitSizes =\n        new Array(numOrSizeSplits).fill(x.shape[axis] / numOrSizeSplits);\n  } else {\n    const numOfNegs = numOrSizeSplits.reduce((count, value) => {\n      if (value === -1) {\n        count += 1;\n      }\n      return count;\n    }, 0);\n    assert(\n        numOfNegs <= 1,\n        () => 'There should be only one negative value in split array.');\n    const negIndex = numOrSizeSplits.indexOf(-1);\n    // Allow the number of split array to be -1, which indicates the rest\n    // of dimension is allocated to that split.\n    if (negIndex !== -1) {\n      const total = numOrSizeSplits.reduce((a, b) => b > 0 ? a + b : a);\n      numOrSizeSplits[negIndex] = x.shape[axis] - total;\n    }\n    assert(\n        x.shape[axis] === numOrSizeSplits.reduce((a, b) => a + b),\n        () => 'The sum of sizes must match the size of the axis dimension.');\n    splitSizes = numOrSizeSplits;\n  }\n\n  return splitSizes;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SplitV, SplitVAttrs, SplitVInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {prepareSplitSize} from './split_util';\n\n/**\n * Splits a `tf.Tensor` into sub tensors.\n *\n * If `numOrSizeSplits` is a number, splits `x` along dimension `axis`\n * into `numOrSizeSplits` smaller tensors.\n * Requires that `numOrSizeSplits` evenly divides `x.shape[axis]`.\n *\n * If `numOrSizeSplits` is a number array, splits `x` into\n * `numOrSizeSplits.length` pieces. The shape of the `i`-th piece has the\n * same size as `x` except along dimension `axis` where the size is\n * `numOrSizeSplits[i]`.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4, 5, 6, 7, 8], [2, 4]);\n * const [a, b] = tf.split(x, 2, 1);\n * a.print();\n * b.print();\n *\n * const [c, d, e] = tf.split(x, [1, 2, 1], 1);\n * c.print();\n * d.print();\n * e.print();\n * ```\n *\n * @param x The input tensor to split.\n * @param numOrSizeSplits Either an integer indicating the number of\n * splits along the axis or an array of integers containing the sizes of\n * each output tensor along the axis. If a number then it must evenly divide\n * `x.shape[axis]`; otherwise the sum of sizes must match `x.shape[axis]`.\n * Can contain one -1 indicating that dimension is to be inferred.\n * @param axis The dimension along which to split. Defaults to 0 (the first\n * dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction split_<T extends Tensor>(\n    x: Tensor|TensorLike, numOrSizeSplits: number[]|number, axis = 0): T[] {\n  const $x = convertToTensor(x, 'x', 'split');\n\n  const forward: ForwardFunc<Tensor> = (backend, _) => {\n    const $axis = parseAxisParam(axis, $x.shape)[0];\n    const splitSizes = prepareSplitSize($x, numOrSizeSplits, $axis);\n    return backend.split($x, splitSizes, $axis) as {} as T;\n  };\n\n  const inputs: SplitVInputs = {x: $x};\n  const attr: SplitVAttrs = {numOrSizeSplits, axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, SplitV,\n             attr as {} as NamedAttrMap) as {} as T[];\n}\n\nexport const split = op({split_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {assert} from '../../util';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {split} from '../split';\nimport {zeros} from '../zeros';\nimport {zerosLike} from '../zeros_like';\n\nimport {fft} from './fft';\n\n/**\n * Real value input fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n *\n * real.rfft().print();\n * ```\n * @param input The real value input to compute an rfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction rfft_(input: Tensor, fftLength?: number): Tensor {\n  assert(\n      input.dtype === 'float32',\n      () => `The dtype for rfft() must be real value but got ${input.dtype}`);\n\n  let innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n\n  let adjustedInput: Tensor;\n  if (fftLength != null && fftLength < innerDimensionSize) {\n    // Need to crop\n    const begin = input.shape.map(v => 0);\n    const size = input.shape.map(v => v);\n    size[input.shape.length - 1] = fftLength;\n    adjustedInput = slice(input, begin, size);\n    innerDimensionSize = fftLength;\n  } else if (fftLength != null && fftLength > innerDimensionSize) {\n    // Need to pad with zeros\n    const zerosShape = input.shape.map(v => v);\n    zerosShape[input.shape.length - 1] = fftLength - innerDimensionSize;\n    adjustedInput = concat([input, zeros(zerosShape)], input.shape.length - 1);\n    innerDimensionSize = fftLength;\n  } else {\n    adjustedInput = input;\n  }\n\n  // Complement the input with zero imaginary numbers.\n  const zerosInput = zerosLike(adjustedInput);\n  const complexInput =\n      reshape(complex(adjustedInput, zerosInput), [batch, innerDimensionSize]);\n\n  const ret = fft(complexInput);\n\n  // Exclude complex conjugations. These conjugations are put symmetrically.\n  const half = Math.floor(innerDimensionSize / 2) + 1;\n  const realValues = real(ret);\n  const imagValues = imag(ret);\n  const realComplexConjugate = split(\n      realValues, [half, innerDimensionSize - half],\n      realValues.shape.length - 1);\n  const imagComplexConjugate = split(\n      imagValues, [half, innerDimensionSize - half],\n      imagValues.shape.length - 1);\n\n  const outputShape = adjustedInput.shape.slice();\n  outputShape[adjustedInput.shape.length - 1] = half;\n\n  return reshape(\n      complex(realComplexConjugate[0], imagComplexConjugate[0]), outputShape);\n}\n\nexport const rfft = op({rfft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sqrt, SqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes square root of the input `tf.Tensor` element-wise: `y = sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.sqrt().print();  // or tf.sqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sqrt');\n\n  const inputs: SqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sqrt);\n}\nexport const sqrt = op({sqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SquaredDifference, SquaredDifferenceInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns (a - b) * (a - b) element-wise.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * ```js\n * // Broadcast squared difference  a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction squaredDifference_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'squaredDifference');\n  let $b = convertToTensor(b, 'b', 'squaredDifference');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.squaredDifference($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SquaredDifferenceInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             SquaredDifference, attrs) as T;\n}\n\nexport const squaredDifference = op({squaredDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {squeezeShape} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Removes dimensions of size 1 from the shape of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4], [1, 1, 4]);\n * x.squeeze().print();\n * ```\n *\n * @param x The input tensor to be squeezed.\n * @param axis An optional list of numbers. If specified, only\n *     squeezes the dimensions listed. The dimension index starts at 0. It\n * is an error to squeeze a dimension that is not 1.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction squeeze_<T extends Tensor>(x: Tensor|TensorLike, axis?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'squeeze');\n  return reshape($x, squeezeShape($x.shape, axis).newShape) as T;\n}\n\nexport const squeeze = op({squeeze_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {concat} from './concat';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\n\n/**\n * Stacks a list of rank-`R` `tf.Tensor`s into one rank-`(R+1)` `tf.Tensor`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.stack([a, b, c]).print();\n * ```\n *\n * @param tensors A list of tensor objects with the same shape and dtype.\n * @param axis The axis to stack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction stack_<T extends Tensor>(\n    tensors: Array<T|TensorLike>, axis = 0): Tensor {\n  const $tensors = convertToTensorArray(tensors, 'tensors', 'stack');\n\n  util.assert(\n      $tensors.length >= 1, () => 'Pass at least one tensor to tf.stack');\n\n  if ($tensors.length === 1) {\n    return expandDims($tensors[0], axis);\n  }\n\n  const rank = $tensors[0].rank;\n  const shape = $tensors[0].shape;\n  const dtype = $tensors[0].dtype;\n\n  util.assert(axis <= rank, () => 'Axis must be <= rank of the tensor');\n\n  $tensors.forEach(t => {\n    util.assertShapesMatch(\n        shape, t.shape,\n        'All tensors passed to stack must have matching shapes');\n    util.assert(\n        dtype === t.dtype,\n        () => 'All tensors passed to stack must have matching dtypes');\n  });\n\n  const expandedTensors = $tensors.map(t => expandDims(t, axis));\n  // Stack exists in the TensorFlow C++ API\n  // (https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/stack) but not\n  // in\n  // https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/core/ops/ops.pbtxt.\n  // Therefore we are treating it like a high-level op rather than\n  // creating a dedicated stack kernel.\n  return concat(expandedTensors, axis);\n}\n\nexport const stack = op({stack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Step, StepAttrs, StepInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes step of the input `tf.Tensor` element-wise: `x > 0 ? 1 : alpha * x`\n *\n * ```js\n * const x = tf.tensor1d([0, 2, -1, -3]);\n *\n * x.step(.5).print();  // or tf.step(x, .5)\n * ```\n * @param x The input tensor.\n * @param alpha The gradient when input is negative.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction step_<T extends Tensor>(x: T|TensorLike, alpha = 0.0): T {\n  const $x = convertToTensor(x, 'x', 'step');\n\n  const inputs: StepInputs = {x: $x};\n  const attrs: StepAttrs = {alpha};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.step($x, alpha), inputs as {} as NamedTensorMap,\n      null /* grad */, Step, attrs as {} as NamedAttrMap);\n}\nexport const step = op({step_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {StridedSlice, StridedSliceAttrs, StridedSliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {slice} from './slice';\nimport {computeOutShape, getNormalizedAxes, maskToAxes} from './slice_util';\n\n/**\n * Extracts a strided slice of a tensor.\n *\n * Roughly speaking, this op extracts a slice of size (end-begin)/stride from\n * the given input tensor (x). Starting at the location specified by begin the\n * slice continues by adding stride to the index until all dimensions are not\n * less than end. Note that a stride can be negative, which causes a reverse\n * slice.\n *\n * ```js\n * const t = tf.tensor3d([1, 1, 1 ,2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6],\n *    [3, 2, 3]);\n * t.stridedSlice([1, 0, 0], [2, 1, 3], [1, 1, 1]).print()  // [[[3, 3, 3]]]\n * t.stridedSlice([1, 0, 0], [2, 2, 3], [1, 1, 1]).print()  // [[[3, 3, 3],\n *                                                     // [4, 4, 4]]]\n * t.stridedSlice([1, -1, 0], [2, -3, 3], [1, -1, 1]).print() // [[[4, 4, 4],\n *                                                     // [3, 3, 3]]]\n * ```\n *\n * @param x The tensor to stride slice.\n * @param begin The coordinates to start the slice from.\n * @param end: The coordinates to end the slice at.\n * @param strides: The size of the slice.\n * @param beginMask: If the ith bit of beginMask is set, begin[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param endMask: If the ith bit of endMask is set, end[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param shrinkAxisMask: a bitmask where bit i implies that\n * the ith specification should shrink the dimensionality. begin and end must\n * imply a slice of size 1 in the dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction stridedSlice_(\n    x: Tensor|TensorLike, begin: number[], end: number[], strides?: number[],\n    beginMask = 0, endMask = 0, ellipsisMask = 0, newAxisMask = 0,\n    shrinkAxisMask = 0): Tensor {\n  let $x = convertToTensor(x, 'x', 'stridedSlice');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if (strides == null) {\n      strides = new Array(begin.length);\n    }\n\n    const ellipsisAxes = maskToAxes(ellipsisMask);\n    if (ellipsisAxes.length > 1) {\n      throw new Error('Multiple ellipses in slice is not allowed.');\n    }\n\n    if (ellipsisMask !== 0 && newAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and newAxisMask is not yet supported.');\n    }\n\n    if (ellipsisMask !== 0 && shrinkAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and shrinkAxisMask is not yet supported.');\n    }\n\n    const numInterpolatedAxes = $x.rank - begin.length;\n\n    // Expand the dims of x based on the newAxisMask.\n    const expandAxes = maskToAxes(newAxisMask);\n    const newShape = $x.shape.slice();\n    expandAxes.forEach(axis => {\n      begin[axis] = 0;\n      end[axis] = 1;\n      newShape.splice(axis, 0, 1);\n    });\n    $x = reshape($x, newShape);\n\n    const {\n      begin: normalizedBegin,\n      end: normalizedEnd,\n      strides: normalizedStrides\n    } =\n        getNormalizedAxes(\n            $x.shape, ellipsisAxes, numInterpolatedAxes, begin, end, strides,\n            beginMask, endMask, ellipsisMask);\n    begin = normalizedBegin;\n    end = normalizedEnd;\n    strides = normalizedStrides;\n\n    const shrinkAxes = maskToAxes(shrinkAxisMask);\n    // Adjust the ends based on the shrink mask.\n    shrinkAxes.forEach(axis => {\n      end[axis] = begin[axis] + 1;\n      strides[axis] = 1;\n    });\n\n    // Figure out the output shape.\n    const size = computeOutShape(begin, end, strides);\n    // Remove the axes based on shrinkMask.\n    const outShape = size.filter((_, axis) => shrinkAxes.indexOf(axis) === -1);\n\n    const nonStrided = strides.every(v => v === 1);\n    if (nonStrided) {\n      return reshape(slice($x, begin, size), outShape);\n    }\n\n    const res = backend.stridedSlice($x, begin, end, strides);\n    return reshape(res, outShape);\n  };\n\n  const inputs: StridedSliceInputs = {x: $x};\n  const attrs: StridedSliceAttrs = {\n    begin,\n    end,\n    strides,\n    beginMask,\n    endMask,\n    ellipsisMask,\n    newAxisMask,\n    shrinkAxisMask\n  };\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, StridedSlice,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const stridedSlice = op({stridedSlice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tan, TanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes tan of the input `tf.Tensor` element-wise, `tan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.tan().print();  // or tf.tan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tan');\n\n  const inputs: TanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.tan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tan);\n}\nexport const tan = op({tan_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike2D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-2 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor2d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor2d([[1, 2], [3, 4]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor2d(\n    values: TensorLike2D, shape?: [number, number],\n    dtype?: DataType): Tensor2D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 2) {\n    throw new Error('tensor2d() requires shape to have two numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 2 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor2d() requires values to be number[][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor2d() requires shape to be provided when `values` ' +\n        'are a flat/TypedArray');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-4 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor4d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor4d([[[[1], [2]], [[3], [4]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor4d(\n    values: TensorLike4D, shape?: [number, number, number, number],\n    dtype?: DataType): Tensor4D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 4) {\n    throw new Error('tensor4d() requires shape to have four numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 4 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor4d() requires values to be number[][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor4d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor4D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor5D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike5D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-5 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor5d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor5d([[[[[1], [2]], [[3], [4]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor5d(\n    values: TensorLike5D, shape?: [number, number, number, number, number],\n    dtype?: DataType): Tensor5D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 5) {\n    throw new Error('tensor5d() requires shape to have five numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 5 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor5d() requires values to be ' +\n        'number[][][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor5d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor5D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor6D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike6D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-6 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor6d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor6d([[[[[[1],[2]],[[3],[4]]],[[[5],[6]],[[7],[8]]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor6d([1, 2, 3, 4, 5, 6, 7, 8], [1, 1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor6d(\n    values: TensorLike6D,\n    shape?: [number, number, number, number, number, number],\n    dtype?: DataType): Tensor6D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 6) {\n    throw new Error('tensor6d() requires shape to have six numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 6 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor6d() requires values to be number[][][][][][] or ' +\n        'flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor6d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  shape = shape ||\n      inferredShape as [number, number, number, number, number, number];\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor6D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {TopK, TopKAttrs, TopKInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {NumericTensor, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Finds the values and indices of the `k` largest entries along the last\n * dimension.\n *\n * If the input is a vector (rank=1), finds the k largest entries in the vector\n * and outputs their values and indices as vectors. Thus values[j] is the j-th\n * largest entry in input, and its index is indices[j].\n * For higher rank inputs, computes the top k entries along the last dimension.\n *\n * If two elements are equal, the lower-index element appears first.\n *\n * ```js\n * const a = tf.tensor2d([[1, 5], [4, 3]]);\n * const {values, indices} = tf.topk(a);\n * values.print();\n * indices.print();\n * ```\n * @param x 1-D or higher `tf.Tensor` with last dimension being at least `k`.\n * @param k Number of top elements to look for along the last dimension.\n * @param sorted If true, the resulting `k` elements will be sorted by the\n *     values in descending order.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction topk_<T extends Tensor>(\n    x: T|TensorLike, k = 1, sorted = true): {values: T, indices: T} {\n  const $x = convertToTensor(x, 'x', 'topk');\n  if ($x.rank === 0) {\n    throw new Error('topk() expects the input to be of rank 1 or higher');\n  }\n  const lastDim = $x.shape[$x.shape.length - 1];\n  if (k > lastDim) {\n    throw new Error(\n        `'k' passed to topk() must be <= the last dimension (${lastDim}) ` +\n        `but got ${k}`);\n  }\n\n  const inputs: TopKInputs = {x: $x};\n  const attrs: TopKAttrs = {k, sorted};\n\n  const [values, indices] = ENGINE.runKernelFunc(\n      b => b.topk($x as NumericTensor, k, sorted),\n      inputs as {} as NamedTensorMap, null /* grad */, TopK,\n      attrs as {} as NamedAttrMap);\n\n  return {values, indices} as {values: T, indices: T};\n}\n\nexport const topk = op({topk_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a truncated normal\n * distribution.\n *\n * ```js\n * tf.truncatedNormal([2, 2]).print();\n * ```\n *\n * The generated values follow a normal distribution with specified mean and\n * standard deviation, except that values whose magnitude is more than 2\n * standard deviations from the mean are dropped and re-picked.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output tensor.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction truncatedNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type $ { dtype }`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, true /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const truncatedNormal = op({truncatedNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Unique, UniqueAttrs, UniqueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Finds unique elements along an axis of a tensor.\n *\n * It returns a tensor `values` containing all of the unique elements along the\n * `axis` of the given tensor `x` in the same order that they occur along the\n * `axis` in `x`; `x` does not need to be sorted. It also returns a tensor\n * `indices` the same size as the number of the elements in `x` along the `axis`\n * dimension. It contains the index in the unique output `values`.\n *\n * ```js\n * // A 1-D tensor\n * const a = tf.tensor1d([1, 1, 2, 4, 4, 4, 7, 8, 8]);\n * const {values, indices} = tf.unique(a);\n * values.print();   // [1, 2, 4, 7, 8,]\n * indices.print();  // [0, 0, 1, 2, 2, 2, 3, 4, 4]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=0\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 0)\n * values.print();   // [[1, 0, 0],\n *                   //  [2, 0, 0]]\n * indices.print();  // [0, 0, 1]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=1\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 1)\n * values.print();   // [[1, 0],\n *                   //  [1, 0],\n *                   //  [2, 0]]\n * indices.print();  // [0, 1, 1]\n * ```\n * @param x A tensor (int32, string, bool).\n * @param axis The axis of the tensor to find the unique elements.\n * @returns [uniqueElements, indices] (see above for details)\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction unique_<T extends Tensor>(\n    x: T|TensorLike, axis = 0): {values: T, indices: Tensor1D} {\n  // x can be of any dtype, thus null as the last argument.\n  const $x = convertToTensor(x, 'x', 'unique', null);\n  assert($x.rank > 0, () => 'The input tensor must be at least 1D');\n\n  const inputs: UniqueInputs = {x: $x};\n  const attrs: UniqueAttrs = {axis};\n  const [values, indices] = ENGINE.runKernel(\n                                Unique, inputs as {} as NamedTensorMap,\n                                attrs as {} as NamedAttrMap) as [T, Tensor1D];\n  return {values, indices};\n}\n\nexport const unique = op({unique_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {UnsortedSegmentSum, UnsortedSegmentSumAttrs, UnsortedSegmentSumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, isInt} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Computes the sum along segments of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const segmentIds = tf.tensor1d([1, 2, 0, 1], 'int32');\n * const numSegments = 3;\n *\n * x.unsortedSegmentSum(segmentIds, numSegments).print()\n * //or tf.unsortedSegmentSum(x, segmentIds, numSegments)\n * ```\n * @param x The `tf.Tensor` that will be summed along its segments.\n * @param segmentIds A `tf.Tensor1D` whose rank is equal to the rank of `x`'s\n * dimension along the `axis`.  Maps each element of `x` to a segment.\n * @param numSegments The number of distinct `segmentIds`.\n *\n * @doc {heading: 'Operations', subheading: 'Segment'}\n */\nfunction unsortedSegmentSum_<T extends Tensor>(\n    x: T|TensorLike, segmentIds: Tensor1D|TensorLike, numSegments: number): T {\n  const $x = convertToTensor(x, 'x', 'unsortedSegmentSum');\n  const $segmentIds =\n      convertToTensor(segmentIds, 'segmentIds', 'unsortedSegmentSum', 'int32');\n  assert(isInt(numSegments), () => 'numSegments must be of dtype int');\n\n  const inputs: UnsortedSegmentSumInputs = {x: $x, segmentIds: $segmentIds};\n  const attrs: UnsortedSegmentSumAttrs = {numSegments};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.unsortedSegmentSum($x, $segmentIds, numSegments);\n    save([$segmentIds]);\n    return res;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             UnsortedSegmentSum, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const unsortedSegmentSum = op({unsortedSegmentSum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Unpack, UnpackAttrs, UnpackInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Unstacks a `tf.Tensor` of rank-`R` into a list of rank-`(R-1)` `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * tf.unstack(a).forEach(tensor => tensor.print());\n * ```\n *\n * @param x A tensor object.\n * @param axis The axis to unstack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction unstack_(x: Tensor|TensorLike, axis = 0): Tensor[] {\n  const $x = convertToTensor(x, 'x', 'unstack');\n  util.assert(\n      axis >= -$x.shape.length && axis < $x.shape.length,\n      () =>\n          `Axis = ${axis} is not in [-${$x.shape.length}, ${$x.shape.length})`);\n  if (axis < 0) {\n    axis += $x.shape.length;\n  }\n  const inputs: UnpackInputs = {value: $x};\n  const attrs: UnpackAttrs = {axis};\n  const forward: ForwardFunc<Tensor[]> = (backend: KernelBackend) =>\n      backend.unstack($x, axis);\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Unpack,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const unstack = op({unstack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor, Variable} from '../tensor';\nimport {DataType, Rank} from '../types';\n\n/**\n * Creates a new variable with the provided initial value.\n * ```js\n * const x = tf.variable(tf.tensor([1, 2, 3]));\n * x.assign(tf.tensor([4, 5, 6]));\n *\n * x.print();\n * ```\n *\n * @param initialValue Initial value for the tensor.\n * @param trainable If true, optimizers are allowed to update it.\n * @param name Name of the variable. Defaults to a unique id.\n * @param dtype If set, initialValue will be converted to the given type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function variable<R extends Rank>(\n    initialValue: Tensor<R>, trainable = true, name?: string,\n    dtype?: DataType): Variable<R> {\n  return ENGINE.makeVariable(initialValue, trainable, name, dtype) as\n      Variable<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the Where kernel shared between cpu and webgl */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor2D} from '../tensor';\nimport {TypedArray} from '../types';\n\nexport function whereImpl(condShape: number[], condVals: TypedArray): Tensor2D {\n  const indices = [];\n  for (let i = 0; i < condVals.length; i++) {\n    if (condVals[i]) {\n      indices.push(i);\n    }\n  }\n\n  const inBuffer = buffer(condShape, 'int32');\n\n  const out = buffer([indices.length, condShape.length], 'int32');\n  for (let i = 0; i < indices.length; i++) {\n    const loc = inBuffer.indexToLoc(indices[i]);\n    const offset = i * condShape.length;\n    out.values.set(loc, offset);\n  }\n  return out.toTensor() as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {whereImpl} from '../backends/where_impl';\nimport {Tensor, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\n/**\n * Returns the coordinates of true elements of condition.\n *\n * The coordinates are returned in a 2-D tensor where the first dimension (rows)\n * represents the number of true elements, and the second dimension (columns)\n * represents the coordinates of the true elements. Keep in mind, the shape of\n * the output tensor can vary depending on how many true values there are in\n * input. Indices are output in row-major order. The resulting tensor has the\n * shape `[numTrueElems, condition.rank]`.\n *\n * This is analogous to calling the python `tf.where(cond)` without an x or y.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const result = await tf.whereAsync(cond);\n * result.print();\n * ```\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nasync function whereAsync_(condition: Tensor|TensorLike): Promise<Tensor2D> {\n  const $condition =\n      convertToTensor(condition, 'condition', 'whereAsync', 'bool');\n  const vals = await $condition.data();\n  const res = whereImpl($condition.shape, vals);\n  if (condition !== $condition) {\n    $condition.dispose();\n  }\n  return res;\n}\n\nexport const whereAsync = whereAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {gather} from './gather';\nimport {reshape} from './reshape';\nimport {squeeze} from './squeeze';\nimport {whereAsync} from './where_async';\n\n/**\n * Apply boolean mask to tensor.\n *\n * ```js\n * const tensor = tf.tensor2d([1, 2, 3, 4, 5, 6], [3, 2]);\n * const mask = tf.tensor1d([1, 0, 1], 'bool');\n * const result = await tf.booleanMaskAsync(tensor, mask);\n * result.print();\n * ```\n *\n * @param tensor N-D tensor.\n * @param mask K-D boolean tensor, K <= N and K must be known statically.\n * @param axis A 0-D int Tensor representing the axis in tensor to mask from.\n *     By default, axis is 0 which will mask from the first dimension.\n *     Otherwise K + axis <= N.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nasync function booleanMaskAsync_(\n    tensor: Tensor|TensorLike, mask: Tensor|TensorLike,\n    axis?: number): Promise<Tensor> {\n  const $tensor = convertToTensor(tensor, 'tensor', 'boolMask');\n  const $mask = convertToTensor(mask, 'mask', 'boolMask', 'bool');\n\n  const axisFrom = axis == null ? 0 : axis;\n  const maskDim = $mask.rank;\n  const tensorShape = $tensor.shape;\n\n  util.assert(maskDim > 0, () => 'mask cannot be scalar');\n  util.assertShapesMatch(\n      tensorShape.slice(axisFrom, axisFrom + maskDim), $mask.shape,\n      `mask's shape must match the first K dimensions of tensor's shape,`);\n\n  let leadingSize = 1;\n  for (let i = axisFrom; i < axisFrom + maskDim; i++) {\n    leadingSize *= tensorShape[i];\n  }\n  const targetTensorShape =\n      tensorShape.slice(0, axisFrom)\n          .concat([leadingSize], tensorShape.slice(axisFrom + maskDim));\n  const reshapedTensor = reshape($tensor, targetTensorShape);\n  const reshapedMask = reshape($mask, [-1]);\n  const positivePositions = await whereAsync(reshapedMask);\n  const indices = squeeze(positivePositions, [1]);\n\n  const res = gather(reshapedTensor, indices, axisFrom);\n\n  // Ensure no memory leak.\n  if (tensor !== $tensor) {\n    $tensor.dispose();\n  }\n  if (mask !== $mask) {\n    $mask.dispose();\n  }\n  indices.dispose();\n  reshapedTensor.dispose();\n  reshapedMask.dispose();\n  positivePositions.dispose();\n\n  return res;\n}\n\nexport const booleanMaskAsync = booleanMaskAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertShapesMatch} from '../util';\n\nimport {equal} from './equal';\nimport {greater} from './greater';\nimport {greaterEqual} from './greater_equal';\nimport {less} from './less';\nimport {lessEqual} from './less_equal';\nimport {notEqual} from './not_equal';\nimport {op} from './operation';\n\n/**\n * @deprecated\n * Strict version of `tf.notEqual` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction notEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'notEqualStrict');\n  const $b = convertToTensor(b, 'b', 'notEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in notEqualStrict: ');\n  return notEqual($a, $b);\n}\n\n/**\n * @deprecated\n * Strict version of `tf.less` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction lessStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessStrict');\n  const $b = convertToTensor(b, 'b', 'lessStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessStrict: ');\n  return less($a, $b);\n}\n\nfunction equalStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'equalStrict');\n  const $b = convertToTensor(b, 'b', 'equalStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in equalStrict: ');\n  return equal($a, $b);\n}\n\nfunction lessEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessEqualStrict');\n  const $b = convertToTensor(b, 'b', 'lessEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessEqualStrict: ');\n  return lessEqual($a, $b);\n}\n\nfunction greaterStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterStrict');\n  const $b = convertToTensor(b, 'b', 'greaterStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterStrict: ');\n  return greater($a, $b);\n}\n\nfunction greaterEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterEqualStrict');\n  const $b = convertToTensor(b, 'b', 'greaterEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterEqualStrict: ');\n  return greaterEqual($a, $b);\n}\n\nexport const equalStrict = op({equalStrict_});\nexport const greaterEqualStrict = op({greaterEqualStrict_});\nexport const greaterStrict = op({greaterStrict_});\nexport const lessEqualStrict = op({lessEqualStrict_});\nexport const lessStrict = op({lessStrict_});\nexport const notEqualStrict = op({notEqualStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {maximum} from './maximum';\nimport {minimum} from './minimum';\nimport {mod} from './mod';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {squaredDifference} from './squared_difference';\nimport {sub} from './sub';\n\n/**\n * @deprecated\n * Adds two `tf.Tensor`s element-wise, A + B.\n *\n * Inputs must be the same shape. For broadcasting support, use add() instead.\n *\n * @param a The first Tensor to add element-wise.\n * @param b The second Tensor to add element-wise.\n */\nfunction addStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'addStrict');\n  const $b = convertToTensor(b, 'b', 'addStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in addStrict: ');\n  return add($a, $b);\n}\n\n/**\n * @deprecated\n * Subtracts two `tf.Tensor`s element-wise, A - B. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.sub` instead.\n *\n * @param a The first Tensor to subtract element-wise.\n * @param b The second Tensor to subtract element-wise.\n */\nfunction subStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'subStrict');\n  const $b = convertToTensor(b, 'b', 'subStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in subStrict: ');\n  return sub($a, $b);\n}\n\n/**\n * @deprecated\n * Computes the power of one `tf.Tensor` to another. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.pow` instead.\n *\n * @param base The base tensor to pow element-wise.\n * @param exp The exponent tensor to pow element-wise.\n */\nfunction powStrict_<T extends Tensor>(base: T, exp: Tensor): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  util.assertShapesMatch(base.shape, exp.shape, 'Error in powStrict: ');\n  return pow(base, exp);\n}\n\n/**\n * @deprecated\n * Multiplies two `tf.Tensor`s element-wise, A * B.\n *\n * Inputs must be the same shape. For broadcasting support, use `tf.mul`.\n *\n * @param a The first tensor to multiply.\n * @param b The first tensor to multiply. Must have the same\n *    dtype as `a`.\n */\nfunction mulStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'mul');\n  const $b = convertToTensor(b, 'b', 'mul');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in multiplyStrict: ');\n  return mul($a, $b);\n}\n\n/**\n * @deprecated\n * Divides two `tf.Tensor`s element-wise, A / B. Inputs must\n * be the same shape.\n *\n * @param a The first tensor as the numerator for element-wise division.\n * @param b The second tensor as the denominator for element-wise division.\n */\nfunction divStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'div');\n  const $b = convertToTensor(b, 'b', 'div');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in divideStrict: ');\n  return div($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the mod of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use mod().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction modStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'modStrict');\n  const $b = convertToTensor(b, 'b', 'modStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in modStrict: ');\n  return mod($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the min of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use minimum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction minimumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'minimumStrict');\n  const $b = convertToTensor(b, 'b', 'minimumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in minimumStrict: ');\n  return minimum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the max of a and b (`a > b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use maximum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction maximumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'maximumStrict');\n  const $b = convertToTensor(b, 'b', 'maximumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in maximumStrict: ');\n  return maximum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns (a - b) * (a - b) element-wise.\n *\n * Inputs must be the same shape. For broadcasting support, use\n * `tf.squaredDifference` instead.\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n */\nfunction squaredDifferenceStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'squaredDifferenceStrict');\n  const $b = convertToTensor(b, 'b', 'squaredDifferenceStrict');\n  util.assertShapesMatch(\n      $a.shape, $b.shape, 'Error in squaredDifferenceStrict: ');\n  return squaredDifference($a, $b);\n}\n\nexport const addStrict = op({addStrict_});\nexport const divStrict = op({divStrict_});\nexport const maximumStrict = op({maximumStrict_});\nexport const minimumStrict = op({minimumStrict_});\nexport const modStrict = op({modStrict_});\nexport const mulStrict = op({mulStrict_});\nexport const powStrict = op({powStrict_});\nexport const squaredDifferenceStrict = op({squaredDifferenceStrict_});\nexport const subStrict = op({subStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {abs} from './abs';\nimport * as axis_util from './axis_util';\nimport {max} from './max';\nimport {min} from './min';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sqrt} from './sqrt';\nimport {square} from './square';\nimport {sum} from './sum';\n\n/**\n * Computes the norm of scalar, vectors, and matrices.\n * This function can compute several different vector norms (the 1-norm, the\n * Euclidean or 2-norm, the inf-norm, and in general the p-norm for p > 0)\n * and matrix norms (Frobenius, 1-norm, and inf-norm).\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.norm().print();  // or tf.norm(x)\n * ```\n *\n * @param x The input array.\n * @param ord Optional. Order of the norm. Supported norm types are\n * following:\n *\n *  | ord        | norm for matrices         | norm for vectors\n *  |------------|---------------------------|---------------------\n *  |'euclidean' |Frobenius norm             |2-norm\n *  |'fro'       |Frobenius norm\t           |\n *  |Infinity    |max(sum(abs(x), axis=1))   |max(abs(x))\n *  |-Infinity   |min(sum(abs(x), axis=1))   |min(abs(x))\n *  |1           |max(sum(abs(x), axis=0))   |sum(abs(x))\n *  |2           |                           |sum(abs(x)^2)^1/2*\n *\n * @param axis Optional. If axis is null (the default), the input is\n * considered a vector and a single vector norm is computed over the entire\n * set of values in the Tensor, i.e. norm(x, ord) is equivalent\n * to norm(x.reshape([-1]), ord). If axis is a integer, the input\n * is considered a batch of vectors, and axis determines the axis in x\n * over which to compute vector norms. If axis is a 2-tuple of integer it is\n * considered a batch of matrices and axis determines the axes in NDArray\n * over which to compute a matrix norm.\n * @param keepDims Optional. If true, the norm have the same dimensionality\n * as the input.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction norm_(\n    x: Tensor|TensorLike, ord: number|'euclidean'|'fro' = 'euclidean',\n    axis: number|number[] = null, keepDims = false): Tensor {\n  x = convertToTensor(x, 'x', 'norm');\n\n  const norm = normImpl(x, ord, axis);\n  let keepDimsShape = norm.shape;\n  if (keepDims) {\n    const axes = parseAxisParam(axis, x.shape);\n    keepDimsShape = axis_util.expandShapeToKeepDim(norm.shape, axes);\n  }\n  return reshape(norm, keepDimsShape);\n}\n\nfunction normImpl(\n    x: Tensor, p: number|string, axis: number|number[] = null): Tensor {\n  if (x.rank === 0) {\n    return abs(x);\n  }\n\n  // consider vector when no axis is specified\n  if (x.rank !== 1 && axis === null) {\n    return normImpl(reshape(x, [-1]), p, axis);\n  }\n\n  // vector\n  if (x.rank === 1 || typeof axis === 'number' ||\n      Array.isArray(axis) && axis.length === 1) {\n    if (p === 1) {\n      return sum(abs(x), axis);\n    }\n    if (p === Infinity) {\n      return max(abs(x), axis);\n    }\n    if (p === -Infinity) {\n      return min(abs(x), axis);\n    }\n    if (p === 'euclidean' || p === 2) {\n      // norm(x, 2) = sum(abs(xi) ^ 2) ^ 1/2\n      return sqrt(sum(pow(abs(x), scalar(2, 'int32')), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  // matrix (assumption axis[0] < axis[1])\n  if (Array.isArray(axis) && axis.length === 2) {\n    if (p === 1) {\n      return max(sum(abs(x), axis[0]), axis[1] - 1);\n    }\n    if (p === Infinity) {\n      return max(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === -Infinity) {\n      return min(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === 'fro' || p === 'euclidean') {\n      // norm(x) = sqrt(sum(pow(x, 2)))\n      return sqrt(sum(square(x), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  throw new Error(`Error in norm: invalid axis: ${axis}`);\n}\n\nexport const norm = op({norm_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor} from '../tensor';\nimport {assertTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {scalar} from './scalar';\nimport {sub} from './sub';\n\n/**\n * Compute the moving average of a variable.\n *\n * Without zeroDebias, the moving average operation is defined by:\n *   `v += delta`\n * where\n *   `delta = (1 - decay) * (x - v)`\n *\n * With zeroDebias (default), the `delta` term is scaled to debias the\n * effect of the (assumed) zero-initialization of `v`.\n *   `delta /= (1 - decay ^ step)`\n *\n * For more details on the zero-debiasing algorithm, see:\n *   https://arxiv.org/abs/1412.6980\n *\n * Note that this function is completely stateless and does not keep track of\n * step count. The step count needs to be maintained by the caller and passed\n * in as `step`.\n *\n * @param v The current moving average value.\n * @param x New input value, must have the same shape and dtype as `v`.\n * @param decay The decay factor. Typical values are 0.95 and 0.99.\n * @param step Step count.\n * @param zeroDebias: Whether zeroDebias is to be performed (default: `true`).\n * @returns The new moving average value.\n *\n * @doc {heading: 'Operations', subheading: 'Moving Average'}\n */\nfunction movingAverage_<T extends Tensor>(\n    v: T|TensorLike, x: T|TensorLike, decay: number|Scalar,\n    step?: number|Scalar, zeroDebias = true): T {\n  const $v = convertToTensor(v, 'v', 'movingAverage');\n  const $x = convertToTensor(x, 'x', 'movingAverage');\n  const $decay = convertToTensor(decay, 'decay', 'movingAverage');\n\n  assertTypesMatch($v, $x);\n  util.assert(\n      util.arraysEqual($v.shape, $x.shape), () => 'Shape mismatch in v and x');\n\n  const one = scalar(1);\n  const oneMinusDecay = sub(one, $decay);\n\n  let update = mul(sub($x, $v), oneMinusDecay);\n  if (zeroDebias) {\n    util.assert(\n        step != null, () => 'When using zeroDebias: true, step is required.');\n    const $step = convertToTensor(step, 'step', 'movingAverage');\n    update = div(update, sub(one, pow($decay, $step)));\n  }\n  return add($v, update);\n}\n\nexport const movingAverage = op({movingAverage_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ScatterNd, ScatterNdAttrs, ScatterNdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as scatter_nd_util from './scatter_nd_util';\n\n/**\n * Creates a new tensor by applying sparse updates to individual\n * values or slices within a zero tensor of the given shape tensor according to\n * indices. This operator is the inverse of the `tf.gatherND` operator which\n * extracts values or slices from a given tensor.\n *\n * ```js\n * const indices = tf.tensor2d([4, 3, 1, 7], [4, 1], 'int32');\n * const updates = tf.tensor1d([9, 10, 11, 12]);\n * const shape = [8];\n * tf.scatterND(indices, updates, shape).print() //[0, 11, 0, 10, 9, 0, 0, 12]\n * ```\n *\n * @param indices The tensor contains the indices into the output tensor.\n * @param updates The tensor contains the value for the indices.\n * @param shape: The shape of the output tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction scatterND_<R extends Rank>(\n    indices: Tensor|TensorLike, updates: Tensor|TensorLike,\n    shape: ShapeMap[R]): Tensor<R> {\n  const $indices = convertToTensor(indices, 'indices', 'scatterND', 'int32');\n  const $updates = convertToTensor(updates, 'updates', 'scatterND');\n  scatter_nd_util.validateInput($updates, $indices, shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.scatterND($indices, $updates, shape);\n  };\n\n  const inputs: ScatterNdInputs = {indices: $indices, updates: $updates};\n  const attrs: ScatterNdAttrs = {shape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             ScatterNd, attrs as {} as NamedAttrMap) as Tensor<R>;\n}\n\nexport const scatterND = op({scatterND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\n\n/**\n * Validate sparseToDense inputs.\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape number[]. Shape of the dense output tensor.\n * @param validateIndices boolean. indice validation is not supported, error\n * will be thrown if it is set.\n */\nexport function validateInput(\n    sparseIndices: Tensor, sparseValues: Tensor, outputShape: number[],\n    defaultValues: Tensor) {\n  if (sparseIndices.dtype !== 'int32') {\n    throw new Error(\n        'tf.sparseToDense() expects the indices to be int32 type,' +\n        ` but the dtype was ${sparseIndices.dtype}.`);\n  }\n  if (sparseIndices.rank > 2) {\n    throw new Error(\n        'sparseIndices should be a scalar, vector, or matrix,' +\n        ` but got shape ${sparseIndices.shape}.`);\n  }\n\n  const numElems = sparseIndices.rank > 0 ? sparseIndices.shape[0] : 1;\n  const numDims = sparseIndices.rank > 1 ? sparseIndices.shape[1] : 1;\n\n  if (outputShape.length !== numDims) {\n    throw new Error(\n        'outputShape has incorrect number of elements:,' +\n        ` ${outputShape.length}, should be: ${numDims}.`);\n  }\n\n  const numValues = sparseValues.size;\n  if (!(sparseValues.rank === 0 ||\n        sparseValues.rank === 1 && numValues === numElems)) {\n    throw new Error(\n        'sparseValues has incorrect shape ' +\n        `${sparseValues.shape}, should be [] or [${numElems}]`);\n  }\n\n  if (sparseValues.dtype !== defaultValues.dtype) {\n    throw new Error('sparseValues.dtype must match defaultValues.dtype');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {SparseToDense, SparseToDenseAttrs, SparseToDenseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport * as sparse_to_dense from '../ops/sparse_to_dense_util';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ScalarLike, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Converts a sparse representation into a dense tensor.\n *\n * Builds an array dense with shape outputShape such that:\n *\n * // If sparseIndices is scalar\n * dense[i] = (i == sparseIndices ? sparseValues : defaultValue)\n *\n * // If sparseIndices is a vector, then for each i\n * dense[sparseIndices[i]] = sparseValues[i]\n *\n * // If sparseIndices is an n by d matrix, then for each i in [0, n)\n * dense[sparseIndices[i][0], ..., sparseIndices[i][d-1]] = sparseValues[i]\n * All other values in dense are set to defaultValue. If sparseValues is a\n * scalar, all sparse indices are set to this single value.\n *\n * If indices are repeated the final value is summed over all values for those\n * indices.\n *\n * ```js\n * const indices = tf.tensor1d([4, 5, 6, 1, 2, 3], 'int32');\n * const values = tf.tensor1d([10, 11, 12, 13, 14, 15], 'float32');\n * const shape = [8];\n * tf.sparseToDense(indices, values, shape).print();\n * ```\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape Shape of the dense output tensor. the type is inferred.\n * @param defaultValue Scalar. Value to set for indices not specified in\n * sparseIndices. Defaults to zero.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction sparseToDense_<R extends Rank>(\n    sparseIndices: Tensor|TensorLike, sparseValues: Tensor|TensorLike,\n    outputShape: ShapeMap[R], defaultValue: Scalar|ScalarLike = 0): Tensor<R> {\n  const $sparseIndices =\n      convertToTensor(sparseIndices, 'sparseIndices', 'sparseToDense', 'int32');\n  const $sparseValues =\n      convertToTensor(sparseValues, 'sparseValues', 'sparseToDense');\n  const $defaultValue = convertToTensor(\n      defaultValue, 'defaultValue', 'sparseToDense', $sparseValues.dtype);\n\n  sparse_to_dense.validateInput(\n      $sparseIndices, $sparseValues, outputShape, $defaultValue);\n\n  const inputs: SparseToDenseInputs = {\n    sparseIndices: $sparseIndices,\n    sparseValues: $sparseValues,\n    defaultValue: $defaultValue\n  };\n\n  const attrs: SparseToDenseAttrs = {outputShape};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.sparseToDense(\n          $sparseIndices, $sparseValues, outputShape, $defaultValue),\n      inputs as {} as NamedTensorMap, null /* grad */, SparseToDense,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const sparseToDense = op({sparseToDense_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherNd, GatherNdInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Gather slices from input tensor into a Tensor with shape specified by\n * `indices`.\n *\n * `indices` is an K-dimensional integer tensor, best thought of as a\n * (K-1)-dimensional tensor of indices into input, where each element defines a\n * slice of input:\n * output[\\\\(i_0, ..., i_{K-2}\\\\)] = input[indices[\\\\(i_0, ..., i_{K-2}\\\\)]]\n *\n * Whereas in `tf.gather`, `indices` defines slices into the first dimension of\n * input, in `tf.gatherND`, `indices` defines slices into the first N dimensions\n * of input, where N = indices.shape[-1].\n *\n * The last dimension of indices can be at most the rank of input:\n * indices.shape[-1] <= input.rank\n *\n * The last dimension of `indices` corresponds to elements\n * (if indices.shape[-1] == input.rank) or slices\n * (if indices.shape[-1] < input.rank) along dimension indices.shape[-1] of\n * input.\n * The output tensor has shape\n * indices.shape[:-1] + input.shape[indices.shape[-1]:]\n *\n * Note that on CPU, if an out of bound index is found, an error is returned. On\n * GPU, if an out of bound index is found, a 0 is stored in the corresponding\n * output value.\n *\n * ```js\n * const indices = tf.tensor2d([0, 1, 1, 0], [2,2], 'int32');\n * const input = tf.tensor2d([9, 10, 11, 12], [2, 2]);\n * tf.gatherND(input, indices).print() // [10, 11]\n * ```\n *\n * @param x The tensor from which to gather values.\n * @param indices Index tensor, must be of type int32.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction gatherND_(x: Tensor|TensorLike, indices: Tensor|TensorLike): Tensor {\n  const $indices = convertToTensor(indices, 'indices', 'gatherND', 'int32');\n  const $x = convertToTensor(x, 'x', 'gatherND');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.gatherND($x, $indices);\n  };\n\n  const inputs: GatherNdInputs = {params: $x, indices: $indices};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */, GatherNd);\n}\n\nexport const gatherND = op({gatherND_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\n/**\n * Normalize noise shape based on provided tensor and noise shape.\n *\n * @param x Tensor.\n * @param noiseShape The shape for the randomly generated keep/drop flags, as\n *   an array of numbers. Optional.\n * @returns Normalized noise shape.\n */\nexport function getNoiseShape(x: Tensor, noiseShape?: number[]): number[] {\n  if (noiseShape == null) {\n    return x.shape.slice();\n  }\n  if (util.arraysEqual(x.shape, noiseShape)) {\n    return noiseShape;\n  }\n  if (x.shape.length === noiseShape.length) {\n    const newDimension: number[] = [];\n    for (let i = 0; i < x.shape.length; i++) {\n      if (noiseShape[i] == null && x.shape[i] != null) {\n        newDimension.push(x.shape[i]);\n      } else {\n        newDimension.push(noiseShape[i]);\n      }\n    }\n    return newDimension;\n  }\n\n  return noiseShape;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {getNoiseShape} from './dropout_util';\nimport {floor} from './floor';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {randomUniform} from './random_uniform';\n\n/**\n * Computes dropout.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 2, 1]);\n * const rate = 0.75;\n * const output = tf.dropout(x, rate);\n * output.print();\n * ```\n *\n * @param x A floating point Tensor or TensorLike.\n * @param rate A float in the range [0, 1). The probability that each element\n *   of x is discarded.\n * @param noiseShape An array of numbers of type int32, representing the\n * shape for randomly generated keep/drop flags. If the noiseShape has null\n * value, it will be automatically replaced with the x's relative dimension\n * size. Optional.\n * @param seed Used to create random seeds. Optional.\n * @returns A Tensor of the same shape of x.\n *\n * @doc {heading: 'Operations', subheading: 'Dropout'}\n */\nfunction dropout_(\n    x: Tensor|TensorLike, rate: number, noiseShape?: number[],\n    seed?: number|string): Tensor {\n  const $x = convertToTensor(x, 'x', 'dropout');\n\n  util.assert(\n      $x.dtype === 'float32',\n      () => `x has to be a floating point tensor since it's going to be ` +\n          `scaled, but got a ${$x.dtype} tensor instead.`);\n  util.assert(\n      rate >= 0 && rate < 1,\n      () => `rate must be a float in the range [0, 1), but got ${rate}.`);\n\n  if (rate === 0) {\n    return x instanceof Tensor ? $x.clone() : $x;\n  }\n\n  const $noiseShape = getNoiseShape($x, noiseShape);\n  const keepProb = 1 - rate;\n  const multiplier = div(\n      floor(add(randomUniform($noiseShape, 0, 1, 'float32', seed), keepProb)),\n      keepProb);\n\n  return mul($x, multiplier);\n}\n\nexport const dropout = op({dropout_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {tensor1d} from './tensor1d';\n\nexport function enclosingPowerOfTwo(value: number) {\n  // Return 2**N for integer N such that 2**N >= value.\n  return Math.floor(Math.pow(2, Math.ceil(Math.log(value) / Math.log(2.0))));\n}\n\nexport function cosineWindow(\n    windowLength: number, a: number, b: number): Tensor1D {\n  const even = 1 - windowLength % 2;\n  const newValues = new Float32Array(windowLength);\n  for (let i = 0; i < windowLength; ++i) {\n    const cosArg = (2.0 * Math.PI * i) / (windowLength + even - 1);\n    newValues[i] = a - b * Math.cos(cosArg);\n  }\n  return tensor1d(newValues, 'float32');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch, getTypedArrayFromDType} from '../util';\nimport {tensor} from './tensor';\n\n/**\n * Returns whether the targets are in the top K predictions.\n *\n * ```js\n * const predictions = tf.tensor2d([[20, 10, 40, 30], [30, 50, -20, 10]]);\n * const targets = tf.tensor1d([2, 0]);\n * const precision = await tf.inTopKAsync(predictions, targets);\n * precision.print();\n * ```\n * @param predictions 2-D or higher `tf.Tensor` with last dimension being\n *     at least `k`.\n * @param targets 1-D or higher `tf.Tensor`.\n * @param k Optional Number of top elements to look at for computing precision,\n *     default to 1.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nasync function inTopKAsync_<T extends Tensor, U extends Tensor>(\n    predictions: T|TensorLike, targets: U|TensorLike, k = 1): Promise<U> {\n  const $predictions = convertToTensor(predictions, 'predictions', 'inTopK');\n  const $targets = convertToTensor(targets, 'targets', 'inTopK');\n\n  assert(\n      $predictions.rank > 1,\n      () => 'inTopK() expects the predictions to be of rank 2 or higher, ' +\n          `but got ${$predictions.rank}`);\n  assert(\n      $predictions.rank - 1 === $targets.rank,\n      () => `predictions rank should be 1 larger than ` +\n          `targets rank, but got predictions rank ` +\n          `${$predictions.rank} and targets rank ${$targets.rank}`);\n  assertShapesMatch(\n      $predictions.shape.slice(0, $predictions.shape.length - 1),\n      $targets.shape,\n      `predictions's shape should be align with the targets' shape, ` +\n          'except the last dimension.');\n  const lastDim = $predictions.shape[$predictions.shape.length - 1];\n  assert(\n      k > 0 && k <= lastDim,\n      () => `'k' passed to inTopK() must be > 0 && <= the predictions last ` +\n          `dimension (${lastDim}), but got ${k}`);\n\n  const predictionsVals = await $predictions.data();\n  const targetsVals = await $targets.data();\n\n  // Reshape predictionsVals into a 2d tensor [batch, lastDim]\n  // and look up topK along lastDim.\n  const [batch, size] = [predictionsVals.length / lastDim, lastDim];\n  const precision = getTypedArrayFromDType('bool', batch);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = predictionsVals.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    precision[b] = 0;\n    for (let i = 0; i < k; i++) {\n      if (valAndInd[i].index === targetsVals[b]) {\n        precision[b] = 1;\n        break;\n      }\n    }\n  }\n\n  if (predictions !== $predictions) {\n    $predictions.dispose();\n  }\n  if (targets !== $targets) {\n    $targets.dispose();\n  }\n\n  // Output precision has the same shape as targets.\n  return tensor(precision, $targets.shape, 'bool') as U;\n}\n\nexport const inTopKAsync = inTopKAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropFilter, Conv2DBackpropFilterAttrs, Conv2DBackpropFilterInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 2D convolution.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     [batch, height, width, inChannels]. If rank 3, batch of 1 is assumed.\n * @param dy The dy image, of rank 4 or rank 3, of shape\n *     [batch, height, width, outDepth]. If rank 3, batch of 1 is assumed.\n * @param filterShape The shape of the filter, length 4,\n *     [filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction conv2DBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2dDerFilter: input must be rank 4, but got shape ` +\n          `${x4D.shape}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerFilter: dy must be rank 4, but got shape ` +\n          `${dy4D.shape}.`);\n  util.assert(\n      filterShape.length === 4,\n      () => `Error in conv2dDerFilter: filterShape must be length 4, but got ` +\n          `${filterShape}.`);\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filterShape[2],\n      () => `Error in conv2dDerFilter: depth of input ${inDepth}) must ` +\n          `match input depth in filter (${filterShape[2]}.`);\n  util.assert(\n      outDepth === filterShape[3],\n      () => `Error in conv2dDerFilter: depth of dy (${outDepth}) must ` +\n          `match output depth for filter (${filterShape[3]}).`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerFilter: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, filterShape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    return backend.conv2dDerFilter(x4D, dy4D, convInfo);\n  };\n\n  const inputs: Conv2DBackpropFilterInputs = {x: x4D, dy: dy4D};\n  const attrs:\n      Conv2DBackpropFilterAttrs = {strides, pad, dataFormat, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv2DBackpropFilter, attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const conv2DBackpropFilter = op({conv2DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\nimport * as broadcast_util from './broadcast_util';\nimport {elu} from './elu';\nimport {Activation} from './fused_types';\nimport {mul} from './mul';\nimport {prelu} from './prelu';\nimport {relu} from './relu';\nimport {relu6} from './relu6';\nimport {reshape} from './reshape';\nimport {step} from './step';\nimport {sum} from './sum';\n\n// Returns gradient for fused activation.\nexport function getFusedDyActivation(\n    dy: Tensor, y: Tensor, activation: Activation): Tensor {\n  if (activation == null || activation === 'linear') {\n    return dy;\n  }\n  if (activation === 'relu') {\n    return mul(dy, step(y));\n  }\n  throw new Error(\n      `Cannot compute gradient for fused activation ${activation}.`);\n}\n\n// Returns gradient for fused bias.\nexport function getFusedBiasGradient(\n    bias: Tensor, dyActivation: Tensor): Tensor {\n  let res = dyActivation;\n  const reduceAxes =\n      broadcast_util.getReductionAxes(bias.shape, dyActivation.shape);\n  if (reduceAxes.length > 0) {\n    res = sum(res, reduceAxes);\n  }\n  return reshape(res, bias.shape);\n}\n\nexport function applyActivation(\n    x: Tensor, activation: Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return x;\n  } else if (activation === 'relu') {\n    return relu(x);\n  } else if (activation === 'elu') {\n    return elu(x);\n  } else if (activation === 'relu6') {\n    return relu6(x);\n  } else if (activation === 'prelu') {\n    return prelu(x, preluActivationWeights);\n  }\n  throw new Error(`Unknown fused activation ${activation}.`);\n}\n\n// Whether we should call fused ops.\nexport const shouldFuse = (gradientDepth: number, activation: Activation) => {\n  const gradientMode = gradientDepth > 0;\n  return !gradientMode || activation === 'linear';\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedConv2D, FusedConv2DAttrs, FusedConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {conv2d as unfusedConv2d} from '../conv2d';\nimport {conv2DBackpropFilter} from '../conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../conv2d_backprop_input';\nimport * as conv_util from '../conv_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes a 2D convolution over the input x, optionally fused with adding a\n * bias and applying an activation.\n *\n * ```js\n * const inputDepth = 2;\n * const inShape = [2, 2, 2, inputDepth];\n * const outputDepth = 2;\n * const fSize = 1;\n * const pad = 0;\n * const strides = 1;\n *\n * const x = tf.tensor4d( [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,\n * 16], inShape);\n * const w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth,\n * outputDepth]);\n *\n * tf.fused.conv2d({ x, filter: w, strides, pad, dataFormat: 'NHWC',\n * dilations: [1, 1], bias: tf.scalar(5), activation: 'relu' }).print();\n * ```\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid` output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`) to be\n *     applied\n *      after biasAdd.\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  activation = activation || 'linear';\n\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused conv2d: input must be rank 4, but got rank ` +\n          `${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${x4D.shape[3]}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in conv2d: got dataFormat of ${\n          dataFormat} but only NHWC is currently supported.`);\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused conv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    const [$filter, x4D, y, $bias] =\n        saved as [Tensor4D, Tensor4D, Tensor4D, Tensor];\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused conv2D: ' +\n            `dilation rates greater than 1 ` +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const xDer =\n        conv2DBackpropInput(x4D.shape, dyActivation, $filter, strides, pad);\n    const filterDer =\n        conv2DBackpropFilter(x4D, dyActivation, $filter.shape, strides, pad);\n    const der: Tensor[] = [xDer, filterDer];\n\n    if ($bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      der.push(biasDer);\n    }\n    return der;\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedConv2d({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n\n  const attrs: FusedConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const conv2d = op({fusedConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropFilterInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    convInfo: conv_util.Conv2DInfo): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerFilter(x4D, dy4D, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropFilterInputs = {x: x4D, dy: dy4D};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             DepthwiseConv2dNativeBackpropFilter) as Tensor4D;\n}\n\nexport const depthwiseConv2dNativeBackpropFilter =\n    op({depthwiseConv2dNativeBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropInput, DepthwiseConv2dNativeBackpropInputInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, convInfo: conv_util.Conv2DInfo): T {\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerInput(dy4D, filter, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropInputInputs = {dy: dy4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null,\n      DepthwiseConv2dNativeBackpropInput);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2dNativeBackpropInput =\n    op({depthwiseConv2dNativeBackpropInput_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedDepthwiseConv2D, FusedDepthwiseConv2DAttrs, FusedDepthwiseConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport * as conv_util from '../conv_util';\nimport {depthwiseConv2d as unfusedDepthwiseConv2d} from '../depthwise_conv2d';\nimport {depthwiseConv2dNativeBackpropFilter} from '../depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../depthwise_conv2d_native_backprop_input';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes depthwise 2D convolution, optionally fused with adding a\n * bias and applying an activation.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`).\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedDepthwiseConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedDepthwiseConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused depthwiseConv2d: filter must be rank 4, ` +\n          `but got rank ${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in fused depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () =>\n          'Error in fused depthwiseConv2d: Either strides or dilations must ' +\n          `be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused depthwiseConv2d: pad must be an integer when ` +\n            `using dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n      true /* depthwise */);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused depthwiseConv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused depthwiseConv2d: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${dilations}'`);\n    const [$filter, x4D, y, bias] = saved;\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    const xDer = depthwiseConv2dNativeBackpropInput(\n        (x4D as Tensor4D).shape, dyActivation, $filter as Tensor4D, convInfo);\n    const filterDer = depthwiseConv2dNativeBackpropFilter(\n        x4D as Tensor4D, dyActivation, ($filter as Tensor4D).shape, convInfo);\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [xDer, filterDer, biasDer];\n    }\n    return [xDer, filterDer];\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedDepthwiseConv2D({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedDepthwiseConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: FusedDepthwiseConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const depthwiseConv2d = op({fusedDepthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {_FusedMatMul, _FusedMatMulAttrs, _FusedMatMulInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {matMul as unfusedMatMul} from '../mat_mul';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes the dot product of two matrices with optional activation and bias.\n *\n * ```js\n * const a = tf.tensor2d([-1, -2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const bias = tf.tensor2d([1, 2], [1, 2]);\n *\n * tf.fused.matMul({a, b, bias, activation: 'relu'}).print();\n * ```\n *\n * @param obj An object with the following properties:\n * - `a` First matrix in dot product operation.\n * - `b` Second matrix in dot product operation.\n * - `transposeA` If true, `a` is transposed before multiplication.\n * - `transposeB` If true, `b` is transposed before multiplication.\n * - `bias` Matrix to be added to the result.\n * - `activation` Name of activation kernel (defaults to `linear`).\n * - `preluActivationWeights` Tensor of prelu weights.\n */\nfunction fusedMatMul_<T extends Tensor>({\n  a,\n  b,\n  transposeA = false,\n  transposeB = false,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  a: T|TensorLike,\n  b: T|TensorLike,\n  transposeA?: boolean,\n  transposeB?: boolean,\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedMatMul(a, b, transposeA, transposeB);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  let $a = convertToTensor(a, 'a', 'fused matMul');\n  let $b = convertToTensor(b, 'b', 'fused matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () =>\n          `Error in fused matMul: inputs must have the same rank of at least ` +\n          `2, got ranks ${$a.rank} and ${$b.rank}.`);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in fused matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in fused matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D: Tensor3D = transposeA ?\n      reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n      reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D: Tensor3D = transposeB ?\n      reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n      reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused matMul');\n    [$bias] = makeTypesMatch($bias, $a);\n\n    broadcast_util.assertAndGetBroadcastShape(outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused matMul');\n  }\n\n  const grad = (dy: Tensor3D, saved: Tensor[]) => {\n    const [a3D, b3D, y, $bias] = saved;\n    // we reshape dy because the result of the forward is not\n    // necessarily going to be a 3d tensor due to a reshape done at the end of\n    // the customOp.\n    const dyActivation =\n        getFusedDyActivation(reshape(dy, y.shape), y, activation);\n    let aDer: Tensor;\n    let bDer: Tensor;\n\n    if (!transposeA && !transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, true, false);\n    } else if (!transposeA && transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, false);\n      bDer = unfusedMatMul(dyActivation, a3D, true, false);\n    } else if (transposeA && !transposeB) {\n      aDer = unfusedMatMul(b3D, dyActivation, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, false, false);\n    } else {\n      aDer = unfusedMatMul(b3D, dyActivation, true, true);\n      bDer = unfusedMatMul(dyActivation, a3D, true, true);\n    }\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [aDer, bDer, biasDer];\n    } else {\n      return [aDer, bDer];\n    }\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const y = backend.fusedBatchMatMul({\n      a: a3D,\n      b: b3D,\n      transposeA,\n      transposeB,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return y;\n  };\n\n  const inputs: _FusedMatMulInputs = {\n    a: a3D,\n    b: b3D,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: _FusedMatMulAttrs = {transposeA, transposeB, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((a3D: Tensor3D, b3D: Tensor3D, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n    return customOp(a3D, b3D) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (a3D: Tensor3D, b3D: Tensor3D, $bias: Tensor, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res, $bias]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n\n    return customOpWithBias(a3D, b3D, $bias) as T;\n  }\n}\n\nexport const matMul = op({fusedMatMul_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {conv2d} from './fused/conv2d';\nimport {depthwiseConv2d} from './fused/depthwise_conv2d';\nimport {matMul} from './fused/mat_mul';\nimport {Activation} from './fused_types';\n\nexport {Activation, conv2d, depthwiseConv2d, matMul};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a hamming window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hammingWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hammingWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.54, 0.46);\n}\nexport const hammingWindow = op({hammingWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a Hann window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hannWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hannWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.5, 0.5);\n}\n\nexport const hannWindow = op({hannWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {fill} from '../fill';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {tensor2d} from '../tensor2d';\n\n/**\n * Expands input into frames of frameLength.\n * Slides a window size with frameStep.\n *\n * ```js\n * tf.signal.frame([1, 2, 3], 2, 1).print();\n * ```\n * @param signal The input tensor to be expanded\n * @param frameLength Length of each frame\n * @param frameStep The frame hop size in samples.\n * @param padEnd Whether to pad the end of signal with padValue.\n * @param padValue An number to use where the input signal does\n *     not exist when padEnd is True.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction frame_(\n    signal: Tensor1D, frameLength: number, frameStep: number, padEnd = false,\n    padValue = 0): Tensor {\n  let start = 0;\n  const output: Tensor[] = [];\n  while (start + frameLength <= signal.size) {\n    output.push(slice(signal, start, frameLength));\n    start += frameStep;\n  }\n\n  if (padEnd) {\n    while (start < signal.size) {\n      const padLen = (start + frameLength) - signal.size;\n      const pad = concat([\n        slice(signal, start, frameLength - padLen), fill([padLen], padValue)\n      ]);\n      output.push(pad);\n      start += frameStep;\n    }\n  }\n\n  if (output.length === 0) {\n    return tensor2d([], [0, frameLength]);\n  }\n\n  return reshape(concat(output), [output.length, frameLength]);\n}\nexport const frame = op({frame_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {enclosingPowerOfTwo} from '../signal_ops_util';\nimport {slice} from '../slice';\nimport {rfft} from '../spectral/rfft';\n\nimport {frame} from './frame';\nimport {hannWindow} from './hann_window';\n\n/**\n * Computes the Short-time Fourier Transform of signals\n * See: https://en.wikipedia.org/wiki/Short-time_Fourier_transform\n *\n * ```js\n * const input = tf.tensor1d([1, 1, 1, 1, 1])\n * tf.signal.stft(input, 3, 1).print();\n * ```\n * @param signal 1-dimensional real value tensor.\n * @param frameLength The window length of samples.\n * @param frameStep The number of samples to step.\n * @param fftLength The size of the FFT to apply.\n * @param windowFn A callable that takes a window length and returns 1-d tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction stft_(\n    signal: Tensor1D, frameLength: number, frameStep: number,\n    fftLength?: number,\n    windowFn: (length: number) => Tensor1D = hannWindow): Tensor {\n  if (fftLength == null) {\n    fftLength = enclosingPowerOfTwo(frameLength);\n  }\n  const framedSignal = frame(signal, frameLength, frameStep);\n  const windowedSignal = mul(framedSignal, windowFn(frameLength));\n  const output: Tensor[] = [];\n  for (let i = 0; i < framedSignal.shape[0]; i++) {\n    output.push(\n        rfft(slice(windowedSignal, [i, 0], [1, frameLength]), fftLength));\n  }\n  return concat(output);\n}\nexport const stft = op({stft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {CropAndResize, CropAndResizeAttrs, CropAndResizeInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor1D, Tensor2D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Extracts crops from the input image tensor and resizes them using bilinear\n * sampling or nearest neighbor sampling (possibly with aspect ratio change)\n * to a common output size specified by crop_size.\n *\n * @param image 4d tensor of shape `[batch,imageHeight,imageWidth, depth]`,\n *     where imageHeight and imageWidth must be positive, specifying the\n *     batch of images from which to take crops\n * @param boxes 2d float32 tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the normalized\n *     coordinates of the box in the boxInd[i]'th image in the batch\n * @param boxInd 1d int32 tensor of shape `[numBoxes]` with values in range\n *     `[0, batch)` that specifies the image that the `i`-th box refers to.\n * @param cropSize 1d int32 tensor of 2 elements `[cropHeigh, cropWidth]`\n *     specifying the size to which all crops are resized to.\n * @param method Optional string from `'bilinear' | 'nearest'`,\n *     defaults to bilinear, which specifies the sampling method for resizing\n * @param extrapolationValue A threshold for deciding when to remove boxes based\n *     on score. Defaults to 0.\n * @return A 4D tensor of the shape `[numBoxes,cropHeight,cropWidth,depth]`\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction cropAndResize_(\n    image: Tensor4D|TensorLike,\n    boxes: Tensor2D|TensorLike,\n    boxInd: Tensor1D|TensorLike,\n    cropSize: [number, number],\n    method?: 'bilinear'|'nearest',\n    extrapolationValue?: number,\n    ): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'cropAndResize');\n  const $boxes = convertToTensor(boxes, 'boxes', 'cropAndResize', 'float32');\n  const $boxInd = convertToTensor(boxInd, 'boxInd', 'cropAndResize', 'int32');\n  method = method || 'bilinear';\n  extrapolationValue = extrapolationValue || 0;\n\n  const numBoxes = $boxes.shape[0];\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in cropAndResize: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n  util.assert(\n      $boxes.rank === 2 && $boxes.shape[1] === 4,\n      () => `Error in cropAndResize: boxes must be have size [${numBoxes},4] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      $boxInd.rank === 1 && $boxInd.shape[0] === numBoxes,\n      () => `Error in cropAndResize: boxInd must be have size [${numBoxes}] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      cropSize.length === 2,\n      () => `Error in cropAndResize: cropSize must be of length 2, but got ` +\n          `length ${cropSize.length}.`);\n  util.assert(\n      cropSize[0] >= 1 && cropSize[1] >= 1,\n      () => `cropSize must be atleast [1,1], but was ${cropSize}`);\n  util.assert(\n      method === 'bilinear' || method === 'nearest',\n      () => `method must be bilinear or nearest, but was ${method}`);\n\n  const forward: ForwardFunc<Tensor4D> = (backend) => backend.cropAndResize(\n      $image, $boxes, $boxInd, cropSize, method, extrapolationValue);\n\n  const inputs:\n      CropAndResizeInputs = {image: $image, boxes: $boxes, boxInd: $boxInd};\n  const attrs: CropAndResizeAttrs = {method, extrapolationValue, cropSize};\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, CropAndResize,\n      attrs as {} as NamedAttrMap);\n  return res;\n}\n\nexport const cropAndResize = op({cropAndResize_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FlipLeftRight, FlipLeftRightInputs} from '../../kernel_names';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {op} from '../operation';\n\n/**\n * Flips the image left to right. Currently available in the CPU, WebGL, and\n * WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n */\n/** @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'} */\nfunction flipLeftRight_(image: Tensor4D|TensorLike): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'flipLeftRight', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in flipLeftRight: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: FlipLeftRightInputs = {image: $image};\n  const res =\n      ENGINE.runKernel(FlipLeftRight, inputs as {} as NamedTensorMap, {});\n  return res as Tensor4D;\n}\n\nexport const flipLeftRight = op({flipLeftRight_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Rotates the input image tensor counter-clockwise with an optional offset\n * center of rotation. Currently available in the CPU, WebGL, and WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n * @param radians The amount of rotation.\n * @param fillValue The value to fill in the empty space leftover\n *     after rotation. Can be either a single grayscale value (0-255), or an\n *     array of three numbers `[red, green, blue]` specifying the red, green,\n *     and blue channels. Defaults to `0` (black).\n * @param center The center of rotation. Can be either a single value (0-1), or\n *     an array of two numbers `[centerX, centerY]`. Defaults to `0.5` (rotates\n *     the image around its center).\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction rotateWithOffset_(\n    image: Tensor4D|TensorLike, radians: number,\n    fillValue: number|[number, number, number] = 0,\n    center: number|[number, number] = 0.5): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'rotateWithOffset', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in rotateWithOffset: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: RotateWithOffsetInputs = {image: $image};\n  const attrs: RotateWithOffsetAttrs = {radians, fillValue, center};\n  const res = ENGINE.runKernel(\n      RotateWithOffset, inputs as {} as NamedTensorMap,\n      attrs as {} as NamedAttrMap);\n  return res as Tensor4D;\n}\n\nexport const rotateWithOffset = op({rotateWithOffset_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport * as util from '../util';\n\nfunction nonMaxSuppSanityCheck(\n    boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma?: number): {\n  maxOutputSize: number,\n  iouThreshold: number,\n  scoreThreshold: number,\n  softNmsSigma: number\n} {\n  if (iouThreshold == null) {\n    iouThreshold = 0.5;\n  }\n  if (scoreThreshold == null) {\n    scoreThreshold = Number.NEGATIVE_INFINITY;\n  }\n  if (softNmsSigma == null) {\n    softNmsSigma = 0.0;\n  }\n\n  const numBoxes = boxes.shape[0];\n  maxOutputSize = Math.min(maxOutputSize, numBoxes);\n\n  util.assert(\n      0 <= iouThreshold && iouThreshold <= 1,\n      () => `iouThreshold must be in [0, 1], but was '${iouThreshold}'`);\n  util.assert(\n      boxes.rank === 2,\n      () => `boxes must be a 2D tensor, but was of rank '${boxes.rank}'`);\n  util.assert(\n      boxes.shape[1] === 4,\n      () =>\n          `boxes must have 4 columns, but 2nd dimension was ${boxes.shape[1]}`);\n  util.assert(scores.rank === 1, () => 'scores must be a 1D tensor');\n  util.assert(\n      scores.shape[0] === numBoxes,\n      () => `scores has incompatible shape with boxes. Expected ${numBoxes}, ` +\n          `but was ${scores.shape[0]}`);\n  util.assert(\n      0 <= softNmsSigma && softNmsSigma <= 1,\n      () => `softNmsSigma must be in [0, 1], but was '${softNmsSigma}'`);\n  return {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n}\n\nexport {nonMaxSuppSanityCheck};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV3} from '../../kernel_names';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\nfunction nonMaxSuppression_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Tensor1D {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const attrs = {maxOutputSize, iouThreshold, scoreThreshold};\n  return ENGINE.runKernelFunc(\n      b => b.nonMaxSuppression(\n          $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold),\n      {boxes: $boxes, scores: $scores}, null /* grad */, NonMaxSuppressionV3,\n      attrs);\n}\n\nexport const nonMaxSuppression = op({nonMaxSuppression_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inserts a value into a sorted array. This method allows duplicate, meaning it\n * allows inserting duplicate value, in which case, the element will be inserted\n * at the lowest index of the value.\n * @param arr The array to modify.\n * @param element The element to insert.\n * @param comparator Optional. If no comparator is specified, elements are\n * compared using array_util.defaultComparator, which is suitable for Strings\n * and Numbers in ascending arrays. If the array contains multiple instances of\n * the target value, the left-most instance will be returned. To provide a\n * comparator, it should take 2 arguments to compare and return a negative,\n * zero, or a positive number.\n */\nexport function binaryInsert<T>(\n    arr: T[], element: T, comparator?: (a: T, b: T) => number) {\n  const index = binarySearch(arr, element, comparator);\n  const insertionPoint = index < 0 ? -(index + 1) : index;\n  arr.splice(insertionPoint, 0, element);\n}\n\n/**\n * Searches the array for the target using binary search, returns the index\n * of the found element, or position to insert if element not found. If no\n * comparator is specified, elements are compared using array_\n * util.defaultComparator, which is suitable for Strings and Numbers in\n * ascending arrays. If the array contains multiple instances of the target\n * value, the left-most instance will be returned.\n * @param arr The array to be searched in.\n * @param target The target to be searched for.\n * @param comparator Should take 2 arguments to compare and return a negative,\n *    zero, or a positive number.\n * @return Lowest index of the target value if found, otherwise the insertion\n *    point where the target should be inserted, in the form of\n *    (-insertionPoint - 1).\n */\nexport function binarySearch<T>(\n    arr: T[], target: T, comparator?: (a: T, b: T) => number) {\n  return binarySearch_(arr, target, comparator || defaultComparator);\n}\n\n/**\n * Compares its two arguments for order.\n * @param a The first element to be compared.\n * @param b The second element to be compared.\n * @return A negative number, zero, or a positive number as the first\n *     argument is less than, equal to, or greater than the second.\n */\nfunction defaultComparator<T>(a: T, b: T): number {\n  return a > b ? 1 : a < b ? -1 : 0;\n}\n\nfunction binarySearch_<T>(\n    arr: T[], target: T, comparator: (a: T, b: T) => number) {\n  let left = 0;\n  let right = arr.length;\n  let middle = 0;\n  let found = false;\n  while (left < right) {\n    middle = left + ((right - left) >>> 1);\n    const compareResult = comparator(target, arr[middle]);\n    if (compareResult > 0) {\n      left = middle + 1;\n    } else {\n      right = middle;\n      // If compareResult is 0, the value is found. We record it is found,\n      // and then keep looking because there may be duplicate.\n      found = !compareResult;\n    }\n  }\n\n  return found ? left : -left - 1;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Implementation of the NonMaxSuppression kernel shared between webgl and cpu.\n */\n\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\n\nimport {binaryInsert} from './array_util';\n\ninterface Candidate {\n  score: number;\n  boxIndex: number;\n  suppressBeginIndex: number;\n}\n\nexport function nonMaxSuppressionV3Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number): Tensor1D {\n  return nonMaxSuppressionImpl_(\n             boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n             0 /* softNmsSigma */)\n             .selectedIndices as Tensor1D;\n}\n\nexport function nonMaxSuppressionV4Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    padToMaxOutputSize: boolean): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n      0 /* softNmsSigma */, false /* returnScoresTensor */,\n      padToMaxOutputSize /* padToMaxOutputSize */, true\n      /* returnValidOutputs */);\n}\n\nexport function nonMaxSuppressionV5Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    softNmsSigma: number): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma,\n      true /* returnScoresTensor */);\n}\n\nfunction nonMaxSuppressionImpl_(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma: number,\n    returnScoresTensor = false, padToMaxOutputSize = false,\n    returnValidOutputs = false): NamedTensorMap {\n  // The list is sorted in ascending order, so that we can always pop the\n  // candidate with the largest score in O(1) time.\n  const candidates = [];\n\n  for (let i = 0; i < scores.length; i++) {\n    if (scores[i] > scoreThreshold) {\n      candidates.push({score: scores[i], boxIndex: i, suppressBeginIndex: 0});\n    }\n  }\n\n  candidates.sort(ascendingComparator);\n\n  // If softNmsSigma is 0, the outcome of this algorithm is exactly same as\n  // before.\n  const scale = softNmsSigma > 0 ? (-0.5 / softNmsSigma) : 0.0;\n\n  const selectedIndices: number[] = [];\n  const selectedScores: number[] = [];\n\n  while (selectedIndices.length < maxOutputSize && candidates.length > 0) {\n    const candidate = candidates.pop();\n    const {score: originalScore, boxIndex, suppressBeginIndex} = candidate;\n\n    if (originalScore < scoreThreshold) {\n      break;\n    }\n\n    // Overlapping boxes are likely to have similar scores, therefore we\n    // iterate through the previously selected boxes backwards in order to\n    // see if candidate's score should be suppressed. We use\n    // suppressBeginIndex to track and ensure a candidate can be suppressed\n    // by a selected box no more than once. Also, if the overlap exceeds\n    // iouThreshold, we simply ignore the candidate.\n    let ignoreCandidate = false;\n    for (let j = selectedIndices.length - 1; j >= suppressBeginIndex; --j) {\n      const iou = intersectionOverUnion(boxes, boxIndex, selectedIndices[j]);\n\n      if (iou >= iouThreshold) {\n        ignoreCandidate = true;\n        break;\n      }\n\n      candidate.score =\n          candidate.score * suppressWeight(iouThreshold, scale, iou);\n\n      if (candidate.score <= scoreThreshold) {\n        break;\n      }\n    }\n\n    // At this point, if `candidate.score` has not dropped below\n    // `scoreThreshold`, then we know that we went through all of the\n    // previous selections and can safely update `suppressBeginIndex` to the\n    // end of the selected array. Then we can re-insert the candidate with\n    // the updated score and suppressBeginIndex back in the candidate list.\n    // If on the other hand, `candidate.score` has dropped below the score\n    // threshold, we will not add it back to the candidates list.\n    candidate.suppressBeginIndex = selectedIndices.length;\n\n    if (!ignoreCandidate) {\n      // Candidate has passed all the tests, and is not suppressed, so\n      // select the candidate.\n      if (candidate.score === originalScore) {\n        selectedIndices.push(boxIndex);\n        selectedScores.push(candidate.score);\n      } else if (candidate.score > scoreThreshold) {\n        // Candidate's score is suppressed but is still high enough to be\n        // considered, so add back to the candidates list.\n        binaryInsert(candidates, candidate, ascendingComparator);\n      }\n    }\n  }\n\n  // NonMaxSuppressionV4 feature: padding output to maxOutputSize.\n  const validOutputs = selectedIndices.length;\n  const elemsToPad = maxOutputSize - validOutputs;\n\n  if (padToMaxOutputSize && elemsToPad > 0) {\n    selectedIndices.push(...new Array(elemsToPad).fill(0));\n    selectedScores.push(...new Array(elemsToPad).fill(0.0));\n  }\n\n  const result:\n      NamedTensorMap = {selectedIndices: tensor1d(selectedIndices, 'int32')};\n\n  if (returnScoresTensor) {\n    result['selectedScores'] = tensor1d(selectedScores, 'float32');\n  }\n\n  if (returnValidOutputs) {\n    result['validOutputs'] = scalar(validOutputs, 'int32');\n  }\n\n  return result;\n}\n\nfunction intersectionOverUnion(boxes: TypedArray, i: number, j: number) {\n  const iCoord = boxes.subarray(i * 4, i * 4 + 4);\n  const jCoord = boxes.subarray(j * 4, j * 4 + 4);\n  const yminI = Math.min(iCoord[0], iCoord[2]);\n  const xminI = Math.min(iCoord[1], iCoord[3]);\n  const ymaxI = Math.max(iCoord[0], iCoord[2]);\n  const xmaxI = Math.max(iCoord[1], iCoord[3]);\n  const yminJ = Math.min(jCoord[0], jCoord[2]);\n  const xminJ = Math.min(jCoord[1], jCoord[3]);\n  const ymaxJ = Math.max(jCoord[0], jCoord[2]);\n  const xmaxJ = Math.max(jCoord[1], jCoord[3]);\n  const areaI = (ymaxI - yminI) * (xmaxI - xminI);\n  const areaJ = (ymaxJ - yminJ) * (xmaxJ - xminJ);\n  if (areaI <= 0 || areaJ <= 0) {\n    return 0.0;\n  }\n  const intersectionYmin = Math.max(yminI, yminJ);\n  const intersectionXmin = Math.max(xminI, xminJ);\n  const intersectionYmax = Math.min(ymaxI, ymaxJ);\n  const intersectionXmax = Math.min(xmaxI, xmaxJ);\n  const intersectionArea = Math.max(intersectionYmax - intersectionYmin, 0.0) *\n      Math.max(intersectionXmax - intersectionXmin, 0.0);\n  return intersectionArea / (areaI + areaJ - intersectionArea);\n}\n\n// A Gaussian penalty function, this method always returns values in [0, 1].\n// The weight is a function of similarity, the more overlap two boxes are, the\n// smaller the weight is, meaning highly overlapping boxe will be significantly\n// penalized. On the other hand, a non-overlapping box will not be penalized.\nfunction suppressWeight(iouThreshold: number, scale: number, iou: number) {\n  const weight = Math.exp(scale * iou * iou);\n  return iou <= iouThreshold ? weight : 0.0;\n}\n\nfunction ascendingComparator(c1: Candidate, c2: Candidate) {\n  // For objects with same scores, we make the object with the larger index go\n  // first. In an array that pops from the end, this means that the object with\n  // the smaller index will be popped first. This ensures the same output as\n  // the TensorFlow python version.\n  return (c1.score - c2.score) ||\n      ((c1.score === c2.score) && (c2.boxIndex - c1.boxIndex));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {nonMaxSuppressionV3Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This is the async version of `nonMaxSuppression`\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @return A 1D tensor with the selected box indices.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Promise<Tensor1D> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV3Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionAsync = nonMaxSuppressionAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionWithScore_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const inputs: NonMaxSuppressionV5Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV5Attrs =\n      {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV5, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], selectedScores: result[1]};\n}\n\nexport const nonMaxSuppressionWithScore = op({nonMaxSuppressionWithScore_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV5Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionWithScoreAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV5Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionWithScoreAsync = nonMaxSuppressionWithScoreAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionPadded_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const inputs: NonMaxSuppressionV4Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV4Attrs = {\n    maxOutputSize: $maxOutputSize,\n    iouThreshold: $iouThreshold,\n    scoreThreshold: $scoreThreshold,\n    padToMaxOutputSize\n  };\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV4, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], validOutputs: result[1]};\n}\n\nexport const nonMaxSuppressionPadded = op({nonMaxSuppressionPadded_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV4Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionPaddedAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const [boxesVals, scoresVals] =\n      await Promise.all([$boxes.data(), $scores.data()]);\n\n  // We call a cpu based impl directly with the typedarray data here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV4Impl(\n      boxesVals, scoresVals, $maxOutputSize, $iouThreshold, $scoreThreshold,\n      padToMaxOutputSize);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionPaddedAsync = nonMaxSuppressionPaddedAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Bilinear resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeBilinear_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeBilinear');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeBilinear: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () => `Error in resizeBilinear: new shape must 2D, but got shape ` +\n          `${size}.`);\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n\n  const [newHeight, newWidth] = size;\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeBilinear(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const inputs: ResizeBilinearInputs = {images: batchImages};\n  const attrs: ResizeBilinearAttrs = {alignCorners, size};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeBilinear, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeBilinear = op({resizeBilinear_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * NearestNeighbor resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeNearestNeighbor_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeNearestNeighbor');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeNearestNeighbor: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () =>\n          `Error in resizeNearestNeighbor: new shape must 2D, but got shape ` +\n          `${size}.`);\n  util.assert(\n      $images.dtype === 'float32' || $images.dtype === 'int32',\n      () => '`images` must have `int32` or `float32` as dtype');\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n  const [newHeight, newWidth] = size;\n\n  const inputs: ResizeNearestNeighborInputs = {images: batchImages};\n  const attrs: ResizeNearestNeighborAttrs = {alignCorners, size};\n\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeNearestNeighbor(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeNearestNeighbor, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeNearestNeighbor = op({resizeNearestNeighbor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assert} from '../../util';\n\nimport {greaterEqual} from '../greater_equal';\nimport {lessEqual} from '../less_equal';\nimport {logicalAnd} from '../logical_and';\nimport {op} from '../operation';\nimport {range} from '../range';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\nimport {zeros} from '../zeros';\n\n/**\n * Copy a tensor setting everything outside a central band in each innermost\n * matrix to zero.\n *\n * The band part is computed as follows: Assume input has `k` dimensions\n * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where\n * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`.\n * The indicator function\n * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))`\n * `&& (num_upper < 0 || (n-m) <= num_upper)`\n *\n * ```js\n * const x = tf.tensor2d([[ 0,  1,  2, 3],\n *                        [-1,  0,  1, 2],\n *                        [-2, -1,  0, 1],\n *                        [-3, -2, -1, 0]]);\n * let y = tf.linalg.bandPart(x, 1, -1);\n * y.print(); // [[ 0,  1,  2, 3],\n *            //  [-1,  0,  1, 2],\n *            //  [ 0, -1,  0, 1],\n *            //  [ 0, 0 , -1, 0]]\n * let z = tf.linalg.bandPart(x, 2, 1);\n * z.print(); // [[ 0,  1,  0, 0],\n *            //  [-1,  0,  1, 0],\n *            //  [-2, -1,  0, 1],\n *            //  [ 0, -2, -1, 0]]\n * ```\n *\n * @param x Rank `k` tensor\n * @param numLower Number of subdiagonals to keep.\n *   If negative, keep entire lower triangle.\n * @param numUpper Number of subdiagonals to keep.\n *   If negative, keep entire upper triangle.\n * @returns Rank `k` tensor of the same shape as input.\n *   The extracted banded tensor.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction bandPart_<T extends Tensor>(\n    a: T|TensorLike, numLower: number, numUpper: number): T {\n  assert(\n      numLower % 1 === 0,\n      () => `bandPart(): numLower must be an integer, got ${numLower}.`);\n  assert(\n      numUpper % 1 === 0,\n      () => `bandPart(): numUpper must be an integer, got ${numUpper}.`);\n\n  const $a = convertToTensor(a, 'a', 'bandPart');\n\n  assert(\n      $a.rank >= 2,\n      () => `bandPart(): Rank must be at least 2, got ${$a.rank}.`);\n\n  const shape = $a.shape;\n  const [M, N] = $a.shape.slice(-2);\n\n  if (!(numLower <= M)) {\n    throw new Error(\n        `bandPart(): numLower (${numLower})` +\n        ` must not be greater than the number of rows (${M}).`);\n  }\n  if (!(numUpper <= N)) {\n    throw new Error(\n        `bandPart(): numUpper (${numUpper})` +\n        ` must not be greater than the number of columns (${N}).`);\n  }\n\n  if (numLower < 0) {\n    numLower = M;\n  }\n  if (numUpper < 0) {\n    numUpper = N;\n  }\n\n  const i = reshape(range(0, M, 1, 'int32'), [-1, 1]);\n  const j = range(0, N, 1, 'int32');\n  const ij = sub(i, j);\n\n  const inBand = logicalAnd(\n      lessEqual(ij, scalar(+numLower, 'int32')),\n      greaterEqual(ij, scalar(-numUpper, 'int32')));\n\n  const zero = zeros([M, N], $a.dtype);\n\n  return reshape(\n             stack(unstack(reshape($a, [-1, M, N]))\n                       .map(mat => where(inBand, mat, zero))),\n             shape) as T;\n}\n\nexport const bandPart = op({bandPart_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {div} from '../div';\nimport {mul} from '../mul';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {split} from '../split';\nimport {squeeze} from '../squeeze';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\n/**\n * Gram-Schmidt orthogonalization.\n *\n * ```js\n * const x = tf.tensor2d([[1, 2], [3, 4]]);\n * let y = tf.linalg.gramSchmidt(x);\n * y.print();\n * console.log('Othogonalized:');\n * y.dot(y.transpose()).print();  // should be nearly the identity matrix.\n * console.log('First row direction maintained:');\n * const data = await y.array();\n * console.log(data[0][1] / data[0][0]);  // should be nearly 2.\n * ```\n *\n * @param xs The vectors to be orthogonalized, in one of the two following\n *   formats:\n *   - An Array of `tf.Tensor1D`.\n *   - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows\n *     of `xs`.\n *   In each case, all the vectors must have the same length and the length\n *   must be greater than or equal to the number of vectors.\n * @returns The orthogonalized and normalized vectors or matrix.\n *   Orthogonalization means that the vectors or the rows of the matrix\n *   are orthogonal (zero inner products). Normalization means that each\n *   vector or each row of the matrix has an L2 norm that equals `1`.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D {\n  let inputIsTensor2D: boolean;\n  if (Array.isArray(xs)) {\n    inputIsTensor2D = false;\n    assert(\n        xs != null && xs.length > 0,\n        () => 'Gram-Schmidt process: input must not be null, undefined, or ' +\n            'empty');\n    const dim = xs[0].shape[0];\n    for (let i = 1; i < xs.length; ++i) {\n      assert(\n          xs[i].shape[0] === dim,\n          () =>\n              'Gram-Schmidt: Non-unique lengths found in the input vectors: ' +\n              `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`);\n    }\n  } else {\n    inputIsTensor2D = true;\n    xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0]));\n  }\n\n  assert(\n      xs.length <= xs[0].shape[0],\n      () => `Gram-Schmidt: Number of vectors (${\n                (xs as Tensor1D[]).length}) exceeds ` +\n          `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`);\n\n  const ys: Tensor1D[] = [];\n  const xs1d = xs;\n  for (let i = 0; i < xs.length; ++i) {\n    ys.push(ENGINE.tidy(() => {\n      let x = xs1d[i];\n      if (i > 0) {\n        for (let j = 0; j < i; ++j) {\n          const proj = mul(sum(mul(ys[j], x)), ys[j]);\n          x = sub(x, proj);\n        }\n      }\n      return div(x, norm(x, 'euclidean'));\n    }));\n  }\n\n  if (inputIsTensor2D) {\n    return stack(ys, 0) as Tensor2D;\n  } else {\n    return ys;\n  }\n}\n\nexport const gramSchmidt = op({gramSchmidt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../../engine';\nimport {dispose} from '../../globals';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {clone} from '../clone';\nimport {concat} from '../concat';\nimport {div} from '../div';\nimport {eye} from '../eye';\nimport {greater} from '../greater';\nimport {matMul} from '../mat_mul';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {tensor2d} from '../tensor2d';\nimport {transpose} from '../transpose';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\n\n/**\n * Compute QR decomposition of m-by-n matrix using Householder transformation.\n *\n * Implementation based on\n *   [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf]\n * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf)\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [3, 4]]);\n * let [q, r] = tf.linalg.qr(a);\n * console.log('Q');\n * q.print();\n * console.log('R');\n * r.print();\n * console.log('Orthogonalized');\n * q.dot(q.transpose()).print()  // should be nearly the identity matrix.\n * console.log('Reconstructed');\n * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]];\n * ```\n *\n * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose\n *   it has the shape `[..., M, N]`.\n * @param fullMatrices An optional boolean parameter. Defaults to `false`.\n *   If `true`, compute full-sized `Q`. If `false` (the default),\n *   compute only the leading N columns of `Q` and `R`.\n * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix,\n *   i.e., its columns all have unit norm and are mutually orthogonal.\n *   If `M >= N`,\n *     If `fullMatrices` is `false` (default),\n *       - `Q` has a shape of `[..., M, N]`,\n *       - `R` has a shape of `[..., N, N]`.\n *     If `fullMatrices` is `true` (default),\n *       - `Q` has a shape of `[..., M, M]`,\n *       - `R` has a shape of `[..., M, N]`.\n *   If `M < N`,\n *     - `Q` has a shape of `[..., M, M]`,\n *     - `R` has a shape of `[..., M, N]`.\n * @throws If the rank of `x` is less than 2.\n *\n * @doc {heading:'Operations',\n *       subheading:'Linear Algebra',\n *       namespace:'linalg'}\n */\nfunction qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] {\n  assert(\n      x.rank >= 2,\n      () => `qr() requires input tensor to have a rank >= 2, but got rank ${\n          x.rank}`);\n\n  if (x.rank === 2) {\n    return qr2d(x as Tensor2D, fullMatrices);\n  } else {\n    // Rank > 2.\n    // TODO(cais): Below we split the input into individual 2D tensors,\n    //   perform QR decomposition on them and then stack the results back\n    //   together. We should explore whether this can be parallelized.\n    const outerDimsProd = x.shape.slice(0, x.shape.length - 2)\n                              .reduce((value, prev) => value * prev);\n    const x2ds = unstack(\n        reshape(\n            x,\n            [\n              outerDimsProd, x.shape[x.shape.length - 2],\n              x.shape[x.shape.length - 1]\n            ]),\n        0);\n    const q2ds: Tensor2D[] = [];\n    const r2ds: Tensor2D[] = [];\n    x2ds.forEach(x2d => {\n      const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices);\n      q2ds.push(q2d);\n      r2ds.push(r2d);\n    });\n    const q = reshape(stack(q2ds, 0), x.shape);\n    const r = reshape(stack(r2ds, 0), x.shape);\n    return [q, r];\n  }\n}\n\nfunction qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] {\n  return ENGINE.tidy(() => {\n    assert(\n        x.shape.length === 2,\n        () => `qr2d() requires a 2D Tensor, but got a ${\n            x.shape.length}D Tensor.`);\n\n    const m = x.shape[0];\n    const n = x.shape[1];\n\n    let q = eye(m);    // Orthogonal transform so far.\n    let r = clone(x);  // Transformed matrix so far.\n\n    const one2D = tensor2d([[1]], [1, 1]);\n    let w: Tensor2D = clone(one2D);\n\n    const iters = m >= n ? n : m;\n    for (let j = 0; j < iters; ++j) {\n      // This tidy within the for-loop ensures we clean up temporary\n      // tensors as soon as they are no longer needed.\n      const rTemp = r;\n      const wTemp = w;\n      const qTemp = q;\n      [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => {\n        // Find H = I - tau * w * w', to put zeros below R(j, j).\n        const rjEnd1 = slice(r, [j, j], [m - j, 1]);\n        const normX = norm(rjEnd1);\n        const rjj = slice(r, [j, j], [1, 1]);\n\n        // The sign() function returns 0 on 0, which causes division by zero.\n        const s = where(greater(rjj, 0), tensor2d([[-1]]), tensor2d([[1]]));\n\n        const u1 = sub(rjj, mul(s, normX));\n        const wPre = div(rjEnd1, u1);\n        if (wPre.shape[0] === 1) {\n          w = clone(one2D);\n        } else {\n          w = concat(\n              [\n                one2D,\n                slice(wPre, [1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as\n                    Tensor2D\n              ],\n              0);\n        }\n        const tau = neg(div(matMul(s, u1), normX)) as Tensor2D;\n\n        // -- R := HR, Q := QH.\n        const rjEndAll = slice(r, [j, 0], [m - j, n]);\n        const tauTimesW: Tensor2D = mul(tau, w);\n        const wT: Tensor2D = transpose(w);\n        if (j === 0) {\n          r = sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n        } else {\n          const rTimesTau: Tensor2D =\n              sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n          r = concat([slice(r, [0, 0], [j, n]), rTimesTau], 0);\n        }\n        const tawTimesWT: Tensor2D = transpose(tauTimesW);\n        const qAllJEnd = slice(q, [0, j], [m, q.shape[1] - j]);\n        if (j === 0) {\n          q = sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n        } else {\n          const qTimesTau: Tensor2D =\n              sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n          q = concat([slice(q, [0, 0], [m, j]), qTimesTau], 1);\n        }\n        return [w, r, q];\n      });\n      dispose([rTemp, wTemp, qTemp]);\n    }\n\n    if (!fullMatrices && m > n) {\n      q = slice(q, [0, 0], [m, n]);\n      r = slice(r, [0, 0], [n, n]);\n    }\n\n    return [q, r];\n  }) as [Tensor2D, Tensor2D];\n}\n\nexport const qr = op({qr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport enum Reduction {\n  NONE,\n  MEAN,\n  SUM,\n  SUM_BY_NONZERO_WEIGHTS\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {Reduction} from '../loss_ops_utils';\nimport {mean} from '../mean';\nimport {mul} from '../mul';\nimport {notEqual} from '../not_equal';\nimport {ones} from '../ones';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sum} from '../sum';\n\n/**\n * Computes the weighted loss between two tensors.\n *\n * @param losses Tensor of shape `[batch_size, d1, ... dN]`.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `losses`, and must be broadcastable to `losses` (i.e., all\n *    dimensions must be either `1`, or the same as the corresponding\n *    `losses` dimension).\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction computeWeightedLoss_<T extends Tensor, O extends Tensor>(\n    losses: T|TensorLike, weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $losses = convertToTensor(losses, 'losses', 'computeWeightedLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'computeWeightedLoss');\n  }\n\n  const weightedLoss = ($weights == null) ? $losses : mul($losses, $weights);\n\n  if (reduction === Reduction.NONE) {\n    return weightedLoss as O;\n  }\n  if (reduction === Reduction.SUM) {\n    return sum(weightedLoss);\n  }\n  if (reduction === Reduction.MEAN) {\n    if ($weights == null) {\n      return mean(weightedLoss);\n    } else {\n      const broadcastFactor = $losses.size / $weights.size;\n      const result = div(sum(weightedLoss), sum($weights));\n      return broadcastFactor > 1 ? div(result, scalar(broadcastFactor)) :\n                                   result as O;\n    }\n  }\n  if (reduction === Reduction.SUM_BY_NONZERO_WEIGHTS) {\n    if ($weights == null) {\n      return div(sum(weightedLoss), scalar($losses.size));\n    } else {\n      const broadcastedWeights = mul($weights, ones($losses.shape));\n\n      const numNonZeros =\n          cast(sum(notEqual(broadcastedWeights, scalar(0))), 'float32');\n      return div(sum(weightedLoss), numNonZeros);\n    }\n  }\n\n  throw Error(`Unknown reduction: ${reduction}`);\n}\nexport const computeWeightedLoss = op({computeWeightedLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the absolute difference loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction absoluteDifference_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'absoluteDifference');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'absoluteDifference');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'absoluteDifference');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in absoluteDifference: ');\n\n  const losses = abs(sub($labels, $predictions));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const absoluteDifference = op({absoluteDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the cosine distance loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param axis The dimension along which the cosine distance is computed.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction cosineDistance_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike, axis: number,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'cosineDistance');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'cosineDistance');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'cosineDistance');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in cosineDistance: ');\n\n  const one = scalar(1);\n  const losses = sub(one, sum(mul($labels, $predictions), axis, true));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const cosineDistance = op({cosineDistance_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the Hinge loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction hingeLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $labels = convertToTensor(labels, 'labels', 'hingeLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'hingeLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'hingeLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in hingeLoss: ');\n\n  const one = scalar(1);\n  // Convert binary labels to (-1, 1)\n  $labels = sub(mul(scalar(2), $labels), one);\n  const losses = relu(sub(one, mul($labels, $predictions)));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const hingeLoss = op({hingeLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {Reduction} from '../loss_ops_utils';\nimport {minimum} from '../minimum';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {square} from '../square';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the huber loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param delta Point where huber loss changes from quadratic to linear.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`.\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction huberLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, delta = 1.0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'huberLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'huberLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'huberLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in huberLoss: ');\n\n  const deltaScalar = scalar(delta);\n  const error = abs(sub($predictions, $labels));\n  const quadratic = minimum(error, deltaScalar);\n  const linear = sub(error, quadratic);\n\n  const losses =\n      add(mul(scalar(0.5), square(quadratic)), mul(deltaScalar, linear));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const huberLoss = op({huberLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {log} from '../log';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the log loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param epsilon A small increment to avoid taking log of zero\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction logLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, epsilon = 1e-7,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'logLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'logLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'logLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in logLoss: ');\n\n  const one = scalar(1);\n  const epsilonScalar = scalar(epsilon);\n\n  const l1 = neg(mul($labels, log(add($predictions, epsilonScalar))));\n  const l2 =\n      mul(sub(one, $labels), log(add(sub(one, $predictions), epsilonScalar)));\n  const losses = sub(l1, l2);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const logLoss = op({logLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {squaredDifference} from '../squared_difference';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the mean squared error between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction meanSquaredError_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'meanSquaredError');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'meanSquaredError');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'meanSquaredError');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in meanSquaredError: ');\n\n  const losses = squaredDifference($labels, $predictions);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const meanSquaredError = op({meanSquaredError_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {exp} from '../exp';\nimport {log1p} from '../log1p';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\nfunction sigmoidCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, logits: T|TensorLike): O {\n  const $labels =\n      convertToTensor(labels, 'labels', 'sigmoidCrossEntropyWithLogits');\n  const $logits =\n      convertToTensor(logits, 'logits', 'sigmoidCrossEntropyWithLogits');\n  assertShapesMatch(\n      $labels.shape, $logits.shape, 'Error in sigmoidCrossEntropyWithLogits: ');\n\n  /**\n   * Implementation Details:\n   *\n   * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n   *     z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n   *   = (1 - z) * x + log(1 + exp(-x))\n   *   = x - x * z + log(1 + exp(-x))\n   *\n   *   For x < 0, to avoid overflow in exp(-x), we reformulate the above\n   *     x - x * z + log(1 + exp(-x))\n   *   = log(exp(x)) - x * z + log(1 + exp(-x))\n   *   = - x * z + log(1 + exp(x))\n   *\n   * Hence, to ensure stability and avoid overflow, the implementation uses\n   * this equivalent formulation:\n   *     max(x, 0) - x * z + log(1 + exp(-abs(x)))\n   */\n  const maxOutput = relu($logits);\n  const outputXTarget = mul($logits, $labels);\n  const sigmoidOutput = log1p(exp(neg(abs($logits))));\n\n  return add(sub(maxOutput, outputXTarget), sigmoidOutput);\n}\n\n/**\n * Computes the sigmoid cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newMulticlassLabels = multiclassLabels * (1 - labelSmoothing)\n *                         + 0.5 * labelSmoothing\n *\n * @param multiClassLabels The ground truth output tensor of shape\n * [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction sigmoidCrossEntropy_<T extends Tensor, O extends Tensor>(\n    multiClassLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $multiClassLabels = convertToTensor(\n      multiClassLabels, 'multiClassLabels', 'sigmoidCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'sigmoidCrossEntropy');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'sigmoidCrossEntropy');\n  }\n  assertShapesMatch(\n      $multiClassLabels.shape, $logits.shape, 'Error in sigmoidCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const half = scalar(0.5);\n\n    $multiClassLabels =\n        add(mul($multiClassLabels, sub(one, labelSmoothingScalar)),\n            mul(half, labelSmoothingScalar));\n  }\n  const losses = sigmoidCrossEntropyWithLogits_($multiClassLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const sigmoidCrossEntropy = op({sigmoidCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../../gradients';\nimport {Tensor} from '../../tensor';\nimport {GradSaveFunc} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {expandShapeToKeepDim} from '../axis_util';\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {exp} from '../exp';\nimport {logSumExp} from '../log_sum_exp';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes softmax cross entropy between logits and labels.\n *\n * Measures the probability error in discrete classification tasks in which\n * the classes are mutually exclusive (each entry is in exactly one class).\n * For example, each CIFAR-10 image is labeled with one and only one label: an\n * image can be a dog or a truck, but not both.\n *\n * `NOTE`: While the classes are mutually exclusive, their probabilities need\n * not be. All that is required is that each row of labels is a valid\n * probability distribution. If they are not, the computation of the gradient\n * will be incorrect.\n *\n * `WARNING`: This op expects unscaled logits, since it performs a softmax on\n * logits internally for efficiency. Do not call this op with the output of\n * softmax, as it will produce incorrect results.\n *\n * logits and labels must have the same shape, e.g. [batch_size, num_classes]\n * and the same dtype.\n * @param labels The labels array.\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n */\nfunction softmaxCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T, logits: T, dim = -1): O {\n  if (dim === -1) {\n    dim = logits.rank - 1;\n  }\n\n  if (dim !== logits.rank - 1) {\n    throw Error(\n        `Softmax cross entropy along a non-last dimension is not yet ` +\n        `supported. Labels / logits was rank ${logits.rank} ` +\n        `and dim was ${dim}`);\n  }\n  // Use a custom gradient for numerical stability.\n  const customOp =\n      customGrad((labels: Tensor, logits: Tensor, save: GradSaveFunc) => {\n        // Reference:\n        //   1. http://cs231n.github.io/linear-classify/#softmax\n        //   2. https://blog.feedly.com/tricks-of-the-trade-logsumexp/\n        const keepDims = true;\n        const lse = logSumExp(logits, [dim], keepDims);\n        const logResult = sub(cast(logits, 'float32'), lse);\n        save([labels, logResult]);\n\n        const costVector = neg(mul(logResult, labels));\n        const value: O = sum(costVector, [dim]);\n\n        const gradFunc = (dy: O, saved: Tensor[]) => {\n          const [labels, logResult] = saved;\n          const dyShape = expandShapeToKeepDim(dy.shape, [dim]);\n          return [\n            mul(reshape(dy, dyShape),\n                sub(cast(labels, 'float32'), exp(logResult))),\n            mul(reshape(dy, dyShape),\n                sub(exp(logResult), cast(labels, 'float32'))),\n          ];\n        };\n        return {value, gradFunc};\n      });\n\n  return customOp(labels, logits);\n}\n\n/**\n * Computes the softmax cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newOnehotLabels = onehotLabels * (1 - labelSmoothing)\n *                         + labelSmoothing / numClasses\n *\n * @param onehotLabels One hot encoded labels\n *    [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or 1, and must be\n *    broadcastable to `loss`  of shape [batch_size]\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction softmaxCrossEntropy_<T extends Tensor, O extends Tensor>(\n    onehotLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $onehotLabels =\n      convertToTensor(onehotLabels, 'onehotLabels', 'softmaxCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'softmaxCrossEntropy');\n  let $weights: Tensor = null;\n\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'softmaxCrossEntropy');\n  }\n\n  assertShapesMatch(\n      $onehotLabels.shape, $logits.shape, 'Error in softmaxCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const numClasses = scalar($onehotLabels.shape[1]);\n\n    $onehotLabels =\n        add(mul($onehotLabels, sub(one, labelSmoothingScalar)),\n            div(labelSmoothingScalar, numClasses));\n  }\n\n  const losses = softmaxCrossEntropyWithLogits_($onehotLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const softmaxCrossEntropy = op({softmaxCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Modularized ops.\nexport {abs} from './abs';\nexport {acos} from './acos';\nexport {acosh} from './acosh';\nexport {add} from './add';\nexport {addN} from './add_n';\nexport {all} from './all';\nexport {any} from './any';\nexport {argMax} from './arg_max';\nexport {argMin} from './arg_min';\nexport {asin} from './asin';\nexport {asinh} from './asinh';\nexport {atan} from './atan';\nexport {atan2} from './atan2';\nexport {atanh} from './atanh';\nexport {avgPool} from './avg_pool';\nexport {avgPool3d} from './avg_pool_3d';\nexport {basicLSTMCell} from './basic_lstm_cell';\nexport {batchToSpaceND} from './batch_to_space_nd';\nexport {batchNorm} from './batchnorm';\nexport {batchNorm2d} from './batchnorm2d';\nexport {batchNorm3d} from './batchnorm3d';\nexport {batchNorm4d} from './batchnorm4d';\nexport {broadcastTo} from './broadcast_to';\nexport {buffer} from './buffer';\nexport {cast} from './cast';\nexport {ceil} from './ceil';\nexport {clipByValue} from './clip_by_value';\nexport {clone} from './clone';\nexport {complex} from './complex';\nexport {concat} from './concat';\nexport {concat1d} from './concat_1d';\nexport {concat2d} from './concat_2d';\nexport {concat3d} from './concat_3d';\nexport {concat4d} from './concat_4d';\nexport {conv1d} from './conv1d';\nexport {conv2d} from './conv2d';\nexport {conv2dTranspose} from './conv2d_transpose';\nexport {conv3d} from './conv3d';\nexport {conv3dTranspose} from './conv3d_transpose';\nexport {cos} from './cos';\nexport {cosh} from './cosh';\nexport {cumsum} from './cumsum';\nexport {depthToSpace} from './depth_to_space';\nexport {depthwiseConv2d} from './depthwise_conv2d';\nexport {diag} from './diag';\nexport {dilation2d} from './dilation2d';\nexport {div} from './div';\nexport {divNoNan} from './div_no_nan';\nexport {dot} from './dot';\nexport {elu} from './elu';\nexport {equal} from './equal';\nexport {erf} from './erf';\nexport {exp} from './exp';\nexport {expandDims} from './expand_dims';\nexport {expm1} from './expm1';\nexport {eye} from './eye';\nexport {fill} from './fill';\nexport {floor} from './floor';\nexport {floorDiv} from './floorDiv';\nexport {gather} from './gather';\nexport {greater} from './greater';\nexport {greaterEqual} from './greater_equal';\nexport {imag} from './imag';\nexport {isFinite} from './is_finite';\nexport {isInf} from './is_inf';\nexport {isNaN} from './is_nan';\nexport {leakyRelu} from './leaky_relu';\nexport {less} from './less';\nexport {lessEqual} from './less_equal';\nexport {linspace} from './linspace';\nexport {localResponseNormalization} from './local_response_normalization';\nexport {log} from './log';\nexport {log1p} from './log1p';\nexport {logSigmoid} from './log_sigmoid';\nexport {logSoftmax} from './log_softmax';\nexport {logSumExp} from './log_sum_exp';\nexport {logicalAnd} from './logical_and';\nexport {logicalNot} from './logical_not';\nexport {logicalOr} from './logical_or';\nexport {logicalXor} from './logical_xor';\nexport {matMul} from './mat_mul';\nexport {max} from './max';\nexport {maxPool} from './max_pool';\nexport {maxPool3d} from './max_pool_3d';\nexport {maxPoolWithArgmax} from './max_pool_with_argmax';\nexport {maximum} from './maximum';\nexport {mean} from './mean';\nexport {min} from './min';\nexport {minimum} from './minimum';\nexport {mod} from './mod';\nexport {moments} from './moments';\nexport {mul} from './mul';\nexport {LSTMCellFunc, multiRNNCell} from './multi_rnn_cell';\nexport {multinomial} from './multinomial';\nexport {neg} from './neg';\nexport {notEqual} from './not_equal';\nexport {oneHot} from './one_hot';\nexport {ones} from './ones';\nexport {onesLike} from './ones_like';\nexport {outerProduct} from './outer_product';\nexport {pad} from './pad';\nexport {pad1d} from './pad1d';\nexport {pad2d} from './pad2d';\nexport {pad3d} from './pad3d';\nexport {pad4d} from './pad4d';\nexport {pool} from './pool';\nexport {pow} from './pow';\nexport {prelu} from './prelu';\nexport {print} from './print';\nexport {prod} from './prod';\nexport {rand} from './rand';\nexport {randomGamma} from './random_gamma';\nexport {randomNormal} from './random_normal';\nexport {randomUniform} from './random_uniform';\nexport {range} from './range';\nexport {real} from './real';\nexport {reciprocal} from './reciprocal';\nexport {relu} from './relu';\nexport {relu6} from './relu6';\nexport {reshape} from './reshape';\nexport {reverse} from './reverse';\nexport {reverse1d} from './reverse_1d';\nexport {reverse2d} from './reverse_2d';\nexport {reverse3d} from './reverse_3d';\nexport {reverse4d} from './reverse_4d';\nexport {round} from './round';\nexport {rsqrt} from './rsqrt';\nexport {scalar} from './scalar';\nexport {selu} from './selu';\nexport {separableConv2d} from './separable_conv2d';\nexport {setdiff1dAsync} from './setdiff1d_async';\nexport {sigmoid} from './sigmoid';\nexport {sign} from './sign';\nexport {sin} from './sin';\nexport {sinh} from './sinh';\nexport {slice} from './slice';\nexport {slice1d} from './slice1d';\nexport {slice2d} from './slice2d';\nexport {slice3d} from './slice3d';\nexport {slice4d} from './slice4d';\nexport {softmax} from './softmax';\nexport {softplus} from './softplus';\nexport {spaceToBatchND} from './space_to_batch_nd';\nexport {fft} from './spectral/fft';\nexport {ifft} from './spectral/ifft';\nexport {irfft} from './spectral/irfft';\nexport {rfft} from './spectral/rfft';\nexport {split} from './split';\nexport {sqrt} from './sqrt';\nexport {square} from './square';\nexport {squaredDifference} from './squared_difference';\nexport {squeeze} from './squeeze';\nexport {stack} from './stack';\nexport {step} from './step';\nexport {stridedSlice} from './strided_slice';\nexport {sub} from './sub';\nexport {sum} from './sum';\nexport {tan} from './tan';\nexport {tanh} from './tanh';\nexport {tensor} from './tensor';\nexport {tensor1d} from './tensor1d';\nexport {tensor2d} from './tensor2d';\nexport {tensor3d} from './tensor3d';\nexport {tensor4d} from './tensor4d';\nexport {tensor5d} from './tensor5d';\nexport {tensor6d} from './tensor6d';\nexport {tile} from './tile';\nexport {topk} from './topk';\nexport {truncatedNormal} from './truncated_normal';\nexport {unique} from './unique';\nexport {unsortedSegmentSum} from './unsorted_segment_sum';\nexport {unstack} from './unstack';\nexport {variable} from './variable';\nexport {where} from './where';\nexport {whereAsync} from './where_async';\nexport {zeros} from './zeros';\nexport {zerosLike} from './zeros_like';\n\nexport * from './boolean_mask';\nexport * from './compare';\nexport * from './binary_ops';\nexport * from './transpose';\nexport * from './norm';\nexport * from './moving_average';\nexport * from './scatter_nd';\nexport * from './sparse_to_dense';\nexport * from './gather_nd';\nexport * from './dropout';\nexport * from './signal_ops_util';\nexport * from './in_top_k';\n\nexport {op, OP_SCOPE_SUFFIX} from './operation';\n\nimport {rfft} from './spectral/rfft';\nimport {fft} from './spectral/fft';\nimport {ifft} from './spectral/ifft';\nimport {irfft} from './spectral/irfft';\nconst spectral = {\n  fft,\n  ifft,\n  rfft,\n  irfft\n};\n\nimport * as fused from './fused_ops';\n\nimport {hammingWindow} from './signal/hamming_window';\nimport {hannWindow} from './signal/hann_window';\nimport {frame} from './signal/frame';\nimport {stft} from './signal/stft';\nconst signal = {\n  hammingWindow,\n  hannWindow,\n  frame,\n  stft,\n};\n\n// Image Ops namespace\nimport {cropAndResize} from './image/crop_and_resize';\nimport {flipLeftRight} from './image/flip_left_right';\nimport {rotateWithOffset} from './image/rotate_with_offset';\nimport {nonMaxSuppression} from './image/non_max_suppression';\nimport {nonMaxSuppressionAsync} from './image/non_max_suppression_async';\nimport {nonMaxSuppressionWithScore} from './image/non_max_suppression_with_score';\nimport {nonMaxSuppressionWithScoreAsync} from './image/non_max_suppression_with_score_async';\nimport {nonMaxSuppressionPadded} from './image/non_max_suppression_padded';\nimport {nonMaxSuppressionPaddedAsync} from './image/non_max_suppression_padded_async';\nimport {resizeBilinear} from './image/resize_bilinear';\nimport {resizeNearestNeighbor} from './image/resize_nearest_neighbor';\nconst image = {\n  flipLeftRight,\n  resizeNearestNeighbor,\n  resizeBilinear,\n  rotateWithOffset,\n  cropAndResize,\n  nonMaxSuppression,\n  nonMaxSuppressionAsync,\n  nonMaxSuppressionWithScore,\n  nonMaxSuppressionWithScoreAsync,\n  nonMaxSuppressionPadded,\n  nonMaxSuppressionPaddedAsync\n};\n\n// linalg namespace\nimport {bandPart} from './linalg/band_part';\nimport {gramSchmidt} from './linalg/gram_schmidt';\nimport {qr} from './linalg/qr';\nconst linalg = {\n  bandPart,\n  gramSchmidt,\n  qr\n};\n\n// losses namespace;\nimport {absoluteDifference} from './losses/absolute_difference';\nimport {computeWeightedLoss} from './losses/compute_weighted_loss';\nimport {cosineDistance} from './losses/cosine_distance';\nimport {hingeLoss} from './losses/hinge_loss';\nimport {huberLoss} from './losses/huber_loss';\nimport {logLoss} from './losses/log_loss';\nimport {meanSquaredError} from './losses/mean_squared_error';\nimport {sigmoidCrossEntropy} from './losses/sigmoid_cross_entropy';\nimport {softmaxCrossEntropy} from './losses/softmax_cross_entropy';\nconst losses = {\n  absoluteDifference,\n  computeWeightedLoss,\n  cosineDistance,\n  hingeLoss,\n  huberLoss,\n  logLoss,\n  meanSquaredError,\n  sigmoidCrossEntropy,\n  softmaxCrossEntropy\n};\n\n// Second level exports.\nexport {image, linalg, losses, spectral, fused, signal};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {dispose} from '../globals';\nimport {variableGrads} from '../gradients';\nimport {scalar} from '../ops/ops';\nimport {Serializable} from '../serialization';\nimport {Scalar, Variable} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\n/**\n * A variable that belongs to an optimizer.\n *\n * The `originalName` field is required for keeping track of the canonical\n * name of the variable, which is usually the name of the model weight that\n * the variable is related to plus a suffix, e.g., 'dense1/kernel/momentum'.\n * The name of the `Variable` object itself cannot be used directly due to\n * possible deduplication: Every `Variable` must have a unique name but more\n * than one optimizer objects of the same type may be created for the same model\n * or the same `Variable`.\n */\nexport interface OptimizerVariable {\n  originalName: string;\n  variable: Variable;\n}\n\n/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */\nexport abstract class Optimizer extends Serializable {\n  protected iterations_: number;\n\n  /**\n   * Executes `f()` and minimizes the scalar output of `f()` by computing\n   * gradients of y with respect to the list of trainable variables provided by\n   * `varList`. If no list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to minimize.\n   * @param returnCost Whether to return the scalar cost value produced by\n   * executing `f()`.\n   * @param varList An optional list of variables to update. If specified, only\n   * the trainable variables in varList will be updated by minimize. Defaults to\n   * all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  minimize(f: () => Scalar, returnCost = false, varList?: Variable[]): Scalar\n      |null {\n    const {value, grads} = this.computeGradients(f, varList);\n\n    if (varList != null) {\n      const gradArray: NamedTensor[] =\n          varList.map(v => ({name: v.name, tensor: grads[v.name]}));\n      this.applyGradients(gradArray);\n    } else {\n      this.applyGradients(grads);\n    }\n\n    // Dispose gradients.\n    dispose(grads);\n\n    if (returnCost) {\n      return value;\n    } else {\n      value.dispose();\n      return null;\n    }\n  }\n\n  /**\n   * The number of iterations that this optimizer instance has been invoked for.\n   */\n  get iterations(): number {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return this.iterations_;\n  }\n\n  protected incrementIterations() {\n    this.iterations_ = this.iterations + 1;\n  }\n\n  /**\n   * Executes f() and computes the gradient of the scalar output of f() with\n   * respect to the list of trainable variables provided by `varList`. If no\n   * list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to use for computing\n   * gradients with respect to variables.\n   * @param varList An optional list of variables to compute gradients with\n   * respect to. If specified, only the trainable variables in varList will have\n   * gradients computed with respect to. Defaults to all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  computeGradients(f: () => Scalar, varList?: Variable[]):\n      {value: Scalar, grads: NamedTensorMap} {\n    return variableGrads(f, varList);\n  }\n\n  /**\n   * Updates variables by using the computed gradients.\n   *\n   * @param variableGradients A mapping of variable name to its gradient value.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  abstract applyGradients(variableGradients: NamedTensorMap|\n                          NamedTensor[]): void;\n\n  /**\n   * Dispose the variables (if any) owned by this optimizer instance.\n   */\n  dispose(): void {\n    if (this.iterations_ != null) {\n      dispose(this.iterations_);\n    }\n  }\n\n  async saveIterations(): Promise<NamedTensor> {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return {\n      name: 'iter',  // Named for Python compatibility.\n      // TODO(cais): Use 'int64' type when available.\n      tensor: scalar(this.iterations_, 'int32')\n    };\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for this optimizer yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error(\n        `setWeights() is not implemented for this optimizer class ` +\n        `${this.getClassName()}`);\n  }\n\n  /**\n   * Extract the first element of the weight values and set it\n   * as the iterations counter variable of this instance of optimizer.\n   *\n   * @param weightValues\n   * @returns Weight values with the first element consumed and excluded.\n   */\n  protected async extractIterations(weightValues: NamedTensor[]):\n      Promise<NamedTensor[]> {\n    this.iterations_ = (await weightValues[0].tensor.data())[0];\n    return weightValues.slice(1);\n  }\n}\n\nObject.defineProperty(Optimizer, Symbol.hasInstance, {\n  value: (instance: Optimizer) => {\n    return instance.minimize != null && instance.computeGradients != null &&\n        instance.applyGradients != null;\n  }\n});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/ops';\nimport {square} from '../ops/square';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdadeltaOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adadelta';  // Name matters for Python compatibility.\n  private accumulatedGrads: OptimizerVariable[] = [];\n  private accumulatedUpdates: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected rho: number,\n      protected epsilon: number = null) {\n    super();\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedGrads[i] == null) {\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accum_grad`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedUpdates[i] == null) {\n        this.accumulatedUpdates[i] = {\n          originalName: `${name}/accum_var`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n      const accumulatedUpdate = this.accumulatedUpdates[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad =\n            add(mul(accumulatedGrad, this.rho),\n                mul(square(gradient), 1 - this.rho));\n\n        const updates =\n            mul(div(sqrt(add(accumulatedUpdate, this.epsilon)),\n                    sqrt(add(accumulatedGrad, this.epsilon))),\n                gradient);\n\n        const newAccumulatedUpdate =\n            add(mul(accumulatedUpdate, this.rho),\n                mul(square(updates), 1 - this.rho));\n\n        accumulatedGrad.assign(newAccumulatedGrad);\n        accumulatedUpdate.assign(newAccumulatedUpdate);\n\n        const newValue = add(mul(updates, -this.learningRate), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedUpdates != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n      dispose(this.accumulatedUpdates.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedGrads, ...this.accumulatedUpdates];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedGrads =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedUpdates =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'rho': this.rho,\n      'epsilon': this.epsilon\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['rho'], config['epsilon']);\n  }\n}\nregisterClass(AdadeltaOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {fill} from '../ops/fill';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdagradOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adagrad';  // Note: Name matters for Python compatibility.\n\n  private accumulatedGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private initialAccumulatorValue = 0.1) {\n    super();\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulatedGrads[i] == null) {\n        const trainable = false;\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accumulator`,\n          variable: tidy(\n              () => fill(value.shape, this.initialAccumulatorValue)\n                        .variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad = add(accumulatedGrad, square(gradient));\n        accumulatedGrad.assign(newAccumulatedGrad);\n\n        const newValue = add(\n            mul(div(gradient,\n                    sqrt(add(newAccumulatedGrad, ENGINE.backend.epsilon()))),\n                -this.learningRate),\n            value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedGrads != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulatedGrads.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulatedGrads = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'initialAccumulatorValue': this.initialAccumulatorValue,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['initialAccumulatorValue']);\n  }\n}\nregisterClass(AdagradOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adam';  // Note: Name matters for Python compatibility.\n  private accBeta1: Variable;\n  private accBeta2: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedSecondMoment: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null) {\n    super();\n    tidy(() => {\n      // accB* will be updated by batch.\n      this.accBeta1 = scalar(beta1).variable();\n      this.accBeta2 = scalar(beta2).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const oneMinusAccBeta2 = sub(1, this.accBeta2);\n\n      varNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n        if (this.accumulatedSecondMoment[i] == null) {\n          this.accumulatedSecondMoment[i] = {\n            originalName: `${name}/v`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const secondMoment = this.accumulatedSecondMoment[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n        const newSecondMoment =\n            add(mul(secondMoment, this.beta2),\n                mul(square(gradient), 1 - this.beta2));\n\n        const biasCorrectedFirstMoment = div(newFirstMoment, oneMinusAccBeta1);\n        const biasCorrectedSecondMoment =\n            div(newSecondMoment, oneMinusAccBeta2);\n\n        firstMoment.assign(newFirstMoment);\n        secondMoment.assign(newSecondMoment);\n\n        const newValue =\n            add(mul(div(biasCorrectedFirstMoment,\n                        add(sqrt(biasCorrectedSecondMoment), this.epsilon)),\n                    -this.learningRate),\n                value);\n        value.assign(newValue);\n      });\n\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n      this.accBeta2.assign(mul(this.accBeta2, this.beta2));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.accBeta2.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedSecondMoment != null) {\n      dispose(this.accumulatedSecondMoment.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedFirstMoment, ...this.accumulatedSecondMoment];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    tidy(() => {\n      this.accBeta1.assign(pow(this.beta1, this.iterations_ + 1));\n      this.accBeta2.assign(pow(this.beta2, this.iterations_ + 1));\n    });\n\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedFirstMoment =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedSecondMoment =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon']);\n  }\n}\nregisterClass(AdamOptimizer);\n", "\uFEFF/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {abs} from '../ops/abs';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {maximum} from '../ops/maximum';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamaxOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adamax';  // Note: Name matters for Python compatbility.\n  private accBeta1: Variable;\n  private iteration: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null,\n      protected decay = 0.0) {\n    super();\n\n    tidy(() => {\n      this.iteration = scalar(0).variable();\n      this.accBeta1 = scalar(beta1).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const lr =\n          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));\n\n      variableNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n        if (this.accumulatedWeightedInfNorm[i] == null) {\n          this.accumulatedWeightedInfNorm[i] = {\n            originalName: `${name}/v`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n\n        const ut0 = mul(weightedInfNorm, this.beta2);\n        const ut1 = abs(gradient);\n\n        const newWeightedInfNorm = maximum(ut0, ut1);\n\n        firstMoment.assign(newFirstMoment);\n        weightedInfNorm.assign(newWeightedInfNorm);\n\n        const newValue =\n            add(mul(div(lr, oneMinusAccBeta1),\n                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),\n                value);\n\n        value.assign(newValue);\n      });\n\n      this.iteration.assign(add(this.iteration, 1));\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.iteration.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedWeightedInfNorm != null) {\n      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for Adamax yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error('setWeights() is not implemented for Adamax yet.');\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n      'decay': this.decay\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon'], config['decay']);\n  }\n}\nregisterClass(AdamaxOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {keep, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer} from './optimizer';\n\n/** @doclink Optimizer */\nexport class SGDOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'SGD';  // Note: Name matters for Python compatibility.\n  protected c: Scalar;\n\n  constructor(protected learningRate: number) {\n    super();\n    this.setLearningRate(learningRate);\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    varNames.forEach((name, i) => {\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n      const value = ENGINE.registeredVariables[name];\n      tidy(() => {\n        const newValue = add(mul(this.c, gradient), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  /**\n   * Sets the learning rate of the optimizer.\n   */\n  setLearningRate(learningRate: number) {\n    this.learningRate = learningRate;\n    if (this.c != null) {\n      this.c.dispose();\n    }\n    this.c = keep(scalar(-learningRate));\n  }\n\n  dispose() {\n    this.c.dispose();\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    return [await this.saveIterations()];\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    if (weightValues.length !== 0) {\n      throw new Error('SGD optimizer does not have settable weights.');\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {'learningRate': this.learningRate};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate']);\n  }\n}\nregisterClass(SGDOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {OptimizerVariable} from './optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\n/** @doclink Optimizer */\nexport class MomentumOptimizer extends SGDOptimizer {\n  /** @nocollapse */\n  static className = 'Momentum';  // Name matters for Python compatibility.\n  private m: Scalar;\n  private accumulations: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private momentum: number,\n      private useNesterov = false) {\n    super(learningRate);\n    this.m = scalar(this.momentum);\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulations[i] == null) {\n        const trainable = false;\n        this.accumulations[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const accumulation = this.accumulations[i].variable;\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      tidy(() => {\n        let newValue: Tensor;\n        const newAccumulation = add(mul(this.m, accumulation), gradient);\n        if (this.useNesterov) {\n          newValue = add(\n              mul(this.c, add(gradient, mul(newAccumulation, this.m))), value);\n        } else {\n          newValue = add(mul(this.c, newAccumulation), value);\n        }\n        accumulation.assign(newAccumulation);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.m.dispose();\n    if (this.accumulations != null) {\n      dispose(this.accumulations.map(v => v.variable));\n    }\n  }\n\n  /**\n   * Sets the momentum of the optimizer.\n   *\n   * @param momentum\n   */\n  setMomentum(momentum: number) {\n    this.momentum = momentum;\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulations.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulations = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'momentum': this.momentum,\n      'useNesterov': this.useNesterov\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['momentum'], config['useNesterov']);\n  }\n}\nregisterClass(MomentumOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class RMSPropOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'RMSProp';  // Note: Name matters for Python compatibility.\n  private centered: boolean;\n\n  private accumulatedMeanSquares: OptimizerVariable[] = [];\n  private accumulatedMoments: OptimizerVariable[] = [];\n  private accumulatedMeanGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected decay = 0.9,\n      protected momentum = 0.0, protected epsilon: number = null,\n      centered = false) {\n    super();\n\n    this.centered = centered;\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n    if (learningRate == null) {\n      throw new Error(`learningRate for RMSPropOptimizer must be defined.`);\n    }\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedMeanSquares[i] == null) {\n        this.accumulatedMeanSquares[i] = {\n          originalName: `${name}/rms`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMoments[i] == null) {\n        this.accumulatedMoments[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMeanGrads[i] == null && this.centered) {\n        this.accumulatedMeanGrads[i] = {\n          originalName: `${name}/mg`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedMeanSquare = this.accumulatedMeanSquares[i].variable;\n      const accumulatedMoments = this.accumulatedMoments[i].variable;\n      tidy(() => {\n        const newAccumulatedMeanSquare =\n            add(mul(accumulatedMeanSquare, this.decay),\n                mul(square(gradient), 1 - this.decay));\n\n        if (this.centered) {\n          const accumulatedMeanGrad = this.accumulatedMeanGrads[i].variable;\n          // Centered gradient\n          const newAccumulatedMeanGrad =\n              add(mul(accumulatedMeanGrad, this.decay),\n                  mul(gradient, 1 - this.decay));\n\n          const gradContribution =\n              div(mul(gradient, this.learningRate),\n                  sqrt(\n                      sub(newAccumulatedMeanSquare,\n                          add(square(newAccumulatedMeanGrad), this.epsilon))));\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum), gradContribution);\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMeanGrad.assign(newAccumulatedMeanGrad);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        } else {\n          // Plain gradient\n          const newAccumulatedMeanSquare =\n              add(mul(accumulatedMeanSquare, this.decay),\n                  mul(square(gradient), 1 - this.decay));\n\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum),\n                  div(mul(gradient, this.learningRate),\n                      sqrt(add(newAccumulatedMeanSquare, this.epsilon))));\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        }\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedMeanSquares != null) {\n      dispose(this.accumulatedMeanSquares.map(v => v.variable));\n    }\n    if (this.accumulatedMeanGrads != null && this.centered) {\n      dispose(this.accumulatedMeanGrads.map(v => v.variable));\n    }\n    if (this.accumulatedMoments != null) {\n      dispose(this.accumulatedMoments.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedMeanSquares, ...this.accumulatedMoments];\n    if (this.centered) {\n      variables.push(...this.accumulatedMeanGrads);\n    }\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount =\n        this.centered ? weightValues.length / 3 : weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedMeanSquares =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedMoments =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n    if (this.centered) {\n      this.accumulatedMeanGrads =\n          weightValues.slice(variableCount * 2, variableCount * 3)\n              .map(v => ({\n                     originalName: v.name,\n                     variable: v.tensor.variable(trainable)\n                   }));\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'decay': this.decay,\n      'momentum': this.momentum,\n      'epsilon': this.epsilon,\n      'centered': this.centered\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['decay'], config['momentum'],\n        config['epsilon'], config['centered']);\n  }\n}\nregisterClass(RMSPropOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AdadeltaOptimizer} from './adadelta_optimizer';\nimport {AdagradOptimizer} from './adagrad_optimizer';\nimport {AdamOptimizer} from './adam_optimizer';\nimport {AdamaxOptimizer} from './adamax_optimizer';\nimport {MomentumOptimizer} from './momentum_optimizer';\nimport {RMSPropOptimizer} from './rmsprop_optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\nexport class OptimizerConstructors {\n  /**\n   * Constructs a `tf.SGDOptimizer` that uses stochastic gradient descent.\n   *\n   * ```js\n   * // Fit a quadratic function by learning the coefficients a, b, c.\n   * const xs = tf.tensor1d([0, 1, 2, 3]);\n   * const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);\n   *\n   * const a = tf.scalar(Math.random()).variable();\n   * const b = tf.scalar(Math.random()).variable();\n   * const c = tf.scalar(Math.random()).variable();\n   *\n   * // y = a * x^2 + b * x + c.\n   * const f = x => a.mul(x.square()).add(b.mul(x)).add(c);\n   * const loss = (pred, label) => pred.sub(label).square().mean();\n   *\n   * const learningRate = 0.01;\n   * const optimizer = tf.train.sgd(learningRate);\n   *\n   * // Train the model.\n   * for (let i = 0; i < 10; i++) {\n   *   optimizer.minimize(() => loss(f(xs), ys));\n   * }\n   *\n   * // Make predictions.\n   * console.log(\n   *     `a: ${a.dataSync()}, b: ${b.dataSync()}, c: ${c.dataSync()}`);\n   * const preds = f(xs).dataSync();\n   * preds.forEach((pred, i) => {\n   *   console.log(`x: ${i}, pred: ${pred}`);\n   * });\n   * ```\n   *\n   * @param learningRate The learning rate to use for the SGD algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static sgd(learningRate: number): SGDOptimizer {\n    return new SGDOptimizer(learningRate);\n  }\n\n  /**\n   * Constructs a `tf.MomentumOptimizer` that uses momentum gradient\n   * descent.\n   *\n   * See\n   * [http://proceedings.mlr.press/v28/sutskever13.pdf](\n   * http://proceedings.mlr.press/v28/sutskever13.pdf)\n   *\n   * @param learningRate The learning rate to use for the Momentum gradient\n   * descent algorithm.\n   * @param momentum The momentum to use for the momentum gradient descent\n   * algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static momentum(learningRate: number, momentum: number, useNesterov = false):\n      MomentumOptimizer {\n    return new MomentumOptimizer(learningRate, momentum, useNesterov);\n  }\n\n  /**\n   * Constructs a `tf.RMSPropOptimizer` that uses RMSProp gradient\n   * descent. This implementation uses plain momentum and is not centered\n   * version of RMSProp.\n   *\n   * See\n   * [http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf](\n   * http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)\n   *\n   * @param learningRate The learning rate to use for the RMSProp gradient\n   * descent algorithm.\n   * @param decay The discounting factor for the history/coming gradient.\n   * @param momentum The momentum to use for the RMSProp gradient descent\n   * algorithm.\n   * @param epsilon Small value to avoid zero denominator.\n   * @param centered If true, gradients are normalized by the estimated\n   * variance of the gradient.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static rmsprop(\n      learningRate: number, decay = .9, momentum = 0.0, epsilon: number = null,\n      centered = false): RMSPropOptimizer {\n    return new RMSPropOptimizer(\n        learningRate, decay, momentum, epsilon, centered);\n  }\n\n  /**\n   * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adam gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adam(\n      learningRate = 0.001, beta1 = 0.9, beta2 = 0.999,\n      epsilon: number = null): AdamOptimizer {\n    return new AdamOptimizer(learningRate, beta1, beta2, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.\n   * See [https://arxiv.org/abs/1212.5701](https://arxiv.org/abs/1212.5701)\n   *\n   * @param learningRate The learning rate to use for the Adadelta gradient\n   * descent algorithm.\n   * @param rho The learning rate decay over each update.\n   * @param epsilon A constant epsilon used to better condition the grad\n   * update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adadelta(learningRate = .001, rho = .95, epsilon: number = null):\n      AdadeltaOptimizer {\n    return new AdadeltaOptimizer(learningRate, rho, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdamaxOptimizer` that uses the Adamax algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adamax gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   * @param decay The learning rate decay over each update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adamax(\n      learningRate = 0.002, beta1 = 0.9, beta2 = 0.999, epsilon: number = null,\n      decay = 0.0): AdamaxOptimizer {\n    return new AdamaxOptimizer(learningRate, beta1, beta2, epsilon, decay);\n  }\n\n  /**\n   * Constructs a `tf.AdagradOptimizer` that uses the Adagrad algorithm.\n   * See\n   * [http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf](\n   * http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\n   * or\n   * [http://ruder.io/optimizing-gradient-descent/index.html#adagrad](\n   * http://ruder.io/optimizing-gradient-descent/index.html#adagrad)\n   *\n   * @param learningRate The learning rate to use for the Adagrad gradient\n   * descent algorithm.\n   * @param initialAccumulatorValue Starting value for the accumulators, must be\n   * positive.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adagrad(learningRate: number, initialAccumulatorValue = 0.1):\n      AdagradOptimizer {\n    return new AdagradOptimizer(learningRate, initialAccumulatorValue);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// So typings can propagate.\nimport {AdadeltaOptimizer} from './optimizers/adadelta_optimizer';\nimport {AdagradOptimizer} from './optimizers/adagrad_optimizer';\nimport {AdamOptimizer} from './optimizers/adam_optimizer';\nimport {AdamaxOptimizer} from './optimizers/adamax_optimizer';\nimport {MomentumOptimizer} from './optimizers/momentum_optimizer';\nimport {OptimizerConstructors} from './optimizers/optimizer_constructors';\nimport {RMSPropOptimizer} from './optimizers/rmsprop_optimizer';\nimport {SGDOptimizer} from './optimizers/sgd_optimizer';\n\n// tslint:disable-next-line:no-unused-expression\n[MomentumOptimizer, SGDOptimizer, AdadeltaOptimizer, AdagradOptimizer,\n RMSPropOptimizer, AdamaxOptimizer, AdamOptimizer];\n\nexport const train = {\n  sgd: OptimizerConstructors.sgd,\n  momentum: OptimizerConstructors.momentum,\n  adadelta: OptimizerConstructors.adadelta,\n  adagrad: OptimizerConstructors.adagrad,\n  rmsprop: OptimizerConstructors.rmsprop,\n  adamax: OptimizerConstructors.adamax,\n  adam: OptimizerConstructors.adam\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst delayCallback: Function = (() => {\n  if (typeof requestAnimationFrame !== 'undefined') {\n    return requestAnimationFrame;\n  } else if (typeof setImmediate !== 'undefined') {\n    return setImmediate;\n  }\n  return (f: Function) => f();  // no delays\n})();\n\n/**\n * Returns a promise that resolve when a requestAnimationFrame has completed.\n *\n * On Node.js this uses setImmediate instead of requestAnimationFrame.\n *\n * This is simply a sugar method so that users can do the following:\n * `await tf.nextFrame();`\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nfunction nextFrame(): Promise<void> {\n  return new Promise<void>(resolve => delayCallback(() => resolve()));\n}\n\nexport {nextFrame};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Returns the image center in pixels.\nexport function getImageCenter(\n    center: number|[number, number], imageHeight: number,\n    imageWidth: number): [number, number] {\n  const centerX =\n      imageWidth * (typeof center === 'number' ? center : center[0]);\n  const centerY =\n      imageHeight * (typeof center === 'number' ? center : center[1]);\n  return [centerX, centerY];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Gets the new shape of the input Tensor after it's been reshaped\n * to:\n * [blockShape[0], ..., blockShape[M-1], batch / prod(blockShape),\n * inputShape[1], ..., inputShape[N-1]]\n *\n * See step 1: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshaped(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  let reshaped: number[] = [];\n  if (batchToSpace) {\n    reshaped = reshaped.concat(blockShape.slice(0));\n    reshaped.push(inputShape[0] / prod);\n    reshaped = reshaped.concat(inputShape.slice(1));\n  } else {\n    reshaped = reshaped.concat(inputShape[0]);\n    const spatialLength = blockShape.length;\n    for (let i = 0; i < spatialLength; ++i) {\n      reshaped =\n          reshaped.concat([inputShape[i + 1] / blockShape[i], blockShape[i]]);\n    }\n    reshaped = reshaped.concat(inputShape.slice(spatialLength + 1));\n  }\n  return reshaped;\n}\n\n/**\n * Gets the permutation that will transpose the dimensions of the\n * reshaped tensor to shape:\n *\n * [batch / prod(block_shape),inputShape[1], blockShape[0], ...,\n * inputShape[M], blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * see step 2: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getPermuted(\n    reshapedRank: number, blockShapeRank: number,\n    batchToSpace = true): number[] {\n  const permuted = [];\n  if (batchToSpace) {\n    permuted.push(blockShapeRank);\n    for (let i = blockShapeRank + 1; i < reshapedRank; ++i) {\n      if (i <= 2 * blockShapeRank) {\n        permuted.push(i);\n        permuted.push(i - (blockShapeRank + 1));\n      } else {\n        permuted.push(i);\n      }\n    }\n  } else {\n    const permutedBeforeBatch = [];\n    const permutedAfterBatch = [];\n    for (let i = 1; i < reshapedRank; ++i) {\n      if (i >= blockShapeRank * 2 + 1 || i % 2 === 1) {\n        permutedAfterBatch.push(i);\n      } else {\n        permutedBeforeBatch.push(i);\n      }\n    }\n    permuted.push(...permutedBeforeBatch);\n    permuted.push(0);\n    permuted.push(...permutedAfterBatch);\n  }\n  return permuted;\n}\n\n/**\n * Gets the shape of the reshaped and permuted input Tensor before any cropping\n * is applied.  The new shape will be:\n *\n * [batch / prod(blockShape),inputShape[1] * blockShape[0], ...,\n * inputShape[M] * blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 3: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshapedPermuted(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  const reshapedPermuted = [];\n\n  if (batchToSpace) {\n    reshapedPermuted.push(inputShape[0] / prod);\n  } else {\n    reshapedPermuted.push(inputShape[0] * prod);\n  }\n\n  for (let i = 1; i < inputShape.length; ++i) {\n    if (i <= blockShape.length) {\n      if (batchToSpace) {\n        reshapedPermuted.push(blockShape[i - 1] * inputShape[i]);\n      } else {\n        reshapedPermuted.push(inputShape[i] / blockShape[i - 1]);\n      }\n    } else {\n      reshapedPermuted.push(inputShape[i]);\n    }\n  }\n\n  return reshapedPermuted;\n}\n\n/**\n * Converts the crops argument into the beginning coordinates of a slice\n * operation.\n */\nexport function getSliceBeginCoords(\n    crops: number[][], blockShape: number): number[] {\n  const sliceBeginCoords = [0];\n  for (let i = 0; i < blockShape; ++i) {\n    sliceBeginCoords.push(crops[i][0]);\n  }\n  return sliceBeginCoords;\n}\n\n/**\n * Converts the crops argument into the size of a slice operation.  When\n * combined with getSliceBeginCoords this function allows the reshaped and\n * permuted Tensor to be cropped to its final output shape of:\n *\n * inputShape[1] * blockShape[0] - crops[0,0] - crops[0,1], ...,\n * inputShape[M] * blockShape[M-1] -crops[M-1,0] -\n * crops[M-1,1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 4: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getSliceSize(\n    uncroppedShape: number[], crops: number[][], blockShape: number): number[] {\n  const sliceSize = uncroppedShape.slice(0, 1);\n  for (let i = 0; i < blockShape; ++i) {\n    sliceSize.push(uncroppedShape[i + 1] - crops[i][0] - crops[i][1]);\n  }\n\n  return sliceSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const SELU_SCALEALPHA = 1.7580993408473768599402175208123;\nexport const SELU_SCALE = 1.0507009873554804934193349852946;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const ERF_P = 0.3275911;\nexport const ERF_A1 = 0.254829592;\nexport const ERF_A2 = -0.284496736;\nexport const ERF_A3 = 1.421413741;\nexport const ERF_A4 = -1.453152027;\nexport const ERF_A5 = 1.061405429;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\n\nexport function warn(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.warn(...msg);\n  }\n}\n\nexport function log(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.log(...msg);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TypedArray} from '../types';\n/**\n * Merges real and imaginary Float32Arrays into a single complex Float32Array.\n *\n * The memory layout is interleaved as follows:\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n * complex: [r0, i0, r1, i1, r2, i2]\n *\n * This is the inverse of splitRealAndImagArrays.\n *\n * @param real The real values of the complex tensor values.\n * @param imag The imag values of the complex tensor values.\n * @returns A complex tensor as a Float32Array with merged values.\n */\nexport function mergeRealAndImagArrays(\n    real: Float32Array, imag: Float32Array): Float32Array {\n  if (real.length !== imag.length) {\n    throw new Error(\n        `Cannot merge real and imag arrays of different lengths. real:` +\n        `${real.length}, imag: ${imag.length}.`);\n  }\n  const result = new Float32Array(real.length * 2);\n  for (let i = 0; i < result.length; i += 2) {\n    result[i] = real[i / 2];\n    result[i + 1] = imag[i / 2];\n  }\n  return result;\n}\n\n/**\n * Splits a complex Float32Array into real and imag parts.\n *\n * The memory layout is interleaved as follows:\n * complex: [r0, i0, r1, i1, r2, i2]\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n *\n * This is the inverse of mergeRealAndImagArrays.\n *\n * @param complex The complex tensor values.\n * @returns An object with real and imag Float32Array components of the complex\n *     tensor.\n */\nexport function splitRealAndImagArrays(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(complex.length / 2);\n  const imag = new Float32Array(complex.length / 2);\n  for (let i = 0; i < complex.length; i += 2) {\n    real[i / 2] = complex[i];\n    imag[i / 2] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts even indexed complex values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithEvenIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.ceil(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 0; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts odd indexed comple values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithOddIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.floor(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 2; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Get the map representing a complex value in the given array.\n * @param complex The complex tensor values.\n * @param index An index of the target complex value.\n */\nexport function getComplexWithIndex(\n    complex: Float32Array, index: number): {real: number, imag: number} {\n  const real = complex[index * 2];\n  const imag = complex[index * 2 + 1];\n  return {real, imag};\n}\n\n/**\n * Insert a given complex value into the TypedArray.\n * @param data The array in which the complex value is inserted.\n * @param c The complex value to be inserted.\n * @param index An index of the target complex value.\n */\nexport function assignToTypedArray(\n    data: TypedArray, real: number, imag: number, index: number) {\n  data[index * 2] = real;\n  data[index * 2 + 1] = imag;\n}\n\n/**\n * Make the list of exponent terms used by FFT.\n */\nexport function exponents(\n    n: number, inverse: boolean): {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(n / 2);\n  const imag = new Float32Array(n / 2);\n  for (let i = 0; i < Math.ceil(n / 2); i++) {\n    const x = (inverse ? 2 : -2) * Math.PI * (i / n);\n    real[i] = Math.cos(x);\n    imag[i] = Math.sin(x);\n  }\n  return {real, imag};\n}\n\n/**\n * Make the exponent term used by FFT.\n */\nexport function exponent(\n    k: number, n: number, inverse: boolean): {real: number, imag: number} {\n  const x = (inverse ? 2 : -2) * Math.PI * (k / n);\n  const real = Math.cos(x);\n  const imag = Math.sin(x);\n  return {real, imag};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {cast} from '../ops/cast';\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\nimport {Rank} from '../types';\nimport {DataType, ShapeMap} from '../types';\nimport {hasEncodingLoss, makeZerosTypedArray} from '../util';\n\nimport {KernelBackend} from './backend';\n\n// Utilities needed by backend consumers of tf-core.\nexport * from '../ops/axis_util';\nexport * from '../ops/broadcast_util';\nexport * from '../ops/concat_util';\nexport * from '../ops/conv_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/fused_types';\nexport * from '../ops/reduce_util';\n\nimport * as slice_util from '../ops/slice_util';\nexport {slice_util};\n\nexport {BackendValues, TypedArray, upcastType, PixelData} from '../types';\nexport {MemoryInfo, TimingInfo} from '../engine';\nexport * from '../ops/rotate_util';\nexport * from '../ops/array_ops_util';\nexport * from '../ops/gather_nd_util';\nexport * from '../ops/scatter_nd_util';\nexport * from '../ops/selu_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/erf_util';\nexport * from '../log';\nexport * from '../backends/complex_util';\nexport * from '../ops/split_util';\n\nimport * as segment_util from '../ops/segment_util';\nexport {segment_util};\n\nexport function castTensor<T extends Tensor>(\n    x: T, dtype: DataType, backend: KernelBackend): T {\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return x.clone();\n    }\n    const zerosTensor = zeros(x.shape);\n    const floatX = cast(x, 'float32');\n    const result = backend.complex(floatX, zerosTensor);\n    zerosTensor.dispose();\n    floatX.dispose();\n    return result as T;\n  }\n\n  if (!hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    return ENGINE.makeTensorFromDataId(x.dataId, x.shape, dtype) as T;\n  }\n  if (x.dtype === 'complex64') {\n    const real = backend.real(x);\n    const result = cast(real, dtype);\n    real.dispose();\n    return result;\n  }\n  if (dtype === 'int32') {\n    return backend.int(x);\n  } else if (dtype === 'bool') {\n    const zero = scalar(0, x.dtype);\n    const result = backend.notEqual(x, zero) as T;\n    zero.dispose();\n    return result;\n  } else {\n    throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n  }\n}\n\nexport function reshapeTensor<T extends Tensor, R extends Rank>(\n    x: T, shape: ShapeMap[R]): Tensor<R> {\n  return ENGINE.makeTensorFromDataId(x.dataId, shape, x.dtype) as Tensor<R>;\n}\n\nexport function linspaceImpl(start: number, stop: number, num: number) {\n  const step = (stop - start) / (num - 1);\n\n  const values = makeZerosTypedArray(num, 'float32');\n  values[0] = start;\n  for (let i = 1; i < values.length; i++) {\n    values[i] = values[i - 1] + step;\n  }\n\n  return tensor1d(values, 'float32');\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\n// TODO(annxingyuan): Use this helper in WASM Split kernel once intermediate\n// kernels have been modularized in WebGL and CPU\n// https://github.com/tensorflow/tfjs/issues/2822.\n/** Shared implementation of the split kernel across WebGL and CPU. */\nexport function split<T extends Tensor>(\n    x: T, sizeSplits: number[], axis: number): T[] {\n  const begin = new Array(x.rank).fill(0);\n  const size = x.shape.slice();\n  return sizeSplits.map(s => {\n    const sliceSize = [...size];\n    sliceSize[axis] = s;\n    const sliceT = slice(x, begin, sliceSize);\n    begin[axis] += s;\n    return sliceT;\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * An implementation of the tile kernel shared between webgl and cpu for string\n * tensors only.\n */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {DataType, Rank} from '../types';\n\nexport function tile<R extends Rank>(\n    xBuf: TensorBuffer<R, DataType>, reps: number[]): Tensor<R> {\n  const newShape: number[] = new Array(xBuf.rank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = xBuf.shape[i] * reps[i];\n  }\n  const result = buffer(newShape, xBuf.dtype);\n  for (let i = 0; i < result.values.length; ++i) {\n    const newLoc = result.indexToLoc(i);\n\n    const originalLoc: number[] = new Array(xBuf.rank);\n    for (let j = 0; j < originalLoc.length; j++) {\n      originalLoc[j] = newLoc[j] % xBuf.shape[j];\n    }\n\n    const originalIndex = xBuf.locToIndex(originalLoc);\n\n    result.values[i] = xBuf.values[originalIndex];\n  }\n  return result.toTensor() as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the TopK kernel shared between webgl and cpu. */\n\nimport {tensor} from '../ops/tensor';\nimport {Tensor} from '../tensor';\nimport {NumericDataType, TypedArray} from '../types';\nimport {getTypedArrayFromDType} from '../util';\n\nexport function topkImpl<T extends Tensor>(\n    x: TypedArray, xShape: number[], xDtype: NumericDataType, k: number,\n    sorted: boolean): [T, T] {\n  // Reshape into a 2d tensor [batch, lastDim] and compute topk along lastDim.\n  const lastDim = xShape[xShape.length - 1];\n  const [batch, size] = [x.length / lastDim, lastDim];\n  const allTopKVals = getTypedArrayFromDType(xDtype, batch * k);\n  const allTopKIndices = getTypedArrayFromDType('int32', batch * k);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = x.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    const outOffset = b * k;\n    const topKVals = allTopKVals.subarray(outOffset, outOffset + k);\n    const topKIndices = allTopKIndices.subarray(outOffset, outOffset + k);\n    for (let i = 0; i < k; i++) {\n      topKVals[i] = valAndInd[i].value;\n      topKIndices[i] = valAndInd[i].index;\n    }\n  }\n  // Reshape back to the original input shape, except that the last\n  // dimension is k.\n  const outputShape = xShape.slice();\n  outputShape[outputShape.length - 1] = k;\n  return [\n    tensor(allTopKVals, outputShape, xDtype) as T,\n    tensor(allTopKIndices, outputShape, 'int32') as T\n  ];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport {nonMaxSuppressionV3Impl, nonMaxSuppressionV4Impl, nonMaxSuppressionV5Impl} from './non_max_suppression_impl';\nexport {split} from './split_shared';\nexport {tile} from './tile_impl';\nexport {topkImpl} from './topk_impl';\nexport {whereImpl} from './where_impl';\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const absGradConfig: GradConfig = {\n  kernelName: Abs,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, step(cast(x, 'float32'), -1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acosGradConfig: GradConfig = {\n  kernelName: Acos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = square(cast(x, 'float32'));\n        const b = sqrt(sub(scalar(1), a));\n        return neg(div(dy, b));\n      }\n\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acoshGradConfig: GradConfig = {\n  kernelName: Acosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(sub(square(cast(x, 'float32')), 1));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Add} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const addGradConfig: GradConfig = {\n  kernelName: Add,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AddN} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const addNGradConfig: GradConfig = {\n  kernelName: AddN,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const ders: {[key: string]: () => Tensor} = {};\n    saved.forEach((_, i) => {\n      ders[i] = () => dy.clone();\n    });\n    return ders;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMax} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMaxGradConfig: GradConfig = {\n  kernelName: ArgMax,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMinGradConfig: GradConfig = {\n  kernelName: ArgMin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const asinGradConfig: GradConfig = {\n  kernelName: Asin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, sqrt(sub(scalar(1), square(cast(x, 'float32')))))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const asinhGradConfig: GradConfig = {\n  kernelName: Asinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(add(scalar(1), square(cast(x, 'float32'))));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const atan2GradConfig: GradConfig = {\n  kernelName: Atan2,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const d = add(square(a), square(b));\n      let res = mul(dy, div(b, d));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      const d = add(square(a), square(b));\n      let res = neg(mul(dy, div(a, d)));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const atanGradConfig: GradConfig = {\n  kernelName: Atan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, add(square(cast(x, 'float32')), 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {scalar} from '../ops/scalar';\nimport {Tensor} from '../tensor';\n\nexport const atanhGradConfig: GradConfig = {\n  kernelName: Atanh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, sub(scalar(1), square(cast(x, 'float32'))))};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool3DBackprop, AvgPool3DBackpropAttrs, AvgPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d avg pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0. The dilation\n *     rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction avgPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in avgPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in avgPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.avgPool3dBackprop(dy5D, input5D, convInfo);\n  };\n\n  const inputs: AvgPool3DBackpropInputs = {dy: dy5D, input: input5D};\n\n  const attrs: AvgPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      AvgPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3dBackprop = op({avgPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool3D, AvgPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPool3dBackprop} from '../ops/avg_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const avgPool3DGradConfig: GradConfig = {\n  kernelName: AvgPool3D,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as AvgPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => avgPool3dBackprop(\n          dy as Tensor5D, x, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of an 2D avg pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The input image, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction avgPoolBackprop_<T extends Tensor3D|Tensor4D>(\n    dy: T|TensorLike, input: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy (${\n          $dy.rank})`);\n\n  let input4D = $input as Tensor4D;\n  let dy4D = $dy as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($input.rank === 3) {\n    reshapedTo4D = true;\n    input4D =\n        reshape($input, [1, $input.shape[0], $input.shape[1], $input.shape[2]]);\n    dy4D = reshape($dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2]]);\n  }\n\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in avgPoolBackprop: dy must be rank 4 but got rank ` +\n          `${dy4D.rank}.`);\n  util.assert(\n      input4D.rank === 4,\n      () => `Error in avgPoolBackprop: input must be rank 4 but got rank ` +\n          `${input4D.rank}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        input4D.shape, filterSize, strides, 1 /* dilations */, pad);\n\n    return backend.avgPoolBackprop(dy4D, input4D, convInfo);\n  };\n\n  const inputs: AvgPoolBackpropInputs = {dy: dy4D, input: input4D};\n\n  const attrs: AvgPoolBackpropAttrs = {filterSize, strides, pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, AvgPoolBackprop,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const avgPoolBackprop = op({avgPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool, AvgPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPoolBackprop} from '../ops/avg_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const avgPoolGradConfig: GradConfig = {\n  kernelName: AvgPool,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as AvgPoolAttrs;\n    return {\n      x: () => avgPoolBackprop(dy as Tensor4D, x, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchMatMul, BatchMatMulAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {matMul} from '../ops/mat_mul';\nimport {Tensor, Tensor3D} from '../tensor';\n\nexport const batchMatMulGradConfig: GradConfig = {\n  kernelName: BatchMatMul,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor3D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [a, b] = saved as Tensor3D[];\n\n    const {transposeA, transposeB} = attrs as {} as BatchMatMulAttrs;\n\n    if (!transposeA && !transposeB) {\n      return {\n        a: () => matMul(dy, b, false, true),\n        b: () => matMul(a, dy, true, false)\n      };\n    } else if (!transposeA && transposeB) {\n      return {\n        a: () => matMul(dy, b, false, false),\n        b: () => matMul(dy, a, true, false)\n      };\n    } else if (transposeA && !transposeB) {\n      return {\n        a: () => matMul(b, dy, false, true),\n        b: () => matMul(a, dy, false, false)\n      };\n    } else {\n      return {\n        a: () => matMul(b, dy, true, true),\n        b: () => matMul(dy, a, true, true)\n      };\n    }\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchToSpaceND, BatchToSpaceNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {spaceToBatchND} from '../ops/space_to_batch_nd';\nimport {Tensor} from '../tensor';\n\nexport const batchToSpaceNDGradConfig: GradConfig = {\n  kernelName: BatchToSpaceND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, crops} = attrs as {} as BatchToSpaceNDAttrs;\n    return {x: () => spaceToBatchND(dy, blockShape, crops)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BroadcastTo, BroadCastToAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const broadcastToGradConfig: GradConfig = {\n  kernelName: BroadcastTo,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const broadCastToAttrs: BroadCastToAttrs =\n        attrs as unknown as BroadCastToAttrs;\n\n    const inputShape = broadCastToAttrs.inputShape;\n    const outputShape = broadCastToAttrs.shape;\n\n    const reps: number[] = Array.from(outputShape);\n    for (let i = inputShape.length - 1; i >= 0; i--) {\n      if (inputShape[i] === outputShape[i]) {\n        reps[i] = 1;\n      } else if (inputShape[i] !== 1) {\n        throw new Error(`broadcastTo(): [${\n            inputShape}] cannot be broadcast to [${outputShape}].`);\n      }\n    }\n    const axes: number[] = [];\n    for (let i = 0; i < reps.length; i++) {\n      if (reps[i] > 1) {\n        axes.push(i);\n      }\n    }\n\n    return {x: () => sum(dy, axes, true /* keepDims */)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cast} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const castGradConfig: GradConfig = {\n  kernelName: Cast,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => dy.clone()};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const ceilGradConfig: GradConfig = {\n  kernelName: Ceil,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {lessEqual} from '../ops/less_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const clipByValueGradConfig: GradConfig = {\n  kernelName: ClipByValue,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {clipValueMin, clipValueMax} = attrs as {} as ClipByValueAttrs;\n    return {\n      x: () => where(\n          logicalAnd(greaterEqual(x, clipValueMin), lessEqual(x, clipValueMax)),\n          dy, zerosLike(dy)),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Concat, ConcatAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {split} from '../ops/split';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const concatGradConfig: GradConfig = {\n  kernelName: Concat,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const shapes = saved.map(t => t.shape);\n    const {axis} = attrs as {} as ConcatAttrs;\n    const $axis = parseAxisParam(axis, saved[0].shape)[0];\n    const sizeSplits = shapes.map(s => s[$axis]);\n    const derTensors = split(dy, sizeSplits, $axis);\n    return derTensors.map(t => () => t) as {};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2D, Conv2DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../ops/conv2d_backprop_input';\nimport * as conv_util from '../ops/conv_util';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv2DGradConfig: GradConfig = {\n  kernelName: Conv2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x4D, $filter] = saved as [Tensor4D, Tensor4D];\n    const {dilations, strides, pad, dataFormat} = attrs as {} as Conv2DAttrs;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of conv2D: dilation rates greater than 1 ' +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    return {\n      x: () =>\n          conv2DBackpropInput(x4D.shape, dy, $filter, strides, pad, dataFormat),\n      filter: () =>\n          conv2DBackpropFilter(x4D, dy, $filter.shape, strides, pad, dataFormat)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2d} from '../ops/conv2d';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const conv2DBackpropInputGradConfig: GradConfig = {\n  kernelName: Conv2DBackpropInput,\n  inputsToSave: ['dy', 'filter'],\n  gradFunc: (ddx: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [dy, filter] = saved as [Tensor4D, Tensor4D];\n\n    const {strides, pad, dataFormat, dimRoundingMode} =\n        attrs as {} as Conv2DBackpropInputAttrs;\n\n    return {\n      dy: () => conv2d(\n          ddx, filter, strides, pad, dataFormat, 1 /* dilations */,\n          dimRoundingMode),\n      filter: () => conv2DBackpropFilter(\n          ddx, dy, filter.shape, strides, pad, dataFormat, dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropFilterAttrs, Conv3DBackpropFilterInputs, Conv3DBackpropFilterV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 3D convolution.\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     [batch, depth, height, width, inChannels]. If rank 4, batch of 1 is\n *     assumed.\n * @param dy The dy image, of rank 5 or rank 4, of shape\n *     [batch, depth, height, width, outDepth]. If rank 4, batch of 1 is\n *     assumed.\n * @param filterShape The shape of the filter, length 5,\n *     [filterDepth, filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideDepth, strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction conv3DBackpropFilter_<T extends Tensor4D|Tensor5D>(\n    x: T, dy: T, filterShape: [number, number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): Tensor5D {\n  let x5D = x as Tensor5D;\n  if (x.rank === 4) {\n    x5D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2], x.shape[3]]);\n  }\n  let dy5D = dy as Tensor5D;\n  if (dy5D.rank === 4) {\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3dDerFilter: input must be rank 5, but got shape ` +\n          `${x5D.shape}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerFilter: dy must be rank 5, but got shape ` +\n          `${dy5D.shape}.`);\n  util.assert(\n      filterShape.length === 5,\n      () => `Error in conv3dDerFilter: filterShape must be length 5, but got ` +\n          `${filterShape}.`);\n  util.assert(\n      x5D.shape[4] === filterShape[3],\n      () => `Error in conv3dDerFilter: depth of input ${x5D.shape[4]}) must ` +\n          `match input depth in filter (${filterShape[3]}.`);\n  util.assert(\n      dy5D.shape[4] === filterShape[4],\n      () => `Error in conv3dDerFilter: depth of dy (${dy5D.shape[4]}) must ` +\n          `match output depth for filter (${filterShape[4]}).`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, filterShape, strides, dilations, pad);\n\n    return backend.conv3dDerFilter(x5D, dy5D, convInfo);\n  };\n\n  const inputs: Conv3DBackpropFilterInputs = {x: x5D, y: dy5D};\n\n  const attrs: Conv3DBackpropFilterAttrs = {strides, pad};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv3DBackpropFilterV2, attrs as {} as NamedAttrMap) as Tensor5D;\n}\n\nexport const conv3DBackpropFilter = op({conv3DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv3D, Conv3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv3DBackpropFilter} from '../ops/conv3d_backprop_filter';\nimport {conv3DBackpropInput} from '../ops/conv3d_backprop_input';\nimport {tupleValuesAreOne} from '../ops/conv_util';\nimport {Tensor, Tensor5D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv3DGradConfig: GradConfig = {\n  kernelName: Conv3D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor5D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad} = attrs as {} as Conv3DAttrs;\n    util.assert(\n        tupleValuesAreOne(dilations),\n        () =>\n            'Error in gradient of conv3D: dilation rates greater than 1 are ' +\n            `not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const [x5D, $filter] = saved;\n\n    return {\n      x: () => conv3DBackpropInput(\n          (x5D as Tensor5D).shape, dy, $filter as Tensor5D, strides, pad),\n      filter: () => conv3DBackpropFilter(\n          x5D as Tensor5D, dy, ($filter as Tensor5D).shape, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {sin} from '../ops/sin';\nimport {Tensor} from '../tensor';\n\nexport const cosGradConfig: GradConfig = {\n  kernelName: Cos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(neg(sin(cast(x, 'float32'))), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {sinh} from '../ops/sinh';\nimport {Tensor} from '../tensor';\n\nexport const coshGradConfig: GradConfig = {\n  kernelName: Cosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(sinh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cumsum, CumsumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getAxesPermutation} from '../ops/axis_util';\nimport {cumsum} from '../ops/cumsum';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const cumsumGradConfig: GradConfig = {\n  kernelName: Cumsum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {axis, exclusive, reverse}: CumsumAttrs = attrs as {} as CumsumAttrs;\n\n    return {\n      x: () => {\n        const permutation = getAxesPermutation([axis], x.rank);\n\n        let out = cumsum(dy, axis, exclusive, !reverse);\n\n        if (permutation != null) {\n          out = transpose(out, permutation);\n        }\n\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as conv_util from '../ops/conv_util';\nimport {depthwiseConv2dNativeBackpropFilter} from '../ops/depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../ops/depthwise_conv2d_native_backprop_input';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const depthwiseConv2dNativeGradConfig: GradConfig = {\n  kernelName: DepthwiseConv2dNative,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad, dimRoundingMode} =\n        attrs as {} as DepthwiseConv2dNativeAttrs;\n\n    const $dilations = dilations == null ? [1, 1] : dilations;\n\n    util.assert(\n        conv_util.tupleValuesAreOne($dilations),\n        () => 'Error in gradient of depthwiseConv2dNative: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${$dilations}'`);\n\n    const [x, filter] = saved as [Tensor4D, Tensor4D];\n\n    util.assert(\n        x.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: input must be ` +\n            `rank 4, but got rank ${x.rank}.`);\n    util.assert(\n        filter.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: filter must be ` +\n            `rank 4, but got rank ${filter.rank}.`);\n    util.assert(\n        x.shape[3] === filter.shape[2],\n        () => `Error in gradient of depthwiseConv2d: number of input ` +\n            `channels (${x.shape[3]}) must match the inChannels dimension ` +\n            `in filter ${filter.shape[2]}.`);\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, $dilations),\n        () => 'Error in gradient of depthwiseConv2d: Either strides or ' +\n            `dilations must be  1. Got strides ${strides} and dilations ` +\n            `'${$dilations}'.`);\n\n    if (dimRoundingMode != null) {\n      util.assert(\n          util.isInt(pad as number),\n          () =>\n              `Error in depthwiseConv2d: pad must be an integer when using, ` +\n              `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n    }\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x.shape, filter.shape, strides, $dilations as number | [number, number],\n        pad, dimRoundingMode, true /* depthwise */);\n\n    return {\n      x: () =>\n          depthwiseConv2dNativeBackpropInput(x.shape, dy, filter, convInfo),\n      filter: () =>\n          depthwiseConv2dNativeBackpropFilter(x, dy, filter.shape, convInfo),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DBackpropFilter, Dilation2DBackpropFilterInputs, Dilation2DBackpropInput, Dilation2DBackpropInputInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const dilation2dGradConfig: GradConfig = {\n  kernelName: Dilation2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, filter] = saved as [Tensor4D, Tensor3D];\n\n    const inputInputs: Dilation2DBackpropInputInputs = {x, filter, dy};\n    const filterInputs: Dilation2DBackpropFilterInputs = {x, filter, dy};\n\n    return {\n      x: () => ENGINE.runKernel(\n                   Dilation2DBackpropInput, inputInputs as {} as NamedTensorMap,\n                   attrs) as Tensor,\n      filter: () => ENGINE.runKernel(\n                        Dilation2DBackpropFilter,\n                        filterInputs as {} as NamedTensorMap, attrs) as Tensor\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const divGradConfig: GradConfig = {\n  kernelName: Div,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluGrad, EluGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const eluGradConfig: GradConfig = {\n  kernelName: Elu,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      return backend.eluDer(dy, y);\n    };\n\n    const inputs: EluGradInputs = {dy, y};\n\n    return {\n      x: () => ENGINE.runKernelFunc(\n          backPropKernelFunc, inputs as {} as NamedTensorMap, null /* grad */,\n          EluGrad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Erf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const erfGradConfig: GradConfig = {\n  kernelName: Erf,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const a = mul(exp(neg(square(x))), 2 / Math.sqrt(Math.PI));\n    return {x: () => mul(dy, a)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expGradConfig: GradConfig = {\n  kernelName: Exp,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n    return {x: () => mul(dy, y)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expm1GradConfig: GradConfig = {\n  kernelName: Expm1,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, exp(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const floorGradConfig: GradConfig = {\n  kernelName: Floor,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FloorDiv} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const floorDivGradConfig: GradConfig = {\n  kernelName: FloorDiv,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {FusedBatchNorm, FusedBatchNormAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {rsqrt} from '../ops/rsqrt';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {tile} from '../ops/tile';\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nexport const fusedBatchNormGradConfig: GradConfig = {\n  kernelName: FusedBatchNorm,\n  inputsToSave: ['x', 'mean', 'variance', 'scale'],\n  gradFunc: <R extends Rank>(\n      dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {varianceEpsilon} = attrs as {} as FusedBatchNormAttrs;\n    const [x, mean, variance, scale] = saved;\n\n    const scaleValue = scale == null ? scalar(1) : scale;\n    const reductionAxes = getReductionAxes(mean.shape, x.shape);\n    const tileShape: number[] = [];\n    if (mean.rank === 1) {\n      for (let i = 0; i < x.shape.length - 1; ++i) {\n        tileShape.push(x.shape[i]);\n      }\n      tileShape.push(1);\n    }\n\n    const xMinusMean = sub(x, mean);\n    const dyTimesScaleValue = mul(dy, scaleValue);\n    const oneOverSqrtVariance = rsqrt(add(variance, scalar(varianceEpsilon)));\n    const minusHalfRCube = mul(\n        mul(mul(oneOverSqrtVariance, oneOverSqrtVariance), oneOverSqrtVariance),\n        scalar(-0.5));\n\n    const derX = () => {\n      if (mean.rank === 1) {\n        return reshape(\n            mul(mul(dy,\n                    tile(\n                        reshape(oneOverSqrtVariance, [1, 1, 1, mean.shape[0]]),\n                        tileShape)),\n                scaleValue),\n            x.shape);\n      } else {\n        return reshape(mul(mul(dy, oneOverSqrtVariance), scaleValue), x.shape);\n      }\n    };\n    const derMean = () => {\n      let meanDer =\n          mul(mul(oneOverSqrtVariance, scalar(-1)), dyTimesScaleValue);\n      if (mean.rank === 1) {\n        meanDer = sum(meanDer, reductionAxes);\n      }\n      return reshape(meanDer, mean.shape as ShapeMap[R]);\n    };\n    const derVariance = () => {\n      let varianceDer = mul(mul(minusHalfRCube, xMinusMean), dyTimesScaleValue);\n\n      if (mean.rank === 1) {\n        varianceDer = sum(varianceDer, reductionAxes);\n      }\n      return reshape(varianceDer, mean.shape as ShapeMap[R]);\n    };\n    const derScale = () => {\n      const xMinusMean2TimesRsqrt = mul(xMinusMean, oneOverSqrtVariance);\n\n      let scaleDer = mul(dy, xMinusMean2TimesRsqrt);\n      if (mean.rank === 1) {\n        scaleDer = sum(scaleDer, reductionAxes);\n      }\n      return reshape(scaleDer, mean.shape as ShapeMap[R]);\n    };\n    const derOffset = () => {\n      let offsetDer = dy;\n      if (mean.rank === 1) {\n        offsetDer = sum(offsetDer, reductionAxes);\n      }\n      return reshape(offsetDer, mean.shape as ShapeMap[R]);\n    };\n\n    return {\n      x: derX,\n      mean: derMean,\n      variance: derVariance,\n      scale: derScale,\n      offset: derOffset\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GatherV2, GatherV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getUndoAxesPermutation} from '../ops/axis_util';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {unsortedSegmentSum} from '../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const gatherGradConfig: GradConfig = {\n  kernelName: GatherV2,\n  inputsToSave: ['x', 'indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, indices] = saved;\n    const {axis} = attrs as {} as GatherV2Attrs;\n\n    const parsedAxis = parseAxisParam(axis, x.shape)[0];\n\n    const derX = () => {\n      const paramsShape = x.shape;\n      const indicesSize = indices.size;\n\n      const outerShape = paramsShape.slice(0, parsedAxis);\n      const outerDims = outerShape.length;\n      const innerShape = paramsShape.slice(axis, paramsShape.length).slice(1);\n      const innerDims = innerShape.length;\n\n      const outerAxesIndices = arrayRange(0, outerDims);\n      const innerAxesIndices =\n          arrayRange(outerDims + 1, outerDims + 1 + innerDims);\n\n      const valuesShape = arrayConcat([outerShape, [indicesSize], innerShape]);\n\n      const values = reshape(dy, valuesShape);\n      const reshapedIndices = reshape(indices, [indicesSize]);\n\n      const transposeDims =\n          arrayConcat([[outerDims], outerAxesIndices, innerAxesIndices]);\n      const valuesTranspose = transpose(values, transposeDims);\n      let paramsGrad = unsortedSegmentSum(\n          valuesTranspose, reshapedIndices as Tensor1D, x.shape[parsedAxis]);\n\n      const invertTransposeDims = getUndoAxesPermutation(transposeDims);\n      paramsGrad = transpose(paramsGrad, invertTransposeDims);\n\n      return paramsGrad;\n    };\n    return {x: derX, indices: () => indices};\n  }\n};\n\nfunction arrayRange(start: number, stop: number): number[] {\n  const result = [];\n  for (let i = start; i < stop; ++i) {\n    result.push(i);\n  }\n  return result;\n}\n\nfunction arrayConcat(arrays: number[][]): number[] {\n  const result = [];\n  for (let i = 0; i < arrays.length; ++i) {\n    for (let j = 0; j < arrays[i].length; ++j) {\n      result.push(arrays[i][j]);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GreaterEqual} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const greaterEqualGradConfig: GradConfig = {\n  kernelName: GreaterEqual,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    return {a: () => zerosLike(a), b: () => zerosLike(b)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {Tensor} from '../tensor';\n\nexport const identityGradConfig: GradConfig = {\n  kernelName: Identity,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => cast(dy, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isFiniteGradConfig: GradConfig = {\n  kernelName: IsFinite,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isInfGradConfig: GradConfig = {\n  kernelName: IsInf,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isNanGradConfig: GradConfig = {\n  kernelName: IsNan,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log1p} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const log1pGradConfig: GradConfig = {\n  kernelName: Log1p,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, add(x, 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const logGradConfig: GradConfig = {\n  kernelName: Log,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, cast(x, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {LogSoftmax, LogSoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const logSoftmaxGradConfig: GradConfig = {\n  kernelName: LogSoftmax,\n  inputsToSave: [],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [value] = saved;\n    const {axis} = attrs as {} as LogSoftmaxAttrs;\n    return {\n      logits: () => {\n        const keepDims = true;\n        const softmax = exp(value);\n        return sub(dy, mul(sum(dy, axis, keepDims), softmax));\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRNBackprop, LRNBackpropAttrs, LRNBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport {op} from './operation';\n\nfunction localResponseNormalizationBackprop_<T extends Tensor4D>(\n    x: T, y: T, dy: T, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.LRNGrad(dy, x, y, depthRadius, bias, alpha, beta);\n\n  const inputs: LRNBackpropInputs = {x, y, dy};\n\n  const attrs: LRNBackpropAttrs = {depthRadius, bias, alpha, beta};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LRNBackprop, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const localResponseNormalizationBackprop =\n    op({localResponseNormalizationBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {LRN, LRNAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {localResponseNormalizationBackprop} from '../ops/local_response_normalization_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const lrnGradConfig: GradConfig = {\n  kernelName: LRN,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {depthRadius, bias, alpha, beta} = attrs as {} as LRNAttrs;\n\n    return {\n      x: () => localResponseNormalizationBackprop(\n          x, y, dy, depthRadius, bias, alpha, beta)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as axis_util from '../ops/axis_util';\nimport {cast} from '../ops/cast';\nimport {equal} from '../ops/equal';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\n/**\n * Gradient helper function for the min and max operations.\n */\nexport function gradForMinAndMax<T extends Tensor>(\n    dy: T, y: T, xOrig: Tensor, origAxes: number[], permutedAxes: number[]) {\n  if (y.rank < xOrig.rank) {\n    y = reshape(y, axis_util.expandShapeToKeepDim(y.shape, origAxes)) as T;\n  }\n  if (dy.rank < xOrig.rank) {\n    dy = reshape(dy, axis_util.expandShapeToKeepDim(dy.shape, origAxes)) as T;\n  }\n  return {\n    x: () => {\n      const dx = mul(dy, cast(equal(xOrig, y), dy.dtype));\n      return permutedAxes == null ? dx : transpose(dx, permutedAxes);\n    }\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const maxGradConfig: GradConfig = {\n  kernelName: Max,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const maxAttrs: MaxAttrs = attrs as {} as MaxAttrs;\n    const {reductionIndices} = maxAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const maxGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = maxGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Maximum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {less} from '../ops/less';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const maximumGradConfig: GradConfig = {\n  kernelName: Maximum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(greaterEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(less(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool3DBackprop, MaxPool3DBackpropAttrs, MaxPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d max pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank 4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param output The original output image, of rank 5 of shape\n *     [batchSize, outDepth, outHeight, outWidth, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike, output: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'maxPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPool3dBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let output5D = $output as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n    output5D = reshape($output, [\n      1, $output.shape[0], $output.shape[1], $output.shape[2], $output.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in maxPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in maxPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n  util.assert(\n      output5D.rank === 5,\n      () => `Error in maxPool3dBackprop: output must be rank 5 but got rank ` +\n          `${output5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.maxPool3dBackprop(dy5D, input5D, output5D, convInfo);\n  };\n\n  const inputs:\n      MaxPool3DBackpropInputs = {dy: dy5D, input: input5D, output: output5D};\n\n  const attrs: MaxPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      MaxPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3dBackprop = op({maxPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool3D, MaxPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPool3dBackprop} from '../ops/max_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const maxPool3DGradConfig: GradConfig = {\n  kernelName: MaxPool3D,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor5D, Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as MaxPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => maxPool3dBackprop(\n          dy as Tensor5D, x, y, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\n\n/**\n * Computes the backprop of a 2D max pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The original input image, of rank 4, of shape\n *     [batchSize, height, width, channels].\n * @param output The original output image, of rank 4, of shape\n *     [batchSize, outHeight, outWidth, channels].\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPoolBackprop_(\n    dy: Tensor4D|TensorLike, input: Tensor4D|TensorLike,\n    output: Tensor4D|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  const $dy = convertToTensor(dy, 'dy', 'maxPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPoolBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy ` +\n          `(${$dy.rank})`);\n\n  util.assert(\n      $dy.rank === 4,\n      () => `Error in maxPoolBackprop: dy must be rank 4 but got rank ` +\n          `${$dy.rank}.`);\n  util.assert(\n      $input.rank === 4,\n      () => `Error in maxPoolBackprop: input must be rank 4 but got rank ` +\n          `${$input.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPoolBackprop: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        $input.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    return backend.maxPoolBackprop($dy, $input, $output, convInfo);\n  };\n\n  const inputs:\n      MaxPoolBackpropInputs = {dy: $dy, input: $input, output: $output};\n\n  const attrs:\n      MaxPoolBackpropAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, MaxPoolBackprop,\n             attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const maxPoolBackprop = op({maxPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool, MaxPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPoolBackprop} from '../ops/max_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const maxPoolGradConfig: GradConfig = {\n  kernelName: MaxPool,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as MaxPoolAttrs;\n\n    return {\n      x: () => maxPoolBackprop(dy as Tensor4D, x, y, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Min, MinAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const minGradConfig: GradConfig = {\n  kernelName: Min,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const minAttrs: MinAttrs = attrs as {} as MinAttrs;\n    const {axis} = minAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(axis, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const minGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = minGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Minimum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const minimumGradConfig: GradConfig = {\n  kernelName: Minimum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(lessEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(greater(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Mod} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {floor} from '../ops/floor';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const modGradConfig: GradConfig = {\n  kernelName: Mod,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(dy, reduceAxes), a.shape);\n      }\n      return dy;\n    };\n    const derB = () => {\n      const res = mul(dy, neg(floor(div(a, b))));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Multiply} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const multiplyGradConfig: GradConfig = {\n  kernelName: Multiply,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = mul(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      const res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Negate} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {neg} from '../ops/neg';\nimport {Tensor} from '../tensor';\n\nexport const negateGradConfig: GradConfig = {\n  kernelName: Negate,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => neg(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OneHot} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\n\nexport const oneHotGradConfig: GradConfig = {\n  kernelName: OneHot,\n  inputsToSave: ['indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const indices = saved[0];\n    return {indices: () => zeros(indices.shape, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OnesLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const onesLikeGradConfig: GradConfig = {\n  kernelName: OnesLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PadV2, PadV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\nexport const padV2GradConfig: GradConfig = {\n  kernelName: PadV2,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    // Pad introduces values around the original tensor, so the gradient\n    // slices the original shape out of the gradient.\n    const x = saved[0];\n    const {paddings} = attrs as unknown as PadV2Attrs;\n    const begin = paddings.map(p => p[0]);\n    return {x: () => slice(dy, begin, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Pow} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {log} from '../ops/log';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {reshape} from '../ops/reshape';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const powGradConfig: GradConfig = {\n  kernelName: Pow,\n  inputsToSave: ['a', 'b'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b, y] = saved;\n    const base = a;\n    const exp = b;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(base.shape, exp.shape);\n\n    const derBase = () => {\n      const expFloat = cast(exp, 'float32');\n      let res = mul(dy, mul(expFloat, pow(base, sub(expFloat, scalar(1)))));\n      const reduceAxes = broadcast_util.getReductionAxes(base.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, base.shape);\n    };\n    const derExp = () => {\n      const condition = greater(base, 0);\n      const logBase = where(condition, log(base), zerosLike(base));\n      let res = mul(dy, mul(y, logBase));\n      const reduceAxes = broadcast_util.getReductionAxes(exp.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, exp.shape);\n    };\n    return {a: derBase, b: derExp};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Prelu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const preluGradConfig: GradConfig = {\n  kernelName: Prelu,\n  inputsToSave: ['x', 'alpha'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x, alpha] = saved;\n    const mask = greater(x, 0);\n\n    return {\n      x: () => where(mask, dy, mul(dy, alpha)),\n      alpha: () => {\n        let res = where(mask, zerosLike(dy), mul(dy, x));\n        const reduceAxes = getReductionAxes(alpha.shape, dy.shape);\n        if (reduceAxes.length > 0) {\n          res = sum(res, reduceAxes);\n        }\n        return reshape(res, alpha.shape);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Reciprocal} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const reciprocalGradConfig: GradConfig = {\n  kernelName: Reciprocal,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, neg(square(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu6} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const relu6GradConfig: GradConfig = {\n  kernelName: Relu6,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const mask = mul(lessEqual(x, 6), step(x));\n\n    return {x: () => mul(dy, cast(mask, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const reluGradConfig: GradConfig = {\n  kernelName: Relu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, cast(step(x), 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reshape} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\n\nexport const reshapeGradConfig: GradConfig = {\n  kernelName: Reshape,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => reshape(dy, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearGrad, ResizeBilinearGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeBilinearGradConfig: GradConfig = {\n  kernelName: ResizeBilinear,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeBilinearAttrs;\n      return backend.resizeBilinearBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeBilinearGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeBilinearGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborGrad, ResizeNearestNeighborGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeNearestNeighborGradConfig: GradConfig = {\n  kernelName: ResizeNearestNeighbor,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeNearestNeighborAttrs;\n      return backend.resizeNearestNeighborBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeNearestNeighborGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeNearestNeighborGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reverse, ReverseAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {reverse} from '../ops/reverse';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const reverseGradConfig: GradConfig = {\n  kernelName: Reverse,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dims} = attrs as {} as ReverseAttrs;\n    const axes = parseAxisParam(dims, dy.shape);\n    return {x: () => reverse(dy, axes)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Round} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const roundGradConfig: GradConfig = {\n  kernelName: Round,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rsqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {pow} from '../ops/pow';\nimport {Tensor} from '../tensor';\n\nexport const rsqrtGradConfig: GradConfig = {\n  kernelName: Rsqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => neg(div(dy, mul(pow(x, 1.5), 2)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SelectV2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {logicalNot} from '../ops/logical_not';\nimport {mul} from '../ops/mul';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const selectV2PoolGradConfig: GradConfig = {\n  kernelName: SelectV2,\n  inputsToSave: ['condition'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [condition] = saved;\n    return {\n      // TODO(julianoks): Return null for condition gradient\n      // when backprop supports it.\n      condition: () => cast(zerosLike(condition), 'float32'),\n      t: () => mul(dy, cast(condition, dy.dtype)),\n      e: () => mul(dy, cast(logicalNot(condition), dy.dtype))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Selu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {exp} from '../ops/exp';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {SELU_SCALE, SELU_SCALEALPHA} from '../ops/selu_util';\nimport {where} from '../ops/where';\nimport {Tensor} from '../tensor';\n\nexport const seluGradConfig: GradConfig = {\n  kernelName: Selu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {\n      x: () => {\n        const mask = greater(x, scalar(0));\n\n        const scaleAlpha = scalar(SELU_SCALEALPHA);\n        const scale = scalar(SELU_SCALE);\n\n        const greaterThanZeroDer = mul(dy, scale);\n        const lessEqualZeroDer =\n            mul(mul(dy, scaleAlpha), exp(cast(x, 'float32')));\n\n        return where(mask, greaterThanZeroDer, lessEqualZeroDer);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sigmoid} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const sigmoidGradConfig: GradConfig = {\n  kernelName: Sigmoid,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(dy, mul(y, sub(scalar(1), y)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sign} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const signGradConfig: GradConfig = {\n  kernelName: Sign,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cos} from '../ops/cos';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinGradConfig: GradConfig = {\n  kernelName: Sin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(cos(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cosh} from '../ops/cosh';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinhGradConfig: GradConfig = {\n  kernelName: Sinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(cosh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Slice, SliceAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {pad} from '../ops/pad';\nimport {parseSliceParams} from '../ops/slice_util';\nimport {Tensor} from '../tensor';\n\nexport const sliceGradConfig: GradConfig = {\n  kernelName: Slice,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {begin, size} = attrs as {} as SliceAttrs;\n\n    const inputShape = x.shape;\n    const [begin_, size_] = parseSliceParams(x, begin, size);\n\n    // Create an Nx2 padding where the first column represents how many\n    // zeros are prepended (at start) for each dimension, and the second\n    // column indicates how many zeros are appended (at end).\n\n    // The number of zeros to append is the shape of the input\n    // elementwise-subtracted by both the begin vector and sizes vector.\n    const paddings: Array<[number, number]> = [];\n    for (let i = 0; i < dy.rank; i++) {\n      paddings.push([begin_[i], inputShape[i] - begin_[i] - size_[i]]);\n    }\n    return {x: () => pad(dy, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softmax, SoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const softmaxGradConfig: GradConfig = {\n  kernelName: Softmax,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [y] = saved;\n    const {dim} = attrs as {} as SoftmaxAttrs;\n    const keepDims = true;\n\n    const dyTimesY = mul(dy, y);\n    return {\n      logits: () => sub(dyTimesY, mul(sum(dyTimesY, [dim], keepDims), y))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softplus} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sigmoid} from '../ops/sigmoid';\nimport {Tensor} from '../tensor';\n\nexport const softplusGradConfig: GradConfig = {\n  kernelName: Softplus,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(dy, sigmoid(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SpaceToBatchND, SpaceToBatchNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {batchToSpaceND} from '../ops/batch_to_space_nd';\nimport {Tensor} from '../tensor';\n\nexport const spaceToBatchNDGradConfig: GradConfig = {\n  kernelName: SpaceToBatchND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, paddings} = attrs as {} as SpaceToBatchNDAttrs;\n    return {x: () => batchToSpaceND(dy, blockShape, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {SplitV, SplitVAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {concat} from '../ops/concat';\nimport {Tensor} from '../tensor';\n\nexport const splitVGradConfig: GradConfig = {\n  kernelName: SplitV,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const {axis} = attrs as {} as SplitVAttrs;\n\n    return {x: () => concat(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {Tensor} from '../tensor';\n\nexport const sqrtGradConfig: GradConfig = {\n  kernelName: Sqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, mul(sqrt(cast(x, 'float32')), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const squareGradConfig: GradConfig = {\n  kernelName: Square,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, mul(cast(x, 'float32'), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SquaredDifference} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const squaredDifferenceGradConfig: GradConfig = {\n  kernelName: SquaredDifference,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const two = scalar(2);\n    const derA = () => mul(dy, mul(two, sub(a, b)));\n    const derB = () => mul(dy, mul(two, sub(b, a)));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Step} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const stepGradConfig: GradConfig = {\n  kernelName: Step,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports\n    // it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Sub} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const subGradConfig: GradConfig = {\n  kernelName: Sub,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(neg(res), b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sum, SumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {ones} from '../ops/ones';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const sumGradConfig: GradConfig = {\n  kernelName: Sum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const expandedDyShape = x.shape.slice();\n    const {axis} = attrs as {} as SumAttrs;\n\n    const axes = parseAxisParam(axis, x.shape);\n    axes.forEach(axis => {\n      expandedDyShape[axis] = 1;\n    });\n    const expandedDy = reshape(dy, expandedDyShape);\n    const derX = mul(expandedDy, ones(x.shape, 'float32'));\n\n    return {x: () => derX};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cos} from '../ops/cos';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const tanGradConfig: GradConfig = {\n  kernelName: Tan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, square(cos(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const tanhGradConfig: GradConfig = {\n  kernelName: Tanh,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(sub(scalar(1), square(y)), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tile, TileAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {slice} from '../ops/slice';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const tileGradConfig: GradConfig = {\n  kernelName: Tile,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {reps} = attrs as unknown as TileAttrs;\n\n    const derX = () => {\n      let xGrad = zerosLike(x);\n      // TODO(cais): Maybe reduce memory footprint by avoiding repeated\n      // slicing.\n      if (x.rank === 1) {\n        for (let i = 0; i < reps[0]; ++i) {\n          xGrad = add(xGrad, slice(dy, [i * x.shape[0]], [x.shape[0]]));\n        }\n      } else if (x.rank === 2) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            xGrad = add(xGrad, slice(dy, [i * x.shape[0], j * x.shape[1]], [\n                          x.shape[0], x.shape[1]\n                        ]));\n          }\n        }\n      } else if (x.rank === 3) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              xGrad =\n                  add(xGrad,\n                      slice(\n                          dy, [i * x.shape[0], j * x.shape[1], k * x.shape[2]],\n                          [x.shape[0], x.shape[1], x.shape[2]]));\n            }\n          }\n        }\n      } else if (x.rank === 4) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              for (let l = 0; l < reps[3]; ++l) {\n                xGrad =\n                    add(xGrad,\n                        slice(\n                            dy,\n                            [\n                              i * x.shape[0], j * x.shape[1], k * x.shape[2],\n                              l * x.shape[3]\n                            ],\n                            [x.shape[0], x.shape[1], x.shape[2], x.shape[3]]));\n              }\n            }\n          }\n        }\n      } else {\n        throw new Error(\n            `Gradient for tile operation is not implemented for rank-` +\n            `${x.rank} tensors yet.`);\n      }\n      return xGrad;\n    };\n    return {x: derX};\n  },\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Transpose, TransposeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const transposeGradConfig: GradConfig = {\n  kernelName: Transpose,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const transposeAttrs: TransposeAttrs = attrs as {} as TransposeAttrs;\n    const {perm} = transposeAttrs;\n    const undoPerm = axis_util.getUndoAxesPermutation(perm);\n    return {x: () => transpose(dy, undoPerm)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Unpack, UnpackAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {stack} from '../ops/stack';\nimport {Tensor} from '../tensor';\n\nexport const unpackGradConfig: GradConfig = {\n  kernelName: Unpack,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const unpackAttrs: UnpackAttrs = attrs as {} as UnpackAttrs;\n    const {axis} = unpackAttrs;\n    return {value: () => stack(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {UnsortedSegmentSum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {expandDims} from '../ops/expand_dims';\nimport {gather} from '../ops/gather';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {maximum} from '../ops/maximum';\nimport {ones} from '../ops/ones';\nimport {scalar} from '../ops/scalar';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor, Tensor1D} from '../tensor';\n\nexport const unsortedSegmentSumGradConfig: GradConfig = {\n  kernelName: UnsortedSegmentSum,\n  inputsToSave: ['segmentIds'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [segmentIds] = saved;\n\n    const derX = () => {\n      return gatherDropNegatives(dy, segmentIds as Tensor1D);\n    };\n    return {x: derX};\n  }\n};\n\nfunction gatherDropNegatives<T extends Tensor>(x: T, indices: Tensor1D) {\n  // Helper function for unsorted segment ops. Gathers params for\n  // positive segment ids and gathers 0 for inputs with negative segment id.\n  // Mirrors _GatherDropNegatives from tensorflow/python/ops/math_grad.py\n  const zeroClippedIndices = maximum(indices, zerosLike(indices));\n  const gathered = gather(x, zeroClippedIndices as Tensor1D);\n  let isPositive = greaterEqual(indices, scalar(0, 'int32'));\n  const numIters = gathered.rank - isPositive.rank;\n  for (let i = 0; i < numIters; ++i) {\n    isPositive = expandDims(isPositive, i + 1);\n  }\n  isPositive = logicalAnd(isPositive, ones(gathered.shape, 'bool'));\n  const zeroSlice = zerosLike(gathered);\n  return where(isPositive, gathered, zeroSlice);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ZerosLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const zerosLikeGradConfig: GradConfig = {\n  kernelName: ZerosLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {absGradConfig} from './gradients/Abs_grad';\nimport {acosGradConfig} from './gradients/Acos_grad';\nimport {acoshGradConfig} from './gradients/Acosh_grad';\nimport {addGradConfig} from './gradients/Add_grad';\nimport {addNGradConfig} from './gradients/AddN_grad';\nimport {argMaxGradConfig} from './gradients/ArgMax_grad';\nimport {argMinGradConfig} from './gradients/ArgMin_grad';\nimport {asinGradConfig} from './gradients/Asin_grad';\nimport {asinhGradConfig} from './gradients/Asinh_grad';\nimport {atan2GradConfig} from './gradients/Atan2_grad';\nimport {atanGradConfig} from './gradients/Atan_grad';\nimport {atanhGradConfig} from './gradients/Atanh_grad';\nimport {avgPool3DGradConfig} from './gradients/AvgPool3D_grad';\nimport {avgPoolGradConfig} from './gradients/AvgPool_grad';\nimport {batchMatMulGradConfig} from './gradients/BatchMatMul_grad';\nimport {batchToSpaceNDGradConfig} from './gradients/BatchToSpaceND_grad';\nimport {broadcastToGradConfig} from './gradients/BroadcastTo_grad';\nimport {castGradConfig} from './gradients/Cast_grad';\nimport {ceilGradConfig} from './gradients/Ceil_grad';\nimport {clipByValueGradConfig} from './gradients/ClipByValue_grad';\nimport {concatGradConfig} from './gradients/Concat_grad';\nimport {conv2DGradConfig} from './gradients/Conv2D_grad';\nimport {conv2DBackpropInputGradConfig} from './gradients/Conv2DBackpropInput_grad';\nimport {conv3DGradConfig} from './gradients/Conv3D_grad';\nimport {cosGradConfig} from './gradients/Cos_grad';\nimport {coshGradConfig} from './gradients/Cosh_grad';\nimport {cumsumGradConfig} from './gradients/Cumsum_grad';\nimport {depthwiseConv2dNativeGradConfig} from './gradients/DepthwiseConv2dNative_grad';\nimport {dilation2dGradConfig} from './gradients/Dilation2D_grad';\nimport {divGradConfig} from './gradients/Div_grad';\nimport {eluGradConfig} from './gradients/Elu_grad';\nimport {erfGradConfig} from './gradients/Erf_grad';\nimport {expGradConfig} from './gradients/Exp_grad';\nimport {expm1GradConfig} from './gradients/Expm1_grad';\nimport {floorGradConfig} from './gradients/Floor_grad';\nimport {floorDivGradConfig} from './gradients/FloorDiv_grad';\nimport {fusedBatchNormGradConfig} from './gradients/FusedBatchNorm_grad';\nimport {gatherGradConfig} from './gradients/GatherV2_grad';\nimport {greaterEqualGradConfig} from './gradients/GreaterEqual_grad';\nimport {identityGradConfig} from './gradients/Identity_grad';\nimport {isFiniteGradConfig} from './gradients/IsFinite_grad';\nimport {isInfGradConfig} from './gradients/IsInf_grad';\nimport {isNanGradConfig} from './gradients/IsNan_grad';\nimport {log1pGradConfig} from './gradients/Log1p_grad';\nimport {logGradConfig} from './gradients/Log_grad';\nimport {logSoftmaxGradConfig} from './gradients/LogSoftmax_grad';\nimport {lrnGradConfig} from './gradients/LRN_grad';\nimport {maxGradConfig} from './gradients/Max_grad';\nimport {maximumGradConfig} from './gradients/Maximum_grad';\nimport {maxPool3DGradConfig} from './gradients/MaxPool3D_grad';\nimport {maxPoolGradConfig} from './gradients/MaxPool_grad';\nimport {minGradConfig} from './gradients/Min_grad';\nimport {minimumGradConfig} from './gradients/Minimum_grad';\nimport {modGradConfig} from './gradients/Mod_grad';\nimport {multiplyGradConfig} from './gradients/Multiply_grad';\nimport {negateGradConfig} from './gradients/Negate_grad';\nimport {oneHotGradConfig} from './gradients/OneHot_grad';\nimport {onesLikeGradConfig} from './gradients/OnesLike_grad';\nimport {padV2GradConfig} from './gradients/PadV2_grad';\nimport {powGradConfig} from './gradients/Pow_grad';\nimport {preluGradConfig} from './gradients/Prelu_grad';\nimport {reciprocalGradConfig} from './gradients/Reciprocal_grad';\nimport {relu6GradConfig} from './gradients/Relu6_grad';\nimport {reluGradConfig} from './gradients/Relu_grad';\nimport {reshapeGradConfig} from './gradients/Reshape_grad';\nimport {resizeBilinearGradConfig} from './gradients/ResizeBilinear_grad';\nimport {resizeNearestNeighborGradConfig} from './gradients/ResizeNearestNeighbor_grad';\nimport {reverseGradConfig} from './gradients/Reverse_grad';\nimport {roundGradConfig} from './gradients/Round_grad';\nimport {rsqrtGradConfig} from './gradients/Rsqrt_grad';\nimport {selectV2PoolGradConfig} from './gradients/SelectV2_grad';\nimport {seluGradConfig} from './gradients/Selu_grad';\nimport {sigmoidGradConfig} from './gradients/Sigmoid_grad';\nimport {signGradConfig} from './gradients/Sign_grad';\nimport {sinGradConfig} from './gradients/Sin_grad';\nimport {sinhGradConfig} from './gradients/Sinh_grad';\nimport {sliceGradConfig} from './gradients/Slice_grad';\nimport {softmaxGradConfig} from './gradients/Softmax_grad';\nimport {softplusGradConfig} from './gradients/Softplus_grad';\nimport {spaceToBatchNDGradConfig} from './gradients/SpaceToBatchND_grad';\nimport {splitVGradConfig} from './gradients/SplitV_grad';\nimport {sqrtGradConfig} from './gradients/Sqrt_grad';\nimport {squareGradConfig} from './gradients/Square_grad';\nimport {squaredDifferenceGradConfig} from './gradients/SquaredDifference_grad';\nimport {stepGradConfig} from './gradients/Step_grad';\nimport {subGradConfig} from './gradients/Sub_grad';\nimport {sumGradConfig} from './gradients/Sum_grad';\nimport {tanGradConfig} from './gradients/Tan_grad';\nimport {tanhGradConfig} from './gradients/Tanh_grad';\nimport {tileGradConfig} from './gradients/Tile_grad';\nimport {transposeGradConfig} from './gradients/Transpose_grad';\nimport {unpackGradConfig} from './gradients/Unpack_grad';\nimport {unsortedSegmentSumGradConfig} from './gradients/UnsortedSegmentSum_grad';\nimport {zerosLikeGradConfig} from './gradients/ZerosLike_grad';\nimport {GradConfig} from './kernel_registry';\nimport {registerGradient} from './kernel_registry';\n\n// Export all kernel configs here so that the package can auto register them\nconst gradConfigs: GradConfig[] = [\n  absGradConfig,\n  acosGradConfig,\n  acoshGradConfig,\n  addGradConfig,\n  addNGradConfig,\n  argMaxGradConfig,\n  argMinGradConfig,\n  asinGradConfig,\n  asinhGradConfig,\n  atan2GradConfig,\n  atanGradConfig,\n  atanhGradConfig,\n  avgPool3DGradConfig,\n  avgPoolGradConfig,\n  batchMatMulGradConfig,\n  batchToSpaceNDGradConfig,\n  broadcastToGradConfig,\n  castGradConfig,\n  ceilGradConfig,\n  clipByValueGradConfig,\n  concatGradConfig,\n  conv2DBackpropInputGradConfig,\n  conv2DGradConfig,\n  conv3DGradConfig,\n  cosGradConfig,\n  coshGradConfig,\n  cumsumGradConfig,\n  depthwiseConv2dNativeGradConfig,\n  dilation2dGradConfig,\n  divGradConfig,\n  eluGradConfig,\n  erfGradConfig,\n  expGradConfig,\n  expm1GradConfig,\n  floorDivGradConfig,\n  floorGradConfig,\n  fusedBatchNormGradConfig,\n  gatherGradConfig,\n  greaterEqualGradConfig,\n  identityGradConfig,\n  isFiniteGradConfig,\n  isInfGradConfig,\n  isNanGradConfig,\n  log1pGradConfig,\n  logGradConfig,\n  logSoftmaxGradConfig,\n  lrnGradConfig,\n  maxGradConfig,\n  maxGradConfig,\n  maximumGradConfig,\n  maxPool3DGradConfig,\n  maxPoolGradConfig,\n  minGradConfig,\n  minimumGradConfig,\n  modGradConfig,\n  multiplyGradConfig,\n  negateGradConfig,\n  oneHotGradConfig,\n  onesLikeGradConfig,\n  padV2GradConfig,\n  padV2GradConfig,\n  powGradConfig,\n  preluGradConfig,\n  reciprocalGradConfig,\n  relu6GradConfig,\n  reluGradConfig,\n  reshapeGradConfig,\n  resizeBilinearGradConfig,\n  resizeNearestNeighborGradConfig,\n  reverseGradConfig,\n  roundGradConfig,\n  rsqrtGradConfig,\n  selectV2PoolGradConfig,\n  seluGradConfig,\n  sigmoidGradConfig,\n  signGradConfig,\n  sinGradConfig,\n  sinhGradConfig,\n  sliceGradConfig,\n  softmaxGradConfig,\n  softplusGradConfig,\n  spaceToBatchNDGradConfig,\n  spaceToBatchNDGradConfig,\n  splitVGradConfig,\n  splitVGradConfig,\n  sqrtGradConfig,\n  squaredDifferenceGradConfig,\n  squareGradConfig,\n  stepGradConfig,\n  subGradConfig,\n  sumGradConfig,\n  tanGradConfig,\n  tanhGradConfig,\n  tileGradConfig,\n  transposeGradConfig,\n  unpackGradConfig,\n  unsortedSegmentSumGradConfig,\n  zerosLikeGradConfig\n];\n\nfor (const gradientConfig of gradConfigs) {\n  registerGradient(gradientConfig);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {abs} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    abs<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.abs = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return abs(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acos = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acosh = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {addStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    addStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.addStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return addStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {add} from '../../ops/add';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    add<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.add = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return add(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {all} from '../../ops/all';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    all<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.all = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return all(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {any} from '../../ops/any';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    any<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.any = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return any(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMax} from '../../ops/arg_max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMax<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMax = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return argMax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMin} from '../../ops/arg_min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMin<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMin = function<T extends Tensor>(axis: number): T {\n  this.throwIfDisposed();\n  return argMin(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Scalar, Tensor} from '../../tensor';\nimport {Rank} from '../../types';\nimport {assert} from '../../util';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asScalar<T extends Tensor>(): Scalar;\n  }\n}\n\n/** Converts a size-1 `tf.Tensor` to a `tf.Scalar`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asScalar = function<T extends Tensor>(this: T): Scalar {\n  this.throwIfDisposed();\n  assert(this.size === 1, () => 'The array must have only 1 element.');\n  return reshape(this, []);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asType<T extends Tensor>(this: T, dtype: DataType): T;\n  }\n}\n\n/**\n * Casts a `tf.Tensor` to a specified dtype.\n *\n * @param dtype Data-type to cast the tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asType = function<T extends Tensor>(\n    this: T, dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast<T>(this, dtype);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as1D<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Converts a `tf.Tensor` to a `tf.Tensor1D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as1D = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as2D<T extends Tensor>(rows: number, columns: number): Tensor2D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor2D`.\n *\n * @param rows Number of rows in `tf.Tensor2D`.\n * @param columns Number of columns in `tf.Tensor2D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as2D = function<T extends Tensor>(\n    rows: number, columns: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as3D<T extends Tensor>(rows: number, columns: number, depth: number):\n        Tensor3D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor3D`.\n *\n * @param rows Number of rows in `tf.Tensor3D`.\n * @param columns Number of columns in `tf.Tensor3D`.\n * @param depth Depth of `tf.Tensor3D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as3D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as4D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number): Tensor4D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor4D`.\n *\n * @param rows Number of rows in `tf.Tensor4D`.\n * @param columns Number of columns in `tf.Tensor4D`.\n * @param depth Depth of `tf.Tensor4D`.\n * @param depth2 4th dimension of `tf.Tensor4D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as4D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as5D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number,\n        depth3: number): Tensor5D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor5D`.\n *\n * @param rows Number of rows in `tf.Tensor5D`.\n * @param columns Number of columns in `tf.Tensor5D`.\n * @param depth Depth of `tf.Tensor5D`.\n * @param depth2 4th dimension of `tf.Tensor5D`.\n * @param depth3 5th dimension of 'tf.Tensor5D'\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as5D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number,\n    depth3: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2, depth3]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {atan2} from '../../ops/atan2';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan2<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.atan2 = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return atan2(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {avgPool} from '../../ops/avg_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    avgPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.avgPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return avgPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchToSpaceND} from '../../ops/batch_to_space_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchToSpaceND<R extends Rank>(blockShape: number[], crops: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.batchToSpaceND = function<R extends Rank>(\n    blockShape: number[], crops: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return batchToSpaceND(this, blockShape, crops);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchNorm} from '../../ops/batchnorm';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchNorm<T extends Tensor>(\n        mean: Tensor<R>|Tensor1D|TensorLike,\n        variance: Tensor<R>|Tensor1D|TensorLike,\n        offset?: Tensor<R>|Tensor1D|TensorLike,\n        scale?: Tensor<R>|Tensor1D|TensorLike,\n        varianceEpsilon?: number): Tensor<R>;\n  }\n}\n\nTensor.prototype.batchNorm = function<R extends Rank>(\n    mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  this.throwIfDisposed();\n  return batchNorm(this, mean, variance, offset, scale, varianceEpsilon);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {broadcastTo} from '../../ops/broadcast_to';\nimport {Tensor} from '../../tensor';\nimport {Rank, ShapeMap} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    broadcastTo<R extends Rank>(shape: ShapeMap[R]): Tensor<R>;\n  }\n}\n\nTensor.prototype.broadcastTo = function<R extends Rank>(shape: ShapeMap[R]):\n    Tensor<R> {\n  this.throwIfDisposed();\n  return broadcastTo(this, shape);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cast<T extends Tensor>(dtype: DataType): T;\n  }\n}\n\nTensor.prototype.cast = function<T extends Tensor>(dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast(this, dtype) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ceil} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ceil<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.ceil = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return ceil(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {clipByValue} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    clipByValue<T extends Tensor>(min: number, max: number): Tensor;\n  }\n}\n\nTensor.prototype.clipByValue = function<T extends Tensor>(\n    min: number, max: number): T {\n  this.throwIfDisposed();\n  return clipByValue(this, min, max) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {concat} from '../../ops/concat';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    concat<T extends Tensor>(tensors: T|Array<T|TensorLike>, axis?: number): T;\n  }\n}\n\nTensor.prototype.concat = function<T extends Tensor>(\n    x: T|Array<T|TensorLike>, axis?: number): T {\n  this.throwIfDisposed();\n  if (x instanceof Tensor) {\n    x = [x];\n  }\n  return concat([this, ...x], axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv1d} from '../../ops/conv1d';\nimport {ExplicitPadding} from '../../ops/conv_util';\nimport {Tensor, Tensor2D, Tensor3D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv1d<T extends Tensor2D|Tensor3D>(\n        filter: Tensor3D|TensorLike3D, stride: number,\n        pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n        dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv1d = function<T extends Tensor2D|Tensor3D>(\n    filter: Tensor3D|TensorLike3D, stride: number,\n    pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n    dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv1d(\n             this, filter, stride, pad, dataFormat, dilation,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2dTranspose} from '../../ops/conv2d_transpose';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2dTranspose<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D,\n        outputShape: [number, number, number, number]|[number, number, number],\n        strides: [number, number]|number, pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2dTranspose = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2dTranspose(\n             this, filter, outputShape, strides, pad, dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2d} from '../../ops/conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cos = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cosh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {cumsum} from '../../ops/cumsum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cumsum<R extends Rank>(\n        axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R>;\n  }\n}\n\nTensor.prototype.cumsum = function<R extends Rank>(\n    axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R> {\n  this.throwIfDisposed();\n  return cumsum(this, axis, exclusive, reverse);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthToSpace} from '../../ops/depth_to_space';\nimport {Tensor, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthToSpace<T extends Tensor4D>(\n        blockSize: number, dataFormat: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.depthToSpace = function<T extends Tensor4D>(\n    blockSize: number, dataFormat: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return depthToSpace(this, blockSize, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../../globals';\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2D<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\n/**\n * @deprecated Use `depthwiseConv2d` instead.\n */\nTensor.prototype.depthwiseConv2D = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  deprecationWarn('depthwiseConv2D is deprecated, use depthwiseConv2d instead');\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.depthwiseConv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dilation2d} from '../../ops/dilation2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dilation2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n        pad: 'valid'|'same', dilations?: [number, number]|number,\n        dataFormat?: 'NHWC'): T;\n  }\n}\n\nTensor.prototype.dilation2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n    pad: 'valid'|'same', dilations?: [number, number]|number,\n    dataFormat?: 'NHWC'): T {\n  this.throwIfDisposed();\n  return dilation2d(this, filter, strides, pad, dilations, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {divNoNan} from '../../ops/div_no_nan';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divNoNan<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.divNoNan = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return divNoNan(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {divStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.divStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return divStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {div} from '../../ops/div';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    div<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.div = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return div(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dot} from '../../ops/dot';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dot<T extends Tensor>(b: Tensor|TensorLike): Tensor;\n  }\n}\n\nTensor.prototype.dot = function<T extends Tensor>(b: T|TensorLike): Tensor {\n  this.throwIfDisposed();\n  return dot(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {elu} from '../../ops/elu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    elu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.elu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return elu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {equalStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equalStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.equalStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return equalStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {equal} from '../../ops/equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equal<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.equal = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return equal(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {erf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    erf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.erf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return erf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {exp} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    exp<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.exp = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return exp(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {expandDims} from '../../ops/expand_dims';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expandDims<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.expandDims = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return expandDims(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {expm1} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expm1<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.expm1 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return expm1(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {fft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    fft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.fft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return fft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    flatten<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Flatten a Tensor to a 1D array.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.flatten = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {floor} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floor<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.floor = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return floor(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {floorDiv} from '../../ops/floorDiv';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floorDiv<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.floorDiv = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return floorDiv(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {gather} from '../../ops/gather';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    gather<T extends Tensor>(\n        this: T, indices: Tensor|TensorLike, axis?: number): T;\n  }\n}\n\nTensor.prototype.gather = function<T extends Tensor>(\n    this: T, indices: Tensor|TensorLike, axis?: number): T {\n  this.throwIfDisposed();\n  return gather(this, indices, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return greaterEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greaterEqual} from '../../ops/greater_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greaterEqual = function<T extends Tensor>(b: Tensor|\n                                                           TensorLike): T {\n  this.throwIfDisposed();\n  return greaterEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return greaterStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greater} from '../../ops/greater';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greater<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greater = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return greater(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ifft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ifft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.ifft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return ifft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {irfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    irfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.irfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return irfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isFinite} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isFinite<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isFinite = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isFinite(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isInf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isInf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isInf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isInf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isNaN} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isNaN<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isNaN = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isNaN(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {leakyRelu} from '../../ops/leaky_relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    leakyRelu<T extends Tensor>(alpha: number): T;\n  }\n}\n\nTensor.prototype.leakyRelu = function<T extends Tensor>(\n    this: T, alpha: number): T {\n  this.throwIfDisposed();\n  return leakyRelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.lessEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {lessEqual} from '../../ops/less_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessEqual = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return lessEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {less} from '../../ops/less';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    less<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.less = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return less(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {localResponseNormalization} from '../../ops/local_response_normalization';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    localResponseNormalization<T extends Tensor>(\n        depthRadius?: number, bias?: number, alpha?: number, beta?: number): T;\n  }\n}\n\nTensor.prototype.localResponseNormalization = function<T extends Tensor>(\n    depthRadius?: number, bias?: number, alpha?: number, beta?: number): T {\n  this.throwIfDisposed();\n  return localResponseNormalization(this, depthRadius, bias, alpha, beta) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.logSigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return logSigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSoftmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSoftmax<T extends Tensor>(this: T, axis?: number): T;\n  }\n}\n\nTensor.prototype.logSoftmax = function<T extends Tensor>(\n    this: T, axis?: number): T {\n  this.throwIfDisposed();\n  return logSoftmax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logSumExp} from '../../ops/log_sum_exp';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSumExp<T extends Tensor>(\n        this: T, axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.logSumExp = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return logSumExp(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log1p} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log1p<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log1p = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log1p(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalAnd} from '../../ops/logical_and';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalAnd<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalAnd = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalAnd(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalNot} from '../../ops/logical_not';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalNot<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.logicalNot = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return logicalNot(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalOr} from '../../ops/logical_or';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalOr<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalOr = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return logicalOr(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalXor} from '../../ops/logical_xor';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalXor<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalXor = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalXor(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {matMul} from '../../ops/mat_mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    matMul<T extends Tensor>(\n        b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T;\n  }\n}\n\nTensor.prototype.matMul = function<T extends Tensor>(\n    this: T, b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T {\n  this.throwIfDisposed();\n  return matMul(this, b, transposeA, transposeB);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maxPool} from '../../ops/max_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maxPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.maxPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return maxPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {max} from '../../ops/max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    max<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.max = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return max(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {maximumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.maximumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return maximumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maximum} from '../../ops/maximum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.maximum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return maximum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mean} from '../../ops/mean';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mean<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.mean = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return mean(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {min} from '../../ops/min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    min<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.min = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return min(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {minimumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.minimumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return minimumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {minimum} from '../../ops/minimum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.minimum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return minimum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {modStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    modStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.modStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return modStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mod} from '../../ops/mod';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mod<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mod = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mod(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {mulStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mulStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.mulStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return mulStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mul} from '../../ops/mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mul<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mul = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mul(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {neg} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    neg<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.neg = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return neg(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {norm} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    norm<T extends Tensor>(\n        ord?: number|'euclidean'|'fro', axis?: number|number[],\n        keepDims?: boolean): Tensor;\n  }\n}\n\nTensor.prototype.norm = function<T extends Tensor>(\n    ord?: number|'euclidean'|'fro', axis?: number|number[],\n    keepDims?: boolean) {\n  this.throwIfDisposed();\n  return norm(this, ord, axis, keepDims) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {notEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.notEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return notEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {notEqual} from '../../ops/not_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.notEqual = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return notEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {oneHot} from '../../ops/one_hot';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    oneHot(depth: number, onValue: number, offValue: number): Tensor;\n  }\n}\n\nTensor.prototype.oneHot = function(\n    depth: number, onValue = 1, offValue = 0): Tensor {\n  this.throwIfDisposed();\n  return oneHot(this, depth, onValue, offValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {onesLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    onesLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.onesLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return onesLike(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {pad} from '../../ops/pad';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pad<T extends Tensor>(\n        paddings: Array<[number, number]>, constantValue?: number): T;\n  }\n}\n\nTensor.prototype.pad = function<T extends Tensor>(\n    this: T, paddings: Array<[number, number]>, constantValue: number): T {\n  this.throwIfDisposed();\n  return pad(this, paddings, constantValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pool} from '../../ops/pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pool<T extends Tensor3D|Tensor4D>(\n        windowShape: [number, number]|number, poolingType: 'avg'|'max',\n        padding: 'valid'|'same'|number, diationRate?: [number, number]|number,\n        strides?: [number, number]|number): T;\n  }\n}\n\nTensor.prototype.pool = function<T extends Tensor3D|Tensor4D>(\n    this: T, windowShape: [number, number]|number, poolingType: 'max'|'avg',\n    padding: 'valid'|'same'|number, dilationRate?: [number, number]|number,\n    strides?: [number, number]|number): T {\n  this.throwIfDisposed();\n  return pool(this, windowShape, poolingType, padding, dilationRate, strides);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {powStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    powStrict<T extends Tensor>(exp: Tensor): Tensor;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.powStrict = function<T extends Tensor>(exp: Tensor): T {\n  this.throwIfDisposed();\n  return powStrict(this, exp) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pow} from '../../ops/pow';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pow<T extends Tensor>(exp: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.pow = function<T extends Tensor>(exp: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return pow(this, exp);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prelu} from '../../ops/prelu';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prelu<T extends Tensor>(alpha: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.prelu = function<T extends Tensor>(\n    this: T, alpha: T|TensorLike): T {\n  this.throwIfDisposed();\n  return prelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prod} from '../../ops/prod';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prod<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.prod = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return prod(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {reciprocal} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reciprocal<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.reciprocal = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return reciprocal(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu} from '../../ops/relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu6} from '../../ops/relu6';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu6<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu6 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu6(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshapeAs<T extends Tensor>(x: T): T;\n  }\n}\n\n/**\n * Reshapes the tensor into the shape of the provided tensor.\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.reshapeAs = function<T extends Tensor>(x: T): T {\n  this.throwIfDisposed();\n  return reshape(this, x.shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshape<T extends Tensor>(shape: number[]): T;\n  }\n}\n\nTensor.prototype.reshape = function<T extends Tensor>(shape: number[]): T {\n  this.throwIfDisposed();\n  return reshape(this, shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeBilinear} from '../../ops/image/resize_bilinear';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeBilinear<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeBilinear = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeBilinear(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeNearestNeighbor} from '../../ops/image/resize_nearest_neighbor';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeNearestNeighbor<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeNearestNeighbor = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeNearestNeighbor(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reverse} from '../../ops/reverse';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reverse<T extends Tensor>(this: T, axis?: number|number[]): T;\n  }\n}\n\nTensor.prototype.reverse = function<T extends Tensor>(\n    this: T, axis?: number|number[]): T {\n  this.throwIfDisposed();\n  return reverse(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.rfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return rfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {round} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    round<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.round = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return round(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rsqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rsqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.rsqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return rsqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {selu} from '../../ops/selu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    selu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.selu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return selu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {separableConv2d} from '../../ops/separable_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    separableConv2d<T extends Tensor3D|Tensor4D>(\n        depthwiseFilter: Tensor4D|TensorLike4D,\n        pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n        pad: 'valid'|'same', dilation?: [number, number]|number,\n        dataFormat?: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.separableConv2d = function<T extends Tensor3D|Tensor4D>(\n    depthwiseFilter: Tensor4D|TensorLike4D,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation?: [number, number]|number,\n    dataFormat?: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return separableConv2d(\n             this, depthwiseFilter, pointwiseFilter, strides, pad, dilation,\n             dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sign} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sign<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sign = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sign(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {slice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    slice<T extends Tensor>(\n        this: T, begin: number|number[], size?: number|number[]): T;\n  }\n}\n\nTensor.prototype.slice = function<T extends Tensor>(\n    this: T, begin: number|number[], size?: number|number[]): T {\n  this.throwIfDisposed();\n  return slice(this, begin, size);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softmax<T extends Tensor>(this: T, dim?: number): T;\n  }\n}\n\nTensor.prototype.softmax = function<T extends Tensor>(this: T, dim: number): T {\n  this.throwIfDisposed();\n  return softmax(this, dim);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softplus} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softplus<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.softplus = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return softplus(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {spaceToBatchND} from '../../ops/space_to_batch_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    spaceToBatchND<R extends Rank>(blockShape: number[], paddings: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.spaceToBatchND = function<R extends Rank>(\n    blockShape: number[], paddings: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return spaceToBatchND(this, blockShape, paddings);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {split} from '../../ops/split';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    split<T extends Tensor>(numOrSizeSplits: number[]|number, axis?: number):\n        T[];\n  }\n}\n\nTensor.prototype.split = function<T extends Tensor>(\n    numOrSizeSplits: number[]|number, axis?: number): T[] {\n  this.throwIfDisposed();\n  return split(this, numOrSizeSplits, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {square} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    square<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.square = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return square(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {squaredDifference} from '../../ops/squared_difference';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifference<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.squaredDifference = function<T extends Tensor>(b: Tensor|\n                                                                TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifference(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {squaredDifferenceStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifferenceStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.squaredDifferenceStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifferenceStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {squeeze} from '../../ops/squeeze';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squeeze<T extends Tensor>(axis?: number[]): T;\n  }\n}\n\nTensor.prototype.squeeze = function<T extends Tensor>(axis?: number[]): T {\n  this.throwIfDisposed();\n  return squeeze(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {stack} from '../../ops/stack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stack<T extends Tensor>(x: Tensor|Tensor[], axis?: number): T;\n  }\n}\n\nTensor.prototype.stack = function<T extends Tensor>(\n    x: Tensor|Tensor[], axis?: number): T {\n  this.throwIfDisposed();\n  const tensorsToBeStacked = x instanceof Tensor ? [this, x] : [this, ...x];\n  return stack(tensorsToBeStacked, axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {step} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    step<T extends Tensor>(this: T, alpha?: number): T;\n  }\n}\n\nTensor.prototype.step = function<T extends Tensor>(this: T, alpha?: number) {\n  this.throwIfDisposed();\n  return step(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {stridedSlice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stridedSlice<T extends Tensor>(\n        this: Tensor, begin: number[], end: number[], strides: number[],\n        beginMask?: number, endMask?: number, ellipsisMask?: number,\n        newAxisMask?: number, shrinkAxisMask?: number): Tensor;\n  }\n}\n\nTensor.prototype.stridedSlice = function<T extends Tensor>(\n    this: Tensor, begin: number[], end: number[], strides: number[],\n    beginMask?: number, endMask?: number, ellipsisMask?: number,\n    newAxisMask?: number, shrinkAxisMask?: number): T {\n  this.throwIfDisposed();\n  return stridedSlice(\n             this, begin, end, strides, beginMask, endMask, ellipsisMask,\n             newAxisMask, shrinkAxisMask) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {subStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    subStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.subStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return subStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sub} from '../../ops/sub';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sub<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.sub = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return sub(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sum} from '../../ops/sum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sum<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.sum = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return sum(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tile} from '../../ops/tile';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tile<T extends Tensor>(b: number[]): T;\n  }\n}\n\nTensor.prototype.tile = function<T extends Tensor>(reps: number[]): T {\n  this.throwIfDisposed();\n  return tile(this, reps) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toBool<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `bool`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toBool = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'bool');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toFloat<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `float32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toFloat = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'float32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toInt<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `int32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toInt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'int32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {topk} from '../../ops/topk';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    topk<T extends Tensor>(this: T, k?: number, sorted?: boolean):\n        {values: T, indices: T};\n  }\n}\n\nTensor.prototype.topk = function<T extends Tensor>(\n    this: T, k?: number, sorted?: boolean): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return topk(this, k, sorted) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {transpose} from '../../ops/transpose';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    transpose<T extends Tensor>(perm?: number[]): T;\n  }\n}\n\nTensor.prototype.transpose = function<T extends Tensor>(\n    this: T, perm?: number[]): T {\n  this.throwIfDisposed();\n  return transpose(this, perm);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unique} from '../../ops/unique';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unique<T extends Tensor>(this: T, axis?: number): {values: T, indices: T};\n  }\n}\n\nTensor.prototype.unique = function<T extends Tensor>(\n    this: T, axis?: number): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return unique(this, axis) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unsortedSegmentSum} from '../../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike1D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unsortedSegmentSum<T extends Tensor>(\n        this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T;\n  }\n}\n\nTensor.prototype.unsortedSegmentSum = function<T extends Tensor>(\n    this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T {\n  this.throwIfDisposed();\n  return unsortedSegmentSum(this, segmentIds, numSegments);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {unstack} from '../../ops/unstack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unstack<T extends Tensor>(axis?: number): T[];\n  }\n}\n\nTensor.prototype.unstack = function<T extends Tensor>(axis?: number): T[] {\n  this.throwIfDisposed();\n  return unstack(this, axis) as T[];\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {where} from '../../ops/where';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    where<T extends Tensor>(condition: Tensor|TensorLike, x: Tensor|TensorLike):\n        T;\n  }\n}\n\nTensor.prototype.where = function<T extends Tensor>(\n    condition: Tensor|TensorLike, x: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return where(condition, this, x) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {zerosLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    zerosLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.zerosLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return zerosLike(this);\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {backend} from '@tensorflow/tfjs-core';\nimport {DataFormat} from '../keras_format/common';\n\nlet _epsilon: number;\n\n/**\n * Returns the value of the fuzz factor used in numeric expressions.\n */\nexport function epsilon() {\n  if (_epsilon == null) {\n    _epsilon = backend().epsilon();\n  }\n  return _epsilon;\n}\n\n/**\n * Sets the value of the fuzz factor used in numeric expressions.\n * @param e New value of epsilon.\n */\nexport function setEpsilon(e: number) {\n  _epsilon = e;\n}\n\n/**\n * Returns the default image data format convention.\n */\nexport function imageDataFormat(): DataFormat {\n  return 'channelsLast';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Explicit error types.\n *\n * See the following link for more information about why the code includes\n * calls to setPrototypeOf:\n *\n * https://github.com/Microsoft/TypeScript-wiki/blob/master/Breaking-Changes.md#extending-built-ins-like-error-array-and-map-may-no-longer-work\n */\n// tslint:enable\n\n/**\n * Equivalent of Python's AttributeError.\n */\nexport class AttributeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AttributeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's RuntimeError.\n */\nexport class RuntimeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, RuntimeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's ValueError.\n */\nexport class ValueError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, ValueError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's NotImplementedError.\n */\nexport class NotImplementedError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, NotImplementedError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's AssertionError.\n */\nexport class AssertionError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AssertionError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's IndexError.\n */\nexport class IndexError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, IndexError.prototype);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {DataType, fused, serialization, util} from '@tensorflow/tfjs-core';\n\nimport {AssertionError, ValueError} from '../errors';\n\n// tslint:enable\n\n/**\n * If `value` is an Array, equivalent to Python's `value * numValues`.\n * If `value` is not an Array, equivalent to Python's `[value] * numValues`\n */\n// tslint:disable-next-line:no-any\nexport function pyListRepeat(value: any, numValues: number): any[] {\n  if (Array.isArray(value)) {\n    // tslint:disable-next-line:no-any\n    let newArray: any[] = [];\n    for (let i = 0; i < numValues; i++) {\n      newArray = newArray.concat(value);\n    }\n    return newArray;\n  } else {\n    const newArray = new Array(numValues);\n    newArray.fill(value);\n    return newArray;\n  }\n}\n\nexport function assert(val: boolean, message?: string): void {\n  if (!val) {\n    throw new AssertionError(message);\n  }\n}\n\n/**\n * Count the number of elements of the `array` that are equal to `reference`.\n */\nexport function count<T>(array: T[], refernce: T) {\n  let counter = 0;\n  for (const item of array) {\n    if (item === refernce) {\n      counter++;\n    }\n  }\n  return counter;\n}\n\n/**\n * If an array is of length 1, just return the first element. Otherwise, return\n * the full array.\n * @param tensors\n */\nexport function singletonOrArray<T>(xs: T[]): T|T[] {\n  if (xs.length === 1) {\n    return xs[0];\n  }\n  return xs;\n}\n\n/**\n * Normalizes a list/tensor into a list.\n *\n * If a tensor is passed, we return\n * a list of size 1 containing the tensor.\n *\n * @param x target object to be normalized.\n */\n// tslint:disable-next-line:no-any\nexport function toList(x: any): any[] {\n  if (Array.isArray(x)) {\n    return x;\n  }\n  return [x];\n}\n\n/**\n * Generate a UID for a list\n */\n// tslint:disable-next-line:no-any\nexport function objectListUid(objs: any|any[]): string {\n  const objectList = toList(objs);\n  let retVal = '';\n  for (const obj of objectList) {\n    if (obj.id == null) {\n      throw new ValueError(\n          `Object ${obj} passed to objectListUid without an id`);\n    }\n    if (retVal !== '') {\n      retVal = retVal + ', ';\n    }\n    retVal = `${retVal}${Math.abs(obj.id)}`;\n  }\n  return retVal;\n}\n/**\n * Converts string to snake-case.\n * @param name\n */\nexport function toSnakeCase(name: string): string {\n  const intermediate = name.replace(/(.)([A-Z][a-z0-9]+)/g, '$1_$2');\n  const insecure =\n      intermediate.replace(/([a-z])([A-Z])/g, '$1_$2').toLowerCase();\n  /*\n   If the class is private the name starts with \"_\" which is not secure\n   for creating scopes. We prefix the name with \"private\" in this case.\n   */\n  if (insecure[0] !== '_') {\n    return insecure;\n  }\n  return 'private' + insecure;\n}\n\nexport function toCamelCase(identifier: string): string {\n  // quick return for empty string or single character strings\n  if (identifier.length <= 1) {\n    return identifier;\n  }\n  // Check for the underscore indicating snake_case\n  if (identifier.indexOf('_') === -1) {\n    return identifier;\n  }\n  return identifier.replace(/[_]+(\\w|$)/g, (m, p1) => p1.toUpperCase());\n}\n\n// tslint:disable-next-line:no-any\nlet _GLOBAL_CUSTOM_OBJECTS = {} as {[objName: string]: any};\n\nexport function serializeKerasObject(instance: serialization.Serializable):\n    serialization.ConfigDictValue {\n  if (instance === null || instance === undefined) {\n    return null;\n  }\n  const dict: serialization.ConfigDictValue = {};\n  dict['className'] = instance.getClassName();\n  dict['config'] = instance.getConfig();\n  return dict;\n}\n\n/**\n * Replace ndarray-style scalar objects in serialization objects with numbers.\n *\n * Background: In some versions of tf.keras, certain scalar values in the HDF5\n * model save file can be serialized as: `{'type': 'ndarray', 'value': num}`,\n * where in `num` is a plain number. This method converts such serialization\n * to a `number`.\n *\n * @param config The keras-format serialization object to be processed\n *   (in place).\n */\nfunction convertNDArrayScalarsInConfig(config: serialization.ConfigDictValue):\n    void {\n  if (config == null || typeof config !== 'object') {\n    return;\n  } else if (Array.isArray(config)) {\n    config.forEach(configItem => convertNDArrayScalarsInConfig(configItem));\n  } else {\n    const fields = Object.keys(config);\n    for (const field of fields) {\n      const value = config[field];\n      if (value != null && typeof value === 'object') {\n        if (!Array.isArray(value) && value['type'] === 'ndarray' &&\n            typeof value['value'] === 'number') {\n          config[field] = value['value'];\n        } else {\n          convertNDArrayScalarsInConfig(value as serialization.ConfigDict);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Deserialize a saved Keras Object\n * @param identifier either a string ID or a saved Keras dictionary\n * @param moduleObjects a list of Python class names to object constructors\n * @param customObjects a list of Python class names to object constructors\n * @param printableModuleName debug text for the object being reconstituted\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns a TensorFlow.js Layers object\n */\n// tslint:disable:no-any\nexport function deserializeKerasObject(\n    identifier: string|serialization.ConfigDict,\n    moduleObjects = {} as {[objName: string]: any},\n    customObjects = {} as {[objName: string]: any},\n    printableModuleName = 'object', fastWeightInit = false): any {\n  // tslint:enable\n  if (typeof identifier === 'string') {\n    const functionName = identifier;\n    let fn;\n    if (functionName in customObjects) {\n      fn = customObjects[functionName];\n    } else if (functionName in _GLOBAL_CUSTOM_OBJECTS) {\n      fn = _GLOBAL_CUSTOM_OBJECTS[functionName];\n    } else {\n      fn = moduleObjects[functionName];\n      if (fn == null) {\n        throw new ValueError(\n            `Unknown ${printableModuleName}: ${identifier}. ` +\n            `This may be due to one of the following reasons:\\n` +\n            `1. The ${printableModuleName} is defined in Python, in which ` +\n            `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n            `code.\\n` +\n            `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n            `but is not registered properly with ` +\n            `tf.serialization.registerClass().`);\n        // TODO(cais): Add link to tutorial page on custom layers.\n      }\n    }\n    return fn;\n  } else {\n    // In this case we are dealing with a Keras config dictionary.\n    const config = identifier;\n    if (config['className'] == null || config['config'] == null) {\n      throw new ValueError(\n          `${printableModuleName}: Improper config format: ` +\n          `${JSON.stringify(config)}.\\n` +\n          `'className' and 'config' must set.`);\n    }\n    const className = config['className'] as string;\n    let cls, fromConfig;\n    if (className in customObjects) {\n      [cls, fromConfig] = customObjects[className];\n    } else if (className in _GLOBAL_CUSTOM_OBJECTS) {\n      [cls, fromConfig] = _GLOBAL_CUSTOM_OBJECTS['className'];\n    } else if (className in moduleObjects) {\n      [cls, fromConfig] = moduleObjects[className];\n    }\n    if (cls == null) {\n      throw new ValueError(\n          `Unknown ${printableModuleName}: ${className}. ` +\n          `This may be due to one of the following reasons:\\n` +\n          `1. The ${printableModuleName} is defined in Python, in which ` +\n          `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n          `code.\\n` +\n          `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n          `but is not registered properly with ` +\n          `tf.serialization.registerClass().`);\n      // TODO(cais): Add link to tutorial page on custom layers.\n    }\n    if (fromConfig != null) {\n      // Porting notes: Instead of checking to see whether fromConfig accepts\n      // customObjects, we create a customObjects dictionary and tack it on to\n      // config['config'] as config['config'].customObjects. Objects can use it,\n      // if they want.\n\n      // tslint:disable-next-line:no-any\n      const customObjectsCombined = {} as {[objName: string]: any};\n      for (const key of Object.keys(_GLOBAL_CUSTOM_OBJECTS)) {\n        customObjectsCombined[key] = _GLOBAL_CUSTOM_OBJECTS[key];\n      }\n      for (const key of Object.keys(customObjects)) {\n        customObjectsCombined[key] = customObjects[key];\n      }\n      // Add the customObjects to config\n      const nestedConfig = config['config'] as serialization.ConfigDict;\n      nestedConfig['customObjects'] = customObjectsCombined;\n\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      convertNDArrayScalarsInConfig(config['config']);\n      const returnObj =\n          fromConfig(cls, config['config'], customObjects, fastWeightInit);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n\n      return returnObj;\n    } else {\n      // Then `cls` may be a function returning a class.\n      // In this case by convention `config` holds\n      // the kwargs of the function.\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      // In python this is **config['config'], for tfjs-layers we require\n      // classes that use this fall-through construction method to take\n      // a config interface that mimics the expansion of named parameters.\n      const returnObj = new cls(config['config']);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n      return returnObj;\n    }\n  }\n}\n\n/**\n * Compares two numbers for sorting.\n * @param a\n * @param b\n */\nexport function numberCompare(a: number, b: number) {\n  return (a < b) ? -1 : ((a > b) ? 1 : 0);\n}\n\n/**\n * Comparison of two numbers for reverse sorting.\n * @param a\n * @param b\n */\nexport function reverseNumberCompare(a: number, b: number) {\n  return -1 * numberCompare(a, b);\n}\n\n/**\n * Convert a string into the corresponding DType.\n * @param dtype\n * @returns An instance of DType.\n */\nexport function stringToDType(dtype: string): DataType {\n  switch (dtype) {\n    case 'float32':\n      return 'float32';\n    default:\n      throw new ValueError(`Invalid dtype: ${dtype}`);\n  }\n}\n\n/**\n * Test the element-by-element equality of two Arrays of strings.\n * @param xs First array of strings.\n * @param ys Second array of strings.\n * @returns Wether the two arrays are all equal, element by element.\n */\nexport function stringsEqual(xs: string[], ys: string[]): boolean {\n  if (xs == null || ys == null) {\n    return xs === ys;\n  }\n  if (xs.length !== ys.length) {\n    return false;\n  }\n  for (let i = 0; i < xs.length; ++i) {\n    if (xs[i] !== ys[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Get the unique elements of an array.\n * @param xs Array.\n * @returns An Array consisting of the unique elements in `xs`.\n */\nexport function unique<T>(xs: T[]): T[] {\n  if (xs == null) {\n    return xs;\n  }\n  const out: T[] = [];\n  // TODO(cais): Maybe improve performance by sorting.\n  for (const x of xs) {\n    if (out.indexOf(x) === -1) {\n      out.push(x);\n    }\n  }\n  return out;\n}\n\n/**\n * Determine if an Object is empty (i.e., does not have own properties).\n * @param obj Object\n * @returns Whether the Object is empty.\n * @throws ValueError: If object is `null` or `undefined`.\n */\nexport function isObjectEmpty(obj: {}): boolean {\n  if (obj == null) {\n    throw new ValueError(`Invalid value in obj: ${JSON.stringify(obj)}`);\n  }\n  for (const key in obj) {\n    if (obj.hasOwnProperty(key)) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Helper function used to build type union/enum run-time checkers.\n * @param values The list of allowed values.\n * @param label A string name for the type\n * @param value The value to test.\n * @throws ValueError: If the value is not in values nor `undefined`/`null`.\n */\nexport function checkStringTypeUnionValue(\n    values: string[], label: string, value: string): void {\n  if (value == null) {\n    return;\n  }\n  if (values.indexOf(value) < 0) {\n    throw new ValueError(`${value} is not a valid ${label}.  Valid values are ${\n        values} or null/undefined.`);\n  }\n}\n\n/**\n * Helper function for verifying the types of inputs.\n *\n * Ensures that the elements of `x` are all of type `expectedType`.\n * Also verifies that the length of `x` is within bounds.\n *\n * @param x Object to test.\n * @param expectedType The string expected type of all of the elements in the\n * Array.\n * @param minLength Return false if x.length is less than this.\n * @param maxLength Return false if x.length is greater than this.\n * @returns true if and only if `x` is an `Array<expectedType>` with\n * length >= `minLength` and <= `maxLength`.\n */\n// tslint:disable:no-any\nexport function checkArrayTypeAndLength(\n    x: any, expectedType: string, minLength = 0,\n    maxLength = Infinity): boolean {\n  assert(minLength >= 0);\n  assert(maxLength >= minLength);\n  return (\n      Array.isArray(x) && x.length >= minLength && x.length <= maxLength &&\n      x.every(e => typeof e === expectedType));\n}\n// tslint:enable:no-any\n\n/**\n * Assert that a value or an array of value are positive integer.\n *\n * @param value The value being asserted on. May be a single number or an array\n *   of numbers.\n * @param name Name of the value, used to make the error message.\n */\nexport function assertPositiveInteger(value: number|number[], name: string) {\n  if (Array.isArray(value)) {\n    util.assert(\n        value.length > 0, () => `${name} is unexpectedly an empty array.`);\n    value.forEach(\n        (v, i) => assertPositiveInteger(v, `element ${i + 1} of ${name}`));\n  } else {\n    util.assert(\n        Number.isInteger(value) && value > 0,\n        () => `Expected ${name} to be a positive integer, but got ` +\n            `${formatAsFriendlyString(value)}.`);\n  }\n}\n\n/**\n * Format a value into a display-friendly, human-readable fashion.\n *\n * - `null` is formatted as `'null'`\n * - Strings are formated with flanking pair of quotes.\n * - Arrays are formatted with flanking pair of square brackets.\n *\n * @param value The value to display.\n * @return Formatted string.\n */\n// tslint:disable-next-line:no-any\nexport function formatAsFriendlyString(value: any): string {\n  if (value === null) {\n    return 'null';\n  } else if (Array.isArray(value)) {\n    return '[' + value.map(v => formatAsFriendlyString(v)).join(',') + ']';\n  } else if (typeof value === 'string') {\n    return `\"${value}\"`;\n  } else {\n    return `${value}`;\n  }\n}\n\n/**\n * Returns a function `f2` (decorator) which wraps the original function\n * `f`. `f2` guarantees that `f` can be called at most once\n * every `waitMs` ms. If `f2` is called more often, it will return\n * the last returned result of `f`.\n *\n * @param f The original function `f` to wrap.\n * @param waitMs The time between two consecutive calls to `f` in ms.\n */\nexport function debounce<T>(\n    f: (...args: Array<{}>) => T, waitMs: number): (...args: Array<{}>) => T {\n  let lastTime = util.now();\n  let lastResult: T;\n  const f2 = (...args: Array<{}>) => {\n    const now = util.now();\n    if (now - lastTime < waitMs) {\n      return lastResult;\n    }\n    lastTime = now;\n    lastResult = f(...args);\n    return lastResult;\n  };\n  return f2;\n}\n\n/**\n * Returns the fusable activation given a layers identifier.\n *\n * @param activationName The layers identifier string.\n * @return The name of the fusable activation.\n */\nexport function mapActivationToFusedKernel(activationName: string):\n    fused.Activation {\n  if (activationName === 'relu') {\n    return 'relu';\n  }\n  if (activationName === 'linear') {\n    return 'linear';\n  }\n  if (activationName === 'elu') {\n    return 'elu';\n  }\n  return null;\n}\n\ntype PossibleValues = Array<Array<boolean|string|number>>;\n\n/**\n * Returns the cartesian product of sets of values.\n * This works the same as itertools.product in Python.\n *\n * Example:\n *\n * filters = [128, 256, 512]\n * paddings = ['same', 'valid']\n *\n * product = [ [128, 'same'], [128, 'valid'], [256, 'same'], [256, 'valid'],\n * [512, 'same'], [512, 'valid']]\n *\n * @param arrayOfValues List/array of values.\n * @return The cartesian product.\n */\nexport function getCartesianProductOfValues(...arrayOfValues: PossibleValues):\n    PossibleValues {\n  assert(arrayOfValues.length > 0, 'arrayOfValues is empty');\n\n  for (const values of arrayOfValues) {\n    assert(Array.isArray(values), 'one of the values is not an array');\n    assert(values.length > 0, 'one of the values is empty');\n  }\n\n  return arrayOfValues.reduce((products, values) => {\n    if (products.length === 0) {\n      return values.map(value => [value]);\n    }\n\n    return values\n        .map(value => {\n          return products.map((prevValue) => [...prevValue, value]);\n        })\n        .reduce((flattenedProduct, unflattenedProduct) => {\n          return flattenedProduct.concat(unflattenedProduct);\n        }, []);\n  }, [] as PossibleValues);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/contraints.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport {epsilon} from './backend/common';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\n/**\n * Helper function used by many of the Constraints to find the L2Norms.\n */\nfunction calcL2Norms(w: Tensor, axis: number): Tensor {\n  return tidy(() => tfc.sqrt(tfc.sum(tfc.mul(w, w), axis, true)));\n}\n\n/**\n * Base class for functions that impose constraints on weight values\n *\n * @doc {\n *   heading: 'Constraints',\n *   subheading: 'Classes',\n *   namespace: 'constraints'\n * }\n */\nexport abstract class Constraint extends serialization.Serializable {\n  /* Porting note: was __call__, apply chosen to match other similar choices */\n  abstract apply(w: Tensor): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport interface MaxNormArgs {\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   *\n   *  For instance, in a `Dense` layer the weight matrix\n   *  has shape `[inputDim, outputDim]`,\n   *  set `axis` to `0` to constrain each weight vector\n   *  of length `[inputDim,]`.\n   *  In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   *  the weight tensor has shape\n   *  `[rows, cols, inputDepth, outputDepth]`,\n   *  set `axis` to `[0, 1, 2]`\n   *  to constrain the weights of each filter tensor of size\n   *  `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class MaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MaxNorm';\n  private maxValue: number;\n  private axis: number;\n  private readonly defaultMaxValue = 2;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MaxNormArgs) {\n    super();\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.clipByValue(norms, 0, this.maxValue);\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {maxValue: this.maxValue, axis: this.axis};\n  }\n}\nserialization.registerClass(MaxNorm);\n\nexport interface UnitNormArgs {\n  /**\n   * Axis along which to calculate norms.\n   *\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * [rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class UnitNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'UnitNorm';\n  private axis: number;\n  private readonly defaultAxis = 0;\n  constructor(args: UnitNormArgs) {\n    super();\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(\n        () => tfc.div(w, tfc.add(epsilon(), calcL2Norms(w, this.axis))));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {axis: this.axis};\n  }\n}\nserialization.registerClass(UnitNorm);\n\nexport class NonNeg extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'NonNeg';\n\n  apply(w: Tensor): Tensor {\n    return tfc.relu(w);\n  }\n}\nserialization.registerClass(NonNeg);\n\nexport interface MinMaxNormArgs {\n  /**\n   * Minimum norm for incoming weights\n   */\n  minValue?: number;\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * `[rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n  /**\n   * Rate for enforcing the constraint: weights will be rescaled to yield:\n   * `(1 - rate) * norm + rate * norm.clip(minValue, maxValue)`.\n   * Effectively, this means that rate=1.0 stands for strict\n   * enforcement of the constraint, while rate<1.0 means that\n   * weights will be rescaled at each step to slowly move\n   * towards a value inside the desired interval.\n   */\n  rate?: number;\n}\n\nexport class MinMaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MinMaxNorm';\n  private minValue: number;\n  private maxValue: number;\n  private rate: number;\n  private axis: number;\n  private readonly defaultMinValue = 0.0;\n  private readonly defaultMaxValue = 1.0;\n  private readonly defaultRate = 1.0;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MinMaxNormArgs) {\n    super();\n    this.minValue =\n        args.minValue != null ? args.minValue : this.defaultMinValue;\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.rate = args.rate != null ? args.rate : this.defaultRate;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.add(\n          tfc.mul(\n              this.rate, tfc.clipByValue(norms, this.minValue, this.maxValue)),\n          tfc.mul(1.0 - this.rate, norms));\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      minValue: this.minValue,\n      maxValue: this.maxValue,\n      rate: this.rate,\n      axis: this.axis\n    };\n  }\n}\nserialization.registerClass(MinMaxNorm);\n\n/** @docinline */\nexport type ConstraintIdentifier =\n    'maxNorm'|'minMaxNorm'|'nonNeg'|'unitNorm'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in ConstraintIdentifier]: string} = {\n      'maxNorm': 'MaxNorm',\n      'minMaxNorm': 'MinMaxNorm',\n      'nonNeg': 'NonNeg',\n      'unitNorm': 'UnitNorm'\n    };\n\nexport function serializeConstraint(constraint: Constraint):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeConstraint(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Constraint {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'constraint');\n}\n\nexport function getConstraint(identifier: ConstraintIdentifier|\n                              serialization.ConfigDict|Constraint): Constraint {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeConstraint(config);\n  } else if (identifier instanceof Constraint) {\n    return identifier;\n  } else {\n    return deserializeConstraint(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constraint, MaxNorm, MaxNormArgs, MinMaxNorm, MinMaxNormArgs, NonNeg, UnitNorm, UnitNormArgs} from './constraints';\n\n/**\n * MaxNorm weight constraint.\n *\n * Constrains the weights incident to each hidden unit\n * to have a norm less than or equal to a desired value.\n *\n * References\n *       - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting\n * Srivastava, Hinton, et al.\n * 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Constraints',namespace: 'constraints'}\n */\nexport function maxNorm(args: MaxNormArgs): Constraint {\n  return new MaxNorm(args);\n}\n\n/**\n * Constrains the weights incident to each hidden unit to have unit norm.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function unitNorm(args: UnitNormArgs): Constraint {\n  return new UnitNorm(args);\n}\n\n/**\n * Constains the weight to be non-negative.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function nonNeg(): Constraint {\n  return new NonNeg();\n}\n\n/** @doc {heading: 'Constraints', namespace: 'constraints'} */\nexport function minMaxNorm(config: MinMaxNormArgs): Constraint {\n  return new MinMaxNorm(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// TODO(huan): add layer-specific input shape types (see: https://github.com/tensorflow/tfjs-layers/pull/492)\n/** @docalias (null | number)[] */\nexport type Shape = Array<null | number>;\n\n// The tfjs-core version of DataType must stay synced with this.\nexport type DataType = 'float32'|'int32'|'bool'|'complex64'|'string';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type DataFormat = 'channelsFirst'|'channelsLast';\nexport const VALID_DATA_FORMAT_VALUES = ['channelsFirst', 'channelsLast'];\n\n// These constants have a snake vs. camel distinction.\nexport type DataFormatSerialization = 'channels_first'|'channels_last';\n\n/** @docinline */\nexport type PaddingMode = 'valid'|'same'|'causal';\nexport const VALID_PADDING_MODE_VALUES = ['valid', 'same', 'causal'];\n\n/** @docinline */\nexport type PoolMode = 'max'|'avg';\nexport const VALID_POOL_MODE_VALUES = ['max', 'avg'];\n\n/** @docinline */\nexport type BidirectionalMergeMode = 'sum'|'mul'|'concat'|'ave';\nexport const VALID_BIDIRECTIONAL_MERGE_MODES = ['sum', 'mul', 'concat', 'ave'];\n\n/** @docinline */\nexport type SampleWeightMode = 'temporal';\nexport const VALID_SAMPLE_WEIGHT_MODES = ['temporal'];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Common functions for TensorFlow.js Layers.\n */\nimport {VALID_DATA_FORMAT_VALUES, VALID_PADDING_MODE_VALUES, VALID_POOL_MODE_VALUES} from './keras_format/common';\nimport {checkStringTypeUnionValue} from './utils/generic_utils';\n\n// A map from the requested scoped name of a Tensor to the number of Tensors\n// wanting that name so far.  This allows enforcing name uniqueness by appending\n// an incrementing index, e.g. scope/name, scope/name_1, scope/name_2, etc.\nconst nameMap: Map<string, number> = new Map<string, number>();\n\nexport function checkDataFormat(value?: string): void {\n  checkStringTypeUnionValue(VALID_DATA_FORMAT_VALUES, 'DataFormat', value);\n}\n\nexport function checkPaddingMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_PADDING_MODE_VALUES, 'PaddingMode', value);\n}\n\nexport function checkPoolMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_POOL_MODE_VALUES, 'PoolMode', value);\n}\n\nconst _nameScopeStack: string[] = [];\nconst _nameScopeDivider = '/';\n\n/**\n * Enter namescope, which can be nested.\n */\nexport function nameScope<T>(name: string, fn: () => T): T {\n  _nameScopeStack.push(name);\n  try {\n    const val: T = fn();\n    _nameScopeStack.pop();\n    return val;\n  } catch (e) {\n    _nameScopeStack.pop();\n    throw e;\n  }\n}\n\n/**\n * Get the current namescope as a flat, concatenated string.\n */\nfunction currentNameScopePrefix(): string {\n  if (_nameScopeStack.length === 0) {\n    return '';\n  } else {\n    return _nameScopeStack.join(_nameScopeDivider) + _nameScopeDivider;\n  }\n}\n\n/**\n * Get the name a Tensor (or Variable) would have if not uniqueified.\n * @param tensorName\n * @return Scoped name string.\n */\nexport function getScopedTensorName(tensorName: string): string {\n  if (!isValidTensorName(tensorName)) {\n    throw new Error('Not a valid tensor name: \\'' + tensorName + '\\'');\n  }\n  return currentNameScopePrefix() + tensorName;\n}\n\n/**\n * Get unique names for Tensors and Variables.\n * @param scopedName The fully-qualified name of the Tensor, i.e. as produced by\n *  `getScopedTensorName()`.\n * @return A unique version of the given fully scoped name.\n *   If this is the first time that the scoped name is seen in this session,\n *   then the given `scopedName` is returned unaltered.  If the same name is\n *   seen again (producing a collision), an incrementing suffix is added to the\n *   end of the name, so it takes the form 'scope/name_1', 'scope/name_2', etc.\n */\nexport function getUniqueTensorName(scopedName: string): string {\n  if (!isValidTensorName(scopedName)) {\n    throw new Error('Not a valid tensor name: \\'' + scopedName + '\\'');\n  }\n  if (!nameMap.has(scopedName)) {\n    nameMap.set(scopedName, 0);\n  }\n  const index = nameMap.get(scopedName);\n  nameMap.set(scopedName, nameMap.get(scopedName) + 1);\n\n  if (index > 0) {\n    const result = `${scopedName}_${index}`;\n    // Mark the composed name as used in case someone wants\n    // to call getUniqueTensorName(\"name_1\").\n    nameMap.set(result, 1);\n    return result;\n  } else {\n    return scopedName;\n  }\n}\n\nconst tensorNameRegex = new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\\._\\/]*$/);\n\n/**\n * Determine whether a string is a valid tensor name.\n * @param name\n * @returns A Boolean indicating whether `name` is a valid tensor name.\n */\nexport function isValidTensorName(name: string): boolean {\n  return !!name.match(tensorNameRegex);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Math utility functions.\n *\n * This file contains some frequently used math function that operates on\n * number[] or Float32Array and return a number. Many of these functions are\n * not-so-thick wrappers around TF.js Core functions. But they offer the\n * convenience of\n * 1) not having to convert the inputs into Tensors,\n * 2) not having to convert the returned Tensors to numbers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar, Tensor1D, tensor1d} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\n\nexport type ArrayTypes = Uint8Array|Int32Array|Float32Array;\n\n/**\n * Determine if a number is an integer.\n */\nexport function isInteger(x: number): boolean {\n  return x === parseInt(x.toString(), 10);\n}\n\n/**\n * Calculate the product of an array of numbers.\n * @param array The array to calculate the product over.\n * @param begin Beginning index, inclusive.\n * @param end Ending index, exclusive.\n * @return The product.\n */\nexport function arrayProd(\n    array: number[]|ArrayTypes, begin?: number, end?: number): number {\n  if (begin == null) {\n    begin = 0;\n  }\n  if (end == null) {\n    end = array.length;\n  }\n\n  let prod = 1;\n  for (let i = begin; i < end; ++i) {\n    prod *= array[i];\n  }\n  return prod;\n}\n\n/**\n * A helper function transforms the two input types to an instance of Tensor1D,\n * so the return value can be fed directly into various TF.js Core functions.\n * @param array\n */\nfunction toArray1D(array: number[]|Float32Array): Tensor1D {\n  array = Array.isArray(array) ? new Float32Array(array) : array;\n  return tensor1d(array);\n}\n\n/**\n * Compute minimum value.\n * @param array\n * @return minimum value.\n */\nexport function min(array: number[]|Float32Array): number {\n  return tfc.min(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute maximum value.\n * @param array\n * @return maximum value\n */\nexport function max(array: number[]|Float32Array): number {\n  return tfc.max(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute sum of array.\n * @param array\n * @return The sum.\n */\nexport function sum(array: number[]|Float32Array): number {\n  return tfc.sum(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute mean of array.\n * @param array\n * @return The mean.\n */\nexport function mean(array: number[]|Float32Array): number {\n  return sum(array) / array.length;\n}\n\n/**\n * Compute variance of array.\n * @param array\n * @return The variance.\n */\nexport function variance(array: number[]|Float32Array): number {\n  const demeaned = tfc.sub(toArray1D(array), scalar(mean(array)));\n  const sumSquare = tfc.sum(tfc.mul(demeaned, demeaned)).dataSync()[0];\n  return sumSquare / array.length;\n}\n\n/**\n * Compute median of array.\n * @param array\n * @return The median value.\n */\nexport function median(array: number[]|Float32Array): number {\n  const arraySorted = array.slice().sort((a, b) => a - b);\n  const lowIdx = Math.floor((arraySorted.length - 1) / 2);\n  const highIdx = Math.ceil((arraySorted.length - 1) / 2);\n  if (lowIdx === highIdx) {\n    return arraySorted[lowIdx];\n  }\n  return (arraySorted[lowIdx] + arraySorted[highIdx]) / 2;\n}\n\n/**\n * Generate an array of integers in [begin, end).\n * @param begin Beginning integer, inclusive.\n * @param end Ending integer, exclusive.\n * @returns Range array.\n * @throws ValueError, iff `end` < `begin`.\n */\nexport function range(begin: number, end: number): number[] {\n  if (end < begin) {\n    throw new ValueError(`end (${end}) < begin (${begin}) is forbidden.`);\n  }\n  const out: number[] = [];\n  for (let i = begin; i < end; ++i) {\n    out.push(i);\n  }\n  return out;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * deeplearn.js backend.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {onesLike as coreOnesLike, scalar, Tensor, Tensor1D, tensor1d, Tensor2D, Tensor3D, Tensor4D, Tensor5D, tidy, where, zerosLike as coreZerosLike} from '@tensorflow/tfjs-core';\nimport {checkDataFormat} from '../common';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {HasShape} from '../types';\nimport * as math_utils from '../utils/math_utils';\n\nimport {imageDataFormat} from './common';\n\n// tslint:enable\n\n/* Setting and getting backend from deeplearn.js. */\n\n// Default deeplearn.js backend is WebGL (GPU).\nlet backend: 'cpu'|'webgl' = 'webgl';\n\nexport function setBackend(requestedBackend: 'cpu'|'webgl') {\n  tfc.setBackend(requestedBackend);\n  backend = requestedBackend;\n}\n\nexport function getBackend(): 'cpu'|'webgl' {\n  return backend;\n}\n\n/**\n * Indicates whether the backend is operating symbolically.\n *\n * This function will be used to determine how to interpret user code. If\n * it returns true, calls to the backend construct a symbolic graph; if\n * it returns false, calls to the backend execute immediately.\n */\nexport function isBackendSymbolic(): boolean {\n  return false;\n}\n\n/**\n * Get the number of elements in a Tensor.\n * @param x The Tensor.\n * @return Number of elements in `x`.\n */\nexport function countParams(x: HasShape): number {\n  const shape = x.shape;\n  if (shape.length > 0) {\n    return shape.reduce((a: number, b: number) => a * b);\n  } else {\n    // Scalar.\n    return 1;\n  }\n}\n\n/**\n * Casts a tensor to a different dtype and returns it.\n * @param x Input tensor.\n * @param dtype String: 'float32'|'int32'|'bool'.\n * @returns Tensor of the specified `dtype`.\n */\nexport function cast(x: Tensor, dtype: tfc.DataType): Tensor {\n  return x.asType(dtype);\n}\n\n/**\n * Adds a 1-sized dimension at index \"axis\".\n * @param x Input tensor.\n * @param axis Position where to add the new axis.\n * @returns Result of the dimension expansion.\n */\nexport function expandDims(x: Tensor, axis = -1): Tensor {\n  const outShape = x.shape.slice();\n  if (axis < 0) {\n    axis = outShape.length + axis + 1;\n  }\n  outShape.splice(axis, 0, 1);\n  return x.reshape(outShape);\n}\n\n/**\n * Repeats a 2D tensor.\n *\n * If `x` has shape `[samples, dim]` and `n` is 2, for example, the output\n * will have shape `[samples, 2, dim]`.\n *\n * @param x Input tensor.\n * @param n Integer, number of times to repeat.\n * @returns The result of the repeat operation.\n * @throws ValueError: If input tensor is not 2D.\n */\nexport function repeat(x: Tensor, n: number): Tensor {\n  return tidy(() => {\n    if (x.shape.length !== 2) {\n      throw new ValueError(\n          `repeat() expects a rank-2 tensor, but received a ` +\n          `rank-${x.shape.length} tensor.`);\n    }\n    const y = expandDims(x, 1);\n    return tile(y, [1, n, 1]);\n  });\n}\n\n/**\n * Flatten a Tensor into 1D.\n * @param x Input tensor.\n * @return The result of the flattening `x`.\n */\nexport function flatten(x: Tensor): Tensor {\n  const newShape = [math_utils.arrayProd(x.shape)];\n  return x.reshape(newShape);\n}\n\n/**\n * Turn a nD tensor into a 2D tensor with same 0th dimension.\n * In other words, it flattens each data samples of a batch.\n *\n * @param x The tensor to flatten. The rank of this tensor is required to be 2\n *   or higher.\n * @return The result of the flattening.\n */\nexport function batchFlatten(x: Tensor): Tensor {\n  if (x.rank <= 1) {\n    throw new ValueError(\n        `batchFlatten requires a minimum rank of 2. Got rank: ${x.rank}.`);\n  }\n  const newShape = [x.shape[0], math_utils.arrayProd(x.shape, 1)];\n  return x.reshape(newShape);\n}\n\n/**\n * Do slicing along the first axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the first axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongFirstAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [start, 0], [size, array.shape[1]]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [start, 0, 0],\n            [size, array.shape[1], array.shape[2]]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [start, 0, 0, 0],\n            [size, array.shape[1], array.shape[2], array.shape[3]]);\n      case 5:\n        return tfc.slice(array as Tensor5D, [start, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4]\n        ]);\n      case 6:\n        return tfc.slice(array, [start, 0, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4],\n          array.shape[5]\n        ]);\n      default:\n        throw new ValueError(\n            `sliceAlongFirstAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the last axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the last axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongLastAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [0, start], [array.shape[0], size]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [0, 0, start],\n            [array.shape[0], array.shape[1], size]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [0, 0, 0, start],\n            [array.shape[0], array.shape[1], array.shape[2], size]);\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the sepcified axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size of the slice along the chosen axis.\n * @param choose an axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongAxis(\n    array: Tensor, start: number, size: number, axis: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 3:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice3d(\n                array as Tensor3D, [0, start, 0],\n                [array.shape[0], size, array.shape[2]]);\n          case 3:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 4:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice4d(\n                array as Tensor4D, [0, start, 0, 0],\n                [array.shape[0], size, array.shape[2], array.shape[3]]);\n          case 3:\n            return tfc.slice4d(\n                array as Tensor4D, [0, 0, start, 0],\n                [array.shape[0], array.shape[1], size, array.shape[3]]);\n          case 4:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Concatenates a list of tensors alongside the specified axis.\n * @param tensors `Array` of tensors to concatenate.\n * @param axis Concatenation axis.\n * @returns The result of the concatenation.\n */\nexport function concatenate(tensors: Tensor[], axis = -1): Tensor {\n  let rank: number;\n  if (axis < 0) {\n    rank = tensors[0].rank;\n    if (rank !== 0) {\n      axis = rank;\n    } else {\n      axis = 0;\n    }\n  }\n  if (axis === tensors[0].rank) {\n    // Porting Note: This is necessary because tfc.concat() requires axis to be\n    //   in the interval [-rank, rank).\n    axis = -1;\n  }\n  // Porting Note: Sparse concat is not supported yet.\n  return tfc.concat(tensors, axis);\n}\n\n/**\n * Concatenate two arrays along the first dimension.\n * @param a The 1st `tf.Tensor` to concatenate.\n * @param b The 2nd `tf.Tensor` to concatenate.\n * @returns Result of the concatenation.\n * @throws ValueError: If `a` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function concatAlongFirstAxis(a: Tensor, b: Tensor): Tensor {\n  switch (a.rank) {\n    case 1:\n      return tfc.concat1d([a as Tensor1D, b as Tensor1D]);\n    case 2:\n      return tfc.concat2d([a as Tensor2D, b as Tensor2D], 0);\n    case 3:\n      return tfc.concat3d([a as Tensor3D, b as Tensor3D], 0);\n    case 4:\n      return tfc.concat4d([a as Tensor4D, b as Tensor4D], 0);\n    default:\n      throw new ValueError(\n          `concatAlongFirstAxis() received an unsupported ` +\n          `tensor rank: ${a.rank}`);\n  }\n}\n\n/**\n * Creates a tensor by tiling `x` by `n`.\n * @param x A tensor.\n * @param n An Array of integers or a single integer. If an Array, the length\n *   must be the same as the number of dimensions in `x`. If a single integer,\n *   it will be treated as an Array of length 1.\n */\nexport function tile(x: Tensor, n: number|number[]): Tensor {\n  if (!Array.isArray(n)) {\n    n = [n];\n  }\n  if (x.rank !== n.length) {\n    throw new ValueError(\n        `The length of input n (${n.length}) does not match ` +\n        `the number of dimensions in input x (${x.rank})`);\n  }\n  return tfc.tile(x, n);\n}\n\n/* Creation of random tensors. */\n\n/**\n * Get a tensor with normal distribution of values.\n *\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @return The normal tensor.\n */\nexport function randomNormal(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: 'float32'|'int32',\n    seed?: number): Tensor {\n  return tfc.randomNormal(shape, mean, stddev, dtype, seed);\n}\n\n/* Linear Algebra */\n\n/**\n * Multiply two tensors and returns the result as a tensor.\n *\n * For 2D tensors, this is equivalent to matrix multiplication (matMul).\n * For tensors of higher ranks, it follows the Theano behavior,\n * (e.g. `(2, 3) * (4, 3, 5) -> (2, 4, 5)`).  From the Theano documentation:\n *\n * For N dimensions it is a sum product over the last axis of x and the\n * second-to-last of y:\n *\n * @param a A tensor of at least rank 2.\n * @param b A tensor of at least rank 2.\n * @param activation (optional) A string identifying the activation\n *   function.\n * @return Result of the dot operation.\n */\nexport function dot(\n    a: Tensor, b: Tensor, activation?: tfc.fused.Activation,\n    bias?: Tensor): Tensor {\n  if ((a.rank < 2) || (b.rank < 2)) {\n    throw new NotImplementedError(\n        `dot requires both inputs to be rank >= 2` +\n        ` but got x shape = ${a.shape} and y shape = ${b.shape}`);\n  }\n  if (b.rank >= 3) {\n    const xLastDim = a.shape.slice(-1)[0];\n    const ySecondLastDim = b.shape.slice(-2)[0];\n    if (xLastDim !== ySecondLastDim) {\n      throw new NotImplementedError(\n          `If rank y >= 3, then the second last dim` +\n          ` of y must equal the last dim of x but got x shape = ${\n              a.shape} and ` +\n          ` y shape = ${b.shape}`);\n    }\n  }\n  // Handle basic 2D x 2D case.\n  if ((a.rank === 2) && (b.rank === 2)) {\n    const transposeA = false;\n    const transposeB = false;\n    // tfc.fused.matMul only fuses certain activation functions. Unsupported\n    // activation functions are treated as 'linear' activations, which is\n    // equivalent to a no-op.\n    return tfc.fused.matMul({\n      a,\n      b: b as Tensor2D,\n      transposeA,\n      transposeB,\n      bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n      activation\n    });\n  } else {\n    // Reshape x into the analogous 2D Tensor.\n    const aFirstDims = a.shape.slice();  // Holds all but the last dim of x.\n    const aLastDim = aFirstDims.pop();\n    a = a.reshape([-1, aLastDim]);\n\n    // Reshape y into the analogous 2D Tensor, and keep track of the\n    // required dimensions to reproduce the output shape.\n    const bShape = b.shape.slice();\n    const bLastDim = bShape.pop();\n    const ySecondLastDim = bShape.pop();\n    const yOtherDims = [...bShape, bLastDim];\n    // permutation should be like [r-2, 0, 1, 2, ... r-4, r-3, r-1]\n    // where r is the rank of y.\n    const perm = Array.from({length: b.rank}, (_, i) => {\n      if (i === 0) {\n        return b.rank - 2;\n      } else if (i <= b.rank - 2) {\n        return i - 1;\n      }\n      return i;\n    });\n    b = b.transpose(perm).reshape([ySecondLastDim, -1]);\n\n    // Multiply x and y as 2D Tensors, and then reshape back to original.\n    const outputShape = [...aFirstDims, ...yOtherDims];\n    const transposeA = false;\n    const transposeB = false;\n    return tfc.fused\n        .matMul({\n          a,\n          b,\n          transposeA,\n          transposeB,\n          bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n          activation\n        })\n        .reshape(outputShape);\n  }\n}\n\n/**\n * Compute the sign Tensor of an input Tensor.\n *\n * Elements of the input `tf.Tensor` that are === 0 are mapped to 0.\n * Elements of the input `tf.Tensor` that are > 0 are mapped to 1.\n * Elements of the input `tf.Tensor` that are < 0 are mapped to -1.\n *\n * @param x Input `tf.Tensor`.\n * @return The sign `tf.Tensor`.\n */\nexport function sign(x: Tensor): Tensor {\n  // TODO(cais): Move to the core.\n  return tidy(() => {\n    const zerosLikeX = coreZerosLike(x);\n    const onesLikeX = coreOnesLike(x);\n    return where(\n        tfc.equal(x, zerosLikeX), zerosLikeX,\n        where(\n            tfc.greater(x, coreZerosLike(x)), onesLikeX,\n            tfc.mul(-1, onesLikeX)));\n  });\n}\n\n/**\n * Computes the one-hot representation of an integer tensor.\n * @param indices nD integer tensor of shape\n *   `(batch_size, dim1, dim2, ... dim(n-1))`\n * @param numClasses Integer, number of classes to consider.\n * @returns (n + 1)D one hot representation of the input\n *   with shape `(batch_size, dim1, dim2, ... dim(n-1), num_classes)`\n */\nexport function oneHot(indices: Tensor, numClasses: number): Tensor {\n  return tidy(() => {\n    if (indices.rank !== 1) {\n      throw new Error(\n          'Only 1D one-hot tensors are supported in the ' +\n          'deeplearn backend, at present.');\n    }\n    indices = indices.toInt();\n    return tfc.oneHot(indices as Tensor1D, numClasses).toFloat();\n  });\n}\n\n/* Elementary math functions. */\n\n/**\n * Retrieves the elements of indices `indices` in the tensor `reference`.\n * @param reference A tensor.\n * @param indices An integer tensor of indices or an `Array` of integers.\n * @param axis Axis along which to perform the gather operation.\n * @returns The result of the gathering as a tensor.\n */\nexport function gather(\n    reference: Tensor, indices: number[]|Tensor1D, axis?: number): Tensor {\n  return tidy(() => {\n    if (Array.isArray(indices)) {\n      indices = tensor1d(indices, 'int32');\n    } else {\n      indices = indices.toInt();\n    }\n    return tfc.gather(reference, indices, axis);\n  });\n}\n\n/**\n * Element-wise square.\n * @param x Input tensor.\n * @return element-wise x^2\n */\nexport function square(x: Tensor): Tensor {\n  return tfc.mul(x, x);\n}\n\n/**\n * Element-wise exponentiation.\n *\n * Porting Note: In PyKeras, `a` (the exponent) is a Python integer, which\n *   takes advatnage of the backend's (e.g., TensorFlow's) automatic\n * conversion to tensor. Here we allow `a` to be either a number or a tensor.\n *\n * @param x The base tensor.\n * @param a The exponent, tensor or number. If a number, it is rounded to the\n *   nearest integer and converted to a tensor.\n * @returns A tensor of the same shape as `x`.\n */\nexport function pow(x: Tensor, a: Tensor|number): Tensor {\n  return tidy(() => {\n    if (typeof (a) === 'number') {\n      a = scalar(Math.round(a), 'int32');\n    }\n    if (a.dtype !== 'int32') {\n      throw new NotImplementedError(\n          `Non-int32 dtype (${a.dtype}) is not supported by pow() yet`);\n    }\n    return tfc.pow(x, a);\n  });\n}\n\n/**\n * Reshapes bias tensor according to rank of x.\n */\nfunction reshapeBias(xRank: number, bias: Tensor, dataFormat: string) {\n  const biasShape = bias.shape;\n\n  if (bias.rank !== 1 && bias.rank !== xRank) {\n    throw new ValueError(\n        `Unexpected bias dimensions: ${bias.rank}` +\n        `; expected it to be 1 or ${xRank}`);\n  }\n\n  if (xRank === 5) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1, 1]);\n      } else {\n        return bias.reshape(\n            [1, biasShape[3], biasShape[0], biasShape[1], biasShape[2]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 4) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1]);\n      } else {\n        return bias.reshape([1, biasShape[2], biasShape[0], biasShape[1]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 3) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1]);\n      } else {\n        return bias.reshape([1, biasShape[1], biasShape[0]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank < 3) {\n    return bias;\n  }\n  throw new ValueError(`Unsupported input rank by biasAdd: ${bias.rank}`);\n}\n\n/* Neural-network operations. */\n\n/**\n * Add a bias to a tensor.\n *\n * @param x The tensor to add the bias to.\n * @param bias The bias to add to `x`. Must be 1D or the same rank as `x`.\n * @return Result of the bias adding.\n * @throws ValueError: If the rank of `bias` is incorrect.\n */\nexport function biasAdd(\n    x: Tensor, bias: Tensor, dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n\n    return x.add(reshapeBias(x.rank, bias, dataFormat));\n  });\n}\n\n/**\n * Exponential linear unit (ELU).\n * @param x A tensor or variable to compute the activation function for.\n * @param alpha: A scalar, a scaling factor for the negative section.\n * @return Output of the ELU operation.\n */\nexport function elu(x: Tensor, alpha = 1): Tensor {\n  // TODO(cais): Add support for alpha values other than 1.\n  if (alpha !== 1) {\n    throw new NotImplementedError(\n        `Support for alpha values other than 1 (${alpha}) is not implemented ` +\n        `yet.`);\n  }\n  return tfc.elu(x);\n}\n\n/**\n * Softsign of a tensor.\n *\n * Defined as x / (abs(x) + 1), element-wise.\n *\n * @param x: Input.\n * @returns Output.\n */\nexport function softsign(x: Tensor): Tensor {\n  return tidy(() => tfc.div(x, tfc.abs(x).add(1)));\n}\n\n/**\n * Sets entries in `x` to zero at random, while scaling the entire tensor.\n *\n * @param x input tensor.\n * @param level fraction of the entries in the tensor that will be set to 0.\n * @param noiseShape shape of randomly generated keep/drop flags, must be\n *   broadcastable to the shape of `x`. Optional.\n * @param seed random seed to ensure determinism. Optional.\n * @returns Result of the dropout operation.\n */\nexport function dropout(\n    x: Tensor, level: number, noiseShape?: number[], seed?: number): Tensor {\n  return tidy(() => tfc.dropout(x, level, noiseShape, seed));\n}\n\n/**\n * Element-wise, segment-wise linear approximation of sigmoid.\n *\n * Returns `0.` if `x < -2.5`, `1.` if `x > 2.5`.\n * In `-2.5 <= x <= 2.5`, returns `0.2 * x + 0.5`.\n *\n * @param x Input tensor.\n * @returns Output tensor.\n */\nexport function hardSigmoid(x: Tensor): Tensor {\n  return tidy(() => {\n    const y = tfc.add(.5, tfc.mul(.2, x));\n    return tfc.clipByValue(y, 0, 1);\n  });\n}\n\n/**\n * Invoke `x` in the training phase, and `alt` otherwise.\n *\n * Porting Note: We do not create placeholder tensors for the `training`\n * boolean flag here, because there is no such thing in the TF.js imperative\n * backend.\n *\n * @param x The function to invoke iff `training` is `true`.\n * @param alt The function to invoke iff `training` is `false`.\n * @param training Boolean flag for whether training phase is active.\n * @returns The return value of `x()` if `training` is `true`, or the return\n *   value of `alt()` if `training` is `false`.\n */\nexport function inTrainPhase<T>(x: () => T, alt: () => T, training = false): T {\n  return training ? x() : alt();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {BaseSerialization} from './types';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type FanMode = 'fanIn'|'fanOut'|'fanAvg';\nexport const VALID_FAN_MODE_VALUES = ['fanIn', 'fanOut', 'fanAvg'];\n\n// These constants have a snake vs. camel distinction.\nexport type FanModeSerialization = 'fan_in'|'fan_out'|'fan_avg';\n\n/** @docinline */\nexport type Distribution = 'normal'|'uniform'|'truncatedNormal';\nexport const VALID_DISTRIBUTION_VALUES =\n    ['normal', 'uniform', 'truncatedNormal'];\n// These constants have a snake vs. camel distinction.\nexport type DistributionSerialization = 'normal'|'uniform'|'truncated_normal';\n\nexport type ZerosSerialization = BaseSerialization<'Zeros', {}>;\n\nexport type OnesSerialization = BaseSerialization<'Ones', {}>;\n\nexport type ConstantConfig = {\n  value: number;\n};\n\nexport type ConstantSerialization =\n    BaseSerialization<'Constant', ConstantConfig>;\n\nexport type RandomNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type RandomNormalSerialization =\n    BaseSerialization<'RandomNormal', RandomNormalConfig>;\n\nexport type RandomUniformConfig = {\n  minval?: number;\n  maxval?: number;\n  seed?: number;\n};\n\nexport type RandomUniformSerialization =\n    BaseSerialization<'RandomUniform', RandomUniformConfig>;\n\nexport type TruncatedNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type TruncatedNormalSerialization =\n    BaseSerialization<'TruncatedNormal', TruncatedNormalConfig>;\n\nexport type VarianceScalingConfig = {\n  scale?: number;\n\n  mode?: FanModeSerialization;\n  distribution?: DistributionSerialization;\n  seed?: number;\n};\n\nexport type VarianceScalingSerialization =\n    BaseSerialization<'VarianceScaling', VarianceScalingConfig>;\n\nexport type OrthogonalConfig = {\n  seed?: number;\n  gain?: number;\n};\n\nexport type OrthogonalSerialization =\n    BaseSerialization<'Orthogonal', OrthogonalConfig>;\n\nexport type IdentityConfig = {\n  gain?: number;\n};\n\nexport type IdentitySerialization =\n    BaseSerialization<'Identity', IdentityConfig>;\n\n// Update initializerClassNames below in concert with this.\nexport type InitializerSerialization = ZerosSerialization|OnesSerialization|\n    ConstantSerialization|RandomUniformSerialization|RandomNormalSerialization|\n    TruncatedNormalSerialization|IdentitySerialization|\n    VarianceScalingSerialization|OrthogonalSerialization;\n\nexport type InitializerClassName = InitializerSerialization['class_name'];\n\n// We can't easily extract a string[] from the string union type, but we can\n// recapitulate the list, enforcing at compile time that the values are valid\n// and that we have the right number of them.\n\n/**\n * A string array of valid Initializer class names.\n *\n * This is guaranteed to match the `InitializerClassName` union type.\n */\nexport const initializerClassNames: InitializerClassName[] = [\n  'Zeros', 'Ones', 'Constant', 'RandomNormal', 'RandomUniform',\n  'TruncatedNormal', 'VarianceScaling', 'Orthogonal', 'Identity'\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, eye, linalg, mul, ones, randomUniform, scalar, serialization, Tensor, Tensor2D, tidy, truncatedNormal, zeros} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {checkDataFormat} from './common';\nimport {NotImplementedError, ValueError} from './errors';\nimport {DataFormat, Shape} from './keras_format/common';\nimport {Distribution, FanMode, VALID_DISTRIBUTION_VALUES, VALID_FAN_MODE_VALUES} from './keras_format/initializer_config';\nimport {checkStringTypeUnionValue, deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\nimport {arrayProd} from './utils/math_utils';\n\nexport function checkFanMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_FAN_MODE_VALUES, 'FanMode', value);\n}\n\nexport function checkDistribution(value?: string): void {\n  checkStringTypeUnionValue(VALID_DISTRIBUTION_VALUES, 'Distribution', value);\n}\n\n/**\n * Initializer base class.\n *\n * @doc {\n *   heading: 'Initializers', subheading: 'Classes', namespace: 'initializers'}\n */\nexport abstract class Initializer extends serialization.Serializable {\n  public fromConfigUsesCustomObjects(): boolean {\n    return false;\n  }\n  /**\n   * Generate an initial value.\n   * @param shape\n   * @param dtype\n   * @return The init value.\n   */\n  abstract apply(shape: Shape, dtype?: DataType): Tensor;\n\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport class Zeros extends Initializer {\n  /** @nocollapse */\n  static className = 'Zeros';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return zeros(shape, dtype);\n  }\n}\nserialization.registerClass(Zeros);\n\nexport class Ones extends Initializer {\n  /** @nocollapse */\n  static className = 'Ones';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return ones(shape, dtype);\n  }\n}\nserialization.registerClass(Ones);\n\nexport interface ConstantArgs {\n  /** The value for each element in the variable. */\n  value: number;\n}\n\nexport class Constant extends Initializer {\n  /** @nocollapse */\n  static className = 'Constant';\n  private value: number;\n  constructor(args: ConstantArgs) {\n    super();\n    if (typeof args !== 'object') {\n      throw new ValueError(\n          `Expected argument of type ConstantConfig but got ${args}`);\n    }\n    if (args.value === undefined) {\n      throw new ValueError(`config must have value set but got ${args}`);\n    }\n    this.value = args.value;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => mul(scalar(this.value), ones(shape, dtype)));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      value: this.value,\n    };\n  }\n}\nserialization.registerClass(Constant);\n\nexport interface RandomUniformArgs {\n  /** Lower bound of the range of random values to generate. */\n  minval?: number;\n  /** Upper bound of the range of random values to generate. */\n  maxval?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomUniform extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomUniform';\n  readonly DEFAULT_MINVAL = -0.05;\n  readonly DEFAULT_MAXVAL = 0.05;\n  private minval: number;\n  private maxval: number;\n  private seed: number;\n\n  constructor(args: RandomUniformArgs) {\n    super();\n    this.minval = args.minval || this.DEFAULT_MINVAL;\n    this.maxval = args.maxval || this.DEFAULT_MAXVAL;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return randomUniform(shape, this.minval, this.maxval, dtype);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {minval: this.minval, maxval: this.maxval, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomUniform);\n\nexport interface RandomNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomNormal';\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: RandomNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `randomNormal does not support dType ${dtype}.`);\n    }\n\n    return K.randomNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomNormal);\n\nexport interface TruncatedNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class TruncatedNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'TruncatedNormal';\n\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: TruncatedNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `truncatedNormal does not support dType ${dtype}.`);\n    }\n    return truncatedNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(TruncatedNormal);\n\nexport interface IdentityArgs {\n  /**\n   * Multiplicative factor to apply to the identity matrix.\n   */\n  gain?: number;\n}\n\nexport class Identity extends Initializer {\n  /** @nocollapse */\n  static className = 'Identity';\n  private gain: number;\n  constructor(args: IdentityArgs) {\n    super();\n    this.gain = args.gain != null ? args.gain : 1.0;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length !== 2 || shape[0] !== shape[1]) {\n        throw new ValueError(\n            'Identity matrix initializer can only be used for' +\n            ' 2D square matrices.');\n      } else {\n        return mul(this.gain, eye(shape[0]));\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {gain: this.gain};\n  }\n}\nserialization.registerClass(Identity);\n\n/**\n * Computes the number of input and output units for a weight shape.\n * @param shape Shape of weight.\n * @param dataFormat data format to use for convolution kernels.\n *   Note that all kernels in Keras are standardized on the\n *   CHANNEL_LAST ordering (even when inputs are set to CHANNEL_FIRST).\n * @return An length-2 array: fanIn, fanOut.\n */\nfunction computeFans(\n    shape: Shape, dataFormat: DataFormat = 'channelsLast'): number[] {\n  let fanIn: number;\n  let fanOut: number;\n  checkDataFormat(dataFormat);\n  if (shape.length === 2) {\n    fanIn = shape[0];\n    fanOut = shape[1];\n  } else if ([3, 4, 5].indexOf(shape.length) !== -1) {\n    if (dataFormat === 'channelsFirst') {\n      const receptiveFieldSize = arrayProd(shape, 2);\n      fanIn = shape[1] * receptiveFieldSize;\n      fanOut = shape[0] * receptiveFieldSize;\n    } else if (dataFormat === 'channelsLast') {\n      const receptiveFieldSize = arrayProd(shape, 0, shape.length - 2);\n      fanIn = shape[shape.length - 2] * receptiveFieldSize;\n      fanOut = shape[shape.length - 1] * receptiveFieldSize;\n    }\n  } else {\n    const shapeProd = arrayProd(shape);\n    fanIn = Math.sqrt(shapeProd);\n    fanOut = Math.sqrt(shapeProd);\n  }\n\n  return [fanIn, fanOut];\n}\n\nexport interface VarianceScalingArgs {\n  /** Scaling factor (positive float). */\n  scale?: number;\n\n  /** Fanning mode for inputs and outputs. */\n  mode?: FanMode;\n\n  /** Probabilistic distribution of the values. */\n  distribution?: Distribution;\n\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class VarianceScaling extends Initializer {\n  /** @nocollapse */\n  static className = 'VarianceScaling';\n  private scale: number;\n  private mode: FanMode;\n  private distribution: Distribution;\n  private seed: number;\n\n  /**\n   * Constructor of VarianceScaling.\n   * @throws ValueError for invalid value in scale.\n   */\n  constructor(args: VarianceScalingArgs) {\n    super();\n    if (args.scale < 0.0) {\n      throw new ValueError(\n          `scale must be a positive float. Got: ${args.scale}`);\n    }\n    this.scale = args.scale == null ? 1.0 : args.scale;\n    this.mode = args.mode == null ? 'fanIn' : args.mode;\n    checkFanMode(this.mode);\n    this.distribution =\n        args.distribution == null ? 'normal' : args.distribution;\n    checkDistribution(this.distribution);\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    const fans = computeFans(shape);\n    const fanIn = fans[0];\n    const fanOut = fans[1];\n    let scale = this.scale;\n    if (this.mode === 'fanIn') {\n      scale /= Math.max(1, fanIn);\n    } else if (this.mode === 'fanOut') {\n      scale /= Math.max(1, fanOut);\n    } else {\n      scale /= Math.max(1, (fanIn + fanOut) / 2);\n    }\n\n    if (this.distribution === 'normal') {\n      const stddev = Math.sqrt(scale);\n      dtype = dtype || 'float32';\n      if (dtype !== 'float32' && dtype !== 'int32') {\n        throw new NotImplementedError(\n            `${this.getClassName()} does not support dType ${dtype}.`);\n      }\n      return truncatedNormal(shape, 0, stddev, dtype, this.seed);\n    } else {\n      const limit = Math.sqrt(3 * scale);\n      return randomUniform(shape, -limit, limit, dtype);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      scale: this.scale,\n      mode: this.mode,\n      distribution: this.distribution,\n      seed: this.seed\n    };\n  }\n}\nserialization.registerClass(VarianceScaling);\n\nexport interface SeedOnlyInitializerArgs {\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class GlorotUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotUniform';\n\n  /**\n   * Constructor of GlorotUniform\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotUniform);\n\nexport class GlorotNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotNormal';\n\n  /**\n   * Constructor of GlorotNormal.\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotNormal);\n\nexport class HeNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeNormal);\n\nexport class HeUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeUniform';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeUniform);\n\nexport class LeCunNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunNormal);\n\nexport class LeCunUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunUniform);\n\nexport interface OrthogonalArgs extends SeedOnlyInitializerArgs {\n  /**\n   * Multiplicative factor to apply to the orthogonal matrix. Defaults to 1.\n   */\n  gain?: number;\n}\n\nexport class Orthogonal extends Initializer {\n  /** @nocollapse */\n  static className = 'Orthogonal';\n  readonly DEFAULT_GAIN = 1;\n  protected readonly gain: number;\n  protected readonly seed: number;\n\n  constructor(args?: OrthogonalArgs) {\n    super();\n    this.gain = args.gain == null ? this.DEFAULT_GAIN : args.gain;\n    this.seed = args.seed;\n\n    if (this.seed != null) {\n      throw new NotImplementedError(\n          'Random seed is not implemented for Orthogonal Initializer yet.');\n    }\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length < 2) {\n        throw new NotImplementedError('Shape must be at least 2D.');\n      }\n      if (shape[0] * shape[1] > 2000) {\n        console.warn(\n            `Orthogonal initializer is being called on a matrix with more ` +\n            `than 2000 (${shape[0] * shape[1]}) elements: ` +\n            `Slowness may result.`);\n      }\n\n      // TODO(cais): Add seed support.\n      const normalizedShape =\n          shape[0] > shape[1] ? [shape[1], shape[0]] : shape;\n      const a = K.randomNormal(normalizedShape, 0, 1, 'float32') as Tensor2D;\n      let q = linalg.gramSchmidt(a) as Tensor2D;\n      if (shape[0] > shape[1]) {\n        q = q.transpose();\n      }\n      return mul(this.gain, q);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      gain: this.gain,\n      seed: this.seed,\n    };\n  }\n}\nserialization.registerClass(Orthogonal);\n\n/** @docinline */\nexport type InitializerIdentifier =\n    'constant'|'glorotNormal'|'glorotUniform'|'heNormal'|'heUniform'|'identity'|\n    'leCunNormal'|'leCunUniform'|'ones'|'orthogonal'|'randomNormal'|\n    'randomUniform'|'truncatedNormal'|'varianceScaling'|'zeros'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in InitializerIdentifier]: string} = {\n      'constant': 'Constant',\n      'glorotNormal': 'GlorotNormal',\n      'glorotUniform': 'GlorotUniform',\n      'heNormal': 'HeNormal',\n      'heUniform': 'HeUniform',\n      'identity': 'Identity',\n      'leCunNormal': 'LeCunNormal',\n      'leCunUniform': 'LeCunUniform',\n      'ones': 'Ones',\n      'orthogonal': 'Orthogonal',\n      'randomNormal': 'RandomNormal',\n      'randomUniform': 'RandomUniform',\n      'truncatedNormal': 'TruncatedNormal',\n      'varianceScaling': 'VarianceScaling',\n      'zeros': 'Zeros'\n    };\n\nfunction deserializeInitializer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Initializer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'initializer');\n}\n\nexport function serializeInitializer(initializer: Initializer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(initializer);\n}\n\nexport function getInitializer(identifier: InitializerIdentifier|Initializer|\n                               serialization.ConfigDict): Initializer {\n  if (typeof identifier === 'string') {\n    const className = identifier in INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    /* We have four 'helper' classes for common initializers that\n    all get serialized as 'VarianceScaling' and shouldn't go through\n    the deserializeInitializer pathway. */\n    if (className === 'GlorotNormal') {\n      return new GlorotNormal();\n    } else if (className === 'GlorotUniform') {\n      return new GlorotUniform();\n    } else if (className === 'HeNormal') {\n      return new HeNormal();\n    } else if (className === 'HeUniform') {\n      return new HeUniform();\n    } else if (className === 'LeCunNormal') {\n      return new LeCunNormal();\n    } else if (className === 'LeCunUniform') {\n      return new LeCunUniform();\n    } else {\n      const config: serialization.ConfigDict = {};\n      config['className'] = className;\n      config['config'] = {};\n      return deserializeInitializer(config);\n    }\n  } else if (identifier instanceof Initializer) {\n    return identifier;\n  } else {\n    return deserializeInitializer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constant, ConstantArgs, GlorotNormal, GlorotUniform, HeNormal, HeUniform, Identity, IdentityArgs, Initializer, LeCunNormal, LeCunUniform, Ones, Orthogonal, OrthogonalArgs, RandomNormal, RandomNormalArgs, RandomUniform, RandomUniformArgs, SeedOnlyInitializerArgs, TruncatedNormal, TruncatedNormalArgs, VarianceScaling, VarianceScalingArgs, Zeros} from './initializers';\n\n/**\n * Initializer that generates tensors initialized to 0.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function zeros(): Zeros {\n  return new Zeros();\n}\n\n/**\n * Initializer that generates tensors initialized to 1.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function ones(): Initializer {\n  return new Ones();\n}\n\n/**\n * Initializer that generates values initialized to some constant.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function constant(args: ConstantArgs): Initializer {\n  return new Constant(args);\n}\n\n/**\n * Initializer that generates random values initialized to a uniform\n * distribution.\n *\n * Values will be distributed uniformly between the configured minval and\n * maxval.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomUniform(args: RandomUniformArgs): Initializer {\n  return new RandomUniform(args);\n}\n\n/**\n * Initializer that generates random values initialized to a normal\n * distribution.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomNormal(args: RandomNormalArgs): Initializer {\n  return new RandomNormal(args);\n}\n\n/**\n * Initializer that generates random values initialized to a truncated normal.\n * distribution.\n *\n * These values are similar to values from a `RandomNormal` except that values\n * more than two standard deviations from the mean are discarded and re-drawn.\n * This is the recommended initializer for neural network weights and filters.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function truncatedNormal(args: TruncatedNormalArgs): Initializer {\n  return new TruncatedNormal(args);\n}\n\n/**\n * Initializer that generates the identity matrix.\n * Only use for square 2D matrices.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function identity(args: IdentityArgs): Initializer {\n  return new Identity(args);\n}\n\n/**\n * Initializer capable of adapting its scale to the shape of weights.\n * With distribution=NORMAL, samples are drawn from a truncated normal\n * distribution centered on zero, with `stddev = sqrt(scale / n)` where n is:\n *   - number of input units in the weight tensor, if mode = FAN_IN.\n *   - number of output units, if mode = FAN_OUT.\n *   - average of the numbers of input and output units, if mode = FAN_AVG.\n * With distribution=UNIFORM,\n * samples are drawn from a uniform distribution\n * within [-limit, limit], with `limit = sqrt(3 * scale / n)`.\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function varianceScaling(config: VarianceScalingArgs): Initializer {\n  return new VarianceScaling(config);\n}\n\n/**\n * Glorot uniform initializer, also called Xavier uniform initializer.\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotUniform(args);\n}\n\n/**\n * Glorot normal initializer, also called Xavier normal initializer.\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor.\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotNormal(args);\n}\n\n/**\n * He normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function heNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeNormal(args);\n}\n\n/**\n * He uniform initializer.\n *\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / fan_in)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function heUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeUniform(args);\n}\n\n/**\n * LeCun normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(1 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * References:\n *   [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *   [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunNormal(args);\n}\n\n/**\n * LeCun uniform initializer.\n *\n * It draws samples from a uniform distribution in the interval\n * `[-limit, limit]` with `limit = sqrt(3 / fanIn)`,\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunUniform(args);\n}\n\n/**\n * Initializer that generates a random orthogonal matrix.\n *\n * Reference:\n * [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function orthogonal(args: OrthogonalArgs): Initializer {\n  return new Orthogonal(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Utilities related to persistent state in the backend.\n */\n\n/**\n * An ID to track `tf.SymbolicTensor`s and derived classes.\n * Required in different places in engine/topology.ts to identify unique\n * tensors.\n */\nlet _nextUniqueTensorId = 0;\n\nexport function getNextUniqueTensorId(): number {\n  return _nextUniqueTensorId++;\n}\n\nconst _uidPrefixes: {[prefix: string]: number} = {};\n\n/**\n * Provides a unique UID given a string prefix.\n *\n * @param prefix\n */\nexport function getUid(prefix = ''): string {\n  if (!(prefix in _uidPrefixes)) {\n    _uidPrefixes[prefix] = 0;\n  }\n  _uidPrefixes[prefix] += 1;\n  return prefix + _uidPrefixes[prefix].toString();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\n// tslint:enable\n\n/**\n * Determine whether the input is an Array of Shapes.\n */\nexport function isArrayOfShapes(x: Shape|Shape[]): boolean {\n  return Array.isArray(x) && Array.isArray(x[0]);\n}\n\n/**\n * Special case of normalizing shapes to lists.\n *\n * @param x A shape or list of shapes to normalize into a list of Shapes.\n * @return A list of Shapes.\n */\nexport function normalizeShapeList(x: Shape|Shape[]): Shape[] {\n  if (x.length === 0) {\n    return [];\n  }\n  if (!Array.isArray(x[0])) {\n    return [x] as Shape[];\n  }\n  return x as Shape[];\n}\n\n/**\n * Helper function to obtain exactly one Tensor.\n * @param xs: A single `tf.Tensor` or an `Array` of `tf.Tensor`s.\n * @return A single `tf.Tensor`. If `xs` is an `Array`, return the first one.\n * @throws ValueError: If `xs` is an `Array` and its length is not 1.\n */\nexport function getExactlyOneTensor(xs: Tensor|Tensor[]): Tensor {\n  let x: Tensor;\n  if (Array.isArray(xs)) {\n    if (xs.length !== 1) {\n      throw new ValueError(`Expected Tensor length to be 1; got ${xs.length}`);\n    }\n    x = xs[0];\n  } else {\n    x = xs;\n  }\n  return x;\n}\n\n/**\n * Helper function to obtain exactly on instance of Shape.\n *\n * @param shapes Input single `Shape` or Array of `Shape`s.\n * @returns If input is a single `Shape`, return it unchanged. If the input is\n *   an `Array` containing exactly one instance of `Shape`, return the instance.\n *   Otherwise, throw a `ValueError`.\n * @throws ValueError: If input is an `Array` of `Shape`s, and its length is not\n *   1.\n */\nexport function getExactlyOneShape(shapes: Shape|Shape[]): Shape {\n  if (Array.isArray(shapes) && Array.isArray(shapes[0])) {\n    if (shapes.length === 1) {\n      shapes = shapes as Shape[];\n      return shapes[0];\n    } else {\n      throw new ValueError(`Expected exactly 1 Shape; got ${shapes.length}`);\n    }\n  } else {\n    return shapes as Shape;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {LayerVariable} from '../variables';\n\n/**\n * Count the elements in an Array of LayerVariables.\n *\n * @param weights: The LayerVariables of which the constituent numbers are to\n *   be counted.\n * @returns A count of the elements in all the LayerVariables\n */\nexport function countParamsInWeights(weights: LayerVariable[]): number {\n  let count = 0;\n  for (const weight of weights) {\n    if (weight.shape.length === 0) {\n      count += 1;\n    } else {\n      count += weight.shape.reduce((a, b) => a * b);\n    }\n  }\n  return count;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, Tensor, variableGrads} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId} from './backend/state';\nimport {getScopedTensorName, getUniqueTensorName} from './common';\nimport {Constraint} from './constraints';\nimport {NotImplementedError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {HasShape} from './types';\n\nconst DEFAULT_VARIABLE_NAME_PREFIX = 'Variable';\n\n/**\n * A `tf.layers.LayerVariable` is similar to a `tf.Tensor` in that it has a\n * dtype and shape, but its value is mutable.  The value is itself represented\n * as a`tf.Tensor`, and can be read with the `read()` method and updated with\n * the `write()` method.\n */\nexport class LayerVariable {\n  readonly dtype: DataType;\n  readonly shape: Shape;\n\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName: string;\n  private trainable_: boolean;\n\n  protected readonly val: tfc.Variable;\n  readonly constraint: Constraint;\n  /**\n   * Construct Variable from a `tf.Tensor`.\n   *\n   * If not explicitly named, the Variable will be given a name with the\n   * prefix 'Variable'. Variable names are unique. In the case of name\n   * collision, suffixies '_<num>' will be added to the name.\n   *\n   * @param val Initial value of the Variable.\n   * @param name Name of the variable. If `null` or `undefined` is provided, it\n   *   will default a name with the prefix 'Variable'.\n   * @param constraint Optional, projection function to be applied to the\n   * variable after optimize updates\n   * @throws ValueError if `name` is `null` or `undefined`.\n   */\n  constructor(\n      val: Tensor, dtype: DataType = 'float32',\n      name = DEFAULT_VARIABLE_NAME_PREFIX, trainable = true,\n      constraint: Constraint = null) {\n    this.dtype = dtype == null ? 'float32' : dtype;\n    this.shape = val.shape;\n    this.id = getNextUniqueTensorId();\n\n    name = name == null ? DEFAULT_VARIABLE_NAME_PREFIX : name;\n    this.originalName = getScopedTensorName(name);\n    this.name = getUniqueTensorName(this.originalName);\n\n    this.trainable_ = trainable;\n    this.constraint = constraint;\n\n    this.val = tfc.variable(val, this.trainable_, this.name, this.dtype);\n  }\n\n  /**\n   * Get a snapshot of the Variable's value.\n   *\n   * The returned value is a snapshot of the Variable's value at the time of\n   * the invocation. Future mutations in the value of the tensor will only\n   * be reflected by future calls to this method.\n   */\n  read(): Tensor {\n    this.assertNotDisposed();\n    return this.val;\n  }\n\n  /**\n   * Update the value of the Variable.\n   *\n   * @param newVal: The new value to update to. Must be consistent with the\n   *   dtype and shape of the Variable.\n   * @return This Variable.\n   */\n  write(newVal: Tensor) {\n    // TODO(cais): Once  TF.js Core supports Tensor.dtype, check dtype match.\n    this.assertNotDisposed();\n    checkShapesMatch(this.val, newVal);\n    // Skip updating if this is the exact same tensor.\n    if (this.val.id !== newVal.id) {\n      this.val.assign(newVal);\n      if (this.constraint != null) {\n        this.val.assign(this.constraint.apply(this.val));\n      }\n    }\n    return this;\n  }\n\n  /**\n   * Dispose this LayersVariable instance from memory.\n   */\n  dispose(): void {\n    this.assertNotDisposed();\n    this.val.dispose();\n  }\n\n  protected assertNotDisposed(): void {\n    if (this.val.isDisposed) {\n      throw new Error(`LayersVariable ${this.name} is already disposed.`);\n    }\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.trainable_ = trainable;\n    this.val.trainable = trainable;\n  }\n}\n\nfunction checkShapesMatch(x: HasShape, y: HasShape): void {\n  if (x.shape.toString() !== y.shape.toString()) {\n    throw new Error(\n        'Shape mismatch: ' + JSON.stringify(x.shape) + ' vs. ' +\n        JSON.stringify(y.shape));\n  }\n}\n\n/**\n * Create a Variable.\n * @param x The initial value of the `Variable`.\n * @param dtype optional, the type of the variable.\n * @param name optional, the name of the variable, default provided by\n * Variable.\n * @param constraint optional, a constraint to be applied after every update.\n * @return The newly instantiated `Variable`.\n */\nexport function variable(\n    x: Tensor, dtype?: DataType, name?: string,\n    constraint?: Constraint): LayerVariable {\n  return new LayerVariable(x, dtype, name, true, constraint);\n}\n\n/**\n * Instantiates an all-zeros Variable and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-zero Variable.\n */\nexport function zerosVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  return new LayerVariable(tfc.zeros(shape), dtype, name);\n}\n\n/**\n * Instantiates an all-zeros tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function zerosLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.zerosLike(x), dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-ones Variable.\n */\nexport function onesVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  const allocated = tfc.ones(shape);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function onesLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  const allocated = tfc.onesLike(x);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiate an identity matrix and returns it, as a Variable\n *\n * @param size Number of rows/columns.\n * @param dtype Data type of returned Variable.\n * @param name Name of returned Variable.\n * @return A Variable, an identity matrix.\n */\nexport function eyeVariable(\n    size: number, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.eye(size), dtype, name);\n}\n\n/**\n * Get a Variable with uniform distribution of values.\n * @param shape Shape of the tensor.\n * @param minval Lower bound of the uniform distribution.\n * @param maxval Upper bound of the uniform distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The uniform-random Variable.\n */\nexport function randomUniformVariable(\n    shape: Shape, minval: number, maxval: number, dtype?: DataType,\n    seed?: number, name = 'randomUniform'): LayerVariable {\n  return new LayerVariable(\n      tfc.randomUniform(shape, minval, maxval, dtype), dtype, name);\n}\n\n/**\n * Get a Variable with truncated-normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function truncatedNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'truncatedNormal'): LayerVariable {\n  // TODO(cais): Implement logic for dtype and seed once they are supported\n  // by deeplearn.js.\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormal does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.truncatedNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n/**\n * Get a Variable with normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function randomNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'randomNormal'): LayerVariable {\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormalVariable does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.randomNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n\n/**\n * Update the value of a Variable.\n * @param x The Variable to be updated.\n * @param xNew The new value to update to.\n * @return The Variable updated.\n */\nexport function update(x: LayerVariable, xNew: Tensor): LayerVariable {\n  return x.write(xNew);\n}\n\n/**\n * Update the value of a Variable by adding an increment.\n * @param x The Variable to be updated.\n * @param increment The incrment to add to `x`.\n * @return The Variable updated.\n */\nexport function updateAdd(x: LayerVariable, increment: Tensor): LayerVariable {\n  return x.write(tfc.add(x.read(), increment));\n}\n\n/**\n * Update the value of a Variable by subtracting a decrement.\n * @param x The Variable to be updated.\n * @param decrement The decrement to subtract from `x`.\n * @return The Variable updated.\n */\nexport function updateSub(x: LayerVariable, decrement: Tensor): LayerVariable {\n  return x.write(tfc.sub(x.read(), decrement));\n}\n\n/**\n * Get the values of an array of Variables.\n *\n * @param tensors An `Array` of `Variable`s to get the values of.\n * @return The values of the inputs, as an `Array` of`tf.Tensor`s.\n */\nexport function batchGetValue(xs: LayerVariable[]): Tensor[] {\n  return xs.map(x => x.read());\n}\n\n/**\n * Update the value of multiple Variables at once.\n *\n * @param variablesAndValues An `Array`, each element is of type\n *   [Variable, Tensor]. The first item is the\n *   `Variable` of which the value is to be updated. The second item\n *   carries the new value.\n */\nexport function batchSetValue(\n    variablesAndValues: Array<[LayerVariable, Tensor]>): void {\n  variablesAndValues.forEach(variableAndValue => {\n    const variable: LayerVariable = variableAndValue[0];\n    variable.write(variableAndValue[1]);\n  });\n}\n\n/**\n * Returns the gradients of `variables` w.r.t. the return value of `lossFn`.\n * @param lossFn A function which returns a Scalar to be used as the function\n *   value (i.e., numerator) for differentiation.\n * @param variables List of variables to be used as the independent variables\n *   (i.e., denominator) for differentiation.\n * @returns An Array of gradients tensors.\n */\nexport function gradients(\n    lossFn: () => tfc.Scalar, variables: LayerVariable[]): Tensor[] {\n  // TODO(cais): The return type signature can be simplified if deeplearn makes\n  //   the corresponding type public.\n  const variableList =\n      variables.map(variable => variable.read() as tfc.Variable);\n  const valudAndGrads = variableGrads(lossFn, variableList);\n  return variables.map(variable => valudAndGrads.grads[variable.name]);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {DataType, Scalar, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId, getUid} from '../backend/state';\nimport {getScopedTensorName, getUniqueTensorName, nameScope} from '../common';\nimport {Constraint} from '../constraints';\nimport {AttributeError, NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {getInitializer, Initializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs, RegularizerFn} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as types_utils from '../utils/types_utils';\nimport * as variable_utils from '../utils/variable_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\n\n// TODO(michaelterry): This is a stub until it's defined.\nexport type Op = (x: LayerVariable) => LayerVariable;\n\n/**\n * Constructor arguments for InputSpec.\n */\nexport interface InputSpecArgs {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n}\n\n/**\n * Specifies the ndim, dtype and shape of every input to a layer.\n *\n * Every layer should expose (if appropriate) an `inputSpec` attribute:\n * a list of instances of InputSpec (one per input tensor).\n *\n * A null entry in a shape is compatible with any dimension,\n * a null shape is compatible with any shape.\n */\nexport class InputSpec {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n\n  constructor(args: InputSpecArgs) {\n    this.dtype = args.dtype;\n    this.shape = args.shape;\n    /*\n      TODO(michaelterry): Could throw error if ndim and shape are both defined\n        (then backport).\n    */\n    if (args.shape != null) {\n      this.ndim = args.shape.length;\n    } else {\n      this.ndim = args.ndim;\n    }\n    this.maxNDim = args.maxNDim;\n    this.minNDim = args.minNDim;\n    this.axes = args.axes || {};\n  }\n}\n\n/**\n * `tf.SymbolicTensor` is a placeholder for a Tensor without any concrete value.\n *\n * They are most often encountered when building a graph of `Layer`s for a\n * a `tf.LayersModel` and the input data's shape, but not values are known.\n *\n * @doc {heading: 'Models', 'subheading': 'Classes'}\n */\nexport class SymbolicTensor {\n  /* A unique ID for the tensor to be able to differentiate tensors. */\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName?: string;\n  /**\n   * Rank/dimensionality of the tensor.\n   */\n  readonly rank: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  nodeIndex: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  tensorIndex: number;\n\n  /**\n   *\n   * @param dtype\n   * @param shape\n   * @param sourceLayer The Layer that produced this symbolic tensor.\n   * @param inputs The inputs passed to sourceLayer's __call__() method.\n   * @param nodeIndex\n   * @param tensorIndex\n   * @param callArgs The keyword arguments passed to the __call__() method.\n   * @param name\n   * @param outputTensorIndex The index of this tensor in the list of outputs\n   *   returned by apply().\n   */\n  constructor(\n      readonly dtype: DataType, readonly shape: Shape,\n      public sourceLayer: Layer, readonly inputs: SymbolicTensor[],\n      readonly callArgs: Kwargs, name?: string,\n      readonly outputTensorIndex?: number) {\n    this.id = getNextUniqueTensorId();\n    if (name != null) {\n      this.originalName = getScopedTensorName(name);\n      this.name = getUniqueTensorName(this.originalName);\n    }\n    this.rank = shape.length;\n  }\n}\n\n/**\n * Constructor arguments for Node.\n */\nexport interface NodeArgs {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`.\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n}\n\n/**\n * The type of the return value of Layer.dispose() and Container.dispose().\n */\nexport interface DisposeResult {\n  /**\n   * Reference count after the dispose call.\n   */\n  refCountAfterDispose: number;\n\n  /**\n   * Number of variables dispose in this dispose call.\n   */\n  numDisposedVariables: number;\n}\n\nlet _nextNodeID = 0;\n\n/**\n * A `Node` describes the connectivity between two layers.\n *\n * Each time a layer is connected to some new input,\n * a node is added to `layer.inboundNodes`.\n *\n * Each time the output of a layer is used by another layer,\n * a node is added to `layer.outboundNodes`.\n *\n * `nodeIndices` and `tensorIndices` are basically fine-grained coordinates\n * describing the origin of the `inputTensors`, verifying the following:\n *\n * `inputTensors[i] ==\n * inboundLayers[i].inboundNodes[nodeIndices[i]].outputTensors[\n *   tensorIndices[i]]`\n *\n * A node from layer A to layer B is added to:\n *     A.outboundNodes\n *     B.inboundNodes\n */\nexport class Node {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n\n  readonly id: number;\n\n  constructor(\n      args: NodeArgs,\n      // TODO(michaelterry): Define actual type for this.\n      public callArgs?: Kwargs) {\n    this.id = _nextNodeID++;\n    /*\n      Layer instance (NOT a list).\n      this is the layer that takes a list of input tensors\n      and turns them into a list of output tensors.\n      the current node will be added to\n      the inboundNodes of outboundLayer.\n    */\n    this.outboundLayer = args.outboundLayer;\n\n    /*\n        The following 3 properties describe where\n        the input tensors come from: which layers,\n        and for each layer, which node and which\n        tensor output of each node.\n    */\n\n    // List of layer instances.\n    this.inboundLayers = args.inboundLayers;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.nodeIndices = args.nodeIndices;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.tensorIndices = args.tensorIndices;\n\n    /*\n        Following 2 properties:\n        tensor inputs and outputs of outboundLayer.\n    */\n\n    // List of tensors. 1:1 mapping with inboundLayers.\n    this.inputTensors = args.inputTensors;\n    // List of tensors, created by outboundLayer.call().\n    this.outputTensors = args.outputTensors;\n\n    /*\n        Following 2 properties: input and output masks.\n        List of tensors, 1:1 mapping with inputTensor.\n    */\n    this.inputMasks = args.inputMasks;\n    // List of tensors, created by outboundLayer.computeMask().\n    this.outputMasks = args.outputMasks;\n\n    // Following 2 properties: input and output shapes.\n\n    // List of shape tuples, shapes of inputTensors.\n    this.inputShapes = args.inputShapes;\n    // List of shape tuples, shapes of outputTensors.\n    this.outputShapes = args.outputShapes;\n\n    // Add nodes to all layers involved.\n    for (const layer of args.inboundLayers) {\n      if (layer != null) {\n        layer.outboundNodes.push(this);\n      }\n    }\n    args.outboundLayer.inboundNodes.push(this);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const inboundNames: string[] = [];\n    for (const layer of this.inboundLayers) {\n      if (layer != null) {\n        inboundNames.push(layer.name);\n      } else {\n        inboundNames.push(null);\n      }\n    }\n    return {\n      outboundLayer: this.outboundLayer ? this.outboundLayer.name : null,\n      inboundLayers: inboundNames,\n      nodeIndices: this.nodeIndices,\n      tensorIndices: this.tensorIndices\n    };\n  }\n}\n\n/** Constructor arguments for Layer. */\nexport declare interface LayerArgs {\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  inputShape?: Shape;\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  batchInputShape?: Shape;\n  /**\n   * If `inputShape` is specified and `batchInputShape` is *not* specified,\n   * `batchSize` is used to construct the `batchInputShape`: `[batchSize,\n   * ...inputShape]`\n   */\n  batchSize?: number;\n  /**\n   * The data-type for this layer. Defaults to 'float32'.\n   * This argument is only applicable to input layers (the first layer of a\n   * model).\n   */\n  dtype?: DataType;\n  /** Name for this layer. */\n  name?: string;\n  /**\n   * Whether the weights of this layer are updatable by `fit`.\n   * Defaults to true.\n   */\n  trainable?: boolean;\n  /**\n   * Initial weight values of the layer.\n   */\n  weights?: Tensor[];\n  /** Legacy support. Do not use for new code. */\n  inputDType?: DataType;\n}\n\n// If necessary, add `output` arguments to the CallHook function.\n// This is currently used for testing only, but may be used for debugger-related\n// purposes in the future.\nexport type CallHook = (inputs: Tensor|Tensor[], kwargs: Kwargs) => void;\n\nlet _nextLayerID = 0;\n\n/**\n * A layer is a grouping of operations and weights that can be composed to\n * create a `tf.LayersModel`.\n *\n * Layers are constructed by using the functions under the\n * [tf.layers](#Layers-Basic) namespace.\n *\n * @doc {heading: 'Layers', subheading: 'Classes', namespace: 'layers'}\n */\nexport abstract class Layer extends serialization.Serializable {\n  /** Name for this layer. Must be unique within a model. */\n  name: string;\n  /**\n   * List of InputSpec class instances.\n   *\n   * Each entry describes one required input:\n   * - ndim\n   * - dtype\n   * A layer with `n` input tensors must have an `inputSpec` of length `n`.\n   */\n  inputSpec: InputSpec[];\n  supportsMasking: boolean;\n  /** Whether the layer weights will be updated during training. */\n  protected trainable_: boolean;\n  batchInputShape: Shape;\n  dtype: DataType;\n  initialWeights: Tensor[];\n\n  inboundNodes: Node[];\n  outboundNodes: Node[];\n\n  activityRegularizer: Regularizer;\n\n  protected _trainableWeights: LayerVariable[];\n  private _nonTrainableWeights: LayerVariable[];\n  private _losses: RegularizerFn[];\n  // TODO(cais): _updates is currently unused.\n  private _updates: Tensor[];\n  private _built: boolean;\n  private _callHook: CallHook = null;\n\n  private _addedWeightNames: string[] = [];\n\n  readonly id: number;\n\n  // Porting Notes: PyKeras does not have this property in this base Layer\n  //   class. Instead lets Layer subclass set it dynamically and checks the\n  //   value with `hasattr`. In tfjs-layers, we let this be a member of this\n  //   base class.\n  protected _stateful = false;\n\n  protected _refCount: number|null;\n\n  // A flag for whether fast (i.e., all-zero) weight initialization is to\n  // be used during `build()` call. This speeds up weight initialization\n  // by saving unnecessary calls to expensive initializers in cases where\n  // the initialized values will be overwritten by loaded weight values\n  // during model loading.\n  private fastWeightInitDuringBuild: boolean;\n\n  constructor(args: LayerArgs = {}) {\n    super();\n    this.id = _nextLayerID++;\n\n    this.activityRegularizer = null;\n\n    this.inputSpec = null;\n    this.supportsMasking = false;\n\n    // These properties will be set upon call of this.build()\n    this._trainableWeights = [];\n    this._nonTrainableWeights = [];\n    this._losses = [];\n    this._updates = [];\n    this._built = false;\n\n    /*\n      These lists will be filled via successive calls\n      to this.addInboundNode().\n     */\n    this.inboundNodes = [];\n    this.outboundNodes = [];\n\n    let name = args.name;\n    if (!name) {\n      const prefix = this.getClassName();\n      name = generic_utils.toSnakeCase(prefix) + '_' + getUid(prefix);\n    }\n    this.name = name;\n\n    this.trainable_ = args.trainable == null ? true : args.trainable;\n\n    if (args.inputShape != null || args.batchInputShape != null) {\n      /*\n        In this case we will later create an input layer\n        to insert before the current layer\n       */\n      let batchInputShape: Shape;\n      if (args.batchInputShape != null) {\n        batchInputShape = args.batchInputShape;\n      } else if (args.inputShape != null) {\n        let batchSize: number = null;\n        if (args.batchSize != null) {\n          batchSize = args.batchSize;\n        }\n        batchInputShape = [batchSize].concat(args.inputShape);\n      }\n      this.batchInputShape = batchInputShape;\n\n      // Set dtype.\n      let dtype = args.dtype;\n      if (dtype == null) {\n        dtype = args.inputDType;\n      }\n      if (dtype == null) {\n        dtype = 'float32';\n      }\n      this.dtype = dtype;\n    }\n\n    if (args.weights != null) {\n      this.initialWeights = args.weights;\n    } else {\n      this.initialWeights = null;\n    }\n\n    // The value of `_refCount` is initialized to null. When the layer is used\n    // in a symbolic way for the first time, it will be set to 1.\n    this._refCount = null;\n\n    this.fastWeightInitDuringBuild = false;\n  }\n\n  /**\n   * Converts a layer and its index to a unique (immutable type) name.\n   * This function is used internally with `this.containerNodes`.\n   * @param layer The layer.\n   * @param nodeIndex The layer's position (e.g. via enumerate) in a list of\n   *   nodes.\n   *\n   * @returns The unique name.\n   */\n  protected static nodeKey(layer: Layer, nodeIndex: number) {\n    return layer.name + '_ib-' + nodeIndex.toString();\n  }\n\n  /**\n   * Returns this.inboundNode at index nodeIndex.\n   *\n   * Porting note: This is a replacement for _get_node_attribute_at_index()\n   * @param nodeIndex\n   * @param attrName The name of the attribute related to request for this node.\n   */\n  private getNodeAtIndex(nodeIndex: number, attrName: string): Node {\n    if (this.inboundNodes.length === 0) {\n      throw new RuntimeError(\n          'The layer has never been called ' +\n          `and thus has no defined ${attrName}.`);\n    }\n    if (this.inboundNodes.length <= nodeIndex) {\n      throw new ValueError(\n          `Asked to get ${attrName} at node ${nodeIndex}, ` +\n          `but the layer has only ${this.inboundNodes.length} inbound nodes.`);\n    }\n    return this.inboundNodes[nodeIndex];\n  }\n\n  /**\n   * Retrieves the input tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple inputs).\n   */\n  getInputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple outputs).\n   */\n  getOutputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'output').outputTensors);\n  }\n\n  // Properties\n\n  /**\n   * Retrieves the input tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Input tensor or list of input tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get input(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer input\" ' +\n          'is ill-defined. ' +\n          'Use `getInputAt(nodeIndex)` instead.');\n    } else if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' is not connected, no input to return.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Output tensor or list of output tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get output(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has no inbound nodes.');\n    }\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer output\" ' +\n          'is ill-defined. ' +\n          'Use `getOutputAt(nodeIndex)` instead.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'output').outputTensors);\n  }\n\n  get losses(): RegularizerFn[] {\n    return this._losses;\n  }\n\n  /**\n   * Retrieves the Layer's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Layer.loss in PyKeras.\n    //   In PyKeras, Layer.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return this.losses.map(lossFn => lossFn());\n  }\n\n  get updates(): Tensor[] {\n    return this._updates;\n  }\n\n  get built(): boolean {\n    return this._built;\n  }\n\n  set built(built: boolean) {\n    this._built = built;\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this._trainableWeights.forEach(w => w.trainable = trainable);\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (this.trainable_) {\n      return this._trainableWeights.filter(w => w.trainable);\n    } else {\n      return [];\n    }\n  }\n\n  set trainableWeights(weights: LayerVariable[]) {\n    this._trainableWeights = weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    if (this.trainable) {\n      return this._trainableWeights.filter(w => !w.trainable)\n          .concat(this._nonTrainableWeights);\n    } else {\n      return this._trainableWeights.concat(this._nonTrainableWeights);\n    }\n  }\n\n  set nonTrainableWeights(weights: LayerVariable[]) {\n    this._nonTrainableWeights = weights;\n  }\n\n  /**\n   * The concatenation of the lists trainableWeights and nonTrainableWeights\n   * (in this order).\n   */\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  get stateful(): boolean {\n    return this._stateful;\n  }\n\n  /**\n   * Reset the states of the layer.\n   *\n   * This method of the base Layer class is essentially a no-op.\n   * Subclasses that are stateful (e.g., stateful RNNs) should override this\n   * method.\n   */\n  resetStates(): void {\n    if (!this.stateful) {\n      throw new Error(\n          'Cannot call the resetStates() method of a non-stateful Layer ' +\n          'object.');\n    }\n  }\n\n  /**\n   * Checks compatibility between the layer and provided inputs.\n   *\n   * This checks that the tensor(s) `input`\n   * verify the input assumptions of the layer\n   * (if any). If not, exceptions are raised.\n   *\n   * @param inputs Input tensor or list of input tensors.\n   *\n   * @exception ValueError in case of mismatch between\n   *   the provided inputs and the expectations of the layer.\n   */\n  protected assertInputCompatibility(inputs: Tensor|Tensor[]|SymbolicTensor|\n                                     SymbolicTensor[]): void {\n    inputs = generic_utils.toList(inputs);\n    if (this.inputSpec == null || this.inputSpec.length === 0) {\n      return;\n    }\n    const inputSpec = generic_utils.toList(this.inputSpec);\n    if (inputs.length !== inputSpec.length) {\n      throw new ValueError(\n          `Layer ${this.name} expects ${inputSpec.length} inputs, ` +\n          `but it received ${inputs.length} input tensors. ` +\n          `Input received: ${inputs}`);\n    }\n    for (let inputIndex = 0; inputIndex < inputs.length; inputIndex++) {\n      const x = inputs[inputIndex];\n      const spec: InputSpec = inputSpec[inputIndex];\n      if (spec == null) {\n        continue;\n      }\n\n      // Check ndim.\n      const ndim = x.rank;\n      if (spec.ndim != null) {\n        if (ndim !== spec.ndim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}: ` +\n              `expected ndim=${spec.ndim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.maxNDim != null) {\n        if (ndim > spec.maxNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected max_ndim=${spec.maxNDim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.minNDim != null) {\n        if (ndim < spec.minNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected min_ndim=${spec.minNDim}, found ndim=${ndim}.`);\n        }\n      }\n\n      // Check dtype.\n      if (spec.dtype != null) {\n        if (x.dtype !== spec.dtype) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name} ` +\n              `: expected dtype=${spec.dtype}, found dtype=${x.dtype}.`);\n        }\n      }\n\n      // Check specific shape axes.\n      if (spec.axes) {\n        const xShape = x.shape;\n        for (const key in spec.axes) {\n          const axis = Number(key);\n          const value = spec.axes[key];\n          // Perform Python-style slicing in case axis < 0;\n          // TODO(cais): Use https://github.com/alvivi/typescript-underscore to\n          // ensure type safety through Underscore calls.\n          const xShapeAtAxis =\n              axis >= 0 ? xShape[axis] : xShape[xShape.length + axis];\n          if (value != null && [value, null].indexOf(xShapeAtAxis) === -1) {\n            throw new ValueError(\n                `Input ${inputIndex} is incompatible with layer ` +\n                `${this.name}: expected axis ${axis} of input shape to ` +\n                `have value ${value} but got shape ${xShape}.`);\n          }\n        }\n      }\n\n      // Check shape.\n      if (spec.shape != null) {\n        for (let i = 0; i < spec.shape.length; ++i) {\n          const specDim = spec.shape[i];\n          const dim = x.shape[i];\n          if (specDim != null && dim != null) {\n            if (specDim !== dim) {\n              throw new ValueError(\n                  `Input ${inputIndex} is incompatible with layer ` +\n                  `${this.name}: expected shape=${spec.shape}, ` +\n                  `found shape=${x.shape}.`);\n            }\n          }\n        }\n      }\n    }\n  }\n\n  /**\n   * This is where the layer's logic lives.\n   *\n   * @param inputs Input tensor, or list/tuple of input tensors.\n   * @param kwargs Additional keyword arguments.\n   *\n   * @return A tensor or list/tuple of tensors.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return inputs;\n  }\n\n  protected invokeCallHook(inputs: Tensor|Tensor[], kwargs: Kwargs) {\n    if (this._callHook != null) {\n      this._callHook(inputs, kwargs);\n    }\n  }\n\n  /**\n   * Set call hook.\n   * This is currently used for testing only.\n   * @param callHook\n   */\n  setCallHook(callHook: CallHook) {\n    this._callHook = callHook;\n  }\n\n  /**\n   * Clear call hook.\n   * This is currently used for testing only.\n   */\n  clearCallHook() {\n    this._callHook = null;\n  }\n\n  /**\n   * Builds or executes a `Layer's logic.\n   *\n   * When called with `tf.Tensor`(s), execute the `Layer`s computation and\n   * return Tensor(s). For example:\n   *\n   * ```js\n   * const denseLayer = tf.layers.dense({\n   *   units: 1,\n   *   kernelInitializer: 'zeros',\n   *   useBias: false\n   * });\n   *\n   * // Invoke the layer's apply() method with a `tf.Tensor` (with concrete\n   * // numeric values).\n   * const input = tf.ones([2, 2]);\n   * const output = denseLayer.apply(input);\n   *\n   * // The output's value is expected to be [[0], [0]], due to the fact that\n   * // the dense layer has a kernel initialized to all-zeros and does not have\n   * // a bias.\n   * output.print();\n   * ```\n   *\n   * When called with `tf.SymbolicTensor`(s), this will prepare the layer for\n   * future execution.  This entails internal book-keeping on shapes of\n   * expected Tensors, wiring layers together, and initializing weights.\n   *\n   * Calling `apply` with `tf.SymbolicTensor`s are typically used during the\n   * building of non-`tf.Sequential` models. For example:\n   *\n   * ```js\n   * const flattenLayer = tf.layers.flatten();\n   * const denseLayer = tf.layers.dense({units: 1});\n   *\n   * // Use tf.layers.input() to obtain a SymbolicTensor as input to apply().\n   * const input = tf.input({shape: [2, 2]});\n   * const output1 = flattenLayer.apply(input);\n   *\n   * // output1.shape is [null, 4]. The first dimension is the undetermined\n   * // batch size. The second dimension comes from flattening the [2, 2]\n   * // shape.\n   * console.log(JSON.stringify(output1.shape));\n   *\n   * // The output SymbolicTensor of the flatten layer can be used to call\n   * // the apply() of the dense layer:\n   * const output2 = denseLayer.apply(output1);\n   *\n   * // output2.shape is [null, 1]. The first dimension is the undetermined\n   * // batch size. The second dimension matches the number of units of the\n   * // dense layer.\n   * console.log(JSON.stringify(output2.shape));\n   *\n   * // The input and output and be used to construct a model that consists\n   * // of the flatten and dense layers.\n   * const model = tf.model({inputs: input, outputs: output2});\n   * ```\n   *\n   * @param inputs a `tf.Tensor` or `tf.SymbolicTensor` or an Array of them.\n   * @param kwargs Additional keyword arguments to be passed to `call()`.\n   *\n   * @return Output of the layer's `call` method.\n   *\n   * @exception ValueError error in case the layer is missing shape information\n   *   for its `build` call.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  // Porting Note: This is a replacement for __call__() in Python.\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    kwargs = kwargs || {};\n\n    this.assertNotDisposed();\n\n    // Ensure inputs are all the same type.\n    const inputsList = generic_utils.toList(inputs);\n\n    let allAreSymbolic = true;\n    for (const input of inputsList) {\n      if (!(input instanceof SymbolicTensor)) {\n        allAreSymbolic = false;\n        break;\n      }\n    }\n    let noneAreSymbolic = true;\n    for (const input of inputsList) {\n      if (input instanceof SymbolicTensor) {\n        noneAreSymbolic = false;\n        break;\n      }\n    }\n\n    if (allAreSymbolic === noneAreSymbolic) {\n      throw new ValueError(\n          'Arguments to apply() must be all ' +\n          'SymbolicTensors or all Tensors');\n    }\n\n    // TODO(michaelterry): nameScope() may not be necessary.\n    return nameScope(this.name, () => {\n      // Handle laying building (weight creating, input spec locking).\n      if (!this.built) {\n        /*\n          Throw exceptions in case the input is not compatible\n          with the inputSpec specified in the layer constructor.\n         */\n        this.assertInputCompatibility(inputs);\n\n        // Collect input shapes to build layer.\n        const inputShapes: Shape[] = [];\n        for (const xElem of generic_utils.toList(inputs)) {\n          inputShapes.push(xElem.shape);\n        }\n        this.build(generic_utils.singletonOrArray(inputShapes));\n        this.built = true;\n\n        // Load weights that were specified at layer instantiation.\n        if (this.initialWeights) {\n          this.setWeights(this.initialWeights);\n        }\n\n        if (this._refCount === null && noneAreSymbolic) {\n          // The first use of this layer is a non-symbolic call, set ref count\n          // to 1 so the Layer can be properly disposed if its dispose() method\n          // is called.\n          this._refCount = 1;\n        }\n      }\n\n      /*\n        Throw exceptions in case the input is not compatible\n        with the inputSpec set at build time.\n      */\n      this.assertInputCompatibility(inputs);\n\n      // Handle mask propagation.\n      // TODO(michaelterry): Mask propagation not currently implemented.\n\n      // Actually call the layer, collecting output(s), mask(s), and shape(s).\n      if (noneAreSymbolic) {\n        let output = this.call(inputs as Tensor | Tensor[], kwargs);\n        // TODO(michaelterry): Compute the outputMask\n\n        // If the layer returns tensors from its inputs, unmodified,\n        // we copy them to avoid loss of tensor metadata.\n        const outputList: Tensor[] = generic_utils.toList(output);\n        const outputListCopy: Tensor[] = [];\n        // TODO(michaelterry): This copying may not be necessary given our eager\n        // backend.\n        for (let x of outputList) {\n          if (inputsList.indexOf(x) !== -1) {\n            x = x.clone();\n          }\n          outputListCopy.push(x);\n        }\n        output = generic_utils.singletonOrArray(outputListCopy);\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        // TODO(michaelterry): Call addInboundNode()?\n        return output;\n      } else {\n        const inputShape = collectInputShape(inputs);\n        const outputShape = this.computeOutputShape(inputShape);\n        let output: SymbolicTensor|SymbolicTensor[];\n        const outputDType = guessOutputDType(inputs);\n        this.warnOnIncompatibleInputShape(\n            Array.isArray(inputs) ? inputShape[0] as Shape :\n                                    inputShape as Shape);\n\n        if (outputShape != null && outputShape.length > 0 &&\n            Array.isArray(outputShape[0])) {\n          // We have multiple output shapes. Create multiple output tensors.\n          output = (outputShape as Shape[])\n                       .map(\n                           (shape, index) => new SymbolicTensor(\n                               outputDType, shape, this,\n                               generic_utils.toList(inputs), kwargs, this.name,\n                               index));\n        } else {\n          output = new SymbolicTensor(\n              outputDType, outputShape as Shape, this,\n              generic_utils.toList(inputs), kwargs, this.name);\n        }\n\n        /*\n          Add an inbound node to the layer, so that it keeps track\n          of the call and of all new variables created during the call.\n          This also updates the layer history of the output tensor(s).\n          If the input tensor(s) had no previous history,\n          this does nothing.\n        */\n        this.addInboundNode(\n            inputs as SymbolicTensor | SymbolicTensor[], output, null, null,\n            inputShape, outputShape, kwargs);\n        this._refCount++;\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        return output;\n      }\n    });\n  }\n\n  /**\n   * Check compatibility between input shape and this layer's batchInputShape.\n   *\n   * Print warning if any incompatibility is found.\n   *\n   * @param inputShape Input shape to be checked.\n   */\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {\n    if (this.batchInputShape == null) {\n      return;\n    } else if (inputShape.length !== this.batchInputShape.length) {\n      console.warn(\n          `The rank of the input tensor provided (shape: ` +\n          `${JSON.stringify(inputShape)}) does not match that of the ` +\n          `batchInputShape (${JSON.stringify(this.batchInputShape)}) ` +\n          `of the layer ${this.name}`);\n    } else {\n      let dimMismatch = false;\n      this.batchInputShape.forEach((dimension, i) => {\n        if (dimension != null && inputShape[i] != null &&\n            inputShape[i] !== dimension) {\n          dimMismatch = true;\n        }\n      });\n      if (dimMismatch) {\n        console.warn(\n            `The shape of the input tensor ` +\n            `(${JSON.stringify(inputShape)}) does not ` +\n            `match the expectation of layer ${this.name}: ` +\n            `${JSON.stringify(this.batchInputShape)}`);\n      }\n    }\n  }\n\n  /**\n   * Retrieves the output shape(s) of a layer.\n   *\n   * Only applicable if the layer has only one inbound node, or if all inbound\n   * nodes have the same output shape.\n   *\n   * @returns Output shape or shapes.\n   * @throws AttributeError: if the layer is connected to more than one incoming\n   *   nodes.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  get outputShape(): Shape|Shape[] {\n    if (this.inboundNodes == null || this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `The layer ${this.name} has never been called and thus has no ` +\n          `defined output shape.`);\n    }\n    const allOutputShapes: string[] = [];\n    for (const node of this.inboundNodes) {\n      const shapeString = JSON.stringify(node.outputShapes);\n      if (allOutputShapes.indexOf(shapeString) === -1) {\n        allOutputShapes.push(shapeString);\n      }\n    }\n    if (allOutputShapes.length === 1) {\n      const outputShapes = this.inboundNodes[0].outputShapes;\n      if (Array.isArray(outputShapes) && Array.isArray(outputShapes[0]) &&\n          outputShapes.length === 1) {\n        return (outputShapes as Shape[])[0];\n      } else {\n        return outputShapes;\n      }\n\n    } else {\n      throw new AttributeError(\n          `The layer ${this.name} has multiple inbound nodes with different ` +\n          `output shapes. Hence the notion of \"output shape\" is ill-defined ` +\n          `for the layer.`);\n      // TODO(cais): Implement getOutputShapeAt().\n    }\n  }\n\n  /**\n   * Counts the total number of numbers (e.g., float32, int32) in the\n   * weights.\n   *\n   * @returns An integer count.\n   * @throws RuntimeError: If the layer is not built yet (in which case its\n   *   weights are not defined yet.)\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  countParams(): number {\n    if (!this.built) {\n      throw new RuntimeError(\n          `You tried to call countParams() on ${this.name}, ` +\n          `but the layer is not built yet. Build it first by calling ` +\n          `build(batchInputShape).`);\n    }\n    return variable_utils.countParamsInWeights(this.weights);\n  }\n\n  /**\n   * Creates the layer weights.\n   *\n   * Must be implemented on all layers that have weights.\n   *\n   * Called when apply() is called to construct the weights.\n   *\n   * @param inputShape A `Shape` or array of `Shape` (unused).\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  build(inputShape: Shape|Shape[]) {\n    this.built = true;\n  }\n\n  /**\n   * Returns the current values of the weights of the layer.\n   *\n   * @param trainableOnly Whether to get the values of only trainable weights.\n   * @returns Weight values as an `Array` of `tf.Tensor`s.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getWeights(trainableOnly = false): Tensor[] {\n    return batchGetValue(trainableOnly ? this.trainableWeights : this.weights);\n  }\n\n  /**\n   * Sets the weights of the layer, from Tensors.\n   *\n   * @param weights a list of Tensors. The number of arrays and their shape\n   *   must match number of the dimensions of the weights of the layer (i.e.\n   *   it should match the output of `getWeights`).\n   *\n   * @exception ValueError If the provided weights list does not match the\n   *   layer's specifications.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  setWeights(weights: Tensor[]): void {\n    tidy(() => {\n      const params = this.weights;\n      if (params.length !== weights.length) {\n        // TODO(cais): Restore the following and use `providedWeights`, instead\n        // of `weights` in the error message, once the deeplearn.js bug is\n        // fixed: https://github.com/PAIR-code/deeplearnjs/issues/498 const\n        // providedWeights = JSON.stringify(weights).substr(0, 50);\n        throw new ValueError(\n            `You called setWeights(weights) on layer \"${this.name}\" ` +\n            `with a weight list of length ${weights.length}, ` +\n            `but the layer was expecting ${params.length} weights. ` +\n            `Provided weights: ${weights}...`);\n      }\n      if (params.length === 0) {\n        return;\n      }\n      const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n      const paramValues = batchGetValue(params);\n      for (let i = 0; i < paramValues.length; ++i) {\n        const pv = paramValues[i];\n        const p = params[i];\n        const w = weights[i];\n        if (!util.arraysEqual(pv.shape, w.shape)) {\n          throw new ValueError(\n              `Layer weight shape ${pv.shape} ` +\n              `not compatible with provided weight shape ${w.shape}`);\n        }\n        weightValueTuples.push([p, w]);\n      }\n      batchSetValue(weightValueTuples);\n    });\n  }\n\n  /**\n   * Adds a weight variable to the layer.\n   *\n   * @param name Name of the new weight variable.\n   * @param shape The shape of the weight.\n   * @param dtype The dtype of the weight.\n   * @param initializer An initializer instance.\n   * @param regularizer A regularizer instance.\n   * @param trainable Whether the weight should be trained via backprop or not\n   *   (assuming that the layer itself is also trainable).\n   * @param constraint An optional trainable.\n   * @return The created weight variable.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  protected addWeight(\n      name: string, shape: Shape, dtype?: DataType, initializer?: Initializer,\n      regularizer?: Regularizer, trainable?: boolean,\n      constraint?: Constraint): LayerVariable {\n    // Reject duplicate weight names.\n    if (this._addedWeightNames.indexOf(name) !== -1) {\n      throw new ValueError(\n          `Duplicate weight name ${name} for layer ${this.name}`);\n    }\n    this._addedWeightNames.push(name);\n\n    if (dtype == null) {\n      dtype = 'float32';\n    }\n\n    if (this.fastWeightInitDuringBuild) {\n      initializer = getInitializer('zeros');\n    }\n    const initValue = initializer.apply(shape, dtype);\n    const weight =\n        new LayerVariable(initValue, dtype, name, trainable, constraint);\n    initValue.dispose();\n    // Request backend not to dispose the weights of the model on scope() exit.\n    if (regularizer != null) {\n      this.addLoss(() => regularizer.apply(weight.read()));\n    }\n    if (trainable == null) {\n      trainable = true;\n    }\n    if (trainable) {\n      this._trainableWeights.push(weight);\n    } else {\n      this._nonTrainableWeights.push(weight);\n    }\n    return weight;\n  }\n\n  /**\n   * Set the fast-weight-initialization flag.\n   *\n   * In cases where the initialized weight values will be immediately\n   * overwritten by loaded weight values during model loading, setting\n   * the flag to `true` saves unnecessary calls to potentially expensive\n   * initializers and speeds up the loading process.\n   *\n   * @param value Target value of the flag.\n   */\n  setFastWeightInitDuringBuild(value: boolean) {\n    this.fastWeightInitDuringBuild = value;\n  }\n\n  /**\n   * Add losses to the layer.\n   *\n   * The loss may potentionally be conditional on some inputs tensors,\n   * for instance activity losses are conditional on the layer's inputs.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  addLoss(losses: RegularizerFn|RegularizerFn[]): void {\n    if (losses == null || Array.isArray(losses) && losses.length === 0) {\n      return;\n    }\n    // Update this.losses\n    losses = generic_utils.toList(losses);\n    if (this._losses !== undefined && this._losses !== null) {\n      this.losses.push(...losses);\n    }\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (!this.supportsMasking) {\n      if (mask != null) {\n        if (Array.isArray(mask)) {\n          mask.forEach(maskElement => {\n            if (maskElement != null) {\n              throw new TypeError(\n                  `Layer ${this.name} does not support masking, ` +\n                  'but was passed an inputMask.');\n            }\n          });\n        } else {\n          throw new TypeError(\n              `Layer ${this.name} does not support masking, ` +\n              'but was passed an inputMask.');\n        }\n      }\n      // masking not explicitly supported: return null as mask\n      return null;\n    }\n    // if masking is explictly supported, by default\n    // carry over the input mask\n    return mask;\n  }\n\n  /**\n   * Internal method to create an inbound node for the layer.\n   *\n   * @param inputTensors List of input tensors.\n   * @param outputTensors List of output tensors.\n   * @param inputMasks List of input masks (a mask can be a tensor, or null).\n   * @param outputMasks List of output masks (a mask can be a tensor, or null).\n   * @param inputShapes List of input shape tuples.\n   * @param outputShapes List of output shape tuples.\n   * @param kwargs Dictionary of keyword arguments that were passed to the\n   *   `call` method of the layer at the call that created the node.\n   */\n  private addInboundNode(\n      inputTensors: SymbolicTensor|SymbolicTensor[],\n      outputTensors: SymbolicTensor|SymbolicTensor[],\n      inputMasks: Tensor|Tensor[], outputMasks: Tensor|Tensor[],\n      inputShapes: Shape|Shape[], outputShapes: Shape|Shape[],\n      kwargs: {} = null): void {\n    const inputTensorList: SymbolicTensor[] =\n        generic_utils.toList(inputTensors);\n    outputTensors = generic_utils.toList(outputTensors);\n    inputMasks = generic_utils.toList(inputMasks);\n    outputMasks = generic_utils.toList(outputMasks);\n    inputShapes = types_utils.normalizeShapeList(inputShapes);\n    outputShapes = types_utils.normalizeShapeList(outputShapes);\n\n    // Collect input tensor(s) coordinates.\n    const inboundLayers: Layer[] = [];\n    const nodeIndices: number[] = [];\n    const tensorIndices: number[] = [];\n    for (const x of inputTensorList) {\n      /*\n       * TODO(michaelterry): Keras adds this value to tensors; it's not\n       * clear whether we'll use this or not.\n       */\n      inboundLayers.push(x.sourceLayer);\n      nodeIndices.push(x.nodeIndex);\n      tensorIndices.push(x.tensorIndex);\n    }\n\n    // Create node, add it to inbound nodes.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node(\n        {\n          outboundLayer: this,\n          inboundLayers,\n          nodeIndices,\n          tensorIndices,\n          inputTensors: inputTensorList,\n          outputTensors,\n          inputMasks,\n          outputMasks,\n          inputShapes,\n          outputShapes\n        },\n        kwargs);\n\n    // Update tensor history\n    for (let i = 0; i < outputTensors.length; i++) {\n      // TODO(michaelterry: _uses_learning_phase not tracked.\n      outputTensors[i].sourceLayer = this;\n      outputTensors[i].nodeIndex = this.inboundNodes.length - 1;\n      outputTensors[i].tensorIndex = i;\n    }\n  }\n\n  /**\n   * Returns the config of the layer.\n   *\n   * A layer config is a TS dictionary (serializable)\n   * containing the configuration of a layer.\n   * The same layer can be reinstantiated later\n   * (without its trained weights) from this configuration.\n   *\n   * The config of a layer does not include connectivity\n   * information, nor the layer class name.  These are handled\n   * by 'Container' (one layer of abstraction above).\n   *\n   * Porting Note: The TS dictionary follows TS naming standrds for\n   * keys, and uses tfjs-layers type-safe Enums.  Serialization methods\n   * should use a helper function to convert to the pythonic storage\n   * standard. (see serialization_utils.convertTsToPythonic)\n   *\n   * @returns TS dictionary of configuration.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getConfig(): serialization.ConfigDict {\n    const config:\n        serialization.ConfigDict = {name: this.name, trainable: this.trainable};\n    if (this.batchInputShape != null) {\n      config['batchInputShape'] = this.batchInputShape;\n    }\n    if (this.dtype != null) {\n      config['dtype'] = this.dtype;\n    }\n    return config;\n  }\n\n  /**\n   * Dispose the weight variables that this Layer instance holds.\n   *\n   * @returns {number} Number of disposed variables.\n   */\n  protected disposeWeights(): number {\n    this.weights.forEach(weight => weight.dispose());\n    return this.weights.length;\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Layer '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose layer's weights.\n   *\n   * This method decrease the reference count of the Layer object by 1.\n   *\n   * A Layer is reference-counted. Its reference count is incremented by 1\n   * the first item its `apply()` method is called and when it becomes a part\n   * of a new `Node` (through calling the `apply()`) method on a\n   * `tf.SymbolicTensor`).\n   *\n   * If the reference count of a Layer becomes 0, all the weights will be\n   * disposed and the underlying memory (e.g., the textures allocated in WebGL)\n   * will be freed.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * weights of the Layer will *not* be disposed.\n   *\n   * After a Layer is disposed, it cannot be used in calls such as `apply()`,\n   * `getWeights()` or `setWeights()` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the Container after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the layer has already\n   *   been disposed.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  dispose(): DisposeResult {\n    if (!this.built) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been ` +\n          `built yet.`);\n    }\n\n    if (this._refCount === null) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been used ` +\n          `yet.`);\n    }\n\n    this.assertNotDisposed();\n\n    let numDisposedVariables = 0;\n    if (--this._refCount === 0) {\n      numDisposedVariables = this.disposeWeights();\n    }\n\n    return {refCountAfterDispose: this._refCount, numDisposedVariables};\n  }\n}\n\n/**\n * Collects the input shape(s) of a list of `tf.Tensor`s or\n * `tf.SymbolicTensor`s.\n *\n * TODO(michaelterry): Update PyKeras docs (backport).\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return List of shape tuples (or single tuple), one tuple per input.\n */\nfunction collectInputShape(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                           Tensor[]): Shape|Shape[] {\n  inputTensors =\n      generic_utils.toList(inputTensors) as SymbolicTensor[] | Tensor[];\n  const shapes: Shape[] = [];\n  for (const x of inputTensors) {\n    shapes.push(x.shape);\n  }\n  return generic_utils.singletonOrArray(shapes);\n}\n\n/**\n * Guesses output dtype based on inputs.\n *\n * At present, just returns 'float32' for any input.\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return The guessed DType. At present, always returns 'float32'.\n */\nfunction guessOutputDType(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                          Tensor[]): DataType {\n  return 'float32';\n}\n\n/**\n * Returns the list of input tensors necessary to compute `tensor`.\n *\n * Output will always be a list of tensors (potentially with 1 element).\n *\n * @param tensor The tensor to start from.\n * @param layer Origin layer of the tensor.\n * @param nodeIndex Origin node index of the tensor.\n *\n * @return Array of input tensors.\n */\nexport function getSourceInputs(\n    tensor: SymbolicTensor, layer?: Layer,\n    nodeIndex?: number): SymbolicTensor[] {\n  if (layer == null || (nodeIndex != null && nodeIndex > 0)) {\n    layer = tensor.sourceLayer;\n    nodeIndex = tensor.nodeIndex;\n  }\n  if (layer.inboundNodes.length === 0) {\n    return [tensor];\n  } else {\n    const node = layer.inboundNodes[nodeIndex];\n    if (node.inboundLayers.length === 0) {\n      return node.inputTensors;\n    } else {\n      const sourceTensors: SymbolicTensor[] = [];\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const x = node.inputTensors[i];\n        const layer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const previousSources = getSourceInputs(x, layer, nodeIndex);\n        // Avoid input redundancy.\n        for (const x of previousSources) {\n          if (sourceTensors.indexOf(x) === -1) {\n            sourceTensors.push(x);\n          }\n        }\n      }\n      return sourceTensors;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\n\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/**\n * Constructor arguments for InputLayer.\n *\n * Note: You should provide only inputShape or batchInputShape (not both).\n * If only inputShape is provided, then the batchInputShape is determined by\n * the batchSize argument and the inputShape: [batchSize].concat(inputShape).\n */\nexport declare interface InputLayerArgs {\n  /** Input shape, not including the batch axis. */\n  inputShape?: Shape;\n  /** Optional input batch size (integer or null). */\n  batchSize?: number;\n  /** Batch input shape, including the batch axis. */\n  batchInputShape?: Shape;\n  /** Datatype of the input.  */\n  dtype?: DataType;\n  /**\n   * Whether the placeholder created is meant to be sparse.\n   */\n  sparse?: boolean;  // TODO(michaelterry): Not clear whether we'll need this.\n\n  /** Name of the layer. */\n  name?: string;\n}\n\nexport class InputLayer extends Layer {\n  /** @nocollapse */\n  static readonly className = 'InputLayer';\n  sparse: boolean;\n  constructor(args: InputLayerArgs) {\n    super({\n      dtype: args.dtype,\n      name: args.name != null ? args.name : getUid('input').toString()\n    });\n    // Normalize config.batchSize and config.sparse\n    if (args.batchSize == null) {\n      args.batchSize = null;\n    }\n    if (args.sparse == null) {\n      args.sparse = false;\n    }\n\n    this.trainable = false;\n    this.built = true;\n    this.sparse = args.sparse;\n\n    if (args.inputShape != null && args.batchInputShape != null) {\n      throw new ValueError(\n          'Only provide the inputShape OR ' +\n          'batchInputShape argument to inputLayer, not both at the same time.');\n    }\n    let batchInputShape = args.batchInputShape;\n    if (batchInputShape == null) {\n      if (args.inputShape == null) {\n        throw new ValueError(\n            'An InputLayer should be passed either a ' +\n            '`batchInputShape` or an `inputShape`.');\n      } else {\n        batchInputShape = [args.batchSize].concat(args.inputShape);\n      }\n    } else {\n      // TODO(michaelterry): Backport to PyKeras\n      if (args.batchSize != null) {\n        throw new ValueError(\n            'Cannot specify batchSize if batchInputShape is ' +\n            'specified when creating an InputLayer.');\n      }\n    }\n\n    const dtype = args.dtype || 'float32';\n\n    this.batchInputShape = batchInputShape;\n    this.dtype = dtype;\n    // TODO(michaelterry): Backport this to PyKeras?\n    this.inputSpec = [{shape: batchInputShape}];\n\n    const inputTensor = new SymbolicTensor(\n        this.dtype, this.batchInputShape, this, [], {}, this.name);\n    inputTensor.nodeIndex = 0;\n    inputTensor.tensorIndex = 0;\n\n    // Create an input node to add to this.outboundNode.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: [inputTensor],\n      outputTensors: [inputTensor],\n      inputMasks: [null],\n      outputMasks: [null],\n      inputShapes: [batchInputShape],\n      outputShapes: [batchInputShape]\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor {\n    throw new ValueError(\n        'Cannot pass any input to an ' +\n        `InputLayer's apply() method. InputLayer name: ${this.name}`);\n  }\n\n  dispose(): DisposeResult {\n    // dispose() for InputLayer is overridden as no-op.\n    return {refCountAfterDispose: this._refCount, numDisposedVariables: 0};\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      batchInputShape: this.batchInputShape,\n      dtype: this.dtype,\n      sparse: this.sparse,\n      name: this.name\n    };\n  }\n}\nserialization.registerClass(InputLayer);\n\n/**\n * Config for the Input function.\n *\n * Note: You should provide only shape or batchShape (not both).\n * If only shape is provided, then the batchShape becomes\n * [null].concat(inputShape).\n */\nexport interface InputConfig {\n  /**\n   * A shape, not including the batch size. For instance, `shape=[32]`\n   * indicates that the expected input will be batches of 32-dimensional\n   * vectors.\n   */\n  shape?: Shape;\n  /**\n   * A shape tuple (integer), including the batch size. For instance,\n   * `batchShape=[10, 32]` indicates that the expected input will be batches of\n   * 10 32-dimensional vectors. `batchShape=[null, 32]` indicates batches of an\n   * arbitrary number of 32-dimensional vectors.\n   */\n  batchShape?: Shape;\n  /**\n   * An optional name string for the layer. Should be unique in a model (do not\n   * reuse the same name twice). It will be autogenerated if it isn't provided.\n   */\n  name?: string;\n  dtype?: DataType;\n  /**\n   * A boolean specifying whether the placeholder to be created is sparse.\n   */\n  sparse?: boolean;\n}\n\nexport function Input(config: InputConfig): SymbolicTensor {\n  if (config.batchShape == null && config.shape == null) {\n    throw new Error(\n        'Please provide to Input either a `shape`' +\n        ' or a `batchShape` argument. Note that ' +\n        '`shape` does not include the batch ' +\n        'dimension.');\n  }\n  if (config.batchShape != null && config.shape != null) {\n    // TODO(michaelterry): Backport to PyKeras.\n    throw new ValueError(\n        'Please provide either a `shape` or `batchShape` ' +\n        'argument to Input, but not both.');\n  }\n  let batchShape = config.batchShape;\n  if (config.shape != null && batchShape == null) {\n    batchShape = [null].concat(config.shape);\n  }\n\n  let dtype = config.dtype;\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n\n  const inputLayer = new InputLayer({\n    batchInputShape: batchShape,\n    name: config.name,\n    dtype,\n    sparse: config.sparse\n  });\n\n  const outputs = inputLayer.inboundNodes[0].outputTensors;\n  return outputs[0];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, Scalar} from '@tensorflow/tfjs-core';\n\n/**\n * Logs in which values can be either numbers or Tensors (Scalars).\n *\n * Used internally.\n */\nexport type UnresolvedLogs = {\n  [key: string]: number|Scalar;\n};\n\n/**\n * Turn any Scalar values in a Logs object into actual number values.\n *\n * @param logs The `Logs` object to be resolved in place.\n */\nexport async function resolveScalarsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n  const keys: string[] = [];\n  const scalarsToDispose: Scalar[] = [];\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      const valueScalar = value;\n      promises.push(valueScalar.data());\n      keys.push(key);\n      scalarsToDispose.push(valueScalar);\n    }\n  }\n  if (promises.length > 0) {\n    const values = await Promise.all(promises);\n    for (let i = 0; i < values.length; ++i) {\n      logs[keys[i]] = values[i][0];\n    }\n    // Dispose the original scalar tensors.\n    dispose(scalarsToDispose);\n  }\n}\n\n/**\n * Dispose all Tensors in an UnresolvedLogs object.\n *\n * @param logs An `UnresolvedLogs` object potentially containing `tf.Tensor`s in\n *   places where the values can be `tf.Tensor` or `number`.\n */\nexport function disposeTensorsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      value.dispose();\n    }\n  }\n}\n\n/**\n * Logs in which values can only be numbers.\n *\n * Used when calling client-provided custom callbacks.\n */\nexport type Logs = {\n  [key: string]: number;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {add, div, keep, mul, nextFrame, Scalar, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Container} from './engine/container';\nimport {ValueError} from './errors';\nimport {Logs, resolveScalarsInLogs, UnresolvedLogs} from './logs';\nimport * as generic_utils from './utils/generic_utils';\n\n/** Verbosity logging level when fitting a model. */\nexport enum ModelLoggingVerbosity {\n  SILENT = 0,\n  VERBOSE = 1\n}\n\n/** How often to yield to the main thread when training (in ms). */\nexport const DEFAULT_YIELD_EVERY_MS = 125;\n\nexport type Params = {\n  [key: string]: number|string|boolean|number[]|string[]|boolean[];\n};\n\nexport type YieldEveryOptions = 'auto'|'batch'|'epoch'|'never'|number;\n\n/**\n * Abstract base class used to build new callbacks.\n *\n * The `logs` dictionary that callback methods take as argument will contain\n * keys for quantities relevant to the current batch or epoch.\n *\n * Currently, the `.fit()` method of the `Sequential` model class\n * will include the following quantities in the `logs` that\n * it passes to its callbacks:\n *\n * onEpochEnd: Logs include `acc` and `loss`, and optionally include `valLoss`\n *   (if validation is enabled in `fit`), and `valAcc` (if validation and\n *   accuracy monitoring are enabled).\n * onBatchBegin: Logs include `size`, the number of samples in the current\n *   batch.\n * onBatchEnd: Logs include `loss`, and optionally `acc` (if accuracy monitoring\n *   is enabled).\n */\nexport abstract class BaseCallback {\n  // TODO(michaelterry): This type is a best guess.\n  validationData: Tensor|Tensor[] = null;\n  /**\n   * Training parameters (eg. verbosity, batch size, number of epochs...).\n   */\n  params: Params;\n\n  setParams(params: Params): void {\n    this.params = params;\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {}\n\n  async onTrainBegin(logs?: UnresolvedLogs) {}\n\n  async onTrainEnd(logs?: UnresolvedLogs) {}\n\n  // LayersModel needs to call Callback.setModel(), but cannot actually depend\n  // on Callback because that creates a cyclic dependency.  Providing this no-op\n  // method on BaseCallback breaks the cycle: this way LayersModel can depend on\n  // BaseCallback but not on Callback.  The argument is typed as `Container`\n  // (the superclass of LayersModel) to avoid recapitulating the cycle. Callback\n  // overrides this method and enforces that the argument is really a\n  // LayersModel.\n  setModel(model: Container): void {\n    // Do nothing. Use Callback instead of BaseCallback to track the model.\n  }\n}\n\n/**\n * Container abstracting a list of callbacks.\n */\nexport class CallbackList {\n  callbacks: BaseCallback[];\n  queueLength: number;\n\n  // TODO(cais): When the need arises, uncomment the following lines and\n  // implement the queue for time values.\n  // private deltaTBatch: number;\n  // private deltaTsBatchBegin: Array<number>;\n  // private deltaTsBatchEnd: Array<number>;\n\n  /**\n   * Constructor of CallbackList.\n   * @param callbacks Array of `Callback` instances.\n   * @param queueLength Queue length for keeping running statistics over\n   *   callback execution time.\n   */\n  constructor(callbacks?: BaseCallback[], queueLength = 10) {\n    // TODO(cais): Make use of queueLength when implementing the queue for time\n    // values.\n    if (callbacks == null) {\n      callbacks = [];\n    }\n    this.callbacks = callbacks;\n    this.queueLength = queueLength;\n  }\n\n  append(callback: BaseCallback): void {\n    this.callbacks.push(callback);\n  }\n\n  setParams(params: Params): void {\n    for (const callback of this.callbacks) {\n      callback.setParams(params);\n    }\n  }\n\n  setModel(model: Container): void {\n    for (const callback of this.callbacks) {\n      callback.setModel(model);\n    }\n  }\n\n  /**\n   * Called at the start of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochBegin(epoch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochEnd(epoch, logs);\n    }\n  }\n\n  /**\n   * Called  right before processing a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchBegin(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchEnd(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the beginning of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainBegin(logs);\n    }\n  }\n\n  /**\n   * Called at the end of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainEnd(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainEnd(logs);\n    }\n  }\n}\n\n/**\n * Callback that accumulates epoch averages of metrics.\n *\n * This callback is automatically applied to every LayersModel.\n */\nexport class BaseLogger extends BaseCallback {\n  private seen: number;\n  private totals: UnresolvedLogs;\n\n  constructor() {\n    super();\n  }\n\n  async onEpochBegin(epoch: number) {\n    this.seen = 0;\n    this.totals = {};\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const batchSize = logs['size'] == null ? 0 : logs['size'] as number;\n    this.seen += batchSize;\n    for (const key in logs) {\n      const value = logs[key];\n      if (typeof value === 'number') {\n        if (!this.totals.hasOwnProperty(key)) {\n          this.totals[key] = 0;\n        }\n        this.totals[key] = this.totals[key] as number + value * batchSize;\n      } else {\n        let oldTotalsToDispose: Scalar;\n        if (key in this.totals) {\n          oldTotalsToDispose = this.totals[key] as Scalar;\n        } else {\n          this.totals[key] = 0;\n        }\n        const total: Scalar =\n            tidy(() => add((this.totals[key]), mul(value, batchSize)));\n        this.totals[key] = total;\n        if (oldTotalsToDispose != null) {\n          oldTotalsToDispose.dispose();\n        }\n      }\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs != null) {\n      for (const key of this.params['metrics'] as string[]) {\n        if (this.totals[key] == null) {\n          continue;\n        }\n        if (typeof this.totals[key] === 'number') {\n          logs[key] = this.totals[key] as number / this.seen;\n        } else {\n          tidy(() => {\n            const log: Scalar = mul(div(1, this.seen), this.totals[key]);\n            logs[key] = log;\n            (this.totals[key] as Tensor).dispose();\n            keep(logs[key] as Scalar);\n          });\n        }\n      }\n    }\n  }\n}\n\n/**\n * Callback that records events into a `History` object. This callback is\n * automatically applied to every TF.js Layers model. The `History` object\n * gets returned by the `fit` method of models.\n */\nexport class History extends BaseCallback {\n  epoch: number[];\n  history: {[key: string]: Array<number|Tensor>};\n\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    this.epoch = [];\n    this.history = {};\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    this.epoch.push(epoch);\n    for (const key in logs) {\n      if (this.history[key] == null) {\n        this.history[key] = [];\n      }\n      this.history[key].push(logs[key]);\n    }\n  }\n\n  /**\n   * Await the values of all losses and metrics.\n   */\n  async syncData() {\n    const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n    const keys: string[] = [];\n    const indices: number[] = [];\n    for (const key in this.history) {\n      const valueArray = this.history[key];\n      for (let i = 0; i < valueArray.length; ++i) {\n        if (typeof valueArray[i] !== 'number') {\n          const valueScalar = valueArray[i] as Tensor;\n          promises.push(valueScalar.data());\n          keys.push(key);\n          indices.push(i);\n        }\n      }\n    }\n    const values = await Promise.all(promises);\n    for (let n = 0; n < values.length; ++n) {\n      const tensorToDispose = this.history[keys[n]][indices[n]] as Tensor;\n      tensorToDispose.dispose();\n      this.history[keys[n]][indices[n]] = values[n][0];\n    }\n  }\n}\n\nexport interface CustomCallbackArgs {\n  onTrainBegin?: (logs?: Logs) => void | Promise<void>;\n  onTrainEnd?: (logs?: Logs) => void | Promise<void>;\n  onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n}\n\n/**\n * Custom callback for training.\n */\nexport class CustomCallback extends BaseCallback {\n  protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;\n  protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;\n  protected readonly epochBegin:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly epochEnd:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchBegin:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchEnd:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly yield:\n      (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n\n  private yieldEvery: YieldEveryOptions;\n  private currentEpoch = 0;\n\n  constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions) {\n    super();\n    this.yieldEvery = yieldEvery || 'auto';\n    if (this.yieldEvery === 'auto') {\n      this.yieldEvery = DEFAULT_YIELD_EVERY_MS;\n    }\n    if (this.yieldEvery === 'never' && args.onYield != null) {\n      throw new Error(\n          'yieldEvery is `never` but you provided an `onYield` callback. ' +\n          'Either change `yieldEvery` or remove the callback');\n    }\n    if (util.isNumber(this.yieldEvery)) {\n      // Decorate `maybeWait` so it will be called at most once every\n      // `yieldEvery` ms.\n      this.maybeWait = generic_utils.debounce(\n          this.maybeWait.bind(this), this.yieldEvery as number);\n    }\n    this.trainBegin = args.onTrainBegin;\n    this.trainEnd = args.onTrainEnd;\n    this.epochBegin = args.onEpochBegin;\n    this.epochEnd = args.onEpochEnd;\n    this.batchBegin = args.onBatchBegin;\n    this.batchEnd = args.onBatchEnd;\n    this.yield = args.onYield;\n  }\n\n  async maybeWait(epoch: number, batch: number, logs: UnresolvedLogs) {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.yield != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.yield(epoch, batch, logs as Logs));\n    }\n    ps.push(nextFrame());\n    await Promise.all(ps);\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    this.currentEpoch = epoch;\n    if (this.epochBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.epochBegin(epoch, logs as Logs);\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.epochEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.epochEnd(epoch, logs as Logs));\n    }\n    if (this.yieldEvery === 'epoch') {\n      ps.push(nextFrame());\n    }\n    await Promise.all(ps);\n  }\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    if (this.batchBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.batchBegin(batch, logs as Logs);\n    }\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.batchEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.batchEnd(batch, logs as Logs));\n    }\n    if (this.yieldEvery === 'batch') {\n      ps.push(nextFrame());\n    } else if (util.isNumber(this.yieldEvery)) {\n      ps.push(this.maybeWait(this.currentEpoch, batch, logs));\n    }\n    await Promise.all(ps);\n  }\n\n  async onTrainBegin(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainBegin(logs as Logs);\n    }\n  }\n\n  async onTrainEnd(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainEnd != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainEnd(logs as Logs);\n    }\n  }\n}\n\n/**\n * Standardize callbacks or configurations of them to an Array of callbacks.\n */\nexport function standardizeCallbacks(\n    callbacks: BaseCallback|BaseCallback[]|CustomCallbackArgs|\n    CustomCallbackArgs[],\n    yieldEvery: YieldEveryOptions): BaseCallback[] {\n  if (callbacks == null) {\n    callbacks = {} as BaseCallback;\n  }\n  if (callbacks instanceof BaseCallback) {\n    return [callbacks];\n  }\n  if (Array.isArray(callbacks) && callbacks[0] instanceof BaseCallback) {\n    return callbacks as BaseCallback[];\n  }\n  // Convert custom callback configs to custom callback objects.\n  const callbackConfigs =\n      generic_utils.toList(callbacks) as CustomCallbackArgs[];\n  return callbackConfigs.map(\n      callbackConfig => new CustomCallback(callbackConfig, yieldEvery));\n}\n\nexport declare type BaseCallbackConstructor = {\n  new (): BaseCallback\n};\n\n/**\n * A global registry for callback constructors to be used during\n * LayersModel.fit().\n */\nexport class CallbackConstructorRegistry {\n  private static constructors:\n      {[verbosityLevel: number]: BaseCallbackConstructor[]} = {};\n\n  /**\n   * Blocks public access to constructor.\n   */\n  private constructor() {}\n\n  /**\n   * Register a tf.LayersModel.fit() callback constructor.\n   *\n   * The registered callback constructor will be used to instantiate\n   * callbacks for every tf.LayersModel.fit() call afterwards.\n   *\n   * @param verbosityLevel Level of verbosity at which the `callbackConstructor`\n   *   is to be reigstered.\n   * @param callbackConstructor A no-arg constructor for `tf.Callback`.\n   * @throws Error, if the same callbackConstructor has been registered before,\n   *   either at the same or a different `verbosityLevel`.\n   */\n  static registerCallbackConstructor(\n      verbosityLevel: number, callbackConstructor: BaseCallbackConstructor) {\n    util.assert(\n        verbosityLevel >= 0 && Number.isInteger(verbosityLevel),\n        () => `Verbosity level is expected to be an integer >= 0, ` +\n            `but got ${verbosityLevel}`);\n    CallbackConstructorRegistry.checkForDuplicate(callbackConstructor);\n    if (CallbackConstructorRegistry.constructors[verbosityLevel] == null) {\n      CallbackConstructorRegistry.constructors[verbosityLevel] = [];\n    }\n    CallbackConstructorRegistry.constructors[verbosityLevel].push(\n        callbackConstructor);\n  }\n\n  private static checkForDuplicate(callbackConstructor:\n                                       BaseCallbackConstructor) {\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const constructors = CallbackConstructorRegistry.constructors[+levelName];\n      constructors.forEach(ctor => {\n        if (ctor === callbackConstructor) {\n          throw new ValueError('Duplicate callback constructor.');\n        }\n      });\n    }\n  }\n\n  /**\n   * Clear all registered callback constructors.\n   */\n  protected static clear() {\n    CallbackConstructorRegistry.constructors = {};\n  }\n\n  /**\n   * Create callbacks using the registered callback constructors.\n   *\n   * Given `verbosityLevel`, all constructors registered at that level or above\n   * will be called and the instantiated callbacks will be used.\n   *\n   * @param verbosityLevel: Level of verbosity.\n   */\n  static createCallbacks(verbosityLevel: number): BaseCallback[] {\n    const constructors: BaseCallbackConstructor[] = [];\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const level = +levelName;\n      if (verbosityLevel >= level) {\n        constructors.push(...CallbackConstructorRegistry.constructors[level]);\n      }\n    }\n    return constructors.map(ctor => new ctor());\n  }\n}\n\nexport function configureCallbacks(\n    callbacks: BaseCallback[], verbose: ModelLoggingVerbosity, epochs: number,\n    initialEpoch: number, numTrainSamples: number, stepsPerEpoch: number,\n    batchSize: number, doValidation: boolean,\n    callbackMetrics: string[]): {callbackList: CallbackList, history: History} {\n  const history = new History();\n  const actualCallbacks: BaseCallback[] = [\n    new BaseLogger(), ...CallbackConstructorRegistry.createCallbacks(verbose)\n  ];\n  if (callbacks != null) {\n    actualCallbacks.push(...callbacks);\n  }\n  actualCallbacks.push(history);\n  const callbackList = new CallbackList(actualCallbacks);\n\n  // TODO(cais): Figure out when this LayersModel instance can have a\n  // dynamically\n  //   set property called 'callback_model' as in PyKeras.\n\n  callbackList.setParams({\n    epochs,\n    initialEpoch,\n    samples: numTrainSamples,\n    steps: stepsPerEpoch,\n    batchSize,\n    verbose,\n    doValidation,\n    metrics: callbackMetrics,\n  });\n  return {callbackList, history};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source layers/__init__.py */\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {deserializeKerasObject} from '../utils/generic_utils';\n\n/**\n * Instantiate a layer from a config dictionary.\n * @param config dict of the form {class_name: str, config: dict}\n * @param customObjects dict mapping class names (or function names)\n *   of custom (non-Keras) objects to class/functions\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns Layer instance (may be LayersModel, Sequential, Layer...)\n */\nexport function deserialize(\n    config: serialization.ConfigDict,\n    customObjects = {} as serialization.ConfigDict,\n    fastWeightInit = false): serialization.Serializable {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'layer', fastWeightInit);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: losses.py */\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\nimport * as K from './backend/tfjs_backend';\nimport {ValueError} from './errors';\nimport {LossOrMetricFn} from './types';\n\n/**\n * Normalizes a tensor wrt the L2 norm alongside the specified axis.\n * @param x\n * @param axis Axis along which to perform normalization.\n */\nexport function l2Normalize(x: Tensor, axis?: number): Tensor {\n  return tidy(() => {\n    if (x.dtype !== 'float32') {\n      x = x.asType('float32');\n    }\n    const squareSum = tfc.sum(K.square(x), axis, true);\n    const epsilonTensor = tfc.fill(squareSum.shape, epsilon());\n    const norm = tfc.sqrt(tfc.maximum(squareSum, epsilonTensor));\n    return tfc.div(x, norm);\n  });\n}\n\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(K.square(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(tfc.abs(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const diff = tfc.sub(yTrue, yPred);\n    const clippedTrue =\n        tfc.clipByValue(tfc.abs(yTrue), epsilon(), Number.MAX_VALUE);\n    const absResult = tfc.abs(tfc.div(diff, clippedTrue));\n    return tfc.mul(100, tfc.mean(absResult, -1));\n  });\n}\n\nexport function meanSquaredLogarithmicError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), Number.MAX_VALUE);\n    const firstLog = tfc.log(tfc.add(1, clippedPred));\n\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), Number.MAX_VALUE);\n    const secondLog = tfc.log(tfc.add(1, clippedTrue));\n\n    return tfc.mean(K.square(tfc.sub(firstLog, secondLog)), -1);\n  });\n}\n\nexport function squaredHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(K.square(maxResult), -1);\n  });\n}\n\nexport function hinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(maxResult, -1);\n  });\n}\n\nexport function categoricalHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const pos = tfc.sum(tfc.mul(yTrue, yPred), -1);\n    const neg = tfc.max(tfc.mul(tfc.sub(1, yTrue), yPred), -1);\n    return tfc.maximum(0, tfc.add(1, tfc.sub(neg, pos)));\n  });\n}\n\n/**\n * Logarithm of the hyperbolic cosine of the prediction error.\n *\n * `log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and\n * to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly\n * like the mean squared error, but will not be so strongly affected by the\n * occasional wildly incorrect prediction.\n */\nexport function logcosh(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const log2 = Math.log(2);\n    const predictionDiff = tfc.sub(yPred, yTrue);\n    const logcoshResult = tfc.sub(\n        tfc.add(predictionDiff, tfc.softplus(tfc.mul(-2, predictionDiff))),\n        log2);\n    return tfc.mean(logcoshResult, -1);\n  });\n}\n\nexport function categoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    if (fromLogits) {\n      output = tfc.softmax(output);\n    } else {\n      // scale preds so that the class probabilities of each sample sum to 1.\n      const outputSum = tfc.sum(output, output.shape.length - 1, true);\n      output = tfc.div(output, outputSum);\n    }\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    return tfc.neg(tfc.sum(\n        tfc.mul(target.toFloat(), tfc.log(output)), output.shape.length - 1));\n  });\n}\n\n/**\n * Categorical crossentropy with integer targets.\n *\n * @param target An integer tensor.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n */\nexport function sparseCategoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    const flatTarget = tfc.floor(K.flatten(target)).toInt() as Tensor1D;\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    const outputShape = output.shape;\n    const oneHotTarget =\n        tfc.oneHot(flatTarget, outputShape[outputShape.length - 1])\n            .reshape(outputShape);\n    return categoricalCrossentropy(oneHotTarget, output, fromLogits);\n  });\n}\n\n/**\n * From TensorFlow's implementation in nn_impl.py:\n *\n * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n *      z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n *    = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n *    = (1 - z) * x + log(1 + exp(-x))\n *    = x - x * z + log(1 + exp(-x))\n * For x < 0, to avoid overflow in exp(-x), we reformulate the above\n *      x - x * z + log(1 + exp(-x))\n *    = log(exp(x)) - x * z + log(1 + exp(-x))\n *    = - x * z + log(1 + exp(x))\n * Hence, to ensure stability and avoid overflow, the implementation uses this\n * equivalent formulation\n *    max(x, 0) - x * z + log(1 + exp(-abs(x)))\n *\n * @param labels The labels.\n * @param logits The logits.\n */\nexport function sigmoidCrossEntropyWithLogits(\n    labels: Tensor, logits: Tensor): Tensor {\n  if (!util.arraysEqual(labels.shape, logits.shape)) {\n    throw new ValueError(\n        `logits and labels must have the same shape, but got shapes ` +\n        `${JSON.stringify(labels.shape)} and ${JSON.stringify(logits.shape)}`);\n  }\n  return tidy(() => {\n    // The logistic loss formula from above is\n    //   x - x * z + log(1 + exp(-x))\n    // For x < 0, a more numerically stable formula is\n    //   -x * z + log(1 + exp(x))\n    // Note that these two expressions can be combined into the following:\n    //   max(x, 0) - x * z + log(1 + exp(-abs(x)))\n    const reluLogits = logits.relu();\n    const negAbsLogits = logits.abs().neg();\n    return reluLogits.sub(logits.mul(labels)).add(negAbsLogits.exp().log1p());\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    let y: Tensor;\n    y = tfc.clipByValue(yPred, epsilon(), 1 - epsilon());\n    y = tfc.log(tfc.div(y, tfc.sub(1, y)));\n    return tfc.mean(sigmoidCrossEntropyWithLogits(yTrue, y), -1);\n  });\n}\n\nexport function kullbackLeiblerDivergence(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), 1);\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), 1);\n    return tfc.sum(\n        tfc.mul(yTrue, tfc.log(tfc.div(clippedTrue, clippedPred))), -1);\n  });\n}\n\nexport function poisson(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const logPred = tfc.log(tfc.add(epsilon(), yPred));\n    return tfc.mean(tfc.sub(yPred, tfc.mul(yTrue, logPred)), -1);\n  });\n}\n\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const trueNormalized = l2Normalize(yTrue, -1);\n    const predNormalized = l2Normalize(yPred, -1);\n    const trueXPred = tfc.mul(trueNormalized, predNormalized);\n    return tfc.neg(tfc.sum(trueXPred, -1));\n  });\n}\n\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const msle = meanSquaredLogarithmicError;\nexport const MSLE = meanSquaredLogarithmicError;\nexport const kld = kullbackLeiblerDivergence;\nexport const KLD = kullbackLeiblerDivergence;\nexport const cosine = cosineProximity;\n\n// TODO(michaelterry): Add deserialize() function.\n\nexport const lossesMap: {[functionName: string]: LossOrMetricFn} = {\n  meanSquaredError,\n  meanAbsoluteError,\n  meanAbsolutePercentageError,\n  meanSquaredLogarithmicError,\n  squaredHinge,\n  hinge,\n  categoricalHinge,\n  logcosh,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  binaryCrossentropy,\n  kullbackLeiblerDivergence,\n  poisson,\n  cosineProximity\n};\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function get(identifierOrFn: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifierOrFn === 'string') {\n    if (identifierOrFn in lossesMap) {\n      return lossesMap[identifierOrFn];\n    }\n    let errMsg = `Unknown loss ${identifierOrFn}`;\n    if (identifierOrFn.toLowerCase().includes('softmaxcrossentropy')) {\n      errMsg = `Unknown loss ${identifierOrFn}. ` +\n          'Use \"categoricalCrossentropy\" as the string name for ' +\n          'tf.losses.softmaxCrossEntropy';\n    }\n    throw new ValueError(errMsg);\n  } else {\n    return identifierOrFn;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Built-in metrics.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {NotImplementedError, ValueError} from './errors';\nimport {categoricalCrossentropy as categoricalCrossentropyLoss, cosineProximity, meanAbsoluteError, meanAbsolutePercentageError, meanSquaredError, sparseCategoricalCrossentropy as sparseCategoricalCrossentropyLoss} from './losses';\nimport {binaryCrossentropy as lossBinaryCrossentropy} from './losses';\nimport {lossesMap} from './losses';\nimport {LossOrMetricFn} from './types';\nimport * as util from './utils/generic_utils';\n\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const threshold = tfc.mul(.5, tfc.onesLike(yPred));\n    const yPredThresholded = K.cast(tfc.greater(yPred, threshold), yTrue.dtype);\n    return tfc.mean(tfc.equal(yTrue, yPredThresholded), -1);\n  });\n}\n\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(\n      () => K.cast(\n          tfc.equal(tfc.argMax(yTrue, -1), tfc.argMax(yPred, -1)), 'float32'));\n}\n\nfunction truePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nfunction falseNegatives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(0)).sum().cast('float32');\n  });\n}\n\nfunction falsePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(0), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fp = falsePositives(yTrue, yPred);\n\n    const denominator = tp.add(fp);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fn = falseNegatives(yTrue, yPred);\n\n    const denominator = tp.add(fn);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return lossBinaryCrossentropy(yTrue, yPred);\n}\n\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  if (yTrue.rank === yPred.rank) {\n    yTrue = yTrue.squeeze([yTrue.rank - 1]);\n  }\n  yPred = yPred.argMax(-1);\n  if (yPred.dtype !== yTrue.dtype) {\n    yPred = yPred.asType(yTrue.dtype);\n  }\n  return tfc.equal(yTrue, yPred).asType('float32');\n}\n\nexport function topKCategoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\nexport function sparseTopKCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\n// Aliases.\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const categoricalCrossentropy = categoricalCrossentropyLoss;\nexport const cosine = cosineProximity;\nexport const sparseCategoricalCrossentropy = sparseCategoricalCrossentropyLoss;\n\n// TODO(cais, nielsene): Add serialize().\n\nexport const metricsMap: {[functionName: string]: LossOrMetricFn} = {\n  binaryAccuracy,\n  categoricalAccuracy,\n  precision,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  mse,\n  MSE,\n  mae,\n  MAE,\n  mape,\n  MAPE,\n  cosine\n};\n\nexport function get(identifier: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifier === 'string' && identifier in metricsMap) {\n    return metricsMap[identifier];\n  } else if (typeof identifier !== 'string' && identifier != null) {\n    return identifier;\n  } else {\n    throw new ValueError(`Unknown metric ${identifier}`);\n  }\n}\n\n/**\n * Get the shortcut function name.\n *\n * If the fn name is a string,\n *   directly return the string name.\n * If the function is included in metricsMap or lossesMap,\n *   return key of the map.\n *   - If the function relative to multiple keys,\n *     return the first found key as the function name.\n *   - If the function exists in both lossesMap and metricsMap,\n *     search lossesMap first.\n * If the function is not included in metricsMap or lossesMap,\n *   return the function name.\n *\n * @param fn loss function, metric function, or short cut name.\n * @returns Loss or Metric name in string.\n */\nexport function getLossOrMetricName(fn: string|LossOrMetricFn): string {\n  util.assert(fn !== null, `Unknown LossOrMetricFn ${fn}`);\n  if (typeof fn === 'string') {\n    return fn;\n  } else {\n    let fnName;\n    for (const key of Object.keys(lossesMap)) {\n      if (lossesMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    for (const key of Object.keys(metricsMap)) {\n      if (metricsMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    return (fn as Function).name;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Optimizers.\n */\n\nimport {Optimizer, train} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\n\nimport {ValueError} from './errors';\n\n// Add (de)serialize()\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function getOptimizer(identifier: string): Optimizer {\n  const optimizerMap: {[optimizerName: string]: () => Optimizer} = {\n    'Adagrad': () => train.adagrad(0.01),\n    'Adadelta': () => train.adadelta(1, 0.95, epsilon()),\n    'Adam': () => train.adam(0.001, 0.9, 0.999, epsilon()),\n    'Adamax': () => train.adamax(0.002, 0.9, 0.999, epsilon(), 0),\n    'RMSProp': () => train.rmsprop(0.001, 0.9, 0, epsilon()),\n    'SGD': () => train.sgd(0.01)\n  };\n  optimizerMap['adagrad'] = optimizerMap['Adagrad'];\n  optimizerMap['adadelta'] = optimizerMap['Adadelta'];\n  optimizerMap['adam'] = optimizerMap['Adam'];\n  optimizerMap['adamax'] = optimizerMap['Adamax'];\n  optimizerMap['rmsprop'] = optimizerMap['RMSProp'];\n  optimizerMap['sgd'] = optimizerMap['SGD'];\n\n  if (identifier in optimizerMap) {\n    return optimizerMap[identifier]();\n  }\n  throw new ValueError(`Unknown Optimizer ${identifier}`);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/** Utility functions related to user-defined metadata. */\n\n// Maximum recommended serialized size for user-defined metadata.\n// Beyond this limit, a warning message will be printed during model loading and\n// saving.\nexport const MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH = 1 * 1024 * 1024;\n\n/**\n * Check validity of user-defined metadata.\n *\n * @param userDefinedMetadata\n * @param modelName Name of the model that the user-defined metadata belongs to.\n *   Used during construction of error messages.\n * @param checkSize Whether to check the size of the metadata is under\n *   recommended limit. Default: `false`. If `true`, will try stringify the\n *   JSON object and print a console warning if the serialzied size is above the\n *   limit.\n * @throws Error if `userDefinedMetadata` is not a plain JSON object.\n */\nexport function checkUserDefinedMetadata(\n    userDefinedMetadata: {}, modelName: string, checkSize = false): void {\n  if (userDefinedMetadata == null ||\n      typeof userDefinedMetadata !== 'object' ||\n      Object.getPrototypeOf(userDefinedMetadata) !== Object.prototype ||\n      !plainObjectCheck(userDefinedMetadata)) {\n    throw new Error(\n        'User-defined metadata is expected to be a JSON object, but is not.');\n  }\n\n  if (checkSize) {\n    const out = JSON.stringify(userDefinedMetadata);\n    if (out.length > MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH) {\n      console.warn(\n          `User-defined metadata of model \"${modelName}\" is too large in ` +\n          `size (length=${out.length} when serialized). It is not ` +\n          `recommended to store such large objects in user-defined metadata. ` +\n          `Please make sure its serialized length is <= ` +\n          `${MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH}.`);\n    }\n  }\n}\n\n/**\n * Check if an input is plain JSON object or any valid subfield of it.\n *\n * @param x The input to be checked.\n * @param assertObject Whether to assert `x` is a JSON object, i.e., reject\n *   cases of arrays and primitives.\n * @return Returns `true` if and only if `x` is a plain JSON object,\n *   a JSON-valid primitive including string, number, boolean and null,\n *   or an array of the said types.\n */\n// tslint:disable-next-line:no-any\nexport function plainObjectCheck(x: any): boolean {\n  if (x === null) {\n    // Note: typeof `null` is 'object', and `null` is valid in JSON.\n    return true;\n  } else if (typeof x === 'object') {\n    if (Object.getPrototypeOf(x) === Object.prototype) {\n      // `x` is a JavaScript object and its prototype is Object.\n      const keys = Object.keys(x);\n      for (const key of keys) {\n        if (typeof key !== 'string') {\n          // JSON keys must be strings.\n          return false;\n        }\n        if (!plainObjectCheck(x[key])) {  // Recursive call.\n          return false;\n        }\n      }\n      return true;\n    } else {\n      // `x` is a JavaScript object but its prototype is not Object.\n      if (Array.isArray(x)) {\n        // `x` is a JavaScript array.\n        for (const item of x) {\n          if (!plainObjectCheck(item)) {  // Recursive call.\n            return false;\n          }\n        }\n        return true;\n      } else {\n        // `x` is a JavaScript object and its prototype is not Object,\n        // and it's not an Array. I.e., it's a complex object such as\n        // `Error` and `Date`.\n        return false;\n      }\n    }\n  } else {\n    // `x` is not a JavaScript object or `null`.\n    const xType = typeof x;\n    return xType === 'string' || xType === 'number' || xType === 'boolean';\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {Container} from '../engine/container';\nimport {Layer, Node} from '../engine/topology';\nimport {countParamsInWeights} from './variable_utils';\n\n/**\n * Print the summary of a LayersModel object.\n *\n * @param model tf.LayersModel instance.\n * @param lineLength Total length of printed lines. Set this to adapt to the\n *   display to different terminal or console sizes.\n * @param positions Relative or absolute positions of log elements in each\n *   line. Each number corresponds to right-most (i.e., ending) position of a\n *   column.\n *   If not provided, defaults to `[0.45, 0.85, 1]` for sequential-like\n *   models and `[0.33, 0.55, 0.67, 1]` for non-sequential like models.\n * @param printFn Print function to use.\n *   It will be called on each line of the summary. You can provide a custom\n *   function in order to capture the string summary. Defaults to `console.log`.\n */\nexport function printSummary(\n    model: Container, lineLength?: number, positions?: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void =\n        console.log): void {\n  const sequentialLike = isModelSequentialLike(model);\n\n  // Header names for different log elements.\n  const toDisplay: string[] = ['Layer (type)', 'Output shape', 'Param #'];\n  if (sequentialLike) {\n    lineLength = lineLength || 65;\n    positions = positions || [0.45, 0.85, 1];\n  } else {\n    lineLength = lineLength || 98;\n    positions = positions || [0.33, 0.55, 0.67, 1];\n    // Header names for different log elements.\n  }\n\n  if (positions[positions.length - 1] <= 1) {\n    // `positions` is relative. Convert it to absolute positioning.\n    positions = positions.map(p => Math.floor(lineLength * p));\n  }\n\n  let relevantNodes: Node[];\n  if (!sequentialLike) {\n    toDisplay.push('Receives inputs');\n    relevantNodes = [];\n    for (const depth in model.nodesByDepth) {\n      relevantNodes.push(...model.nodesByDepth[depth]);\n    }\n  }\n\n  printFn('_'.repeat(lineLength));\n  printRow(toDisplay, positions, printFn);\n  printFn('='.repeat(lineLength));\n\n  const layers = model.layers;\n  for (let i = 0; i < layers.length; ++i) {\n    if (sequentialLike) {\n      printLayerSummary(layers[i], positions, printFn);\n    } else {\n      printLayerSummaryWithConnections(\n          layers[i], positions, relevantNodes, printFn);\n    }\n    printFn((i === layers.length - 1 ? '=' : '_').repeat(lineLength));\n  }\n\n  // tslint:disable-next-line:no-any\n  (model as any).checkTrainableWeightsConsistency();\n\n  const trainableCount = countTrainableParams(model);\n  const nonTrainableCount = countParamsInWeights(model.nonTrainableWeights);\n\n  printFn(`Total params: ${trainableCount + nonTrainableCount}`);\n  printFn(`Trainable params: ${trainableCount}`);\n  printFn(`Non-trainable params: ${nonTrainableCount}`);\n  printFn('_'.repeat(lineLength));\n}\n\nfunction countTrainableParams(model: Container): number {\n  let trainableCount: number;\n  // tslint:disable:no-any\n  if ((model as any).collectedTrainableWeights != null) {\n    trainableCount =\n        countParamsInWeights((model as any).collectedTrainableWeights);\n  } else {\n    trainableCount = countParamsInWeights(model.trainableWeights);\n  }\n  // tslint:enable:no-any\n  return trainableCount;\n}\n\nfunction isModelSequentialLike(model: Container): boolean {\n  let sequentialLike = true;\n  const nodesByDepth: Node[][] = [];\n  const nodes: Node[] = [];\n  for (const depth in model.nodesByDepth) {\n    nodesByDepth.push(model.nodesByDepth[depth]);\n  }\n  for (const depthNodes of nodesByDepth) {\n    if (depthNodes.length > 1 ||\n        depthNodes.length === 1 && depthNodes[0].inboundLayers.length > 1) {\n      sequentialLike = false;\n      break;\n    }\n    nodes.push(...depthNodes);\n  }\n  if (sequentialLike) {\n    // Search for shared layers.\n    for (const layer of model.layers) {\n      let flag = false;\n      for (const node of layer.inboundNodes) {\n        if (nodes.indexOf(node) !== -1) {\n          if (flag) {\n            sequentialLike = false;\n            break;\n          } else {\n            flag = true;\n          }\n        }\n      }\n      if (!sequentialLike) {\n        break;\n      }\n    }\n  }\n  return sequentialLike;\n}\n\nfunction printRow(\n    fields: string[], positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void = console.log) {\n  let line = '';\n  for (let i = 0; i < fields.length; ++i) {\n    if (i > 0) {\n      line = line.slice(0, line.length - 1) + ' ';\n    }\n    line += fields[i];\n    line = line.slice(0, positions[i]);\n    line += ' '.repeat(positions[i] - line.length);\n  }\n  printFn(line);\n}\n\n/**\n * Prints a summary for a single Layer, without connectivity information.\n *\n * @param layer: Layer instance to print.\n */\nfunction printLayerSummary(\n    layer: Layer, positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const name = layer.name;\n  const className = layer.getClassName();\n  const fields: string[] =\n      [`${name} (${className})`, outputShape, layer.countParams().toString()];\n  printRow(fields, positions, printFn);\n}\n\n/**\n * Prints a summary for a single Layer, with connectivity information.\n */\nfunction printLayerSummaryWithConnections(\n    layer: Layer, positions: number[], relevantNodes: Node[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const connections: string[] = [];\n  for (const node of layer.inboundNodes) {\n    if (relevantNodes != null && relevantNodes.length > 0 &&\n        relevantNodes.indexOf(node) === -1) {\n      continue;\n    }\n    for (let i = 0; i < node.inboundLayers.length; ++i) {\n      const inboundLayer = node.inboundLayers[i].name;\n      const inboundLayerIndex = node.nodeIndices[i];\n      const inboundTensorIndex = node.tensorIndices[i];\n      connections.push(\n          `${inboundLayer}[${inboundLayerIndex}][${inboundTensorIndex}]`);\n    }\n  }\n  const name = layer.name;\n  const className = layer.getClassName();\n  const firstConnection = connections.length === 0 ? '' : connections[0];\n  const fields: string[] = [\n    `${name} (${className})`, outputShape, layer.countParams().toString(),\n    firstConnection\n  ];\n\n  printRow(fields, positions, printFn);\n  for (let i = 1; i < connections.length; ++i) {\n    printRow(['', '', '', connections[i]], positions, printFn);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Porting note: This file doesn't exist in PyKeras.\n// Its purpose here is to centralize the boundary layer between\n// tfjs-layers's internal Config TS-Centric format and PyKeras's\n// serialized Python Config format.\n\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {PyJsonValue} from '../keras_format/types';\nimport * as generic_utils from '../utils/generic_utils';\n// tslint:enable\n\n/**\n * Test whether a value in an array is the name of a LayersModel or Layer.\n * @param key The key name that the value is found under. Note that the key\n *   may not be at the level immediately above the value, if the value is in a\n *   nested array.\n * @param index Index of the value in the Array that it is found in.\n * @param value The value object.\n * @returns A boolean indicating whether value is a name.\n */\nfunction isArrayItemInputOrOutputName<T>(\n    key: string, index: number, value: T): boolean {\n  return (key === 'inboundNodes' || key === 'outputLayers' ||\n          key === 'inputLayers') &&\n      index === 0 && typeof value === 'string';\n}\n\n/**\n * Convert a Pythonic config object to TypeScript config object.\n * @param pythonicConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertPythonicToTs(\n    pythonicConfig: PyJsonValue, key?: string): serialization.ConfigDictValue {\n  if (pythonicConfig === null) {\n    return null;\n  } else if (typeof pythonicConfig === 'string') {\n    return generic_utils.toCamelCase(pythonicConfig);\n  } else if (\n      (typeof pythonicConfig === 'number') ||\n      (typeof pythonicConfig === 'boolean')) {\n    return pythonicConfig;\n  } else if (pythonicConfig instanceof Array) {\n    const tsArray = [];\n    const arrayLength = pythonicConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = pythonicConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        tsArray.push(item);\n      } else {\n        tsArray.push(convertPythonicToTs(item, key));\n      }\n    }\n    return tsArray;\n  } else {\n    const tsDict: serialization.ConfigDict = {};\n    for (const pythonicKey of Object.keys(pythonicConfig)) {\n      const pythonicValue = pythonicConfig[pythonicKey];\n      if (pythonicKey === 'name' && typeof pythonicValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // camel-case conversion.\n        tsDict[pythonicKey] = pythonicValue;\n      } else {\n        const tsKey = generic_utils.toCamelCase(pythonicKey);\n        tsDict[tsKey] = convertPythonicToTs(pythonicValue, tsKey);\n      }\n    }\n    return tsDict;\n  }\n}\n\n/**\n * Convert a TypeScript config object to Python config object.\n * @param tsConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertTsToPythonic(\n    tsConfig: serialization.ConfigDictValue, key?: string): PyJsonValue {\n  if (tsConfig === null || tsConfig === undefined) {\n    return null;\n  } else if (typeof tsConfig === 'string') {\n    return generic_utils.toSnakeCase(tsConfig);\n  } else if (\n      (typeof tsConfig === 'number') || (typeof tsConfig === 'boolean')) {\n    return tsConfig;\n  } else if (tsConfig instanceof Array) {\n    const pyArray = [];\n    const arrayLength = tsConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = tsConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        pyArray.push(item);\n      } else {\n        pyArray.push(convertTsToPythonic(item, key));\n      }\n    }\n    return pyArray;\n  } else {\n    const pyDict: serialization.ConfigDict = {};\n    for (const tsKey of Object.keys(tsConfig)) {\n      const tsValue = tsConfig[tsKey];\n      const pyKey = generic_utils.toSnakeCase(tsKey);\n      if ((tsKey === 'name' || tsKey === 'className') &&\n          typeof tsValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // snake-case conversion.\n        pyDict[pyKey] = tsValue;\n      } else {\n        pyDict[pyKey] = convertTsToPythonic(tsValue, tsKey);\n      }\n    }\n    return pyDict;\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Executor: Evaluates SymbolicTensor based on feeds.\n */\n\nimport {cast, dispose, memory, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {ValueError} from '../errors';\nimport {Kwargs} from '../types';\nimport {toList} from '../utils/generic_utils';\n\nimport {InputLayer} from './input_layer';\nimport {SymbolicTensor} from './topology';\n\n/**\n * Helper function to check the dtype and shape compatibility of a feed value.\n */\nfunction assertFeedCompatibility(key: SymbolicTensor, val: Tensor): Tensor {\n  // Check dtype compatibility.\n  if (key.dtype == null || key.dtype === val.dtype) {\n    //  a.  If types match, return val tensor as is.\n    return val;\n  }\n  try {\n    //  b. Attempt to convert to expected type.\n    return cast(val, key.dtype);\n  } catch (err) {\n    //  c. If conversion fails, return helpful error.\n    throw new ValueError(\n        `The dtype of the feed (${val.dtype}) can not be cast to the dtype ` +\n        `of the key '${key.name}' (${key.dtype}).`);\n  }\n}\n\n/**\n * A concrete Tensor value for a symbolic tensor as the key.\n */\nexport interface Feed {\n  key: SymbolicTensor;\n  value: Tensor;\n}\n\n/**\n * FeedDict: A mapping from unique SymbolicTensors to feed values for them.\n * A feed value is a concrete value represented as an `Tensor`.\n */\nexport class FeedDict {\n  private id2Value: {[id: number]: Tensor} = {};\n  private id2Mask: {[id: number]: Tensor} = {};\n  private name2Id: {[name: string]: number} = {};\n\n  /**\n   * Constructor, optionally does copy-construction.\n   * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case\n   *   copy-construction will be performed.\n   */\n  constructor(feeds?: Feed[]|FeedDict) {\n    if (feeds instanceof FeedDict) {\n      for (const id in feeds.id2Value) {\n        this.id2Value[id] = feeds.id2Value[id];\n        if (id in feeds.id2Mask) {\n          this.id2Mask[id] = feeds.id2Mask[id];\n        }\n      }\n    } else {\n      if (feeds == null) {\n        return;\n      }\n      for (const feed of feeds) {\n        this.add(feed.key, feed.value);\n      }\n    }\n  }\n\n  /**\n   * Add a key-value pair to the FeedDict.\n   *\n   * @param key The key of the feed.\n   * @param value The value of the tensor feed.\n   * @param mask The value of the mask feed (optional).\n   * @returns This `FeedDict`.\n   * @throws ValueError: If the key `SymbolicTensor` already exists in the\n   *   `FeedDict`.\n   */\n  add(key: SymbolicTensor, value: Tensor, mask?: Tensor): FeedDict {\n    if (this.id2Value[key.id] == null) {\n      this.id2Value[key.id] = assertFeedCompatibility(key, value);\n      this.name2Id[key.name] = key.id;\n      if (mask != null) {\n        this.id2Mask[key.id] = mask;\n      }\n    } else {\n      throw new ValueError(`Duplicate key: name=${key.name}, id=${key.id}`);\n    }\n    return this;\n  }\n\n  /**\n   * Add a Feed to the FeedDict.\n   * @param feed The new `Feed` to add.\n   * @returns This `FeedDict`.\n   */\n  addFeed(feed: Feed) {\n    this.add(feed.key, feed.value);\n  }\n\n  /**\n   * Probe whether a key already exists in the FeedDict.\n   * @param key\n   */\n  hasKey(key: SymbolicTensor): boolean {\n    return this.id2Value[key.id] != null;\n  }\n\n  /**\n   * Get all the SymbolicTensor available in this FeedDict.\n   */\n  names(): string[] {\n    return Object.keys(this.name2Id);\n  }\n\n  /**\n   * Get the feed value for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed value.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getValue(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Value[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Value[id];\n    }\n  }\n\n  /**\n   * Get the feed mask for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed mask.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getMask(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Mask[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Mask[id];\n    }\n  }\n\n  /** Dispose all mask Tensors held by this object. */\n  disposeMasks() {\n    if (this.id2Mask != null) {\n      dispose(this.id2Mask);\n    }\n  }\n}\n\n// Cache for topologically sorted SymbolicTensors for given execution\n// targets (i.e., fetches).\nconst cachedSorted: {[concatFetchNames: string]: SymbolicTensor[]} = {};\n\n// Cache for recipient count maps for given execution targets (i.e., fetches).\nconst cachedRecipientCounts:\n    {[concatFetchNames: string]: {[fetchName: string]: number}} = {};\n\n/**\n * Interface for the optional object used for probing the memory\n * usage and other statistics during execution.\n */\nexport interface ExecutionProbe {\n  /**\n   * Maximum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  maxNumTensors?: number;\n\n  /**\n   * Minimum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  minNumTensors?: number;\n}\n\n/**\n * Execute a SymbolicTensor by using concrete feed values.\n *\n * A `SymbolicTensor` object is a node in a computation graph of TF.js\n * Layers. The object is backed by a source layer and input\n * `SymbolicTensor`s to the source layer. This method evaluates\n * the `call()` method of the source layer, using concrete values of the\n * inputs obtained from either\n * * `feedDict`, if the input key exists in `feedDict`, or else,\n * * a recursive call to `execute()` itself.\n *\n * @param x: The `SymbolicTensor` to execute.\n * @param feedDict: The feed values, as base condition of the recursion.\n *   execution.\n * @param kwargs: Optional keyword arguments.\n * @param probe: A probe object (of interface `ExecutionProbe`) used for\n *   testing memory footprint of `execute` calls.\n * @returns Result of the execution.\n * @throws ValueError: If any `SymbolicTensor`s from `InputLayer`s\n *   encountered during the execution lacks a feed value in `feedDict`.\n */\nexport function execute(\n    fetches: SymbolicTensor|SymbolicTensor[], feedDict: FeedDict,\n    kwargs?: Kwargs, probe?: ExecutionProbe): Tensor|\n    Tensor[]|[Tensor | Tensor[]] {\n  const training: boolean = kwargs == null ? false : kwargs['training'];\n\n  const arrayFetches = Array.isArray(fetches);\n  const fetchArray: SymbolicTensor[] =\n      arrayFetches ? fetches as SymbolicTensor[] : [fetches as SymbolicTensor];\n\n  const outputNames = fetchArray.map(t => t.name);\n  const finalOutputs: Tensor[] = [];\n  const feedNames = feedDict.names();\n  for (const outputName of outputNames) {\n    if (feedNames.indexOf(outputName) !== -1) {\n      finalOutputs.push(feedDict.getValue(outputName));\n    } else {\n      finalOutputs.push(null);\n    }\n  }\n\n  if (probe != null) {\n    // For optional probing of memory footprint during execution.\n    probe.maxNumTensors = -Infinity;\n    probe.minNumTensors = Infinity;\n  }\n\n  // Check cache.\n  const fetchAndFeedKey =\n      outputNames.join(',') + '|' + feedDict.names().join(',');\n  let sorted: SymbolicTensor[];\n  let recipientCounts: {[fetchName: string]: number};\n  if (cachedSorted[fetchAndFeedKey] == null) {\n    // Cache doesn't contain the desired combination of fetches. Compute\n    // topological sort for the combination for the first time.\n    const out = getTopologicalSortAndRecipientCounts(fetchArray, feedDict);\n    sorted = out.sorted;\n    recipientCounts = out.recipientCounts;\n\n    // Store results in cache for future use.\n    cachedSorted[fetchAndFeedKey] = sorted;\n    cachedRecipientCounts[fetchAndFeedKey] = recipientCounts;\n  }\n  sorted = cachedSorted[fetchAndFeedKey];\n  recipientCounts = {};\n  if (!training) {\n    Object.assign(recipientCounts, cachedRecipientCounts[fetchAndFeedKey]);\n  }\n\n  const internalFeedDict = new FeedDict(feedDict);\n\n  // Start iterative execution on the topologically-sorted SymbolicTensors.\n  for (let i = 0; i < sorted.length; ++i) {\n    if (probe != null) {\n      // For optional probing of memory usage during execution.\n      const numTensors = memory().numTensors;\n      if (numTensors > probe.maxNumTensors) {\n        probe.maxNumTensors = numTensors;\n      }\n      if (numTensors < probe.minNumTensors) {\n        probe.minNumTensors = numTensors;\n      }\n    }\n\n    const symbolic = sorted[i];\n    const srcLayer = symbolic.sourceLayer;\n    if (srcLayer instanceof InputLayer) {\n      continue;\n    }\n    const inputValues: Tensor[] = [];\n    const inputMasks: Tensor[] = [];\n    const tensorsToDispose: Tensor[] = [];\n\n    let maskExists = false;\n    for (const input of symbolic.inputs) {\n      const value = internalFeedDict.getValue(input);\n      const mask = internalFeedDict.getMask(input);\n      inputValues.push(value);\n      inputMasks.push(mask);\n      if (mask != null) {\n        maskExists = true;\n      }\n      if (!training) {\n        recipientCounts[input.name]--;\n        if (recipientCounts[input.name] === 0 && !feedDict.hasKey(input) &&\n            outputNames.indexOf(input.name) === -1 && !value.isDisposed &&\n            input.sourceLayer.stateful !== true) {\n          tensorsToDispose.push(value);\n        }\n      }\n    }\n\n    if (maskExists) {\n      kwargs = kwargs || {};\n      kwargs['mask'] = inputMasks[0];\n    }\n    const outputTensors =\n        toList(srcLayer.apply(inputValues, kwargs)) as Tensor[];\n    let outputMask: Tensor|Tensor[] = null;\n    if (srcLayer.supportsMasking) {\n      outputMask = srcLayer.computeMask(inputValues, inputMasks);\n    }\n    const layerOutputs = getNodeOutputs(symbolic);\n    const outputSymbolicTensors =\n        Array.isArray(layerOutputs) ? layerOutputs : [layerOutputs];\n    for (let i = 0; i < outputSymbolicTensors.length; ++i) {\n      if (!internalFeedDict.hasKey(outputSymbolicTensors[i])) {\n        internalFeedDict.add(\n            outputSymbolicTensors[i], outputTensors[i],\n            Array.isArray(outputMask) ? outputMask[0] : outputMask);\n      }\n      const index = outputNames.indexOf(outputSymbolicTensors[i].name);\n      if (index !== -1) {\n        finalOutputs[index] = outputTensors[i];\n      }\n    }\n\n    if (!training) {\n      // Clean up Tensors that are no longer needed.\n      dispose(tensorsToDispose);\n    }\n  }\n  // NOTE(cais): Unlike intermediate tensors, we don't discard mask\n  // tensors as we go, because these tensors are sometimes passed over a\n  // series of mutliple layers, i.e., not obeying the immediate input\n  // relations in the graph. If this becomes a memory-usage concern,\n  // we can improve this in the future.\n  internalFeedDict.disposeMasks();\n\n  return arrayFetches ? finalOutputs : finalOutputs[0];\n}\n\ntype RecipientCounts = {\n  [fetchName: string]: number\n};\n\nexport type RecipientMap = {\n  [fetchName: string]: Set<string>;\n};\n\n/**\n * Sort the `SymbolicTensor`s topologically, for an array of fetches.\n *\n * This function calls getTopologicalSortAndRecipientCountsForOneFetch and\n * merges their results.\n *\n * @param fetch The array of fetches requested. Must be a non-empty array.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientCounts: Recipient counts for all SymbolicTensors in `sorted`.\n */\nfunction getTopologicalSortAndRecipientCounts(\n    fetches: SymbolicTensor[], feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientCounts: RecipientCounts} {\n  util.assert(\n      fetches != null && fetches.length > 0,\n      () => `Expected at least one fetch, got none`);\n\n  let finalSorted: SymbolicTensor[] = [];\n  let finalRecipientMap: RecipientMap = {};\n  if (fetches.length === 1) {\n    // Special-casing 1 fetch for efficiency.\n    const out =\n        getTopologicalSortAndRecipientCountsForOneFetch(fetches[0], feedDict);\n    finalSorted = out.sorted;\n    finalRecipientMap = out.recipientMap;\n  } else {\n    const visited = new Set<string>();\n    for (const fetch of fetches) {\n      const {sorted, recipientMap} =\n          getTopologicalSortAndRecipientCountsForOneFetch(fetch, feedDict);\n\n      // Merge sorted SymbolicTensor Arrays.\n      for (const symbolicTensor of sorted) {\n        if (!visited.has(symbolicTensor.name)) {\n          finalSorted.push(symbolicTensor);\n          visited.add(symbolicTensor.name);\n        }\n      }\n\n      // Merge recipient maps.\n      for (const name in recipientMap) {\n        if (finalRecipientMap[name] == null) {\n          finalRecipientMap[name] = new Set<string>();\n        }\n        recipientMap[name].forEach(\n            recipient => finalRecipientMap[name].add(recipient));\n      }\n    }\n  }\n  return {\n    sorted: finalSorted,\n    recipientCounts: recipientMap2Counts(finalRecipientMap)\n  };\n}\n\nfunction recipientMap2Counts(recipientMap: RecipientMap): RecipientCounts {\n  const recipientCounts: RecipientCounts = {};\n  for (const name in recipientMap) {\n    recipientCounts[name] = recipientMap[name].size;\n  }\n  return recipientCounts;\n}\n\n/**\n * Sort the `SymbolicTensor`s topologically, for a single fetch.\n *\n * This helper function processes the upstream SymbolicTensors of a single\n * fetch.\n *\n * @param fetch The single fetch requested.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientMap: Recipient names for all SymbolicTensors in `sorted`.\n */\nexport function getTopologicalSortAndRecipientCountsForOneFetch(\n    fetch: SymbolicTensor, feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientMap: RecipientMap} {\n  const visited = new Set<string>();\n  const sorted: SymbolicTensor[] = [];\n  const recipientMap: RecipientMap = {};\n\n  // Put keys of the feedDict into visited first, so they don't have to be\n  // walked. This is needed in case where there are feeds for intermediate\n  // SymbolicTensors of the graph.\n  for (const key of feedDict.names()) {\n    visited.add(key);\n  }\n\n  const stack: SymbolicTensor[] = [];\n  const marks: number[] = [];\n\n  // Initial population of stack and marks.\n  stack.push(fetch);\n\n  while (stack.length > 0) {\n    const top = stack[stack.length - 1];\n    if (visited.has(top.name)) {\n      stack.pop();\n      continue;\n    }\n    const topIsMarked = marks[marks.length - 1] === stack.length - 1;\n    if (top.inputs.length === 0 || topIsMarked) {\n      // Input SymbolicTensor or all children have been visited.\n      stack.pop();\n      sorted.push(top);\n      visited.add(top.name);\n      if (topIsMarked) {\n        marks.pop();\n      }\n    } else {\n      // A non-input SymbolicTensor whose upstream SymbolicTensors haven't\n      // been visited yet. Push them onto the stack.\n      marks.push(stack.length - 1);\n      for (const input of top.inputs) {\n        // Increment the recipient count. Note that this needs to happen\n        // regardless of whether the SymbolicTensor has been visited before.\n        if (recipientMap[input.name] == null) {\n          recipientMap[input.name] = new Set<string>();\n        }\n        recipientMap[input.name].add(top.name);\n\n        if (visited.has(input.name)) {\n          continue;  // Avoid repeated visits to the same SymbolicTensor.\n        }\n        stack.push(input);\n      }\n    }\n  }\n  return {sorted, recipientMap};\n}\n\n/**\n * Get the symbolic output tensors of the node to which a given fetch belongs.\n * @param fetch The fetched symbolic tensor.\n * @returns The Array of symbolic tensors output by the node to which `fetch`\n *   belongs.\n */\nfunction getNodeOutputs(fetch: SymbolicTensor): SymbolicTensor|\n    SymbolicTensor[] {\n  let layerOutputs: SymbolicTensor|SymbolicTensor[];\n  if (fetch.sourceLayer.inboundNodes.length === 1) {\n    layerOutputs = fetch.sourceLayer.output;\n  } else {\n    let nodeIndex: number = null;\n    for (let i = 0; i < fetch.sourceLayer.inboundNodes.length; ++i) {\n      for (const outputTensor of fetch.sourceLayer.inboundNodes[i]\n               .outputTensors) {\n        if (outputTensor.id === fetch.id) {\n          nodeIndex = i;\n          break;\n        }\n      }\n    }\n    layerOutputs = fetch.sourceLayer.getOutputAt(nodeIndex);\n  }\n  return layerOutputs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {NamedTensorMap, Scalar, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {TensorKeyWithArgsArray} from '../keras_format/node_config';\nimport {PyJsonDict} from '../keras_format/types';\nimport {deserialize as deserializeLayer} from '../layers/serialization';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {convertTsToPythonic} from '../utils/serialization_utils';\nimport * as types_utils from '../utils/types_utils';\nimport {batchSetValue, LayerVariable} from '../variables';\nimport {version as layersVersion} from '../version';\n\nimport {execute, FeedDict} from './executor';\nimport {InputLayer} from './input_layer';\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/** Constructor config for Container. */\nexport interface ContainerArgs {\n  inputs: SymbolicTensor|SymbolicTensor[];\n  outputs: SymbolicTensor|SymbolicTensor[];\n  name?: string;\n}\n\n/**\n * A Container is a directed acyclic graph of layers.\n *\n * It is the topological form of a \"model\". A LayersModel\n * is simply a Container with added training routines.\n *\n */\nexport abstract class Container extends Layer {\n  inputs: SymbolicTensor[];\n  outputs: SymbolicTensor[];\n\n  inputLayers: Layer[];\n  inputLayersNodeIndices: number[];\n  inputLayersTensorIndices: number[];\n\n  outputLayers: Layer[];\n  outputLayersNodeIndices: number[];\n  outputLayersTensorIndices: number[];\n\n  layers: Layer[];\n  layersByDepth: {[depth: string]: Layer[]};\n  nodesByDepth: {[depth: string]: Node[]};\n\n  internalContainerRefs: Container[];\n\n  containerNodes = new Set<string>();\n\n  // TODO(michaelterry): Add cache support\n  // private outputMaskCache: any;\n  // private outputTensorCache: any;\n  // private outputShapeCache: any;\n\n  inputNames: string[];\n  outputNames: string[];\n  feedInputShapes: Shape[];\n\n  protected internalInputShapes: Shape[];\n  protected internalOutputShapes: Shape[];\n  // TODO(cais): Maybe 'feed' should not in the names of these variables,\n  //   due to the fact that our backend is not symbolic.\n  protected feedInputNames: string[];\n  protected feedOutputNames: string[];\n\n  constructor(args: ContainerArgs) {\n    // No args passed to super's constructor.\n    super({});\n    this.name = args.name;\n    if (this.name == null) {\n      const prefix = this.getClassName().toLowerCase();\n      this.name = getUid(prefix);\n    }\n\n    this.supportsMasking = false;\n    this.trainable_ = true;\n\n    // TODO(michaelterry): Initialize perInputLosses/Updates here.\n\n    // Container-specific properties.\n    if (Array.isArray(args.inputs)) {\n      this.inputs = args.inputs.slice();\n    } else {\n      this.inputs = [args.inputs];\n    }\n    if (Array.isArray(args.outputs)) {\n      this.outputs = args.outputs.slice();\n    } else {\n      this.outputs = [args.outputs];\n    }\n\n    // Check for redundancy in inputs.\n    if (generic_utils.unique(this.inputs).length !== this.inputs.length) {\n      throw new ValueError(\n          'The list of inputs passed to the model is ' +\n          'redundant. All inputs should only appear once. Found: ' +\n          `${this.inputs.map(x => x.name)}`);\n    }\n\n    // Check for redundancy in outputs.\n    if (generic_utils.unique(this.outputs).length !== this.outputs.length) {\n      console.warn(\n          'The list of outputs passed to the model is redundant. ' +\n          'All outputs should only appear once. Found: ' +\n          `${this.outputs.map(x => x.name)}`);\n    }\n\n    /*\n      List of initial layers (1 to 1 mapping with this.inputs, hence the same\n      layer might appear twice)\n    */\n    this.inputLayers = [];\n    this.inputLayersNodeIndices = [];\n    this.inputLayersTensorIndices = [];\n    /*\n      List of layers (1 to 1 mapping with this.outputs, hence the same layer\n      might appear twice)\n    */\n    this.outputLayers = [];\n    this.outputLayersNodeIndices = [];\n    this.outputLayersTensorIndices = [];\n    /*\n      All layers in order of horizontal graph traversal. Entries are unique.\n      Includes input and output layers.\n    */\n    this.layers = [];\n\n    /*\n      References to container layers that were constructed internally. We need\n      these to properly dispose of tensors from nested containers.\n    */\n    this.internalContainerRefs = [];\n\n    // TODO(michaelterry): Determine if caching still needed with eager\n    // backend.\n    /*\n      This is for performance optimization when calling the Container on new\n      inputs. Every time the Container is called on a set on input tensors,\n      we compute the output tensors, output masks and output shapes in one pass,\n      then cache them here. When one of these outputs is queried later,\n      we retrieve it from there instead of recomputing it.\n    */\n    // this.outputTensorCache = {};\n    // this.outputShapeCache = {};\n\n    // Build this.outputLayers:\n    for (const x of this.outputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      this.outputLayers.push(layer);\n      this.outputLayersNodeIndices.push(nodeIndex);\n      this.outputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // TODO(michaelterry): Add output mask cache code.\n\n    // Build this.inputLayers:\n    for (const x of this.inputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      /*\n        It's supposed to be an input layer, so only one node\n        and one tensor output.\n      */\n      generic_utils.assert(nodeIndex === 0, 'input layer has >1 nodes');\n      generic_utils.assert(tensorIndex === 0, 'input layer has >1 tensors');\n      this.inputLayers.push(layer);\n      this.inputLayersNodeIndices.push(nodeIndex);\n      this.inputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // Build this.inputNames and this.outputNames.\n    this.inputNames = [];\n    this.outputNames = [];\n    this.feedInputShapes = [];\n    this.feedInputNames = [];\n    this.feedOutputNames = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      // Check that layer is an InputLayer.\n      if (!(layer instanceof InputLayer)) {\n        throw new TypeError(\n            'Input layers to a LayersModel must be InputLayer objects. ' +\n            `Received inputs: ${args.inputs}. ` +\n            `Input ${i} (0-based) originates ` +\n            `from layer type ${layer.getClassName()}.`);\n      }\n      this.inputNames.push(layer.name);\n      this.feedInputShapes.push(layer.batchInputShape);\n\n      this.feedInputNames.push(layer.name);\n    }\n    for (const layer of this.outputLayers) {\n      this.outputNames.push(layer.name);\n    }\n\n    this.internalInputShapes = this.inputs.map(x => x.shape);\n    this.internalOutputShapes = this.outputs.map(x => x.shape);\n\n    /*\n      Container_nodes: set of nodes included in the graph (not all nodes\n      included in the layers are relevant to the current graph).\n    */\n    // ids of all nodes relevant to the Container:\n    const nodesDepths: {[nodeID: string]: number} = {};\n    // To recover nodes from their ID.\n    const nodeIDToNode: {[nodeID: string]: Node} = {};\n    const layersDepths: {[layerID: string]: number} = {};\n    // To layers from their ID.\n    const layerIDToLayer: {[layerID: string]: Layer} = {};\n    const layerIndices: {[layerID: string]: number} = {};\n    const nodesInDecreasingDepth: Node[] = [];\n\n    /**\n     * Builds a map of the graph of layers.\n     *\n     * This recursively updates the map `layerIndices`,\n     * the list `nodesInDecreasingDepth` and the set `containerNodes`.\n     *\n     * @param tensor Some tensor in a graph.\n     * @param finishedNodes Set of nodes whose subgraphs have been traversed\n     *         completely. Useful to prevent duplicated work.\n     * @param nodesInProgress Set of nodes that are currently active on the\n     *         recursion stack. Useful to detect cycles.\n     * @param layer Layer from which `tensor` comes from. If not provided,\n     *   will be obtained from tensor.sourceLayer.\n     * @param nodeIndex Node index from which `tensor` comes from.\n     * @param tensorIndex TensorIndex from which `tensor` comes from.\n     *\n     * @exception RuntimeError if a cycle is detected.\n     */\n    const buildMapOfGraph =\n        (tensor: SymbolicTensor, finishedNodes: Node[], nodesInProgress: Node[],\n         layer?: Layer, nodeIndex?: number, tensorIndex?: number) => {\n          if (layer == null || nodeIndex == null || tensorIndex == null) {\n            layer = tensor.sourceLayer;\n            nodeIndex = tensor.nodeIndex;\n            tensorIndex = tensor.tensorIndex;\n          }\n          const node = layer.inboundNodes[nodeIndex];\n\n          // Prevent cycles.\n          if (nodesInProgress.indexOf(node) !== -1) {\n            throw new RuntimeError(\n                `The tensor ${tensor.name} at layer \"${layer.name}\" ` +\n                'is part of a cycle.');\n          }\n\n          // Don't repeat work for shared subgraphs\n          if (finishedNodes.indexOf(node) !== -1) {\n            return;\n          }\n\n          // Update containerNodes.\n          this.containerNodes.add(Container.nodeKey(layer, nodeIndex));\n\n          // Store the traversal order for layer sorting.\n          if (!(layer.id in layerIndices)) {\n            layerIndices[layer.id] = Object.keys(layerIndices).length;\n          }\n\n          if (nodesInProgress.indexOf(node) === -1) {\n            nodesInProgress.push(node);\n          }\n\n          // Propagate to all previous tensors connected to this node.\n          const numInboundLayers = node.inboundLayers.length;\n          for (let i = 0; i < numInboundLayers; i++) {\n            const x = node.inputTensors[i];\n            const layer = node.inboundLayers[i];\n            const nodeIndex = node.nodeIndices[i];\n            const tensorIndex = node.tensorIndices[i];\n            buildMapOfGraph(\n                x, finishedNodes, nodesInProgress, layer, nodeIndex,\n                tensorIndex);\n          }\n          finishedNodes.push(node);\n          while (nodesInProgress.indexOf(node) >= 0) {\n            nodesInProgress.splice(nodesInProgress.indexOf(node), 1);\n          }\n          nodesInDecreasingDepth.push(node);\n        };\n\n    const finishedNodes: Node[] = [];\n    const nodesInProgress: Node[] = [];\n    for (const x of this.outputs) {\n      buildMapOfGraph(x, finishedNodes, nodesInProgress);\n    }\n\n    const reversedNodesInDecreasingDepth =\n        nodesInDecreasingDepth.slice().reverse();\n    for (const node of reversedNodesInDecreasingDepth) {\n      nodeIDToNode[node.id] = node;\n      // If the depth is not set, the node has no outbound nodes (depth 0).\n      if (!(node.id in nodesDepths)) {\n        nodesDepths[node.id] = 0;\n      }\n      let depth = nodesDepths[node.id];\n\n      // Update the depth of the corresponding layer\n      const previousDepth =\n          (layersDepths[node.outboundLayer.id] == null ?\n               0 :\n               layersDepths[node.outboundLayer.id]);\n\n      /*\n        If we've seen this layer before at a higher depth, we should use that\n        depth instead of the node depth.  This is necessary for shared layers\n        that have inputs at different depth levels in the graph.\n      */\n      depth = Math.max(depth, previousDepth);\n      layersDepths[node.outboundLayer.id] = depth;\n      layerIDToLayer[node.outboundLayer.id] = node.outboundLayer;\n      nodesDepths[node.id] = depth;\n\n      // Update the depth of inbound nodes.\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const inboundLayer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const inboundNode = inboundLayer.inboundNodes[nodeIndex];\n        const previousDepth =\n            (nodesDepths[inboundNode.id] == null ? 0 :\n                                                   nodesDepths[inboundNode.id]);\n        nodesDepths[inboundNode.id] = Math.max(depth + 1, previousDepth);\n        nodeIDToNode[inboundNode.id] = inboundNode;\n      }\n    }\n\n    // Build a dict {depth: list of nodes with this depth}\n    const nodesByDepth: {[depth: string]: Node[]} = {};\n    for (const nodeID in nodesDepths) {\n      const depth = nodesDepths[nodeID];\n      if (!(depth in nodesByDepth)) {\n        nodesByDepth[depth] = [];\n      }\n      nodesByDepth[depth].push(nodeIDToNode[nodeID]);\n    }\n\n    // Build a dict {depth: list of layers with this depth}\n    const layersByDepth: {[depth: string]: Layer[]} = {};\n    for (const layerID in layersDepths) {\n      const depth = layersDepths[layerID];\n      if (!(depth in layersByDepth)) {\n        layersByDepth[depth] = [];\n      }\n      layersByDepth[depth].push(layerIDToLayer[layerID]);\n    }\n\n    // Get sorted list of layer depths.\n    let depthKeys = Object.keys(layersByDepth)\n                        .map(x => parseInt(x, 10))\n                        .sort(generic_utils.reverseNumberCompare);\n\n    // Set this.layers and this.layersByDepth.\n    this.layers = [];\n    for (const depth of depthKeys) {\n      const layersForDepth = layersByDepth[depth];\n      // Container.layers needs to have a deterministic order:\n      // here we order them by traversal order.\n      layersForDepth.sort((a, b) => {\n        const aIndex = layerIndices[a.id];\n        const bIndex = layerIndices[b.id];\n        if (aIndex < bIndex) {\n          return -1;\n        }\n        if (aIndex > bIndex) {\n          return 1;\n        }\n        return 0;\n      });\n      for (const layer of layersForDepth) {\n        if (layer instanceof Container) {\n          this.internalContainerRefs.push(layer);\n        }\n        this.layers.push(layer);\n      }\n    }\n    this.layersByDepth = layersByDepth;\n\n    // Get sorted list of node depths;\n    depthKeys = Object.keys(nodesByDepth)\n                    .map(x => parseInt(x, 10))\n                    .sort(generic_utils.reverseNumberCompare);\n\n    // Check that all tensors required are computable.\n    // computable_tensors: all tensors in the graph\n    // that can be computed from the inputs provided.\n    const computableTensors = this.inputs.slice();\n\n    // To provide a better error msg.\n    const layersWithCompleteInput: string[] = [];\n    for (const depth of depthKeys) {\n      for (const node of nodesByDepth[depth]) {\n        const layer = node.outboundLayer;\n        if (layer != null) {\n          for (const x of node.inputTensors) {\n            if (computableTensors.indexOf(x) === -1) {\n              throw new RuntimeError(\n                  `Graph disconnected: cannot obtain value for tensor ${x}` +\n                  ` at layer \"${layer.name}\". ` +\n                  'The following previous layers were accessed without ' +\n                  `issue: ${layersWithCompleteInput}`);\n            }\n          }\n          for (const x of node.outputTensors) {\n            computableTensors.push(x);\n          }\n          layersWithCompleteInput.push(layer.name);\n        }\n      }\n    }\n\n    // Set this.containerNodes and this.nodesByDepth.\n    this.nodesByDepth = nodesByDepth;\n\n    // Ensure name unicity, which will be crucial for serialization\n    // (since serialized nodes refer to layers by their name).\n    const allNames = this.layers.map(x => x.name);\n    for (const name of allNames) {\n      const numOccurrences = allNames.filter(x => x === name).length;\n      if (numOccurrences !== 1) {\n        throw new RuntimeError(\n            `The name \"${name}\" is used ${numOccurrences} times ` +\n            'in the model. All layer names should be unique. Layer names: ' +\n            JSON.stringify(allNames));\n      }\n    }\n\n    // Layer parameters.\n    // The new container starts with a single inbound node\n    // for its inputs, and no outbound nodes.\n    // Will be appended to by future calls to apply().\n    this.outboundNodes = [];\n    // Will be appended to below, and by future calls to apply().\n    this.inboundNodes = [];\n\n    // Create the node linking internal inputs to internal outputs.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: this.inputs,\n      outputTensors: this.outputs,\n      inputMasks: this.inputs.map(x => null),\n      outputMasks: this.outputs.map(x => null),\n      inputShapes: this.inputs.map(x => x.shape),\n      outputShapes: this.outputs.map(x => x.shape)\n    });\n    this.built = true;\n    this._refCount = 1;  // The ref count of a container always start at 1.\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Container '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose a LayersModel's weights.\n   *\n   * This method decrease the reference count of the LayersModel object by 1.\n   *\n   * A LayersModel is reference-counted. Its reference count is incremented by 1\n   * when it is first constructed and when it is used as a Layer of another\n   * LayersModel.\n   *\n   * If the reference count of a LayersModel becomes 0, the `dispose` method of\n   * all its constituent `Layer`s will be called.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * `dispose` method of its constituent `Layer`s will *not* be called.\n   *\n   * After a LayersModel is disposed, it cannot be used in calls such as\n   * 'predict`, `evaluate` or `fit` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the LayersModel after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the LayersModel has\n   *   already been disposed.\n   */\n  dispose(): DisposeResult {\n    this.assertNotDisposed();\n    const result:\n        DisposeResult = {refCountAfterDispose: null, numDisposedVariables: 0};\n    if (--this._refCount === 0) {\n      for (const layer of this.layers) {\n        result.numDisposedVariables += layer.dispose().numDisposedVariables;\n      }\n\n      // Call dispose on each internally created container layer again to ensure\n      // their refCounts hit zero and their tensors are subsequently deleted.\n      for (const container of this.internalContainerRefs) {\n        result.numDisposedVariables += container.dispose().numDisposedVariables;\n      }\n    }\n    result.refCountAfterDispose = this._refCount;\n    return result;\n  }\n\n  get trainable() {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.layers.forEach(layer => {\n      // tslint:disable-next-line:no-any\n      ((layer as any)._trainableWeights as LayerVariable[])\n          .forEach(w => w.trainable = trainable);\n    });\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    // Porting Note: This check below is to prevent errors where the\n    //   _trainableWeights inherited from the parent class (Layer) gets\n    //   inadvertently used.\n    if (this._trainableWeights.length > 0) {\n      throw new ValueError(\n          'Container instance unexpectedly contains _trainableWeights.' +\n          'The trainable weights of a Container are a union of the ' +\n          'trainable weights of its consituent Layers. Its own ' +\n          '_trainableWeights must remain an empty Array.');\n    }\n\n    if (!this.trainable) {\n      return [];\n    }\n    let weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights = weights.concat(layer.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights.push(...layer.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const layer of this.layers) {\n        trainableWeights.push(...layer.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  /**\n   * Loads all layer weights from a JSON object.\n   *\n   * Porting Note: HDF5 weight files cannot be directly loaded in JavaScript /\n   *   TypeScript. The utility script at `scripts/pykeras.py` offers means\n   *   to convert them into JSON strings compatible with this method.\n   * Porting Note: TensorFlow.js Layers supports only loading by name currently.\n   *\n   * @param weights A JSON mapping weight names to weight values as nested\n   *   arrays of numbers, or a `NamedTensorMap`, i.e., a JSON mapping weight\n   *   names to `tf.Tensor` objects.\n   * @param strict Require that the provided weights exactly match those\n   *   required by the container.  Default: `true`.  Passing `false` means that\n   *   extra weights and missing weights will be silently ignored.\n   */\n  loadWeights(weights: NamedTensorMap, strict = true) {\n    const nameToWeight: {[name: string]: LayerVariable} = {};\n    let totalWeightsCount = 0;\n    for (const layer of this.layers) {\n      for (const weight of layer.weights) {\n        if (nameToWeight[weight.originalName] != null) {\n          throw new ValueError(`Duplicate weight name: ${weight.originalName}`);\n        }\n        nameToWeight[weight.originalName] = weight;\n        totalWeightsCount++;\n      }\n    }\n\n    const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n    for (const name in weights) {\n      // TF 2.2.0 added cell name to the weight name in the format of\n      // layer_name/cell_name/weight_name, we need to remove\n      // the inner cell name.\n      let validatedName = name;\n      if (nameToWeight[name] == null) {\n        const tokens = name.split('/');\n        const shortenNameArray =\n            tokens.slice(0, -2).concat([tokens[tokens.length - 1]]);\n        validatedName = shortenNameArray.join('/');\n      }\n      if (nameToWeight[validatedName] != null) {\n        weightValueTuples.push([nameToWeight[validatedName], weights[name]]);\n      } else if (strict) {\n        throw new ValueError(\n            `Provided weight data has no target variable: ${name}`);\n      }\n      delete nameToWeight[validatedName];\n    }\n\n    if (strict) {\n      // Check that all weights are set.\n      const unsetNames: string[] = [];\n      for (const name in nameToWeight) {\n        unsetNames.push(name);\n      }\n      if (unsetNames.length > 0) {\n        throw new ValueError(\n            `${unsetNames.length} of ${\n                totalWeightsCount} weights are not set: ` +\n            `${unsetNames}`);\n      }\n    }\n\n    batchSetValue(weightValueTuples);\n  }\n\n  /**\n   * Util shared between different serialization methods.\n   * @returns LayersModel config with Keras version information added.\n   */\n  protected updatedConfig(): serialization.ConfigDict {\n    const theConfig = this.getConfig();\n    const modelConfig: serialization.ConfigDict = {};\n    modelConfig['className'] = this.getClassName();\n    modelConfig['config'] = theConfig;\n    modelConfig['kerasVersion'] = `tfjs-layers ${layersVersion}`;\n    // TODO(nielsene): Replace something like K.backend() once\n    // possible.\n    modelConfig['backend'] = 'TensorFlow.js';\n    return modelConfig;\n  }\n\n  /**\n   * Returns a JSON string containing the network configuration.\n   *\n   * To load a network from a JSON save file, use\n   * models.modelFromJSON(jsonString);\n   * @param extraJsonArgs Unused in tfjs-layers, maintained for PyKeras\n   * @param returnString Whether the return value should be stringified\n   *    (default: `true`).\n   * @returns a JSON string if `returnString` (default), or a JSON object if\n   *   `!returnString`.\n   */\n  // tslint:disable-next-line:no-any\n  toJSON(unused?: any, returnString = true): string|PyJsonDict {\n    const modelConfig = convertTsToPythonic(this.updatedConfig()) as PyJsonDict;\n    return returnString ? JSON.stringify(modelConfig) : modelConfig;\n  }\n\n  /**\n   * Call the model on new inputs.\n   *\n   * In this case `call` just reapplies all ops in the graph to the new inputs\n   * (e.g. build a new computational graph from the provided inputs).\n   *\n   * @param inputs A tensor or list of tensors.\n   * @param mask A mask or list of masks. A mask can be either a tensor or null\n   *   (no mask).\n   *\n   * @return A tensor if there is a single output, or a list of tensors if there\n   *   are more than one outputs.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      const feedDict = new FeedDict();\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n      return execute(this.outputs, feedDict, kwargs) as Tensor | Tensor[];\n    });\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      let masks: Tensor[];\n      if (mask == null) {\n        masks = generic_utils.pyListRepeat(null, inputs.length);\n      } else {\n        masks = generic_utils.toList(mask);\n      }\n      // TODO(michaelterry): Add support for mask caching.\n      return this.runInternalGraph(inputs, masks)[1];\n    });\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    const inputShapes = types_utils.normalizeShapeList(inputShape);\n    if (inputShapes.length !== this.inputLayers.length) {\n      throw new ValueError(\n          `Invalid inputShape argument ${inputShape}: ` +\n          `model has ${this.inputLayers.length} tensor inputs.`);\n    }\n\n    // TODO(michaelterry): Add caching\n    const layersToOutputShapes: {[shapeKey: string]: Shape} = {};\n    for (let i = 0; i < inputShapes.length; i++) {\n      const layer = this.inputLayers[i];\n      const inputShape = inputShapes[i];\n      // It's an input layer: computeOutputShape is identity,\n      // and there is only one node and one tensor output.\n      const shapeKey = layer.name + '_0_0';\n      layersToOutputShapes[shapeKey] = inputShape;\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    // Iterate over nodes, by depth level.\n    if (depthKeys.length > 1) {\n      for (const depth of depthKeys) {\n        const nodes = this.nodesByDepth[depth];\n        for (const node of nodes) {\n          // This is always a single layer, never a list.\n          const layer = node.outboundLayer;\n          if (this.inputLayers.map(x => x.id).indexOf(layer.id) !== -1) {\n            // We've already covered the input layers a few lines above.\n            continue;\n          }\n          // Potentially redundant list, same size of node.inputTensors.\n          const inputShapes: Shape[] = [];\n          for (let j = 0; j < node.inboundLayers.length; j++) {\n            const inboundLayer = node.inboundLayers[j];\n            const nodeIndex = node.nodeIndices[j];\n            const tensorIndex = node.tensorIndices[j];\n            const shapeKey = `${inboundLayer.name}_${nodeIndex}_${tensorIndex}`;\n            const inputShape = layersToOutputShapes[shapeKey];\n            inputShapes.push(inputShape);\n          }\n\n          const outputShape = layer.computeOutputShape(\n              generic_utils.singletonOrArray(inputShapes));\n\n          const outputShapes = types_utils.normalizeShapeList(outputShape);\n          const nodeIndex = layer.inboundNodes.indexOf(node);\n          for (let j = 0; j < outputShapes.length; j++) {\n            const shapeKey = `${layer.name}_${nodeIndex}_${j}`;\n            layersToOutputShapes[shapeKey] = outputShapes[j];\n          }\n        }\n      }\n    }\n\n    // Read final output shapes from layersToOutputShapes.\n    const outputShapes: Shape[] = [];\n    const outputShapeKeys: string[] = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      const shapeKey = `${layer.name}_${nodeIndex}_${tensorIndex}`;\n      outputShapeKeys.push(shapeKey);\n    }\n\n    for (let i = 0; i < outputShapeKeys.length; i++) {\n      const key = outputShapeKeys[i];\n      generic_utils.assert(key in layersToOutputShapes);\n      outputShapes.push(layersToOutputShapes[key]);\n    }\n\n    // TODO(michaelterry): Update cache\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  /**\n   * Computes output tensors for new inputs.\n   *\n   * Note:\n   *   - Expects `inputs` to be a list (potentially with 1 element).\n   *\n   * @param inputs List of tensors\n   * @param masks List of masks (tensors or null).\n   * @return Three lists: outputTensors, outputMasks, outputShapes\n   */\n  protected runInternalGraph(inputs: Tensor[], masks?: Tensor[]):\n      [Tensor[], Tensor[], Shape[]] {\n    if (masks == null) {\n      masks = generic_utils.pyListRepeat(null, inputs.length);\n    }\n\n    // Dictionary mapping reference tensors to tuples\n    // (computed tensor, compute mask)\n    // we assume a 1:1 mapping from tensor to mask\n    // TODO: raise exception when a `.computeMask()` call\n    // does not return a list the same size as `call`\n    const tensorMap: {[tensorID: string]: [Tensor, Tensor]} = {};\n    for (let i = 0; i < this.inputs.length; ++i) {\n      const x = this.inputs[i];\n      const y = inputs[i];\n      const mask = masks[i];\n      tensorMap[x.id] = [y, mask];\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    for (const depth of depthKeys) {\n      const nodes = this.nodesByDepth[depth];\n      for (const node of nodes) {\n        // This is always a single layer, never a list.\n        const layer = node.outboundLayer;\n        const referenceInputTensors = node.inputTensors;\n        const referenceOutputTensors = node.outputTensors;\n\n        // If all previous input tensors are available in tensorMap,\n        // then call node.inboundLayer on them.\n        // List of tuples [input, mask]:\n        const computedData = new Array<[Tensor, Tensor]>();\n        for (const x of referenceInputTensors) {\n          if (x.id in tensorMap) {\n            computedData.push(tensorMap[x.id]);\n          }\n        }\n        if (computedData.length === referenceInputTensors.length) {\n          // TODO(michaelterry): Add K.name_scope here, if we need it.\n          let kwargs: Kwargs = {};\n          let computedTensors: Tensor[];\n          let computedMasks: Tensor[];\n          let outputTensors: Tensor[];\n          let outputMasks: Tensor[];\n          // call layer\n          if (node.callArgs != null) {\n            kwargs = node.callArgs;\n          }\n          if (computedData.length === 1) {\n            const [computedTensor, computedMask] = computedData[0];\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMask;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensor, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensor, computedMask));\n            computedTensors = [computedTensor];\n            computedMasks = [computedMask];\n          } else {\n            computedTensors = computedData.map(x => x[0]);\n            computedMasks = computedData.map(x => x[1]);\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMasks;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensors, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensors, computedMasks));\n          }\n\n          if (layer.activityRegularizer) {\n            throw new NotImplementedError(\n                'LayersModel invocation with concrete Tensor value(s) in the ' +\n                'presence of activity regularizer(s) is not supported yet.');\n          }\n          // TODO(michaelterry): Add model updates and losses\n\n          // Update tensor map.\n          for (let i = 0; i < referenceOutputTensors.length; ++i) {\n            const x = referenceOutputTensors[i];\n            const y = outputTensors[i];\n            const mask = outputMasks[i];\n            tensorMap[x.id] = [y, mask];\n          }\n        }\n      }\n    }\n\n    const outputTensors: Tensor[] = [];\n    const outputMasks: Tensor[] = [];\n    const outputShapes: Shape[] = [];\n    for (const x of this.outputs) {\n      generic_utils.assert(\n          x.id in tensorMap, `Could not compute output ${x.name} : ${x.id}`);\n      const [tensor, mask] = tensorMap[x.id];\n      outputShapes.push(tensor.shape);\n      outputTensors.push(tensor);\n      outputMasks.push(mask);\n    }\n\n    // TODO(michaelterry): Add support for caches.\n    return [outputTensors, outputMasks, outputShapes];\n  }\n\n  /**\n   * Builds a map of internal node keys to node ordering.\n   * Used in serializaion a node orderings may change as unused nodes are\n   * dropped. Porting Note:  This helper method was pulled out of getConfig to\n   * improve readability.\n   * @param layers An array of Layers in the model.\n   * @returns Map of Node Keys to index order within the layer.\n   */\n  private buildNodeConversionMap(layers: Layer[]): {[nodeKey: string]: number} {\n    const nodeConversionMap: {[nodeKey: string]: number} = {};\n    let keptNodes: number;\n    for (const layer of this.layers) {\n      keptNodes = layer instanceof Container ? 1 : 0;\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        if (this.containerNodes.has(nodeKey)) {\n          // i.e. we mark it to be saved\n          nodeConversionMap[nodeKey] = keptNodes;\n          keptNodes += 1;\n        }\n      }\n    }\n    return nodeConversionMap;\n  }\n\n  /**\n   * Retrieves a layer based on either its name (unique) or index.\n   *\n   * Indices are based on order of horizontal graph traversal (bottom-up).\n   *\n   * If both `name` and `index` are specified, `index` takes precedence.\n   *\n   * @param name Name of layer.\n   * @param index Index of layer.\n   * @returns A Layer instance.\n   * @throws ValueError: In case of invalid layer name or index.\n   *\n   * @doc {\n   *    heading: 'Layers',\n   *    subheading: 'Classes',\n   *    namespace: 'layers',\n   *    subclasses: ['LayersModel']\n   * }\n   */\n  getLayer(name?: string, index?: number): Layer {\n    if (index != null) {\n      if (this.layers.length <= index) {\n        throw new ValueError(\n            `Was asked to retrieve layer at index ${index}, but model only ` +\n            `has ${this.layers.length} layer(s).`);\n      } else {\n        return this.layers[index];\n      }\n    } else {\n      if (name == null) {\n        throw new ValueError('Provide either a layer name or layer index');\n      }\n    }\n\n    for (const layer of this.layers) {\n      if (layer.name === name) {\n        return layer;\n      }\n    }\n    throw new ValueError(`No such layer: ${name}`);\n  }\n\n  /**\n   * Retrieves the Container's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Container.loss in PyKeras.\n    //   In PyKeras, Container.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return tidy(() => {\n      const losses: Scalar[] = [];\n      for (const layer of this.layers) {\n        for (let nodeIndex = 0; nodeIndex < layer.inboundNodes.length;\n             ++nodeIndex) {\n          const nodeKey = Container.nodeKey(layer, nodeIndex);\n          if (this.containerNodes.has(nodeKey)) {\n            losses.push(...layer.calculateLosses());\n          }\n        }\n      }\n      // TODO(cais): Add any unconditional model-level losses?\n      return losses;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {name: this.name};\n\n    // Build a map from layer unique name (self._node_key)\n    // to the index of the nodes that are saved in the config.\n    // Only nodes in container_nodes are saved.\n    const nodeConversionMap: {[nodeKey: string]: number} =\n        this.buildNodeConversionMap(this.layers);\n\n    // Serialize and save the layers in layerConfigs\n    const layerConfigs = [];\n    for (const layer of this.layers) {\n      const layerClassName = layer.getClassName();\n      const layerConfig = layer.getConfig();\n      const filteredInboundNodes = [];\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const node = layer.inboundNodes[originalNodeIndex];\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        let kwargs = {};\n        if (this.containerNodes.has(nodeKey)) {\n          // The node is relevant to the model:\n          // add to filteredInboundNodes.\n          if (node.callArgs) {\n            try {\n              JSON.stringify(node.callArgs);\n              kwargs = node.callArgs;\n            } catch (err) {\n              console.warn(\n                  `Layer ${layer.name} was passed ` +\n                  `non-serializable keyword arguments: ` +\n                  `${node.callArgs}. They will not be included ` +\n                  `in the serialized model (and thus will be ` +\n                  `missing at deserialization time).`);\n              kwargs = {};\n            }\n          }\n          if (node.inboundLayers.length > 0) {\n            const nodeData = [];\n            for (let i = 0; i < node.inboundLayers.length; i++) {\n              const inboundLayer = node.inboundLayers[i];\n              const nodeIndex = node.nodeIndices[i];\n              const tensorIndex = node.tensorIndices[i];\n              const nodeKey = Container.nodeKey(inboundLayer, nodeIndex);\n              let newNodeIndex = nodeConversionMap[nodeKey];\n              if (newNodeIndex == null) {\n                newNodeIndex = 0;\n              }\n              nodeData.push(\n                  [inboundLayer.name, newNodeIndex, tensorIndex, kwargs]);\n            }\n            filteredInboundNodes.push(nodeData);\n          }\n        }\n      }\n      const dict: serialization.ConfigDict = {};\n      dict['name'] = layer.name;\n      dict['className'] = layerClassName;\n      dict['config'] = layerConfig;\n      dict['inboundNodes'] = filteredInboundNodes;\n      layerConfigs.push(dict);\n    }\n    config['layers'] = layerConfigs;\n    // Gather info about inputs and outputs\n    const modelInputs = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      const nodeIndex = this.inputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.inputLayersTensorIndices[i];\n      modelInputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['inputLayers'] = modelInputs;\n\n    const modelOutputs = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      modelOutputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['outputLayers'] = modelOutputs;\n    return config;\n  }\n\n  /**\n   * Instantiates a LayersModel from its config (output of `get_config()`).\n   * @param cls the class to create\n   * @param config LayersModel config dictionary.\n   * @param customObjects An optional dictionary of custom objects.\n   * @param fastWeightInit Optional flag to use fast weight initialization\n   *   during deserialization. This is applicable to cases in which\n   *   the initialization will be immediately overwritten by loaded weight\n   *   values. Default: `false`.\n   * @returns A LayersModel instance.\n   * @throws ValueError: In case of improperly formatted config dict.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    // Layer instances created during\n    // the graph reconstruction process\n    const createdLayers: {[layerName: string]: Layer} = {};\n\n    // Dictionary mapping layer instances to\n    // node data that specifies a layer call.\n    // It acts as a queue that maintains any unprocessed\n    // layer call until it becomes possible to process it\n    // (i.e. until the input tensors to the call all exist).\n    const unprocessedNodes: {[layer: string]: TensorKeyWithArgsArray[][]} = {};\n    function addUnprocessedNode(\n        layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      if (!(layer.name in unprocessedNodes)) {\n        unprocessedNodes[layer.name] = [nodeData];\n      } else {\n        unprocessedNodes[layer.name].push(nodeData);\n      }\n    }\n\n    function processNode(layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      const inputTensors: SymbolicTensor[] = [];\n      let kwargs;\n      for (const inputData of nodeData) {\n        const inboundLayerName = inputData[0];\n        const inboundNodeIndex = inputData[1];\n        const inboundTensorIndex = inputData[2];\n\n        kwargs = inputData[3] == null ?\n            {} :\n            inputData[3] as serialization.ConfigDict;\n        if (!(inboundLayerName in createdLayers)) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundLayer = createdLayers[inboundLayerName];\n        if (inboundLayer.inboundNodes.length <= inboundNodeIndex) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundNode = inboundLayer.inboundNodes[inboundNodeIndex];\n        inputTensors.push(inboundNode.outputTensors[inboundTensorIndex]);\n      }\n      // Call layer on its inputs, thus creating the node\n      // and building the layer if needed.\n      // Note: This has Eager vs Graph Implications.\n      if (inputTensors.length > 0) {\n        layer.apply(\n            generic_utils.singletonOrArray(inputTensors),\n            kwargs);  // was ** kwargs\n      }\n    }\n\n    /**\n     * Deserialize a layer, then call it on appropriate inputs.\n     * @param layerData: layer config dict.\n     * @throws ValueError: In case of improperly formatted `layer_data`\n     * dict.\n     */\n    function processLayer(layerData: serialization.ConfigDict|null) {\n      const layerName = layerData['name'] as string;\n      // Instantiate layer.\n      const layer =\n          deserializeLayer(\n              layerData,\n              config['customObjects'] != null ?\n                  config['customObjects'] as serialization.ConfigDict :\n                  {}) as Layer;\n      layer.setFastWeightInitDuringBuild(fastWeightInit);\n      createdLayers[layerName] = layer;\n      // Gather layer inputs.\n      const inboundNodesData =\n          layerData['inboundNodes'] as TensorKeyWithArgsArray[][];\n      inboundNodesData.forEach(nodeData => {\n        if (!(nodeData instanceof Array)) {\n          throw new ValueError(\n              `Corrupted configuration, expected array for nodeData: ${\n                  nodeData}`);\n        }\n        // We don't process nodes (i.e. make layer calls)\n        // on the fly because the inbound node may not yet exist,\n        // in case of layer shared at different topological depths\n        // (e.g.a model such as A(B(A(B(x)))))\n        addUnprocessedNode(layer, nodeData);\n      });\n    }\n\n    // First, we create all layers and enqueue nodes to be processed.\n    const name = config['name'];\n    const layersFromConfig = config['layers'] as serialization.ConfigDict[];\n    for (const layerData of layersFromConfig) {\n      processLayer(layerData);\n    }\n\n    // Then we process nodes in order of layer depth.\n    // Nodes that cannot yet be processed(if the inbound node\n    // does not yet exist) are re - enqueued, and the process\n    // is repeated until all nodes are processed.\n    while (!generic_utils.isObjectEmpty(unprocessedNodes)) {\n      for (const layerData of layersFromConfig) {\n        const layer = createdLayers[layerData['name'] as string];\n        if (layer.name in unprocessedNodes) {\n          const currentUnprocessedNodesForLayer = unprocessedNodes[layer.name];\n          delete unprocessedNodes[layer.name];\n          for (const nodeData of currentUnprocessedNodesForLayer) {\n            processNode(layer, nodeData);\n          }\n        }\n      }\n    }\n\n    const inputTensors: SymbolicTensor[] = [];\n    const outputTensors: SymbolicTensor[] = [];\n    const inputLayersFromConfig =\n        config['inputLayers'] as serialization.ConfigDict[];\n    for (const layerData of inputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      inputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    const outputLayersFromConfig =\n        config['outputLayers'] as serialization.ConfigDict[];\n    for (const layerData of outputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      outputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    return new cls({inputs: inputTensors, outputs: outputTensors, name});\n  }\n\n  /**\n   * Determine whether the container is stateful.\n   *\n   * Porting Note: this is the equivalent of the stateful @property of\n   *   the Container class in PyKeras.\n   */\n  get stateful(): boolean {\n    // Porting Note: This check is to prevent inadvertent setting of the\n    //   _stateful property of the Container instance.\n    if (this._stateful) {\n      throw new ValueError(\n          'Container instance unexpectedly has _stateful = true. The ' +\n          'statefulness of a Container is determined by the Layers it ' +\n          'contains. Its _stateful property must remain the default false.');\n    }\n    for (const layer of this.layers) {\n      if (layer.stateful) {\n        return true;\n      }\n    }\n    return false;\n  }\n\n  /**\n   * Reset the state of all stateful constituent layers (if any).\n   *\n   * Examples of stateful layers include RNN layers whose `stateful` property\n   * is set as `true`.\n   */\n  resetStates() {\n    tidy(() => {\n      this.layers.forEach(layer => {\n        // tslint:disable:no-any\n        if (layer.stateful) {\n          layer.resetStates();\n        }\n        // tslint:enable:no-any\n      });\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, mul, Tensor, Tensor1D, tensor1d, tidy} from '@tensorflow/tfjs-core';\n\n/**\n * For multi-class classification problems, this object is designed to store a\n * mapping from class index to the \"weight\" of the class, where higher weighted\n * classes have larger impact on loss, accuracy, and other metrics.\n *\n * This is useful for cases in which you want the model to \"pay more attention\"\n * to examples from an under-represented class, e.g., in unbalanced datasets.\n */\nexport type ClassWeight = {\n  [classIndex: number]: number\n};\n\n/**\n * Class weighting for a model with multiple outputs.\n *\n * This object maps each output name to a class-weighting object.\n */\nexport type ClassWeightMap = {\n  [outputName: string]: ClassWeight\n};\n\nfunction standardizeSampleOrClassWeights(\n    xWeight: ClassWeight|ClassWeight[]|ClassWeightMap, outputNames: string[],\n    weightType: 'sampleWeight'|'classWeight'): ClassWeight[] {\n  const numOutputs = outputNames.length;\n  if (xWeight == null || (Array.isArray(xWeight) && xWeight.length === 0)) {\n    return outputNames.map(name => null);\n  }\n  if (numOutputs === 1) {\n    if (Array.isArray(xWeight) && xWeight.length === 1) {\n      return xWeight;\n    } else if (typeof xWeight === 'object' && outputNames[0] in xWeight) {\n      return [(xWeight as ClassWeightMap)[outputNames[0]]];\n    } else {\n      return [xWeight as ClassWeight];\n    }\n  }\n  if (Array.isArray(xWeight)) {\n    if (xWeight.length !== numOutputs) {\n      throw new Error(\n          `Provided ${weightType} is an array of ${xWeight.length} ` +\n          `element(s), but the model has ${numOutputs} outputs. ` +\n          `Make sure a set of weights is provided for each model output.`);\n    }\n    return xWeight;\n  } else if (\n      typeof xWeight === 'object' && Object.keys(xWeight).length > 0 &&\n      typeof (xWeight as ClassWeightMap)[Object.keys(xWeight)[0]] ===\n          'object') {\n    const output: ClassWeight[] = [];\n    outputNames.forEach(outputName => {\n      if (outputName in xWeight) {\n        output.push((xWeight as ClassWeightMap)[outputName]);\n      } else {\n        output.push(null);\n      }\n    });\n    return output;\n  } else {\n    throw new Error(\n        `The model has multiple (${numOutputs}) outputs, ` +\n        `so ${weightType} must be either an array with ` +\n        `${numOutputs} elements or an object with ${outputNames} keys. ` +\n        `Provided ${weightType} not understood: ${JSON.stringify(xWeight)}`);\n  }\n}\n\n/**\n * Standardize class weighting objects.\n *\n * This function takes a single class-weighting object, an array of them,\n * or a map from output name to class-weighting object. It compares it to the\n * output name(s) of the model, base on which it outputs an array of\n * class-weighting objects of which the length matches the number of outputs.\n *\n * @param classWeight Input class-weighting object(s).\n * @param outputNames All output name(s) of the model.\n * @return An array of class-weighting objects. The length of the array matches\n *   the model's number of outputs.\n */\nexport function standardizeClassWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'classWeight');\n}\n\nexport function standardizeSampleWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'sampleWeight');\n}\n\n/**\n * Standardize by-sample and/or by-class weights for training.\n *\n * Note that this function operates on one model output at a time. For a model\n * with multiple outputs, you must call this function multiple times.\n *\n * @param y The target tensor that the by-sample and/or by-class weight is for.\n *     The values of y are assumed to encode the classes, either directly\n *     as an integer index, or as one-hot encoding.\n * @param sampleWeight By-sample weights.\n * @param classWeight By-class weights: an object mapping class indices\n *     (integers) to a weight (float) to apply to the model's loss for the\n *     samples from this class during training. This can be useful to tell the\n *     model to \"pay more attention\" to samples from an under-represented class.\n * @param sampleWeightMode The mode for the sample weights.\n * @return A Promise of weight tensor, of which the size of the first dimension\n *     matches that of `y`.\n */\nexport async function standardizeWeights(\n    y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight,\n    sampleWeightMode?: 'temporal'): Promise<Tensor> {\n  if (sampleWeight != null || sampleWeightMode != null) {\n    // TODO(cais): Once 'temporal' mode is implemented, document it in the doc\n    // string.\n    throw new Error('Support sampleWeight is not implemented yet');\n  }\n\n  if (classWeight != null) {\n    // Apply class weights per sample.\n    const yClasses: Tensor1D = tidy(() => {\n      if (y.shape.length === 1) {\n        // Assume class indices.\n        return y.clone() as Tensor1D;\n      } else if (y.shape.length === 2) {\n        if (y.shape[1] > 1) {\n          // Assume one-hot encoding of classes.\n          const axis = 1;\n          return y.argMax(axis);\n        } else if (y.shape[1] === 1) {\n          // Class index.\n          return y.reshape([y.shape[0]]);\n        } else {\n          throw new Error(\n              `Encountered unexpected last-dimension size (${y.shape[1]}) ` +\n              `during handling of class weights. The size is expected to be ` +\n              `>= 1.`);\n        }\n      } else {\n        throw new Error(\n            `Unexpected rank of target (y) tensor (${y.rank}) during ` +\n            `handling of class weights. The rank is expected to be 1 or 2.`);\n      }\n    });\n\n    const yClassIndices = Array.from(await yClasses.data());\n    dispose(yClasses);\n    const classSampleWeight: number[] = [];\n    yClassIndices.forEach(classIndex => {\n      if (classWeight[classIndex] == null) {\n        throw new Error(\n            `classWeight must contain all classes in the training data. ` +\n            `The class ${classIndex} exists in the data but not in ` +\n            `classWeight`);\n      } else {\n        classSampleWeight.push(classWeight[classIndex]);\n      }\n    });\n\n    return tensor1d(classSampleWeight, 'float32');\n  } else {\n    return null;\n  }\n}\n\n/**\n * Apply per-sample weights on the loss values from a number of samples.\n *\n * @param losses Loss tensor of shape `[batchSize]`.\n * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.\n * @returns Tensor of the same shape as`losses`.\n */\nexport function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor) {\n  return mul(losses, sampleWeights);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using TensorFlow.js datasets.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar} from '@tensorflow/tfjs-core';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {TensorOrArrayOrMap} from '../types';\nimport {singletonOrArray, toList} from '../utils/generic_utils';\n\nimport {Dataset, LazyIterator} from './dataset_stub';\nimport {ClassWeight, ClassWeightMap, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Interface for configuring model training based on a dataset object.\n */\nexport interface ModelFitDatasetArgs<T> {\n  /**\n   * (Optional) Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. It should\n   * typically be equal to the number of samples of your dataset divided by\n   * the batch size, so that `fitDataset`() call can utilize the entire dataset.\n   * If it is not provided, use `done` return value in `iterator.next()` as\n   * signal to finish an epoch.\n   */\n  batchesPerEpoch?: number;\n\n  /**\n   * Integer number of times to iterate over the training dataset.\n   */\n  epochs: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be any of the following:\n   *\n   *   - An array `[xVal, yVal]`, where the two values may be `tf.Tensor`,\n   *     an array of Tensors, or a map of string to Tensor.\n   *   - Similarly, an array ` [xVal, yVal, valSampleWeights]`\n   *     (not implemented yet).\n   *   - a `Dataset` object with elements of the form `{xs: xVal, ys: yVal}`,\n   *     where `xs` and `ys` are the feature and label tensors, respectively.\n   *\n   * If `validationData` is an Array of Tensor objects, each `tf.Tensor` will be\n   * sliced into batches during validation, using the parameter\n   * `validationBatchSize` (which defaults to 32). The entirety of the\n   * `tf.Tensor` objects will be used in the validation.\n   *\n   * If `validationData` is a dataset object, and the `validationBatches`\n   * parameter is specified, the validation will use `validationBatches` batches\n   * drawn from the dataset object. If `validationBatches` parameter is not\n   * specified, the validation will stop when the dataset is exhausted.\n   *\n   * The model will not be trained on this data.\n   */\n  validationData?: [\n    TensorOrArrayOrMap, TensorOrArrayOrMap\n  ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|Dataset<T>;\n\n  /**\n   * Optional batch size for validation.\n   *\n   * Used only if `validationData` is an array of `tf.Tensor` objects, i.e., not\n   * a dataset object.\n   *\n   * If not specified, its value defaults to 32.\n   */\n  validationBatchSize?: number;\n\n  /**\n   * (Optional) Only relevant if `validationData` is specified and is a dataset\n   * object.\n   *\n   * Total number of batches of samples to draw from `validationData` for\n   * validation purpose before stopping at the end of every epoch. If not\n   * specified, `evaluateDataset` will use `iterator.next().done` as signal to\n   * stop validation.\n   */\n  validationBatches?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - a `number`: Will yield every `number` milliseconds.\n   *   - `'never'`: never yield. (But yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n}\n\nexport interface FitDatasetElement {\n  xs: TensorOrArrayOrMap;\n  ys: TensorOrArrayOrMap;\n}\n\n/**\n * Interface for configuring model evaluation based on a dataset object.\n */\nexport interface ModelEvaluateDatasetArgs {\n  /**\n   * Number of batches to draw from the dataset object before ending the\n   * evaluation.\n   */\n  batches?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n}\n\n// Default batch size used during tensor-based validation.\nconst DEFAULT_VALIDATION_BATCH_SIZE = 32;\n\n/**\n * Standardize the output of a dataset iterator for use by\n * LayersModel.fitDataset().\n *\n * @param model: A `tf.LayersModel` object.\n * @param iteratorOut The output of a dataset iterator. It is required to be\n *   an object of the form `{xs: TensorOrArrayOrMap, ys:\n * TensorOrArrayOrMap}`, where `TensorOrArrayOrMap` is a single `tf.Tensor`,\n * a `tf.Tensor[]`, or a flat map from string names to `tf.Tensor`s.\n * @returns A flat array of `tf.Tensor` objects: the input `tf.Tensor`s\n *   followed by the target `tf.Tensor`s.  When `tf.Tensor`s are provided\n *   as a map, the order in the resulting array is taken from the `inputNames`\n *   and `outputNames` of the model.\n */\nfunction standardizeDataIteratorOutput(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, iteratorOut: {}): {xs: tfc.Tensor[], ys: tfc.Tensor[]} {\n  let xs: TensorOrArrayOrMap;\n  let ys: TensorOrArrayOrMap;\n\n  const iteratorOutObj = iteratorOut as FitDatasetElement;\n  xs = iteratorOutObj['xs'];\n  ys = iteratorOutObj['ys'];\n  tfc.util.assert(\n      xs != null && ys != null,\n      () => 'A Dataset iterator for fitDataset() is expected to generate ' +\n          'objects of the form `{xs: xVal, ys: yVal}`, where the two ' +\n          'values may be `tf.Tensor`, an array of Tensors, or a map of ' +\n          'string to Tensor.  The provided Dataset instead generates ' +\n          `${iteratorOut}`);\n\n  const flattenedXs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('input', model.inputNames, xs);\n  const flattenedYs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('output', model.outputNames, ys);\n\n  const batchSize: number = flattenedXs[0].shape[0];\n\n  tfc.util.assert(\n      flattenedXs.length === model.inputs.length,\n      () => `LayersModel has ${model.inputs.length} inputs, but the dataset ` +\n          `provides ${flattenedXs.length} inputs.  (Expected input keys: ` +\n          `${JSON.stringify(model.inputNames)})`);\n\n  tfc.util.assert(\n      flattenedYs.length === model.outputs.length,\n      () =>\n          `LayersModel has ${model.outputs.length} outputs, but the dataset ` +\n          `provides ${flattenedYs.length} outputs.  (Expected output keys: ` +\n          `${JSON.stringify(model.outputNames)})`);\n\n  for (let xIndex = 0; xIndex < flattenedXs.length; xIndex++) {\n    tfc.util.assert(\n        flattenedXs[xIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: input ` +\n            `${model.inputNames[xIndex]} has ${\n                  flattenedXs[xIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  for (let yIndex = 0; yIndex < flattenedYs.length; yIndex++) {\n    tfc.util.assert(\n        flattenedYs[yIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: output ` +\n            `${model.outputNames[yIndex]} has ${\n                  flattenedYs[yIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  return {xs: flattenedXs, ys: flattenedYs};\n}\n\nfunction flattenTensorOrArrayOrMap(\n    inputOrOutput: string, names: string[], values: TensorOrArrayOrMap) {\n  if (values instanceof tfc.Tensor) {\n    return [values];\n  } else if (Array.isArray(values)) {\n    tfc.util.assert(\n        values.length === names.length,\n        () => `Received an array of ${values.length} Tensors, but expected ${\n            names.length} to match the ${inputOrOutput} keys ${names}.`);\n    return values;\n  } else {\n    const result: tfc.Tensor[] = [];\n    // Check that all the required keys are available.\n    for (const name of names) {\n      if (values[name] == null) {\n        throw new ValueError(\n            `The feature data generated by the dataset lacks the required ` +\n            `${inputOrOutput} key '${name}'.`);\n      }\n      result.push(values[name]);\n    }\n    return result;\n  }\n}\n\nfunction standardizeTensorValidationData<T>(\n    data:\n        [\n          tfc.Tensor|tfc.Tensor[], tfc.Tensor|tfc.Tensor[]\n        ]|[tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n           tfc.Tensor | tfc.Tensor[]]):\n    {xs: tfc.Tensor|tfc.Tensor[], ys: tfc.Tensor|tfc.Tensor[]} {\n  if (data.length === 3) {\n    throw new NotImplementedError(\n        'Validation with sample weights is not implemented yet.');\n  }\n  return {xs: data[0], ys: data[1]};\n}\n\nexport async function fitDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>,\n    args: ModelFitDatasetArgs<T>): Promise<History> {\n  const hasBatchesPerEpoch = args.batchesPerEpoch != null;\n  tfc.util.assert(\n      model.optimizer != null,\n      () => 'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileConfig).');\n\n  tfc.util.assert(\n      args != null,\n      () => `For fitDataset(), the 2nd argument (config) is required, ` +\n          `but it is not provided in this call.`);\n  tfc.util.assert(\n      args.epochs != null && args.epochs > 0 && Number.isInteger(args.epochs),\n      () => `For fitDataset(), config.epochs is expected to be a positive ` +\n          `integer, but got ${args.epochs}`);\n  tfc.util.assert(\n      !hasBatchesPerEpoch ||\n          (args.batchesPerEpoch > 0 && Number.isInteger(args.batchesPerEpoch)),\n      () => `For fitDataset(), config.batchesPerEpoch is expected to be a ` +\n          `positive integer if specified, but got ${args.batchesPerEpoch}`);\n  tfc.util.assert(\n      // tslint:disable-next-line:no-any\n      (args as any)['validationSplit'] == null,\n      () => '`validationSplit` is not supported by `fitDataset()`. ' +\n          'Use validationData instead.');\n\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n\n  try {\n    const doValidation = args.validationData != null;\n    let valXs: tfc.Tensor|tfc.Tensor[];\n    let valYs: tfc.Tensor|tfc.Tensor[];\n    if (doValidation) {\n      if (isDatasetObject(args.validationData)) {\n        tfc.util.assert(\n            args.validationBatches == null ||\n                (args.validationBatches > 0 &&\n                 Number.isInteger(args.validationBatches)),\n            () => `For fitDataset() with dataset-based validation, ` +\n                `config.validationBatches is expected not to be provided, ` +\n                `or to be a positive integer, ` +\n                `but got ${args.validationBatches}`);\n      } else {\n        const validationData = standardizeTensorValidationData(\n            args.validationData as\n                    [tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[]] |\n            [\n              tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n              tfc.Tensor | tfc.Tensor[]\n            ]);\n        valXs = validationData.xs;\n        valYs = validationData.ys;\n      }\n    }\n\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let callbackMetrics: string[];\n    if (doValidation) {\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const verbose = args.verbose == null ? 1 : args.verbose;\n    const {callbackList, history} = configureCallbacks(\n        callbacks, verbose, args.epochs, null, null,\n        getStepsPerEpoch(dataset, args),\n        null,  // Batch size determined by the dataset itself.\n        doValidation, callbackMetrics);\n    callbackList.setModel(model);\n    model.history = history;\n\n    await callbackList.onTrainBegin();\n    model.stopTraining_ = false;\n    let epoch = args.initialEpoch == null ? 0 : args.initialEpoch;\n\n    let dataIterator = await dataset.iterator();\n    while (epoch < args.epochs) {\n      const epochLogs: UnresolvedLogs = {};\n      await callbackList.onEpochBegin(epoch);\n      let stepsDone = 0;\n      let batchIndex = 0;\n      if (!hasBatchesPerEpoch) {\n        dataIterator = await dataset.iterator();\n      }\n      while (hasBatchesPerEpoch ? stepsDone < args.batchesPerEpoch : true) {\n        const iteratorOut = await dataIterator.next();\n\n        // If `batchesPerEpoch` is specified, the dataset should not be\n        // exhausted until all epoches are done.\n        if (hasBatchesPerEpoch && iteratorOut.done) {\n          console.warn(\n              'You provided `batchesPerEpoch` as ' +\n              `${args.batchesPerEpoch}, ` +\n              'but your dataset iterator ran out of data after ' +\n              `${stepsDone} batches; ` +\n              'interrupting training. Make sure that your ' +\n              'dataset can generate at least `batchesPerEpoch * epochs` ' +\n              'batches (in this case, ' +\n              `${args.batchesPerEpoch * args.epochs} batches). ` +\n              'You may need to use the repeat() function when building ' +\n              'your dataset.');\n          break;\n        }\n\n        if (iteratorOut.value != null) {\n          const {xs, ys} =\n              standardizeDataIteratorOutput(model, iteratorOut.value);\n          const batchLogs: UnresolvedLogs = {};\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = xs[0].shape[0];\n\n          await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n          const sampleWeights: tfc.Tensor[] = [];\n          if (args.classWeight != null) {\n            const standardClassWeights =\n                standardizeClassWeights(args.classWeight, model.outputNames);\n            for (let i = 0; i < standardClassWeights.length; ++i) {\n              sampleWeights.push(await standardizeWeights(\n                  ys[i], null, standardClassWeights[i]));\n            }\n          }\n\n          // Train on batch.\n          const ins = xs.concat(ys).concat(sampleWeights);\n          const outs = trainFunction(ins);\n          tfc.dispose(ins);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n          }\n\n          await callbackList.onBatchEnd(batchIndex, batchLogs);\n          disposeTensorsInLogs(batchLogs);\n\n          batchIndex++;\n          stepsDone++;\n        }\n\n        if (hasBatchesPerEpoch ? stepsDone >= args.batchesPerEpoch :\n                                 iteratorOut.done) {\n          // Epoch finished. Perform validation.\n          if (doValidation) {\n            let valOuts: tfc.Scalar[];\n            if (isDatasetObject(args.validationData)) {\n              valOuts = toList(await model.evaluateDataset(\n                  args.validationData, {batches: args.validationBatches}));\n            } else {\n              valOuts = toList(model.evaluate(valXs, valYs, {\n                batchSize: args.validationBatchSize == null ?\n                    DEFAULT_VALIDATION_BATCH_SIZE :\n                    args.validationBatchSize,\n                verbose: 0\n              }));\n            }\n            for (let i = 0; i < model.metricsNames.length; ++i) {\n              epochLogs[`val_${model.metricsNames[i]}`] = valOuts[i];\n            }\n          }\n          // Call `break` to exit one epoch lopp after validation is done. If\n          // config.batchesPerEpoch is specified, an epoch while loop will\n          // stop when `stepsDone >= config.batchesPerEpoch`. When\n          // config.batchesPerEpoch is not provided, the following `break` is\n          // required to exit the while lopp after dataset is exhausted.\n          break;\n        }\n\n        if (model.stopTraining_) {\n          break;\n        }\n      }\n      await callbackList.onEpochEnd(epoch, epochLogs);\n      epoch++;\n      if (model.stopTraining_) {\n        break;\n      }\n    }\n    await callbackList.onTrainEnd();\n    await model.history.syncData();\n    return model.history;\n  } finally {\n    model.isTraining = false;\n  }\n}\n\n/** Helper function that determines number of steps (batches) per epoch. */\nfunction getStepsPerEpoch<T>(\n    dataset: Dataset<T>, args: ModelFitDatasetArgs<T>): number {\n  // Attempt to determine # of batches in an epoch.\n  let stepsPerEpoch: number = null;\n  if (args.batchesPerEpoch != null) {\n    stepsPerEpoch = args.batchesPerEpoch;\n  } else if (Number.isFinite(dataset.size)) {\n    stepsPerEpoch = dataset.size;\n  }\n  return stepsPerEpoch;\n}\n\n// Check if provided object is a Dataset object by checking its .iterator\n// element.\nfunction isDatasetObject<T>(\n    dataset:\n        [\n          TensorOrArrayOrMap, TensorOrArrayOrMap\n        ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|\n    Dataset<T>): boolean {\n  return (typeof (dataset as Dataset<T>).iterator === 'function');\n}\n\n// Check if provided object is a LazyIterator object by checking it's .next\n// element.\nfunction isLazyIteratorObject<T>(iterator: Dataset<T>|\n                                 LazyIterator<T>): boolean {\n  return (typeof (iterator as LazyIterator<T>).next === 'function');\n}\n\nexport async function evaluateDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>|LazyIterator<T>,\n    args: ModelEvaluateDatasetArgs): Promise<tfc.Scalar|tfc.Scalar[]> {\n  args = args || {};\n  const hasBatches = args.batches != null;\n  const f = model.testFunction;\n  let outs: tfc.Scalar[] = [];\n  if (args.verbose > 0) {\n    throw new NotImplementedError('Verbose mode is not implemented yet.');\n  }\n\n  tfc.util.assert(\n      !hasBatches || (args.batches > 0 && Number.isInteger(args.batches)),\n      () => 'Test loop expects `batches` to be a positive integer, but ' +\n          `received ${JSON.stringify(args.batches)}`);\n  const dataIterator = isLazyIteratorObject(dataset) ?\n      dataset as LazyIterator<T>:\n      await (dataset as Dataset<T>).iterator();\n  // Keeps track of number of examples used in this evaluation.\n  let numExamples = 0;\n  let batch = 0;\n\n  while (hasBatches ? batch < args.batches : true) {\n    const iteratorOut = await dataIterator.next();\n    outs = tfc.tidy(() => {\n      if (iteratorOut.value) {\n        // TODO(cais): Once real dataset is available, use\n        //   `map(x => standardizeDataIteratorOutput(model, x).map(f)`.\n        const {xs, ys} =\n            standardizeDataIteratorOutput(model, iteratorOut.value);\n        const xsAndYs = xs.concat(ys);\n        const batchOuts = tfc.tidy(() => f(xsAndYs));\n        tfc.dispose(xsAndYs);\n\n        if (batch === 0) {\n          for (let i = 0; i < batchOuts.length; ++i) {\n            outs.push(scalar(0));\n          }\n        }\n\n        const batchSize = xsAndYs[0].shape[0];\n        for (let i = 0; i < batchOuts.length; ++i) {\n          const batchOut = batchOuts[i];\n          const oldScalar = outs[i];\n          outs[i] =\n              tfc.tidy(() => tfc.add(outs[i], tfc.mul(batchSize, batchOut)));\n          if (batch > 0) {\n            tfc.dispose(oldScalar);\n          }\n        }\n        tfc.dispose(batchOuts);\n        numExamples += batchSize;\n\n        ++batch;\n      }\n      return outs;\n    });\n\n    if (iteratorOut.done) {\n      if (hasBatches) {\n        console.warn(\n            'Your dataset iterator ran out of data during evaluateDataset(). ' +\n            'Interrupting evalution. Make sure that your ' +\n            'dataset can generate at least `batches` ' +\n            `batches (in this case, ${args.batches} batches). ` +\n            'You may need to use the repeat() function when building ' +\n            'your dataset.');\n      }\n      break;\n    }\n  }\n\n  for (let i = 0; i < outs.length; ++i) {\n    const oldScalar = outs[i];\n    outs[i] = tfc.div(outs[i], numExamples);\n    tfc.dispose(oldScalar);\n  }\n\n  return singletonOrArray(outs);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using tf.Tensor objects.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Scalar, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport {expandDims, gather, sliceAlongFirstAxis} from '../backend/tfjs_backend';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {range} from '../utils/math_utils';\nimport {ClassWeight, ClassWeightMap} from './training_utils';\n\n/**\n * Interface configuration model training based on data as `tf.Tensor`s.\n */\nexport interface ModelFitArgs {\n  /**\n   * Number of samples per gradient update. If unspecified, it\n   * will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Integer number of times to iterate over the training data arrays.\n   */\n  epochs?: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Float between 0 and 1: fraction of the training data\n   * to be used as validation data. The model will set apart this fraction of\n   * the training data, will not train on it, and will evaluate the loss and\n   * any model metrics on this data at the end of each epoch.\n   * The validation data is selected from the last samples in the `x` and `y`\n   * data provided, before shuffling.\n   */\n  validationSplit?: number;\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal,\n   * valSampleWeights]. The model will not be trained on this data.\n   * `validationData` will override `validationSplit`.\n   */\n  validationData?: [\n    Tensor|Tensor[], Tensor|Tensor[]\n  ]|[Tensor | Tensor[], Tensor|Tensor[], Tensor|Tensor[]];\n\n  /**\n   * Whether to shuffle the training data before each epoch. Has\n   * no effect when `stepsPerEpoch` is not `null`.\n   */\n  shuffle?: boolean;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n\n  /**\n   * Optional array of the same length as x, containing\n   * weights to apply to the model's loss for each sample. In the case of\n   * temporal data, you can pass a 2D array with shape (samples,\n   * sequenceLength), to apply a different weight to every timestep of every\n   * sample. In this case you should make sure to specify\n   * sampleWeightMode=\"temporal\" in compile().\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. When training\n   * with Input Tensors such as TensorFlow data tensors, the default `null` is\n   * equal to the number of unique samples in your dataset divided by the\n   * batch size, or 1 if that cannot be determined.\n   */\n  stepsPerEpoch?: number;\n\n  /**\n   * Only relevant if `stepsPerEpoch` is specified. Total number of steps\n   * (batches of samples) to validate before stopping.\n   */\n  validationSteps?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - any `number`: yield every `number` milliseconds.\n   *   - `'never'`: never yield. (yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n}\n\nexport function checkBatchSize(batchSize: number) {\n  tfc.util.assert(\n      batchSize > 0 && Number.isInteger(batchSize),\n      () => `batchSize is required to be a positive integer, but got ${\n          batchSize}`);\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by start and stop indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArraysByIndices()` together.\n *\n * @param arrays: the input.\n * @param start: the starting index (inclusive).\n * @param stop: the stopping index (exclusive).\n * @returns The result of the slicing. If `arrays` is an `Array` of\n *   `tf.Tensor`s, the slicing will be applied to all elements of the `Array`\n *   in the same way.\n */\nexport function sliceArrays(\n    arrays: Tensor|Tensor[], start: number, stop: number): Tensor|Tensor[] {\n  if (arrays == null) {\n    return [null];\n  } else if (Array.isArray(arrays)) {\n    return arrays.map(array => sliceAlongFirstAxis(array, start, stop - start));\n  } else {  // Tensor.\n    return sliceAlongFirstAxis(arrays, start, stop - start);\n  }\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by random-order indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArrays()` together.\n *\n * @param arrays The input `tf.Tensor` or `Array` of `tf.Tensor`s to slice.\n *   If an `Array` of `tf.Tensor`s, all `tf.Tensor`s will be sliced in the\n *   same fashion.\n * @param indices The indices to use for slicing along the first (batch)\n *   dimension.\n * @returns Result(s) of the slicing.\n */\nexport function sliceArraysByIndices(\n    arrays: Tensor|Tensor[], indices: Tensor1D): Tensor|Tensor[] {\n  return tfc.tidy(() => {\n    if (arrays == null) {\n      return null;\n    } else if (Array.isArray(arrays)) {\n      return arrays.map(\n          array => (sliceArraysByIndices(array, indices) as Tensor));\n    } else {\n      // TODO(cais): indices should be a pre-constructed Tensor1D to avoid\n      //   tensor1d() calls.\n      return gather(\n          arrays, indices.dtype === 'int32' ? indices : indices.toInt());\n    }\n  });\n}\n\n/**\n * Returns a list of batch indices (tuples of indices).\n * @param size: Integer, total size of the data to slice into batches.\n * @param batchSize: Integer, batch size.\n * @returns An Array of [batchStart, batchEnd] tuples. batchStart is\n *   inclusive; batchEnd is exclusive. I.e., each batch consists of indices x\n *   that satisfy batchStart <= x < batchEnd.\n */\nexport function makeBatches(\n    size: number, batchSize: number): Array<[number, number]> {\n  const output: Array<[number, number]> = [];\n  let batchStart = 0;\n  let batchEnd: number = null;\n  while (batchStart < size) {\n    batchEnd = batchStart + batchSize;\n    if (batchEnd >= size) {\n      batchEnd = size;\n    }\n    output.push([batchStart, batchEnd]);\n    batchStart = batchEnd;\n  }\n  return output;\n}\n\n/**\n * Abstract fit function for `f(ins)`.\n * @param f A Function returning a list of tensors. For training, this\n *   function is expected to perform the updates to the variables.\n * @param ins List of tensors to be fed to `f`.\n * @param outLabels List of strings, display names of the outputs of `f`.\n * @param batchSize Integer batch size or `== null` if unknown. Default : 32.\n * @param epochs Number of times to iterate over the data. Default : 1.\n * @param verbose Verbosity mode: 0, 1, or 2. Default: 1.\n * @param callbacks List of callbacks to be called during training.\n * @param valF Function to call for validation.\n * @param valIns List of tensors to be fed to `valF`.\n * @param shuffle Whether to shuffle the data at the beginning of every\n * epoch. Default : true.\n * @param callbackMetrics List of strings, the display names of the metrics\n *   passed to the callbacks. They should be the concatenation of the\n *   display names of the outputs of `f` and the list of display names\n *   of the outputs of `valF`.\n * @param initialEpoch Epoch at which to start training (useful for\n *   resuming a previous training run). Default : 0.\n * @param stepsPerEpoch Total number of steps (batches on samples) before\n *   declaring one epoch finished and starting the next epoch. Ignored with\n *   the default value of `undefined` or `null`.\n * @param validationSteps Number of steps to run validation for (only if\n *   doing validation from data tensors). Not applicable for tfjs-layers.\n * @returns A `History` object.\n */\nasync function fitLoop(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, f: (data: Tensor[]) => Scalar[], ins: Tensor[],\n    outLabels?: string[], batchSize?: number, epochs?: number, verbose?: number,\n    callbacks?: BaseCallback[], valF?: (data: Tensor[]) => Scalar[],\n    valIns?: Tensor[], shuffle?: boolean|string, callbackMetrics?: string[],\n    initialEpoch?: number, stepsPerEpoch?: number,\n    validationSteps?: number): Promise<History> {\n  if (batchSize == null) {\n    batchSize = 32;\n  }\n  if (epochs == null) {\n    epochs = 1;\n  }\n  if (shuffle == null) {\n    shuffle = true;\n  }\n  if (initialEpoch == null) {\n    initialEpoch = 0;\n  }\n\n  // TODO(cais): Change const to let below when implementing validation.\n  let doValidation = false;\n  if (valF != null && valIns != null) {\n    doValidation = true;\n    // TODO(cais): verbose message.\n  }\n  if (validationSteps != null) {\n    doValidation = true;\n    if (stepsPerEpoch == null) {\n      throw new ValueError(\n          'Can only use `validationSteps` when doing step-wise training, ' +\n          'i.e., `stepsPerEpoch` must be set.');\n    }\n  }\n\n  const numTrainSamples =\n      model.checkNumSamples(ins, batchSize, stepsPerEpoch, 'steps_per_epoch');\n  let indexArray: number[];\n  if (numTrainSamples != null) {\n    indexArray = range(0, numTrainSamples);\n  }\n\n  if (verbose == null) {\n    verbose = 1;\n  }\n\n  const {callbackList, history} = configureCallbacks(\n      callbacks, verbose, epochs, initialEpoch, numTrainSamples, stepsPerEpoch,\n      batchSize, doValidation, callbackMetrics);\n  callbackList.setModel(model);\n  model.history = history;\n  await callbackList.onTrainBegin();\n  model.stopTraining_ = false;\n  // TODO(cais): Take care of callbacks.validation_data as in PyKeras.\n  // TODO(cais): Pre-convert feeds for performance as in PyKeras.\n\n  for (let epoch = initialEpoch; epoch < epochs; ++epoch) {\n    await callbackList.onEpochBegin(epoch);\n    const epochLogs: UnresolvedLogs = {};\n    if (stepsPerEpoch != null) {\n      throw new NotImplementedError(\n          'stepsPerEpoch mode is not implemented yet.');\n    } else {\n      if (shuffle === 'batch') {\n        throw new NotImplementedError('batch shuffling is not implemneted yet');\n      } else if (shuffle) {\n        util.shuffle(indexArray);\n      }\n      // Convert the potentially shuffled indices to Tensor1D, to avoid the\n      // cost of repeated creation of Array1Ds later on.\n      const epochIndexArray1D = tensor1d(indexArray);\n\n      const batches = makeBatches(numTrainSamples, batchSize);\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchLogs: UnresolvedLogs = {};\n        await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n        tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds = sliceAlongFirstAxis(\n                               epochIndexArray1D, batchStart,\n                               batchEnd - batchStart) as Tensor1D;\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = batchEnd - batchStart;\n\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Tensor[];\n          const outs = f(insBatch);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n            // TODO(cais): Use scope() to avoid ownership.\n          }\n\n          if (batchIndex === batches.length - 1) {  // Last batch.\n            if (doValidation) {\n              const valOuts = model.testLoop(valF, valIns, batchSize);\n              // Porting Notes: In tfjs-layers, valOuts is always an Array.\n              for (let i = 0; i < outLabels.length; ++i) {\n                const label = outLabels[i];\n                const out = valOuts[i];\n                tfc.keep(out);\n                // TODO(cais): Use scope() to avoid ownership.\n                epochLogs['val_' + label] = out;\n              }\n            }\n          }\n        });\n\n        await callbackList.onBatchEnd(batchIndex, batchLogs);\n        disposeTensorsInLogs(batchLogs);\n\n        if (model.stopTraining_) {\n          break;\n        }\n        // TODO(cais): return outs as list of Tensor.\n      }\n\n      epochIndexArray1D.dispose();\n    }\n    // TODO(cais): Run validation at the end of the epoch.\n    await callbackList.onEpochEnd(epoch, epochLogs);\n    if (model.stopTraining_) {\n      break;\n    }\n  }\n  await callbackList.onTrainEnd();\n\n  await model.history.syncData();\n  return model.history;\n}\n\nexport async function fitTensors(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    args: ModelFitArgs = {}): Promise<History> {\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n  let inputs: Tensor[];\n  let targets: Tensor[];\n  let inputValX: Tensor|Tensor[];\n  let inputValY: Tensor|Tensor[];\n  let valX: Tensor|Tensor[];\n  let valY: Tensor|Tensor[];\n  let sampleWeights: Tensor[];\n  try {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // Validate user data.\n    // TODO(cais): Support sampleWeight.\n    const checkBatchAxis = false;\n    const standardizedOuts =\n        await model.standardizeUserData(\n            x, y, args.sampleWeight, args.classWeight, checkBatchAxis,\n            batchSize) as [Tensor[], Tensor[], Tensor[]];\n    inputs = standardizedOuts[0];\n    targets = standardizedOuts[1];\n    sampleWeights = standardizedOuts[2];\n\n    // Prepare validation data.\n    let doValidation = false;\n    let valIns: Tensor[];\n    if (args.validationData != null && args.validationData.length > 0) {\n      doValidation = true;\n      if (args.validationData.length === 2) {\n        // config.validationData consists of valX and valY.\n        inputValX = args.validationData[0];\n        inputValY = args.validationData[1];\n      } else if (args.validationData.length === 3) {\n        throw new NotImplementedError(\n            'validationData including sample weights is not supported yet.');\n      } else {\n        throw new ValueError(\n            `When passing validation data, it must contain 2 (valX, valY) ` +\n            `or 3 (valX, valY, valSampleWeight) items; ` +\n            `${args.validationData} is invalid.`);\n      }\n\n      const checkBatchAxis = true;\n      const valStandardized =\n          await model.standardizeUserData(\n              inputValX, inputValY, null, /** Unused sample weights. */\n              null,                       /** Unused class weights. */\n              checkBatchAxis, batchSize) as [Tensor[], Tensor[], Tensor[]];\n      valX = valStandardized[0];\n      valY = valStandardized[1];\n      valIns = valX.concat(valY);\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (\n        args.validationSplit != null && args.validationSplit > 0 &&\n        args.validationSplit < 1) {\n      doValidation = true;\n      // Porting Note: In tfjs-layers, inputs[0] is always a Tensor.\n      const splitAt =\n          Math.floor(inputs[0].shape[0] * (1 - args.validationSplit));\n      const originalBatchSize = inputs[0].shape[0];\n      valX = sliceArrays(inputs, splitAt, originalBatchSize) as Tensor[];\n      inputs = sliceArrays(inputs, 0, splitAt) as Tensor[];\n      valY = sliceArrays(targets, splitAt, originalBatchSize) as Tensor[];\n      targets = sliceArrays(targets, 0, splitAt) as Tensor[];\n      // TODO(cais): Once sampleWeights becomes available, slice it to get\n      //   valSampleWeights.\n      valIns = valX.concat(valY);\n\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (args.validationSteps != null) {\n      doValidation = true;\n      // TODO(cais): Add useLearningPhase.\n    }\n\n    const ins = inputs.concat(targets).concat(sampleWeights);\n\n    model.checkTrainableWeightsConsistency();\n\n    // TODO(cais): Handle use_learning_phase and learning_phase?\n\n    // Porting Note: Here we see a key deviation of tfjs-layers from\n    // Keras.\n    //  Due to the imperative nature of tfjs-layers' backend (tfjs-core),\n    //  we do not construct symbolic computation graphs to embody the\n    //  training process. Instead, we define a function that performs the\n    //  training action. In PyKeras, the data (inputs and targets) are fed\n    //  through graph placeholders. In tfjs-layers, the data are fed as\n    //  function arguments. Since the function are defined below in the\n    //  scope, we don't have equivalents of PyKeras's\n    //  `_make_train_funciton`.\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let valFunction: (data: Tensor[]) => Scalar[];\n    let callbackMetrics: string[];\n    if (doValidation) {\n      model.makeTestFunction();\n      valFunction = model.testFunction;\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      valFunction = null;\n      valIns = [];\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const out = await fitLoop(\n        model, trainFunction, ins, outLabels, batchSize, args.epochs,\n        args.verbose, callbacks, valFunction, valIns, args.shuffle,\n        callbackMetrics, args.initialEpoch, null, null);\n    return out;\n  } finally {\n    model.isTraining = false;\n    // Memory clean up.\n    disposeNewTensors(inputs, x);\n    disposeNewTensors(targets, y);\n    disposeNewTensors(valX as Tensor[], inputValX);\n    disposeNewTensors(valY as Tensor[], inputValY);\n    if (sampleWeights != null) {\n      tfc.dispose(sampleWeights);\n    }\n  }\n  // TODO(cais): Add value to outLabels.\n}\n\n/**\n * Ensure tensors all have a rank of at least 2.\n *\n * If a tensor has a rank of 1, it is dimension-expanded to rank 2.\n * If any tensor has a rank of 0 (i.e., is a scalar), an error will be thrown.\n */\nexport function ensureTensorsRank2OrHigher(tensors: Tensor|Tensor[]): Tensor[] {\n  const outs: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    tensors = [tensors];\n  }\n\n  // Make Tensors at least 2D.\n  for (let i = 0; i < tensors.length; ++i) {\n    const tensor = tensors[i];\n    if (tensor.rank === 1) {\n      outs.push(expandDims(tensor, 1));\n    } else if (tensor.rank === 0) {\n      throw new Error(\n          'Expected tensor to be at least 1D, but received a 0D tensor ' +\n          '(scalar).');\n    } else {\n      outs.push(tensor);\n    }\n  }\n  return outs;\n}\n\n/**\n * Compare a set of tensors with a reference (old) set, discard the ones\n * in the new set that are not present in the reference set.\n *\n * This method is used for memory clenaup during calls such as\n * LayersModel.fit().\n *\n * @param tensors New set which may contain Tensors not present in\n *   `refTensors`.\n * @param refTensors Reference Tensor set.\n */\n// TODO(cais, kangyizhang): Deduplicate with tfjs-data.\nexport function disposeNewTensors(\n    tensors: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    refTensors: Tensor|Tensor[]|{[inputName: string]: Tensor}): void {\n  if (tensors == null) {\n    return;\n  }\n  const oldTensorIds: number[] = [];\n  if (refTensors instanceof Tensor) {\n    oldTensorIds.push(refTensors.id);\n  } else if (Array.isArray(refTensors)) {\n    refTensors.forEach(t => oldTensorIds.push(t.id));\n  } else if (refTensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in refTensors) {\n      const oldTensor = refTensors[name];\n      oldTensorIds.push(oldTensor.id);\n    }\n  }\n\n  const tensorsToDispose: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    if (oldTensorIds.indexOf(tensors.id) === -1) {\n      tensorsToDispose.push(tensors);\n    }\n  } else if (Array.isArray(tensors)) {\n    tensors.forEach(t => {\n      if (oldTensorIds.indexOf(t.id) === -1) {\n        tensorsToDispose.push(t);\n      }\n    });\n  } else if (tensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in tensors) {\n      const tensor = tensors[name];\n      if (oldTensorIds.indexOf(tensor.id) === -1) {\n        tensorsToDispose.push(tensor);\n      }\n    }\n  }\n\n  tensorsToDispose.forEach(t => {\n    if (!t.isDisposed) {\n      t.dispose();\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: engine/training.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {io, ModelPredictConfig as ModelPredictArgs, NamedTensorMap, Optimizer, Scalar, scalar, serialization, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {History, ModelLoggingVerbosity} from '../base_callbacks';\nimport {nameScope} from '../common';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {LossIdentifier} from '../keras_format/loss_config';\nimport {OptimizerSerialization} from '../keras_format/optimizer_config';\nimport {MetricsIdentifier, TrainingConfig} from '../keras_format/training_config';\nimport {deserialize} from '../layers/serialization';\nimport * as losses from '../losses';\nimport * as Metrics from '../metrics';\nimport * as optimizers from '../optimizers';\nimport {LossOrMetricFn, NamedTensor} from '../types';\nimport {checkUserDefinedMetadata} from '../user_defined_metadata';\nimport {count, pyListRepeat, singletonOrArray, toCamelCase, toSnakeCase, unique} from '../utils/generic_utils';\nimport {printSummary} from '../utils/layer_utils';\nimport {range} from '../utils/math_utils';\nimport {convertPythonicToTs} from '../utils/serialization_utils';\nimport {LayerVariable} from '../variables';\nimport {version} from '../version';\n\nimport {Container, ContainerArgs} from './container';\nimport {Dataset} from './dataset_stub';\nimport {execute, FeedDict} from './executor';\nimport {DisposeResult, SymbolicTensor} from './topology';\nimport {evaluateDataset, fitDataset, ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './training_dataset';\nimport {checkBatchSize, disposeNewTensors, ensureTensorsRank2OrHigher, fitTensors, makeBatches, ModelFitArgs, sliceArrays, sliceArraysByIndices} from './training_tensors';\nimport {ClassWeight, ClassWeightMap, computeWeightedLoss, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Helper function for polymorphic input data: 1. singleton Tensor.\n */\nexport function isDataTensor(x: Tensor|Tensor[]|{[inputName: string]: Tensor}|\n                             {[inputName: string]: Tensor[]}): boolean {\n  return x instanceof Tensor;\n}\n\n/**\n * Helper function for polymorphic input data: 2. Array of Tensor.\n */\nexport function isDataArray(x: Tensor|Tensor[]|\n                            {[inputName: string]: Tensor}): boolean {\n  return Array.isArray(x);\n}\n\n/**\n * Helper function for polymorphic input data: 3. \"dict\" of Tensor.\n */\nexport function isDataDict(x: Tensor|Tensor[]|\n                           {[inputName: string]: Tensor}): boolean {\n  return !isDataTensor(x) && !isDataArray(x);\n}\n\n/**\n * Normalizes inputs and targets provided by users.\n * @param data User-provided input data (polymorphic).\n * @param names An Array of expected Tensor names.\n * @param shapes Optional Array of expected Tensor shapes.\n * @param checkBatchAxis Whether to check that the batch axis of the arrays\n *   match  the expected value found in `shapes`.\n * @param exceptionPrefix String prefix used for exception formatting.\n * @returns List of standardized input Tensors (one Tensor per model input).\n * @throws ValueError: in case of improperly formatted user data.\n */\nexport function standardizeInputData(\n    data: Tensor|Tensor[]|{[inputName: string]: Tensor}, names: string[],\n    shapes?: Shape[], checkBatchAxis = true, exceptionPrefix = ''): Tensor[] {\n  if (names == null || names.length === 0) {\n    // Check for the case where the model expected no data, but some data got\n    // sent.\n    if (data != null) {\n      let gotUnexpectedData = false;\n      if (isDataArray(data) && (data as Tensor[]).length > 0) {\n        gotUnexpectedData = true;\n      } else if (isDataDict(data)) {\n        for (const key in data) {\n          if (data.hasOwnProperty(key)) {\n            gotUnexpectedData = true;\n            break;\n          }\n        }\n      } else {\n        // `data` is a singleton Tensor in this case.\n        gotUnexpectedData = true;\n      }\n      if (gotUnexpectedData) {\n        throw new ValueError(\n            `Error when checking model ${exceptionPrefix} expected no data, ` +\n            `but got ${data}`);\n      }\n    }\n    return [];\n  }\n  if (data == null) {\n    return names.map(name => null);\n  }\n\n  let arrays: Tensor[];\n  if (isDataDict(data)) {\n    data = data as {[inputName: string]: Tensor};\n    arrays = [];\n    for (const name of names) {\n      if (data[name] == null) {\n        throw new ValueError(\n            `No data provided for \"${name}\". Need data for each key in: ` +\n            `${names}`);\n      }\n      arrays.push(data[name]);\n    }\n  } else if (isDataArray(data)) {\n    data = data as Tensor[];\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `model expected. Expected to see ${names.length} Tensor(s), but ` +\n          `instead got the following list of Tensor(s): ${data}`);\n    }\n    arrays = data;\n  } else {\n    data = data as Tensor;\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model ${exceptionPrefix} expects ${names.length} Tensor(s), ` +\n          `but only received one Tensor. Found: Tensor with shape ${\n              data.shape}`);\n    }\n    arrays = [data];\n  }\n\n  arrays = ensureTensorsRank2OrHigher(arrays);\n\n  // Check shape compatibility.\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s). but got array with ` +\n            `shape ${array.shape}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          // Skip the first (batch) axis.\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null && refDim >= 0 && dim !== refDim) {\n          throw new ValueError(\n              `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n              `to have shape [${shapes[i]}], but got array with shape ` +\n              `[${array.shape}].`);\n        }\n      }\n    }\n  }\n  return arrays;\n}\n\n/**\n * User input validation for Tensors.\n * @param inputs `Array` of `tf.Tensor`s for inputs.\n * @param targets `Array` of `tf.Tensor`s for targets.\n * @param weights Optional `Array` of `tf.Tensor`s for sample weights.\n * @throws ValueError: in case of incorrectly formatted data.\n */\nexport function checkArrayLengths(\n    inputs: Tensor[], targets: Tensor[], weights?: Tensor[]) {\n  const setX = unique(inputs.map(input => input.shape[0]));\n  setX.sort();\n  const setY = unique(targets.map(target => target.shape[0]));\n  setY.sort();\n  // TODO(cais): Check `weights` as well.\n  if (setX.length > 1) {\n    throw new ValueError(\n        `All input Tensors (x) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(inputs.map(input => input.shape))}`);\n  }\n  if (setY.length > 1) {\n    throw new ValueError(\n        `All target Tensors (y) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(targets.map(target => target.shape))}`);\n  }\n  if (setX.length > 0 && setY.length > 0 && !util.arraysEqual(setX, setY)) {\n    throw new ValueError(\n        `Input Tensors should have the same number of samples as target ` +\n        `Tensors. Found ${setX[0]} input sample(s) and ${setY[0]} target ` +\n        `sample(s).`);\n  }\n}\n\n/**\n * Validation on the compatibility of targes and loss functions.\n *\n * This helps prevent users from using loss functions incorrectly.\n *\n * @param targets `Array` of `tf.Tensor`s of targets.\n * @param lossFns `Array` of loss functions.\n * @param outputShapes `Array` of shapes of model outputs.\n */\nfunction checkLossAndTargetCompatibility(\n    targets: Tensor[], lossFns: LossOrMetricFn[], outputShapes: Shape[]) {\n  // TODO(cais): Dedicated test coverage?\n  const keyLosses = [\n    losses.meanSquaredError, losses.binaryCrossentropy,\n    losses.categoricalCrossentropy\n  ];\n  for (let i = 0; i < targets.length; ++i) {\n    const y = targets[i];\n    const loss = lossFns[i];\n    const shape = outputShapes[i];\n    if (loss == null) {\n      continue;\n    }\n    if (loss === losses.categoricalCrossentropy) {\n      if (y.shape[y.shape.length - 1] === 1) {\n        throw new ValueError(\n            `You are passing a target array of shape ${y.shape} while using ` +\n            `a loss 'categorical_crossentropy'. 'categorical_crossentropy'` +\n            `expects targets to be binary matrices (1s and 0s) of shape ` +\n            `[samples, classes].`);\n        // TODO(cais): Example code in error message.\n      }\n    }\n    if (keyLosses.indexOf(loss) !== -1) {\n      const slicedYShape = y.shape.slice(1);\n      const slicedShape = shape.slice(1);\n      for (let j = 0; j < slicedYShape.length; ++j) {\n        const targetDim = slicedYShape[j];\n        const outDim = slicedShape[j];\n        if (outDim != null && targetDim !== outDim) {\n          throw new ValueError(\n              `A target Tensor with shape ${y.shape} was passed for an ` +\n              `output of shape ${shape}, while using a loss function that ` +\n              `expects targets to have the same shape as the output.`);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Check inputs provided by the user.\n *\n * Porting Note: This corresponds to _standardize_input_data() in Python\n *   Keras. Because of the strong typing in TF.js, we do not need to convert\n *   the data. Specifically:\n *   1) in PyKeras, `data` can be `DataFrame` instances from pandas, for\n *      example. We don't need to worry about that here because there is no\n *      widely popular javascript/typesdcript equivalent of pandas (so far).\n *      If one becomes available in the future, we can add support.\n *   2) in PyKeras, inputs can be Python dict. But here we are stipulating\n * that the data is either a single `tf.Tensor` or an Array of `tf.Tensor`s. We\n * may add support for `Object` data inputs in the future when the need\n * arises.\n *\n * Instead, we perform basic checks for number of parameters and shapes.\n *\n * @param data: The input data.\n * @param names: Name for the inputs, from the model.\n * @param shapes: Expected shapes for the input data, from the model.\n * @param checkBatchAxis: Whether the size along the batch axis (i.e., the\n *   first dimension) will be checked for matching.\n * @param exceptionPrefix: Execption prefix message, used in generating error\n *   messages.\n * @throws ValueError: on incorrect number of inputs or mismatches in shapes.\n */\nfunction checkInputData(\n    data: Tensor|Tensor[], names: string[], shapes?: Shape[],\n    checkBatchAxis = true, exceptionPrefix = '') {\n  let arrays: Tensor[];\n  if (Array.isArray(data)) {\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `the model expected. Expected to see ${names.length} Tensor(s),` +\n          ` but instead got ${data.length} Tensors(s).`);\n    }\n    arrays = data;\n  } else {\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model expects ${names.length} ${exceptionPrefix} Tensors, ` +\n          `but only received one Tensor. Found: array with shape ` +\n          `${JSON.stringify(data.shape)}.`);\n    }\n    arrays = [data];\n  }\n\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s), but got array with ` +\n            `shape ${JSON.stringify(array.shape)}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null) {\n          if (refDim !== dim) {\n            throw new ValueError(\n                `Error when checking ${exceptionPrefix}: expected ` +\n                `${names[i]} to have shape ${JSON.stringify(shapes[i])} but ` +\n                `got array with shape ${JSON.stringify(array.shape)}.`);\n          }\n        }\n      }\n    }\n  }\n}\n\n/**\n * Maps metric functions to model outputs.\n * @param metrics An shortcut strings name, metric function, `Array` or dict\n *   (`Object`) of metric functions.\n * @param outputNames An `Array` of the names of model outputs.\n * @returns An `Array` (one entry per model output) of `Array` of metric\n *   functions. For instance, if the model has 2 outputs, and for the first\n *   output we want to compute `binaryAccuracy` and `binaryCrossentropy`,\n *   and just `binaryAccuracy` for the second output, the `Array` would look\n *   like:\n *     `[[binaryAccuracy, binaryCrossentropy],  [binaryAccuracy]]`\n * @throws TypeError: incompatible metrics format.\n */\nexport function collectMetrics(\n    metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n    {[outputName: string]: string | LossOrMetricFn},\n    outputNames: string[]): Array<Array<string|LossOrMetricFn>> {\n  if (metrics == null || Array.isArray(metrics) && metrics.length === 0) {\n    return outputNames.map(name => []);\n  }\n\n  let wrappedMetrics: Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  if (typeof metrics === 'string' || typeof metrics === 'function') {\n    wrappedMetrics = [metrics];\n  } else if (Array.isArray(metrics) || typeof metrics === 'object') {\n    wrappedMetrics = metrics as Array<string|LossOrMetricFn>|\n        {[outputName: string]: string} | {[outputName: string]: LossOrMetricFn};\n  } else {\n    throw new TypeError(\n        'Type of metrics argument not understood. Expected an string,' +\n        `function, Array, or Object, found: ${metrics}`);\n  }\n\n  if (Array.isArray(wrappedMetrics)) {\n    // We then apply all metrics to all outputs.\n    return outputNames.map(\n        name => wrappedMetrics as Array<string|LossOrMetricFn>);\n  } else {\n    // In this case, metrics is a dict.\n    const nestedMetrics: Array<Array<string|LossOrMetricFn>> = [];\n    for (const name of outputNames) {\n      let outputMetrics: string|LossOrMetricFn|Array<string|LossOrMetricFn> =\n          wrappedMetrics.hasOwnProperty(name) ? wrappedMetrics[name] : [];\n      if (!Array.isArray(outputMetrics)) {\n        outputMetrics = [outputMetrics];\n      }\n      nestedMetrics.push(outputMetrics);\n    }\n    return nestedMetrics;\n  }\n}\n\nexport interface ModelEvaluateArgs {\n  /**\n   * Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * Tensor of weights to weight the contribution of different samples to the\n   * loss and metrics.\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * integer: total number of steps (batches of samples)\n   * before declaring the evaluation round finished. Ignored with the default\n   * value of `undefined`.\n   */\n  steps?: number;\n}\n\n/**\n * Configuration for calls to `LayersModel.compile()`.\n */\nexport interface ModelCompileArgs {\n  /**\n   * An instance of `tf.train.Optimizer` or a string name for an Optimizer.\n   */\n  optimizer: string|Optimizer;\n\n  /**\n   * Object function(s) or name(s) of object function(s).\n   * If the model has multiple outputs, you can use a different loss\n   * on each output by passing a dictionary or an Array of losses.\n   * The loss value that will be minimized by the model will then be the sum\n   * of all individual losses.\n   */\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n\n  /**\n   * List of metrics to be evaluated by the model during training and testing.\n   * Typically you will use `metrics=['accuracy']`.\n   * To specify different metrics for different outputs of a multi-output\n   * model, you could also pass a dictionary.\n   */\n  metrics?: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n\n  // TODO(cais): Add lossWeights, sampleWeightMode, weightedMetrics, and\n  //   targetTensors.\n}\n\nconst LAYERS_MODEL_FORMAT_NAME = 'layers-model';\n\n/**\n * A `tf.LayersModel` is a directed, acyclic graph of `tf.Layer`s plus methods\n * for training, evaluation, prediction and saving.\n *\n * `tf.LayersModel` is the basic unit of training, inference and evaluation in\n * TensorFlow.js. To create a `tf.LayersModel`, use `tf.LayersModel`.\n *\n * See also:\n *   `tf.Sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class LayersModel extends Container implements tfc.InferenceModel {\n  // The class name is 'Model' rather than 'LayersModel' for backwards\n  // compatibility since this class name shows up in the serialization format.\n  /** @nocollapse */\n  static className = 'Model';\n  protected optimizer_: Optimizer;\n  // Whether the model instance owns the optimizer: `true` if and only if\n  // `optimizer` is created from a string parameter during `compile()` call.\n  protected isOptimizerOwned: boolean;\n\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n  lossFunctions: LossOrMetricFn[];\n\n  // TODO(cais): These private variables should probably not have the string\n  //   'feed' in their names, because we are not dealing with a symbolic\n  //   backend.\n  private feedOutputShapes: Shape[];\n  private feedLossFns: LossOrMetricFn[];\n  private collectedTrainableWeights: LayerVariable[];\n  private testFunction: (data: Tensor[]) => Scalar[];\n  history: History;\n\n  // A public property that can be set by Callbacks to order early stopping\n  // during `fit()` calls.\n  protected stopTraining_: boolean;\n  protected isTraining: boolean;\n\n  metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  metricsNames: string[];\n  // Porting Note: `metrics_tensors` in PyKeras is a symbolic tensor. But given\n  //   the imperative nature of tfjs-core, `metricsTensors` is a\n  //   TypeScript function here.\n  //   Also note that due to the imperative nature of tfjs-core, `metricsTensor`\n  //   here needs an output index to keep track of which output of the\n  //   LayersModel a metric belongs to. This is unlike `metrics_tensors` in\n  //   PyKeras, which is a `list` of symbolic tensors, each of which has\n  //   implicit \"knowledge\" of the outputs it depends on.\n  metricsTensors: Array<[LossOrMetricFn, number]>;\n\n  // User defind metadata (if any).\n  private userDefinedMetadata: {};\n\n  constructor(args: ContainerArgs) {\n    super(args);\n    this.isTraining = false;\n  }\n\n  /**\n   * Print a text summary of the model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - If the model has non-sequential-like topology, the inputs each layer\n   *   receives\n   * - The total number of trainable and non-trainable parameters of the model.\n   *\n   * ```js\n   * const input1 = tf.input({shape: [10]});\n   * const input2 = tf.input({shape: [20]});\n   * const dense1 = tf.layers.dense({units: 4}).apply(input1);\n   * const dense2 = tf.layers.dense({units: 8}).apply(input2);\n   * const concat = tf.layers.concatenate().apply([dense1, dense2]);\n   * const output =\n   *     tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat);\n   *\n   * const model = tf.model({inputs: [input1, input2], outputs: output});\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      throw new ValueError(\n          `This model has never been called, thus its weights have not been ` +\n          `created yet. So no summary can be displayed. Build the model ` +\n          `first (e.g., by calling it on some test data).`);\n    }\n    printSummary(this, lineLength, positions, printFn);\n  }\n\n  /**\n   * Configures and prepares the model for training and evaluation.  Compiling\n   * outfits the model with an optimizer, loss, and/or metrics.  Calling `fit`\n   * or `evaluate` on an un-compiled model will throw an error.\n   *\n   * @param args a `ModelCompileArgs` specifying the loss, optimizer, and\n   * metrics to be used for fitting and evaluating this model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  compile(args: ModelCompileArgs): void {\n    if (args.loss == null) {\n      args.loss = [];\n    }\n    this.loss = args.loss;\n\n    if (typeof args.optimizer === 'string') {\n      this.optimizer_ = optimizers.getOptimizer(args.optimizer);\n      this.isOptimizerOwned = true;\n    } else {\n      if (!(args.optimizer instanceof Optimizer)) {\n        throw new ValueError(\n            `User-defined optimizer must be an instance of tf.Optimizer.`);\n      }\n      this.optimizer_ = args.optimizer;\n      this.isOptimizerOwned = false;\n    }\n\n    // TODO(cais): Add lossWeights.\n    // TODO(cais): Add sampleWeightMode.\n\n    // Prepare loss functions.\n    let lossFunctions: LossOrMetricFn[] = [];\n    if (!Array.isArray(args.loss) && typeof args.loss !== 'string' &&\n        typeof args.loss !== 'function') {\n      args.loss = args.loss as {[outputName: string]: string};\n      for (const name in args.loss) {\n        if (this.outputNames.indexOf(name) === -1) {\n          throw new ValueError(\n              `Unknown entry in loss dictionary: \"${name}\". ` +\n              `Only expected the following keys: ${this.outputNames}`);\n        }\n      }\n      for (const name of this.outputNames) {\n        if (args.loss[name] == null) {\n          console.warn(\n              `Output \"${name}\" is missing from loss dictionary. We assume ` +\n              `this was done on purpose, and we will not be expecting data ` +\n              `to be passed to ${name} during training`);\n        }\n        lossFunctions.push(losses.get(args.loss[name]));\n      }\n    } else if (Array.isArray(args.loss)) {\n      if (args.loss.length !== this.outputs.length) {\n        throw new ValueError(\n            `When passing an Array as loss, it should have one entry per ` +\n            `model output. The model has ${this.outputs.length} output(s), ` +\n            `but you passed loss=${args.loss}.`);\n      }\n      const theLosses = args.loss as Array<string|LossOrMetricFn>;\n      lossFunctions = theLosses.map(l => losses.get(l));\n    } else {\n      const lossFunction = losses.get(args.loss);\n      this.outputs.forEach(_ => {\n        lossFunctions.push(lossFunction);\n      });\n    }\n\n    this.lossFunctions = lossFunctions;\n\n    this.feedOutputNames = [];\n    this.feedOutputShapes = [];\n    this.feedLossFns = [];\n    for (let i = 0; i < this.outputs.length; ++i) {\n      // TODO(cais): Logic for skipping target(s).\n      const shape = this.internalOutputShapes[i];\n      const name = this.outputNames[i];\n      this.feedOutputNames.push(name);\n      this.feedOutputShapes.push(shape);\n      this.feedLossFns.push(this.lossFunctions[i]);\n    }\n\n    // TODO(cais): Add logic for output masks.\n    // TODO(cais): Add logic for sample weights.\n    const skipTargetIndices: number[] = [];\n\n    // Prepare metrics.\n    this.metrics = args.metrics;\n    // TODO(cais): Add weightedMetrics.\n    this.metricsNames = ['loss'];\n    this.metricsTensors = [];\n\n    // Compute total loss.\n    // Porting Note: In PyKeras, metrics_tensors are symbolic tensor objects.\n    //   Here, metricsTensors are TypeScript functions. This difference is due\n    //   to the difference in symbolic/imperative property of the backends.\n    nameScope('loss', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        // TODO(cais): Add weightedLoss, sampleWeight and mask.\n        //   The following line should be weightedLoss\n        const weightedLoss = this.lossFunctions[i];\n        if (this.outputs.length > 1) {\n          this.metricsTensors.push([weightedLoss, i]);\n          this.metricsNames.push(this.outputNames[i] + '_loss');\n        }\n      }\n\n      // Porting Note: Due to the imperative nature of the backend, we calculate\n      //   the regularizer penalties in the totalLossFunction, instead of here.\n    });\n\n    const nestedMetrics = collectMetrics(args.metrics, this.outputNames);\n    // TODO(cais): Add nestedWeightedMetrics.\n\n    /**\n     * Helper function used in loop below.\n     */\n    const appendMetric =\n        (outputIndex: number, metricName: string,\n         metricTensor: LossOrMetricFn) => {\n          if (this.outputNames.length > 1) {\n            metricName = this.outputNames[outputIndex] + '_' + metricName;\n          }\n          this.metricsNames.push(metricName);\n          this.metricsTensors.push([metricTensor, outputIndex]);\n        };\n\n    nameScope('metric', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        const outputMetrics = nestedMetrics[i];\n        // TODO(cais): Add weights and outputWeightedMetrics.\n\n        // TODO(cais): Add optional arg `weights` to the following function.\n        const handleMetrics = (metrics: Array<string|LossOrMetricFn>) => {\n          const metricNamePrefix = '';\n          let metricName: string;\n          let accFn: LossOrMetricFn;\n          let weightedMetricFn: LossOrMetricFn;\n          //  TODO(cais): Use 'weights_' for weighted metrics.\n\n          for (const metric of metrics) {\n            if (typeof metric === 'string' &&\n                ['accuracy', 'acc', 'crossentropy', 'ce'].indexOf(metric) !==\n                    -1) {\n              const outputShape = this.internalOutputShapes[i];\n\n              if (outputShape[outputShape.length - 1] === 1 ||\n                  this.lossFunctions[i] === losses.binaryCrossentropy) {\n                // case: binary accuracy/crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryCrossentropy;\n                }\n              } else if (\n                  this.lossFunctions[i] ===\n                  losses.sparseCategoricalCrossentropy) {\n                // case: categorical accuracy / crossentropy with sparse\n                // targets.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalCrossentropy;\n                }\n              } else {\n                // case: categorical accuracy / crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalCrossentropy;\n                }\n              }\n              let suffix: string;\n              if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                suffix = 'acc';\n              } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                suffix = 'ce';\n              }\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = accFn;\n              metricName = metricNamePrefix + suffix;\n            } else {\n              const metricFn = Metrics.get(metric);\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = metricFn;\n              metricName =\n                  metricNamePrefix + Metrics.getLossOrMetricName(metric);\n            }\n\n            // TODO(cais): Add weighting and masking to metricResult.\n            let metricResult: LossOrMetricFn;\n            nameScope(metricName, () => {\n              metricResult = weightedMetricFn;\n            });\n            appendMetric(i, metricName, metricResult);\n          }\n        };\n\n        handleMetrics(outputMetrics);\n        // TODO(cais): Call handleMetrics with weights.\n      }\n    });\n\n    // Porting Notes: Given the imperative backend of tfjs-core,\n    //   there is no need for constructing the symbolic graph and placeholders.\n    this.collectedTrainableWeights = this.trainableWeights;\n  }\n\n  /**\n   * Check trainable weights count consistency.\n   *\n   * This will raise a warning if `this.trainableWeights` and\n   * `this.collectedTrainableWeights` are inconsistent (i.e., have different\n   * numbers of parameters).\n   * Inconsistency will typically arise when one modifies `model.trainable`\n   * without calling `model.compile()` again.\n   */\n  protected checkTrainableWeightsConsistency(): void {\n    if (this.collectedTrainableWeights == null) {\n      return;\n    }\n    if (this.trainableWeights.length !==\n        this.collectedTrainableWeights.length) {\n      console.warn(\n          'Discrepancy between trainableweights and collected trainable ' +\n          'weights. Did you set `model.trainable` without calling ' +\n          '`model.compile()` afterwards?');\n    }\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(\n   *     tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4});\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateArgs`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // TODO(cais): Standardize `config.sampleWeights` as well.\n    // Validate user data.\n    const checkBatchAxis = true;\n    const standardizedOuts =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    try {\n      // TODO(cais): If uses `useLearningPhase`, set the corresponding element\n      // of the input to 0.\n      const ins = standardizedOuts[0].concat(standardizedOuts[1]);\n      this.makeTestFunction();\n      const f = this.testFunction;\n      const testOuts =\n          this.testLoop(f, ins, batchSize, args.verbose, args.steps);\n      return singletonOrArray(testOuts);\n    } finally {\n      disposeNewTensors(standardizedOuts[0], x);\n      disposeNewTensors(standardizedOuts[1], y);\n    }\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args?: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    this.makeTestFunction();\n    return evaluateDataset(this, dataset, args);\n  }\n\n  /**\n   * Get number of samples provided for training, evaluation or prediction.\n   *\n   * @param ins Input `tf.Tensor`.\n   * @param batchSize Integer batch size, optional.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring loop finished. Optional.\n   * @param stepsName The public API's parameter name for `steps`.\n   * @returns Number of samples provided.\n   */\n  private checkNumSamples(\n      ins: Tensor|Tensor[], batchSize?: number, steps?: number,\n      stepsName = 'steps'): number {\n    let numSamples: number;\n    if (steps != null) {\n      numSamples = null;\n      if (batchSize != null) {\n        throw new ValueError(\n            `If ${stepsName} is set, batchSize must be null or undefined.` +\n            `Got batchSize = ${batchSize}`);\n      }\n    } else if (ins != null) {\n      if (Array.isArray(ins)) {\n        numSamples = ins[0].shape[0];\n      } else {\n        numSamples = ins.shape[0];\n      }\n    } else {\n      throw new ValueError(\n          `Either the input data should have a defined shape, or ` +\n          `${stepsName} shoud be specified.`);\n    }\n    return numSamples;\n  }\n\n  /**\n   * Execute internal tensors of the model with input data feed.\n   * @param inputs Input data feed. Must match the inputs of the model.\n   * @param outputs Names of the output tensors to be fetched. Must match\n   *   names of the SymbolicTensors that belong to the graph.\n   * @returns Fetched values for `outputs`.\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs: string|string[]):\n      Tensor|Tensor[] {\n    if (Array.isArray(outputs) && outputs.length === 0) {\n      throw new ValueError(\n          '`outputs` is an empty Array, which is not allowed.');\n    }\n\n    const outputsIsArray = Array.isArray(outputs);\n    const outputNames =\n        (outputsIsArray ? outputs as string[] : [outputs as string]);\n    const outputSymbolicTensors = this.retrieveSymbolicTensors(outputNames);\n\n    // Format the input into a FeedDict.\n    const feedDict = new FeedDict();\n    if (inputs instanceof Tensor) {\n      inputs = [inputs];\n    }\n    if (Array.isArray(inputs)) {\n      if (inputs.length !== this.inputs.length) {\n        throw new ValueError(\n            `The number of inputs provided (${inputs.length}) ` +\n            `does not match the number of inputs of this model ` +\n            `(${this.inputs.length}).`);\n      }\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n    } else {\n      for (const input of this.inputs) {\n        const tensorValue = inputs[input.name];\n        if (tensorValue == null) {\n          throw new ValueError(\n              `No value is provided for the model's input ${input.name}`);\n        }\n        feedDict.add(input, tensorValue);\n      }\n    }\n\n    // Run execution.\n    const executeOutputs = execute(outputSymbolicTensors, feedDict) as Tensor[];\n    return outputsIsArray ? executeOutputs : executeOutputs[0];\n  }\n\n  /**\n   * Retrieve the model's internal symbolic tensors from symbolic-tensor names.\n   */\n  private retrieveSymbolicTensors(symbolicTensorNames: string[]):\n      SymbolicTensor[] {\n    const outputSymbolicTensors: SymbolicTensor[] =\n        pyListRepeat(null, symbolicTensorNames.length);\n    let outputsRemaining = symbolicTensorNames.length;\n    for (const layer of this.layers) {\n      const layerOutputs: SymbolicTensor[] =\n          Array.isArray(layer.output) ? layer.output : [layer.output];\n      const layerOutputNames = layerOutputs.map(output => output.name);\n      for (let i = 0; i < symbolicTensorNames.length; ++i) {\n        const index = layerOutputNames.indexOf(symbolicTensorNames[i]);\n        if (index !== -1) {\n          outputSymbolicTensors[i] = layerOutputs[index];\n          outputsRemaining--;\n        }\n        if (outputsRemaining === 0) {\n          break;\n        }\n      }\n      if (outputsRemaining === 0) {\n        break;\n      }\n    }\n\n    if (outputsRemaining > 0) {\n      const remainingNames: string[] = [];\n      outputSymbolicTensors.forEach((tensor, i) => {\n        if (tensor == null) {\n          remainingNames.push(symbolicTensorNames[i]);\n        }\n      });\n      throw new ValueError(\n          `Cannot find SymbolicTensors for output name(s): ` +\n          `${JSON.stringify(remainingNames)}`);\n    }\n    return outputSymbolicTensors;\n  }\n\n  /**\n   * Helper method to loop over some data in batches.\n   *\n   * Porting Note: Not using the functional approach in the Python equivalent\n   *   due to the imperative backend.\n   * Porting Note: Does not support step mode currently.\n   *\n   * @param ins: input data\n   * @param batchSize: integer batch size.\n   * @param verbose: verbosity model\n   * @returns: Predictions as `tf.Tensor` (if a single output) or an `Array` of\n   *   `tf.Tensor` (if multipe outputs).\n   */\n  private predictLoop(ins: Tensor|Tensor[], batchSize = 32, verbose = false):\n      Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins);\n      if (verbose) {\n        throw new NotImplementedError(\n            'Verbose predictLoop() is not implemented yet.');\n      }\n\n      // Sample-based predictions.\n      // Porting Note: Tensor currently does not support sliced assignments as\n      //   in numpy, e.g., x[1:3] = y. Therefore we use concatenation while\n      //   iterating over the batches.\n\n      const batches = makeBatches(numSamples, batchSize);\n      const outsBatches: Tensor[][] = this.outputs.map(output => []);\n\n      // TODO(cais): Can the scope() be pushed down inside the for loop?\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchOuts = tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          // TODO(cais): Take care of the case of the last element is a flag for\n          //   training/test.\n          const insBatch = sliceArrays(ins, batchStart, batchEnd);\n\n          // Construct the feeds for execute();\n          const feeds = [];\n          if (Array.isArray(insBatch)) {\n            for (let i = 0; i < insBatch.length; ++i) {\n              feeds.push({key: this.inputs[i], value: insBatch[i]});\n            }\n          } else {\n            feeds.push({key: this.inputs[0], value: insBatch});\n          }\n          const feedDict = new FeedDict(feeds);\n          return execute(this.outputs, feedDict) as Tensor[];\n        });\n        batchOuts.forEach((batchOut, i) => outsBatches[i].push(batchOut));\n      }\n      return singletonOrArray(\n          outsBatches.map(batches => tfc.concat(batches, 0)));\n    });\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFlow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([8, 10]), {batchSize: 4}).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param args A `ModelPredictArgs` object containing optional fields.\n   *\n   * @return Prediction results as a `tf.Tensor`(s).\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    const xsRank2OrHigher = ensureTensorsRank2OrHigher(x);\n    checkInputData(\n        xsRank2OrHigher, this.inputNames, this.feedInputShapes, false);\n    try {\n      // TODO(cais): Take care of stateful models.\n      //   if (this.stateful) ...\n      // TODO(cais): Take care of the learning_phase boolean flag.\n      //   if (this.useLearningPhase) ...\n      const batchSize = args.batchSize == null ? 32 : args.batchSize;\n      checkBatchSize(batchSize);\n      return this.predictLoop(xsRank2OrHigher, batchSize);\n    } finally {\n      disposeNewTensors(xsRank2OrHigher, x);\n    }\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predictOnBatch(tf.ones([8, 10])).print();\n   * ```\n   * @param x: Input samples, as a Tensor (for models with exactly one\n   *   input) or an array of Tensors (for models with more than one input).\n   * @return Tensor(s) of predictions\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predictOnBatch(x: Tensor|Tensor[]): Tensor|Tensor[] {\n    checkInputData(x, this.inputNames, this.feedInputShapes, true);\n    // TODO(cais): Take care of the learning_phase boolean flag.\n    //   if (this.useLearningPhase) ...\n    const batchSize = (Array.isArray(x) ? x[0] : x).shape[0];\n    return this.predictLoop(x, batchSize);\n  }\n\n  protected standardizeUserDataXY(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor}, checkBatchAxis = true,\n      batchSize?: number): [Tensor[], Tensor[]] {\n    // TODO(cais): Add sampleWeight, classWeight\n    if (this.optimizer_ == null) {\n      throw new RuntimeError(\n          'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileArgs).');\n    }\n    const outputShapes: Shape[] = [];\n    for (let i = 0; i < this.feedOutputShapes.length; ++i) {\n      const outputShape = this.feedOutputShapes[i];\n      const lossFn = this.feedLossFns[i];\n      if (lossFn === losses.sparseCategoricalCrossentropy) {\n        outputShapes.push(\n            outputShape.slice(0, outputShape.length - 1).concat([1]));\n      } else {\n        // Porting Note: Because of strong typing `lossFn` must be a function.\n        outputShapes.push(outputShape);\n      }\n    }\n    x = standardizeInputData(\n        x, this.feedInputNames, this.feedInputShapes, false, 'input');\n    y = standardizeInputData(\n        y, this.feedOutputNames, outputShapes, false, 'target');\n    // TODO(cais): Standardize sampleWeights & classWeights.\n    checkArrayLengths(x, y, null);\n    // TODO(cais): Check sampleWeights as well.\n    checkLossAndTargetCompatibility(y, this.feedLossFns, this.feedOutputShapes);\n    if (this.stateful && batchSize != null && batchSize > 0) {\n      if (x[0].shape[0] % batchSize !== 0) {\n        throw new ValueError(\n            `In a stateful network, you should only pass inputs with a ` +\n            `number of samples that is divisible by the batch size ` +\n            `${batchSize}. Found: ${x[0].shape[0]} sample(s).`);\n      }\n    }\n    return [x, y];\n  }\n\n  protected async standardizeUserData(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      sampleWeight?: Tensor|Tensor[]|{[outputName: string]: Tensor},\n      classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap,\n      checkBatchAxis = true,\n      batchSize?: number): Promise<[Tensor[], Tensor[], Tensor[]]> {\n    const [standardXs, standardYs] =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    // TODO(cais): Handle sampleWeights.\n    if (sampleWeight != null) {\n      throw new Error('sample weight is not supported yet.');\n    }\n\n    let standardSampleWeights: Tensor[] = null;\n    if (classWeight != null) {\n      const classWeights =\n          standardizeClassWeights(classWeight, this.outputNames);\n      standardSampleWeights = [];\n      for (let i = 0; i < classWeights.length; ++i) {\n        standardSampleWeights.push(\n            await standardizeWeights(standardYs[i], null, classWeights[i]));\n      }\n    }\n\n    // TODO(cais): Deal with the case of model.stateful == true.\n    return [standardXs, standardYs, standardSampleWeights];\n  }\n\n  /**\n   * Loop over some test data in batches.\n   * @param f A Function returning a list of tensors.\n   * @param ins Array of tensors to be fed to `f`.\n   * @param batchSize Integer batch size or `null` / `undefined`.\n   * @param verbose verbosity mode.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring test finished. Ignored with the default value of `null` /\n   * `undefined`.\n   * @returns Array of Scalars.\n   */\n  private testLoop(\n      f: (data: Tensor[]) => Scalar[], ins: Tensor[], batchSize?: number,\n      verbose = 0, steps?: number): Scalar[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins, batchSize, steps, 'steps');\n      const outs: Scalar[] = [];\n      if (verbose > 0) {\n        throw new NotImplementedError('Verbose mode is not implemented yet.');\n      }\n      // TODO(cais): Use `indicesForConversionToDense' to prevent slow down.\n      if (steps != null) {\n        throw new NotImplementedError(\n            'steps mode in testLoop() is not implemented yet');\n      } else {\n        const batches = makeBatches(numSamples, batchSize);\n        const indexArray = tensor1d(range(0, numSamples));\n        for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds =\n              K.sliceAlongFirstAxis(\n                  indexArray, batchStart, batchEnd - batchStart) as Tensor1D;\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Scalar[];\n          const batchOuts = f(insBatch);\n          if (batchIndex === 0) {\n            for (let i = 0; i < batchOuts.length; ++i) {\n              outs.push(scalar(0));\n            }\n          }\n          for (let i = 0; i < batchOuts.length; ++i) {\n            const batchOut = batchOuts[i];\n            outs[i] =\n                tfc.add(outs[i], tfc.mul(batchEnd - batchStart, batchOut));\n          }\n        }\n        for (let i = 0; i < outs.length; ++i) {\n          outs[i] = tfc.div(outs[i], numSamples);\n        }\n      }\n      return outs;\n    });\n  }\n\n  protected getDedupedMetricsNames(): string[] {\n    const outLabels = this.metricsNames;\n    // Rename duplicated metrics names (can happen with an output layer\n    // shared among multiple dataflows).\n    const dedupedOutLabels = [];\n    for (let i = 0; i < outLabels.length; ++i) {\n      const label = outLabels[i];\n      let newLabel = label;\n      if (count(outLabels, label) > 1) {\n        const dupIndex = count(outLabels.slice(0, i), label);\n        newLabel += `_${dupIndex}`;\n      }\n      dedupedOutLabels.push(newLabel);\n    }\n    return dedupedOutLabels;\n  }\n\n  /**\n   * Creates a function that performs the following actions:\n   *\n   * 1. computes the losses\n   * 2. sums them to get the total loss\n   * 3. call the optimizer computes the gradients of the LayersModel's\n   *    trainable weights w.r.t. the total loss and update the variables\n   * 4. calculates the metrics\n   * 5. returns the values of the losses and metrics.\n   */\n  protected makeTrainFunction(): (data: Tensor[]) => Scalar[] {\n    return (data: Tensor[]) => {\n      const lossValues: Scalar[] = [];\n\n      const inputs = data.slice(0, this.inputs.length);\n      const targets = data.slice(\n          this.inputs.length, this.inputs.length + this.outputs.length);\n      const sampleWeights = data.slice(\n          this.inputs.length + this.outputs.length,\n          this.inputs.length + this.outputs.length * 2);\n\n      const metricsValues: Scalar[] = [];\n\n      // Create a function that computes the total loss based on the\n      // inputs. This function is used for obtaining gradients through\n      // backprop.\n      const totalLossFunction = () => {\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs =\n            execute(this.outputs, feedDict, {'training': true}) as Tensor[];\n        // TODO(cais): Take care of the case of multiple outputs from a\n        //   single layer?\n\n        let totalLoss: Tensor;\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          let loss = lossFunction(targets[i], outputs[i]);\n          if (sampleWeights[i] != null) {\n            loss = computeWeightedLoss(loss, sampleWeights[i]);\n          }\n\n          // TODO(cais): push Scalar instead.\n          const meanLoss: Scalar = tfc.mean(loss);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          lossValues.push(meanLoss);\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n        }\n\n        // Compute the metrics.\n        // TODO(cais): These should probably be calculated outside\n        //   totalLossFunction to benefit speed?\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          let weightedMetric: Scalar;\n\n          if (this.outputs.length > 1 && i < this.outputs.length) {\n            weightedMetric = lossValues[i];\n          } else {\n            const metric = this.metricsTensors[i][0];\n            const outputIndex = this.metricsTensors[i][1];\n            weightedMetric =\n                tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          }\n\n          tfc.keep(weightedMetric);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          metricsValues.push(weightedMetric);\n        }\n\n        totalLoss = tfc.mean(totalLoss);\n\n        // Add regularizer penalties.\n        this.calculateLosses().forEach(regularizerLoss => {\n          totalLoss = tfc.add(totalLoss, regularizerLoss);\n        });\n\n        return totalLoss as Scalar;\n      };\n\n      const variables = this.collectedTrainableWeights.map(\n          param => param.read() as tfc.Variable);\n      const returnCost = true;\n      const totalLossValue =\n          this.optimizer_.minimize(totalLossFunction, returnCost, variables);\n\n      return [totalLossValue].concat(metricsValues);\n    };\n  }\n\n  /**\n   * Create a function which, when invoked with an array of `tf.Tensor`s as a\n   * batch of inputs, returns the prespecified loss and metrics of the model\n   * under the batch of input data.\n   */\n  private makeTestFunction() {\n    this.testFunction = (data: Tensor[]) => {\n      return tfc.tidy(() => {\n        const valOutputs: Scalar[] = [];\n        let totalLoss: Scalar;\n        const inputs = data.slice(0, this.inputs.length);\n        const targets = data.slice(\n            this.inputs.length, this.inputs.length + this.outputs.length);\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs = execute(this.outputs, feedDict) as Tensor[];\n        // Compute total loss.\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          // TODO(cais): Add sample weighting and replace the simple\n          // averaging.\n          const loss: Scalar = tfc.mean(lossFunction(targets[i], outputs[i]));\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n          valOutputs.push(totalLoss);\n        }\n        // Compute the metrics.\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          const metric = this.metricsTensors[i][0];\n          const outputIndex = this.metricsTensors[i][1];\n          // TODO(cais): Replace K.mean() with a proper weighting function.\n          const meanMetric =\n              tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          valOutputs.push(meanMetric as Scalar);\n        }\n        return valOutputs;\n      });\n    };\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a\n   * dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *     layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * for (let i = 1; i < 5 ; ++i) {\n   *   const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *       batchSize: 4,\n   *       epochs: 3\n   *   });\n   *   console.log(\"Loss after Epoch \" + i + \" : \" + h.history.loss[0]);\n   * }\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you\n   * can also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named,\n   * you can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args A `ModelFitArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input\n   * data and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    return fitTensors(this, x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below when it's possible to instantiate\n  //   actual dataset objects.\n  /**\n   * Trains the model using a dataset object.\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for training. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs.\n   *   Of the two items in the array, the first is the input feature(s) and\n   *   the second is the output target(s).\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    return fitDataset(this, dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    // TODO(cais): Support sampleWeight and classWeight.\n    // TODO(cais): Support Dataset objects.\n    const standardizeOut = await this.standardizeUserData(x, y);\n    const inputs = standardizeOut[0];\n    const targets = standardizeOut[1];\n    const trainFunction = this.makeTrainFunction();\n    const losses = trainFunction(inputs.concat(targets));\n    const lossValues: number[] = [];\n    for (const loss of losses) {\n      const v = await loss.data();\n      lossValues.push(v[0]);\n    }\n    tfc.dispose(losses);\n    return singletonOrArray(lossValues);\n  }\n\n  /**\n   * Extract weight values of the model.\n   *\n   * @param config: An instance of `io.SaveConfig`, which specifies\n   * model-saving options such as whether only trainable weights are to be\n   * saved.\n   * @returns A `NamedTensorMap` mapping original weight names (i.e.,\n   *   non-uniqueified weight names) to their values.\n   */\n  protected getNamedWeights(config?: io.SaveConfig): NamedTensor[] {\n    const namedWeights: NamedTensor[] = [];\n\n    const trainableOnly = config != null && config.trainableOnly;\n    const weights = trainableOnly ? this.trainableWeights : this.weights;\n    const weightValues = this.getWeights(trainableOnly);\n    for (let i = 0; i < weights.length; ++i) {\n      if (trainableOnly && !weights[i].trainable) {\n        // Optionally skip non-trainable weights.\n        continue;\n      }\n      namedWeights.push(\n          {name: weights[i].originalName, tensor: weightValues[i]});\n    }\n    return namedWeights;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const input = tf.input({shape: [10]});\n   * const output = tf.layers.dense({units: 1}).apply(input);\n   * const model = tf.model({inputs: [input], outputs: [output]});\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10\n   * values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    this.stopTraining_ = stop;\n  }\n\n  get stopTraining(): boolean {\n    return this.stopTraining_;\n  }\n\n  get optimizer(): Optimizer {\n    return this.optimizer_;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    if (this.optimizer_ !== optimizer) {\n      this.optimizer_ = optimizer;\n      this.isOptimizerOwned = false;\n    }\n  }\n\n  dispose(): DisposeResult {\n    const result = super.dispose();\n    if (result.refCountAfterDispose === 0 && this.optimizer != null &&\n        this.isOptimizerOwned) {\n      const numTensorsBeforeOptmizerDisposal = tfc.memory().numTensors;\n      this.optimizer_.dispose();\n      result.numDisposedVariables +=\n          numTensorsBeforeOptmizerDisposal - tfc.memory().numTensors;\n    }\n    return result;\n  }\n\n  private getLossIdentifiers(): LossIdentifier|LossIdentifier[]|\n      {[outputName: string]: LossIdentifier} {\n    let lossNames: LossIdentifier|LossIdentifier[]|\n        {[outputName: string]: LossIdentifier};\n    if (typeof this.loss === 'string') {\n      lossNames = toSnakeCase(this.loss) as LossIdentifier;\n    } else if (Array.isArray(this.loss)) {\n      for (const loss of this.loss) {\n        if (typeof loss !== 'string') {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n      lossNames = (this.loss as string[]).map(name => toSnakeCase(name)) as\n          LossIdentifier[];\n    } else {\n      const outputNames = Object.keys(this.loss);\n      lossNames = {} as {[outputName: string]: LossIdentifier};\n      const losses =\n          this.loss as {[outputName: string]: LossOrMetricFn | string};\n      for (const outputName of outputNames) {\n        if (typeof losses[outputName] === 'string') {\n          lossNames[outputName] =\n              toSnakeCase(losses[outputName] as string) as LossIdentifier;\n        } else {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n    }\n    return lossNames;\n  }\n\n  private getMetricIdentifiers(): MetricsIdentifier[]|\n      {[key: string]: MetricsIdentifier} {\n    if (typeof this.metrics === 'string' ||\n        typeof this.metrics === 'function') {\n      return [toSnakeCase(Metrics.getLossOrMetricName(this.metrics))];\n    } else if (Array.isArray(this.metrics)) {\n      return this.metrics.map(\n          metric => toSnakeCase(Metrics.getLossOrMetricName(metric)));\n    } else {\n      const metricsIdentifiers: {[key: string]: MetricsIdentifier} = {};\n      for (const key in this.metrics) {\n        metricsIdentifiers[key] =\n            toSnakeCase(Metrics.getLossOrMetricName(this.metrics[key]));\n      }\n      return metricsIdentifiers;\n    }\n  }\n\n  protected getTrainingConfig(): TrainingConfig {\n    return {\n      loss: this.getLossIdentifiers(),\n      metrics: this.getMetricIdentifiers(),\n      optimizer_config: {\n        class_name: this.optimizer.getClassName(),\n        config: this.optimizer.getConfig()\n      } as OptimizerSerialization\n    };\n    // TODO(cais): Add weight_metrics when they are supported.\n    // TODO(cais): Add sample_weight_mode when it's supported.\n    // TODO(cais): Add loss_weights when it's supported.\n  }\n\n  loadTrainingConfig(trainingConfig: TrainingConfig) {\n    if (trainingConfig.weighted_metrics != null) {\n      throw new Error('Loading weight_metrics is not supported yet.');\n    }\n    if (trainingConfig.loss_weights != null) {\n      throw new Error('Loading loss_weights is not supported yet.');\n    }\n    if (trainingConfig.sample_weight_mode != null) {\n      throw new Error('Loading sample_weight_mode is not supported yet.');\n    }\n\n    const tsConfig = convertPythonicToTs(trainingConfig.optimizer_config) as\n        serialization.ConfigDict;\n    const optimizer = deserialize(tsConfig) as Optimizer;\n\n    let loss;\n    if (typeof trainingConfig.loss === 'string') {\n      loss = toCamelCase(trainingConfig.loss);\n    } else if (Array.isArray(trainingConfig.loss)) {\n      loss = trainingConfig.loss.map(lossEntry => toCamelCase(lossEntry));\n    } else if (trainingConfig.loss != null) {\n      loss = {} as {[outputName: string]: LossIdentifier};\n      for (const key in trainingConfig.loss) {\n        loss[key] = toCamelCase(trainingConfig.loss[key]) as LossIdentifier;\n      }\n    }\n\n    let metrics;\n    if (Array.isArray(trainingConfig.metrics)) {\n      metrics = trainingConfig.metrics.map(metric => toCamelCase(metric));\n    } else if (trainingConfig.metrics != null) {\n      metrics = {} as {[outputName: string]: MetricsIdentifier};\n      for (const key in trainingConfig.metrics) {\n        metrics[key] = toCamelCase(trainingConfig.metrics[key]);\n      }\n    }\n\n    this.compile({loss, metrics, optimizer});\n  }\n\n  /**\n   * Save the configuration and/or weights of the LayersModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 2. Saving `model`'s topology and weights to browser\n   * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('indexeddb://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 3. Saving `model`'s topology and weights as two files\n   * (`my-model-1.json` and `my-model-1.weights.bin`) downloaded from\n   * browser.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('downloads://my-model-1');\n   * ```\n   *\n   * Example 4. Send  `model`'s topology and weights to an HTTP server.\n   * See the documentation of `tf.io.http` for more details\n   * including specifying request parameters and implementation of the\n   * server.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('http://my-server/model/upload');\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new ValueError(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new ValueError(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new ValueError(\n          'LayersModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    const weightDataAndSpecs =\n        await io.encodeWeights(this.getNamedWeights(config));\n\n    const returnString = false;\n    const unusedArg: {} = null;\n    const modelConfig = this.toJSON(unusedArg, returnString);\n    const modelArtifacts: io.ModelArtifacts = {\n      modelTopology: modelConfig,\n      format: LAYERS_MODEL_FORMAT_NAME,\n      generatedBy: `TensorFlow.js tfjs-layers v${version}`,\n      convertedBy: null,\n    };\n\n    const includeOptimizer = config == null ? false : config.includeOptimizer;\n    if (includeOptimizer && this.optimizer != null) {\n      modelArtifacts.trainingConfig = this.getTrainingConfig();\n      const weightType = 'optimizer';\n      const {data: optimizerWeightData, specs: optimizerWeightSpecs} =\n          await io.encodeWeights(await this.optimizer.getWeights(), weightType);\n      weightDataAndSpecs.specs.push(...optimizerWeightSpecs);\n      weightDataAndSpecs.data = io.concatenateArrayBuffers(\n          [weightDataAndSpecs.data, optimizerWeightData]);\n    }\n\n    if (this.userDefinedMetadata != null) {\n      // Check serialized size of user-defined metadata.\n      const checkSize = true;\n      checkUserDefinedMetadata(this.userDefinedMetadata, this.name, checkSize);\n      modelArtifacts.userDefinedMetadata = this.userDefinedMetadata;\n    }\n\n    modelArtifacts.weightData = weightDataAndSpecs.data;\n    modelArtifacts.weightSpecs = weightDataAndSpecs.specs;\n    return handlerOrURL.save(modelArtifacts);\n  }\n\n  /**\n   * Set user-defined metadata.\n   *\n   * The set metadata will be serialized together with the topology\n   * and weights of the model during `save()` calls.\n   *\n   * @param setUserDefinedMetadata\n   */\n  setUserDefinedMetadata(userDefinedMetadata: {}): void {\n    checkUserDefinedMetadata(userDefinedMetadata, this.name);\n    this.userDefinedMetadata = userDefinedMetadata;\n  }\n\n  /**\n   * Get user-defined metadata.\n   *\n   * The metadata is supplied via one of the two routes:\n   *   1. By calling `setUserDefinedMetadata()`.\n   *   2. Loaded during model loading (if the model is constructed\n   *      via `tf.loadLayersModel()`.)\n   *\n   * If no user-defined metadata is available from either of the\n   * two routes, this function will return `undefined`.\n   */\n  getUserDefinedMetadata(): {} {\n    return this.userDefinedMetadata;\n  }\n}\nserialization.registerClass(LayersModel);\n\n/**\n * A `tf.Functional` is an alias to `tf.LayersModel`.\n *\n * See also:\n *   `tf.LayersModel`, `tf.Sequential`, `tf.loadLayersModel`.\n */\n/** @doc {heading: 'Models', subheading: 'Classes'} */\nexport class Functional extends LayersModel {\n  static className = 'Functional';\n}\nserialization.registerClass(Functional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source keras/models.py */\n\nimport {dispose, io, NamedTensorMap, Optimizer, Scalar, serialization, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {getUid} from './backend/state';\nimport {History} from './base_callbacks';\nimport {Dataset} from './engine/dataset_stub';\nimport {Input} from './engine/input_layer';\nimport {getSourceInputs, Layer, Node, SymbolicTensor} from './engine/topology';\nimport {LayersModel, ModelCompileArgs, ModelEvaluateArgs} from './engine/training';\nimport {ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './engine/training_dataset';\nimport {ModelFitArgs} from './engine/training_tensors';\nimport {NotImplementedError, RuntimeError, ValueError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {TrainingConfig} from './keras_format/training_config';\nimport {PyJsonDict} from './keras_format/types';\nimport {deserialize} from './layers/serialization';\nimport {Kwargs, NamedTensor} from './types';\nimport * as generic_utils from './utils/generic_utils';\nimport {convertPythonicToTs} from './utils/serialization_utils';\nimport {getExactlyOneShape} from './utils/types_utils';\n\n/**\n * Parses a JSON model configuration file and returns a model instance.\n *\n * ```js\n * // This example shows how to serialize a model using `toJSON()` and\n * // deserialize it as another model using `tf.models.modelFromJSON()`.\n * // Note: this example serializes and deserializes only the topology\n * // of the model; the weights of the loaded model will be different\n * // from those of the the original model, due to random weight\n * // initialization.\n * // To load the topology and weights of a model, use `tf.loadLayersModel()`.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.repeatVector({inputShape: [2], n: 4}));\n * // Serialize `model1` as a JSON object.\n * const model1JSON = model1.toJSON(null, false);\n * model1.summary();\n *\n * const model2 = await tf.models.modelFromJSON(model1JSON);\n * model2.summary();\n * ```\n *\n *  @param modelAndWeightsConfig JSON object or string encoding a model and\n *       weights configuration. It can also be only the topology JSON of the\n *       model, in which case the weights will not be loaded.\n *  @param custom_objects Optional dictionary mapping names\n *       (strings) to custom classes or functions to be\n *       considered during deserialization.\n * @returns A TensorFlow.js Layers `tf.LayersModel` instance (uncompiled).\n */\nexport async function modelFromJSON(\n    modelAndWeightsConfig: ModelAndWeightsConfig|PyJsonDict,\n    customObjects?: serialization.ConfigDict): Promise<LayersModel> {\n  if (!('modelTopology' in modelAndWeightsConfig)) {\n    modelAndWeightsConfig = {modelTopology: modelAndWeightsConfig};\n  }\n  modelAndWeightsConfig = modelAndWeightsConfig as ModelAndWeightsConfig;\n\n  let modelTopology = modelAndWeightsConfig.modelTopology;\n  if (modelTopology['model_config'] != null) {\n    // If the model-topology JSON contains a 'model_config' field, then it is\n    // a full model JSON (e.g., from `keras.Model.save()`), which contains\n    // not only the model's architecture in its 'model_config' field, but\n    // additional information such as the model's optimizer. We use only the\n    // 'model_config' field currently.\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n  const tsConfig =\n      convertPythonicToTs(modelTopology) as serialization.ConfigDict;\n  const model = deserialize(tsConfig, customObjects) as LayersModel;\n\n  if (modelAndWeightsConfig.weightsManifest != null) {\n    // Load the weight values keyed by the original tensor names in the model\n    // file that was loaded.  These should match the keys of the weight\n    // manifest.\n    const weightValues = await io.loadWeights(\n        modelAndWeightsConfig.weightsManifest, modelAndWeightsConfig.pathPrefix,\n        model.weights.map(weight => weight.originalName));\n\n    // Map the weights to the unique tensor names generated during model loading\n    const uniqueWeightValues: NamedTensorMap = {};\n    for (const weight of model.weights) {\n      uniqueWeightValues[weight.originalName] =\n          weightValues[weight.originalName];\n    }\n\n    model.loadWeights(uniqueWeightValues);\n    // Dispose temporary weight values.\n    dispose(weightValues);\n  }\n  return model;\n}\n\n/**\n * Options for loading a saved mode in TensorFlow.js format.\n */\nexport interface ModelAndWeightsConfig {\n  /**\n   * A JSON object or JSON string containing the model config.\n   *\n   * This can be either of the following two formats:\n   *   - A model archiecture-only config,  i.e., a format consistent with the\n   *     return value of`keras.Model.to_json()`.\n   *   - A full model config, containing not only model architecture, but also\n   *     training options and state, i.e., a format consistent with the return\n   *     value of `keras.models.save_model()`.\n   */\n  modelTopology: PyJsonDict;\n\n  /**\n   * A weights manifest in TensorFlow.js format.\n   */\n  weightsManifest?: io.WeightsManifestConfig;\n\n  /**\n   * Path to prepend to the paths in `weightManifest` before fetching.\n   *\n   * The path may optionally end in a slash ('/').\n   */\n  pathPrefix?: string;\n}\n\n// TODO(nielsene): Remove after: https://github.com/tensorflow/tfjs/issues/400\nexport interface ModelPredictArgs {\n  /**\n   * Optional. Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Optional. Verbosity mode. Defaults to false.\n   */\n  verbose?: boolean;\n}\n\n/**\n * Load a model, including its topology and optionally weights.  See the\n * Tutorial named \"How to import a Keras Model\" for usage examples.\n *\n * Example 1: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 2. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * Example 4. Load a model from an HTTP server.\n *\n * ```js\n * const model = await\n *     tf.loadLayersModel('https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. This path will be\n *      interpreted as a relative HTTP path, to which `fetch` will be used to\n *      request the model topology and weight manifest JSON.\n *      The content of the JSON file is assumed to be a JSON object with the\n *      following fields and values:\n *      - 'modelTopology': A JSON object that can be either of:\n *        1. a model architecture JSON consistent with the format of the return\n *            value of `keras.Model.to_json()`\n *        2. a full model JSON in the format of `keras.models.save_model()`.\n *      - 'weightsManifest': A TensorFlow.js weights manifest.\n *      See the Python converter function `save_model()` for more details.\n *      It is also assumed that model weights can be accessed from relative\n *      paths described by the `paths` fields in weights manifest.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A progress callback of the form:\n *     `(fraction: number) => void`. This callback can be used to monitor the\n *     model-loading process.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n */\nexport async function loadLayersModelInternal(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (typeof pathOrIOHandler === 'string') {\n    const handlers = io.getLoadHandlers(pathOrIOHandler, options);\n    if (handlers.length === 0) {\n      // For backward compatibility: if no load handler can be found,\n      // assume it is a relative http path.\n      // TODO(cais): Reformat the args into a single `LoadOptions` once the core\n      // is refactored.\n      handlers.push(io.browserHTTPRequest(pathOrIOHandler, options));\n    } else if (handlers.length > 1) {\n      throw new ValueError(\n          `Found more than one (${handlers.length}) load handlers for ` +\n          `URL '${pathOrIOHandler}'`);\n    }\n    pathOrIOHandler = handlers[0];\n  }\n  return loadLayersModelFromIOHandler(pathOrIOHandler, undefined, options);\n}\n\n/**\n * Load a model and optionally its weights, using an IOHandler object.\n *\n * @param handler The instance of `IOHandler` to be used during the model\n *   loading.\n * @param customObjects Any optional custom objects to be used during model\n *   loading.\n * @param strict Whether the weight loading will be done in strict mode.\n *   Default: `true`.\n */\nexport async function loadLayersModelFromIOHandler(\n    handler: io.IOHandler, customObjects?: serialization.ConfigDict,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (handler.load == null) {\n    throw new ValueError(\n        'Cannot proceed with model loading because the IOHandler provided ' +\n        'does not have the `load` method implemented.');\n  }\n  const artifacts = await handler.load();\n  let modelTopology = artifacts.modelTopology as PyJsonDict;\n  if (modelTopology['model_config'] != null) {\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n\n  const strict = options.strict == null ? true : options.strict;\n  // If weights are provided and the weight-loading mode is strict, use\n  // fast weight initialization. This skips costly initializers such as\n  // 'orthogonal' and saves unnecessary computation in cases where\n  // the initialized weight values will immediately be overwritten by\n  // loaded weight values.\n  const fastWeightInit =\n      artifacts.weightData != null && artifacts.weightSpecs != null && strict;\n  const model =\n      deserialize(\n          convertPythonicToTs(modelTopology) as serialization.ConfigDict,\n          customObjects, fastWeightInit) as LayersModel;\n\n  const trainingConfig = artifacts.trainingConfig as TrainingConfig;\n  if (trainingConfig != null) {\n    model.loadTrainingConfig(trainingConfig);\n  }\n  if (artifacts.userDefinedMetadata != null) {\n    model.setUserDefinedMetadata(artifacts.userDefinedMetadata);\n  }\n\n  // If weightData is present, load the weights into the model.\n  if (artifacts.weightData != null) {\n    // Loading weights requires weightSpecs.\n    if (artifacts.weightSpecs == null) {\n      throw new ValueError(\n          'LayersModel artifacts contains weight data, but not weight specs. ' +\n          'Therefore loading of weights cannot proceed.');\n    }\n\n    const {modelWeights, optimizerWeights} = decodeModelAndOptimizerWeights(\n        artifacts.weightData, artifacts.weightSpecs);\n    model.loadWeights(modelWeights, strict);\n\n    if (model.optimizer != null && optimizerWeights.length > 0) {\n      await model.optimizer.setWeights(optimizerWeights);\n    }\n\n    // Dispose temporary weight values.\n    dispose(modelWeights);\n    dispose(optimizerWeights.map(w => w.tensor));\n  }\n  return model;\n}\n\nfunction decodeModelAndOptimizerWeights(\n    buffer: ArrayBuffer, specs: io.WeightsManifestEntry[]):\n    {modelWeights: NamedTensorMap, optimizerWeights: NamedTensor[]} {\n  const name2Tensor = io.decodeWeights(buffer, specs);\n  const modelWeights: NamedTensorMap = {};\n  const optimizerWeights: NamedTensor[] = [];\n  specs.forEach(spec => {\n    if (spec.group === 'optimizer') {\n      optimizerWeights.push({name: spec.name, tensor: name2Tensor[spec.name]});\n    } else {\n      modelWeights[spec.name] = name2Tensor[spec.name];\n    }\n  });\n  return {modelWeights, optimizerWeights};\n}\n\n/**\n * Configuration for a Sequential model.\n */\nexport interface SequentialArgs {\n  /** Stack of layers for the model. */\n  layers?: Layer[];\n\n  /** The name of this model. */\n  name?: string;\n}\n\n/**\n * A model with a stack of layers, feeding linearly from one to the next.\n *\n * `tf.sequential` is a factory function that creates an instance of\n * `tf.Sequential`.\n *\n * ```js\n *  // Define a model for linear regression.\n *  const model = tf.sequential();\n *  model.add(tf.layers.dense({units: 1, inputShape: [1]}));\n *\n *  // Prepare the model for training: Specify the loss and the optimizer.\n *  model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n *\n *  // Generate some synthetic data for training.\n *  const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);\n *  const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);\n *\n *  // Train the model using the data then do inference on a data point the\n *  // model hasn't seen:\n *  await model.fit(xs, ys);\n *  model.predict(tf.tensor2d([5], [1, 1])).print();\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class Sequential extends LayersModel {\n  /** @nocollapse */\n  static className = 'Sequential';\n  private model: LayersModel;\n  constructor(args?: SequentialArgs) {\n    super({inputs: [], outputs: []});\n    args = args || {};\n\n    this.trainable = true;\n    this.built = false;\n\n    // Set model name.\n    this.name = (args.name != null) ? args.name : getUid('sequential_');\n\n    // Add to the model any layers passed to the constructor.\n    if (args.layers != null) {\n      for (const layer of args.layers) {\n        this.add(layer);\n      }\n    }\n  }\n\n  // Helper function to Sequential.add  Throws if the new output shape will be\n  // invalid.\n  private checkShape(layer: Layer) {\n    const shape = layer.inboundNodes[0].outputTensors[0].shape;\n    if (shape.some(x => x < 0)) {\n      throw new ValueError(\n          'Negative dimension size caused by adding layer ' +\n          `${layer.name} with input shape [` +\n          `${layer.inboundNodes[0].inputTensors[0].shape}]`);\n    }\n  }\n\n  /**\n   * Adds a layer instance on top of the layer stack.\n   *\n   * ```js\n   *  const model = tf.sequential();\n   *  model.add(tf.layers.dense({units: 8, inputShape: [1]}));\n   *  model.add(tf.layers.dense({units: 4, activation: 'relu6'}));\n   *  model.add(tf.layers.dense({units: 1, activation: 'relu6'}));\n   *  // Note that the untrained model is random at this point.\n   *  model.predict(tf.randomNormal([10, 1])).print();\n   * ```\n   * @param layer Layer instance.\n   *\n   * @exception ValueError In case the `layer` argument does not know its\n   * input shape.\n   * @exception ValueError In case the `layer` argument has multiple output\n   *   tensors, or is already connected somewhere else (forbidden in\n   *   `Sequential` models).\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  add(layer: Layer): void {\n    const isLayerModelInstance =\n        layer instanceof Sequential || layer instanceof LayersModel;\n    let modelLayer: LayersModel;\n    if (isLayerModelInstance) {\n      modelLayer = layer as LayersModel;\n      if (modelLayer.outputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      if (modelLayer.inputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single input tensor. ' +\n            'For multi-input layers, ' +\n            'use the functional API.');\n      }\n    }\n\n    if (this.outputs.length === 0) {\n      // first layer in model: check that it is an input layer\n      if (layer.inboundNodes.length === 0) {\n        // create an input layer\n        if (layer.batchInputShape == null) {\n          throw new ValueError(\n              'The first layer in a Sequential model must ' +\n              'get an `inputShape` or `batchInputShape` argument.');\n        }\n        // Instantiate the input layer.\n        const x = Input({\n          batchShape: layer.batchInputShape,\n          dtype: layer.dtype,\n          name: layer.name + '_input'\n        });\n        // This will build the current layer and create the node connecting\n        // the current layer to the input layer we just created.\n        layer.apply(x);\n      }\n\n      if (isLayerModelInstance) {\n        this.outputs = modelLayer.outputs;\n        this.inputs = modelLayer.inputs;\n      } else {\n        if (layer.inboundNodes.length !== 1) {\n          throw new ValueError(\n              'A layer added to a Sequential model must not already be ' +\n              `connected somewhere else. LayersModel received layer ${\n                  layer.name} ` +\n              `which has ${layer.inboundNodes.length} pre-existing inbound ` +\n              'connections.');\n        }\n\n        if (layer.inboundNodes[0].outputTensors.length !== 1) {\n          throw new ValueError(\n              'All layers in a Sequential model ' +\n              'should have a single output tensor. ' +\n              'For multi-output layers, ' +\n              'use the functional API.');\n        }\n        this.checkShape(layer);\n        this.outputs = [layer.inboundNodes[0].outputTensors[0]];\n        this.inputs = getSourceInputs(this.outputs[0]);\n      }\n\n      this.inboundNodes = [];\n      // We create an input node, which we will keep updated\n      // as we add more layers.\n      // (This call has side effects.)\n      // tslint:disable-next-line:no-unused-expression\n      new Node({\n        outboundLayer: this,\n        inboundLayers: [],\n        nodeIndices: [],\n        tensorIndices: [],\n        inputTensors: this.inputs,\n        outputTensors: this.outputs,\n        // no model-level masking for now\n        inputMasks: generic_utils.pyListRepeat(null, this.inputs.length),\n        outputMasks: [null],\n        inputShapes: this.inputs.map(x => x.shape),\n        outputShapes: this.outputs[0].shape\n      });\n    } else {\n      const outputTensor = layer.apply(this.outputs[0]);\n      if (Array.isArray(outputTensor)) {\n        throw new TypeError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      this.checkShape(layer);\n      this.outputs = [outputTensor as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n\n    this.layers.push(layer);\n    this.built = false;\n  }\n\n  /**\n   * Removes the last layer in the model.\n   *\n   * @exception TypeError if there are no layers in the model.\n   */\n  pop(): void {\n    if (this.layers.length === 0) {\n      throw new TypeError('There are no layers in the model.');\n    }\n\n    this.layers.pop();\n    if (this.layers.length === 0) {\n      this.outputs = [];\n      this.inboundNodes = [];\n      this.outboundNodes = [];\n    } else {\n      const lastLayerIndex = this.layers.length - 1;\n      this.layers[lastLayerIndex].outboundNodes = [];\n      this.outputs = [this.layers[lastLayerIndex].output as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.call(inputs, kwargs);\n  }\n\n  build(inputShape?: Shape|Shape[]) {\n    // Call `getExactlyOneShape` without using its return value,\n    // to verify that exactly one input shape is provided.\n    getExactlyOneShape(inputShape);\n\n    if (this.inputs.length === 0 || this.outputs.length === 0) {\n      throw new TypeError(\n          'Sequential model cannot be built: model is empty.' +\n          ' Add some layers first.');\n    }\n    // actually create the model\n    this.model = new LayersModel({\n      inputs: this.inputs,\n      outputs: this.outputs[0],\n      name: this.name + '_model'\n    });\n    this.model.trainable = this.trainable;\n\n    // mirror model attributes\n    this.supportsMasking = this.model.supportsMasking;\n    // TODO(michaelterry): Add caches\n    this.inputLayers = this.model.inputLayers;\n    this.inputLayersNodeIndices = this.model.inputLayersNodeIndices;\n    this.inputLayersTensorIndices = this.model.inputLayersTensorIndices;\n    this.outputLayers = this.model.outputLayers;\n    this.outputLayersNodeIndices = this.model.outputLayersNodeIndices;\n    this.outputLayersTensorIndices = this.model.outputLayersTensorIndices;\n    this.nodesByDepth = this.model.nodesByDepth;\n    this.containerNodes = this.model.containerNodes;\n    this.outputNames = this.model.outputNames;\n    this.inputNames = this.model.inputNames;\n    // TODO(michaelterry): Add feedInputNames, feedInputs, if needed.\n    // TODO(michaelterry): Add callbackModel if needed.\n    this.built = true;\n  }\n\n  countParams(): number {\n    if (!this.built) {\n      this.build();\n    }\n    return super.countParams();\n  }\n\n  /**\n   * Print a text summary of the Sequential model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - The total number of trainable and non-trainable parameters of the\n   * model.\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(\n   *     tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'}));\n   * model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n   *\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      this.build();\n    }\n    super.summary(lineLength, positions, printFn);\n  }\n\n  /**\n   * Sets the weights of the model.\n   *\n   * @param weights Should be a list of Tensors with shapes and types matching\n   *   the output of `model.getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    if (this.model == null) {\n      this.build();\n    }\n    this.model.setWeights(weights);\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   * });\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateConfig`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluate(x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluateDataset(dataset, args);\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([2, 10])).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param conifg A `ModelPredictConfig` object containing optional fields.\n   *\n   * @return `tf.Tensor`(s) of predictions.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predict(x, args);\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * @param x: Input samples, as a Tensor, or list of Tensors (if the model\n   *   has multiple inputs).\n   * @return Tensor(s) of predictions\n   */\n  predictOnBatch(x: Tensor): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predictOnBatch(x);\n  }\n\n  /**\n   * See `LayersModel.compile`.\n   *\n   * @param args\n   */\n  compile(args: ModelCompileArgs): void {\n    this.build();\n    this.model.compile(args);\n    this.optimizer_ = this.model.optimizer;\n    // tslint:disable-next-line:no-any\n    this.isOptimizerOwned = (this.model as any).isOptimizerOwned;\n    this.loss = this.model.loss;\n    this.metrics = this.model.metrics;\n    // TODO(cais): Add this.lossWeights, this.sampleWeightMode,\n    //   this.weightedMetrics, this.targets.\n    this.metricsTensors = this.model.metricsTensors;\n    this.metricsNames = this.model.metricsNames;\n    // TODO(cais): Add sampleWeights.\n  }\n\n  get optimizer(): Optimizer {\n    return this.model == null ? undefined : this.model.optimizer;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    this.model.optimizer = optimizer;\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   *   epochs: 3\n   * });\n   * console.log(history.history.loss[0]);\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you can\n   * also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named, you\n   *  can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args  A `ModelFitConfig`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fit(x, y, args);\n  }\n\n  /**\n   * Trains the model using a dataset object.\n   *\n   * ```js\n   * const xArray = [\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   * ];\n   * const yArray = [1, 1, 1, 1];\n   * // Create a dataset from the JavaScript array.\n   * const xDataset = tf.data.array(xArray);\n   * const yDataset = tf.data.array(yArray);\n   * // Zip combines the `x` and `y` Datasets into a single Dataset, the\n   * // iterator of which will return an object containing of two tensors,\n   * // corresponding to `x` and `y`.  The call to `batch(4)` will bundle\n   * // four such samples into a single object, with the same keys now pointing\n   * // to tensors that hold 4 examples, organized along the batch dimension.\n   * // The call to `shuffle(4)` causes each iteration through the dataset to\n   * // happen in a different order.  The size of the shuffle window is 4.\n   * const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset})\n   *     .batch(4)\n   *     .shuffle(4);\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [9]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fitDataset(xyDataset, {\n   *   epochs: 4,\n   *   callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)}\n   * });\n   * ```\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected to\n   *   generate a dataset iterator object, the `next()` method of which is\n   *   expected to produce data batches for evaluation. The return value of the\n   *   `next()` call ought to contain a boolean `done` field and a `value`\n   *   field.\n   *\n   *   The `value` field is expected to be an object of with fields\n   *   `xs` and `ys`, which point to the feature tensor and the target tensor,\n   *   respectively. This case is for models with exactly one input and one\n   *   output (e.g.. a sequential model). For example:\n   *   ```js\n   *   {value: {xs: xsTensor, ys: ysTensor}, done: false}\n   *   ```\n   *\n   *   If the model has multiple inputs, the `xs` field of `value` should\n   *   be an object mapping input names to their respective feature tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: {\n   *         input_1: xsTensor1,\n   *         input_2: xsTensor2\n   *       },\n   *       ys: ysTensor\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   *   If the model has multiple outputs, the `ys` field of `value` should\n   *   be an object mapping output names to their respective target tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: xsTensor,\n   *       ys: {\n   *         output_1: ysTensor1,\n   *         output_2: ysTensor2\n   *       },\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fitDataset(dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    return this.model.trainOnBatch(x, y);\n  }\n\n  /* See parent class for JsDoc */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    let configArray: serialization.ConfigDictArray;\n    let extraModelConfig: serialization.ConfigDict = {};\n    if (config instanceof Array) {\n      if (!(config[0].className != null) ||\n          config[0]['className'] === 'Merge') {\n        throw new ValueError('Legacy serialization format not supported yet.');\n      }\n      configArray = config;\n    } else {\n      util.assert(\n          config['layers'] != null,\n          () =>\n              `When the config data for a Sequential model is not an Array, ` +\n              `it must be an Object that contains the 'layers' field.`);\n      configArray = config['layers'] as serialization.ConfigDictArray;\n      delete config['layers'];\n      extraModelConfig = config;\n    }\n\n    const model = new cls(extraModelConfig);\n    if (!(model instanceof Sequential)) {\n      throw new NotImplementedError(\n          `Sequential.fromConfig called on non-Sequential input: ${model}`);\n    }\n    for (const conf of configArray) {\n      const customObjects: serialization.ConfigDict = undefined;\n      const layer = deserialize(\n                        conf as serialization.ConfigDict, customObjects,\n                        fastWeightInit) as Layer;\n      if (fastWeightInit) {\n        layer.setFastWeightInitDuringBuild(true);\n      }\n      model.add(layer);\n    }\n    return model;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(tf.layers.dense({units: 1, inputShape: [10]}));\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10 values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    // TODO(cais): When refactoring to remove the composition pattern happens,\n    // remove this method overriding.\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot set the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    this.model.stopTraining = stop;\n  }\n\n  get stopTraining(): boolean {\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot get the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    return this.model.stopTraining;\n  }\n\n  // TODO(cais): Override get trainableWeights() here\n\n  // tslint:disable-next-line:no-any\n  getConfig(): any {\n    // NOTE(cais): We override the return type of getConfig() to `any` here,\n    //   because the `Sequential` class is a special case among `Container`\n    //   subtypes in that its getConfig() method returns an Array (not a\n    //   dict).\n    const layers: serialization.ConfigDict[] = [];\n    for (const layer of this.layers) {\n      const dict: serialization.ConfigDict = {};\n      dict['className'] = layer.getClassName();\n      dict['config'] = layer.getConfig();\n      layers.push(dict);\n    }\n    return {name: this.name, layers};\n  }\n}\nserialization.registerClass(Sequential);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Exported functions.\n */\n\nimport {io} from '@tensorflow/tfjs-core';\n\nimport {BaseCallbackConstructor, CallbackConstructorRegistry} from './base_callbacks';\nimport {ContainerArgs} from './engine/container';\nimport {Input, InputConfig,} from './engine/input_layer';\nimport {SymbolicTensor} from './engine/topology';\nimport {LayersModel} from './engine/training';\nimport {loadLayersModelInternal, Sequential, SequentialArgs} from './models';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// LayersModel and related factory methods.\n\n/**\n * A model is a data structure that consists of `Layers` and defines inputs\n * and outputs.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.model` is more generic, supporting an arbitrary graph (without\n * cycles) of layers. `tf.sequential` is less generic and supports only a linear\n * stack of layers.\n *\n * When creating a `tf.LayersModel`, specify its input(s) and output(s). Layers\n * are used to wire input(s) to output(s).\n *\n * For example, the following code snippet defines a model consisting of\n * two `dense` layers, with 10 and 4 units, respectively.\n *\n * ```js\n * // Define input, which has a size of 5 (not including batch dimension).\n * const input = tf.input({shape: [5]});\n *\n * // First dense layer uses relu activation.\n * const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});\n * // Second dense layer uses softmax activation.\n * const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});\n *\n * // Obtain the output symbolic tensor by applying the layers on the input.\n * const output = denseLayer2.apply(denseLayer1.apply(input));\n *\n * // Create the model based on the inputs.\n * const model = tf.model({inputs: input, outputs: output});\n *\n * // The model can be used for training, evaluation and prediction.\n * // For example, the following line runs prediction with the model on\n * // some fake data.\n * model.predict(tf.ones([2, 5])).print();\n * ```\n * See also:\n *   `tf.sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function model(args: ContainerArgs): LayersModel {\n  return new LayersModel(args);\n}\n\n/**\n * Creates a `tf.Sequential` model.  A sequential model is any model where the\n * outputs of one layer are the inputs to the next layer, i.e. the model\n * topology is a simple 'stack' of layers, with no branching or skipping.\n *\n * This means that the first layer passed to a `tf.Sequential` model should have\n * a defined input shape. What that means is that it should have received an\n * `inputShape` or `batchInputShape` argument, or for some type of layers\n * (recurrent, Dense...) an `inputDim` argument.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.sequential` is less generic, supporting only a linear stack of layers.\n * `tf.model` is more generic and supports an arbitrary graph (without\n * cycles) of layers.\n *\n * Examples:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have an input shape defined.\n * model.add(tf.layers.dense({units: 32, inputShape: [50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output, which equals\n * // `[null, 4]`. The 1st dimension is the undetermined batch dimension; the\n * // 2nd is the output size of the model's last layer.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * It is also possible to specify a batch size (with potentially undetermined\n * batch dimension, denoted by \"null\") for the first layer using the\n * `batchInputShape` key. The following example is equivalent to the above:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have a defined input shape\n * model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * You can also use an `Array` of already-constructed `Layer`s to create\n * a `tf.Sequential` model:\n *\n * ```js\n * const model = tf.sequential({\n *   layers: [tf.layers.dense({units: 32, inputShape: [50]}),\n *            tf.layers.dense({units: 4})]\n * });\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function sequential(config?: SequentialArgs): Sequential {\n  return new Sequential(config);\n}\n\n/**\n * Load a model composed of Layer objects, including its topology and optionally\n * weights. See the Tutorial named \"How to import a Keras Model\" for usage\n * examples.\n *\n * This method is applicable to:\n *\n * 1. Models created with the `tf.layers.*`, `tf.sequential`, and\n * `tf.model` APIs of TensorFlow.js and later saved with the\n * `tf.LayersModel.save` method.\n * 2. Models converted from Keras or TensorFlow tf.keras using the\n * [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).\n *\n * This mode is *not* applicable to TensorFlow `SavedModel`s or their converted\n * forms. For those models, use `tf.loadGraphModel`.\n *\n * Example 1. Load a model from an HTTP server.\n *\n * ```js\n * const model = await tf.loadLayersModel(\n *     'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * Example 2: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 4. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. For file://\n *      (tfjs-node-only), http:// and https:// schemas, the path can be\n *      either absolute or relative.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A function of the signature `(fraction: number) => void',\n *     that can be used as the progress callback for the model loading.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport function loadLayersModel(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  return loadLayersModelInternal(pathOrIOHandler, options);\n}\n\n/**\n * Used to instantiate an input to a model as a `tf.SymbolicTensor`.\n *\n * Users should call the `input` factory function for\n * consistency with other generator functions.\n *\n * Example:\n *\n * ```js\n * // Defines a simple logistic regression model with 32 dimensional input\n * // and 3 dimensional output.\n * const x = tf.input({shape: [32]});\n * const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x);\n * const model = tf.model({inputs: x, outputs: y});\n * model.predict(tf.ones([2, 32])).print();\n * ```\n *\n * Note: `input` is only necessary when using `model`. When using\n * `sequential`, specify `inputShape` for the first layer or use `inputLayer`\n * as the first layer.\n *\n * @doc {heading: 'Models', subheading: 'Inputs'}\n */\nexport function input(config: InputConfig): SymbolicTensor {\n  return Input(config);\n}\n\nexport function registerCallbackConstructor(\n    verbosityLevel: number,\n    callbackConstructor: BaseCallbackConstructor): void {\n  CallbackConstructorRegistry.registerCallbackConstructor(\n      verbosityLevel, callbackConstructor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Layer activation functions\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {ActivationIdentifier} from './keras_format/activation_config';\nimport {deserializeKerasObject} from './utils/generic_utils';\n\n/**\n * Base class for Activations.\n *\n * Special note: due to cross-language compatibility reasons, the\n * static readonly className field in this family of classes must be set to\n * the initialLowerCamelCase name of the activation.\n */\nexport abstract class Activation extends serialization.Serializable {\n  abstract apply(tensor: Tensor, axis?: number): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\n/**\n * Exponential linear unit (ELU).\n * Reference: https://arxiv.org/abs/1511.07289\n */\nexport class Elu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'elu';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x: Input.\n   * @param alpha: Scaling factor the negative section.\n   * @return Output of the ELU activation.\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return K.elu(x, alpha);\n  }\n}\nserialization.registerClass(Elu);\n\n/**\n * Scaled Exponential Linear Unit. (Klambauer et al., 2017).\n * Reference: Self-Normalizing Neural Networks, https://arxiv.org/abs/1706.02515\n * Notes:\n *   - To be used together with the initialization \"lecunNormal\".\n *   - To be used together with the dropout variant \"AlphaDropout\".\n */\nexport class Selu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'selu';\n  apply(x: Tensor): Tensor {\n    return tfc.selu(x);\n  }\n}\nserialization.registerClass(Selu);\n\n/**\n *  Rectified linear unit\n */\nexport class Relu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu';\n  apply(x: Tensor): Tensor {\n    return tfc.relu(x);\n  }\n}\nserialization.registerClass(Relu);\n\n/**\n * Rectified linear unit activation maxing out at 6.0.\n */\nexport class Relu6 extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu6';\n  apply(x: Tensor): Tensor {\n    return tidy(() => tfc.minimum(6.0, tfc.relu(x)));\n  }\n}\nserialization.registerClass(Relu6);\n\n//* Linear activation (no-op) */\nexport class Linear extends Activation {\n  /** @nocollapse */\n  static readonly className = 'linear';\n  apply(x: Tensor): Tensor {\n    return x;\n  }\n}\nserialization.registerClass(Linear);\n\n/**\n * Sigmoid activation function.\n */\nexport class Sigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'sigmoid';\n  apply(x: Tensor): Tensor {\n    return tfc.sigmoid(x);\n  }\n}\nserialization.registerClass(Sigmoid);\n\n/**\n * Segment-wise linear approximation of sigmoid.\n */\nexport class HardSigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'hardSigmoid';\n  apply(x: Tensor): Tensor {\n    return K.hardSigmoid(x);\n  }\n}\nserialization.registerClass(HardSigmoid);\n\n/**\n * Softplus activation function.\n */\nexport class Softplus extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softplus';\n  apply(x: Tensor): Tensor {\n    return tfc.softplus(x);\n  }\n}\nserialization.registerClass(Softplus);\n\n/**\n * Softsign activation function.\n */\nexport class Softsign extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softsign';\n  apply(x: Tensor): Tensor {\n    return K.softsign(x);\n  }\n}\nserialization.registerClass(Softsign);\n\n/**\n * Hyperbolic tangent function.\n */\nexport class Tanh extends Activation {\n  /** @nocollapse */\n  static readonly className = 'tanh';\n  apply(x: Tensor): Tensor {\n    return tfc.tanh(x);\n  }\n}\nserialization.registerClass(Tanh);\n\n/**\n * Softmax activation function\n */\nexport class Softmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softmax';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.softmax(x, axis);\n  }\n}\nserialization.registerClass(Softmax);\n\n/**\n * Log softmax activation function\n */\nexport class LogSoftmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'logSoftmax';\n  /**\n   * Calculate the activation function of log softmax:\n   * log( exp(x_i) / sum(exp(x)) )\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.logSoftmax(x, axis);\n  }\n}\nserialization.registerClass(LogSoftmax);\n\n/**\n * Swish activation function\n */\nexport class Swish extends Activation {\n  /** @nocollapse */\n  static readonly className = 'swish';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param alpha Scaling factor for the sigmoid function.\n   * @returns a Tensor of the same shape as x\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return tidy(() => tfc.sigmoid(x.mul(alpha)).mul(x));\n  }\n}\nserialization.registerClass(Swish);\n\nexport function serializeActivation(activation: Activation): string {\n  return activation.getClassName();\n}\n\nexport function deserializeActivation(\n   config: serialization.ConfigDict,\n   customObjects: serialization.ConfigDict = {}): Activation {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'activation');\n}\n\nexport function getActivation(identifier: ActivationIdentifier|\n                              serialization.ConfigDict|Activation): Activation {\n  if (identifier == null) {\n    const config: serialization.ConfigDict = {};\n    config['className'] = 'linear';\n    config['config'] = {};\n    return deserializeActivation(config);\n  }\n  if (typeof identifier === 'string') {\n    const config: serialization.ConfigDict = {};\n    config['className'] = identifier;\n    config['config'] = {};\n    return deserializeActivation(config);\n  } else if (identifier instanceof Activation) {\n    return identifier;\n  } else {\n    return deserializeActivation(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* original source: keras/regularizers.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {abs, add, Scalar, serialization, sum, Tensor, tidy, zeros} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\nfunction assertObjectArgs(args: L1Args | L2Args | L1L2Args): void {\n  if (args != null && typeof args !== 'object') {\n    throw new Error(\n        `Argument to L1L2 regularizer's constructor is expected to be an ` +\n        `object, but received: ${args}`);\n  }\n}\n\n/**\n * Regularizer base class.\n */\nexport abstract class Regularizer extends serialization.Serializable {\n  abstract apply(x: Tensor): Scalar;\n}\n\nexport interface L1L2Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1?: number;\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2?: number;\n}\n\nexport interface L1Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1: number;\n}\n\nexport interface L2Args {\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2: number;\n}\n\nexport class L1L2 extends Regularizer {\n  /** @nocollapse */\n  static className = 'L1L2';\n\n  private readonly l1: number;\n  private readonly l2: number;\n  private readonly hasL1: boolean;\n  private readonly hasL2: boolean;\n  constructor(args?: L1L2Args) {\n    super();\n\n    assertObjectArgs(args);\n\n    this.l1 = args == null || args.l1 == null ? 0.01 : args.l1;\n    this.l2 = args == null || args.l2 == null ? 0.01 : args.l2;\n    this.hasL1 = this.l1 !== 0;\n    this.hasL2 = this.l2 !== 0;\n  }\n\n  /**\n   * Porting note: Renamed from __call__.\n   * @param x Variable of which to calculate the regularization score.\n   */\n  apply(x: Tensor): Scalar {\n    return tidy(() => {\n      let regularization: Tensor = zeros([1]);\n      if (this.hasL1) {\n        regularization = add(regularization, sum(tfc.mul(this.l1, abs(x))));\n      }\n      if (this.hasL2) {\n        regularization =\n            add(regularization, sum(tfc.mul(this.l2, K.square(x))));\n      }\n      return regularization.asScalar();\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {'l1': this.l1, 'l2': this.l2};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls({l1: config['l1'] as number, l2: config['l2'] as number});\n  }\n}\nserialization.registerClass(L1L2);\n\nexport function l1(args?: L1Args) {\n  assertObjectArgs(args);\n  return new L1L2({l1: args != null ? args.l1 : null, l2: 0});\n}\n\nexport function l2(args: L2Args) {\n  assertObjectArgs(args);\n  return new L1L2({l2: args != null ? args.l2 : null, l1: 0});\n}\n\n/** @docinline */\nexport type RegularizerIdentifier = 'l1l2'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding keras symbols.\nexport const REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in RegularizerIdentifier]: string} = {\n      'l1l2': 'L1L2'\n    };\n\nexport function serializeRegularizer(constraint: Regularizer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeRegularizer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Regularizer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'regularizer');\n}\n\nexport function getRegularizer(identifier: RegularizerIdentifier|\n                               serialization.ConfigDict|\n                               Regularizer): Regularizer {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeRegularizer(config);\n  } else if (identifier instanceof Regularizer) {\n    return identifier;\n  } else {\n    return deserializeRegularizer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n *  Advanced activation layers.\n */\n\nimport {clipByValue, elu, leakyRelu, prelu, relu, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {Softmax as softmaxActivation} from '../activations';\nimport {cast} from '../backend/tfjs_backend';\nimport {Constraint, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface ReLULayerArgs extends LayerArgs {\n  /**\n   * Float, the maximum output value.\n   */\n  maxValue?: number;\n}\n\nexport class ReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ReLU';\n  maxValue: number;\n\n  constructor(args?: ReLULayerArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maxValue = args.maxValue;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    let output = relu(inputs);\n    if (this.maxValue != null) {\n      output = clipByValue(output, 0, this.maxValue);\n    }\n    return output;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {maxValue: this.maxValue};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ReLU);\n\nexport declare interface LeakyReLULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `0.3`.\n   */\n  alpha?: number;\n}\n\nexport class LeakyReLU extends Layer {\n  /** @nocollapse */\n  static className = 'LeakyReLU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 0.3;\n\n  constructor(args?: LeakyReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return leakyRelu(x, this.alpha);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LeakyReLU);\n\nexport declare interface PReLULayerArgs extends LayerArgs {\n  /**\n   * Initializer for the learnable alpha.\n   */\n  alphaInitializer?: Initializer|InitializerIdentifier;\n\n  /**\n   * Regularizer for the learnable alpha.\n   */\n  alphaRegularizer?: Regularizer;\n\n  /**\n   * Constraint for the learnable alpha.\n   */\n  alphaConstraint?: Constraint;\n\n  /**\n   * The axes along which to share learnable parameters for the activation\n   * function. For example, if the incoming feature maps are from a 2D\n   * convolution with output shape `[numExamples, height, width, channels]`,\n   * and you wish to share parameters across space (height and width) so that\n   * each filter channels has only one set of parameters, set\n   * `shared_axes: [1, 2]`.\n   */\n  sharedAxes?: number|number[];\n}\n\nexport class PReLU extends Layer {\n  /** @nocollapse */\n  static className = 'PReLU';\n  private readonly alphaInitializer: Initializer;\n  private readonly alphaRegularizer: Regularizer;\n  private readonly alphaConstraint: Constraint;\n  private readonly sharedAxes: number[];\n  private alpha: LayerVariable;\n\n  readonly DEFAULT_ALPHA_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args?: PReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.supportsMasking = true;\n    this.alphaInitializer =\n        getInitializer(args.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER);\n    this.alphaRegularizer = getRegularizer(args.alphaRegularizer);\n    this.alphaConstraint = getConstraint(args.alphaConstraint);\n    if (args.sharedAxes == null) {\n      this.sharedAxes = null;\n    } else if (Array.isArray(args.sharedAxes)) {\n      this.sharedAxes = args.sharedAxes;\n    } else if (typeof args.sharedAxes === 'number') {\n      this.sharedAxes = [args.sharedAxes];\n    } else {\n      throw new ValueError(\n          `Expected sharedAxes to be a number or an array of numbers, ` +\n          `but got ${args.sharedAxes}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]) {\n    inputShape = getExactlyOneShape(inputShape);\n    const paramShape: Shape = inputShape.slice(1);\n    if (this.sharedAxes != null) {\n      for (const i of this.sharedAxes) {\n        paramShape[i - 1] = 1;\n      }\n    }\n    this.alpha = this.addWeight(\n        'alpha', paramShape, 'float32', this.alphaInitializer,\n        this.alphaRegularizer, true, this.alphaConstraint);\n    // Set input spec.\n    const axes: {[axis: number]: number} = {};\n    if (this.sharedAxes != null) {\n      for (let i = 1; i < inputShape.length; ++i) {\n        axes[i] = inputShape[i];\n      }\n    }\n    this.inputSpec = [new InputSpec({\n      ndim: inputShape.length,\n      axes,\n    })];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    return prelu(inputs, this.alpha.read());\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      alphaInitializer: serializeInitializer(this.alphaInitializer),\n      alphaRegularizer: serializeRegularizer(this.alphaRegularizer),\n      alphaConstraint: serializeConstraint(this.alphaConstraint),\n      sharedAxes: this.sharedAxes\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(PReLU);\n\nexport declare interface ELULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `1.0`.\n   */\n  alpha?: number;\n}\n\nexport class ELU extends Layer {\n  /** @nocollapse */\n  static className = 'ELU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 1.0;\n\n  constructor(args?: ELULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    if (args.alpha != null && args.alpha !== this.DEFAULT_ALPHA) {\n      throw new NotImplementedError(\n          `Non-default alpha value (${args.alpha}) is not supported by the ` +\n          `ELU layer yet.`);\n    }\n\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return elu(x);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ELU);\n\nexport declare interface ThresholdedReLULayerArgs extends LayerArgs {\n  /**\n   * Float >= 0. Threshold location of activation.\n   */\n  theta?: number;\n}\n\nexport class ThresholdedReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ThresholdedReLU';\n  readonly theta: number;\n\n  readonly DEFAULT_THETA = 1.0;\n\n  constructor(args?: ThresholdedReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.theta = args.theta == null ? this.DEFAULT_THETA : args.theta;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return x.mul(cast(x.greater(this.theta), 'float32'));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {theta: this.theta};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ThresholdedReLU);\n\nexport declare interface SoftmaxLayerArgs extends LayerArgs {\n  /**\n   * Integer, axis along which the softmax normalization is applied.\n   * Defaults to `-1` (i.e., the last axis).\n   */\n  axis?: number;\n}\n\nexport class Softmax extends Layer {\n  /** @nocollapse */\n  static className = 'Softmax';\n  readonly axis: number;\n  readonly softmax: (t: Tensor, a?: number) => Tensor;\n  readonly DEFAULT_AXIS = 1.0;\n\n  constructor(args?: SoftmaxLayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.softmax = new softmaxActivation().apply;\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return this.softmax(x, this.axis);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {axis: this.axis};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Softmax);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {ValueError} from '../errors';\nimport {PaddingMode} from '../keras_format/common';\n\nimport {pyListRepeat} from './generic_utils';\nimport {isInteger, max} from './math_utils';\n\n/**\n * Transforms a single number of array of numbers into an array of numbers.\n * @param value\n * @param n: The size of the tuple to be returned.\n * @param name: Name of the parameter, used for generating error messages.\n * @returns An array of numbers.\n */\nexport function normalizeArray(\n    value: number|number[], n: number, name: string): number[] {\n  if (typeof value === 'number') {\n    return pyListRepeat(value, n);\n  } else {\n    if (value.length !== n) {\n      throw new ValueError(\n          `The ${name} argument must be an integer or tuple of ${n} integers.` +\n          ` Received: ${value.length} elements.`);\n    }\n    for (let i = 0; i < n; ++i) {\n      const singleValue = value[i];\n      if (!isInteger(singleValue)) {\n        throw new ValueError(\n            `The ${name} argument must be an integer or tuple of ${n}` +\n            ` integers. Received: ${JSON.stringify(value)} including a` +\n            ` non-integer number ${singleValue}`);\n      }\n    }\n    return value;\n  }\n}\n\n/**\n * Determines output length of a convolution given input length.\n * @param inputLength\n * @param filterSize\n * @param padding\n * @param stride\n * @param dilation: dilation rate.\n */\nexport function convOutputLength(\n    inputLength: number, filterSize: number, padding: PaddingMode,\n    stride: number, dilation = 1): number {\n  if (inputLength == null) {\n    return inputLength;\n  }\n  const dilatedFilterSize = filterSize + (filterSize - 1) * (dilation - 1);\n  let outputLength: number;\n  if (padding === 'same') {\n    outputLength = inputLength;\n  } else {  // VALID\n    outputLength = inputLength - dilatedFilterSize + 1;\n  }\n  return Math.floor((outputLength + stride - 1) / stride);\n}\n\nexport function deconvLength(\n    dimSize: number, strideSize: number, kernelSize: number,\n    padding: PaddingMode): number {\n  if (dimSize == null) {\n    return null;\n  }\n\n  if (padding === 'valid') {\n    dimSize = dimSize * strideSize + max([kernelSize - strideSize, 0]);\n  } else if (padding === 'same') {\n    dimSize = dimSize * strideSize;\n  } else {\n    throw new ValueError(`Unsupport padding mode: ${padding}.`);\n  }\n  return dimSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {fused, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, PaddingMode, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, deconvLength, normalizeArray} from '../utils/conv_utils';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Transpose and cast the input before the conv2d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv2DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  // TODO(cais): Cast type to float32 if not.\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 1]);  // NCHW -> NHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * Transpose and cast the input before the conv3d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv3DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 4, 1]);  // NCDHW -> NDHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * 1D-convolution with bias added.\n *\n * Porting Note: This function does not exist in the Python Keras backend.\n *   It is exactly the same as `conv2d`, except the added `bias`.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.\n * @param bias Bias, rank-3, of shape `[outDepth]`.\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    // Check the ranks of x, kernel and bias.\n    if (x.shape.length !== 3) {\n      throw new ValueError(\n          `The input of a conv1dWithBias operation should be 3, but is ` +\n          `${x.shape.length} instead.`);\n    }\n    if (kernel.shape.length !== 3) {\n      throw new ValueError(\n          `The kernel for a conv1dWithBias operation should be 3, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    if (bias != null && bias.shape.length !== 1) {\n      throw new ValueError(\n          `The bias for a conv1dWithBias operation should be 1, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    // TODO(cais): Support CAUSAL padding mode.\n    if (dataFormat === 'channelsFirst') {\n      x = tfc.transpose(x, [0, 2, 1]);  // NCW -> NWC.\n    }\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    let y: Tensor = tfc.conv1d(\n        x as Tensor2D | Tensor3D, kernel as Tensor3D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias);\n    }\n    return y;\n  });\n}\n\n/**\n * 1D-convolution.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.s\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1d(\n    x: Tensor, kernel: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv1dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 2D pooling.\n */\nexport function conv2d(\n    x: Tensor, kernel: Tensor, strides = [1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv2dWithBiasActivation(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution with an added bias and optional activation.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv2d`, except the added `bias`.\n */\nexport function conv2dWithBiasActivation(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat, dilationRate?: [number, number],\n    activation: fused.Activation = null): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 3 && x.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects input to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    if (kernel.rank !== 3 && kernel.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects kernel to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.fused.conv2d({\n      x: y as Tensor3D | Tensor4D,\n      filter: kernel as Tensor4D,\n      strides: strides as [number, number],\n      pad: padding === 'same' ? 'same' : 'valid',\n      dilations: dilationRate,\n      dataFormat: 'NHWC',\n      bias,\n      activation\n    });\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\n/**\n * 3D Convolution.\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 3D convolution.\n */\nexport function conv3d(\n    x: Tensor, kernel: Tensor, strides = [1, 1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv3dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 3D Convolution with an added bias.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv3d`, except the added `bias`.\n */\nexport function conv3dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 4 && x.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects input to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    if (kernel.rank !== 4 && kernel.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects kernel to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    let y = preprocessConv3DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv3dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.conv3d(\n        y as Tensor4D | tfc.Tensor<tfc.Rank.R5>,\n        kernel as tfc.Tensor<tfc.Rank.R5>, strides as [number, number, number],\n        padding === 'same' ? 'same' : 'valid', 'NDHWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias as Tensor1D);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);\n    }\n    return y;\n  });\n}\n\n/**\n * Base LayerConfig for depthwise and non-depthwise convolutional layers.\n */\nexport declare interface BaseConvLayerArgs extends LayerArgs {\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number]|[number, number, number];\n\n  /**\n   * Activation function of the layer.\n   *\n   * If you don't specify the activation, none is applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector. Defaults to `true`.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the convolutional kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the convolutional kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\n/**\n * LayerConfig for non-depthwise convolutional layers.\n * Applies to non-depthwise convolution of all ranks (e.g, Conv1D, Conv2D,\n * Conv3D).\n */\nexport declare interface ConvLayerArgs extends BaseConvLayerArgs {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n}\n\n/**\n * Abstract convolution layer.\n */\nexport abstract class BaseConv extends Layer {\n  protected readonly rank: number;\n  protected readonly kernelSize: number[];\n  protected readonly strides: number[];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n  protected readonly activation: Activation;\n  protected readonly useBias: boolean;\n  protected readonly dilationRate: number[];\n\n  // Bias-related members are here because all convolution subclasses use the\n  // same configuration parmeters to control bias.  Kernel-related members\n  // are in subclass `Conv` because some subclasses use different parameters to\n  // control kernel properties, for instance, `DepthwiseConv2D` uses\n  // `depthwiseInitializer` instead of `kernelInitializer`.\n  protected readonly biasInitializer?: Initializer;\n  protected readonly biasConstraint?: Constraint;\n  protected readonly biasRegularizer?: Regularizer;\n\n  protected bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(rank: number, args: BaseConvLayerArgs) {\n    super(args as LayerArgs);\n    BaseConv.verifyArgs(args);\n    this.rank = rank;\n    generic_utils.assertPositiveInteger(this.rank, 'rank');\n    if (this.rank !== 1 && this.rank !== 2 && this.rank !== 3) {\n      throw new NotImplementedError(\n          `Convolution layer for rank other than 1, 2, or 3 (${\n              this.rank}) is ` +\n          `not implemented yet.`);\n    }\n    this.kernelSize = normalizeArray(args.kernelSize, rank, 'kernelSize');\n    this.strides = normalizeArray(\n        args.strides == null ? 1 : args.strides, rank, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.activation = getActivation(args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.dilationRate = normalizeArray(\n        args.dilationRate == null ? 1 : args.dilationRate, rank,\n        'dilationRate');\n    if (this.rank === 1 &&\n        (Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)) {\n      throw new ValueError(\n          `dilationRate must be a number or an array of a single number ` +\n          `for 1D convolution, but received ` +\n          `${JSON.stringify(this.dilationRate)}`);\n    } else if (this.rank === 2) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate = [this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 2) {\n        throw new ValueError(\n            `dilationRate must be a number or array of two numbers for 2D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    } else if (this.rank === 3) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate =\n            [this.dilationRate, this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 3) {\n        throw new ValueError(\n            `dilationRate must be a number or array of three numbers for 3D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    }\n  }\n\n  protected static verifyArgs(args: BaseConvLayerArgs) {\n    // Check config.kernelSize type and shape.\n    generic_utils.assert(\n        'kernelSize' in args, `required key 'kernelSize' not in config`);\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 3)) {\n      throw new ValueError(\n          `BaseConv expects config.kernelSize to be number or number[] with ` +\n          `length 1, 2, or 3, but received ${\n              JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      kernelSize: this.kernelSize,\n      strides: this.strides,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\n/**\n * Abstract nD convolution layer.  Ancestor of convolution layers which reduce\n * across channels, i.e., Conv1D and Conv2D, but not DepthwiseConv2D.\n */\nexport abstract class Conv extends BaseConv {\n  protected readonly filters: number;\n\n  protected kernel: LayerVariable = null;\n\n  // Bias-related properties are stored in the superclass `BaseConv` because all\n  // convolution subclasses use the same configuration parameters to control\n  // bias. Kernel-related properties are defined here rather than in the\n  // superclass because some convolution subclasses use different names and\n  // configuration parameters for their internal kernel state.\n  protected readonly kernelInitializer?: Initializer;\n  protected readonly kernelConstraint?: Constraint;\n  protected readonly kernelRegularizer?: Regularizer;\n\n  constructor(rank: number, args: ConvLayerArgs) {\n    super(rank, args as BaseConvLayerArgs);\n    Conv.verifyArgs(args);\n    this.filters = args.filters;\n    generic_utils.assertPositiveInteger(this.filters, 'filters');\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n    const inputDim = inputShape[channelAxis];\n\n    const kernelShape = this.kernelSize.concat([inputDim, this.filters]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.inputSpec = [{ndim: this.rank + 2, axes: {[channelAxis]: inputDim}}];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs: Tensor;\n      const biasValue = this.bias == null ? null : this.bias.read();\n      const fusedActivationName = generic_utils.mapActivationToFusedKernel(\n                                            this.activation.getClassName());\n\n      if (fusedActivationName != null && this.rank === 2) {\n        outputs = conv2dWithBiasActivation(\n            inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n            this.dataFormat, this.dilationRate as [number, number],\n            fusedActivationName);\n      } else {\n        if (this.rank === 1) {\n          outputs = conv1dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides[0],\n              this.padding, this.dataFormat, this.dilationRate[0]);\n        } else if (this.rank === 2) {\n          // TODO(cais): Move up to constructor.\n          outputs = conv2dWithBiasActivation(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number]);\n        } else if (this.rank === 3) {\n          outputs = conv3dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number, number]);\n        } else {\n          throw new NotImplementedError(\n              'convolutions greater than 3D are not implemented yet.');\n        }\n\n        if (this.activation != null) {\n          outputs = this.activation.apply(outputs);\n        }\n      }\n\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const newSpace: number[] = [];\n    const space = (this.dataFormat === 'channelsLast') ?\n        inputShape.slice(1, inputShape.length - 1) :\n        inputShape.slice(2);\n    for (let i = 0; i < space.length; ++i) {\n      const newDim = convOutputLength(\n          space[i], this.kernelSize[i], this.padding, this.strides[i],\n          typeof this.dilationRate === 'number' ? this.dilationRate :\n                                                  this.dilationRate[i]);\n      newSpace.push(newDim);\n    }\n\n    let outputShape = [inputShape[0]];\n    if (this.dataFormat === 'channelsLast') {\n      outputShape = outputShape.concat(newSpace);\n      outputShape.push(this.filters);\n    } else {\n      outputShape.push(this.filters);\n      outputShape = outputShape.concat(newSpace);\n    }\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      filters: this.filters,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // Check config.filters type, shape, and value.\n    if (!('filters' in args) || typeof args.filters !== 'number' ||\n        args.filters < 1) {\n      throw new ValueError(\n          `Convolution layer expected config.filters to be a 'number' > 0 ` +\n          `but got ${JSON.stringify(args.filters)}`);\n    }\n  }\n}\n\nexport class Conv2D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv2D';\n  constructor(args: ConvLayerArgs) {\n    super(2, args);\n    Conv2D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if ((typeof args.kernelSize !== 'number') &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 2)) {\n      throw new ValueError(\n          `Conv2D expects config.kernelSize to be number or number[] with ` +\n          `length 1 or 2, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv2D);\n\nexport class Conv3D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv3D';\n  constructor(args: ConvLayerArgs) {\n    super(3, args);\n    Conv3D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number') {\n      if (!(Array.isArray(args.kernelSize) &&\n            (args.kernelSize.length === 1 || args.kernelSize.length === 3))) {\n        throw new ValueError(\n            `Conv3D expects config.kernelSize to be number or` +\n            ` [number, number, number], but received ${\n                JSON.stringify(args.kernelSize)}.`);\n      }\n    }\n  }\n}\nserialization.registerClass(Conv3D);\n\nexport class Conv2DTranspose extends Conv2D {\n  /** @nocollapse */\n  static className = 'Conv2DTranspose';\n  inputSpec: InputSpec[];\n\n  constructor(args: ConvLayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n\n    if (this.padding !== 'same' && this.padding !== 'valid') {\n      throw new ValueError(\n          `Conv2DTranspose currently supports only padding modes 'same' ` +\n          `and 'valid', but received padding mode ${this.padding}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    if (inputShape.length !== 4) {\n      throw new ValueError(\n          'Input should have rank 4; Received input shape: ' +\n          JSON.stringify(inputShape));\n    }\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          'The channel dimension of the inputs should be defined. ' +\n          'Found `None`.');\n    }\n    const inputDim = inputShape[channelAxis];\n    const kernelShape = this.kernelSize.concat([this.filters, inputDim]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, 'float32', this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    // Set input spec.\n    this.inputSpec =\n        [new InputSpec({ndim: 4, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs);\n      if (input.shape.length !== 4) {\n        throw new ValueError(\n            `Conv2DTranspose.call() expects input tensor to be rank-4, but ` +\n            `received a tensor of rank-${input.shape.length}`);\n      }\n\n      const inputShape = input.shape;\n      const batchSize = inputShape[0];\n\n      let hAxis: number;\n      let wAxis: number;\n      if (this.dataFormat === 'channelsFirst') {\n        hAxis = 2;\n        wAxis = 3;\n      } else {\n        hAxis = 1;\n        wAxis = 2;\n      }\n\n      const height = inputShape[hAxis];\n      const width = inputShape[wAxis];\n      const kernelH = this.kernelSize[0];\n      const kernelW = this.kernelSize[1];\n      const strideH = this.strides[0];\n      const strideW = this.strides[1];\n\n      // Infer the dynamic output shape.\n      const outHeight = deconvLength(height, strideH, kernelH, this.padding);\n      const outWidth = deconvLength(width, strideW, kernelW, this.padding);\n\n      // Porting Note: We don't branch based on `this.dataFormat` here,\n      // because\n      //   the tjfs-core function `conv2dTranspose` called below always\n      //   assumes channelsLast.\n      const outputShape: [number, number, number, number] =\n          [batchSize, outHeight, outWidth, this.filters];\n\n      if (this.dataFormat !== 'channelsLast') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n      }\n      let outputs = tfc.conv2dTranspose(\n          input as Tensor4D, this.kernel.read() as Tensor4D, outputShape,\n          this.strides as [number, number], this.padding as 'same' | 'valid');\n      if (this.dataFormat !== 'channelsLast') {\n        outputs = tfc.transpose(outputs, [0, 3, 1, 2]);\n      }\n\n      if (this.bias != null) {\n        outputs =\n            K.biasAdd(outputs, this.bias.read(), this.dataFormat) as Tensor4D;\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs) as Tensor4D;\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n\n    let channelAxis: number;\n    let heightAxis: number;\n    let widthAxis: number;\n    if (this.dataFormat === 'channelsFirst') {\n      channelAxis = 1;\n      heightAxis = 2;\n      widthAxis = 3;\n    } else {\n      channelAxis = 3;\n      heightAxis = 1;\n      widthAxis = 2;\n    }\n\n    const kernelH = this.kernelSize[0];\n    const kernelW = this.kernelSize[1];\n    const strideH = this.strides[0];\n    const strideW = this.strides[1];\n\n    outputShape[channelAxis] = this.filters;\n    outputShape[heightAxis] =\n        deconvLength(outputShape[heightAxis], strideH, kernelH, this.padding);\n    outputShape[widthAxis] =\n        deconvLength(outputShape[widthAxis], strideW, kernelW, this.padding);\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['dilationRate'];\n    return config;\n  }\n}\nserialization.registerClass(Conv2DTranspose);\n\nexport declare interface SeparableConvLayerArgs extends ConvLayerArgs {\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal\n   * to `filtersIn * depthMultiplier`. Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the pointwise kernel matrix.\n   */\n  pointwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the pointwise kernel matrix.\n   */\n  pointwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the pointwise kernel matrix.\n   */\n  pointwiseConstraint?: ConstraintIdentifier|Constraint;\n}\n\nexport class SeparableConv extends Conv {\n  /** @nocollapse */\n  static className = 'SeparableConv';\n\n  readonly depthMultiplier: number;\n\n  protected readonly depthwiseInitializer?: Initializer;\n  protected readonly depthwiseRegularizer?: Regularizer;\n  protected readonly depthwiseConstraint?: Constraint;\n  protected readonly pointwiseInitializer?: Initializer;\n  protected readonly pointwiseRegularizer?: Regularizer;\n  protected readonly pointwiseConstraint?: Constraint;\n\n  readonly DEFAULT_DEPTHWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n  readonly DEFAULT_POINTWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n\n  protected depthwiseKernel: LayerVariable = null;\n  protected pointwiseKernel: LayerVariable = null;\n\n  constructor(rank: number, config?: SeparableConvLayerArgs) {\n    super(rank, config);\n\n    if (config.filters == null) {\n      throw new ValueError(\n          'The `filters` configuration field is required by SeparableConv, ' +\n          'but is unspecified.');\n    }\n    if (config.kernelInitializer != null || config.kernelRegularizer != null ||\n        config.kernelConstraint != null) {\n      throw new ValueError(\n          'Fields kernelInitializer, kernelRegularizer and kernelConstraint ' +\n          'are invalid for SeparableConv2D. Use depthwiseInitializer, ' +\n          'depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, ' +\n          'pointwiseRegularizer and pointwiseConstraint instead.');\n    }\n    if (config.padding != null && config.padding !== 'same' &&\n        config.padding !== 'valid') {\n      throw new ValueError(\n          `SeparableConv${this.rank}D supports only padding modes: ` +\n          `'same' and 'valid', but received ${JSON.stringify(config.padding)}`);\n    }\n\n    this.depthMultiplier =\n        config.depthMultiplier == null ? 1 : config.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER);\n    this.depthwiseRegularizer = getRegularizer(config.depthwiseRegularizer);\n    this.depthwiseConstraint = getConstraint(config.depthwiseConstraint);\n    this.pointwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER);\n    this.pointwiseRegularizer = getRegularizer(config.pointwiseRegularizer);\n    this.pointwiseConstraint = getConstraint(config.pointwiseConstraint);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < this.rank + 2) {\n      throw new ValueError(\n          `Inputs to SeparableConv${this.rank}D should have rank ` +\n          `${this.rank + 2}, but received input shape: ` +\n          `${JSON.stringify(inputShape)}`);\n    }\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          `The channel dimension of the inputs should be defined, ` +\n          `but found ${JSON.stringify(inputShape[channelAxis])}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape =\n        this.kernelSize.concat([inputDim, this.depthMultiplier]);\n    const pointwiseKernelShape = [];\n    for (let i = 0; i < this.rank; ++i) {\n      pointwiseKernelShape.push(1);\n    }\n    pointwiseKernelShape.push(inputDim * this.depthMultiplier, this.filters);\n\n    const trainable = true;\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, 'float32',\n        this.depthwiseInitializer, this.depthwiseRegularizer, trainable,\n        this.depthwiseConstraint);\n    this.pointwiseKernel = this.addWeight(\n        'pointwise_kernel', pointwiseKernelShape, 'float32',\n        this.pointwiseInitializer, this.pointwiseRegularizer, trainable,\n        this.pointwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, trainable, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n\n    this.inputSpec =\n        [new InputSpec({ndim: this.rank + 2, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      let output: Tensor;\n      if (this.rank === 1) {\n        throw new NotImplementedError(\n            '1D separable convolution is not implemented yet.');\n      } else if (this.rank === 2) {\n        if (this.dataFormat === 'channelsFirst') {\n          inputs = tfc.transpose(inputs, [0, 2, 3, 1]);  // NCHW -> NHWC.\n        }\n\n        output = tfc.separableConv2d(\n            inputs as Tensor4D, this.depthwiseKernel.read() as Tensor4D,\n            this.pointwiseKernel.read() as Tensor4D,\n            this.strides as [number, number], this.padding as 'same' | 'valid',\n            this.dilationRate as [number, number], 'NHWC');\n      }\n\n      if (this.useBias) {\n        output = K.biasAdd(output, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      if (this.dataFormat === 'channelsFirst') {\n        output = tfc.transpose(output, [0, 3, 1, 2]);  // NHWC -> NCHW.\n      }\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['kernelInitializer'];\n    delete config['kernelRegularizer'];\n    delete config['kernelConstraint'];\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['pointwiseInitializer'] =\n        serializeInitializer(this.pointwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['pointwiseRegularizer'] =\n        serializeRegularizer(this.pointwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseConstraint);\n    config['pointwiseConstraint'] =\n        serializeConstraint(this.pointwiseConstraint);\n    return config;\n  }\n}\n\nexport class SeparableConv2D extends SeparableConv {\n  /** @nocollapse */\n  static className = 'SeparableConv2D';\n  constructor(args?: SeparableConvLayerArgs) {\n    super(2, args);\n  }\n}\nserialization.registerClass(SeparableConv2D);\n\nexport class Conv1D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv1D';\n  constructor(args: ConvLayerArgs) {\n    super(1, args);\n    Conv1D.verifyArgs(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['dataFormat'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 1)) {\n      throw new ValueError(\n          `Conv1D expects config.kernelSize to be number or number[] with ` +\n          `length 1, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv1D);\n\nexport declare interface Cropping2DLayerArgs extends LayerArgs {\n  /**\n   * Dimension of the cropping along the width and the height.\n   * - If integer: the same symmetric cropping\n   *  is applied to width and height.\n   * - If list of 2 integers:\n   *   interpreted as two different\n   *   symmetric cropping values for height and width:\n   *   `[symmetric_height_crop, symmetric_width_crop]`.\n   * - If a list of 2 list of 2 integers:\n   *   interpreted as\n   *   `[[top_crop, bottom_crop], [left_crop, right_crop]]`\n   */\n  cropping: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   * `channels_first` corresponds to inputs with shape\n   *   `(batch, channels, ...)`\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class Cropping2D extends Layer {\n  /** @nocollapse */\n  static className = 'Cropping2D';\n  protected readonly cropping: [[number, number], [number, number]];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Cropping2DLayerArgs) {\n    super(args);\n    if (typeof args.cropping === 'number') {\n      this.cropping =\n          [[args.cropping, args.cropping], [args.cropping, args.cropping]];\n    } else if (typeof args.cropping[0] === 'number') {\n      this.cropping = [\n        [args.cropping[0], args.cropping[0]],\n        [args.cropping[1] as number, args.cropping[1] as number]\n      ];\n    } else {\n      this.cropping = args.cropping as [[number, number], [number, number]];\n    }\n    this.dataFormat =\n        args.dataFormat === undefined ? 'channelsLast' : args.dataFormat;\n    this.inputSpec = [{ndim: 4}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      return [\n        inputShape[0], inputShape[1],\n        inputShape[2] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[3] - this.cropping[1][0] - this.cropping[1][1]\n      ];\n    } else {\n      return [\n        inputShape[0],\n        inputShape[1] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[2] - this.cropping[1][0] - this.cropping[1][1], inputShape[3]\n      ];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      if (this.dataFormat === 'channelsLast') {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[1] - this.cropping[0][0] - this.cropping[0][1], 2);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[2] - this.cropping[1][1] - this.cropping[1][0], 3);\n      } else {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[2] - this.cropping[0][0] - this.cropping[0][1], 3);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[3] - this.cropping[1][1] - this.cropping[1][0], 4);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {cropping: this.cropping, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Cropping2D);\n\nexport declare interface UpSampling2DLayerArgs extends LayerArgs {\n  /**\n   * The upsampling factors for rows and columns.\n   *\n   * Defaults to `[2, 2]`.\n   */\n  size?: number[];\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `\"channelsLast\"` corresponds to inputs with shape\n   *   `[batch, ..., channels]`\n   *\n   *  `\"channelsFirst\"` corresponds to inputs with shape `[batch, channels,\n   * ...]`.\n   *\n   * Defaults to `\"channelsLast\"`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class UpSampling2D extends Layer {\n  /** @nocollapse */\n  static className = 'UpSampling2D';\n  protected readonly DEFAULT_SIZE = [2, 2];\n  protected readonly size: number[];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: UpSampling2DLayerArgs) {\n    super(args);\n    this.inputSpec = [{ndim: 4}];\n    this.size = args.size == null ? this.DEFAULT_SIZE : args.size;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      const height =\n          inputShape[2] == null ? null : this.size[0] * inputShape[2];\n      const width = inputShape[3] == null ? null : this.size[1] * inputShape[3];\n      return [inputShape[0], inputShape[1], height, width];\n    } else {\n      const height =\n          inputShape[1] == null ? null : this.size[0] * inputShape[1];\n      const width = inputShape[2] == null ? null : this.size[1] * inputShape[2];\n      return [inputShape[0], height, width, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs) as Tensor4D;\n      const inputShape = input.shape;\n\n      if (this.dataFormat === 'channelsFirst') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n        const height = this.size[0] * inputShape[2];\n        const width = this.size[1] * inputShape[3];\n        const resized = input.resizeNearestNeighbor([height, width]);\n        return tfc.transpose(resized, [0, 3, 1, 2]);\n      } else {\n        const height = this.size[0] * inputShape[1];\n        const width = this.size[1] * inputShape[2];\n        return input.resizeNearestNeighbor([height, width]);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {size: this.size, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(UpSampling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Depthwise Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {BaseConv, BaseConvLayerArgs, ConvLayerArgs, preprocessConv2DInput} from './convolutional';\n\n/**\n * 2D convolution with separable filters.\n * @param x Input tensor.\n * @param depthwiseKernel Convolution kernel for depthwise convolution.\n * @param strides Strides (Array of two integers).\n * @param padding Padding model.\n * @param dataFormat Data format.\n * @param dilationRate Array of two integers, dilation rates for the separable\n *   convolution.\n * @returns Output tensor.\n * @throws ValueError If depthwiseKernel is not a 4D array.\n */\nexport function depthwiseConv2d(\n    x: Tensor, depthwiseKernel: Tensor, strides: [number, number] = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `Input for depthwiseConv2d is required to be 4-D, but is instead ` +\n          `${x.rank}-D`);\n    }\n    if (depthwiseKernel.rank !== 4) {\n      throw new ValueError(\n          `depthwiseKernel is required to be 4-D, but is instead ` +\n          `${depthwiseKernel.rank}-D`);\n    }\n    y = tfc.depthwiseConv2d(\n        y as Tensor4D, depthwiseKernel as Tensor4D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NHWC', dilationRate);\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\nexport declare interface DepthwiseConv2DLayerArgs extends BaseConvLayerArgs {\n  /**\n   * An integer or Array of 2 integers, specifying the width and height of the\n   * 2D convolution window. Can be a single integer to specify the same value\n   * for all spatial dimensions.\n   */\n  kernelSize: number|[number, number];\n\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal to\n   * `filtersIn * depthMultiplier`.\n   * Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   * Default: GlorotNormal.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regulzarizer function for the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class DepthwiseConv2D extends BaseConv {\n  /** @nocollapse */\n  static className = 'DepthwiseConv2D';\n  private readonly depthMultiplier: number;\n  private readonly depthwiseInitializer: Initializer;\n  private readonly depthwiseConstraint: Constraint;\n  private readonly depthwiseRegularizer: Regularizer;\n\n  private depthwiseKernel: LayerVariable = null;\n\n  constructor(args: DepthwiseConv2DLayerArgs) {\n    super(2, args as ConvLayerArgs);\n    this.depthMultiplier =\n        args.depthMultiplier == null ? 1 : args.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        args.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.depthwiseConstraint = getConstraint(args.depthwiseConstraint);\n    this.depthwiseRegularizer = getRegularizer(args.depthwiseRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 4) {\n      throw new ValueError(\n          `Inputs to DepthwiseConv2D should have rank 4. ` +\n          `Received input shape: ${JSON.stringify(inputShape)}.`);\n    }\n    const channelAxis = this.dataFormat === 'channelsFirst' ? 1 : 3;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          'The channel dimension of the inputs to DepthwiseConv2D should ' +\n          `be defined, but is not (${inputShape[channelAxis]}).`);\n    }\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape: Shape = [\n      this.kernelSize[0], this.kernelSize[1], inputDim, this.depthMultiplier\n    ];\n\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, null,\n        this.depthwiseInitializer, this.depthwiseRegularizer, true,\n        this.depthwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [inputDim * this.depthMultiplier], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs = depthwiseConv2d(\n          inputs, this.depthwiseKernel.read(), this.strides as [number, number],\n          this.padding, this.dataFormat, null);\n      // TODO(cais): Add support for dilation.\n      if (this.useBias) {\n        outputs = K.biasAdd(outputs, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs);\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    const cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    const outFilters = this.dataFormat === 'channelsFirst' ?\n        inputShape[1] * this.depthMultiplier :\n        inputShape[3] * this.depthMultiplier;\n    const outRows = convOutputLength(\n        rows, this.kernelSize[0], this.padding, this.strides[0]);\n    const outCols = convOutputLength(\n        cols, this.kernelSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], outFilters, outRows, outCols];\n    } else {\n      // In this case, assume 'channelsLast'.\n      return [inputShape[0], outRows, outCols, outFilters];\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    config['depthMultiplier'] = this.depthMultiplier;\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseRegularizer);\n    return config;\n  }\n}\nserialization.registerClass(DepthwiseConv2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Recurrent Neural Network Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, SymbolicTensor} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, Ones, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs, RnnStepFunction} from '../types';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor, isArrayOfShapes} from '../utils/types_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\nimport {deserialize} from './serialization';\n\n/**\n * Standardize `apply()` args to a single list of tensor inputs.\n *\n * When running a model loaded from file, the input tensors `initialState` and\n * `constants` are passed to `RNN.apply()` as part of `inputs` instead of the\n * dedicated kwargs fields. `inputs` consists of\n * `[inputs, initialState0, initialState1, ..., constant0, constant1]` in this\n * case.\n * This method makes sure that arguments are\n * separated and that `initialState` and `constants` are `Array`s of tensors\n * (or None).\n *\n * @param inputs Tensor or `Array` of  tensors.\n * @param initialState Tensor or `Array` of tensors or `null`/`undefined`.\n * @param constants Tensor or `Array` of tensors or `null`/`undefined`.\n * @returns An object consisting of\n *   inputs: A tensor.\n *   initialState: `Array` of tensors or `null`.\n *   constants: `Array` of tensors or `null`.\n * @throws ValueError, if `inputs` is an `Array` but either `initialState` or\n *   `constants` is provided.\n */\nexport function standardizeArgs(\n    inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    initialState: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    constants: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    numConstants?: number): {\n  inputs: Tensor|SymbolicTensor,\n  initialState: Tensor[]|SymbolicTensor[],\n  constants: Tensor[]|SymbolicTensor[]\n} {\n  if (Array.isArray(inputs)) {\n    if (initialState != null || constants != null) {\n      throw new ValueError(\n          'When inputs is an array, neither initialState or constants ' +\n          'should be provided');\n    }\n    if (numConstants != null) {\n      constants = inputs.slice(inputs.length - numConstants, inputs.length);\n      inputs = inputs.slice(0, inputs.length - numConstants);\n    }\n    if (inputs.length > 1) {\n      initialState = inputs.slice(1, inputs.length);\n    }\n    inputs = inputs[0];\n  }\n\n  function toListOrNull(x: Tensor|Tensor[]|SymbolicTensor|\n                        SymbolicTensor[]): Tensor[]|SymbolicTensor[] {\n    if (x == null || Array.isArray(x)) {\n      return x as Tensor[] | SymbolicTensor[];\n    } else {\n      return [x] as Tensor[] | SymbolicTensor[];\n    }\n  }\n\n  initialState = toListOrNull(initialState);\n  constants = toListOrNull(constants);\n\n  return {inputs, initialState, constants};\n}\n\n/**\n * Iterates over the time dimension of a tensor.\n *\n * @param stepFunction RNN step function.\n *   Parameters:\n *     inputs: tensor with shape `[samples, ...]` (no time dimension),\n *       representing input for the batch of samples at a certain time step.\n *     states: an Array of tensors.\n *   Returns:\n *     outputs: tensor with shape `[samples, outputDim]` (no time dimension).\n *     newStates: list of tensors, same length and shapes as `states`. The first\n *       state in the list must be the output tensor at the previous timestep.\n * @param inputs Tensor of temporal data of shape `[samples, time, ...]` (at\n *   least 3D).\n * @param initialStates Tensor with shape `[samples, outputDim]` (no time\n *   dimension), containing the initial values of the states used in the step\n *   function.\n * @param goBackwards If `true`, do the iteration over the time dimension in\n *   reverse order and return the reversed sequence.\n * @param mask Binary tensor with shape `[sample, time, 1]`, with a zero for\n *   every element that is masked.\n * @param constants An Array of constant values passed at each step.\n * @param unroll Whether to unroll the RNN or to use a symbolic loop. *Not*\n *   applicable to this imperative deeplearn.js backend. Its value is ignored.\n * @param needPerStepOutputs Whether the per-step outputs are to be\n *   concatenated into a single tensor and returned (as the second return\n *   value). Default: `false`. This arg is included so that the relatively\n *   expensive concatenation of the stepwise outputs can be omitted unless\n *   the stepwise outputs need to be kept (e.g., for an LSTM layer of which\n *   `returnSequence` is `true`.)\n * @returns An Array: `[lastOutput, outputs, newStates]`.\n *   lastOutput: the lastest output of the RNN, of shape `[samples, ...]`.\n *   outputs: tensor with shape `[samples, time, ...]` where each entry\n *     `output[s, t]` is the output of the step function at time `t` for sample\n *     `s`. This return value is provided if and only if the\n *     `needPerStepOutputs` is set as `true`. If it is set as `false`, this\n *     return value will be `undefined`.\n *   newStates: Array of tensors, latest states returned by the step function,\n *      of shape `(samples, ...)`.\n * @throws ValueError If input dimension is less than 3.\n *\n * TODO(nielsene): This needs to be tidy-ed.\n */\nexport function rnn(\n    stepFunction: RnnStepFunction, inputs: Tensor, initialStates: Tensor[],\n    goBackwards = false, mask?: Tensor, constants?: Tensor[], unroll = false,\n    needPerStepOutputs = false): [Tensor, Tensor, Tensor[]] {\n  return tfc.tidy(() => {\n    const ndim = inputs.shape.length;\n    if (ndim < 3) {\n      throw new ValueError(`Input should be at least 3D, but is ${ndim}D.`);\n    }\n\n    // Transpose to time-major, i.e., from [batch, time, ...] to [time, batch,\n    // ...].\n    const axes = [1, 0].concat(math_utils.range(2, ndim));\n    inputs = tfc.transpose(inputs, axes);\n\n    if (constants != null) {\n      throw new NotImplementedError(\n          'The rnn() functoin of the deeplearn.js backend does not support ' +\n          'constants yet.');\n    }\n\n    // Porting Note: the unroll option is ignored by the imperative backend.\n    if (unroll) {\n      console.warn(\n          'Backend rnn(): the unroll = true option is not applicable to the ' +\n          'imperative deeplearn.js backend.');\n    }\n\n    if (mask != null) {\n      mask = mask.asType('bool').asType('float32');\n      if (mask.rank === ndim - 1) {\n        mask = tfc.expandDims(mask, -1);\n      }\n      mask = tfc.transpose(mask, axes);\n    }\n\n    if (goBackwards) {\n      inputs = tfc.reverse(inputs, 0);\n      if (mask != null) {\n        mask = tfc.reverse(mask, 0);\n      }\n    }\n\n    // Porting Note: PyKeras with TensorFlow backend uses a symbolic loop\n    //   (tf.while_loop). But for the imperative deeplearn.js backend, we just\n    //   use the usual TypeScript control flow to iterate over the time steps in\n    //   the inputs.\n    // Porting Note: PyKeras patches a \"_use_learning_phase\" attribute to\n    // outputs.\n    //   This is not idiomatic in TypeScript. The info regarding whether we are\n    //   in a learning (i.e., training) phase for RNN is passed in a different\n    //   way.\n\n    const perStepOutputs: Tensor[] = [];\n    let lastOutput: Tensor;\n    let states = initialStates;\n    const timeSteps = inputs.shape[0];\n    const perStepInputs = tfc.unstack(inputs);\n    let perStepMasks: Tensor[];\n    if (mask != null) {\n      perStepMasks = tfc.unstack(mask);\n    }\n\n    for (let t = 0; t < timeSteps; ++t) {\n      const currentInput = perStepInputs[t];\n      const stepOutputs = tfc.tidy(() => stepFunction(currentInput, states));\n\n      if (mask == null) {\n        lastOutput = stepOutputs[0];\n        states = stepOutputs[1];\n      } else {\n        const maskedOutputs = tfc.tidy(() => {\n          const stepMask = perStepMasks[t];\n          const negStepMask = tfc.onesLike(stepMask).sub(stepMask);\n          // TODO(cais): Would tfc.where() be better for performance?\n          const output =\n              stepOutputs[0].mul(stepMask).add(states[0].mul(negStepMask));\n          const newStates = states.map((state, i) => {\n            return stepOutputs[1][i].mul(stepMask).add(state.mul(negStepMask));\n          });\n          return {output, newStates};\n        });\n        lastOutput = maskedOutputs.output;\n        states = maskedOutputs.newStates;\n      }\n\n      if (needPerStepOutputs) {\n        perStepOutputs.push(lastOutput);\n      }\n    }\n    let outputs: Tensor;\n    if (needPerStepOutputs) {\n      const axis = 1;\n      outputs = tfc.stack(perStepOutputs, axis);\n    }\n    return [lastOutput, outputs, states] as [Tensor, Tensor, Tensor[]];\n  });\n}\n\nexport declare interface BaseRNNLayerArgs extends LayerArgs {\n  /**\n   * A RNN cell instance. A RNN cell is a class that has:\n   *   - a `call()` method, which takes `[Tensor, Tensor]` as the\n   *     first input argument. The first item is the input at time t, and\n   *     second item is the cell state at time t.\n   *     The `call()` method returns `[outputAtT, statesAtTPlus1]`.\n   *     The `call()` method of the cell can also take the argument `constants`,\n   *     see section \"Note on passing external constants\" below.\n   *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,\n   *       which are Layer subtypes, to accept two arguments. tfjs-layers does\n   *       not do such overriding. Instead we preseve the `call()` signature,\n   *       which due to its `Tensor|Tensor[]` argument and return value, is\n   *       flexible enough to handle the inputs and states.\n   *   - a `stateSize` attribute. This can be a single integer (single state)\n   *     in which case it is the size of the recurrent state (which should be\n   *     the same as the size of the cell output). This can also be an Array of\n   *     integers (one size per state). In this case, the first entry\n   *     (`stateSize[0]`) should be the same as the size of the cell output.\n   * It is also possible for `cell` to be a list of RNN cell instances, in which\n   * case the cells get stacked on after the other in the RNN, implementing an\n   * efficient stacked RNN.\n   */\n  cell?: RNNCell|RNNCell[];\n\n  /**\n   * Whether to return the last output in the output sequence, or the full\n   * sequence.\n   */\n  returnSequences?: boolean;\n\n  /**\n   * Whether to return the last state in addition to the output.\n   */\n  returnState?: boolean;\n\n  /**\n   * If `true`, process the input sequence backwards and return the reversed\n   * sequence (default: `false`).\n   */\n  goBackwards?: boolean;\n\n  /**\n   * If `true`, the last state for each sample at index i in a batch will be\n   * used as initial state of the sample of index i in the following batch\n   * (default: `false`).\n   *\n   * You can set RNN layers to be \"stateful\", which means that the states\n   * computed for the samples in one batch will be reused as initial states\n   * for the samples in the next batch. This assumes a one-to-one mapping\n   * between samples in different successive batches.\n   *\n   * To enable \"statefulness\":\n   *   - specify `stateful: true` in the layer constructor.\n   *   - specify a fixed batch size for your model, by passing\n   *     - if sequential model:\n   *       `batchInputShape: [...]` to the first layer in your model.\n   *     - else for functional model with 1 or more Input layers:\n   *       `batchShape: [...]` to all the first layers in your model.\n   *     This is the expected shape of your inputs\n   *     *including the batch size*.\n   *     It should be a tuple of integers, e.g., `[32, 10, 100]`.\n   *   - specify `shuffle: false` when calling `LayersModel.fit()`.\n   *\n   * To reset the state of your model, call `resetStates()` on either the\n   * specific layer or on the entire model.\n   */\n  stateful?: boolean;\n  // TODO(cais): Explore whether we can warn users when they fail to set\n  //   `shuffle: false` when training a model consisting of stateful RNNs\n  //   and any stateful Layers in general.\n\n  /**\n   * If `true`, the network will be unrolled, else a symbolic loop will be\n   * used. Unrolling can speed-up a RNN, although it tends to be more memory-\n   * intensive. Unrolling is only suitable for short sequences (default:\n   * `false`).\n   * Porting Note: tfjs-layers has an imperative backend. RNNs are executed with\n   *   normal TypeScript control flow. Hence this property is inapplicable and\n   *   ignored in tfjs-layers.\n   */\n  unroll?: boolean;\n\n  /**\n   * Dimensionality of the input (integer).\n   *   This option (or alternatively, the option `inputShape`) is required when\n   *   this layer is used as the first layer in a model.\n   */\n  inputDim?: number;\n\n  /**\n   * Length of the input sequences, to be specified when it is constant.\n   * This argument is required if you are going to connect `Flatten` then\n   * `Dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed). Note that if the recurrent layer is not the first layer in\n   * your model, you would need to specify the input length at the level of the\n   * first layer (e.g., via the `inputShape` option).\n   */\n  inputLength?: number;\n}\n\nexport class RNN extends Layer {\n  /** @nocollapse */\n  static className = 'RNN';\n  public readonly cell: RNNCell;\n  public readonly returnSequences: boolean;\n  public readonly returnState: boolean;\n  public readonly goBackwards: boolean;\n  public readonly unroll: boolean;\n\n  public stateSpec: InputSpec[];\n  protected states_: Tensor[];\n\n  // NOTE(cais): For stateful RNNs, the old states cannot be disposed right\n  // away when new states are set, because the old states may need to be used\n  // later for backpropagation through time (BPTT) and other purposes. So we\n  // keep them here for final disposal when the state is reset completely\n  // (i.e., through no-arg call to `resetStates()`).\n  protected keptStates: Tensor[][];\n\n  private numConstants: number;\n\n  constructor(args: RNNLayerArgs) {\n    super(args);\n    let cell: RNNCell;\n    if (args.cell == null) {\n      throw new ValueError(\n          'cell property is missing for the constructor of RNN.');\n    } else if (Array.isArray(args.cell)) {\n      cell = new StackedRNNCells({cells: args.cell});\n    } else {\n      cell = args.cell;\n    }\n    if (cell.stateSize == null) {\n      throw new ValueError(\n          'The RNN cell should have an attribute `stateSize` (tuple of ' +\n          'integers, one integer per RNN state).');\n    }\n    this.cell = cell;\n    this.returnSequences =\n        args.returnSequences == null ? false : args.returnSequences;\n    this.returnState = args.returnState == null ? false : args.returnState;\n    this.goBackwards = args.goBackwards == null ? false : args.goBackwards;\n    this._stateful = args.stateful == null ? false : args.stateful;\n    this.unroll = args.unroll == null ? false : args.unroll;\n\n    this.supportsMasking = true;\n    this.inputSpec = [new InputSpec({ndim: 3})];\n    this.stateSpec = null;\n    this.states_ = null;\n    // TODO(cais): Add constantsSpec and numConstants.\n    this.numConstants = null;\n    // TODO(cais): Look into the use of initial_state in the kwargs of the\n    //   constructor.\n\n    this.keptStates = [];\n  }\n\n  // Porting Note: This is the equivalent of `RNN.states` property getter in\n  //   PyKeras.\n  getStates(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      return math_utils.range(0, numStates).map(x => null);\n    } else {\n      return this.states_;\n    }\n  }\n\n  // Porting Note: This is the equivalent of the `RNN.states` property setter in\n  //   PyKeras.\n  setStates(states: Tensor[]): void {\n    this.states_ = states;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    // TODO(cais): Remove the casting once stacked RNN cells become supported.\n    let stateSize = this.cell.stateSize;\n    if (!Array.isArray(stateSize)) {\n      stateSize = [stateSize];\n    }\n    const outputDim = stateSize[0];\n    let outputShape: Shape|Shape[];\n    if (this.returnSequences) {\n      outputShape = [inputShape[0], inputShape[1], outputDim];\n    } else {\n      outputShape = [inputShape[0], outputDim];\n    }\n\n    if (this.returnState) {\n      const stateShape: Shape[] = [];\n      for (const dim of stateSize) {\n        stateShape.push([inputShape[0], dim]);\n      }\n      return [outputShape].concat(stateShape);\n    } else {\n      return outputShape;\n    }\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tfc.tidy(() => {\n      if (Array.isArray(mask)) {\n        mask = mask[0];\n      }\n      const outputMask = this.returnSequences ? mask : null;\n\n      if (this.returnState) {\n        const stateMask = this.states.map(s => null);\n        return [outputMask].concat(stateMask);\n      } else {\n        return outputMask;\n      }\n    });\n  }\n\n  /**\n   * Get the current state tensors of the RNN.\n   *\n   * If the state hasn't been set, return an array of `null`s of the correct\n   * length.\n   */\n  get states(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      const output: Tensor[] = [];\n      for (let i = 0; i < numStates; ++i) {\n        output.push(null);\n      }\n      return output;\n    } else {\n      return this.states_;\n    }\n  }\n\n  set states(s: Tensor[]) {\n    this.states_ = s;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    // Note inputShape will be an Array of Shapes of initial states and\n    // constants if these are passed in apply().\n    const constantShape: Shape[] = null;\n    if (this.numConstants != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    }\n\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    const batchSize: number = this.stateful ? inputShape[0] : null;\n    const inputDim = inputShape.slice(2);\n    this.inputSpec[0] = new InputSpec({shape: [batchSize, null, ...inputDim]});\n\n    // Allow cell (if RNNCell Layer) to build before we set or validate\n    // stateSpec.\n    const stepInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (constantShape != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    } else {\n      this.cell.build(stepInputShape);\n    }\n\n    // Set or validate stateSpec.\n    let stateSize: number[];\n    if (Array.isArray(this.cell.stateSize)) {\n      stateSize = this.cell.stateSize;\n    } else {\n      stateSize = [this.cell.stateSize];\n    }\n\n    if (this.stateSpec != null) {\n      if (!util.arraysEqual(\n              this.stateSpec.map(spec => spec.shape[spec.shape.length - 1]),\n              stateSize)) {\n        throw new ValueError(\n            `An initialState was passed that is not compatible with ` +\n            `cell.stateSize. Received stateSpec=${this.stateSpec}; ` +\n            `However cell.stateSize is ${this.cell.stateSize}`);\n      }\n    } else {\n      this.stateSpec =\n          stateSize.map(dim => new InputSpec({shape: [null, dim]}));\n    }\n    if (this.stateful) {\n      this.resetStates();\n    }\n  }\n\n  /**\n   * Reset the state tensors of the RNN.\n   *\n   * If the `states` argument is `undefined` or `null`, will set the\n   * state tensor(s) of the RNN to all-zero tensors of the appropriate\n   * shape(s).\n   *\n   * If `states` is provided, will set the state tensors of the RNN to its\n   * value.\n   *\n   * @param states Optional externally-provided initial states.\n   * @param training Whether this call is done during training. For stateful\n   *   RNNs, this affects whether the old states are kept or discarded. In\n   *   particular, if `training` is `true`, the old states will be kept so\n   *   that subsequent backpropgataion through time (BPTT) may work properly.\n   *   Else, the old states will be discarded.\n   */\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n      const batchSize = this.inputSpec[0].shape[0];\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n      // Initialize state if null.\n      if (this.states_ == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_ = [tfc.zeros([batchSize, this.cell.stateSize])];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_[0] = tfc.zeros([batchSize, this.cell.stateSize]);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training === true) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n          const dim = Array.isArray(this.cell.stateSize) ?\n              this.cell.stateSize[index] :\n              this.cell.stateSize;\n          const expectedShape = [batchSize, dim];\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n          this.states_[index] = value;\n        }\n      }\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    // TODO(cais): Figure out whether initialState is in kwargs or inputs.\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    // If any of `initial_state` or `constants` are specified and are\n    // `tf.SymbolicTensor`s, then add them to the inputs and temporarily modify\n    // the input_spec to include them.\n\n    let additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    let additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      kwargs['initialState'] = initialState;\n      additionalInputs = additionalInputs.concat(initialState);\n      this.stateSpec = [];\n      for (const state of initialState) {\n        this.stateSpec.push(new InputSpec({shape: state.shape}));\n      }\n      // TODO(cais): Use the following instead.\n      // this.stateSpec = initialState.map(state => new InputSpec({shape:\n      // state.shape}));\n      additionalSpecs = additionalSpecs.concat(this.stateSpec);\n    }\n    if (constants != null) {\n      kwargs['constants'] = constants;\n      additionalInputs = additionalInputs.concat(constants);\n      // TODO(cais): Add this.constantsSpec.\n      this.numConstants = constants.length;\n    }\n\n    const isTensor = additionalInputs[0] instanceof SymbolicTensor;\n    if (isTensor) {\n      // Compute full input spec, including state and constants.\n      const fullInput =\n          [inputs].concat(additionalInputs) as Tensor[] | SymbolicTensor[];\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call with temporarily replaced inputSpec.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output = super.apply(fullInput, kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  // tslint:disable-next-line:no-any\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    // Input shape: `[samples, time (padded with zeros), input_dim]`.\n    // Note that the .build() method of subclasses **must** define\n    // this.inputSpec and this.stateSpec owith complete input shapes.\n    return tidy(() => {\n      const mask = kwargs == null ? null : kwargs['mask'] as Tensor;\n      const training = kwargs == null ? null : kwargs['training'];\n      let initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      inputs = getExactlyOneTensor(inputs);\n      if (initialState == null) {\n        if (this.stateful) {\n          initialState = this.states_;\n        } else {\n          initialState = this.getInitialState(inputs);\n        }\n      }\n\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      if (initialState.length !== numStates) {\n        throw new ValueError(\n            `RNN Layer has ${numStates} state(s) but was passed ` +\n            `${initialState.length} initial state(s).`);\n      }\n      if (this.unroll) {\n        console.warn(\n            'Ignoring unroll = true for RNN layer, due to imperative backend.');\n      }\n\n      const cellCallKwargs: Kwargs = {training};\n\n      // TODO(cais): Add support for constants.\n      const step = (inputs: Tensor, states: Tensor[]) => {\n        // `inputs` and `states` are concatenated to form a single `Array` of\n        // `tf.Tensor`s as the input to `cell.call()`.\n        const outputs =\n            this.cell.call([inputs].concat(states), cellCallKwargs) as Tensor[];\n        // Marshall the return value into output and new states.\n        return [outputs[0], outputs.slice(1)] as [Tensor, Tensor[]];\n      };\n\n      // TODO(cais): Add support for constants.\n\n      const rnnOutputs =\n          rnn(step, inputs, initialState, this.goBackwards, mask, null,\n              this.unroll, this.returnSequences);\n      const lastOutput = rnnOutputs[0];\n      const outputs = rnnOutputs[1];\n      const states = rnnOutputs[2];\n\n      if (this.stateful) {\n        this.resetStates(states, training);\n      }\n\n      const output = this.returnSequences ? outputs : lastOutput;\n\n      // TODO(cais): Porperty set learning phase flag.\n\n      if (this.returnState) {\n        return [output].concat(states);\n      } else {\n        return output;\n      }\n    });\n  }\n\n  getInitialState(inputs: Tensor): Tensor[] {\n    return tidy(() => {\n      // Build an all-zero tensor of shape [samples, outputDim].\n      // [Samples, timeSteps, inputDim].\n      let initialState = tfc.zeros(inputs.shape);\n      // [Samples].\n      initialState = tfc.sum(initialState, [1, 2]);\n      initialState = K.expandDims(initialState);  // [Samples, 1].\n\n      if (Array.isArray(this.cell.stateSize)) {\n        return this.cell.stateSize.map(\n            dim => dim > 1 ? K.tile(initialState, [1, dim]) : initialState);\n      } else {\n        return this.cell.stateSize > 1 ?\n            [K.tile(initialState, [1, this.cell.stateSize])] :\n            [initialState];\n      }\n    });\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    return this.cell.trainableWeights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    if (!this.trainable) {\n      return this.cell.weights;\n    }\n    return this.cell.nonTrainableWeights;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.cell != null) {\n      this.cell.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      returnSequences: this.returnSequences,\n      returnState: this.returnState,\n      goBackwards: this.goBackwards,\n      stateful: this.stateful,\n      unroll: this.unroll,\n    };\n\n    if (this.numConstants != null) {\n      config['numConstants'] = this.numConstants;\n    }\n\n    const cellConfig = this.cell.getConfig();\n\n    if (this.getClassName() === RNN.className) {\n      config['cell'] = {\n        'className': this.cell.getClassName(),\n        'config': cellConfig,\n      } as serialization.ConfigDictValue;\n    }\n\n    // this order is necessary, to prevent cell name from replacing layer name\n    return {...cellConfig, ...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cellConfig = config['cell'] as serialization.ConfigDict;\n    const cell = deserialize(cellConfig, customObjects) as RNNCell;\n    return new cls(Object.assign(config, {cell}));\n  }\n}\nserialization.registerClass(RNN);\n\n// Porting Note: This is a common parent class for RNN cells. There is no\n// equivalent of this in PyKeras. Having a common parent class forgoes the\n//  need for `has_attr(cell, ...)` checks or its TypeScript equivalent.\n/**\n * An RNNCell layer.\n *\n * @doc {heading: 'Layers', subheading: 'Classes'}\n */\nexport abstract class RNNCell extends Layer {\n  /**\n   * Size(s) of the states.\n   * For RNN cells with only a single state, this is a single integer.\n   */\n  public stateSize: number|number[];\n  public dropoutMask: Tensor|Tensor[];\n  public recurrentDropoutMask: Tensor|Tensor[];\n}\n\nexport declare interface SimpleRNNCellLayerArgs extends LayerArgs {\n  /**\n   * units: Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   * Default: hyperbolic tangent ('tanh').\n   * If you pass `null`,  'linear' activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the `kernel` weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the `recurrent_kernel` weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the `kernel` weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the `recurrentKernel` weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraintfunction applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\nexport class SimpleRNNCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'SimpleRNNCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args: SimpleRNNCellLayerArgs) {\n    super(args);\n    this.units = args.units;\n    assertPositiveInteger(this.units, `units`);\n    this.activation = getActivation(\n        args.activation == null ? this.DEFAULT_ACTIVATION : args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    // TODO(cais): Use regularizer.\n    this.kernel = this.addWeight(\n        'kernel', [inputShape[inputShape.length - 1], this.units], null,\n        this.kernelInitializer, this.kernelRegularizer, true,\n        this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  // Porting Note: PyKeras' equivalent of this method takes two tensor inputs:\n  //   `inputs` and `states`. Here, the two tensors are combined into an\n  //   `Tensor[]` Array as the first input argument.\n  //   Similarly, PyKeras' equivalent of this method returns two values:\n  //    `output` and `[output]`. Here the two are combined into one length-2\n  //    `Tensor[]`, consisting of `output` repeated.\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `SimpleRNNCell expects 2 input Tensors, got ${inputs.length}.`);\n      }\n      let prevOutput = inputs[1];\n      inputs = inputs[0];\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training\n                           }) as Tensor;\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(prevOutput),\n                                      rate: this.recurrentDropout,\n                                      training\n                                    }) as Tensor;\n      }\n      let h: Tensor;\n      const dpMask: Tensor = this.dropoutMask as Tensor;\n      const recDpMask: Tensor = this.recurrentDropoutMask as Tensor;\n      if (dpMask != null) {\n        h = K.dot(tfc.mul(inputs, dpMask), this.kernel.read());\n      } else {\n        h = K.dot(inputs, this.kernel.read());\n      }\n      if (this.bias != null) {\n        h = K.biasAdd(h, this.bias.read());\n      }\n      if (recDpMask != null) {\n        prevOutput = tfc.mul(prevOutput, recDpMask);\n      }\n      let output = tfc.add(h, K.dot(prevOutput, this.recurrentKernel.read()));\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      // TODO(cais): Properly set learning phase on output tensor?\n      return [output, output];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(SimpleRNNCell);\n\nexport declare interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {\n  /**\n   * Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   *\n   * Defaults to  hyperbolic tangent (`tanh`)\n   *\n   * If you pass `null`, no activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the recurrentKernel weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the kernel weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the recurrentKernel weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\n/**\n * RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the\n * `cell` property required. This interface is to be used with constructors\n * of concrete RNN layer subtypes.\n */\nexport declare interface RNNLayerArgs extends BaseRNNLayerArgs {\n  cell: RNNCell|RNNCell[];\n}\n\nexport class SimpleRNN extends RNN {\n  /** @nocollapse */\n  static className = 'SimpleRNN';\n  constructor(args: SimpleRNNLayerArgs) {\n    args.cell = new SimpleRNNCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\nserialization.registerClass(SimpleRNN);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n\n  /**\n   * GRU convention (whether to apply reset gate after or before matrix\n   * multiplication). false = \"before\", true = \"after\" (only false is\n   * supported).\n   */\n  resetAfter?: boolean;\n}\n\nexport class GRUCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'GRUCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION: ActivationIdentifier = 'hardSigmoid';\n\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: GRUCellLayerArgs) {\n    super(args);\n    if (args.resetAfter) {\n      throw new ValueError(\n          `GRUCell does not support reset_after parameter set to true.`);\n    }\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 3], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 3], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units * 3], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `GRUCell expects 2 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      let hTMinus1 = inputs[1];  // Previous memory state.\n      inputs = inputs[0];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2, regardless of the actual value of\n      // config.implementation.\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 3\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 3\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor];\n      const recDpMask = this.recurrentDropoutMask as [Tensor, Tensor, Tensor];\n      let z: Tensor;\n      let r: Tensor;\n      let hh: Tensor;\n\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let matrixX = K.dot(inputs, this.kernel.read());\n      if (this.useBias) {\n        matrixX = K.biasAdd(matrixX, this.bias.read());\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n\n      const recurrentKernelValue = this.recurrentKernel.read();\n      const [rk1, rk2] = tfc.split(\n          recurrentKernelValue, [2 * this.units, this.units],\n          recurrentKernelValue.rank - 1);\n      const matrixInner = K.dot(hTMinus1, rk1);\n\n      const [xZ, xR, xH] = tfc.split(matrixX, 3, matrixX.rank - 1);\n      const [recurrentZ, recurrentR] =\n          tfc.split(matrixInner, 2, matrixInner.rank - 1);\n      z = this.recurrentActivation.apply(tfc.add(xZ, recurrentZ));\n      r = this.recurrentActivation.apply(tfc.add(xR, recurrentR));\n\n      const recurrentH = K.dot(tfc.mul(r, hTMinus1), rk2);\n      hh = this.activation.apply(tfc.add(xH, recurrentH));\n\n      const h =\n          tfc.add(tfc.mul(z, hTMinus1), tfc.mul(tfc.add(1, tfc.neg(z)), hh));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n      resetAfter: false\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(GRUCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface GRULayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   * smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class GRU extends RNN {\n  /** @nocollapse */\n  static className = 'GRU';\n  constructor(args: GRULayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new GRUCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(GRU);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class LSTMCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'LSTMCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n  readonly unitForgetBias: boolean;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number[];\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION = 'hardSigmoid';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n\n  readonly DEFAULT_BIAS_INITIALIZER = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: LSTMCellLayerArgs) {\n    super(args);\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.unitForgetBias = args.unitForgetBias;\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = [this.units, this.units];\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 4], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 4], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    let biasInitializer: Initializer;\n    if (this.useBias) {\n      if (this.unitForgetBias) {\n        const capturedBiasInit = this.biasInitializer;\n        const capturedUnits = this.units;\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): Tensor {\n            // TODO(cais): More informative variable names?\n            const bI = capturedBiasInit.apply([capturedUnits]);\n            const bF = (new Ones()).apply([capturedUnits]);\n            const bCAndH = capturedBiasInit.apply([capturedUnits * 2]);\n            return K.concatAlongFirstAxis(\n                K.concatAlongFirstAxis(bI, bF), bCAndH);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n      this.bias = this.addWeight(\n          'bias', [this.units * 4], null, biasInitializer, this.biasRegularizer,\n          true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `LSTMCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n      let hTMinus1 = inputs[1];    // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n      inputs = inputs[0];\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 4\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 4\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor, Tensor];\n      const recDpMask =\n          this.recurrentDropoutMask as [Tensor, Tensor, Tensor, Tensor];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2 regardless of the actual value of\n      // config.implementation.\n      let i: Tensor;\n      let f: Tensor;\n      let c: Tensor;\n      let o: Tensor;\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let z = K.dot(inputs, this.kernel.read());\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n      z = tfc.add(z, K.dot(hTMinus1, this.recurrentKernel.read()));\n      if (this.useBias) {\n        z = K.biasAdd(z, this.bias.read());\n      }\n\n      const [z0, z1, z2, z3] = tfc.split(z, 4, z.rank - 1);\n\n      i = this.recurrentActivation.apply(z0);\n      f = this.recurrentActivation.apply(z1);\n      c = tfc.add(tfc.mul(f, cTMinus1), tfc.mul(i, this.activation.apply(z2)));\n      o = this.recurrentActivation.apply(z3);\n\n      const h = tfc.mul(o, this.activation.apply(c));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      unitForgetBias: this.unitForgetBias,\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(LSTMCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface LSTMLayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *   Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions, whereas mode 2 will\n   *   batch them into fewer, larger operations. These modes will\n   *   have different performance profiles on different hardware and\n   *   for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this config field.\n   */\n  implementation?: number;\n}\n\nexport class LSTM extends RNN {\n  /** @nocollapse */\n  static className = 'LSTM';\n  constructor(args: LSTMLayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new LSTMCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(LSTM);\n\nexport declare interface StackedRNNCellsArgs extends LayerArgs {\n  /**\n   * A `Array` of `RNNCell` instances.\n   */\n  cells: RNNCell[];\n}\n\nexport class StackedRNNCells extends RNNCell {\n  /** @nocollapse */\n  static className = 'StackedRNNCells';\n  protected cells: RNNCell[];\n\n  constructor(args: StackedRNNCellsArgs) {\n    super(args);\n    this.cells = args.cells;\n  }\n\n  get stateSize(): number[] {\n    // States are a flat list in reverse order of the cell stack.\n    // This allows perserving the requirement `stack.statesize[0] ===\n    // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,\n    // assuming one LSTM has states `[h, c]`.\n    const stateSize: number[] = [];\n    for (const cell of this.cells.slice().reverse()) {\n      if (Array.isArray(cell.stateSize)) {\n        stateSize.push(...cell.stateSize);\n      } else {\n        stateSize.push(cell.stateSize);\n      }\n    }\n    return stateSize;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      let states = inputs.slice(1);\n\n      // Recover per-cell states.\n      const nestedStates: Tensor[][] = [];\n      for (const cell of this.cells.slice().reverse()) {\n        if (Array.isArray(cell.stateSize)) {\n          nestedStates.push(states.splice(0, cell.stateSize.length));\n        } else {\n          nestedStates.push(states.splice(0, 1));\n        }\n      }\n      nestedStates.reverse();\n\n      // Call the cells in order and store the returned states.\n      const newNestedStates: Tensor[][] = [];\n      let callInputs: Tensor[];\n      for (let i = 0; i < this.cells.length; ++i) {\n        const cell = this.cells[i];\n        states = nestedStates[i];\n        // TODO(cais): Take care of constants.\n        if (i === 0) {\n          callInputs = [inputs[0]].concat(states);\n        } else {\n          callInputs = [callInputs[0]].concat(states);\n        }\n        callInputs = cell.call(callInputs, kwargs) as Tensor[];\n        newNestedStates.push(callInputs.slice(1));\n      }\n\n      // Format the new states as a flat list in reverse cell order.\n      states = [];\n      for (const cellStates of newNestedStates.slice().reverse()) {\n        states.push(...cellStates);\n      }\n      return [callInputs[0]].concat(states);\n    });\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    if (isArrayOfShapes(inputShape)) {\n      // TODO(cais): Take care of input constants.\n      // const constantShape = inputShape.slice(1);\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n    let outputDim: number;\n    this.cells.forEach((cell, i) => {\n      nameScope(`RNNCell_${i}`, () => {\n        // TODO(cais): Take care of input constants.\n\n        cell.build(inputShape);\n        if (Array.isArray(cell.stateSize)) {\n          outputDim = cell.stateSize[0];\n        } else {\n          outputDim = cell.stateSize;\n        }\n        inputShape = [inputShape[0], outputDim] as Shape;\n      });\n    });\n    this.built = true;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const getCellConfig = (cell: RNNCell) => {\n      return {\n        'className': cell.getClassName(),\n        'config': cell.getConfig(),\n      };\n    };\n\n    const cellConfigs = this.cells.map(getCellConfig);\n\n    const config = {'cells': cellConfigs};\n\n    return {...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cells: RNNCell[] = [];\n    for (const cellConfig of (config['cells'] as serialization.ConfigDict[])) {\n      cells.push(deserialize(cellConfig, customObjects) as RNNCell);\n    }\n    return new cls({cells});\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const cell of this.cells) {\n        trainableWeights.push(...cell.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  /**\n   * Retrieve the weights of a the model.\n   *\n   * @returns A flat `Array` of `tf.Tensor`s.\n   */\n  getWeights(): Tensor[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.weights);\n    }\n    return batchGetValue(weights);\n  }\n\n  /**\n   * Set the weights of the model.\n   *\n   * @param weights An `Array` of `tf.Tensor`s with shapes and types matching\n   *     the output of `getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    const tuples: Array<[LayerVariable, Tensor]> = [];\n    for (const cell of this.cells) {\n      const numParams = cell.weights.length;\n      const inputWeights = weights.splice(numParams);\n      for (let i = 0; i < cell.weights.length; ++i) {\n        tuples.push([cell.weights[i], inputWeights[i]]);\n      }\n    }\n    batchSetValue(tuples);\n  }\n\n  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.\n}\nserialization.registerClass(StackedRNNCells);\n\nexport function generateDropoutMask(args: {\n  ones: () => tfc.Tensor,\n  rate: number,\n  training?: boolean,\n  count?: number,\n}): tfc.Tensor|tfc.Tensor[] {\n  const {ones, rate, training = false, count = 1} = args;\n\n  const droppedInputs = () => K.dropout(ones(), rate);\n\n  const createMask = () => K.inTrainPhase(droppedInputs, ones, training);\n\n  // just in case count is provided with null or undefined\n  if (!count || count <= 1) {\n    return tfc.keep(createMask().clone());\n  }\n\n  const masks = Array(count).fill(undefined).map(createMask);\n\n  return masks.map(m => tfc.keep(m.clone()));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {Activation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint} from '../constraints';\nimport {InputSpec} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {Initializer} from '../initializers';\nimport {DataFormat, DataType, PaddingMode, Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, normalizeArray} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\nimport {BaseRNNLayerArgs, generateDropoutMask, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNNCellLayerArgs} from './recurrent';\n\ndeclare interface ConvRNN2DCellArgs extends\n    Omit<SimpleRNNCellLayerArgs, 'units'> {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number];\n}\n\nabstract class ConvRNN2DCell extends RNNCell {\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n}\n\ndeclare interface ConvRNN2DLayerArgs extends BaseRNNLayerArgs,\n                                             ConvRNN2DCellArgs {}\n\n/**\n * Base class for convolutional-recurrent layers.\n */\nclass ConvRNN2D extends RNN {\n  /** @nocollapse */\n  static className = 'ConvRNN2D';\n\n  readonly cell: ConvRNN2DCell;\n\n  constructor(args: ConvRNN2DLayerArgs) {\n    if (args.unroll) {\n      throw new NotImplementedError(\n          'Unrolling is not possible with convolutional RNNs.');\n    }\n\n    if (Array.isArray(args.cell)) {\n      throw new NotImplementedError(\n          'It is not possible at the moment to stack convolutional cells.');\n    }\n\n    super(args as RNNLayerArgs);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n\n        this.cell.dropoutMask = null;\n      }\n\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n\n        this.cell.recurrentDropoutMask = null;\n      }\n\n      if (kwargs && kwargs['constants']) {\n        throw new ValueError('ConvRNN2D cell does not support constants');\n      }\n\n      const mask = kwargs == null ? null : kwargs['mask'];\n\n      const training = kwargs == null ? null : kwargs['training'];\n\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  computeOutputShape(inputShape: Shape): Shape|Shape[] {\n    let outShape: Shape = this.computeSingleOutputShape(inputShape);\n\n    if (!this.returnSequences) {\n      outShape = [outShape[0], ...outShape.slice(2)];\n    }\n\n    if (this.returnState) {\n      outShape =\n          [outShape, ...Array(2).fill([inputShape[0], ...outShape.slice(-3)])];\n    }\n\n    return outShape;\n  }\n\n  getInitialState(inputs: tfc.Tensor): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      const {stateSize} = this.cell;\n\n      const inputShape = inputs.shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const initialState = tfc.zeros(stateShape);\n\n      if (Array.isArray(stateSize)) {\n        return Array(stateSize.length).fill(initialState);\n      }\n\n      return [initialState];\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tfc.tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n\n      const inputShape = this.inputSpec[0].shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const batchSize = inputShape[0];\n\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n\n      // Initialize state if null.\n      if (this.getStates() == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_ = [tfc.zeros(stateShape)];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_[0] = tfc.zeros(stateShape);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n\n          const expectedShape = stateShape;\n\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n\n          this.states_[index] = value;\n        }\n      }\n\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  protected computeSingleOutputShape(inputShape: Shape): Shape {\n    const {dataFormat, filters, kernelSize, padding, strides, dilationRate} =\n        this.cell;\n\n    const isChannelsFirst = dataFormat === 'channelsFirst';\n\n    const h = inputShape[isChannelsFirst ? 3 : 2];\n    const w = inputShape[isChannelsFirst ? 4 : 3];\n\n    const hOut = convOutputLength(\n        h, kernelSize[0], padding, strides[0], dilationRate[0]);\n    const wOut = convOutputLength(\n        w, kernelSize[1], padding, strides[1], dilationRate[1]);\n\n    const outShape: Shape = [\n      ...inputShape.slice(0, 2),\n      ...(isChannelsFirst ? [filters, hOut, wOut] : [hOut, wOut, filters])\n    ];\n\n    return outShape;\n  }\n}\n\nexport declare interface ConvLSTM2DCellArgs extends\n    Omit<LSTMCellLayerArgs, 'units'>, ConvRNN2DCellArgs {}\n\nexport class ConvLSTM2DCell extends LSTMCell implements ConvRNN2DCell {\n  /** @nocollapse */\n  static className = 'ConvLSTM2DCell';\n\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  constructor(args: ConvLSTM2DCellArgs) {\n    const {\n      filters,\n      kernelSize,\n      strides,\n      padding,\n      dataFormat,\n      dilationRate,\n    } = args;\n\n    super({...args, units: filters});\n\n    this.filters = filters;\n    assertPositiveInteger(this.filters, 'filters');\n\n    this.kernelSize = normalizeArray(kernelSize, 2, 'kernelSize');\n    this.kernelSize.forEach(size => assertPositiveInteger(size, 'kernelSize'));\n\n    this.strides = normalizeArray(strides || 1, 2, 'strides');\n    this.strides.forEach(stride => assertPositiveInteger(stride, 'strides'));\n\n    this.padding = padding || 'valid';\n    checkPaddingMode(this.padding);\n\n    this.dataFormat = dataFormat || 'channelsLast';\n    checkDataFormat(this.dataFormat);\n\n    this.dilationRate = normalizeArray(dilationRate || 1, 2, 'dilationRate');\n    this.dilationRate.forEach(\n        rate => assertPositiveInteger(rate, 'dilationRate'));\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n\n    const numOfKernels = 4;\n\n    const kernelShape =\n        this.kernelSize.concat([inputDim, this.filters * numOfKernels]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n\n    const recurrentKernelShape =\n        this.kernelSize.concat([this.filters, this.filters * numOfKernels]);\n\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', recurrentKernelShape, null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n\n    if (this.useBias) {\n      let biasInitializer: Initializer;\n\n      if (this.unitForgetBias) {\n        const init = this.biasInitializer;\n\n        const filters = this.filters;\n\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): tfc.Tensor {\n            const biasI = init.apply([filters]);\n            const biasF = tfc.ones([filters]);\n            const biasCAndO = init.apply([filters * 2]);\n            return K.concatenate([biasI, biasF, biasCAndO]);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n\n      this.bias = this.addWeight(\n          'bias', [this.filters * numOfKernels], null, biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: tfc.Tensor[], kwargs: Kwargs): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] || false;\n\n      const x = inputs[0];         // Current input\n      const hTMinus1 = inputs[1];  // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n\n      const numOfKernels = 4;\n\n      type DropoutMasks = [tfc.Tensor, tfc.Tensor, tfc.Tensor, tfc.Tensor];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(x),\n                             rate: this.dropout,\n                             training,\n                             count: numOfKernels\n                           }) as tfc.Tensor[];\n      }\n\n      const dropoutMask = this.dropoutMask as DropoutMasks;\n\n      const applyDropout =\n          (x: tfc.Tensor, mask: tfc.Tensor[], index: number) => {\n            if (!mask || !mask[index]) {\n              return x;\n            }\n\n            return tfc.mul(mask[index], x);\n          };\n\n      let xI = applyDropout(x, dropoutMask, 0);\n      let xF = applyDropout(x, dropoutMask, 1);\n      let xC = applyDropout(x, dropoutMask, 2);\n      let xO = applyDropout(x, dropoutMask, 3);\n\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: numOfKernels\n                                    }) as tfc.Tensor[];\n      }\n\n      const recDropoutMask = this.recurrentDropoutMask as DropoutMasks;\n\n      let hI = applyDropout(hTMinus1, recDropoutMask, 0);\n      let hF = applyDropout(hTMinus1, recDropoutMask, 1);\n      let hC = applyDropout(hTMinus1, recDropoutMask, 2);\n      let hO = applyDropout(hTMinus1, recDropoutMask, 3);\n\n      const kernelChannelAxis = 3;\n\n      const [kernelI, kernelF, kernelC, kernelO]: tfc.Tensor[] =\n          tfc.split(this.kernel.read(), numOfKernels, kernelChannelAxis);\n\n      const [biasI, biasF, biasC, biasO]: tfc.Tensor[] = this.useBias ?\n          tfc.split(this.bias.read(), numOfKernels) :\n          [null, null, null, null];\n\n      xI = this.inputConv(xI, kernelI, biasI, this.padding);\n      xF = this.inputConv(xF, kernelF, biasF, this.padding);\n      xC = this.inputConv(xC, kernelC, biasC, this.padding);\n      xO = this.inputConv(xO, kernelO, biasO, this.padding);\n\n      const [recKernelI, recKernelF, recKernelC, recKernelO]: tfc.Tensor[] =\n          tfc.split(\n              this.recurrentKernel.read(), numOfKernels, kernelChannelAxis);\n\n      hI = this.recurrentConv(hI, recKernelI);\n      hF = this.recurrentConv(hF, recKernelF);\n      hC = this.recurrentConv(hC, recKernelC);\n      hO = this.recurrentConv(hO, recKernelO);\n\n      const i = this.recurrentActivation.apply(tfc.add(xI, hI));\n      const f = this.recurrentActivation.apply(tfc.add(xF, hF));\n      const c = tfc.add(\n          tfc.mul(f, cTMinus1),\n          tfc.mul(i, this.activation.apply(tfc.add(xC, hC))));\n      const h = tfc.mul(\n          this.recurrentActivation.apply(tfc.add(xO, hO)),\n          this.activation.apply(c));\n\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): tfc.serialization.ConfigDict {\n    const {'units': _, ...baseConfig} = super.getConfig();\n\n    const config: tfc.serialization.ConfigDict = {\n      filters: this.filters,\n      kernelSize: this.kernelSize,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      strides: this.strides,\n    };\n\n    return {...baseConfig, ...config};\n  }\n\n  inputConv(x: Tensor, w: Tensor, b?: Tensor, padding?: PaddingMode) {\n    const out = tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, this.strides as [number, number],\n        (padding || 'valid') as 'same' | 'valid',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC',\n        this.dilationRate as [number, number]);\n\n    if (b) {\n      return K.biasAdd(out, b, this.dataFormat) as tfc.Tensor3D;\n    }\n\n    return out;\n  }\n\n  recurrentConv(x: Tensor, w: Tensor) {\n    const strides = 1;\n\n    return tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, strides, 'same',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC');\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2DCell);\n\nexport declare interface ConvLSTM2DArgs extends\n    Omit<LSTMLayerArgs, 'units'|'cell'>, ConvRNN2DLayerArgs {}\n\nexport class ConvLSTM2D extends ConvRNN2D {\n  /** @nocollapse */\n  static className = 'ConvLSTM2D';\n\n  constructor(args: ConvLSTM2DArgs) {\n    const cell = new ConvLSTM2DCell(args);\n\n    super({...args, cell} as ConvRNN2DLayerArgs);\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends tfc.serialization.Serializable>(\n      cls: tfc.serialization.SerializableConstructor<T>,\n      config: tfc.serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Basic Layers.\n */\n\nimport {any, notEqual, serialization, Tensor, tidy, transpose, util} from '@tensorflow/tfjs-core';\n\nimport {Activation as ActivationFn, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {DisposeResult, InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {LayerConfig} from '../keras_format/topology_config';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {assertPositiveInteger, mapActivationToFusedKernel} from '../utils/generic_utils';\nimport {arrayProd, range} from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface DropoutLayerArgs extends LayerArgs {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /**\n   * Integer array representing the shape of the binary dropout mask that will\n   * be multiplied with the input.\n   *\n   * For instance, if your inputs have shape `(batchSize, timesteps, features)`\n   * and you want the dropout mask to be the same for all timesteps, you can use\n   * `noise_shape=(batch_size, 1, features)`.\n   */\n  noiseShape?: number[];\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class Dropout extends Layer {\n  /** @nocollapse */\n  static className = 'Dropout';\n  private readonly rate: number;\n  private readonly noiseShape: number[];\n  private readonly seed: number;\n\n  constructor(args: DropoutLayerArgs) {\n    super(args);\n    this.rate = Math.max(Math.min(args.rate, 1), 0);\n    // So that the scalar doesn't get tidied up between executions.\n    this.noiseShape = args.noiseShape;\n    this.seed = args.seed;\n    this.supportsMasking = true;\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    if (this.noiseShape == null) {\n      return this.noiseShape;\n    }\n    const inputShape = input.shape;\n    const noiseShape: Shape = [];\n    for (let i = 0; i < this.noiseShape.length; ++i) {\n      noiseShape.push(\n          this.noiseShape[i] == null ? inputShape[i] : this.noiseShape[i]);\n    }\n    return noiseShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      if (0 < this.rate && this.rate < 1) {\n        const training =\n            kwargs['training'] == null ? false : kwargs['training'];\n        const noiseShape = this.getNoiseShape(input);\n        const output = K.inTrainPhase(\n            () => K.dropout(input, this.rate, noiseShape, this.seed),\n            () => input, training);\n        return output;\n      }\n      return inputs;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      rate: this.rate,\n      noiseShape: this.noiseShape,\n      seed: this.seed,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  dispose(): DisposeResult {\n    return super.dispose();\n  }\n}\nserialization.registerClass(Dropout);\n\nexport declare interface DenseLayerArgs extends LayerArgs {\n  /** Positive integer, dimensionality of the output space. */\n  units: number;\n  /**\n   * Activation function to use.\n   *\n   * If unspecified, no activation is applied.\n   */\n  activation?: ActivationIdentifier;\n  /** Whether to apply a bias. */\n  useBias?: boolean;\n  /**\n   * Initializer for the dense kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * If specified, defines inputShape as `[inputDim]`.\n   */\n  inputDim?: number;\n\n  /**\n   * Constraint for the kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the dense kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport interface SpatialDropout1DLayerConfig extends LayerConfig {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class SpatialDropout1D extends Dropout {\n  /** @nocollapse */\n  static className = 'SpatialDropout1D';\n\n  constructor(args: SpatialDropout1DLayerConfig) {\n    super(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    const inputShape = input.shape;\n    return [inputShape[0], 1, inputShape[2]];\n  }\n}\nserialization.registerClass(SpatialDropout1D);\n\nexport class Dense extends Layer {\n  /** @nocollapse */\n  static className = 'Dense';\n  private units: number;\n  // Default activation: Linear (none).\n  private activation: ActivationFn = null;\n  private useBias = true;\n  private kernelInitializer: Initializer;\n  private biasInitializer: Initializer;\n  private kernel: LayerVariable = null;\n  private bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n  private readonly kernelConstraint?: Constraint;\n  private readonly biasConstraint?: Constraint;\n  private readonly kernelRegularizer?: Regularizer;\n  private readonly biasRegularizer?: Regularizer;\n\n  constructor(args: DenseLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null &&\n        args.inputDim != null) {\n      // This logic is copied from Layer's constructor, since we can't\n      // do exactly what the Python constructor does for Dense().\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      this.batchInputShape = [batchSize, args.inputDim];\n    }\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(args.activation);\n    if (args.useBias != null) {\n      this.useBias = args.useBias;\n    }\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.supportsMasking = true;\n\n    this.inputSpec = [{minNDim: 2}];\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputLastDim = inputShape[inputShape.length - 1];\n    if (this.kernel == null) {\n      this.kernel = this.addWeight(\n          'kernel', [inputLastDim, this.units], null, this.kernelInitializer,\n          this.kernelRegularizer, true, this.kernelConstraint);\n      if (this.useBias) {\n        this.bias = this.addWeight(\n            'bias', [this.units], null, this.biasInitializer,\n            this.biasRegularizer, true, this.biasConstraint);\n      }\n    }\n\n    this.inputSpec = [{minNDim: 2, axes: {[-1]: inputLastDim}}];\n    this.built = true;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    outputShape[outputShape.length - 1] = this.units;\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Dense layer accepts only a single input.\n      const input = getExactlyOneTensor(inputs);\n      const fusedActivationName =\n          mapActivationToFusedKernel(this.activation.getClassName());\n      let output: Tensor;\n\n      if (fusedActivationName != null) {\n        output = K.dot(\n            input, this.kernel.read(), fusedActivationName,\n            this.bias ? this.bias.read() : null);\n      } else {\n        output = K.dot(input, this.kernel.read());\n        if (this.bias != null) {\n          output = K.biasAdd(output, this.bias.read());\n        }\n        if (this.activation != null) {\n          output = this.activation.apply(output);\n        }\n      }\n\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dense);\n\nexport declare interface FlattenLayerArgs extends LayerArgs {\n  /** Image data format: channeLast (default) or channelFirst. */\n  dataFormat?: DataFormat;\n}\n\nexport class Flatten extends Layer {\n  private dataFormat: DataFormat;\n\n  /** @nocollapse */\n  static className = 'Flatten';\n  constructor(args?: FlattenLayerArgs) {\n    args = args || {};\n    super(args);\n    this.inputSpec = [{minNDim: 3}];\n    this.dataFormat = args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    for (const dim of inputShape.slice(1)) {\n      if (dim == null) {\n        throw new ValueError(\n            `The shape of the input to \"Flatten\" is not fully defined ` +\n            `(got ${inputShape.slice(1)}). Make sure to pass a complete ` +\n            `\"input_shape\" or \"batch_input_shape\" argument to the first ` +\n            `layer in your model.`);\n      }\n    }\n    return [inputShape[0], arrayProd(inputShape, 1)];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n\n      let input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsFirst' && input.rank > 1) {\n        const permutation: number[] = [0];\n        for (let i = 2; i < input.rank; ++i) {\n          permutation.push(i);\n        }\n        permutation.push(1);\n        input = input.transpose(permutation);\n      }\n\n      return K.batchFlatten(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {};\n    if (this.dataFormat != null) {\n      config['dataFormat'] = this.dataFormat;\n    }\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Flatten);\n\nexport declare interface ActivationLayerArgs extends LayerArgs {\n  /**\n   * Name of the activation function to use.\n   */\n  activation: ActivationIdentifier;\n}\n\nexport class Activation extends Layer {\n  /** @nocollapse */\n  static className = 'Activation';\n  activation: ActivationFn;\n\n  constructor(args: ActivationLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n    this.activation = getActivation(args.activation);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      return this.activation.apply(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {activation: serializeActivation(this.activation)};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Activation);\n\nexport declare interface ReshapeLayerArgs extends LayerArgs {\n  /** The target shape. Does not include the batch axis. */\n  targetShape: Shape;\n}\n\nexport declare interface RepeatVectorLayerArgs extends LayerArgs {\n  /**\n   * The integer number of times to repeat the input.\n   */\n  n: number;\n}\n\nexport class RepeatVector extends Layer {\n  /** @nocollapse */\n  static className = 'RepeatVector';\n  readonly n: number;\n\n  constructor(args: RepeatVectorLayerArgs) {\n    super(args);\n    this.n = args.n;\n    this.inputSpec = [{ndim: 2}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], this.n, inputShape[1]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      return K.repeat(inputs, this.n);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      n: this.n,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(RepeatVector);\n\nexport class Reshape extends Layer {\n  /** @nocollapse */\n  static className = 'Reshape';\n  private targetShape: Shape;\n\n  constructor(args: ReshapeLayerArgs) {\n    super(args);\n    this.targetShape = args.targetShape;\n\n    // Make sure that all unknown dimensions are represented as `null`.\n    for (let i = 0; i < this.targetShape.length; ++i) {\n      if (this.isUnknown(this.targetShape[i])) {\n        this.targetShape[i] = null;\n      }\n    }\n  }\n\n  private isUnknown(dim: number): boolean {\n    return dim < 0 || dim == null;\n  }\n\n  /**\n   * Finds and replaces a missing dimension in output shape.\n   *\n   * This is a near direct port of the internal Numpy function\n   * `_fix_unknown_dimension` in `numpy/core/src/multiarray/shape.c`.\n   *\n   * @param inputShape: Original shape of array begin reshape.\n   * @param outputShape: Target shape of the array, with at most a single\n   * `null` or negative number, which indicates an underdetermined dimension\n   * that should be derived from `inputShape` and the known dimensions of\n   *   `outputShape`.\n   * @returns: The output shape with `null` replaced with its computed value.\n   * @throws: ValueError: If `inputShape` and `outputShape` do not match.\n   */\n  private fixUnknownDimension(inputShape: Shape, outputShape: Shape): Shape {\n    const errorMsg = 'Total size of new array must be unchanged.';\n    const finalShape = outputShape.slice();\n    let known = 1;\n    let unknown = null;\n    for (let i = 0; i < finalShape.length; ++i) {\n      const dim = finalShape[i];\n      if (this.isUnknown(dim)) {\n        if (unknown === null) {\n          unknown = i;\n        } else {\n          throw new ValueError('Can only specifiy one unknown dimension.');\n        }\n      } else {\n        known *= dim;\n      }\n    }\n\n    const originalSize = arrayProd(inputShape);\n    if (unknown !== null) {\n      if (known === 0 || originalSize % known !== 0) {\n        throw new ValueError(errorMsg);\n      }\n      finalShape[unknown] = originalSize / known;\n    } else if (originalSize !== known) {\n      throw new ValueError(errorMsg);\n    }\n\n    return finalShape;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    let anyUnknownDims = false;\n    for (let i = 0; i < inputShape.length; ++i) {\n      if (this.isUnknown(inputShape[i])) {\n        anyUnknownDims = true;\n        break;\n      }\n    }\n\n    if (anyUnknownDims) {\n      return inputShape.slice(0, 1).concat(this.targetShape);\n    } else {\n      return inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const outputShape = inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n      return input.reshape(outputShape);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      targetShape: this.targetShape,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Reshape);\n\nexport declare interface PermuteLayerArgs extends LayerArgs {\n  /**\n   * Array of integers. Permutation pattern. Does not include the\n   * sample (batch) dimension. Index starts at 1.\n   * For instance, `[2, 1]` permutes the first and second dimensions\n   * of the input.\n   */\n  dims: number[];\n}\n\nexport class Permute extends Layer {\n  /** @nocollapse */\n  static className = 'Permute';\n  readonly dims: number[];\n  private readonly dimsIncludingBatch: number[];\n\n  constructor(args: PermuteLayerArgs) {\n    super(args);\n    if (args.dims == null) {\n      throw new Error(\n          'Required configuration field `dims` is missing during Permute ' +\n          'constructor call.');\n    }\n    if (!Array.isArray(args.dims)) {\n      throw new Error(\n          'Permute constructor requires `dims` to be an Array, but received ' +\n          `${args.dims} instead.`);\n    }\n\n    // Check the validity of the permutation indices.\n    const expectedSortedIndices = range(1, args.dims.length + 1);\n    if (!util.arraysEqual(args.dims.slice().sort(), expectedSortedIndices)) {\n      throw new Error(\n          'Invalid permutation `dims`: ' + JSON.stringify(args.dims) +\n          ' `dims` must contain consecutive integers starting from 1.');\n    }\n\n    this.dims = args.dims;\n    this.dimsIncludingBatch = [0].concat(this.dims);\n    this.inputSpec = [new InputSpec({ndim: this.dims.length + 1})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    this.dims.forEach((dim: number, i: number) => {\n      outputShape[i + 1] = (inputShape as Shape)[dim];\n    });\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return transpose(getExactlyOneTensor(inputs), this.dimsIncludingBatch);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      dims: this.dims,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Permute);\n\nexport declare interface MaskingArgs extends LayerArgs {\n  /**\n   * Masking Value. Defaults to `0.0`.\n   */\n  maskValue?: number;\n}\n\nexport class Masking extends Layer {\n  /** @nocollapse */\n  static className = 'Masking';\n  maskValue: number;\n\n  constructor(args?: MaskingArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maskValue = args.maskValue == null ? 0 : args.maskValue;\n    } else {\n      this.maskValue = 0;\n    }\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig() {\n    const baseConfig = super.getConfig();\n    const config = {maskValue: this.maskValue};\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    const input = getExactlyOneTensor(inputs);\n    const axis = -1;\n    return any(notEqual(input, this.maskValue), axis);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const axis = -1;\n      const keepDims = true;\n      const booleanMask = any(notEqual(input, this.maskValue), axis, keepDims);\n      const output = input.mul(booleanMask.asType(input.dtype));\n      return output;\n    });\n  }\n}\nserialization.registerClass(Masking);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Embedding Layer.\n *\n * Original source: keras/constraints.py\n */\nimport {notEqual, serialization, Tensor, tidy, zerosLike} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface EmbeddingLayerArgs extends LayerArgs {\n  /**\n   * Integer > 0. Size of the vocabulary, i.e. maximum integer index + 1.\n   */\n  inputDim: number;\n  /**\n   * Integer >= 0. Dimension of the dense embedding.\n   */\n  outputDim: number;\n  /**\n   * Initializer for the `embeddings` matrix.\n   */\n  embeddingsInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Regularizer function applied to the `embeddings` matrix.\n   */\n  embeddingsRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Constraint function applied to the `embeddings` matrix.\n   */\n  embeddingsConstraint?: ConstraintIdentifier|Constraint;\n  /**\n   * Whether the input value 0 is a special \"padding\" value that should be\n   * masked out. This is useful when using recurrent layers which may take\n   * variable length input.\n   *\n   * If this is `True` then all subsequent layers in the model need to support\n   * masking or an exception will be raised. If maskZero is set to `True`, as a\n   * consequence, index 0 cannot be used in the vocabulary (inputDim should\n   * equal size of vocabulary + 1).\n   */\n  maskZero?: boolean;\n  /**\n   * Length of input sequences, when it is constant.\n   *\n   * This argument is required if you are going to connect `flatten` then\n   * `dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed).\n   */\n  inputLength?: number|number[];\n}\n\nexport class Embedding extends Layer {\n  /** @nocollapse */\n  static className = 'Embedding';\n  private inputDim: number;\n  private outputDim: number;\n  private embeddingsInitializer: Initializer;\n  private maskZero: boolean;\n  private inputLength: number|number[];\n\n  private embeddings: LayerVariable = null;\n\n  readonly DEFAULT_EMBEDDINGS_INITIALIZER: InitializerIdentifier =\n      'randomUniform';\n  private readonly embeddingsRegularizer?: Regularizer;\n  private readonly embeddingsConstraint?: Constraint;\n\n  constructor(args: EmbeddingLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null) {\n      // Porting Note: This logic is copied from Layer's constructor, since we\n      // can't do exactly what the Python constructor does for Embedding().\n      // Specifically, the super constructor can not be called after the\n      // mutation of the `config` argument.\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      if (args.inputLength == null) {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (None, )\n        this.batchInputShape = [batchSize, null];\n      } else {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (config.inputLength, )\n        this.batchInputShape =\n            [batchSize].concat(generic_utils.toList(args.inputLength));\n      }\n    }\n    this.inputDim = args.inputDim;\n    generic_utils.assertPositiveInteger(this.inputDim, 'inputDim');\n    this.outputDim = args.outputDim;\n    generic_utils.assertPositiveInteger(this.outputDim, 'outputDim');\n    this.embeddingsInitializer = getInitializer(\n        args.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER);\n    this.embeddingsRegularizer = getRegularizer(args.embeddingsRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.embeddingsConstraint = getConstraint(args.embeddingsConstraint);\n    this.maskZero = args.maskZero;\n    this.supportsMasking = args.maskZero;\n    this.inputLength = args.inputLength;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    this.embeddings = this.addWeight(\n        'embeddings', [this.inputDim, this.outputDim], this.dtype,\n        this.embeddingsInitializer, this.embeddingsRegularizer, true,\n        this.embeddingsConstraint);\n    this.built = true;\n  }\n\n  // Override warnOnIncompatibleInputShape because an embedding layer allows\n  // the input to have varying ranks.\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {}\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tidy(() => {\n      if (!this.maskZero) {\n        return null;\n      } else {\n        inputs = getExactlyOneTensor(inputs);\n        return notEqual(inputs, zerosLike(inputs));\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    if (this.inputLength == null) {\n      return [...inputShape, this.outputDim];\n    }\n    // inputLength can be an array if input is 3D or higher.\n    const inLens: number[] = generic_utils.toList(this.inputLength);\n    if (inLens.length !== inputShape.length - 1) {\n      throw new ValueError(\n          `\"inputLength\" is ${this.inputLength}, but received ` +\n          `input shape has shape ${inputShape}`);\n    } else {\n      let i = 0;\n      for (let k = 0; k < inLens.length; ++k) {\n        const s1 = inLens[k];\n        const s2 = inputShape[k + 1];\n        if ((s1 != null) && (s2 != null) && (s1 !== s2)) {\n          throw new ValueError(\n              `\"inputLength\" is ${this.inputLength}, but received ` +\n              `input shape has shape ${inputShape}`);\n        } else if (s1 == null) {\n          inLens[i] = s2;\n        }\n        i++;\n      }\n    }\n    return [inputShape[0], ...inLens, this.outputDim];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Embedding layer accepts only a single input.\n      let input = getExactlyOneTensor(inputs);\n      if (input.dtype !== 'int32') {\n        input = K.cast(input, 'int32');\n      }\n      const output = K.gather(this.embeddings.read(), input.as1D());\n      return output.reshape(\n          getExactlyOneShape(this.computeOutputShape(input.shape)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      inputDim: this.inputDim,\n      outputDim: this.outputDim,\n      embeddingsInitializer: serializeInitializer(this.embeddingsInitializer),\n      embeddingsRegularizer: serializeRegularizer(this.embeddingsRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      embeddingsConstraint: serializeConstraint(this.embeddingsConstraint),\n      maskZero: this.maskZero,\n      inputLength: this.inputLength\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Embedding);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Merge Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {l2Normalize} from '../losses';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as mathUtils from '../utils/math_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\n/**\n * Generic Merge layer for element-wise merge functions.\n *\n * Used to implement `Sum`, `Average`, `Concatenate`, etc.\n */\nexport abstract class Merge extends Layer {\n  protected reshapeRequired: boolean;\n\n  constructor(args?: LayerArgs) {\n    super(args || {});\n    this.supportsMasking = true;\n  }\n\n  /**\n   * Logic for merging multiple tensors, to be overridden by subclasses.\n   * @param inputs\n   */\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    throw new NotImplementedError();\n  }\n\n  /**\n   * Computes the shape of the result of an elementwise operation.\n   *\n   * @param shape1: Shape of the first tensor.\n   * @param shape2: Shape of the second tensor.\n   * @returns Expected output shape when an elementwise operation is carried\n   *   out on 2 tensors with shapes `shape1` and `shape2`.\n   * @throws ValueError: If `shape1` and `shape2` are not compatible for\n   *   element-wise operations.\n   */\n  private computeElementwiseOpOutputShape(shape1: Shape, shape2: Shape): Shape {\n    if (shape1 == null || shape2 == null) {\n      return null;\n    } else if (shape1.length < shape2.length) {\n      return this.computeElementwiseOpOutputShape(shape2, shape1);\n    } else if (shape2.length === 0) {\n      return shape1;\n    }\n    const outputShape: Shape = shape1.slice(0, shape1.length - shape2.length);\n    for (let k = 0; k < shape2.length; ++k) {\n      const i = shape1[shape1.length - shape2.length + k];\n      const j = shape2[k];\n      if (i == null || j == null || i < 0 || j < 0) {\n        outputShape.push(null);\n      } else if (i === 1) {\n        outputShape.push(j);\n      } else if (j === 1) {\n        outputShape.push(i);\n      } else {\n        if (i !== j) {\n          throw new ValueError(\n              'Operands could not be broadcast together with shapes ' +\n              JSON.stringify(shape1) + ' ' + JSON.stringify(shape2));\n        }\n        outputShape.push(i);\n      }\n    }\n    return outputShape;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.\n    if (Array.isArray(inputShape) && !Array.isArray(inputShape[0])) {\n      // Make sure that inputShape is an Array of shape.\n      inputShape = [getExactlyOneShape(inputShape)];\n    }\n    inputShape = inputShape as Shape[];\n    if (inputShape.length < 2) {\n      throw new ValueError(\n          'A merge layer should be called on an Array of at least 2 inputs.' +\n          ` Got ${inputShape.length} input(s).`);\n    }\n\n    // Make sure that there is at most one unique batch size among the input\n    // shapes.\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length > 1) {\n      throw new ValueError(\n          `Can not merge tensors with different batch sizes. ` +\n          `Got tensors with shapes: ${JSON.stringify(inputShape)}.`);\n    }\n\n    let outputShape: Shape =\n        inputShape[0] == null ? null : inputShape[0].slice(1);\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n    // If the inputs have different ranks, we have to reshape them to make them\n    // broadcastable.\n    const allRanks = inputShape.map(shape => shape.length);\n    if (inputShape.indexOf(null) === -1 &&\n        generic_utils.unique(allRanks).length === 1) {\n      this.reshapeRequired = false;\n    } else {\n      this.reshapeRequired = true;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (this.reshapeRequired) {\n        const reshapedInputs: Tensor[] = [];\n        const inputDims = inputs.map(input => input.rank);\n        if (inputDims.indexOf(null) === -1) {\n          // If ranks of all inputs are available, we simply expand each of them\n          // at axis=1 until all of them have the same rank.\n          const maxNDim = mathUtils.max(inputDims);\n          for (let x of inputs) {\n            const xNDim = x.rank;\n            for (let k = 0; k < maxNDim - xNDim; ++k) {\n              x = K.expandDims(x, 1);\n            }\n            reshapedInputs.push(x);\n          }\n          return this.mergeFunction(reshapedInputs);\n        } else {\n          // Transpose all inputs so that batch size is the last dimension.\n          // [batchSize, dim1, dim2, ...] -> [dim1, dim2, ..., batchSize]\n          let transposed = false;\n          for (const x of inputs) {\n            const xNDim = x.rank;\n            if (xNDim == null) {\n              const xShape = x.shape;\n              const batchSize = xShape[0];\n              const newShape = xShape.slice(1).concat([batchSize]);\n              let xTransposed = x.reshape(\n                  [batchSize].concat(mathUtils.arrayProd(xShape.slice(1))));\n              xTransposed = tfc.transpose(xTransposed, [1, 0]);\n              xTransposed = xTransposed.reshape(newShape);\n              reshapedInputs.push(xTransposed);\n              transposed = true;\n            } else if (xNDim > 1) {\n              const dims = mathUtils.range(1, xNDim).concat([0]);\n              reshapedInputs.push(tfc.transpose(x, dims));\n              transposed = true;\n            } else {\n              // We don't transpose inputs if they are 1D vectors or scalars.\n              reshapedInputs.push(x);\n            }\n          }\n          let y = this.mergeFunction(reshapedInputs);\n          const yNDim = y.rank;\n          if (transposed) {\n            // If inputs have been transposed, we have to transpose the output\n            // too.\n            if (yNDim == null) {\n              const yShape = y.shape;\n              const yNDim = yShape.length;\n              const batchSize = yShape[yNDim - 1];\n              const newShape =\n                  [batchSize].concat(yShape.slice(0, yShape.length - 1));\n              y = tfc.transpose(y.reshape([-1, batchSize]), [1, 0])\n                      .reshape(newShape);\n            } else if (yNDim > 1) {\n              const dims = [yNDim - 1].concat(mathUtils.range(0, yNDim - 1));\n              y = tfc.transpose(y, dims);\n            }\n          }\n          return y;\n        }\n      } else {\n        return this.mergeFunction(inputs);\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape[];\n    let outputShape: Shape;\n    if (inputShape[0] == null) {\n      outputShape = null;\n    } else {\n      outputShape = inputShape[0].slice(1);\n    }\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length === 1) {\n      outputShape = batchSizes.concat(outputShape);\n    } else {\n      outputShape = [null].concat(outputShape);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tfc.tidy(() => {\n      if (mask == null) {\n        return null;\n      }\n      if (!Array.isArray(mask)) {\n        throw new ValueError('`mask` should be an Array');\n      }\n      if (!Array.isArray(inputs)) {\n        throw new ValueError('`inputs` should be an Array');\n      }\n      if (mask.length !== inputs.length) {\n        throw new ValueError(\n            `The Array 'inputs' and 'mask' are expected to have the same ` +\n            `length, but have different lengths ` +\n            `(${inputs.length} vs ${mask.length})`);\n      }\n      if (mask.every(m => m == null)) {\n        return null;\n      }\n      mask = mask.map(m => m == null ? m : tfc.expandDims(m, 0));\n      let output = mask[0];\n      for (let i = 1; i < mask.length - 1; ++i) {\n        output = tfc.logicalAnd(output, mask[i]);\n      }\n      return output;\n    });\n  }\n}\n\nexport class Add extends Merge {\n  /** @nocollapse */\n  static className = 'Add';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Add);\n\n/**\n * Calculate the element-wise sum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Add` layer, by using no input argument\n *    or a single configuration argument. The resultant `Add` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const addLayer = tf.layers.add();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = addLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.add([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.add([input1, input2]).print();\n * // Gives [[11, 22], [33, 44]].\n *\n */\nexport function add(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Add({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Add(config);\n  }\n}\n\nexport class Multiply extends Merge {\n  /** @nocollapse */\n  static className = 'Multiply';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.mul(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Multiply);\n\n/**\n * Calculate the element-wise product of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Multiply` layer, by using no input argument\n *    or a single configuration argument. The resultant `Multiply` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const multiplyLayer = tf.layers.multiply();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = multiplyLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.multiply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.multiply([input1, input2]).print();\n * // Gives [[10, 40], [90, 160]].\n *\n */\nexport function multiply(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Multiply({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Multiply(config);\n  }\n}\n\nexport class Average extends Merge {\n  /** @nocollapse */\n  static className = 'Average';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return tfc.mul(1 / inputs.length, output);\n    });\n  }\n}\nserialization.registerClass(Average);\n\n/**\n * Calculate the element-wise arithmetic mean of inputs, which all have the same\n * shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Average` layer, by using no input argument\n *    or a single configuration argument. The resultant `Average` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const averageLayer = tf.layers.average();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = averageLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.average([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.average([input1, input2]).print();\n * // Gives [[5.5, 11], [16.5, 22]].\n *\n */\nexport function average(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Average({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Average(config);\n  }\n}\n\nexport class Maximum extends Merge {\n  /** @nocollapse */\n  static className = 'Maximum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.maximum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Maximum);\n\n/**\n * Calculate the element-wise maximum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Maximum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Maximum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const maximumLayer = tf.layers.maximum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = maximumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.maximum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.maximum([input1, input2]).print();\n * // Gives [[10, 20], [30, 40]].\n *\n */\nexport function maximum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Maximum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Maximum(config);\n  }\n}\n\nexport class Minimum extends Merge {\n  /** @nocollapse */\n  static className = 'Minimum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.minimum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Minimum);\n\n/**\n * Calculate the element-wise minimum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Minimum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Minimum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const minimumLayer = tf.layers.minimum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = minimumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.minimum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.minimum([input1, input2]).print();\n * // Gives [[1, 2], [3, 4]].\n *\n */\nexport function minimum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Minimum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Minimum(config);\n  }\n}\n\nexport declare interface ConcatenateLayerArgs extends LayerArgs {\n  /**\n   * Axis along which to concatenate.\n   */\n  axis?: number;\n}\n\nexport class Concatenate extends Merge {\n  /** @nocollapse */\n  static className = 'Concatenate';\n  readonly DEFAULT_AXIS = -1;\n  private readonly axis: number;\n\n  constructor(args?: ConcatenateLayerArgs) {\n    super(args);\n    if (args == null) {\n      args = {};\n    }\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.]\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0])) ||\n        inputShape.length === 1) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of at least 2 ' +\n          'inputs');\n    }\n    inputShape = inputShape as Shape[];\n\n    let allNoneShape = true;\n    for (const shape of inputShape) {\n      if (shape != null) {\n        allNoneShape = false;\n        break;\n      }\n    }\n    if (allNoneShape) {\n      return;\n    }\n\n    const shapeSet: Shape[] = [];\n    for (let i = 0; i < inputShape.length; ++i) {\n      const shapeWithoutConcatAxis = inputShape[i].slice();\n      shapeWithoutConcatAxis.splice(this.axis, 1);\n      let exists = false;\n      for (const shape of shapeSet) {\n        if (util.arraysEqual(shape, shapeWithoutConcatAxis)) {\n          exists = true;\n          break;\n        }\n      }\n      if (!exists) {\n        shapeSet.push(shapeWithoutConcatAxis);\n      }\n    }\n    if (shapeSet.length > 1) {\n      throw new ValueError(\n          'A `Concatenate` layer requires inputs with matching shapes ' +\n          'except for the concat axis. Got input shapes: ' +\n          JSON.stringify(inputShape));\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      return K.concatenate(inputs, this.axis);\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0]))) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of inputs.');\n    }\n    const inputShapes = inputShape as Shape[];\n    const outputShape = inputShapes[0].slice();\n    const axis = this.axis < 0 ? outputShape.length + this.axis : this.axis;\n    // Porting Note: the line above is because TypeScript doesn't support\n    //   negative indices.\n    for (const shape of inputShapes.slice(1)) {\n      if (outputShape[axis] == null || shape[axis] == null) {\n        outputShape[axis] = null;\n        break;\n      }\n      outputShape[axis] += shape[axis];\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    if (mask == null) {\n      return null;\n    }\n    if (!Array.isArray(mask)) {\n      throw new ValueError('`mask` should be an array for Concatenate');\n    }\n    if (!Array.isArray(inputs)) {\n      throw new ValueError('`inputs` should be an array for Concatenate');\n    }\n    if (mask.length !== inputs.length) {\n      throw new ValueError(\n          `Mismatch in the length of mask (${mask.length}) ` +\n          `and the legnth of inputs (${inputs.length})`);\n    }\n    return tfc.tidy(() => {\n      let allNullMasks = true;\n      mask.forEach(m => {\n        if (m != null) {\n          allNullMasks = false;\n          return;\n        }\n      });\n      if (allNullMasks) {\n        return null;\n      }\n      const outputMasks: Tensor[] = [];\n      for (let i = 0; i < inputs.length; ++i) {\n        if (mask[i] == null) {\n          // Input is unmasked. Append all 1's to masks.\n          outputMasks.push(tfc.onesLike(inputs[i]).asType('bool'));\n        } else if (mask[i].rank < inputs[i].rank) {\n          // Mask is smaller than the input, expand it.\n          outputMasks.push(tfc.expandDims(mask[i], -1));\n        } else {\n          outputMasks.push(mask[i]);\n        }\n      }\n      const concatenatedMasks = tfc.concat(outputMasks, this.axis);\n      return tfc.all(concatenatedMasks, -1, false);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axis': this.axis,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Concatenate);\n\n/**\n * Concatenate an `Array` of inputs.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Concatenate` layer, by using no input argument\n *    or a single configuration argument. The resultant `Concatenate` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const concatLayer = tf.layers.concatenate();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = concatLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 7], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = tf.layers.concatenate([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([[1, 2], [3, 4]], [2, 2]);\n * const input2 = tf.tensor2d([[10, 20], [30, 40]], [2, 2]);\n * tf.layers.concatenate([input1, input2]).print();\n * // Gives [[1, 2, 10, 20], [3, 4, 30, 40]].\n *\n */\nexport function concatenate(config?: SymbolicTensor[]|Tensor[]|\n                            ConcatenateLayerArgs): Layer|SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Concatenate({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Concatenate(config);\n  }\n}\n\nexport declare interface DotLayerArgs extends LayerArgs {\n  /**\n   * Axis or axes along which the dot product will be taken.\n   *\n   * Integer or an Array of integers.\n   */\n  axes: number|[number, number];\n\n  /**\n   * Whether to L2-normalize samples along the dot product axis\n   * before taking the dot product.\n   *\n   * If set to `true`, the output of the dot product isthe cosine\n   * proximity between the two samples.\n   */\n  normalize?: boolean;\n}\n\n/**\n * Interpretable potentially negative axis index.\n *\n * For example, given axis = -1, and dim = 3, this function will return 2.\n *\n * @param axis The axis index, may be a positive, zero or negative integer.\n * @param dim Total number of dimensions, a positive integer.\n * @returns A non-negative axis index equivalent to the input `axis`.\n */\nfunction interpretAxis(axis: number, dim: number): number {\n  while (axis < 0) {\n    axis += dim;\n  }\n  return axis;\n}\n\nfunction batchDot(x: Tensor, y: Tensor, axes: number|[number, number]): Tensor {\n  if (x.shape.length > 3 || y.shape.length > 3) {\n    throw new NotImplementedError(\n        'batchDot is not implemented for tensors of 4D or higher rank yet');\n  }\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of x to be >= 2, ` +\n          `but got ${x.shape.length}`);\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of y to be >= 2, ` +\n          `but got ${y.shape.length}`);\n\n  if (typeof axes === 'number') {\n    axes = [axes, axes];\n  }\n\n  if (x.dtype === 'complex64' || y.dtype === 'complex64') {\n    throw new NotImplementedError(\n        'batchDot is not implemented for complex64-type Tensors yet.');\n  }\n\n  const xNDim = x.shape.length;\n  const yNDim = y.shape.length;\n  if (axes == null) {\n    // Behave like batchMatmul by default.\n    axes = [xNDim - 1, yNDim - 2];\n  }\n  const axesArray = axes as [number, number];\n\n  return tfc.tidy(() => {\n    let diff: number;\n    if (xNDim > yNDim) {\n      diff = xNDim - yNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      y = y.reshape(y.shape.concat(diffShape));\n    } else if (yNDim > xNDim) {\n      diff = yNDim - xNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      x = x.reshape(x.shape.concat(diffShape));\n    } else {\n      diff = 0;\n    }\n\n    let out: Tensor;\n    if (x.shape.length === 2 && y.shape.length === 2) {\n      if (axesArray[0] === axesArray[1]) {\n        out = x.mul(y).sum(axesArray[0]);\n      } else {\n        out = x.transpose([1, 0]).mul(y).sum(axesArray[1]);\n      }\n    } else {\n      const adjX = axesArray[0] !== x.shape.length - 1;\n      const adjY = axesArray[1] === y.shape.length - 1;\n      out = x.matMul(y, adjX, adjY);\n    }\n\n    if (diff > 0) {\n      let idx: number;\n      if (xNDim > yNDim) {\n        idx = xNDim + yNDim - 3;\n      } else {\n        idx = xNDim - 1;\n      }\n      const squeezeAxes: number[] = [];\n      for (let i = idx; i < idx + diff; ++i) {\n        squeezeAxes.push(i);\n      }\n      out = out.squeeze(squeezeAxes);\n    }\n    if (out.shape.length === 1) {\n      out = out.expandDims(1);\n    }\n    return out;\n  });\n}\n\nexport class Dot extends Merge {\n  /** @nocollapse */\n  static className = 'Dot';\n\n  private axes: number|[number, number];\n  private normalize: boolean;\n\n  constructor(args: DotLayerArgs) {\n    super(args);\n    this.axes = args.axes;\n    this.normalize = args.normalize == null ? false : args.normalize;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = inputShape[0] as Shape;\n    const shape2 = inputShape[1] as Shape;\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    if (shape1[axes[0]] !== shape2[axes[1]]) {\n      throw new ValueError(\n          `Dimension incompatibility: ` +\n          `${shape1[axes[0]]} !== ${shape2[axes[1]]}`);\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    if (inputs.length !== 2) {\n      throw new ValueError(\n          'A `Dot` layer must be called on exactly 2 inputs, ' +\n          `but received ${inputs.length} input(s).`);\n    }\n\n    let x1 = inputs[0];\n    let x2 = inputs[1];\n    let axes: [number, number];\n    if (!Array.isArray(this.axes)) {\n      axes = [\n        interpretAxis(this.axes, x1.shape.length),\n        interpretAxis(this.axes, x2.shape.length)\n      ];\n    } else {\n      axes = this.axes.map(\n                 (axis, i) => interpretAxis(\n                     axis, inputs[i].shape.length)) as [number, number];\n    }\n    if (this.normalize) {\n      x1 = l2Normalize(x1, axes[0]);\n      x2 = l2Normalize(x2, axes[1]);\n    }\n    return batchDot(x1, x2, axes);\n  }\n\n  private interpretAxes(shape1: Shape, shape2: Shape): number[] {\n    let axes: number[];\n    if (!Array.isArray(this.axes)) {\n      // `this.axes` is a single integer.\n      axes = [\n        interpretAxis(this.axes, shape1.length),\n        interpretAxis(this.axes, shape2.length)\n      ];\n    } else {\n      // `this.axes` is an Array of integers.\n      axes = this.axes;\n    }\n    return axes;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = (inputShape[0] as Shape).slice();\n    const shape2 = (inputShape[1] as Shape).slice();\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    shape1.splice(axes[0], 1);\n    shape2.splice(axes[1], 1);\n    shape2.splice(0, 1);\n    const outputShape = shape1.concat(shape2);\n    if (outputShape.length === 1) {\n      outputShape.push(1);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return null;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axes': this.axes,\n      'normalize': this.normalize\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dot);\n\n// TODO(cais): Add functional interfaces for the merge layers.\n", "/**\r\n * @license\r\n * Copyright 2018 Google LLC\r\n *\r\n * Use of this source code is governed by an MIT-style\r\n * license that can be found in the LICENSE file or at\r\n * https://opensource.org/licenses/MIT.\r\n * =============================================================================\r\n */\r\n\r\n/**\r\n * TensorFlow.js Layers: Noise Layers.\r\n */\r\n\r\nimport {greaterEqual, randomUniform, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\r\n\r\nimport * as K from '../backend/tfjs_backend';\r\nimport {Layer, LayerArgs} from '../engine/topology';\r\nimport {Shape} from '../keras_format/common';\r\nimport {Kwargs} from '../types';\r\nimport {getExactlyOneTensor} from '../utils/types_utils';\r\n\r\nexport declare interface GaussianNoiseArgs extends LayerArgs {\r\n  /** Standard Deviation.  */\r\n  stddev: number;\r\n}\r\n\r\nexport class GaussianNoise extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianNoise';\r\n  readonly stddev: number;\r\n\r\n  constructor(args: GaussianNoiseArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.stddev = args.stddev;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {stddev: this.stddev};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      const noised = () =>\r\n          K.randomNormal(input.shape, 0, this.stddev).add(input);\r\n      const output =\r\n          K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      return output;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianNoise);\r\n\r\nexport declare interface GaussianDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n}\r\n\r\nexport class GaussianDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianDropout';\r\n  readonly rate: number;\r\n\r\n  constructor(args: GaussianDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      if (this.rate > 0 && this.rate < 1) {\r\n        const noised = () => {\r\n          const stddev = Math.sqrt(this.rate / (1 - this.rate));\r\n          return input.mul(K.randomNormal(input.shape, 1, stddev));\r\n        };\r\n        return K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      }\r\n      return input;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianDropout);\r\n\r\nexport declare interface AlphaDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n  /**\r\n   * A 1-D `Tensor` of type `int32`, representing the\r\n   * shape for randomly generated keep/drop flags.\r\n   */\r\n  noiseShape?: Shape;\r\n}\r\n\r\n/**\r\n * Applies Alpha Dropout to the input.\r\n *\r\n * As it is a regularization layer, it is only active at training time.\r\n *\r\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\r\n * to their original values, in order to ensure the self-normalizing property\r\n * even after this dropout.\r\n * Alpha Dropout fits well to Scaled Exponential Linear Units\r\n * by randomly setting activations to the negative saturation value.\r\n *\r\n * Arguments:\r\n *   - `rate`: float, drop probability (as with `Dropout`).\r\n *     The multiplicative noise will have\r\n *     standard deviation `sqrt(rate / (1 - rate))`.\r\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\r\n *     shape for randomly generated keep/drop flags.\r\n *\r\n * Input shape:\r\n *   Arbitrary. Use the keyword argument `inputShape`\r\n *   (tuple of integers, does not include the samples axis)\r\n *   when using this layer as the first layer in a model.\r\n *\r\n * Output shape:\r\n *   Same shape as input.\r\n *\r\n * References:\r\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\r\n */\r\nexport class AlphaDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'AlphaDropout';\r\n  readonly rate: number;\r\n  readonly noiseShape: Shape;\r\n\r\n  constructor(args: AlphaDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n    this.noiseShape = args.noiseShape;\r\n  }\r\n\r\n  _getNoiseShape(inputs: Tensor|Tensor[]) {\r\n    return this.noiseShape || getExactlyOneTensor(inputs).shape;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      if (this.rate < 1 && this.rate > 0) {\r\n        const noiseShape = this._getNoiseShape(inputs);\r\n\r\n        const droppedInputs = () => {\r\n          const input = getExactlyOneTensor(inputs);\r\n\r\n          const alpha = 1.6732632423543772848170429916717;\r\n          const scale = 1.0507009873554804934193349852946;\r\n\r\n          const alphaP = -alpha * scale;\r\n\r\n          let keptIdx = greaterEqual(randomUniform(noiseShape), this.rate);\r\n\r\n          keptIdx = K.cast(keptIdx, 'float32');  // get default dtype.\r\n\r\n          // Get affine transformation params.\r\n          const a = ((1 - this.rate) * (1 + this.rate * alphaP ** 2)) ** -0.5;\r\n          const b = -a * alphaP * this.rate;\r\n\r\n          // Apply mask.\r\n          const x = input.mul(keptIdx).add(keptIdx.add(-1).mul(alphaP));\r\n\r\n          return x.mul(a).add(b);\r\n        };\r\n        return K.inTrainPhase(\r\n            droppedInputs, () => getExactlyOneTensor(inputs),\r\n            kwargs['training'] || false);\r\n      }\r\n      return inputs;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(AlphaDropout);\r\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Normalization layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {moments, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Applies batch normalization on x given mean, var, beta and gamma.\n *\n * I.e. returns:\n *   `output = (x - mean) / (sqrt(var) + epsilon) * gamma + beta`\n *\n * @param x Input tensor.\n * @param mean Mean of batch.\n * @param variance Variance of batch.\n * @param beta Tensor with which to center the input.\n * @param gamma Tensor by which to scale the input.\n * @param epsilon Fuzz factor.\n * @returns The result of the batch normalization.\n */\nexport function batchNormalization(\n    x: Tensor, mean: Tensor, variance: Tensor, beta?: Tensor, gamma?: Tensor,\n    epsilon = 1e-3): Tensor {\n  let out: Tensor;\n  if (x.rank === 2) {\n    out = tfc.batchNorm2d(\n        x as Tensor2D, mean as Tensor2D | Tensor1D,\n        variance as Tensor2D | Tensor1D, beta as Tensor2D | Tensor1D,\n        gamma as Tensor2D | Tensor1D, epsilon);\n  } else if (x.rank === 3) {\n    // TODO(cais): Check rank; give proper error message.\n    out = tfc.batchNorm3d(\n        x as Tensor3D, mean as Tensor3D | Tensor1D,\n        variance as Tensor3D | Tensor1D, beta as Tensor3D | Tensor1D,\n        gamma as Tensor3D | Tensor1D, epsilon);\n  } else if (x.rank === 4) {\n    out = tfc.batchNorm4d(\n        x as Tensor4D, mean as Tensor4D | Tensor1D,\n        variance as Tensor4D | Tensor1D, beta as Tensor4D | Tensor1D,\n        gamma as Tensor4D | Tensor1D, epsilon);\n  } else {\n    throw new NotImplementedError(\n        `batchNormalization is not implemented for array of rank ${x.rank} ` +\n        `yet`);\n  }\n  return out;\n}\n\n/**\n * Non-broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction regularNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const normed =\n               batchNormalization(x, mean, variance, beta, gamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction broadcastNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const targetShape: number[] = [];\n           for (const axis of math_utils.range(0, x.rank)) {\n             if (reductionAxes.indexOf(axis) !== -1) {\n               targetShape.push(1);\n             } else {\n               targetShape.push(x.shape[axis]);\n             }\n           }\n           const broadcastMean = mean.reshape(targetShape);\n           const broadcastVariance = variance.reshape(targetShape);\n           const broadcastGamma =\n               gamma == null ? null : gamma.reshape(targetShape);\n           const broadcastBeta =\n               beta == null ? null : beta.reshape(targetShape);\n           const normed = batchNormalization(\n               x, broadcastMean, broadcastVariance, broadcastBeta,\n               broadcastGamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Batch normalization for use in training (not inference).\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nexport function normalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  if (util.arraysEqual(\n          reductionAxes.slice().sort(), math_utils.range(0, x.rank - 1))) {\n    return regularNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  } else {\n    return broadcastNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  }\n}\n\nexport declare interface BatchNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The integer axis that should be normalized (typically the features axis).\n   * Defaults to -1.\n   *\n   * For instance, after a `Conv2D` layer with `data_format=\"channels_first\"`,\n   * set `axis=1` in `batchNormalization`.\n   */\n  axis?: number;\n\n  /**\n   * Momentum of the moving average. Defaults to 0.99.\n   */\n  momentum?: number;\n\n  /**\n   * Small float added to the variance to avoid dividing by zero. Defaults to\n   * 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Defaults to `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear (also e.g. `nn.relu`),\n   * this can be disabled since the scaling will be done by the next layer.\n   * Defaults to `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   *  Defaults to 'zeros'.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   *  Defaults to `ones`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving mean.\n   * Defaults to `zeros`\n   */\n  movingMeanInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving variance.\n   *  Defaults to 'Ones'.\n   */\n  movingVarianceInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the beta weight.\n   */\n  betaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for gamma weight.\n   */\n  gammaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer for the beta weight.\n   */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer for the gamma weight.\n   */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class BatchNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'BatchNormalization';\n  private readonly axis: number;\n  private readonly momentum: number;\n  private readonly epsilon: number;\n  private readonly center: boolean;\n  private readonly scale: boolean;\n  private readonly betaInitializer: Initializer;\n  private readonly gammaInitializer: Initializer;\n  private readonly movingMeanInitializer: Initializer;\n  private readonly movingVarianceInitializer: Initializer;\n  private readonly betaConstraint: Constraint;\n  private readonly gammaConstraint: Constraint;\n  private readonly betaRegularizer: Regularizer;\n  private readonly gammaRegularizer: Regularizer;\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n  private movingMean: LayerVariable;\n  private movingVariance: LayerVariable;\n\n  constructor(args?: BatchNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.supportsMasking = true;\n    this.axis = args.axis == null ? -1 : args.axis;\n    this.momentum = args.momentum == null ? 0.99 : args.momentum;\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.movingMeanInitializer =\n        getInitializer(args.movingMeanInitializer || 'zeros');\n    this.movingVarianceInitializer =\n        getInitializer(args.movingVarianceInitializer || 'ones');\n    this.betaConstraint = getConstraint(args.betaConstraint);\n    this.gammaConstraint = getConstraint(args.gammaConstraint);\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const axis = this.axis >= 0 ? this.axis : (this.axis + inputShape.length);\n    const dim = inputShape[axis];\n    if (dim == null) {\n      throw new ValueError(\n          `Axis ${axis} of input tensor should have a defined dimension but ` +\n          `the layer received an input with shape ` +\n          `${JSON.stringify(inputShape)}.`);\n    }\n    this.inputSpec =\n        [new InputSpec({ndim: inputShape.length, axes: {[axis]: dim}})];\n    const shape = [dim];\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', shape, null, this.gammaInitializer, this.gammaRegularizer,\n          true, this.gammaConstraint);\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', shape, null, this.betaInitializer, this.betaRegularizer, true,\n          this.betaConstraint);\n    }\n    this.movingMean = this.addWeight(\n        'moving_mean', shape, null, this.movingMeanInitializer, null, false);\n    this.movingVariance = this.addWeight(\n        'moving_variance', shape, null, this.movingVarianceInitializer, null,\n        false);\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const ndim = inputShape.length;\n      const reductionAxes = math_utils.range(0, ndim);\n      const axis = this.axis >= 0 ? this.axis : (this.axis + ndim);\n      reductionAxes.splice(axis, 1);\n      const broadcastShape = generic_utils.pyListRepeat(1, ndim);\n      broadcastShape[axis] = inputShape[axis];\n\n      const sortedReductionAxes = reductionAxes.slice();\n      sortedReductionAxes.sort();\n      const needsBroadcasting = !util.arraysEqual(\n          sortedReductionAxes, math_utils.range(0, ndim).slice(0, ndim - 1));\n\n      const normalizeInference: () => Tensor = () => {\n        if (needsBroadcasting) {\n          const broadcastMovingMean =\n              this.movingMean.read().reshape(broadcastShape);\n          const broadcastMovingVariance =\n              this.movingVariance.read().reshape(broadcastShape);\n          const broadcastBeta =\n              this.center ? this.beta.read().reshape(broadcastShape) : null;\n          const broadcastGamma =\n              this.scale ? this.gamma.read().reshape(broadcastShape) : null;\n          return batchNormalization(\n              input, broadcastMovingMean, broadcastMovingVariance,\n              broadcastBeta, broadcastGamma, this.epsilon);\n        } else {\n          return batchNormalization(\n              input, this.movingMean.read(), this.movingVariance.read(),\n              this.beta == null ? null : this.beta.read(),\n              this.gamma == null ? null : this.gamma.read(), this.epsilon);\n        }\n      };\n\n      if (!training) {\n        return normalizeInference();\n      }\n\n      const [normedTraining, mean, variance] = normalizeBatchInTraining(\n          input, this.gamma.read(), this.beta.read(), reductionAxes,\n          this.epsilon);\n\n      const doMovingAverage =\n          (variable: LayerVariable, value: Tensor, momentum: number): void => {\n            tfc.tidy(() => {\n              const decay = 1 - momentum;\n              const origValue = variable.read();\n              const updateDelta = origValue.sub(value).mul(decay);\n              variable.write(origValue.sub(updateDelta));\n            });\n          };\n\n      // Perform updates to moving mean and moving variance for training.\n      // Porting Note: In PyKeras, these updates to `movingMean` and\n      //   `movingAverage` are done as a deferred Graph, added to the `Layer`'s\n      //   `update`s using the `add_update()` method. Here we do it imperatively\n      //   and encapsulate the updates in a function that is invoked\n      //   immediately.\n      const updateMovingMeanAndVariance = () => {\n        doMovingAverage(this.movingMean, mean, this.momentum);\n        doMovingAverage(this.movingVariance, variance, this.momentum);\n      };\n      updateMovingMeanAndVariance();\n\n      return normedTraining;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      momentum: this.momentum,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      movingMeanInitializer: serializeInitializer(this.movingMeanInitializer),\n      movingVarianceInitializer:\n          serializeInitializer(this.movingVarianceInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer),\n      betaConstraint: serializeConstraint(this.betaConstraint),\n      gammaConstraint: serializeConstraint(this.gammaConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(BatchNormalization);\n\nexport interface LayerNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The axis or axes that should be normalized (typically, the feature axis.)\n   * Defaults to -1 (the last axis.)\n   */\n  axis?: number|number[];\n\n  /**\n   * A small positive float added to variance to avoid divison by zero.\n   * Defaults to 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Default: `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply output by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear, this can be disabled since scaling will\n   * be done by the next layer.\n   * Default: `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   * Default: `'zeros'`.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   * Default: `'ones'`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /** Regularizer for the beta weight. */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /** Regularizer for the gamma weight. */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class LayerNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'LayerNormalization';\n\n  private axis: number|number[];\n  readonly epsilon: number;\n  readonly center: boolean;\n  readonly scale: boolean;\n  readonly betaInitializer: Initializer;\n  readonly gammaInitializer: Initializer;\n  readonly betaRegularizer: Regularizer;\n  readonly gammaRegularizer: Regularizer;\n\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n\n  constructor(args?: LayerNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.axis = args.axis == null ? -1 : args.axis;\n    if (typeof this.axis === 'number') {\n      if (!Number.isInteger(this.axis)) {\n        throw new Error(\n            `Expected axis to be an integer, but received ${this.axis}`);\n      }\n    } else if (Array.isArray(this.axis)) {\n      for (const axis of this.axis) {\n        if (!Number.isInteger(axis)) {\n          throw new Error(\n              `Expected axis to be an array of integers, ` +\n              `but received ${JSON.stringify(this.axis)}`);\n        }\n      }\n    } else {\n      throw new Error(\n          `Expected axis to be an integer or an array of integers, ` +\n          `but received ${JSON.stringify(this.axis)}`);\n    }\n\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n\n    this.supportsMasking = true;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const nDims = inputShape.length;\n\n    // Convert axis to array and resolve negatives.\n    if (typeof this.axis === 'number') {\n      this.axis = [this.axis];\n    }\n    for (let i = 0; i < this.axis.length; ++i) {\n      if (this.axis[i] < 0) {\n        this.axis[i] += nDims;\n      }\n    }\n\n    // Further validate axes.\n    for (const axis of this.axis) {\n      if (axis < 0 || axis >= nDims) {\n        throw new Error(`Invalid axis: ${axis}`);\n      }\n    }\n    if (this.axis.length !== generic_utils.unique(this.axis).length) {\n      throw new Error(`Found duplicate axes in: ${this.axis}`);\n    }\n\n    const paramShape = this.axis.map(axis => inputShape[axis]) as number[];\n\n    const trainable = true;\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', paramShape, 'float32', this.gammaInitializer,\n          this.gammaRegularizer, trainable);\n    } else {\n      this.gamma = null;\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', paramShape, 'float32', this.betaInitializer,\n          this.betaRegularizer, trainable);\n    } else {\n      this.beta = null;\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const input = getExactlyOneTensor(inputs);\n    const inputShape = input.shape;\n    const nDims = inputShape.length;\n\n    return tidy(() => {\n      const keepDims = true;\n      let {mean, variance} = moments(input, this.axis, keepDims);\n      const broadcastShape = generic_utils.pyListRepeat(1, nDims);\n      for (const dim of this.axis as number[]) {\n        broadcastShape[dim] = inputShape[dim];\n      }\n\n      const broadcast = (v: Tensor) => {\n        if (v != null && v.shape.length !== nDims &&\n            this.axis !== [nDims - 1]) {\n          return v.reshape(broadcastShape);\n        } else {\n          return v;\n        }\n      };\n\n      let scale = broadcast(this.gamma.read());\n      let offset = broadcast(this.beta.read());\n\n      // TODO(https://github.com/tensorflow/tfjs/issues/2120): The tiling below\n      // is a workaround for the limitation of core's batchNormalization?d don't\n      // support broadcasting in their gradients. In addition, the tiling is\n      // necessary to ensure correctness on the browser CPU backend regardless\n      // of forward or backward computation. Remove this workaround once the\n      // limitation is addressed. See .\n      const momentsTiling: number[] = [];\n      const scaleOffsetTiling: number[] = [];\n      for (let i = 0; i < nDims; ++i) {\n        if ((this.axis as number[]).indexOf(i) !== -1) {\n          momentsTiling.push(inputShape[i]);\n          scaleOffsetTiling.push(1);\n        } else {\n          momentsTiling.push(1);\n          scaleOffsetTiling.push(inputShape[i]);\n        }\n      }\n      mean = mean.tile(momentsTiling);\n      variance = variance.tile(momentsTiling);\n      scale = scale.tile(scaleOffsetTiling);\n      offset = offset.tile(scaleOffsetTiling);\n\n      return batchNormalization(\n          input, mean, variance, offset, scale, this.epsilon);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LayerNormalization);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Padding Layers.\n */\n\n// Porting Note: In Python Keras, the padding layers are in convolutional.py,\n//   but we decided to put them in a separate file (padding.ts) for clarity.\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\n/**\n * Pads the middle dimension of a 3D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of 2 integers, how many zeros to add at the start and\n *   end of the middle dimension (i.e., dimension 1).\n * @return A padded 3D `tf.Tensor`.\n */\nexport function temporalPadding(x: Tensor, padding?: [number, number]): Tensor {\n  return tidy(() => {\n    if (x.rank !== 3) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 3-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [1, 1];\n    }\n    if (padding.length !== 2) {\n      throw new ValueError(\n          `temporalPadding expects input padding pattern to be a length-2 ` +\n          `array, but received a length-${padding.length} array.`);\n    }\n\n    const pattern: Array<[number, number]> = [[0, 0], padding, [0, 0]];\n    return tfc.pad(x, pattern);\n  });\n}\n\n/**\n * Pads the 2nd and 3rd dimensions of a 4D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of two `Array`s, each of which is an `Array` of two\n *   integers. The amount of padding at the beginning and end of the 2nd and 3rd\n *   dimensions, respectively.\n * @param dataFormat 'channelsLast' (default) or 'channelsFirst'.\n * @return Padded 4D `tf.Tensor`.\n */\nexport function spatial2dPadding(\n    x: Tensor, padding?: [[number, number], [number, number]],\n    dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 4-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [[1, 1], [1, 1]];\n    }\n    if (padding.length !== 2 || padding[0].length !== 2 ||\n        padding[1].length !== 2) {\n      throw new ValueError(\n          'spatial2dPadding expects `padding` to be an Array of two Arrays, ' +\n          'each of which is an Array of two integers.');\n    }\n\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (dataFormat !== 'channelsLast' && dataFormat !== 'channelsFirst') {\n      throw new ValueError(\n          `Unknown data format: ${dataFormat}. ` +\n          `Supported data formats are 'channelsLast' and 'channelsFirst.`);\n    }\n\n    let pattern: Array<[number, number]>;\n    if (dataFormat === 'channelsFirst') {\n      pattern = [[0, 0], [0, 0], padding[0], padding[1]];\n    } else {\n      pattern = [[0, 0], padding[0], padding[1], [0, 0]];\n    }\n\n    return tfc.pad(x, pattern);\n  });\n}\n\nexport declare interface ZeroPadding2DLayerArgs extends LayerArgs {\n  /**\n   * Integer, or `Array` of 2 integers, or `Array` of 2 `Array`s, each of\n   * which is an `Array` of 2 integers.\n   * - If integer, the same symmetric padding is applied to width and height.\n   * - If Array` of 2 integers, interpreted as two different symmetric values\n   *   for height and width:\n   *   `[symmetricHeightPad, symmetricWidthPad]`.\n   * - If `Array` of 2 `Array`s, interpreted as:\n   *   `[[topPad, bottomPad], [leftPad, rightPad]]`.\n   */\n  padding?: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * One of `'channelsLast'` (default) and `'channelsFirst'`.\n   *\n   * The ordering of the dimensions in the inputs.\n   * `channelsLast` corresponds to inputs with shape\n   * `[batch, height, width, channels]` while `channelsFirst`\n   * corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class ZeroPadding2D extends Layer {\n  /** @nocollapse */\n  static className = 'ZeroPadding2D';\n  readonly dataFormat: DataFormat;\n  readonly padding: [[number, number], [number, number]];\n\n  constructor(args?: ZeroPadding2DLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.dataFormat =\n        args.dataFormat == null ? imageDataFormat() : args.dataFormat;\n    // TODO(cais): Maybe refactor the following logic surrounding `padding`\n    //   into a helper method.\n    if (args.padding == null) {\n      this.padding = [[1, 1], [1, 1]];\n    } else if (typeof args.padding === 'number') {\n      this.padding =\n          [[args.padding, args.padding], [args.padding, args.padding]];\n    } else {\n      args.padding = args.padding;\n      if (args.padding.length !== 2) {\n        throw new ValueError(\n            `ZeroPadding2D expects padding to be a length-2 array, but ` +\n            `received a length-${args.padding.length} array.`);\n      }\n\n      let heightPadding: [number, number];\n      let widthPadding: [number, number];\n      if (typeof args.padding[0] === 'number') {\n        heightPadding = [args.padding[0], args.padding[0]];\n        widthPadding = [args.padding[1] as number, args.padding[1] as number];\n      } else {\n        args.padding = args.padding as [[number, number], [number, number]];\n\n        if (args.padding[0].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects height padding to be a length-2 array, ` +\n              `but received a length-${args.padding[0].length} array.`);\n        }\n        heightPadding = args.padding[0] as [number, number];\n\n        if (args.padding[1].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects width padding to be a length-2 array, ` +\n              `but received a length-${args.padding[1].length} array.`);\n        }\n        widthPadding = args.padding[1] as [number, number];\n      }\n      this.padding = [heightPadding, widthPadding];\n    }\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n\n    let rows: number;\n    let cols: number;\n    if (this.dataFormat === 'channelsFirst') {\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        rows = inputShape[2] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[3] != null && inputShape[3] >= 0) {\n        cols = inputShape[3] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      if (inputShape[1] != null && inputShape[1] >= 0) {\n        rows = inputShape[1] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        cols = inputShape[2] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(\n        () => spatial2dPadding(\n            getExactlyOneTensor(inputs), this.padding, this.dataFormat));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ZeroPadding2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Pooling Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor3D, Tensor4D, Tensor5D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode, checkPoolMode} from '../common';\nimport {InputSpec} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, PaddingMode, PoolMode, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\nimport {preprocessConv2DInput, preprocessConv3DInput} from './convolutional';\n\n/**\n * 2D pooling.\n * @param x\n * @param poolSize\n * @param stridesdes strides. Defaults to [1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 2D pooling.\n */\nexport function pool2d(\n    x: Tensor, poolSize: [number, number], strides?: [number, number],\n    padding?: PaddingMode, dataFormat?: DataFormat,\n    poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // TODO(cais): Remove the preprocessing step once deeplearn.js supports\n    // dataFormat as an input argument.\n    x = preprocessConv2DInput(x, dataFormat);  // x is NHWC after preprocessing.\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      // TODO(cais): Rank check?\n      y = tfc.maxPool(x as Tensor4D, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      // TODO(cais): Check the dtype and rank of x and give clear error message\n      //   if those are incorrect.\n      y = tfc.avgPool(\n          // TODO(cais): Rank check?\n          x as Tensor3D | Tensor4D, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);  // NHWC -> NCHW.\n    }\n    return y;\n  });\n}\n\n/**\n * 3D pooling.\n * @param x\n * @param poolSize. Default to [1, 1, 1].\n * @param strides strides. Defaults to [1, 1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 3D pooling.\n */\nexport function pool3d(\n    x: Tensor5D, poolSize: [number, number, number],\n    strides?: [number, number, number], padding?: PaddingMode,\n    dataFormat?: DataFormat, poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // x is NDHWC after preprocessing.\n    x = preprocessConv3DInput(x as Tensor, dataFormat) as Tensor5D;\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      y = tfc.maxPool3d(x, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      y = tfc.avgPool3d(x, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);  // NDHWC -> NCDHW.\n    }\n    return y;\n  });\n}\n\nexport declare interface Pooling1DLayerArgs extends LayerArgs {\n  /**\n   * Size of the window to pool over, should be an integer.\n   */\n  poolSize?: number|[number];\n  /**\n   * Period at which to sample the pooled values.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number];\n  /** How to fill in data that's not an integer multiple of poolSize. */\n  padding?: PaddingMode;\n}\n\n/**\n * Abstract class for different pooling 1D layers.\n */\nexport abstract class Pooling1D extends Layer {\n  protected readonly poolSize: [number];\n  protected readonly strides: [number];\n  protected readonly padding: PaddingMode;\n\n  /**\n   *\n   * @param args Parameters for the Pooling layer.\n   *\n   * config.poolSize defaults to 2.\n   */\n  constructor(args: Pooling1DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = 2;\n    }\n    super(args);\n    if (typeof args.poolSize === 'number') {\n      this.poolSize = [args.poolSize];\n    } else if (\n        Array.isArray(args.poolSize) &&\n        (args.poolSize as number[]).length === 1 &&\n        typeof (args.poolSize as number[])[0] === 'number') {\n      this.poolSize = args.poolSize;\n    } else {\n      throw new ValueError(\n          `poolSize for 1D convolutional layer must be a number or an ` +\n          `Array of a single number, but received ` +\n          `${JSON.stringify(args.poolSize)}`);\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else {\n      if (typeof args.strides === 'number') {\n        this.strides = [args.strides];\n      } else if (\n          Array.isArray(args.strides) &&\n          (args.strides as number[]).length === 1 &&\n          typeof (args.strides as number[])[0] === 'number') {\n        this.strides = args.strides;\n      } else {\n        throw new ValueError(\n            `strides for 1D convolutional layer must be a number or an ` +\n            `Array of a single number, but received ` +\n            `${JSON.stringify(args.strides)}`);\n      }\n    }\n    assertPositiveInteger(this.strides, 'strides');\n\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const length = convOutputLength(\n        inputShape[1], this.poolSize[0], this.padding, this.strides[0]);\n    return [inputShape[0], length, inputShape[2]];\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Add dummy last dimension.\n      inputs = K.expandDims(getExactlyOneTensor(inputs), 2);\n      const output = this.poolingFunction(\n          getExactlyOneTensor(inputs), [this.poolSize[0], 1],\n          [this.strides[0], 1], this.padding, 'channelsLast');\n      // Remove dummy last dimension.\n      return tfc.squeeze(output, [2]);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'MaxPooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling1D);\n\nexport class AveragePooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'AveragePooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling1D);\n\nexport declare interface Pooling2DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [vertical, horizontal].\n   * Expects an integer or an array of 2 integers.\n   *\n   * For example, `[2, 2]` will halve the input in both spatial dimension.\n   * If only one integer is specified, the same window length\n   * will be used for both dimensions.\n   */\n  poolSize?: number|[number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 2 integers. Integer, tuple of 2 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 2D layers.\n */\nexport abstract class Pooling2D extends Layer {\n  protected readonly poolSize: [number, number];\n  protected readonly strides: [number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling2DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 2) {\n        throw new ValueError(\n            `If the strides property of a 2D pooling layer is an Array, ` +\n            `it is expected to have a length of 2, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    rows =\n        convOutputLength(rows, this.poolSize[0], this.padding, this.strides[0]);\n    cols =\n        convOutputLength(cols, this.poolSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'MaxPooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling2D);\n\nexport class AveragePooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'AveragePooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling2D);\n\nexport declare interface Pooling3DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [depth, height, width].\n   * Expects an integer or an array of 3 integers.\n   *\n   * For example, `[2, 2, 2]` will halve the input in three dimensions.\n   * If only one integer is specified, the same window length\n   * will be used for all dimensions.\n   */\n  poolSize?: number|[number, number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 3 integers. Integer, tuple of 3 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 3D layers.\n */\nexport abstract class Pooling3D extends Layer {\n  protected readonly poolSize: [number, number, number];\n  protected readonly strides: [number, number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling3DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 3) {\n        throw new ValueError(\n            `If the strides property of a 3D pooling layer is an Array, ` +\n            `it is expected to have a length of 3, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let depths =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[4] : inputShape[3];\n    depths = convOutputLength(\n        depths, this.poolSize[0], this.padding, this.strides[0]);\n    rows =\n        convOutputLength(rows, this.poolSize[1], this.padding, this.strides[1]);\n    cols =\n        convOutputLength(cols, this.poolSize[2], this.padding, this.strides[2]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], depths, rows, cols];\n    } else {\n      return [inputShape[0], depths, rows, cols, inputShape[4]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'MaxPooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling3D);\n\nexport class AveragePooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'AveragePooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling3D);\n\n/**\n * Abstract class for different global pooling 1D layers.\n */\nexport abstract class GlobalPooling1D extends Layer {\n  constructor(args: LayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], inputShape[2]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n}\n\nexport class GlobalAveragePooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling1D';\n  constructor(args?: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.mean(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling1D);\n\nexport class GlobalMaxPooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling1D';\n  constructor(args: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.max(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling1D);\n\nexport declare interface GlobalPooling2DLayerArgs extends LayerArgs {\n  /**\n   * One of `CHANNEL_LAST` (default) or `CHANNEL_FIRST`.\n   *\n   * The ordering of the dimensions in the inputs. `CHANNEL_LAST` corresponds\n   * to inputs with shape `[batch, height, width, channels[` while\n   * `CHANNEL_FIRST` corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different global pooling 2D layers.\n */\nexport abstract class GlobalPooling2D extends Layer {\n  protected dataFormat: DataFormat;\n  constructor(args: GlobalPooling2DLayerArgs) {\n    super(args);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape;\n    if (this.dataFormat === 'channelsLast') {\n      return [inputShape[0], inputShape[3]];\n    } else {\n      return [inputShape[0], inputShape[1]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class GlobalAveragePooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.mean(input, [1, 2]);\n      } else {\n        return tfc.mean(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling2D);\n\nexport class GlobalMaxPooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.max(input, [1, 2]);\n      } else {\n        return tfc.max(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Layers that augment the functionality of a base layer.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {InputSpec, Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {BidirectionalMergeMode, Shape, VALID_BIDIRECTIONAL_MERGE_MODES} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {RegularizerFn, RnnStepFunction} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {rnn, RNN, standardizeArgs} from './recurrent';\nimport {deserialize} from './serialization';\n\nexport declare interface WrapperLayerArgs extends LayerArgs {\n  /**\n   * The layer to be wrapped.\n   */\n  layer: Layer;\n}\n\n/**\n * Abstract wrapper base class.\n *\n * Wrappers take another layer and augment it in various ways.\n * Do not use this class as a layer, it is only an abstract base class.\n * Two usable wrappers are the `TimeDistributed` and `Bidirectional` wrappers.\n */\nexport abstract class Wrapper extends Layer {\n  readonly layer: Layer;\n\n  constructor(args: WrapperLayerArgs) {\n    // Porting Note: In PyKeras, `self.layer` is set prior to the calling\n    //   `super()`. But we can't do that here due to TypeScript's restriction.\n    //   See: https://github.com/Microsoft/TypeScript/issues/8277\n    //   As a result, we have to add checks in `get trainable()` and\n    //   `set trainable()` below in order to prevent using `this.layer` when\n    //   its value is `undefined`. The super constructor does use the getter\n    //   and the setter of `this.layer`.\n    super(args);\n    this.layer = args.layer;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    this.built = true;\n  }\n\n  // TODO(cais): Implement activityRegularizer getter.\n\n  get trainable(): boolean {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      return this.layer.trainable;\n    } else {\n      return false;\n    }\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      this.layer.trainable = value;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.layer.trainableWeights;\n  }\n  // TODO(cais): Implement setter for trainableWeights.\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.layer.nonTrainableWeights;\n  }\n  // TODO(cais): Implement setter for nonTrainableWeights.\n\n  get updates(): Tensor[] {\n    // tslint:disable-next-line:no-any\n    return (this.layer as any)._updates;\n  }\n\n  // TODO(cais): Implement getUpdatesFor().\n\n  get losses(): RegularizerFn[] {\n    return this.layer.losses;\n  }\n\n  // TODO(cais): Implement getLossesFor().\n\n  getWeights(): Tensor[] {\n    return this.layer.getWeights();\n  }\n\n  setWeights(weights: Tensor[]): void {\n    this.layer.setWeights(weights);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'layer': {\n        'className': this.layer.getClassName(),\n        'config': this.layer.getConfig(),\n      }\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.layer != null) {\n      this.layer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const layerConfig = config['layer'] as serialization.ConfigDict;\n    const layer = deserialize(layerConfig, customObjects) as Layer;\n    delete config['layer'];\n    const newConfig = {layer};\n    Object.assign(newConfig, config);\n    return new cls(newConfig);\n  }\n}\n\nexport class TimeDistributed extends Wrapper {\n  /** @nocollapse */\n  static className = 'TimeDistributed';\n  constructor(args: WrapperLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 3) {\n      throw new ValueError(\n          `TimeDistributed layer expects an input shape >= 3D, but received ` +\n          `input shape ${JSON.stringify(inputShape)}`);\n    }\n    this.inputSpec = [{shape: inputShape}];\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (!this.layer.built) {\n      this.layer.build(childInputShape);\n      this.layer.built = true;\n    }\n    super.build(inputShape);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    const childOutputShape =\n        this.layer.computeOutputShape(childInputShape) as Shape;\n    const timesteps = inputShape[1];\n    return [childOutputShape[0], timesteps].concat(childOutputShape.slice(1));\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      // TODO(cais): Add 'training' and 'useLearningPhase' to kwargs.\n      inputs = getExactlyOneTensor(inputs);\n      // Porting Note: In tfjs-layers, `inputs` are always concrete tensor\n      // values. Hence the inputs can't have an undetermined first (batch)\n      // dimension, which is why we always use the K.rnn approach here.\n      const step: RnnStepFunction = (inputs: Tensor, states: Tensor[]) => {\n        // TODO(cais): Add useLearningPhase.\n        // NOTE(cais): `layer.call` may return a length-1 array of Tensor in\n        //   some cases (e.g., `layer` is a `Sequential` instance), which is\n        //   why `getExactlyOneTensor` is used below.\n        const output = getExactlyOneTensor(this.layer.call(inputs, kwargs));\n        return [output, []];\n      };\n      const rnnOutputs =\n          rnn(step, inputs, [], false /* goBackwards */, null /* mask */,\n              null /* constants */, false /* unroll */,\n              true /* needPerStepOutputs */);\n      const y = rnnOutputs[1];\n      // TODO(cais): Add activity regularization.\n      // TODO(cais): Add useLearningPhase.\n      return y;\n    });\n  }\n\n  // TODO(cais): Implement detailed computeMask() logic.\n}\nserialization.registerClass(TimeDistributed);\n\nexport function checkBidirectionalMergeMode(value?: string): void {\n  generic_utils.checkStringTypeUnionValue(\n      VALID_BIDIRECTIONAL_MERGE_MODES, 'BidirectionalMergeMode', value);\n}\n\nexport declare interface BidirectionalLayerArgs extends WrapperLayerArgs {\n  /**\n   * The instance of an `RNN` layer to be wrapped.\n   */\n  layer: RNN;\n\n  /**\n   * Mode by which outputs of the forward and backward RNNs are\n   * combined. If `null` or `undefined`, the output will not be\n   * combined, they will be returned as an `Array`.\n   *\n   * If `undefined` (i.e., not provided), defaults to `'concat'`.\n   */\n  mergeMode?: BidirectionalMergeMode;\n}\n\nconst DEFAULT_BIDIRECTIONAL_MERGE_MODE: BidirectionalMergeMode = 'concat';\n\nexport class Bidirectional extends Wrapper {\n  /** @nocollapse */\n  static className = 'Bidirectional';\n  mergeMode: BidirectionalMergeMode;\n  private forwardLayer: RNN;\n  private backwardLayer: RNN;\n  private returnSequences: boolean;\n  private returnState: boolean;\n  private numConstants?: number;\n  private _trainable: boolean;\n\n  constructor(args: BidirectionalLayerArgs) {\n    super(args);\n\n    // Note: When creating `this.forwardLayer`, the original Layer object\n    //   (`config.layer`) ought to be cloned. This is why we call\n    //   `getConfig()` followed by `deserialize()`. Without this cloning,\n    //   the layer names saved during serialization will incorrectly contain\n    //   the 'forward_' prefix. In Python Keras, this is done using\n    //   `copy.copy` (shallow copy), which does not have a simple equivalent\n    //   in JavaScript. JavaScript's `Object.assign()` does not copy\n    //   methods.\n    const layerConfig = args.layer.getConfig();\n    const forwDict: serialization.ConfigDict = {};\n    forwDict['className'] = args.layer.getClassName();\n    forwDict['config'] = layerConfig;\n    this.forwardLayer = deserialize(forwDict) as RNN;\n    layerConfig['goBackwards'] =\n        layerConfig['goBackwards'] === true ? false : true;\n    const backDict: serialization.ConfigDict = {};\n    backDict['className'] = args.layer.getClassName();\n    backDict['config'] = layerConfig;\n    this.backwardLayer = deserialize(backDict) as RNN;\n    this.forwardLayer.name = 'forward_' + this.forwardLayer.name;\n    this.backwardLayer.name = 'backward_' + this.backwardLayer.name;\n\n    this.mergeMode = args.mergeMode === undefined ?\n        DEFAULT_BIDIRECTIONAL_MERGE_MODE :\n        args.mergeMode;\n    checkBidirectionalMergeMode(this.mergeMode);\n    if (args.weights) {\n      throw new NotImplementedError(\n          'weights support is not implemented for Bidirectional layer yet.');\n    }\n    this._stateful = args.layer.stateful;\n    this.returnSequences = args.layer.returnSequences;\n    this.returnState = args.layer.returnState;\n    this.supportsMasking = true;\n    this._trainable = true;\n    this.inputSpec = args.layer.inputSpec;\n    this.numConstants = null;\n  }\n\n  get trainable(): boolean {\n    return this._trainable;\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    this._trainable = value;\n    if (this.forwardLayer != null) {\n      this.forwardLayer.trainable = value;\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.trainable = value;\n    }\n  }\n\n  getWeights(): Tensor[] {\n    return this.forwardLayer.getWeights().concat(\n        this.backwardLayer.getWeights());\n  }\n\n  setWeights(weights: Tensor[]): void {\n    const numWeights = weights.length;\n    const numeightsOver2 = Math.floor(numWeights / 2);\n    this.forwardLayer.setWeights(weights.slice(0, numeightsOver2));\n    this.backwardLayer.setWeights(weights.slice(numeightsOver2));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    let layerShapes: Shape|Shape[] =\n        this.forwardLayer.computeOutputShape(inputShape);\n    if (!(Array.isArray(layerShapes) && Array.isArray(layerShapes[0]))) {\n      layerShapes = [layerShapes as Shape];\n    }\n    layerShapes = layerShapes as Shape[];\n\n    let outputShape: Shape;\n    let outputShapes: Shape[];\n    let stateShape: Shape[];\n    if (this.returnState) {\n      stateShape = layerShapes.slice(1);\n      outputShape = layerShapes[0];\n    } else {\n      outputShape = layerShapes[0];\n    }\n    outputShape = outputShape;\n    if (this.mergeMode === 'concat') {\n      outputShape[outputShape.length - 1] *= 2;\n      outputShapes = [outputShape];\n    } else if (this.mergeMode == null) {\n      outputShapes = [outputShape, outputShape.slice()];\n    } else {\n      outputShapes = [outputShape];\n    }\n\n    if (this.returnState) {\n      if (this.mergeMode == null) {\n        return outputShapes.concat(stateShape).concat(stateShape.slice());\n      }\n      return [outputShape].concat(stateShape).concat(stateShape.slice());\n    }\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    if (Array.isArray(inputs)) {\n      initialState = (inputs as Tensor[] | SymbolicTensor[]).slice(1);\n      inputs = (inputs as Tensor[] | SymbolicTensor[])[0];\n    }\n\n    if ((initialState == null || initialState.length === 0) &&\n        constants == null) {\n      return super.apply(inputs, kwargs);\n    }\n    const additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    const additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      const numStates = initialState.length;\n      if (numStates % 2 > 0) {\n        throw new ValueError(\n            'When passing `initialState` to a Bidrectional RNN, ' +\n            'the state should be an Array containing the states of ' +\n            'the underlying RNNs.');\n      }\n      kwargs['initialState'] = initialState;\n      additionalInputs.push(...initialState);\n      const stateSpecs = (initialState as Array<Tensor|SymbolicTensor>)\n                             .map(state => new InputSpec({shape: state.shape}));\n      this.forwardLayer.stateSpec = stateSpecs.slice(0, numStates / 2);\n      this.backwardLayer.stateSpec = stateSpecs.slice(numStates / 2);\n      additionalSpecs.push(...stateSpecs);\n    }\n    if (constants != null) {\n      throw new NotImplementedError(\n          'Support for constants in Bidirectional layers is not ' +\n          'implemented yet.');\n    }\n\n    const isSymbolicTensor = additionalInputs[0] instanceof SymbolicTensor;\n    for (const tensor of additionalInputs) {\n      if (tensor instanceof SymbolicTensor !== isSymbolicTensor) {\n        throw new ValueError(\n            'The initial state of a Bidirectional layer cannot be ' +\n            'specified as a mix of symbolic and non-symbolic tensors');\n      }\n    }\n\n    if (isSymbolicTensor) {\n      // Compute the full input and specs, including the states.\n      const fullInput = [inputs].concat(additionalInputs);\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call temporarily and replace inputSpec.\n      // Note: with initial states symbolic calls and non-symbolic calls to\n      // this method differ in how the initial states are passed. For\n      // symbolic calls, the initial states are passed in the first arg, as\n      // an Array of SymbolicTensors; for non-symbolic calls, they are\n      // passed in the second arg as a part of the kwargs. Hence the need to\n      // temporarily modify inputSpec here.\n      // TODO(cais): Make refactoring so that this hacky code below is no\n      // longer needed.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output =\n          super.apply(fullInput as Tensor[] | SymbolicTensor[], kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const initialState = kwargs['initialState'];\n\n      let y: Tensor|Tensor[];\n      let yRev: Tensor|Tensor[];\n      if (initialState == null) {\n        y = this.forwardLayer.call(inputs, kwargs);\n        yRev = this.backwardLayer.call(inputs, kwargs);\n      } else {\n        const forwardState = initialState.slice(0, initialState.length / 2);\n        const backwardState = initialState.slice(initialState.length / 2);\n        y = this.forwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: forwardState}));\n        yRev = this.backwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: backwardState}));\n      }\n\n      let states: Tensor[];\n      if (this.returnState) {\n        if (Array.isArray(y)) {\n          states = y.slice(1).concat((yRev as Tensor[]).slice(1));\n        } else {\n        }\n        y = (y as Tensor[])[0];\n        yRev = (yRev as Tensor[])[0];\n      }\n\n      if (this.returnSequences) {\n        yRev = tfc.reverse(yRev as Tensor, 1);\n      }\n\n      let output: Tensor|Tensor[];\n      if (this.mergeMode === 'concat') {\n        output = K.concatenate([y as Tensor, yRev as Tensor]);\n      } else if (this.mergeMode === 'sum') {\n        output = tfc.add(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode === 'ave') {\n        output = tfc.mul(.5, tfc.add(y as Tensor, yRev as Tensor));\n      } else if (this.mergeMode === 'mul') {\n        output = tfc.mul(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode == null) {\n        output = [y as Tensor, yRev as Tensor];\n      }\n\n      // TODO(cais): Properly set learning phase.\n      if (this.returnState) {\n        if (this.mergeMode == null) {\n          return (output as Tensor[]).concat(states);\n        }\n        return [output as Tensor].concat(states);\n      }\n      return output;\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[]): void {\n    this.forwardLayer.resetStates();\n    this.backwardLayer.resetStates();\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    nameScope(this.forwardLayer.name, () => {\n      this.forwardLayer.build(inputShape);\n    });\n    nameScope(this.backwardLayer.name, () => {\n      this.backwardLayer.build(inputShape);\n    });\n    this.built = true;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (Array.isArray(mask)) {\n      mask = mask[0];\n    }\n    let outputMask: Tensor|Tensor[];\n    if (this.returnSequences) {\n      if (this.mergeMode == null) {\n        outputMask = [mask, mask];\n      } else {\n        outputMask = mask;\n      }\n    } else {\n      if (this.mergeMode == null) {\n        outputMask = [null, null];\n      } else {\n        outputMask = null;\n      }\n    }\n    if (this.returnState) {\n      const states = this.forwardLayer.states;\n      const stateMask: Tensor[] = states.map(state => null);\n      if (Array.isArray(outputMask)) {\n        return outputMask.concat(stateMask).concat(stateMask);\n      } else {\n        return [outputMask].concat(stateMask).concat(stateMask);\n      }\n    } else {\n      return outputMask;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.forwardLayer.trainableWeights.concat(\n        this.backwardLayer.trainableWeights);\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.forwardLayer.nonTrainableWeights.concat(\n        this.backwardLayer.nonTrainableWeights);\n  }\n\n  // TODO(cais): Implement constraints().\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.forwardLayer != null) {\n      this.forwardLayer.setFastWeightInitDuringBuild(value);\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'mergeMode': this.mergeMode,\n    };\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    const rnnLayer =\n        deserialize(config['layer'] as serialization.ConfigDict) as RNN;\n    delete config['layer'];\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    if (config['numConstants'] != null) {\n      throw new NotImplementedError(\n          `Deserialization of a Bidirectional layer with numConstants ` +\n          `present is not supported yet.`);\n    }\n    // tslint:disable-next-line:no-any\n    const newConfig: {[key: string]: any} = config;\n    newConfig['layer'] = rnnLayer;\n    return new cls(newConfig);\n  }\n}\nserialization.registerClass(Bidirectional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {InputLayer, InputLayerArgs} from './engine/input_layer';\nimport {Layer, LayerArgs} from './engine/topology';\nimport {input} from './exports';\nimport {ELU, ELULayerArgs, LeakyReLU, LeakyReLULayerArgs, PReLU, PReLULayerArgs, ReLU, ReLULayerArgs, Softmax, SoftmaxLayerArgs, ThresholdedReLU, ThresholdedReLULayerArgs} from './layers/advanced_activations';\nimport {Conv1D, Conv2D, Conv2DTranspose, Conv3D, ConvLayerArgs, Cropping2D, Cropping2DLayerArgs, SeparableConv2D, SeparableConvLayerArgs, UpSampling2D, UpSampling2DLayerArgs} from './layers/convolutional';\nimport {DepthwiseConv2D, DepthwiseConv2DLayerArgs} from './layers/convolutional_depthwise';\nimport {ConvLSTM2D, ConvLSTM2DArgs, ConvLSTM2DCell, ConvLSTM2DCellArgs} from './layers/convolutional_recurrent';\nimport {Activation, ActivationLayerArgs, Dense, DenseLayerArgs, Dropout, DropoutLayerArgs, Flatten, FlattenLayerArgs, Masking, MaskingArgs, Permute, PermuteLayerArgs, RepeatVector, RepeatVectorLayerArgs, Reshape, ReshapeLayerArgs, SpatialDropout1D, SpatialDropout1DLayerConfig} from './layers/core';\nimport {Embedding, EmbeddingLayerArgs} from './layers/embeddings';\nimport {Add, Average, Concatenate, ConcatenateLayerArgs, Dot, DotLayerArgs, Maximum, Minimum, Multiply} from './layers/merge';\nimport {AlphaDropout, AlphaDropoutArgs, GaussianDropout, GaussianDropoutArgs, GaussianNoise, GaussianNoiseArgs} from './layers/noise';\nimport {BatchNormalization, BatchNormalizationLayerArgs, LayerNormalization, LayerNormalizationLayerArgs} from './layers/normalization';\nimport {ZeroPadding2D, ZeroPadding2DLayerArgs} from './layers/padding';\nimport {AveragePooling1D, AveragePooling2D, AveragePooling3D, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalPooling2DLayerArgs, MaxPooling1D, MaxPooling2D, MaxPooling3D, Pooling1DLayerArgs, Pooling2DLayerArgs, Pooling3DLayerArgs} from './layers/pooling';\nimport {GRU, GRUCell, GRUCellLayerArgs, GRULayerArgs, LSTM, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNN, SimpleRNNCell, SimpleRNNCellLayerArgs, SimpleRNNLayerArgs, StackedRNNCells, StackedRNNCellsArgs} from './layers/recurrent';\nimport {Bidirectional, BidirectionalLayerArgs, TimeDistributed, WrapperLayerArgs} from './layers/wrappers';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// Input Layer.\n/**\n * An input layer is an entry point into a `tf.LayersModel`.\n *\n * `InputLayer` is generated automatically for `tf.Sequential`` models by\n * specifying the `inputshape` or `batchInputShape` for the first layer.  It\n * should not be specified explicitly. However, it can be useful sometimes,\n * e.g., when constructing a sequential model from a subset of another\n * sequential model's layers. Like the code snippet below shows.\n *\n * ```js\n * // Define a model which simply adds two inputs.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));\n * model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n * model1.summary();\n * model1.predict(tf.zeros([1, 4])).print();\n *\n * // Construct another model, reusing the second layer of `model1` while\n * // not using the first layer of `model1`. Note that you cannot add the second\n * // layer of `model` directly as the first layer of the new sequential model,\n * // because doing so will lead to an error related to the fact that the layer\n * // is not an input layer. Instead, you need to create an `inputLayer` and add\n * // it to the new sequential model before adding the reused layer.\n * const model2 = tf.sequential();\n * // Use an inputShape that matches the input shape of `model1`'s second\n * // layer.\n * model2.add(tf.layers.inputLayer({inputShape: [3]}));\n * model2.add(model1.layers[1]);\n * model2.summary();\n * model2.predict(tf.zeros([1, 3])).print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}\n */\nexport function inputLayer(args: InputLayerArgs): Layer {\n  return new InputLayer(args);\n}\n\n// Advanced Activation Layers.\n\n/**\n * Exponetial Linear Unit (ELU).\n *\n * It follows:\n * `f(x) =  alpha * (exp(x) - 1.) for x < 0`,\n * `f(x) = x for x >= 0`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Fast and Accurate Deep Network Learning by Exponential Linear Units\n * (ELUs)](https://arxiv.org/abs/1511.07289v1)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function elu(args?: ELULayerArgs): Layer {\n  return new ELU(args);\n}\n\n/**\n * Rectified Linear Unit activation function.\n *\n * Input shape:\n *   Arbitrary. Use the config field `inputShape` (Array of integers, does\n *   not include the sample axis) when using this layer as the first layer\n *   in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function reLU(args?: ReLULayerArgs): Layer {\n  return new ReLU(args);\n}\n\n/**\n * Leaky version of a rectified linear unit.\n *\n * It allows a small gradient when the unit is not active:\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function leakyReLU(args?: LeakyReLULayerArgs): Layer {\n  return new LeakyReLU(args);\n}\n\n/**\n * Parameterized version of a leaky rectified linear unit.\n *\n * It follows\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n * wherein `alpha` is a trainable weight.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function prelu(args?: PReLULayerArgs): Layer {\n  return new PReLU(args);\n}\n\n/**\n * Softmax activation layer.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function softmax(args?: SoftmaxLayerArgs): Layer {\n  return new Softmax(args);\n}\n\n/**\n * Thresholded Rectified Linear Unit.\n *\n * It follows:\n * `f(x) = x for x > theta`,\n * `f(x) = 0 otherwise`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Zero-Bias Autoencoders and the Benefits of Co-Adapting\n * Features](http://arxiv.org/abs/1402.3337)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function thresholdedReLU(args?: ThresholdedReLULayerArgs): Layer {\n  return new ThresholdedReLU(args);\n}\n\n// Convolutional Layers.\n\n/**\n * 1D convolution layer (e.g., temporal convolution).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input over a single spatial (or temporal) dimension\n * to produce a tensor of outputs.\n *\n * If `use_bias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model, provide an\n * `inputShape` argument `Array` or `null`.\n *\n * For example, `inputShape` would be:\n * - `[10, 128]` for sequences of 10 vectors of 128-dimensional vectors\n * - `[null, 128]` for variable-length sequences of 128-dimensional vectors.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional',  namespace: 'layers'}\n */\nexport function conv1d(args: ConvLayerArgs): Layer {\n  return new Conv1D(args);\n}\n\n/**\n * 2D convolution layer (e.g. spatial convolution over images).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 3]` for 128x128 RGB pictures\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2d(args: ConvLayerArgs): Layer {\n  return new Conv2D(args);\n}\n\n/**\n * Transposed convolutional layer (sometimes called Deconvolution).\n *\n * The need for transposed convolutions generally arises\n * from the desire to use a transformation going in the opposite direction of\n * a normal convolution, i.e., from something that has the shape of the output\n * of some convolution to something that has the shape of its input while\n * maintaining a connectivity pattern that is compatible with said\n * convolution.\n *\n * When using this layer as the first layer in a model, provide the\n * configuration `inputShape` (`Array` of integers, does not include the\n * sample axis), e.g., `inputShape: [128, 128, 3]` for 128x128 RGB pictures in\n * `dataFormat: 'channelsLast'`.\n *\n * Input shape:\n *   4D tensor with shape:\n *   `[batch, channels, rows, cols]` if `dataFormat` is `'channelsFirst'`.\n *   or 4D tensor with shape\n *   `[batch, rows, cols, channels]` if `dataFormat` is `'channelsLast`.\n *\n * Output shape:\n *   4D tensor with shape:\n *   `[batch, filters, newRows, newCols]` if `dataFormat` is\n * `'channelsFirst'`. or 4D tensor with shape:\n *   `[batch, newRows, newCols, filters]` if `dataFormat` is `'channelsLast'`.\n *\n * References:\n *   - [A guide to convolution arithmetic for deep\n * learning](https://arxiv.org/abs/1603.07285v1)\n *   - [Deconvolutional\n * Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2dTranspose(args: ConvLayerArgs): Layer {\n  return new Conv2DTranspose(args);\n}\n\n/**\n * 3D convolution layer (e.g. spatial convolution over volumes).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 128, 1]` for 128x128x128 grayscale volumes\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv3d(args: ConvLayerArgs): Layer {\n  return new Conv3D(args);\n}\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Separable convolution consists of first performing\n * a depthwise spatial convolution\n * (which acts on each input channel separately)\n * followed by a pointwise convolution which mixes together the resulting\n * output channels. The `depthMultiplier` argument controls how many\n * output channels are generated per input channel in the depthwise step.\n *\n * Intuitively, separable convolutions can be understood as\n * a way to factorize a convolution kernel into two smaller kernels,\n * or as an extreme version of an Inception block.\n *\n * Input shape:\n *   4D tensor with shape:\n *     `[batch, channels, rows, cols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, rows, cols, channels]` if data_format='channelsLast'.\n *\n * Output shape:\n *   4D tensor with shape:\n *     `[batch, filters, newRows, newCols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, newRows, newCols, filters]` if data_format='channelsLast'.\n *     `rows` and `cols` values might have changed due to padding.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function separableConv2d(args: SeparableConvLayerArgs): Layer {\n  return new SeparableConv2D(args);\n}\n\n/**\n * Cropping layer for 2D input (e.g., image).\n *\n * This layer can crop an input\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, croppedRows, croppedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, croppedRows, croppedCols]`.\n *\n * Examples\n * ```js\n *\n * const model = tf.sequential();\n * model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]],\n *                                inputShape: [128, 128, 3]}));\n * //now output shape is [batch, 124, 124, 3]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function cropping2D(args: Cropping2DLayerArgs): Layer {\n  return new Cropping2D(args);\n}\n\n/**\n * Upsampling layer for 2D inputs.\n *\n * Repeats the rows and columns of the data\n * by size[0] and size[1] respectively.\n *\n *\n * Input shape:\n *    4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *         `[batch, rows, cols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *        `[batch, channels, rows, cols]`\n *\n * Output shape:\n *     4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *        `[batch, upsampledRows, upsampledCols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *         `[batch, channels, upsampledRows, upsampledCols]`\n *\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function upSampling2d(args: UpSampling2DLayerArgs): Layer {\n  return new UpSampling2D(args);\n}\n\n// Convolutional(depthwise) Layers.\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Depthwise Separable convolutions consists in performing just the first step\n * in a depthwise spatial convolution (which acts on each input channel\n * separately). The `depthMultplier` argument controls how many output channels\n * are generated per input channel in the depthwise step.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function depthwiseConv2d(args: DepthwiseConv2DLayerArgs): Layer {\n  return new DepthwiseConv2D(args);\n}\n\n// Basic Layers.\n\n/**\n * Applies an activation function to an output.\n *\n * This layer applies element-wise activation function.  Other layers, notably\n * `dense` can also apply activation functions.  Use this isolated activation\n * function to extract the values before and after the\n * activation. For instance:\n *\n * ```js\n * const input = tf.input({shape: [5]});\n * const denseLayer = tf.layers.dense({units: 1});\n * const activationLayer = tf.layers.activation({activation: 'relu6'});\n *\n * // Obtain the output symbolic tensors by applying the layers in order.\n * const denseOutput = denseLayer.apply(input);\n * const activationOutput = activationLayer.apply(denseOutput);\n *\n * // Create the model based on the inputs.\n * const model = tf.model({\n *     inputs: input,\n *     outputs: [denseOutput, activationOutput]\n * });\n *\n * // Collect both outputs and print separately.\n * const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));\n * denseOut.print();\n * activationOut.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function activation(args: ActivationLayerArgs): Layer {\n  return new Activation(args);\n}\n\n/**\n * Creates a dense (fully connected) layer.\n *\n * This layer implements the operation:\n *   `output = activation(dot(input, kernel) + bias)`\n *\n * `activation` is the element-wise activation function\n *   passed as the `activation` argument.\n *\n * `kernel` is a weights matrix created by the layer.\n *\n * `bias` is a bias vector created by the layer (only applicable if `useBias`\n * is `true`).\n *\n * **Input shape:**\n *\n *   nD `tf.Tensor` with shape: `(batchSize, ..., inputDim)`.\n *\n *   The most common situation would be\n *   a 2D input with shape `(batchSize, inputDim)`.\n *\n * **Output shape:**\n *\n *   nD tensor with shape: `(batchSize, ..., units)`.\n *\n *   For instance, for a 2D input with shape `(batchSize, inputDim)`,\n *   the output would have shape `(batchSize, units)`.\n *\n * Note: if the input to the layer has a rank greater than 2, then it is\n * flattened prior to the initial dot product with the kernel.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dense(args: DenseLayerArgs): Layer {\n  return new Dense(args);\n}\n\n/**\n * Applies\n * [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to\n * the input.\n *\n * Dropout consists in randomly setting a fraction `rate` of input units to 0 at\n * each update during training time, which helps prevent overfitting.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dropout(args: DropoutLayerArgs): Layer {\n  return new Dropout(args);\n}\n\n/**\n * Spatial 1D version of Dropout.\n *\n * This Layer type performs the same function as the Dropout layer, but it drops\n * entire 1D feature maps instead of individual elements. For example, if an\n * input example consists of 3 timesteps and the feature map for each timestep\n * has a size of 4, a `spatialDropout1d` layer may zero out the feature maps\n * of the 1st timesteps and 2nd timesteps completely while sparing all feature\n * elements of the 3rd timestep.\n *\n * If adjacent frames (timesteps) are strongly correlated (as is normally the\n * case in early convolution layers), regular dropout will not regularize the\n * activation and will otherwise just result in merely an effective learning\n * rate decrease. In this case, `spatialDropout1d` will help promote\n * independence among feature maps and should be used instead.\n *\n * **Arguments:**\n *   rate: A floating-point number >=0 and <=1. Fraction of the input elements\n *     to drop.\n *\n * **Input shape:**\n *   3D tensor with shape `(samples, timesteps, channels)`.\n *\n * **Output shape:**\n *   Same as the input shape.\n *\n * References:\n *   - [Efficient Object Localization Using Convolutional\n *      Networks](https://arxiv.org/abs/1411.4280)\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function spatialDropout1d(args: SpatialDropout1DLayerConfig): Layer {\n  return new SpatialDropout1D(args);\n}\n\n/**\n * Flattens the input. Does not affect the batch size.\n *\n * A `Flatten` layer flattens each batch in its inputs to 1D (making the output\n * 2D).\n *\n * For example:\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const flattenLayer = tf.layers.flatten();\n * // Inspect the inferred output shape of the flatten layer, which\n * // equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the\n * // flattening. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(flattenLayer.apply(input).shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function flatten(args?: FlattenLayerArgs): Layer {\n  return new Flatten(args);\n}\n\n/**\n * Repeats the input n times in a new dimension.\n *\n * ```js\n *  const model = tf.sequential();\n *  model.add(tf.layers.repeatVector({n: 4, inputShape: [2]}));\n *  const x = tf.tensor2d([[10, 20]]);\n *  // Use the model to do inference on a data point the model hasn't see\n *  model.predict(x).print();\n *  // output shape is now [batch, 2, 4]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function repeatVector(args: RepeatVectorLayerArgs): Layer {\n  return new RepeatVector(args);\n}\n\n/**\n * Reshapes an input to a certain shape.\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const reshapeLayer = tf.layers.reshape({targetShape: [2, 6]});\n * // Inspect the inferred output shape of the Reshape layer, which\n * // equals `[null, 2, 6]`. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(reshapeLayer.apply(input).shape));\n * ```\n *\n * Input shape:\n *   Arbitrary, although all dimensions in the input shape must be fixed.\n *   Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n *\n * Output shape:\n *   [batchSize, targetShape[0], targetShape[1], ...,\n *    targetShape[targetShape.length - 1]].\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function reshape(args: ReshapeLayerArgs): Layer {\n  return new Reshape(args);\n}\n\n/**\n * Permutes the dimensions of the input according to a given pattern.\n *\n * Useful for, e.g., connecting RNNs and convnets together.\n *\n * Example:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.permute({\n *   dims: [2, 1],\n *   inputShape: [10, 64]\n * }));\n * console.log(model.outputShape);\n * // Now model's output shape is [null, 64, 10], where null is the\n * // unpermuted sample (batch) dimension.\n * ```\n *\n * Input shape:\n *   Arbitrary. Use the configuration field `inputShape` when using this\n *   layer as the first layer in a model.\n *\n * Output shape:\n *   Same rank as the input shape, but with the dimensions re-ordered (i.e.,\n *   permuted) according to the `dims` configuration of this layer.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function permute(args: PermuteLayerArgs): Layer {\n  return new Permute(args);\n}\n\n/**\n * Maps positive integers (indices) into dense vectors of fixed size.\n * eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]\n *\n * **Input shape:** 2D tensor with shape: `[batchSize, sequenceLength]`.\n *\n * **Output shape:** 3D tensor with shape: `[batchSize, sequenceLength,\n * outputDim]`.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function embedding(args: EmbeddingLayerArgs): Layer {\n  return new Embedding(args);\n}\n\n// Merge Layers.\n\n/**\n * Layer that performs element-wise addition on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). The inputs are specified as an\n * `Array` when the `apply` method of the `Add` layer instance is called. For\n * example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const addLayer = tf.layers.add();\n * const sum = addLayer.apply([input1, input2]);\n * console.log(JSON.stringify(sum.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function add(args?: LayerArgs): Layer {\n  return new Add(args);\n}\n\n/**\n * Layer that performs element-wise averaging on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const averageLayer = tf.layers.average();\n * const average = averageLayer.apply([input1, input2]);\n * console.log(JSON.stringify(average.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function average(args?: LayerArgs): Layer {\n  return new Average(args);\n}\n\n/**\n * Layer that concatenates an `Array` of inputs.\n *\n * It takes a list of tensors, all of the same shape except for the\n * concatenation axis, and returns a single tensor, the concatenation\n * of all inputs. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 3]});\n * const concatLayer = tf.layers.concatenate();\n * const output = concatLayer.apply([input1, input2]);\n * console.log(JSON.stringify(output.shape));\n * // You get [null, 2, 5], with the first dimension as the undetermined batch\n * // dimension. The last dimension (5) is the result of concatenating the\n * // last dimensions of the inputs (2 and 3).\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function concatenate(args?: ConcatenateLayerArgs): Layer {\n  return new Concatenate(args);\n}\n\n/**\n * Layer that computes the element-wise maximum an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const maxLayer = tf.layers.maximum();\n * const max = maxLayer.apply([input1, input2]);\n * console.log(JSON.stringify(max.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function maximum(args?: LayerArgs): Layer {\n  return new Maximum(args);\n}\n\n/**\n * Layer that computes the element-wise minimum of an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const minLayer = tf.layers.minimum();\n * const min = minLayer.apply([input1, input2]);\n * console.log(JSON.stringify(min.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function minimum(args?: LayerArgs): Layer {\n  return new Minimum(args);\n}\n\n/**\n * Layer that multiplies (element-wise) an `Array` of inputs.\n *\n * It takes as input an Array of tensors, all of the same\n * shape, and returns a single tensor (also of the same shape).\n * For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const input3 = tf.input({shape: [2, 2]});\n * const multiplyLayer = tf.layers.multiply();\n * const product = multiplyLayer.apply([input1, input2, input3]);\n * console.log(product.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function multiply(args?: LayerArgs): Layer {\n  return new Multiply(args);\n}\n\n/**\n * Layer that computes a dot product between samples in two tensors.\n *\n * E.g., if applied to a list of two tensors `a` and `b` both of shape\n * `[batchSize, n]`, the output will be a tensor of shape `[batchSize, 1]`,\n * where each entry at index `[i, 0]` will be the dot product between\n * `a[i, :]` and `b[i, :]`.\n *\n * Example:\n *\n * ```js\n * const dotLayer = tf.layers.dot({axes: -1});\n * const x1 = tf.tensor2d([[10, 20], [30, 40]]);\n * const x2 = tf.tensor2d([[-1, -2], [-3, -4]]);\n *\n * // Invoke the layer's apply() method in eager (imperative) mode.\n * const y = dotLayer.apply([x1, x2]);\n * y.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function dot(args: DotLayerArgs): Layer {\n  return new Dot(args);\n}\n\n// Normalization Layers.\n\n/**\n * Batch normalization layer (Ioffe and Szegedy, 2014).\n *\n * Normalize the activations of the previous layer at each batch,\n * i.e. applies a transformation that maintains the mean activation\n * close to 0 and the activation standard deviation close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape` (Array of integers, does\n *   not include the sample axis) when calling the constructor of this class,\n *   if this layer is used as a first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Batch Normalization: Accelerating Deep Network Training by Reducing\n * Internal Covariate Shift](https://arxiv.org/abs/1502.03167)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function batchNormalization(args?: BatchNormalizationLayerArgs): Layer {\n  return new BatchNormalization(args);\n}\n\n/**\n * Layer-normalization layer (Ba et al., 2016).\n *\n * Normalizes the activations of the previous layer for each given example in a\n * batch independently, instead of across a batch like in `batchNormalization`.\n * In other words, this layer applies a transformation that maintanis the mean\n * activation within each example close to0 and activation variance close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the argument `inputShape` when using this layer as the first\n *   layer in a model.\n *\n * Output shape:\n *   Same as input.\n *\n * References:\n *   - [Layer Normalization](https://arxiv.org/abs/1607.06450)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function layerNormalization(args?: LayerNormalizationLayerArgs): Layer {\n  return new LayerNormalization(args);\n}\n\n// Padding Layers.\n\n/**\n * Zero-padding layer for 2D input (e.g., image).\n *\n * This layer can add rows and columns of zeros\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, paddedRows, paddedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, paddedRows, paddedCols]`.\n *\n * @doc {heading: 'Layers', subheading: 'Padding', namespace: 'layers'}\n */\nexport function zeroPadding2d(args?: ZeroPadding2DLayerArgs): Layer {\n  return new ZeroPadding2D(args);\n}\n\n// Pooling Layers.\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape: `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * `tf.avgPool1d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling1d(args: Pooling1DLayerArgs): Layer {\n  return new AveragePooling1D(args);\n}\nexport function avgPool1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape:\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, rows, cols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, pooleRows, pooledCols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, pooleRows, pooledCols]`\n *\n * `tf.avgPool2d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling2d(args: Pooling2DLayerArgs): Layer {\n  return new AveragePooling2D(args);\n}\nexport function avgPool2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n\n/**\n * Average pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      4D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling3d(args: Pooling3DLayerArgs): Layer {\n  return new AveragePooling3D(args);\n}\nexport function avgPool3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n\n/**\n * Global average pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling1d(args?: LayerArgs): Layer {\n  return new GlobalAveragePooling1D(args);\n}\n\n/**\n * Global average pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalAveragePooling2D(args);\n}\n\n/**\n * Global max pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling1d(args?: LayerArgs): Layer {\n  return new GlobalMaxPooling1D(args);\n}\n\n/**\n * Global max pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalMaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for temporal data.\n *\n * Input shape:  `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling1d(args: Pooling1DLayerArgs): Layer {\n  return new MaxPooling1D(args);\n}\n\n/**\n * Max pooling operation for spatial data.\n *\n * Input shape\n *   - If `dataFormat === CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, rows, cols, channels]`\n *   - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *       `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, pooleRows, pooledCols, channels]`\n *   - If `dataFormat=CHANNEL_FIRST`:\n *       4D tensor with shape:\n *       `[batchSize, channels, pooleRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling2d(args: Pooling2DLayerArgs): Layer {\n  return new MaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      5D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling3d(args: Pooling3DLayerArgs): Layer {\n  return new MaxPooling3D(args);\n}\n\n// Recurrent Layers.\n\n/**\n * Gated Recurrent Unit - Cho et al. 2014.\n *\n * This is an `RNN` layer consisting of one `GRUCell`. However, unlike\n * the underlying `GRUCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.gru({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `GRUCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gru(args: GRULayerArgs): Layer {\n  return new GRU(args);\n}\n\n/**\n * Cell class for `GRU`.\n *\n * `GRUCell` is distinct from the `RNN` subclass `GRU` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `GRU` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.gruCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `GRUCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.gruCell({units: 4}),\n *   tf.layers.gruCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `gruCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `GRUCell`, use the\n * `tf.layers.gru`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gruCell(args: GRUCellLayerArgs): RNNCell {\n  return new GRUCell(args);\n}\n\n/**\n * Long-Short Term Memory layer - Hochreiter 1997.\n *\n * This is an `RNN` layer consisting of one `LSTMCell`. However, unlike\n * the underlying `LSTMCell`, the `apply` method of `LSTM` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const lstm = tf.layers.lstm({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = lstm.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `LSTMCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstm(args: LSTMLayerArgs): Layer {\n  return new LSTM(args);\n}\n\n/**\n * Cell class for `LSTM`.\n *\n * `LSTMCell` is distinct from the `RNN` subclass `LSTM` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `LSTM` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.lstmCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `LSTMCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.lstmCell({units: 4}),\n *   tf.layers.lstmCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `lstmCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `LSTMCell`, use the\n * `tf.layers.lstm`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstmCell(args: LSTMCellLayerArgs): RNNCell {\n  return new LSTMCell(args);\n}\n\n/**\n * Fully-connected RNN where the output is to be fed back to input.\n *\n * This is an `RNN` layer consisting of one `SimpleRNNCell`. However, unlike\n * the underlying `SimpleRNNCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.simpleRNN({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `SimpleRNNCell`'s number of units.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNN(args: SimpleRNNLayerArgs): Layer {\n  return new SimpleRNN(args);\n}\n\n/**\n * Cell class for `SimpleRNN`.\n *\n * `SimpleRNNCell` is distinct from the `RNN` subclass `SimpleRNN` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `SimpleRNN` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.simpleRNNCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `SimpleRNNCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.simpleRNNCell({units: 4}),\n *   tf.layers.simpleRNNCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `SimpleRNNCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `SimpleRNNCell`, use the\n * `tf.layers.simpleRNN`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNNCell(args: SimpleRNNCellLayerArgs): RNNCell {\n  return new SimpleRNNCell(args);\n}\n\n/**\n * Convolutional LSTM layer - Xingjian Shi 2015.\n *\n * This is an `ConvRNN2D` layer consisting of one `ConvLSTM2DCell`. However,\n * unlike the underlying `ConvLSTM2DCell`, the `apply` method of `ConvLSTM2D`\n * operates on a sequence of inputs. The shape of the input (not including the\n * first, batch dimension) needs to be 4-D, with the first dimension being time\n * steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const batchSize = 4;\n * const sequenceLength = 2;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [batchSize, sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const layer = tf.layers.convLstm2d({filters, kernelSize});\n *\n * const output = layer.apply(input);\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2d(args: ConvLSTM2DArgs): ConvLSTM2D {\n  return new ConvLSTM2D(args);\n}\n\n/**\n * Cell class for `ConvLSTM2D`.\n *\n * `ConvLSTM2DCell` is distinct from the `ConvRNN2D` subclass `ConvLSTM2D` in\n * that its `call` method takes the input data of only a single time step and\n * returns the cell's output at the time step, while `ConvLSTM2D` takes the\n * input data over a number of time steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const sequenceLength = 1;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const cell = tf.layers.convLstm2dCell({filters, kernelSize});\n *\n * cell.build(input.shape);\n *\n * const outputSize = size - kernelSize + 1;\n * const outShape = [sequenceLength, outputSize, outputSize, filters];\n *\n * const initialH = tf.zeros(outShape);\n * const initialC = tf.zeros(outShape);\n *\n * const [o, h, c] = cell.call([input, initialH, initialC], {});\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2dCell(args: ConvLSTM2DCellArgs): ConvLSTM2DCell {\n  return new ConvLSTM2DCell(args);\n}\n\n/**\n * Base class for recurrent layers.\n *\n * Input shape:\n *   3D tensor with shape `[batchSize, timeSteps, inputDim]`.\n *\n * Output shape:\n *   - if `returnState`, an Array of tensors (i.e., `tf.Tensor`s). The first\n *     tensor is the output. The remaining tensors are the states at the\n *     last time step, each with shape `[batchSize, units]`.\n *   - if `returnSequences`, the output will have shape\n *     `[batchSize, timeSteps, units]`.\n *   - else, the output will have shape `[batchSize, units]`.\n *\n * Masking:\n *   This layer supports masking for input data with a variable number\n *   of timesteps. To introduce masks to your data,\n *   use an embedding layer with the `mask_zero` parameter\n *   set to `True`.\n *\n * Notes on using statefulness in RNNs:\n *   You can set RNN layers to be 'stateful', which means that the states\n *   computed for the samples in one batch will be reused as initial states\n *   for the samples in the next batch. This assumes a one-to-one mapping\n *   between samples in different successive batches.\n *\n *   To enable statefulness:\n *     - specify `stateful: true` in the layer constructor.\n *     - specify a fixed batch size for your model, by passing\n *       if sequential model:\n *         `batchInputShape=[...]` to the first layer in your model.\n *       else for functional model with 1 or more Input layers:\n *         `batchShape=[...]` to all the first layers in your model.\n *       This is the expected shape of your inputs *including the batch size*.\n *       It should be a tuple of integers, e.g. `(32, 10, 100)`.\n *     - specify `shuffle=False` when calling fit().\n *\n *   To reset the states of your model, call `.resetStates()` on either\n *   a specific layer, or on your entire model.\n *\n * Note on specifying the initial state of RNNs\n *   You can specify the initial state of RNN layers symbolically by\n *   calling them with the option `initialState`. The value of\n *   `initialState` should be a tensor or list of tensors representing\n *   the initial state of the RNN layer.\n *\n *   You can specify the initial state of RNN layers numerically by\n *   calling `resetStates` with the keyword argument `states`. The value of\n *   `states` should be a numpy array or list of numpy arrays representing\n *   the initial state of the RNN layer.\n *\n * Note on passing external constants to RNNs\n *   You can pass \"external\" constants to the cell using the `constants`\n *   keyword argument of `RNN.call` method. This requires that the `cell.call`\n *   method accepts the same keyword argument `constants`. Such constants\n *   can be used to conditon the cell transformation on additional static inputs\n *   (not changing over time), a.k.a an attention mechanism.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function rnn(args: RNNLayerArgs): Layer {\n  return new RNN(args);\n}\n\n/**\n * Wrapper allowing a stack of RNN cells to behave as a single cell.\n *\n * Used to implement efficient stacked RNNs.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function stackedRNNCells(args: StackedRNNCellsArgs): RNNCell {\n  return new StackedRNNCells(args);\n}\n\n// Wrapper Layers.\n\n/** @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'} */\nexport function bidirectional(args: BidirectionalLayerArgs): Bidirectional {\n  return new Bidirectional(args);\n}\n\n/**\n * This wrapper applies a layer to every temporal slice of an input.\n *\n * The input should be at least 3D,  and the dimension of the index `1` will be\n * considered to be the temporal dimension.\n *\n * Consider a batch of 32 samples, where each sample is a sequence of 10 vectors\n * of 16 dimensions. The batch input shape of the layer is then `[32,  10,\n * 16]`, and the `inputShape`, not including the sample dimension, is\n * `[10, 16]`.\n *\n * You can then use `TimeDistributed` to apply a `Dense` layer to each of the 10\n * timesteps, independently:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.dense({units: 8}),\n *   inputShape: [10, 16],\n * }));\n *\n * // Now model.outputShape = [null, 10, 8].\n * // The output will then have shape `[32, 10, 8]`.\n *\n * // In subsequent layers, there is no need for `inputShape`:\n * model.add(tf.layers.timeDistributed({layer: tf.layers.dense({units: 32})}));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * // Now model.outputShape = [null, 10, 32].\n * ```\n *\n * The output will then have shape `[32, 10, 32]`.\n *\n * `TimeDistributed` can be used with arbitrary layers, not just `Dense`, for\n * instance a `Conv2D` layer.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.conv2d({filters: 64, kernelSize: [3, 3]}),\n *   inputShape: [10, 299, 299, 3],\n * }));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}\n */\nexport function timeDistributed(args: WrapperLayerArgs): Layer {\n  return new TimeDistributed(args);\n}\n\n// Aliases for pooling.\nexport const globalMaxPool1d = globalMaxPooling1d;\nexport const globalMaxPool2d = globalMaxPooling2d;\nexport const maxPool1d = maxPooling1d;\nexport const maxPool2d = maxPooling2d;\n\nexport {Layer, RNN, RNNCell, input /* alias for tf.input */};\n\n/**\n * Apply additive zero-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * This is useful to mitigate overfitting\n * (you could see it as a form of random data augmentation).\n * Gaussian Noise (GS) is a natural choice as corruption process\n * for real valued inputs.\n *\n * # Arguments\n *     stddev: float, standard deviation of the noise distribution.\n *\n * # Input shape\n *         Arbitrary. Use the keyword argument `input_shape`\n *         (tuple of integers, does not include the samples axis)\n *         when using this layer as the first layer in a model.\n *\n * # Output shape\n *         Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianNoise(args: GaussianNoiseArgs): GaussianNoise {\n  return new GaussianNoise(args);\n}\n\n/**\n * Apply multiplicative 1-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting](\n *      http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianDropout(args: GaussianDropoutArgs): GaussianDropout {\n  return new GaussianDropout(args);\n}\n\n/**\n * Applies Alpha Dropout to the input.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\n * to their original values, in order to ensure the self-normalizing property\n * even after this dropout.\n * Alpha Dropout fits well to Scaled Exponential Linear Units\n * by randomly setting activations to the negative saturation value.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\n *     shape for randomly generated keep/drop flags.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function alphaDropout(args: AlphaDropoutArgs): AlphaDropout {\n  return new AlphaDropout(args);\n}\n\n/**\n * Masks a sequence by using a mask value to skip timesteps.\n *\n * If all features for a given sample timestep are equal to `mask_value`,\n * then the sample timestep will be masked (skipped) in all downstream layers\n * (as long as they support masking).\n *\n * If any downstream layer does not support masking yet receives such\n * an input mask, an exception will be raised.\n *\n * Arguments:\n *   - `maskValue`: Either None or mask value to skip.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Mask', namespace: 'layers'}\n */\nexport function masking(args?: MaskingArgs): Layer {\n  return new Masking(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport * as losses from './losses';\nimport * as metrics from './metrics';\n\n/**\n * Binary accuracy metric function.\n *\n * `yTrue` and `yPred` can have 0-1 values. Example:\n * ```js\n * const x = tf.tensor2d([[1, 1, 1, 1], [0, 0, 0, 0]], [2, 4]);\n * const y = tf.tensor2d([[1, 0, 1, 0], [0, 0, 0, 1]], [2, 4]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * `yTrue` and `yPred` can also have floating-number values between 0 and 1, in\n * which case the values will be thresholded at 0.5 to yield 0-1 values (i.e.,\n * a value >= 0.5 and <= 1.0 is interpreted as 1.\n * )\n * Example:\n * ```js\n * const x = tf.tensor1d([1, 1, 1, 1, 0, 0, 0, 0]);\n * const y = tf.tensor1d([0.2, 0.4, 0.6, 0.8, 0.2, 0.3, 0.4, 0.7]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryAccuracy(yTrue, yPred);\n}\n\n/**\n * Binary crossentropy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0], [1], [1], [1]]);\n * const y = tf.tensor2d([[0], [0], [0.5], [1]]);\n * const crossentropy = tf.metrics.binaryCrossentropy(x, y);\n * crossentropy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction, probabilities for the `1` case.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryCrossentropy(yTrue, yPred);\n}\n\n/**\n * Sparse categorical accuracy metric function.\n *\n * Example:\n * ```js\n *\n * const yTrue = tf.tensor1d([1, 1, 2, 2, 0]);\n * const yPred = tf.tensor2d(\n *      [[0, 1, 0], [1, 0, 0], [0, 0.4, 0.6], [0, 0.6, 0.4], [0.7, 0.3, 0]]);\n * const crossentropy = tf.metrics.sparseCategoricalAccuracy(yTrue, yPred);\n * crossentropy.print();\n * ```\n *\n * @param yTrue True labels: indices.\n * @param yPred Predicted probabilities or logits.\n * @returns Accuracy tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.sparseCategoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical accuracy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0, 0, 0, 1], [0, 0, 0, 1]]);\n * const y = tf.tensor2d([[0.1, 0.8, 0.05, 0.05], [0.1, 0.05, 0.05, 0.8]]);\n * const accuracy = tf.metrics.categoricalAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth: one-hot encoding of categories.\n * @param yPred Binary Tensor of prediction: probabilities or logits for the\n *   same categories as in `yTrue`.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical crossentropy between an output tensor and a target tensor.\n *\n * @param target A tensor of the same shape as `output`.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalCrossentropy(yTrue, yPred);\n}\n\n/**\n * Computes the precision of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const precision = tf.metrics.precision(x, y);\n * precision.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Precision Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.precision(yTrue, yPred);\n}\n\n/**\n * Computes the recall of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const recall = tf.metrics.recall(x, y);\n * recall.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Recall Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.recall(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Cosine proximity.\n *\n * Mathematically, cosine proximity is defined as:\n *   `-sum(l2Normalize(yTrue) * l2Normalize(yPred))`,\n * wherein `l2Normalize()` normalizes the L2 norm of the input to 1 and `*`\n * represents element-wise multiplication.\n *\n * ```js\n * const yTrue = tf.tensor2d([[1, 0], [1, 0]]);\n * const yPred = tf.tensor2d([[1 / Math.sqrt(2), 1 / Math.sqrt(2)], [0, 1]]);\n * const proximity = tf.metrics.cosineProximity(yTrue, yPred);\n * proximity.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Cosine proximity Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.cosineProximity(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute error.\n *\n * Mathematically, mean absolute error is defined as:\n *   `mean(abs(yPred - yTrue))`,\n * wherein the `mean` is applied over feature dimensions.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [0, 0], [2, 3]]);\n * const yPred = tf.tensor2d([[0, 1], [0, 1], [-2, -3]]);\n * const mse = tf.metrics.meanAbsoluteError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsoluteError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute percentage error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [10, 20]]);\n * const yPred = tf.tensor2d([[0, 1], [11, 24]]);\n * const mse = tf.metrics.meanAbsolutePercentageError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MAPE`, `tf.metrics.mape`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute percentage error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function MAPE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function mape(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean squared error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [3, 4]]);\n * const yPred = tf.tensor2d([[0, 1], [-3, -4]]);\n * const mse = tf.metrics.meanSquaredError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MSE`, `tf.metrics.mse`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean squared error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function MSE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function mse(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nexport {modelFromJSON} from './models';\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport * as regularizers from './regularizers';\n// tslint:disable-next-line:max-line-length\nimport {L1Args, L1L2, L1L2Args, L2Args, Regularizer} from './regularizers';\n\n/**\n * Regularizer for L1 and L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x)) + sum(l2 * x^2)\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1l2(config?: L1L2Args): Regularizer {\n  return new L1L2(config);\n}\n\n/**\n * Regularizer for L1 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x))\n * @param args l1 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1(config?: L1Args): Regularizer {\n  return regularizers.l1(config);\n}\n\n/**\n * Regularizer for L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l2 * x^2)\n * @param args l2 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l2(config?: L2Args): Regularizer {\n  return regularizers.l2(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {BaseCallback} from './base_callbacks';\nimport {Container} from './engine/container';\nimport {LayersModel} from './engine/training';\nimport {NotImplementedError} from './errors';\nimport {Logs, resolveScalarsInLogs} from './logs';\n\nexport abstract class Callback extends BaseCallback {\n  /** Instance of `keras.models.Model`. Reference of the model being trained. */\n  model: LayersModel = null;\n\n  setModel(model: Container): void {\n    if (!(model instanceof LayersModel)) {\n      throw new Error('model must be a LayersModel, not some other Container');\n    }\n    this.model = model;\n  }\n}\n\nexport interface EarlyStoppingCallbackArgs {\n  /**\n   * Quantity to be monitored.\n   *\n   * Defaults to 'val_loss'.\n   */\n  monitor?: string;\n\n  /**\n   * Minimum change in the monitored quantity to qualify as improvement,\n   * i.e., an absolute change of less than `minDelta` will count as no\n   * improvement.\n   *\n   * Defaults to 0.\n   */\n  minDelta?: number;\n\n  /**\n   * Number of epochs with no improvement after which training will be stopped.\n   *\n   * Defaults to 0.\n   */\n  patience?: number;\n\n  /** Verbosity mode. */\n  verbose?: number;\n\n  /**\n   * Mode: one of 'min', 'max', and 'auto'.\n   * - In 'min' mode, training will be stopped when the quantity monitored has\n   *   stopped decreasing.\n   * - In 'max' mode, training will be stopped when the quantity monitored has\n   *   stopped increasing.\n   * - In 'auto' mode, the direction is inferred automatically from the name of\n   *   the monitored quantity.\n   *\n   * Defaults to 'auto'.\n   */\n  mode?: 'auto'|'min'|'max';\n\n  /**\n   * Baseline value of the monitored quantity.\n   *\n   * If specified, training will be stopped if the model doesn't show\n   * improvement over the baseline.\n   */\n  baseline?: number;\n\n  /**\n   * Whether to restore model weights from the epoch with the best value\n   * of the monitored quantity. If `False`, the model weights obtained at the\n   * at the last step of training are used.\n   *\n   * **`True` is not supported yet.**\n   */\n  restoreBestWeights?: boolean;\n}\n\nfunction less(currVal: number, prevVal: number) {\n  return currVal < prevVal;\n}\n\nfunction greater(currVal: number, prevVal: number) {\n  return currVal > prevVal;\n}\n\n/**\n * A Callback that stops training when a monitored quantity has stopped\n * improving.\n */\nexport class EarlyStopping extends Callback {\n  protected readonly monitor: string;\n  protected readonly minDelta: number;\n  protected readonly patience: number;\n  protected readonly baseline: number;\n  protected readonly verbose: number;\n  protected readonly mode: 'auto'|'min'|'max';\n\n  protected monitorFunc: (currVal: number, prevVal: number) => boolean;\n\n  private wait: number;\n  private stoppedEpoch: number;\n  private best: number;\n\n  constructor(args?: EarlyStoppingCallbackArgs) {\n    super();\n    if (args == null) {\n      args = {};\n    }\n    if (args.restoreBestWeights) {\n      throw new NotImplementedError(\n          'restoreBestWeights = True is not implemented in EarlyStopping yet.');\n    }\n\n    this.monitor = args.monitor || 'val_loss';\n    this.minDelta = Math.abs(args.minDelta || 0);\n    this.patience = args.patience || 0;\n    this.verbose = args.verbose || 0;\n    this.mode = args.mode || 'auto';\n    this.baseline = args.baseline;\n\n    if (['auto', 'min', 'max'].indexOf(this.mode) === -1) {\n      console.warn(\n          `EarlyStopping mode '${this.mode}' is invalid. ` +\n          `Falling back to mode 'auto'.`);\n      this.mode = 'auto';\n    }\n\n    if (this.mode === 'min') {\n      this.monitorFunc = less;\n    } else if (this.mode === 'max') {\n      this.monitorFunc = greater;\n    } else {\n      // For mode === 'auto'.\n      if (this.monitor.indexOf('acc') !== -1) {\n        this.monitorFunc = greater;\n      } else {\n        this.monitorFunc = less;\n      }\n    }\n\n    if (this.monitorFunc === less) {\n      this.minDelta *= -1;\n    }\n  }\n\n  async onTrainBegin(logs?: Logs) {\n    this.wait = 0;\n    this.stoppedEpoch = 0;\n    if (this.baseline != null) {\n      this.best = this.baseline;\n    } else {\n      this.best = this.monitorFunc === less ? Infinity : -Infinity;\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: Logs) {\n    await resolveScalarsInLogs(logs);\n    const current = this.getMonitorValue(logs);\n    if (current == null) {\n      return;\n    }\n\n    if (this.monitorFunc(current - this.minDelta, this.best)) {\n      this.best = current;\n      this.wait = 0;\n      // TODO(cais): Logic for restoreBestWeights.\n    } else {\n      this.wait++;\n      if (this.wait >= this.patience) {\n        this.stoppedEpoch = epoch;\n        this.model.stopTraining = true;\n      }\n      // TODO(cais): Logic for restoreBestWeights.\n    }\n  }\n\n  async onTrainEnd(logs?: Logs) {\n    if (this.stoppedEpoch > 0 && this.verbose) {\n      console.log(`Epoch ${this.stoppedEpoch}: early stopping.`);\n    }\n  }\n\n  private getMonitorValue(logs: Logs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const monitorValue = logs[this.monitor];\n    if (monitorValue == null) {\n      console.warn(\n          `Metric for EarlyStopping ${this.monitor} is not available. ` +\n          `Available metrics are: ${Object.keys(logs)}`);\n    }\n    return monitorValue;\n  }\n}\n\n/**\n * Factory function for a Callback that stops training when a monitored\n * quantity has stopped improving.\n *\n * Early stopping is a type of regularization, and protects model against\n * overfitting.\n *\n * The following example based on fake data illustrates how this callback\n * can be used during `tf.LayersModel.fit()`:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense({\n *   units: 3,\n *   activation: 'softmax',\n *   kernelInitializer: 'ones',\n *   inputShape: [2]\n * }));\n * const xs = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const ys = tf.tensor2d([[1, 0, 0], [0, 1, 0]], [2, 3]);\n * const xsVal = tf.tensor2d([4, 3, 2, 1], [2, 2]);\n * const ysVal = tf.tensor2d([[0, 0, 1], [0, 1, 0]], [2, 3]);\n * model.compile(\n *     {loss: 'categoricalCrossentropy', optimizer: 'sgd', metrics: ['acc']});\n *\n * // Without the EarlyStopping callback, the val_acc value would be:\n * //   0.5, 0.5, 0.5, 0.5, ...\n * // With val_acc being monitored, training should stop after the 2nd epoch.\n * const history = await model.fit(xs, ys, {\n *   epochs: 10,\n *   validationData: [xsVal, ysVal],\n *   callbacks: tf.callbacks.earlyStopping({monitor: 'val_acc'})\n * });\n *\n * // Expect to see a length-2 array.\n * console.log(history.history.val_acc);\n * ```\n *\n * @doc {\n *   heading: 'Callbacks',\n *   namespace: 'callbacks'\n * }\n */\nexport function earlyStopping(args?: EarlyStoppingCallbackArgs) {\n  return new EarlyStopping(args);\n}\n\nexport const callbacks = {earlyStopping};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/* tslint:disable */\n\n/** Properties of an Any. */\nexport declare interface IAny {\n  /** Any typeUrl */\n  typeUrl?: (string|null);\n\n  /** Any value */\n  value?: (Uint8Array|null);\n}\n\n/** DataType enum. */\nexport enum DataType {\n  'DT_INVALID' = 0,\n  'DT_FLOAT' = 1,\n  'DT_DOUBLE' = 2,\n  'DT_INT32' = 3,\n  'DT_UINT8' = 4,\n  'DT_INT16' = 5,\n  'DT_INT8' = 6,\n  'DT_STRING' = 7,\n  'DT_COMPLEX64' = 8,\n  'DT_INT64' = 9,\n  'DT_BOOL' = 10,\n  'DT_QINT8' = 11,\n  'DT_QUINT8' = 12,\n  'DT_QINT32' = 13,\n  'DT_BFLOAT16' = 14,\n  'DT_FLOAT_REF' = 101,\n  'DT_DOUBLE_REF' = 102,\n  'DT_INT32_REF' = 103,\n  'DT_UINT8_REF' = 104,\n  'DT_INT16_REF' = 105,\n  'DT_INT8_REF' = 106,\n  'DT_STRING_REF' = 107,\n  'DT_COMPLEX64_REF' = 108,\n  'DT_INT64_REF' = 109,\n  'DT_BOOL_REF' = 110,\n  'DT_QINT8_REF' = 111,\n  'DT_QUINT8_REF' = 112,\n  'DT_QINT32_REF' = 113,\n  'DT_BFLOAT16_REF' = 114\n}\n\n/** Properties of a TensorShape. */\nexport declare interface ITensorShape {\n  /** TensorShape dim */\n  dim?: (TensorShape.IDim[]|null);\n\n  /** TensorShape unknownRank */\n  unknownRank?: (boolean|null);\n}\n\nexport namespace TensorShape {\n  /** Properties of a Dim. */\n  export declare interface IDim {\n    /** Dim size */\n    size?: (number|string|null);\n\n    /** Dim name */\n    name?: (string|null);\n  }\n}\n\n/** Properties of a Tensor. */\nexport declare interface ITensor {\n  /** Tensor dtype */\n  dtype?: (DataType|null);\n\n  /** Tensor tensorShape */\n  tensorShape?: (ITensorShape|null);\n\n  /** Tensor versionNumber */\n  versionNumber?: (number|null);\n\n  /** Tensor tensorContent */\n  tensorContent?: (Uint8Array|null);\n\n  /** Tensor floatVal */\n  floatVal?: (number[]|null);\n\n  /** Tensor doubleVal */\n  doubleVal?: (number[]|null);\n\n  /** Tensor intVal */\n  intVal?: (number[]|null);\n\n  /** Tensor stringVal */\n  stringVal?: (Uint8Array[]|null);\n\n  /** Tensor scomplexVal */\n  scomplexVal?: (number[]|null);\n\n  /** Tensor int64Val */\n  int64Val?: ((number | string)[]|null);\n\n  /** Tensor boolVal */\n  boolVal?: (boolean[]|null);\n\n  /** Tensor uint32Val */\n  uint32Val?: (number[]|null);\n\n  /** Tensor uint64Val */\n  uint64Val?: ((number | string)[]|null);\n}\n\n/** Properties of an AttrValue. */\nexport declare interface IAttrValue {\n  /** AttrValue list */\n  list?: (AttrValue.IListValue|null);\n\n  /** AttrValue s */\n  s?: (string|null);\n\n  /** AttrValue i */\n  i?: (number|string|null);\n\n  /** AttrValue f */\n  f?: (number|null);\n\n  /** AttrValue b */\n  b?: (boolean|null);\n\n  /** AttrValue type */\n  type?: (DataType|null);\n\n  /** AttrValue shape */\n  shape?: (ITensorShape|null);\n\n  /** AttrValue tensor */\n  tensor?: (ITensor|null);\n\n  /** AttrValue placeholder */\n  placeholder?: (string|null);\n\n  /** AttrValue func */\n  func?: (INameAttrList|null);\n}\n\nexport namespace AttrValue {\n  /** Properties of a ListValue. */\n  export declare interface IListValue {\n    /** ListValue s */\n    s?: (string[]|null);\n\n    /** ListValue i */\n    i?: ((number | string)[]|null);\n\n    /** ListValue f */\n    f?: (number[]|null);\n\n    /** ListValue b */\n    b?: (boolean[]|null);\n\n    /** ListValue type */\n    type?: (DataType[]|null);\n\n    /** ListValue shape */\n    shape?: (ITensorShape[]|null);\n\n    /** ListValue tensor */\n    tensor?: (ITensor[]|null);\n\n    /** ListValue func */\n    func?: (INameAttrList[]|null);\n  }\n}\n\n/** Properties of a NameAttrList. */\nexport declare interface INameAttrList {\n  /** NameAttrList name */\n  name?: (string|null);\n\n  /** NameAttrList attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a NodeDef. */\nexport declare interface INodeDef {\n  /** NodeDef name */\n  name?: (string|null);\n\n  /** NodeDef op */\n  op?: (string|null);\n\n  /** NodeDef input */\n  input?: (string[]|null);\n\n  /** NodeDef device */\n  device?: (string|null);\n\n  /** NodeDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a VersionDef. */\nexport declare interface IVersionDef {\n  /** VersionDef producer */\n  producer?: (number|null);\n\n  /** VersionDef minConsumer */\n  minConsumer?: (number|null);\n\n  /** VersionDef badConsumers */\n  badConsumers?: (number[]|null);\n}\n\n/** Properties of a GraphDef. */\nexport declare interface IGraphDef {\n  /** GraphDef node */\n  node?: (INodeDef[]|null);\n\n  /** GraphDef versions */\n  versions?: (IVersionDef|null);\n\n  /** GraphDef library */\n  library?: (IFunctionDefLibrary|null);\n}\n\n/** Properties of a CollectionDef. */\nexport declare interface ICollectionDef {\n  /** CollectionDef nodeList */\n  nodeList?: (CollectionDef.INodeList|null);\n\n  /** CollectionDef bytesList */\n  bytesList?: (CollectionDef.IBytesList|null);\n\n  /** CollectionDef int64List */\n  int64List?: (CollectionDef.IInt64List|null);\n\n  /** CollectionDef floatList */\n  floatList?: (CollectionDef.IFloatList|null);\n\n  /** CollectionDef anyList */\n  anyList?: (CollectionDef.IAnyList|null);\n}\n\nexport namespace CollectionDef {\n  /** Properties of a NodeList. */\n  export declare interface INodeList {\n    /** NodeList value */\n    value?: (string[]|null);\n  }\n\n  /** Properties of a BytesList. */\n  export declare interface IBytesList {\n    /** BytesList value */\n    value?: (Uint8Array[]|null);\n  }\n\n  /** Properties of an Int64List. */\n  export declare interface IInt64List {\n    /** Int64List value */\n    value?: ((number | string)[]|null);\n  }\n\n  /** Properties of a FloatList. */\n  export declare interface IFloatList {\n    /** FloatList value */\n    value?: (number[]|null);\n  }\n\n  /** Properties of an AnyList. */\n  export declare interface IAnyList {\n    /** AnyList value */\n    value?: (IAny[]|null);\n  }\n}\n\n/** Properties of a SaverDef. */\nexport declare interface ISaverDef {\n  /** SaverDef filenameTensorName */\n  filenameTensorName?: (string|null);\n\n  /** SaverDef saveTensorName */\n  saveTensorName?: (string|null);\n\n  /** SaverDef restoreOpName */\n  restoreOpName?: (string|null);\n\n  /** SaverDef maxToKeep */\n  maxToKeep?: (number|null);\n\n  /** SaverDef sharded */\n  sharded?: (boolean|null);\n\n  /** SaverDef keepCheckpointEveryNHours */\n  keepCheckpointEveryNHours?: (number|null);\n\n  /** SaverDef version */\n  version?: (SaverDef.CheckpointFormatVersion|null);\n}\n\nexport namespace SaverDef {\n  /** CheckpointFormatVersion enum. */\n  export enum CheckpointFormatVersion {'LEGACY' = 0, 'V1' = 1, 'V2' = 2}\n}\n\n/** Properties of a TensorInfo. */\nexport declare interface ITensorInfo {\n  /** TensorInfo name */\n  name?: (string|null);\n\n  /** TensorInfo cooSparse */\n  cooSparse?: (TensorInfo.ICooSparse|null);\n\n  /** TensorInfo dtype */\n  dtype?: (DataType|null);\n\n  /** TensorInfo tensorShape */\n  tensorShape?: (ITensorShape|null);\n}\n\nexport namespace TensorInfo {\n  /** Properties of a CooSparse. */\n  export declare interface ICooSparse {\n    /** CooSparse valuesTensorName */\n    valuesTensorName?: (string|null);\n\n    /** CooSparse indicesTensorName */\n    indicesTensorName?: (string|null);\n\n    /** CooSparse denseShapeTensorName */\n    denseShapeTensorName?: (string|null);\n  }\n}\n\n/** Properties of a SignatureDef. */\nexport declare interface ISignatureDef {\n  /** SignatureDef inputs */\n  inputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef outputs */\n  outputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef methodName */\n  methodName?: (string|null);\n}\n\n/** Properties of an AssetFileDef. */\nexport declare interface IAssetFileDef {\n  /** AssetFileDef tensorInfo */\n  tensorInfo?: (ITensorInfo|null);\n\n  /** AssetFileDef filename */\n  filename?: (string|null);\n}\n\n/** Properties of an OpDef. */\nexport declare interface IOpDef {\n  /** OpDef name */\n  name?: (string|null);\n\n  /** OpDef inputArg */\n  inputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef outputArg */\n  outputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef attr */\n  attr?: (OpDef.IAttrDef[]|null);\n\n  /** OpDef deprecation */\n  deprecation?: (OpDef.IOpDeprecation|null);\n\n  /** OpDef summary */\n  summary?: (string|null);\n\n  /** OpDef description */\n  description?: (string|null);\n\n  /** OpDef isCommutative */\n  isCommutative?: (boolean|null);\n\n  /** OpDef isAggregate */\n  isAggregate?: (boolean|null);\n\n  /** OpDef isStateful */\n  isStateful?: (boolean|null);\n\n  /** OpDef allowsUninitializedInput */\n  allowsUninitializedInput?: (boolean|null);\n}\n\nexport namespace OpDef {\n  /** Properties of an ArgDef. */\n  export declare interface IArgDef {\n    /** ArgDef name */\n    name?: (string|null);\n\n    /** ArgDef description */\n    description?: (string|null);\n\n    /** ArgDef type */\n    type?: (DataType|null);\n\n    /** ArgDef typeAttr */\n    typeAttr?: (string|null);\n\n    /** ArgDef numberAttr */\n    numberAttr?: (string|null);\n\n    /** ArgDef typeListAttr */\n    typeListAttr?: (string|null);\n\n    /** ArgDef isRef */\n    isRef?: (boolean|null);\n  }\n\n  /** Properties of an AttrDef. */\n  export declare interface IAttrDef {\n    /** AttrDef name */\n    name?: (string|null);\n\n    /** AttrDef type */\n    type?: (string|null);\n\n    /** AttrDef defaultValue */\n    defaultValue?: (IAttrValue|null);\n\n    /** AttrDef description */\n    description?: (string|null);\n\n    /** AttrDef hasMinimum */\n    hasMinimum?: (boolean|null);\n\n    /** AttrDef minimum */\n    minimum?: (number|string|null);\n\n    /** AttrDef allowedValues */\n    allowedValues?: (IAttrValue|null);\n  }\n\n  /** Properties of an OpDeprecation. */\n  export declare interface IOpDeprecation {\n    /** OpDeprecation version */\n    version?: (number|null);\n\n    /** OpDeprecation explanation */\n    explanation?: (string|null);\n  }\n}\n\n/** Properties of an OpList. */\nexport declare interface IOpList {\n  /** OpList op */\n  op?: (IOpDef[]|null);\n}\n\n/** Properties of a MetaGraphDef. */\nexport declare interface IMetaGraphDef {\n  /** MetaGraphDef metaInfoDef */\n  metaInfoDef?: (MetaGraphDef.IMetaInfoDef|null);\n\n  /** MetaGraphDef graphDef */\n  graphDef?: (IGraphDef|null);\n\n  /** MetaGraphDef saverDef */\n  saverDef?: (ISaverDef|null);\n\n  /** MetaGraphDef collectionDef */\n  collectionDef?: ({[k: string]: ICollectionDef}|null);\n\n  /** MetaGraphDef signatureDef */\n  signatureDef?: ({[k: string]: ISignatureDef}|null);\n\n  /** MetaGraphDef assetFileDef */\n  assetFileDef?: (IAssetFileDef[]|null);\n}\n\nexport namespace MetaGraphDef {\n  /** Properties of a MetaInfoDef. */\n  export declare interface IMetaInfoDef {\n    /** MetaInfoDef metaGraphVersion */\n    metaGraphVersion?: (string|null);\n\n    /** MetaInfoDef strippedOpList */\n    strippedOpList?: (IOpList|null);\n\n    /** MetaInfoDef anyInfo */\n    anyInfo?: (IAny|null);\n\n    /** MetaInfoDef tags */\n    tags?: (string[]|null);\n\n    /** MetaInfoDef tensorflowVersion */\n    tensorflowVersion?: (string|null);\n\n    /** MetaInfoDef tensorflowGitVersion */\n    tensorflowGitVersion?: (string|null);\n  }\n}\n\n/** Properties of a SavedModel. */\nexport declare interface ISavedModel {\n  /** SavedModel savedModelSchemaVersion */\n  savedModelSchemaVersion?: (number|string|null);\n\n  /** SavedModel metaGraphs */\n  metaGraphs?: (IMetaGraphDef[]|null);\n}\n\n/** Properties of a FunctionDefLibrary. */\nexport declare interface IFunctionDefLibrary {\n  /** FunctionDefLibrary function */\n  'function'?: (IFunctionDef[]|null);\n\n  /** FunctionDefLibrary gradient */\n  gradient?: (IGradientDef[]|null);\n}\n\n/** Properties of a FunctionDef. */\nexport declare interface IFunctionDef {\n  /** FunctionDef signature */\n  signature?: (IOpDef|null);\n\n  /** FunctionDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n\n  /** FunctionDef nodeDef */\n  nodeDef?: (INodeDef[]|null);\n\n  /** FunctionDef ret */\n  ret?: ({[k: string]: string}|null);\n}\n\n/** Properties of a GradientDef. */\nexport declare interface IGradientDef {\n  /** GradientDef functionName */\n  functionName?: (string|null);\n\n  /** GradientDef gradientFunc */\n  gradientFunc?: (string|null);\n}\n", "\n/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpExecutor, OpMapper} from '../types';\n\nconst CUSTOM_OPS: {[key: string]: OpMapper} = {};\n\n/**\n * Register an Op for graph model executor. This allow you to register\n * TensorFlow custom op or override existing op.\n *\n * Here is an example of registering a new MatMul Op.\n * ```js\n * const customMatmul = (node) =>\n *    tf.matMul(\n *        node.inputs[0], node.inputs[1],\n *        node.attrs['transpose_a'], node.attrs['transpose_b']);\n *\n * tf.registerOp('MatMul', customMatmul);\n * ```\n * The inputs and attrs of the node object is based on the TensorFlow op\n * registry.\n *\n * @param name The Tensorflow Op name.\n * @param opFunc An op function which is called with the current graph node\n * during execution and needs to return a tensor or a list of tensors. The node\n * has the following attributes:\n *    - attr: A map from attribute name to its value\n *    - inputs: A list of input tensors\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function registerOp(name: string, opFunc: OpExecutor) {\n  const opMapper: OpMapper = {\n    tfOpName: name,\n    category: 'custom',\n    inputs: [],\n    attrs: [],\n    customExecutor: opFunc\n  };\n\n  CUSTOM_OPS[name] = opMapper;\n}\n\n/**\n * Retrieve the OpMapper object for the registered op.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function getRegisteredOp(name: string): OpMapper {\n  return CUSTOM_OPS[name];\n}\n\n/**\n * Deregister the Op for graph model executor.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function deregisterOp(name: string) {\n  delete CUSTOM_OPS[name];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {clone, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {Node, ValueType} from '../types';\n\nexport function getParamValue(\n    paramName: string, node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  const inputParam = node.inputParams[paramName];\n  if (inputParam && inputParam.inputIndexStart !== undefined) {\n    const start = inputParam.inputIndexStart;\n    const end = inputParam.inputIndexEnd === 0 ?\n        undefined :\n        (inputParam.inputIndexEnd === undefined ? start + 1 :\n                                                  inputParam.inputIndexEnd);\n    if (inputParam.type === 'tensor') {\n      return getTensor(\n          node.inputNames[inputParam.inputIndexStart], tensorMap, context);\n    }\n    if (inputParam.type === 'tensors') {\n      const inputs = node.inputNames.slice(start, end);\n\n      return inputs.map(name => getTensor(name, tensorMap, context));\n    }\n    const tensor =\n        getTensor(node.inputNames.slice(start)[0], tensorMap, context);\n    const data = tensor.dataSync();\n    return inputParam.type === 'number' ?\n        data[0] :\n        util.toNestedArray(tensor.shape, data);\n  }\n  const attrParam = node.attrParams[paramName];\n  return attrParam && attrParam.value;\n}\n\n/**\n * Retrieve the tensor based on input name by extracting the node name and\n * output index information.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensor(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor {\n  const [nodeName, index] = parseNodeName(name);\n  const contextId = context.currentContextIds.find(contextId => {\n    return !!tensorsMap[getNodeNameWithContextId(nodeName, contextId)];\n  });\n\n  return contextId !== undefined ?\n      tensorsMap[getNodeNameWithContextId(nodeName, contextId)][index] :\n      undefined;\n}\n\n/**\n * Retrieve the tensors based on input name for current context.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensorsForCurrentContenxt(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor[] {\n  return tensorsMap[getNodeNameWithContextId(name, context.currentContextId)];\n}\n\n/**\n * Returns the node name and index from the Node input name.\n * @param inputName The input name of the node, in format of\n * node_name:output_index, i.e. MatMul:0, if the output_index is not set, it is\n * default to 0.\n */\nexport function getNodeNameAndIndex(\n    inputName: string, context?: ExecutionContext): [string, number] {\n  const [nodeName, index] = parseNodeName(inputName);\n\n  return [\n    getNodeNameWithContextId(nodeName, context && context.currentContextId),\n    index\n  ];\n}\n\nfunction getNodeNameWithContextId(name: string, contextId?: string): string {\n  return !!contextId ? `${name}-${contextId}` : name;\n}\n\nexport function parseNodeName(name: string): [string, number] {\n  const parts = name.split(':');\n  if (parts.length === 1) {\n    return [name, 0];\n  }\n\n  const nodeName = parts[0];\n  return [nodeName, Number(parts[parts.length - 1])];\n}\n\nexport function split(arr: number[], size: number) {\n  const res = [];\n  for (let i = 0; i < arr.length; i += size) {\n    res.push(arr.slice(i, i + size));\n  }\n  return res;\n}\nexport function getPadding(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  let pad = getParamValue('pad', node, tensorMap, context);\n  if (pad === 'explicit') {\n    // This is 1d array, we need to convert it to 2d array\n    pad = getParamValue('explicitPaddings', node, tensorMap, context);\n    const explicitPadding: [\n      [number, number], [number, number], [number, number], [number, number]\n    ] = [[0, 0], [0, 0], [0, 0], [0, 0]];\n    for (let i = 0; i < 4; i++) {\n      explicitPadding[i][0] = (pad as number[])[i * 2];\n      explicitPadding[i][1] = (pad as number[])[i * 2 + 1];\n    }\n    return explicitPadding;\n  }\n  return pad;\n}\n\n/**\n *  Reuse the tensor if it is marked as keep, otherwise clone the tensor to\n *  avoid disposal. This is important for TensorArray and TensorList ops, since\n *  internally they use a tensor as the id for TensorArray and TensorList, and\n * to simplify lookup, they also use Tensor.id as the key to the internal map.\n * These id tensors have been marked as kept in the backend, we need avoid clone\n * them in order to create new Tensor.id.\n * @param tensor\n */\nexport function cloneTensor(tensor: Tensor): Tensor {\n  return tensor.kept ? tensor : clone(tensor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Add',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddV2',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddN',\n    'category': 'arithmetic',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'BiasAdd',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sub',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'RealDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Div',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'DivNoNan',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mul',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Maximum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Minimum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Pow',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SquaredDifference',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorMod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Abs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan2',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ceil',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ClipByValue',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'clip_value_min', 'name': 'clipValueMin', 'type': 'number'},\n      {'tfName': 'clip_value_max', 'name': 'clipValueMax', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Complex',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'real', 'type': 'tensor'},\n      {'start': 1, 'name': 'imag', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ComplexAbs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Elu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Exp',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Floor',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Imag',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Neg',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Real',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Prelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'alpha', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu6',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'clipValueMin',\n        'name': 'clipValueMin',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'clipValueMax',\n        'name': 'clipValueMax',\n        'type': 'number',\n        'defaultValue': 6\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Selu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sigmoid',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Rsqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Square',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sign',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Round',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Expm1',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log1p',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Reciprocal',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Softplus',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Erf',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axes', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'keep_dims',\n        'name': 'keepDims',\n        'type': 'bool',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LeakyRelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 0.2\n      },\n      {\n        'tfName': 'T',\n        'name': 'dtype',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'LoopCond',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'pred', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Switch',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'data', 'type': 'tensor'},\n      {'start': 1, 'name': 'pred', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Merge',\n    'category': 'control',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Enter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'frame_name', 'name': 'frameName', 'type': 'string'},\n      {'tfName': 'is_constant', 'name': 'isConstant', 'type': 'bool'}\n    ]\n  },\n  {\n    'tfOpName': 'Exit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NextIteration',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'size', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'dynamic_size', 'name': 'dynamicSize', 'type': 'bool'},\n      {'tfName': 'clear_after_read', 'name': 'clearAfterRead', 'type': 'bool'},\n      {\n        'tfName': 'identical_element_shapes',\n        'name': 'identicalElementShapes',\n        'type': 'bool'\n      },\n      {'tfName': 'tensor_array_name', 'name': 'name', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayWriteV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayReadV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{\n      'tfName': 'dtype',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  },\n  {\n    'tfOpName': 'TensorArrayGatherV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayScatterV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArrayConcatV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}, {\n        'tfName': 'element_shape_except0',\n        'name': 'elementShapeExcept0',\n        'type': 'shape',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'TensorArraySplitV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArraySizeV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayCloseV3',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'StatelessIf',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'If',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'StatelessWhile',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'While',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListScatter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListScatterV2',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 3, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGather',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListSetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListReserve',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 1, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListFromTensor',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListStack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs': [\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'},\n      {'tfName': 'num_elements', 'name': 'numElements', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListSplit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListConcat',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListPopBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListPushBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n];\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'AvgPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPoolWithArgmax',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'}, {\n        'tfName': 'include_batch_in_index',\n        'name': 'includeBatchInIndex',\n        'type': 'bool'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AvgPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Conv1D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'stride', 'name': 'stride', 'type': 'number'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NWC'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'dilation',\n        'name': 'dilation',\n        'type': 'number',\n        'defaultValue': 1\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'useCudnnOnGpu', 'name': 'useCudnnOnGpu', 'type': 'bool'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': '_FusedConv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'use_cudnn_on_gpu',\n        'name': 'useCudnnOnGpu',\n        'type': 'bool',\n        'defaultValue': true\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n    ]\n  },\n  {\n    'tfOpName': 'Conv2DBackpropInput',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 2, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 0, 'name': 'outputShape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2d',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'FusedDepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ],\n  },\n  {\n    'tfOpName': 'Dilation2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'rates', 'name': 'dilations', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Fill',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n      {'start': 1, 'name': 'value', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'LinSpace',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'num', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'OneHot',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'depth', 'type': 'number'},\n      {'start': 2, 'name': 'onValue', 'type': 'number', 'defaultValue': 1},\n      {'start': 3, 'name': 'offValue', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [\n      {\n        'tfName': 'axis',\n        'name': 'axis',\n        'type': 'number',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ones',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'OnesLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'RandomUniform',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'minval',\n        'name': 'minval',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'maxval',\n        'name': 'maxval',\n        'type': 'number',\n        'defaultValue': 1\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Range',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'step', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [{'tfName': 'Tidx', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TruncatedNormal',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'means',\n        'name': 'mean',\n        'type': 'number',\n        'defaultValue': 0.0\n      },\n      {\n        'tfName': 'stddev',\n        'name': 'stdDev',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Zeros',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'ZerosLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'Multinomial',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'logits', 'type': 'tensor'},\n      {'start': 1, 'name': 'numSamples', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'},\n      {'tfName': 'seed2', 'name': 'seed2', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'output_dtype', 'name': 'output_dtype', 'type': 'dtype'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'NonMaxSuppressionV2',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV3',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV4',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'T_threshold',\n        'name': 'threshold',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {\n        'tfName': 'pad_to_max_output_size',\n        'name': 'padToMaxOutputSize',\n        'type': 'bool'\n      }\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV5',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'},\n      {'start': 5, 'name': 'softNmsSigma', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Where',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ListDiff',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'TopKV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'k', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'sorted', 'name': 'sorted', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Unique',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n  },\n  {\n    'tfOpName': 'UniqueV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n  },\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'PlaceholderWithDefault',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'default', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'Placeholder',\n    'category': 'graph',\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {'tfOpName': 'Const', 'category': 'graph'}, {\n    'tfOpName': 'Identity',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IdentityN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Snapshot',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Rank',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Size',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Shape',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'ShapeN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Print',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'data', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'message', 'name': 'message', 'type': 'string'}, {\n        'tfName': 'first_n',\n        'name': 'firstN',\n        'type': 'number',\n        'notSupported': true\n      },\n      {\n        'tfName': 'summarize',\n        'name': 'summarize',\n        'type': 'number',\n        'defaultValue': 3\n      }\n    ]\n  },\n  {'tfOpName': 'NoOp', 'category': 'graph', 'inputs': []}, {\n    'tfOpName': 'StopGradient',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'FakeQuantWithMinMaxVars',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'min', 'name': 'min', 'type': 'number'},\n      {'tfName': 'max', 'name': 'max', 'type': 'number'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ResizeBilinear',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ResizeNearestNeighbor',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'CropAndResize',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'image', 'type': 'tensor'},\n      {'start': 1, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 2, 'name': 'boxInd', 'type': 'tensor'},\n      {'start': 3, 'name': 'cropSize', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'method', 'name': 'method', 'type': 'string'}, {\n        'tfName': 'extrapolation_value',\n        'name': 'extrapolationValue',\n        'type': 'number'\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Equal',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NotEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Greater',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'GreaterEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Less',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LessEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalAnd',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalNot',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalOr',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Select',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SelectV2',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': '_FusedMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'}, {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMulV2',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Transpose',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'perm', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FusedBatchNorm',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV2',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV3',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'LRN',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'depth_radius',\n        'name': 'radius',\n        'type': 'number',\n        'defaultValue': 5\n      },\n      {'tfName': 'bias', 'name': 'bias', 'type': 'number', 'defaultValue': 1.0},\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {\n        'tfName': 'beta',\n        'name': 'beta',\n        'type': 'number',\n        'defaultValue': 0.5\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Softmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'LogSoftmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': true,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Max',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Mean',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Min',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Sum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'All',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Any',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'ArgMax',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'ArgMin',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Cumsum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'exclusive', 'name': 'exclusive', 'type': 'bool'},\n      {'tfName': 'reverse', 'name': 'reverse', 'type': 'bool'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ConcatV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': -1, 'name': 'tensors', 'type': 'tensors'},\n      {'start': -1, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs':\n        [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n  },\n  {\n    'tfOpName': 'Concat',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 1, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n      {'start': 0, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs': [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n\n  },\n  {\n    'tfOpName': 'GatherV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Gather',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'validate_indices',\n        'name': 'validateIndices',\n        'type': 'bool',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reverse',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'dims', 'type': 'bool', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ReverseV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Slice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'size', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'StridedSlice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'end', 'type': 'number[]'},\n      {'start': 3, 'name': 'strides', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'begin_mask',\n        'name': 'beginMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'end_mask',\n        'name': 'endMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'new_axis_mask',\n        'name': 'newAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'ellipsis_mask',\n        'name': 'ellipsisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'shrink_axis_mask',\n        'name': 'shrinkAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Pack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Unpack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'num',\n        'name': 'num',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Tile',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'reps', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Split',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'axis', 'type': 'number', 'defaultValue': 0},\n      {'start': 1, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'num_split',\n      'name': 'numOrSizeSplits',\n      'type': 'number',\n      'defaultValue': 1\n    }]\n  },\n  {\n    'tfOpName': 'SplitV',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'numOrSizeSplits', 'type': 'number[]'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'ScatterNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'values', 'type': 'tensor'},\n      {'start': 2, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'GatherNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': false,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IFFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'RFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'IRFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Cast',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'SrcT',\n        'name': 'sdtype',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {'tfName': 'DstT', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'ExpandDims',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Pad',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'constant_value',\n      'name': 'constantValue',\n      'type': 'number',\n      'defaultValue': 0\n    }]\n  },\n  {\n    'tfOpName': 'PadV2',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'}, {\n        'start': 2,\n        'name': 'constantValue',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reshape',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Squeeze',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'axis',\n      'tfDeprecatedName': 'squeeze_dims',\n      'name': 'axis',\n      'type': 'number[]'\n    }]\n  },\n  {\n    'tfOpName': 'SpaceToBatchND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'paddings', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'BatchToSpaceND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'crops', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthToSpace',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'block_size', 'name': 'blockSize', 'type': 'number'},\n      {'tfName': 'data_format', 'name': 'dataFormat', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'BroadcastTo',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': []\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, env} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\n\nimport {getRegisteredOp} from './custom_op/register';\nimport {getNodeNameAndIndex} from './executors/utils';\nimport * as arithmetic from './op_list/arithmetic';\nimport * as basicMath from './op_list/basic_math';\nimport * as control from './op_list/control';\nimport * as convolution from './op_list/convolution';\nimport * as creation from './op_list/creation';\nimport * as dynamic from './op_list/dynamic';\nimport * as evaluation from './op_list/evaluation';\nimport * as graph from './op_list/graph';\nimport * as image from './op_list/image';\nimport * as logical from './op_list/logical';\nimport * as matrices from './op_list/matrices';\nimport * as normalization from './op_list/normalization';\nimport * as reduction from './op_list/reduction';\nimport * as sliceJoin from './op_list/slice_join';\nimport * as spectral from './op_list/spectral';\nimport * as transformation from './op_list/transformation';\nimport {Graph, InputParamValue, Node, OpMapper, ParamValue} from './types';\n\nexport class OperationMapper {\n  private static _instance: OperationMapper;\n\n  private opMappers: {[key: string]: OpMapper};\n\n  // Singleton instance for the mapper\n  public static get Instance() {\n    return this._instance || (this._instance = new this());\n  }\n\n  // Loads the op mapping from the JSON file.\n  private constructor() {\n    const ops = [\n      arithmetic, basicMath, control, convolution, creation, dynamic,\n      evaluation, logical, image, graph, matrices, normalization, reduction,\n      sliceJoin, spectral, transformation\n    ];\n    const mappersJson: OpMapper[] = [].concat(...ops.map(op => op.json));\n\n    this.opMappers = mappersJson.reduce<{[key: string]: OpMapper}>(\n        (map, mapper: OpMapper) => {\n          map[mapper.tfOpName] = mapper;\n          return map;\n        },\n        {});\n  }\n\n  // Converts the model inference graph from Tensorflow GraphDef to local\n  // representation for TensorFlow.js API\n  transformGraph(\n      graph: tensorflow.IGraphDef,\n      signature: tensorflow.ISignatureDef = {}): Graph {\n    const tfNodes = graph.node;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    const initNodes: Node[] = [];\n    const nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n      map[node.name] = this.mapNode(node);\n      if (node.op.startsWith('Placeholder')) {\n        placeholders.push(map[node.name]);\n      } else if (node.op === 'Const') {\n        weights.push(map[node.name]);\n      } else if (node.input == null || node.input.length === 0) {\n        initNodes.push(map[node.name]);\n      }\n      return map;\n    }, {});\n\n    let inputs: Node[] = [];\n    const outputs: Node[] = [];\n    let inputNodeNameToKey: {[key: string]: string} = {};\n    let outputNodeNameToKey: {[key: string]: string} = {};\n    if (signature != null) {\n      inputNodeNameToKey = this.mapSignatureEntries(signature.inputs);\n      outputNodeNameToKey = this.mapSignatureEntries(signature.outputs);\n    }\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    // if signature has not outputs set, add any node that does not have\n    // outputs.\n    if (Object.keys(outputNodeNameToKey).length === 0) {\n      allNodes.forEach(key => {\n        const node = nodes[key];\n        if (node.children.length === 0) {\n          outputs.push(node);\n        }\n      });\n    } else {\n      Object.keys(outputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node != null) {\n          node.signatureKey = outputNodeNameToKey[name];\n          outputs.push(node);\n        }\n      });\n    }\n\n    if (Object.keys(inputNodeNameToKey).length > 0) {\n      Object.keys(inputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node) {\n          node.signatureKey = inputNodeNameToKey[name];\n          inputs.push(node);\n        }\n      });\n    } else {\n      inputs = placeholders;\n    }\n\n    let functions = {};\n    if (graph.library != null && graph.library.function != null) {\n      functions = graph.library.function.reduce((functions, func) => {\n        functions[func.signature.name] = this.mapFunction(func);\n        return functions;\n      }, {} as {[key: string]: Graph});\n    }\n\n    const result: Graph =\n        {nodes, inputs, outputs, weights, placeholders, signature, functions};\n\n    if (initNodes.length > 0) {\n      result.initNodes = initNodes;\n    }\n\n    return result;\n  }\n\n  private mapSignatureEntries(entries: {[k: string]: tensorflow.ITensorInfo}) {\n    return Object.keys(entries || {})\n        .reduce<{[key: string]: string}>((prev, curr) => {\n          prev[entries[curr].name] = curr;\n          return prev;\n        }, {});\n  }\n\n  private mapNode(node: tensorflow.INodeDef): Node {\n    // Unsupported ops will cause an error at run-time (not parse time), since\n    // they may not be used by the actual execution subgraph.\n    const mapper =\n        getRegisteredOp(node.op) || this.opMappers[node.op] || {} as OpMapper;\n    if (node.attr == null) {\n      node.attr = {};\n    }\n\n    const newNode: Node = {\n      name: node.name,\n      op: node.op,\n      category: mapper.category,\n      inputNames:\n          (node.input ||\n           []).map(input => input.startsWith('^') ? input.substr(1) : input),\n      inputs: [],\n      children: [],\n      inputParams: {},\n      attrParams: {},\n      rawAttrs: node.attr\n    };\n\n    if (mapper.inputs != null) {\n      newNode.inputParams =\n          mapper.inputs.reduce<{[key: string]: InputParamValue}>(\n              (map, param) => {\n                map[param.name] = {\n                  type: param.type,\n                  inputIndexStart: param.start,\n                  inputIndexEnd: param.end\n                };\n                return map;\n              },\n              {});\n    }\n    if (mapper.attrs != null) {\n      newNode.attrParams =\n          mapper.attrs.reduce<{[key: string]: ParamValue}>((map, param) => {\n            const type = param.type;\n            let value = undefined;\n            switch (param.type) {\n              case 'string':\n                value = getStringParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'string[]':\n                value = getStringArrayParam(\n                    node.attr, param.tfName, param.defaultValue as string[]);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string[]);\n                }\n                break;\n              case 'number':\n                value = getNumberParam(\n                    node.attr, param.tfName,\n                    (param.defaultValue || 0) as number);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumberParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number);\n                }\n                break;\n              case 'number[]':\n                value = getNumericArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumericArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'bool':\n                value = getBoolParam(\n                    node.attr, param.tfName, param.defaultValue as boolean);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean);\n                }\n                break;\n              case 'bool[]':\n                value = getBoolArrayParam(\n                    node.attr, param.tfName, param.defaultValue as boolean[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean[]);\n                }\n                break;\n              case 'shape':\n                value = getTensorShapeParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'shape[]':\n                value = getTensorShapeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[][]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[][]);\n                }\n                break;\n              case 'dtype':\n                value = getDtypeParam(\n                    node.attr, param.tfName, param.defaultValue as DataType);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType);\n                }\n                break;\n              case 'dtype[]':\n                value = getDtypeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as DataType[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType[]);\n                }\n                break;\n              case 'func':\n                value = getFuncParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getFuncParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'tensor':\n              case 'tensors':\n                break;\n              default:\n                throw new Error(\n                    `Unsupported param type: ${param.type} for op: ${node.op}`);\n            }\n            map[param.name] = {value, type};\n            return map;\n          }, {});\n    }\n    return newNode;\n  }\n\n  // map the TFunctionDef to TFJS graph object\n  private mapFunction(functionDef: tensorflow.IFunctionDef): Graph {\n    const tfNodes = functionDef.nodeDef;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    let nodes: {[key: string]: Node} = {};\n    if (tfNodes != null) {\n      nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n        map[node.name] = this.mapNode(node);\n        if (node.op === 'Const') {\n          weights.push(map[node.name]);\n        }\n        return map;\n      }, {});\n    }\n    const inputs: Node[] = [];\n    const outputs: Node[] = [];\n\n    functionDef.signature.inputArg.forEach(arg => {\n      const [nodeName, ] = getNodeNameAndIndex(arg.name);\n      const node: Node = {\n        name: nodeName,\n        op: 'Placeholder',\n        inputs: [],\n        inputNames: [],\n        category: 'graph',\n        inputParams: {},\n        attrParams: {dtype: {value: parseDtypeParam(arg.type), type: 'dtype'}},\n        children: []\n      };\n      node.signatureKey = arg.name;\n      inputs.push(node);\n      nodes[nodeName] = node;\n    });\n\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    const returnNodeMap = functionDef.ret;\n\n    functionDef.signature.outputArg.forEach(output => {\n      const [nodeName, index] = getNodeNameAndIndex(returnNodeMap[output.name]);\n      const node = nodes[nodeName];\n      if (node != null) {\n        node.defaultOutput = index;\n        outputs.push(node);\n      }\n    });\n\n    const signature = this.mapArgsToSignature(functionDef);\n    return {nodes, inputs, outputs, weights, placeholders, signature};\n  }\n\n  private mapArgsToSignature(functionDef: tensorflow.IFunctionDef):\n      tensorflow.ISignatureDef {\n    return {\n      methodName: functionDef.signature.name,\n      inputs: functionDef.signature.inputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n      outputs: functionDef.signature.outputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg, functionDef.ret);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n    };\n  }\n\n  private mapArgToTensorInfo(\n      arg: tensorflow.OpDef.IArgDef,\n      nameMap?: {[key: string]: string}): tensorflow.ITensorInfo {\n    let name = arg.name;\n    if (nameMap != null) {\n      name = nameMap[name];\n    }\n    return {name, dtype: arg.type};\n  }\n}\n\nexport function decodeBase64(text: string): string {\n  const global = env().global;\n  if (typeof global.atob !== 'undefined') {\n    return global.atob(text);\n  } else if (typeof Buffer !== 'undefined') {\n    return new Buffer(text, 'base64').toString();\n  } else {\n    throw new Error(\n        'Unable to decode base64 in this environment. ' +\n        'Missing built-in atob() or Buffer()');\n  }\n}\n\nexport function parseStringParam(s: []|string, keepCase: boolean): string {\n  const value =\n      Array.isArray(s) ? String.fromCharCode.apply(null, s) : decodeBase64(s);\n  return keepCase ? value : value.toLowerCase();\n}\n\nexport function getStringParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string,\n    keepCase = false): string {\n  const param = attrs[name];\n  if (param != null) {\n    return parseStringParam(param.s, keepCase);\n  }\n  return def;\n}\n\nexport function getBoolParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean): boolean {\n  const param = attrs[name];\n  return param ? param.b : def;\n}\n\nexport function getNumberParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number): number {\n  const param = attrs[name] || {};\n  const value =\n      param['i'] != null ? param['i'] : (param['f'] != null ? param['f'] : def);\n  return (typeof value === 'number') ? value : parseInt(value, 10);\n}\n\nexport function parseDtypeParam(value: string|tensorflow.DataType): DataType {\n  if (typeof (value) === 'string') {\n    // tslint:disable-next-line:no-any\n    value = tensorflow.DataType[value as any];\n  }\n  switch (value) {\n    case tensorflow.DataType.DT_FLOAT:\n      return 'float32';\n    case tensorflow.DataType.DT_INT32:\n    case tensorflow.DataType.DT_INT64:\n    case tensorflow.DataType.DT_INT8:\n    case tensorflow.DataType.DT_UINT8:\n      return 'int32';\n    case tensorflow.DataType.DT_BOOL:\n      return 'bool';\n    case tensorflow.DataType.DT_DOUBLE:\n      return 'float32';\n    case tensorflow.DataType.DT_STRING:\n      return 'string';\n    default:\n      // Unknown dtype error will happen at runtime (instead of parse time),\n      // since these nodes might not be used by the actual subgraph execution.\n      return null;\n  }\n}\n\nexport function getFuncParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: string): string {\n  const param = attrs[name];\n  if (param && param.func) {\n    return param.func.name;\n  }\n  return def;\n}\n\nexport function getDtypeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType): DataType {\n  const param = attrs[name];\n  if (param && param.type) {\n    return parseDtypeParam(param.type);\n  }\n  return def;\n}\n\nexport function getDtypeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType[]): DataType[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.type) {\n    return param.list.type.map(v => parseDtypeParam(v));\n  }\n  return def;\n}\n\nexport function parseTensorShapeParam(shape: tensorflow.ITensorShape): number[]|\n    undefined {\n  if (shape.unknownRank) {\n    return undefined;\n  }\n  if (shape.dim != null) {\n    return shape.dim.map(\n        dim =>\n            (typeof dim.size === 'number') ? dim.size : parseInt(dim.size, 10));\n  }\n  return [];\n}\n\nexport function getTensorShapeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def?: number[]): number[]|undefined {\n  const param = attrs[name];\n  if (param && param.shape) {\n    return parseTensorShapeParam(param.shape);\n  }\n  return def;\n}\n\nexport function getNumericArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[]): number[] {\n  const param = attrs[name];\n  if (param) {\n    return ((param.list.f && param.list.f.length ? param.list.f :\n                                                   param.list.i) ||\n            [])\n        .map(v => (typeof v === 'number') ? v : parseInt(v, 10));\n  }\n  return def;\n}\n\nexport function getStringArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string[],\n    keepCase = false): string[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.s) {\n    return param.list.s.map((v) => {\n      return parseStringParam(v, keepCase);\n    });\n  }\n  return def;\n}\n\nexport function getTensorShapeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[][]): number[][] {\n  const param = attrs[name];\n  if (param && param.list && param.list.shape) {\n    return param.list.shape.map((v) => {\n      return parseTensorShapeParam(v);\n    });\n  }\n  return def;\n}\n\nexport function getBoolArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean[]): boolean[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.b) {\n    return param.list.b;\n  }\n  return def;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {getTensor} from '../executors/utils';\nimport {getBoolArrayParam, getBoolParam, getDtypeArrayParam, getDtypeParam, getNumberParam, getNumericArrayParam, getStringArrayParam, getStringParam, getTensorShapeArrayParam, getTensorShapeParam} from '../operation_mapper';\nimport {GraphNode, Node, ValueType} from '../types';\n\n/**\n * Helper class for lookup inputs and params for nodes in the model graph.\n */\nexport class NodeValueImpl implements GraphNode {\n  public readonly inputs: Tensor[] = [];\n  public readonly attrs: {[key: string]: ValueType} = {};\n  constructor(\n      private node: Node, private tensorMap: NamedTensorsMap,\n      private context: ExecutionContext) {\n    this.inputs = node.inputNames.map(name => this.getInput(name));\n    if (node.rawAttrs != null) {\n      this.attrs = Object.keys(node.rawAttrs)\n                       .reduce((attrs: {[key: string]: ValueType}, key) => {\n                         attrs[key] = this.getAttr(key);\n                         return attrs;\n                       }, {});\n    }\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getInput(name: string): Tensor {\n    return getTensor(name, this.tensorMap, this.context);\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getAttr(name: string, defaultValue?: ValueType): ValueType {\n    const value = this.node.rawAttrs[name];\n    if (value.tensor != null) {\n      return getTensor(name, this.tensorMap, this.context);\n    }\n    if (value.i != null || value.f != null) {\n      return getNumberParam(this.node.rawAttrs, name, defaultValue as number);\n    }\n    if (value.s != null) {\n      return getStringParam(this.node.rawAttrs, name, defaultValue as string);\n    }\n    if (value.b != null) {\n      return getBoolParam(this.node.rawAttrs, name, defaultValue as boolean);\n    }\n    if (value.shape != null) {\n      return getTensorShapeParam(\n          this.node.rawAttrs, name, defaultValue as number[]);\n    }\n    if (value.type != null) {\n      return getDtypeParam(this.node.rawAttrs, name, defaultValue as DataType);\n    }\n    if (value.list != null) {\n      if (value.list.i != null || value.list.f != null) {\n        return getNumericArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[]);\n      }\n      if (value.list.s != null) {\n        return getStringArrayParam(\n            this.node.rawAttrs, name, defaultValue as string[]);\n      }\n      if (value.list.shape != null) {\n        return getTensorShapeArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[][]);\n      }\n      if (value.list.b != null) {\n        return getBoolArrayParam(\n            this.node.rawAttrs, name, defaultValue as boolean[]);\n      }\n      if (value.list.type != null) {\n        return getDtypeArrayParam(\n            this.node.rawAttrs, name, defaultValue as DataType[]);\n      }\n    }\n\n    return defaultValue;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BiasAdd':\n        case 'AddV2':\n        case 'Add': {\n          return [tfOps.add(\n              (getParamValue('a', node, tensorMap, context) as Tensor),\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'AddN': {\n          return [tfOps.addN((\n              getParamValue('tensors', node, tensorMap, context) as Tensor[]))];\n        }\n        case 'FloorMod':\n        case 'Mod':\n          return [tfOps.mod(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'Mul':\n          return [tfOps.mul(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'RealDiv':\n        case 'Div': {\n          return [tfOps.div(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'DivNoNan': {\n          return [tfOps.divNoNan(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'FloorDiv': {\n          return [tfOps.floorDiv(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sub': {\n          return [tfOps.sub(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Minimum': {\n          return [tfOps.minimum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Maximum': {\n          return [tfOps.maximum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Pow': {\n          return [tfOps.pow(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'SquaredDifference': {\n          return [tfOps.squaredDifference(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'arithmetic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Abs':\n        case 'ComplexAbs':\n          return [tfOps.abs(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acos':\n          return [tfOps.acos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acosh':\n          return [tfOps.acosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asin':\n          return [tfOps.asin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asinh':\n          return [tfOps.asinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan':\n          return [tfOps.atan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan2':\n          return [tfOps.atan2(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('y', node, tensorMap, context) as Tensor)];\n        case 'Atanh':\n          return [tfOps.atanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Ceil':\n          return [tfOps.ceil(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Complex':\n          return [tfOps.complex(\n              getParamValue('real', node, tensorMap, context) as Tensor,\n              getParamValue('imag', node, tensorMap, context) as Tensor)];\n        case 'Cos':\n          return [tfOps.cos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Cosh':\n          return [tfOps.cosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Elu':\n          return [tfOps.elu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Erf':\n          return [tfOps.erf(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Exp':\n          return [tfOps.exp(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Expm1': {\n          return [tfOps.expm1(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Floor':\n          return [tfOps.floor(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log':\n          return [tfOps.log(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log1p': {\n          return [tfOps.log1p(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Imag':\n          return [tfOps.imag(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n\n        case 'Neg':\n          return [tfOps.neg(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Reciprocal': {\n          return [tfOps.reciprocal(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Real':\n          return [tfOps.real(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu':\n          return [tfOps.relu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Round': {\n          return [tfOps.round(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Selu':\n          return [tfOps.selu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sigmoid':\n          return [tfOps.sigmoid(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sin':\n          return [tfOps.sin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sign': {\n          return [tfOps.sign(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sinh': {\n          return [tfOps.sinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Softplus': {\n          return [tfOps.softplus(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sqrt': {\n          return [tfOps.sqrt(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Square': {\n          return [tfOps.square(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tanh': {\n          return [tfOps.tanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tan':\n          return [tfOps.tan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu6':\n        case 'ClipByValue':\n          return [tfOps.clipByValue(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('clipValueMin', node, tensorMap, context) as number,\n              getParamValue('clipValueMax', node, tensorMap, context) as\n                  number)];\n        case 'Rsqrt':\n          return [tfOps.rsqrt(\n              getTensor(node.inputNames[0], tensorMap, context))];\n        case 'Prod':\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('axes', node, tensorMap, context) as number[])];\n        case 'LeakyRelu':\n          return [tfOps.leakyRelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as number)];\n        case 'Prelu':\n          return [tfOps.prelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as Tensor)];\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'basic_math';\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n/**\n * This differs from util.assertShapesMatch in that it allows values of\n * negative one, an undefined size of a dimensinon, in a shape to match\n * anything.\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function assertShapesMatchAllowUndefinedSize(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  util.assert(\n      shapesEqualAllowUndefinedSize(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function shapesEqualAllowUndefinedSize(n1: number[], n2: number[]) {\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== -1 && n2[i] !== -1 && n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\nexport interface TensorWithState {\n  tensor?: Tensor;\n  written?: boolean;\n  read?: boolean;\n  cleared?: boolean;\n}\n/**\n * The TensorArray object keeps an array of Tensors.  It\n * allows reading from the array and writing to the array.\n */\nexport class TensorArray {\n  private tensors: TensorWithState[] = [];\n  private closed_ = false;\n  readonly idTensor: Tensor;\n  constructor(\n      readonly name: string, readonly dtype: DataType, private maxSize: number,\n      private elementShape: number[], readonly identicalElementShapes: boolean,\n      readonly dynamicSize: boolean, readonly clearAfterRead: boolean) {\n    this.idTensor = scalar(0);\n    keep(this.idTensor);\n  }\n\n  get id() {\n    return this.idTensor.id;\n  }\n\n  get closed() {\n    return this.closed_;\n  }\n\n  /**\n   * Dispose the tensors and idTensor and mark the TensoryArray as closed.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.tensor.id)) {\n        tensor.tensor.dispose();\n      }\n    });\n    this.tensors = [];\n    this.closed_ = true;\n    this.idTensor.dispose();\n  }\n\n  size(): number {\n    return this.tensors.length;\n  }\n\n  /**\n   * Read the value at location index in the TensorArray.\n   * @param index Number the index to read from.\n   */\n  read(index: number): Tensor {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || index >= this.size()) {\n      throw new Error(`Tried to read from index ${index}, but array size is: ${\n          this.size()}`);\n    }\n\n    const tensorWithState = this.tensors[index];\n    if (tensorWithState.cleared) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not read index ${\n              index} twice because it was cleared after a previous read ` +\n          `(perhaps try setting clear_after_read = false?).`);\n    }\n\n    if (this.clearAfterRead) {\n      tensorWithState.cleared = true;\n    }\n\n    tensorWithState.read = true;\n    return tensorWithState.tensor;\n  }\n\n  /**\n   * Helper method to read multiple tensors from the specified indices.\n   */\n  readMany(indices: number[]): Tensor[] {\n    return indices.map(index => this.read(index));\n  }\n\n  /**\n   * Write value into the index of the TensorArray.\n   * @param index number the index to write to.\n   * @param tensor\n   */\n  write(index: number, tensor: Tensor) {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || !this.dynamicSize && index >= this.maxSize) {\n      throw new Error(`Tried to write to index ${\n          index}, but array is not resizeable and size is: ${this.maxSize}`);\n    }\n\n    const t = this.tensors[index] || {};\n\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray ${\n          this.name}: Could not write to TensorArray index ${index},\n          because the value dtype is ${\n          tensor.dtype}, but TensorArray dtype is ${this.dtype}.`);\n    }\n\n    // Set the shape for the first time write to unknow shape tensor array\n    if (this.size() === 0 &&\n        (this.elementShape == null || this.elementShape.length === 0)) {\n      this.elementShape = tensor.shape;\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape,\n        `TensorArray ${this.name}: Could not write to TensorArray index ${\n            index}.`);\n\n    if (t.read) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been read.`);\n    }\n\n    if (t.written) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been written.`);\n    }\n\n    t.tensor = tensor;\n    keep(tensor);\n    t.written = true;\n\n    this.tensors[index] = t;\n  }\n\n  /**\n   * Helper method to write multiple tensors to the specified indices.\n   */\n  writeMany(indices: number[], tensors: Tensor[]) {\n    if (indices.length !== tensors.length) {\n      throw new Error(\n          `TensorArray ${this.name}: could not write multiple tensors,` +\n          `because the index size: ${\n              indices.length} is not the same as tensors size: ${\n              tensors.length}.`);\n    }\n\n    indices.forEach((i, index) => this.write(i, tensors[index]));\n  }\n\n  /**\n   * Return selected values in the TensorArray as a packed Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param [indices] number[] Optional. Taking values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size(). If not specified returns\n   *    all tensors in the original order.\n   * @param [dtype]\n   */\n  gather(indices?: number[], dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but gather requested dtype ${dtype}`);\n    }\n\n    if (!indices) {\n      indices = [];\n      for (let i = 0; i < this.size(); i++) {\n        indices.push(i);\n      }\n    } else {\n      indices = indices.slice(0, this.size());\n    }\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    // Read all the PersistentTensors into a vector to keep track of\n    // their memory.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape, 'TensorArray shape mismatch: ');\n\n    return stack(tensors, 0);\n  }\n\n  /**\n   * Return the values in the TensorArray as a concatenated Tensor.\n   */\n  concat(dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but concat requested dtype ${dtype}`);\n    }\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    const indices = [];\n    for (let i = 0; i < this.size(); i++) {\n      indices.push(i);\n    }\n    // Collect all the tensors from the tensors array.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape,\n        `TensorArray shape mismatch: tensor array shape (${\n            this.elementShape}) vs first tensor shape (${tensors[0].shape})`);\n\n    return concat(tensors, 0);\n  }\n\n  /**\n   * Scatter the values of a Tensor in specific indices of a TensorArray.\n   * @param indices nummber[] values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size().\n   * @param tensor Tensor input tensor.\n   */\n  scatter(indices: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n\n    if (indices.length !== tensor.shape[0]) {\n      throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n          indices.length} vs. ${tensor.shape[0]}`);\n    }\n\n    const maxIndex = Math.max(...indices);\n\n    if (!this.dynamicSize && maxIndex >= this.maxSize) {\n      throw new Error(\n          `Max index must be < array size (${maxIndex}  vs. ${this.maxSize})`);\n    }\n\n    this.writeMany(indices, unstack(tensor, 0));\n  }\n\n  /**\n   * Split the values of a Tensor into the TensorArray.\n   * @param length number[] with the lengths to use when splitting value along\n   *    its first dimension.\n   * @param tensor Tensor, the tensor to split.\n   */\n  split(length: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n    let totalLength = 0;\n    const cumulativeLengths = length.map(len => {\n      totalLength += len;\n      return totalLength;\n    });\n\n    if (totalLength !== tensor.shape[0]) {\n      throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n    }\n\n    if (!this.dynamicSize && length.length !== this.maxSize) {\n      throw new Error(\n          `TensorArray's size is not equal to the size of lengths (${\n              this.maxSize} vs. ${length.length}), ` +\n          'and the TensorArray is not marked as dynamically resizeable');\n    }\n\n    const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n    const tensors: Tensor[] = [];\n    tidy(() => {\n      tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n      for (let i = 0; i < length.length; ++i) {\n        const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n        const indices = [0, previousLength, 0];\n        const sizes = [1, length[i], elementPerRow];\n        tensors[i] = reshape(slice(tensor, indices, sizes), this.elementShape);\n      }\n      return tensors;\n    });\n    const indices = [];\n    for (let i = 0; i < length.length; i++) {\n      indices[i] = i;\n    }\n    this.writeMany(indices, tensors);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\n/**\n * TensorList stores a container of `tf.Tensor` objects, which are accessible\n * via tensors field.\n *\n * In order to get a copy of the underlying list, use the copy method:\n * ```\n *    TensorList b = a.copy();\n *    b.tensors().pushBack(t);  // This does not modify a.tensors().\n * ```\n *\n * Note that this is not a deep copy: the memory locations of the underlying\n * tensors will still point to the same locations of the corresponding tensors\n * in the original.\n */\n\nexport class TensorList {\n  readonly idTensor: Tensor;\n  maxNumElements: number;\n\n  get id() {\n    return this.idTensor.id;\n  }\n  /**\n   *\n   * @param tensors list of tensors\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param maxNumElements The maximum allowed size of `tensors`. Defaults to -1\n   *   meaning that the size of `tensors` is unbounded.\n   */\n  constructor(\n      readonly tensors: Tensor[], readonly elementShape: number[],\n      readonly elementDtype: DataType, maxNumElements = -1) {\n    if (tensors != null) {\n      tensors.forEach(tensor => {\n        if (elementDtype !== tensor.dtype) {\n          throw new Error(`Invalid data types; op elements ${\n              elementDtype}, but list elements ${tensor.dtype}`);\n        }\n        assertShapesMatchAllowUndefinedSize(\n            elementShape, tensor.shape, 'TensorList shape mismatch: ');\n\n        keep(tensor);\n      });\n    }\n    this.idTensor = scalar(0);\n    this.maxNumElements = maxNumElements;\n    keep(this.idTensor);\n  }\n\n  /**\n   * Get a new TensorList containing a copy of the underlying tensor container.\n   */\n  copy(): TensorList {\n    return new TensorList(\n        [...this.tensors], this.elementShape, this.elementDtype);\n  }\n\n  /**\n   * Dispose the tensors and idTensor and clear the tensor list.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.id)) {\n        tensor.dispose();\n      }\n    });\n    this.tensors.length = 0;\n    this.idTensor.dispose();\n  }\n  /**\n   * The size of the tensors in the tensor list.\n   */\n  size() {\n    return this.tensors.length;\n  }\n\n  /**\n   * Return a tensor that stacks a list of rank-R tf.Tensors into one rank-(R+1)\n   * tf.Tensor.\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param numElements the number of elements to stack\n   */\n  stack(elementShape: number[], elementDtype: DataType, numElements = -1):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (numElements !== -1 && this.tensors.length !== numElements) {\n      throw new Error(`Operation expected a list with ${\n          numElements} elements but got a list with ${\n          this.tensors.length} elements.`);\n    }\n    assertShapesMatchAllowUndefinedSize(\n        elementShape, this.elementShape, 'TensorList shape mismatch: ');\n    return tidy(() => {\n      const reshapedTensors =\n          this.tensors.map(tensor => reshape(tensor, elementShape));\n      return stack(reshapedTensors, 0);\n    });\n  }\n\n  /**\n   * Pop a tensor from the end of the list.\n   * @param elementShape shape of the tensor\n   * @param elementDtype data type of the tensor\n   */\n  popBack(elementShape: number[], elementDtype: DataType): Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (this.size() === 0) {\n      throw new Error('Trying to pop from an empty list.');\n    }\n\n    const tensor = this.tensors.pop();\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, elementShape, 'TensorList shape mismatch: ');\n    return reshape(tensor, elementShape);\n  }\n\n  /**\n   * Push a tensor to the end of the list.\n   * @param tensor Tensor to be pushed.\n   */\n  pushBack(tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, this.elementShape, 'TensorList shape mismatch: ');\n\n    if (this.maxNumElements === this.size()) {\n      throw new Error(`Trying to push element into a full list.`);\n    }\n    keep(tensor);\n    this.tensors.push(tensor);\n  }\n\n  /**\n   * Update the size of the list.\n   * @param size the new size of the list.\n   */\n  resize(size: number) {\n    if (size < 0) {\n      throw new Error(\n          `TensorListResize expects size to be non-negative. Got: ${size}`);\n    }\n\n    if (this.maxNumElements !== -1 && size > this.maxNumElements) {\n      throw new Error(`TensorListResize input size ${\n          size} is greater maxNumElement ${this.maxNumElements}.`);\n    }\n    this.tensors.length = size;\n  }\n\n  /**\n   * Retrieve the element at the provided index\n   * @param elementShape shape of the tensor\n   * @param elementDtype dtype of the tensor\n   * @param elementIndex index of the tensor\n   */\n  getItem(elementIndex: number, elementShape: number[], elementDtype: DataType):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (elementIndex < 0 || elementIndex > this.tensors.length) {\n      throw new Error(`Trying to access element ${\n          elementIndex} in a list with ${this.tensors.length} elements.`);\n    }\n\n    if (this.tensors[elementIndex] == null) {\n      throw new Error(`element at index ${elementIndex} is null.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.tensors[elementIndex].shape, elementShape,\n        'TensorList shape mismatch: ');\n\n    return this.tensors[elementIndex];\n  }\n\n  /**\n   * Set the tensor at the index\n   * @param elementIndex index of the tensor\n   * @param tensor the tensor to be inserted into the list\n   */\n  setItem(elementIndex: number, tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (elementIndex < 0 ||\n        this.maxNumElements !== -1 && elementIndex >= this.maxNumElements) {\n      throw new Error(`Trying to set element ${\n          elementIndex} in a list with max ${this.maxNumElements} elements.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape, 'TensorList shape mismatch: ');\n    keep(tensor);\n    this.tensors[elementIndex] = tensor;\n  }\n\n  /**\n   * Return selected values in the TensorList as a stacked Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param indices indices of tensors to gather\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  gather(indices: number[], elementDtype: DataType, elementShape: number[]):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    // When indices is greater than the size of the list, indices beyond the\n    // size of the list are ignored.\n    indices = indices.slice(0, this.size());\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = indices.map(i => reshape(this.tensors[i], elementShape));\n      return stack(tensors, 0);\n    });\n  }\n\n  /**\n   * Return the values in the TensorList as a concatenated Tensor.\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  concat(elementDtype: DataType, elementShape: number[]): Tensor {\n    if (!!elementDtype && elementDtype !== this.elementDtype) {\n      throw new Error(`TensorList dtype is ${\n          this.elementDtype} but concat requested dtype ${elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = this.tensors.map(t => reshape(t, elementShape));\n      return concat(tensors, 0);\n    });\n  }\n}\n\n/**\n * Creates a TensorList which, when stacked, has the value of tensor.\n * @param tensor from tensor\n * @param elementShape output tensor element shape\n */\nexport function fromTensor(\n    tensor: Tensor, elementShape: number[], elementDtype: DataType) {\n  const dtype = tensor.dtype;\n  if (tensor.shape.length < 1) {\n    throw new Error(\n        `Tensor must be at least a vector, but saw shape: ${tensor.shape}`);\n  }\n  if (tensor.dtype !== elementDtype) {\n    throw new Error(`Invalid data types; op elements ${\n        tensor.dtype}, but list elements ${elementDtype}`);\n  }\n  const outputShape = tensor.shape.slice(1);\n  assertShapesMatchAllowUndefinedSize(\n      outputShape, elementShape, 'TensorList shape mismatch: ');\n\n  const tensorList: Tensor[] = unstack(tensor);\n  return new TensorList(tensorList, elementShape, dtype);\n}\n\n/**\n * Return a TensorList of the given size with empty elements.\n * @param elementShape the shape of the future elements of the list\n * @param elementDtype the desired type of elements in the list\n * @param numElements the number of elements to reserve\n */\nexport function reserve(\n    elementShape: number[], elementDtype: DataType, numElements: number) {\n  return new TensorList([], elementShape, elementDtype, numElements);\n}\n\n/**\n * Put tensors at specific indices of a stacked tensor into a TensorList.\n * @param indices list of indices on how to scatter the tensor.\n * @param tensor input tensor.\n * @param elementShape the shape of the future elements of the list\n * @param numElements the number of elements to scatter\n */\nexport function scatter(\n    tensor: Tensor, indices: number[], elementShape: number[],\n    numElements?: number): TensorList {\n  if (indices.length !== tensor.shape[0]) {\n    throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n        indices.length} vs. ${tensor.shape[0]}`);\n  }\n\n  const maxIndex = Math.max(...indices);\n\n  if (numElements != null && numElements !== -1 && maxIndex >= numElements) {\n    throw new Error(\n        `Max index must be < array size (${maxIndex}  vs. ${numElements})`);\n  }\n\n  const list = new TensorList([], elementShape, tensor.dtype, numElements);\n  const tensors = unstack(tensor, 0);\n  indices.forEach((value, index) => {\n    list.setItem(value, tensors[index]);\n  });\n  return list;\n}\n\n/**\n * Split the values of a Tensor into a TensorList.\n * @param length the lengths to use when splitting value along\n *    its first dimension.\n * @param tensor the tensor to split.\n * @param elementShape the shape of the future elements of the list\n */\nexport function split(\n    tensor: Tensor, length: number[], elementShape: number[]) {\n  let totalLength = 0;\n  const cumulativeLengths = length.map(len => {\n    totalLength += len;\n    return totalLength;\n  });\n\n  if (totalLength !== tensor.shape[0]) {\n    throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n  }\n\n  const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n  const tensors: Tensor[] = tidy(() => {\n    const tensors = [];\n    tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n    for (let i = 0; i < length.length; ++i) {\n      const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n      const indices = [0, previousLength, 0];\n      const sizes = [1, length[i], elementPerRow];\n      tensors[i] = reshape(slice(tensor, indices, sizes), elementShape);\n    }\n    tensor.dispose();\n    return tensors;\n  });\n\n  const list = new TensorList([], elementShape, tensor.dtype, length.length);\n\n  for (let i = 0; i < tensors.length; i++) {\n    list.setItem(i, tensors[i]);\n  }\n  return list;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, scalar, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {TensorArray} from '../../executor/tensor_array';\nimport {fromTensor, reserve, scatter, split} from '../../executor/tensor_list';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'If':\n    case 'StatelessIf': {\n      const thenFunc =\n          getParamValue('thenBranch', node, tensorMap, context) as string;\n      const elseFunc =\n          getParamValue('elseBranch', node, tensorMap, context) as string;\n      const cond = getParamValue('cond', node, tensorMap, context) as Tensor;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n      const condValue = await cond.data();\n      if (condValue[0]) {\n        return context.functionMap[thenFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      } else {\n        return context.functionMap[elseFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      }\n    }\n    case 'While':\n    case 'StatelessWhile': {\n      const bodyFunc =\n          getParamValue('body', node, tensorMap, context) as string;\n      const condFunc =\n          getParamValue('cond', node, tensorMap, context) as string;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n\n      // Calculate the condition of the loop\n      const condResult =\n          (await context.functionMap[condFunc].executeFunctionAsync(\n              args, context.tensorArrayMap, context.tensorListMap));\n      const argIds = args.map(tensor => tensor.id);\n      let condValue = await condResult[0].data();\n      // Dispose the intermediate tensors for condition function\n      condResult.forEach(tensor => {\n        if (!tensor.kept && argIds.indexOf(tensor.id) === -1) {\n          tensor.dispose();\n        }\n      });\n\n      let result: Tensor[] = args;\n\n      while (condValue[0]) {\n        // Record the previous result for intermediate tensor tracking\n        const origResult = result;\n        // Execution the body of the loop\n        result = await context.functionMap[bodyFunc].executeFunctionAsync(\n            result, context.tensorArrayMap, context.tensorListMap);\n        const resultIds = result.map(tensor => tensor.id);\n\n        // Dispose the intermediate tensor for body function that is not global\n        // kept, not input/output of the body function\n        origResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n\n        // Recalcuate the condition of the loop using the latest results.\n        const condResult =\n            (await context.functionMap[condFunc].executeFunctionAsync(\n                result, context.tensorArrayMap, context.tensorListMap));\n        condValue = await condResult[0].data();\n        // Dispose the intermediate tensors for condition function\n        condResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n      }\n      return result;\n    }\n    case 'LoopCond': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      return [cloneTensor(pred)];\n    }\n    case 'Switch': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      let data = getParamValue('data', node, tensorMap, context) as Tensor;\n      if (!data.kept) {\n        data = cloneTensor(data);\n      }\n      // Outputs nodes :0 => false, :1 => true\n      return (await pred.data())[0] ? [undefined, data] : [data, undefined];\n    }\n    case 'Merge': {\n      const inputName = node.inputNames.find(\n          name => getTensor(name, tensorMap, context) !== undefined);\n      if (inputName) {\n        const data = getTensor(inputName, tensorMap, context);\n        return [cloneTensor(data)];\n      }\n      return undefined;\n    }\n    case 'Enter': {\n      const frameId =\n          getParamValue('frameName', node, tensorMap, context) as string;\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.enterFrame(frameId);\n      return [cloneTensor(data)];\n    }\n    case 'Exit': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.exitFrame();\n      return [cloneTensor(data)];\n    }\n    case 'NextIteration': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.nextIteration();\n      return [cloneTensor(data)];\n    }\n    case 'TensorArrayV3': {\n      const size = getParamValue('size', node, tensorMap, context) as number;\n      const dtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const dynamicSize =\n          getParamValue('dynamicSize', node, tensorMap, context) as boolean;\n      const clearAfterRead =\n          getParamValue('clearAfterRead', node, tensorMap, context) as boolean;\n      const identicalElementShapes =\n          getParamValue('identicalElementShapes', node, tensorMap, context) as\n          boolean;\n      const name = getParamValue('name', node, tensorMap, context) as string;\n      const tensorArray = new TensorArray(\n          name, dtype, size, elementShape, identicalElementShapes, dynamicSize,\n          clearAfterRead);\n      context.addTensorArray(tensorArray);\n      return [tensorArray.idTensor, scalar(1.0)];\n    }\n    case 'TensorArrayWriteV3': {\n      const id =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const writeTensorArray = context.getTensorArray(id.id);\n      writeTensorArray.write(index, writeTensor);\n      return [writeTensorArray.idTensor];\n    }\n    case 'TensorArrayReadV3': {\n      const readId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const readTensorArray = context.getTensorArray(readId.id);\n      return [readTensorArray.read(readIndex)];\n    }\n    case 'TensorArrayGatherV3': {\n      const gatherId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const gatherDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const gatherTensorArray = context.getTensorArray(gatherId.id);\n      return [gatherTensorArray.gather(gatherIndices, gatherDtype)];\n    }\n    case 'TensorArrayScatterV3': {\n      const scatterId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const scatterTensorArray = context.getTensorArray(scatterId.id);\n      scatterTensorArray.scatter(scatterIndices, scatterTensor);\n      return [scatterTensorArray.idTensor];\n    }\n    case 'TensorArrayConcatV3': {\n      const concatId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const concatTensorArray = context.getTensorArray(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      return [concatTensorArray.concat(concatDtype)];\n    }\n    case 'TensorArraySplitV3': {\n      const splitId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n      const splitTensorArray = context.getTensorArray(splitId.id);\n      splitTensorArray.split(lengths, splitTensor);\n      return [splitTensorArray.idTensor];\n    }\n    case 'TensorArraySizeV3': {\n      const sizeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const sizeTensorArray = context.getTensorArray(sizeId.id);\n      return [scalar(sizeTensorArray.size(), 'int32')];\n    }\n    case 'TensorArrayCloseV3': {\n      const closeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const closeTensorArray = context.getTensorArray(closeId.id);\n      closeTensorArray.clearAndClose();\n      return [closeTensorArray.idTensor];\n    }\n    case 'TensorListSetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.setItem(index, writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.getItem(readIndex, elementShape, elementDType)];\n    }\n    case 'TensorListScatterV2':\n    case 'TensorListScatter': {\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList =\n          scatter(scatterTensor, scatterIndices, elementShape, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListReserve': {\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = reserve(elementShape, elementDtype, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGather': {\n      const gatherId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(gatherId.id);\n      return [tensorList.gather(gatherIndices, elementDtype, elementShape)];\n    }\n    case 'TensorListStack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.stack(elementShape, elementDtype, numElements)];\n    }\n    case 'TensorListFromTensor': {\n      const tensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = fromTensor(tensor, elementShape, elementDtype);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListConcat': {\n      const concatId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      return [tensorList.concat(concatDtype, elementShape)];\n    }\n    case 'TensorListPushBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.pushBack(writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListPopBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.popBack(elementShape, elementDType)];\n    }\n    case 'TensorListSplit': {\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n\n      const tensorList = split(splitTensor, lengths, elementShape);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'control';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rank, Tensor, Tensor3D, Tensor4D, Tensor5D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getPadding, getParamValue} from './utils';\n\nfunction fusedConvAndDepthWiseParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const [extraOp, activationFunc] =\n      (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n  const isBiasAdd = extraOp === 'biasadd';\n  const isPrelu = activationFunc === 'prelu';\n  const isBatchNorm = extraOp === 'fusedbatchnorm';\n\n  const numArgs =\n      (getParamValue('numArgs', node, tensorMap, context) as number);\n  if (isBiasAdd) {\n    if (isPrelu && numArgs !== 2) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu ' +\n          'must have two extra arguments: bias and alpha.');\n    }\n    if (!isPrelu && numArgs !== 1) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd must have ' +\n          'one extra argument: bias.');\n    }\n  }\n  if (isBatchNorm) {\n    throw new Error(\n        'FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.');\n  }\n  const stride = getParamValue('strides', node, tensorMap, context) as number[];\n  const pad = getPadding(node, tensorMap, context);\n  const dataFormat =\n      (getParamValue('dataFormat', node, tensorMap, context) as string)\n          .toUpperCase();\n  const dilations =\n      getParamValue('dilations', node, tensorMap, context) as number[];\n  const [biasArg, preluArg] =\n      getParamValue('args', node, tensorMap, context) as Tensor[];\n\n  return {\n    stride,\n    pad,\n    dataFormat,\n    dilations,\n    biasArg,\n    preluArg,\n    activationFunc\n  };\n}\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Conv1D': {\n          const stride =\n              getParamValue('stride', node, tensorMap, context) as number;\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilation =\n              getParamValue('dilation', node, tensorMap, context) as number;\n          return [tfOps.conv1d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              stride, pad as 'valid' | 'same', dataFormat as 'NWC' | 'NCW',\n              dilation)];\n        }\n        case 'Conv2D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case '_FusedConv2D': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.conv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n\n        case 'FusedDepthwiseConv2dNative': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.depthwiseConv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n        case 'Conv2DBackpropInput':\n        case 'Conv2dTranspose': {\n          const shape = getParamValue(\n                            'outputShape', node, tensorMap,\n                            context) as [number, number, number] |\n              [number, number, number, number];\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          return [tfOps.conv2dTranspose(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              shape, [stride[1], stride[2]], pad as 'valid' | 'same')];\n        }\n        case 'DepthwiseConv2dNative':\n        case 'DepthwiseConv2d': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n\n          return [tfOps.depthwiseConv2d(\n              getParamValue('input', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case 'Conv3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv3d(\n              getParamValue('x', node, tensorMap, context) as Tensor4D |\n                  Tensor<Rank.R5>,\n              getParamValue('filter', node, tensorMap, context) as\n                  Tensor<Rank.R5>,\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same',\n              dataFormat as 'NDHWC' | 'NCDHW',\n              [dilations[1], dilations[2], dilations[3]])];\n        }\n        case 'AvgPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPoolWithArgmax': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n          const includeBatchInIndex =\n              getParamValue('includeBatchInIndex', node, tensorMap, context) as\n              boolean;\n          const {result, indexes} = tfOps.maxPoolWithArgmax(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same', includeBatchInIndex);\n          return [result, indexes];\n        }\n        case 'AvgPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'MaxPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'Dilation2D': {\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n\n          // strides: [1, stride_height, stride_width, 1].\n          const strideHeight = strides[1];\n          const strideWidth = strides[2];\n\n          // dilations: [1, dilation_height, dilation_width, 1].\n          const dilationHeight = dilations[1];\n          const dilationWidth = dilations[2];\n\n          return [tfOps.dilation2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              [strideHeight, strideWidth], pad as 'valid' | 'same',\n              [dilationHeight, dilationWidth], 'NHWC' /* dataFormat */)];\n        }\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'convolution';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor, Tensor1D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Fill': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const dtype =\n              getParamValue('dtype', node, tensorMap, context) as DataType;\n          const value =\n              getParamValue('value', node, tensorMap, context) as number;\n          return [tfOps.fill(shape, value, dtype)];\n        }\n        case 'LinSpace': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const num = getParamValue('num', node, tensorMap, context) as number;\n          return [tfOps.linspace(start, stop, num)];\n        }\n        case 'Multinomial': {\n          const logits =\n              getParamValue('logits', node, tensorMap, context) as Tensor1D;\n          const numSamples =\n              getParamValue('numSamples', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.multinomial(logits, numSamples, seed)];\n        }\n        case 'OneHot': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          const depth =\n              getParamValue('depth', node, tensorMap, context) as number;\n          const onValue =\n              getParamValue('onValue', node, tensorMap, context) as number;\n          const offValue =\n              getParamValue('offValue', node, tensorMap, context) as number;\n          return [tfOps.oneHot(indices, depth, onValue, offValue)];\n        }\n        case 'Ones': {\n          return [tfOps.ones(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'OnesLike': {\n          return [tfOps.onesLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'RandomUniform': {\n          return [tfOps.randomUniform(\n              // tslint:disable-next-line:no-any\n              getParamValue('shape', node, tensorMap, context) as any,\n              getParamValue('minval', node, tensorMap, context) as number,\n              getParamValue('maxval', node, tensorMap, context) as number,\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'Range': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const step =\n              getParamValue('step', node, tensorMap, context) as number;\n          return [tfOps.range(\n              start, stop, step,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32')];\n        }\n        case 'TruncatedNormal': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const mean =\n              getParamValue('mean', node, tensorMap, context) as number;\n          const stdDev =\n              getParamValue('stdDev', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.truncatedNormal(\n              shape, mean, stdDev,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32',\n              seed)];\n        }\n        case 'Zeros': {\n          return [tfOps.zeros(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'ZerosLike': {\n          return [tfOps.zerosLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'creation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nfunction nmsParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const boxes = getParamValue('boxes', node, tensorMap, context) as Tensor;\n  const scores = getParamValue('scores', node, tensorMap, context) as Tensor;\n  const maxOutputSize =\n      getParamValue('maxOutputSize', node, tensorMap, context) as number;\n  const iouThreshold =\n      getParamValue('iouThreshold', node, tensorMap, context) as number;\n  const scoreThreshold =\n      getParamValue('scoreThreshold', node, tensorMap, context) as number;\n  const softNmsSigma =\n      getParamValue('softNmsSigma', node, tensorMap, context) as number;\n\n  return {\n    boxes,\n    scores,\n    maxOutputSize,\n    iouThreshold,\n    scoreThreshold,\n    softNmsSigma\n  };\n}\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'NonMaxSuppressionV5': {\n      const {\n        boxes,\n        scores,\n        maxOutputSize,\n        iouThreshold,\n        scoreThreshold,\n        softNmsSigma\n      } = nmsParams(node, tensorMap, context);\n\n      const result = await tfOps.image.nonMaxSuppressionWithScoreAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, softNmsSigma);\n\n      return [result.selectedIndices, result.selectedScores];\n    }\n    case 'NonMaxSuppressionV4': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      const padToMaxOutputSize =\n          getParamValue('padToMaxOutputSize', node, tensorMap, context) as\n          boolean;\n\n      const result = await tfOps.image.nonMaxSuppressionPaddedAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, padToMaxOutputSize);\n\n      return [result.selectedIndices, result.validOutputs];\n    }\n    case 'NonMaxSuppressionV3':\n    case 'NonMaxSuppressionV2': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      return [await tfOps.image.nonMaxSuppressionAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold)];\n    }\n    case 'Where': {\n      const condition = tfOps.cast(\n          (getParamValue('condition', node, tensorMap, context) as Tensor),\n          'bool');\n      const result = [await tfOps.whereAsync(condition)];\n      condition.dispose();\n      return result;\n    }\n    case 'ListDiff': {\n      return tfOps.setdiff1dAsync(\n          getParamValue('x', node, tensorMap, context) as Tensor,\n          getParamValue('y', node, tensorMap, context) as Tensor);\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'dynamic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'TopKV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const k = getParamValue('k', node, tensorMap, context) as number;\n              const sorted =\n                  getParamValue('sorted', node, tensorMap, context) as boolean;\n              const result = tfOps.topk(x, k, sorted);\n              return [result.values, result.indices];\n            }\n            case 'Unique': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const result = tfOps.unique(x);\n              return [result.values, result.indices];\n            }\n            case 'UniqueV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const axis =\n                  getParamValue('axis', node, tensorMap, context) as number;\n              const result = tfOps.unique(x, axis);\n              return [result.values, result.indices];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'evaluation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Const': {\n          return tensorMap[node.name];\n        }\n        case 'PlaceholderWithDefault':\n          const def =\n              getParamValue('default', node, tensorMap, context) as Tensor;\n          return [getTensor(node.name, tensorMap, context) || def];\n        case 'Placeholder':\n          return [getTensor(node.name, tensorMap, context)];\n        case 'Identity':\n        case 'StopGradient':\n        case 'FakeQuantWithMinMaxVars': {  // This op is currently ignored.\n          const data = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [cloneTensor(data)];\n        }\n        case 'IdentityN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => cloneTensor(t));\n        case 'Snapshot':\n          const snapshot =\n              (getParamValue('x', node, tensorMap, context) as Tensor);\n          return [cloneTensor(snapshot)];\n        case 'Shape':\n          return [tfOps.tensor1d(\n              (getParamValue('x', node, tensorMap, context) as Tensor).shape,\n              'int32')];\n        case 'ShapeN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => tfOps.tensor1d(t.shape));\n        case 'Size':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).size,\n              'int32')];\n        case 'Rank':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).rank,\n              'int32')];\n        case 'NoOp':\n          return [tfOps.scalar(1)];\n        case 'Print':\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const data =\n              getParamValue('data', node, tensorMap, context) as Tensor[];\n          const message =\n              getParamValue('message', node, tensorMap, context) as string;\n          const summarize =\n              getParamValue('summarize', node, tensorMap, context) as number;\n          console.warn(\n              'The graph has a tf.print() operation,' +\n              'usually used for debugging, which slows down performance.');\n          console.log(message);\n          for (let i = 0; i < data.length; i++) {\n            console.log(Array.prototype.slice.call(data[i].dataSync())\n                            .slice(0, summarize));\n          }\n          return [input];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'graph';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ResizeBilinear': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeBilinear(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'ResizeNearestNeighbor': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeNearestNeighbor(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'CropAndResize': {\n          const image =\n              getParamValue('image', node, tensorMap, context) as Tensor;\n          const boxes =\n              getParamValue('boxes', node, tensorMap, context) as Tensor;\n          const boxInd =\n              getParamValue('boxInd', node, tensorMap, context) as Tensor;\n          const cropSize =\n              getParamValue('cropSize', node, tensorMap, context) as number[];\n          const method =\n              getParamValue('method', node, tensorMap, context) as string;\n          const extrapolationValue =\n              getParamValue('extrapolationValue', node, tensorMap, context) as\n              number;\n          return [tfOps.image.cropAndResize(\n              image as Tensor4D, boxes as Tensor2D, boxInd as Tensor1D,\n              cropSize as [number, number], method as 'bilinear' | 'nearest',\n              extrapolationValue)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'image';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Equal': {\n          return [tfOps.equal(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'NotEqual': {\n          return [tfOps.notEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Greater': {\n          return [tfOps.greater(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'GreaterEqual': {\n          return [tfOps.greaterEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Less': {\n          return [tfOps.less(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LessEqual': {\n          return [tfOps.lessEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalAnd': {\n          return [tfOps.logicalAnd(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalNot': {\n          return [tfOps.logicalNot(\n              getParamValue('a', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalOr': {\n          return [tfOps.logicalOr(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Select':\n        case 'SelectV2': {\n          return [tfOps.where(\n              getParamValue('condition', node, tensorMap, context) as Tensor,\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'logical';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BatchMatMul':\n        case 'BatchMatMulV2':\n        case 'MatMul':\n          return [tfOps.matMul(\n              getParamValue('a', node, tensorMap, context) as Tensor2D,\n              getParamValue('b', node, tensorMap, context) as Tensor2D,\n              getParamValue('transposeA', node, tensorMap, context) as boolean,\n              getParamValue('transposeB', node, tensorMap, context) as\n                  boolean)];\n\n        case 'Transpose':\n          return [tfOps.transpose(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('perm', node, tensorMap, context) as number[])];\n\n        case '_FusedMatMul':\n          const [extraOp, activationFunc] =\n              (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n          const isBiasAdd = extraOp === 'biasadd';\n          const isPrelu = activationFunc === 'prelu';\n\n          const numArgs =\n              (getParamValue('numArgs', node, tensorMap, context) as number);\n          if (isBiasAdd) {\n            if (isPrelu && numArgs !== 2) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd and Prelu must have two ' +\n                  'extra arguments: bias and alpha.');\n            }\n            if (!isPrelu && numArgs !== 1) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd must have one extra argument: bias.');\n            }\n          }\n          const [biasArg, preluArg] =\n              getParamValue('args', node, tensorMap, context) as Tensor[];\n          return [tfOps.fused.matMul({\n            a: getParamValue('a', node, tensorMap, context) as Tensor2D,\n            b: getParamValue('b', node, tensorMap, context) as Tensor2D,\n            transposeA: getParamValue('transposeA', node, tensorMap, context) as\n                boolean,\n            transposeB: getParamValue('transposeB', node, tensorMap, context) as\n                boolean,\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'matrices';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'FusedBatchNorm':\n        case 'FusedBatchNormV2': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'FusedBatchNormV3': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'LRN': {\n          return [tfOps.localResponseNormalization(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('radius', node, tensorMap, context) as number,\n              getParamValue('bias', node, tensorMap, context) as number,\n              getParamValue('alpha', node, tensorMap, context) as number,\n              getParamValue('beta', node, tensorMap, context) as number)];\n        }\n        case 'Softmax': {\n          return [tfOps.softmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogSoftmax': {\n          return [tfOps.logSoftmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'SparseToDense': {\n          return [tfOps.sparseToDense(\n              getParamValue('sparseIndices', node, tensorMap, context) as\n                  Tensor,\n              getParamValue('outputShape', node, tensorMap, context) as Tensor,\n              getParamValue('sparseValues', node, tensorMap, context) as\n                  number[],\n              getParamValue('defaultValue', node, tensorMap, context) as\n                  Scalar)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'normalization';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Max': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.max(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Mean': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.mean(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Min': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.min(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Sum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.sum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'All': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.all(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Any': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.any(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'ArgMax': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMax(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'ArgMin': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMin(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Prod': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Cumsum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const exclusive =\n              getParamValue('exclusive', node, tensorMap, context) as boolean;\n          const reverse =\n              getParamValue('reverse', node, tensorMap, context) as boolean;\n          return [tfOps.cumsum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              exclusive, reverse)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'reduction';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ConcatV2':\n        case 'Concat': {\n          const n = getParamValue('n', node, tensorMap, context) as number;\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          let inputs =\n              getParamValue('tensors', node, tensorMap, context) as Tensor[];\n          inputs = inputs.slice(0, n);\n          return [tfOps.concat(inputs, axis)];\n        }\n        case 'GatherV2':\n        case 'Gather': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          return [tfOps.gather(input, tfOps.cast(indices, 'int32'), axis)];\n        }\n        case 'ReverseV2':\n        case 'Reverse': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [tfOps.reverse(input, axis)];\n        }\n        case 'Slice': {\n          // tslint:disable-next-line:no-any\n          const begin = getParamValue('begin', node, tensorMap, context) as any;\n          // tslint:disable-next-line:no-any\n          const size = getParamValue('size', node, tensorMap, context) as any;\n          return [tfOps.slice(\n              getParamValue('x', node, tensorMap, context) as Tensor, begin,\n              size)];\n        }\n        case 'StridedSlice': {\n          const begin =\n              getParamValue('begin', node, tensorMap, context) as number[];\n          const end =\n              getParamValue('end', node, tensorMap, context) as number[];\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const beginMask =\n              getParamValue('beginMask', node, tensorMap, context) as number;\n          const endMask =\n              getParamValue('endMask', node, tensorMap, context) as number;\n          const ellipsisMask =\n              getParamValue('ellipsisMask', node, tensorMap, context) as number;\n          const newAxisMask =\n              getParamValue('newAxisMask', node, tensorMap, context) as number;\n          const shrinkAxisMask =\n              getParamValue('shrinkAxisMask', node, tensorMap, context) as\n              number;\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return [tfOps.stridedSlice(\n              tensor, begin, end, strides, beginMask, endMask, ellipsisMask,\n              newAxisMask, shrinkAxisMask)];\n        }\n        case 'Pack': {\n          return tidy(() => {\n            const axis =\n                getParamValue('axis', node, tensorMap, context) as number;\n            const tensors =\n                getParamValue('tensors', node, tensorMap, context) as Tensor[];\n            // Reshape the tensors to the first tensor's shape if they don't\n            // match.\n            const shape = tensors[0].shape;\n            const squeezedShape = tfOps.squeeze(tensors[0]).shape;\n            const mapped = tensors.map(tensor => {\n              const sameShape = util.arraysEqual(tensor.shape, shape);\n              if (!sameShape &&\n                  !util.arraysEqual(\n                      tfOps.squeeze(tensor).shape, squeezedShape)) {\n                throw new Error('the input tensors shape does not match');\n              }\n              return sameShape ? tensor : tfOps.reshape(tensor, shape);\n            });\n            return [tfOps.stack(mapped, axis)];\n          });\n        }\n        case 'Unpack': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const tensor =\n              getParamValue('tensor', node, tensorMap, context) as Tensor;\n          return tfOps.unstack(tensor, axis);\n        }\n        case 'Tile': {\n          const reps =\n              getParamValue('reps', node, tensorMap, context) as number[];\n          return [tfOps.tile(\n              getParamValue('x', node, tensorMap, context) as Tensor, reps)];\n        }\n        case 'Split':\n        case 'SplitV': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const numOrSizeSplits =\n              getParamValue('numOrSizeSplits', node, tensorMap, context) as\n                  number |\n              number[];\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return tfOps.split(tensor, numOrSizeSplits, axis);\n        }\n        case 'ScatterNd': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          const values =\n              getParamValue('values', node, tensorMap, context) as Tensor;\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          return [tfOps.scatterND(indices, values, shape)];\n        }\n        case 'GatherNd': {\n          const x = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          return [tfOps.gatherND(x, indices)];\n        }\n        case 'SparseToDense': {\n          const indices =\n              getParamValue('sparseIndices', node, tensorMap, context) as\n              Tensor;\n          const shape =\n              getParamValue('outputShape', node, tensorMap, context) as\n              number[];\n          const sparseValues =\n              getParamValue('sparseValues', node, tensorMap, context) as Tensor;\n          const defaultValue =\n              getParamValue('defaultValue', node, tensorMap, context) as Scalar;\n          return [tfOps.sparseToDense(\n              indices, sparseValues, shape,\n              sparseValues.dtype === defaultValue.dtype ?\n                  defaultValue :\n                  tfOps.cast(defaultValue, sparseValues.dtype))];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'slice_join';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'FFT': {\n              return [tfOps.fft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IFFT': {\n              return [tfOps.ifft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'RFFT': {\n              return [tfOps.rfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IRFFT': {\n              return [tfOps.irfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'spectral';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Cast': {\n          return [tfOps.cast(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('dtype', node, tensorMap, context) as 'int32' |\n                  'float32' | 'bool')];\n        }\n        case 'ExpandDims': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.expandDims(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Squeeze': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          return [tfOps.squeeze(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n\n        case 'Reshape': {\n          return [tfOps.reshape(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        case 'PadV2':\n        case 'Pad': {\n          return [tfOps.pad(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('padding', node, tensorMap, context) as\n                  Array<[number, number]>,\n              getParamValue('constantValue', node, tensorMap, context) as\n                  number)];\n        }\n        case 'SpaceToBatchND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const paddings =\n              getParamValue('paddings', node, tensorMap, context) as number[][];\n          return [tfOps.spaceToBatchND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, paddings)];\n        }\n        case 'BatchToSpaceND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const crops =\n              getParamValue('crops', node, tensorMap, context) as number[][];\n          return [tfOps.batchToSpaceND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, crops)];\n        }\n        case 'DepthToSpace': {\n          const blockSize =\n              getParamValue('blockSize', node, tensorMap, context) as number;\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as\n               string).toUpperCase() as 'NHWC' |\n              'NCHW';\n          return [tfOps.depthToSpace(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              blockSize, dataFormat)];\n        }\n        case 'BroadcastTo': {\n          return [tfOps.broadcastTo(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'transformation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {ExecutionContext} from '../executor/execution_context';\n\nimport {NodeValueImpl} from './custom_op/node_value_impl';\nimport {getRegisteredOp} from './custom_op/register';\nimport * as arithmetic from './executors/arithmetic_executor';\nimport * as basicMath from './executors/basic_math_executor';\nimport * as control from './executors/control_executor';\nimport * as convolution from './executors/convolution_executor';\nimport * as creation from './executors/creation_executor';\nimport * as dynamic from './executors/dynamic_executor';\nimport * as evaluation from './executors/evaluation_executor';\nimport * as graph from './executors/graph_executor';\nimport * as image from './executors/image_executor';\nimport * as logical from './executors/logical_executor';\nimport * as matrices from './executors/matrices_executor';\nimport * as normalization from './executors/normalization_executor';\nimport * as reduction from './executors/reduction_executor';\nimport * as sliceJoin from './executors/slice_join_executor';\nimport * as spectral from './executors/spectral_executor';\nimport * as transformation from './executors/transformation_executor';\nimport {Node} from './types';\n\n/**\n * Executes the op defined by the node object.\n * @param node\n * @param tensorMap contains tensors for executed nodes and weights\n */\nexport function executeOp(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): tfc.Tensor[]|Promise<tfc.Tensor[]> {\n  const value =\n      ((node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) => {\n        switch (node.category) {\n          case 'arithmetic':\n            return tfc.tidy(\n                () => arithmetic.executeOp(node, tensorMap, context));\n          case 'basic_math':\n            return tfc.tidy(\n                () => basicMath.executeOp(node, tensorMap, context));\n          case 'control':\n            return control.executeOp(node, tensorMap, context);\n          case 'convolution':\n            return tfc.tidy(\n                () => convolution.executeOp(node, tensorMap, context));\n          case 'creation':\n            return tfc.tidy(() => creation.executeOp(node, tensorMap, context));\n          case 'dynamic':\n            return dynamic.executeOp(node, tensorMap, context);\n          case 'evaluation':\n            return tfc.tidy(\n                () => evaluation.executeOp(node, tensorMap, context));\n          case 'image':\n            return tfc.tidy(() => image.executeOp(node, tensorMap, context));\n          case 'graph':\n            return tfc.tidy(() => graph.executeOp(node, tensorMap, context));\n          case 'logical':\n            return tfc.tidy(() => logical.executeOp(node, tensorMap, context));\n          case 'matrices':\n            return tfc.tidy(() => matrices.executeOp(node, tensorMap, context));\n          case 'normalization':\n            return tfc.tidy(\n                () => normalization.executeOp(node, tensorMap, context));\n          case 'reduction':\n            return tfc.tidy(\n                () => reduction.executeOp(node, tensorMap, context));\n          case 'slice_join':\n            return tfc.tidy(\n                () => sliceJoin.executeOp(node, tensorMap, context));\n          case 'spectral':\n            return tfc.tidy(() => spectral.executeOp(node, tensorMap, context));\n          case 'transformation':\n            return tfc.tidy(\n                () => transformation.executeOp(node, tensorMap, context));\n          case 'custom':\n            const opMapper = getRegisteredOp(node.op);\n            if (opMapper && opMapper.customExecutor) {\n              return opMapper.customExecutor(\n                  new NodeValueImpl(node, tensorMap, context));\n            } else {\n              throw TypeError(`Custom op ${node.op} is not registered.`);\n            }\n          default:\n            throw TypeError(\n                `Unknown op '${node.op}'. File an issue at ` +\n                `https://github.com/tensorflow/tfjs/issues so we can add it` +\n                `, or register a custom execution with tf.registerOp()`);\n        }\n      })(node, tensorMap, context);\n  if (value instanceof Promise) {\n    return value.then((data) => [].concat(data));\n  }\n  return [].concat(value);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap, TensorArrayMap, TensorListMap} from '../data/types';\n\nimport {TensorArray} from './tensor_array';\nimport {TensorList} from './tensor_list';\nimport {FunctionExecutor} from './types';\n\nexport interface ExecutionContextInfo {\n  id: number;           // the unique id of the context info\n  frameName: string;    // The frame name of the loop, this comes from\n                        // the TensorFlow NodeDef.\n  iterationId: number;  // The iteration id of the loop\n}\n\n/**\n * ExecutionContext captures the runtime environment of the node. It keeps\n * track of the current frame and iteration for the control flow ops.\n *\n * For example, typical Dynamic RNN model may contain loops, for which\n * TensorFlow will generate graphs with Enter/Exit nodes to control the\n * current execution frame, and NextIteration Nodes for iteration id increment.\n * For model with branch logic, TensorFLow will generate Switch/Merge ops.\n */\nexport class ExecutionContext {\n  private rootContext = {id: 0, frameName: '', iterationId: 0};\n  private contexts: ExecutionContextInfo[] = [this.rootContext];\n  private lastId = 0;\n  private _currentContextIds: string[];\n\n  constructor(\n      readonly weightMap: NamedTensorsMap = {},\n      readonly tensorArrayMap: TensorArrayMap = {},\n      readonly tensorListMap: TensorListMap = {},\n      readonly functionMap: {[key: string]: FunctionExecutor} = {}) {\n    this.generateCurrentContextIds();\n  }\n\n  private newFrame(id: number, frameName: string) {\n    return {id, frameName, iterationId: 0};\n  }\n\n  /**\n   * Set the current context\n   * @param contexts: ExecutionContextInfo[] the current path of execution\n   * frames\n   */\n  set currentContext(contexts: ExecutionContextInfo[]) {\n    if (this.contexts !== contexts) {\n      this.contexts = contexts;\n      this.generateCurrentContextIds();\n    }\n  }\n\n  get currentContext(): ExecutionContextInfo[] {\n    return this.contexts;\n  }\n\n  /**\n   * Returns the current context in string format.\n   */\n  get currentContextId(): string {\n    return this._currentContextIds[0];\n  }\n\n  /**\n   * Returns the current context and all parent contexts in string format.\n   * This allow access to the nodes in the current and parent frames.\n   */\n  get currentContextIds(): string[] {\n    return this._currentContextIds;\n  }\n\n  private generateCurrentContextIds() {\n    const names = [];\n    for (let i = 0; i < this.contexts.length - 1; i++) {\n      const contexts = this.contexts.slice(0, this.contexts.length - i);\n      names.push(this.contextIdforContexts(contexts));\n    }\n    names.push('');\n    this._currentContextIds = names;\n  }\n\n  private contextIdforContexts(contexts: ExecutionContextInfo[]) {\n    return contexts ?\n        contexts\n            .map(\n                context => (context.id === 0 && context.iterationId === 0) ?\n                    '' :\n                    `${context.frameName}-${context.iterationId}`)\n            .join('/') :\n        '';\n  }\n\n  /**\n   * Enter a new frame, a new context is pushed on the current context list.\n   * @param frameId new frame id\n   */\n  enterFrame(frameId: string) {\n    if (this.contexts) {\n      this.lastId++;\n      this.contexts = this.contexts.slice();\n      this.contexts.push(this.newFrame(this.lastId, frameId));\n      this._currentContextIds.unshift(this.contextIdforContexts(this.contexts));\n    }\n  }\n\n  /**\n   * Exit the current frame, the last context is removed from the current\n   * context list.\n   */\n  exitFrame() {\n    if (this.contexts && this.contexts.length > 1) {\n      this.contexts = this.contexts.slice();\n      this.contexts.splice(-1);\n      this.currentContextIds.shift();\n    } else {\n      throw new Error('Cannot exit frame, the context is empty');\n    }\n  }\n\n  /**\n   * Enter the next iteration of a loop, the iteration id of last context is\n   * increased.\n   */\n  nextIteration() {\n    if (this.contexts && this.contexts.length > 0) {\n      this.contexts = this.contexts.slice();\n      this.lastId++;\n      const context =\n          Object.assign({}, this.contexts[this.contexts.length - 1]);\n      context.iterationId += 1;\n      context.id = this.lastId;\n      this.contexts.splice(-1, 1, context);\n      this._currentContextIds.splice(\n          0, 1, this.contextIdforContexts(this.contexts));\n    } else {\n      throw new Error('Cannot increase frame iteration, the context is empty');\n    }\n  }\n\n  getWeight(name: string): Tensor[] {\n    return this.weightMap[name];\n  }\n\n  addTensorArray(tensorArray: TensorArray) {\n    this.tensorArrayMap[tensorArray.id] = tensorArray;\n  }\n\n  getTensorArray(id: number): TensorArray {\n    return this.tensorArrayMap[id];\n  }\n\n  addTensorList(tensorList: TensorList) {\n    this.tensorListMap[tensorList.id] = tensorList;\n  }\n\n  getTensorList(id: number): TensorList {\n    return this.tensorListMap[id];\n  }\n\n  dispose(keepIds: Set<number>) {\n    for (const key in this.tensorArrayMap) {\n      this.tensorArrayMap[key].clearAndClose(keepIds);\n    }\n\n    for (const key in this.tensorListMap) {\n      this.tensorListMap[key].clearAndClose(keepIds);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NamedTensorMap} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {parseNodeName} from '../operations/executors/utils';\nimport {Graph, Node} from '../operations/types';\n\nexport interface ExecutionInfo {\n  inputs: NamedTensorMap;\n  outputs: Node[];\n  usedNodes: Set<string>;\n  missingInputs: string[];\n  dynamicNode: Node;\n  syncInputs: string[];\n}\n\n/**\n * Given graph inputs and desired outputs, find the minimal set of nodes\n * to execute in order to compute the outputs. In addition return other useful\n * info such:\n * - Missing inputs needed to compute the output.\n * - Whether the subgraph contains dynamic ops (control flow, dynamic shape).\n * - Alternative inputs in order to avoid async (dynamic op) execution.\n */\nexport function getExecutionSubgraph(\n    inputs: NamedTensorMap, outputs: Node[], weightMap: NamedTensorsMap,\n    initNodes?: Node[]): ExecutionInfo {\n  const usedNodes = new Set<string>();\n  const missingInputs: string[] = [];\n  let dynamicNode: Node = null;\n  let syncInputs: string[] = null;\n\n  // Start with the outputs, going backwards and find all the nodes that are\n  // needed to compute those outputs.\n  const seen = new Set<string>();\n  const inputNodeNames =\n      Object.keys(inputs).map(name => parseNodeName(name)[0]);\n\n  let initNodeNames: string[] = [];\n  if (initNodes != null) {\n    initNodeNames = initNodes.map(node => parseNodeName(node.name)[0]);\n  }\n\n  const frontier = [...outputs];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    if (isControlFlow(node) || isDynamicShape(node)) {\n      if (dynamicNode == null) {\n        dynamicNode = node;\n        syncInputs = dynamicNode.children.map(child => child.name)\n                         .filter(name => usedNodes.has(name));\n      }\n    }\n    usedNodes.add(node.name);\n\n    // Weights are dead end since we already have their values.\n    if (weightMap[node.name] != null) {\n      continue;\n    }\n    // This node is a dead end since it's one of the user-provided inputs.\n    if (inputNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    // This node is a dead end since it doesn't have any inputs.\n    if (initNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    if (node.inputs.length === 0) {\n      missingInputs.push(node.name);\n      continue;\n    }\n    node.inputs.forEach(input => {\n      // Don't add to the frontier if it is already there.\n      if (seen.has(input.name)) {\n        return;\n      }\n      seen.add(input.name);\n      frontier.push(input);\n    });\n  }\n  return {inputs, outputs, usedNodes, missingInputs, dynamicNode, syncInputs};\n}\n\n/**\n * Given the execution info, return a list of nodes in topological order that\n * need to be executed to compute the output.\n */\nexport function getNodesInTopologicalOrder(\n    graph: Graph, weightMap: NamedTensorsMap,\n    executionInfo: ExecutionInfo): Node[] {\n  const {usedNodes, inputs} = executionInfo;\n  const frontier: Node[] = [];\n  const inputNodes = Object.keys(inputs)\n                         .map(name => parseNodeName(name)[0])\n                         .map(name => graph.nodes[name]);\n  const initNodes = graph.initNodes;\n\n  inputNodes.forEach(input => {\n    if (usedNodes.has(input.name)) {\n      frontier.push(input);\n    }\n  });\n  graph.weights.forEach(weight => {\n    if (usedNodes.has(weight.name)) {\n      frontier.push(weight);\n    }\n  });\n  if (initNodes != null) {\n    initNodes.forEach(node => {\n      if (usedNodes.has(node.name)) {\n        frontier.push(node);\n      }\n    });\n  }\n  const seen = new Set<string>();\n  const orderedNodes: Node[] = [];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    seen.add(node.name);\n    if (!weightMap[node.name]) {\n      orderedNodes.push(node);\n    }\n    node.children.forEach(child => {\n      if (!seen.has(child.name) && usedNodes.has(child.name) &&\n          child.inputs.every(input => seen.has(input.name))) {\n        frontier.push(child);\n      }\n    });\n  }\n  return orderedNodes;\n}\n\nconst CONTROL_FLOW_OPS = [\n  'Switch', 'Merge', 'Enter', 'Exit', 'NextIteration', 'StatelessIf',\n  'StatelessWhile', 'if', 'While'\n];\nconst DYNAMIC_SHAPE_OPS = [\n  'NonMaxSuppressionV2', 'NonMaxSuppressionV3', 'NonMaxSuppressionV5', 'Where'\n];\n\nexport function isControlFlow(node: Node) {\n  return CONTROL_FLOW_OPS.indexOf(node.op) >= 0;\n}\n\nexport function isDynamicShape(node: Node) {\n  return DYNAMIC_SHAPE_OPS.indexOf(node.op) >= 0;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NamedTensorMap, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {ISignatureDef} from '../data/compiled_api';\nimport {NamedTensorsMap, TensorArrayMap, TensorInfo, TensorListMap} from '../data/types';\nimport {getNodeNameAndIndex, getParamValue, getTensor, getTensorsForCurrentContenxt, parseNodeName} from '../operations/executors/utils';\nimport {executeOp} from '../operations/operation_executor';\nimport {Graph, Node} from '../operations/types';\n\nimport {ExecutionContext, ExecutionContextInfo} from './execution_context';\nimport {getExecutionSubgraph, getNodesInTopologicalOrder, isControlFlow} from './model_analysis';\nimport {FunctionExecutor} from './types';\n\ninterface NodeWithContexts {\n  contexts: ExecutionContextInfo[];\n  node: Node;\n}\n\nexport class GraphExecutor implements FunctionExecutor {\n  private compiledMap: Map<string, Node[]> = new Map();\n  private _weightMap: NamedTensorsMap = {};\n  private _weightIds: number[];\n  private _signature: ISignatureDef;\n  private _inputs: Node[];\n  private _outputs: Node[];\n  private _initNodes: Node[];\n  private SEPERATOR = ',';\n  private _functions: {[key: string]: Graph} = {};\n  private _functionExecutorMap: {[key: string]: FunctionExecutor} = {};\n\n  get weightIds(): number[] {\n    return this.parent ? this.parent.weightIds : this._weightIds;\n  }\n\n  get functionExecutorMap(): {[key: string]: FunctionExecutor} {\n    return this.parent ? this.parent.functionExecutorMap :\n                         this._functionExecutorMap;\n  }\n\n  get weightMap(): NamedTensorsMap {\n    return this.parent ? this.parent.weightMap : this._weightMap;\n  }\n\n  set weightMap(weightMap: NamedTensorsMap) {\n    const weightIds = Object.keys(weightMap).map(\n        key => weightMap[key].map(tensor => tensor.id));\n    this._weightIds = [].concat(...weightIds);\n    this._weightMap = weightMap;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this._inputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get outputs(): TensorInfo[] {\n    return this._outputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get inputNodes(): string[] {\n    return this._inputs.map(node => node.signatureKey || node.name);\n  }\n\n  get outputNodes(): string[] {\n    return this._outputs.map((node) => {\n      const name = node.signatureKey || node.name;\n      return node.defaultOutput ? (`${name}:${node.defaultOutput}`) : name;\n    });\n  }\n\n  get functions(): {[key: string]: ISignatureDef} {\n    return Object.keys(this._functions).reduce((map, key) => {\n      map[key] = this._functions[key].signature;\n      return map;\n    }, {} as {[key: string]: ISignatureDef});\n  }\n\n  /**\n   *\n   * @param graph Graph the model or function graph to be executed.\n   * @param parent When building function exector you need to set the parent\n   * executor. Since the weights and function executor maps are set at parant\n   * level, that function executor can access the function maps and weight maps\n   * through the parent.\n   */\n  constructor(private graph: Graph, private parent?: GraphExecutor) {\n    this._outputs = graph.outputs;\n    this._inputs = graph.inputs;\n    this._initNodes = graph.initNodes;\n    this._signature = graph.signature;\n    this._functions = graph.functions;\n    // create sub-graph executors\n    if (graph.functions != null) {\n      Object.keys(graph.functions).forEach(name => {\n        this._functionExecutorMap[name] =\n            new GraphExecutor(graph.functions[name], this);\n      });\n    }\n  }\n\n  private getCompilationKey(inputs: Node[], outputs: Node[]): string {\n    const sortedInputs = inputs.map(node => node.name).sort();\n    const sortedOutputs = outputs.map(node => node.name).sort();\n    return sortedInputs.join(this.SEPERATOR) + '--' +\n        sortedOutputs.join(this.SEPERATOR);\n  }\n\n  /**\n   * Compiles the inference graph and returns the minimal set of nodes that are\n   * required for execution, in the correct execution order.\n   */\n  private compile(inputs: NamedTensorMap, outputs: Node[]): Node[] {\n    const executionInfo =\n        getExecutionSubgraph(inputs, outputs, this.weightMap, this._initNodes);\n    const {missingInputs, dynamicNode, syncInputs} = executionInfo;\n    if (dynamicNode != null) {\n      throw new Error(\n          `This execution contains the node '${dynamicNode.name}', which has ` +\n          `the dynamic op '${dynamicNode.op}'. Please use ` +\n          `model.executeAsync() instead. Alternatively, to avoid the ` +\n          `dynamic ops, specify the inputs [${syncInputs}]`);\n    }\n\n    if (missingInputs.length > 0) {\n      const outNames = outputs.map(n => n.name);\n      const inNames = Object.keys(inputs);\n      throw new Error(\n          `Cannot compute the outputs [${outNames}] from the provided inputs ` +\n          `[${inNames}]. Missing the following inputs: [${missingInputs}]`);\n    }\n\n    return getNodesInTopologicalOrder(\n        this.graph, this.weightMap, executionInfo);\n  }\n\n  /**\n   * Executes the inference for given input tensors.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  execute(inputs: NamedTensorMap, outputs: string[]): Tensor[] {\n    inputs = this.mapInputs(inputs);\n    const names = Object.keys(inputs).sort();\n    this.checkInputs(inputs);\n    this.checkInputShapeAndType(inputs);\n    outputs = this.mapOutputs(outputs);\n    this.checkOutputs(outputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputs.map(name => parseNodeName(name)[0]);\n    let outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n\n    // If no outputs are specified, then use the default outputs of the model.\n    if (outputNodes.length === 0) {\n      outputNodes = this._outputs;\n    }\n\n    const compilationKey = this.getCompilationKey(inputNodes, outputNodes);\n\n    // Do nothing if the compiled graph cache contains the input.\n    let orderedNodes = this.compiledMap.get(compilationKey);\n    if (orderedNodes == null) {\n      orderedNodes = this.compile(inputs, outputNodes);\n      this.compiledMap.set(compilationKey, orderedNodes);\n    }\n\n    const tensorArrayMap: TensorArrayMap = {};\n    const tensorListMap: TensorListMap = {};\n    return tidy(() => {\n      const context = new ExecutionContext(\n          this.weightMap, tensorArrayMap, tensorListMap,\n          this.functionExecutorMap);\n      const tensorsMap: NamedTensorsMap = {...this.weightMap};\n\n      Object.keys(inputs).forEach(name => {\n        const [nodeName, index] = parseNodeName(name);\n        const tensors: Tensor[] = [];\n        tensors[index] = inputs[name];\n        tensorsMap[nodeName] = tensors;\n      });\n\n      const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n      const intermediateTensorConsumerCount: {[key: number]: number} = {};\n      for (let i = 0; i < orderedNodes.length; i++) {\n        const node = orderedNodes[i];\n        if (!tensorsMap[node.name]) {\n          const tensors = executeOp(node, tensorsMap, context) as Tensor[];\n          if (tensors instanceof Promise) {\n            throw new Error(\n                `The execution of the op '${node.op}' returned a promise. ` +\n                `Please use model.executeAsync() instead.`);\n          }\n          tensorsMap[node.name] = tensors;\n          this.checkTensorForDisposal(\n              node.name, node, tensorsMap, context, tensorsToKeep,\n              outputNodeNames, intermediateTensorConsumerCount);\n        }\n      }\n      // dispose the context for the root executor\n      if (this.parent == null) {\n        context.dispose(tensorsToKeep);\n      }\n      return outputs.map(name => getTensor(name, tensorsMap, context));\n    });\n  }\n\n  private getFrozenTensorIds(tensorMap: NamedTensorsMap): Set<number> {\n    const ids = [].concat.apply(\n        [],\n        Object.keys(tensorMap)\n            .map(key => tensorMap[key])\n            .map(tensors => tensors.map(tensor => tensor.id)));\n    return new Set(ids);\n  }\n  private checkTensorForDisposal(\n      nodeName: string, node: Node, tensorMap: NamedTensorsMap,\n      context: ExecutionContext, tensorsToKeep: Set<number>,\n      outputNames: string[],\n      intermediateTensorConsumerCount: {[key: string]: number}) {\n    // Skip output nodes and any control flow nodes, since its dependency is\n    // tricky to track correctly.\n    if (node.category === 'control' || outputNames.indexOf(nodeName) !== -1) {\n      return;\n    }\n\n    tensorMap[nodeName].forEach(tensor => {\n      if (tensor != null) {\n        intermediateTensorConsumerCount[tensor.id] =\n            (intermediateTensorConsumerCount[tensor.id] || 0) +\n            node.children.length;\n      }\n    });\n    node.inputs.forEach(input => {\n      // Skip any control flow nodes, since its dependency is tricky to track\n      // correctly.\n      if (input.category !== 'control') {\n        const tensors =\n            getTensorsForCurrentContenxt(input.name, tensorMap, context);\n        if (tensors != null) {\n          tensors.forEach(tensor => {\n            if (tensor && !tensorsToKeep.has(tensor.id)) {\n              const count = intermediateTensorConsumerCount[tensor.id];\n              if (count === 1) {\n                tensor.dispose();\n                delete intermediateTensorConsumerCount[tensor.id];\n              } else if (count != null) {\n                // only intermediate nodes has count set, inputs and weights are\n                // not.\n                intermediateTensorConsumerCount[tensor.id]--;\n              }\n            }\n          });\n        }\n      }\n    });\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  async executeAsync(inputs: NamedTensorMap, outputs: string[]):\n      Promise<Tensor[]> {\n    return this._executeAsync(inputs, outputs);\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   * @param isFunctionExecution Flag for executing a function.\n   * @param tensorArrayMap Optional, global TensorArray map by id. Used for\n   * function execution.\n   * @param tensorArrayMap Optinal global TensorList map by id. Used for\n   * function execution.\n   */\n  private async _executeAsync(\n      inputs: NamedTensorMap, outputs: string[], isFunctionExecution = false,\n      tensorArrayMap: TensorArrayMap = {},\n      tensorListMap: TensorListMap = {}): Promise<Tensor[]> {\n    if (!isFunctionExecution) {\n      inputs = this.mapInputs(inputs);\n      this.checkInputs(inputs);\n      this.checkInputShapeAndType(inputs);\n      outputs = this.mapOutputs(outputs);\n      this.checkOutputs(outputs);\n    }\n\n    const context = new ExecutionContext(\n        this.weightMap, tensorArrayMap, tensorListMap,\n        this.functionExecutorMap);\n\n    // Graph with control flow op requires runtime evaluation of the execution\n    // order, while without control flow the execution order is pre-determined\n    // in the compile method.\n    const tensorMap = await this.executeWithControlFlow(\n        inputs, context, outputs, isFunctionExecution);\n    const results = outputs.map(name => getTensor(name, tensorMap, context));\n\n    // dispose all the intermediate tensors\n    const outputIds = results.map(t => t.id);\n    const inputIds = Object.keys(inputs).map(name => inputs[name].id);\n    const keepIds =\n        new Set<number>([...outputIds, ...inputIds, ...this.weightIds]);\n    Object.keys(tensorMap).forEach(key => {\n      const tensorArray = tensorMap[key];\n      tensorArray.forEach(tensor => {\n        if (tensor && !tensor.isDisposed && !keepIds.has(tensor.id)) {\n          tensor.dispose();\n        }\n      });\n    });\n    // dispose the context for the root executor\n    if (this.parent == null) {\n      context.dispose(keepIds);\n    }\n\n    return results;\n  }\n\n  async executeFunctionAsync(\n      inputs: Tensor[], tensorArrayMap: TensorArrayMap,\n      tensorListMap: TensorListMap): Promise<Tensor[]> {\n    const mappedInputs = inputs.reduce((map, tensor, index) => {\n      map[this.inputs[index].name] = tensor;\n      return map;\n    }, {} as NamedTensorMap);\n\n    return this._executeAsync(\n        mappedInputs, this.outputNodes, true, tensorArrayMap, tensorListMap);\n  }\n  /**\n   * When there are control flow nodes in the graph, the graph execution use\n   * ExecutionContext to keep track of the frames and loop iterators.\n   * @param inputs placeholder tensors for the graph.\n   * @param context the execution context object for current execution.\n   * @param isFunctionExecution Flag for executing a function.\n   */\n  private async executeWithControlFlow(\n      inputs: NamedTensorMap, context: ExecutionContext, outputNames: string[],\n      isFunctionExecution: boolean): Promise<NamedTensorsMap> {\n    const names = Object.keys(inputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputNames.map(name => parseNodeName(name)[0]);\n    const outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n    const {usedNodes, missingInputs, dynamicNode, syncInputs} =\n        getExecutionSubgraph(inputs, outputNodes, this.weightMap);\n\n    const stack: NodeWithContexts[] =\n        [...inputNodes, ...this.graph.weights].map(node => {\n          return {node, contexts: context.currentContext};\n        });\n    const tensorsMap: NamedTensorsMap = {...this.weightMap};\n    Object.keys(inputs).forEach(name => {\n      const [nodeName, index] = parseNodeName(name);\n      const tensors: Tensor[] = [];\n      tensors[index] = inputs[name];\n      tensorsMap[nodeName] = tensors;\n    });\n    const intermediateTensorConsumerCount: {[key: number]: number} = {};\n    const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n    const added: {[key: string]: boolean} = {};\n    while (stack.length > 0) {\n      const promises = this.processStack(\n          inputNodes, stack, context, tensorsMap, added, tensorsToKeep,\n          outputNodeNames, intermediateTensorConsumerCount, usedNodes);\n      await Promise.all(promises);\n    }\n    if (dynamicNode == null && !isFunctionExecution) {\n      console.warn(\n          `This model execution did not contain any nodes with control flow ` +\n          `or dynamic output shapes. You can use model.execute() instead.`);\n    }\n    const missingOutputs =\n        outputNodes\n            .filter(\n                node => !isControlFlow(node) &&\n                    !getTensor(node.name, tensorsMap, context))\n            .map(node => node.name);\n    if (missingOutputs.length > 0) {\n      let alternativeMsg = '';\n      if (dynamicNode != null) {\n        alternativeMsg =\n            `Alternatively, to avoid the dynamic ops, use model.execute() ` +\n            `and specify the inputs [${syncInputs}]`;\n      }\n      throw new Error(\n          `Cannot compute the outputs [${missingOutputs}] from the provided ` +\n          `inputs [${names}]. Consider providing the following inputs: ` +\n          `[${missingInputs}]. ${alternativeMsg}`);\n    }\n    return tensorsMap;\n  }\n\n  private processStack(\n      inputNodes: Node[], stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      tensorsToKeep: Set<number>, outputNames: string[],\n      intermediateTensorConsumerCount: {[key: number]: number},\n      usedNodes: Set<string>) {\n    const promises: Array<Promise<Tensor[]>> = [];\n    while (stack.length > 0) {\n      const item = stack.pop();\n      context.currentContext = item.contexts;\n      let nodeName = '';\n      // The tensor of the Enter op with isConstant set should be set\n      // in the parent scope, so it will be available as constant for the\n      // whole loop.\n      if (item.node.op === 'Enter' &&\n          getParamValue('isConstant', item.node, tensorMap, context)) {\n        [nodeName] = getNodeNameAndIndex(item.node.name, context);\n      }\n\n      // only process nodes that are not provided as input nodes.\n      if (inputNodes.indexOf(item.node) === -1) {\n        const tensors = executeOp(item.node, tensorMap, context);\n        if (!nodeName) {\n          [nodeName] = getNodeNameAndIndex(item.node.name, context);\n        }\n        const currentContext = context.currentContext;\n        if (tensors instanceof Promise) {\n          promises.push(tensors.then(t => {\n            tensorMap[nodeName] = t;\n            context.currentContext = currentContext;\n            this.checkTensorForDisposal(\n                nodeName, item.node, tensorMap, context, tensorsToKeep,\n                outputNames, intermediateTensorConsumerCount);\n            this.processChildNodes(\n                item.node, stack, context, tensorMap, added, usedNodes);\n            return t;\n          }));\n        } else {\n          tensorMap[nodeName] = tensors;\n          this.checkTensorForDisposal(\n              nodeName, item.node, tensorMap, context, tensorsToKeep,\n              outputNames, intermediateTensorConsumerCount);\n          this.processChildNodes(\n              item.node, stack, context, tensorMap, added, usedNodes);\n        }\n      } else {\n        this.processChildNodes(\n            item.node, stack, context, tensorMap, added, usedNodes);\n      }\n    }\n    return promises;\n  }\n\n  private processChildNodes(\n      node: Node, stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      usedNodes: Set<string>) {\n    node.children.forEach((childNode) => {\n      const [nodeName, ] = getNodeNameAndIndex(childNode.name, context);\n      if (added[nodeName] || !usedNodes.has(childNode.name)) {\n        return;\n      }\n      // Merge op can be pushed if any of its inputs has value.\n      if (childNode.op === 'Merge') {\n        if (childNode.inputNames.some(name => {\n              return !!getTensor(name, tensorMap, context);\n            })) {\n          added[nodeName] = true;\n          stack.push({contexts: context.currentContext, node: childNode});\n        }\n      } else  // Otherwise all inputs must to have value.\n          if (childNode.inputNames.every(name => {\n                return !!getTensor(name, tensorMap, context);\n              })) {\n        added[nodeName] = true;\n        stack.push({contexts: context.currentContext, node: childNode});\n      }\n    });\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   */\n  dispose() {\n    Object.keys(this.weightMap)\n        .forEach(\n            key => this.weightMap[key].forEach(tensor => tensor.dispose()));\n  }\n\n  private checkInputShapeAndType(inputs: NamedTensorMap) {\n    Object.keys(inputs).forEach(name => {\n      const input = inputs[name];\n      const [nodeName, ] = parseNodeName(name);\n      const node = this.graph.nodes[nodeName];\n      if (node.attrParams['shape'] && node.attrParams['shape'].value) {\n        const shape = node.attrParams['shape'].value as number[];\n        const match = shape.length === input.shape.length &&\n            input.shape.every(\n                (dim, index) => shape[index] === -1 || shape[index] === dim);\n        util.assert(\n            match,\n            () => `The shape of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be [${shape}], but was ` +\n                `[${input.shape}]`);\n      }\n      if (node.attrParams['dtype'] && node.attrParams['dtype'].value) {\n        util.assert(\n            input.dtype === node.attrParams['dtype'].value as string,\n            () => `The dtype of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be ` +\n                `${node.attrParams['dtype'].value}, but was ${input.dtype}`);\n      }\n    });\n  }\n\n  private mapInputs(inputs: NamedTensorMap) {\n    const result: NamedTensorMap = {};\n    for (const inputName in inputs) {\n      if (this._signature != null && this._signature.inputs != null &&\n          this._signature.inputs[inputName] != null) {\n        const tensor = this._signature.inputs[inputName];\n        result[tensor.name] = inputs[inputName];\n      } else {\n        result[inputName] = inputs[inputName];\n      }\n    }\n    return result;\n  }\n\n  private checkInputs(inputs: NamedTensorMap) {\n    const notInGraph = Object.keys(inputs).filter(name => {\n      const [nodeName] = parseNodeName(name);\n      return this.graph.nodes[nodeName] == null;\n    });\n    if (notInGraph.length > 0) {\n      throw new Error(\n          `The dict provided in model.execute(dict) has ` +\n          `keys: [${notInGraph}] that are not part of graph`);\n    }\n  }\n\n  private mapOutputs(outputs: string[]) {\n    return outputs.map(name => {\n      if (this._signature != null && this._signature.outputs != null &&\n          this._signature.outputs[name] != null) {\n        const tensor = this._signature.outputs[name];\n        return tensor.name;\n      }\n      return name;\n    }, {});\n  }\n\n  private checkOutputs(outputs: string[]): void {\n    outputs.forEach(name => {\n      const [normalizedName] = parseNodeName(name);\n      if (!this.graph.nodes[normalizedName]) {\n        throw new Error(`The output '${name}' is not found in the graph`);\n      }\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {InferenceModel, io, ModelPredictConfig, NamedTensorMap, Tensor} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\nimport {NamedTensorsMap, TensorInfo} from '../data/types';\nimport {OperationMapper} from '../operations/operation_mapper';\n\nimport {GraphExecutor} from './graph_executor';\n\nexport const TFHUB_SEARCH_PARAM = '?tfjs-format=file';\nexport const DEFAULT_MODEL_NAME = 'model.json';\n/**\n * A `tf.GraphModel` is a directed, acyclic graph built from a\n * SavedModel GraphDef and allows inference execution.\n *\n * A `tf.GraphModel` can only be created by loading from a model converted from\n * a [TensorFlow SavedModel](https://www.tensorflow.org/guide/saved_model) using\n * the command line converter tool and loaded via `tf.loadGraphModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class GraphModel implements InferenceModel {\n  private executor: GraphExecutor;\n  private version = 'n/a';\n  private handler: io.IOHandler;\n  private artifacts: io.ModelArtifacts;\n  private initializer: GraphExecutor;\n\n  // Returns the version information for the tensorflow model GraphDef.\n  get modelVersion(): string {\n    return this.version;\n  }\n\n  get inputNodes(): string[] {\n    return this.executor.inputNodes;\n  }\n\n  get outputNodes(): string[] {\n    return this.executor.outputNodes;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this.executor.inputs;\n  }\n\n  get outputs(): TensorInfo[] {\n    return this.executor.outputs;\n  }\n\n  get weights(): NamedTensorsMap {\n    return this.executor.weightMap;\n  }\n\n  /**\n   * @param modelUrl url for the model, or an `io.IOHandler`.\n   * @param weightManifestUrl url for the weight file generated by\n   * scripts/convert.py script.\n   * @param requestOption options for Request, which allows to send credentials\n   * and custom headers.\n   * @param onProgress Optional, progress callback function, fired periodically\n   * before the load is completed.\n   */\n  constructor(\n      private modelUrl: string|io.IOHandler,\n      private loadOptions: io.LoadOptions = {}) {\n    if (loadOptions == null) {\n      this.loadOptions = {};\n    }\n  }\n\n  private findIOHandler() {\n    const path = this.modelUrl;\n    if ((path as io.IOHandler).load != null) {\n      // Path is an IO Handler.\n      this.handler = path as io.IOHandler;\n    } else if (this.loadOptions.requestInit != null) {\n      this.handler = io.browserHTTPRequest(path as string, this.loadOptions);\n    } else {\n      const handlers = io.getLoadHandlers(path as string, this.loadOptions);\n      if (handlers.length === 0) {\n        // For backward compatibility: if no load handler can be found,\n        // assume it is a relative http path.\n        handlers.push(io.browserHTTPRequest(path as string, this.loadOptions));\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) load handlers for ` +\n            `URL '${[path]}'`);\n      }\n      this.handler = handlers[0];\n    }\n  }\n\n  /**\n   * Loads the model and weight files, construct the in memory weight map and\n   * compile the inference graph.\n   */\n  async load(): Promise<boolean> {\n    this.findIOHandler();\n    if (this.handler.load == null) {\n      throw new Error(\n          'Cannot proceed with model loading because the IOHandler provided ' +\n          'does not have the `load` method implemented.');\n    }\n    const artifacts = await this.handler.load();\n\n    return this.loadSync(artifacts);\n  }\n\n  /**\n   * Synchronously construct the in memory weight map and\n   * compile the inference graph. Also initialize hashtable if any.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  loadSync(artifacts: io.ModelArtifacts) {\n    this.artifacts = artifacts;\n    const graph = this.artifacts.modelTopology as tensorflow.IGraphDef;\n    let signature = {};\n    if (this.artifacts.userDefinedMetadata != null) {\n      signature =  // tslint:disable-next-line:no-any\n          (this.artifacts.userDefinedMetadata as any).signature as\n          tensorflow.ISignatureDef;\n    }\n\n    this.version = `${graph.versions.producer}.${graph.versions.minConsumer}`;\n    const weightMap =\n        io.decodeWeights(this.artifacts.weightData, this.artifacts.weightSpecs);\n    this.executor = new GraphExecutor(\n        OperationMapper.Instance.transformGraph(graph, signature));\n    this.executor.weightMap = this.convertTensorMapToTensorsMap(weightMap);\n\n    if (artifacts.modelInitializer != null) {\n      const initializer =\n          OperationMapper.Instance.transformGraph(artifacts.modelInitializer);\n      this.initializer = new GraphExecutor(initializer);\n      this.initializer.weightMap = this.executor.weightMap;\n      this.initializer.execute({}, []);\n    }\n\n    return true;\n  }\n\n  /**\n   * Save the configuration and/or weights of the GraphModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const modelUrl =\n   *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n   * const model = await tf.loadGraphModel(modelUrl);\n   * const zeros = tf.zeros([1, 224, 224, 3]);\n   * model.predict(zeros).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * model.predict(zeros).print();\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new Error(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new Error(\n          'GraphModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    return handlerOrURL.save(this.artifacts);\n  }\n\n  /**\n   * Execute the inference for the input tensors.\n   *\n   * @param input The input tensors, when there is single input for the model,\n   * inputs param should be a `tf.Tensor`. For models with mutliple inputs,\n   * inputs params should be in either `tf.Tensor`[] if the input order is\n   * fixed, or otherwise NamedTensorMap format.\n   *\n   * For model with multiple inputs, we recommend you use NamedTensorMap as the\n   * input type, if you use `tf.Tensor`[], the order of the array needs to\n   * follow the\n   * order of inputNodes array. @see {@link GraphModel.inputNodes}\n   *\n   * You can also feed any intermediate nodes using the NamedTensorMap as the\n   * input type. For example, given the graph\n   *    InputNode => Intermediate => OutputNode,\n   * you can execute the subgraph Intermediate => OutputNode by calling\n   *    model.execute('IntermediateNode' : tf.tensor(...));\n   *\n   * This is useful for models that uses tf.dynamic_rnn, where the intermediate\n   * state needs to be fed manually.\n   *\n   * For batch inference execution, the tensors for each input need to be\n   * concatenated together. For example with mobilenet, the required input shape\n   * is [1, 244, 244, 3], which represents the [batch, height, width, channel].\n   * If we are provide a batched data of 100 images, the input tensor should be\n   * in the shape of [100, 244, 244, 3].\n   *\n   * @param config Prediction configuration for specifying the batch size and\n   * output node names. Currently the batch size option is ignored for graph\n   * model.\n   *\n   * @returns Inference result tensors. The output would be single `tf.Tensor`\n   * if model has single output node, otherwise Tensor[] or NamedTensorMap[]\n   * will be returned for model with multiple outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(inputs: Tensor|Tensor[]|NamedTensorMap, config?: ModelPredictConfig):\n      Tensor|Tensor[]|NamedTensorMap {\n    return this.execute(inputs, this.outputNodes);\n  }\n\n  private normalizeInputs(inputs: Tensor|Tensor[]|\n                          NamedTensorMap): NamedTensorMap {\n    if (!(inputs instanceof Tensor) && !Array.isArray(inputs)) {\n      // The input is already a NamedTensorMap.\n      return inputs;\n    }\n    inputs = Array.isArray(inputs) ? inputs : [inputs];\n    if (inputs.length !== this.inputNodes.length) {\n      throw new Error(\n          'Input tensor count mismatch,' +\n          `the graph model has ${this.inputNodes.length} placeholders, ` +\n          `while there are ${inputs.length} input tensors.`);\n    }\n    return this.inputNodes.reduce((map, inputName, i) => {\n      map[inputName] = (inputs as Tensor[])[i];\n      return map;\n    }, {} as NamedTensorMap);\n  }\n\n  private normalizeOutputs(outputs: string|string[]): string[] {\n    outputs = outputs || this.outputNodes;\n    return !Array.isArray(outputs) ? [outputs] : outputs;\n  }\n\n  /**\n   * Executes inference for the model for given input tensors.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no\n   * outputs are specified, the default outputs of the model would be used.\n   * You can inspect intermediate nodes of the model by adding them to the\n   * outputs array.\n   *\n   * @returns A single tensor if provided with a single output or no outputs\n   * are provided and there is only one default output, otherwise return a\n   * tensor array. The order of the tensor array is the same as the outputs\n   * if provided, otherwise the order of outputNodes attribute of the model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs?: string|string[]):\n      Tensor|Tensor[] {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = this.executor.execute(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n  /**\n   * Executes inference for the model for given input tensors in async\n   * fashion, use this method when your model contains control flow ops.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   *\n   * @returns A Promise of single tensor if provided with a single output or\n   * no outputs are provided and there is only one default output, otherwise\n   * return a tensor map.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async executeAsync(\n      inputs: Tensor|Tensor[]|NamedTensorMap,\n      outputs?: string|string[]): Promise<Tensor|Tensor[]> {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = await this.executor.executeAsync(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n\n  private convertTensorMapToTensorsMap(map: NamedTensorMap): NamedTensorsMap {\n    return Object.keys(map).reduce((newMap: NamedTensorsMap, key) => {\n      newMap[key] = [map[key]];\n      return newMap;\n    }, {});\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  dispose() {\n    this.executor.dispose();\n\n    if (this.initializer) {\n      this.initializer.dispose();\n    }\n  }\n}\n\n/**\n * Load a graph model given a URL to the model definition.\n *\n * Example of loading MobileNetV2 from a URL and making a prediction with a\n * zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n * const model = await tf.loadGraphModel(modelUrl);\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n *\n * Example of loading MobileNetV2 from a TF Hub URL and making a prediction with\n * a zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/2';\n * const model = await tf.loadGraphModel(modelUrl, {fromTFHub: true});\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n * @param modelUrl The url or an `io.IOHandler` that loads the model.\n * @param options Options for the HTTP request, which allows to send credentials\n *    and custom headers.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport async function loadGraphModel(\n    modelUrl: string|io.IOHandler,\n    options: io.LoadOptions = {}): Promise<GraphModel> {\n  if (modelUrl == null) {\n    throw new Error(\n        'modelUrl in loadGraphModel() cannot be null. Please provide a url ' +\n        'or an IOHandler that loads the model');\n  }\n  if (options == null) {\n    options = {};\n  }\n\n  if (options.fromTFHub) {\n    if ((modelUrl as io.IOHandler).load == null) {\n      if (!(modelUrl as string).endsWith('/')) {\n        modelUrl = (modelUrl as string) + '/';\n      }\n      modelUrl = `${modelUrl}${DEFAULT_MODEL_NAME}${TFHUB_SEARCH_PARAM}`;\n    }\n  }\n  const model = new GraphModel(modelUrl, options);\n  await model.load();\n  return model;\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\n\n// tslint:disable:no-any\n\n/**\n * A return value for a mapping function that can be applied via deepMap.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapResult = {\n  value: any,\n  recurse: boolean\n};\n\n/**\n * Apply a mapping function to a nested structure in a recursive manner.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapResult`.  The `DeepMapResult` either provides a\n *   replacement value for that node (i.e., replacing the subtree), or indicates\n *   that the node should be processed recursively.\n */\nexport function deepMap(input: any, mapFn: (x: any) => DeepMapResult): any|\n    any[] {\n  return deepMapInternal(input, mapFn);\n}\n\n/**\n * @param seen: A Map of known object mappings (i.e., memoized results of\n *   `mapFn()`)\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepMapInternal(\n    input: any, mapFn: (x: any) => DeepMapResult,\n    seen: Map<any, any> = new Map(), containedIn: Set<{}> = new Set()): any|\n    any[] {\n  if (input == null) {\n    return null;\n  }\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  if (seen.has(input)) {\n    return seen.get(input);\n  }\n  const result = mapFn(input);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep map function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    seen.set(input, result.value);\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const child = input[k];\n      const childResult = deepMapInternal(child, mapFn, seen, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// TODO(soergel, kangyizhang) Reconsider naming of deepZip() to avoid confusion\n// with zip()\n\n/**\n * Zip nested structures together in a recursive manner.\n *\n * This has the effect of transposing or pivoting data, e.g. converting it from\n * a row-major representation to a column-major representation.\n *\n * For example, `deepZip([{a: 1, b: 2}, {a: 3, b: 4}])` returns\n * `{a: [1, 3], b: [2, 4]}`.\n *\n * The inputs should all have the same nested structure (i.e., of arrays and\n * dicts).  The result is a single object with the same nested structure, where\n * the leaves are arrays collecting the values of the inputs at that location\n * (or, optionally, the result of a custom function applied to those arrays).\n *\n * @param inputs: An array of the objects to zip together.\n * @param zipFn: (optional) A function that expects an array of elements at a\n *   single node of the object tree, and returns a `DeepMapResult`.  The\n *   `DeepMapResult` either provides a result value for that node (i.e.,\n *   representing the subtree), or indicates that the node should be processed\n *   recursively.  The default zipFn recurses as far as possible and places\n *   arrays at the leaves.\n */\nexport function deepZip(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult = zipToList): any|any[] {\n  return deepZipInternal(inputs, zipFn);\n}\n\n/**\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepZipInternal(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult,\n    containedIn: Set<{}> = new Set()): any|any[] {\n  // The recursion follows the structure of input 0; it's assumed that all the\n  // other inputs have the same structure.\n  const input = inputs[0];\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  const result = zipFn(inputs);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep zip function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const children = inputs.map(x => x[k]);\n      const childResult = deepZipInternal(children, zipFn, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// tslint:disable-next-line:no-any\nexport function zipToList(x: any[]): DeepMapResult {\n  if (x === null) {\n    return null;\n  }\n  // TODO(soergel): validate array type?\n\n  if (isIterable(x[0])) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: x, recurse: false};\n  }\n}\n\n/**\n * A return value for an async map function for use with deepMapAndAwaitAll.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapAsyncResult = {\n  value: Promise<any>,\n  recurse: boolean\n};\n\n/**\n * Apply an async mapping function to a nested structure in a recursive manner.\n *\n * This first creates a nested structure of Promises, and then awaits all of\n * those, resulting in a single Promise for a resolved nested structure.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapAsyncResult`.  The `DeepMapAsyncResult` either provides\n *   a `Promise` for a replacement value for that node (i.e., replacing the\n *   subtree), or indicates that the node should be processed recursively.  Note\n *   that the decision whether or not to recurse must be made immediately; only\n *   the mapped value may be promised.\n */\nexport async function deepMapAndAwaitAll(\n    input: any, mapFn: (x: any) => DeepMapAsyncResult): Promise<any|any[]> {\n  const seen: Map<any, Promise<any>> = new Map();\n\n  // First do a normal deepMap, collecting Promises in 'seen' as a side effect.\n  deepMapInternal(input, mapFn, seen);\n\n  // Replace the Promises in 'seen' in place.\n  // Note TypeScript provides no async map iteration, and regular map iteration\n  // is broken too, so sadly we have to do Array.from() to make it work.\n  // (There's no advantage to Promise.all(), and that would be tricky anyway.)\n  for (const key of Array.from(seen.keys())) {\n    const value = seen.get(key);\n    if (value instanceof Promise) {\n      const mappedValue = await value;\n      seen.set(key, mappedValue);\n    }\n  }\n\n  // Normal deepMap again, this time filling in the resolved values.\n  // It's unfortunate that we have to do two passes.\n  // TODO(soergel): test performance and think harder about a fast solution.\n  const result = deepMapInternal(input, mapFn, seen);\n  return result;\n}\n\n/**\n * Determine whether the argument is iterable.\n *\n * @returns true if the argument is an array or any non-Tensor object.\n */\n// tslint:disable-next-line:no-any\nexport function isIterable(obj: any): boolean {\n  return obj != null && (!ArrayBuffer.isView(obj)) &&\n      (Array.isArray(obj) ||\n       (typeof obj === 'object' && !(obj instanceof tf.Tensor)));\n}\n\n/**\n * Determine whether the argument can be converted to Tensor.\n *\n * Tensors, primitives, arrays, and TypedArrays all qualify; anything else does\n * not.\n *\n * @returns true if the argument can be converted to Tensor.\n */\n// tslint:disable-next-line:no-any\nexport function canTensorify(obj: any): boolean {\n  return obj == null || isPrimitive(obj) || Array.isArray(obj) ||\n      (typeof obj === 'object' && (obj instanceof tf.Tensor)) ||\n      tf.util.isTypedArray(obj);\n}\n\n/**\n * Returns true if the given `value` is a primitive type. Otherwise returns\n * false. This is equivalant to node util.isPrimitive\n */\nfunction isPrimitive(value: any): boolean {\n  return (\n      value === null ||\n      (typeof value !== 'object' && typeof value !== 'function'));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {deepMap, DeepMapResult, isIterable} from './deep_map';\n\nexport function deepClone<T>(container: T): T {\n  return deepMap(container, cloneIfTensor);\n}\n\n// tslint:disable-next-line: no-any\nfunction cloneIfTensor(item: any): DeepMapResult {\n  if (item instanceof tf.Tensor) {\n    return ({value: item.clone(), recurse: false});\n  } else if (isIterable(item)) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: item, recurse: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/**\n * A ring buffer, providing O(1) FIFO, LIFO, and related operations.\n */\nexport class RingBuffer<T> {\n  // Note we store the indices in the range 0 <= index < 2*capacity.\n  // This allows us to distinguish the full from the empty case.\n  // See https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/\n  protected begin = 0;  // inclusive\n  protected end = 0;    // exclusive\n  protected doubledCapacity: number;\n\n  protected data: T[];\n\n  /**\n   * Constructs a `RingBuffer`.\n   * @param capacity The number of items that the buffer can accomodate.\n   */\n  constructor(public capacity: number) {\n    if (capacity == null) {\n      throw new RangeError('Can\\'t create a ring buffer of unknown capacity.');\n    }\n    if (capacity < 1) {\n      throw new RangeError('Can\\'t create ring buffer of capacity < 1.');\n    }\n    this.data = new Array<T>(capacity);\n    this.doubledCapacity = 2 * capacity;\n  }\n\n  /**\n   * Map any index into the range 0 <= index < 2*capacity.\n   */\n  protected wrap(index: number) {\n    // don't trust % on negative numbers\n    while (index < 0) {\n      index += this.doubledCapacity;\n    }\n    return index % this.doubledCapacity;\n  }\n\n  protected get(index: number) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t get item at a negative index.');\n    }\n    return this.data[index % this.capacity];\n  }\n\n  protected set(index: number, value: T) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t set item at a negative index.');\n    }\n    this.data[index % this.capacity] = value;\n  }\n\n  /**\n   * Returns the current number of items in the buffer.\n   */\n  length(): number {\n    let length = this.end - this.begin;\n    if (length < 0) {\n      length = this.doubledCapacity + length;\n    }\n    return length;\n  }\n\n  /**\n   * Reports whether the buffer is full.\n   * @returns true if the number of items in the buffer equals its capacity, and\n   *   false otherwise.\n   */\n  isFull() {\n    return this.length() === this.capacity;\n  }\n\n  /**\n   * Reports whether the buffer is empty.\n   * @returns true if the number of items in the buffer equals zero, and\n   *   false otherwise.\n   */\n  isEmpty() {\n    return this.length() === 0;\n  }\n\n  /**\n   * Adds an item to the end of the buffer.\n   */\n  push(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.set(this.end, value);\n    this.end = this.wrap(this.end + 1);\n  }\n\n  /**\n   * Adds many items to the end of the buffer, in order.\n   */\n  pushAll(values: T[]) {\n    for (const value of values) {\n      this.push(value);\n    }\n  }\n\n  /**\n   * Removes and returns the last item in the buffer.\n   */\n  pop(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    this.end = this.wrap(this.end - 1);\n    const result = this.get(this.end);\n    this.set(this.end, undefined);\n    return result;\n  }\n\n  /**\n   * Adds an item to the beginning of the buffer.\n   */\n  unshift(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.begin = this.wrap(this.begin - 1);\n    this.set(this.begin, value);\n  }\n\n  /**\n   * Removes and returns the first item in the buffer.\n   */\n  shift(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const result = this.get(this.begin);\n    this.set(this.begin, undefined);\n    this.begin = this.wrap(this.begin + 1);\n    return result;\n  }\n\n  /**\n   * Removes and returns a specific item in the buffer, and moves the last item\n   * to the vacated slot.  This is useful for implementing a shuffling stream.\n   * Note that this operation necessarily scrambles the original order.\n   *\n   * @param relativeIndex: the index of the item to remove, relative to the\n   *   first item in the buffer (e.g., hiding the ring nature of the underlying\n   *   storage).\n   */\n  shuffleExcise(relativeIndex: number): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const index = this.wrap(this.begin + relativeIndex);\n    const result = this.get(index);\n    this.set(index, this.pop());\n    return result;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {RingBuffer} from './ring_buffer';\n\nexport class GrowingRingBuffer<T> extends RingBuffer<T> {\n  private static INITIAL_CAPACITY = 32;\n\n  /**\n   * Constructs a `GrowingRingBuffer`.\n   */\n  constructor() {\n    super(GrowingRingBuffer.INITIAL_CAPACITY);\n  }\n\n  isFull() {\n    return false;\n  }\n\n  push(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.push(value);\n  }\n\n  unshift(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.unshift(value);\n  }\n\n  /**\n   * Doubles the capacity of the buffer.\n   */\n  private expand() {\n    const newCapacity = this.capacity * 2;\n    const newData = new Array<T>(newCapacity);\n    const len = this.length();\n\n    // Rotate the buffer to start at index 0 again, since we can't just\n    // allocate more space at the end.\n    for (let i = 0; i < len; i++) {\n      newData[i] = this.get(this.wrap(this.begin + i));\n    }\n\n    this.data = newData;\n    this.capacity = newCapacity;\n    this.doubledCapacity = 2 * this.capacity;\n    this.begin = 0;\n    this.end = len;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {Container} from '../types';\nimport {deepClone} from '../util/deep_clone';\nimport {deepMapAndAwaitAll, DeepMapAsyncResult, DeepMapResult, deepZip, zipToList} from '../util/deep_map';\nimport {GrowingRingBuffer} from '../util/growing_ring_buffer';\nimport {RingBuffer} from '../util/ring_buffer';\n\n/**\n * A nested structure of LazyIterators, used as the input to zip().\n */\nexport type IteratorContainer = Container<LazyIterator<tf.TensorContainer>>;\n\n// Here we implement a simple asynchronous iterator.\n// This lets us avoid using either third-party stream libraries or\n// recent TypeScript language support requiring polyfills.\n\n/**\n * Create a `LazyIterator` from an array of items.\n */\nexport function iteratorFromItems<T>(items: T[]): LazyIterator<T> {\n  return new ArrayIterator(items);\n}\n\n/**\n * Create a `LazyIterator` of incrementing integers.\n */\nexport function iteratorFromIncrementing(start: number): LazyIterator<number> {\n  let i = start;\n  return iteratorFromFunction(() => ({value: i++, done: false}));\n}\n\n/**\n * Create a `LazyIterator` from a function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * await iter.forEachAsync(e => console.log(e));\n * ```\n *\n * @param func A function that produces data on each call.\n */\nexport function iteratorFromFunction<T>(\n    func: () =>\n        IteratorResult<T>| Promise<IteratorResult<T>>): LazyIterator<T> {\n  return new FunctionCallIterator(func);\n}\n\n/**\n * Create a `LazyIterator` by concatenating underlying streams, which are\n * themselves provided as a stream.\n *\n * This can also be thought of as a \"stream flatten\" operation.\n *\n * @param baseIterators A stream of streams to be concatenated.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenated<T>(\n    baseIterators: LazyIterator<LazyIterator<T>>,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return new ChainedIterator(baseIterators, baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by concatenating streams produced by calling a\n * stream-generating function a given number of times.\n *\n * Since a `LazyIterator` is read-once, it cannot be repeated, but this\n * function can be used to achieve a similar effect:\n *\n *   LazyIterator.ofConcatenatedFunction(() => new MyIterator(), 6);\n *\n * @param iteratorFunc: A function that produces a new stream on each call.\n * @param count: The number of times to call the function.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenatedFunction<T>(\n    iteratorFunc: () => IteratorResult<LazyIterator<T>>, count: number,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return iteratorFromConcatenated(\n      iteratorFromFunction(iteratorFunc).take(count), baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by zipping together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nexport function iteratorFromZipped<O extends tf.TensorContainer>(\n    iterators: IteratorContainer,\n    mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL): LazyIterator<O> {\n  return new ZipIterator<O>(iterators, mismatchMode);\n}\n\n/**\n * An asynchronous iterator, providing lazy access to a potentially\n * unbounded stream of elements.\n *\n * Iterator can be obtained from a dataset:\n * `const iter = await dataset.iterator();`\n */\nexport abstract class LazyIterator<T> {\n  // This class implements AsyncIterator<T>, but we have not yet set the\n  // TypeScript --downlevelIteration flag to enable that.\n\n  abstract summary(): string;\n\n  /**\n   * Returns a `Promise` for the next element in the stream.\n   *\n   * When an item can be provided successfully, the return value is\n   * `{value:T, done:false}`.\n   *\n   * Calling next() on a closed stream returns `{value:null, done:true}`.\n   */\n  abstract async next(): Promise<IteratorResult<T>>;\n\n  /**\n   * Collect all remaining elements of a bounded stream into an array.\n   * Obviously this will succeed only for small streams that fit in memory.\n   * Useful for testing.\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArray(): Promise<T[]> {\n    const result: T[] = [];\n    let x = await this.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await this.next();\n    }\n    return result;\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArrayForTest(): Promise<T[]> {\n    const stream = this.prefetch(100);\n    const result: T[] = [];\n    let x = await stream.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await stream.next();\n    }\n    return result;\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveFully(): Promise<void> {\n    let x = await this.next();\n    while (!x.done) {\n      x = await this.next();\n    }\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted, or a predicate fails.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveWhile(predicate: (r: T) => boolean): Promise<void> {\n    let x = await this.next();\n    let shouldContinue = predicate(x.value);\n    while ((!x.done) && shouldContinue) {\n      x = await this.next();\n      shouldContinue = predicate(x.value);\n    }\n  }\n\n  /**\n   * Handles errors thrown on this stream using a provided handler function.\n   *\n   * @param handler A function that handles any `Error` thrown during a `next()`\n   *   call and returns true if the stream should continue (dropping the failed\n   *   call) or false if the stream should quietly terminate.  If the handler\n   *   itself throws (or rethrows) an `Error`, that will be propagated.\n   *\n   * @returns A `LazyIterator` of elements passed through from upstream,\n   *   possibly filtering or terminating on upstream `next()` calls that\n   *   throw an `Error`.\n   */\n  handleErrors(handler: (error: Error) => boolean): LazyIterator<T> {\n    return new ErrorHandlingLazyIterator(this, handler);\n  }\n\n  // TODO(soergel): Implement reduce() etc.\n\n  /**\n   * Filters this stream according to `predicate`.\n   *\n   * @param predicate A function mapping a stream element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `LazyIterator` of elements for which the predicate was true.\n   */\n  filter(predicate: (value: T) => boolean): LazyIterator<T> {\n    return new FilterIterator(this, predicate);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  map<O>(transform: (value: T) => O): LazyIterator<O> {\n    return new MapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through an async 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a `Promise` for a\n   *   transformed stream element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  mapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform, forcing serial execution.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  serialMapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform).serial();\n  }\n\n  /**\n   * Maps this stream through a 1-to-many transform.\n   *\n   * @param transform A function mapping a stream element to an array of\n   *   transformed elements.\n   *\n   * @returns A `DataStream` of transformed elements.\n   */\n  flatmap<O>(transform: (value: T) => O[]): LazyIterator<O> {\n    return new FlatmapIterator(this, transform);\n  }\n\n  /**\n   * Apply a function to every element of the stream.\n   *\n   * @param f A function to apply to each stream element.\n   */\n  async forEachAsync(f: (value: T) => void): Promise<void> {\n    return this.map(f).resolveFully();\n  }\n\n  /**\n   * Apply a function to every element of the stream, forcing serial execution.\n   *\n   * @param f A function to apply to each stream element.  Should return 'true'\n   *   to indicate that the stream should continue, or 'false' to cause it to\n   *   terminate.\n   */\n  async serialForEach(f: (value: T) => Promise<boolean>): Promise<void> {\n    return this.serialMapAsync(f).resolveWhile(x => (x === true));\n  }\n\n  /**\n   * Groups elements into batches, represented as arrays of elements.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"Row-major\" means that the resulting batch is simply a collection of\n   * rows: `[row1, row2, row3, ...]`.  This is contrast to the column-major\n   * form, which is needed for vectorized computation.\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `LazyIterator` of batches of elements, represented as arrays\n   *   of the original element type.\n   */\n  rowMajorBatch(batchSize: number, smallLastBatch = true): LazyIterator<T[]> {\n    return new RowMajorBatchIterator(this, batchSize, smallLastBatch);\n  }\n\n  /**\n   * Groups elements into batches, represented in column-major form.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"column-major\" means that the resulting batch is a (potentially\n   * nested) structure representing the columns.  Each column entry, then,\n   * contains a collection of the values found in that column for a range of\n   * input elements.  This representation allows for vectorized computation, in\n   * contrast to the row-major form.\n   *\n   * The inputs should all have the same nested structure (i.e., of arrays and\n   * dicts).  The result is a single object with the same nested structure,\n   * where the leaves are arrays collecting the values of the inputs at that\n   * location (or, optionally, the result of a custom function applied to those\n   * arrays).\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @param zipFn: (optional) A function that expects an array of elements at a\n   *   single node of the object tree, and returns a `DeepMapResult`.  The\n   *   `DeepMapResult` either provides a result value for that node (i.e.,\n   *   representing the subtree), or indicates that the node should be processed\n   *   recursively.  The default zipFn recurses as far as possible and places\n   *   arrays at the leaves.\n   * @returns A `LazyIterator` of batches of elements, represented as an object\n   *   with collections at the leaves.\n   */\n  columnMajorBatch(\n      batchSize: number, smallLastBatch = true,\n      // tslint:disable-next-line:no-any\n      zipFn: (xs: any[]) => DeepMapResult = zipToList):\n      LazyIterator<tf.TensorContainer> {\n    // First collect the desired number of input elements as a row-major batch.\n    const rowBatches = this.rowMajorBatch(batchSize, smallLastBatch);\n    // Now 'rotate' or 'pivot' the data, collecting all values from each column\n    // in the batch (i.e., for each key within the elements) into an array.\n    return rowBatches.map(x => deepZip(x, zipFn));\n  }\n\n  /**\n   * Concatenate this `LazyIterator` with another.\n   *\n   * @param iterator A `LazyIterator` to be concatenated onto this one.\n   * @param baseErrorHandler An optional function that can intercept `Error`s\n   *   raised during a `next()` call on the base stream.  This function can\n   *   decide whether the error should be propagated, whether the error should\n   *   be ignored, or whether the base stream should be terminated.\n   * @returns A `LazyIterator`.\n   */\n  concatenate(\n      iterator: LazyIterator<T>,\n      baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n    return new ChainedIterator(\n        iteratorFromItems([this, iterator]), baseErrorHandler);\n  }\n\n  /**\n   * Limits this stream to return at most `count` items.\n   *\n   * @param count The maximum number of items to provide from the stream. If\n   * a negative or undefined value is given, the entire stream is returned\n   *   unaltered.\n   */\n  take(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new TakeIterator(this, count);\n  }\n\n  /**\n   * Skips the first `count` items in this stream.\n   *\n   * @param count The number of items to skip.  If a negative or undefined\n   * value is given, the entire stream is returned unaltered.\n   */\n  skip(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new SkipIterator(this, count);\n  }\n\n  /**\n   * Prefetch the first `bufferSize` items in this stream.\n   *\n   * Note this prefetches Promises, but makes no guarantees about when those\n   * Promises resolve.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   */\n  prefetch(bufferSize: number): LazyIterator<T> {\n    return new PrefetchIterator(this, bufferSize);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  /**\n   * Randomly shuffles the elements of this stream.\n   *\n   * @param bufferSize: An integer specifying the number of elements from\n   * this stream from which the new stream will sample.\n   * @param seed: (Optional.) An integer specifying the random seed that\n   * will be used to create the distribution.\n   */\n  shuffle(windowSize: number, seed?: string): LazyIterator<T> {\n    return new ShuffleIterator(this, windowSize, seed);\n  }\n\n  /**\n   * Force an iterator to execute serially: each next() call will await the\n   * prior one, so that they cannot execute concurrently.\n   */\n  serial(): LazyIterator<T> {\n    return new SerialIterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on LazyIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// Iterators that just extend LazyIterator directly\n// ============================================================================\n\nclass ArrayIterator<T> extends LazyIterator<T> {\n  private trav = 0;\n  constructor(protected items: T[]) {\n    super();\n  }\n\n  summary() {\n    return `Array of ${this.items.length} items`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.trav >= this.items.length) {\n      return {value: null, done: true};\n    }\n    const item = this.items[this.trav];\n    this.trav++;\n    return {value: deepClone(item), done: false};\n  }\n}\n\nclass FunctionCallIterator<T> extends LazyIterator<T> {\n  constructor(\n      protected nextFn: () => IteratorResult<T>| Promise<IteratorResult<T>>) {\n    super();\n  }\n\n  summary() {\n    return `Function call`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    try {\n      return this.nextFn();\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message =\n          `Error thrown while iterating through a dataset: ${e.message}`;\n      throw e;\n    }\n  }\n}\n\nclass SerialIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(protected upstream: LazyIterator<T>) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Serial`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    return this.upstream.next();\n  }\n}\n\nclass SkipIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  count = 0;\n\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Skip`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider tradeoffs of reading in parallel, eg.\n    // collecting next() promises in an Array and then waiting for\n    // Promise.all() of those. Benefit: pseudo-parallel execution.  Drawback:\n    // maybe delayed GC.\n    while (this.count++ < this.maxCount) {\n      const skipped = await this.upstream.next();\n      // short-circuit if upstream is already empty\n      if (skipped.done) {\n        return skipped;\n      }\n      tf.dispose(skipped.value as {});\n    }\n    return this.upstream.next();\n  }\n}\n\nclass TakeIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Take`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.count++ >= this.maxCount) {\n      return {value: null, done: true};\n    }\n    return this.upstream.next();\n  }\n}\n\n// Note this batch just groups items into row-wise element arrays.\n// Rotating these to a column-wise representation happens only at the dataset\n// level.\nclass RowMajorBatchIterator<T> extends LazyIterator<T[]> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T[]>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected batchSize: number,\n      protected enableSmallLastBatch = true) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> RowMajorBatch`;\n  }\n\n  async next(): Promise<IteratorResult<T[]>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T[]>> {\n    const batch: T[] = [];\n    while (batch.length < this.batchSize) {\n      const item = await this.upstream.next();\n      if (item.done) {\n        if (this.enableSmallLastBatch && batch.length > 0) {\n          return {value: batch, done: false};\n        }\n        return {value: null, done: true};\n      }\n      batch.push(item.value);\n    }\n    return {value: batch, done: false};\n  }\n}\n\nclass FilterIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected predicate: (value: T) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Filter`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      const item = await this.upstream.next();\n      if (item.done || this.predicate(item.value)) {\n        return item;\n      }\n      tf.dispose(item.value as {});\n    }\n  }\n}\n\nclass MapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Map`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\nclass ErrorHandlingLazyIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected handler: (error: Error) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> handleErrors`;\n  }\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      try {\n        return await this.upstream.next();\n      } catch (e) {\n        if (!this.handler(e)) {\n          return {value: null, done: true};\n        }\n        // If the handler returns true, loop and fetch the next upstream item.\n\n        // If the upstream iterator throws an endless stream of errors, and if\n        // the handler says to ignore them, then we loop forever here.  That is\n        // the correct behavior-- it's up to the handler to decide when to stop.\n      }\n    }\n  }\n}\n\nclass AsyncMapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => Promise<O>) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> AsyncMap`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = await this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\n// Iterators that maintain a queue of pending items\n// ============================================================================\n\n/**\n * A base class for transforming streams that operate by maintaining an\n * output queue of elements that are ready to return via next().  This is\n * commonly required when the transformation is 1-to-many:  A call to next()\n * may trigger a call to the underlying stream, which will produce many\n * mapped elements of this stream-- of which we need to return only one, so\n * we have to queue the rest.\n */\nexport abstract class OneToManyIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  protected outputQueue: RingBuffer<T>;\n\n  constructor() {\n    super();\n    this.outputQueue = new GrowingRingBuffer<T>();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  /**\n   * Read one or more chunks from upstream and process them, possibly\n   * reading or writing a carryover, and adding processed items to the\n   * output queue.  Note it's possible that no items are added to the queue\n   * on a given pump() call, even if the upstream stream is not closed\n   * (e.g., because items are filtered).\n   *\n   * @return `true` if any action was taken, i.e. fetching items from the\n   *   upstream source OR adding items to the output queue.  `false` if the\n   *   upstream source is exhausted AND nothing was added to the queue\n   * (i.e., any remaining carryover).\n   */\n  protected abstract async pump(): Promise<boolean>;\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // Fetch so that the queue contains at least one item if possible.\n    // If the upstream source is exhausted, AND there are no items left in\n    // the output queue, then this stream is also exhausted.\n    while (this.outputQueue.length() === 0) {\n      // TODO(soergel): consider parallel reads.\n      if (!await this.pump()) {\n        return {value: null, done: true};\n      }\n    }\n    return {value: this.outputQueue.shift(), done: false};\n  }\n}\nclass FlatmapIterator<I, O> extends OneToManyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O[]) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Flatmap`;\n  }\n\n  async pump(): Promise<boolean> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return false;\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // that's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying any\n    // intermediate Tensors.  Here we are concerned only about the inputs.\n    const mappedArray = this.transform(item.value);\n    const outputTensors =\n        tf.tensor_util.getTensorsInContainer(mappedArray as {});\n    this.outputQueue.pushAll(mappedArray);\n\n    // TODO(soergel) faster intersection, and deduplicate outputTensors\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n\n    return true;\n  }\n}\n\n/**\n * Provides a `LazyIterator` that concatenates a stream of underlying\n * streams.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n */\nexport class ChainedIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>> = null;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private iterator: LazyIterator<T> = null;\n  private moreIterators: LazyIterator<LazyIterator<T>>;\n\n  constructor(\n      iterators: LazyIterator<LazyIterator<T>>,\n      private readonly baseErrorHandler?: (e: Error) => boolean) {\n    super();\n    this.moreIterators = iterators;\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of chained summaries';\n    return `${upstreamSummaries} -> Chained`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    this.lastRead = this.readFromChain(this.lastRead);\n    return this.lastRead;\n  }\n\n  private async readFromChain(lastRead: Promise<IteratorResult<T>>):\n      Promise<IteratorResult<T>> {\n    // Must await on the previous read since the previous read may have advanced\n    // the stream of streams, from which we need to read.\n    // This is unfortunate since we can't parallelize reads. Which means\n    // prefetching of chained streams is a no-op.\n    // One solution is to prefetch immediately upstream of this.\n    await lastRead;\n    if (this.iterator == null) {\n      const iteratorResult = await this.moreIterators.next();\n      if (iteratorResult.done) {\n        // No more streams to stream from.\n        return {value: null, done: true};\n      }\n      this.iterator = iteratorResult.value;\n      if (this.baseErrorHandler != null) {\n        this.iterator = this.iterator.handleErrors(this.baseErrorHandler);\n      }\n    }\n    const itemResult = await this.iterator.next();\n    if (itemResult.done) {\n      this.iterator = null;\n      return this.readFromChain(lastRead);\n    }\n    return itemResult;\n  }\n}\n\nexport enum ZipMismatchMode {\n  FAIL,      // require zipped streams to have the same length\n  SHORTEST,  // terminate zip when the first stream is exhausted\n  LONGEST    // use nulls for exhausted streams; use up the longest stream.\n}\n\n/**\n * Provides a `LazyIterator` that zips together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nclass ZipIterator<O extends tf.TensorContainer> extends LazyIterator<O> {\n  private count = 0;\n  private currentPromise: Promise<IteratorResult<O>> = null;\n\n  constructor(\n      protected readonly iterators: IteratorContainer,\n      protected readonly mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL) {\n    super();\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of zip summaries';\n    return `{${upstreamSummaries}} -> Zip`;\n  }\n\n  private async nextState(afterState: Promise<IteratorResult<O>>):\n      Promise<IteratorResult<O>> {\n    // This chaining ensures that the underlying next() are not even called\n    // before the previous ones have resolved.\n    await afterState;\n\n    // Collect underlying iterator \"done\" signals as a side effect in\n    // getNext()\n    let numIterators = 0;\n    let iteratorsDone = 0;\n\n    function getNext(container: IteratorContainer): DeepMapAsyncResult {\n      if (container instanceof LazyIterator) {\n        const result = container.next();\n        return {\n          value: result.then(x => {\n            numIterators++;\n            if (x.done) {\n              iteratorsDone++;\n            }\n            return x.value;\n          }),\n          recurse: false\n        };\n      } else {\n        return {value: null, recurse: true};\n      }\n    }\n\n    const mapped: O = await deepMapAndAwaitAll(this.iterators, getNext);\n\n    if (numIterators === iteratorsDone) {\n      // The streams have all ended.\n      return {value: null, done: true};\n    }\n    if (iteratorsDone > 0) {\n      switch (this.mismatchMode) {\n        case ZipMismatchMode.FAIL:\n          throw new Error(\n              'Zipped streams should have the same length. ' +\n              `Mismatched at element ${this.count}.`);\n        case ZipMismatchMode.SHORTEST:\n          return {value: null, done: true};\n        case ZipMismatchMode.LONGEST:\n        default:\n          // Continue.  The exhausted streams already produced value: null.\n      }\n    }\n\n    this.count++;\n    return {value: mapped, done: false};\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    this.currentPromise = this.nextState(this.currentPromise);\n    return this.currentPromise;\n  }\n}\n\n// Iterators that maintain a ring buffer of pending promises\n// ============================================================================\n\n/**\n * A stream that prefetches a given number of items from an upstream source,\n * returning them in FIFO order.\n *\n * Note this prefetches Promises, but makes no guarantees about when those\n * Promises resolve.\n */\nexport class PrefetchIterator<T> extends LazyIterator<T> {\n  protected buffer: RingBuffer<Promise<IteratorResult<T>>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected bufferSize: number) {\n    super();\n    this.buffer = new RingBuffer<Promise<IteratorResult<T>>>(bufferSize);\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Prefetch`;\n  }\n\n  /**\n   * Refill the prefetch buffer.  Returns only after the buffer is full, or\n   * the upstream source is exhausted.\n   */\n  protected refill() {\n    while (!this.buffer.isFull()) {\n      const v = this.upstream.next();\n      this.buffer.push(v);\n    }\n  }\n\n  next(): Promise<IteratorResult<T>> {\n    this.refill();\n    // This shift will never throw an error because the buffer is always\n    // full after a refill. If the stream is exhausted, the buffer will be\n    // full of Promises that will resolve to the end-of-stream signal.\n    return this.buffer.shift();\n  }\n}\n\n/**\n * A stream that performs a sliding-window random shuffle on an upstream\n * source. This is like a `PrefetchIterator` except that the items are\n * returned in randomized order.  Mixing naturally improves as the buffer\n * size increases.\n */\nexport class ShuffleIterator<T> extends PrefetchIterator<T> {\n  private readonly random: seedrandom.prng;\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private upstreamExhausted = false;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected windowSize: number,\n      seed?: string) {\n    super(upstream, windowSize);\n    this.random = seedrandom.alea(seed || tf.util.now().toString());\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private randomInt(max: number) {\n    return Math.floor(this.random() * max);\n  }\n\n  protected chooseIndex(): number {\n    return this.randomInt(this.buffer.length());\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider performance\n    if (!this.upstreamExhausted) {\n      this.refill();\n    }\n    while (!this.buffer.isEmpty()) {\n      const chosenIndex = this.chooseIndex();\n      const result = await this.buffer.shuffleExcise(chosenIndex);\n      if (result.done) {\n        this.upstreamExhausted = true;\n      } else {\n        this.refill();\n        return result;\n      }\n    }\n    return {value: null, done: true};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {TensorContainer, TensorLike} from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {iteratorFromConcatenated, iteratorFromFunction, iteratorFromItems, iteratorFromZipped, LazyIterator, ZipMismatchMode} from './iterators/lazy_iterator';\nimport {Container} from './types';\nimport {canTensorify, deepMapAndAwaitAll, DeepMapResult, isIterable} from './util/deep_map';\n\n/**\n * A nested structure of Datasets, used as the input to zip().\n */\nexport type DatasetContainer = Container<Dataset<TensorContainer>>;\n\n// TODO(soergel): consider vectorized operations within the pipeline.\n\n/**\n * Represents a potentially large list of independent data elements (typically\n * 'samples' or 'examples').\n *\n * A 'data example' may be a primitive, an array, a map from string keys to\n * values, or any nested structure of these.\n *\n * A `Dataset` represents an ordered collection of elements, together with a\n * chain of transformations to be performed on those elements. Each\n * transformation is a method of `Dataset` that returns another `Dataset`, so\n * these may be chained, e.g.\n * `const processedDataset = rawDataset.filter(...).map(...).batch(...)`.\n *\n * Data loading and transformation is done in a lazy, streaming fashion.  The\n * dataset may be iterated over multiple times; each iteration starts the data\n * loading anew and recapitulates the transformations.\n *\n * A `Dataset` is typically processed as a stream of unbatched examples --i.e.,\n * its transformations are applied one example at a time. Batching produces a\n * new `Dataset` where each element is a batch. Batching should usually come\n * last in a pipeline, because data transformations are easier to express on a\n * per-example basis than on a per-batch basis.\n *\n * The following code examples are calling `await dataset.forEachAsync(...)` to\n * iterate once over the entire dataset in order to print out the data.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport abstract class Dataset<T extends tf.TensorContainer> {\n  /*\n   * Provide a new stream of elements.  Note this will also start new streams\n   * from any underlying `Dataset`s.\n   *\n   * CAUTION: Any Tensors contained within the elements returned from\n   * this stream *must* be manually disposed to avoid a GPU memory leak.\n   * The tf.tidy() approach cannot be used in an asynchronous context.\n   */\n  abstract async iterator(): Promise<LazyIterator<T>>;\n\n  readonly size: number = null;\n\n  // TODO(soergel): Make Datasets report whether repeated iterator() calls\n  // produce the same result (e.g., reading from a file) or different results\n  // (e.g., from the webcam).  Currently we don't make this distinction but it\n  // could be important for the user to know.\n  // abstract isDeterministic(): boolean;\n\n  /**\n   * Groups elements into batches.\n   *\n   * It is assumed that each of the incoming dataset elements has the same\n   * structure-- i.e. the same set of keys at each location in an object\n   * hierarchy.  For each key, the resulting `Dataset` provides a batched\n   * element collecting all of the incoming values for that key.\n   *\n   *  * Incoming primitives are grouped into a 1-D Tensor.\n   *  * Incoming Tensors are grouped into a new Tensor where the 0'th axis is\n   *    the batch dimension.\n   *  * Incoming arrays are converted to Tensor and then batched.\n   *  * A nested array is interpreted as an n-D Tensor, so the batched result\n   *    has n+1 dimensions.\n   *  * An array that cannot be converted to Tensor produces an error.\n   *\n   * If an array should not be batched as a unit, it should first be converted\n   * to an object with integer keys.\n   *\n   * Here are a few examples:\n   *\n   * Batch a dataset of numbers:\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8]).batch(4);\n   * await a.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of arrays:\n   * ```js\n   * const b = tf.data.array([[1], [2], [3], [4], [5], [6], [7], [8]]).batch(4);\n   * await b.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of objects:\n   * ```js\n   * const c = tf.data.array([{a: 1, b: 11}, {a: 2, b: 12}, {a: 3, b: 13},\n   *   {a: 4, b: 14}, {a: 5, b: 15}, {a: 6, b: 16}, {a: 7, b: 17},\n   *   {a: 8, b: 18}]).batch(4);\n   * await c.forEachAsync(e => {\n   *   console.log('{');\n   *   for(var key in e) {\n   *     console.log(key+':');\n   *     e[key].print();\n   *   }\n   *   console.log('}');\n   * })\n   * ```\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `Dataset`, from which a stream of batches can be obtained.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  batch(batchSize: number, smallLastBatch = true): Dataset<tf.TensorContainer> {\n    const base = this;\n    tf.util.assert(\n        batchSize > 0, () => `batchSize needs to be positive, but it is\n      ${batchSize}`);\n    let size;\n    if (this.size === Infinity || this.size == null) {\n      // If the size of this dataset is infinity or null, the new size keeps the\n      // same.\n      size = this.size;\n    } else if (smallLastBatch) {\n      // If the size of this dataset is known and include small last batch, the\n      // new size is full batch count plus last batch.\n      size = Math.ceil(this.size / batchSize);\n    } else {\n      // If the size of this dataset is known and not include small last batch,\n      // the new size is full batch count.\n      size = Math.floor(this.size / batchSize);\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator())\n          .columnMajorBatch(batchSize, smallLastBatch, deepBatchConcat);\n    }, size);\n  }\n\n  /**\n   * Concatenates this `Dataset` with another.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * const b = tf.data.array([4, 5, 6]);\n   * const c = a.concatenate(b);\n   * await c.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param dataset A `Dataset` to be concatenated onto this one.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  concatenate(dataset: Dataset<T>): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity || dataset.size === Infinity) {\n      // If the size of any of these two dataset is infinity, new size is\n      // infinity.\n      size = Infinity;\n    } else if (this.size != null && dataset.size != null) {\n      // If the size of both datasets are known and not infinity, new size is\n      // sum the size of these two datasets.\n      size = this.size + dataset.size;\n    } else {\n      // If neither of these two datasets has infinite size and any of these two\n      // datasets' size is null, the new size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () =>\n            (await base.iterator()).concatenate(await dataset.iterator()),\n        size);\n  }\n\n  /**\n   * Filters this dataset according to `predicate`.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])\n   *   .filter(x => x%2 === 0);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param predicate A function mapping a dataset element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `Dataset` of elements for which the predicate was true.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  filter(predicate: (value: T) => boolean): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity) {\n      // If the size of this dataset is infinity, new size is infinity\n      size = Infinity;\n    } else {\n      // If this dataset has limited elements, new size is null because it might\n      // exhausted randomly.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).filter(x => tf.tidy(() => predicate(x)));\n    }, size);\n  }\n\n  /**\n   * Apply a function to every element of the dataset.\n   *\n   * After the function is applied to a dataset element, any Tensors contained\n   * within that element are disposed.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param f A function to apply to each dataset element.\n   * @returns A `Promise` that resolves after all elements have been processed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async forEachAsync(f: (input: T) => void): Promise<void> {\n    return (await this.iterator()).forEachAsync(f);\n  }\n\n  /**\n   * Maps this dataset through a 1-to-1 transform.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).map(x => x*x);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a transformed\n   *   dataset element.\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  map<O extends tf.TensorContainer>(transform: (value: T) => O): Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).map(x => tf.tidy(() => transform(x)));\n    }, this.size);\n  }\n\n  /**\n   * Maps this dataset through an async 1-to-1 transform.\n   *\n   * ```js\n   * const a =\n   *  tf.data.array([1, 2, 3]).mapAsync(x => new Promise(function(resolve){\n   *    setTimeout(() => {\n   *      resolve(x * x);\n   *    }, Math.random()*1000 + 500);\n   *  }));\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a `Promise` for a\n   *   transformed dataset element.  This transform is responsible for disposing\n   *   any intermediate `Tensor`s, i.e. by wrapping its computation in\n   *   `tf.tidy()`; that cannot be automated here (as it is in the synchronous\n   *   `map()` case).\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  mapAsync<O extends tf.TensorContainer>(transform: (value: T) => Promise<O>):\n      Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).mapAsync(transform);\n    }, this.size);\n  }\n\n  /**\n   *  Creates a `Dataset` that prefetches elements from this dataset.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  prefetch(bufferSize: number): Dataset<T> {\n    if (bufferSize == null) {\n      throw new RangeError(\n          '`Dataset.prefetch()` requires bufferSize to be specified.');\n    }\n\n    const base = this;\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).prefetch(bufferSize), this.size);\n  }\n\n  /**\n   * Repeats this dataset `count` times.\n   *\n   * NOTE: If this dataset is a function of global state (e.g. a random number\n   * generator), then different repetitions may produce different elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).repeat(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: (Optional) An integer, representing the number of times\n   *   the dataset should be repeated. The default behavior (if `count` is\n   *   `undefined` or negative) is for the dataset be repeated indefinitely.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  repeat(count?: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count > 0) {\n      // If this dataset has size and count is positive, new size is current\n      // size multiply count. This also covers the case that current size is\n      // infinity.\n      size = this.size * count;\n    } else if (count === 0) {\n      // If count is 0, new size is 0.\n      size = 0;\n    } else if (this.size != null && (count === undefined || count < 0)) {\n      // If this dataset has size and count is undefined or negative, the\n      // dataset will be repeated indefinitely and new size is infinity.\n      size = Infinity;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      const iteratorIterator = iteratorFromFunction(\n          async () => ({value: await base.iterator(), done: false}));\n      return iteratorFromConcatenated(iteratorIterator.take(count));\n    }, size);\n  }\n\n  /**\n   * Creates a `Dataset` that skips `count` initial elements from this dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).skip(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be skipped\n   *   to form the new dataset.  If `count` is greater than the size of this\n   *   dataset, the new dataset will contain no elements.  If `count`\n   *   is `undefined` or negative, skips the entire dataset.\n   *\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  skip(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count >= 0 && this.size >= count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is current size minus skipped size.This also covers the case that\n      // current size is infinity.\n      size = this.size - count;\n    } else if (\n        this.size != null &&\n        (this.size < count || count === undefined || count < 0)) {\n      // If the size of this dataset is smaller than count, or count is\n      // undefined or negative, skips the entire dataset and the new size is 0.\n      size = 0;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).skip(count), size);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  static readonly MAX_BUFFER_SIZE = 10000;\n\n  /**\n   * Pseudorandomly shuffles the elements of this dataset. This is done in a\n   * streaming manner, by sampling from a given number of prefetched elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).shuffle(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param bufferSize: An integer specifying the number of elements from this\n   *   dataset from which the new dataset will sample.\n   * @param seed: (Optional) An integer specifying the random seed that will\n   *   be used to create the distribution.\n   * @param reshuffleEachIteration: (Optional) A boolean, which if true\n   *   indicates that the dataset should be pseudorandomly reshuffled each time\n   *   it is iterated over. If false, elements will be returned in the same\n   *   shuffled order on each iteration. (Defaults to `true`.)\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  shuffle(bufferSize: number, seed?: string, reshuffleEachIteration = true):\n      Dataset<T> {\n    if (bufferSize == null || bufferSize < 0) {\n      if (this.size == null) {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.');\n      } else {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.  ' +\n            'If your data fits in main memory (for regular JS objects), ' +\n            'and/or GPU memory (for `tf.Tensor`s), consider setting ' +\n            `bufferSize to the dataset size (${this.size} elements)`);\n      }\n    }\n    const base = this;\n    const random = seedrandom.alea(seed || tf.util.now().toString());\n    return datasetFromIteratorFn(async () => {\n      let seed2 = random.int32();\n      if (reshuffleEachIteration) {\n        seed2 += random.int32();\n      }\n      return (await base.iterator()).shuffle(bufferSize, seed2.toString());\n    }, this.size);\n  }\n\n  /**\n   * Creates a `Dataset` with at most `count` initial elements from this\n   * dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).take(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be taken\n   *   to form the new dataset.  If `count` is `undefined` or negative, or if\n   *   `count` is greater than the size of this dataset, the new dataset will\n   *   contain all elements of this dataset.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  take(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && this.size > count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is count.\n      size = count;\n    } else if (this.size != null && this.size <= count) {\n      // If the size of this dataset is equal or smaller than count, the new\n      // dataset's size is the size of this dataset.\n      size = this.size;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).take(count), size);\n  }\n\n  /**\n   * Collect all elements of this dataset into an array.\n   *\n   * Obviously this will succeed only for small datasets that fit in memory.\n   * Useful for testing and generally should be avoided if possible.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]);\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async toArray() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArray();\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   */\n  async toArrayForTest() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArrayForTest();\n  }\n}\n\n/**\n * Create a `Dataset` defined by a provided iterator() function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * const ds = tf.data.datasetFromIteratorFn(iter);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n */\nexport function datasetFromIteratorFn<T extends tf.TensorContainer>(\n    iteratorFn: () => Promise<LazyIterator<T>>,\n    size: number = null): Dataset<T> {\n  return new class extends Dataset<T> {\n    size = size;\n\n    /*\n     * Provide a new stream of elements.  Note this will also start new streams\n     * from any underlying `Dataset`s.\n     */\n    async iterator(): Promise<LazyIterator<T>> {\n      return iteratorFn();\n    }\n  }\n  ();\n}\n\n/**\n * Create a `Dataset` from an array of elements.\n *\n * Create a Dataset from an array of objects:\n * ```js\n * const a = tf.data.array([{'item': 1}, {'item': 2}, {'item': 3}]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n *\n * Create a Dataset from an array of numbers:\n * ```js\n * const a = tf.data.array([4, 5, 6]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n * @param items An array of elements that will be parsed as items in a dataset.\n *\n * @doc {heading: 'Data', subheading: 'Creation', namespace: 'data'}\n */\nexport function array<T extends tf.TensorContainer>(items: T[]): Dataset<T> {\n  return datasetFromIteratorFn(\n      async () => iteratorFromItems(items), items.length);\n}\n\n/**\n * Create a `Dataset` by zipping together an array, dict, or nested\n * structure of `Dataset`s (and perhaps additional constants).\n * The underlying datasets must provide elements in a consistent order such that\n * they correspond.\n *\n * The number of elements in the resulting dataset is the same as the size of\n * the smallest dataset in datasets.\n *\n * The nested structure of the `datasets` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Note this means that, given an array of two datasets that produce dict\n * elements, the result is a dataset that produces elements that are arrays\n * of two dicts:\n *\n * Zip an array of datasets:\n * ```js\n * console.log('Zip two datasets of objects:');\n * const ds1 = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const ds2 = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const ds3 = tf.data.zip([ds1, ds2]);\n * await ds3.forEachAsync(e => console.log(JSON.stringify(e)));\n *\n * // If the goal is to merge the dicts in order to produce elements like\n * // {a: ..., b: ...}, this requires a second step such as:\n * console.log('Merge the objects:');\n * const ds4 = ds3.map(x => {return {a: x[0].a, b: x[1].b}});\n * await ds4.forEachAsync(e => console.log(e));\n * ```\n *\n * Zip a dict of datasets:\n * ```js\n * const a = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const b = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const c = tf.data.zip({c: a, d: b});\n * await c.forEachAsync(e => console.log(JSON.stringify(e)));\n * ```\n *\n * @doc {heading: 'Data', subheading: 'Operations', namespace: 'data'}\n */\nexport function zip<O extends tf.TensorContainer>(datasets: DatasetContainer):\n    Dataset<O> {\n  // manually type-check the argument for JS users\n  if (!isIterable(datasets)) {\n    throw new Error('The argument to zip() must be an object or array.');\n  }\n  let size;\n  if (Array.isArray(datasets)) {\n    for (let i = 0; i < datasets.length; i++) {\n      size = size == null ? (datasets[i] as Dataset<O>).size :\n                            Math.min(size, (datasets[i] as Dataset<O>).size);\n    }\n  } else if (datasets instanceof Object) {\n    for (const ds in datasets) {\n      size = size == null ? (datasets[ds] as Dataset<O>).size :\n                            Math.min(size, (datasets[ds] as Dataset<O>).size);\n    }\n  }\n  return datasetFromIteratorFn<O>(async () => {\n    const streams = await deepMapAndAwaitAll(datasets, d => {\n      if (d instanceof Dataset) {\n        return {value: d.iterator(), recurse: false};\n      } else if (isIterable(d)) {\n        return {value: null, recurse: true};\n      } else {\n        throw new Error(\n            'Leaves of the structure passed to zip() must be Datasets, ' +\n            'not primitives.');\n      }\n    });\n    return iteratorFromZipped<O>(streams, ZipMismatchMode.SHORTEST);\n  }, size);\n}\n\n/**\n * A zip function for use with deepZip, passed via the columnMajorBatch call.\n *\n * Accepts an array of identically-structured nested elements and either batches\n * them (if they are primitives, numeric arrays, or Tensors) or requests\n * recursion (if not).\n */\n// tslint:disable-next-line:no-any\nfunction deepBatchConcat(rows: any[]): DeepMapResult {\n  if (rows === null) {\n    return null;\n  }\n\n  // use the first item to decide whether to recurse or batch here.\n  const exampleRow = rows[0];\n\n  if (canTensorify(exampleRow)) {\n    // rows is an array of primitives, Tensors, or arrays.  Batch them.\n    const value = batchConcat(rows);\n    return {value, recurse: false};\n  }\n\n  // the example row is an object, so recurse into it.\n  return {value: null, recurse: true};\n}\n\n/**\n * Assembles a list of same-shaped numbers, number arrays, or Tensors\n * into a single new Tensor where axis 0 is the batch dimension.\n */\nfunction batchConcat<T extends(TensorLike | tf.Tensor)>(arrays: T[]):\n    tf.Tensor {\n  if (arrays.length === 0) {\n    // We can't return an empty Tensor because we don't know the element shape.\n    throw new Error('Can\\'t make a batch of zero elements.');\n  }\n\n  if (arrays[0] instanceof tf.Tensor) {\n    // Input is an array of Tensors\n    return tf.stack(arrays as tf.Tensor[]);\n  } else {\n    // Input is a possibly-nested array of numbers.\n    return tf.tensor(arrays as TensorLike);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\n\n/**\n * Represents a potentially large collection of text lines.\n *\n * The results are not batched.\n */\nexport class TextLineDataset extends Dataset<string> {\n  /**\n   * Create a `TextLineDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   */\n  constructor(protected readonly input: DataSource) {\n    super();\n  }\n\n  async iterator(): Promise<LazyIterator<string>> {\n    const inputIterator = await this.input.iterator();\n    const utf8Iterator = inputIterator.decodeUTF8();\n    const lineIterator = utf8Iterator.split('\\n').map(line => {\n      // Windows/DOS format text file has extra line breaker at the end of line.\n      if (line.endsWith('\\r')) {\n        line = line.slice(0, -1);\n      }\n      return line;\n    });\n    return lineIterator;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\nimport {ColumnConfig, CSVConfig} from '../types';\nimport {TextLineDataset} from './text_line_dataset';\n\nconst CODE_QUOTE = '\"';\nconst STATE_OUT = Symbol('out');\nconst STATE_FIELD = Symbol('field');\nconst STATE_QUOTE = Symbol('quote');\nconst STATE_QUOTE_AFTER_QUOTE = Symbol('quoteafterquote');\nconst STATE_WITHIN_QUOTE_IN_QUOTE = Symbol('quoteinquote');\n\n/**\n * Represents a potentially large collection of delimited text records.\n *\n * The produced `TensorContainer`s each contain one key-value pair for\n * every column of the table.  When a field is empty in the incoming data, the\n * resulting value is `undefined`, or throw error if it is required.  Values\n * that can be parsed as numbers are emitted as type `number`, other values\n * are parsed as `string`.\n *\n * The results are not batched.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport class CSVDataset extends Dataset<TensorContainer> {\n  base: TextLineDataset;\n  private hasHeader = true;\n  private fullColumnNames: string[] = null;\n  private columnNamesValidated = false;\n  private columnConfigs: {[key: string]: ColumnConfig} = null;\n  private configuredColumnsOnly = false;\n  private delimiter = ',';\n  private delimWhitespace = false;\n\n  /**\n   * Returns column names of the csv dataset. If `configuredColumnsOnly` is\n   * true, return column names in `columnConfigs`. If `configuredColumnsOnly` is\n   * false and `columnNames` is provided, `columnNames`. If\n   * `configuredColumnsOnly` is false and `columnNames` is not provided, return\n   * all column names parsed from the csv file. For example usage please go to\n   * `tf.data.csv`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async columnNames() {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    return this.configuredColumnsOnly ? Object.keys(this.columnConfigs) :\n                                        this.fullColumnNames;\n  }\n\n  /* 1) If `columnNames` is provided as string[], use this string[] as output\n   * keys in corresponding order. The length must match the number of inferred\n   * columns if `hasHeader` is true .\n   * 2) If `columnNames` is not provided, parse header line as `columnNames` if\n   * hasHeader is true. If `hasHeader` is false, throw an error.\n   * 3) If `columnConfigs` is provided, all the keys in `columnConfigs` must\n   * exist in parsed `columnNames`.\n   */\n  private async setColumnNames() {\n    const columnNamesFromFile = await this.maybeReadHeaderLine();\n    if (!this.fullColumnNames && !columnNamesFromFile) {\n      // Throw an error if columnNames is not provided and no header line.\n      throw new Error(\n          'Column names must be provided if there is no header line.');\n    } else if (this.fullColumnNames && columnNamesFromFile) {\n      // Check provided columnNames match header line.\n      util.assert(\n          columnNamesFromFile.length === this.fullColumnNames.length,\n          () => 'The length of provided columnNames (' +\n              this.fullColumnNames.length.toString() +\n              ') does not match the length of the header line read from ' +\n              'file (' + columnNamesFromFile.length.toString() + ').');\n    }\n    if (!this.fullColumnNames) {\n      this.fullColumnNames = columnNamesFromFile;\n    }\n    // Check if there are duplicate column names.\n    const counts: {[key: string]: number} = this.fullColumnNames.reduce(\n        (countAcc: {[key: string]: number}, name) => {\n          countAcc[name] = (countAcc[name] + 1) || 1;\n          return countAcc;\n        },\n        {});\n    const duplicateNames =\n        Object.keys(counts).filter((name) => (counts[name] > 1));\n    util.assert(\n        duplicateNames.length === 0,\n        () => 'Duplicate column names found: ' + duplicateNames.toString());\n    // Check if keys in columnConfigs match columnNames.\n    if (this.columnConfigs) {\n      for (const key of Object.keys(this.columnConfigs)) {\n        const index = this.fullColumnNames.indexOf(key);\n        if (index === -1) {\n          throw new Error(\n              'The key \"' + key +\n              '\" provided in columnConfigs does not match any of the column ' +\n              'names (' + this.fullColumnNames.toString() + ').');\n        }\n      }\n    }\n    this.columnNamesValidated = true;\n  }\n\n  private async maybeReadHeaderLine() {\n    if (this.hasHeader) {\n      const iter = await this.base.iterator();\n      const firstElement = await iter.next();\n      if (firstElement.done) {\n        throw new Error('No data was found for CSV parsing.');\n      }\n      const firstLine: string = firstElement.value;\n      const headers = this.parseRow(firstLine, false);\n      return headers;\n    } else {\n      return null;\n    }\n  }\n\n  /**\n   * Create a `CSVDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   * @param csvConfig (Optional) A CSVConfig object that contains configurations\n   *     of reading and decoding from CSV file(s).\n   *\n   *     hasHeader: (Optional) A boolean value that indicates whether the first\n   *     row of provided CSV file is a header line with column names, and should\n   *     not be included in the data. Defaults to `true`.\n   *\n   *     columnNames: (Optional) A list of strings that corresponds to\n   *     the CSV column names, in order. If provided, it ignores the column\n   *     names inferred from the header row. If not provided, infers the column\n   *     names from the first row of the records. If hasHeader is false and\n   *     columnNames is not provided, this method throws an error.\n   *\n   *     columnConfigs: (Optional) A dictionary whose key is column names, value\n   *     is an object stating if this column is required, column's data type,\n   *     default value, and if this column is label. If provided, keys must\n   *     correspond to names provided in columnNames or inferred from the file\n   *     header lines. If isLabel is true any column, returns an array of two\n   *     items: the first item is a dict of features key/value pairs, the second\n   *     item is a dict of labels key/value pairs. If no feature is marked as\n   *     label, returns a dict of features only.\n   *\n   *     configuredColumnsOnly (Optional) If true, only columns provided in\n   *     columnConfigs will be parsed and provided during iteration.\n   *\n   *     delimiter (Optional) The string used to parse each line of the input\n   *     file. Defaults to `,`.\n   */\n  constructor(protected readonly input: DataSource, csvConfig?: CSVConfig) {\n    super();\n    this.base = new TextLineDataset(input);\n    if (!csvConfig) {\n      csvConfig = {};\n    }\n    this.hasHeader = csvConfig.hasHeader === false ? false : true;\n    this.fullColumnNames = csvConfig.columnNames;\n    this.columnConfigs = csvConfig.columnConfigs;\n    this.configuredColumnsOnly = csvConfig.configuredColumnsOnly;\n    if (csvConfig.delimWhitespace) {\n      util.assert(\n          csvConfig.delimiter == null,\n          () =>\n              'Delimiter should not be provided when delimWhitespace is true.');\n      this.delimWhitespace = true;\n      this.delimiter = ' ';\n    } else {\n      this.delimiter = csvConfig.delimiter ? csvConfig.delimiter : ',';\n    }\n  }\n\n  async iterator(): Promise<LazyIterator<TensorContainer>> {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    let lines = await this.base.iterator();\n    if (this.hasHeader) {\n      // We previously read the first line to get the columnNames.\n      // Now that we're providing data, skip it.\n      lines = lines.skip(1);\n    }\n    return lines.map(x => this.makeDataElement(x));\n  }\n\n  makeDataElement(line: string): TensorContainer {\n    const values = this.parseRow(line);\n    const features: {[key: string]: TensorContainer} = {};\n    const labels: {[key: string]: TensorContainer} = {};\n\n    for (let i = 0; i < this.fullColumnNames.length; i++) {\n      const key = this.fullColumnNames[i];\n      const config = this.columnConfigs ? this.columnConfigs[key] : null;\n      if (this.configuredColumnsOnly && !config) {\n        // This column is not selected.\n        continue;\n      } else {\n        const value = values[i];\n        let parsedValue = null;\n        if (value === '') {\n          // If default value is provided, use it. If default value is not\n          // provided, set as undefined.\n          if (config && config.default !== undefined) {\n            parsedValue = config.default;\n          } else if (config && (config.required || config.isLabel)) {\n            throw new Error(\n                `Required column ${key} is empty in this line: ${line}`);\n          } else {\n            parsedValue = undefined;\n          }\n        } else {\n          // A value is present, so parse it based on type\n          const valueAsNum = Number(value);\n          if (isNaN(valueAsNum)) {\n            // The value is a string and this column is declared as boolean\n            // in config, parse it as boolean.\n            if (config && config.dtype === 'bool') {\n              parsedValue = this.getBoolean(value);\n            } else {\n              // Set value as string\n              parsedValue = value;\n            }\n          } else if (!config || !config.dtype) {\n            // If this value is a number and no type config is provided, return\n            // it as number.\n            parsedValue = valueAsNum;\n          } else {\n            // If this value is a number and data type is provided, parse it\n            // according to provided data type.\n            switch (config.dtype) {\n              case 'float32':\n                parsedValue = valueAsNum;\n                break;\n              case 'int32':\n                parsedValue = Math.floor(valueAsNum);\n                break;\n              case 'bool':\n                parsedValue = this.getBoolean(value);\n                break;\n              default:\n                parsedValue = valueAsNum;\n            }\n          }\n        }\n        // Check if this column is label.\n        (config && config.isLabel) ? labels[key] = parsedValue :\n                                     features[key] = parsedValue;\n      }\n    }\n    // If label exists, return an object of features and labels as {xs:features,\n    // ys:labels}, otherwise return features only.\n    if (Object.keys(labels).length === 0) {\n      return features;\n\n    } else {\n      return {xs: features, ys: labels};\n    }\n  }\n\n  private getBoolean(value: string): number {\n    if (value === '1' || value.toLowerCase() === 'true') {\n      return 1;\n    } else {\n      return 0;\n    }\n  }\n\n  // adapted from https://beta.observablehq.com/@mbostock/streaming-csv\n  private parseRow(line: string, validateElementCount = true): string[] {\n    const result: string[] = [];\n    let readOffset = 0;\n    const readLength = line.length;\n    let currentState = STATE_OUT;\n    // Goes through the line to parse quote.\n    for (let i = 0; i < readLength; i++) {\n      switch (currentState) {\n        // Before enter a new field\n        case STATE_OUT:\n          switch (line.charAt(i)) {\n            // Enter a quoted field\n            case CODE_QUOTE:\n              readOffset = i + 1;\n              currentState = STATE_QUOTE;\n              break;\n            // Read an empty field\n            case this.delimiter:\n              readOffset = i + 1;\n              // If delimiter is white space and configured to collapse\n              // multiple white spaces, ignore this white space.\n              if (this.delimiter === ' ' && this.delimWhitespace) {\n                break;\n              }\n              result.push('');\n              currentState = STATE_OUT;\n              break;\n            // Enter an unquoted field\n            default:\n              currentState = STATE_FIELD;\n              readOffset = i;\n              break;\n          }\n          break;\n        // In an unquoted field\n        case STATE_FIELD:\n          switch (line.charAt(i)) {\n            // Exit an unquoted field, add it to result\n            case this.delimiter:\n              result.push(line.substring(readOffset, i));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            default:\n          }\n          break;\n        // In a quoted field\n        case STATE_QUOTE:\n          switch (line.charAt(i)) {\n            // Read a quote after a quote\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE_AFTER_QUOTE;\n              break;\n            default:\n          }\n          break;\n        // This state means it's right after a second quote in a field\n        case STATE_QUOTE_AFTER_QUOTE:\n          switch (line.charAt(i)) {\n            // Finished a quoted field\n            case this.delimiter:\n              result.push(line.substring(readOffset, i - 1));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            // Finished a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            // In a quoted part in a quoted field\n            default:\n              currentState = STATE_WITHIN_QUOTE_IN_QUOTE;\n              break;\n          }\n          break;\n        case STATE_WITHIN_QUOTE_IN_QUOTE:\n          switch (line.charAt(i)) {\n            // Exit a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            default:\n          }\n          break;\n        default:\n      }\n    }\n    // Adds last item based on if it is quoted.\n    if (currentState === STATE_QUOTE_AFTER_QUOTE) {\n      result.push(line.substring(readOffset, readLength - 1));\n    } else {\n      result.push(line.substring(readOffset));\n    }\n    // Check if each row has the same number of elements as column names.\n    if (validateElementCount && result.length !== this.fullColumnNames.length) {\n      throw new Error(`Invalid row in csv file. Should have ${\n          this.fullColumnNames.length} elements in a row, but got ${result}`);\n    }\n    return result;\n  }\n}\n\n// TODO(soergel): add more basic datasets for parity with tf.data\n// tf.data.FixedLengthRecordDataset()\n// tf.data.TFRecordDataset()\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env, Tensor, tensor, Tensor2D, Tensor3D, TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {MicrophoneConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of tensors from microphone audio stream. The tensors are\n * representing audio data as frequency-domain spectrogram generated with\n * browser's native FFT. Tensors representing time-domain waveform is available\n * based on configuration. Only works in browser environment.\n */\nexport class MicrophoneIterator extends LazyIterator<TensorContainer> {\n  private isClosed = false;\n  private stream: MediaStream;\n  private readonly fftSize: number;\n  private readonly columnTruncateLength: number;\n  private freqData: Float32Array;\n  private timeData: Float32Array;\n  private readonly numFrames: number;\n  private analyser: AnalyserNode;\n  private audioContext: AudioContext;\n  private sampleRateHz: number;\n  private readonly audioTrackConstraints: MediaTrackConstraints;\n  private readonly smoothingTimeConstant: number;\n  private readonly includeSpectrogram: boolean;\n  private readonly includeWaveform: boolean;\n\n  private constructor(protected readonly microphoneConfig: MicrophoneConfig) {\n    super();\n    this.fftSize = microphoneConfig.fftSize || 1024;\n    const fftSizeLog2 = Math.log2(this.fftSize);\n    if (this.fftSize < 0 || fftSizeLog2 < 4 || fftSizeLog2 > 14 ||\n        !Number.isInteger(fftSizeLog2)) {\n      throw new Error(\n          `Invalid fftSize: it must be a power of 2 between ` +\n          `2 to 4 and 2 to 14, but got ${this.fftSize}`);\n    }\n\n    this.numFrames = microphoneConfig.numFramesPerSpectrogram || 43;\n    this.sampleRateHz = microphoneConfig.sampleRateHz;\n    this.columnTruncateLength =\n        microphoneConfig.columnTruncateLength || this.fftSize;\n    this.audioTrackConstraints = microphoneConfig.audioTrackConstraints;\n    this.smoothingTimeConstant = microphoneConfig.smoothingTimeConstant || 0;\n\n    this.includeSpectrogram =\n        microphoneConfig.includeSpectrogram === false ? false : true;\n    this.includeWaveform =\n        microphoneConfig.includeWaveform === true ? true : false;\n    if (!this.includeSpectrogram && !this.includeWaveform) {\n      throw new Error(\n          'Both includeSpectrogram and includeWaveform are false. ' +\n          'At least one type of data should be returned.');\n    }\n  }\n\n  summary() {\n    return `microphone`;\n  }\n\n  // Construct a MicrophoneIterator and start the audio stream.\n  static async create(microphoneConfig: MicrophoneConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'microphone API is only supported in browser environment.');\n    }\n\n    const microphoneIterator = new MicrophoneIterator(microphoneConfig);\n\n    // Call async function start() to initialize the audio stream.\n    await microphoneIterator.start();\n\n    return microphoneIterator;\n  }\n\n  // Start the audio stream and FFT.\n  async start(): Promise<void> {\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        audio: this.audioTrackConstraints == null ? true :\n                                                    this.audioTrackConstraints,\n        video: false\n      });\n    } catch (e) {\n      throw new Error(\n          `Error thrown while initializing video stream: ${e.message}`);\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain audio from microphone.');\n    }\n\n    const ctxConstructor =\n        // tslint:disable-next-line:no-any\n        (window as any).AudioContext || (window as any).webkitAudioContext;\n    this.audioContext = new ctxConstructor();\n\n    if (!this.sampleRateHz) {\n      // If sample rate is not provided, use the available sample rate on\n      // device.\n      this.sampleRateHz = this.audioContext.sampleRate;\n    } else if (this.audioContext.sampleRate !== this.sampleRateHz) {\n      throw new Error(\n          `Mismatch in sampling rate: ` +\n          `Expected: ${this.sampleRateHz}; ` +\n          `Actual: ${this.audioContext.sampleRate}`);\n    }\n\n    const streamSource = this.audioContext.createMediaStreamSource(this.stream);\n    this.analyser = this.audioContext.createAnalyser();\n    this.analyser.fftSize = this.fftSize * 2;\n    this.analyser.smoothingTimeConstant = this.smoothingTimeConstant;\n    streamSource.connect(this.analyser);\n    this.freqData = new Float32Array(this.fftSize);\n    this.timeData = new Float32Array(this.fftSize);\n    return;\n  }\n\n  async next(): Promise<IteratorResult<TensorContainer>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let spectrogramTensor: Tensor;\n    let waveformTensor: Tensor;\n\n    const audioDataQueue = await this.getAudioData();\n    if (this.includeSpectrogram) {\n      const freqData = this.flattenQueue(audioDataQueue.freqDataQueue);\n      spectrogramTensor = this.getTensorFromAudioDataArray(\n          freqData, [this.numFrames, this.columnTruncateLength, 1]);\n    }\n    if (this.includeWaveform) {\n      const timeData = this.flattenQueue(audioDataQueue.timeDataQueue);\n      waveformTensor = this.getTensorFromAudioDataArray(\n          timeData, [this.numFrames * this.fftSize, 1]);\n    }\n\n    return {\n      value: {'spectrogram': spectrogramTensor, 'waveform': waveformTensor},\n      done: false\n    };\n  }\n\n  // Capture one result from the audio stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<{spectrogram: Tensor3D, waveform: Tensor2D}> {\n    return (await this.next()).value as\n        {spectrogram: Tensor3D, waveform: Tensor2D};\n  }\n\n  private async getAudioData():\n      Promise<{freqDataQueue: Float32Array[], timeDataQueue: Float32Array[]}> {\n    const freqDataQueue: Float32Array[] = [];\n    const timeDataQueue: Float32Array[] = [];\n    let currentFrames = 0;\n    return new Promise(resolve => {\n      const intervalID = setInterval(() => {\n        if (this.includeSpectrogram) {\n          this.analyser.getFloatFrequencyData(this.freqData);\n          // If the audio stream is initializing, return empty queue.\n          if (this.freqData[0] === -Infinity) {\n            resolve({freqDataQueue, timeDataQueue});\n          }\n          freqDataQueue.push(this.freqData.slice(0, this.columnTruncateLength));\n        }\n        if (this.includeWaveform) {\n          this.analyser.getFloatTimeDomainData(this.timeData);\n          timeDataQueue.push(this.timeData.slice());\n        }\n\n        // Clean interval and return when all frames have been collected\n        if (++currentFrames === this.numFrames) {\n          clearInterval(intervalID);\n          resolve({freqDataQueue, timeDataQueue});\n        }\n      }, this.fftSize / this.sampleRateHz * 1e3);\n    });\n  }\n\n  // Stop the audio stream and pause the iterator.\n  stop(): void {\n    if (!this.isClosed) {\n      this.isClosed = true;\n      this.analyser.disconnect();\n      this.audioContext.close();\n      if (this.stream != null && this.stream.getTracks().length > 0) {\n        this.stream.getTracks()[0].stop();\n      }\n    }\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor[]> {\n    throw new Error('Can not convert infinite audio stream to array.');\n  }\n\n  // Return audio sampling rate in Hz\n  getSampleRate(): number {\n    return this.sampleRateHz;\n  }\n\n  private flattenQueue(queue: Float32Array[]): Float32Array {\n    const frameSize = queue[0].length;\n    const freqData = new Float32Array(queue.length * frameSize);\n    queue.forEach((data, i) => freqData.set(data, i * frameSize));\n    return freqData;\n  }\n\n  private getTensorFromAudioDataArray(freqData: Float32Array, shape: number[]):\n      Tensor {\n    const vals = new Float32Array(util.sizeFromShape(shape));\n    // If the data is less than the output shape, the rest is padded with zeros.\n    vals.set(freqData, vals.length - freqData.length);\n    return tensor(vals, shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {browser, env, image, tensor1d, Tensor1D, tensor2d, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\nimport {WebcamConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of image tensors from webcam video stream. Only works in\n * browser environment.\n */\nexport class WebcamIterator extends LazyIterator<Tensor3D> {\n  private isClosed = true;\n  private stream: MediaStream;\n  private resize = false;\n  private cropSize: [number, number];\n  private cropBox: Tensor2D;\n  private cropBoxInd: Tensor1D;\n\n  private constructor(\n      protected readonly webcamVideoElement: HTMLVideoElement,\n      protected readonly webcamConfig: WebcamConfig) {\n    super();\n    if (this.needToResize()) {\n      this.resize = true;\n      this.cropSize =\n          [this.webcamConfig.resizeHeight, this.webcamConfig.resizeWidth];\n      this.cropBoxInd = tensor1d([0], 'int32');\n      if (this.webcamConfig.centerCrop) {\n        // Calculate the box based on resizing shape.\n        const widthCroppingRatio =\n            this.webcamConfig.resizeWidth * 1.0 / this.webcamVideoElement.width;\n        const heightCroppingRatio = this.webcamConfig.resizeHeight * 1.0 /\n            this.webcamVideoElement.height;\n        const widthCropStart = (1 - widthCroppingRatio) / 2;\n        const heightCropStart = (1 - heightCroppingRatio) / 2;\n        const widthCropEnd = widthCropStart + widthCroppingRatio;\n        const heightCropEnd = heightCroppingRatio + heightCropStart;\n        this.cropBox = tensor2d(\n            [heightCropStart, widthCropStart, heightCropEnd, widthCropEnd],\n            [1, 4]);\n      } else {\n        this.cropBox = tensor2d([0, 0, 1, 1], [1, 4]);\n      }\n    }\n  }\n\n  summary() {\n    return `webcam`;\n  }\n\n  // Construct a WebcamIterator and start it's video stream.\n  static async create(\n      webcamVideoElement?: HTMLVideoElement, webcamConfig: WebcamConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'tf.data.webcam is only supported in browser environment.');\n    }\n\n    if (!webcamVideoElement) {\n      // If webcam video element is not provided, create a hidden video element\n      // with provided width and height.\n      webcamVideoElement = document.createElement('video');\n      if (!webcamConfig.resizeWidth || !webcamConfig.resizeHeight) {\n        throw new Error(\n            'Please provide webcam video element, or resizeWidth and ' +\n            'resizeHeight to create a hidden video element.');\n      }\n      webcamVideoElement.width = webcamConfig.resizeWidth;\n      webcamVideoElement.height = webcamConfig.resizeHeight;\n    }\n    const webcamIterator = new WebcamIterator(webcamVideoElement, webcamConfig);\n\n    // Call async function to initialize the video stream.\n    await webcamIterator.start();\n\n    return webcamIterator;\n  }\n\n  // Async function to start video stream.\n  async start(): Promise<void> {\n    if (this.webcamConfig.facingMode) {\n      util.assert(\n          (this.webcamConfig.facingMode === 'user') ||\n              (this.webcamConfig.facingMode === 'environment'),\n          () =>\n              `Invalid webcam facing mode: ${this.webcamConfig.facingMode}. ` +\n              `Please provide 'user' or 'environment'`);\n    }\n\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        video: {\n          deviceId: this.webcamConfig.deviceId,\n          facingMode: this.webcamConfig.facingMode ?\n              this.webcamConfig.facingMode :\n              'user',\n          width: this.webcamVideoElement.width,\n          height: this.webcamVideoElement.height\n        }\n      });\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message = `Error thrown while initializing video stream: ${e.message}`;\n      throw e;\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain video from webcam.');\n    }\n\n    // Older browsers may not have srcObject\n    try {\n      this.webcamVideoElement.srcObject = this.stream;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = window.URL.createObjectURL(this.stream);\n    }\n    // Start the webcam video stream\n    this.webcamVideoElement.play();\n\n    this.isClosed = false;\n\n    return new Promise<void>(resolve => {\n      // Add event listener to make sure the webcam has been fully initialized.\n      this.webcamVideoElement.onloadedmetadata = () => {\n        resolve();\n      };\n    });\n  }\n\n  async next(): Promise<IteratorResult<Tensor3D>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let img;\n    try {\n      img = browser.fromPixels(this.webcamVideoElement);\n    } catch (e) {\n      throw new Error(\n          `Error thrown converting video to pixels: ${JSON.stringify(e)}`);\n    }\n    if (this.resize) {\n      try {\n        return {value: this.cropAndResizeFrame(img), done: false};\n      } catch (e) {\n        throw new Error(`Error thrown cropping the video: ${e.message}`);\n      } finally {\n        img.dispose();\n      }\n    } else {\n      return {value: img, done: false};\n    }\n  }\n\n  private needToResize() {\n    // If resizeWidth and resizeHeight are provided, and different from the\n    // width and height of original HTMLVideoElement, then resizing and cropping\n    // is required.\n    if (this.webcamConfig.resizeWidth && this.webcamConfig.resizeHeight &&\n        (this.webcamVideoElement.width !== this.webcamConfig.resizeWidth ||\n         this.webcamVideoElement.height !== this.webcamConfig.resizeHeight)) {\n      return true;\n    }\n    return false;\n  }\n\n  // Cropping and resizing each frame based on config\n  cropAndResizeFrame(img: Tensor3D): Tensor3D {\n    return tidy(() => {\n      const expandedImage: Tensor4D = img.toFloat().expandDims(0);\n      let resizedImage;\n      resizedImage = image.cropAndResize(\n          expandedImage, this.cropBox, this.cropBoxInd, this.cropSize,\n          'bilinear');\n      // Extract image from batch cropping.\n      const shape = resizedImage.shape;\n      return resizedImage.reshape(shape.slice(1) as [number, number, number]);\n    });\n  }\n\n  // Capture one frame from the video stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<Tensor3D> {\n    return (await this.next()).value;\n  }\n\n  // Stop the video stream and pause webcam iterator.\n  stop(): void {\n    const tracks = this.stream.getTracks();\n\n    tracks.forEach(track => track.stop());\n\n    try {\n      this.webcamVideoElement.srcObject = null;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = null;\n    }\n    this.isClosed = true;\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor3D[]> {\n    throw new Error('Can not convert infinite video stream to array.');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {ByteChunkIterator} from './iterators/byte_chunk_iterator';\n\n/**\n * Represents a data source readable as a stream of binary data chunks.\n *\n * Because `Dataset`s can be read repeatedly (via `Dataset.iterator()`), this\n * provides a means to repeatedly create streams from the underlying data\n * sources.\n */\nexport abstract class DataSource {\n  /**\n   * Obtain a new stream of binary data chunks.\n   *\n   * Starts the new stream from the beginning of the data source, even if other\n   * streams have been obtained previously.\n   */\n  abstract async iterator(): Promise<ByteChunkIterator>;\n\n  // TODO(soergel): consider chainable Dataset construction here\n}\n\n// TODO(soergel): consider convenience factory functions here\n// in combination with chainable source->dataset above, e.g.:\n// tf.data.url(...).asCsvDataset().shuffle().batch()\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\n\nexport abstract class StringIterator extends LazyIterator<string> {\n  /**\n   * Splits a string stream on a given separator.\n   *\n   * It is assumed that the incoming chunk boundaries have no semantic meaning,\n   * so conceptually the incoming stream is treated simply as the concatenation\n   * of its elements.\n   *\n   * The outgoing stream provides chunks corresponding to the results of the\n   * standard string split() operation (even if such a chunk spanned incoming\n   * chunks).  The separators are not included.\n   *\n   * A typical usage is to split a text file (represented as a stream with\n   * arbitrary chunk boundaries) into lines.\n   *\n   * @param upstream A readable stream of strings that can be treated as\n   *   concatenated.\n   * @param separator A character to split on.\n   */\n  split(separator: string): StringIterator {\n    return new SplitIterator(this, separator);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on StringIterator.  Unfortunately they can't be placed in separate files, due\n// to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class SplitIterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass SplitIterator extends StringIterator {\n  private impl: SplitIteratorImpl;\n\n  constructor(protected upstream: LazyIterator<string>, separator: string) {\n    super();\n    this.impl = new SplitIteratorImpl(upstream, separator);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\nclass SplitIteratorImpl extends OneToManyIterator<string> {\n  // A partial string at the end of an upstream chunk\n  carryover = '';\n\n  constructor(\n      protected upstream: LazyIterator<string>, protected separator: string) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Split('${this.separator}')`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    if (chunkResult.done) {\n      if (this.carryover === '') {\n        return false;\n      }\n\n      // Pretend that the pump succeeded in order to emit the small last batch.\n      // The next pump() call will actually fail.\n      this.outputQueue.push(this.carryover);\n      this.carryover = '';\n      return true;\n    }\n    const lines = chunkResult.value.split(this.separator);\n    // Note the behavior: \" ab \".split(' ') === ['', 'ab', '']\n    // Thus the carryover may be '' if the separator falls on a chunk\n    // boundary; this produces the correct result.\n\n    lines[0] = this.carryover + lines[0];\n    for (const line of lines.slice(0, -1)) {\n      this.outputQueue.push(line);\n    }\n    this.carryover = lines[lines.length - 1];\n\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\nimport {StringIterator} from './string_iterator';\n\nexport abstract class ByteChunkIterator extends LazyIterator<Uint8Array> {\n  /**\n   * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n   *\n   * The byte arrays producetd from the ByteChunkIterator on which this is\n   * called will be interpreted as concatenated.  No assumptions are made about\n   * the boundaries of the incoming chunks, so a multi-byte UTF8 encoding of a\n   * character may span the boundary between chunks.  This naturally happens,\n   * for instance, when reading fixed-size byte arrays from a file.\n   */\n  decodeUTF8(): StringIterator {\n    return new Utf8Iterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on ByteChunkIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class Utf8Iterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass Utf8Iterator extends StringIterator {\n  private impl: Utf8IteratorImpl;\n\n  constructor(protected upstream: LazyIterator<Uint8Array>) {\n    super();\n    this.impl = new Utf8IteratorImpl(upstream);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\n/**\n * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n *\n * This is tricky because the incoming byte array boundaries may disrupt a\n * multi-byte UTF8 character. Thus any incomplete character data at the end of\n * a chunk must be carried over and prepended to the next chunk before\n * decoding. Luckily with native decoder, TextDecoder in browser and\n * string_decoder in node, byte array boundaries are handled automatically.\n *\n * In the context of an input pipeline for machine learning, UTF8 decoding is\n * needed to parse text files containing training examples or prediction\n * requests (e.g., formatted as CSV or JSON). We cannot use the built-in\n * decoding provided by FileReader.readAsText() because here we are in a\n * streaming context, which FileReader does not support.\n *\n * @param upstream A `LazyIterator` of `Uint8Arrays` containing UTF8-encoded\n *   text, which should be interpreted as concatenated.  No assumptions are\n *   made about the boundaries of the incoming chunks, so a multi-byte UTF8\n *   encoding of a character may span the boundary between chunks.  This\n *   naturally happens, for instance, when reading fixed-size byte arrays from a\n *   file.\n */\nclass Utf8IteratorImpl extends OneToManyIterator<string> {\n  // `decoder` as `any` here to dynamically assign value based on the\n  // environment.\n  // tslint:disable-next-line:no-any\n  decoder: any;\n\n  constructor(protected readonly upstream: LazyIterator<Uint8Array>) {\n    super();\n    if (env().get('IS_BROWSER')) {\n      this.decoder = new TextDecoder('utf-8');\n    } else {\n      // tslint:disable-next-line:no-require-imports\n      const {StringDecoder} = require('string_decoder');\n      this.decoder = new StringDecoder('utf8');\n    }\n  }\n  summary() {\n    return `${this.upstream.summary()} -> Utf8`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    let chunk;\n    if (chunkResult.done) {\n      return false;\n    } else {\n      chunk = chunkResult.value;\n    }\n\n    let text: string;\n    if (env().get('IS_BROWSER')) {\n      text = this.decoder.decode(chunk, {stream: true});\n    } else {\n      text = this.decoder.write(Buffer.from(chunk.buffer));\n    }\n    this.outputQueue.push(text);\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// inspired by https://github.com/maxogden/filereader-stream\nimport {env, util} from '@tensorflow/tfjs-core';\nimport {FileElement} from '../types';\nimport {ByteChunkIterator} from './byte_chunk_iterator';\n\nexport interface FileChunkIteratorOptions {\n  /** The byte offset at which to begin reading the File or Blob. Default 0. */\n  offset?: number;\n  /** The number of bytes to read at a time. Default 1MB. */\n  chunkSize?: number;\n}\n\n/**\n * Provide a stream of chunks from a File, Blob, or Uint8Array.\n * @param file The source File, Blob or Uint8Array.\n * @param options Optional settings controlling file reading.\n * @returns a lazy Iterator of Uint8Arrays containing sequential chunks of the\n *   input File, Blob or Uint8Array.\n */\nexport class FileChunkIterator extends ByteChunkIterator {\n  offset: number;\n  chunkSize: number;\n\n  constructor(\n      protected file: FileElement,\n      protected options: FileChunkIteratorOptions = {}) {\n    super();\n    util.assert(\n        (file instanceof Uint8Array) ||\n            (env().get('IS_BROWSER') ?\n                 (file instanceof File || file instanceof Blob) :\n                 false),\n        () => 'FileChunkIterator only supports File, Blob and Uint8Array ' +\n            'right now.');\n    this.offset = options.offset || 0;\n    // default 1MB chunk has tolerable perf on large files\n    this.chunkSize = options.chunkSize || 1024 * 1024;\n  }\n\n  summary() {\n    return `FileChunks ${this.file}`;\n  }\n\n  async next(): Promise<IteratorResult<Uint8Array>> {\n    if (this.offset >= ((this.file instanceof Uint8Array) ?\n                            this.file.byteLength :\n                            this.file.size)) {\n      return {value: null, done: true};\n    }\n    const chunk = new Promise<Uint8Array>((resolve, reject) => {\n      const end = this.offset + this.chunkSize;\n      if (this.file instanceof Uint8Array) {\n        // Note if end > this.uint8Array.byteLength, we just get a small last\n        // chunk.\n        resolve(new Uint8Array(this.file.slice(this.offset, end)));\n      } else {\n        // This branch assumes that this.file type is File or Blob, which\n        // means it is in the browser environment.\n\n        // TODO(soergel): is this a performance issue?\n        const fileReader = new FileReader();\n        fileReader.onload = (event) => {\n          let data: string|ArrayBuffer|Uint8Array = fileReader.result;\n          // Not sure we can trust the return type of\n          // FileReader.readAsArrayBuffer See e.g.\n          // https://github.com/node-file-api/FileReader/issues/2\n          if (data instanceof ArrayBuffer) {\n            data = new Uint8Array(data);\n          }\n          if (!(data instanceof Uint8Array)) {\n            return reject(new TypeError('FileReader returned unknown type.'));\n          }\n          resolve(data);\n        };\n        fileReader.onabort = (event) => {\n          return reject(new Error('Aborted'));\n        };\n        fileReader.onerror = (event) => {\n          return reject(new Error(event.type));\n        };\n        // TODO(soergel): better handle onabort, onerror\n        // Note if end > this.file.size, we just get a small last chunk.\n        const slice = this.file.slice(this.offset, end);\n        // We can't use readAsText here (even if we know the file is text)\n        // because the slice boundary may fall within a multi-byte character.\n        fileReader.readAsArrayBuffer(slice);\n      }\n      this.offset = end;\n    });\n    return {value: (await chunk), done: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\nimport {FileChunkIterator, FileChunkIteratorOptions} from './file_chunk_iterator';\n\n/**\n * Provide a stream of chunks from a URL.\n *\n * Note this class first downloads the entire file into memory before providing\n * the first element from the stream.  This is because the Fetch API does not\n * yet reliably provide a reader stream for the response body.\n */\nexport async function urlChunkIterator(\n    url: RequestInfo, options: FileChunkIteratorOptions = {}) {\n  let urlString;\n  let requestInit;\n  if ((typeof url) === 'string') {\n    urlString = url as string;\n  } else {\n    urlString = (url as Request).url;\n    requestInit = getRequestInitFromRequest(url as Request);\n  }\n  const response = await util.fetch(urlString, requestInit);\n  if (response.ok) {\n    const uint8Array = new Uint8Array(await response.arrayBuffer());\n    return new FileChunkIterator(uint8Array, options);\n  } else {\n    throw new Error(response.statusText);\n  }\n}\n\n// Generate RequestInit from Request to match tf.util.fetch signature.\nconst getRequestInitFromRequest = (request: Request) => {\n  const init = {\n    method: request.method,\n    headers: request.headers,\n    body: request.body,\n    mode: request.mode,\n    credentials: request.credentials,\n    cache: request.cache,\n    redirect: request.redirect,\n    referrer: request.referrer,\n    integrity: request.integrity,\n  };\n  return init;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// Skip tslint any type check cause this method is aiming to check type of\n// input.\n// tslint:disable-next-line:no-any\nexport function isLocalPath(source: any): boolean {\n  return (typeof source === 'string') && source.substr(0, 7) === 'file://';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIterator, FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {FileElement} from '../types';\nimport {isLocalPath} from '../util/source_util';\n\n/**\n * Represents a file, blob, or Uint8Array readable as a stream of binary data\n * chunks.\n */\nexport class FileDataSource extends DataSource {\n  /**\n   * Create a `FileDataSource`.\n   *\n   * @param input Local file path, or `File`/`Blob`/`Uint8Array` object to\n   *     read. Local file only works in node environment.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected input: FileElement|string,\n      protected readonly options: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.input) && env().get('IS_NODE')) {\n      // tslint:disable-next-line:no-require-imports\n      const fs = require('fs');\n      this.input = fs.readFileSync((this.input as string).substr(7));\n    }\n    // TODO(kangyizhang): Add LocalFileChunkIterator to split local streaming\n    // with file in browser.\n    return new FileChunkIterator(this.input as FileElement, this.options);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {urlChunkIterator} from '../iterators/url_chunk_iterator';\nimport {isLocalPath} from '../util/source_util';\nimport {FileDataSource} from './file_data_source';\n\n/*\n * Represents a URL readable as a stream of binary data chunks.\n */\nexport class URLDataSource extends DataSource {\n  /**\n   * Create a `URLDataSource`.\n   *\n   * @param url A source URL string, or a `Request` object.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected readonly url: RequestInfo,\n      protected readonly fileOptions: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  // TODO(soergel): provide appropriate caching options.  Currently this\n  // will download the URL anew for each call to iterator().  Since we have\n  // to treat the downloaded file as a blob/buffer anyway, we may as well retain\n  // it-- but that raises GC issues.  Also we may want a persistent disk cache.\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.url)) {\n      return (new FileDataSource(this.url as string, this.fileOptions))\n          .iterator();\n    } else {\n      return urlChunkIterator(this.url, this.fileOptions);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer} from '@tensorflow/tfjs-core';\nimport {Dataset, datasetFromIteratorFn} from './dataset';\nimport {CSVDataset} from './datasets/csv_dataset';\nimport {iteratorFromFunction} from './iterators/lazy_iterator';\nimport {MicrophoneIterator} from './iterators/microphone_iterator';\nimport {WebcamIterator} from './iterators/webcam_iterator';\nimport {URLDataSource} from './sources/url_data_source';\nimport {CSVConfig, MicrophoneConfig, WebcamConfig} from './types';\n\n/**\n * Create a `CSVDataset` by reading and decoding CSV file(s) from provided URL\n * or local path if it's in Node environment.\n *\n * Note: If isLabel in columnConfigs is `true` for at least one column, the\n * element in returned `CSVDataset` will be an object of\n * `{xs:features, ys:labels}`: xs is a dict of features key/value pairs, ys\n * is a dict of labels key/value pairs. If no column is marked as label,\n * returns a dict of features only.\n *\n * ```js\n * const csvUrl =\n * 'https://storage.googleapis.com/tfjs-examples/multivariate-linear-regression/data/boston-housing-train.csv';\n *\n * async function run() {\n *   // We want to predict the column \"medv\", which represents a median value of\n *   // a home (in $1000s), so we mark it as a label.\n *   const csvDataset = tf.data.csv(\n *     csvUrl, {\n *       columnConfigs: {\n *         medv: {\n *           isLabel: true\n *         }\n *       }\n *     });\n *\n *   // Number of features is the number of column names minus one for the label\n *   // column.\n *   const numOfFeatures = (await csvDataset.columnNames()).length - 1;\n *\n *   // Prepare the Dataset for training.\n *   const flattenedDataset =\n *     csvDataset\n *     .map(({xs, ys}) =>\n *       {\n *         // Convert xs(features) and ys(labels) from object form (keyed by\n *         // column name) to array form.\n *         return {xs:Object.values(xs), ys:Object.values(ys)};\n *       })\n *     .batch(10);\n *\n *   // Define the model.\n *   const model = tf.sequential();\n *   model.add(tf.layers.dense({\n *     inputShape: [numOfFeatures],\n *     units: 1\n *   }));\n *   model.compile({\n *     optimizer: tf.train.sgd(0.000001),\n *     loss: 'meanSquaredError'\n *   });\n *\n *   // Fit the model using the prepared Dataset\n *   return model.fitDataset(flattenedDataset, {\n *     epochs: 10,\n *     callbacks: {\n *       onEpochEnd: async (epoch, logs) => {\n *         console.log(epoch + ':' + logs.loss);\n *       }\n *     }\n *   });\n * }\n *\n * await run();\n * ```\n *\n * @param source URL or local path to get CSV file. If it's a local path, it\n * must have prefix `file://` and it only works in node environment.\n * @param csvConfig (Optional) A CSVConfig object that contains configurations\n *     of reading and decoding from CSV file(s).\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function csv(\n    source: RequestInfo, csvConfig: CSVConfig = {}): CSVDataset {\n  return new CSVDataset(new URLDataSource(source), csvConfig);\n}\n\n/**\n * Create a `Dataset` that produces each element by calling a provided function.\n *\n * Note that repeated iterations over this `Dataset` may produce different\n * results, because the function will be called anew for each element of each\n * iteration.\n *\n * Also, beware that the sequence of calls to this function may be out of order\n * in time with respect to the logical order of the Dataset. This is due to the\n * asynchronous lazy nature of stream processing, and depends on downstream\n * transformations (e.g. .shuffle()). If the provided function is pure, this is\n * no problem, but if it is a closure over a mutable state (e.g., a traversal\n * pointer), then the order of the produced elements may be scrambled.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const ds = tf.data.func(func);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param f A function that produces one data element on each call.\n */\nexport function func<T extends TensorContainer>(\n    f: () => IteratorResult<T>| Promise<IteratorResult<T>>): Dataset<T> {\n  const iter = iteratorFromFunction(f);\n  return datasetFromIteratorFn(async () => iter);\n}\n\n/**\n * Create a `Dataset` that produces each element from provided JavaScript\n * generator, which is a function*\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions),\n * or a function that returns an\n * iterator\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions).\n *\n * The returned iterator should have `.next()` function that returns element in\n * format of `{value: TensorContainer, done:boolean}`.\n *\n * Example of creating a dataset from an iterator factory:\n * ```js\n * function makeIterator() {\n *   const numElements = 10;\n *   let index = 0;\n *\n *   const iterator = {\n *     next: () => {\n *       let result;\n *       if (index < numElements) {\n *         result = {value: index, done: false};\n *         index++;\n *         return result;\n *       }\n *       return {value: index, done: true};\n *     }\n *   };\n *   return iterator;\n * }\n * const ds = tf.data.generator(makeIterator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * Example of creating a dataset from a generator:\n * ```js\n * function* dataGenerator() {\n *   const numElements = 10;\n *   let index = 0;\n *   while (index < numElements) {\n *     const x = index;\n *     index++;\n *     yield x;\n *   }\n * }\n *\n * const ds = tf.data.generator(dataGenerator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param generator A Javascript generator function that returns a JavaScript\n *     iterator.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function generator<T extends TensorContainer>(\n    generator: () => Iterator<T>| Promise<Iterator<T>>): Dataset<T> {\n  return datasetFromIteratorFn(async () => {\n    const gen = await generator();\n    return iteratorFromFunction(() => gen.next());\n  });\n}\n\n/**\n * Create an iterator that generate `Tensor`s from webcam video stream. This API\n * only works in Browser environment when the device has webcam.\n *\n * Note: this code snippet only works when the device has a webcam. It will\n * request permission to open the webcam when running.\n * ```js\n * const videoElement = document.createElement('video');\n * videoElement.width = 100;\n * videoElement.height = 100;\n * const cam = await tf.data.webcam(videoElement);\n * const img = await cam.capture();\n * img.print();\n * cam.stop();\n * ```\n *\n * @param webcamVideoElement A `HTMLVideoElement` used to play video from\n *     webcam. If this element is not provided, a hidden `HTMLVideoElement` will\n *     be created. In that case, `resizeWidth` and `resizeHeight` must be\n *     provided to set the generated tensor shape.\n * @param webcamConfig A `WebcamConfig` object that contains configurations of\n *     reading and manipulating data from webcam video stream.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function webcam(\n    webcamVideoElement?: HTMLVideoElement,\n    webcamConfig?: WebcamConfig): Promise<WebcamIterator> {\n  return WebcamIterator.create(webcamVideoElement, webcamConfig);\n}\n\n/**\n * Create an iterator that generate frequency-domain spectrogram `Tensor`s from\n * microphone audio stream with browser's native FFT. This API only works in\n * browser environment when the device has microphone.\n *\n * Note: this code snippet only works when the device has a microphone. It will\n * request permission to open the microphone when running.\n * ```js\n * const mic = await tf.data.microphone({\n *   fftSize: 1024,\n *   columnTruncateLength: 232,\n *   numFramesPerSpectrogram: 43,\n *   sampleRateHz:44100,\n *   includeSpectrogram: true,\n *   includeWaveform: true\n * });\n * const audioData = await mic.capture();\n * const spectrogramTensor = audioData.spectrogram;\n * spectrogramTensor.print();\n * const waveformTensor = audioData.waveform;\n * waveformTensor.print();\n * mic.stop();\n * ```\n *\n * @param microphoneConfig A `MicrophoneConfig` object that contains\n *     configurations of reading audio data from microphone.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function microphone(microphoneConfig?: MicrophoneConfig):\n    Promise<MicrophoneIterator> {\n  return MicrophoneIterator.create(microphoneConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${\n              opName} does not support complex64 tensors in the CPU backend.`);\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {backend_util, BackendTimingInfo, DataStorage, DataType, DataValues, engine, env, kernel_impls, KernelBackend, max, NumericDataType, Rank, Scalar, ShapeMap, slice_util, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, TensorInfo, TypedArray, upcastType, util} from '@tensorflow/tfjs-core';\n\nconst nonMaxSuppressionV3Impl = kernel_impls.nonMaxSuppressionV3Impl;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\nimport * as seedrandom from 'seedrandom';\nimport {assertNotComplex} from './cpu_util';\n\ninterface DataId {}\n\nfunction mapActivation(\n    backend: MathBackendCPU, x: Tensor, activation: backend_util.Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return backend.linear(x);\n  } else if (activation === 'relu') {\n    return backend.relu(x);\n  } else if (activation === 'elu') {\n    return tf.elu(x);\n  } else if (activation === 'relu6') {\n    return backend.relu6(x);\n  } else if (activation === 'prelu') {\n    return backend.prelu(x, preluActivationWeights);\n  }\n  throw new Error(\n      `Activation ${activation} has not been implemented for the CPU backend.`);\n}\n\nexport interface TensorData<D extends DataType> {\n  values?: backend_util.BackendValues;\n  dtype: D;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensorInfos field.\n  complexTensorInfos?: {real: TensorInfo, imag: TensorInfo};\n  // refCount keeps track of how many tensors reference it. Used for memory\n  // management.\n  refCount: number;\n}\n\nexport class MathBackendCPU extends KernelBackend {\n  public blockSize = 48;\n\n  data: DataStorage<TensorData<DataType>>;\n  private firstUse = true;\n\n  constructor() {\n    super();\n    this.data = new DataStorage(this, engine());\n  }\n\n  write(values: backend_util.BackendValues, shape: number[], dtype: DataType):\n      DataId {\n    if (this.firstUse) {\n      this.firstUse = false;\n      if (env().get('IS_NODE')) {\n        backend_util.warn(\n            '\\n============================\\n' +\n            'Hi there \uD83D\uDC4B. Looks like you are running TensorFlow.js in ' +\n            'Node.js. To speed things up dramatically, install our node ' +\n            'backend, which binds to TensorFlow C++, by running ' +\n            'npm i @tensorflow/tfjs-node, ' +\n            'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +\n            'Then call require(\\'@tensorflow/tfjs-node\\'); (-gpu ' +\n            'suffix for CUDA) at the start of your program. ' +\n            'Visit https://github.com/tensorflow/tfjs-node for more details.' +\n            '\\n============================');\n      }\n    }\n    const dataId = {};\n\n    this.data.set(dataId, {values, dtype, refCount: 1});\n\n    return dataId;\n  }\n\n  /**\n   * Create a data bucket in cpu backend.\n   * @param shape Shape of the `TensorInfo`.\n   * @param dtype DType of the `TensorInfo`.\n   * @param values The value of the `TensorInfo` stored as a flattened array.\n   */\n  makeTensorInfo(\n      shape: number[], dtype: DataType,\n      values?: backend_util.BackendValues): TensorInfo {\n    const outId = this.write(values, shape, dtype);\n\n    return {dataId: outId, shape, dtype};\n  }\n\n  /** Increase refCount of a `TensorData`. */\n  incRef(dataId: DataId): void {\n    const tensorData = this.data.get(dataId);\n    tensorData.refCount++;\n  }\n\n  /** Decrease refCount of a `TensorData`. */\n  decRef(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n      tensorData.refCount--;\n    }\n  }\n\n  move(\n      dataId: DataId, values: backend_util.BackendValues, shape: number[],\n      dtype: DataType): void {\n    this.data.set(dataId, {values, dtype, refCount: 1});\n  }\n\n  numDataIds(): number {\n    return this.data.numDataIds();\n  }\n\n  async read(dataId: DataId): Promise<backend_util.BackendValues> {\n    return this.readSync(dataId);\n  }\n  readSync(dataId: DataId): backend_util.BackendValues {\n    const {dtype, complexTensorInfos} = this.data.get(dataId);\n\n    if (dtype === 'complex64') {\n      const realValues =\n          this.readSync(complexTensorInfos.real.dataId) as Float32Array;\n      const imagValues =\n          this.readSync(complexTensorInfos.imag.dataId) as Float32Array;\n      return backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    }\n\n    return this.data.get(dataId).values;\n  }\n\n  private bufferSync<R extends Rank>(t: Tensor<R>): TensorBuffer<R> {\n    const data = this.readSync(t.dataId);\n    let decodedData = data as DataValues;\n    if (t.dtype === 'string') {\n      try {\n        // Decode the bytes into string.\n        decodedData = (data as Uint8Array[]).map(d => util.decodeString(d));\n      } catch {\n        throw new Error('Failed to decode encoded string bytes into utf-8');\n      }\n    }\n    return tf.buffer(t.shape, t.dtype, decodedData) as TensorBuffer<R>;\n  }\n\n  makeOutput<T extends Tensor>(\n      values: backend_util.BackendValues, shape: number[], dtype: DataType): T {\n    const dataId = this.write(values, shape, dtype);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  disposeData(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const {complexTensorInfos} = this.data.get(dataId);\n\n      if (complexTensorInfos != null) {\n        this.disposeData(complexTensorInfos.real.dataId);\n        this.disposeData(complexTensorInfos.imag.dataId);\n      }\n\n      this.data.delete(dataId);\n    }\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n\n      tensorData.refCount--;\n\n      if (tensorData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  async time(f: () => void): Promise<BackendTimingInfo> {\n    const start = util.now();\n    f();\n    const kernelMs = util.now() - start;\n    return {kernelMs};\n  }\n\n  memory() {\n    return {\n      // Unreliable due to automatic gc. The numbers above are cumulative.\n      unreliable: true,\n      reasons:\n          ['The reported memory is an upper bound. Due to automatic garbage ' +\n           'collection, the true allocated memory may be less.']\n    };\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    assertNotComplex(x, 'stridedSlice');\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tf.tensor([], outShape) as T;\n    }\n\n    const buffer = tf.buffer(outShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n    for (let i = 0; i < buffer.size; i++) {\n      const loc = buffer.indexToLoc(i);\n\n      const newLoc: number[] = new Array(loc.length);\n      for (let j = 0; j < newLoc.length; j++) {\n        newLoc[j] = loc[j] * strides[j] + begin[j];\n      }\n      buffer.set(xBuf.get(...newLoc), ...loc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  diag(x: Tensor): Tensor {\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const buffer = tf.buffer([x.size, x.size], x.dtype);\n    const vals = buffer.values;\n    for (let i = 0; i < xVals.length; i++) {\n      vals[i * x.size + i] = xVals[i];\n    }\n    return buffer.toTensor();\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = tf.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    assertNotComplex(x, 'reverse');\n\n    const buffer = tf.buffer(x.shape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < buffer.size; i++) {\n      const outLoc = buffer.indexToLoc(i);\n      const inLoc = outLoc.slice();\n      axis.forEach(ax => inLoc[ax] = x.shape[ax] - 1 - inLoc[ax]);\n      buffer.set(xBuf.get(...inLoc), ...outLoc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'neg');\n\n    // TODO(lina128): Use mul directly once neg is modularized.\n    return tf.mul(tf.scalar(-1), x);\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    assertNotComplex(tensors, 'addN');\n\n    const vals = tensors.map(t => this.readSync(t.dataId) as TypedArray);\n    const result = tf.buffer(tensors[0].shape, tensors[0].dtype as 'float32');\n    const resultVals = result.values;\n    for (let i = 0; i < tensors.length; i++) {\n      const currVals = vals[i];\n      for (let j = 0; j < resultVals.length; j++) {\n        resultVals[j] += currVals[j];\n      }\n    }\n    return result.toTensor() as T;\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n\n    // TODO(lina128): Use sub directly once softmax is modularized.\n    const a = tf.sub(logits, maxLogit.reshape(expandedShape));\n    const b = tf.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax\n    // kernel modularization.\n    return tf.div(b, sumExp);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    assertNotComplex([a, b], 'pow');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.pow(aValue, bValue)) as\n        T;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    assertNotComplex([a, b], 'matMul');\n\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const leftDim = transposeA ? a.shape[2] : a.shape[1];\n    const rightDim = transposeB ? b.shape[1] : b.shape[2];\n    const batchDim = a.shape[0];\n\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const [aBatch, aOuterStep, aInnerStep] = transposeA ?\n        [a.strides[0], 1, a.strides[1]] :\n        [a.strides[0], a.strides[1], 1];\n    const [bInnerStep, bOuterStep, bBatch] = transposeB ?\n        [1, b.strides[1], b.strides[0]] :\n        [b.strides[1], 1, b.strides[0]];\n\n    const size = leftDim * rightDim;\n    const result = tf.buffer([batchDim, leftDim, rightDim], a.dtype);\n    const resVals = result.values as TypedArray;\n    const blockSize = this.blockSize;\n\n    for (let b = 0; b < batchDim; b++) {\n      for (let i0 = 0; i0 < leftDim; i0 += blockSize) {\n        for (let j0 = 0; j0 < rightDim; j0 += blockSize) {\n          for (let k0 = 0; k0 < sharedDim; k0 += blockSize) {\n            // for when blockSize doesn't evenly divide the input\n            const iBlock = Math.min(i0 + blockSize, leftDim);\n            const jBlock = Math.min(j0 + blockSize, rightDim);\n            const kBlock = Math.min(k0 + blockSize, sharedDim);\n\n            for (let i = i0; i < iBlock; i++) {\n              for (let j = j0; j < jBlock; j++) {\n                let sum = 0.0;\n\n                for (let k = k0; k < kBlock; k++) {\n                  sum += aValues[b * aBatch + i * aOuterStep + k * aInnerStep] *\n                      bValues[k * bInnerStep + j * bOuterStep + b * bBatch];\n                }\n                resVals[b * size + (i * rightDim + j)] += sum;\n              }\n            }\n          }\n        }\n      }\n    }\n    return result.toTensor() as Tensor3D;\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    let result = this.batchMatMul(a, b, transposeA, transposeB);\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedBatchMatMul is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor3D;\n    }\n\n    return result;\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'floorDiv');\n\n    const op = (a: number, b: number) => Math.floor(a / b);\n    const outputDtype = 'int32';\n    return this.broadcastedBinaryOp(a, b, outputDtype, op);\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let sum = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        sum += aVals[offset + j];\n      }\n      vals[i] = sum;\n    }\n    return result;\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let prod = 1;\n      for (let j = 0; j < reduceSize; ++j) {\n        prod *= aVals[offset + j];\n      }\n      vals[i] = prod;\n    }\n    return result;\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    assertNotComplex(x, 'unsortedSegmentSum');\n\n    const res = [];\n\n    // Reshape the segment id's so that they can be broadcast with\n    // x. The new shape should be [segmentIds.shape, 1, ..., 1]\n    const numIters = x.rank - segmentIds.rank;\n    for (let i = 0; i < numIters; ++i) {\n      segmentIds = segmentIds.expandDims(i + 1);\n    }\n\n    for (let i = 0; i < numSegments; ++i) {\n      const segmentId = tf.scalar(i, 'int32');\n      const mask = tf.equal(segmentId, segmentIds).asType('float32');\n      const sum = mask.mul(x).sum(0);\n      res.push(sum);\n    }\n\n    return tf.stack(res);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMin');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMin', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      let minIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n          minIndex = j;\n        }\n      }\n      vals[i] = minIndex;\n    }\n    return result;\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMax');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMax', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let max = aVals[offset];\n      let maxIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value > max) {\n          max = value;\n          maxIndex = j;\n        }\n      }\n      vals[i] = maxIndex;\n    }\n    return result;\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    assertNotComplex(x, 'cumsum');\n\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `backend.cumsum in CPU expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(x.shape, resultDtype);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    const finalDim = x.shape[x.rank - 1];\n    const indexAdjuster = reverse ?\n        (i: number, j: number) => i + finalDim - j - 1 :\n        (i: number, j: number) => i + j;\n    for (let i = 0; i < aVals.length; i += finalDim) {\n      for (let j = 0; j < finalDim; j++) {\n        const idx = indexAdjuster(i, j);\n        if (j === 0) {\n          vals[idx] = exclusive ? 0 : aVals[idx];\n        } else {\n          const prevIdx = indexAdjuster(i, j - 1);\n          vals[idx] = exclusive ? aVals[prevIdx] + vals[prevIdx] :\n                                  aVals[idx] + vals[prevIdx];\n        }\n      }\n    }\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'equal');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal === bVal) ? 1 : 0;\n    });\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'notEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal !== bVal) ? 1 : 0;\n    });\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'less');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal < bVal) ? 1 : 0;\n    });\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'lessEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal <= bVal) ? 1 : 0;\n    });\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greater');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal > bVal) ? 1 : 0;\n    });\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greaterEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal >= bVal) ? 1 : 0;\n    });\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalAnd');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal && bVal;\n    });\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalOr');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal || bVal;\n    });\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([condition, a, b], 'select');\n\n    const values = this.readSync(condition.dataId) as TypedArray;\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const result = tf.zeros(a.shape, upcastType(a.dtype, b.dtype));\n    const newValues = this.readSync(result.dataId) as TypedArray;\n    let index = 0;\n    const offset = condition.rank === 0 || condition.rank > 1 || a.rank === 1 ?\n        1 :\n        util.sizeFromShape(a.shape.slice(1));\n\n    for (let i = 0; i < values.length; i++) {\n      for (let j = 0; j < offset; j++) {\n        if (values[i] === 1) {\n          newValues[index++] = aValues[i];\n        } else {\n          newValues[index++] = bValues[i];\n        }\n      }\n    }\n\n    return result;\n  }\n\n  where(condition: Tensor): Tensor2D {\n    assertNotComplex([condition], 'where');\n\n    const condVals = this.readSync(condition.dataId) as TypedArray;\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    assertNotComplex(x, 'topk');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'min');\n\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n        }\n      }\n      vals[i] = min;\n    }\n    return result;\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'minimum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.min(aVal, bVal));\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'mod');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const rem = aVal % bVal;\n      if ((aVal < 0 && bVal < 0) || (aVal >= 0 && bVal >= 0)) {\n        return rem;\n      } else {\n        return (rem + bVal) % bVal;\n      }\n    });\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'maximum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.max(aVal, bVal));\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'all');\n\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let all = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        all = all && value;\n      }\n      vals[i] = all;\n    }\n    return result;\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'any');\n\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let anyVal = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        anyVal = anyVal || value;\n      }\n      vals[i] = anyVal;\n    }\n    return result;\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'squaredDifference');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const diff = aVal - bVal;\n      return diff * diff;\n    });\n  }\n\n  linear<T extends Tensor>(x: T): T {\n    return x;\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.max(0, inVals[i]);\n    }\n    return res as T;\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.min(Math.max(0, inVals[i]), 6);\n    }\n    return res as T;\n  }\n\n  prelu<T extends Tensor>(x: T, a: T): T {\n    assertNotComplex([x, a], 'prelu');\n\n    return this.broadcastedBinaryOp(\n               x, a, x.dtype,\n               (xValue, aValue) => xValue < 0 ? aValue * xValue : xValue) as T;\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    assertNotComplex([dy, y], 'eluDer');\n\n    const resultValues = new Float32Array(y.size);\n    const values = this.readSync(y.dataId) as TypedArray;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    for (let i = 0; i < values.length; ++i) {\n      const v = values[i];\n      if (v >= 1) {\n        resultValues[i] = dyValues[i];\n      } else {\n        resultValues[i] = dyValues[i] * (v + 1);\n      }\n    }\n    return this.makeOutput(resultValues, y.shape, 'float32');\n  }\n\n  atan2<T extends Tensor>(a: T, b: T): T {\n    assertNotComplex([a, b], 'atan2');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.atan2(aValue, bValue)) as\n        T;\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.conv2d(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedConv2d is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, filter], 'conv2d');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n\n    const xBatchStride = x.strides[0];\n    const xRowStride = isChannelsLast ? x.strides[1] : x.strides[2];\n    const xColStride = isChannelsLast ? x.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : x.strides[1];\n    const yBatchStride = y.strides[0];\n    const yRowStride = isChannelsLast ? y.strides[1] : y.strides[2];\n    const yColStride = isChannelsLast ? y.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : y.strides[1];\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * xBatchStride;\n      const yOffset1 = b * yBatchStride;\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * yRowStride;\n        const xRCorner = yR * convInfo.strideHeight - padTop;\n        for (let wR = 0; wR < filterHeight; wR++) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * xRowStride;\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * yColStride;\n            const xCCorner = yC * convInfo.strideWidth - padLeft;\n            for (let wC = 0; wC < filterWidth; wC++) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * xColStride;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1 * xChannelStride];\n                for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                  yVals[yOffset3 + d2 * yChannelStride] +=\n                      xVal * wVals[wOffset3 + d2];\n                }\n                wOffset3 += convInfo.outChannels;\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor() as Tensor4D;\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padFront = convInfo.padInfo.front;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const y = tf.buffer<Rank.R5>(convInfo.outShape, x.dtype as 'float32');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yF = 0; yF < convInfo.outDepth; ++yF) {\n        const yOffset2 = yOffset1 + yF * y.strides[1];\n        const xFCorner = yF * convInfo.strideDepth - padFront;\n        for (let wF = 0; wF < filterDepth; wF++) {\n          const xF = xFCorner + wF * dilationDepth;\n          if (xF < 0 || xF >= convInfo.inDepth) {\n            continue;\n          }\n          const wOffset1 = wF * filter.strides[0];\n          const xOffset2 = xOffset1 + xF * x.strides[1];\n\n          for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n            const yOffset3 = yOffset2 + yR * y.strides[2];\n            const xRCorner = yR * convInfo.strideHeight - padTop;\n            for (let wR = 0; wR < filterHeight; wR++) {\n              const xR = xRCorner + wR * dilationHeight;\n              if (xR < 0 || xR >= convInfo.inHeight) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wR * filter.strides[1];\n              const xOffset3 = xOffset2 + xR * x.strides[2];\n              for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n                const yOffset4 = yOffset3 + yC * convInfo.outChannels;\n                const xCCorner = yC * convInfo.strideWidth - padLeft;\n                for (let wC = 0; wC < filterWidth; wC++) {\n                  const xC = xCCorner + wC * dilationWidth;\n                  if (xC < 0 || xC >= convInfo.inWidth) {\n                    continue;\n                  }\n                  const wOffset3 = wOffset2 + wC * filter.strides[2];\n                  const xOffset4 = xOffset3 + xC * convInfo.inChannels;\n                  let wOffset4 = wOffset3;\n                  for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                    const xVal = xVals[xOffset4 + d1];\n                    for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                      yVals[yOffset4 + d2] += xVal * wVals[wOffset4 + d2];\n                    }\n                    wOffset4 += convInfo.outChannels;\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor();\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'conv2dDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth,\n      dataFormat\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const xBatchStride = dx.strides[0];\n    const xRowStride = isChannelsLast ? dx.strides[1] : dx.strides[2];\n    const xColStride = isChannelsLast ? dx.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : dx.strides[1];\n    const yBatchStride = dy.strides[0];\n    const yRowStride = isChannelsLast ? dy.strides[1] : dy.strides[2];\n    const yColStride = isChannelsLast ? dy.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : dy.strides[1];\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset =\n                    yBatchStride * b + yRowStride * yR + yColStride * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let d2 = 0; d2 < outChannels; ++d2) {\n                  const pixel = dyValues[dyOffset + yChannelStride * d2];\n                  const weight = fltValues[fltOffset + d2];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            const dxOffset = xBatchStride * b + xRowStride * xR +\n                xColStride * xC + xChannelStride * d1;\n            dxValues[dxOffset] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const dx = tf.buffer<Rank.R5>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2, dxS3] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2, fltS3] = filter.strides;\n    const {\n      batchSize,\n      filterDepth,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inDepth,\n      inHeight,\n      inWidth,\n      outChannels,\n      outDepth,\n      outHeight,\n      outWidth,\n      strideDepth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const frontPad = filterDepth - 1 - convInfo.padInfo.front;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        // Frames of depth\n        for (let xF = 0; xF < inDepth; ++xF) {\n          const xFCorner = xF - frontPad;\n          const xFMin = Math.max(0, Math.ceil(xFCorner / strideDepth));\n          const yFMax =\n              Math.min(outDepth, (filterDepth + xFCorner) / strideDepth);\n\n          // Rows as per standard 2d matrix notation\n          for (let xR = 0; xR < inHeight; ++xR) {\n            const xRCorner = xR - topPad;\n            const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n            const yRMax =\n                Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n            // Columns as per standard 2d matrix notation\n            for (let xC = 0; xC < inWidth; ++xC) {\n              const xCCorner = xC - leftPad;\n              const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n              const yCMax =\n                  Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n              let dotProd = 0;\n              for (let yF = xFMin; yF < yFMax; ++yF) {\n                const wF = yF * strideDepth - xFCorner;\n\n                for (let yR = xRMin; yR < yRMax; ++yR) {\n                  const wR = yR * strideHeight - xRCorner;\n\n                  for (let yC = xCMin; yC < yCMax; ++yC) {\n                    const wC = yC * strideWidth - xCCorner;\n                    const dyOffset =\n                        dyS0 * b + dyS1 * yF + dyS2 * yR + dyS3 * yC;\n                    const fltOffset = fltS0 * (filterDepth - 1 - wF) +\n                        fltS1 * (filterHeight - 1 - wR) +\n                        fltS2 * (filterWidth - 1 - wC) + fltS3 * d1;\n\n                    for (let d2 = 0; d2 < outChannels; ++d2) {\n                      const pixel = dyValues[dyOffset + d2];\n                      const weight = fltValues[fltOffset + d2];\n                      dotProd += pixel * weight;\n                    }\n                  }\n                }\n              }\n              dxValues[dxS0 * b + dxS1 * xF + dxS2 * xR + dxS3 * xC + d1] =\n                  dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, dy], 'conv2dDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n          for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n            // Need to convolve.\n            let dotProd = 0;\n            for (let b = 0; b < convInfo.batchSize; ++b) {\n              for (let yR = yRMin; yR < yRMax; ++yR) {\n                const xR = wR + yR * strideHeight - topPad;\n                for (let yC = yCMin; yC < yCMax; ++yC) {\n                  const xC = wC + yC * strideWidth - leftPad;\n                  if (isChannelsLast) {\n                    dotProd +=\n                        xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n                  } else {\n                    dotProd +=\n                        xBuf.get(b, d1, xR, xC) * dyBuf.get(b, d2, yR, yC);\n                  }\n                }\n              }\n            }\n            dW.set(dotProd, wR, wC, d1, d2);\n          }\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const dw = tf.buffer<Rank.R5>(convInfo.filterShape, 'float32');\n    const dwValues = dw.values;\n    const [dwS0, dwS1, dwS2, dwS3] = dw.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const [xS0, xS1, xS2, xS3] = x.strides;\n\n    const frontPad = convInfo.padInfo.front;\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n\n    for (let wF = 0; wF < filterDepth; ++wF) {\n      const yFMin = Math.max(0, Math.ceil((frontPad - wF) / strideDepth));\n      const yFMax = Math.min(\n          convInfo.outDepth, (convInfo.inDepth + frontPad - wF) / strideDepth);\n      const wOffset1 = wF * dwS0;\n\n      for (let wR = 0; wR < filterHeight; ++wR) {\n        const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n        const yRMax = Math.min(\n            convInfo.outHeight,\n            (convInfo.inHeight + topPad - wR) / strideHeight);\n        const wOffset2 = wR * dwS1 + wOffset1;\n\n        for (let wC = 0; wC < filterWidth; ++wC) {\n          const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n          const yCMax = Math.min(\n              convInfo.outWidth,\n              (convInfo.inWidth + leftPad - wC) / strideWidth);\n          const wOffset3 = wC * dwS2 + wOffset2;\n\n          for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n            const wOffset4 = d1 * dwS3 + wOffset3;\n\n            for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n              let dotProd = 0;\n              for (let b = 0; b < convInfo.batchSize; ++b) {\n                const xOffset1 = b * xS0;\n                const yOffset1 = b * dyS0;\n\n                for (let yF = yFMin; yF < yFMax; ++yF) {\n                  const xF = wF + yF * strideDepth - frontPad;\n                  const xOffset2 = xF * xS1 + xOffset1;\n                  const yOffset2 = yF * dyS1 + yOffset1;\n\n                  for (let yR = yRMin; yR < yRMax; ++yR) {\n                    const xR = wR + yR * strideHeight - topPad;\n                    const xOffset3 = xR * xS2 + xOffset2;\n                    const yOffset3 = yR * dyS2 + yOffset2;\n\n                    for (let yC = yCMin; yC < yCMax; ++yC) {\n                      const xC = wC + yC * strideWidth - leftPad;\n                      const xOffset4 = xC * xS3 + xOffset3;\n                      const yOffset4 = yC * dyS3 + yOffset3;\n\n                      dotProd +=\n                          xValues[xOffset4 + d1] * dyValues[yOffset4 + d2];\n                    }\n                  }\n                }\n              }\n              dwValues[wOffset4 + d2] = dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dw.toTensor();\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.depthwiseConv2D(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedDepthwiseConv2D is\n      // modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, filter], 'depthwiseConv2D');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * y.strides[1];\n        const xRCorner = yR * convInfo.strideHeight - padLeft;\n        for (let wR = 0; wR < filterHeight; ++wR) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * x.strides[1];\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * y.strides[2];\n            const xCCorner = yC * convInfo.strideWidth - padTop;\n            for (let wC = 0; wC < filterWidth; ++wC) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * convInfo.inChannels;\n              let yOffset4 = yOffset3;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1];\n                for (let q = 0; q < chMul; ++q) {\n                  yVals[yOffset4 + q] += xVal * wVals[wOffset3 + q];\n                }\n                yOffset4 += chMul;\n                wOffset3 += chMul;\n              }\n            }\n          }\n        }\n      }\n    }\n\n    return y.toTensor() as Tensor4D;\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'depthwiseConv2DDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n    const chMul = outChannels / inChannels;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset = dyS0 * b + dyS1 * yR + dyS2 * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let dm = 0; dm < chMul; ++dm) {\n                  const d2 = d1 * chMul + dm;\n                  const pixel = dyValues[dyOffset + d2];\n                  const weight = fltValues[fltOffset + dm];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            dxValues[dxS0 * b + dxS1 * xR + dxS2 * xC + d1] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, dy], 'depthwiseConv2DDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n          const d1 = Math.trunc(d2 / chMul);\n          const dm = d2 % chMul;\n\n          let dotProd = 0;\n          for (let b = 0; b < convInfo.batchSize; ++b) {\n            for (let yR = yRMin; yR < yRMax; ++yR) {\n              const xR = wR + yR * strideHeight - topPad;\n              for (let yC = yCMin; yC < yCMax; ++yC) {\n                const xC = wC + yC * strideWidth - leftPad;\n                dotProd += xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n              }\n            }\n          }\n          dW.set(dotProd, wR, wC, d1, dm);\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    assertNotComplex(x, 'tile');\n    return tile(this.bufferSync(x), reps) as T;\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    assertNotComplex([x, indices], 'gather');\n\n    const newShape: number[] = x.shape.slice();\n    const indicesValues = this.readSync(indices.dataId) as TypedArray;\n    newShape[axis] = indicesValues.length;\n    const result = tf.buffer(newShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < result.size; ++i) {\n      const newLoc = result.indexToLoc(i);\n\n      const originalLoc: number[] = newLoc.slice();\n      originalLoc[axis] = indicesValues[newLoc[axis]];\n\n      const originalIndex = xBuf.locToIndex(originalLoc);\n      result.values[i] = xBuf.values[originalIndex];\n    }\n    return result.toTensor() as T;\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    assertNotComplex([x], 'batchToSpaceND');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return tf.transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  private pool3d(\n      x: Tensor5D, convInfo: backend_util.Conv3DInfo,\n      poolType: 'max'|'avg'): Tensor5D {\n    assertNotComplex(x, 'pool3d');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const initialValue =\n        (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                              Number.POSITIVE_INFINITY);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = tf.buffer(convInfo.outShape, x.dtype);\n    const outputVals = output.values;\n\n    const outputBatchStrides = convInfo.outShape[1] * convInfo.outShape[2] *\n        convInfo.outShape[3] * convInfo.outShape[4];\n    const outputDepthStrides =\n        convInfo.outShape[2] * convInfo.outShape[3] * convInfo.outShape[4];\n    const outputRowStrides = convInfo.outShape[3] * convInfo.outShape[4];\n    const outputColStrides = convInfo.outShape[4];\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      const outputBatchOffset = batch * outputBatchStrides;\n      const inputBatchOffset = batch * x.strides[0];\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          const outputDepthOffset =\n              outputBatchOffset + yDepth * outputDepthStrides;\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            const outputRowOffset = outputDepthOffset + yRow * outputRowStrides;\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n              // Shader code begins\n              const outputColOffset = outputRowOffset + yCol * outputColStrides;\n              let minMaxValue = initialValue;\n              let avgValue = 0;\n              let count = 0;\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const xDepthOffset = inputBatchOffset + xDepth * x.strides[1];\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const xRowOffset = xDepthOffset + xRow * x.strides[2];\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const xColOffset = xRowOffset + xCol * x.strides[3];\n                    const pixel = xValues[xColOffset + channel];\n                    if ((poolType === 'max' && pixel > minMaxValue)) {\n                      minMaxValue = pixel;\n                    } else if (poolType === 'avg') {\n                      avgValue += pixel;\n                      count++;\n                    }\n                    if (isNaN(minMaxValue)) {\n                      break;\n                    }\n                  }\n                  if (isNaN(minMaxValue)) {\n                    break;\n                  }\n                }\n                if (isNaN(minMaxValue)) {\n                  break;\n                }\n              }\n              const outputOffset = outputColOffset + channel;\n              outputVals[outputOffset] =\n                  poolType === 'avg' ? avgValue / count : minMaxValue;\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor5D;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'avgPool3d');\n\n    return this.pool3d(x, convInfo, 'avg').toFloat();\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([dy, x], 'avgPool3dBackprop');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins.\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel;\n                  }\n                }\n              }\n              dx.set(\n                  dotProd * avgMultiplier, batch, dxDepth, dxRow, dxCol,\n                  channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'maxPool3d');\n\n    return this.pool3d(x, convInfo, 'max').toFloat();\n  }\n\n  private maxPool3dPositions(x: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const maxPositions = tf.buffer(convInfo.outShape, 'int32');\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const xBuf = this.bufferSync(x);\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n\n              // Shader code begins\n              let maxValue = Number.NEGATIVE_INFINITY;\n              let maxPosition = -1;\n\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const wDepth = xDepth - xDepthCorner;\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const wRow = xRow - xRowCorner;\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const wCol = xCol - xColCorner;\n                    const pixel = xBuf.get(batch, xDepth, xRow, xCol, channel);\n                    if (pixel >= maxValue) {\n                      maxValue = pixel;\n                      maxPosition = wDepth * effectiveFilterHeight *\n                              effectiveFilterWidth +\n                          wRow * effectiveFilterHeight + wCol;\n                    }\n                  }\n                }\n              }\n\n              maxPositions.set(maxPosition, batch, yDepth, yRow, yCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return maxPositions.toTensor() as Tensor5D;\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([x, y], 'maxPool3dBackprop');\n\n    const maxPositions = this.maxPool3dPositions(x, convInfo);\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const maxPosBuf = this.bufferSync(maxPositions);\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const maxPos = effectiveFilterDepth *\n                            effectiveFilterHeight * effectiveFilterWidth -\n                        1 -\n                        maxPosBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    const curPos =\n                        wDepth * effectiveFilterHeight * effectiveFilterWidth +\n                        wRow * effectiveFilterWidth + wCol;\n\n                    const mask = maxPos === curPos ? 1 : 0;\n                    if (mask === 0) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel * mask;\n                  }\n                }\n              }\n              dx.set(dotProd, batch, dxDepth, dxRow, dxCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeBilinear');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result = new Float32Array(\n        util.sizeFromShape([batch, newHeight, newWidth, numChannels]));\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n    let outputIdx = 0;\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n    for (let b = 0; b < batch; b++) {\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceRowFloor = Math.floor(sourceFracRow);\n        const rowFrac = sourceFracRow - sourceRowFloor;\n        const sourceRowCeil = Math.min(oldHeight - 1, Math.ceil(sourceFracRow));\n        const topRowOffset = b * x.strides[0] + sourceRowFloor * x.strides[1];\n        const botRowOffset = b * x.strides[0] + sourceRowCeil * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceColFloor = Math.floor(sourceFracCol);\n          const colFrac = sourceFracCol - sourceColFloor;\n          const sourceColCeil =\n              Math.min(oldWidth - 1, Math.ceil(sourceFracCol));\n          const topLeftOffest = topRowOffset + sourceColFloor * x.strides[2];\n          const botLeftOffset = botRowOffset + sourceColFloor * x.strides[2];\n          const topRightOffset = topRowOffset + sourceColCeil * x.strides[2];\n          const botRightOffest = botRowOffset + sourceColCeil * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n\n            // Compute the fractional index of the source.\n            const topLeft = xValues[topLeftOffest + d];\n            const bottomLeft = xValues[botLeftOffset + d];\n            const topRight = xValues[topRightOffset + d];\n            const bottomRight = xValues[botRightOffest + d];\n\n            const top = topLeft + (topRight - topLeft) * colFrac;\n            const bottom = bottomLeft + (bottomRight - bottomLeft) * colFrac;\n            const newValue = top + (bottom - top) * rowFrac;\n\n            result[outputIdx++] = newValue;\n          }\n        }\n      }\n    }\n    return tf.tensor(result, [batch, newHeight, newWidth, numChannels]);\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeBilinearBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass and add the\n    // corresponding coefficient from dy to the gradient (with some\n    // interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/3039375c86a5bbc9610c7725dcaa95d635f87ba2/tensorflow/core/kernels/resize_bilinear_op.cc#L275\n\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    let offset = 0;\n    for (let b = 0; b < batch; b++) {\n      const bOffset = b * x.strides[0];\n      for (let r = 0; r < yHeight; r++) {\n        const dxR = r * heightScale;\n        const topDxRIndex = Math.floor(dxR);\n        const bottomDxRIndex = Math.min(Math.ceil(dxR), xHeight - 1);\n\n        const topDxROffset = bOffset + topDxRIndex * x.strides[1];\n        const bottomDxROffset = bOffset + bottomDxRIndex * x.strides[1];\n\n        const dxRLerp = dxR - topDxRIndex;\n        const inverseDxRLerp = 1.0 - dxRLerp;\n        for (let c = 0; c < yWidth; c++) {\n          const dxC = c * widthScale;\n          const leftDxCIndex = Math.floor(dxC);\n          const rightDxCIndex = Math.min(Math.ceil(dxC), xWidth - 1);\n          const dxCLerp = dxC - leftDxCIndex;\n          const inverseDxCLerp = 1.0 - dxCLerp;\n\n          const topLeftRCOffset = topDxROffset + leftDxCIndex * x.strides[2];\n          const topRightRCOffset = topDxROffset + rightDxCIndex * x.strides[2];\n          const bottomLeftRCOffset =\n              bottomDxROffset + leftDxCIndex * x.strides[2];\n          const bottomRightRCOffset =\n              bottomDxROffset + rightDxCIndex * x.strides[2];\n\n          const inverseDxRLerpTimesInverseDxCLerp =\n              inverseDxRLerp * inverseDxCLerp;\n          const inverseDxRLerpTimesDxCLerp = inverseDxRLerp * dxCLerp;\n          const dxRLerpTimesInverseDxCLerp = dxRLerp * inverseDxCLerp;\n          const dxRLerpTimesDxCLerp = dxRLerp * dxCLerp;\n          for (let d = 0; d < depth; d++) {\n            const dyVal = dyValues[offset++];\n            output[topLeftRCOffset + d] +=\n                dyVal * inverseDxRLerpTimesInverseDxCLerp;\n            output[topRightRCOffset + d] += dyVal * inverseDxRLerpTimesDxCLerp;\n            output[bottomLeftRCOffset + d] +=\n                dyVal * dxRLerpTimesInverseDxCLerp;\n            output[bottomRightRCOffset + d] += dyVal * dxRLerpTimesDxCLerp;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, [batch, xWidth, xHeight, depth], x.dtype);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeNearestNeighbor');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = new Float32Array(batch * newHeight * newWidth * numChannels);\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n\n    let outputOffset = 0;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceNearestRow = Math.min(\n            oldHeight - 1,\n            alignCorners ? Math.round(sourceFracRow) :\n                           Math.floor(sourceFracRow));\n        const rowOffset = batchOffset + sourceNearestRow * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceNearestCol = Math.min(\n              oldWidth - 1,\n              alignCorners ? Math.round(sourceFracCol) :\n                             Math.floor(sourceFracCol));\n          const colOffset = rowOffset + sourceNearestCol * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n            // Compute the fractional index of the source.\n            const newVal = xValues[colOffset + d];\n            output[outputOffset++] = newVal;\n          }\n        }\n      }\n    }\n    return tf.tensor(\n        output, [batch, newHeight, newWidth, numChannels], x.dtype);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeNearestNeighborBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    // Loop over the output space.\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < xHeight; r++) {\n        const rowOffset = batchOffset + r * x.strides[1];\n\n        // Compute bounds for where in dy we will look\n        const startRLerp = Math.floor(r * invHeightScale);\n        const startDyR = Math.floor(startRLerp - (winHeight / 2));\n        for (let c = 0; c < xWidth; c++) {\n          const colOffset = rowOffset + c * x.strides[2];\n\n          // Compute bounds for where in dy we will look\n          const startCLerp = Math.floor(c * invWidthScale);\n          const startDyC = Math.floor(startCLerp - (winWidth / 2));\n\n          for (let d = 0; d < depth; d++) {\n            let accum = 0;\n            // loop over dy\n\n            for (let dyRIndex = 0; dyRIndex < winHeight; dyRIndex++) {\n              const dyR = dyRIndex + startDyR;\n              // Guard against the window exceeding the bounds of dy\n              if (dyR < 0 || dyR >= yHeight) {\n                continue;\n              }\n\n              const dyROffset = batchOffset + dyR * dy.strides[1];\n              const sourceFracRow = dyR * heightScale;\n              const sourceNearestRow = Math.min(\n                  xHeight - 1,\n                  alignCorners ? Math.round(sourceFracRow) :\n                                 Math.floor(sourceFracRow));\n              if (r !== sourceNearestRow) {\n                continue;\n              }\n              for (let dyCIndex = 0; dyCIndex < winWidth; dyCIndex++) {\n                const dyC = dyCIndex + startDyC;\n                // Guard against the window exceeding the bounds of dy\n                if (dyC < 0 || dyC >= yWidth) {\n                  continue;\n                }\n\n                const dyCOffset = dyROffset + dyC * dy.strides[2];\n                const sourceFracCol = dyC * widthScale;\n                const sourceNearestCol = Math.min(\n                    xWidth - 1,\n                    alignCorners ? Math.round(sourceFracCol) :\n                                   Math.floor(sourceFracCol));\n\n                if (c === sourceNearestCol) {\n                  accum += dyValues[dyCOffset + d];\n                }\n              }\n            }\n            output[colOffset + d] = accum;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, x.shape, x.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(x, 'localResponseNormalization4D');\n\n    const channels = x.shape[3];\n    const maxD = channels - 1;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const size = x.size;\n    const result = new Float32Array(size);\n\n    function sumAcrossChannels(offset: number) {\n      const currentChannel = offset % channels;\n      let beginSumOffset =\n          offset - currentChannel + Math.max(0, currentChannel - depthRadius);\n      const endSumOffset = offset - currentChannel +\n          Math.min(currentChannel + depthRadius, maxD);\n\n      let sum = 0.0;\n      for (; beginSumOffset <= endSumOffset; beginSumOffset++) {\n        const z = xValues[beginSumOffset];\n        sum += z * z;\n      }\n      return sum;\n    }\n\n    for (let offset = 0; offset < size; offset++) {\n      const sum = sumAcrossChannels(offset);\n      const val = xValues[offset] * Math.pow(bias + alpha * sum, -beta);\n      result[offset] = val;\n    }\n\n    return tf.tensor4d(result, x.shape);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(dy, 'LRNGrad');\n    const channels = dy.shape[3];\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const inputImageValues = this.readSync(inputImage.dataId) as TypedArray;\n    const outputImageValues = this.readSync(outputImage.dataId) as TypedArray;\n    const result = new Float32Array(dy.size);\n    const size = dy.size;\n\n    for (let offset = 0; offset < size; offset++) {\n      const currentChannel = offset % channels;\n      const depthBegin =\n          (offset - currentChannel) + Math.max(0, currentChannel - depthRadius);\n      const depthEnd = (offset - currentChannel) +\n          Math.min(channels, currentChannel + depthRadius + 1);\n\n      let norm = 0;\n      for (let k = depthBegin; k < depthEnd; k++) {\n        norm += Math.pow(inputImageValues[k], 2);\n      }\n      norm = alpha * norm + bias;\n\n      for (let k = depthBegin; k < depthEnd; k++) {\n        let dyi = -2 * alpha * beta * inputImageValues[k] *\n            outputImageValues[offset] / norm;\n        if (offset === k) {\n          dyi += Math.pow(norm, -beta);\n        }\n        dyi *= dyValues[offset];\n        result[k] += dyi;\n      }\n    }\n    return tf.tensor4d(result, dy.shape);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    assertNotComplex(logits, 'multinomial');\n\n    const probabilities = normalized ? logits : tf.softmax(logits);\n    const batchSize = probabilities.shape[0];\n    const numEvents = probabilities.shape[1];\n    const res = tf.zeros<Rank.R2>([batchSize, numSamples], 'int32');\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const probVals = this.readSync(probabilities.dataId) as TypedArray;\n\n    for (let b = 0; b < batchSize; ++b) {\n      const offset = b * numEvents;\n      // The cdf won't include the last event. It will be implicit if no other\n      // event happened.\n      const cdf = new Float32Array(numEvents - 1);\n      cdf[0] = probVals[offset];\n      for (let event = 1; event < cdf.length; ++event) {\n        cdf[event] = cdf[event - 1] + probVals[offset + event];\n      }\n\n      const random = seedrandom.alea(seed.toString());\n      const outOffset = b * numSamples;\n      for (let sampleId = 0; sampleId < numSamples; ++sampleId) {\n        const r = random();\n\n        // Assume last event happened by default.\n        resVals[outOffset + sampleId] = cdf.length;\n\n        for (let event = 0; event < cdf.length; event++) {\n          if (r < cdf[event]) {\n            resVals[outOffset + sampleId] = event;\n            break;\n          }\n        }\n      }\n    }\n    return res;\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    assertNotComplex(indices, 'oneHot');\n\n    const res = new Float32Array(indices.size * depth);\n    res.fill(offValue);\n    const indicesVal = this.readSync(indices.dataId) as TypedArray;\n\n    for (let event = 0; event < indices.size; ++event) {\n      if (indicesVal[event] >= 0 && indicesVal[event] < depth) {\n        res[event * depth + indicesVal[event]] = onValue;\n      }\n    }\n    return tf.tensor2d(res, [indices.size, depth], 'int32');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold: number): Tensor1D {\n    assertNotComplex(boxes, 'nonMaxSuppression');\n\n    const boxesVals = this.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = this.readSync(scores.dataId) as TypedArray;\n    return nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        dataFormat === 'NHWC',\n        () => `Only NHWC dataFormat supported on CPU for depthToSpace. Got ${\n            dataFormat}`);\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = x.shape[1];\n    const inputWidth = x.shape[2];\n    const inputDepth = x.shape[3];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result =\n        new Float32Array(batchSize * outputHeight * outputWidth * outputDepth);\n\n    let outputIdx = 0;\n    for (let b = 0; b < batchSize; ++b) {\n      for (let h = 0; h < outputHeight; ++h) {\n        const inH = Math.floor(h / blockSize);\n        const offsetH = (h % blockSize);\n        for (let w = 0; w < outputWidth; ++w) {\n          const inW = Math.floor(w / blockSize);\n          const offsetW = (w % blockSize);\n          const offsetD = (offsetH * blockSize + offsetW) * outputDepth;\n          for (let d = 0; d < outputDepth; ++d) {\n            const inD = d + offsetD;\n            const inputIdx =\n                inD + inputDepth * (inW + inputWidth * (inH + inputHeight * b));\n            result[outputIdx++] = xValues[inputIdx];\n          }\n        }\n      }\n    }\n    return tf.tensor4d(\n        result, [batchSize, outputHeight, outputWidth, outputDepth]);\n  }\n\n  private broadcastedBinaryOp(\n      a: Tensor, b: Tensor, dtype: DataType,\n      op: (a: number, b: number) => number): Tensor {\n    const newShape = backend_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const result = tf.buffer(newShape, dtype);\n    const aVals = this.readSync(a.dataId) as TypedArray;\n    const bVals = this.readSync(b.dataId) as TypedArray;\n    const aBroadcastDims = backend_util.getBroadcastDims(a.shape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(b.shape, newShape);\n\n    const resVals = result.values;\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resVals.length; ++i) {\n        resVals[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      const aBuf = this.bufferSync(a);\n      const bBuf = this.bufferSync(b);\n      for (let i = 0; i < resVals.length; ++i) {\n        const loc = result.indexToLoc(i);\n\n        const aLoc = loc.slice(-a.rank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = aBuf.locToIndex(aLoc);\n\n        const bLoc = loc.slice(-b.rank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = bBuf.locToIndex(bLoc);\n\n        resVals[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n    return result.toTensor();\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  dispose() {}\n\n  floatPrecision(): 16|32 {\n    return 32;\n  }\n\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return super.epsilon();\n  }\n\n  cropAndResize(\n      images: Tensor4D,\n      boxes: Tensor2D,\n      boxIndex: Tensor1D,\n      cropSize: [number, number],\n      method: string,\n      extrapolationValue: number,\n  ) {\n    const [batch, imageHeight, imageWidth, numChannels] = images.shape;\n    const numBoxes = boxes.shape[0];\n\n    const [cropHeight, cropWidth] = cropSize;\n    const output =\n        tf.buffer([numBoxes, cropHeight, cropWidth, numChannels], 'float32');\n\n    const boxVals = this.readSync(boxes.dataId) as TypedArray;\n    const boxIndVals = this.readSync(boxIndex.dataId) as TypedArray;\n    const imageVals = this.readSync(images.dataId) as TypedArray;\n\n    const inStride = images.strides;   // to calculate flat indexes into image\n    const outStride = output.strides;  // to calculate flat indexes into output\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op.cc\n    for (let b = 0; b < numBoxes; b++) {\n      const startInd = b * 4;\n      const y1 = boxVals[startInd];\n      const x1 = boxVals[startInd + 1];\n      const y2 = boxVals[startInd + 2];\n      const x2 = boxVals[startInd + 3];\n\n      const bInd: number = boxIndVals[b];\n      if (bInd >= batch) {\n        continue;\n      }\n\n      const heightScale = (cropHeight > 1) ?\n          (y2 - y1) * (imageHeight - 1) / (cropHeight - 1) :\n          0;\n      const widthScale =\n          (cropWidth > 1) ? (x2 - x1) * (imageWidth - 1) / (cropWidth - 1) : 0;\n\n      for (let y = 0; y < cropHeight; y++) {\n        const yInd: number = (cropHeight > 1) ?\n            y1 * (imageHeight - 1) + y * (heightScale) :\n            0.5 * (y1 + y2) * (imageHeight - 1);\n\n        if (yInd < 0 || yInd > imageHeight - 1) {\n          for (let x = 0; x < cropWidth; x++) {\n            for (let c = 0; c < numChannels; c++) {\n              const ind =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = extrapolationValue;\n            }\n          }\n          continue;\n        }\n\n        if (method === 'bilinear') {\n          const topInd = Math.floor(yInd);\n          const bottomInd = Math.ceil(yInd);\n          const yLerp = yInd - topInd;\n\n          for (let x = 0; x < cropWidth; x++) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const leftInd = Math.floor(xInd);\n            const rightInd = Math.ceil(xInd);\n            const xLerp = xInd - leftInd;\n\n            for (let c = 0; c < numChannels; c++) {\n              let ind = c + leftInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topRight = imageVals[ind];\n\n              ind = c + leftInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomRight = imageVals[ind];\n\n              const top = topLeft + (topRight - topLeft) * xLerp;\n              const bottom = bottomLeft + (bottomRight - bottomLeft) * xLerp;\n\n              ind = c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = top + ((bottom - top) * yLerp);\n            }\n          }\n        } else {  // method == \"nearest\"\n          for (let x = 0; x < cropWidth; ++x) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const closestX = Math.round(xInd);\n            const closestY = Math.round(yInd);\n            for (let c = 0; c < numChannels; c++) {\n              const inInd = c + closestX * inStride[2] +\n                  closestY * inStride[1] + bInd * inStride[0];\n              const outInd =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[outInd] = imageVals[inInd];\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor4D;\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n    const sumDupeIndices = false;\n    return this.scatter(\n        sparseIndices, sparseValues, outputShape, outputSize, sliceSize,\n        numUpdates, sliceRank, strides, defaultValue, sumDupeIndices);\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n    if (numSlices === 0) {\n      return tf.tensor([], resultShape, x.dtype);\n    }\n\n    const buffer = new TensorBuffer([numSlices, sliceSize], x.dtype);\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const xData = this.readSync(x.dataId) as TypedArray;\n\n    for (let i = 0; i < numSlices; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        flattenIndex += dim * strides[j];\n        index.push(dim);\n      }\n      if (flattenIndex < 0 || flattenIndex >= x.size / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${x.shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        buffer.values[i * sliceSize + k] = xData[flattenIndex * sliceSize + k];\n      }\n    }\n    return buffer.toTensor().reshape(resultShape);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n    const defaultValue = tf.scalar(0);\n    const sumDupeIndices = true;\n    return this.scatter(\n        indices, updates, shape, outputSize, sliceSize, numUpdates, sliceRank,\n        strides, defaultValue, sumDupeIndices);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n    const values =\n        util.getArrayFromDType(dtype, util.sizeFromShape(shape)) as TypedArray;\n    values.fill(value as number);\n    return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported for string tensors');\n    } else {\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    const values = util.getArrayFromDType(\n                       x.dtype, util.sizeFromShape(x.shape)) as TypedArray;\n    return this.makeOutput(values, x.shape, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  private scatter<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R], outputSize: number,\n      sliceSize: number, numUpdates: number, sliceRank: number,\n      strides: number[], defaultValue: Scalar,\n      sumDupeIndices: boolean): Tensor<R> {\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const updatesData = this.readSync(updates.dataId) as TypedArray;\n\n    if (outputSize === 0) {\n      return tf.tensor([], shape, updates.dtype);\n    }\n\n    const buffer = new TensorBuffer(flattenShape, updates.dtype as 'float32');\n    buffer.values.fill((this.readSync(defaultValue.dataId) as TypedArray)[0]);\n\n    for (let i = 0; i < numUpdates; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        index.push(dim);\n        flattenIndex += dim * strides[j];\n      }\n\n      if (flattenIndex < 0 || flattenIndex >= outputSize / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        if (sumDupeIndices) {\n          buffer.values[flattenIndex * sliceSize + k] +=\n              updatesData[i * sliceSize + k];\n        } else {\n          buffer.values[flattenIndex * sliceSize + k] = updates.rank === 0 ?\n              updatesData[0] :\n              updatesData[i * sliceSize + k];\n        }\n      }\n    }\n    return buffer.toTensor().reshape(shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs, AbsInputs, KernelConfig, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function simpleAbsImpl(vals: TypedArray): Float32Array {\n  const resultValues = new Float32Array(vals.length);\n  for (let i = 0; i < vals.length; ++i) {\n    resultValues[i] = Math.abs(vals[i]);\n  }\n  return resultValues;\n}\n\nexport const absKernelFunc =\n    (args: {inputs: AbsInputs, backend: MathBackendCPU}) => {\n      const {x} = args.inputs;\n      const cpuBackend = args.backend;\n      let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n      if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n        resultValues = simpleAbsImpl(values);\n      } else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals =\n            cpuBackend.data.get(real.dataId).values as Float32Array;\n        const imagVals =\n            cpuBackend.data.get(imag.dataId).values as Float32Array;\n        for (let i = 0; i < realVals.length; i++) {\n          const real = realVals[i];\n          const imag = imagVals[i];\n          resultValues[i] = Math.hypot(real, imag);\n        }\n      }\n      return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n    };\n\nexport const absConfig: KernelConfig = {\n  kernelName: Abs,\n  backendName: 'cpu',\n  kernelFunc: absKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleBinaryKernelImpl, SimpleBinaryOperation} from './binary_types';\n\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op: SimpleBinaryOperation):\n    SimpleBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aVals: TypedArray,\n          bVals: TypedArray, dtype: DataType): [TypedArray, number[]] => {\n    const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    const resultRank = newShape.length;\n    const resultStrides = util.computeStrides(newShape);\n    const resultSize = util.sizeFromShape(newShape);\n\n    const result =\n        util.getTypedArrayFromDType(dtype as NumericDataType, resultSize);\n\n    const aRank = aShape.length;\n    const bRank = bShape.length;\n\n    const aStrides = util.computeStrides(aShape);\n    const bStrides = util.computeStrides(bShape);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < result.length; ++i) {\n        result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      for (let i = 0; i < result.length; ++i) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        result[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n\n    return [result, newShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Complex, ComplexInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function complex(args: {inputs: ComplexInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {real, imag} = inputs;\n\n  const realVals = backend.data.get(real.dataId).values as TypedArray;\n  const imagVals = backend.data.get(imag.dataId).values as TypedArray;\n\n  const complexInfo = backend.makeTensorInfo(real.shape, 'complex64');\n\n  const complex = backend.data.get(complexInfo.dataId);\n\n  // The complex tensor owns the underlying real and imag tensorInfos, only the\n  // complex tensor tracks refCount, when complexData is disposed the\n  // underlying tensorData will be disposed.\n  complex.complexTensorInfos = {\n    real: backend.makeTensorInfo(real.shape, 'float32', realVals),\n    imag: backend.makeTensorInfo(imag.shape, 'float32', imagVals)\n  };\n\n  return complexInfo;\n}\n\nexport const complexConfig: KernelConfig = {\n  kernelName: Complex,\n  backendName: 'cpu',\n  kernelFunc: complex as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendCPU}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'cpu',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Real, RealInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function real(args: {inputs: RealInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const real = backend.data.get(input.dataId).complexTensorInfos.real;\n  const realVal = backend.data.get(real.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the real value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(real.shape, real.dtype, realVal);\n}\n\nexport const realConfig: KernelConfig = {\n  kernelName: Real,\n  backendName: 'cpu',\n  kernelFunc: real as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport * as tf from '@tensorflow/tfjs-core';\nimport {Cast, CastAttrs, CastInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\n\nimport {complex} from './Complex';\nimport {identity} from './Identity';\nimport {real} from './Real';\n\nexport function cast(\n    args: {inputs: CastInputs, backend: MathBackendCPU, attrs: CastAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {dtype} = attrs;\n\n  // Casting to complex64.\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return identity({inputs: {x}, backend});\n    }\n\n    // TODO(lina128): Import kernel function once zeros is modularized.\n    const zerosTensor = tf.zeros(x.shape);\n    const floatX = cast({inputs: {x}, backend, attrs: {dtype: 'float32'}});\n\n    const result =\n        complex({inputs: {real: floatX, imag: zerosTensor}, backend});\n\n    zerosTensor.dispose();\n    backend.disposeIntermediateTensorInfo(floatX);\n\n    return result;\n  }\n\n  // Casting from complex64\n  if (x.dtype === 'complex64') {\n    const realPart = real({inputs: {input: x}, backend});\n    const result = cast({inputs: {x: realPart}, backend, attrs: {dtype}});\n\n    backend.disposeIntermediateTensorInfo(realPart);\n\n    return result;\n  }\n\n  if (!util.hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    const result = identity({inputs: {x}, backend});\n    return {dataId: result.dataId, shape: result.shape, dtype};\n  }\n\n  if (dtype === 'int32') {\n    const values = backend.data.get(x.dataId).values as TypedArray;\n    const resultValues = Int32Array.from(values);\n    return backend.makeTensorInfo(x.shape, 'int32', resultValues);\n  }\n\n  if (dtype === 'bool') {\n    // This is essentially the result of notEqual(x, 0). We avoid using\n    // kernel notEqual to avoid circular dependency, i.e. binary_utils ->\n    // cast -> notEqual -> binary_utils.\n    const xVals = backend.data.get(x.dataId).values as TypedArray;\n    const zero = util.toTypedArray([0], x.dtype);\n\n    const [resultData, resultShape] = createSimpleBinaryKernelImpl(\n        (a, b) => (a !== b) ? 1 : 0)(x.shape, [], xVals, zero, 'bool');\n\n    return backend.makeTensorInfo(resultShape, 'bool', resultData);\n  }\n\n  throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n}\n\nexport const castConfig: KernelConfig = {\n  kernelName: Cast,\n  backendName: 'cpu',\n  kernelFunc: cast as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, BinaryInputs, DataType, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {cast} from '../kernels/Cast';\nimport {complex} from '../kernels/Complex';\n\nimport {ComplexBinaryKernelImpl, ComplexBinaryOperation, SimpleBinaryKernelImpl} from './binary_types';\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param name Kernel name.\n * @param binaryKernelImpl A `SimpleBinaryKernelImpl` for the kernel.\n * @param binaryKernelComplexImpl Optional. If exists, represents a\n *     `ComplexBinaryKernelImpl` for the kernel, will be used when input dtype\n *     is `complex64`.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    name: string, simpleImpl: SimpleBinaryKernelImpl,\n    complexImpl?: ComplexBinaryKernelImpl, dtype?: DataType): KernelFunc {\n  if (complexImpl == null) {\n    return ({inputs, backend}) => {\n      const {a, b} = inputs as BinaryInputs;\n      const cpuBackend = backend as MathBackendCPU;\n\n      assertNotComplex([a, b], name);\n\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    };\n  }\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    if (a.dtype === 'complex64' || b.dtype === 'complex64') {\n      const $aComplex = cast(\n          {inputs: {x: a}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $aComplexVals = cpuBackend.data.get($aComplex.dataId);\n\n      const aReal = $aComplexVals.complexTensorInfos.real;\n      const aImag = $aComplexVals.complexTensorInfos.imag;\n\n      const aRealVals =\n          cpuBackend.data.get(aReal.dataId).values as Float32Array;\n      const aImagVals =\n          cpuBackend.data.get(aImag.dataId).values as Float32Array;\n\n      const $bComplex = cast(\n          {inputs: {x: b}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $bComplexVals = cpuBackend.data.get($bComplex.dataId);\n\n      const bReal = $bComplexVals.complexTensorInfos.real;\n      const bImag = $bComplexVals.complexTensorInfos.imag;\n\n      const bRealVals =\n          cpuBackend.data.get(bReal.dataId).values as Float32Array;\n      const bImagVals =\n          cpuBackend.data.get(bImag.dataId).values as Float32Array;\n\n      const [resultRealData, resultImagData, resultShape] = complexImpl(\n          a.shape, b.shape, aRealVals, aImagVals, bRealVals, bImagVals);\n\n      const resultReal =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultRealData);\n\n      const resultImag =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultImagData);\n\n      const result = complex(\n          {inputs: {real: resultReal, imag: resultImag}, backend: cpuBackend});\n\n      cpuBackend.disposeIntermediateTensorInfo($aComplex);\n      cpuBackend.disposeIntermediateTensorInfo($bComplex);\n      cpuBackend.disposeIntermediateTensorInfo(resultReal);\n      cpuBackend.disposeIntermediateTensorInfo(resultImag);\n\n      return result;\n    } else {\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    }\n  };\n}\n\n/**\n * Template that creates the complex type implementation for binary ops.\n * Supports broadcast.\n */\nexport function createComplexBinaryKernelImpl(op: ComplexBinaryOperation):\n    ComplexBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aRealVals: Float32Array,\n          aImagVals: Float32Array, bRealVals: Float32Array,\n          bImagVals: Float32Array): [TypedArray, TypedArray, number[]] => {\n    const resultShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const resultSize = util.sizeFromShape(resultShape);\n    const resultRank = resultShape.length;\n    const resultStrides = util.computeStrides(resultShape);\n\n    const resultRealVals = util.getTypedArrayFromDType('float32', resultSize);\n    const resultImagVals = util.getTypedArrayFromDType('float32', resultSize);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, resultShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, resultShape);\n\n    const aVals = backend_util.mergeRealAndImagArrays(aRealVals, aImagVals);\n    const bVals = backend_util.mergeRealAndImagArrays(bRealVals, bImagVals);\n\n    const aRank = aShape.length;\n    const aStrides = util.computeStrides(aShape);\n\n    const bRank = bShape.length;\n    const bStrides = util.computeStrides(bShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const aIdx = i % aVals.length;\n        const bIdx = i % bVals.length;\n\n        const result =\n            op(aVals[aIdx * 2], aVals[aIdx * 2 + 1], bVals[bIdx * 2],\n               bVals[bIdx * 2 + 1]);\n\n        resultRealVals[i] = result.real;\n        resultImagVals[i] = result.imag;\n      }\n    } else {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        const opResult =\n            op(aVals[aIndex * 2], aVals[aIndex * 2 + 1], bVals[bIndex * 2],\n               bVals[bIndex * 2 + 1]);\n\n        resultRealVals[i] = opResult.real;\n        resultImagVals[i] = opResult.imag;\n      }\n    }\n    return [resultRealVals, resultImagVals, resultShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Add, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal + bReal, imag: aImag + bImag};\n    }));\n\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\n\nexport const addConfig: KernelConfig = {\n  kernelName: Add,\n  backendName: 'cpu',\n  kernelFunc: add\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NumericDataType, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op: SimpleUnaryOperation):\n    SimpleUnaryImpl {\n  return (values, dtype, attrs) => {\n    const newValues =\n        util.getTypedArrayFromDType(dtype as NumericDataType, values.length);\n    for (let i = 0; i < values.length; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return newValues;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelFunc, TypedArray, UnaryInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param name Kernel name.\n * @param op A `SimpleUnaryOperation` for the kernel.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFunc(\n    name: string, op: SimpleUnaryOperation, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xSize = util.sizeFromShape(x.shape);\n    const $dtype = dtype || x.dtype;\n    const newValues = util.getArrayFromDType($dtype, xSize);\n    for (let i = 0; i < xSize; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for unary ops from the given\n * `SimpleUnaryImpl`..\n * @param name Kernel name.\n * @param unaryImpl A `SimpleUnaryImpl` that implements the op.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFuncFromImpl(\n    name: string, unaryImpl: SimpleUnaryImpl, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const $dtype = dtype || x.dtype;\n    const newValues = unaryImpl(values, $dtype, attrs);\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\n\nexport const ceilConfig: KernelConfig = {\n  kernelName: Ceil,\n  backendName: 'cpu',\n  kernelFunc: ceilKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\n\nexport const expConfig: KernelConfig = {\n  kernelName: Exp,\n  backendName: 'cpu',\n  kernelFunc: expKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\n\nexport const expm1Config: KernelConfig = {\n  kernelName: Expm1,\n  backendName: 'cpu',\n  kernelFunc: expm1KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\n\nexport const floorConfig: KernelConfig = {\n  kernelName: Floor,\n  backendName: 'cpu',\n  kernelFunc: floorKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\n\nexport const logConfig: KernelConfig = {\n  kernelName: Log,\n  backendName: 'cpu',\n  kernelFunc: logKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function maxImpl(\n    aVals: TypedArray, reduceSize: number, outShape: number[],\n    dtype: DataType): TypedArray {\n  const vals = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(outShape));\n\n  for (let i = 0; i < vals.length; ++i) {\n    const offset = i * reduceSize;\n    let max = aVals[offset];\n    for (let j = 0; j < reduceSize; ++j) {\n      const value = aVals[offset + j];\n      if (value > max) {\n        max = value;\n      }\n    }\n    vals[i] = max;\n  }\n  return vals;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Multiply} from '@tensorflow/tfjs-core';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const multiplyImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n      };\n    }));\n\nexport const multiply =\n    binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\n\nexport const multiplyConfig: KernelConfig = {\n  kernelName: Multiply,\n  backendName: 'cpu',\n  kernelFunc: multiply\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Rsqrt} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\n\nexport const rsqrtConfig: KernelConfig = {\n  kernelName: Rsqrt,\n  backendName: 'cpu',\n  kernelFunc: rsqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelConfig, KernelFunc, NumericDataType, Slice, slice_util, SliceAttrs, SliceInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function sliceImpl(\n    vals: TypedArray, begin: number[], size: number[], shape: number[],\n    dtype: DataType): TypedArray {\n  const isContinous = slice_util.isSliceContinous(shape, begin, size);\n  const length = util.sizeFromShape(size);\n  const xStrides = util.computeStrides(shape);\n\n  if (isContinous) {\n    const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n    return vals.subarray(flatOffset, flatOffset + length);\n  }\n\n  const outVals = util.getTypedArrayFromDType(dtype as NumericDataType, length);\n  for (let i = 0; i < length; ++i) {\n    const rank = size.length;\n    const strides = util.computeStrides(size);\n    const loc = util.indexToLoc(i, rank, strides);\n    const xLoc = loc.map((idx: number, j) => idx + begin[j]);\n    const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n    outVals[i] = vals[xIndex];\n  }\n  return outVals;\n}\n\nexport function slice(\n    args: {inputs: SliceInputs, backend: MathBackendCPU, attrs: SliceAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {begin, size} = attrs;\n\n  assertNotComplex(x, 'slice');\n\n  const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n  slice_util.assertParamsValid(x, $begin, $size);\n\n  const vals = backend.data.get(x.dataId).values as TypedArray;\n  const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n  return backend.makeTensorInfo($size, x.dtype, outVals);\n}\n\nexport const sliceConfig: KernelConfig = {\n  kernelName: Slice,\n  backendName: 'cpu',\n  kernelFunc: slice as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sub} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const subImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal - bReal, imag: aImag - bImag};\n    }));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\n\nexport const subConfig: KernelConfig = {\n  kernelName: Sub,\n  backendName: 'cpu',\n  kernelFunc: sub\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function transposeImpl(\n    xVals: TypedArray, xShape: number[], dtype: DataType, perm: number[],\n    newShape: number[]): TypedArray {\n  const xRank = xShape.length;\n  const xSize = util.sizeFromShape(xShape);\n  const xStrides = util.computeStrides(xShape);\n  const newStrides = util.computeStrides(newShape);\n\n  const result = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(newShape));\n\n  for (let i = 0; i < xSize; ++i) {\n    const loc = util.indexToLoc(i, xRank, xStrides);\n\n    // Permute location.\n    const newLoc: number[] = new Array(loc.length);\n    for (let i = 0; i < newLoc.length; i++) {\n      newLoc[i] = loc[perm[i]];\n    }\n\n    const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n    result[newIndex] = xVals[i];\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendValues, DataType, TensorBuffer, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function uniqueImpl(\n    values: BackendValues, axis: number, shape: number[], dtype: DataType): {\n  outputValues: BackendValues,\n  outputShape: number[],\n  indices: BackendValues\n} {\n  // Normalize and validate axis.\n  const $axis = util.parseAxisParam(axis, shape)[0];\n\n  // Calculate the new shape that is suitable for extracting data along the\n  // given axis.\n  //\n  // The rank is 3.\n  // The size of the 1st dimension is the size of all the axes < the given axis.\n  // The size of the 2nd dimension is the same as the size of the given axis.\n  // The size of the 3rd dimension is the size of all the axes > the given axis.\n  //\n  // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n  // newShape would be: [2*3, 5, 4].\n  //\n  // Note that this is not the final output shape. This will be the shape for an\n  // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n  // values along the given axis. To demonstrate how it works, consider the\n  // following example:\n  //\n  // Input: a 3D tensor, with shape [1, 2, 3]\n  // [\n  //   [\n  //      [1,2,3],\n  //      [4,5,6]\n  //   ]\n  // ]\n  // Axis: 2 (the last axis).\n  // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n  //\n  // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n  // 1*2. The re-shaped data would look like:\n  //\n  // [\n  //   [\n  //     [1], [2], [3]\n  //   ],\n  //   [\n  //     [4], [5], [6]\n  //   ]\n  // ]\n  //\n  // Then, we can construct a 3-level nested loop by the following dimension\n  // order to extract the values along the axis (dimension1):\n  // i: dimension1       // 0,1,2 (newShape[1])\n  //   m: dimension0     // 0,1   (newShape[0])\n  //     n: dimension2   // 0     (newShape[2])\n  //\n  //                       m, i, n\n  //                      ---------\n  // Iteration 0: data at [0, 0, 0] => \"1\"\n  // Iteration 1: data at [1, 0, 0] => \"4\"\n  // We got [1,4].\n  // Iteration 2: data at [0, 1, 0] => \"2\"\n  // Iteration 3: data at [1, 1, 0] => \"5\"\n  // We got [2,5].\n  // Iteration 4: data at [0, 2, 0] => \"3\"\n  // Iteration 5: data at [1, 2, 0] => \"6\"\n  // We got [3,6].\n  const newShape = [1, shape[0], 1];\n  for (let i = 0; i < $axis; i++) {\n    newShape[0] *= shape[i];\n  }\n  newShape[1] = shape[$axis];\n  for (let i = $axis + 1; i < shape.length; i++) {\n    newShape[2] *= shape[i];\n  }\n\n  // A map from unique elements (their string representations) to their values\n  // in \"indices\" (below).\n  const uniqueElements: {[key: string]: number} = {};\n  // The indices of each unique element in the original tensor along the given\n  // axis. It is 1D and has the same size as the given axis.\n  const indices = new Int32Array(shape[$axis]);\n  // Create a buffer so we can easily extract value at a given location.\n  const inputBuffer = new TensorBuffer(newShape, dtype, values as TypedArray);\n  // The indices along the given axis that have unique elements. This is a\n  // de-duped version of \"indices\" above.\n  const uniqueIndices: number[] = [];\n  const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n  for (let i = 0; i < shape[$axis]; i++) {\n    // Extract values along the axis.\n    let element: string;\n    if (is1DTensor) {\n      // Fast path for 1D tensor input.\n      element = values[i].toString();\n    } else {\n      const axisValues = [];\n      for (let m = 0; m < newShape[0]; m++) {\n        for (let n = 0; n < newShape[2]; n++) {\n          axisValues.push(inputBuffer.get(m, i, n));\n        }\n      }\n      element = axisValues.join(',');\n    }\n\n    // Dedup and update various indices.\n    if (uniqueElements[element] !== undefined) {\n      indices[i] = uniqueElements[element];\n    } else {\n      const uniqueIndex = Object.keys(uniqueElements).length;\n      uniqueElements[element] = uniqueIndex;\n      indices[i] = uniqueIndex;\n      uniqueIndices.push(i);\n    }\n  }\n\n  // Now we know where each of the unique elements are located along the axis\n  // (uniqueIndices). Extract them from input buffer and store them in the\n  // output buffer.\n  const outputTmpShape = newShape.slice();\n  outputTmpShape[1] = Object.keys(uniqueElements).length;\n  const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n  uniqueIndices.forEach((uniqueElementIndex, i) => {\n    for (let m = 0; m < newShape[0]; m++) {\n      for (let n = 0; n < newShape[2]; n++) {\n        outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n      }\n    }\n  });\n\n  // The output shape can be calculated from the input shape with the size of\n  // the given axis replaced by the number of unique elements along that axis.\n  const outputShape = shape.slice();\n  outputShape[$axis] = outputTmpShape[1];\n\n  return {\n    outputValues: outputBuffer.values as BackendValues,\n    outputShape,\n    indices,\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Shared kernel impls for use in other backends.\nexport {simpleAbsImpl} from './kernels/Abs';\nexport {addImpl} from './kernels/Add';\nexport {ceilImpl} from './kernels/Ceil';\nexport {expImpl} from './kernels/Exp';\nexport {expm1Impl} from './kernels/Expm1';\nexport {floorImpl} from './kernels/Floor';\nexport {logImpl} from './kernels/Log';\nexport {maxImpl} from './kernels/Max_impl';\nexport {multiplyImpl} from './kernels/Multiply';\nexport {rsqrtImpl} from './kernels/Rsqrt';\nexport {sliceImpl} from './kernels/Slice';\nexport {subImpl} from './kernels/Sub';\nexport {transposeImpl} from './kernels/Transpose_impl';\nexport {uniqueImpl} from './kernels/Unique_impl';\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/*\n * base.ts contains all the exports from tfjs-backend-cpu\n * without auto-kernel registration\n */\nimport {registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from './backend_cpu';\nimport * as shared from './shared';\n\nexport {MathBackendCPU} from './backend_cpu';\nexport {version as version_cpu} from './version';\nexport {shared};\n\n// Side effects for default initialization of MathBackendCPU\nregisterBackend('cpu', () => new MathBackendCPU(), 1 /* priority */);\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acosKernelFunc = unaryKernelFunc(Acos, (xi) => Math.acos(xi));\n\nexport const acosConfig: KernelConfig = {\n  kernelName: Acos,\n  backendName: 'cpu',\n  kernelFunc: acosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acoshKernelFunc = unaryKernelFunc(Acosh, (xi) => Math.acosh(xi));\n\nexport const acoshConfig: KernelConfig = {\n  kernelName: Acosh,\n  backendName: 'cpu',\n  kernelFunc: acoshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinKernelFunc = unaryKernelFunc(Asin, (xi) => Math.asin(xi));\n\nexport const asinConfig: KernelConfig = {\n  kernelName: Asin,\n  backendName: 'cpu',\n  kernelFunc: asinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinhKernelFunc = unaryKernelFunc(Asinh, (xi) => Math.asinh(xi));\n\nexport const asinhConfig: KernelConfig = {\n  kernelName: Asinh,\n  backendName: 'cpu',\n  kernelFunc: asinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanKernelFunc = unaryKernelFunc(Atan, (xi) => Math.atan(xi));\n\nexport const atanConfig: KernelConfig = {\n  kernelName: Atan,\n  backendName: 'cpu',\n  kernelFunc: atanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanhKernelFunc = unaryKernelFunc(Atanh, (xi) => Math.atanh(xi));\n\nexport const atanhConfig: KernelConfig = {\n  kernelName: Atanh,\n  backendName: 'cpu',\n  kernelFunc: atanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, buffer, DataType, Rank, TensorBuffer, TypedArray} from '@tensorflow/tfjs-core';\n\nexport function pool(\n    xValues: TypedArray, xShape: number[], dtype: DataType, strides: number[],\n    convInfo: backend_util.Conv2DInfo,\n    poolType: 'max'|'avg'): TensorBuffer<Rank, DataType> {\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const initialValue =\n      (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                            Number.POSITIVE_INFINITY);\n\n  const output = buffer(convInfo.outShape, dtype);\n  const outputVals = output.values;\n\n  const outputBatchStrides =\n      convInfo.outShape[1] * convInfo.outShape[2] * convInfo.outShape[3];\n  const outputRowStrides = convInfo.outShape[2] * convInfo.outShape[3];\n  const outputColStrides = convInfo.outShape[3];\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    const outputBatchOffset = b * outputBatchStrides;\n    const inputBatchOffset = b * strides[0];\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        const outputRowOffset = outputBatchOffset + yR * outputRowStrides;\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          const xCMin = Math.max(0, xCCorner);\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let minMaxValue = initialValue;\n          let avgValue = 0;\n          let count = 0;\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const xROffset = inputBatchOffset + xR * strides[1];\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const xCOffset = xROffset + xC * strides[2];\n              const pixel = xValues[xCOffset + d];\n              if ((poolType === 'max' && pixel > minMaxValue)) {\n                minMaxValue = pixel;\n              } else if (poolType === 'avg') {\n                avgValue += pixel;\n                count++;\n              }\n            }\n            if (isNaN(minMaxValue)) {\n              break;\n            }\n          }\n          const outputOffset = outputRowOffset + yC * outputColStrides + d;\n          outputVals[outputOffset] =\n              poolType === 'avg' ? avgValue / count : minMaxValue;\n        }\n      }\n    }\n  }\n  return output;\n}\n\nexport function maxPoolPositions(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    convInfo: backend_util.Conv2DInfo, flattenPositions = false,\n    includeBatchInIndex = false): TensorBuffer<Rank, 'int32'> {\n  const maxPositions = buffer(convInfo.outShape, 'int32');\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const xBuf = buffer(xShape, dtype, xValues);\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        let xRMin = xRCorner;\n        while (xRMin < 0) {\n          xRMin += dilationHeight;\n        }\n        // const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          let xCMin = xCCorner;\n          while (xCMin < 0) {\n            xCMin += dilationWidth;\n          }\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let maxValue = Number.NEGATIVE_INFINITY;\n          let maxPosition = -1;\n\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const wR = xR - xRCorner;\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const wC = xC - xCCorner;\n              const pixel = xBuf.get(b, xR, xC, d);\n              if (pixel > maxValue) {\n                maxValue = pixel as number;\n                if (flattenPositions) {\n                  maxPosition = includeBatchInIndex ?\n                      ((b * convInfo.inHeight + xR) * convInfo.inWidth + xC) *\n                              convInfo.inChannels +\n                          d :\n                      (xR * convInfo.inWidth + xC) * convInfo.inChannels + d;\n                } else {\n                  maxPosition = wR * effectiveFilterWidth + wC;\n                }\n              }\n            }\n          }\n          maxPositions.set(maxPosition, b, yR, yC, d);\n        }\n      }\n    }\n  }\n  return maxPositions;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function avgPool(\n    args:\n        {inputs: AvgPoolInputs, backend: MathBackendCPU, attrs: AvgPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'avg');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'cpu',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, buffer, KernelConfig, KernelFunc, Rank, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const filterHeight = convInfo.filterHeight;\n  const filterWidth = convInfo.filterWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel;\n            }\n          }\n          dx.set(dotProd * avgMultiplier, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function batchNormKernelFunc(args: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendCPU,\n  attrs: FusedBatchNormAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x, scale, offset, mean, variance} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  assertNotComplex([x, mean, variance, scale, offset], 'batchNorm');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const mVals = backend.data.get(mean.dataId).values as TypedArray;\n  const varVals = backend.data.get(variance.dataId).values as TypedArray;\n  const sVals = scale ? backend.data.get(scale.dataId).values as TypedArray :\n                        new Float32Array([1]);\n  const offVals = offset ?\n      backend.data.get(offset.dataId).values as TypedArray :\n      new Float32Array([0]);\n  const outVals = new Float32Array(xVals.length);\n\n  const offValsLength = offVals.length;\n  const sValsLength = sVals.length;\n  const varValsLength = varVals.length;\n  const mValsLength = mVals.length;\n\n  let offi = 0;\n  let mi = 0;\n  let si = 0;\n  let vi = 0;\n  for (let i = 0; i < xVals.length; ++i) {\n    outVals[i] = offVals[offi++] +\n        (xVals[i] - mVals[mi++]) * sVals[si++] /\n            Math.sqrt(varVals[vi++] + varianceEpsilon);\n    if (offi >= offValsLength) {\n      offi = 0;\n    }\n    if (mi >= mValsLength) {\n      mi = 0;\n    }\n    if (si >= sValsLength) {\n      si = 0;\n    }\n    if (vi >= varValsLength) {\n      vi = 0;\n    }\n  }\n  return backend.makeTensorInfo(x.shape, x.dtype, outVals);\n}\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'cpu',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const clipKernelFunc = unaryKernelFunc(ClipByValue, (xi, attrs) => {\n  const clipAttrs = attrs as {} as ClipByValueAttrs;\n  if (xi > clipAttrs.clipValueMax) {\n    return clipAttrs.clipValueMax;\n  }\n  return xi < clipAttrs.clipValueMin ? clipAttrs.clipValueMin : xi;\n});\n\nexport const clipConfig: KernelConfig = {\n  kernelName: ClipByValue,\n  backendName: 'cpu',\n  kernelFunc: clipKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Imag, ImagInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function imag(args: {inputs: ImagInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const imag = backend.data.get(input.dataId).complexTensorInfos.imag;\n  const imagVal = backend.data.get(imag.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the imag value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(imag.shape, imag.dtype, imagVal);\n}\n\nexport const imagConfig: KernelConfig = {\n  kernelName: Imag,\n  backendName: 'cpu',\n  kernelFunc: imag as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function reshape(\n    args:\n        {inputs: ReshapeInputs, backend: MathBackendCPU, attrs: ReshapeAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  backend.incRef(x.dataId);\n\n  const xData = backend.data.get(x.dataId);\n\n  if (xData.complexTensorInfos != null) {\n    const real = xData.complexTensorInfos.real;\n    const imag = xData.complexTensorInfos.imag;\n\n    real.shape = $shape;\n    imag.shape = $shape;\n  }\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'cpu',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Concat, ConcatAttrs, ConcatInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nimport {complex} from './Complex';\nimport {imag} from './Imag';\nimport {real} from './Real';\nimport {reshape} from './Reshape';\n\nexport function concat(\n    args: {inputs: ConcatInputs, backend: MathBackendCPU, attrs: ConcatAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {axis} = attrs;\n\n  const $axis = util.parseAxisParam(axis, inputs[0].shape)[0];\n  let outShape = backend_util.computeOutShape(inputs.map(t => t.shape), $axis);\n\n  if (util.sizeFromShape(outShape) === 0) {\n    return backend.makeTensorInfo(outShape, inputs[0].dtype, []);\n  }\n\n  // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n  const $inputs = inputs.filter(t => util.sizeFromShape(t.shape) > 0);\n  if ($inputs.length === 1) {\n    return $inputs[0];\n  }\n\n  const shapes = $inputs.map(t => t.shape);\n  backend_util.assertParamsConsistent(shapes, $axis);\n\n  if ($inputs[0].dtype === 'complex64') {\n    const reals = $inputs.map((t) => real({inputs: {input: t}, backend}));\n    const imags = $inputs.map((t) => imag({inputs: {input: t}, backend}));\n\n    const realConcated = concat({inputs: reals, backend, attrs: {axis}});\n    const imagConcated = concat({inputs: imags, backend, attrs: {axis}});\n\n    const result =\n        complex({inputs: {real: realConcated, imag: imagConcated}, backend});\n\n    reals.forEach(r => backend.disposeIntermediateTensorInfo(r));\n    imags.forEach(i => backend.disposeIntermediateTensorInfo(i));\n    backend.disposeIntermediateTensorInfo(realConcated);\n    backend.disposeIntermediateTensorInfo(imagConcated);\n\n    return result;\n  }\n\n  // Any concat of n-dimensional tensors across any axis can be reduced to\n  // a concatenation of two-dimensional tensors across the axis 1 by first\n  // partitioning the axes of the original tensors into those less than the\n  // axis to be concatenated and the rest. Then reshape the tensors\n  // into a two-dimensional tensor by collapsing these two sets of axes and\n  // concatenate the resulting matrices across the axis 1, finally reshaping\n  // the result to have the proper shape.\n  const inputs2D = $inputs.map(t => {\n    const innerSize = util.sizeFromShape(t.shape.slice($axis));\n    const shape = [-1, innerSize];\n    return reshape({inputs: {x: t}, backend, attrs: {shape}});\n  });\n\n  // Concats 2d tensors along axis=1.\n  outShape =\n      backend_util.computeOutShape(inputs2D.map(t => t.shape), 1 /* axis */);\n\n  const outVals = util.getTypedArrayFromDType(\n      $inputs[0].dtype as 'float32', util.sizeFromShape(outShape));\n\n  if (inputs2D[0].shape[0] === 1) {\n    // Use built-in TypedArray.set() method for speed.\n    let offset = 0;\n    inputs2D.forEach(t => {\n      const val = backend.data.get(t.dataId).values as TypedArray;\n      const size = util.sizeFromShape(t.shape);\n\n      outVals.set(val, offset);\n      offset += size;\n    });\n  } else {\n    let colOffset = 0;\n\n    inputs2D.forEach(t => {\n      const tVals = backend.data.get(t.dataId).values as TypedArray;\n\n      let tIdx = 0;\n\n      for (let row = 0; row < t.shape[0]; ++row) {\n        const resIdx = row * outShape[1] + colOffset;\n        for (let col = 0; col < t.shape[1]; ++col) {\n          outVals[resIdx + col] = tVals[tIdx++];\n        }\n      }\n\n      colOffset += t.shape[1];\n    });\n  }\n\n  const finalOutShape =\n      backend_util.computeOutShape($inputs.map(t => t.shape), $axis);\n\n  const outInfo =\n      backend.makeTensorInfo(finalOutShape, inputs[0].dtype, outVals);\n\n  inputs2D.forEach(t => backend.disposeIntermediateTensorInfo(t));\n\n  return outInfo;\n}\n\nexport const concatConfig: KernelConfig = {\n  kernelName: Concat,\n  backendName: 'cpu',\n  kernelFunc: concat as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const cosKernelFunc = unaryKernelFunc(Cos, (xi) => Math.cos(xi));\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'cpu',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const coshKernelFunc = unaryKernelFunc(Cosh, (xi) => Math.cosh(xi));\n\nexport const coshConfig: KernelConfig = {\n  kernelName: Cosh,\n  backendName: 'cpu',\n  kernelFunc: coshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2D, Dilation2DAttrs, Dilation2DInputs, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dConfig: KernelConfig = {\n  kernelName: Dilation2D,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter} = inputs as Dilation2DInputs;\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xRank = x.shape.length;\n\n    const filterVals = cpuBackend.data.get(filter.dataId).values as TypedArray;\n    const filterRank = filter.shape.length;\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    const outSize = util.sizeFromShape(outShape);\n    const outRank = outShape.length;\n    const outputVals = util.getArrayFromDType(x.dtype, outSize);\n\n    // Upsampling the input by fill in `dilation size - 1` values between each\n    // input value.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const xIndex = util.locToIndex(\n                        [b, hIn, wIn, d], xRank, util.computeStrides(x.shape));\n                    const filterIndex = util.locToIndex(\n                        [h, w, d], filterRank,\n                        util.computeStrides(filter.shape));\n                    const val = xVals[xIndex] + filterVals[filterIndex];\n                    if (val > curVal) {\n                      curVal = val;\n                    }\n                  }\n                }\n              }\n            }\n            const outputIndex = util.locToIndex(\n                [b, hOut, wOut, d], outRank, util.computeStrides(outShape));\n            outputVals[outputIndex] = curVal;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(outputVals, x.dtype), outShape, x.dtype);\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropFilter, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropFilterConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropFilter,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropFilter}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed filter gradients has the same dimensions as the filter:\n    // [filterHeight, filterWidth, depth]\n    const gradients = util.makeZerosNestedTypedArray(\n                          filter.shape, filter.dtype) as number[][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hMax = 0;\n            let wMax = 0;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hMax = h;\n                      wMax = w;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[hMax][wMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), filter.shape, filter.dtype);\n\n    return {dataId, shape: filter.shape, dtype: filter.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropInput, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropInputConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropInput,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropInput}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed gradients has the same dimensions as the input:\n    // [batch, inputHeight, inputCols, inChannel]\n    const gradients =\n        util.makeZerosNestedTypedArray(x.shape, x.dtype) as number[][][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hInMax = (hBeg < 0) ? 0 : hBeg;\n            let wInMax = (wBeg < 0) ? 0 : wBeg;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hInMax = hIn;\n                      wInMax = wIn;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[b][hInMax][wInMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), x.shape, x.dtype);\n\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const divImpl =\n    createSimpleBinaryKernelImpl((a: number, b: number) => a / b);\nexport const div = binaryKernelFunc(Div, divImpl);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'cpu',\n  kernelFunc: div\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Elu, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const eluKernelFunc =\n    unaryKernelFunc(Elu, (xi) => xi >= 0 ? xi : (Math.exp(xi) - 1));\n\nexport const eluConfig: KernelConfig = {\n  kernelName: Elu,\n  backendName: 'cpu',\n  kernelFunc: eluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Erf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst p = backend_util.ERF_P;\nconst a1 = backend_util.ERF_A1;\nconst a2 = backend_util.ERF_A2;\nconst a3 = backend_util.ERF_A3;\nconst a4 = backend_util.ERF_A4;\nconst a5 = backend_util.ERF_A5;\n\nexport const erfKernelFunc = unaryKernelFunc(\n    Erf,\n    (xi) => {\n      const sign = Math.sign(xi);\n      const v = Math.abs(xi);\n      const t = 1.0 / (1.0 + p * v);\n      return sign *\n          (1.0 -\n           (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t *\n               Math.exp(-v * v));\n    },\n);\n\nexport const erfConfig: KernelConfig = {\n  kernelName: Erf,\n  backendName: 'cpu',\n  kernelFunc: erfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Tensor, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {add} from '../kernels/Add';\nimport {complex} from '../kernels/Complex';\nimport {concat} from '../kernels/Concat';\nimport {divConfig} from '../kernels/Div';\nimport {identity} from '../kernels/Identity';\nimport {imag} from '../kernels/Imag';\nimport {multiply} from '../kernels/Multiply';\nimport {real} from '../kernels/Real';\nimport {slice} from '../kernels/Slice';\nimport {sub} from '../kernels/Sub';\n\n/**\n * Calculate FFT of inner most elements of batch tensor.\n */\nexport function fftBatch(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): TensorInfo {\n  const inputShape = input.shape;\n  const batch = inputShape[0];\n  const innerDim = inputShape[1];\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const real2D = inputVals.complexTensorInfos.real;\n  const imag2D = inputVals.complexTensorInfos.imag;\n\n  // Collects real and imaginary values separately.\n  const resultShape = [batch, innerDim];\n  const resultSize = util.sizeFromShape(resultShape);\n  const resultReal = util.getTypedArrayFromDType('float32', resultSize);\n  const resultImag = util.getTypedArrayFromDType('float32', resultSize);\n\n  for (let b = 0; b < batch; b++) {\n    // TODO: Support slice ops for complex type.\n    const r = slice({\n      inputs: {x: real2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n    const i = slice({\n      inputs: {x: imag2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n\n    const input = complex({inputs: {real: r, imag: i}, backend: cpuBackend});\n\n    // Run FFT by batch element.\n    const {real, imag} = fftImpl(input, inverse, cpuBackend);\n    const res = backend_util.mergeRealAndImagArrays(real, imag);\n\n    for (let d = 0; d < innerDim; d++) {\n      const c = backend_util.getComplexWithIndex(res, d);\n      resultReal[b * innerDim + d] = c.real;\n      resultImag[b * innerDim + d] = c.imag;\n    }\n\n    cpuBackend.disposeIntermediateTensorInfo(r);\n    cpuBackend.disposeIntermediateTensorInfo(i);\n    cpuBackend.disposeIntermediateTensorInfo(input);\n  }\n\n  const $realInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultReal);\n  const $imagInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultImag);\n\n  const result = complex(\n      {inputs: {real: $realInfo, imag: $imagInfo}, backend: cpuBackend});\n\n  cpuBackend.disposeIntermediateTensorInfo($realInfo);\n  cpuBackend.disposeIntermediateTensorInfo($imagInfo);\n\n  return result;\n}\n\nexport function fftImpl(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  const inputSize = util.sizeFromShape(input.shape);\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const realVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.real.dataId).values as\n      Float32Array;\n\n  const imagVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.imag.dataId).values as\n      Float32Array;\n\n  if (isExponentOf2(inputSize)) {\n    const result =\n        fftRadix2(realVals, imagVals, inputSize, inverse, cpuBackend);\n\n    const resultShape = [input.shape[0], input.shape[1]];\n\n    if (inverse) {\n      const realInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.real);\n      const imagInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.imag);\n\n      const sizeInfo: TensorInfo = cpuBackend.makeTensorInfo(\n          [], 'float32',\n          util.createScalarValue(inputSize as {} as 'float32', 'float32'));\n      const sizeInfoCopy =\n          identity({inputs: {x: sizeInfo}, backend: cpuBackend});\n\n      const divRealInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: realInfo, b: sizeInfo}, backend: cpuBackend}) as\n          TensorInfo;\n      const divImagInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: imagInfo, b: sizeInfoCopy}, backend: cpuBackend}) as\n          TensorInfo;\n\n      const divRealVals =\n          cpuBackend.data.get(divRealInfo.dataId).values as Float32Array;\n      const divImagVals =\n          cpuBackend.data.get(divImagInfo.dataId).values as Float32Array;\n\n      cpuBackend.disposeIntermediateTensorInfo(realInfo);\n      cpuBackend.disposeIntermediateTensorInfo(imagInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfoCopy);\n      cpuBackend.disposeIntermediateTensorInfo(divRealInfo);\n      cpuBackend.disposeIntermediateTensorInfo(divImagInfo);\n\n      return {real: divRealVals, imag: divImagVals};\n    }\n\n    return result;\n  } else {\n    const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n    const rawOutput =\n        fourierTransformByMatmul(data, inputSize, inverse) as Float32Array;\n\n    return backend_util.splitRealAndImagArrays(rawOutput);\n  }\n}\n\nfunction isExponentOf2(size: number): boolean {\n  return (size & size - 1) === 0;\n}\n\n// FFT using Cooley-Tukey algorithm on radix 2 dimensional input.\nfunction fftRadix2(\n    realVals: Float32Array, imagVals: Float32Array, size: number,\n    inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  if (size === 1) {\n    return {real: realVals, imag: imagVals};\n  }\n\n  const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n  const half = size / 2;\n\n  const evenComplex = backend_util.complexWithEvenIndex(data);\n\n  const evenRealVals = evenComplex.real;\n  const evenImagVals = evenComplex.imag;\n\n  const evenShape = [evenRealVals.length];\n\n  const evenRealInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenRealVals);\n  const evenImagInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenImagVals);\n\n  const evenTensorInfo = complex(\n      {inputs: {real: evenRealInfo, imag: evenImagInfo}, backend: cpuBackend});\n\n  const oddComplex = backend_util.complexWithOddIndex(data);\n\n  const oddRealVals = oddComplex.real;\n  const oddImagVals = oddComplex.imag;\n\n  const oddShape = [oddRealVals.length];\n\n  const oddRealInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddRealVals);\n  const oddImagInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddImagVals);\n\n  const oddTensorInfo = complex(\n      {inputs: {real: oddRealInfo, imag: oddImagInfo}, backend: cpuBackend});\n\n  // Recursive call for half part of original input.\n  const $evenComplex =\n      fftRadix2(evenRealVals, evenImagVals, half, inverse, cpuBackend);\n\n  const $evenRealVals = $evenComplex.real;\n  const $evenImagVals = $evenComplex.imag;\n\n  const $evenShape = [$evenRealVals.length];\n\n  const $evenRealInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenRealVals);\n  const $evenImagInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenImagVals);\n\n  const $evenTensorInfo = complex({\n    inputs: {real: $evenRealInfo, imag: $evenImagInfo},\n    backend: cpuBackend\n  });\n\n  const $oddComplex =\n      fftRadix2(oddRealVals, oddImagVals, half, inverse, cpuBackend);\n\n  const $oddRealVals = $oddComplex.real;\n  const $oddImagVals = $oddComplex.imag;\n\n  const $oddShape = [$oddRealVals.length];\n\n  const $oddRealInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddRealVals);\n  const $oddImagInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddImagVals);\n\n  const $oddTensorInfo = complex(\n      {inputs: {real: $oddRealInfo, imag: $oddImagInfo}, backend: cpuBackend});\n\n  const e = backend_util.exponents(size, inverse);\n  const eShape = [e.real.length];\n\n  const eRealInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.real);\n  const eImagInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.imag);\n\n  const complexInfo = complex(\n      {inputs: {real: eRealInfo, imag: eImagInfo}, backend: cpuBackend});\n\n  const exponentInfo =\n      multiply(\n          {inputs: {a: complexInfo, b: $oddTensorInfo}, backend: cpuBackend}) as\n      TensorInfo;\n\n  const addPart = add({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n  const subPart = sub({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n\n  const addPartReal = real({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartReal = real({inputs: {input: subPart}, backend: cpuBackend});\n\n  const addPartImag = imag({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartImag = imag({inputs: {input: subPart}, backend: cpuBackend});\n\n  const $real = concat({\n    inputs: [addPartReal as Tensor, subPartReal as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n  const $imag = concat({\n    inputs: [addPartImag as Tensor, subPartImag as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n\n  const $realVals = cpuBackend.data.get($real.dataId).values as Float32Array;\n  const $imagVals = cpuBackend.data.get($imag.dataId).values as Float32Array;\n\n  cpuBackend.disposeIntermediateTensorInfo(evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(complexInfo);\n  cpuBackend.disposeIntermediateTensorInfo(exponentInfo);\n  cpuBackend.disposeIntermediateTensorInfo(addPart);\n  cpuBackend.disposeIntermediateTensorInfo(subPart);\n  cpuBackend.disposeIntermediateTensorInfo(addPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(addPartImag);\n  cpuBackend.disposeIntermediateTensorInfo(subPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(subPartImag);\n  cpuBackend.disposeIntermediateTensorInfo($real);\n  cpuBackend.disposeIntermediateTensorInfo($imag);\n\n  return {real: $realVals, imag: $imagVals};\n}\n\n// Calculate fourier transform by multplying sinusoid matrix.\nfunction fourierTransformByMatmul(\n    data: TypedArray, size: number, inverse: boolean): TypedArray {\n  const ret = new Float32Array(size * 2);\n  // TODO: Use matmul instead once it supports complex64 type.\n  for (let r = 0; r < size; r++) {\n    let real = 0.0;\n    let imag = 0.0;\n    for (let c = 0; c < size; c++) {\n      const e = backend_util.exponent(r * c, size, inverse);\n      const term = backend_util.getComplexWithIndex(data as Float32Array, c);\n      real += term.real * e.real - term.imag * e.imag;\n      imag += term.real * e.imag + term.imag * e.real;\n    }\n    if (inverse) {\n      real /= size;\n      imag /= size;\n    }\n    backend_util.assignToTypedArray(ret, real, imag, r);\n  }\n  return ret;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FFT, FFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function fft(args: {inputs: FFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, false, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const fftConfig: KernelConfig = {\n  kernelName: FFT,\n  backendName: 'cpu',\n  kernelFunc: fft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n\n            const coordX = Math.round(imageWidth - x);\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n\n            let outputValue = imageVals[outIdx];\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth) {\n              // set the output to the image value at the coordinate position.\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n            output[outIdx] = outputValue;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IFFT, IFFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function ifft(args: {inputs: IFFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, true, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const ifftConfig: KernelConfig = {\n  kernelName: IFFT,\n  backendName: 'cpu',\n  kernelFunc: ifft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isFiniteKernelFunc =\n    unaryKernelFunc(IsFinite, (xi) => Number.isFinite(xi) ? 1 : 0, 'bool');\n\nexport const isFiniteConfig: KernelConfig = {\n  kernelName: IsFinite,\n  backendName: 'cpu',\n  kernelFunc: isFiniteKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isInfKernelFunc =\n    unaryKernelFunc(IsInf, (xi) => Math.abs(xi) === Infinity ? 1 : 0, 'bool');\n\nexport const isInfConfig: KernelConfig = {\n  kernelName: IsInf,\n  backendName: 'cpu',\n  kernelFunc: isInfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isNaNKernelFunc =\n    unaryKernelFunc(IsNan, (xi) => Number.isNaN(xi) ? 1 : 0, 'bool');\n\nexport const isNaNConfig: KernelConfig = {\n  kernelName: IsNan,\n  backendName: 'cpu',\n  kernelFunc: isNaNKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log1p} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const log1pKernelFunc = unaryKernelFunc(Log1p, (xi) => Math.log1p(xi));\n\nexport const log1pConfig: KernelConfig = {\n  kernelName: Log1p,\n  backendName: 'cpu',\n  kernelFunc: log1pKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, LogicalNot} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const logicalNotKernelFunc =\n    unaryKernelFunc(LogicalNot, (xi) => xi ? 0 : 1, 'bool');\n\nexport const logicalNotConfig: KernelConfig = {\n  kernelName: LogicalNot,\n  backendName: 'cpu',\n  kernelFunc: logicalNotKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig} from '@tensorflow/tfjs-core';\nimport {TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    let xShape = x.shape;\n    const xRank = xShape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, xShape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    let xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    if (permutedAxes != null) {\n      const newShape: number[] = new Array(xRank);\n      for (let i = 0; i < newShape.length; i++) {\n        newShape[i] = xShape[permutedAxes[i]];\n      }\n\n      xVals = transposeImpl(xVals, xShape, x.dtype, permutedAxes, newShape);\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n\n      xShape = newShape;\n    }\n\n    assertNotComplex(x, 'max');\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(xShape, axes);\n\n    const reduceSize = util.sizeFromShape(reduceShape);\n\n    const result = maxImpl(xVals, reduceSize, maxOutShape, x.dtype);\n    const dataId = cpuBackend.write(result, maxOutShape, x.dtype);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // reshape\n      const newShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n      outShape = newShape;\n    }\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function maxPool(\n    args:\n        {inputs: MaxPoolInputs, backend: MathBackendCPU, attrs: MaxPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'max');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'cpu',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, buffer, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, Rank, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {maxPoolPositions} from '../utils/pool_utils';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const xValues = backend.data.get(x.dataId).values as TypedArray;\n  const maxPosBuf = buffer(\n      convInfo.outShape, x.dtype,\n      maxPoolPositions(xValues, x.shape, x.dtype, convInfo).values);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n              const maxPos = effectiveFilterHeight * effectiveFilterWidth - 1 -\n                  (maxPosBuf.get(b, dyR, dyC, d) as number);\n              const curPos = wR * effectiveFilterWidth + wC;\n\n              const mask = maxPos === curPos ? 1 : 0;\n              if (mask === 0) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel * mask;\n            }\n          }\n          dx.set(dotProd, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, DataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {maxPoolPositions, pool} from '../utils/pool_utils';\nexport function maxPoolWithArgmaxImpl(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    includeBatchInIndex: boolean, convInfo: backend_util.Conv2DInfo) {\n  const strides = util.computeStrides(xShape);\n  const maxPools = pool(xValues, xShape, dtype, strides, convInfo, 'max');\n  const maxPositions = maxPoolPositions(\n      xValues, xShape, dtype, convInfo, true, includeBatchInIndex);\n\n  return [maxPools.values, maxPositions.values];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'MaxPoolWithArgmax');\n\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        [1, 1], pad);\n    const [pooled, indexes] = maxPoolWithArgmaxImpl(\n        values, x.shape, x.dtype, includeBatchInIndex, convInfo);\n\n    const pooledDataId =\n        cpuBackend.write(pooled as Float32Array, convInfo.outShape, x.dtype);\n    const indexesDataId =\n        cpuBackend.write(indexes as Int32Array, convInfo.outShape, x.dtype);\n    return [\n      {dataId: pooledDataId, shape: convInfo.outShape, dtype: x.dtype},\n      {dataId: indexesDataId, shape: convInfo.outShape, dtype: 'int32'}\n    ];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionPadded');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionWithScore');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NotEqual} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const notEqualImpl =\n    createSimpleBinaryKernelImpl(((a, b) => (a !== b) ? 1 : 0));\nexport const notEqual =\n    binaryKernelFunc(NotEqual, notEqualImpl, null /* complexOp */, 'bool');\n\nexport const notEqualConfig: KernelConfig = {\n  kernelName: NotEqual,\n  backendName: 'cpu',\n  kernelFunc: notEqual\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, NumericDataType, PadV2, PadV2Attrs, PadV2Inputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function padV2(\n    args: {inputs: PadV2Inputs, backend: MathBackendCPU, attrs: PadV2Attrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {paddings, constantValue} = attrs;\n\n  assertNotComplex(x, 'pad');\n\n  const outShape = paddings.map(\n      (p, i) => p[0] /* beforePad */ + x.shape[i] + p[1] /* afterPad */);\n\n  const start = paddings.map(p => p[0]);\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const xSize = util.sizeFromShape(x.shape);\n  const xRank = x.shape.length;\n  const xStrides = util.computeStrides(x.shape);\n\n  const resultSize = util.sizeFromShape(outShape);\n  const resultRank = outShape.length;\n  const resultStrides = util.computeStrides(outShape);\n  const resVals =\n      util.getTypedArrayFromDType(x.dtype as NumericDataType, resultSize);\n\n  if (constantValue !== 0) {\n    resVals.fill(constantValue);\n  }\n\n  for (let i = 0; i < xSize; i++) {\n    const coords = util.indexToLoc(i, xRank, xStrides);\n    const outCoords = coords.map((c, i) => c + start[i]);\n    const outIndex = util.locToIndex(outCoords, resultRank, resultStrides);\n\n    resVals[outIndex] = xVals[i];\n  }\n\n  const outId = backend.write(resVals, outShape, x.dtype);\n\n  return {dataId: outId, shape: outShape, dtype: x.dtype};\n}\n\nexport const padV2Config: KernelConfig = {\n  kernelName: PadV2,\n  backendName: 'cpu',\n  kernelFunc: padV2 as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Reciprocal} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const reciprocalKernelFunc = unaryKernelFunc(Reciprocal, (xi) => 1 / xi);\n\nexport const reciprocalConfig: KernelConfig = {\n  kernelName: Reciprocal,\n  backendName: 'cpu',\n  kernelFunc: reciprocalKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {backend_util, RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const fullOpacityValue = 255;\n\n    const sinFactor = Math.sin(radians);\n    const cosFactor = Math.cos(radians);\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n            const y = coords[1];\n\n            // coordX/coordY are the result of rotating and translating x/y.\n            let coordX = (x - centerX) * cosFactor - (y - centerY) * sinFactor;\n            let coordY = (x - centerX) * sinFactor + (y - centerY) * cosFactor;\n            coordX = Math.round(coordX + centerX);\n            coordY = Math.round(coordY + centerY);\n\n            let outputValue = fillValue;\n            if (typeof fillValue !== 'number') {\n              if (channel === 3) {\n                outputValue = fullOpacityValue;\n              } else {\n                outputValue = fillValue[channel];\n              }\n            }\n\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth && coordY >= 0 &&\n                coordY < imageHeight) {\n              // set the output to the image value at the coordinate position.\n              const rotatedRowOffset = coordY * (imageWidth * numChannels);\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rotatedRowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n            output[outIdx] = outputValue as number;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Round} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const roundKernelFunc = unaryKernelFunc(Round, (xi) => {\n  // The algorithm is based on banker's rounding.\n  const base = Math.floor(xi);\n  if (xi - base < 0.5) {\n    return Math.floor(xi);\n  } else if (xi - base > 0.5) {\n    return Math.ceil(xi);\n  } else {\n    if (base % 2.0 === 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n});\n\nexport const roundConfig: KernelConfig = {\n  kernelName: Round,\n  backendName: 'cpu',\n  kernelFunc: roundKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, Selu} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst scaleAlpha = backend_util.SELU_SCALEALPHA;\nconst scale = backend_util.SELU_SCALE;\n\nexport const seluKernelFunc = unaryKernelFunc(Selu, (xi) => {\n  if (xi >= 0) {\n    return scale * xi;\n  } else {\n    return scaleAlpha * (Math.exp(xi) - 1);\n  }\n});\n\nexport const seluConfig: KernelConfig = {\n  kernelName: Selu,\n  backendName: 'cpu',\n  kernelFunc: seluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sigmoid} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sigmoidKernelFunc =\n    unaryKernelFunc(Sigmoid, (xi) => 1 / (1 + Math.exp(-xi)));\n\nexport const sigmoidConfig: KernelConfig = {\n  kernelName: Sigmoid,\n  backendName: 'cpu',\n  kernelFunc: sigmoidKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sign} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const signKernelFunc = unaryKernelFunc(Sign, (xi) => {\n  if (xi < 0) {\n    return -1;\n  } else if (xi > 0) {\n    return 1;\n  } else {\n    return 0;\n  }\n});\n\nexport const signConfig: KernelConfig = {\n  kernelName: Sign,\n  backendName: 'cpu',\n  kernelFunc: signKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinKernelFunc = unaryKernelFunc(Sin, (xi) => Math.sin(xi));\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'cpu',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sinh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinhKernelFunc = unaryKernelFunc(Sinh, (xi) => Math.sinh(xi));\n\nexport const sinhConfig: KernelConfig = {\n  kernelName: Sinh,\n  backendName: 'cpu',\n  kernelFunc: sinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Softplus} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\n// mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n\n// epsilon is the difference between 1.0 and the next representable float.\n// For a single precision 32 bit float this should be 2^-23, see:\n// https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\nconst epsilon = 1.1920928955078125e-7;\nconst threshold = Math.log(epsilon) + 2.0;\n\nexport const softplusKernelFunc = unaryKernelFunc(Softplus, (xi) => {\n  // Value above which exp(x) may overflow, but softplus(x) == x\n  // is within machine epsilon.\n  const tooLarge = xi > -threshold;\n\n  // Value below which exp(x) may underflow, but softplus(x) == exp(x)\n  // is within machine epsilon.\n  const tooSmall = xi < threshold;\n\n  const expX = Math.exp(xi);\n  let result;\n\n  if (tooSmall) {\n    result = expX;\n  } else if (tooLarge) {\n    result = xi;\n  } else {\n    result = Math.log(1.0 + expX);\n  }\n  return result;\n});\n\nexport const softplusConfig: KernelConfig = {\n  kernelName: Softplus,\n  backendName: 'cpu',\n  kernelFunc: softplusKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {transposeImpl} from './Transpose_impl';\n\nexport function transpose(args: {\n  inputs: TransposeInputs,\n  attrs: TransposeAttrs,\n  backend: MathBackendCPU\n}): TensorInfo {\n  const {inputs, attrs, backend} = args;\n  const {x} = inputs;\n  const {perm} = attrs;\n\n  assertNotComplex(x, 'transpose');\n\n  const xRank = x.shape.length;\n\n  const newShape: number[] = new Array(xRank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = x.shape[perm[i]];\n  }\n\n  const values = backend.data.get(x.dataId).values as TypedArray;\n  const result = transposeImpl(values, x.shape, x.dtype, perm, newShape);\n\n  const dataId = backend.write(result, newShape, x.dtype);\n  return {dataId, shape: newShape, dtype: x.dtype};\n}\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'cpu',\n  kernelFunc: transpose as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, KernelFunc, ReshapeAttrs, ReshapeInputs, SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs, TensorInfo, TransposeAttrs, TransposeInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {padV2Config} from './PadV2';\nimport {reshape} from './Reshape';\nimport {transpose} from './Transpose';\n\nexport function spaceToBatchND(args: {\n  inputs: SpaceToBatchNDInputs,\n  backend: MathBackendCPU,\n  attrs: SpaceToBatchNDAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {blockShape, paddings} = attrs;\n\n  assertNotComplex([x], 'spaceToBatchND');\n\n  const prod = util.sizeFromShape(blockShape);\n\n  const completePaddings: Array<[number, number]> = [[0, 0]];\n  completePaddings.push(...(paddings as Array<[number, number]>));\n\n  for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n    completePaddings.push([0, 0]);\n  }\n\n  const paddedX = padV2Config.kernelFunc({\n    inputs: {x},\n    backend,\n    attrs: {paddings: completePaddings, constantValue: 0}\n  }) as TensorInfo;\n\n  const reshapedPaddedShape =\n      backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n  const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n      reshapedPaddedShape.length, blockShape.length, false);\n\n  const flattenShape =\n      backend_util.getReshapedPermuted(paddedX.shape, blockShape, prod, false);\n\n  const reshapeInputs: ReshapeInputs = {x: paddedX};\n  const reshapeAttrs: ReshapeAttrs = {shape: reshapedPaddedShape};\n  const paddedXReshaped =\n      reshape({inputs: reshapeInputs, backend, attrs: reshapeAttrs});\n\n  const transposeInputs: TransposeInputs = {x: paddedXReshaped};\n  const transposeAttrs:\n      TransposeAttrs = {perm: permutedReshapedPaddedPermutation};\n  const paddedXT =\n      transpose({inputs: transposeInputs, backend, attrs: transposeAttrs});\n\n  const resultReshapeInputs: ReshapeInputs = {x: paddedXT};\n  const resultReshapeAttrs: ReshapeAttrs = {shape: flattenShape};\n  const result = reshape(\n      {inputs: resultReshapeInputs, backend, attrs: resultReshapeAttrs});\n\n  backend.disposeIntermediateTensorInfo(paddedX);\n  backend.disposeIntermediateTensorInfo(paddedXReshaped);\n  backend.disposeIntermediateTensorInfo(paddedXT);\n\n  return result;\n}\n\nexport const spaceToBatchNDConfig: KernelConfig = {\n  kernelName: SpaceToBatchND,\n  backendName: 'cpu',\n  kernelFunc: spaceToBatchND as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sqrt} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sqrtKernelFunc = unaryKernelFunc(Sqrt, (xi) => Math.sqrt(xi));\n\nexport const sqrtConfig: KernelConfig = {\n  kernelName: Sqrt,\n  backendName: 'cpu',\n  kernelFunc: sqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square, SquareInputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend}) => {\n    const {x} = inputs as SquareInputs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'square');\n\n    const values = cpuBackend.data.get(x.dataId).values as Float32Array;\n    const newValues = new Float32Array(values.length);\n    for (let i = 0; i < values.length; ++i) {\n      const value = values[i];\n      newValues[i] = value * value;\n    }\n    const dataId = cpuBackend.write(newValues, x.shape, x.dtype);\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const squaredDifferenceImpl = createSimpleBinaryKernelImpl(((a, b) => {\n  const diff = a - b;\n  return diff * diff;\n}));\nexport const squaredDifference =\n    binaryKernelFunc(SquaredDifference, squaredDifferenceImpl);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'cpu',\n  kernelFunc: squaredDifference\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Step, StepAttrs} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const stepKernelFunc = unaryKernelFunc(Step, (xi, attrs) => {\n  const stepAttrs = attrs as {} as StepAttrs;\n  if (isNaN(xi)) {\n    return NaN;\n  } else {\n    return xi > 0 ? 1 : stepAttrs.alpha;\n  }\n});\n\nexport const stepConfig: KernelConfig = {\n  kernelName: Step,\n  backendName: 'cpu',\n  kernelFunc: stepKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanKernelFunc = unaryKernelFunc(Tan, (xi) => Math.tan(xi));\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'cpu',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tanh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanhKernelFunc = unaryKernelFunc(Tanh, (xi) => Math.tanh(xi));\n\nexport const tanhConfig: KernelConfig = {\n  kernelName: Tanh,\n  backendName: 'cpu',\n  kernelFunc: tanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {uniqueImpl} from './Unique_impl';\n\nexport function unique(\n    args: {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendCPU}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  const values = backend.data.get(x.dataId).values;\n  const {outputValues, outputShape, indices} =\n      uniqueImpl(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'cpu',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// We explicitly import the modular kernels so they get registered in the\n// global registry when we compile the library. A modular build would replace\n// the contents of this file and import only the kernels that are needed.\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {absConfig} from './kernels/Abs';\nimport {acosConfig} from './kernels/Acos';\nimport {acoshConfig} from './kernels/Acosh';\nimport {addConfig} from './kernels/Add';\nimport {asinConfig} from './kernels/Asin';\nimport {asinhConfig} from './kernels/Asinh';\nimport {atanConfig} from './kernels/Atan';\nimport {atanhConfig} from './kernels/Atanh';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {castConfig} from './kernels/Cast';\nimport {ceilConfig} from './kernels/Ceil';\nimport {clipConfig} from './kernels/Clip';\nimport {complexConfig} from './kernels/Complex';\nimport {concatConfig} from './kernels/Concat';\nimport {cosConfig} from './kernels/Cos';\nimport {coshConfig} from './kernels/Cosh';\nimport {dilation2dConfig} from './kernels/Dilation2D';\nimport {dilation2dBackpropFilterConfig} from './kernels/Dilation2DBackpropFilter';\nimport {dilation2dBackpropInputConfig} from './kernels/Dilation2DBackpropInput';\nimport {divConfig} from './kernels/Div';\nimport {eluConfig} from './kernels/Elu';\nimport {erfConfig} from './kernels/Erf';\nimport {expConfig} from './kernels/Exp';\nimport {expm1Config} from './kernels/Expm1';\nimport {fftConfig} from './kernels/FFT';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {floorConfig} from './kernels/Floor';\nimport {identityConfig} from './kernels/Identity';\nimport {ifftConfig} from './kernels/IFFT';\nimport {imagConfig} from './kernels/Imag';\nimport {isFiniteConfig} from './kernels/IsFinite';\nimport {isInfConfig} from './kernels/IsInf';\nimport {isNaNConfig} from './kernels/IsNaN';\nimport {logConfig} from './kernels/Log';\nimport {log1pConfig} from './kernels/Log1p';\nimport {logicalNotConfig} from './kernels/LogicalNot';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {multiplyConfig} from './kernels/Multiply';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {notEqualConfig} from './kernels/NotEqual';\nimport {padV2Config} from './kernels/PadV2';\nimport {realConfig} from './kernels/Real';\nimport {reciprocalConfig} from './kernels/Reciprocal';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {roundConfig} from './kernels/Round';\nimport {rsqrtConfig} from './kernels/Rsqrt';\nimport {seluConfig} from './kernels/Selu';\nimport {sigmoidConfig} from './kernels/Sigmoid';\nimport {signConfig} from './kernels/Sign';\nimport {sinConfig} from './kernels/Sin';\nimport {sinhConfig} from './kernels/Sinh';\nimport {sliceConfig} from './kernels/Slice';\nimport {softplusConfig} from './kernels/Softplus';\nimport {spaceToBatchNDConfig} from './kernels/SpaceToBatchND';\nimport {sqrtConfig} from './kernels/Sqrt';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {stepConfig} from './kernels/Step';\nimport {subConfig} from './kernels/Sub';\nimport {tanConfig} from './kernels/Tan';\nimport {tanhConfig} from './kernels/Tanh';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  absConfig,\n  acosConfig,\n  acoshConfig,\n  addConfig,\n  asinConfig,\n  asinhConfig,\n  atanConfig,\n  atanhConfig,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  castConfig,\n  ceilConfig,\n  clipConfig,\n  complexConfig,\n  concatConfig,\n  cosConfig,\n  coshConfig,\n  dilation2dConfig,\n  dilation2dBackpropInputConfig,\n  dilation2dBackpropFilterConfig,\n  divConfig,\n  eluConfig,\n  erfConfig,\n  expConfig,\n  expm1Config,\n  fftConfig,\n  flipLeftRightConfig,\n  floorConfig,\n  identityConfig,\n  ifftConfig,\n  imagConfig,\n  isFiniteConfig,\n  isInfConfig,\n  isNaNConfig,\n  logConfig,\n  log1pConfig,\n  logicalNotConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  maxConfig,\n  multiplyConfig,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  notEqualConfig,\n  padV2Config,\n  realConfig,\n  reciprocalConfig,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  roundConfig,\n  rsqrtConfig,\n  seluConfig,\n  sigmoidConfig,\n  signConfig,\n  sinConfig,\n  sinhConfig,\n  sliceConfig,\n  softplusConfig,\n  spaceToBatchNDConfig,\n  sqrtConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  stepConfig,\n  subConfig,\n  tanConfig,\n  tanhConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst contexts: {[key: string]: WebGLRenderingContext} = {};\n\nconst WEBGL_ATTRIBUTES: WebGLContextAttributes = {\n  alpha: false,\n  antialias: false,\n  premultipliedAlpha: false,\n  preserveDrawingBuffer: false,\n  depth: false,\n  stencil: false,\n  failIfMajorPerformanceCaveat: true\n};\n\nexport function setWebGLContext(\n    webGLVersion: number, gl: WebGLRenderingContext) {\n  contexts[webGLVersion] = gl;\n}\n\nexport function getWebGLContext(webGLVersion: number): WebGLRenderingContext {\n  if (!(webGLVersion in contexts)) {\n    const newCtx = getWebGLRenderingContext(webGLVersion);\n    if (newCtx !== null) {\n      contexts[webGLVersion] = newCtx;\n    } else {\n      console.log('Could not get context for WebGL version', webGLVersion);\n      return null;\n    }\n  }\n  const gl = contexts[webGLVersion];\n  if (gl.isContextLost()) {\n    delete contexts[webGLVersion];\n    return getWebGLContext(webGLVersion);\n  }\n\n  gl.disable(gl.DEPTH_TEST);\n  gl.disable(gl.STENCIL_TEST);\n  gl.disable(gl.BLEND);\n  gl.disable(gl.DITHER);\n  gl.disable(gl.POLYGON_OFFSET_FILL);\n  gl.disable(gl.SAMPLE_COVERAGE);\n  gl.enable(gl.SCISSOR_TEST);\n  gl.enable(gl.CULL_FACE);\n  gl.cullFace(gl.BACK);\n\n  return contexts[webGLVersion];\n}\n\nfunction createCanvas(webGLVersion: number) {\n  if (typeof OffscreenCanvas !== 'undefined' && webGLVersion === 2) {\n    return new OffscreenCanvas(300, 150);\n  } else if (typeof document !== 'undefined') {\n    return document.createElement('canvas');\n  } else {\n    throw new Error('Cannot create a canvas in this context');\n  }\n}\n\nfunction getWebGLRenderingContext(webGLVersion: number): WebGLRenderingContext {\n  if (webGLVersion !== 1 && webGLVersion !== 2) {\n    throw new Error('Cannot get WebGL rendering context, WebGL is disabled.');\n  }\n  const canvas = createCanvas(webGLVersion);\n\n  canvas.addEventListener('webglcontextlost', (ev: Event) => {\n    ev.preventDefault();\n    delete contexts[webGLVersion];\n  }, false);\n  if (webGLVersion === 1) {\n    return (canvas.getContext('webgl', WEBGL_ATTRIBUTES) ||\n            canvas.getContext('experimental-webgl', WEBGL_ATTRIBUTES)) as\n        WebGLRenderingContext;\n  }\n  return canvas.getContext('webgl2', WEBGL_ATTRIBUTES) as WebGLRenderingContext;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {backend_util, DataId, DataType, Tensor, util} from '@tensorflow/tfjs-core';\n\nexport enum PackingScheme {\n  /**\n   * All values in a single texel are densely packed without any constraints.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 4]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   010|011   020|021\n   * -------   -------   -------\n   * 002|003   012|013   022|023\n   *\n   * 100|101   110|111   120|121\n   * -------   -------   -------\n   * 102|103   112|113   122|123\n   *\n   */\n  DENSE,\n\n  /**\n   * Single texels contain only values from the same batch, and from adjacent\n   * rows and columns.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 5]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   002|003   004|xxx   020|021   022|023   024|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   * 100|101   102|103   104|xxx   120|121   122|123   124|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   */\n  SHARED_BATCH\n}\n\nexport enum TextureUsage {\n  RENDER,\n  UPLOAD,\n  PIXELS,\n  DOWNLOAD\n}\n\nexport enum PhysicalTextureType {\n  UNPACKED_FLOAT16,\n  UNPACKED_FLOAT32,\n  PACKED_4X1_UNSIGNED_BYTE,\n  PACKED_2X2_FLOAT32,\n  PACKED_2X2_FLOAT16\n}\n\nexport interface TextureData {\n  // Required.\n  shape: number[];\n  dtype: DataType;\n\n  // Optional.\n  values?: backend_util.BackendValues;\n  texture?: WebGLTexture;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensors field. When this is defined, texture will be null.\n  complexTensors?: {real: Tensor, imag: Tensor};\n  /** [rows, columns] shape of the texture. */\n  texShape?: [number, number];\n  usage?: TextureUsage;\n  isPacked?: boolean;\n\n  refCount: number;\n\n  // Available when the tensor has been sliced.\n  slice?: {\n    // Offset in the 'flat index' space.\n    flatOffset: number;\n    // Used for counting how many sliced tensors point to the same texture.\n    origDataId: DataId;\n  };\n}\n\nexport function getUnpackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns, rows];\n}\n\nexport function getUnpackedArraySizeFromMatrixSize(\n    matrixSize: number, channelsPerTexture: number): number {\n  return matrixSize * channelsPerTexture;\n}\n\nexport function getColorMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns * 4, rows];\n}\n\n/**\n * Get shape for densely packed RGBA texture.\n */\nexport function getDenseTexShape(shape: number[]): [number, number] {\n  const size = util.sizeFromShape(shape);\n  const texelsNeeded = Math.ceil(size / 4);\n  return util.sizeToSquarishShape(texelsNeeded);\n}\n\nexport function getMatrixSizeFromUnpackedArraySize(\n    unpackedSize: number, channelsPerTexture: number): number {\n  if (unpackedSize % channelsPerTexture !== 0) {\n    throw new Error(\n        `unpackedSize (${unpackedSize}) must be a multiple of ` +\n        `${channelsPerTexture}`);\n  }\n  return unpackedSize / channelsPerTexture;\n}\n\nexport function decodeMatrixFromUnpackedColorRGBAArray(\n    unpackedArray: Float32Array, matrix: Float32Array, channels: number) {\n  const requiredSize = unpackedArray.length * channels / 4;\n  if (matrix.length < requiredSize) {\n    throw new Error(\n        `matrix length (${matrix.length}) must be >= ${requiredSize}`);\n  }\n  let dst = 0;\n  for (let src = 0; src < unpackedArray.length; src += 4) {\n    for (let c = 0; c < channels; c++) {\n      matrix[dst++] = unpackedArray[src + c];\n    }\n  }\n}\n\nexport function getPackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [\n    Math.max(1, Math.ceil(columns / 2)), Math.max(1, Math.ceil(rows / 2))\n  ];\n}\n\nexport function getPackedRGBAArraySizeFromMatrixShape(\n    rows: number, columns: number): number {\n  const [w, h] = getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return w * h * 4;\n}\n\nexport interface TextureConfig {\n  internalFormatFloat: number;\n  textureFormatFloat: number;\n  internalFormatPackedHalfFloat: number;\n  internalFormatHalfFloat: number;\n  internalFormatPackedFloat: number;\n\n  // The format to use during a gl.readPixels call.\n  downloadTextureFormat: number;\n  // How many channels need to be unpacked after a gl.readPixels call.\n  downloadUnpackNumChannels: number;\n\n  defaultNumChannels: number;\n  textureTypeHalfFloat: number;\n  textureTypeFloat: number;\n}\n\nexport function getTextureConfig(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension?: any): TextureConfig {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n\n  let internalFormatFloat: number;\n  let internalFormatHalfFloat: number;\n  let internalFormatPackedHalfFloat: number;\n  let internalFormatPackedFloat: number;\n  let textureFormatFloat: number;\n\n  let downloadTextureFormat: number;\n  let downloadUnpackNumChannels: number;\n\n  let defaultNumChannels: number;\n  let textureTypeHalfFloat: number;\n  let textureTypeFloat: number;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    internalFormatFloat = glany.R32F;\n    internalFormatHalfFloat = glany.R16F;\n    internalFormatPackedHalfFloat = glany.RGBA16F;\n    internalFormatPackedFloat = glany.RGBA32F;\n    textureFormatFloat = glany.RED;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 1;\n    textureTypeHalfFloat = glany.HALF_FLOAT;\n    textureTypeFloat = glany.FLOAT;\n  } else {\n    internalFormatFloat = gl.RGBA;\n    internalFormatHalfFloat = gl.RGBA;\n    internalFormatPackedHalfFloat = gl.RGBA;\n    internalFormatPackedFloat = glany.RGBA;\n    textureFormatFloat = gl.RGBA;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 4;\n    textureTypeHalfFloat = textureHalfFloatExtension != null ?\n        textureHalfFloatExtension.HALF_FLOAT_OES :\n        null;\n    textureTypeFloat = gl.FLOAT;\n  }\n  downloadTextureFormat = gl.RGBA;\n\n  return {\n    internalFormatFloat,\n    internalFormatHalfFloat,\n    internalFormatPackedHalfFloat,\n    internalFormatPackedFloat,\n    textureFormatFloat,\n    downloadTextureFormat,\n    downloadUnpackNumChannels,\n    defaultNumChannels,\n    textureTypeHalfFloat,\n    textureTypeFloat\n  };\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext} from './canvas_util';\nimport {getTextureConfig} from './tex_util';\n\nexport function callAndCheck<T>(gl: WebGLRenderingContext, func: () => T): T {\n  const returnValue = func();\n  if (env().getBool('DEBUG')) {\n    checkWebGLError(gl);\n  }\n  return returnValue;\n}\n\nfunction checkWebGLError(gl: WebGLRenderingContext) {\n  const error = gl.getError();\n  if (error !== gl.NO_ERROR) {\n    throw new Error('WebGL Error: ' + getWebGLErrorMessage(gl, error));\n  }\n}\n\n// https://en.wikipedia.org/wiki/Half-precision_floating-point_format\nconst MIN_FLOAT16 = 5.96e-8;\nconst MAX_FLOAT16 = 65504;\n\nexport function canBeRepresented(num: number): boolean {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED') || num === 0 ||\n      (MIN_FLOAT16 < Math.abs(num) && Math.abs(num) < MAX_FLOAT16)) {\n    return true;\n  }\n  return false;\n}\n\nexport function getWebGLErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.NO_ERROR:\n      return 'NO_ERROR';\n    case gl.INVALID_ENUM:\n      return 'INVALID_ENUM';\n    case gl.INVALID_VALUE:\n      return 'INVALID_VALUE';\n    case gl.INVALID_OPERATION:\n      return 'INVALID_OPERATION';\n    case gl.INVALID_FRAMEBUFFER_OPERATION:\n      return 'INVALID_FRAMEBUFFER_OPERATION';\n    case gl.OUT_OF_MEMORY:\n      return 'OUT_OF_MEMORY';\n    case gl.CONTEXT_LOST_WEBGL:\n      return 'CONTEXT_LOST_WEBGL';\n    default:\n      return `Unknown error code ${status}`;\n  }\n}\n\nexport function getExtensionOrThrow(\n    gl: WebGLRenderingContext, extensionName: string): {} {\n  return throwIfNull<{}>(\n      gl, () => gl.getExtension(extensionName),\n      'Extension \"' + extensionName + '\" not supported on this browser.');\n}\n\nexport function createVertexShader(\n    gl: WebGLRenderingContext, vertexShaderSource: string): WebGLShader {\n  const vertexShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.VERTEX_SHADER),\n      'Unable to create vertex WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(vertexShader, vertexShaderSource));\n  callAndCheck(gl, () => gl.compileShader(vertexShader));\n  if (gl.getShaderParameter(vertexShader, gl.COMPILE_STATUS) === false) {\n    console.log(gl.getShaderInfoLog(vertexShader));\n    throw new Error('Failed to compile vertex shader.');\n  }\n  return vertexShader;\n}\n\nexport function createFragmentShader(\n    gl: WebGLRenderingContext, fragmentShaderSource: string): WebGLShader {\n  const fragmentShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.FRAGMENT_SHADER),\n      'Unable to create fragment WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(fragmentShader, fragmentShaderSource));\n  callAndCheck(gl, () => gl.compileShader(fragmentShader));\n  if (gl.getShaderParameter(fragmentShader, gl.COMPILE_STATUS) === false) {\n    logShaderSourceAndInfoLog(\n        fragmentShaderSource, gl.getShaderInfoLog(fragmentShader));\n    throw new Error('Failed to compile fragment shader.');\n  }\n  return fragmentShader;\n}\n\nconst lineNumberRegex = /ERROR: [0-9]+:([0-9]+):/g;\nfunction logShaderSourceAndInfoLog(\n    shaderSource: string, shaderInfoLog: string) {\n  const lineNumberRegexResult = lineNumberRegex.exec(shaderInfoLog);\n  if (lineNumberRegexResult == null) {\n    console.log(`Couldn't parse line number in error: ${shaderInfoLog}`);\n    console.log(shaderSource);\n    return;\n  }\n\n  const lineNumber = +lineNumberRegexResult[1];\n\n  const shaderLines = shaderSource.split('\\n');\n  const pad = shaderLines.length.toString().length + 2;\n  const linesWithLineNumbers = shaderLines.map(\n      (line, lineNumber) =>\n          util.rightPad((lineNumber + 1).toString(), pad) + line);\n  let maxLineLength = 0;\n  for (let i = 0; i < linesWithLineNumbers.length; i++) {\n    maxLineLength = Math.max(linesWithLineNumbers[i].length, maxLineLength);\n  }\n\n  const beforeErrorLines = linesWithLineNumbers.slice(0, lineNumber - 1);\n  const errorLine = linesWithLineNumbers.slice(lineNumber - 1, lineNumber);\n  const afterErrorLines = linesWithLineNumbers.slice(lineNumber);\n\n  console.log(beforeErrorLines.join('\\n'));\n  console.log(shaderInfoLog.split('\\n')[0]);\n  console.log(\n      `%c ${util.rightPad(errorLine[0], maxLineLength)}`,\n      'border:1px solid red; background-color:#e3d2d2; color:#a61717');\n  console.log(afterErrorLines.join('\\n'));\n}\n\nexport function createProgram(gl: WebGLRenderingContext): WebGLProgram {\n  return throwIfNull<WebGLProgram>(\n      gl, () => gl.createProgram(), 'Unable to create WebGLProgram.');\n}\n\nexport function linkProgram(gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.linkProgram(program));\n  if (gl.getProgramParameter(program, gl.LINK_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Failed to link vertex and fragment shaders.');\n  }\n}\n\nexport function validateProgram(\n    gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.validateProgram(program));\n  if (gl.getProgramParameter(program, gl.VALIDATE_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Shader program validation failed.');\n  }\n}\n\nexport function createStaticVertexBuffer(\n    gl: WebGLRenderingContext, data: Float32Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(gl, () => gl.bufferData(gl.ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function createStaticIndexBuffer(\n    gl: WebGLRenderingContext, data: Uint16Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl, () => gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function getNumChannels(): number {\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    return 1;\n  }\n  return 4;\n}\n\nexport function createTexture(gl: WebGLRenderingContext): WebGLTexture {\n  return throwIfNull<WebGLTexture>(\n      gl, () => gl.createTexture(), 'Unable to create WebGLTexture.');\n}\n\nexport function validateTextureSize(width: number, height: number) {\n  const maxTextureSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if ((width <= 0) || (height <= 0)) {\n    const requested = `[${width}x${height}]`;\n    throw new Error('Requested texture size ' + requested + ' is invalid.');\n  }\n  if ((width > maxTextureSize) || (height > maxTextureSize)) {\n    const requested = `[${width}x${height}]`;\n    const max = `[${maxTextureSize}x${maxTextureSize}]`;\n    throw new Error(\n        'Requested texture size ' + requested +\n        ' greater than WebGL maximum on this browser / GPU ' + max + '.');\n  }\n}\n\nexport function createFramebuffer(gl: WebGLRenderingContext): WebGLFramebuffer {\n  return throwIfNull<WebGLFramebuffer>(\n      gl, () => gl.createFramebuffer(), 'Unable to create WebGLFramebuffer.');\n}\n\nexport function bindVertexBufferToProgramAttribute(\n    gl: WebGLRenderingContext, program: WebGLProgram, attribute: string,\n    buffer: WebGLBuffer, arrayEntriesPerItem: number, itemStrideInBytes: number,\n    itemOffsetInBytes: number): boolean {\n  const loc = gl.getAttribLocation(program, attribute);\n  if (loc === -1) {\n    // The GPU compiler decided to strip out this attribute because it's unused,\n    // thus no need to bind.\n    return false;\n  }\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl,\n      () => gl.vertexAttribPointer(\n          loc, arrayEntriesPerItem, gl.FLOAT, false, itemStrideInBytes,\n          itemOffsetInBytes));\n  callAndCheck(gl, () => gl.enableVertexAttribArray(loc));\n  return true;\n}\n\nexport function bindTextureUnit(\n    gl: WebGLRenderingContext, texture: WebGLTexture, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n}\n\nexport function unbindTextureUnit(\n    gl: WebGLRenderingContext, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function getProgramUniformLocationOrThrow(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return throwIfNull<WebGLUniformLocation>(\n      gl, () => gl.getUniformLocation(program, uniformName),\n      'uniform \"' + uniformName + '\" not present in program.');\n}\n\nexport function getProgramUniformLocation(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return gl.getUniformLocation(program, uniformName);\n}\n\nexport function bindTextureToProgramUniformSampler(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    uniformSamplerLocation: WebGLUniformLocation, textureUnit: number) {\n  callAndCheck(gl, () => bindTextureUnit(gl, texture, textureUnit));\n  callAndCheck(gl, () => gl.uniform1i(uniformSamplerLocation, textureUnit));\n}\n\nexport function bindCanvasToFramebuffer(gl: WebGLRenderingContext) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n  callAndCheck(gl, () => gl.viewport(0, 0, gl.canvas.width, gl.canvas.height));\n  callAndCheck(gl, () => gl.scissor(0, 0, gl.canvas.width, gl.canvas.height));\n}\n\nexport function bindColorTextureToFramebuffer(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0));\n}\n\nexport function unbindColorTextureFromFramebuffer(\n    gl: WebGLRenderingContext, framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, null, 0));\n}\n\nexport function validateFramebuffer(gl: WebGLRenderingContext) {\n  const status = gl.checkFramebufferStatus(gl.FRAMEBUFFER);\n  if (status !== gl.FRAMEBUFFER_COMPLETE) {\n    throw new Error(\n        'Error binding framebuffer: ' + getFramebufferErrorMessage(gl, status));\n  }\n}\n\nexport function getFramebufferErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:\n      return 'FRAMEBUFFER_INCOMPLETE_DIMENSIONS';\n    case gl.FRAMEBUFFER_UNSUPPORTED:\n      return 'FRAMEBUFFER_UNSUPPORTED';\n    default:\n      return `unknown error ${status}`;\n  }\n}\n\nfunction throwIfNull<T>(\n    gl: WebGLRenderingContext, returnTOrNull: () => T | null,\n    failureMessage: string): T {\n  const tOrNull: T|null = callAndCheck(gl, () => returnTOrNull());\n  if (tOrNull == null) {\n    throw new Error(failureMessage);\n  }\n  return tOrNull;\n}\n\nfunction validateTextureUnit(gl: WebGLRenderingContext, textureUnit: number) {\n  const maxTextureUnit = gl.MAX_COMBINED_TEXTURE_IMAGE_UNITS - 1;\n  const glTextureUnit = textureUnit + gl.TEXTURE0;\n  if (glTextureUnit < gl.TEXTURE0 || glTextureUnit > maxTextureUnit) {\n    const textureUnitRange = `[gl.TEXTURE0, gl.TEXTURE${maxTextureUnit}]`;\n    throw new Error(`textureUnit must be in ${textureUnitRange}.`);\n  }\n}\n\nexport function getBatchDim(shape: number[], dimsToSkip = 2): number {\n  return util.sizeFromShape(shape.slice(0, shape.length - dimsToSkip));\n}\n\nexport function getRowsCols(shape: number[]): [number, number] {\n  if (shape.length === 0) {\n    throw Error('Cannot get rows and columns of an empty shape array.');\n  }\n\n  return [\n    shape.length > 1 ? shape[shape.length - 2] : 1, shape[shape.length - 1]\n  ];\n}\n\nexport function getShapeAs3D(shape: number[]): [number, number, number] {\n  let shapeAs3D: [number, number, number] = [1, 1, 1];\n  const isScalar = shape.length === 0 || (shape.length === 1 && shape[0] === 1);\n  if (!isScalar) {\n    shapeAs3D =\n        [getBatchDim(shape), ...getRowsCols(shape)] as [number, number, number];\n  }\n  return shapeAs3D;\n}\n\nexport function getTextureShapeFromLogicalShape(\n    logShape: number[], isPacked = false): [number, number] {\n  let maxTexSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if (isPacked) {\n    maxTexSize = maxTexSize * 2;\n\n    // This logic ensures we accurately count the number of packed texels needed\n    // to accommodate the tensor. We can only pack values in the same texel if\n    // they are from adjacent pairs of rows/cols within the same batch. So if a\n    // tensor has 3 rows, we pretend it has 4 rows in order to account for the\n    // fact that the texels containing the third row are half empty.\n    logShape = logShape.map(\n        (d, i) => i >= logShape.length - 2 ?\n            util.nearestLargerEven(logShape[i]) :\n            logShape[i]);\n\n    // Packed texture height is at least 2 (the channel height of a single\n    // texel).\n    if (logShape.length === 1) {\n      logShape = [2, logShape[0]];\n    }\n  }\n\n  // If logical shape is 2, we don't squeeze, since we want to match physical.\n  if (logShape.length !== 2) {\n    const squeezeResult = util.squeezeShape(logShape);\n    logShape = squeezeResult.newShape;\n  }\n\n  let size = util.sizeFromShape(logShape);\n  if (logShape.length <= 1 && size <= maxTexSize) {\n    return [1, size];\n  } else if (\n      logShape.length === 2 && logShape[0] <= maxTexSize &&\n      logShape[1] <= maxTexSize) {\n    return logShape as [number, number];\n  } else if (\n      logShape.length === 3 && logShape[0] * logShape[1] <= maxTexSize &&\n      logShape[2] <= maxTexSize) {\n    return [logShape[0] * logShape[1], logShape[2]];\n  } else if (\n      logShape.length === 3 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2]];\n  } else if (\n      logShape.length === 4 &&\n      logShape[0] * logShape[1] * logShape[2] <= maxTexSize &&\n      logShape[3] <= maxTexSize) {\n    return [logShape[0] * logShape[1] * logShape[2], logShape[3]];\n  } else if (\n      logShape.length === 4 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] * logShape[3] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2] * logShape[3]];\n  } else {\n    if (isPacked) {\n      // For packed textures size equals the number of channels required to\n      // accommodate the texture data. However in order to squarify such that\n      // inner dimensions stay even, we rewrite size to equal the number of\n      // texels. Then in the return statement we rehydrate the squarified\n      // dimensions to channel units.\n\n      const batchDim = getBatchDim(logShape);\n      let rows = 2, cols = 2;\n      if (logShape.length) {\n        [rows, cols] = getRowsCols(logShape);\n      }\n      size = batchDim * (rows / 2) * (cols / 2);\n      return util.sizeToSquarishShape(size).map(d => d * 2) as [number, number];\n    }\n    return util.sizeToSquarishShape(size);\n  }\n}\n\nfunction isEven(n: number): boolean {\n  return n % 2 === 0;\n}\n\n/**\n * This determines whether reshaping a packed texture requires rearranging\n * the data within the texture, assuming 2x2 packing.\n */\nexport function isReshapeFree(shape1: number[], shape2: number[]): boolean {\n  shape1 = shape1.slice(-2);\n  shape2 = shape2.slice(-2);\n\n  if (util.arraysEqual(shape1, shape2)) {\n    return true;\n  }\n\n  if (!shape1.length || !shape2.length) {  // One of the shapes is a scalar.\n    return true;\n  }\n\n  if (shape1[0] === 0 || shape1[1] === 0 || shape2[0] === 0 ||\n      shape2[1] === 0) {\n    return true;\n  }\n\n  if (shape1.length !== shape2.length) {  // One of the shapes is a vector.\n    const shape1Cols = shape1.slice(-1)[0];\n    const shape2Cols = shape2.slice(-1)[0];\n    if (shape1Cols === shape2Cols) {\n      return true;\n    }\n\n    if (isEven(shape1Cols) && isEven(shape2Cols) &&\n        (shape1[0] === 1 || shape2[0] === 1)) {\n      return true;\n    }\n  }\n  return shape1[1] === shape2[1] && isEven(shape1[0]) && isEven(shape2[0]);\n}\n\n// We cache webgl params because the environment gets reset between\n// unit tests and we don't want to constantly query the WebGLContext for\n// MAX_TEXTURE_SIZE.\nlet MAX_TEXTURE_SIZE: number;\nlet MAX_TEXTURES_IN_SHADER: number;\n\nexport function getWebGLMaxTextureSize(webGLVersion: number): number {\n  if (MAX_TEXTURE_SIZE == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURE_SIZE = gl.getParameter(gl.MAX_TEXTURE_SIZE);\n  }\n  return MAX_TEXTURE_SIZE;\n}\n\nexport function resetMaxTextureSize() {\n  MAX_TEXTURE_SIZE = null;\n}\nexport function resetMaxTexturesInShader() {\n  MAX_TEXTURES_IN_SHADER = null;\n}\n\nexport function getMaxTexturesInShader(webGLVersion: number): number {\n  if (MAX_TEXTURES_IN_SHADER == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURES_IN_SHADER = gl.getParameter(gl.MAX_TEXTURE_IMAGE_UNITS);\n  }\n  // We cap at 16 to avoid spurious runtime \"memory exhausted\" error.\n  return Math.min(16, MAX_TEXTURES_IN_SHADER);\n}\n\nexport function getWebGLDisjointQueryTimerVersion(webGLVersion: number):\n    number {\n  if (webGLVersion === 0) {\n    return 0;\n  }\n\n  let queryTimerVersion: number;\n  const gl = getWebGLContext(webGLVersion);\n\n  if (hasExtension(gl, 'EXT_disjoint_timer_query_webgl2') &&\n      webGLVersion === 2) {\n    queryTimerVersion = 2;\n  } else if (hasExtension(gl, 'EXT_disjoint_timer_query')) {\n    queryTimerVersion = 1;\n  } else {\n    queryTimerVersion = 0;\n  }\n  return queryTimerVersion;\n}\n\nexport function hasExtension(gl: WebGLRenderingContext, extensionName: string) {\n  const ext = gl.getExtension(extensionName);\n  return ext != null;\n}\n\nexport function isWebGLVersionEnabled(webGLVersion: 1|2) {\n  try {\n    const gl = getWebGLContext(webGLVersion);\n    if (gl != null) {\n      return true;\n    }\n  } catch (e) {\n    console.log('Error when getting WebGL context: ', e);\n    return false;\n  }\n  return false;\n}\n\nexport function isCapableOfRenderingToFloatTexture(webGLVersion: number):\n    boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n  } else {\n    if (!hasExtension(gl, 'EXT_color_buffer_float')) {\n      return false;\n    }\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\n/**\n * Check if we can download values from a float/half-float texture.\n *\n * Note that for performance reasons we use binding a texture to a framebuffer\n * as a proxy for ability to download float values later using readPixels. The\n * texture params of this texture will not match those in readPixels exactly\n * but if we are unable to bind some kind of float texture to the frameBuffer\n * then we definitely will not be able to read float values from it.\n */\nexport function isDownloadFloatTextureEnabled(webGLVersion: number): boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n    if (!hasExtension(gl, 'WEBGL_color_buffer_float')) {\n      return false;\n    }\n  } else {\n    if (hasExtension(gl, 'EXT_color_buffer_float')) {\n      return createFloatTextureAndBindToFramebuffer(gl);\n    }\n\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (hasExtension(gl, COLOR_BUFFER_HALF_FLOAT)) {\n      const textureHalfFloatExtension =\n          gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      return createHalfFloatTextureAndBindToFramebuffer(\n          gl, textureHalfFloatExtension);\n    }\n\n    return false;\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\nfunction createFloatTextureAndBindToFramebuffer(gl: WebGLRenderingContext):\n    boolean {\n  const texConfig = getTextureConfig(gl);\n\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nfunction createHalfFloatTextureAndBindToFramebuffer(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension: any): boolean {\n  const texConfig = getTextureConfig(gl, textureHalfFloatExtension);\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatHalfFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeHalfFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nexport function isWebGLFenceEnabled(webGLVersion: number) {\n  if (webGLVersion !== 2) {\n    return false;\n  }\n  const gl = getWebGLContext(webGLVersion);\n\n  // tslint:disable-next-line:no-any\n  const isEnabled = (gl as any).fenceSync != null;\n  return isEnabled;\n}\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${opName} does not support complex64 tensors ` +\n              'in the WebGL backend.');\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {device_util, env} from '@tensorflow/tfjs-core';\n\nimport {getMaxTexturesInShader, getWebGLDisjointQueryTimerVersion, getWebGLMaxTextureSize, isCapableOfRenderingToFloatTexture, isDownloadFloatTextureEnabled, isWebGLFenceEnabled, isWebGLVersionEnabled} from './webgl_util';\n\nconst ENV = env();\n\n/**\n * This file contains WebGL-specific flag registrations.\n */\n\n/**\n * True if WebGL is supported.\n */\nENV.registerFlag('HAS_WEBGL', () => ENV.getNumber('WEBGL_VERSION') > 0);\n\n/** 0: No WebGL, 1: WebGL 1.0, 2: WebGL 2.0. */\nENV.registerFlag('WEBGL_VERSION', () => {\n  if (isWebGLVersionEnabled(2)) {\n    return 2;\n  } else if (isWebGLVersionEnabled(1)) {\n    return 1;\n  }\n  return 0;\n});\n\n/** Whether to check for numerical representation problems. */\nENV.registerFlag('WEBGL_CHECK_NUMERICAL_PROBLEMS', () => false);\n\nENV.registerFlag(\n    'WEBGL_BUFFER_SUPPORTED', () => ENV.get('WEBGL_VERSION') === 2);\n\n/** Whether the WebGL backend will sometimes forward ops to the CPU. */\nENV.registerFlag('WEBGL_CPU_FORWARD', () => true);\n\n/** Whether the WebGL backend will always use f16 textures for rendering. */\nENV.registerFlag('WEBGL_FORCE_F16_TEXTURES', () => false);\n\n/** Whether to turn all packing related flags on. */\nENV.registerFlag('WEBGL_PACK', () => ENV.getBool('HAS_WEBGL'));\n\n/** Whether we will pack the batchnormalization op. */\nENV.registerFlag('WEBGL_PACK_NORMALIZATION', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the clip op. */\nENV.registerFlag('WEBGL_PACK_CLIP', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the depthwise conv op. */\n// TODO: https://github.com/tensorflow/tfjs/issues/1679\nENV.registerFlag('WEBGL_PACK_DEPTHWISECONV', () => false);\n\n/** Whether we will pack binary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_BINARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack unary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_UNARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack array ops. */\nENV.registerFlag(\n    'WEBGL_PACK_ARRAY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack image ops. */\nENV.registerFlag(\n    'WEBGL_PACK_IMAGE_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack reduce ops. */\nENV.registerFlag('WEBGL_PACK_REDUCE', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether packed WebGL kernels lazily unpack their outputs. */\nENV.registerFlag('WEBGL_LAZILY_UNPACK', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will use the im2col algorithm to speed up convolutions. */\nENV.registerFlag('WEBGL_CONV_IM2COL', () => ENV.getBool('WEBGL_PACK'));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURE_SIZE',\n    () => getWebGLMaxTextureSize(ENV.getNumber('WEBGL_VERSION')));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURES_IN_SHADER',\n    () => getMaxTexturesInShader(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * The disjoint_query_timer extension version.\n * 0: disabled, 1: EXT_disjoint_timer_query, 2:\n * EXT_disjoint_timer_query_webgl2.\n * In Firefox with WebGL 2.0,\n * EXT_disjoint_timer_query_webgl2 is not available, so we must use the\n * WebGL 1.0 extension.\n */\nENV.registerFlag('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION', () => {\n  const webGLVersion = ENV.getNumber('WEBGL_VERSION');\n\n  if (webGLVersion === 0) {\n    return 0;\n  }\n  return getWebGLDisjointQueryTimerVersion(webGLVersion);\n});\n\n/**\n * Whether the timer object from the disjoint_query_timer extension gives\n * timing information that is reliable.\n */\nENV.registerFlag(\n    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE',\n    () => ENV.getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0 &&\n        !device_util.isMobile());\n\n/**\n * Whether the device is physically capable of rendering to float32 textures.\n */\nENV.registerFlag(\n    'WEBGL_RENDER_FLOAT32_CAPABLE',\n    () => isCapableOfRenderingToFloatTexture(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Whether rendering to float32 textures is enabled. If disabled, renders to\n * float16 textures.\n */\nENV.registerFlag('WEBGL_RENDER_FLOAT32_ENABLED', () => {\n  return ENV.getBool('WEBGL_FORCE_F16_TEXTURES') ?\n      false :\n      ENV.getBool('WEBGL_RENDER_FLOAT32_CAPABLE');\n});\n\n/**\n * Whether downloading float textures is enabled (16 or 32 bit). If disabled,\n * uses IEEE 754 encoding of the float32 values to 4 uint8 when downloading.\n */\nENV.registerFlag(\n    'WEBGL_DOWNLOAD_FLOAT_ENABLED',\n    () => isDownloadFloatTextureEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/** Whether the fence API is available. */\nENV.registerFlag(\n    'WEBGL_FENCE_API_ENABLED',\n    () => isWebGLFenceEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Tensors with size <= than this will be uploaded as uniforms, not textures.\n */\nENV.registerFlag('WEBGL_SIZE_UPLOAD_UNIFORM', () => {\n  // Use uniform uploads only when 32bit floats are supported. In\n  // 16bit\n  // environments there are problems with comparing a 16bit texture value\n  // with a 32bit uniform value.\n  const useUniforms = ENV.getBool('WEBGL_RENDER_FLOAT32_ENABLED');\n  return useUniforms ? 4 : 0;\n});\n\n/**\n * If the total number of bytes allocated on the GPU is greater than this\n * number, we will aggressively delete textures upon disposal with\n * gl.deleteMatrixTexture, rather than making them available for reuse.\n *\n * Default value -1 indicates that we will never aggressively delete textures.\n */\nENV.registerFlag(\n    'WEBGL_DELETE_TEXTURE_THRESHOLD',\n    () => {\n      return -1;\n    },\n    threshold => {\n      if (threshold < 0 && threshold !== -1) {\n        throw new Error(\n            `WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never ` +\n            `delete) or at least 0, but got ${threshold}.`);\n      }\n    });\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Abs, util } from '@tensorflow/tfjs-core';\nexport function simpleAbsImpl(vals) {\n    const resultValues = new Float32Array(vals.length);\n    for (let i = 0; i < vals.length; ++i) {\n        resultValues[i] = Math.abs(vals[i]);\n    }\n    return resultValues;\n}\nexport const absKernelFunc = (args) => {\n    const { x } = args.inputs;\n    const cpuBackend = args.backend;\n    let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n    if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values;\n        resultValues = simpleAbsImpl(values);\n    }\n    else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals = cpuBackend.data.get(real.dataId).values;\n        const imagVals = cpuBackend.data.get(imag.dataId).values;\n        for (let i = 0; i < realVals.length; i++) {\n            const real = realVals[i];\n            const imag = imagVals[i];\n            resultValues[i] = Math.hypot(real, imag);\n        }\n    }\n    return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n};\nexport const absConfig = {\n    kernelName: Abs,\n    backendName: 'cpu',\n    kernelFunc: absKernelFunc,\n};\n//# sourceMappingURL=Abs.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { backend_util, util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op) {\n    return (aShape, bShape, aVals, bVals, dtype) => {\n        const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n        const resultRank = newShape.length;\n        const resultStrides = util.computeStrides(newShape);\n        const resultSize = util.sizeFromShape(newShape);\n        const result = util.getTypedArrayFromDType(dtype, resultSize);\n        const aRank = aShape.length;\n        const bRank = bShape.length;\n        const aStrides = util.computeStrides(aShape);\n        const bStrides = util.computeStrides(bShape);\n        const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n        const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n        if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n            for (let i = 0; i < result.length; ++i) {\n                result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n            }\n        }\n        else {\n            for (let i = 0; i < result.length; ++i) {\n                const loc = util.indexToLoc(i, resultRank, resultStrides);\n                const aLoc = loc.slice(-aRank);\n                aBroadcastDims.forEach(d => aLoc[d] = 0);\n                const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n                const bLoc = loc.slice(-bRank);\n                bBroadcastDims.forEach(d => bLoc[d] = 0);\n                const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n                result[i] = op(aVals[aIndex], bVals[bIndex]);\n            }\n        }\n        return [result, newShape];\n    };\n}\n//# sourceMappingURL=binary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Add } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal + bReal, imag: aImag + bImag };\n}));\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\nexport const addConfig = {\n    kernelName: Add,\n    backendName: 'cpu',\n    kernelFunc: add\n};\n//# sourceMappingURL=Add.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op) {\n    return (values, dtype, attrs) => {\n        const newValues = util.getTypedArrayFromDType(dtype, values.length);\n        for (let i = 0; i < values.length; ++i) {\n            newValues[i] = op(values[i], attrs);\n        }\n        return newValues;\n    };\n}\n//# sourceMappingURL=unary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Ceil } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\nexport const ceilConfig = {\n    kernelName: Ceil,\n    backendName: 'cpu',\n    kernelFunc: ceilKernelFunc,\n};\n//# sourceMappingURL=Ceil.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Exp } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\nexport const expConfig = {\n    kernelName: Exp,\n    backendName: 'cpu',\n    kernelFunc: expKernelFunc,\n};\n//# sourceMappingURL=Exp.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Expm1 } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\nexport const expm1Config = {\n    kernelName: Expm1,\n    backendName: 'cpu',\n    kernelFunc: expm1KernelFunc,\n};\n//# sourceMappingURL=Expm1.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Floor } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\nexport const floorConfig = {\n    kernelName: Floor,\n    backendName: 'cpu',\n    kernelFunc: floorKernelFunc,\n};\n//# sourceMappingURL=Floor.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Log } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\nexport const logConfig = {\n    kernelName: Log,\n    backendName: 'cpu',\n    kernelFunc: logKernelFunc,\n};\n//# sourceMappingURL=Log.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function maxImpl(aVals, reduceSize, outShape, dtype) {\n    const vals = util.getTypedArrayFromDType(dtype, util.sizeFromShape(outShape));\n    for (let i = 0; i < vals.length; ++i) {\n        const offset = i * reduceSize;\n        let max = aVals[offset];\n        for (let j = 0; j < reduceSize; ++j) {\n            const value = aVals[offset + j];\n            if (value > max) {\n                max = value;\n            }\n        }\n        vals[i] = max;\n    }\n    return vals;\n}\n//# sourceMappingURL=Max_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Multiply } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const multiplyImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n    };\n}));\nexport const multiply = binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\nexport const multiplyConfig = {\n    kernelName: Multiply,\n    backendName: 'cpu',\n    kernelFunc: multiply\n};\n//# sourceMappingURL=Multiply.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Rsqrt } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\nexport const rsqrtConfig = {\n    kernelName: Rsqrt,\n    backendName: 'cpu',\n    kernelFunc: rsqrtKernelFunc,\n};\n//# sourceMappingURL=Rsqrt.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Slice, slice_util, util } from '@tensorflow/tfjs-core';\nimport { assertNotComplex } from '../cpu_util';\nexport function sliceImpl(vals, begin, size, shape, dtype) {\n    const isContinous = slice_util.isSliceContinous(shape, begin, size);\n    const length = util.sizeFromShape(size);\n    const xStrides = util.computeStrides(shape);\n    if (isContinous) {\n        const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n        return vals.subarray(flatOffset, flatOffset + length);\n    }\n    const outVals = util.getTypedArrayFromDType(dtype, length);\n    for (let i = 0; i < length; ++i) {\n        const rank = size.length;\n        const strides = util.computeStrides(size);\n        const loc = util.indexToLoc(i, rank, strides);\n        const xLoc = loc.map((idx, j) => idx + begin[j]);\n        const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n        outVals[i] = vals[xIndex];\n    }\n    return outVals;\n}\nexport function slice(args) {\n    const { inputs, backend, attrs } = args;\n    const { x } = inputs;\n    const { begin, size } = attrs;\n    assertNotComplex(x, 'slice');\n    const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n    slice_util.assertParamsValid(x, $begin, $size);\n    const vals = backend.data.get(x.dataId).values;\n    const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n    return backend.makeTensorInfo($size, x.dtype, outVals);\n}\nexport const sliceConfig = {\n    kernelName: Slice,\n    backendName: 'cpu',\n    kernelFunc: slice\n};\n//# sourceMappingURL=Slice.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Sub } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const subImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal - bReal, imag: aImag - bImag };\n}));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\nexport const subConfig = {\n    kernelName: Sub,\n    backendName: 'cpu',\n    kernelFunc: sub\n};\n//# sourceMappingURL=Sub.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function transposeImpl(xVals, xShape, dtype, perm, newShape) {\n    const xRank = xShape.length;\n    const xSize = util.sizeFromShape(xShape);\n    const xStrides = util.computeStrides(xShape);\n    const newStrides = util.computeStrides(newShape);\n    const result = util.getTypedArrayFromDType(dtype, util.sizeFromShape(newShape));\n    for (let i = 0; i < xSize; ++i) {\n        const loc = util.indexToLoc(i, xRank, xStrides);\n        // Permute location.\n        const newLoc = new Array(loc.length);\n        for (let i = 0; i < newLoc.length; i++) {\n            newLoc[i] = loc[perm[i]];\n        }\n        const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n        result[newIndex] = xVals[i];\n    }\n    return result;\n}\n//# sourceMappingURL=Transpose_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { TensorBuffer, util } from '@tensorflow/tfjs-core';\nexport function uniqueImpl(values, axis, shape, dtype) {\n    // Normalize and validate axis.\n    const $axis = util.parseAxisParam(axis, shape)[0];\n    // Calculate the new shape that is suitable for extracting data along the\n    // given axis.\n    //\n    // The rank is 3.\n    // The size of the 1st dimension is the size of all the axes < the given axis.\n    // The size of the 2nd dimension is the same as the size of the given axis.\n    // The size of the 3rd dimension is the size of all the axes > the given axis.\n    //\n    // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n    // newShape would be: [2*3, 5, 4].\n    //\n    // Note that this is not the final output shape. This will be the shape for an\n    // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n    // values along the given axis. To demonstrate how it works, consider the\n    // following example:\n    //\n    // Input: a 3D tensor, with shape [1, 2, 3]\n    // [\n    //   [\n    //      [1,2,3],\n    //      [4,5,6]\n    //   ]\n    // ]\n    // Axis: 2 (the last axis).\n    // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n    //\n    // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n    // 1*2. The re-shaped data would look like:\n    //\n    // [\n    //   [\n    //     [1], [2], [3]\n    //   ],\n    //   [\n    //     [4], [5], [6]\n    //   ]\n    // ]\n    //\n    // Then, we can construct a 3-level nested loop by the following dimension\n    // order to extract the values along the axis (dimension1):\n    // i: dimension1       // 0,1,2 (newShape[1])\n    //   m: dimension0     // 0,1   (newShape[0])\n    //     n: dimension2   // 0     (newShape[2])\n    //\n    //                       m, i, n\n    //                      ---------\n    // Iteration 0: data at [0, 0, 0] => \"1\"\n    // Iteration 1: data at [1, 0, 0] => \"4\"\n    // We got [1,4].\n    // Iteration 2: data at [0, 1, 0] => \"2\"\n    // Iteration 3: data at [1, 1, 0] => \"5\"\n    // We got [2,5].\n    // Iteration 4: data at [0, 2, 0] => \"3\"\n    // Iteration 5: data at [1, 2, 0] => \"6\"\n    // We got [3,6].\n    const newShape = [1, shape[0], 1];\n    for (let i = 0; i < $axis; i++) {\n        newShape[0] *= shape[i];\n    }\n    newShape[1] = shape[$axis];\n    for (let i = $axis + 1; i < shape.length; i++) {\n        newShape[2] *= shape[i];\n    }\n    // A map from unique elements (their string representations) to their values\n    // in \"indices\" (below).\n    const uniqueElements = {};\n    // The indices of each unique element in the original tensor along the given\n    // axis. It is 1D and has the same size as the given axis.\n    const indices = new Int32Array(shape[$axis]);\n    // Create a buffer so we can easily extract value at a given location.\n    const inputBuffer = new TensorBuffer(newShape, dtype, values);\n    // The indices along the given axis that have unique elements. This is a\n    // de-duped version of \"indices\" above.\n    const uniqueIndices = [];\n    const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n    for (let i = 0; i < shape[$axis]; i++) {\n        // Extract values along the axis.\n        let element;\n        if (is1DTensor) {\n            // Fast path for 1D tensor input.\n            element = values[i].toString();\n        }\n        else {\n            const axisValues = [];\n            for (let m = 0; m < newShape[0]; m++) {\n                for (let n = 0; n < newShape[2]; n++) {\n                    axisValues.push(inputBuffer.get(m, i, n));\n                }\n            }\n            element = axisValues.join(',');\n        }\n        // Dedup and update various indices.\n        if (uniqueElements[element] !== undefined) {\n            indices[i] = uniqueElements[element];\n        }\n        else {\n            const uniqueIndex = Object.keys(uniqueElements).length;\n            uniqueElements[element] = uniqueIndex;\n            indices[i] = uniqueIndex;\n            uniqueIndices.push(i);\n        }\n    }\n    // Now we know where each of the unique elements are located along the axis\n    // (uniqueIndices). Extract them from input buffer and store them in the\n    // output buffer.\n    const outputTmpShape = newShape.slice();\n    outputTmpShape[1] = Object.keys(uniqueElements).length;\n    const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n    uniqueIndices.forEach((uniqueElementIndex, i) => {\n        for (let m = 0; m < newShape[0]; m++) {\n            for (let n = 0; n < newShape[2]; n++) {\n                outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n            }\n        }\n    });\n    // The output shape can be calculated from the input shape with the size of\n    // the given axis replaced by the number of unique elements along that axis.\n    const outputShape = shape.slice();\n    outputShape[$axis] = outputTmpShape[1];\n    return {\n        outputValues: outputBuffer.values,\n        outputShape,\n        indices,\n    };\n}\n//# sourceMappingURL=Unique_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import shared functionality from tfjs-backend-cpu without triggering\n// side effects.\n// tslint:disable-next-line: no-imports-from-dist\nimport * as shared from '@tensorflow/tfjs-backend-cpu/dist/shared';\n\nconst {\n  simpleAbsImpl: simpleAbsImplCPU,\n  addImpl: addImplCPU,\n  ceilImpl: ceilImplCPU,\n  expImpl: expImplCPU,\n  expm1Impl: expm1ImplCPU,\n  floorImpl: floorImplCPU,\n  logImpl: logImplCPU,\n  maxImpl: maxImplCPU,\n  multiplyImpl: multiplyImplCPU,\n  rsqrtImpl: rsqrtImplCPU,\n  sliceImpl: sliceImplCPU,\n  subImpl: subImplCPU,\n  transposeImpl: transposeImplCPU,\n  uniqueImpl: uniqueImplCPU,\n} = shared;\n\nexport {\n  simpleAbsImplCPU,\n  addImplCPU,\n  ceilImplCPU,\n  expImplCPU,\n  expm1ImplCPU,\n  logImplCPU,\n  multiplyImplCPU,\n  sliceImplCPU,\n  subImplCPU,\n  floorImplCPU,\n  maxImplCPU,\n  rsqrtImplCPU,\n  transposeImplCPU,\n  uniqueImplCPU,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`float v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        float result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`vec4 v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        vec4 result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ArgMinMaxProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo, op: 'max'|'min',\n      firstPass: boolean) {\n    const {windowSize, batchSize, outSize} = reduceInfo;\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    this.outputShape = [batchSize, outSize];\n    const compOp = (op === 'max') ? '>' : '<';\n    const indexSnippet = firstPass ?\n        'inOffset + i;' :\n        'round(getBestIndicesA(batch, inOffset + i));';\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        int bestIndex = inOffset;\n        float bestValue = getA(batch, bestIndex);\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          int inIdx = ${indexSnippet};\n          float candidate = getA(batch, inIdx);\n          if (candidate ${compOp} bestValue) {\n            bestValue = candidate;\n            bestIndex = inIdx;\n          }\n        }\n        setOutput(float(bestIndex));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport function getVecChannels(name: string, rank: number): string[] {\n  return ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank).map(d => `${name}.${d}`);\n}\n\nexport function getChannels(name: string, rank: number): string[] {\n  if (rank === 1) {\n    return [name];\n  }\n  return getVecChannels(name, rank);\n}\n\nexport function getSourceCoords(rank: number, dims: string[]): string {\n  if (rank === 1) {\n    return 'rc';\n  }\n\n  let coords = '';\n  for (let i = 0; i < rank; i++) {\n    coords += dims[i];\n    if (i < rank - 1) {\n      coords += ',';\n    }\n  }\n  return coords;\n}", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '@tensorflow/tfjs-core';\n\nexport type GLSL = {\n  version: string,\n  attribute: string,\n  varyingVs: string,\n  varyingFs: string,\n  texture2D: string,\n  output: string,\n  defineOutput: string,\n  defineSpecialNaN: string,\n  defineSpecialInf: string,\n  defineRound: string\n};\n\nexport function getGlslDifferences(): GLSL {\n  let version: string;\n  let attribute: string;\n  let varyingVs: string;\n  let varyingFs: string;\n  let texture2D: string;\n  let output: string;\n  let defineOutput: string;\n  let defineSpecialNaN: string;\n  let defineSpecialInf: string;\n  let defineRound: string;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    version = '#version 300 es';\n    attribute = 'in';\n    varyingVs = 'out';\n    varyingFs = 'in';\n    texture2D = 'texture';\n    output = 'outputColor';\n    defineOutput = 'out vec4 outputColor;';\n\n    // Use custom isnan definition to work across differences between\n    // implementations on various platforms. While this should happen in ANGLE\n    // we still see differences between android and windows (on chrome) when\n    // using isnan directly.\n    defineSpecialNaN = `\n      bool isnan_custom(float val) {\n        return (val > 0.0 || val < 0.0) ? false : val != 0.0;\n      }\n\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan_custom(val.x),\n          isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));\n      }\n\n      #define isnan(value) isnan_custom(value)\n    `;\n    // In webgl 2 we do not need to specify a custom isinf so there is no\n    // need for a special INFINITY constant.\n    defineSpecialInf = ``;\n    defineRound = `\n      #define round(value) newRound(value)\n      int newRound(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 newRound(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  } else {\n    version = '';\n    attribute = 'attribute';\n    varyingVs = 'varying';\n    varyingFs = 'varying';\n    texture2D = 'texture2D';\n    output = 'gl_FragColor';\n    defineOutput = '';\n    // WebGL1 has no built in isnan so we define one here.\n    defineSpecialNaN = `\n      #define isnan(value) isnan_custom(value)\n      bool isnan_custom(float val) {\n        return (val > 0. || val < 1. || val == 0.) ? false : true;\n      }\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));\n      }\n    `;\n    defineSpecialInf = `\n      uniform float INFINITY;\n\n      bool isinf(float val) {\n        return abs(val) == INFINITY;\n      }\n      bvec4 isinf(vec4 val) {\n        return equal(abs(val), vec4(INFINITY));\n      }\n    `;\n    defineRound = `\n      int round(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 round(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  }\n\n  return {\n    version,\n    attribute,\n    varyingVs,\n    varyingFs,\n    texture2D,\n    output,\n    defineOutput,\n    defineSpecialNaN,\n    defineSpecialInf,\n    defineRound\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\n/**\n * Produces GLSL code that derives logical coordinates from a flat\n * index. The code performs integer division with each stride and decrements\n * the index until the index equals the final dimension coordinate.\n */\nexport function getLogicalCoordinatesFromFlatIndex(\n    coords: string[], shape: number[], index = 'index'): string {\n  const strides = util.computeStrides(shape);\n  return strides\n      .map((stride, i) => {\n        const line1 = `int ${coords[i]} = ${index} / ${stride}`;\n        const line2 = i === strides.length - 1 ?\n            `int ${coords[i + 1]} = ${index} - ${coords[i]} * ${stride}` :\n            `index -= ${coords[i]} * ${stride}`;\n        return `${line1}; ${line2};`;\n      })\n      .join('');\n}\n\nfunction buildVec(x: string[]): string {\n  if (x.length === 1) {\n    return `${x[0]}`;\n  }\n  return `vec${x.length}(${x.join(',')})`;\n}\n\n/**\n * Produces GLSL code that computes the dot product of the input x and y\n * vectors. Handles splitting inputs into increments of vec4s when necessary.\n */\nexport function dotify(x: string[], y: string[]): string {\n  if (x.length !== y.length) {\n    throw new Error(\n        `Vectors to be dotted must be of the same length -` +\n        `got ${x.length} and ${y.length}`);\n  }\n\n  const slices: string[] = [];\n  const nearestVec4 = Math.floor(x.length / 4);\n  const nearestVec4Remainder = x.length % 4;\n\n  for (let i = 0; i < nearestVec4; i++) {\n    const xSlice = x.slice(i * 4, i * 4 + 4);\n    const ySlice = y.slice(i * 4, i * 4 + 4);\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  if (nearestVec4Remainder !== 0) {\n    let xSlice = x.slice(nearestVec4 * 4);\n    let ySlice = y.slice(nearestVec4 * 4);\n    if (xSlice.length === 1) {\n      xSlice = xSlice.map(d => `float(${d})`);\n      ySlice = ySlice.map(d => `float(${d})`);\n    }\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  return slices.map((d, i) => `dot(${d})`).join('+');\n}\n\n/**\n * Produces GLSL that computes the flat index from 3D coordinates.\n */\nexport function getFlatIndexFrom3D(shape: [number, number, number]): string {\n  const strides = util.computeStrides(shape).map(d => d.toString());\n\n  return `\n  int getFlatIndex(ivec3 coords) {\n    return coords.x * ${strides[0]} + coords.y * ${strides[1]} + coords.z;\n  }\n`;\n}\n\nexport const ENCODE_FLOAT_SNIPPET = `\n  const float FLOAT_MAX = 1.70141184e38;\n  const float FLOAT_MIN = 1.17549435e-38;\n\n  lowp vec4 encode_float(highp float v) {\n    if (isnan(v)) {\n      return vec4(255, 255, 255, 255);\n    }\n\n    highp float av = abs(v);\n\n    if(av < FLOAT_MIN) {\n      return vec4(0.0, 0.0, 0.0, 0.0);\n    } else if(v > FLOAT_MAX) {\n      return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;\n    } else if(v < -FLOAT_MAX) {\n      return vec4(0.0, 0.0,  128.0, 255.0) / 255.0;\n    }\n\n    highp vec4 c = vec4(0,0,0,0);\n\n    highp float e = floor(log2(av));\n    highp float m = exp2(fract(log2(av))) - 1.0;\n\n    c[2] = floor(128.0 * m);\n    m -= c[2] / 128.0;\n    c[1] = floor(32768.0 * m);\n    m -= c[1] / 32768.0;\n    c[0] = floor(8388608.0 * m);\n\n    highp float ebias = e + 127.0;\n    c[3] = floor(ebias / 2.0);\n    ebias -= c[3] * 2.0;\n    c[2] += floor(ebias) * 128.0;\n\n    c[3] += 128.0 * step(0.0, -v);\n\n    return c / 255.0;\n  }\n`;\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\nconst {getBroadcastDims} = backend_util;\nimport {getGlslDifferences, GLSL} from './glsl_version';\nimport * as shader_util from './shader_compiler_util';\n\nexport type ShapeInfo = {\n  logicalShape: number[],\n  texShape: [number, number],\n  isUniform: boolean,\n  isPacked: boolean,\n  flatOffset: number\n};\n\nexport type InputInfo = {\n  name: string,\n  shapeInfo: ShapeInfo\n};\n\nexport function makeShader(\n    inputsInfo: InputInfo[], outputShape: ShapeInfo, userCode: string,\n    usesPackedTextures: boolean): string {\n  const prefixSnippets: string[] = [];\n  inputsInfo.forEach(x => {\n    const size = util.sizeFromShape(x.shapeInfo.logicalShape);\n\n    // Snippet when we decided to upload the values as uniform.\n    if (x.shapeInfo.isUniform) {\n      prefixSnippets.push(\n          `uniform float ${x.name}${size > 1 ? `[${size}]` : ''};`);\n    } else {\n      prefixSnippets.push(`uniform sampler2D ${x.name};`);\n      prefixSnippets.push(`uniform int offset${x.name};`);\n    }\n  });\n  const inputPrefixSnippet = prefixSnippets.join('\\n');\n\n  const inputSamplingSnippet =\n      inputsInfo\n          .map(x => getInputSamplingSnippet(x, outputShape, usesPackedTextures))\n          .join('\\n');\n  const outTexShape = outputShape.texShape;\n  const glsl = getGlslDifferences();\n  const floatTextureSampleSnippet = getFloatTextureSampleSnippet(glsl);\n  let outputSamplingSnippet: string;\n  let floatTextureSetOutputSnippet: string;\n  let shaderPrefix = getShaderPrefix(glsl);\n\n  if (outputShape.isPacked) {\n    outputSamplingSnippet =\n        getPackedOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRGBASnippet(glsl);\n  } else {\n    outputSamplingSnippet =\n        getOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRSnippet(glsl);\n  }\n\n  if (usesPackedTextures) {\n    shaderPrefix += SHADER_PACKED_PREFIX;\n  }\n\n  const source = [\n    shaderPrefix, floatTextureSampleSnippet, floatTextureSetOutputSnippet,\n    inputPrefixSnippet, outputSamplingSnippet, inputSamplingSnippet, userCode\n  ].join('\\n');\n  return source;\n}\n\nfunction getSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getSamplerScalar(inInfo);\n    case 1:\n      return getSampler1D(inInfo);\n    case 2:\n      return getSampler2D(inInfo);\n    case 3:\n      return getSampler3D(inInfo);\n    case 4:\n      return getSampler4D(inInfo);\n    case 5:\n      return getSampler5D(inInfo);\n    case 6:\n      return getSampler6D(inInfo);\n    default:\n      throw new Error(\n          `${shape.length}-D input sampling` +\n          ` is not yet supported`);\n  }\n}\n\nfunction getPackedSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getPackedSamplerScalar(inInfo);\n    case 1:\n      return getPackedSampler1D(inInfo);\n    case 2:\n      return getPackedSampler2D(inInfo);\n    case 3:\n      return getPackedSampler3D(inInfo);\n    default:\n      return getPackedSamplerND(inInfo);\n  }\n}\n\nfunction getInputSamplingSnippet(\n    inInfo: InputInfo, outShapeInfo: ShapeInfo,\n    usesPackedTextures = false): string {\n  let res = '';\n  if (usesPackedTextures) {\n    res += getPackedSamplerFromInInfo(inInfo);\n  } else {\n    res += getSamplerFromInInfo(inInfo);\n  }\n\n  const inShape = inInfo.shapeInfo.logicalShape;\n  const outShape = outShapeInfo.logicalShape;\n  if (inShape.length <= outShape.length) {\n    if (usesPackedTextures) {\n      res += getPackedSamplerAtOutputCoords(inInfo, outShapeInfo);\n    } else {\n      res += getSamplerAtOutputCoords(inInfo, outShapeInfo);\n    }\n  }\n  return res;\n}\n\nfunction getPackedOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutputPacked1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutputPacked2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutputPacked3DCoords(\n          outShape as [number, number, number], outTexShape);\n    default:\n      return getOutputPackedNDCoords(outShape, outTexShape);\n  }\n}\n\nfunction getOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutput1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutput2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutput3DCoords(\n          outShape as [number, number, number], outTexShape);\n    case 4:\n      return getOutput4DCoords(\n          outShape as [number, number, number, number], outTexShape);\n    case 5:\n      return getOutput5DCoords(\n          outShape as [number, number, number, number, number], outTexShape);\n    case 6:\n      return getOutput6DCoords(\n          outShape as [number, number, number, number, number, number],\n          outTexShape);\n    default:\n      throw new Error(\n          `${outShape.length}-D output sampling is not yet supported`);\n  }\n}\n\nfunction getFloatTextureSampleSnippet(glsl: GLSL): string {\n  return `\n    float sampleTexture(sampler2D textureSampler, vec2 uv) {\n      return ${glsl.texture2D}(textureSampler, uv).r;\n    }\n  `;\n}\n\nfunction getFloatTextureSetRSnippet(glsl: GLSL): string {\n  return `\n    void setOutput(float val) {\n      ${glsl.output} = vec4(val, 0, 0, 0);\n    }\n  `;\n}\n\nfunction getFloatTextureSetRGBASnippet(glsl: GLSL): string {\n  return `\n    void setOutput(vec4 val) {\n      ${glsl.output} = val;\n    }\n  `;\n}\n\nfunction getShaderPrefix(glsl: GLSL): string {\n  const SHADER_PREFIX = `${glsl.version}\n    precision highp float;\n    precision highp int;\n    precision highp sampler2D;\n    ${glsl.varyingFs} vec2 resultUV;\n    ${glsl.defineOutput}\n    const vec2 halfCR = vec2(0.5, 0.5);\n\n    struct ivec5\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n    };\n\n    struct ivec6\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n      int v;\n    };\n\n    uniform float NAN;\n    ${glsl.defineSpecialNaN}\n    ${glsl.defineSpecialInf}\n    ${glsl.defineRound}\n\n    int imod(int x, int y) {\n      return x - y * (x / y);\n    }\n\n    int idiv(int a, int b, float sign) {\n      int res = a / b;\n      int mod = imod(a, b);\n      if (sign < 0. && mod != 0) {\n        res -= 1;\n      }\n      return res;\n    }\n\n    //Based on the work of Dave Hoskins\n    //https://www.shadertoy.com/view/4djSRW\n    #define HASHSCALE1 443.8975\n    float random(float seed){\n      vec2 p = resultUV * seed;\n      vec3 p3  = fract(vec3(p.xyx) * HASHSCALE1);\n      p3 += dot(p3, p3.yzx + 19.19);\n      return fract((p3.x + p3.y) * p3.z);\n    }\n\n    ${SAMPLE_1D_SNIPPET}\n    ${SAMPLE_2D_SNIPPET}\n    ${SAMPLE_3D_SNIPPET}\n  `;\n\n  return SHADER_PREFIX;\n}\n\nconst SAMPLE_1D_SNIPPET = `\nvec2 uvFromFlat(int texNumR, int texNumC, int index) {\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\nvec2 packedUVfrom1D(int texNumR, int texNumC, int index) {\n  int texelIndex = index / 2;\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_2D_SNIPPET = `\nvec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,\n  int texNumC, int row, int col) {\n  int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_3D_SNIPPET = `\nvec2 packedUVfrom3D(int texNumR, int texNumC,\n    int texelsInBatch, int texelsInLogicalRow, int b,\n    int row, int col) {\n  int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SHADER_PACKED_PREFIX = `\n  float getChannel(vec4 frag, vec2 innerDims) {\n    vec2 modCoord = mod(innerDims, 2.);\n    return modCoord.x == 0. ?\n      (modCoord.y == 0. ? frag.r : frag.g) :\n      (modCoord.y == 0. ? frag.b : frag.a);\n  }\n  float getChannel(vec4 frag, int dim) {\n    float modCoord = mod(float(dim), 2.);\n    return modCoord == 0. ? frag.r : frag.g;\n  }\n`;\n\nfunction getOutputScalarCoords() {\n  return `\n    int getOutputCoords() {\n      return 0;\n    }\n  `;\n}\n\nfunction getOutputPacked1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (packedTexShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.x * ${packedTexShape[1]}.0);\n      }\n    `;\n  }\n\n  if (packedTexShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.y * ${packedTexShape[0]}.0);\n      }\n    `;\n  }\n\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      return 2 * (resTexRC.x * ${packedTexShape[1]} + resTexRC.y);\n    }\n  `;\n}\n\nfunction getOutput1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  if (texShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.x * ${texShape[1]}.0);\n      }\n    `;\n  }\n  if (texShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.y * ${texShape[0]}.0);\n      }\n    `;\n  }\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      return resTexRC.x * ${texShape[1]} + resTexRC.y;\n    }\n  `;\n}\n\nfunction getOutputPacked3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texelsInLogicalRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = texelsInLogicalRow * Math.ceil(shape[1] / 2);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec3(b, r, c);\n    }\n  `;\n}\n\nfunction getOutput3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n\nfunction getOutputPackedNDCoords(\n    shape: number[], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  const texelsInLogicalRow = Math.ceil(shape[shape.length - 1] / 2);\n  const texelsInBatch =\n      texelsInLogicalRow * Math.ceil(shape[shape.length - 2] / 2);\n  let texelsInBatchN = texelsInBatch;\n  let batches = ``;\n  let coords = 'b, r, c';\n\n  for (let b = 2; b < shape.length - 1; b++) {\n    texelsInBatchN *= shape[shape.length - b - 1];\n    batches = `\n      int b${b} = index / ${texelsInBatchN};\n      index -= b${b} * ${texelsInBatchN};\n    ` + batches;\n    coords = `b${b}, ` + coords;\n  }\n\n  return `\n    ivec${shape.length} getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      ${batches}\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec${shape.length}(${coords});\n    }\n  `;\n}\n\nfunction getOutput4DCoords(\n    shape: [number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2'], shape);\n\n  return `\n    ivec4 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec4(r, c, d, d2);\n    }\n  `;\n}\n\nfunction getOutput5DCoords(\n    shape: [number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3'], shape);\n\n  return `\n    ivec5 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx * vec2(${texShape[0]},\n                             ${texShape[1]}));\n\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec5 outShape = ivec5(r, c, d, d2, d3);\n      return outShape;\n    }\n  `;\n}\n\nfunction getOutput6DCoords(\n    shape: [number, number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3', 'd4'], shape);\n\n  return `\n    ivec6 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec6 result = ivec6(r, c, d, d2, d3, d4);\n      return result;\n    }\n  `;\n}\n\nfunction getOutputPacked2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return 2 * ivec2(resultUV.yx * vec2(${packedTexShape[0]}, ${\n        packedTexShape[1]}));\n      }\n    `;\n  }\n\n  // texels needed to accommodate a logical row\n  const texelsInLogicalRow = Math.ceil(shape[1] / 2);\n\n  /**\n   * getOutputCoords\n   *\n   * resTexRC: The rows and columns of the texels. If you move over one\n   * texel to the right in the packed texture, you are moving over one column\n   * (not two).\n   *\n   * index: The texel index\n   */\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getOutput2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return ivec2(resultUV.yx * vec2(${texShape[0]}, ${texShape[1]}));\n      }\n    `;\n  }\n  if (shape[1] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(index, 0);\n      }\n    `;\n  }\n  if (shape[0] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(0, index);\n      }\n    `;\n  }\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      int r = index / ${shape[1]};\n      int c = index - r * ${shape[1]};\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getFlatOffsetUniformName(texName: string): string {\n  return `offset${texName}`;\n}\n\nfunction getPackedSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}() {\n      return ${glsl.texture2D}(${texName}, halfCR);\n    }\n  `;\n}\n\nfunction getSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  if (inputInfo.shapeInfo.isUniform) {\n    return `float ${funcName}() {return ${texName};}`;\n  }\n  const [texNumR, texNumC] = inputInfo.shapeInfo.texShape;\n  if (texNumR === 1 && texNumC === 1) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n\n  const [tNumR, tNumC] = inputInfo.shapeInfo.texShape;\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}() {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int index) {\n      vec2 uv = packedUVfrom1D(\n        ${packedTexShape[0]}, ${packedTexShape[1]}, index);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int index) {\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const tNumR = texShape[0];\n  const tNumC = texShape[1];\n\n  if (tNumC === 1 && tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  if (tNumC === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2(0.5, (float(index + ${offset}) + 0.5) / ${tNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2((float(index + ${offset}) + 0.5) / ${tNumC}.0, 0.5);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  return `\n    float ${funcName}(int index) {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const glsl = getGlslDifferences();\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    return `\n      vec4 ${funcName}(int row, int col) {\n        vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n\n        return ${glsl.texture2D}(${texName}, uv);\n      }\n    `;\n  }\n\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const valuesPerRow = Math.ceil(shape[1] / 2);\n\n  return `\n    vec4 ${funcName}(int row, int col) {\n      vec2 uv = packedUVfrom2D(${valuesPerRow}, ${packedTexShape[0]}, ${\n      packedTexShape[1]}, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    const texNumR = texShape[0];\n    const texNumC = texShape[1];\n    return `\n    float ${funcName}(int row, int col) {\n      vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col) {\n        int index = round(dot(vec2(row, col), vec2(${shape[1]}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const offset = getFlatOffsetUniformName(texName);\n  if (texNumC === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2(0.5, (index + 0.5) / ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n  if (texNumR === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2((index + 0.5) / ${texNumC}.0, 0.5);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  return `\n  float ${funcName}(int row, int col) {\n    // Explicitly use integer operations as dot() only works on floats.\n    int index = row * ${shape[1]} + col + ${offset};\n    vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n    return sampleTexture(${texName}, uv);\n  }\n`;\n}\n\nfunction getPackedSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  if (shape[0] === 1) {\n    const squeezedShape = shape.slice(1);\n    const keptDims = [1, 2];\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['b', 'row', 'col'];\n    return `\n        ${getPackedSamplerFromInInfo(newInputInfo)}\n        vec4 ${funcName}(int b, int row, int col) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = valuesPerRow * Math.ceil(shape[1] / 2);\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int b, int row, int col) {\n      vec2 uv = packedUVfrom3D(\n        ${texNumR}, ${texNumC}, ${texelsInBatch}, ${valuesPerRow}, b, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride0 = shape[1] * shape[2];\n  const stride1 = shape[2];\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col', 'depth'];\n    return `\n        ${getSamplerFromInInfo(newInputInfo)}\n        float ${funcName}(int row, int col, int depth) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth) {\n        int index = round(dot(vec3(row, col, depth),\n                          vec3(${stride0}, ${stride1}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n        float ${funcName}(int row, int col, int depth) {\n          float texR = float(row);\n          float texC = dot(vec2(col, depth), vec2(${stride1}, 1));\n          vec2 uv = (vec2(texC, texR) + halfCR) /\n                     vec2(${texNumC}.0, ${texNumR}.0);\n          return sampleTexture(${texName}, uv);\n        }\n      `;\n  }\n\n  if (texNumC === stride1 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col, int depth) {\n      float texR = dot(vec2(row, col), vec2(${shape[1]}, 1));\n      float texC = float(depth);\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n      float ${funcName}(int row, int col, int depth) {\n        // Explicitly use integer operations as dot() only works on floats.\n        int index = row * ${stride0} + col * ${stride1} + depth + ${offset};\n        vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n        return sampleTexture(${texName}, uv);\n      }\n  `;\n}\n\nfunction getPackedSamplerND(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const rank = shape.length;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[rank - 1] / 2);\n  let texelsInBatch = valuesPerRow * Math.ceil(shape[rank - 2] / 2);\n  let params = `int b, int row, int col`;\n  let index = `b * ${texelsInBatch} + (row / 2) * ${valuesPerRow} + (col / 2)`;\n  for (let b = 2; b < rank - 1; b++) {\n    params = `int b${b}, ` + params;\n    texelsInBatch *= shape[rank - b - 1];\n    index = `b${b} * ${texelsInBatch} + ` + index;\n  }\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}(${params}) {\n      int index = ${index};\n      int texR = index / ${texNumC};\n      int texC = index - texR * ${texNumC};\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}, ${texNumR});\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler4D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride2 = shape[3];\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        int index = round(dot(vec4(row, col, depth, depth2),\n                          vec4(${stride0}, ${stride1}, ${stride2}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = float(row);\n        float texC =\n            dot(vec3(col, depth, depth2),\n                vec3(${stride1}, ${stride2}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride2 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = dot(vec3(row, col, depth),\n                         vec3(${shape[1] * shape[2]}, ${shape[2]}, 1));\n        float texC = float(depth2);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} +\n          depth * ${stride2} + depth2;\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler5D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride3 = shape[4];\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float index = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          depth3;\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n                         vec4(${stride1}, ${stride2}, ${stride3}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  if (texNumC === stride3 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float texR = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3]},\n               ${shape[2] * shape[3]}, ${shape[3]}, 1));\n        int texC = depth3;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler6D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3', 'depth4'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  const stride4 = shape[5];\n  const stride3 = shape[4] * stride4;\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n        int index = round(dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          dot(\n            vec2(depth3, depth4),\n            vec2(${stride4}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n          vec4(${stride1}, ${stride2}, ${stride3}, ${stride4})) +\n               float(depth4);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride4 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        float texR = dot(vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3] * shape[4]},\n               ${shape[2] * shape[3] * shape[4]},\n               ${shape[3] * shape[4]},\n               ${shape[4]})) + float(depth3);\n        int texC = depth4;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 * ${stride4} + depth4 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getUniformSampler(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n\n  if (inSize < 2) {\n    return `return ${texName};`;\n  }\n  return `\n    for (int i = 0; i < ${inSize}; i++) {\n      if (i == index) {\n        return ${texName}[i];\n      }\n    }\n  `;\n}\n\nfunction getPackedSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n\n  const type = getCoordsDataType(outRank);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  let output = `return outputValue;`;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n  const isInputScalar = inSize === 1;\n  const outSize = util.sizeFromShape(outShapeInfo.logicalShape);\n  const isOutputScalar = outSize === 1;\n\n  if (inRank === 1 && !isInputScalar && !isOutputScalar) {\n    output = `\n      return vec4(outputValue.xy, outputValue.xy);\n    `;\n  } else if (isInputScalar && !isOutputScalar) {\n    if (outRank === 1) {\n      output = `\n        return vec4(outputValue.x, outputValue.x, 0., 0.);\n      `;\n    } else {\n      output = `\n        return vec4(outputValue.x);\n      `;\n    }\n  } else if (broadcastDims.length) {\n    const rows = inRank - 2;\n    const cols = inRank - 1;\n\n    if (broadcastDims.indexOf(rows) > -1 && broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.x);`;\n    } else if (broadcastDims.indexOf(rows) > -1) {\n      output = `return vec4(outputValue.x, outputValue.y, ` +\n          `outputValue.x, outputValue.y);`;\n    } else if (broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.xx, outputValue.zz);`;\n    }\n  }\n\n  return `\n    vec4 ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      vec4 outputValue = get${texFuncSnippet}(${unpackedCoordsSnippet});\n      ${output}\n    }\n  `;\n}\n\nfunction getSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const outTexShape = outShapeInfo.texShape;\n  const inTexShape = inputInfo.shapeInfo.texShape;\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  if (!inputInfo.shapeInfo.isUniform && inRank === outRank &&\n      inputInfo.shapeInfo.flatOffset == null &&\n      util.arraysEqual(inTexShape, outTexShape)) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, resultUV);\n      }\n    `;\n  }\n\n  const type = getCoordsDataType(outRank);\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  return `\n    float ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      return get${texFuncSnippet}(${unpackedCoordsSnippet});\n    }\n  `;\n}\n\nexport function getCoordsDataType(rank: number): string {\n  if (rank <= 1) {\n    return 'int';\n  } else if (rank === 2) {\n    return 'ivec2';\n  } else if (rank === 3) {\n    return 'ivec3';\n  } else if (rank === 4) {\n    return 'ivec4';\n  } else if (rank === 5) {\n    return 'ivec5';\n  } else if (rank === 6) {\n    return 'ivec6';\n  } else {\n    throw Error(`GPU for rank ${rank} is not yet supported`);\n  }\n}\n\n/** Returns a new input info (a copy) that has a squeezed logical shape. */\nfunction squeezeInputInfo(\n    inInfo: InputInfo, squeezedShape: number[]): InputInfo {\n  // Deep copy.\n  const newInputInfo: InputInfo = JSON.parse(JSON.stringify(inInfo));\n  newInputInfo.shapeInfo.logicalShape = squeezedShape;\n  return newInputInfo;\n}\n\nfunction getSqueezedParams(params: string[], keptDims: number[]): string {\n  return keptDims.map(d => params[d]).join(', ');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ArgMinMaxPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      shape: number[], windowSize: number, op: 'max'|'min',\n      firstPass: boolean) {\n    util.assert(\n        shape.length > 2,\n        () => `Packed arg${\n            op.charAt(0).toUpperCase() +\n            op.slice(1)} supports only inputs with rank above 2.`);\n    const inSize = shape[shape.length - 1];\n    const outSize = Math.ceil(inSize / windowSize);\n    this.outputShape = shape.slice(0, -1);\n    if (outSize > 1) {\n      this.outputShape.push(outSize);\n    }\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    const outShape = this.outputShape;\n    const rank = outShape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n\n    let sourceLocSetup;\n    let sourceRank;\n    if (outSize === 1) {\n      sourceRank = rank + 1;\n      const sourceLocDType = getCoordsDataType(sourceRank);\n      sourceLocSetup = `\n        ${sourceLocDType} sourceLocR = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 1]};\n        ${sourceLocDType} sourceLocG = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 2]};\n        ${sourceLocDType} sourceLocA = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 1]};\n        ${sourceLocDType} sourceLocB = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 2]};`;\n    } else {\n      sourceRank = rank;\n      sourceLocSetup = `\n        ${dtype} sourceLocR = coords;\n        ++${coords[rank - 1]};\n        ${dtype} sourceLocG = coords;\n        ++${coords[rank - 2]};\n        ${dtype} sourceLocA = coords;\n        --${coords[rank - 1]};\n        ${dtype} sourceLocB = coords;\n        --${coords[rank - 2]};`;\n    }\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, sourceRank);\n    const inChannel = '.' + channels[sourceRank - 1];  // e.g. \".b\" for rank 3.\n    const intChannels = channels.map(x => 'int ' + x);\n    const srcRCoords =\n        getChannels('sourceLocR', sourceRank - 1).concat('inIdx.r');\n    const srcGCoords =\n        getChannels('sourceLocG', sourceRank - 1).concat('inIdx.g');\n    const srcBCoords =\n        getChannels('sourceLocB', sourceRank - 1).concat('inIdx.b');\n    const srcACoords =\n        getChannels('sourceLocA', sourceRank - 1).concat('inIdx.a');\n\n    const compOp = (op === 'max') ? 'greaterThan' : 'lessThan';\n    const fetchCandidateIdx = firstPass ? '' : `\n          inIdx = round(vec4(getBestIndicesAChannel(${srcRCoords.join()}),\n                             getBestIndicesAChannel(${srcGCoords.join()}),\n                             getBestIndicesAChannel(${srcBCoords.join()}),\n                             getBestIndicesAChannel(${srcACoords.join()})));`;\n\n    const fetchValue = `vec4(\n            getAChannel(${srcRCoords.join()}),\n            hasNextCol ? getAChannel(${srcGCoords.join()}) : 0.,\n            hasNextRow ? getAChannel(${srcBCoords.join()}) : 0.,\n            hasNextRow && hasNextCol ? getAChannel(${srcACoords.join()}) : 0.)`;\n\n    const getBestIndicesAChannelSnippet = firstPass ? '' : `\n      float getBestIndicesAChannel(${intChannels.join()}) {\n        return getChannel(getBestIndicesA(${channels.join()}),\n                                          vec2(${channels.slice(-2).join()}));\n      }`;\n\n    this.userCode = `\n      float getAChannel(${intChannels.join()}) {\n        return getChannel(getA(${channels.join()}),\n                               vec2(${channels.slice(-2).join()}));\n      }\n      ${getBestIndicesAChannelSnippet}\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        bool hasNextCol = ${coords[rank - 1]} < ${outShape[rank - 1] - 1};\n        bool hasNextRow = ${coords[rank - 2]} < ${outShape[rank - 2] - 1};\n        ${sourceLocSetup}\n        ivec4 srcIdx = ivec4(sourceLocR${inChannel}, sourceLocG${inChannel},\n          sourceLocB${inChannel}, sourceLocA${inChannel}) * ${windowSize};\n        ivec4 inIdx = srcIdx;\n        vec4 bestIndex = vec4(inIdx);\n        vec4 bestValue = ${fetchValue};\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          inIdx = srcIdx;\n          ${fetchCandidateIdx}\n          vec4 candidate = ${fetchValue};\n          bvec4 nan = isnan(candidate);\n          bvec4 replace = bvec4(\n            vec4(${compOp}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));\n\n          bestValue = vec4(replace.x  ? candidate.x : bestValue.x,\n                           replace.y  ? candidate.y : bestValue.y,\n                           replace.z  ? candidate.z : bestValue.z,\n                           replace.w  ? candidate.w : bestValue.w);\n          bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));\n          srcIdx++;\n        }\n        setOutput(bestIndex);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AvgPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth};\n            wC+= ${dilationWidth}) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n\n            dotProd += dyValue * avgMultiplier;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class AvgPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n\n              dotProd += dyValue * avgMultiplier;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\n// (Ar + Ai)(Br + Bi) =\n// ArBr + ArBi + AiBr + AiBi = ArBr - AB + ArBi + AiBr\n// Yr = ArBr - AB\n// Yi = ArBi + AiBr\nexport const COMPLEX_MULTIPLY = {\n  REAL: 'return areal * breal - aimag * bimag;',\n  IMAG: 'return areal * bimag + aimag * breal;'\n};\n\nexport class BinaryOpComplexProgram implements GPGPUProgram {\n  variableNames = ['AReal', 'AImag', 'BReal', 'BImag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    this.userCode = `\n      float binaryOpComplex(\n          float areal, float aimag, float breal, float bimag) {\n        ${op}\n      }\n\n      void main() {\n        float areal = getARealAtOutCoords();\n        float aimag = getAImagAtOutCoords();\n        float breal = getBRealAtOutCoords();\n        float bimag = getBImagAtOutCoords();\n        setOutput(binaryOpComplex(areal, aimag, breal, bimag));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nconst CHECK_NAN_SNIPPET = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const ADD = 'return a + b;';\nexport const SUB = 'return a - b;';\nexport const MUL = 'return a * b;';\n\n// We use native integer division to deal with floating point imprecision. Since\n// we implement floor division and glsl implements truncated division, we\n// correct for this by subtracting 1 from result when the result is negative and\n// there is a remainder.\nexport const INT_DIV = `\n  float s = sign(a) * sign(b);\n  int ia = round(a);\n  int ib = round(b);\n  if (ib != 0) {\n    // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n    return float(idiv(ia, ib, s));\n  } else {\n    return NAN;\n  }\n`;\n\nexport const POW = `\nif(a < 0.0 && floor(b) < b){\n  return NAN;\n}\nif (b == 0.0) {\n  return 1.0;\n}\nreturn (round(mod(b, 2.0)) != 1) ?\n    pow(abs(a), b) : sign(a) * pow(abs(a), b);\n`;\nexport const SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const EQUAL = `return float(a == b);`;\n\nexport const NOT_EQUAL = `return float(a != b);`;\n\nexport const LESS = `return float(a < b);`;\n\nexport const LESS_EQUAL = `return float(a <= b);`;\n\nexport const GREATER = `return float(a > b);`;\n\nexport const GREATER_EQUAL = `return float(a >= b);`;\n\nexport const LOGICAL_AND = `return float(a >= 1.0 && b >= 1.0);`;\n\nexport const LOGICAL_OR = `return float(a >= 1.0 || b >= 1.0);`;\n\nexport const MAX = CHECK_NAN_SNIPPET + `\n  return max(a, b);\n`;\nexport const MIN = CHECK_NAN_SNIPPET + `\n  return min(a, b);\n`;\nexport const MOD = `if (b == 0.0) return NAN;\n  return mod(a, b);`;\n\nexport const ELU_DER = `return (b >= 1.0) ? a : a * (b + 1.0);`;\n\nexport const PRELU = `return (a < 0.) ? b * a : a;`;\n\nexport class BinaryOpProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    this.userCode = `\n      float binaryOperation(float a, float b) {\n        ${op}\n      }\n\n      void main() {\n        float a = getAAtOutCoords();\n        float b = getBAtOutCoords();\n        setOutput(binaryOperation(a, b));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nconst CHECK_NAN_SNIPPET = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\nexport const INT_DIV = `\n  ivec4 ia = round(a);\n  ivec4 ib = round(b);\n  bvec4 cond = notEqual(ib, ivec4(0));\n  ivec4 result = ivec4(0);\n  vec4 s = sign(a) * sign(b);\n\n  // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n  if (cond[0]) {\n    result[0] = idiv(ia[0], ib[0], s[0]);\n  }\n  if (cond[1]) {\n    result[1] = idiv(ia[1], ib[1], s[1]);\n  }\n  if (cond[2]) {\n    result[2] = idiv(ia[2], ib[2], s[2]);\n  }\n  if (cond[3]) {\n    result[3] = idiv(ia[3], ib[3], s[3]);\n  }\n  return vec4(result);\n`;\n\nexport const POW = `\n  // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.\n  vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));\n  vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);\n  vec4 result = multiplier * pow(abs(a), b);\n\n  // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS\n  bvec4 isExpZero = equal(b, vec4(0.0));\n  result.r = isExpZero.r ? 1.0 : result.r;\n  result.g = isExpZero.g ? 1.0 : result.g;\n  result.b = isExpZero.b ? 1.0 : result.b;\n  result.a = isExpZero.a ? 1.0 : result.a;\n\n  vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const PRELU = `\n  vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));\n  return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);\n`;\n\nexport const ELU_DER = `\n  vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));\n  return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));\n`;\n\nexport const EQUAL = `\n  return vec4(equal(a, b));\n`;\n\nexport const NOT_EQUAL = `\n  return vec4(notEqual(a, b));\n`;\n\nexport const LESS = `\n  return vec4(lessThan(a, b));\n`;\n\nexport const LESS_EQUAL = `\n  return vec4(lessThanEqual(a, b));\n`;\n\nexport const GREATER = `\n  return vec4(greaterThan(a, b));\n`;\n\nexport const GREATER_EQUAL = `\n  return vec4(greaterThanEqual(a, b));\n`;\n\nexport const LOGICAL_AND = `\n  return vec4(\n    vec4(greaterThanEqual(a, vec4(1.0))) *\n    vec4(greaterThanEqual(b, vec4(1.0))));\n`;\n\nexport const LOGICAL_OR = `\n  return min(\n    vec4(greaterThanEqual(a, vec4(1.0))) +\n    vec4(greaterThanEqual(b, vec4(1.0))),\n    vec4(1.0));\n`;\n\nexport const MAX = `\n  vec4 result = vec4(max(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MIN = `\n  vec4 result = vec4(min(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MOD = `\n  vec4 result = mod(a, b);\n  vec4 isNaN = vec4(equal(b, vec4(0.0)));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport class BinaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n  supportsBroadcasting = true;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      op: string, aShape: number[], bShape: number[],\n      checkOutOfBounds = false) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const rank = this.outputShape.length;\n    let checkOutOfBoundsString = '';\n    if (checkOutOfBounds) {\n      if (rank === 0 || util.sizeFromShape(this.outputShape) === 1) {\n        checkOutOfBoundsString = `\n          result.y = 0.;\n          result.z = 0.;\n          result.w = 0.;\n        `;\n      } else {\n        const dtype = getCoordsDataType(rank);\n        checkOutOfBoundsString = `\n          ${dtype} coords = getOutputCoords();\n        `;\n        if (rank === 1) {\n          checkOutOfBoundsString += `\n            result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;\n            result.z = 0.;\n            result.w = 0.;\n          `;\n        } else {\n          const channels = getChannels('coords', rank);\n          checkOutOfBoundsString += `\n            bool nextRowOutOfBounds =\n              (${channels[rank - 2]} + 1) >= ${this.outputShape[rank - 2]};\n            bool nextColOutOfBounds =\n              (${channels[rank - 1]} + 1) >= ${this.outputShape[rank - 1]};\n            result.y = nextColOutOfBounds ? 0. : result.y;\n            result.z = nextRowOutOfBounds ? 0. : result.z;\n            result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;\n          `;\n        }\n      }\n    }\n\n    this.userCode = `\n      vec4 binaryOperation(vec4 a, vec4 b) {\n        ${op}\n      }\n\n      void main() {\n        vec4 a = getAAtOutCoords();\n        vec4 b = getBAtOutCoords();\n\n        vec4 result = binaryOperation(a, b);\n        ${checkOutOfBoundsString}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        float value = getAAtOutCoords();\n        if (isnan(value)) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, minVal, maxVal));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        vec4 value = getAAtOutCoords();\n\n        if (any(isnan(value))) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, vec4(minVal), vec4(maxVal)));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ComplexAbsProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(shape: number[]) {\n    this.outputShape = shape;\n    this.userCode = `\n      void main() {\n        float re = abs(getRealAtOutCoords());\n        float im = abs(getImagAtOutCoords());\n        float mx = max(re, im);\n\n        // sadly the length function in glsl is not underflow-safe\n        // (at least not on Intel GPUs). So the safe solution is\n        // to ensure underflow-safety in all cases.\n        setOutput(\n          mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))\n        );\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ConcatProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  // Concats 2d tensors along axis=1. See comments in MathBackendWebGL.concat().\n  constructor(shapes: Array<[number, number]>) {\n    this.outputShape = backend_util.computeOutShape(shapes, 1 /* axis */);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][1];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][1];\n    }\n\n    const snippets = [`if (yC < ${offsets[0]}) setOutput(getT0(yR, yC));`];\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      snippets.push(\n          `else if (yC < ${offsets[i]}) ` +\n          `setOutput(getT${i}(yR, yC-${shift}));`);\n    }\n    const lastIndex = offsets.length;\n    const lastShift = offsets[offsets.length - 1];\n    snippets.push(`else setOutput(getT${lastIndex}(yR, yC-${lastShift}));`);\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int yR = coords.x;\n        int yC = coords.y;\n\n        ${snippets.join('\\n        ')}\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ConcatPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(shapes: number[][], axis: number) {\n    this.outputShape = backend_util.computeOutShape(shapes, axis);\n    const shape = this.outputShape;\n    const rank = shape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][axis];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][axis];\n    }\n\n    const channel = channels[axis];\n    const lastChannels = channels.slice(-2);\n    const allChannels = channels.join();\n\n    let getValueSnippet = `if (${channel} < ${offsets[0]}) {\n        return getChannel(\n            getT0(${allChannels}), vec2(${lastChannels.join()}));\n        }`;\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      // Note: the >= comparison below may seem unnecessary given the check\n      // above but is needed to workaround branch execution issues on some\n      // devices. It makes all the conditions exclusive without relying on\n      // execution order.\n      getValueSnippet += `\n        if (${channel} < ${offsets[i]}  && ${channel} >= ${offsets[i - 1]}) {\n          return getChannel(\n            getT${i}(${shiftedChannels(channels, channel, shift)}),\n            vec2(${shiftedChannels(lastChannels, channel, shift)}));\n        }`;\n    }\n    const lastIndex = offsets.length;\n    const shift = offsets[offsets.length - 1];\n    getValueSnippet += `\n        return getChannel(\n          getT${lastIndex}(${shiftedChannels(channels, channel, shift)}),\n          vec2(${shiftedChannels(lastChannels, channel, shift)}));`;\n\n    this.userCode = `\n      float getValue(${channels.map(x => 'int ' + x)}) {\n        ${getValueSnippet}\n      }\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        vec4 result = vec4(getValue(${coords}), 0., 0., 0.);\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} + 1;\n        if (${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.g = getValue(${coords});\n        }\n\n        ${coords[rank - 2]} = ${coords[rank - 2]} + 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]}) {\n          result.a = getValue(${coords});\n        }\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} - 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]} &&\n            ${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.b = getValue(${coords});\n        }\n        setOutput(result);\n      }\n    `;\n  }\n}\n\n/**\n * Return an expression for coordinates into a vector where a given channel\n * will be offset by [shift].\n *\n * @param channels the channels to consider\n * @param channel the channel we want shifted\n * @param shift  the amount to subtract from the channel.\n *\n * @returns a string of the form 'x, y-[shift], z' where any one channel can\n * have the shift applied.\n */\nfunction shiftedChannels(channels: string[], channel: string, shift: number) {\n  const channelIdx = channels.indexOf(channel);\n  const res = channels.map((c, idx) => {\n    if (idx === channelIdx) {\n      return `${c} - ${shift}`;\n    } else {\n      return c;\n    }\n  });\n  return res.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int d2 = coords.w;\n\n        // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              if (${isChannelsLast}) {\n                float dyValue = getDy(b, yR, yC, d2);\n                float xValue = getX(b, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              } else {\n                float dyValue = getDy(b, d2, yR, yC);\n                float xValue = getX(b, d1, xR, xC);\n                dotProd += (xValue * dyValue);\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[${channelDim}];\n\n        ivec2 dyCorner = ivec2(coords[${rowDim}], coords[${colDim}]) - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n\n              if (${isChannelsLast}) {\n                float xValue = getDy(batch, idyR, idyC, d2);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              } else {\n                float xValue = getDy(batch, d2, idyR, idyC);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    this.userCode = `\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int wF = coords.x;\n        int wR = coords.y;\n        int wC = coords.z;\n        int d1 = coords.w;\n        int d2 = coords.u;\n\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yF = 0; yF < ${convInfo.outDepth}; yF++) {\n            int xF = wF + yF * ${strideDepth} - ${padFront};\n\n            if (xF < 0 || xF >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n              int xR = wR + yR * ${strideHeight} - ${padTop};\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n                int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float dyValue = getDy(b, yF, yR, yC, d2);\n                float xValue = getX(b, xF, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padFront = filterDepth - 1 - convInfo.padInfo.front;\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d1 = coords.u;\n\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyFCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          float dyF = float(dyFCorner + wF) / ${strideDepth}.0;\n\n          if (dyF < 0.0 || dyF >= ${convInfo.outDepth}.0 || fract(dyF) > 0.0) {\n            continue;\n          }\n          int idyF = int(dyF);\n\n          int wFPerm = ${filterDepth} - 1 - wF;\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n              fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            int wRPerm = ${filterHeight} - 1 - wR;\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              int wCPerm = ${filterWidth} - 1 - wC;\n\n              for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n                float xValue = getDy(batch, idyF, idyR, idyC, d2);\n                float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int dm = coords.w;\n        int d2 = d1 * ${channelMul} + dm;\n\n        float dotProd = 0.0;\n\n        // TO DO: Vec4 over the batch size\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float dyValue = getDy(b, yR, yC, d2);\n              float xValue = getX(b, xR, xC, d1);\n              dotProd += (xValue * dyValue);\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class DepthwiseConv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[3];\n        ivec2 dyCorner = coords.yz - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        float dotProd = 0.0;\n\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            // TO DO: Vec4 over the channelMul\n            for (int dm = 0; dm < ${channelMul}; dm++) {\n              int d2 = d1 * ${channelMul} + dm;\n              float xValue = getDy(batch, idyR, idyC, d2);\n              float wValue = getW(wRPerm, wCPerm, d1, dm);\n              dotProd += xValue * wValue;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivationWeights = false) {\n    this.outputShape = convInfo.outShape;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivationWeights) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivationWeights) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d2 = coords[${channelDim}];\n\n        ivec2 xRCCorner =\n            ivec2(coords[${rowDim}], coords[${colDim}]) * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${convInfo.inWidth}) {\n              continue;\n            }\n\n            for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n              vec4 wValues = vec4(\n                getW(wR, wC, d1, d2),\n                getW(wR, wC, d1 + 1, d2),\n                getW(wR, wC, d1 + 2, d2),\n                getW(wR, wC, d1 + 3, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec4 xValues = vec4(\n                  getX(batch, xR, xC, d1),\n                  getX(batch, xR, xC, d1 + 1),\n                  getX(batch, xR, xC, d1 + 2),\n                  getX(batch, xR, xC, d1 + 3)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec4 xValues = vec4(\n                  getX(batch, d1, xR, xC),\n                  getX(batch, d1 + 1, xR, xC),\n                  getX(batch, d1 + 2, xR, xC),\n                  getX(batch, d1 + 3, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n\n            if (${inputDepthVec4Remainder === 1}) {\n\n              if (${isChannelsLast}) {\n                dotProd +=\n                    getX(batch, xR, xC, ${inputDepthNearestVec4}) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              } else {\n                dotProd +=\n                    getX(batch, ${inputDepthNearestVec4}, xR, xC) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              }\n\n            } else if (${inputDepthVec4Remainder === 2}) {\n              vec2 wValues = vec2(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec2 xValues = vec2(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec2 xValues = vec2(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            } else if (${inputDepthVec4Remainder === 3}) {\n              vec3 wValues = vec3(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 2, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec3 xValues = vec3(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec3 xValues = vec3(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 2, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            }\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nexport class Conv3DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.outShape;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n\n    this.userCode = `\n      const ivec3 strides = ivec3(${strideDepth}, ${strideHeight}, ${\n        strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d2 = coords.u;\n\n        ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xFCorner = xFRCCorner.x;\n        int xRCorner = xFRCCorner.y;\n        int xCCorner = xFRCCorner.z;\n\n        // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get\n        // y(yF, yR, yC, d2). ? = to be determined. : = across all\n        // values in that axis.\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          int xF = xFCorner + wF * ${dilationDepth};\n\n          if (xF < 0 || xF >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            int xR = xRCorner + wR * ${dilationHeight};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n                vec4 xValues = vec4(\n                  getX(batch, xF, xR, xC, d1),\n                  getX(batch, xF, xR, xC, d1 + 1),\n                  getX(batch, xF, xR, xC, d1 + 2),\n                  getX(batch, xF, xR, xC, d1 + 3)\n                );\n                vec4 wValues = vec4(\n                  getW(wF, wR, wC, d1, d2),\n                  getW(wF, wR, wC, d1 + 1, d2),\n                  getW(wF, wR, wC, d1 + 2, d2),\n                  getW(wF, wR, wC, d1 + 3, d2)\n                );\n\n                dotProd += dot(xValues, wValues);\n              }\n\n              if (${inputDepthVec4Remainder === 1}) {\n                dotProd +=\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}) *\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2);\n              } else if (${inputDepthVec4Remainder === 2}) {\n                vec2 xValues = vec2(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                vec2 wValues = vec2(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else if (${inputDepthVec4Remainder === 3}) {\n                vec3 xValues = vec3(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                vec3 wValues = vec3(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 2, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2 / ${channelMul};\n        int q = d2 - d1 * ${channelMul};\n\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${xNumRows}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${xNumCols}) {\n              continue;\n            }\n\n            float xVal = getX(batch, xR, xC, d1);\n            float wVal = getW(wR, wC, d1, q);\n            dotProd += xVal * wVal;\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConvPacked2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const texelsAcross = filterWidth;\n\n    let mainLoop = `int xR; int xC; int xCOffset;`;\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `\n          vec4 xTexelR${r}C${c * 2} = vec4(0.);\n          vec4 wR${r}C${c} = vec4(0.);\n          vec4 xR${r}C${c} = vec4(0.);`;\n      }\n    }\n\n    /**\n     * This vectorized implementation works by gathering the values needed for\n     * each output channel's dot product into vec4's and then multiplying them\n     * all together (this happens in the final double for-loop below). Most of\n     * the main loop consists of constructing these vec4's with the minimum\n     * number of texture2D calls, which means making use of all four returned\n     * values from a texture2D call at once.\n     */\n    for (let r = 0; r < filterHeight; r++) {\n      for (let texelC = 0; texelC < texelsAcross; texelC++) {\n        const c = texelC * 2;\n\n        mainLoop += `\n          xR = xRCorner + ${r * dilationHeight};\n          xC = xCCorner + ${c * dilationWidth};\n        `;\n\n        if (strideWidth === 1) {\n          if (c < filterWidth) {\n            // If padding is odd, the outer texels have to be composed.\n            if (padLeft % 2 === 1) {\n              // TODO: Ensure vec4 previous does not result in redundant sample,\n              // and avoid setting xTexelRC's that exceed the boundary in the\n              // first place rather than resetting them to vec4(0)).\n\n              // To compute xCOffset:\n              // - If padding is odd, we must add 1 to ensure we ask for an\n              // even-numbered row.\n              // - We subtract 2 to access the previous texel.\n\n              mainLoop += `\n                xCOffset = xC + 1;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    xTexelR${r}C${c}.zw = vec2(0.);\n                  }\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + 1 - 2;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  vec4 previous = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    previous.zw = vec2(0.);\n                  }\n\n                  xR${r}C${c} = vec4(previous.zw, xTexelR${r}C${c}.xy);\n                } else {\n                  xR${r}C${c} = vec4(0, 0, xTexelR${r}C${c}.xy);\n                }\n              `;\n            } else {\n              // Padding is even, so xRC corresponds to a single texel.\n              mainLoop += `\n                if(xR >= 0 && xR < ${xNumRows} && xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xR${r}C${c} = xTexelR${r}C${c};\n              `;\n            }\n\n            if (c + 1 < filterWidth) {\n              // If dilation is even, the second entry should match the first\n              // (either both are composed or both are single samples). But if\n              // dilation is odd, then the second entry should be the opposite\n              // of the first (if the first is composed, the second is a single\n              // sample, and vice versa.)\n\n              const nextTexelOffset = padLeft % 2 === 0 ?\n                  util.nearestLargerEven(dilationWidth) :\n                  dilationWidth;\n\n              if ((dilationWidth % 2 === 0 && padLeft % 2 === 1) ||\n                  (dilationWidth % 2 !== 0 && padLeft % 2 !== 1)) {\n                mainLoop += `\n                  xCOffset = xC + ${padLeft % 2} + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n                `;\n\n                // If dilation > 1 then the xRC's will not be able to share any\n                // values, so each xRC will require two unique calls to getX.\n                if (dilationWidth > 1) {\n                  mainLoop += `\n                    xCOffset -= 2;\n                    if(xR >= 0 && xR < ${xNumRows} &&\n                      xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                      xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                    } else {\n                      xTexelR${r}C${c} = vec4(0.);\n                    }\n                  `;\n                }\n\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.xy);\n                `;\n              } else {\n                mainLoop += `\n                  xCOffset = xC + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n\n                  xR${r}C${c + 1} = xTexelR${r}C${c + 2};\n                `;\n              }\n            }\n          }\n        } else {  // stride > 1\n          if (c < filterWidth) {\n            mainLoop += `\n              if(xR >= 0 && xR < ${xNumRows}) {\n            `;\n\n            // Depending on whether padLeft is even or odd, we want either the\n            // xy or zw channels from X texels for xR${r}C${c}. If padLeft is\n            // even, xR${r}C${c + 1} is simply the zw channels of texels we've\n            // already sampled. But if padLeft is odd, xR${r}C{$c + 1}.zw will\n            // need to come from the xy channels of a new texel, hence the `vec4\n            // final` initialized below.\n            if (padLeft % 2 === 1) {\n              mainLoop += `\n                xCOffset = xC + 1 - ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                if(xC + 1 >= 0 && xC + 1 < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xC + 1, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  vec4 final = vec4(0.);\n                  xCOffset = xC + 1 + ${strideWidth};\n                  if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    final = getX(batch, xR, xCOffset, d1);\n                  }\n                  xR${r}C${c + 1} = vec4(xTexelR${r}C${c + 2}.xy, final.xy);\n                `;\n              }\n            } else {\n              mainLoop += `\n                if(xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.xy, xTexelR${r}C${c + 2}.xy);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n                `;\n              }\n            }\n\n            mainLoop += `}`;\n          }\n        }\n\n        if (c < filterWidth) {\n          mainLoop += `\n            vec4 wTexelR${r}C${c} = getW(${r}, ${c}, d1, q);\n            wR${r}C${c} = vec4(wTexelR${r}C${c}.xz, wTexelR${r}C${c}.xz);\n          `;\n\n          if (c + 1 < filterWidth) {\n            mainLoop += `\n              vec4 wTexelR${r}C${c + 1} = getW(${r}, ${c + 1}, d1, q);\n              wR${r}C${c + 1} =\n                vec4(wTexelR${r}C${c + 1}.xz, wTexelR${r}C${c + 1}.xz);`;\n          }\n        }\n      }\n    }\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `dotProd += xR${r}C${c} * wR${r}C${c};`;\n      }\n    }\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2;\n        int q = 0;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        vec4 dotProd = vec4(0.);\n\n        ${mainLoop}\n\n        vec4 result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class CropAndResizeProgram implements GPGPUProgram {\n  variableNames = ['Image', 'Boxes', 'BoxInd'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], boxShape: [number, number],\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number) {\n    const [batch, imageHeight, imageWidth, depth] = imageShape;\n    const [numBoxes, ] = boxShape;\n    const [cropHeight, cropWidth] = cropSize;\n    this.outputShape = [numBoxes, cropHeight, cropWidth, depth];\n    const methodId = method === 'bilinear' ? 1 : 0;\n\n    const [inputHeightFloat, inputWidthFloat] =\n        [`${imageHeight - 1}.0`, `${imageWidth - 1}.0`];\n\n    const [heightRatio, heightScale, inY] = cropHeight > 1 ?\n        [\n          `${(imageHeight - 1) / (cropHeight - 1)}`,\n          '(y2-y1) * height_ratio',\n          `y1*${inputHeightFloat} + float(y)*(height_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (y1+y2) * ${inputHeightFloat}`,\n        ];\n    const [widthRatio, widthScale, inX] = cropWidth > 1 ?\n        [\n          `${(imageWidth - 1) / (cropWidth - 1)}`,\n          '(x2-x1) * width_ratio',\n          `x1*${inputWidthFloat} + float(x)*(width_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (x1+x2) * ${inputWidthFloat}`,\n        ];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc\n    this.userCode = `\n      const float height_ratio = float(${heightRatio});\n      const float width_ratio = float(${widthRatio});\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int y = coords[1];\n        int x = coords[2];\n        int d = coords[3];\n\n        // get box vals\n        float y1 = getBoxes(b,0);\n        float x1 = getBoxes(b,1);\n        float y2 = getBoxes(b,2);\n        float x2 = getBoxes(b,3);\n\n        // get image in batch index\n        int bInd = round(getBoxInd(b));\n        if(bInd < 0 || bInd >= ${batch}) {\n          return;\n        }\n\n        float height_scale = ${heightScale};\n        float width_scale = ${widthScale};\n\n        float in_y = ${inY};\n        if( in_y < 0.0 || in_y > ${inputHeightFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n        float in_x = ${inX};\n        if( in_x < 0.0 || in_x > ${inputWidthFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n\n        vec2 sourceFracIndexCR = vec2(in_x,in_y);\n        if(${methodId} == 1) {\n          // Compute the four integer indices.\n          ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);\n          ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));\n\n          float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);\n          float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);\n          float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);\n          float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);\n\n          vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);\n\n          float top = topLeft + (topRight - topLeft) * fracCR.x;\n          float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;\n          float newValue = top + (bottom - top) * fracCR.y;\n          setOutput(newValue);\n        } else {\n          // Compute the coordinators of nearest neighbor point.\n          ivec2 sourceNearestCR = ivec2(floor(\n            sourceFracIndexCR + vec2(0.5,0.5)));\n          float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);\n          setOutput(newValue);\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class CumSumProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  index: WebGLUniformLocation;\n\n  constructor(shape: number[], exclusive: boolean, reverse: boolean) {\n    this.outputShape = shape;\n    const rank = shape.length;\n    const val = exclusive ? '0.0' : `getX(${getCoords(rank, 'coords')})`;\n    const length = shape[shape.length - 1];\n    let condition = '';\n    let idxString = '';\n    // When exclusive is set, the cumsum op becomes roll op that copies the\n    // value from the previous index based on the direction specified by the\n    // reverse flag.\n    if (exclusive) {\n      condition = reverse ? `end != ${length - 1}` : 'end != 0';\n      idxString = reverse ? 'end + 1' : 'end - 1';\n    } else {\n      condition = reverse ? `end + pow2 < ${length}` : 'end >= pow2';\n      idxString = (reverse ? 'end + pow2' : 'end - pow2');\n    }\n\n    this.userCode = `\n      uniform float index;\n      void main() {\n        ${getCoordsDataType(rank)} coords = getOutputCoords();\n        int end = ${getFinalCoord(rank, 'coords')};\n        float val = ${val};\n        int pow2 = int(pow(2.0, index));\n        if (${condition}) {\n          int idx = ${idxString};\n          ${getFinalCoord(rank, 'coords')} = idx;\n          val += getX(${getCoords(rank, 'coords')});\n        }\n        setOutput(val);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(index: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.index == null) {\n        this.index = gpgpu.getUniformLocation(webGLProgram, 'index');\n      }\n      gpgpu.gl.uniform1f(this.index, index);\n    };\n  }\n}\n\nfunction getCoords(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.x, ${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.x, ${name}.y, ${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.x, ${name}.y, ${name}.z, ${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n\nfunction getFinalCoord(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: [number, number, number];\n  packedInputs = false;\n  packedOutput = true;\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getA(rc.x, rc.y, rc.z);\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: [number, number, number];\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthToSpaceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  blockSize: number;\n  dataFormat: string;\n\n  constructor(\n      outputShape: number[], blockSize: number, dataFormat: 'NHWC'|'NCHW') {\n    this.outputShape = outputShape;\n    this.blockSize = blockSize;\n    this.dataFormat = dataFormat;\n    this.userCode = `\n    void main() {\n      ivec4 coords = getOutputCoords();\n      int b = coords[0];\n      int h = ${this.getHeightCoordString()};\n      int w = ${this.getWidthCoordString()};\n      int d = ${this.getDepthCoordString()};\n\n      int in_h = h / ${blockSize};\n      int offset_h = imod(h, ${blockSize});\n      int in_w = w / ${blockSize};\n      int offset_w = imod(w, ${blockSize});\n      int offset_d = (offset_h * ${blockSize} + offset_w) *\n        ${this.getOutputDepthSize()};\n      int in_d = d + offset_d;\n\n      float result = ${this.getInputSamplingString()};\n      setOutput(result);\n    }\n  `;\n  }\n\n  private getHeightCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[1]`;\n    } else {\n      return `coords[2]`;\n    }\n  }\n\n  private getWidthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[2]`;\n    } else {\n      return `coords[3]`;\n    }\n  }\n\n  private getDepthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[3]`;\n    } else {\n      return `coords[1]`;\n    }\n  }\n\n  private getOutputDepthSize(): number {\n    if (this.dataFormat === 'NHWC') {\n      return this.outputShape[3];\n    } else {\n      return this.outputShape[1];\n    }\n  }\n\n  private getInputSamplingString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `getX(b, in_h, in_w, in_d)`;\n    } else {\n      return `getX(b, in_d, in_h, in_w)`;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DiagProgram implements GPGPUProgram {\n  variableNames = ['X'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(size: number) {\n    this.outputShape = [size, size];\n    this.userCode = `\n      void main() {\n          ivec2 coords = getOutputCoords();\n          float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;\n          setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        float x = getAAtOutCoords();\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = false;\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: [number, number, number]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n        float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class EncodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let output = `result`;\n    if (inputIsUnsignedByte) {\n      output = `floor(result * 255. + 0.5)`;\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        int flatIndex = getFlatIndex(coords);\n        int offset = imod(flatIndex, 4);\n\n        flatIndex = idiv(flatIndex, 4, 1.);\n\n        int r = flatIndex / ${width};\n        int c = imod(flatIndex, ${width});\n        vec2 uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n        vec4 values = ${glsl.texture2D}(A, uv);\n\n        float result;\n\n        if(offset == 0) {\n          result = values[0];\n        } else if(offset == 1) {\n          result = values[1];\n        } else if(offset == 2) {\n          result = values[2];\n        } else {\n          result = values[3];\n        }\n\n        ${glsl.output} = vec4(${output}, 0., 0., 0.);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\n/*\nThis is how the shader encodes a tensor with shape = [2, 3, 5]\n(indices are [batch, row, col]).\n\n000|001   002|003   004|xxx   020|021   022|023   024|xxx\n-------   -------   -------   -------   -------   -------\n010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\n100|101   102|103   104|xxx   120|121   122|123   124|xxx\n-------   -------   -------   -------   -------   -------\n110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\nSingle texels contain only values from the same batch, and from adjacent rows\nand columns.\n */\n\nexport class EncodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let mainLoop = '';\n    let output = 'result';\n    if (inputIsUnsignedByte) {\n      output = 'floor(result * 255. + 0.5)';\n    }\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        const channel = row * 2 + col;\n\n        mainLoop += `\n          localCoords = coords;\n          if(localCoords[2] + ${col} < ${outputShape[2]}) {\n            localCoords[2] += ${col};\n            if(localCoords[1] + ${row} < ${outputShape[1]}) {\n              localCoords[1] += ${row};\n\n              flatIndex = getFlatIndex(localCoords);\n              offset = imod(flatIndex, 4);\n\n              flatIndex = idiv(flatIndex, 4, 1.);\n\n              r = flatIndex / ${width};\n              c = imod(flatIndex, ${width});\n              uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n              values = ${glsl.texture2D}(A, uv);\n\n              if(offset == 0) {\n                result[${channel}] = values[0];\n              } else if(offset == 1) {\n                result[${channel}] = values[1];\n              } else if(offset == 2) {\n                result[${channel}] = values[2];\n              } else {\n                result[${channel}] = values[3];\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        vec4 result = vec4(0.);\n        int flatIndex, r, c, offset;\n        ivec3 localCoords;\n        vec2 uv;\n        vec4 values;\n\n        ${mainLoop}\n\n        ${glsl.output} = ${output};\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const COMPLEX_FFT = {\n  REAL: 'return real * expR - imag * expI;',\n  IMAG: 'return real * expI + imag * expR;'\n};\n\nexport class FFTProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, inputShape: [number, number], inverse: boolean) {\n    const innerDim = inputShape[1];\n    this.outputShape = inputShape;\n\n    const exponentMultiplierSnippet =\n        inverse ? `2.0 * ${Math.PI}` : `-2.0 * ${Math.PI}`;\n    const resultDenominator = inverse ? `${innerDim}.0` : '1.0';\n\n    this.userCode = `\n      const float exponentMultiplier = ${exponentMultiplierSnippet};\n\n      float unaryOpComplex(float real, float expR, float imag, float expI) {\n        ${op}\n      }\n\n      float mulMatDFT(int batch, int index) {\n        float indexRatio = float(index) / float(${innerDim});\n        float exponentMultiplierTimesIndexRatio =\n            exponentMultiplier * indexRatio;\n\n        float result = 0.0;\n\n        for (int i = 0; i < ${innerDim}; i++) {\n          // x = (-2|2 * PI / N) * index * i;\n          float x = exponentMultiplierTimesIndexRatio * float(i);\n          float expR = cos(x);\n          float expI = sin(x);\n          float real = getReal(batch, i);\n          float imag = getImag(batch, i);\n\n          result +=\n              unaryOpComplex(real, expR, imag, expI) / ${resultDenominator};\n        }\n\n        return result;\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        setOutput(mulMatDFT(coords[0], coords[1]));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FillProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  valueLoc: WebGLUniformLocation;\n\n  constructor(shape: number[], value: number) {\n    this.variableNames = ['x'];\n    this.outputShape = shape;\n\n    this.userCode = `\n      uniform float value;\n      void main() {\n        // Input can be obtained from uniform value.\n        setOutput(value);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(value: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.valueLoc == null) {\n        this.valueLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'value');\n      }\n      gpgpu.gl.uniform1f(this.valueLoc, value);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherProgram implements GPGPUProgram {\n  variableNames = ['A', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], indicesLength: number, axis: number) {\n    const outputShape: number[] = aShape.slice();\n    outputShape[axis] = indicesLength;\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape, axis);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[], axis: number): string {\n  const rank = aShape.length;\n  if (rank > 4) {\n    throw Error(`Gather for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `int(getIndices(resRC))`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    if (i === axis) {\n      sourceCoords.push(`int(getIndices(${currentCoords[i]}))`);\n    } else {\n      sourceCoords.push(`${currentCoords[i]}`);\n    }\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherNDProgram implements GPGPUProgram {\n  variableNames = ['x', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  constructor(\n      private sliceDim: number, private strides: number[], shape: number[]) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    const strideString = this.sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${this.strides});\n         void main() {\n          ${dtype} coords = getOutputCoords();\n          int flattenIndex = 0;\n          for (int j = 0; j < ${this.sliceDim}; j++) {\n            int index = round(getIndices(coords[0], j));\n            flattenIndex += index * ${strideString};\n          }\n          setOutput(getX(flattenIndex, coords[1]));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PixelData, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {getGlslDifferences} from './glsl_version';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport * as webgl_util from './webgl_util';\n\nexport function createVertexShader(gl: WebGLRenderingContext): WebGLShader {\n  const glsl = getGlslDifferences();\n  const vertexShaderSource = `${glsl.version}\n    precision highp float;\n    ${glsl.attribute} vec3 clipSpacePos;\n    ${glsl.attribute} vec2 uv;\n    ${glsl.varyingVs} vec2 resultUV;\n\n    void main() {\n      gl_Position = vec4(clipSpacePos, 1);\n      resultUV = uv;\n    }`;\n  return webgl_util.createVertexShader(gl, vertexShaderSource);\n}\n\nexport function createVertexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // [x y z u v] * [upper-left, lower-left, upper-right, lower-right]\n  const vertexArray = new Float32Array(\n      [-1, 1, 0, 0, 1, -1, -1, 0, 0, 0, 1, 1, 0, 1, 1, 1, -1, 0, 1, 0]);\n  return webgl_util.createStaticVertexBuffer(gl, vertexArray);\n}\n\nexport function createIndexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // OpenGL (and WebGL) have \"CCW == front\" winding\n  const triangleVertexIndices = new Uint16Array([0, 1, 2, 2, 1, 3]);\n  return webgl_util.createStaticIndexBuffer(gl, triangleVertexIndices);\n}\n\nfunction createAndConfigureTexture(\n    gl: WebGLRenderingContext, width: number, height: number,\n    internalFormat: number, textureFormat: number,\n    textureType: number): WebGLTexture {\n  webgl_util.validateTextureSize(width, height);\n  const texture = webgl_util.createTexture(gl);\n\n  const tex2d = gl.TEXTURE_2D;\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(tex2d, texture));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MIN_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MAG_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          tex2d, 0, internalFormat, width, height, 0, textureFormat,\n          textureType, null));\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n  return texture;\n}\n\nexport function getInternalFormatForFloat32MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatFloat;\n}\n\nexport function createFloat32MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat32MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatHalfFloat;\n}\n\nexport function createFloat16MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, textureConfig.textureTypeHalfFloat);\n}\n\nexport function getInternalFormatForUnsignedBytesMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.downloadTextureFormat;\n}\n\nexport function createUnsignedBytesMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForUnsignedBytesMatrixTexture(textureConfig), gl.RGBA,\n      gl.UNSIGNED_BYTE);\n}\n\nexport function getInternalFormatForPackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedFloat;\n}\n\nexport function createPackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height, getInternalFormatForPackedMatrixTexture(textureConfig),\n      gl.RGBA, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16PackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedHalfFloat;\n}\n\nexport function createFloat16PackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16PackedMatrixTexture(textureConfig), gl.RGBA,\n      textureConfig.textureTypeHalfFloat);\n}\n\nexport function bindVertexProgramAttributeStreams(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    vertexBuffer: WebGLBuffer): boolean {\n  const posOffset = 0;               // x is the first buffer element\n  const uvOffset = 3 * 4;            // uv comes after [x y z]\n  const stride = (3 * 4) + (2 * 4);  // xyz + uv, each entry is 4-byte float.\n  webgl_util.callAndCheck(\n      gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, vertexBuffer));\n  const success = webgl_util.bindVertexBufferToProgramAttribute(\n      gl, program, 'clipSpacePos', vertexBuffer, 3, stride, posOffset);\n  return success &&\n      webgl_util.bindVertexBufferToProgramAttribute(\n          gl, program, 'uv', vertexBuffer, 2, stride, uvOffset);\n}\n\nexport function uploadDenseMatrixToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture, width: number,\n    height: number, data: TypedArray, textureConfig: TextureConfig) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n\n  let dataForUpload: TypedArray, texelDataType: number, internalFormat: number;\n  if (data instanceof Uint8Array) {\n    dataForUpload = new Uint8Array(width * height * 4);\n    texelDataType = gl.UNSIGNED_BYTE;\n    internalFormat = gl.RGBA;\n  } else {\n    dataForUpload = new Float32Array(width * height * 4);\n    texelDataType = gl.FLOAT;\n    internalFormat = textureConfig.internalFormatPackedFloat;\n  }\n\n  dataForUpload.set(data);\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          gl.TEXTURE_2D, 0, internalFormat, width, height, 0, gl.RGBA,\n          texelDataType, dataForUpload));\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function uploadPixelDataToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, pixels.width, pixels.height, 0, gl.RGBA,\n            gl.UNSIGNED_BYTE, (pixels as PixelData).data));\n  } else {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, gl.RGBA, gl.UNSIGNED_BYTE,\n            pixels as ImageData | HTMLImageElement | HTMLCanvasElement |\n                HTMLVideoElement));\n  }\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function createBufferFromOutputTexture(\n    gl2: WebGL2RenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLBuffer {\n  // Create and bind the buffer.\n  const buffer = gl2.createBuffer();\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer));\n\n  // Initialize the buffer to the size of the texture in bytes.\n  const bytesPerFloat = 4;\n  const valuesPerTexel = 4;\n  const bufferSizeBytes = bytesPerFloat * valuesPerTexel * rows * columns;\n\n  webgl_util.callAndCheck(\n      gl2,\n      () => gl2.bufferData(\n          gl2.PIXEL_PACK_BUFFER, bufferSizeBytes, gl2.STREAM_READ));\n\n  // Enqueue a command on the GPU command queue to copy of texture into the\n  // buffer.\n  webgl_util.callAndCheck(\n      gl2, () => gl2.readPixels(0, 0, columns, rows, gl2.RGBA, gl2.FLOAT, 0));\n\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null));\n\n  return buffer;\n}\n\nexport function downloadFloat32MatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer,\n    size: number): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget = new Float32Array(size);\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadByteEncodedFloatMatrixFromOutputTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig) {\n  const [w, h] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n\n  const numChannels = 4;\n  const downloadTarget = new Uint8Array(\n      tex_util.getUnpackedArraySizeFromMatrixSize(rows * columns, numChannels));\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, w, h, textureConfig.downloadTextureFormat, gl.UNSIGNED_BYTE,\n          downloadTarget));\n\n  // By wrapping the buffer in a Float32Array, we use native browser IEEE 754\n  // decoding of the 4 bytes that back each 32 bit float.\n  return new Float32Array(downloadTarget.buffer);\n}\n\nexport function downloadPackedMatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer, batch: number, rows: number,\n    cols: number, physicalRows: number, physicalCols: number,\n    textureConfig: TextureConfig): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget =\n      new Float32Array(tex_util.getPackedRGBAArraySizeFromMatrixShape(\n          physicalRows, physicalCols));\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadMatrixFromPackedOutputTexture(\n    gl: WebGLRenderingContext, physicalRows: number,\n    physicalCols: number): Float32Array {\n  const packedRGBA = new Float32Array(physicalRows * physicalCols * 4);\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, physicalCols, physicalRows, gl.RGBA, gl.FLOAT, packedRGBA));\n\n  return packedRGBA;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, PixelData, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext, setWebGLContext} from './canvas_util';\nimport * as gpgpu_util from './gpgpu_util';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport {WebGL1DisjointQueryTimerExtension, WebGL2DisjointQueryTimerExtension} from './webgl_types';\nimport * as webgl_util from './webgl_util';\n\nexport interface FenceContext {\n  query: WebGLQuery|WebGLSync;\n  isFencePassed(): boolean;\n}\n\nexport class GPGPUContext {\n  gl: WebGLRenderingContext;\n  textureFloatExtension: {};\n  textureHalfFloatExtension: {};\n  colorBufferFloatExtension: {};\n  colorBufferHalfFloatExtension: {};\n  disjointQueryTimerExtension: WebGL2DisjointQueryTimerExtension|\n      WebGL1DisjointQueryTimerExtension;\n  vertexBuffer: WebGLBuffer;\n  indexBuffer: WebGLBuffer;\n  framebuffer: WebGLFramebuffer;\n  outputTexture: WebGLTexture|null = null;\n  program: WebGLProgram|null = null;\n  private disposed = false;\n  private disjoint: boolean;\n  textureConfig: TextureConfig;\n\n  constructor(gl?: WebGLRenderingContext) {\n    const glVersion = env().getNumber('WEBGL_VERSION');\n    if (gl != null) {\n      this.gl = gl;\n      setWebGLContext(glVersion, gl);\n    } else {\n      this.gl = getWebGLContext(glVersion);\n    }\n    // WebGL 2.0 enables texture floats without an extension.\n    let COLOR_BUFFER_FLOAT = 'WEBGL_color_buffer_float';\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (env().getNumber('WEBGL_VERSION') === 1) {\n      const TEXTURE_FLOAT = 'OES_texture_float';\n      const TEXTURE_HALF_FLOAT = 'OES_texture_half_float';\n\n      this.textureFloatExtension =\n          webgl_util.getExtensionOrThrow(this.gl, TEXTURE_FLOAT);\n      if (webgl_util.hasExtension(this.gl, TEXTURE_HALF_FLOAT)) {\n        this.textureHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, TEXTURE_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support half float textures, yet the ' +\n            'environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n\n      this.colorBufferFloatExtension = this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, COLOR_BUFFER_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support color renderable half floats, yet ' +\n            'the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n    } else {\n      COLOR_BUFFER_FLOAT = 'EXT_color_buffer_float';\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_FLOAT)) {\n        this.colorBufferFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      } else if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      } else {\n        throw new Error('GL context does not support color renderable floats');\n      }\n    }\n\n    this.vertexBuffer = gpgpu_util.createVertexBuffer(this.gl);\n    this.indexBuffer = gpgpu_util.createIndexBuffer(this.gl);\n    this.framebuffer = webgl_util.createFramebuffer(this.gl);\n\n    this.textureConfig =\n        tex_util.getTextureConfig(this.gl, this.textureHalfFloatExtension);\n  }\n\n  private get debug(): boolean {\n    return env().getBool('DEBUG');\n  }\n\n  public dispose() {\n    if (this.disposed) {\n      return;\n    }\n    if (this.program != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound WebGLProgram.' +\n          ' This is probably a resource leak, delete the program with ' +\n          'GPGPUContext.deleteProgram before disposing.');\n    }\n    if (this.outputTexture != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound output matrix ' +\n          'texture.  This is probably a resource leak, delete the output ' +\n          'matrix texture with GPGPUContext.deleteMatrixTexture before ' +\n          'disposing.');\n    }\n    const gl = this.gl;\n    webgl_util.callAndCheck(gl, () => gl.finish());\n    webgl_util.callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteFramebuffer(this.framebuffer));\n    webgl_util.callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(\n        gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteBuffer(this.indexBuffer));\n    this.disposed = true;\n  }\n\n  public createFloat32MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat32MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createFloat16MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createUnsignedBytesMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createUnsignedBytesMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public uploadPixelDataToTexture(\n      texture: WebGLTexture,\n      pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadPixelDataToTexture(this.gl, texture, pixels);\n  }\n\n  public uploadDenseMatrixToTexture(\n      texture: WebGLTexture, width: number, height: number, data: TypedArray) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadDenseMatrixToTexture(\n        this.gl, texture, width, height, data, this.textureConfig);\n  }\n\n  public createFloat16PackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16PackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createPackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createPackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public deleteMatrixTexture(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    if (this.outputTexture === texture) {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n      this.outputTexture = null;\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.deleteTexture(texture));\n  }\n\n  public downloadByteEncodedFloatMatrixFromOutputTexture(\n      texture: WebGLTexture, rows: number, columns: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadByteEncodedFloatMatrixFromOutputTexture(\n            this.gl, rows, columns, this.textureConfig));\n  }\n\n  public downloadPackedMatrixFromBuffer(\n      buffer: WebGLBuffer, batch: number, rows: number, columns: number,\n      physicalRows: number, physicalCols: number): Float32Array {\n    return gpgpu_util.downloadPackedMatrixFromBuffer(\n        this.gl, buffer, batch, rows, columns, physicalRows, physicalCols,\n        this.textureConfig);\n  }\n\n  public downloadFloat32MatrixFromBuffer(buffer: WebGLBuffer, size: number):\n      Float32Array {\n    return gpgpu_util.downloadFloat32MatrixFromBuffer(this.gl, buffer, size);\n  }\n\n  public createBufferFromTexture(\n      texture: WebGLTexture, rows: number, columns: number): WebGLBuffer {\n    this.bindTextureToFrameBuffer(texture);\n    const result = gpgpu_util.createBufferFromOutputTexture(\n        this.gl as WebGL2RenderingContext, rows, columns, this.textureConfig);\n    this.unbindTextureToFrameBuffer();\n    return result;\n  }\n\n  public createAndWaitForFence(): Promise<void> {\n    const fenceContext = this.createFence(this.gl);\n    return this.pollFence(fenceContext);\n  }\n\n  private createFence(gl: WebGLRenderingContext): FenceContext {\n    let query: WebGLQuery|WebGLSync;\n    let isFencePassed: () => boolean;\n\n    if (env().getBool('WEBGL_FENCE_API_ENABLED')) {\n      const gl2 = gl as WebGL2RenderingContext;\n\n      const sync = gl2.fenceSync(gl2.SYNC_GPU_COMMANDS_COMPLETE, 0);\n      gl.flush();\n\n      isFencePassed = () => {\n        const status = gl2.clientWaitSync(sync, 0, 0);\n        return status === gl2.ALREADY_SIGNALED ||\n            status === gl2.CONDITION_SATISFIED;\n      };\n\n      query = sync;\n    } else if (\n        env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0) {\n      query = this.beginQuery();\n      this.endQuery();\n      isFencePassed = () => this.isQueryAvailable(\n          query,\n          env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n    } else {\n      // If we have no way to fence, return true immediately. This will fire in\n      // WebGL 1.0 when there is no disjoint query timer. In this case, because\n      // the fence passes immediately, we'll immediately ask for a download of\n      // the texture, which will cause the UI thread to hang.\n      isFencePassed = () => true;\n    }\n\n    return {query, isFencePassed};\n  }\n\n  public downloadMatrixFromPackedTexture(\n      texture: WebGLTexture, physicalRows: number,\n      physicalCols: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadMatrixFromPackedOutputTexture(\n            this.gl, physicalRows, physicalCols));\n  }\n\n  private vertexAttrsAreBound = false;\n\n  public createProgram(fragmentShaderSource: string): WebGLProgram {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    const fragmentShader: WebGLShader =\n        webgl_util.createFragmentShader(gl, fragmentShaderSource);\n    const vertexShader: WebGLShader = gpgpu_util.createVertexShader(gl);\n    const program: WebGLProgram = webgl_util.createProgram(gl);\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, vertexShader));\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, fragmentShader));\n    webgl_util.linkProgram(gl, program);\n    if (this.debug) {\n      webgl_util.validateProgram(gl, program);\n    }\n    if (!this.vertexAttrsAreBound) {\n      this.setProgram(program);\n      this.vertexAttrsAreBound = gpgpu_util.bindVertexProgramAttributeStreams(\n          gl, this.program, this.vertexBuffer);\n    }\n    return program;\n  }\n\n  public deleteProgram(program: WebGLProgram) {\n    this.throwIfDisposed();\n    if (program === this.program) {\n      this.program = null;\n    }\n    if (program != null) {\n      webgl_util.callAndCheck(this.gl, () => this.gl.deleteProgram(program));\n    }\n  }\n\n  public setProgram(program: WebGLProgram|null) {\n    this.throwIfDisposed();\n    this.program = program;\n    if ((this.program != null) && this.debug) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.useProgram(program));\n  }\n\n  public getUniformLocation(\n      program: WebGLProgram, uniformName: string,\n      shouldThrow = true): WebGLUniformLocation {\n    this.throwIfDisposed();\n    if (shouldThrow) {\n      return webgl_util.getProgramUniformLocationOrThrow(\n          this.gl, program, uniformName);\n    } else {\n      return webgl_util.getProgramUniformLocation(\n          this.gl, program, uniformName);\n    }\n  }\n\n  public getAttributeLocation(program: WebGLProgram, attribute: string):\n      number {\n    this.throwIfDisposed();\n    return webgl_util.callAndCheck(\n        this.gl, () => this.gl.getAttribLocation(program, attribute));\n  }\n\n  public getUniformLocationNoThrow(program: WebGLProgram, uniformName: string):\n      WebGLUniformLocation {\n    this.throwIfDisposed();\n    return this.gl.getUniformLocation(program, uniformName);\n  }\n\n  public setInputMatrixTexture(\n      inputMatrixTexture: WebGLTexture, uniformLocation: WebGLUniformLocation,\n      textureUnit: number) {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    webgl_util.bindTextureToProgramUniformSampler(\n        this.gl, inputMatrixTexture, uniformLocation, textureUnit);\n  }\n\n  public setOutputMatrixTexture(\n      outputMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.setOutputMatrixTextureDriver(outputMatrixTexture, columns, rows);\n  }\n\n  public setOutputPackedMatrixTexture(\n      outputPackedMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.throwIfDisposed();\n    const [width, height] =\n        tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n    this.setOutputMatrixTextureDriver(outputPackedMatrixTexture, width, height);\n  }\n\n  public setOutputMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    this.setOutputMatrixWriteRegionDriver(\n        startColumn, startRow, numColumns, numRows);\n  }\n\n  public setOutputPackedMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    throw new Error('setOutputPackedMatrixWriteRegion not implemented.');\n  }\n\n  public debugValidate() {\n    if (this.program != null) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.validateFramebuffer(this.gl);\n  }\n\n  public executeProgram() {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    const gl = this.gl;\n    if (this.debug) {\n      this.debugValidate();\n    }\n    webgl_util.callAndCheck(\n        gl, () => gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_SHORT, 0));\n  }\n\n  public blockUntilAllProgramsCompleted() {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(this.gl, () => this.gl.finish());\n  }\n\n  private getQueryTimerExtension(): WebGL1DisjointQueryTimerExtension\n      |WebGL2DisjointQueryTimerExtension {\n    if (this.disjointQueryTimerExtension == null) {\n      this.disjointQueryTimerExtension =\n          webgl_util.getExtensionOrThrow(\n              this.gl,\n              env().getNumber(\n                  'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2 ?\n                  'EXT_disjoint_timer_query_webgl2' :\n                  'EXT_disjoint_timer_query') as\n              WebGL1DisjointQueryTimerExtension |\n          WebGL2DisjointQueryTimerExtension;\n    }\n    return this.disjointQueryTimerExtension;\n  }\n\n  private getQueryTimerExtensionWebGL2(): WebGL2DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension();\n  }\n\n  private getQueryTimerExtensionWebGL1(): WebGL1DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension() as WebGL1DisjointQueryTimerExtension;\n  }\n\n  beginQuery(): WebGLQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const query = gl2.createQuery();\n      gl2.beginQuery(ext.TIME_ELAPSED_EXT, query);\n      return query;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    const query = ext.createQueryEXT() as WebGLQuery;\n    ext.beginQueryEXT(ext.TIME_ELAPSED_EXT, query);\n    return query;\n  }\n\n  endQuery() {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n      gl2.endQuery(ext.TIME_ELAPSED_EXT);\n      return;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    ext.endQueryEXT(ext.TIME_ELAPSED_EXT);\n  }\n\n  public async waitForQueryAndGetTime(query: WebGLQuery): Promise<number> {\n    await util.repeatedTry(\n        () => this.disposed ||  // while testing contexts are created / disposed\n                                // in rapid succession, so without this check we\n                                // may poll for the query timer indefinitely\n            this.isQueryAvailable(\n                query,\n                env().getNumber(\n                    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION')));\n    return this.getQueryTime(\n        query, env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n  }\n\n  private getQueryTime(query: WebGLQuery, queryTimerVersion: number): number {\n    if (queryTimerVersion === 0) {\n      return null;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n\n      const timeElapsedNanos = gl2.getQueryParameter(query, gl2.QUERY_RESULT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const timeElapsedNanos =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_EXT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    }\n  }\n\n  private isQueryAvailable(query: WebGLQuery, queryTimerVersion: number):\n      boolean {\n    if (queryTimerVersion === 0) {\n      return true;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const available =\n          gl2.getQueryParameter(query, gl2.QUERY_RESULT_AVAILABLE);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const available =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_AVAILABLE_EXT);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    }\n  }\n\n  pollFence(fenceContext: FenceContext) {\n    return new Promise<void>(resolve => {\n      this.addItemToPoll(() => fenceContext.isFencePassed(), () => resolve());\n    });\n  }\n\n  private itemsToPoll: PollItem[] = [];\n\n  pollItems(): void {\n    // Find the last query that has finished.\n    const index = linearSearchLastTrue(this.itemsToPoll.map(x => x.isDoneFn));\n    for (let i = 0; i <= index; ++i) {\n      const {resolveFn} = this.itemsToPoll[i];\n      resolveFn();\n    }\n    this.itemsToPoll = this.itemsToPoll.slice(index + 1);\n  }\n\n  private addItemToPoll(isDoneFn: () => boolean, resolveFn: () => void) {\n    this.itemsToPoll.push({isDoneFn, resolveFn});\n    if (this.itemsToPoll.length > 1) {\n      // We already have a running loop that polls.\n      return;\n    }\n    // Start a new loop that polls.\n    util.repeatedTry(() => {\n      this.pollItems();\n      // End the loop if no more items to poll.\n      return this.itemsToPoll.length === 0;\n    });\n  }\n\n  private bindTextureToFrameBuffer(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    webgl_util.bindColorTextureToFramebuffer(\n        this.gl, texture, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(this.gl);\n    }\n  }\n\n  private unbindTextureToFrameBuffer() {\n    if (this.outputTexture != null) {\n      webgl_util.bindColorTextureToFramebuffer(\n          this.gl, this.outputTexture, this.framebuffer);\n      if (this.debug) {\n        webgl_util.validateFramebuffer(this.gl);\n      }\n    } else {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n    }\n  }\n\n  private downloadMatrixDriver(\n      texture: WebGLTexture,\n      downloadAndDecode: () => Float32Array): Float32Array {\n    this.bindTextureToFrameBuffer(texture);\n    const result = downloadAndDecode();\n    this.unbindTextureToFrameBuffer();\n\n    return result;\n  }\n\n  private setOutputMatrixTextureDriver(\n      outputMatrixTextureMaybePacked: WebGLTexture, width: number,\n      height: number) {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    webgl_util.bindColorTextureToFramebuffer(\n        gl, outputMatrixTextureMaybePacked, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(gl);\n    }\n    this.outputTexture = outputMatrixTextureMaybePacked;\n    webgl_util.callAndCheck(gl, () => gl.viewport(0, 0, width, height));\n    webgl_util.callAndCheck(gl, () => gl.scissor(0, 0, width, height));\n  }\n\n  private setOutputMatrixWriteRegionDriver(\n      x: number, y: number, width: number, height: number) {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(\n        this.gl, () => this.gl.scissor(x, y, width, height));\n  }\n\n  private throwIfDisposed() {\n    if (this.disposed) {\n      throw new Error('Attempted to use disposed GPGPUContext.');\n    }\n  }\n\n  private throwIfNoProgram() {\n    if (this.program == null) {\n      throw new Error('No GPU program is currently set.');\n    }\n  }\n}\n\ntype PollItem = {\n  isDoneFn: () => boolean,\n  resolveFn: () => void\n};\n\n/**\n * Finds the index of the last true element using linear search.\n * Note: We can't do binary search because Chrome expects us to explicitly\n * test all fences before download:\n * https://github.com/tensorflow/tfjs/issues/1145\n */\nexport function linearSearchLastTrue(arr: Array<() => boolean>): number {\n  let i = 0;\n  for (; i < arr.length; ++i) {\n    const isDone = arr[i]();\n    if (!isDone) {\n      break;\n    }\n  }\n  return i - 1;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, Tensor, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport * as shader_compiler from './shader_compiler';\nimport {InputInfo, ShapeInfo} from './shader_compiler';\nimport {PackingScheme, TextureData, TextureUsage} from './tex_util';\n\nexport interface GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  /** If true, this program expects packed input textures. Defaults to false. */\n  packedInputs?: boolean;\n  /** If true, this program produces a packed texture. Defaults to false. */\n  packedOutput?: boolean;\n  /**\n   * Affects what type of texture we allocate for the output. Defaults to\n   * `TextureUsage.RENDER`.\n   */\n  outTexUsage?: TextureUsage;\n  /**\n   * The type of scheme to use when packing texels for the output values.\n   * See `PackingScheme` for details. Defaults to `PackingScheme.SHARED_BATCH`.\n   */\n  outPackingScheme?: PackingScheme;\n}\n\nexport interface GPGPUBinary {\n  webGLProgram: WebGLProgram;\n  program: GPGPUProgram;\n  uniformLocations: {[name: string]: WebGLUniformLocation};\n  source: string;\n  inShapeInfos: ShapeInfo[];\n  outShapeInfo: ShapeInfo;\n  infLoc: WebGLUniformLocation;\n  nanLoc: WebGLUniformLocation;\n}\n\nexport interface TensorData {\n  shape: number[];\n  texData: TextureData;\n  isUniform: boolean;\n  // Available when we decide to upload as uniform instead of texture.\n  uniformValues?: TypedArray;\n}\n\nexport function compileProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, program: GPGPUProgram, inputs: TensorData[],\n    output: TensorData): GPGPUBinary {\n  const userCode = program.userCode;\n  const inputInfos: InputInfo[] = inputs.map((input, i) => {\n    const shapeInfo: ShapeInfo = {\n      logicalShape: input.shape,\n      texShape: input.isUniform ? null : input.texData.texShape,\n      isUniform: input.isUniform,\n      isPacked: input.isUniform ? false : input.texData.isPacked,\n      flatOffset: null\n    };\n    if (input.texData != null && input.texData.slice != null &&\n        input.texData.slice.flatOffset > 0) {\n      shapeInfo.flatOffset = input.texData.slice.flatOffset;\n    }\n    return {name: program.variableNames[i], shapeInfo};\n  });\n  const inShapeInfos = inputInfos.map(x => x.shapeInfo);\n  const outShapeInfo: ShapeInfo = {\n    logicalShape: output.shape,\n    texShape: output.texData.texShape,\n    isUniform: false,\n    isPacked: output.texData.isPacked,\n    flatOffset: null\n  };\n  const source = shader_compiler.makeShader(\n      inputInfos, outShapeInfo, userCode, program.packedInputs);\n\n  const webGLProgram = gpgpu.createProgram(source);\n\n  // Add special uniforms (NAN, INFINITY)\n  let infLoc: WebGLUniformLocation = null;\n  const nanLoc = gpgpu.getUniformLocation(webGLProgram, 'NAN', false);\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    infLoc = gpgpu.getUniformLocation(webGLProgram, 'INFINITY', false);\n  }\n\n  // Add user-defined uniforms\n  const uniformLocations: {[name: string]: WebGLUniformLocation} = {};\n  for (let i = 0; i < program.variableNames.length; i++) {\n    const varName = program.variableNames[i];\n    const shouldThrow = false;\n    uniformLocations[varName] =\n        gpgpu.getUniformLocation(webGLProgram, varName, shouldThrow);\n    uniformLocations[`offset${varName}`] =\n        gpgpu.getUniformLocation(webGLProgram, `offset${varName}`, shouldThrow);\n  }\n\n  return {\n    program,\n    source,\n    webGLProgram,\n    uniformLocations,\n    inShapeInfos,\n    outShapeInfo,\n    infLoc,\n    nanLoc,\n  };\n}\n\nfunction validateBinaryAndProgram(\n    shapeInfos: ShapeInfo[], inputs: TensorData[]) {\n  if (shapeInfos.length !== inputs.length) {\n    throw Error(\n        `Binary was compiled with ${shapeInfos.length} inputs, but ` +\n        `was executed with ${inputs.length} inputs`);\n  }\n\n  shapeInfos.forEach((s, i) => {\n    const shapeA = s.logicalShape;\n    const input = inputs[i];\n    const shapeB = input.shape;\n\n    if (!util.arraysEqual(shapeA, shapeB)) {\n      throw Error(\n          `Binary was compiled with different shapes than ` +\n          `the current args. Shapes ${shapeA} and ${shapeB} must match`);\n    }\n    // The input is uploaded as uniform.\n    if (s.isUniform && input.isUniform) {\n      return;\n    }\n\n    const texShapeA = s.texShape;\n    const texShapeB = input.isUniform ? null : input.texData.texShape;\n    if (!util.arraysEqual(texShapeA, texShapeB)) {\n      throw Error(\n          `Binary was compiled with different texture shapes than the` +\n          ` current args. Shape ${texShapeA} and ${texShapeB} must match`);\n    }\n  });\n}\n\nexport function runProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, binary: GPGPUBinary, inputs: TensorData[],\n    output: TensorData,\n    customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) =>\n        void): void {\n  validateBinaryAndProgram(binary.inShapeInfos, inputs);\n  validateBinaryAndProgram([binary.outShapeInfo], [output]);\n\n  const outTex = output.texData.texture;\n  const outTexShape = output.texData.texShape;\n  if (output.texData.isPacked) {\n    gpgpu.setOutputPackedMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  } else {\n    gpgpu.setOutputMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  }\n  gpgpu.setProgram(binary.webGLProgram);\n\n  // Set special uniforms (NAN, INFINITY)\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    if (binary.infLoc !== null) {\n      gpgpu.gl.uniform1f(binary.infLoc, Infinity);\n    }\n  }\n  if (binary.nanLoc !== null) {\n    gpgpu.gl.uniform1f(binary.nanLoc, NaN);\n  }\n\n  // Set user-defined inputs\n  inputs.forEach((input, i) => {\n    const varName = binary.program.variableNames[i];\n    const varLoc = binary.uniformLocations[varName];\n    const varOffsetLoc = binary.uniformLocations[`offset${varName}`];\n\n    if (varLoc == null) {\n      // The compiler inferred that this variable is not used in this shader.\n      return;\n    }\n\n    if (input.isUniform) {\n      // Upload the values of the tensor as uniform.\n      if (util.sizeFromShape(input.shape) < 2) {\n        gpgpu.gl.uniform1f(varLoc, input.uniformValues[0]);\n      } else {\n        let vals = input.uniformValues;\n        if (!(vals instanceof Float32Array)) {\n          vals = new Float32Array(vals);\n        }\n        gpgpu.gl.uniform1fv(varLoc, vals);\n      }\n      return;\n    }\n\n    // If the input was sliced, upload the flat offset index.\n    if (input.texData.slice != null && varOffsetLoc != null) {\n      gpgpu.gl.uniform1i(varOffsetLoc, input.texData.slice.flatOffset);\n    }\n\n    gpgpu.setInputMatrixTexture(input.texData.texture, varLoc, i);\n  });\n\n  if (customSetup != null) {\n    customSetup(gpgpu, binary.webGLProgram);\n  }\n  gpgpu.executeProgram();\n}\n\nexport function makeShaderKey(\n    program: GPGPUProgram, inputs: TensorData[], output: TensorData): string {\n  let keyInputs = '';\n  inputs.concat(output).forEach(x => {\n    const hasOffset = x.texData != null && x.texData.slice != null &&\n        x.texData.slice.flatOffset > 0;\n    const texShape = x.isUniform ? 'uniform' : x.texData.texShape;\n    keyInputs += `${x.shape}_${texShape}_${hasOffset}`;\n  });\n  const keyUserCode = program.userCode;\n  let key = program.constructor.name;\n  // Fast string concat. See https://jsperf.com/string-concatenation/14.\n  key += '_' + keyInputs + '_' + keyUserCode;\n  return key;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Im2ColPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      outputShape: number[], inputShape: number[],\n      convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = outputShape;\n\n    const {\n      filterWidth,\n      inChannels,\n      strideWidth,\n      strideHeight,\n      padInfo,\n      outWidth,\n      dilationWidth,\n      dilationHeight,\n      dataFormat\n    } = convInfo;\n    const {left, top} = padInfo;\n    const itemsPerBlockRow = inChannels * filterWidth;\n    const glsl = getGlslDifferences();\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const rowDim = isChannelsLast ? 0 : 1;\n    const colDim = isChannelsLast ? 1 : 2;\n\n    let unrolled = ``;\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        unrolled += `\n          blockIndex = rc.y + ${col};\n          pos = rc.x + ${row};\n\n          if(blockIndex < ${outputShape[1]} && pos < ${outputShape[0]}) {\n            offsetY = int(blockIndex / (${outWidth})) * ${strideHeight} - ${\n            top};\n            d0 = offsetY + ${dilationHeight} * (pos / ${itemsPerBlockRow});\n\n            if(d0 < ${inputShape[rowDim]} && d0 >= 0) {\n\n              offsetX = int(mod(float(blockIndex), ${outWidth}.) * ${\n            strideWidth}. - ${left}.);\n              d1 = offsetX + ${dilationWidth} * (int(mod(float(pos), ${\n            itemsPerBlockRow}.) / ${inChannels}.));\n\n              if(d1 < ${inputShape[colDim]} && d1 >= 0) {\n\n                ch = int(mod(float(pos), ${inChannels}.));\n\n                if (${isChannelsLast}) {\n                  innerDims = vec2(d1, ch);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(d0, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                } else {\n                  innerDims = vec2(d0, d1);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(ch, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                }\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      void main() {\n        ivec2 rc = getOutputCoords();\n\n        vec4 result = vec4(0);\n\n        int blockIndex, pos, offsetY, d0, offsetX, d1, ch;\n        vec2 innerDims;\n\n        ${unrolled}\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n        int d = coords[3];\n        float x = getX(b, r, c, d);\n        float sum = 0.0;\n        for (int j = -${rad}; j <= ${rad}; j++) {\n          int idx = d + j;\n          if (idx >= 0 && idx <=  ${maxD}) {\n            float z = getX(b, r, c, idx);\n            sum += z * z;\n          }\n        }\n        float val = x * ${powOperator};\n        setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNGradProgram implements GPGPUProgram {\n  variableNames = ['inputImage', 'outputImage', 'dy'];\n  outputShape: number[] = [];\n  userCode: string;\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n  depth: number;\n\n  constructor(\n      inputShape: number[], depthRadius: number, bias: number, alpha: number,\n      beta: number) {\n    this.outputShape = inputShape;\n    this.depth = inputShape[3];\n    this.depthRadius = depthRadius;\n    this.bias = bias;\n    this.alpha = alpha;\n    this.beta = beta;\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n\n        float result = 0.0;\n        for (int d = 0; d < ${this.depth}; ++d) {\n          int depthBegin = int(max(0.0, float(d - ${depthRadius})));\n          int depthEnd = int(min(float(${this.depth}),\n              float(d + ${depthRadius} + 1)));\n\n          const int MIN_DEPTH_BEGIN = 0;\n          const int MAX_DEPTH_END = ${this.depth};\n\n          float norm = 0.0;\n          for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd) {\n              norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);\n            }\n            else {\n              break;\n            }\n          }\n\n          norm = float(${alpha}) * norm + float(${bias});\n\n          for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd){\n              float dyi = -2.0 * float(${alpha})\n                * float(${beta})\n                * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)\n                / norm;\n              if (k == d) {\n                dyi += pow(norm, -1.0 * ${beta});\n              }\n              if (k == coords[3]) {\n                dyi *= getDy(b, r, c, d);\n                result += dyi;\n              }\n            }\n            else {\n              break;\n            }\n          }\n      }\n      setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords.x;\n        int r = coords.y;\n        int c = coords.z;\n        int d = coords.w;\n\n        bool hasNextCol = d < ${this.outputShape[3]};\n        bool hasNextRow = c < ${this.outputShape[2]};\n\n        vec4 sum = vec4(0.);\n        vec4 xFragAtOutputCoords = getX(b, r, c, d);\n\n        vec4 xAtOutputCoords = vec4(\n          getChannel(xFragAtOutputCoords, vec2(c, d)),\n          hasNextCol ?\n            getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,\n          hasNextRow ?\n            getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0\n        );\n\n        int firstChannel = d - ${rad};\n        vec2 cache = vec2(0.);\n        if(firstChannel >= 0){\n          vec4 firstChannelFrag = getX(b, r, c, firstChannel);\n          cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));\n            if(hasNextRow){\n              cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));\n            }\n        }\n\n        ivec2 depth = ivec2(d, d + 1);\n        for (int j = - ${rad}; j <= ${rad}; j++) {\n          ivec2 idx = depth + j;\n          bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));\n          bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${maxD}));\n\n          bool depthInRange = aboveLowerBound.x && belowUpperBound.x;\n          bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;\n\n          if(depthInRange || depthPlusOneInRange){\n            vec4 z = vec4(0.);\n            vec4 xFragAtCurrentDepth;\n            z.xz = cache.xy;\n            if(depthPlusOneInRange && hasNextCol){\n              xFragAtCurrentDepth = idx.y != d ?\n                getX(b, r, c, idx.y) : xFragAtOutputCoords;\n              z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));\n              if(hasNextRow){\n                z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));\n              }\n            }\n            cache.xy = z.yw;\n            sum += z * z;\n          }\n        }\n        vec4 result = xAtOutputCoords * ${powOperator};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MaxPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex = effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n          wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n            int maxPosValue = ${lastIndex} - int(getMaxPos(b, idyR, idyC, d));\n\n            // Get the current value, check it against the value from the\n            // position matrix.\n            int curPosValue = wR * ${effectiveFilterWidth} + wC;\n            float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n            dotProd += dyValue * mask;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class MaxPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex =\n        effectiveFilterDepth * effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n           wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n              int maxPosValue = ${lastIndex} -\n                  int(getMaxPos(batch, idyD, idyR, idyC, ch));\n\n              // Get the current value, check it against the value from the\n              // position matrix.\n              int curPosValue =\n                  wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                  wR * ${effectiveFilterWidth} + wC;\n              float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n              dotProd += dyValue * mask;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MatMulPackedProgram implements GPGPUProgram {\n  variableNames = ['matrixA', 'matrixB'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      aShape: [number, number, number], outputShape: [number, number, number],\n      transposeA = false, transposeB = false, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = outputShape;\n\n    const sharedDim = transposeA ? aShape[1] : aShape[2];\n    const sharedDimensionPacked = Math.ceil(sharedDim / 2);\n\n    const aSample = transposeA ? 'i * 2, rc.y' : 'rc.y, i * 2';\n    const bSample = transposeB ? 'rc.z, i * 2' : 'i * 2, rc.z';\n    const aSwizzle = transposeA ? ['a.xxyy', 'a.zzww'] : ['a.xxzz', 'a.yyww'];\n    const bSwizzle = transposeB ? ['b.xzxz', 'b.ywyw'] : ['b.xyxy', 'b.zwzw'];\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const float sharedDimension = ${sharedDimensionPacked}.0;\n\n      vec4 dot2x2ARowBCol(ivec3 rc) {\n        vec4 result = vec4(0);\n        for (int i = 0; i < ${sharedDimensionPacked}; i++) {\n          vec4 a = getMatrixA(rc.x, ${aSample});\n          vec4 b = getMatrixB(rc.x, ${bSample});\n\n          // These swizzled products need to be separately added.\n          // See: https://github.com/tensorflow/tfjs/issues/1735\n          result += (${aSwizzle[0]} * ${bSwizzle[0]});\n          result += (${aSwizzle[1]} * ${bSwizzle[1]});\n        }\n        return result;\n      }\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n        vec4 result = dot2x2ARowBCol(rc);\n\n        ${addBiasSnippet}\n\n        ${applyActivationSnippet}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MultinomialProgram implements GPGPUProgram {\n  variableNames = ['probs'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(batchSize: number, numOutcomes: number, numSamples: number) {\n    this.outputShape = [batchSize, numSamples];\n\n    this.userCode = `\n      uniform float seed;\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n\n        float r = random(seed);\n        float cdf = 0.0;\n\n        for (int i = 0; i < ${numOutcomes - 1}; i++) {\n          cdf += getProbs(batch, i);\n\n          if (r < cdf) {\n            setOutput(float(i));\n            return;\n          }\n        }\n\n        // If no other event happened, last event happened.\n        setOutput(float(${numOutcomes - 1}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(seed: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.seedLoc == null) {\n        this.seedLoc = gpgpu.getUniformLocation(webGLProgram, 'seed');\n      }\n      gpgpu.gl.uniform1f(this.seedLoc, seed);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class OneHotProgram implements GPGPUProgram {\n  variableNames = ['indices'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(\n      numIndices: number, depth: number, onValue: number, offValue: number) {\n    this.outputShape = [numIndices, depth];\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int index = round(getIndices(coords.x));\n        setOutput(mix(float(${offValue}), float(${onValue}),\n                      float(index == coords.y)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape:\n          number[]) {  // TODO(https://github.com/tensorflow/tfjs/issues/893):\n                       // Only input / output 3D tensors.\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    if (rank === 0) {\n      this.userCode = `\n        void main() {\n          setOutput(vec4(getA(), 0., 0., 0.));\n        }\n      `;\n    } else {\n      const channels = getChannels('rc', rank);\n      const dtype = getCoordsDataType(rank);\n      const outOfBoundsCondition =\n          getOutOfBoundsCondition(rank, outputShape, channels);\n      const setup = getSetup(\n          rank, outputShape[outputShape.length - 1],\n          outputShape[outputShape.length - 2], channels);\n      const output = getOutput(outputShape, channels);\n\n      this.userCode = `\n        void main() {\n          ${dtype} rc = getOutputCoords();\n\n          if(${outOfBoundsCondition}) {\n            setOutput(vec4(0));\n          } else {\n            ${setup}\n\n            setOutput(vec4(${output}));\n          }\n        }\n      `;\n    }\n  }\n}\n\nfunction getSourceCoordsArr(rank: number, dims: string[]): string[] {\n  const coords = [];\n\n  for (let row = 0; row <= 1; row++) {\n    for (let col = 0; col <= 1; col++) {\n      let coord = `${row === 0 ? 'r' : 'rp1'}, ${col === 0 ? 'c' : 'cp1'}`;\n\n      for (let d = 2; d < rank; d++) {\n        coord = `${dims[dims.length - 1 - d]},` + coord;\n      }\n\n      coords.push(coord);\n    }\n  }\n  return coords;\n}\n\nfunction getOutOfBoundsCondition(\n    rank: number, shape: number[], dims: string[]): string {\n  if (rank === 1) {\n    return `rc > ${shape[0]}`;\n  }\n\n  let cond = '';\n  for (let i = rank - 2; i < rank; i++) {\n    cond += `${dims[i]} >= ${shape[i]}`;\n    if (i < rank - 1) {\n      cond += '||';\n    }\n  }\n\n  return cond;\n}\n\nfunction getSetup(\n    rank: number, cols: number, rows: number, dims: string[]): string {\n  if (rank === 1) {\n    return '';\n  }\n\n  const innerDims = dims.slice(-2);\n\n  return `\n    int r = ${innerDims[0]};\n    int c = ${innerDims[1]};\n    int rp1 = r + 1;\n    int cp1 = c + 1;\n\n    bool cEdge = cp1 >= ${cols};\n    bool rEdge = rp1 >= ${rows};\n  `;\n}\n\nfunction getOutput(shape: number[], dims: string[]): string {\n  const rank = shape.length;\n  const sourceCoords = getSourceCoordsArr(rank, dims);\n  if (rank === 1) {\n    return `getA(rc),\n            rc + 1 >= ${shape[0]} ? 0. : getA(rc + 1),\n            0, 0`;\n  }\n\n  return `getA(${sourceCoords[0]}),\n          cEdge ? 0. : getA(${sourceCoords[1]}),\n          rEdge ? 0. : getA(${sourceCoords[2]}),\n          rEdge || cEdge ? 0. : getA(${sourceCoords[3]})`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const type = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const unpackedCoords =\n        ['coords[0]', 'coords[1]', 'coords[2]', 'coords[3]'].slice(0, rank);\n\n    if (rank === 1) {\n      this.userCode = `\n        int start = ${start};\n        int end = ${end};\n\n        void main() {\n          int outC = getOutputCoords();\n          if (outC < start || outC >= end) {\n            setOutput(float(${constantValue}));\n          } else {\n            setOutput(getX(outC - start));\n          }\n        }\n      `;\n      return;\n    }\n    this.userCode = `\n      ${type} start = ${type}(${start});\n      ${type} end = ${type}(${end});\n\n      void main() {\n        ${type} outC = getOutputCoords();\n        if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {\n          setOutput(float(${constantValue}));\n        } else {\n          ${type} coords = outC - start;\n          setOutput(getX(${unpackedCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const dtype = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const coords = getChannels('rc', rank);\n    const source = getChannels('source', rank);\n    const cLimit = `${coords[rank - 1]} < ${this.outputShape[rank - 1]}`;\n    const innerDims =\n        rank === 1 ? 'source' : `vec2(${source.slice(-2).join()})`;\n\n    const componentSetup = [\n      `${dtype} rc = outputLoc;`, `${coords[rank - 1]} += 1;\n       if(${cLimit}) {\n      `,\n      rank === 1 ? '' : `}\n       rc = outputLoc;\n       ${coords[rank - 2]} += 1;\n       if(${coords[rank - 2]} < ${this.outputShape[rank - 2]}) {`,\n      rank === 1 ? '' : `  ${coords[rank - 1]} += 1;\n         if(${cLimit}) {`\n    ];\n\n    const paddingArea = rank === 1 ?\n        'rc < start || rc >= end' :\n        'any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))';\n    let mainLoop = '';\n    for (let i = 0, j = rank === 1 ? 2 : 4; i < j; i++) {\n      mainLoop += `\n        ${componentSetup[i]}\n        if (${paddingArea}) {\n          result[${i}] = float(${constantValue});\n        } else {\n          ${dtype} source = rc - start;\n          result[${i}] = getChannel(getX(${source.join()}), ${innerDims});\n        }\n      `;\n    }\n    mainLoop += (rank === 1 ? `} ` : `}}`);\n\n    this.userCode = `\n      const ${dtype} start = ${dtype}(${start});\n      const ${dtype} end = ${dtype}(${end});\n\n      void main() {\n        ${dtype} outputLoc = getOutputCoords();\n        vec4 result = vec4(0.);\n        ${mainLoop}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Pool2DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n    const batchFlattenPositionStr = `((batch  * ${convInfo.inHeight} + xR) * ${\n        convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n    const flattenPositionStr =\n        `(xR * ${convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n        const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int batch = coords[0];\n          int d = coords[3];\n\n          ivec2 xRCCorner = coords.yz * strides - pads;\n          int xRCorner = xRCCorner.x;\n          int xCCorner = xRCCorner.y;\n\n          // max/min x(?, ?, d) to get y(yR, yC, d).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n          float avgValue = 0.0;\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              int xC = xCCorner + wC;\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float value = getX(batch, xR, xC, d);\n\n              // If a min / max value has already been found, use it. If not,\n              // use the current value.\n              float currMinMaxValue = mix(\n                  value, minMaxValue, minMaxValueFound);\n              if (value ${compareOp} currMinMaxValue) {\n                minMaxValue = value;\n                minMaxValueFound = 1.0;\n                minMaxPosition = ${\n          flattenPositions ? (includeBatchInIndex ? batchFlattenPositionStr :\n                                                    flattenPositionStr) :\n                             `wR * ${effectiveFilterWidth} + wC`};\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xR, int xC, int d) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xR, xC, d);\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d = coords[3];\n\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // max/min x(?, ?, d) to get y(yR, yC, d).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          int xR = xRCorner + wR;\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              getValue(batch, xR, xC + 3 * ${dilationWidth}, d)\n            );\n\n            ${updateSnippet}\n          }\n\n          int xC = xCCorner + ${filterWidthNearestVec4};\n          if (${filterWidthVec4Remainder === 1}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              initializationValue,\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 2}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 3}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              initializationValue\n            );\n\n            ${updateSnippet}\n          }\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n\nexport class Pool3DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv3DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec3 strides =\n            ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n        const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n        void main() {\n          ivec5 coords = getOutputCoords();\n          int batch = coords.x;\n          int ch = coords.u;\n\n          ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n          int xDCorner = xCorner.x;\n          int xRCorner = xCorner.y;\n          int xCCorner = xCorner.z;\n\n          // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n\n          for (int wD = 0; wD < ${effectiveFilterDepth};\n              wD += ${dilationDepth}) {\n            int xD = xDCorner + wD;\n\n            if (xD < 0 || xD >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int wR = 0; wR < ${effectiveFilterHeight};\n                wR += ${dilationHeight}) {\n              int xR = xRCorner + wR;\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int wC = 0; wC < ${effectiveFilterWidth};\n                  wC += ${dilationWidth}) {\n                int xC = xCCorner + wC;\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float value = getX(batch, xD, xR, xC, ch);\n\n                // If a min / max value has already been found, use it. If not,\n                // use the current value.\n                float currMinMaxValue = mix(\n                    value, minMaxValue, minMaxValueFound);\n                if (value ${compareOp} currMinMaxValue) {\n                  minMaxValue = value;\n                  minMaxValueFound = 1.0;\n                  minMaxPosition = ${\n          flattenPositions ?\n              (includeBatchInIndex ?\n                   `(((batch * ${convInfo.inDepth} + xD) * ${\n                       convInfo.inHeight} + xR) * ${convInfo.inWidth} + xC) * ${\n                       convInfo.inChannels} + ch` :\n                   `((xD * ${convInfo.inHeight} + xR) * ${\n                       convInfo.inWidth} + xC) * ${convInfo.inChannels} + ch`) :\n              `wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                      wR * ${effectiveFilterWidth} + wC`};\n                }\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec3 strides =\n        ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xD, int xR, int xC, int ch) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xD, xR, xC, ch);\n      }\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xDCorner = xCorner.x;\n        int xRCorner = xCorner.y;\n        int xCCorner = xCorner.z;\n\n        // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          int xD = xDCorner + wD;\n\n          if (xD < 0 || xD >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 3 * ${dilationWidth}, ch)\n              );\n\n              ${updateSnippet}\n            }\n\n            int xC = xCCorner + ${filterWidthNearestVec4};\n            if (${filterWidthVec4Remainder === 1}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                initializationValue,\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 2}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 3}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                initializationValue\n              );\n\n              ${updateSnippet}\n            }\n          }\n          setOutput(${returnValue});\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ReduceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo,\n      reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod') {\n    const {windowSize, batchSize, inSize, outSize} = reduceInfo;\n    this.outputShape = [batchSize, outSize];\n\n    let initializationValue = '0.0';\n    let compareOp = ``;\n\n    if (reduceType === 'prod') {\n      initializationValue = '1.0';\n    } else if (reduceType === 'min') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '1.0 / 1e-20';\n      compareOp = `min`;\n    } else if (reduceType === 'max') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n      compareOp = `max`;\n    }\n\n    let returnValue = `${reduceType}(${reduceType}(${reduceType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n\n    if (reduceType === 'sum') {\n      returnValue = `sumValue`;\n    } else if (reduceType === 'prod') {\n      returnValue = `prodValue`;\n    } else if (reduceType === 'all') {\n      returnValue = `allValue`;\n    } else if (reduceType === 'any') {\n      returnValue = `anyValue`;\n    }\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    let updateSnippet = `\n      if (${reduceType === 'sum'}) {\n        sumValue += dot(values, ones);\n      } else if (${reduceType === 'prod'}) {\n        vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);\n        prodValue *= tmp[0] * tmp[1];\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    let vecType = `vec4`;\n\n    if (reduceType === 'all') {\n      initializationValue = '1.0';\n      updateSnippet = `\n        bool reducedAllValue = all(values);\n        float floatedReducedAllValue = float(reducedAllValue);\n        allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    } else if (reduceType === 'any') {\n      initializationValue = '0.0';\n      updateSnippet = `\n        bool reducedAnyValue = any(values);\n        float floatedReducedAnyValue = float(reducedAnyValue);\n        anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    }\n\n    let checkOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float getValue(int batch, int inIdx) {\n        ${checkOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        vec4 minMaxValue = vec4(${initializationValue});\n        float prodValue = 1.0;\n        float sumValue = 0.0;\n        float allValue = 1.0;\n        float anyValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class ReshapePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: [number, number, number], inputShape: [\n    number, number, number\n  ]) {\n    this.outputShape = outputShape;\n\n    let mainLoop = ``;\n    for (let i = 0; i < 4; i++) {\n      let thisRC = `thisRC = rc;`;\n      if (i % 2 === 1) {\n        thisRC += `thisRC.z += 1;`;\n      }\n      if (i > 1) {\n        thisRC += `thisRC.y += 1;`;\n      }\n\n      mainLoop += `\n        ${thisRC}\n        ${i > 0 ? `if(thisRC.y < rows && thisRC.z < cols){` : ''}\n          int flatIndex = getFlatIndex(thisRC);\n\n          ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);\n          vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));\n\n          result[${i}] =\n            getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);\n        ${i > 0 ? '}' : ''}\n      `;\n    }\n\n    this.userCode = `\n      ${getReshapedInputCoords(inputShape)}\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n\n        vec4 result = vec4(0.);\n\n        ivec3 thisRC;\n        int rows = ${outputShape[1]};\n        int cols = ${outputShape[2]};\n\n        ${mainLoop}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nfunction getReshapedInputCoords(shape: [number, number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 inputCoordsFromReshapedOutCoords(int index) {\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(startRLerp - float(winHeight / 2));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(startCLerp - float(winWidth / 2));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float dxR = float(dyR) * heightScale;\n            int topDxRIndex = int(floor(dxR));\n            int bottomDxRIndex = int(min(ceil(dxR), ${xHeight - 1}.0));\n            float dxRLerp = dxR - float(topDxRIndex);\n            float inverseDxRLerp = 1.0 - dxRLerp;\n\n            float dxC = float(dyC) * widthScale;\n            int leftDxCIndex = int(floor(dxC));\n            int rightDxCIndex = int(min(ceil(dxC), ${xWidth - 1}.0));\n            float dxCLerp = dxC - float(leftDxCIndex);\n            float inverseDxCLerp = 1.0 - dxCLerp;\n\n            if (r == topDxRIndex && c == leftDxCIndex) {\n              // topLeft\n              accumulator +=\n                getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;\n            }\n\n            if (r == topDxRIndex && c == rightDxCIndex) {\n              // topRight\n              accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == leftDxCIndex) {\n              // bottomLeft\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == rightDxCIndex) {\n              // bottomRight\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec2 sourceFloorRC = ivec2(sourceFracIndexRC);\n        ivec2 sourceCeilRC = ivec2(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);\n        float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);\n        float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);\n        float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);\n\n        vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);\n\n        float top = topLeft + (topRight - topLeft) * fracRC.y;\n        float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;\n        float newValue = top + (bottom - top) * fracRC.x;\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec3 effectiveInputOverOutputRatioRC = vec3(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec3 inputShapeRC = vec3(${oldHeight}.0, ${oldWidth}.0,\n                                     ${oldWidth}.0);\n\n      float getAValue(int b, int r, int c, int d) {\n        return getChannel(getA(b, r, c, d), vec2(c, d));\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        // Calculate values for next column in yRC.z.\n        ivec3 yRC = coords.yzz + ivec3(0, 0, 1);\n\n        // Fractional source index.\n        vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec3 sourceFloorRC = ivec3(sourceFracIndexRC);\n        ivec3 sourceCeilRC = ivec3(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        // Should we calculate next column and row elements in 2x2 packed cell.\n        bool hasNextCol = d < ${depth - 1};\n        bool hasNextRow = coords.z < ${newWidth - 1};\n\n        // In parallel, construct four corners for all four components in\n        // packed 2x2 cell.\n        vec4 topLeft = vec4(\n          getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 bottomLeft = vec4(\n          getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 topRight = vec4(\n          getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec4 bottomRight = vec4(\n          getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);\n\n        vec4 top = mix(topLeft, topRight, fracRC.yyzz);\n        vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);\n        vec4 newValue = mix(top, bottom, fracRC.x);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeigborBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(floor(startRLerp - float(winHeight / 2)));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(floor(startCLerp - float(winWidth / 2)));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float sourceFracRow =\n              float(${effectiveXSize[0]}) *\n                (float(dyR) / float(${effectiveYSize[0]}));\n\n            float sourceFracCol =\n                float(${effectiveXSize[1]}) *\n                  (float(dyC) / float(${effectiveYSize[1]}));\n\n            int sourceNearestRow = int(min(\n                float(int(${xHeight}) - 1),\n                ${alignCorners} ? float(round(sourceFracRow)) :\n                                  float(floor(sourceFracRow))));\n\n            int sourceNearestCol = int(min(\n                float(int(${xWidth}) - 1),\n                ${alignCorners} ? float(round(sourceFracCol)) :\n                                  float(floor(sourceFracCol))));\n\n            if (r == sourceNearestRow && c == sourceNearestCol) {\n              accumulator += getDy(b, dyR, dyC, d);\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeighborProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    // When align corners is false, we rounds the value with floor.\n    const roundBase = alignCorners ? '0.5' : '0.0';\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the coordinators of nearest neighbor point.\n        ivec2 sourceNearestRC = ivec2(\n          min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${roundBase})));\n\n        float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReverseProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n\n    if (rank === 1) {\n      this.userCode = `\n        void main() {\n          int coord = getOutputCoords();\n          setOutput(getX(${xShape[0]} - coord - 1));\n        }\n      `;\n      return;\n    }\n    const getInCoord = (i: number) => {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - coords[${i}] - 1`;\n      }\n      return `coords[${i}]`;\n    };\n    const inCoords = xShape.map((_, i) => getInCoord(i)).join(',');\n    const type = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${type} coords = getOutputCoords();\n        setOutput(getX(${inCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReversePackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n    const channels = getChannels('rc', rank);\n    const nextColumn =\n        `${channels[rank - 1]} + 1 < ${this.outputShape[rank - 1]}`;\n    const nextRow = `${channels[rank - 2]} + 1 < ${this.outputShape[rank - 2]}`;\n    const type = getCoordsDataType(rank);\n    if (rank === 1) {\n      this.userCode = `\n        void main(){\n          int rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = getChannel(getX(${xShape[0]} - rc - 1),\n            ${xShape[0]} - rc - 1);\n          if(${nextColumn}){\n              result.g = getChannel(getX(${xShape[0]} - (rc  + 1) - 1),\n                ${xShape[0]} - (rc  + 1) - 1);\n          }\n          setOutput(result);\n        }\n      `;\n    } else {\n      this.userCode = `\n        void main() {\n          ${type} rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = ${getR(channels.slice())};\n          if(${nextColumn}){\n            result.g = ${getG(channels.slice())};\n          }\n          if(${nextRow}) {\n            result.b = ${getB(channels.slice())};\n            if(${nextColumn}) {\n              result.a = ${getA(channels.slice())};\n            }\n          }\n          setOutput(result);\n        }\n    `;\n    }\n\n    function getR(channels: string[]): string {\n      return getChannel(channels);\n    }\n\n    function getG(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getB(channels: string[]): string {\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getA(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getChannel(channels: string[]): string {\n      const inCoordsArray = xShape.map((_, i) => getInCoord(i, channels));\n      const inCoords = inCoordsArray.join(',');\n      const innerDims = inCoordsArray.slice(-2).join(',');\n      return `getChannel(getX(${inCoords}), vec2(${innerDims}))`;\n    }\n\n    function getInCoord(i: number, channels1: string[]): string {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - ${channels1[i]} - 1`;\n      } else {\n        return `${channels1[i]}`;\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ScatterProgram implements GPGPUProgram {\n  variableNames = ['updates', 'indices', 'defaultValue'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      updateSize: number, sliceDim: number, indicesRank: number,\n      updatesRank: number, strides: number[], shape: number[],\n      summingDupeIndex = true) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    let indicesString = '';\n    if (indicesRank === 1) {\n      indicesString = 'i';\n    } else if (indicesRank === 2) {\n      indicesString = 'i, j';\n    }\n    const indicesSnippet = `getIndices(${indicesString})`;\n\n    let updatesString = '';\n    if (updatesRank === 1) {\n      updatesString = 'i';\n    } else if (updatesRank === 2) {\n      updatesString = 'i, coords[1]';\n    }\n    const updatesSnippet = `getUpdates(${updatesString})`;\n\n    const strideString = sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${strides});\n\n        void main() {\n          ${dtype} coords = getOutputCoords();\n          float sum = 0.0;\n          bool found = false;\n          for (int i = 0; i < ${updateSize}; i++) {\n            int flattenedIndex = 0;\n            for (int j = 0; j < ${sliceDim}; j++) {\n              int index = round(${indicesSnippet});\n              flattenedIndex += index * ${strideString};\n            }\n            if (flattenedIndex == coords[0]) {\n              sum += ${updatesSnippet};\n              found = true;\n            }\n          }\n          setOutput(mix(getDefaultValue(), sum, float(found)));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class SegmentOpProgram implements GPGPUProgram {\n  variableNames = ['x', 'segmentIds'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      segOpInfo: backend_util.segment_util.SegOpInfo,\n      segOpType: 'unsortedSegmentSum') {\n    const windowSize = segOpInfo.windowSize;\n    const batchSize = segOpInfo.batchSize;\n    const inSize = segOpInfo.inSize;\n    const numSegments = segOpInfo.numSegments;\n    const outSize = numSegments * Math.ceil(inSize / windowSize);\n    this.outputShape = [batchSize, outSize];\n\n    const initializationValue = '0.0';\n    const returnValue = `sumValue`;\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    const updateSnippet = `\n        sumValue += dot(values, segFilter);\n    `;\n\n    let checkValueOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkValueOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n\n    let checkSegmentIdOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkSegmentIdOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return -1.0;\n        }\n      `;\n    }\n\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n\n      float getValue(int batch, int inIdx) {\n        ${checkValueOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      float getSegmentIdAtIndex(int inIdx) {\n        ${checkSegmentIdOutOfBounds}\n        return getSegmentIds(inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = int(floor(float(outIdx) / float(\n          ${numSegments})) * float(${windowSize}));\n        int currentSeg = int(mod(float(outIdx), float(${numSegments})));\n\n        float sumValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          int inIdxSeg = int(getSegmentIdAtIndex(inIdx));\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            0,\n            0,\n            0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n              0,\n              0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            0\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SelectProgram implements GPGPUProgram {\n  variableNames = ['c', 'a', 'b'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(cRank: number, shape: number[], rank: number) {\n    this.outputShape = shape;\n\n    let cCoords;\n    let abCoords;\n    if (rank > 4) {\n      throw Error(`Where for rank ${rank} is not yet supported`);\n    }\n\n    if (rank === 1) {\n      abCoords = `resRC`;\n      cCoords = `resRC`;\n    } else {\n      const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n      const cCoordVars = [];\n      const abCoordVars = [];\n      for (let i = 0; i < shape.length; i++) {\n        abCoordVars.push(`${currentCoords[i]}`);\n        if (i < cRank) {\n          cCoordVars.push(`${currentCoords[i]}`);\n        }\n      }\n      cCoords = cCoordVars.join();\n      abCoords = abCoordVars.join();\n    }\n\n    const dtype = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        float cVal = getC(${cCoords});\n        if (cVal >= 1.0) {\n          setOutput(getA(${abCoords}));\n        } else {\n          setOutput(getB(${abCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SliceProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const uniformPart = `uniform int start[${this.rank}];`;\n    const sourceCoords = getCoords(this.rank);\n\n    let body: string;\n    const coordSum = destSize.map((_, i) => {\n      return `sourceLoc.${coords[i]} = start[${i}] + coords.${coords[i]};`;\n    });\n    body = `\n        ${dtype} sourceLoc;\n        ${dtype} coords = getOutputCoords();\n        ${coordSum.join('\\n')}\n      `;\n    this.userCode = `\n      ${uniformPart}\n      void main() {\n        ${body}\n        setOutput(getSource(${sourceCoords}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n\nconst coords = ['x', 'y', 'z', 'w', 'u', 'v'];\n\nfunction getCoords(rank: number): string {\n  if (rank === 1) {\n    return 'sourceLoc';\n  } else if (rank <= 6) {\n    return coords.slice(0, rank).map(x => 'sourceLoc.' + x).join(',');\n  } else {\n    throw Error(`Slicing for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SlicePackedProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const coords = getChannels('coords', this.rank);\n    const sourceLoc = getChannels('sourceLoc', this.rank);\n\n    const innerDims =\n        this.rank === 1 ? 'sourceLoc' : `vec2(${sourceLoc.slice(-2).join()})`;\n    const getChannel =\n        `getChannel(getSource(${sourceLoc.join()}), ${innerDims})`;\n    const upperRow = `\n      result.x = ${getChannel};\n      if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n        ++${sourceLoc[this.rank - 1]};\n        result.y = ${getChannel};\n        --${sourceLoc[this.rank - 1]};\n      }\n    `;\n    const lowerRow = this.rank === 1 ? '' : `\n      --${coords[this.rank - 1]};\n      if (++${coords[this.rank - 2]} < ${destSize[this.rank - 2]}) {\n        ++${sourceLoc[this.rank - 2]};\n        result.z = ${getChannel};\n        if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n          ++${sourceLoc[this.rank - 1]};\n          result.w = ${getChannel};\n        }\n      }\n    `;\n\n    const sourceLocSetup = this.rank <= 4 ?\n        `sourceLoc = coords +\n            ${dtype}(${destSize.map((_, i) => `start[${i}]`).join()});` :\n        destSize.map((_, i) => `${sourceLoc[i]} = ${coords[i]} + start[${i}];`)\n            .join('\\n');\n    this.userCode = `\n      uniform int start[${this.rank}];\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        ${dtype} sourceLoc;\n        ${sourceLocSetup}\n        vec4 result = vec4(0.);\n        ${upperRow}\n        ${lowerRow}\n        setOutput(result);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class StridedSliceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(begin: number[], strides: number[], size: number[]) {\n    this.outputShape = size;\n    const rank = size.length;\n    const inputDtype = getCoordsDataType(size.length);\n    const dtype = getCoordsDataType(size.length);\n\n    let newCoords = '';\n    if (rank === 1) {\n      newCoords = 'coords * strides + begin';\n    } else {\n      let outputAxis = 0;\n      newCoords =\n          size.map((_, i) => {\n                outputAxis++;\n                return size.length === 1 ?\n                    `coords * strides[${i}] + begin[${i}]` :\n                    `coords[${outputAxis - 1}] * strides[${i}] + begin[${i}]`;\n              })\n              .join(',');\n    }\n\n    this.userCode = `\n      ${inputDtype} begin = ${inputDtype}(${begin});\n      ${inputDtype} strides = ${inputDtype}(${strides});\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        setOutput(getX(${newCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {getInternalFormatForFloat16MatrixTexture, getInternalFormatForFloat16PackedMatrixTexture, getInternalFormatForFloat32MatrixTexture, getInternalFormatForPackedMatrixTexture, getInternalFormatForUnsignedBytesMatrixTexture} from './gpgpu_util';\nimport {getPackedMatrixTextureShapeWidthHeight, getUnpackedMatrixTextureShapeWidthHeight, PhysicalTextureType, TextureConfig, TextureUsage} from './tex_util';\n\nexport class TextureManager {\n  private numUsedTextures = 0;\n  private numFreeTextures = 0;\n  private _numBytesAllocated = 0;\n  private _numBytesFree = 0;  // How many bytes that have been allocated\n                              // are available for reuse.\n  private freeTextures: {[shape: string]: WebGLTexture[]} = {};\n  private logEnabled = false;\n  private usedTextures: {[shape: string]: WebGLTexture[]} = {};\n\n  constructor(private gpgpu: GPGPUContext) {}\n\n  acquireTexture(\n      shapeRC: [number, number], usage: TextureUsage,\n      isPacked: boolean): WebGLTexture {\n    const physicalTexType = getPhysicalFromLogicalTextureType(usage, isPacked);\n\n    const shapeKey = getKeyFromTextureShape(shapeRC, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n    if (!(shapeKey in this.usedTextures)) {\n      this.usedTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shapeRC, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n\n    if (this.freeTextures[shapeKey].length > 0) {\n      this.numFreeTextures--;\n      this.numUsedTextures++;\n      this._numBytesFree -= texBytes;\n      this.log();\n      const newTexture = this.freeTextures[shapeKey].shift();\n      this.usedTextures[shapeKey].push(newTexture);\n      return newTexture;\n    }\n\n    let newTexture: WebGLTexture;\n    if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT32) {\n      newTexture = this.gpgpu.createPackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16PackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT32) {\n      newTexture =\n          this.gpgpu.createFloat32MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (\n        physicalTexType === PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE) {\n      newTexture =\n          this.gpgpu.createUnsignedBytesMatrixTexture(shapeRC[0], shapeRC[1]);\n    }\n    this.usedTextures[shapeKey].push(newTexture);\n\n    this.numUsedTextures++;\n    this._numBytesAllocated += texBytes;\n    this.log();\n\n    return newTexture;\n  }\n\n  releaseTexture(\n      texture: WebGLTexture, shape: [number, number],\n      logicalTexType: TextureUsage, isPacked: boolean): void {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    const physicalTexType =\n        getPhysicalFromLogicalTextureType(logicalTexType, isPacked);\n    const shapeKey = getKeyFromTextureShape(shape, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shape, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n    const deleteTexThreshold = env().get('WEBGL_DELETE_TEXTURE_THRESHOLD');\n    if (deleteTexThreshold !== -1 &&\n        this._numBytesAllocated > deleteTexThreshold) {\n      this.gpgpu.deleteMatrixTexture(texture);\n      this._numBytesAllocated -= texBytes;\n    } else {\n      this.freeTextures[shapeKey].push(texture);\n      this.numFreeTextures++;\n      this._numBytesFree += texBytes;\n    }\n\n    this.numUsedTextures--;\n\n    const texList = this.usedTextures[shapeKey];\n    const texIndex = texList.indexOf(texture);\n    if (texIndex < 0) {\n      throw new Error(\n          'Cannot release a texture that was never provided by this ' +\n          'texture manager');\n    }\n    texList.splice(texIndex, 1);\n    this.log();\n  }\n\n  private log() {\n    if (!this.logEnabled) {\n      return;\n    }\n    const total = this.numFreeTextures + this.numUsedTextures;\n    console.log(\n        'Free/Used', `${this.numFreeTextures} / ${this.numUsedTextures}`,\n        `(${total})`);\n    const freeRatio = this._numBytesFree / this._numBytesAllocated;\n    console.log(`Bytes allocated: ${this._numBytesAllocated}`);\n    console.log(`Bytes unused: ${this._numBytesFree} (${\n        Math.round(100 * freeRatio)}%)`);\n  }\n\n  get numBytesAllocated(): number {\n    return this._numBytesAllocated;\n  }\n\n  get numBytesFree(): number {\n    return this._numBytesFree;\n  }\n\n  getNumUsedTextures(): number {\n    return this.numUsedTextures;\n  }\n\n  getNumFreeTextures(): number {\n    return this.numFreeTextures;\n  }\n\n  dispose() {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    for (const texShape in this.freeTextures) {\n      this.freeTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    for (const texShape in this.usedTextures) {\n      this.usedTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    this.freeTextures = null;\n    this.usedTextures = null;\n    this.numUsedTextures = 0;\n    this.numFreeTextures = 0;\n    this._numBytesAllocated = 0;\n    this._numBytesFree = 0;\n  }\n}\n\nfunction numBytesForInternalFormat(\n    gl: WebGLRenderingContext, internalFormat: number): number {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n  if (internalFormat === glany.R32F) {\n    return 4;\n  } else if (internalFormat === glany.R16F) {\n    return 2;\n  } else if (internalFormat === glany.RGBA32F) {\n    return 16;\n  } else if (internalFormat === gl.RGBA) {\n    return 16;\n  } else if (internalFormat === glany.RGBA16F) {\n    return 8;\n  }\n  throw new Error(`Unknown internal format ${internalFormat}`);\n}\n\nexport function computeBytes(\n    shape: [number, number], physicalTexType: PhysicalTextureType,\n    gl: WebGLRenderingContext, textureConfig: TextureConfig,\n    isPacked: boolean): number {\n  // It is not possible to infer packed status from the texture type because\n  // depending on the textureConfig, different  texture types may resolve to the\n  // same internal format (e.g. in WebGL1, the internal format for\n  // UNPACKED_FLOAT16 textures is gl.RGBA). Therefore we pass in `isPacked`\n  // explicitly.\n  const internalFormat =\n      internalFormatForPhysicalTexType(physicalTexType, textureConfig);\n\n  let numElements: number;\n  if (isPacked) {\n    const [packedWidth, packedHeight] =\n        getPackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = packedWidth * packedHeight;\n\n  } else {\n    const [width, height] =\n        getUnpackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = width * height;\n  }\n\n  const bytesPerElement = numBytesForInternalFormat(gl, internalFormat);\n  return numElements * bytesPerElement;\n}\n\nfunction internalFormatForPhysicalTexType(\n    physicalTexType: PhysicalTextureType,\n    textureConfig: TextureConfig): number {\n  switch (physicalTexType) {\n    case PhysicalTextureType.PACKED_2X2_FLOAT32:\n      return getInternalFormatForPackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_2X2_FLOAT16:\n      return getInternalFormatForFloat16PackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT32:\n      return getInternalFormatForFloat32MatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT16:\n      return getInternalFormatForFloat16MatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE:\n      return getInternalFormatForUnsignedBytesMatrixTexture(textureConfig);\n    default:\n      throw new Error(`Unknown physical texture type ${physicalTexType}`);\n  }\n}\n\nfunction getPhysicalTextureForRendering(isPacked: boolean):\n    PhysicalTextureType {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED')) {\n    if (isPacked) {\n      return PhysicalTextureType.PACKED_2X2_FLOAT32;\n    }\n    return PhysicalTextureType.UNPACKED_FLOAT32;\n  }\n\n  if (isPacked) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT16;\n  }\n  return PhysicalTextureType.UNPACKED_FLOAT16;\n}\n\nfunction getPhysicalFromLogicalTextureType(\n    logicalTexType: TextureUsage, isPacked: boolean): PhysicalTextureType {\n  if (logicalTexType === TextureUsage.UPLOAD) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT32;\n  } else if (logicalTexType === TextureUsage.RENDER || logicalTexType == null) {\n    return getPhysicalTextureForRendering(isPacked);\n  } else if (\n      logicalTexType === TextureUsage.DOWNLOAD ||\n      logicalTexType === TextureUsage.PIXELS) {\n    return PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE;\n  }\n  throw new Error(`Unknown logical texture type ${logicalTexType}`);\n}\n\nfunction getKeyFromTextureShape(\n    shapeRowsCol: [number, number], physicalTexType: PhysicalTextureType,\n    isPacked: boolean): string {\n  return `${shapeRowsCol[0]}_${shapeRowsCol[1]}_${physicalTexType}_${isPacked}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TileProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], reps: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[i] * reps[i];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[]): string {\n  const rank = aShape.length;\n  if (rank > 5) {\n    throw Error(`Tile for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `imod(resRC, ${aShape[0]})`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    sourceCoords.push(`imod(${currentCoords[i]}, ${aShape[i]})`);\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class UnaryOpProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      float unaryOperation(float x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        float x = getAAtOutCoords();\n        float y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n\nconst CHECK_NAN_SNIPPET = `if (isnan(x)) return x;`;\n\nexport const LINEAR = `return x;`;\n\nexport const ABS = `return abs(x);`;\n\nexport const RELU = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : x;\n`;\n\nexport const RELU6 = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : min(6.0, x);\n`;\n\nexport const ELU = `return (x >= 0.0) ? x : (exp(x) - 1.0);`;\n\nexport const SELU = `\n  // Stable and Attracting Fixed Point (0, 1) for Normalized Weights.\n  // see: https://arxiv.org/abs/1706.02515\n  float scaleAlpha = ${backend_util.SELU_SCALEALPHA};\n  float scale = ${backend_util.SELU_SCALE};\n  return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);\n`;\n\nexport function STEP(alpha = 0.0) {\n  return CHECK_NAN_SNIPPET + `\n    return x > 0.0 ? 1.0 : float(${alpha});\n  `;\n}\n\nexport const NEG = `return -x;`;\n\nexport const CEIL = `return ceil(x);`;\n\nexport const FLOOR = `return floor(x);`;\n\nexport const SIGN = `\n  if (isnan(x)) { return 0.0; }\n  return sign(x);\n`;\n\nexport const IS_NAN = `return float(isnan(x));`;\n\nexport const IS_INF = `return float(isinf(x));`;\n\nexport const IS_FINITE = `return float(!isnan(x) && !isinf(x));`;\n\nexport const ROUND = `\n  // OpenGL ES does not support round function.\n  // The algorithm is based on banker's rounding.\n  float base = floor(x);\n  if ((x - base) < 0.5) {\n    return floor(x);\n  } else if ((x - base) > 0.5) {\n    return ceil(x);\n  } else {\n    if (mod(base, 2.0) == 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n`;\n\nexport const EXP = `return exp(x);`;\n\nexport const EXPM1 = `return exp(x) - 1.0;`;\n\nexport const LOG = `if (x < 0.0) return NAN;\n  return log(x);`;\n\nexport const LOG1P = `return log(1.0 + x);`;\n\nexport const SQRT = `return sqrt(x);`;\n\nexport const RSQRT = `return inversesqrt(x);`;\n\nexport const SIGMOID = `return 1.0 / (1.0 + exp(-1.0 * x));`;\n\n/**\n * mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n *\n * epsilon is the difference between 1.0 and the next representable\n * float. For a single precision 32 bit float this should be 2^-23, see:\n * https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\n *\n * too_large = (x > -threshold) is value above which exp(x) may overflow\n * but softplus(x) == x is within machine epsilon\n *\n * too_small = (x < threshold) is value below which exp(x) may underflow,\n * but softplus(x) == exp(x) is within machine epsilon.\n */\nexport const SOFTPLUS = `\n  float epsilon = 1.1920928955078125e-7;\n  float threshold = log(epsilon) + 2.0;\n\n  bool too_large = x > -threshold;\n  bool too_small = x < threshold;\n\n  float result;\n  float exp_x = exp(x);\n\n  if (too_large){\n    result = x;\n  }\n  else if (too_small){\n    result = exp_x;\n  }\n  else{\n    result = log(exp_x + 1.0);\n  }\n  return result;\n`;\n\nexport const ASIN = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return asin(x);\n`;\n\nexport const ACOS = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return acos(x);\n`;\n\nexport const ATAN = CHECK_NAN_SNIPPET + `\n  return atan(x);\n`;\n\nexport const SINH = `\n  float e2x = exp(x);\n  return (e2x - 1.0 / e2x) / 2.0;\n`;\n\nexport const COSH = `\n  float e2x = exp(-x);\n  return (e2x + 1.0 / e2x) / 2.0;\n`;\n\nexport const TANH = `\n  float e2x = exp(-2.0 * abs(x));\n  return sign(x) * (1.0 - e2x) / (1.0 + e2x);\n`;\n\nexport const ASINH = CHECK_NAN_SNIPPET + `return log(x + sqrt(x * x + 1.0));`;\n\nexport const ACOSH = CHECK_NAN_SNIPPET + `\n  if (x < 1.0) return NAN;\n  return log(x + sqrt(x * x - 1.0));`;\n\nexport const ATANH = CHECK_NAN_SNIPPET + `\n  if ((x < -1.0) || (x > 1.0)) return NAN;\n  return (log(1.0 + x) - log(1.0 - x)) / 2.0;`;\n\nexport const ERF = `\n  // Error function is calculated approximately with elementary function.\n  // See \"Handbook of Mathematical Functions with Formulas,\n  // Graphs, and Mathematical Tables\", Abramowitz and Stegun.\n  float p = ${backend_util.ERF_P};\n  float a1 = ${backend_util.ERF_A1};\n  float a2 = ${backend_util.ERF_A2};\n  float a3 = ${backend_util.ERF_A3};\n  float a4 = ${backend_util.ERF_A4};\n  float a5 = ${backend_util.ERF_A5};\n\n  float sign = sign(x);\n  x = abs(x);\n  float t = 1.0 / (1.0 + p * x);\n  return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));\n`;\n\nexport const RECIPROCAL = `return 1.0 / x;`;\n\nexport const LOGICAL_NOT = `return float(!(x >= 1.0));`;\n\nexport const TO_INT = `return float(int(x));`;\n\nexport const CLONE = 'return x;';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const LINEAR = `return x;`;\n\nexport const LOG = `\n  vec4 result = log(x);\n  vec4 isNaN = vec4(lessThan(x, vec4(0.0)));\n  result.r = isNaN.r == 1.0 ? NAN : result.r;\n  result.g = isNaN.g == 1.0 ? NAN : result.g;\n  result.b = isNaN.b == 1.0 ? NAN : result.b;\n  result.a = isNaN.a == 1.0 ? NAN : result.a;\n\n  return result;\n`;\n\nexport const RELU = `\n  vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const RELU6 = `\n  vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const ELU = `\n  vec4 result;\n\n  result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);\n  result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);\n  result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);\n  result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);\n\n  return result;\n`;\n\nexport class UnaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      vec4 unaryOperation(vec4 x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        vec4 x = getAAtOutCoords();\n        vec4 y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels, getSourceCoords} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class UnpackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = false;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: number[]) {\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    const channels = getChannels('rc', rank);\n    const dtype = getCoordsDataType(rank);\n    const sourceCoords = getSourceCoords(rank, channels);\n    const innerDims = channels.slice(-2);\n    const coords = rank <= 1 ? 'rc' : `vec2(${innerDims.join(',')})`;\n\n    this.userCode = `\n      void main() {\n        ${dtype} rc = getOutputCoords();\n        vec4 packedInput = getA(${sourceCoords});\n\n        setOutput(getChannel(packedInput, ${coords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import webgl flags.\nimport './flags_webgl';\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {complex, DataId, div, engine, env, imag, max, MemoryInfo, range, real, RecursiveArray, reshape, scalar, softmax, tensor, tidy, TimingInfo, transpose} from '@tensorflow/tfjs-core';\nimport {backend_util, buffer, kernel_impls, slice_util, util} from '@tensorflow/tfjs-core';\nimport {DataStorage, DataType, KernelBackend, NumericDataType, Rank, Scalar, ShapeMap, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorInfo, TypedArray, upcastType} from '@tensorflow/tfjs-core';\n\nimport {addImplCPU, ceilImplCPU, expImplCPU, expm1ImplCPU, floorImplCPU, logImplCPU, multiplyImplCPU, rsqrtImplCPU, simpleAbsImplCPU, sliceImplCPU, subImplCPU} from './kernel_utils/shared';\n\nconst {segment_util} = backend_util;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\n\nimport {AddNProgram} from './addn_gpu';\nimport {AddNPackedProgram} from './addn_packed_gpu';\nimport {ArgMinMaxProgram} from './argminmax_gpu';\nimport {ArgMinMaxPackedProgram} from './argminmax_packed_gpu';\nimport {AvgPool3DBackpropProgram} from './avg_pool_backprop_gpu';\nimport * as binaryop_complex_gpu from './binaryop_complex_gpu';\nimport {BinaryOpComplexProgram} from './binaryop_complex_gpu';\nimport * as binaryop_gpu from './binaryop_gpu';\nimport {BinaryOpProgram} from './binaryop_gpu';\nimport * as binaryop_packed_gpu from './binaryop_packed_gpu';\nimport {BinaryOpPackedProgram} from './binaryop_packed_gpu';\nimport {getWebGLContext} from './canvas_util';\nimport {ClipProgram} from './clip_gpu';\nimport {ClipPackedProgram} from './clip_packed_gpu';\nimport {ComplexAbsProgram} from './complex_abs_gpu';\nimport {ConcatProgram} from './concat_gpu';\nimport {ConcatPackedProgram} from './concat_packed_gpu';\nimport {Conv2DDerFilterProgram, Conv2DDerInputProgram, Conv3DDerFilterProgram, Conv3DDerInputProgram} from './conv_backprop_gpu';\nimport {DepthwiseConv2DDerFilterProgram, DepthwiseConv2DDerInputProgram} from './conv_backprop_gpu_depthwise';\nimport {Conv2DProgram, Conv3DProgram} from './conv_gpu';\nimport {DepthwiseConv2DProgram} from './conv_gpu_depthwise';\nimport {DepthwiseConvPacked2DProgram} from './conv_packed_gpu_depthwise';\nimport {CropAndResizeProgram} from './crop_and_resize_gpu';\nimport {CumSumProgram} from './cumsum_gpu';\nimport {DecodeMatrixProgram} from './decode_matrix_gpu';\nimport {DecodeMatrixPackedProgram} from './decode_matrix_packed_gpu';\nimport {DepthToSpaceProgram} from './depth_to_space_gpu';\nimport {DiagProgram} from './diag_gpu';\nimport {EncodeFloatProgram} from './encode_float_gpu';\nimport {EncodeFloatPackedProgram} from './encode_float_packed_gpu';\nimport {EncodeMatrixProgram} from './encode_matrix_gpu';\nimport {EncodeMatrixPackedProgram} from './encode_matrix_packed_gpu';\nimport * as fft_gpu from './fft_gpu';\nimport {FFTProgram} from './fft_gpu';\nimport {FillProgram} from './fill_gpu';\nimport {GatherProgram} from './gather_gpu';\nimport {GatherNDProgram} from './gather_nd_gpu';\nimport {GPGPUContext} from './gpgpu_context';\nimport * as gpgpu_math from './gpgpu_math';\nimport {GPGPUBinary, GPGPUProgram, TensorData} from './gpgpu_math';\nimport {Im2ColPackedProgram} from './im2col_packed_gpu';\nimport {LRNProgram} from './lrn_gpu';\nimport {LRNGradProgram} from './lrn_grad_gpu';\nimport {LRNPackedProgram} from './lrn_packed_gpu';\nimport {MaxPool3DBackpropProgram} from './max_pool_backprop_gpu';\nimport {MatMulPackedProgram} from './mulmat_packed_gpu';\nimport {MultinomialProgram} from './multinomial_gpu';\nimport {OneHotProgram} from './onehot_gpu';\nimport {PackProgram} from './pack_gpu';\nimport {PadProgram} from './pad_gpu';\nimport {PadPackedProgram} from './pad_packed_gpu';\nimport {Pool3DProgram} from './pool_gpu';\nimport {ReduceProgram} from './reduce_gpu';\nimport {ReshapePackedProgram} from './reshape_packed_gpu';\nimport {ResizeBilinearBackpropProgram} from './resize_bilinear_backprop_gpu';\nimport {ResizeBilinearProgram} from './resize_bilinear_gpu';\nimport {ResizeBilinearPackedProgram} from './resize_bilinear_packed_gpu';\nimport {ResizeNearestNeigborBackpropProgram} from './resize_nearest_neighbor_backprop_gpu';\nimport {ResizeNearestNeighborProgram} from './resize_nearest_neighbor_gpu';\nimport {ReverseProgram} from './reverse_gpu';\nimport {ReversePackedProgram} from './reverse_packed_gpu';\nimport {ScatterProgram} from './scatter_gpu';\nimport {SegmentOpProgram} from './segment_gpu';\nimport {SelectProgram} from './select_gpu';\nimport {SliceProgram} from './slice_gpu';\nimport {SlicePackedProgram} from './slice_packed_gpu';\nimport {StridedSliceProgram} from './strided_slice_gpu';\nimport * as tex_util from './tex_util';\nimport {TextureData, TextureUsage} from './tex_util';\nimport {TextureManager} from './texture_manager';\nimport {TileProgram} from './tile_gpu';\nimport * as unary_op from './unaryop_gpu';\nimport {UnaryOpProgram} from './unaryop_gpu';\nimport * as unary_packed_op from './unaryop_packed_gpu';\nimport {UnaryOpPackedProgram} from './unaryop_packed_gpu';\nimport {UnpackProgram} from './unpack_gpu';\nimport * as webgl_util from './webgl_util';\nimport {BackendValues} from '@tensorflow/tfjs-core';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\ntype KernelInfo = {\n  name: string; query: Promise<number>;\n};\n\nexport type TimerNode = RecursiveArray<KernelInfo>|KernelInfo;\nexport interface CPUTimerQuery {\n  startMs: number;\n  endMs?: number;\n}\n\nexport interface WebGLMemoryInfo extends MemoryInfo {\n  numBytesInGPU: number;\n  // Tracks the total number of bytes allocated on the GPU, accounting for the\n  // physical texture type.\n  numBytesInGPUAllocated: number;\n  // Tracks byte size of textures that were created and then made available for\n  // reuse (disposed).\n  numBytesInGPUFree: number;\n  unreliable: boolean;\n}\n\nexport interface WebGLTimingInfo extends TimingInfo {\n  uploadWaitMs: number;\n  downloadWaitMs: number;\n}\n\nconst binaryCaches: {[webGLVersion: string]: {[key: string]: GPGPUBinary}} = {};\n\nexport function getBinaryCache(webGLVersion: number) {\n  if (webGLVersion in binaryCaches) {\n    return binaryCaches[webGLVersion];\n  }\n  binaryCaches[webGLVersion] = {};\n  return binaryCaches[webGLVersion];\n}\n\nfunction mapActivationToShaderProgram(\n    activation: backend_util.Activation, packed = false): string {\n  if (activation === 'linear') {\n    if (packed) {\n      return unary_packed_op.LINEAR;\n    }\n    return unary_op.LINEAR;\n  } else if (activation === 'relu') {\n    if (packed) {\n      return unary_packed_op.RELU;\n    }\n    return unary_op.RELU;\n  } else if (activation === 'elu') {\n    if (packed) {\n      return unary_packed_op.ELU;\n    }\n    return unary_op.ELU;\n  } else if (activation === 'relu6') {\n    if (packed) {\n      return unary_packed_op.RELU6;\n    }\n    return unary_op.RELU6;\n  } else if (activation === 'prelu') {\n    if (packed) {\n      return binaryop_packed_gpu.PRELU;\n    }\n    return binaryop_gpu.PRELU;\n  }\n  throw new Error(`Activation ${\n      activation} has not been implemented for the WebGL backend.`);\n}\n\n// Empirically determined constant used to determine size threshold for handing\n// off execution to the CPU.\nconst CPU_HANDOFF_SIZE_THRESHOLD = 128;\n\n// Empirically determined constant used to decide the number of MB on GPU\n// before we warn about high memory use. The MB are this constant * screen area\n// * dpi / 1024 / 1024.\nconst BEFORE_PAGING_CONSTANT = 600;\nfunction numMBBeforeWarning(): number {\n  if (env().global.screen == null) {\n    return 1024;  // 1 GB.\n  }\n  return (env().global.screen.height * env().global.screen.width *\n          window.devicePixelRatio) *\n      BEFORE_PAGING_CONSTANT / 1024 / 1024;\n}\n\n// Empirically determined minimal shared dimension in matmul before we forward\n// to a.mul(b).sum() in order to take advantage of GPU parallelism. See\n// https://github.com/tensorflow/tfjs-core/pull/1379 for benchmarks.\nexport const MATMUL_SHARED_DIM_THRESHOLD = 1000;\n\nexport class MathBackendWebGL extends KernelBackend {\n  texData: DataStorage<TextureData>;\n  gpgpu: GPGPUContext;\n\n  // Maps data ids that have a pending read operation, to list of subscribers.\n  private pendingRead = new WeakMap<DataId, Array<(arr: TypedArray) => void>>();\n  // List of data ids that are scheduled for disposal, but are waiting on a\n  // pending read operation.\n  private pendingDisposal = new WeakSet<DataId>();\n  // Used to count the number of 'shallow' sliced tensors that point to the\n  // same data id.\n  private dataRefCount = new WeakMap<DataId, number>();\n  private numBytesInGPU = 0;\n\n  private canvas: HTMLCanvasElement|OffscreenCanvas;\n\n  private programTimersStack: TimerNode[];\n  private activeTimers: TimerNode[];\n  // Accumulated time spent (including blocking) in uploading data to webgl.\n  private uploadWaitMs = 0;\n  // Accumulated time spent (including blocking in downloading data from webgl.\n  private downloadWaitMs = 0;\n  private cpuBackend: KernelBackend;\n\n  // Number of bits of precision of this backend.\n  private floatPrecisionValue: 32|16;\n\n  private textureManager: TextureManager;\n  private binaryCache: {[key: string]: GPGPUBinary};\n  private gpgpuCreatedLocally: boolean;\n  private numMBBeforeWarning: number;\n  private warnedAboutMemory = false;\n  private warnedAboutCPUBackend = false;\n\n  constructor(gpgpu?: GPGPUContext) {\n    super();\n    if (!env().getBool('HAS_WEBGL')) {\n      throw new Error('WebGL is not supported on this device');\n    }\n\n    if (gpgpu == null) {\n      const gl = getWebGLContext(env().getNumber('WEBGL_VERSION'));\n      this.binaryCache = getBinaryCache(env().getNumber('WEBGL_VERSION'));\n      this.gpgpu = new GPGPUContext(gl);\n      this.canvas = gl.canvas;\n      this.gpgpuCreatedLocally = true;\n    } else {\n      this.gpgpu = gpgpu;\n      this.binaryCache = {};\n      this.gpgpuCreatedLocally = false;\n      this.canvas = gpgpu.gl.canvas;\n    }\n    this.textureManager = new TextureManager(this.gpgpu);\n    this.numMBBeforeWarning = numMBBeforeWarning();\n\n    this.texData = new DataStorage(this, engine());\n  }\n\n  numDataIds() {\n    return this.texData.numDataIds() +\n        (this.cpuBackend ? this.cpuBackend.numDataIds() : 0) -\n        this.pendingDeletes;\n  }\n\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    if (env().getBool('WEBGL_CHECK_NUMERICAL_PROBLEMS') ||\n        env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64' && values != null) {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    const dataId = {};\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n    return dataId;\n  }\n\n  /** Increase refCount of a `TextureData`. */\n  incRef(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    texData.refCount++;\n  }\n\n  /** Decrease refCount of a `TextureData`. */\n  decRef(dataId: DataId): void {\n    if (this.texData.has(dataId)) {\n      const texData = this.texData.get(dataId);\n      texData.refCount--;\n    }\n  }\n\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    if (env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.texData.has(dataId)) {\n      const textureData = this.texData.get(dataId);\n\n      textureData.refCount--;\n\n      if (textureData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    const texData = this.texData.get(dataId);\n    const {values, dtype, complexTensors, slice, shape, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.readSync(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n    if (dtype === 'string') {\n      return values;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let result: Float32Array;\n    if (dtype === 'complex64') {\n      const realValues = complexTensors.real.dataSync() as Float32Array;\n      const imagValues = complexTensors.imag.dataSync() as Float32Array;\n      result = backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    } else {\n      result = this.getValuesFromTexture(dataId);\n    }\n\n    if (shouldTimeProgram) {\n      this.downloadWaitMs += util.now() - start;\n    }\n    return this.convertAndCacheOnCPU(dataId, result);\n  }\n\n  async read(dataId: DataId): Promise<BackendValues> {\n    if (this.pendingRead.has(dataId)) {\n      const subscribers = this.pendingRead.get(dataId);\n      return new Promise<TypedArray>(resolve => subscribers.push(resolve));\n    }\n    const texData = this.texData.get(dataId);\n    const {values, shape, slice, dtype, complexTensors, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.read(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n\n    if (!env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED') &&\n        env().getNumber('WEBGL_VERSION') === 2) {\n      throw new Error(\n          `tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and ` +\n          `WEBGL_VERSION=2 not yet supported.`);\n    }\n\n    let buffer = null;\n    let tmpDownloadTarget: TensorInfo;\n\n    if (dtype !== 'complex64' && env().get('WEBGL_BUFFER_SUPPORTED')) {\n      // Possibly copy the texture into a buffer before inserting a fence.\n      tmpDownloadTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpDownloadTarget.dataId);\n\n      buffer = this.gpgpu.createBufferFromTexture(\n          tmpData.texture, ...tex_util.getDenseTexShape(shape));\n    }\n\n    this.pendingRead.set(dataId, []);\n\n    if (dtype !== 'complex64') {\n      // Create a fence and wait for it to resolve.\n      await this.gpgpu.createAndWaitForFence();\n    }\n\n    // Download the values from the GPU.\n    let vals: Float32Array;\n    if (dtype === 'complex64') {\n      const ps = await Promise.all(\n          [complexTensors.real.data(), complexTensors.imag.data()]);\n      const realValues = ps[0];\n      const imagValues = ps[1];\n      vals = backend_util.mergeRealAndImagArrays(\n          realValues as Float32Array, imagValues as Float32Array);\n    } else if (buffer == null) {\n      vals = this.getValuesFromTexture(dataId);\n    } else {\n      const size = util.sizeFromShape(shape);\n      vals = this.gpgpu.downloadFloat32MatrixFromBuffer(buffer, size);\n    }\n    if (tmpDownloadTarget != null) {\n      this.disposeIntermediateTensorInfo(tmpDownloadTarget);\n    }\n    const dTypeVals = this.convertAndCacheOnCPU(dataId, vals);\n\n    const subscribers = this.pendingRead.get(dataId);\n    this.pendingRead.delete(dataId);\n\n    // Notify all pending reads.\n    subscribers.forEach(resolve => resolve(dTypeVals));\n    if (this.pendingDisposal.has(dataId)) {\n      this.pendingDisposal.delete(dataId);\n      this.disposeData(dataId);\n      this.pendingDeletes--;\n    }\n    return dTypeVals;\n  }\n\n  private checkNumericalProblems(values: BackendValues): void {\n    if (values == null) {\n      return;\n    }\n    for (let i = 0; i < values.length; i++) {\n      const num = values[i] as number;\n      if (!webgl_util.canBeRepresented(num)) {\n        if (env().getBool('WEBGL_RENDER_FLOAT32_CAPABLE')) {\n          throw Error(\n              `The value ${num} cannot be represented with your ` +\n              `current settings. Consider enabling float32 rendering: ` +\n              `'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`);\n        }\n        throw Error(`The value ${num} cannot be represented on this device.`);\n      }\n    }\n  }\n\n  private getValuesFromTexture(dataId: DataId): Float32Array {\n    const {shape, dtype, isPacked} = this.texData.get(dataId);\n    const size = util.sizeFromShape(shape);\n    if (env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED')) {\n      const tmpTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpTarget.dataId);\n      const vals = this.gpgpu\n                       .downloadMatrixFromPackedTexture(\n                           tmpData.texture, ...tex_util.getDenseTexShape(shape))\n                       .subarray(0, size);\n\n      this.disposeIntermediateTensorInfo(tmpTarget);\n\n      return vals;\n    }\n\n    const shouldUsePackedProgram =\n        env().getBool('WEBGL_PACK') && isPacked === true;\n    const outputShape =\n        shouldUsePackedProgram ? webgl_util.getShapeAs3D(shape) : shape;\n    const program = shouldUsePackedProgram ?\n        new EncodeFloatPackedProgram(outputShape as [number, number, number]) :\n        new EncodeFloatProgram(outputShape);\n    const output = this.runWebGLProgram(\n        program, [{shape: outputShape, dtype, dataId}], 'float32');\n    const tmpData = this.texData.get(output.dataId);\n    const vals =\n        this.gpgpu\n            .downloadByteEncodedFloatMatrixFromOutputTexture(\n                tmpData.texture, tmpData.texShape[0], tmpData.texShape[1])\n            .subarray(0, size);\n    this.disposeIntermediateTensorInfo(output);\n\n    return vals;\n  }\n\n  async time(f: () => void): Promise<WebGLTimingInfo> {\n    const oldActiveTimers = this.activeTimers;\n    const newActiveTimers: TimerNode[] = [];\n\n    let outerMostTime = false;\n    if (this.programTimersStack == null) {\n      this.programTimersStack = newActiveTimers;\n      outerMostTime = true;\n    } else {\n      this.activeTimers.push(newActiveTimers);\n    }\n    this.activeTimers = newActiveTimers;\n\n    f();\n\n    // needing to split these up because util.flatten only accepts certain types\n    const flattenedActiveTimerQueries =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.query))\n            .filter(d => d != null);\n    const flattenedActiveTimerNames =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.name))\n            .filter(d => d != null);\n\n    this.activeTimers = oldActiveTimers;\n\n    if (outerMostTime) {\n      this.programTimersStack = null;\n    }\n\n    const res: WebGLTimingInfo = {\n      uploadWaitMs: this.uploadWaitMs,\n      downloadWaitMs: this.downloadWaitMs,\n      kernelMs: null,\n      wallMs: null  // will be filled by the engine\n    };\n\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      const kernelMs = await Promise.all(flattenedActiveTimerQueries);\n\n      res['kernelMs'] = util.sum(kernelMs);\n      res['getExtraProfileInfo'] = () =>\n          kernelMs.map((d, i) => ({name: flattenedActiveTimerNames[i], ms: d}))\n              .map(d => `${d.name}: ${d.ms}`)\n              .join(', ');\n    } else {\n      res['kernelMs'] = {\n        error: 'WebGL query timers are not supported in this environment.'\n      };\n    }\n\n    this.uploadWaitMs = 0;\n    this.downloadWaitMs = 0;\n    return res;\n  }\n  memory(): WebGLMemoryInfo {\n    return {\n      unreliable: false,\n      numBytesInGPU: this.numBytesInGPU,\n      numBytesInGPUAllocated: this.textureManager.numBytesAllocated,\n      numBytesInGPUFree: this.textureManager.numBytesFree\n    } as WebGLMemoryInfo;\n  }\n\n  private startTimer(): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.beginQuery();\n    }\n    return {startMs: util.now(), endMs: null};\n  }\n\n  private endTimer(query: WebGLQuery|CPUTimerQuery): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      this.gpgpu.endQuery();\n      return query;\n    }\n    (query as CPUTimerQuery).endMs = util.now();\n    return query;\n  }\n\n  private async getQueryTime(query: WebGLQuery|CPUTimerQuery): Promise<number> {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.waitForQueryAndGetTime(query as WebGLQuery);\n    }\n    const timerQuery = query as CPUTimerQuery;\n    return timerQuery.endMs - timerQuery.startMs;\n  }\n\n  private pendingDeletes = 0;\n\n  disposeData(dataId: DataId): void {\n    if (this.pendingDisposal.has(dataId)) {\n      return;\n    }\n    if (this.pendingRead.has(dataId)) {\n      this.pendingDisposal.add(dataId);\n      this.pendingDeletes++;\n      return;\n    }\n    // No-op if already disposed.\n    if (!this.texData.has(dataId)) {\n      return;\n    }\n\n    this.releaseGPUData(dataId);\n    const {complexTensors} = this.texData.get(dataId);\n    if (complexTensors != null) {\n      complexTensors.real.dispose();\n      complexTensors.imag.dispose();\n    }\n    this.texData.delete(dataId);\n  }\n\n  private releaseGPUData(dataId: DataId): void {\n    const {texture, dtype, texShape, usage, isPacked, slice} =\n        this.texData.get(dataId);\n    const key = slice && slice.origDataId || dataId;\n    const refCount = this.dataRefCount.get(key);\n    if (refCount > 1) {\n      this.dataRefCount.set(key, refCount - 1);\n    } else {\n      this.dataRefCount.delete(key);\n      if (texture != null) {\n        this.numBytesInGPU -= this.computeBytes(texShape, dtype);\n        this.textureManager.releaseTexture(texture, texShape, usage, isPacked);\n      }\n    }\n    const texData = this.texData.get(dataId);\n    texData.texture = null;\n    texData.texShape = null;\n    texData.isPacked = false;\n    texData.slice = null;\n  }\n\n  getTexture(dataId: DataId): WebGLTexture {\n    this.uploadToGPU(dataId);\n    return this.texData.get(dataId).texture;\n  }\n\n  /**\n   * Returns internal information for the specific data bucket. Used in unit\n   * tests.\n   */\n  getDataInfo(dataId: DataId): TextureData {\n    return this.texData.get(dataId);\n  }\n\n  private getCPUBackend(): KernelBackend|null {\n    if (!env().getBool('WEBGL_CPU_FORWARD')) {\n      return null;\n    }\n\n    if (this.cpuBackend == null) {\n      this.cpuBackend = engine().findBackend('cpu');\n    }\n\n    return this.cpuBackend;\n  }\n\n  /*\n  Tests whether all the inputs to an op are small and on the CPU. This heuristic\n  determines when it would be faster to execute a kernel on the CPU. WebGL\n  kernels opt into running this check and forwarding when appropriate.\n  TODO(https://github.com/tensorflow/tfjs/issues/872): Develop a more\n  sustainable strategy for optimizing backend execution of ops.\n   */\n  shouldExecuteOnCPU(\n      inputs: TensorInfo[],\n      sizeThreshold = CPU_HANDOFF_SIZE_THRESHOLD): boolean {\n    const cpuBackend = this.getCPUBackend();\n    if (!this.warnedAboutCPUBackend && cpuBackend == null) {\n      console.warn(\n          'Your application contains ops that are small enough to be ' +\n          'executed on the CPU backend, however the CPU backend cannot ' +\n          'be found. Consider importing the CPU backend ' +\n          '(@tensorflow/tfjs-backend-cpu) for better performance.');\n\n      this.warnedAboutCPUBackend = true;\n    }\n\n    return cpuBackend != null &&\n        inputs.every(\n            input => this.texData.get(input.dataId).texture == null &&\n                util.sizeFromShape(input.shape) < sizeThreshold);\n  }\n\n  getGPGPUContext(): GPGPUContext {\n    return this.gpgpu;\n  }\n\n  complex<T extends Tensor>(real: T, imag: T): T {\n    const result = this.makeOutput(real.shape, 'complex64');\n    const resultData = this.texData.get(result.dataId);\n    // The backend owns the reference to the underlying real and imaginary\n    // clones. These will explicitly get disposed when the complex tensor is\n    // disposed.\n    resultData.complexTensors = {\n      real: engine().keep(real.clone()),\n      imag: engine().keep(imag.clone())\n    };\n\n    return result as T;\n  }\n  real<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.real.clone() as T;\n  }\n  imag<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.imag.clone() as T;\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = sliceImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, begin, size, x.shape,\n          x.dtype);\n      return this.makeOutput(size, x.dtype, outValues);\n    }\n    // Short-circuit computation if the slice is zero-sized.\n    if (util.sizeFromShape(size) === 0) {\n      return tensor([], size, x.dtype) as T;\n    }\n    const {isPacked} = this.texData.get(x.dataId);\n    const isContinous = slice_util.isSliceContinous(x.shape, begin, size);\n    if (isPacked || !isContinous) {\n      const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n          new SlicePackedProgram(size) :\n          new SliceProgram(size);\n      const customSetup = program.getCustomSetupFunc(begin);\n      return this.compileAndRun(program, [x], null, customSetup);\n    }\n    this.uploadToGPU(x.dataId);\n    return this.shallowSlice(x, begin, size) as T;\n  }\n\n  private shallowSlice(x: Tensor, begin: number[], size: number[]): Tensor {\n    const xTexData = this.texData.get(x.dataId);\n    const t = this.makeOutput(size, x.dtype);\n    const newTexData = this.texData.get(t.dataId);\n    // Copy texture data from the original tensor.\n    Object.assign(newTexData, xTexData);\n    newTexData.shape = size;\n    newTexData.dtype = x.dtype;\n    let flatOffset = slice_util.computeFlatOffset(begin, x.strides);\n    if (xTexData.slice) {\n      // We are slicing an already sliced tensor, so we have to accumulate\n      // the offset.\n      flatOffset += xTexData.slice.flatOffset;\n    }\n    newTexData.slice = {\n      flatOffset,\n      // Point to the original dataId, which is used to do ref counting.\n      origDataId: xTexData.slice && xTexData.slice.origDataId || x.dataId\n    };\n\n    // Increase the ref count for that data bucket.\n    const refCount = this.dataRefCount.get(newTexData.slice.origDataId) || 1;\n    this.dataRefCount.set(newTexData.slice.origDataId, refCount + 1);\n\n    return t;\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x], () => this.cpuBackend.stridedSlice(x, begin, end, strides));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tensor([], outShape) as T;\n    }\n\n    const program = new StridedSliceProgram(begin, strides, outShape);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new ReversePackedProgram(x.shape, axis) :\n        new ReverseProgram(x.shape, axis);\n    return this.compileAndRun(program, [x]);\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    if (tensors[0].dtype === 'complex64') {\n      const reals = tensors.map((t) => real(t));\n      const imags = tensors.map((t) => imag(t));\n      return complex(this.concat(reals, axis), this.concat(imags, axis));\n    }\n\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n    if (tensors.length > env().getNumber('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.concat(tensors.slice(0, midIndex), axis);\n      const rightSide = this.concat(tensors.slice(midIndex), axis);\n      return this.concat([leftSide, rightSide], axis);\n    }\n    if (env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') && tensors[0].rank > 1) {\n      const program = new ConcatPackedProgram(tensors.map(t => t.shape), axis);\n      return this.compileAndRun(program, tensors);\n    }\n    // Any concat of n-dimensional tensors across any axis can be reduced to\n    // a concatenation of two-dimensional tensors across the axis 1 by first\n    // partitioning the axes of the original tensors into those less than the\n    // axis to be concatenated and the rest. Then reshape the tensors\n    // into a two-dimensional tensor by collapsing these two sets of axes and\n    // concatenate the resulting matrices across the axis 1, finally reshaping\n    // the result to have the proper shape.\n    const outShape =\n        backend_util.computeOutShape(tensors.map(t => t.shape), axis);\n    const tensors2D =\n        tensors.map(t => t.as2D(-1, util.sizeFromShape(t.shape.slice(axis))));\n    const program = new ConcatProgram(tensors2D.map(t => t.shape));\n    const res: Tensor = this.compileAndRun(program, tensors2D);\n    return res.reshape(outShape);\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    const cpuRes = this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.neg(x));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.NEG, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.NEG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const [batch, , ] = a.shape;\n\n    // Since the matrices are vectors, it is faster to call mul().sum()\n    // because sum() is O(sqrt(N)) due to divide-and-conquer.\n    if ((outerShapeA === 1 || outerShapeB === 1) &&\n        sharedDim > MATMUL_SHARED_DIM_THRESHOLD) {\n      if (transposeA) {\n        a = transpose(a, [0, 2, 1]);\n      }\n      if (transposeB) {\n        b = transpose(b, [0, 2, 1]);\n      }\n\n      const a3D = outerShapeB === 1 ? a : a.as3D(batch, sharedDim, 1);\n      const axis = outerShapeB === 1 ? 2 : 1;\n      const b3D = outerShapeB === 1 ? b.as3D(batch, 1, sharedDim) : b;\n      return this.multiply(a3D, b3D).sum(axis, true /* keepDims */);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB);\n    return this.compileAndRun<Tensor3D>(program, [a, b], dtype);\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const [batch, , ] = a.shape;\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB,\n        hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [a, b];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun<Tensor3D>(program, inputs, dtype);\n  }\n\n  multiply(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64') {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n\n      const realProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.REAL, a.shape, b.shape);\n      const imagProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.IMAG, a.shape, b.shape);\n\n      const inputs = [\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.imag),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.imag)\n      ];\n      const real = this.compileAndRun<Tensor>(realProgram, inputs);\n      const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n\n      const complex = this.complex(real, imag);\n      real.dispose();\n      imag.dispose();\n      return complex;\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = multiplyImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.MUL, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.MUL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], a.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n        new LRNPackedProgram(x.shape, radius, bias, alpha, beta) :\n        new LRNProgram(x.shape, radius, bias, alpha, beta);\n    return this.compileAndRun(program, [x]);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program =\n        new LRNGradProgram(inputImage.shape, depthRadius, bias, alpha, beta);\n    return this.compileAndRun(program, [inputImage, outputImage, dy]);\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    if (x.dtype === 'string') {\n      const data = this.readSync(x.dataId) as Uint8Array[];\n      const decodedData = data.map(d => util.decodeString(d));\n      const buf = buffer(x.shape, x.dtype, decodedData);\n      return tile(buf, reps) as T;\n    }\n    const program = new TileProgram(x.shape, reps);\n    return this.compileAndRun(program, [x]);\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new PadPackedProgram(x.shape, paddings, constantValue) :\n        new PadProgram(x.shape, paddings, constantValue);\n    return this.compileAndRun(program, [x]);\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x, indices], () => this.cpuBackend.gather(x, indices, axis));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = new GatherProgram(x.shape, indices.size, axis);\n    return this.compileAndRun(program, [x, indices]);\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'batchToSpaceND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: Array<[number, number]>): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'spaceToBatchND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const completePaddings: Array<[number, number]> = [[0, 0]];\n    completePaddings.push(...paddings);\n    for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n      completePaddings.push([0, 0]);\n    }\n\n    const paddedX = x.pad(completePaddings);\n\n    const reshapedPaddedShape =\n        backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n    const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n        reshapedPaddedShape.length, blockShape.length, false);\n\n    const flattenShape = backend_util.getReshapedPermuted(\n        paddedX.shape, blockShape, prod, false);\n\n    const paddedXT = transpose(\n        paddedX.reshape(reshapedPaddedShape),\n        permutedReshapedPaddedPermutation);\n    return reshape(paddedXT, flattenShape) as T;\n  }\n\n  private reduce(\n      x: Tensor2D, reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod',\n      dtype: DataType): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const outSize = Math.ceil(inSize / windowSize);\n    const reduceInfo = {windowSize, inSize, batchSize, outSize};\n    const program = new ReduceProgram(reduceInfo, reduceType);\n    const output = this.compileAndRun<Tensor2D>(program, [x], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.reduce(output, reduceType, dtype);\n  }\n\n  private argReduce(\n      x: Tensor2D, reduceType: 'max'|'min',\n      bestIndicesA: Tensor2D = null): Tensor2D {\n    let batchSize = x.shape[0];\n    let inSize = x.shape[1];\n    if (bestIndicesA != null) {\n      batchSize = bestIndicesA.shape[0];\n      inSize = bestIndicesA.shape[1];\n    }\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const reduceInfo = {\n      windowSize,\n      inSize,\n      batchSize,\n      outSize: Math.ceil(inSize / windowSize)\n    };\n    const program =\n        new ArgMinMaxProgram(reduceInfo, reduceType, bestIndicesA == null);\n    const inputs = [x];\n    if (bestIndicesA != null) {\n      inputs.push(bestIndicesA);\n    }\n    const output = this.compileAndRun<Tensor2D>(program, inputs, 'int32');\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.argReduce(x, reduceType, output);\n  }\n\n  private argReducePacked(\n      x: Tensor, reduceType: 'max'|'min', bestIndicesA: Tensor = null): Tensor {\n    const inShape = bestIndicesA != null ? bestIndicesA.shape : x.shape;\n    const inSize = inShape[inShape.length - 1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const program = new ArgMinMaxPackedProgram(\n        inShape, windowSize, reduceType, bestIndicesA == null);\n    const inputs = bestIndicesA == null ? [x] : [x, bestIndicesA];\n    const output = this.compileAndRun<Tensor>(program, inputs, 'int32');\n    if (output.rank === x.rank) {\n      return this.argReducePacked(x, reduceType, output);\n    }\n    return output;\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'sum', outputDType).reshape(outShape);\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.prod(x, axes));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'prod', outputDType).reshape(outShape);\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    let axis = 0;\n    const permutation = backend_util.getAxesPermutation([axis], x.rank);\n    let permutedX = x;\n    if (permutation != null) {\n      permutedX = transpose(x, permutation);\n      axis = backend_util.getInnerMostAxes(1, x.rank)[0];\n    }\n\n    const outShape =\n        segment_util.computeOutShape(permutedX.shape, axis, numSegments);\n    const inSize = util.sizeFromShape([permutedX.shape[axis]]);\n    const a2D = permutedX.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    let result =\n        this.segOpCompute(\n                a2D, 'unsortedSegmentSum', segmentIds, outputDType, numSegments)\n            .reshape(outShape);\n    if (permutation != null) {\n      result =\n          transpose(result, backend_util.getUndoAxesPermutation(permutation));\n    }\n    return result;\n  }\n\n  private segOpCompute(\n      x: Tensor2D, segOpType: 'unsortedSegmentSum', segmentIds: Tensor1D,\n      dtype: DataType, numSegments: number): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize =\n        segment_util.segOpComputeOptimalWindowSize(inSize, numSegments);\n    const segOpInfo = {windowSize, inSize, batchSize, numSegments};\n    const program = new SegmentOpProgram(segOpInfo, segOpType);\n    const output =\n        this.compileAndRun<Tensor2D>(program, [x, segmentIds], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === numSegments) {\n      return output;\n    }\n    segmentIds = range(0, numSegments).tile([inSize / windowSize]);\n    return this.segOpCompute(output, segOpType, segmentIds, dtype, numSegments);\n  }\n\n  private argMinMaxReduce(x: Tensor, axis: number, reduceType: 'min'|'max'):\n      Tensor {\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims(\n        'arg' + reduceType.charAt(0).toUpperCase() + reduceType.slice(1), axes,\n        x.rank);\n    if (!env().getBool('WEBGL_PACK_REDUCE') || x.rank <= 2) {\n      const [outShape, reduceShape] =\n          backend_util.computeOutAndReduceShapes(x.shape, axes);\n      const inSize = util.sizeFromShape(reduceShape);\n      const a2D = x.as2D(-1, inSize);\n      return this.argReduce(a2D, reduceType).reshape(outShape);\n    }\n    return this.argReducePacked(x, reduceType);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'min');\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'max');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `WebGL cumsum shader expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const size = x.shape[axis];\n    let result = x;\n    // Use cumsum parallel algorithm, ref:\n    // https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-39-parallel-prefix-sum-scan-cuda\n    for (let i = 0; i <= Math.ceil(Math.log2(size)) - 1; i++) {\n      const program = new CumSumProgram(x.shape, false, reverse);\n      const customSetup = program.getCustomSetupFunc(i);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result], result.dtype, customSetup);\n      prevResult.dispose();\n    }\n    // For exclusive cumsum, shift the end result in the direction of sum and\n    // add 0 to the front index.\n    if (exclusive) {\n      const program = new CumSumProgram(x.shape, exclusive, reverse);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result]);\n      prevResult.dispose();\n    }\n\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.EQUAL, 'bool');\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.NOT_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.NOT_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.less(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.LESS, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LESS_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.greater(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.GREATER, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.GREATER, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.GREATER_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.GREATER_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalNot<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOGICAL_NOT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_AND, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_AND, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_OR, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_OR, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    const program = new SelectProgram(condition.rank, a.shape, a.rank);\n    return this.compileAndRun(\n        program, [condition, a, b], upcastType(a.dtype, b.dtype));\n  }\n\n  where(condition: Tensor): Tensor2D {\n    backend_util.warn(\n        'tf.where() in webgl locks the UI thread. ' +\n        'Call tf.whereAsync() instead');\n    const condVals = condition.dataSync();\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    const xVals = x.dataSync();\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'min', a2D.dtype).reshape(outShape);\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.minimum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MIN, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MIN, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MOD, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MOD, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.maximum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MAX, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MAX, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'all', a2D.dtype).reshape(outShape);\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'any', a2D.dtype).reshape(outShape);\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    const op = binaryop_gpu.INT_DIV;\n    const outputDtype = 'int32';\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.INT_DIV, outputDtype);\n    }\n    const program = new BinaryOpProgram(op, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], outputDtype);\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.ADD);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = addImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.ADD, dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.ADD, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  private packedUnaryOp(x: TensorInfo, op: string, dtype: DataType) {\n    const program = new UnaryOpPackedProgram(x.shape, op);\n    return this.compileAndRun<Tensor>(program, [x], dtype);\n  }\n\n  private packedBinaryOp(\n      a: TensorInfo, b: TensorInfo, op: string, dtype: DataType,\n      checkOutOfBounds = false) {\n    const program =\n        new BinaryOpPackedProgram(op, a.shape, b.shape, checkOutOfBounds);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  /**\n   * Computes a complex binary operation that can be decomposed into a simple\n   * binary operation on both the real and imagary parts.\n   */\n  private complexSeparableBinaryOp(a: Tensor, b: Tensor, op: string): Tensor {\n    const aData = this.texData.get(a.dataId);\n    const bData = this.texData.get(b.dataId);\n\n    const [real, imag] = [\n      [aData.complexTensors.real, bData.complexTensors.real],\n      [aData.complexTensors.imag, bData.complexTensors.imag]\n    ].map(complexParts => {\n      const [aPart, bPart] = complexParts;\n\n      const aHandle = this.makeComplexComponentTensorInfo(a, aPart);\n      const bHandle = this.makeComplexComponentTensorInfo(b, bPart);\n\n      const program = new BinaryOpProgram(op, a.shape, b.shape);\n      return this.compileAndRun<Tensor>(\n          program, [aHandle, bHandle], upcastType(aPart.dtype, bPart.dtype));\n    });\n\n    const complex = this.complex(real, imag);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  // Returns a TensorInfo with the complex shape and the dataId of the\n  // underlying part. We need to do this because a reshaped complex tensor is\n  // not reflected in its parts.\n  private makeComplexComponentTensorInfo(\n      complexTensor: Tensor, complexPart: Tensor): TensorInfo {\n    return {\n      dataId: complexPart.dataId,\n      dtype: complexPart.dtype,\n      shape: complexTensor.shape\n    };\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n\n    // Limit the number of uploaded textures for optimization.\n    if (tensors.length > env().get('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.addN(tensors.slice(0, midIndex));\n      const rightSide = this.addN(tensors.slice(midIndex));\n      return this.addN([leftSide, rightSide]);\n    }\n\n    const dtype =\n        tensors.map(t => t.dtype).reduce((d1, d2) => upcastType(d1, d2));\n    const shapes = tensors.map(t => t.shape);\n    // We can make sure shapes are identical in op level.\n    const usePackedOp = env().getBool('WEBGL_PACK');\n    const program = usePackedOp ?\n        new AddNPackedProgram(tensors[0].shape, shapes) :\n        new AddNProgram(tensors[0].shape, shapes);\n    return this.compileAndRun<T>(program, tensors, dtype);\n  }\n\n  subtract(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.SUB);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = subImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.SUB, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.SUB, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    const usePackedOp = env().getBool('WEBGL_PACK_BINARY_OPERATIONS');\n    const program = usePackedOp ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.POW, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.POW, a.shape, b.shape);\n    const dtype = upcastType(a.dtype, b.dtype);\n    return this.compileAndRun<T>(program, [a, b], dtype);\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          ceilImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.CEIL, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.CEIL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  floor<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = floorImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.FLOOR, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.FLOOR);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_NAN);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_INF);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_FINITE);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n\n  round<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ROUND);\n    return this.compileAndRun(program, [x]);\n  }\n\n  exp<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          expImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXP, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXP);\n    return this.compileAndRun(program, [x]);\n  }\n\n  expm1<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = expm1ImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXPM1, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXPM1);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n    const a = this.subtract(logits, maxLogit.reshape(expandedShape));\n    const b = this.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax kernel\n    // modularization.\n    return div(b, sumExp);\n  }\n\n  log<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          logImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.LOG, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  log1p<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG1P);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sqrt<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  rsqrt<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = rsqrtImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.RSQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reciprocal<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.RECIPROCAL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU6);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU6);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  prelu<T extends Tensor>(x: T, alpha: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.PRELU, x.shape, alpha.shape) :\n        new BinaryOpProgram(binaryop_gpu.PRELU, x.shape, alpha.shape);\n    return this.compileAndRun(program, [x, alpha]);\n  }\n\n  elu<T extends Tensor>(x: T): T {\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.ELU, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.ELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.ELU_DER, dy.shape, y.shape) :\n        new BinaryOpProgram(binaryop_gpu.ELU_DER, dy.shape, y.shape);\n    return this.compileAndRun(program, [dy, y]);\n  }\n\n  selu<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  int<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TO_INT);\n    return this.compileAndRun(program, [x], 'int32');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    let program;\n    if (env().getBool('WEBGL_PACK_CLIP')) {\n      program = new ClipPackedProgram(x.shape);\n    } else {\n      program = new ClipProgram(x.shape);\n    }\n    const customSetup = program.getCustomSetupFunc(min, max);\n    return this.compileAndRun(program, [x], null, customSetup);\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    // TODO: handle cases when x is complex.\n    if (this.shouldExecuteOnCPU([x]) && x.dtype !== 'complex64') {\n      const outValues =\n          simpleAbsImplCPU(this.texData.get(x.dataId).values as TypedArray);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.ABS, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.ABS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  complexAbs<T extends Tensor>(x: T): T {\n    const xData = this.texData.get(x.dataId);\n\n    const program = new ComplexAbsProgram(x.shape);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    return this.compileAndRun<Tensor>(program, inputs) as T;\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGMOID);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SOFTPLUS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASIN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acos<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atan<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATAN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.COSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  tanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ERF);\n    return this.compileAndRun(program, [x]);\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.STEP(alpha));\n    return this.compileAndRun(program, [x]);\n  }\n\n  private conv2dByMatMul(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Reshapes conv2D input to 2D tensors, uses matMul and then reshape the\n    // result from 2D to 4D.\n    const xShape = x.shape;\n    const xTexData = this.texData.get(x.dataId);\n    const sharedMatMulDim = convInfo.inChannels;\n    const outerShapeX = xShape[0] * xShape[1] * xShape[2];\n    const outerShapeFilter = convInfo.outChannels;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const transposeA = false;\n    const transposeB = false;\n\n    // TODO: Once reduction ops are packed, batchMatMul will always be packed\n    // and we can remove this condition.\n    const batchMatMulWillBeUnpacked =\n        (outerShapeX === 1 || outerShapeFilter === 1) &&\n        sharedMatMulDim > MATMUL_SHARED_DIM_THRESHOLD;\n    const reshapeWillBeExpensive = xShape[2] % 2 !== 0 && !!xTexData.isPacked;\n\n    if (batchMatMulWillBeUnpacked || !env().getBool('WEBGL_LAZILY_UNPACK') ||\n        !env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ||\n        !reshapeWillBeExpensive) {\n      const targetShape = isChannelsLast ? xShape[0] * xShape[1] * xShape[2] :\n                                           xShape[0] * xShape[2] * xShape[3];\n      const xReshaped = reshape(x, [1, targetShape, convInfo.inChannels]);\n      const filterReshaped =\n          reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n      const result = this.fusedBatchMatMul({\n        a: xReshaped as Tensor3D,\n        b: filterReshaped as Tensor3D,\n        transposeA,\n        transposeB,\n        bias,\n        activation,\n        preluActivationWeights\n      });\n      return reshape(result, convInfo.outShape);\n    }\n\n    // Following optimization is specific to packed |x| with odd row count\n    // (For example, in channelLast mode, 'row count' refers to x.shape[2]):\n    // we avoid expensive packed 2x2 reshape by padding row count to next,\n    // even number. When x.shape[2] is odd, the result of packed batchMatMul is\n    // the same (has the same texture layout and and values in the texture) as\n    // it is for even x.shape[2] + 1. We make the odd-rows tensor to look like\n    // even-rows tensor before the operation and, after the batchMatMul,\n    // fix the even-rows result to have odd number of rows.\n    const targetShape = isChannelsLast ?\n        xShape[0] * xShape[1] * (xShape[2] + 1) :\n        xShape[0] * xShape[2] * (xShape[3] + 1);\n    const xReshaped: TensorInfo = {\n      dataId: x.dataId,\n      shape: [1, targetShape, convInfo.inChannels],\n      dtype: x.dtype\n    };\n    // xTexData.shape gets referenced from GPGPUBinary.inShapeInfos.\n    // Decrementing row count, after batchMatMul->...->compileProgram leads to\n    // invalid row count within the reference in GPGPUBinary.inShapeInfos.\n    // Alternative fix would be to provide a copy to GPGPUBinary.inShapeInfos\n    // in compileProgram method, but that would affect compilation of all\n    // programs - instead, provide a copy here, with even row count, before\n    // calling batchMatMul->...->compileProgram and after that, the original\n    // xTexData.shape is restored.\n    const originalXTexDataShape = xTexData.shape;\n    xTexData.shape = xTexData.shape.slice();\n    xTexData.shape[xTexData.shape.length - 2]++;\n    util.assert(\n        webgl_util.isReshapeFree(xTexData.shape, xReshaped.shape),\n        () => `packed reshape ${xTexData.shape} to ${\n            xReshaped.shape} isn't free`);\n    const filterReshaped =\n        reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n    const pointwiseConv = this.fusedBatchMatMul({\n      a: xReshaped as Tensor3D,\n      b: filterReshaped as Tensor3D,\n      transposeA,\n      transposeB,\n      bias,\n      activation,\n      preluActivationWeights\n    });\n    const pointwiseConvTexData = this.texData.get(pointwiseConv.dataId);\n    util.assert(\n        pointwiseConvTexData.isPacked,\n        () => 'batchMatMul result is expected to be packed');\n    // Restore the input shape to original.\n    xTexData.shape = originalXTexDataShape;\n    // Set the output shape - there is no need for expensive reshape as data\n    // layout is already correct.\n    pointwiseConvTexData.shape = convInfo.outShape;\n    return engine().makeTensorFromDataId(\n               pointwiseConv.dataId, convInfo.outShape, pointwiseConv.dtype) as\n        Tensor4D;\n  }\n\n  private conv2dWithIm2Row(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Rearranges conv2d input so each block to be convolved over forms the\n    // column of a new matrix with shape [filterWidth * filterHeight *\n    // inChannels, outHeight * outWidth]. The filter is also rearranged so each\n    // output channel forms a row of a new matrix with shape [outChannels,\n    // filterWidth * filterHeight * inChannels]. The convolution is then\n    // computed by multiplying these matrices and reshaping the result.\n    const {\n      filterWidth,\n      filterHeight,\n      inChannels,\n      outWidth,\n      outHeight,\n      dataFormat\n    } = convInfo;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n\n    const sharedDim = filterWidth * filterHeight * inChannels;\n    const numCols = outHeight * outWidth;\n    const x2ColShape = [sharedDim, numCols];\n    const transposeA = true;\n    const transposeB = false;\n\n    const xSqueezed = x.squeeze([0]);\n    const w2Row = filter.reshape([1, sharedDim, -1]);\n\n    const im2ColProgram =\n        new Im2ColPackedProgram(x2ColShape, xSqueezed.shape, convInfo);\n    const im2Col: Tensor3D =\n        this.compileAndRun<Tensor2D>(im2ColProgram, [xSqueezed]).reshape([\n          1, x2ColShape[0], x2ColShape[1]\n        ]);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const matmulProgram = new MatMulPackedProgram(\n        im2Col.shape, [1, numCols, convInfo.outChannels], transposeA,\n        transposeB, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [im2Col, w2Row];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    const product = this.compileAndRun<Tensor4D>(matmulProgram, inputs);\n\n    if (isChannelsLast) {\n      return product.reshape([1, outHeight, outWidth, convInfo.outChannels]);\n    } else {\n      return product.reshape([1, convInfo.outChannels, outHeight, outWidth]);\n    }\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && input.shape[0] === 1) {\n      return this.conv2dWithIm2Row(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, false) : null;\n    const program = new Conv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [input, filter];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun(program, inputs);\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(x, filter, convInfo);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && x.shape[0] === 1) {\n      return this.conv2dWithIm2Row(x, filter, convInfo);\n    }\n    const program = new Conv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new Conv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    const program = new Conv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    const shouldPackDepthwiseConv = env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1;\n    const fusedActivation = activation ?\n        mapActivationToShaderProgram(activation, shouldPackDepthwiseConv) :\n        null;\n    const inputs: Tensor[] = [input, filter];\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    if (hasBias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (shouldPackDepthwiseConv) {\n      program = new DepthwiseConvPacked2DProgram(\n          convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n      return this.compileAndRun(program, inputs);\n    }\n\n    program = new DepthwiseConv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    return this.compileAndRun(program, inputs);\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1) {\n      program = new DepthwiseConvPacked2DProgram(convInfo);\n      return this.compileAndRun(program, [x, filter]);\n    }\n\n    program = new DepthwiseConv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Conv3DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return backend_util.castTensor(x, dtype, this);\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = this.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'avg', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const avgPool3dBackpropProgram = new AvgPool3DBackpropProgram(convInfo);\n    return this.compileAndRun(avgPool3dBackpropProgram, [dy], x.dtype);\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'max', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const getPositions = true;\n    const maxPool3dPositionsProgram =\n        new Pool3DProgram(convInfo, 'max', getPositions);\n    const maxPool3dPositions: Tensor5D =\n        this.compileAndRun(maxPool3dPositionsProgram, [x]);\n    const maxPool3dBackPropProgram = new MaxPool3DBackpropProgram(convInfo);\n    const result = this.compileAndRun(\n        maxPool3dBackPropProgram, [dy, maxPool3dPositions], x.dtype);\n    maxPool3dPositions.dispose();\n    return result as Tensor5D;\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_IMAGE_OPERATIONS') ?\n        new ResizeBilinearPackedProgram(\n            x.shape, newHeight, newWidth, alignCorners) :\n        new ResizeBilinearProgram(x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    const program = new ResizeBilinearBackpropProgram(dy, x, alignCorners);\n\n    return this.compileAndRun(program, [dy]);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = new ResizeNearestNeighborProgram(\n        x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x]);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    const program =\n        new ResizeNearestNeigborBackpropProgram(dy, x, alignCorners);\n    return this.compileAndRun(program, [dy]);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    const probs = normalized ? logits : softmax(logits);\n    const batchSize = probs.shape[0];\n    const numOutcomes = probs.shape[1];\n    const program = new MultinomialProgram(batchSize, numOutcomes, numSamples);\n    const customSetup = program.getCustomSetupFunc(seed);\n    return this.compileAndRun(program, [probs], 'int32', customSetup);\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    const program = new OneHotProgram(indices.size, depth, onValue, offValue);\n    return this.compileAndRun(program, [indices]);\n  }\n\n  diag(x: Tensor): Tensor {\n    const program = new DiagProgram(x.size);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    const program = new CropAndResizeProgram(\n        image.shape, boxes.shape, cropSize, method, extrapolationValue);\n    return this.compileAndRun(program, [image, boxes, boxIndex], 'float32');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = (dataFormat === 'NHWC') ? x.shape[1] : x.shape[2];\n    const inputWidth = (dataFormat === 'NHWC') ? x.shape[2] : x.shape[3];\n    const inputDepth = (dataFormat === 'NHWC') ? x.shape[3] : x.shape[1];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const outputShape = (dataFormat === 'NHWC') ?\n        [batchSize, outputHeight, outputWidth, outputDepth] :\n        [batchSize, outputDepth, outputHeight, outputWidth];\n\n    const program = new DepthToSpaceProgram(outputShape, blockSize, dataFormat);\n    return this.compileAndRun(program, [x]);\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n    const flattenIndices = indices.reshape([numUpdates, sliceRank]);\n    const flattenX = updates.reshape([numUpdates, sliceSize]);\n\n    if (outputSize === 0) {\n      return backend_util.reshapeTensor(tensor([]), shape);\n    }\n    const defaultValue = scalar(0);\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, flattenIndices.rank, flattenX.rank, strides,\n        flattenShape);\n    const res: Tensor =\n        this.compileAndRun(program, [flattenX, flattenIndices, defaultValue]);\n    return res.reshape(shape);\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n\n    const sumDupeIndices = false;\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, sparseIndices.rank, sparseValues.rank, strides,\n        [outputSize, 1], sumDupeIndices);\n    const res: Tensor = this.compileAndRun(\n        program, [sparseValues, sparseIndices, defaultValue]);\n    return res.reshape(outputShape);\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    const inverse = false;\n    return this.fftImpl(x, inverse);\n  }\n\n  ifft(x: Tensor2D): Tensor2D {\n    const inverse = true;\n    return this.fftImpl(x, inverse);\n  }\n\n  private fftImpl(x: Tensor2D, inverse: boolean): Tensor2D {\n    const xData = this.texData.get(x.dataId);\n\n    const realProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.REAL, x.shape, inverse);\n    const imagProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.IMAG, x.shape, inverse);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    const real = this.compileAndRun<Tensor>(realProgram, inputs);\n    const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n    const complex = this.complex(real, imag).as2D(x.shape[0], x.shape[1]);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n\n    const flattenIndices = indices.reshape([numSlices, sliceRank]);\n    const flattenX = x.reshape([x.size / sliceSize, sliceSize]);\n    const program =\n        new GatherNDProgram(sliceRank, strides, [numSlices, sliceSize]);\n    const res: Tensor = this.compileAndRun(program, [flattenX, flattenIndices]);\n    return res.reshape(resultShape);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n\n    if (dtype === 'string') {\n      // String type should be handled in CPU memory.\n      const values = util.getArrayFromDType(dtype, util.sizeFromShape(shape));\n      values.fill(value as string);\n      return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n    } else {\n      const program = new FillProgram(shape, value as number);\n      const customSetup = program.getCustomSetupFunc(value as number);\n      return this.compileAndRun(program, [], dtype, customSetup);\n    }\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported under string dtype');\n    } else {\n      // TODO(cais, smilkov): Add WebGL shader for onesLike:\n      //   https://github.com/tensorflow/tfjs/issues/1293\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return this.fill(x.shape, x.dtype === 'string' ? '' : 0, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    // TODO: Use CPU implementation due to the precision problem in Safari.\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  makeTensorInfo(shape: number[], dtype: DataType, values?: BackendValues):\n      TensorInfo {\n    const dataId = this.write(values, shape, dtype);\n    this.texData.get(dataId).usage = null;\n    return {dataId, shape, dtype};\n  }\n\n  private makeOutput<T extends Tensor>(\n      shape: number[], dtype: DataType, values?: BackendValues): T {\n    const {dataId} = this.makeTensorInfo(shape, dtype, values);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  private unpackTensor(input: TensorInfo): TensorInfo {\n    const program = new UnpackProgram(input.shape);\n    return this.runWebGLProgram(program, [input], input.dtype);\n  }\n\n  private packTensor(input: TensorInfo): TensorInfo {\n    const program = new PackProgram(input.shape);\n    const preventEagerUnpackingOutput = true;\n    return this.runWebGLProgram(\n        program, [input], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOutput);\n  }\n\n  private packedReshape(input: TensorInfo, afterShape: number[]): TensorInfo {\n    const input3DShape = [\n      webgl_util.getBatchDim(input.shape),\n      ...webgl_util.getRowsCols(input.shape)\n    ] as [number, number, number];\n    const input3D: TensorInfo = {\n      dtype: input.dtype,\n      shape: input3DShape,\n      dataId: input.dataId\n    };\n    const afterShapeAs3D = [\n      webgl_util.getBatchDim(afterShape), ...webgl_util.getRowsCols(afterShape)\n    ] as [number, number, number];\n\n    const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n    const preventEagerUnpackingOfOutput = true;\n    const output = this.runWebGLProgram(\n        program, [input3D], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOfOutput);\n    return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n  }\n\n  private decode(dataId: DataId): TensorInfo {\n    const texData = this.texData.get(dataId);\n    const {isPacked, shape, dtype} = texData;\n    const shapeAs3D =\n        webgl_util.getShapeAs3D(shape) as [number, number, number];\n    let program;\n    if (isPacked) {\n      program = new DecodeMatrixPackedProgram(shapeAs3D);\n    } else {\n      program = new DecodeMatrixProgram(shapeAs3D);\n    }\n    const preventEagerUnpackingOfOutput = true;\n    const out = this.runWebGLProgram(\n        program, [{shape: shapeAs3D, dtype, dataId}], dtype,\n        null /* customSetup */, preventEagerUnpackingOfOutput);\n    return {dtype, shape, dataId: out.dataId};\n  }\n\n  runWebGLProgram(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): TensorInfo {\n    const output = this.makeTensorInfo(program.outputShape, outputDtype);\n    const outData = this.texData.get(output.dataId);\n    if (program.packedOutput) {\n      outData.isPacked = true;\n    }\n    if (program.outPackingScheme === tex_util.PackingScheme.DENSE) {\n      const texelShape = tex_util.getDenseTexShape(program.outputShape);\n      // For a densely packed output, we explicitly set texShape\n      // so it doesn't get assigned later according to our typical packing\n      // scheme wherein a single texel can only contain values from adjacent\n      // rows/cols.\n      outData.texShape = texelShape.map(d => d * 2) as [number, number];\n    }\n    if (program.outTexUsage != null) {\n      outData.usage = program.outTexUsage;\n    }\n    if (util.sizeFromShape(output.shape) === 0) {\n      // Short-circuit the computation since the result is empty (has 0 in its\n      // shape).\n      outData.values =\n          util.getTypedArrayFromDType(output.dtype as 'float32', 0);\n      return output;\n    }\n\n    const dataToDispose: TensorInfo[] = [];\n    const inputsData: TensorData[] = inputs.map(input => {\n      if (input.dtype === 'complex64') {\n        throw new Error(\n            `GPGPUProgram does not support complex64 input. For complex64 ` +\n            `dtypes, please separate the program into real and imaginary ` +\n            `parts.`);\n      }\n\n      let texData = this.texData.get(input.dataId);\n\n      if (texData.texture == null) {\n        if (!program.packedInputs &&\n            util.sizeFromShape(input.shape) <=\n                env().getNumber('WEBGL_SIZE_UPLOAD_UNIFORM')) {\n          // Upload small tensors that live on the CPU as uniforms, not as\n          // textures. Do this only when the environment supports 32bit floats\n          // due to problems when comparing 16bit floats with 32bit floats.\n          // TODO(https://github.com/tensorflow/tfjs/issues/821): Make it\n          // possible for packed shaders to sample from uniforms.\n          return {\n            shape: input.shape,\n            texData: null,\n            isUniform: true,\n            uniformValues: texData.values as TypedArray\n          };\n        }\n\n        // This ensures that if a packed program's inputs have not yet been\n        // uploaded to the GPU, they get uploaded as packed right off the bat.\n        if (program.packedInputs) {\n          texData.isPacked = true;\n          texData.shape = input.shape;\n        }\n      } else if (!!texData.isPacked !== !!program.packedInputs) {\n        input = texData.isPacked ? this.unpackTensor(input) :\n                                   this.packTensor(input);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n      } else if (\n          texData.isPacked &&\n          !webgl_util.isReshapeFree(texData.shape, input.shape)) {\n        // This is a special case where a texture exists for a tensor\n        // but the shapes are incompatible (due to packing constraints) because\n        // the tensor did not have a chance to go through the packed reshape\n        // shader. This only happens when we reshape the *same* tensor to form\n        // *distinct* inputs to an op, e.g. dotting a vector with itself. This\n        // case will disappear once packed uploading is the default.\n\n        const savedInput = input;\n        const targetShape = input.shape;\n\n        input.shape = texData.shape;\n        input = this.packedReshape(input as Tensor, targetShape);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n\n        savedInput.shape = targetShape;\n      }\n\n      this.uploadToGPU(input.dataId);\n      return {shape: input.shape, texData, isUniform: false};\n    });\n\n    this.uploadToGPU(output.dataId);\n    const outputData:\n        TensorData = {shape: output.shape, texData: outData, isUniform: false};\n    const key = gpgpu_math.makeShaderKey(program, inputsData, outputData);\n    const binary = this.getAndSaveBinary(key, () => {\n      return gpgpu_math.compileProgram(\n          this.gpgpu, program, inputsData, outputData);\n    });\n    const shouldTimeProgram = this.activeTimers != null;\n    let query: WebGLQuery|CPUTimerQuery;\n    if (shouldTimeProgram) {\n      query = this.startTimer();\n    }\n\n    gpgpu_math.runProgram(\n        this.gpgpu, binary, inputsData, outputData, customSetup);\n\n    dataToDispose.forEach(info => this.disposeIntermediateTensorInfo(info));\n\n    if (shouldTimeProgram) {\n      query = this.endTimer(query);\n      this.activeTimers.push(\n          {name: program.constructor.name, query: this.getQueryTime(query)});\n    }\n\n    if (!env().getBool('WEBGL_LAZILY_UNPACK') && outData.isPacked &&\n        preventEagerUnpackingOfOutput === false) {\n      const unpacked = this.unpackTensor(output);\n      this.disposeIntermediateTensorInfo(output);\n      return unpacked;\n    }\n    return output;\n  }\n\n  compileAndRun<K extends TensorInfo>(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype?: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): K {\n    outputDtype = outputDtype || inputs[0].dtype;\n    const outInfo = this.runWebGLProgram(\n        program, inputs, outputDtype, customSetup,\n        preventEagerUnpackingOfOutput);\n    return engine().makeTensorFromDataId(\n               outInfo.dataId, outInfo.shape, outInfo.dtype) as {} as K;\n  }\n\n  private getAndSaveBinary(key: string, getBinary: () => GPGPUBinary):\n      GPGPUBinary {\n    if (!(key in this.binaryCache)) {\n      this.binaryCache[key] = getBinary();\n    }\n    return this.binaryCache[key];\n  }\n\n  getTextureManager(): TextureManager {\n    return this.textureManager;\n  }\n\n  private disposed = false;\n\n  dispose() {\n    if (this.disposed) {\n      return;\n    }\n    // Avoid disposing the compiled webgl programs during unit testing because\n    // it slows down test execution.\n    if (!env().getBool('IS_TEST')) {\n      const allKeys = Object.keys(this.binaryCache);\n      allKeys.forEach(key => {\n        this.gpgpu.deleteProgram(this.binaryCache[key].webGLProgram);\n        delete this.binaryCache[key];\n      });\n    }\n    this.textureManager.dispose();\n    if (this.canvas != null &&\n        (typeof (HTMLCanvasElement) !== 'undefined' &&\n         this.canvas instanceof HTMLCanvasElement)) {\n      this.canvas.remove();\n    } else {\n      this.canvas = null;\n    }\n    if (this.gpgpuCreatedLocally) {\n      this.gpgpu.program = null;\n      this.gpgpu.dispose();\n    }\n    this.disposed = true;\n  }\n\n  floatPrecision(): 16|32 {\n    if (this.floatPrecisionValue == null) {\n      this.floatPrecisionValue = tidy(() => {\n        if (!env().get('WEBGL_RENDER_FLOAT32_ENABLED')) {\n          // Momentarily switching DEBUG flag to false so we don't throw an\n          // error trying to upload a small value.\n          const debugFlag = env().getBool('DEBUG');\n          env().set('DEBUG', false);\n          const underflowCheckValue = this.abs(scalar(1e-8)).dataSync()[0];\n          env().set('DEBUG', debugFlag);\n\n          if (underflowCheckValue > 0) {\n            return 32;\n          }\n        }\n        return 16;\n      });\n    }\n    return this.floatPrecisionValue;\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  private uploadToGPU(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    const {shape, dtype, values, texture, usage, isPacked} = texData;\n\n    if (texture != null) {\n      // Array is already on GPU. No-op.\n      return;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let texShape = texData.texShape;\n    if (texShape == null) {\n      texShape = webgl_util.getTextureShapeFromLogicalShape(shape, isPacked);\n      texData.texShape = texShape;\n    }\n\n    if (values != null) {\n      const shapeAs3D = webgl_util.getShapeAs3D(shape);\n\n      let program;\n      let width = texShape[1], height = texShape[0];\n      const isByteArray = values instanceof Uint8Array;\n\n      if (isPacked) {\n        [width, height] = tex_util.getPackedMatrixTextureShapeWidthHeight(\n            texShape[0], texShape[1]);\n        program = new EncodeMatrixPackedProgram(\n            shapeAs3D, [height, width], isByteArray);\n      } else {\n        program =\n            new EncodeMatrixProgram(shapeAs3D, [height, width], isByteArray);\n      }\n\n      const tempDenseInputHandle = this.makeTensorInfo([height, width], dtype);\n      if (isByteArray) {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.PIXELS;\n      } else {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.UPLOAD;\n      }\n      this.gpgpu.uploadDenseMatrixToTexture(\n          this.getTexture(tempDenseInputHandle.dataId), width, height,\n          values as TypedArray);\n\n      // We want the output to remain packed regardless of the value of\n      // WEBGL_PACK.\n      const preventEagerUnpacking = true;\n      const encodedOutputTarget = this.runWebGLProgram(\n          program, [tempDenseInputHandle], dtype, null, preventEagerUnpacking);\n\n      // Have the original texture assume the identity of the encoded output.\n      const outputTexData = this.texData.get(encodedOutputTarget.dataId);\n      texData.texture = outputTexData.texture;\n      texData.texShape = outputTexData.texShape;\n      texData.isPacked = outputTexData.isPacked;\n      texData.usage = outputTexData.usage;\n\n      this.disposeIntermediateTensorInfo(tempDenseInputHandle);\n      this.texData.delete(encodedOutputTarget.dataId);\n\n      // Once uploaded, don't store the values on cpu.\n      texData.values = null;\n      if (shouldTimeProgram) {\n        this.uploadWaitMs += util.now() - start;\n      }\n    } else {\n      const newTexture = this.acquireTexture(texShape, usage, dtype, isPacked);\n      texData.texture = newTexture;\n    }\n  }\n\n  private convertAndCacheOnCPU(dataId: DataId, float32Values?: Float32Array):\n      TypedArray {\n    const texData = this.texData.get(dataId);\n    const {dtype} = texData;\n\n    this.releaseGPUData(dataId);\n\n    if (float32Values != null) {\n      texData.values = float32ToTypedArray(float32Values, dtype as 'float32');\n    }\n    return texData.values as TypedArray;\n  }\n\n  private acquireTexture(\n      texShape: [number, number], texType: TextureUsage, dtype: DataType,\n      isPacked: boolean): WebGLTexture {\n    this.numBytesInGPU += this.computeBytes(texShape, dtype);\n    if (!this.warnedAboutMemory &&\n        this.numBytesInGPU > this.numMBBeforeWarning * 1024 * 1024) {\n      const mb = (this.numBytesInGPU / 1024 / 1024).toFixed(2);\n      this.warnedAboutMemory = true;\n      console.warn(\n          `High memory usage in GPU: ${mb} MB, ` +\n          `most likely due to a memory leak`);\n    }\n    return this.textureManager.acquireTexture(texShape, texType, isPacked);\n  }\n\n  private computeBytes(shape: [number, number], dtype: DataType) {\n    return shape[0] * shape[1] * util.bytesPerElement(dtype);\n  }\n\n  private tryRunOnCpuOrThrow<T extends Tensor>(\n      inputs: TensorInfo[], fn: () => T): T|null {\n    if (this.shouldExecuteOnCPU(inputs)) {\n      try {\n        return fn();\n      } catch (e) {\n        if (env().getBool('IS_TEST')) {\n          throw new Error('CPU forwarding failed');\n        }\n      }\n    }\n    return null;\n  }\n}\n\nfunction float32ToTypedArray<D extends NumericDataType>(\n    a: Float32Array, dtype: D): tf.DataTypeMap[D] {\n  if (dtype === 'float32' || dtype === 'complex64') {\n    return a as tf.DataTypeMap[D];\n  } else if (dtype === 'int32' || dtype === 'bool') {\n    const result = (dtype === 'int32') ? new Int32Array(a.length) :\n                                         new Uint8Array(a.length);\n    for (let i = 0; i < result.length; ++i) {\n      result[i] = Math.round(a[i]);\n    }\n    return result as tf.DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport * as gpgpu_util from './gpgpu_util';\nimport * as webgl_util from './webgl_util';\n\nexport {MathBackendWebGL, WebGLMemoryInfo, WebGLTimingInfo} from './backend_webgl';\nexport {setWebGLContext} from './canvas_util';\nexport {GPGPUContext} from './gpgpu_context';\nexport {GPGPUProgram} from './gpgpu_math';\n// WebGL specific utils.\nexport {gpgpu_util, webgl_util};\n\n/**\n * Enforce use of half precision textures if available on the platform.\n *\n * @doc {heading: 'Environment', namespace: 'webgl'}\n */\nexport function forceHalfFloat(): void {\n  env().set('WEBGL_FORCE_F16_TEXTURES', true);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// base.ts is the webgl backend without auto kernel registration.\n\nimport {device_util, registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from './backend_webgl';\nexport {version as version_webgl} from './version';\n\nif (device_util.isBrowser()) {\n  registerBackend('webgl', () => new MathBackendWebGL(), 2 /* priority */);\n}\n\n// Export webgl utilities\nexport * from './webgl';\n\n// Export forceHalfFlost under webgl namespace for the union bundle.\nimport {forceHalfFloat} from './webgl';\nexport const webgl = {forceHalfFloat};\n", "import {BinaryInputs, DataType, env, KernelFunc, UnaryInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BinaryOpProgram} from '../binaryop_gpu';\nimport {BinaryOpPackedProgram} from '../binaryop_packed_gpu';\nimport {UnaryOpProgram} from '../unaryop_gpu';\n\nexport const CHECK_NAN_SNIPPET_UNARY = `if (isnan(x)) return x;`;\n\nexport const CHECK_NAN_SNIPPET_BINARY = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const CHECK_NAN_SNIPPET_BINARY_PACKED = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param opSnippets Op snippet to create `UnaryOpProgram`.\n */\nexport function unaryKernelFunc(opSnippet: string): KernelFunc {\n  return ({inputs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = new UnaryOpProgram(x.shape, opSnippet);\n    return webglBackend.runWebGLProgram(program, [x], x.dtype);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param opSnippet Op snippet to create `BinaryOpProgram`.\n * @param packedOpSnippet Op snippet to create `BinaryOpPackedProgram`.\n * @param checkOutOfBoundsForPackedProgram Whether to set checkOutOfBounds=true\n *     when creating BinaryOpPackedProgram.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    opSnippet: string, packedOpSnippet: string,\n    checkOutOfBoundsForPackedProgram?: boolean, dtype?: DataType): KernelFunc {\n  // TODO(jingjin): handle complex64.\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            packedOpSnippet, a.shape, b.shape,\n            !!checkOutOfBoundsForPackedProgram) :\n        new BinaryOpProgram(opSnippet, a.shape, b.shape);\n    const $dtype = dtype || a.dtype;\n    const output = webglBackend.runWebGLProgram(program, [a, b], $dtype);\n    return output;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc, CHECK_NAN_SNIPPET_BINARY, CHECK_NAN_SNIPPET_BINARY_PACKED} from '../kernel_utils/kernel_funcs_utils';\n\nconst ATAN2 = CHECK_NAN_SNIPPET_BINARY + `\n  return atan(a, b);\n`;\n\nconst ATAN2_PACKED = `\n  vec4 result = atan(a, b);\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET_BINARY_PACKED + `\n  return result;\n`;\n\nexport const atan2KernelFunc = binaryKernelFunc(ATAN2, ATAN2_PACKED);\n\nexport const atan2Config: KernelConfig = {\n  kernelName: Atan2,\n  backendName: 'webgl',\n  kernelFunc: atan2KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendWebGL}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'webgl',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function avgPool(args: {\n  inputs: AvgPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const avgPoolProgram = new Pool2DProgram(convInfo, 'avg', false);\n  return backend.runWebGLProgram(avgPoolProgram, [x], 'float32');\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'webgl',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {AvgPool2DBackpropProgram} from '../avg_pool_backprop_gpu';\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const avgPoolBackpropProgram = new AvgPool2DBackpropProgram(convInfo);\n  return backend.runWebGLProgram(avgPoolBackpropProgram, [dy], x.dtype);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = '0.0';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = '1.0';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        float x = getXAtOutCoords();\n        float mean = getMeanAtOutCoords();\n        float variance = getVarianceAtOutCoords();\n        float offset = ${offsetSnippet};\n        float scale = ${scaleSnippet};\n        float inv = scale * inversesqrt(variance + float(${varianceEpsilon}));\n        setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = 'vec4(0.0)';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = 'vec4(1.0)';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        vec4 offset = ${offsetSnippet};\n        vec4 scale = ${scaleSnippet};\n\n        vec4 x = getXAtOutCoords();\n        vec4 mean = getMeanAtOutCoords();\n        vec4 variance = getVarianceAtOutCoords();\n\n        vec4 inv = scale * inversesqrt(variance + vec4(${varianceEpsilon}));\n\n        setOutput((x - mean) * inv + offset);\n      }\n    `;\n  }\n}\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BatchNormProgram} from '../batchnorm_gpu';\nimport {BatchNormPackedProgram} from '../batchnorm_packed_gpu';\n\nexport const batchNormKernelFunc: (params: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendWebGL,\n  attrs: FusedBatchNormAttrs\n}) => TensorInfo = ({inputs, backend, attrs}) => {\n  const {x, mean, variance, offset, scale} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const finalInputs = [x, mean, variance];\n\n  let offsetShape = null;\n  if (offset != null) {\n    offsetShape = offset.shape;\n    finalInputs.push(offset);\n  }\n\n  let scaleShape = null;\n  if (scale != null) {\n    scaleShape = scale.shape;\n    finalInputs.push(scale);\n  }\n\n  const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n      new BatchNormPackedProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon) :\n      new BatchNormProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon);\n  const output =\n      backend.runWebGLProgram(program, finalInputs, finalInputs[0].dtype);\n\n  return output;\n};\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'webgl',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst COS = CHECK_NAN_SNIPPET_UNARY + `\n  return cos(x);\n`;\n\nexport const cosKernelFunc = unaryKernelFunc(COS);\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'webgl',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\n// Without the equality check div produces 0.9999 for a = b, which when\n// floored can cause errors.\nconst DIV = `\nif (a == b) {\n  return 1.0;\n};\nreturn a / b;`;\n\n// We do the same as in ./binaryop_gpu, with vec4 and ivec4.\n// On Linux, the vectorized implementation produces NaNs when a and b are 0.\nconst DIV_PACKED = `\n  // vec4 one = vec4(equal(a, b));\n  // return one + (vec4(1.0) - one) * a / b;\n  vec4 result = a / b;\n  if(a.x == b.x) {\n    result.x = 1.;\n  }\n  if(a.y == b.y) {\n    result.y = 1.;\n  }\n  if(a.z == b.z) {\n    result.z = 1.;\n  }\n  if(a.w == b.w) {\n    result.w = 1.;\n  }\n\n  return result;\n`;\n\nexport const divKernelFunc = binaryKernelFunc(\n    DIV, DIV_PACKED, true /* checkOutOfBoundsForPackedProgram */);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'webgl',\n  kernelFunc: divKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FlipLeftRightProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(imageShape: [number, number, number, number]) {\n    const imageWidth = imageShape[2];\n    this.outputShape = imageShape;\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n\n          int coordX = ${imageWidth} - x;\n          float outputValue;\n          if(coordX >= 0 && coordX < ${imageWidth}) {\n            outputValue = getImage(coords[0], coords[1], coordX, coords[3]);\n          } else {\n            outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {FlipLeftRightProgram} from '../flip_left_right_gpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new FlipLeftRightProgram((image as Tensor4D).shape);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n        vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${width}.0, ${height}.0);\n\n        vec4 values = ${glsl.texture2D}(A, uv);\n        float value;\n        if (depth == 0) {\n          value = values.r;\n        } else if (depth == 1) {\n          value = values.g;\n        } else if (depth == 2) {\n          value = values.b;\n        } else if (depth == 3) {\n          value = values.a;\n        }\n\n        setOutput(floor(value * 255.0 + 0.5));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n\n        vec4 result = vec4(0.);\n\n        for(int row=0; row<=1; row++) {\n          for(int col=0; col<=1; col++) {\n            texC = coords[1] + row;\n            depth = coords[2] + col;\n\n            vec2 uv = (vec2(texC, texR) + halfCR) /\n                       vec2(${width}.0, ${height}.0);\n            vec4 values = ${glsl.texture2D}(A, uv);\n            float value;\n            if (depth == 0) {\n              value = values.r;\n            } else if (depth == 1) {\n              value = values.g;\n            } else if (depth == 2) {\n              value = values.b;\n            } else if (depth == 3) {\n              value = values.a;\n            }\n\n            result[row * 2 + col] = floor(value * 255.0 + 0.5);\n          }\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {TextureUsage} from '../tex_util';\n\nimport {FromPixelsProgram} from './FromPixels_utils/from_pixels_gpu';\nimport {FromPixelsPackedProgram} from './FromPixels_utils/from_pixels_packed_gpu';\n\nexport const fromPixelsConfig: KernelConfig = {\n  kernelName: FromPixels,\n  backendName: 'webgl',\n  kernelFunc: fromPixels as {} as KernelFunc,\n};\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\nfunction fromPixels(args: {\n  inputs: FromPixelsInputs,\n  backend: MathBackendWebGL,\n  attrs: FromPixelsAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  let {pixels} = inputs;\n  const {numChannels} = attrs;\n\n  const isVideo = typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement;\n  const isImage = typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement;\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n\n  const texShape: [number, number] = [height, width];\n  const outShape = [height, width, numChannels];\n\n  if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement | HTMLImageElement, 0, 0, width, height);\n    pixels = fromPixels2DContext.canvas;\n  }\n\n  const tempPixelHandle = backend.makeTensorInfo(texShape, 'int32');\n  // This is a byte texture with pixels.\n  backend.texData.get(tempPixelHandle.dataId).usage = TextureUsage.PIXELS;\n  backend.gpgpu.uploadPixelDataToTexture(\n      backend.getTexture(tempPixelHandle.dataId), pixels as ImageData);\n  const program = env().getBool('WEBGL_PACK') ?\n      new FromPixelsPackedProgram(outShape) :\n      new FromPixelsProgram(outShape);\n  const res = backend.runWebGLProgram(program, [tempPixelHandle], 'int32');\n  backend.disposeData(tempPixelHandle.dataId);\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReduceProgram} from '../reduce_gpu';\n\ntype ReduceTypes = 'all'|'any'|'max'|'min'|'sum'|'prod';\n\n// Returns an array of configuration objects that describe each stage of the\n// reduction.\nfunction getReductionStages(inShape: number[]):\n    Array<{inSize: number, windowSize: number, outSize: number}> {\n  const stages = [];\n\n  while (stages.length === 0 || stages[stages.length - 1].outSize !== 1) {\n    const outSize: number =\n        stages.length ? stages[stages.length - 1].outSize : inShape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(outSize);\n    stages.push({\n      inSize: outSize,\n      windowSize,\n      outSize: Math.ceil(outSize / windowSize)\n    });\n  }\n\n  return stages;\n}\n\nexport function reduce(\n    x: TensorInfo, dtype: DataType, reductionType: ReduceTypes,\n    backend: MathBackendWebGL): TensorInfo {\n  const reductionStages = getReductionStages(x.shape);\n\n  let result = x;\n  for (let i = 0; i < reductionStages.length; i++) {\n    const {inSize, windowSize, outSize} = reductionStages[i];\n\n    const program = new ReduceProgram(\n        {windowSize, inSize, batchSize: x.shape[0], outSize}, reductionType);\n    const previousResult = result;\n    result = backend.runWebGLProgram(program, [result], dtype);\n\n    if (previousResult.dataId !== x.dataId) {\n      backend.disposeData(previousResult.dataId);\n    }\n  }\n\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReshapePackedProgram} from '../reshape_packed_gpu';\nimport {getBatchDim, getRowsCols} from '../webgl_util';\n\nexport function packedReshape(\n    input: TensorInfo, afterShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const input3DShape =\n      [getBatchDim(input.shape),\n       ...getRowsCols(input.shape)] as [number, number, number];\n  const input3D: TensorInfo = {\n    dtype: input.dtype,\n    shape: input3DShape,\n    dataId: input.dataId\n  };\n  const afterShapeAs3D =\n      [getBatchDim(afterShape),\n       ...getRowsCols(afterShape)] as [number, number, number];\n\n  const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n  const preventEagerUnpackingOfOutput = true;\n  const output = backend.runWebGLProgram(\n      program, [input3D], input.dtype, null /* customSetup */,\n      preventEagerUnpackingOfOutput);\n  return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {packedReshape} from '../kernel_utils/reshape';\nimport {isReshapeFree} from '../webgl_util';\n\nexport function reshape(args: {\n  inputs: ReshapeInputs,\n  backend: MathBackendWebGL,\n  attrs: ReshapeAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n  const webglBackend = backend;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  const xTexData = webglBackend.texData.get(x.dataId);\n  if (xTexData.isPacked && !isReshapeFree(x.shape, $shape) &&\n      !(xTexData.texture !== null && isReshapeFree(xTexData.shape, $shape))) {\n    return packedReshape(x, $shape, webglBackend);\n  }\n\n  webglBackend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'webgl',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {reduce} from '../kernel_utils/reduce';\nimport {reshape} from '../kernels/Reshape';\n\nexport function maxImpl(\n    x: TensorInfo, reduceShape: number[], outShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const inSize = util.sizeFromShape(reduceShape);\n  const xSize = util.sizeFromShape(x.shape);\n  const batchSize = xSize / inSize;\n  const reshapedInput =\n      reshape({inputs: {x}, attrs: {shape: [batchSize, inSize]}, backend});\n\n  const reduced = reduce(reshapedInput, x.dtype, 'max', backend);\n  const reshapedOutput =\n      reshape({inputs: {x: reduced}, attrs: {shape: outShape}, backend});\n\n  backend.disposeIntermediateTensorInfo(reshapedInput);\n  backend.disposeIntermediateTensorInfo(reduced);\n\n  return reshapedOutput;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposeProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const switched = getSwitchedCoords(newDim);\n\n    this.userCode = `\n    void main() {\n      ${dtype} resRC = getOutputCoords();\n      setOutput(getA(${switched}));\n    }\n    `;\n  }\n}\n\nfunction getSwitchedCoords(newDim: number[]): string {\n  const rank = newDim.length;\n  if (rank > 6) {\n    throw Error(`Transpose for rank ${rank} is not yet supported`);\n  }\n  const originalOrder =\n      ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u', 'resRC.v'];\n  const switchedCoords = new Array(rank);\n  for (let i = 0; i < newDim.length; i++) {\n    switchedCoords[newDim[i]] = originalOrder[i];\n  }\n  return switchedCoords.join();\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getVecChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    if (this.rank > 6) {\n      throw Error(\n          `Packed transpose for rank ${this.rank} is not yet supported.`);\n    }\n    const dtype = getCoordsDataType(this.rank);\n\n    const outputOrder = getVecChannels('rc', this.rank);\n    const switchedOrder = new Array(this.rank);\n    for (let i = 0; i < newDim.length; i++) {\n      switchedOrder[newDim[i]] = outputOrder[i];\n    }\n    const innerDims = `vec2(${switchedOrder.slice(-2).join()})`;\n    const nextColumn =\n        `++${outputOrder[this.rank - 1]} < ${outputShape[this.rank - 1]}`;\n    const getc = `getChannel(getA(${switchedOrder.join()}), ${innerDims})`;\n\n    this.userCode = `\n    void main() {\n      ${dtype} rc = getOutputCoords();\n      vec4 result = vec4(0.);\n      result[0] = ${getc};\n      if(${nextColumn}) {\n        result[1] = ${getc};\n      }\n      --${outputOrder[this.rank - 1]};\n      if(++${outputOrder[this.rank - 2]} < ${outputShape[this.rank - 2]}) {\n        result[2] = ${getc};\n        if(${nextColumn}) {\n          result[3] = ${getc};\n        }\n      }\n      setOutput(result);\n    }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {transposeImplCPU} from '../kernel_utils/shared';\nimport {TransposeProgram} from '../transpose_gpu';\nimport {TransposePackedProgram} from '../transpose_packed_gpu';\n\nexport function transposeImpl(\n    x: TensorInfo, perm: number[], backend: MathBackendWebGL): TensorInfo {\n  const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n      new TransposePackedProgram(x.shape, perm) :\n      new TransposeProgram(x.shape, perm);\n  return backend.runWebGLProgram(program, [x], x.dtype);\n}\n\nexport {transposeImplCPU};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {maxImplCPU} from '../kernel_utils/shared';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl, transposeImplCPU} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    const maxInputIsTransposed = permutedAxes != null;\n    const shouldExecuteOnCPU = webglBackend.shouldExecuteOnCPU([x]);\n\n    let maxInput = x;\n    if (maxInputIsTransposed) {\n      if (shouldExecuteOnCPU) {\n        const xTexData = webglBackend.texData.get(maxInput.dataId);\n        const values = xTexData.values as TypedArray;\n\n        const newShape: number[] = new Array(xRank);\n        for (let i = 0; i < newShape.length; i++) {\n          newShape[i] = x.shape[permutedAxes[i]];\n        }\n        const maxInputValues =\n            transposeImplCPU(values, x.shape, x.dtype, permutedAxes, newShape);\n\n        maxInput = webglBackend.makeTensorInfo(newShape, x.dtype);\n        const maxInputData = webglBackend.texData.get(maxInput.dataId);\n        maxInputData.values = maxInputValues;\n      } else {\n        maxInput = transposeImpl(x, permutedAxes, webglBackend);\n      }\n\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n    }\n\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(maxInput.shape, axes);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // rather than reshape at the end, set the target shape here.\n      outShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n    }\n\n    let out;\n    if (shouldExecuteOnCPU) {\n      const xTexData = webglBackend.texData.get(maxInput.dataId);\n      const values = xTexData.values as TypedArray;\n\n      const outValues = maxImplCPU(\n          values, util.sizeFromShape(reduceShape), outShape, x.dtype);\n\n      out = webglBackend.makeTensorInfo(outShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = maxImpl(maxInput, reduceShape, outShape, webglBackend);\n    }\n\n    if (maxInputIsTransposed) {\n      webglBackend.disposeIntermediateTensorInfo(maxInput);\n    }\n\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function maxPool(args: {\n  inputs: MaxPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const maxPoolProgram = new Pool2DProgram(convInfo, 'max', false);\n  return backend.runWebGLProgram(maxPoolProgram, [x], x.dtype);\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'webgl',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {MaxPool2DBackpropProgram} from '../max_pool_backprop_gpu';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const getPositions = true;\n  const maxPoolPositionsProgram =\n      new Pool2DProgram(convInfo, 'max', getPositions);\n  const maxPoolPositions: TensorInfo =\n      backend.runWebGLProgram(maxPoolPositionsProgram, [x], x.dtype);\n\n  const maxPoolBackPropProgram = new MaxPool2DBackpropProgram(convInfo);\n  const result = backend.runWebGLProgram(\n      maxPoolBackPropProgram, [dy, maxPoolPositions], x.dtype);\n  backend.disposeIntermediateTensorInfo(maxPoolPositions);\n  return result;\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\n\nexport function maxPoolWithArgmaxImpl(\n    x: TensorInfo, includeBatchInIndex: boolean,\n    convInfo: backend_util.Conv2DInfo,\n    backend: MathBackendWebGL): TensorInfo[] {\n  let program = new Pool2DProgram(convInfo, 'max', false);\n  const poolOutput = backend.runWebGLProgram(program, [x], 'float32');\n\n  program = new Pool2DProgram(convInfo, 'max', true, true, includeBatchInIndex);\n  const indexOutput = backend.runWebGLProgram(program, [x], 'float32');\n  return [poolOutput, indexOutput];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    util.assert(\n        x.shape.length === 4,\n        () => `Error in maxPool: input must be rank 4 but got rank ${\n            x.shape.length}.`);\n    const dilations: [number, number] = [1, 1];\n    util.assert(\n        backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n            `Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        dilations, pad);\n\n    const [result, indexes] =\n        maxPoolWithArgmaxImpl(x, includeBatchInIndex, convInfo, webglBackend);\n    return [result, indexes];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV3, NonMaxSuppressionV3Attrs, NonMaxSuppressionV3Inputs, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV3Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV3,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV3Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold} =\n        attrs as unknown as NonMaxSuppressionV3Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n\n    return kernel_impls.nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal);\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs, TypedArray} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class RotateProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], radians: number,\n      fillValue: number|[number, number, number],\n      center: number|[number, number]) {\n    const imageHeight = imageShape[1];\n    const imageWidth = imageShape[2];\n    const sinFactor = Math.sin(radians).toFixed(3);\n    const cosFactor = Math.cos(radians).toFixed(3);\n    this.outputShape = imageShape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const centerXString = centerX.toFixed(3);\n    const centerYString = centerY.toFixed(3);\n\n    let fillSnippet = '';\n    if (typeof fillValue === 'number') {\n      fillSnippet = `float outputValue = ${fillValue.toFixed(2)};`;\n    } else {\n      fillSnippet = `\n        vec3 fill = vec3(${fillValue.join(',')});\n        float outputValue = fill[coords[3]];`;\n    }\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n          int y = coords[1];\n          float coordXFloat = (float(x) - ${centerXString}) * ${\n        cosFactor} - (float(y) - ${centerYString}) * ${sinFactor};\n          float coordYFloat = (float(x) - ${centerXString}) * ${\n        sinFactor} + (float(y) - ${centerYString}) * ${cosFactor};\n          int coordX = int(round(coordXFloat + ${centerXString}));\n          int coordY = int(round(coordYFloat + ${centerYString}));\n          ${fillSnippet}\n          if(coordX >= 0 && coordX < ${imageWidth} && coordY >= 0 && coordY < ${\n        imageHeight}) {\n            outputValue = getImage(coords[0], coordY, coordX, coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {RotateProgram} from '../rotate_gpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new RotateProgram(\n        (image as Tensor4D).shape, radians, fillValue, center);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SIN = CHECK_NAN_SNIPPET_UNARY + `\n  return sin(x);\n`;\n\nexport const sinKernelFunc = unaryKernelFunc(SIN);\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'webgl',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Square} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARE = `return x * x;`;\n\nexport const squareKernelFunc = unaryKernelFunc(SQUARE);\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'webgl',\n  kernelFunc: squareKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const squaredDifferenceKernelFunc =\n    binaryKernelFunc(SQUARED_DIFFERENCE, SQUARED_DIFFERENCE);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'webgl',\n  kernelFunc: squaredDifferenceKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst TAN = `return tan(x);`;\n\nexport const tanKernelFunc = unaryKernelFunc(TAN);\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'webgl',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {transposeImpl} from './Transpose_impl';\nimport {transposeImplCPU as cpuTranspose} from './Transpose_impl';\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as TransposeInputs;\n    const {perm} = attrs as {} as TransposeAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const newShape: number[] = new Array(xRank);\n    for (let i = 0; i < newShape.length; i++) {\n      newShape[i] = x.shape[perm[i]];\n    }\n\n    let out: TensorInfo;\n    if (webglBackend.shouldExecuteOnCPU([x])) {\n      const xTexData = webglBackend.texData.get(x.dataId);\n      const values = xTexData.values as TypedArray;\n      const outValues = cpuTranspose(values, x.shape, x.dtype, perm, newShape);\n\n      out = webglBackend.makeTensorInfo(newShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = transposeImpl(x, perm, webglBackend);\n    }\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {uniqueImplCPU} from '../kernel_utils/shared';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function unique(\n    args:\n        {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendWebGL}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  // For now, always forward calculation to the CPU backend.\n  console.warn(\n      'WARNING: ',\n      'UI might be locked temporarily as data is being downloaded');\n  const values = backend.readSync(x.dataId);\n  const {outputValues, outputShape, indices} =\n      uniqueImplCPU(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'webgl',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {atan2Config} from './kernels/Atan2';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {cosConfig} from './kernels/Cos';\nimport {divConfig} from './kernels/Div';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {fromPixelsConfig} from './kernels/FromPixels';\nimport {identityConfig} from './kernels/Identity';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {nonMaxSuppressionV3Config} from './kernels/NonMaxSuppressionV3';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {sinConfig} from './kernels/Sin';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {tanConfig} from './kernels/Tan';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  atan2Config,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  cosConfig,\n  divConfig,\n  flipLeftRightConfig,\n  fromPixelsConfig,\n  identityConfig,\n  maxConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  nonMaxSuppressionV3Config,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  sinConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  tanConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport * from '@tensorflow/tfjs-core';\nexport * from '@tensorflow/tfjs-layers';\nexport * from '@tensorflow/tfjs-converter';\n\n// Export data api as tf.data\nimport * as data from '@tensorflow/tfjs-data';\nexport {data};\n\n// Import and register backends.\nimport '@tensorflow/tfjs-backend-cpu';\nimport '@tensorflow/tfjs-backend-webgl';\n\n// Import versions of all sub-packages.\nimport {version_core} from '@tensorflow/tfjs-core';\nimport {version_cpu} from '@tensorflow/tfjs-backend-cpu';\nimport {version_webgl} from '@tensorflow/tfjs-backend-webgl';\nimport {version_data} from '@tensorflow/tfjs-data';\nimport {version_layers} from '@tensorflow/tfjs-layers';\nimport {version_converter} from '@tensorflow/tfjs-converter';\nimport {version as version_union} from './version';\n\nexport const version = {\n  'tfjs-core': version_core,\n  'tfjs-backend-cpu': version_cpu,\n  'tfjs-backend-webgl': version_webgl,\n  'tfjs-data': version_data,\n  'tfjs-layers': version_layers,\n  'tfjs-converter': version_converter,\n  'tfjs': version_union\n};\n", "const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  if (frame > config.face.age.skipFrames) {\n    frame = 0;\n  } else {\n    frame += 1;\n  }\n  if (frame === 0) return last;\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;\n    if (confidence > config.face.gender.minConfidence) {\n      obj.gender = data[0] <= 0.5 ? 'female' : 'male';\n      obj.confidence = confidence;\n    }\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "/* eslint-disable indent */\n/* eslint-disable no-multi-spaces */\n\nexport default {\n  backend: 'webgl',          // select tfjs backend to use\n  console: true,             // enable debugging output to console\n  face: {\n    enabled: true,           // controls if specified modul is enabled\n                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion\n                             // (note: module is not loaded until it is required)\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.\n                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.\n      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'\n      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192,        // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3,    // empiric tuning\n      inputSize: 64,         // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'\n                                                         // which determines training set for model\n      inputSize: 64,         // fixed value\n      skipFrames: 10,        // how many frames to go without re-running the detector\n    },\n    gender: {\n      enabled: true,\n      minConfidence: 0.8,    // threshold for discarding a prediction\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64,         // fixed value\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      skipFrames: 10,        // how many frames to go without re-running the detector\n      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257,    // fixed value\n    outputStride: 16,        // fixed value\n    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256,          // fixed value\n    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n    minConfidence: 0.5,      // threshold for discarding a prediction\n    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace\n    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    // let savedWebglPackDepthwiseConvFlag;\n    // if (tf.getBackend() === 'webgl') {\n    //   savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n    //  tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    // }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          agConfidence: ssrData.confidence,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    // tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
+  "mappings": "oMAAA,kBCAA,kBCAA,0sDCAA,AAsBO,GAAM,IAAkB,KAClB,GAAkB,mBA0B7B,WAAoB,EAAgC,GAAhC,KAAA,QAAA,EAAgC,KAAA,UAAA,EAH5C,KAAA,KAAO,GAAI,SACX,KAAA,aAAe,EAIvB,SAAA,UAAA,IAAA,SAAI,GACF,MAAK,MAAK,KAAK,IAAI,IACjB,KAAK,UAAU,SAAS,KAAK,QAAS,GAEjC,KAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,IAAA,SAAI,EAAgB,GAClB,KAAK,eACL,KAAK,KAAK,IAAI,EAAQ,IAGxB,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,MAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,OAAA,SAAO,GACL,YAAK,eACE,KAAK,KAAK,OAAO,IAG1B,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,cAEhB,mBAqBA,cACE,SAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,WACE,MAAO,GAAkB,eAE3B,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,GAAkB,gBAE3B,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAC5C,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAE3D,MAAO,GAAkB,SAE3B,EAAA,UAAA,OAAA,WACE,MAAO,GAAkB,WAG3B,EAAA,UAAA,eAAA,WACE,MAAO,GAAkB,mBAG3B,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAG1D,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAEnD,MAAO,GAAkB,qBAG3B,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,MAAO,GAAkB,UAE3B,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,MAAO,GAAkB,iBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,GAAkB,WAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAG3B,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,uBAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAE3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAG3B,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,MAAO,GAAkB,iBAG3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,MAAO,GAAkB,WAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,MAAO,GAAkB,sBAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,GAAkB,WAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,MAAO,GAAkB,WAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAG3B,EAAA,UAAA,QAAA,SAA0B,GACxB,MAAO,GAAkB,YAG3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAG3B,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,yBAG3B,EAAA,UAAA,gBAAA,SAAgB,EAAiB,EAAkB,GAEjD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,wBAAA,SAAwB,EAAc,EAAkB,GAEtD,MAAO,GAAkB,4BAE3B,EAAA,UAAA,yBAAA,SAAyB,EAAa,EAAc,GAElD,MAAO,GAAkB,6BAE3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,EAAa,GAEtD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SAAkB,EAAc,EAAa,GAC3C,MAAO,GAAkB,sBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAAa,GAC1C,MAAO,GAAkB,sBAG3B,EAAA,UAAA,QAAA,SAA0C,EAAM,GAE9C,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,MAAO,GAAkB,QAG3B,EAAA,UAAA,UAAA,SAA4B,EAAM,GAChC,MAAO,GAAkB,cAG3B,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,MAAO,GAAkB,WAG3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GACpC,MAAO,GAAkB,cAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,mBAG3B,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,MAAO,GAAkB,2BAG3B,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,0BAG3B,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,MAAO,GAAkB,kCAG3B,EAAA,UAAA,UAAA,SACI,EAAa,EAAyB,EACtC,EAA4B,EAC5B,GACF,MAAO,GAAkB,cAG3B,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,MAAO,GAAkB,iCAG3B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EAAuB,EAC3D,EAAc,EAAe,GAC/B,MAAO,GAAkB,YAG3B,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,MAAO,GAAkB,WAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,MAAO,GAAkB,WAG3B,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,MAAO,GAAkB,sBAG3B,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAC3C,MAAO,GAAkB,iBAI3B,EAAA,UAAA,MAAA,SAAwB,EAAU,EAAsB,GACtD,MAAO,GAAkB,UAG3B,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,MAAO,GAAkB,SAG3B,EAAA,UAAA,SAAA,SAAyB,GACvB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,GAAkB,cAG3B,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAkB,aAG3B,EAAA,UAAA,QAAA,WACE,MAAO,GAAkB,YAE7B,KAEA,WAA2B,GACzB,KAAM,IAAI,OACN,IAAI,EAAU,4HCzpBpB,AAoBA,GAAM,IAA4B,0BA6BhC,WAAmB,GAAA,KAAA,OAAA,EATX,KAAA,MAAe,GACf,KAAA,aAAwD,GAExD,KAAA,SAAkB,GAOxB,KAAK,mBAGP,SAAA,UAAA,YAAA,SAAY,EAAsB,GAChC,AAAI,KAAK,UAAY,MACnB,QAAQ,KACJ,YAAY,KAAK,aAAY,0BAC7B,kCAAiC,EAAQ,MAE/C,KAAK,aAAe,EACpB,KAAK,SAAW,GAGlB,EAAA,UAAA,aAAA,SACI,EAAkB,EAClB,GAKF,GAJA,KAAK,aAAa,GAAY,CAAC,aAAY,EAAE,QAAO,GAIhD,KAAK,SAAS,IAAa,MAC7B,GAAM,GAAY,KAAK,SAAS,GAChC,QAAQ,KACJ,qCAAqC,EAAQ,KAAK,EAAS,KAC/D,KAAK,IAAI,EAAU,KAIjB,EAAA,UAAA,SAAN,SAAe,qGACb,MAAI,KAAY,MAAK,MACnB,CAAA,EAAO,KAAK,MAAM,IAGpB,GAAA,KAAK,MAAM,EAAA,EAAY,CAAA,EAAM,KAAK,aAAa,YAA/C,SAAA,GAAuB,EAAA,OACvB,CAAA,EAAO,KAAK,MAAM,UAGpB,EAAA,UAAA,IAAA,SAAI,GACF,GAAI,IAAY,MAAK,MACnB,MAAO,MAAK,MAAM,GAGpB,GAAM,GAAY,KAAK,aAAa,GACpC,GAAI,YAAqB,SACvB,KAAM,IAAI,OACN,QAAQ,EAAQ,sEAItB,YAAK,MAAM,GAAY,EAEhB,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,QAAA,SAAQ,GACN,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,OAGd,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,uCAGd,EAAA,UAAA,IAAA,SAAI,EAAkB,GACpB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,mBAAmB,EAAQ,mCAEjC,KAAK,MAAM,GAAY,EACvB,AAAI,KAAK,aAAa,GAAU,SAAW,MACzC,KAAK,aAAa,GAAU,QAAQ,IAIhC,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,yBAAyB,EAAQ,oCAEvC,MAAO,MAAK,aAAa,GAAU,gBAGrC,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,MAAQ,OAAO,OAAO,GAAI,IAGjC,EAAA,UAAA,MAAA,WACE,KAAK,MAAQ,GACb,KAAK,SAAW,GAChB,KAAK,oBAGC,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,GAAI,MAAO,MAAK,QAAW,aACvB,MAAO,MAAK,OAAO,UAAa,aAChC,MAAO,MAAK,OAAO,SAAS,QAAW,YACzC,OAGF,GAAM,GAAY,GAAe,KAAK,OAAO,SAAS,QACtD,GAAI,KAA6B,IAC/B,GAAM,GAAY,EAAU,IAA2B,MAAM,KAC7D,EAAU,QAAQ,SAAA,GACV,GAAA,GAAA,EAAA,MAAA,KAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,SAAS,GAAO,GAAW,EAAK,OAI7C,iBAE+B,GAC7B,GAAM,GAAS,GACf,SAAY,QAAQ,8BAA+B,SAAC,UAAG,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GACrD,UAAY,EAAQ,EAAE,GAAI,EAAE,IACrB,EAAE,KAAK,OAET,EAGT,YACI,EAAiC,EAAc,GACjD,EAAO,mBAAmB,IAAS,mBAAmB,GAAS,IAGjE,YAAoB,EAAkB,GAEpC,GADA,EAAQ,EAAM,cACV,IAAU,QAAU,IAAU,QAChC,MAAO,KAAU,OACZ,GAAI,GAAG,EAAE,IAAY,EAC1B,MAAO,CAAC,EAEV,KAAM,IAAI,OACN,oCAAoC,EAAK,aAAa,EAAQ,mBAYlE,MAAO,GAAA,IAGE,EAAA,IAAmB,iBACO,GACnC,EAAA,IAAM,EC9MR,AAqBA,GAAI,kBAGF,GAAI,IAAmB,MAErB,GAAI,GAAE,OACN,GAAI,MAAQ,SAAY,YACtB,EAAK,eACI,MAAQ,SAAY,YAC7B,EAAK,eACI,MAAQ,UAAa,YAC9B,EAAK,gBACI,MAAQ,OAAU,YAC3B,EAAK,SAEL,MAAM,IAAI,OAAM,kCAElB,GAAkB,EAEpB,MAAO,IAIT,cACE,GAAM,GAAK,KACX,MAAI,GAAG,YAAc,MACnB,GAAG,WAAa,GAAI,MAEf,EAAG,uBAUiB,EAAa,GACxC,GAAM,GAAY,KAClB,GAAI,EAAU,IAAI,GAChB,MAAO,GAAU,IAAI,GAErB,GAAM,GAAY,IAClB,SAAU,IAAI,EAAK,GACZ,EAAU,IAAI,MCzCZ,IAAM,MAGN,GAAO,OAGP,GAAQ,QAGR,GAAM,MAGN,GAAO,OAGP,GAAM,MAON,GAAM,MAON,GAAS,SAMT,GAAS,SAMT,GAAO,OAGP,GAAQ,QAGR,GAAO,OAGP,GAAQ,QAGR,GAAQ,QAGR,GAAU,UASV,GAAkB,kBAQlB,GAAY,YAWZ,GAAoB,oBAUpB,GAAc,cAOd,GAAiB,iBASjB,GAAc,cAOd,GAAO,OAMP,GAAO,OAGP,GAAc,cAOd,GAAU,UAGV,GAAS,SAMT,GAAS,SAUT,GAAuB,uBASvB,GAAsB,sBAUtB,GAAS,SAST,GAAyB,yBAQzB,GAAwB,wBAMxB,GAAM,MAGN,GAAO,OAGP,GAAS,SAQT,GAAgB,gBAShB,GAAe,eAOf,GAAwB,wBAWxB,GACT,sCAIS,GACT,qCAIS,GAAO,OAGP,GAAa,aAQb,GAA0B,0BAI1B,GAA2B,2BAI3B,GAAM,MAGN,GAAM,MAGN,GAAU,UAGV,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAO,OAOP,GAAgB,gBAGhB,GAAQ,QAGR,GAAW,WAGX,GAAiB,iBAOjB,GAAW,WAMX,GAAW,WAGX,GAAU,UAGV,GAAe,eAGf,GAAW,WAGX,GAAO,OAGP,GAAO,OAGP,GAAW,WAGX,GAAQ,QAGR,GAAQ,QAGR,GAAO,OAGP,GAAY,YAGZ,GAAW,WAMX,GAAM,MAGN,GAAQ,QAGR,GAAa,aAGb,GAAa,aAGb,GAAY,YAGZ,GAAa,aAMb,GAAM,MASN,GAAc,cASd,GAAM,MAON,GAAU,UAGV,GAAU,UASV,GAAkB,kBAUlB,GAAY,YAWZ,GAAoB,oBAWpB,GAAoB,oBASpB,GAAO,OAOP,GAAM,MAON,GAAU,UAGV,GAAM,MAGN,GAAW,WAGX,GAAS,SAGT,GAAW,WAGX,GAAsB,sBAStB,GAAsB,sBAUtB,GAAsB,sBAUtB,GAAW,WAGX,GAAS,SAQT,GAAQ,QAOR,GAAO,OAGP,GAAM,MAGN,GAAQ,QAGR,GAAO,OAOP,GAAQ,QAQR,GAAO,OAGP,GAAa,aAGb,GAAO,OAGP,GAAU,UAMV,GAAwB,wBAOxB,GAA4B,4BAI5B,GAAiB,iBAOjB,GAAqB,qBAGrB,GAAQ,QAGR,GAAU,UAMV,GAAQ,QAGR,GAAQ,QAGR,GAAY,YAMZ,GAAW,WAGX,GAAO,OAGP,GAAQ,QAMR,GAAM,MAGN,GAAO,OAGP,GAAO,OAGP,GAAU,UAGV,GAAW,WAGX,GAAO,OAGP,GAAM,MAON,GAAiB,iBAOjB,GAAS,SAOT,GAAU,UAMV,GAAoB,oBAGpB,GAAS,SAGT,GAAM,MAGN,GAAgB,gBAOhB,GAAe,eAaf,GAAM,MAGN,GAAO,OAGP,GAAO,OAMP,GAAO,OAOP,GAAY,YAMZ,GAAS,SAQT,GAAS,SAMT,GAAqB,qBAOrB,GAAY,YAMZ,GAAO,OAMP,GAAa,aASb,GAAmB,mBAQnB,GAAe,eAef,GAAc,cAgBd,GAAuB,uBC7zBpC,AAuBA,GAAM,IACF,GAAU,iBAAkB,WAAM,MAAA,IAAI,OACpC,GACF,GAAU,eAAgB,WAAM,MAAA,IAAI,mBAqEpC,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,MAAO,IAAe,IAAI,eAOA,GAC1B,MAAO,IAAa,IAAI,eAGW,GAInC,OAHM,GAAK,GAAe,UACpB,EAAyB,MAGvB,GAAA,GAAA,EAAA,OAAC,EAAA,EAAA,KAAM,EAAA,EAAA,MACb,GAAI,EACF,MAEK,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GACL,EAAA,EAAA,MAAA,KAAA,GACP,AAAI,IAAY,GACd,EAAO,KAAK,GAGhB,MAAO,eAcsB,GACtB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,YACb,EAAM,GAAQ,EAAY,GAChC,AAAI,GAAe,IAAI,IACrB,QAAQ,KACJ,eAAe,EAAU,iBACzB,KAAI,EAAW,4BAErB,GAAe,IAAI,EAAK,eAWO,GACxB,GAAA,GAAA,EAAA,WAEP,AAAI,GAAa,IAAI,IAGnB,CAAI,KAAM,QAAQ,UAChB,QAAQ,KAAK,gCAAgC,EAAU,MAG3D,GAAa,IAAI,EAAY,eAW3B,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,GAAI,CAAC,GAAe,IAAI,GACtB,KAAM,IAAI,OACN,eAAe,EAAU,iBACzB,KAAI,EAAW,wBAErB,GAAe,OAAO,eAIW,GACjC,GAAI,CAAC,GAAa,IAAI,GACpB,KAAM,IAAI,OACN,iBAAiB,EAAU,mCAEjC,GAAa,OAAO,eAUlB,EAA+B,GACjC,GAAM,GAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,GAAM,GACF,OAAO,OAAO,GAAI,EAAc,CAAC,YAAa,IAClD,GAAe,KAInB,YAAiB,EAAoB,GACnC,MAAU,GAAW,IAAI,ECrN3B,YAkCwB,GAMtB,OAJI,GAAU,EAAM,OAChB,EAAO,EACP,EAAQ,EAEL,EAAU,GAEf,EAAS,KAAK,SAAW,EAAW,EAEpC,IAEA,EAAO,EAAM,GACb,EAAM,GAAW,EAAM,GACvB,EAAM,GAAS,cAKG,EAAa,EAAW,GAC5C,MAAO,MAAK,IAAI,EAAK,KAAK,IAAI,EAAG,gBAGD,GAChC,MAAO,GAAM,IAAM,EAAI,EAAM,EAAM,cAGjB,GAElB,OADI,GAAM,EACD,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,GAAO,EAAI,GAEb,MAAO,eAUmB,EAAW,GACrC,GAAM,GAAI,KAAK,SACf,MAAQ,GAAI,EAAM,GAAI,GAAK,cAID,EAAe,GAEzC,OADI,GAAS,EACJ,EAAI,EAAG,EAAI,EAAE,OAAQ,KAC5B,GAAM,GAAO,OAAO,EAAE,IAAM,OAAO,EAAE,IACrC,GAAU,EAAO,EAEnB,MAAO,cAkBc,EAAe,GACpC,GAAI,CAAC,EACH,KAAM,IAAI,OAAM,MAAO,IAAQ,SAAW,EAAM,iBAKhD,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EACI,GAAY,EAAQ,GACpB,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlC,GAC5B,EACI,GAAK,KACL,WAAM,MAAA,8EAwBR,EAA0B,EAAkB,GAI9C,GAJ4B,AAAA,IAAA,QAAA,GAAA,IAAkB,AAAA,IAAA,QAAA,GAAA,IAC9C,AAAI,GAAU,MACZ,GAAS,IAEP,MAAM,QAAQ,IAAQ,GAAa,IAAQ,CAAC,EAC9C,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAAQ,EAAI,GAAI,EAAQ,OAG1B,GAAO,KAAK,GAEd,MAAO,eAcqB,GAC5B,GAAI,EAAM,SAAW,EAEnB,MAAO,GAGT,OADI,GAAO,EAAM,GACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,GAAQ,EAAM,GAEhB,MAAO,eAGqB,GAC5B,MAAO,GAAM,SAAW,cAGE,EAAgB,GAC1C,GAAI,IAAO,EACT,MAAO,GAET,GAAI,GAAM,MAAQ,GAAM,KACtB,MAAO,GAGT,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,EAAG,GACf,MAAO,GAGX,MAAO,eAGa,GACpB,MAAO,GAAI,IAAM,cAGE,GAEnB,GAAK,KAAa,MAAQ,KAExB,MAAQ,MAAa,KAAK,GAE5B,GAAI,IAAM,SACR,MAAO,GACF,GAAI,IAAM,UACf,MAAO,GAEP,GAAM,GAAM,KAAK,IAAI,EAAI,GACzB,MAAQ,GAAM,GAAM,GAAM,eAIM,GAClC,GAAM,GAAQ,KAAK,KAAK,KAAK,KAAK,IAClC,MAAO,CAAC,EAAO,KAAK,KAAK,EAAO,gBAeI,GAEpC,OADM,GAAkB,GAAI,aAAY,GAC/B,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAgB,GAAK,EAEvB,UAAQ,GACD,cAGgB,EAAW,GAClC,MAAI,IAAQ,EAAE,OACL,EAEF,EAAI,IAAI,OAAO,EAAO,EAAE,oBAI7B,EAAwB,EACxB,GADwB,MAAA,KAAA,QAAA,GAAA,SAAW,GAAoB,MAAA,KAElD,GAAI,SAAc,SAAC,EAAS,GACjC,GAAI,GAAW,EAET,EAAQ,WACZ,GAAI,KACF,IACA,OAGF,IAEA,GAAM,GAAc,EAAQ,GAE5B,GAAI,GAAc,MAAQ,GAAY,GACpC,IACA,OAEF,WAAW,EAAO,IAGpB,kBAcA,EAAiB,GAInB,OAHI,GAAY,EACZ,EAAc,GAET,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,GAAI,EAAM,IAAM,EACd,GAAa,EAAM,WACV,EAAM,KAAO,IACtB,GAAI,IAAgB,GAClB,KAAM,OACF,yCACA,oBAAmB,EAAW,YAAY,IAEhD,EAAc,UACL,EAAM,GAAK,EACpB,KAAM,OAAM,gCAAgC,EAAM,GAAE,WAAW,GAInE,GAAI,IAAgB,IAClB,GAAI,EAAO,GAAK,IAAS,EACvB,KAAM,OAAM,QAAQ,EAAI,qCAAqC,GAE/D,MAAO,GAGT,GAAI,IAAc,EAChB,KAAM,OACF,qCAAqC,EAAK,+BAGhD,GAAI,EAAO,IAAc,EACvB,KAAM,OACF,oDACA,QAAO,EAAI,MAAM,IAGvB,GAAM,GAAW,EAAM,QACvB,SAAS,GAAe,EAAO,EACxB,cAIL,EAAuB,GACzB,GAAM,GAAO,EAAM,OAGnB,SAAO,GAAQ,KAAO,EAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAK,GAAG,OAAO,GAGzD,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,CAAC,GAAQ,EAAK,IACrC,WACI,MAAA,+CAA+C,EAAI,KAAK,EAAI,SAC5D,aAAY,KAGpB,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,KACvB,WAAM,MAAA,iDACF,aAAY,KAGb,EAAK,IAAI,SAAA,GAAK,MAAA,GAAI,EAAI,EAAO,EAAI,gBAIb,EAAiB,GAS5C,OAPM,GAAqB,GACrB,EAAqB,GACrB,EAAe,GAAQ,MAAQ,MAAM,QAAQ,IAAS,EAAK,SAAW,EACtE,EAAQ,GAAQ,MAAQ,EAC1B,KACA,GAAe,EAAM,GAAO,OAC5B,EAAI,EACC,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,GAAQ,MACV,GAAI,EAAK,KAAO,GAAK,EAAM,KAAO,EAChC,KAAM,IAAI,OACN,sBAAsB,EAAC,mBAAmB,EAAM,GAAE,cAExD,AAAK,GAAK,IAAM,MAAQ,EAAK,GAAK,IAAM,EAAM,KAAO,GACnD,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAEhB,AAAI,EAAK,IAAM,GACb,IAGJ,AAAI,EAAM,KAAO,GACf,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAGlB,MAAO,CAAC,SAAQ,EAAE,SAAQ,eAIxB,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,SACnB,EAAS,GAAI,OAAgB,OAE7B,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAA+B,GACjC,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAC1B,KAAM,OAAM,oBAAoB,EAAK,4BAA4B,EAAG,kBAM7C,GAC3B,MAAO,KAAU,QAAU,IAAU,aAAe,IAAU,WAC1D,IAAU,SAAW,IAAU,qBAOL,EAAmB,GACjD,MAAI,KAAY,aAGZ,KAAY,WAAa,IAAY,aAGrC,IAAY,SAAW,IAAY,WAAa,IAAY,aALvD,GAQL,MAAY,QAAU,IAAY,oBAMX,GAC3B,MAAO,aAAa,eAAgB,YAAa,aAC7C,YAAa,wBAGa,GAC9B,GAAI,IAAU,WAAa,IAAU,QACnC,MAAO,GACF,GAAI,IAAU,YACnB,MAAO,GACF,GAAI,IAAU,OACnB,MAAO,GAEP,KAAM,IAAI,OAAM,iBAAiB,eAUA,GACnC,GAAI,GAAO,KACT,MAAO,GAET,GAAI,GAAQ,EACZ,SAAI,QAAQ,SAAA,GAAK,MAAA,IAAS,EAAE,SACrB,cAIgB,GACvB,MAAO,OAAO,IAAU,UAAY,YAAiB,oBAG7B,GACxB,MAAO,OAAO,IAAU,sBAGD,GACvB,MAAO,OAAO,IAAU,qBAGC,GACzB,MAAI,OAAM,QAAQ,GACT,GAAW,EAAO,IAEvB,YAAkB,cACb,UACE,YAAkB,aAAc,YAAkB,YACpD,QACE,GAAS,GACX,UACE,GAAS,GACX,SACE,GAAU,GACZ,OAEF,sBAGkB,GACzB,MAAO,CAAC,CAAE,IAAK,EAAE,aAAe,EAAE,MAAQ,EAAE,mBAGf,EAAc,GAC3C,OAAS,GAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,GAAI,EAAO,IAAM,EACf,MAAO,GAGX,MAAO,eAGsB,GAC7B,GAAM,GAAO,EAAM,OACnB,GAAI,EAAO,EACT,MAAO,GAKT,GAAM,GAAU,GAAI,OAAM,EAAO,GACjC,EAAQ,EAAO,GAAK,EAAM,EAAO,GACjC,OAAS,GAAI,EAAO,EAAG,GAAK,EAAG,EAAE,EAC/B,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAM,EAAI,GAE1C,MAAO,eAOL,EAAiB,GACnB,MAAI,KAAU,SACL,GAAa,GAGf,GAAa,CAAC,GAAQ,eAGF,EAAe,GAC1C,GAAI,IAAU,SACZ,KAAM,IAAI,OAAM,6CASlB,GAPA,AAAI,MAAM,QAAQ,IAChB,GAAI,GAAQ,IAGd,AAAI,KAAM,QAAQ,UAChB,GAAyB,EAAe,GAEtC,GAAmB,EAAG,GACxB,MAAO,GAET,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,QAEnB,OADM,GAAO,GAAI,YAAY,EAAe,QACnC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,AAAI,KAAK,MAAO,EAAe,MAAQ,GACrC,GAAK,GAAK,GAGd,MAAO,OAEP,MAAM,IAAI,OAAM,qBAAqB,GAIzC,YAA2B,EAAgB,EAAiB,GAC1D,GAAM,GAAM,GAAI,OAChB,GAAI,EAAM,SAAW,EAEnB,OADM,GAAI,EAAM,GACP,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,EAAE,EAAS,OAMtB,QAHM,GAAI,EAAM,GACV,EAAO,EAAM,MAAM,GACnB,EAAM,EAAK,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IACjC,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,GAAkB,EAAS,EAAI,EAAK,EAAM,GAGvD,MAAO,eAIqB,EAAiB,GAC7C,GAAI,EAAM,SAAW,EAEnB,MAAO,GAAE,GAEX,GAAM,GAAO,EAAM,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IAC5C,GAAI,IAAS,EAEX,MAAO,GAET,GAAI,IAAS,EAAE,OACb,KAAM,IAAI,OAAM,IAAI,EAAK,mCAAmC,EAAE,OAAM,KAGtE,MAAO,IAAkB,EAAG,EAAO,GAGrC,YAA4B,EAAe,GACzC,MAAQ,aAAa,eAAgB,IAAU,WAC1C,YAAa,aAAc,IAAU,SACrC,YAAa,aAAc,IAAU,mBAIxC,EAAc,GAEhB,OADM,GAAQ,GAAoB,EAAM,GAC/B,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAM,GAAK,EAEb,MAAO,eAIL,EAAc,GAChB,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,OACnB,MAAO,IAAI,YAAW,GAEtB,KAAM,IAAI,OAAM,qBAAqB,eAUrC,EAAiB,GACnB,GAAM,GAAO,EAAM,OAAO,SAAC,EAAM,GAAS,MAAA,GAAO,GAAM,GACvD,GAAI,GAAS,MAAQ,IAAU,UAC7B,MAAO,IAAc,EAAO,GAAI,cAAa,IACxC,GAAI,IAAU,QACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IACtC,GAAI,IAAU,OACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IAE3C,KAAM,IAAI,OAAM,qBAAqB,iBAgBvC,MAAO,MAAM,SAAS,kBAG2B,GACjD,EAAM,QAAQ,SAAA,GACZ,EACI,OAAO,UAAU,IAAY,GAAW,EACxC,WACI,MAAA,mEACA,WAAU,EAAK,sBAqBvB,EAAc,GAChB,MAAO,MAAM,SAAS,MAAM,EAAM,eAWP,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,SACtC,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAG,eAWL,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,SAC9C,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAO,eAYlC,EAAgB,EAAc,GAChC,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,EAAQ,GAAK,EAAK,GAE7B,MAAO,eAYL,EAAe,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,EAAQ,IACrC,GAAS,EAAK,GAAK,EAAQ,GAE7B,SAAK,EAAK,OAAS,GAAK,EACjB,22BC3xBT,AA+BA,GAAA,IAAA,WACE,WAAoB,EAAoC,GAApC,KAAA,aAAA,EAAoC,KAAA,OAAA,EACtD,AAAI,GAAU,MACZ,MAAK,OAAS,GAAI,KAItB,SAAA,UAAA,cAAA,SAAc,EAAoB,EAAwB,GAQxD,OANI,GACE,EAAsB,WAC1B,EAAU,KAEN,EAAQ,KAAK,aAAa,KAAK,cAE5B,GACP,GAAM,GAAS,EAAQ,GAGvB,EAAO,OAAO,KAAK,SAAA,GACjB,GAA0B,EAAY,EAAO,MAAO,MAL/C,EAAI,EAAG,EAAI,EAAQ,OAAQ,MAA3B,GAST,GAAM,GAAgB,CACpB,WAAU,EACV,QAAO,EACP,OAAM,EACN,OAAQ,EAAM,KAAK,SAAA,GAAU,MAAA,GAAO,WACpC,UAAW,EAAM,KACb,SAAA,GAAU,MAAA,GAAO,qBAAuB,KACpC,EAAO,sBACP,MAEV,MAAO,IAGT,EAAA,UAAA,iBAAA,SAAiB,GAAjB,GAAA,GAAA,KACS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,OAAQ,EAAA,EAAA,OAAQ,EAAA,EAAA,UAE5C,EAAQ,QAAQ,SAAA,GACd,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,IAAY,KAAK,SAAA,GACnD,EAAK,OAAO,iBACR,EAAY,EAAQ,EAAe,GAAI,EAAe,GAAI,EAC1D,EAAe,SAI3B,iBAGI,EAAsB,EAAU,GAClC,GAAI,IAAU,UAEZ,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAE1B,eAAQ,KAAK,SAAS,EAAG,sBAAsB,EAAU,KAClD,GAGX,MAAO,GAGT,GAAA,IAAA,WAAA,cACE,SAAA,UAAA,iBAAA,SACI,EAAc,EAAgB,EAC9B,EAAgC,EAChC,GACF,GAAM,GAAO,MAAO,IAAW,SAAW,GAAiB,EAAM,KAAM,GAC7B,EAAO,MAC3C,EAAa,GAAc,EAAM,IACjC,EAAO,EAAO,KACd,EAAO,EAAO,KACd,EAAQ,GAAc,EAAO,MAAM,WAAY,IACjD,EAAyB,GAE7B,OAAW,KAAQ,IACjB,GAAM,GAAQ,EAAO,GACrB,GAAI,GAAS,MAGX,GAAM,GAAa,EAAM,OAAS,EAAO,MACnC,EAAY,EAAW,OAC7B,GACO,EAAI,KAAK,EAAS,KAAK,GAAY,EAAI,EAAa,IAAE,KAIjE,QAAQ,IACJ,KAAK,EAAU,MAAO,EAAI,MAAO,EAAI,KAAK,EAAK,MAAO,EAAI,MACtD,EAAsB,MAAO,EACjC,mBAAoB,YAAa,aAAc,gBAC/C,eAAgB,qBAExB,KCjIA,YA4CI,EAAkB,EAAc,GAKlC,OAFM,GAA8C,GAC9C,EAA0C,GACvC,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,EAAa,EAAG,GAAG,IAAM,GAG3B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GACZ,EAAa,EAAK,OACxB,OAAW,KAAa,IAItB,OAHM,GAAQ,EAAW,GAErB,EAAgB,GACX,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAa,EAAM,KACrB,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAa,EAAO,IAAM,KACzD,EAAgB,GAChB,EAAW,EAAK,IAAM,GACtB,MAIJ,GAAI,EACF,OAMN,GAAM,GAAgD,GACtD,EAAe,EAAE,IAAM,GAGvB,OAFM,GAAwC,GAErC,EAAI,EAAK,OAAS,EAAG,GAAK,EAAG,IAKpC,OAJM,GAAO,EAAK,GACZ,EAAa,EAAK,OAGf,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,IACvC,GAAI,EAAe,EAAK,QAAQ,GAAG,KACjC,OAAW,KAAa,GACtB,EAAe,EAAW,GAAW,IAAM,GAC3C,EAAS,EAAK,IAAM,GAEtB,MAON,OADM,GAA2B,GACxB,EAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GAElB,GAAI,EAAW,EAAK,KAAO,EAAS,EAAK,KAEvC,GAAM,GAA8C,GACpD,OAAW,KAAa,GAAK,QAC3B,GAAM,GAAY,EAAK,OAAO,GAC9B,AAAI,EAAa,EAAU,KACzB,GAAa,GAAa,GAK9B,GAAM,GAAa,OAAO,OAAO,GAAI,GACrC,EAAW,OAAS,EACpB,EAAW,QAAU,EAAK,QAE1B,EAAa,KAAK,IAItB,MAAO,eAWL,EACA,EAA0B,EAC1B,GAEF,mBAAS,GACP,GAAM,GAAO,EAAa,GAEpB,EAAgB,GAYtB,GAXA,EAAK,QAAQ,QAAQ,SAAA,GACnB,GAAM,GAAa,EAA6B,EAAE,IAClD,AAAI,GAAc,KAChB,EAAI,KAAK,GAIT,EAAI,KAAK,QAIT,EAAK,UAAY,KACnB,KAAM,IAAI,OACN,wDACA,QAAO,EAAK,WAAU,MAI5B,GAAM,GAAiB,EAAK,SAAS,cAE1B,GACT,GAAI,CAAE,KAAa,IACjB,KAAM,IAAI,OACN,iCAAiC,EAAS,KAC1C,+BAA8B,OAAO,KAAK,GAAe,MAI/D,GAAM,GAAK,EAAK,WAAM,MAAA,GAAe,OACrC,GAAI,EAAG,QAAU,UACf,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BAChB,GAAS,wCAAwC,EAAG,MAAK,MAElE,GAAM,GAAI,EAAK,OAAO,GACtB,GAAI,CAAC,GAAiB,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BACnB,KAAI,EAAS,gBAAgB,EAAG,MAAK,4BACrC,4BAA2B,EAAE,MAAK,MAGxC,GAAI,EAA6B,EAAE,KAAO,KACxC,EAA6B,EAAE,IAAM,OAErC,GAAM,GAAc,EAA6B,EAAE,IACnD,EAA6B,EAAE,IAAM,EAAI,EAAa,GACtD,EAAY,YA7BhB,OAAW,KAAa,GAAK,SAAlB,IAxBJ,EAAI,EAAa,OAAS,EAAG,GAAK,EAAG,MAArC,GCvIX,AAqBA,GAAM,IAAwB,GAExB,GAA6B,EAE7B,GAAwB,cAG1B,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAU,GAAe,GACzB,EAAY,GAAwB,EAAM,EAAO,EAAO,GACxD,EAAO,EAAM,OACb,EAAY,GAAkB,EAAM,EAAO,EAAO,EAAS,GAC3D,EAAQ,CAAC,UACf,MAAI,IACF,GAAM,KAAK,YAAY,GACvB,EAAM,KAAK,WAAW,GACtB,EAAM,KAAK,aAAa,EAAK,KAC7B,EAAM,KAAK,cAEb,EAAM,KAAK,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,IAAG,KAAK;IACxC,EAAM,KAAK;GAGpB,YACI,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAI,GAAc,GAClB,EAAU,EAAQ,EAAQ,OAAS,GACnC,EAAY,GAAI,OAAM,GAAS,KAAK,GACpC,EAAO,EAAM,OACb,EACF,IAAU,YAAc,GAAoB,GAAQ,EAExD,GAAI,EAAO,EACT,OAAS,GAAM,EAAG,EAAM,EAAI,EAAS,IAEnC,OADM,GAAS,EAAM,EACZ,EAAI,EAAG,EAAI,EAAS,IAC3B,EAAU,GAAK,KAAK,IAChB,EAAU,GACV,GAAY,EAAe,EAAS,GAAI,EAAG,GAAO,QAI5D,MAAO,GAGT,YACI,EAAqC,EAAa,GACpD,GAAI,GACJ,MAAI,OAAM,QAAQ,GAChB,EAAY,WAAW,EAAI,GAAG,QAAQ,KAAuB,MACtD,YAAW,EAAI,GAAG,QAAQ,KAAuB,KACnD,AAAI,GAAS,GAClB,EAAS,IAAI,EAAG,IACX,AAAI,IAAU,OACnB,EAAS,GAAgB,GAEzB,EAAS,WAAW,EAAI,QAAQ,KAAwB,WAGnD,GAAS,EAAQ,GAG1B,YAAyB,GACvB,MAAO,KAAM,EAAI,QAAU,OAG7B,YACI,EAA2B,EAAiB,EAC5C,EAAmB,EAAqB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAM,GAAoB,IAAU,YAAc,EAAI,EAEhD,EAAO,EAAM,GACb,EAAO,EAAM,OACnB,GAAI,IAAS,GACX,GAAI,IAAU,aACZ,GAAM,GAAe,GAAoB,GACzC,MAAO,CAAC,GAAY,EAAa,GAAI,EAAG,IAE1C,MAAI,KAAU,OACL,CAAC,GAAgB,EAAK,KAExB,CAAC,EAAK,GAAG,YAGlB,GAAI,IAAS,GACX,GAAI,EAAO,IACT,GAAM,GAAgB,GAA6B,EAE/C,EAAY,MAAM,KAClB,EAAK,MAAM,EAAG,IACd,EAAW,MAAM,KAAqC,EAAK,MAC1D,GAAO,IAA8B,EACtC,EAAO,IACX,MAAI,KAAU,aACZ,GAAY,GAAoB,GAChC,EAAW,GAAoB,IAE1B,CACL,IACA,EAAU,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAChD,KAAK,MACV,UACA,EACK,IACG,SAAC,EAAG,GAAM,MAAA,IACN,EAAG,EAAU,EAAO,GAA6B,GAAI,KAC5D,KAAK,MACV,KAGJ,GAAM,GACF,IAAU,YAAc,GAAoB,GACpB,MAAM,KAAoB,GAEtD,MAAO,CACL,IACA,EAAY,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAClD,KAAK,MACV,KAKJ,GAAM,GAAW,EAAM,MAAM,GACvB,EAAa,EAAQ,MAAM,GAC3B,EAAS,EAAQ,GAAK,EACtB,EAAkB,GACxB,GAAI,EAAO,IACT,OAAS,GAAI,EAAG,EAAI,GAA4B,KAC9C,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,KAEN,EAAM,KAAK,OACX,OAAS,GAAI,EAAO,GAA4B,EAAI,EAAM,KACxD,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,SAGnB,QAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,IAGrB,GAAM,GAAM,IAAS,EAAI,IAAM,GAC/B,EAAM,GAAK,IAAM,EAAM,GAAK,EAC5B,OAAS,GAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,EAAM,GAAK,IAAM,EAAM,GAAK,EAG9B,OADI,GAAa;EACR,EAAI,EAAG,EAAI,EAAM,IACxB,GAAc;EAEhB,SAAM,EAAM,OAAS,GACjB,IAAM,EAAM,EAAM,OAAS,GAAK,IAAO,GAAS,GAAK,GAClD,EAGT,YAA6B,GAG3B,OADM,GAAyC,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,GAAK,EACpC,EAAc,KAAK,CAAC,EAAK,GAAI,EAAK,EAAI,KAExC,MAAO,GCnMT,kBA6CE,WAAY,EAA2B,EAAU,GAAjD,GAAA,GAAA,KAIE,GAJqC,KAAA,MAAA,EACrC,KAAK,MAAQ,EAAM,QACnB,KAAK,KAAO,GAAmB,GAE3B,GAAU,MACZ,GAAM,GAAI,EAAO,OACjB,EACI,IAAM,KAAK,KACX,WAAM,MAAA,qBAAqB,EAAC,6BACxB,2BAA0B,EAAK,KAAI,QAE7C,GAAI,IAAU,YACZ,KAAM,IAAI,OACN,8JAIN,KAAK,OAAS,GAAU,GAAuB,EAAO,KAAK,MAC3D,KAAK,QAAU,GAAe,GAWhC,SAAA,UAAA,IAAA,SAAI,UAAJ,GAAA,KAA8B,EAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GAC5B,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAEV,EACI,EAAK,SAAW,KAAK,KACrB,WAAM,MAAA,uCAAuC,EAAK,OAAM,UACpD,oBAAmB,EAAK,KAAI,OAEpC,GAAM,GAAQ,KAAK,WAAW,GAC9B,KAAK,OAAO,GAAS,GAUvB,EAAA,UAAA,IAAA,kBAAI,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACF,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAGV,OADI,GAAI,EACU,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,EAAM,GAAK,GAAO,KAAK,MAAM,IAC/B,GAAM,GAAM,qCAAqC,EAAI,KACjD,mBAAkB,KAAK,OAC3B,KAAM,IAAI,OAAM,GAElB,IAGF,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,MAAK,OAAO,IAGrB,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,IAGT,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,KAAK,MAAM,QACnC,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,KAAK,QAAQ,IAC1C,GAAS,EAAK,GAAK,KAAK,QAAQ,GAElC,SAAK,EAAK,OAAS,GAAK,EACjB,GAGT,OAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAY,WAAW,KAAK,OAAQ,KAAK,MAAO,KAAK,QAGhE,KA8BI,GAAiC,KAEjC,GAAuB,iBAaM,GAC/B,GAAY,cAQe,GAC3B,GAAY,mBA4DZ,WAAY,EAAoB,EAAiB,EAAgB,GAXjE,KAAA,KAAO,GAoIG,KAAA,mBAAqB,GAxH7B,KAAK,MAAQ,EAAM,QACnB,KAAK,MAAQ,GAAS,UACtB,KAAK,KAAO,GAAmB,GAC/B,KAAK,QAAU,GAAe,GAC9B,KAAK,OAAS,EACd,KAAK,GAAK,EACV,KAAK,SAAY,KAAK,KAAO,EAAI,KAAK,KAAK,WAAa,SAG1D,cAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQd,EAAA,UAAA,OAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAU,OAAO,KAAK,MAAO,KAAK,MAAY,UAOvD,EAAA,UAAA,WAAA,WACE,MAAO,IAAU,OAAO,KAAK,MAAO,KAAK,MAAY,KAAK,aAStD,EAAA,UAAA,MAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAc,KAAK,MAAO,UASnC,EAAA,UAAA,UAAA,WACE,MAAO,IAAc,KAAK,MAAO,KAAK,aASlC,EAAA,UAAA,KAAN,mHACE,MAAK,kBACC,EAAO,KAAY,KAAK,KAAK,QAC/B,KAAK,QAAU,SACH,CAAA,EAAM,GADlB,CAAA,EAAA,UACI,EAAQ,EAAA,OACd,IACE,MAAA,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,IAAkB,YACxC,GACA,KAAM,IAAI,OACN,kHAIR,MAAA,CAAA,EAAO,SAST,EAAA,UAAA,SAAA,WACE,KAAK,kBACL,GAAM,GAAO,KAAY,SAAS,KAAK,QACvC,GAAI,KAAK,QAAU,SACjB,IACE,MAAQ,GAAsB,IAAI,SAAA,GAAK,MAAA,IAAkB,WAEzD,GACA,KAAM,IAAI,OACN,iGAIR,MAAO,IAIH,EAAA,UAAA,MAAN,2GACE,YAAK,kBACQ,CAAA,EAAM,KAAY,KAAK,KAAK,gBACzC,MADM,GAAO,EAAA,OACT,KAAK,QAAU,SACjB,CAAA,EAAO,GAEP,CAAA,EAAO,GAAI,YAAY,EAAoB,eAS/C,EAAA,UAAA,QAAA,WACE,GAAI,KAAK,WACP,OAEF,KAAY,cAAc,MAC1B,KAAK,mBAAqB,IAI5B,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,oDAGd,EAAA,UAAA,gBAAA,WACE,GAAI,KAAK,WACP,KAAM,IAAI,OAAM,wBAYpB,EAAA,UAAA,MAAA,SAAM,GAAA,MAAA,KAAA,QAAA,GAAA,IACG,GAAU,MAAM,KAAM,IAO/B,EAAA,UAAA,MAAA,WACE,YAAK,kBACE,GAAU,MAAM,OAQzB,EAAA,UAAA,SAAA,SAAS,GAAA,AAAA,IAAA,QAAA,GAAA,IACP,GAAM,GAAO,KAAK,WAClB,MAAO,IAAe,EAAM,KAAK,MAAO,KAAK,MAAO,IAGtD,EAAA,UAAA,KAAA,SAAqB,GACnB,YAAK,kBACE,GAAU,KAAK,KAAW,IAEnC,EAAA,UAAA,SAAA,SAAS,EAAkB,EAAe,GAAjC,MAAA,KAAA,QAAA,GAAA,IACP,KAAK,kBACE,KAAY,aAAa,KAAM,EAAW,EAAM,IAG3D,KACA,OAAO,eAAe,EAAQ,OAAO,YAAa,CAChD,MAAO,SAAC,GAMN,MAAO,CAAC,CAAC,GAAY,EAAS,MAAQ,MAAQ,EAAS,UAAY,MAC/D,EAAS,iBAAmB,2BAoCiB,GAAA,EAAA,GAGnD,WACI,EAAgC,EAAoB,EACpD,GAFJ,GAAA,GAGE,EAAA,KAAA,KACI,EAAa,MAAO,EAAa,MAAO,EAAa,OAAQ,IAAS,KAHxC,SAAA,UAAA,EAIlC,EAAK,KAAO,IAWd,SAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAS,QAAU,KAAK,MAC1B,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,GAAI,CAAC,GAAiB,EAAS,MAAO,KAAK,OACzC,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,KAAY,cAAc,MAC1B,KAAK,OAAS,EAAS,OACvB,KAAY,OAAO,KAAM,OAG3B,EAAA,UAAA,QAAA,WACE,KAAY,gBAAgB,MAC5B,KAAK,mBAAqB,IAE9B,GAvCqD,GAyCrD,OAAO,eAAe,GAAU,OAAO,YAAa,CAClD,MAAO,SAAC,GACN,MAAO,aAAoB,IAAU,EAAS,QAAU,MACpD,EAAS,iBAAkB,aCnhBnC,AAgEA,AAAA,UAAY,GACV,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,OAPU,EAAA,MAAA,GAAA,KAAI,KAqBhB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,QACA,EAAA,UAAA,cAJG,IAAA,IAAiB,KAOtB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,OACA,EAAA,UAAA,cAJG,IAAA,IAAgB,KAOrB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,UACA,EAAA,KAAA,UACA,EAAA,UAAA,cAJG,IAAA,IAAmB,KAOxB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,YACA,EAAA,MAAA,YACA,EAAA,KAAA,YACA,EAAA,UAAA,cAJG,IAAA,IAAqB,KAO1B,GAAM,IAAgB,CACpB,QAAW,GACX,MAAS,GACT,KAAQ,GACR,UAAa,gBAGY,EAAiB,GAC1C,GAAI,IAAU,UAAY,IAAU,UAClC,GAAI,IAAU,UAAY,IAAU,SAClC,MAAO,SAET,KAAM,IAAI,OAAM,kBAAkB,EAAK,SAAS,GAElD,MAAO,IAAc,GAAO,eAIH,GACzB,MAAO,IAAW,EAAM,SCpI1B,YAsBiD,EAAM,GACrD,GAAI,EAAE,QAAU,EAAE,MAChB,MAAO,CAAC,EAAG,GAEb,GAAM,GAAQ,GAAW,EAAE,MAAO,EAAE,OACpC,MAAO,CAAC,EAAE,KAAK,GAAQ,EAAE,KAAK,gBAGC,EAAW,GAC1C,EACI,EAAE,QAAU,EAAE,MACd,WAAM,MAAA,2BAA2B,EAAE,MAAK,QACpC,YAAW,EAAE,MAAK,oCAGG,EAAgB,GAC7C,MAAO,GAAW,KAAK,SAAA,GAAK,MAAA,GAAE,KAAO,EAAO,iBAeR,GACpC,GAAM,GAAiB,GACjB,EAAO,GAAI,KACjB,UAAoB,EAAQ,EAAM,GAC3B,EAGT,YACI,EAA4B,EAAgB,GAC9C,GAAI,GAAa,KACf,OAEF,GAAI,YAAqB,IACvB,EAAK,KAAK,GACV,OAEF,GAAI,CAAC,GAAW,GACd,OAGF,GAAM,GAAW,EACjB,OAAW,KAAK,IACd,GAAM,GAAM,EAAS,GACrB,AAAK,EAAK,IAAI,IACZ,GAAK,IAAI,GACT,GAAoB,EAAK,EAAM,KAMrC,YAAoB,GAClB,MAAO,OAAM,QAAQ,IAAQ,MAAO,IAAQ,kHCrF9C,AAmFA,GAAA,IAAA,WAAA,aAEE,KAAA,oBAAwC,GAExC,KAAA,eAAiB,EACjB,KAAA,SAAW,EACX,KAAA,WAAa,EACb,KAAA,iBAAmB,EACnB,KAAA,eAAiB,EAMjB,KAAA,cAAgB,EAGhB,KAAA,YAAc,EAId,KAAA,WAA2B,GAK3B,KAAA,kBAA8B,GAC9B,KAAA,YAAc,EAEd,KAAA,WAAa,GAAI,SAQjB,KAAA,UAAY,GACZ,KAAA,cACI,CAAC,SAAU,EAAG,WAAY,EAAG,UAAW,EAAG,QAAS,GAAI,OAAQ,MAEpE,SAAA,UAAA,QAAA,WACE,OAAW,KAAgB,MAAK,oBAC9B,KAAK,oBAAoB,GAAc,WAG7C,KAEA,GAAA,WAgBE,WAAmB,GAAA,KAAA,IAAA,EAbnB,KAAA,SAA0C,GAC1C,KAAA,gBAKI,GAKI,KAAA,qBAAuB,EAG7B,KAAK,MAAQ,GAAI,IAGb,SAAA,UAAA,MAAN,iHACE,GAAI,KAAK,oBAAsB,KAC7B,MAAA,CAAA,EAAO,KAAK,mBAAmB,KAAK,eAEtC,GAAI,KAAK,iBAAmB,KAC1B,MAAA,CAAA,GAEI,EAAiB,KAAK,oBAEnB,EAAI,yBAAG,GAAI,EAAe,OAC3B,GAAc,EAAe,GACnB,CAAA,EAAM,KAAK,kBAAkB,GAAa,UAFnB,CAAA,EAAA,gBAEjC,GAAU,EAAA,OACZ,EACF,CAAA,EAAM,KAAK,WAAW,IADpB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,UALuC,wBAS3C,KAAM,IAAI,OACN,gFAIN,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,GAAI,KAAK,oBAAsB,KAC7B,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,uHAIlC,GAAI,KAAK,iBAAmB,MACpB,GAAA,GAAA,KAAA,kCAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UACb,GAAI,EACF,KAAM,IAAI,OACN,iCAAiC,EAAI,uHAI3C,KAAK,WAAW,GAElB,MAAO,MAAK,iDAGd,EAAA,UAAA,aAAA,WACE,MAAO,QAAO,KAAK,KAAK,kBAG1B,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,CAAE,KAAe,MAAK,UAGxB,GAAI,IAAe,MAAK,iBACf,GAAA,GAAA,KAAA,kBAAA,GAAA,UACP,GAAI,EAEF,MAAO,UAGT,OAAO,MAGX,MAAO,MAAK,SAAS,IAGvB,EAAA,UAAA,mBAAA,SAAmB,GAEjB,MAAM,KAAe,MAAK,gBAGnB,KAAK,gBAAgB,GAAa,QAFhC,MAKX,EAAA,UAAA,gBAAA,SACI,EACA,EACA,GACF,MADE,AAAA,KAAA,QAAA,GAAA,GACE,IAAe,MAAK,gBACtB,SAAQ,KACD,EAAW,sEAEX,IAET,MAAK,gBAAgB,GAAe,CAAC,QAAO,EAAE,SAAQ,GAC/C,KAGH,EAAA,UAAA,WAAN,SAAiB,2GACf,GAAI,KAAK,gBAAgB,IAAgB,KACvC,KAAM,IAAI,OAAM,iBAAiB,EAAW,iCAE9C,MAAK,YAAc,EACf,KAAK,SAAS,IAAgB,KAChC,MAAK,gBAAkB,KACjB,EAAuB,KAAK,kBAAkB,GAA7C,EAAO,EAAA,QAAE,EAAS,EAAA,UACV,EAAY,CAAA,EAAM,GAAlB,CAAA,EAAA,IAHb,CAAA,EAAA,UAGyB,SAAA,EAAA,oBAAgB,EAAA,mBAC3C,GADM,EAAM,EACR,CAAC,EACH,MAAA,CAAA,EAAO,qBAGX,YAAK,gBAAkB,KAAK,SAAS,GACrC,KAAK,yBAEL,KAAK,SAAW,GAAI,IAAS,KAAK,iBAElC,CAAA,EAAO,UAGD,EAAA,UAAA,uBAAR,WAAA,GAAA,GAAA,KACQ,EAAU,GAAqB,KAAK,aAC1C,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,WAAa,MACtB,EAAO,UAAU,EAAK,oBAKpB,EAAA,UAAA,yBAAR,SAAiC,GAAjC,GAAA,GAAA,KACQ,EAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,aAAe,MACxB,EAAO,YAAY,EAAK,SAAS,OAW/B,EAAA,UAAA,kBAAR,SAA0B,GAA1B,GAAA,GAAA,KAEQ,EAAuB,KAAK,gBAAgB,GAClD,GAAI,GAAwB,KAC1B,KAAM,IAAI,OACN,6BAA6B,EAAW,4BAG9C,IACE,GAAM,GAAU,EAAqB,UAMrC,GAAI,GAAW,CAAE,aAAmB,MAC7B,MAAO,GAAQ,MAAS,YAC7B,GAAM,GAAY,EAAE,KAAK,qBACnB,EACF,EACK,KAAK,SAAA,GAEJ,MAAI,GAAY,EAAK,qBACZ,GAET,GAAK,SAAS,GAAe,EAC7B,EAAK,mBAAqB,KACnB,MAER,MAAM,SAAA,GAEL,MAAI,GAAY,EAAK,sBAGrB,GAAK,mBAAqB,KAC1B,QAAQ,KACJ,6BAA6B,EAAW,WAC5C,QAAQ,KAAK,EAAI,OAAS,EAAI,UACvB,KAEjB,YAAK,mBAAqB,EACnB,CAAC,QAAO,EAAE,UAAW,QAE5B,aAAK,SAAS,GAAe,EACtB,CAAC,QAAS,GAAM,UAAW,UAE7B,GACP,eAAQ,KAAK,6BAA6B,EAAW,WACrD,QAAQ,KAAK,EAAI,OAAS,EAAI,SACvB,CAAC,QAAS,GAAO,UAAW,MAIvC,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAE,KAAe,MAAK,iBACxB,KAAM,IAAI,OAAS,EAAW,kCAEhC,AAAI,KAAK,cAAgB,GAAe,KAAK,oBAAsB,MAGjE,KAAK,uBAGP,AAAI,IAAe,MAAK,UACtB,MAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,IAGvB,MAAO,MAAK,gBAAgB,GAG5B,AAAI,KAAK,cAAgB,GACvB,MAAK,mBAAqB,KAC1B,KAAK,YAAc,KACnB,KAAK,gBAAkB,OAInB,EAAA,UAAA,kBAAR,WAAA,GAAA,GAAA,KACE,GAAI,OAAO,KAAK,KAAK,iBAAiB,SAAW,EAC/C,KAAM,IAAI,OAAM,iCAElB,MAAO,QAAO,KAAK,KAAK,iBAAiB,KAAK,SAAC,EAAW,GAExD,MAAO,GAAK,gBAAgB,GAAG,SAC3B,EAAK,gBAAgB,GAAG,YAIxB,EAAA,UAAA,gCAAR,WAIE,OAFM,GAAiB,KAAK,oBAEnB,EAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAc,EAAe,GAC7B,EAAA,KAAA,kBAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAChB,GAAI,GAAa,EACf,MAAO,CAAC,KAAM,EAAa,UAAS,GAGxC,KAAM,IAAI,OACN,2EAIN,EAAA,UAAA,SAAA,SAAS,EAAwB,GAC/B,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAa,EAAK,QAClB,EAAS,KAAK,SAAS,GAG7B,EAAW,YAAY,GACvB,EAAK,QAAU,EACf,EAAQ,KAAK,EAAQ,EAAQ,EAAK,MAAO,EAAK,OAC9C,AAAI,KAAK,0BAGP,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,MAIvE,EAAA,UAAA,KAAA,SAAgC,EAA6B,GAA7D,GAAA,GAAA,KAEM,EAAe,KACnB,GAAI,GAAM,MAER,GAAI,MAAO,IAAa,WACtB,KAAM,IAAI,OAAM,uCAElB,EAAK,OAGL,GAAI,MAAO,IAAa,UAAY,CAAE,aAAoB,SACxD,KAAM,IAAI,OACN,kFAGN,GAAI,MAAO,IAAO,WAChB,KAAM,IAAI,OACN,kFAGN,EAAO,EAIT,GAAI,GACJ,MAAO,MAAK,UACR,WAAM,MAAA,GAAK,WAAW,IAAO,WAAM,MAAA,GAAK,SAAS,IAAS,WACxD,SAAS,IACT,AAAI,YAAkB,UACpB,QAAQ,MAAM,2CAET,KAIP,EAAA,UAAA,UAAR,SAAqB,EAAmB,EAAiB,GACvD,IACA,IACE,GAAM,GAAM,IACZ,WACO,QACA,GACP,UACM,IAKF,EAAA,UAAA,aAAR,WACE,MAAO,GAAO,gBAIR,EAAA,UAAA,eAAR,WACE,MAAO,GAAO,kBAYR,EAAA,UAAA,MAAR,SAAc,GACZ,GAAM,GAAI,KAAK,qBAAqB,EAAE,OAAQ,EAAE,MAAO,EAAE,OACnD,EAAS,CAAC,EAAC,GACX,EAAO,SAAC,GAAe,MAAC,CAC5B,EAAG,WACD,GAAM,GAAQ,UACR,EAAa,CAAC,EAAG,GACjB,EAAQ,CAAC,MAAK,GAEpB,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAC5B,EAAoC,KAAiB,GACrD,MAGF,EAAkB,GACxB,YAAK,YAAY,KAAK,MAAM,YAAY,KAAM,EAAQ,CAAC,GAAI,EAAM,EAAO,IACjE,GAgBT,EAAA,UAAA,UAAA,SACI,EAAoB,EAAwB,EAC5C,EAAyB,GAC3B,GAAM,GAAoB,KACpB,EAAsB,KAI5B,MAAO,MAAK,cACR,EAAa,EAAQ,EAAe,EAAY,EAAO,EACvD,IAGE,EAAA,UAAA,uBAAR,WACE,MAAO,MAAK,IAAI,QAAQ,YAGlB,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAkB,KAAK,QAAQ,aAGjC,EAAmB,EACvB,EAAS,QAAQ,SAAA,GAGf,GAAqB,EAAK,QAAU,YAAc,EAAI,IAQxD,GAAM,GACF,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,GACjE,EACF,EAAkB,EAAmB,EAAmB,EAC5D,GAAI,EAAgB,EAClB,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,iCAC5B,KAAI,EAAa,6BAA6B,EAAU,OAQhE,EAAA,UAAA,cAAA,SACI,EAA6B,EAC7B,EACA,EAAqB,EAAsB,EAC3C,GAJJ,GAAA,GAAA,KAKM,EACA,EAAkB,GAChB,EAAW,KAAK,WACtB,AAAI,GAAc,MAChB,GACI,KAAK,MAAM,aAAe,KAAO,KAAK,MAAM,YAAY,KAAO,IAGrE,GAAM,GAAoB,KAAK,MAAM,SAC/B,EAAqB,KAAK,MAAM,WAEtC,AAAI,KAAK,0BACP,KAAK,MAAM,kBAAkB,KAAK,GAGpC,GAAI,GACE,EAAS,GAAU,EAAY,KAAK,aACtC,EACJ,GAAI,GAAU,KACZ,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAO,WAAW,CAAC,OAAM,EAAE,MAAK,EAAE,QAAS,EAAK,UACtD,GAAM,GAAW,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC7C,AAAI,EAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAE3D,GAAM,GAAa,EAAS,IACxB,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,MACb,MAAA,GAAK,qBAAqB,EAAQ,EAAO,KAMjD,GAAI,GACF,GAAI,GACA,EAAK,sBAAsB,EAAY,EAAQ,GACnD,GAAI,GAAiB,MAKnB,AAAI,GAAiB,MACnB,GAAgB,IAElB,GAAM,GAAa,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAC7D,EAAiB,IAAgB,IAAI,QAAQ,OAAO,GAEtD,EAAQ,EAAK,2BAA2B,GAE1C,MAAO,SAGT,GAAM,GAAyB,SAAC,GAI9B,GAAI,CAAC,EACH,OAEF,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,OAGrD,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAK,KAAK,WAAM,MAAA,GAAY,EAAK,QAAS,KAChD,GAAM,GAAQ,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC1C,MAAI,GAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAEpD,GAKX,GAAI,GACJ,YAAK,UACD,WAAM,MAAA,GAAK,MAAM,eAAe,WAAM,MAAA,GAAK,MAAM,eAAe,WAC9D,AAAI,CAAC,EAAK,IAAI,QAAQ,UAAY,CAAC,EAAK,MAAM,UAC5C,EAAU,IAEV,GAAgB,EAAK,SAAS,cAC1B,EAAY,EAAQ,WAAM,MAAA,OAC9B,AAAI,EAAK,IAAI,QAAQ,UACnB,EAAK,SAAS,iBAAiB,GAEjC,EAAU,EAAc,WAIhC,AAAI,GACF,KAAK,YACD,EAAY,EAAQ,EAAS,EAAe,EAAO,GAGzD,AAAI,KAAK,MAAM,WACb,KAAK,MAAM,cAAc,QAAQ,KAAK,CACpC,KAAM,EACN,WAAY,KAAK,MAAM,SAAW,EAClC,mBAAoB,KAAK,MAAM,SAC/B,aAAc,KAAK,MAAM,WAAa,EACtC,qBAAsB,KAAK,MAAM,WACjC,YAAa,OAAO,KAAK,GAAQ,IAC7B,SAAA,GAAO,MAAA,GAAO,IAAQ,KAAO,EAAO,GAAK,MAAQ,OACrD,aAAc,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,QACvC,aAAc,EAAc,OAC5B,UAAW,EAAc,YAGrB,MAAM,QAAQ,GAAO,EAAU,EAAQ,IAQzC,EAAA,UAAA,2BAAR,SAAmC,GAAnC,GAAA,GAAA,KACQ,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,MACzD,MAAO,IAaD,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAa,GAAY,GAC/B,GAAI,GAAc,MAChB,GAAM,GAAyB,EAAW,cAAgB,GACpD,EAA2B,EAAW,eAAiB,GAIzD,EAAkB,OACtB,AAAI,EAAW,cACb,GACI,MAAM,QAAQ,GACd,WAAM,MAAA,2DAEV,EAAqB,OAAO,KAAK,GAAQ,IAAI,SAAC,GAAQ,MAAA,GAAO,MAE7D,EAAqB,EAAa,IAAI,SAAC,GAAc,MAAA,GAAO,KAG9D,GAAM,GACF,EAAQ,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAE3C,MAAO,GAAmB,OAAO,GAInC,MAAO,OAQT,EAAA,UAAA,WAAA,SACI,EAAoB,EAAiB,EACrC,GACF,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,iDAElB,EAAQ,GAAS,UACjB,EAAU,GAAW,KAAK,QAC1B,GAAI,GAAc,EAClB,AAAI,IAAU,UAAY,GAAc,EAAO,KAC7C,GAAe,EAAoB,IAAI,SAAA,GAAK,MAAA,IAAkB,MAEhE,GAAM,GAAS,EAAQ,MAAM,EAAa,EAAO,GAC3C,EAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAIhD,GAHA,KAAK,OAAO,EAAG,GAGX,IAAU,UACZ,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAW,GAAqB,GACtC,KAAK,MAAM,UAAY,EAAW,EAAK,MACvC,EAAK,MAAQ,EAEf,MAAO,IAQT,EAAA,UAAA,qBAAA,SACI,EAAgB,EAAiB,EACjC,GACF,EAAQ,GAAS,UACjB,GAAM,GAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAChD,YAAK,OAAO,EAAG,GACR,GAGT,EAAA,UAAA,aAAA,SACI,EAAsB,EAAkB,EACxC,GADsB,AAAA,IAAA,QAAA,GAAA,IAExB,EAAO,GAAQ,KAAK,iBAAiB,WACrC,AAAI,GAAS,MAAQ,IAAU,EAAa,OAC1C,GAAe,EAAa,KAAK,IAEnC,GAAM,GAAI,GAAI,IAAS,EAAc,EAAW,EAAM,KAAK,gBAC3D,GAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,KAC5C,KAAM,IAAI,OAAM,sBAAsB,EAAE,KAAI,2BAE9C,YAAK,MAAM,oBAAoB,EAAE,MAAQ,EACzC,KAAK,OAAO,EAAG,KAAK,SACb,GAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAM,GAAW,KAAK,MAAM,WAAW,IAAI,EAAE,QACzC,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,SACpC,EAKJ,GAJA,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAET,IAAa,GACf,KAAK,MAAM,iBAIX,GAAI,GAAQ,EACZ,AAAI,EAAE,QAAU,aAAe,EAAE,QAAU,UACzC,GAAQ,EAAE,KAAO,GAAqB,EAAE,QAE1C,KAAK,MAAM,WAAW,IAAI,EAAE,OAAQ,CAClC,QAAS,GAAW,KAAK,QACzB,MAAO,EAAE,MACT,MAAO,EAAE,MACT,MAAK,EACL,SAAU,IAEZ,KAAK,MAAM,UAAY,EAGzB,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,WAEpC,AAAM,YAAa,KACjB,KAAK,MAAM,IAIf,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAC,KAAK,MAAM,WAAW,IAAI,EAAE,QAC/B,OAGF,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAEb,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,EAAE,QACnC,EAAW,EAAK,SAEtB,AAAI,GAAY,EAGd,CAAI,EAAE,QAAU,aACd,MAAK,MAAM,UAAY,EAAK,OAE9B,KAAK,MAAM,iBAEX,EAAK,QAAQ,YAAY,EAAE,QAC3B,KAAK,MAAM,WAAW,OAAO,EAAE,SAE/B,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,YAOxC,EAAA,UAAA,iBAAA,WACE,OAAW,KAAW,MAAK,MAAM,qBAC/B,GAAM,GAAI,KAAK,MAAM,oBAAoB,GACzC,KAAK,gBAAgB,KAIzB,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,cAAc,GACnB,AAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,MAC5C,MAAO,MAAK,MAAM,oBAAoB,EAAE,OAI5C,EAAA,UAAA,OAAA,WACE,GAAM,GAAO,KAAK,QAAQ,SAC1B,SAAK,WAAa,KAAK,MAAM,WAC7B,EAAK,eAAiB,KAAK,MAAM,eACjC,EAAK,SAAW,KAAK,MAAM,SAC3B,AAAI,KAAK,MAAM,iBAAmB,GAChC,GAAK,WAAa,GAClB,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,IAEjB,EAAK,QAAQ,KACT,0EAGC,GAGH,EAAA,UAAA,QAAN,SAAc,iHAEZ,YAAK,MAAM,UAAY,GAEjB,EAAa,KAAK,MAAM,SACxB,EAAkB,KAAK,MAAM,WAEnC,KAAK,MAAM,cAAc,QAAU,GACnC,EAAA,KAAK,MAAM,cAAuB,CAAA,EAAM,YAAxC,EAAyB,OAAS,EAAA,OAElC,KAAK,MAAM,UAAY,GAEvB,KAAK,MAAM,cAAc,UAAY,KAAK,IAAG,MAAR,KAC9B,KAAK,MAAM,cAAc,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,sBACnD,KAAK,MAAM,cAAc,SAAW,KAAK,MAAM,SAAW,EAC1D,KAAK,MAAM,cAAc,WACrB,KAAK,MAAM,WAAa,MACP,EAAA,KAAK,MAAM,cAAc,+BAAzB,GAAA,EAAA,OAAV,GAAM,EAAA,GACf,EAAA,EAAsB,CAAA,EAAM,EAAO,eADgB,CAAA,EAAA,UACnD,SAAO,aAAe,EAAA,OACtB,EAAA,EAAmB,CAAA,EAAM,EAAO,kBAAhC,EAAO,UAAY,EAAA,wBAFA,wBAIrB,MAAA,CAAA,EAAO,KAAK,MAAM,qBAGpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,MAAM,cAAgB,GAAK,KAAK,MAAM,cAAgB,GAG5D,EAAA,UAAA,YAAR,SACI,EAAoB,EAAwB,EAC5C,EAAyB,EAAiB,GAF9C,GAAA,GAAA,KAGQ,EACF,CAAC,GAAI,KAAK,MAAM,iBAAkB,WAAU,EAAE,OAAM,EAAE,QAAO,EAAE,MAAK,GAElE,EAAa,GAAY,GAC/B,AAAI,GAAc,MAChB,GAAgB,EAAW,UAE7B,AAAI,GAAiB,MACnB,GAAS,SAAW,SAAC,GAGnB,SAAM,EAAI,IAAI,SAAC,EAAI,GACjB,GAAI,GAAM,MACR,GAAM,GAAS,EAAQ,GACjB,EAAO,GAAyB,EAAO,KAAM,EAAO,OAC1D,MAAO,GAAK,WAAW,EAAM,EAAO,MAAO,EAAO,OAEpD,MAAO,KAIF,EAAc,EAAI,OAAS,EAAI,EAAM,EAAI,GAAI,EAAO,KAG/D,KAAK,MAAM,WAAW,KAAK,IAG7B,EAAA,UAAA,KAAA,SAAuB,GACrB,SAAO,KAAO,GACP,GAGD,EAAA,UAAA,UAAR,WACE,AAAI,KAAK,MAAM,gBAAkB,GAC/B,MAAK,MAAM,WAAa,IAE1B,KAAK,MAAM,iBAGL,EAAA,UAAA,QAAR,WACE,KAAK,MAAM,iBAOb,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAwB,CAC5B,MAAO,GACP,KAAM,gBACN,GAAI,KAAK,MAAM,eAEjB,AAAI,GACF,GAAU,KAAO,GAEnB,KAAK,MAAM,WAAW,KAAK,GAC3B,KAAK,MAAM,YAAc,GAO3B,EAAA,UAAA,SAAA,SAAS,GAMP,OANF,GAAA,KACQ,EAAyB,GAAsB,GAC/C,EACF,GAAI,KAAI,EAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,MAGrC,EAAI,EAAG,EAAI,KAAK,MAAM,YAAY,MAAM,OAAQ,KACvD,GAAM,GAAS,KAAK,MAAM,YAAY,MAAM,GAC5C,AAAI,CAAC,EAAO,MAAQ,CAAC,EAA0B,IAAI,EAAO,KACxD,EAAO,UAIX,GAAM,GAAW,KAAK,MAAM,WAAW,MACvC,KAAK,MAAM,YAAc,KAAK,MAAM,WAAW,SAAW,EACtD,KACA,KAAK,MAAM,WAAW,KAAK,MAAM,WAAW,OAAS,GAGzD,EAAuB,QAAQ,SAAA,GAG7B,AAAI,CAAC,EAAO,MAAQ,EAAO,UAAY,EAAS,IAC9C,EAAK,MAAM,MAWjB,EAAA,UAAA,UAAA,SACI,EAAY,EAAc,EAC1B,GAFJ,GAAA,GAAA,KAKE,GAHE,AAAA,IAAA,QAAA,GAAA,IACF,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,8CACrB,GAAM,MAAQ,EAAG,QAAU,UAC7B,KAAM,IAAI,OAAM,0CAA0C,EAAG,MAAK,KAGpE,GAAM,GAAI,KAAK,UACX,WAAM,MAAA,GAAK,aAAa,WAAM,MAAA,GAAK,WACnC,WAAM,MAAA,GAAK,KAAK,UAAW,KAE/B,EACI,YAAa,GACb,WAAM,MAAA,mDAEV,GAAM,GAAe,GAAqB,KAAK,MAAM,WAAY,EAAI,GACrE,GAAI,CAAC,GAAoB,EAAa,SAAW,GAAK,EAAG,OAAS,EAChE,KAAM,IAAI,OACN,uIAKN,MAAO,MAAK,KAAK,WAAY,WAC3B,GAAM,GAAuD,GAC7D,EAAuB,EAAE,IAAO,GAAc,GAAK,EAAE,OAGrD,GACI,EAAwB,EAExB,SAAA,GAAK,MAAA,GAAK,KAAK,IAEf,IACJ,GAAM,GAAQ,EAAG,IAAI,SAAA,GAAK,MAAA,GAAuB,EAAE,MAEnD,MAAI,GAAK,MAAM,gBAAkB,GAG/B,GAAK,MAAM,WAAW,QAAQ,SAAA,GAC5B,OAAqB,GAAA,EAAA,EAAA,EAAK,MAAL,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,EAAO,aAGX,EAAK,MAAM,WAAa,MAEnB,CAAC,MAAO,EAAG,MAAK,MAI3B,EAAA,UAAA,WAAA,SAA6B,GAA7B,GAAA,GAAA,KAEE,SACI,GAAgB,GAChB,WAAM,MAAA,sDACH,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACN,EACI,EAAO,MAAM,SAAA,GAAK,MAAA,aAAa,KAC/B,WAAM,MAAA,qEAGV,GAAI,GAIE,EAA2B,GACjC,SAAO,QAAQ,SAAC,EAAO,GACrB,EAAS,GAAK,IAET,EAAK,cACR,SAAC,EAAG,GACF,SAAM,EAAC,MAAA,OAAQ,EAAM,OAAA,CAAE,KACvB,EACI,EAAI,gBAAiB,GACrB,WAAM,MAAA,+FAEV,EACI,GAAgB,EAAI,UACpB,WAAM,MAAA,qGAEH,EAAI,OAEb,EACA,SAAC,EAAO,GACN,GAAM,GAAU,EAAI,SAAS,EAAI,GAC3B,EACF,MAAM,QAAQ,GAAW,EAAU,CAAC,GACxC,EACI,EAAM,SAAW,EAAO,OACxB,WAAM,MAAA,wKAGV,EACI,EAAM,MAAM,SAAA,GAAK,MAAA,aAAa,KAC9B,WAAM,MAAA,yIAGV,GAAM,GAAyC,GAC/C,SAAM,QAAQ,SAAC,EAAM,GACnB,EAAQ,GAAK,WAAM,MAAA,MAEd,MAKjB,EAAA,UAAA,SAAA,SAAS,GAEP,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,SAAS,IAE/B,EAAA,UAAA,KAAA,SAAK,GAEH,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,KAAK,IAGrB,EAAA,UAAA,KAAN,SAAW,qGACH,SAAQ,KACK,CAAA,EAAM,KAAK,QAAQ,KAAK,WAArC,SAAa,EAAA,OACnB,EAAW,OAAS,KAAQ,EAC5B,CAAA,EAAO,SASD,EAAA,UAAA,MAAR,SAAgC,GAC9B,MAAI,MAAK,MAAM,aAAe,MAC5B,GAAO,QAAU,KAAK,MAAM,YAAY,GACxC,KAAK,MAAM,YAAY,MAAM,KAAK,IAG7B,GAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAOpB,EAAA,UAAA,MAAA,WAEE,KAAK,uBAEL,KAAK,MAAM,UACX,KAAK,IAAI,QACT,KAAK,MAAQ,GAAI,IAEjB,OAAW,KAAe,MAAK,SAC7B,KAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,GAEvB,KAAK,YAAc,KACnB,KAAK,gBAAkB,KACvB,KAAK,mBAAqB,MA/sBb,EAAA,aAAe,EAKf,EAAA,eAAiB,EA4sBlC,KAEA,YAAc,GACZ,GAAM,GAAS,GAAmB,GAAc,GAAQ,WACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,yBAIxC,GAAM,GAAK,KACX,GAAI,EAAG,WAAa,MAClB,GAAM,GAAc,GAAI,IAAY,GACpC,EAAG,UAAY,GAAI,IAAO,GAE5B,UAAqB,EAAG,UAAU,KAIlC,GAAiB,WAAM,MAAA,GAAG,YACnB,EAAG,UAGL,GAAM,GAAS,iBAQF,EAAW,GAE7B,GAAM,GAAS,CAAC,EAAC,EAAE,EAAC,GACpB,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,EAAG,GAC3B,SAAK,CAAC,EAAG,IACF,GACN,EAAgC,KAAqB,ICzrC1D,AAkBA,cACE,MAAO,OAAO,YAAc,aAAe,WAAa,mBAIxD,GAAI,MAEF,GAAM,GAAI,UAAU,WAAa,UAAU,QAAW,OAAe,MAErE,MAAO,2TACK,KAAK,IAEb,0kDACK,KAAK,EAAE,OAAO,EAAG,IAE5B,MAAO,iBAIP,MAAQ,OAAO,SAAW,aAAe,OAAO,UAAY,MAEvD,MAAO,oBAAsB,6DCvCpC,AAqBA,GAAM,IAAM,KAOZ,GAAI,aAAa,QAAS,WAAM,MAAA,IAAO,SAAA,GACrC,AAAI,GACF,QAAQ,KACJ,iJAOR,GAAI,aAAa,aAAc,WAAM,MAAA,QAGrC,GAAI,aACA,UACA,WAAM,MAAC,OAAO,UAAY,aACrB,MAAO,SAAQ,UAAa,aAC5B,MAAO,SAAQ,SAAS,MAAS,cAG1C,GAAI,aACA,YACA,WAAM,MAAA,OAAO,YAAc,aAAe,WAAa,MACnD,UAAU,WAAa,MAAQ,SAAS,KAAK,UAAU,YACvD,aAAa,KAAK,UAAU,UAMpC,GAAI,aAAa,OAAQ,WAAM,MAAA,KAM/B,GAAI,aACA,qCAAsC,WAAM,MAAA,IAAI,QAAQ,WAG5D,GAAI,aAAa,+BAAgC,WAAM,MAAA,KAGvD,GAAI,aAAa,UAAW,WAAM,MAAA,KCvElC,YAuB2B,EAAiB,GAC1C,GAAI,GAAwB,EAE5B,GAAI,GAAa,GACf,MAAO,KAAU,SAAW,GAAK,CAAC,EAAI,QAExC,GAAI,CAAC,MAAM,QAAQ,GACjB,MAAO,GAIT,OAFM,GAAkB,GAEjB,MAAM,QAAQ,IACd,GAAa,IAAc,IAAU,UAC1C,EAAM,KAAK,EAAU,QACrB,EAAY,EAAU,GAExB,MAAI,OAAM,QAAQ,IACd,KAAM,QAAQ,uCAChB,GAA2B,EAAK,EAAO,IAGlC,EAGT,YACI,EAAiB,EAAiB,GAEpC,GADA,EAAU,GAAW,GACjB,CAAE,MAAM,QAAQ,IAAS,CAAC,GAAa,IACzC,EACI,EAAM,SAAW,EACjB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,qBACnC,yCAAwC,EAAM,GAAE,eACxD,OAEF,EACI,EAAM,OAAS,EACf,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,4BACnC,uBAAsB,EAAI,OAAM,eACxC,EACI,EAAI,SAAW,EAAM,GACrB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,iBAAiB,EAAM,GAAE,IAC5D,sBAAqB,EAAI,OAAM,eAEvC,OADM,GAAW,EAAM,MAAM,GACpB,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAA2B,EAAI,GAAI,EAAU,EAAQ,OAAO,IAIhE,YACI,EAAmC,EAAuB,EAC1D,GACF,GAAI,GAAiB,KACnB,OAEF,GAAI,IAAkB,WAAa,IAAkB,GACjD,IAAkB,WAAa,IAAgB,SACjD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,UAChD,OAAM,EAAa,oBAAoB,EAAW,uBAKtD,EAAiB,EAAiB,EAClC,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,YAAa,GACf,UAAY,EAAc,EAAE,MAAO,EAAS,GACrC,EAET,GAAI,GAAgB,GAAW,GAS/B,GANA,AAAI,IAAkB,UAClB,CAAC,OAAQ,QAAS,WAAW,QAAQ,IAAiB,GACxD,GAAgB,GAElB,GAAY,EAAc,EAAe,EAAS,GAE7C,GAAK,MACL,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IAAM,MAAO,IAAM,UACtD,MAAO,IAAM,WAAa,MAAO,IAAM,UAC1C,GAAM,GAAO,GAAK,KAAO,OAAU,EAAS,YAAY,KACxD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,eAChD,mCAAkC,EAAI,MAE5C,GAAM,GAAgB,GAAW,EAAG,GACpC,AAAI,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IACrC,GAAI,CAAC,IAEP,GAAM,GAAiB,GACjB,EAAS,IAAkB,SAC7B,GAAa,EAAG,GAChB,GAAQ,EAAe,GAAI,GAC/B,MAAO,GAAO,WAAW,EAAQ,EAAe,eAI9C,EAA0B,EAAiB,EAC3C,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,OACN,YAAY,EAAO,cAAc,EAAY,2CAGnD,GAAM,GAAU,EAChB,MAAO,GAAQ,IACX,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAM,EAAO,IAAI,EAAC,IAAK,IACjD,GCnIN,GAkBa,IAAkB,kBAOQ,GACrC,GAAM,GAAO,OAAO,KAAK,GACzB,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,OACN,yGAEG,GAAK,OAAM,WAGpB,GAAI,GAAS,EAAK,GACZ,EAAK,EAAE,GAGb,AAAI,EAAO,SAAS,MAClB,GAAS,EAAO,UAAU,EAAG,EAAO,OAAS,IAI/C,EAAS,EAAS,GAGlB,GAAM,GAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,EAAO,WAAW,GAClB,IACE,GAAM,GAAS,EAAE,MAAA,OAAI,GACrB,MAAI,aAAkB,UACpB,QAAQ,MAAM,2CAEhB,EAAO,SAAS,GACT,QACA,GACP,QAAO,SAAS,MACV,IAGV,cAAO,eAAe,EAAI,OAAQ,CAAC,MAAO,EAAQ,aAAc,KAGzD,EC/DT,AA8CA,YAAoC,EAAoB,GACtD,GAAM,GAAQ,EAAgB,EAAM,OAAQ,WACtC,EAAQ,EAAgB,EAAM,OAAQ,WAC5C,GACI,EAAM,MAAO,EAAM,MACnB,yBAAyB,EAAM,MAAK,QAAQ,EAAM,MAAK,yCAG3D,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,QAAQ,EAAO,IAE1B,EAAwB,CAAC,KAAM,EAAO,KAAM,GAClD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC/DnC,YAyBI,EAAoB,EAAiB,EACrC,GAIF,GAHA,AAAI,GAAS,MACX,GAAQ,GAAW,IAEjB,IAAU,YACZ,KAAM,IAAI,OACN,oFAGN,GAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IACxC,MAAO,IAAW,UAAY,MAAO,IAAW,WAChD,MAAO,IAAW,SACpB,KAAM,IAAI,OACN,4HAGN,GAAI,GAAS,MACX,GAAmC,GAEnC,GAAM,GAAe,GAAc,GAC7B,EAAe,GAAc,GACnC,EACI,IAAiB,EACjB,WACI,MAAA,iCAAiC,EAAK,6BACnC,GAAY,mBAAmB,KAE1C,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,EAAE,GAC1C,GAAM,GAAW,EAAc,GACzB,EAAoB,IAAM,EAAc,OAAS,EACnD,IAAa,GAAc,EAAM,MAAM,IACvC,GACJ,EACI,EAAc,KAAO,EAAM,IAAM,CAAC,EAClC,WAAM,MAAA,+CACF,KAAI,EAAa,kCACjB,WAAU,EAAK,UAI3B,MAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IAC1C,GAAS,CAAC,IAGZ,EAAQ,GAAS,EACjB,EAAS,IAAU,SACf,GAAa,EAAQ,GACrB,GAAQ,EAAoB,GAAI,IAC7B,EAAO,WAAW,EAAsB,EAAO,GC1ExD,YAqDI,EAAoB,EAAqB,GAC3C,GAAM,GAAgB,GAAW,EAAQ,GACzC,MAAO,IAAW,EAAQ,EAAO,EAAe,GCvDlD,AAsBO,GAAM,IAAkD,CAC7D,QAAW,EACX,QAAW,EACX,MAAS,EACT,OAAU,EACV,MAAS,EACT,KAAQ,EACR,UAAa,GC7Bf,AA2BA,GAAM,IAA0B,cAoB5B,EAAuC,oHAUzC,IAPM,EAAgC,GAChC,EAA2C,GAE3C,EAAkB,MAAM,QAAQ,GAClC,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,OAC7B,OAAO,KAAK,cAEP,GACP,GAAM,GAAO,EAAM,GACb,EAAI,MAAM,QAAQ,GAAW,EAAQ,GAAG,OAAS,EAAQ,GAC/D,GAAI,EAAE,QAAU,WAAa,EAAE,QAAU,SAAW,EAAE,QAAU,QAC5D,EAAE,QAAU,UAAY,EAAE,QAAU,YACtC,KAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,EAAE,OAE9D,GAAM,GAA6B,CAAC,KAAI,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,OACnE,GAAI,EAAE,QAAU,UACd,GAAM,GAAY,GAAI,SAAoB,SAAM,GAAO,MAAA,IAAA,EAAA,OAAA,OAAA,+EACxC,MAAA,CAAA,EAAM,EAAE,gBAKrB,IALM,EAAO,EAAA,OACP,EAAgB,EAAK,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAE,QAAQ,GACtD,GAA0B,EAAK,OAC7B,EAAQ,GAAI,YAAW,GACzB,EAAS,EACJ,EAAI,EAAG,EAAI,EAAK,OAAQ,IACzB,EAAM,EAAK,GACX,EACF,GAAI,YAAW,GAAI,aAAY,CAAC,EAAI,SAAS,QACjD,EAAM,IAAI,EAAe,GACzB,GAAU,GACV,EAAM,IAAI,EAAK,GACf,GAAU,EAAI,OAEhB,SAAQ,cAEV,EAAa,KAAK,OAElB,GAAa,KAAK,EAAE,QAEtB,AAAI,GAAS,MACX,GAAK,MAAQ,GAEf,EAAM,KAAK,IAjCJ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,IAA3B,GAoCY,MAAA,CAAA,EAAM,QAAQ,IAAI,WAAjC,SAAe,EAAA,OACrB,CAAA,EAAO,CAAC,KAAM,GAAuB,GAAe,MAAK,qBAmBvD,EAAqB,GAKvB,OAHM,GAAsB,GACxB,EACA,EAAS,EACM,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAO,EAAK,KACZ,EAAQ,EAAK,MACb,EAAQ,EAAK,MACb,EAAO,GAAc,GACvB,EAAM,OAEV,GAAI,gBAAkB,IACpB,GAAM,GAAe,EAAK,aAC1B,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,GAAI,CAAE,QAAS,IAAgB,SAAW,IACxC,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,+DAGxD,EAAa,QAAU,WAChC,GAAI,IAAU,UACZ,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,IAC3D,oDAAmD,EAAK,UAG9D,MAAM,IAAI,OACN,UAAU,EAAK,KAAI,gBACnB,uBAAsB,EAAa,MAAK,MACxC,wEAGN,GAAM,GAAyB,GAAqB,EAAa,OAC3D,EACF,EAAO,MAAM,EAAQ,EAAS,EAAO,GACnC,EAAkB,EAAa,QAAU,QAC3C,GAAI,YAAW,GACf,GAAI,aAAY,GACpB,GAAI,IAAU,UACZ,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,EAAS,GAAI,cAAa,EAAe,QACzC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,EAAI,EAAa,MAAQ,EAAa,aAE3C,EAAa,QAAU,UAChC,AAAI,IAAkB,QACpB,GAAgB,MAElB,EAAS,EAAc,OAEvB,MAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,qCAGhD,IAAU,SACnB,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,SAC3D,KAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,2BAGzD,EAAS,GAAI,YAAW,EAAe,QACvC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,KAAK,MAAM,EAAI,EAAa,MAAQ,EAAa,UAG/D,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,UACR,IAAU,UACnB,GAAM,GAAO,GAAc,EAAK,OAChC,EAAS,GACT,OAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAa,GAAI,aACnB,EAAO,MAAM,EAAQ,EAAS,KAA0B,GAC5D,GAAU,GACV,GAAM,GAAQ,GAAI,YAAW,EAAO,MAAM,EAAQ,EAAS,IAC1D,EAAwB,KAAK,GAC9B,GAAU,QAGZ,GAAM,GAAc,GAAqB,GACnC,EAAa,EAAO,MAAM,EAAQ,EAAS,EAAO,GAExD,GAAI,IAAU,UACZ,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,aACnB,EAAS,GAAI,cAAa,GAG1B,OAFM,GAAO,GAAI,cAAa,EAAO,OAAS,GACxC,EAAQ,GAAI,cAAa,EAAO,OAAS,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAK,GAAK,EAAO,EAAI,GACrB,EAAM,GAAK,EAAO,EAAI,EAAI,GAE5B,GAAM,GAAa,GAAO,EAAM,EAAO,WACjC,EAAc,GAAO,EAAO,EAAO,WACzC,EAAI,GAAQ,GAAQ,EAAY,OAEhC,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,EAEnB,AAAI,IAAU,aACZ,GAAI,GAAQ,GAAO,EAAQ,EAAO,IAGtC,MAAO,eAM8B,GAErC,GAAI,IAAO,KACT,KAAM,IAAI,OAAM,wBAAwB,KAAK,UAAU,IAGzD,GAAI,GAAkB,EAShB,EAA6B,GACnC,EAAG,QAAQ,SAAC,GAMV,GALA,GAAmB,EAAE,WAErB,EAAa,KACT,EAAE,aAAe,EAAE,OAAO,WAAa,EACA,GAAK,GAAE,YAAoB,IAClE,CAAE,aAAoB,eAAgB,YAAoB,aACxD,YAAoB,aACxB,KAAM,IAAI,OAAM,mCAAmC,EAAE,YAAY,QAKrE,GAAM,GAAI,GAAI,YAAW,GACrB,EAAS,EACb,SAAa,QAAQ,SAAC,GACpB,EAAE,IAAI,GAAI,YAAW,EAAE,QAAS,GAChC,GAAU,EAAE,aAGP,EAAE,OAIX,GAAM,IAAgB,MAAO,SAAW,aACnC,OAAO,OAAS,aAAe,MAAO,OAAS,aAC/C,MAAO,OAAS,yBAWY,GAC/B,MAAI,IACK,OAAO,WAAW,GAEpB,GAAI,MAAK,CAAC,IAAM,iBASiB,GACxC,GAAI,GACF,MAAO,QAAO,KAAK,GAAQ,SAAS,UAItC,OAFM,GAAM,GAAI,YAAW,GACvB,EAAI,GACC,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAI,EAAG,IACrC,GAAK,OAAO,aAAa,EAAI,IAE/B,MAAO,MAAK,eAS4B,GACxC,GAAI,IACF,GAAM,GAAM,OAAO,KAAK,EAAK,UAC7B,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YAI/D,OAFM,GAAI,KAAK,GACT,EAAS,GAAI,YAAW,EAAE,QACvB,EAAI,EAAG,EAAI,EAAE,OAAQ,EAAE,EAC9B,EAAO,IAAI,CAAC,EAAE,WAAW,IAAK,GAEhC,MAAO,GAAO,mBASwB,GACtC,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAI,GAAkB,EACtB,EAAQ,QAAQ,SAAC,GACf,GAAmB,EAAO,aAG5B,GAAM,GAAO,GAAI,YAAW,GACxB,EAAS,EACb,SAAQ,QAAQ,SAAC,GACf,EAAK,IAAI,GAAI,YAAW,GAAS,GACjC,GAAU,EAAO,aAEZ,EAAK,mBAUW,GACvB,GAAM,GAAY,IAElB,IADA,EAAO,EAAK,OACL,EAAK,SAAS,IACnB,EAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAErC,GAAM,GAAQ,EAAK,MAAM,GACzB,MAAO,GAAM,EAAM,OAAS,eAQe,GAE3C,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OAAM,uDAGlB,MAAO,CACL,UAAW,GAAI,MACf,kBAAmB,OACnB,mBAAoB,EAAe,eAAiB,KAChD,EACA,GAAiB,KAAK,UAAU,EAAe,gBACnD,iBAAkB,EAAe,aAAe,KAC5C,EACA,GAAiB,KAAK,UAAU,EAAe,cACnD,gBAAiB,EAAe,YAAc,KAC1C,EACA,EAAe,WAAW,YAUlC,cACE,GAAM,GAAkB,SAAC,GAIvB,OAHI,GAAI,GAAK,GACT,EAAI,EAEA,GAAI,WAAgB,GAC1B,GAAK,QACL,IAAM,EAER,UAAK,CAAC,QACN,GAAK,UAEE,EAAI,GAGP,EAAe,GAAI,aAAY,MAErC,EAAa,GAAK,EAClB,OAAS,GAAI,EAAG,EAAI,KAAM,IACxB,EAAa,GAAK,EAAgB,GAEpC,OAAS,GAAI,KAAM,EAAI,KAAM,IAC3B,EAAa,GAAK,UAAe,GAAI,MAAS,IAGhD,MAAO,GAST,cACE,GAAM,GAAgB,GAAI,aAAY,IAEtC,EAAc,GAAK,EACnB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,OAAS,GAAI,EAAG,EAAI,GAAI,IACtB,EAAc,GAAK,GAAK,GAE1B,OAAS,GAAI,GAAI,EAAI,GAAI,IACvB,EAAc,GAAK,WAAe,GAAI,IAAO,IAG/C,MAAO,GAST,cAGE,OAFM,GAAc,GAAI,aAAY,IAE3B,EAAI,EAAG,EAAI,GAAI,IACtB,EAAY,GAAK,KAEnB,SAAY,GAAK,EAAY,IAAM,EAE5B,gBAeP,GAAM,GAAe,KACf,EAAgB,KAChB,EAAc,KAEpB,MAAO,UAAC,GAGN,OAFM,GAAS,GAAI,aAAY,EAAI,EAAe,QAC5C,EAAmB,GAAI,aAAY,GAChC,EAAQ,EAAG,EAAQ,EAAe,OAAQ,KACjD,GAAM,GAAc,EAAe,GAC7B,EACF,EAAa,EAAY,GAAe,IAAO,GAAc,OAC7D,EAAc,GAAe,IACjC,EAAiB,GAAS,EAE5B,MAAO,IAAI,cAAa,IChf5B,AAsBA,GAAA,IAAA,WAOE,aACE,KAAK,YAAc,GACnB,KAAK,YAAc,GAGN,SAAA,YAAf,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UASnB,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAS3C,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAW3C,EAAA,gBAAP,SAAuB,GACrB,MAAO,GAAiB,YAAY,EAAK,SAWpC,EAAA,gBAAP,SAAuB,EAAsB,GAE3C,MAAO,GAAiB,YAAY,EAAK,OAAQ,IAGpC,EAAA,YAAf,SACI,EAAsB,EACtB,GACF,GAAM,GAA6B,GAC7B,EAAU,IAAgB,OAC5B,EAAiB,cAAc,YAC/B,EAAiB,cAAc,YACnC,SAAQ,QAAQ,SAAA,GACd,GAAM,GAAU,EAAO,EAAK,GAC5B,AAAI,IAAY,MACd,EAAc,KAAK,KAGhB,GAEX,KAEa,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAkB,SAAC,GAC5B,MAAA,IAAiB,gBAAgB,IACxB,GACT,SAAC,EAAsB,GACnB,MAAA,IAAiB,gBAAgB,EAAK,IC/G9C,AAyBA,GAAM,IAAgB,eAChB,GAAmB,EAKnB,GAAmB,eAInB,GAAkB,mBAexB,cACE,GAAI,CAAC,KAAM,QAAQ,cAIjB,KAAM,IAAI,OACN,2FAIN,GAAM,GAAiB,MAAO,SAAW,YAAc,KAAO,OACxD,EAAU,EAAU,WAAa,EAAU,cAC7C,EAAU,iBAAmB,EAAU,aACvC,EAAU,cACd,GAAI,GAAW,KACb,KAAM,IAAI,OACN,6DAEN,MAAO,GAGT,YAAuB,GACrB,GAAM,GAAK,EAAY,OACvB,EAAG,kBAAkB,GAAkB,CAAC,QAAS,cACjD,EAAG,kBAAkB,GAAiB,CAAC,QAAS,cAQlD,GAAA,IAAA,WAME,WAAY,GAGV,GAFA,KAAK,UAAY,KAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,kEAEN,KAAK,UAAY,EAGb,SAAA,UAAA,KAAN,SAAW,sEAET,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAIN,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,UAAW,SAIvC,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,iBAiB1B,EAAA,UAAA,eAAR,SAAuB,EAAmB,GAA1C,GAAA,GAAA,KAEE,MAAO,IAAI,SAAmC,SAAC,EAAS,GACtD,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OAEvB,GAAI,GAAkB,MAEpB,GAAM,GAAU,EAAG,YAAY,GAAkB,YAC3C,EAAa,EAAQ,YAAY,IACjC,EAAa,EAAW,IAAI,EAAK,WACvC,EAAW,UAAY,WACrB,GAAI,EAAW,QAAU,KACvB,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAK,UAAS,oBAGlD,EAAQ,EAAW,OAAO,iBAG9B,EAAW,QAAU,SAAA,GACnB,SAAG,QACI,EAAO,EAAW,QAE3B,EAAQ,WAAa,WAAM,MAAA,GAAG,cAG9B,GAAM,GACF,GAA6B,GAE3B,EAAS,EAAG,YAAY,GAAiB,aAC3C,EAAY,EAAO,YAAY,IAC7B,EACF,EAAU,IAAI,CAAC,UAAW,EAAK,UAAW,mBAAkB,IAC5D,EACJ,EAAe,UAAY,WAEzB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAkB,EAAW,IAAI,CACrC,UAAW,EAAK,UAChB,eAAc,EACd,mBAAkB,IAEpB,EAAgB,UAAY,WAAM,MAAA,GAAQ,CAAC,mBAAkB,KAC7D,EAAgB,QAAU,SAAA,GAGxB,EAAY,EAAO,YAAY,IAC/B,GAAM,GAAoB,EAAU,OAAO,EAAK,WAChD,EAAkB,UAAY,WAC5B,SAAG,QACI,EAAO,EAAgB,QAEhC,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAgB,UAIpC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAE/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,YAKtC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,WAvHtC,EAAA,WAAa,eA0H/B,KAEa,GAA4B,SAAC,GACxC,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAmBH,GAC/B,MAAO,IAAI,IAAiB,GAG9B,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAiB,YACnC,EAAI,MAAM,GAAiB,WAAW,QACtC,EAGN,GAAA,IAAA,WAGE,aACE,KAAK,UAAY,KAGb,SAAA,UAAA,WAAN,yFACE,MAAA,CAAA,EAAO,GAAI,SACP,SAAC,EAAS,GACR,GAAM,GACF,EAAK,UAAU,KAAK,GAAe,IACvC,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAK,EAAG,YAAY,GAAiB,YACrC,EAAQ,EAAG,YAAY,IASvB,EAAqB,EAAc,SACzC,EAAkB,UAAY,WAE5B,OADM,GAA4C,GAC/B,EAAA,EAAA,EAAA,EAAkB,OAAlB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAI,EAAK,WAAa,EAAK,mBAE7B,EAAQ,IAEV,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAkB,QAElC,EAAG,WAAa,WAAM,MAAA,GAAG,UAE3B,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAIpD,EAAA,UAAA,YAAN,SAAkB,iFAChB,SAAO,GAAiB,GACxB,CAAA,EAAO,GAAI,SAA4B,SAAC,EAAS,GAC/C,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAS,EAAG,YAAY,GAAiB,aACzC,EAAY,EAAO,YAAY,IAE/B,EAAiB,EAAU,IAAI,GACjC,EACJ,EAAe,UAAY,WACzB,GAAI,EAAe,QAAU,KAC3B,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAI,oBAIxC,GAAM,GAAoB,EAAU,OAAO,GACrC,EAAkB,WAEtB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAqB,EAAW,OAAO,GAC7C,EAAmB,UAAY,WAC3B,MAAA,GAAQ,EAAe,OAAO,qBAClC,EAAmB,QAAU,SAAA,GACzB,MAAA,GAAO,EAAe,SAI5B,EAAkB,UAAY,EAC9B,EAAkB,QAAU,SAAA,GAC1B,WACA,EAAG,QACI,EAAO,EAAe,SAInC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAG/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,WAIpC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAGxD,KCjWA,AAyBA,GAAM,IAAiB,IACjB,GAAc,sBACd,GAAc,OACd,GAAwB,iBACxB,GAAsB,eACtB,GAAqB,cACrB,GAAwB,iBA8B9B,YAAsB,GAOpB,MAAO,CACL,KAAM,CAAC,GAAa,EAAM,IAAa,KAAK,IAC5C,SAAU,CAAC,GAAa,EAAM,IAAuB,KAAK,IAC1D,YAAa,CAAC,GAAa,EAAM,IAAqB,KAAK,IAC3D,WAAY,CAAC,GAAa,EAAM,IAAoB,KAAK,IACzD,cACI,CAAC,GAAa,EAAM,IAAuB,KAAK,KAWxD,YAA6B,GAC3B,GAAM,GAAQ,EAAI,MAAM,IACxB,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OAAM,uBAAuB,GAEzC,MAAO,GAAM,MAAM,EAAG,EAAM,OAAS,GAAG,KAAK,IAG/C,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAoB,YACtC,EAAI,MAAM,GAAoB,WAAW,QACzC,EAgBN,GAAA,IAAA,WAOE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,eAAiB,MAAO,SAAW,aAClD,MAAO,QAAO,cAAiB,YAKjC,KAAM,IAAI,OACN,2DAIN,GAFA,KAAK,GAAK,OAAO,aAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,sEAEN,KAAK,UAAY,EACjB,KAAK,KAAO,GAAa,KAAK,WAY1B,SAAA,UAAA,KAAN,SAAW,gFACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAGE,EAAW,KAAK,UAAU,EAAe,eACzC,EAAc,KAAK,UAAU,EAAe,aAE5C,EACF,GAA6B,GAEjC,IACE,YAAK,GAAG,QAAQ,KAAK,KAAK,KAAM,KAAK,UAAU,IAC/C,KAAK,GAAG,QAAQ,KAAK,KAAK,SAAU,GACpC,KAAK,GAAG,QAAQ,KAAK,KAAK,YAAa,GACvC,KAAK,GAAG,QACJ,KAAK,KAAK,WACV,GAA0B,EAAe,aAC7C,KAAK,GAAG,QAAQ,KAAK,KAAK,cAAe,KAAK,UAAU,CACtD,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,uBAGtC,CAAA,EAAO,CAAC,mBAAkB,UACnB,GAEP,WAAK,GAAG,WAAW,KAAK,KAAK,MAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,UAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,aAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,YAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,eAEvB,GAAI,OACN,yBAAyB,KAAK,UAAS,sFAEvC,uBAAsB,EAAmB,mBAAkB,MAC3D,qBAAoB,EAAmB,iBAAgB,MACvD,oBAAmB,EAAmB,gBAAe,qBAazD,EAAA,UAAA,KAAN,gGAGE,GAFM,EACF,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,OACrC,GAAQ,KACV,KAAM,IAAI,OACN,kDAAkD,KAAK,UAAS,KAGtE,GAAI,EAAK,oBAAsB,OAC7B,KAAM,IAAI,OACN,6EAQN,GAJM,EAAsB,GAGtB,EAAW,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,WAClD,GAAY,KACd,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAS,iBAOhE,GAJA,EAAI,cAAgB,EAGd,EAAc,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,cACrD,GAAe,KACjB,KAAM,IAAI,OACN,gDAAgD,KAAK,UAAS,kBAiBpE,GAdA,EAAI,YAAc,EAGZ,EAAiB,KAAK,GAAG,QAAQ,KAAK,KAAK,eACjD,AAAI,GAAkB,MACd,GAAW,KAAK,MAAM,GAC5B,EAAI,OAAS,EAAS,OACtB,EAAI,YAAc,EAAS,YAC3B,EAAI,YAAc,EAAS,YAC3B,EAAI,oBAAsB,EAAS,qBAI/B,EAAmB,KAAK,GAAG,QAAQ,KAAK,KAAK,YAC/C,GAAoB,KACtB,KAAM,IAAI,OACN,uDACA,KAAI,KAAK,UAAS,mBAExB,SAAI,WAAa,GAA0B,GAE3C,CAAA,EAAO,QAzIO,EAAA,WAAa,kBA2I/B,KAEa,GAA+B,SAAC,GAC3C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAoB,aACrD,GACH,EAAI,MAAM,GAAoB,WAAW,SAJxC,MAUX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBA0BA,GAClC,MAAO,IAAI,IAAoB,GAGjC,GAAA,IAAA,WAGE,aACE,EACI,KAAM,QAAQ,cACd,WAAM,MAAA,6CACV,EACI,MAAO,SAAW,aACd,MAAO,QAAO,cAAiB,YACnC,WAAM,MAAA,4DACV,KAAK,GAAK,OAAO,aAGb,SAAA,UAAA,WAAN,8FAIE,IAHM,EAA4C,GAC5C,EAAS,GAAc,GACvB,EAAS,GAAiB,GACvB,EAAI,EAAG,EAAI,KAAK,GAAG,OAAQ,EAAE,EAC9B,EAAM,KAAK,GAAG,IAAI,GACxB,AAAI,EAAI,WAAW,IAAW,EAAI,SAAS,IACnC,GAAY,GAAoB,GACtC,EAAI,GAAa,KAAK,MAAM,KAAK,GAAG,QAAQ,KAGhD,MAAA,CAAA,EAAO,QAGH,EAAA,UAAA,YAAN,SAAkB,8EAGhB,GAFA,EAAO,GAAiB,GAClB,EAAO,GAAa,GACtB,KAAK,GAAG,QAAQ,EAAK,OAAS,KAChC,KAAM,IAAI,OAAM,8BAA8B,EAAI,KAE9C,SAAO,KAAK,MAAM,KAAK,GAAG,QAAQ,EAAK,OAE7C,KAAK,GAAG,WAAW,EAAK,MACxB,KAAK,GAAG,WAAW,EAAK,UACxB,KAAK,GAAG,WAAW,EAAK,aACxB,KAAK,GAAG,WAAW,EAAK,YACxB,CAAA,EAAO,QAEX,KCvVA,AAiCA,GAAM,IAAoB,MAE1B,GAAA,WAME,aACE,KAAK,SAAW,GAGH,SAAA,YAAf,WACE,MAAI,GAA0B,UAAY,MACxC,GAA0B,SAAW,GAAI,IAEpC,EAA0B,UAS5B,EAAA,gBAAP,SAAuB,EAAgB,GACrC,EAAO,GAAU,KAAM,WAAM,MAAA,0CAC7B,AAAI,EAAO,SAAS,KAClB,GAAS,EAAO,MAAM,EAAG,EAAO,QAAQ,MAE1C,EAAO,EAAO,OAAS,EAAG,WAAM,MAAA,wCAChC,GAAM,GAAW,EAA0B,cAC3C,EACI,EAAS,SAAS,IAAW,KAC7B,WAAM,MAAA,2DACF,EAAM,OACd,EAAS,SAAS,GAAU,GAGvB,EAAA,WAAP,SAAkB,GAChB,GAAM,GAAU,KAAK,cAAc,SAAS,GAC5C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,yCAAyC,EAAM,KAEjE,MAAO,IAGF,EAAA,WAAP,WACE,MAAO,QAAO,KAAK,KAAK,cAAc,WAE1C,KAUA,YAAkB,GAChB,GAAI,EAAI,QAAQ,MAAuB,GACrC,KAAM,IAAI,OACN,6EAEA,IAAG,GAA0B,aAAa,KAAK,OAErD,MAAO,CACL,OAAQ,EAAI,MAAM,IAAmB,GACrC,KAAM,EAAI,MAAM,IAAmB,IAIvC,YACI,EAAmB,EACnB,GAAA,MAAA,KAAA,QAAA,GAAA,6GACF,SACI,IAAc,EACd,WAAM,MAAA,wCAAwC,EAAS,MAErD,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,kEACF,EAAS,MACjB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,iCAAgC,EAAS,OAC3C,EAAc,EAAa,GAE3B,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,mEACF,QAAO,EAAO,OACtB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,sCAAqC,EAAO,OAC9C,EAAc,EAAa,GAE3B,EAAe,GAAS,GAAW,OACnC,EAAa,GAAS,GAAW,KACjC,EAAa,IAAiB,GAAS,GAAW,OAEjC,CAAA,EAAM,EAAY,qBAAnC,GAAiB,EAAA,OAKnB,GAAgB,EAClB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIiB,MAAA,CAAA,EAAM,EAAY,KAAK,iBAApC,GAAa,EAAA,OAKf,GAAgB,CAAC,EACnB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIF,MAAA,CAAA,EAAO,EAAW,yBAsCpB,4HACQ,EAAU,GAA0B,aACpC,EAA2C,OAC5B,EAAA,yBAAA,GAAA,EAAA,OAAV,GAAM,EAAA,GAEX,CAAA,EAAM,GAA0B,WAAW,GAAQ,eAF7B,CAAA,EAAA,UACpB,EACF,EAAA,OACJ,IAAW,IAAQ,GACX,EAAM,EAAS,GAAoB,EACzC,EAAI,GAAO,EAAU,oBALJ,wBAQrB,MAAA,CAAA,EAAO,QAoCT,YAA2B,8EACnB,SAAgB,GAAS,GACzB,EAAU,GAA0B,WAAW,EAAc,QACnE,CAAA,EAAO,EAAQ,YAAY,EAAc,WAkD3C,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QAiDhD,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QChWhD,AA0BA,GAAA,IAAA,WAAA,cAKE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,OAAM,EAAM,IAGrB,EAAA,UAAA,IAAA,WACE,MAAO,aAAY,OAGrB,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,kDAAkD,GAExD,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAAI,cAElB,KAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,IAAI,aAAY,GAAU,OAAO,IAE5C,KAEA,GAAI,KAAM,IAAI,eACZ,KAAM,YAAY,UAAW,GAAI,KAGjC,IACE,GAA0B,gBACtB,GAAoB,WAAY,GAAI,WACjC,IAIT,IACE,GAA0B,gBACtB,GAAiB,WAAY,GAAI,WAC9B,KCpEX,AAqBO,GAAM,IAAe,CAE1B,YAAa,WAAM,MAAA,QAIjB,GAaJ,GAAA,WAKE,aAEE,KAAK,KAAO,KAGZ,KAAK,YAAc,GAAI,MAAK,KAAK,YAGnC,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAI,MAAM,OAAO,OAAS,KACjB,KAAM,OAAO,MAAM,EAAM,GAGlC,CAAI,IAAe,MACjB,IAAc,GAAa,eAEtB,GAAY,EAAM,KAG3B,EAAA,UAAA,IAAA,WACE,GAAM,GAAO,QAAQ,SACrB,MAAO,GAAK,GAAK,IAAO,EAAK,GAAK,KAGpC,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,sDAAsD,GAE5D,MAAO,MAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAI,GAAM,SAAW,EACZ,GAEF,GAAI,MAAK,KAAK,YAAY,GAAU,OAAO,IAEtD,KAEA,AAAI,KAAM,IAAI,YACZ,KAAM,YAAY,OAAQ,GAAI,KCrFhC,YAgDI,EAAoB,EACpB,GADoB,MAAA,KAAA,QAAA,GAAW,WAEjC,EAAQ,GAAS,UACjB,GAAwC,GACjC,GAAI,IAAmB,EAAO,EAAO,GCpD9C,AAuCA,YAAiC,EAAiB,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAGnC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,mCAAmC,GAErD,GAAI,IAAU,UAAY,EAAG,QAAU,UACnC,IAAU,UAAY,EAAG,QAAU,SACrC,KAAM,IAAI,OAAM,yCAGlB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAGhB,IAAO,EAAG,CAAC,MAAK,KC3D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,QAAS,MACtC,EAAU,WACZ,MAAA,GAAO,qBAAqB,EAAG,OAAQ,EAAG,MAAO,EAAG,QAElD,EAAyB,CAAC,EAAG,GAInC,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAQ,EAAG,CAAC,OAAM,KCrD/B,YAgCwC,EAAM,GAAA,AAAA,IAAA,QAAA,GAAA,IAC5C,QAAQ,IAAI,EAAE,SAAS,ICjCzB,AAqBA,KAcA,GAAM,IAAuB,CAC3B,OAAM,GACN,KAAI,GACJ,MAAK,GACL,MAAK,IAEP,GAAa,ICzCb,AA6BA,GAAM,IAA2B,QAC3B,GAA8B,QAC9B,GAAqC,eAE3C,YAAkB,GAChB,MAAO,IAAI,SAAQ,SAAA,GAAW,MAAA,YAAW,KAAU,KAAK,GAG1D,GAAA,IAAA,WAQE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,cAGjB,KAAM,IAAI,OACN,uFAIN,AAAI,EAAe,WAAW,EAAiB,aAC7C,GAAiB,EAAe,MAAM,EAAiB,WAAW,SAEpE,AAAI,IAAkB,MAAQ,EAAe,SAAW,IACtD,GAAiB,IAGnB,KAAK,sBAAwB,EAAiB,GAC9C,KAAK,mBACD,EAAiB,GAGjB,SAAA,UAAA,KAAN,SAAW,6GACT,GAAI,MAAQ,WAAc,YACxB,KAAM,IAAI,OACN,8FAGA,EAAa,OAAO,IAAI,gBAAgB,GAAI,MAC9C,CAAC,EAAe,YAAa,CAAC,KAAM,+BAEpC,GAAe,wBAAyB,cAAxC,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,OACN,gGAGE,SAAyC,CAAC,CAC9C,MAAO,CAAC,KAAO,KAAK,oBACpB,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,gBAAe,GAEX,EACF,OAAO,IAAI,gBAAgB,GAAI,MAC3B,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,sBAIT,EAAa,KAAK,YAAc,KAAO,SAAS,cAAc,KACvB,KAAK,WAClD,EAAW,SAAW,KAAK,sBAC3B,EAAW,KAAO,EAIlB,CAAA,EAAM,GAAM,WAAM,MAAA,GAAW,cAAc,GAAI,YAAW,0BAA1D,GAAA,OAEI,EAAe,YAAc,KACzB,GAAmB,KAAK,kBAAoB,KAC9C,SAAS,cAAc,KACvB,KAAK,iBACT,EAAiB,SAAW,KAAK,mBACjC,EAAiB,KAAO,EACxB,CAAA,EAAM,GACF,WAAM,MAAA,GAAiB,cAAc,GAAI,YAAW,cAPtD,CAAA,EAAA,UAMF,EAAA,wBAIF,MAAA,CAAA,EAAO,CAAC,mBAAoB,GAA6B,WA1E7C,EAAA,WAAa,eA6E/B,KAEA,GAAA,WAGE,WAAY,GACV,GAAI,GAAS,MAAQ,EAAM,OAAS,EAClC,KAAM,IAAI,OACN,2DACA,iBAAgB,IAEtB,KAAK,MAAQ,EAGT,SAAA,UAAA,KAAN,6FACQ,SAAW,KAAK,MAAM,GACtB,EAAc,KAAK,MAAM,MAAM,GAErC,CAAA,EAAO,GAAI,SAAwB,SAAC,EAAS,GAC3C,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GAEnB,GAAM,GAAY,KAAK,MAAO,EAAM,OAAe,QAC7C,EAAgB,EAAU,cAChC,GAAI,GAAiB,MACnB,EAAO,GAAI,OACP,4CAA4C,EAAS,OACzD,OAGF,AAAI,EAAY,SAAW,GACzB,EAAQ,CAAC,cAAa,IAGxB,GAAM,GAAkB,EAAU,gBAClC,GAAI,GAAmB,MACrB,EAAO,GAAI,OACP,6CAA6C,EAAS,OAC1D,OAGF,GAAI,GACJ,IACE,EACI,EAAK,4BAA4B,EAAiB,SAC/C,GACP,EAAO,GACP,OAGF,GAAM,GAAsC,GACtC,EAAkB,GAClB,EAAgC,GACtC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,EAAM,KAAK,GACX,EAAe,KAAK,QAEtB,EAAY,KAAI,MAAhB,EAAoB,EAAa,WAGnC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,GAAM,GAAmB,GAAI,YAC7B,EAAiB,OAAS,SAAC,GAEzB,GAAM,GAAc,EAAM,OAAe,OACnC,EAAQ,EAAM,QAAQ,GAC5B,EAAe,GAAS,EACxB,AAAI,EAAe,QAAQ,QAAU,IACnC,EAAQ,CACN,cAAa,EACb,YAAW,EACX,WAAY,GAAwB,GACpC,OAAQ,EAAU,OAClB,YAAa,EAAU,YACvB,YAAa,EAAU,YACvB,oBAAqB,EAAU,uBAIrC,EAAiB,QAAU,SAAA,GACvB,MAAA,GAAO,6CAA6C,EAAI,OAC5D,EAAiB,kBAAkB,EAAW,SAIpD,EAAW,QAAU,SAAA,GAAS,MAAA,GAC1B,2DACA,eAAc,EAAS,KAAI,qCAC3B,yCACJ,EAAW,WAAW,WAOlB,EAAA,UAAA,4BAAR,SACI,EAAiC,GAInC,OAHM,GAAsB,GACtB,EAAY,EAAM,IAAI,SAAA,GAAQ,MAAA,IAAS,EAAK,QAC5C,EAAqC,GACvB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAM,MAAM,QAAQ,SAAA,GAClB,GAAM,GAAe,GAAS,GAC9B,GAAI,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,sDACA,KAAI,EAAY,MAGtB,GADA,EAAU,KAAK,GACX,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,8BAA8B,EAAY,sBAE9C,EAAW,GAAQ,EAAM,EAAU,QAAQ,MAKjD,GAAI,EAAU,SAAW,EAAM,OAC7B,KAAM,IAAI,OACN,uDACA,KAAI,EAAU,OAAM,8CACpB,KAAI,EAAM,OAAM,OAEtB,MAAO,IAEX,KAEa,GAAmC,SAAC,GAC/C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,gBAyCH,GAAA,MAAA,KAAA,QAAA,GAAA,SACxB,GAAI,IAAiB,eA2CD,GAC3B,MAAO,IAAI,IAAa,GC5V1B,YA8BI,EAAmC,EACnC,EAAwB,GAC1B,EAAc,GACd,EAAgB,GAAwB,EACxC,EAAc,GAAsB,EACpC,EAAc,EAAe,GAC7B,GAAI,GAAkB,EAEhB,EAAkB,SAAC,GACvB,SAAQ,KAAK,SAAA,GACX,GAAM,GAAW,EACb,GAAE,EAAkB,EAAS,OAAU,GAAc,GAEzD,SAAW,GACJ,IAEF,GAGT,WAAuB,GACrB,EACI,GAAY,MAAQ,MAAM,QAAQ,IAAa,EAAS,OAAS,EACjE,WAAM,MAAA,wCAGZ,WAAuB,EAAuB,GAC5C,EACI,GAAiB,GAAK,GAAiB,EACvC,WAAM,MAAA,kDACF,sBAAqB,KAC7B,EACI,GAAe,GAAK,GAAe,EACnC,WAAM,MAAA,kDACF,oBAAmB,KAC3B,EACI,GAAe,EACf,WAAM,MAAA,uDACF,sBAAqB,EAAa,qBAClC,IAAG,KAGb,MAAO,SAAQ,IAAI,EAAS,IAAI,ICvElC,YAqCI,EAAqB,6HACvB,AAAI,IAAe,MACjB,GAAc,IAGV,EAAY,EAAY,WAAa,KAAO,KAAM,SAAS,MACf,EAAY,UAGxD,EAAW,EAAU,IACvB,SAAA,GACI,MAAA,GAAU,EAAU,EAAY,YAAa,CAAC,SAAU,OAE1D,EAAqB,EACrB,EAAmB,GAEP,EAAY,YAAc,KACxC,CAAA,EAAM,QAAQ,IAAI,IADJ,CAAA,EAAA,UACd,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAU,EAAY,WAAY,EAClC,WAFJ,EAAA,EAAA,8BAFE,GAAS,EAMT,EAAiB,EAAU,IAAI,SAAA,GAAY,MAAA,GAAS,gBAEpD,EAAsB,GACtB,EAAoB,EAEV,EAAY,YAAc,KACtC,CAAA,EAAM,QAAQ,IAAI,IADN,CAAA,EAAA,UACZ,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAgB,EAAY,WAAY,EACxC,WAFJ,EAAA,EAAA,wBAFE,SAAO,EAKb,CAAA,EAAO,oBAaL,EAAiC,EACjC,EACA,GAFiC,MAAA,KAAA,QAAA,GAAA,wEAS7B,SAAe,SAAC,GAClB,MAAA,IAAyB,EAAW,CAAC,YAAW,KAC9C,EAAc,GAAqB,GAEzC,CAAA,EAAO,EAAY,EAAU,EAAgB,oBA4B3C,GADJ,GAAA,GAAA,KAIE,MAAO,UACI,EAAiC,EACjC,GADiC,MAAA,KAAA,QAAA,GAAA,4GAqD1C,GAjDM,EAAyB,EAAS,IAAI,WAAM,MAAA,KAC5C,EAKF,GACE,EACF,GAAe,KAAO,EAAY,IAAI,WAAM,MAAA,KAAS,GACnD,EAAmC,GACzC,EAAS,QAAQ,SAAC,EAAqB,GACrC,GAAI,GAAc,EAClB,EAAoB,QAAQ,QAAQ,SAAA,GAClC,GAAM,GAAY,gBAAkB,GAChC,EAAa,aAAa,MAC1B,EAAa,MAEX,EAAe,GAAqB,GACtC,GAAmB,EAAa,OAE9B,EAA8B,WAClC,EAAuB,GAAc,GACrC,AAAI,EAAoB,IAAe,MACrC,GAAoB,GAAc,IAGpC,EAAoB,GAAY,KAAK,CACnC,cAAe,EACf,YAAW,EACX,UAAW,KAIf,AAAI,GAAe,KACjB,EAAY,QAAQ,SAAC,EAAY,GAC/B,AAAI,IAAe,EAAa,MAC9B,KACA,EAAa,GAAe,MAIhC,IAGF,EAAuB,KAAK,EAAa,MACzC,GAAe,MAIf,CAAC,EAAa,MAAM,SAAA,GAAS,MAAA,KACzB,QAAkB,EAAY,OAAO,SAAC,EAAG,GAAM,MAAA,CAAC,EAAa,KAC7D,GAAI,OACN,kDACG,GAAgB,KAAK,MAAK;GAC7B,yCACG,GAAuB,KAAK,MAAK,MAKpC,SACF,EAAuB,OAAO,SAAC,EAAa,EAAa,GACvD,MAAI,IACF,EAAY,KAAK,GAEZ,GACN,IAED,EAAsB,GAC5B,EAAoB,QAAQ,SAAA,GAC1B,EAAS,GAAG,MAAM,QAAQ,SAAA,GACxB,GAAM,GAAW,EACZ,CAAC,EAAe,SAAS,KAAa,GAAN,KAAY,EACjD,EAAU,KAAK,OAGH,CAAA,EAAM,EAAqB,WAArC,SAAU,EAAA,OAEV,EAAmC,GACrC,EAAoB,EACxB,EAAoB,QAAQ,SAAA,GAI1B,OAHM,GAAa,EAAS,GAAG,MAAM,OAEjC,EAAa,EACR,EAAI,EAAG,EAAI,EAAY,IAC9B,GAAc,EAAQ,EAAoB,GAAG,WAO/C,OAHM,GAAc,GAAI,aAAY,GAC9B,EAAkB,GAAI,YAAW,GACnC,EAAoB,EACf,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAS,GAAI,YAAW,EAAQ,EAAoB,IAC1D,EAAgB,IAAI,EAAQ,GAC5B,GAAqB,EAAO,WAG9B,GAAM,GAAiB,EAAoB,GAC3C,EAAe,QAAQ,SAAA,GACrB,GAAM,GAAa,EAAY,MAC3B,EAAa,YACb,EAAa,YAAc,EAAa,WACtC,EACF,GAAc,EAAY,CAAC,EAAa,gBAC5C,OAAW,KAAQ,GACjB,EAAiB,GAAQ,EAAgB,KAI7C,GAAqB,IAGvB,CAAA,EAAO,SCpPX,AA+BA,GAAM,IAAyB,2BACzB,GAAY,mBAClB,GAAA,WAcE,WAAY,EAAc,GAgCxB,GAvCO,KAAA,eAAiB,OAQxB,AAAI,GAAe,MACjB,GAAc,IAEhB,KAAK,iBAAmB,EAAY,iBACpC,KAAK,WAAa,EAAY,WAC9B,KAAK,mBAAqB,EAAY,mBAEtC,AAAI,EAAY,WAAa,KAC3B,GACI,MAAO,GAAY,WAAc,WACjC,WAAM,MAAA,gIAGV,KAAK,MAAQ,EAAY,WAEzB,KAAK,MAAQ,KAAM,SAAS,MAG9B,EACI,GAAQ,MAAQ,EAAK,OAAS,EAC9B,WAAM,MAAA,4DAGV,AAAI,MAAM,QAAQ,IAChB,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,+CACF,sBAAqB,EAAK,OAAM,QAE1C,KAAK,KAAO,EAER,EAAY,aAAe,MAC3B,EAAY,YAAY,MAAQ,KAClC,KAAM,IAAI,OACN,sEAEN,KAAK,YAAc,EAAY,aAAe,GAG1C,SAAA,UAAA,KAAN,SAAW,yGACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,2FAIA,SAAO,OAAO,OAAO,CAAC,OAAQ,KAAK,gBAAiB,KAAK,aAC/D,EAAK,KAAO,GAAI,UAEV,EAAyC,CAAC,CAC9C,MAAO,CAAC,uBACR,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,oBACpC,gBAAe,GAGjB,EAAK,KAAK,OACN,aACA,GAAI,MACA,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,KACX,cAEJ,AAAI,EAAe,YAAc,MAC/B,EAAK,KAAK,OACN,oBACA,GAAI,MAAK,CAAC,EAAe,YAAa,CAAC,KAAM,KAC7C,qBAGW,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,WAE7C,GAFM,EAAW,EAAA,OAEb,EAAS,GACX,MAAA,CAAA,EAAO,CACL,mBAAoB,GAA6B,GACjD,UAAW,CAAC,KAGd,KAAM,IAAI,OACN,gEACG,GAAS,OAAM,YAYpB,EAAA,UAAA,KAAN,uIAC6B,MAAA,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,KAAK,qBAE5D,GAFM,EAAqB,EAAA,OAEvB,CAAC,EAAmB,GACtB,KAAM,IAAI,OACN,cAAc,KAAK,KAAI,4BACpB,GAAmB,OAAM,uCAC5B,sFAIU,CAAA,EAAM,EAAmB,eAAvC,SAAc,EAAA,qCAEV,EAAU,+CAA+C,KAAK,KAAI,IAGtE,AAAI,KAAK,KAAK,SAAS,OACrB,GAAW,+UAOX,GAAW,uEAGP,GAAI,OAAM,UAUlB,GARM,EAAgB,EAAY,cAC5B,EAAkB,EAAY,gBAC9B,EAAc,EAAY,YAC1B,EAAc,EAAY,YAC1B,EAAS,EAAY,OACrB,EAAsB,EAAY,oBAGpC,GAAiB,MAAQ,GAAmB,KAC9C,KAAM,IAAI,OACN,2BAA2B,KAAK,KAAI,mEAMtC,IAAmB,KACL,CAAA,EAAM,KAAK,YAAY,IADrC,CAAA,EAAA,UACI,EAAU,EAAA,OACf,EAAA,EAAA,GAAa,EAAA,EAAA,oBAGV,SAA4B,CAChC,cAAa,EACb,YAAW,EACX,WAAU,EACV,oBAAmB,EACnB,YAAW,EACX,YAAW,EACX,OAAM,GAGF,EAAc,EAAY,iBAChC,AAAI,GACF,GAAU,iBAAmB,GAG/B,CAAA,EAAO,SAGK,EAAA,UAAA,YAAd,SAA0B,2IAOxB,IALM,EAAa,MAAM,QAAQ,KAAK,MAAQ,KAAK,KAAK,GAAK,KAAK,KAC5D,EAAmB,GAAS,GAA3B,EAAM,EAAA,GAAE,EAAM,EAAA,GACf,EAAa,KAAK,kBAAoB,EAEtC,EAAc,GACpB,EAAA,EAAoB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAT,EAAK,EAAA,GACd,EAAY,KAAI,MAAhB,EAAoB,EAAM,SAK5B,IAFM,EAAsB,GACtB,EAAsC,GAC5C,EAAA,EAA2B,EAAA,EAAA,EAAA,EAAA,OAAA,IACzB,IADS,EAAY,EAAA,GACrB,EAAA,EAAmB,EAAA,EAAa,MAAb,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,AAAI,KAAK,oBAAsB,KAC7B,EAAY,KAAK,KAAK,mBAAmB,IAEzC,EAAU,KAAK,EAAa,EAAO,SAKrC,MAAK,sBACP,GAAA,EAAU,MAAI,SAAd,GAAkB,CAAA,EAAM,QAAQ,IAAI,KADlC,CAAA,EAAA,UACF,EAAA,MAAA,EAAA,EAAA,OAAA,CAAkB,EAAA,2BAGJ,MAAA,CAAA,EAAM,GAAyB,EAAW,CACxD,YAAa,KAAK,YAClB,UAAW,KAAK,MAChB,WAAY,KAAK,qBAHb,SAAU,EAAA,OAKhB,CAAA,EAAO,CAAC,EAAa,GAAwB,WA5M/B,EAAA,iBAAmB,eA8MrC,iBAayB,GACvB,GAAM,GAAY,EAAI,YAAY,KAC5B,EAAkB,EAAI,YAAY,KAClC,EAAS,EAAI,UAAU,EAAG,GAC1B,EACF,EAAkB,EAAY,EAAI,UAAU,GAAmB,GACnE,MAAO,CAAC,EAAS,IAAK,eAGK,GAC3B,MAAO,GAAI,MAAM,GAAY,mBAAqB,KAG7C,GAAM,IACT,SAAC,EAAa,GACZ,GAAI,MAAO,QAAU,aAChB,IAAe,MAAQ,EAAY,WAAa,MAInD,MAAO,MAEP,GAAI,GAAS,GAMb,MALA,AAAI,OAAM,QAAQ,GAChB,EAAS,EAAI,MAAM,SAAA,GAAW,MAAA,IAAa,KAE3C,EAAS,GAAa,GAEpB,EACK,GAAK,EAAK,GAGd,MAEb,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAwEf,EAAc,GACjC,MAAO,IAAI,IAAY,EAAM,eAS3B,EAAc,GAChB,MAAO,IAAK,EAAM,GC3XpB,AAuBA,GAAA,IAAA,WACE,WAA6B,GAAA,KAAA,eAAA,EAEvB,SAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,qBAEhB,KAEA,GAAA,WACE,WACqB,GAAA,KAAA,YAAA,EAGf,SAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,YAAY,SAE5B,iBAwBI,EAAmC,EACnC,EAA0B,GAC5B,GAAI,UAAU,SAAW,GACvB,GAAM,GACD,EAAkC,eAAiB,MACnD,EAAkC,aAAe,KACtD,MAAI,GACK,GAAI,IAAkB,GAI7B,SAAQ,KACJ,yNAIG,GAAI,IAAkB,CAAC,cAAe,SAK/C,gBAAQ,KACJ,yNAIG,GAAI,IAAkB,CAC3B,cAAe,EACf,YAAW,EACX,WAAU,EACV,eAAc,gBAqBhB,GAEF,MAAO,IAAI,IAAiB,GCpH9B,oYCAA,AAuDA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MAExC,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,MAAK,GAC5B,EACW,SAAC,EAAwB,GACxC,SAAQ,GAA4B,EAAO,EAAG,MAC9C,EACI,EAAG,OAAS,GAAmB,GAC/B,WAAM,MAAA,mEACV,EAAK,CAAC,IACC,EAAQ,QAAQ,EAAI,IAE7B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAEO,GAAU,EAAG,CAAC,SAAQ,KC1EnC,AA6CA,YACI,EAAiB,EAAiB,EAClC,SADkC,AAAA,IAAA,QAAA,GAAA,IAClC,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAK,EAAgB,EAAG,IAAK,UACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WAAM,MAAA,kEACF,cAAa,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE3C,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,sCAAsC,EAAU,UAC/C,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,kCAAkC,EAAW,UAC5C,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACxD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAExD,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,IAEJ,EAAQ,YACX,EAAiB,EAAiB,EAAY,IAG9C,EAA4B,CAAC,EAAG,EAAK,EAAG,GAExC,EAA0B,CAAC,WAAU,EAAE,WAAU,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAO,GAAQ,EAAK,MAGT,IAAS,EAAG,CAAC,QAAO,KC9GjC,AA+CA,YACI,EAA4B,EAAe,EAC3C,GACF,GAF6C,AAAA,IAAA,QAAA,GAAA,GAC3C,AAAA,IAAA,QAAA,GAAA,GACE,EAAQ,EACV,KAAM,IAAI,OAAM,iDAAiD,GAEnE,GAAM,GAAW,EAAgB,EAAS,UAAW,SAAU,SACzD,EAAe,EAAS,MAAK,OAAA,CAAE,IAE/B,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,IACC,EACH,EAAQ,OACJ,EAAQ,EAAU,CAAC,EAAS,OAAQ,EAAO,EAAS,GACxD,IAGA,EAAuB,CAAC,QAAS,GACjC,EAAqB,CAAC,MAAK,EAAE,QAAO,EAAE,SAAQ,GAEpD,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAS,EAAG,CAAC,QAAO,KCxEjC,AA+CA,YAAsC,EAAiB,GACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAgBnC,GAdA,AAAI,GAAQ,MACV,GAAO,EAAG,MAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAG,WAEnC,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OACzC,EAAK,QAAQ,SAAA,GACX,EACI,GAAQ,GAAK,EAAO,EAAG,KACvB,WAAM,MAAA,+CAA+C,GAAG,KAAO,GAC3D,aAAY,OAGlB,EAAG,MAAQ,EACb,MAAO,GAAG,QAGZ,GAAM,GAA0B,CAAC,EAAG,GAC9B,EAAwB,CAAC,KAAI,GAEnC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IAAO,EACxC,KAAqB,GAAW,MAGzB,IAAY,EAAG,CAAC,WAAU,KC5EvC,YA0DI,EAA6B,EAC7B,GACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,mBAC5C,EACF,EAAgB,EAAa,cAAe,mBAEhD,EACI,GAAc,MAAQ,EAAa,GAAK,OAAO,UAAU,GACzD,WAAM,MAAA,uDACF,YAAW,KACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,gDAAgD,EAAQ,OAClE,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,6CACF,YAAW,EAAa,QAChC,EACI,EAAQ,MAAM,KAAO,EAAa,MAAM,GACxC,WAAM,MAAA,uCACC,GAAQ,MAAM,GAAE,QAAQ,EAAa,MAAM,GAAE,MAChD,oEACR,EACI,EAAa,GAAK,OAAO,UAAU,GACnC,WAAM,MAAA,4DACF,IAAG,KAIX,GAAM,GAAe,GAAO,GAAK,EAAS,SAAU,GAC9C,EACF,GAAO,GAAK,EAAc,SAAU,GAClC,EAA0B,GAAU,GAC1C,MAAO,IAAK,GAAO,EAAe,GAAoB,SAGjD,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KC9FnD,2CCAA,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,mDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,oEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,AA6BA,GAAI,IAyBJ,YACI,EAEA,GAEF,GAFE,AAAA,IAAA,QAAA,GAAA,GAEE,EAAc,EAChB,KAAM,IAAI,OACN,kEAEN,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,4DAElB,GAAI,GAAc,GACd,EAAc,GACd,EAAU,GACV,EAAU,GACV,EAAe,GACnB,GAAK,EAAqB,eAAgB,YACxC,EAAc,WAEZ,MAAQ,YAAe,aAAe,YAAkB,WAC1D,EAAc,WAEZ,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAER,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAEA,EAAe,YAAc,KACvC,EAAe,OAEf,MAAM,IAAI,OACN,6OAIA,YAAY,EAAc,YAAY,OAE5C,GAAI,GACF,GAAM,GAAgC,EACtC,GAAI,GACC,EAA4B,WACzB,EACN,KAAM,IAAI,OACN,yGAMR,GAAM,GAAS,GAAU,GAAY,EAAO,aAC5C,GAAI,GAAU,MACZ,GAAM,GAA2B,CAAC,OAAM,GAClC,EAAyB,CAAC,YAAW,GAC3C,MAAO,GAAO,UACH,GAAY,EACZ,GAGP,GAAA,GAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAMV,EAEJ,AAAI,EACF,EAEK,EAAe,WAAW,MAAM,aAAa,EAAG,EAAG,EAAO,GAAQ,KAClE,AAAI,GAAe,EACxB,EAAQ,EAAiC,KACpC,AAAI,IAAW,IACpB,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAEpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA4B,EAAG,EAAG,EAAO,GAC7C,EAAO,GAAoB,aAAa,EAAG,EAAG,EAAO,GAAQ,MAE/D,GAAI,GACJ,GAAI,IAAgB,EAClB,EAAS,GAAI,YAAW,QAExB,GAAM,GAAY,EAAQ,EAC1B,EAAS,GAAI,YAAW,EAAY,GACpC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAU,EAAG,EAAU,EAAa,EAAE,EAC7C,EAAO,EAAI,EAAc,GAAW,EAAK,EAAI,EAAI,GAIvD,GAAM,GAAqC,CAAC,EAAQ,EAAO,GAC3D,MAAO,IAAS,EAAQ,EAAU,qBAuBhC,EACA,iIAQF,GAPI,EAAO,EAAgB,EAAK,MAAO,YACvC,AAAM,YAAe,IAEb,GAAoB,EAC1B,EAAO,GAAK,EAAmB,SAC/B,EAAkB,WAEhB,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,OACN,wDAAwD,EAAK,KAAI,KAKvE,GAHM,EAAkB,EAAK,MAAM,MAAM,EAAG,GAArC,EAAM,EAAA,GAAE,EAAK,EAAA,GACd,EAAQ,EAAK,OAAS,EAAI,EAAI,EAAK,MAAM,GAE3C,EAAQ,GAAK,IAAU,EACzB,KAAM,IAAI,OACN,wCACA,sBAAqB,IAG3B,GAAI,EAAK,QAAU,WAAa,EAAK,QAAU,QAC7C,KAAM,IAAI,OACN,kCAAkC,EAAK,MAAK,0CAIrC,MAAA,CAAA,EAAM,EAAK,eAIxB,IAJM,EAAO,EAAA,OACP,EAAa,EAAK,QAAU,UAAY,IAAM,EAC9C,EAAQ,GAAI,mBAAkB,EAAQ,EAAS,GAE5C,EAAI,EAAG,EAAI,EAAS,EAAO,EAAE,GAGpC,IAFM,EAAO,CAAC,EAAG,EAAG,EAAG,KAEd,EAAI,EAAG,EAAI,EAAO,KAGzB,GAFM,EAAQ,EAAK,EAAI,EAAQ,GAE3B,EAAK,QAAU,WACjB,GAAI,EAAQ,GAAK,EAAQ,EACvB,KAAM,IAAI,OACN,qDACA,kCAAiC,EAAK,cAEnC,EAAK,QAAU,SACpB,GAAQ,GAAK,EAAQ,KACvB,KAAM,IAAI,OACN,mDACA,oCAAmC,EAAK,MAIhD,AAAI,IAAU,EACZ,GAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,GAElB,EAAK,GAAK,EAAQ,EAIhB,EAAI,EAAI,EACd,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAGjC,MAAI,IAAU,MACZ,GAAO,MAAQ,EACf,EAAO,OAAS,EACV,EAAM,EAAO,WAAW,MACxB,EAAY,GAAI,WAAU,EAAO,EAAO,GAC9C,EAAI,aAAa,EAAW,EAAG,IAEjC,AAAI,IAAS,GACX,EAAK,UAEP,CAAA,EAAO,QAGF,GAAM,IAAa,EAAG,CAAC,YAAW,+DCrOrC,EAAgB,GAClB,GAAI,EAAO,KAAO,EAChB,KAAM,IAAI,OACN,0DACA,sBAAqB,EAAO,KAAI,MAEtC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,4DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OACN,sDACA,uBAAsB,EAAQ,MAAK,MAEzC,GAAI,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAAO,KAC3C,KAAM,IAAI,OACN,iEACG,GAAQ,MAAM,EAAQ,KAAO,GAAE,QAAQ,EAAO,OAGvD,GAAI,EAAO,OAAS,EAClB,KAAM,IAAI,OACN,qDACA,kBAAiB,EAAO,MAAK,MASnC,OANM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAIjD,EAAU,EACL,EAAI,EAAG,EAAI,EAAa,OAAS,EAAG,EAAE,EAC7C,GAAW,EAAa,GAG1B,GAAM,GAAa,EAAO,MAEpB,EAAc,EAAa,QACjC,EAAY,MAGZ,OADI,GAAY,EACP,EAAI,EAAW,EAAI,EAAO,KAAM,EAAE,EACzC,GAAa,EAAW,GACxB,EAAY,KAAK,EAAW,IAG9B,GAAM,GACE,GAAe,EAAO,OAAO,IAAI,SAAA,GAAU,MAAA,GAAS,IAAU,OAAA,CACjE,IAAG,MAAM,EAAG,GAEjB,MAAO,CAAC,EAAa,EAAS,EAAW,6DCtDvC,EAAiB,EAAiB,GACpC,GAAM,GAAY,EAAQ,KAAO,EAAK,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAClE,EAAY,EAAQ,KAAO,EAAK,EAAQ,KAAO,EAAI,EAEnD,EAAa,wDACf,yCAAwC,EAAQ,OAChD,qBAAoB,EAAQ,MAAK,YAAY,GAC7C,gBAAe,EAAQ,mBAAmB,EAAQ,KAEtD,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OAAM,EAAa,mBAAkB,EAAQ,OAEzD,GAAI,EAAM,OAAS,EAAY,GAAQ,KAAO,GAC5C,KAAM,IAAI,OACN,EACA,2BAA0B,GAAY,GAAQ,KAAO,MAE3D,GAAI,EAAQ,OAAS,EAAW,EAAM,OAAS,EAC7C,KAAM,IAAI,OACN,EAAa,oBAAmB,GAAW,EAAM,OAAS,KAEhE,OAAS,GAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,GAAI,EAAQ,MAAM,KAAO,EAAQ,MAAM,GACrC,KAAM,IAAI,OACN,EACA,mBAAkB,EAAC,MAAM,EAAQ,MAAM,GAAE,sBAAsB,EAAC,MAC5D,EAAQ,MAAM,GAAE,OAG5B,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAO,EAAU,EAAE,EAC7C,GAAI,EAAQ,MAAM,EAAI,KAAc,EAAM,EAAI,GAC5C,KAAM,IAAI,OACN,EACA,mBAAkB,GAAI,GAAQ,MAC1B,EAAQ,MAAM,EAAI,GAAS,cAAc,GAAI,GAAQ,MACrD,EAAM,EAAI,GAAS,kBAoB7B,EAAiB,EAAiB,GACpC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OAAM,0DACZ,EAAQ,OAEd,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OACN,6DAA6D,GAGnE,GAAI,EAAM,SAAW,GACnB,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAEd,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAIhB,GAAoB,EAAO,EAAS,eAalC,EAAqB,EACrB,GAWF,OATM,GAAc,EAAQ,MAAM,OAC5B,EAAa,EAAc,EAAK,EAAQ,MAAM,EAAc,GAAK,EAKjE,EAAU,EAAM,OAElB,EAAY,EACP,EAAI,EAAW,EAAI,EAAS,EAAE,EACrC,GAAa,EAAM,GAGrB,GAAM,GAAgB,EAAY,EAAK,EAAI,EACrC,EAAa,GAAc,EAAQ,OAAS,EAE5C,EAAc,GAAe,EAAM,MAAM,EAAG,IAAW,OAAA,CAAE,IACzD,EAAa,GAAc,GACjC,MAAO,CAAC,UAAS,EAAE,WAAU,EAAE,UAAS,EAAE,QAAO,EAAE,WAAU,sFCnJ/D,YAqBI,EAAmB,EAAiB,GACtC,GAAM,GAAY,EAAM,MAAM,OAC9B,EACI,IAAc,EAAM,OACpB,WAAM,MAAA,iBAAiB,EAAS,sBAAsB,EAAK,SACvD,iCAAgC,EAAS,QACjD,EACI,IAAc,EAAK,OACnB,WAAM,MAAA,iBAAiB,EAAS,qBAAqB,EAAI,SACrD,iCAAgC,EAAS,QAEjD,mBAAS,GACP,EACI,EAAM,GAAK,EAAK,IAAM,EAAM,MAAM,GAClC,WAAM,MAAA,iBAAiB,EAAS,YAAY,EAAC,YAAY,EAAC,KACtD,KAAI,GAAM,GAAK,EAAK,IAAE,gCAAgC,EAAC,MACjD,EAAM,MAAM,GAAE,QALrB,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,eAUgB,GAGzB,OAFM,GAAO,GACT,EAAO,EACJ,EAAO,GACZ,AAAI,EAAO,GACT,EAAK,KAAK,GAEZ,GAAQ,EACR,IAEF,MAAO,eAKL,EAAiB,EAAe,GAElC,OADM,GAAO,GACJ,EAAO,EAAG,EAAO,EAAM,OAAQ,IACtC,EAAK,GAAQ,KAAK,KAAM,GAAI,GAAQ,EAAM,IAAS,EAAQ,IAE7D,MAAO,eAML,EAAmB,EAAgC,EACnD,GAEF,OADM,GAAiB,EAAO,QACrB,EAAI,EAAW,OAAQ,EAAI,EAAW,OAAQ,IACrD,EAAW,KAAK,GAElB,OAAS,GAAI,EAAG,EAAI,EAAe,IACjC,AAAI,IAAM,EACR,EAAW,GAA0B,EAErC,GAAW,OACP,EAAwB,EACxB,GACJ,EAAW,OAGf,MAAO,GAGT,YACI,EAAgC,EAChC,GACF,MAAI,IAAkB,EACb,EAGF,EAAkB,GAAgB,GAG3C,YAAuB,EAAuB,GAE5C,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAe,IACjC,EAAW,KAAK,EAAyB,GAE3C,MAAO,eAKL,EAAsB,EAAwB,EAC9C,EAAiB,EAAe,EAAmB,EACnD,EACA,GACF,GAAM,GAAY,EAAW,OACzB,EAAkB,GAAI,OAAM,GAC5B,EAAgB,GAAI,OAAM,GAC1B,EAAoB,GAAI,OAAM,GAClC,GAAI,EAAa,QAAU,EAAsB,GAC/C,GAAM,GAAY,EAAa,GAIzB,EAAgB,EAAsB,EAC5C,EAAkB,GACd,EAAW,EAAW,EAAe,EAAO,GAChD,EAAgB,GACZ,EAAS,EAAW,EAAe,EAAK,GAC5C,EACI,GAAsB,EAAS,EAAW,EAAe,OAE7D,QAAS,GAAO,EAAG,EAAO,EAAW,IACnC,EAAgB,GAAQ,GACpB,EAAW,EAAO,EAAS,EAAY,EAAM,GACjD,EAAc,GACV,GAAY,EAAS,EAAK,EAAS,EAAY,EAAM,GACzD,EAAkB,GAAQ,GAAe,EAAS,EAAM,GAI5D,MAAO,CACL,MAAO,EACP,IAAK,EACL,QAAS,eAOT,EAAmB,EAAgC,EACnD,EAAyB,GAI3B,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAEnB,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAc,GAClC,AAAI,EAAY,GAAK,GACnB,GAAgB,GAGlB,EAAW,GAAQ,EAGvB,MAAO,eAML,EAAiB,EAAgC,EACjD,EAAuB,GAIzB,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAAO,sBAE1B,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAY,GAChC,AAAI,EAAU,GAAK,GACjB,GAAgB,OAAO,kBAEzB,EAAW,GAAQ,EAIvB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,KAErC,GAAM,GAAW,EAAW,GAC5B,AAAI,EAAW,GAAK,GAClB,GAAW,IAAM,GAEnB,EAAW,GAAK,GAAW,EAAG,EAAW,GAAI,EAAW,IAE1D,MAAO,eAIL,EAAmB,EAAc,GACnC,GAAI,GAAS,EAAQ,GACrB,MAAI,GAAgB,GAAK,GAAS,GAAU,OAC1C,GAAS,GAGJ,cAIL,EAAmB,EAAwB,EAC3C,EAAsB,EAAc,GAEtC,GAAI,GAAQ,EAAa,GACnB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAY,GAAK,GAAQ,EAAe,GAAK,GAAQ,GAAS,OAChE,CAAI,EAAS,EAIX,EAAQ,OAAO,iBAGf,EAAQ,OAAO,kBAKnB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAQ,GACV,IAAS,GAIX,EAAQ,GAAW,EAAG,EAAO,EAAW,GAEjC,cAIL,EAAiB,EAAuB,EACxC,EAAsB,EAAc,GAEtC,GAAI,GAAO,EAAY,GACjB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAW,GAAK,GAAS,EAAgB,GAAK,GAAS,GAAQ,OACjE,CAAI,EAAS,EAGX,EAAO,OAAO,iBAGd,EAAO,OAAO,kBAKlB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAO,GACT,IAAQ,GAMV,AAAI,EAAS,EAEX,EAAO,GAAW,EAAG,EAAM,GAG3B,EAAO,GAAW,GAAI,EAAM,EAAW,GAGlC,cAQL,EAAiB,EAAiB,GAGpC,OADI,GAAkB,EAAK,OAClB,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,GAAI,EAAK,GAAK,GACZ,EAAkB,EAClB,MAIJ,OAAS,GAAI,EAAkB,EAAG,EAAI,EAAK,OAAQ,IACjD,GAAI,EAAM,GAAK,GAAK,EAAK,KAAO,EAAM,GACpC,MAAO,GAGX,MAAO,eAGyB,EAAiB,GAEjD,OADI,GAAa,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EACrD,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAc,EAAM,GAAK,EAAQ,GAEnC,MAAO,eAIL,EAAe,EAAwB,GAEzC,GAAI,GACE,EAAQ,EAAE,MAAM,OACtB,AAAI,MAAO,IAAU,SACnB,EAAM,CAAI,GAAK,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,IACzC,AAAI,EAAM,OAAS,EACxB,EAAS,EAAM,OAAO,GAAI,OAAM,EAAQ,EAAM,QAAQ,KAAK,IAE3D,EAAS,EAAM,QAEjB,EAAO,QAAQ,SAAA,GACb,EACI,IAAM,GAAI,WAAM,MAAA,wDAEtB,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAAI,OAAM,GAAO,KAAK,IACzB,AAAI,MAAO,IAAS,SACzB,EAAK,CAAI,GAAI,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,KACvC,AAAI,EAAK,OAAS,EACvB,EAAQ,EAAK,OAAO,GAAI,OAAM,EAAQ,EAAK,QAAQ,KAAK,KAExD,EAAQ,EAEV,EAAQ,EAAM,IAAI,SAAC,EAAG,GACpB,MAAI,IAAK,EACA,EAEP,GACI,IAAM,GACN,WAAM,MAAA,qDACC,GAAC,kCAAkC,EAAC,OACxC,EAAE,MAAM,GAAK,EAAO,MAGxB,CAAC,EAAQ,wSCjWlB,AA8DA,GAAA,IAAA,WAAA,cAYE,SAAA,UAAA,aAAA,WACE,MAAQ,MAAK,YACR,WAiBA,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,IAEnB,KASA,GAAA,WAOE,aACE,KAAK,aAAe,GAMf,SAAA,OAAP,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UAMnB,EAAA,SAAP,SAAwC,GACtC,EAAiB,SAAS,aAAa,EAAI,WACvC,CAAC,EAAK,EAAI,aAElB,iBA2BI,GACF,EACI,EAAI,WAAa,KACjB,WAAM,MAAA,gFAEV,EACI,MAAO,GAAI,WAAc,SACzB,WAAM,MAAA,sDACF,MAAO,GAAI,YACnB,EACI,EAAI,UAAU,OAAS,EACvB,WAAM,MAAA,sFAGV,GAAiB,SAAS,gFC/K5B,AAsBA,GAAM,IAAuB,KAChB,GAAuB,eAGhC,EACA,EAAoD,GACtD,MAAI,IAAW,MACb,GAAU,MAEL,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,mBAInE,MAAO,GAAO,QAAQ,mBAAqB,GAAK,GACA,GAGlD,YACI,EAAoB,EACpB,GACF,GAAI,GAAiB,GAOrB,GANA,AAAI,IAAa,IAAW,GAAa,KACvC,GAAiB,IAEnB,AAAI,GAAa,IAAW,GAAa,IACvC,GAAiB,IAEf,GACF,GAAM,GAAQ,EAAO,YAAY,KAC3B,EAAQ,EAAS,YAAY,KAEnC,GAAI,IAAU,EACZ,KAAM,IAAI,OACN,yCAAyC,EAAK,KAC9C,cAAa,IAIrB,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,IACzC,GAAM,GAAc,GAAW,GACzB,EAAgB,GAAW,GACjC,GAAI,CAAC,GAAY,EAAa,GAC5B,KAAM,IAAI,OACN,iCACA,aAAY,EAAW,iBAAiB,EAAa,MAI7D,GAAM,GACF,GAAa,GAAU,EAAS,GAAQ,GACtC,EAAe,GAAa,GAC9B,EACA,GAAQ,GAEZ,GAAI,EAAW,SAAW,EAAa,OACrC,KAAM,IAAI,OACN,yCAAyC,EAAW,OAAM,OAC1D,cAAa,EAAa,OAAM;GAChC,cAAa,EAAU;GACvB,cAAa,EAAY,MAE/B,OAAS,GAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAI,EAAW,GACf,EAAI,EAAa,GAEvB,GAAI,CAAC,EAAU,EAAG,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAC,OAAO,EAAC,cAAc,EAAC,OAAO,EAAC;EACzD,cAAa,EAAU;GACvB,cAAa,EAAY,mBAUC,EAAuB,GACzD,IAAK,KAAK,WAAM,MAAA,GAAK,QAAQ,WAAM,MAAA,mBAGH,EAAoB,GACpD,GAAM,GAAM,MAAO,IAAa,UAAY,MAAO,IAAa,UACxD,MAAO,IAAa,UACxB,CAAC,GACD,EACJ,MAAI,IAAS,IAAW,GAAU,EAAoB,KAClD,GAAS,IAAa,GAAU,EAAsB,IAEjD,GAAsB,EAAQ,EAAK,SAAC,EAAG,GAAM,MAAA,IAAK,IAEpD,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,iBAGlC,EAAW,EAAW,GAIvD,GAHA,AAAI,GAAW,MACb,GAAU,MAER,CAAC,GAAS,EAAG,EAAG,GAClB,KAAM,IAAI,OAAM,8BAA8B,EAAC,kBAAkB,GAIrE,YAAkB,EAAW,EAAW,GACtC,MAAI,CAAC,SAAS,IAAM,CAAC,SAAS,GACrB,GAEL,QAAM,IAAM,MAAM,IAAM,KAAK,IAAI,EAAI,GAAK,eAO5C,EAA6B,EAAa,GAC5C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,GAAI,EAAO,GAAK,GAAO,EAAO,GAAK,EACjC,KAAM,IAAI,OACN,sBAAsB,EAAO,GAAE,SAAS,EAAG,WAAW,eAM5D,EAAqB,GAGvB,OAAO,GAAI,cAAa,IAAS,QAAQ,GAAI,cAAa,4MCzJ5D,GAGM,IAAU,QCHhB,cAiCE,KAAM,IAAI,OAAQ,kBAkBlB,KAAM,IAAI,QAAS,kBAKnB,KAAM,IAAI,+BAAgC,IAC1C,QAAQ,KAAK,sEAIiB,GAC9B,AAAI,KAAM,QAAQ,iCAChB,QAAQ,KACJ,EAAM,6FAYZ,EAAO,iCASP,MAAO,iBAyBP,MAAO,GAAO,qBAgCQ,GAEtB,MAAO,GAAO,QAAQ,eA4CpB,EAA6B,GAC/B,MAAO,GAAO,KAAK,EAAU,eAcP,GACtB,GAAM,GAAU,GAAsB,GACtC,EAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,wBAmCI,GACrC,MAAO,GAAO,KAAK,eA4BA,GACnB,MAAO,GAAO,KAAK,eAkBM,GACzB,MAAO,GAAO,WAAW,iBAWzB,MAAO,GAAO,sBAUd,MAAO,GAAO,wBAQc,GAC5B,EAAO,cAAc,eAOK,GAC1B,MAAO,GAAO,YAAY,eAQO,GAEjC,MAAO,GAAO,mBAAmB,eAmB/B,EAAc,EACd,GAAA,MAAA,KAAA,QAAA,GAAA,GACK,EAAO,gBAAgB,EAAM,EAAS,iBAY7C,MAAO,GAAO,oBASY,EAAsB,GAChD,KAAM,YAAY,EAAc,GC5XlC,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCnE3B,AAqDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCvErC,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OAGjC,GAFA,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAED,EAAG,QAAU,SAAW,EAAG,QAAU,QACvC,MAAO,IAAS,EAAI,GAGtB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,WAAW,EAAI,GACnC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAC/B,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KC3E3B,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAEA,GAAM,EAAG,CAAC,KAAI,KCpE3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GAEpC,MADA,GAAK,CAAC,IACF,EAAG,QAAU,YACR,EAAQ,WAAW,GAGrB,EAAQ,IAAI,IAClB,EAAgC,KAAiB,OAGzC,IAAM,EAAG,CAAC,KAAI,KCrD3B,AAoCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAuCA,YAAiC,GAC/B,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,+DACV,EACI,EAAQ,QAAU,EAClB,WAAM,MAAA,uDACF,IAAG,EAAQ,UAEnB,GAAM,GACF,EAAQ,IAAI,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAG,UAAU,EAAK,UAEtD,EAAc,EAAS,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAI,EAAE,QAAU,EAAY,MAC1B,KAAM,IAAI,OACN,8DAIR,EAAS,QAAQ,SAAA,GACf,GAAI,CAAC,GAAiB,EAAE,MAAO,EAAY,OACzC,KAAM,IAAI,OACN,8DAIR,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,GACE,GAGH,EAAqB,EAE3B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC/E7B,YAuBqC,EAAgB,GACnD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,GAAI,EAAK,EAAK,OAAS,EAAI,KAAO,EAAO,EAAI,EAC3C,MAAO,GAGX,MAAO,eAIL,EAAqB,EAAqB,GAK1C,OAJI,GAAO,EAAU,OAAS,EAAU,OACpC,EAAM,GACR,EAAS,EACT,EAAY,EACL,EAAM,EAAG,EAAM,EAAM,IAC9B,AAAI,EAAK,QAAQ,KAAS,GACxB,EAAI,KAAK,EAAU,MAEnB,EAAI,KAAK,EAAU,MAGvB,MAAO,eAIL,EAAkB,GAGpB,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,EAAK,QAAQ,KAAS,IACxB,EAAS,KAAK,EAAO,IAGzB,GAAM,GAAc,EAAK,IAAI,SAAA,GAAO,MAAA,GAAO,KAC3C,MAAO,CAAC,EAAU,eAIhB,EAAiB,GACnB,GAAM,GAAiB,EAAK,IAAI,SAAA,GAAK,MAAA,KACrC,MAAO,IAAiB,EAAO,EAAgB,eAI7C,EAAa,EAAgB,GAC/B,EACI,GAAqB,EAAM,GAC3B,WAAM,MAAG,GAAG,2CACR,aAAY,EAAI,aAAa,EAAI,yBAQR,EAAgB,GAEjD,GAAI,GAAqB,EAAM,GAC7B,MAAO,MAGT,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,AAAI,EAAK,QAAQ,KAAO,IACtB,EAAO,KAAK,GAGhB,SAAK,QAAQ,SAAA,GAAQ,MAAA,GAAO,KAAK,KAC1B,cAI8B,GACrC,MAAO,GAAK,IAAI,SAAC,EAAM,GAAM,MAAA,CAAC,EAAG,KAC5B,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAE,KACxB,IAAI,SAAA,GAAK,MAAA,GAAE,iBAGe,EAAiB,GAEhD,OADM,GAAgB,GACb,EAAI,EAAO,EAAS,EAAI,EAAM,EAAE,EACvC,EAAI,KAAK,GAEX,MAAO,GC3GT,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KCxF3B,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAIA,IAAM,EAAG,CAAC,KAAI,KCzF3B,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC7EjC,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,AAAI,GAAQ,MACV,GAAO,GAET,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KChFjC,AAqCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC/C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA2CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAEH,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAQ,EAAG,CAAC,OAAM,KC7D/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,YAoGI,EACA,EAAuC,EACvC,EAA4B,EAC5B,GAD4B,AAAA,IAAA,QAAA,GAAA,QAM9B,GAAM,GAAgB,EAAW,GAC3B,EACE,EAAW,OAAA,CAAE,IACf,EAAc,GAAwB,GAE5C,MAAO,IACH,EAAY,EAAc,EAAS,EAAW,EAC9C,KAAyB,KAAsB,eAIjD,EACA,EAAqC,EACrC,EAAoC,EACpC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,gBACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GAEjB,EACJ,GAAI,IAAe,eACjB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACrD,IAAe,gBACxB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAE9D,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,EAAc,GAC7D,eAOF,EACA,EACA,EACA,EAA4C,EAC5C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,SACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAE9B,EACA,EACJ,GAAI,IAAe,QACjB,EAAc,eACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACxD,IAAe,QACxB,EAAc,gBACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAEjE,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,GAAO,EACtD,eAQF,EACA,EACA,EAAkC,EAClC,EACA,EAAuC,EACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACvC,AAAA,IAAA,QAAA,GAAA,gBACE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GACnC,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WACtB,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAU,EAAA,EAAA,OAElC,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAC9B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GACf,EAAA,GAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAEjB,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,SAIrB,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAW,GAC1C,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAW,EAAU,IAGvC,CACL,UAAS,EACT,WAAU,EACV,SAAQ,EACR,QAAO,EACP,WAAU,EACV,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,aAAY,EACZ,YAAW,EACX,aAAY,EACZ,YAAW,EACX,sBAAqB,EACrB,qBAAoB,EACpB,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,eA2CX,EACA,EACA,EACA,EAA4C,EAC5C,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,gBAEE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GAE5C,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WAC/B,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,OAE3C,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAE3C,EAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAC5B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAe,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGhC,EACF,GAAuB,EAAa,GAClC,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,SAAU,EAAA,EAAA,UAAW,EAAA,EAAA,SAK/B,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAU,EAAW,GACpD,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAU,EAAW,EAAU,IAGjD,CACL,UAAS,EACT,WAAU,EACV,QAAO,EACP,SAAQ,EACR,QAAO,EACP,WAAU,EACV,SAAQ,EACR,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,YAAW,EACX,aAAY,EACZ,YAAW,EACX,YAAW,EACX,aAAY,EACZ,YAAW,EACX,qBAAoB,EACpB,sBAAqB,EACrB,qBAAoB,EACpB,cAAa,EACb,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,GAIf,YACI,EAA2B,EAAmB,EAC9C,EAAkB,GACpB,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAY,GAGtB,YACI,EAA2C,EAC3C,EAAqB,EAAgB,EACrC,GACF,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAa,EAAQ,GACrB,EAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAe,GAChB,GAAa,EAAY,EAAI,GAAW,EAAS,EAAG,GACzD,EACI,GAAW,GACX,WAAM,MAAA,2BAA2B,EAAY,uEAGjD,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAc,EAAY,EAAY,eAI5C,EACA,EAAmB,EAAgB,GAAA,AAAA,IAAA,QAAA,GAAA,GACrC,GAAM,GAAqB,GAAuB,EAAW,GAC7D,MAAO,MAAK,MACP,GAAW,GAAM,GAAS,GAAK,EAAS,GAAsB,GAGrE,YAAyB,GACvB,MAAI,OAAO,IAAU,SACZ,CAAC,EAAO,EAAO,GAEpB,EAAM,SAAW,EACZ,CAAC,EAAM,GAAI,EAAM,GAAI,GAEvB,EAGT,YAA0B,GAExB,MAAO,OAAO,IAAU,SAAW,CAAC,EAAO,EAAO,GAAS,EAc7D,YAAgC,EAAoB,GAClD,MAAI,IAAY,EACP,EAGF,EAAc,GAAa,GAAM,GAAW,GAGrD,YACI,EAA4C,EAC5C,EAAiB,EAAsB,EACvC,EAAsB,EACtB,EACA,GAEF,GAAI,GACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CAAC,IAAK,EAAK,OAAQ,EAAK,KAAM,EAAK,MAAO,EAAK,KAAM,GAC/D,GAAM,GAAW,GACb,CAAC,EAAU,GAAU,EAAc,EAAc,EAAK,GAC1D,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GACF,KAAK,IAAI,EAAI,GAAY,GAAK,EAAe,EAAe,GAC1D,EACF,KAAK,IAAI,EAAI,GAAW,GAAK,EAAc,EAAc,GACvD,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAC9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,gBAClC,IAAQ,QACjB,EAAU,CAAC,IAAK,EAAG,OAAQ,EAAG,KAAM,EAAG,MAAO,EAAG,KAAM,SACvD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,WAC1C,MAAO,IAAQ,UACxB,GAAM,GAAM,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GACzD,EAAS,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC5D,EAAO,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC1D,EAAQ,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC3D,EAAW,IAAQ,GAAK,IAAW,GAAK,IAAS,GAAK,IAAU,EAClE,QACA,WACJ,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,GAC3C,EAAY,GACP,GAAW,EAAe,EAAM,GAAU,EAAe,EAC1D,GACJ,EAAW,GACN,GAAU,EAAc,EAAO,GAAS,EAAc,EAAG,OAE9D,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,UAAS,EAAE,SAAQ,GAGtC,YACI,EAA4B,EAAiB,EAC7C,EAAiB,EAAqB,EACtC,EAAqB,EAAqB,EAC1C,EAAqB,GAMvB,GAAI,GACA,EACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,GAER,GAAM,GAAW,GACb,CAAC,EAAS,EAAU,EAAS,GAAI,EAAa,EAAG,EAAa,EAC9D,GACJ,EAAW,EAAS,GACpB,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAW,KAAK,KAAK,EAAU,GAC/B,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EACD,GAAY,GAAK,EAAe,EAAe,EAC9C,EAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EAAQ,KAAK,MAAM,EAAgB,GACnC,EAAO,EAAgB,EACvB,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAE9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,MAAK,EAAE,KAAI,EAAE,KAAM,gBAC/C,IAAQ,QACjB,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,SAER,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,GACnD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,OAEnD,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,SAAQ,EAAE,UAAS,EAAE,SAAQ,GAQhD,YACI,EAAe,GACjB,GAAI,CAAC,EACH,MAAO,GAET,OAAQ,OACD,QAEH,MAAO,MAAK,MAAM,OACf,OAEH,MAAO,MAAK,KAAK,OACd,QACH,MAAO,MAAK,MAAM,WAElB,KAAM,IAAI,OAAM,wBAAwB,gBAIZ,GAC1B,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,MAAO,KAAS,GAAK,IAAS,GAAK,IAAS,cAI1C,EAA0B,GAC5B,MAAO,IAAkB,IAAY,GAAkB,eAUjB,GAEtC,GAAI,IAAe,OACjB,MAAO,eACF,GAAI,IAAe,OACxB,MAAO,gBAEP,KAAM,IAAI,OAAM,sBAAsB,GC5mB1C,AAoDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,WACxC,EAAY,EAElB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,mDAAmD,EAAI,KAAI,MAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAIJ,MAFA,GAAK,CAAC,IAEF,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UACvC,EAAI,QAGN,EAAQ,QAAQ,EAAK,IAGxB,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAElE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAG,OAEf,EACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KClHnC,AA6EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,WAE5C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAGrB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GAEJ,SAAK,CAAC,IAEC,EAAQ,UAAU,EAAK,IAG1B,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAEjE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAI,OAEhB,EACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KCzJvC,YAmBuC,EAAoB,GACzD,GAAM,GAAO,EAAO,GAAG,OACvB,EAAO,QAAQ,SAAC,EAAO,GACrB,EACI,EAAM,SAAW,EACjB,WACI,MAAA,kBAAkB,EAAI,sBAAsB,EAAC,sBAC7C,6BAA4B,EAAI,SAG1C,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,kBAAkB,EAAI,iCAAiC,GAAO,GAAC,MAEzE,GAAM,GAAa,EAAO,GAC1B,EAAO,QAAQ,SAAC,EAAO,GACrB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EACK,IAAM,GAAU,EAAM,KAAO,EAAW,GACzC,WAAM,MAAA,kBAAkB,EAAI,uBAAuB,EAAC,MAAM,EAAK,KAC3D,0CAAyC,EAAU,MACnD,oCAAmC,EAAC,qBAKlB,EAAoB,GAElD,OADM,GAAc,EAAO,GAAG,QACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAY,IAAS,EAAO,GAAG,GAEjC,MAAO,GClDT,AAqEA,YAAmC,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/D,EAAO,EAAQ,QAAU,EAAG,WAAM,MAAA,uCAElC,GAAI,GAAW,GAAqB,EAAS,UAAW,UAExD,AAAI,EAAS,GAAG,QAAU,aACxB,EAAS,QAAQ,SAAA,GACf,GAAI,EAAO,QAAU,YACnB,KAAM,IAAI,OAAM;uBACD,EAAO,MAAK,QAKjC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAS,GAAG,OAAO,GAChD,EAAW,GAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAC7D,GAAI,GAAc,KAAc,EAC9B,MAAO,IAAO,GAAI,GAIpB,GADA,EAAW,EAAS,OAAO,SAAA,GAAK,MAAA,GAAE,KAAO,IACrC,EAAS,SAAW,EACtB,MAAO,GAAS,GAGlB,GAAM,GAAS,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QACnC,GAAuB,EAAQ,GAE/B,GAAM,GAAM,EAAQ,OAAO,EAAU,GACrC,SAAK,GACE,GAGH,EAAuB,EACvB,EAAoB,CAAC,KAAI,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC/GjC,AAsCA,YAAoC,GAClC,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAAwB,CAAC,EAAG,GAElC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,QAAQ,GAC5B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAU,EAAG,CAAC,SAAQ,KCjDnC,AA8DA,YACI,EAAiB,EAAwB,GAC3C,GAAM,GAAK,EAAgB,EAAG,IAAK,SAEnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,kCAGlB,GAAM,GAA+B,SAAC,EAAS,GACvC,GAAA,GAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,UAA6B,EAAI,EAAQ,GACzC,EAAK,CAAC,IACC,EAAQ,MAAM,EAAI,EAAQ,IAG7B,EAAsB,CAAC,EAAG,GAC1B,EAAoB,CAAC,MAAK,EAAE,KAAI,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCrF/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAI,EAAQ,KAAK,GACvB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA8CA,YACI,EAA+B,EAC/B,EAA+B,EAC/B,EAAwB,GAC1B,GAAM,GACF,EAAgB,EAAY,aAAc,iBACxC,EACF,EAAgB,EAAY,aAAc,iBACxC,EAAY,EAAgB,EAAU,WAAY,iBAClD,EAAQ,EAAgB,EAAM,OAAQ,iBACtC,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAE7B,EAAW,GAAO,CAAC,EAAO,GAAK,GAC/B,EAAW,GAAO,EAAU,GAC5B,EAAgB,GAAI,EAAU,GAG9B,EAAY,EAAI,MAAM,GACtB,EAAY,EAAI,MAAM,GAAK,EAC3B,EAA8B,CAAC,EAAW,GAC1C,EAAI,GAAM,EAAK,CAAC,EAAG,GAAI,GACvB,EAAI,GAAM,EAAK,CAAC,EAAG,GAAY,GAC/B,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GACnC,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GAEnC,EACF,GAAI,EAAI,GAAQ,GAAI,GAAK,IACrB,EAAI,EAAI,GAAQ,GAAI,EAAa,MACnC,EAAiB,EAAI,GAAK,GAAO,GAAQ,IAC/C,MAAO,CAAC,EAAM,MAGH,IAAgB,EAAG,CAAC,eAAc,KC/E/C,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAE7C,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,iBAAiB,EAAG,KAAI,2CAC1B,EAAW,SAEnB,EACI,EAAM,SAAW,EAAW,OAC5B,WAAM,MAAA,mBACF,EAAM,OAAM,8CACZ,EAAW,SAEnB,EACI,EAAG,MAAM,GAAK,IAAS,EACvB,WAAM,MAAA,yBACI,EAAG,MAAM,GAAE,2CACjB,+BAA8B,EAAW,KAAK,OAAM,QAAQ,KAEpE,GAAM,GAA0B,SAAA,GAC9B,MAAO,GAAQ,eAAe,EAAI,EAAY,IAG1C,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,MAAK,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,iBC1FX,GACpC,GAAI,GACJ,MAAI,GAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAG,EAAE,OACxB,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACvC,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAErD,EAAM,EAGD,EChCT,AAwDA,YACI,EAAyB,EACzB,EACA,EACA,EACA,GACF,AAAI,GAAmB,MACrB,GAAkB,MAEpB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,AAAI,GAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAG9C,EACI,EAAM,OAAS,EAAU,KACzB,WAAM,MAAA,iFAEV,EACI,GAAW,MAAQ,EAAM,OAAS,EAAQ,KAC1C,WAAM,MAAA,+EAEV,EACI,GAAU,MAAQ,EAAM,OAAS,EAAO,KACxC,WAAM,MAAA,8EAGV,GAAM,GAAgB,GAAM,GAEtB,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,EAAO,EAAW,IAEtB,EAAQ,UACX,EAAK,GAAS,GAAQ,GAAS,GAAY,GAAS,GACpD,GAAS,GAAS,IAGlB,EAA+B,CACnC,EAAG,EACH,MAAO,EACP,OAAQ,EACR,KAAM,EACN,SAAU,GAGN,EAA6B,CAAC,gBAAe,GAE7C,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAO,GAAQ,EAAK,EAAG,OAGzB,YAAkB,GAChB,MAAI,IAAK,KACA,KAEL,EAAE,OAAS,EAEN,EAAQ,EAAG,CAAC,EAAE,OACZ,EAAE,OAAS,EACb,EACE,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACpC,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAEjD,KAGI,IAAY,EAAG,CAAC,WAAU,KCrGvC,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAE/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC9E3C,AA4CA,YACI,EAAsB,GACxB,GAAI,GAAQ,EAAgB,EAAG,cAAe,KACxC,EAAS,EAAM,MAErB,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,CAAE,GAAI,IAAM,EAAI,IAAM,IACxC,KAAM,IAAI,OAAM,2CAA2C,EAAK,MAGlE,GAAI,EAAM,OAAS,EAAM,KACvB,KAAM,IAAI,OAAM,+BAA+B,EAAM,OAAM,iBACvD,EAAM,KAAI,KAGhB,GAAI,EAAM,OAAS,EAAM,MAEvB,OADM,GAAW,EAAM,MAAM,QACtB,EAAS,OAAS,EAAM,QAC7B,EAAS,QAAQ,GAEnB,EAAQ,EAAQ,EAAO,GAKzB,OAFM,GAAa,EAAM,MACnB,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAM,OAAS,EAAG,GAAK,EAAG,IACrC,GAAI,EAAW,KAAO,EAAM,GAC1B,EAAK,GAAK,UACD,EAAM,MAAM,KAAO,EAC5B,KAAM,IAAI,OACN,mBAAmB,EAAM,6BAA6B,EAAK,MAGnE,GAAM,GAAO,EAAK,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,KAAI,OAAO,SAAA,GAAK,MAAA,IAAK,IAEjE,GAAI,EAAK,SAAW,EAClB,MAAO,IAAM,GAGf,GAAM,GAAU,SAAC,GAA2B,MAAA,GAAQ,KAAK,EAAO,IAE1D,EAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,MAAK,EAAE,WAAU,GAElD,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAa,MAGb,IAAc,EAAG,CAAC,aAAY,KC5F3C,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAyCA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,eACnC,EACK,GAAgB,EACjB,WAAM,MAAA,uBAAuB,EAAY,aACrC,+BAA8B,EAAY,QAElD,GAAM,GAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,aAAY,EAAE,aAAY,GAE3D,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAM,EAAQ,KAAK,EAAI,EAAc,GAC3C,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAc,EAAG,CAAC,aAAY,KC7B3C,YAAmB,GACjB,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCYrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCDrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KC5BrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCnCrC,AA8DA,YACI,EAAiB,EACjB,EACA,EACA,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GACjE,EACI,IAAY,EAAQ,MAAM,GAC1B,WAAM,MAAA,oCAAoC,EAAO,gBAC7C,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,GAAO,GACL,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GACxC,EACY,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAS,EAAG,CAAC,QAAO,KC3EjC,YACI,EAAiB,EAA6B,EAC9C,EACA,EAAiC,EACjC,GADA,AAAA,IAAA,QAAA,GAAA,OAAiC,AAAA,IAAA,QAAA,GAAA,GAEnC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,MAG9C,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAQ,GACjD,WAAM,MAAA,yDACF,eAAc,EAAM,kBAAkB,EAAQ,OACtD,EACI,IAAe,MACf,WAAM,MAAA,sCACF,EAAU,0CAElB,GAAM,GAAW,EACb,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,KAC7D,EAAU,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAA4B,CAAC,EAAG,GAChC,EAA8B,CAAC,EAAG,GAElC,EAAmB,OAEnB,EAAM,GACP,EAAuB,EAAuB,EAAS,EACxD,EAAkB,EAAW,GAEjC,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAGxC,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,QAGhD,IAAS,EAAG,CAAC,QAAO,KCpHjC,AAoDA,YACI,EAAmE,EACnE,EAAkB,EAClB,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KAC1D,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,KAG9C,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,GAAM,GAAU,IAAe,OAAS,EAAS,GAAK,EAAS,GACzD,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAY,EAEZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEE,EAAM,EAAQ,eAAe,EAAM,EAAQ,GAEjD,SAAK,CAAC,EAAM,IAEL,GAGH,EAAoC,CAAC,GAAI,EAAM,OAAM,GAErD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,EAAE,WAAY,GAEtD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAqB,GAEzB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxF3D,YACI,EAAiB,EACjB,EACA,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IACH,EAAa,EAAI,EAAS,EAAS,EAAK,OAAQ,MAGzC,IAAkB,EAAG,CAAC,iBAAgB,KCxDnD,AA8DA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAExE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAA+B,EAAS,GACxC,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,QACf,WAAM,MAAA,sCACF,EAAU,4CAElB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,GAC5C,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GAExC,EAAqB,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,GAEzD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,KAGI,IAAS,EAAG,CAAC,QAAO,KC3HjC,AA6CA,YACI,EAGA,EAAO,EAAkB,EACzB,GACF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACvE,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,EAAO,KAGzD,GAAM,GAAU,EAAS,GACnB,EAAW,EAAK,MAAM,GAC5B,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,GAEhD,MAAO,GAAQ,eAAe,EAAM,EAAQ,IAGxC,EAAoC,CAAC,GAAI,GAEzC,EAAkC,CAAC,IAAG,GAEtC,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxE3D,YACI,EAAiB,EACjB,EAGA,EAA0C,GAC5C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IAAoB,EAAa,EAAI,EAAS,EAAS,MAGnD,IAAkB,EAAG,CAAC,iBAAgB,KCtDnD,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AAqDA,YACI,EAAsB,EAAU,EAAmB,GAA7B,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,IAAmB,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAG,MAC9C,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,IAE5B,GAAM,GAAe,GAAiB,EAAG,EAAG,MAAM,GAC9C,EAAQ,EAAQ,OAAO,EAAW,EAAc,EAAW,GAG/D,GAFA,EAAK,CAAC,IAEF,GAAe,MACjB,GAAM,GAAqB,GAAuB,GAClD,EAAQ,GAAU,EAAO,GAE3B,MAAO,IAGP,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,EAAE,UAAS,EAAE,QAAO,GAEpD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KCnFjC,AAiEA,YACI,EAA0B,EAC1B,GAAA,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,gBAE7B,EAAe,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC/D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC9D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAEpE,EACI,EAAc,GAAa,EAC3B,WAAM,MAAA;MACN,EAAW,QAAQ,EAAS;MAC5B,EAAG,QAEP,EACI,EAAa,GAAa,EAC1B,WAAM,MAAA;MACN,EAAU,QAAQ,EAAS;UACvB,EAAG,QAEX,EACK,EAAc,GAAY,KAAe,EAC1C,WAAM,MAAA,8CACF,EAAY,EAAS,WACrB,EAAU,sCAAsC,EAAG,QAE3D,GAAM,GAAiC,SAAA,GACnC,MAAA,GAAQ,aAAa,EAAI,EAAW,IAElC,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAAC,UAAS,EAAE,WAAU,GAEvD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAc,MAGP,IAAe,EAAG,CAAC,cAAa,KCvG7C,AA2EA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,sDACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAElC,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAGlB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,iEACF,mBAAkB,EAAO,mBAAmB,EAAS,OAE7D,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IACE,EAAM,EAAQ,gBAAgB,EAAK,EAAS,GAClD,SAAK,CAAC,EAAK,IACJ,GAGH,EAAsC,CAAC,EAAG,EAAK,OAAQ,GACvD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KC/InD,AA+CA,YAAe,GACb,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAA,GACnC,GAAM,GAAO,EAAQ,EAAI,CAAC,EAAG,OACvB,EAAS,EAAQ,KAAK,GACtB,EAAe,EAAE,MAAK,OAAK,EAAE,OAEnC,MAAO,GAAQ,EAAQ,IAGnB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAO,EAAG,CAAC,MAAK,KChE7B,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,GADA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IACzC,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAU,EAAgB,EAAQ,SAAU,cAElD,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA,gEACC,GAAG,KAAI,OAClB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,4DACC,GAAQ,KAAI,OACvB,EACI,IAAe,OACf,WAAM,MAAA,0DACF,0BAAyB,KAEjC,GAAI,GAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACzD,EAAe,IAGjB,GAAM,GAA2B,CAAC,EAAG,EAAK,OAAQ,GAC5C,EAAyB,CAAC,QAAO,EAAE,IAAG,EAAE,UAAS,GAEjD,EAAM,EAAO,UACH,GAAY,EACZ,GAEhB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAa,EAAG,CAAC,YAAW,KCvGzC,YA2BI,EAAmB,GAGrB,OAFM,GAAS,EAAQ,OACjB,EAAiB,GACd,EAAI,EAAG,EAAI,EAAQ,KAC1B,GAAM,GAAM,EAAS,EAAI,EACnB,EAAI,EAAQ,IAAQ,EACpB,EAAI,EAAS,EAAS,OAAS,EAAI,IAAM,EAC/C,AAAI,EAAI,GAAK,IAAM,GACjB,EAAK,QAAQ,GAGjB,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAQ,EAAQ,EAAQ,OAAS,EAAI,GACrC,EAAU,EAAS,OAAS,EAAI,EAChC,EAAS,EAAS,GACxB,AAAI,IAAS,MAAS,IAAU,GAAK,EAAS,IAC5C,EAAO,QAAQ,GAGnB,MAAO,eAIL,EAAkB,GAIpB,OAHM,GAAmB,GACnB,EAAI,KAAK,IAAI,EAAO,OAAQ,EAAO,QAEhC,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GACnC,AAAI,GAAK,MACP,GAAI,GAEN,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GAInC,GAHA,AAAI,GAAK,MACP,GAAI,GAEF,IAAM,EACR,EAAO,QAAQ,WACN,IAAM,EACf,EAAO,QAAQ,WACN,IAAM,GACf,GAAM,GAAS,wDACR,GAAM,QAAQ,EAAM,KAC3B,KAAM,OAAM,OAEZ,GAAO,QAAQ,GAGnB,MAAO,GCrFT,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,MAAM,EAAI,IAE5D,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,OAG/C,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AAoDA,YACI,EAA8B,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAa,EAAgB,EAAW,YAAa,QAAS,QAI9D,EAAiB,GAA2B,EAAG,MAAO,EAAG,OACzD,EAAgB,GAAY,EAAI,GAChC,EAAgB,GAAY,EAAI,GACtC,AAAI,EAAW,OAAS,GAGtB,EACI,EAAW,MAAM,KAAO,EAAG,MAAM,GACjC,WAAM,MAAA,mEAGZ,AAAI,EAAW,OAAS,GAEtB,GACI,EAAW,MAAO,EAAc,MAAO,oBAG7C,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,OAAO,EAAY,EAAe,GACtD,SAAK,CAAC,IACC,GAEH,EAAyB,CAC7B,UAAW,EACX,EAAG,EACH,EAAG,GAEL,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,OAGA,IAAQ,EAAG,CAAC,OAAM,KC5F/B,AAuCA,YAAsC,GACpC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAA0B,CAAC,EAAG,GACpC,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,IAAK,EAClC,KAAiB,OAEjB,IAAY,EAAG,CAAC,WAAU,KC9CvC,AA0DA,YACI,EAAsB,SAEpB,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAAY,GAAI,EAAI,GACpB,EAAQ,GAAU,GAClB,EAAc,GAAM,EAAI,GAC9B,MAAO,IAAM,EAAa,EAAO,MAGtB,IAAW,EAAG,CAAC,UAAS,KCvErC,AA2CA,YAAc,EAAuB,GACnC,GAAM,GAAM,EAAgB,EAAI,KAAM,OAChC,EAAM,EAAgB,EAAI,KAAM,OAEtC,EACK,GAAI,OAAS,GAAK,EAAI,OAAS,IAAO,GAAI,OAAS,GAAK,EAAI,OAAS,GACtE,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GACjD,EAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GAOvD,GALA,EACI,IAAY,EACZ,WAAM,MAAA,gEACC,GAAO,QAAQ,EAAO,OAE7B,EAAI,OAAS,GAAK,EAAI,OAAS,GACjC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,YACZ,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,CAAC,EAAK,eAClB,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAK,GACzB,MAAO,GAAQ,EAAM,CAAC,EAAK,YAE3B,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAK,GACzB,MAAO,OAIE,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,GACtB,SAAK,CAAC,IACC,GAGH,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCtD3B,AAyCA,YAAgC,GAC9B,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,EACI,EAAG,QAAU,SAAW,EAAG,QAAU,UACrC,WAAM,MAAA,8CAEV,AAAI,EAAG,QAAU,SACf,GAAK,GAAK,EAAI,YAGhB,GAAM,GAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KC1D3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAyCA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GAC3D,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,aAAc,GAEjD,EAAY,GAAQ,EAAG,KAAM,WAAM,MAAA,uCACnC,GAAM,GAAW,EAAG,MAAM,QAC1B,MAAI,GAAO,GAET,GACI,CAAE,GAAG,KAAO,IAAM,EAClB,WAAM,MAAA,iCAAiC,CAAG,GAAG,KAAO,GAAE,KAAK,EAAG,KAAI,MACtE,EAAO,EAAG,KAAO,EAAO,GAE1B,EAAS,OAAO,EAAM,EAAG,GAClB,EAAQ,EAAI,MAGR,IAAa,EAAG,CAAC,YAAW,KC1DzC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAqDA,YAAiC,EAAiB,GAChD,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,OAAQ,GAC3C,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OAEzC,GAAM,GAA0B,SAAC,EAAS,GACxC,GAAM,GAAM,EAAQ,KAAK,EAAI,GAC7B,SAAK,CAAC,IACC,GAGH,EAAe,CAAC,GAChB,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,GAE9B,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,EAAkC,MAG3B,IAAO,EAAG,CAAC,MAAK,KC5E7B,AAwCA,YACI,EAAiB,EACjB,EAKA,GAAA,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAc,MAChB,GAAa,GAIf,OAFM,GAAO,GAAO,CAAC,EAAS,GAAa,GACrC,EAAI,GAAW,EAAa,EAAU,EACnC,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAK,IAAI,EAAG,EAAG,GAEjB,GAAM,GAAgB,EAAQ,EAAK,WAAY,CAAC,EAAS,IACzD,GAAI,GAAc,KAChB,MAAO,GAEP,GAAI,EAAW,SAAW,EACxB,MAAO,IAAK,GAAW,EAAK,GAAI,CAAC,EAAW,GAAI,EAAG,IAC9C,GAAI,EAAW,SAAW,EAE/B,MAAO,IACI,GAAW,GAAW,EAAK,GAAI,GAC/B,CAAC,EAAW,GAAI,EAAW,GAAI,EAAG,IACxC,GAAI,EAAW,SAAW,EAE/B,MAAO,IAAK,GAAW,GAAW,GAAW,EAAK,GAAI,GAAI,GAAI,CACrD,EAAW,GAAI,EAAW,GAAI,EAAW,GAAI,EAAG,IAGzD,KAAM,IAAI,OACN,2CAEA,8BAA8B,EAAmB,OAAM,UAKpD,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAqCA,YACI,EAAoB,EAAsB,GAC5C,GAAM,GAAmB,CAAC,MAAK,EAAE,MAAK,EAAE,MAAK,GAE7C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAO,EAAO,IAAQ,GAAI,KAAM,GACxD,GC3CN,AAqCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC7C/B,AAuBO,GAAM,IAAwB,eASI,GACvC,MAAI,IAAU,GACL,EAEF,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KCpCrD,YA8BI,EAAgB,GAClB,GAAI,GAAO,GACP,EASJ,IAPA,AAAI,GAAU,GACZ,GAAM,EACN,EAAO,IAEP,EAAM,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KAG7C,CAAC,GACN,AAAI,EAAM,GAAe,IAAQ,EAC/B,EAAO,GAEP,EAAM,GAAe,EAAQ,EAAM,GAGvC,MAAO,eAIL,EAAkB,EAAc,GAGlC,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,IAAQ,EACV,EAAS,KAAK,EAAO,IAErB,EAAS,KAAK,GAGlB,MAAO,eAUL,EAAW,EAAiB,GAM9B,OALM,GAAU,EAAE,MAAM,GAElB,EAAwB,GAC1B,EAAY,EACZ,EAAY,EACP,EAAI,EAAG,EAAI,EAAM,IACxB,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAM,IAChC,EAAY,KAAK,EAAQ,MAAM,IAGjC,OAAS,GAAI,EAAO,EAAG,EAAI,EAAE,KAAM,IACjC,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,MAAO,CAAC,UAAS,EAAE,UAAS,EAAE,QAAO,EAAE,YAAW,2GC5FpD,AAoDA,YACI,EAAiB,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAW,EAAgB,EAAS,UAAW,SAAU,SAEzD,EAAyB,CAAC,EAAG,EAAI,QAAS,GAC1C,EAAuB,CAAC,KAAI,GAE5B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAa,GAAe,EAAM,EAAG,OAAO,GAC5C,EAAY,GAAyB,EAAI,EAAU,GAEnD,EACF,EAAQ,OAAO,EAAI,EAAQ,EAAU,CAAC,EAAS,OAAQ,GAC3D,SAAK,CAAC,EAAI,IAEH,EAAQ,EAAK,EAAU,cAGhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC5EjC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,QAAQ,EAAI,IAE9D,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC3DnC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,aAAa,EAAI,GACrC,SAAK,CAAC,EAAI,IACH,GAGH,EAA6B,CAAC,EAAG,EAAI,EAAG,GAE9C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAe,EAAG,CAAC,cAAa,KC/D7C,AAsCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCnD7B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GAEnC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,SAAS,IAAK,EACnC,KAAiB,OAEV,IAAW,EAAG,CAAC,UAAS,KC/CrC,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC/C/B,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC9C/B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAEhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAGH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,YAsCI,EAAyC,GAC3C,GAAM,IAAa,IAAU,IAAU,UAAa,MAAM,QAAQ,KAC9D,IAAU,YACZ,KAAM,IAAI,OACN,kFAGN,GAAI,IAAU,UAAY,GAAa,IACnC,CAAE,aAAiB,aACrB,KAAM,IAAI,OACN,6EAGN,GAAM,GAAkB,GAClB,EAA0B,GAChC,MAAO,IAAW,EAAO,EAAO,EAAe,GCrDjD,AA2CA,YAAsC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aACnC,MAAO,IAAQ,EAAI,GAAO,GAAQ,GAAK,MAG5B,IAAY,EAAG,CAAC,WAAU,KChDvC,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,QAC7B,EAAK,EAAgB,EAAG,IAAK,QACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAE3D,EAAqB,CAAC,EAAG,EAAI,EAAG,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC1D7B,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,UAAU,EAAI,GAClC,SAAK,CAAC,EAAI,IACH,GAGH,EAA0B,CAAC,EAAG,EAAI,EAAG,GAE3C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAY,EAAG,CAAC,WAAU,KC/DvC,YAkCyB,EAAe,EAAc,GACpD,GAAI,GAAO,EACT,KAAM,IAAI,OAAM,4CAGlB,GAAM,GAAuB,CAAC,MAAK,EAAE,KAAI,EAAE,IAAG,GAC9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,SAAS,EAAO,EAAM,IAAM,GAC/C,KAAiB,GAAU,GC1CjC,AA4CA,YACI,EAAiB,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IACzD,GAAM,GAAK,EAAgB,EAAG,IAAK,8BACnC,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA;sBACU,EAAG,KAAI,MAC3B,EACI,GAAW,GACX,WAAM,MAAA,+DACF,gCAA+B,EAAW,OAElD,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,6BACd,EAAK,EAAa,EAAM,EAAO,GAEnC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAoB,CAAC,EAAG,GAExB,EAAkB,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEpD,KAIE,IAA6B,EAAG,CAAC,4BAA2B,KCvFzE,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AA0DA,YAAc,GAEZ,SACI,GAAgB,GAAI,WAAM,MAAA,+CACvB,SAAC,EAAsB,GAE5B,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MACxC,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,WAAa,KAC1D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,kFAGN,GAAW,GACJ,EAAM,MAkCnB,YAAe,GAEb,SACI,GAAgB,GAAI,WAAM,MAAA,gDACvB,SAAC,EAAgC,GACtC,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,qFAGV,GAAM,GAAQ,GAAqB,EAAM,OAAQ,WAAY,MACvD,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,YAAc,KAC3D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,MAAA,OAAA,IAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,iGAGN,GAAW,GACJ,KA8Bb,YAA0D,GAKxD,SACI,GAAgB,GAChB,WAAM,MAAA,uDACH,SAAC,EAAM,GACZ,EACI,YAAa,GACb,WAAM,MAAA,wDACV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,6DACJ,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,UAAW,GACJ,CAAC,KAAM,EAAM,GAAS,MAAK,IAmCtC,YAAyC,GAKvC,SACI,GAAgB,GAChB,WAAM,MAAA,wDACH,SAAC,EAAgB,GACtB,EACI,MAAM,QAAQ,IAAS,EAAK,MAAM,SAAA,GAAO,MAAA,aAAe,KACxD,WAAM,MAAA,uEAEV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,iEACV,GAAM,GAAM,EAAO,UAAU,WAAM,MAAA,GAAC,MAAA,OAAI,IAAO,EAAM,GACrD,MAAI,IAAM,MACR,GACI,EAAI,MAAM,MAAO,EAAG,MACpB,yGAGN,GAAW,EAAI,OACR,GAkCX,YAAuB,EAAiB,GAEtC,EACI,GAAgB,GAChB,WAAM,MAAA,wDACV,EACI,GAAW,MACP,MAAM,QAAQ,IAAY,EAAQ,MAAM,SAAA,GAAK,MAAA,aAAa,MAC9D,WACI,MAAA,kFAGR,GAAM,GAAmB,GAAW,KACpC,GAAI,CAAC,GAEH,EAAU,GACV,OAAW,KAAW,GAAO,oBAC3B,EAAQ,KAAK,EAAO,oBAAoB,IAI5C,GAAM,GACF,EAAmB,EAAQ,OAAO,SAAA,GAAY,MAAA,CAAC,EAAS,YAAa,KAGnE,EAAmB,EAAQ,OACjC,EAAU,EAAQ,OAAO,SAAA,GAAY,MAAA,GAAS,YAC9C,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,kEACF,kCAAiC,EAAgB,kBACjD,eAER,GAAM,GAAmB,GACnB,EAAA,EAAA,UAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,EACI,EAAM,KAAK,SAAA,GAAK,MAAA,IAAK,OACrB,WAAM,MAAA,iMAGV,EACI,EAAM,OAAS,EACf,WAAM,MAAA,iEACF,oBAAmB,EAAM,KAAI,aAErC,GAAM,GAA6B,GACnC,SAAQ,QAAQ,SAAC,EAAG,GAClB,AAAI,EAAM,IAAM,MACd,GAAW,EAAE,MAAQ,EAAM,MAG/B,AAAI,GAAyB,MAG3B,EAAsB,QAAQ,SAAA,GAAK,MAAA,GAAW,EAAE,MAAQ,OAEnD,CAAC,MAAK,EAAE,MAAO,GA2CxB,YAAsC,GAEpC,MAAO,GAAO,WAAW,GAG3B,YAAoB,GAClB,GAAM,GAAmB,EAAM,OAAO,SAAA,GAAK,MAAA,IAAK,OAAM,OACtD,GAAI,EAAmB,EACrB,KAAM,IAAI,OACN;sEC7XR,AAuCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAuB,CAAC,EAAG,GACjC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,IAAI,IAAK,EAC5B,KAAiB,OAEV,IAAM,EAAG,CAAC,KAAI,KC/C3B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,SAAS,GAC7B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAW,EAAG,CAAC,UAAS,KChDrC,AAyCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAK7B,EAAW,GAAW,SAAC,GAI3B,GAAM,GAAQ,GAAI,GAAS,GAAI,KAEzB,EAAW,SAAC,GAChB,GAAM,GAAO,EAAI,EAAI,GAAQ,GAAI,KACjC,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAEL,IAAa,EAAG,CAAC,YAAW,KC9DzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAoB,EAAM,EAAG,OAC1C,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAS,OAG1D,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GAAgB,GAClB,EAAI,MAAO,GAAoB,EAAM,EAAG,QAC5C,EAAM,EAAQ,EAAK,GACnB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,iBAAkB,EAAM,SAAQ,GAEzD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCnE3B,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAGhB,GAAM,GAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IACN,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,IAAI,EAAW,GACnC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAEzB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KC/F3B,AAsDA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAAU,EAAgB,EAAQ,SAAU,cAKlD,GAHA,AAAI,IAAS,IACX,GAAO,EAAQ,KAAO,GAEpB,IAAS,EAAQ,KAAO,EAC1B,KAAM,OACF,gEACA,oBAAmB,EAAQ,KAAI,iBAAiB,IAGtD,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACX,EAAO,GAAI,EAAQ,EAAM,IACzB,EAAU,GAAI,EAAQ,GACtB,EACF,GAAI,GAAK,EAAS,WAAY,GAAI,GAAI,GAAI,GAAU,EAAM,KAC9D,SAAK,CAAC,IACC,GAGH,EAA2B,CAAC,OAAQ,GACpC,EAAyB,CAAC,KAAI,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAY,MAGZ,IAAa,EAAG,CAAC,YAAW,KCpFzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAO,GAAI,EAAI,EAAM,IACrB,EAAI,GAAI,EAAI,GACZ,EAAI,GAAI,GACR,EAAI,GAAI,EAAG,GACX,EAAI,GAAI,GACR,EAAM,GAAI,EAAQ,EAAM,EAAE,OAAQ,GAExC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,MAGI,IAAY,EAAG,CAAC,WAAU,KChFvC,AAyCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA2B,CAAC,EAAG,EAAI,EAAG,GAE5C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,WAAW,EAAI,IAClC,EAAgC,KAAiB,OAGjD,IAAa,EAAG,CAAC,YAAW,KCtDzC,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,WAAW,IAAK,EACnC,KAAiB,OAGV,IAAa,EAAG,CAAC,YAAW,KC9CzC,AAwCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,QAC1C,EAAK,EAAgB,EAAG,IAAK,YAAa,QAChD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA0B,CAAC,EAAG,EAAI,EAAG,GAC3C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IACjC,EAAgC,KAAiB,OAEjD,IAAY,EAAG,CAAC,WAAU,KCnDvC,AA0CA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,UAA2B,EAAG,MAAO,EAAG,OAGjC,GAAW,GAAU,EAAG,GAAI,GAAW,GAAW,EAAG,QAGjD,IAAa,EAAG,CAAC,YAAW,KCpDzC,AAwDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAC7B,EAAY,EAEd,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAEA,EAEJ,MAAI,GAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UAC9C,EAAI,EAAI,QAER,EAAI,EAAQ,QAAQ,EAAK,GAG3B,EAAK,CAAC,EAAK,IAEJ,GAGH,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAEhE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KCrHnC,AA2EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GAJiB,AAAA,IAAA,QAAA,GAAA,CAA+C,EAAG,EAAG,IAGtE,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE/B,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAErB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GACE,EAAI,EAAQ,UAAU,EAAK,GACjC,SAAK,CAAC,EAAK,IACJ,GAGH,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAE/D,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KClJvC,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAK,EAAgB,EAAG,IAAK,qBAE7B,EAAkC,CAAC,EAAG,GACtC,EACuB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,oBAAmB,GAErE,EAAS,EAAO,UACH,GAAmB,EACnB,GAEnB,MAAO,CAAC,OAAQ,EAAO,GAAI,QAAS,EAAO,OAGhC,IAAoB,EAAG,CAAC,mBAAkB,KC7EvD,YAsCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAM,EAAO,WACpB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAoB,GAAc,GAAQ,GACzD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC7C1C,YAuCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAK,EAAO,WACnB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAmB,GAAc,GAAQ,GACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC9C1C,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAS,GAA0B,EAAG,MAAO,GAC7C,EAAc,EAAO,GACrB,EAAa,GAAc,GAI3B,EAAW,GAAW,SAAC,GAC3B,GAAM,GAAmB,GAAO,GAE1B,EAAU,EAAiB,QAAU,EAAE,MACzC,EACA,GAAK,EAAG,EAAiB,OACvB,EAAM,GAAI,EAAS,GACnB,EAAQ,GAAI,EAAK,EAAM,GAEvB,EAAW,SAAC,GAChB,GAAM,GAAkB,EAAE,MAAM,QAChC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,GAAI,EAAI,EAAY,GAAK,EAAE,MAAO,YAAa,GAC5D,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAGL,IAAO,EAAG,CAAC,MAAK,KCpC7B,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAGpD,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GACF,GAA+B,EAAI,MAAO,GAC9C,EAAM,EAAQ,EAAG,GACjB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAGhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAEH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,AAuDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAEH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AAmCA,YAAmC,GACjC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAQ,GACR,EAAe,CAAC,GAChB,EAA2B,GACjC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,SAAK,CAAC,IACC,EAAQ,OAAO,IACrB,CAAC,EAAG,GAAK,KAAiB,SAAU,EAAO,EAAc,MAGjD,IAAS,EAAG,CAAC,QAAO,KC9CjC,AA4CA,YACI,EAAsB,EACtB,GADsB,AAAA,IAAA,QAAA,GAAA,MACtB,AAAA,IAAA,QAAA,GAAA,IACF,EAAI,EAAgB,EAAG,IAAK,WAC5B,GAAM,GAAO,GAAe,EAAM,EAAE,OAC9B,EAAQ,GAAK,EAAG,EAAM,GACxB,EAAgB,EAAM,MAC1B,AAAK,GACH,GAAgB,GAAqB,EAAM,MAAO,IAEpD,GAAM,GACF,GAAO,GAAI,GAAK,EAAG,WAAY,EAAQ,EAAO,KAC5C,EAAW,GAAK,EAAY,EAAM,GACxC,MAAO,CAAC,KAAM,EAAO,SAAQ,MAGlB,IAAU,EAAG,CAAC,SAAQ,KChBnC,YACI,EAA2B,EAC3B,EACA,GAOF,OANM,GAAQ,EAAgB,EAAM,OAAQ,gBACtC,EAAK,GAAqB,EAAG,IAAK,gBAClC,EAAK,GAAqB,EAAG,IAAK,gBAEpC,EAAQ,EACN,EAAY,GACT,EAAI,EAAG,EAAI,EAAU,OAAQ,KACpC,GAAM,GAAS,EAAU,GAAG,EAAO,EAAG,GAAI,EAAG,IAC7C,EAAU,KAAK,EAAO,IACtB,EAAU,KAAK,EAAO,IACtB,EAAQ,EAAO,GAIjB,OAFM,GAAmB,GACnB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,GAAK,EACzC,EAAK,KAAK,EAAU,IACpB,EAAK,KAAK,EAAU,EAAI,IAE1B,MAAO,CAAC,EAAM,MAEH,IAAe,EAAG,CAAC,cAAa,KCpE7C,AA4CA,YACI,EAAsC,EAAoB,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,eAC5C,EAAc,EAAQ,KACtB,EAAW,EAAQ,KACzB,GAAI,EAAc,EAChB,KAAM,IAAI,OACN,+DACG,GAAW,MAEpB,GAAI,EAAW,EACb,KAAM,IAAI,OAAM,gDAAgD,GAElE,EAAO,GAAQ,KAAK,SACpB,GAAM,GACF,IAAa,EAAI,EAAQ,EAAS,CAAC,EAAG,KAAO,EAC3C,EAAM,EAAO,cACf,SAAA,GAAW,MAAA,GAAQ,YAAY,EAAU,EAAY,EAAY,IACjE,CAAC,SAAQ,IAEb,MAAO,KAAa,EAAI,EAAQ,EAAK,CAAC,EAAI,OAAqB,KAGpD,IAAc,EAAG,CAAC,aAAY,KCpE3C,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,GAAY,MAAA,GAAQ,SAAS,EAAI,IAEjE,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KC1DrC,AAwCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCrD7B,AA0CA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAI,EAAG,QAAU,aACf,GAAM,GAAI,GAAS,GAAK,IAClB,EAAI,GAAU,GAAK,IACzB,MAAO,IAAQ,EAAG,GAGpB,MAAO,GAAQ,SAAS,IAGpB,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCtBrC,YACI,EAAyB,GAC3B,GAAM,GAAM,EAAgB,EAAI,KAAM,gBAChC,EAAM,EAAgB,EAAI,KAAM,gBAEtC,EACI,EAAI,OAAS,GAAK,EAAI,OAAS,EAC/B,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,EAAQ,EAAK,CAAC,EAAG,KAC9B,MAAO,IAAO,EAAM,MAGT,IAAe,EAAG,CAAC,cAAa,KCtD7C,AAmDA,YACI,EAAiB,EAAmC,GAAA,AAAA,IAAA,QAAA,GAAA,GACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAA0B,SAAC,EAAS,GACxC,SAAK,CAAC,IACC,EAAQ,IAAI,EAAI,EAAU,IAG7B,EAAoB,CAAC,SAAQ,EAAE,cAAa,GAC5C,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAM,EAAG,CAAC,KAAI,KC5C3B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,qDACH,GAAI,EAAG,CAAC,GAAW,MAGf,IAAQ,EAAG,CAAC,OAAM,KCT/B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCV/B,YACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,EACvD,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCX/B,YACI,EACA,EAIA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,GACnD,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCxC/B,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAEnC,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,cAAc,EAAG,KAAI,kCACvB,EAAW,SAEnB,EACI,EAAS,SAAW,EAAW,OAC/B,WAAM,MAAA,qBACF,EAAS,OAAM,kCAAkC,EAAW,SAEpE,EACI,EAAG,MAAM,OACL,SAAC,EAAG,EAAG,GACL,MAAI,GAAI,GAAK,GAAK,EAAW,OACpB,GACD,GAAI,EAAS,EAAI,GAAG,GAAK,EAAS,EAAI,GAAG,IACtC,EAAW,EAAI,KACnB,EAEA,GAET,IACJ,WAAM,MAAA,4BAA4B,EAAG,MAAM,MAAM,GAAE,kBAC/C,EAAS,WAAU,qCACnB,EAAW,aAEnB,GAAM,GAA0B,SAAA,GAC5B,MAAA,GAAQ,eAAe,EAAI,EAAY,IAErC,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,SAAQ,GAExD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,KCrHjD,AAwDA,YACI,EAAqB,EACrB,EAA0B,EAC1B,EAAqC,GACvC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,AAAI,GAAW,MACb,GAAU,GAEZ,AAAI,IAAQ,GACV,GAAM,SAGR,GAAM,GAAK,EAAgB,EAAO,IAAK,WACnC,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,yDACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAC1C,EACF,CAAC,EAAS,eAAgB,EAAS,eAOnC,EACJ,AAAI,IAAQ,OACV,EAAc,GACV,CAAC,EAAS,aAAc,EAAS,aAAc,GAEnD,EAAc,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,IAG7B,GAAM,GAAgB,EAAS,KAAO,GAAK,EAAS,KAAO,EACrD,EAAA,GAAA,CAAA,EAAA,SAAA,EAAA,SAAA,EAAA,GAAC,EAAA,EAAA,GAAiB,EAAA,EAAA,GAElB,EAAe,EAAgB,EAAM,QACrC,EACF,EAAgB,EAAM,GAAe,EAAK,EAAU,GAElD,EAAY,IAAgB,MAC9B,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAChD,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAC9C,EAAI,IAEJ,EAAM,EAAgB,EAAI,GAAe,EAAG,EAAU,GAE5D,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,EAMT,YACI,EAA8B,EAC9B,GACF,GAAM,GAAW,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,EAAa,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KACpC,EAAiB,EAAW,OAAO,EAAU,GAC7C,EAAc,EAAW,IAAI,SAAC,EAAG,GAAM,MAAC,GAAI,EAAe,GAAK,GAAK,IACrE,EAAS,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAY,KAClD,EAAW,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAS,GAAI,EAAO,MACzD,EAAQ,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAG,EAAY,MACvD,MAAO,CAAC,EAAU,GAMpB,YACI,EAA+B,GAGjC,GAAM,GAAqB,EAAY,IAAI,SAAC,EAAG,GAC7C,MAAO,GAAK,GAAI,GAAM,GAAS,GAAK,KAEhC,EAAgB,EAAmB,IAAI,SAAA,GAAK,MAAA,GAAI,IAIhD,EAAgB,EAAc,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAI,KACtD,EAAc,EAAc,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAc,KAClE,MAAO,GAAc,IAAI,SAAC,EAAG,GAC3B,MAAO,CAAC,EAAc,GAAI,EAAY,SAI7B,IAAO,EAAG,CAAC,MAAK,KChK7B,AAsDA,YACI,EAAyB,SACvB,EAAQ,EAAgB,EAAM,OAAQ,OACtC,EAAO,EAAgB,EAAK,MAAO,OACvC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,GAAM,GAAoB,CAAC,EAAG,EAAO,EAAG,GAClC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,EAAO,GAC7B,SAAK,CAAC,EAAO,EAAM,IACZ,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AA0CA,YAAkC,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAS,EAAgB,EAAO,QAAS,SAEzC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAGH,EAAsB,CAAC,EAAG,EAAI,MAAO,GAC3C,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AA8DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,QAE3B,EAA+B,SAAC,GACpC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAEhB,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,KAAK,EAAW,GACpC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAGzB,MAAO,IAGH,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,EAAE,SAAQ,GAExC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAO,EAAG,CAAC,MAAK,KChG7B,AAiCA,YACI,EAAoB,EACpB,GACF,GAAM,GAAO,GAAc,GACvB,EAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,IAEd,MAAO,GAAO,WAAW,EAAQ,EAAO,MAG7B,IAAO,EAAG,CAAC,MAAK,6OC1B7B,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,cAEJ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,AAyCA,GAAA,IAAA,WAUE,WACI,EAAc,EAAsB,EACpC,EAAqB,GACvB,KAAK,KAAO,EACZ,KAAK,OAAS,EACd,KAAK,MAAQ,EACb,KAAK,QAAU,IACf,KAAK,UAAY,EACjB,AAAI,KAAK,WACP,MAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,EACvC,KAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,GAEzC,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,OAAS,GAAgB,EAAU,YAInC,SAAA,UAAA,UAAP,WACE,GAAI,CAAC,MAAM,KAAK,UACd,GAAM,GAAQ,KAAK,QACnB,YAAK,QAAU,IACR,EAKT,OAFI,GAAiB,EACjB,EAAU,GACP,CAAC,IACN,GAAI,GAAE,OAAU,EAAE,OAAU,EAAC,OAC7B,EACE,GAAK,EAAI,KAAK,SAAW,EACzB,EAAK,EAAI,KAAK,SAAW,EACzB,EAAI,EAAK,EAAK,EAAK,QACZ,GAAK,GAAK,IAAM,GAEzB,GAAM,GAAM,KAAK,KAAK,GAAO,KAAK,IAAI,GAAK,GAC3C,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EACzC,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EAEzC,AAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,GAAU,IAId,MAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,MAAK,QAAU,KAAK,aAAa,IAE5B,KAAK,aAAa,IAInB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,OAAS,MAAQ,KAAK,QAAU,UAChC,EAEF,KAAK,MAAM,IAIZ,EAAA,UAAA,iBAAR,SAAyB,GACvB,MAAO,IAAS,KAAK,OAAS,GAAS,KAAK,OAEhD,KAIA,GAAA,WASE,WACI,EAAe,EAAc,EAC7B,GACF,KAAK,MAAQ,EACb,KAAK,KAAO,EAAI,EAChB,KAAK,MAAQ,EAEb,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,MAAQ,GAAgB,EAAU,YACvC,KAAK,MAAQ,GAAI,IAAY,EAAG,EAAG,EAAO,GAAO,KAAK,SAEtD,AAAI,EAAQ,EACV,KAAK,EAAI,EAAS,EAAI,EAEtB,KAAK,EAAI,EAAS,EAAI,EAExB,KAAK,EAAI,EAAI,KAAK,KAAK,EAAI,KAAK,GAI3B,SAAA,UAAA,UAAP,WAEE,OADI,GAAY,EAAY,EAAY,EAAW,EAAW,KAE5D,EACE,GAAI,KAAK,MAAM,YACf,EAAI,EAAK,KAAK,EAAI,QACX,GAAK,GAMd,GALA,GAAK,EAAI,EACT,EAAK,EAAI,EACT,EAAK,EAAK,KAAQ,EAAK,EACvB,EAAM,GAAM,EAAO,KAAK,EAAK,GAAI,EAAI,KAAK,IAAI,IAC9C,EAAI,KAAK,QACL,EAAI,GAAM,KAAK,IAAI,GAAK,EAC1B,MAGJ,SAAK,EAAI,KAAK,KAAQ,KAAK,EAAI,EAC/B,AAAI,KAAK,MAAQ,GACf,IAAK,KAAK,IAAI,KAAK,QAAS,EAAI,KAAK,QAEhC,KAAK,aAAa,IAGnB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,QAAU,UACV,EAEF,KAAK,MAAM,IAEtB,KAEA,GAAA,WAME,WACI,EAAS,EAAS,EAClB,GAFJ,GAAA,GAAA,KAaE,GAZE,AAAA,IAAA,QAAA,GAAA,GAAS,AAAA,IAAA,QAAA,GAAA,GAoBL,KAAA,eAAiB,WACrB,MAAC,GAAK,OAAS,MAAQ,EAAK,QAAU,WAnBxC,KAAK,IAAM,EACX,KAAK,MAAQ,EAAM,EACnB,KAAK,MAAQ,EACb,AAAI,GAAQ,MACV,GAAO,KAAK,UAEd,AAAI,MAAO,IAAS,UAClB,GAAO,EAAK,YAGV,CAAC,KAAK,kBAAoB,KAAK,OAAS,EAC1C,KAAM,IAAI,OACN,0BAA0B,EAAG,MAAM,EAAG,gCAE5C,KAAK,OAAS,GAAgB,GAOxB,SAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,iBACA,EAEF,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,WACE,MAAO,MAAK,aAAa,KAAK,IAAM,KAAK,MAAQ,KAAK,WAE1D,KCxNA,AAwCA,YACI,EAAoB,EAAe,EACnC,EAAsC,GAOxC,GARqC,AAAA,IAAA,QAAA,GAAA,GACnC,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAQ,MACV,GAAO,GAET,AAAI,GAAS,MACX,GAAQ,WAEN,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,OAAM,yBAAyB,GAI3C,OAFM,GAAS,GAAI,IAAU,EAAO,EAAM,EAAO,GAC3C,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAc,EAAG,CAAC,aAAY,KC5D3C,AAuCA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,yBAAyB,GAK3C,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAuB,GAC1D,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAe,EAAG,CAAC,cAAa,KCtD7C,AA4CA,YACI,EAAoB,EAAY,EAAY,EAC5C,GADoB,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,WAI9C,OAFM,GAAM,GAAO,EAAO,GACpB,EAAS,GAAI,IAAc,EAAQ,EAAQ,KAAM,GAC9C,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAgB,EAAG,CAAC,eAAc,KCvD/C,YAwCyB,EAAsB,GAC7C,GAAc,GACd,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,EAC3B,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAkB,KACxB,MAAO,IAAW,EAAQ,EAAO,EAAe,GC/ClD,YA8CI,EAAe,EAAc,EAC7B,GACF,GAF+B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAA,WACE,IAAS,EACX,KAAM,IAAI,OAAM,8BAGlB,GAAM,GAA+B,WACnC,GAAM,GAAgB,IAAU,EAC1B,EAA8B,EAAQ,GAAQ,EAAO,EACrD,EAA8B,EAAO,GAAS,EAAO,EAE3D,GAAI,GAAiB,GACjB,EACF,MAAO,IAAM,CAAC,GAAI,GAGpB,GAAM,GAAc,KAAK,IAAI,KAAK,KAAM,GAAO,GAAS,IAClD,EAAS,GAAoB,EAAa,GAEhD,AAAI,EAAO,GAAS,IAAS,GAG3B,GAAO,IAGT,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,IAGpB,EAAoB,CAAC,MAAK,EAAE,KAAI,EAAE,KAAI,EAAE,MAAK,GAEnD,MAAO,GAAO,cACH,EAAS,GAAiB,KAAiB,GAC3C,GCnFb,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAE7B,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,WAAW,GAC/B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAa,EAAG,CAAC,YAAW,KChDzC,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,KAAK,IAGhB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,MAAM,IAGjB,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC5D/B,AA6DA,YACI,EAAiB,GACnB,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAA+B,SAAC,GACpC,GAAM,GAAO,GAAe,EAAM,EAAG,OACrC,GAAI,EAAG,OAAS,EACd,MAAO,IAAM,GAEf,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,MAAO,GAAQ,EAAK,EAAG,QAGnB,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,KAAM,GAEnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAS,MAGT,IAAU,EAAG,CAAC,SAAQ,KClFnC,AA6BA,YAAoB,GAClB,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCrCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAGV,IAAQ,EAAG,CAAC,OAAM,KChD/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GAGH,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KCe7B,YACI,EAAiB,EACjB,EAAsC,EACtC,EAAqB,EACrB,GADqB,AAAA,IAAA,QAAA,GAAA,CAAqC,EAAG,IAC7D,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EACF,EAAgB,EAAiB,kBAAmB,mBAClD,EACF,EAAgB,EAAiB,kBAAmB,mBAEpD,EAAM,EACN,EAAe,GAMnB,GALA,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGvD,IAAe,OACjB,KAAM,IAAI,OACN,sFAIN,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,MAAM,KAAO,EAC9B,WACI,MAAA,qEACA,wBAAuB,EAAiB,MAAM,GAAE,OACxD,EACI,EAAiB,MAAM,KAAO,EAC9B,WAAM,MAAA,+DACF,8BAA6B,EAAiB,MAAM,GAAE,OAE9D,GAAM,GAAa,EAAiB,MAAM,GACpC,EAAoB,EAAiB,MAAM,GACjD,EACI,EAAiB,MAAM,KAAO,EAAa,EAC3C,WACI,MAAA,qEACA,YAAW,EAAa,EAAiB,MACzC,YAAW,EAAiB,MAAM,GAAE,OAE5C,GAAM,GAAY,GACd,EAAK,EAAkB,EAAS,EAAK,EAAY,GAC/C,EAAkB,EAClB,EACF,GAAO,EAAW,EAAkB,EAAiB,QAAS,GAElE,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KCxFnD,YACI,EAAsB,uHAClB,SAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aAEnC,EACI,EAAG,QAAU,EAAG,MAChB,WAAM,MAAA,kDACF,EAAG,MAAK,YAAY,EAAG,MAAK,OAEpC,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAEtE,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAExD,CAAA,EAAM,EAAG,eAAjB,SAAQ,EAAA,OACA,CAAA,EAAM,EAAG,eAIvB,IAJM,EAAQ,EAAA,OACR,EAAO,GAAI,KAAI,GAEjB,EAAa,EACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,AAAK,EAAK,IAAI,EAAM,KAClB,IAMJ,IAFM,EAAS,GAAI,IAAa,CAAC,GAAa,EAAG,OAC3C,EAAU,GAAI,IAAa,CAAC,GAAa,SACtC,EAAI,EAAG,EAAI,EAAG,EAAI,EAAM,OAAQ,IACvC,AAAK,EAAK,IAAI,EAAM,KAClB,GAAO,OAAO,GAAK,EAAM,GACzB,EAAQ,OAAO,GAAK,EACpB,KAGJ,MAAA,CAAA,EAAO,CAAC,EAAO,WAAY,EAAQ,qBAExB,IAAiB,GCxF9B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC7C7B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AA6BA,YACI,EAAwB,EAAe,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,CAAC,GAAQ,CAAC,OAEhB,IAAU,EAAG,CAAC,SAAQ,KCtCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AAgDA,YAAoC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxD,GAAM,GAAU,EAAgB,EAAQ,SAAU,UAAW,WAK7D,GAHA,AAAI,IAAQ,IACV,GAAM,EAAQ,KAAO,GAEnB,IAAQ,EAAQ,KAAO,EACzB,KAAM,OACF,4DACA,oBAAmB,EAAQ,KAAI,gBAAgB,IAGrD,GAAM,GAAwB,CAAC,OAAQ,GACjC,EAAsB,CAAC,IAAG,GAEhC,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAI,EAAQ,QAAQ,EAAS,GACnC,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAU,EAAG,CAAC,SAAQ,KCzEnC,AAyCA,YAAc,GACZ,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,qDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAoB,CAAC,MAAK,GAEhC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAU,EAAM,KAAK,EAAO,GAC5B,EAAS,EAAQ,IAAI,GAC3B,MAAO,GAAO,QAAQ,EAAM,QAC3B,EAAgC,KAAqB,OAG7C,IAAM,EAAG,CAAC,KAAI,KC5D3B,AA0CA,YAAe,GACb,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,sDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAqB,CAAC,MAAK,GAEjC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAoB,EAAQ,EAAO,CAAC,EAAO,IAC3C,EAAS,EAAQ,KAAK,GAC5B,MAAO,GAAQ,EAAQ,EAAM,QAC5B,EAAgC,KAAqB,OAG7C,IAAO,EAAG,CAAC,MAAK,KC7D7B,AAgDA,YAAgB,GACd,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EACvB,EACJ,GAAI,GAAsB,GACxB,GAAM,GAAe,EAAQ,EAAO,CAAC,EAAO,IAC5C,EAAM,GAAK,QAIX,GAAM,GAAc,CAAC,EAAO,EAAK,GAAqB,IAChD,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IACzC,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IAEzC,EACF,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACjE,EAA0B,EAC5B,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACnE,GAAO,KAEL,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EACF,EAAQ,GAAQ,EAAG,GAAI,CAAC,EAAY,GAAI,EAAY,KACxD,EAAM,GAAK,GAIb,GAFA,EAAM,GAAK,GAEP,EAAM,OAAS,GAAK,EAAM,MAAM,KAAO,GACzC,GAAM,GAAO,EACP,EAAQ,EAAM,MAAM,GAC1B,EAAM,EAAQ,EAAK,CAAC,EAAO,EAAI,MAAM,GAAK,EAAO,EAAI,MAAM,KAC3D,EAAK,UAEP,MAAO,MAGI,IAAQ,EAAG,CAAC,OAAM,iBC3D3B,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAa,GACjB,GAAI,MAAQ,IAAqB,SAC/B,EACI,EAAE,MAAM,GAAQ,IAAoB,EACpC,WAAM,MAAA,kDACV,EACI,GAAI,OAAM,GAAiB,KAAK,EAAE,MAAM,GAAQ,QAEpD,GAAM,GAAY,EAAgB,OAAO,SAAC,EAAO,GAC/C,MAAI,KAAU,IACZ,IAAS,GAEJ,GACN,GACH,EACI,GAAa,EACb,WAAM,MAAA,4DACV,GAAM,GAAW,EAAgB,QAAQ,IAGzC,GAAI,IAAa,IACf,GAAM,GAAQ,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,EAAI,IAC/D,EAAgB,GAAY,EAAE,MAAM,GAAQ,EAE9C,EACI,EAAE,MAAM,KAAU,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IACvD,WAAM,MAAA,gEACV,EAAa,EAGf,MAAO,GC1DT,AA+DA,YACI,EAAsB,EAAkC,GAAA,AAAA,IAAA,QAAA,GAAA,GAC1D,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAG,OAAO,GACvC,EAAa,GAAiB,EAAI,EAAiB,GACzD,MAAO,GAAQ,MAAM,EAAI,EAAY,IAGjC,EAAuB,CAAC,EAAG,GAC3B,EAAoB,CAAC,gBAAe,EAAE,KAAI,GAEhD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCjF/B,AA+CA,YAAe,EAAe,GAC5B,EACI,EAAM,QAAU,UAChB,WAAM,MAAA,mDAAmD,EAAM,QAEnE,GAAI,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACpD,EAAQ,EAAM,KAAO,EAEvB,EACJ,GAAI,GAAa,MAAQ,EAAY,GAEnC,GAAM,GAAQ,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAC7B,EAAO,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAClC,EAAK,EAAM,MAAM,OAAS,GAAK,EAC/B,EAAgB,GAAM,EAAO,EAAO,GACpC,EAAqB,UACZ,GAAa,MAAQ,EAAY,GAE1C,GAAM,GAAa,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KACxC,EAAW,EAAM,MAAM,OAAS,GAAK,EAAY,EACjD,EAAgB,GAAO,CAAC,EAAO,GAAM,IAAc,EAAM,MAAM,OAAS,GACxE,EAAqB,MAErB,GAAgB,EAIlB,GAAM,GAAa,GAAU,GACvB,EACF,EAAQ,GAAQ,EAAe,GAAa,CAAC,EAAO,IAElD,EAAM,GAAI,GAGV,EAAO,KAAK,MAAM,EAAqB,GAAK,EAC5C,EAAa,GAAK,GAClB,EAAa,GAAK,GAClB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GACxB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GAExB,EAAc,EAAc,MAAM,QACxC,SAAY,EAAc,MAAM,OAAS,GAAK,EAEvC,EACH,GAAQ,EAAqB,GAAI,EAAqB,IAAK,MAGpD,IAAO,EAAG,CAAC,MAAK,KClG7B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AAoDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,qBAC7B,EAAK,EAAgB,EAAG,IAAK,qBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,kBAAkB,EAAI,GAC1C,SAAK,CAAC,EAAI,IACH,GAGH,EAAkC,CAAC,EAAG,EAAI,EAAG,GAC7C,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAmB,MAGnB,IAAoB,EAAG,CAAC,mBAAkB,KC1EvD,AAwCA,YAAoC,EAAsB,GACxD,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,MAAO,GAAQ,EAAI,GAAa,EAAG,MAAO,GAAM,aAGrC,IAAU,EAAG,CAAC,SAAQ,KC7CnC,AAyCA,YACI,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAChC,GAAM,GAAW,GAAqB,EAAS,UAAW,SAK1D,GAHA,EACI,EAAS,QAAU,EAAG,WAAM,MAAA,yCAE5B,EAAS,SAAW,EACtB,MAAO,IAAW,EAAS,GAAI,GAGjC,GAAM,GAAO,EAAS,GAAG,KACnB,EAAQ,EAAS,GAAG,MACpB,EAAQ,EAAS,GAAG,MAE1B,EAAY,GAAQ,EAAM,WAAM,MAAA,uCAEhC,EAAS,QAAQ,SAAA,GACf,GACI,EAAO,EAAE,MACT,yDACJ,EACI,IAAU,EAAE,MACZ,WAAM,MAAA,4DAGZ,GAAM,GAAkB,EAAS,IAAI,SAAA,GAAK,MAAA,IAAW,EAAG,KAOxD,MAAO,IAAO,EAAiB,MAGpB,IAAQ,EAAG,CAAC,OAAM,KC7E/B,AAwCA,YAAiC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAEhB,IAAO,EAAG,CAAC,MAAK,KClD7B,AA+DA,YACI,EAAsB,EAAiB,EAAe,EACtD,EAAe,EAAa,EAAkB,EAC9C,GADA,AAAA,IAAA,QAAA,GAAA,GAAe,AAAA,IAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAAkB,AAAA,IAAA,QAAA,GAAA,GAC9C,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAE3B,EAA+B,SAAC,GACpC,AAAI,GAAW,MACb,GAAU,GAAI,OAAM,EAAM,SAG5B,GAAM,GAAe,GAAW,GAChC,GAAI,EAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAI,IAAiB,GAAK,IAAgB,EACxC,KAAM,IAAI,OACN,iEAGN,GAAI,IAAiB,GAAK,IAAmB,EAC3C,KAAM,IAAI,OACN,oEAGN,GAAM,GAAsB,EAAG,KAAO,EAAM,OAGtC,EAAa,GAAW,GACxB,EAAW,EAAG,MAAM,QAC1B,EAAW,QAAQ,SAAA,GACjB,EAAM,GAAQ,EACd,EAAI,GAAQ,EACZ,EAAS,OAAO,EAAM,EAAG,KAE3B,EAAK,EAAQ,EAAI,GAEX,GAAA,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GACJ,EAAA,EAAA,MACA,EAAA,EAAA,IACA,EAAA,EAAA,QAKF,EAAQ,EACR,EAAM,EACN,EAAU,EAEV,GAAM,GAAa,GAAW,GAE9B,EAAW,QAAQ,SAAA,GACjB,EAAI,GAAQ,EAAM,GAAQ,EAC1B,EAAQ,GAAQ,IAIlB,GAAM,GAAO,GAAgB,EAAO,EAAK,GAEnC,EAAW,EAAK,OAAO,SAAC,EAAG,GAAS,MAAA,GAAW,QAAQ,KAAU,KAEjE,EAAa,EAAQ,MAAM,SAAA,GAAK,MAAA,KAAM,IAC5C,GAAI,EACF,MAAO,GAAQ,GAAM,EAAI,EAAO,GAAO,GAGzC,GAAM,GAAM,EAAQ,aAAa,EAAI,EAAO,EAAK,GACjD,MAAO,GAAQ,EAAK,IAGhB,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAC/B,MAAK,EACL,IAAG,EACH,QAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,YAAW,EACX,eAAc,GAGhB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAe,EAAG,CAAC,cAAa,KCvJ7C,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,kEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,gFAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,sEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,wEAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GCjElD,YAgDI,EACA,EACA,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,0EAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,SAAQ,GACJ,EACG,GAAW,EAAQ,EAAO,EAAe,GCpElD,AAmDA,YACI,EAAiB,EAAO,GAAP,AAAA,IAAA,QAAA,GAAA,GAAO,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAK,EAAgB,EAAG,IAAK,QACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAU,EAAG,MAAM,EAAG,MAAM,OAAS,GAC3C,GAAI,EAAI,EACN,KAAM,IAAI,OACN,uDAAuD,EAAO,KAC9D,YAAW,IAGjB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,EAAC,EAAE,OAAM,GAE7B,EAAA,EAAA,cAAA,SAAA,GAAA,MAAA,GAAA,KAAA,EAAA,EAAA,IAAA,EAAA,KAAA,GAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAKf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAO,EAAG,CAAC,MAAK,KC3E7B,AA4CA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,qCAKlB,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAsB,GACzD,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAkB,EAAG,CAAC,iBAAgB,KC3DnD,AA6EA,YACI,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAEnB,GAAM,GAAK,EAAgB,EAAG,IAAK,SAAU,MAC7C,EAAO,EAAG,KAAO,EAAG,WAAM,MAAA,yCAE1B,GAAM,GAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAC1B,EAAA,EAAA,UAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAS,EAAG,CAAC,QAAO,KC3FjC,AA8CA,YACI,EAAiB,EAAiC,GACpD,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EACF,EAAgB,EAAY,aAAc,qBAAsB,SACpE,EAAO,GAAM,GAAc,WAAM,MAAA,qCAEjC,GAAM,GAAmC,CAAC,EAAG,EAAI,WAAY,GACvD,EAAiC,CAAC,YAAW,GAE7C,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,mBAAmB,EAAI,EAAa,GACxD,SAAK,CAAC,IACC,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAoB,MAGpB,IAAqB,EAAG,CAAC,oBAAmB,KCnEzD,AA2CA,YAAkB,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACtC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,EACI,GAAQ,CAAC,EAAG,MAAM,QAAU,EAAO,EAAG,MAAM,OAC5C,WACI,MAAA,UAAU,EAAI,gBAAgB,EAAG,MAAM,OAAM,KAAK,EAAG,MAAM,OAAM,MACzE,AAAI,EAAO,GACT,IAAQ,EAAG,MAAM,QAEnB,GAAM,GAAuB,CAAC,MAAO,GAC/B,EAAqB,CAAC,KAAI,GAC1B,EAAiC,SAAC,GACpC,MAAA,GAAQ,QAAQ,EAAI,IACxB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAU,EAAG,CAAC,SAAQ,KC7DnC,YAsCI,EAAyB,EAAkB,EAC3C,GADyB,MAAA,KAAA,QAAA,GAAA,IAEpB,EAAO,aAAa,EAAc,EAAW,EAAM,GCxC5D,YAuB0B,EAAqB,GAE7C,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,AAAI,EAAS,IACX,EAAQ,KAAK,GAOjB,OAHM,GAAW,GAAO,EAAW,SAE7B,EAAM,GAAO,CAAC,EAAQ,OAAQ,EAAU,QAAS,SAC9C,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAM,EAAS,WAAW,EAAQ,IAClC,EAAS,EAAI,EAAU,OAC7B,EAAI,OAAO,IAAI,EAAK,GAEtB,MAAO,GAAI,WCEb,YAA2B,uGACnB,SACF,EAAgB,EAAW,YAAa,aAAc,QAC7C,CAAA,EAAM,EAAW,eAAxB,SAAO,EAAA,OACP,EAAM,GAAU,EAAW,MAAO,GACxC,AAAI,IAAc,GAChB,EAAW,UAEb,CAAA,EAAO,WAGI,IAAa,GCpD1B,AA6CA,YACI,EAA2B,EAC3B,2HAcF,IAbM,EAAU,EAAgB,EAAQ,SAAU,YAC5C,EAAQ,EAAgB,EAAM,OAAQ,WAAY,QAElD,EAAW,GAAe,EAC1B,EAAU,EAAM,KAChB,EAAc,EAAQ,MAE5B,EAAY,EAAU,EAAG,WAAM,MAAA,0BAC/B,GACI,EAAY,MAAM,EAAU,EAAW,GAAU,EAAM,MACvD,qEAEA,EAAc,EACT,EAAI,EAAU,EAAI,EAAW,EAAS,IAC7C,GAAe,EAAY,GAEvB,SACF,EAAY,MAAM,EAAG,GAChB,OAAO,CAAC,GAAc,EAAY,MAAM,EAAW,IACtD,EAAiB,EAAQ,EAAS,GAClC,EAAe,EAAQ,EAAO,CAAC,KACX,CAAA,EAAM,GAAW,WAArC,SAAoB,EAAA,OACpB,EAAU,GAAQ,EAAmB,CAAC,IAEtC,EAAM,GAAO,EAAgB,EAAS,GAG5C,AAAI,IAAW,GACb,EAAQ,UAEV,AAAI,IAAS,GACX,EAAM,UAER,EAAQ,UACR,EAAe,UACf,EAAa,UACb,EAAkB,UAElB,CAAA,EAAO,WAGI,IAAmB,GCzFhC,AAuCA,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAK,EAAgB,EAAG,IAAK,kBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,6BAC/B,GAAS,EAAI,GAYtB,YAAuC,EAAiB,GACtD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAK,EAAgB,EAAG,IAAK,cACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,yBAC/B,GAAK,EAAI,GAGlB,YAAwC,EAAiB,GACvD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,eAC7B,EAAK,EAAgB,EAAG,IAAK,eACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,0BAC/B,GAAM,EAAI,GAGnB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAK,EAAgB,EAAG,IAAK,mBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,8BAC/B,GAAU,EAAI,GAGvB,YAA0C,EAAiB,GACzD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,4BAC/B,GAAQ,EAAI,GAGrB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EAAK,EAAgB,EAAG,IAAK,sBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,iCAC/B,GAAa,EAAI,MAGb,IAAc,EAAG,CAAC,aAAY,KAC9B,GAAqB,EAAG,CAAC,oBAAmB,KAC5C,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAkB,EAAG,CAAC,iBAAgB,KACtC,GAAa,EAAG,CAAC,YAAW,KAC5B,GAAiB,EAAG,CAAC,gBAAe,KCpHjD,AA2CA,YAAsC,EAAiB,GACrD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAS,GAC7C,UACI,6EAGJ,GAAuB,EAAK,MAAO,EAAI,MAAO,wBACvC,GAAI,EAAM,GAanB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,6BACpC,EAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,2BACpC,GAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAWjB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAWrB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAarB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,2BAC7B,EAAK,EAAgB,EAAG,IAAK,2BACnC,UACI,EAAG,MAAO,EAAG,MAAO,sCACjB,GAAkB,EAAI,MAGlB,IAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAA0B,EAAG,CAAC,yBAAwB,KACtD,GAAY,EAAG,CAAC,WAAU,KC5NvC,AAwEA,YACI,EAAsB,EACtB,EAA8B,GADR,AAAA,IAAA,QAAA,GAAA,aACtB,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IAChC,EAAI,EAAgB,EAAG,IAAK,QAE5B,GAAM,GAAO,GAAS,EAAG,EAAK,GAC1B,EAAgB,EAAK,MACzB,GAAI,GACF,GAAM,GAAO,GAAe,EAAM,EAAE,OACpC,EAAgB,GAA+B,EAAK,MAAO,GAE7D,MAAO,GAAQ,EAAM,GAGvB,YACI,EAAW,EAAkB,GAC/B,GAD+B,AAAA,IAAA,QAAA,GAAA,MAC3B,EAAE,OAAS,EACb,MAAO,IAAI,GAIb,GAAI,EAAE,OAAS,GAAK,IAAS,KAC3B,MAAO,IAAS,EAAQ,EAAG,CAAC,KAAM,EAAG,GAIvC,GAAI,EAAE,OAAS,GAAK,MAAO,IAAS,UAChC,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,aAAe,IAAM,EAE7B,MAAO,IAAK,GAAI,GAAI,GAAI,GAAI,GAAO,EAAG,UAAW,IAGnD,KAAM,IAAI,OAAM,qCAAqC,GAIvD,GAAI,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,GAAK,GAE7C,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,OAAS,IAAM,YAEvB,MAAO,IAAK,GAAI,GAAO,GAAI,IAG7B,KAAM,IAAI,OAAM,qCAAqC,GAGvD,KAAM,IAAI,OAAM,gCAAgC,MAGrC,IAAO,EAAG,CAAC,MAAK,KC3I7B,AA2DA,YACI,EAAiB,EAAiB,EAClC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAS,EAAgB,EAAO,QAAS,iBAE/C,GAAiB,EAAI,GACrB,EACI,GAAiB,EAAG,MAAO,EAAG,OAAQ,WAAM,MAAA,8BAEhD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAI,EAAK,GAE3B,EAAS,EAAI,GAAI,EAAI,GAAK,GAC9B,GAAI,GACF,EACI,GAAQ,KAAM,WAAM,MAAA,mDACxB,GAAM,GAAQ,EAAgB,EAAM,OAAQ,iBAC5C,EAAS,GAAI,EAAQ,GAAI,EAAK,GAAI,EAAQ,KAE5C,MAAO,IAAI,EAAI,MAGJ,IAAgB,EAAG,CAAC,eAAc,KCnF/C,AA+CA,YACI,EAA4B,EAC5B,GACF,GAAM,GAAW,EAAgB,EAAS,UAAW,YAAa,SAC5D,EAAW,EAAgB,EAAS,UAAW,aACrD,GAA8B,EAAU,EAAU,GAElD,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,UAAU,EAAU,EAAU,IAGzC,EAA0B,CAAC,QAAS,EAAU,QAAS,GACvD,EAAwB,CAAC,MAAK,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAW,MAGX,IAAY,EAAG,CAAC,WAAU,iBClCnC,EAAuB,EAAsB,EAC7C,GACF,GAAI,EAAc,QAAU,QAC1B,KAAM,IAAI,OACN,2DACA,uBAAsB,EAAc,MAAK,MAE/C,GAAI,EAAc,KAAO,EACvB,KAAM,IAAI,OACN,uDACA,mBAAkB,EAAc,MAAK,MAG3C,GAAM,GAAW,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAC7D,EAAU,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAElE,GAAI,EAAY,SAAW,EACzB,KAAM,IAAI,OACN,iDACA,KAAI,EAAY,OAAM,gBAAgB,EAAO,MAGnD,GAAM,GAAY,EAAa,KAC/B,GAAI,CAAE,GAAa,OAAS,GACtB,EAAa,OAAS,GAAK,IAAc,GAC7C,KAAM,IAAI,OACN,oCACG,GAAa,MAAK,sBAAsB,EAAQ,MAGzD,GAAI,EAAa,QAAU,EAAc,MACvC,KAAM,IAAI,OAAM,qDC/DpB,AAkEA,YACI,EAAkC,EAClC,EAA0B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC5B,GAAM,GACF,EAAgB,EAAe,gBAAiB,gBAAiB,SAC/D,EACF,EAAgB,EAAc,eAAgB,iBAC5C,EAAgB,EAClB,EAAc,eAAgB,gBAAiB,EAAc,OAEjE,GACI,EAAgB,EAAe,EAAa,GAEhD,GAAM,GAA8B,CAClC,cAAe,EACf,aAAc,EACd,aAAc,GAGV,EAA4B,CAAC,YAAW,GAE9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,cACf,EAAgB,EAAe,EAAa,IAChD,EAAgC,KAAiB,GACjD,MAGO,IAAgB,EAAG,CAAC,eAAc,KC9F/C,AA8DA,YAAmB,EAAsB,GACvC,GAAM,GAAW,EAAgB,EAAS,UAAW,WAAY,SAC3D,EAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,GACpC,MAAO,GAAQ,SAAS,EAAI,IAGxB,EAAyB,CAAC,OAAQ,EAAI,QAAS,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KAAqB,OAGvD,IAAW,EAAG,CAAC,UAAS,KC5ErC,YA4B8B,EAAW,GACvC,GAAI,GAAc,KAChB,MAAO,GAAE,MAAM,QAEjB,GAAI,GAAiB,EAAE,MAAO,GAC5B,MAAO,GAET,GAAI,EAAE,MAAM,SAAW,EAAW,QAEhC,OADM,GAAyB,GACtB,EAAI,EAAG,EAAI,EAAE,MAAM,OAAQ,IAClC,AAAI,EAAW,IAAM,MAAQ,EAAE,MAAM,IAAM,KACzC,EAAa,KAAK,EAAE,MAAM,IAE1B,EAAa,KAAK,EAAW,IAGjC,MAAO,GAGT,MAAO,GC/CT,AAoDA,YACI,EAAsB,EAAc,EACpC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAUnC,GARA,EACI,EAAG,QAAU,UACb,WAAM,MAAA,8DACF,sBAAqB,EAAG,MAAK,sBACrC,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,qDAAqD,EAAI,MAE/D,IAAS,EACX,MAAO,aAAa,GAAS,EAAG,QAAU,EAG5C,GAAM,GAAc,GAAc,EAAI,GAChC,EAAW,EAAI,EACf,EAAa,GACf,GAAM,GAAI,GAAc,EAAa,EAAG,EAAG,UAAW,GAAO,IAC7D,GAEJ,MAAO,GAAI,EAAI,MAGJ,IAAU,EAAG,CAAC,SAAQ,KC9EnC,YAoBoC,GAElC,MAAO,MAAK,MAAM,KAAK,IAAI,EAAG,KAAK,KAAK,KAAK,IAAI,GAAS,KAAK,IAAI,kBAIjE,EAAsB,EAAW,GAGnC,OAFM,GAAO,EAAI,EAAe,EAC1B,EAAY,GAAI,cAAa,GAC1B,EAAI,EAAG,EAAI,EAAc,EAAE,GAClC,GAAM,GAAU,EAAM,KAAK,GAAK,EAAM,GAAe,EAAO,GAC5D,EAAU,GAAK,EAAI,EAAI,KAAK,IAAI,GAElC,MAAO,IAAS,EAAW,WCjC7B,AAwCA,YACI,EAA2B,EAAuB,GAAA,MAAA,KAAA,QAAA,GAAA,wHAC9C,SAAe,EAAgB,EAAa,cAAe,UAC3D,EAAW,EAAgB,EAAS,UAAW,UAErD,EACI,EAAa,KAAO,EACpB,WAAM,MAAA,+DACF,YAAW,EAAa,QAChC,EACI,EAAa,KAAO,IAAM,EAAS,KACnC,WAAM,MAAA,mFAEC,GAAa,KAAI,qBAAqB,EAAS,QAC1D,GACI,EAAa,MAAM,MAAM,EAAG,EAAa,MAAM,OAAS,GACxD,EAAS,MACT,2FAEE,EAAU,EAAa,MAAM,EAAa,MAAM,OAAS,GAC/D,EACI,EAAI,GAAK,GAAK,EACd,WAAM,MAAA,iEACF,eAAc,EAAO,cAAc,KAEnB,CAAA,EAAM,EAAa,eAArC,SAAkB,EAAA,OACJ,CAAA,EAAM,EAAS,eAOnC,IAPM,EAAc,EAAA,OAId,EAAgB,CAAC,EAAgB,OAAS,EAAS,GAAlD,EAAK,EAAA,GAAE,EAAI,EAAA,GACZ,EAAY,GAAuB,OAAQ,GAExC,EAAI,EAAG,EAAI,EAAO,KAIzB,IAHM,EAAS,EAAI,EACb,EAAO,EAAgB,SAAS,EAAQ,EAAS,GACjD,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAKzC,IAHA,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAErC,EAAU,GAAK,EACN,EAAI,EAAG,EAAI,EAAG,IACrB,GAAI,EAAU,GAAG,QAAU,EAAY,IACrC,EAAU,GAAK,EACf,OAKN,MAAI,KAAgB,GAClB,EAAa,UAEf,AAAI,IAAY,GACd,EAAS,UAIX,CAAA,EAAO,GAAO,EAAW,EAAS,MAAO,iBAG9B,IAAc,GCtG3B,AAiDA,YACI,EAAM,EAAO,EACb,EACA,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE5D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GAC3D,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAY,GACxB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAY,GAAE,OACtD,EACI,IAAa,EAAY,GACzB,WAAM,MAAA,0CAA0C,EAAQ,UACpD,mCAAkC,EAAY,GAAE,QACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EACZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,GAAI,GAClD,EAC0B,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,GAE1E,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAsB,GAG5B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC/G7D,YAgCI,EAAY,EAAW,GACzB,GAAI,GAAc,MAAQ,IAAe,SACvC,MAAO,GAET,GAAI,IAAe,OACjB,MAAO,GAAI,EAAI,GAAK,IAEtB,KAAM,IAAI,OACN,gDAAgD,EAAU,iBAK5D,EAAc,GAChB,GAAI,GAAM,EACJ,EACF,GAAgC,EAAK,MAAO,EAAa,OAC7D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,mBAIvB,EAAW,EACX,GACF,GAAI,IAAe,SACjB,MAAO,GACF,GAAI,IAAe,OACxB,MAAO,IAAK,GACP,GAAI,IAAe,MACxB,MAAO,IAAI,GACN,GAAI,IAAe,QACxB,MAAO,IAAM,GACR,GAAI,IAAe,QACxB,MAAO,IAAM,EAAG,GAElB,KAAM,IAAI,OAAM,4BAA4B,EAAU,KAIjD,GAAM,IAAa,SAAC,EAAuB,GAChD,GAAM,GAAe,EAAgB,EACrC,MAAO,CAAC,GAAgB,IAAe,UC3EzC,AA8FA,YAAmD,MACjD,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAeA,GAFA,EAAa,GAAc,SAEvB,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,6DACC,GAAI,KAAI,OACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,8DACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,6DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,OACf,WAAM,MAAA,sCACF,EAAU,2CAElB,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,GAEnD,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACpB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAGlB,EAAe,GAAqB,EAAI,EAAG,GAEjD,EACI,GAA4B,GAC5B,WAAM,MAAA,oEAEF,uDAAsD,EAAS,OAEvE,GAAM,GACF,GAAoB,EAAI,MAAO,EAAc,EAAS,EAAS,GAC7D,EACF,GAAqB,EAAK,EAAc,EAAQ,MAAO,EAAS,GAC9D,GAAgB,CAAC,EAAM,GAE7B,GAAI,GAAS,MACX,GAAM,IAAU,GAAqB,EAAO,GAC5C,GAAI,KAAK,IAEX,MAAO,KAGH,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,YAAY,CAC9B,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA4B,CAChC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAGpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAS,EAAG,CAAC,aAAY,KC5QtC,AAyBA,YACI,EAAM,EAAO,EACb,GACF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,yBAAyB,EAAK,EAAM,IAE1C,EAAoD,CAAC,EAAG,EAAK,GAAI,GAEvE,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,IAGN,GAAM,IACT,EAAG,CAAC,qCAAoC,KChD5C,AAyBA,YACI,EAAmE,EACnE,EAAkB,GACpB,GAAI,GAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,wBAAwB,EAAM,EAAQ,IAE5C,EAAmD,CAAC,GAAI,GAExD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,IAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IACT,EAAG,CAAC,oCAAmC,KCnD3C,AAyFA,YAA4D,MAC1D,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAaA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,0DACF,iBAAgB,EAAQ,KAAI,OACpC,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,4DACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAClC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,EACI,GAAyC,EAAS,GAClD,WACI,MAAA,oEACA,sBAAqB,EAAO,mBAAmB,EAAS,OAEhE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,0BAAyB,EAAe,gBAAgB,EAAG,OAGrE,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IAEA,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,0BAG/C,GAAM,GAAO,SAAC,EAAc,GAC1B,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAS,OACd,GAAA,GAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAElB,EAAe,GAAqB,EAAI,EAAG,GAE3C,EAAO,GACR,EAAiB,MAAO,EAAc,EAAqB,GAC1D,EAAY,GACd,EAAiB,EAAe,EAAqB,MAAO,GAEhE,GAAI,GAAQ,MACV,GAAM,GAAU,GAAqB,EAAO,GAC5C,MAAO,CAAC,EAAM,EAAW,GAE3B,MAAO,CAAC,EAAM,IAGV,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,qBAAqB,CACvC,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAAqC,CACzC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAEpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAkB,EAAG,CAAC,sBAAqB,KClQxD,AAwDA,YAAwC,SACtC,EAAA,EAAA,EACA,EAAA,EAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAUA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GAAc,EAAG,EAAG,EAAY,GAC7C,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WACI,MAAA,qEACA,iBAAgB,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE9C,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAU,UACrD,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,wCAAwC,EAAW,UAClD,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACnC,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAErC,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAU,EAAM,QAG5D,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACnB,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAAK,GAAA,EAAA,GAAG,GAAA,EAAA,GAId,GACF,GAAqB,EAAQ,EAAI,GAAE,OAAQ,GAAG,GAC9C,GACA,GAgBJ,GAdA,AAAI,CAAC,GAAc,CAAC,EAClB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAM,KACzC,AAAI,CAAC,GAAc,EACxB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,GAAc,EAAK,GAAM,KACzC,AAAI,GAAc,CAAC,EACxB,IAAO,GAAc,EAAK,GAAc,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAO,KAE/C,IAAO,GAAc,EAAK,GAAc,GAAM,IAC9C,GAAO,GAAc,GAAc,EAAK,GAAM,KAG5C,GAAQ,MACV,GAAM,IAAU,GAAqB,GAAO,IAC5C,MAAO,CAAC,GAAM,GAAM,QAEpB,OAAO,CAAC,GAAM,KAIZ,EAA+B,SAAC,GACpC,GAAM,GAAI,EAAQ,iBAAiB,CACjC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA6B,CACjC,EAAG,EACH,EAAG,EACH,KAAM,EACN,uBAAwB,GAEpB,EAA2B,CAAC,WAAU,EAAE,WAAU,EAAE,WAAU,GAIpE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAe,GACxC,GAAM,GAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,IAET,CAAC,MAAO,EAAQ,EAAK,GAAW,SAAU,KAEvD,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAe,EAAe,GAC5C,GAAM,IAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,GAAK,IAEd,CAAC,MAAO,EAAQ,GAAK,GAAW,SAAU,KAGvD,MAAO,GAAiB,EAAK,EAAK,IAI/B,GAAM,IAAS,EAAG,CAAC,aAAY,KClOtC,+DCAA,AAiCA,YAAwB,GACtB,MAAO,IAAa,EAAc,IAAM,KAEnC,GAAM,IAAgB,EAAG,CAAC,eAAc,KCpC/C,AAiCA,YAAqB,GACnB,MAAO,IAAa,EAAc,GAAK,IAGlC,GAAM,IAAa,EAAG,CAAC,YAAW,KCrCzC,AAyCA,YACI,EAAkB,EAAqB,EAAmB,EAC1D,GAD0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,GAGF,OAFI,GAAQ,EACN,EAAmB,GAClB,EAAQ,GAAe,EAAO,MACnC,EAAO,KAAK,GAAM,EAAQ,EAAO,IACjC,GAAS,EAGX,GAAI,EACF,KAAO,EAAQ,EAAO,OACpB,GAAM,GAAU,EAAQ,EAAe,EAAO,KACxC,EAAM,GAAO,CACjB,GAAM,EAAQ,EAAO,EAAc,GAAS,GAAK,CAAC,GAAS,KAE7D,EAAO,KAAK,GACZ,GAAS,EAIb,MAAI,GAAO,SAAW,EACb,GAAS,GAAI,CAAC,EAAG,IAGnB,EAAQ,GAAO,GAAS,CAAC,EAAO,OAAQ,IAE1C,GAAM,IAAQ,EAAG,CAAC,OAAM,KCpE/B,AA4CA,YACI,EAAkB,EAAqB,EACvC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,AAAI,GAAa,MACf,GAAY,GAAoB,IAKlC,OAHM,GAAe,GAAM,EAAQ,EAAa,GAC1C,EAAiB,EAAI,EAAc,EAAS,IAC5C,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAa,MAAM,GAAI,IACzC,EAAO,KACH,GAAK,GAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,EAAG,IAAe,IAE5D,MAAO,IAAO,GAET,GAAM,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAmDA,YACI,EACA,EACA,EACA,EACA,EACA,GAEF,GAAM,GAAS,EAAgB,EAAO,QAAS,iBACzC,EAAS,EAAgB,EAAO,QAAS,gBAAiB,WAC1D,EAAU,EAAgB,EAAQ,SAAU,gBAAiB,SACnE,EAAS,GAAU,WACnB,EAAqB,GAAsB,EAE3C,GAAM,GAAW,EAAO,MAAM,GAE9B,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OACnC,EACI,EAAO,OAAS,GAAK,EAAO,MAAM,KAAO,EACzC,WAAM,MAAA,oDAAoD,EAAQ,OAC9D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAQ,OAAS,GAAK,EAAQ,MAAM,KAAO,EAC3C,WAAM,MAAA,qDAAqD,EAAQ,KAC/D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,iEACF,WAAU,EAAS,OAAM,OACjC,EACI,EAAS,IAAM,GAAK,EAAS,IAAM,EACnC,WAAM,MAAA,2CAA2C,IACrD,EACI,IAAW,YAAc,IAAW,UACpC,WAAM,MAAA,+CAA+C,IAEzD,GAAM,GAAiC,SAAC,GAAY,MAAA,GAAQ,cACxD,EAAQ,EAAQ,EAAS,EAAU,EAAQ,IAEzC,EACoB,CAAC,MAAO,EAAQ,MAAO,EAAQ,OAAQ,GAC3D,EAA4B,CAAC,OAAM,EAAE,mBAAkB,EAAE,SAAQ,GACjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GACJ,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KCtG/C,AAiCA,YAAwB,GACtB,GAAM,GAAS,EAAgB,EAAO,QAAS,gBAAiB,WAEhE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAA8B,CAAC,MAAO,GACtC,EACF,EAAO,UAAU,GAAe,EAAgC,IACpE,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KC/C/C,AA4CA,YACI,EAA4B,EAC5B,EACA,GADA,AAAA,IAAA,QAAA,GAAA,GACA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,mBAAoB,WAEnE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,mDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAAiC,CAAC,MAAO,GACzC,EAA+B,CAAC,QAAO,EAAE,UAAS,EAAE,OAAM,GAC1D,EAAM,EAAO,UACf,GAAkB,EAClB,GACJ,MAAO,GAGF,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC/DrD,AAoBA,YACI,EAAiB,EAAkB,EACnC,EAAsB,EAAwB,GAMhD,AAAI,GAAgB,MAClB,GAAe,IAEjB,AAAI,GAAkB,MACpB,GAAiB,OAAO,mBAE1B,AAAI,GAAgB,MAClB,GAAe,GAGjB,GAAM,GAAW,EAAM,MAAM,GAC7B,SAAgB,KAAK,IAAI,EAAe,GAExC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAClE,EACI,EAAM,OAAS,EACf,WAAM,MAAA,+CAA+C,EAAM,KAAI,MACnE,EACI,EAAM,MAAM,KAAO,EACnB,WACI,MAAA,oDAAoD,EAAM,MAAM,KACxE,EAAY,EAAO,OAAS,EAAG,WAAM,MAAA,+BACrC,EACI,EAAO,MAAM,KAAO,EACpB,WAAM,MAAA,sDAAsD,EAAQ,KAChE,YAAW,EAAO,MAAM,MAChC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAC3D,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GC3DnE,AA0BA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBAC1B,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAExB,GAAM,GAAQ,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,GAC1D,MAAO,GAAO,cACV,SAAA,GAAK,MAAA,GAAE,kBACH,EAAQ,EAAS,EAAe,EAAc,IAClD,CAAC,MAAO,EAAQ,OAAQ,GAAU,KAAiB,GACnD,GAGC,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC/CvD,YA+BI,EAAU,EAAY,GACxB,GAAM,GAAQ,GAAa,EAAK,EAAS,GACnC,EAAiB,EAAQ,EAAI,CAAE,GAAQ,GAAK,EAClD,EAAI,OAAO,EAAgB,EAAG,eAmB5B,EAAU,EAAW,GACvB,MAAO,IAAc,EAAK,EAAQ,GAAc,IAUlD,YAA8B,EAAM,GAClC,MAAO,GAAI,EAAI,EAAI,EAAI,EAAI,GAAK,EAGlC,YACI,EAAU,EAAW,GAKvB,OAJI,GAAO,EACP,EAAQ,EAAI,OACZ,EAAS,EACT,EAAQ,GACL,EAAO,IACZ,EAAS,EAAS,GAAQ,IAAU,GACpC,GAAM,GAAgB,EAAW,EAAQ,EAAI,IAC7C,AAAI,EAAgB,EAClB,EAAO,EAAS,EAEhB,GAAQ,EAGR,EAAQ,CAAC,GAIb,MAAO,GAAQ,EAAO,CAAC,EAAO,ECvFhC,YAoCI,EAAmB,EAAoB,EACvC,EAAsB,GACxB,MAAO,IACI,EAAO,EAAQ,EAAe,EAAc,EAC5C,GACC,4BAIV,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAC5C,EAAsB,GACtB,EAA6C,gBAK/C,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAAgB,EAC5D,IAGN,YACI,EAAmB,EAAoB,EACvC,EAAsB,EAAwB,EAC9C,EAA4B,EAC5B,GADA,AAAA,IAAA,QAAA,GAAA,IAA4B,AAAA,IAAA,QAAA,GAAA,IAC5B,AAAA,IAAA,QAAA,GAAA,IAKF,OAFM,GAAa,GAEV,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,EAAO,GAAK,GACd,EAAW,KAAK,CAAC,MAAO,EAAO,GAAI,SAAU,EAAG,mBAAoB,IAIxE,EAAW,KAAK,IAShB,OALM,GAAQ,EAAe,EAAK,IAAO,EAAgB,EAEnD,EAA4B,GAC5B,EAA2B,GAE1B,EAAgB,OAAS,GAAiB,EAAW,OAAS,IACnE,GAAM,GAAY,EAAW,MACtB,EAAA,EAAA,MAAsB,EAAA,EAAA,SAAU,EAAA,EAAA,mBAEvC,GAAI,EAAgB,EAClB,MAUF,OADI,GAAkB,GACb,EAAI,EAAgB,OAAS,EAAG,GAAK,EAAoB,EAAE,GAClE,GAAM,GAAM,GAAsB,EAAO,EAAU,EAAgB,IAEnE,GAAI,GAAO,GACT,EAAkB,GAClB,MAMF,GAHA,EAAU,MACN,EAAU,MAAQ,GAAe,EAAc,EAAO,GAEtD,EAAU,OAAS,EACrB,MAWJ,EAAU,mBAAqB,EAAgB,OAE/C,AAAK,GAGH,CAAI,EAAU,QAAU,EACtB,GAAgB,KAAK,GACrB,EAAe,KAAK,EAAU,QACzB,AAAI,EAAU,MAAQ,GAG3B,GAAa,EAAY,EAAW,KAM1C,GAAM,GAAe,EAAgB,OAC/B,EAAa,EAAgB,EAEnC,AAAI,GAAsB,EAAa,GACrC,GAAgB,KAAI,MAApB,EAAwB,GAAI,OAAM,GAAY,KAAK,IACnD,EAAe,KAAI,MAAnB,EAAuB,GAAI,OAAM,GAAY,KAAK,KAGpD,GAAM,GACe,CAAC,gBAAiB,GAAS,EAAiB,UAEjE,MAAI,IACF,GAAO,eAAoB,GAAS,EAAgB,YAGtD,AAAI,GACF,GAAO,aAAkB,GAAO,EAAc,UAGzC,EAGT,YAA+B,EAAmB,EAAW,GAC3D,GAAM,GAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAS,GAAQ,GAAU,GAAQ,GACnC,EAAS,GAAQ,GAAU,GAAQ,GACzC,GAAI,GAAS,GAAK,GAAS,EACzB,MAAO,GAET,GAAM,GAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAmB,EAAkB,GACnE,KAAK,IAAI,EAAmB,EAAkB,GAClD,MAAO,GAAoB,GAAQ,EAAQ,GAO7C,YAAwB,EAAsB,EAAe,GAC3D,GAAM,GAAS,KAAK,IAAI,EAAQ,EAAM,GACtC,MAAO,IAAO,EAAe,EAAS,EAGxC,YAA6B,EAAe,GAK1C,MAAQ,GAAG,MAAQ,EAAG,OAChB,EAAG,QAAU,EAAG,OAAW,EAAG,SAAW,EAAG,SC7MpD,AA2CA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,wHACpB,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAED,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,GACxD,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAyB,GC1EtC,AAyDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,GACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEtB,GAAM,GAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EACF,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GAExD,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,eAAgB,EAAO,IAGtD,GAAM,IAA6B,EAAG,CAAC,4BAA2B,KChCzE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,wGACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEC,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,EACpD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAkC,GCxF/C,AAmDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAEzB,EAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EAAkC,CACtC,cAAe,EACf,aAAc,EACd,eAAgB,EAChB,mBAAkB,GAGd,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,aAAc,EAAO,IAGpD,GAAM,IAA0B,EAAG,CAAC,yBAAwB,KCnCnE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,+GACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAG3B,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBADxC,SACF,EAAA,OADG,EAAS,EAAA,GAAE,EAAU,EAAA,GAMtB,EAAM,GACR,EAAW,EAAY,EAAgB,EAAe,EACtD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAA+B,GChF5C,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,2DACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,6DACC,GAAI,OAEf,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAG9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GACZ,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,eACX,EAAa,EAAW,EAAU,IAGlC,EAA+B,CAAC,OAAQ,GACxC,EAA6B,CAAC,aAAY,EAAE,KAAI,GAEhD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCpFjD,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,yBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,kEACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WACI,MAAA,oEACG,GAAI,OACf,EACI,EAAQ,QAAU,WAAa,EAAQ,QAAU,QACjD,WAAM,MAAA,qDAEV,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAE9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GAEZ,EAAsC,CAAC,OAAQ,GAC/C,EAAoC,CAAC,aAAY,EAAE,KAAI,GAEvD,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,sBACX,EAAa,EAAW,EAAU,IAGlC,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAwB,EAAG,CAAC,uBAAsB,KCxF/D,AAyEA,YACI,EAAiB,EAAkB,GACrC,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAClE,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAElE,GAAM,GAAK,EAAgB,EAAG,IAAK,YAEnC,EACI,EAAG,MAAQ,EACX,WAAM,MAAA,4CAA4C,EAAG,KAAI,MAE7D,GAAM,GAAQ,EAAG,MACX,EAAA,EAAA,MAAA,MAAA,IAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,kDAAiD,EAAC,OAExD,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,qDAAoD,EAAC,OAG3D,AAAI,EAAW,GACb,GAAW,GAEb,AAAI,EAAW,GACb,GAAW,GAGb,GAAM,GAAI,EAAQ,GAAM,EAAG,EAAG,EAAG,SAAU,CAAC,GAAI,IAC1C,EAAI,GAAM,EAAG,EAAG,EAAG,SACnB,EAAK,GAAI,EAAG,GAEZ,EAAS,GACX,GAAU,EAAI,GAAO,CAAC,EAAU,UAChC,GAAa,EAAI,GAAO,CAAC,EAAU,WAEjC,EAAO,GAAM,CAAC,EAAG,GAAI,EAAG,OAE9B,MAAO,GACI,GAAM,GAAQ,EAAQ,EAAI,CAAC,GAAI,EAAG,KACvB,IAAI,SAAA,GAAO,MAAA,IAAM,EAAQ,EAAK,MACzC,GAGN,GAAM,IAAW,EAAG,CAAC,UAAS,KC7HrC,AA2DA,YAAsB,GACpB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,EAAkB,GAClB,EACI,GAAM,MAAQ,EAAG,OAAS,EAC1B,WAAM,MAAA,sEAGV,OADM,GAAM,EAAG,GAAG,MAAM,cACf,GACP,EACI,EAAG,GAAG,MAAM,KAAO,EACnB,WACI,MAAA,gEACA,KAAK,EAAkB,GAAG,MAAM,GAAE,QAAQ,EAAG,QAL9C,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,OAQT,GAAkB,GAClB,EAAK,GAAM,EAAI,EAAG,MAAM,GAAI,GAAG,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,CAAC,MAGtD,EACI,EAAG,QAAU,EAAG,GAAG,MAAM,GACzB,WAAM,MAAA,oCACK,EAAkB,OAAM,aAC/B,0BAA0B,EAAkB,GAAG,MAAM,GAAE,QAI/D,OAFM,GAAiB,GACjB,EAAO,aACJ,GACP,EAAG,KAAK,EAAO,KAAK,WAClB,GAAI,GAAI,EAAK,GACb,GAAI,EAAI,EACN,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAO,EAAI,GAAI,EAAI,EAAG,GAAI,IAAK,EAAG,IACxC,EAAI,GAAI,EAAG,GAGf,MAAO,IAAI,EAAG,GAAK,EAAG,kBATjB,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,GAaT,MAAI,GACK,GAAM,EAAI,GAEV,EAIJ,GAAM,IAAc,EAAG,CAAC,aAAY,KC5G3C,AAmFA,YAAa,EAAW,GAMtB,GANsB,AAAA,IAAA,QAAA,GAAA,IACtB,EACI,EAAE,MAAQ,EACV,WAAM,MAAA,gEACF,EAAE,OAEN,EAAE,OAAS,EACb,MAAO,IAAK,EAAe,GAM3B,GAAM,GAAgB,EAAE,MAAM,MAAM,EAAG,EAAE,MAAM,OAAS,GAC7B,OAAO,SAAC,EAAO,GAAS,MAAA,GAAQ,IACrD,EAAO,GACT,EACI,EACA,CACE,EAAe,EAAE,MAAM,EAAE,MAAM,OAAS,GACxC,EAAE,MAAM,EAAE,MAAM,OAAS,KAE/B,GACE,EAAmB,GACnB,EAAmB,GACzB,EAAK,QAAQ,SAAA,GACL,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,KAAK,GACV,EAAK,KAAK,KAEZ,GAAM,GAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OAC9B,EAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OACpC,MAAO,CAAC,EAAG,GAIf,YAAc,EAAa,GAAA,MAAA,KAAA,QAAA,GAAA,IAClB,EAAO,KAAK,WACjB,EACI,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,0CACF,EAAE,MAAM,OAAM,cAYtB,OAVM,GAAI,EAAE,MAAM,GACZ,EAAI,EAAE,MAAM,GAEd,EAAI,GAAI,GACR,EAAI,GAAM,GAER,EAAQ,GAAS,CAAC,CAAC,IAAK,CAAC,EAAG,IAC9B,EAAc,GAAM,GAElB,EAAQ,GAAK,EAAI,EAAI,aAClB,SAGD,EAAQ,EACR,EAAQ,EACR,EAAQ,EACd,EAAA,EAAA,KAAA,wfAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GA8CP,GAAQ,CAAC,EAAO,EAAO,KApDhB,EAAI,EAAG,EAAI,EAAO,EAAE,IAApB,GAuDT,MAAI,CAAC,GAAgB,EAAI,GACvB,GAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,IACzB,EAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,KAGpB,CAAC,EAAG,KAIR,GAAM,IAAK,EAAG,CAAC,IAAG,KCxMzB,AAiBA,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,IAAA,GAAA,MACA,EAAA,EAAA,uBAAA,GAAA,2BAJU,EAAA,WAAA,GAAA,UAAS,KCyBrB,YACI,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GAAM,GAAgB,GAAY,KAAQ,EAAU,EAAI,EAAS,GAEjE,GAAI,IAAc,EAAA,UAAU,KAC1B,MAAO,GAET,GAAI,IAAc,EAAA,UAAU,IAC1B,MAAO,IAAI,GAEb,GAAI,IAAc,EAAA,UAAU,MAC1B,GAAI,GAAY,KACd,MAAO,IAAK,GAEZ,GAAM,GAAkB,EAAQ,KAAO,EAAS,KAC1C,EAAS,GAAI,GAAI,GAAe,GAAI,IAC1C,MAAO,GAAkB,EAAI,GAAI,EAAQ,GAAO,IACnB,EAGjC,GAAI,IAAc,EAAA,UAAU,wBAC1B,GAAI,GAAY,KACd,MAAO,IAAI,GAAI,GAAe,GAAO,EAAQ,OAE7C,GAAM,GAAqB,EAAI,EAAU,GAAK,EAAQ,QAEhD,EACF,GAAK,GAAI,GAAS,EAAoB,GAAO,KAAM,WACvD,MAAO,IAAI,GAAI,GAAe,GAIlC,KAAM,OAAM,sBAAsB,GAE7B,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCnF3D,AA2CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,sBAC5C,EACF,EAAgB,EAAa,cAAe,sBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,uBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,iCAEvC,GAAM,GAAS,GAAI,GAAI,EAAS,IAChC,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAqB,EAAG,CAAC,oBAAmB,KChBzD,YACI,EAAsB,EAA2B,EACjD,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAC5C,EACF,EAAgB,EAAa,cAAe,kBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,mBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,6BAEvC,GAAM,GAAM,GAAO,GACb,EAAS,GAAI,EAAK,GAAI,EAAI,EAAS,GAAe,EAAM,KAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCnBjD,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAU,EAAgB,EAAQ,SAAU,aAC1C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAM,GAAO,GAEnB,EAAU,GAAI,EAAI,GAAO,GAAI,GAAU,GACvC,GAAM,GAAS,GAAK,GAAI,EAAK,EAAI,EAAS,KAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KC9DvC,AAiDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,aAC5C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAc,GAAO,GACrB,EAAQ,GAAI,GAAI,EAAc,IAC9B,EAAY,GAAQ,EAAO,GAC3B,EAAS,GAAI,EAAO,GAEpB,EACF,GAAI,EAAI,GAAO,IAAM,GAAO,IAAa,EAAI,EAAa,IAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KCtEvC,AAgDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,MAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,WAC5C,EAAe,EAAgB,EAAa,cAAe,WAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,YAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,sBAErD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAO,GAEvB,EAAK,GAAI,EAAI,EAAS,GAAI,GAAI,EAAc,MAC5C,EACF,EAAI,GAAI,EAAK,GAAU,GAAI,GAAI,GAAI,EAAK,GAAe,KACrD,EAAS,GAAI,EAAI,GACvB,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAU,EAAG,CAAC,SAAQ,KCrEnC,AA0CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,oBAC5C,EACF,EAAgB,EAAa,cAAe,oBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,qBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,+BAEvC,GAAM,GAAS,GAAkB,EAAS,GAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC3DrD,AAmCA,YACI,EAAsB,GACxB,GAAM,GACF,EAAgB,EAAQ,SAAU,iCAChC,EACF,EAAgB,EAAQ,SAAU,iCACtC,GACI,EAAQ,MAAO,EAAQ,MAAO,4CAsBlC,GAAM,GAAY,GAAK,GACjB,EAAgB,EAAI,EAAS,GAC7B,EAAgB,GAAM,GAAI,GAAI,GAAI,MAExC,MAAO,IAAI,GAAI,EAAW,GAAgB,GAwB5C,YACI,EAAgC,EAChC,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAoB,EACpB,EAAkB,mBAAoB,uBACpC,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KAOvB,GANA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAEjD,GACI,EAAkB,MAAO,EAAQ,MAAO,kCAExC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAO,GAAO,IAEpB,EACI,GAAI,EAAI,EAAmB,GAAI,EAAK,IAChC,EAAI,EAAM,IAEpB,GAAM,GAAS,GAA+B,EAAmB,GAEjE,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxH3D,AA+DA,YACI,EAAW,EAAW,GAKxB,GALwB,AAAA,IAAA,QAAA,GAAA,IACxB,AAAI,IAAQ,IACV,GAAM,EAAO,KAAO,GAGlB,IAAQ,EAAO,KAAO,EACxB,KAAM,OACF,+DACA,wCAAuC,EAAO,KAAI,KAClD,gBAAe,IAGrB,GAAM,GACF,GAAW,SAAC,EAAgB,EAAgB,GAI1C,GAAM,GAAW,GACX,EAAM,GAAU,EAAQ,CAAC,GAAM,GAC/B,EAAY,GAAI,GAAK,EAAQ,WAAY,GAC/C,EAAK,CAAC,EAAQ,IAEd,GAAM,GAAa,GAAI,EAAI,EAAW,IAChC,EAAW,GAAI,EAAY,CAAC,IAE5B,EAAW,SAAC,EAAO,GAChB,GAAA,GAAA,EAAA,GAAQ,EAAA,EAAA,GACT,EAAU,GAAqB,EAAG,MAAO,CAAC,IAChD,MAAO,CACL,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAK,EAAQ,WAAY,GAAI,KACrC,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAI,GAAY,GAAK,EAAQ,eAGzC,MAAO,CAAC,MAAK,EAAE,SAAQ,KAG7B,MAAO,GAAS,EAAQ,GAsB1B,YACI,EAA4B,EAC5B,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GACA,EAAgB,EAAc,eAAgB,uBAC5C,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KASvB,GAPA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GACI,EAAc,MAAO,EAAQ,MAAO,kCAEpC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAa,GAAO,EAAc,MAAM,IAE9C,EACI,GAAI,EAAI,EAAe,GAAI,EAAK,IAC5B,GAAI,EAAsB,IAGpC,GAAM,GAAS,GAA+B,EAAe,GAE7D,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KC3J3D,GAuNM,IAAW,CACf,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,MAAK,IASD,GAAS,CACb,cAAa,GACb,WAAU,GACV,MAAK,GACL,KAAI,IAeA,GAAQ,CACZ,cAAa,GACb,sBAAqB,GACrB,eAAc,GACd,iBAAgB,GAChB,cAAa,GACb,kBAAiB,GACjB,uBAAsB,GACtB,2BAA0B,GAC1B,gCAA+B,GAC/B,wBAAuB,GACvB,6BAA4B,IAOxB,GAAS,CACb,SAAQ,GACR,YAAW,GACX,GAAE,IAaE,GAAS,CACb,mBAAkB,GAClB,oBAAmB,GACnB,eAAc,GACd,UAAS,GACT,UAAS,GACT,QAAO,GACP,iBAAgB,GAChB,oBAAmB,GACnB,oBAAmB,IClSrB,mBAyCwC,GAAA,EAAA,GAAxC,4DAiBE,SAAA,UAAA,SAAA,SAAS,EAAiB,EAAoB,GAApB,AAAA,IAAA,QAAA,GAAA,IAElB,GAAA,GAAA,KAAA,iBAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,GAAI,GAAW,MACb,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,KAAM,OAAQ,EAAM,EAAE,SACrD,KAAK,eAAe,OAEpB,MAAK,eAAe,GAMtB,MAFA,IAAQ,GAEJ,EACK,EAEP,GAAM,UACC,OAOX,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAEd,KAAK,6CAGJ,EAAA,UAAA,oBAAV,WACE,KAAK,YAAc,KAAK,WAAa,GAgBvC,EAAA,UAAA,iBAAA,SAAiB,EAAiB,GAEhC,MAAO,IAAc,EAAG,IAgB1B,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,aAAe,MACtB,GAAQ,KAAK,cAIX,EAAA,UAAA,eAAN,8EACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAErB,CAAA,EAAO,CACL,KAAM,OAEN,OAAQ,GAAO,KAAK,YAAa,gBAI/B,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,gEAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OACN,4DACA,IAAG,KAAK,sBAUE,EAAA,UAAA,kBAAhB,SAAkC,mGAEhC,SAAA,KAAoB,CAAA,EAAM,EAAa,GAAG,OAAO,eAAjD,SAAK,YAAe,EAAA,OAAqC,GACzD,CAAA,EAAO,EAAa,MAAM,UAE9B,GA5HwC,IA8HxC,OAAO,eAAe,GAAW,OAAO,YAAa,CACnD,MAAO,SAAC,GACN,MAAO,GAAS,UAAY,MAAQ,EAAS,kBAAoB,MAC7D,EAAS,gBAAkB,QC1KnC,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAAgC,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,MAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,IAAA,EAChC,EAAA,QAAA,EALN,EAAA,iBAAwC,GACxC,EAAA,mBAA0C,GAOhD,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,cACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,aACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAC3C,EAAoB,EAAK,mBAAmB,GAAG,SAErD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAiB,EAAK,KAC1B,EAAI,GAAO,GAAW,EAAI,EAAK,MAEjC,EACF,EAAI,GAAI,GAAK,GAAI,EAAmB,EAAK,UACjC,GAAK,GAAI,EAAiB,EAAK,WACnC,GAEF,EACF,GAAI,EAAI,EAAmB,EAAK,KAC5B,EAAI,GAAO,GAAU,EAAI,EAAK,MAEtC,EAAgB,OAAO,GACvB,EAAkB,OAAO,GAEzB,GAAM,GAAW,GAAI,EAAI,EAAS,CAAC,EAAK,cAAe,GACvD,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,oBAAsB,MAC7B,IAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,YACzC,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,cAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,iBAAgB,OAAK,KAAK,oBAC/B,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,iBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,IAAO,KAAK,IACZ,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,IAAQ,EAAO,UAjHxD,EAAA,UAAY,WAmHrB,GArHuC,IAsHvC,GAAc,ICrJd,mBA+BsC,GAAA,EAAA,GAMpC,WACc,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,IAD5C,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,aAAA,EAA8B,EAAA,wBAAA,EAHpC,EAAA,iBAAwC,KAOhD,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAM,GAAY,GAClB,EAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,eACrB,SAAU,GACN,WAAM,MAAA,IAAK,EAAM,MAAO,EAAK,yBAClB,SAAS,MAI5B,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAEjD,GAAK,WACH,GAAM,GAAqB,GAAI,EAAiB,GAAO,IACvD,EAAgB,OAAO,GAEvB,GAAM,GAAW,GACb,EAAI,GAAI,EACA,GAAK,GAAI,EAAoB,EAAO,QAAQ,aAChD,CAAC,EAAK,cACV,GACJ,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,kBAAoB,MAC3B,GAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIvC,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,iBAAiB,IAC9D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,iBAAmB,EAAa,IACjC,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,wBAA2B,KAAK,0BAK7B,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,0BA/EzC,EAAA,UAAY,UAiFrB,GAnFsC,IAoFtC,GAAc,ICnHd,mBAkCmC,GAAA,EAAA,GASjC,WACc,EAAgC,EAChC,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MAFvC,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EAL/B,EAAA,uBAA8C,GAC9C,EAAA,wBAA+C,GAMrD,GAAK,WAEH,EAAK,SAAW,GAAO,GAAO,WAC9B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EAAmB,GAAI,EAAG,EAAK,UAErC,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,wBAAwB,IAAM,MACrC,GAAK,wBAAwB,GAAK,CAChC,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAe,EAAK,wBAAwB,GAAG,SAE/C,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QACvD,EACF,GAAI,EAAI,EAAc,EAAK,OACvB,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EAA2B,GAAI,EAAgB,GAC/C,EACF,GAAI,EAAiB,GAEzB,EAAY,OAAO,GACnB,EAAa,OAAO,GAEpB,GAAM,GACF,GAAI,EAAI,GAAI,EACA,GAAI,GAAK,GAA4B,EAAK,UAC9C,CAAC,EAAK,cACV,GACR,EAAM,OAAO,KAGf,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,QAC7C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,SAAS,UAEd,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,yBAA2B,MAClC,GAAQ,KAAK,wBAAwB,IAAI,SAAA,GAAK,MAAA,GAAE,aAI9C,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,yBACrC,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,4GACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACf,GAAK,WACH,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,IACxD,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,MAGpD,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,wBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,UA5IN,EAAA,UAAY,OA8IrB,GAhJmC,IAiJnC,GAAc,ICnLb,mBAiCoC,GAAA,EAAA,GASnC,WACc,EAAgC,EAChC,EAAyB,EACzB,GADyB,AAAA,IAAA,QAAA,GAAA,MACzB,AAAA,IAAA,QAAA,GAAA,GAHd,GAAA,GAIE,EAAA,KAAA,OAAO,KAHK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EACzB,EAAA,MAAA,EANN,EAAA,uBAA8C,GAC9C,EAAA,2BAAkD,GAQxD,GAAK,WACH,EAAK,UAAY,GAAO,GAAG,WAC3B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EACF,GAAI,CAAC,EAAK,aAAc,GAAI,EAAI,EAAK,UAAW,EAAK,OAAQ,IAEjE,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAGxC,AAAI,EAAK,2BAA2B,IAAM,MACxC,GAAK,2BAA2B,GAAK,CACnC,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAIxC,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAkB,EAAK,2BAA2B,GAAG,SAErD,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QAEvD,EAAM,EAAI,EAAiB,EAAK,OAChC,EAAM,GAAI,GAEV,EAAqB,GAAQ,EAAK,GAExC,EAAY,OAAO,GACnB,EAAgB,OAAO,GAEvB,GAAM,GACF,GAAI,EAAI,GAAI,EAAI,GACR,GAAI,EAAgB,GAAI,EAAoB,EAAK,WACrD,GAER,EAAM,OAAO,KAGf,EAAK,UAAU,OAAO,GAAI,EAAK,UAAW,IAC1C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,UAAU,UAEf,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,4BAA8B,MACrC,GAAQ,KAAK,2BAA2B,IAAI,SAAA,GAAK,MAAA,GAAE,aAIjD,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,wDAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OAAM,wDAGlB,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,QAChB,MAAS,KAAK,QAKX,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,QAAY,EAAO,QAvHzB,EAAA,UAAY,SAyHrB,GA3HqC,IA4HrC,GAAc,IC7Jd,mBA6BkC,GAAA,EAAA,GAKhC,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,aAAA,EAEpB,EAAK,gBAAgB,KAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAEF,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAK,WACH,GAAM,GAAW,GAAI,EAAI,EAAK,EAAG,GAAW,GAC5C,EAAM,OAAO,OAGjB,KAAK,uBAMP,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,aAAe,EACpB,AAAI,KAAK,GAAK,MACZ,KAAK,EAAE,UAET,KAAK,EAAI,GAAK,GAAO,CAAC,KAGxB,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,WAGH,EAAA,UAAA,WAAN,qGACU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAA,CAAQ,EAAA,eAGJ,EAAA,UAAA,WAAN,SAAiB,6FACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAC5C,GADA,EAAe,EAAA,OACX,EAAa,SAAW,EAC1B,KAAM,IAAI,OAAM,iEAIpB,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,aAAgB,KAAK,eAIxB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,eA7DjB,EAAA,UAAY,MA+DrB,GAjEkC,IAkElC,GAAc,IC/Fd,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAA8B,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFZ,GAAA,GAGE,EAAA,KAAA,KAAM,IAAa,KAFP,SAAA,aAAA,EAA8B,EAAA,SAAA,EAChC,EAAA,YAAA,EAJJ,EAAA,cAAqC,GAM3C,EAAK,EAAI,GAAO,EAAK,YAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,cAAc,IAAM,MAC3B,GAAM,GAAY,GAClB,EAAK,cAAc,GAAK,CACtB,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,MAInD,GAAM,GAAe,EAAK,cAAc,GAAG,SACrC,EAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAK,WACH,GAAI,GACE,EAAkB,GAAI,EAAI,EAAK,EAAG,GAAe,GACvD,AAAI,EAAK,YACP,EAAW,GACP,EAAI,EAAK,EAAG,GAAI,EAAU,EAAI,EAAiB,EAAK,KAAM,GAE9D,EAAW,GAAI,EAAI,EAAK,EAAG,GAAkB,GAE/C,EAAa,OAAO,GACpB,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,UACP,AAAI,KAAK,eAAiB,MACxB,GAAQ,KAAK,cAAc,IAAI,SAAA,GAAK,MAAA,GAAE,aAS1C,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,SAAW,GAGZ,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,cAAc,IAC3D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,cAAgB,EAAa,IAC9B,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,SAAY,KAAK,SACjB,YAAe,KAAK,cAKjB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,SAAa,EAAO,cA3FlD,EAAA,UAAY,WA6FrB,GA/FuC,IAgGvC,GAAc,IC/Hd,mBAgCsC,GAAA,EAAA,GASpC,WACc,EAAgC,EAChC,EAA0B,EACpC,GAF0C,AAAA,IAAA,QAAA,GAAA,IAChC,AAAA,IAAA,QAAA,GAAA,GAA0B,AAAA,IAAA,QAAA,GAAA,MACpC,AAAA,IAAA,QAAA,GAAA,IAHJ,GAAA,GAIE,EAAA,KAAA,OAAO,KAOP,GAVY,EAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,SAAA,EAA0B,EAAA,QAAA,EANhC,EAAA,uBAA8C,GAC9C,EAAA,mBAA0C,GAC1C,EAAA,qBAA4C,GAQlD,EAAK,SAAW,EAEhB,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,WAE5B,GAAgB,KAClB,KAAM,IAAI,OAAM,+DAIpB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,OACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,qBAAqB,IAAM,MAAQ,EAAK,UAC/C,GAAK,qBAAqB,GAAK,CAC7B,aAAiB,EAAI,MACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAwB,EAAK,uBAAuB,GAAG,SACvD,EAAqB,EAAK,mBAAmB,GAAG,SACtD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEvC,GAAI,EAAK,UACP,GAAM,GAAsB,EAAK,qBAAqB,GAAG,SAEnD,EACF,GAAI,EAAI,EAAqB,EAAK,OAC9B,EAAI,EAAU,EAAI,EAAK,QAEzB,EACF,GAAI,EAAI,EAAU,EAAK,cACnB,GACI,GAAI,EACA,GAAI,GAAO,GAAyB,EAAK,YACnD,EACF,GAAI,EAAI,EAAoB,EAAK,UAAW,GAEhD,EAAsB,OAAO,GAC7B,EAAoB,OAAO,GAC3B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAGb,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EACF,GAAI,EAAI,EAAoB,EAAK,UAC7B,GAAI,EAAI,EAAU,EAAK,cACnB,GAAK,GAAI,EAA0B,EAAK,YAEpD,EAAsB,OAAO,GAC7B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAInB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,sBAAwB,MAAQ,KAAK,UAC5C,GAAQ,KAAK,qBAAqB,IAAI,SAAA,GAAK,MAAA,GAAE,YAE/C,AAAI,KAAK,oBAAsB,MAC7B,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,oBAC7C,AAAI,KAAK,UACP,EAAU,KAAI,MAAd,EAAkB,KAAK,sBAEjB,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EACF,KAAK,SAAW,EAAa,OAAS,EAAI,EAAa,OAAS,EAC9D,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,MAE3C,AAAI,KAAK,UACP,MAAK,qBACD,EAAa,MAAM,EAAgB,EAAG,EAAgB,GACjD,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,iBAK/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,SAAY,KAAK,SACjB,QAAW,KAAK,QAChB,SAAY,KAAK,WAKd,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,SAChD,EAAO,QAAY,EAAO,WA5KzB,EAAA,UAAY,UA8KrB,GAhLsC,IAiLtC,GAAc,ICjNd,AAyBA,GAAA,IAAA,WAAA,cAsCS,SAAA,IAAP,SAAW,GACT,MAAO,IAAI,IAAa,IAkBnB,EAAA,SAAP,SAAgB,EAAsB,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE/C,GAAI,IAAkB,EAAc,EAAU,IAuBhD,EAAA,QAAP,SACI,EAAsB,EAAY,EAAgB,EAClD,GADsB,MAAA,KAAA,QAAA,GAAA,IAAY,AAAA,IAAA,QAAA,GAAA,GAAgB,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,IACK,GAAI,IACP,EAAc,EAAO,EAAU,EAAS,IAevC,EAAA,KAAP,SACI,EAAsB,EAAa,EACnC,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MACnC,AAAA,IAAA,QAAA,GAAA,MACK,GAAI,IAAc,EAAc,EAAO,EAAO,IAehD,EAAA,SAAP,SAAgB,EAAqB,EAAW,GAAhC,MAAA,KAAA,QAAA,GAAA,MAAqB,AAAA,IAAA,QAAA,GAAA,KAAW,AAAA,IAAA,QAAA,GAAA,MAEvC,GAAI,IAAkB,EAAc,EAAK,IAgB3C,EAAA,OAAP,SACI,EAAsB,EAAa,EAAe,EAClD,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MAAe,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,GACK,GAAI,IAAgB,EAAc,EAAO,EAAO,EAAS,IAmB3D,EAAA,QAAP,SAAe,EAAsB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE5B,GAAI,IAAiB,EAAc,IAE9C,KC5LA,GA+Ba,IAAQ,CACnB,IAAK,GAAsB,IAC3B,SAAU,GAAsB,SAChC,SAAU,GAAsB,SAChC,QAAS,GAAsB,QAC/B,QAAS,GAAsB,QAC/B,OAAQ,GAAsB,OAC9B,KAAM,GAAsB,MCtC9B,AAiBA,GAAM,IAA2B,WAC/B,MAAI,OAAO,wBAA0B,YAC5B,sBACE,MAAO,eAAiB,YAC1B,aAEF,SAAC,GAAgB,MAAA,SAa1B,cACE,MAAO,IAAI,SAAc,SAAA,GAAW,MAAA,IAAc,WAAM,MAAA,SCrC1D,YAmBI,EAAiC,EACjC,GACF,GAAM,GACF,EAAc,OAAO,IAAW,SAAW,EAAS,EAAO,IACzD,EACF,EAAe,OAAO,IAAW,SAAW,EAAS,EAAO,IAChE,MAAO,CAAC,EAAS,GCzBnB,YA0BI,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAqB,GACzB,GAAI,EACF,EAAW,EAAS,OAAO,EAAW,MAAM,IAC5C,EAAS,KAAK,EAAW,GAAK,GAC9B,EAAW,EAAS,OAAO,EAAW,MAAM,SAE5C,EAAW,EAAS,OAAO,EAAW,IAEtC,OADM,GAAgB,EAAW,OACxB,EAAI,EAAG,EAAI,EAAe,EAAE,EACnC,EACI,EAAS,OAAO,CAAC,EAAW,EAAI,GAAK,EAAW,GAAI,EAAW,KAErE,EAAW,EAAS,OAAO,EAAW,MAAM,EAAgB,IAE9D,MAAO,eAaL,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAW,GACjB,GAAI,GACF,EAAS,KAAK,GACd,OAAS,GAAI,EAAiB,EAAG,EAAI,EAAc,EAAE,EACnD,AAAI,GAAK,EAAI,EACX,GAAS,KAAK,GACd,EAAS,KAAK,EAAK,GAAiB,KAEpC,EAAS,KAAK,QAMlB,OAFM,GAAsB,GACtB,EAAqB,GAClB,EAAI,EAAG,EAAI,EAAc,EAAE,EAClC,AAAI,GAAK,EAAiB,EAAI,GAAK,EAAI,IAAM,EAC3C,EAAmB,KAAK,GAExB,EAAoB,KAAK,GAG7B,EAAS,KAAI,MAAb,EAAiB,GACjB,EAAS,KAAK,GACd,EAAS,KAAI,MAAb,EAAiB,GAEnB,MAAO,eAaL,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAmB,GAEzB,AAAI,EACF,EAAiB,KAAK,EAAW,GAAK,GAEtC,EAAiB,KAAK,EAAW,GAAK,GAGxC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,GAAK,EAAW,OAClB,AAAI,EACF,EAAiB,KAAK,EAAW,EAAI,GAAK,EAAW,IAErD,EAAiB,KAAK,EAAW,GAAK,EAAW,EAAI,IAGvD,EAAiB,KAAK,EAAW,IAIrC,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,CAAC,GACjB,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAiB,KAAK,EAAM,GAAG,IAEjC,MAAO,eAeL,EAA0B,EAAmB,GAE/C,OADM,GAAY,EAAe,MAAM,EAAG,GACjC,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAU,KAAK,EAAe,EAAI,GAAK,EAAM,GAAG,GAAK,EAAM,GAAG,IAGhE,MAAO,GCvJT,AAiBO,GAAM,IAAkB,mBAClB,GAAa,mBClB1B,AAiBO,GAAM,IAAQ,SACR,GAAS,WACT,GAAS,YACT,GAAS,YACT,GAAS,aACT,GAAS,YCtBtB,qBAmBqB,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACnB,AAAK,KAAM,QAAQ,YACjB,QAAQ,KAAI,MAAZ,QAAgB,wBAIA,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GAClB,AAAK,KAAM,QAAQ,YACjB,QAAQ,IAAG,MAAX,QAAe,GC3BnB,YAiCI,EAAoB,GACtB,GAAI,EAAK,SAAW,EAAK,OACvB,KAAM,IAAI,OACN,gEACG,GAAK,OAAM,WAAW,EAAK,OAAM,MAG1C,OADM,GAAS,GAAI,cAAa,EAAK,OAAS,GACrC,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EACtC,EAAO,GAAK,EAAK,EAAI,GACrB,EAAO,EAAI,GAAK,EAAK,EAAI,GAE3B,MAAO,eAiB8B,GAIrC,OAFM,GAAO,GAAI,cAAa,EAAQ,OAAS,GACzC,EAAO,GAAI,cAAa,EAAQ,OAAS,GACtC,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,EAAI,GAAK,EAAQ,GACtB,EAAK,EAAI,GAAK,EAAQ,EAAI,GAE5B,MAAO,CAAC,KAAI,EAAE,KAAI,eAOiB,GAKnC,OAHM,GAAM,KAAK,KAAK,EAAQ,OAAS,GACjC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAOgB,GAKlC,OAHM,GAAM,KAAK,MAAM,EAAQ,OAAS,GAClC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAShB,EAAuB,GACzB,GAAM,GAAO,EAAQ,EAAQ,GACvB,EAAO,EAAQ,EAAQ,EAAI,GACjC,MAAO,CAAC,KAAI,EAAE,KAAI,eAUhB,EAAkB,EAAc,EAAc,GAChD,EAAK,EAAQ,GAAK,EAClB,EAAK,EAAQ,EAAI,GAAK,cAOpB,EAAW,GAGb,OAFM,GAAO,GAAI,cAAa,EAAI,GAC5B,EAAO,GAAI,cAAa,EAAI,GACzB,EAAI,EAAG,EAAI,KAAK,KAAK,EAAI,GAAI,KACpC,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GAC9C,EAAK,GAAK,KAAK,IAAI,GACnB,EAAK,GAAK,KAAK,IAAI,GAErB,MAAO,CAAC,KAAI,EAAE,KAAI,eAOhB,EAAW,EAAW,GACxB,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GACxC,EAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GACtB,MAAO,CAAC,KAAI,EAAE,KAAI,GCvJpB,YAyDI,EAAM,EAAiB,GACzB,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,GAAE,QAEX,GAAM,GAAc,GAAM,EAAE,OACtB,EAAS,GAAK,EAAG,WACjB,EAAS,EAAQ,QAAQ,EAAQ,GACvC,SAAY,UACZ,EAAO,UACA,EAGT,GAAI,CAAC,GAAgB,EAAE,MAAO,GAG5B,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAE,MAAO,GAExD,GAAI,EAAE,QAAU,aACd,GAAM,GAAO,EAAQ,KAAK,GACpB,EAAS,GAAK,EAAM,GAC1B,SAAK,UACE,EAET,GAAI,IAAU,QACZ,MAAO,GAAQ,IAAI,GACd,GAAI,IAAU,QACnB,GAAM,GAAO,GAAO,EAAG,EAAE,OACnB,EAAS,EAAQ,SAAS,EAAG,GACnC,SAAK,UACE,MAEP,MAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,eAK/D,EAAM,GACR,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAO,EAAE,mBAG3B,EAAe,EAAc,GACxD,GAAM,GAAQ,GAAO,GAAU,GAAM,GAE/B,EAAS,GAAoB,EAAK,WACxC,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,qvCC3G1B,YAyBI,EAAM,EAAsB,GAC9B,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,MAAO,GAAW,IAAI,SAAA,GACpB,GAAM,GAAgB,EAAI,QAC1B,EAAU,GAAQ,EAClB,GAAM,GAAS,GAAM,EAAG,EAAO,GAC/B,SAAM,IAAS,EACR,ICjCX,YA2BI,EAAiC,GAEnC,OADM,GAAqB,GAAI,OAAM,EAAK,MACjC,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAK,MAAM,GAAK,EAAK,GAGrC,OADM,GAAS,GAAO,EAAU,EAAK,OAC5B,EAAI,EAAG,EAAI,EAAO,OAAO,OAAQ,EAAE,GAI1C,OAHM,GAAS,EAAO,WAAW,GAE3B,EAAwB,GAAI,OAAM,EAAK,MACpC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,MAAM,GAG1C,GAAM,GAAgB,EAAK,WAAW,GAEtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,WC7ChB,YAyBI,EAAe,EAAkB,EAAyB,EAC1D,GAOF,OALM,GAAU,EAAO,EAAO,OAAS,GACjC,EAAA,CAAA,EAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAc,GAAuB,EAAQ,EAAQ,GACrD,EAAiB,GAAuB,QAAS,EAAQ,GAEtD,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHM,GAAS,EAAI,EACb,EAAO,EAAE,SAAS,EAAQ,EAAS,GACnC,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAEzC,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAKrC,OAHM,GAAY,EAAI,EAChB,EAAW,EAAY,SAAS,EAAW,EAAY,GACvD,EAAc,EAAe,SAAS,EAAW,EAAY,GAC1D,EAAI,EAAG,EAAI,EAAG,IACrB,EAAS,GAAK,EAAU,GAAG,MAC3B,EAAY,GAAK,EAAU,GAAG,MAKlC,GAAM,GAAc,EAAO,QAC3B,SAAY,EAAY,OAAS,GAAK,EAC/B,CACL,GAAO,EAAa,EAAa,GACjC,GAAO,EAAgB,EAAa,UCxDxC,mJCAA,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,WAAY,SC7BtD,AA4BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAO,GAAK,EAAG,YACnB,EAAI,GAAK,GAAI,GAAO,GAAI,IAC9B,MAAO,IAAI,GAAI,EAAI,QCtC3B,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAK,EAAG,YAAa,IAC/C,MAAO,IAAI,EAAI,OCnCvB,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAGxB,MAAO,CAAC,EAAS,EAAG,KChDxB,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAsC,GAC5C,SAAM,QAAQ,SAAC,EAAG,GAChB,EAAK,GAAK,WAAM,MAAA,GAAG,WAEd,IC7BX,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,mBChChE,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,cAC7C,MAAO,IAAI,EAAI,OCpCvB,AA6BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,EAAI,EAAI,GAAI,EAAG,IACnB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,GAAI,EAAI,EAAI,GAAI,EAAG,KACvB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAExB,MAAO,CAAC,EAAS,EAAG,KCtDxB,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAK,EAAG,YAAa,QC/B7D,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,kBC/B3D,AA2DA,YACI,EAAkB,EAClB,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBAE3C,EAAO,EACP,EAAU,EACV,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,MAIvE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OAEvB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,IAG5C,EAAkC,CAAC,GAAI,EAAM,MAAO,GAEpD,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KChIvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAY,EAAS,EAAY,EACpD,OCpCV,AA8CA,YACI,EAAkB,EAAqB,EACvC,EAAkC,GACpC,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBAE/C,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,gCAC/B,EAAI,KAAI,MAEhB,GAAI,GAAU,EACV,EAAO,EACP,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EACI,EAAQ,EAAQ,CAAC,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KACvE,EAAO,EAAQ,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGhE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,4DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,+DACC,GAAQ,KAAI,OAEvB,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAmB,GAE3D,MAAO,GAAQ,gBAAgB,EAAM,EAAS,IAG1C,EAAgC,CAAC,GAAI,EAAM,MAAO,GAElD,EAA8B,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,GAEvD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAC5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAY,EAAS,OC7BvE,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,WAEnB,MAAI,CAAC,GAAc,CAAC,EACX,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,MAEtB,CAAC,GAAc,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,MAEtB,GAAc,CAAC,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,MAGzB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,KAC7B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,QChDrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,MACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAQtC,OAPM,GACF,EAEE,EAAa,EAAiB,WAC9B,EAAc,EAAiB,MAE/B,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAW,OAAS,EAAG,GAAK,EAAG,IAC1C,GAAI,EAAW,KAAO,EAAY,GAChC,EAAK,GAAK,UACD,EAAW,KAAO,EAC3B,KAAM,IAAI,OAAM,mBACZ,EAAU,6BAA6B,EAAW,MAI1D,OADM,GAAiB,GACd,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,EAAK,GAAK,GACZ,EAAK,KAAK,GAId,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAM,QC/CnC,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,GAAG,YCxBxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAET,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AA0BO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,aAAc,EAAA,EAAA,aACrB,MAAO,CACL,EAAG,WAAM,MAAA,IACL,GAAW,GAAa,EAAG,GAAe,GAAU,EAAG,IACvD,EAAI,GAAU,QCnCxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAS,EAAM,IAAI,SAAA,GAAK,MAAA,GAAE,QACzB,EAAA,EAAA,KACD,EAAQ,GAAe,EAAM,EAAM,GAAG,OAAO,GAC7C,EAAa,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAa,GAAM,EAAI,EAAY,GACzC,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,YAAM,MAAA,QC/BrC,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACN,EAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAEhC,SACI,GAA4B,GAC5B,WAAM,MAAA,8DACF,uDAAsD,EAAS,OAEhE,CACL,EAAG,WACC,MAAA,IAAoB,EAAI,MAAO,EAAI,EAAS,EAAS,EAAK,IAC9D,OAAQ,WACJ,MAAA,IAAqB,EAAK,EAAI,EAAQ,MAAO,EAAS,EAAK,OCxCrE,AAsBO,GAAM,IAA4C,CACvD,WAAY,GACZ,aAAc,CAAC,KAAM,UACrB,SAAU,SAAC,EAAe,EAAiB,GACnC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAAY,EAAA,EAAA,gBAGjC,MAAO,CACL,GAAI,WAAM,MAAA,IACN,EAAK,EAAQ,EAAS,EAAK,EAAY,EACvC,IACJ,OAAQ,WAAM,MAAA,IACV,EAAK,EAAI,EAAO,MAAO,EAAS,EAAK,EAAY,OCpC3D,AA2CA,YACI,EAAM,EAAO,EACb,EAA0C,GAC5C,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEnE,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAEzE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,EACI,EAAI,MAAM,KAAO,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAI,MAAM,GAAE,UAC1D,iCAAgC,EAAY,GAAE,OACtD,EACI,EAAK,MAAM,KAAO,EAAY,GAC9B,WAAM,MAAA,0CAA0C,EAAK,MAAM,GAAE,UACzD,mCAAkC,EAAY,GAAE,QAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAEhD,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,EAAG,GAEjD,EAAmC,CAAC,QAAO,EAAE,IAAG,GAEtD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAwB,GAG9B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC7F7D,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAC3B,EACI,GAAkB,GAClB,WACI,MAAA,kEACA,mDAAkD,EAAS,OAE5D,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAEZ,MAAO,CACL,EAAG,WAAM,MAAA,IACJ,EAAiB,MAAO,EAAI,EAAqB,EAAS,IAC/D,OAAQ,WAAM,MAAA,IACV,EAAiB,EAAK,EAAqB,MAAO,EAAS,OCzCrE,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAI,GAAK,EAAG,aAAc,OC/BvD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UAAW,EAAA,EAAA,QAExB,MAAO,CACL,EAAG,WACD,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAE,MAE7C,EAAM,GAAO,EAAI,EAAM,EAAW,CAAC,GAEvC,MAAI,IAAe,MACjB,GAAM,GAAU,EAAK,IAGhB,MCzCf,AAwBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAG1B,EAAa,GAAoB,CAAC,EAAG,GAE3C,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAU,OAEhB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,EACI,EAAE,OAAS,EACX,WAAM,MAAA,6DACF,yBAAwB,EAAE,KAAI,OACtC,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,8DACF,yBAAwB,EAAO,KAAI,OAC3C,EACI,EAAE,MAAM,KAAO,EAAO,MAAM,GAC5B,WAAM,MAAA,yDACF,cAAa,EAAE,MAAM,GAAE,0CACvB,cAAa,EAAO,MAAM,GAAE,OAEpC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,sCAAqC,EAAO,mBAC5C,KAAI,EAAU,QAEtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WACI,MAAA,gEACA,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAW,GACb,EAAE,MAAO,EAAO,MAAO,EAAS,EAChC,EAAK,EAAiB,IAE1B,MAAO,CACL,EAAG,WACC,MAAA,IAAmC,EAAE,MAAO,EAAI,EAAQ,IAC5D,OAAQ,WACJ,MAAA,IAAoC,EAAG,EAAI,EAAO,MAAO,OC7EnE,AAuBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAA6C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAC3D,EAA+C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAEnE,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,UACH,GAAyB,EACzB,IACb,OAAQ,WAAM,MAAA,GAAO,UACH,GACA,EAAsC,OCtC9D,AA6BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OACnD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAgC,EAAE,MAAO,GAC5D,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GAC/C,MAAO,GAAQ,OAAO,EAAI,IAGtB,EAAwB,CAAC,GAAE,EAAE,EAAC,GAEpC,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,cACZ,EAAoB,EAAgC,KACpD,QCrCV,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAI,EAAI,GAAI,GAAI,GAAO,KAAM,EAAI,KAAK,KAAK,KAAK,KACtD,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC/B7B,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC3B7B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAI,QC5BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AA6BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAiB,EAAE,MAAO,GAC7C,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AA8BO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,IAAK,OAAQ,WAAY,SACxC,SAAU,SACN,EAAY,EAAiB,GACxB,GAAA,GAAA,EAAA,gBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GAAM,EAAA,EAAA,GAAU,EAAA,EAAA,GAEpB,EAAa,GAAgB,GAAO,GACpC,EAAgB,GAAiB,EAAK,MAAO,EAAE,OAC/C,EAAsB,GAC5B,GAAI,EAAK,OAAS,GAChB,OAAS,GAAI,EAAG,EAAI,EAAE,MAAM,OAAS,EAAG,EAAE,EACxC,EAAU,KAAK,EAAE,MAAM,IAEzB,EAAU,KAAK,GAGjB,GAAM,GAAa,GAAI,EAAG,GACpB,EAAoB,EAAI,EAAI,GAC5B,EAAsB,GAAM,GAAI,EAAU,GAAO,KACjD,EAAiB,EACnB,EAAI,EAAI,EAAqB,GAAsB,GACnD,GAAO,MAEL,EAAO,WACX,MAAI,GAAK,OAAS,EACT,EACH,EAAI,EAAI,EACA,GACI,EAAQ,EAAqB,CAAC,EAAG,EAAG,EAAG,EAAK,MAAM,KAClD,IACR,GACJ,EAAE,OAEC,EAAQ,EAAI,EAAI,EAAI,GAAsB,GAAa,EAAE,QAG9D,EAAU,WACd,GAAI,GACA,EAAI,EAAI,EAAqB,GAAO,KAAM,GAC9C,MAAI,GAAK,OAAS,GAChB,GAAU,GAAI,EAAS,IAElB,EAAQ,EAAS,EAAK,QAEzB,EAAc,WAClB,GAAI,GAAc,EAAI,EAAI,EAAgB,GAAa,GAEvD,MAAI,GAAK,OAAS,GAChB,GAAc,GAAI,EAAa,IAE1B,EAAQ,EAAa,EAAK,QAE7B,EAAW,WACf,GAAM,GAAwB,EAAI,EAAY,GAE1C,EAAW,EAAI,EAAI,GACvB,MAAI,GAAK,OAAS,GAChB,GAAW,GAAI,EAAU,IAEpB,EAAQ,EAAU,EAAK,QAE1B,EAAY,WAChB,GAAI,GAAY,EAChB,MAAI,GAAK,OAAS,GAChB,GAAY,GAAI,EAAW,IAEtB,EAAQ,EAAW,EAAK,QAGjC,MAAO,CACL,EAAG,EACH,KAAM,EACN,SAAU,EACV,MAAO,EACP,OAAQ,KC1Gd,AA0BO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,WACpB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACH,EAAA,EAAA,KAED,EAAa,GAAe,EAAM,EAAE,OAAO,GAE3C,EAAO,WACX,GAAM,GAAc,EAAE,MAChB,EAAc,EAAQ,KAEtB,EAAa,EAAY,MAAM,EAAG,GAClC,EAAY,EAAW,OACvB,EAAa,EAAY,MAAM,EAAM,EAAY,QAAQ,MAAM,GAC/D,EAAY,EAAW,OAEvB,EAAmB,GAAW,EAAG,GACjC,EACF,GAAW,EAAY,EAAG,EAAY,EAAI,GAExC,EAAc,GAAY,CAAC,EAAY,CAAC,GAAc,IAEtD,EAAS,EAAQ,EAAI,GACrB,EAAkB,EAAQ,EAAS,CAAC,IAEpC,EACF,GAAY,CAAC,CAAC,GAAY,EAAkB,IAC1C,EAAkB,GAAU,EAAQ,GACtC,EAAa,GACb,EAAiB,EAA6B,EAAE,MAAM,IAEpD,EAAsB,GAAuB,GACnD,SAAa,GAAU,EAAY,GAE5B,GAET,MAAO,CAAC,EAAG,EAAM,QAAS,WAAM,MAAA,OAIpC,YAAoB,EAAe,GAEjC,OADM,GAAS,GACN,EAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,EAAO,KAAK,GAEd,MAAO,GAGT,YAAqB,GAEnB,OADM,GAAS,GACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,EACtC,EAAO,KAAK,EAAO,GAAG,IAG1B,MAAO,GCnFT,AAqBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,IAAI,EAAG,WAAM,MAAA,IAAU,OC1BtD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAK,EAAI,eCzB9B,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC5B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,EAAG,QC5BpC,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,EAAG,gBC5BrC,AAyBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,GACd,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KACP,MAAO,CACL,OAAQ,WACN,GAAM,GAAW,GACX,EAAU,GAAI,GACpB,MAAO,IAAI,EAAI,EAAI,GAAI,EAAI,EAAM,GAAW,QCpCpD,AAyBA,YACI,EAAM,EAAM,EAAO,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,QAAQ,EAAI,EAAG,EAAG,EAAa,EAAM,EAAO,IAElD,EAA4B,CAAC,EAAC,EAAE,EAAC,EAAE,GAAE,GAErC,EAA0B,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAE/D,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAa,GAGnB,GAAM,IACT,EAAG,CAAC,oCAAmC,KCxC3C,AAqBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,YAAa,EAAA,EAAA,KAAM,EAAA,EAAA,MAAO,EAAA,EAAA,KAEjC,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAG,EAAG,EAAI,EAAa,EAAM,EAAO,OC/B9C,YA6BI,EAAO,EAAM,EAAe,EAAoB,GAClD,MAAI,GAAE,KAAO,EAAM,MACjB,GAAI,EAAQ,EAAG,GAA+B,EAAE,MAAO,KAEzD,AAAI,EAAG,KAAO,EAAM,MAClB,GAAK,EAAQ,EAAI,GAA+B,EAAG,MAAO,KAErD,CACL,EAAG,WACD,GAAM,GAAK,EAAI,EAAI,GAAK,GAAM,EAAO,GAAI,EAAG,QAC5C,MAAO,IAAgB,KAAO,EAAK,GAAU,EAAI,KCvCvD,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,iBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAkB,EAAE,OACnD,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAa,EAAG,GAAI,aAC9C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,GAAI,aAC5C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA6DA,YACI,EAAkB,EAAqB,EACvC,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE9C,EAAO,EACP,EAAU,EACV,EAAW,EACX,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KAErE,EAAW,EAAQ,EAAS,CAC1B,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAI3E,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAS,OAAS,EAClB,WAAM,MAAA,kEACC,GAAS,KAAI,OAExB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OACrE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,EAAU,IAGtD,EACwB,CAAC,GAAI,EAAM,MAAO,EAAS,OAAQ,GAE3D,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC1IvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAG,EAAY,EAAS,EAAY,EACvD,OCrCV,AAkDA,YACI,EAAyB,EACzB,EAA6B,EAC7B,EAAkC,EAClC,GACF,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBACzC,EAAU,EAAgB,EAAQ,SAAU,mBAElD,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,+BAC/B,KAAI,EAAI,KAAI,OAEpB,EACI,EAAI,OAAS,EACb,WAAM,MAAA,4DACC,GAAI,KAAI,OACnB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,+DACC,GAAO,KAAI,OACtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAO,MAAO,EAAY,EAAS,EAAmB,EACtD,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAQ,EAAS,IAGjD,EACsB,CAAC,GAAI,EAAK,MAAO,EAAQ,OAAQ,GAEvD,EACqB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAErE,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,GAC/C,GAGN,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAE5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAG,EAAY,EAAS,OC/B1E,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,KACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAM,EAAE,OACvC,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAU,EAAG,GAAI,aAC3C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAQ,EAAG,GAAI,aAC/C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA4BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAI,GAAa,EAAE,OAEjC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAI,GAAM,GAAI,EAAG,MAC/B,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KClDxB,AA0BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,OCzBzB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,WACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAU,EAAM,GACtB,MAAO,CAAC,QAAS,WAAM,MAAA,IAAM,EAAQ,MAAO,eC3BhD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAGtC,GAAM,GAAI,EAAM,GACT,EAAA,EAAA,SACD,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAM,EAAI,EAAO,EAAE,WC/BxC,AAgCO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GACP,EAAO,EACP,EAAM,EACN,EACF,GAA0C,EAAK,MAAO,EAAI,OAExD,EAAU,WACd,GAAM,GAAW,GAAK,EAAK,WACvB,EAAM,EAAI,EAAI,EAAI,EAAU,GAAI,EAAM,GAAI,EAAU,GAAO,OACzD,EAAa,GAAgC,EAAK,MAAO,GAC/D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,QAErB,EAAS,WACb,GAAM,GAAY,GAAQ,EAAM,GAC1B,EAAU,GAAM,EAAW,GAAI,GAAO,GAAU,IAClD,EAAM,EAAI,EAAI,EAAI,EAAG,IACnB,EAAa,GAAgC,EAAI,MAAO,GAC9D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAI,QAE1B,MAAO,CAAC,EAAG,EAAS,EAAG,KC9D3B,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,SACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,GAAQ,EAAG,GAExB,MAAO,CACL,EAAG,WAAM,MAAA,IAAM,EAAM,EAAI,EAAI,EAAI,KACjC,MAAO,WACL,GAAI,GAAM,GAAM,EAAM,GAAU,GAAK,EAAI,EAAI,IACvC,EAAa,GAAiB,EAAM,MAAO,EAAG,OACpD,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAM,WC1ClC,AAwBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,SC7BxC,AAwBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAO,EAAI,GAAU,EAAG,GAAI,GAAK,IAEvC,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAM,gBC/BxC,AAuBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,GAAI,gBC5B3C,AAqBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAQ,EAAI,EAAE,WC1BnC,AAuBO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,uBACX,EAAI,EAAoB,IAGxB,EAAmC,CAAC,OAAM,GAC1C,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAAoB,IAExB,MAAO,CAAC,OAAQ,KCxCpB,AAuBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,8BACX,EAAI,EAAoB,IAGxB,EAA0C,CAAC,OAAM,GACjD,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAA2B,IAE/B,MAAO,CAAC,OAAQ,KCxCpB,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,KACD,EAAO,GAAe,EAAM,EAAG,OACrC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAQ,EAAI,OC3BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAyBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,GAAI,EAAI,EAAI,GAAI,EAAG,KAAM,SC9BlD,AAyBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,aACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAGL,UAAW,WAAM,MAAA,IAAK,GAAU,GAAY,YAC5C,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAW,EAAG,SACpC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAW,GAAY,EAAG,YCnCtD,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CACL,EAAG,WACD,GAAM,GAAO,GAAQ,EAAG,GAAO,IAEzB,EAAa,GAAO,IACpB,EAAQ,GAAO,IAEf,EAAqB,EAAI,EAAI,GAC7B,EACF,EAAI,EAAI,EAAI,GAAa,GAAI,GAAK,EAAG,aAEzC,MAAO,IAAM,EAAM,EAAoB,OC3C/C,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,EAAG,GAAI,GAAO,GAAI,SC9BnD,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAK,EAAG,YAAa,OC7BlD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GActC,OAbO,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,KAER,EAAa,EAAE,MACf,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAQT,EAAoC,GACjC,EAAI,EAAG,EAAI,EAAG,KAAM,IAC3B,EAAS,KAAK,CAAC,EAAO,GAAI,EAAW,GAAK,EAAO,GAAK,EAAM,KAE9D,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,OC3C7B,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,IACD,EAAW,GAEX,EAAW,EAAI,EAAI,GACzB,MAAO,CACL,OAAQ,WAAM,MAAA,IAAI,EAAU,EAAI,GAAI,EAAU,CAAC,GAAM,GAAW,QClCtE,AAuBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAQ,QC7BrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,SACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,KAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAO,EAAI,OC1BhC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAI,GAAK,GAAK,EAAG,YAAa,QC/B3D,AAuBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,GAAK,EAAG,WAAY,QC5BrD,AAwBO,GAAM,IAA0C,CACrD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAM,GAAO,GACb,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MACrC,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MAC3C,MAAO,CAAC,EAAS,EAAG,KChCxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,GAAI,GAAM,EAAE,QAG7B,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAkB,EAAE,MAAM,QACzB,EAAA,EAAA,KAED,EAAO,GAAe,EAAM,EAAE,OACpC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,EAAI,EAAY,GAAK,EAAE,MAAO,YAE3C,MAAO,CAAC,EAAG,WAAM,MAAA,OCxCrB,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAO,GAAI,SC9BxC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAO,GAAI,GAAO,IAAK,OC/BpD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KAED,EAAO,WACX,GAAI,GAAQ,GAAU,GAGtB,GAAI,EAAE,OAAS,EACb,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,IAAK,CAAC,EAAE,MAAM,cAEjD,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IAAK,CACjD,EAAE,MAAM,GAAI,EAAE,MAAM,cAI7B,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IACjD,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,cAI1C,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EACA,CACE,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAC5C,EAAI,EAAE,MAAM,IAEd,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,UAMjE,MAAM,IAAI,OACN,2DACG,GAAE,KAAI,kBAEf,MAAO,IAET,MAAO,CAAC,EAAG,KCpFf,AAuBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAiC,EAChC,EAAA,EAAA,KACD,EAAW,GAAiC,GAClD,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,EAAI,OC7BnC,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACxC,GAAM,GAA2B,EAC1B,EAAA,EAAA,KACP,MAAO,CAAC,MAAO,WAAM,MAAA,IAAM,EAAI,OC1BnC,AA8BO,GAAM,IAA2C,CACtD,WAAY,GACZ,aAAc,CAAC,cACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAAO,WACX,MAAO,IAAoB,EAAI,IAEjC,MAAO,CAAC,EAAG,KAIf,YAA+C,EAAM,GAQnD,OAJM,GAAqB,GAAQ,EAAS,GAAU,IAChD,EAAW,GAAO,EAAG,GACvB,EAAa,GAAa,EAAS,GAAO,EAAG,UAC3C,EAAW,EAAS,KAAO,EAAW,KACnC,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,GAAW,EAAY,EAAI,GAE1C,EAAa,GAAW,EAAY,GAAK,EAAS,MAAO,SACzD,GAAM,GAAY,GAAU,GAC5B,MAAO,IAAM,EAAY,EAAU,GCxDrC,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAkHA,GAAM,IAA4B,CAChC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA6B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAxB,GAAM,IAAc,GAAA,IACvB,GAAiB,ICxNnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA+BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACL,EAAO,KAAK,OAAS,EAAG,WAAM,MAAA,wCACvB,EAAQ,KAAM,KClCvB,AAmCA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAQ,KAAM,ICtCvB,AA8BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AAkCA,EAAO,UAAU,KAAO,SACpB,EAAc,GAChB,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,KCrC9B,AAoCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,GACjC,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,KCvCvC,AAqCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,GAChD,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,KCxC9C,AAwCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,EAC9C,GACF,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,EAAQ,KC5CtD,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA+BA,EAAO,UAAU,UAAY,SACzB,EACA,EACA,EACA,EACA,GACF,YAAK,kBACE,GAAU,KAAM,EAAM,EAAU,EAAQ,EAAO,ICtCxD,AA2BA,EAAO,UAAU,YAAc,SAAyB,GAEtD,YAAK,kBACE,GAAY,KAAM,IC9B3B,AA4BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,YAAc,SAC3B,EAAa,GACf,YAAK,kBACE,GAAY,KAAM,EAAK,IC/BhC,AA0BA,EAAO,UAAU,OAAS,SACtB,EAA0B,GAC5B,YAAK,kBACL,AAAI,YAAa,IACf,GAAI,CAAC,IAEA,GAAM,CAAE,MAAI,OAAK,GAAI,IChC9B,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4C,EAC5C,EAAmB,GACrB,YAAK,kBACE,GACI,KAAM,EAAQ,EAAQ,EAAK,EAAY,EACvC,ICrCb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EACA,EAAkC,EAClC,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAa,EAAS,EAAK,ICrCtD,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAqB,GACtC,YAAK,kBACE,GAAO,KAAM,EAAM,EAAW,IC9BvC,AA2BA,EAAO,UAAU,aAAe,SAC5B,EAAmB,GACrB,YAAK,kBACE,GAAa,KAAM,EAAW,IC9BvC,AAkCA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,UAAgB,8DAChB,KAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,IC3Cb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA6BA,EAAO,UAAU,WAAa,SAC1B,EAA+B,EAC/B,EAAqB,EACrB,GACF,YAAK,kBACE,GAAW,KAAM,EAAQ,EAAS,EAAK,EAAW,IClC3D,AA2BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC9BxB,AA4BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC/BzB,AA2BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC7BnB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC5Bb,AA+BA,EAAO,UAAU,YAAc,SAClB,GACX,YAAK,kBACE,GAAY,KAAM,IClC3B,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA0BA,EAAO,UAAU,WAAa,SAA2B,GACvD,YAAK,kBACE,GAAW,KAAM,IC5B1B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA4BA,EAAO,UAAU,OAAS,SACb,EAA4B,GACvC,YAAK,kBACE,GAAO,KAAM,EAAS,IC/B/B,AA+BA,EAAO,UAAU,mBAAqB,SACzB,GACX,YAAK,kBACE,GAAmB,KAAM,IClClC,AA0BA,EAAO,UAAU,aAAe,SAA2B,GAEzD,YAAK,kBACE,GAAa,KAAM,IC7B5B,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA+BA,EAAO,UAAU,gBAAkB,SACtB,GACX,YAAK,kBACE,GAAgB,KAAM,IClC/B,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA0BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC5BpB,AA2BA,EAAO,UAAU,2BAA6B,SAC1C,EAAsB,EAAe,EAAgB,GACvD,YAAK,kBACE,GAA2B,KAAM,EAAa,EAAM,EAAO,IC9BpE,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA2BA,EAAO,UAAU,UAAY,SAChB,EAAwB,GACnC,YAAK,kBACE,GAAU,KAAM,EAAM,IC9B/B,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA0BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC5BpB,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA2BA,EAAO,UAAU,OAAS,SACb,EAAiB,EAAsB,GAClD,YAAK,kBACE,GAAO,KAAM,EAAG,EAAY,IC9BrC,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,KAAO,SACpB,EAAwB,GAC1B,YAAK,kBACE,GAAK,KAAM,EAAM,IC7B1B,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,EAAI,KAAM,IC5BnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,KAAO,SACpB,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAK,EAAM,IClC/B,AA+BA,EAAO,UAAU,eAAiB,SACrB,GACX,YAAK,kBACE,GAAe,KAAM,IClC9B,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAa,GAAb,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAC9B,KAAK,kBACE,GAAO,KAAM,EAAO,EAAS,IC9BtC,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,IAAM,SACV,EAAmC,GAC9C,YAAK,kBACE,GAAI,KAAM,EAAU,IC/B7B,AA6BA,EAAO,UAAU,KAAO,SACX,EAAsC,EAC/C,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAa,EAAa,EAAS,EAAc,IClCrE,AA+BA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,GAAU,KAAM,ICjCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,MAAQ,SACZ,GACX,YAAK,kBACE,GAAM,KAAM,IC7BrB,AA2BA,EAAO,UAAU,KAAO,SACX,EAAwB,GACnC,YAAK,kBACE,GAAK,KAAM,EAAM,IC9B1B,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA0BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC5Bf,AAkCA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,EAAQ,KAAM,EAAE,QCpCzB,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,EAAQ,KAAM,IC5BvB,AA2BA,EAAO,UAAU,eAAiB,SACrB,EAA8B,GACzC,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,sBAAwB,SAC5B,EAA8B,GACzC,YAAK,kBACE,GAAsB,KAAM,EAAY,IC9BjD,AA0BA,EAAO,UAAU,QAAU,SACd,GACX,YAAK,kBACE,GAAQ,KAAM,IC7BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EAAsC,EACtC,EAAqB,EACrB,GACF,YAAK,kBACE,GACI,KAAM,EAAiB,EAAiB,EAAS,EAAK,EACtD,ICtCb,AA4BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,OC9BjB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA6BA,EAAO,UAAU,MAAQ,SACZ,EAAwB,GACnC,YAAK,kBACE,GAAM,KAAM,EAAO,IChC5B,AA4BA,EAAO,UAAU,QAAU,SAAoC,GAC7D,YAAK,kBACE,GAAQ,KAAM,IC9BvB,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,MAAQ,SACrB,EAAkC,GACpC,YAAK,kBACE,GAAM,KAAM,EAAiB,IC9BtC,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAO,OC9BhB,AA2BA,EAAO,UAAU,kBAAoB,SAA2B,GAE9D,YAAK,kBACE,GAAkB,KAAM,IC9BjC,AA+BA,EAAO,UAAU,wBAA0B,SAC9B,GACX,YAAK,kBACE,GAAwB,KAAM,IClCvC,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,MAAQ,SACrB,EAAoB,GACtB,KAAK,kBACL,GAAM,GAAqB,YAAa,GAAS,CAAC,KAAM,GAAE,CAAI,MAAI,OAAK,GACvE,MAAO,IAAM,EAAoB,IC9BnC,AA4BA,EAAO,UAAU,KAAO,SAAoC,GAC1D,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA+BA,EAAO,UAAU,aAAe,SACd,EAAiB,EAAe,EAC9C,EAAoB,EAAkB,EACtC,EAAsB,GACxB,YAAK,kBACE,GACI,KAAM,EAAO,EAAK,EAAS,EAAW,EAAS,EAC/C,EAAa,ICtC1B,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC7BzB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC7BpB,AAgCA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAQ,KAAM,SClCvB,AAgCA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,KAAM,YClCvB,AAgCA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAQ,KAAM,UClCvB,AA4BA,EAAO,UAAU,KAAO,SACX,EAAY,GACvB,YAAK,kBACE,GAAK,KAAM,EAAG,IC/BvB,AA2BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC9BzB,AA2BA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAO,KAAM,IC9BtB,AA4BA,EAAO,UAAU,mBAAqB,SACzB,EAAmC,GAC9C,YAAK,kBACE,GAAmB,KAAM,EAAY,IC/B9C,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,MAAQ,SACrB,EAA8B,GAChC,YAAK,kBACE,GAAM,EAAW,KAAM,IC/BhC,AA4BA,EAAO,UAAU,UAAY,WAC3B,YAAK,kBACE,GAAU,4vQC9BnB,AAaA,GAAI,kBAMF,MAAI,KAAY,MACd,IAAW,EAAA,UAAU,WAEhB,iBAeP,MAAO,eCrCT,AAuBA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAChC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAa,aAE7C,MAAA,IANkC,OAWlC,EAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OAWhC,GAAA,SAAA,GAAyC,EAAA,EAAA,GACvC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAoB,aAEpD,MAAA,IANyC,OAWzC,GAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OC9EhC,YAuB6B,EAAY,GACvC,GAAI,MAAM,QAAQ,IAGhB,OADI,GAAkB,GACb,EAAI,EAAG,EAAI,EAAW,IAC7B,EAAW,EAAS,OAAO,GAE7B,MAAO,QAEP,GAAM,GAAW,GAAI,OAAM,GAC3B,SAAS,KAAK,GACP,eAIY,EAAc,GACnC,GAAI,CAAC,EACH,KAAM,IAAI,IAAe,eAOJ,EAAY,GAEnC,OADI,GAAU,EACK,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,IAAS,GACX,IAGJ,MAAO,eAQ2B,GAClC,MAAI,GAAG,SAAW,EACT,EAAG,GAEL,cAYc,GACrB,MAAI,OAAM,QAAQ,GACT,EAEF,CAAC,eA0BkB,GAC1B,GAAM,GAAe,EAAK,QAAQ,uBAAwB,SACpD,EACF,EAAa,QAAQ,kBAAmB,SAAS,cAKrD,MAAI,GAAS,KAAO,IACX,EAEF,UAAY,cAGO,GAE1B,MAAI,GAAW,QAAU,GAIrB,EAAW,QAAQ,OAAS,GAHvB,EAMF,EAAW,QAAQ,cAAe,SAAC,EAAG,GAAO,MAAA,GAAG,gBAIzD,GAAI,IAAyB,eAEQ,GAEnC,GAAI,GAAa,KACf,MAAO,MAET,GAAM,GAAsC,GAC5C,SAAK,UAAe,EAAS,eAC7B,EAAK,OAAY,EAAS,YACnB,EAcT,YAAuC,GAErC,GAAI,GAAU,MAAQ,MAAO,IAAW,SACtC,OACK,GAAI,MAAM,QAAQ,GACvB,EAAO,QAAQ,SAAA,GAAc,MAAA,IAA8B,SAG3D,QADM,GAAS,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAO,GACrB,AAAI,GAAS,MAAQ,MAAO,IAAU,UACpC,CAAI,CAAC,MAAM,QAAQ,IAAU,EAAM,OAAY,WAC3C,MAAO,GAAM,OAAa,SAC5B,EAAO,GAAS,EAAM,MAEtB,GAA8B,iBAqBpC,EACA,EACA,EACA,EAAgC,aAElC,GAJE,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,UAAgC,AAAA,IAAA,QAAA,GAAA,IAE9B,MAAO,IAAe,UACxB,GAAM,GAAe,EACjB,EAAE,OACN,GAAI,IAAgB,GAClB,EAAK,EAAc,WACV,IAAgB,IACzB,EAAK,GAAuB,WAE5B,EAAK,EAAc,GACf,GAAM,KACR,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAU;EAE7C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAKR,MAAO,QAGP,GAAM,GAAS,EACf,GAAI,EAAO,WAAgB,MAAQ,EAAO,QAAa,KACrD,KAAM,IAAI,GACH,EAAmB,6BACnB,MAAK,UAAU,GAAO;GACzB,sCAEN,GAAM,GAAY,EAAO,UACrB,EAAG,OAAE,EAAU,OAQnB,GAPA,AAAI,IAAa,GACf,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,GAAA,UAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IAEJ,GAAO,KACT,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAS;EAE5C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAIN,GAAI,GAAc,MAQhB,OADM,GAAwB,GACZ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,GAAuB,GAEtD,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,EAAc,GAG7C,GAAM,GAAe,EAAO,OAC5B,EAAa,cAAmB,EAGhC,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAE9C,GAA8B,EAAO,QACrC,GAAM,GACF,EAAW,EAAK,EAAO,OAAW,EAAe,GACrD,UAAsB,GAAA,GAAO,GAEtB,OAMP,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAK9C,GAAM,GAAY,GAAI,GAAI,EAAO,QACjC,UAAsB,GAAA,GAAO,GACtB,gBAUiB,EAAW,GACvC,MAAQ,GAAI,EAAK,GAAO,EAAI,EAAK,EAAI,cAQF,EAAW,GAC9C,MAAO,GAAK,GAAc,EAAG,eA2CL,GACxB,GAAI,GAAM,KACR,MAAO,GAIT,OAFM,GAAW,GAED,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAI,QAAQ,KAAO,IACrB,EAAI,KAAK,GAGb,MAAO,eASqB,GAC5B,GAAI,GAAO,KACT,KAAM,IAAI,GAAW,yBAAyB,KAAK,UAAU,IAE/D,OAAW,KAAO,GAChB,GAAI,EAAI,eAAe,GACrB,MAAO,GAGX,MAAO,eAWL,EAAkB,EAAe,GACnC,GAAI,GAAS,KACX,OAEF,GAAI,EAAO,QAAQ,GAAS,EAC1B,KAAM,IAAI,GAAc,EAAK,mBAAmB,EAAK,uBACjD,EAAM,mCAoBV,EAAQ,EAAsB,EAC9B,GAD8B,MAAA,KAAA,QAAA,GAAA,GAC9B,AAAA,IAAA,QAAA,GAAA,UACF,GAAO,GAAa,GACpB,GAAO,GAAa,GAEhB,MAAM,QAAQ,IAAM,EAAE,QAAU,GAAa,EAAE,QAAU,GACzD,EAAE,MAAM,SAAA,GAAK,MAAA,OAAO,KAAM,gBAWM,EAAwB,GAC5D,AAAI,MAAM,QAAQ,GAChB,GAAA,KAAK,OACD,EAAM,OAAS,EAAG,WAAM,MAAG,GAAI,qCACnC,EAAM,QACF,SAAC,EAAG,GAAM,MAAA,IAAsB,EAAG,WAAW,GAAI,GAAC,OAAO,MAE9D,EAAA,KAAK,OACD,OAAO,UAAU,IAAU,EAAQ,EACnC,WAAM,MAAA,YAAY,EAAI,sCACf,IAAuB,GAAM,mBAeL,GACrC,MAAI,KAAU,KACL,OACE,MAAM,QAAQ,GAChB,IAAM,EAAM,IAAI,SAAA,GAAK,MAAA,IAAuB,KAAI,KAAK,KAAO,IAC1D,MAAO,IAAU,SACnB,IAAI,EAAK,IAET,GAAG,cAcV,EAA8B,GAChC,GAAI,GAAW,EAAA,KAAK,MAChB,EACE,EAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,GAAM,GAAM,EAAA,KAAK,MACjB,MAAI,GAAM,EAAW,GAGrB,GAAW,EACX,EAAa,EAAC,MAAA,OAAI,IACX,GAET,MAAO,eASkC,GAEzC,MAAI,KAAmB,OACd,OAEL,IAAmB,SACd,SAEL,IAAmB,MACd,MAEF,KCvgBT,AAoBA,YAAqB,EAAW,GAC9B,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAI,EAAM,OAY1D,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAGE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GANyC,EAAA,cAAc,cA8BvD,GAAA,SAAA,GAA6B,EAAA,EAAA,GAQ3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAJQ,SAAA,gBAAkB,EAClB,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,YAAgB,EAAO,EAAG,EAAK,UAC/C,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,SAAU,KAAK,SAAU,KAAM,KAAK,OAtB9B,EAAA,UAAY,UAwB9B,GA1B6B,IA2B7B,EAAA,cAAc,cAAc,IAoB5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAK5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAFQ,SAAA,YAAc,EAG7B,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,KAAW,GAAY,EAAG,EAAK,WAG9D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OAdL,EAAA,UAAY,WAgB9B,GAlB8B,IAmB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAIE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAHF,EAAA,UAAY,SAK9B,GAP4B,IAQ5B,EAAA,cAAc,cAAc,IAoC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAY9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KANQ,SAAA,gBAAkB,EAClB,EAAA,gBAAkB,EAClB,EAAA,YAAc,EACd,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,YACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,IACZ,EAAA,IACI,EAAK,KAAM,EAAA,YAAgB,EAAO,EAAK,SAAU,EAAK,WAC1D,EAAA,IAAQ,EAAM,EAAK,KAAM,IAC7B,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,SAAU,KAAK,SACf,SAAU,KAAK,SACf,KAAM,KAAK,KACX,KAAM,KAAK,OApCC,EAAA,UAAY,aAuC9B,GAzCgC,IA0ChC,EAAA,cAAc,cAAc,IAQrB,GAAM,IACwC,CAC/C,QAAW,UACX,WAAc,aACd,OAAU,SACV,SAAY,wBAGkB,GAElC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA0C,GAC1C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCpQjC,YAyBwB,GACtB,MAAO,IAAI,IAAQ,eAQI,GACvB,MAAO,IAAI,IAAS,iBASpB,MAAO,IAAI,gBAIc,GACzB,MAAO,IAAI,IAAW,0ECjDxB,AAqBO,GAAM,IAA2B,CAAC,gBAAiB,gBAO7C,GAA4B,CAAC,QAAS,OAAQ,UAI9C,GAAyB,CAAC,MAAO,OAIjC,GAAkC,CAAC,MAAO,MAAO,SAAU,OCpCxE,AAmBA,GAAM,IAA+B,GAAI,iBAET,GAC9B,GAA0B,GAA0B,aAAc,eAGnC,GAC/B,GAA0B,GAA2B,cAAe,eAGxC,GAC5B,GAA0B,GAAwB,WAAY,GAGhE,GAAM,IAA4B,GAC5B,GAAoB,gBAKG,EAAc,GACzC,GAAgB,KAAK,GACrB,IACE,GAAM,GAAS,IACf,UAAgB,MACT,QACA,GACP,SAAgB,MACV,GAOV,cACE,MAAI,IAAgB,SAAW,EACtB,GAEA,GAAgB,KAAK,IAAqB,eASjB,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,MAAO,MAA2B,cAaA,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,AAAK,GAAQ,IAAI,IACf,GAAQ,IAAI,EAAY,GAE1B,GAAM,GAAQ,GAAQ,IAAI,GAG1B,GAFA,GAAQ,IAAI,EAAY,GAAQ,IAAI,GAAc,GAE9C,EAAQ,GACV,GAAM,GAAY,EAAU,IAAI,EAGhC,UAAQ,IAAI,EAAQ,GACb,MAEP,OAAO,GAIX,GAAM,IAAkB,GAAI,QAAO,+CAOD,GAChC,MAAO,CAAC,CAAC,EAAK,MAAM,ICjHtB,YA8B0B,GACxB,MAAO,KAAM,SAAS,EAAE,WAAY,gBAWlC,EAA4B,EAAgB,GAC9C,AAAI,GAAS,MACX,GAAQ,GAEV,AAAI,GAAO,MACT,GAAM,EAAM,QAId,OADI,GAAO,EACF,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,GAAQ,EAAM,GAEhB,MAAO,GAQT,YAAmB,GACjB,SAAQ,MAAM,QAAQ,GAAS,GAAI,cAAa,GAAS,EAClD,EAAA,SAAS,eAQE,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAQ1B,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAsDxB,EAAe,GACnC,GAAI,EAAM,EACR,KAAM,IAAI,GAAW,QAAQ,EAAG,cAAc,EAAK,mBAGrD,OADM,GAAgB,GACb,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,EAAI,KAAK,GAEX,MAAO,GChJT,YAwEqB,EAAW,GAC9B,MAAO,GAAE,OAAO,eASS,EAAW,GAAA,AAAA,IAAA,QAAA,GAAA,IACpC,GAAM,GAAW,EAAE,MAAM,QACzB,MAAI,GAAO,GACT,GAAO,EAAS,OAAS,EAAO,GAElC,EAAS,OAAO,EAAM,EAAG,GAClB,EAAE,QAAQ,eAcI,EAAW,GAChC,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,oDACA,SAAQ,EAAE,MAAM,OAAM,aAE5B,GAAM,GAAI,GAAW,EAAG,GACxB,MAAO,IAAK,EAAG,CAAC,EAAG,EAAG,kBASF,GACtB,GAAM,GAAW,CAAC,GAAqB,EAAE,QACzC,MAAO,GAAE,QAAQ,eAWU,GAC3B,GAAI,EAAE,MAAQ,EACZ,KAAM,IAAI,GACN,wDAAwD,EAAE,KAAI,KAEpE,GAAM,GAAW,CAAC,EAAE,MAAM,GAAI,GAAqB,EAAE,MAAO,IAC5D,MAAO,GAAE,QAAQ,eAYf,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,GAAI,CAAC,EAAM,EAAM,MAAM,SACnD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,GAC9B,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,EAAG,GACjC,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,MAAU,EAAmB,CAAC,EAAO,EAAG,EAAG,EAAG,GAAI,CACvD,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SAEjE,GACH,MAAO,GAAA,MAAU,EAAO,CAAC,EAAO,EAAG,EAAG,EAAG,EAAG,GAAI,CAC9C,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAClE,EAAM,MAAM,aAGd,KAAM,IAAI,GACN,8DACA,IAAG,EAAM,sBAcjB,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,GAAQ,CAAC,EAAM,MAAM,GAAI,QACjD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,GAC1B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,QAClC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAG,GAC7B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,YAErD,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAejB,EAAe,EAAe,EAAc,GAC9C,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,GAC9B,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,EAAG,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAO,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpD,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,YAGX,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAWO,EAAmB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC7C,GAAI,GACJ,MAAI,GAAO,GACT,GAAO,EAAQ,GAAG,KAClB,AAAI,IAAS,EACX,EAAO,EAEP,EAAO,GAGX,AAAI,IAAS,EAAQ,GAAG,MAGtB,GAAO,IAGF,EAAA,OAAW,EAAS,eAUQ,EAAW,GAC9C,OAAQ,EAAE,UACH,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,QACjC,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,WAEpD,KAAM,IAAI,GACN,kDACA,iBAAgB,EAAE,oBAWP,EAAW,GAI9B,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAI,CAAC,IAEH,EAAE,OAAS,EAAE,OACf,KAAM,IAAI,GACN,0BAA0B,EAAE,OAAM,oBAClC,yCAAwC,EAAE,KAAI,MAEpD,MAAO,GAAA,KAAS,EAAG,eAgBjB,EAAc,EAAY,EAAc,EACxC,GADc,MAAA,KAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAErB,EAAA,aAAiB,EAAO,EAAM,EAAQ,EAAO,eAsBlD,EAAW,EAAW,EACtB,GACF,GAAK,EAAE,KAAO,GAAO,EAAE,KAAO,EAC5B,KAAM,IAAI,IACN,2CACA,uBAAsB,EAAE,MAAK,kBAAkB,EAAE,QAEvD,GAAI,EAAE,MAAQ,GACZ,GAAM,GAAW,EAAE,MAAM,MAAM,IAAI,GAC7B,EAAiB,EAAE,MAAM,MAAM,IAAI,GACzC,GAAI,IAAa,EACf,KAAM,IAAI,IACN,2CACA,yDACI,EAAE,MAAK,SACX,eAAc,EAAE,QAIxB,GAAK,EAAE,OAAS,GAAO,EAAE,OAAS,GAChC,GAAM,GAAa,GACb,EAAa,GAInB,MAAO,GAAA,MAAU,OAAO,CACtB,EAAC,EACD,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,SAIZ,GAAM,GAAa,EAAE,MAAM,QACrB,EAAW,EAAW,MAC5B,EAAI,EAAE,QAAQ,CAAC,GAAI,IAInB,GAAM,GAAS,EAAE,MAAM,QACjB,EAAW,EAAO,MAClB,EAAiB,EAAO,MACxB,EAAiB,EAAM,OAAA,CAAE,IAGzB,EAAO,MAAM,KAAK,CAAC,OAAQ,EAAE,MAAO,SAAC,EAAG,GAC5C,MAAI,KAAM,EACD,EAAE,KAAO,EACP,GAAK,EAAE,KAAO,EAChB,EAAI,EAEN,IAET,EAAI,EAAE,UAAU,GAAM,QAAQ,CAAC,EAAgB,KAG/C,GAAM,GAAkB,EAAU,OAAK,GACjC,EAAa,GACb,EAAa,GACnB,MAAO,GAAA,MACF,OAAO,CACN,EAAC,EACD,EAAC,EACD,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,IAEX,QAAQ,gBAyDb,EAAmB,EAA4B,GACjD,MAAO,GAAA,KAAK,WACV,MAAI,OAAM,QAAQ,GAChB,EAAU,EAAA,SAAS,EAAS,SAE5B,EAAU,EAAQ,QAEb,EAAA,OAAW,EAAW,EAAS,iBASnB,GACrB,MAAO,GAAA,IAAQ,EAAG,GA+BpB,YAAqB,EAAe,EAAc,GAChD,GAAM,GAAY,EAAK,MAEvB,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,GACN,+BAA+B,EAAK,KACpC,6BAA4B,IAGlC,GAAI,IAAU,GACZ,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,EAAG,IAErC,EAAK,QACR,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,GAAI,EAAU,KAEzD,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAG,EAAU,KAEpC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,IAElC,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,KAE3D,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAU,KAEjC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,IAE/B,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,KAE7C,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAU,KAE9B,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,EAAQ,EACjB,MAAO,GAET,KAAM,IAAI,GAAW,sCAAsC,EAAK,kBAc9D,EAAW,EAAc,GAC3B,MAAO,GAAA,KAAK,WACV,MAAI,IAAc,MAChB,GAAa,MAEf,GAAgB,GAET,EAAE,IAAI,GAAY,EAAE,KAAM,EAAM,kBAUvB,EAAW,GAE7B,GAF6B,AAAA,IAAA,QAAA,GAAA,GAEzB,IAAU,EACZ,KAAM,IAAI,IACN,0CAA0C,EAAK,6BAGrD,MAAO,GAAA,IAAQ,eAWQ,GACvB,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,GAAG,IAAI,kBAc1C,EAAW,EAAe,EAAuB,GACnD,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAG,EAAO,EAAY,iBAY1B,GAC1B,MAAO,GAAA,KAAK,WACV,GAAM,GAAI,EAAA,IAAQ,GAAI,EAAA,IAAQ,GAAI,IAClC,MAAO,GAAA,YAAgB,EAAG,EAAG,iBAiBD,EAAY,EAAc,GAAA,MAAA,KAAA,QAAA,GAAA,IACjD,EAAW,IAAM,IC7sB1B,AAgBO,GAAM,IAAwB,CAAC,QAAS,SAAU,UAO5C,GACT,CAAC,SAAU,UAAW,mBCxB1B,YAoB6B,GAC3B,GAA0B,GAAuB,UAAW,eAG5B,GAChC,GAA0B,GAA2B,eAAgB,GASvE,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DACS,SAAA,UAAA,4BAAP,WACE,MAAO,IAUT,EAAA,UAAA,UAAA,WACE,MAAO,IAEX,GAf0C,EAAA,cAAc,cAiBxD,GAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,MAAM,EAAO,IAHf,EAAA,UAAY,QAKrB,GAP2B,IAQ3B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,KAAK,EAAO,IAHd,EAAA,UAAY,OAKrB,GAP0B,IAQ1B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,GACN,oDAAoD,GAE1D,GAAI,EAAK,QAAU,OACjB,KAAM,IAAI,GAAW,sCAAsC,GAE7D,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAI,EAAA,OAAO,EAAK,OAAQ,EAAA,KAAK,EAAO,OAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,QApBT,EAAA,UAAY,WAuBrB,GAzB8B,IA0B9B,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GASjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,eAAiB,KACjB,EAAA,eAAiB,IAOxB,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,cAAc,EAAO,KAAK,OAAQ,KAAK,OAAQ,IAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,OAAQ,KAAK,OAAQ,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAnBxD,EAAA,UAAY,gBAqBrB,GAvBmC,IAwBnC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAShC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,uCAAuC,EAAK,KAGlD,MAAO,IAAe,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGnE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,eA2BrB,GA7BkC,IA8BlC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,0CAA0C,EAAK,KAErD,MAAO,GAAA,gBAAgB,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGpE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,kBA2BrB,GA7BqC,IA8BrC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,IAG9C,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,SAAW,GAAK,EAAM,KAAO,EAAM,GAC3C,KAAM,IAAI,GACN,wEAGJ,MAAO,GAAA,IAAI,EAAK,KAAM,EAAA,IAAI,EAAM,QAKtC,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OApBd,EAAA,UAAY,WAsBrB,GAxB8B,IAyB9B,EAAA,cAAc,cAAc,IAU5B,YACI,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,gBAChB,GAAI,GACA,EAEJ,GADA,GAAgB,GACZ,EAAM,SAAW,EACnB,EAAQ,EAAM,GACd,EAAS,EAAM,WACN,CAAC,EAAG,EAAG,GAAG,QAAQ,EAAM,UAAY,IAC7C,GAAI,IAAe,iBACjB,GAAM,GAAqB,GAAU,EAAO,GAC5C,EAAQ,EAAM,GAAK,EACnB,EAAS,EAAM,GAAK,UACX,IAAe,gBACxB,GAAM,GAAqB,GAAU,EAAO,EAAG,EAAM,OAAS,GAC9D,EAAQ,EAAM,EAAM,OAAS,GAAK,EAClC,EAAS,EAAM,EAAM,OAAS,GAAK,QAGrC,GAAM,GAAY,GAAU,GAC5B,EAAQ,KAAK,KAAK,GAClB,EAAS,KAAK,KAAK,GAGrB,MAAO,CAAC,EAAO,GAiBjB,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAYnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,EAAK,MAAQ,EACf,KAAM,IAAI,GACN,wCAAwC,EAAK,OAEnD,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAM,EAAK,MAC7C,EAAK,KAAO,EAAK,MAAQ,KAAO,QAAU,EAAK,KAC/C,GAAa,EAAK,MAClB,EAAK,aACD,EAAK,cAAgB,KAAO,SAAW,EAAK,aAChD,GAAkB,EAAK,cACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAO,GAAY,GACnB,EAAQ,EAAK,GACb,EAAS,EAAK,GAChB,EAAQ,KAAK,MASjB,GARA,AAAI,KAAK,OAAS,QAChB,GAAS,KAAK,IAAI,EAAG,GAChB,AAAI,KAAK,OAAS,SACvB,GAAS,KAAK,IAAI,EAAG,GAErB,GAAS,KAAK,IAAI,EAAI,GAAQ,GAAU,GAGtC,KAAK,eAAiB,UACxB,GAAM,GAAS,KAAK,KAAK,GAEzB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACH,KAAK,eAAc,2BAA2B,EAAK,KAE5D,MAAO,GAAA,gBAAgB,EAAO,EAAG,EAAQ,EAAO,KAAK,WAErD,GAAM,GAAQ,KAAK,KAAK,EAAI,GAC5B,MAAO,GAAA,cAAc,EAAO,CAAC,EAAO,EAAO,KAI/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,MACZ,KAAM,KAAK,KACX,aAAc,KAAK,aACnB,KAAM,KAAK,OAzDR,EAAA,UAAY,kBA4DrB,GA9DqC,IA+DrC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,gBAwBrB,GA1BmC,IA2BnC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAWhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,eAwBrB,GA1BkC,IA2BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,WAiBrB,GAnB8B,IAoB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAI7B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,YAiBrB,GAnB+B,IAoB/B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAI/B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBiC,IAoBjC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAIhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBkC,IAoBlC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAO9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GATO,EAAA,aAAe,EAMtB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,KAAO,EAAK,KAEb,EAAK,MAAQ,KACf,KAAM,IAAI,IACN,2EAIR,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,IAAoB,8BAEhC,AAAI,EAAM,GAAK,EAAM,GAAK,KACxB,QAAQ,KACJ,gEACA,eAAc,EAAM,GAAK,EAAM,GAAE,gBACjC,wBAIN,GAAM,GACF,EAAM,GAAK,EAAM,GAAK,CAAC,EAAM,GAAI,EAAM,IAAM,EAC3C,EAAI,GAAe,EAAiB,EAAG,EAAG,WAC5C,EAAI,EAAA,OAAO,YAAY,GAC3B,MAAI,GAAM,GAAK,EAAM,IACnB,GAAI,EAAE,aAED,EAAA,IAAI,EAAK,KAAM,MAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CACL,KAAM,KAAK,KACX,KAAM,KAAK,OA3CR,EAAA,UAAY,aA8CrB,GAhDgC,IAiDhC,EAAA,cAAc,cAAc,IAUrB,GAAM,IACyC,CAChD,SAAY,WACZ,aAAgB,eAChB,cAAiB,gBACjB,SAAY,WACZ,UAAa,YACb,SAAY,WACZ,YAAe,cACf,aAAgB,eAChB,KAAQ,OACR,WAAc,aACd,aAAgB,eAChB,cAAiB,gBACjB,gBAAmB,kBACnB,gBAAmB,kBACnB,MAAS,SAGf,YACI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGgB,GAEnC,MAAO,IAAqB,eAGC,GAE7B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EAIJ,GAAI,IAAc,eAChB,MAAO,IAAI,IACN,GAAI,IAAc,gBACvB,MAAO,IAAI,IACN,GAAI,IAAc,WACvB,MAAO,IAAI,IACN,GAAI,IAAc,YACvB,MAAO,IAAI,IACN,GAAI,IAAc,cACvB,MAAO,IAAI,IACN,GAAI,IAAc,eACvB,MAAO,IAAI,IAEX,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAuB,OAE3B,OAAI,aAAsB,IACxB,EAEA,GAAuB,GCxoBlC,cAkBE,MAAO,IAAI,kBASX,MAAO,IAAI,gBAQY,GACvB,MAAO,IAAI,IAAS,eAYQ,GAC5B,MAAO,IAAI,IAAc,eASE,GAC3B,MAAO,IAAI,IAAa,eAaM,GAC9B,MAAO,IAAI,IAAgB,eASJ,GACvB,MAAO,IAAI,IAAS,eAgBU,GAC9B,MAAO,IAAI,IAAgB,eAgBC,GAC5B,MAAO,IAAI,IAAc,eAgBE,GAC3B,MAAO,IAAI,IAAa,eAeD,GACvB,MAAO,IAAI,IAAS,eAeI,GACxB,MAAO,IAAI,IAAU,eAgBK,GAC1B,MAAO,IAAI,IAAY,eAYI,GAC3B,MAAO,IAAI,IAAa,eAWC,GACzB,MAAO,IAAI,IAAW,kPChNxB,AAmBA,GAAI,IAAsB,gBAGxB,MAAO,MAGT,GAAM,IAA2C,eAO1B,GAAA,MAAA,KAAA,QAAA,GAAA,IACrB,AAAM,IAAU,KACd,IAAa,GAAU,GAEzB,GAAa,IAAW,EACjB,EAAS,GAAa,GAAQ,WCrCvC,YAoBgC,GAC9B,MAAO,OAAM,QAAQ,IAAM,MAAM,QAAQ,EAAE,gBASV,GACjC,MAAI,GAAE,SAAW,EACR,GAEJ,MAAM,QAAQ,EAAE,IAGd,EAFE,CAAC,eAWwB,GAClC,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAG,SAAW,EAChB,KAAM,IAAI,GAAW,uCAAuC,EAAG,QAEjE,EAAI,EAAG,OAEP,GAAI,EAEN,MAAO,eAa0B,GACjC,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,EAAO,KAChD,GAAI,EAAO,SAAW,EACpB,SAAS,EACF,EAAO,GAEd,KAAM,IAAI,GAAW,iCAAiC,EAAO,YAG/D,OAAO,GC9EX,YAmBqC,GAEnC,OADI,GAAQ,EACS,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,AAAI,EAAO,MAAM,SAAW,EAC1B,GAAS,EAET,GAAS,EAAO,MAAM,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAG/C,MAAO,GC5BT,AAoBA,GAAM,IAA+B,yBAqCnC,WACI,EAAa,EACb,EAAqC,EACrC,GAFa,AAAA,IAAA,QAAA,GAAA,WACb,AAAA,IAAA,QAAA,GAAA,IAAqC,AAAA,IAAA,QAAA,GAAA,IACrC,AAAA,IAAA,QAAA,GAAA,MACF,KAAK,MAAQ,GAAgB,UAC7B,KAAK,MAAQ,EAAI,MACjB,KAAK,GAAK,KAEV,EAAO,GAAe,GACtB,KAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,cAErC,KAAK,WAAa,EAClB,KAAK,WAAa,EAElB,KAAK,IAAM,EAAA,SAAa,EAAK,KAAK,WAAY,KAAK,KAAM,KAAK,OAUhE,SAAA,UAAA,KAAA,WACE,YAAK,oBACE,KAAK,KAUd,EAAA,UAAA,MAAA,SAAM,GAEJ,YAAK,oBACL,GAAiB,KAAK,IAAK,GAE3B,AAAI,KAAK,IAAI,KAAO,EAAO,IACzB,MAAK,IAAI,OAAO,GAChB,AAAI,KAAK,YAAc,MACrB,KAAK,IAAI,OAAO,KAAK,WAAW,MAAM,KAAK,OAGxC,MAMT,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,KAAK,IAAI,WAGD,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,IAAI,WACX,KAAM,IAAI,OAAM,kBAAkB,KAAK,KAAI,0BAI/C,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,WAAa,EAClB,KAAK,IAAI,UAAY,mCAEzB,KAEA,YAA0B,EAAa,GACrC,GAAI,EAAE,MAAM,aAAe,EAAE,MAAM,WACjC,KAAM,IAAI,OACN,mBAAqB,KAAK,UAAU,EAAE,OAAS,QAC/C,KAAK,UAAU,EAAE,oBA0LK,GAC5B,MAAO,GAAG,IAAI,SAAA,GAAK,MAAA,GAAE,qBAYnB,GACF,EAAmB,QAAQ,SAAA,GACzB,GAAM,GAA0B,EAAiB,GACjD,EAAS,MAAM,EAAiB,MClVpC,kBAuEE,WAAY,GACV,KAAK,MAAQ,EAAK,MAClB,KAAK,MAAQ,EAAK,MAKlB,AAAI,EAAK,OAAS,KAChB,KAAK,KAAO,EAAK,MAAM,OAEvB,KAAK,KAAO,EAAK,KAEnB,KAAK,QAAU,EAAK,QACpB,KAAK,QAAU,EAAK,QACpB,KAAK,KAAO,EAAK,MAAQ,GAE7B,MAAA,oBA6CE,WACa,EAA0B,EAC5B,EAA6B,EAC3B,EAAkB,EAClB,GAHA,KAAA,MAAA,EAA0B,KAAA,MAAA,EAC5B,KAAA,YAAA,EAA6B,KAAA,OAAA,EAC3B,KAAA,SAAA,EACA,KAAA,kBAAA,EACX,KAAK,GAAK,KACV,AAAI,GAAQ,MACV,MAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,eAEvC,KAAK,KAAO,EAAM,OAEtB,MAAA,MA2DI,GAAc,EAsBlB,GAAA,WAwCE,WACI,EAEO,GAAA,KAAA,SAAA,EACT,KAAK,GAAK,KAQV,KAAK,cAAgB,EAAK,cAU1B,KAAK,cAAgB,EAAK,cAE1B,KAAK,YAAc,EAAK,YAExB,KAAK,cAAgB,EAAK,cAQ1B,KAAK,aAAe,EAAK,aAEzB,KAAK,cAAgB,EAAK,cAM1B,KAAK,WAAa,EAAK,WAEvB,KAAK,YAAc,EAAK,YAKxB,KAAK,YAAc,EAAK,YAExB,KAAK,aAAe,EAAK,aAGzB,OAAoB,GAAA,EAAA,EAAA,EAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MACX,EAAM,cAAc,KAAK,MAG7B,EAAK,cAAc,aAAa,KAAK,MAGvC,SAAA,UAAA,UAAA,WAEE,OADM,GAAyB,GACX,EAAA,EAAA,EAAA,KAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,KACX,EAAa,KAAK,EAAM,MAExB,EAAa,KAAK,MAGtB,MAAO,CACL,cAAe,KAAK,cAAgB,KAAK,cAAc,KAAO,KAC9D,cAAe,EACf,YAAa,KAAK,YAClB,cAAe,KAAK,gBAG1B,KAkDI,GAAe,EAWnB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAmDlC,WAAY,GAAA,AAAA,IAAA,QAAA,GAAA,IAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAtBD,EAAA,UAAsB,KAEtB,EAAA,kBAA8B,GAQ5B,EAAA,UAAY,GAapB,EAAK,GAAK,KAEV,EAAK,oBAAsB,KAE3B,EAAK,UAAY,KACjB,EAAK,gBAAkB,GAGvB,EAAK,kBAAoB,GACzB,EAAK,qBAAuB,GAC5B,EAAK,QAAU,GACf,EAAK,SAAW,GAChB,EAAK,OAAS,GAMd,EAAK,aAAe,GACpB,EAAK,cAAgB,GAErB,GAAI,GAAO,EAAK,KAChB,GAAI,CAAC,GACH,GAAM,GAAS,EAAK,eACpB,EAAO,GAA0B,GAAU,IAAM,GAAO,GAM1D,GAJA,EAAK,KAAO,EAEZ,EAAK,WAAa,EAAK,WAAa,KAAO,GAAO,EAAK,UAEnD,EAAK,YAAc,MAAQ,EAAK,iBAAmB,MAKrD,GAAI,GAAe,OACnB,GAAI,EAAK,iBAAmB,KAC1B,EAAkB,EAAK,wBACd,EAAK,YAAc,MAC5B,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAkB,CAAC,GAAW,OAAO,EAAK,YAE5C,EAAK,gBAAkB,EAGvB,GAAI,GAAQ,EAAK,MACjB,AAAI,GAAS,MACX,GAAQ,EAAK,YAEf,AAAI,GAAS,MACX,GAAQ,WAEV,EAAK,MAAQ,EAGf,MAAI,GAAK,SAAW,KAClB,EAAK,eAAiB,EAAK,QAE3B,EAAK,eAAiB,KAKxB,EAAK,UAAY,KAEjB,EAAK,0BAA4B,KAYlB,SAAA,QAAjB,SAAyB,EAAc,GACrC,MAAO,GAAM,KAAO,OAAS,EAAU,YAUjC,EAAA,UAAA,eAAR,SAAuB,EAAmB,GACxC,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,mCACA,4BAA2B,EAAQ,MAEzC,GAAI,KAAK,aAAa,QAAU,EAC9B,KAAM,IAAI,GACN,gBAAgB,EAAQ,YAAY,EAAS,KAC7C,2BAA0B,KAAK,aAAa,OAAM,oBAExD,MAAO,MAAK,aAAa,IAY3B,EAAA,UAAA,WAAA,SAAW,GACT,MAAO,IACH,KAAK,eAAe,EAAW,SAAS,eAY9C,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,IACH,KAAK,eAAe,EAAW,UAAU,gBAgB/C,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,uHAIC,GAAI,KAAK,aAAa,SAAW,EACtC,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0CAEN,MAAO,IACH,KAAK,eAAe,EAAG,SAAS,+CActC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0BAEN,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,yHAKN,MAAO,IACH,KAAK,eAAe,EAAG,UAAU,gDAGvC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAQd,EAAA,UAAA,gBAAA,WAKE,MAAO,MAAK,OAAO,IAAI,SAAA,GAAU,MAAA,QAGnC,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,0CAGd,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,MAAO,MAAK,YAGd,SAAU,GACR,KAAK,OAAS,mCAGhB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,kBAAkB,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,IAClD,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAI,MAAK,WACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,GAAE,YAErC,QAIX,SAAqB,GACnB,KAAK,kBAAoB,mCAG3B,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAI,MAAK,UACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,CAAC,EAAE,YACxC,OAAO,KAAK,sBAEV,KAAK,kBAAkB,OAAO,KAAK,2BAI9C,SAAwB,GACtB,KAAK,qBAAuB,mCAO9B,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAG3C,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,2CAUd,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,SACR,KAAM,IAAI,OACN,yEAiBE,EAAA,UAAA,yBAAV,SAAmC,GAGjC,GADA,EAAS,GAAqB,GAC1B,KAAK,WAAa,MAAQ,KAAK,UAAU,SAAW,EACtD,OAEF,GAAM,GAAY,GAAqB,KAAK,WAC5C,GAAI,EAAO,SAAW,EAAU,OAC9B,KAAM,IAAI,GACN,SAAS,KAAK,KAAI,YAAY,EAAU,OAAM,YAC9C,oBAAmB,EAAO,OAAM,oBAChC,oBAAmB,IAEzB,OAAS,GAAa,EAAG,EAAa,EAAO,OAAQ,KACnD,GAAM,GAAI,EAAO,GACX,EAAkB,EAAU,GAClC,GAAI,GAAQ,KACV,SAIF,GAAM,GAAO,EAAE,KACf,GAAI,EAAK,MAAQ,MACX,IAAS,EAAK,KAChB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,KAC3D,kBAAiB,EAAK,KAAI,gBAAgB,IAGlD,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,IAG3D,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,EAAI,MAK/D,GAAI,EAAK,OAAS,MACZ,EAAE,QAAU,EAAK,MACnB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,IAC3D,qBAAoB,EAAK,MAAK,iBAAiB,EAAE,MAAK,MAK9D,GAAI,EAAK,MACP,GAAM,GAAS,EAAE,MACjB,OAAW,KAAO,GAAK,MACrB,GAAM,GAAO,OAAO,GACd,EAAQ,EAAK,KAAK,GAIlB,EACF,GAAQ,EAAI,EAAO,GAAQ,EAAO,EAAO,OAAS,GACtD,GAAI,GAAS,MAAQ,CAAC,EAAO,MAAM,QAAQ,KAAkB,GAC3D,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,mBAAmB,EAAI,uBACnC,eAAc,EAAK,kBAAkB,EAAM,OAMrD,GAAI,EAAK,OAAS,KAChB,OAAS,GAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAU,EAAK,MAAM,GACrB,EAAM,EAAE,MAAM,GACpB,GAAI,GAAW,MAAQ,GAAO,MACxB,IAAY,EACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,oBAAoB,EAAK,MAAK,MAC1C,gBAAe,EAAE,MAAK,SAgBtC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,IAGC,EAAA,UAAA,eAAV,SAAyB,EAAyB,GAChD,AAAI,KAAK,WAAa,MACpB,KAAK,UAAU,EAAQ,IAS3B,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,UAAY,GAOnB,EAAA,UAAA,cAAA,WACE,KAAK,UAAY,MAwEnB,EAAA,UAAA,MAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAGE,EAAS,GAAU,GAEnB,KAAK,oBAML,OAHM,GAAa,GAAqB,GAEpC,EAAiB,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,CAAE,aAAiB,MACrB,EAAiB,GACjB,OAIJ,OADI,GAAkB,GACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,YAAiB,KACnB,EAAkB,GAClB,OAIJ,GAAI,IAAmB,EACrB,KAAM,IAAI,GACN,mEAKN,MAAO,IAAU,KAAK,KAAM,WAE1B,GAAI,CAAC,EAAK,OAKR,EAAK,yBAAyB,GAI9B,OADM,GAAuB,GACT,EAAA,EAAA,EAAA,GAAqB,GAArB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,KAAK,EAAM,OAEzB,EAAK,MAAM,GAA+B,IAC1C,EAAK,MAAQ,GAGb,AAAI,EAAK,gBACP,EAAK,WAAW,EAAK,gBAGvB,AAAI,EAAK,YAAc,MAAQ,GAI7B,GAAK,UAAY,GAcrB,GANA,EAAK,yBAAyB,GAM1B,GAUF,OATI,GAAS,EAAK,KAAK,EAA6B,GAK9C,EAAuB,GAAqB,GAC5C,EAA2B,GAGnB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAT,GAAI,GAAC,EAAA,GACR,AAAI,EAAW,QAAQ,KAAO,IAC5B,GAAI,EAAE,SAER,EAAe,KAAK,GAItB,GAFA,EAAS,GAA+B,GAEpC,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAKN,MAAO,QAEP,GAAM,GAAa,GAAkB,GAC/B,EAAc,EAAK,mBAAmB,GACxC,EAAM,OACJ,EAAc,GAAiB,GAgCrC,GA/BA,EAAK,6BACD,MAAM,QAAQ,GAAU,EAAW,GACX,GAE5B,AAAI,GAAe,MAAQ,EAAY,OAAS,GAC5C,MAAM,QAAQ,EAAY,IAE5B,EAAU,EACI,IACG,SAAC,EAAO,GAAU,MAAA,IAAI,IAClB,EAAa,EAAO,EACpB,GAAqB,GAAS,EAAQ,EAAK,KAC3C,KAErB,EAAS,GAAI,IACT,EAAa,EAAsB,EACnC,GAAqB,GAAS,EAAQ,EAAK,MAUjD,EAAK,eACD,EAA6C,EAAQ,KAAM,KAC3D,EAAY,EAAa,GAC7B,EAAK,YAED,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAIN,MAAO,OAYH,EAAA,UAAA,6BAAV,SAAuC,GACrC,GAAI,KAAK,iBAAmB,KAC1B,OACK,GAAI,EAAW,SAAW,KAAK,gBAAgB,OACpD,QAAQ,KACJ,iDACG,MAAK,UAAU,GAAW,iCAC7B,qBAAoB,KAAK,UAAU,KAAK,iBAAgB,MACxD,iBAAgB,KAAK,YAEzB,GAAI,GAAc,GAClB,KAAK,gBAAgB,QAAQ,SAAC,EAAW,GACvC,AAAI,GAAa,MAAQ,EAAW,IAAM,MACtC,EAAW,KAAO,GACpB,GAAc,MAGlB,AAAI,GACF,QAAQ,KACJ,iCACA,KAAI,KAAK,UAAU,GAAW,eAC9B,mCAAkC,KAAK,KAAI,MAC3C,IAAG,KAAK,UAAU,KAAK,qBAiBjC,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,GAAI,KAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,EAC5D,KAAM,IAAI,IACN,aAAa,KAAK,KAAI,gEAI5B,OADM,GAA4B,GACf,EAAA,EAAA,EAAA,KAAK,aAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAc,KAAK,UAAU,EAAK,cACxC,AAAI,EAAgB,QAAQ,KAAiB,IAC3C,EAAgB,KAAK,GAGzB,GAAI,EAAgB,SAAW,GAC7B,GAAM,GAAe,KAAK,aAAa,GAAG,aAC1C,MAAI,OAAM,QAAQ,IAAiB,MAAM,QAAQ,EAAa,KAC1D,EAAa,SAAW,EAClB,EAAyB,GAE1B,MAIT,MAAM,IAAI,IACN,aAAa,KAAK,KAAI,+JAiB9B,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,sCAAsC,KAAK,KAAI,uFAIrD,MAAO,IAAoC,KAAK,UAclD,EAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAWf,EAAA,UAAA,WAAA,SAAW,GAAA,MAAA,KAAA,QAAA,GAAA,IACF,GAAc,EAAgB,KAAK,iBAAmB,KAAK,UAepE,EAAA,UAAA,WAAA,SAAW,GAAX,GAAA,GAAA,KACE,EAAA,KAAK,WACH,GAAM,GAAS,EAAK,QACpB,GAAI,EAAO,SAAW,EAAQ,OAK5B,KAAM,IAAI,GACN,4CAA4C,EAAK,KAAI,KACrD,iCAAgC,EAAQ,OAAM,MAC9C,gCAA+B,EAAO,OAAM,cAC5C,sBAAqB,EAAO,QAElC,GAAI,EAAO,SAAW,EACpB,OAIF,OAFM,GAAoD,GACpD,EAAc,GAAc,GACzB,EAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,GACxC,GAAM,GAAK,EAAY,GACjB,EAAI,EAAO,GACX,EAAI,EAAQ,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,GACN,sBAAsB,EAAG,MAAK,IAC9B,8CAA6C,EAAE,QAErD,EAAkB,KAAK,CAAC,EAAG,IAE7B,GAAc,MAmBR,EAAA,UAAA,UAAV,SACI,EAAc,EAAc,EAAkB,EAC9C,EAA2B,EAC3B,GAEF,GAAI,KAAK,kBAAkB,QAAQ,KAAU,GAC3C,KAAM,IAAI,GACN,yBAAyB,EAAI,cAAc,KAAK,MAEtD,KAAK,kBAAkB,KAAK,GAE5B,AAAI,GAAS,MACX,GAAQ,WAGV,AAAI,KAAK,2BACP,GAAc,GAAe,UAE/B,GAAM,GAAY,EAAY,MAAM,EAAO,GACrC,EACF,GAAI,IAAc,EAAW,EAAO,EAAM,EAAW,GACzD,SAAU,UAEV,AAAI,GAAe,MACjB,KAAK,QAAQ,WAAM,MAAA,GAAY,MAAM,EAAO,UAE9C,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,EACF,KAAK,kBAAkB,KAAK,GAE5B,KAAK,qBAAqB,KAAK,GAE1B,GAaT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,KAAK,0BAA4B,GAWnC,EAAA,UAAA,QAAA,SAAQ,SACN,GAAI,GAAU,MAAQ,MAAM,QAAQ,IAAW,EAAO,SAAW,EAC/D,OAGF,EAAS,GAAqB,GAC9B,AAAI,KAAK,UAAY,QAAa,KAAK,UAAY,MACjD,GAAA,KAAK,QAAO,KAAI,MAAA,EAAI,IAexB,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAYT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,GAAI,CAAC,KAAK,iBACR,GAAI,GAAQ,KACV,GAAI,MAAM,QAAQ,GAChB,EAAK,QAAQ,SAAA,GACX,GAAI,GAAe,KACjB,KAAM,IAAI,WACN,SAAS,EAAK,KAAI,iEAK1B,MAAM,IAAI,WACN,SAAS,KAAK,KAAI,2DAK1B,MAAO,MAIT,MAAO,IAeD,EAAA,UAAA,eAAR,SACI,EACA,EACA,EAA6B,EAC7B,EAA4B,EAC5B,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAM,GACF,GAAqB,GACzB,EAAgB,GAAqB,GACrC,EAAa,GAAqB,GAClC,EAAc,GAAqB,GACnC,EAAc,GAA+B,GAC7C,EAAe,GAA+B,GAM9C,OAHM,GAAyB,GACzB,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GAKV,EAAc,KAAK,EAAE,aACrB,EAAY,KAAK,EAAE,WACnB,EAAc,KAAK,EAAE,aAMvB,GAAI,IACA,CACE,cAAe,KACf,cAAa,EACb,YAAW,EACX,cAAa,EACb,aAAc,EACd,cAAa,EACb,WAAU,EACV,YAAW,EACX,YAAW,EACX,aAAY,GAEd,GAGJ,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,IAExC,EAAc,GAAG,YAAc,KAC/B,EAAc,GAAG,UAAY,KAAK,aAAa,OAAS,EACxD,EAAc,GAAG,YAAc,GAyBnC,EAAA,UAAA,UAAA,WACE,GAAM,GACyB,CAAC,KAAM,KAAK,KAAM,UAAW,KAAK,WACjE,MAAI,MAAK,iBAAmB,MAC1B,GAAO,gBAAqB,KAAK,iBAEnC,AAAI,KAAK,OAAS,MAChB,GAAO,MAAW,KAAK,OAElB,GAQC,EAAA,UAAA,eAAV,WACE,YAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,YAC/B,KAAK,QAAQ,QAGZ,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,UAAU,KAAK,KAAI,2BAkCvC,EAAA,UAAA,QAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,uCAIvC,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,sCAIvC,KAAK,oBAEL,GAAI,GAAuB,EAC3B,MAAI,EAAE,KAAK,YAAc,GACvB,GAAuB,KAAK,kBAGvB,CAAC,qBAAsB,KAAK,UAAW,qBAAoB,IAEtE,GAxoCoC,EAAA,cAAc,cAopClD,YAA2B,GAEzB,EACI,GAAqB,GAEzB,OADM,GAAkB,GACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAO,KAAK,EAAE,OAEhB,MAAO,IAA+B,GAYxC,YAA0B,GAExB,MAAO,sBAeL,EAAwB,EACxB,GAKF,GAJA,AAAI,IAAS,MAAS,GAAa,MAAQ,EAAY,IACrD,GAAQ,EAAO,YACf,EAAY,EAAO,WAEjB,EAAM,aAAa,SAAW,EAChC,MAAO,CAAC,GAER,GAAM,GAAO,EAAM,aAAa,GAChC,GAAI,EAAK,cAAc,SAAW,EAChC,MAAO,GAAK,aAGZ,OADM,GAAkC,GAC/B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,IAM7C,OALM,GAAI,EAAK,aAAa,GACtB,EAAQ,EAAK,cAAc,GAC3B,EAAY,EAAK,YAAY,GAC7B,EAAkB,GAAgB,EAAG,EAAO,GAElC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAc,QAAQ,KAAO,IAC/B,EAAc,KAAK,GAIzB,MAAO,GCvmDb,AA4CA,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CACJ,MAAO,EAAK,MACZ,KAAM,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAO,SAAS,cACtD,KAaF,GAXA,AAAI,EAAK,WAAa,MACpB,GAAK,UAAY,MAEnB,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,IAGhB,EAAK,UAAY,GACjB,EAAK,MAAQ,GACb,EAAK,OAAS,EAAK,OAEf,EAAK,YAAc,MAAQ,EAAK,iBAAmB,KACrD,KAAM,IAAI,GACN,qGAGN,GAAI,GAAkB,EAAK,gBAC3B,GAAI,GAAmB,MACrB,GAAI,EAAK,YAAc,KACrB,KAAM,IAAI,GACN,iFAGJ,EAAkB,CAAC,EAAK,WAAW,OAAO,EAAK,oBAI7C,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,yFAKR,GAAM,GAAQ,EAAK,OAAS,UAE5B,EAAK,gBAAkB,EACvB,EAAK,MAAQ,EAEb,EAAK,UAAY,CAAC,CAAC,MAAO,IAE1B,GAAM,GAAc,GAAI,IACpB,EAAK,MAAO,EAAK,gBAAiB,EAAM,GAAI,GAAI,EAAK,MACzD,SAAY,UAAY,EACxB,EAAY,YAAc,EAK1B,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,CAAC,GACf,cAAe,CAAC,GAChB,WAAY,CAAC,MACb,YAAa,CAAC,MACd,YAAa,CAAC,GACd,aAAc,CAAC,OAInB,SAAA,UAAA,MAAA,SACI,EACA,GACF,KAAM,IAAI,GACN,+BACA,kDAAiD,KAAK,QAG5D,EAAA,UAAA,QAAA,WAEE,MAAO,CAAC,qBAAsB,KAAK,UAAW,qBAAsB,IAGtE,EAAA,UAAA,UAAA,WACE,MAAO,CACL,gBAAiB,KAAK,gBACtB,MAAO,KAAK,MACZ,OAAQ,KAAK,OACb,KAAM,KAAK,OAzFC,EAAA,UAAY,aA4F9B,GA9FgC,IA+FhC,EAAA,cAAc,cAAc,gBAmCN,GACpB,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAC/C,KAAM,IAAI,OACN,gIAKN,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAE/C,KAAM,IAAI,GACN,oFAGN,GAAI,GAAa,EAAO,WACxB,AAAI,EAAO,OAAS,MAAQ,GAAc,MACxC,GAAa,CAAC,MAAM,OAAO,EAAO,QAGpC,GAAI,GAAQ,EAAO,MACnB,AAAI,GAAS,MACX,GAAQ,WAGV,GAAM,GAAa,GAAI,IAAW,CAChC,gBAAiB,EACjB,KAAM,EAAO,KACb,MAAK,EACL,OAAQ,EAAO,SAGX,EAAU,EAAW,aAAa,GAAG,cAC3C,MAAO,GAAQ,GC9MjB,YA0B2C,iHACzC,GAAI,GAAQ,KACV,MAAA,CAAA,GAEI,EAA+D,GAC/D,EAAiB,GACjB,EAA6B,GACnC,IAAW,IAAO,GACV,EAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACb,GAAc,EACpB,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAiB,KAAK,UAGtB,GAAS,OAAS,EACL,CAAA,EAAM,QAAQ,IAAI,IAD/B,CAAA,EAAA,UAEF,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAK,EAAK,IAAM,EAAO,GAAG,GAG5B,EAAA,QAAQ,+CAUyB,GACnC,GAAI,GAAQ,KACV,OAEF,OAAW,KAAO,IAChB,GAAM,GAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACnB,EAAM,WCjEZ,AAoBA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,QAAA,GAAA,YAFU,IAAA,IAAqB,KAM1B,GAAM,IAAyB,IA0BtC,GAAA,WAAA,aAEE,KAAA,eAAkC,KAMlC,SAAA,UAAA,UAAA,SAAU,GACR,KAAK,OAAS,GAGV,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,qFAEb,EAAA,UAAA,WAAN,SAAiB,qFASjB,EAAA,UAAA,SAAA,SAAS,KAGX,mBAqBE,WAAY,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,IAGtC,AAAI,GAAa,MACf,GAAY,IAEd,KAAK,UAAY,EACjB,KAAK,YAAc,EAGrB,SAAA,UAAA,OAAA,SAAO,GACL,KAAK,UAAU,KAAK,IAGtB,EAAA,UAAA,UAAA,SAAU,GACR,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,UAAU,KAIvB,EAAA,UAAA,SAAA,SAAS,GACP,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,SAAS,KAShB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,aAAN,SAAmB,uGACjB,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,KADO,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,WAAN,SAAiB,uGACf,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,KADS,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAI3B,KAOA,GAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,mBACE,GAAA,KAAA,OAAO,KAGH,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,OAAS,YAGV,EAAA,UAAA,WAAN,SAAiB,EAAe,yFAC9B,AAAI,GAAQ,MACV,GAAO,IAEH,EAAY,EAAK,MAAW,KAAO,EAAI,EAAK,KAClD,KAAK,MAAQ,aACF,GACT,GAAM,GAAQ,EAAK,GACnB,GAAI,MAAO,IAAU,SACnB,AAAK,EAAK,OAAO,eAAe,IAC9B,GAAK,OAAO,GAAO,GAErB,EAAK,OAAO,GAAO,EAAK,OAAO,GAAiB,EAAQ,OAExD,GAAI,GAAkB,OACtB,AAAI,IAAO,GAAK,OACd,EAAqB,EAAK,OAAO,GAEjC,EAAK,OAAO,GAAO,EAErB,GAAM,GACF,EAAA,KAAK,WAAM,MAAA,GAAA,IAAK,EAAK,OAAO,GAAO,EAAA,IAAI,EAAO,MAClD,EAAK,OAAO,GAAO,EACnB,AAAI,GAAsB,MACxB,EAAmB,mBAlBzB,IAAW,IAAO,KAAP,kBAwBP,EAAA,UAAA,WAAN,SAAiB,EAAe,2FAC9B,GAAI,GAAQ,KACV,eAAW,GACT,GAAI,EAAK,OAAO,IAAQ,sBAGxB,AAAI,MAAO,GAAK,OAAO,IAAS,SAC9B,EAAK,GAAO,EAAK,OAAO,GAAiB,EAAK,KAE9C,EAAA,KAAK,WACH,GAAM,GAAc,EAAA,IAAI,EAAA,IAAI,EAAG,EAAK,MAAO,EAAK,OAAO,IACvD,EAAK,GAAO,EACX,EAAK,OAAO,GAAgB,UAC7B,EAAA,KAAK,EAAK,cAXhB,EAAA,EAAkB,EAAA,KAAK,OAAO,QAAZ,EAAA,EAAA,OAAA,IAAP,EAAG,EAAA,KAAH,kBAiBjB,GA9DgC,mBAqEH,EAAA,EAAA,GAA7B,4DAIQ,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,MAAQ,GACb,KAAK,QAAU,YAGX,EAAA,UAAA,WAAN,SAAiB,EAAe,4EAC9B,AAAI,GAAQ,MACV,GAAO,IAET,KAAK,MAAM,KAAK,GAChB,IAAW,IAAO,GAChB,AAAI,KAAK,QAAQ,IAAQ,MACvB,MAAK,QAAQ,GAAO,IAEtB,KAAK,QAAQ,GAAK,KAAK,EAAK,mBAO1B,EAAA,UAAA,SAAN,6HACQ,EAA+D,GAC/D,EAAiB,GACjB,EAAoB,GAC1B,IAAW,IAAO,MAAK,QAErB,IADM,EAAa,KAAK,QAAQ,GACvB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,MAAO,GAAW,IAAO,UACrB,GAAc,EAAW,GAC/B,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAQ,KAAK,IAIJ,MAAA,CAAA,EAAM,QAAQ,IAAI,WACjC,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EAC7B,EAAkB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IACtD,EAAgB,UAChB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IAAM,EAAO,GAAG,mBAGpD,GA/C6B,mBA8DO,EAAA,EAAA,GAiBlC,WAAY,EAA0B,GAAtC,GAAA,GACE,EAAA,KAAA,OAAO,KAKP,GARM,EAAA,aAAe,EAIrB,EAAK,WAAa,GAAc,OAChC,AAAI,EAAK,aAAe,QACtB,GAAK,WAAa,IAEhB,EAAK,aAAe,SAAW,EAAK,SAAW,KACjD,KAAM,IAAI,OACN,mHAGN,MAAI,GAAA,KAAK,SAAS,EAAK,aAGrB,GAAK,UAAY,GACb,EAAK,UAAU,KAAK,GAAO,EAAK,aAEtC,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,MAAQ,EAAK,UAGd,SAAA,UAAA,UAAN,SAAgB,EAAe,EAAe,yGACtC,GAAgC,GAClC,KAAK,OAAS,KAChB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,MAAM,EAAO,EAAO,qBAEnC,SAAG,KAAK,EAAA,aACR,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAChC,MAAK,aAAe,EAChB,KAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,SACtB,EAAG,KAAK,EAAA,aAEV,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAC5B,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,QACtB,EAAG,KAAK,EAAA,aACH,AAAI,EAAA,KAAK,SAAS,KAAK,aAC5B,EAAG,KAAK,KAAK,UAAU,KAAK,aAAc,EAAO,IAEnD,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,mGACb,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,WAAtB,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,mGACX,MAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,SAAS,WAApB,EAAA,wCAGN,GA3GoC,gBAiHhC,EAEA,GAIF,GAHA,AAAI,GAAa,MACf,GAAY,IAEV,YAAqB,IACvB,MAAO,CAAC,GAEV,GAAI,MAAM,QAAQ,IAAc,EAAU,YAAc,IACtD,MAAO,GAGT,GAAM,GACF,GAAqB,GACzB,MAAO,GAAgB,IACnB,SAAA,GAAkB,MAAA,IAAI,IAAe,EAAgB,KAW3D,GAAA,IAAA,WAOE,cAcO,SAAA,4BAAP,SACI,EAAwB,GAC1B,EAAA,KAAK,OACD,GAAkB,GAAK,OAAO,UAAU,GACxC,WAAM,MAAA,sDACF,YAAW,KACnB,EAA4B,kBAAkB,GAC9C,AAAI,EAA4B,aAAa,IAAmB,MAC9D,GAA4B,aAAa,GAAkB,IAE7D,EAA4B,aAAa,GAAgB,KACrD,IAGS,EAAA,kBAAf,SAAiC,GAE/B,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAe,EAA4B,aAAa,CAAC,GAC/D,EAAa,QAAQ,SAAA,GACnB,GAAI,IAAS,EACX,KAAM,IAAI,GAAW,uCASZ,EAAA,MAAjB,WACE,EAA4B,aAAe,IAWtC,EAAA,gBAAP,SAAuB,GACrB,GAAM,GAA0C,GAChD,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAQ,CAAC,EACf,AAAI,GAAkB,GACpB,EAAa,KAAI,MAAjB,EAAqB,EAA4B,aAAa,IAGlE,MAAO,GAAa,IAAI,SAAA,GAAQ,MAAA,IAAI,MArEvB,EAAA,aAC6C,GAsE9D,iBAGI,EAA2B,EAAgC,EAC3D,EAAsB,EAAyB,EAC/C,EAAmB,EACnB,GACF,GAAM,GAAU,GAAI,IACd,EAAe,CACnB,GAAI,YAAiB,GAA4B,gBAAgB,IAEnE,AAAI,GAAa,MACf,EAAgB,KAAI,MAApB,EAAwB,GAE1B,EAAgB,KAAK,GACrB,GAAM,GAAe,GAAI,IAAa,GAMtC,SAAa,UAAU,CACrB,OAAM,EACN,aAAY,EACZ,QAAS,EACT,MAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,QAAS,IAEJ,CAAC,aAAY,EAAE,QAAO,GCnlB/B,YA2BI,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,QAAS,GChC9B,YAwB4B,EAAW,GACrC,MAAO,GAAA,KAAK,WACV,AAAI,EAAE,QAAU,WACd,GAAI,EAAE,OAAO,YAEf,GAAM,GAAY,EAAA,IAAQ,GAAS,GAAI,EAAM,IACvC,EAAgB,EAAA,KAAS,EAAU,MAAO,MAC1C,EAAO,EAAA,KAAS,EAAA,QAAY,EAAW,IAC7C,MAAO,GAAA,IAAQ,EAAG,iBAIW,EAAe,GAC9C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAO,IAAS,kBAG5B,EAAe,GAC/C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAO,IAAS,kBAIzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,EAAA,IAAQ,EAAO,GACtB,EACF,EAAA,YAAgB,EAAA,IAAQ,GAAQ,KAAW,OAAO,WAChD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAM,IACxC,MAAO,GAAA,IAAQ,IAAK,EAAA,KAAS,EAAW,mBAKxC,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAE9B,EAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAErC,MAAO,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAU,IAAa,kBAI/B,EAAe,GAC1C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,GAAS,GAAY,kBAInB,EAAe,GACnC,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,EAAW,kBAIE,EAAe,GAC9C,MAAO,GAAA,KAAK,WACV,GAAM,GAAM,EAAA,IAAQ,EAAA,IAAQ,EAAO,GAAQ,IACrC,EAAM,EAAA,IAAQ,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAQ,GAAQ,IACvD,MAAO,GAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAK,mBAY1B,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,KAAK,IAAI,GAChB,EAAiB,EAAA,IAAQ,EAAO,GAChC,EAAgB,EAAA,IAClB,EAAA,IAAQ,EAAgB,EAAA,SAAa,EAAA,IAAQ,GAAI,KACjD,GACJ,MAAO,GAAA,KAAS,EAAe,kBAK/B,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAI,EACF,EAAS,EAAA,QAAY,QAGrB,GAAM,GAAY,EAAA,IAAQ,EAAQ,EAAO,MAAM,OAAS,EAAG,IAC3D,EAAS,EAAA,IAAQ,EAAQ,GAE3B,SAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MACzC,EAAA,IAAQ,EAAA,IACX,EAAA,IAAQ,EAAO,UAAW,EAAA,IAAQ,IAAU,EAAO,MAAM,OAAS,kBActE,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAM,GAAa,EAAA,MAAU,GAAU,IAAS,QAChD,EAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MAChD,GAAM,GAAc,EAAO,MACrB,EACF,EAAA,OAAW,EAAY,EAAY,EAAY,OAAS,IACnD,QAAQ,GACjB,MAAO,IAAwB,EAAc,EAAQ,iBA0BrD,EAAgB,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAO,MAAO,EAAO,OACzC,KAAM,IAAI,GACN,8DACG,MAAK,UAAU,EAAO,OAAM,QAAQ,KAAK,UAAU,EAAO,SAEnE,MAAO,GAAA,KAAK,WAOV,GAAM,GAAa,EAAO,OACpB,EAAe,EAAO,MAAM,MAClC,MAAO,GAAW,IAAI,EAAO,IAAI,IAAS,IAAI,EAAa,MAAM,uBAIlC,EAAe,GAChD,MAAO,GAAA,KAAK,WACV,GAAI,GACJ,SAAI,EAAA,YAAgB,EAAO,KAAW,EAAI,MAC1C,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAG,KAC3B,EAAA,KAAS,GAA8B,EAAO,GAAI,kBAKzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,GAChD,EAAc,EAAA,YAAgB,EAAO,KAAW,GACtD,MAAO,GAAA,IACH,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAA,IAAQ,EAAa,KAAgB,kBAI5C,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,EAAA,IAAQ,EAAA,IAAQ,KAAW,IAC3C,MAAO,GAAA,KAAS,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAO,IAAW,kBAI7B,EAAe,GAC7C,MAAO,GAAA,KAAK,WACV,GAAM,GAAiB,GAAY,EAAO,IACpC,EAAiB,GAAY,EAAO,IACpC,EAAY,EAAA,IAAQ,EAAgB,GAC1C,MAAO,GAAA,IAAQ,EAAA,IAAQ,EAAW,OAkB/B,GAAM,IAAsD,CACjE,iBAAgB,GAChB,kBAAiB,GACjB,4BAA2B,GAC3B,4BAA2B,GAC3B,aAAY,GACZ,MAAK,GACL,iBAAgB,GAChB,QAAO,GACP,wBAAuB,GACvB,8BAA6B,GAC7B,mBAAkB,GAClB,0BAAyB,GACzB,QAAO,GACP,gBAAe,gBAKG,GAClB,GAAI,MAAO,IAAmB,UAC5B,GAAI,IAAkB,IACpB,MAAO,IAAU,GAEnB,GAAI,GAAS,gBAAgB,EAC7B,KAAI,GAAe,cAAc,SAAS,wBACxC,GAAS,gBAAgB,EAAc,wFAInC,GAAI,GAAW,OAErB,OAAO,GC7QX,YAyB+B,EAAe,GAC5C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,IAAQ,GAAI,EAAA,SAAa,IACrC,EAAmB,GAAO,EAAA,QAAY,EAAO,GAAY,EAAM,OACrE,MAAO,GAAA,KAAS,EAAA,MAAU,EAAO,GAAmB,kBAIpB,EAAe,GACjD,MAAO,GAAA,KACH,WAAM,MAAA,IACF,EAAA,MAAU,EAAA,OAAW,EAAO,IAAK,EAAA,OAAW,EAAO,KAAM,aAGnE,YAAuB,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,yBAI3C,EAAe,GACvC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIS,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIqB,EAAe,GAChD,MAAO,IAAuB,EAAO,eAInC,EAAe,GACjB,MAAI,GAAM,OAAS,EAAM,MACvB,GAAQ,EAAM,QAAQ,CAAC,EAAM,KAAO,KAEtC,EAAQ,EAAM,OAAO,IACrB,AAAI,EAAM,QAAU,EAAM,OACxB,GAAQ,EAAM,OAAO,EAAM,QAEtB,EAAA,MAAU,EAAO,GAAO,OAAO,WAajC,GAAM,IAAM,GACN,GAAM,GACN,GAAM,GACN,GAAM,GACN,GAAO,GACP,GAAO,GACP,GAA0B,GAC1B,GAAS,GACT,GAAgC,GAIhC,GAAuD,CAClE,eAAc,GACd,oBAAmB,GACnB,UAAS,GACT,wBAAuB,GACvB,8BAA6B,GAC7B,IAAG,GACH,IAAG,GACH,IAAG,GACH,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,OAAM,gBAGY,GAClB,GAAI,MAAO,IAAe,UAAY,IAAc,IAClD,MAAO,IAAW,GACb,GAAI,MAAO,IAAe,UAAY,GAAc,KACzD,MAAO,GAEP,KAAM,IAAI,GAAW,kBAAkB,eAqBP,GAElC,GADA,GAAY,IAAO,KAAM,0BAA0B,GAC/C,MAAO,IAAO,SAChB,MAAO,GAGP,OADI,GAAM,OACQ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAU,KAAS,GACrB,EAAS,EACT,OAGJ,GAAI,IAAW,OACb,MAAO,GAET,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAW,KAAS,GACtB,EAAS,EACT,OAGJ,MAAI,KAAW,OACN,EAED,EAAgB,KCzL5B,YAwB6B,GAC3B,GAAM,GAA2D,CAC/D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,MAC/B,SAAY,WAAM,MAAA,GAAA,MAAM,SAAS,EAAG,IAAM,OAC1C,KAAQ,WAAM,MAAA,GAAA,MAAM,KAAK,KAAO,GAAK,KAAO,OAC5C,OAAU,WAAM,MAAA,GAAA,MAAM,OAAO,KAAO,GAAK,KAAO,KAAW,IAC3D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,KAAO,GAAK,EAAG,OAC9C,IAAO,WAAM,MAAA,GAAA,MAAM,IAAI,OASzB,GAPA,EAAa,QAAa,EAAa,QACvC,EAAa,SAAc,EAAa,SACxC,EAAa,KAAU,EAAa,KACpC,EAAa,OAAY,EAAa,OACtC,EAAa,QAAa,EAAa,QACvC,EAAa,IAAS,EAAa,IAE/B,IAAc,GAChB,MAAO,GAAa,KAEtB,KAAM,IAAI,GAAW,qBAAqB,GC3C5C,AAeO,GAAM,IAA8C,EAAI,KAAO,iBAelE,EAAyB,EAAmB,GAC9C,GAD8C,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAuB,MACvB,MAAO,IAAwB,UAC/B,OAAO,eAAe,KAAyB,OAAO,WACtD,CAAC,GAAiB,GACpB,KAAM,IAAI,OACN,sEAGN,GAAI,GACF,GAAM,GAAM,KAAK,UAAU,GAC3B,AAAI,EAAI,OAAS,IACf,QAAQ,KACJ,mCAAmC,EAAS,qBAC5C,iBAAgB,EAAI,OAAM,iCAC1B,kHAEG,IAA2C,mBAgBvB,GAC/B,GAAI,IAAM,KAER,MAAO,GACF,GAAI,MAAO,IAAM,SACtB,GAAI,OAAO,eAAe,KAAO,OAAO,WAGtC,OADM,GAAO,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,MAAO,IAAQ,SAEjB,MAAO,GAET,GAAI,CAAC,GAAiB,EAAE,IACtB,MAAO,GAGX,MAAO,WAGH,MAAM,QAAQ,IAEhB,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,GAAiB,GACpB,MAAO,GAGX,MAAO,OAKP,OAAO,QAKX,GAAM,GAAQ,MAAO,GACrB,MAAO,KAAU,UAAY,IAAU,UAAY,IAAU,WCrGjE,YA8BI,EAAkB,EAAqB,EAEvC,GAAA,AAAA,IAAA,QAAA,GACI,QAAQ,KACd,GAAM,GAAiB,GAAsB,GAGvC,EAAsB,CAAC,eAAgB,eAAgB,WAC7D,AAAI,EACF,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAEtC,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAAM,IAI9C,AAAI,EAAU,EAAU,OAAS,IAAM,GAErC,GAAY,EAAU,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAa,MAGzD,GAAI,GACJ,GAAI,CAAC,GACH,EAAU,KAAK,mBACf,EAAgB,GAChB,OAAW,KAAS,GAAM,aACxB,EAAc,KAAI,MAAlB,EAAsB,EAAM,aAAa,IAI7C,EAAQ,IAAI,OAAO,IACnB,GAAS,EAAW,EAAW,GAC/B,EAAQ,IAAI,OAAO,IAGnB,OADM,GAAS,EAAM,OACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EACF,GAAkB,EAAO,GAAI,EAAW,GAExC,GACI,EAAO,GAAI,EAAW,EAAe,GAE3C,EAAS,KAAM,EAAO,OAAS,EAAI,IAAM,KAAK,OAAO,IAItD,EAAc,mCAEf,GAAM,GAAiB,GAAqB,GACtC,EAAoB,GAAqB,EAAM,qBAErD,EAAQ,iBAAiB,GAAiB,IAC1C,EAAQ,qBAAqB,GAC7B,EAAQ,yBAAyB,GACjC,EAAQ,IAAI,OAAO,IAGrB,YAA8B,GAC5B,GAAI,GAEJ,MAAK,GAAc,2BAA6B,KAC9C,EACI,GAAsB,EAAc,2BAExC,EAAiB,GAAqB,EAAM,kBAGvC,EAGT,YAA+B,GAC7B,GAAI,GAAiB,GACf,EAAyB,GACzB,EAAgB,GACtB,OAAW,KAAS,GAAM,aACxB,EAAa,KAAK,EAAM,aAAa,IAEvC,OAAyB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,EAAW,OAAS,GACpB,EAAW,SAAW,GAAK,EAAW,GAAG,cAAc,OAAS,GAClE,EAAiB,GACjB,MAEF,EAAM,KAAI,MAAV,EAAc,GAEhB,GAAI,EAEF,OAAoB,GAAA,EAAA,EAAA,EAAM,OAAN,EAAA,EAAA,OAAA,KAElB,OAFS,GAAK,EAAA,GACV,EAAO,GACQ,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAM,QAAQ,KAAU,GAC1B,GAAI,GACF,EAAiB,GACjB,UAEA,GAAO,GAIb,GAAI,CAAC,EACH,MAIN,MAAO,GAGT,YACI,EAAkB,EAElB,GAAA,AAAA,IAAA,QAAA,GAA6D,QAAQ,KAEvE,OADI,GAAO,GACF,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAI,GACN,GAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAAK,KAE1C,GAAQ,EAAO,GACf,EAAO,EAAK,MAAM,EAAG,EAAU,IAC/B,GAAQ,IAAI,OAAO,EAAU,GAAK,EAAK,QAEzC,EAAQ,GAQV,YACI,EAAc,EAEd,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAGhB,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EACF,CAAI,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,YAChE,GAAS,EAAQ,EAAW,GAM9B,YACI,EAAc,EAAqB,EAEnC,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAIhB,OADM,GAAwB,GACX,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,GAAiB,MAAQ,EAAc,OAAS,GAChD,EAAc,QAAQ,KAAU,GAClC,SAEF,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAAG,KACrC,EAAoB,EAAK,YAAY,GACrC,EAAqB,EAAK,cAAc,GAC9C,EAAY,KACL,EAAY,IAAI,EAAiB,KAAK,EAAkB,MAGnE,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EAAkB,EAAY,SAAW,EAAI,GAAK,EAAY,GAC9D,EAAmB,CACpB,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,WAC3D,GAGF,GAAS,EAAQ,EAAW,GAC5B,OAAS,GAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,EACxC,GAAS,CAAC,GAAI,GAAI,GAAI,EAAY,IAAK,EAAW,GCvNtD,AA8BA,YACI,EAAa,EAAe,GAC9B,MAAQ,KAAQ,gBAAkB,IAAQ,gBAClC,IAAQ,gBACZ,IAAU,GAAK,MAAO,IAAU,qBAUlC,EAA6B,GAC/B,GAAI,IAAmB,KACrB,MAAO,MACF,GAAI,MAAO,IAAmB,SACnC,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAmB,UAC1B,MAAO,IAAmB,UAC7B,MAAO,GACF,GAAI,YAA0B,QAGnC,OAFM,GAAU,GACV,EAAc,EAAe,OAC1B,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAe,GAC5B,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACf,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAArB,GAAM,GAAW,EAAA,GACd,EAAgB,EAAe,GACrC,GAAI,IAAgB,QAAU,MAAO,IAAkB,SAIrD,EAAO,GAAe,OAEtB,GAAM,GAAQ,GAA0B,GACxC,EAAO,GAAS,GAAoB,EAAe,IAGvD,MAAO,gBAWP,EAAyC,GAC3C,GAAI,GAAa,KACf,MAAO,MACF,GAAI,MAAO,IAAa,SAC7B,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAa,UAAc,MAAO,IAAa,UACzD,MAAO,GACF,GAAI,YAAoB,QAG7B,OAFM,GAAU,GACV,EAAc,EAAS,OACpB,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAS,GACtB,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACrB,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAU,EAAS,GACnB,EAAQ,GAA0B,GACxC,AAAK,KAAU,QAAU,IAAU,cAC/B,MAAO,IAAY,SAIrB,EAAO,GAAS,EAEhB,EAAO,GAAS,GAAoB,EAAS,GAGjD,MAAO,IC7HX,GAGM,IAAU,QCHhB,AA0BA,YAAiC,EAAqB,GAEpD,GAAI,EAAI,OAAS,MAAQ,EAAI,QAAU,EAAI,MAEzC,MAAO,GAET,IAEE,MAAO,GAAA,KAAK,EAAK,EAAI,aACd,GAEP,KAAM,IAAI,GACN,0BAA0B,EAAI,MAAK,kCACnC,gBAAe,EAAI,KAAI,MAAM,EAAI,MAAK,QAgB9C,GAAA,IAAA,WAUE,WAAY,GACV,GAVM,KAAA,SAAmC,GACnC,KAAA,QAAkC,GAClC,KAAA,QAAoC,GAQtC,YAAiB,GACnB,OAAW,KAAM,GAAM,SACrB,KAAK,SAAS,GAAM,EAAM,SAAS,GACnC,AAAI,IAAM,GAAM,SACd,MAAK,QAAQ,GAAM,EAAM,QAAQ,SAIrC,GAAI,GAAS,KACX,OAEF,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,KAAK,IAAI,EAAK,IAAK,EAAK,SAe9B,SAAA,UAAA,IAAA,SAAI,EAAqB,EAAe,GACtC,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAK,SAAS,EAAI,IAAM,GAAwB,EAAK,GACrD,KAAK,QAAQ,EAAI,MAAQ,EAAI,GAC7B,AAAI,GAAQ,MACV,MAAK,QAAQ,EAAI,IAAM,OAGzB,MAAM,IAAI,GAAW,uBAAuB,EAAI,KAAI,QAAQ,EAAI,IAElE,MAAO,OAQT,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,IAAI,EAAK,IAAK,EAAK,QAO1B,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,MAAK,SAAS,EAAI,KAAO,MAMlC,EAAA,UAAA,MAAA,WACE,MAAO,QAAO,KAAK,KAAK,UAU1B,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,SAAS,EAAI,SAG3B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,SAAS,KAWzB,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,QAAQ,EAAI,SAG1B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,QAAQ,KAKxB,EAAA,UAAA,aAAA,WACE,AAAI,KAAK,SAAW,MAClB,EAAA,QAAQ,KAAK,UAGnB,KAIM,GAA+D,GAG/D,GAC4D,eA4C9D,EAA0C,EAC1C,EAAiB,GAWnB,OATM,GAAoB,GAAU,KAAO,GAAQ,EAAO,SAEpD,EAAe,MAAM,QAAQ,GAC7B,EACF,EAAe,EAA8B,CAAC,GAE5C,EAAc,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,OACpC,EAAyB,GACzB,EAAY,EAAS,QACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,AAAI,EAAU,QAAQ,KAAgB,GACpC,EAAa,KAAK,EAAS,SAAS,IAEpC,EAAa,KAAK,MAItB,AAAI,GAAS,MAEX,GAAM,cAAgB,UACtB,EAAM,cAAgB,UAIxB,GAAM,GACF,EAAY,KAAK,KAAO,IAAM,EAAS,QAAQ,KAAK,KACpD,EACA,EACJ,GAAI,GAAa,IAAoB,MAGnC,GAAM,GAAM,GAAqC,EAAY,GAC7D,EAAS,EAAI,OACb,EAAkB,EAAI,gBAGtB,GAAa,GAAmB,EAChC,GAAsB,GAAmB,EAE3C,EAAS,GAAa,GACtB,EAAkB,GAClB,AAAK,GACH,OAAO,OAAO,EAAiB,GAAsB,IAMvD,OAHM,GAAmB,GAAI,IAAS,GAG7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAI,GAAS,MAEX,GAAM,GAAa,EAAA,SAAS,WAC5B,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAExB,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAI1B,GAAM,GAAW,EAAO,GAClB,EAAW,EAAS,YAC1B,GAAI,YAAoB,IACtB,SAOF,OALM,GAAwB,GACxB,EAAuB,GACvB,EAA6B,GAE/B,EAAa,GACG,EAAA,EAAA,EAAA,EAAS,OAAT,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAiB,SAAS,GAClC,EAAO,EAAiB,QAAQ,GACtC,EAAY,KAAK,GACjB,EAAW,KAAK,GAChB,AAAI,GAAQ,MACV,GAAa,IAEf,AAAK,GACH,GAAgB,EAAM,QACtB,AAAI,EAAgB,EAAM,QAAU,GAAK,CAAC,EAAS,OAAO,IACtD,EAAY,QAAQ,EAAM,QAAU,IAAM,CAAC,EAAM,YACjD,EAAM,YAAY,WAAa,IACjC,EAAiB,KAAK,IAK5B,AAAI,GACF,GAAS,GAAU,GACnB,EAAO,KAAU,EAAW,IAE9B,GAAM,GACF,GAAO,EAAS,MAAM,EAAa,IACnC,EAA8B,KAClC,AAAI,EAAS,iBACX,GAAa,EAAS,YAAY,EAAa,IAKjD,OAHM,GAAe,GAAe,GAC9B,EACF,MAAM,QAAQ,GAAgB,EAAe,CAAC,GACzC,EAAI,EAAG,EAAI,EAAsB,OAAQ,EAAE,GAClD,AAAK,EAAiB,OAAO,EAAsB,KACjD,EAAiB,IACb,EAAsB,GAAI,EAAc,GACxC,MAAM,QAAQ,GAAc,EAAW,GAAK,GAElD,GAAM,GAAQ,EAAY,QAAQ,EAAsB,GAAG,MAC3D,AAAI,IAAU,IACZ,GAAa,GAAS,EAAc,IAIxC,AAAK,GAEH,EAAA,QAAQ,GAQZ,SAAiB,eAEV,EAAe,EAAe,EAAa,GAsBpD,YACI,EAA2B,GAE7B,EAAA,KAAK,OACD,GAAW,MAAQ,EAAQ,OAAS,EACpC,WAAM,MAAA,0CAEV,GAAI,GAAgC,GAChC,EAAkC,GACtC,GAAI,EAAQ,SAAW,GAErB,GAAM,GACF,GAAgD,EAAQ,GAAI,GAChE,EAAc,EAAI,OAClB,EAAoB,EAAI,iBAGxB,QADM,GAAU,GAAI,KACA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAKlB,OALS,GAAK,EAAA,GACR,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,aAIc,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAxB,GAAM,GAAc,EAAA,GACvB,AAAK,EAAQ,IAAI,EAAe,OAC9B,GAAY,KAAK,GACjB,EAAQ,IAAI,EAAe,sBAKpB,GACT,AAAI,EAAkB,IAAS,MAC7B,GAAkB,GAAQ,GAAI,MAEhC,EAAa,GAAM,QACf,SAAA,GAAa,MAAA,GAAkB,GAAM,IAAI,MAL/C,OAAW,KAAQ,KAAR,GASf,MAAO,CACL,OAAQ,EACR,gBAAiB,GAAoB,IAIzC,YAA6B,GAC3B,GAAM,GAAmC,GACzC,OAAW,KAAQ,GACjB,EAAgB,GAAQ,EAAa,GAAM,KAE7C,MAAO,eAeL,EAAuB,GASzB,OAPM,GAAU,GAAI,KACd,EAA2B,GAC3B,EAA6B,GAKjB,EAAA,EAAA,EAAA,EAAS,QAAT,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAQ,IAAI,GAGd,GAAM,GAA0B,GAC1B,EAAkB,GAKxB,IAFA,EAAM,KAAK,GAEJ,EAAM,OAAS,IACpB,GAAM,GAAM,EAAM,EAAM,OAAS,GACjC,GAAI,EAAQ,IAAI,EAAI,OAClB,EAAM,MACN,SAEF,GAAM,GAAc,EAAM,EAAM,OAAS,KAAO,EAAM,OAAS,EAC/D,GAAI,EAAI,OAAO,SAAW,GAAK,EAE7B,EAAM,MACN,EAAO,KAAK,GACZ,EAAQ,IAAI,EAAI,MAChB,AAAI,GACF,EAAM,WAKR,EAAM,KAAK,EAAM,OAAS,GAC1B,OAAoB,GAAA,EAAA,EAAA,EAAI,OAAJ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GAQd,GALA,AAAI,EAAa,EAAM,OAAS,MAC9B,GAAa,EAAM,MAAQ,GAAI,MAEjC,EAAa,EAAM,MAAM,IAAI,EAAI,MAE7B,EAAQ,IAAI,EAAM,MACpB,SAEF,EAAM,KAAK,KAIjB,MAAO,CAAC,OAAM,EAAE,aAAY,GAS9B,YAAwB,GAEtB,GAAI,GACJ,GAAI,EAAM,YAAY,aAAa,SAAW,EAC5C,EAAe,EAAM,YAAY,YAGjC,OADI,GAAoB,KACf,EAAI,EAAG,EAAI,EAAM,YAAY,aAAa,OAAQ,EAAE,EAC3D,OAA2B,GAAA,EAAA,EAAA,EAAM,YAAY,aAAa,GAChD,cADiB,EAAA,EAAA,OAAA,KAAtB,GAAM,GAAY,EAAA,GAErB,GAAI,EAAa,KAAO,EAAM,IAC5B,EAAY,EACZ,OAIN,EAAe,EAAM,YAAY,YAAY,GAE/C,MAAO,GCjhBT,AA6CA,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAoCtC,WAAY,GAAZ,GAAA,GAEE,EAAA,KAAA,KAAM,KAAG,KAET,GAtBF,EAAA,eAAiB,GAAI,KAqBnB,EAAK,KAAO,EAAK,KACb,EAAK,MAAQ,MACf,GAAM,GAAS,EAAK,eAAe,cACnC,EAAK,KAAO,GAAO,GAqBrB,GAlBA,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAKlB,AAAI,MAAM,QAAQ,EAAK,QACrB,EAAK,OAAS,EAAK,OAAO,QAE1B,EAAK,OAAS,CAAC,EAAK,QAEtB,AAAI,MAAM,QAAQ,EAAK,SACrB,EAAK,QAAU,EAAK,QAAQ,QAE5B,EAAK,QAAU,CAAC,EAAK,SAInB,GAAqB,EAAK,QAAQ,SAAW,EAAK,OAAO,OAC3D,KAAM,IAAI,GACN,mGAEA,IAAG,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,SAIhC,AAAI,GAAqB,EAAK,SAAS,SAAW,EAAK,QAAQ,QAC7D,QAAQ,KACJ,qGAEA,IAAG,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,SAOjC,EAAK,YAAc,GACnB,EAAK,uBAAyB,GAC9B,EAAK,yBAA2B,GAKhC,EAAK,aAAe,GACpB,EAAK,wBAA0B,GAC/B,EAAK,0BAA4B,GAKjC,EAAK,OAAS,GAMd,EAAK,sBAAwB,GAe7B,OAAgB,GAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YACtB,EAAK,aAAa,KAAK,GACvB,EAAK,wBAAwB,KAAK,GAClC,EAAK,0BAA0B,KAAK,GAMtC,OAAgB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YAKtB,GAAqB,IAAc,EAAG,4BACtC,GAAqB,IAAgB,EAAG,8BACxC,EAAK,YAAY,KAAK,GACtB,EAAK,uBAAuB,KAAK,GACjC,EAAK,yBAAyB,KAAK,GAIrC,EAAK,WAAa,GAClB,EAAK,YAAc,GACnB,EAAK,gBAAkB,GACvB,EAAK,eAAiB,GACtB,EAAK,gBAAkB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,EAAK,YAAY,GAE/B,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,WACN,6DACA,qBAAoB,EAAK,OAAM,MAC/B,UAAS,EAAC,0BACV,oBAAmB,EAAM,eAAc,MAE7C,EAAK,WAAW,KAAK,EAAM,MAC3B,EAAK,gBAAgB,KAAK,EAAM,iBAEhC,EAAK,eAAe,KAAK,EAAM,MAEjC,OAAoB,GAAA,EAAA,EAAA,EAAK,aAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,YAAY,KAAK,EAAM,MAG9B,EAAK,oBAAsB,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QAClD,EAAK,qBAAuB,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,QAwFpD,OAjFM,GAA0C,GAE1C,EAAyC,GACzC,EAA4C,GAE5C,EAA6C,GAC7C,EAA4C,GAC5C,EAAiC,GAoBjC,EACF,SAAC,GAAwB,GAAuB,GAC/C,GAAe,GAAoB,IAClC,AAAI,KAAS,MAAQ,IAAa,MAAQ,IAAe,OACvD,IAAQ,GAAO,YACf,GAAY,GAAO,UACnB,GAAc,GAAO,aAEvB,GAAM,IAAO,GAAM,aAAa,IAGhC,GAAI,GAAgB,QAAQ,MAAU,GACpC,KAAM,IAAI,IACN,cAAc,GAAO,KAAI,cAAc,GAAM,KAAI,yBAKvD,GAAI,GAAc,QAAQ,MAAU,GAClC,OAIF,EAAK,eAAe,IAAI,EAAU,QAAQ,GAAO,KAGjD,AAAM,GAAM,KAAM,IAChB,GAAa,GAAM,IAAM,OAAO,KAAK,GAAc,QAGrD,AAAI,GAAgB,QAAQ,MAAU,IACpC,GAAgB,KAAK,IAKvB,OADM,IAAmB,GAAK,cAAc,OACnC,GAAI,EAAG,GAAI,GAAkB,MACpC,GAAM,IAAI,GAAK,aAAa,IACtB,GAAQ,GAAK,cAAc,IAC3B,GAAY,GAAK,YAAY,IAC7B,GAAc,GAAK,cAAc,IACvC,EACI,GAAG,GAAe,GAAiB,GAAO,GAC1C,IAGN,IADA,GAAc,KAAK,IACZ,GAAgB,QAAQ,KAAS,GACtC,GAAgB,OAAO,GAAgB,QAAQ,IAAO,GAExD,EAAuB,KAAK,KAG5B,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAgB,EAAG,EAAe,GAKpC,OAFM,GACF,EAAuB,QAAQ,UAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAa,EAAK,IAAM,EAExB,AAAM,EAAK,KAAM,IACf,GAAY,EAAK,IAAM,GAEzB,GAAI,GAAQ,EAAY,EAAK,IAGvB,EACD,EAAa,EAAK,cAAc,KAAO,KACnC,EACA,EAAa,EAAK,cAAc,IAOzC,EAAQ,KAAK,IAAI,EAAO,GACxB,EAAa,EAAK,cAAc,IAAM,EACtC,EAAe,EAAK,cAAc,IAAM,EAAK,cAC7C,EAAY,EAAK,IAAM,EAGvB,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAa,aAAa,GACxC,EACD,EAAY,EAAY,KAAO,KAAO,EACA,EAAY,EAAY,IACnE,EAAY,EAAY,IAAM,KAAK,IAAI,EAAQ,EAAG,GAClD,EAAa,EAAY,IAAM,GAKnC,GAAM,GAA0C,GAChD,OAAW,KAAU,IACnB,GAAM,GAAQ,EAAY,GAC1B,AAAM,IAAS,IACb,GAAa,GAAS,IAExB,EAAa,GAAO,KAAK,EAAa,IAIxC,GAAM,IAA4C,GAClD,OAAW,MAAW,IACpB,GAAM,GAAQ,EAAa,IAC3B,AAAM,IAAS,KACb,IAAc,GAAS,IAEzB,GAAc,GAAO,KAAK,EAAe,KAI3C,GAAI,IAAY,OAAO,KAAK,IACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAG1B,EAAK,OAAS,GACd,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACR,GAAiB,GAAc,GAGrC,GAAe,KAAK,SAAC,GAAG,IACtB,GAAM,IAAS,EAAa,GAAE,IACxB,GAAS,EAAa,GAAE,IAC9B,MAAI,IAAS,GACJ,GAEL,GAAS,GACJ,EAEF,IAET,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACd,AAAI,YAAiB,IACnB,EAAK,sBAAsB,KAAK,GAElC,EAAK,OAAO,KAAK,IAGrB,EAAK,cAAgB,GAGrB,GAAY,OAAO,KAAK,GACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAStB,OAJM,IAAoB,EAAK,OAAO,QAGhC,GAAoC,GACtB,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,KAClB,OADS,GAAK,GAAA,IACK,GAAA,EAAA,GAAA,EAAa,GAAb,GAAA,GAAA,OAAA,MAAd,GAAM,GAAI,GAAA,IACP,EAAQ,EAAK,cACnB,GAAI,GAAS,MACX,OAAgB,IAAA,EAAA,GAAA,EAAK,aAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAI,GAAkB,QAAQ,KAAO,GACnC,KAAM,IAAI,IACN,sDAAsD,EACtD,eAAc,EAAM,KAAI,OACxB,uDACA,WAAU,KAGlB,OAAgB,IAAA,EAAA,GAAA,EAAK,cAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAkB,KAAK,GAEzB,GAAwB,KAAK,EAAM,OAMzC,EAAK,aAAe,EAKpB,OADM,IAAW,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,mBAC7B,IACT,GAAM,IAAiB,GAAS,OAAO,SAAA,IAAK,MAAA,MAAM,KAAM,OACxD,GAAI,KAAmB,EACrB,KAAM,IAAI,IACN,aAAa,GAAI,aAAa,GAAc,uEAE5C,KAAK,UAAU,MANJ,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAd,GAAM,IAAI,GAAA,OAAJ,IAcX,SAAK,cAAgB,GAErB,EAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,EAAK,OACnB,cAAe,EAAK,QACpB,WAAY,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,QACjC,YAAa,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,QACnC,YAAa,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QACpC,aAAc,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,UAExC,EAAK,MAAQ,GACb,EAAK,UAAY,IAGT,SAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,cAAc,KAAK,KAAI,2BA8B3C,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,GAAM,GACc,CAAC,qBAAsB,KAAM,qBAAsB,GACvE,GAAI,EAAE,KAAK,YAAc,GACvB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAO,sBAAwB,EAAM,UAAU,qBAKjD,OAAwB,GAAA,EAAA,EAAA,KAAK,sBAAL,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAO,sBAAwB,EAAU,UAAU,sBAGvD,SAAO,qBAAuB,KAAK,UAC5B,GAGT,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,OAAO,QAAQ,SAAA,GAEhB,EAAc,kBACX,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,MAElC,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WAIE,GAAI,KAAK,kBAAkB,OAAS,EAClC,KAAM,IAAI,GACN,wNAMN,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADI,GAA2B,GACX,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAU,EAAQ,OAAO,EAAM,kBAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACb,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAQ,KAAI,MAAZ,EAAgB,EAAM,qBAExB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACtB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAiB,KAAI,MAArB,EAAyB,EAAM,kBAEjC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAkB3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,IAGnC,OAFM,GAAgD,GAClD,EAAoB,EACJ,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACO,EAAA,EAAA,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,EAAa,EAAO,eAAiB,KACvC,KAAM,IAAI,GAAW,0BAA0B,EAAO,cAExD,EAAa,EAAO,cAAgB,EACpC,IAIJ,GAAM,GAAoD,GAC1D,OAAW,KAAQ,IAIjB,GAAI,GAAgB,EACpB,GAAI,EAAa,IAAS,MACxB,GAAM,GAAS,EAAK,MAAM,KACpB,EACF,EAAO,MAAM,EAAG,IAAI,OAAO,CAAC,EAAO,EAAO,OAAS,KACvD,EAAgB,EAAiB,KAAK,KAExC,GAAI,EAAa,IAAkB,KACjC,EAAkB,KAAK,CAAC,EAAa,GAAgB,EAAQ,aACpD,EACT,KAAM,IAAI,GACN,gDAAgD,GAEtD,MAAO,GAAa,GAGtB,GAAI,GAEF,GAAM,GAAuB,GAC7B,OAAW,KAAQ,GACjB,EAAW,KAAK,GAElB,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,GACH,EAAW,OAAM,OAChB,EAAiB,yBACrB,IAAG,IAIX,GAAc,IAON,EAAA,UAAA,cAAV,WACE,GAAM,GAAY,KAAK,YACjB,EAAwC,GAC9C,SAAY,UAAe,KAAK,eAChC,EAAY,OAAY,EACxB,EAAY,aAAkB,eAAe,GAG7C,EAAY,QAAa,gBAClB,GAeT,EAAA,UAAA,OAAA,SAAO,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IACnB,GAAM,GAAc,GAAoB,KAAK,iBAC7C,MAAO,GAAe,KAAK,UAAU,GAAe,GAgBtD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAE9B,OADM,GAAW,GAAI,IACZ,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,EAAK,OAAO,GAAI,EAAO,IAEtC,MAAO,IAAQ,EAAK,QAAS,EAAU,MAa3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAC9B,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAA2B,KAAM,EAAO,QAEhD,EAAQ,GAAqB,GAGxB,EAAK,iBAAiB,EAAQ,GAAO,MAahD,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAM,GAAc,GAA+B,GACnD,GAAI,EAAY,SAAW,KAAK,YAAY,OAC1C,KAAM,IAAI,GACN,+BAA+B,EAAU,KACzC,cAAa,KAAK,YAAY,OAAM,oBAK1C,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,EAAY,OAAQ,KACtC,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAa,EAAY,GAGzB,EAAW,EAAM,KAAO,OAC9B,EAAqB,GAAY,EAGnC,GAAM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,GAAK,MAAA,UAAS,EAAG,MACrB,KAAK,IAE5B,GAAI,EAAU,OAAS,EACrB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GAEP,EAAQ,EAAK,cACnB,GAAI,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,QAAQ,EAAM,MAAQ,GAExD,SAIF,OADM,GAAuB,GACpB,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAc,EAAa,KAAI,IAAI,EAAS,IAAI,EAChD,EAAa,EAAqB,GACxC,EAAY,KAAK,GAQnB,OALM,GAAc,EAAM,mBACtB,GAA+B,IAE7B,EAAe,GAA+B,GAC9C,EAAY,EAAM,aAAa,QAAQ,GACpC,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAqB,GAAY,EAAa,IAStD,OAFM,GAAwB,GACxB,EAA4B,GACzB,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GACzC,EAAc,KAAK,0BAA0B,GAC7C,EAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAgB,KAAK,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAgB,OAAQ,KAC1C,GAAM,GAAM,EAAgB,GAC5B,GAAqB,IAAO,IAC5B,EAAa,KAAK,EAAqB,IAIzC,MAAO,IAA+B,IAa9B,EAAA,UAAA,iBAAV,SAA2B,EAAkB,GAE3C,AAAI,GAAS,MACX,GAAQ,GAA2B,KAAM,EAAO,SASlD,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,GACxC,GAAM,GAAI,KAAK,OAAO,GAChB,EAAI,EAAO,GACX,EAAO,EAAM,GACnB,EAAU,EAAE,IAAM,CAAC,EAAG,GAMxB,OAHM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAUjB,OAVS,GAAI,EAAA,GAEP,EAAQ,EAAK,cACb,EAAwB,EAAK,aAC7B,EAAyB,EAAK,cAK9B,EAAe,GAAI,OACT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAE,KAAM,IACV,EAAa,KAAK,EAAU,EAAE,KAGlC,GAAI,EAAa,SAAW,EAAsB,QAEhD,GAAI,GAAiB,GACjB,EAAe,OACf,EAAa,OACb,EAAa,OACb,EAAW,OAKf,GAHA,AAAI,EAAK,UAAY,MACnB,GAAS,EAAK,UAEZ,EAAa,SAAW,GACpB,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GACvB,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAgB,IACpD,EAAc,GACV,EAAM,YAAY,EAAgB,IACtC,EAAkB,CAAC,GACnB,EAAgB,CAAC,OAEjB,GAAkB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KAC1C,EAAgB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KACxC,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAiB,IACrD,EAAc,GACV,EAAM,YAAY,EAAiB,IAGzC,GAAI,EAAM,oBACR,KAAM,IAAI,IACN,yHAMN,OAAS,GAAI,EAAG,EAAI,EAAuB,OAAQ,EAAE,GACnD,GAAM,GAAI,EAAuB,GAC3B,EAAI,EAAc,GAClB,EAAO,EAAY,GACzB,EAAU,EAAE,IAAM,CAAC,EAAG,KAS9B,OAHM,GAA0B,GAC1B,EAAwB,GACxB,EAAwB,GACd,EAAA,EAAA,EAAA,KAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,GACI,EAAE,KAAM,GAAW,4BAA4B,EAAE,KAAI,MAAM,EAAE,IAC3D,GAAA,GAAA,EAAA,EAAA,IAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAa,KAAK,EAAO,OACzB,EAAc,KAAK,GACnB,EAAY,KAAK,GAInB,MAAO,CAAC,EAAe,EAAa,IAW9B,EAAA,UAAA,uBAAR,SAA+B,GAG7B,OAFM,GAAiD,GACnD,EACgB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,YAAiB,GAAY,EAAI,EAC7C,OAAS,GAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,KAAK,eAAe,IAAI,IAE1B,GAAkB,GAAW,EAC7B,GAAa,IAInB,MAAO,IAsBT,EAAA,UAAA,SAAA,SAAS,EAAe,GACtB,GAAI,GAAS,MACX,GAAI,KAAK,OAAO,QAAU,EACxB,KAAM,IAAI,GACN,wCAAwC,EAAK,oBAC7C,QAAO,KAAK,OAAO,OAAM,eAE7B,MAAO,MAAK,OAAO,WAGjB,GAAQ,KACV,KAAM,IAAI,GAAW,8CAIzB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,OAAS,EACjB,MAAO,GAGX,KAAM,IAAI,GAAW,kBAAkB,IAQzC,EAAA,UAAA,gBAAA,WAAA,GAAA,GAAA,KAKE,MAAO,GAAA,KAAK,WAEV,OADM,GAAmB,GACL,EAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACL,EAAY,EAAG,EAAY,EAAM,aAAa,OAClD,EAAE,GACL,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,EAAK,eAAe,IAAI,IAC1B,EAAO,KAAI,MAAX,EAAe,EAAM,mBAK3B,MAAO,MAIX,EAAA,UAAA,UAAA,WAWE,OAVM,GAAmC,CAAC,KAAM,KAAK,MAK/C,EACF,KAAK,uBAAuB,KAAK,QAG/B,EAAe,GACD,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EAAiB,EAAM,eACvB,EAAc,EAAM,YACpB,EAAuB,GACpB,EAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAO,EAAM,aAAa,GAC1B,EAAU,EAAU,QAAQ,EAAO,GACrC,EAAS,GACb,GAAI,KAAK,eAAe,IAAI,IAG1B,GAAI,EAAK,SACP,IACE,KAAK,UAAU,EAAK,UACpB,EAAS,EAAK,eACP,GACP,QAAQ,KACJ,SAAS,EAAM,KAAI,mDAEhB,GAAK,SAAQ,gCAChB,+EAEJ,EAAS,GAGb,GAAI,EAAK,cAAc,OAAS,GAE9B,OADM,GAAW,GACR,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAU,EAAU,QAAQ,EAAc,GAC5C,EAAe,EAAkB,GACrC,AAAI,GAAgB,MAClB,GAAe,GAEjB,EAAS,KACL,CAAC,EAAa,KAAM,EAAc,EAAa,IAErD,EAAqB,KAAK,KAIhC,GAAM,GAAiC,GACvC,EAAK,KAAU,EAAM,KACrB,EAAK,UAAe,EACpB,EAAK,OAAY,EACjB,EAAK,aAAkB,EACvB,EAAa,KAAK,GAEpB,EAAO,OAAY,EAGnB,OADM,GAAc,GACX,EAAI,EAAG,EAAI,KAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAY,KAAK,uBAAuB,GAExC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,yBAAyB,GAClD,EAAY,KAAK,CAAC,EAAM,KAAM,EAAc,IAE9C,EAAO,YAAiB,EAGxB,OADM,GAAe,GACZ,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GAEzC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,0BAA0B,GACnD,EAAa,KAAK,CAAC,EAAM,KAAM,EAAc,IAE/C,SAAO,aAAkB,EAClB,GAgBF,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAA8C,GAO9C,EAAkE,GACxE,WACI,EAAc,GAChB,AAAM,EAAM,OAAQ,GAGlB,EAAiB,EAAM,MAAM,KAAK,GAFlC,EAAiB,EAAM,MAAQ,CAAC,GAMpC,WAAqB,EAAc,GAGjC,OAFM,GAAiC,GACnC,GACoB,GAAA,EAAA,GAAA,EAAA,GAAA,GAAA,OAAA,MAAnB,GAAM,IAAS,GAAA,IACZ,GAAmB,GAAU,GAC7B,GAAmB,GAAU,GAC7B,GAAqB,GAAU,GAKrC,GAHA,GAAS,GAAU,IAAM,KACrB,GACA,GAAU,GACV,CAAE,MAAoB,KACxB,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAe,EAAc,IACnC,GAAI,GAAa,aAAa,QAAU,IACtC,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAc,GAAa,aAAa,IAC9C,EAAa,KAAK,GAAY,cAAc,KAK9C,AAAI,EAAa,OAAS,GACxB,EAAM,MACF,GAA+B,GAC/B,IAUR,WAAsB,GACpB,GAAM,GAAY,EAAU,KAEtB,EACF,GACI,EACA,EAAO,eAAoB,KACvB,EAAO,cACP,IACZ,EAAM,6BAA6B,GACnC,EAAc,GAAa,EAE3B,GAAM,IACF,EAAU,aACd,GAAiB,QAAQ,SAAA,IACvB,GAAI,CAAE,cAAoB,QACxB,KAAM,IAAI,GACN,yDACI,IAMV,EAAmB,EAAO,MAO9B,OAFM,GAAO,EAAO,KACd,EAAmB,EAAO,OACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAa,GAOf,KAAO,CAAC,GAA4B,IAClC,OAAwB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAQ,EAAc,EAAU,MACtC,GAAI,EAAM,OAAQ,IAChB,GAAM,GAAkC,EAAiB,EAAM,MAC/D,MAAO,GAAiB,EAAM,MAC9B,OAAuB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAY,EAAO,KAU3B,OAJM,GAAiC,GACjC,EAAkC,GAClC,EACF,EAAO,YACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAa,KAAK,EAAmB,IAIvC,OAFM,GACF,EAAO,aACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAc,KAAK,EAAmB,IAExC,MAAO,IAAI,GAAI,CAAC,OAAQ,EAAc,QAAS,EAAe,KAAI,KASpE,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WAGE,GAAI,KAAK,UACP,KAAM,IAAI,GACN,wLAIN,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,SACR,MAAO,GAGX,MAAO,oCAST,EAAA,UAAA,YAAA,WAAA,GAAA,GAAA,KACE,EAAA,KAAK,WACH,EAAK,OAAO,QAAQ,SAAA,GAElB,AAAI,EAAM,UACR,EAAM,mBAMhB,GA3vCwC,IC7CxC,AAiCA,YACI,EAAmD,EACnD,GACF,GAAM,GAAa,EAAY,OAC/B,GAAI,GAAW,MAAS,MAAM,QAAQ,IAAY,EAAQ,SAAW,EACnE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,QAEjC,GAAI,IAAe,EACjB,MAAI,OAAM,QAAQ,IAAY,EAAQ,SAAW,EACxC,EACE,MAAO,IAAY,UAAY,EAAY,IAAM,GACnD,CAAE,EAA2B,EAAY,KAEzC,CAAC,GAGZ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAQ,SAAW,EACrB,KAAM,IAAI,OACN,YAAY,EAAU,mBAAmB,EAAQ,OAAM,IACvD,kCAAiC,EAAU,cAC3C,iEAEN,MAAO,WAEL,MAAO,IAAY,UAAY,OAAO,KAAK,GAAS,OAAS,GAC7D,MAAQ,GAA2B,OAAO,KAAK,GAAS,KACpD,UACN,GAAM,GAAwB,GAC9B,SAAY,QAAQ,SAAA,GAClB,AAAI,IAAc,GAChB,EAAO,KAAM,EAA2B,IAExC,EAAO,KAAK,QAGT,MAEP,MAAM,IAAI,OACN,2BAA2B,EAAU,cACrC,OAAM,EAAU,kCACb,GAAU,+BAA+B,EAAW,WACvD,aAAY,EAAU,oBAAoB,KAAK,UAAU,iBAkB7D,EACA,GACF,MAAO,IACH,EAAa,EAAa,2BA6B5B,EAAW,EAAuB,EAClC,2GACF,GAAI,GAAgB,MAAQ,GAAoB,KAG9C,KAAM,IAAI,OAAM,qDAGd,IAAe,KAEX,GAAqB,EAAA,KAAK,WAC9B,GAAI,EAAE,MAAM,SAAW,EAErB,MAAO,GAAE,QACJ,GAAI,EAAE,MAAM,SAAW,EAC5B,GAAI,EAAE,MAAM,GAAK,GAEf,GAAM,GAAO,EACb,MAAO,GAAE,OAAO,QACX,GAAI,EAAE,MAAM,KAAO,EAExB,MAAO,GAAE,QAAQ,CAAC,EAAE,MAAM,KAE1B,KAAM,IAAI,OACN,+CAA+C,EAAE,MAAM,GAAE,4EAK/D,MAAM,IAAI,OACN,yCAAyC,EAAE,KAAI,4EAKjC,EAAA,GAAA,OAAM,KAAK,CAAA,EAAM,EAAS,SA3B9C,CAAA,EAAA,UA2BI,SAAgB,EAAA,MAAA,EAAA,CAAW,EAAA,SACjC,EAAA,QAAQ,GACF,EAA8B,GACpC,EAAc,QAAQ,SAAA,GACpB,GAAI,EAAY,IAAe,KAC7B,KAAM,IAAI,OACN,8DACA,cAAa,EAAU,mCACvB,eAEJ,EAAkB,KAAK,EAAY,MAIvC,CAAA,EAAO,EAAA,SAAS,EAAmB,mBAEnC,MAAA,CAAA,EAAO,uBAWyB,EAAgB,GAClD,MAAO,GAAA,IAAI,EAAQ,GC5LrB,AAyLA,GAAM,IAAgC,GAgBtC,YAII,EAAY,GACd,GAAI,GACA,EAEE,EAAiB,EACvB,EAAK,EAAe,GACpB,EAAK,EAAe,GACpB,EAAA,KAAS,OACL,GAAM,MAAQ,GAAM,KACpB,WAAM,MAAA,+OAIF,IAAG,KAEX,GAAM,GACF,GAA0B,QAAS,EAAM,WAAY,GACnD,EACF,GAA0B,SAAU,EAAM,YAAa,GAErD,EAAoB,EAAY,GAAG,MAAM,GAE/C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,OAAO,OACpC,WAAM,MAAA,mBAAmB,EAAM,OAAO,OAAM,4BACxC,aAAY,EAAY,OAAM,oCAC3B,MAAK,UAAU,EAAM,YAAW,OAE3C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,QAAQ,OACrC,WACI,MAAA,mBAAmB,EAAM,QAAQ,OAAM,6BACvC,aAAY,EAAY,OAAM,sCAC3B,MAAK,UAAU,EAAM,aAAY,OAE5C,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,8BACC,GAAM,WAAW,GAAO,QACrB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,+BACC,GAAM,YAAY,GAAO,QACtB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,MAAO,CAAC,GAAI,EAAa,GAAI,GAG/B,YACI,EAAuB,EAAiB,GAC1C,GAAI,YAAkB,GAAA,OACpB,MAAO,CAAC,GACH,GAAI,MAAM,QAAQ,GACvB,SAAA,KAAS,OACL,EAAO,SAAW,EAAM,OACxB,WAAM,MAAA,wBAAwB,EAAO,OAAM,0BACvC,EAAM,OAAM,iBAAiB,EAAa,SAAS,EAAK,MACzD,EAIP,OAFM,GAAuB,GAEV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,IAAS,KAClB,KAAM,IAAI,GACN,gEACG,GAAa,SAAS,EAAI,OAEnC,EAAO,KAAK,EAAO,IAErB,MAAO,GAIX,YACI,GAMF,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,IACN,0DAEN,MAAO,CAAC,GAAI,EAAK,GAAI,GAAI,EAAK,gBAO5B,EAAY,EACZ,yKA0BF,GAzBM,EAAqB,EAAK,iBAAmB,KACnD,EAAA,KAAS,OACL,EAAM,WAAa,KACnB,WAAM,MAAA,mGAGV,EAAA,KAAS,OACL,GAAQ,KACR,WAAM,MAAA,kGAEV,EAAA,KAAS,OACL,EAAK,QAAU,MAAQ,EAAK,OAAS,GAAK,OAAO,UAAU,EAAK,QAChE,WAAM,MAAA,gEACF,qBAAoB,EAAK,UACjC,EAAA,KAAS,OACL,CAAC,GACI,EAAK,gBAAkB,GAAK,OAAO,UAAU,EAAK,iBACvD,WAAM,MAAA,gEACF,2CAA0C,EAAK,mBACvD,EAAA,KAAS,OAEJ,EAAa,iBAAsB,KACpC,WAAM,MAAA,sFAGN,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,mDAGX,EAAe,EAAK,gBAAkB,KACxC,EAAK,OACL,EAAK,OACT,AAAI,GACF,CAAI,GAAgB,EAAK,gBACvB,EAAA,KAAS,OACL,EAAK,mBAAqB,MACrB,EAAK,kBAAoB,GACzB,OAAO,UAAU,EAAK,mBAC3B,WAAM,MAAA,yIAGF,YAAW,EAAK,qBAElB,GAAiB,GACnB,EAAK,gBAMT,EAAQ,EAAe,GACvB,EAAQ,EAAe,KAIrB,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAe,OACnB,AAAI,EACF,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,IAAK,MAAA,OAAS,MAEzD,EAAkB,EAAU,QAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YACtD,EAAU,EAAK,SAAW,KAAO,EAAI,EAAK,QAC1C,EAA0B,GAC5B,EAAW,EAAS,EAAK,OAAQ,KAAM,KACvC,GAAiB,EAAS,GAC1B,KACA,EAAc,GAJX,EAAY,EAAA,aAAE,EAAA,EAAA,QAKrB,EAAa,SAAS,GACtB,EAAM,QAAU,EAEhB,CAAA,EAAM,EAAa,uBAAnB,SAAA,OACA,EAAM,cAAgB,GAClB,EAAQ,EAAK,cAAgB,KAAO,EAAI,EAAK,aAE9B,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BACZ,GAAQ,EAAK,OACZ,GAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,KAFR,CAAA,EAAA,iBAExB,GAAA,OACI,EAAY,EACZ,EAAa,EACZ,EAAD,CAAA,EAAA,GACa,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BAEV,GAAqB,EAAY,EAAK,gBAAkB,IACzC,CAAA,EAAM,EAAa,QAD0B,CAAA,EAAA,WAKjE,MAJM,GAAc,EAAA,OAIhB,GAAsB,EAAY,KACpC,SAAQ,KACJ,qCACG,GAAK,gBAAe,MACvB,mDACG,GAAS,cACZ,8HAGG,GAAK,gBAAkB,EAAK,OAAM,eACrC,yEAEJ,CAAA,EAAA,KAGE,EAAY,OAAS,KACjB,GACF,GAA8B,EAAO,EAAY,OAD9C,EAAE,EAAA,GAAE,EAAE,EAAA,GAEP,EAA4B,GAClC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAG,GAAG,MAAM,GAEhC,CAAA,EAAM,EAAa,aAAa,EAAY,KAP1C,CAAA,EAAA,cAOF,EAAA,OAEM,EAA8B,IAChC,GAAK,aAAe,MAApB,MAAA,CAAA,EAAA,IACI,EACF,GAAwB,EAAK,YAAa,EAAM,aAC3C,EAAI,2BAAG,GAAI,EAAqB,OACvC,GAAA,GAAA,GAAc,KAAK,CAAA,EAAM,GACrB,EAAG,GAAI,KAAM,EAAqB,MAFO,CAAA,EAAA,YAC7C,EAAA,MAAA,EAAA,CAAmB,EAAA,4BAD4B,QAAE,iBAUrD,IAHM,EAAM,EAAG,OAAO,GAAI,OAAO,GAC3B,EAAO,EAAc,GAC3B,EAAA,QAAY,GACH,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EAChC,EAAQ,EAAU,GAClB,EAAM,EAAK,GACjB,EAAU,GAAS,EACnB,EAAA,KAAS,GAGX,MAAA,CAAA,EAAM,EAAa,WAAW,EAAY,YAA1C,EAAA,OACA,GAAqB,GAErB,IACA,6BAGE,GAAqB,GAAa,EAAK,gBAClB,EAAY,MAE/B,EACE,GAAO,OACP,GAAgB,EAAK,gBACb,GAAA,GAAO,CAAA,EAAM,EAAM,gBACzB,EAAK,eAAgB,CAAC,QAAS,EAAK,sBAFtC,CAAA,EAAA,KAFF,CAAA,EAAA,IAHF,CAAA,EAAA,YAME,SAAU,EAAA,MAAA,OAAA,CAAO,EAAA,wBAGjB,EAAU,GAAO,EAAM,SAAS,EAAO,EAAO,CAC5C,UAAW,EAAK,qBAAuB,KACnC,GACA,EAAK,oBACT,QAAS,wBAGb,IAAS,EAAI,EAAG,EAAI,EAAM,aAAa,OAAQ,EAAE,EAC/C,EAAU,OAAO,EAAM,aAAa,IAAQ,EAAQ,sBAQxD,MAAA,CAAA,EAAA,YAGF,MAAI,GAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,YAErC,MAFA,GAAA,OACA,IACI,EAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,sBAAnB,SAAA,OACA,CAAA,EAAM,EAAM,QAAQ,oBAApB,SAAA,OACA,CAAA,EAAO,EAAM,iBAEb,SAAM,WAAa,8BAKvB,YACI,EAAqB,GAEvB,GAAI,GAAwB,KAC5B,MAAI,GAAK,iBAAmB,KAC1B,EAAgB,EAAK,gBAChB,AAAI,OAAO,SAAS,EAAQ,OACjC,GAAgB,EAAQ,MAEnB,EAKT,YACI,GAKF,MAAQ,OAAQ,GAAuB,UAAa,WAKtD,YAAiC,GAE/B,MAAQ,OAAQ,GAA6B,MAAS,uBAOpD,EAAY,EACZ,uHAKF,GAJA,EAAO,GAAQ,GACT,EAAa,EAAK,SAAW,KAC7B,EAAI,EAAM,aACZ,EAAqB,GACrB,EAAK,QAAU,EACjB,KAAM,IAAI,IAAoB,8CAGhC,GAAA,KAAS,OACL,CAAC,GAAe,EAAK,QAAU,GAAK,OAAO,UAAU,EAAK,SAC1D,WAAM,MAAA,6DACF,aAAY,KAAK,UAAU,EAAK,YACnB,GAAqB,GACtC,GAAA,SADiB,CAAA,EAAA,UAEjB,MAAA,CAAA,EAAO,EAAuB,mBAA9B,EAAA,EAAA,wBAFE,EAAY,EAId,EAAc,EACd,EAAQ,uEAGU,MAAA,CAAA,EAAM,EAAa,eAmCvC,MAnCM,GAAc,EAAA,OACpB,EAAO,EAAA,KAAS,WACd,GAAI,EAAY,OAGR,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAU,EAAG,OAAO,GACpB,EAAY,EAAA,KAAS,WAAM,MAAA,GAAE,KAGnC,GAFA,EAAA,QAAY,GAER,IAAU,EACZ,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAKrB,OADM,GAAY,EAAQ,GAAG,MAAM,cAC1B,GACP,GAAM,GAAW,EAAU,GACrB,EAAY,EAAK,GACvB,EAAK,GACD,EAAA,KAAS,WAAM,MAAA,GAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,MACvD,AAAI,EAAQ,GACV,EAAA,QAAY,IANP,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,IAA/B,GAST,EAAA,QAAY,GACZ,GAAe,EAEf,EAAE,EAEJ,MAAO,KAGL,EAAY,KACd,CAAI,GACF,QAAQ,KACJ,uJAGA,2BAA0B,EAAK,QAAO,eACtC,qHA3CH,GAAa,EAAQ,EAAK,QAAU,YAAI,CAAA,EAAA,2DAkD/C,IAAS,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC3B,EAAY,EAAK,GACvB,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAC3B,EAAA,QAAY,GAGd,MAAA,CAAA,EAAO,GAAiB,SCrmB1B,YAiK+B,GAC7B,EAAA,KAAS,OACL,EAAY,GAAK,OAAO,UAAU,GAClC,WAAM,MAAA,2DACF,gBAiBN,EAAyB,EAAe,GAC1C,MAAI,IAAU,KACL,CAAC,MACC,MAAM,QAAQ,GAChB,EAAO,IAAI,SAAA,GAAS,MAAA,IAAoB,EAAO,EAAO,EAAO,KAE7D,GAAoB,EAAQ,EAAO,EAAO,eAkBjD,EAAyB,GAC3B,MAAO,GAAA,KAAS,WACd,MAAI,IAAU,KACL,KACE,MAAM,QAAQ,GAChB,EAAO,IACV,SAAA,GAAS,MAAC,IAAqB,EAAO,KAInC,GACH,EAAQ,EAAQ,QAAU,QAAU,EAAU,EAAQ,uBAc5D,EAAc,GAIhB,OAHM,GAAkC,GACpC,EAAa,EACb,EAAmB,KAChB,EAAa,GAClB,EAAW,EAAa,EACxB,AAAI,GAAY,GACd,GAAW,GAEb,EAAO,KAAK,CAAC,EAAY,IACzB,EAAa,EAEf,MAAO,GA8BT,YAGI,EAAY,EAAiC,EAC7C,EAAsB,EAAoB,EAAiB,EAC3D,EAA4B,EAC5B,EAAmB,EAA0B,EAC7C,EAAuB,EACvB,mHAoBF,GAnBA,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,GAAU,MACZ,GAAS,GAEX,AAAI,GAAW,MACb,GAAU,IAEZ,AAAI,GAAgB,MAClB,GAAe,GAIb,EAAe,GACnB,AAAI,GAAQ,MAAQ,GAAU,MAC5B,GAAe,IAGb,GAAmB,MACrB,GAAe,GACX,GAAiB,MACnB,KAAM,IAAI,GACN,oGAKF,SACF,EAAM,gBAAgB,EAAK,EAAW,EAAe,mBAEzD,AAAI,GAAmB,MACrB,GAAa,GAAM,EAAG,IAGxB,AAAI,GAAW,MACb,GAAU,GAGN,EAA0B,GAC5B,EAAW,EAAS,EAAQ,EAAc,EAAiB,EAC3D,EAAW,EAAc,GAFtB,EAAY,EAAA,aAAE,EAAO,EAAA,QAG5B,EAAa,SAAS,GACtB,EAAM,QAAU,EAChB,CAAA,EAAM,EAAa,uBAAnB,EAAA,OACA,EAAM,cAAgB,cAIb,qEACP,MAAA,CAAA,EAAM,EAAa,aAAa,cAAhC,EAAA,OACM,EAA4B,IAC9B,IAAiB,MAAjB,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,IACN,qDAEJ,GAAI,IAAY,QACd,KAAM,IAAI,IAAoB,0CACzB,AAAI,GACT,EAAA,KAAK,QAAQ,GAIT,EAAoB,EAAA,SAAS,GAE7B,EAAU,GAAY,EAAiB,cACpC,2DACD,SAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,EAAY,WAA5C,SAAA,OAEA,EAAA,KAAS,WACP,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,GAAW,GACI,EAAmB,EACnB,EAAW,GAChC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAW,EAM/B,OAFM,IAAW,GAAqB,EAAK,IACrC,GAAO,EAAE,IACN,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAK,IACjB,EAAU,IAAS,GACnB,EAAA,KAAS,IAIX,GAAI,IAAe,EAAQ,OAAS,GAC9B,EAGF,OAFM,IAAU,EAAM,SAAS,EAAM,EAAQ,GAEpC,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAQ,IACpB,EAAA,KAAS,IAET,EAAU,OAAS,IAAS,MAMpC,CAAA,EAAM,EAAa,WAAW,EAAY,WAG1C,MAHA,GAAA,OACA,GAAqB,GAEjB,EAAM,kCA3CH,EAAa,yBAAG,GAAa,EAAQ,YAArC,IAA2C,CAAA,EAAA,iEAAE,QAAE,eAiDxD,EAAkB,2BAGpB,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,WACrC,MADA,GAAA,OACI,EAAM,kCAtEH,EAAQ,yBAAc,GAAQ,OAA9B,IAAoC,CAAA,EAAA,iEAAE,QAAE,eA0EjD,MAAA,CAAA,EAAM,EAAa,qBAAnB,SAAA,OAEA,CAAA,EAAM,EAAM,QAAQ,mBAApB,SAAA,OACA,CAAA,EAAO,EAAM,0BAMX,EAAY,EACZ,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,yIACF,GAAI,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,iDASX,EAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIT,EAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAG,EAAG,EAAK,aAAc,EAAK,YAAa,EAC3C,cAHF,EACF,EAAA,OAGJ,EAAS,EAAiB,GAC1B,EAAU,EAAiB,GAC3B,EAAgB,EAAiB,GAG7B,EAAe,GACf,EAAM,QACN,GAAK,gBAAkB,MAAQ,EAAK,eAAe,OAAS,GAA5D,MAAA,CAAA,EAAA,GAEF,GADA,EAAe,GACX,EAAK,eAAe,SAAW,EAEjC,EAAY,EAAK,eAAe,GAChC,EAAY,EAAK,eAAe,OAC3B,MAAI,GAAK,eAAe,SAAW,EAClC,GAAI,IACN,iEAEE,GAAI,GACN,0GAEG,GAAK,eAAc,iBAGtB,SAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAW,EAAW,KACtB,KACA,EAAgB,WAJlB,SACF,EAAA,OAIJ,EAAO,EAAgB,GACvB,EAAO,EAAgB,GACvB,EAAS,EAAK,OAAO,gBAEhB,AACH,EAAK,iBAAmB,MAAQ,EAAK,gBAAkB,GACvD,EAAK,gBAAkB,EACzB,GAAe,GAET,EACF,KAAK,MAAM,EAAO,GAAG,MAAM,GAAM,GAAI,EAAK,kBACxC,EAAoB,EAAO,GAAG,MAAM,GAC1C,EAAO,GAAY,EAAQ,EAAS,GACpC,EAAS,GAAY,EAAQ,EAAG,GAChC,EAAO,GAAY,EAAS,EAAS,GACrC,EAAU,GAAY,EAAS,EAAG,GAGlC,EAAS,EAAK,OAAO,IAGhB,AAAI,EAAK,iBAAmB,MACjC,GAAe,qBAIX,SAAM,EAAO,OAAO,GAAS,OAAO,GAE1C,EAAM,mCAcA,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAW,OACX,EAAe,OACnB,AAAI,EACF,GAAM,mBACN,EAAc,EAAM,aACpB,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,MAEzD,GAAc,KACd,EAAS,GACT,EAAkB,EAAU,SAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YAChD,CAAA,EAAM,GACd,EAAO,EAAe,EAAK,EAAW,EAAW,EAAK,OACtD,EAAK,QAAS,EAAW,EAAa,EAAQ,EAAK,QACnD,EAAiB,EAAK,aAAc,KAAM,cAHxC,SAAM,EAAA,OAIZ,CAAA,EAAO,UAEP,SAAM,WAAa,GAEnB,GAAkB,EAAQ,GAC1B,GAAkB,EAAS,GAC3B,GAAkB,EAAkB,GACpC,GAAkB,EAAkB,GACpC,AAAI,GAAiB,MACnB,EAAA,QAAY,yCAYyB,GACzC,GAAM,GAAiB,GACvB,AAAI,YAAmB,GAAA,QACrB,GAAU,CAAC,IAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAS,EAAQ,GACvB,GAAI,EAAO,OAAS,EAClB,EAAK,KAAK,GAAW,EAAQ,SACxB,GAAI,EAAO,OAAS,EACzB,KAAM,IAAI,OACN,yEAGJ,EAAK,KAAK,IAGd,MAAO,eAgBL,EACA,GACF,GAAI,GAAW,KACb,OAEF,GAAM,GAAyB,GAC/B,GAAI,YAAsB,GAAA,OACxB,EAAa,KAAK,EAAW,YACpB,MAAM,QAAQ,GACvB,EAAW,QAAQ,SAAA,GAAK,MAAA,GAAa,KAAK,EAAE,cACnC,GAAc,KAEvB,OAAW,KAAQ,IACjB,GAAM,GAAY,EAAW,GAC7B,EAAa,KAAK,EAAU,IAIhC,GAAM,GAA6B,GACnC,GAAI,YAAmB,GAAA,OACrB,AAAI,EAAa,QAAQ,EAAQ,MAAQ,IACvC,EAAiB,KAAK,WAEf,MAAM,QAAQ,GACvB,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAa,QAAQ,EAAE,MAAQ,IACjC,EAAiB,KAAK,aAGjB,GAAW,KAEpB,OAAW,KAAQ,IACjB,GAAM,GAAS,EAAQ,GACvB,AAAI,EAAa,QAAQ,EAAO,MAAQ,IACtC,EAAiB,KAAK,GAK5B,EAAiB,QAAQ,SAAA,GACvB,AAAK,EAAE,YACL,EAAE,YCrnBR,YA+C6B,GAE3B,MAAO,aAAa,GAAA,mBAMM,GAE1B,MAAO,OAAM,QAAQ,eAMI,GAEzB,MAAO,CAAC,GAAa,IAAM,CAAC,GAAY,eAetC,EAAqD,EACrD,EAAkB,EAAuB,GAC3C,GADoB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACvC,GAAS,MAAQ,EAAM,SAAW,GAGpC,GAAI,GAAQ,MACV,GAAI,GAAoB,GACxB,GAAI,GAAY,IAAU,EAAkB,OAAS,EACnD,EAAoB,WACX,GAAW,IACpB,OAAW,KAAO,GAChB,GAAI,EAAK,eAAe,IACtB,EAAoB,GACpB,WAKJ,GAAoB,GAEtB,GAAI,EACF,KAAM,IAAI,GACN,6BAA6B,EAAe,sBAC5C,YAAW,IAGnB,MAAO,GAET,GAAI,GAAQ,KACV,MAAO,GAAM,IAAI,SAAA,GAAQ,MAAA,QAG3B,GAAI,GACJ,GAAI,GAAW,IACb,EAAO,EACP,EAAS,GACT,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAK,IAAS,KAChB,KAAM,IAAI,GACN,yBAAyB,EAAI,iCAC7B,IAAG,IAET,EAAO,KAAK,EAAK,aAEV,GAAY,IAErB,GADA,EAAO,EACH,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,oCAAmC,EAAM,OAAM,oBAC/C,iDAAgD,IAEtD,EAAS,OAGT,GADA,EAAO,EACH,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,aAAa,EAAe,YAAY,EAAM,OAAM,eACpD,2DACI,EAAK,QAEf,EAAS,CAAC,GAMZ,GAHA,EAAS,GAA2B,GAGhC,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,EAAM,QAErB,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EAEd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MAAQ,GAAU,GAAK,IAAQ,EAC3C,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,mBAAkB,EAAO,GAAE,gCAC3B,KAAI,EAAM,MAAK,QAK3B,MAAO,eAWL,EAAkB,EAAmB,GACvC,GAAM,GAAO,GAAO,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,MAAM,MACpD,EAAK,OACL,GAAM,GAAO,GAAO,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,MAAM,MAGvD,GAFA,EAAK,OAED,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,mFAEA,IAAG,KAAK,UAAU,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,WAElD,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,oFAEA,IAAG,KAAK,UAAU,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,WAErD,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,CAAC,EAAA,KAAK,YAAY,EAAM,GAChE,KAAM,IAAI,GACN,kEACA,mBAAkB,EAAK,GAAE,wBAAwB,EAAK,GAAE,YACxD,cAaR,YACI,EAAmB,EAA2B,GAMhD,OAJM,GAAY,CAChB,GAAyB,GACzB,IAEO,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAI,EAAQ,GACZ,EAAO,EAAQ,GACf,EAAQ,EAAa,GAC3B,GAAI,GAAQ,KACV,SAEF,GAAI,IAAS,IACP,EAAE,MAAM,EAAE,MAAM,OAAS,KAAO,EAClC,KAAM,IAAI,GACN,2CAA2C,EAAE,MAAK,4JAO1D,GAAI,EAAU,QAAQ,KAAU,GAG9B,OAFM,GAAe,EAAE,MAAM,MAAM,GAC7B,EAAc,EAAM,MAAM,GACvB,EAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAY,EAAa,GACzB,EAAS,EAAY,GAC3B,GAAI,GAAU,MAAQ,IAAc,EAClC,KAAM,IAAI,GACN,8BAA8B,EAAE,MAAK,sBACrC,oBAAmB,EAAK,uCACxB,2DAiCd,YACI,EAAuB,EAAiB,EACxC,EAAuB,GAAvB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACzB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,wCAAuC,EAAM,OAAM,eACnD,qBAAoB,EAAK,OAAM,iBAErC,EAAS,OAET,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,qBAAqB,EAAM,OAAM,IAAI,EAAe,mEAEjD,MAAK,UAAU,EAAK,OAAM,MAEnC,EAAS,CAAC,GAGZ,GAAI,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,KAAK,UAAU,EAAM,SAEpC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EACd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MACR,IAAW,EACb,KAAM,IAAI,GACN,uBAAuB,EAAe,cACnC,GAAM,GAAE,kBAAkB,KAAK,UAAU,EAAO,IAAG,SACtD,yBAAwB,KAAK,UAAU,EAAM,OAAM,oBAsB/D,EAEA,GACF,GAAI,GAAW,MAAQ,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAClE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,KAGjC,GAAI,GAEJ,GAAI,MAAO,IAAY,UAAY,MAAO,IAAY,WACpD,EAAiB,CAAC,WACT,MAAM,QAAQ,IAAY,MAAO,IAAY,SACtD,EAAiB,MAGjB,MAAM,IAAI,WACN,+DACA,uCAAsC,IAG5C,GAAI,MAAM,QAAQ,GAEhB,MAAO,GAAY,IACf,SAAA,GAAQ,MAAA,KAIZ,OADM,GAAqD,GACxC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACT,EACA,EAAe,eAAe,GAAQ,EAAe,GAAQ,GACjE,AAAK,MAAM,QAAQ,IACjB,GAAgB,CAAC,IAEnB,EAAc,KAAK,GAErB,MAAO,GA6DX,GAAM,IAA2B,8BAcA,EAAA,EAAA,GA4C/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WAAa,KAsCpB,SAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAGF,GAHE,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC1D,CAAC,KAAK,MACR,KAAM,IAAI,GACN,gLAIN,GAAa,KAAM,EAAY,EAAW,IAa5C,EAAA,UAAA,QAAA,SAAQ,GAAR,GAAA,GAAA,KAME,GALA,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAEd,KAAK,KAAO,EAAK,KAEb,MAAO,GAAK,WAAc,SAC5B,KAAK,WAAa,GAAwB,EAAK,WAC/C,KAAK,iBAAmB,QAExB,GAAI,CAAE,GAAK,oBAAqB,GAAA,WAC9B,KAAM,IAAI,GACN,+DAEN,KAAK,WAAa,EAAK,UACvB,KAAK,iBAAmB,GAO1B,GAAI,GAAkC,GACtC,GAAI,CAAC,MAAM,QAAQ,EAAK,OAAS,MAAO,GAAK,MAAS,UAClD,MAAO,GAAK,MAAS,YACvB,EAAK,KAAO,EAAK,KACjB,OAAW,KAAQ,GAAK,KACtB,GAAI,KAAK,YAAY,QAAQ,KAAU,GACrC,KAAM,IAAI,GACN,sCAAsC,EAAI,MAC1C,sCAAqC,KAAK,cAGlD,OAAmB,GAAA,EAAA,EAAA,KAAK,YAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAK,KAAK,IAAS,MACrB,QAAQ,KACJ,WAAW,EAAI,4GAEf,oBAAmB,EAAI,qBAE7B,EAAc,KAAK,GAAW,EAAK,KAAK,cAEjC,MAAM,QAAQ,EAAK,OAC5B,GAAI,EAAK,KAAK,SAAW,KAAK,QAAQ,OACpC,KAAM,IAAI,GACN,+DACA,gCAA+B,KAAK,QAAQ,OAAM,gBAClD,wBAAuB,EAAK,KAAI,MAEtC,GAAM,GAAY,EAAK,KACvB,EAAgB,EAAU,IAAI,SAAA,GAAK,MAAA,IAAW,UAE9C,GAAM,GAAe,GAAW,EAAK,MACrC,KAAK,QAAQ,QAAQ,SAAA,GACnB,EAAc,KAAK,KAIvB,KAAK,cAAgB,EAErB,KAAK,gBAAkB,GACvB,KAAK,iBAAmB,GACxB,KAAK,YAAc,GACnB,OAAS,GAAI,EAAG,EAAI,KAAK,QAAQ,OAAQ,EAAE,GAEzC,GAAM,GAAQ,KAAK,qBAAqB,GAClC,EAAO,KAAK,YAAY,GAC9B,KAAK,gBAAgB,KAAK,GAC1B,KAAK,iBAAiB,KAAK,GAC3B,KAAK,YAAY,KAAK,KAAK,cAAc,IAK3C,GAAM,GAA8B,GAGpC,KAAK,QAAU,EAAK,QAEpB,KAAK,aAAe,CAAC,QACrB,KAAK,eAAiB,GAMtB,GAAU,OAAQ,WAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,GACzC,GAAI,EAAkB,QAAQ,KAAO,GACnC,SAIF,GAAM,GAAe,EAAK,cAAc,GACxC,AAAI,EAAK,QAAQ,OAAS,GACxB,GAAK,eAAe,KAAK,CAAC,EAAc,IACxC,EAAK,aAAa,KAAK,EAAK,YAAY,GAAK,aAQnD,GAAM,GAAgB,GAAe,EAAK,QAAS,KAAK,aAMlD,EACF,SAAC,EAAqB,EACrB,GACC,AAAI,EAAK,YAAY,OAAS,GAC5B,GAAa,EAAK,YAAY,GAAe,IAAM,GAErD,EAAK,aAAa,KAAK,GACvB,EAAK,eAAe,KAAK,CAAC,EAAc,KAG9C,GAAU,SAAU,WAClB,mBAAS,GACP,GAAI,EAAkB,QAAQ,KAAO,oBAGrC,GAAM,GAAgB,EAAc,GAI9B,EAAgB,SAAC,GAOrB,OANM,GAAmB,GACrB,EACA,EACA,aAGO,GACT,GAAI,MAAO,IAAW,UAClB,CAAC,WAAY,MAAO,eAAgB,MAAM,QAAQ,KAC9C,IACN,GAAM,GAAc,EAAK,qBAAqB,GAE9C,AAAI,EAAY,EAAY,OAAS,KAAO,GACxC,EAAK,cAAc,KAAO,GAE5B,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAEL,AACH,EAAK,cAAc,KACnB,GAGF,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAIV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAGZ,GAAI,GAAM,OACV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAS,MACJ,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAS,MAGX,EAAmB,EACnB,EAAa,EAAmB,OAEhC,GAAM,GAAW,GAAY,GAE7B,EAAmB,EACnB,EACI,EAAmB,GAA4B,GAIrD,GAAI,GACJ,GAAU,EAAY,WACpB,EAAe,IAEjB,EAAa,EAAG,EAAY,IAtDT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,KAAN,KA0Db,EAAc,IAzEP,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,IAAlC,KAgFX,KAAK,0BAA4B,KAAK,kBAY9B,EAAA,UAAA,iCAAV,WACE,GAAI,KAAK,2BAA6B,KACpC,OAEF,AAAI,KAAK,iBAAiB,SACtB,KAAK,0BAA0B,QACjC,QAAQ,KACJ,sJAqCR,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIf,GAAM,GAAiB,GACjB,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GACrD,IAGE,GAAM,GAAM,EAAiB,GAAG,OAAO,EAAiB,IACxD,KAAK,mBACL,GAAM,GAAI,KAAK,aACT,EACF,KAAK,SAAS,EAAG,EAAK,EAAW,EAAK,QAAS,EAAK,OACxD,MAAO,IAAiB,WAExB,GAAkB,EAAiB,GAAI,GACvC,GAAkB,EAAiB,GAAI,KA0BrC,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,YAAK,mBACL,CAAA,EAAO,GAAgB,KAAM,EAAS,SAahC,EAAA,UAAA,gBAAR,SACI,EAAsB,EAAoB,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,SACF,GAAI,GACJ,GAAI,GAAS,MAEX,GADA,EAAa,KACT,GAAa,KACf,KAAM,IAAI,GACN,MAAM,EAAS,gDACf,oBAAmB,YAEhB,GAAO,KAChB,AAAI,MAAM,QAAQ,GAChB,EAAa,EAAI,GAAG,MAAM,GAE1B,EAAa,EAAI,MAAM,OAGzB,MAAM,IAAI,GACN,yDACG,GAAS,yBAElB,MAAO,IAUT,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,GAAI,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAC/C,KAAM,IAAI,GACN,sDAGN,GAAM,GAAiB,MAAM,QAAQ,GAC/B,EACD,EAAiB,EAAsB,CAAC,GACvC,EAAwB,KAAK,wBAAwB,GAGrD,EAAW,GAAI,IAIrB,GAHA,AAAI,YAAkB,GAAA,QACpB,GAAS,CAAC,IAER,MAAM,QAAQ,IAChB,GAAI,EAAO,SAAW,KAAK,OAAO,OAChC,KAAM,IAAI,GACN,kCAAkC,EAAO,OAAM,uDAE/C,KAAI,KAAK,OAAO,OAAM,OAE5B,OAAS,GAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,KAAK,OAAO,GAAI,EAAO,QAGtC,QAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAc,EAAO,EAAM,MACjC,GAAI,GAAe,KACjB,KAAM,IAAI,GACN,8CAA8C,EAAM,MAE1D,EAAS,IAAI,EAAO,GAKxB,GAAM,GAAiB,GAAQ,EAAuB,GACtD,MAAO,GAAiB,EAAiB,EAAe,IAMlD,EAAA,UAAA,wBAAR,SAAgC,GAK9B,OAHM,GACF,GAAa,KAAM,EAAoB,QACvC,EAAmB,EAAoB,OACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EACF,MAAM,QAAQ,EAAM,QAAU,EAAM,OAAS,CAAC,EAAM,QAClD,EAAmB,EAAa,IAAI,SAAA,GAAU,MAAA,GAAO,OAClD,EAAI,EAAG,EAAI,EAAoB,OAAQ,EAAE,GAChD,GAAM,GAAQ,EAAiB,QAAQ,EAAoB,IAK3D,GAJA,AAAI,IAAU,IACZ,GAAsB,GAAK,EAAa,GACxC,KAEE,IAAqB,EACvB,MAGJ,GAAI,IAAqB,EACvB,MAIJ,GAAI,EAAmB,GACrB,GAAM,GAA2B,GACjC,QAAsB,QAAQ,SAAC,EAAQ,GACrC,AAAI,GAAU,MACZ,EAAe,KAAK,EAAoB,MAGtC,GAAI,GACN,mDACA,IAAG,KAAK,UAAU,KAExB,MAAO,IAgBD,EAAA,UAAA,YAAR,SAAoB,EAAsB,EAAgB,GAA1D,GAAA,GAAA,KAA0C,MAAA,KAAA,QAAA,GAAA,IAAgB,AAAA,IAAA,QAAA,GAAA,IAEjD,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,GACxC,GAAI,EACF,KAAM,IAAI,IACN,iDAYN,OAJM,GAAU,GAAY,EAAY,GAClC,EAA0B,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,gBAGlD,GACP,GAAM,GAAY,EAAA,KAAS,WACzB,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAG/B,EAAW,GAAY,EAAK,EAAY,GAGxC,EAAQ,GACd,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,EAAE,EACrC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAS,SAGnD,GAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,IAE1C,GAAM,GAAW,GAAI,IAAS,GAC9B,MAAO,IAAQ,EAAK,QAAS,KAE/B,EAAU,QAAQ,SAAC,EAAU,GAAM,MAAA,GAAY,GAAG,KAAK,MApBhD,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,IAA/C,GAsBT,MAAO,IACH,EAAY,IAAI,SAAA,GAAW,MAAA,GAAA,OAAW,EAAS,SA+BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAkB,GAA2B,GACnD,GACI,EAAiB,KAAK,WAAY,KAAK,gBAAiB,IAC5D,IAKE,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,UAAe,GACR,KAAK,YAAY,EAAiB,WAEzC,GAAkB,EAAiB,KAmBvC,EAAA,UAAA,eAAA,SAAe,GACb,GAAe,EAAG,KAAK,WAAY,KAAK,gBAAiB,IAGzD,GAAM,GAAa,OAAM,QAAQ,GAAK,EAAE,GAAK,GAAG,MAAM,GACtD,MAAO,MAAK,YAAY,EAAG,IAGnB,EAAA,UAAA,sBAAV,SACI,EACA,EAAkD,EAClD,GAEF,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,IACN,gGAIN,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,KAAK,iBAAiB,OAAQ,EAAE,GAClD,GAAM,GAAc,KAAK,iBAAiB,GACpC,EAAS,KAAK,YAAY,GAChC,AAAI,IAAW,GACb,EAAa,KACT,EAAY,MAAM,EAAG,EAAY,OAAS,GAAG,OAAO,CAAC,KAGzD,EAAa,KAAK,GAWtB,GARA,EAAI,GACA,EAAG,KAAK,eAAgB,KAAK,gBAAiB,GAAO,SACzD,EAAI,GACA,EAAG,KAAK,gBAAiB,EAAc,GAAO,UAElD,GAAkB,EAAG,GAErB,GAAgC,EAAG,KAAK,YAAa,KAAK,kBACtD,KAAK,UAAY,GAAa,MAAQ,EAAY,GAChD,EAAE,GAAG,MAAM,GAAK,IAAc,EAChC,KAAM,IAAI,GACN,mHAEG,GAAS,YAAY,EAAE,GAAG,MAAM,GAAE,gBAG7C,MAAO,CAAC,EAAG,IAGG,EAAA,UAAA,oBAAhB,SACI,EACA,EACA,EACA,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAA,2GAKF,GAHM,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GAD9C,EAAU,EAAA,GAAE,EAAU,EAAA,GAGzB,GAAgB,KAClB,KAAM,IAAI,OAAM,0CAGd,EAAkC,MAClC,IAAe,MAAf,MAAA,CAAA,EAAA,GACI,EACF,GAAwB,EAAa,KAAK,aAC9C,EAAwB,GACf,EAAI,yBAAG,GAAI,EAAa,OAC/B,GAAA,GAAA,GAAsB,KAClB,CAAA,EAAM,GAAmB,EAAW,GAAI,KAAM,EAAa,MAF1B,CAAA,EAAA,UACrC,EAAA,MAAA,EAAA,CACI,EAAA,0BAFmC,QAAE,eAO7C,MAAA,CAAA,EAAO,CAAC,EAAY,EAAY,UAc1B,EAAA,UAAA,SAAR,SACI,EAAiC,EAAe,EAChD,EAAa,GAFjB,GAAA,GAAA,KAEI,MAAA,KAAA,QAAA,GAAA,GACK,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,EAAK,EAAW,EAAO,SACzD,EAAiB,GACvB,GAAI,EAAU,EACZ,KAAM,IAAI,IAAoB,wCAGhC,GAAI,GAAS,KACX,KAAM,IAAI,IACN,mDAIJ,OAFM,GAAU,GAAY,EAAY,GAClC,EAAa,EAAA,SAAS,GAAM,EAAG,IAC5B,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,GACtD,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,EACF,GACI,EAAY,EAAY,EAAW,GAGrC,EAAW,GAAqB,EAAK,GACrC,EAAY,EAAE,GACpB,GAAI,IAAe,EACjB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAGrB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAW,EAAU,GAC3B,EAAK,GACD,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,EAAY,KAGxD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAG/B,MAAO,MAID,EAAA,UAAA,uBAAV,WAKE,OAJM,GAAY,KAAK,aAGjB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAQ,EAAU,GACpB,EAAW,EACf,GAAI,GAAM,EAAW,GAAS,GAC5B,GAAM,GAAW,GAAM,EAAU,MAAM,EAAG,GAAI,GAC9C,GAAY,IAAI,EAElB,EAAiB,KAAK,GAExB,MAAO,IAaC,EAAA,UAAA,kBAAV,WAAA,GAAA,GAAA,KACE,MAAO,UAAC,GACN,GAAM,GAAuB,GAEvB,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAgB,EAAK,MACvB,EAAK,OAAO,OAAS,EAAK,QAAQ,OAClC,EAAK,OAAO,OAAS,EAAK,QAAQ,OAAS,GAEzC,EAA0B,GAK1B,EAAoB,WAExB,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KASjD,OAPM,GAAW,GAAI,IAAS,GACxB,EACF,GAAQ,EAAK,QAAS,EAAU,CAAC,SAAY,KAI7C,EACK,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GACpC,EAAO,EAAa,EAAQ,GAAI,EAAQ,IAC5C,AAAI,EAAc,IAAM,MACtB,GAAO,GAAoB,EAAM,EAAc,KAIjD,GAAM,GAAmB,EAAA,KAAS,GAElC,EAAW,KAAK,GAChB,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAOnC,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAI,GAAc,OAElB,GAAI,EAAK,QAAQ,OAAS,GAAK,EAAI,EAAK,QAAQ,OAC9C,EAAiB,EAAW,QAE5B,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAC3C,EACI,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAGpD,EAAA,KAAS,GAET,EAAc,KAAK,GAGrB,SAAY,EAAA,KAAS,GAGrB,EAAK,kBAAkB,QAAQ,SAAA,GAC7B,EAAY,EAAA,IAAQ,EAAW,KAG1B,GAGH,EAAY,EAAK,0BAA0B,IAC7C,SAAA,GAAS,MAAA,GAAM,SACb,EAAa,GACb,EACF,EAAK,WAAW,SAAS,EAAmB,EAAY,GAE5D,MAAO,CAAC,GAAgB,OAAO,KAS3B,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,KAAK,aAAe,SAAC,GACnB,MAAO,GAAA,KAAS,WAOd,OANM,GAAuB,GACzB,EACE,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KAKjD,OAHM,GAAW,GAAI,IAAS,GACxB,EAAU,GAAQ,EAAK,QAAS,GAE7B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAGlC,EAAe,EAAA,KAAS,EAAa,EAAQ,GAAI,EAAQ,KAC/D,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAEjC,EAAW,KAAK,GAGlB,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAErC,EACF,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAClD,EAAW,KAAK,GAElB,MAAO,OAuCP,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,MAAA,CAAA,EAAO,GAAW,KAAM,EAAG,EAAG,SA0B1B,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,MAAA,CAAA,EAAO,GAAW,KAAM,EAAS,SA0B7B,EAAA,UAAA,aAAN,SACI,EACA,qHAIqB,MAAA,CAAA,EAAM,KAAK,oBAAoB,EAAG,WAAnD,EAAiB,EAAA,OACjB,EAAS,EAAe,GACxB,EAAU,EAAe,GACzB,EAAgB,KAAK,oBACrB,EAAS,EAAc,EAAO,OAAO,IACrC,EAAuB,OACV,EAAA,yBAAA,GAAA,EAAA,OAAR,GAAI,EAAA,GACH,CAAA,EAAM,EAAK,SADE,CAAA,EAAA,UACjB,EAAI,EAAA,OACV,EAAW,KAAK,EAAE,qBAFD,wBAInB,SAAA,QAAY,GACZ,CAAA,EAAO,GAAiB,UAYhB,EAAA,UAAA,gBAAV,SAA0B,GAMxB,OALM,GAA8B,GAE9B,EAAgB,GAAU,MAAQ,EAAO,cACzC,EAAU,EAAgB,KAAK,iBAAmB,KAAK,QACvD,EAAe,KAAK,WAAW,GAC5B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAI,GAAiB,CAAC,EAAQ,GAAG,UAE/B,SAEF,EAAa,KACT,CAAC,KAAM,EAAQ,GAAG,aAAc,OAAQ,EAAa,KAE3D,MAAO,IAiCT,OAAA,eAAI,EAAA,UAAA,eAAY,KAIhB,WACE,MAAO,MAAK,mBALd,SAAiB,GACf,KAAK,cAAgB,mCAOvB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,AAAI,KAAK,aAAe,GACtB,MAAK,WAAa,EAClB,KAAK,iBAAmB,qCAI5B,EAAA,UAAA,QAAA,WACE,GAAM,GAAS,EAAA,UAAM,QAAO,KAAA,MAC5B,GAAI,EAAO,uBAAyB,GAAK,KAAK,WAAa,MACvD,KAAK,kBACP,GAAM,GAAmC,EAAA,SAAa,WACtD,KAAK,WAAW,UAChB,EAAO,sBACH,EAAmC,EAAA,SAAa,WAEtD,MAAO,IAGD,EAAA,UAAA,mBAAR,WAEE,GAAI,GAEJ,GAAI,MAAO,MAAK,MAAS,SACvB,EAAY,GAAY,KAAK,cACpB,MAAM,QAAQ,KAAK,OAC5B,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,OAAM,sDAGpB,EAAa,KAAK,KAAkB,IAAI,SAAA,GAAQ,MAAA,IAAY,UAG5D,GAAM,GAAc,OAAO,KAAK,KAAK,MACrC,EAAY,GAGZ,OAFM,GACF,KAAK,KACgB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,MAAO,GAAO,IAAgB,SAChC,EAAU,GACN,GAAY,EAAO,QAEvB,MAAM,IAAI,OAAM,uDAItB,MAAO,IAGD,EAAA,UAAA,qBAAR,WAEE,GAAI,MAAO,MAAK,SAAY,UACxB,MAAO,MAAK,SAAY,WAC1B,MAAO,CAAC,GAAY,GAA4B,KAAK,WAChD,GAAI,MAAM,QAAQ,KAAK,SAC5B,MAAO,MAAK,QAAQ,IAChB,SAAA,GAAU,MAAA,IAAY,GAA4B,MAEtD,GAAM,GAAyD,GAC/D,OAAW,KAAO,MAAK,QACrB,EAAmB,GACf,GAAY,GAA4B,KAAK,QAAQ,KAE3D,MAAO,IAID,EAAA,UAAA,kBAAV,WACE,MAAO,CACL,KAAM,KAAK,qBACX,QAAS,KAAK,uBACd,iBAAkB,CAChB,WAAY,KAAK,UAAU,eAC3B,OAAQ,KAAK,UAAU,eAQ7B,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,EAAe,kBAAoB,KACrC,KAAM,IAAI,OAAM,gDAElB,GAAI,EAAe,cAAgB,KACjC,KAAM,IAAI,OAAM,8CAElB,GAAI,EAAe,oBAAsB,KACvC,KAAM,IAAI,OAAM,oDAGlB,GAAM,GAAW,GAAoB,EAAe,kBAE9C,EAAY,GAAY,GAE1B,EACJ,GAAI,MAAO,GAAe,MAAS,SACjC,EAAO,GAAY,EAAe,cACzB,MAAM,QAAQ,EAAe,MACtC,EAAO,EAAe,KAAK,IAAI,SAAA,GAAa,MAAA,IAAY,aAC/C,EAAe,MAAQ,MAChC,EAAO,GACP,OAAW,KAAO,GAAe,KAC/B,EAAK,GAAO,GAAY,EAAe,KAAK,IAIhD,GAAI,GACJ,GAAI,MAAM,QAAQ,EAAe,SAC/B,EAAU,EAAe,QAAQ,IAAI,SAAA,GAAU,MAAA,IAAY,aAClD,EAAe,SAAW,MACnC,EAAU,GACV,OAAW,KAAO,GAAe,QAC/B,EAAQ,GAAO,GAAY,EAAe,QAAQ,IAItD,KAAK,QAAQ,CAAC,KAAI,EAAE,QAAO,EAAE,UAAS,KAoFlC,EAAA,UAAA,KAAN,SAAW,EAAmC,+HAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,GACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,GACN,gHAKF,MAAA,CAAA,EAAM,EAAA,GAAG,cAAc,KAAK,gBAAgB,kBAD1C,GACF,EAAA,OAEE,EAAe,GACf,EAAgB,KAChB,EAAc,KAAK,OAAO,EAAW,GACrC,EAAoC,CACxC,cAAe,EACf,OAAQ,GACR,YAAa,8BAA8B,GAC3C,YAAa,MAGT,EAAmB,GAAU,KAAO,GAAQ,EAAO,iBACrD,GAAoB,KAAK,WAAa,KACxC,GAAe,eAAiB,KAAK,oBAC/B,EAAa,YAET,EAAA,GAAA,EAAA,IAAG,cAAc,CAAA,EAAM,KAAK,UAAU,eAJ9C,CAAA,EAAA,UAIE,MAAA,CAAA,EAAM,EAAA,MAAA,EAAA,CAAiB,EAAA,OAAmC,YADxD,EACF,EAAA,OADS,EAAmB,EAAA,KAAS,EAAoB,EAAA,MAE7D,GAAA,EAAmB,OAAM,KAAI,MAAA,EAAI,GACjC,EAAmB,KAAO,EAAA,GAAG,wBACzB,CAAC,EAAmB,KAAM,qBAGhC,MAAI,MAAK,qBAAuB,MAExB,GAAY,GAClB,GAAyB,KAAK,oBAAqB,KAAK,KAAM,GAC9D,EAAe,oBAAsB,KAAK,qBAG5C,EAAe,WAAa,EAAmB,KAC/C,EAAe,YAAc,EAAmB,MAChD,CAAA,EAAO,EAAa,KAAK,UAW3B,EAAA,UAAA,uBAAA,SAAuB,GACrB,GAAyB,EAAqB,KAAK,MACnD,KAAK,oBAAsB,GAc7B,EAAA,UAAA,uBAAA,WACE,MAAO,MAAK,qBAv4CP,EAAA,UAAY,QAy4CrB,GA74CiC,IA84CjC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DACS,SAAA,UAAY,aACrB,GAFgC,IAGhC,EAAA,cAAc,cAAc,IC72D5B,YA8DI,EACA,uHACF,AAAM,iBAAmB,IACvB,GAAwB,CAAC,cAAe,IAE1C,EAAwB,EAEpB,EAAgB,EAAsB,cAC1C,AAAI,EAAc,cAAmB,MAMnC,GAAgB,EAAc,cAE1B,EACF,GAAoB,GAClB,EAAQ,GAAY,EAAU,GAEhC,EAAsB,iBAAmB,KAItB,CAAA,EAAM,EAAA,GAAG,YAC1B,EAAsB,gBAAiB,EAAsB,WAC7D,EAAM,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,iBANrC,CAAA,EAAA,UAUF,IANM,EAAe,EAAA,OAKf,EAAqC,GAC3C,EAAA,EAAqB,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,IAAV,EAAM,EAAA,GACf,EAAmB,EAAO,cACtB,EAAa,EAAO,cAG1B,EAAM,YAAY,GAElB,EAAA,QAAQ,oBAEV,MAAA,CAAA,EAAO,oBAsIL,EACA,4EAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,MAAO,IAAoB,UAE7B,GADM,EAAW,EAAA,GAAG,gBAAgB,EAAiB,GACjD,EAAS,SAAW,EAKtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAiB,YAC5C,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAe,MAE7B,EAAkB,EAAS,GAE7B,MAAA,CAAA,EAAO,GAA6B,EAAiB,OAAW,oBAc9D,EAAuB,EACvB,mHAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,EAAQ,MAAQ,KAClB,KAAM,IAAI,GACN,iHAGY,MAAA,CAAA,EAAM,EAAQ,kBAA1B,EAAY,EAAA,OACd,EAAgB,EAAU,cAC9B,AAAI,EAAc,cAAmB,MACnC,GAAgB,EAAc,cAG1B,EAAS,EAAQ,QAAU,KAAO,GAAO,EAAQ,OAMjD,EACF,EAAU,YAAc,MAAQ,EAAU,aAAe,MAAQ,EAC/D,EACF,GACI,GAAoB,GACpB,EAAe,GAEjB,EAAiB,EAAU,eACjC,AAAI,GAAkB,MACpB,EAAM,mBAAmB,GAE3B,AAAI,EAAU,qBAAuB,MACnC,EAAM,uBAAuB,EAAU,sBAIrC,GAAU,YAAc,MAAxB,MAAA,CAAA,EAAA,GAEF,GAAI,EAAU,aAAe,KAC3B,KAAM,IAAI,GACN,wHAIA,GAAmC,GACrC,EAAU,WAAY,EAAU,aAD7B,EAAY,EAAA,aAAE,EAAgB,EAAA,iBAErC,EAAM,YAAY,EAAc,GAE5B,EAAM,WAAa,MAAQ,EAAiB,OAAS,EACvD,CAAA,EAAM,EAAM,UAAU,WAAW,IAD/B,CAAA,EAAA,UACF,EAAA,wBAIF,EAAA,QAAQ,GACR,EAAA,QAAQ,EAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,2BAEtC,MAAA,CAAA,EAAO,QAGT,YACI,EAAqB,GAEvB,GAAM,GAAc,EAAA,GAAG,cAAc,EAAQ,GACvC,EAA+B,GAC/B,EAAkC,GACxC,SAAM,QAAQ,SAAA,GACZ,AAAI,EAAK,QAAU,YACjB,EAAiB,KAAK,CAAC,KAAM,EAAK,KAAM,OAAQ,EAAY,EAAK,QAEjE,EAAa,EAAK,MAAQ,EAAY,EAAK,QAGxC,CAAC,aAAY,EAAE,iBAAgB,sBAwCR,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CAAC,OAAQ,GAAI,QAAS,MAAI,KAUhC,GATA,EAAO,GAAQ,GAEf,EAAK,UAAY,GACjB,EAAK,MAAQ,GAGb,EAAK,KAAQ,EAAK,MAAQ,KAAQ,EAAK,KAAO,GAAO,eAGjD,EAAK,QAAU,KACjB,OAAoB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,IAAI,YAOP,SAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAQ,EAAM,aAAa,GAAG,cAAc,GAAG,MACrD,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,GAAI,IACtB,KAAM,IAAI,GACN,kDACG,GAAM,KAAI,uBACV,GAAM,aAAa,GAAG,aAAa,GAAG,MAAK,OAyBtD,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GACF,YAAiB,IAAc,YAAiB,IAChD,EACJ,GAAI,GAEF,GADA,EAAa,EACT,EAAW,QAAQ,SAAW,EAChC,KAAM,IAAI,GACN,yHAKN,GAAI,EAAW,OAAO,SAAW,EAC/B,KAAM,IAAI,GACN,uHAOR,GAAI,KAAK,QAAQ,SAAW,GAE1B,GAAI,EAAM,aAAa,SAAW,GAEhC,GAAI,EAAM,iBAAmB,KAC3B,KAAM,IAAI,GACN,iGAIN,GAAM,GAAI,GAAM,CACd,WAAY,EAAM,gBAClB,MAAO,EAAM,MACb,KAAM,EAAM,KAAO,WAIrB,EAAM,MAAM,GAGd,GAAI,EACF,KAAK,QAAU,EAAW,QAC1B,KAAK,OAAS,EAAW,YAEzB,GAAI,EAAM,aAAa,SAAW,EAChC,KAAM,IAAI,GACN,2DACA,yDACI,EAAM,KAAI,KACd,cAAa,EAAM,aAAa,OAAM,0BACtC,gBAGN,GAAI,EAAM,aAAa,GAAG,cAAc,SAAW,EACjD,KAAM,IAAI,GACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,EAAM,aAAa,GAAG,cAAc,IACpD,KAAK,OAAS,GAAgB,KAAK,QAAQ,IAG7C,KAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,KACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,KAAK,OACnB,cAAe,KAAK,QAEpB,WAAY,GAA2B,KAAM,KAAK,OAAO,QACzD,YAAa,CAAC,MACd,YAAa,KAAK,OAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QACpC,aAAc,KAAK,QAAQ,GAAG,aAGhC,GAAM,GAAe,EAAM,MAAM,KAAK,QAAQ,IAC9C,GAAI,MAAM,QAAQ,GAChB,KAAM,IAAI,WACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,GAEhB,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,OAGvD,KAAK,OAAO,KAAK,GACjB,KAAK,MAAQ,IAQf,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,OAAO,SAAW,EACzB,KAAM,IAAI,WAAU,qCAItB,GADA,KAAK,OAAO,MACR,KAAK,OAAO,SAAW,EACzB,KAAK,QAAU,GACf,KAAK,aAAe,GACpB,KAAK,cAAgB,QAErB,GAAM,GAAiB,KAAK,OAAO,OAAS,EAC5C,KAAK,OAAO,GAAgB,cAAgB,GAC5C,KAAK,QAAU,CAAC,KAAK,OAAO,GAAgB,QAE5C,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,SAIzD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,KAAK,EAAQ,IAGjC,EAAA,UAAA,MAAA,SAAM,GAKJ,GAFA,GAAmB,GAEf,KAAK,OAAO,SAAW,GAAK,KAAK,QAAQ,SAAW,EACtD,KAAM,IAAI,WACN,4EAIN,KAAK,MAAQ,GAAI,IAAY,CAC3B,OAAQ,KAAK,OACb,QAAS,KAAK,QAAQ,GACtB,KAAM,KAAK,KAAO,WAEpB,KAAK,MAAM,UAAY,KAAK,UAG5B,KAAK,gBAAkB,KAAK,MAAM,gBAElC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,uBAAyB,KAAK,MAAM,uBACzC,KAAK,yBAA2B,KAAK,MAAM,yBAC3C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,wBAA0B,KAAK,MAAM,wBAC1C,KAAK,0BAA4B,KAAK,MAAM,0BAC5C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,WAAa,KAAK,MAAM,WAG7B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,WACE,MAAK,MAAK,OACR,KAAK,QAEA,EAAA,UAAM,YAAW,KAAA,OAiC1B,EAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAAA,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC9D,AAAK,KAAK,OACR,KAAK,QAEP,EAAA,UAAM,QAAO,KAAA,KAAC,EAAY,EAAW,IASvC,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,OAAS,MAChB,KAAK,QAEP,KAAK,MAAM,WAAW,IAmCxB,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GACF,GADE,AAAA,IAAA,QAAA,GAAA,IACE,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAO,MAAK,MAAM,SAAS,EAAG,EAAG,IAyB7B,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAA,CAAA,EAAO,KAAK,MAAM,gBAAgB,EAAS,SA8B7C,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC1B,AAAI,KAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,QAAQ,EAAG,IAU/B,EAAA,UAAA,eAAA,SAAe,GACb,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,eAAe,IAQnC,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,QACL,KAAK,MAAM,QAAQ,GACnB,KAAK,WAAa,KAAK,MAAM,UAE7B,KAAK,iBAAoB,KAAK,MAAc,iBAC5C,KAAK,KAAO,KAAK,MAAM,KACvB,KAAK,QAAU,KAAK,MAAM,QAG1B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,aAAe,KAAK,MAAM,cAIjC,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAO,OAAY,KAAK,MAAM,eAGrD,SAAc,GACZ,KAAK,MAAM,UAAY,mCAkCnB,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,IAAI,EAAG,EAAG,SAwFxB,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,WAAW,EAAS,SA0BlC,EAAA,UAAA,aAAN,SACI,EACA,sEAEF,MAAA,CAAA,EAAO,KAAK,MAAM,aAAa,EAAG,SAK7B,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GACA,EAA6C,GACjD,GAAI,YAAkB,QACpB,GAAI,CAAE,GAAO,GAAG,WAAa,OACzB,EAAO,GAAG,YAAiB,QAC7B,KAAM,IAAI,GAAW,kDAEvB,EAAc,MAEd,GAAA,KAAK,OACD,EAAO,QAAa,KACpB,WACI,MAAA,wHAER,EAAc,EAAO,OACrB,MAAO,GAAO,OACd,EAAmB,EAGrB,GAAM,GAAQ,GAAI,GAAI,GACtB,GAAI,CAAE,aAAiB,IACrB,KAAM,IAAI,IACN,yDAAyD,GAE/D,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAA0C,OAC1C,EAAQ,GACI,EAAkC,EAClC,GAClB,AAAI,GACF,EAAM,6BAA6B,IAErC,EAAM,IAAI,GAEZ,MAAO,IA+BT,OAAA,eAAI,EAAA,UAAA,eAAY,KAWhB,WACE,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,MAAO,MAAK,MAAM,kBAjBpB,SAAiB,GAGf,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,KAAK,MAAM,aAAe,mCAe5B,EAAA,UAAA,UAAA,WAME,OADM,GAAqC,GACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAiC,GACvC,EAAK,UAAe,EAAM,eAC1B,EAAK,OAAY,EAAM,YACvB,EAAO,KAAK,GAEd,MAAO,CAAC,KAAM,KAAK,KAAM,OAAM,IAvsB1B,EAAA,UAAY,aAysBrB,GA3sBgC,IA4sBhC,EAAA,cAAc,cAAc,IC3kC5B,YAqEsB,GACpB,MAAO,IAAI,IAAY,eA+DE,GACzB,MAAO,IAAI,IAAW,eA+FpB,EACA,GACF,MAAI,IAAW,MACb,GAAU,IAEL,GAAwB,EAAiB,eA0B5B,GACpB,MAAO,IAAM,eAIX,EACA,GACF,GAA4B,4BACxB,EAAgB,GC5QtB,AAwBA,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAEE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GALyC,EAAA,cAAc,cAWvD,GAAA,SAAA,GAAyB,EAAA,EAAA,GAAzB,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,GAAM,EAAG,IATF,EAAA,UAAY,MAW9B,GAbyB,IAczB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAK,EAAA,KAAS,OAF9B,EAAA,UAAY,QAI9B,GAN2B,IAO3B,EAAA,cAAc,cAAc,IAG5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAFO,EAAA,UAAY,SAI9B,GAN4B,IAO5B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,QAAY,IAFL,EAAA,UAAY,UAI9B,GAN6B,IAO7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAAjC,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAc,IAFP,EAAA,UAAY,cAI9B,GANiC,IAOjC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,SAAa,IAFN,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAW,IAFJ,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAeE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,QAAY,EAAG,IAdR,EAAA,UAAY,UAgB9B,GAlB6B,IAmB7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DAgBE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,WAAe,EAAG,IAfX,EAAA,UAAY,aAiB9B,GAnBgC,IAoBhC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,EAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAE,IAAI,IAAQ,IAAI,MATlC,EAAA,UAAY,QAW9B,GAb2B,IAc3B,EAAA,cAAc,cAAc,gBAEQ,GAClC,MAAO,GAAW,2BAIjB,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACM,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,MAChB,GAAM,GAAmC,GACzC,SAAO,UAAe,SACtB,EAAO,OAAY,GACZ,GAAsB,GAE/B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCjQjC,AAiBA,YAA0B,GACxB,GAAI,GAAQ,MAAQ,MAAO,IAAS,SAClC,KAAM,IAAI,OACN,mEACA,0BAAyB,IAOjC,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DAEA,MAAA,IAF0C,EAAA,cAAc,cAqBxD,GAAA,SAAA,GAA0B,EAAA,EAAA,GAQxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAEP,UAAiB,GAEjB,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,MAAQ,EAAK,KAAO,EACzB,EAAK,MAAQ,EAAK,KAAO,IAO3B,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,GAAyB,EAAA,MAAM,CAAC,IACpC,MAAI,GAAK,OACP,GAAiB,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAI,OAEhE,AAAI,EAAK,OACP,GACI,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,GAAS,OAEjD,EAAe,cAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,GAAM,KAAK,GAAI,GAAM,KAAK,KAI7B,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,CAAC,GAAI,EAAO,GAAiB,GAAI,EAAO,MA3ClD,EAAA,UAAY,OA6CrB,GA/C0B,IAgD1B,EAAA,cAAc,cAAc,gBAET,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,gBAGvC,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,IAOnD,GAAM,IACyC,CAChD,KAAQ,oBAGuB,GAEnC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGU,GAG7B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAuB,OACzB,OAAI,aAAsB,IACxB,EAEA,GAAuB,GClJlC,AAmCA,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAKxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,MACV,GAAK,SAAW,EAAK,YAIzB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,EAAS,GAAoB,GAC7B,GAAI,GAAS,EAAA,KAAK,GAClB,MAAI,MAAK,UAAY,MACnB,GAAS,EAAA,YAAY,EAAQ,EAAG,KAAK,WAEhC,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,SAAU,KAAK,UACnD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA5BF,EAAA,UAAY,OA8BrB,GAhC0B,IAiC1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAO7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,GAIvB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,UAAU,EAAG,KAAK,QAG3B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,YA4BrB,GA9B+B,IA+B/B,EAAA,cAAc,cAAc,IA6B5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAWzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAU/B,GAbO,EAAA,0BAAmD,QAI1D,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,gBAAkB,GACvB,EAAK,iBACD,GAAe,EAAK,kBAAoB,EAAK,2BACjD,EAAK,iBAAmB,GAAe,EAAK,kBAC5C,EAAK,gBAAkB,GAAc,EAAK,iBACtC,EAAK,YAAc,KACrB,EAAK,WAAa,aACT,MAAM,QAAQ,EAAK,YAC5B,EAAK,WAAa,EAAK,mBACd,MAAO,GAAK,YAAe,SACpC,EAAK,WAAa,CAAC,EAAK,gBAExB,MAAM,IAAI,GACN,8DACA,YAAW,EAAK,sBAIxB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAChC,GAAM,GAAoB,EAAW,MAAM,GAC3C,GAAI,KAAK,YAAc,KACrB,OAAgB,GAAA,EAAA,EAAA,KAAK,WAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAW,EAAI,GAAK,EAGxB,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAAM,KAAK,iBAEtC,GAAM,GAAiC,GACvC,GAAI,KAAK,YAAc,KACrB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,EAAK,GAAK,EAAW,GAGzB,KAAK,UAAY,CAAC,GAAI,IAAU,CAC9B,KAAM,EAAW,OACjB,KAAI,KAEN,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,SAAS,GAAoB,GACtB,EAAA,MAAM,EAAQ,KAAK,MAAM,SAGlC,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,iBAAkB,GAAqB,KAAK,kBAC5C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAoB,KAAK,iBAC1C,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAxEF,EAAA,UAAY,QA0ErB,GA5E2B,IA6E3B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAK/B,GARO,EAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGL,EAAK,OAAS,MAAQ,EAAK,QAAU,EAAK,cAC5C,KAAM,IAAI,IACN,4BAA4B,EAAK,MAAK,4CAI5C,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,IAAI,IAGb,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjCF,EAAA,UAAY,MAmCrB,GArCyB,IAsCzB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAOnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAE,IAAI,GAAK,EAAE,QAAQ,KAAK,OAAQ,aAG3C,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,kBA6BrB,GA/BqC,IAgCrC,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,aAAe,EAItB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,QAAU,GAAI,MAAoB,MACvC,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,OAG3D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,MAAK,QAAQ,EAAG,KAAK,OAG9B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,KAAM,KAAK,MAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,UA6BrB,GA/B6B,IAgC7B,EAAA,cAAc,cAAc,ICxV5B,YAwBI,EAAwB,EAAW,GACrC,GAAI,MAAO,IAAU,SACnB,MAAO,IAAa,EAAO,GAE3B,GAAI,EAAM,SAAW,EACnB,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EAAC,aACxD,eAAc,EAAM,OAAM,eAEhC,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAc,EAAM,GAC1B,GAAI,CAAC,GAAU,GACb,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EACvD,yBAAwB,KAAK,UAAU,GAAM,gBAC7C,wBAAuB,IAG/B,MAAO,eAaP,EAAqB,EAAoB,EACzC,EAAgB,GAClB,GADkB,AAAA,IAAA,QAAA,GAAA,GACd,GAAe,KACjB,MAAO,GAET,GAAM,GAAoB,EAAc,GAAa,GAAM,GAAW,GAClE,EACJ,MAAI,KAAY,OACd,EAAe,EAEf,EAAe,EAAc,EAAoB,EAE5C,KAAK,MAAO,GAAe,EAAS,GAAK,eAI9C,EAAiB,EAAoB,EACrC,GACF,GAAI,GAAW,KACb,MAAO,MAGT,GAAI,IAAY,QACd,EAAU,EAAU,EAAa,GAAI,CAAC,EAAa,EAAY,YACtD,IAAY,OACrB,EAAU,EAAU,MAEpB,MAAM,IAAI,GAAW,2BAA2B,EAAO,KAEzD,MAAO,GCpFT,YAwCI,EAAW,GAEb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,IAE3B,gBAWT,EAAW,GACb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,IAE9B,gBAsBT,EAAW,EAAgB,EAAc,EAAa,EACtD,EAAyB,GADgB,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,SAC7B,AAAA,IAAA,QAAA,GAAA,GACpB,EAAA,KAAK,WAMV,GALA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAEZ,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,+DACG,GAAE,MAAM,OAAM,cAEvB,GAAI,EAAO,MAAM,SAAW,EAC1B,KAAM,IAAI,GACN,iEACG,GAAO,MAAM,OAAM,aAE5B,GAAI,GAAQ,MAAQ,EAAK,MAAM,SAAW,EACxC,KAAM,IAAI,GACN,+DACG,GAAO,MAAM,OAAM,aAM5B,GAHA,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,KAE1B,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,GAAI,GAAY,EAAA,OACZ,EAA0B,EAAoB,EAC9C,IAAY,OAAS,OAAS,QAAS,MAAO,GAClD,MAAI,IAAQ,MACV,GAAI,GAAU,EAAG,IAEZ,gBAoDP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EAAyB,EAC5C,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,IACvD,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,gEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,iEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,MAAU,OAAO,CACnB,EAAG,EACH,OAAQ,EACR,QAAS,EACT,IAAK,IAAY,OAAS,OAAS,QACnC,UAAW,EACX,WAAY,OACZ,KAAI,EACJ,WAAU,IAEZ,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBA8BP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EACnB,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,EAAG,IAC1D,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,MAEf,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,oEACG,GAAE,KAAI,MAEf,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,OACA,EACA,EAAmC,EACnC,IAAY,OAAS,OAAS,QAAS,QAAS,GACpD,AAAI,GAAQ,MACV,GAAI,GAAU,EAAG,IAEnB,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAmHX,GAAA,IAAA,SAAA,GAAuC,EAAA,EAAA,GAwBrC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,IAAkB,KAIxB,GAVQ,EAAA,KAAsB,KAEvB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAIzD,EAAS,WAAW,GACpB,EAAK,KAAO,EACZ,GAAoC,EAAK,KAAM,QAC3C,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,EAAK,OAAS,EACtD,KAAM,IAAI,IACN,qDACI,EAAK,KAAI,6BAqBnB,GAlBA,EAAK,WAAa,GAAe,EAAK,WAAY,EAAM,cACxD,EAAK,QAAU,GACX,EAAK,SAAW,KAAO,EAAI,EAAK,QAAS,EAAM,WACnD,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,WAAa,GAAc,EAAK,YACrC,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAClD,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,aAAe,GAChB,EAAK,cAAgB,KAAO,EAAI,EAAK,aAAc,EACnD,gBACA,EAAK,OAAS,GACb,MAAM,QAAQ,EAAK,eAAiB,EAAK,aAAa,SAAW,EACpE,KAAM,IAAI,GACN,iGAEA,IAAG,KAAK,UAAU,EAAK,gBACtB,GAAI,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aAAe,CAAC,EAAK,aAAc,EAAK,sBACpC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,gEACA,8BAA6B,KAAK,UAAU,EAAK,wBAE9C,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aACD,CAAC,EAAK,aAAc,EAAK,aAAc,EAAK,sBACvC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,kEACA,8BAA6B,KAAK,UAAU,EAAK,yBAK1C,SAAA,WAAjB,SAA4B,GAI1B,GAFA,GACI,cAAgB,GAAM,2CACtB,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,oEACA,oCACI,KAAK,UAAU,EAAK,YAAW,OAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,gBAAiB,GAAqB,KAAK,iBAC3C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA/GuC,IAqHvC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAcjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAA0B,KAZ9B,SAAA,OAAwB,KAahC,EAAK,WAAW,GAChB,EAAK,QAAU,EAAK,QACpB,GAAoC,EAAK,QAAS,WAClD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,kBAAoB,GAAe,EAAK,qBAG/C,SAAA,UAAA,MAAA,SAAM,SACJ,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAE1B,GAAM,GAAW,EAAW,GAEtB,EAAc,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,UAE3D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,KAAM,KAAK,gBACnC,KAAK,gBAAiB,GAAM,KAAK,iBAGvC,KAAK,UAAY,CAAC,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,KACtE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GACE,EAAY,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACjD,EAAsB,GACU,EAAK,WAAW,gBAEtD,GAAI,GAAuB,MAAQ,EAAK,OAAS,EAC/C,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,aACtB,QAEJ,GAAI,EAAK,OAAS,EAChB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAQ,GACpD,EAAK,QAAS,EAAK,WAAY,EAAK,aAAa,YAC5C,EAAK,OAAS,EAEvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,sBACjB,EAAK,OAAS,EACvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,kBAE1B,MAAM,IAAI,IACN,yDAGN,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAIpC,MAAO,MAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAKhC,OAJM,GAAqB,GACrB,EAAS,KAAK,aAAe,eAC/B,EAAW,MAAM,EAAG,EAAW,OAAS,GACxC,EAAW,MAAM,GACZ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAM,GAAS,GACX,EAAM,GAAI,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,GACzD,MAAO,MAAK,cAAiB,SAAW,KAAK,aACL,KAAK,aAAa,IAC9D,EAAS,KAAK,GAGhB,GAAI,GAAc,CAAC,EAAW,IAC9B,MAAI,MAAK,aAAe,eACtB,GAAc,EAAY,OAAO,GACjC,EAAY,KAAK,KAAK,UAEtB,GAAY,KAAK,KAAK,SACtB,EAAc,EAAY,OAAO,IAE5B,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,kBAAmB,GAAqB,KAAK,mBAC7C,iBAAkB,GAAoB,KAAK,mBAEvC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,CAAE,YAAa,KAAS,MAAO,GAAK,SAAY,UAChD,EAAK,QAAU,EACjB,KAAM,IAAI,GACN,kEACA,YAAW,KAAK,UAAU,EAAK,YAGzC,GA1ImC,IA4InC,GAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAK,MAAO,GAAK,YAAe,UAC5B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,gCAA+B,KAAK,UAAU,EAAK,YAAW,OAnB/D,EAAA,UAAY,SAsBrB,GAxB4B,IAyB5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UACzB,CAAE,OAAM,QAAQ,EAAK,aAClB,GAAK,WAAW,SAAW,GAAK,EAAK,WAAW,SAAW,IAChE,KAAM,IAAI,GACN,mDACA,4CACI,KAAK,UAAU,EAAK,YAAW,OApBtC,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAGX,GAFA,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KAEnC,EAAK,UAAY,QAAU,EAAK,UAAY,QAC9C,KAAM,IAAI,GACN,gEACA,2CAA0C,EAAK,mBAIvD,SAAA,UAAA,MAAA,SAAM,SAGJ,GAFA,EAAa,GAAmB,GAE5B,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,mDACA,KAAK,UAAU,IAGrB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,wEAGN,GAAM,GAAW,EAAW,GACtB,EAAc,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,IAE1D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,UAAW,KAAK,kBACvC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,GAAM,KAAK,iBAIvC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MAC3D,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAM,MAAM,SAAW,EACzB,KAAM,IAAI,GACN,iEACA,8BAA6B,EAAM,MAAM,SAG/C,GAAM,GAAa,EAAM,MACnB,EAAY,EAAW,GAEzB,EACA,EACJ,AAAI,EAAK,aAAe,gBACtB,GAAQ,EACR,EAAQ,GAER,GAAQ,EACR,EAAQ,GAGV,GAAM,GAAS,EAAW,GACpB,EAAQ,EAAW,GACnB,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,QAAQ,GACvB,EAAU,EAAK,QAAQ,GAGvB,EAAY,GAAa,EAAQ,EAAS,EAAS,EAAK,SACxD,EAAW,GAAa,EAAO,EAAS,EAAS,EAAK,SAMtD,EACF,CAAC,EAAW,EAAW,EAAU,EAAK,SAE1C,AAAI,EAAK,aAAe,gBACtB,GAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,KAEzC,GAAI,GAAU,EAAA,gBACV,EAAmB,EAAK,OAAO,OAAoB,EACnD,EAAK,QAA6B,EAAK,SAC3C,MAAI,GAAK,aAAe,gBACtB,GAAU,EAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,KAG7C,AAAI,EAAK,MAAQ,MACf,GACI,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEhD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAE3B,EACA,EACA,EACJ,AAAI,KAAK,aAAe,gBACtB,GAAc,EACd,EAAa,EACb,EAAY,GAEZ,GAAc,EACd,EAAa,EACb,EAAY,GAGd,GAAM,GAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,QAAQ,GACvB,EAAU,KAAK,QAAQ,GAE7B,SAAY,GAAe,KAAK,QAChC,EAAY,GACR,GAAa,EAAY,GAAa,EAAS,EAAS,KAAK,SACjE,EAAY,GACR,GAAa,EAAY,GAAY,EAAS,EAAS,KAAK,SACzD,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,aACP,GA9IF,EAAA,UAAY,kBAgJrB,GAlJqC,IAmJrC,EAAA,cAAc,cAAc,IA0C5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAqBjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAAO,KAEnB,GAXO,EAAA,8BACL,gBACK,EAAA,8BACL,gBAEM,EAAA,gBAAiC,KACjC,EAAA,gBAAiC,KAKrC,EAAO,SAAW,KACpB,KAAM,IAAI,GACN,uFAGN,GAAI,EAAO,mBAAqB,MAAQ,EAAO,mBAAqB,MAChE,EAAO,kBAAoB,KAC7B,KAAM,IAAI,GACN,sPAKN,GAAI,EAAO,SAAW,MAAQ,EAAO,UAAY,QAC7C,EAAO,UAAY,QACrB,KAAM,IAAI,GACN,gBAAgB,EAAK,KAAI,kCACzB,qCAAoC,KAAK,UAAU,EAAO,WAGhE,SAAK,gBACD,EAAO,iBAAmB,KAAO,EAAI,EAAO,gBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,qBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,uBAGlD,SAAA,UAAA,MAAA,SAAM,SAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,KAAK,KAAO,EAClC,KAAM,IAAI,GACN,0BAA0B,KAAK,KAAI,sBAChC,MAAK,KAAO,EAAC,gCAChB,IAAG,KAAK,UAAU,KAExB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,0DACA,cAAa,KAAK,UAAU,EAAW,MAO7C,OAJM,GAAW,EAAW,GACtB,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,kBACrC,EAAuB,GACpB,EAAI,EAAG,EAAI,KAAK,KAAM,EAAE,EAC/B,EAAqB,KAAK,GAE5B,EAAqB,KAAK,EAAW,KAAK,gBAAiB,KAAK,SAEhE,GAAM,GAAY,GAClB,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,EAAW,KAAK,gBAE1C,KAAK,KAAO,KAGd,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MACvE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAE7B,GAAI,GACJ,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,IACN,oDACC,MAAI,GAAK,OAAS,GACvB,CAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAG3C,EAAS,EAAA,gBACL,EAAoB,EAAK,gBAAgB,OACzC,EAAK,gBAAgB,OACrB,EAAK,QAA6B,EAAK,QACvC,EAAK,aAAkC,SAG7C,AAAI,EAAK,SACP,GAAS,GAAU,EAAQ,EAAK,KAAK,OAAQ,EAAK,aAEpD,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAGjC,AAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAEpC,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,kBACd,MAAO,GAAO,kBACd,MAAO,GAAO,iBACd,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,qBAC7B,EAAO,oBACH,GAAoB,KAAK,qBACtB,GAzJF,EAAA,UAAY,gBA2JrB,GA7JmC,IA+JnC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,SACV,GAAA,KAAA,KAAM,EAAG,IAAK,KAFT,SAAA,UAAY,kBAIrB,GANqC,IAOrC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,GAClB,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,WACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,2BAA0B,KAAK,UAAU,EAAK,YAAW,OArB1D,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAgC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAM9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,MAAI,OAAO,GAAK,UAAa,SAC3B,EAAK,SACD,CAAC,CAAC,EAAK,SAAU,EAAK,UAAW,CAAC,EAAK,SAAU,EAAK,WACrD,AAAI,MAAO,GAAK,SAAS,IAAO,SACrC,EAAK,SAAW,CACd,CAAC,EAAK,SAAS,GAAI,EAAK,SAAS,IACjC,CAAC,EAAK,SAAS,GAAc,EAAK,SAAS,KAG7C,EAAK,SAAW,EAAK,SAEvB,EAAK,WACD,EAAK,aAAe,OAAY,eAAiB,EAAK,WAC1D,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAI,MAAK,aAAe,gBACf,CACL,EAAW,GAAI,EAAW,GAC1B,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,IAGlD,CACL,EAAW,GACX,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GAAI,EAAW,KAK5E,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAFA,EAAS,GAAoB,GAEzB,EAAK,aAAe,gBACtB,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,QAEjE,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,OAKvE,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,SAAU,KAAK,SAAU,WAAY,KAAK,YACpD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAhEF,EAAA,UAAY,aAkErB,GApEgC,IAqEhC,EAAA,cAAc,cAAc,IAwB5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAOhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KALM,SAAA,aAAe,CAAC,EAAG,GAMpC,EAAK,UAAY,CAAC,CAAC,KAAM,IACzB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,aAGtD,SAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,KAAK,aAAe,iBACtB,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,QAE9C,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAO,EAAW,MAIrD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAC1B,EAAa,EAAM,MAEzB,GAAI,EAAK,aAAe,iBACtB,EAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,IACvC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GAClC,EAAU,EAAM,sBAAsB,CAAC,EAAQ,IACrD,MAAO,GAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,SAExC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GACxC,MAAO,GAAM,sBAAsB,CAAC,EAAQ,QAKlD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,KAAM,KAAK,KAAM,WAAY,KAAK,YAC5C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlDF,EAAA,UAAY,eAoDrB,GAtDkC,IAuDlC,EAAA,cAAc,cAAc,ICrvC5B,YA6CI,EAAW,EAAyB,EACpC,EAAmB,EACnB,GAFoC,MAAA,KAAA,QAAA,GAAA,CAA6B,EAAG,IACpE,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WACV,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAChB,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,OAEf,GAAI,EAAgB,OAAS,EAC3B,KAAM,IAAI,GACN,yDACG,GAAgB,KAAI,OAE7B,SAAI,EAAA,gBACA,EAAe,EAA6B,EAC5C,IAAY,OAAS,OAAS,QAAS,OAAQ,GACnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,IAsCX,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAsB,KAHzB,SAAA,gBAAiC,KAIvC,EAAK,gBACD,EAAK,iBAAmB,KAAO,EAAI,EAAK,gBAC5C,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,4BACtC,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,qBAAuB,GAAe,EAAK,wBAGlD,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,iDACA,0BAAyB,KAAK,UAAU,GAAW,MAEzD,GAAM,GAAc,KAAK,aAAe,gBAAkB,EAAI,EAC9D,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,iEACA,4BAA2B,EAAW,GAAY,OAExD,GAAM,GAAW,EAAW,GACtB,EAA8B,CAClC,KAAK,WAAW,GAAI,KAAK,WAAW,GAAI,EAAU,KAAK,iBAGzD,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,EAAW,KAAK,iBAAkB,KAAM,KAAK,gBACtD,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GAAU,GACV,EAAQ,EAAK,gBAAgB,OAAQ,EAAK,QAC1C,EAAK,QAAS,EAAK,WAAY,MAEnC,MAAI,GAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEtD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EAAa,KAAK,aAAe,gBACnC,EAAW,GAAK,KAAK,gBACrB,EAAW,GAAK,KAAK,gBACnB,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACnD,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,MAAI,MAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAY,EAAS,GAGrC,CAAC,EAAW,GAAI,EAAS,EAAS,IAI7C,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,SAAO,gBAAqB,KAAK,gBACjC,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,sBACtB,GAjGF,EAAA,UAAY,kBAmGrB,GArGqC,IAsGrC,EAAA,cAAc,cAAc,IClN5B,YA0DI,EACA,EACA,EACA,GAKF,GAAI,MAAM,QAAQ,IAChB,GAAI,GAAgB,MAAQ,GAAa,KACvC,KAAM,IAAI,GACN,iFAGN,AAAI,GAAgB,MAClB,GAAY,EAAO,MAAM,EAAO,OAAS,EAAc,EAAO,QAC9D,EAAS,EAAO,MAAM,EAAG,EAAO,OAAS,IAE3C,AAAI,EAAO,OAAS,GAClB,GAAe,EAAO,MAAM,EAAG,EAAO,SAExC,EAAS,EAAO,GAGlB,WAAsB,GAEpB,MAAI,IAAK,MAAQ,MAAM,QAAQ,GACtB,EAEA,CAAC,GAIZ,SAAe,EAAa,GAC5B,EAAY,EAAa,GAElB,CAAC,OAAM,EAAE,aAAY,EAAE,UAAS,eA+CrC,EAA+B,EAAgB,EAC/C,EAAqB,EAAe,EAAsB,EAC1D,GADA,MAAA,KAAA,QAAA,GAAA,IAA0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,IACK,EAAA,KAAS,WACd,GAAM,GAAO,EAAO,MAAM,OAC1B,GAAI,EAAO,EACT,KAAM,IAAI,GAAW,uCAAuC,EAAI,MAKlE,GAAM,GAAO,CAAC,EAAG,GAAG,OAAO,GAAiB,EAAG,IAG/C,GAFA,EAAS,EAAA,UAAc,EAAQ,GAE3B,GAAa,KACf,KAAM,IAAI,IACN,kFAKN,AAAI,GACF,QAAQ,KACJ,qGAIN,AAAI,GAAQ,MACV,GAAO,EAAK,OAAO,QAAQ,OAAO,WAClC,AAAI,EAAK,OAAS,EAAO,GACvB,GAAO,EAAA,WAAe,EAAM,KAE9B,EAAO,EAAA,UAAc,EAAM,IAG7B,AAAI,GACF,GAAS,EAAA,QAAY,EAAQ,GAC7B,AAAI,GAAQ,MACV,GAAO,EAAA,QAAY,EAAM,KAc7B,GAAM,GAA2B,GAC7B,EACA,EAAS,EACP,EAAY,EAAO,MAAM,GACzB,EAAgB,EAAA,QAAY,GAC9B,EACJ,AAAI,GAAQ,MACV,GAAe,EAAA,QAAY,IAG7B,mBAAS,GACP,GAAM,GAAe,EAAc,GAC7B,EAAc,EAAA,KAAS,WAAM,MAAA,GAAa,EAAc,KAE9D,GAAI,GAAQ,KACV,EAAa,EAAY,GACzB,EAAS,EAAY,QAErB,GAAM,GAAgB,EAAA,KAAS,WAC7B,GAAM,GAAW,EAAa,GACxB,EAAc,EAAA,SAAa,GAAU,IAAI,GAEzC,EACF,EAAY,GAAG,IAAI,GAAU,IAAI,EAAO,GAAG,IAAI,IAC7C,EAAY,EAAO,IAAI,SAAC,EAAO,GACnC,MAAO,GAAY,GAAG,GAAG,IAAI,GAAU,IAAI,EAAM,IAAI,MAEvD,MAAO,CAAC,OAAM,EAAE,UAAS,KAE3B,EAAa,EAAc,OAC3B,EAAS,EAAc,UAGzB,AAAI,GACF,EAAe,KAAK,IAxBf,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,GA2BT,GAAI,GACJ,GAAI,GACF,GAAM,GAAO,EACb,EAAU,EAAA,MAAU,EAAgB,GAEtC,MAAO,CAAC,EAAY,EAAS,wBAyGR,EAAA,EAAA,GAqBvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACP,EACJ,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,GACN,wDAMN,GALO,AAAI,MAAM,QAAQ,EAAK,MAC5B,EAAO,GAAI,IAAgB,CAAC,MAAO,EAAK,OAExC,EAAO,EAAK,KAEV,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,qGAGN,SAAK,KAAO,EACZ,EAAK,gBACD,EAAK,iBAAmB,KAAO,GAAQ,EAAK,gBAChD,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,UAAY,EAAK,UAAY,KAAO,GAAQ,EAAK,SACtD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAQ,EAAK,OAEjD,EAAK,gBAAkB,GACvB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KACvC,EAAK,UAAY,KACjB,EAAK,QAAU,KAEf,EAAK,aAAe,KAIpB,EAAK,WAAa,KAKpB,SAAA,UAAA,UAAA,WACE,GAAI,KAAK,SAAW,MAClB,GAAM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EACtE,MAAO,IAAiB,EAAG,GAAW,IAAI,SAAA,GAAK,MAAA,YAE/C,OAAO,MAAK,SAMhB,EAAA,UAAA,UAAA,SAAU,GACR,KAAK,QAAU,GAGjB,EAAA,UAAA,mBAAA,SAAmB,GACjB,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAGb,GAAI,GAAY,KAAK,KAAK,UAC1B,AAAK,MAAM,QAAQ,IACjB,GAAY,CAAC,IAEf,GAAM,GAAY,EAAU,GACxB,EAOJ,GANA,AAAI,KAAK,gBACP,EAAc,CAAC,EAAW,GAAI,EAAW,GAAI,GAE7C,EAAc,CAAC,EAAW,GAAI,GAG5B,KAAK,aAEP,OADM,GAAsB,GACV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAW,KAAK,CAAC,EAAW,GAAI,IAElC,MAAO,CAAC,GAAa,OAAO,OAE5B,OAAO,IAIX,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAS,WACd,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAM,GAAa,EAAK,gBAAkB,EAAO,KAEjD,GAAI,EAAK,aACP,GAAM,GAAY,EAAK,OAAO,IAAI,SAAA,GAAK,MAAA,QACvC,MAAO,CAAC,GAAY,OAAO,OAE3B,OAAO,MAWb,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,SAAW,MAIlB,OAHM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EAChE,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,EAAO,KAAK,MAEd,MAAO,OAEP,OAAO,MAAK,aAIhB,SAAW,GACT,KAAK,QAAU,mCAGV,EAAA,UAAA,MAAP,SAAa,GAIX,GAAI,KAAK,cAAgB,KACvB,KAAM,IAAI,IACN,oDAGN,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAEb,GAAM,GAAoB,KAAK,SAAW,EAAW,GAAK,KACpD,EAAW,EAAW,MAAM,GAClC,KAAK,UAAU,GAAK,GAAI,IAAU,CAAC,MAAK,CAAG,EAAW,MAAI,OAAK,KAI/D,GAAM,GAAiB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAK7D,KAAK,KAAK,MAAM,GAIlB,GAAI,GAOJ,GANA,AAAI,MAAM,QAAQ,KAAK,KAAK,WAC1B,EAAY,KAAK,KAAK,UAEtB,EAAY,CAAC,KAAK,KAAK,WAGrB,KAAK,WAAa,MACpB,GAAI,CAAC,EAAA,KAAK,YACF,KAAK,UAAU,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,EAAK,MAAM,OAAS,KAC1D,GACN,KAAM,IAAI,GACN,0DACA,uCAAsC,KAAK,UAAS,MACpD,8BAA6B,KAAK,KAAK,gBAG7C,MAAK,UACD,EAAU,IAAI,SAAA,GAAO,MAAA,IAAI,IAAU,CAAC,MAAO,CAAC,KAAM,OAExD,AAAI,KAAK,UACP,KAAK,eAqBT,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAK,WACH,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAEN,GAAM,GAAY,EAAK,UAAU,GAAG,MAAM,GAC1C,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUAQN,GAAI,EAAK,SAAW,KAClB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAU,CAAC,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,qBAEzC,GAAU,KAEnB,EAAA,QAAY,EAAK,SAEjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAQ,GAAK,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,iBAMpD,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAER,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,IAAa,GAKf,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GACf,EAAM,MAAM,QAAQ,EAAK,KAAK,WAChC,EAAK,KAAK,UAAU,GACpB,EAAK,KAAK,UACR,EAAgB,CAAC,EAAW,GAClC,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAEhB,EAAK,QAAQ,GAAS,GAG1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAI5D,EAAA,UAAA,MAAA,SACI,EACA,GAEF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAGX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAC1D,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAMzB,GAAI,GAAiD,GACjD,EAA+B,GACnC,GAAI,GAAgB,MAClB,EAAO,aAAkB,EACzB,EAAmB,EAAiB,OAAO,GAC3C,KAAK,UAAY,GACjB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,UAAU,KAAK,GAAI,IAAU,CAAC,MAAO,EAAM,SAKlD,EAAkB,EAAgB,OAAO,KAAK,WAEhD,AAAI,GAAa,MACf,GAAO,UAAe,EACtB,EAAmB,EAAiB,OAAO,GAE3C,KAAK,aAAe,EAAU,QAGhC,GAAM,GAAW,EAAiB,YAAc,IAChD,GAAI,GAEF,GAAM,GACF,CAAC,GAAQ,OAAO,GACd,EAAgB,KAAK,UAAU,OAAO,GAEtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GAAS,EAAA,UAAM,MAAK,KAAA,KAAC,EAAW,GACtC,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAK/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KAIE,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC5C,EACA,GAAU,KAAO,KAAO,EAAO,aAEnC,EAAS,GAAoB,GAC7B,AAAI,GAAgB,MAClB,CAAI,EAAK,SACP,EAAe,EAAK,QAEpB,EAAe,EAAK,gBAAgB,IAIxC,GAAM,GACF,MAAM,QAAQ,EAAK,KAAK,WAAa,EAAK,KAAK,UAAU,OAAS,EACtE,GAAI,EAAa,SAAW,EAC1B,KAAM,IAAI,GACN,iBAAiB,EAAS,4BACvB,GAAa,OAAM,uBAE5B,AAAI,EAAK,QACP,QAAQ,KACJ,oEAGN,GAAM,GAAyB,CAAC,SAAQ,GAGlC,EAAO,SAAC,EAAgB,GAG5B,GAAM,GACF,EAAK,KAAK,KAAK,CAAC,GAAQ,OAAO,GAAS,GAE5C,MAAO,CAAC,EAAQ,GAAI,EAAQ,MAAM,KAK9B,EACF,GAAI,EAAM,EAAQ,EAAc,EAAK,YAAa,EAAM,KACpD,EAAK,OAAQ,EAAK,iBACpB,EAAa,EAAW,GACxB,EAAU,EAAW,GACrB,EAAS,EAAW,GAE1B,AAAI,EAAK,UACP,EAAK,YAAY,EAAQ,GAG3B,GAAM,GAAS,EAAK,gBAAkB,EAAU,EAIhD,MAAI,GAAK,YACA,CAAC,GAAQ,OAAO,GAEhB,KAKb,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAAI,GAAe,EAAA,MAAU,EAAO,OAKpC,MAHA,GAAe,EAAA,IAAQ,EAAc,CAAC,EAAG,IACzC,EAAe,GAAa,GAExB,MAAM,QAAQ,EAAK,KAAK,WACnB,EAAK,KAAK,UAAU,IACvB,SAAA,GAAO,MAAA,GAAM,EAAI,GAAO,EAAc,CAAC,EAAG,IAAQ,IAE/C,EAAK,KAAK,UAAY,EACzB,CAAC,GAAO,EAAc,CAAC,EAAG,EAAK,KAAK,aACpC,CAAC,MAKX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAK,MAAK,UAIH,KAAK,KAAK,iBAHR,oCAMX,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,MAAK,MAAK,UAGH,KAAK,KAAK,oBAFR,KAAK,KAAK,yCAKrB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,MAAQ,MACf,KAAK,KAAK,6BAA6B,IAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,gBAAiB,KAAK,gBACtB,YAAa,KAAK,YAClB,YAAa,KAAK,YAClB,SAAU,KAAK,SACf,OAAQ,KAAK,QAGf,AAAI,KAAK,cAAgB,MACvB,GAAO,aAAkB,KAAK,cAGhC,GAAM,GAAa,KAAK,KAAK,YAE7B,MAAI,MAAK,iBAAmB,EAAI,WAC9B,GAAO,KAAU,CACf,UAAa,KAAK,KAAK,eACvB,OAAU,IAKd,GAAA,GAAW,EAAe,EAAe,IAIpC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAa,EAAO,KACpB,EAAO,GAAY,EAAY,GACrC,MAAO,IAAI,GAAI,OAAO,OAAO,EAAQ,CAAC,KAAI,MArfrC,EAAA,UAAY,MAufrB,GAzfyB,IA0fzB,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAAtC,4DAQA,MAAA,IARsC,IAwFtC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAiCjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KANJ,SAAA,mBAAqB,OACrB,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAIzD,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,YAAc,KAAO,EAAK,mBAAqB,EAAK,YAC7D,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAG9B,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAEhC,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAW,EAAW,OAAS,GAAI,KAAK,OAAQ,KAC3D,KAAK,kBAAmB,KAAK,kBAAmB,GAChD,KAAK,kBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,OAAQ,KAC9C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IASf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8CAA8C,EAAO,OAAM,KAEjE,GAAI,GAAa,EAAO,GACxB,EAAS,EAAO,GAChB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,KAG/B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,KAGxC,GAAI,GACE,EAAiB,EAAK,YACtB,EAAoB,EAAK,qBAC/B,AAAI,GAAU,KACZ,EAAI,GAAM,EAAA,IAAQ,EAAQ,GAAS,EAAK,OAAO,QAE/C,EAAI,GAAM,EAAQ,EAAK,OAAO,QAEhC,AAAI,EAAK,MAAQ,MACf,GAAI,GAAU,EAAG,EAAK,KAAK,SAE7B,AAAI,GAAa,MACf,GAAa,EAAA,IAAQ,EAAY,IAEnC,GAAI,GAAS,EAAA,IAAQ,EAAG,GAAM,EAAY,EAAK,gBAAgB,SAC/D,MAAI,GAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAI1B,CAAC,EAAQ,MAIpB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,kBAGzB,MAAA,IAAA,GAAW,EAAe,IArKrB,EAAA,UAAY,gBAuKrB,GAzKmC,IA0KnC,EAAA,cAAc,cAAc,IA2F5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAG7B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAK,KAAO,GAAI,IAAc,GAC9B,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IA7BV,EAAA,UAAY,YA+BrB,GAjC+B,IAkC/B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAqC3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAbO,EAAA,mBAAqB,OACrB,EAAA,6BAAqD,cAErD,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAQrD,EAAK,WACP,KAAM,IAAI,GACN,+DAEN,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,KAAK,gBACrC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,uDACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACzD,EAAW,EAAO,GACtB,EAAS,EAAO,GAKhB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EAAY,EAAK,qBACnB,EACA,EACA,EAEJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAU,GAAM,EAAQ,EAAK,OAAO,QACxC,AAAI,EAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,SAEzC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAGzC,GAAM,GAAuB,EAAK,gBAAgB,OAC5C,EAAA,EAAA,MAAA,EAAA,CAAA,EAAA,EAAA,MAAA,EAAA,OAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAGN,EAAc,GAAM,EAAU,GAE9B,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GACT,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAE/C,GAAM,GAAa,GAAM,EAAA,IAAQ,EAAG,GAAW,GAC/C,EAAK,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,IAEvC,GAAM,GACF,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,IAAK,IAElE,MAAO,CAAC,EAAG,MAIf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,eACrB,WAAY,IAGd,MAAA,IAAA,GAAW,EAAe,IAvMrB,EAAA,UAAY,UAyMrB,GA3M6B,IA4M7B,EAAA,cAAc,cAAc,IA8B5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAQ,GACxB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,MAuCrB,GAzCyB,IA0CzB,EAAA,cAAc,cAAc,IAuC5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAsC5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAZJ,SAAA,mBAAqB,OACrB,EAAA,6BAA+B,cAC/B,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAEhC,EAAA,yBAA2B,QASlC,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,EAAK,eAE3B,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,CAAC,EAAK,MAAO,EAAK,OACnC,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,GAAI,GACJ,GAAI,KAAK,SACP,GAAI,KAAK,gBACP,GAAM,GAAmB,KAAK,gBACxB,EAAgB,KAAK,MAC3B,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GAAM,GAAK,EAAiB,MAAM,CAAC,IAC7B,EAAM,GAAI,MAAQ,MAAM,CAAC,IACzB,EAAS,EAAiB,MAAM,CAAC,EAAgB,IACvD,MAAO,IACH,GAAuB,EAAI,GAAK,IAExC,GAZgD,IAEvC,EAAA,UAAY,oBAYrB,GAAkB,KAAK,gBAEzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,EAAiB,KAAK,gBACtD,GAAM,KAAK,oBAEf,MAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,wDACG,GAAO,OAAM,MAEtB,GAAI,GAAW,EAAO,GAChB,EAAW,EAAO,GACxB,EAAS,EAAO,GAChB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EACF,EAAK,qBAKL,EACA,EACA,EACA,EACJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAI,GAAM,EAAQ,EAAK,OAAO,QAClC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAEzC,EAAI,EAAA,IAAQ,EAAG,GAAM,EAAU,EAAK,gBAAgB,SACpD,AAAI,EAAK,SACP,GAAI,GAAU,EAAG,EAAK,KAAK,SAGvB,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,KACnE,EAAI,EAAK,oBAAoB,MAAM,GAEnC,GAAM,GAAI,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,IAE3C,MAAO,CAAC,EAAG,EAAG,MAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,eAAgB,KAAK,eACrB,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,gBAGvB,MAAA,IAAA,GAAW,EAAe,IAnNrB,EAAA,UAAY,WAqNrB,GAvN8B,IAwN9B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAGxB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAS,GACzB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,OAuCrB,GAzC0B,IA0C1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAME,OADM,GAAsB,GACT,EAAA,EAAA,EAAA,KAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAU,KAAI,MAAd,EAAkB,EAAK,WAEvB,EAAU,KAAK,EAAK,WAGxB,MAAO,oCAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,EAKT,OAJI,GAAS,EAAO,MAAM,GAGpB,EAA2B,GACd,EAAA,EAAA,EAAA,EAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAa,KAAK,EAAO,OAAO,EAAG,EAAK,UAAU,SAElD,EAAa,KAAK,EAAO,OAAO,EAAG,IAGvC,EAAa,UAKb,OAFM,GAA8B,GAChC,EACK,EAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAO,EAAK,MAAM,GACxB,EAAS,EAAa,GAEtB,AAAI,IAAM,EACR,EAAa,CAAC,EAAO,IAAI,OAAO,GAEhC,EAAa,CAAC,EAAW,IAAI,OAAO,GAEtC,EAAa,EAAK,KAAK,EAAY,GACnC,EAAgB,KAAK,EAAW,MAAM,IAIxC,EAAS,GACT,OAAyB,GAAA,EAAA,EAAA,EAAgB,QAAQ,UAAxB,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAO,KAAI,MAAX,EAAe,GAEjB,MAAO,CAAC,EAAW,IAAI,OAAO,MAI3B,EAAA,UAAA,MAAP,SAAa,GACX,AAAI,GAAgB,IAGlB,GAAc,EAAuB,IAEvC,EAAa,EACb,GAAI,GACJ,KAAK,MAAM,QAAQ,SAAC,EAAM,GACxB,GAAU,WAAW,EAAK,WAGxB,EAAK,MAAM,GACX,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAY,EAAK,UAAU,GAE3B,EAAY,EAAK,UAEnB,EAAa,CAAC,EAAW,GAAI,OAGjC,KAAK,MAAQ,IAGf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAgB,SAAC,GACrB,MAAO,CACL,UAAa,EAAK,eAClB,OAAU,EAAK,cAIb,EAAc,KAAK,MAAM,IAAI,GAE7B,EAAS,CAAC,MAAS,GAEzB,MAAA,IAAA,GAAW,EAAe,IAIrB,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAElB,OADM,GAAmB,GACA,EAAA,EAAA,EAAC,EAAO,MAAR,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAM,KAAK,GAAY,EAAY,IAErC,MAAO,IAAI,GAAI,CAAC,MAAK,KAGvB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,kBAEvB,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,qBAEvB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACvB,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAiB,KAAI,MAArB,EAAyB,EAAK,kBAEhC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAQT,EAAA,UAAA,WAAA,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,SAEvB,MAAO,IAAc,IASvB,EAAA,UAAA,WAAA,SAAW,GAET,OADM,GAAyC,GAC5B,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,IAGjB,OAHS,GAAI,EAAA,GACP,EAAY,EAAK,QAAQ,OACzB,EAAe,EAAQ,OAAO,GAC3B,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,EACzC,EAAO,KAAK,CAAC,EAAK,QAAQ,GAAI,EAAa,KAG/C,GAAc,IA5KT,EAAA,UAAY,kBAgLrB,GAlLqC,IAmLrC,EAAA,cAAc,cAAc,gBAEQ,GAM3B,GAAA,GAAA,EAAA,KAAM,EAAA,EAAA,KAAM,EAAA,EAAA,SAAA,EAAA,IAAA,OAAA,GAAA,EAAkB,EAAA,EAAA,MAAA,EAAA,IAAA,OAAA,EAAA,EAE/B,EAAgB,WAAM,MAAA,IAAU,IAAQ,IAExC,EAAa,WAAM,MAAA,IAAe,EAAe,EAAM,IAG7D,GAAI,CAAC,GAAS,GAAS,EACrB,MAAO,GAAA,KAAS,IAAa,SAG/B,GAAM,GAAQ,MAAM,GAAO,KAAK,QAAW,IAAI,GAE/C,MAAO,GAAM,IAAI,SAAA,GAAK,MAAA,GAAA,KAAS,EAAE,WC/iEnC,AAiFA,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAArC,4DAyBA,MAAA,IAzBqC,IAiCrC,GAAA,SAAA,GAAwB,EAAA,EAAA,GAMtB,WAAY,GAAZ,GAAA,GAAA,KACE,GAAI,EAAK,OACP,KAAM,IAAI,IACN,sDAGN,GAAI,MAAM,QAAQ,EAAK,MACrB,KAAM,IAAI,IACN,kEAGN,SAAA,EAAA,KAAA,KAAM,IAAqB,KAE3B,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WAad,GAZA,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aAEtB,EAAK,KAAK,YAAc,MAG1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBAEtB,EAAK,KAAK,qBAAuB,MAG/B,GAAU,EAAO,UACnB,KAAM,IAAI,GAAW,6CAGvB,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KAEtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAE1C,EACF,GAAU,KAAO,KAAO,EAAO,aAEnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAI3D,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GAAkB,KAAK,yBAAyB,GAEpD,MAAK,MAAK,iBACR,GAAQ,CAAI,EAAS,IAAE,OAAK,EAAS,MAAM,KAG7C,AAAI,KAAK,aACP,GAAQ,CACH,GAAQ,OAAK,MAAM,GAAG,KAAI,CAAE,EAAW,IAAE,OAAK,EAAS,MAAM,QAG7D,GAGT,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACP,GAAA,GAAA,EAAA,KAAA,UAED,EAAa,EAAO,MAEpB,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAe,EAAA,MAAU,GAE/B,MAAI,OAAM,QAAQ,GACT,MAAM,EAAU,QAAQ,KAAK,GAG/B,CAAC,MAIZ,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAS,WACP,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAGN,GAAM,GAAa,EAAK,UAAU,GAAG,MAE/B,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAY,EAAW,GAE7B,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUASN,GAAI,EAAK,aAAe,KACtB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAU,CAAC,EAAA,MAAU,YAEnB,GAAU,KAEnB,EAAA,QAAY,EAAK,SAGjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAQ,GAAK,EAAA,MAAU,QAO9B,GAJA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAGR,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,EAKF,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GAEf,EAAgB,EAEtB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAGhB,EAAK,QAAQ,GAAS,GAI1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAIlD,EAAA,UAAA,yBAAV,SAAmC,GAC3B,GAAA,GAAA,KAAA,KAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,QAAS,EAAA,EAAA,aAGpD,EAAkB,IAAe,gBAEjC,EAAI,EAAW,EAAkB,EAAI,GACrC,EAAI,EAAW,EAAkB,EAAI,GAErC,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAClD,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAElD,EACD,EAAW,MAAM,EAAG,GAAE,OACrB,EAAkB,CAAC,EAAS,EAAM,GAAQ,CAAC,EAAM,EAAM,IAG7D,MAAO,IAhMF,EAAA,UAAY,YAkMrB,GApMwB,IAyMxB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAWlC,WAAY,GAAZ,GAAA,GAAA,KAEI,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,aAGF,SAAA,EAAA,KAAA,KAAA,GAAA,GAAU,EAAI,CAAE,MAAO,MAAS,KAEhC,EAAK,QAAU,EACf,GAAsB,EAAK,QAAS,WAEpC,EAAK,WAAa,GAAe,EAAY,EAAG,cAChD,EAAK,WAAW,QAAQ,SAAA,GAAQ,MAAA,IAAsB,EAAM,gBAE5D,EAAK,QAAU,GAAe,GAAW,EAAG,EAAG,WAC/C,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,IAAsB,EAAQ,aAE7D,EAAK,QAAU,GAAW,QAC1B,GAAiB,EAAK,SAEtB,EAAK,WAAa,GAAc,eAChC,GAAgB,EAAK,YAErB,EAAK,aAAe,GAAe,GAAgB,EAAG,EAAG,gBACzD,EAAK,aAAa,QACd,SAAA,GAAQ,MAAA,IAAsB,EAAM,oBAGnC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAEhC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAElE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAG1B,GAAM,GAAW,EAAW,GAEtB,EAAe,EAEf,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,QAAU,IAErD,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBAEvC,GAAM,GACF,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,KAAK,QAAU,IAOzD,GALA,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBAEL,KAAK,SACP,GAAI,GAAe,OAEnB,GAAI,KAAK,gBACP,GAAM,GAAO,KAAK,gBAEZ,EAAU,KAAK,QAErB,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAQ,EAAK,MAAM,CAAC,IACpB,EAAQ,EAAA,KAAS,CAAC,IAClB,EAAY,EAAK,MAAM,CAAC,EAAU,IACxC,MAAO,IAAc,CAAC,EAAO,EAAO,KAExC,GAVgD,IAEvC,EAAA,UAAY,oBAUrB,GAAkB,KAAK,gBAGzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,QAAU,GAAe,KAAM,EAC7C,KAAK,gBAAiB,GAAM,KAAK,gBAGvC,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAsB,GAA3B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8DACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,GAEjC,EAAI,EAAO,GACX,EAAW,EAAO,GAClB,EAAW,EAAO,GAElB,EAAe,EAIrB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAI9B,GAAM,GAAc,EAAK,YAEnB,EACF,SAAC,GAAe,GAAoB,IAClC,MAAI,CAAC,IAAQ,CAAC,GAAK,IACV,GAGF,EAAA,IAAQ,GAAK,IAAQ,KAG9B,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAEtC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAIvC,GAAM,GAAiB,EAAK,qBAExB,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAE1C,EAAoB,EAEpB,EAAA,EAAA,MAAA,EAAA,OAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAG5B,EAAA,EAAA,uDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAI5B,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAEvC,GAAA,GAAA,EAAA,MAAA,EAAA,gBAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAI3C,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAE5B,GAAM,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAA,IACN,EAAA,IAAQ,GAAG,GACX,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,MAC3C,GAAI,EAAA,IACN,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC3C,EAAK,WAAW,MAAM,KAE1B,MAAO,CAAC,GAAG,GAAG,OAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAA,EAAA,UAAA,UAAA,KAAA,MAAC,EAAA,EAAA,MAAY,EAAA,GAAA,EAAA,CAAA,UAEb,EAAuC,CAC3C,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,QAAS,KAAK,SAGhB,MAAA,IAAA,GAAW,EAAe,IAG5B,EAAA,UAAA,UAAA,SAAU,EAAW,EAAW,EAAY,GAC1C,GAAM,GAAM,EAAA,OACR,EAAmB,EAAmB,KAAK,QAC1C,GAAW,QACZ,KAAK,aAAe,gBAAkB,OAAS,OAC/C,KAAK,cAET,MAAI,GACK,GAAU,EAAK,EAAG,KAAK,YAGzB,GAGT,EAAA,UAAA,cAAA,SAAc,EAAW,GACvB,GAAM,GAAU,EAEhB,MAAO,GAAA,OACH,EAAmB,EAAmB,EAAS,OAC/C,KAAK,aAAe,gBAAkB,OAAS,SAzO9C,EAAA,UAAY,iBA2OrB,GA7OoC,IA+OpC,EAAA,cAAkB,cAAc,IAKhC,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,GAAI,IAAe,GAEhC,SAAA,EAAA,KAAA,KAAM,GAAA,GAAI,EAAI,CAAE,KAAI,MAAwB,OAIvC,SAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IAZV,EAAA,UAAY,aAcrB,GAhBgC,IAkBhC,EAAA,cAAkB,cAAc,ICjkBhC,AAkDA,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,KAAK,IAAI,KAAK,IAAI,EAAK,KAAM,GAAI,GAE7C,EAAK,WAAa,EAAK,WACvB,EAAK,KAAO,EAAK,KACjB,EAAK,gBAAkB,KAGf,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,KAAK,YAAc,KACrB,MAAO,MAAK,WAId,OAFM,GAAa,EAAM,MACnB,EAAoB,GACjB,EAAI,EAAG,EAAI,KAAK,WAAW,OAAQ,EAAE,EAC5C,EAAW,KACP,KAAK,WAAW,IAAM,KAAO,EAAW,GAAK,KAAK,WAAW,IAEnE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAI,EAAK,MAAQ,EAAK,KAAO,GAC/B,GAAM,GACF,EAAO,UAAe,KAAO,GAAQ,EAAO,SAC1C,EAAa,EAAK,cAAc,GAChC,EAAS,GACX,WAAM,MAAA,IAAU,EAAO,EAAK,KAAM,EAAY,EAAK,OACnD,WAAM,MAAA,IAAO,GACjB,MAAO,GAET,MAAO,MAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,KACX,WAAY,KAAK,WACjB,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAxDf,EAAA,UAAY,UA0DrB,GA5D6B,IA6D7B,EAAA,cAAc,cAAc,IA4D5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAIpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,CAAC,KAAM,MAGjB,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAM,GAAa,EAAM,MACzB,MAAO,CAAC,EAAW,GAAI,EAAG,EAAW,KAThC,EAAA,UAAY,mBAWrB,GAbsC,IActC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAmBzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAhBM,EAAA,WAA2B,KAC3B,EAAA,QAAU,GAGV,EAAA,OAAwB,KACxB,EAAA,KAAsB,KAErB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAQrD,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MACnD,EAAK,UAAY,MAGnB,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAK,gBAAkB,CAAC,EAAW,EAAK,UAG1C,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GAAc,EAAK,YACrC,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAK,SAEtB,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,gBAAkB,GAEvB,EAAK,UAAY,CAAC,CAAC,QAAS,MAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAe,EAAW,EAAW,OAAS,GACpD,AAAI,KAAK,QAAU,MACjB,MAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAc,KAAK,OAAQ,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,kBAIzC,KAAK,UAAY,CAAC,CAAC,QAAS,EAAG,KAAI,GAAA,GAAG,EAAC,IAAK,EAAY,KACxD,KAAK,MAAQ,IAGf,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,SAAY,EAAY,OAAS,GAAK,KAAK,MACpC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAM,GAAQ,GAAoB,GAC5B,EACF,GAA2B,EAAK,WAAW,gBAC3C,EAEJ,MAAI,IAAuB,KACzB,EAAS,GACL,EAAO,EAAK,OAAO,OAAQ,EAC3B,EAAK,KAAO,EAAK,KAAK,OAAS,MAEnC,GAAS,GAAM,EAAO,EAAK,OAAO,QAClC,AAAI,EAAK,MAAQ,MACf,GAAS,GAAU,EAAQ,EAAK,KAAK,SAEvC,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,KAI5B,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArHF,EAAA,UAAY,QAuHrB,GAzH2B,IA0H3B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAO,GAAQ,GACf,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,UAAY,CAAC,CAAC,QAAS,IAC5B,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,OAAkB,GAAA,EAAA,EAAA,EAAW,MAAM,GAAjB,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAO,KACT,KAAM,IAAI,GACN,4DACA,SAAQ,EAAW,MAAM,GAAE,oCAC3B,mFAIR,MAAO,CAAC,EAAW,GAAI,GAAU,EAAY,KAG/C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAK,aAAe,iBAAmB,EAAM,KAAO,GAEtD,OADM,GAAwB,CAAC,GACtB,EAAI,EAAG,EAAI,EAAM,KAAM,EAAE,EAChC,EAAY,KAAK,GAEnB,EAAY,KAAK,GACjB,EAAQ,EAAM,UAAU,GAG1B,MAAO,IAAe,MAI1B,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,GACzC,AAAI,KAAK,YAAc,MACrB,GAAO,WAAgB,KAAK,YAE9B,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA/CF,EAAA,UAAY,UAiDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAK9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,WAAa,GAAc,EAAK,cAGvC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAK,WAAW,MAAM,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,GAAoB,KAAK,aAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArBF,EAAA,UAAY,aAuBrB,GAzBgC,IA0BhC,EAAA,cAAc,cAAc,IAc5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAKhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,EAAI,EAAK,EACd,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,KAAK,EAAG,EAAW,KAG5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAS,GAAoB,GACtB,GAAS,EAAQ,EAAK,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,EAAG,KAAK,GAEJ,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,eA4BrB,GA9BkC,IA+BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,YAAc,EAAK,YAGxB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,EAAE,EAC7C,AAAI,EAAK,UAAU,EAAK,YAAY,KAClC,GAAK,YAAY,GAAK,eAKpB,SAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,GAAM,GAAK,GAAO,MAiBnB,EAAA,UAAA,oBAAR,SAA4B,EAAmB,GAK7C,OAJM,GAAW,6CACX,EAAa,EAAY,QAC3B,EAAQ,EACR,EAAU,KACL,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAM,EAAW,GACvB,GAAI,KAAK,UAAU,GACjB,GAAI,IAAY,KACd,EAAU,MAEV,MAAM,IAAI,GAAW,gDAGvB,IAAS,EAIb,GAAM,GAAe,GAAU,GAC/B,GAAI,IAAY,MACd,GAAI,IAAU,GAAK,EAAe,IAAU,EAC1C,KAAM,IAAI,GAAW,GAEvB,EAAW,GAAW,EAAe,UAC5B,IAAiB,EAC1B,KAAM,IAAI,GAAW,GAGvB,MAAO,IAGT,EAAA,UAAA,mBAAA,SAAmB,GAEjB,OADI,GAAiB,GACZ,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,GAAI,KAAK,UAAU,EAAW,KAC5B,EAAiB,GACjB,MAIJ,MAAI,GACK,EAAW,MAAM,EAAG,GAAG,OAAO,KAAK,aAEnC,EAAW,MAAM,EAAG,GAAG,OAC1B,KAAK,oBAAoB,EAAW,MAAM,GAAI,KAAK,eAI3D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAc,EAAW,MAAM,EAAG,GAAG,OACvC,EAAK,oBAAoB,EAAW,MAAM,GAAI,EAAK,cACvD,MAAO,GAAM,QAAQ,MAIzB,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlGF,EAAA,UAAY,UAoGrB,GAtG6B,IAuG7B,EAAA,cAAc,cAAc,IAY5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAM3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,OACN,mFAGN,GAAI,CAAC,MAAM,QAAQ,EAAK,MACtB,KAAM,IAAI,OACN,oEACG,GAAK,KAAI,cAIlB,GAAM,GAAwB,GAAM,EAAG,EAAK,KAAK,OAAS,GAC1D,GAAI,CAAC,EAAA,KAAK,YAAY,EAAK,KAAK,QAAQ,OAAQ,GAC9C,KAAM,IAAI,OACN,+BAAiC,KAAK,UAAU,EAAK,MACrD,8DAGN,SAAK,KAAO,EAAK,KACjB,EAAK,mBAAqB,CAAC,GAAG,OAAO,EAAK,MAC1C,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,EAAK,KAAK,OAAS,OAG5D,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,YAAK,KAAK,QAAQ,SAAC,EAAa,GAC9B,EAAY,EAAI,GAAM,EAAqB,KAEtC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,UAAU,GAAoB,GAAS,KAAK,qBAGrD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjDF,EAAA,UAAY,UAmDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,KACV,EAAK,UAAY,EAAK,WAAa,KAAO,EAAI,EAAK,UAEnD,EAAK,UAAY,IAIrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,UAAW,KAAK,WAChC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACb,MAAO,GAAA,IAAI,EAAA,SAAS,EAAO,KAAK,WAAY,IAG9C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACP,EAAW,GACX,EAAc,EAAA,IAAI,EAAA,SAAS,EAAO,EAAK,WAAY,EAAM,GACzD,EAAS,EAAM,IAAI,EAAY,OAAO,EAAM,QAClD,MAAO,MAtCJ,EAAA,UAAY,UAyCrB,GA3C6B,IA4C7B,EAAA,cAAc,cAAc,IClqB5B,AA2EA,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAgB7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GATM,EAAA,WAA4B,KAE3B,EAAA,+BACL,gBAME,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MAKrD,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,AAAI,EAAK,aAAe,KAGtB,EAAK,gBAAkB,CAAC,EAAW,MAInC,EAAK,gBACD,CAAC,GAAW,OAAO,GAAqB,EAAK,cAGrD,SAAK,SAAW,EAAK,SACrB,GAAoC,EAAK,SAAU,YACnD,EAAK,UAAY,EAAK,UACtB,GAAoC,EAAK,UAAW,aACpD,EAAK,sBAAwB,GACzB,EAAK,uBAAyB,EAAK,gCACvC,EAAK,sBAAwB,GAAe,EAAK,uBACjD,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,qBAAuB,GAAc,EAAK,sBAC/C,EAAK,SAAW,EAAK,SACrB,EAAK,gBAAkB,EAAK,SAC5B,EAAK,YAAc,EAAK,cAGnB,SAAA,UAAA,MAAP,SAAa,GACX,KAAK,WAAa,KAAK,UACnB,aAAc,CAAC,KAAK,SAAU,KAAK,WAAY,KAAK,MACpD,KAAK,sBAAuB,KAAK,sBAAuB,GACxD,KAAK,sBACT,KAAK,MAAQ,IAKL,EAAA,UAAA,6BAAV,SAAuC,KAEvC,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAK,GAAK,SAGR,GAAS,GAAoB,GACtB,EAAA,SAAS,EAAQ,EAAA,UAAU,KAH3B,QAQb,EAAA,UAAA,mBAAA,SAAmB,GAEjB,GADA,EAAa,GAAmB,GAC5B,KAAK,aAAe,KACtB,MAAW,GAAU,OAAA,CAAE,KAAK,YAG9B,GAAM,GAAmB,GAAqB,KAAK,aACnD,GAAI,EAAO,SAAW,EAAW,OAAS,EACxC,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IAG7B,OADI,GAAI,EACC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAK,EAAO,GACZ,EAAK,EAAW,EAAI,GAC1B,GAAK,GAAM,MAAU,GAAM,MAAU,IAAO,EAC1C,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IACxB,AAAI,GAAM,MACf,GAAO,GAAK,GAEd,IAGJ,MAAA,CAAQ,EAAW,IAAE,OAAK,EAAM,CAAE,KAAK,aAGzC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,AAAI,EAAM,QAAU,SAClB,GAAQ,GAAO,EAAO,UAExB,GAAM,GAAS,GAAS,EAAK,WAAW,OAAQ,EAAM,QACtD,MAAO,GAAO,QACV,GAAmB,EAAK,mBAAmB,EAAM,YAIzD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,UAAW,KAAK,UAChB,sBAAuB,GAAqB,KAAK,uBACjD,sBAAuB,GAAqB,KAAK,uBACjD,oBAAqB,GAAqB,KAAK,qBAC/C,qBAAsB,GAAoB,KAAK,sBAC/C,SAAU,KAAK,SACf,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjIF,EAAA,UAAY,YAmIrB,GArI+B,IAsI/B,EAAA,cAAc,cAAc,ICjN5B,AA+BA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAGlC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAQ,KAAG,KACjB,SAAK,gBAAkB,KAOf,SAAA,UAAA,cAAV,SAAwB,GACtB,KAAM,IAAI,KAaJ,EAAA,UAAA,gCAAR,SAAwC,EAAe,GACrD,GAAI,GAAU,MAAQ,GAAU,KAC9B,MAAO,MACF,GAAI,EAAO,OAAS,EAAO,OAChC,MAAO,MAAK,gCAAgC,EAAQ,GAC/C,GAAI,EAAO,SAAW,EAC3B,MAAO,GAGT,OADM,GAAqB,EAAO,MAAM,EAAG,EAAO,OAAS,EAAO,QACzD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,EAAO,OAAS,EAAO,OAAS,GAC3C,EAAI,EAAO,GACjB,GAAI,GAAK,MAAQ,GAAK,MAAQ,EAAI,GAAK,EAAI,EACzC,EAAY,KAAK,cACR,IAAM,EACf,EAAY,KAAK,WACR,IAAM,EACf,EAAY,KAAK,QAEjB,GAAI,IAAM,EACR,KAAM,IAAI,GACN,wDACA,KAAK,UAAU,GAAU,IAAM,KAAK,UAAU,IAEpD,EAAY,KAAK,IAGrB,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GAOJ,GALA,AAAI,MAAM,QAAQ,IAAe,CAAC,MAAM,QAAQ,EAAW,KAEzD,GAAa,CAAC,GAAmB,KAEnC,EAAa,EACT,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,mEACA,SAAQ,EAAW,OAAM,eAM/B,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAI1B,GADA,EAAa,GAAqB,GAC9B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,qDACA,6BAA4B,KAAK,UAAU,GAAW,MAK5D,OAFI,GACA,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GAC9C,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,GAAM,GAAW,EAAW,IAAI,SAAA,GAAS,MAAA,GAAM,SAC/C,AAAI,EAAW,QAAQ,QAAU,IAC7B,GAAqB,GAAU,SAAW,EAC5C,KAAK,gBAAkB,GAEvB,KAAK,gBAAkB,IAI3B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAK,iBACP,GAAM,GAA2B,GAC3B,EAAY,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,OAC5C,GAAI,EAAU,QAAQ,QAAU,IAI9B,OADM,GAAU,GAAc,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAEZ,OAFO,GAAC,EAAA,GACF,EAAQ,EAAE,KACP,EAAI,EAAG,EAAI,EAAU,EAAO,EAAE,EACrC,EAAI,GAAa,EAAG,GAEtB,EAAe,KAAK,GAEtB,MAAO,GAAK,cAAc,QAK1B,OADI,GAAa,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,KAChB,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAY,EAAO,GACnB,EAAW,EAAO,MAAM,GAAG,OAAO,CAAC,IACrC,EAAc,EAAE,QAChB,CAAC,GAAW,OAAO,GAAoB,EAAO,MAAM,MACxD,EAAc,EAAA,UAAc,EAAa,CAAC,EAAG,IAC7C,EAAc,EAAY,QAAQ,GAClC,EAAe,KAAK,GACpB,EAAa,WACJ,EAAQ,GACjB,GAAM,GAAO,GAAgB,EAAG,GAAO,OAAO,CAAC,IAC/C,EAAe,KAAK,EAAA,UAAc,EAAG,IACrC,EAAa,OAGb,GAAe,KAAK,GAGxB,GAAI,GAAI,EAAK,cAAc,GACrB,EAAQ,EAAE,KAChB,GAAI,GAGF,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAQ,EAAO,OACf,EAAY,EAAO,EAAQ,GAC3B,EACF,CAAC,GAAW,OAAO,EAAO,MAAM,EAAG,EAAO,OAAS,IACvD,EAAI,EAAA,UAAc,EAAE,QAAQ,CAAC,GAAI,IAAa,CAAC,EAAG,IACzC,QAAQ,WACR,EAAQ,GACjB,GAAM,GAAO,CAAC,EAAQ,GAAG,OAAO,GAAgB,EAAG,EAAQ,IAC3D,EAAI,EAAA,UAAc,EAAG,IAGzB,MAAO,QAGT,OAAO,GAAK,cAAc,MAKhC,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,EACb,GAAI,GACJ,AAAI,EAAW,IAAM,KACnB,EAAc,KAEd,EAAc,EAAW,GAAG,MAAM,GAEpC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAG1B,SAAa,GAAqB,GAClC,AAAI,EAAW,SAAW,EACxB,EAAc,EAAW,OAAO,GAEhC,EAAc,CAAC,MAAM,OAAO,GAEvB,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6BAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+BAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,kGAEA,KAAI,EAAO,OAAM,OAAO,EAAK,OAAM,MAEzC,GAAI,EAAK,MAAM,SAAA,GAAK,MAAA,IAAK,OACvB,MAAO,MAET,EAAO,EAAK,IAAI,SAAA,GAAK,MAAA,IAAK,KAAO,EAAI,EAAA,WAAe,EAAG,KAEvD,OADI,GAAS,EAAK,GACT,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAS,EAAA,WAAe,EAAQ,EAAK,IAEvC,MAAO,MAGb,GAlOoC,IAoOpC,GAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,MAcrB,GAhByB,IAiBzB,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAG5B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,WAcrB,GAhB8B,IAiB9B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,GAAA,IAAQ,EAAI,EAAO,OAAQ,MAX/B,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA2D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IAiE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAM/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAJJ,SAAA,aAAe,GAKtB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,MACxD,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,yEAGN,EAAa,EAGb,OADI,GAAe,GACC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,GAAS,MACX,EAAe,GACf,OAGJ,GAAI,EACF,OAIF,OADM,GAAoB,GACjB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAyB,EAAW,GAAG,QAC7C,EAAuB,OAAO,KAAK,KAAM,GAEzC,OADI,GAAS,GACO,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAA,KAAK,YAAY,EAAO,IAC1B,EAAS,GACT,OAGJ,AAAK,GACH,EAAS,KAAK,GAGlB,GAAI,EAAS,OAAS,EACpB,KAAM,IAAI,GACN,4GAEA,KAAK,UAAU,KAIb,EAAA,UAAA,cAAV,SAAwB,GAAxB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAO,IAAc,EAAQ,EAAK,SAItC,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,KAC1D,KAAM,IAAI,GACN,+DAON,OALM,GAAc,EACd,EAAc,EAAY,GAAG,QAC7B,EAAO,KAAK,KAAO,EAAI,EAAY,OAAS,KAAK,KAAO,KAAK,KAG/C,EAAA,EAAA,EAAA,EAAY,MAAM,GAAlB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAY,IAAS,MAAQ,EAAM,IAAS,MAC9C,EAAY,GAAQ,KACpB,MAEF,EAAY,IAAS,EAAM,GAE7B,MAAO,IAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6CAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+CAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,mCAAmC,EAAK,OAAM,KAC9C,8BAA6B,EAAO,OAAM,MAEhD,MAAO,GAAA,KAAS,WACd,GAAI,GAAe,GAOnB,GANA,EAAK,QAAQ,SAAA,GACX,GAAI,GAAK,MACP,EAAe,GACf,UAGA,EACF,MAAO,MAGT,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAK,IAAM,KAEb,EAAY,KAAK,EAAA,SAAa,EAAO,IAAI,OAAO,SAC3C,AAAI,EAAK,GAAG,KAAO,EAAO,GAAG,KAElC,EAAY,KAAK,EAAA,WAAe,EAAK,GAAI,KAEzC,EAAY,KAAK,EAAK,IAG1B,GAAM,GAAoB,EAAA,OAAW,EAAa,EAAK,MACvD,MAAO,GAAA,IAAQ,EAAmB,GAAI,OAI1C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,MAET,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArIF,EAAA,UAAY,cAuIrB,GAzIiC,IA0IjC,EAAA,cAAc,cAAc,IAuF5B,YAAuB,EAAc,GACnC,KAAO,EAAO,GACZ,GAAQ,EAEV,MAAO,GAGT,YAAkB,EAAW,EAAW,GACtC,GAAI,EAAE,MAAM,OAAS,GAAK,EAAE,MAAM,OAAS,EACzC,KAAM,IAAI,IACN,oEAeN,GAbA,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAC3B,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAE3B,AAAI,MAAO,IAAS,UAClB,GAAO,CAAC,EAAM,IAGZ,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,KAAM,IAAI,IACN,+DAGN,GAAM,GAAQ,EAAE,MAAM,OAChB,EAAQ,EAAE,MAAM,OACtB,AAAI,GAAQ,MAEV,GAAO,CAAC,EAAQ,EAAG,EAAQ,IAE7B,GAAM,GAAY,EAElB,MAAO,GAAA,KAAS,WACd,GAAI,GACJ,GAAI,EAAQ,GACV,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,YACpB,EAAQ,GACjB,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,QAE7B,GAAO,EAGT,GAAI,GACJ,GAAI,EAAE,MAAM,SAAW,GAAK,EAAE,MAAM,SAAW,EAC7C,AAAI,EAAU,KAAO,EAAU,GAC7B,EAAM,EAAE,IAAI,GAAG,IAAI,EAAU,IAE7B,EAAM,EAAE,UAAU,CAAC,EAAG,IAAI,IAAI,GAAG,IAAI,EAAU,SAGjD,GAAM,GAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EACzC,EAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EAC/C,EAAM,EAAE,OAAO,EAAG,EAAM,GAG1B,GAAI,EAAO,GACT,GAAI,GAAG,OACP,AAAI,EAAQ,EACV,EAAM,EAAQ,EAAQ,EAEtB,EAAM,EAAQ,EAGhB,OADM,GAAwB,GACrB,EAAI,EAAK,EAAI,EAAM,EAAM,EAAE,EAClC,EAAY,KAAK,GAEnB,EAAM,EAAI,QAAQ,GAEpB,MAAI,GAAI,MAAM,SAAW,GACvB,GAAM,EAAI,WAAW,IAEhB,IAIX,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,EAAK,KACjB,EAAK,UAAY,EAAK,WAAa,KAAO,GAAQ,EAAK,UACvD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAS,EAAW,GACpB,EAAS,EAAW,GAC1B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,GAAI,EAAO,EAAK,MAAQ,EAAO,EAAK,IAClC,KAAM,IAAI,GACN,8BACG,GAAO,EAAK,IAAG,QAAQ,EAAO,EAAK,OAIpC,EAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,qDACA,iBAAgB,EAAO,OAAM,eAGnC,GAAI,GAAK,EAAO,GACZ,EAAK,EAAO,GACZ,EACJ,MAAK,OAAM,QAAQ,KAAK,MAMtB,EAAO,KAAK,KAAK,IACN,SAAC,EAAM,GAAM,MAAA,IACT,EAAM,EAAO,GAAG,MAAM,UAPrC,EAAO,CACL,GAAc,KAAK,KAAM,EAAG,MAAM,QAClC,GAAc,KAAK,KAAM,EAAG,MAAM,SAOtC,AAAI,KAAK,WACP,GAAK,GAAY,EAAI,EAAK,IAC1B,EAAK,GAAY,EAAI,EAAK,KAErB,GAAS,EAAI,EAAI,IAGlB,EAAA,UAAA,cAAR,SAAsB,EAAe,GACnC,GAAI,GACJ,MAAK,OAAM,QAAQ,KAAK,MAQtB,EAAO,KAAK,KANZ,EAAO,CACL,GAAc,KAAK,KAAM,EAAO,QAChC,GAAc,KAAK,KAAM,EAAO,SAM7B,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAU,EAAW,GAAa,QAClC,EAAU,EAAW,GAAa,QACxC,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAG,GACjB,GAAM,GAAc,EAAO,OAAO,GAClC,MAAI,GAAY,SAAW,GACzB,EAAY,KAAK,GAEZ,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,OAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,KACb,UAAa,KAAK,WAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA7GF,EAAA,UAAY,MA+GrB,GAjHyB,IAkHzB,EAAA,cAAc,cAAc,IChjC5B,AA2BA,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAKjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,OAAS,EAAK,SAGrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,OAAQ,KAAK,QAC7B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAS,WACX,MAAA,IAAe,EAAM,MAAO,EAAG,EAAK,QAAQ,IAAI,IAC9C,EACF,GAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAC9D,MAAO,MA5BJ,EAAA,UAAY,gBA+BrB,GAjCmC,IAkCnC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAS,WACb,GAAM,GAAS,KAAK,KAAK,EAAK,KAAQ,GAAI,EAAK,OAC/C,MAAO,GAAM,IAAI,GAAe,EAAM,MAAO,EAAG,KAElD,MAAO,IAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAEnE,MAAO,MA/BJ,EAAA,UAAY,kBAkCrB,GApCqC,IAqCrC,EAAA,cAAc,cAAc,IAyC5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAMhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,KACjB,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,YAAc,GAAoB,GAAQ,OAGxD,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAa,EAAK,eAAe,GAEjC,EAAgB,WACpB,GAAM,GAAQ,GAAoB,GAE5B,EAAQ,mBACR,EAAQ,mBAER,EAAS,CAAC,EAAQ,EAEpB,EAAU,EAAA,aAAa,EAAA,cAAc,GAAa,EAAK,MAE3D,EAAU,GAAO,EAAS,WAG1B,GAAM,GAAI,KAAA,IAAE,GAAI,EAAK,MAAS,GAAI,EAAK,KAAO,KAAA,IAAA,EAAU,IAAO,KACzD,EAAI,CAAC,EAAI,EAAS,EAAK,KAGvB,EAAI,EAAM,IAAI,GAAS,IAAI,EAAQ,IAAI,IAAI,IAAI,IAErD,MAAO,GAAE,IAAI,GAAG,IAAI,IAEtB,MAAO,IACH,EAAe,WAAM,MAAA,IAAoB,IACzC,EAAO,UAAe,IAE5B,MAAO,MAxDJ,EAAA,UAAY,eA2DrB,GA7DkC,IA8DlC,EAAA,cAAc,cAAc,IChN5B,YA4CI,EAAW,EAAc,EAAkB,EAAe,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GACJ,GAAI,EAAE,OAAS,EACb,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EAEpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EACpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,OAElC,MAAM,IAAI,IACN,2DAA2D,EAAE,KAAI,QAGvE,MAAO,GAoBT,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WACH,GAAM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EACF,GAAmB,EAAG,EAAM,EAAU,EAAM,EAAO,GACvD,MAAO,CAAC,EAAQ,EAAM,KAqBjC,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKH,OAJM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EAAwB,GACX,EAAA,EAAA,EAAA,GAAiB,EAAG,EAAE,MAAtB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAc,QAAQ,KAAU,GAClC,EAAY,KAAK,GAEjB,EAAY,KAAK,EAAE,MAAM,IAG7B,GAAM,GAAgB,EAAK,QAAQ,GAC7B,EAAoB,EAAS,QAAQ,GACrC,EACF,GAAS,KAAO,KAAO,EAAM,QAAQ,GACnC,EACF,GAAQ,KAAO,KAAO,EAAK,QAAQ,GACjC,EAAS,GACX,EAAG,EAAe,EAAmB,EACrC,EAAgB,GACpB,MAAO,CAAC,EAAQ,EAAM,iBAgB7B,EAAW,EAAe,EAAc,EACxC,GACF,MADE,AAAA,KAAA,QAAA,GAAA,MACE,EAAA,KAAK,YACD,EAAc,QAAQ,OAAQ,GAAiB,EAAG,EAAE,KAAO,IAC1D,GACH,EAAG,EAAO,EAAM,EAAe,GAE5B,GACH,EAAG,EAAO,EAAM,EAAe,GAsFvC,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAqBtC,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,IAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KAC1C,EAAK,SAAW,EAAK,UAAY,KAAO,IAAO,EAAK,SACpD,EAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,sBACD,GAAe,EAAK,uBAAyB,SACjD,EAAK,0BACD,GAAe,EAAK,2BAA6B,QACrD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAc,EAAK,iBAC1C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,oBAGvC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAO,KAAK,MAAQ,EAAI,KAAK,KAAQ,KAAK,KAAO,EAAW,OAC5D,EAAM,EAAW,GACvB,GAAI,GAAO,KACT,KAAM,IAAI,GACN,QAAQ,EAAI,+FAET,MAAK,UAAU,GAAW,MAEnC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAW,OAAQ,KAAI,GAAA,GAAG,EAAC,GAAO,EAAG,MAC/D,GAAM,GAAQ,CAAC,GACf,AAAI,KAAK,OACP,MAAK,MAAQ,KAAK,UACd,QAAS,EAAO,KAAM,KAAK,iBAAkB,KAAK,iBAClD,GAAM,KAAK,kBAEjB,AAAI,KAAK,QACP,MAAK,KAAO,KAAK,UACb,OAAQ,EAAO,KAAM,KAAK,gBAAiB,KAAK,gBAAiB,GACjE,KAAK,iBAEX,KAAK,WAAa,KAAK,UACnB,cAAe,EAAO,KAAM,KAAK,sBAAuB,KAAM,IAClE,KAAK,eAAiB,KAAK,UACvB,kBAAmB,EAAO,KAAM,KAAK,0BAA2B,KAChE,IACJ,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACvD,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAO,EAAW,OAClB,EAAgB,GAAiB,EAAG,GACpC,EAAO,EAAK,MAAQ,EAAI,EAAK,KAAQ,EAAK,KAAO,EACvD,EAAc,OAAO,EAAM,GAC3B,GAAM,GAAiB,GAA2B,EAAG,GACrD,EAAe,GAAQ,EAAW,GAElC,GAAM,GAAsB,EAAc,QAC1C,EAAoB,OACpB,GAAM,GAAoB,CAAC,EAAA,KAAK,YAC5B,EAAqB,GAAiB,EAAG,GAAM,MAAM,EAAG,EAAO,IAE7D,EAAmC,WACvC,GAAI,GACF,GAAM,GACF,EAAK,WAAW,OAAO,QAAQ,GAC7B,EACF,EAAK,eAAe,OAAO,QAAQ,GACjC,EACF,EAAK,OAAS,EAAK,KAAK,OAAO,QAAQ,GAAkB,KACvD,EACF,EAAK,MAAQ,EAAK,MAAM,OAAO,QAAQ,GAAkB,KAC7D,MAAO,IACH,EAAO,EAAqB,EAC5B,EAAe,EAAgB,EAAK,aAExC,OAAO,IACH,EAAO,EAAK,WAAW,OAAQ,EAAK,eAAe,OACnD,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACrC,EAAK,OAAS,KAAO,KAAO,EAAK,MAAM,OAAQ,EAAK,UAI5D,GAAI,CAAC,EACH,MAAO,KAGH,GAAA,GAAA,GAAA,EAAA,EAAA,MAAA,OAAA,EAAA,KAAA,OAAA,EAAA,EAAA,SAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAM,EAAA,EAAA,GAIvB,EACF,SAAC,EAAyB,EAAe,GACvC,EAAA,KAAS,WACP,GAAM,GAAQ,EAAI,EACZ,EAAY,EAAS,OACrB,EAAc,EAAU,IAAI,GAAO,IAAI,GAC7C,EAAS,MAAM,EAAU,IAAI,OAU/B,EAA8B,WAClC,EAAgB,EAAK,WAAY,EAAM,EAAK,UAC5C,EAAgB,EAAK,eAAgB,EAAU,EAAK,WAEtD,WAEO,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,SAAU,KAAK,SACf,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,sBAAuB,GAAqB,KAAK,uBACjD,0BACI,GAAqB,KAAK,2BAC9B,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,eAAgB,GAAoB,KAAK,gBACzC,gBAAiB,GAAoB,KAAK,kBAEtC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArKF,EAAA,UAAY,qBAuKrB,GAzKwC,IA0KxC,EAAA,cAAc,cAAc,IAkD5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAgBtC,WAAY,GAAZ,GAAA,GAAA,KAOE,GANA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KACtC,MAAO,GAAK,MAAS,UACvB,GAAI,CAAC,OAAO,UAAU,EAAK,MACzB,KAAM,IAAI,OACN,gDAAgD,EAAK,cAElD,MAAM,QAAQ,EAAK,MAC5B,OAAmB,GAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,6CACA,iBAAgB,KAAK,UAAU,EAAK,YAI5C,MAAM,IAAI,OACN,2DACA,iBAAgB,KAAK,UAAU,EAAK,QAG1C,SAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,kBAE5C,EAAK,gBAAkB,KAGlB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAQ,EAAW,OAGzB,AAAI,MAAO,MAAK,MAAS,UACvB,MAAK,KAAO,CAAC,KAAK,OAEpB,OAAS,GAAI,EAAG,EAAI,KAAK,KAAK,OAAQ,EAAE,EACtC,AAAI,KAAK,KAAK,GAAK,GACjB,MAAK,KAAK,IAAM,GAKpB,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,GAAK,GAAQ,EACtB,KAAM,IAAI,OAAM,iBAAiB,GAGrC,GAAI,KAAK,KAAK,SAAW,GAAqB,KAAK,MAAM,OACvD,KAAM,IAAI,OAAM,4BAA4B,KAAK,MAGnD,GAAM,GAAa,KAAK,KAAK,IAAI,SAAA,GAAQ,MAAA,GAAW,KAE9C,EAAY,GAClB,AAAI,KAAK,MACP,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAE3B,KAAK,MAAQ,KAEf,AAAI,KAAK,OACP,KAAK,KAAO,KAAK,UACb,OAAQ,EAAY,UAAW,KAAK,gBACpC,KAAK,gBAAiB,GAE1B,KAAK,KAAO,KAGd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACQ,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAQ,EAAW,OAEzB,MAAO,GAAA,KAAK,WAIV,OAHM,GAAW,GACb,EAAA,EAAA,QAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,SACL,EAAiB,GAA2B,EAAG,GACnC,EAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAe,GAAO,EAAW,GAuBnC,OApBM,GAAY,SAAC,GACjB,MAAI,IAAK,MAAQ,EAAE,MAAM,SAAW,GAChC,EAAK,OAAS,CAAC,EAAQ,GAClB,EAAE,QAAQ,GAEV,GAIP,EAAQ,EAAU,EAAK,MAAM,QAC7B,EAAS,EAAU,EAAK,KAAK,QAQ3B,EAA0B,GAC1B,EAA8B,GAC3B,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,AAAK,EAAK,KAAkB,QAAQ,KAAO,GACzC,GAAc,KAAK,EAAW,IAC9B,EAAkB,KAAK,IAEvB,GAAc,KAAK,GACnB,EAAkB,KAAK,EAAW,KAGtC,SAAO,EAAK,KAAK,GACjB,EAAW,EAAS,KAAK,GACzB,EAAQ,EAAM,KAAK,GACnB,EAAS,EAAO,KAAK,GAEd,GACH,EAAO,EAAM,EAAU,EAAQ,EAAO,EAAK,YAInD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,mBAExC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjKF,EAAA,UAAY,qBAmKrB,GArKwC,IAsKxC,EAAA,cAAc,cAAc,IC9nB5B,YAoEI,EAAW,EACX,GACF,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,kEACG,GAAE,KAAI,eAMf,GAHA,AAAI,GAAW,MACb,GAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,KAErB,EAAQ,SAAW,GAAK,EAAQ,GAAG,SAAW,GAC9C,EAAQ,GAAG,SAAW,EACxB,KAAM,IAAI,GACN,+GAON,GAHA,AAAI,GAAc,MAChB,GAAa,MAEX,IAAe,gBAAkB,IAAe,gBAClD,KAAM,IAAI,GACN,wBAAwB,EAAU,mEAIxC,GAAI,GACJ,MAAI,KAAe,gBACjB,EAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,IAE/C,EAAU,CAAC,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,GAAI,CAAC,EAAG,IAG1C,EAAA,IAAQ,EAAG,KA6BtB,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAMjC,WAAY,GAAZ,GAAA,GAAA,KAUE,GATA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,WACD,EAAK,YAAc,KAAO,KAAoB,EAAK,WAGnD,EAAK,SAAW,KAClB,EAAK,QAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,YACnB,MAAO,GAAK,SAAY,SACjC,EAAK,QACD,CAAC,CAAC,EAAK,QAAS,EAAK,SAAU,CAAC,EAAK,QAAS,EAAK,eAGvD,GADA,EAAK,QAAU,EAAK,QAChB,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,6DACA,sBAAqB,EAAK,QAAQ,OAAM,YAG9C,GAAI,GAAa,OACb,EAAY,OAChB,GAAI,MAAO,GAAK,QAAQ,IAAO,SAC7B,EAAgB,CAAC,EAAK,QAAQ,GAAI,EAAK,QAAQ,IAC/C,EAAe,CAAC,EAAK,QAAQ,GAAc,EAAK,QAAQ,SAIxD,GAFA,EAAK,QAAU,EAAK,QAEhB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,gEACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAIrD,GAFA,EAAgB,EAAK,QAAQ,GAEzB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,+DACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAErD,EAAe,EAAK,QAAQ,GAE9B,EAAK,QAAU,CAAC,EAAe,GAEjC,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAEhC,GAAI,GACA,EACJ,MAAI,MAAK,aAAe,gBACtB,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,IAE5C,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,MAIlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,IACF,GAAoB,GAAS,EAAK,QAAS,EAAK,eAG1D,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAnGF,EAAA,UAAY,gBAqGrB,GAvGmC,IAwGnC,EAAA,cAAc,cAAc,IC5O5B,YA0CI,EAAW,EAA4B,EACvC,EAAuB,EACvB,GACF,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,IAEhB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAKb,EAAI,GAAsB,EAAG,GAC7B,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MAEf,EAAI,EAAA,QAAY,EAAe,EAAU,EAAS,GAIlD,EAAI,EAAA,QAEA,EAA0B,EAAU,EAAS,GAEnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBAeP,EAAa,EACb,EAAoC,EACpC,EAAyB,GAC3B,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,EAAG,IAEnB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAIb,EAAI,GAAsB,EAAa,GACvC,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MACf,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAExC,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAE1C,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAsBX,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAWtC,WAAY,GAAZ,GAAA,GAAA,KAKE,GAJA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,GAElB,EAAA,EAAA,KAAA,KAAM,IAAK,KACP,MAAO,GAAK,UAAa,SAC3B,EAAK,SAAW,CAAC,EAAK,kBAEpB,MAAM,QAAQ,EAAK,WAClB,EAAK,SAAsB,SAAW,GACvC,MAAQ,GAAK,SAAsB,IAAO,SAC5C,EAAK,SAAW,EAAK,aAErB,MAAM,IAAI,GACN,qGAEA,IAAG,KAAK,UAAU,EAAK,YAG7B,GADA,GAAsB,EAAK,SAAU,YACjC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBAEhB,MAAO,GAAK,SAAY,SAC1B,EAAK,QAAU,CAAC,EAAK,iBAEnB,MAAM,QAAQ,EAAK,UAClB,EAAK,QAAqB,SAAW,GACtC,MAAQ,GAAK,QAAqB,IAAO,SAC3C,EAAK,QAAU,EAAK,YAEpB,MAAM,IAAI,GACN,oGAEA,IAAG,KAAK,UAAU,EAAK,WAG/B,UAAsB,EAAK,QAAS,WAEpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAS,GACX,EAAW,GAAI,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IAChE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAW,KAO5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,EAAS,GAAa,GAAoB,GAAS,GACnD,GAAM,GAAS,EAAK,gBAChB,GAAoB,GAAS,CAAC,EAAK,SAAS,GAAI,GAChD,CAAC,EAAK,QAAQ,GAAI,GAAI,EAAK,QAAS,gBAExC,MAAO,GAAA,QAAY,EAAQ,CAAC,OAIhC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,SAEV,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAxFwC,IA0FxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,IAEtB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,UACrB,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,SAErC,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAKrE,MAJA,GACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,GAErC,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,KAQlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAhFwC,IAkFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,EAAG,IAEzB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,SAAU,EAAK,UACpC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,QAAS,EAAK,SAEnD,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAOrE,MANA,GAAS,GACL,EAAQ,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,EAAM,GAE7C,CAAC,EAAW,GAAI,EAAQ,EAAM,EAAM,EAAW,KAS1D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GArFwC,IAuFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,eAcrB,GAhBkC,IAiBlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,mBAcrB,GAhBsC,IAiBtC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAC5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,EAAW,KAGpC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAEd,GAb8C,IAe9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAG1C,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,KAAS,EAAO,MARpB,EAAA,UAAY,yBAWrB,GAb4C,IAc5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAGtC,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,IAAQ,EAAO,MARnB,EAAA,UAAY,qBAWrB,GAbwC,IAcxC,EAAA,cAAc,cAAc,IAiB5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAE5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GAEjB,MADA,GAAa,EACT,KAAK,aAAe,eACf,CAAC,EAAW,GAAI,EAAW,IAE3B,CAAC,EAAW,GAAI,EAAW,KAItC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAGZ,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,KAAK,YAC3B,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA7B8C,IA+B9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAA5C,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,KAAS,EAAO,CAAC,EAAG,IAEpB,EAAA,KAAS,EAAO,CAAC,EAAG,OAR1B,EAAA,UAAY,yBAYrB,GAd4C,IAe5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAAxC,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,IAAQ,EAAO,CAAC,EAAG,IAEnB,EAAA,IAAQ,EAAO,CAAC,EAAG,OARzB,EAAA,UAAY,qBAYrB,GAdwC,IAexC,EAAA,cAAc,cAAc,IC5rB5B,AA4CA,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,GAAZ,GAAA,GAQE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAKf,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAIE,MAAI,MAAK,OAAS,KACT,KAAK,MAAM,UAEX,QAIX,SAAc,GAIZ,AAAI,KAAK,OAAS,MAChB,MAAK,MAAM,UAAY,oCAI3B,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,MAAM,kDAIpB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAIpB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WAEE,MAAQ,MAAK,MAAc,0CAK7B,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,MAAM,wCAKpB,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,MAAM,cAGpB,EAAA,UAAA,WAAA,SAAW,GACT,KAAK,MAAM,WAAW,IAGxB,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAS,CACP,UAAa,KAAK,MAAM,eACxB,OAAU,KAAK,MAAM,cAGnB,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,OAAS,MAChB,KAAK,MAAM,6BAA6B,IAKrC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAc,EAAO,MACrB,EAAQ,GAAY,EAAa,GACvC,MAAO,GAAO,MACd,GAAM,GAAY,CAAC,MAAK,GACxB,cAAO,OAAO,EAAW,GAClB,GAAI,GAAI,IAEnB,GAvGsC,IAyGtC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,oEACA,gBAAe,KAAK,UAAU,KAEpC,KAAK,UAAY,CAAC,CAAC,MAAO,IAC1B,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAChE,AAAK,KAAK,MAAM,OACd,MAAK,MAAM,MAAM,GACjB,KAAK,MAAM,MAAQ,IAErB,EAAA,UAAM,MAAK,KAAA,KAAC,IAGd,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAC1D,EACF,KAAK,MAAM,mBAAmB,GAC5B,EAAY,EAAW,GAC7B,MAAO,CAAC,EAAiB,GAAI,GAAW,OAAO,EAAiB,MAAM,KAGxE,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,EAAS,GAAoB,GAI7B,GAAM,GAAwB,SAAC,EAAgB,GAK7C,GAAM,GAAS,GAAoB,EAAK,MAAM,KAAK,EAAQ,IAC3D,MAAO,CAAC,EAAQ,KAEZ,EACF,GAAI,EAAM,EAAQ,GAAI,GAAyB,KAC3C,KAAsB,GACtB,IACF,EAAI,EAAW,GAGrB,MAAO,MArDJ,EAAA,UAAY,kBA0DrB,GA5DqC,IA6DrC,EAAA,cAAc,cAAc,gBAEgB,GAC1C,GACI,GAAiC,yBAA0B,GAmBjE,GAAM,IAA2D,SAEjE,GAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAUL,EAAc,EAAK,MAAM,YACzB,EAAqC,GAC3C,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,aAAe,GAAY,GAChC,EAAY,YACR,IAAY,cAAmB,IACnC,GAAM,GAAqC,GAW3C,GAVA,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,cAAgB,GAAY,GACjC,EAAK,aAAa,KAAO,WAAa,EAAK,aAAa,KACxD,EAAK,cAAc,KAAO,YAAc,EAAK,cAAc,KAE3D,EAAK,UAAY,EAAK,YAAc,OAChC,GACA,EAAK,UACT,GAA4B,EAAK,WAC7B,EAAK,QACP,KAAM,IAAI,IACN,mEAEN,SAAK,UAAY,EAAK,MAAM,SAC5B,EAAK,gBAAkB,EAAK,MAAM,gBAClC,EAAK,YAAc,EAAK,MAAM,YAC9B,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAClB,EAAK,UAAY,EAAK,MAAM,UAC5B,EAAK,aAAe,OAGtB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GAIZ,KAAK,WAAa,EAClB,AAAI,KAAK,cAAgB,MACvB,MAAK,aAAa,UAAY,GAEhC,AAAI,KAAK,eAAiB,MACxB,MAAK,cAAc,UAAY,oCAInC,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,aAAa,aAAa,OAClC,KAAK,cAAc,eAGzB,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAa,EAAQ,OACrB,EAAiB,KAAK,MAAM,EAAa,GAC/C,KAAK,aAAa,WAAW,EAAQ,MAAM,EAAG,IAC9C,KAAK,cAAc,WAAW,EAAQ,MAAM,KAG9C,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GACA,KAAK,aAAa,mBAAmB,GACzC,AAAM,MAAM,QAAQ,IAAgB,MAAM,QAAQ,EAAY,KAC5D,GAAc,CAAC,IAEjB,EAAc,EAEd,GAAI,GACA,EACA,EAiBJ,MAhBA,AAAI,MAAK,aACP,GAAa,EAAY,MAAM,IAC/B,EAAc,EAAY,GAI5B,EAAc,EACd,AAAI,KAAK,YAAc,SACrB,GAAY,EAAY,OAAS,IAAM,EACvC,EAAe,CAAC,IACX,AAAI,KAAK,WAAa,KAC3B,EAAe,CAAC,EAAa,EAAY,SAEzC,EAAe,CAAC,GAGd,KAAK,YACH,KAAK,WAAa,KACb,EAAa,OAAO,GAAY,OAAO,EAAW,SAEpD,CAAC,GAAa,OAAO,GAAY,OAAO,EAAW,SAErD,GAA+B,IAGxC,EAAA,UAAA,MAAA,SACI,EACA,GACF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAEX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAU1D,GATA,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAEzB,AAAI,MAAM,QAAQ,IAChB,GAAgB,EAAuC,MAAM,GAC7D,EAAU,EAAuC,IAG9C,IAAgB,MAAQ,EAAa,SAAW,IACjD,GAAa,KACf,MAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,GAE7B,GAAM,GAAiD,GACjD,EAA+B,GACrC,GAAI,GAAgB,MAClB,GAAM,GAAY,EAAa,OAC/B,GAAI,EAAY,EAAI,EAClB,KAAM,IAAI,GACN,iIAIN,EAAO,aAAkB,EACzB,EAAiB,KAAI,MAArB,EAAyB,GACzB,GAAM,GAAc,EACI,IAAI,SAAA,GAAS,MAAA,IAAI,IAAU,CAAC,MAAO,EAAM,UACjE,KAAK,aAAa,UAAY,EAAW,MAAM,EAAG,EAAY,GAC9D,KAAK,cAAc,UAAY,EAAW,MAAM,EAAY,GAC5D,EAAgB,KAAI,MAApB,EAAwB,GAE1B,GAAI,GAAa,KACf,KAAM,IAAI,IACN,yEAKN,OADM,GAAmB,EAAiB,YAAc,IACnC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,YAAkB,MAAmB,EACvC,KAAM,IAAI,GACN,gHAKR,GAAI,GAEF,GAAM,GAAY,CAAC,GAAQ,OAAO,GAC5B,EAAgB,KAAK,UAAU,OAAO,GAUtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GACF,EAAA,UAAM,MAAK,KAAA,KAAC,EAA0C,GAC1D,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAI/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAe,EAAO,aAExB,EACA,EACJ,GAAI,GAAgB,KAClB,EAAI,EAAK,aAAa,KAAK,EAAQ,GACnC,EAAO,EAAK,cAAc,KAAK,EAAQ,QAEvC,GAAM,GAAe,EAAa,MAAM,EAAG,EAAa,OAAS,GAC3D,EAAgB,EAAa,MAAM,EAAa,OAAS,GAC/D,EAAI,EAAK,aAAa,KAClB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KACjD,EAAO,EAAK,cAAc,KACtB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KAGnD,GAAI,GACJ,AAAI,EAAK,aACP,CAAI,MAAM,QAAQ,IAChB,GAAS,EAAE,MAAM,GAAG,OAAQ,EAAkB,MAAM,KAGtD,EAAK,EAAe,GACpB,EAAQ,EAAkB,IAG5B,AAAI,EAAK,iBACP,GAAO,EAAA,QAAY,EAAgB,IAGrC,GAAI,GAcJ,MAbA,AAAI,GAAK,YAAc,SACrB,EAAS,GAAc,CAAC,EAAa,IAChC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,GAAI,EAAA,IAAQ,EAAa,IACrC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,WAAa,MAC3B,GAAS,CAAC,EAAa,IAIrB,EAAK,YACH,EAAK,WAAa,KACZ,EAAoB,OAAO,GAE9B,CAAC,GAAkB,OAAO,GAE5B,KAIX,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,aAAa,cAClB,KAAK,cAAc,eAGrB,EAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,GAAU,KAAK,aAAa,KAAM,WAChC,EAAK,aAAa,MAAM,KAE1B,GAAU,KAAK,cAAc,KAAM,WACjC,EAAK,cAAc,MAAM,KAE3B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,SAAY,EAAyB,GAEnC,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAI,GAcJ,GAbA,AAAI,KAAK,gBACP,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,EAAM,GAEpB,EAAa,EAGf,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,KAAM,MAEpB,EAAa,KAGb,KAAK,aACP,GAAM,GAAS,KAAK,aAAa,OAC3B,EAAsB,EAAO,IAAI,SAAA,GAAS,MAAA,QAChD,MAAI,OAAM,QAAQ,GACT,EAAW,OAAO,GAAW,OAAO,GAEpC,CAAC,GAAY,OAAO,GAAW,OAAO,OAG/C,OAAO,IAIX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,aAAa,iBAAiB,OACtC,KAAK,cAAc,mDAGzB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,aAAa,oBAAoB,OACzC,KAAK,cAAc,sDAKzB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,cAAgB,MACvB,KAAK,aAAa,6BAA6B,GAEjD,AAAI,KAAK,eAAiB,MACxB,KAAK,cAAc,6BAA6B,IAIpD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,UAAa,KAAK,WAGd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAIF,EAAA,WAAP,SACI,EACA,GACF,GAAM,GACF,GAAY,EAAO,OAGvB,GAFA,MAAO,GAAO,MAEV,EAAO,cAAmB,KAC5B,KAAM,IAAI,IACN,4FAIN,GAAM,GAAkC,EACxC,SAAU,MAAW,EACd,GAAI,GAAI,IA7VV,EAAA,UAAY,gBA+VrB,GAjWmC,IAkWnC,EAAA,cAAc,cAAc,IC7kB5B,YAkE2B,GACzB,MAAO,IAAI,IAAW,eA6BJ,GAClB,MAAO,IAAI,IAAI,eAoBI,GACnB,MAAO,IAAI,IAAK,eAuBQ,GACxB,MAAO,IAAI,IAAU,eAwBD,GACpB,MAAO,IAAI,IAAM,eAmBK,GACtB,MAAO,IAAI,IAAQ,eA2BW,GAC9B,MAAO,IAAI,IAAgB,eAyBN,GACrB,MAAO,IAAI,IAAO,eAqBG,GACrB,MAAO,IAAI,IAAO,eAsCY,GAC9B,MAAO,IAAI,IAAgB,eAqBN,GACrB,MAAO,IAAI,IAAO,eAgCY,GAC9B,MAAO,IAAI,IAAgB,eAkCF,GACzB,MAAO,IAAI,IAAW,eA2BK,GAC3B,MAAO,IAAI,IAAa,eAeM,GAC9B,MAAO,IAAI,IAAgB,eAoCF,GACzB,MAAO,IAAI,IAAW,eAoCF,GACpB,MAAO,IAAI,IAAM,eAaK,GACtB,MAAO,IAAI,IAAQ,eAmCY,GAC/B,MAAO,IAAI,IAAiB,eAsBN,GACtB,MAAO,IAAI,IAAQ,eAiBQ,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,eA+BG,GACtB,MAAO,IAAI,IAAQ,eAcK,GACxB,MAAO,IAAI,IAAU,eAyBH,GAClB,MAAO,IAAI,IAAI,eAqBO,GACtB,MAAO,IAAI,IAAQ,eAuBO,GAC1B,MAAO,IAAI,IAAY,eAqBD,GACtB,MAAO,IAAI,IAAQ,eAqBG,GACtB,MAAO,IAAI,IAAQ,eAsBI,GACvB,MAAO,IAAI,IAAS,eAyBF,GAClB,MAAO,IAAI,IAAI,eA0BkB,GACjC,MAAO,IAAI,IAAmB,eAuBG,GACjC,MAAO,IAAI,IAAmB,eA2BF,GAC5B,MAAO,IAAI,IAAc,eAgBM,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eA0BO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAwBO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAYa,GACrC,MAAO,IAAI,IAAuB,eAiBG,GACrC,MAAO,IAAI,IAAuB,eAYD,GACjC,MAAO,IAAI,IAAmB,eAiBG,GACjC,MAAO,IAAI,IAAmB,eAYH,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eA4BN,GAClB,MAAO,IAAI,IAAI,eAgDO,GACtB,MAAO,IAAI,IAAQ,eA0BA,GACnB,MAAO,IAAI,IAAK,eAgDO,GACvB,MAAO,IAAI,IAAS,eA2BI,GACxB,MAAO,IAAI,IAAU,eAgDO,GAC5B,MAAO,IAAI,IAAc,eA8BA,GACzB,MAAO,IAAI,IAAW,eAoCO,GAC7B,MAAO,IAAI,IAAe,eA+DR,GAClB,MAAO,IAAI,IAAI,eAUe,GAC9B,MAAO,IAAI,IAAgB,eAMC,GAC5B,MAAO,IAAI,IAAc,eAiDK,GAC9B,MAAO,IAAI,IAAgB,GAItB,GAAM,IAAkB,GAClB,GAAkB,GAClB,GAAY,GACZ,GAAY,eA2BK,GAC5B,MAAO,IAAI,IAAc,eA2BK,GAC9B,MAAO,IAAI,IAAgB,eAkCA,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,glCClnDU,EAAe,GAC5C,MAAO,IAAuB,EAAO,eAoBJ,EAAe,GAChD,MAAO,IAA2B,EAAO,eAuBvC,EAAe,GACjB,MAAO,IAAkC,EAAO,eAqBd,EAAe,GACjD,MAAO,IAA4B,EAAO,eAcJ,EAAe,GACrD,MAAO,IAAgC,EAAO,eAsCtB,EAAe,GACvC,MAAO,IAAkB,EAAO,eAsCX,EAAe,GACpC,MAAO,IAAe,EAAO,eAwBC,EAAe,GAC7C,MAAO,IAAuB,EAAO,eAuBL,EAAe,GAC/C,MAAO,IAAyB,EAAO,eAsBrC,EAAe,GACjB,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAqBlB,EAAe,GAC9C,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,kSC1TxC,yCCAA,YAqBqB,GACnB,MAAO,IAAI,IAAK,eAYC,GACjB,MAAO,IAAgB,eAYN,GACjB,MAAO,IAAgB,+CChDzB,mBAkBuC,EAAA,EAAA,GAAvC,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KAEE,SAAA,MAAqB,OAErB,SAAA,UAAA,SAAA,SAAS,GACP,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,OAAM,yDAElB,KAAK,MAAQ,GAEjB,GAVuC,IAsEvC,YAAc,EAAiB,GAC7B,MAAO,GAAU,EAGnB,YAAiB,EAAiB,GAChC,MAAO,GAAU,qBAOgB,EAAA,EAAA,GAcjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GAHA,AAAI,GAAQ,MACV,GAAO,IAEL,EAAK,mBACP,KAAM,IAAI,IACN,sEAGN,SAAK,QAAU,EAAK,SAAW,WAC/B,EAAK,SAAW,KAAK,IAAI,EAAK,UAAY,GAC1C,EAAK,SAAW,EAAK,UAAY,EACjC,EAAK,QAAU,EAAK,SAAW,EAC/B,EAAK,KAAO,EAAK,MAAQ,OACzB,EAAK,SAAW,EAAK,SAErB,AAAI,CAAC,OAAQ,MAAO,OAAO,QAAQ,EAAK,QAAU,IAChD,SAAQ,KACJ,uBAAuB,EAAK,KAAI,8CAEpC,EAAK,KAAO,QAGd,AAAI,EAAK,OAAS,MAChB,EAAK,YAAc,GACd,AAAI,EAAK,OAAS,OAInB,EAAK,QAAQ,QAAQ,SAAW,GAHpC,EAAK,YAAc,GAMjB,EAAK,YAAc,GAIvB,AAAI,EAAK,cAAgB,IACvB,GAAK,UAAY,MAIf,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,aAAe,EACpB,AAAI,KAAK,UAAY,KACnB,KAAK,KAAO,KAAK,SAEjB,KAAK,KAAO,KAAK,cAAgB,GAAO,SAAW,mBAIjD,EAAA,UAAA,WAAN,SAAiB,EAAe,mGAC9B,MAAA,CAAA,EAAM,GAAqB,WAE3B,MAFA,GAAA,OACM,EAAU,KAAK,gBAAgB,GACjC,GAAW,KACb,CAAA,GAGF,CAAI,KAAK,YAAY,EAAU,KAAK,SAAU,KAAK,MACjD,MAAK,KAAO,EACZ,KAAK,KAAO,GAGZ,MAAK,OACL,AAAI,KAAK,MAAQ,KAAK,UACpB,MAAK,aAAe,EACpB,KAAK,MAAM,aAAe,gBAM1B,EAAA,UAAA,WAAN,SAAiB,sEACf,MAAI,MAAK,aAAe,GAAK,KAAK,SAChC,QAAQ,IAAI,SAAS,KAAK,aAAY,8BAIlC,EAAA,UAAA,gBAAR,SAAwB,GACtB,AAAI,GAAQ,MACV,GAAO,IAET,GAAM,GAAe,EAAK,KAAK,SAC/B,MAAI,IAAgB,MAClB,QAAQ,KACJ,4BAA4B,KAAK,QAAO,sBACxC,2BAA0B,OAAO,KAAK,KAErC,GAEX,GAzGmC,gBAsJL,GAC5B,MAAO,IAAI,IAAc,MAGd,IAAY,CAAC,cAAa,i8DC9PvC,AA8BA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,WAAA,GAAA,aACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,UACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,aAAA,GAAA,eACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,IAAA,UACA,EAAA,EAAA,SAAA,IAAA,WACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,YAAA,IAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,iBAAA,KAAA,mBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,gBAAA,KAAA,oBA7BU,IAAA,IAAQ,QAyRH,IAAjB,AAAA,UAAiB,GAEf,GAAY,GAAZ,AAAA,UAAY,GAAyB,EAAA,EAAA,OAAA,GAAA,SAAc,EAAA,EAAA,GAAA,GAAA,KAAU,EAAA,EAAA,GAAA,GAAA,OAAjD,EAAA,EAAA,yBAAA,GAAA,wBAAuB,OAFpB,IAAA,IAAQ,KCtTzB,AAmBA,GAAM,IAAwC,eA2BnB,EAAc,GACvC,GAAM,GAAqB,CACzB,SAAU,EACV,SAAU,SACV,OAAQ,GACR,MAAO,GACP,eAAgB,GAGlB,GAAW,GAAQ,cAUW,GAC9B,MAAO,IAAW,eAUS,GAC3B,MAAO,IAAW,GC9EpB,WAwBI,EAAmB,EAAY,EAC/B,GACF,GAAM,GAAa,EAAK,YAAY,GACpC,GAAI,GAAc,EAAW,kBAAoB,QAC/C,GAAM,GAAQ,EAAW,gBACnB,EAAM,EAAW,gBAAkB,EACrC,OACC,EAAW,gBAAkB,OAAY,EAAQ,EACR,EAAW,cACzD,GAAI,EAAW,OAAS,SACtB,MAAO,IACH,EAAK,WAAW,EAAW,iBAAkB,EAAW,GAE9D,GAAI,EAAW,OAAS,WACtB,GAAM,GAAS,EAAK,WAAW,MAAM,EAAO,GAE5C,MAAO,GAAO,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAEvD,GAAM,GACF,GAAU,EAAK,WAAW,MAAM,GAAO,GAAI,EAAW,GACpD,EAAO,EAAO,WACpB,MAAO,GAAW,OAAS,SACvB,EAAK,GACL,EAAA,KAAK,cAAc,EAAO,MAAO,GAEvC,GAAM,GAAY,EAAK,WAAW,GAClC,MAAO,IAAa,EAAU,kBAU5B,EAAc,EACd,GACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAY,EAAQ,kBAAkB,KAAK,SAAA,GAC/C,MAAO,CAAC,CAAC,EAAW,GAAyB,EAAU,MAGzD,MAAO,KAAc,OACjB,EAAW,GAAyB,EAAU,IAAY,GAC1D,mBASF,EAAc,EACd,GACF,MAAO,GAAW,GAAyB,EAAM,EAAQ,+BAUvD,EAAmB,GACf,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEjB,MAAO,CACL,GAAyB,EAAU,GAAW,EAAQ,kBACtD,GAIJ,YAAkC,EAAc,GAC9C,MAAO,AAAE,GAAe,EAAI,IAAI,EAAc,cAGlB,GAC5B,GAAM,GAAQ,EAAK,MAAM,KACzB,GAAI,EAAM,SAAW,EACnB,MAAO,CAAC,EAAM,GAGhB,GAAM,GAAW,EAAM,GACvB,MAAO,CAAC,EAAU,OAAO,EAAM,EAAM,OAAS,iBAW5C,EAAY,EACZ,GACF,GAAI,GAAM,EAAc,MAAO,EAAM,EAAW,GAChD,GAAI,IAAQ,YAEV,EAAM,EAAc,mBAAoB,EAAM,EAAW,GAIzD,OAHM,GAEF,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,IACxB,EAAI,EAAG,EAAI,EAAG,IACrB,EAAgB,GAAG,GAAM,EAAiB,EAAI,GAC9C,EAAgB,GAAG,GAAM,EAAiB,EAAI,EAAI,GAEpD,MAAO,GAET,MAAO,eAYmB,GAC1B,MAAO,GAAO,KAAO,EAAS,EAAA,MAAM,GCrJtC,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,mCCnLxE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,UAC7D,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,YAGjE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC7dxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAElD,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,QAC1D,CAAC,OAAU,mBAAoB,KAAQ,iBAAkB,KAAQ,QACjE,CACE,OAAU,2BACV,KAAQ,yBACR,KAAQ,QAEV,CAAC,OAAU,oBAAqB,KAAQ,OAAQ,KAAQ,YAG5D,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CACR,OAAU,QACV,KAAQ,QACR,KAAQ,QACR,aAAgB,MAGpB,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAAU,CACrD,OAAU,wBACV,KAAQ,sBACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,YAG3C,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,YAE3D,CACE,SAAY,cACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,KACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,iBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,WAG9D,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,wCCnWrE,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAAa,CAC7D,OAAU,yBACV,KAAQ,sBACR,KAAQ,QAEV,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAC/C,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,OAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,WACV,KAAQ,WACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,gBAAiB,KAAQ,gBAAiB,KAAQ,QAAS,CACpE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,eACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,mBACV,KAAQ,gBACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,QAItB,CACE,SAAY,sBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,aAE9C,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,kBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,wBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,6BACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,aACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,QAAS,KAAQ,YAAa,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,yCCpVnD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,WAEtC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,SAAU,aAAgB,GAClE,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,SAAU,aAAgB,IAErE,MAAS,CACP,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAEzD,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,IAEjE,MAAS,CAAC,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAExD,CACE,SAAY,kBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,OACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAAW,CACpD,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,WAE7C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAC3C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,UAC7C,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,SACzC,CAAC,OAAU,eAAgB,KAAQ,eAAgB,KAAQ,wCC3KjE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,YAGnD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,WAEjD,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,cACV,KAAQ,YACR,KAAQ,QACR,aAAgB,IAElB,CACE,OAAU,yBACV,KAAQ,qBACR,KAAQ,UAId,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,UAC/C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,WAE5C,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCChGtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAE3D,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,yCCvC3C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,yBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CACE,SAAY,cACZ,SAAY,QACZ,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CAAC,SAAY,QAAS,SAAY,SAAU,CAC1C,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,YACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,SACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAEvC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,UAAW,CAC1D,OAAU,UACV,KAAQ,SACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,SACR,aAAgB,KAItB,CAAC,SAAY,OAAQ,SAAY,QAAS,OAAU,IAAK,CACvD,SAAY,eACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,0BACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,UACzC,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,yCC3G/C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,wBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,aAE3C,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAAW,CACxD,OAAU,sBACV,KAAQ,qBACR,KAAQ,yCCvDhB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,eACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC9ItB,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,eACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAAW,CAC3D,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCCtItB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,eACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GACrE,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,aACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,gBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCCjJtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,QACrD,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,uCC1GvD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,GAAI,KAAQ,UAAW,KAAQ,WACnD,CAAC,MAAS,GAAI,KAAQ,OAAQ,KAAQ,WAExC,MACI,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAEtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,WAClD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAG3E,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,OAAQ,aAAgB,MAGjE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,eACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,YACpC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CACP,CACE,OAAU,aACV,KAAQ,YACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,WACV,KAAQ,UACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,cACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,mBACV,KAAQ,iBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,YAEpD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGzE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,MACV,KAAQ,MACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,MAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAC/D,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,YACV,KAAQ,kBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,kBAAmB,KAAQ,YAChD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAG5C,CACE,SAAY,gBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCC1NtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,mCChDxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,OACV,KAAQ,SACR,KAAQ,QACR,aAAgB,IAElB,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAGhD,CACE,SAAY,aACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,MACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CAAC,CACR,OAAU,iBACV,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,QACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAAa,CACnD,MAAS,EACT,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,OACV,iBAAoB,eACpB,KAAQ,OACR,KAAQ,cAGZ,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,cAG7C,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,eACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,YAG5D,CACE,SAAY,cACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,iCChIb,AAyCA,GAAA,IAAA,WAWE,aACE,GAAM,GAAM,CACV,GAAY,GAAW,GAAS,GAAa,GAAU,GACvD,GAAY,GAAS,GAAO,GAAO,GAAU,GAAe,GAC5D,GAAW,GAAU,IAEjB,EAA0B,GAAG,OAAM,MAAT,GAAa,EAAI,IAAI,SAAA,GAAM,MAAA,GAAG,QAE9D,KAAK,UAAY,EAAY,OACzB,SAAC,EAAK,GACJ,SAAI,EAAO,UAAY,EAChB,GAET,IAlBN,cAAA,eAAkB,EAAA,WAAQ,KAA1B,WACE,MAAO,MAAK,WAAc,MAAK,UAAY,GAAI,wCAsBjD,EAAA,UAAA,eAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAM,KAChB,EAAuB,GACvB,EAAkB,GAClB,EAAoB,GACpB,EAAQ,EAAQ,OAA8B,SAAC,EAAK,GACxD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,GAAG,WAAW,eACrB,EAAa,KAAK,EAAI,EAAK,OACtB,AAAI,EAAK,KAAO,QACrB,EAAQ,KAAK,EAAI,EAAK,OACjB,AAAI,GAAK,OAAS,MAAQ,EAAK,MAAM,SAAW,IACrD,EAAU,KAAK,EAAI,EAAK,OAEnB,GACN,IAEC,EAAiB,GACf,EAAkB,GACpB,EAA8C,GAC9C,EAA+C,GACnD,AAAI,GAAa,MACf,GAAqB,KAAK,oBAAoB,EAAU,QACxD,EAAsB,KAAK,oBAAoB,EAAU,UAE3D,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAMlC,AAAI,OAAO,KAAK,GAAqB,SAAW,EAC9C,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,AAAI,EAAK,SAAS,SAAW,GAC3B,EAAQ,KAAK,KAIjB,OAAO,KAAK,GAAqB,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,aAAe,EAAoB,GACxC,EAAQ,KAAK,MAKnB,AAAI,OAAO,KAAK,GAAoB,OAAS,EAC3C,OAAO,KAAK,GAAoB,QAAQ,SAAA,GAC/B,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GACF,GAAK,aAAe,EAAmB,GACvC,EAAO,KAAK,MAIhB,EAAS,EAGX,GAAI,GAAY,GAChB,AAAI,EAAM,SAAW,MAAQ,EAAM,QAAQ,UAAY,MACrD,GAAY,EAAM,QAAQ,SAAS,OAAO,SAAC,EAAW,GACpD,SAAU,EAAK,UAAU,MAAQ,EAAK,YAAY,GAC3C,GACN,KAGL,GAAM,GACF,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,EAAE,UAAS,GAExE,MAAI,GAAU,OAAS,GACrB,GAAO,UAAY,GAGd,GAGD,EAAA,UAAA,oBAAR,SAA4B,GAC1B,MAAO,QAAO,KAAK,GAAW,IACzB,OAAgC,SAAC,EAAM,GACtC,SAAK,EAAQ,GAAM,MAAQ,EACpB,GACN,KAGD,EAAA,UAAA,QAAR,SAAgB,GAGd,GAAM,GACF,GAAgB,EAAK,KAAO,KAAK,UAAU,EAAK,KAAO,GAC3D,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAGd,GAAM,GAAgB,CACpB,KAAM,EAAK,KACX,GAAI,EAAK,GACT,SAAU,EAAO,SACjB,WACK,GAAK,OACL,IAAI,IAAI,SAAA,GAAS,MAAA,GAAM,WAAW,KAAO,EAAM,OAAO,GAAK,IAChE,OAAQ,GACR,SAAU,GACV,YAAa,GACb,WAAY,GACZ,SAAU,EAAK,MAGjB,MAAI,GAAO,QAAU,MACnB,GAAQ,YACJ,EAAO,OAAO,OACV,SAAC,EAAK,GACJ,SAAI,EAAM,MAAQ,CAChB,KAAM,EAAM,KACZ,gBAAiB,EAAM,MACvB,cAAe,EAAM,KAEhB,GAET,KAEV,AAAI,EAAO,OAAS,MAClB,GAAQ,WACJ,EAAO,MAAM,OAAoC,SAAC,EAAK,GACrD,GAAM,GAAO,EAAM,KACf,EAAQ,OACZ,OAAQ,EAAM,UACP,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAChB,EAAM,cAAgB,GAC3B,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,aACA,UACH,cAEA,KAAM,IAAI,OACN,2BAA2B,EAAM,KAAI,YAAY,EAAK,IAE9D,SAAI,EAAM,MAAQ,CAAC,MAAK,EAAE,KAAI,GACvB,GACN,KAEF,GAID,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAU,EAAY,QACtB,EAAuB,GACvB,EAAkB,GACpB,EAA+B,GACnC,AAAI,GAAW,MACb,GAAQ,EAAQ,OAA8B,SAAC,EAAK,GAClD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,KAAO,SACd,EAAQ,KAAK,EAAI,EAAK,OAEjB,GACN,KAEL,GAAM,GAAiB,GACjB,EAAkB,GAExB,EAAY,UAAU,SAAS,QAAQ,SAAA,GAC9B,GAAA,GAAA,GAAA,EAAA,MAAA,GACD,EAAa,CACjB,KAAM,EACN,GAAI,cACJ,OAAQ,GACR,WAAY,GACZ,SAAU,QACV,YAAa,GACb,WAAY,CAAC,MAAO,CAAC,MAAO,GAAgB,EAAI,MAAO,KAAM,UAC7D,SAAU,IAEZ,EAAK,aAAe,EAAI,KACxB,EAAO,KAAK,GACZ,EAAM,GAAY,IAGpB,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAIlC,GAAM,GAAgB,EAAY,IAElC,EAAY,UAAU,UAAU,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,cAAgB,EACrB,EAAQ,KAAK,MAIjB,GAAM,GAAY,KAAK,mBAAmB,GAC1C,MAAO,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,IAG1D,EAAA,UAAA,mBAAR,SAA2B,GAA3B,GAAA,GAAA,KAEE,MAAO,CACL,WAAY,EAAY,UAAU,KAClC,OAAQ,EAAY,UAAU,SAAS,OACnC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,GACjC,GAET,IACJ,QAAS,EAAY,UAAU,UAAU,OACrC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,EAAK,EAAY,KAClD,GAET,MAIA,EAAA,UAAA,mBAAR,SACI,EACA,GACF,GAAI,GAAO,EAAI,KACf,MAAI,IAAW,MACb,GAAO,EAAQ,IAEV,CAAC,KAAI,EAAE,MAAO,EAAI,OAE7B,iBAE6B,GAC3B,GAAM,GAAS,EAAA,MAAM,OACrB,GAAI,MAAO,GAAO,MAAS,YACzB,MAAO,GAAO,KAAK,GACd,GAAI,MAAO,SAAW,YAC3B,MAAO,IAAI,QAAO,EAAM,UAAU,WAElC,KAAM,IAAI,OACN,gGAKyB,EAAc,GAC7C,GAAM,GACF,MAAM,QAAQ,GAAK,OAAO,aAAa,MAAM,KAAM,GAAK,GAAa,GACzE,MAAO,GAAW,EAAQ,EAAM,0BAI9B,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,KACJ,GAAiB,EAAM,EAAG,GAE5B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAO,GAAQ,EAAM,EAAI,cAIvB,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,IAAS,GACvB,EACF,EAAM,GAAQ,KAAO,EAAM,EAAQ,EAAM,GAAQ,KAAO,EAAM,EAAO,EACzE,MAAQ,OAAO,IAAU,SAAY,EAAQ,SAAS,EAAO,gBAG/B,GAC9B,AAAI,MAAQ,IAAW,UAErB,GAAQ,GAAoB,IAE9B,OAAQ,OACD,IAAoB,SACvB,MAAO,cACJ,IAAoB,aACpB,IAAoB,aACpB,IAAoB,YACpB,IAAoB,SACvB,MAAO,YACJ,IAAoB,QACvB,MAAO,WACJ,IAAoB,UACvB,MAAO,cACJ,IAAoB,UACvB,MAAO,iBAIP,MAAO,mBAKT,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,EAAM,KAAK,KAEb,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,GAAgB,EAAM,MAExB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,KAC7B,EAAM,KAAK,KAAK,IAAI,SAAA,GAAK,MAAA,IAAgB,KAE3C,cAG6B,GAEpC,MAAI,GAAM,YACR,OAEE,EAAM,KAAO,KACR,EAAM,IAAI,IACb,SAAA,GACI,MAAC,OAAO,GAAI,MAAS,SAAY,EAAI,KAAO,SAAS,EAAI,KAAM,MAElE,eAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MACV,GAAsB,EAAM,OAE9B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,GACO,IAAM,KAAK,GAAK,EAAM,KAAK,EAAE,OAAS,EAAM,KAAK,EACX,EAAM,KAAK,IAClD,IACH,IAAI,SAAA,GAAK,MAAC,OAAO,IAAM,SAAY,EAAI,SAAS,EAAG,MAEnD,cAIL,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAAE,IAAI,SAAC,GACvB,MAAO,IAAiB,EAAG,KAGxB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,MAC7B,EAAM,KAAK,MAAM,IAAI,SAAC,GAC3B,MAAO,IAAsB,KAG1B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAEb,ECtkBT,AA4BA,GAAA,IAAA,WAGE,WACY,EAAoB,EACpB,GAFZ,GAAA,GAAA,KACY,KAAA,KAAA,EAAoB,KAAA,UAAA,EACpB,KAAA,QAAA,EAJI,KAAA,OAAmB,GACnB,KAAA,MAAoC,GAIlD,KAAK,OAAS,EAAK,WAAW,IAAI,SAAA,GAAQ,MAAA,GAAK,SAAS,KACxD,AAAI,EAAK,UAAY,MACnB,MAAK,MAAQ,OAAO,KAAK,EAAK,UACZ,OAAO,SAAC,EAAmC,GAC1C,SAAM,GAAO,EAAK,QAAQ,GACnB,GACN,KAQhB,SAAA,UAAA,SAAR,SAAiB,GACf,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,UAOtC,EAAA,UAAA,QAAR,SAAgB,EAAc,GAC5B,GAAM,GAAQ,KAAK,KAAK,SAAS,GACjC,GAAI,EAAM,QAAU,KAClB,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,SAE9C,GAAI,EAAM,GAAK,MAAQ,EAAM,GAAK,KAChC,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAa,KAAK,KAAK,SAAU,EAAM,GAEhD,GAAI,EAAM,OAAS,KACjB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,MAAQ,KAChB,MAAO,IAAc,KAAK,KAAK,SAAU,EAAM,GAEjD,GAAI,EAAM,MAAQ,MAChB,GAAI,EAAM,KAAK,GAAK,MAAQ,EAAM,KAAK,GAAK,KAC1C,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,OAAS,KACtB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,MAAQ,KACrB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAIlC,MAAO,IAEX,KCtGA,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,cACA,YACA,MACH,MAAO,CAAC,EAAA,IACH,EAAc,IAAK,EAAM,EAAW,GACrC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,UAAW,EAAM,EAAW,SAE3C,eACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,cACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,oBACH,MAAO,CAAC,EAAA,kBACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC7F9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACA,aACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,OAAQ,EAAM,EAAW,SACxC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,YACA,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,eAAgB,EAAM,EAAW,GAC/C,EAAc,eAAgB,EAAM,EAAW,SAEhD,QACH,MAAO,CAAC,EAAA,MACJ,GAAU,EAAK,WAAW,GAAI,EAAW,SAC1C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SACxC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SACzC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAE5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC5K9C,YAyBI,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EAAA,KAAK,OACD,GAA8B,EAAQ,GACtC,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlB,EAAc,GAC1D,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,IAAM,EAAG,KAAO,IAAM,EAAG,KAAO,EAAG,GAC/C,MAAO,GAGX,MAAO,GCzCT,AA+BA,GAAA,IAAA,WAIE,WACa,EAAuB,EAAyB,EACjD,EAAiC,EAChC,EAA+B,GAF/B,KAAA,KAAA,EAAuB,KAAA,MAAA,EAAyB,KAAA,QAAA,EACjD,KAAA,aAAA,EAAiC,KAAA,uBAAA,EAChC,KAAA,YAAA,EAA+B,KAAA,eAAA,EANpC,KAAA,QAA6B,GAC7B,KAAA,QAAU,GAMhB,KAAK,SAAW,EAAA,OAAO,GACvB,EAAA,KAAK,KAAK,UAGZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAMd,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,OAAO,MAChD,EAAO,OAAO,YAGlB,KAAK,QAAU,GACf,KAAK,QAAU,GACf,KAAK,SAAS,WAGhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAOtB,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,GAAS,KAAK,OAC7B,KAAM,IAAI,OAAM,4BAA4B,EAAK,wBAC7C,KAAK,QAGX,GAAM,GAAkB,KAAK,QAAQ,GACrC,GAAI,EAAgB,QAClB,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0BACpB,EAAK,wGAIf,MAAI,MAAK,gBACP,GAAgB,QAAU,IAG5B,EAAgB,KAAO,GAChB,EAAgB,QAMzB,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GAAS,MAAA,GAAK,KAAK,MAQxC,EAAA,UAAA,MAAA,SAAM,EAAe,GACnB,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,CAAC,KAAK,aAAe,GAAS,KAAK,QAClD,KAAM,IAAI,OAAM,2BACZ,EAAK,8CAA8C,KAAK,SAG9D,GAAM,GAAI,KAAK,QAAQ,IAAU,GAEjC,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,eACZ,KAAK,KAAI,0CAA0C,EAAK;uCAExD,EAAO,MAAK,8BAA8B,KAAK,MAAK,KAc1D,GAVA,AAAI,KAAK,SAAW,GACf,MAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,IAC7D,MAAK,aAAe,EAAO,OAG7B,GACI,KAAK,aAAc,EAAO,MAC1B,eAAe,KAAK,KAAI,0CACpB,EAAK,KAET,EAAE,KACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,uCAGf,GAAI,EAAE,QACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,0CAGf,EAAE,OAAS,EACX,EAAA,KAAK,GACL,EAAE,QAAU,GAEZ,KAAK,QAAQ,GAAS,GAMxB,EAAA,UAAA,UAAA,SAAU,EAAmB,GAA7B,GAAA,GAAA,KACE,GAAI,EAAQ,SAAW,EAAQ,OAC7B,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,sCACxB,4BACI,EAAQ,OAAM,qCACd,EAAQ,OAAM,MAGxB,EAAQ,QAAQ,SAAC,EAAG,GAAU,MAAA,GAAK,MAAM,EAAG,EAAQ,OAWtD,EAAA,UAAA,OAAA,SAAO,EAAoB,GACzB,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAK,EAMH,EAAU,EAAQ,MAAM,EAAG,KAAK,aALhC,EAAU,GACV,OAAS,GAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAMjB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAKpC,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAAO,gCAElC,EAAA,MAAM,EAAS,IAMxB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAI,KAAK,SAAW,EAClB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAIpC,OADM,GAAU,GACP,EAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAGf,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAC9B,mDACI,KAAK,aAAY,4BAA4B,EAAQ,GAAG,MAAK,KAE9D,EAAA,OAAO,EAAS,IASzB,EAAA,UAAA,QAAA,SAAQ,EAAmB,GACzB,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAGhD,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,CAAC,KAAK,aAAe,GAAY,KAAK,QACxC,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,KAAK,QAAO,KAGtE,KAAK,UAAU,EAAS,EAAA,QAAQ,EAAQ,KAS1C,EAAA,UAAA,MAAA,SAAM,EAAkB,GAAxB,GAAA,GAAA,KACE,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAEhD,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEZ,EAAW,4BAA4B,EAAO,OAGpD,GAAI,CAAC,KAAK,aAAe,EAAO,SAAW,KAAK,QAC9C,KAAM,IAAI,OACN,2DACI,KAAK,QAAO,QAAQ,EAAO,OAAM,kEAI3C,GAAM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,GAC1B,EAAA,KAAK,WACH,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,EAAK,cAE3D,MAAO,KAGT,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAQ,GAAK,EAEf,KAAK,UAAU,EAAS,IAE5B,KC3TA,AAoCA,GAAA,IAAA,WAeE,WACa,EAA4B,EAC5B,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IADxB,KAAA,QAAA,EAA4B,KAAA,aAAA,EAC5B,KAAA,aAAA,EACX,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,IAAiB,EAAO,MAC1B,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,EAAO,OAEhD,GACI,EAAc,EAAO,MAAO,+BAEhC,EAAA,KAAK,KAGT,KAAK,SAAW,EAAA,OAAO,GACvB,KAAK,eAAiB,EACtB,EAAA,KAAK,KAAK,UA5BZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAiCvB,EAAA,UAAA,KAAA,WACE,MAAO,IAAI,GACH,KAAK,QAAO,QAAG,KAAK,aAAc,KAAK,eAMjD,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,MACzC,EAAO,YAGX,KAAK,QAAQ,OAAS,EACtB,KAAK,SAAS,WAKhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAUtB,EAAA,UAAA,MAAA,SAAM,EAAwB,EAAwB,GAAtD,GAAA,GAAA,KAEE,GAFoD,AAAA,IAAA,QAAA,GAAA,IAEhD,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,IAAgB,IAAM,KAAK,QAAQ,SAAW,EAChD,KAAM,IAAI,OAAM,kCACZ,EAAW,iCACX,KAAK,QAAQ,OAAM,cAEzB,UACI,EAAc,KAAK,aAAc,+BAC9B,EAAA,KAAK,WACV,GAAM,GACF,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAA,QAAQ,EAAQ,KAC/C,MAAO,GAAA,MAAM,EAAiB,MASlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAC9B,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAG9C,GAAI,KAAK,SAAW,EAClB,KAAM,IAAI,OAAM,qCAGlB,GAAM,GAAS,KAAK,QAAQ,MAC5B,UACI,EAAO,MAAO,EAAc,+BACzB,EAAA,QAAQ,EAAQ,IAOzB,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAM9C,GAHA,GACI,EAAO,MAAO,KAAK,aAAc,+BAEjC,KAAK,iBAAmB,KAAK,OAC/B,KAAM,IAAI,OAAM,4CAElB,EAAA,KAAK,GACL,KAAK,QAAQ,KAAK,IAOpB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAO,EACT,KAAM,IAAI,OACN,0DAA0D,GAGhE,GAAI,KAAK,iBAAmB,IAAM,EAAO,KAAK,eAC5C,KAAM,IAAI,OAAM,+BACZ,EAAI,6BAA6B,KAAK,eAAc,KAE1D,KAAK,QAAQ,OAAS,GASxB,EAAA,UAAA,QAAA,SAAQ,EAAsB,EAAwB,GAEpD,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,EAAe,GAAK,EAAe,KAAK,QAAQ,OAClD,KAAM,IAAI,OAAM,4BACZ,EAAY,mBAAmB,KAAK,QAAQ,OAAM,cAGxD,GAAI,KAAK,QAAQ,IAAiB,KAChC,KAAM,IAAI,OAAM,oBAAoB,EAAY,aAGlD,UACI,KAAK,QAAQ,GAAc,MAAO,EAClC,+BAEG,KAAK,QAAQ,IAQtB,EAAA,UAAA,QAAA,SAAQ,EAAsB,GAC5B,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAG9C,GAAI,EAAe,GACf,KAAK,iBAAmB,IAAM,GAAgB,KAAK,eACrD,KAAM,IAAI,OAAM,yBACZ,EAAY,uBAAuB,KAAK,eAAc,cAG5D,GACI,KAAK,aAAc,EAAO,MAAO,+BACrC,EAAA,KAAK,GACL,KAAK,QAAQ,GAAgB,GAU/B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAwB,GAAlD,GAAA,GAAA,KAEE,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAU9C,MAPA,IACI,KAAK,aAAc,EAAc,+BAIrC,EAAU,EAAQ,MAAM,EAAG,KAAK,QAE5B,EAAQ,SAAW,EACd,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAK,QAAQ,GAAI,KAC1D,MAAO,GAAA,MAAM,EAAS,MAS1B,EAAA,UAAA,OAAA,SAAO,EAAwB,GAA/B,GAAA,GAAA,KACE,GAAI,CAAC,CAAC,GAAgB,IAAiB,KAAK,aAC1C,KAAM,IAAI,OAAM,uBACZ,KAAK,aAAY,+BAA+B,GAMtD,MAHA,IACI,KAAK,aAAc,EAAc,+BAEjC,KAAK,SAAW,EACX,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAK,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAG,KACjD,MAAO,GAAA,OAAO,EAAS,MAG7B,iBAQI,EAAgB,EAAwB,GAC1C,GAAM,GAAQ,EAAO,MACrB,GAAI,EAAO,MAAM,OAAS,EACxB,KAAM,IAAI,OACN,oDAAoD,EAAO,OAEjE,GAAI,EAAO,QAAU,EACnB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,GAEzC,GAAM,GAAc,EAAO,MAAM,MAAM,GACvC,GACI,EAAa,EAAc,+BAE/B,GAAM,GAAuB,EAAA,QAAQ,GACrC,MAAO,IAAI,IAAW,EAAY,EAAc,eAU9C,EAAwB,EAAwB,GAClD,MAAO,IAAI,IAAW,GAAI,EAAc,EAAc,eAWpD,EAAgB,EAAmB,EACnC,GACF,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,GAAe,MAAQ,IAAgB,IAAM,GAAY,EAC3D,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,EAAW,KAGrE,GAAM,GAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,GACtD,EAAU,EAAA,QAAQ,EAAQ,GAChC,SAAQ,QAAQ,SAAC,EAAO,GACtB,EAAK,QAAQ,EAAO,EAAQ,MAEvB,cAWL,EAAgB,EAAkB,GACpC,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEV,EAAW,4BAA4B,EAAO,OAmBtD,OAhBM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,EAAA,KAAK,WAC7B,GAAM,GAAU,GAChB,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,GAEtD,SAAO,UACA,IAGH,EAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,EAAO,QAE1D,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAK,QAAQ,EAAG,EAAQ,IAE1B,MAAO,GC3YT,AAiBA,GAAA,IAAA,OAUa,GAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,kUACnB,EAAA,EAAK,iBACN,KAAA,MAAA,CAAA,EAAA,OACA,cAAA,MAAA,CAAA,EAAA,OAgBA,QAAA,MAAA,CAAA,EAAA,OACA,iBAAA,MAAA,CAAA,EAAA,OAsDA,WAAA,MAAA,CAAA,EAAA,OAIA,SAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QAOA,OAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAoBA,qBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAQA,sBAAA,MAAA,CAAA,EAAA,QAUA,uBAAA,MAAA,CAAA,EAAA,QAWA,sBAAA,MAAA,CAAA,EAAA,QAQA,qBAAA,MAAA,CAAA,EAAA,QAWA,oBAAA,MAAA,CAAA,EAAA,QAMA,qBAAA,MAAA,CAAA,EAAA,QAOA,oBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAaA,sBAAA,MAAA,CAAA,EAAA,QACA,oBAAA,MAAA,CAAA,EAAA,QAcA,oBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAYA,kBAAA,MAAA,CAAA,EAAA,QAYA,uBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAUA,qBAAA,MAAA,CAAA,EAAA,QASA,oBAAA,MAAA,CAAA,EAAA,QAUA,kBAAA,MAAA,CAAA,EAAA,wBAzTG,SACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClC,CAAA,EAAM,EAAK,eAC7B,MADM,GAAY,GAAA,OACd,EAAU,GACZ,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,gBAE1C,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAKtC,SACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAO,EAAc,OAAQ,EAAM,EAAW,GAI/C,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAFxC,SACD,GAAA,OAEC,EAAS,EAAK,IAAI,SAAA,IAAU,MAAA,IAAO,KACzB,CAAA,EAAM,EAAW,GAAG,eAAhC,EAAY,GAAA,OAEhB,EAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAChD,GAAO,YAIP,EAAmB,gFAIf,UAAa,EAEV,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACzC,EAAQ,EAAQ,eAAgB,EAAQ,uBAD5C,SAAS,GAAA,OAEH,GAAY,EAAO,IAAI,SAAA,IAAU,MAAA,IAAO,KAI9C,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,YAMN,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAQ,EAAQ,eAAgB,EAAQ,uBAF1C,UACD,GAAA,OAEO,CAAA,EAAM,GAAW,GAAG,eAAhC,SAAY,GAAA,OAEZ,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,4CA1BN,GAAU,WAAE,CAAA,EAAA,wCA8BnB,MAAA,CAAA,EAAO,UAGD,SAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAChD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClD,AAAK,EAAK,MACR,GAAO,GAAY,IAGb,CAAA,EAAM,EAAK,gBAAnB,MAAA,CAAA,EAAQ,GAAA,OAAmB,GAAK,CAAC,OAAW,GAAQ,CAAC,EAAM,iBAK3D,MAFM,GAAY,EAAK,WAAW,KAC9B,SAAA,IAAQ,MAAA,IAAU,GAAM,EAAW,KAAa,SAChD,EACI,GAAO,GAAU,EAAW,EAAW,GAC7C,CAAA,EAAO,CAAC,GAAY,MAEtB,CAAA,EAAO,gBAGD,SACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,WAAW,GACnB,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,YACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,gBACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,yBAA0B,EAAM,EAAW,GAEvD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAc,GAAI,IACpB,EAAM,EAAO,EAAM,EAAc,EAAwB,EACzD,GACJ,EAAQ,eAAe,GACvB,CAAA,EAAO,CAAC,EAAY,SAAU,EAAA,OAAO,aAG/B,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAmB,EAAQ,eAAe,EAAG,IACnD,EAAiB,MAAM,EAAO,GAC9B,CAAA,EAAO,CAAC,EAAiB,mBAGnB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAkB,EAAQ,eAAe,EAAO,IACtD,CAAA,EAAO,CAAC,EAAgB,KAAK,aAGvB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAoB,EAAQ,eAAe,EAAS,IAC1D,CAAA,EAAO,CAAC,EAAkB,OAAO,EAAe,aAG1C,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAqB,EAAQ,eAAe,GAAU,IAC5D,GAAmB,QAAQ,GAAgB,IAC3C,CAAA,EAAO,CAAC,GAAmB,mBAGrB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAoB,EAAQ,eAAe,GAAS,IACpD,GACF,EAAc,QAAS,EAAM,EAAW,GAC5C,CAAA,EAAO,CAAC,GAAkB,OAAO,cAG3B,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,MAAM,GAAS,IAChC,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAkB,EAAQ,eAAe,GAAO,IACtD,CAAA,EAAO,CAAC,EAAA,OAAO,GAAgB,OAAQ,mBAGjC,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,gBACjB,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,QAAQ,EAAO,GAC1B,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAE7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAW,EAAc,cAI9C,UACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GACF,GAAQ,GAAe,GAAgB,EAAc,IACzD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,GAAQ,EAAc,GAAc,IACvD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,EAAS,IAClD,CAAA,EAAO,CAAC,GAAW,OAAO,EAAe,GAAc,aAGjD,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,MAAM,EAAc,GAAc,cAG/C,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,GAAW,GAAQ,EAAc,IACpD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAC5C,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,CAAA,EAAO,CAAC,GAAW,OAAO,GAAa,aAGjC,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,SAAS,GACpB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAc,cAGnC,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,UAAW,EAAM,EAAW,GAExC,GAAa,GAAM,GAAa,GAAS,GAC/C,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGnB,KAAM,WAAU,aAAa,EAAK,GAAE,6BCvW1C,AA2BA,YACI,EAAY,EAA4B,GACpC,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAC7B,EAAc,IAAY,iBAE1B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,yGAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,oFAIR,GAAI,EACF,KAAM,IAAI,OACN,yEAEN,GAAM,GAAS,EAAc,UAAW,EAAM,EAAW,GACnD,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGhB,MAAO,CACL,OAAM,EACN,IAAG,EACH,WAAU,EACV,UAAS,EACT,QAAO,EACP,SAAQ,EACR,eAAc,GAIX,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAQ,EAAyB,EACjC,QAED,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,gBACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAIvB,8BACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,gBAAgB,CAClC,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAGvB,0BACA,mBACH,GAAM,GAAQ,EACI,cAAe,EAAM,EACrB,GAEZ,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GACxC,MAAO,CAAC,EAAA,gBACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAO,CAAC,EAAO,GAAI,EAAO,IAAK,QAEhC,4BACA,mBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cAET,MAAO,CAAC,EAAA,gBACJ,EAAc,QAAS,EAAM,EAAW,GAExC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GAEzC,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,EACnC,EACA,CAAC,EAAU,GAAI,EAAU,GAAI,EAAU,UAExC,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,qBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,sBAAuB,EAAM,EAAW,GAEpD,EAAA,EAAA,kBAAA,EAAA,IAAA,EAAA,EAAA,GAAA,CAAA,EAAA,GAAA,EAAA,IAAA,CAAA,EAAA,GAAA,EAAA,IAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,QAIf,MAAO,CAAC,EAAQ,OAEb,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,cACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,YAAa,EAAM,EAAW,GAG1C,EAAe,EAAQ,GACvB,EAAc,EAAQ,GAGtB,EAAiB,EAAU,GAC3B,EAAgB,EAAU,GAEhC,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAc,GAAc,EAC7B,CAAC,EAAgB,GAAgB,iBAIrC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC/S9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,KAAW,EAAO,EAAO,QAE9B,YACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAM,EAAc,MAAO,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,SAAe,EAAO,EAAM,QAEjC,eACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,YAAkB,EAAQ,EAAY,QAE3C,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OAAa,EAAS,EAAO,EAAS,QAE3C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cAEJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,SAEzC,SACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,MACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,SAGzC,mBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,gBACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,GAExC,QAED,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1H9C,AAiBA,GAAA,IAAA,OAUA,YACI,EAAY,EAA4B,GAC1C,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EAAS,EAAc,SAAU,EAAM,EAAW,GAClD,EACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CACL,MAAK,EACL,OAAM,EACN,cAAa,EACb,aAAY,EACZ,eAAc,EACd,aAAY,GAIT,GAAM,IAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,mHACnB,EAAA,EAAK,iBACN,sBAAA,MAAA,CAAA,EAAA,OAgBA,sBAAA,MAAA,CAAA,EAAA,OAcA,sBAAA,MAAA,CAAA,EAAA,OACA,sBAAA,MAAA,CAAA,EAAA,OAQA,QAAA,MAAA,CAAA,EAAA,OAQA,WAAA,MAAA,CAAA,EAAA,uBA9CG,SAOF,GAAU,EAAM,EAAW,GAN7B,EAAK,EAAA,MACL,EAAM,EAAA,OACN,EAAa,EAAA,cACb,EAAY,EAAA,aACZ,EAAc,EAAA,eACd,EAAY,EAAA,aAGC,CAAA,EAAM,EAAA,MAAY,gCAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,wBAGjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAG3D,EACF,EAAc,qBAAsB,EAAM,EAAW,GAG1C,CAAA,EAAM,EAAA,MAAY,6BAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,sBAIjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAGzD,CAAA,EAAM,EAAA,MAAY,uBACtB,EAAmB,EAAoB,EAAe,EACtD,WAFJ,MAAA,CAAA,EAAA,CAAQ,EAAA,gBAKF,SAAY,EAAA,KACb,EAAc,YAAa,EAAM,EAAW,GAC7C,QACY,CAAA,EAAM,EAAA,WAAiB,WAAjC,SAAM,CAAI,EAAA,QAChB,EAAU,UACV,CAAA,EAAO,UAGP,MAAA,CAAA,EAAO,EAAA,eACH,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,6BC3G1C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAS,EAAA,KAAW,EAAG,EAAG,GAChC,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAS,EAAA,OAAa,GAC5B,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAS,EAAA,OAAa,EAAG,GAC/B,MAAO,CAAC,EAAO,OAAQ,EAAO,iBAG9B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCpDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,GAAU,EAAK,UAEnB,yBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,IAAY,OACjD,cACH,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,QACrC,eACA,mBACA,2BACH,GAAM,GAAO,EAAc,IAAK,EAAM,EAAW,GACjD,MAAO,CAAC,GAAY,QAEjB,YACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,IAAY,SACjC,WACH,GAAM,GACD,EAAc,IAAK,EAAM,EAAW,GACzC,MAAO,CAAC,GAAY,QACjB,QACH,MAAO,CAAC,EAAA,SACH,EAAc,IAAK,EAAM,EAAW,GAAoB,MACzD,cACD,SACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,GAAA,SAAe,EAAE,aACtC,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OAAa,QAClB,QACH,GAAM,GAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,QAAQ,KACJ,kGAEJ,QAAQ,IAAI,GACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,QAAQ,IAAI,MAAM,UAAU,MAAM,KAAK,EAAK,GAAG,YAC9B,MAAM,EAAG,IAE5B,MAAO,CAAC,WAGR,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,eAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,yBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,sBAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,iBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,WAAY,EAAM,EAAW,GACzC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,qBAAsB,EAAM,EAAW,GAEzD,MAAO,CAAC,EAAA,MAAY,cAChB,EAAmB,EAAmB,EACtC,EAA8B,EAC9B,YAGJ,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzE9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,eACH,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACA,WACH,MAAO,CAAC,EAAA,MACJ,EAAc,YAAa,EAAM,EAAW,GAC5C,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACA,oBACA,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,aAAc,EAAM,EAAW,GAC7C,EAAc,aAAc,EAAM,EAAW,SAG9C,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SAExC,eACG,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAE7B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,sFAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,iEAGF,GAAA,GAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEhB,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,KAAM,EACN,WAAY,EACZ,uBAAwB,aAI1B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCjF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,qBACA,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,MACH,MAAO,CAAC,EAAA,2BACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,OAAQ,EAAM,EAAW,SAExC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cACJ,EAAc,gBAAiB,EAAM,EAAW,GAEhD,EAAc,cAAe,EAAM,EAAW,GAC9C,EAAc,eAAgB,EAAM,EAAW,GAE/C,EAAc,eAAgB,EAAM,EAAW,aAInD,KAAM,WAAU,aAAa,EAAK,GAAE,yBC9E9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,EAAW,YAGf,KAAM,WAAU,aAAa,EAAK,GAAE,yBCtH9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,eACA,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACvC,EACA,EAAc,UAAW,EAAM,EAAW,GAC9C,SAAS,EAAO,MAAM,EAAG,GAClB,CAAC,EAAA,OAAa,EAAQ,QAE1B,eACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OAAa,EAAO,EAAA,KAAW,EAAS,SAAU,QAEvD,gBACA,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,QAAc,EAAO,QAE1B,SAEH,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAEhD,EAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,gBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,MAAO,EAAM,EAAW,GACpC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAE/C,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,aACJ,EAAQ,EAAO,EAAK,EAAS,EAAW,EAAS,EACjD,EAAa,QAEd,OACH,MAAO,GAAA,KAAK,WACV,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GAGxC,EAAQ,EAAQ,GAAG,MACnB,EAAgB,EAAA,QAAc,EAAQ,IAAI,MAC1C,EAAS,EAAQ,IAAI,SAAA,GACzB,GAAM,GAAY,EAAA,KAAK,YAAY,EAAO,MAAO,GACjD,GAAI,CAAC,GACD,CAAC,EAAA,KAAK,YACF,EAAA,QAAc,GAAQ,MAAO,GACnC,KAAM,IAAI,OAAM,0CAElB,MAAO,GAAY,EAAS,EAAA,QAAc,EAAQ,KAEpD,MAAO,CAAC,EAAA,MAAY,EAAQ,UAG3B,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GAC7C,MAAO,GAAA,QAAc,EAAQ,OAE1B,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,YACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,kBAAmB,EAAM,EAAW,GAGhD,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,GAAA,MAAY,EAAQ,EAAiB,OAEzC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,UAAgB,EAAS,EAAQ,QAEtC,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,SAAe,EAAG,QAEvB,iBACH,GAAM,GACF,EAAc,gBAAiB,EAAM,EAAW,GAE9C,EACF,EAAc,cAAe,EAAM,EAAW,GAE5C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,MAAO,CAAC,EAAA,cACJ,EAAS,EAAc,EACvB,EAAa,QAAU,EAAa,MAChC,EACA,EAAA,KAAW,EAAc,EAAa,iBAG9C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1K9C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBChDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAGzC,cACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAGzD,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,UAAW,EAAM,EAAW,GAE1C,EAAc,gBAAiB,EAAM,EAAW,SAGjD,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,gBACH,GAAM,GACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACrC,cAEb,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAW,QAEZ,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAG5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnG9C,YAgDI,EAAY,EACZ,GACF,GAAM,GACD,SAAC,EAAY,EAA4B,GACxC,OAAQ,EAAK,cACN,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,cACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAsB,EAAM,EAAW,SAC9C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,UACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAkB,EAAM,EAAW,SACtD,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,gBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAwB,EAAM,EAAW,SAChD,YACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,iBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAyB,EAAM,EAAW,SACjD,SACH,GAAM,GAAW,GAAgB,EAAK,IACtC,GAAI,GAAY,EAAS,eACvB,MAAO,GAAS,eACZ,GAAI,IAAc,EAAM,EAAW,IAEvC,KAAM,WAAU,aAAa,EAAK,GAAE,+BAGtC,KAAM,WACF,eAAe,EAAK,GAAE,yIAI7B,EAAM,EAAW,GACxB,MAAI,aAAiB,SACZ,EAAM,KAAK,SAAC,GAAS,MAAA,GAAG,OAAO,KAEjC,GAAG,OAAO,GCvEnB,GAAA,IAAA,WAME,WACa,EACA,EACA,EACA,GAHA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IAHA,KAAA,UAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,YAAA,EATL,KAAA,YAAc,CAAC,GAAI,EAAG,UAAW,GAAI,YAAa,GAClD,KAAA,SAAmC,CAAC,KAAK,aACzC,KAAA,OAAS,EAQf,KAAK,4BAGC,SAAA,UAAA,SAAR,SAAiB,EAAY,GAC3B,MAAO,CAAC,GAAE,EAAE,UAAS,EAAE,YAAa,IAQtC,OAAA,eAAI,EAAA,UAAA,iBAAc,KAOlB,WACE,MAAO,MAAK,cARd,SAAmB,GACjB,AAAI,KAAK,WAAa,GACpB,MAAK,SAAW,EAChB,KAAK,8DAWT,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,mBAAmB,oCAOjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGN,EAAA,UAAA,0BAAR,WAEE,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,KAAK,SAAS,OAAS,EAAG,KAC5C,GAAM,GAAW,KAAK,SAAS,MAAM,EAAG,KAAK,SAAS,OAAS,GAC/D,EAAM,KAAK,KAAK,qBAAqB,IAEvC,EAAM,KAAK,IACX,KAAK,mBAAqB,GAGpB,EAAA,UAAA,qBAAR,SAA6B,GAC3B,MAAO,GACH,EACK,IACG,SAAA,GAAW,MAAC,GAAQ,KAAO,GAAK,EAAQ,cAAgB,EACpD,GACG,EAAQ,UAAS,IAAI,EAAQ,cACvC,KAAK,KACV,IAON,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,UACP,MAAK,SACL,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,KAAK,KAAK,SAAS,KAAK,OAAQ,IAC9C,KAAK,mBAAmB,QAAQ,KAAK,qBAAqB,KAAK,aAQnE,EAAA,UAAA,UAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,EAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,OAAO,IACrB,KAAK,kBAAkB,YAEvB,MAAM,IAAI,OAAM,4CAQpB,EAAA,UAAA,cAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,GAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SACL,GAAM,GACF,OAAO,OAAO,GAAI,KAAK,SAAS,KAAK,SAAS,OAAS,IAC3D,EAAQ,aAAe,EACvB,EAAQ,GAAK,KAAK,OAClB,KAAK,SAAS,OAAO,GAAI,EAAG,GAC5B,KAAK,mBAAmB,OACpB,EAAG,EAAG,KAAK,qBAAqB,KAAK,eAEzC,MAAM,IAAI,OAAM,0DAIpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,UAAU,IAGxB,EAAA,UAAA,eAAA,SAAe,GACb,KAAK,eAAe,EAAY,IAAM,GAGxC,EAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,eAAe,IAG7B,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,cAAc,EAAW,IAAM,GAGtC,EAAA,UAAA,cAAA,SAAc,GACZ,MAAO,MAAK,cAAc,IAG5B,EAAA,UAAA,QAAA,SAAQ,GACN,OAAW,KAAO,MAAK,eACrB,KAAK,eAAe,GAAK,cAAc,GAGzC,OAAW,KAAO,MAAK,cACrB,KAAK,cAAc,GAAK,cAAc,IAG5C,KC1LA,YAyCI,EAAwB,EAAiB,EACzC,GACF,GAAM,GAAY,GAAI,KAChB,EAA0B,GAC5B,EAAoB,KACpB,EAAuB,KAIrB,EAAO,GAAI,KACX,EACF,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAEpD,EAA0B,GAC9B,AAAI,GAAa,MACf,GAAgB,EAAU,IAAI,SAAA,GAAQ,MAAA,IAAc,EAAK,MAAM,MAIjE,OADM,GAAe,EAAO,QACrB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MAWtB,GAVA,AAAI,IAAc,IAAS,GAAe,KACxC,CAAI,GAAe,MACjB,GAAc,EACd,EAAa,EAAY,SAAS,IAAI,SAAA,GAAS,MAAA,GAAM,OACnC,OAAO,SAAA,GAAQ,MAAA,GAAU,IAAI,OAGnD,EAAU,IAAI,EAAK,MAGf,EAAU,EAAK,OAAS,KAC1B,SAGF,GAAI,EAAe,QAAQ,EAAK,QAAU,GACxC,SAGF,GAAI,EAAc,QAAQ,EAAK,QAAU,GACvC,SAEF,GAAI,EAAK,OAAO,SAAW,GACzB,EAAc,KAAK,EAAK,MACxB,SAEF,EAAK,OAAO,QAAQ,SAAA,GAElB,GAAI,EAAK,IAAI,EAAM,MACjB,OAEF,EAAK,IAAI,EAAM,MACf,EAAS,KAAK,KAGlB,MAAO,CAAC,OAAM,EAAE,QAAO,EAAE,UAAS,EAAE,cAAa,EAAE,YAAW,EAAE,WAAU,eAQxE,EAAc,EACd,GACK,GAAA,GAAA,EAAA,UAAW,EAAA,EAAA,OACZ,EAAmB,GACnB,EAAa,OAAO,KAAK,GACP,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAChC,IAAI,SAAA,GAAQ,MAAA,GAAM,MAAM,KAC1C,EAAY,EAAM,UAExB,EAAW,QAAQ,SAAA,GACjB,AAAI,EAAU,IAAI,EAAM,OACtB,EAAS,KAAK,KAGlB,EAAM,QAAQ,QAAQ,SAAA,GACpB,AAAI,EAAU,IAAI,EAAO,OACvB,EAAS,KAAK,KAGlB,AAAI,GAAa,MACf,EAAU,QAAQ,SAAA,GAChB,AAAI,EAAU,IAAI,EAAK,OACrB,EAAS,KAAK,KAMpB,OAFM,GAAO,GAAI,KACX,EAAuB,GACtB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MACtB,EAAK,IAAI,EAAK,MACd,AAAK,EAAU,EAAK,OAClB,EAAa,KAAK,GAEpB,EAAK,SAAS,QAAQ,SAAA,GACpB,AAAI,CAAC,EAAK,IAAI,EAAM,OAAS,EAAU,IAAI,EAAM,OAC7C,EAAM,OAAO,MAAM,SAAA,GAAS,MAAA,GAAK,IAAI,EAAM,SAC7C,EAAS,KAAK,KAIpB,MAAO,GAGT,GAAM,IAAmB,CACvB,SAAU,QAAS,QAAS,OAAQ,gBAAiB,cACrD,iBAAkB,KAAM,SAEpB,GAAoB,CACxB,sBAAuB,sBAAuB,sBAAuB,qBAGzC,GAC5B,MAAO,IAAiB,QAAQ,EAAK,KAAO,cAGf,GAC7B,MAAO,IAAkB,QAAQ,EAAK,KAAO,ECjK/C,AAkCA,GAAA,IAAA,WAsFE,WAAoB,EAAsB,GAA1C,GAAA,GAAA,KAAoB,KAAA,MAAA,EAAsB,KAAA,OAAA,EArFlC,KAAA,YAAmC,GAAI,KACvC,KAAA,WAA8B,GAM9B,KAAA,UAAY,IACZ,KAAA,WAAqC,GACrC,KAAA,qBAA0D,GA6EhE,KAAK,SAAW,EAAM,QACtB,KAAK,QAAU,EAAM,OACrB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UAExB,AAAI,EAAM,WAAa,MACrB,OAAO,KAAK,EAAM,WAAW,QAAQ,SAAA,GACnC,EAAK,qBAAqB,GACtB,GAAI,GAAc,EAAM,UAAU,GAAO,KApFnD,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,4CAGpD,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,oBACZ,KAAK,sDAG5B,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,gBAGpD,SAAc,GACZ,GAAM,GAAY,OAAO,KAAK,GAAW,IACrC,SAAA,GAAO,MAAA,GAAU,GAAK,IAAI,SAAA,GAAU,MAAA,GAAO,OAC/C,KAAK,WAAa,GAAG,OAAM,MAAT,GAAa,GAC/B,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GACtB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,IAAI,SAAA,GACvB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,cAAgB,EAAK,wCAG5D,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,IAAI,SAAC,GACxB,GAAM,GAAO,EAAK,cAAgB,EAAK,KACvC,MAAO,GAAK,cAAoB,EAAI,IAAI,EAAK,cAAmB,qCAIpE,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAAA,GAAA,GAAA,KACE,MAAO,QAAO,KAAK,KAAK,YAAY,OAAO,SAAC,EAAK,GAC/C,SAAI,GAAO,EAAK,WAAW,GAAK,UACzB,GACN,qCA0BG,EAAA,UAAA,kBAAR,SAA0B,EAAgB,GACxC,GAAM,GAAe,EAAO,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OAC7C,EAAgB,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OACrD,MAAO,GAAa,KAAK,KAAK,WAAa,KACvC,EAAc,KAAK,KAAK,YAOtB,EAAA,UAAA,QAAR,SAAgB,EAAwB,GACtC,GAAM,GACF,GAAqB,EAAQ,EAAS,KAAK,UAAW,KAAK,YACxD,EAAA,EAAA,cAAe,EAAA,EAAA,YAAa,EAAA,EAAA,WACnC,GAAI,GAAe,KACjB,KAAM,IAAI,OACN,qCAAqC,EAAY,KAAI,gBACrD,oBAAmB,EAAY,GAAE,kBACjC,6DACA,qCAAoC,EAAU,MAGpD,GAAI,EAAc,OAAS,GACzB,GAAM,GAAW,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,OAC9B,EAAU,OAAO,KAAK,GAC5B,KAAM,IAAI,OACN,+BAA+B,EAAQ,8BACvC,KAAI,EAAO,qCAAqC,EAAa,MAGnE,MAAO,IACH,KAAK,MAAO,KAAK,UAAW,IAYlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAAhC,GAAA,GAAA,KACE,EAAS,KAAK,UAAU,GACxB,GAAM,GAAQ,OAAO,KAAK,GAAQ,OAClC,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,GAClB,GAAM,GACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC5D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAG/D,AAAI,EAAY,SAAW,GACzB,GAAc,KAAK,UAGrB,GAAM,GAAiB,KAAK,kBAAkB,EAAY,GAGtD,EAAe,KAAK,YAAY,IAAI,GACxC,AAAI,GAAgB,MAClB,GAAe,KAAK,QAAQ,EAAQ,GACpC,KAAK,YAAY,IAAI,EAAgB,IAGvC,GAAM,GAAiC,GACjC,EAA+B,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,GAAI,IAChB,EAAK,UAAW,EAAgB,EAChC,EAAK,qBACH,EAAU,GAAA,GAAwB,EAAK,WAE7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAKzB,OAFM,GAAgB,EAAK,mBAAmB,GACxC,EAA2D,GACxD,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAO,EAAa,GAC1B,GAAI,CAAC,EAAW,EAAK,OACnB,GAAM,GAAU,GAAU,EAAM,EAAY,GAC5C,GAAI,YAAmB,SACrB,KAAM,IAAI,OACN,4BAA4B,EAAK,GAAE,kEAGzC,EAAW,EAAK,MAAQ,EACxB,EAAK,uBACD,EAAK,KAAM,EAAM,EAAY,EAAS,EACtC,EAAiB,IAIzB,MAAI,GAAK,QAAU,MACjB,EAAQ,QAAQ,GAEX,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAY,QAInD,EAAA,UAAA,mBAAR,SAA2B,GACzB,GAAM,GAAM,GAAG,OAAO,MAClB,GACA,OAAO,KAAK,GACP,IAAI,SAAA,GAAO,MAAA,GAAU,KACrB,IAAI,SAAA,GAAW,MAAA,GAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,QACrD,MAAO,IAAI,KAAI,IAET,EAAA,UAAA,uBAAR,SACI,EAAkB,EAAY,EAC9B,EAA2B,EAC3B,EACA,GAGF,GAAI,EAAK,WAAa,WAAa,EAAY,QAAQ,KAAc,GACnE,OAGF,EAAU,GAAU,QAAQ,SAAA,GAC1B,AAAI,GAAU,MACZ,GAAgC,EAAO,IAClC,GAAgC,EAAO,KAAO,GAC/C,EAAK,SAAS,UAGtB,EAAK,OAAO,QAAQ,SAAA,GAGlB,GAAI,EAAM,WAAa,WACrB,GAAM,GACF,GAA6B,EAAM,KAAM,EAAW,GACxD,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,GAAU,CAAC,EAAc,IAAI,EAAO,KACtC,GAAM,GAAQ,EAAgC,EAAO,IACrD,AAAI,IAAU,EACZ,GAAO,UACP,MAAO,GAAgC,EAAO,KACzC,AAAI,GAAS,MAGlB,EAAgC,EAAO,aAkB/C,EAAA,UAAA,aAAN,SAAmB,EAAwB,sEAEzC,MAAA,CAAA,EAAO,KAAK,cAAc,EAAQ,SAiBtB,EAAA,UAAA,cAAd,SACI,EAAwB,EAAmB,EAC3C,EACA,GAF2C,MAAA,KAAA,QAAA,GAAA,IAC3C,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,uGACF,MAAK,IACH,GAAS,KAAK,UAAU,GACxB,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,IAGd,EAAU,GAAI,IAChB,KAAK,UAAW,EAAgB,EAChC,KAAK,qBAKS,CAAA,EAAM,KAAK,uBACzB,EAAQ,EAAS,EAAS,WADxB,SAAY,EAAA,OAEZ,EAAU,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAGzD,EAAY,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAW,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,GAAO,GAAM,KACxD,EACF,GAAI,KAAgB,EAAS,OAAK,EAAa,KAAK,YACxD,OAAO,KAAK,GAAW,QAAQ,SAAA,GAC7B,GAAM,GAAc,EAAU,GAC9B,EAAY,QAAQ,SAAA,GAClB,AAAI,GAAU,CAAC,EAAO,YAAc,CAAC,EAAQ,IAAI,EAAO,KACtD,EAAO,cAKb,AAAI,KAAK,QAAU,MACjB,EAAQ,QAAQ,GAGlB,CAAA,EAAO,SAGH,EAAA,UAAA,qBAAN,SACI,EAAkB,EAClB,mFACI,SAAe,EAAO,OAAO,SAAC,EAAK,EAAQ,GAC/C,SAAI,EAAK,OAAO,GAAO,MAAQ,EACxB,GACN,IAEH,CAAA,EAAO,KAAK,cACR,EAAc,KAAK,YAAa,GAAM,EAAgB,SAS9C,EAAA,UAAA,uBAAd,SACI,EAAwB,EAA2B,EACnD,0IACI,EAAQ,OAAO,KAAK,GACpB,EACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAY,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC9D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAC3D,EACF,GAAqB,EAAQ,EAAa,KAAK,WAD5C,EAAS,EAAA,UAAE,EAAa,EAAA,cAAE,EAAW,EAAA,YAAE,EAAU,EAAA,WAGlD,EACE,EAAU,OAAK,KAAK,MAAM,SAAS,IAAI,SAAA,GACzC,MAAO,CAAC,KAAI,EAAE,SAAU,EAAQ,kBAEhC,EAAU,GAAA,GAAwB,KAAK,WAC7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAEnB,EAA2D,GAC3D,EAAgB,KAAK,mBAAmB,GACxC,EAAkC,0BACjC,GAAM,OAAS,EACd,GAAW,KAAK,aAClB,EAAY,EAAO,EAAS,EAAY,EAAO,EAC/C,EAAiB,EAAiC,GACtD,CAAA,EAAM,QAAQ,IAAI,KAJG,CAAA,EAAA,UAIrB,SAAA,oBAaF,GAXA,AAAI,GAAe,MAAQ,CAAC,GAC1B,QAAQ,KACJ,mIAGA,EACF,EACK,OACG,SAAA,GAAQ,MAAA,CAAC,GAAc,IACnB,CAAC,GAAU,EAAK,KAAM,EAAY,KACzC,IAAI,SAAA,GAAQ,MAAA,GAAK,OACtB,EAAe,OAAS,EACtB,QAAiB,GACrB,AAAI,GAAe,MACjB,GACI,gEACA,4BAA2B,EAAU,MAErC,GAAI,OACN,+BAA+B,EAAc,uBAC7C,YAAW,EAAK,gDAChB,KAAI,EAAa,MAAM,IAE7B,MAAA,CAAA,EAAO,SAGD,EAAA,UAAA,aAAR,SACI,EAAoB,EAA2B,EAC/C,EAA4B,EAC5B,EAA4B,EAC5B,EACA,GAEF,OAPF,GAAA,KAMQ,EAAqC,gBAEzC,GAAM,GAAO,EAAM,MACnB,EAAQ,eAAiB,EAAK,SAC9B,GAAI,GAAW,GAUf,GANA,AAAI,EAAK,KAAK,KAAO,SACjB,EAAc,aAAc,EAAK,KAAM,EAAW,IACnD,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAIC,EAAW,QAAQ,EAAK,QAAU,IACpC,GAAM,GAAU,GAAU,EAAK,KAAM,EAAW,GAChD,AAAK,GACF,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAEH,GAAM,GAAiB,EAAQ,eAC/B,AAAI,YAAmB,SACrB,EAAS,KAAK,EAAQ,KAAK,SAAA,GACzB,SAAU,GAAY,EACtB,EAAQ,eAAiB,EACzB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,GAC1C,KAGT,GAAU,GAAY,EACtB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,QAGnD,GAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,WAxC9C,EAAM,OAAS,OA2CtB,MAAO,IAGD,EAAA,UAAA,kBAAR,SACI,EAAY,EAA2B,EACvC,EAA4B,EAC5B,GACF,EAAK,SAAS,QAAQ,SAAC,GACd,GAAA,GAAA,GAAA,EAAA,KAAA,GAAA,GACP,GAAI,EAAM,IAAa,CAAC,EAAU,IAAI,EAAU,MAC9C,OAGF,AAAI,EAAU,KAAO,QACnB,AAAI,EAAU,WAAW,KAAK,SAAA,GACxB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAExC,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,KAGpD,AAAI,EAAU,WAAW,MAAM,SAAA,GACzB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAE5C,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,QAQ1D,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,OAAO,KAAK,KAAK,WACZ,QACG,SAAA,GAAO,MAAA,GAAK,UAAU,GAAK,QAAQ,SAAA,GAAU,MAAA,GAAO,eAGtD,EAAA,UAAA,uBAAR,SAA+B,GAA/B,GAAA,GAAA,KACE,OAAO,KAAK,GAAQ,QAAQ,SAAA,GAC1B,GAAM,GAAQ,EAAO,GACd,EAAA,GAAA,GAAA,GACD,EAAO,EAAK,MAAM,MAAM,GAC9B,GAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,GAAM,GAAQ,EAAK,WAAW,MAAS,MACjC,EAAQ,EAAM,SAAW,EAAM,MAAM,QACvC,EAAM,MAAM,MACR,SAAC,EAAK,GAAU,MAAA,GAAM,KAAW,IAAM,EAAM,KAAW,IAChE,EAAA,KAAK,OACD,EACA,WAAM,MAAA,sBAAsB,EAAK,KAAI,kBACjC,iCAAgC,EAAK,eACrC,KAAI,EAAM,MAAK,OAEzB,AAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,EAAA,KAAK,OACD,EAAM,QAAU,EAAK,WAAW,MAAS,MACzC,WAAM,MAAA,sBAAsB,EAAK,KAAI,8CAE9B,GAAK,WAAW,MAAS,MAAK,aAAa,EAAM,YAK1D,EAAA,UAAA,UAAR,SAAkB,GAChB,GAAM,GAAyB,GAC/B,OAAW,KAAa,GACtB,GAAI,KAAK,YAAc,MAAQ,KAAK,WAAW,QAAU,MACrD,KAAK,WAAW,OAAO,IAAc,MACvC,GAAM,GAAS,KAAK,WAAW,OAAO,GACtC,EAAO,EAAO,MAAQ,EAAO,OAE7B,GAAO,GAAa,EAAO,GAG/B,MAAO,IAGD,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAa,OAAO,KAAK,GAAQ,OAAO,SAAA,GACrC,GAAA,GAAA,GAAA,GAAA,GACP,MAAO,GAAK,MAAM,MAAM,IAAa,OAEvC,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,OACN,gDACA,WAAU,EAAU,kCAIpB,EAAA,UAAA,WAAR,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GACjB,GAAI,EAAK,YAAc,MAAQ,EAAK,WAAW,SAAW,MACtD,EAAK,WAAW,QAAQ,IAAS,MACnC,GAAM,GAAS,EAAK,WAAW,QAAQ,GACvC,MAAO,GAAO,KAEhB,MAAO,IACN,KAGG,EAAA,UAAA,aAAR,SAAqB,GAArB,GAAA,GAAA,KACE,EAAQ,QAAQ,SAAA,GACP,GAAA,GAAA,GAAA,GAAA,GACP,GAAI,CAAC,EAAK,MAAM,MAAM,GACpB,KAAM,IAAI,OAAM,eAAe,EAAI,kCAI3C,KC3lBA,AAyBO,GAAM,IAAqB,oBACrB,GAAqB,2BAoDhC,WACY,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IADA,KAAA,SAAA,EACA,KAAA,YAAA,EAzCJ,KAAA,QAAU,MA0ChB,AAAI,GAAe,MACjB,MAAK,YAAc,IArCvB,cAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,yCAGd,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,SAAS,4CAGvB,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,6CAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,SAAS,wCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,yCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,2CAoBf,EAAA,UAAA,cAAR,WACE,GAAM,GAAO,KAAK,SAClB,GAAK,EAAsB,MAAQ,KAEjC,KAAK,QAAU,UACN,KAAK,YAAY,aAAe,KACzC,KAAK,QAAU,EAAA,GAAG,mBAAmB,EAAgB,KAAK,kBAE1D,GAAM,GAAW,EAAA,GAAG,gBAAgB,EAAgB,KAAK,aACzD,GAAI,EAAS,SAAW,EAGtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAgB,KAAK,sBAChD,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,CAAC,GAAK,MAEpB,KAAK,QAAU,EAAS,KAQtB,EAAA,UAAA,KAAN,2GAEE,GADA,KAAK,gBACD,KAAK,QAAQ,MAAQ,KACvB,KAAM,IAAI,OACN,iHAGY,MAAA,CAAA,EAAM,KAAK,QAAQ,eAA/B,SAAY,EAAA,OAElB,CAAA,EAAO,KAAK,SAAS,UASvB,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,UAAY,EACjB,GAAM,GAAQ,KAAK,UAAU,cACzB,EAAY,GAChB,AAAI,KAAK,UAAU,qBAAuB,MACxC,GACK,KAAK,UAAU,oBAA4B,WAIlD,KAAK,QAAa,EAAM,SAAS,SAAQ,IAAI,EAAM,SAAS,YAC5D,GAAM,GACF,EAAA,GAAG,cAAc,KAAK,UAAU,WAAY,KAAK,UAAU,aAK/D,GAJA,KAAK,SAAW,GAAI,IAChB,GAAgB,SAAS,eAAe,EAAO,IACnD,KAAK,SAAS,UAAY,KAAK,6BAA6B,GAExD,EAAU,kBAAoB,MAChC,GAAM,GACF,GAAgB,SAAS,eAAe,EAAU,kBACtD,KAAK,YAAc,GAAI,IAAc,GACrC,KAAK,YAAY,UAAY,KAAK,SAAS,UAC3C,KAAK,YAAY,QAAQ,GAAI,IAG/B,MAAO,IA+CH,EAAA,UAAA,KAAN,SAAW,EAAmC,4EAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,OACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,OACN,+GAIN,MAAA,CAAA,EAAO,EAAa,KAAK,KAAK,iBAyChC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,MAAO,MAAK,QAAQ,EAAQ,KAAK,cAG3B,EAAA,UAAA,gBAAR,SAAwB,GAEtB,GAAI,CAAE,aAAkB,GAAA,SAAW,CAAC,MAAM,QAAQ,GAEhD,MAAO,GAGT,GADA,EAAS,MAAM,QAAQ,GAAU,EAAS,CAAC,GACvC,EAAO,SAAW,KAAK,WAAW,OACpC,KAAM,IAAI,OACN,+BACA,wBAAuB,KAAK,WAAW,OAAM,mBAC7C,oBAAmB,EAAO,OAAM,oBAEtC,MAAO,MAAK,WAAW,OAAO,SAAC,EAAK,EAAW,GAC7C,SAAI,GAAc,EAAoB,GAC/B,GACN,KAGG,EAAA,UAAA,iBAAR,SAAyB,GACvB,SAAU,GAAW,KAAK,YACnB,AAAC,MAAM,QAAQ,GAAuB,EAAZ,CAAC,IAmBpC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,EAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GAChC,GAAM,GAAS,KAAK,SAAS,QAAQ,EAAQ,GAC7C,MAAO,GAAO,OAAS,EAAI,EAAS,EAAO,IAkBvC,EAAA,UAAA,aAAN,SACI,EACA,mGACF,SAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GACjB,CAAA,EAAM,KAAK,SAAS,aAAa,EAAQ,WAAlD,SAAS,EAAA,OACf,CAAA,EAAO,EAAO,OAAS,EAAI,EAAS,EAAO,UAGrC,EAAA,UAAA,6BAAR,SAAqC,GACnC,MAAO,QAAO,KAAK,GAAK,OAAO,SAAC,EAAyB,GACvD,SAAO,GAAO,CAAC,EAAI,IACZ,GACN,KAQL,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UAEd,AAAI,KAAK,aACP,KAAK,YAAY,WAGvB,iBAiCI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,6FACF,GAAI,GAAY,KACd,KAAM,IAAI,OACN,0GAGN,MAAI,IAAW,MACb,GAAU,IAGZ,AAAI,EAAQ,WACV,CAAK,EAA0B,MAAQ,MACrC,CAAM,EAAoB,SAAS,MACjC,GAAY,EAAsB,KAEpC,EAAW,GAAG,EAAW,GAAqB,KAG5C,EAAQ,GAAI,IAAW,EAAU,GACvC,CAAA,EAAM,EAAM,eAAZ,SAAA,OACA,CAAA,EAAO,QC7ZT,GAGM,IAAU,4GCHhB,o8DC2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAEZ,GAAQ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,YAkDwB,EAAY,GAElC,MAAO,IAAgB,EAAO,GAShC,YACI,EAAY,EACZ,EAAiC,GAEnC,GAFE,AAAA,IAAA,QAAA,GAAA,GAA0B,MAAO,AAAA,IAAA,QAAA,GAAA,GAA2B,MAE1D,GAAS,KACX,MAAO,MAET,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAI,EAAK,IAAI,GACX,MAAO,GAAK,IAAI,GAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAGL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,GAChB,OAAW,KAAK,IACd,GAAM,GAAQ,EAAM,GACd,EAAc,GAAgB,EAAO,EAAO,EAAM,GACxD,EAAe,GAAK,EAEtB,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAdzD,UAAK,IAAI,EAAO,EAAO,OAChB,EAAO,kBA2Cd,EAAe,GAAA,MAAA,KAAA,QAAA,GAAA,IACV,GAAgB,EAAQ,GAOjC,YACI,EAAe,EACf,GAAA,AAAA,IAAA,QAAA,GAAA,GAA2B,MAG7B,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAEL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,kBACL,GACT,GAAM,GAAW,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC7B,EAAc,GAAgB,EAAU,EAAO,GACrD,EAAe,GAAK,GAHtB,OAAW,KAAK,KAAL,GAKX,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAbzD,OAAO,GAAO,kBAkBQ,GACxB,MAAI,KAAM,KACD,KAIL,GAAW,EAAE,IACR,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAG,QAAS,gBAsC3B,EAAY,+GACR,EAA+B,GAAI,KAGzC,GAAgB,EAAO,EAAO,OAMZ,EAAA,MAAM,KAAK,EAAK,+BAAhB,GAAA,EAAA,OAAP,GAAG,EAAA,GACN,EAAQ,EAAK,IAAI,GACnB,YAAiB,SACC,CAAA,EAAM,GADxB,CAAA,EAAA,IAFmC,CAAA,EAAA,UAG/B,EAAc,EAAA,OACpB,EAAK,IAAI,EAAK,oBAJA,wBAWZ,SAAS,GAAgB,EAAO,EAAO,GAC7C,CAAA,EAAO,oBASkB,GACzB,MAAO,IAAO,MAAS,CAAC,YAAY,OAAO,IACtC,OAAM,QAAQ,IACb,MAAO,IAAQ,UAAY,CAAE,aAAe,IAAA,qBAYvB,GAC3B,MAAO,IAAO,MAAQ,GAAY,IAAQ,MAAM,QAAQ,IACnD,MAAO,IAAQ,UAAa,YAAe,IAAA,QAC5C,GAAA,KAAQ,aAAa,GAO3B,YAAqB,GACnB,MACI,KAAU,MACT,MAAO,IAAU,UAAY,MAAO,IAAU,WCnRrD,YAqB6B,GAC3B,MAAO,IAAQ,EAAW,IAI5B,YAAuB,GACrB,MAAI,aAAgB,IAAA,OACV,CAAC,MAAO,EAAK,QAAS,QAAS,IAC9B,GAAW,GACb,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAM,QAAS,IChClC,AAqBA,GAAA,IAAA,WAcE,WAAmB,GACjB,GADiB,KAAA,SAAA,EAVT,KAAA,MAAQ,EACR,KAAA,IAAM,EAUV,GAAY,KACd,KAAM,IAAI,YAAW,mDAEvB,GAAI,EAAW,EACb,KAAM,IAAI,YAAW,6CAEvB,KAAK,KAAO,GAAI,OAAS,GACzB,KAAK,gBAAkB,EAAI,EAMnB,SAAA,UAAA,KAAV,SAAe,GAEb,KAAO,EAAQ,GACb,GAAS,KAAK,gBAEhB,MAAO,GAAQ,KAAK,iBAGZ,EAAA,UAAA,IAAV,SAAc,GACZ,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,MAAO,MAAK,KAAK,EAAQ,KAAK,WAGtB,EAAA,UAAA,IAAV,SAAc,EAAe,GAC3B,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,KAAK,KAAK,EAAQ,KAAK,UAAY,GAMrC,EAAA,UAAA,OAAA,WACE,GAAI,GAAS,KAAK,IAAM,KAAK,MAC7B,MAAI,GAAS,GACX,GAAS,KAAK,gBAAkB,GAE3B,GAQT,EAAA,UAAA,OAAA,WACE,MAAO,MAAK,WAAa,KAAK,UAQhC,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,WAAa,GAM3B,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,IAAI,KAAK,IAAK,GACnB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,IAMlC,EAAA,UAAA,QAAA,SAAQ,GACN,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,KAAK,KAOd,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,GAChC,GAAM,GAAS,KAAK,IAAI,KAAK,KAC7B,YAAK,IAAI,KAAK,IAAK,QACZ,GAMT,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GACpC,KAAK,IAAI,KAAK,MAAO,IAMvB,EAAA,UAAA,MAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAS,KAAK,IAAI,KAAK,OAC7B,YAAK,IAAI,KAAK,MAAO,QACrB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GAC7B,GAYT,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAQ,KAAK,KAAK,KAAK,MAAQ,GAC/B,EAAS,KAAK,IAAI,GACxB,YAAK,IAAI,EAAO,KAAK,OACd,GAEX,KC/KA,AAoBA,GAAA,IAAA,SAAA,GAA0C,GAAA,EAAA,GAMxC,mBACE,GAAA,KAAA,KAAM,EAAkB,mBAAiB,KAG3C,SAAA,UAAA,OAAA,WACE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,GACH,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,KAAI,KAAA,KAAC,IAGb,EAAA,UAAA,QAAA,SAAQ,GACN,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,QAAO,KAAA,KAAC,IAMR,EAAA,UAAA,OAAR,WAOE,OANM,GAAc,KAAK,SAAW,EAC9B,EAAU,GAAI,OAAS,GACvB,EAAM,KAAK,SAIR,EAAI,EAAG,EAAI,EAAK,IACvB,EAAQ,GAAK,KAAK,IAAI,KAAK,KAAK,KAAK,MAAQ,IAG/C,KAAK,KAAO,EACZ,KAAK,SAAW,EAChB,KAAK,gBAAkB,EAAI,KAAK,SAChC,KAAK,MAAQ,EACb,KAAK,IAAM,GA7CE,EAAA,iBAAmB,GA+CpC,GAhD0C,ICpB1C,YAuCqC,GACnC,MAAO,IAAI,IAAc,eAyBvB,GAEF,MAAO,IAAI,IAAqB,eAgB9B,EACA,GACF,MAAO,IAAI,IAAgB,EAAe,eAmDxC,EACA,GAAA,MAAA,KAAA,QAAA,GAAgC,GAAgB,MAC3C,GAAI,IAAe,EAAW,GAUvC,GAAA,IAAA,WAAA,cAwBQ,SAAA,UAAA,QAAN,6GACQ,SAAc,GACZ,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,KAAK,gBAAf,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAcH,EAAA,UAAA,eAAN,+GACQ,SAAS,KAAK,SAAS,KACvB,EAAc,GACZ,CAAA,EAAM,EAAO,eAAjB,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,EAAO,gBAAjB,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAUH,EAAA,UAAA,aAAN,2GACU,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACR,CAAA,EAAM,KAAK,eAAf,SAAI,EAAA,oCAWF,EAAA,UAAA,aAAN,SAAmB,qGACT,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,8BACzB,CAAC,EAAE,MAAS,EACd,CAAA,EAAM,KAAK,QADiB,CAAA,EAAA,UAChC,SAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,oCAgBjC,EAAA,UAAA,aAAA,SAAa,GACX,MAAO,IAAI,IAA0B,KAAM,IAa7C,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,IAAI,IAAe,KAAM,IAWlC,EAAA,UAAA,IAAA,SAAO,GACL,MAAO,IAAI,IAAY,KAAM,IAW/B,EAAA,UAAA,SAAA,SAAY,GACV,MAAO,IAAI,IAAiB,KAAM,IAWpC,EAAA,UAAA,eAAA,SAAkB,GAChB,MAAO,IAAI,IAAiB,KAAM,GAAW,UAW/C,EAAA,UAAA,QAAA,SAAW,GACT,MAAO,IAAI,IAAgB,KAAM,IAQ7B,EAAA,UAAA,aAAN,SAAmB,sEACjB,MAAA,CAAA,EAAO,KAAK,IAAI,GAAG,qBAUf,EAAA,UAAA,cAAN,SAAoB,sEAClB,MAAA,CAAA,EAAO,KAAK,eAAe,GAAG,aAAa,SAAA,GAAK,MAAC,KAAM,WAqBzD,EAAA,UAAA,cAAA,SAAc,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,IACxB,GAAI,IAAsB,KAAM,EAAW,IAmCpD,EAAA,UAAA,iBAAA,SACI,EAAmB,EAEnB,GAFmB,AAAA,IAAA,QAAA,GAAA,IAEnB,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAAa,KAAK,cAAc,EAAW,GAGjD,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,MAaxC,EAAA,UAAA,YAAA,SACI,EACA,GACF,MAAO,IAAI,IACP,GAAkB,CAAC,KAAM,IAAY,IAU3C,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAShC,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAYhC,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,IAAI,IAAiB,KAAM,IAapC,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAC1B,MAAO,IAAI,IAAgB,KAAM,EAAY,IAO/C,EAAA,UAAA,OAAA,WACE,MAAO,IAAI,IAAe,OAE9B,KAWA,GAAA,SAAA,GAA+B,GAAA,EAAA,GAE7B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,MAAA,EADd,EAAA,KAAO,IAKf,SAAA,UAAA,QAAA,WACE,MAAO,YAAY,KAAK,MAAM,OAAM,UAGhC,EAAA,UAAA,KAAN,oFACE,MAAI,MAAK,MAAQ,KAAK,MAAM,OAC1B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAO,KAAK,MAAM,KAAK,MAC7B,KAAK,OACL,CAAA,EAAO,CAAC,MAAO,GAAU,GAAO,KAAM,WAE1C,GAlB+B,IAoB/B,GAAA,SAAA,GAAsC,GAAA,EAAA,GACpC,WACc,GADd,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,OAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAO,iBAGH,EAAA,UAAA,KAAN,8EACE,IACE,MAAA,CAAA,EAAO,KAAK,gBACL,GAEP,QAAE,QACE,mDAAmD,EAAE,QACnD,iBAGZ,GApBsC,IAsBtC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,8EACE,MAAA,CAAA,EAAO,KAAK,SAAS,aAEzB,GA1BgC,IA4BhC,GAAA,SAAA,GAA8B,GAAA,EAAA,GAQ5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAF3D,EAAA,MAAQ,EAIN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,iHAKS,MAAK,QAAU,KAAK,SACT,CAAA,EAAM,KAAK,SAAS,QADH,CAAA,EAAA,UAGjC,MAFM,GAAU,EAAA,OAEZ,EAAQ,KACV,CAAA,EAAO,GAET,IAAA,QAAW,EAAQ,qBAErB,MAAA,CAAA,EAAO,KAAK,SAAS,cAEzB,GAzC8B,IA2C9B,GAAA,SAAA,GAA8B,GAAA,EAAA,GAE5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAD3D,EAAA,MAAQ,IAKR,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,8EACE,MAAI,MAAK,SAAW,KAAK,SACvB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,CAAA,EAAO,KAAK,SAAS,aAEzB,GAhB8B,IAqB9B,GAAA,SAAA,GAAuC,GAAA,EAAA,GAKrC,WACc,EAAqC,EACrC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EAAqC,EAAA,UAAA,EACrC,EAAA,qBAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,qBAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,6GACQ,EAAa,0BACZ,GAAM,OAAS,KAAK,UACZ,CAAA,EAAM,KAAK,SAAS,QADC,CAAA,EAAA,UAElC,MADM,GAAO,EAAA,OACT,EAAK,KACH,KAAK,sBAAwB,EAAM,OAAS,EAC9C,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,KAE9B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAM,KAAK,EAAK,qBAElB,MAAA,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,WAEhC,GAvCuC,IAyCvC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,2GAEiB,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,MAAQ,KAAK,UAAU,EAAK,OACnC,CAAA,EAAO,GAET,IAAA,QAAW,EAAK,qCAGtB,GAlCgC,IAoChC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAC9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,WAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAc7B,IAZM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAOzD,EAAS,KAAK,UAAU,EAAK,OAC7B,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCgC,IAqChC,GAAA,SAAA,GAA2C,GAAA,EAAA,GAEzC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,QAAA,EAHd,EAAA,MAAQ,EAKN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,oBAO7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGR,EAAA,UAAA,WAAN,yJAGa,CAAA,EAAM,KAAK,SAAS,eAA3B,MAAA,CAAA,EAAO,EAAA,eAEP,kBAAI,AAAC,KAAK,QAAQ,SAChB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,+CAUrC,GA1C2C,IA4C3C,GAAA,SAAA,GAAqC,GAAA,EAAA,GACnC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,KACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAe,GAAA,YAAe,sBAAsB,EAAK,OAOhD,CAAA,EAAM,KAAK,UAAU,EAAK,gBAKzC,IALM,EAAS,EAAA,OACT,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCqC,IAgDrC,GAAA,SAAA,GAAmD,GAAA,EAAA,GAQjD,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,YAAc,GAAI,IACvB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAiBR,EAAA,UAAA,WAAN,2GAIS,MAAK,YAAY,WAAa,EAE9B,CAAA,EAAM,KAAK,QAFoB,CAAA,EAAA,UAEpC,MAAK,GAAA,aACH,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,YAG/B,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,YAAY,QAAS,KAAM,WAEnD,GAjDmD,IAkDnD,GAAA,SAAA,GAAoC,GAAA,EAAA,GAClC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,eAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,IAeT,IAbM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAMzD,EAAc,KAAK,UAAU,EAAK,OAClC,EACF,GAAA,YAAe,sBAAsB,GACzC,KAAK,YAAY,QAAQ,GAIzB,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAIN,MAAA,CAAA,EAAO,UAEX,GArCoC,IAgDpC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAStC,WACI,EACiB,GAFrB,GAAA,GAGE,EAAA,KAAA,OAAO,KADY,SAAA,iBAAA,EARb,EAAA,SAAuC,KAGvC,EAAA,SAA4B,KAOlC,EAAK,cAAgB,IAGvB,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,8CAC1B,MAAU,GAAiB,eAGvB,EAAA,UAAA,KAAN,8EACE,YAAK,SAAW,KAAK,cAAc,KAAK,UACxC,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,cAAd,SAA4B,qGAO1B,MAAA,CAAA,EAAM,gBAAN,GAAA,OACI,KAAK,UAAY,KACI,CAAA,EAAM,KAAK,cAAc,QAD9C,CAAA,EAAA,UAEF,GADM,EAAiB,EAAA,OACnB,EAAe,KAEjB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,KAAK,SAAW,EAAe,MAC/B,AAAI,KAAK,kBAAoB,MAC3B,MAAK,SAAW,KAAK,SAAS,aAAa,KAAK,oCAGjC,MAAA,CAAA,EAAM,KAAK,SAAS,eACvC,MADM,GAAa,EAAA,OACf,EAAW,KACb,MAAK,SAAW,KAChB,CAAA,EAAO,KAAK,cAAc,KAE5B,CAAA,EAAO,SAEX,GApDwC,IAsD5B,GAAZ,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,YAHU,IAAA,IAAe,KAmC3B,GAAA,IAAA,SAAA,GAAwD,GAAA,EAAA,GAItD,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAgC,GAAgB,MAFvE,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,UAAA,EACA,EAAA,aAAA,EALf,EAAA,MAAQ,EACR,EAAA,eAA6C,OAQrD,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,0CAC1B,MAAO,IAAI,EAAiB,YAGhB,EAAA,UAAA,UAAd,SAAwB,2CAWtB,WAAiB,GACf,GAAI,YAAqB,KACvB,GAAM,GAAS,EAAU,OACzB,MAAO,CACL,MAAO,EAAO,KAAK,SAAA,GACjB,WACA,AAAI,EAAE,MACJ,IAEK,EAAE,QAEX,QAAS,QAGX,OAAO,CAAC,MAAO,KAAM,QAAS,gEArBlC,MAAA,CAAA,EAAM,UAAN,SAAA,OAII,EAAe,EACf,EAAgB,EAoBF,CAAA,EAAM,GAAmB,KAAK,UAAW,WAE3D,GAFM,EAAY,EAAA,OAEd,IAAiB,EAEnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAI,EAAgB,EAClB,OAAQ,KAAK,kBACN,IAAgB,KACnB,KAAM,IAAI,OACN,+CACA,0BAAyB,KAAK,MAAK,UACpC,IAAgB,SACnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,SACxB,IAAgB,SAMzB,YAAK,QACL,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAGzB,EAAA,UAAA,KAAN,8EACE,YAAK,eAAiB,KAAK,UAAU,KAAK,gBAC1C,CAAA,EAAO,KAAK,qBAEhB,GAxEwD,IAoFxD,GAAA,SAAA,GAAyC,GAAA,EAAA,GAGvC,WACc,EAAqC,GADnD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAAqC,EAAA,WAAA,EAEjD,EAAK,OAAS,GAAI,IAAuC,KAG3D,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAOzB,EAAA,UAAA,OAAV,WACE,KAAO,CAAC,KAAK,OAAO,WAClB,GAAM,GAAI,KAAK,SAAS,OACxB,KAAK,OAAO,KAAK,KAIrB,EAAA,UAAA,KAAA,WACE,YAAK,SAIE,KAAK,OAAO,SAEvB,GA/ByC,IAuCzC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAUtC,WACc,EAAqC,EAC/C,GAFJ,GAAA,GAGE,EAAA,KAAA,KAAM,EAAU,IAAW,KAFf,SAAA,SAAA,EAAqC,EAAA,WAAA,EAH3C,EAAA,kBAAoB,GAM1B,EAAK,OAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACpD,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGN,EAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,MAAK,MAAM,KAAK,SAAW,IAG1B,EAAA,UAAA,YAAV,WACE,MAAO,MAAK,UAAU,KAAK,OAAO,WAG9B,EAAA,UAAA,WAAN,6GAEE,AAAK,KAAK,mBACR,KAAK,gCAEC,MAAK,OAAO,UAAS,CAAA,EAAA,GACrB,GAAc,KAAK,cACV,CAAA,EAAM,KAAK,OAAO,cAAc,YAC/C,GADM,EAAS,EAAA,OACX,EAAO,KACT,KAAK,kBAAoB,OAEzB,aAAK,SACL,CAAA,EAAO,sBAGX,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,WAE/B,GApDwC,ICjnCxC,kBA6DA,aAWW,KAAA,KAAe,KA+DxB,SAAA,UAAA,MAAA,SAAM,EAAmB,GAAzB,GAAA,GAAA,KAAyB,AAAA,IAAA,QAAA,GAAA,IACvB,GAAM,GAAO,KACb,GAAA,KAAQ,OACJ,EAAY,EAAG,WAAM,MAAA;QACrB,IACJ,GAAI,GACJ,MAAI,MAAK,OAAS,UAAY,KAAK,MAAQ,KAGzC,EAAO,KAAK,KACP,AAAI,EAGT,EAAO,KAAK,KAAK,KAAK,KAAO,GAI7B,EAAO,KAAK,MAAM,KAAK,KAAO,GAEzB,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OACH,iBAAiB,EAAW,EAAgB,WAChD,IAkBL,EAAA,UAAA,YAAA,SAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,UAAY,EAAQ,OAAS,SAG7C,EAAO,SACF,AAAI,KAAK,MAAQ,MAAQ,EAAQ,MAAQ,KAG9C,EAAO,KAAK,KAAO,EAAQ,KAI3B,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,EAAA,MAAA,IAAA,KAAA,SAAA,0BACK,MAAA,CAAA,EAAM,EAAK,mBAAZ,SAAA,GAAC,EAAA,QAAuB,YAAY,CAAA,EAAM,EAAQ,mBAAlD,MAAA,CAAA,EAAA,EAAA,MAAA,EAAA,CAAoC,EAAA,gBACxC,IAmBN,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,SAEhB,EAAO,SAIP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,OAAO,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAClE,IAmBC,EAAA,UAAA,aAAN,SAAmB,6FACT,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,aAAa,UAkB9C,EAAA,UAAA,IAAA,SAAkC,GAAlC,GAAA,GAAA,KACQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,IAAI,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAC/D,KAAK,OA0BV,EAAA,UAAA,SAAA,SAAuC,GAAvC,GAAA,GAAA,KAEQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,SAAS,UACvC,KAAK,OAYV,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,GAAI,GAAc,KAChB,KAAM,IAAI,YACN,6DAGN,GAAM,GAAO,KACb,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,SAAS,UAAa,KAAK,OAqBrE,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,EAAQ,EAI/B,EAAO,KAAK,KAAO,EACd,AAAI,IAAU,EAEnB,EAAO,EACF,AAAI,KAAK,MAAQ,MAAS,KAAU,QAAa,EAAQ,GAG9D,EAAO,SAGP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mDACrB,SAAmB,GACrB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,MAAA,IAAA,KAAA,SAAA,sCAAqB,CAAA,EAAM,EAAK,mBAApB,MAAA,CAAA,EAAE,GAAA,MAAO,EAAA,OAAuB,EAAA,KAAM,GAAK,WAC3D,CAAA,EAAO,GAAyB,EAAiB,KAAK,UACrD,IAoBL,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,GAAS,GAAK,KAAK,MAAQ,EAIlD,EAAO,KAAK,KAAO,EACd,AACH,KAAK,MAAQ,MACZ,MAAK,KAAO,GAAS,IAAU,QAAa,EAAQ,GAGvD,EAAO,EAGP,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IA4BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,EAAe,GAA3C,GAAA,GAAA,KAEE,GAFyC,AAAA,IAAA,QAAA,GAAA,IAErC,GAAc,MAAQ,EAAa,EACrC,KAAI,MAAK,MAAQ,KACT,GAAI,YACN,4DAEE,GAAI,YACN,+KAGA,oCAAmC,KAAK,KAAI,eAGpD,GAAM,GAAO,KACP,EAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACrD,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACvB,SAAQ,EAAO,QACnB,AAAI,GACF,IAAS,EAAO,SAEV,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,QAAQ,EAAY,EAAM,mBACxD,KAAK,OAoBV,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,KAAK,KAAO,EAGnC,EAAO,EACF,AAAI,KAAK,MAAQ,MAAQ,KAAK,MAAQ,EAG3C,EAAO,KAAK,KAGZ,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IAmBjD,EAAA,UAAA,QAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,iBAc3B,EAAA,UAAA,eAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,wBA1HjB,EAAA,gBAAkB,IA4HpC,iBAeI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,GAAA,UAAA,GAAkB,GAAA,EAAA,GAAd,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KACT,SAAA,KAAO,IAMD,SAAA,UAAA,SAAN,8EACE,MAAA,CAAA,EAAO,UAEX,GAVyB,iBAgCyB,GAApD,GAAA,GAAA,KACE,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,GAAkB,SAAQ,EAAM,oBA4CA,GAAlD,GAAA,GAAA,KAGE,GAAI,CAAC,GAAW,GACd,KAAM,IAAI,OAAM,qDAElB,GAAI,GACJ,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAO,GAAQ,KAAQ,EAAS,GAAkB,KAC5B,KAAK,IAAI,EAAO,EAAS,GAAkB,cAE1D,YAAoB,QAC7B,OAAW,KAAM,GACf,EAAO,GAAQ,KAAQ,EAAS,GAAmB,KAC7B,KAAK,IAAI,EAAO,EAAS,GAAmB,MAGtE,MAAO,IAAyB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACd,MAAA,CAAA,EAAM,GAAmB,EAAU,SAAA,GACjD,GAAI,YAAa,IACf,MAAO,CAAC,MAAO,EAAE,WAAY,QAAS,IACjC,GAAI,GAAW,GACpB,MAAO,CAAC,MAAO,KAAM,QAAS,IAE9B,KAAM,IAAI,OACN,uFAPF,SAAU,EAAA,OAWhB,CAAA,EAAO,GAAsB,EAAS,GAAgB,iBACrD,GAWL,YAAyB,GACvB,GAAI,IAAS,KACX,MAAO,MAIT,GAAM,GAAa,EAAK,GAExB,GAAI,GAAa,IAEf,GAAM,GAAQ,GAAY,GAC1B,MAAO,CAAC,MAAK,EAAE,QAAS,IAI1B,MAAO,CAAC,MAAO,KAAM,QAAS,IAOhC,YAAwD,GAEtD,GAAI,EAAO,SAAW,EAEpB,KAAM,IAAI,OAAM,wCAGlB,MAAI,GAAO,YAAc,IAAA,OAEhB,GAAA,MAAS,GAGT,GAAA,OAAU,GC9rBrB,mBA2BqC,GAAA,EAAA,GAMnC,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,IAIzB,SAAA,UAAA,SAAN,+GACwB,MAAA,CAAA,EAAM,KAAK,MAAM,mBAAjC,SAAgB,EAAA,OAChB,EAAe,EAAc,aAC7B,EAAe,EAAa,MAAM;GAAM,IAAI,SAAA,GAEhD,MAAI,GAAK,SAAS,OAChB,GAAO,EAAK,MAAM,EAAG,KAEhB,IAET,CAAA,EAAO,SAEX,GAtBqC,IC3BrC,AAyBA,GAAM,IAAa,IACb,GAAY,OAAO,OACnB,GAAc,OAAO,SACrB,GAAc,OAAO,SACrB,GAA0B,OAAO,mBACjC,GAA8B,OAAO,+BAeX,GAAA,EAAA,GAgI9B,WAA+B,EAAmB,GAAlD,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,EA9HvB,EAAA,UAAY,GACZ,EAAA,gBAA4B,KAC5B,EAAA,qBAAuB,GACvB,EAAA,cAA+C,KAC/C,EAAA,sBAAwB,GACxB,EAAA,UAAY,IACZ,EAAA,gBAAkB,GA0HxB,EAAK,KAAO,GAAI,IAAgB,GAChC,AAAK,GACH,GAAY,IAEd,EAAK,UAAY,IAAU,YAAc,IACzC,EAAK,gBAAkB,EAAU,YACjC,EAAK,cAAgB,EAAU,cAC/B,EAAK,sBAAwB,EAAU,sBACvC,AAAI,EAAU,gBACZ,IAAA,KAAK,OACD,EAAU,WAAa,KACvB,WACI,MAAA,mEACR,EAAK,gBAAkB,GACvB,EAAK,UAAY,KAEjB,EAAK,UAAY,EAAU,UAAY,EAAU,UAAY,MA9H3D,SAAA,UAAA,YAAN,2GACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEF,MAAA,CAAA,EAAO,KAAK,sBAAwB,OAAO,KAAK,KAAK,eACjB,KAAK,uBAW7B,EAAA,UAAA,eAAd,8HAC8B,MAAA,CAAA,EAAM,KAAK,8BACvC,GADM,EAAsB,EAAA,OACxB,CAAC,KAAK,iBAAmB,CAAC,EAE5B,KAAM,IAAI,OACN,6DA0BN,GAzBO,AAAI,KAAK,iBAAmB,GAEjC,GAAA,KAAK,OACD,EAAoB,SAAW,KAAK,gBAAgB,OACpD,WAAM,MAAA,uCACF,EAAK,gBAAgB,OAAO,WAC5B,kEACW,EAAoB,OAAO,WAAa,OAE7D,AAAK,KAAK,iBACR,MAAK,gBAAkB,GAGnB,EAAkC,KAAK,gBAAgB,OACzD,SAAC,EAAmC,GAClC,SAAS,GAAS,EAAS,GAAQ,GAAM,EAClC,GAET,IACE,EACF,OAAO,KAAK,GAAQ,OAAO,SAAC,GAAS,MAAC,GAAO,GAAQ,IACzD,GAAA,KAAK,OACD,EAAe,SAAW,EAC1B,WAAM,MAAA,iCAAmC,EAAe,aAExD,KAAK,eACP,IAAA,EAAA,EAAkB,EAAA,OAAO,KAAK,KAAK,eAAjB,EAAA,EAAA,OAAA,IAEhB,GAFS,EAAG,EAAA,GACN,EAAQ,KAAK,gBAAgB,QAAQ,GACvC,IAAU,GACZ,KAAM,IAAI,OACN,YAAc,EACd,uEACY,KAAK,gBAAgB,WAAa,MAIxD,YAAK,qBAAuB,aAGhB,EAAA,UAAA,oBAAd,uHACM,MAAK,UACM,CAAA,EAAM,KAAK,KAAK,YAD3B,CAAA,EAAA,UACI,SAAO,EAAA,OACQ,CAAA,EAAM,EAAK,eAChC,GADM,EAAe,EAAA,OACjB,EAAa,KACf,KAAM,IAAI,OAAM,sCAEZ,SAAoB,EAAa,MACjC,EAAU,KAAK,SAAS,EAAW,IACzC,CAAA,EAAO,UAEP,MAAA,CAAA,EAAO,YA0DL,EAAA,UAAA,SAAN,wHACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEU,MAAA,CAAA,EAAM,KAAK,KAAK,mBAAxB,SAAQ,EAAA,OACZ,AAAI,KAAK,WAGP,GAAQ,EAAM,KAAK,IAErB,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,GAAK,gBAAgB,YAG7C,EAAA,UAAA,gBAAA,SAAgB,GAKd,OAJM,GAAS,KAAK,SAAS,GACvB,EAA6C,GAC7C,EAA2C,GAExC,EAAI,EAAG,EAAI,KAAK,gBAAgB,OAAQ,KAC/C,GAAM,GAAM,KAAK,gBAAgB,GAC3B,EAAS,KAAK,cAAgB,KAAK,cAAc,GAAO,KAC9D,GAAI,KAAK,uBAAyB,CAAC,EAEjC,SAEA,GAAM,GAAQ,EAAO,GACjB,EAAc,KAClB,GAAI,IAAU,GAGZ,GAAI,GAAU,EAAO,UAAY,OAC/B,EAAc,EAAO,aAChB,GAAI,GAAW,GAAO,UAAY,EAAO,SAC9C,KAAM,IAAI,OACN,mBAAmB,EAAG,2BAA2B,GAErD,EAAc,YAIhB,GAAM,GAAa,OAAO,GAC1B,GAAI,MAAM,GAGR,AAAI,GAAU,EAAO,QAAU,OAC7B,EAAc,KAAK,WAAW,GAG9B,EAAc,UAEP,CAAC,GAAU,CAAC,EAAO,MAG5B,EAAc,MAId,QAAQ,EAAO,WACR,UACH,EAAc,EACd,UACG,QACH,EAAc,KAAK,MAAM,GACzB,UACG,OACH,EAAc,KAAK,WAAW,GAC9B,cAEA,EAAc,GAKtB,AAAC,GAAU,EAAO,QAAW,EAAO,GAAO,EACd,EAAS,GAAO,EAKjD,MAAI,QAAO,KAAK,GAAQ,SAAW,EAC1B,EAGA,CAAC,GAAI,EAAU,GAAI,IAItB,EAAA,UAAA,WAAR,SAAmB,GACjB,MAAI,KAAU,KAAO,EAAM,gBAAkB,OACpC,EAEA,GAKH,EAAA,UAAA,SAAR,SAAiB,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IAM7B,OALM,GAAmB,GACrB,EAAa,EACX,EAAa,EAAK,OACpB,EAAe,GAEV,EAAI,EAAG,EAAI,EAAY,IAC9B,OAAQ,OAED,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAa,EAAI,EACjB,EAAe,GACf,UAEG,MAAK,UAIR,GAHA,EAAa,EAAI,EAGb,KAAK,YAAc,KAAO,KAAK,gBACjC,MAEF,EAAO,KAAK,IACZ,EAAe,GACf,cAGA,EAAe,GACf,EAAa,EACb,MAEJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,IACvC,EAAe,GACf,EAAa,EAAI,EACjB,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,EAAI,IAC3C,EAAe,GACf,EAAa,EAAI,EACjB,UAEG,IACH,EAAe,GACf,cAGA,EAAe,GACf,MAEJ,UACG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,MAWN,GANA,AAAI,IAAiB,GACnB,EAAO,KAAK,EAAK,UAAU,EAAY,EAAa,IAEpD,EAAO,KAAK,EAAK,UAAU,IAGzB,GAAwB,EAAO,SAAW,KAAK,gBAAgB,OACjE,KAAM,IAAI,OAAM,wCACZ,KAAK,gBAAgB,OAAM,+BAA+B,GAEhE,MAAO,IAEX,GA1VgC,IC7ChC,AA4BA,GAAA,IAAA,SAAA,GAAwC,GAAA,EAAA,GAgBtC,WAAuC,GAAvC,GAAA,GACE,EAAA,KAAA,OAAO,KAD8B,EAAA,iBAAA,EAf/B,EAAA,SAAW,GAiBjB,EAAK,QAAU,EAAiB,SAAW,KAC3C,GAAM,GAAc,KAAK,KAAK,EAAK,SACnC,GAAI,EAAK,QAAU,GAAK,EAAc,GAAK,EAAc,IACrD,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,oDACA,gCAA+B,EAAK,UAc1C,GAXA,EAAK,UAAY,EAAiB,yBAA2B,GAC7D,EAAK,aAAe,EAAiB,aACrC,EAAK,qBACD,EAAiB,sBAAwB,EAAK,QAClD,EAAK,sBAAwB,EAAiB,sBAC9C,EAAK,sBAAwB,EAAiB,uBAAyB,EAEvE,EAAK,mBACD,IAAiB,qBAAuB,IAC5C,EAAK,gBACD,EAAiB,kBAAoB,GACrC,CAAC,EAAK,oBAAsB,CAAC,EAAK,gBACpC,KAAM,IAAI,OACN,iHAKR,SAAA,UAAA,QAAA,WACE,MAAO,cAII,EAAA,OAAb,SAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,6FAClB,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGA,SAAqB,GAAI,GAAmB,GAGlD,CAAA,EAAM,EAAmB,gBAAzB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,8IAEI,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,KAAK,uBAAyB,KAAO,GACA,KAAK,sBACjD,MAAO,aAHT,SAAK,OAAS,EAAA,qCAMR,GAAI,OACN,iDAAiD,EAAE,gBAGzD,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,2CAQlB,GALM,EAED,OAAe,cAAiB,OAAe,mBACpD,KAAK,aAAe,GAAI,GAEpB,CAAC,KAAK,aAGR,KAAK,aAAe,KAAK,aAAa,mBAC7B,KAAK,aAAa,aAAe,KAAK,aAC/C,KAAM,IAAI,OACN,8BACA,cAAa,KAAK,aAAY,MAC9B,YAAW,KAAK,aAAa,aAG7B,SAAe,KAAK,aAAa,wBAAwB,KAAK,QACpE,KAAK,SAAW,KAAK,aAAa,iBAClC,KAAK,SAAS,QAAU,KAAK,QAAU,EACvC,KAAK,SAAS,sBAAwB,KAAK,sBAC3C,EAAa,QAAQ,KAAK,UAC1B,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,CAAA,SAGI,EAAA,UAAA,KAAN,mHACE,MAAI,MAAK,SACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAMN,CAAA,EAAM,KAAK,uBAA5B,SAAiB,EAAA,OACvB,AAAI,KAAK,oBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAoB,KAAK,4BACrB,EAAU,CAAC,KAAK,UAAW,KAAK,qBAAsB,KAE5D,AAAI,KAAK,iBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAiB,KAAK,4BAClB,EAAU,CAAC,KAAK,UAAY,KAAK,QAAS,KAGhD,CAAA,EAAO,CACL,MAAO,CAAC,YAAe,EAAmB,SAAY,GACtD,KAAM,WAMJ,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAIf,EAAA,UAAA,aAAd,+FAEQ,SAAgC,GAChC,EAAgC,GAClC,EAAgB,EACpB,CAAA,EAAO,GAAI,SAAQ,SAAA,GACjB,GAAM,GAAa,YAAY,WAC7B,AAAI,EAAK,oBACP,GAAK,SAAS,sBAAsB,EAAK,UAEzC,AAAI,EAAK,SAAS,KAAO,WACvB,EAAQ,CAAC,cAAa,EAAE,cAAa,IAEvC,EAAc,KAAK,EAAK,SAAS,MAAM,EAAG,EAAK,wBAEjD,AAAI,EAAK,iBACP,GAAK,SAAS,uBAAuB,EAAK,UAC1C,EAAc,KAAK,EAAK,SAAS,UAInC,AAAI,EAAE,IAAkB,EAAK,WAC3B,eAAc,GACd,EAAQ,CAAC,cAAa,EAAE,cAAa,MAEtC,EAAK,QAAU,EAAK,aAAe,aAK1C,EAAA,UAAA,KAAA,WACE,AAAK,KAAK,UACR,MAAK,SAAW,GAChB,KAAK,SAAS,aACd,KAAK,aAAa,QAClB,AAAI,KAAK,QAAU,MAAQ,KAAK,OAAO,YAAY,OAAS,GAC1D,KAAK,OAAO,YAAY,GAAG,SAMjC,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAIlB,EAAA,UAAA,cAAA,WACE,MAAO,MAAK,cAGN,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAY,EAAM,GAAG,OACrB,EAAW,GAAI,cAAa,EAAM,OAAS,GACjD,SAAM,QAAQ,SAAC,EAAM,GAAM,MAAA,GAAS,IAAI,EAAM,EAAI,KAC3C,GAGD,EAAA,UAAA,4BAAR,SAAoC,EAAwB,GAE1D,GAAM,GAAO,GAAI,cAAa,GAAA,KAAK,cAAc,IAEjD,SAAK,IAAI,EAAU,EAAK,OAAS,EAAS,QACnC,GAAA,OAAO,EAAM,IAExB,GA7MwC,IC5BxC,AA0BA,GAAA,IAAA,SAAA,GAAoC,GAAA,EAAA,GAQlC,WACuB,EACA,GAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KACP,GAHqB,EAAA,mBAAA,EACA,EAAA,aAAA,EATf,EAAA,SAAW,GAEX,EAAA,OAAS,GASX,EAAK,eAKP,GAJA,EAAK,OAAS,GACd,EAAK,SACD,CAAC,EAAK,aAAa,aAAc,EAAK,aAAa,aACvD,EAAK,WAAa,GAAA,SAAS,CAAC,GAAI,SAC5B,EAAK,aAAa,YAEpB,GAAM,GACF,EAAK,aAAa,YAAc,EAAM,EAAK,mBAAmB,MAC5D,EAAsB,EAAK,aAAa,aAAe,EACzD,EAAK,mBAAmB,OACtB,EAAkB,GAAI,GAAsB,EAC5C,EAAmB,GAAI,GAAuB,EAC9C,EAAe,EAAiB,EAChC,EAAgB,EAAsB,EAC5C,EAAK,QAAU,GAAA,SACX,CAAC,EAAiB,EAAgB,EAAe,GACjD,CAAC,EAAG,QAER,GAAK,QAAU,GAAA,SAAS,CAAC,EAAG,EAAG,EAAG,GAAI,CAAC,EAAG,aAKhD,SAAA,UAAA,QAAA,WACE,MAAO,UAII,EAAA,OAAb,SACI,EAAuC,GAAA,MAAA,KAAA,QAAA,GAAA,6FACzC,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGN,GAAI,CAAC,GAIH,GADA,EAAqB,SAAS,cAAc,SACxC,CAAC,EAAa,aAAe,CAAC,EAAa,aAC7C,KAAM,IAAI,OACN,0GAGN,EAAmB,MAAQ,EAAa,YACxC,EAAmB,OAAS,EAAa,aAErC,SAAiB,GAAI,GAAe,EAAoB,GAG9D,CAAA,EAAM,EAAe,gBAArB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,oHACE,AAAI,KAAK,aAAa,YACpB,GAAA,KAAK,OACA,KAAK,aAAa,aAAe,QAC7B,KAAK,aAAa,aAAe,cACtC,WACI,MAAA,+BAA+B,EAAK,aAAa,WAAU,2FAKnE,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,CACL,SAAU,KAAK,aAAa,SAC5B,WAAY,KAAK,aAAa,WAC1B,KAAK,aAAa,WAClB,OACJ,MAAO,KAAK,mBAAmB,MAC/B,OAAQ,KAAK,mBAAmB,kBAPpC,SAAK,OAAS,EAAA,qCAYd,EAAE,QAAU,iDAAiD,EAAE,QACzD,SAGR,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,uCAIlB,IACE,KAAK,mBAAmB,UAAY,KAAK,aAClC,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,OAAO,IAAI,gBAAgB,KAAK,QAGhE,YAAK,mBAAmB,OAExB,KAAK,SAAW,GAEhB,CAAA,EAAO,GAAI,SAAc,SAAA,GAEvB,EAAK,mBAAmB,iBAAmB,WACzC,cAKA,EAAA,UAAA,KAAN,oFACE,GAAI,KAAK,SACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAI7B,IACE,EAAM,GAAA,QAAQ,WAAW,KAAK,0BACvB,GACP,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAU,IAEjE,GAAI,KAAK,OACP,IACE,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,mBAAmB,GAAM,KAAM,WAC5C,GACP,KAAM,IAAI,OAAM,oCAAoC,EAAE,iBAEtD,EAAI,cAGN,OAAA,CAAA,EAAO,CAAC,MAAO,EAAK,KAAM,oBAItB,EAAA,UAAA,aAAR,WAIE,MAAI,QAAK,aAAa,aAAe,KAAK,aAAa,cAClD,MAAK,mBAAmB,QAAU,KAAK,aAAa,aACpD,KAAK,mBAAmB,SAAW,KAAK,aAAa,gBAO5D,EAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,IAAA,KAAK,WACV,GAAM,GAA0B,EAAI,UAAU,WAAW,GACrD,EACJ,EAAe,GAAA,MAAM,cACjB,EAAe,EAAK,QAAS,EAAK,WAAY,EAAK,SACnD,YAEJ,GAAM,GAAQ,EAAa,MAC3B,MAAO,GAAa,QAAQ,EAAM,MAAM,OAMtC,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAI7B,EAAA,UAAA,KAAA,WACE,GAAM,GAAS,KAAK,OAAO,YAE3B,EAAO,QAAQ,SAAA,GAAS,MAAA,GAAM,SAE9B,IACE,KAAK,mBAAmB,UAAY,WAC7B,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,KAEhC,KAAK,SAAW,IAIlB,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAEpB,GApMoC,IC1BpC,AA2BA,GAAA,IAAA,WAAA,cAUA,MAAA,MCrCA,AAoBA,GAAA,IAAA,SAAA,GAA6C,GAAA,EAAA,GAA7C,4DAmBE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAI,IAAc,KAAM,IAEnC,GAtB6C,IAmC7C,GAAA,SAAA,GAA4B,GAAA,EAAA,GAG1B,WAAsB,EAAgC,GAAtD,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAkB,EAAU,KAG9C,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf4B,IAiB5B,GAAA,SAAA,GAAgC,GAAA,EAAA,GAI9B,WACc,EAA0C,GADxD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAA0C,EAAA,UAAA,EAHxD,EAAA,UAAY,KAOZ,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAAc,KAAK,UAAS,MAGzD,EAAA,UAAA,KAAN,mHACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eACxC,GADM,EAAc,EAAA,OAChB,EAAY,KACd,MAAI,MAAK,YAAc,GACrB,CAAA,EAAO,IAKT,MAAK,YAAY,KAAK,KAAK,WAC3B,KAAK,UAAY,GACjB,CAAA,EAAO,KAQT,IANM,EAAQ,EAAY,MAAM,MAAM,KAAK,WAK3C,EAAM,GAAK,KAAK,UAAY,EAAM,GAClC,EAAA,EAAmB,EAAA,EAAM,MAAM,EAAG,IAAf,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,KAAK,YAAY,KAAK,GAExB,YAAK,UAAY,EAAM,EAAM,OAAS,GAEtC,CAAA,EAAO,UAEX,GAvCgC,ICxEhC,AAsBA,GAAA,IAAA,SAAA,GAAgD,GAAA,EAAA,GAAhD,4DAUE,SAAA,UAAA,WAAA,WACE,MAAO,IAAI,IAAa,OAE5B,GAbgD,IA0BhD,GAAA,SAAA,GAA2B,GAAA,EAAA,GAGzB,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAiB,KAGnC,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf2B,IAuC3B,GAAA,SAAA,GAA+B,GAAA,EAAA,GAM7B,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAF6B,EAAA,SAAA,EAEzB,GAAA,MAAM,IAAI,cACZ,EAAK,QAAU,GAAI,aAAY,cAGxB,GAAA,GAAA,KAAA,cACP,EAAK,QAAU,GAAI,GAAc,iBAGrC,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,+GACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eAExC,MAFM,GAAc,EAAA,OAEhB,EAAY,KACd,CAAA,EAAO,IAEP,GAAQ,EAAY,MAItB,AAAI,GAAA,MAAM,IAAI,cACZ,EAAO,KAAK,QAAQ,OAAO,EAAO,CAAC,OAAQ,KAE3C,EAAO,KAAK,QAAQ,MAAM,OAAO,KAAK,EAAM,SAE9C,KAAK,YAAY,KAAK,GACtB,CAAA,EAAO,WAEX,GAtC+B,ICvF/B,AAqCA,GAAA,IAAA,SAAA,GAAuC,GAAA,EAAA,GAIrC,WACc,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,KAAA,EACA,EAAA,QAAA,EAEZ,GAAA,KAAK,OACA,YAAgB,aACZ,IAAA,MAAM,IAAI,cACL,YAAgB,OAAQ,YAAgB,MACzC,IACT,WAAM,MAAA,yEAEV,EAAK,OAAS,EAAQ,QAAU,EAEhC,EAAK,UAAY,EAAQ,WAAa,KAAO,OAG/C,SAAA,UAAA,QAAA,WACE,MAAO,cAAc,KAAK,MAGtB,EAAA,UAAA,KAAN,oHACE,MAAI,MAAK,QAAY,MAAK,eAAgB,YAClB,KAAK,KAAK,WACV,KAAK,KAAK,MAChC,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAQ,GAAI,SAAoB,SAAC,EAAS,GAC9C,GAAM,GAAM,EAAK,OAAS,EAAK,UAC/B,GAAI,EAAK,eAAgB,YAGvB,EAAQ,GAAI,YAAW,EAAK,KAAK,MAAM,EAAK,OAAQ,UAMpD,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GACnB,GAAI,GAAsC,EAAW,OAOrD,GAHA,AAAI,YAAgB,cAClB,GAAO,GAAI,YAAW,IAEpB,CAAE,aAAgB,aACpB,MAAO,GAAO,GAAI,WAAU,sCAE9B,EAAQ,IAEV,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,aAE1B,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,EAAM,QAIhC,GAAM,GAAQ,EAAK,KAAK,MAAM,EAAK,OAAQ,GAG3C,EAAW,kBAAkB,GAE/B,EAAK,OAAS,SAEA,CAAA,EAAM,WAAtB,MAAA,CAAA,EAAQ,GAAA,MAAQ,EAAA,OAAc,EAAA,KAAM,GAAK,UAE7C,GAxEuC,ICrCvC,YA6BI,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,qGAGpB,MAAK,OAAO,IAAS,SACnB,EAAY,EAEZ,GAAa,EAAgB,IAC7B,EAAc,GAA0B,IAEzB,CAAA,EAAM,GAAA,KAAK,MAAM,EAAW,iBAAvC,GAAW,EAAA,OACb,EAAS,MACY,WAAU,KAAC,CAAA,EAAM,EAAS,gBAD/C,CAAA,EAAA,UACI,SAAa,GAAA,GAAA,MAAI,WAAU,CAAA,OAAC,EAAA,UAClC,CAAA,EAAO,GAAI,IAAkB,EAAY,WAEzC,KAAM,IAAI,OAAM,EAAS,iBAK7B,GAAM,IAA4B,SAAC,GACjC,GAAM,GAAO,CACX,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,KACd,KAAM,EAAQ,KACd,YAAa,EAAQ,YACrB,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,SAAU,EAAQ,SAClB,UAAW,EAAQ,WAErB,MAAO,IC5DT,YAqB4B,GAC1B,MAAQ,OAAO,IAAW,UAAa,EAAO,OAAO,EAAG,KAAO,UCtBjE,mBA6BoC,GAAA,EAAA,GASlC,WACc,EACS,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,MAAA,EACS,EAAA,QAAA,IAIjB,SAAA,UAAA,SAAN,oFACE,MAAI,IAAY,KAAK,QAAU,GAAA,MAAM,IAAI,YAEjC,GAAK,cACX,KAAK,MAAQ,EAAG,aAAc,KAAK,MAAiB,OAAO,KAI7D,CAAA,EAAO,GAAI,IAAkB,KAAK,MAAsB,KAAK,eAEjE,GAzBoC,IC7BpC,mBA4BmC,GAAA,EAAA,GAQjC,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,IAAA,EACA,EAAA,YAAA,IAQjB,SAAA,UAAA,SAAN,8EACE,MAAI,IAAY,KAAK,KACnB,CAAA,EAAQ,GAAI,IAAe,KAAK,IAAe,KAAK,aAC/C,YAEL,CAAA,EAAO,GAAiB,KAAK,IAAK,KAAK,mBAG7C,GA1BmC,IC5BnC,YA0GI,EAAqB,GAAA,MAAA,KAAA,QAAA,GAAA,IAChB,GAAI,IAAW,GAAI,IAAc,GAAS,eA4B/C,GADJ,GAAA,GAAA,KAEQ,EAAO,GAAqB,GAClC,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,qBAgEvC,GADJ,GAAA,GAAA,KAEE,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACf,MAAA,CAAA,EAAM,YAAZ,SAAM,EAAA,OACZ,CAAA,EAAO,GAAqB,WAAM,MAAA,GAAI,6BAmCtC,EACA,sEACF,MAAA,CAAA,EAAO,GAAe,OAAO,EAAoB,oBAqClB,sEAE/B,MAAA,CAAA,EAAO,GAAmB,OAAO,QCxRnC,GAGM,IAAU,oNCHhB,oBA2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9GnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC7EnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UClFnC,oBAKA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAER,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAG,EAC9B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9FnC,oBAyBA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UChJnC,oBAIA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UCnGnC,oBAwBA,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMjB,MAAO,KAAW,UAAY,GAAO,SACxC,GAAO,QAAU,EAEjB,IACE,EAAa,WACN,SACJ,AAAK,OAAO,SAAW,YAAc,OAAO,KACjD,OAAO,WAAa,MAAO,OAK3B,GACA,QCrPF,oBAYA,GAAI,IAAO,KAKP,GAAS,KAKT,GAAS,KAQT,GAAY,KASZ,GAAU,KAOV,GAAS,KAIT,GAAK,KAET,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAO,QAAU,6vDC3DjB,YAoBI,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MACF,GAAM,8DC7BpB,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBACvC,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAM/B,YACI,EAAyB,EAAW,EACpC,GACF,GAAI,IAAe,SACjB,MAAO,GAAQ,OAAO,GACjB,GAAI,IAAe,OACxB,MAAO,GAAQ,KAAK,GACf,GAAI,IAAe,MACxB,MAAO,GAAA,IAAO,GACT,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,GAChB,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,EAAG,GAE1B,KAAM,IAAI,OACN,cAAc,EAAU,qEAeM,GAAA,EAAA,GAMlC,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KANF,SAAA,UAAY,GAGX,EAAA,SAAW,GAIjB,EAAK,KAAO,GAAI,GAAA,YAAY,EAAM,EAAA,YAGpC,SAAA,UAAA,MAAA,SAAM,EAAoC,EAAiB,GAEzD,AAAI,KAAK,UACP,MAAK,SAAW,GAChB,AAAI,EAAA,MAAM,IAAI,YACZ,EAAA,aAAa,KACT;;;gCAYR,GAAM,GAAS,GAEf,YAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,IAEzC,GAST,EAAA,UAAA,eAAA,SACI,EAAiB,EACjB,GACF,GAAM,GAAQ,KAAK,MAAM,EAAQ,EAAO,GAExC,MAAO,CAAC,OAAQ,EAAO,MAAK,EAAE,MAAK,IAIrC,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,YAIb,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,aAIf,EAAA,UAAA,KAAA,SACI,EAAgB,EAAoC,EACpD,GACF,KAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,KAGlD,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,KAAK,cAGb,EAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,SAAS,SAEvB,EAAA,UAAA,SAAA,SAAS,GACD,GAAA,GAAA,KAAA,KAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,mBAEd,GAAI,IAAU,aACZ,GAAM,GACF,KAAK,SAAS,EAAmB,KAAK,QACpC,EACF,KAAK,SAAS,EAAmB,KAAK,QAC1C,MAAO,GAAA,aAAa,uBAAuB,EAAY,GAGzD,MAAO,MAAK,KAAK,IAAI,GAAQ,QAGvB,EAAA,UAAA,WAAR,SAAmC,GACjC,GAAM,GAAO,KAAK,SAAS,EAAE,QACzB,EAAc,EAClB,GAAI,EAAE,QAAU,SACd,IAEE,EAAe,EAAsB,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,WAChE,GACA,KAAM,IAAI,OAAM,oDAGpB,MAAO,GAAA,OAAU,EAAE,MAAO,EAAE,MAAO,IAGrC,EAAA,UAAA,WAAA,SACI,EAAoC,EAAiB,GACvD,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAG7D,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,KAAK,IAAI,IACT,GAAA,GAAA,KAAA,KAAA,IAAA,GAAA,mBAEP,AAAI,GAAsB,MACxB,MAAK,YAAY,EAAmB,KAAK,QACzC,KAAK,YAAY,EAAmB,KAAK,SAG3C,KAAK,KAAK,OAAO,KAIrB,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GAEjC,EAAW,WAEX,AAAI,EAAW,SAAW,GACxB,KAAK,YAAY,KAKjB,EAAA,UAAA,KAAN,SAAW,8EACH,SAAQ,EAAA,KAAK,MACnB,IACM,EAAW,EAAA,KAAK,MAAQ,EAC9B,CAAA,EAAO,CAAC,SAAQ,SAGlB,EAAA,UAAA,OAAA,WACE,MAAO,CAEL,WAAY,GACZ,QACI,CAAC,wHAKT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,GAAiB,EAAG,gBAEpB,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAU,GAAI,GAKvB,OAFM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GACpB,EAAI,EAAG,EAAI,EAAO,KAAM,KAI/B,OAHM,GAAM,EAAO,WAAW,GAExB,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,GAAK,EAAQ,GAAK,EAAM,GAE1C,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAM,OAAM,IAGrC,MAAO,GAAO,YAGhB,EAAA,UAAA,KAAA,SAAK,GAIH,OAHM,GAAQ,KAAK,SAAS,EAAE,QACxB,EAAS,EAAA,OAAU,CAAC,EAAE,KAAM,EAAE,MAAO,EAAE,OACvC,EAAO,EAAO,OACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAK,EAAI,EAAE,KAAO,GAAK,EAAM,GAE/B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,EAAA,MAAS,EAAG,EAAO,GAAM,QAAQ,GAE5C,MAAO,IAGT,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAiB,EAAG,WAKpB,OAHM,GAAS,EAAA,OAAU,EAAE,MAAO,EAAE,OAC9B,EAAO,KAAK,WAAW,cAEpB,GACP,GAAM,GAAS,EAAO,WAAW,GAC3B,EAAQ,EAAO,QACrB,EAAK,QAAQ,SAAA,GAAM,MAAA,GAAM,GAAM,EAAE,MAAM,GAAM,EAAI,EAAM,KACvD,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAK,OAAM,KAJ3B,EAAI,EAAG,EAAI,EAAO,KAAM,MAAxB,GAOT,MAAO,GAAO,YAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,UAAiB,EAAG,OAGb,EAAA,IAAO,EAAA,OAAU,IAAK,IAG/B,EAAA,UAAA,KAAA,SAAuB,GAAvB,GAAA,GAAA,KACE,GAAiB,EAAS,QAK1B,OAHM,GAAO,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAK,SAAS,EAAE,UACxC,EAAS,EAAA,OAAU,EAAQ,GAAG,MAAO,EAAQ,GAAG,OAChD,EAAa,EAAO,OACjB,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAElC,OADM,GAAW,EAAK,GACb,EAAI,EAAG,EAAI,EAAW,OAAQ,IACrC,EAAW,IAAM,EAAS,GAG9B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAGhD,EAAI,EAAA,IAAO,EAAQ,EAAS,QAAQ,IACpC,EAAI,EAAA,IAAO,GACX,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAO,EAAG,IAGnB,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,IAAI,EAAQ,MAIjE,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAiB,CAAC,EAAG,GAAI,UAqBzB,OAnBM,GAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAU,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5C,EAAW,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7C,EAAW,EAAE,MAAM,GAEnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,8DAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAAY,EAAA,EAAA,GAGrB,EAAA,8DAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAIzB,EAAO,EAAU,EACjB,EAAS,EAAA,OAAU,CAAC,EAAU,EAAS,GAAW,EAAE,OACpD,EAAU,EAAO,OACjB,EAAY,KAAK,UAEd,EAAI,EAAG,EAAI,EAAU,IAC5B,OAAS,GAAK,EAAG,EAAK,EAAS,GAAM,EACnC,OAAS,GAAK,EAAG,EAAK,EAAU,GAAM,EACpC,OAAS,GAAK,EAAG,EAAK,EAAW,GAAM,EAMrC,OAJM,GAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAE/B,EAAI,EAAI,EAAI,EAAQ,IAC3B,OAAS,GAAI,EAAI,EAAI,EAAQ,KAG3B,OAFI,GAAM,EAED,EAAI,EAAI,EAAI,EAAQ,IAC3B,GAAO,EAAQ,EAAI,EAAS,EAAI,EAAa,EAAI,GAC7C,EAAQ,EAAI,EAAa,EAAI,EAAa,EAAI,GAEpD,EAAQ,EAAI,EAAQ,GAAI,EAAW,KAAO,EAOtD,MAAO,GAAO,YAGhB,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE/C,EAAS,KAAK,YAAY,EAAG,EAAG,EAAY,GAChD,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAIvC,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAiB,CAAC,EAAG,GAAI,YAEzB,GAAM,GAAK,SAAC,EAAW,GAAc,MAAA,MAAK,MAAM,EAAI,IAC9C,EAAc,QACpB,MAAO,MAAK,oBAAoB,EAAG,EAAG,EAAa,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MASvD,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EACD,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAO,EAAM,EAAS,GAExB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,GAAiB,EAAG,OAUpB,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAO,EACF,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAQ,EAAM,EAAS,GAEzB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,EAAG,sBAOpB,OALM,GAAM,GAIN,EAAW,EAAE,KAAO,EAAW,KAC5B,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,EAAW,WAAW,EAAI,GAGzC,OAAS,GAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAY,EAAA,OAAU,EAAG,SACzB,EAAO,EAAA,MAAS,EAAW,GAAY,OAAO,WAC9C,EAAM,EAAK,IAAI,GAAG,IAAI,GAC5B,EAAI,KAAK,GAGX,MAAO,GAAA,MAAS,IAGlB,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAIlD,GAFA,GAAiB,EAAG,UAEhB,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,oDAAoD,GAAE,KAAO,GAAC,IAC9D,iBAAgB,IAWtB,OATM,GAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAE,MAAO,GAC3B,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAW,EAAE,MAAM,EAAE,KAAO,GAC5B,EAAgB,EAClB,SAAC,EAAW,GAAc,MAAA,GAAI,EAAW,EAAI,GAC7C,SAAC,EAAW,GAAc,MAAA,GAAI,GACzB,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACrC,OAAS,GAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAM,EAAc,EAAG,GAC7B,GAAI,IAAM,EACR,EAAK,GAAO,EAAY,EAAI,EAAM,QAElC,GAAM,GAAU,EAAc,EAAG,EAAI,GACrC,EAAK,GAAO,EAAY,EAAM,GAAW,EAAK,GACtB,EAAM,GAAO,EAAK,IAIhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,UAAiB,CAAC,EAAG,GAAI,YAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,UAAiB,CAAC,EAAG,GAAI,QAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,UAAiB,CAAC,EAAG,GAAI,gBAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,UAAiB,CAAC,EAAG,GAAI,cAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAiB,CAAC,EAAW,EAAG,GAAI,UAYpC,OAVM,GAAS,KAAK,SAAS,EAAU,QACjC,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,MAAS,EAAE,MAAO,EAAA,WAAW,EAAE,MAAO,EAAE,QACjD,EAAY,KAAK,SAAS,EAAO,QACnC,EAAQ,EACN,EAAS,EAAU,OAAS,GAAK,EAAU,KAAO,GAAK,EAAE,OAAS,EACpE,EACA,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAE5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,OAAS,GAAI,EAAG,EAAI,EAAQ,IAC1B,AAAI,EAAO,KAAO,EAChB,EAAU,KAAW,EAAQ,GAE7B,EAAU,KAAW,EAAQ,GAKnC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GACJ,GAAiB,CAAC,GAAY,SAE9B,GAAM,GAAW,KAAK,SAAS,EAAU,QACzC,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAiB,EAAG,QAEpB,GAAM,GAAQ,KAAK,SAAS,EAAE,QAC9B,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAM,EAAO,EACnB,MAAK,GAAO,GAAK,EAAO,GAAO,GAAQ,GAAK,GAAQ,EAC3C,EAEC,GAAM,GAAQ,KAK5B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAM,GAAO,EAEf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAS,EAAM,GACV,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAS,GAAU,EAErB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,UAAiB,CAAC,EAAG,GAAI,qBAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAO,EAAO,EACpB,MAAO,GAAO,KAIlB,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,IAGT,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,EAAG,EAAO,IAElC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,KAAK,IAAI,EAAG,EAAO,IAAK,GAEhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MACR,SAAC,EAAQ,GAAW,MAAA,GAAS,EAAI,EAAS,EAAS,KAGhE,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAiB,CAAC,EAAI,GAAI,UAK1B,OAHM,GAAe,GAAI,cAAa,EAAE,MAClC,EAAS,KAAK,SAAS,EAAE,QACzB,EAAW,KAAK,SAAS,EAAG,QACzB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,GACjB,AAAI,GAAK,EACP,EAAa,GAAK,EAAS,GAE3B,EAAa,GAAK,EAAS,GAAM,GAAI,GAGzC,MAAO,MAAK,WAAW,EAAc,EAAE,MAAO,YAGhD,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,MAAM,EAAQ,MAInE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,OAAO,EAAO,EAAQ,GAExC,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAiB,CAAC,EAAG,GAAS,UAyB9B,OAvBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAiB,EAAS,aAAe,eAEzC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OAEnC,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAChD,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAEhD,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EACf,EAAW,EAAI,EACZ,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EACxB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,IAAK,EAAW,EAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAC7B,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,GAAK,GAC1B,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,GAAK,IAClB,GAAO,EAAM,GAAW,IAE9B,IAAY,EAAS,cAOjC,MAAO,GAAE,YAGX,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAiBpC,OAfM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAI,EAAA,OAAmB,EAAS,SAAU,EAAE,OAE5C,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAKF,OAHM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAElC,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAS,YACpC,EAAW,EAAK,EAAS,YAAc,EACpC,GAAK,EAAG,GAAK,EAAa,MACjC,GAAM,IAAK,EAAW,GAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,GAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAAS,WACtC,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,IACrB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,KAAO,GAAO,EAAM,GAAW,IAElD,IAAY,EAAS,eASrC,MAAO,GAAE,YAGX,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,kBAkC/B,OAhCM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAiB,IAAe,eAChC,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GACjD,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GAE9C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,IAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAOjC,OANM,IAAK,GAAK,EAAc,GACxB,GACF,EAAe,EAAI,EAAa,GAAK,EAAa,GAChD,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,EAAiB,IAC7C,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,GAAM,IAAW,EAAe,EAAI,EAAa,GAC7C,EAAa,GAAK,EAAiB,EACvC,EAAS,IAAY,GAK7B,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GA6BF,OA5BM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAE1B,EAAA,EAAA,UACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,GAAA,EAAA,YAEI,GAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,GAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,GAAU,EAAc,EAAI,EAAS,QAAQ,KAE1C,GAAI,EAAG,GAAI,EAAW,EAAE,GAC/B,OAAS,IAAK,EAAG,GAAK,EAAY,EAAE,GAElC,OAAS,IAAK,EAAG,GAAK,EAAS,EAAE,GAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAGzC,GAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,KACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,IAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAc,GAErB,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAQjC,OAPM,IAAK,GAAK,GAAc,GACxB,GACF,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GACxC,GAAY,EAAS,GAAc,EAAI,IACzC,EAAS,GAAe,EAAI,IAC5B,EAAS,GAAc,EAAI,IAAM,EAAQ,GAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAK3B,EAAS,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GAAK,IACpD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAiB,CAAC,EAAG,GAAK,mBAa1B,OAXM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eACzC,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,EAC3C,OAAS,GAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAG5C,OADI,GAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,AAAI,EACF,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAEnD,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAK3D,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAKpC,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAqBzC,OAnBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAC9C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAEhB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAEvB,EAAK,EAAG,EAAK,EAAa,EAAE,EAMnC,OALM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAW,GAAM,IAChD,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAW,GAAM,GACtD,EAAW,EAAK,EAEb,EAAK,EAAG,EAAK,EAAc,EAAE,EAOpC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,GAAQ,KAAK,IACf,EAAS,UACR,GAAS,SAAW,EAAS,GAAM,GAClC,GAAW,EAAK,EAAO,EAEpB,GAAK,EAAG,GAAK,EAAa,EAAE,GAOnC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,IAAM,IAC/C,GAAQ,KAAK,IACf,EAAS,SACR,GAAS,QAAU,EAAU,IAAM,GAClC,GAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,GAG3C,OAFM,IAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,IAE5C,OADI,IAAU,EACL,GAAI,EAAG,GAAI,EAAS,UAAW,EAAE,GAIxC,OAHM,IAAW,GAAI,EACf,GAAW,GAAI,EAEZ,GAAK,EAAO,GAAK,EAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAe,EAC9B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,IACjC,GAAM,IAAK,GAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAE7B,IACI,EAAQ,GAAW,IAAM,EAAS,GAAW,IAKzD,EAAS,GAAW,IAAM,GAMpC,MAAO,GAAG,YAGZ,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,gBAAgB,EAAO,EAAQ,GAEjD,MAAI,IAGF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAiB,CAAC,EAAG,GAAS,mBAc9B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WACxC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OACnC,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,EAAE,GACpC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,EAAE,GACnC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAMF,OAJM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAS,WACtC,EAAW,EACX,EAAW,EACN,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,GAE3C,OADM,GAAO,EAAM,EAAW,GACrB,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAM,EAAW,IAAM,EAAO,EAAM,EAAW,GAEjD,GAAY,EACZ,GAAY,IAQxB,MAAO,GAAE,YAGX,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,2BA0B/B,OAxBM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAQ,EAAc,EAEnB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,GAAK,EAAG,EAAK,EAAU,EAAE,EAMhC,OALM,IAAW,EAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAMjC,OALM,IAAK,GAAK,EAAc,GACxB,GAAW,EAAO,EAAI,EAAO,GAAK,EAAO,GACzC,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAO,EAAE,IAC7B,GAAM,IAAK,EAAK,EAAQ,GAClB,GAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,EAAS,EAAO,EAAI,EAAO,EAAK,EAAO,GAAK,GAAM,GAK1D,MAAO,GAAG,YAGZ,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAiB,CAAC,EAAG,GAAK,4BAc1B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WAExC,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAK5C,OAJM,GAAK,KAAK,MAAM,EAAK,GACrB,EAAK,EAAK,EAEZ,EAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,GAAW,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAIhE,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAIlC,MAAO,GAAG,YAGZ,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,UAAiB,EAAG,QACb,GAAK,KAAK,WAAW,GAAI,IAGlC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,GAAiB,CAAC,EAAG,GAAU,UAE/B,GAAM,GAAqB,EAAE,MAAM,QAC7B,EAAgB,KAAK,SAAS,EAAQ,QAC5C,EAAS,GAAQ,EAAc,OAI/B,OAHM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GAEpB,EAAI,EAAG,EAAI,EAAO,KAAM,EAAE,GACjC,GAAM,GAAS,EAAO,WAAW,GAE3B,EAAwB,EAAO,QACrC,EAAY,GAAQ,EAAc,EAAO,IAEzC,GAAM,GAAgB,EAAK,WAAW,GACtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,YAGhB,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAa,EAAE,QAAQ,GAAW,GAC7B,QAAQ,GACR,MAAM,EAAkB,IAG9B,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAiB,EAAG,UA8BpB,OA5BM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,OAAU,EAAS,SAAU,EAAE,OACxC,EAAa,EAAO,OAEpB,EAAqB,EAAS,SAAS,GAAK,EAAS,SAAS,GAChE,EAAS,SAAS,GAAK,EAAS,SAAS,GACvC,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAGhD,OAFM,GAAoB,EAAQ,EAC5B,EAAmB,EAAQ,EAAE,QAAQ,GAClC,EAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAMf,OAJM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAChD,EACF,EAAoB,EAAS,EACxB,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAKb,OAHM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,GAAkB,EAAoB,EAAO,EAC1C,GAAO,EAAG,GAAO,EAAS,SAAU,EAAE,IAG7C,OAFM,IAAa,GAAO,EAAc,EACpC,GAAU,GACP,GAAU,GACf,IAAW,EASb,OAPM,IACF,KAAK,IAAI,EAAS,QAAS,EAAuB,IAEhD,GAAkB,GAAkB,GAAO,EAC7C,GAAc,EACd,GAAW,EACX,GAAQ,EACH,GAAS,EAAW,GAAS,EACjC,IAAU,GAEb,OADM,IAAe,EAAmB,GAAS,EAAE,QAAQ,GAClD,GAAO,EAAS,GAAO,EAC3B,IAAQ,GAEX,OADM,IAAa,GAAe,GAAO,EAAE,QAAQ,GAC1C,GAAO,GAAS,GAAO,GAC3B,IAAQ,GACX,GAAM,IAAa,GAAa,GAAO,EAAE,QAAQ,GAC3C,GAAQ,EAAQ,GAAa,GAOnC,GANA,AAAK,IAAa,OAAS,GAAQ,GACjC,GAAc,GACT,AAAI,IAAa,OACtB,KAAY,GACZ,MAEE,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAM,IAAe,GAAkB,EACvC,EAAW,IACP,IAAa,MAAQ,GAAW,GAAQ,KAMtD,MAAO,GAAO,YAGhB,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGzC,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,qBAuB1B,OArBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAgB,EAAK,GAAc,EAAe,GAElD,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,KAIjB,EAAG,IACC,EAAU,EAAe,EAAO,EAAS,EAAO,EAChD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGjC,EAAA,UAAA,mBAAR,SAA2B,EAAa,GAiBtC,OAfM,GAAe,EAAA,OAAU,EAAS,SAAU,SAC5C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,KAAK,WAAW,GACpB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAIf,OAFM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAC7C,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAIb,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAO,EAAG,EAAO,EAAS,SAAU,EAAE,GAG7C,OAFM,GAAa,EAAO,EAAc,EACpC,EAAU,EACP,EAAU,GACf,GAAW,EASb,OAPM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAGlD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAS,EAAW,EAAS,EACjC,GAAU,EAEb,OADM,GAAS,EAAS,EACf,EAAO,EAAS,EAAO,EAC3B,GAAQ,EAEX,OADM,GAAO,EAAO,EACX,GAAO,EAAS,GAAO,EAC3B,IAAQ,GACX,GAAM,IAAO,GAAO,EACd,GAAQ,EAAK,IAAI,EAAO,EAAQ,EAAM,GAAM,GAClD,AAAI,IAAS,GACX,GAAW,GACX,EAAc,EAAS,EACf,EACJ,EAAO,EAAwB,IAM3C,EAAa,IAAI,EAAa,EAAO,EAAQ,EAAM,EAAM,KAMnE,MAAO,GAAa,YAGtB,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAiB,CAAC,EAAG,GAAI,qBAoBzB,OAlBM,GAAe,KAAK,mBAAmB,EAAG,GAC1C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAY,KAAK,WAAW,GAC5B,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,GAAS,EACP,EAAwB,EAC5B,EACA,EAAU,IAAI,EAAO,EAAS,EAAO,EAAO,GAC1C,GACF,EAAS,EAAwB,EACjC,EAAO,EAAuB,EAE5B,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,GAAQ,KAIzB,EAAG,IAAI,EAAS,EAAO,EAAS,EAAO,EAAO,GAMxD,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,kBAqBpB,OAnBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cACf,EAAA,KAAK,cAAc,CAAC,EAAO,EAAW,EAAU,KAE9C,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAE9C,EAAY,EACV,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GACvC,EAAI,EAAG,EAAI,EAAO,IACzB,OAAS,GAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EAAgB,KAAK,IAAI,EAAY,EAAG,KAAK,KAAK,IAClD,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAiB,EAAE,QAAQ,GAC7D,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAgB,EAAE,QAAQ,GACzD,EAAI,EAAG,EAAI,EAAU,IAU5B,OATM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EACF,KAAK,IAAI,EAAW,EAAG,KAAK,KAAK,IAC/B,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GACvD,EAAI,EAAG,EAAI,EAAa,KAI/B,GAAM,GAAU,EAAQ,EAAgB,GAClC,EAAa,EAAQ,EAAgB,GACrC,GAAW,EAAQ,EAAiB,GACpC,GAAc,EAAQ,EAAiB,GAEvC,GAAM,EAAW,IAAW,GAAW,EACvC,GAAS,EAAc,IAAc,GAAc,EACnD,GAAW,GAAO,IAAS,IAAO,EAExC,EAAO,KAAe,GAK9B,MAAO,GAAA,OAAU,EAAQ,CAAC,EAAO,EAAW,EAAU,KAGxD,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAChD,GAAiB,CAAC,EAAI,GAAI,0BA+B1B,OA7BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GAOrD,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAMhD,EAAW,KAAK,SAAS,EAAG,QAC9B,EAAS,EACJ,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAU,EAAI,EAAE,QAAQ,GACrB,EAAI,EAAG,EAAI,EAAS,IAU3B,OATM,GAAM,EAAI,EACV,EAAc,KAAK,MAAM,GACzB,EAAiB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAU,GAEpD,EAAe,EAAU,EAAc,EAAE,QAAQ,GACjD,EAAkB,EAAU,EAAiB,EAAE,QAAQ,GAEvD,EAAU,EAAM,EAChB,EAAiB,EAAM,EACpB,EAAI,EAAG,EAAI,EAAQ,IAmB1B,OAlBM,GAAM,EAAI,EACV,EAAe,KAAK,MAAM,GAC1B,EAAgB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAS,GAClD,EAAU,EAAM,EAChB,EAAiB,EAAM,EAEvB,EAAkB,EAAe,EAAe,EAAE,QAAQ,GAC1D,EAAmB,EAAe,EAAgB,EAAE,QAAQ,GAC5D,GACF,EAAkB,EAAe,EAAE,QAAQ,GACzC,GACF,EAAkB,EAAgB,EAAE,QAAQ,GAE1C,GACF,EAAiB,EACf,GAA6B,EAAiB,EAC9C,GAA6B,EAAU,EACvC,GAAsB,EAAU,EAC7B,GAAI,EAAG,GAAI,EAAO,MACzB,GAAM,IAAQ,EAAS,KACvB,EAAO,EAAkB,KACrB,GAAQ,GACZ,EAAO,EAAmB,KAAM,GAAQ,GACxC,EAAO,GAAqB,KACxB,GAAQ,GACZ,EAAO,GAAsB,KAAM,GAAQ,GAKnD,MAAO,GAAA,SAAY,EAAQ,CAAC,EAAO,EAAQ,EAAS,GAAQ,EAAE,QAGhE,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,yBAsBpB,OApBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cAAa,EAAQ,EAAY,EAAW,GAEzD,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GAE5C,EAAe,EACV,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAY,EACZ,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAc,EAAmB,EAAE,QAAQ,GACpD,EAAI,EAAG,EAAI,EAAU,IAO5B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAW,EACX,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAY,EAAmB,EAAE,QAAQ,GAClD,EAAI,EAAG,EAAI,EAAa,KAG/B,GAAM,GAAS,EAAQ,EAAY,GACnC,EAAO,KAAkB,EAKjC,MAAO,GAAA,OACH,EAAQ,CAAC,EAAO,EAAW,EAAU,GAAc,EAAE,QAG3D,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,iCAiC1B,OA/BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GACrD,EAAW,KAAK,SAAS,EAAG,QAK5B,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAGzC,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAS,IAM3B,OALM,GAAY,EAAc,EAAI,EAAE,QAAQ,GAGxC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAY,GAC7C,EAAI,EAAG,EAAI,EAAQ,IAO1B,OANM,GAAY,EAAY,EAAI,EAAE,QAAQ,GAGtC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAW,GAE5C,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHI,GAAQ,EAGH,EAAW,EAAG,EAAW,EAAW,KAC3C,GAAM,GAAM,EAAW,EAEvB,GAAI,EAAM,GAAK,GAAO,EACpB,SAGF,GAAM,GAAY,EAAc,EAAM,EAAG,QAAQ,GAC3C,GAAgB,EAAM,EACtB,GAAmB,KAAK,IAC1B,EAAU,EACV,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAC9B,GAAI,IAAM,GACR,SAEF,OAAS,IAAW,EAAG,GAAW,EAAU,MAC1C,GAAM,IAAM,GAAW,EAEvB,GAAI,GAAM,GAAK,IAAO,EACpB,SAGF,GAAM,IAAY,EAAY,GAAM,EAAG,QAAQ,GACzC,GAAgB,GAAM,EACtB,GAAmB,KAAK,IAC1B,EAAS,EACT,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAE9B,AAAI,IAAM,IACR,IAAS,EAAS,GAAY,KAIpC,EAAO,EAAY,GAAK,EAKhC,MAAO,GAAA,SAAY,EAAQ,EAAE,MAAO,EAAE,QAGxC,EAAA,UAAA,6BAAA,SACI,EAAa,EAAqB,EAAc,EAChD,GACF,GAAiB,EAAG,gCAEpB,GAAM,GAAW,EAAE,MAAM,GACnB,EAAO,EAAW,EAClB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAO,EAAE,KACT,EAAS,GAAI,cAAa,GAEhC,WAA2B,GAQzB,OAPM,GAAiB,EAAS,EAC5B,EACA,EAAS,EAAiB,KAAK,IAAI,EAAG,EAAiB,GACrD,EAAe,EAAS,EAC1B,KAAK,IAAI,EAAiB,EAAa,GAEvC,EAAM,EACH,GAAkB,EAAc,KACrC,GAAM,GAAI,EAAQ,GAClB,GAAO,EAAI,EAEb,MAAO,GAGT,OAAS,GAAS,EAAG,EAAS,EAAM,KAClC,GAAM,GAAM,EAAkB,GACxB,EAAM,EAAQ,GAAU,KAAK,IAAI,EAAO,EAAQ,EAAK,CAAC,GAC5D,EAAO,GAAU,EAGnB,MAAO,GAAA,SAAY,EAAQ,EAAE,QAG/B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAiB,EAAI,WAQrB,OAPM,GAAW,EAAG,MAAM,GACpB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAmB,KAAK,SAAS,EAAW,QAC5C,EAAoB,KAAK,SAAS,EAAY,QAC9C,EAAS,GAAI,cAAa,EAAG,MAC7B,EAAO,EAAG,KAEP,EAAS,EAAG,EAAS,EAAM,KAQlC,OAPM,GAAiB,EAAS,EAC1B,EACD,EAAS,EAAkB,KAAK,IAAI,EAAG,EAAiB,GACvD,EAAY,EAAS,EACvB,KAAK,IAAI,EAAU,EAAiB,EAAc,GAElD,EAAO,EACF,EAAI,EAAY,EAAI,EAAU,IACrC,GAAQ,KAAK,IAAI,EAAiB,GAAI,GAExC,EAAO,EAAQ,EAAO,EAEtB,OAAS,GAAI,EAAY,EAAI,EAAU,KACrC,GAAI,GAAM,GAAK,EAAQ,EAAO,EAAiB,GAC3C,EAAkB,GAAU,EAChC,AAAI,IAAW,GACb,IAAO,KAAK,IAAI,EAAM,CAAC,IAEzB,GAAO,EAAS,GAChB,EAAO,IAAM,GAGjB,MAAO,GAAA,SAAY,EAAQ,EAAG,QAGhC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAiB,EAAQ,eASzB,OAPM,GAAgB,EAAa,EAAS,EAAA,QAAW,GACjD,EAAY,EAAc,MAAM,GAChC,EAAY,EAAc,MAAM,GAChC,EAAM,EAAA,MAAkB,CAAC,EAAW,GAAa,SACjD,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAW,KAAK,SAAS,EAAc,QAEpC,EAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,GAAS,EAAI,EAGb,EAAM,GAAI,cAAa,EAAY,GACzC,EAAI,GAAK,EAAS,GAClB,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,EAAE,EACxC,EAAI,GAAS,EAAI,EAAQ,GAAK,EAAS,EAAS,GAKlD,OAFM,GAAS,GAAA,KAAgB,EAAK,YAC9B,EAAY,EAAI,EACb,EAAW,EAAG,EAAW,EAAY,EAAE,GAC9C,GAAM,GAAI,IAGV,EAAQ,EAAY,GAAY,EAAI,OAEpC,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,IACtC,GAAI,EAAI,EAAI,IACV,EAAQ,EAAY,GAAY,EAChC,QAKR,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAiB,EAAS,UAE1B,GAAM,GAAM,GAAI,cAAa,EAAQ,KAAO,GAC5C,EAAI,KAAK,GAGT,OAFM,GAAa,KAAK,SAAS,EAAQ,QAEhC,EAAQ,EAAG,EAAQ,EAAQ,KAAM,EAAE,EAC1C,AAAI,EAAW,IAAU,GAAK,EAAW,GAAS,GAChD,GAAI,EAAQ,EAAQ,EAAW,IAAU,GAG7C,MAAO,GAAA,SAAY,EAAK,CAAC,EAAQ,KAAM,GAAQ,UAGjD,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,GAAiB,EAAO,qBAExB,GAAM,GAAY,KAAK,SAAS,EAAM,QAChC,EAAa,KAAK,SAAS,EAAO,QACxC,MAAO,IACH,EAAW,EAAY,EAAe,EAAc,IAG1D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,IAAe,OACf,WAAM,MAAA,+DACF,IACR,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAgB9D,OAdM,GAAY,EAAE,MAAM,GACpB,EAAc,EAAE,MAAM,GACtB,EAAa,EAAE,MAAM,GACrB,EAAa,EAAE,MAAM,GAErB,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAU,KAAK,SAAS,EAAE,QAC1B,EACF,GAAI,cAAa,EAAY,EAAe,EAAc,GAE1D,EAAY,EACP,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAI,EAAG,EAAI,EAAc,EAAE,EAGlC,OAFM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACZ,EAAI,EAAG,EAAI,EAAa,EAAE,EAIjC,OAHM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACf,EAAW,GAAU,EAAY,GAAW,EACzC,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAM,EAAI,EACV,EACF,EAAM,EAAc,GAAM,EAAc,GAAM,EAAc,IAChE,EAAO,KAAe,EAAQ,GAKtC,MAAO,GAAA,SACH,EAAQ,CAAC,EAAW,EAAc,EAAa,KAG7C,EAAA,UAAA,oBAAR,SACI,EAAW,EAAW,EACtB,GACF,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAE,MAAO,EAAE,OAC9D,EAAS,EAAA,OAAU,EAAU,GAC7B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GACxD,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GAExD,EAAU,EAAO,OACvB,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,EACpC,EAAQ,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAK3D,QAFM,GAAO,KAAK,WAAW,GACvB,EAAO,KAAK,WAAW,cACpB,GACP,GAAM,GAAM,EAAO,WAAW,GAExB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAEzB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAE/B,EAAQ,GAAK,EAAG,EAAM,GAAS,EAAM,KAX9B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,IAA7B,GAcX,MAAO,GAAO,YAGhB,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,QAAA,aAEA,EAAA,UAAA,eAAA,WACE,MAAO,KAIT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAGtB,EAAA,UAAA,cAAA,SACI,EACA,EACA,EACA,EACA,EACA,GAmBF,OAjBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GACjC,EAAW,EAAM,MAAM,GAEtB,EAAA,EAAA,GAAY,EAAA,EAAA,GACb,EACF,EAAA,OAAU,CAAC,EAAU,EAAY,EAAW,GAAc,WAExD,EAAU,KAAK,SAAS,EAAM,QAC9B,EAAa,KAAK,SAAS,EAAS,QACpC,EAAY,KAAK,SAAS,EAAO,QAEjC,EAAW,EAAO,QAClB,EAAY,EAAO,QAKhB,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAW,EAAI,EACf,EAAK,EAAQ,GACb,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GAExB,EAAe,EAAW,GAChC,GAAI,GAAQ,EACV,SASF,OANM,GAAe,EAAa,EAC7B,GAAK,GAAO,GAAc,GAAM,GAAa,GAC9C,EACE,EACD,EAAY,EAAM,GAAK,GAAO,GAAa,GAAM,GAAY,GAAK,EAE9D,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAgB,EAAa,EAC/B,EAAM,GAAc,GAAK,EAAK,EAC9B,GAAO,GAAK,GAAO,GAAc,GAErC,GAAI,EAAO,GAAK,EAAO,EAAc,GACnC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAGzB,SAGF,GAAI,IAAW,WAKb,OAJM,GAAS,KAAK,MAAM,GACpB,EAAY,KAAK,KAAK,GACtB,GAAQ,EAAO,EAEZ,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAOF,OAJM,IAAU,KAAK,MAAM,IACrB,GAAW,KAAK,KAAK,IACrB,GAAQ,GAAO,GAEZ,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAI,GAAM,EAAI,GAAU,EAAS,GAAK,EAAS,EAAS,GACpD,EAAO,EAAS,GACd,GAAU,EAAU,GAE1B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAS,EAAS,GACjD,EAAO,EAAS,GACpB,GAAM,IAAW,EAAU,GAE3B,EAAM,EAAI,GAAU,EAAS,GAAK,EAAY,EAAS,GACnD,EAAO,EAAS,GACpB,GAAM,IAAa,EAAU,GAE7B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAY,EAAS,GACpD,EAAO,EAAS,GACpB,GAAM,IAAc,EAAU,GAExB,GAAM,GAAW,IAAW,IAAW,GACvC,GAAS,GAAc,IAAc,IAAc,GAEzD,EAAM,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC9D,EAAO,OAAO,GAAO,GAAQ,IAAS,IAAO,QAIjD,QAAS,GAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAKF,OAFM,IAAW,KAAK,MAAM,IACtB,GAAW,KAAK,MAAM,GACnB,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,IAAQ,EAAI,GAAW,EAAS,GAClC,GAAW,EAAS,GAAK,EAAO,EAAS,GACvC,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,IAAU,EAAU,OAM5C,MAAO,GAAO,YAGhB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAiB,GACvB,MAAO,MAAK,QACR,EAAe,EAAc,EAAa,EAAY,EACtD,EAAY,EAAW,EAAS,EAAc,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAE1C,GAAI,IAAc,EAChB,MAAO,GAAA,OAAU,GAAI,EAAa,EAAE,OAOtC,OAJM,GAAS,GAAI,GAAA,aAAa,CAAC,EAAW,GAAY,EAAE,OACpD,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAQ,KAAK,SAAS,EAAE,QAErB,EAAI,EAAG,EAAI,EAAW,KAG7B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,GAAgB,EAAM,EAAQ,GAC9B,EAAM,KAAK,GAEb,GAAI,EAAe,GAAK,GAAgB,EAAE,KAAO,EAC/C,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,EAAE,OAGzD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,EAAO,OAAO,EAAI,EAAY,GAAK,EAAM,EAAe,EAAY,GAGxE,MAAO,GAAO,WAAW,QAAQ,IAGnC,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAe,EAAA,OAAU,GACzB,EAAiB,GACvB,MAAO,MAAK,QACR,EAAS,EAAS,EAAO,EAAY,EAAW,EAAY,EAC5D,EAAS,EAAc,IAG7B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,EAAQ,GAAS,EAAA,KAAK,WAAW,GACjC,GAAM,GACF,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IACrD,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,OAGnD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAEhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,GAAM,GAAS,EAAA,KAAK,kBACD,EAAE,MAAO,EAAA,KAAK,cAAc,EAAE,QACjD,MAAO,MAAK,WAAW,EAAQ,EAAE,MAAO,EAAE,QAG5C,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGxC,EAAA,UAAA,QAAR,SACI,EAAiB,EAAiB,EAAoB,EACtD,EAAmB,EAAoB,EACvC,EAAmB,EACnB,GACF,GAAM,GAAe,CAAC,EAAa,EAAW,GAExC,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAc,KAAK,SAAS,EAAQ,QAE1C,GAAI,IAAe,EACjB,MAAO,GAAA,OAAU,GAAI,EAAO,EAAQ,OAGtC,GAAM,GAAS,GAAI,GAAA,aAAa,EAAc,EAAQ,OACtD,EAAO,OAAO,KAAM,KAAK,SAAS,EAAa,QAAuB,IAEtE,OAAS,GAAI,EAAG,EAAI,EAAY,KAG9B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,EAAM,KAAK,GACX,GAAgB,EAAM,EAAQ,GAGhC,GAAI,EAAe,GAAK,GAAgB,EAAa,EACnD,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,GAGvD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,AAAI,EACF,EAAO,OAAO,EAAe,EAAY,IACrC,EAAY,EAAI,EAAY,GAEhC,EAAO,OAAO,EAAe,EAAY,GAAK,EAAQ,OAAS,EAC3D,EAAY,GACZ,EAAY,EAAI,EAAY,GAItC,MAAO,GAAO,WAAW,QAAQ,IAErC,GAzjFoC,EAAA,eC5DpC,YAqB8B,GAE5B,OADM,GAAe,GAAI,cAAa,EAAK,QAClC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAa,GAAK,KAAK,IAAI,EAAK,IAElC,MAAO,GAGF,GAAM,IACT,SAAC,GACQ,GAAA,GAAA,EAAA,OAAA,EACD,EAAa,EAAK,QACpB,EAAe,GAAI,cAAa,EAAA,KAAK,cAAc,EAAE,QACzD,GAAI,EAAE,QAAU,aACd,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC7C,EAAe,GAAc,OAS7B,QAPM,GAAc,EAAW,KAAK,IAAI,EAAE,QACpC,EAAO,EAAY,mBAAmB,KACtC,EAAO,EAAY,mBAAmB,KACtC,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC/B,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAO,EAAS,GAChB,EAAO,EAAS,GACtB,EAAa,GAAK,KAAK,MAAM,EAAM,GAGvC,MAAO,GAAW,WAAW,EAAc,EAAE,MAAO,YAG7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICzDd,YAwB6C,GAE3C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAmB,GACzB,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAE3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAEhC,EACF,EAAA,KAAK,uBAAuB,EAA0B,GAEpD,EAAQ,EAAO,OACf,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAE7D,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAG1D,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAE5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,KAX7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,IAA5B,GAeX,MAAO,CAAC,EAAQ,IClEpB,YAqBwB,GAEf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,KAAM,EAAA,EAAA,KAEP,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACzC,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAEzC,EAAc,EAAQ,eAAe,EAAK,MAAO,aAEjD,EAAU,EAAQ,KAAK,IAAI,EAAY,QAK7C,SAAQ,mBAAqB,CAC3B,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,GACpD,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,IAG/C,EAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC/Cd,YAsBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IClCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAGP,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAIvC,GAAM,GAAc,EAAA,MAAS,EAAE,OACzB,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,EAAE,MAAO,CAAC,MAAO,aAEpD,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAQ,KAAM,GAAc,QAAO,IAE/D,SAAY,UACZ,EAAQ,8BAA8B,GAE/B,EAIT,GAAI,EAAE,QAAU,aACd,GAAM,GAAW,GAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,IAC5C,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAO,EAAE,MAAO,CAAC,MAAK,KAElE,SAAQ,8BAA8B,GAE/B,EAGT,GAAI,CAAC,EAAA,KAAK,gBAAgB,EAAE,MAAO,IAGjC,GAAM,GAAS,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAC7C,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAO,MAAO,MAAK,GAG3D,GAAI,IAAU,SACZ,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAe,WAAW,KAAK,GACrC,MAAO,GAAQ,eAAe,EAAE,MAAO,QAAS,GAGlD,GAAI,IAAU,QAIZ,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAO,EAAA,KAAK,aAAa,CAAC,GAAI,EAAE,OAEhC,EAAA,GAAA,SAAA,EAAA,GAAA,MAAA,KAAA,EAAA,EAAA,IAAA,EAAA,MAAA,GAAA,EAAA,EAAA,QAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAQ,eAAe,EAAa,OAAQ,GAGrD,KAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,GAG1D,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC9Fd,YAsCI,EAAc,EACd,EAAuC,GACzC,MAAI,IAAe,KACV,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAiB,CAAC,EAAG,GAAI,GAEzB,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,IAInD,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,aACzC,GAAM,GAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGjC,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAY,KAAM,GAAa,QAAS,IAE5D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,OAEP,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,iBASd,GAE5C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAyB,EACzB,GACN,GAAM,GAAc,EAAA,aAAa,2BAA2B,EAAQ,GAC9D,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAY,OACzB,EAAgB,EAAA,KAAK,eAAe,GAEpC,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GACxD,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GAExD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAEvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GACvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GAEvD,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAO,EAAI,EAAM,OACjB,EAAO,EAAI,EAAM,OAEjB,EACF,EAAG,EAAM,EAAO,GAAI,EAAM,EAAO,EAAI,GAAI,EAAM,EAAO,GACnD,EAAM,EAAO,EAAI,IAExB,EAAe,GAAK,EAAO,KAC3B,EAAe,GAAK,EAAO,SAG7B,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EACF,EAAG,EAAM,EAAS,GAAI,EAAM,EAAS,EAAI,GAAI,EAAM,EAAS,GACzD,EAAM,EAAS,EAAI,IAE1B,EAAe,GAAK,EAAS,KAC7B,EAAe,GAAK,EAAS,MAhBtB,EAAI,EAAG,EAAI,EAAe,OAAQ,MAAlC,GAmBX,MAAO,CAAC,EAAgB,EAAgB,ICvL5C,AAsBO,GAAM,IAAU,GAA8B,SAAC,EAAG,GAAM,MAAA,GAAI,IACtD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAGpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAwBsC,GAEpC,MAAO,UAAC,EAAQ,EAAO,GAGrB,OAFM,GACF,EAAA,KAAK,uBAAuB,EAA0B,EAAO,QACxD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,IChCX,YAiCI,EAAc,EAA0B,GAC1C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAQlB,OALM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,GAAS,EAAE,MACpB,EAAY,EAAA,KAAK,kBAAkB,EAAQ,GACxC,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,gBAclD,EAAc,EAA4B,GAC5C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAGlB,GAAM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAS,GAAS,EAAE,MACpB,EAAY,EAAU,EAAQ,EAAQ,GAC5C,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,IC3EtD,AAsBO,GAAM,IAAW,GAAsB,SAAC,GAAO,MAAA,MAAK,KAAK,KACnD,GAAiB,GAAwB,EAAA,KAAM,IAE/C,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,YAoBI,EAAmB,EAAoB,EACvC,GAIF,OAHM,GAAO,EAAA,KAAK,uBACd,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,GCpCT,AAqBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CACL,KAAM,EAAQ,EAAQ,EAAQ,EAC9B,KAAM,EAAQ,EAAQ,EAAQ,KAIzB,GACT,GAAiB,EAAA,SAAU,GAAc,IAEhC,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICrCd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,GAAI,KAAK,KAAK,KACxD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,YAuBI,EAAkB,EAAiB,EAAgB,EACnD,GACF,GAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,GACF,GAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAIhD,OADM,GAAU,EAAA,KAAK,uBAAuB,EAA0B,GAC7D,EAAI,EAAG,EAAI,EAAQ,EAAE,GAC5B,GAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,SAAC,EAAa,GAAM,MAAA,GAAM,EAAM,KAC/C,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEpB,MAAO,eAIL,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAAO,EAAA,EAAA,KAEd,GAAiB,EAAG,SAEd,GAAA,GAAA,EAAA,WAAA,iBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAA,WAAW,kBAAkB,EAAG,EAAQ,GAExC,GAAM,GAAO,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAClC,EAAU,GAAU,EAAM,EAAQ,EAAO,EAAE,MAAO,EAAE,OAC1D,MAAO,GAAQ,eAAe,EAAO,EAAE,MAAO,GAGzC,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IClEd,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAEpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAqBI,EAAmB,EAAkB,EAAiB,EACtD,GASF,OARM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GAEjC,EAAS,EAAA,KAAK,uBAChB,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAO,EAAE,GAK3B,OAJM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAGhC,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,EAAK,IAGvB,GAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE3B,MAAO,GC3CT,YAoBI,EAAuB,EAAc,EAAiB,GAgExD,OA1DM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAyDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GACtB,EAAI,EAAG,EAAI,EAAO,IACzB,EAAS,IAAM,EAAM,GAEvB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACxC,EAAS,IAAM,EAAM,GAevB,OAVM,GAA0C,GAG1C,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAA0B,GAC1B,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EAC/C,EAAI,EAAG,EAAI,EAAM,GAAQ,KAEhC,GAAI,GAAO,OACX,GAAI,EAEF,EAAU,EAAO,GAAG,gBAGpB,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG1C,EAAU,EAAW,KAAK,KAI5B,GAAI,EAAe,KAAa,OAC9B,EAAQ,GAAK,EAAe,QAE5B,GAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAOvB,GAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,GAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,SAAC,EAAoB,GACzC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAOxE,GAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAE7B,CACL,aAAc,EAAa,OAC3B,YAAW,EACX,QAAO,GCzJX,+MCAA,GAGM,IAAU,QCHhB,AA8BA,EAAA,gBAAgB,MAAO,WAAM,MAAA,IAAI,KAAkB,GC9BnD,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,YAoBI,EAAqB,EAAkB,EAAiB,EACxD,EACA,GAsBF,OArBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAS,EAAA,OAAO,EAAS,SAAU,GACnC,EAAa,EAAO,OAEpB,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAoB,EAAI,EACxB,EAAmB,EAAI,EAAQ,GAC5B,EAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAM1C,OALM,GAAW,EAAK,EAAe,EAC/B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,EAAkB,EAAoB,EAAK,EACxC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAQzC,OAPM,GAAW,EAAK,EAAc,EAC9B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAc,EACd,EAAW,EACX,EAAQ,EACH,EAAK,EAAO,EAAK,EAAO,GAAM,GAErC,OADM,GAAW,EAAmB,EAAK,EAAQ,GACxC,GAAK,EAAO,GAAK,EAAO,IAAM,GACrC,GAAM,IAAW,EAAW,GAAK,EAAQ,GACnC,GAAQ,EAAQ,GAAW,GACjC,AAAK,IAAa,OAAS,GAAQ,EACjC,EAAc,GACT,AAAI,IAAa,OACtB,IAAY,GACZ,KAGJ,GAAI,MAAM,GACR,MAGJ,GAAM,IAAe,EAAkB,EAAK,EAAmB,EAC/D,EAAW,IACP,IAAa,MAAQ,EAAW,EAAQ,EAKpD,MAAO,eAIL,EAAqB,EAAkB,EACvC,EAAmC,EACnC,GADmC,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,IAYF,OAXM,GAAe,EAAA,OAAO,EAAS,SAAU,SACzC,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,EAAA,OAAO,EAAQ,EAAO,GAC1B,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,GAG1C,OAFM,GAAW,EAAK,EAAe,EACjC,EAAQ,EACL,EAAQ,GACb,GAAS,EAKX,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAGzC,OAFM,GAAW,EAAK,EAAc,EAChC,EAAQ,EACL,EAAQ,GACb,GAAS,EAOX,OALM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAK,EAAO,EAAK,EAAO,GAAM,EAErC,OADM,GAAK,EAAK,EACP,EAAK,EAAO,EAAK,EAAO,GAAM,GACrC,GAAM,GAAK,EAAK,EACV,EAAQ,EAAK,IAAI,EAAG,EAAI,EAAI,GAClC,AAAI,EAAQ,GACV,GAAW,EACX,AAAI,EACF,EAAc,EACR,IAAI,EAAS,SAAW,GAAM,EAAS,QAAU,GAC3C,EAAS,WACb,EACH,GAAK,EAAS,QAAU,GAAM,EAAS,WAAa,EAEzD,EAAc,EAAK,EAAuB,GAKlD,EAAa,IAAI,EAAa,EAAG,EAAI,EAAI,IAKjD,MAAO,GCrJT,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAqBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBAyB9B,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAgB,EAAK,GAAe,GAEpC,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAGF,GAAM,GAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAGf,EAAG,IAAI,EAAU,EAAe,EAAG,EAAK,EAAK,GAKrD,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,IC1Fd,YAsBoC,GAK3B,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EAAG,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,KAAM,EAAA,EAAA,SAE/B,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGV,GAAiB,CAAC,EAAG,EAAM,EAAU,EAAO,GAAS,aAEhD,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAsBpB,OAnBM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACtC,EAAU,EAAQ,KAAK,IAAI,EAAS,QAAQ,OAC5C,EAAQ,EAAQ,EAAQ,KAAK,IAAI,EAAM,QAAQ,OAC/B,GAAI,cAAa,CAAC,IAClC,EAAU,EACZ,EAAQ,KAAK,IAAI,EAAO,QAAQ,OAChC,GAAI,cAAa,CAAC,IAChB,EAAU,GAAI,cAAa,EAAM,QAEjC,EAAgB,EAAQ,OACxB,EAAc,EAAM,OACpB,EAAgB,EAAQ,OACxB,EAAc,EAAM,OAEtB,EAAO,EACP,EAAK,EACL,EAAK,EACL,EAAK,EACA,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,EAAQ,GAAK,EAAQ,KAChB,GAAM,GAAK,EAAM,MAAS,EAAM,KAC7B,KAAK,KAAK,EAAQ,KAAQ,GAClC,AAAI,GAAQ,GACV,GAAO,GAET,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAGT,MAAO,GAAQ,eAAe,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAM,IAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,IC5Fd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,YAAa,SAAC,EAAI,GAC9D,GAAM,GAAY,EAClB,MAAI,GAAK,EAAU,aACV,EAAU,aAEZ,EAAK,EAAU,aAAe,EAAU,aAAe,IAGnD,GAA2B,CACtC,WAAY,EAAA,YACZ,YAAa,MACb,WAAY,IChCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YAsBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAED,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,EAAQ,OAAO,EAAE,QAEjB,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAEjC,GAAI,EAAM,oBAAsB,MAC9B,GAAM,GAAO,EAAM,mBAAmB,KAChC,EAAO,EAAM,mBAAmB,KAEtC,EAAK,MAAQ,EACb,EAAK,MAAQ,EAGf,MAAO,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,OAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,ICzDd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,KAED,EAAQ,EAAA,KAAK,eAAe,EAAM,EAAO,GAAG,OAAO,GACrD,EAAW,EAAA,aAAa,gBAAgB,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtE,GAAI,EAAA,KAAK,cAAc,KAAc,EACnC,MAAO,GAAQ,eAAe,EAAU,EAAO,GAAG,MAAO,IAI3D,GAAM,GAAU,EAAO,OAAO,SAAA,GAAK,MAAA,GAAA,KAAK,cAAc,EAAE,OAAS,IACjE,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAM,GAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAGlC,GAFA,EAAA,aAAa,uBAAuB,EAAQ,GAExC,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAC5D,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAE5D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAC3D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAE3D,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAO,IAEtE,SAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAUT,GAAM,GAAW,EAAQ,IAAI,SAAA,GAC3B,GAAM,GAAY,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAC7C,EAAQ,CAAC,GAAI,GACnB,MAAO,IAAQ,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAO,EAAE,MAAO,CAAC,MAAK,OAIxD,EACI,EAAA,aAAa,gBAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAE7D,GAAM,GAAU,EAAA,KAAK,uBACjB,EAAQ,GAAG,MAAoB,EAAA,KAAK,cAAc,IAEtD,GAAI,EAAS,GAAG,MAAM,KAAO,GAE3B,GAAI,GAAS,EACb,EAAS,QAAQ,SAAA,GACf,GAAM,GAAM,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACjC,EAAO,EAAA,KAAK,cAAc,EAAE,OAElC,EAAQ,IAAI,EAAK,GACjB,GAAU,SAGZ,GAAI,GAAY,EAEhB,EAAS,QAAQ,SAAA,GAKf,OAJM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAErC,EAAO,EAEF,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EAEpC,OADM,GAAS,EAAM,EAAS,GAAK,EAC1B,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EACpC,EAAQ,EAAS,GAAO,EAAM,KAIlC,GAAa,EAAE,MAAM,KAIzB,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtD,EACF,EAAQ,eAAe,EAAe,EAAO,GAAG,MAAO,GAE3D,SAAS,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KAErD,EAGF,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,ICjId,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,SAAC,GAwCX,OAxCY,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OACJ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAE,MAAM,OAEhB,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAChD,EAAa,EAAO,MAAM,OAE1B,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOI,EAAU,EAAA,KAAK,cAAc,GAC7B,EAAU,EAAS,OACnB,EAAa,EAAA,KAAK,kBAAkB,EAAE,MAAO,GAM1C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,IAAO,EAAO,EAAc,EAAQ,KACjC,GAAI,EAAG,GAAI,EAAY,EAAE,IAEhC,OADI,IAAS,OAAO,iBACX,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,GAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAS,EAAA,KAAK,WAChB,CAAC,EAAG,GAAK,GAAK,IAAI,EAAO,EAAA,KAAK,eAAe,EAAE,QAC7C,GAAc,EAAA,KAAK,WACrB,CAAC,GAAG,GAAG,IAAI,EACX,EAAA,KAAK,eAAe,EAAO,QACzB,GAAM,EAAM,IAAU,EAAW,IACvC,AAAI,GAAM,IACR,IAAS,MAMnB,GAAM,IAAc,EAAA,KAAK,WACrB,CAAC,EAAG,EAAM,EAAM,IAAI,EAAS,EAAA,KAAK,eAAe,IACrD,EAAW,IAAe,GAMlC,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAY,EAAE,OAAQ,EAAU,EAAE,OAExD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAU,MAAO,EAAE,SCrG9C,AAsBO,GAAM,IAA+C,CAC1D,WAAY,EAAA,yBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,yBAAwB,QACtC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EAAY,EAAA,KAAK,0BACD,EAAO,MAAO,EAAO,OAOlC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAO,EACP,GAAO,EACF,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAO,GACP,GAAO,MAMjB,EAAU,IAAM,IAAM,IAAM,EAAI,GAAG,GAAM,GAAM,GAMvD,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAO,MAAO,EAAO,OAEhE,MAAO,CAAC,OAAM,GAAE,MAAO,EAAO,MAAO,MAAO,EAAO,SCrHvD,AAsBO,GAAM,IAA8C,CACzD,WAAY,EAAA,wBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,wBAAuB,QACrC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EACF,EAAA,KAAK,0BAA0B,EAAE,MAAO,EAAE,OAOrC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAU,EAAO,EAAK,EAAI,EAC1B,GAAU,EAAO,EAAK,EAAI,EACrB,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAS,GACT,GAAS,MAMnB,EAAU,GAAG,IAAQ,IAAQ,IAAM,EAAI,GAAG,GAAM,GAAM,GAM9D,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAE,MAAO,EAAE,OAEtD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrH7C,AAsBO,GAAM,IACT,GAA6B,SAAC,EAAW,GAAc,MAAA,GAAI,IAClD,GAAM,GAAiB,EAAA,IAAK,IAE5B,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC7Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,IAAM,EAAI,EAAM,KAAK,IAAI,GAAM,IAEnD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC3Bd,AAqBA,GAAM,IAAI,EAAA,aAAa,MACjB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAEX,GAAgB,GACzB,EAAA,IACA,SAAC,GACC,GAAM,GAAO,KAAK,KAAK,GACjB,EAAI,KAAK,IAAI,GACb,EAAI,EAAO,GAAM,GAAI,GAC3B,MAAO,GACF,GACK,OAAK,EAAI,IAAM,EAAK,IAAM,EAAI,IAAM,EAAI,IAAM,EAC/C,KAAK,IAAI,CAAC,EAAI,MAIhB,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Cd,YAmCI,EAAmB,EACnB,GAgBF,OAfM,GAAa,EAAM,MACnB,EAAQ,EAAW,GACnB,EAAW,EAAW,GAEtB,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EAAS,EAAU,mBAAmB,KACtC,EAAS,EAAU,mBAAmB,KAGtC,EAAc,CAAC,EAAO,GACtB,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAA,KAAK,uBAAuB,UAAW,GACpD,EAAa,EAAA,KAAK,uBAAuB,UAAW,GAEjD,EAAI,EAAG,EAAI,EAAO,KAmBzB,OAjBM,GAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAE7B,EAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAG7B,EAAQ,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAG,KAAM,GAAI,QAAS,IAGtD,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,KACP,EAAM,EAAA,aAAa,uBAAuB,EAAM,GAE7C,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAI,EAAA,aAAa,oBAAoB,EAAK,GAChD,EAAW,EAAI,EAAW,GAAK,EAAE,KACjC,EAAW,EAAI,EAAW,GAAK,EAAE,KAGnC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAG3C,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,GAChD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,GAAY,QAAS,IAE1D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,cAIL,EAAmB,EACnB,GACF,GAAM,GAAY,EAAA,KAAK,cAAc,EAAM,OAErC,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAG5D,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAGlE,GAAI,GAAc,IAChB,GAAM,GACF,GAAU,EAAU,EAAU,EAAW,EAAS,GAEhD,EAAc,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAEjD,GAAI,GACF,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,EAAO,MACvD,EACF,EAAW,eAAe,EAAa,UAAW,EAAO,MAEvD,EAAuB,EAAW,eACpC,GAAI,UACJ,EAAA,KAAK,kBAAkB,EAA8B,YACnD,EACF,GAAS,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAS,IAExC,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAW,QAAS,IAEhD,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAe,QAAS,IAGpD,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OACtC,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OAE5C,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,CAAC,KAAM,EAAa,KAAM,GAGnC,MAAO,QAEP,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EACF,GAAyB,EAAM,EAAW,GAE9C,MAAO,GAAA,aAAa,uBAAuB,IAI/C,YAAuB,GACrB,MAAQ,GAAO,EAAO,KAAO,EAI/B,YACI,EAAwB,EAAwB,EAChD,EACA,GACF,GAAI,IAAS,EACX,MAAO,CAAC,KAAM,EAAU,KAAM,GAGhC,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EAAO,EAAO,EAEd,EAAc,EAAA,aAAa,qBAAqB,GAEhD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAa,EAAA,aAAa,oBAAoB,GAE9C,EAAc,EAAW,KACzB,EAAc,EAAW,KAEzB,EAAW,CAAC,EAAY,QAExB,EACF,EAAW,eAAe,EAAU,UAAW,GAC7C,EACF,EAAW,eAAe,EAAU,UAAW,GAE7C,EAAgB,GAClB,CAAC,OAAQ,CAAC,KAAM,EAAa,KAAM,GAAc,QAAS,IAGxD,EACF,GAAU,EAAc,EAAc,EAAM,EAAS,GAEnD,EAAgB,EAAa,KAC7B,EAAgB,EAAa,KAE7B,EAAa,CAAC,EAAc,QAE5B,EACF,EAAW,eAAe,EAAY,UAAW,GAC/C,EACF,EAAW,eAAe,EAAY,UAAW,GAE/C,EAAkB,GAAQ,CAC9B,OAAQ,CAAC,KAAM,EAAe,KAAM,GACpC,QAAS,IAGL,EACF,GAAU,EAAa,EAAa,EAAM,EAAS,GAEjD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAI,EAAA,aAAa,UAAU,EAAM,GACjC,EAAS,CAAC,EAAE,KAAK,QAEjB,EAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAC3D,GAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAE3D,GAAc,GAChB,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,IAAY,QAAS,IAEpD,GACF,GACI,CAAC,OAAQ,CAAC,EAAG,GAAa,EAAG,GAAiB,QAAS,IAGzD,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAErB,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAGrB,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAEV,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAGV,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAC9C,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAEpD,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IAElC,CAAC,KAAM,GAAW,KAAM,IAIjC,YACI,EAAkB,EAAc,GAGlC,OAFM,GAAM,GAAI,cAAa,EAAO,GAE3B,EAAI,EAAG,EAAI,EAAM,KAGxB,OAFI,GAAO,EACP,EAAO,EACF,EAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAI,EAAA,aAAa,SAAS,EAAI,EAAG,EAAM,GACvC,EAAO,EAAA,aAAa,oBAAoB,EAAsB,GACpE,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAC3C,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAE7C,AAAI,GACF,IAAQ,EACR,GAAQ,GAEV,EAAA,aAAa,mBAAmB,EAAK,EAAM,EAAM,GAEnD,MAAO,GCjVT,YAuBoB,GAEX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAO,GAElC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICtDd,AAsBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,MACb,WAAY,SAAC,GAUX,OAVY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GAEX,EAAS,KAAK,MAAM,EAAa,GACjC,EAAS,EAAc,EAAY,EAAY,EAEjD,EAAc,EAAU,GAE5B,GAAI,GAAU,GAAK,EAAS,GAE1B,GAAM,GAAmB,EAAS,EAC5B,EACF,EAAc,EAAY,EAAmB,EACjD,EAAc,EAAU,GAE1B,EAAO,GAAU,EAMzB,GAAM,GAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,EAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SCpErD,YAuBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAM,GAEjC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtDd,AAqBO,GAAM,IACT,GAAgB,EAAA,SAAU,SAAC,GAAO,MAAA,QAAO,SAAS,GAAM,EAAI,GAAG,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,IAAI,KAAQ,SAAW,EAAI,GAAG,QAEzD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,QAAO,MAAM,GAAM,EAAI,GAAG,QAEhD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAK,EAAI,GAAG,QAEvC,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC3Bd,AA2BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAa,EACf,EAAS,EAAE,MACT,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,EAAkB,GACnD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACvD,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC1C,GAAI,GAAgB,MAElB,OADM,GAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAO,EAAa,IAGpC,EAAQ,GAAc,EAAO,EAAQ,EAAE,MAAO,EAAc,GAC5D,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAElD,EAAS,EAGX,GAAiB,EAAG,OACpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGd,EAAa,EAAA,KAAK,cAAc,GAEhC,EAAS,GAAQ,EAAO,EAAY,EAAa,EAAE,OACnD,EAAS,EAAW,MAAM,EAAQ,EAAa,EAAE,OAEnD,EAAW,EACf,GAAI,GAEF,GAAM,GAAW,EAAA,aAAa,qBAAqB,EAAa,GAChE,EAAW,EAGb,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,SCtE9C,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAyBlC,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAY,EAAA,OACd,EAAS,SAAU,EAAE,MACrB,GAAiB,EAAS,EAAE,MAAO,EAAE,MAAO,GAAU,QACpD,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,GAAM,GAAS,EAAwB,EAAuB,EACzD,EAAU,IAAI,EAAG,EAAK,EAAK,GAC1B,GAAS,EAAK,EAAuB,EAErC,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAAQ,IAGvB,EAAG,IAAI,EAAS,EAAG,EAAK,EAAK,GAKrC,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,ICnGd,YAoBI,EAAqB,EAAkB,EACvC,EAA8B,GAChC,GAAM,GAAU,EAAA,KAAK,eAAe,GAC9B,EAAW,GAAK,EAAS,EAAQ,EAAO,EAAS,EAAU,OAC3D,EAAe,GACjB,EAAS,EAAQ,EAAO,EAAU,GAAM,GAE5C,MAAO,CAAC,EAAS,OAAQ,EAAa,QC3BxC,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAa,EACnB,GAAiB,EAAG,qBAEpB,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,CAAC,EAAG,GAAI,GACN,EAAA,GAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGT,EACF,EAAW,MAAM,EAAwB,EAAS,SAAU,EAAE,OAC5D,EACF,EAAW,MAAM,EAAuB,EAAS,SAAU,EAAE,OACjE,MAAO,CACL,CAAC,OAAQ,EAAc,MAAO,EAAS,SAAU,MAAO,EAAE,OAC1D,CAAC,OAAQ,EAAe,MAAO,EAAS,SAAU,MAAO,YC/C/D,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEnB,GAAiB,EAAO,2BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC3C7B,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEnB,GAAiB,EAAO,8BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAG,GAAM,MAAC,KAAM,EAAK,EAAI,IAC/C,GACT,GAAiB,EAAA,SAAU,GAAc,KAAsB,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC9Bd,YAuBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,SAAU,EAAA,EAAA,cAEjB,GAAiB,EAAG,OAEpB,GAAM,GAAW,EAAS,IACtB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAE,MAAM,GAAK,EAAE,KAE9C,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAE5B,EAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAQ,EAAE,MAAM,OAChB,EAAW,EAAA,KAAK,eAAe,EAAE,OAEjC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EACF,EAAA,KAAK,uBAAuB,EAAE,MAA0B,GAE5D,AAAI,IAAkB,GACpB,EAAQ,KAAK,GAGf,OAAS,GAAI,EAAG,EAAI,EAAO,KACzB,GAAM,GAAS,EAAA,KAAK,WAAW,EAAG,EAAO,GACnC,EAAY,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAM,KAC3C,EAAW,EAAA,KAAK,WAAW,EAAW,EAAY,GAExD,EAAQ,GAAY,EAAM,GAG5B,GAAM,GAAQ,EAAQ,MAAM,EAAS,EAAU,EAAE,OAEjD,MAAO,CAAC,OAAQ,EAAO,MAAO,EAAU,MAAO,EAAE,OAG5C,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICnEd,AAqBO,GAAM,IAAuB,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAI,IAE/D,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,MACb,WAAY,SAAC,GAiBX,OAjBY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAmB,IAEnB,EAAY,KAAK,IAAI,GACrB,EAAY,KAAK,IAAI,GACrB,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GACX,EAAI,EAAO,GAGb,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACrD,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACzD,EAAS,KAAK,MAAM,EAAS,GAC7B,EAAS,KAAK,MAAM,EAAS,GAE7B,GAAI,GAAc,EAUlB,GATA,AAAI,MAAO,IAAc,UACvB,CAAI,IAAY,EACd,EAAc,EAEd,EAAc,EAAU,IAKxB,GAAU,GAAK,EAAS,GAAc,GAAU,GAChD,EAAS,GAEX,GAAM,GAAmB,EAAU,GAAa,GAC1C,EAAmB,EAAS,EAC5B,GACF,EAAc,EAAmB,EAAmB,EACxD,EAAc,EAAU,IAG1B,GAAM,IAAS,EAAc,EAAY,EAAY,EACrD,EAAO,IAAU,EAMzB,GAAM,IAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,GAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SC3FrD,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAErD,GAAM,GAAO,KAAK,MAAM,GACxB,MAAI,GAAK,EAAO,GACP,KAAK,MAAM,GACT,EAAK,EAAO,GACd,KAAK,KAAK,GAEb,EAAO,IAAQ,EACV,EAEA,EAAO,IAKP,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICxCd,AAqBA,GAAM,IAAa,EAAA,aAAa,gBAC1B,GAAQ,EAAA,aAAa,WAEd,GAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,IAAM,EACD,GAAQ,EAER,GAAc,MAAK,IAAI,GAAM,KAI3B,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICnCd,AAqBO,GAAM,IACT,GAAgB,EAAA,QAAS,SAAC,GAAO,MAAA,GAAK,GAAI,KAAK,IAAI,CAAC,MAE3C,GAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,GAAK,EACA,GACE,EAAK,EACP,EAEA,IAIE,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IClCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AA0BA,GAAM,IAAU,sBACV,GAAY,KAAK,IAAI,IAAW,EAEzB,GAAqB,GAAgB,EAAA,SAAU,SAAC,GAG3D,GAAM,GAAW,EAAK,CAAC,GAIjB,EAAW,EAAK,GAEhB,EAAO,KAAK,IAAI,GAClB,EAEJ,MAAI,GACF,EAAS,EACJ,AAAI,EACT,EAAS,EAET,EAAS,KAAK,IAAI,EAAM,GAEnB,IAGI,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICtDd,YAwB0B,GAKjB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EACA,EAAA,EAAA,KAEP,GAAiB,EAAG,aAKpB,OAHM,GAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAS,GAAc,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAEvD,EAAS,EAAQ,MAAM,EAAQ,EAAU,EAAE,OACjD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,OAGrC,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,MACb,WAAY,ICpDd,YA0B+B,GAKtB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,WAAY,EAAA,EAAA,SAEnB,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAA,KAAK,cAAc,GAE1B,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAA0B,GAE1B,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,GAAY,WAAW,CACrC,OAAQ,CAAC,EAAC,GACV,QAAO,EACP,MAAO,CAAC,SAAU,EAAkB,cAAe,KAG/C,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EACF,EAAA,aAAa,oBAAoB,EAAQ,MAAO,EAAY,EAAM,IAEhE,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,MAAO,GACrC,EACF,GAAQ,CAAC,OAAQ,EAAe,QAAO,EAAE,MAAO,IAE9C,EAAmC,CAAC,EAAG,GACvC,EACe,CAAC,KAAM,GACtB,EACF,GAAU,CAAC,OAAQ,EAAiB,QAAO,EAAE,MAAO,IAElD,EAAqC,CAAC,EAAG,GACzC,EAAmC,CAAC,MAAO,GAC3C,EAAS,GACX,CAAC,OAAQ,EAAqB,QAAO,EAAE,MAAO,IAElD,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAAqC,CAChD,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,ICvFd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,EACD,EAAa,EACnB,GAAiB,EAAG,UAIpB,OAFM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAY,GAAI,cAAa,EAAO,QACjC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAQ,EAAO,GACrB,EAAU,GAAK,EAAQ,EAEzB,GAAM,GAAS,EAAW,MAAM,EAAW,EAAE,MAAO,EAAE,OACtD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrC7C,AAsBO,GAAM,IAAwB,GAA8B,SAAC,EAAG,GACrE,GAAM,GAAO,EAAI,EACjB,MAAO,GAAO,IAEH,GACT,GAAiB,EAAA,kBAAmB,IAE3B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,IChCd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,EAAI,GACvD,GAAM,GAAY,EAClB,MAAI,OAAM,GACD,IAEA,EAAK,EAAI,EAAI,EAAU,QAIrB,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICjCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,YAyBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAEpB,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,IC5Cd,AA6FA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,0yDCvKjB,AAiBA,GAAM,IAAmD,GAEnD,GAA2C,CAC/C,MAAO,GACP,UAAW,GACX,mBAAoB,GACpB,sBAAuB,GACvB,MAAO,GACP,QAAS,GACT,6BAA8B,gBAI5B,EAAsB,GACxB,GAAS,GAAgB,cAGK,GAC9B,GAAI,CAAE,KAAgB,MACpB,GAAM,GAAS,GAAyB,GACxC,GAAI,IAAW,KACb,GAAS,GAAgB,MAEzB,gBAAQ,IAAI,0CAA2C,GAChD,KAGX,GAAM,GAAK,GAAS,GACpB,MAAI,GAAG,gBACL,OAAO,IAAS,GACT,GAAgB,IAGzB,GAAG,QAAQ,EAAG,YACd,EAAG,QAAQ,EAAG,cACd,EAAG,QAAQ,EAAG,OACd,EAAG,QAAQ,EAAG,QACd,EAAG,QAAQ,EAAG,qBACd,EAAG,QAAQ,EAAG,iBACd,EAAG,OAAO,EAAG,cACb,EAAG,OAAO,EAAG,WACb,EAAG,SAAS,EAAG,MAER,GAAS,IAGlB,YAAsB,GACpB,GAAI,MAAO,kBAAoB,aAAe,IAAiB,EAC7D,MAAO,IAAI,iBAAgB,IAAK,KAC3B,GAAI,MAAO,WAAa,YAC7B,MAAO,UAAS,cAAc,UAE9B,KAAM,IAAI,OAAM,0CAIpB,YAAkC,GAChC,GAAI,IAAiB,GAAK,IAAiB,EACzC,KAAM,IAAI,OAAM,0DAElB,GAAM,GAAS,GAAa,GAM5B,MAJA,GAAO,iBAAiB,mBAAoB,SAAC,GAC3C,EAAG,iBACH,MAAO,IAAS,IACf,IACC,IAAiB,EACX,EAAO,WAAW,QAAS,KAC3B,EAAO,WAAW,qBAAsB,IAG3C,EAAO,WAAW,SAAU,ICxFrC,AAoBA,GAAY,IAAZ,AAAA,UAAY,GAgBV,EAAA,EAAA,MAAA,GAAA,QAkBA,EAAA,EAAA,aAAA,GAAA,iBAlCU,IAAA,IAAa,KAqCzB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,SAAA,GAAA,aAJU,IAAA,IAAY,KAOxB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,yBAAA,GAAA,2BACA,EAAA,EAAA,mBAAA,GAAA,qBACA,EAAA,EAAA,mBAAA,GAAA,uBALU,IAAA,IAAmB,iBAqC3B,EAAc,GAChB,MAAO,CAAC,EAAS,eAIf,EAAoB,GACtB,MAAO,GAAa,cAWW,GAC/B,GAAM,GAAO,EAAA,KAAK,cAAc,GAC1B,EAAe,KAAK,KAAK,EAAO,GACtC,MAAO,GAAA,KAAK,oBAAoB,eA6B9B,EAAc,GAChB,MAAO,CACL,KAAK,IAAI,EAAG,KAAK,KAAK,EAAU,IAAK,KAAK,IAAI,EAAG,KAAK,KAAK,EAAO,iBAKlE,EAAc,GACV,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,GAAI,EAAI,cAsBb,EAA2B,GAE7B,GAAM,GAAQ,EAEV,EACA,EACA,EACA,EACA,EAEA,EACA,EAEA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAsB,EAAM,KAC5B,EAA0B,EAAM,KAChC,EAAgC,EAAM,QACtC,EAA4B,EAAM,QAClC,EAAqB,EAAM,IAC3B,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,EAAM,WAC7B,EAAmB,EAAM,OAEzB,GAAsB,EAAG,KACzB,EAA0B,EAAG,KAC7B,EAAgC,EAAG,KACnC,EAA4B,EAAM,KAClC,EAAqB,EAAG,KACxB,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,GAA6B,KAChD,EAA0B,eAC1B,KACJ,EAAmB,EAAG,OAExB,EAAwB,EAAG,KAEpB,CACL,oBAAmB,EACnB,wBAAuB,EACvB,8BAA6B,EAC7B,0BAAyB,EACzB,mBAAkB,EAClB,sBAAqB,EACrB,0BAAyB,EACzB,mBAAkB,EAClB,qBAAoB,EACpB,iBAAgB,GCzOpB,YAsBgC,EAA2B,GACzD,GAAM,GAAc,IACpB,MAAI,GAAA,MAAM,QAAQ,UAChB,GAAgB,GAEX,EAGT,YAAyB,GACvB,GAAM,GAAQ,EAAG,WACjB,GAAI,IAAU,EAAG,SACf,KAAM,IAAI,OAAM,gBAAkB,GAAqB,EAAI,IAK/D,GAAM,IAAc,QACd,GAAc,kBAEa,GAC/B,MAAI,KAAA,MAAM,QAAQ,iCAAmC,IAAQ,GACxD,GAAc,KAAK,IAAI,IAAQ,KAAK,IAAI,GAAO,gBAOlD,EAA2B,GAC7B,OAAQ,OACD,GAAG,SACN,MAAO,eACJ,GAAG,aACN,MAAO,mBACJ,GAAG,cACN,MAAO,oBACJ,GAAG,kBACN,MAAO,wBACJ,GAAG,8BACN,MAAO,oCACJ,GAAG,cACN,MAAO,oBACJ,GAAG,mBACN,MAAO,6BAEP,MAAO,sBAAsB,eAK/B,EAA2B,GAC7B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,aAAa,IAC1B,cAAgB,EAAgB,gDAIlC,EAA2B,GAC7B,GAAM,GAA4B,GAC9B,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,gBAC7B,wCAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAc,KACrD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAc,EAAG,kBAAoB,GAC7D,cAAQ,IAAI,EAAG,iBAAiB,IAC1B,GAAI,OAAM,oCAElB,MAAO,eAIL,EAA2B,GAC7B,GAAM,GAA8B,GAChC,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,kBAC7B,0CAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAgB,KACvD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAgB,EAAG,kBAAoB,GAC/D,SACI,EAAsB,EAAG,iBAAiB,IACxC,GAAI,OAAM,sCAElB,MAAO,GAGT,GAAM,IAAkB,2BACxB,YACI,EAAsB,GACxB,GAAM,GAAwB,GAAgB,KAAK,GACnD,GAAI,GAAyB,MAC3B,QAAQ,IAAI,wCAAwC,GACpD,QAAQ,IAAI,GACZ,OAWF,OARM,GAAa,CAAC,EAAsB,GAEpC,EAAc,EAAa,MAAM;GACjC,EAAM,EAAY,OAAO,WAAW,OAAS,EAC7C,EAAuB,EAAY,IACrC,SAAC,EAAM,GACH,MAAA,GAAA,KAAK,SAAU,GAAa,GAAG,WAAY,GAAO,IACtD,EAAgB,EACX,EAAI,EAAG,EAAI,EAAqB,OAAQ,IAC/C,EAAgB,KAAK,IAAI,EAAqB,GAAG,OAAQ,GAG3D,GAAM,GAAmB,EAAqB,MAAM,EAAG,EAAa,GAC9D,EAAY,EAAqB,MAAM,EAAa,EAAG,GACvD,EAAkB,EAAqB,MAAM,GAEnD,QAAQ,IAAI,EAAiB,KAAK;IAClC,QAAQ,IAAI,EAAc,MAAM;GAAM,IACtC,QAAQ,IACJ,MAAM,EAAA,KAAK,SAAS,EAAU,GAAI,GAClC,iEACJ,QAAQ,IAAI,EAAgB,KAAK;gBAGL,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGR,EAA2B,GAErD,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,KAClC,EAAG,oBAAoB,EAAS,EAAG,eAAiB,GACtD,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,2DAKhB,EAA2B,GAE7B,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,KACtC,EAAG,oBAAoB,EAAS,EAAG,mBAAqB,GAC1D,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,iDAKhB,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,EAAM,EAAG,eACxD,cAIL,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,KAC9D,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,EAAM,EAAG,eACvD,gBAIP,MAAI,GAAA,MAAM,UAAU,mBAAqB,EAChC,EAEF,cAGqB,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGA,EAAe,GACjD,GAAM,GAAiB,EAAA,MAAM,UAAU,0BACvC,GAAK,GAAS,GAAO,GAAU,GAC7B,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IACrC,KAAM,IAAI,OAAM,0BAA4B,EAAY,gBAE1D,GAAK,EAAQ,GAAoB,EAAS,GACxC,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IAC/B,EAAM,IAAI,EAAc,IAAI,EAAc,IAChD,KAAM,IAAI,OACN,0BAA4B,EAC5B,qDAAuD,EAAM,kBAInC,GAChC,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,qBAAqB,kDAIpC,EAA2B,EAAuB,EAClD,EAAqB,EAA6B,EAClD,GACF,GAAM,GAAM,EAAG,kBAAkB,EAAS,GAC1C,MAAI,KAAQ,GAGH,GAET,IAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GACI,EACA,WAAM,MAAA,GAAG,oBACL,EAAK,EAAqB,EAAG,MAAO,GAAO,EAC3C,KACR,GAAa,EAAI,WAAM,MAAA,GAAG,wBAAwB,KAC3C,gBAIL,EAA2B,EAAuB,GACpD,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,iBAInD,EAA2B,GAC7B,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAInD,EAA2B,EAC3B,GACF,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,mBAAmB,EAAS,IACzC,YAAc,EAAc,yCAI9B,EAA2B,EAC3B,GACF,MAAO,GAAG,mBAAmB,EAAS,eAIpC,EAA2B,EAC3B,EAA8C,GAChD,GAAa,EAAI,WAAM,MAAA,IAAgB,EAAI,EAAS,KACpD,GAAa,EAAI,WAAM,MAAA,GAAG,UAAU,EAAwB,iBAGtB,GACtC,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QAC1D,GAAa,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,UACpE,GAAa,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,sBAIjE,EAA2B,EAC3B,GACF,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,iBAIpE,EAA2B,GAC7B,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,KAAM,iBAGjC,GAClC,GAAM,GAAS,EAAG,uBAAuB,EAAG,aAC5C,GAAI,IAAW,EAAG,qBAChB,KAAM,IAAI,OACN,8BAAgC,GAA2B,EAAI,gBAKnE,EAA2B,GAC7B,OAAQ,OACD,GAAG,kCACN,MAAO,wCACJ,GAAG,0CACN,MAAO,gDACJ,GAAG,kCACN,MAAO,wCACJ,GAAG,wBACN,MAAO,kCAEP,MAAO,iBAAiB,GAI9B,YACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAa,EAAI,WAAM,MAAA,OAC/C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,GAElB,MAAO,GAGT,YAA6B,EAA2B,GACtD,GAAM,GAAiB,EAAG,iCAAmC,EACvD,EAAgB,EAAc,EAAG,SACvC,GAAI,EAAgB,EAAG,UAAY,EAAgB,GACjD,GAAM,GAAmB,2BAA2B,EAAc,IAClE,KAAM,IAAI,OAAM,0BAA0B,EAAgB,kBAIlC,EAAiB,GAAA,MAAA,KAAA,QAAA,GAAA,GACpC,EAAA,KAAK,cAAc,EAAM,MAAM,EAAG,EAAM,OAAS,gBAG9B,GAC1B,GAAI,EAAM,SAAW,EACnB,KAAM,OAAM,wDAGd,MAAO,CACL,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EAAG,EAAM,EAAM,OAAS,gBAI5C,GAC3B,GAAI,GAAsC,CAAC,EAAG,EAAG,GAC3C,EAAW,EAAM,SAAW,GAAM,EAAM,SAAW,GAAK,EAAM,KAAO,EAC3E,MAAK,IACH,GACI,CAAC,GAAY,IAAM,OAAK,GAAY,KAEnC,cAIL,EAAoB,SAAA,AAAA,IAAA,QAAA,GAAA,IACtB,GAAI,GAAa,EAAA,MAAM,UAAU,0BAsBjC,GArBA,AAAI,GACF,GAAa,EAAa,EAO1B,EAAW,EAAS,IAChB,SAAC,EAAG,GAAM,MAAA,IAAK,EAAS,OAAS,EAC7B,EAAA,KAAK,kBAAkB,EAAS,IAChC,EAAS,KAIjB,AAAI,EAAS,SAAW,GACtB,GAAW,CAAC,EAAG,EAAS,MAKxB,EAAS,SAAW,GACtB,GAAM,GAAgB,EAAA,KAAK,aAAa,GACxC,EAAW,EAAc,SAG3B,GAAI,GAAO,EAAA,KAAK,cAAc,GAC9B,GAAI,EAAS,QAAU,GAAK,GAAQ,EAClC,MAAO,CAAC,EAAG,GACN,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,IAAM,EACjB,MAAO,GACF,GACH,EAAS,SAAW,GAAK,EAAS,GAAK,EAAS,IAAM,GACtD,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAI,EAAS,IACvC,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,IAAM,EAC/B,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,IACvC,GACH,EAAS,SAAW,GACpB,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,GAC3C,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAK,EAAS,GAAI,EAAS,IACrD,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,EAC7C,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,GAAK,EAAS,IAE1D,GAAI,GAOF,GAAM,GAAW,GAAY,GACzB,EAAO,EAAG,EAAO,EACrB,MAAI,GAAS,QACX,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,IAET,EAAO,EAAY,GAAO,GAAM,GAAO,GAChC,EAAA,KAAK,oBAAoB,GAAM,IAAI,SAAA,GAAK,MAAA,GAAI,IAErD,MAAO,GAAA,KAAK,oBAAoB,GAIpC,YAAgB,GACd,MAAO,GAAI,IAAM,cAOW,EAAkB,GAI9C,GAHA,EAAS,EAAO,MAAM,IACtB,EAAS,EAAO,MAAM,IAElB,EAAA,KAAK,YAAY,EAAQ,GAC3B,MAAO,GAGT,GAAI,CAAC,EAAO,QAAU,CAAC,EAAO,OAC5B,MAAO,GAGT,GAAI,EAAO,KAAO,GAAK,EAAO,KAAO,GAAK,EAAO,KAAO,GACpD,EAAO,KAAO,EAChB,MAAO,GAGT,GAAI,EAAO,SAAW,EAAO,QAC3B,GAAM,GAAa,EAAO,MAAM,IAAI,GAC9B,EAAa,EAAO,MAAM,IAAI,GACpC,GAAI,IAAe,EACjB,MAAO,GAGT,GAAI,GAAO,IAAe,GAAO,IAC5B,GAAO,KAAO,GAAK,EAAO,KAAO,GACpC,MAAO,GAGX,MAAO,GAAO,KAAO,EAAO,IAAM,GAAO,EAAO,KAAO,GAAO,EAAO,IAMvE,GAAI,IACA,eAEmC,GACrC,GAAI,IAAoB,MACtB,GAAM,GAAK,GAAgB,GAC3B,GAAmB,EAAG,aAAa,EAAG,kBAExC,MAAO,kBAIP,GAAmB,mBAGnB,GAAyB,iBAGY,GACrC,GAAI,IAA0B,MAC5B,GAAM,GAAK,GAAgB,GAC3B,GAAyB,EAAG,aAAa,EAAG,yBAG9C,MAAO,MAAK,IAAI,GAAI,gBAG4B,GAEhD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAI,GACE,EAAK,GAAgB,GAE3B,MAAI,IAAa,EAAI,oCACjB,IAAiB,EACnB,EAAoB,EACf,AAAI,GAAa,EAAI,4BAC1B,EAAoB,EAEpB,EAAoB,EAEf,cAGoB,EAA2B,GACtD,GAAM,GAAM,EAAG,aAAa,GAC5B,MAAO,IAAO,iBAGsB,GACpC,IACE,GAAM,GAAK,GAAgB,GAC3B,GAAI,GAAM,KACR,MAAO,SAEF,GACP,eAAQ,IAAI,qCAAsC,GAC3C,GAET,MAAO,eAG0C,GAEjD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,WAGL,CAAC,GAAa,EAAI,0BACpB,MAAO,GAIX,GAAM,GAAwB,GAAuC,GACrE,MAAO,eAYqC,GAC5C,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,GAET,GAAI,CAAC,GAAa,EAAI,4BACpB,MAAO,QAGT,GAAI,GAAa,EAAI,0BACnB,MAAO,IAAuC,GAGhD,GAAM,GAA0B,8BAChC,GAAI,GAAa,EAAI,IACnB,GAAM,GACF,EAAG,aAAa,GACpB,MAAO,IACH,EAAI,GAGV,MAAO,GAGT,GAAM,GAAwB,GAAuC,GACrE,MAAO,GAGT,YAAgD,GAE9C,GAAM,GAAY,GAAiB,GAE7B,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,oBAAqB,EAAO,EAAQ,EAChE,EAAU,mBAAoB,EAAU,iBAAkB,MAE9D,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,EAGT,YAEI,EAA2B,GAC7B,GAAM,GAAY,GAAiB,EAAI,GACjC,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,wBAAyB,EAAO,EAAQ,EACpE,EAAU,mBAAoB,EAAU,qBAAsB,MAElE,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,cAG2B,GAClC,GAAI,IAAiB,EACnB,MAAO,GAET,GAAM,GAAK,GAAgB,GAGrB,EAAa,EAAW,WAAa,KAC3C,MAAO,eAIL,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MAAG,GAAM,smCC7qBzB,AAqBA,GAAM,IAAM,EAAA,MASZ,GAAI,aAAa,YAAa,WAAM,MAAA,IAAI,UAAU,iBAAmB,IAGrE,GAAI,aAAa,gBAAiB,WAChC,MAAI,IAAsB,GACjB,EACE,GAAsB,GACxB,EAEF,IAIT,GAAI,aAAa,iCAAkC,WAAM,MAAA,KAEzD,GAAI,aACA,yBAA0B,WAAM,MAAA,IAAI,IAAI,mBAAqB,IAGjE,GAAI,aAAa,oBAAqB,WAAM,MAAA,KAG5C,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aAAa,aAAc,WAAM,MAAA,IAAI,QAAQ,eAGjD,GAAI,aAAa,2BAA4B,WAAM,MAAA,IAAI,QAAQ,gBAG/D,GAAI,aAAa,kBAAmB,WAAM,MAAA,IAAI,QAAQ,gBAItD,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aACA,+BAAgC,WAAM,MAAA,IAAI,QAAQ,gBAGtD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aAAa,sBAAuB,WAAM,MAAA,IAAI,QAAQ,gBAG1D,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aACA,yBACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAG/C,GAAI,aACA,+BACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAU/C,GAAI,aAAa,+CAAgD,WAC/D,GAAM,GAAe,GAAI,UAAU,iBAEnC,MAAI,KAAiB,EACZ,EAEF,GAAkC,KAO3C,GAAI,aACA,gDACA,WAAM,MAAA,IAAI,UAAU,gDAAkD,GAClE,CAAC,EAAA,YAAY,aAKrB,GAAI,aACA,+BACA,WAAM,MAAA,IAAmC,GAAI,UAAU,oBAM3D,GAAI,aAAa,+BAAgC,WAC/C,MAAO,IAAI,QAAQ,4BACf,GACA,GAAI,QAAQ,kCAOlB,GAAI,aACA,+BACA,WAAM,MAAA,IAA8B,GAAI,UAAU,oBAGtD,GAAI,aACA,0BACA,WAAM,MAAA,IAAoB,GAAI,UAAU,oBAK5C,GAAI,aAAa,4BAA6B,WAK5C,GAAM,GAAc,GAAI,QAAQ,gCAChC,MAAO,GAAc,EAAI,IAU3B,GAAI,aACA,iCACA,WACE,MAAO,IAET,SAAA,GACE,GAAI,EAAY,GAAK,IAAc,GACjC,KAAM,IAAI,OACN,+DACA,mCAAkC,EAAS,QC1LvD,AAiBO,YAAuB,GAC1B,KAAM,GAAe,GAAI,cAAa,EAAK,QAC3C,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC/B,EAAa,GAAK,KAAK,IAAI,EAAK,IAEpC,MAAO,GCtBX,AAoBO,YAAsC,GACzC,MAAO,CAAC,EAAQ,EAAQ,EAAO,EAAO,KAClC,KAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAC3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAQ,EAAO,OACf,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAC7D,GAAI,EAAe,OAAS,EAAe,SAAW,EAClD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAI5D,QAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACjC,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GACrC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GACtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAC5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,IAG5C,MAAO,CAAC,EAAQ,IClDxB,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAG,IAAM,EAAI,GCnBnE,AAoBO,YAA+B,GAClC,MAAO,CAAC,EAAQ,EAAO,KACnB,KAAM,GAAY,EAAA,KAAK,uBAAuB,EAAO,EAAO,QAC5D,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAU,GAAK,EAAG,EAAO,GAAI,GAEjC,MAAO,IC1Bf,AAmBO,KAAM,IAAW,GAAsB,AAAC,GAAO,KAAK,KAAK,ICnBhE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAiBO,YAAiB,EAAO,EAAY,EAAU,GACjD,KAAM,GAAO,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACnE,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAC/B,KAAM,GAAS,EAAI,EACnB,GAAI,GAAM,EAAM,GAChB,OAAS,GAAI,EAAG,EAAI,EAAY,EAAE,GAC9B,KAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACR,GAAM,GAGd,EAAK,GAAK,EAEd,MAAO,GC9BX,AAmBO,KAAM,IAAe,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBvF,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,EAAI,KAAK,KAAK,ICnBrE,AAkBO,YAAmB,EAAM,EAAO,EAAM,EAAO,GAChD,KAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GACrC,GAAI,GACA,KAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAElD,KAAM,GAAU,EAAA,KAAK,uBAAuB,EAAO,GACnD,OAAS,GAAI,EAAG,EAAI,EAAQ,EAAE,GAC1B,KAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,CAAC,EAAK,IAAM,EAAM,EAAM,IACvC,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEtB,MAAO,GCnCX,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBlF,AAiBO,YAAuB,EAAO,EAAQ,EAAO,EAAM,GACtD,KAAM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GACjC,EAAS,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACrE,OAAS,GAAI,EAAG,EAAI,EAAO,EAAE,GACzB,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAEhC,EAAS,GAAI,OAAM,EAAI,QAC7B,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAO,GAAK,EAAI,EAAK,IAEzB,KAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE7B,MAAO,GCjCX,AAiBO,YAAoB,EAAQ,EAAM,EAAO,GAE5C,KAAM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAwDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GAC/B,OAAS,GAAI,EAAG,EAAI,EAAO,IACvB,EAAS,IAAM,EAAM,GAEzB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACtC,EAAS,IAAM,EAAM,GAIzB,KAAM,GAAiB,GAGjB,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAAgB,GAChB,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EACxD,OAAS,GAAI,EAAG,EAAI,EAAM,GAAQ,KAE9B,GAAI,GACJ,GAAI,EAEA,EAAU,EAAO,GAAG,gBAGpB,KAAM,GAAa,GACnB,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG9C,EAAU,EAAW,KAAK,KAG9B,GAAI,EAAe,KAAa,OAC5B,EAAQ,GAAK,EAAe,QAG5B,KAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAM3B,KAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,KAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,CAAC,EAAoB,KACvC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAM9E,KAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAC7B,CACH,aAAc,EAAa,OAC3B,cACA,WC9IR,AAuBE,GAAA,IAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GCpCF,AAmBA,GAAA,IAAA,WAKE,WAAY,EAAuB,GAHnC,KAAA,YAAwB,GAItB,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,UAAU,EAAQ,SAAS,EAAQ,oBAInD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;yBAEZ,EAAS;;;MAKlC,MAAA,MClDA,AAmBA,GAAA,IAAA,WAOE,WAAY,EAAuB,GALnC,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,SAAS,EAAQ,SAAS,EAAQ,oBAIlD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;wBAEb,EAAS;;;MAKjC,MAAA,MCpDA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAqC,EACrC,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAC9B,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,KAAK,YAAc,CAAC,EAAW,GAC/B,GAAM,GAAU,IAAO,MAAS,IAAM,IAChC,EAAe,EACjB,gBACA,+CAEJ,KAAK,SAAW;;;;;kCAKc,EAAU;;;;;8BAKd,EAAU;wBAChB,EAAY;;0BAEV,EAAM;;;;;;;MAShC,MAAA,MC5DA,YAiB+B,EAAc,GAC3C,MAAO,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAG,GAAI,IAAI,gBAG/C,EAAc,GACxC,MAAI,KAAS,EACJ,CAAC,GAEH,GAAe,EAAM,eAGE,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,KAIT,OADI,GAAS,GACJ,EAAI,EAAG,EAAI,EAAM,IACxB,GAAU,EAAK,GACf,AAAI,EAAI,EAAO,GACb,IAAU,KAGd,MAAO,GCxCT,cAgCE,GAAI,GACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAU,kBACV,EAAY,KACZ,EAAY,MACZ,EAAY,KACZ,EAAY,UACZ,EAAS,cACT,EAAe,wBAMf,EAAmB;;;;;;;;;;;MAcnB,EAAmB,GACnB,EAAc;;;;;;;;;OAWd,GAAU,GACV,EAAY,YACZ,EAAY,UACZ,EAAY,UACZ,EAAY,YACZ,EAAS,eACT,EAAe,GAEf,EAAmB;;;;;;;;MASnB,EAAmB;;;;;;;;;MAUnB,EAAc;;;;;;;;OAWT,CACL,QAAO,EACP,UAAS,EACT,UAAS,EACT,UAAS,EACT,UAAS,EACT,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,iBAAgB,EAChB,YAAW,GClIf,YAyBI,EAAkB,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,SACrC,GAAM,GAAU,EAAA,KAAK,eAAe,GACpC,MAAO,GACF,IAAI,SAAC,EAAQ,GACZ,GAAM,GAAQ,OAAO,EAAO,GAAE,MAAM,EAAK,MAAM,EACzC,EAAQ,IAAM,EAAQ,OAAS,EACjC,OAAO,EAAO,EAAI,GAAE,MAAM,EAAK,MAAM,EAAO,GAAE,MAAM,EACpD,YAAY,EAAO,GAAE,MAAM,EAC/B,MAAU,GAAK,KAAK,EAAK,MAE1B,KAAK,gBA+CuB,GACjC,GAAM,GAAU,EAAA,KAAK,eAAe,GAAO,IAAI,SAAA,GAAK,MAAA,GAAE,aAEtD,MAAO;;wBAEe,EAAQ,GAAE,iBAAiB,EAAQ,GAAE;;EAKtD,GAAM,IAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EC5FpC,AAkBO,GAAA,IAAA,EAAA,aAAA,6BAkBH,EAAyB,EAAwB,EACjD,GACF,GAAM,GAA2B,GACjC,EAAW,QAAQ,SAAA,GACjB,GAAM,GAAO,EAAA,KAAK,cAAc,EAAE,UAAU,cAG5C,AAAI,EAAE,UAAU,UACd,EAAe,KACX,iBAAiB,EAAE,KAAO,GAAO,EAAI,IAAI,EAAI,IAAM,IAAE,KAEzD,GAAe,KAAK,qBAAqB,EAAE,KAAI,KAC/C,EAAe,KAAK,qBAAqB,EAAE,KAAI,QAGnD,GAAM,GAAqB,EAAe,KAAK;GAEzC,EACF,EACK,IAAI,SAAA,GAAK,MAAA,IAAwB,EAAG,EAAa,KACjD,KAAK;GACR,EAAc,EAAY,SAC1B,EAAO,KACP,EAA4B,GAA6B,GAC3D,EACA,EACA,EAAe,GAAgB,GAEnC,AAAI,EAAY,SACd,GACI,GAA+B,EAAY,aAAc,GAC7D,EAA+B,GAA8B,IAE7D,GACI,GAAyB,EAAY,aAAc,GACvD,EAA+B,GAA2B,IAG5D,AAAI,GACF,IAAgB,IAGlB,GAAM,GAAS,CACb,EAAc,EAA2B,EACzC,EAAoB,EAAuB,EAAsB,GACjE,KAAK;GACP,MAAO,GAGT,YAA8B,GAC5B,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAiB,OACrB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,WAEpB,KAAM,IAAI,OACH,EAAM,OAAM,2CAKzB,YAAoC,GAClC,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAuB,OAC3B,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,WAE1B,MAAO,IAAmB,IAIhC,YACI,EAAmB,EACnB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAM,GACV,AAAI,EACF,GAAO,GAA2B,GAElC,GAAO,GAAqB,GAG9B,GAAM,GAAU,EAAO,UAAU,aAC3B,EAAW,EAAa,aAC9B,MAAI,GAAQ,QAAU,EAAS,QAC7B,CAAI,EACF,GAAO,GAA+B,EAAQ,GAE9C,GAAO,GAAyB,EAAQ,IAGrC,EAGT,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAwB,EAAsB,OAClD,GACH,MAAO,IAAwB,EAA8B,OAC1D,GACH,MAAO,IACH,EAAsC,WAE1C,MAAO,IAAwB,EAAU,IAI/C,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAkB,EAAsB,OAC5C,GACH,MAAO,IAAkB,EAA8B,OACpD,GACH,MAAO,IACH,EAAsC,OACvC,GACH,MAAO,IACH,EAA8C,OAC/C,GACH,MAAO,IACH,EAAsD,OACvD,GACH,MAAO,IACH,EACA,WAEJ,KAAM,IAAI,OACH,EAAS,OAAM,4CAI5B,YAAsC,GACpC,MAAO;;eAEM,EAAK,UAAS;;IAK7B,YAAoC,GAClC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAuC,GACrC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAyB,GACvB,GAAM,GAAmB,EAAK,QAAO;;;;MAIjC,EAAK,UAAS;MACd,EAAK,aAAY;;;;;;;;;;;;;;;;;;;;;;;MAuBjB,EAAK,iBAAgB;MACrB,EAAK,iBAAgB;MACrB,EAAK,YAAW;;;;;;;;;;;;;;;;;;;;;;;;;MAyBhB,GAAiB;MACjB,GAAiB;MACjB,GAAiB;IAGrB,MAAO,GAGT,GAAM,IAAoB;;;;;;;;;;;;EAcpB,GAAoB;;;;;;;;EAUpB,GAAoB;;;;;;;;;EAWpB,GAAuB;;;;;;;;;;;EAa7B,cACE,MAAO;;;;IAOT,YACI,EAAiB,GACnB,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,MAAI,GAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAKjD,EAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAK9C;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;IAKlD,YACI,EAAiB,GACnB,MAAI,GAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIvC,EAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIpC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;4BACnC,EAAS,GAAE;;IAKvC,YACI,EAAiC,GACnC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAqB,KAAK,KAAK,EAAM,GAAK,GAC1C,EAAgB,EAAqB,KAAK,KAAK,EAAM,GAAK,GAEhE,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;wBAE1B,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAiC,GACnC,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EAAiB,GAWnB,OAVM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEnD,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EACF,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EAAiB,EACjB,EAAU,GACV,EAAS,UAEJ,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAkB,EAAM,EAAM,OAAS,EAAI,GAC3C,EAAU;aACD,EAAC,cAAc,EAAc;kBACxB,EAAC,MAAM,EAAc;MAC/B,EACJ,EAAS,IAAI,EAAC,KAAO,EAGvB,MAAO;UACC,EAAM,OAAM;;oCAEc,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;QAE1C,EAAO;;wBAES,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;mBAE3B,EAAM,OAAM,IAAI,EAAM;;IAKzC,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,MAAO,GAE3B,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,MAAO,GAEjC,MAAO;;kDAEyC,EAAS,GAAE;+BAC9B,EAAS,GAAE;;iCAET,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,KAAM,MAAO,GAEvC,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EAAyB,GAC3B,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,GAAI,EAAA,KAAK,YAAY,EAAO,GAC1B,MAAO;;8CAEmC,EAAe,GAAE,KACvD,EAAe,GAAE;;MAMvB,GAAM,GAAqB,KAAK,KAAK,EAAM,GAAK,GAWhD,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;;iCAE1C,EAAe,GAAE;6BACrB,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAyB,GAC3B,MAAI,GAAA,KAAK,YAAY,EAAO,GACnB;;0CAE+B,EAAS,GAAE,KAAK,EAAS,GAAE;;MAI/D,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKxC,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKrC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;wBACpB,EAAM,GAAE;4BACJ,EAAM,GAAE;;;IAMpC,YAAkC,GAChC,MAAO,SAAS,EAGlB,YAAgC,GAC9B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAO,KACb,MAAO;WACE,EAAQ;eACJ,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAA0B,GACxB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACzE,GAAI,EAAU,UAAU,UACtB,MAAO,SAAS,EAAQ,cAAc,EAAO,KAEzC,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAChB,GAAI,IAAY,GAAK,IAAY,EAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAK9B,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,KAAK,EAAM;6BAC1B,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAe,GAAE,KAAK,EAAe,GAAE;eAClC,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEzE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;UACZ,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAQ,EAAS,GACjB,EAAQ,EAAS,GAEvB,GAAI,IAAU,GAAK,IAAU,EAC3B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAI,KAAU,EACL;cACG,EAAQ;6CACuB,EAAM,cAAc,EAAK;+BACvC,EAAO;;MAIhC,IAAU,EACL;cACG,EAAQ;wCACkB,EAAM,cAAc,EAAK;+BAClC,EAAO;;MAI7B;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,aAAa,EAAM;6BAClC,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAE/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAO,KACb,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,GAC9C,MAAO;aACE,EAAQ;qDACgC,EAAO,OAAO,EAAO;;iBAEzD,EAAK,UAAS,IAAI,EAAO;;MAKxC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAe,KAAK,KAAK,EAAM,GAAK,GAE1C,MAAO;WACE,EAAQ;iCACc,EAAY,KAAK,EAAe,GAAE,KAC7D,EAAe,GAAE;eACR,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAErC,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,IAC9C,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACzB,MAAO;YACC,EAAQ;mDAC+B,EAAO,OAAO,EAAO;6BAC3C,EAAO;;IAK5B,GAAA,GAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,OACvB,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;qDAC+B,EAAM,GAAE;UACnD,GAAkB,GAAU;;MAKpC,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAS,GAAyB,GACxC,MAAI,KAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;4CACtB,EAAO;6BACtB,EAAO;;IAI9B,IAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;uCAC3B,EAAO;6BACjB,EAAO;;IAK3B;UACC,EAAQ;;wBAEM,EAAM,GAAE,YAAY,EAAM;2BACvB,EAAO,KAAK,EAAO;2BACnB,EAAO;;EAKlC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEzD,GAAI,EAAM,KAAO,GACf,GAAM,GAAgB,EAAM,MAAM,GAC5B,EAAW,CAAC,EAAG,GACf,EAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,IAAK,MAAO,OAC5B,MAAO;UACD,GAA2B,GAAa;eACnC,EAAQ;mBACJ,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAM,GAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpC,EAAgB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAO,KAAK,EAAO,KAAK,EAAa,KAAK,EAAY;eACjD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAAK,EAAM,GAC3B,EAAU,EAAM,GAEhB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,SAC9B,MAAO;UACD,GAAqB,GAAa;gBAC5B,EAAQ;mBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO;UAC1C,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAa,EAAU,UAAU,WACvC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;gBACK,EAAQ;;oDAE4B,EAAO;;4BAE/B,EAAO,OAAO,EAAO;iCAChB,EAAO;;QAKtC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;YACC,EAAQ;8CAC0B,EAAM,GAAE;;qDAED,EAAO,OAAO,EAAO;6BAC7C,EAAO;;IAKlC,GAAM,GAAS,GAAyB,GACxC,MAAO;cACK,EAAQ;;4BAEM,EAAO,YAAY,EAAO,cAAc,EAAM;+BAC3C,EAAO,KAAK,EAAO;+BACnB,EAAO;;IAKtC,YAA4B,GAe1B,OAdM,GAAQ,EAAU,UAAU,aAC5B,EAAO,EAAM,OACb,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC7C,EAAgB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC3D,EAAS,0BACT,EAAQ,OAAO,EAAa,kBAAkB,EAAY,eACrD,EAAI,EAAG,EAAI,EAAO,EAAG,IAC5B,EAAS,QAAQ,EAAC,KAAO,EACzB,GAAiB,EAAM,EAAO,EAAI,GAClC,EAAQ,IAAI,EAAC,MAAM,EAAa,MAAQ,EAE1C,GAAM,GAAO,KACb,MAAO;WACE,EAAQ,IAAI,EAAM;oBACT,EAAK;2BACE,EAAO;kCACA,EAAO;qDACY,EAAO,KAAK,EAAO;eACzD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,UACvC,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO,KAAK,EAAO;UACtD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;uBAIC,EAAO,KAAK,EAAO;;0BAEhB,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;gCAEU,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGvC,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO;oBAChC,EAAO;6BACE,EAAO,KAAK,EAAO,aAAa,EAAM;6BACtC,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,UACjD,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;iBAGL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;UAElD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;gCAGU,EAAO,KAAK,EAAO,KAAK,EAAO;;0BAErC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAKpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGxB,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAM;6BACpB,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEnE,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,SAAU,UAC3D,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;;iBAEL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAM,GAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAE3B,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;mBAGzC,EAAO;UAChB,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;0BAGlC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBACzC,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE;iBACnB,EAAM,GAAE;;;yBAGA,EAAO,OAAO,EAAO;+BACf,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;;0BAGM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAO,eAAe,EAAM;6BAC1C,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAA2B,GACzB,GAAM,GAAU,EAAU,KACpB,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAEtD,MAAI,GAAS,EACJ,UAAU,EAAO,IAEnB;0BACiB,EAAM;;iBAEf,EAAO;;;IAMxB,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAEpC,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAE7C,EAAO,GAAkB,GACzB,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,AAAI,EAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAGpC,GAAI,GAAS,sBACP,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAChD,EAAgB,IAAW,EAC3B,EAAU,EAAA,KAAK,cAAc,EAAa,cAC1C,EAAiB,IAAY,EAEnC,GAAI,IAAW,GAAK,CAAC,GAAiB,CAAC,EACrC,EAAS;;cAGA,GAAiB,CAAC,EAC3B,AAAI,IAAY,EACd,EAAS;;QAIT,EAAS;;gBAIF,EAAc,QACvB,GAAM,GAAO,EAAS,EAChB,EAAO,EAAS,EAEtB,AAAI,EAAc,QAAQ,GAAQ,IAAM,EAAc,QAAQ,GAAQ,GACpE,EAAS,8BACJ,AAAI,EAAc,QAAQ,GAAQ,GACvC,EAAS,2EAEJ,AAAI,EAAc,QAAQ,GAAQ,IACvC,GAAS,gDAIb,MAAO;WACE,EAAQ;QACX,EAAI;QACJ,EAAa;8BACS,EAAc,IAAI,EAAqB;QAC7D,EAAM;;IAKd,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAc,EAAa,SAC3B,EAAa,EAAU,UAAU,SACjC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAE1C,GAAI,CAAC,EAAU,UAAU,WAAa,IAAW,GAC7C,EAAU,UAAU,YAAc,MAClC,EAAA,KAAK,YAAY,EAAY,GAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAKpC,GAAM,GAAO,GAAkB,GACzB,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAC7C,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,MAAI,GAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAG7B;YACG,EAAQ;QACZ,EAAI;QACJ,EAAa;kBACH,EAAc,IAAI,EAAqB;;gBAKvB,GAChC,GAAI,GAAQ,EACV,MAAO,MACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QAEP,KAAM,OAAM,gBAAgB,EAAI,yBAKpC,YACI,EAAmB,GAErB,GAAM,GAA0B,KAAK,MAAM,KAAK,UAAU,IAC1D,SAAa,UAAU,aAAe,EAC/B,EAGT,YAA2B,EAAkB,GAC3C,MAAO,GAAS,IAAI,SAAA,GAAK,MAAA,GAAO,KAAI,KAAK,MCt2C3C,AAuBA,GAAA,IAAA,WAOE,WACI,EAAiB,EAAoB,EACrC,GARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,EAAA,KAAK,OACD,EAAM,OAAS,EACf,WAAM,MAAA,aACF,GAAG,OAAO,GAAG,cACb,EAAG,MAAM,IAAE,6CACnB,GAAM,GAAS,EAAM,EAAM,OAAS,GAC9B,EAAU,KAAK,KAAK,EAAS,GACnC,KAAK,YAAc,EAAM,MAAM,EAAG,IAClC,AAAI,EAAU,GACZ,KAAK,YAAY,KAAK,GAExB,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,GAAM,GAAW,KAAK,YAChB,EAAO,EAAS,OAChB,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAEjC,EACA,EACJ,GAAI,IAAY,GACd,EAAa,EAAO,EACpB,GAAM,GAAiB,GAAkB,GACzC,EAAiB;UACb,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE,QAEtB,GAAa,EACb,EAAiB;UACb,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE,IAExB,GAAM,GAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACnD,EAAY,IAAM,EAAS,EAAa,GACxC,EAAc,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IACzC,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAE/C,EAAU,IAAO,MAAS,cAAgB,WAC1C,EAAoB,EAAY,GAAK;sDACO,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM,OAE7D,EAAa;0BACG,EAAW,OAAM;uCACJ,EAAW,OAAM;uCACjB,EAAW,OAAM;qDACH,EAAW,OAAM,UAE5D,EAAgC,EAAY,GAAK;qCACtB,EAAY,OAAM;4CACX,EAAS,OAAM;iDACV,EAAS,MAAM,IAAI,OAAM;SAGtE,KAAK,SAAW;0BACM,EAAY,OAAM;iCACX,EAAS,OAAM;sCACV,EAAS,MAAM,IAAI,OAAM;;QAEvD,EAA6B;;UAE3B,EAAK;4BACa,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;4BAC5C,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;UAC9D,EAAc;yCACiB,EAAS,eAAe,EAAS;sBACpD,EAAS,eAAe,EAAS,OAAO,EAAU;;;2BAG7C,EAAU;;8BAEP,EAAU;;YAE5B,EAAiB;6BACA,EAAU;;;mBAGpB,EAAM;;;;;;;;;;;MAazB,MAAA,MCjJA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAe,GAE1C,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;0CACT,EAAa;;;;;;;;;;;;;;gCAcvB,EAAqB;oBACjC,EAAc;gDACc,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;mBACnC,EAAa;kDACkB,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;;;;;;;MAevD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAc,EAAe,GAExD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAa;;;;;;;;;;;;;;;;;gCAiBvB,EAAoB;oBAChC,EAAa;gDACe,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;;;;;;;;MAgBzD,MAAA,MC3KA,AAwBO,GAAM,IAAmB,CAC9B,KAAM,wCACN,KAAM,yCAGR,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,QAAS,QAAS,QAAS,SAK1C,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GAEnE,KAAK,SAAW;;;UAGV,EAAE;;;;;;;;;;MAYZ,MAAA,MCpDA,AAqBA,GAAM,IAAoB;;;EAKb,GAAM,gBACN,GAAM,gBACN,GAAM,gBAMN,GAAU;;;;;;;;;;EAYV,GAAM;;;;;;;;;EAYN,GAAQ,wBAER,GAAY,wBAEZ,GAAO,uBAEP,GAAa,wBAEb,GAAU,uBAEV,GAAgB,wBAEhB,GAAc,sCAEd,GAAa,sCAEb,GAAM,GAAoB;;EAG1B,GAAM,GAAoB;;EAG1B,GAAM;qBAGN,GAAU,yCAEV,GAAQ,+BAErB,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;MAUZ,MAAA,MC1GA,AAuBA,GAAM,IAAoB;;;;;EAOb,GAAU;;;;;;;;;;;;;;;;;;;;;EAuBV,GAAM;;;;;;;;;;;;;;IAef,GAAoB;;EAIX,GAAQ;;;EAKR,GAAU;;;EAKV,GAAQ;;EAIR,GAAY;;EAIZ,GAAO;;EAIP,GAAa;;EAIb,GAAU;;EAIV,GAAgB;;EAIhB,GAAc;;;;EAMd,GAAa;;;;;EAOb,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIxB,GAAA,WAQE,WACI,EAAY,EAAkB,EAC9B,GAAA,AAAA,IAAA,QAAA,GAAA,IATJ,KAAA,cAAgB,CAAC,IAAK,KAGtB,KAAA,qBAAuB,GACvB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,GAAM,GAAO,KAAK,YAAY,OAC1B,EAAyB,GAC7B,GAAI,EACF,GAAI,IAAS,GAAK,EAAA,KAAK,cAAc,KAAK,eAAiB,EACzD,EAAyB;;;;eAMzB,GAAM,GAAQ,GAAkB,GAIhC,GAHA,EAAyB;YACrB,EAAK;UAEL,IAAS,EACX,GAA0B;yCACK,KAAK,YAAY,GAAE;;;iBAKlD,GAAM,GAAW,GAAY,SAAU,GACvC,GAA0B;;iBAEnB,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;iBAExD,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;;;aASrE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;UAQF,EAAsB;;;;MAMhC,MAAA,MC9MA,AAoBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,KASf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAgBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KCzDA,AAoBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;MAiBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KC5DA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,OAAQ,QAKvB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAepB,MAAA,MCzCA,AAoBA,GAAA,IAAA,WAME,WAAY,GAJZ,KAAA,YAAwB,GAKtB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAI1C,OADM,GAAW,CAAC,YAAY,EAAQ,GAAE,+BAC/B,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAC1B,EAAS,KACL,iBAAiB,EAAQ,GAAE,KAC3B,kBAAiB,EAAC,WAAW,EAAK,QAExC,GAAM,GAAY,EAAQ,OACpB,EAAY,EAAQ,EAAQ,OAAS,GAC3C,EAAS,KAAK,sBAAsB,EAAS,WAAW,EAAS,OAEjE,KAAK,SAAW;;;;;;UAMV,EAAS,KAAK;WAAa;;MAIrC,MAAA,MCzDA,AAuBA,GAAA,IAAA,WAOE,WAAY,EAAoB,GALhC,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,GAAM,GAAQ,KAAK,YACb,EAAO,EAAM,OACb,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAC/B,EAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACzD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAW1C,OARM,GAAU,EAAS,GACnB,EAAe,EAAS,MAAM,IAC9B,EAAc,EAAS,OAEzB,EAAkB,OAAO,EAAO,MAAM,EAAQ,GAAE;;oBAEpC,EAAW,WAAW,EAAa,OAAM;WAEhD,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAK1B,GAAmB;cACX,EAAO,MAAM,EAAQ,GAAE,QAAQ,EAAO,OAAO,EAAQ,EAAI,GAAE;;kBAEvD,EAAC,IAAI,GAAgB,EAAU,EAAS,GAAM;mBAC7C,GAAgB,EAAc,EAAS,GAAM;WAG5D,GAAM,GAAY,EAAQ,OACpB,EAAQ,EAAQ,EAAQ,OAAS,GACvC,GAAmB;;gBAEP,EAAS,IAAI,GAAgB,EAAU,EAAS,GAAM;iBACrD,GAAgB,EAAc,EAAS,GAAM,MAE1D,KAAK,SAAW;uBACG,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IAAE;UAC1C,EAAe;;;;UAIf,EAAK;sCACuB,EAAM;;UAElC,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;cACrC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;;MAMtC,MAAA,MAaA,YAAyB,EAAoB,EAAiB,GAC5D,GAAM,GAAa,EAAS,QAAQ,GAC9B,EAAM,EAAS,IAAI,SAAC,EAAG,GAC3B,MAAI,KAAQ,EACA,EAAC,MAAM,EAEV,IAGX,MAAO,GAAI,OC3Hb,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAiB,EAAS,aAAe,eAE/C,KAAK,SAAW;;;;;;;;;;;;8BAYU,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;oBAIhC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAExC,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;wCAEI,EAAM,aAAa,EAAM;;;;;;;gCAOjC,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;oCAEF,EAAS,YAAW;;oBAEpC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAEjC,KAAK,SAAW;;;;;;;;;;;8BAWU,EAAS,UAAS;kCACd,EAAS,SAAQ;iCAClB,EAAW,MAAM,EAAQ;;kCAExB,EAAS,QAAO;;;;oCAId,EAAS,UAAS;mCACnB,EAAY,MAAM,EAAM;;oCAEvB,EAAS,SAAQ;;;;sCAIf,EAAS,SAAQ;qCAClB,EAAW,MAAM,EAAO;;sCAEvB,EAAS,QAAO;;;;;;;;;;;;;MAetD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAEnD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAchC,EAAW;gDACK,EAAW;;oCAEvB,EAAS,SAAQ;;;;;yBAK5B,EAAW;;kCAEF,EAAY;kDACI,EAAY;;sCAExB,EAAS,UAAS;;;;;;2BAM7B,EAAY;;oCAEH,EAAW;oDACK,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;6BAM5B,EAAW;;sCAEF,EAAS,YAAW;;;;;;;;;;MAY1D,MAAA,MCzSA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;;;;;;;wBAOI,EAAU;;;;;8BAKJ,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;;;;;;;;;MAcpD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;gCAYnB,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;;oCAGF,EAAU;8BAChB,EAAU;;;;;;;;;MAWxC,MAAA,MCzIA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAChD,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAEpC,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;;2BAGT,EAAM,aAAa,EAAM;;;;;;;gCAOpB,EAAY;qCACP,EAAc;;gCAEnB,EAAS,SAAQ;;;;kCAIf,EAAW;uCACN,EAAa;;kCAElB,EAAS,QAAO;;;;oCAId,EAAqB;;;;;;;;oBAQrC,EAAc;;;;;;;;;;;;;;;;;;;kBAmBhB,KAA4B,GAAC;;oBAE3B,EAAc;;0CAEQ,EAAqB;mCAC5B,EAAqB;;;kCAGtB,EAAqB;mCACpB,EAAqB;;;yBAG/B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;;;;;yBAK5B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;gCACrB,EAAqB;;;;;;;;;;UAU3C,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAEtD,KAAK,SAAW;oCACgB,EAAW,KAAK,EAAY,KACxD,EAAW;iCACc,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;gCAgBhC,EAAW;qCACN,EAAa;;gCAElB,EAAS,QAAO;;;;kCAId,EAAY;uCACP,EAAc;;kCAEnB,EAAS,SAAQ;;;;oCAIf,EAAW;yCACN,EAAa;;oCAElB,EAAS,QAAO;;;;sCAId,EAAqB;;;;;;;;;;;;;;;;;oBAiBvC,KAA4B,GAAC;;4CAEL,EAAqB;qCAC5B,EAAqB;2BAC/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;;;2BAG/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;qCACrB,EAAqB;;;;;;;;;MAW1D,MAAA,MC3TA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAa,EAAS,YAAc,EAAS,WAE/C,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;wBAO3B,EAAU;4BACN,EAAU;;;;;;;;;gCASN,EAAY;qCACP,EAAc;;gCAEnB,EAAQ;;;;kCAIN,EAAW;uCACN,EAAa;;kCAElB,EAAQ;;;;;;;;;;;UAWhC,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MCrHA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAR/B,KAAA,cAAgB,CAAC,IAAK,KACtB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,SAgB5B,OAdM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAEjB,EAAW,gCAEN,EAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY;wBACI,EAAC,IAAI,EAAI,EAAC;mBACf,EAAC,IAAI,EAAC;mBACN,EAAC,IAAI,EAAC,eAYrB,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAS,EAAG,EAAS,EAAc,KAC1C,GAAM,GAAI,EAAS,EAOnB,GALA,GAAY;4BACQ,EAAI,EAAc;4BAClB,EAAI,EAAa;UAGjC,IAAgB,GAClB,GAAI,EAAI,GAEN,CAAI,EAAU,IAAM,EAUlB,GAAY;;qCAEW,EAAQ,mCAC3B,EAAQ;2BACC,EAAC,IAAI,EAAC;;;;uCAIM,EAAQ;6BAClB,EAAC,IAAI,EAAC;;;2BAGR,EAAC,IAAI,EAAC;;;;qCAII,EAAQ,mCAC3B,EAAQ;;;;;uCAKa,EAAQ;;;;sBAIzB,EAAC,IAAI,EAAC,+BAA+B,EAAC,IAAI,EAAC;;sBAE3C,EAAC,IAAI,EAAC,wBAAwB,EAAC,IAAI,EAAC;;gBAK5C,GAAY;qCACW,EAAQ,uBAAuB,EAAQ;2BACjD,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;oBAGb,EAAC,IAAI,EAAC,aAAa,EAAC,IAAI,EAAC;gBAI7B,EAAI,EAAI,IAOV,GAAM,GAAkB,EAAU,IAAM,EACpC,EAAA,KAAK,kBAAkB,GACvB,EAEJ,AAAK,EAAgB,IAAM,GAAK,EAAU,IAAM,GAC3C,EAAgB,IAAM,GAAK,EAAU,IAAM,EAC9C,IAAY;oCACQ,EAAU,EAAC,MAAM,EAAe;;uCAE7B,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;kBAMvB,AAAI,EAAgB,GAClB,IAAY;;yCAEW,EAAQ;oDACG,EAAQ;+BAC7B,EAAC,IAAI,EAAC;;+BAEN,EAAC,IAAI,EAAC;;qBAKrB,GAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;mBAG5C,GAAY;oCACQ,EAAe;;uCAEZ,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;;sBAGjB,EAAC,IAAI,GAAI,GAAC,aAAa,EAAC,IAAI,GAAI,GAAC;uBAM7C,AAAI,GAAI,GACN,IAAY;mCACW,EAAQ;cAS/B,AAAI,EAAU,IAAM,EAClB,IAAY;sCACY,EAAW;iDACA,EAAQ;2BAC9B,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;6CAGY,EAAQ;2BAC1B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;;wCAEY,EAAW;mDACA,EAAQ;;;sBAGrC,EAAC,IAAI,GAAI,GAAC,kBAAkB,EAAC,IAAI,GAAI,GAAC;oBAI9C,IAAY;qCACW,EAAQ;2BAClB,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;kCAGC,EAAW;iDACI,EAAQ;2BAC9B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;oBAKhD,GAAY,KAIhB,AAAI,EAAI,GACN,IAAY;0BACI,EAAC,IAAI,EAAC,WAAW,EAAC,KAAK,EAAC;gBAClC,EAAC,IAAI,EAAC,kBAAkB,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,EAAC;YAGzD,AAAI,EAAI,EAAI,GACV,IAAY;4BACI,EAAC,IAAI,GAAI,GAAC,WAAW,EAAC,KAAK,GAAI,GAAC;kBAC1C,EAAC,IAAI,GAAI,GAAC;8BACE,EAAC,IAAI,GAAI,GAAC,eAAe,EAAC,IAAI,GAAI,GAAC,UAM7D,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY,gBAAgB,EAAC,IAAI,EAAC,QAAQ,EAAC,IAAI,EAAC,IAIpD,GAAI,GAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;UAezC,EAAQ;;;UAGR,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MC3UA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAA4B,EAC5B,GAPJ,KAAA,cAAgB,CAAC,QAAS,QAAS,UACnC,KAAA,YAAwB,GAOf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAChC,EAAA,EAAA,GACA,EAAA,EAAA,GAAY,EAAA,EAAA,GACnB,KAAK,YAAc,CAAC,EAAU,EAAY,EAAW,GACrD,GAAM,GAAW,IAAW,WAAa,EAAI,EAEvC,EAAA,CAAA,EAAA,EAAA,KAAA,EAAA,EAAA,MAAC,EAAA,EAAA,GAAkB,EAAA,EAAA,GAGnB,EAAA,EAAA,oHAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAa,EAAA,EAAA,GAW3B,EAAA,EAAA,kHAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAe/B,KAAK,SAAW;yCACqB,EAAW;wCACZ,EAAU;;;;;;;;;;;;;;;;iCAgBjB,EAAK;;;;+BAIP,EAAW;8BACZ,EAAU;;uBAEjB,EAAG;mCACS,EAAgB;4BACvB,EAAkB;;;uBAGvB,EAAG;mCACS,EAAe;4BACtB,EAAkB;;;;;aAKjC,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;MA0BrB,MAAA,MC1GA,GAAA,WAQE,WAAY,EAAiB,EAAoB,GAPjD,KAAA,cAAgB,CAAC,KAQf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAM,OACb,EAAM,EAAY,MAAQ,QAAQ,GAAU,EAAM,UAAS,IAC3D,EAAS,EAAM,EAAM,OAAS,GAChC,EAAY,GACZ,EAAY,GAIhB,AAAI,EACF,GAAY,EAAU,UAAU,GAAS,GAAM,WAC/C,EAAY,EAAU,UAAY,WAElC,GAAY,EAAU,gBAAgB,EAAW,cACjD,EAAa,EAAU,aAAe,cAGxC,KAAK,SAAW;;;UAGV,GAAkB,GAAK;oBACb,GAAc,EAAM,UAAS;sBAC3B,EAAG;;cAEX,EAAS;sBACD,EAAS;YACnB,GAAc,EAAM,UAAS;wBACjB,GAAU,EAAM,UAAS;;;;MAO/C,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,OAAS,MAChB,GAAK,MAAQ,EAAM,mBAAmB,EAAc,UAEtD,EAAM,GAAG,UAAU,EAAK,MAAO,KAGrC,KAEA,YAAmB,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,KACpB,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,KAC/B,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,OAAO,EAAI,KAE/C,KAAM,OAAM,2BAA2B,EAAI,yBAI/C,YAAuB,EAAc,GACnC,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KAEd,KAAM,OAAM,2BAA2B,EAAI,yBCjG/C,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAEjB,KAAA,aAAe,GACf,KAAA,aAAe,GAEf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAuB,EAAmB,GAP9C,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAOtB,KAAK,YAAc,EACnB,KAAK,UAAY,EACjB,KAAK,WAAa,EAClB,KAAK,SAAW;;;;gBAIJ,KAAK,uBAAsB;gBAC3B,KAAK,sBAAqB;gBAC1B,KAAK,sBAAqB;;uBAEnB,EAAS;+BACD,EAAS;uBACjB,EAAS;+BACD,EAAS;mCACL,EAAS;UAClC,KAAK,qBAAoB;;;uBAGZ,KAAK,yBAAwB;;;IAM1C,SAAA,UAAA,qBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,mBAAR,WACE,MAAI,MAAK,aAAe,OACf,KAAK,YAAY,GAEjB,KAAK,YAAY,IAIpB,EAAA,UAAA,uBAAR,WACE,MAAI,MAAK,aAAe,OACf,4BAEA,6BAGb,KC5FA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,CAAC,EAAM,GAC1B,KAAK,SAAW;;;;;;MAQpB,MAAA,MClCA,AAsBA,GAAA,IAAA,WAME,WAAY,GALZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;UAIlB,EAAK,OAAM;;MAIrB,MAAA,MCxCA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;;UAKlB,EAAK,OAAM;;MAIrB,MAAA,MC3CA,AAqBA,GAAA,IAAA,WAKE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IANJ,KAAA,cAAgB,CAAC,KAOf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAS,SACb,AAAI,GACF,GAAS,8BAGX,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;8BAUrB,EAAK;kCACD,EAAK;iDACU,EAAK,OAAO,EAAM;wBAC3C,EAAK,UAAS;;;;;;;;;;;;;;UAc5B,EAAK,OAAM,WAAW,EAAM;;MAItC,MAAA,MCtEA,AAqCA,GAAA,IAAA,WAOE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAW,GACX,EAAS,SACb,AAAI,GACF,GAAS,8BAGX,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAC1B,GAAM,GAAU,EAAM,EAAI,EAE1B,GAAY;;gCAEY,EAAG,MAAM,EAAY,GAAE;gCACvB,EAAG;kCACD,EAAG,MAAM,EAAY,GAAE;kCACvB,EAAG;;;;;;;gCAOL,EAAK;oCACD,EAAK;kDACS,EAAK,OAAO,EAAM;yBAC3C,EAAK,UAAS;;;yBAGd,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;;;UAQ5B,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;;UAWzC,EAAQ;;UAER,EAAK,OAAM,MAAM,EAAM;;MAIjC,MAAA,MC/GA,AAmBO,GAAM,IAAc,CACzB,KAAM,oCACN,KAAM,qCAGR,GAAA,WAKE,WAAY,EAAY,EAA8B,GAJtD,KAAA,cAAgB,CAAC,OAAQ,QAKvB,GAAM,GAAW,EAAW,GAC5B,KAAK,YAAc,EAEnB,GAAM,GACF,EAAU,SAAS,KAAK,GAAO,UAAU,KAAK,GAC5C,EAAoB,EAAa,EAAQ,KAAO,MAEtD,KAAK,SAAW;yCACqB,EAAyB;;;UAGxD,EAAE;;;;kDAIsC,EAAQ;;;;;;8BAM5B,EAAQ;;;;;;;;;yDASmB,EAAiB;;;;;;;;;;MAY1E,MAAA,MCxEA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAiB,GAL7B,KAAA,YAAwB,GAMtB,KAAK,cAAgB,CAAC,KACtB,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;;MASlB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,UAEhE,EAAM,GAAG,UAAU,EAAK,SAAU,KAGxC,KChDA,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,EAAuB,GALrD,KAAA,cAAgB,CAAC,IAAK,WAMpB,GAAM,GAAwB,EAAO,QACrC,EAAY,GAAQ,EACpB,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,EAAQ,GAE7C,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,EAAkB,GACzC,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,mBAAmB,EAAI,yBAErC,GAAI,IAAS,EACX,MAAO,yBAMT,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAElD,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,IAAM,EACR,EAAa,KAAK,kBAAkB,EAAc,GAAE,MAEpD,EAAa,KAAK,GAAG,EAAc,IAGvC,MAAO,GAAa,OC3CtB,GAAA,IAAA,WAIE,WACY,EAA0B,EAAmB,GAA7C,KAAA,SAAA,EAA0B,KAAA,QAAA,EAJtC,KAAA,cAAgB,CAAC,IAAK,WAKpB,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAChC,EAAe,KAAK,SAAW,EAAI,aAAe,UACxD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,KAAK,QAAO;;YAElD,EAAK;;gCAEe,KAAK,SAAQ;;sCAEP,EAAY;;;;QAMlD,MAAA,MC1CA,YAwBmC,GACjC,GAAM,GAAO,KACP,EAAwB,EAAK,QAAO;;MAEtC,EAAK,UAAS;MACd,EAAK,UAAS;MACd,EAAK,UAAS;;;;;OAMlB,MAAO,IAA8B,EAAI,eAGR,GAEjC,GAAM,GAAc,GAAI,cACpB,CAAC,GAAI,EAAG,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,GAAI,EAAG,EAAG,IAClE,MAAO,IAAoC,EAAI,eAGf,GAEhC,GAAM,GAAwB,GAAI,aAAY,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,IAC9D,MAAO,IAAmC,EAAI,GAGhD,YACI,EAA2B,EAAe,EAC1C,EAAwB,EACxB,GACF,GAA+B,EAAO,GACtC,GAAM,GAAU,GAAyB,GAEnC,EAAQ,EAAG,WACjB,UAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAO,KACxD,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAO,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAC5C,EAAa,QACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,QACzD,cAIL,GACF,MAAO,GAAc,gCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAG,mBAIvC,GACF,MAAO,GAAc,oCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAc,kCAIlD,GACF,MAAO,GAAc,kCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAG,2BAIL,GACF,MAAO,GAAc,sCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EAAQ,GAAwC,GAC3D,EAAG,KAAM,EAAG,mBAId,GACF,MAAO,GAAc,0CAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAc,kCAIhB,EAA2B,EAC3B,GACF,GAAM,GAAY,EACZ,EAAW,EAAI,EACf,EAAU,EAAI,EAAM,EAAI,EAC9B,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KAC7C,GAAM,GAAU,GACZ,EAAI,EAAS,eAAgB,EAAc,EAAG,EAAQ,GAC1D,MAAO,IACH,GACI,EAAI,EAAS,KAAM,EAAc,EAAG,EAAQ,eAIlD,EAA2B,EAAuB,EAClD,EAAgB,EAAkB,GACpC,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAEhE,GAAI,GAA2B,EAAuB,EACtD,AAAI,YAAgB,YAClB,GAAgB,GAAI,YAAW,EAAQ,EAAS,GAChD,EAAgB,EAAG,cACnB,EAAiB,EAAG,MAEpB,GAAgB,GAAI,cAAa,EAAQ,EAAS,GAClD,EAAgB,EAAG,MACnB,EAAiB,EAAc,2BAGjC,EAAc,IAAI,GAElB,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAAG,KACvD,EAAe,KAEvB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA2B,EAC3B,GAEF,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAChE,AAAK,EAAqB,eAAgB,YACxC,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAO,MAAO,EAAO,OAAQ,EAAG,EAAG,KAC9D,EAAG,cAAgB,EAAqB,QAEhD,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAG,KAAM,EAAG,cACvC,KAIV,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA6B,EAAc,EAC3C,GAEF,GAAM,GAAS,EAAI,eACnB,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,KAGrD,GAAM,GAAgB,EAChB,EAAiB,EACjB,EAAkB,EAAgB,EAAiB,EAAO,EAEhE,UACI,EACA,WAAM,MAAA,GAAI,WACN,EAAI,kBAAmB,EAAiB,EAAI,eAIpD,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAG,EAAG,EAAS,EAAM,EAAI,KAAM,EAAI,MAAO,KAExE,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,QAE9C,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAM,EAEN,EAAiB,GAAI,cAAa,GAExC,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAGJ,EAAc,EACd,EAAiB,GAAI,YACvB,GAA4C,EAAO,EAAS,IAEhE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAG,EAAG,EAAc,sBAAuB,EAAG,cACpD,KAID,GAAI,cAAa,EAAe,oBAIrC,EAA2B,EAAqB,EAAe,EAC/D,EAAc,EAAsB,EACpC,GACF,GAAM,GAAM,EAEN,EACF,GAAI,cAAa,GACb,EAAc,IAEtB,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAa,GAAI,cAAa,EAAe,EAAe,GAClE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAc,EAAc,EAAG,KAAM,EAAG,MAAO,KAEtD,owBCxTT,kBAgDE,WAAY,GANZ,KAAA,cAAmC,KACnC,KAAA,QAA6B,KACrB,KAAA,SAAW,GAqOX,KAAA,oBAAsB,GAsPtB,KAAA,YAA0B,GAtdhC,GAAM,GAAY,EAAA,MAAM,UAAU,iBAClC,AAAI,GAAM,KACR,MAAK,GAAK,EACV,GAAgB,EAAW,IAE3B,KAAK,GAAK,GAAgB,GAG5B,GAAI,GAAqB,2BACnB,EAA0B,8BAChC,GAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAM,GAAgB,oBAChB,EAAqB,yBAI3B,GAFA,KAAK,sBACD,GAA+B,KAAK,GAAI,GACxC,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,sHAKN,GADA,KAAK,0BAA4B,KAAK,GAAG,aAAa,GAClD,GAAwB,KAAK,GAAI,GACnC,KAAK,8BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,uIAIN,EAAqB,yBACjB,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,KAAK,GAAG,aAAa,WAChB,GAAwB,KAAK,GAAI,GAC1C,KAAK,8BACD,KAAK,GAAG,aAAa,OAEzB,MAAM,IAAI,OAAM,uDAIpB,KAAK,aAAe,GAA8B,KAAK,IACvD,KAAK,YAAc,GAA6B,KAAK,IACrD,KAAK,YAAc,GAA6B,KAAK,IAErD,KAAK,cACD,GAA0B,KAAK,GAAI,KAAK,2BAG9C,cAAA,eAAY,EAAA,UAAA,QAAK,KAAjB,WACE,MAAO,GAAA,MAAM,QAAQ,0CAGhB,EAAA,UAAA,QAAP,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAEF,AAAI,KAAK,SAAW,MAClB,QAAQ,KACJ,wKAIN,AAAI,KAAK,eAAiB,MACxB,QAAQ,KACJ,sMAKN,GAAM,GAAK,KAAK,GAChB,GAAwB,EAAI,WAAM,MAAA,GAAG,WACrC,GAAwB,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QACrE,GAAwB,EAAI,WAAM,MAAA,GAAG,kBAAkB,EAAK,eAC5D,GAAwB,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,QACjE,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,QACrD,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAK,eACvD,KAAK,SAAW,IAGX,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,yBAAP,SACI,EACA,GACF,KAAK,kBACL,GAAoC,KAAK,GAAI,EAAS,IAGjD,EAAA,UAAA,2BAAP,SACI,EAAuB,EAAe,EAAgB,GACxD,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,EAAO,EAAQ,EAAM,KAAK,gBAG3C,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,0BAAP,SAAiC,EAAc,GAE7C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,oBAAP,SAA2B,GAA3B,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,KAAK,gBAAkB,GACzB,IAA6C,KAAK,GAAI,KAAK,aAC3D,KAAK,cAAgB,MAEvB,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAGxD,EAAA,UAAA,gDAAP,SACI,EAAuB,EAAc,GADzC,GAAA,GAAA,KAEE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAM,EAAS,EAAK,kBAGhC,EAAA,UAAA,+BAAP,SACI,EAAqB,EAAe,EAAc,EAClD,EAAsB,GACxB,MAAO,IACH,KAAK,GAAI,EAAQ,EAAO,EAAM,EAAS,EAAc,EACrD,KAAK,gBAGJ,EAAA,UAAA,gCAAP,SAAuC,EAAqB,GAE1D,MAAO,IAA2C,KAAK,GAAI,EAAQ,IAG9D,EAAA,UAAA,wBAAP,SACI,EAAuB,EAAc,GACvC,KAAK,yBAAyB,GAC9B,GAAM,GAAS,GACX,KAAK,GAA8B,EAAM,EAAS,KAAK,eAC3D,YAAK,6BACE,GAGF,EAAA,UAAA,sBAAP,WACE,GAAM,GAAe,KAAK,YAAY,KAAK,IAC3C,MAAO,MAAK,UAAU,IAGhB,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACM,EACA,EAEJ,GAAI,EAAA,MAAM,QAAQ,4BAChB,GAAM,GAAM,EAEN,EAAO,EAAI,UAAU,EAAI,2BAA4B,GAC3D,EAAG,QAEH,EAAgB,WACd,GAAM,GAAS,EAAI,eAAe,EAAM,EAAG,GAC3C,MAAO,KAAW,EAAI,kBAClB,IAAW,EAAI,qBAGrB,EAAQ,MACH,AACH,GAAA,MAAM,UAAU,gDAAkD,EACpE,GAAQ,KAAK,aACb,KAAK,WACL,EAAgB,WAAM,MAAA,GAAK,iBACvB,EACA,EAAA,MAAM,UAAU,mDAMpB,EAAgB,WAAM,MAAA,IAGxB,MAAO,CAAC,MAAK,EAAE,cAAa,IAGvB,EAAA,UAAA,gCAAP,SACI,EAAuB,EACvB,GAFJ,GAAA,GAAA,KAGE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAc,MAK1B,EAAA,UAAA,cAAP,SAAqB,GACnB,KAAK,kBACL,GAAM,GAAK,KAAK,GACV,EACF,GAAgC,EAAI,GAClC,EAA4B,GAA8B,GAC1D,EAAwB,GAAyB,GACvD,UAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAuB,EAAI,GAC3B,AAAI,KAAK,OACP,GAA2B,EAAI,GAEjC,AAAK,KAAK,qBACR,MAAK,WAAW,GAChB,KAAK,oBAAsB,GACvB,EAAI,KAAK,QAAS,KAAK,eAEtB,GAGF,EAAA,UAAA,cAAP,SAAqB,GAArB,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,IAAY,KAAK,SACnB,MAAK,QAAU,MAEjB,AAAI,GAAW,MACb,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAI1D,EAAA,UAAA,WAAP,SAAkB,GAAlB,GAAA,GAAA,KACE,KAAK,kBACL,KAAK,QAAU,EACf,AAAK,KAAK,SAAW,MAAS,KAAK,OACjC,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,WAAW,MAGrD,EAAA,UAAA,mBAAP,SACI,EAAuB,EACvB,GAEF,MAFE,AAAA,KAAA,QAAA,GAAA,IACF,KAAK,kBACD,EACK,GACH,KAAK,GAAI,EAAS,GAEf,GACH,KAAK,GAAI,EAAS,IAInB,EAAA,UAAA,qBAAP,SAA4B,EAAuB,GAAnD,GAAA,GAAA,KAEE,YAAK,kBACE,GACH,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,kBAAkB,EAAS,MAGjD,EAAA,UAAA,0BAAP,SAAiC,EAAuB,GAEtD,YAAK,kBACE,KAAK,GAAG,mBAAmB,EAAS,IAGtC,EAAA,UAAA,sBAAP,SACI,EAAkC,EAClC,GACF,KAAK,kBACL,KAAK,mBACL,GACI,KAAK,GAAI,EAAoB,EAAiB,IAG7C,EAAA,UAAA,uBAAP,SACI,EAAmC,EAAc,GACnD,KAAK,6BAA6B,EAAqB,EAAS,IAG3D,EAAA,UAAA,6BAAP,SACI,EAAyC,EAAc,GACzD,KAAK,kBACC,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,KAAK,6BAA6B,EAA2B,EAAO,IAG/D,EAAA,UAAA,2BAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAK,iCACD,EAAa,EAAU,EAAY,IAGlC,EAAA,UAAA,iCAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAM,IAAI,OAAM,sDAGX,EAAA,UAAA,cAAP,WACE,AAAI,KAAK,SAAW,MAClB,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAA+B,KAAK,KAG/B,EAAA,UAAA,eAAP,WACE,KAAK,kBACL,KAAK,mBACL,GAAM,GAAK,KAAK,GAChB,AAAI,KAAK,OACP,KAAK,gBAEP,GACI,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,UAAW,EAAG,EAAG,eAAgB,MAG7D,EAAA,UAAA,+BAAP,WAAA,GAAA,GAAA,KACE,KAAK,kBACL,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,YAGzC,EAAA,UAAA,uBAAR,WAEE,MAAI,MAAK,6BAA+B,MACtC,MAAK,4BACD,GACI,KAAK,GACL,EAAA,MAAM,UACF,kDAAoD,EACpD,kCACA,6BAIP,KAAK,6BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGd,EAAA,UAAA,WAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EAAQ,EAAI,cAClB,SAAI,WAAW,EAAI,iBAAkB,GAC9B,EAET,GAAM,GAAM,KAAK,+BACX,EAAQ,EAAI,iBAClB,SAAI,cAAc,EAAI,iBAAkB,GACjC,GAGT,EAAA,UAAA,SAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BACjB,EAAI,SAAS,EAAI,kBACjB,OAEF,GAAM,GAAM,KAAK,+BACjB,EAAI,YAAY,EAAI,mBAGT,EAAA,UAAA,uBAAb,SAAoC,wGAClC,MAAA,CAAA,EAAM,EAAA,KAAK,YACP,WAAM,MAAA,GAAK,UAGP,EAAK,iBACD,EACA,EAAA,MAAM,UACF,2DAPhB,SAAA,OAQA,CAAA,EAAO,KAAK,aACR,EAAO,EAAA,MAAM,UAAU,wDAGrB,EAAA,UAAA,aAAR,SAAqB,EAAmB,GACtC,GAAI,IAAsB,EACxB,MAAO,MAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GAEX,EAAmB,EAAI,kBAAkB,EAAO,EAAI,cAE1D,MAAO,GAAmB,SAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,kBAErC,MAAO,GAAmB,MAItB,EAAA,UAAA,iBAAR,SAAyB,EAAmB,GAE1C,GAAI,IAAsB,EACxB,MAAO,GAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,wBACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,cAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,4BACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,WAI9B,EAAA,UAAA,UAAA,SAAU,GAAV,GAAA,GAAA,KACE,MAAO,IAAI,SAAc,SAAA,GACvB,EAAK,cAAc,WAAM,MAAA,GAAa,iBAAiB,WAAM,MAAA,UAMjE,EAAA,UAAA,UAAA,WAGE,OADM,GAAQ,GAAqB,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,YACtD,EAAI,EAAG,GAAK,EAAO,EAAE,GACrB,GAAA,GAAA,KAAA,YAAA,GAAA,UACP,IAEF,KAAK,YAAc,KAAK,YAAY,MAAM,EAAQ,IAG5C,EAAA,UAAA,cAAR,SAAsB,EAAyB,GAA/C,GAAA,GAAA,KAEE,GADA,KAAK,YAAY,KAAK,CAAC,SAAQ,EAAE,UAAS,IACtC,KAAK,YAAY,OAAS,EAE5B,OAGF,EAAA,KAAK,YAAY,WACf,SAAK,YAEE,EAAK,YAAY,SAAW,KAI/B,EAAA,UAAA,yBAAR,SAAiC,GAC/B,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,KAAK,aAC3B,AAAI,KAAK,OACP,GAA+B,KAAK,KAIhC,EAAA,UAAA,2BAAR,WACE,AAAI,KAAK,eAAiB,KACxB,IACI,KAAK,GAAI,KAAK,cAAe,KAAK,aACtC,AAAI,KAAK,OACP,GAA+B,KAAK,KAGtC,GAA6C,KAAK,GAAI,KAAK,cAIvD,EAAA,UAAA,qBAAR,SACI,EACA,GACF,KAAK,yBAAyB,GAC9B,GAAM,GAAS,IACf,YAAK,6BAEE,GAGD,EAAA,UAAA,6BAAR,SACI,EAA8C,EAC9C,GACF,KAAK,kBACL,GAAM,GAAK,KAAK,GAChB,GACI,EAAI,EAAgC,KAAK,aAC7C,AAAI,KAAK,OACP,GAA+B,GAEjC,KAAK,cAAgB,EACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAO,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAO,MAGpD,EAAA,UAAA,iCAAR,SACI,EAAW,EAAW,EAAe,GADzC,GAAA,GAAA,KAEE,KAAK,kBACL,GACI,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,QAAQ,EAAG,EAAG,EAAO,MAG1C,EAAA,UAAA,gBAAR,WACE,GAAI,KAAK,SACP,KAAM,IAAI,OAAM,4CAIZ,EAAA,UAAA,iBAAR,WACE,GAAI,KAAK,SAAW,KAClB,KAAM,IAAI,OAAM,qCAGtB,iBAaqC,GAEnC,OADI,GAAI,EACD,EAAI,EAAI,OAAQ,EAAE,GACvB,GAAM,GAAS,EAAI,KACnB,GAAI,CAAC,EACH,MAGJ,MAAO,GAAI,ECtnBb,YAgEI,EAAqB,EAAuB,EAC5C,GACF,GAAM,GAAW,EAAQ,SACnB,EAA0B,EAAO,IAAI,SAAC,EAAO,GACjD,GAAM,GAAuB,CAC3B,aAAc,EAAM,MACpB,SAAU,EAAM,UAAY,KAAO,EAAM,QAAQ,SACjD,UAAW,EAAM,UACjB,SAAU,EAAM,UAAY,GAAQ,EAAM,QAAQ,SAClD,WAAY,MAEd,MAAI,GAAM,SAAW,MAAQ,EAAM,QAAQ,OAAS,MAChD,EAAM,QAAQ,MAAM,WAAa,GACnC,GAAU,WAAa,EAAM,QAAQ,MAAM,YAEtC,CAAC,KAAM,EAAQ,cAAc,GAAI,UAAS,KAE7C,EAAe,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,YACrC,EAA0B,CAC9B,aAAc,EAAO,MACrB,SAAU,EAAO,QAAQ,SACzB,UAAW,GACX,SAAU,EAAO,QAAQ,SACzB,WAAY,MAER,EAAS,GACX,EAAY,EAAc,EAAU,EAAQ,cAE1C,EAAe,EAAM,cAAc,GAGrC,EAA+B,KAC7B,EAAS,EAAM,mBAAmB,EAAc,MAAO,IAC7D,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAS,EAAM,mBAAmB,EAAc,WAAY,KAK9D,OADM,GAA2D,GACxD,EAAI,EAAG,EAAI,EAAQ,cAAc,OAAQ,KAChD,GAAM,GAAU,EAAQ,cAAc,GAChC,EAAc,GACpB,EAAiB,GACb,EAAM,mBAAmB,EAAc,EAAS,GACpD,EAAiB,SAAS,GACtB,EAAM,mBAAmB,EAAc,SAAS,EAAW,GAGjE,MAAO,CACL,QAAO,EACP,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,aAAY,EACZ,aAAY,EACZ,OAAM,EACN,OAAM,GAIV,YACI,EAAyB,GAC3B,GAAI,EAAW,SAAW,EAAO,OAC/B,KAAM,OACF,4BAA4B,EAAW,OAAM,gBAC7C,sBAAqB,EAAO,OAAM,YAGxC,EAAW,QAAQ,SAAC,EAAG,GACrB,GAAM,GAAS,EAAE,aACX,EAAQ,EAAO,GACf,EAAS,EAAM,MAErB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAQ,GAC5B,KAAM,OACF,kDACA,6BAA4B,EAAM,QAAQ,EAAM,gBAGtD,GAAI,EAAE,WAAa,EAAM,UACvB,OAGF,GAAM,GAAY,EAAE,SACd,EAAY,EAAM,UAAY,KAAO,EAAM,QAAQ,SACzD,GAAI,CAAC,EAAA,KAAK,YAAY,EAAW,GAC/B,KAAM,OACF,6DACA,yBAAwB,EAAS,QAAQ,EAAS,8BAMxD,EAAqB,EAAqB,EAC1C,EACA,GAEF,GAAyB,EAAO,aAAc,GAC9C,GAAyB,CAAC,EAAO,cAAe,CAAC,IAEjD,GAAM,GAAS,EAAO,QAAQ,QACxB,EAAc,EAAO,QAAQ,SACnC,AAAI,EAAO,QAAQ,SACjB,EAAM,6BAA6B,EAAQ,EAAY,GAAI,EAAY,IAEvE,EAAM,uBAAuB,EAAQ,EAAY,GAAI,EAAY,IAEnE,EAAM,WAAW,EAAO,cAGxB,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,CAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,WAGtC,AAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,KAIpC,EAAO,QAAQ,SAAC,EAAO,GACrB,GAAM,GAAU,EAAO,QAAQ,cAAc,GACvC,EAAS,EAAO,iBAAiB,GACjC,EAAe,EAAO,iBAAiB,SAAS,GAEtD,GAAI,GAAU,KAEZ,OAGF,GAAI,EAAM,WAER,GAAI,EAAA,KAAK,cAAc,EAAM,OAAS,EACpC,EAAM,GAAG,UAAU,EAAQ,EAAM,cAAc,SAE/C,GAAI,GAAO,EAAM,cACjB,AAAM,YAAgB,eACpB,GAAO,GAAI,cAAa,IAE1B,EAAM,GAAG,WAAW,EAAQ,GAE9B,OAIF,AAAI,EAAM,QAAQ,OAAS,MAAQ,GAAgB,MACjD,EAAM,GAAG,UAAU,EAAc,EAAM,QAAQ,MAAM,YAGvD,EAAM,sBAAsB,EAAM,QAAQ,QAAS,EAAQ,KAG7D,AAAI,GAAe,MACjB,EAAY,EAAO,EAAO,cAE5B,EAAM,6BAIJ,EAAuB,EAAsB,GAC/C,GAAI,GAAY,GAChB,EAAO,OAAO,GAAQ,QAAQ,SAAA,GAC5B,GAAM,GAAY,EAAE,SAAW,MAAQ,EAAE,QAAQ,OAAS,MACtD,EAAE,QAAQ,MAAM,WAAa,EAC3B,EAAW,EAAE,UAAY,UAAY,EAAE,QAAQ,SACrD,GAAgB,EAAE,MAAK,IAAI,EAAQ,IAAI,IAEzC,GAAM,GAAc,EAAQ,SACxB,EAAM,EAAQ,YAAY,KAE9B,UAAO,IAAM,EAAY,IAAM,EACxB,EC5OT,AAqBA,GAAA,IAAA,WAOE,WACI,EAAuB,EACvB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAsBnB,OAnBE,GAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,cACA,EAAA,EAAA,eACA,EAAA,EAAA,WAEK,EAAA,EAAA,KAAM,EAAA,EAAA,IACP,EAAmB,EAAa,EAChC,EAAO,KACP,EAAiB,IAAe,eAChC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAEhC,EAAW,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,GAAY;gCACY,EAAG;yBACV,EAAG;;4BAEA,EAAY,GAAE,aAAa,EAAY,GAAE;0CAC3B,EAAQ,QAAQ,EAAY,MAC1D,EAAG;6BACc,EAAc,aAAa,EAAgB;;sBAElD,EAAW,GAAO;;qDAEa,EAAQ,QACjD,EAAW,OAAO,EAAI;+BACH,EAAa,2BAChC,EAAgB,QAAQ,EAAU;;wBAEtB,EAAW,GAAO;;2CAEC,EAAU;;sBAE/B,EAAc;;2BAET,GAAM,EAAI,GAAG;;;;;2BAKb,GAAM,EAAI,GAAG;;;;;;;UAWpC,KAAK,SAAW;;;;;;;;;UASV,EAAQ;;UAER,EAAK,OAAM;;MAIrB,MAAA,MC5GA,AAmBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAgB,EAAc,EAChD,GANJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMtB,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;;wBASI,EAAG,UAAU,EAAG;;oCAEJ,EAAI;;;;;0BAKd,EAAW;;;MAKrC,MAAA,MClEA,AAmBA,GAAA,IAAA,WAUE,WACI,EAAsB,EAAqB,EAAc,EACzD,GAXJ,KAAA,cAAgB,CAAC,aAAc,cAAe,MAC9C,KAAA,YAAwB,GAWtB,KAAK,YAAc,EACnB,KAAK,MAAQ,EAAW,GACxB,KAAK,YAAc,EACnB,KAAK,KAAO,EACZ,KAAK,MAAQ,EACb,KAAK,KAAO,EACZ,KAAK,SAAW;;;;;;;;8BAQU,KAAK,MAAK;oDACY,EAAW;yCACtB,KAAK,MAAK;0BACzB,EAAW;;;sCAGC,KAAK,MAAK;;;;;;;;;;;;;;;yBAevB,EAAK,oBAAoB,EAAI;;;;;;;yCAOb,EAAK;0BACpB,EAAI;;;;0CAIY,EAAI;;;;;;;;;;;;;;MAgB9C,MAAA,MC/FA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAgB,EAAc,EAChD,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;gCAQY,KAAK,YAAY,GAAE;gCACnB,KAAK,YAAY,GAAE;;;;;;;;;;;;;;;iCAelB,EAAG;;;;;;;;;;;yBAWX,EAAG,UAAU,EAAG;;;6DAGoB,EAAI;;;;;;;;;;;;;;;;;;;;;0CAqBvB,EAAW;;;MAKrD,MAAA,MC/GA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAY,EAAwB,EAAuB,EACjE,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAcnB,EAAqB;kBACnC,EAAc;gDACgB,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;kDACJ,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;gCAOvB,EAAS;;;;qCAIJ,EAAoB;;;;;;;;MAUzD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EACF,EAAuB,EAAwB,EAAuB,EAC1E,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;gCAiBhC,EAAoB;mBACjC,EAAa;gDACgB,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;kCAOvB,EAAS;;;;;;yBAMlB,EAAqB,MAAM,EAAoB;yBAC/C,EAAoB;;;;;;;;;MAW7C,MAAA,MChLA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkC,EAClC,EAAoB,EAAoB,EACxC,EAA2B,GAD3B,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IACxC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAT/B,KAAA,cAAgB,CAAC,UAAW,WAC5B,KAAA,aAAe,GACf,KAAA,aAAe,GAQb,KAAK,YAAc,EAEnB,GAAM,GAAY,EAAa,EAAO,GAAK,EAAO,GAC5C,EAAwB,KAAK,KAAK,EAAY,GAE9C,EAAU,EAAa,cAAgB,cACvC,EAAU,EAAa,cAAgB,cACvC,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAC1D,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAE5D,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;sCAEa,EAAqB;;;;8BAI7B,EAAqB;sCACb,EAAO;sCACP,EAAO;;;;uBAItB,EAAS,GAAE,MAAM,EAAS,GAAE;uBAC5B,EAAS,GAAE,MAAM,EAAS,GAAE;;;;;;;;;UASzC,EAAc;;UAEd,EAAsB;;;;MAMhC,MAAA,MChGA,AAoBA,GAAA,IAAA,WAQE,WAAY,EAAmB,EAAqB,GAPpD,KAAA,cAAgB,CAAC,SAQf,KAAK,YAAc,CAAC,EAAW,GAE/B,KAAK,SAAW;;;;;;;;;;8BAUU,GAAc,GAAC;;;;;;;;;;0BAUnB,GAAc,GAAC;;MAKvC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAM,mBAAmB,EAAc,SAExD,EAAM,GAAG,UAAU,EAAK,QAAS,KAGvC,KChEA,AAmBA,GAAA,IAAA,WAQE,WACI,EAAoB,EAAe,EAAiB,GARxD,KAAA,cAAgB,CAAC,WASf,KAAK,YAAc,CAAC,EAAY,GAEhC,KAAK,SAAW;;;;8BAIU,EAAQ,YAAY,EAAO;;;MAKzD,MAAA,MCxCA,AAqBA,GAAA,IAAA,WAOE,WACI,GAPJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEzB,GAAI,IAAS,EACX,KAAK,SAAW;;;;aAMhB,GAAM,GAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EACF,GAAwB,EAAM,EAAa,GACzC,EAAQ,GACV,EAAM,EAAY,EAAY,OAAS,GACvC,EAAY,EAAY,OAAS,GAAI,GACnC,EAAS,GAAU,EAAa,GAEtC,KAAK,SAAW;;YAEV,EAAK;;eAEF,EAAoB;;;cAGrB,EAAK;;6BAEU,EAAM;;;SAMnC,MAAA,MAEA,YAA4B,EAAc,GAGxC,OAFM,GAAS,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAG1B,OAFI,GAAW,KAAQ,EAAI,IAAM,OAAK,KAAK,KAAQ,EAAI,IAAM,OAEpD,EAAI,EAAG,EAAI,EAAM,IACxB,EAAW,EAAK,EAAK,OAAS,EAAI,GAAE,IAAM,EAG5C,EAAO,KAAK,GAGhB,MAAO,GAGT,YACI,EAAc,EAAiB,GACjC,GAAI,IAAS,EACX,MAAO,QAAQ,EAAM,GAIvB,OADI,GAAO,GACF,EAAI,EAAO,EAAG,EAAI,EAAM,IAC/B,GAAW,EAAK,GAAE,OAAO,EAAM,GAC/B,AAAI,EAAI,EAAO,GACb,IAAQ,MAIZ,MAAO,GAGT,YACI,EAAc,EAAc,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,GAGT,GAAM,GAAY,EAAK,MAAM,IAE7B,MAAO;cACK,EAAU,GAAE;cACZ,EAAU,GAAE;;;;0BAIA,EAAI;0BACJ,EAAI;IAI9B,YAAmB,EAAiB,GAClC,GAAM,GAAO,EAAM,OACb,EAAe,GAAmB,EAAM,GAC9C,MAAI,KAAS,EACJ;wBACa,EAAM,GAAE;kBAIvB,QAAQ,EAAa,GAAE;8BACF,EAAa,GAAE;8BACf,EAAa,GAAE;uCACN,EAAa,GAAE,ICrItD,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAClB,GANJ,KAAA,cAAgB,CAAC,KAOf,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KACnD,GAAM,GAAO,EAAO,OACd,EAAO,GAAkB,GAEzB,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EACF,CAAC,YAAa,YAAa,YAAa,aAAa,MAAM,EAAG,GAElE,GAAI,IAAS,GACX,KAAK,SAAW;sBACA,EAAK;oBACP,EAAG;;;;;8BAKO,EAAa;;;;;QAMrC,OAEF,KAAK,SAAW;QACZ,EAAI,YAAY,EAAI,IAAI,EAAK;QAC7B,EAAI,UAAU,EAAI,IAAI,EAAG;;;UAGvB,EAAI;;4BAEc,EAAa;;YAE7B,EAAI;2BACW,EAAc;;;MAKzC,MAAA,MCrEA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAClB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KA4BnD,OA3BM,GAAO,EAAO,OACd,EAAQ,GAAkB,GAE1B,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EAAS,GAAY,KAAM,GAC3B,EAAS,GAAY,SAAU,GAC/B,EAAY,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAC1D,EACF,IAAS,EAAI,SAAW,QAAQ,EAAO,MAAM,IAAI,OAAM,IAErD,EAAiB,CAClB,EAAK,mBAAuB,EAAO,EAAO,GAAE;YACzC,EAAM;QAEZ,IAAS,EAAI,GAAK;;SAEf,EAAO,EAAO,GAAE;YACb,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAAE,MACtD,IAAS,EAAI,GAAK,KAAK,EAAO,EAAO,GAAE;cAC/B,EAAM,OAGV,EAAc,IAAS,EACzB,0BACA,6DACA,EAAW,GACN,EAAI,EAAG,EAAI,IAAS,EAAI,EAAI,EAAG,EAAI,EAAG,IAC7C,GAAY;UACR,EAAe,GAAE;cACb,EAAW;mBACN,EAAC,aAAa,EAAa;;YAElC,EAAK;mBACE,EAAC,uBAAuB,EAAO,OAAM,MAAM,EAAS;;QAInE,GAAa,IAAS,EAAI,KAAO,KAEjC,KAAK,SAAW;cACN,EAAK,YAAY,EAAK,IAAI,EAAK;cAC/B,EAAK,UAAU,EAAK,IAAI,EAAG;;;UAG/B,EAAK;;UAEL,EAAQ;;;MAKlB,MAAA,MCrFA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MACzB,EAA0B,cAAc,EAAS,SAAQ,YAC3D,EAAS,QAAO,YAAY,EAAS,WAAU,OAC7C,EACF,SAAS,EAAS,QAAO,YAAY,EAAS,WAAU,OAExD,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;sCACgB,EAAY,KAAK,EAAW;mCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBnB,EAAqB;sBACjC,EAAc;;;kCAGF,EAAS,SAAQ;;;;oCAIf,EAAoB;wBAChC,EAAa;;;oCAGD,EAAS,QAAO;;;;;;;;;;0BAU1B,EAAS;;;mCAIzB,GAAoB,EAAsB,EACA,EACvB,QAAQ,EAAoB,SAAO;;;;;;QAO1D,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;oCACgB,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;0CACT,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;kCAkBZ,EAAmB;;;;gCAIrB,EAAqB;oBACjC,EAAc;;;gCAGF,EAAS,SAAQ;;;;kCAIf,EAAsB;uCACjB,EAAa;;;;yCAIX,EAAa;6CACT,EAAa;6CACb,EAAa;;;cAG5C,EAAa;;;gCAGK,EAAsB;gBACtC,KAA6B,GAAC;;;;;;;;cAQhC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;;;;;cAKxC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;6CACT,EAAa;;;;cAI5C,EAAa;;;oBAGP,EAAW;;MAI/B,MAAA,MAEA,GAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MAE3B,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;;oBAEF,EAAW,KAAK,EAAY,KAAK,EAAW;mCAC7B,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBhC,EAAoB;sBAChC,EAAa;;;kCAGD,EAAS,QAAO;;;;oCAId,EAAqB;wBACjC,EAAc;;;oCAGF,EAAS,SAAQ;;;;sCAIf,EAAoB;0BAChC,EAAa;;;sCAGD,EAAS,QAAO;;;;;;;;;;4BAU1B,EAAS;;;qCAI3B,GACK,EACI,cAAc,EAAS,QAAO,YAC1B,EAAS,SAAQ,YAAY,EAAS,QAAO,YAC7C,EAAS,WAAU,QACvB,UAAU,EAAS,SAAQ,YACvB,EAAS,QAAO,YAAY,EAAS,WAAU,QACxD,QAAQ,EAAqB,MAAM,EAAoB;6BACxC,EAAoB,SAAO;;;;;;;QAQlD,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;;gBAEJ,EAAW,KAAK,EAAY,KAAK,EAAW;iCAC3B,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;;kCAmBZ,EAAmB;;;;gCAIrB,EAAoB;oBAChC,EAAa;;;gCAGD,EAAS,QAAO;;;;kCAId,EAAqB;oBACnC,EAAc;;;kCAGA,EAAS,SAAQ;;;;oCAIf,EAAsB;yCACjB,EAAa;;;;+CAIP,EAAa;mDACT,EAAa;mDACb,EAAa;;;gBAGhD,EAAa;;;kCAGK,EAAsB;kBACtC,KAA6B,GAAC;;;;;;;;gBAQhC,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;;;;;gBAK5C,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;mDACT,EAAa;;;;gBAIhD,EAAa;;;sBAGP,EAAW;;;MAKjC,MAAA,MCvcA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,OAAQ,EAAA,EAAA,QACtC,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAI,GAAsB,MACtB,EAAY,GAEhB,AAAI,IAAe,OACjB,EAAsB,MACjB,AAAI,IAAe,MAExB,GAAsB,cACtB,EAAY,OACP,AAAI,IAAe,OAExB,GAAsB,eACtB,EAAY,OAGd,GAAI,GAAiB,EAAU,IAAI,EAAU,IAAI,EAAU,qEAG3D,AAAI,IAAe,MACjB,EAAc,WACT,AAAI,IAAe,OACxB,EAAc,YACT,AAAI,IAAe,MACxB,EAAc,WACT,AAAI,IAAe,OACxB,GAAc,YAGhB,GAAM,GAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEzC,EAAgB;YACZ,KAAe,OAAK;;mBAEb,KAAe,QAAM;;;;wBAIhB,EAAS;;MAIzB,EAAU,OAEd,AAAI,IAAe,MACjB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SACL,AAAI,IAAe,OACxB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SAGZ,GAAI,GAAmB,GACvB,AAAI,EAAS,EAAa,GACxB,GAAmB;oCACW,EAAM;;;SAKtC,KAAK,SAAW;0CACsB,EAAmB;;;;UAInD,EAAgB;;;;;;;;kCAQQ,EAAU;;kCAEV,EAAmB;;;;;;8BAMvB,EAAqB;;YAEvC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;YAC/B,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCvKA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAuC,GANnD,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAGnB,OADI,GAAW,GACN,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAS,eACb,AAAI,EAAI,IAAM,GACZ,IAAU,kBAEZ,AAAI,EAAI,GACN,IAAU,kBAGZ,GAAY;UACR,EAAM;UACN,GAAI,EAAI,0CAA4C,IAAE;;;;;;mBAM7C,EAAC;;UAEV,GAAI,EAAI,IAAM,IAAE;QAItB,KAAK,SAAW;QACZ,GAAuB,GAAW;QAClC,GAA+B,GAAY;;;;;;;;qBAQ9B,EAAY,GAAE;qBACd,EAAY,GAAE;;UAEzB,EAAQ;;;;MAMlB,MAAA,MAEA,YAAgC,GAC9B,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;QAED,EAAsB;;;ICnF9B,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;;sDAMY,GAAU,GAAC;;;;;;qDAMZ,GAAS,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAgC/D,MAAA,MC1IA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MA+B/D,MAAA,MC3EA,AAmBA,GAAA,IAAA,WAOE,WACI,EAA8C,EAC9C,EAAkB,GARtB,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;uCACxB,EAAQ;;;;;;;;;;;;;;;;;;;;;;gCAsBf,GAAQ,GAAC;uCACF,GAAW,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAkDnD,MAAA,MCzHA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;sBAKpB,EAAe,GAAE;sCACD,EAAe,GAAE;;;wBAG/B,EAAe,GAAE;wCACD,EAAe,GAAE;;;4BAG7B,EAAO;kBACjB,EAAY;;;;4BAIF,EAAM;kBAChB,EAAY;;;;;;;;;;;;MAc9B,MAAA,MC/HA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAI5C,EAAY,EAAe,MAAQ,MAEzC,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;8DAaD,EAAS;;;;;;MAQvE,MAAA,MCpEA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAI5C,GAFA,KAAK,YAAc,EAEf,IAAS,GACX,KAAK,SAAW;;;2BAGK,EAAO,GAAE;;QAG9B,OAEF,GAAM,GAAa,SAAC,GAClB,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,aAAa,EAAC,QAE5B,UAAU,EAAC,KAEd,EAAW,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,KAAI,KAAK,KACpD,EAAO,GAAkB,GAE/B,KAAK,SAAW;;UAEV,EAAI;yBACW,EAAQ;;MAIjC,MAAA,MC1DA,AAqBA,GAAA,IAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAE5C,KAAK,YAAc,EACnB,GAAM,GAAW,GAAY,KAAM,GAC7B,EACC,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACrD,EAAa,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACjE,EAAO,GAAkB,GAC/B,AAAI,IAAS,EACX,KAAK,SAAW;;;;uCAIiB,EAAO,GAAE;cAClC,EAAO,GAAE;eACR,EAAU;2CACkB,EAAO,GAAE;kBAClC,EAAO,GAAE;;;;QAMrB,KAAK,SAAW;;YAEV,EAAI;;uBAEO,EAAK,EAAS,SAAQ;eAC9B,EAAU;yBACA,EAAK,EAAS,SAAQ;;eAEhC,EAAO;yBACG,EAAK,EAAS,SAAQ;iBAC9B,EAAU;2BACA,EAAK,EAAS,SAAQ;;;;;MAQ7C,WAAc,GACZ,MAAO,GAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QAChD,EAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAoB,GAClB,GAAM,GAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,EAAG,KACnD,EAAW,EAAc,KAAK,KAC9B,EAAY,EAAc,MAAM,IAAI,KAAK,KAC/C,MAAO,mBAAmB,EAAQ,WAAW,EAAS,KAGxD,WAAoB,EAAW,GAC7B,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,MAAM,EAAU,GAAE,OAE9B,GAAG,EAAU,IAI5B,MAAA,MC7GA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAoB,EAAkB,EACtC,EAAqB,EAAmB,EACxC,GAPJ,KAAA,cAAgB,CAAC,UAAW,UAAW,gBAQrC,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAClC,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,QAElB,GAAM,GAAiB,cAAc,EAAa,IAE9C,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,gBAElB,GAAM,GAAiB,cAAc,EAAa,IAE5C,EAAe,EAAW,EAAI,aAAe,UACnD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,EAAO;;;YAG7C,EAAK;;;gCAGe,EAAU;;kCAER,EAAQ;kCACR,EAAc;0CACN,EAAY;;;uBAG/B,EAAc;;;;;;QAQrC,MAAA,MCvEA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,IAAK,cAOpB,GAAM,GAAa,EAAU,WACvB,EAAY,EAAU,UACtB,EAAS,EAAU,OACnB,EAAc,EAAU,YACxB,EAAU,EAAc,KAAK,KAAK,EAAS,GACjD,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAM,GAAsB,MACtB,EAAc,WAEd,EAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEvC,EAAgB;;MAIlB,EAAwB,GAC5B,AAAI,EAAS,EAAa,GACxB,GAAwB;oCACM,EAAM;;;SAMtC,GAAI,GAA4B,GAChC,AAAI,EAAS,EAAa,GACxB,GAA4B;oCACE,EAAM;;;SAMtC,KAAK,SAAW;0CACsB,EAAmB;;;UAGnD,EAAqB;;;;;UAKrB,EAAyB;;;;;;;;;YASvB,EAAW,cAAc,EAAU;wDACS,EAAW;;;;8BAIrC,EAAqB;;;;;;;;;;;;;;;;YAgBvC,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;;;;;;;;;;;;;;;;;YAiB/B,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCjKA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAe,EAAiB,GAJ5C,KAAA,cAAgB,CAAC,IAAK,IAAK,KAKzB,KAAK,YAAc,EAEnB,GAAI,GACA,EACJ,GAAI,EAAO,EACT,KAAM,OAAM,kBAAkB,EAAI,yBAGpC,GAAI,IAAS,EACX,EAAW,QACX,EAAU,aAKV,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAClD,EAAa,GACb,EAAc,GACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAY,KAAK,GAAG,EAAc,IAClC,AAAI,EAAI,GACN,EAAW,KAAK,GAAG,EAAc,IAGrC,EAAU,EAAW,OACrB,EAAW,EAAY,OAGzB,GAAM,GAAQ,GAAkB,GAEhC,KAAK,SAAW;;UAEV,EAAK;4BACa,EAAO;;2BAER,EAAQ;;2BAER,EAAQ;;;MAKnC,MAAA,MCjEA,AAqBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,UASf,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAc,qBAAqB,KAAK,KAAI,KAC5C,EAAe,GAAU,KAAK,MAEhC,EACE,EAAW,EAAS,IAAI,SAAC,EAAG,GAChC,MAAO,aAAa,GAAO,GAAE,YAAY,EAAC,cAAc,GAAO,GAAE,MAEnE,EAAO;UACD,EAAK;UACL,EAAK;UACL,EAAS,KAAK;GAAK;QAEzB,KAAK,SAAW;QACZ,EAAW;;UAET,EAAI;8BACgB,EAAY;;MAKxC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KAEM,GAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,YAAmB,GACjB,GAAI,IAAS,EACX,MAAO,YACF,GAAI,GAAQ,EACjB,MAAO,IAAO,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAA,aAAe,IAAG,KAAK,KAE7D,KAAM,OAAM,oBAAoB,EAAI,yBCpFxC,AAsBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,UACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAS,GAAY,SAAU,KAAK,MACpC,EAAY,GAAY,YAAa,KAAK,MAE1C,EACF,KAAK,OAAS,EAAI,YAAc,QAAQ,EAAU,MAAM,IAAI,OAAM,IAChE,EACF,wBAAwB,EAAU,OAAM,MAAM,EAAS,IACrD,EAAW;mBACF,EAAU;cACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;YACnB,EAAU,KAAK,KAAO,GAAE;;MAG1B,EAAW,KAAK,OAAS,EAAI,GAAK;UAClC,EAAO,KAAK,KAAO,GAAE;cACjB,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;gBACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;cACpD,EAAU,KAAK,KAAO,GAAE;uBACf,EAAU;;;MAKvB,EAAiB,KAAK,MAAQ,EAChC;cACM,EAAK,IAAI,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,SAAS,EAAC,MAAK,OAAM,KAC3D,EAAS,IAAI,SAAC,EAAG,GAAM,MAAG,GAAU,GAAE,MAAM,EAAO,GAAE,YAAY,EAAC,OAC7D,KAAK;GACd,KAAK,SAAW;0BACM,KAAK,KAAI;;UAEzB,EAAK;UACL,EAAK;UACL,EAAc;;UAEd,EAAQ;UACR,EAAQ;;;MAMhB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KCtGA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAiB,EAAmB,GAJhD,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAK,OACZ,EAAa,GAAkB,EAAK,QACpC,EAAQ,GAAkB,EAAK,QAEjC,EAAY,GAChB,GAAI,IAAS,EACX,EAAY,gCAEZ,GAAI,GAAa,EACjB,EACI,EAAK,IAAI,SAAC,EAAG,GACP,WACO,EAAK,SAAW,EACnB,oBAAoB,EAAC,aAAa,EAAC,IACnC,UAAU,GAAa,GAAC,eAAe,EAAC,aAAa,EAAC,MAE3D,KAAK,KAGhB,KAAK,SAAW;QACZ,EAAU,YAAY,EAAU,IAAI,EAAK;QACzC,EAAU,cAAc,EAAU,IAAI,EAAO;;;UAG3C,EAAK;yBACU,EAAS;;MAIlC,MAAA,MCxDA,AAuBA,GAAA,IAAA,WAUE,WAAoB,GAAA,KAAA,MAAA,EATZ,KAAA,gBAAkB,EAClB,KAAA,gBAAkB,EAClB,KAAA,mBAAqB,EACrB,KAAA,cAAgB,EAEhB,KAAA,aAAkD,GAClD,KAAA,WAAa,GACb,KAAA,aAAkD,GAI1D,SAAA,UAAA,eAAA,SACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAkC,EAAO,GAE3D,EAAW,GAAuB,EAAS,EAAiB,GAClE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAEhC,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAS,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cACpD,GAEJ,GAAI,KAAK,aAAa,GAAU,OAAS,GACvC,KAAK,kBACL,KAAK,kBACL,KAAK,eAAiB,EACtB,KAAK,MACL,GAAM,GAAa,KAAK,aAAa,GAAU,QAC/C,YAAK,aAAa,GAAU,KAAK,GAC1B,EAGT,GAAI,GACJ,MAAI,KAAoB,GAAoB,mBAC1C,EAAa,KAAK,MAAM,0BAA0B,EAAQ,GAAI,EAAQ,IACjE,AAAI,IAAoB,GAAoB,mBACjD,EACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,IAC/D,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AACH,IAAoB,GAAoB,0BAC1C,GACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,KAEtE,KAAK,aAAa,GAAU,KAAK,GAEjC,KAAK,kBACL,KAAK,oBAAsB,EAC3B,KAAK,MAEE,GAGT,EAAA,UAAA,eAAA,SACI,EAAuB,EACvB,EAA8B,GAChC,GAAI,KAAK,cAAgB,KAEvB,OAEF,GAAM,GACF,GAAkC,EAAgB,GAChD,EAAW,GAAuB,EAAO,EAAiB,GAChE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAO,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cAClD,GACE,EAAqB,EAAA,MAAM,IAAI,kCACrC,AAAI,IAAuB,IACvB,KAAK,mBAAqB,EAC5B,MAAK,MAAM,oBAAoB,GAC/B,KAAK,oBAAsB,GAE3B,MAAK,aAAa,GAAU,KAAK,GACjC,KAAK,kBACL,KAAK,eAAiB,GAGxB,KAAK,kBAEL,GAAM,GAAU,KAAK,aAAa,GAC5B,EAAW,EAAQ,QAAQ,GACjC,GAAI,EAAW,EACb,KAAM,IAAI,OACN,4EAGN,EAAQ,OAAO,EAAU,GACzB,KAAK,OAGC,EAAA,UAAA,IAAR,WACE,GAAI,CAAC,KAAK,WACR,OAEF,GAAM,GAAQ,KAAK,gBAAkB,KAAK,gBAC1C,QAAQ,IACJ,YAAgB,KAAK,gBAAe,MAAM,KAAK,gBAC/C,IAAI,EAAK,KACb,GAAM,GAAY,KAAK,cAAgB,KAAK,mBAC5C,QAAQ,IAAI,oBAAoB,KAAK,oBACrC,QAAQ,IAAI,iBAAiB,KAAK,cAAa,KAC3C,KAAK,MAAM,IAAM,GAAU,OAGjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGd,OAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,+CAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,cAAgB,KAEvB,OAEF,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,KAAK,aAAe,KACpB,KAAK,aAAe,KACpB,KAAK,gBAAkB,EACvB,KAAK,gBAAkB,EACvB,KAAK,mBAAqB,EAC1B,KAAK,cAAgB,GAEzB,KAEA,YACI,EAA2B,GAE7B,GAAM,GAAQ,EACd,GAAI,IAAmB,EAAM,KAC3B,MAAO,GACF,GAAI,IAAmB,EAAM,KAClC,MAAO,GACF,GAAI,IAAmB,EAAM,QAClC,MAAO,IACF,GAAI,IAAmB,EAAG,KAC/B,MAAO,IACF,GAAI,IAAmB,EAAM,QAClC,MAAO,GAET,KAAM,IAAI,OAAM,2BAA2B,eAIzC,EAAyB,EACzB,EAA2B,EAC3B,GAMF,GAAM,GACF,GAAiC,EAAiB,GAElD,EACJ,GAAI,GACI,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAEpB,EAAc,EAAc,OAGtB,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,EAAc,EAAQ,EAGxB,GAAM,GAAkB,GAA0B,EAAI,GACtD,MAAO,GAAc,EAGvB,YACI,EACA,GACF,OAAQ,OACD,IAAoB,mBACvB,MAAO,IAAwC,OAC5C,IAAoB,mBACvB,MAAO,IAA+C,OACnD,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,yBACvB,MAAO,IAA+C,WAEtD,KAAM,IAAI,OAAM,iCAAiC,IAIvD,YAAwC,GAEtC,MAAI,GAAA,MAAM,QAAQ,gCACZ,EACK,GAAoB,mBAEtB,GAAoB,iBAGzB,EACK,GAAoB,mBAEtB,GAAoB,iBAG7B,YACI,EAA8B,GAChC,GAAI,IAAmB,GAAa,OAClC,MAAO,IAAoB,mBACtB,GAAI,IAAmB,GAAa,QAAU,GAAkB,KACrE,MAAO,IAA+B,GACjC,GACH,IAAmB,GAAa,UAChC,IAAmB,GAAa,OAClC,MAAO,IAAoB,yBAE7B,KAAM,IAAI,OAAM,gCAAgC,GAGlD,YACI,EAAgC,EAChC,GACF,MAAU,GAAa,GAAE,IAAI,EAAa,GAAE,IAAI,EAAe,IAAI,ECxRrE,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,GAEpC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,GAErC,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,GACvB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,iBAAiB,EAAI,yBAEnC,GAAI,IAAS,EACX,MAAO,eAAe,EAAO,GAAE,IAMjC,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,UAAW,WAE7D,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAa,KAAK,QAAQ,EAAc,GAAE,KAAK,EAAO,GAAE,KAE1D,MAAO,GAAa,OC5DtB,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MAEM,GAAoB,0BAEb,GAAS,YAET,GAAM,iBAEN,GAAO,GAAoB;;EAI3B,GAAQ,GAAoB;;EAI5B,GAAM,0CAEN,GAAO;;;uBAGG,EAAA,aAAa,gBAAe;kBACjC,EAAA,aAAa,WAAU;;cAIpB,GAAA,MAAA,KAAA,QAAA,GAAA,GACZ,GAAoB;mCACM,EAAK;KAIjC,GAAM,IAAM,aAEN,GAAO,kBAEP,GAAQ,mBAER,GAAO;;;EAKP,GAAS,0BAET,GAAS,0BAET,GAAY,wCAEZ,GAAQ;;;;;;;;;;;;;;;EAiBR,GAAM,iBAEN,GAAQ,uBAER,GAAM;kBAGN,GAAQ,uBAER,GAAO,kBAEP,GAAQ,yBAER,GAAU,sCAeV,GAAW;;;;;;;;;;;;;;;;;;;;EAsBX,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;EAI3B,GAAO;;;EAKP,GAAO;;;EAKP,GAAO;;;EAKP,GAAQ,GAAoB,qCAE5B,GAAQ,GAAoB;;sCAI5B,GAAQ,GAAoB;;+CAI5B,GAAM;;;;cAIL,EAAA,aAAa,MAAK;eACjB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;;;;;;EAQrB,GAAa,kBAEb,GAAc,6BAEd,GAAS,wBAET,GAAQ,YC9NrB,AAmBO,GAAM,IAAS,YAET,GAAM;;;;;;;;;EAWN,GAAO;;;;;;;;;;EAYP,GAAQ;;;;;;;;;;EAYR,GAAM;;;;;;;;;EAWnB,GAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MCzFA,AAqBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEnB,EAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EAAe,GAAgB,EAAM,GACrC,EAAY,EAAS,MAAM,IAC3B,EAAS,GAAQ,EAAI,KAAO,QAAQ,EAAU,KAAK,KAAI,IAE7D,KAAK,SAAW;;UAEV,EAAK;kCACmB,EAAY;;4CAEF,EAAM;;MAIlD,MAAA,MC/CA,AA2BO,GAAA,IAAA,EAAA,aAAA,aACD,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAiFlB,GAAkB,KAClB,GAAkB,KA4BzB,GAAuE,eAE9C,GAC7B,MAAI,KAAgB,KAGpB,IAAa,GAAgB,IACtB,GAAa,GAGtB,YACI,EAAqC,GACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACnC,IAAe,SACjB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,OACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,MACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GAET,KAAM,IAAI,OAAM,cACZ,EAAU,oDAKhB,GAAM,IAA6B,IAK7B,GAAyB,IAC/B,cACE,MAAI,GAAA,MAAM,OAAO,QAAU,KAClB,KAED,EAAA,MAAM,OAAO,OAAO,OAAS,EAAA,MAAM,OAAO,OAAO,MACjD,OAAO,iBACX,GAAyB,KAAO,KAM/B,GAAM,IAA8B,mBAEL,GAAA,EAAA,GAkCpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GA/BM,EAAA,YAAc,GAAI,SAGlB,EAAA,gBAAkB,GAAI,SAGtB,EAAA,aAAe,GAAI,SACnB,EAAA,cAAgB,EAOhB,EAAA,aAAe,EAEf,EAAA,eAAiB,EAUjB,EAAA,kBAAoB,GACpB,EAAA,sBAAwB,GAgXxB,EAAA,eAAiB,EAq+DjB,EAAA,SAAW,GAj1Eb,CAAC,EAAA,MAAM,QAAQ,aACjB,KAAM,IAAI,OAAM,yCAGlB,GAAI,GAAS,MACX,GAAM,GAAK,GAAgB,EAAA,MAAM,UAAU,kBAC3C,EAAK,YAAc,GAAe,EAAA,MAAM,UAAU,kBAClD,EAAK,MAAQ,GAAI,IAAa,GAC9B,EAAK,OAAS,EAAG,OACjB,EAAK,oBAAsB,OAE3B,GAAK,MAAQ,EACb,EAAK,YAAc,GACnB,EAAK,oBAAsB,GAC3B,EAAK,OAAS,EAAM,GAAG,OAEzB,SAAK,eAAiB,GAAI,IAAe,EAAK,OAC9C,EAAK,mBAAqB,KAE1B,EAAK,QAAU,GAAI,GAAA,YAAY,EAAM,EAAA,YAGvC,SAAA,UAAA,WAAA,WACE,MAAO,MAAK,QAAQ,aACf,MAAK,WAAa,KAAK,WAAW,aAAe,GAClD,KAAK,gBAGX,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAK5C,GAJA,AAAI,GAAA,MAAM,QAAQ,mCACd,EAAA,MAAM,QAAQ,WAChB,KAAK,uBAAuB,GAE1B,IAAU,aAAe,GAAU,KACrC,KAAM,IAAI,OACN,yEAGN,GAAM,GAAS,GACf,YAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,IAC1D,GAIT,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,YAIV,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,aAIZ,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAK3D,GAHA,AAAI,EAAA,MAAM,QAAQ,UAChB,KAAK,uBAAuB,GAE1B,IAAU,YACZ,KAAM,IAAI,OACN,yEAGN,KAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,KAGnE,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAc,KAAK,QAAQ,IAAI,GAErC,EAAY,WAEZ,AAAI,EAAY,SAAW,GACzB,KAAK,YAAY,KAKvB,EAAA,UAAA,SAAA,SAAS,GACP,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,eAAgB,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SAKpD,GAAI,GAAS,MACX,GAAI,GAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEtC,GAAM,GACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,SAAS,EAAI,QAC/B,YAAK,8BAA8B,GAC5B,EAET,GAAI,GAAU,KACZ,MAAO,MAAK,qBAAqB,GAEnC,GAAI,IAAU,SACZ,MAAO,GAET,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GACJ,GAAI,IAAU,aACZ,GAAM,GAAa,EAAe,KAAK,WACjC,EAAa,EAAe,KAAK,WACvC,EAAS,EAAA,aAAa,uBAAuB,EAAY,OAEzD,GAAS,KAAK,qBAAqB,GAGrC,MAAI,IACF,MAAK,gBAAkB,EAAA,KAAK,MAAQ,GAE/B,KAAK,qBAAqB,EAAQ,IAGrC,EAAA,UAAA,KAAN,SAAW,6IACT,GAAI,KAAK,YAAY,IAAI,GACjB,SAAc,KAAK,YAAY,IAAI,GACzC,CAAA,EAAO,GAAI,SAAoB,SAAA,GAAW,MAAA,GAAY,KAAK,MAQ7D,GANM,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAyD,EAAO,OAAxD,EAAiD,EAAO,MAAjD,EAA0C,EAAO,MAA1C,EAAmC,EAAO,MAAnC,EAA4B,EAAO,eAAnB,EAAY,EAAO,SAKnE,GAAS,KACP,SAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEhC,EACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,KAAK,EAAI,QAC3B,KAAK,8BAA8B,GACnC,CAAA,EAAO,GAGT,GAAI,GAAU,KACZ,MAAA,CAAA,EAAO,KAAK,qBAAqB,IAGnC,GAAI,CAAC,EAAA,MAAM,QAAQ,iCACf,EAAA,MAAM,UAAU,mBAAqB,EACvC,KAAM,IAAI,OACN,sGAIF,GAAS,KAGb,AAAI,IAAU,aAAe,EAAA,MAAM,IAAI,2BAErC,GAAoB,KAAK,OAAO,GAC1B,EAAU,KAAK,QAAQ,IAAI,EAAkB,QAEnD,EAAS,GAAA,KAAK,OAAM,wBAAuB,MAAA,EAAA,CACvC,EAAQ,SAAO,OAAK,GAA0B,MAGpD,KAAK,YAAY,IAAI,EAAQ,IAEzB,IAAU,YAEZ,CAAA,EAAM,KAAK,MAAM,yBAFf,CAAA,EAAA,UAEF,EAAA,8BAKE,KAAU,YACD,CAAA,EAAM,QAAQ,IACrB,CAAC,EAAe,KAAK,OAAQ,EAAe,KAAK,UAFnD,CAAA,EAAA,UACI,SAAK,EAAA,OAEL,EAAa,EAAG,GAChB,EAAa,EAAG,GACtB,EAAO,EAAA,aAAa,uBAChB,EAA4B,gBAC3B,AAAI,GAAU,KACnB,EAAO,KAAK,qBAAqB,GAE3B,GAAO,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,MAAM,gCAAgC,EAAQ,qBAE5D,MAAI,IAAqB,MACvB,KAAK,8BAA8B,GAE/B,EAAY,KAAK,qBAAqB,EAAQ,GAE9C,EAAc,KAAK,YAAY,IAAI,GACzC,KAAK,YAAY,OAAO,GAGxB,EAAY,QAAQ,SAAA,GAAW,MAAA,GAAQ,KACvC,AAAI,KAAK,gBAAgB,IAAI,IAC3B,MAAK,gBAAgB,OAAO,GAC5B,KAAK,YAAY,GACjB,KAAK,kBAEP,CAAA,EAAO,SAGD,EAAA,UAAA,uBAAR,SAA+B,GAC7B,GAAI,GAAU,KACZ,OAEF,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,KACjC,GAAM,GAAM,EAAO,GACnB,GAAI,CAAC,GAA4B,GAC/B,KAAI,GAAA,MAAM,QAAQ,gCACV,MACF,aAAa,EAAG,iJAIhB,MAAM,aAAa,EAAG,4CAK1B,EAAA,UAAA,qBAAR,SAA6B,SACrB,EAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SACf,EAAO,EAAA,KAAK,cAAc,GAChC,GAAI,EAAA,MAAM,QAAQ,iCAChB,GAAM,GAAY,KAAK,OAAO,GACxB,EAAU,KAAK,QAAQ,IAAI,EAAU,QACrC,EAAO,GAAA,KAAK,OACA,gCAA+B,MAAA,EAAA,CAC5B,EAAQ,SAAO,OAAK,GAA0B,KACjD,SAAS,EAAG,GAE9B,YAAK,8BAA8B,GAE5B,EAGT,GAAM,GACF,EAAA,MAAM,QAAQ,eAAiB,IAAa,GAC1C,EACF,EAAyB,GAAwB,GAAS,EACxD,EAAU,EACZ,GAAI,IAAyB,GAC7B,GAAI,IAAmB,GACrB,EAAS,KAAK,gBAChB,EAAS,CAAC,CAAC,MAAO,EAAa,MAAK,EAAE,OAAM,IAAI,WAC9C,EAAU,KAAK,QAAQ,IAAI,EAAO,QAClC,EACF,KAAK,MACA,gDACG,EAAQ,QAAS,EAAQ,SAAS,GAAI,EAAQ,SAAS,IAC1D,SAAS,EAAG,GACrB,YAAK,8BAA8B,GAE5B,GAGH,EAAA,UAAA,KAAN,SAAW,qHACH,GAAkB,KAAK,aACvB,EAA+B,GAEjC,EAAgB,GACpB,AAAI,KAAK,oBAAsB,KAC7B,MAAK,mBAAqB,EAC1B,EAAgB,IAEhB,KAAK,aAAa,KAAK,GAEzB,KAAK,aAAe,EAEpB,IAGM,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,SACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OACpB,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,QACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OAE1B,KAAK,aAAe,EAEpB,AAAI,GACF,MAAK,mBAAqB,MAGtB,EAAuB,CAC3B,aAAc,KAAK,aACnB,eAAgB,KAAK,eACrB,SAAU,KACV,OAAQ,MAGN,EAAA,MAAM,UAAU,iDAAmD,EACpD,CAAA,EAAM,QAAQ,IAAI,IADjC,CAAA,EAAA,UACI,SAAW,EAAA,OAEjB,EAAI,SAAc,EAAA,KAAK,IAAI,GAC3B,EAAI,oBAAyB,WACzB,MAAA,GAAS,IAAI,SAAC,EAAG,GAAM,MAAC,CAAC,KAAM,EAA0B,GAAI,GAAI,KAC5D,IAAI,SAAA,GAAK,MAAG,GAAE,KAAI,KAAK,EAAE,KACzB,KAAK,oBAEd,EAAI,SAAc,CAChB,MAAO,8EAIX,YAAK,aAAe,EACpB,KAAK,eAAiB,EACtB,CAAA,EAAO,SAET,EAAA,UAAA,OAAA,WACE,MAAO,CACL,WAAY,GACZ,cAAe,KAAK,cACpB,uBAAwB,KAAK,eAAe,kBAC5C,kBAAmB,KAAK,eAAe,eAInC,EAAA,UAAA,WAAR,WACE,MAAI,GAAA,MAAM,UAAU,iDAAmD,EAC9D,KAAK,MAAM,aAEb,CAAC,QAAS,EAAA,KAAK,MAAO,MAAO,OAG9B,EAAA,UAAA,SAAR,SAAiB,GACf,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,MAAK,MAAM,WACJ,GAER,GAAwB,MAAQ,EAAA,KAAK,MAC/B,IAGK,EAAA,UAAA,aAAd,SAA2B,4EACzB,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,CAAA,EAAO,KAAK,MAAM,uBAAuB,IAErC,GAAa,EACnB,CAAA,EAAO,EAAW,MAAQ,EAAW,eAKvC,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,gBAAgB,IAAI,GAC3B,OAEF,GAAI,KAAK,YAAY,IAAI,IACvB,KAAK,gBAAgB,IAAI,GACzB,KAAK,iBACL,OAGF,GAAI,CAAC,KAAK,QAAQ,IAAI,GACpB,OAGF,KAAK,eAAe,GACb,GAAA,GAAA,KAAA,QAAA,IAAA,GAAA,eACP,AAAI,GAAkB,MACpB,GAAe,KAAK,UACpB,EAAe,KAAK,WAEtB,KAAK,QAAQ,OAAO,IAGd,EAAA,UAAA,eAAR,SAAuB,GACf,GAAA,GAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAE5C,EAAM,GAAS,EAAM,YAAc,EACnC,EAAW,KAAK,aAAa,IAAI,GACvC,AAAI,EAAW,EACb,KAAK,aAAa,IAAI,EAAK,EAAW,GAEtC,MAAK,aAAa,OAAO,GACzB,AAAI,GAAW,MACb,MAAK,eAAiB,KAAK,aAAa,EAAU,GAClD,KAAK,eAAe,eAAe,EAAS,EAAU,EAAO,KAGjE,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,QAAU,KAClB,EAAQ,SAAW,KACnB,EAAQ,SAAW,GACnB,EAAQ,MAAQ,MAGlB,EAAA,UAAA,WAAA,SAAW,GACT,YAAK,YAAY,GACV,KAAK,QAAQ,IAAI,GAAQ,SAOlC,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,MAAK,QAAQ,IAAI,IAGlB,EAAA,UAAA,cAAR,WACE,MAAK,GAAA,MAAM,QAAQ,qBAInB,CAAI,KAAK,YAAc,MACrB,MAAK,WAAa,EAAA,SAAS,YAAY,QAGlC,KAAK,YAPH,MAiBX,EAAA,UAAA,mBAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAa,KAAK,gBACxB,MAAI,CAAC,KAAK,uBAAyB,GAAc,MAC/C,SAAQ,KACJ,6NAKJ,KAAK,sBAAwB,IAGxB,GAAc,MACjB,EAAO,MACH,SAAA,GAAS,MAAA,GAAK,QAAQ,IAAI,EAAM,QAAQ,SAAW,MAC/C,EAAA,KAAK,cAAc,EAAM,OAAS,KAGhD,EAAA,UAAA,gBAAA,WACE,MAAO,MAAK,OAGd,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,GAAM,GAAS,KAAK,WAAW,EAAK,MAAO,aACrC,EAAa,KAAK,QAAQ,IAAI,EAAO,QAI3C,SAAW,eAAiB,CAC1B,KAAM,EAAA,SAAS,KAAK,EAAK,SACzB,KAAM,EAAA,SAAS,KAAK,EAAK,UAGpB,GAET,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAExC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAGxC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAO,EAAM,EAAE,MAChE,EAAE,OACN,MAAO,MAAK,WAAW,EAAM,EAAE,MAAO,GAGxC,GAAI,EAAA,KAAK,cAAc,KAAU,EAC/B,MAAO,GAAA,OAAO,GAAI,EAAM,EAAE,OAErB,GAAA,GAAA,KAAA,QAAA,IAAA,EAAA,QAAA,SACD,EAAc,EAAA,WAAW,iBAAiB,EAAE,MAAO,EAAO,GAChE,GAAI,GAAY,CAAC,GACf,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAmB,GACvB,GAAI,IAAa,GACf,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,GAEhD,YAAK,YAAY,EAAE,QACZ,KAAK,aAAa,EAAG,EAAO,IAG7B,EAAA,UAAA,aAAR,SAAqB,EAAW,EAAiB,GAC/C,GAAM,GAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAI,KAAK,WAAW,EAAM,EAAE,OAC5B,EAAa,KAAK,QAAQ,IAAI,EAAE,QAEtC,OAAO,OAAO,EAAY,GAC1B,EAAW,MAAQ,EACnB,EAAW,MAAQ,EAAE,MACrB,GAAI,GAAa,EAAA,WAAW,kBAAkB,EAAO,EAAE,SACvD,AAAI,EAAS,OAGX,IAAc,EAAS,MAAM,YAE/B,EAAW,MAAQ,CACjB,WAAU,EAEV,WAAY,EAAS,OAAS,EAAS,MAAM,YAAc,EAAE,QAI/D,GAAM,GAAW,KAAK,aAAa,IAAI,EAAW,MAAM,aAAe,EACvE,YAAK,aAAa,IAAI,EAAW,MAAM,WAAY,EAAW,GAEvD,GAGT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GAD1C,GAAA,GAAA,KAEQ,EAAS,KAAK,mBAChB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,aAAa,EAAG,EAAO,EAAK,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAO,GAAI,GAGpB,GAAM,GAAU,GAAI,IAAoB,EAAO,EAAS,GACxD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAqB,EAAE,MAAO,GAClC,GAAI,IAAe,EAAE,MAAO,GAChC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,GAAI,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KAChC,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KACtC,MAAO,GAAA,QAAQ,KAAK,OAAO,EAAO,GAAO,KAAK,OAAO,EAAO,IAG9D,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAEjB,GAAI,EAAQ,OAAS,EAAA,MAAM,UAAU,iCACnC,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,OAAO,EAAQ,MAAM,EAAG,GAAW,GACnD,EAAY,KAAK,OAAO,EAAQ,MAAM,GAAW,GACvD,MAAO,MAAK,OAAO,CAAC,EAAU,GAAY,GAE5C,GAAI,EAAA,MAAM,QAAQ,gCAAkC,EAAQ,GAAG,KAAO,GACpE,GAAM,GAAU,GAAI,IAAoB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACnE,MAAO,MAAK,cAAc,EAAS,GASrC,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACtD,EACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAAK,GAAI,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,OAC3D,EAAU,GAAI,IAAc,EAAU,IAAI,SAAA,GAAK,MAAA,GAAE,SACjD,EAAc,KAAK,cAAc,EAAS,GAChD,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAsB,GAAtB,GAAA,GAAA,KACQ,EAAS,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,IAAI,KACtE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAE/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAM,GAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAA,EAAA,MAAC,EAAA,EAAA,GAIP,GAAK,KAAgB,GAAK,IAAgB,IACtC,EAAY,IACd,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAE1B,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAG1B,GAAM,GAAM,IAAgB,EAAI,EAAI,EAAE,KAAK,EAAO,EAAW,GACvD,EAAO,IAAgB,EAAI,EAAI,EAC/B,EAAM,IAAgB,EAAI,EAAE,KAAK,EAAO,EAAG,GAAa,EAC9D,MAAO,MAAK,SAAS,EAAK,GAAK,IAAI,EAAM,IAG3C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,GAC5D,MAAO,MAAK,cAAwB,EAAS,CAAC,EAAG,GAAI,IAGvD,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE7C,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAA,EAAA,MAAC,EAAA,EAAA,GAED,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,EACxD,EAAS,EAAiB,GACxB,EAAuB,CAAC,EAAG,GACjC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAwB,EAAS,EAAQ,IAGvD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aACd,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OACrD,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OAErD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAExD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAE/C,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,EAGT,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,EAAE,QAG/C,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAQ,EAAM,EAAO,GACnD,GAAI,IAAW,EAAE,MAAO,EAAQ,EAAM,EAAO,GACjD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAM,GACF,GAAI,IAAe,EAAW,MAAO,EAAa,EAAM,EAAO,GACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAY,EAAa,KAG/D,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAI,EAAE,QAAU,UACd,GAAM,GAAO,KAAK,SAAS,EAAE,QACvB,EAAc,EAAK,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,KAC9C,EAAM,EAAA,OAAO,EAAE,MAAO,EAAE,MAAO,GACrC,MAAO,IAAK,EAAK,GAEnB,GAAM,GAAU,GAAI,IAAY,EAAE,MAAO,GACzC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAU,GACxC,GAAI,IAAW,EAAE,MAAO,EAAU,GACtC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAAlD,GAAA,GAAA,KACQ,EAAS,KAAK,mBAChB,CAAC,EAAG,GAAU,WAAM,MAAA,GAAK,WAAW,OAAO,EAAG,EAAS,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAQ,KAAM,GACzD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAEV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAU,EAAE,QAAQ,GAAW,GAC1B,QAAQ,GACR,MAAM,EAAkB,IAGtC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAGV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAAyB,GACzB,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,EAAE,IAAI,GAEhB,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EAAe,EAAA,aAAa,oBAC9B,EAAQ,MAAO,EAAY,EAAM,IAE/B,EAAW,EAAA,UACb,EAAQ,QAAQ,GAChB,GACJ,MAAO,GAAA,QAAQ,EAAU,IAGnB,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,KAAK,KAAK,EAAS,GAC7B,EAAa,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,QAAO,GACpD,EAAU,GAAI,IAAc,EAAY,GACxC,EAAS,KAAK,cAAwB,EAAS,CAAC,GAAI,GAE1D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,OAAO,EAAQ,EAAY,IAGjC,EAAA,UAAA,UAAR,SACI,EAAa,EACb,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACrB,AAAI,GAAgB,MAClB,GAAY,EAAa,MAAM,GAC/B,EAAS,EAAa,MAAM,IAE9B,GAAM,GAAa,EAAA,aAAa,yBAAyB,GACnD,EAAa,CACjB,WAAU,EACV,OAAM,EACN,UAAS,EACT,QAAS,KAAK,KAAK,EAAS,IAExB,EACF,GAAI,IAAiB,EAAY,EAAY,GAAgB,MAC3D,EAAS,CAAC,GAChB,AAAI,GAAgB,MAClB,EAAO,KAAK,GAEd,GAAM,GAAS,KAAK,cAAwB,EAAS,EAAQ,SAE7D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,UAAU,EAAG,EAAY,IAG/B,EAAA,UAAA,gBAAR,SACI,EAAW,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MACtC,GAAM,GAAU,GAAgB,KAAO,EAAa,MAAQ,EAAE,MACxD,EAAS,EAAQ,EAAQ,OAAS,GAClC,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,GAAI,IAChB,EAAS,EAAY,EAAY,GAAgB,MAC/C,EAAS,GAAgB,KAAO,CAAC,GAAK,CAAC,EAAG,GAC1C,EAAS,KAAK,cAAsB,EAAS,EAAQ,SAC3D,MAAI,GAAO,OAAS,EAAE,KACb,KAAK,gBAAgB,EAAG,EAAY,GAEtC,GAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,MAAO,GAAa,QAAQ,IAGtD,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAC/D,GAAI,EACF,MAAO,GAGH,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,OAAQ,GAAa,QAAQ,IAGvD,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAI,GAAO,EACL,EAAc,EAAA,aAAa,mBAAmB,CAAC,GAAO,EAAE,MAC1D,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,EAAA,UAAU,EAAG,GACzB,EAAO,EAAA,aAAa,iBAAiB,EAAG,EAAE,MAAM,IAGlD,GAAM,GACF,GAAa,gBAAgB,EAAU,MAAO,EAAM,GAClD,EAAS,EAAA,KAAK,cAAc,CAAC,EAAU,MAAM,KAC7C,EAAM,EAAU,KAAK,GAAI,GACzB,EAAc,EAAA,WAAc,EAAE,OAChC,EACA,KAAK,aACG,EAAK,qBAAsB,EAAY,EAAa,GACvD,QAAQ,GACjB,MAAI,IAAe,MACjB,GACI,EAAA,UAAU,EAAQ,EAAA,aAAa,uBAAuB,KAErD,GAGD,EAAA,UAAA,aAAR,SACI,EAAa,EAAiC,EAC9C,EAAiB,GACnB,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EACF,GAAa,8BAA8B,EAAQ,GACjD,EAAY,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,YAAW,GACvD,EAAU,GAAI,IAAiB,EAAW,GAC1C,EACF,KAAK,cAAwB,EAAS,CAAC,EAAG,GAAa,GAE3D,MAAI,GAAO,MAAM,KAAO,EACf,EAET,GAAa,EAAA,MAAM,EAAG,GAAa,KAAK,CAAC,EAAS,IAC3C,KAAK,aAAa,EAAQ,EAAW,EAAY,EAAO,KAGzD,EAAA,UAAA,gBAAR,SAAwB,EAAW,EAAc,GAE/C,GAAM,GAAO,CAAC,GAId,GAHA,EAAA,aAAa,2BACT,MAAQ,EAAW,OAAO,GAAG,cAAgB,EAAW,MAAM,GAAI,EAClE,EAAE,MACF,CAAC,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAQ,GAC7C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,UAAU,EAAK,GAAY,QAAQ,GAEjD,MAAO,MAAK,gBAAgB,EAAG,IAGjC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,GAAI,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,kDAAkD,GAAE,KAAO,GAAC,IAC5D,iBAAgB,IAMtB,OAJM,GAAO,EAAE,MAAM,GACjB,EAAS,EAGJ,EAAI,EAAG,GAAK,KAAK,KAAK,KAAK,KAAK,IAAS,EAAG,KACnD,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,GAAO,GAC5C,EAAc,EAAQ,mBAAmB,GACzC,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,GAAS,EAAO,MAAO,GAC7D,EAAW,UAIb,GAAI,GACF,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAW,GAChD,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,IACtC,EAAW,UAGb,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA2B,QAE9D,GAAM,GAAU,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAE,OACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA+B,QAElE,GAAM,GACF,GAAI,IAAgB,GAAwB,EAAE,MAAO,EAAE,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAClE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA0B,QAG7D,GAAM,GAAU,GAAI,IAAgB,GAAmB,EAAE,MAAO,EAAE,OAClE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA6B,QAGhE,GAAM,GAAU,GAAI,IAAgB,GAAsB,EAAE,MAAO,EAAE,OACrE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAAmC,QAE/C,GAAM,GACF,GAAI,IAAgB,GAA4B,EAAE,MAAO,EAAE,OAC/D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAiC,QAEpE,GAAM,GACF,GAAI,IAAgB,GAA0B,EAAE,MAAO,EAAE,OAC7D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAM,GAAU,GAAI,IAAc,EAAU,KAAM,EAAE,MAAO,EAAE,MAC7D,MAAO,MAAK,cACR,EAAS,CAAC,EAAW,EAAG,GAAI,EAAA,WAAW,EAAE,MAAO,EAAE,SAGxD,EAAA,UAAA,MAAA,SAAM,GACJ,EAAA,aAAa,KACT,yEAEJ,GAAM,GAAW,EAAU,WAC3B,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAM,GAAQ,EAAE,WAChB,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAK,GACL,EAAc,QACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAA6B,GAEzC,GAAM,GAAU,GAAI,IAAgB,EAAI,EAAE,MAAO,EAAE,OACnD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAG1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,GAErD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAG7C,EAAA,UAAA,cAAR,SAAsB,EAAe,EAAY,GAC/C,GAAM,GAAU,GAAI,IAAqB,EAAE,MAAO,GAClD,MAAO,MAAK,cAAsB,EAAS,CAAC,GAAI,IAG1C,EAAA,UAAA,eAAR,SACI,EAAe,EAAe,EAAY,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GACF,GAAI,IAAsB,EAAI,EAAE,MAAO,EAAE,MAAO,GACpD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAO7C,EAAA,UAAA,yBAAR,SAAiC,EAAW,EAAW,GAAvD,GAAA,GAAA,KACQ,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAA,4SAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAcP,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,GAMD,EAAA,UAAA,+BAAR,SACI,EAAuB,GACzB,MAAO,CACL,OAAQ,EAAY,OACpB,MAAO,EAAY,MACnB,MAAO,EAAc,QAIzB,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAIjB,GAAI,EAAQ,OAAS,EAAA,MAAM,IAAI,iCAC7B,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,KAAK,EAAQ,MAAM,EAAG,IACtC,EAAY,KAAK,KAAK,EAAQ,MAAM,IAC1C,MAAO,MAAK,KAAK,CAAC,EAAU,IAG9B,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAO,OAAO,SAAC,EAAI,GAAO,MAAA,GAAA,WAAW,EAAI,KAC1D,EAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAE5B,EAAc,EAAA,MAAM,QAAQ,cAC5B,EAAU,EACZ,GAAI,IAAkB,EAAQ,GAAG,MAAO,GACxC,GAAI,IAAY,EAAQ,GAAG,MAAO,GACtC,MAAO,MAAK,cAAiB,EAAS,EAAS,IAGjD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,GAAM,GAAc,EAAA,MAAM,QAAQ,gCAC5B,EAAU,EACZ,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OAC/C,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,MAAO,MAAK,cAAiB,EAAS,CAAC,EAAG,GAAI,IAGhD,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAY,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACnE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAe,EAAE,OAGhD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAG1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAChD,EAAI,KAAK,SAAS,EAAQ,EAAS,QAAQ,IAC3C,EAAI,KAAK,IAAI,GACb,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAI,EAAG,IAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAGtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAE3C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA2B,EAAE,MAAO,EAAM,OAC9C,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAM,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAEtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA6B,EAAG,MAAO,EAAE,OAC7C,GAAI,IAAgB,GAAsB,EAAG,MAAO,EAAE,OAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,UAG1C,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,GAAI,GACJ,AAAI,EAAA,MAAM,QAAQ,mBAChB,EAAU,GAAI,IAAkB,EAAE,OAElC,EAAU,GAAI,IAAY,EAAE,OAE9B,GAAM,GAAc,EAAQ,mBAAmB,EAAK,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,IAGhD,EAAA,UAAA,IAAA,SAAsB,GAEpB,GAAI,KAAK,mBAAmB,CAAC,KAAO,EAAE,QAAU,aAC9C,GAAM,GACF,GAAiB,KAAK,QAAQ,IAAI,EAAE,QAAQ,QAChD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAU,GAAI,IAAkB,EAAE,OAClC,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAG9D,MAAO,MAAK,cAAsB,EAAS,IAG7C,EAAA,UAAA,QAAA,SAA0B,GACxB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,GAAc,IAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,KAG9B,EAAA,UAAA,eAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAGF,GAAM,GAAS,EAAE,MACX,EAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAkB,EAAS,WAC3B,EAAc,EAAO,GAAK,EAAO,GAAK,EAAO,GAC7C,EAAmB,EAAS,YAC5B,EAAiB,EAAS,aAAe,eACzC,EAAa,GACb,EAAa,GAIb,EACD,KAAgB,GAAK,IAAqB,IAC3C,EAAkB,GAChB,EAAyB,EAAO,GAAK,IAAM,GAAK,CAAC,CAAC,EAAS,SAEjE,GAAI,GAA6B,CAAC,EAAA,MAAM,QAAQ,wBAC5C,CAAC,EAAA,MAAM,QAAQ,iCACf,CAAC,GACH,GAAM,GAAc,EAAiB,EAAO,GAAK,EAAO,GAAK,EAAO,GAC/B,EAAO,GAAK,EAAO,GAAK,EAAO,GAC9D,EAAY,EAAA,QAAQ,EAAG,CAAC,EAAG,EAAa,EAAS,aACjD,EACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAS,KAAK,iBAAiB,CACnC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAExB,MAAO,GAAA,QAAQ,EAAQ,EAAS,UAWlC,GAAM,GAAc,EAChB,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACrC,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACnC,EAAwB,CAC5B,OAAQ,EAAE,OACV,MAAO,CAAC,EAAG,EAAa,EAAS,YACjC,MAAO,EAAE,OAUL,EAAwB,EAAS,MACvC,EAAS,MAAQ,EAAS,MAAM,QAChC,EAAS,MAAM,EAAS,MAAM,OAAS,KACvC,EAAA,KAAK,OACD,GAAyB,EAAS,MAAO,EAAU,OACnD,WAAM,MAAA,kBAAkB,EAAS,MAAK,OAClC,EAAU,MAAK,gBACvB,GAAM,GACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAgB,KAAK,iBAAiB,CAC1C,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAElB,EAAuB,KAAK,QAAQ,IAAI,EAAc,QAC5D,SAAA,KAAK,OACD,EAAqB,SACrB,WAAM,MAAA,gDAEV,EAAS,MAAQ,EAGjB,EAAqB,MAAQ,EAAS,SAC/B,EAAA,SAAS,qBACL,EAAc,OAAQ,EAAS,SAAU,EAAc,QAI5D,EAAA,UAAA,iBAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAQA,GAAA,GAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,WAGI,EAAiB,IAAe,eAEhC,EAAY,EAAc,EAAe,EACzC,EAAU,EAAY,EACtB,EAAa,CAAC,EAAW,GACzB,EAAa,GACb,EAAa,GAEb,EAAY,EAAE,QAAQ,CAAC,IACvB,EAAQ,EAAO,QAAQ,CAAC,EAAG,EAAW,KAEtC,EACF,GAAI,IAAoB,EAAY,EAAU,MAAO,GACnD,EACF,KAAK,cAAwB,EAAe,CAAC,IAAY,QAAQ,CAC/D,EAAG,EAAW,GAAI,EAAW,KAG7B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAgB,GAAI,IACtB,EAAO,MAAO,CAAC,EAAG,EAAS,EAAS,aAAc,EAClD,EAAY,EAAS,EAAiB,GACpC,EAAuB,CAAC,EAAQ,GACtC,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEd,GAAM,GAAU,KAAK,cAAwB,EAAe,GAE5D,MAAI,GACK,EAAQ,QAAQ,CAAC,EAAG,EAAW,EAAU,EAAS,cAElD,EAAQ,QAAQ,CAAC,EAAG,EAAS,YAAa,EAAW,KAIhE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAEjD,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAM,MAAM,KAAO,EAC3D,MAAO,MAAK,iBACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAGjD,GAAM,GAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAS,KAC7D,EAAU,GAAI,IAChB,EAAU,EAAS,EAAiB,GAClC,EAAuB,CAAC,EAAO,GACrC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAc,EAAS,IAGrC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eAAe,EAAG,EAAQ,GAExC,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAM,KAAO,EACvD,MAAO,MAAK,iBAAiB,EAAG,EAAQ,GAE1C,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAExC,EAA0B,EAAA,MAAM,QAAQ,6BAC1C,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EAC7C,EAAkB,EACpB,GAA6B,EAAY,GACzC,KACE,EAAmB,CAAC,EAAO,GAE3B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KAC5D,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAGd,GAAI,GACJ,MAAI,GACF,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,IAGrC,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,KAGrC,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,6BACd,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EACjD,GAAU,GAAI,IAA6B,GACpC,KAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,GAAU,GAAI,IAAuB,GAC9B,KAAK,cAAc,EAAS,CAAC,EAAG,MAGzC,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAA+B,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAM,GAAU,GAAI,IAAgC,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAA,aAAa,WAAW,EAAG,EAAO,OAG3C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,KAAK,MAAM,EAAG,EAAO,GAAM,QAAQ,GAE9C,MAAO,IAGT,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GAA2B,GAAI,IAAyB,GAC9D,MAAO,MAAK,cAAc,EAA0B,CAAC,GAAK,EAAE,QAG9D,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAM,GAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,KAAK,cAAc,EAA2B,CAAC,IAC7C,EAA2B,GAAI,IAAyB,GACxD,EAAS,KAAK,cAChB,EAA0B,CAAC,EAAI,GAAqB,EAAE,OAC1D,SAAmB,UACZ,GAGT,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IACA,EAAE,MAAO,EAAW,EAAU,GAClC,GAAI,IAAsB,EAAE,MAAO,EAAW,EAAU,GAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,GAAM,GAAU,GAAI,IAA8B,EAAI,EAAG,GAEzD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,GAAI,IAChB,EAAE,MAAO,EAAW,EAAU,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GACF,GAAI,IAAoC,EAAI,EAAG,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAM,GAAQ,EAAa,EAAS,EAAA,QAAQ,GACtC,EAAY,EAAM,MAAM,GACxB,EAAc,EAAM,MAAM,GAC1B,EAAU,GAAI,IAAmB,EAAW,EAAa,GACzD,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAQ,QAAS,IAGvD,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAM,GAAU,GAAI,IAAc,EAAQ,KAAM,EAAO,EAAS,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAAI,IAAY,EAAE,MAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,GAAM,GAAU,GAAI,IAChB,EAAM,MAAO,EAAM,MAAO,EAAU,EAAQ,GAChD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAO,EAAO,GAAW,YAG/D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAE9D,GAAM,GAAY,EAAE,MAAM,GACpB,EAAe,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAE5D,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAe,IAAe,OAChC,CAAC,EAAW,EAAc,EAAa,GACvC,CAAC,EAAW,EAAa,EAAc,GAErC,EAAU,GAAI,IAAoB,EAAa,EAAW,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAG5C,EAAe,CAAC,EAAa,EAAW,GACxC,EAAiB,EAAQ,QAAQ,CAAC,EAAY,IAC9C,EAAW,EAAQ,QAAQ,CAAC,EAAY,IAE9C,GAAI,IAAe,EACjB,MAAO,GAAA,aAAa,cAAc,EAAA,OAAO,IAAK,GAEhD,GAAM,GAAe,EAAA,OAAO,GACtB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAe,KAAM,EAAS,KAAM,EAC3D,GACE,EACF,KAAK,cAAc,EAAS,CAAC,EAAU,EAAgB,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAGjC,EAAiB,GACjB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAc,KAAM,EAAa,KAAM,EAC9D,CAAC,EAAY,GAAI,GACf,EAAc,KAAK,cACrB,EAAS,CAAC,EAAc,EAAe,IAC3C,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGzB,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGjB,EAAA,UAAA,QAAR,SAAgB,EAAa,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAGxD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAU,KAAK,QAAQ,EAAM,GAAM,KAAK,EAAE,MAAM,GAAI,EAAE,MAAM,IAClE,SAAK,UACL,EAAK,UACE,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAGpC,EAAiB,EAAQ,QAAQ,CAAC,EAAW,IAC7C,EAAW,EAAE,QAAQ,CAAC,EAAE,KAAO,EAAW,IAC1C,EACF,GAAI,IAAgB,EAAW,EAAS,CAAC,EAAW,IAClD,EAAc,KAAK,cAAc,EAAS,CAAC,EAAU,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAG5C,GAFA,EAAQ,GAAS,EAAA,KAAK,WAAW,GAE7B,IAAU,UAEZ,GAAM,GAAS,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IAChE,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,WAEjD,GAAM,GAAU,GAAI,IAAY,EAAO,GACjC,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,GAAI,EAAO,KAIlD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAIhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAE,QAAU,SAAW,GAAK,EAAG,EAAE,QAG7D,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GAEpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGhD,EAAA,UAAA,eAAA,SAAe,EAAiB,EAAiB,GAE/C,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,YAAK,QAAQ,IAAI,GAAQ,MAAQ,KAC1B,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAGtB,EAAA,UAAA,WAAR,SACI,EAAiB,EAAiB,GAC7B,GAAA,GAAA,KAAA,eAAA,EAAA,EAAA,GAAA,OACP,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAGrD,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAU,GAAI,IAAc,EAAM,OACxC,MAAO,MAAK,gBAAgB,EAAS,CAAC,GAAQ,EAAM,QAG9C,EAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAU,GAAI,IAAY,EAAM,OAChC,EAA8B,GACpC,MAAO,MAAK,gBACR,EAAS,CAAC,GAAQ,EAAM,MAAO,KAC/B,IAGE,EAAA,UAAA,cAAR,SAAsB,EAAmB,GACvC,GAAM,GAAe,CACnB,GAAuB,EAAM,eAC1B,GAAuB,EAAM,QAE5B,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EAAiB,CACrB,GAAuB,WAAgB,GAAuB,IAG1D,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,KAAK,gBAChB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,QAG1D,EAAA,UAAA,OAAR,SAAe,GACb,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,MAClB,EACF,GAAwB,GACxB,EACJ,AAAI,EACF,EAAU,GAAI,IAA0B,GAExC,EAAU,GAAI,IAAoB,GAEpC,GAAM,GAAgC,GAChC,EAAM,KAAK,gBACb,EAAS,CAAC,CAAC,MAAO,EAAW,MAAK,EAAE,OAAM,IAAI,EAC9C,KAAwB,GAC5B,MAAO,CAAC,MAAK,EAAE,MAAK,EAAE,OAAQ,EAAI,SAGpC,EAAA,UAAA,gBAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAHJ,GAAA,GAAA,KAGI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,KAAK,eAAe,EAAQ,YAAa,GAClD,EAAU,KAAK,QAAQ,IAAI,EAAO,QAIxC,GAHA,AAAI,EAAQ,cACV,GAAQ,SAAW,IAEjB,EAAQ,mBAAqB,GAAuB,OACtD,GAAM,GAAa,GAA0B,EAAQ,aAKrD,EAAQ,SAAW,EAAW,IAAI,SAAA,GAAK,MAAA,GAAI,IAK7C,GAHA,AAAI,EAAQ,aAAe,MACzB,GAAQ,MAAQ,EAAQ,aAEtB,EAAA,KAAK,cAAc,EAAO,SAAW,EAGvC,SAAQ,OACJ,EAAA,KAAK,uBAAuB,EAAO,MAAoB,GACpD,EAGT,GAAM,GAA8B,GAC9B,EAA2B,EAAO,IAAI,SAAA,GAC1C,GAAI,EAAM,QAAU,YAClB,KAAM,IAAI,OACN,mIAKN,GAAI,GAAU,EAAK,QAAQ,IAAI,EAAM,QAErC,GAAI,EAAQ,SAAW,MACrB,GAAI,CAAC,EAAQ,cACT,EAAA,KAAK,cAAc,EAAM,QACrB,EAAA,MAAM,UAAU,6BAMtB,MAAO,CACL,MAAO,EAAM,MACb,QAAS,KACT,UAAW,GACX,cAAe,EAAQ,QAM3B,AAAI,EAAQ,cACV,GAAQ,SAAW,GACnB,EAAQ,MAAQ,EAAM,eAEf,CAAC,CAAC,EAAQ,WAAa,CAAC,CAAC,EAAQ,aAC1C,EAAQ,EAAQ,SAAW,EAAK,aAAa,GAClB,EAAK,WAAW,GAC3C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,gBAE/B,EAAQ,UACR,CAAC,GAAyB,EAAQ,MAAO,EAAM,QAQjD,GAAM,GAAa,EACb,EAAc,EAAM,MAE1B,EAAM,MAAQ,EAAQ,MACtB,EAAQ,EAAK,cAAc,EAAiB,GAC5C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,QAEjC,EAAW,MAAQ,EAGrB,SAAK,YAAY,EAAM,QAChB,CAAC,MAAO,EAAM,MAAO,QAAO,EAAE,UAAW,MAGlD,KAAK,YAAY,EAAO,QACxB,GAAM,GACW,CAAC,MAAO,EAAO,MAAO,QAAS,EAAS,UAAW,IAC9D,EAAM,GAAyB,EAAS,EAAY,GACpD,EAAS,KAAK,iBAAiB,EAAK,WACxC,MAAO,IACH,EAAK,MAAO,EAAS,EAAY,KAEjC,EAAoB,KAAK,cAAgB,KAC3C,EAgBJ,GAfA,AAAI,GACF,GAAQ,KAAK,cAGf,GACI,KAAK,MAAO,EAAQ,EAAY,EAAY,GAEhD,EAAc,QAAQ,SAAA,GAAQ,MAAA,GAAK,8BAA8B,KAEjE,AAAI,GACF,GAAQ,KAAK,SAAS,GACtB,KAAK,aAAa,KACd,CAAC,KAAM,EAAQ,YAAY,KAAM,MAAO,KAAK,aAAa,MAG5D,CAAC,EAAA,MAAM,QAAQ,wBAA0B,EAAQ,UACjD,IAAkC,IACpC,GAAM,GAAW,KAAK,aAAa,GACnC,YAAK,8BAA8B,GAC5B,EAET,MAAO,IAGT,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,EAAc,GAAe,EAAO,GAAG,MACvC,GAAM,GAAU,KAAK,gBACjB,EAAS,EAAQ,EAAa,EAC9B,GACJ,MAAO,GAAA,SAAS,qBACL,EAAQ,OAAQ,EAAQ,MAAO,EAAQ,QAG5C,EAAA,UAAA,iBAAR,SAAyB,EAAa,GAEpC,MAAM,KAAO,MAAK,aAChB,MAAK,YAAY,GAAO,KAEnB,KAAK,YAAY,IAG1B,EAAA,UAAA,kBAAA,WACE,MAAO,MAAK,gBAKd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAIF,GAAI,CAAC,EAAA,MAAM,QAAQ,YACjB,GAAM,GAAU,OAAO,KAAK,KAAK,aACjC,EAAQ,QAAQ,SAAA,GACd,EAAK,MAAM,cAAc,EAAK,YAAY,GAAK,cAC/C,MAAO,GAAK,YAAY,KAG5B,KAAK,eAAe,UACpB,AAAI,KAAK,QAAU,MACd,MAAQ,oBAAuB,aAC/B,KAAK,iBAAkB,mBAC1B,KAAK,OAAO,SAEZ,KAAK,OAAS,KAEhB,AAAI,KAAK,qBACP,MAAK,MAAM,QAAU,KACrB,KAAK,MAAM,WAEb,KAAK,SAAW,IAGlB,EAAA,UAAA,eAAA,WAAA,GAAA,GAAA,KACE,MAAI,MAAK,qBAAuB,MAC9B,MAAK,oBAAsB,EAAA,KAAK,WAC9B,GAAI,CAAC,EAAA,MAAM,IAAI,iCAGb,GAAM,GAAY,EAAA,MAAM,QAAQ,SAChC,EAAA,MAAM,IAAI,QAAS,IACnB,GAAM,GAAsB,EAAK,IAAI,EAAA,OAAO,OAAO,WAAW,GAG9D,GAFA,EAAA,MAAM,IAAI,QAAS,GAEf,EAAsB,EACxB,MAAO,IAGX,MAAO,OAGJ,KAAK,qBAGd,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAGlD,EAAA,UAAA,YAAR,SAAoB,SACZ,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAE7C,GAAI,GAAW,KAEb,OAEF,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GAAW,EAAQ,SAMvB,GALA,AAAI,GAAY,MACd,GAAW,GAA2C,EAAO,GAC7D,EAAQ,SAAW,GAGjB,GAAU,MACZ,GAAM,GAAY,GAAwB,GAEtC,EAAO,OACP,EAAQ,EAAS,GAAI,EAAS,EAAS,GACrC,EAAc,YAAkB,YAEtC,AAAI,EACF,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,EAAU,GAAI,IACV,EAAW,CAAC,EAAQ,GAAQ,IAEhC,EACI,GAAI,IAAoB,EAAW,CAAC,EAAQ,GAAQ,GAG1D,GAAM,GAAuB,KAAK,eAAe,CAAC,EAAQ,GAAQ,GAClE,AAAI,EACF,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEjB,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEnB,KAAK,MAAM,2BACP,KAAK,WAAW,EAAqB,QAAS,EAAO,EACrD,GAIJ,GAAM,GAAwB,GACxB,EAAsB,KAAK,gBAC7B,EAAS,CAAC,GAAuB,EAAO,KAAM,GAG5C,EAAgB,KAAK,QAAQ,IAAI,EAAoB,QAC3D,EAAQ,QAAU,EAAc,QAChC,EAAQ,SAAW,EAAc,SACjC,EAAQ,SAAW,EAAc,SACjC,EAAQ,MAAQ,EAAc,MAE9B,KAAK,8BAA8B,GACnC,KAAK,QAAQ,OAAO,EAAoB,QAGxC,EAAQ,OAAS,KACjB,AAAI,GACF,MAAK,cAAgB,EAAA,KAAK,MAAQ,QAGpC,GAAM,GAAa,KAAK,eAAe,EAAU,EAAO,EAAO,GAC/D,EAAQ,QAAU,IAId,EAAA,UAAA,qBAAR,SAA6B,EAAgB,GAE3C,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAEP,YAAK,eAAe,GAEpB,AAAI,GAAiB,MACnB,GAAQ,OAAS,GAAoB,EAAe,IAE/C,EAAQ,QAGT,EAAA,UAAA,eAAR,SACI,EAA4B,EAAuB,EACnD,GAEF,GADA,KAAK,eAAiB,KAAK,aAAa,EAAU,GAC9C,CAAC,KAAK,mBACN,KAAK,cAAgB,KAAK,mBAAqB,KAAO,MACxD,GAAM,GAAM,MAAK,cAAgB,KAAO,MAAM,QAAQ,GACtD,KAAK,kBAAoB,GACzB,QAAQ,KACJ,6BAA6B,EAAE,yCAGrC,MAAO,MAAK,eAAe,eAAe,EAAU,EAAS,IAGvD,EAAA,UAAA,aAAR,SAAqB,EAAyB,GAC5C,MAAO,GAAM,GAAK,EAAM,GAAK,EAAA,KAAK,gBAAgB,IAG5C,EAAA,UAAA,mBAAR,SACI,EAAsB,GACxB,GAAI,KAAK,mBAAmB,GAC1B,IACE,MAAO,WACA,GACP,GAAI,EAAA,MAAM,QAAQ,WAChB,KAAM,IAAI,OAAM,yBAItB,MAAO,OAEX,GAriFsC,EAAA,eAuiFtC,YACI,EAAiB,GACnB,GAAI,IAAU,WAAa,IAAU,YACnC,MAAO,GACF,GAAI,IAAU,SAAW,IAAU,QAGxC,OAFM,GAAU,IAAU,QAAW,GAAI,YAAW,EAAE,QACjB,GAAI,YAAW,EAAE,QAC7C,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,KAAK,MAAM,EAAE,IAE3B,MAAO,OAEP,MAAM,IAAI,OAAM,iBAAiB,GChwFrC,GAGM,IAAU,QCHhB,cAmCE,EAAA,MAAM,IAAI,2BAA4B,ICnCxC,AAuBA,AAAI,EAAA,YAAY,aACd,EAAA,gBAAgB,QAAS,WAAM,MAAA,IAAI,KAAoB,MAQ5C,IAAQ,CAAC,eAAc,ICzBvB,GAA0B,0BAE1B,GAA2B;;;EAK3B,GAAkC;;;;;cAWf,GAC9B,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EACD,EAAe,EACf,EAAU,GAAI,IAAe,EAAE,MAAO,GAC5C,MAAO,GAAa,gBAAgB,EAAS,CAAC,GAAI,EAAE,oBAepD,EAAmB,EACnB,EAA4C,GAG9C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAe,EACf,EAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,EAAiB,EAAE,MAAO,EAAE,MAC5B,CAAC,CAAC,GACN,GAAI,IAAgB,EAAW,EAAE,MAAO,EAAE,OACxC,EAAS,GAAS,EAAE,MACpB,EAAS,EAAa,gBAAgB,EAAS,CAAC,EAAG,GAAI,GAC7D,MAAO,IC3DX,AAsBA,GAAM,IAAQ,GAA2B;;EAInC,GAAe;;;IAIjB,GAAkC;;EAIzB,GAAkB,GAAiB,GAAO,IAE1C,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,QACb,WAAY,ICvCd,YAqBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,QACb,WAAY,ICjCd,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,WAG/C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBACvB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAyB,GAAI,IAAyB,GAC5D,MAAO,GAAQ,gBAAgB,EAAwB,CAAC,GAAK,EAAE,OAG1D,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,IC3Cd,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,YAAwB,GAOtB,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,MACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,MACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;yBAKK,EAAa;wBACd,EAAY;2DACuB,EAAe;;;MAK1E,MAAA,MC5DA,AAoBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,YACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,YACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;wBAEI,EAAa;uBACd,EAAY;;;;;;yDAMsB,EAAe;;;;MAMxE,MAAA,MChEA,AAuBO,GAAM,IAIM,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,EAAA,EAAG,EAAA,EAAA,KAAM,EAAA,EAAA,SAAU,EAAA,EAAA,OAAQ,EAAA,EAAA,MAElC,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGL,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAGpB,GAAM,GAAc,CAAC,EAAG,EAAM,GAE1B,EAAc,KAClB,AAAI,GAAU,MACZ,GAAc,EAAO,MACrB,EAAY,KAAK,IAGnB,GAAI,GAAa,KACjB,AAAI,GAAS,MACX,GAAa,EAAM,MACnB,EAAY,KAAK,IAGnB,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACJ,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACF,EACF,EAAQ,gBAAgB,EAAS,EAAa,EAAY,GAAG,OAEjE,MAAO,IAGI,GAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,QACb,WAAY,IC/Ed,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAuBA,GAAM,IAAM;;;;eAQN,GAAa;;;;;;;;;;;;;;;;;;EAoBN,GAAgB,GACzB,GAAK,GAAY,IAER,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,ICzDd,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAItB,GAAM,GAAa,EAAW,GAC9B,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;yBAKK,EAAU;;uCAEI,EAAU;;;;;;;MASjD,MAAA,MC5CA,AAuBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,MACD,EAAe,EAEf,EAAU,GAAI,IAAsB,EAAmB,OACvD,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KChCX,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;uDAMmC,EAAK,OAAO,EAAM;;wBAEjD,EAAK,UAAS;;;;;;;;;;;;;;MAgBtC,MAAA,MCrDA,AAoBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;;8BAeU,EAAK,OAAO,EAAM;4BACpB,EAAK,UAAS;;;;;;;;;;;;;;;;UAgBhC,EAAK,OAAM;;MAIrB,MAAA,MCnEA,AA0BO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,QACb,WAAY,IAGV,GAEJ,YAAoB,GAKX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACnB,EAAA,EAAA,OACE,EAAA,EAAA,YAED,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAOR,EAA6B,CAAC,EAAQ,GACtC,EAAW,CAAC,EAAQ,EAAO,GAEjC,AAAI,IAAW,IACb,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAGpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA+C,EAAG,EAAG,EAAO,GAChE,EAAS,GAAoB,QAG/B,GAAM,GAAkB,EAAQ,eAAe,EAAU,SAEzD,EAAQ,QAAQ,IAAI,EAAgB,QAAQ,MAAQ,GAAa,OACjE,EAAQ,MAAM,yBACV,EAAQ,WAAW,EAAgB,QAAS,GAChD,GAAM,GAAU,EAAA,MAAM,QAAQ,cAC1B,GAAI,IAAwB,GAC5B,GAAI,IAAkB,GACpB,EAAM,EAAQ,gBAAgB,EAAS,CAAC,GAAkB,SAChE,SAAQ,YAAY,EAAgB,QAC7B,EC/ET,AA0BA,YAA4B,GAI1B,OAFM,GAAS,GAER,EAAO,SAAW,GAAK,EAAO,EAAO,OAAS,GAAG,UAAY,IAClE,GAAM,GACF,EAAO,OAAS,EAAO,EAAO,OAAS,GAAG,QAAU,EAAQ,GAC1D,EAAa,EAAA,aAAa,yBAAyB,GACzD,EAAO,KAAK,CACV,OAAQ,EACR,WAAU,EACV,QAAS,KAAK,KAAK,EAAU,KAIjC,MAAO,eAIL,EAAe,EAAiB,EAChC,GAIF,OAHM,GAAkB,GAAmB,EAAE,OAEzC,EAAS,EACJ,EAAI,EAAG,EAAI,EAAgB,OAAQ,KACpC,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,WAAY,EAAA,EAAA,QAErB,EAAU,GAAI,IAChB,CAAC,WAAU,EAAE,OAAM,EAAE,UAAW,EAAE,MAAM,GAAI,QAAO,GAAG,GACpD,EAAiB,EACvB,EAAS,EAAQ,gBAAgB,EAAS,CAAC,GAAS,GAEpD,AAAI,EAAe,SAAW,EAAE,QAC9B,EAAQ,YAAY,EAAe,QAIvC,MAAO,GC/DT,YAwBI,EAAmB,EACnB,GACF,GAAM,GACF,CAAC,GAAY,EAAM,QAAM,OACrB,GAAY,EAAM,QACpB,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EACF,CAAC,GAAY,IAAW,OACpB,GAAY,IAEd,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,EAAQ,gBACnB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,OC3ClE,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MACD,EAAe,EAEf,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QAC5C,MAAI,GAAS,UAAY,CAAC,GAAc,EAAE,MAAO,IAC7C,CAAE,GAAS,UAAY,MAAQ,GAAc,EAAS,MAAO,IACxD,GAAc,EAAG,EAAQ,GAGlC,GAAa,OAAO,EAAE,QAEf,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,QAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICzDd,YAwBI,EAAe,EAAuB,EACtC,GACF,GAAM,GAAS,EAAA,KAAK,cAAc,GAC5B,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAY,EAAQ,EACpB,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAC,GAAG,MAAO,CAAC,MAAO,CAAC,EAAW,IAAU,QAAO,IAEhE,EAAU,GAAO,EAAe,EAAE,MAAO,MAAO,GAChD,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAU,MAAO,CAAC,MAAO,GAAW,QAAO,IAEpE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,ECvCT,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAEjC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAW,GAAkB,GAEnC,KAAK,SAAW;;QAEZ,EAAK;uBACU,EAAQ;;MAI/B,MAAA,MAEA,YAA2B,GACzB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,sBAAsB,EAAI,yBAKxC,OAHM,GACF,CAAC,UAAW,UAAW,UAAW,UAAW,UAAW,WACtD,EAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAe,EAAO,IAAM,EAAc,GAE5C,MAAO,GAAe,OCxDxB,AAqBA,GAAA,IAAA,WAQE,WAAY,EAAkB,GAP9B,KAAA,cAAgB,CAAC,KAIjB,KAAA,aAAe,GACf,KAAA,aAAe,GAIb,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAIjC,GAFA,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACpB,KAAK,KAAO,EACd,KAAM,OACF,6BAA6B,KAAK,KAAI,0BAM5C,OAJM,GAAQ,GAAkB,KAAK,MAE/B,EAAc,GAAe,KAAM,KAAK,MACxC,EAAgB,GAAI,OAAM,KAAK,MAC5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAc,EAAO,IAAM,EAAY,GAEzC,GAAM,GAAY,QAAQ,EAAc,MAAM,IAAI,OAAM,IAClD,EACF,KAAK,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAC3D,EAAO,mBAAmB,EAAc,OAAM,MAAM,EAAS,IAEnE,KAAK,SAAW;;QAEZ,EAAK;;oBAEO,EAAI;WACb,EAAU;sBACC,EAAI;;UAEhB,EAAY,KAAK,KAAO,GAAE;aACvB,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAAE;sBACjD,EAAI;aACb,EAAU;wBACC,EAAI;;;;;MAO5B,MAAA,MCvEA,YAyBI,EAAe,EAAgB,GACjC,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAuB,EAAE,MAAO,GACpC,GAAI,IAAiB,EAAE,MAAO,GAClC,MAAO,GAAQ,gBAAgB,EAAS,CAAC,GAAI,EAAE,OC7BjD,AA0BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAW,EAAA,KAAK,eAAe,EAAkB,EAAE,OACrD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACrD,EAAuB,GAAgB,KACvC,EAAqB,EAAa,mBAAmB,CAAC,IAExD,EAAW,EACf,GAAI,GACF,GAAI,GAKF,OAJM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAa,IAErC,GAAM,GACF,GAAiB,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAc,GAE7D,EAAW,EAAa,eAAe,EAAU,EAAE,OACnD,GAAM,GAAe,EAAa,QAAQ,IAAI,EAAS,QACvD,EAAa,OAAS,MAEtB,GAAW,GAAc,EAAG,EAAc,GAG5C,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAGpD,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGhB,EAAW,EACf,AAAI,GAEF,GAAW,EAAA,aAAa,qBAAqB,EAAa,IAG5D,GAAI,GACJ,GAAI,GACF,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAY,GACd,EAAQ,EAAA,KAAK,cAAc,GAAc,EAAU,EAAE,OAEzD,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAQ,EAAU,EAAa,EAAU,GAGjD,MAAI,IACF,EAAa,8BAA8B,GAGtC,IC9FX,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,EAAE,OAGjD,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAuBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAC3B,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,EAAQ,gBAAgB,EAAyB,CAAC,GAAI,EAAE,OAEtD,EAAyB,GAAI,IAAyB,GACtD,EAAS,EAAQ,gBACnB,EAAwB,CAAC,EAAI,GAAmB,EAAE,OACtD,SAAQ,8BAA8B,GAC/B,EAGF,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,ICrDd,YAuBI,EAAe,EACf,EACA,GACF,GAAI,GAAU,GAAI,IAAc,EAAU,MAAO,IAC3C,EAAa,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAEzD,EAAU,GAAI,IAAc,EAAU,MAAO,GAAM,GAAM,GACzD,GAAM,GAAc,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAC1D,MAAO,CAAC,EAAY,GC/BtB,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAe,EAErB,EAAA,KAAK,OACD,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,uDACF,EAAE,MAAM,OAAM,MACtB,GAAM,GAA8B,CAAC,EAAG,GACxC,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,GAET,EAAA,GAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAEf,MAAO,CAAC,EAAQ,KCjDpB,AAoBO,GAAM,IAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAG9B,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EAE1B,MAAO,GAAA,aAAa,wBAChB,EAAW,EAAY,EAAkB,EACzC,KC3CR,AAkBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC1C7B,AAmBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAqBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EACA,GAPJ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAOtB,GAAM,GAAc,EAAW,GACzB,EAAa,EAAW,GACxB,EAAY,KAAK,IAAI,GAAS,QAAQ,GACtC,EAAY,KAAK,IAAI,GAAS,QAAQ,GAC5C,KAAK,YAAc,EAEb,GAAA,GAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAgB,EAAQ,QAAQ,GAChC,EAAgB,EAAQ,QAAQ,GAElC,EAAc,GAClB,AAAI,MAAO,IAAc,SACvB,EAAc,uBAAuB,EAAU,QAAQ,GAAE,IAEzD,EAAc;2BACO,EAAU,KAAK,KAAI;8CAI1C,KAAK,SAAW;;;;;4CAKwB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;4CACpB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;iDACf,EAAa;iDACb,EAAa;YAClD,EAAW;uCACgB,EAAU,+BACzC,EAAW;;;;;MAOnB,MAAA,MCtEA,AAuBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAe,EAEf,EAAU,GAAI,IACf,EAAmB,MAAO,EAAS,EAAW,GAC7C,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KClCX,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAqBA,GAAM,IAAS,gBAEF,GAAmB,GAAgB,IAEnC,GAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IC5Bd,AAqBA,GAAM,IAAqB,4BAEd,GACT,GAAiB,GAAoB,IAE5B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,IC7Bd,AAqBA,GAAM,IAAM,iBAEC,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC5Bd,AAwBO,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,QACb,WAAY,SAAC,GAQX,OARY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACA,EAAA,EAAA,KACD,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAI,GACJ,GAAI,EAAa,mBAAmB,CAAC,KACnC,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QACtC,EAAS,EAAS,OAClB,EAAY,GAAa,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAE/D,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAc,EAAG,EAAM,GAE/B,MAAO,KCnDX,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAGpB,QAAQ,KACJ,YACA,8DACJ,GAAM,GAAS,EAAQ,SAAS,EAAE,QAC5B,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IChDd,AA4CA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,+RCxEjB,AAGA,GAAM,IAAU,QCHhB,GAsCa,IAAU,CACrB,YAAa,GAAA,aACb,mBAAoB,GAAA,YACpB,qBAAsB,GAAA,cACtB,YAAa,GAAA,aACb,cAAe,GAAA,eACf,iBAAkB,GAAA,kBAClB,KAAQ,yZC7CV,oBAAM,IAAK,KAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,GAAG,IAAI,EAAI,WAAY,GAChC,EAAO,GAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,GAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,GAAG,IAAI,EAAW,GAC5B,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,GAAG,IAAI,EAAU,GACtC,EAAoB,GAAG,IAAI,EAAS,GACpC,EAAc,GAAG,IAAI,EAAoB,GACzC,EAAS,GAAG,IAAI,EAAmB,GACnC,EAAO,GAAG,IAAI,EAAmB,GACjC,EAAkB,GAAG,IAAI,EAAQ,GACjC,EAAgB,GAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,IAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,IAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,GAAG,SAAS,KAAK,aAChC,KAAK,UAAY,GAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,GAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,GAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,GAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,GAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,GACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,GAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,IAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,kBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,oBAAM,IAAK,KAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,GAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,YAAc,KClDtB,kBAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,GAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,GAAQ,wBAA0B,KCvFlC,eACA,KAAM,IAAK,KACL,GAAW,KACX,GAAY,KACZ,GAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,GAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,GAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,GAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,GAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,GAAU,yBAAyB,GACtD,EAAkB,GAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,GAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,GAAS,YAAY,GAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,GAAS,WAAW,GACpC,GAAI,GAAO,GAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,GAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,GAAS,oBAAoB,EAAe,GACxD,EAAc,GAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,GAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,GAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,GAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,GAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,GAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,EAAgB,KAAM,GAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,EAAa,QAAS,EAAiB,KAAM,GAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,EAAkB,KAAK,UAAU,QAAQ,GAAG,OAAO,CAAC,EAAa,KACjE,EAAqB,EAAe,WAC1C,EAAe,UACf,KAAM,GAAc,EAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,EAAkB,KAAM,GAAsB,KAAK,aAAa,EAAa,EAAY,EAAgB,IACtH,EAAe,EAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,EAAmB,KAAM,GAAuB,KAAK,aAAa,EAAc,EAAa,GAC1G,EAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,GAAiC,GAC5C,IAAsB,EAAW,EAAkB,QACnD,GAAsB,EAAW,EAAmB,UAE/C,AAAI,EAAgC,EACzC,GAAsB,EAAW,EAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,EAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,GAAyB,KAAK,sBAAsB,EAAW,EAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,EAAoB,SAC1F,EAAY,EAAU,OAAO,GAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,GAAG,QAAQ,GACX,KAAM,GAAe,GAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,GAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,GAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,GAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,kBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,0CAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,oBAAM,IAAK,KACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QAAS,GAAQ,GAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,GAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,GAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,oBAAM,IAAK,KAEL,GAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,GAAG,QAAQ,WAAW,GAC/B,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,IAAO,KAAK,IAAO,IAAM,KAAM,IAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,GAAO,IAGhB,kBAA0B,GACxB,MAAK,IAAO,QAAQ,IAAO,OAAS,KAAM,IAAG,eAAe,EAAO,KAAK,OAAO,YACxE,GAAO,OAGhB,kBAAuB,EAAO,GAM5B,GALA,AAAI,GAAQ,EAAO,KAAK,IAAI,WAC1B,GAAQ,EAER,IAAS,EAEP,KAAU,EAAG,MAAO,IACxB,GAAI,GACJ,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,GAAG,IAAI,EAAQ,CAAC,MAC1B,GAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,IAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,GAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,IAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OACrB,EAAa,KAAK,MAAM,KAAK,IAAI,IAAM,IAAO,GAAK,GAAK,MAAS,IACvE,AAAI,EAAa,EAAO,KAAK,OAAO,eAClC,GAAI,OAAS,EAAK,IAAM,GAAM,SAAW,OACzC,EAAI,WAAa,GAEnB,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KC9DrB,oBAAM,IAAK,KAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,GAAG,KAAK,KACrB,KAAM,GAAS,GAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,IAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,GAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,GAAG,IAAI,EAAG,CAAC,QAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAY,GAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,oBAAM,IAAK,KAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,oBAAM,IAAK,KACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,eACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,oBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,kBAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,GAAQ,cAAgB,GAAQ,UAAU,OAC1C,GAAQ,QAAU,GAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,GAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,GAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,GAAQ,QAAQ,GAAa,GAAQ,QAAQ,KACnI,GAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,oBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,GAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,GAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,GAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,GAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,GAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,GAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,GAAQ,YAAc,KCnDtB,oBAAM,IAAY,KACZ,GAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,GAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,GAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,GAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,GAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,oBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,oBAAM,IAAK,KACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,GAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,GAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,GAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,GAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,GAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,GAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,IAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,GAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,IAAG,OAAS,EAAQ,GAAG,QAAQ,WAAW,GAEpE,GAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,IAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,GAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,GAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,GAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,GAAQ,kBAAoB,KCrH5B,oBAAM,IAAK,KACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,oBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,GAAQ,KAAO,GAAa,KAC5B,GAAQ,QAAU,GAAa,QAE/B,GAAQ,UAAY,GAAe,UACnC,GAAQ,oBAAsB,GAAe,oBAC7C,GAAQ,aAAe,GAAU,aACjC,GAAQ,QAAU,GAAU,QAC5B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,eAAiB,GAAK,eAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,kBAAoB,GAAK,kBACjC,GAAQ,UAAY,GAAK,YCnBzB,oBAAM,IAAK,KAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,GAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,SAAW,KCtEnB,oBAAM,IAAK,KACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,GAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,GAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,GAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,IAAG,KAAK,KACb,KAAM,GAAa,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,GAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,IAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,IAAG,KAAK,KACb,KAAM,GAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,IAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,GAAG,KAAK,IAAM,GAAG,IAAI,GAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,GAAG,KAAK,IAAM,GAAG,QAAQ,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,GAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,GAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,GAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,kBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,2BAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,KCzEtB,oBAAM,IAAK,KACL,GAAW,KACX,GAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,IAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,IAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,GAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,GAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,GAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,GAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,GAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,oBAAM,IAAK,KACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,GAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,IAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,uCAGA,GAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GAGT,SAAU,CACR,UAAW,sCAEX,UAAW,IACX,SAAU,GACV,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCAEX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,cAAe,GACf,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,GACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,GACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,6vEC/EjB,oBAAM,IAAK,KACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,IAGJ,KAAM,IAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,GAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,GAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,GAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,GAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,IAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,GAAG,eAAiB,GAAO,SAC7B,IAAI,iCAAkC,GAAO,SAC7C,KAAM,IAAG,WAAW,GAAO,SAC3B,KAAM,IAAG,SAUX,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,GAAO,OACvF,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,IAAI,SAAW,CAAC,GAAO,KAAK,IAAO,IAAM,KAAM,IAAO,QAAQ,KACrG,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,OAAO,SAAW,CAAC,GAAO,QAAQ,IAAO,OAAS,KAAM,IAAO,WAAW,KACjH,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,QAAQ,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,KAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,QAAQ,cAAc,EAAO,GAAO,MAAQ,GAC/F,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAAQ,GAChG,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,GAAO,KAAK,SACd,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAChE,EAAK,KAAO,KAAK,MAAM,KAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,KACZ,KAAM,GAAW,GAAO,KAAK,IAAI,SAAW,GAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,IAAU,GACrH,EAAK,UAAY,KAAK,MAAM,KAAQ,GAEpC,EAAY,KACZ,KAAM,GAAc,GAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,IAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,KAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,aAAc,EAAQ,WACtB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,GAAG,SAAS,WAOd,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,QAAU,GAClB,GAAQ,SAAW,GACnB,GAAQ,GAAK,GACb,GAAQ,QAAU,GAAI",
   "names": []
 }
diff --git a/dist/human.js b/dist/human.js
index b59f6a39..a4560f7d 100644
--- a/dist/human.js
+++ b/dist/human.js
@@ -1,39 +1,39 @@
-var human=(()=>{var Np=Object.defineProperty;var be=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),CI=n=>Np(n,"__esModule",{value:!0}),xp=(n,t)=>{CI(n);for(var e in t)Np(n,e,{get:t[e],enumerable:!0})};var Cp=be(()=>{});var Rp=be(()=>{});var ds=be(()=>{});var Xi=be(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var el=function(n,t){return el=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},el(n,t)};function qn(n,t){el(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var RI=1e-7,OI=1e-4,EI=function(){function n(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}return n.prototype.get=function(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)},n.prototype.set=function(t,e){this.dataIdsCount++,this.data.set(t,e)},n.prototype.has=function(t){return this.data.has(t)},n.prototype.delete=function(t){return this.dataIdsCount--,this.data.delete(t)},n.prototype.numDataIds=function(){return this.dataIdsCount},n}(),Op=function(){function n(){}return n.prototype.time=function(t){return $("time")},n.prototype.read=function(t){return $("read")},n.prototype.readSync=function(t){return $("readSync")},n.prototype.numDataIds=function(){return $("numDataIds")},n.prototype.disposeData=function(t){return $("disposeData")},n.prototype.write=function(t,e,i){return $("write")},n.prototype.move=function(t,e,i,r){return $("move")},n.prototype.memory=function(){return $("memory")},n.prototype.floatPrecision=function(){return $("floatPrecision")},n.prototype.epsilon=function(){return this.floatPrecision()===32?RI:OI},n.prototype.batchMatMul=function(t,e,i,r){return $("batchMatMul")},n.prototype.fusedBatchMatMul=function(t){var e=t.a,i=t.b,r=t.transposeA,a=t.transposeB,s=t.bias,o=t.activation,l=t.preluActivationWeights;return $("fusedBatchMatMul")},n.prototype.slice=function(t,e,i){return $("slice")},n.prototype.stridedSlice=function(t,e,i,r){return $("stridedSlice")},n.prototype.unstack=function(t,e){return $("unstack")},n.prototype.reverse=function(t,e){return $("reverse")},n.prototype.concat=function(t,e){return $("concat")},n.prototype.neg=function(t){return $("neg")},n.prototype.add=function(t,e){return $("add")},n.prototype.addN=function(t){return $("addN")},n.prototype.subtract=function(t,e){return $("subtract")},n.prototype.multiply=function(t,e){return $("multiply")},n.prototype.realDivide=function(t,e){return $("realDivide")},n.prototype.floorDiv=function(t,e){return $("floorDiv")},n.prototype.sum=function(t,e){return $("sum")},n.prototype.prod=function(t,e){return $("prod")},n.prototype.unsortedSegmentSum=function(t,e,i){return $("unsortedSegmentSum")},n.prototype.argMin=function(t,e){return $("argMin")},n.prototype.argMax=function(t,e){return $("argMax")},n.prototype.equal=function(t,e){return $("equal")},n.prototype.notEqual=function(t,e){return $("notEqual")},n.prototype.less=function(t,e){return $("less")},n.prototype.lessEqual=function(t,e){return $("lessEqual")},n.prototype.greater=function(t,e){return $("greater")},n.prototype.greaterEqual=function(t,e){return $("greaterEqual")},n.prototype.logicalNot=function(t){return $("logicalNot")},n.prototype.logicalAnd=function(t,e){return $("logicalAnd")},n.prototype.logicalOr=function(t,e){return $("logicalOr")},n.prototype.where=function(t){return $("where")},n.prototype.select=function(t,e,i){return $("select")},n.prototype.topk=function(t,e,i){return $("topk")},n.prototype.min=function(t,e){return $("min")},n.prototype.minimum=function(t,e){return $("minimum")},n.prototype.mod=function(t,e){return $("mod")},n.prototype.max=function(t,e){return $("max")},n.prototype.maximum=function(t,e){return $("maximum")},n.prototype.all=function(t,e){return $("all")},n.prototype.any=function(t,e){return $("any")},n.prototype.squaredDifference=function(t,e){return $("squaredDifference")},n.prototype.ceil=function(t){return $("ceil")},n.prototype.floor=function(t){return $("floor")},n.prototype.round=function(t){return $("round")},n.prototype.sign=function(t){return $("sign")},n.prototype.isNaN=function(t){return $("isNaN")},n.prototype.isInf=function(t){return $("isInf")},n.prototype.isFinite=function(t){return $("isFinite")},n.prototype.pow=function(t,e){return $("pow")},n.prototype.exp=function(t){return $("exp")},n.prototype.expm1=function(t){return $("expm1")},n.prototype.softmax=function(t,e){return $("softmax")},n.prototype.log=function(t){return $("log")},n.prototype.log1p=function(t){return $("log1p")},n.prototype.sqrt=function(t){return $("sqrt")},n.prototype.rsqrt=function(t){return $("rsqrt")},n.prototype.square=function(t){return $("square")},n.prototype.reciprocal=function(t){return $("reciprocal")},n.prototype.relu=function(t){return $("relu")},n.prototype.relu6=function(t){return $("relu6")},n.prototype.prelu=function(t,e){return $("prelu")},n.prototype.elu=function(t){return $("elu")},n.prototype.eluDer=function(t,e){return $("eluDer")},n.prototype.selu=function(t){return $("selu")},n.prototype.int=function(t){return $("int")},n.prototype.clip=function(t,e,i){return $("clip")},n.prototype.abs=function(t){return $("abs")},n.prototype.complexAbs=function(t){return $("complexAbs")},n.prototype.sigmoid=function(t){return $("sigmoid")},n.prototype.softplus=function(t){return $("softplus")},n.prototype.sin=function(t){return $("sin")},n.prototype.cos=function(t){return $("cos")},n.prototype.tan=function(t){return $("tan")},n.prototype.asin=function(t){return $("asin")},n.prototype.acos=function(t){return $("acos")},n.prototype.atan=function(t){return $("atan")},n.prototype.atan2=function(t,e){return $("atan2")},n.prototype.sinh=function(t){return $("sinh")},n.prototype.cosh=function(t){return $("cosh")},n.prototype.tanh=function(t){return $("tanh")},n.prototype.asinh=function(t){return $("asinh")},n.prototype.acosh=function(t){return $("acosh")},n.prototype.atanh=function(t){return $("atanh")},n.prototype.erf=function(t){return $("erf")},n.prototype.step=function(t,e){return $("step")},n.prototype.fusedConv2d=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedConv2d")},n.prototype.conv2d=function(t,e,i){return $("conv2d")},n.prototype.conv2dDerInput=function(t,e,i){return $("conv2dDerInput")},n.prototype.conv2dDerFilter=function(t,e,i){return $("conv2dDerFilter")},n.prototype.fusedDepthwiseConv2D=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedDepthwiseConv2D")},n.prototype.depthwiseConv2D=function(t,e,i){return $("depthwiseConv2D")},n.prototype.depthwiseConv2DDerInput=function(t,e,i){return $("depthwiseConv2DDerInput")},n.prototype.depthwiseConv2DDerFilter=function(t,e,i){return $("depthwiseConv2DDerFilter")},n.prototype.conv3d=function(t,e,i){return $("conv3d")},n.prototype.conv3dDerInput=function(t,e,i){return $("conv3dDerInput")},n.prototype.conv3dDerFilter=function(t,e,i){return $("conv3dDerFilter")},n.prototype.maxPool=function(t,e){return $("maxPool")},n.prototype.maxPoolBackprop=function(t,e,i,r){return $("maxPoolBackprop")},n.prototype.avgPool=function(t,e){return $("avgPool")},n.prototype.avgPoolBackprop=function(t,e,i){return $("avgPoolBackprop")},n.prototype.avgPool3d=function(t,e){return $("avgPool3d")},n.prototype.avgPool3dBackprop=function(t,e,i){return $("avgPool3dBackprop")},n.prototype.maxPool3d=function(t,e){return $("maxPool3d")},n.prototype.maxPool3dBackprop=function(t,e,i,r){return $("maxPool3dBackprop")},n.prototype.reshape=function(t,e){return $("reshape")},n.prototype.cast=function(t,e){return $("cast")},n.prototype.tile=function(t,e){return $("tile")},n.prototype.pad=function(t,e,i){return $("pad")},n.prototype.transpose=function(t,e){return $("transpose")},n.prototype.gather=function(t,e,i){return $("gather")},n.prototype.gatherND=function(t,e){return $("gatherND")},n.prototype.scatterND=function(t,e,i){return $("scatterND")},n.prototype.batchToSpaceND=function(t,e,i){return $("batchToSpaceND")},n.prototype.spaceToBatchND=function(t,e,i){return $("spaceToBatchND")},n.prototype.resizeBilinear=function(t,e,i,r){return $("resizeBilinear")},n.prototype.resizeBilinearBackprop=function(t,e,i){return $("resizeBilinearBackprop")},n.prototype.resizeNearestNeighbor=function(t,e,i,r){return $("resizeNearestNeighbor")},n.prototype.resizeNearestNeighborBackprop=function(t,e,i){return $("resizeNearestNeighborBackprop")},n.prototype.batchNorm=function(t,e,i,r,a,s){return $("batchNorm")},n.prototype.localResponseNormalization4D=function(t,e,i,r,a){return $("localResponseNormalization4D")},n.prototype.LRNGrad=function(t,e,i,r,a,s,o){return $("LRNGrad")},n.prototype.multinomial=function(t,e,i,r){return $("multinomial")},n.prototype.oneHot=function(t,e,i,r){return $("oneHot")},n.prototype.cumsum=function(t,e,i,r){return $("cumsum")},n.prototype.nonMaxSuppression=function(t,e,i,r,a){return $("nonMaxSuppression")},n.prototype.fft=function(t){return $("fft")},n.prototype.ifft=function(t){return $("ifft")},n.prototype.complex=function(t,e){return $("complex")},n.prototype.real=function(t){return $("real")},n.prototype.imag=function(t){return $("imag")},n.prototype.cropAndResize=function(t,e,i,r,a,s){return $("cropAndResize")},n.prototype.depthToSpace=function(t,e,i){return $("depthToSpace")},n.prototype.split=function(t,e,i){return $("split")},n.prototype.sparseToDense=function(t,e,i,r){return $("sparseToDense")},n.prototype.diag=function(t){return $("diag")},n.prototype.fill=function(t,e,i){return $("fill")},n.prototype.onesLike=function(t){return $("onesLike")},n.prototype.zerosLike=function(t){return $("zerosLike")},n.prototype.linspace=function(t,e,i){return $("linspace")},n.prototype.dispose=function(){return $("dispose")},n}();function $(n){throw new Error("'"+n+"' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen")}var Ep="tfjsflags",Dp=function(){function n(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}return n.prototype.setPlatform=function(t,e){this.platform!=null&&console.warn("Platform "+this.platformName+" has already been set. "+("Overwriting the platform with "+e+".")),this.platformName=t,this.platform=e},n.prototype.registerFlag=function(t,e,i){if(this.flagRegistry[t]={evaluationFn:e,setHook:i},this.urlFlags[t]!=null){var r=this.urlFlags[t];console.warn("Setting feature override from URL "+t+": "+r+"."),this.set(t,r)}},n.prototype.getAsync=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return t in this.flags?[2,this.flags[t]]:(e=this.flags,i=t,[4,this.evaluateFlag(t)]);case 1:return e[i]=r.sent(),[2,this.flags[t]]}})})},n.prototype.get=function(t){if(t in this.flags)return this.flags[t];var e=this.evaluateFlag(t);if(e instanceof Promise)throw new Error("Flag "+t+" cannot be synchronously evaluated. Please use getAsync() instead.");return this.flags[t]=e,this.flags[t]},n.prototype.getNumber=function(t){return this.get(t)},n.prototype.getBool=function(t){return this.get(t)},n.prototype.getFlags=function(){return this.flags},Object.defineProperty(n.prototype,"features",{get:function(){return this.flags},enumerable:!0,configurable:!0}),n.prototype.set=function(t,e){if(this.flagRegistry[t]==null)throw new Error("Cannot set flag "+t+" as it has not been registered.");this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)},n.prototype.evaluateFlag=function(t){if(this.flagRegistry[t]==null)throw new Error("Cannot evaluate flag '"+t+"': no evaluation function found.");return this.flagRegistry[t].evaluationFn()},n.prototype.setFlags=function(t){this.flags=Object.assign({},t)},n.prototype.reset=function(){this.flags={},this.urlFlags={},this.populateURLFlags()},n.prototype.populateURLFlags=function(){var t=this;if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;var e=DI(this.global.location.search);if(Ep in e){var i=e[Ep].split(",");i.forEach(function(r){var a=r.split(":"),s=a[0],o=a[1];t.urlFlags[s]=kI(s,o)})}},n}();function DI(n){var t={};return n.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,function(e){for(var i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];return FI(t,i[0],i[1]),i.join("=")}),t}function FI(n,t,e){n[decodeURIComponent(t)]=decodeURIComponent(e||"")}function kI(n,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(""+ +t===t)return+t;throw new Error("Could not parse value flag value "+t+" for flag "+n+".")}function He(){return I.ENV}I.ENV=null;function WI(n){I.ENV=n}var tl;function kp(){if(tl==null){var n=void 0;if(typeof window!="undefined")n=window;else if(typeof global!="undefined")n=global;else if(typeof process!="undefined")n=process;else if(typeof self!="undefined")n=self;else throw new Error("Could not find a global object");tl=n}return tl}function UI(){var n=kp();return n._tfGlobals==null&&(n._tfGlobals=new Map),n._tfGlobals}function Fp(n,t){var e=UI();if(e.has(n))return e.get(n);var i=t();return e.set(n,i),e.get(n)}var nl="Abs",il="Acos",rl="Acosh",ps="Add",al="AddN",Wp="All",Up="Any",sl="ArgMax",ol="ArgMin",ll="Asin",ul="Asinh",cl="Atan",hl="Atanh",dl="Atan2",pl="AvgPool",Bp="AvgPoolBackprop",fl="AvgPool3D",zp="AvgPool3DBackprop",ml="BatchMatMul",gl="BatchToSpaceND",vl="BroadcastTo",fs="Cast",yl="Ceil",bl="ClipByValue",Pp="Complex",wl="Concat",Sl="Conv2D",_p="Conv2DBackpropFilter",Ll="Conv2DBackpropInput",Il="Conv3D",Mp="Conv3DBackpropFilterV2",Hp="Conv3DBackpropInputV2",Al="Cos",Tl="Cosh",Nl="Cumsum",Vp="CropAndResize",qp="DepthToSpace",xl="DepthwiseConv2dNative",Gp="DepthwiseConv2dNativeBackpropFilter",Yp="DepthwiseConv2dNativeBackpropInput",Kp="Diag",Cl="Dilation2D",jp="Dilation2DBackpropInput",$p="Dilation2DBackpropFilter",Rl="Div",Ol="Elu",Xp="EluGrad",El="Erf",Jp="Equal",Dl="Exp",kl="Expm1",Zp="FFT",Qp="Fill",ef="FlipLeftRight",Fl="Floor",Wl="FloorDiv",Ul="FusedBatchNorm",Bl="GatherV2",tf="GatherNd",nf="Greater",zl="GreaterEqual",Pl="Identity",rf="IFFT",af="Imag",_l="IsFinite",Ml="IsInf",Hl="IsNan",sf="Less",of="LessEqual",lf="LinSpace",Vl="Log",ql="Log1p",uf="LogicalAnd",cf="LogicalNot",hf="LogicalOr",Gl="LogSoftmax",Yl="LRN",df="LRNBackprop",Kl="Max",jl="Maximum",$l="MaxPool",pf="MaxPoolBackprop",Xl="MaxPool3D",ff="MaxPool3DBackprop",mf="MaxPoolWithArgmax",BI="Mean",Jl="Min",Zl="Minimum",Ql="Mod",eu="Multiply",tu="Negate",gf="NotEqual",vf="NonMaxSuppressionV3",yf="NonMaxSuppressionV4",bf="NonMaxSuppressionV5",nu="OnesLike",iu="OneHot",ru="PadV2",zI="Pool",au="Pow",su="Prelu",wf="Prod",Sf="Range",Lf="Real",ou="Reciprocal",lu="Relu",uu="Reshape",cu="ResizeNearestNeighbor",If="ResizeNearestNeighborGrad",hu="ResizeBilinear",Af="ResizeBilinearGrad",du="Relu6",pu="Reverse",fu="Round",mu="Rsqrt",Tf="ScatterNd",gu="SelectV2",vu="Selu",yu="Slice",bu="Sin",wu="Sinh",Su="Sign",Lu="Sigmoid",Iu="Softplus",Au="Sqrt",Tu="Sum",Nu="SpaceToBatchND",xu="SplitV",Cu="Softmax",Ru="SquaredDifference",Nf="Square",Ou="Sub",xf="SparseToDense",Cf="StridedSlice",Eu="Tan",Du="Tanh",ku="Tile",Rf="TopK",Fu="Transpose",Of="Unique",Wu="Unpack",Uu="UnsortedSegmentSum",Bu="ZerosLike",zu="Step",Pu="FromPixels",Ef="RotateWithOffset",_u="_FusedMatMul",Mu="FusedConv2D",Hu="FusedDepthwiseConv2D";var br=Fp("kernelRegistry",function(){return new Map}),da=Fp("gradRegistry",function(){return new Map});function qu(n,t){var e=Vu(n,t);return br.get(e)}function Gu(n){return da.get(n)}function ms(n){for(var t=br.entries(),e=[];;){var i=t.next(),r=i.done,a=i.value;if(r)break;var s=a[0],o=a[1],l=s.split("_")[0];l===n&&e.push(o)}return e}function Df(n){var t=n.kernelName,e=n.backendName,i=Vu(t,e);br.has(i)&&console.warn("The kernel '"+t+"' for backend "+("'"+e+"' is already registered")),br.set(i,n)}function kf(n){var t=n.kernelName;da.has(t)&&(He().getBool("DEBUG")&&console.warn("Overriding the gradient for '"+t+"'")),da.set(t,n)}function PI(n,t){var e=Vu(n,t);if(!br.has(e))throw new Error("The kernel '"+n+"' for backend "+("'"+t+"' is not registered"));br.delete(e)}function _I(n){if(!da.has(n))throw new Error("The gradient '"+n+"' for backend is not registered");da.delete(n)}function MI(n,t){var e=ms(n);e.forEach(function(i){var r=Object.assign({},i,{backendName:t});Df(r)})}function Vu(n,t){return t+"_"+n}function Ff(n){for(var t=n.length,e=0,i=0;t>0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function pa(n,t,e){return Math.max(n,Math.min(t,e))}function HI(n){return n%2===0?n:n+1}function VI(n){for(var t=0,e=0;e<n.length;e++)t+=n[e];return t}function qI(n,t){var e=Math.random();return t*e+(1-e)*n}function GI(n,t){for(var e=0,i=0;i<n.length;i++){var r=Number(n[i])-Number(t[i]);e+=r*r}return e}function E(n,t){if(!n)throw new Error(typeof t=="string"?t:t())}function Be(n,t,e){e===void 0&&(e=""),E(ln(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function Fi(n){E(n!=null,function(){return"The input to the tensor constructor must be a non-null value."})}function Wi(n,t,e){if(t===void 0&&(t=[]),e===void 0&&(e=!1),t==null&&(t=[]),Array.isArray(n)||Ot(n)&&!e)for(var i=0;i<n.length;++i)Wi(n[i],t,e);else t.push(n);return t}function ot(n){if(n.length===0)return 1;for(var t=n[0],e=1;e<n.length;e++)t*=n[e];return t}function YI(n){return n.length===0}function ln(n,t){if(n===t)return!0;if(n==null||t==null)return!1;if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==t[e])return!1;return!0}function rt(n){return n%1===0}function KI(n){if(Math.tanh!=null)return Math.tanh(n);if(n===Infinity)return 1;if(n===-Infinity)return-1;var t=Math.exp(2*n);return(t-1)/(t+1)}function jI(n){var t=Math.ceil(Math.sqrt(n));return[t,Math.ceil(n/t)]}function $I(n){for(var t=new Uint32Array(n),e=0;e<n;++e)t[e]=e;return Ff(t),t}function fa(n,t){return t<=n.length?n:n+" ".repeat(t-n.length)}function XI(n,t,e){return t===void 0&&(t=function(i){return 0}),new Promise(function(i,r){var a=0,s=function(){if(n()){i();return}a++;var o=t(a);if(e!=null&&a>=e){r();return}setTimeout(s,o)};s()})}function Wf(n,t){for(var e=1,i=-1,r=0;r<n.length;++r)if(n[r]>=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Qe(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&i<e}),function(){return"All values in axis param must be in range [-"+e+", "+e+") but "+("got axis "+n)}),E(n.every(function(i){return rt(i)}),function(){return"All values in axis param must be integers but "+("got axis "+n)}),n.map(function(i){return i<0?e+i:i})}function Uf(n,t){for(var e=[],i=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Qe(t,n).sort(),s=0,o=0;o<n.length;++o){if(a!=null){if(a[s]===o&&n[o]!==1)throw new Error("Can't squeeze axis "+o+" since its dim '"+n[o]+"' is not 1");(a[s]==null||a[s]>o)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function gs(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function Bf(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function zf(n,t){for(var e=0;e<n.length;e++){var i=n[e];if(isNaN(i)||!isFinite(i))throw Error("A tensor of type "+t+" being uploaded contains "+i+".")}}function Pf(n){return n==="bool"||n==="complex64"||n==="float32"||n==="int32"||n==="string"}function _f(n,t){return t==="complex64"||(t==="float32"&&n!=="complex64"||t==="int32"&&n!=="float32"&&n!=="complex64")?!1:!(t==="bool"&&n==="bool")}function Ot(n){return n instanceof Float32Array||n instanceof Int32Array||n instanceof Uint8Array}function Mf(n){if(n==="float32"||n==="int32")return 4;if(n==="complex64")return 8;if(n==="bool")return 1;throw new Error("Unknown dtype "+n)}function Hf(n){if(n==null)return 0;var t=0;return n.forEach(function(e){return t+=e.length}),t}function ni(n){return typeof n=="string"||n instanceof String}function Vf(n){return typeof n=="boolean"}function qf(n){return typeof n=="number"}function vs(n){return Array.isArray(n)?vs(n[0]):n instanceof Float32Array?"float32":n instanceof Int32Array||n instanceof Uint8Array?"int32":qf(n)?"float32":ni(n)?"string":Vf(n)?"bool":"float32"}function ii(n){return!!(n&&n.constructor&&n.call&&n.apply)}function ys(n,t){for(var e=t;e<n;++e)if(n%e===0)return e;return n}function wr(n){var t=n.length;if(t<2)return[];var e=new Array(t-1);e[t-2]=n[t-1];for(var i=t-3;i>=0;--i)e[i]=e[i+1]*n[i+1];return e}function JI(n,t){return t==="string"?Yu(n):bs([n],t)}function bs(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=Wi(n)),He().getBool("DEBUG")&&zf(n,t),ZI(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i<e.length;++i)Math.round(n[i])!==0&&(e[i]=1);return e}else throw new Error("Unknown data type "+t)}function Gf(n,t,e){var i=new Array;if(t.length===1)for(var r=t[0],a=0;a<r;a++)i[a]=e[n+a];else for(var r=t[0],s=t.slice(1),o=s.reduce(function(u,c){return u*c}),a=0;a<r;a++)i[a]=Gf(n+a*o,s,e);return i}function Sr(n,t){if(n.length===0)return t[0];var e=n.reduce(function(i,r){return i*r});if(e===0)return[];if(e!==t.length)throw new Error("["+n+"] does not match the input size "+t.length+".");return Gf(0,n,t)}function ZI(n,t){return n instanceof Float32Array&&t==="float32"||n instanceof Int32Array&&t==="int32"||n instanceof Uint8Array&&t==="bool"}function Ku(n,t){for(var e=Lr(n,t),i=0;i<e.length;i++)e[i]=1;return e}function Lr(n,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool")return new Uint8Array(n);throw new Error("Unknown data type "+t)}function QI(n,t){var e=n.reduce(function(i,r){return i*r},1);if(t==null||t==="float32")return Sr(n,new Float32Array(e));if(t==="int32")return Sr(n,new Int32Array(e));if(t==="bool")return Sr(n,new Uint8Array(e));throw new Error("Unknown data type "+t)}function ju(){return He().platform.now()}function $u(n){n.forEach(function(t){E(Number.isInteger(t)&&t>=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function e2(n,t){return He().platform.fetch(n,t)}function Yu(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.encode(n,t)}function Xu(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.decode(n,t)}function t2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r<n.length-1;++r)i+=e[r]*n[r];return i}function n2(n,t,e){if(t===0)return[];if(t===1)return[n];for(var i=new Array(t),r=0;r<i.length-1;++r)i[r]=Math.floor(n/e[r]),n-=i[r]*e[r];return i[i.length-1]=n,i}var i2={__proto__:null,shuffle:Ff,clamp:pa,nearestLargerEven:HI,sum:VI,randUniform:qI,distSquared:GI,assert:E,assertShapesMatch:Be,assertNonNull:Fi,flatten:Wi,sizeFromShape:ot,isScalarShape:YI,arraysEqual:ln,isInt:rt,tanh:KI,sizeToSquarishShape:jI,createShuffledIndices:$I,rightPad:fa,repeatedTry:XI,inferFromImplicitShape:Wf,parseAxisParam:Qe,squeezeShape:Uf,getTypedArrayFromDType:gs,getArrayFromDType:Bf,checkConversionForErrors:zf,isValidDtype:Pf,hasEncodingLoss:_f,isTypedArray:Ot,bytesPerElement:Mf,bytesFromStringArray:Hf,isString:ni,isBoolean:Vf,isNumber:qf,inferDtype:vs,isFunction:ii,nearestDivisor:ys,computeStrides:wr,createScalarValue:JI,toTypedArray:bs,toNestedArray:Sr,makeOnesTypedArray:Ku,makeZerosTypedArray:Lr,makeZerosNestedTypedArray:QI,now:ju,assertNonNegativeIntegerDimensions:$u,fetch:e2,encodeString:Yu,decodeString:Xu,locToIndex:t2,indexToLoc:n2};var s2=function(){function n(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new a2)}return n.prototype.profileKernel=function(t,e,i){for(var r,a=function(){r=i()},s=this.backendTimer.time(a),o=function(c){var h=r[c];h.data().then(function(d){r2(d,h.dtype,t)})},l=0;l<r.length;l++)o(l);var u={kernelName:t,outputs:r,inputs:e,timeMs:s.then(function(c){return c.kernelMs}),extraInfo:s.then(function(c){return c.getExtraProfileInfo!=null?c.getExtraProfileInfo():""})};return u},n.prototype.logKernelProfile=function(t){var e=this,i=t.kernelName,r=t.outputs,a=t.timeMs,s=t.inputs,o=t.extraInfo;r.forEach(function(l){Promise.all([l.data(),a,o]).then(function(u){e.logger.logKernelProfile(i,l,u[0],u[1],s,u[2])})})},n}();function r2(n,t,e){if(t!=="float32")return!1;for(var i=0;i<n.length;i++){var r=n[i];if(isNaN(r)||!isFinite(r))return console.warn("Found "+r+" in the result of '"+e+"'"),!0}return!1}var a2=function(){function n(){}return n.prototype.logKernelProfile=function(t,e,i,r,a,s){var o=typeof r=="number"?fa(r+"ms",9):r.error,l=fa(t,25),u=e.rank,c=e.size,h=fa(e.shape.toString(),14),d="";for(var p in a){var f=a[p];if(f!=null){var m=f.shape||e.shape,g=m.length;d+=p+": "+g+"D "+(g>0?m:"")+" "}}console.log("%c"+l+"	%c"+o+"	%c"+u+"D "+h+"	%c"+c+"	%c"+d+"	%c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function o2(n,t,e){for(var i={},r={},a=0;a<t.length;a++)i[t[a].id]=!0;for(var a=0;a<n.length;a++){var s=n[a],o=s.inputs;for(var l in o){for(var u=o[l],c=!1,h=0;h<t.length;h++)if(i[u.id]){s.outputs.forEach(function(w){return i[w.id]=!0}),c=!0,r[s.id]=!0;break}if(c)break}}var d={};d[e.id]=!0;for(var p={},a=n.length-1;a>=0;a--)for(var s=n[a],o=s.inputs,h=0;h<s.outputs.length;h++)if(d[s.outputs[h].id]){for(var l in o)d[o[l].id]=!0,p[s.id]=!0;break}for(var f=[],a=0;a<n.length;a++){var s=n[a];if(r[s.id]&&p[s.id]){var m={};for(var l in s.inputs){var g=s.inputs[l];i[g.id]&&(m[l]=g)}var v=Object.assign({},s);v.inputs=m,v.outputs=s.outputs,f.push(v)}}return f}function l2(n,t,e,i){for(var r=function(s){var o=t[s],l=[];if(o.outputs.forEach(function(d){var p=n[d.id];p!=null?l.push(p):l.push(null)}),o.gradient==null)throw new Error("Cannot compute gradient: gradient function not found "+("for "+o.kernelName+"."));var u=o.gradient(l),c=function(d){if(!(d in u))throw new Error("Cannot backprop through input "+d+". "+("Available gradients found: "+Object.keys(u)+"."));var p=e(function(){return u[d]()});if(p.dtype!=="float32")throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+(d+" must have 'float32' dtype, but has '"+p.dtype+"'"));var f=o.inputs[d];if(!ln(p.shape,f.shape))throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+("'"+d+"' has shape '"+p.shape+"', which does not match ")+("the shape of the input '"+f.shape+"'"));if(n[f.id]==null)n[f.id]=p;else{var m=n[f.id];n[f.id]=i(m,p),m.dispose()}};for(var h in o.inputs)c(h)},a=t.length-1;a>=0;a--)r(a)}var Yf=20,ma=3,Ju=7;function c2(n,t,e,i){var r=wr(t),a=u2(n,t,e,r),s=t.length,o=ws(n,t,e,r,a),l=["Tensor"];return i&&(l.push("  dtype: "+e),l.push("  rank: "+s),l.push("  shape: ["+t+"]"),l.push("  values:")),l.push(o.map(function(u){return"    "+u}).join(`
+var human=(()=>{var Np=Object.defineProperty;var be=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),CI=n=>Np(n,"__esModule",{value:!0}),xp=(n,t)=>{CI(n);for(var e in t)Np(n,e,{get:t[e],enumerable:!0})};var Cp=be(()=>{});var Rp=be(()=>{});var ds=be(()=>{});var Xi=be(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var tl=function(n,t){return tl=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},tl(n,t)};function qn(n,t){tl(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var RI=1e-7,OI=1e-4,EI=function(){function n(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}return n.prototype.get=function(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)},n.prototype.set=function(t,e){this.dataIdsCount++,this.data.set(t,e)},n.prototype.has=function(t){return this.data.has(t)},n.prototype.delete=function(t){return this.dataIdsCount--,this.data.delete(t)},n.prototype.numDataIds=function(){return this.dataIdsCount},n}(),Op=function(){function n(){}return n.prototype.time=function(t){return $("time")},n.prototype.read=function(t){return $("read")},n.prototype.readSync=function(t){return $("readSync")},n.prototype.numDataIds=function(){return $("numDataIds")},n.prototype.disposeData=function(t){return $("disposeData")},n.prototype.write=function(t,e,i){return $("write")},n.prototype.move=function(t,e,i,r){return $("move")},n.prototype.memory=function(){return $("memory")},n.prototype.floatPrecision=function(){return $("floatPrecision")},n.prototype.epsilon=function(){return this.floatPrecision()===32?RI:OI},n.prototype.batchMatMul=function(t,e,i,r){return $("batchMatMul")},n.prototype.fusedBatchMatMul=function(t){var e=t.a,i=t.b,r=t.transposeA,a=t.transposeB,s=t.bias,o=t.activation,l=t.preluActivationWeights;return $("fusedBatchMatMul")},n.prototype.slice=function(t,e,i){return $("slice")},n.prototype.stridedSlice=function(t,e,i,r){return $("stridedSlice")},n.prototype.unstack=function(t,e){return $("unstack")},n.prototype.reverse=function(t,e){return $("reverse")},n.prototype.concat=function(t,e){return $("concat")},n.prototype.neg=function(t){return $("neg")},n.prototype.add=function(t,e){return $("add")},n.prototype.addN=function(t){return $("addN")},n.prototype.subtract=function(t,e){return $("subtract")},n.prototype.multiply=function(t,e){return $("multiply")},n.prototype.realDivide=function(t,e){return $("realDivide")},n.prototype.floorDiv=function(t,e){return $("floorDiv")},n.prototype.sum=function(t,e){return $("sum")},n.prototype.prod=function(t,e){return $("prod")},n.prototype.unsortedSegmentSum=function(t,e,i){return $("unsortedSegmentSum")},n.prototype.argMin=function(t,e){return $("argMin")},n.prototype.argMax=function(t,e){return $("argMax")},n.prototype.equal=function(t,e){return $("equal")},n.prototype.notEqual=function(t,e){return $("notEqual")},n.prototype.less=function(t,e){return $("less")},n.prototype.lessEqual=function(t,e){return $("lessEqual")},n.prototype.greater=function(t,e){return $("greater")},n.prototype.greaterEqual=function(t,e){return $("greaterEqual")},n.prototype.logicalNot=function(t){return $("logicalNot")},n.prototype.logicalAnd=function(t,e){return $("logicalAnd")},n.prototype.logicalOr=function(t,e){return $("logicalOr")},n.prototype.where=function(t){return $("where")},n.prototype.select=function(t,e,i){return $("select")},n.prototype.topk=function(t,e,i){return $("topk")},n.prototype.min=function(t,e){return $("min")},n.prototype.minimum=function(t,e){return $("minimum")},n.prototype.mod=function(t,e){return $("mod")},n.prototype.max=function(t,e){return $("max")},n.prototype.maximum=function(t,e){return $("maximum")},n.prototype.all=function(t,e){return $("all")},n.prototype.any=function(t,e){return $("any")},n.prototype.squaredDifference=function(t,e){return $("squaredDifference")},n.prototype.ceil=function(t){return $("ceil")},n.prototype.floor=function(t){return $("floor")},n.prototype.round=function(t){return $("round")},n.prototype.sign=function(t){return $("sign")},n.prototype.isNaN=function(t){return $("isNaN")},n.prototype.isInf=function(t){return $("isInf")},n.prototype.isFinite=function(t){return $("isFinite")},n.prototype.pow=function(t,e){return $("pow")},n.prototype.exp=function(t){return $("exp")},n.prototype.expm1=function(t){return $("expm1")},n.prototype.softmax=function(t,e){return $("softmax")},n.prototype.log=function(t){return $("log")},n.prototype.log1p=function(t){return $("log1p")},n.prototype.sqrt=function(t){return $("sqrt")},n.prototype.rsqrt=function(t){return $("rsqrt")},n.prototype.square=function(t){return $("square")},n.prototype.reciprocal=function(t){return $("reciprocal")},n.prototype.relu=function(t){return $("relu")},n.prototype.relu6=function(t){return $("relu6")},n.prototype.prelu=function(t,e){return $("prelu")},n.prototype.elu=function(t){return $("elu")},n.prototype.eluDer=function(t,e){return $("eluDer")},n.prototype.selu=function(t){return $("selu")},n.prototype.int=function(t){return $("int")},n.prototype.clip=function(t,e,i){return $("clip")},n.prototype.abs=function(t){return $("abs")},n.prototype.complexAbs=function(t){return $("complexAbs")},n.prototype.sigmoid=function(t){return $("sigmoid")},n.prototype.softplus=function(t){return $("softplus")},n.prototype.sin=function(t){return $("sin")},n.prototype.cos=function(t){return $("cos")},n.prototype.tan=function(t){return $("tan")},n.prototype.asin=function(t){return $("asin")},n.prototype.acos=function(t){return $("acos")},n.prototype.atan=function(t){return $("atan")},n.prototype.atan2=function(t,e){return $("atan2")},n.prototype.sinh=function(t){return $("sinh")},n.prototype.cosh=function(t){return $("cosh")},n.prototype.tanh=function(t){return $("tanh")},n.prototype.asinh=function(t){return $("asinh")},n.prototype.acosh=function(t){return $("acosh")},n.prototype.atanh=function(t){return $("atanh")},n.prototype.erf=function(t){return $("erf")},n.prototype.step=function(t,e){return $("step")},n.prototype.fusedConv2d=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedConv2d")},n.prototype.conv2d=function(t,e,i){return $("conv2d")},n.prototype.conv2dDerInput=function(t,e,i){return $("conv2dDerInput")},n.prototype.conv2dDerFilter=function(t,e,i){return $("conv2dDerFilter")},n.prototype.fusedDepthwiseConv2D=function(t){var e=t.input,i=t.filter,r=t.convInfo,a=t.bias,s=t.activation,o=t.preluActivationWeights;return $("fusedDepthwiseConv2D")},n.prototype.depthwiseConv2D=function(t,e,i){return $("depthwiseConv2D")},n.prototype.depthwiseConv2DDerInput=function(t,e,i){return $("depthwiseConv2DDerInput")},n.prototype.depthwiseConv2DDerFilter=function(t,e,i){return $("depthwiseConv2DDerFilter")},n.prototype.conv3d=function(t,e,i){return $("conv3d")},n.prototype.conv3dDerInput=function(t,e,i){return $("conv3dDerInput")},n.prototype.conv3dDerFilter=function(t,e,i){return $("conv3dDerFilter")},n.prototype.maxPool=function(t,e){return $("maxPool")},n.prototype.maxPoolBackprop=function(t,e,i,r){return $("maxPoolBackprop")},n.prototype.avgPool=function(t,e){return $("avgPool")},n.prototype.avgPoolBackprop=function(t,e,i){return $("avgPoolBackprop")},n.prototype.avgPool3d=function(t,e){return $("avgPool3d")},n.prototype.avgPool3dBackprop=function(t,e,i){return $("avgPool3dBackprop")},n.prototype.maxPool3d=function(t,e){return $("maxPool3d")},n.prototype.maxPool3dBackprop=function(t,e,i,r){return $("maxPool3dBackprop")},n.prototype.reshape=function(t,e){return $("reshape")},n.prototype.cast=function(t,e){return $("cast")},n.prototype.tile=function(t,e){return $("tile")},n.prototype.pad=function(t,e,i){return $("pad")},n.prototype.transpose=function(t,e){return $("transpose")},n.prototype.gather=function(t,e,i){return $("gather")},n.prototype.gatherND=function(t,e){return $("gatherND")},n.prototype.scatterND=function(t,e,i){return $("scatterND")},n.prototype.batchToSpaceND=function(t,e,i){return $("batchToSpaceND")},n.prototype.spaceToBatchND=function(t,e,i){return $("spaceToBatchND")},n.prototype.resizeBilinear=function(t,e,i,r){return $("resizeBilinear")},n.prototype.resizeBilinearBackprop=function(t,e,i){return $("resizeBilinearBackprop")},n.prototype.resizeNearestNeighbor=function(t,e,i,r){return $("resizeNearestNeighbor")},n.prototype.resizeNearestNeighborBackprop=function(t,e,i){return $("resizeNearestNeighborBackprop")},n.prototype.batchNorm=function(t,e,i,r,a,s){return $("batchNorm")},n.prototype.localResponseNormalization4D=function(t,e,i,r,a){return $("localResponseNormalization4D")},n.prototype.LRNGrad=function(t,e,i,r,a,s,o){return $("LRNGrad")},n.prototype.multinomial=function(t,e,i,r){return $("multinomial")},n.prototype.oneHot=function(t,e,i,r){return $("oneHot")},n.prototype.cumsum=function(t,e,i,r){return $("cumsum")},n.prototype.nonMaxSuppression=function(t,e,i,r,a){return $("nonMaxSuppression")},n.prototype.fft=function(t){return $("fft")},n.prototype.ifft=function(t){return $("ifft")},n.prototype.complex=function(t,e){return $("complex")},n.prototype.real=function(t){return $("real")},n.prototype.imag=function(t){return $("imag")},n.prototype.cropAndResize=function(t,e,i,r,a,s){return $("cropAndResize")},n.prototype.depthToSpace=function(t,e,i){return $("depthToSpace")},n.prototype.split=function(t,e,i){return $("split")},n.prototype.sparseToDense=function(t,e,i,r){return $("sparseToDense")},n.prototype.diag=function(t){return $("diag")},n.prototype.fill=function(t,e,i){return $("fill")},n.prototype.onesLike=function(t){return $("onesLike")},n.prototype.zerosLike=function(t){return $("zerosLike")},n.prototype.linspace=function(t,e,i){return $("linspace")},n.prototype.dispose=function(){return $("dispose")},n}();function $(n){throw new Error("'"+n+"' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen")}var Ep="tfjsflags",Dp=function(){function n(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}return n.prototype.setPlatform=function(t,e){this.platform!=null&&console.warn("Platform "+this.platformName+" has already been set. "+("Overwriting the platform with "+e+".")),this.platformName=t,this.platform=e},n.prototype.registerFlag=function(t,e,i){if(this.flagRegistry[t]={evaluationFn:e,setHook:i},this.urlFlags[t]!=null){var r=this.urlFlags[t];console.warn("Setting feature override from URL "+t+": "+r+"."),this.set(t,r)}},n.prototype.getAsync=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return t in this.flags?[2,this.flags[t]]:(e=this.flags,i=t,[4,this.evaluateFlag(t)]);case 1:return e[i]=r.sent(),[2,this.flags[t]]}})})},n.prototype.get=function(t){if(t in this.flags)return this.flags[t];var e=this.evaluateFlag(t);if(e instanceof Promise)throw new Error("Flag "+t+" cannot be synchronously evaluated. Please use getAsync() instead.");return this.flags[t]=e,this.flags[t]},n.prototype.getNumber=function(t){return this.get(t)},n.prototype.getBool=function(t){return this.get(t)},n.prototype.getFlags=function(){return this.flags},Object.defineProperty(n.prototype,"features",{get:function(){return this.flags},enumerable:!0,configurable:!0}),n.prototype.set=function(t,e){if(this.flagRegistry[t]==null)throw new Error("Cannot set flag "+t+" as it has not been registered.");this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)},n.prototype.evaluateFlag=function(t){if(this.flagRegistry[t]==null)throw new Error("Cannot evaluate flag '"+t+"': no evaluation function found.");return this.flagRegistry[t].evaluationFn()},n.prototype.setFlags=function(t){this.flags=Object.assign({},t)},n.prototype.reset=function(){this.flags={},this.urlFlags={},this.populateURLFlags()},n.prototype.populateURLFlags=function(){var t=this;if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;var e=DI(this.global.location.search);if(Ep in e){var i=e[Ep].split(",");i.forEach(function(r){var a=r.split(":"),s=a[0],o=a[1];t.urlFlags[s]=kI(s,o)})}},n}();function DI(n){var t={};return n.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,function(e){for(var i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];return FI(t,i[0],i[1]),i.join("=")}),t}function FI(n,t,e){n[decodeURIComponent(t)]=decodeURIComponent(e||"")}function kI(n,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(""+ +t===t)return+t;throw new Error("Could not parse value flag value "+t+" for flag "+n+".")}function He(){return I.ENV}I.ENV=null;function WI(n){I.ENV=n}var nl;function kp(){if(nl==null){var n=void 0;if(typeof window!="undefined")n=window;else if(typeof global!="undefined")n=global;else if(typeof process!="undefined")n=process;else if(typeof self!="undefined")n=self;else throw new Error("Could not find a global object");nl=n}return nl}function UI(){var n=kp();return n._tfGlobals==null&&(n._tfGlobals=new Map),n._tfGlobals}function Fp(n,t){var e=UI();if(e.has(n))return e.get(n);var i=t();return e.set(n,i),e.get(n)}var il="Abs",rl="Acos",al="Acosh",ps="Add",sl="AddN",Wp="All",Up="Any",ol="ArgMax",ll="ArgMin",ul="Asin",cl="Asinh",hl="Atan",dl="Atanh",pl="Atan2",fl="AvgPool",Bp="AvgPoolBackprop",ml="AvgPool3D",zp="AvgPool3DBackprop",gl="BatchMatMul",vl="BatchToSpaceND",yl="BroadcastTo",fs="Cast",bl="Ceil",wl="ClipByValue",Pp="Complex",Sl="Concat",Ll="Conv2D",_p="Conv2DBackpropFilter",Il="Conv2DBackpropInput",Al="Conv3D",Mp="Conv3DBackpropFilterV2",Hp="Conv3DBackpropInputV2",Tl="Cos",Nl="Cosh",xl="Cumsum",Vp="CropAndResize",qp="DepthToSpace",Cl="DepthwiseConv2dNative",Gp="DepthwiseConv2dNativeBackpropFilter",Yp="DepthwiseConv2dNativeBackpropInput",Kp="Diag",Rl="Dilation2D",jp="Dilation2DBackpropInput",$p="Dilation2DBackpropFilter",Ol="Div",El="Elu",Xp="EluGrad",Dl="Erf",Jp="Equal",kl="Exp",Fl="Expm1",Zp="FFT",Qp="Fill",ef="FlipLeftRight",Wl="Floor",Ul="FloorDiv",Bl="FusedBatchNorm",zl="GatherV2",tf="GatherNd",nf="Greater",Pl="GreaterEqual",_l="Identity",rf="IFFT",af="Imag",Ml="IsFinite",Hl="IsInf",Vl="IsNan",sf="Less",of="LessEqual",lf="LinSpace",ql="Log",Gl="Log1p",uf="LogicalAnd",cf="LogicalNot",hf="LogicalOr",Yl="LogSoftmax",Kl="LRN",df="LRNBackprop",jl="Max",$l="Maximum",Xl="MaxPool",pf="MaxPoolBackprop",Jl="MaxPool3D",ff="MaxPool3DBackprop",mf="MaxPoolWithArgmax",BI="Mean",Zl="Min",Ql="Minimum",eu="Mod",tu="Multiply",nu="Negate",gf="NotEqual",vf="NonMaxSuppressionV3",yf="NonMaxSuppressionV4",bf="NonMaxSuppressionV5",iu="OnesLike",ru="OneHot",au="PadV2",zI="Pool",su="Pow",ou="Prelu",wf="Prod",Sf="Range",Lf="Real",lu="Reciprocal",uu="Relu",cu="Reshape",hu="ResizeNearestNeighbor",If="ResizeNearestNeighborGrad",du="ResizeBilinear",Af="ResizeBilinearGrad",pu="Relu6",fu="Reverse",mu="Round",gu="Rsqrt",Tf="ScatterNd",vu="SelectV2",yu="Selu",bu="Slice",wu="Sin",Su="Sinh",Lu="Sign",Iu="Sigmoid",Au="Softplus",Tu="Sqrt",Nu="Sum",xu="SpaceToBatchND",Cu="SplitV",Ru="Softmax",Ou="SquaredDifference",Nf="Square",Eu="Sub",xf="SparseToDense",Cf="StridedSlice",Du="Tan",ku="Tanh",Fu="Tile",Rf="TopK",Wu="Transpose",Of="Unique",Uu="Unpack",Bu="UnsortedSegmentSum",zu="ZerosLike",Pu="Step",_u="FromPixels",Ef="RotateWithOffset",Mu="_FusedMatMul",Hu="FusedConv2D",Vu="FusedDepthwiseConv2D";var br=Fp("kernelRegistry",function(){return new Map}),da=Fp("gradRegistry",function(){return new Map});function Gu(n,t){var e=qu(n,t);return br.get(e)}function Yu(n){return da.get(n)}function ms(n){for(var t=br.entries(),e=[];;){var i=t.next(),r=i.done,a=i.value;if(r)break;var s=a[0],o=a[1],l=s.split("_")[0];l===n&&e.push(o)}return e}function Df(n){var t=n.kernelName,e=n.backendName,i=qu(t,e);br.has(i)&&console.warn("The kernel '"+t+"' for backend "+("'"+e+"' is already registered")),br.set(i,n)}function kf(n){var t=n.kernelName;da.has(t)&&(He().getBool("DEBUG")&&console.warn("Overriding the gradient for '"+t+"'")),da.set(t,n)}function PI(n,t){var e=qu(n,t);if(!br.has(e))throw new Error("The kernel '"+n+"' for backend "+("'"+t+"' is not registered"));br.delete(e)}function _I(n){if(!da.has(n))throw new Error("The gradient '"+n+"' for backend is not registered");da.delete(n)}function MI(n,t){var e=ms(n);e.forEach(function(i){var r=Object.assign({},i,{backendName:t});Df(r)})}function qu(n,t){return t+"_"+n}function Ff(n){for(var t=n.length,e=0,i=0;t>0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function pa(n,t,e){return Math.max(n,Math.min(t,e))}function HI(n){return n%2===0?n:n+1}function VI(n){for(var t=0,e=0;e<n.length;e++)t+=n[e];return t}function qI(n,t){var e=Math.random();return t*e+(1-e)*n}function GI(n,t){for(var e=0,i=0;i<n.length;i++){var r=Number(n[i])-Number(t[i]);e+=r*r}return e}function E(n,t){if(!n)throw new Error(typeof t=="string"?t:t())}function Be(n,t,e){e===void 0&&(e=""),E(on(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function Fi(n){E(n!=null,function(){return"The input to the tensor constructor must be a non-null value."})}function Wi(n,t,e){if(t===void 0&&(t=[]),e===void 0&&(e=!1),t==null&&(t=[]),Array.isArray(n)||Ot(n)&&!e)for(var i=0;i<n.length;++i)Wi(n[i],t,e);else t.push(n);return t}function ot(n){if(n.length===0)return 1;for(var t=n[0],e=1;e<n.length;e++)t*=n[e];return t}function YI(n){return n.length===0}function on(n,t){if(n===t)return!0;if(n==null||t==null)return!1;if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==t[e])return!1;return!0}function rt(n){return n%1===0}function KI(n){if(Math.tanh!=null)return Math.tanh(n);if(n===Infinity)return 1;if(n===-Infinity)return-1;var t=Math.exp(2*n);return(t-1)/(t+1)}function jI(n){var t=Math.ceil(Math.sqrt(n));return[t,Math.ceil(n/t)]}function $I(n){for(var t=new Uint32Array(n),e=0;e<n;++e)t[e]=e;return Ff(t),t}function fa(n,t){return t<=n.length?n:n+" ".repeat(t-n.length)}function XI(n,t,e){return t===void 0&&(t=function(i){return 0}),new Promise(function(i,r){var a=0,s=function(){if(n()){i();return}a++;var o=t(a);if(e!=null&&a>=e){r();return}setTimeout(s,o)};s()})}function Wf(n,t){for(var e=1,i=-1,r=0;r<n.length;++r)if(n[r]>=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Qe(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&i<e}),function(){return"All values in axis param must be in range [-"+e+", "+e+") but "+("got axis "+n)}),E(n.every(function(i){return rt(i)}),function(){return"All values in axis param must be integers but "+("got axis "+n)}),n.map(function(i){return i<0?e+i:i})}function Uf(n,t){for(var e=[],i=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Qe(t,n).sort(),s=0,o=0;o<n.length;++o){if(a!=null){if(a[s]===o&&n[o]!==1)throw new Error("Can't squeeze axis "+o+" since its dim '"+n[o]+"' is not 1");(a[s]==null||a[s]>o)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function gs(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function Bf(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function zf(n,t){for(var e=0;e<n.length;e++){var i=n[e];if(isNaN(i)||!isFinite(i))throw Error("A tensor of type "+t+" being uploaded contains "+i+".")}}function Pf(n){return n==="bool"||n==="complex64"||n==="float32"||n==="int32"||n==="string"}function _f(n,t){return t==="complex64"||(t==="float32"&&n!=="complex64"||t==="int32"&&n!=="float32"&&n!=="complex64")?!1:!(t==="bool"&&n==="bool")}function Ot(n){return n instanceof Float32Array||n instanceof Int32Array||n instanceof Uint8Array}function Mf(n){if(n==="float32"||n==="int32")return 4;if(n==="complex64")return 8;if(n==="bool")return 1;throw new Error("Unknown dtype "+n)}function Hf(n){if(n==null)return 0;var t=0;return n.forEach(function(e){return t+=e.length}),t}function ni(n){return typeof n=="string"||n instanceof String}function Vf(n){return typeof n=="boolean"}function qf(n){return typeof n=="number"}function vs(n){return Array.isArray(n)?vs(n[0]):n instanceof Float32Array?"float32":n instanceof Int32Array||n instanceof Uint8Array?"int32":qf(n)?"float32":ni(n)?"string":Vf(n)?"bool":"float32"}function ii(n){return!!(n&&n.constructor&&n.call&&n.apply)}function ys(n,t){for(var e=t;e<n;++e)if(n%e===0)return e;return n}function wr(n){var t=n.length;if(t<2)return[];var e=new Array(t-1);e[t-2]=n[t-1];for(var i=t-3;i>=0;--i)e[i]=e[i+1]*n[i+1];return e}function JI(n,t){return t==="string"?Ku(n):bs([n],t)}function bs(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=Wi(n)),He().getBool("DEBUG")&&zf(n,t),ZI(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i<e.length;++i)Math.round(n[i])!==0&&(e[i]=1);return e}else throw new Error("Unknown data type "+t)}function Gf(n,t,e){var i=new Array;if(t.length===1)for(var r=t[0],a=0;a<r;a++)i[a]=e[n+a];else for(var r=t[0],s=t.slice(1),o=s.reduce(function(u,c){return u*c}),a=0;a<r;a++)i[a]=Gf(n+a*o,s,e);return i}function Sr(n,t){if(n.length===0)return t[0];var e=n.reduce(function(i,r){return i*r});if(e===0)return[];if(e!==t.length)throw new Error("["+n+"] does not match the input size "+t.length+".");return Gf(0,n,t)}function ZI(n,t){return n instanceof Float32Array&&t==="float32"||n instanceof Int32Array&&t==="int32"||n instanceof Uint8Array&&t==="bool"}function ju(n,t){for(var e=Lr(n,t),i=0;i<e.length;i++)e[i]=1;return e}function Lr(n,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool")return new Uint8Array(n);throw new Error("Unknown data type "+t)}function QI(n,t){var e=n.reduce(function(i,r){return i*r},1);if(t==null||t==="float32")return Sr(n,new Float32Array(e));if(t==="int32")return Sr(n,new Int32Array(e));if(t==="bool")return Sr(n,new Uint8Array(e));throw new Error("Unknown data type "+t)}function $u(){return He().platform.now()}function Xu(n){n.forEach(function(t){E(Number.isInteger(t)&&t>=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function e2(n,t){return He().platform.fetch(n,t)}function Ku(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.encode(n,t)}function Ju(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",He().platform.decode(n,t)}function t2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r<n.length-1;++r)i+=e[r]*n[r];return i}function n2(n,t,e){if(t===0)return[];if(t===1)return[n];for(var i=new Array(t),r=0;r<i.length-1;++r)i[r]=Math.floor(n/e[r]),n-=i[r]*e[r];return i[i.length-1]=n,i}var i2={__proto__:null,shuffle:Ff,clamp:pa,nearestLargerEven:HI,sum:VI,randUniform:qI,distSquared:GI,assert:E,assertShapesMatch:Be,assertNonNull:Fi,flatten:Wi,sizeFromShape:ot,isScalarShape:YI,arraysEqual:on,isInt:rt,tanh:KI,sizeToSquarishShape:jI,createShuffledIndices:$I,rightPad:fa,repeatedTry:XI,inferFromImplicitShape:Wf,parseAxisParam:Qe,squeezeShape:Uf,getTypedArrayFromDType:gs,getArrayFromDType:Bf,checkConversionForErrors:zf,isValidDtype:Pf,hasEncodingLoss:_f,isTypedArray:Ot,bytesPerElement:Mf,bytesFromStringArray:Hf,isString:ni,isBoolean:Vf,isNumber:qf,inferDtype:vs,isFunction:ii,nearestDivisor:ys,computeStrides:wr,createScalarValue:JI,toTypedArray:bs,toNestedArray:Sr,makeOnesTypedArray:ju,makeZerosTypedArray:Lr,makeZerosNestedTypedArray:QI,now:$u,assertNonNegativeIntegerDimensions:Xu,fetch:e2,encodeString:Ku,decodeString:Ju,locToIndex:t2,indexToLoc:n2};var s2=function(){function n(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new a2)}return n.prototype.profileKernel=function(t,e,i){for(var r,a=function(){r=i()},s=this.backendTimer.time(a),o=function(c){var h=r[c];h.data().then(function(d){r2(d,h.dtype,t)})},l=0;l<r.length;l++)o(l);var u={kernelName:t,outputs:r,inputs:e,timeMs:s.then(function(c){return c.kernelMs}),extraInfo:s.then(function(c){return c.getExtraProfileInfo!=null?c.getExtraProfileInfo():""})};return u},n.prototype.logKernelProfile=function(t){var e=this,i=t.kernelName,r=t.outputs,a=t.timeMs,s=t.inputs,o=t.extraInfo;r.forEach(function(l){Promise.all([l.data(),a,o]).then(function(u){e.logger.logKernelProfile(i,l,u[0],u[1],s,u[2])})})},n}();function r2(n,t,e){if(t!=="float32")return!1;for(var i=0;i<n.length;i++){var r=n[i];if(isNaN(r)||!isFinite(r))return console.warn("Found "+r+" in the result of '"+e+"'"),!0}return!1}var a2=function(){function n(){}return n.prototype.logKernelProfile=function(t,e,i,r,a,s){var o=typeof r=="number"?fa(r+"ms",9):r.error,l=fa(t,25),u=e.rank,c=e.size,h=fa(e.shape.toString(),14),d="";for(var p in a){var f=a[p];if(f!=null){var m=f.shape||e.shape,g=m.length;d+=p+": "+g+"D "+(g>0?m:"")+" "}}console.log("%c"+l+"	%c"+o+"	%c"+u+"D "+h+"	%c"+c+"	%c"+d+"	%c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function o2(n,t,e){for(var i={},r={},a=0;a<t.length;a++)i[t[a].id]=!0;for(var a=0;a<n.length;a++){var s=n[a],o=s.inputs;for(var l in o){for(var u=o[l],c=!1,h=0;h<t.length;h++)if(i[u.id]){s.outputs.forEach(function(w){return i[w.id]=!0}),c=!0,r[s.id]=!0;break}if(c)break}}var d={};d[e.id]=!0;for(var p={},a=n.length-1;a>=0;a--)for(var s=n[a],o=s.inputs,h=0;h<s.outputs.length;h++)if(d[s.outputs[h].id]){for(var l in o)d[o[l].id]=!0,p[s.id]=!0;break}for(var f=[],a=0;a<n.length;a++){var s=n[a];if(r[s.id]&&p[s.id]){var m={};for(var l in s.inputs){var g=s.inputs[l];i[g.id]&&(m[l]=g)}var v=Object.assign({},s);v.inputs=m,v.outputs=s.outputs,f.push(v)}}return f}function l2(n,t,e,i){for(var r=function(s){var o=t[s],l=[];if(o.outputs.forEach(function(d){var p=n[d.id];p!=null?l.push(p):l.push(null)}),o.gradient==null)throw new Error("Cannot compute gradient: gradient function not found "+("for "+o.kernelName+"."));var u=o.gradient(l),c=function(d){if(!(d in u))throw new Error("Cannot backprop through input "+d+". "+("Available gradients found: "+Object.keys(u)+"."));var p=e(function(){return u[d]()});if(p.dtype!=="float32")throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+(d+" must have 'float32' dtype, but has '"+p.dtype+"'"));var f=o.inputs[d];if(!on(p.shape,f.shape))throw new Error("Error in gradient for op "+o.kernelName+". The gradient of input "+("'"+d+"' has shape '"+p.shape+"', which does not match ")+("the shape of the input '"+f.shape+"'"));if(n[f.id]==null)n[f.id]=p;else{var m=n[f.id];n[f.id]=i(m,p),m.dispose()}};for(var h in o.inputs)c(h)},a=t.length-1;a>=0;a--)r(a)}var Yf=20,ma=3,Zu=7;function c2(n,t,e,i){var r=wr(t),a=u2(n,t,e,r),s=t.length,o=ws(n,t,e,r,a),l=["Tensor"];return i&&(l.push("  dtype: "+e),l.push("  rank: "+s),l.push("  shape: ["+t+"]"),l.push("  values:")),l.push(o.map(function(u){return"    "+u}).join(`
 `)),l.join(`
-`)}function u2(n,t,e,i){var r=ot(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?va(n):n;if(o>1)for(var u=0;u<r/a;u++)for(var c=u*a,h=0;h<a;h++)s[h]=Math.max(s[h],ga(l[c+h],0,e).length);return s}function ga(n,t,e){var i;return Array.isArray(n)?i=parseFloat(n[0].toFixed(Ju))+" + "+(parseFloat(n[1].toFixed(Ju))+"j"):ni(n)?i="'"+n+"'":e==="bool"?i=Kf(n):i=parseFloat(n.toFixed(Ju)).toString(),fa(i,t)}function Kf(n){return n===0?"false":"true"}function ws(n,t,e,i,r,a){a===void 0&&(a=!0);var s=e==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(e==="complex64"){var u=va(n);return[ga(u[0],0,e)]}return e==="bool"?[Kf(n[0])]:[n[0].toString()]}if(l===1){if(o>Yf){var c=ma*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-ma)*s,o*s));return e==="complex64"&&(h=va(h),d=va(d)),["["+h.map(function(C,R){return ga(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return ga(C,r[o-ma+R],e)}).join(", ")+"]"]}var p=e==="complex64"?va(n):Array.from(n);return["["+p.map(function(C,R){return ga(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>Yf){for(var b=0;b<ma;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,!1))}v.push("...");for(var b=o-ma;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}}else for(var b=0;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}var L=l===2?",":"";v[0]="["+v[0]+L;for(var b=1;b<v.length-1;b++)v[b]=" "+v[b]+L;for(var x=`,
+`)}function u2(n,t,e,i){var r=ot(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?va(n):n;if(o>1)for(var u=0;u<r/a;u++)for(var c=u*a,h=0;h<a;h++)s[h]=Math.max(s[h],ga(l[c+h],0,e).length);return s}function ga(n,t,e){var i;return Array.isArray(n)?i=parseFloat(n[0].toFixed(Zu))+" + "+(parseFloat(n[1].toFixed(Zu))+"j"):ni(n)?i="'"+n+"'":e==="bool"?i=Kf(n):i=parseFloat(n.toFixed(Zu)).toString(),fa(i,t)}function Kf(n){return n===0?"false":"true"}function ws(n,t,e,i,r,a){a===void 0&&(a=!0);var s=e==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(e==="complex64"){var u=va(n);return[ga(u[0],0,e)]}return e==="bool"?[Kf(n[0])]:[n[0].toString()]}if(l===1){if(o>Yf){var c=ma*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-ma)*s,o*s));return e==="complex64"&&(h=va(h),d=va(d)),["["+h.map(function(C,R){return ga(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return ga(C,r[o-ma+R],e)}).join(", ")+"]"]}var p=e==="complex64"?va(n):Array.from(n);return["["+p.map(function(C,R){return ga(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>Yf){for(var b=0;b<ma;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,!1))}v.push("...");for(var b=o-ma;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}}else for(var b=0;b<o;b++){var w=b*g,S=w+g;v.push.apply(v,ws(n.slice(w,S),f,e,m,r,b===o-1))}var L=l===2?",":"";v[0]="["+v[0]+L;for(var b=1;b<v.length-1;b++)v[b]=" "+v[b]+L;for(var x=`,
 `,b=2;b<l;b++)x+=`
-`;return v[v.length-1]=" "+v[v.length-1]+"]"+(a?"":x),v}function va(n){for(var t=[],e=0;e<n.length;e+=2)t.push([n[e],n[e+1]]);return t}var Ss=function(){function n(t,e,i){var r=this;if(this.dtype=e,this.shape=t.slice(),this.size=ot(t),i!=null){var a=i.length;E(a===this.size,function(){return"Length of values '"+a+"' does not match the size "+("inferred by the shape '"+r.size+"'.")})}if(e==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=i||Bf(e,this.size),this.strides=wr(t)}return n.prototype.set=function(t){for(var e=this,i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];i.length===0&&(i=[0]),E(i.length===this.rank,function(){return"The number of provided coordinates ("+i.length+") must "+("match the rank ("+e.rank+")")});var a=this.locToIndex(i);this.values[a]=t},n.prototype.get=function(){for(var t=[],e=0;e<arguments.length;e++)t[e]=arguments[e];t.length===0&&(t=[0]);for(var i=0,r=0,a=t;r<a.length;r++){var s=a[r];if(s<0||s>=this.shape[i]){var o="Requested out of range element at "+t+". "+("  Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u<t.length-1;++u)l+=this.strides[u]*t[u];return this.values[l]},n.prototype.locToIndex=function(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];for(var e=t[t.length-1],i=0;i<t.length-1;++i)e+=this.strides[i]*t[i];return e},n.prototype.indexToLoc=function(t){if(this.rank===0)return[];if(this.rank===1)return[t];for(var e=new Array(this.shape.length),i=0;i<e.length-1;++i)e[i]=Math.floor(t/this.strides[i]),t-=e[i]*this.strides[i];return e[e.length-1]=t,e},Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.toTensor=function(){return Cn().makeTensor(this.values,this.shape,this.dtype)},n}(),Cn=null,Ir=null;function h2(n){Cn=n}function d2(n){Ir=n}var Y=function(){function n(t,e,i,r){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=ot(t),this.strides=wr(t),this.dataId=i,this.id=r,this.rankType=this.rank<5?this.rank.toString():"higher"}return Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.buffer=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Ir.buffer(this.shape,this.dtype,t)]}})})},n.prototype.bufferSync=function(){return Ir.buffer(this.shape,this.dtype,this.dataSync())},n.prototype.array=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Sr(this.shape,t)]}})})},n.prototype.arraySync=function(){return Sr(this.shape,this.dataSync())},n.prototype.data=function(){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){switch(i.label){case 0:return this.throwIfDisposed(),t=Cn().read(this.dataId),this.dtype==="string"?[4,t]:[3,2];case 1:e=i.sent();try{return[2,e.map(function(r){return Xu(r)})]}catch(r){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}i.label=2;case 2:return[2,t]}})})},n.prototype.dataSync=function(){this.throwIfDisposed();var t=Cn().readSync(this.dataId);if(this.dtype==="string")try{return t.map(function(e){return Xu(e)})}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t},n.prototype.bytes=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return this.throwIfDisposed(),[4,Cn().read(this.dataId)];case 1:return t=e.sent(),this.dtype==="string"?[2,t]:[2,new Uint8Array(t.buffer)]}})})},n.prototype.dispose=function(){if(this.isDisposed)return;Cn().disposeTensor(this),this.isDisposedInternal=!0},Object.defineProperty(n.prototype,"isDisposed",{get:function(){return this.isDisposedInternal},enumerable:!0,configurable:!0}),n.prototype.throwIfDisposed=function(){if(this.isDisposed)throw new Error("Tensor is disposed.")},n.prototype.print=function(t){return t===void 0&&(t=!1),Ir.print(this,t)},n.prototype.clone=function(){return this.throwIfDisposed(),Ir.clone(this)},n.prototype.toString=function(t){t===void 0&&(t=!1);var e=this.dataSync();return c2(e,this.shape,this.dtype,t)},n.prototype.cast=function(t){return this.throwIfDisposed(),Ir.cast(this,t)},n.prototype.variable=function(t,e,i){return t===void 0&&(t=!0),this.throwIfDisposed(),Cn().makeVariable(this,t,e,i)},n}();Object.defineProperty(Y,Symbol.hasInstance,{value:function(n){return!!n&&n.data!=null&&n.dataSync!=null&&n.throwIfDisposed!=null}});var ya=function(n){qn(t,n);function t(e,i,r,a){var s=n.call(this,e.shape,e.dtype,e.dataId,a)||this;return s.trainable=i,s.name=r,s}return t.prototype.assign=function(e){if(e.dtype!==this.dtype)throw new Error("dtype of the new value ("+e.dtype+") and "+("previous value ("+this.dtype+") must match"));if(!ln(e.shape,this.shape))throw new Error("shape of the new value ("+e.shape+") and "+("previous value ("+this.shape+") must match"));Cn().disposeTensor(this),this.dataId=e.dataId,Cn().incRef(this,null)},t.prototype.dispose=function(){Cn().disposeVariable(this),this.isDisposedInternal=!0},t}(Y);Object.defineProperty(ya,Symbol.hasInstance,{value:function(n){return n instanceof Y&&n.assign!=null&&n.assign instanceof Function}});(function(n){n.R0="R0",n.R1="R1",n.R2="R2",n.R3="R3",n.R4="R4",n.R5="R5",n.R6="R6"})(I.Rank||(I.Rank={}));var Zu;(function(n){n.float32="float32",n.int32="int32",n.bool="int32",n.complex64="complex64"})(Zu||(Zu={}));var Qu;(function(n){n.float32="float32",n.int32="int32",n.bool="bool",n.complex64="complex64"})(Qu||(Qu={}));var ec;(function(n){n.float32="float32",n.int32="float32",n.bool="float32",n.complex64="complex64"})(ec||(ec={}));var tc;(function(n){n.float32="complex64",n.int32="complex64",n.bool="complex64",n.complex64="complex64"})(tc||(tc={}));var p2={float32:ec,int32:Zu,bool:Qu,complex64:tc};function Ls(n,t){if(n==="string"||t==="string"){if(n==="string"&&t==="string")return"string";throw new Error("Can not upcast "+n+" with "+t)}return p2[n][t]}function f2(n){return Ls(n,"int32")}function at(n,t){if(n.dtype===t.dtype)return[n,t];var e=Ls(n.dtype,t.dtype);return[n.cast(e),t.cast(e)]}function jf(n,t){E(n.dtype===t.dtype,function(){return"The dtypes of the first("+n.dtype+") and"+(" second("+t.dtype+") input must match")})}function m2(n,t){return t.some(function(e){return e.id===n.id})}function nc(n){var t=[],e=new Set;return $f(n,t,e),t}function $f(n,t,e){if(n==null)return;if(n instanceof Y){t.push(n);return}if(!g2(n))return;var i=n;for(var r in i){var a=i[r];e.has(a)||(e.add(a),$f(a,t,e))}}function g2(n){return Array.isArray(n)||typeof n=="object"}var v2={__proto__:null,makeTypesMatch:at,assertTypesMatch:jf,isTensorInList:m2,getTensorsInContainer:nc};var Xf=function(){function n(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null}}return n.prototype.dispose=function(){for(var t in this.registeredVariables)this.registeredVariables[t].dispose()},n}(),w2=function(){function n(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Xf}return n.prototype.ready=function(){return de(this,void 0,void 0,function(){var t,e,i,r;return pe(this,function(a){switch(a.label){case 0:if(this.pendingBackendInit!=null)return[2,this.pendingBackendInit.then(function(){})];if(this.backendInstance!=null)return[2];t=this.getSortedBackends(),e=0,a.label=1;case 1:return e<t.length?(i=t[e],[4,this.initializeBackend(i).success]):[3,5];case 2:return r=a.sent(),r?[4,this.setBackend(i)]:[3,4];case 3:return a.sent(),[2];case 4:return e++,[3,1];case 5:throw new Error("Could not initialize any backends, all backend initializations failed.")}})})},Object.defineProperty(n.prototype,"backend",{get:function(){if(this.pendingBackendInit!=null)throw new Error("Backend '"+this.backendName+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");if(this.backendInstance==null){var t=this.initializeBackendsAndReturnBest(),e=t.name,i=t.asyncInit;if(i)throw new Error("The highest priority backend '"+e+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");this.setBackend(e)}return this.backendInstance},enumerable:!0,configurable:!0}),n.prototype.backendNames=function(){return Object.keys(this.registryFactory)},n.prototype.findBackend=function(t){if(!(t in this.registry))if(t in this.registryFactory){var e=this.initializeBackend(t).asyncInit;if(e)return null}else return null;return this.registry[t]},n.prototype.findBackendFactory=function(t){return t in this.registryFactory?this.registryFactory[t].factory:null},n.prototype.registerBackend=function(t,e,i){return i===void 0&&(i=1),t in this.registryFactory?(console.warn(t+" backend was already registered. Reusing existing backend factory."),!1):(this.registryFactory[t]={factory:e,priority:i},!0)},n.prototype.setBackend=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s;return pe(this,function(o){switch(o.label){case 0:if(this.registryFactory[t]==null)throw new Error("Backend name '"+t+"' not found in registry");return this.backendName=t,this.registry[t]==null?(this.backendInstance=null,e=this.initializeBackend(t),i=e.success,r=e.asyncInit,r?[4,i]:[3,2]):[3,4];case 1:return s=o.sent(),[3,3];case 2:s=i,o.label=3;case 3:if(a=s,!a)return[2,!1];o.label=4;case 4:return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new s2(this.backendInstance),[2,!0]}})})},n.prototype.setupRegisteredKernels=function(){var t=this,e=ms(this.backendName);e.forEach(function(i){i.setupFunc!=null&&i.setupFunc(t.backendInstance)})},n.prototype.disposeRegisteredKernels=function(t){var e=this,i=ms(t);i.forEach(function(r){r.disposeFunc!=null&&r.disposeFunc(e.registry[t])})},n.prototype.initializeBackend=function(t){var e=this,i=this.registryFactory[t];if(i==null)throw new Error("Cannot initialize backend "+t+", no registration found.");try{var r=i.factory();if(r&&!(r instanceof Op)&&typeof r.then=="function"){var a=++this.pendingBackendInitId,s=r.then(function(o){return a<e.pendingBackendInitId?!1:(e.registry[t]=o,e.pendingBackendInit=null,!0)}).catch(function(o){return a<e.pendingBackendInitId||(e.pendingBackendInit=null,console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message)),!1});return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=r,{success:!0,asyncInit:!1}}catch(o){return console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message),{success:!1,asyncInit:!1}}},n.prototype.removeBackend=function(t){if(!(t in this.registryFactory))throw new Error(t+" backend not found in registry");this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)},n.prototype.getSortedBackends=function(){var t=this;if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(function(e,i){return t.registryFactory[i].priority-t.registryFactory[e].priority})},n.prototype.initializeBackendsAndReturnBest=function(){for(var t=this.getSortedBackends(),e=0;e<t.length;e++){var i=t[e],r=this.initializeBackend(i),a=r.success,s=r.asyncInit;if(s||a)return{name:i,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")},n.prototype.moveData=function(t,e){var i=this.state.tensorInfo.get(e),r=i.backend,a=this.readSync(e);r.disposeData(e),i.backend=t,t.move(e,a,i.shape,i.dtype),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++},n.prototype.tidy=function(t,e){var i=this,r=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");r=t}var a;return this.scopedRun(function(){return i.startScope(r)},function(){return i.endScope(a)},function(){return a=e(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a})},n.prototype.scopedRun=function(t,e,i){t();try{var r=i();return e(),r}catch(a){throw e(),a}},n.prototype.nextTensorId=function(){return n.nextTensorId++},n.prototype.nextVariableId=function(){return n.nextVariableId++},n.prototype.clone=function(t){var e=this.makeTensorFromDataId(t.dataId,t.shape,t.dtype),i={x:t},r=function(s){return{x:function(){var o="float32",l={x:s},u={dtype:o};return z.runKernelFunc(function(c){return c.cast(s,o)},l,null,fs,u)}}},a=[];return this.addTapeNode(this.state.activeScope.name,i,[e],r,a,{}),e},n.prototype.runKernel=function(t,e,i,r,a){var s=null,o=null;return this.runKernelFunc(s,e,o,t,i,r,a)},n.prototype.shouldCheckForMemLeaks=function(){return this.ENV.getBool("IS_TEST")},n.prototype.checkKernelForMemLeak=function(t,e,i){var r=this.backend.numDataIds(),a=0;i.forEach(function(l){a+=l.dtype==="complex64"?3:1});var s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=r-e-a-s;if(o>0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=qu(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var x=l.getTensorsForGradient(r,e,L);if(x==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});x=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(x)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=Gu(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&ni(t[0])&&(a=t.map(function(c){return Yu(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Hf(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new ya(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Mf(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof ya||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a<s.length?(o=s[a],l=o,[4,o.kernelTimeMs]):[3,6];case 3:return l.kernelTimeMs=c.sent(),u=o,[4,o.extraInfo];case 4:u.extraInfo=c.sent(),c.label=5;case 5:return a++,[3,2];case 6:return[2,this.state.activeProfile]}})})},n.prototype.isTapeOn=function(){return this.state.gradientDepth>0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=Gu(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=Lr(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=nc(t),r=new Set(i.map(function(l){return l.id})),a=0;a<this.state.activeScope.track.length;a++){var s=this.state.activeScope.track[a];!s.kept&&!r.has(s.id)&&s.dispose()}var o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],i.forEach(function(l){!l.kept&&l.scopeId===o.id&&e.track(l)})},n.prototype.gradients=function(t,e,i,r){var a=this;if(r===void 0&&(r=!1),E(e.length>0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=o2(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??y2(s.shape),l2(l,o,function(c){return a.tidy(c)},b2);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h<d.length;h++){var p=d[h];p.dispose()}}),a.state.activeTape=null),{value:s,grads:u}})},n.prototype.customGrad=function(t){var e=this;return E(ii(t),function(){return"The f passed in customGrad(f) must be a function."}),function(){for(var i=[],r=0;r<arguments.length;r++)i[r]=arguments[r];E(i.every(function(o){return o instanceof Y}),function(){return"The args passed in customGrad(f)(x1, x2,...) must all be tensors"});var a,s={};return i.forEach(function(o,l){s[l]=o}),e.runKernelFunc(function(o,l){return a=t.apply(void 0,i.concat([l])),E(a.value instanceof Y,function(){return"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"}),E(ii(a.gradFunc),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."}),a.value},s,function(o,l){var u=a.gradFunc(o,l),c=Array.isArray(u)?u:[u];E(c.length===i.length,function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."}),E(c.every(function(d){return d instanceof Y}),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."});var h={};return c.forEach(function(d,p){h[p]=function(){return d}}),h})}},n.prototype.readSync=function(t){var e=this.state.tensorInfo.get(t);return e.backend.readSync(t)},n.prototype.read=function(t){var e=this.state.tensorInfo.get(t);return e.backend.read(t)},n.prototype.time=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return e=ju(),[4,this.backend.time(t)];case 1:return i=r.sent(),i.wallMs=ju()-e,[2,i]}})})},n.prototype.track=function(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t},Object.defineProperty(n.prototype,"registeredVariables",{get:function(){return this.state.registeredVariables},enumerable:!0,configurable:!0}),n.prototype.reset=function(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Xf;for(var t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null},n.nextTensorId=0,n.nextVariableId=0,n}();function y2(n){var t=Ku(ot(n),"float32");return z.makeTensor(t,n,"float32")}function Jf(){var n=kp();if(n._tfengine==null){var t=new Dp(n);n._tfengine=new w2(t)}return WI(n._tfengine.ENV),h2(function(){return n._tfengine}),n._tfengine}var z=Jf();function b2(n,t){var e={a:n,b:t};return z.runKernelFunc(function(i,r){var a=i.add(n,t);return r([n,t]),a},e,null,ps)}function S2(){return typeof navigator!="undefined"&&navigator!=null}function L2(){if(S2()){var n=navigator.userAgent||navigator.vendor||window.opera;return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(n)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(n.substr(0,4))}return!1}function Zf(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var I2={__proto__:null,isMobile:L2,isBrowser:Zf};var Gn=He();Gn.registerFlag("DEBUG",function(){return!1},function(n){n&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});Gn.registerFlag("IS_BROWSER",function(){return Zf()});Gn.registerFlag("IS_NODE",function(){return typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined"});Gn.registerFlag("IS_CHROME",function(){return typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor)});Gn.registerFlag("PROD",function(){return!1});Gn.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",function(){return Gn.getBool("DEBUG")});Gn.registerFlag("DEPRECATION_WARNINGS_ENABLED",function(){return!0});Gn.registerFlag("IS_TEST",function(){return!1});function Rn(n,t){var e=n;if(Ot(n))return t==="string"?[]:[n.length];if(!Array.isArray(n))return[];for(var i=[];Array.isArray(e)||Ot(e)&&t!=="string";)i.push(e.length),e=e[0];return Array.isArray(n)&&He().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Qf(n,i,[]),i}function Qf(n,t,e){if(e=e||[],!Array.isArray(n)&&!Ot(n)){E(t.length===0,function(){return"Element arr["+e.join("][")+"] is a primitive, "+("but should be an array/TypedArray of "+t[0]+" elements")});return}E(t.length>0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r<n.length;++r)Qf(n[r],i,e.concat(r))}function em(n,t,e,i){if(n==null)return;if(n!=="numeric"&&n!==t||n==="numeric"&&t==="string")throw new Error("Argument '"+e+"' passed to '"+i+"' must "+("be "+n+" tensor, but got "+t+" tensor"))}function O(n,t,e,i){if(i===void 0&&(i="numeric"),n instanceof Y)return em(i,n.dtype,t,e),n;var r=vs(n);if(r!=="string"&&["bool","int32","float32"].indexOf(i)>=0&&(r=i),em(i,r,t,e),n==null||!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=Rn(n,r);!Ot(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?bs(n,r):Wi(n,[],o);return z.makeTensor(l,s,r)}function ba(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var tm="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+tm;var r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];z.startScope(e);try{var o=i.apply(void 0,a);return o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),z.endScope(o),o}catch(l){throw z.endScope(null),l}};return Object.defineProperty(r,"name",{value:e,configurable:!0}),r}function A2(n,t){var e=O(n,"real","complex"),i=O(t,"imag","complex");Be(e.shape,i.shape,"real and imag shapes, "+e.shape+" and "+i.shape+", must match in call to tf.complex().");var r=function(s){return s.complex(e,i)},a={real:e,imag:i};return z.runKernelFunc(r,a,null,Pp)}var ri=U({complex_:A2});function ai(n,t,e,i){if(i==null&&(i=vs(n)),i==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){$u(t);var r=ot(t),a=ot(e);E(r===a,function(){return"Based on the provided shape, ["+t+"], the tensor should have "+(r+" values but has "+a)});for(var s=0;s<e.length;++s){var o=e[s],l=s===e.length-1?o!==ot(t.slice(s)):!0;E(e[s]===t[s]||!l,function(){return"Error creating a new Tensor. Inferred shape "+("("+e+") does not match the provided ")+("shape ("+t+"). ")})}}return!Ot(n)&&!Array.isArray(n)&&(n=[n]),t=t||e,n=i!=="string"?bs(n,i):Wi(n,[],!0),z.makeTensor(n,t,i)}function si(n,t,e){var i=Rn(n,e);return ai(n,t,i,e)}var ic={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var Is=4;function N2(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l=this;return pe(this,function(u){switch(u.label){case 0:for(e=[],i=[],r=Array.isArray(n)?n.map(function(c){return c.name}):Object.keys(n),a=function(c){var h=r[c],d=Array.isArray(n)?n[c].tensor:n[h];if(d.dtype!=="float32"&&d.dtype!=="int32"&&d.dtype!=="bool"&&d.dtype!=="string"&&d.dtype!=="complex64")throw new Error("Unsupported dtype in weight '"+h+"': "+d.dtype);var p={name:h,shape:d.shape,dtype:d.dtype};if(d.dtype==="string"){var f=new Promise(function(m){return de(l,void 0,void 0,function(){var g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:return[4,d.bytes()];case 1:for(g=C.sent(),v=g.reduce(function(R,D){return R+D.length},0)+Is*g.length,b=new Uint8Array(v),w=0,S=0;S<g.length;S++)L=g[S],x=new Uint8Array(new Uint32Array([L.length]).buffer),b.set(x,w),w+=Is,b.set(L,w),w+=L.length;return m(b),[2]}})})});i.push(f)}else i.push(d.data());t!=null&&(p.group=t),e.push(p)},s=0;s<r.length;++s)a(s);return[4,Promise.all(i)];case 1:return o=u.sent(),[2,{data:T2(o),specs:e}]}})})}function nm(n,t){for(var e={},i,r=0,a=0,s=t;a<s.length;a++){var o=s[a],l=o.name,u=o.dtype,c=o.shape,h=ot(c),d=void 0;if("quantization"in o){var p=o.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error("Weight "+o.name+" with quantization "+p.dtype+" doesn't have corresponding metadata min and scale.")}else if(p.dtype==="float16"){if(u!=="float32")throw new Error("Weight "+o.name+" is quantized with "+p.dtype+" "+("which only supports weights of type float32 not "+u+"."))}else throw new Error("Weight "+o.name+" has unknown "+("quantization dtype "+p.dtype+". ")+"Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.");var f=ic[p.dtype],m=n.slice(r,r+h*f),g=p.dtype==="uint8"?new Uint8Array(m):new Uint16Array(m);if(u==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){d=new Float32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=b*p.scale+p.min}}else if(p.dtype==="float16")i===void 0&&(i=x2()),d=i(g);else throw new Error("Unsupported quantization type "+p.dtype+" for weight type float32.");else if(u==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error("Unsupported quantization type "+p.dtype+" for weight type int32.");d=new Int32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=Math.round(b*p.scale+p.min)}}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*f}else if(u==="string"){var w=ot(o.shape);d=[];for(var v=0;v<w;v++){var S=new Uint32Array(n.slice(r,r+Is))[0];r+=Is;var L=new Uint8Array(n.slice(r,r+S));d.push(L),r+=S}}else{var x=ic[u],m=n.slice(r,r+h*x);if(u==="float32")d=new Float32Array(m);else if(u==="int32")d=new Int32Array(m);else if(u==="bool")d=new Uint8Array(m);else if(u==="complex64"){d=new Float32Array(m);for(var C=new Float32Array(d.length/2),R=new Float32Array(d.length/2),v=0;v<C.length;v++)C[v]=d[v*2],R[v]=d[v*2+1];var D=si(C,c,"float32"),k=si(R,c,"float32");e[l]=ri(D,k)}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*x}u!=="complex64"&&(e[l]=si(d,c,u))}return e}function T2(n){if(n===null)throw new Error("Invalid input value: "+JSON.stringify(n));var t=0,e=[];n.forEach(function(a){if(t+=a.byteLength,e.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error("Unsupported TypedArray subtype: "+a.constructor.name)});var i=new Uint8Array(t),r=0;return e.forEach(function(a){i.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),i.buffer}var rc=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function im(n){return rc?Buffer.byteLength(n):new Blob([n]).size}function C2(n){if(rc)return Buffer.from(n).toString("base64");for(var t=new Uint8Array(n),e="",i=0,r=t.length;i<r;i++)e+=String.fromCharCode(t[i]);return btoa(e)}function R2(n){if(rc){var t=Buffer.from(n,"base64");return t.buffer.slice(t.byteOffset,t.byteOffset+t.byteLength)}for(var e=atob(n),i=new Uint8Array(e.length),r=0;r<e.length;++r)i.set([e.charCodeAt(r)],r);return i.buffer}function ac(n){if(n.length===1)return n[0];var t=0;n.forEach(function(r){t+=r.byteLength});var e=new Uint8Array(t),i=0;return n.forEach(function(r){e.set(new Uint8Array(r),i),i+=r.byteLength}),e.buffer}function rm(n){var t="/";for(n=n.trim();n.endsWith(t);)n=n.slice(0,n.length-1);var e=n.split(t);return e[e.length-1]}function wa(n){if(n.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:n.modelTopology==null?0:im(JSON.stringify(n.modelTopology)),weightSpecsBytes:n.weightSpecs==null?0:im(JSON.stringify(n.weightSpecs)),weightDataBytes:n.weightData==null?0:n.weightData.byteLength}}function O2(){var n=function(i){for(var r=i<<13,a=0;(r&8388608)===0;)a-=8388608,r<<=1;return r&=~8388608,a+=947912704,r|a},t=new Uint32Array(2048);t[0]=0;for(var e=1;e<1024;e++)t[e]=n(e);for(var e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function E2(){var n=new Uint32Array(64);n[0]=0,n[31]=1199570944,n[32]=2147483648,n[63]=3347054592;for(var t=1;t<31;t++)n[t]=t<<23;for(var t=33;t<63;t++)n[t]=2147483648+(t-32<<23);return n}function D2(){for(var n=new Uint32Array(64),t=0;t<64;t++)n[t]=1024;return n[0]=n[32]=0,n}function x2(){var n=O2(),t=E2(),e=D2();return function(i){for(var r=new ArrayBuffer(4*i.length),a=new Uint32Array(r),s=0;s<i.length;s++){var o=i[s],l=n[e[o>>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var Qt=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),k2=function(n){return Qt.registerSaveRouter(n)},F2=function(n){return Qt.registerLoadRouter(n)},W2=function(n){return Qt.getSaveHandlers(n)},U2=function(n,t){return Qt.getLoadHandlers(n,t)};var sc="tensorflowjs",oc=1,Ui="models_store",oi="model_info_store";function am(){if(!He().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function lc(n){var t=n.result;t.createObjectStore(Ui,{keyPath:"modelPath"}),t.createObjectStore(oi,{keyPath:"modelPath"})}var Ar=function(){function n(t){if(this.indexedDB=am(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(sc,oc);s.onupgradeneeded=function(){return lc(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(Ui,"readonly"),u=l.objectStore(Ui),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=wa(e),d=o.transaction(oi,"readwrite"),p=d.objectStore(oi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(Ui,"readwrite");var g=m.objectStore(Ui),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(oi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),sm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Ar.URL_SCHEME))?B2(n.slice(Ar.URL_SCHEME.length)):null};Qt.registerSaveRouter(sm);Qt.registerLoadRouter(sm);function B2(n){return new Ar(n)}function z2(n){return n.startsWith(Ar.URL_SCHEME)?n.slice(Ar.URL_SCHEME.length):n}var P2=function(){function n(){this.indexedDB=am()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(sc,oc);a.onupgradeneeded=function(){return lc(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(oi,"readonly"),l=o.objectStore(oi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h<d.length;h++){var p=d[h];c[p.modelPath]=p.modelArtifactsInfo}i(c)},u.onerror=function(c){return s.close(),r(u.error)},o.oncomplete=function(){return s.close()}},a.onerror=function(s){return r(a.error)}})]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e=this;return pe(this,function(i){return t=z2(t),[2,new Promise(function(r,a){var s=e.indexedDB.open(sc,oc);s.onupgradeneeded=function(){return lc(s)},s.onsuccess=function(){var o=s.result,l=o.transaction(oi,"readwrite"),u=l.objectStore(oi),c=u.get(t),h;c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+t+"' in IndexedDB."));var d=u.delete(t),p=function(){h=o.transaction(Ui,"readwrite");var f=h.objectStore(Ui),m=f.delete(t);m.onsuccess=function(){return r(c.result.modelArtifactsInfo)},m.onerror=function(g){return a(c.error)}};d.onsuccess=p,d.onerror=function(f){return p(),o.close(),a(c.error)}},c.onerror=function(d){return o.close(),a(c.error)},l.oncomplete=function(){h==null?o.close():h.oncomplete=function(){return o.close()}}},s.onerror=function(o){return a(s.error)}})]})})},n}();var Yn="/",Tr="tensorflowjs_models",om="info",_2="model_topology",M2="weight_specs",H2="weight_data",V2="model_metadata";function lm(n){return{info:[Tr,n,om].join(Yn),topology:[Tr,n,_2].join(Yn),weightSpecs:[Tr,n,M2].join(Yn),weightData:[Tr,n,H2].join(Yn),modelMetadata:[Tr,n,V2].join(Yn)}}function q2(n){var t=n.split(Yn);if(t.length<3)throw new Error("Invalid key format: "+n);return t.slice(1,t.length-1).join(Yn)}function G2(n){return n.startsWith(Nr.URL_SCHEME)?n.slice(Nr.URL_SCHEME.length):n}var Nr=function(){function n(t){if(!He().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=lm(this.modelPath)}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r;return pe(this,function(a){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");e=JSON.stringify(t.modelTopology),i=JSON.stringify(t.weightSpecs),r=wa(t);try{return this.LS.setItem(this.keys.info,JSON.stringify(r)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,i),this.LS.setItem(this.keys.weightData,C2(t.weightData)),this.LS.setItem(this.keys.modelMetadata,JSON.stringify({format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata})),[2,{modelArtifactsInfo:r}]}catch(s){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error("Failed to save model '"+this.modelPath+"' to local storage: size quota being exceeded is a possible cause of this failure: "+("modelTopologyBytes="+r.modelTopologyBytes+", ")+("weightSpecsBytes="+r.weightSpecsBytes+", ")+("weightDataBytes="+r.weightDataBytes+"."))}return[2]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o;return pe(this,function(l){if(t=JSON.parse(this.LS.getItem(this.keys.info)),t==null)throw new Error("In local storage, there is no model with name '"+this.modelPath+"'");if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");if(e={},i=JSON.parse(this.LS.getItem(this.keys.topology)),i==null)throw new Error("In local storage, the topology of model '"+this.modelPath+"' is missing.");if(e.modelTopology=i,r=JSON.parse(this.LS.getItem(this.keys.weightSpecs)),r==null)throw new Error("In local storage, the weight specs of model '"+this.modelPath+"' are missing.");if(e.weightSpecs=r,a=this.LS.getItem(this.keys.modelMetadata),a!=null&&(s=JSON.parse(a),e.format=s.format,e.generatedBy=s.generatedBy,e.convertedBy=s.convertedBy,e.userDefinedMetadata=s.userDefinedMetadata),o=this.LS.getItem(this.keys.weightData),o==null)throw new Error("In local storage, the binary weight values of model "+("'"+this.modelPath+"' are missing."));return e.weightData=R2(o),[2,e]})})},n.URL_SCHEME="localstorage://",n}(),um=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Nr.URL_SCHEME))?Y2(n.slice(Nr.URL_SCHEME.length)):null};Qt.registerSaveRouter(um);Qt.registerLoadRouter(um);function Y2(n){return new Nr(n)}var K2=function(){function n(){E(He().getBool("IS_BROWSER"),function(){return"Current environment is not a web browser"}),E(typeof window=="undefined"||typeof window.localStorage!="undefined",function(){return"Current browser does not appear to support localStorage"}),this.LS=window.localStorage}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s;return pe(this,function(o){for(t={},e=Tr+Yn,i=Yn+om,r=0;r<this.LS.length;++r)a=this.LS.key(r),a.startsWith(e)&&a.endsWith(i)&&(s=q2(a),t[s]=JSON.parse(this.LS.getItem(a)));return[2,t]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){if(t=G2(t),e=lm(t),this.LS.getItem(e.info)==null)throw new Error("Cannot find model at path '"+t+"'");return i=JSON.parse(this.LS.getItem(e.info)),this.LS.removeItem(e.info),this.LS.removeItem(e.topology),this.LS.removeItem(e.weightSpecs),this.LS.removeItem(e.weightData),[2,i]})})},n}();var xr="://",li=function(){function n(){this.managers={}}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerManager=function(t,e){E(t!=null,function(){return"scheme must not be undefined or null."}),t.endsWith(xr)&&(t=t.slice(0,t.indexOf(xr))),E(t.length>0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function As(n){if(n.indexOf(xr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+li.getSchemes().join(",")));return{scheme:n.split(xr)[0],path:n.split(xr)[1]}}function cm(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=Qt.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=Qt.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=As(n).scheme,l=As(n).path,u=o===As(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,li.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,li.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function j2(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=li.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e<i.length?(r=i[e],[4,li.getManager(r).listModels()]):[3,4];case 2:a=l.sent();for(s in a)o=r+xr+s,t[o]=a[s];l.label=3;case 3:return e++,[3,1];case 4:return[2,t]}})})}function $2(n){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){return t=As(n),e=li.getManager(t.scheme),[2,e.removeModel(t.path)]})})}function X2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!1,[2,cm(n,t,e)]})})}function J2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!0,[2,cm(n,t,e)]})})}var Z2=function(){function n(){}return n.prototype.fetch=function(t,e){return fetch(t,e)},n.prototype.now=function(){return performance.now()},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Browser's encoder only supports utf-8, but got "+e);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return new TextDecoder(e).decode(t)},n}();if(He().get("IS_BROWSER")){He().setPlatform("browser",new Z2);try{li.registerManager(Nr.URL_SCHEME,new K2)}catch(n){}try{li.registerManager(Ar.URL_SCHEME,new P2)}catch(n){}}var Q2={importFetch:function(){return Cp()}},uc,eA=function(){function n(){this.util=Rp(),this.textEncoder=new this.util.TextEncoder}return n.prototype.fetch=function(t,e){return He().global.fetch!=null?He().global.fetch(t,e):(uc==null&&(uc=Q2.importFetch()),uc(t,e))},n.prototype.now=function(){var t=process.hrtime();return t[0]*1e3+t[1]/1e6},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Node built-in encoder only supports utf-8, but got "+e);return this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)},n}();He().get("IS_NODE")&&He().setPlatform("node",new eA);function On(n,t,e){return t===void 0&&(t="float32"),t=t||"float32",$u(n),new Ss(n,t,e)}function tA(n,t){var e=O(n,"x","cast");if(!Pf(t))throw new Error("Failed to cast to unknown dtype "+t);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");var i={x:e},r={dtype:t};return z.runKernelFunc(function(a){return a.cast(e,t)},i,null,fs,r)}var ue=U({cast_:tA});function nA(n){var t=O(n,"x","clone",null),e=function(){return z.makeTensorFromDataId(t.dataId,t.shape,t.dtype)},i={x:t};return z.runKernelFunc(e,i,null,Pl)}var Bi=U({clone_:nA});function hm(n,t){t===void 0&&(t=!1),console.log(n.toString(t))}Jf();var iA={buffer:On,cast:ue,clone:Bi,print:hm};d2(iA);var rA="model",aA=".json",sA=".weights.bin";function dm(n){return new Promise(function(t){return setTimeout(t)}).then(n)}var cc=function(){function n(t){if(!He().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(n.URL_SCHEME)&&(t=t.slice(n.URL_SCHEME.length)),(t==null||t.length===0)&&(t=rA),this.modelTopologyFileName=t+aA,this.weightDataFileName=t+sA}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");if(e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"})),!(t.modelTopology instanceof ArrayBuffer))return[3,1];throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");case 1:return i=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:i},a=window.URL.createObjectURL(new Blob([JSON.stringify(r)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor,s.download=this.modelTopologyFileName,s.href=a,[4,dm(function(){return s.dispatchEvent(new MouseEvent("click"))})];case 2:return l.sent(),t.weightData!=null?(o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor,o.download=this.weightDataFileName,o.href=e,[4,dm(function(){return o.dispatchEvent(new MouseEvent("click"))})]):[3,4];case 3:l.sent(),l.label=4;case 4:return[2,{modelArtifactsInfo:wa(t)}]}})})},n.URL_SCHEME="downloads://",n}(),oA=function(){function n(t){if(t==null||t.length<1)throw new Error("When calling browserFiles, at least 1 file is required, "+("but received "+t));this.files=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i=this;return pe(this,function(r){return t=this.files[0],e=this.files.slice(1),[2,new Promise(function(a,s){var o=new FileReader;o.onload=function(l){var u=JSON.parse(l.target.result),c=u.modelTopology;if(c==null){s(new Error("modelTopology field is missing from file "+t.name));return}e.length===0&&a({modelTopology:c});var h=u.weightsManifest;if(h==null){s(new Error("weightManifest field is missing from file "+t.name));return}var d;try{d=i.checkManifestAndWeightFiles(h,e)}catch(g){s(g);return}var p=[],f=[],m=[];h.forEach(function(g){g.paths.forEach(function(v){f.push(v),m.push(null)}),p.push.apply(p,g.weights)}),h.forEach(function(g){g.paths.forEach(function(v){var b=new FileReader;b.onload=function(w){var S=w.target.result,L=f.indexOf(v);m[L]=S,m.indexOf(null)===-1&&a({modelTopology:c,weightSpecs:p,weightData:ac(m),format:u.format,generatedBy:u.generatedBy,convertedBy:u.convertedBy,userDefinedMetadata:u.userDefinedMetadata})},b.onerror=function(w){return s("Failed to weights data from file of path '"+v+"'.")},b.readAsArrayBuffer(d[v])})})},o.onerror=function(l){return s("Failed to read model topology and weights manifest JSON "+("from file '"+t.name+"'. BrowserFiles supports loading ")+"Keras-style tf.Model artifacts only.")},o.readAsText(t)})]})})},n.prototype.checkManifestAndWeightFiles=function(t,e){for(var i=[],r=e.map(function(u){return rm(u.name)}),a={},s=0,o=t;s<o.length;s++){var l=o[s];l.paths.forEach(function(u){var c=rm(u);if(i.indexOf(c)!==-1)throw new Error("Duplicate file basename found in weights manifest: "+("'"+c+"'"));if(i.push(c),r.indexOf(c)===-1)throw new Error("Weight file with basename '"+c+"' is not provided.");a[u]=e[r.indexOf(c)]})}if(i.length!==e.length)throw new Error("Mismatch in the number of files in weights manifest "+("("+i.length+") and the number of weight files provided ")+("("+e.length+")."));return a},n}(),uA=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(cc.URL_SCHEME))?lA(n.slice(cc.URL_SCHEME.length)):null};Qt.registerSaveRouter(uA);function lA(n){return n===void 0&&(n="model"),new cc(n)}function cA(n){return new oA(n)}function pm(n,t,e,i){s(n),e=e??0,i=i??1,o(e,i);var r=0,a=function(l){return l.then(function(u){var c=e+ ++r/n.length*(i-e);return t(c),u}),l};function s(l){E(l!=null&&Array.isArray(l)&&l.length>0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function fm(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?He().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,pm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,pm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function hA(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return fm(o,{requestInit:i})},a=mm(r),[2,a(n,t,e)]})})}function mm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=ic[S]*ot(w.shape),x=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(x(),o[R]=!0)}):x(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. 
-`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w<v;w++)b+=d[f+w].byteLength;for(var S=new ArrayBuffer(b),L=new Uint8Array(S),x=0,C=0;C<v;C++){var R=new Uint8Array(d[f+C]);L.set(R,x),x+=R.byteLength}var D=s[g];D.forEach(function(k){var W=S.slice(k.groupOffset,k.groupOffset+k.sizeBytes),F=nm(W,[k.manifestEntry]);for(var P in F)p[P]=F[P]}),f+=v}),[2,p]}})})}}var dA="application/octet-stream",pA="application/json",gm=function(){function n(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(E(typeof e.fetchFunc=="function",function(){return"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"}),this.fetch=e.fetchFunc):this.fetch=He().platform.fetch,E(t!=null&&t.length>0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:pA}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:dA}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:wa(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=fA(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;l<u.length;l++)c=u[l],o.push.apply(o,c.weights);for(h=[],d=[],p=0,f=t;p<f.length;p++)for(m=f[p],g=0,v=m.paths;g<v.length;g++)b=v[g],this.weightUrlConverter!=null?d.push(this.weightUrlConverter(b)):h.push(s+b+a);return this.weightUrlConverter?(S=(w=h.push).apply,L=[h],[4,Promise.all(d)]):[3,2];case 1:S.apply(w,L.concat([C.sent()])),C.label=2;case 2:return[4,fm(h,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress})];case 3:return x=C.sent(),[2,[o,ac(x)]]}})})},n.URL_SCHEME_REGEX=/^https?:\/\//,n}();function fA(n){var t=n.lastIndexOf("/"),e=n.lastIndexOf("?"),i=n.substring(0,t),r=e>t?n.substring(e):"";return[i+"/",r]}function hc(n){return n.match(gm.URL_SCHEME_REGEX)!=null}var vm=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return hc(i)}):e=hc(n),e?dc(n,t):null};Qt.registerSaveRouter(vm);Qt.registerLoadRouter(vm);function dc(n,t){return new gm(n,t)}function mA(n,t){return dc(n,t)}var pc=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),gA=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function vA(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new pc(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new pc({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new pc({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function yA(n){return new gA(n)}var bA={__proto__:null,browserFiles:cA,browserHTTPRequest:mA,concatenateArrayBuffers:ac,decodeWeights:nm,encodeWeights:N2,fromMemory:vA,getLoadHandlers:U2,getModelArtifactsInfoForJSON:wa,getSaveHandlers:W2,http:dc,isHTTPScheme:hc,loadWeights:hA,registerLoadRouter:F2,registerSaveRouter:k2,weightsLoaderFactory:mm,withSaveHandler:yA,copyModel:X2,listModels:j2,moveModel:J2,removeModel:$2};function wA(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=Wf(t,e.size),E(e.size===ot(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,uu,r)}var V=U({reshape_:wA});function SA(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=at(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=ot(h),f=ot(d);E(ln(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(x,C){return C([g,v]),x.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,ml,S);return V(L,m)}var We=U({matMul_:SA});function LA(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,iu,l)}var Ts=U({oneHot_:LA});function IA(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a<e.rank,function(){return"All entries in 'perm' must be between 0 and "+(e.rank-1)+(" but got "+t)})}),e.rank<=1)return e.clone();var i={x:e},r={perm:t};return z.runKernelFunc(function(a){return a.transpose(e,t)},i,null,Fu,r)}var ut=U({transpose_:IA});function AA(n,t,e){var i=O(n,"labels","confusionMatrix"),r=O(t,"predictions","confusionMatrix");E(e==null||e>0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=Ts(ue(i,"int32"),e),s=Ts(ue(r,"int32"),e),o=ut(a);return ue(We(o,s),"int32")}var TA=U({confusionMatrix_:AA});var NA={__proto__:null,confusionMatrix:TA};function ym(n,t,e){if(Fi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=Rn(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}var Cr;function xA(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState<o)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}var l=qu(Pu,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(Pu,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Cr==null&&(Cr=document.createElement("canvas").getContext("2d")),Cr.canvas.width=d,Cr.canvas.height=p,Cr.drawImage(n,0,0,d,p),f=Cr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v<g;v++)for(var b=0;b<t;++b)m[v*t+b]=f[v*4+b]}var w=[p,d,t];return ym(m,w,"int32")}function CA(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v;return pe(this,function(b){switch(b.label){case 0:if(e=O(n,"img","toPixels"),n instanceof Y||(i=e,e=ue(i,"int32"),i.dispose()),e.rank!==2&&e.rank!==3)throw new Error("toPixels only supports rank 2 or 3 tensors, got rank "+e.rank+".");if(r=e.shape.slice(0,2),a=r[0],s=r[1],o=e.rank===2?1:e.shape[2],o>4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h<a*s;++h){for(d=[0,0,0,255],p=0;p<o;p++){if(f=l[h*o+p],e.dtype==="float32"){if(f<0||f>1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var RA=U({fromPixels_:xA}),OA={__proto__:null,toPixels:CA,fromPixels:RA};function bm(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a<e.length-1;++a)r*=e[a];var s=n.shape,o=e.slice();o.pop();for(var l=1,a=i;a<n.rank;++a)l*=s[a],o.push(s[a]);var u=wr(n.shape).map(function(c){return c/l}).concat([1]).slice(0,i);return[o,r,l,u]}var EA={__proto__:null,prepareAndValidate:bm};function fc(n,t,e){var i=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank<r)throw new Error(a+(" update.rank < "+r+". "));if(n.length<i+(e.rank-r))throw new Error(a+(" Output shape length < "+(i+(e.rank-r))));if(e.rank!==r+n.length-i)throw new Error(a+(" update.rank != "+(r+n.length-i)));for(var s=0;s<r;++s)if(e.shape[s]!==t.shape[s])throw new Error(a+(" updates.shape["+s+"] ("+e.shape[s]+") != indices.shape["+s+"] ("+t.shape[s]+")."));for(var s=0;s<e.rank-r;++s)if(e.shape[s+r]!==n[s+i])throw new Error(a+(" updates.shape["+(s+r)+"] ("+e.shape[s+r]+") != shape["+(s+r)+"] ("+n[s+r]+")"))}function mc(n,t,e){if(t.rank<1)throw new Error("tf.scatterND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(n.rank<1)throw new Error("tf.scatterND() expects the updates to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.dtype!=="int32")throw new Error("The dtype of 'indices' should be int32, but got dtype: "+t.dtype);if(e.length<1)throw new Error("Output rank must be greater or equal to 1, but got shape: "+e);if(e.length===0){if(t.size===0)throw new Error("Indices specified for empty output. indices shape: "+t.shape);if(n.size===0)throw new Error("Updates specified for empty output. updates shape: "+n.shape)}fc(e,t,n)}function wm(n,t,e){for(var i=t.shape.length,r=i>1?t.shape[i-1]:1,a=e.length,s=1,o=r;o<a;++o)s*=e[o];var l=r<1?1:r,u=ot(t.shape)/l,c=wr(e.slice(0,r)).concat([1]),h=ot(e);return{sliceRank:r,numUpdates:u,sliceSize:s,strides:c,outputSize:h}}var DA={__proto__:null,validateUpdateShape:fc,validateInput:mc,calculateShapes:wm};function Sm(n,t,e){var i=n.shape.length;E(i===t.length,function(){return"Error in slice"+i+"D: Length of begin "+t+" must "+("match the rank of the array ("+i+").")}),E(i===e.length,function(){return"Error in slice"+i+"D: Length of size "+e+" must "+("match the rank of the array ("+i+").")});for(var r=function(s){E(t[s]+e[s]<=n.shape[s],function(){return"Error in slice"+i+"D: begin["+s+"] + size["+s+"] "+("("+(t[s]+e[s])+") would overflow input.shape["+s+"] ("+n.shape[s]+")")})},a=0;a<i;++a)r(a)}function Ns(n){for(var t=[],e=0;n>0;)n&1&&t.push(e),n/=2,e++;return t}function Lm(n,t,e){for(var i=[],r=0;r<n.length;r++)i[r]=Math.ceil((t[r]-n[r])/e[r]);return i}function Im(n,t,e,i){for(var r=n.slice(),a=r.length;a<i.length;a++)r.push(1);for(var a=0;a<e;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function Am(n,t,e){return e<=n?e:e-(t-1)}function Tm(n,t){for(var e=[],i=0;i<n;i++)e.push(t+i);return e}function Em(n,t,e,i,r,a,s,o,l){var u=n.length,c=new Array(u),h=new Array(u),d=new Array(u);if(t.length&&e>0){var p=t[0],f=e+1;c=Nm(s,p,f,i,n),h=xm(o,p,f,r,n),d=Im(a,p,f,n)}else for(var m=0;m<u;m++)c[m]=Rm(s,i,a,n,m,l),h[m]=Om(o,r,a,n,m,l),d[m]=Cm(a,m,l);return{begin:c,end:h,strides:d}}function Nm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=0;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=0),a[o]=u}return a}function xm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=Number.MAX_SAFE_INTEGER),a[o]=u}for(var c=0;c<a.length;c++){var h=r[c];a[c]<0&&(a[c]+=h),a[c]=pa(0,a[c],r[c])}return a}function Cm(n,t,e){var i=n[t];return(e&1<<t||i==null)&&(i=1),i}function Rm(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=pa(0,s,l-1),s}function Om(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=pa(0,s,l):s=pa(-1,s,l-1),s}function kA(n,t,e){for(var i=e.length,r=0;r<e.length;r++)if(e[r]>1){i=r;break}for(var r=i+1;r<e.length;r++)if(t[r]>0||e[r]!==n[r])return!1;return!0}function FA(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i<n.length-1;i++)e+=n[i]*t[i];return e}function gc(n,t,e){var i,r=n.shape.length;typeof t=="number"?i=[t].concat(new Array(r-1).fill(0)):t.length<r?i=t.concat(new Array(r-t.length).fill(0)):i=t.slice(),i.forEach(function(s){E(s!==-1,function(){return"slice() does not support negative begin indexing."})});var a;return e==null?a=new Array(r).fill(-1):typeof e=="number"?a=[e].concat(new Array(r-1).fill(-1)):e.length<r?a=e.concat(new Array(r-e.length).fill(-1)):a=e,a=a.map(function(s,o){return s>=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var Dm={__proto__:null,assertParamsValid:Sm,maskToAxes:Ns,computeOutShape:Lm,stridesWithElidedDims:Im,getNormalizedAxes:Em,startIndicesWithElidedDims:Nm,stopIndicesWithElidedDims:xm,stridesForAxis:Cm,startForAxis:Rm,stopForAxis:Om,isSliceContinous:kA,computeFlatOffset:FA,parseSliceParams:gc};var km=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),Fm=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function ui(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),Fm.register(n)}var WA={__proto__:null,Serializable:km,SerializationMap:Fm,registerClass:ui};var UA=.001,Wm=.1;function BA(n,t,e){return e==null&&(e=vc()),yc(n,t,function(i,r){return bc(i,r,e)})}function vc(){return z.backend.floatPrecision()===32?UA:Wm}function yc(n,t,e){var i=!0;if((Ot(n)||Ot(t))&&(i=!1),Ot(n)&&Ot(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=Rn(n),o=Rn(t);if(!ln(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Ot(n)?n:Wi(n),u=Ot(t)?t:Wi(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`.
+`;return v[v.length-1]=" "+v[v.length-1]+"]"+(a?"":x),v}function va(n){for(var t=[],e=0;e<n.length;e+=2)t.push([n[e],n[e+1]]);return t}var Ss=function(){function n(t,e,i){var r=this;if(this.dtype=e,this.shape=t.slice(),this.size=ot(t),i!=null){var a=i.length;E(a===this.size,function(){return"Length of values '"+a+"' does not match the size "+("inferred by the shape '"+r.size+"'.")})}if(e==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=i||Bf(e,this.size),this.strides=wr(t)}return n.prototype.set=function(t){for(var e=this,i=[],r=1;r<arguments.length;r++)i[r-1]=arguments[r];i.length===0&&(i=[0]),E(i.length===this.rank,function(){return"The number of provided coordinates ("+i.length+") must "+("match the rank ("+e.rank+")")});var a=this.locToIndex(i);this.values[a]=t},n.prototype.get=function(){for(var t=[],e=0;e<arguments.length;e++)t[e]=arguments[e];t.length===0&&(t=[0]);for(var i=0,r=0,a=t;r<a.length;r++){var s=a[r];if(s<0||s>=this.shape[i]){var o="Requested out of range element at "+t+". "+("  Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u<t.length-1;++u)l+=this.strides[u]*t[u];return this.values[l]},n.prototype.locToIndex=function(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];for(var e=t[t.length-1],i=0;i<t.length-1;++i)e+=this.strides[i]*t[i];return e},n.prototype.indexToLoc=function(t){if(this.rank===0)return[];if(this.rank===1)return[t];for(var e=new Array(this.shape.length),i=0;i<e.length-1;++i)e[i]=Math.floor(t/this.strides[i]),t-=e[i]*this.strides[i];return e[e.length-1]=t,e},Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.toTensor=function(){return Cn().makeTensor(this.values,this.shape,this.dtype)},n}(),Cn=null,Ir=null;function h2(n){Cn=n}function d2(n){Ir=n}var Y=function(){function n(t,e,i,r){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=ot(t),this.strides=wr(t),this.dataId=i,this.id=r,this.rankType=this.rank<5?this.rank.toString():"higher"}return Object.defineProperty(n.prototype,"rank",{get:function(){return this.shape.length},enumerable:!0,configurable:!0}),n.prototype.buffer=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Ir.buffer(this.shape,this.dtype,t)]}})})},n.prototype.bufferSync=function(){return Ir.buffer(this.shape,this.dtype,this.dataSync())},n.prototype.array=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return[4,this.data()];case 1:return t=e.sent(),[2,Sr(this.shape,t)]}})})},n.prototype.arraySync=function(){return Sr(this.shape,this.dataSync())},n.prototype.data=function(){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){switch(i.label){case 0:return this.throwIfDisposed(),t=Cn().read(this.dataId),this.dtype==="string"?[4,t]:[3,2];case 1:e=i.sent();try{return[2,e.map(function(r){return Ju(r)})]}catch(r){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}i.label=2;case 2:return[2,t]}})})},n.prototype.dataSync=function(){this.throwIfDisposed();var t=Cn().readSync(this.dataId);if(this.dtype==="string")try{return t.map(function(e){return Ju(e)})}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t},n.prototype.bytes=function(){return de(this,void 0,void 0,function(){var t;return pe(this,function(e){switch(e.label){case 0:return this.throwIfDisposed(),[4,Cn().read(this.dataId)];case 1:return t=e.sent(),this.dtype==="string"?[2,t]:[2,new Uint8Array(t.buffer)]}})})},n.prototype.dispose=function(){if(this.isDisposed)return;Cn().disposeTensor(this),this.isDisposedInternal=!0},Object.defineProperty(n.prototype,"isDisposed",{get:function(){return this.isDisposedInternal},enumerable:!0,configurable:!0}),n.prototype.throwIfDisposed=function(){if(this.isDisposed)throw new Error("Tensor is disposed.")},n.prototype.print=function(t){return t===void 0&&(t=!1),Ir.print(this,t)},n.prototype.clone=function(){return this.throwIfDisposed(),Ir.clone(this)},n.prototype.toString=function(t){t===void 0&&(t=!1);var e=this.dataSync();return c2(e,this.shape,this.dtype,t)},n.prototype.cast=function(t){return this.throwIfDisposed(),Ir.cast(this,t)},n.prototype.variable=function(t,e,i){return t===void 0&&(t=!0),this.throwIfDisposed(),Cn().makeVariable(this,t,e,i)},n}();Object.defineProperty(Y,Symbol.hasInstance,{value:function(n){return!!n&&n.data!=null&&n.dataSync!=null&&n.throwIfDisposed!=null}});var ya=function(n){qn(t,n);function t(e,i,r,a){var s=n.call(this,e.shape,e.dtype,e.dataId,a)||this;return s.trainable=i,s.name=r,s}return t.prototype.assign=function(e){if(e.dtype!==this.dtype)throw new Error("dtype of the new value ("+e.dtype+") and "+("previous value ("+this.dtype+") must match"));if(!on(e.shape,this.shape))throw new Error("shape of the new value ("+e.shape+") and "+("previous value ("+this.shape+") must match"));Cn().disposeTensor(this),this.dataId=e.dataId,Cn().incRef(this,null)},t.prototype.dispose=function(){Cn().disposeVariable(this),this.isDisposedInternal=!0},t}(Y);Object.defineProperty(ya,Symbol.hasInstance,{value:function(n){return n instanceof Y&&n.assign!=null&&n.assign instanceof Function}});(function(n){n.R0="R0",n.R1="R1",n.R2="R2",n.R3="R3",n.R4="R4",n.R5="R5",n.R6="R6"})(I.Rank||(I.Rank={}));var Qu;(function(n){n.float32="float32",n.int32="int32",n.bool="int32",n.complex64="complex64"})(Qu||(Qu={}));var ec;(function(n){n.float32="float32",n.int32="int32",n.bool="bool",n.complex64="complex64"})(ec||(ec={}));var tc;(function(n){n.float32="float32",n.int32="float32",n.bool="float32",n.complex64="complex64"})(tc||(tc={}));var nc;(function(n){n.float32="complex64",n.int32="complex64",n.bool="complex64",n.complex64="complex64"})(nc||(nc={}));var p2={float32:tc,int32:Qu,bool:ec,complex64:nc};function Ls(n,t){if(n==="string"||t==="string"){if(n==="string"&&t==="string")return"string";throw new Error("Can not upcast "+n+" with "+t)}return p2[n][t]}function f2(n){return Ls(n,"int32")}function at(n,t){if(n.dtype===t.dtype)return[n,t];var e=Ls(n.dtype,t.dtype);return[n.cast(e),t.cast(e)]}function jf(n,t){E(n.dtype===t.dtype,function(){return"The dtypes of the first("+n.dtype+") and"+(" second("+t.dtype+") input must match")})}function m2(n,t){return t.some(function(e){return e.id===n.id})}function ic(n){var t=[],e=new Set;return $f(n,t,e),t}function $f(n,t,e){if(n==null)return;if(n instanceof Y){t.push(n);return}if(!g2(n))return;var i=n;for(var r in i){var a=i[r];e.has(a)||(e.add(a),$f(a,t,e))}}function g2(n){return Array.isArray(n)||typeof n=="object"}var v2={__proto__:null,makeTypesMatch:at,assertTypesMatch:jf,isTensorInList:m2,getTensorsInContainer:ic};var Xf=function(){function n(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null}}return n.prototype.dispose=function(){for(var t in this.registeredVariables)this.registeredVariables[t].dispose()},n}(),w2=function(){function n(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Xf}return n.prototype.ready=function(){return de(this,void 0,void 0,function(){var t,e,i,r;return pe(this,function(a){switch(a.label){case 0:if(this.pendingBackendInit!=null)return[2,this.pendingBackendInit.then(function(){})];if(this.backendInstance!=null)return[2];t=this.getSortedBackends(),e=0,a.label=1;case 1:return e<t.length?(i=t[e],[4,this.initializeBackend(i).success]):[3,5];case 2:return r=a.sent(),r?[4,this.setBackend(i)]:[3,4];case 3:return a.sent(),[2];case 4:return e++,[3,1];case 5:throw new Error("Could not initialize any backends, all backend initializations failed.")}})})},Object.defineProperty(n.prototype,"backend",{get:function(){if(this.pendingBackendInit!=null)throw new Error("Backend '"+this.backendName+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");if(this.backendInstance==null){var t=this.initializeBackendsAndReturnBest(),e=t.name,i=t.asyncInit;if(i)throw new Error("The highest priority backend '"+e+"' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods");this.setBackend(e)}return this.backendInstance},enumerable:!0,configurable:!0}),n.prototype.backendNames=function(){return Object.keys(this.registryFactory)},n.prototype.findBackend=function(t){if(!(t in this.registry))if(t in this.registryFactory){var e=this.initializeBackend(t).asyncInit;if(e)return null}else return null;return this.registry[t]},n.prototype.findBackendFactory=function(t){return t in this.registryFactory?this.registryFactory[t].factory:null},n.prototype.registerBackend=function(t,e,i){return i===void 0&&(i=1),t in this.registryFactory?(console.warn(t+" backend was already registered. Reusing existing backend factory."),!1):(this.registryFactory[t]={factory:e,priority:i},!0)},n.prototype.setBackend=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s;return pe(this,function(o){switch(o.label){case 0:if(this.registryFactory[t]==null)throw new Error("Backend name '"+t+"' not found in registry");return this.backendName=t,this.registry[t]==null?(this.backendInstance=null,e=this.initializeBackend(t),i=e.success,r=e.asyncInit,r?[4,i]:[3,2]):[3,4];case 1:return s=o.sent(),[3,3];case 2:s=i,o.label=3;case 3:if(a=s,!a)return[2,!1];o.label=4;case 4:return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new s2(this.backendInstance),[2,!0]}})})},n.prototype.setupRegisteredKernels=function(){var t=this,e=ms(this.backendName);e.forEach(function(i){i.setupFunc!=null&&i.setupFunc(t.backendInstance)})},n.prototype.disposeRegisteredKernels=function(t){var e=this,i=ms(t);i.forEach(function(r){r.disposeFunc!=null&&r.disposeFunc(e.registry[t])})},n.prototype.initializeBackend=function(t){var e=this,i=this.registryFactory[t];if(i==null)throw new Error("Cannot initialize backend "+t+", no registration found.");try{var r=i.factory();if(r&&!(r instanceof Op)&&typeof r.then=="function"){var a=++this.pendingBackendInitId,s=r.then(function(o){return a<e.pendingBackendInitId?!1:(e.registry[t]=o,e.pendingBackendInit=null,!0)}).catch(function(o){return a<e.pendingBackendInitId||(e.pendingBackendInit=null,console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message)),!1});return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=r,{success:!0,asyncInit:!1}}catch(o){return console.warn("Initialization of backend "+t+" failed"),console.warn(o.stack||o.message),{success:!1,asyncInit:!1}}},n.prototype.removeBackend=function(t){if(!(t in this.registryFactory))throw new Error(t+" backend not found in registry");this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)},n.prototype.getSortedBackends=function(){var t=this;if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(function(e,i){return t.registryFactory[i].priority-t.registryFactory[e].priority})},n.prototype.initializeBackendsAndReturnBest=function(){for(var t=this.getSortedBackends(),e=0;e<t.length;e++){var i=t[e],r=this.initializeBackend(i),a=r.success,s=r.asyncInit;if(s||a)return{name:i,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")},n.prototype.moveData=function(t,e){var i=this.state.tensorInfo.get(e),r=i.backend,a=this.readSync(e);r.disposeData(e),i.backend=t,t.move(e,a,i.shape,i.dtype),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++},n.prototype.tidy=function(t,e){var i=this,r=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");r=t}var a;return this.scopedRun(function(){return i.startScope(r)},function(){return i.endScope(a)},function(){return a=e(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a})},n.prototype.scopedRun=function(t,e,i){t();try{var r=i();return e(),r}catch(a){throw e(),a}},n.prototype.nextTensorId=function(){return n.nextTensorId++},n.prototype.nextVariableId=function(){return n.nextVariableId++},n.prototype.clone=function(t){var e=this.makeTensorFromDataId(t.dataId,t.shape,t.dtype),i={x:t},r=function(s){return{x:function(){var o="float32",l={x:s},u={dtype:o};return z.runKernelFunc(function(c){return c.cast(s,o)},l,null,fs,u)}}},a=[];return this.addTapeNode(this.state.activeScope.name,i,[e],r,a,{}),e},n.prototype.runKernel=function(t,e,i,r,a){var s=null,o=null;return this.runKernelFunc(s,e,o,t,i,r,a)},n.prototype.shouldCheckForMemLeaks=function(){return this.ENV.getBool("IS_TEST")},n.prototype.checkKernelForMemLeak=function(t,e,i){var r=this.backend.numDataIds(),a=0;i.forEach(function(l){a+=l.dtype==="complex64"?3:1});var s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=r-e-a-s;if(o>0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=Gu(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var x=l.getTensorsForGradient(r,e,L);if(x==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});x=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(x)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=Yu(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&ni(t[0])&&(a=t.map(function(c){return Ku(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Hf(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new ya(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Mf(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof ya||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a<s.length?(o=s[a],l=o,[4,o.kernelTimeMs]):[3,6];case 3:return l.kernelTimeMs=c.sent(),u=o,[4,o.extraInfo];case 4:u.extraInfo=c.sent(),c.label=5;case 5:return a++,[3,2];case 6:return[2,this.state.activeProfile]}})})},n.prototype.isTapeOn=function(){return this.state.gradientDepth>0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=Yu(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=Lr(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=ic(t),r=new Set(i.map(function(l){return l.id})),a=0;a<this.state.activeScope.track.length;a++){var s=this.state.activeScope.track[a];!s.kept&&!r.has(s.id)&&s.dispose()}var o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],i.forEach(function(l){!l.kept&&l.scopeId===o.id&&e.track(l)})},n.prototype.gradients=function(t,e,i,r){var a=this;if(r===void 0&&(r=!1),E(e.length>0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=o2(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??y2(s.shape),l2(l,o,function(c){return a.tidy(c)},b2);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h<d.length;h++){var p=d[h];p.dispose()}}),a.state.activeTape=null),{value:s,grads:u}})},n.prototype.customGrad=function(t){var e=this;return E(ii(t),function(){return"The f passed in customGrad(f) must be a function."}),function(){for(var i=[],r=0;r<arguments.length;r++)i[r]=arguments[r];E(i.every(function(o){return o instanceof Y}),function(){return"The args passed in customGrad(f)(x1, x2,...) must all be tensors"});var a,s={};return i.forEach(function(o,l){s[l]=o}),e.runKernelFunc(function(o,l){return a=t.apply(void 0,i.concat([l])),E(a.value instanceof Y,function(){return"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"}),E(ii(a.gradFunc),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."}),a.value},s,function(o,l){var u=a.gradFunc(o,l),c=Array.isArray(u)?u:[u];E(c.length===i.length,function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."}),E(c.every(function(d){return d instanceof Y}),function(){return"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."});var h={};return c.forEach(function(d,p){h[p]=function(){return d}}),h})}},n.prototype.readSync=function(t){var e=this.state.tensorInfo.get(t);return e.backend.readSync(t)},n.prototype.read=function(t){var e=this.state.tensorInfo.get(t);return e.backend.read(t)},n.prototype.time=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){switch(r.label){case 0:return e=$u(),[4,this.backend.time(t)];case 1:return i=r.sent(),i.wallMs=$u()-e,[2,i]}})})},n.prototype.track=function(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t},Object.defineProperty(n.prototype,"registeredVariables",{get:function(){return this.state.registeredVariables},enumerable:!0,configurable:!0}),n.prototype.reset=function(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Xf;for(var t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null},n.nextTensorId=0,n.nextVariableId=0,n}();function y2(n){var t=ju(ot(n),"float32");return z.makeTensor(t,n,"float32")}function Jf(){var n=kp();if(n._tfengine==null){var t=new Dp(n);n._tfengine=new w2(t)}return WI(n._tfengine.ENV),h2(function(){return n._tfengine}),n._tfengine}var z=Jf();function b2(n,t){var e={a:n,b:t};return z.runKernelFunc(function(i,r){var a=i.add(n,t);return r([n,t]),a},e,null,ps)}function S2(){return typeof navigator!="undefined"&&navigator!=null}function L2(){if(S2()){var n=navigator.userAgent||navigator.vendor||window.opera;return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(n)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(n.substr(0,4))}return!1}function Zf(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var I2={__proto__:null,isMobile:L2,isBrowser:Zf};var Gn=He();Gn.registerFlag("DEBUG",function(){return!1},function(n){n&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});Gn.registerFlag("IS_BROWSER",function(){return Zf()});Gn.registerFlag("IS_NODE",function(){return typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined"});Gn.registerFlag("IS_CHROME",function(){return typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor)});Gn.registerFlag("PROD",function(){return!1});Gn.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",function(){return Gn.getBool("DEBUG")});Gn.registerFlag("DEPRECATION_WARNINGS_ENABLED",function(){return!0});Gn.registerFlag("IS_TEST",function(){return!1});function Rn(n,t){var e=n;if(Ot(n))return t==="string"?[]:[n.length];if(!Array.isArray(n))return[];for(var i=[];Array.isArray(e)||Ot(e)&&t!=="string";)i.push(e.length),e=e[0];return Array.isArray(n)&&He().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Qf(n,i,[]),i}function Qf(n,t,e){if(e=e||[],!Array.isArray(n)&&!Ot(n)){E(t.length===0,function(){return"Element arr["+e.join("][")+"] is a primitive, "+("but should be an array/TypedArray of "+t[0]+" elements")});return}E(t.length>0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r<n.length;++r)Qf(n[r],i,e.concat(r))}function em(n,t,e,i){if(n==null)return;if(n!=="numeric"&&n!==t||n==="numeric"&&t==="string")throw new Error("Argument '"+e+"' passed to '"+i+"' must "+("be "+n+" tensor, but got "+t+" tensor"))}function O(n,t,e,i){if(i===void 0&&(i="numeric"),n instanceof Y)return em(i,n.dtype,t,e),n;var r=vs(n);if(r!=="string"&&["bool","int32","float32"].indexOf(i)>=0&&(r=i),em(i,r,t,e),n==null||!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=Rn(n,r);!Ot(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?bs(n,r):Wi(n,[],o);return z.makeTensor(l,s,r)}function ba(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var tm="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+tm;var r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];z.startScope(e);try{var o=i.apply(void 0,a);return o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),z.endScope(o),o}catch(l){throw z.endScope(null),l}};return Object.defineProperty(r,"name",{value:e,configurable:!0}),r}function A2(n,t){var e=O(n,"real","complex"),i=O(t,"imag","complex");Be(e.shape,i.shape,"real and imag shapes, "+e.shape+" and "+i.shape+", must match in call to tf.complex().");var r=function(s){return s.complex(e,i)},a={real:e,imag:i};return z.runKernelFunc(r,a,null,Pp)}var ri=U({complex_:A2});function ai(n,t,e,i){if(i==null&&(i=vs(n)),i==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!Ot(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){Xu(t);var r=ot(t),a=ot(e);E(r===a,function(){return"Based on the provided shape, ["+t+"], the tensor should have "+(r+" values but has "+a)});for(var s=0;s<e.length;++s){var o=e[s],l=s===e.length-1?o!==ot(t.slice(s)):!0;E(e[s]===t[s]||!l,function(){return"Error creating a new Tensor. Inferred shape "+("("+e+") does not match the provided ")+("shape ("+t+"). ")})}}return!Ot(n)&&!Array.isArray(n)&&(n=[n]),t=t||e,n=i!=="string"?bs(n,i):Wi(n,[],!0),z.makeTensor(n,t,i)}function si(n,t,e){var i=Rn(n,e);return ai(n,t,i,e)}var rc={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var Is=4;function N2(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l=this;return pe(this,function(u){switch(u.label){case 0:for(e=[],i=[],r=Array.isArray(n)?n.map(function(c){return c.name}):Object.keys(n),a=function(c){var h=r[c],d=Array.isArray(n)?n[c].tensor:n[h];if(d.dtype!=="float32"&&d.dtype!=="int32"&&d.dtype!=="bool"&&d.dtype!=="string"&&d.dtype!=="complex64")throw new Error("Unsupported dtype in weight '"+h+"': "+d.dtype);var p={name:h,shape:d.shape,dtype:d.dtype};if(d.dtype==="string"){var f=new Promise(function(m){return de(l,void 0,void 0,function(){var g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:return[4,d.bytes()];case 1:for(g=C.sent(),v=g.reduce(function(R,D){return R+D.length},0)+Is*g.length,b=new Uint8Array(v),w=0,S=0;S<g.length;S++)L=g[S],x=new Uint8Array(new Uint32Array([L.length]).buffer),b.set(x,w),w+=Is,b.set(L,w),w+=L.length;return m(b),[2]}})})});i.push(f)}else i.push(d.data());t!=null&&(p.group=t),e.push(p)},s=0;s<r.length;++s)a(s);return[4,Promise.all(i)];case 1:return o=u.sent(),[2,{data:T2(o),specs:e}]}})})}function nm(n,t){for(var e={},i,r=0,a=0,s=t;a<s.length;a++){var o=s[a],l=o.name,u=o.dtype,c=o.shape,h=ot(c),d=void 0;if("quantization"in o){var p=o.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error("Weight "+o.name+" with quantization "+p.dtype+" doesn't have corresponding metadata min and scale.")}else if(p.dtype==="float16"){if(u!=="float32")throw new Error("Weight "+o.name+" is quantized with "+p.dtype+" "+("which only supports weights of type float32 not "+u+"."))}else throw new Error("Weight "+o.name+" has unknown "+("quantization dtype "+p.dtype+". ")+"Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.");var f=rc[p.dtype],m=n.slice(r,r+h*f),g=p.dtype==="uint8"?new Uint8Array(m):new Uint16Array(m);if(u==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){d=new Float32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=b*p.scale+p.min}}else if(p.dtype==="float16")i===void 0&&(i=x2()),d=i(g);else throw new Error("Unsupported quantization type "+p.dtype+" for weight type float32.");else if(u==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error("Unsupported quantization type "+p.dtype+" for weight type int32.");d=new Int32Array(g.length);for(var v=0;v<g.length;v++){var b=g[v];d[v]=Math.round(b*p.scale+p.min)}}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*f}else if(u==="string"){var w=ot(o.shape);d=[];for(var v=0;v<w;v++){var S=new Uint32Array(n.slice(r,r+Is))[0];r+=Is;var L=new Uint8Array(n.slice(r,r+S));d.push(L),r+=S}}else{var x=rc[u],m=n.slice(r,r+h*x);if(u==="float32")d=new Float32Array(m);else if(u==="int32")d=new Int32Array(m);else if(u==="bool")d=new Uint8Array(m);else if(u==="complex64"){d=new Float32Array(m);for(var C=new Float32Array(d.length/2),R=new Float32Array(d.length/2),v=0;v<C.length;v++)C[v]=d[v*2],R[v]=d[v*2+1];var D=si(C,c,"float32"),k=si(R,c,"float32");e[l]=ri(D,k)}else throw new Error("Unsupported dtype in weight '"+l+"': "+u);r+=h*x}u!=="complex64"&&(e[l]=si(d,c,u))}return e}function T2(n){if(n===null)throw new Error("Invalid input value: "+JSON.stringify(n));var t=0,e=[];n.forEach(function(a){if(t+=a.byteLength,e.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error("Unsupported TypedArray subtype: "+a.constructor.name)});var i=new Uint8Array(t),r=0;return e.forEach(function(a){i.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),i.buffer}var ac=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function im(n){return ac?Buffer.byteLength(n):new Blob([n]).size}function C2(n){if(ac)return Buffer.from(n).toString("base64");for(var t=new Uint8Array(n),e="",i=0,r=t.length;i<r;i++)e+=String.fromCharCode(t[i]);return btoa(e)}function R2(n){if(ac){var t=Buffer.from(n,"base64");return t.buffer.slice(t.byteOffset,t.byteOffset+t.byteLength)}for(var e=atob(n),i=new Uint8Array(e.length),r=0;r<e.length;++r)i.set([e.charCodeAt(r)],r);return i.buffer}function sc(n){if(n.length===1)return n[0];var t=0;n.forEach(function(r){t+=r.byteLength});var e=new Uint8Array(t),i=0;return n.forEach(function(r){e.set(new Uint8Array(r),i),i+=r.byteLength}),e.buffer}function rm(n){var t="/";for(n=n.trim();n.endsWith(t);)n=n.slice(0,n.length-1);var e=n.split(t);return e[e.length-1]}function wa(n){if(n.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:n.modelTopology==null?0:im(JSON.stringify(n.modelTopology)),weightSpecsBytes:n.weightSpecs==null?0:im(JSON.stringify(n.weightSpecs)),weightDataBytes:n.weightData==null?0:n.weightData.byteLength}}function O2(){var n=function(i){for(var r=i<<13,a=0;(r&8388608)===0;)a-=8388608,r<<=1;return r&=~8388608,a+=947912704,r|a},t=new Uint32Array(2048);t[0]=0;for(var e=1;e<1024;e++)t[e]=n(e);for(var e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function E2(){var n=new Uint32Array(64);n[0]=0,n[31]=1199570944,n[32]=2147483648,n[63]=3347054592;for(var t=1;t<31;t++)n[t]=t<<23;for(var t=33;t<63;t++)n[t]=2147483648+(t-32<<23);return n}function D2(){for(var n=new Uint32Array(64),t=0;t<64;t++)n[t]=1024;return n[0]=n[32]=0,n}function x2(){var n=O2(),t=E2(),e=D2();return function(i){for(var r=new ArrayBuffer(4*i.length),a=new Uint32Array(r),s=0;s<i.length;s++){var o=i[s],l=n[e[o>>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var Zt=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),k2=function(n){return Zt.registerSaveRouter(n)},F2=function(n){return Zt.registerLoadRouter(n)},W2=function(n){return Zt.getSaveHandlers(n)},U2=function(n,t){return Zt.getLoadHandlers(n,t)};var oc="tensorflowjs",lc=1,Ui="models_store",oi="model_info_store";function am(){if(!He().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function uc(n){var t=n.result;t.createObjectStore(Ui,{keyPath:"modelPath"}),t.createObjectStore(oi,{keyPath:"modelPath"})}var Ar=function(){function n(t){if(this.indexedDB=am(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(oc,lc);s.onupgradeneeded=function(){return uc(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(Ui,"readonly"),u=l.objectStore(Ui),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=wa(e),d=o.transaction(oi,"readwrite"),p=d.objectStore(oi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(Ui,"readwrite");var g=m.objectStore(Ui),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(oi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),sm=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Ar.URL_SCHEME))?B2(n.slice(Ar.URL_SCHEME.length)):null};Zt.registerSaveRouter(sm);Zt.registerLoadRouter(sm);function B2(n){return new Ar(n)}function z2(n){return n.startsWith(Ar.URL_SCHEME)?n.slice(Ar.URL_SCHEME.length):n}var P2=function(){function n(){this.indexedDB=am()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(oc,lc);a.onupgradeneeded=function(){return uc(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(oi,"readonly"),l=o.objectStore(oi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h<d.length;h++){var p=d[h];c[p.modelPath]=p.modelArtifactsInfo}i(c)},u.onerror=function(c){return s.close(),r(u.error)},o.oncomplete=function(){return s.close()}},a.onerror=function(s){return r(a.error)}})]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e=this;return pe(this,function(i){return t=z2(t),[2,new Promise(function(r,a){var s=e.indexedDB.open(oc,lc);s.onupgradeneeded=function(){return uc(s)},s.onsuccess=function(){var o=s.result,l=o.transaction(oi,"readwrite"),u=l.objectStore(oi),c=u.get(t),h;c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+t+"' in IndexedDB."));var d=u.delete(t),p=function(){h=o.transaction(Ui,"readwrite");var f=h.objectStore(Ui),m=f.delete(t);m.onsuccess=function(){return r(c.result.modelArtifactsInfo)},m.onerror=function(g){return a(c.error)}};d.onsuccess=p,d.onerror=function(f){return p(),o.close(),a(c.error)}},c.onerror=function(d){return o.close(),a(c.error)},l.oncomplete=function(){h==null?o.close():h.oncomplete=function(){return o.close()}}},s.onerror=function(o){return a(s.error)}})]})})},n}();var Yn="/",Tr="tensorflowjs_models",om="info",_2="model_topology",M2="weight_specs",H2="weight_data",V2="model_metadata";function lm(n){return{info:[Tr,n,om].join(Yn),topology:[Tr,n,_2].join(Yn),weightSpecs:[Tr,n,M2].join(Yn),weightData:[Tr,n,H2].join(Yn),modelMetadata:[Tr,n,V2].join(Yn)}}function q2(n){var t=n.split(Yn);if(t.length<3)throw new Error("Invalid key format: "+n);return t.slice(1,t.length-1).join(Yn)}function G2(n){return n.startsWith(Nr.URL_SCHEME)?n.slice(Nr.URL_SCHEME.length):n}var Nr=function(){function n(t){if(!He().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=lm(this.modelPath)}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r;return pe(this,function(a){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");e=JSON.stringify(t.modelTopology),i=JSON.stringify(t.weightSpecs),r=wa(t);try{return this.LS.setItem(this.keys.info,JSON.stringify(r)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,i),this.LS.setItem(this.keys.weightData,C2(t.weightData)),this.LS.setItem(this.keys.modelMetadata,JSON.stringify({format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata})),[2,{modelArtifactsInfo:r}]}catch(s){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error("Failed to save model '"+this.modelPath+"' to local storage: size quota being exceeded is a possible cause of this failure: "+("modelTopologyBytes="+r.modelTopologyBytes+", ")+("weightSpecsBytes="+r.weightSpecsBytes+", ")+("weightDataBytes="+r.weightDataBytes+"."))}return[2]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o;return pe(this,function(l){if(t=JSON.parse(this.LS.getItem(this.keys.info)),t==null)throw new Error("In local storage, there is no model with name '"+this.modelPath+"'");if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");if(e={},i=JSON.parse(this.LS.getItem(this.keys.topology)),i==null)throw new Error("In local storage, the topology of model '"+this.modelPath+"' is missing.");if(e.modelTopology=i,r=JSON.parse(this.LS.getItem(this.keys.weightSpecs)),r==null)throw new Error("In local storage, the weight specs of model '"+this.modelPath+"' are missing.");if(e.weightSpecs=r,a=this.LS.getItem(this.keys.modelMetadata),a!=null&&(s=JSON.parse(a),e.format=s.format,e.generatedBy=s.generatedBy,e.convertedBy=s.convertedBy,e.userDefinedMetadata=s.userDefinedMetadata),o=this.LS.getItem(this.keys.weightData),o==null)throw new Error("In local storage, the binary weight values of model "+("'"+this.modelPath+"' are missing."));return e.weightData=R2(o),[2,e]})})},n.URL_SCHEME="localstorage://",n}(),um=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Nr.URL_SCHEME))?Y2(n.slice(Nr.URL_SCHEME.length)):null};Zt.registerSaveRouter(um);Zt.registerLoadRouter(um);function Y2(n){return new Nr(n)}var K2=function(){function n(){E(He().getBool("IS_BROWSER"),function(){return"Current environment is not a web browser"}),E(typeof window=="undefined"||typeof window.localStorage!="undefined",function(){return"Current browser does not appear to support localStorage"}),this.LS=window.localStorage}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s;return pe(this,function(o){for(t={},e=Tr+Yn,i=Yn+om,r=0;r<this.LS.length;++r)a=this.LS.key(r),a.startsWith(e)&&a.endsWith(i)&&(s=q2(a),t[s]=JSON.parse(this.LS.getItem(a)));return[2,t]})})},n.prototype.removeModel=function(t){return de(this,void 0,void 0,function(){var e,i;return pe(this,function(r){if(t=G2(t),e=lm(t),this.LS.getItem(e.info)==null)throw new Error("Cannot find model at path '"+t+"'");return i=JSON.parse(this.LS.getItem(e.info)),this.LS.removeItem(e.info),this.LS.removeItem(e.topology),this.LS.removeItem(e.weightSpecs),this.LS.removeItem(e.weightData),[2,i]})})},n}();var xr="://",li=function(){function n(){this.managers={}}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerManager=function(t,e){E(t!=null,function(){return"scheme must not be undefined or null."}),t.endsWith(xr)&&(t=t.slice(0,t.indexOf(xr))),E(t.length>0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function As(n){if(n.indexOf(xr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+li.getSchemes().join(",")));return{scheme:n.split(xr)[0],path:n.split(xr)[1]}}function cm(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=Zt.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=Zt.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=As(n).scheme,l=As(n).path,u=o===As(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,li.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,li.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function j2(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=li.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e<i.length?(r=i[e],[4,li.getManager(r).listModels()]):[3,4];case 2:a=l.sent();for(s in a)o=r+xr+s,t[o]=a[s];l.label=3;case 3:return e++,[3,1];case 4:return[2,t]}})})}function $2(n){return de(this,void 0,void 0,function(){var t,e;return pe(this,function(i){return t=As(n),e=li.getManager(t.scheme),[2,e.removeModel(t.path)]})})}function X2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!1,[2,cm(n,t,e)]})})}function J2(n,t){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){return e=!0,[2,cm(n,t,e)]})})}var Z2=function(){function n(){}return n.prototype.fetch=function(t,e){return fetch(t,e)},n.prototype.now=function(){return performance.now()},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Browser's encoder only supports utf-8, but got "+e);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return new TextDecoder(e).decode(t)},n}();if(He().get("IS_BROWSER")){He().setPlatform("browser",new Z2);try{li.registerManager(Nr.URL_SCHEME,new K2)}catch(n){}try{li.registerManager(Ar.URL_SCHEME,new P2)}catch(n){}}var Q2={importFetch:function(){return Cp()}},cc,eA=function(){function n(){this.util=Rp(),this.textEncoder=new this.util.TextEncoder}return n.prototype.fetch=function(t,e){return He().global.fetch!=null?He().global.fetch(t,e):(cc==null&&(cc=Q2.importFetch()),cc(t,e))},n.prototype.now=function(){var t=process.hrtime();return t[0]*1e3+t[1]/1e6},n.prototype.encode=function(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error("Node built-in encoder only supports utf-8, but got "+e);return this.textEncoder.encode(t)},n.prototype.decode=function(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)},n}();He().get("IS_NODE")&&He().setPlatform("node",new eA);function On(n,t,e){return t===void 0&&(t="float32"),t=t||"float32",Xu(n),new Ss(n,t,e)}function tA(n,t){var e=O(n,"x","cast");if(!Pf(t))throw new Error("Failed to cast to unknown dtype "+t);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");var i={x:e},r={dtype:t};return z.runKernelFunc(function(a){return a.cast(e,t)},i,null,fs,r)}var ue=U({cast_:tA});function nA(n){var t=O(n,"x","clone",null),e=function(){return z.makeTensorFromDataId(t.dataId,t.shape,t.dtype)},i={x:t};return z.runKernelFunc(e,i,null,_l)}var Bi=U({clone_:nA});function hm(n,t){t===void 0&&(t=!1),console.log(n.toString(t))}Jf();var iA={buffer:On,cast:ue,clone:Bi,print:hm};d2(iA);var rA="model",aA=".json",sA=".weights.bin";function dm(n){return new Promise(function(t){return setTimeout(t)}).then(n)}var hc=function(){function n(t){if(!He().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(n.URL_SCHEME)&&(t=t.slice(n.URL_SCHEME.length)),(t==null||t.length===0)&&(t=rA),this.modelTopologyFileName=t+aA,this.weightDataFileName=t+sA}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");if(e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"})),!(t.modelTopology instanceof ArrayBuffer))return[3,1];throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");case 1:return i=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:i},a=window.URL.createObjectURL(new Blob([JSON.stringify(r)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor,s.download=this.modelTopologyFileName,s.href=a,[4,dm(function(){return s.dispatchEvent(new MouseEvent("click"))})];case 2:return l.sent(),t.weightData!=null?(o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor,o.download=this.weightDataFileName,o.href=e,[4,dm(function(){return o.dispatchEvent(new MouseEvent("click"))})]):[3,4];case 3:l.sent(),l.label=4;case 4:return[2,{modelArtifactsInfo:wa(t)}]}})})},n.URL_SCHEME="downloads://",n}(),oA=function(){function n(t){if(t==null||t.length<1)throw new Error("When calling browserFiles, at least 1 file is required, "+("but received "+t));this.files=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i=this;return pe(this,function(r){return t=this.files[0],e=this.files.slice(1),[2,new Promise(function(a,s){var o=new FileReader;o.onload=function(l){var u=JSON.parse(l.target.result),c=u.modelTopology;if(c==null){s(new Error("modelTopology field is missing from file "+t.name));return}e.length===0&&a({modelTopology:c});var h=u.weightsManifest;if(h==null){s(new Error("weightManifest field is missing from file "+t.name));return}var d;try{d=i.checkManifestAndWeightFiles(h,e)}catch(g){s(g);return}var p=[],f=[],m=[];h.forEach(function(g){g.paths.forEach(function(v){f.push(v),m.push(null)}),p.push.apply(p,g.weights)}),h.forEach(function(g){g.paths.forEach(function(v){var b=new FileReader;b.onload=function(w){var S=w.target.result,L=f.indexOf(v);m[L]=S,m.indexOf(null)===-1&&a({modelTopology:c,weightSpecs:p,weightData:sc(m),format:u.format,generatedBy:u.generatedBy,convertedBy:u.convertedBy,userDefinedMetadata:u.userDefinedMetadata})},b.onerror=function(w){return s("Failed to weights data from file of path '"+v+"'.")},b.readAsArrayBuffer(d[v])})})},o.onerror=function(l){return s("Failed to read model topology and weights manifest JSON "+("from file '"+t.name+"'. BrowserFiles supports loading ")+"Keras-style tf.Model artifacts only.")},o.readAsText(t)})]})})},n.prototype.checkManifestAndWeightFiles=function(t,e){for(var i=[],r=e.map(function(u){return rm(u.name)}),a={},s=0,o=t;s<o.length;s++){var l=o[s];l.paths.forEach(function(u){var c=rm(u);if(i.indexOf(c)!==-1)throw new Error("Duplicate file basename found in weights manifest: "+("'"+c+"'"));if(i.push(c),r.indexOf(c)===-1)throw new Error("Weight file with basename '"+c+"' is not provided.");a[u]=e[r.indexOf(c)]})}if(i.length!==e.length)throw new Error("Mismatch in the number of files in weights manifest "+("("+i.length+") and the number of weight files provided ")+("("+e.length+")."));return a},n}(),uA=function(n){return He().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(hc.URL_SCHEME))?lA(n.slice(hc.URL_SCHEME.length)):null};Zt.registerSaveRouter(uA);function lA(n){return n===void 0&&(n="model"),new hc(n)}function cA(n){return new oA(n)}function pm(n,t,e,i){s(n),e=e??0,i=i??1,o(e,i);var r=0,a=function(l){return l.then(function(u){var c=e+ ++r/n.length*(i-e);return t(c),u}),l};function s(l){E(l!=null&&Array.isArray(l)&&l.length>0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function fm(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?He().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,pm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,pm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function hA(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return fm(o,{requestInit:i})},a=mm(r),[2,a(n,t,e)]})})}function mm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=rc[S]*ot(w.shape),x=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(x(),o[R]=!0)}):x(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. 
+`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w<v;w++)b+=d[f+w].byteLength;for(var S=new ArrayBuffer(b),L=new Uint8Array(S),x=0,C=0;C<v;C++){var R=new Uint8Array(d[f+C]);L.set(R,x),x+=R.byteLength}var D=s[g];D.forEach(function(k){var W=S.slice(k.groupOffset,k.groupOffset+k.sizeBytes),F=nm(W,[k.manifestEntry]);for(var P in F)p[P]=F[P]}),f+=v}),[2,p]}})})}}var dA="application/octet-stream",pA="application/json",gm=function(){function n(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(E(typeof e.fetchFunc=="function",function(){return"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"}),this.fetch=e.fetchFunc):this.fetch=He().platform.fetch,E(t!=null&&t.length>0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:pA}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:dA}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:wa(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=fA(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;l<u.length;l++)c=u[l],o.push.apply(o,c.weights);for(h=[],d=[],p=0,f=t;p<f.length;p++)for(m=f[p],g=0,v=m.paths;g<v.length;g++)b=v[g],this.weightUrlConverter!=null?d.push(this.weightUrlConverter(b)):h.push(s+b+a);return this.weightUrlConverter?(S=(w=h.push).apply,L=[h],[4,Promise.all(d)]):[3,2];case 1:S.apply(w,L.concat([C.sent()])),C.label=2;case 2:return[4,fm(h,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress})];case 3:return x=C.sent(),[2,[o,sc(x)]]}})})},n.URL_SCHEME_REGEX=/^https?:\/\//,n}();function fA(n){var t=n.lastIndexOf("/"),e=n.lastIndexOf("?"),i=n.substring(0,t),r=e>t?n.substring(e):"";return[i+"/",r]}function dc(n){return n.match(gm.URL_SCHEME_REGEX)!=null}var vm=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return dc(i)}):e=dc(n),e?pc(n,t):null};Zt.registerSaveRouter(vm);Zt.registerLoadRouter(vm);function pc(n,t){return new gm(n,t)}function mA(n,t){return pc(n,t)}var fc=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),gA=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function vA(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new fc(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new fc({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new fc({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function yA(n){return new gA(n)}var bA={__proto__:null,browserFiles:cA,browserHTTPRequest:mA,concatenateArrayBuffers:sc,decodeWeights:nm,encodeWeights:N2,fromMemory:vA,getLoadHandlers:U2,getModelArtifactsInfoForJSON:wa,getSaveHandlers:W2,http:pc,isHTTPScheme:dc,loadWeights:hA,registerLoadRouter:F2,registerSaveRouter:k2,weightsLoaderFactory:mm,withSaveHandler:yA,copyModel:X2,listModels:j2,moveModel:J2,removeModel:$2};function wA(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=Wf(t,e.size),E(e.size===ot(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,cu,r)}var V=U({reshape_:wA});function SA(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=at(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=ot(h),f=ot(d);E(on(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(x,C){return C([g,v]),x.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,gl,S);return V(L,m)}var We=U({matMul_:SA});function LA(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,ru,l)}var Ts=U({oneHot_:LA});function IA(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a<e.rank,function(){return"All entries in 'perm' must be between 0 and "+(e.rank-1)+(" but got "+t)})}),e.rank<=1)return e.clone();var i={x:e},r={perm:t};return z.runKernelFunc(function(a){return a.transpose(e,t)},i,null,Wu,r)}var ut=U({transpose_:IA});function AA(n,t,e){var i=O(n,"labels","confusionMatrix"),r=O(t,"predictions","confusionMatrix");E(e==null||e>0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=Ts(ue(i,"int32"),e),s=Ts(ue(r,"int32"),e),o=ut(a);return ue(We(o,s),"int32")}var TA=U({confusionMatrix_:AA});var NA={__proto__:null,confusionMatrix:TA};function ym(n,t,e){if(Fi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=Rn(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}var Cr;function xA(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState<o)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}var l=Gu(_u,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(_u,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Cr==null&&(Cr=document.createElement("canvas").getContext("2d")),Cr.canvas.width=d,Cr.canvas.height=p,Cr.drawImage(n,0,0,d,p),f=Cr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v<g;v++)for(var b=0;b<t;++b)m[v*t+b]=f[v*4+b]}var w=[p,d,t];return ym(m,w,"int32")}function CA(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v;return pe(this,function(b){switch(b.label){case 0:if(e=O(n,"img","toPixels"),n instanceof Y||(i=e,e=ue(i,"int32"),i.dispose()),e.rank!==2&&e.rank!==3)throw new Error("toPixels only supports rank 2 or 3 tensors, got rank "+e.rank+".");if(r=e.shape.slice(0,2),a=r[0],s=r[1],o=e.rank===2?1:e.shape[2],o>4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h<a*s;++h){for(d=[0,0,0,255],p=0;p<o;p++){if(f=l[h*o+p],e.dtype==="float32"){if(f<0||f>1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var RA=U({fromPixels_:xA}),OA={__proto__:null,toPixels:CA,fromPixels:RA};function bm(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a<e.length-1;++a)r*=e[a];var s=n.shape,o=e.slice();o.pop();for(var l=1,a=i;a<n.rank;++a)l*=s[a],o.push(s[a]);var u=wr(n.shape).map(function(c){return c/l}).concat([1]).slice(0,i);return[o,r,l,u]}var EA={__proto__:null,prepareAndValidate:bm};function mc(n,t,e){var i=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank<r)throw new Error(a+(" update.rank < "+r+". "));if(n.length<i+(e.rank-r))throw new Error(a+(" Output shape length < "+(i+(e.rank-r))));if(e.rank!==r+n.length-i)throw new Error(a+(" update.rank != "+(r+n.length-i)));for(var s=0;s<r;++s)if(e.shape[s]!==t.shape[s])throw new Error(a+(" updates.shape["+s+"] ("+e.shape[s]+") != indices.shape["+s+"] ("+t.shape[s]+")."));for(var s=0;s<e.rank-r;++s)if(e.shape[s+r]!==n[s+i])throw new Error(a+(" updates.shape["+(s+r)+"] ("+e.shape[s+r]+") != shape["+(s+r)+"] ("+n[s+r]+")"))}function gc(n,t,e){if(t.rank<1)throw new Error("tf.scatterND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(n.rank<1)throw new Error("tf.scatterND() expects the updates to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.dtype!=="int32")throw new Error("The dtype of 'indices' should be int32, but got dtype: "+t.dtype);if(e.length<1)throw new Error("Output rank must be greater or equal to 1, but got shape: "+e);if(e.length===0){if(t.size===0)throw new Error("Indices specified for empty output. indices shape: "+t.shape);if(n.size===0)throw new Error("Updates specified for empty output. updates shape: "+n.shape)}mc(e,t,n)}function wm(n,t,e){for(var i=t.shape.length,r=i>1?t.shape[i-1]:1,a=e.length,s=1,o=r;o<a;++o)s*=e[o];var l=r<1?1:r,u=ot(t.shape)/l,c=wr(e.slice(0,r)).concat([1]),h=ot(e);return{sliceRank:r,numUpdates:u,sliceSize:s,strides:c,outputSize:h}}var DA={__proto__:null,validateUpdateShape:mc,validateInput:gc,calculateShapes:wm};function Sm(n,t,e){var i=n.shape.length;E(i===t.length,function(){return"Error in slice"+i+"D: Length of begin "+t+" must "+("match the rank of the array ("+i+").")}),E(i===e.length,function(){return"Error in slice"+i+"D: Length of size "+e+" must "+("match the rank of the array ("+i+").")});for(var r=function(s){E(t[s]+e[s]<=n.shape[s],function(){return"Error in slice"+i+"D: begin["+s+"] + size["+s+"] "+("("+(t[s]+e[s])+") would overflow input.shape["+s+"] ("+n.shape[s]+")")})},a=0;a<i;++a)r(a)}function Ns(n){for(var t=[],e=0;n>0;)n&1&&t.push(e),n/=2,e++;return t}function Lm(n,t,e){for(var i=[],r=0;r<n.length;r++)i[r]=Math.ceil((t[r]-n[r])/e[r]);return i}function Im(n,t,e,i){for(var r=n.slice(),a=r.length;a<i.length;a++)r.push(1);for(var a=0;a<e;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function Am(n,t,e){return e<=n?e:e-(t-1)}function Tm(n,t){for(var e=[],i=0;i<n;i++)e.push(t+i);return e}function Em(n,t,e,i,r,a,s,o,l){var u=n.length,c=new Array(u),h=new Array(u),d=new Array(u);if(t.length&&e>0){var p=t[0],f=e+1;c=Nm(s,p,f,i,n),h=xm(o,p,f,r,n),d=Im(a,p,f,n)}else for(var m=0;m<u;m++)c[m]=Rm(s,i,a,n,m,l),h[m]=Om(o,r,a,n,m,l),d[m]=Cm(a,m,l);return{begin:c,end:h,strides:d}}function Nm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=0;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=0),a[o]=u}return a}function xm(n,t,e,i,r){for(var a=r.slice(),s=Tm(e,t),o=0;o<a.length;o++)if(s.indexOf(o)>-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=Am(t,e,o),u=i[l];n&1<<l&&(u=Number.MAX_SAFE_INTEGER),a[o]=u}for(var c=0;c<a.length;c++){var h=r[c];a[c]<0&&(a[c]+=h),a[c]=pa(0,a[c],r[c])}return a}function Cm(n,t,e){var i=n[t];return(e&1<<t||i==null)&&(i=1),i}function Rm(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=pa(0,s,l-1),s}function Om(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<<r||a&1<<r||s==null)&&(o>0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=pa(0,s,l):s=pa(-1,s,l-1),s}function kA(n,t,e){for(var i=e.length,r=0;r<e.length;r++)if(e[r]>1){i=r;break}for(var r=i+1;r<e.length;r++)if(t[r]>0||e[r]!==n[r])return!1;return!0}function FA(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i<n.length-1;i++)e+=n[i]*t[i];return e}function vc(n,t,e){var i,r=n.shape.length;typeof t=="number"?i=[t].concat(new Array(r-1).fill(0)):t.length<r?i=t.concat(new Array(r-t.length).fill(0)):i=t.slice(),i.forEach(function(s){E(s!==-1,function(){return"slice() does not support negative begin indexing."})});var a;return e==null?a=new Array(r).fill(-1):typeof e=="number"?a=[e].concat(new Array(r-1).fill(-1)):e.length<r?a=e.concat(new Array(r-e.length).fill(-1)):a=e,a=a.map(function(s,o){return s>=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var Dm={__proto__:null,assertParamsValid:Sm,maskToAxes:Ns,computeOutShape:Lm,stridesWithElidedDims:Im,getNormalizedAxes:Em,startIndicesWithElidedDims:Nm,stopIndicesWithElidedDims:xm,stridesForAxis:Cm,startForAxis:Rm,stopForAxis:Om,isSliceContinous:kA,computeFlatOffset:FA,parseSliceParams:vc};var km=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),Fm=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function ui(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),Fm.register(n)}var WA={__proto__:null,Serializable:km,SerializationMap:Fm,registerClass:ui};var UA=.001,Wm=.1;function BA(n,t,e){return e==null&&(e=yc()),bc(n,t,function(i,r){return wc(i,r,e)})}function yc(){return z.backend.floatPrecision()===32?UA:Wm}function bc(n,t,e){var i=!0;if((Ot(n)||Ot(t))&&(i=!1),Ot(n)&&Ot(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=Rn(n),o=Rn(t);if(!on(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Ot(n)?n:Wi(n),u=Ot(t)?t:Wi(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`.
 `)+("Actual:   "+l+`.
 `)+("Expected: "+u+"."));for(var c=0;c<u.length;++c){var h=l[c],d=u[c];if(!e(h,d))throw new Error("Arrays differ: actual["+c+"] = "+h+", expected["+c+"] = "+d+`.
 `+("Actual:   "+l+`.
-`)+("Expected: "+u+"."))}}function zA(n,t){n().then(function(){return t.fail()},function(){return t()})}function PA(n,t){var e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return ni(n)||ni(n[0])||ni(t)||ni(t[0])?yc(n,e,function(i,r){return i==r}):yc(n,t,function(i,r){return bc(i,r,0)})}function _A(n,t,e){if(e==null&&(e=vc()),!bc(n,t,e))throw new Error("Numbers differ: actual === "+n+", expected === "+t)}function bc(n,t,e){return!isFinite(n)&&!isFinite(t)?!0:!(isNaN(n)||isNaN(t)||Math.abs(n-t)>e)}function MA(n,t,e){for(var i=0;i<n.length;i++)if(n[i]<t||n[i]>e)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function HA(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var VA={__proto__:null,TEST_EPSILON_FLOAT16:Wm,expectArraysClose:BA,testEpsilon:vc,expectPromiseToFail:zA,expectArraysEqual:PA,expectNumbersClose:_A,expectValuesInRange:MA,expectArrayBuffersEqual:HA};var qA="2.6.0";function GA(){He().set("PROD",!0)}function YA(){He().set("DEBUG",!0)}function KA(){He().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function It(n){He().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function jA(){z.disposeVariables()}function $A(){return z}function XA(){return z.memory()}function JA(n){return z.profile(n)}function ft(n,t){return z.tidy(n,t)}function zt(n){var t=nc(n);t.forEach(function(e){return e.dispose()})}function Um(n){return z.keep(n)}function ZA(n){return z.time(n)}function QA(n){return z.setBackend(n)}function eT(){return z.ready()}function tT(){return z.backendName}function nT(n){z.removeBackend(n)}function iT(n){return z.findBackend(n)}function rT(n){return z.findBackendFactory(n)}function aT(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function sT(){return z.backend}function oT(n,t){He().setPlatform(n,t)}function lT(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,ps)}var fe=U({add_:lT});function uT(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Wl)}var wc=U({floorDiv_:uT});function cT(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=at(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return wc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Rl,o)}var Ie=U({div_:cT});function hT(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,eu)}var J=U({mul_:hT});function dT(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,nl)}var Gt=U({abs_:dT});function pT(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,il)}var Bm=U({acos_:pT});function fT(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,rl)}var zm=U({acosh_:fT});function mT(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!ln(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,al)}var gT=U({addN_:mT});function Sc(n,t){for(var e=0;e<n.length;++e)if(n[n.length-e-1]!==t-1-e)return!1;return!0}function Pm(n,t,e){for(var i=n.length+t.length,r=[],a=0,s=0,o=0;o<i;o++)e.indexOf(o)===-1?r.push(n[a++]):r.push(t[s++]);return r}function _m(n,t){for(var e=[],i=n.length,r=0;r<i;r++)t.indexOf(r)===-1&&e.push(n[r]);var a=t.map(function(s){return n[s]});return[e,a]}function en(n,t){var e=t.map(function(i){return 1});return Pm(n,e,t)}function vT(n,t,e){E(Sc(t,e),function(){return n+" supports only inner-most axes for now. "+("Got axes "+t+" and rank-"+e+" input.")})}function tn(n,t){if(Sc(n,t))return null;for(var e=[],i=0;i<t;++i)n.indexOf(i)===-1&&e.push(i);return n.forEach(function(r){return e.push(r)}),e}function xs(n){return n.map(function(t,e){return[e,t]}).sort(function(t,e){return t[1]-e[1]}).map(function(t){return t[0]})}function En(n,t){for(var e=[],i=t-n;i<t;++i)e.push(i);return e}function yT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","all","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=tn(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.all(i,u);if(e){var d=en(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Wp,s)}var Mm=U({all_:yT});function bT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","any","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=tn(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.any(i,u);if(e){var d=en(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Up,s)}var Hm=U({any_:bT});function wT(n,t){t===void 0&&(t=0);var e=O(n,"x","argMax"),i=function(s,o){o([e]);var l=Qe(t,e.shape),u=tn(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMax(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,sl,a)}var Vm=U({argMax_:wT});function ST(n,t){t===void 0&&(t=0);var e=O(n,"x","argMin"),i=function(s,o){o([e]),t==null&&(t=0);var l=Qe(t,e.shape),u=tn(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMin(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,ol,a)}var qm=U({argMin_:ST});function LT(n){var t=O(n,"x","asin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asin(t);return r([t]),a},e,null,ll)}var Gm=U({asin_:LT});function IT(n){var t=O(n,"x","asinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asinh(t);return r([t]),a},e,null,ul)}var Ym=U({asinh_:IT});function AT(n){var t=O(n,"x","atan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atan(t);return r([t]),a},e,null,cl)}var Km=U({atan_:AT});function TT(n,t){var e,i=O(n,"a","atan2"),r=O(t,"b","atan2");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.atan2(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,dl)}var jm=U({atan2_:TT});function NT(n){var t=O(n,"x","atanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atanh(t);return r([t]),a},e,null,hl)}var $m=U({atanh_:NT});function xT(n,t,e,i,r,a){r===void 0&&(r="NHWC");var s=n[3],o=t.concat([s]),l=Sa(r);return Dn(n,o,e,a,i,null,null,l)}function Rr(n,t,e,i,r,a,s){s===void 0&&(s="channelsLast");var o=Cs(t),l=o[0],u=o[1],c;if(s==="channelsLast")c=[l,u,n[3],n[3]];else if(s==="channelsFirst")c=[l,u,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Dn(n,c,e,i,r,a,!1,s)}function Ia(n,t,e,i,r,a,s){s===void 0&&(s="NDHWC");var o=Lc(t),l=o[0],u=o[1],c=o[2],h,d;if(s==="NDHWC")d="channelsLast",h=[l,u,c,n[4],n[4]];else if(s==="NCDHW")d="channelsFirst",h=[l,u,c,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return La(n,h,e,i,r,!1,d,a)}function Dn(n,t,e,i,r,a,s,o){s===void 0&&(s=!1),o===void 0&&(o="channelsLast");var l=[-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3];if(o==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3];else if(o==="channelsFirst")u=n[0],d=n[1],c=n[2],h=n[3];else throw new Error("Unknown dataFormat "+o);var p=t[0],f=t[1],m=t[3],g=Cs(e),v=g[0],b=g[1],w=Cs(i),S=w[0],L=w[1],x=Or(p,S),C=Or(f,L),R=CT(r,c,h,v,b,x,C,a,o),D=R.padInfo,k=R.outHeight,W=R.outWidth,F=s?m*d:m,P;return o==="channelsFirst"?P=[u,F,k,W]:o==="channelsLast"&&(P=[u,k,W,F]),{batchSize:u,dataFormat:o,inHeight:c,inWidth:h,inChannels:d,outHeight:k,outWidth:W,outChannels:F,padInfo:D,strideHeight:v,strideWidth:b,filterHeight:p,filterWidth:f,effectiveFilterHeight:x,effectiveFilterWidth:C,dilationHeight:S,dilationWidth:L,inShape:n,outShape:P,filterShape:t}}function La(n,t,e,i,r,a,s,o){a===void 0&&(a=!1),s===void 0&&(s="channelsLast");var l=[-1,-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3],p=l[4];if(s==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3],p=n[4];else if(s==="channelsFirst")u=n[0],p=n[1],c=n[2],h=n[3],d=n[4];else throw new Error("Unknown dataFormat "+s);var f=t[0],m=t[1],g=t[2],v=t[4],b=Lc(e),w=b[0],S=b[1],L=b[2],x=Lc(i),C=x[0],R=x[1],D=x[2],k=Or(f,C),W=Or(m,R),F=Or(g,D),P=RT(r,c,h,d,w,S,L,k,W,F,o),H=P.padInfo,_=P.outDepth,K=P.outHeight,j=P.outWidth,q=a?v*p:v,G;return s==="channelsFirst"?G=[u,q,_,K,j]:s==="channelsLast"&&(G=[u,_,K,j,q]),{batchSize:u,dataFormat:s,inDepth:c,inHeight:h,inWidth:d,inChannels:p,outDepth:_,outHeight:K,outWidth:j,outChannels:q,padInfo:H,strideDepth:w,strideHeight:S,strideWidth:L,filterDepth:f,filterHeight:m,filterWidth:g,effectiveFilterDepth:k,effectiveFilterHeight:W,effectiveFilterWidth:F,dilationDepth:C,dilationHeight:R,dilationWidth:D,inShape:n,outShape:G,filterShape:t}}function OT(n,t,e,i,r){i==null&&(i=Ic(n,t,e));var a=n[0],s=n[1],o=zi((a-t+2*i)/e+1,r);E(rt(o),function(){return"The output # of rows ("+o+") must be an integer. Change the stride and/or zero pad parameters"});var l=zi((s-t+2*i)/e+1,r);return E(rt(l),function(){return"The output # of columns ("+l+") must be an integer. Change the stride and/or zero pad parameters"}),[o,l]}function ET(n,t,e,i,r,a){r==null&&(r=Ic(n,t,i));var s=n[0],o=n[1],l=n[2],u=zi((s-t+2*r)/i+1,a);E(rt(u),function(){return"The output # of depths ("+u+") must be an integer. Change the stride and/or zero pad parameters"});var c=zi((o-t+2*r)/i+1,a);E(rt(c),function(){return"The output # of rows ("+c+") must be an integer. Change the stride and/or zero pad parameters"});var h=zi((l-t+2*r)/i+1,a);return E(rt(h),function(){return"The output # of columns ("+h+") must be an integer. Change the stride and/or zero pad parameters"}),[u,c,h,e]}function Ic(n,t,e,i){i===void 0&&(i=1);var r=Or(t,i);return Math.floor((n[0]*(e-1)-e+r)/2)}function Cs(n){return typeof n=="number"?[n,n,n]:n.length===2?[n[0],n[1],1]:n}function Lc(n){return typeof n=="number"?[n,n,n]:n}function Or(n,t){return t<=1?n:n+(n-1)*(t-1)}function CT(n,t,e,i,r,a,s,o,l){var u,c,h;if(typeof n=="number"){var d=n===0?"VALID":"NUMBER";u={top:n,bottom:n,left:n,right:n,type:d};var p=OT([t,e],a,i,n,o);c=p[0],h=p[1]}else if(n==="same"){c=Math.ceil(t/i),h=Math.ceil(e/r);var f=Math.max(0,(c-1)*i+a-t),m=Math.max(0,(h-1)*r+s-e),g=Math.floor(f/2),v=f-g,b=Math.floor(m/2),w=m-b;u={top:g,bottom:v,left:b,right:w,type:"SAME"}}else if(n==="valid")u={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-a+1)/i),h=Math.ceil((e-s+1)/r);else if(typeof n=="object"){var S=l==="channelsLast"?n[1][0]:n[2][0],v=l==="channelsLast"?n[1][1]:n[2][1],b=l==="channelsLast"?n[2][0]:n[3][0],w=l==="channelsLast"?n[2][1]:n[3][1],d=S===0&&v===0&&b===0&&w===0?"VALID":"EXPLICIT";u={top:S,bottom:v,left:b,right:w,type:d},c=zi((t-a+S+v)/i+1,o),h=zi((e-s+b+w)/r+1,o)}else throw Error("Unknown padding parameter: "+n);return{padInfo:u,outHeight:c,outWidth:h}}function RT(n,t,e,i,r,a,s,o,l,u,c){var h,d,p,f;if(typeof n=="number"){var m=n===0?"VALID":"NUMBER";h={top:n,bottom:n,left:n,right:n,front:n,back:n,type:m};var g=ET([t,e,i,1],o,1,r,n,c);d=g[0],p=g[1],f=g[2]}else if(n==="same"){d=Math.ceil(t/r),p=Math.ceil(e/a),f=Math.ceil(i/s);var v=(d-1)*r+o-t,b=(p-1)*a+l-e,w=(f-1)*s+u-i,S=Math.floor(v/2),L=v-S,x=Math.floor(b/2),C=b-x,R=Math.floor(w/2),D=w-R;h={top:x,bottom:C,left:R,right:D,front:S,back:L,type:"SAME"}}else if(n==="valid")h={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),p=Math.ceil((e-l+1)/a),f=Math.ceil((i-u+1)/s);else throw Error("Unknown padding parameter: "+n);return{padInfo:h,outDepth:d,outHeight:p,outWidth:f}}function zi(n,t){if(!t)return n;switch(t){case"round":return Math.round(n);case"ceil":return Math.ceil(n);case"floor":return Math.floor(n);default:throw new Error("Unknown roundingMode "+t)}}function ci(n){var t=Cs(n),e=t[0],i=t[1],r=t[2];return e===1&&i===1&&r===1}function Pt(n,t){return ci(n)||ci(t)}function Sa(n){if(n==="NHWC")return"channelsLast";if(n==="NCHW")return"channelsFirst";throw new Error("Unknown dataFormat "+n)}function DT(n,t,e,i,r){var a=O(n,"x","avgPool","float32"),s=1;E(Pt(e,s),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")});var o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in avgPool: x must be rank 4 but got rank "+o.rank+"."}),r!=null&&E(rt(i),function(){return"Error in avgPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r);return f([o]),m.filterWidth===1&&m.filterHeight===1&&ln(m.inShape,m.outShape)?o.clone():p.avgPool(o,m)},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,pl,h);return d=ue(d,a.dtype),l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Ac=U({avgPool_:DT});function kT(n,t,e,i,r,a,s){a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","avgPool3d","float32"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in avgPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in avgPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in avgPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in avgPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a);return m([l]),f.avgPool3d(l,g)},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,fl,d);return p=ue(p,l.dtype),u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var FT=U({avgPool3d_:kT});function Xm(n,t){var e=n[0].length;n.forEach(function(r,a){E(r.length===e,function(){return"Error in concat"+e+"D: rank of tensors["+a+"] must be the same "+("as the rank of the rest ("+e+")")})}),E(t>=0&&t<e,function(){return"Error in concat"+e+"D: axis must be between 0 and "+(e-1)+"."});var i=n[0];n.forEach(function(r,a){for(var s=0;s<e;s++)E(s===t||r[s]===i[s],function(){return"Error in concat"+e+"D: Shape of tensors["+a+"] ("+r+") "+("does not match the shape of the rest ("+i+") ")+("along the non-concatenated axis "+a+".")})})}function Jm(n,t){for(var e=n[0].slice(),i=1;i<n.length;i++)e[t]+=n[i][t];return e}function WT(n,t){t===void 0&&(t=0),E(n.length>=1,function(){return"Pass at least one tensor to concat"});var e=ba(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
-          with dtype `+s.dtype+". ")});var i=function(s,o){var l=Qe(t,e[0].shape)[0],u=Jm(e.map(function(d){return d.shape}),l);if(ot(u)===0)return si([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Xm(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,wl,a)}var xt=U({concat_:WT});function UT(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,Lu)}var Pi=U({sigmoid_:UT});function BT(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=gc(i,t,e),c=u[0],h=u[1];return Sm(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,yu,s)}var ze=U({slice_:BT});function zT(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,Du)}var Rs=U({tanh_:zT});function PT(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=xt([u,h],1),p=We(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),x=fe(J(Pi(b),Rs(w)),J(c,Pi(fe(s,S)))),C=J(Rs(x),Pi(L));return[x,C]}var _T=U({basicLSTMCell_:PT});function MT(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length  "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,gl,o)}var Os=U({batchToSpaceND_:MT});function HT(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function VT(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=HT(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Es(o),Es(l),Es(c),Es(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,Ul,f);return V(m,s.shape)}function Es(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Aa=U({batchNorm_:VT});function qT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var GT=U({batchNorm2d_:qT});function YT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var KT=U({batchNorm3d_:YT});function jT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var $T=U({batchNorm4d_:jT});function XT(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.length<e.rank)throw new Error("broadcastTo(): shape.length="+t.length+" < input.rank="+e.rank+".");if(t.length>e.rank){for(var r=e.shape.slice();r.length<t.length;)r.unshift(1);e=V(e,r)}for(var a=e.shape,s=Array.from(t),o=t.length-1;o>=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return Bi(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,vl,h)}var Ds=U({broadcastTo_:XT});function JT(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,yl)}var Zm=U({ceil_:JT});function ZT(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,bl,a)}var Qm=U({clipByValue_:ZT});function QT(n){return xt(n,0)}var e1=U({concat1d_:QT});function t1(n,t){return xt(n,t)}var n1=U({concat2d_:t1});function i1(n,t){return xt(n,t)}var r1=U({concat3d_:i1});function a1(n,t){return xt(n,t)}var s1=U({concat4d_:a1});function o1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(Pt(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=Sa(r),w=Dn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,Sl,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Er=U({conv2d_:o1});function l1(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(Pt(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Er(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var eg=U({conv1d_:l1});function u1(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=Sa(a),S=Dn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,Ll,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Tc=U({conv2DBackpropInput_:u1});function c1(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return Tc(e,s,o,i,r,"NHWC",a)}var tg=U({conv2dTranspose_:c1});function h1(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(Pt(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=La(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,Il,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var d1=U({conv3d_:h1});function p1(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=La(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,Hp,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var ng=U({conv3DBackpropInput_:p1});function f1(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return ng(e,a,s,i,r)}var m1=U({conv3dTranspose_:f1});function g1(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,Al)}var ks=U({cos_:g1});function v1(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,Tl)}var Nc=U({cosh_:v1});function y1(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=tn([t],r.rank),h=r;c!=null&&(h=ut(r,c));var d=En(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=xs(c);p=ut(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,Nl,o)}var xc=U({cumsum_:y1});function b1(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
+`)+("Expected: "+u+"."))}}function zA(n,t){n().then(function(){return t.fail()},function(){return t()})}function PA(n,t){var e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return ni(n)||ni(n[0])||ni(t)||ni(t[0])?bc(n,e,function(i,r){return i==r}):bc(n,t,function(i,r){return wc(i,r,0)})}function _A(n,t,e){if(e==null&&(e=yc()),!wc(n,t,e))throw new Error("Numbers differ: actual === "+n+", expected === "+t)}function wc(n,t,e){return!isFinite(n)&&!isFinite(t)?!0:!(isNaN(n)||isNaN(t)||Math.abs(n-t)>e)}function MA(n,t,e){for(var i=0;i<n.length;i++)if(n[i]<t||n[i]>e)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function HA(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var VA={__proto__:null,TEST_EPSILON_FLOAT16:Wm,expectArraysClose:BA,testEpsilon:yc,expectPromiseToFail:zA,expectArraysEqual:PA,expectNumbersClose:_A,expectValuesInRange:MA,expectArrayBuffersEqual:HA};var qA="2.6.0";function GA(){He().set("PROD",!0)}function YA(){He().set("DEBUG",!0)}function KA(){He().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function It(n){He().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function jA(){z.disposeVariables()}function $A(){return z}function XA(){return z.memory()}function JA(n){return z.profile(n)}function ft(n,t){return z.tidy(n,t)}function zt(n){var t=ic(n);t.forEach(function(e){return e.dispose()})}function Um(n){return z.keep(n)}function ZA(n){return z.time(n)}function QA(n){return z.setBackend(n)}function eT(){return z.ready()}function tT(){return z.backendName}function nT(n){z.removeBackend(n)}function iT(n){return z.findBackend(n)}function rT(n){return z.findBackendFactory(n)}function aT(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function sT(){return z.backend}function oT(n,t){He().setPlatform(n,t)}function lT(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,ps)}var fe=U({add_:lT});function uT(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ul)}var Sc=U({floorDiv_:uT});function cT(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=at(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return Sc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Ol,o)}var Ie=U({div_:cT});function hT(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,tu)}var J=U({mul_:hT});function dT(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,il)}var qt=U({abs_:dT});function pT(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,rl)}var Bm=U({acos_:pT});function fT(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,al)}var zm=U({acosh_:fT});function mT(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!on(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,sl)}var gT=U({addN_:mT});function Lc(n,t){for(var e=0;e<n.length;++e)if(n[n.length-e-1]!==t-1-e)return!1;return!0}function Pm(n,t,e){for(var i=n.length+t.length,r=[],a=0,s=0,o=0;o<i;o++)e.indexOf(o)===-1?r.push(n[a++]):r.push(t[s++]);return r}function _m(n,t){for(var e=[],i=n.length,r=0;r<i;r++)t.indexOf(r)===-1&&e.push(n[r]);var a=t.map(function(s){return n[s]});return[e,a]}function Qt(n,t){var e=t.map(function(i){return 1});return Pm(n,e,t)}function vT(n,t,e){E(Lc(t,e),function(){return n+" supports only inner-most axes for now. "+("Got axes "+t+" and rank-"+e+" input.")})}function en(n,t){if(Lc(n,t))return null;for(var e=[],i=0;i<t;++i)n.indexOf(i)===-1&&e.push(i);return n.forEach(function(r){return e.push(r)}),e}function xs(n){return n.map(function(t,e){return[e,t]}).sort(function(t,e){return t[1]-e[1]}).map(function(t){return t[0]})}function En(n,t){for(var e=[],i=t-n;i<t;++i)e.push(i);return e}function yT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","all","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=en(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.all(i,u);if(e){var d=Qt(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Wp,s)}var Mm=U({all_:yT});function bT(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","any","bool"),r=function(o){var l=Qe(t,i.shape),u=l,c=en(u,i.rank);c!=null&&(i=ut(i,c),u=En(u.length,i.rank));var h=o.any(i,u);if(e){var d=Qt(h.shape,l);return V(h,d)}return h},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Up,s)}var Hm=U({any_:bT});function wT(n,t){t===void 0&&(t=0);var e=O(n,"x","argMax"),i=function(s,o){o([e]);var l=Qe(t,e.shape),u=en(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMax(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,ol,a)}var Vm=U({argMax_:wT});function ST(n,t){t===void 0&&(t=0);var e=O(n,"x","argMin"),i=function(s,o){o([e]),t==null&&(t=0);var l=Qe(t,e.shape),u=en(l,e.rank);return u!=null&&(e=ut(e,u),l=En(l.length,e.rank)),s.argMin(e,l[0])},r={x:e},a={axis:t};return z.runKernelFunc(i,r,null,ll,a)}var qm=U({argMin_:ST});function LT(n){var t=O(n,"x","asin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asin(t);return r([t]),a},e,null,ul)}var Gm=U({asin_:LT});function IT(n){var t=O(n,"x","asinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.asinh(t);return r([t]),a},e,null,cl)}var Ym=U({asinh_:IT});function AT(n){var t=O(n,"x","atan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atan(t);return r([t]),a},e,null,hl)}var Km=U({atan_:AT});function TT(n,t){var e,i=O(n,"a","atan2"),r=O(t,"b","atan2");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.atan2(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,pl)}var jm=U({atan2_:TT});function NT(n){var t=O(n,"x","atanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.atanh(t);return r([t]),a},e,null,dl)}var $m=U({atanh_:NT});function xT(n,t,e,i,r,a){r===void 0&&(r="NHWC");var s=n[3],o=t.concat([s]),l=Sa(r);return Dn(n,o,e,a,i,null,null,l)}function Rr(n,t,e,i,r,a,s){s===void 0&&(s="channelsLast");var o=Cs(t),l=o[0],u=o[1],c;if(s==="channelsLast")c=[l,u,n[3],n[3]];else if(s==="channelsFirst")c=[l,u,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return Dn(n,c,e,i,r,a,!1,s)}function Ia(n,t,e,i,r,a,s){s===void 0&&(s="NDHWC");var o=Ic(t),l=o[0],u=o[1],c=o[2],h,d;if(s==="NDHWC")d="channelsLast",h=[l,u,c,n[4],n[4]];else if(s==="NCDHW")d="channelsFirst",h=[l,u,c,n[1],n[1]];else throw new Error("Unknown dataFormat "+s);return La(n,h,e,i,r,!1,d,a)}function Dn(n,t,e,i,r,a,s,o){s===void 0&&(s=!1),o===void 0&&(o="channelsLast");var l=[-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3];if(o==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3];else if(o==="channelsFirst")u=n[0],d=n[1],c=n[2],h=n[3];else throw new Error("Unknown dataFormat "+o);var p=t[0],f=t[1],m=t[3],g=Cs(e),v=g[0],b=g[1],w=Cs(i),S=w[0],L=w[1],x=Or(p,S),C=Or(f,L),R=CT(r,c,h,v,b,x,C,a,o),D=R.padInfo,k=R.outHeight,W=R.outWidth,F=s?m*d:m,P;return o==="channelsFirst"?P=[u,F,k,W]:o==="channelsLast"&&(P=[u,k,W,F]),{batchSize:u,dataFormat:o,inHeight:c,inWidth:h,inChannels:d,outHeight:k,outWidth:W,outChannels:F,padInfo:D,strideHeight:v,strideWidth:b,filterHeight:p,filterWidth:f,effectiveFilterHeight:x,effectiveFilterWidth:C,dilationHeight:S,dilationWidth:L,inShape:n,outShape:P,filterShape:t}}function La(n,t,e,i,r,a,s,o){a===void 0&&(a=!1),s===void 0&&(s="channelsLast");var l=[-1,-1,-1,-1,-1],u=l[0],c=l[1],h=l[2],d=l[3],p=l[4];if(s==="channelsLast")u=n[0],c=n[1],h=n[2],d=n[3],p=n[4];else if(s==="channelsFirst")u=n[0],p=n[1],c=n[2],h=n[3],d=n[4];else throw new Error("Unknown dataFormat "+s);var f=t[0],m=t[1],g=t[2],v=t[4],b=Ic(e),w=b[0],S=b[1],L=b[2],x=Ic(i),C=x[0],R=x[1],D=x[2],k=Or(f,C),W=Or(m,R),F=Or(g,D),P=RT(r,c,h,d,w,S,L,k,W,F,o),H=P.padInfo,_=P.outDepth,K=P.outHeight,j=P.outWidth,q=a?v*p:v,G;return s==="channelsFirst"?G=[u,q,_,K,j]:s==="channelsLast"&&(G=[u,_,K,j,q]),{batchSize:u,dataFormat:s,inDepth:c,inHeight:h,inWidth:d,inChannels:p,outDepth:_,outHeight:K,outWidth:j,outChannels:q,padInfo:H,strideDepth:w,strideHeight:S,strideWidth:L,filterDepth:f,filterHeight:m,filterWidth:g,effectiveFilterDepth:k,effectiveFilterHeight:W,effectiveFilterWidth:F,dilationDepth:C,dilationHeight:R,dilationWidth:D,inShape:n,outShape:G,filterShape:t}}function OT(n,t,e,i,r){i==null&&(i=Ac(n,t,e));var a=n[0],s=n[1],o=zi((a-t+2*i)/e+1,r);E(rt(o),function(){return"The output # of rows ("+o+") must be an integer. Change the stride and/or zero pad parameters"});var l=zi((s-t+2*i)/e+1,r);return E(rt(l),function(){return"The output # of columns ("+l+") must be an integer. Change the stride and/or zero pad parameters"}),[o,l]}function ET(n,t,e,i,r,a){r==null&&(r=Ac(n,t,i));var s=n[0],o=n[1],l=n[2],u=zi((s-t+2*r)/i+1,a);E(rt(u),function(){return"The output # of depths ("+u+") must be an integer. Change the stride and/or zero pad parameters"});var c=zi((o-t+2*r)/i+1,a);E(rt(c),function(){return"The output # of rows ("+c+") must be an integer. Change the stride and/or zero pad parameters"});var h=zi((l-t+2*r)/i+1,a);return E(rt(h),function(){return"The output # of columns ("+h+") must be an integer. Change the stride and/or zero pad parameters"}),[u,c,h,e]}function Ac(n,t,e,i){i===void 0&&(i=1);var r=Or(t,i);return Math.floor((n[0]*(e-1)-e+r)/2)}function Cs(n){return typeof n=="number"?[n,n,n]:n.length===2?[n[0],n[1],1]:n}function Ic(n){return typeof n=="number"?[n,n,n]:n}function Or(n,t){return t<=1?n:n+(n-1)*(t-1)}function CT(n,t,e,i,r,a,s,o,l){var u,c,h;if(typeof n=="number"){var d=n===0?"VALID":"NUMBER";u={top:n,bottom:n,left:n,right:n,type:d};var p=OT([t,e],a,i,n,o);c=p[0],h=p[1]}else if(n==="same"){c=Math.ceil(t/i),h=Math.ceil(e/r);var f=Math.max(0,(c-1)*i+a-t),m=Math.max(0,(h-1)*r+s-e),g=Math.floor(f/2),v=f-g,b=Math.floor(m/2),w=m-b;u={top:g,bottom:v,left:b,right:w,type:"SAME"}}else if(n==="valid")u={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-a+1)/i),h=Math.ceil((e-s+1)/r);else if(typeof n=="object"){var S=l==="channelsLast"?n[1][0]:n[2][0],v=l==="channelsLast"?n[1][1]:n[2][1],b=l==="channelsLast"?n[2][0]:n[3][0],w=l==="channelsLast"?n[2][1]:n[3][1],d=S===0&&v===0&&b===0&&w===0?"VALID":"EXPLICIT";u={top:S,bottom:v,left:b,right:w,type:d},c=zi((t-a+S+v)/i+1,o),h=zi((e-s+b+w)/r+1,o)}else throw Error("Unknown padding parameter: "+n);return{padInfo:u,outHeight:c,outWidth:h}}function RT(n,t,e,i,r,a,s,o,l,u,c){var h,d,p,f;if(typeof n=="number"){var m=n===0?"VALID":"NUMBER";h={top:n,bottom:n,left:n,right:n,front:n,back:n,type:m};var g=ET([t,e,i,1],o,1,r,n,c);d=g[0],p=g[1],f=g[2]}else if(n==="same"){d=Math.ceil(t/r),p=Math.ceil(e/a),f=Math.ceil(i/s);var v=(d-1)*r+o-t,b=(p-1)*a+l-e,w=(f-1)*s+u-i,S=Math.floor(v/2),L=v-S,x=Math.floor(b/2),C=b-x,R=Math.floor(w/2),D=w-R;h={top:x,bottom:C,left:R,right:D,front:S,back:L,type:"SAME"}}else if(n==="valid")h={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),p=Math.ceil((e-l+1)/a),f=Math.ceil((i-u+1)/s);else throw Error("Unknown padding parameter: "+n);return{padInfo:h,outDepth:d,outHeight:p,outWidth:f}}function zi(n,t){if(!t)return n;switch(t){case"round":return Math.round(n);case"ceil":return Math.ceil(n);case"floor":return Math.floor(n);default:throw new Error("Unknown roundingMode "+t)}}function ci(n){var t=Cs(n),e=t[0],i=t[1],r=t[2];return e===1&&i===1&&r===1}function Pt(n,t){return ci(n)||ci(t)}function Sa(n){if(n==="NHWC")return"channelsLast";if(n==="NCHW")return"channelsFirst";throw new Error("Unknown dataFormat "+n)}function DT(n,t,e,i,r){var a=O(n,"x","avgPool","float32"),s=1;E(Pt(e,s),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")});var o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in avgPool: x must be rank 4 but got rank "+o.rank+"."}),r!=null&&E(rt(i),function(){return"Error in avgPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r);return f([o]),m.filterWidth===1&&m.filterHeight===1&&on(m.inShape,m.outShape)?o.clone():p.avgPool(o,m)},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,fl,h);return d=ue(d,a.dtype),l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Tc=U({avgPool_:DT});function kT(n,t,e,i,r,a,s){a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","avgPool3d","float32"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in avgPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in avgPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in avgPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in avgPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a);return m([l]),f.avgPool3d(l,g)},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,ml,d);return p=ue(p,l.dtype),u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var FT=U({avgPool3d_:kT});function Xm(n,t){var e=n[0].length;n.forEach(function(r,a){E(r.length===e,function(){return"Error in concat"+e+"D: rank of tensors["+a+"] must be the same "+("as the rank of the rest ("+e+")")})}),E(t>=0&&t<e,function(){return"Error in concat"+e+"D: axis must be between 0 and "+(e-1)+"."});var i=n[0];n.forEach(function(r,a){for(var s=0;s<e;s++)E(s===t||r[s]===i[s],function(){return"Error in concat"+e+"D: Shape of tensors["+a+"] ("+r+") "+("does not match the shape of the rest ("+i+") ")+("along the non-concatenated axis "+a+".")})})}function Jm(n,t){for(var e=n[0].slice(),i=1;i<n.length;i++)e[t]+=n[i][t];return e}function WT(n,t){t===void 0&&(t=0),E(n.length>=1,function(){return"Pass at least one tensor to concat"});var e=ba(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
+          with dtype `+s.dtype+". ")});var i=function(s,o){var l=Qe(t,e[0].shape)[0],u=Jm(e.map(function(d){return d.shape}),l);if(ot(u)===0)return si([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Xm(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,Sl,a)}var xt=U({concat_:WT});function UT(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,Iu)}var Pi=U({sigmoid_:UT});function BT(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=vc(i,t,e),c=u[0],h=u[1];return Sm(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,bu,s)}var ze=U({slice_:BT});function zT(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,ku)}var Rs=U({tanh_:zT});function PT(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=xt([u,h],1),p=We(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),x=fe(J(Pi(b),Rs(w)),J(c,Pi(fe(s,S)))),C=J(Rs(x),Pi(L));return[x,C]}var _T=U({basicLSTMCell_:PT});function MT(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length  "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,vl,o)}var Os=U({batchToSpaceND_:MT});function HT(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function VT(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=HT(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Es(o),Es(l),Es(c),Es(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,Bl,f);return V(m,s.shape)}function Es(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Aa=U({batchNorm_:VT});function qT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var GT=U({batchNorm2d_:qT});function YT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var KT=U({batchNorm3d_:YT});function jT(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Aa(s,o,l,c,u,a)}var $T=U({batchNorm4d_:jT});function XT(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.length<e.rank)throw new Error("broadcastTo(): shape.length="+t.length+" < input.rank="+e.rank+".");if(t.length>e.rank){for(var r=e.shape.slice();r.length<t.length;)r.unshift(1);e=V(e,r)}for(var a=e.shape,s=Array.from(t),o=t.length-1;o>=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return Bi(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,yl,h)}var Ds=U({broadcastTo_:XT});function JT(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,bl)}var Zm=U({ceil_:JT});function ZT(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,wl,a)}var Qm=U({clipByValue_:ZT});function QT(n){return xt(n,0)}var e1=U({concat1d_:QT});function t1(n,t){return xt(n,t)}var n1=U({concat2d_:t1});function i1(n,t){return xt(n,t)}var r1=U({concat3d_:i1});function a1(n,t){return xt(n,t)}var s1=U({concat4d_:a1});function o1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(Pt(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=Sa(r),w=Dn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,Ll,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Er=U({conv2d_:o1});function l1(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(rt(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(Pt(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Er(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var eg=U({conv1d_:l1});function u1(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=Sa(a),S=Dn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,Il,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Nc=U({conv2DBackpropInput_:u1});function c1(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return Nc(e,s,o,i,r,"NHWC",a)}var tg=U({conv2dTranspose_:c1});function h1(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(Pt(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=La(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,Al,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var d1=U({conv3d_:h1});function p1(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=La(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,Hp,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var ng=U({conv3DBackpropInput_:p1});function f1(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return ng(e,a,s,i,r)}var m1=U({conv3dTranspose_:f1});function g1(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,Tl)}var ks=U({cos_:g1});function v1(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,Nl)}var xc=U({cosh_:v1});function y1(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=en([t],r.rank),h=r;c!=null&&(h=ut(r,c));var d=En(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=xs(c);p=ut(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,xl,o)}var Cc=U({cumsum_:y1});function b1(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
     `+r+" and "+t+`  for depthToSpace with input shape
     `+i.shape}),E(a*t>=0,function(){return`Negative dimension size caused by overflow when multiplying
     `+a+" and "+t+` for depthToSpace with input shape
-        `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,qp,u)}var ig=U({depthToSpace_:b1});function w1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(rt(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(Pt(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Dn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,xl,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ta=U({depthwiseConv2d_:w1});function S1(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Kp)}var L1=U({diag_:S1});function I1(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(Cl,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var rg=U({dilation2d_:I1});function A1(n,t){for(var e=n.length,i=[],r=0;r<e;r++){var a=e-1-r,s=n[a]||1,o=t[t.length-1-r]||1;o>1&&s===1&&i.unshift(a)}return i}function mt(n,t){for(var e=[],i=0;i<t.length;i++){var r=n[n.length-i-1],a=t.length-i-1,s=t[a];(r==null||r===1&&s>1)&&e.unshift(a)}return e}function et(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;r<i;r++){var a=n[n.length-r-1];a==null&&(a=1);var s=t[t.length-r-1];if(s==null&&(s=1),a===1)e.unshift(s);else if(s===1)e.unshift(a);else if(a!==s){var o="Operands could not be broadcast together with shapes "+(n+" and "+t+".");throw Error(o)}else e.unshift(a)}return e}function T1(n,t){var e,i=O(n,"a","equal"),r=O(t,"b","equal");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.equal(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Jp)}var Na=U({equal_:T1});function N1(n,t,e){var i=O(t,"a","where"),r=O(e,"b","where"),a=O(n,"condition","where","bool"),s=et(i.shape,r.shape),o=Ds(i,s),l=Ds(r,s);a.rank===1&&E(a.shape[0]===i.shape[0],function(){return"The first dimension of `a` must match the size of `condition`."}),a.rank!==1&&Be(a.shape,l.shape,"Error in where: ");var u=function(h,d){var p=h.select(a,o,l);return d([a]),p},c={condition:a,t:o,e:l};return z.runKernelFunc(u,c,null,gu)}var pn=U({where_:N1});function x1(n){var t=O(n,"x","zerosLike"),e={x:t};return z.runKernelFunc(function(i){return i.zerosLike(t)},e,null,Bu)}var Ee=U({zerosLike_:x1});function C1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");e=at(i,r),i=e[0],r=e[1];var a=Ie(i,r),s=Ee(a),o=Na(r,s);return pn(o,s,a)}var ag=U({divNoNan_:C1});function R1(n,t){var e=O(n,"t1","dot"),i=O(t,"t2","dot");E((e.rank===1||e.rank===2)&&(i.rank===1||i.rank===2),function(){return"Error in dot: inputs must all be rank 1 or 2, but got ranks "+(e.rank+" and "+i.rank+".")});var r=e.rank===1?e.size:e.shape[1],a=i.rank===1?i.size:i.shape[0];if(E(r===a,function(){return"Error in dot: inner dimensions of inputs must match, but got "+(r+" and "+a+".")}),e.rank===1&&i.rank===1){var s=V(e,[1,-1]),o=V(i,[-1,1]),l=We(s,o);return V(l,[])}else if(e.rank===1&&i.rank===2){var s=V(e,[1,-1]),o=V(i,[i.shape[0],i.shape[1]]),l=We(s,o);return V(l,[l.size])}else if(e.rank===2&&i.rank===1){var o=V(i,[-1,1]),l=We(e,o);return V(l,[l.size])}else{var o=V(i,[i.shape[0],i.shape[1]]),l=We(e,o);return l}}var sg=U({dot_:R1});function O1(n){var t=O(n,"x","elu"),e=function(r,a){var s=r.elu(t);return a([s]),s},i={x:t};return z.runKernelFunc(e,i,null,Ol)}var Cc=U({elu_:O1});function E1(n){var t=O(n,"x","erf");E(t.dtype==="int32"||t.dtype==="float32",function(){return"Input dtype must be `int32` or `float32`."}),t.dtype==="int32"&&(t=ue(t,"float32"));var e={x:t};return z.runKernelFunc(function(i,r){var a=i.erf(t);return r([t]),a},e,null,El)}var og=U({erf_:E1});function D1(n){var t=O(n,"x","exp"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.exp(t);return r([a]),a},e,null,Dl)}var fn=U({exp_:D1});function k1(n,t){t===void 0&&(t=0);var e=null,i=O(n,"x","expandDims",e);E(t<=i.rank,function(){return"Axis must be <= rank of the tensor"});var r=i.shape.slice();return t<0&&(E(-(i.rank+1)<=t,function(){return"Axis must be in the interval ["+-(i.rank+1)+", "+i.rank+"]"}),t=i.rank+t+1),r.splice(t,0,1),V(i,r)}var mn=U({expandDims_:k1});function F1(n){var t=O(n,"x","expm1"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.expm1(t);return r([t]),a},e,null,kl)}var lg=U({expm1_:F1});function W1(n,t){var e=null,i=O(n,"x","tile",e);E(i.rank===t.length,function(){return"Error in transpose: rank of input "+i.rank+" "+("must match length of reps "+t+".")});var r=function(l,u){var c=l.tile(i,t);return u([i]),c},a=[i],s={x:i},o={reps:t};return z.runKernelFunc(r,s,null,ku,o,a)}var Dr=U({tile_:W1});function U1(n,t,e,i){i===void 0&&(i="float32"),t==null&&(t=n);for(var r=On([n,t],i),a=n<=t?n:t,s=0;s<a;++s)r.set(1,s,s);var o=V(r.toTensor(),[n,t]);if(e==null)return o;if(e.length===1)return Dr(mn(o,0),[e[0],1,1]);if(e.length===2)return Dr(mn(mn(o,0),0),[e[0],e[1],1,1]);if(e.length===3)return Dr(mn(mn(mn(o,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error("eye() currently supports only 1D and 2D "+("batchShapes, but received "+e.length+"D."))}var ug=U({eye_:U1});function Rc(n,t,e){var i={shape:n,value:t,dtype:e};return z.runKernelFunc(function(r){return r.fill(n,t,e)},{},null,Qp,i)}function B1(n){var t=O(n,"x","floor"),e={x:t};return z.runKernelFunc(function(i){return i.floor(t)},e,null,Fl)}var Fs=U({floor_:B1});var Oc=30;function z1(n){return n<=Oc?n:ys(n,Math.floor(Math.sqrt(n)))}function P1(n,t){var e=!1,i;for(n<=Oc?(i=n,e=!0):i=ys(n,Math.floor(Math.sqrt(n)));!e;)i>t||i===n?e=!0:i=ys(n,i+1);return i}function _1(n,t,e){for(var i=[],r=n.length,a=0;a<r;a++)a!==t?i.push(n[a]):i.push(e);return i}function cg(n,t,e){for(var i=n.shape[e],r=[],a=1,s=1,o=0;o<e;o++)r.push(n.shape[o]),a*=n.shape[o];for(var o=0;o<t.rank;o++)r.push(t.shape[o]);for(var o=e+1;o<n.rank;o++)r.push(n.shape[o]),s*=n.shape[o];return{batchSize:a,sliceSize:s,dimSize:i,outputShape:r}}var M1={__proto__:null,segOpComputeOptimalWindowSize:P1,computeOutShape:_1,collectGatherOpShapeInfo:cg};function H1(n,t,e){e===void 0&&(e=0);var i=O(n,"x","gather"),r=O(t,"indices","gather","int32"),a={x:i,indices:r},s={axis:e},o=function(l,u){var c=Qe(e,i.shape)[0],h=cg(i,r,c),d=l.gather(i,V(r,[r.size]),c);return u([i,r]),V(d,h.outputShape)};return z.runKernelFunc(o,a,null,Bl,s)}var Ws=U({gather_:H1});function V1(n,t){var e,i=O(n,"a","greater"),r=O(t,"b","greater");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.greater(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,nf)}var hi=U({greater_:V1});function q1(n,t){var e,i=O(n,"a","greaterEqual"),r=O(t,"b","greaterEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.greaterEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,zl)}var _i=U({greaterEqual_:q1});function G1(n){var t=O(n,"input","imag"),e=function(r){return r.imag(t)},i={input:t};return z.runKernelFunc(e,i,null,af)}var Us=U({imag_:G1});function Y1(n){var t=O(n,"x","isFinite"),e={x:t};return z.runKernelFunc(function(i){return i.isFinite(t)},e,null,_l)}var hg=U({isFinite_:Y1});function K1(n){var t=O(n,"x","isInf"),e={x:t};return z.runKernelFunc(function(i){return i.isInf(t)},e,null,Ml)}var dg=U({isInf_:K1});function j1(n){var t=O(n,"x","isNaN"),e={x:t};return z.runKernelFunc(function(i){return i.isNaN(t)},e,null,Hl)}var pg=U({isNaN_:j1});function $1(n,t){var e,i=O(n,"a","maximum"),r=O(t,"b","maximum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.maximum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,jl)}var kr=U({maximum_:$1});function we(n,t){if((Ot(n)&&t!=="string"||Array.isArray(n))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&Ot(n)&&!(n instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");var e=[],i=[];return ai(n,e,i,t)}function X1(n,t){t===void 0&&(t=.2);var e=O(n,"x","leakyRelu");return kr(J(we(t),e),e)}var fg=U({leakyRelu_:X1});function J1(n,t){var e,i=O(n,"a","less"),r=O(t,"b","less");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.less(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,sf)}var Bs=U({less_:J1});function Z1(n,t){var e,i=O(n,"a","lessEqual"),r=O(t,"b","lessEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.lessEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,of)}var Mi=U({lessEqual_:Z1});function Q1(n,t,e){if(e<=0)throw new Error("The number of values should be positive.");var i={start:n,stop:t,num:e};return z.runKernelFunc(function(r){return r.linspace(n,t,e)},{},null,lf,i)}function eN(n,t,e,i,r){t===void 0&&(t=5),e===void 0&&(e=1),i===void 0&&(i=1),r===void 0&&(r=.5);var a=O(n,"x","localResponseNormalization");E(a.rank===4||a.rank===3,function(){return`Error in localResponseNormalization: x must be rank 3 or 4 but got
-               rank `+a.rank+"."}),E(rt(t),function(){return"Error in localResponseNormalization: depthRadius must be an "+("integer but got depthRadius "+t+".")});var s=a,o=!1;a.rank===3&&(o=!0,s=V(a,[1,a.shape[0],a.shape[1],a.shape[2]]));var l=function(d,p){var f=d.localResponseNormalization4D(s,t,e,i,r);return p([s,f]),f},u={x:s},c={depthRadius:t,bias:e,alpha:i,beta:r},h=z.runKernelFunc(l,u,null,Yl,c);return o?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var mg=U({localResponseNormalization_:eN});function tN(n){var t=O(n,"x","log"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log(t);return r([t]),a},e,null,Vl)}var Hi=U({log_:tN});function nN(n){var t=O(n,"x","log1p"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log1p(t);return r([t]),a},e,null,ql)}var Ec=U({log1p_:nN});function iN(n){return E(ii(n),function(){return"The f passed in grad(f) must be a function"}),function(t,e){var i=O(t,"x","tf.grad",null),r=e!=null?O(e,"dy","tf.grad"):null;return z.tidy(function(){var a=z.gradients(function(){return n(i)},[i],r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),zs(o),o[0]})}}function rN(n){return E(ii(n),function(){return"The f passed in grads(f) must be a function"}),function(t,e){E(Array.isArray(t),function(){return"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s"});var i=ba(t,"args","tf.grads",null),r=e!=null?O(e,"dy","tf.grads"):null;return z.tidy(function(){var a=z.gradients(function(){return n.apply(void 0,i)},i,r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(o),o})}}function aN(n){return E(ii(n),function(){return"The f passed in valueAndGrad(f) must be a function"}),function(t,e){E(t instanceof Y,function(){return"The x passed in valueAndGrad(f)(x) must be a tensor"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrad(f)(x, dy) must be a tensor"});var i=z.gradients(function(){return n(t)},[t],e),r=i.grads,a=i.value;return zs(r),{grad:r[0],value:a}}}function sN(n){return E(ii(n),function(){return"The f passed in valueAndGrads(f) must be a function"}),function(t,e){E(Array.isArray(t)&&t.every(function(r){return r instanceof Y}),function(){return"The args passed in valueAndGrads(f)(args) must be array of tensors"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrads(f)(args, dy) must be a tensor"});var i=z.gradients(function(){return n.apply(void 0,t)},t,e);return e!=null&&Be(i.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(i.grads),i}}function gg(n,t){E(ii(n),function(){return"The f passed in variableGrads(f) must be a function"}),E(t==null||Array.isArray(t)&&t.every(function(h){return h instanceof ya}),function(){return"The varList passed in variableGrads(f, varList) must be an array of variables"});var e=t!=null;if(!e){t=[];for(var i in z.registeredVariables)t.push(z.registeredVariables[i])}var r=e?t.filter(function(h){return!h.trainable}):null,a=t.length;t=t.filter(function(h){return h.trainable}),E(t.length>0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function kn(n){return z.customGrad(n)}function zs(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
-    the f you passed encloses all operations that lead from x to y.`)}function oN(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,tu)}var gt=U({neg_:oN});function lN(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,Iu)}var Dc=U({softplus_:lN});function uN(n){var t=O(n,"x","logSigmoid"),e=kn(function(i){var r=gt(Dc(gt(i))),a=function(s){var o=J(s,Pi(gt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var vg=U({logSigmoid_:uN});function cN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=tn(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=en(f.shape,Qe(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,Kl,s)}var Vi=U({max_:cN});function hN(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ou)}var ge=U({sub_:hN});function dN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Qe(t,i.shape),c=tn(u,i.rank),h=u,d=i;c!=null&&(d=ut(i,c),h=En(h.length,i.rank));var p=o.sum(d,h);if(e){var f=en(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Tu,s)}var Ae=U({sum_:dN});function pN(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=Vi(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),Hi(Ae(fn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,Gl,a)}var yg=U({logSoftmax_:pN});function fN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Qe(t,i.shape),a=Vi(i,r,!0),s=ge(i,a),o=fn(s),l=Ae(o,r),u=Hi(l),c=fe(V(a,u.shape),u);if(e){var h=en(c.shape,r);return V(c,h)}return c}var kc=U({logSumExp_:fN});function mN(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,uf)}var qi=U({logicalAnd_:mN});function gN(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,cf)}var Ps=U({logicalNot_:gN});function vN(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,hf)}var Fc=U({logicalOr_:vN});function yN(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return et(e.shape,i.shape),qi(Fc(n,t),Ps(qi(n,t)))}var bg=U({logicalXor_:yN});function bN(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(Pt(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&ln(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,$l,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Wc=U({maxPool_:bN});function wN(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Xl,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var SN=U({maxPool3d_:wN});function LN(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(mf,s,o);return{result:l[0],indexes:l[1]}}var IN=U({maxPoolWithArgmax_:LN});function Kn(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Kn(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=Lr(ot(n),t);return z.makeTensor(r,n,t)}function Fr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Fr(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=Ku(ot(n),t);return z.makeTensor(r,n,t)}function AN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Qe(t,i.shape),a=_m(i.shape,r),s=a[1],o=ot(s),l=kn(function(u){var c=we(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Ie(h,c),p=Ae(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Ie(J(v,Fr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var xa=U({mean_:AN});function TN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=tn(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=en(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Jl,s)}var _s=U({min_:TN});function NN(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Zl)}var Ms=U({minimum_:NN});function xN(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ql)}var Uc=U({mod_:xN});function CN(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ye=U({square_:CN});function RN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Qe(t,n.shape),r=xa(n,i,e),a=r.shape;e||(a=en(r.shape,i));var s=Ye(ge(ue(n,"float32"),V(r,a))),o=xa(s,i,e);return{mean:r,variance:o}}var ON=U({moments_:RN});function EN(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=ba(e,"c","multiRNNCell"),s=ba(i,"h","multiRNNCell"),o=r,l=[],u=0;u<n.length;u++){var c=n[u](o,a[u],s[u]);l.push(c[0]),l.push(c[1]),o=c[1]}for(var h=[],d=[],u=0;u<l.length;u+=2)h.push(l[u]),d.push(l[u+1]);return[h,d]}var DN=U({multiRNNCell_:EN});function kN(n,t,e,i){i===void 0&&(i=!1);var r=O(n,"logits","multinomial"),a=r.size,s=r.rank;if(a<2)throw new Error("Error in multinomial: you need at least 2 outcomes, but got "+(a+"."));if(s>2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var FN=U({multinomial_:kN});function WN(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,gf)}var Hs=U({notEqual_:WN});function UN(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,Lf)}var Ca=U({real_:UN});function BN(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=Bc(Ca(t)),o=Ee(Us(t));return ri(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,nu)}var Bc=U({onesLike_:BN});function zN(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return We(r,a)}var PN=U({outerProduct_:zN});function _N(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,ru,a)}var Gi=U({pad_:_N});function MN(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),Gi(n,[t],e)}var HN=U({pad1d_:MN});function VN(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var qN=U({pad2d_:VN});function GN(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var YN=U({pad3d_:GN});function KN(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var jN=U({pad4d_:KN});function $N(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,Nu,s)}var Vs=U({spaceToBatchND_:$N});function ZN(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(Pt(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=Rr(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=JN([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=XN([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:Vs(o,c,f),b=e==="avg"?function(){return Ac(v,t,a,g)}:function(){return Wc(v,t,a,g)},w=b(),S=d?w:Os(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function XN(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function JN(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var wg=U({pool_:ZN});function QN(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=at(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,au)}var jn=U({pow_:QN});function ex(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,su)}var zc=U({prelu_:ex});function tx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Qe(t,i.shape),u=tn(l,i.rank),c=l,h=i;u!=null&&(h=ut(i,u),c=En(c.length,i.rank));var d=o.prod(h,c);if(e){var p=en(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,wf,s)}var Sg=U({prod_:tx});function nx(n,t,e){var i=ot(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a<i;a++)r[a]=t();return z.makeTensor(r,n,e)}var ix=U({rand_:nx}),Wr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function Yi(n,t){return t={exports:{}},n(t,t.exports),t.exports}var rx=Yi(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(Wr,n,!1)}),ax=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(Wr,n,!1)}),sx=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(Wr,n,!1)}),ox=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(Wr,n,!1)}),lx=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(Wr,n,!1)}),ux=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(Wr,n,!1)}),Ki=Yi(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});Ki.alea=rx;Ki.xor128=ax;Ki.xorwow=sx;Ki.xorshift7=ox;Ki.xor4096=lx;Ki.tychei=ux;var cx=Ki,Pc=cx.alea;var _c=function(){function n(t,e,i,r,a){this.mean=t,this.stdDev=e,this.dtype=i,this.nextVal=NaN,this.truncated=r,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);var s=a||Math.random();this.random=Pc(s.toString())}return n.prototype.nextValue=function(){if(!isNaN(this.nextVal)){var t=this.nextVal;return this.nextVal=NaN,t}for(var e,i,r=!1;!r;){var a=void 0,s=void 0,o=void 0;do a=2*this.random()-1,s=2*this.random()-1,o=a*a+s*s;while(o>=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),hx=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=Pc(a.toString()),this.randn=new _c(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a<e||Math.log(a)<i)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)},n.prototype.convertValue=function(t){return this.dtype==="float32"?t:Math.round(t)},n}(),dx=function(){function n(t,e,i,r){var a=this;if(t===void 0&&(t=0),e===void 0&&(e=1),this.canReturnFloat=function(){return a.dtype==null||a.dtype==="float32"},this.min=t,this.range=e-t,this.dtype=i,r==null&&(r=Math.random()),typeof r=="number"&&(r=r.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error("The difference between "+t+" - "+e+" <= 1 and dtype is not float");this.random=Pc(r)}return n.prototype.convertValue=function(t){return this.canReturnFloat()?t:Math.round(t)},n.prototype.nextValue=function(){return this.convertValue(this.min+this.range*this.random())},n}();function px(n,t,e,i,r){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e==null&&(e=1),i==null&&(i="float32"),i!=="float32"&&i!=="int32")throw new Error("Unsupported data type "+i);for(var a=new hx(t,e,i,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var fx=U({randomGamma_:px});function mx(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type "+i);for(var a=new _c(t,e,i,!1,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var gx=U({randomNormal_:mx});function vx(n,t,e,i,r){t===void 0&&(t=0),e===void 0&&(e=1),i===void 0&&(i="float32");for(var a=On(n,i),s=new dx(t,e,null,r),o=0;o<a.values.length;o++)a.values[o]=s.nextValue();return a.toTensor()}var Lg=U({randomUniform_:vx});function Ur(n,t){Fi(n);var e=Rn(n,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");var i=null;return ai(n,i,e,t)}function Mc(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e===0)throw new Error("Cannot have a step of zero");var r=function(){var s=n===t,o=n<t&&e<0,l=t<n&&e>1;if(s||o||l)return Kn([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=Lr(u,i);t<n&&e===1&&(e=-1),c[0]=n;for(var h=1;h<c.length;h++)c[h]=c[h-1]+e;return Ur(c,i)},a={start:n,stop:t,step:e,dtype:i};return z.runKernelFunc(r,{},null,Sf,a)}function yx(n){var t=O(n,"x","reciprocal"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.reciprocal(t);return r([t]),a},e,null,ou)}var Ig=U({reciprocal_:yx});function bx(n){var t=O(n,"x","relu"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu(t)},i={x:t};return z.runKernelFunc(e,i,null,lu)}var Ra=U({relu_:bx});function wx(n){var t=O(n,"x","relu6"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu6(t)},i={x:t};return z.runKernelFunc(e,i,null,du)}var Hc=U({relu6_:wx});function Sx(n,t){var e=O(n,"x","reverse"),i=function(s){var o=Qe(t,e.shape);if(e.rank===0)return Bi(e);var l=s.reverse(e,o);return V(l,e.shape)},r={x:e},a={dims:t};return z.runKernelFunc(i,r,null,pu,a)}var $n=U({reverse_:Sx});function Lx(n){var t=O(n,"x","reverse");return E(t.rank===1,function(){return"Error in reverse1D: x must be rank 1 but got rank "+t.rank+"."}),$n(t,0)}var Ix=U({reverse1d_:Lx});function Ax(n,t){var e=O(n,"x","reverse");return E(e.rank===2,function(){return"Error in reverse2D: x must be rank 2 but got rank "+e.rank+"."}),$n(e,t)}var Tx=U({reverse2d_:Ax});function Nx(n,t){var e=O(n,"x","reverse");return E(e.rank===3,function(){return"Error in reverse3D: x must be rank 3 but got rank "+e.rank+"."}),$n(e,t)}var xx=U({reverse3d_:Nx});function Cx(n,t){var e=O(n,"x","reverse");return E(e.rank===4,function(){return"Error in reverse4D: x must be rank 4 but got rank "+e.rank+"."}),$n(e,t)}var Rx=U({reverse4d_:Cx});function Ox(n){var t=O(n,"x","round"),e={x:t};return z.runKernelFunc(function(i){return i.round(t)},e,null,fu)}var Ag=U({round_:Ox});function Ex(n){var t=O(n,"x","rsqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.rsqrt(t);return r([t]),a},e,null,mu)}var Vc=U({rsqrt_:Ex});function Dx(n){var t=O(n,"x","selu"),e=function(r,a){var s=r.selu(t);return a([t]),s},i={x:t};return z.runKernelFunc(e,i,null,vu)}var Tg=U({selu_:Dx});function kx(n,t,e,i,r,a,s){a===void 0&&(a=[1,1]),s===void 0&&(s="NHWC");var o=O(n,"x","separableConv2d"),l=O(t,"depthwiseFilter","separableConv2d"),u=O(e,"pointwiseFilter","separableConv2d"),c=o,h=!1;if(o.rank===3&&(h=!0,c=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),s==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");E(c.rank===4,function(){return"Error in separableConv2d: input must be rank 4, but got "+("rank "+c.rank+".")}),E(l.rank===4,function(){return"Error in separableConv2d: depthwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.rank===4,function(){return"Error in separableConv2d: pointwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.shape[0]===1,function(){return"Error in separableConv2d: the first dimension of pointwise filter "+(" must be 1, but got "+u.shape[0]+".")}),E(u.shape[1]===1,function(){return"Error in separableConv2d: the second dimension of pointwise "+("filter must be 1, but got "+u.shape[1]+".")});var d=l.shape[2],p=l.shape[3];E(u.shape[2]===d*p,function(){return"Error in separableConv2d: the third dimension of pointwise filter "+("must be "+d*p+", ")+("but got "+u.shape[2]+".")});var f=Ta(c,l,i,r,s,a),m=1,g=Er(f,u,m,"valid",s);return h?V(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var Ng=U({separableConv2d_:kx});function Fx(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,l,h;return pe(this,function(d){switch(d.label){case 0:return e=O(n,"x","setdiff1d"),i=O(t,"y","setdiff1d"),E(e.dtype===i.dtype,function(){return"x and y should have the same dtype, but got x ("+e.dtype+") and y ("+i.dtype+")."}),E(e.rank===1,function(){return"x should be 1D tensor, but got x ("+e.shape+")."}),E(i.rank===1,function(){return"y should be 1D tensor, but got y ("+i.shape+")."}),[4,e.data()];case 1:return r=d.sent(),[4,i.data()];case 2:for(a=d.sent(),s=new Set(a),o=0,l=0;l<r.length;l++)s.has(r[l])||o++;for(u=new Ss([o],e.dtype),c=new Ss([o],"int32"),l=0,h=0;l<r.length;l++)s.has(r[l])||(u.values[h]=r[l],c.values[h]=l,h++);return[2,[u.toTensor(),c.toTensor()]]}})})}var Wx=Fx;function Ux(n){var t=O(n,"x","sign"),e={x:t};return z.runKernelFunc(function(i){return i.sign(t)},e,null,Su)}var xg=U({sign_:Ux});function Bx(n){var t=O(n,"x","sin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sin(t);return r([t]),a},e,null,bu)}var qc=U({sin_:Bx});function zx(n){var t=O(n,"x","sinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sinh(t);return r([t]),a},e,null,wu)}var Gc=U({sinh_:zx});function Px(n,t,e){var i=O(n,"x","slice1d");return E(i.rank===1,function(){return"slice1d expects a rank-1 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,[t],[e])}var _x=U({slice1d_:Px});function Mx(n,t,e){var i=O(n,"x","slice2d");return E(i.rank===2,function(){return"slice2d expects a rank-2 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Hx=U({slice2d_:Mx});function Vx(n,t,e){var i=O(n,"x","slice3d");return E(i.rank===3,function(){return"slice3d expects a rank-3 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var qx=U({slice3d_:Vx});function Gx(n,t,e){var i=O(n,"x","slice4d");return E(i.rank===4,function(){return"slice4d expects a rank-4 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Yx=U({slice4d_:Gx});function Kx(n,t){t===void 0&&(t=-1);var e=O(n,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and dim was "+t));var i={logits:e},r={dim:t};return z.runKernelFunc(function(a,s){var o=a.softmax(e,t);return s([o]),o},i,null,Cu,r)}var Cg=U({softmax_:Kx});function jx(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.fft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=n.as2D(r,i),s=e.fft(a);return s.reshape(n.shape)},t,null,Zp)}var qs=U({fft_:jx});function $x(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.ifft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=V(n,[r,i]),s=e.ifft(a);return V(s,n.shape)},t,null,rf)}var Oa=U({ifft_:$x});function Xx(n){var t=n.shape[n.shape.length-1],e=n.size/t,i;if(t<=2){var r=V(n,[e,t]);i=Oa(r)}else{var a=[e,2*(t-1)],s=V(Ca(n),[e,t]),o=V(Us(n),[e,t]),l=$n(ze(s,[0,1],[e,t-2]),1),u=J($n(ze(o,[0,1],[e,t-2]),1),we(-1)),c=xt([s,l],1),h=xt([o,u],1),r=V(ri(c,h),[a[0],a[1]]);i=Oa(r)}if(i=Ca(i),n.rank===3&&n.shape[0]!==0){var d=i,p=n.shape[0];i=V(i,[p,i.shape[0]/p,i.shape[1]]),d.dispose()}return i}var Yc=U({irfft_:Xx});function Rg(n,t,e){e===void 0&&(e=0);var i=[];if(typeof t=="number")E(n.shape[e]%t===0,function(){return"Number of splits must evenly divide the axis."}),i=new Array(t).fill(n.shape[e]/t);else{var r=t.reduce(function(o,l){return l===-1&&(o+=1),o},0);E(r<=1,function(){return"There should be only one negative value in split array."});var a=t.indexOf(-1);if(a!==-1){var s=t.reduce(function(o,l){return l>0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function Jx(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Qe(e,i.shape)[0],c=Rg(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,xu,s)}var Br=U({split_:Jx});function Zx(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&t<e){var a=n.shape.map(function(v){return 0}),s=n.shape.map(function(v){return v});s[n.shape.length-1]=t,r=ze(n,a,s),e=t}else if(t!=null&&t>e){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=xt([n,Kn(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(ri(r,l),[i,e]),c=qs(u),h=Math.floor(e/2)+1,d=Ca(c),p=Us(c),f=Br(d,[h,e-h],d.shape.length-1),m=Br(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(ri(f[0],m[0]),g)}var Gs=U({rfft_:Zx});function Qx(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,Au)}var _t=U({sqrt_:Qx});function eC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Ru,o)}var Ys=U({squaredDifference_:eC});function tC(n,t){var e=O(n,"x","squeeze");return V(e,Uf(e.shape,t).newShape)}var Ks=U({squeeze_:tC});function nC(n,t){t===void 0&&(t=0);var e=ba(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return mn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Be(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return mn(o,t)});return xt(s,t)}var ji=U({stack_:nC});function iC(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,zu,r)}var zr=U({step_:iC});function rC(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=Ns(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=Ns(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=Em(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var x=Ns(l);x.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=Lm(t,e,i),R=C.filter(function(W,F){return x.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,Cf,d)}var Og=U({stridedSlice_:rC});function aC(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,Eu)}var Eg=U({tan_:aC});function Ea(n,t,e){if(Fi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=Rn(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return ai(n,t,i,e)}function sC(n,t,e){if(Fi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=Rn(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function oC(n,t,e){if(Fi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=Rn(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function lC(n,t,e){if(Fi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=Rn(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,ai(n,t,i,e)}function uC(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,Rf,s),l=o[0],u=o[1];return{values:l,indices:u}}var Dg=U({topk_:uC});function cC(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new _c(t,e,i,!0,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var hC=U({truncatedNormal_:cC});function dC(n,t){t===void 0&&(t=0);var e=O(n,"x","unique",null);E(e.rank>0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(Of,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var kg=U({unique_:dC});function pC(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(rt(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,Uu,s)}var Kc=U({unsortedSegmentSum_:pC});function fC(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t<e.shape.length,function(){return"Axis = "+t+" is not in [-"+e.shape.length+", "+e.shape.length+")"}),t<0&&(t+=e.shape.length);var i={value:e},r={axis:t},a=function(s){return s.unstack(e,t)};return z.runKernelFunc(a,i,null,Wu,r)}var js=U({unstack_:fC});function mC(n,t,e,i){return t===void 0&&(t=!0),z.makeVariable(n,t,e,i)}function Fg(n,t){for(var e=[],i=0;i<t.length;i++)t[i]&&e.push(i);for(var r=On(n,"int32"),a=On([e.length,n.length],"int32"),i=0;i<e.length;i++){var s=r.indexToLoc(e[i]),o=i*n.length;a.values.set(s,o)}return a.toTensor()}function gC(n){return de(this,void 0,void 0,function(){var t,e,i;return pe(this,function(r){switch(r.label){case 0:return t=O(n,"condition","whereAsync","bool"),[4,t.data()];case 1:return e=r.sent(),i=Fg(t.shape,e),n!==t&&t.dispose(),[2,i]}})})}var Wg=gC;function vC(n,t,e){return de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:for(i=O(n,"tensor","boolMask"),r=O(t,"mask","boolMask","bool"),a=e??0,s=r.rank,o=i.shape,E(s>0,function(){return"mask cannot be scalar"}),Be(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u<a+s;u++)l*=o[u];return c=o.slice(0,a).concat([l],o.slice(a+s)),h=V(i,c),d=V(r,[-1]),[4,Wg(d)];case 1:return p=g.sent(),f=Ks(p,[1]),m=Ws(h,f,a),n!==i&&i.dispose(),t!==r&&r.dispose(),f.dispose(),h.dispose(),d.dispose(),p.dispose(),[2,m]}})})}var yC=vC;function bC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","notEqualStrict"),i=O(t,"b","notEqualStrict");return Be(e.shape,i.shape,"Error in notEqualStrict: "),Hs(e,i)}function wC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessStrict"),i=O(t,"b","lessStrict");return Be(e.shape,i.shape,"Error in lessStrict: "),Bs(e,i)}function SC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","equalStrict"),i=O(t,"b","equalStrict");return Be(e.shape,i.shape,"Error in equalStrict: "),Na(e,i)}function LC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessEqualStrict"),i=O(t,"b","lessEqualStrict");return Be(e.shape,i.shape,"Error in lessEqualStrict: "),Mi(e,i)}function IC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterStrict"),i=O(t,"b","greaterStrict");return Be(e.shape,i.shape,"Error in greaterStrict: "),hi(e,i)}function AC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterEqualStrict"),i=O(t,"b","greaterEqualStrict");return Be(e.shape,i.shape,"Error in greaterEqualStrict: "),_i(e,i)}var Ug=U({equalStrict_:SC}),Bg=U({greaterEqualStrict_:AC}),zg=U({greaterStrict_:IC}),Pg=U({lessEqualStrict_:LC}),_g=U({lessStrict_:wC}),Mg=U({notEqualStrict_:bC});function TC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","addStrict"),i=O(t,"b","addStrict");return Be(e.shape,i.shape,"Error in addStrict: "),fe(e,i)}function NC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","subStrict"),i=O(t,"b","subStrict");return Be(e.shape,i.shape,"Error in subStrict: "),ge(e,i)}function xC(n,t){return It("strict variants of ops have been deprecated and will be removed in future"),Be(n.shape,t.shape,"Error in powStrict: "),jn(n,t)}function CC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","mul"),i=O(t,"b","mul");return Be(e.shape,i.shape,"Error in multiplyStrict: "),J(e,i)}function RC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","div"),i=O(t,"b","div");return Be(e.shape,i.shape,"Error in divideStrict: "),Ie(e,i)}function OC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","modStrict"),i=O(t,"b","modStrict");return Be(e.shape,i.shape,"Error in modStrict: "),Uc(e,i)}function EC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","minimumStrict"),i=O(t,"b","minimumStrict");return Be(e.shape,i.shape,"Error in minimumStrict: "),Ms(e,i)}function DC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","maximumStrict"),i=O(t,"b","maximumStrict");return Be(e.shape,i.shape,"Error in maximumStrict: "),kr(e,i)}function kC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","squaredDifferenceStrict"),i=O(t,"b","squaredDifferenceStrict");return Be(e.shape,i.shape,"Error in squaredDifferenceStrict: "),Ys(e,i)}var Hg=U({addStrict_:TC}),Vg=U({divStrict_:RC}),qg=U({maximumStrict_:DC}),Gg=U({minimumStrict_:EC}),Yg=U({modStrict_:OC}),Kg=U({mulStrict_:CC}),jg=U({powStrict_:xC}),$g=U({squaredDifferenceStrict_:kC}),Xg=U({subStrict_:NC});function FC(n,t,e,i){t===void 0&&(t="euclidean"),e===void 0&&(e=null),i===void 0&&(i=!1),n=O(n,"x","norm");var r=Jg(n,t,e),a=r.shape;if(i){var s=Qe(e,n.shape);a=en(r.shape,s)}return V(r,a)}function Jg(n,t,e){if(e===void 0&&(e=null),n.rank===0)return Gt(n);if(n.rank!==1&&e===null)return Jg(V(n,[-1]),t,e);if(n.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return Ae(Gt(n),e);if(t===Infinity)return Vi(Gt(n),e);if(t===-Infinity)return _s(Gt(n),e);if(t==="euclidean"||t===2)return _t(Ae(jn(Gt(n),we(2,"int32")),e));throw new Error("Error in norm: invalid ord value: "+t)}if(Array.isArray(e)&&e.length===2){if(t===1)return Vi(Ae(Gt(n),e[0]),e[1]-1);if(t===Infinity)return Vi(Ae(Gt(n),e[1]),e[0]);if(t===-Infinity)return _s(Ae(Gt(n),e[1]),e[0]);if(t==="fro"||t==="euclidean")return _t(Ae(Ye(n),e));throw new Error("Error in norm: invalid ord value: "+t)}throw new Error("Error in norm: invalid axis: "+e)}var $s=U({norm_:FC});function WC(n,t,e,i,r){r===void 0&&(r=!0);var a=O(n,"v","movingAverage"),s=O(t,"x","movingAverage"),o=O(e,"decay","movingAverage");jf(a,s),E(ln(a.shape,s.shape),function(){return"Shape mismatch in v and x"});var l=we(1),u=ge(l,o),c=J(ge(s,a),u);if(r){E(i!=null,function(){return"When using zeroDebias: true, step is required."});var h=O(i,"step","movingAverage");c=Ie(c,ge(l,jn(o,h)))}return fe(a,c)}var UC=U({movingAverage_:WC});function BC(n,t,e){var i=O(n,"indices","scatterND","int32"),r=O(t,"updates","scatterND");mc(r,i,e);var a=function(l){return l.scatterND(i,r,e)},s={indices:i,updates:r},o={shape:e};return z.runKernelFunc(a,s,null,Tf,o)}var zC=U({scatterND_:BC});function PC(n,t,e,i){if(n.dtype!=="int32")throw new Error("tf.sparseToDense() expects the indices to be int32 type,"+(" but the dtype was "+n.dtype+"."));if(n.rank>2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function _C(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);PC(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,xf,l)}var MC=U({sparseToDense_:_C});function HC(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,tf)}var VC=U({gatherND_:HC});function qC(n,t){if(t==null)return n.shape.slice();if(ln(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i<n.shape.length;i++)t[i]==null&&n.shape[i]!=null?e.push(n.shape[i]):e.push(t[i]);return e}return t}function GC(n,t,e,i){var r=O(n,"x","dropout");if(E(r.dtype==="float32",function(){return"x has to be a floating point tensor since it's going to be "+("scaled, but got a "+r.dtype+" tensor instead.")}),E(t>=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=qC(r,e),s=1-t,o=Ie(Fs(fe(Lg(a,0,1,"float32",i),s)),s);return J(r,o)}var YC=U({dropout_:GC});function Zg(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function jc(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a<n;++a){var s=2*Math.PI*a/(n+i-1);r[a]=t-e*Math.cos(s)}return Ur(r,"float32")}function KC(n,t,e){return e===void 0&&(e=1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,g;return pe(this,function(v){switch(v.label){case 0:return i=O(n,"predictions","inTopK"),r=O(t,"targets","inTopK"),E(i.rank>1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Be(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=gs("bool",u),d=0;d<u;d++){for(p=d*c,f=s.subarray(p,p+c),m=[],g=0;g<f.length;g++)m.push({value:f[g],index:g});for(m.sort(function(b,w){return w.value-b.value}),h[d]=0,g=0;g<e;g++)if(m[g].index===o[d]){h[d]=1;break}}return n!==i&&i.dispose(),t!==r&&r.dispose(),[2,si(h,r.shape,"bool")]}})})}var jC=KC;function $C(n,t,e,i,r,a,s){a===void 0&&(a="NHWC");var o=n;n.rank===3&&(o=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var l=t;l.rank===3&&(l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]])),E(o.rank===4,function(){return"Error in conv2dDerFilter: input must be rank 4, but got shape "+(o.shape+".")}),E(l.rank===4,function(){return"Error in conv2dDerFilter: dy must be rank 4, but got shape "+(l.shape+".")}),E(e.length===4,function(){return"Error in conv2dDerFilter: filterShape must be length 4, but got "+(e+".")});var u=a==="NHWC"?o.shape[3]:o.shape[1],c=a==="NHWC"?l.shape[3]:l.shape[1];E(u===e[2],function(){return"Error in conv2dDerFilter: depth of input "+u+") must "+("match input depth in filter ("+e[2]+".")}),E(c===e[3],function(){return"Error in conv2dDerFilter: depth of dy ("+c+") must "+("match output depth for filter ("+e[3]+").")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerFilter: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var h=function(f){var m=1,g=Sa(a),v=Dn(o.shape,e,i,m,r,s,!1,g);return f.conv2dDerFilter(o,l,v)},d={x:o,dy:l},p={strides:i,pad:r,dataFormat:a,dimRoundingMode:s};return z.runKernelFunc(h,d,null,_p,p)}var $c=U({conv2DBackpropFilter_:$C});function Xs(n,t,e){if(e==null||e==="linear")return n;if(e==="relu")return J(n,zr(t));throw new Error("Cannot compute gradient for fused activation "+e+".")}function Js(n,t){var e=t,i=mt(n.shape,t.shape);return i.length>0&&(e=Ae(e,i)),V(e,n.shape)}function Zs(n,t,e){if(t==="linear")return n;if(t==="relu")return Ra(n);if(t==="elu")return Cc(n);if(t==="relu6")return Hc(n);if(t==="prelu")return zc(n,e);throw new Error("Unknown fused activation "+t+".")}var Qs=function(n,t){var e=n>0;return!e||t==="linear"};function XC(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",Qs(z.state.gradientDepth,d)===!1){var f=Er(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(rt(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(Pt(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Dn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var x=function(F,P){var H=P,_=H[0],K=H[1],j=H[2],q=H[3],G=Xs(F,j,d);E(ci(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=Tc(K.shape,G,_,i,r),X=$c(K,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=Js(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Mu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Mu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var JC=U({fusedConv2d_:XC});function ZC(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,Gp)}var Qg=U({depthwiseConv2dNativeBackpropFilter_:ZC});function QC(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Yp);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var ev=U({depthwiseConv2dNativeBackpropInput_:QC});function eR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(Qs(z.state.gradientDepth,d)===!1){var f=Ta(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(Pt(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(rt(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Dn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var x=function(F,P){E(ci(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],K=P[2],j=P[3],q=Xs(F,K,d),G=ev(_.shape,q,H,w),Z=Qg(_,q,H.shape,w);if(j!=null){var X=Js(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Hu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Hu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var tR=U({fusedDepthwiseConv2d_:eR});function nR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(Qs(z.state.gradientDepth,c)===!1){var d=We(e,i,a,o);return l!=null&&(d=fe(d,l)),Zs(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=at(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=ot(w),x=ot(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(ln(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[x,b,g]):V(f,[x,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=at(k,p)[0],et(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=Xs(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=We(ie,X,!1,!0),re=We(Z,ie,!0,!1)):!a&&o?(te=We(ie,X,!1,!1),re=We(ie,Z,!0,!1)):a&&!o?(te=We(X,ie,!1,!0),re=We(Z,ie,!1,!1)):(te=We(X,ie,!0,!0),re=We(ie,Z,!0,!0)),l!=null){var le=Js(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var K=kn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,_u,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return K(R,D)}else{var j=kn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,_u,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return j(R,D,k)}}var iR=U({fusedMatMul_:nR});var rR={__proto__:null,conv2d:JC,depthwiseConv2d:tR,matMul:iR};function aR(n){return jc(n,.54,.46)}var sR=U({hammingWindow_:aR});function oR(n){return jc(n,.5,.5)}var tv=U({hannWindow_:oR});function lR(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a<n.size;){var o=a+t-n.size,l=xt([ze(n,a,t-o),Rc([o],r)]);s.push(l),a+=e}return s.length===0?Ea([],[0,t]):V(xt(s),[s.length,t])}var nv=U({frame_:lR});function uR(n,t,e,i,r){r===void 0&&(r=tv),i==null&&(i=Zg(t));for(var a=nv(n,t,e),s=J(a,r(t)),o=[],l=0;l<a.shape[0];l++)o.push(Gs(ze(s,[l,0],[1,t]),i));return xt(o)}var cR=U({stft_:uR});function hR(n,t,e,i,r,a){var s=O(n,"image","cropAndResize"),o=O(t,"boxes","cropAndResize","float32"),l=O(e,"boxInd","cropAndResize","int32");r=r||"bilinear",a=a||0;var u=o.shape[0];E(s.rank===4,function(){return"Error in cropAndResize: image must be rank 4,"+("but got rank "+s.rank+".")}),E(o.rank===2&&o.shape[1]===4,function(){return"Error in cropAndResize: boxes must be have size ["+u+",4] "+("but had shape "+o.shape+".")}),E(l.rank===1&&l.shape[0]===u,function(){return"Error in cropAndResize: boxInd must be have size ["+u+"] "+("but had shape "+o.shape+".")}),E(i.length===2,function(){return"Error in cropAndResize: cropSize must be of length 2, but got "+("length "+i.length+".")}),E(i[0]>=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Vp,d);return p}var dR=U({cropAndResize_:hR});function pR(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(ef,e,{});return i}var fR=U({flipLeftRight_:pR});function mR(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(Ef,a,s);return o}var gR=U({rotateWithOffset_:mR});function Pr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function vR(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Pr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,vf,l)}var yR=U({nonMaxSuppression_:vR});function wR(n,t,e){var i=bR(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function bR(n,t,e){return LR(n,t,e||SR)}function SR(n,t){return n>t?1:n<t?-1:0}function LR(n,t,e){for(var i=0,r=n.length,a=0,s=!1;i<r;){a=i+(r-i>>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function iv(n,t,e,i,r){return Xc(n,t,e,i,r,0).selectedIndices}function rv(n,t,e,i,r,a){return Xc(n,t,e,i,r,0,!1,a,!0)}function av(n,t,e,i,r,a){return Xc(n,t,e,i,r,a,!0)}function Xc(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;c<t.length;c++)t[c]>r&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(sv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length<e&&u.length>0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m<r)break;for(var b=!1,w=d.length-1;w>=v;--w){var S=IR(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*AR(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&wR(u,f,sv))}var L=d.length,x=e-L;o&&x>0&&(d.push.apply(d,new Array(x).fill(0)),p.push.apply(p,new Array(x).fill(0)));var C={selectedIndices:Ur(d,"int32")};return s&&(C.selectedScores=Ur(p,"float32")),l&&(C.validOutputs=we(L,"int32")),C}function IR(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function AR(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function sv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function TR(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Pr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=iv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var NR=TR;function xR(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(bf,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var CR=U({nonMaxSuppressionWithScore_:xR});function RR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=av(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var OR=RR;function ER(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(yf,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var DR=U({nonMaxSuppressionPadded_:ER});function kR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=rv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var FR=kR;function WR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,hu,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var ov=U({resizeBilinear_:WR});function UR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,cu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var lv=U({resizeNearestNeighbor_:UR});function BR(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(Mc(0,s,1,"int32"),[-1,1]),u=Mc(0,o,1,"int32"),c=ge(l,u),h=qi(Mi(c,we(+t,"int32")),_i(c,we(-e,"int32"))),d=Kn([s,o],i.dtype);return V(ji(js(V(i,[-1,s,o])).map(function(p){return pn(h,p,d)})),r)}var zR=U({bandPart_:BR});function PR(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r<n.length;++r)i(r)}else t=!0,n=Br(n,n.shape[0],0).map(function(l){return Ks(l,[0])});E(n.length<=n[0].shape[0],function(){return"Gram-Schmidt: Number of vectors ("+n.length+") exceeds "+("number of dimensions ("+n[0].shape[0]+").")});for(var a=[],s=n,o=function(l){a.push(z.tidy(function(){var u=s[l];if(l>0)for(var c=0;c<l;++c){var h=J(Ae(J(a[c],u)),a[c]);u=ge(u,h)}return Ie(u,$s(u,"euclidean"))}))},r=0;r<n.length;++r)o(r);return t?ji(a,0):a}var _R=U({gramSchmidt_:PR});function MR(n,t){if(t===void 0&&(t=!1),E(n.rank>=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return uv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=js(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=uv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(ji(r,0),n.shape),o=V(ji(a,0),n.shape);return[s,o]}function uv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=ug(e),a=Bi(n),s=Ea([[1]],[1,1]),o=Bi(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=$s(g),b=ze(a,[h,h],[1,1]),w=pn(hi(b,0),Ea([[-1]]),Ea([[1]])),S=ge(b,J(w,v)),L=Ie(g,S);L.shape[0]===1?o=Bi(s):o=xt([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var x=gt(Ie(We(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(x,o),D=ut(o);if(h===0)a=ge(C,We(R,We(D,C)));else{var k=ge(C,We(R,We(D,C)));a=xt([ze(a,[0,0],[h,i]),k],0)}var W=ut(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,We(We(F,o),W));else{var P=ge(F,We(We(F,o),W));r=xt([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],zt([p,f,m])},c=0;c<l;++c)u(c);return!t&&e>i&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var HR=U({qr_:MR});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function VR(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ae(a);if(e===I.Reduction.MEAN){if(r==null)return xa(a);var s=i.size/r.size,o=Ie(Ae(a),Ae(r));return s>1?Ie(o,we(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Ie(Ae(a),we(i.size));var l=J(r,Fr(i.shape)),u=ue(Ae(Hs(l,we(0))),"float32");return Ie(Ae(a),u)}throw Error("Unknown reduction: "+e)}var Xn=U({computeWeightedLoss_:VR});function qR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Be(r.shape,a.shape,"Error in absoluteDifference: ");var o=Gt(ge(r,a));return Xn(o,s,i)}var GR=U({absoluteDifference_:qR});function YR(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Be(a.shape,s.shape,"Error in cosineDistance: ");var l=we(1),u=ge(l,Ae(J(a,s),e,!0));return Xn(u,o,r)}var KR=U({cosineDistance_:YR});function jR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Be(r.shape,a.shape,"Error in hingeLoss: ");var o=we(1);r=ge(J(we(2),r),o);var l=Ra(ge(o,J(r,a)));return Xn(l,s,i)}var $R=U({hingeLoss_:jR});function XR(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Be(a.shape,s.shape,"Error in huberLoss: ");var l=we(i),u=Gt(ge(s,a)),c=Ms(u,l),h=ge(u,c),d=fe(J(we(.5),Ye(c)),J(l,h));return Xn(d,o,r)}var JR=U({huberLoss_:XR});function ZR(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Be(a.shape,s.shape,"Error in logLoss: ");var l=we(1),u=we(i),c=gt(J(a,Hi(fe(s,u)))),h=J(ge(l,a),Hi(fe(ge(l,s),u))),d=ge(c,h);return Xn(d,o,r)}var QR=U({logLoss_:ZR});function eO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Be(r.shape,a.shape,"Error in meanSquaredError: ");var o=Ys(r,a);return Xn(o,s,i)}var tO=U({meanSquaredError_:eO});function nO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Be(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ra(i),a=J(i,e),s=Ec(fn(gt(Gt(i))));return fe(ge(r,a),s)}function iO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Be(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=nO(a,s);return Xn(h,o,r)}var rO=U({sigmoidCrossEntropy_:iO});function aO(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=kn(function(r,a,s){var o=!0,l=kc(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=gt(J(u,r)),h=Ae(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=en(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),fn(g))),J(V(p,v),ge(fn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function sO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Be(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(a.shape[1]);a=fe(J(a,ge(u,l)),Ie(l,c))}var h=aO(a,s);return Xn(h,o,r)}var oO=U({softmaxCrossEntropy_:sO});var lO={fft:qs,ifft:Oa,rfft:Gs,irfft:Yc},uO={hammingWindow:sR,hannWindow:tv,frame:nv,stft:cR},cO={flipLeftRight:fR,resizeNearestNeighbor:lv,resizeBilinear:ov,rotateWithOffset:gR,cropAndResize:dR,nonMaxSuppression:yR,nonMaxSuppressionAsync:NR,nonMaxSuppressionWithScore:CR,nonMaxSuppressionWithScoreAsync:OR,nonMaxSuppressionPadded:DR,nonMaxSuppressionPaddedAsync:FR},hO={bandPart:zR,gramSchmidt:_R,qr:HR},dO={absoluteDifference:GR,computeWeightedLoss:Xn,cosineDistance:KR,hingeLoss:$R,huberLoss:JR,logLoss:QR,meanSquaredError:tO,sigmoidCrossEntropy:rO,softmaxCrossEntropy:oO};var di=function(n){qn(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return zt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return gg(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&zt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:we(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(km);Object.defineProperty(di,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Jc=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;ft(function(){var d=fe(J(c,i.rho),J(Ye(u),1-i.rho)),p=J(Ie(_t(fe(h,i.epsilon)),_t(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ye(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(zt(this.accumulatedGrads.map(function(e){return e.variable})),zt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(di);ui(Jc);var Zc=function(n){qn(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:ft(function(){return Rc(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;ft(function(){var h=fe(c,Ye(u));c.assign(h);var d=fe(J(Ie(u,_t(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&zt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(di);ui(Zc);var Qc=function(n){qn(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],ft(function(){s.accBeta1=we(i).variable(),s.accBeta2=we(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:ft(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:ft(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ye(h),1-i.beta2)),g=Ie(f,a),v=Ie(m,s);d.assign(f),p.assign(m);var b=fe(J(Ie(g,fe(_t(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&zt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),ft(function(){a.accBeta1.assign(jn(a.beta1,a.iterations_+1)),a.accBeta2.assign(jn(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(di);ui(Qc);var eh=function(n){qn(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],ft(function(){o.iteration=we(0).variable(),o.accBeta1=we(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=Ie(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=Gt(h),v=kr(m,g);d.assign(f),p.assign(v);var b=fe(J(Ie(s,a),Ie(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&zt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(di);ui(eh);var eo=function(n){qn(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];ft(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=Um(we(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(di);ui(eo);var th=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=we(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;ft(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&zt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(eo);ui(th);var nh=function(n){qn(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;ft(function(){var d=fe(J(c,i.decay),J(Ye(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Ie(J(u,i.learningRate),_t(ge(d,fe(Ye(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ye(u),1-i.decay)),g=fe(J(h,i.momentum),Ie(J(u,i.learningRate),_t(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&zt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&zt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&zt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(di);ui(nh);var $i=function(){function n(){}return n.sgd=function(t){return new eo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new th(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new nh(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new Qc(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Jc(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new eh(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Zc(t,e)},n}();var pO={sgd:$i.sgd,momentum:$i.momentum,adadelta:$i.adadelta,adagrad:$i.adagrad,rmsprop:$i.rmsprop,adamax:$i.adamax,adam:$i.adam};var fO=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function mO(){return new Promise(function(n){return fO(function(){return n()})})}function gO(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function vO(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s<a;++s)r=r.concat([n[s+1]/t[s],t[s]]);r=r.concat(n.slice(a+1))}return r}function yO(n,t,e){e===void 0&&(e=!0);var i=[];if(e){i.push(t);for(var r=t+1;r<n;++r)r<=2*t?(i.push(r),i.push(r-(t+1))):i.push(r)}else{for(var a=[],s=[],r=1;r<n;++r)r>=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function bO(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a<n.length;++a)a<=t.length?i?r.push(t[a-1]*n[a]):r.push(n[a]/t[a-1]):r.push(n[a]);return r}function wO(n,t){for(var e=[0],i=0;i<t;++i)e.push(n[i][0]);return e}function SO(n,t,e){for(var i=n.slice(0,1),r=0;r<e;++r)i.push(n[r+1]-t[r][0]-t[r][1]);return i}var cv=1.7580993408473768,hv=1.0507009873554805;var LO=.3275911,IO=.254829592,AO=-.284496736,TO=1.421413741,NO=-1.453152027,xO=1.061405429;function CO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.warn.apply(console,n)}function RO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.log.apply(console,n)}function OO(n,t){if(n.length!==t.length)throw new Error("Cannot merge real and imag arrays of different lengths. real:"+(n.length+", imag: "+t.length+"."));for(var e=new Float32Array(n.length*2),i=0;i<e.length;i+=2)e[i]=n[i/2],e[i+1]=t[i/2];return e}function EO(n){for(var t=new Float32Array(n.length/2),e=new Float32Array(n.length/2),i=0;i<n.length;i+=2)t[i/2]=n[i],e[i/2]=n[i+1];return{real:t,imag:e}}function DO(n){for(var t=Math.ceil(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=0;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function kO(n){for(var t=Math.floor(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=2;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function FO(n,t){var e=n[t*2],i=n[t*2+1];return{real:e,imag:i}}function WO(n,t,e,i){n[i*2]=t,n[i*2+1]=e}function UO(n,t){for(var e=new Float32Array(n/2),i=new Float32Array(n/2),r=0;r<Math.ceil(n/2);r++){var a=(t?2:-2)*Math.PI*(r/n);e[r]=Math.cos(a),i[r]=Math.sin(a)}return{real:e,imag:i}}function BO(n,t,e){var i=(e?2:-2)*Math.PI*(n/t),r=Math.cos(i),a=Math.sin(i);return{real:r,imag:a}}function zO(n,t,e){if(t==="complex64"){if(n.dtype==="complex64")return n.clone();var i=Kn(n.shape),r=ue(n,"float32"),a=e.complex(r,i);return i.dispose(),r.dispose(),a}if(!_f(n.dtype,t))return z.makeTensorFromDataId(n.dataId,n.shape,t);if(n.dtype==="complex64"){var s=e.real(n),a=ue(s,t);return s.dispose(),a}if(t==="int32")return e.int(n);if(t==="bool"){var o=we(0,n.dtype),a=e.notEqual(n,o);return o.dispose(),a}else throw new Error("Error in Cast: failed to cast "+n.dtype+" to "+t)}function PO(n,t){return z.makeTensorFromDataId(n.dataId,t,n.dtype)}function _O(n,t,e){var i=(t-n)/(e-1),r=Lr(e,"float32");r[0]=n;for(var a=1;a<r.length;a++)r[a]=r[a-1]+i;return Ur(r,"float32")}var MO={__proto__:null,slice_util:Dm,segment_util:M1,castTensor:zO,reshapeTensor:PO,linspaceImpl:_O,upcastType:Ls,axesAreInnerMostDims:Sc,combineLocations:Pm,computeOutAndReduceShapes:_m,expandShapeToKeepDim:en,assertAxesAreInnerMostDims:vT,getAxesPermutation:tn,getUndoAxesPermutation:xs,getInnerMostAxes:En,getBroadcastDims:A1,getReductionAxes:mt,assertAndGetBroadcastShape:et,assertParamsConsistent:Xm,computeOutShape:Jm,computeDilation2DInfo:xT,computePool2DInfo:Rr,computePool3DInfo:Ia,computeConv2DInfo:Dn,computeConv3DInfo:La,computeDefaultPad:Ic,tupleValuesAreOne:ci,eitherStridesOrDilationsAreOne:Pt,convertConv2DDataFormat:Sa,getFusedDyActivation:Xs,getFusedBiasGradient:Js,applyActivation:Zs,shouldFuse:Qs,PARALLELIZE_THRESHOLD:Oc,computeOptimalWindowSize:z1,getImageCenter:gO,getReshaped:vO,getPermuted:yO,getReshapedPermuted:bO,getSliceBeginCoords:wO,getSliceSize:SO,prepareAndValidate:bm,validateUpdateShape:fc,validateInput:mc,calculateShapes:wm,SELU_SCALEALPHA:cv,SELU_SCALE:hv,ERF_P:LO,ERF_A1:IO,ERF_A2:AO,ERF_A3:TO,ERF_A4:NO,ERF_A5:xO,warn:CO,log:RO,mergeRealAndImagArrays:OO,splitRealAndImagArrays:EO,complexWithEvenIndex:DO,complexWithOddIndex:kO,getComplexWithIndex:FO,assignToTypedArray:WO,exponents:UO,exponent:BO,prepareSplitSize:Rg};function HO(n,t,e){var i=new Array(n.rank).fill(0),r=n.shape.slice();return t.map(function(a){var s=r.slice();s[e]=a;var o=ze(n,i,s);return i[e]+=a,o})}function VO(n,t){for(var e=new Array(n.rank),i=0;i<e.length;i++)e[i]=n.shape[i]*t[i];for(var r=On(e,n.dtype),i=0;i<r.values.length;++i){for(var a=r.indexToLoc(i),s=new Array(n.rank),o=0;o<s.length;o++)s[o]=a[o]%n.shape[o];var l=n.locToIndex(s);r.values[i]=n.values[l]}return r.toTensor()}function qO(n,t,e,i,r){for(var a=t[t.length-1],s=[n.length/a,a],o=s[0],l=s[1],u=gs(e,o*i),c=gs("int32",o*i),h=0;h<o;h++){for(var d=h*l,p=n.subarray(d,d+l),f=[],m=0;m<p.length;m++)f.push({value:p[m],index:m});f.sort(function(S,L){return L.value-S.value});for(var g=h*i,v=u.subarray(g,g+i),b=c.subarray(g,g+i),m=0;m<i;m++)v[m]=f[m].value,b[m]=f[m].index}var w=t.slice();return w[w.length-1]=i,[si(u,w,e),si(c,w,"int32")]}var GO={__proto__:null,nonMaxSuppressionV3Impl:iv,nonMaxSuppressionV4Impl:rv,nonMaxSuppressionV5Impl:av,split:HO,tile:VO,topkImpl:qO,whereImpl:Fg};var YO={kernelName:nl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,zr(ue(e,"float32"),-1))}}}};var KO={kernelName:il,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Ye(ue(e,"float32")),r=_t(ge(we(1),i));return gt(Ie(n,r))}}}};var jO={kernelName:rl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(ge(Ye(ue(e,"float32")),1));return Ie(n,i)}}}};var $O={kernelName:ps,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,i.shape)};return{a,b:s}}};var XO={kernelName:al,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var JO={kernelName:sl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var ZO={kernelName:ol,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var QO={kernelName:ll,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,_t(ge(we(1),Ye(ue(e,"float32")))))}}}};var eE={kernelName:ul,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(fe(we(1),Ye(ue(e,"float32"))));return Ie(n,i)}}}};var tE={kernelName:dl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=fe(Ye(e),Ye(i)),l=J(n,Ie(i,o)),u=mt(e.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,e.shape)},s=function(){var o=fe(Ye(e),Ye(i)),l=gt(J(n,Ie(e,o))),u=mt(i.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,i.shape)};return{a,b:s}}};var nE={kernelName:cl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(Ye(ue(e,"float32")),1))}}}};var iE={kernelName:hl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ge(we(1),Ye(ue(e,"float32"))))}}}};function rE(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(Pt(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(rt(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=Ia(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,zp,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var aE=U({avgPool3dBackprop_:rE});var sE={kernelName:fl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return aE(n,i,a,s,c,l,u)}}}};function oE(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=Rr(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,Bp,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var lE=U({avgPoolBackprop_:oE});var uE={kernelName:pl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return lE(n,i,a,s,o)}}}};var cE={kernelName:ml,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return We(n,a,!1,!0)},b:function(){return We(r,n,!0,!1)}}:!o&&l?{a:function(){return We(n,a,!1,!1)},b:function(){return We(n,r,!0,!1)}}:o&&!l?{a:function(){return We(a,n,!1,!0)},b:function(){return We(r,n,!1,!1)}}:{a:function(){return We(a,n,!0,!0)},b:function(){return We(n,r,!0,!0)}}}};var hE={kernelName:gl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return Vs(n,r,a)}}}};var dE={kernelName:vl,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o<s.length;o++)s[o]>1&&l.push(o);return{x:function(){return Ae(n,l,!0)}}}};var pE={kernelName:fs,gradFunc:function(n){return{x:function(){return n.clone()}}}};var fE={kernelName:yl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var mE={kernelName:bl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return pn(qi(_i(i,a),Mi(i,s)),n,Ee(n))}}}};var gE={kernelName:wl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Qe(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=Br(n,s,a);return o.map(function(l){return function(){return l}})}};var vE={kernelName:Sl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(ci(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return Tc(r.shape,n,a,l,u,c)},filter:function(){return $c(r,n,a.shape,l,u,c)}}}};var yE={kernelName:Ll,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Er(n,a,o,l,u,1,c)},filter:function(){return $c(n,r,a.shape,o,l,u,c)}}}};function bE(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=La(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,Mp,u)}var wE=U({conv3DBackpropFilter_:bE});var SE={kernelName:Il,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(ci(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return ng(o.shape,n,l,a,s)},filter:function(){return wE(o,n,l.shape,a,s)}}}};var LE={kernelName:Al,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(gt(qc(ue(e,"float32"))),n)}}}};var IE={kernelName:Tl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Gc(ue(e,"float32")),n)}}}};var AE={kernelName:Nl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=tn([a],i.rank),u=xc(n,a,s,!o);return l!=null&&(u=ut(u,l)),u}}}};var TE={kernelName:xl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(ci(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(Pt(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be  1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(rt(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Dn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return ev(c.shape,n,h,d)},filter:function(){return Qg(c,n,h.shape,d)}}}};var NE={kernelName:Cl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(jp,s,e)},filter:function(){return z.runKernel($p,o,e)}}}};var xE={kernelName:Rl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var CE={kernelName:Ol,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Xp)}}}};var RE={kernelName:El,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(fn(gt(Ye(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var OE={kernelName:Dl,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var EE={kernelName:kl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,fn(e))}}}};var DE={kernelName:Fl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var kE={kernelName:Wl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var FE={kernelName:Ul,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??we(1),u=mt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h<r.shape.length-1;++h)c.push(r.shape[h]);c.push(1)}var d=ge(r,a),p=J(n,l),f=Vc(fe(s,we(i))),m=J(J(J(f,f),f),we(-.5)),g=function(){return a.rank===1?V(J(J(n,Dr(V(f,[1,1,1,a.shape[0]]),c)),l),r.shape):V(J(J(n,f),l),r.shape)},v=function(){var L=J(J(f,we(-1)),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},b=function(){var L=J(J(m,d),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},w=function(){var L=J(d,f),x=J(n,L);return a.rank===1&&(x=Ae(x,u)),V(x,a.shape)},S=function(){var L=n;return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)};return{x:g,mean:v,variance:b,scale:w,offset:S}}};var WE={kernelName:Bl,inputsToSave:["x","indices"],gradFunc:function(n,t,e){var i=t[0],r=t[1],a=e.axis,s=Qe(a,i.shape)[0],o=function(){var l=i.shape,u=r.size,c=l.slice(0,s),h=c.length,d=l.slice(a,l.length).slice(1),p=d.length,f=dv(0,h),m=dv(h+1,h+1+p),g=pv([c,[u],d]),v=V(n,g),b=V(r,[u]),w=pv([[h],f,m]),S=ut(v,w),L=Kc(S,b,i.shape[s]),x=xs(w);return L=ut(L,x),L};return{x:o,indices:function(){return r}}}};function dv(n,t){for(var e=[],i=n;i<t;++i)e.push(i);return e}function pv(n){for(var t=[],e=0;e<n.length;++e)for(var i=0;i<n[e].length;++i)t.push(n[e][i]);return t}var UE={kernelName:zl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1];return{a:function(){return Ee(e)},b:function(){return Ee(i)}}}};var BE={kernelName:Pl,gradFunc:function(n){return{x:function(){return ue(n,"float32")}}}};var zE={kernelName:_l,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var PE={kernelName:Ml,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var _E={kernelName:Hl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ME={kernelName:ql,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(e,1))}}}};var HE={kernelName:Vl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ue(e,"float32"))}}}};var VE={kernelName:Gl,inputsToSave:[],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.axis;return{logits:function(){var a=!0,s=fn(i);return ge(n,J(Ae(n,r,a),s))}}}};function qE(n,t,e,i,r,a,s){i===void 0&&(i=5),r===void 0&&(r=1),a===void 0&&(a=1),s===void 0&&(s=.5);var o=function(c){return c.LRNGrad(e,n,t,i,r,a,s)},l={x:n,y:t,dy:e},u={depthRadius:i,bias:r,alpha:a,beta:s};return z.runKernelFunc(o,l,null,df,u)}var GE=U({localResponseNormalizationBackprop_:qE});var YE={kernelName:Yl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.depthRadius,l=s.bias,u=s.alpha,c=s.beta;return{x:function(){return GE(r,a,n,o,l,u,c)}}}};function fv(n,t,e,i,r){return t.rank<e.rank&&(t=V(t,en(t.shape,i))),n.rank<e.rank&&(n=V(n,en(n.shape,i))),{x:function(){var a=J(n,ue(Na(e,t),n.dtype));return r==null?a:ut(a,r)}}}var mv={kernelName:Kl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.reductionIndices,a=t[0],s=t[1],o=Qe(r,a.shape),l=tn(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var KE={kernelName:jl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(_i(e,i),"float32"))},a=function(){return J(n,ue(Bs(e,i),"float32"))};return{a:r,b:a}}};function jE(n,t,e,i,r,a,s,o){a===void 0&&(a=[1,1,1]);var l=O(n,"dy","maxPool3dBackprop"),u=O(t,"input","maxPool3dBackprop"),c=O(e,"output","maxPool3dBackprop"),h=l,d=u,p=c,f=!1;u.rank===4&&(f=!0,h=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=V(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),p=V(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),E(h.rank===5,function(){return"Error in maxPool3dBackprop: dy must be rank 5 but got rank "+(h.rank+".")}),E(d.rank===5,function(){return"Error in maxPool3dBackprop: input must be rank 5 but got rank "+(d.rank+".")}),E(p.rank===5,function(){return"Error in maxPool3dBackprop: output must be rank 5 but got rank "+(p.rank+".")}),E(Pt(r,a),function(){return"Error in maxPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+r+" and dilations '"+a+"'")}),o!=null&&E(rt(s),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+o+" but got pad "+s+".")});var m=function(w){var S=Ia(d.shape,i,r,a,s,o);return w.maxPool3dBackprop(h,d,p,S)},g={dy:h,input:d,output:p},v={filterSize:i,strides:r,dilations:a,pad:s,dimRoundingMode:o},b=z.runKernelFunc(m,g,null,ff,v);return f?V(b,[b.shape[1],b.shape[2],b.shape[3],b.shape[4]]):b}var $E=U({maxPool3dBackprop_:jE});var XE={kernelName:Xl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.dilations,c=s.pad,h=s.dimRoundingMode,d=u??[1,1,1];return{x:function(){return $E(n,r,a,o,l,d,c,h)}}}};function JE(n,t,e,i,r,a,s){var o=O(n,"dy","maxPoolBackprop"),l=O(t,"input","maxPoolBackprop"),u=O(e,"output","maxPoolBackprop");E(l.rank===o.rank,function(){return"Rank of input ("+l.rank+") does not match rank of dy "+("("+o.rank+")")}),E(o.rank===4,function(){return"Error in maxPoolBackprop: dy must be rank 4 but got rank "+(o.rank+".")}),E(l.rank===4,function(){return"Error in maxPoolBackprop: input must be rank 4 but got rank "+(l.rank+".")}),s!=null&&E(rt(a),function(){return"Error in maxPoolBackprop: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+a+".")});var c=function(p){var f=Rr(l.shape,i,r,1,a,s);return p.maxPoolBackprop(o,l,u,f)},h={dy:o,input:l,output:u},d={filterSize:i,strides:r,pad:a,dimRoundingMode:s};return z.runKernelFunc(c,h,null,pf,d)}var ZE=U({maxPoolBackprop_:JE});var QE={kernelName:$l,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.pad;return{x:function(){return ZE(n,r,a,o,l,u)}}}};var eD={kernelName:Jl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.axis,a=t[0],s=t[1],o=Qe(r,a.shape),l=tn(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var tD={kernelName:Zl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Mi(e,i),"float32"))},a=function(){return J(n,ue(hi(e,i),"float32"))};return{a:r,b:a}}};var nD={kernelName:Ql,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=mt(e.shape,r);return o.length>0?V(Ae(n,o),e.shape):n},s=function(){var o=J(n,gt(Fs(Ie(e,i)))),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var iD={kernelName:eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var rD={kernelName:tu,gradFunc:function(n){return{x:function(){return gt(n)}}}};var aD={kernelName:iu,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return Kn(e.shape,"float32")}}}};var sD={kernelName:nu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var gv={kernelName:ru,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var oD={kernelName:au,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=et(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,jn(a,ge(c,we(1))))),d=mt(a.shape,o);return d.length>0&&(h=Ae(h,d)),V(h,a.shape)},u=function(){var c=hi(a,0),h=pn(c,Hi(a),Ee(a)),d=J(n,J(r,h)),p=mt(s.shape,o);return p.length>0&&(d=Ae(d,p)),V(d,s.shape)};return{a:l,b:u}}};var lD={kernelName:su,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=hi(e,0);return{x:function(){return pn(r,n,J(n,i))},alpha:function(){var a=pn(r,Ee(n),J(n,e)),s=mt(i.shape,n.shape);return s.length>0&&(a=Ae(a,s)),V(a,i.shape)}}}};var uD={kernelName:ou,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,gt(Ye(e)))}}}};var cD={kernelName:du,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(Mi(e,6),zr(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var hD={kernelName:lu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue(zr(e),"float32"))}}}};var dD={kernelName:uu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var pD={kernelName:hu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,Af,e)};return{images:s}}};var fD={kernelName:cu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,If,e)};return{images:s}}};var mD={kernelName:pu,gradFunc:function(n,t,e){var i=e.dims,r=Qe(i,n.shape);return{x:function(){return $n(n,r)}}}};var gD={kernelName:fu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var vD={kernelName:mu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return gt(Ie(n,J(jn(e,1.5),2)))}}}};var yD={kernelName:gu,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(Ps(e),n.dtype))}}}};var bD={kernelName:vu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=hi(e,we(0)),r=we(cv),a=we(hv),s=J(n,a),o=J(J(n,r),fn(ue(e,"float32")));return pn(i,s,o)}}}};var wD={kernelName:Lu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(we(1),e)))}}}};var SD={kernelName:Su,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var LD={kernelName:bu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ks(ue(e,"float32")),n)}}}};var ID={kernelName:wu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Nc(ue(e,"float32")),n)}}}};var AD={kernelName:yu,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=gc(i,a,s),u=l[0],c=l[1],h=[],d=0;d<n.rank;d++)h.push([u[d],o[d]-u[d]-c[d]]);return{x:function(){return Gi(n,h)}}}};var TD={kernelName:Cu,outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.dim,a=!0,s=J(n,i);return{logits:function(){return ge(s,J(Ae(s,[r],a),i))}}}};var ND={kernelName:Iu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Pi(e))}}}};var vv={kernelName:Nu,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.paddings;return{x:function(){return Os(n,r,a)}}}};var yv={kernelName:xu,gradFunc:function(n,t,e){var i=e.axis;return{x:function(){return xt(n,i)}}}};var xD={kernelName:Au,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,J(_t(ue(e,"float32")),2))}}}};var CD={kernelName:Nf,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(ue(e,"float32"),2))}}}};var RD={kernelName:Ru,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=we(2),a=function(){return J(n,J(r,ge(e,i)))},s=function(){return J(n,J(r,ge(i,e)))};return{a,b:s}}};var OD={kernelName:zu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ED={kernelName:Ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(gt(o),i.shape)};return{a,b:s}}};var DD={kernelName:Tu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Qe(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Fr(i.shape,"float32"));return{x:function(){return l}}}};var kD={kernelName:Eu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Ye(ks(e)))}}}};var FD={kernelName:Du,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(we(1),Ye(e)),n)}}}};var WD={kernelName:ku,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o<r[0];++o)s=fe(s,ze(n,[o*i.shape[0]],[i.shape[0]]));else if(i.rank===2)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1]],[i.shape[0],i.shape[1]]));else if(i.rank===3)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2]],[i.shape[0],i.shape[1],i.shape[2]]));else if(i.rank===4)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)for(var c=0;c<r[3];++c)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2],c*i.shape[3]],[i.shape[0],i.shape[1],i.shape[2],i.shape[3]]));else throw new Error("Gradient for tile operation is not implemented for rank-"+(i.rank+" tensors yet."));return s};return{x:a}}};var UD={kernelName:Fu,gradFunc:function(n,t,e){var i=e,r=i.perm,a=xs(r);return{x:function(){return ut(n,a)}}}};var BD={kernelName:Wu,gradFunc:function(n,t,e){var i=e,r=i.axis;return{value:function(){return ji(n,r)}}}};var PD={kernelName:Uu,inputsToSave:["segmentIds"],gradFunc:function(n,t){var e=t[0],i=function(){return zD(n,e)};return{x:i}}};function zD(n,t){for(var e=kr(t,Ee(t)),i=Ws(n,e),r=_i(t,we(0,"int32")),a=i.rank-r.rank,s=0;s<a;++s)r=mn(r,s+1);r=qi(r,Fr(i.shape,"bool"));var o=Ee(i);return pn(r,i,o)}var _D={kernelName:Bu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var MD=[YO,KO,jO,$O,XO,JO,ZO,QO,eE,tE,nE,iE,sE,uE,cE,hE,dE,pE,fE,mE,gE,yE,vE,SE,LE,IE,AE,TE,NE,xE,CE,RE,OE,EE,kE,DE,FE,WE,UE,BE,zE,PE,_E,ME,HE,VE,YE,mv,mv,KE,XE,QE,eD,tD,nD,iD,rD,aD,sD,gv,gv,oD,lD,uD,cD,hD,dD,pD,fD,mD,gD,vD,yD,bD,wD,SD,LD,ID,AD,TD,ND,vv,vv,yv,yv,xD,RD,CD,OD,ED,DD,kD,FD,WD,UD,BD,PD,_D];for(var ih=0,bv=MD;ih<bv.length;ih++){var HD=bv[ih];kf(HD)}Y.prototype.abs=function(){return this.throwIfDisposed(),Gt(this)};Y.prototype.acos=function(){return this.throwIfDisposed(),Bm(this)};Y.prototype.acosh=function(){return this.throwIfDisposed(),zm(this)};Y.prototype.addStrict=function(n){return this.throwIfDisposed(),Hg(this,n)};Y.prototype.add=function(n){return this.throwIfDisposed(),fe(this,n)};Y.prototype.all=function(n,t){return this.throwIfDisposed(),Mm(this,n,t)};Y.prototype.any=function(n,t){return this.throwIfDisposed(),Hm(this,n,t)};Y.prototype.argMax=function(n){return this.throwIfDisposed(),Vm(this,n)};Y.prototype.argMin=function(n){return this.throwIfDisposed(),qm(this,n)};Y.prototype.asScalar=function(){return this.throwIfDisposed(),E(this.size===1,function(){return"The array must have only 1 element."}),V(this,[])};Y.prototype.asType=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.as1D=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.as2D=function(n,t){return this.throwIfDisposed(),V(this,[n,t])};Y.prototype.as3D=function(n,t,e){return this.throwIfDisposed(),V(this,[n,t,e])};Y.prototype.as4D=function(n,t,e,i){return this.throwIfDisposed(),V(this,[n,t,e,i])};Y.prototype.as5D=function(n,t,e,i,r){return this.throwIfDisposed(),V(this,[n,t,e,i,r])};Y.prototype.asin=function(){return this.throwIfDisposed(),Gm(this)};Y.prototype.asinh=function(){return this.throwIfDisposed(),Ym(this)};Y.prototype.atan=function(){return this.throwIfDisposed(),Km(this)};Y.prototype.atan2=function(n){return this.throwIfDisposed(),jm(this,n)};Y.prototype.atanh=function(){return this.throwIfDisposed(),$m(this)};Y.prototype.avgPool=function(n,t,e,i){return this.throwIfDisposed(),Ac(this,n,t,e,i)};Y.prototype.batchToSpaceND=function(n,t){return this.throwIfDisposed(),Os(this,n,t)};Y.prototype.batchNorm=function(n,t,e,i,r){return this.throwIfDisposed(),Aa(this,n,t,e,i,r)};Y.prototype.broadcastTo=function(n){return this.throwIfDisposed(),Ds(this,n)};Y.prototype.cast=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.ceil=function(){return this.throwIfDisposed(),Zm(this)};Y.prototype.clipByValue=function(n,t){return this.throwIfDisposed(),Qm(this,n,t)};Y.prototype.concat=function(n,t){return this.throwIfDisposed(),n instanceof Y&&(n=[n]),xt([this].concat(n),t)};Y.prototype.conv1d=function(n,t,e,i,r,a){return this.throwIfDisposed(),eg(this,n,t,e,i,r,a)};Y.prototype.conv2dTranspose=function(n,t,e,i,r){return this.throwIfDisposed(),tg(this,n,t,e,i,r)};Y.prototype.conv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Er(this,n,t,e,i,r,a)};Y.prototype.cos=function(){return this.throwIfDisposed(),ks(this)};Y.prototype.cosh=function(){return this.throwIfDisposed(),Nc(this)};Y.prototype.cumsum=function(n,t,e){return this.throwIfDisposed(),xc(this,n,t,e)};Y.prototype.depthToSpace=function(n,t){return this.throwIfDisposed(),ig(this,n,t)};Y.prototype.depthwiseConv2D=function(n,t,e,i,r,a){return It("depthwiseConv2D is deprecated, use depthwiseConv2d instead"),this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.depthwiseConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.dilation2d=function(n,t,e,i,r){return this.throwIfDisposed(),rg(this,n,t,e,i,r)};Y.prototype.divNoNan=function(n){return this.throwIfDisposed(),ag(this,n)};Y.prototype.divStrict=function(n){return this.throwIfDisposed(),Vg(this,n)};Y.prototype.div=function(n){return this.throwIfDisposed(),Ie(this,n)};Y.prototype.dot=function(n){return this.throwIfDisposed(),sg(this,n)};Y.prototype.elu=function(){return this.throwIfDisposed(),Cc(this)};Y.prototype.equalStrict=function(n){return this.throwIfDisposed(),Ug(this,n)};Y.prototype.equal=function(n){return this.throwIfDisposed(),Na(this,n)};Y.prototype.erf=function(){return this.throwIfDisposed(),og(this)};Y.prototype.exp=function(){return this.throwIfDisposed(),fn(this)};Y.prototype.expandDims=function(n){return this.throwIfDisposed(),mn(this,n)};Y.prototype.expm1=function(){return this.throwIfDisposed(),lg(this)};Y.prototype.fft=function(){return this.throwIfDisposed(),qs(this)};Y.prototype.flatten=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.floor=function(){return this.throwIfDisposed(),Fs(this)};Y.prototype.floorDiv=function(n){return this.throwIfDisposed(),wc(this,n)};Y.prototype.gather=function(n,t){return this.throwIfDisposed(),Ws(this,n,t)};Y.prototype.greaterEqualStrict=function(n){return this.throwIfDisposed(),Bg(this,n)};Y.prototype.greaterEqual=function(n){return this.throwIfDisposed(),_i(this,n)};Y.prototype.greaterStrict=function(n){return this.throwIfDisposed(),zg(this,n)};Y.prototype.greater=function(n){return this.throwIfDisposed(),hi(this,n)};Y.prototype.ifft=function(){return this.throwIfDisposed(),Oa(this)};Y.prototype.irfft=function(){return this.throwIfDisposed(),Yc(this)};Y.prototype.isFinite=function(){return this.throwIfDisposed(),hg(this)};Y.prototype.isInf=function(){return this.throwIfDisposed(),dg(this)};Y.prototype.isNaN=function(){return this.throwIfDisposed(),pg(this)};Y.prototype.leakyRelu=function(n){return this.throwIfDisposed(),fg(this,n)};Y.prototype.lessEqualStrict=function(n){return this.throwIfDisposed(),Pg(this,n)};Y.prototype.lessEqual=function(n){return this.throwIfDisposed(),Mi(this,n)};Y.prototype.lessStrict=function(n){return this.throwIfDisposed(),_g(this,n)};Y.prototype.less=function(n){return this.throwIfDisposed(),Bs(this,n)};Y.prototype.localResponseNormalization=function(n,t,e,i){return this.throwIfDisposed(),mg(this,n,t,e,i)};Y.prototype.logSigmoid=function(){return this.throwIfDisposed(),vg(this)};Y.prototype.logSoftmax=function(n){return this.throwIfDisposed(),yg(this,n)};Y.prototype.logSumExp=function(n,t){return this.throwIfDisposed(),kc(this,n,t)};Y.prototype.log=function(){return this.throwIfDisposed(),Hi(this)};Y.prototype.log1p=function(){return this.throwIfDisposed(),Ec(this)};Y.prototype.logicalAnd=function(n){return this.throwIfDisposed(),qi(this,n)};Y.prototype.logicalNot=function(){return this.throwIfDisposed(),Ps(this)};Y.prototype.logicalOr=function(n){return this.throwIfDisposed(),Fc(this,n)};Y.prototype.logicalXor=function(n){return this.throwIfDisposed(),bg(this,n)};Y.prototype.matMul=function(n,t,e){return this.throwIfDisposed(),We(this,n,t,e)};Y.prototype.maxPool=function(n,t,e,i){return this.throwIfDisposed(),Wc(this,n,t,e,i)};Y.prototype.max=function(n,t){return this.throwIfDisposed(),Vi(this,n,t)};Y.prototype.maximumStrict=function(n){return this.throwIfDisposed(),qg(this,n)};Y.prototype.maximum=function(n){return this.throwIfDisposed(),kr(this,n)};Y.prototype.mean=function(n,t){return this.throwIfDisposed(),xa(this,n,t)};Y.prototype.min=function(n,t){return this.throwIfDisposed(),_s(this,n,t)};Y.prototype.minimumStrict=function(n){return this.throwIfDisposed(),Gg(this,n)};Y.prototype.minimum=function(n){return this.throwIfDisposed(),Ms(this,n)};Y.prototype.modStrict=function(n){return this.throwIfDisposed(),Yg(this,n)};Y.prototype.mod=function(n){return this.throwIfDisposed(),Uc(this,n)};Y.prototype.mulStrict=function(n){return this.throwIfDisposed(),Kg(this,n)};Y.prototype.mul=function(n){return this.throwIfDisposed(),J(this,n)};Y.prototype.neg=function(){return this.throwIfDisposed(),gt(this)};Y.prototype.norm=function(n,t,e){return this.throwIfDisposed(),$s(this,n,t,e)};Y.prototype.notEqualStrict=function(n){return this.throwIfDisposed(),Mg(this,n)};Y.prototype.notEqual=function(n){return this.throwIfDisposed(),Hs(this,n)};Y.prototype.oneHot=function(n,t,e){return t===void 0&&(t=1),e===void 0&&(e=0),this.throwIfDisposed(),Ts(this,n,t,e)};Y.prototype.onesLike=function(){return this.throwIfDisposed(),Bc(this)};Y.prototype.pad=function(n,t){return this.throwIfDisposed(),Gi(this,n,t)};Y.prototype.pool=function(n,t,e,i,r){return this.throwIfDisposed(),wg(this,n,t,e,i,r)};Y.prototype.powStrict=function(n){return this.throwIfDisposed(),jg(this,n)};Y.prototype.pow=function(n){return this.throwIfDisposed(),jn(this,n)};Y.prototype.prelu=function(n){return this.throwIfDisposed(),zc(this,n)};Y.prototype.prod=function(n,t){return this.throwIfDisposed(),Sg(this,n,t)};Y.prototype.reciprocal=function(){return this.throwIfDisposed(),Ig(this)};Y.prototype.relu=function(){return this.throwIfDisposed(),Ra(this)};Y.prototype.relu6=function(){return this.throwIfDisposed(),Hc(this)};Y.prototype.reshapeAs=function(n){return this.throwIfDisposed(),V(this,n.shape)};Y.prototype.reshape=function(n){return this.throwIfDisposed(),V(this,n)};Y.prototype.resizeBilinear=function(n,t){return this.throwIfDisposed(),ov(this,n,t)};Y.prototype.resizeNearestNeighbor=function(n,t){return this.throwIfDisposed(),lv(this,n,t)};Y.prototype.reverse=function(n){return this.throwIfDisposed(),$n(this,n)};Y.prototype.rfft=function(){return this.throwIfDisposed(),Gs(this)};Y.prototype.round=function(){return this.throwIfDisposed(),Ag(this)};Y.prototype.rsqrt=function(){return this.throwIfDisposed(),Vc(this)};Y.prototype.selu=function(){return this.throwIfDisposed(),Tg(this)};Y.prototype.separableConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ng(this,n,t,e,i,r,a)};Y.prototype.sigmoid=function(){return this.throwIfDisposed(),Pi(this)};Y.prototype.sign=function(){return this.throwIfDisposed(),xg(this)};Y.prototype.sin=function(){return this.throwIfDisposed(),qc(this)};Y.prototype.sinh=function(){return this.throwIfDisposed(),Gc(this)};Y.prototype.slice=function(n,t){return this.throwIfDisposed(),ze(this,n,t)};Y.prototype.softmax=function(n){return this.throwIfDisposed(),Cg(this,n)};Y.prototype.softplus=function(){return this.throwIfDisposed(),Dc(this)};Y.prototype.spaceToBatchND=function(n,t){return this.throwIfDisposed(),Vs(this,n,t)};Y.prototype.split=function(n,t){return this.throwIfDisposed(),Br(this,n,t)};Y.prototype.sqrt=function(){return this.throwIfDisposed(),_t(this)};Y.prototype.square=function(){return this.throwIfDisposed(),Ye(this)};Y.prototype.squaredDifference=function(n){return this.throwIfDisposed(),Ys(this,n)};Y.prototype.squaredDifferenceStrict=function(n){return this.throwIfDisposed(),$g(this,n)};Y.prototype.squeeze=function(n){return this.throwIfDisposed(),Ks(this,n)};Y.prototype.stack=function(n,t){this.throwIfDisposed();var e=n instanceof Y?[this,n]:[this].concat(n);return ji(e,t)};Y.prototype.step=function(n){return this.throwIfDisposed(),zr(this,n)};Y.prototype.stridedSlice=function(n,t,e,i,r,a,s,o){return this.throwIfDisposed(),Og(this,n,t,e,i,r,a,s,o)};Y.prototype.subStrict=function(n){return this.throwIfDisposed(),Xg(this,n)};Y.prototype.sub=function(n){return this.throwIfDisposed(),ge(this,n)};Y.prototype.sum=function(n,t){return this.throwIfDisposed(),Ae(this,n,t)};Y.prototype.tan=function(){return this.throwIfDisposed(),Eg(this)};Y.prototype.tanh=function(){return this.throwIfDisposed(),Rs(this)};Y.prototype.tile=function(n){return this.throwIfDisposed(),Dr(this,n)};Y.prototype.toBool=function(){return this.throwIfDisposed(),ue(this,"bool")};Y.prototype.toFloat=function(){return this.throwIfDisposed(),ue(this,"float32")};Y.prototype.toInt=function(){return this.throwIfDisposed(),ue(this,"int32")};Y.prototype.topk=function(n,t){return this.throwIfDisposed(),Dg(this,n,t)};Y.prototype.transpose=function(n){return this.throwIfDisposed(),ut(this,n)};Y.prototype.unique=function(n){return this.throwIfDisposed(),kg(this,n)};Y.prototype.unsortedSegmentSum=function(n,t){return this.throwIfDisposed(),Kc(this,n,t)};Y.prototype.unstack=function(n){return this.throwIfDisposed(),js(this,n)};Y.prototype.where=function(n,t){return this.throwIfDisposed(),pn(n,this,t)};Y.prototype.zerosLike=function(){return this.throwIfDisposed(),Ee(this)};I.Abs=nl;I.Acos=il;I.Acosh=rl;I.AdadeltaOptimizer=Jc;I.AdagradOptimizer=Zc;I.AdamOptimizer=Qc;I.AdamaxOptimizer=eh;I.Add=ps;I.AddN=al;I.All=Wp;I.Any=Up;I.ArgMax=sl;I.ArgMin=ol;I.Asin=ll;I.Asinh=ul;I.Atan=cl;I.Atan2=dl;I.Atanh=hl;I.AvgPool=pl;I.AvgPool3D=fl;I.AvgPool3DBackprop=zp;I.AvgPoolBackprop=Bp;I.BatchMatMul=ml;I.BatchToSpaceND=gl;I.BroadcastTo=vl;I.Cast=fs;I.Ceil=yl;I.ClipByValue=bl;I.Complex=Pp;I.Concat=wl;I.Conv2D=Sl;I.Conv2DBackpropFilter=_p;I.Conv2DBackpropInput=Ll;I.Conv3D=Il;I.Conv3DBackpropFilterV2=Mp;I.Conv3DBackpropInputV2=Hp;I.Cos=Al;I.Cosh=Tl;I.CropAndResize=Vp;I.Cumsum=Nl;I.DataStorage=EI;I.DepthToSpace=qp;I.DepthwiseConv2dNative=xl;I.DepthwiseConv2dNativeBackpropFilter=Gp;I.DepthwiseConv2dNativeBackpropInput=Yp;I.Diag=Kp;I.Dilation2D=Cl;I.Dilation2DBackpropFilter=$p;I.Dilation2DBackpropInput=jp;I.Div=Rl;I.Elu=Ol;I.EluGrad=Xp;I.Environment=Dp;I.Equal=Jp;I.Erf=El;I.Exp=Dl;I.Expm1=kl;I.FFT=Zp;I.Fill=Qp;I.FlipLeftRight=ef;I.Floor=Fl;I.FloorDiv=Wl;I.FromPixels=Pu;I.FusedBatchNorm=Ul;I.FusedConv2D=Mu;I.FusedDepthwiseConv2D=Hu;I.GatherNd=tf;I.GatherV2=Bl;I.Greater=nf;I.GreaterEqual=zl;I.IFFT=rf;I.Identity=Pl;I.Imag=af;I.IsFinite=_l;I.IsInf=Ml;I.IsNan=Hl;I.KernelBackend=Op;I.LRN=Yl;I.LRNBackprop=df;I.Less=sf;I.LessEqual=of;I.LinSpace=lf;I.Log=Vl;I.Log1p=ql;I.LogSoftmax=Gl;I.LogicalAnd=uf;I.LogicalNot=cf;I.LogicalOr=hf;I.Max=Kl;I.MaxPool=$l;I.MaxPool3D=Xl;I.MaxPool3DBackprop=ff;I.MaxPoolBackprop=pf;I.MaxPoolWithArgmax=mf;I.Maximum=jl;I.Mean=BI;I.Min=Jl;I.Minimum=Zl;I.Mod=Ql;I.MomentumOptimizer=th;I.Multiply=eu;I.Negate=tu;I.NonMaxSuppressionV3=vf;I.NonMaxSuppressionV4=yf;I.NonMaxSuppressionV5=bf;I.NotEqual=gf;I.OP_SCOPE_SUFFIX=tm;I.OneHot=iu;I.OnesLike=nu;I.Optimizer=di;I.PadV2=ru;I.Pool=zI;I.Pow=au;I.Prelu=su;I.Prod=wf;I.RMSPropOptimizer=nh;I.Range=Sf;I.Real=Lf;I.Reciprocal=ou;I.Relu=lu;I.Relu6=du;I.Reshape=uu;I.ResizeBilinear=hu;I.ResizeBilinearGrad=Af;I.ResizeNearestNeighbor=cu;I.ResizeNearestNeighborGrad=If;I.Reverse=pu;I.RotateWithOffset=Ef;I.Round=fu;I.Rsqrt=mu;I.SGDOptimizer=eo;I.ScatterNd=Tf;I.SelectV2=gu;I.Selu=vu;I.Sigmoid=Lu;I.Sign=Su;I.Sin=bu;I.Sinh=wu;I.Slice=yu;I.Softmax=Cu;I.Softplus=Iu;I.SpaceToBatchND=Nu;I.SparseToDense=xf;I.SplitV=xu;I.Sqrt=Au;I.Square=Nf;I.SquaredDifference=Ru;I.Step=zu;I.StridedSlice=Cf;I.Sub=Ou;I.Sum=Tu;I.Tan=Eu;I.Tanh=Du;I.Tensor=Y;I.TensorBuffer=Ss;I.Tile=ku;I.TopK=Rf;I.Transpose=Fu;I.Unique=Of;I.Unpack=Wu;I.UnsortedSegmentSum=Uu;I.Variable=ya;I.ZerosLike=Bu;I._FusedMatMul=_u;I.abs=Gt;I.acos=Bm;I.acosh=zm;I.add=fe;I.addN=gT;I.addStrict=Hg;I.all=Mm;I.any=Hm;I.argMax=Vm;I.argMin=qm;I.asin=Gm;I.asinh=Ym;I.atan=Km;I.atan2=jm;I.atanh=$m;I.avgPool=Ac;I.avgPool3d=FT;I.backend=sT;I.backend_util=MO;I.basicLSTMCell=_T;I.batchNorm=Aa;I.batchNorm2d=GT;I.batchNorm3d=KT;I.batchNorm4d=$T;I.batchToSpaceND=Os;I.booleanMaskAsync=yC;I.broadcastTo=Ds;I.browser=OA;I.buffer=On;I.cast=ue;I.ceil=Zm;I.clipByValue=Qm;I.clone=Bi;I.complex=ri;I.concat=xt;I.concat1d=e1;I.concat2d=n1;I.concat3d=r1;I.concat4d=s1;I.conv1d=eg;I.conv2d=Er;I.conv2dTranspose=tg;I.conv3d=d1;I.conv3dTranspose=m1;I.copyRegisteredKernels=MI;I.cos=ks;I.cosh=Nc;I.cosineWindow=jc;I.cumsum=xc;I.customGrad=kn;I.deprecationWarn=It;I.depthToSpace=ig;I.depthwiseConv2d=Ta;I.device_util=I2;I.diag=L1;I.dilation2d=rg;I.disableDeprecationWarnings=KA;I.dispose=zt;I.disposeVariables=jA;I.div=Ie;I.divNoNan=ag;I.divStrict=Vg;I.dot=sg;I.dropout=YC;I.elu=Cc;I.enableDebugMode=YA;I.enableProdMode=GA;I.enclosingPowerOfTwo=Zg;I.engine=$A;I.env=He;I.equal=Na;I.equalStrict=Ug;I.erf=og;I.exp=fn;I.expandDims=mn;I.expm1=lg;I.eye=ug;I.fft=qs;I.fill=Rc;I.findBackend=iT;I.findBackendFactory=rT;I.floor=Fs;I.floorDiv=wc;I.fused=rR;I.gather=Ws;I.gatherND=VC;I.gather_util=EA;I.getBackend=tT;I.getGradient=Gu;I.getKernel=qu;I.getKernelsForBackend=ms;I.grad=iN;I.grads=rN;I.greater=hi;I.greaterEqual=_i;I.greaterEqualStrict=Bg;I.greaterStrict=zg;I.ifft=Oa;I.imag=Us;I.image=cO;I.inTopKAsync=jC;I.io=bA;I.irfft=Yc;I.isFinite=hg;I.isInf=dg;I.isNaN=pg;I.keep=Um;I.kernel_impls=GO;I.leakyRelu=fg;I.less=Bs;I.lessEqual=Mi;I.lessEqualStrict=Pg;I.lessStrict=_g;I.linalg=hO;I.linspace=Q1;I.localResponseNormalization=mg;I.log=Hi;I.log1p=Ec;I.logSigmoid=vg;I.logSoftmax=yg;I.logSumExp=kc;I.logicalAnd=qi;I.logicalNot=Ps;I.logicalOr=Fc;I.logicalXor=bg;I.losses=dO;I.matMul=We;I.math=NA;I.max=Vi;I.maxPool=Wc;I.maxPool3d=SN;I.maxPoolWithArgmax=IN;I.maximum=kr;I.maximumStrict=qg;I.mean=xa;I.memory=XA;I.min=_s;I.minimum=Ms;I.minimumStrict=Gg;I.mod=Uc;I.modStrict=Yg;I.moments=ON;I.movingAverage=UC;I.mul=J;I.mulStrict=Kg;I.multiRNNCell=DN;I.multinomial=FN;I.neg=gt;I.nextFrame=mO;I.norm=$s;I.notEqual=Hs;I.notEqualStrict=Mg;I.oneHot=Ts;I.ones=Fr;I.onesLike=Bc;I.op=U;I.outerProduct=PN;I.pad=Gi;I.pad1d=HN;I.pad2d=qN;I.pad3d=YN;I.pad4d=jN;I.pool=wg;I.pow=jn;I.powStrict=jg;I.prelu=zc;I.print=hm;I.prod=Sg;I.profile=JA;I.rand=ix;I.randomGamma=fx;I.randomNormal=gx;I.randomUniform=Lg;I.range=Mc;I.ready=eT;I.real=Ca;I.reciprocal=Ig;I.registerBackend=aT;I.registerGradient=kf;I.registerKernel=Df;I.relu=Ra;I.relu6=Hc;I.removeBackend=nT;I.reshape=V;I.reverse=$n;I.reverse1d=Ix;I.reverse2d=Tx;I.reverse3d=xx;I.reverse4d=Rx;I.rfft=Gs;I.round=Ag;I.rsqrt=Vc;I.scalar=we;I.scatterND=zC;I.scatter_util=DA;I.selu=Tg;I.separableConv2d=Ng;I.serialization=WA;I.setBackend=QA;I.setPlatform=oT;I.setdiff1dAsync=Wx;I.sigmoid=Pi;I.sign=xg;I.signal=uO;I.sin=qc;I.sinh=Gc;I.slice=ze;I.slice1d=_x;I.slice2d=Hx;I.slice3d=qx;I.slice4d=Yx;I.slice_util=Dm;I.softmax=Cg;I.softplus=Dc;I.spaceToBatchND=Vs;I.sparseToDense=MC;I.spectral=lO;I.split=Br;I.sqrt=_t;I.square=Ye;I.squaredDifference=Ys;I.squaredDifferenceStrict=$g;I.squeeze=Ks;I.stack=ji;I.step=zr;I.stridedSlice=Og;I.sub=ge;I.subStrict=Xg;I.sum=Ae;I.sumOutType=f2;I.tan=Eg;I.tanh=Rs;I.tensor=si;I.tensor1d=Ur;I.tensor2d=Ea;I.tensor3d=ym;I.tensor4d=sC;I.tensor5d=oC;I.tensor6d=lC;I.tensor_util=v2;I.test_util=VA;I.tidy=ft;I.tile=Dr;I.time=ZA;I.topk=Dg;I.train=pO;I.transpose=ut;I.truncatedNormal=hC;I.unique=kg;I.unregisterGradient=_I;I.unregisterKernel=PI;I.unsortedSegmentSum=Kc;I.unstack=js;I.upcastType=Ls;I.util=i2;I.valueAndGrad=aN;I.valueAndGrads=sN;I.variable=mC;I.variableGrads=gg;I.version_core=qA;I.where=pn;I.whereAsync=Wg;I.zeros=Kn;I.zerosLike=Ee});var Yb=be(Xe=>{"use strict";Object.defineProperty(Xe,"__esModule",{value:!0});var y=Xi();var rh=function(n,t){return rh=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},rh(n,t)};function Q(n,t){rh(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var Yt=function(){return Yt=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t},Yt.apply(this,arguments)};function VD(n,t){var e={};for(var i in n)Object.prototype.hasOwnProperty.call(n,i)&&t.indexOf(i)<0&&(e[i]=n[i]);if(n!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,i=Object.getOwnPropertySymbols(n);r<i.length;r++)t.indexOf(i[r])<0&&(e[i[r]]=n[i[r]]);return e}function Le(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function ve(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var ah;function vt(){return ah==null&&(ah=y.backend().epsilon()),ah}function gn(){return"channelsLast"}var pi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Fn=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),M=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Te=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),qD=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),SH=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error);function Ji(n,t){if(Array.isArray(n)){for(var e=[],i=0;i<t;i++)e=e.concat(n);return e}else{var e=new Array(t);return e.fill(n),e}}function Wn(n,t){if(!n)throw new qD(t)}function wv(n,t){for(var e=0,i=0,r=n;i<r.length;i++){var a=r[i];a===t&&e++}return e}function Mt(n){return n.length===1?n[0]:n}function Je(n){return Array.isArray(n)?n:[n]}function Jn(n){var t=n.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2"),e=t.replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function Zi(n){return n.length<=1||n.indexOf("_")===-1?n:n.replace(/[_]+(\w|$)/g,function(t,e){return e.toUpperCase()})}var un={};function sh(n){if(n==null)return null;var t={};return t.className=n.getClassName(),t.config=n.getConfig(),t}function oh(n){if(n==null||typeof n!="object")return;if(Array.isArray(n))n.forEach(function(s){return oh(s)});else for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e],a=n[r];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?n[r]=a.value:oh(a))}}function Da(n,t,e,i,r){var a,s,o;if(t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i="object"),r===void 0&&(r=!1),typeof n=="string"){var l=n,u=void 0;if(l in e)u=e[l];else if(l in un)u=un[l];else if(u=t[l],u==null)throw new M("Unknown "+i+": "+n+`. This may be due to one of the following reasons:
+        `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,qp,u)}var ig=U({depthToSpace_:b1});function w1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(rt(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(Pt(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Dn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,Cl,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ta=U({depthwiseConv2d_:w1});function S1(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Kp)}var L1=U({diag_:S1});function I1(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(Rl,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var rg=U({dilation2d_:I1});function A1(n,t){for(var e=n.length,i=[],r=0;r<e;r++){var a=e-1-r,s=n[a]||1,o=t[t.length-1-r]||1;o>1&&s===1&&i.unshift(a)}return i}function mt(n,t){for(var e=[],i=0;i<t.length;i++){var r=n[n.length-i-1],a=t.length-i-1,s=t[a];(r==null||r===1&&s>1)&&e.unshift(a)}return e}function et(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;r<i;r++){var a=n[n.length-r-1];a==null&&(a=1);var s=t[t.length-r-1];if(s==null&&(s=1),a===1)e.unshift(s);else if(s===1)e.unshift(a);else if(a!==s){var o="Operands could not be broadcast together with shapes "+(n+" and "+t+".");throw Error(o)}else e.unshift(a)}return e}function T1(n,t){var e,i=O(n,"a","equal"),r=O(t,"b","equal");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.equal(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Jp)}var Na=U({equal_:T1});function N1(n,t,e){var i=O(t,"a","where"),r=O(e,"b","where"),a=O(n,"condition","where","bool"),s=et(i.shape,r.shape),o=Ds(i,s),l=Ds(r,s);a.rank===1&&E(a.shape[0]===i.shape[0],function(){return"The first dimension of `a` must match the size of `condition`."}),a.rank!==1&&Be(a.shape,l.shape,"Error in where: ");var u=function(h,d){var p=h.select(a,o,l);return d([a]),p},c={condition:a,t:o,e:l};return z.runKernelFunc(u,c,null,vu)}var dn=U({where_:N1});function x1(n){var t=O(n,"x","zerosLike"),e={x:t};return z.runKernelFunc(function(i){return i.zerosLike(t)},e,null,zu)}var Ee=U({zerosLike_:x1});function C1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");e=at(i,r),i=e[0],r=e[1];var a=Ie(i,r),s=Ee(a),o=Na(r,s);return dn(o,s,a)}var ag=U({divNoNan_:C1});function R1(n,t){var e=O(n,"t1","dot"),i=O(t,"t2","dot");E((e.rank===1||e.rank===2)&&(i.rank===1||i.rank===2),function(){return"Error in dot: inputs must all be rank 1 or 2, but got ranks "+(e.rank+" and "+i.rank+".")});var r=e.rank===1?e.size:e.shape[1],a=i.rank===1?i.size:i.shape[0];if(E(r===a,function(){return"Error in dot: inner dimensions of inputs must match, but got "+(r+" and "+a+".")}),e.rank===1&&i.rank===1){var s=V(e,[1,-1]),o=V(i,[-1,1]),l=We(s,o);return V(l,[])}else if(e.rank===1&&i.rank===2){var s=V(e,[1,-1]),o=V(i,[i.shape[0],i.shape[1]]),l=We(s,o);return V(l,[l.size])}else if(e.rank===2&&i.rank===1){var o=V(i,[-1,1]),l=We(e,o);return V(l,[l.size])}else{var o=V(i,[i.shape[0],i.shape[1]]),l=We(e,o);return l}}var sg=U({dot_:R1});function O1(n){var t=O(n,"x","elu"),e=function(r,a){var s=r.elu(t);return a([s]),s},i={x:t};return z.runKernelFunc(e,i,null,El)}var Rc=U({elu_:O1});function E1(n){var t=O(n,"x","erf");E(t.dtype==="int32"||t.dtype==="float32",function(){return"Input dtype must be `int32` or `float32`."}),t.dtype==="int32"&&(t=ue(t,"float32"));var e={x:t};return z.runKernelFunc(function(i,r){var a=i.erf(t);return r([t]),a},e,null,Dl)}var og=U({erf_:E1});function D1(n){var t=O(n,"x","exp"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.exp(t);return r([a]),a},e,null,kl)}var pn=U({exp_:D1});function k1(n,t){t===void 0&&(t=0);var e=null,i=O(n,"x","expandDims",e);E(t<=i.rank,function(){return"Axis must be <= rank of the tensor"});var r=i.shape.slice();return t<0&&(E(-(i.rank+1)<=t,function(){return"Axis must be in the interval ["+-(i.rank+1)+", "+i.rank+"]"}),t=i.rank+t+1),r.splice(t,0,1),V(i,r)}var fn=U({expandDims_:k1});function F1(n){var t=O(n,"x","expm1"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.expm1(t);return r([t]),a},e,null,Fl)}var lg=U({expm1_:F1});function W1(n,t){var e=null,i=O(n,"x","tile",e);E(i.rank===t.length,function(){return"Error in transpose: rank of input "+i.rank+" "+("must match length of reps "+t+".")});var r=function(l,u){var c=l.tile(i,t);return u([i]),c},a=[i],s={x:i},o={reps:t};return z.runKernelFunc(r,s,null,Fu,o,a)}var Dr=U({tile_:W1});function U1(n,t,e,i){i===void 0&&(i="float32"),t==null&&(t=n);for(var r=On([n,t],i),a=n<=t?n:t,s=0;s<a;++s)r.set(1,s,s);var o=V(r.toTensor(),[n,t]);if(e==null)return o;if(e.length===1)return Dr(fn(o,0),[e[0],1,1]);if(e.length===2)return Dr(fn(fn(o,0),0),[e[0],e[1],1,1]);if(e.length===3)return Dr(fn(fn(fn(o,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error("eye() currently supports only 1D and 2D "+("batchShapes, but received "+e.length+"D."))}var ug=U({eye_:U1});function Oc(n,t,e){var i={shape:n,value:t,dtype:e};return z.runKernelFunc(function(r){return r.fill(n,t,e)},{},null,Qp,i)}function B1(n){var t=O(n,"x","floor"),e={x:t};return z.runKernelFunc(function(i){return i.floor(t)},e,null,Wl)}var Fs=U({floor_:B1});var Ec=30;function z1(n){return n<=Ec?n:ys(n,Math.floor(Math.sqrt(n)))}function P1(n,t){var e=!1,i;for(n<=Ec?(i=n,e=!0):i=ys(n,Math.floor(Math.sqrt(n)));!e;)i>t||i===n?e=!0:i=ys(n,i+1);return i}function _1(n,t,e){for(var i=[],r=n.length,a=0;a<r;a++)a!==t?i.push(n[a]):i.push(e);return i}function cg(n,t,e){for(var i=n.shape[e],r=[],a=1,s=1,o=0;o<e;o++)r.push(n.shape[o]),a*=n.shape[o];for(var o=0;o<t.rank;o++)r.push(t.shape[o]);for(var o=e+1;o<n.rank;o++)r.push(n.shape[o]),s*=n.shape[o];return{batchSize:a,sliceSize:s,dimSize:i,outputShape:r}}var M1={__proto__:null,segOpComputeOptimalWindowSize:P1,computeOutShape:_1,collectGatherOpShapeInfo:cg};function H1(n,t,e){e===void 0&&(e=0);var i=O(n,"x","gather"),r=O(t,"indices","gather","int32"),a={x:i,indices:r},s={axis:e},o=function(l,u){var c=Qe(e,i.shape)[0],h=cg(i,r,c),d=l.gather(i,V(r,[r.size]),c);return u([i,r]),V(d,h.outputShape)};return z.runKernelFunc(o,a,null,zl,s)}var Ws=U({gather_:H1});function V1(n,t){var e,i=O(n,"a","greater"),r=O(t,"b","greater");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.greater(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,nf)}var hi=U({greater_:V1});function q1(n,t){var e,i=O(n,"a","greaterEqual"),r=O(t,"b","greaterEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.greaterEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Pl)}var _i=U({greaterEqual_:q1});function G1(n){var t=O(n,"input","imag"),e=function(r){return r.imag(t)},i={input:t};return z.runKernelFunc(e,i,null,af)}var Us=U({imag_:G1});function Y1(n){var t=O(n,"x","isFinite"),e={x:t};return z.runKernelFunc(function(i){return i.isFinite(t)},e,null,Ml)}var hg=U({isFinite_:Y1});function K1(n){var t=O(n,"x","isInf"),e={x:t};return z.runKernelFunc(function(i){return i.isInf(t)},e,null,Hl)}var dg=U({isInf_:K1});function j1(n){var t=O(n,"x","isNaN"),e={x:t};return z.runKernelFunc(function(i){return i.isNaN(t)},e,null,Vl)}var pg=U({isNaN_:j1});function $1(n,t){var e,i=O(n,"a","maximum"),r=O(t,"b","maximum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.maximum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,$l)}var kr=U({maximum_:$1});function we(n,t){if((Ot(n)&&t!=="string"||Array.isArray(n))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&Ot(n)&&!(n instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");var e=[],i=[];return ai(n,e,i,t)}function X1(n,t){t===void 0&&(t=.2);var e=O(n,"x","leakyRelu");return kr(J(we(t),e),e)}var fg=U({leakyRelu_:X1});function J1(n,t){var e,i=O(n,"a","less"),r=O(t,"b","less");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.less(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,sf)}var Bs=U({less_:J1});function Z1(n,t){var e,i=O(n,"a","lessEqual"),r=O(t,"b","lessEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o,l){var u=o.lessEqual(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,of)}var Mi=U({lessEqual_:Z1});function Q1(n,t,e){if(e<=0)throw new Error("The number of values should be positive.");var i={start:n,stop:t,num:e};return z.runKernelFunc(function(r){return r.linspace(n,t,e)},{},null,lf,i)}function eN(n,t,e,i,r){t===void 0&&(t=5),e===void 0&&(e=1),i===void 0&&(i=1),r===void 0&&(r=.5);var a=O(n,"x","localResponseNormalization");E(a.rank===4||a.rank===3,function(){return`Error in localResponseNormalization: x must be rank 3 or 4 but got
+               rank `+a.rank+"."}),E(rt(t),function(){return"Error in localResponseNormalization: depthRadius must be an "+("integer but got depthRadius "+t+".")});var s=a,o=!1;a.rank===3&&(o=!0,s=V(a,[1,a.shape[0],a.shape[1],a.shape[2]]));var l=function(d,p){var f=d.localResponseNormalization4D(s,t,e,i,r);return p([s,f]),f},u={x:s},c={depthRadius:t,bias:e,alpha:i,beta:r},h=z.runKernelFunc(l,u,null,Kl,c);return o?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var mg=U({localResponseNormalization_:eN});function tN(n){var t=O(n,"x","log"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log(t);return r([t]),a},e,null,ql)}var Hi=U({log_:tN});function nN(n){var t=O(n,"x","log1p"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.log1p(t);return r([t]),a},e,null,Gl)}var Dc=U({log1p_:nN});function iN(n){return E(ii(n),function(){return"The f passed in grad(f) must be a function"}),function(t,e){var i=O(t,"x","tf.grad",null),r=e!=null?O(e,"dy","tf.grad"):null;return z.tidy(function(){var a=z.gradients(function(){return n(i)},[i],r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),zs(o),o[0]})}}function rN(n){return E(ii(n),function(){return"The f passed in grads(f) must be a function"}),function(t,e){E(Array.isArray(t),function(){return"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s"});var i=ba(t,"args","tf.grads",null),r=e!=null?O(e,"dy","tf.grads"):null;return z.tidy(function(){var a=z.gradients(function(){return n.apply(void 0,i)},i,r),s=a.value,o=a.grads;return r!=null&&Be(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(o),o})}}function aN(n){return E(ii(n),function(){return"The f passed in valueAndGrad(f) must be a function"}),function(t,e){E(t instanceof Y,function(){return"The x passed in valueAndGrad(f)(x) must be a tensor"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrad(f)(x, dy) must be a tensor"});var i=z.gradients(function(){return n(t)},[t],e),r=i.grads,a=i.value;return zs(r),{grad:r[0],value:a}}}function sN(n){return E(ii(n),function(){return"The f passed in valueAndGrads(f) must be a function"}),function(t,e){E(Array.isArray(t)&&t.every(function(r){return r instanceof Y}),function(){return"The args passed in valueAndGrads(f)(args) must be array of tensors"}),E(e==null||e instanceof Y,function(){return"The dy passed in valueAndGrads(f)(args, dy) must be a tensor"});var i=z.gradients(function(){return n.apply(void 0,t)},t,e);return e!=null&&Be(i.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),zs(i.grads),i}}function gg(n,t){E(ii(n),function(){return"The f passed in variableGrads(f) must be a function"}),E(t==null||Array.isArray(t)&&t.every(function(h){return h instanceof ya}),function(){return"The varList passed in variableGrads(f, varList) must be an array of variables"});var e=t!=null;if(!e){t=[];for(var i in z.registeredVariables)t.push(z.registeredVariables[i])}var r=e?t.filter(function(h){return!h.trainable}):null,a=t.length;t=t.filter(function(h){return h.trainable}),E(t.length>0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function kn(n){return z.customGrad(n)}function zs(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
+    the f you passed encloses all operations that lead from x to y.`)}function oN(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,nu)}var gt=U({neg_:oN});function lN(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,Au)}var kc=U({softplus_:lN});function uN(n){var t=O(n,"x","logSigmoid"),e=kn(function(i){var r=gt(kc(gt(i))),a=function(s){var o=J(s,Pi(gt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var vg=U({logSigmoid_:uN});function cN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=en(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=Qt(f.shape,Qe(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,jl,s)}var Vi=U({max_:cN});function hN(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Eu)}var ge=U({sub_:hN});function dN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Qe(t,i.shape),c=en(u,i.rank),h=u,d=i;c!=null&&(d=ut(i,c),h=En(h.length,i.rank));var p=o.sum(d,h);if(e){var f=Qt(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Nu,s)}var Ae=U({sum_:dN});function pN(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=Vi(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),Hi(Ae(pn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,Yl,a)}var yg=U({logSoftmax_:pN});function fN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Qe(t,i.shape),a=Vi(i,r,!0),s=ge(i,a),o=pn(s),l=Ae(o,r),u=Hi(l),c=fe(V(a,u.shape),u);if(e){var h=Qt(c.shape,r);return V(c,h)}return c}var Fc=U({logSumExp_:fN});function mN(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,uf)}var qi=U({logicalAnd_:mN});function gN(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,cf)}var Ps=U({logicalNot_:gN});function vN(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");et(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,hf)}var Wc=U({logicalOr_:vN});function yN(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return et(e.shape,i.shape),qi(Wc(n,t),Ps(qi(n,t)))}var bg=U({logicalXor_:yN});function bN(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(Pt(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=Rr(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&on(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,Xl,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Uc=U({maxPool_:bN});function wN(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(Pt(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(rt(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=Ia(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Jl,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var SN=U({maxPool3d_:wN});function LN(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(mf,s,o);return{result:l[0],indexes:l[1]}}var IN=U({maxPoolWithArgmax_:LN});function Kn(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Kn(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=Lr(ot(n),t);return z.makeTensor(r,n,t)}function Fr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Fr(n,"float32"),i=Kn(n,"float32");return ri(e,i)}var r=ju(ot(n),t);return z.makeTensor(r,n,t)}function AN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Qe(t,i.shape),a=_m(i.shape,r),s=a[1],o=ot(s),l=kn(function(u){var c=we(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Ie(h,c),p=Ae(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Ie(J(v,Fr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var xa=U({mean_:AN});function TN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Qe(t,i.shape),c=u,h=en(c,i.rank),d=i;h!=null&&(d=ut(i,h),c=En(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=Qt(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Zl,s)}var _s=U({min_:TN});function NN(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=at(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),et(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ql)}var Ms=U({minimum_:NN});function xN(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=at(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,eu)}var Bc=U({mod_:xN});function CN(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ye=U({square_:CN});function RN(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Qe(t,n.shape),r=xa(n,i,e),a=r.shape;e||(a=Qt(r.shape,i));var s=Ye(ge(ue(n,"float32"),V(r,a))),o=xa(s,i,e);return{mean:r,variance:o}}var ON=U({moments_:RN});function EN(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=ba(e,"c","multiRNNCell"),s=ba(i,"h","multiRNNCell"),o=r,l=[],u=0;u<n.length;u++){var c=n[u](o,a[u],s[u]);l.push(c[0]),l.push(c[1]),o=c[1]}for(var h=[],d=[],u=0;u<l.length;u+=2)h.push(l[u]),d.push(l[u+1]);return[h,d]}var DN=U({multiRNNCell_:EN});function kN(n,t,e,i){i===void 0&&(i=!1);var r=O(n,"logits","multinomial"),a=r.size,s=r.rank;if(a<2)throw new Error("Error in multinomial: you need at least 2 outcomes, but got "+(a+"."));if(s>2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var FN=U({multinomial_:kN});function WN(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,gf)}var Hs=U({notEqual_:WN});function UN(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,Lf)}var Ca=U({real_:UN});function BN(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=zc(Ca(t)),o=Ee(Us(t));return ri(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,iu)}var zc=U({onesLike_:BN});function zN(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return We(r,a)}var PN=U({outerProduct_:zN});function _N(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,au,a)}var Gi=U({pad_:_N});function MN(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),Gi(n,[t],e)}var HN=U({pad1d_:MN});function VN(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var qN=U({pad2d_:VN});function GN(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var YN=U({pad3d_:GN});function KN(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Gi(n,t,e)}var jN=U({pad4d_:KN});function $N(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,xu,s)}var Vs=U({spaceToBatchND_:$N});function ZN(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(Pt(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=Rr(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=JN([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=XN([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:Vs(o,c,f),b=e==="avg"?function(){return Tc(v,t,a,g)}:function(){return Uc(v,t,a,g)},w=b(),S=d?w:Os(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function XN(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function JN(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var wg=U({pool_:ZN});function QN(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=at(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,su)}var jn=U({pow_:QN});function ex(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,ou)}var Pc=U({prelu_:ex});function tx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Qe(t,i.shape),u=en(l,i.rank),c=l,h=i;u!=null&&(h=ut(i,u),c=En(c.length,i.rank));var d=o.prod(h,c);if(e){var p=Qt(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,wf,s)}var Sg=U({prod_:tx});function nx(n,t,e){var i=ot(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a<i;a++)r[a]=t();return z.makeTensor(r,n,e)}var ix=U({rand_:nx}),Wr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function Yi(n,t){return t={exports:{}},n(t,t.exports),t.exports}var rx=Yi(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(Wr,n,!1)}),ax=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(Wr,n,!1)}),sx=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(Wr,n,!1)}),ox=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(Wr,n,!1)}),lx=Yi(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(Wr,n,!1)}),ux=Yi(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(Wr,n,!1)}),Ki=Yi(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});Ki.alea=rx;Ki.xor128=ax;Ki.xorwow=sx;Ki.xorshift7=ox;Ki.xor4096=lx;Ki.tychei=ux;var cx=Ki,_c=cx.alea;var Mc=function(){function n(t,e,i,r,a){this.mean=t,this.stdDev=e,this.dtype=i,this.nextVal=NaN,this.truncated=r,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);var s=a||Math.random();this.random=_c(s.toString())}return n.prototype.nextValue=function(){if(!isNaN(this.nextVal)){var t=this.nextVal;return this.nextVal=NaN,t}for(var e,i,r=!1;!r;){var a=void 0,s=void 0,o=void 0;do a=2*this.random()-1,s=2*this.random()-1,o=a*a+s*s;while(o>=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),hx=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=_c(a.toString()),this.randn=new Mc(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a<e||Math.log(a)<i)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)},n.prototype.convertValue=function(t){return this.dtype==="float32"?t:Math.round(t)},n}(),dx=function(){function n(t,e,i,r){var a=this;if(t===void 0&&(t=0),e===void 0&&(e=1),this.canReturnFloat=function(){return a.dtype==null||a.dtype==="float32"},this.min=t,this.range=e-t,this.dtype=i,r==null&&(r=Math.random()),typeof r=="number"&&(r=r.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error("The difference between "+t+" - "+e+" <= 1 and dtype is not float");this.random=_c(r)}return n.prototype.convertValue=function(t){return this.canReturnFloat()?t:Math.round(t)},n.prototype.nextValue=function(){return this.convertValue(this.min+this.range*this.random())},n}();function px(n,t,e,i,r){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e==null&&(e=1),i==null&&(i="float32"),i!=="float32"&&i!=="int32")throw new Error("Unsupported data type "+i);for(var a=new hx(t,e,i,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var fx=U({randomGamma_:px});function mx(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type "+i);for(var a=new Mc(t,e,i,!1,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var gx=U({randomNormal_:mx});function vx(n,t,e,i,r){t===void 0&&(t=0),e===void 0&&(e=1),i===void 0&&(i="float32");for(var a=On(n,i),s=new dx(t,e,null,r),o=0;o<a.values.length;o++)a.values[o]=s.nextValue();return a.toTensor()}var Lg=U({randomUniform_:vx});function Ur(n,t){Fi(n);var e=Rn(n,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");var i=null;return ai(n,i,e,t)}function Hc(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i="float32"),e===0)throw new Error("Cannot have a step of zero");var r=function(){var s=n===t,o=n<t&&e<0,l=t<n&&e>1;if(s||o||l)return Kn([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=Lr(u,i);t<n&&e===1&&(e=-1),c[0]=n;for(var h=1;h<c.length;h++)c[h]=c[h-1]+e;return Ur(c,i)},a={start:n,stop:t,step:e,dtype:i};return z.runKernelFunc(r,{},null,Sf,a)}function yx(n){var t=O(n,"x","reciprocal"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.reciprocal(t);return r([t]),a},e,null,lu)}var Ig=U({reciprocal_:yx});function bx(n){var t=O(n,"x","relu"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu(t)},i={x:t};return z.runKernelFunc(e,i,null,uu)}var Ra=U({relu_:bx});function wx(n){var t=O(n,"x","relu6"),e=function(r,a){return a([t]),t.dtype==="bool"?ue(t,"int32"):r.relu6(t)},i={x:t};return z.runKernelFunc(e,i,null,pu)}var Vc=U({relu6_:wx});function Sx(n,t){var e=O(n,"x","reverse"),i=function(s){var o=Qe(t,e.shape);if(e.rank===0)return Bi(e);var l=s.reverse(e,o);return V(l,e.shape)},r={x:e},a={dims:t};return z.runKernelFunc(i,r,null,fu,a)}var $n=U({reverse_:Sx});function Lx(n){var t=O(n,"x","reverse");return E(t.rank===1,function(){return"Error in reverse1D: x must be rank 1 but got rank "+t.rank+"."}),$n(t,0)}var Ix=U({reverse1d_:Lx});function Ax(n,t){var e=O(n,"x","reverse");return E(e.rank===2,function(){return"Error in reverse2D: x must be rank 2 but got rank "+e.rank+"."}),$n(e,t)}var Tx=U({reverse2d_:Ax});function Nx(n,t){var e=O(n,"x","reverse");return E(e.rank===3,function(){return"Error in reverse3D: x must be rank 3 but got rank "+e.rank+"."}),$n(e,t)}var xx=U({reverse3d_:Nx});function Cx(n,t){var e=O(n,"x","reverse");return E(e.rank===4,function(){return"Error in reverse4D: x must be rank 4 but got rank "+e.rank+"."}),$n(e,t)}var Rx=U({reverse4d_:Cx});function Ox(n){var t=O(n,"x","round"),e={x:t};return z.runKernelFunc(function(i){return i.round(t)},e,null,mu)}var Ag=U({round_:Ox});function Ex(n){var t=O(n,"x","rsqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.rsqrt(t);return r([t]),a},e,null,gu)}var qc=U({rsqrt_:Ex});function Dx(n){var t=O(n,"x","selu"),e=function(r,a){var s=r.selu(t);return a([t]),s},i={x:t};return z.runKernelFunc(e,i,null,yu)}var Tg=U({selu_:Dx});function kx(n,t,e,i,r,a,s){a===void 0&&(a=[1,1]),s===void 0&&(s="NHWC");var o=O(n,"x","separableConv2d"),l=O(t,"depthwiseFilter","separableConv2d"),u=O(e,"pointwiseFilter","separableConv2d"),c=o,h=!1;if(o.rank===3&&(h=!0,c=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),s==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");E(c.rank===4,function(){return"Error in separableConv2d: input must be rank 4, but got "+("rank "+c.rank+".")}),E(l.rank===4,function(){return"Error in separableConv2d: depthwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.rank===4,function(){return"Error in separableConv2d: pointwise filter must be rank 4, but "+("got rank "+l.rank+".")}),E(u.shape[0]===1,function(){return"Error in separableConv2d: the first dimension of pointwise filter "+(" must be 1, but got "+u.shape[0]+".")}),E(u.shape[1]===1,function(){return"Error in separableConv2d: the second dimension of pointwise "+("filter must be 1, but got "+u.shape[1]+".")});var d=l.shape[2],p=l.shape[3];E(u.shape[2]===d*p,function(){return"Error in separableConv2d: the third dimension of pointwise filter "+("must be "+d*p+", ")+("but got "+u.shape[2]+".")});var f=Ta(c,l,i,r,s,a),m=1,g=Er(f,u,m,"valid",s);return h?V(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var Ng=U({separableConv2d_:kx});function Fx(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,l,h;return pe(this,function(d){switch(d.label){case 0:return e=O(n,"x","setdiff1d"),i=O(t,"y","setdiff1d"),E(e.dtype===i.dtype,function(){return"x and y should have the same dtype, but got x ("+e.dtype+") and y ("+i.dtype+")."}),E(e.rank===1,function(){return"x should be 1D tensor, but got x ("+e.shape+")."}),E(i.rank===1,function(){return"y should be 1D tensor, but got y ("+i.shape+")."}),[4,e.data()];case 1:return r=d.sent(),[4,i.data()];case 2:for(a=d.sent(),s=new Set(a),o=0,l=0;l<r.length;l++)s.has(r[l])||o++;for(u=new Ss([o],e.dtype),c=new Ss([o],"int32"),l=0,h=0;l<r.length;l++)s.has(r[l])||(u.values[h]=r[l],c.values[h]=l,h++);return[2,[u.toTensor(),c.toTensor()]]}})})}var Wx=Fx;function Ux(n){var t=O(n,"x","sign"),e={x:t};return z.runKernelFunc(function(i){return i.sign(t)},e,null,Lu)}var xg=U({sign_:Ux});function Bx(n){var t=O(n,"x","sin"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sin(t);return r([t]),a},e,null,wu)}var Gc=U({sin_:Bx});function zx(n){var t=O(n,"x","sinh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sinh(t);return r([t]),a},e,null,Su)}var Yc=U({sinh_:zx});function Px(n,t,e){var i=O(n,"x","slice1d");return E(i.rank===1,function(){return"slice1d expects a rank-1 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,[t],[e])}var _x=U({slice1d_:Px});function Mx(n,t,e){var i=O(n,"x","slice2d");return E(i.rank===2,function(){return"slice2d expects a rank-2 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Hx=U({slice2d_:Mx});function Vx(n,t,e){var i=O(n,"x","slice3d");return E(i.rank===3,function(){return"slice3d expects a rank-3 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var qx=U({slice3d_:Vx});function Gx(n,t,e){var i=O(n,"x","slice4d");return E(i.rank===4,function(){return"slice4d expects a rank-4 tensor, but got a rank-"+i.rank+" tensor"}),ze(i,t,e)}var Yx=U({slice4d_:Gx});function Kx(n,t){t===void 0&&(t=-1);var e=O(n,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and dim was "+t));var i={logits:e},r={dim:t};return z.runKernelFunc(function(a,s){var o=a.softmax(e,t);return s([o]),o},i,null,Ru,r)}var Cg=U({softmax_:Kx});function jx(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.fft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=n.as2D(r,i),s=e.fft(a);return s.reshape(n.shape)},t,null,Zp)}var qs=U({fft_:jx});function $x(n){E(n.dtype==="complex64",function(){return"The dtype for tf.spectral.ifft() must be complex64 "+("but got "+n.dtype+".")});var t={input:n};return z.runKernelFunc(function(e){var i=n.shape[n.shape.length-1],r=n.size/i,a=V(n,[r,i]),s=e.ifft(a);return V(s,n.shape)},t,null,rf)}var Oa=U({ifft_:$x});function Xx(n){var t=n.shape[n.shape.length-1],e=n.size/t,i;if(t<=2){var r=V(n,[e,t]);i=Oa(r)}else{var a=[e,2*(t-1)],s=V(Ca(n),[e,t]),o=V(Us(n),[e,t]),l=$n(ze(s,[0,1],[e,t-2]),1),u=J($n(ze(o,[0,1],[e,t-2]),1),we(-1)),c=xt([s,l],1),h=xt([o,u],1),r=V(ri(c,h),[a[0],a[1]]);i=Oa(r)}if(i=Ca(i),n.rank===3&&n.shape[0]!==0){var d=i,p=n.shape[0];i=V(i,[p,i.shape[0]/p,i.shape[1]]),d.dispose()}return i}var Kc=U({irfft_:Xx});function Rg(n,t,e){e===void 0&&(e=0);var i=[];if(typeof t=="number")E(n.shape[e]%t===0,function(){return"Number of splits must evenly divide the axis."}),i=new Array(t).fill(n.shape[e]/t);else{var r=t.reduce(function(o,l){return l===-1&&(o+=1),o},0);E(r<=1,function(){return"There should be only one negative value in split array."});var a=t.indexOf(-1);if(a!==-1){var s=t.reduce(function(o,l){return l>0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function Jx(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Qe(e,i.shape)[0],c=Rg(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,Cu,s)}var Br=U({split_:Jx});function Zx(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&t<e){var a=n.shape.map(function(v){return 0}),s=n.shape.map(function(v){return v});s[n.shape.length-1]=t,r=ze(n,a,s),e=t}else if(t!=null&&t>e){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=xt([n,Kn(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(ri(r,l),[i,e]),c=qs(u),h=Math.floor(e/2)+1,d=Ca(c),p=Us(c),f=Br(d,[h,e-h],d.shape.length-1),m=Br(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(ri(f[0],m[0]),g)}var Gs=U({rfft_:Zx});function Qx(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,Tu)}var _t=U({sqrt_:Qx});function eC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=at(i,r),i=e[0],r=e[1],et(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,Ou,o)}var Ys=U({squaredDifference_:eC});function tC(n,t){var e=O(n,"x","squeeze");return V(e,Uf(e.shape,t).newShape)}var Ks=U({squeeze_:tC});function nC(n,t){t===void 0&&(t=0);var e=ba(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return fn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Be(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return fn(o,t)});return xt(s,t)}var ji=U({stack_:nC});function iC(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,Pu,r)}var zr=U({step_:iC});function rC(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=Ns(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=Ns(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=Em(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var x=Ns(l);x.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=Lm(t,e,i),R=C.filter(function(W,F){return x.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,Cf,d)}var Og=U({stridedSlice_:rC});function aC(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,Du)}var Eg=U({tan_:aC});function Ea(n,t,e){if(Fi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=Rn(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return ai(n,t,i,e)}function sC(n,t,e){if(Fi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=Rn(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function oC(n,t,e){if(Fi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=Rn(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return ai(n,t,i,e)}function lC(n,t,e){if(Fi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=Rn(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,ai(n,t,i,e)}function uC(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,Rf,s),l=o[0],u=o[1];return{values:l,indices:u}}var Dg=U({topk_:uC});function cC(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new Mc(t,e,i,!0,r),s=On(n,i),o=0;o<s.values.length;o++)s.values[o]=a.nextValue();return s.toTensor()}var hC=U({truncatedNormal_:cC});function dC(n,t){t===void 0&&(t=0);var e=O(n,"x","unique",null);E(e.rank>0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(Of,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var kg=U({unique_:dC});function pC(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(rt(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,Bu,s)}var jc=U({unsortedSegmentSum_:pC});function fC(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t<e.shape.length,function(){return"Axis = "+t+" is not in [-"+e.shape.length+", "+e.shape.length+")"}),t<0&&(t+=e.shape.length);var i={value:e},r={axis:t},a=function(s){return s.unstack(e,t)};return z.runKernelFunc(a,i,null,Uu,r)}var js=U({unstack_:fC});function mC(n,t,e,i){return t===void 0&&(t=!0),z.makeVariable(n,t,e,i)}function Fg(n,t){for(var e=[],i=0;i<t.length;i++)t[i]&&e.push(i);for(var r=On(n,"int32"),a=On([e.length,n.length],"int32"),i=0;i<e.length;i++){var s=r.indexToLoc(e[i]),o=i*n.length;a.values.set(s,o)}return a.toTensor()}function gC(n){return de(this,void 0,void 0,function(){var t,e,i;return pe(this,function(r){switch(r.label){case 0:return t=O(n,"condition","whereAsync","bool"),[4,t.data()];case 1:return e=r.sent(),i=Fg(t.shape,e),n!==t&&t.dispose(),[2,i]}})})}var Wg=gC;function vC(n,t,e){return de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:for(i=O(n,"tensor","boolMask"),r=O(t,"mask","boolMask","bool"),a=e??0,s=r.rank,o=i.shape,E(s>0,function(){return"mask cannot be scalar"}),Be(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u<a+s;u++)l*=o[u];return c=o.slice(0,a).concat([l],o.slice(a+s)),h=V(i,c),d=V(r,[-1]),[4,Wg(d)];case 1:return p=g.sent(),f=Ks(p,[1]),m=Ws(h,f,a),n!==i&&i.dispose(),t!==r&&r.dispose(),f.dispose(),h.dispose(),d.dispose(),p.dispose(),[2,m]}})})}var yC=vC;function bC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","notEqualStrict"),i=O(t,"b","notEqualStrict");return Be(e.shape,i.shape,"Error in notEqualStrict: "),Hs(e,i)}function wC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessStrict"),i=O(t,"b","lessStrict");return Be(e.shape,i.shape,"Error in lessStrict: "),Bs(e,i)}function SC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","equalStrict"),i=O(t,"b","equalStrict");return Be(e.shape,i.shape,"Error in equalStrict: "),Na(e,i)}function LC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","lessEqualStrict"),i=O(t,"b","lessEqualStrict");return Be(e.shape,i.shape,"Error in lessEqualStrict: "),Mi(e,i)}function IC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterStrict"),i=O(t,"b","greaterStrict");return Be(e.shape,i.shape,"Error in greaterStrict: "),hi(e,i)}function AC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","greaterEqualStrict"),i=O(t,"b","greaterEqualStrict");return Be(e.shape,i.shape,"Error in greaterEqualStrict: "),_i(e,i)}var Ug=U({equalStrict_:SC}),Bg=U({greaterEqualStrict_:AC}),zg=U({greaterStrict_:IC}),Pg=U({lessEqualStrict_:LC}),_g=U({lessStrict_:wC}),Mg=U({notEqualStrict_:bC});function TC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","addStrict"),i=O(t,"b","addStrict");return Be(e.shape,i.shape,"Error in addStrict: "),fe(e,i)}function NC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","subStrict"),i=O(t,"b","subStrict");return Be(e.shape,i.shape,"Error in subStrict: "),ge(e,i)}function xC(n,t){return It("strict variants of ops have been deprecated and will be removed in future"),Be(n.shape,t.shape,"Error in powStrict: "),jn(n,t)}function CC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","mul"),i=O(t,"b","mul");return Be(e.shape,i.shape,"Error in multiplyStrict: "),J(e,i)}function RC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","div"),i=O(t,"b","div");return Be(e.shape,i.shape,"Error in divideStrict: "),Ie(e,i)}function OC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","modStrict"),i=O(t,"b","modStrict");return Be(e.shape,i.shape,"Error in modStrict: "),Bc(e,i)}function EC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","minimumStrict"),i=O(t,"b","minimumStrict");return Be(e.shape,i.shape,"Error in minimumStrict: "),Ms(e,i)}function DC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","maximumStrict"),i=O(t,"b","maximumStrict");return Be(e.shape,i.shape,"Error in maximumStrict: "),kr(e,i)}function kC(n,t){It("strict variants of ops have been deprecated and will be removed in future");var e=O(n,"a","squaredDifferenceStrict"),i=O(t,"b","squaredDifferenceStrict");return Be(e.shape,i.shape,"Error in squaredDifferenceStrict: "),Ys(e,i)}var Hg=U({addStrict_:TC}),Vg=U({divStrict_:RC}),qg=U({maximumStrict_:DC}),Gg=U({minimumStrict_:EC}),Yg=U({modStrict_:OC}),Kg=U({mulStrict_:CC}),jg=U({powStrict_:xC}),$g=U({squaredDifferenceStrict_:kC}),Xg=U({subStrict_:NC});function FC(n,t,e,i){t===void 0&&(t="euclidean"),e===void 0&&(e=null),i===void 0&&(i=!1),n=O(n,"x","norm");var r=Jg(n,t,e),a=r.shape;if(i){var s=Qe(e,n.shape);a=Qt(r.shape,s)}return V(r,a)}function Jg(n,t,e){if(e===void 0&&(e=null),n.rank===0)return qt(n);if(n.rank!==1&&e===null)return Jg(V(n,[-1]),t,e);if(n.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return Ae(qt(n),e);if(t===Infinity)return Vi(qt(n),e);if(t===-Infinity)return _s(qt(n),e);if(t==="euclidean"||t===2)return _t(Ae(jn(qt(n),we(2,"int32")),e));throw new Error("Error in norm: invalid ord value: "+t)}if(Array.isArray(e)&&e.length===2){if(t===1)return Vi(Ae(qt(n),e[0]),e[1]-1);if(t===Infinity)return Vi(Ae(qt(n),e[1]),e[0]);if(t===-Infinity)return _s(Ae(qt(n),e[1]),e[0]);if(t==="fro"||t==="euclidean")return _t(Ae(Ye(n),e));throw new Error("Error in norm: invalid ord value: "+t)}throw new Error("Error in norm: invalid axis: "+e)}var $s=U({norm_:FC});function WC(n,t,e,i,r){r===void 0&&(r=!0);var a=O(n,"v","movingAverage"),s=O(t,"x","movingAverage"),o=O(e,"decay","movingAverage");jf(a,s),E(on(a.shape,s.shape),function(){return"Shape mismatch in v and x"});var l=we(1),u=ge(l,o),c=J(ge(s,a),u);if(r){E(i!=null,function(){return"When using zeroDebias: true, step is required."});var h=O(i,"step","movingAverage");c=Ie(c,ge(l,jn(o,h)))}return fe(a,c)}var UC=U({movingAverage_:WC});function BC(n,t,e){var i=O(n,"indices","scatterND","int32"),r=O(t,"updates","scatterND");gc(r,i,e);var a=function(l){return l.scatterND(i,r,e)},s={indices:i,updates:r},o={shape:e};return z.runKernelFunc(a,s,null,Tf,o)}var zC=U({scatterND_:BC});function PC(n,t,e,i){if(n.dtype!=="int32")throw new Error("tf.sparseToDense() expects the indices to be int32 type,"+(" but the dtype was "+n.dtype+"."));if(n.rank>2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function _C(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);PC(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,xf,l)}var MC=U({sparseToDense_:_C});function HC(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,tf)}var VC=U({gatherND_:HC});function qC(n,t){if(t==null)return n.shape.slice();if(on(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i<n.shape.length;i++)t[i]==null&&n.shape[i]!=null?e.push(n.shape[i]):e.push(t[i]);return e}return t}function GC(n,t,e,i){var r=O(n,"x","dropout");if(E(r.dtype==="float32",function(){return"x has to be a floating point tensor since it's going to be "+("scaled, but got a "+r.dtype+" tensor instead.")}),E(t>=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=qC(r,e),s=1-t,o=Ie(Fs(fe(Lg(a,0,1,"float32",i),s)),s);return J(r,o)}var YC=U({dropout_:GC});function Zg(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function $c(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a<n;++a){var s=2*Math.PI*a/(n+i-1);r[a]=t-e*Math.cos(s)}return Ur(r,"float32")}function KC(n,t,e){return e===void 0&&(e=1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,g;return pe(this,function(v){switch(v.label){case 0:return i=O(n,"predictions","inTopK"),r=O(t,"targets","inTopK"),E(i.rank>1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Be(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=gs("bool",u),d=0;d<u;d++){for(p=d*c,f=s.subarray(p,p+c),m=[],g=0;g<f.length;g++)m.push({value:f[g],index:g});for(m.sort(function(b,w){return w.value-b.value}),h[d]=0,g=0;g<e;g++)if(m[g].index===o[d]){h[d]=1;break}}return n!==i&&i.dispose(),t!==r&&r.dispose(),[2,si(h,r.shape,"bool")]}})})}var jC=KC;function $C(n,t,e,i,r,a,s){a===void 0&&(a="NHWC");var o=n;n.rank===3&&(o=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var l=t;l.rank===3&&(l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]])),E(o.rank===4,function(){return"Error in conv2dDerFilter: input must be rank 4, but got shape "+(o.shape+".")}),E(l.rank===4,function(){return"Error in conv2dDerFilter: dy must be rank 4, but got shape "+(l.shape+".")}),E(e.length===4,function(){return"Error in conv2dDerFilter: filterShape must be length 4, but got "+(e+".")});var u=a==="NHWC"?o.shape[3]:o.shape[1],c=a==="NHWC"?l.shape[3]:l.shape[1];E(u===e[2],function(){return"Error in conv2dDerFilter: depth of input "+u+") must "+("match input depth in filter ("+e[2]+".")}),E(c===e[3],function(){return"Error in conv2dDerFilter: depth of dy ("+c+") must "+("match output depth for filter ("+e[3]+").")}),s!=null&&E(rt(r),function(){return"Error in conv2dDerFilter: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var h=function(f){var m=1,g=Sa(a),v=Dn(o.shape,e,i,m,r,s,!1,g);return f.conv2dDerFilter(o,l,v)},d={x:o,dy:l},p={strides:i,pad:r,dataFormat:a,dimRoundingMode:s};return z.runKernelFunc(h,d,null,_p,p)}var Xc=U({conv2DBackpropFilter_:$C});function Xs(n,t,e){if(e==null||e==="linear")return n;if(e==="relu")return J(n,zr(t));throw new Error("Cannot compute gradient for fused activation "+e+".")}function Js(n,t){var e=t,i=mt(n.shape,t.shape);return i.length>0&&(e=Ae(e,i)),V(e,n.shape)}function Zs(n,t,e){if(t==="linear")return n;if(t==="relu")return Ra(n);if(t==="elu")return Rc(n);if(t==="relu6")return Vc(n);if(t==="prelu")return Pc(n,e);throw new Error("Unknown fused activation "+t+".")}var Qs=function(n,t){var e=n>0;return!e||t==="linear"};function XC(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",Qs(z.state.gradientDepth,d)===!1){var f=Er(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(rt(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(Pt(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Dn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var x=function(F,P){var H=P,_=H[0],K=H[1],j=H[2],q=H[3],G=Xs(F,j,d);E(ci(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=Nc(K.shape,G,_,i,r),X=Xc(K,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=Js(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Hu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Hu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var JC=U({fusedConv2d_:XC});function ZC(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,Gp)}var Qg=U({depthwiseConv2dNativeBackpropFilter_:ZC});function QC(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Yp);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var ev=U({depthwiseConv2dNativeBackpropInput_:QC});function eR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(Qs(z.state.gradientDepth,d)===!1){var f=Ta(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),Zs(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(Pt(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(rt(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Dn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=at(S,m)[0],et(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var x=function(F,P){E(ci(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],K=P[2],j=P[3],q=Xs(F,K,d),G=ev(_.shape,q,H,w),Z=Qg(_,q,H.shape,w);if(j!=null){var X=Js(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=kn(function(F,P,H){var _=z.runKernelFunc(C,R,null,Vu,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:x}});return k(v,g)}else{var W=kn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,Vu,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:x}});return W(v,g,S)}}var tR=U({fusedDepthwiseConv2d_:eR});function nR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(Qs(z.state.gradientDepth,c)===!1){var d=We(e,i,a,o);return l!=null&&(d=fe(d,l)),Zs(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=at(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=ot(w),x=ot(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(on(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[x,b,g]):V(f,[x,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=at(k,p)[0],et(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=Xs(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=We(ie,X,!1,!0),re=We(Z,ie,!0,!1)):!a&&o?(te=We(ie,X,!1,!1),re=We(ie,Z,!0,!1)):a&&!o?(te=We(X,ie,!1,!0),re=We(Z,ie,!1,!1)):(te=We(X,ie,!0,!0),re=We(ie,Z,!0,!0)),l!=null){var le=Js(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var K=kn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,Mu,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return K(R,D)}else{var j=kn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,Mu,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return j(R,D,k)}}var iR=U({fusedMatMul_:nR});var rR={__proto__:null,conv2d:JC,depthwiseConv2d:tR,matMul:iR};function aR(n){return $c(n,.54,.46)}var sR=U({hammingWindow_:aR});function oR(n){return $c(n,.5,.5)}var tv=U({hannWindow_:oR});function lR(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a<n.size;){var o=a+t-n.size,l=xt([ze(n,a,t-o),Oc([o],r)]);s.push(l),a+=e}return s.length===0?Ea([],[0,t]):V(xt(s),[s.length,t])}var nv=U({frame_:lR});function uR(n,t,e,i,r){r===void 0&&(r=tv),i==null&&(i=Zg(t));for(var a=nv(n,t,e),s=J(a,r(t)),o=[],l=0;l<a.shape[0];l++)o.push(Gs(ze(s,[l,0],[1,t]),i));return xt(o)}var cR=U({stft_:uR});function hR(n,t,e,i,r,a){var s=O(n,"image","cropAndResize"),o=O(t,"boxes","cropAndResize","float32"),l=O(e,"boxInd","cropAndResize","int32");r=r||"bilinear",a=a||0;var u=o.shape[0];E(s.rank===4,function(){return"Error in cropAndResize: image must be rank 4,"+("but got rank "+s.rank+".")}),E(o.rank===2&&o.shape[1]===4,function(){return"Error in cropAndResize: boxes must be have size ["+u+",4] "+("but had shape "+o.shape+".")}),E(l.rank===1&&l.shape[0]===u,function(){return"Error in cropAndResize: boxInd must be have size ["+u+"] "+("but had shape "+o.shape+".")}),E(i.length===2,function(){return"Error in cropAndResize: cropSize must be of length 2, but got "+("length "+i.length+".")}),E(i[0]>=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Vp,d);return p}var dR=U({cropAndResize_:hR});function pR(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(ef,e,{});return i}var fR=U({flipLeftRight_:pR});function mR(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(Ef,a,s);return o}var gR=U({rotateWithOffset_:mR});function Pr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function vR(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Pr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,vf,l)}var yR=U({nonMaxSuppression_:vR});function wR(n,t,e){var i=bR(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function bR(n,t,e){return LR(n,t,e||SR)}function SR(n,t){return n>t?1:n<t?-1:0}function LR(n,t,e){for(var i=0,r=n.length,a=0,s=!1;i<r;){a=i+(r-i>>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function iv(n,t,e,i,r){return Jc(n,t,e,i,r,0).selectedIndices}function rv(n,t,e,i,r,a){return Jc(n,t,e,i,r,0,!1,a,!0)}function av(n,t,e,i,r,a){return Jc(n,t,e,i,r,a,!0)}function Jc(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;c<t.length;c++)t[c]>r&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(sv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length<e&&u.length>0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m<r)break;for(var b=!1,w=d.length-1;w>=v;--w){var S=IR(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*AR(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&wR(u,f,sv))}var L=d.length,x=e-L;o&&x>0&&(d.push.apply(d,new Array(x).fill(0)),p.push.apply(p,new Array(x).fill(0)));var C={selectedIndices:Ur(d,"int32")};return s&&(C.selectedScores=Ur(p,"float32")),l&&(C.validOutputs=we(L,"int32")),C}function IR(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function AR(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function sv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function TR(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Pr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=iv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var NR=TR;function xR(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(bf,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var CR=U({nonMaxSuppressionWithScore_:xR});function RR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=av(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var OR=RR;function ER(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(yf,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var DR=U({nonMaxSuppressionPadded_:ER});function kR(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Pr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=rv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var FR=kR;function WR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,du,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var ov=U({resizeBilinear_:WR});function UR(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,hu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var lv=U({resizeNearestNeighbor_:UR});function BR(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(Hc(0,s,1,"int32"),[-1,1]),u=Hc(0,o,1,"int32"),c=ge(l,u),h=qi(Mi(c,we(+t,"int32")),_i(c,we(-e,"int32"))),d=Kn([s,o],i.dtype);return V(ji(js(V(i,[-1,s,o])).map(function(p){return dn(h,p,d)})),r)}var zR=U({bandPart_:BR});function PR(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r<n.length;++r)i(r)}else t=!0,n=Br(n,n.shape[0],0).map(function(l){return Ks(l,[0])});E(n.length<=n[0].shape[0],function(){return"Gram-Schmidt: Number of vectors ("+n.length+") exceeds "+("number of dimensions ("+n[0].shape[0]+").")});for(var a=[],s=n,o=function(l){a.push(z.tidy(function(){var u=s[l];if(l>0)for(var c=0;c<l;++c){var h=J(Ae(J(a[c],u)),a[c]);u=ge(u,h)}return Ie(u,$s(u,"euclidean"))}))},r=0;r<n.length;++r)o(r);return t?ji(a,0):a}var _R=U({gramSchmidt_:PR});function MR(n,t){if(t===void 0&&(t=!1),E(n.rank>=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return uv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=js(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=uv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(ji(r,0),n.shape),o=V(ji(a,0),n.shape);return[s,o]}function uv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=ug(e),a=Bi(n),s=Ea([[1]],[1,1]),o=Bi(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=$s(g),b=ze(a,[h,h],[1,1]),w=dn(hi(b,0),Ea([[-1]]),Ea([[1]])),S=ge(b,J(w,v)),L=Ie(g,S);L.shape[0]===1?o=Bi(s):o=xt([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var x=gt(Ie(We(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(x,o),D=ut(o);if(h===0)a=ge(C,We(R,We(D,C)));else{var k=ge(C,We(R,We(D,C)));a=xt([ze(a,[0,0],[h,i]),k],0)}var W=ut(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,We(We(F,o),W));else{var P=ge(F,We(We(F,o),W));r=xt([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],zt([p,f,m])},c=0;c<l;++c)u(c);return!t&&e>i&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var HR=U({qr_:MR});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function VR(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ae(a);if(e===I.Reduction.MEAN){if(r==null)return xa(a);var s=i.size/r.size,o=Ie(Ae(a),Ae(r));return s>1?Ie(o,we(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Ie(Ae(a),we(i.size));var l=J(r,Fr(i.shape)),u=ue(Ae(Hs(l,we(0))),"float32");return Ie(Ae(a),u)}throw Error("Unknown reduction: "+e)}var Xn=U({computeWeightedLoss_:VR});function qR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Be(r.shape,a.shape,"Error in absoluteDifference: ");var o=qt(ge(r,a));return Xn(o,s,i)}var GR=U({absoluteDifference_:qR});function YR(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Be(a.shape,s.shape,"Error in cosineDistance: ");var l=we(1),u=ge(l,Ae(J(a,s),e,!0));return Xn(u,o,r)}var KR=U({cosineDistance_:YR});function jR(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Be(r.shape,a.shape,"Error in hingeLoss: ");var o=we(1);r=ge(J(we(2),r),o);var l=Ra(ge(o,J(r,a)));return Xn(l,s,i)}var $R=U({hingeLoss_:jR});function XR(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Be(a.shape,s.shape,"Error in huberLoss: ");var l=we(i),u=qt(ge(s,a)),c=Ms(u,l),h=ge(u,c),d=fe(J(we(.5),Ye(c)),J(l,h));return Xn(d,o,r)}var JR=U({huberLoss_:XR});function ZR(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Be(a.shape,s.shape,"Error in logLoss: ");var l=we(1),u=we(i),c=gt(J(a,Hi(fe(s,u)))),h=J(ge(l,a),Hi(fe(ge(l,s),u))),d=ge(c,h);return Xn(d,o,r)}var QR=U({logLoss_:ZR});function eO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Be(r.shape,a.shape,"Error in meanSquaredError: ");var o=Ys(r,a);return Xn(o,s,i)}var tO=U({meanSquaredError_:eO});function nO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Be(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ra(i),a=J(i,e),s=Dc(pn(gt(qt(i))));return fe(ge(r,a),s)}function iO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Be(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=nO(a,s);return Xn(h,o,r)}var rO=U({sigmoidCrossEntropy_:iO});function aO(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=kn(function(r,a,s){var o=!0,l=Fc(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=gt(J(u,r)),h=Ae(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=Qt(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),pn(g))),J(V(p,v),ge(pn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function sO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Be(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=we(i),u=we(1),c=we(a.shape[1]);a=fe(J(a,ge(u,l)),Ie(l,c))}var h=aO(a,s);return Xn(h,o,r)}var oO=U({softmaxCrossEntropy_:sO});var lO={fft:qs,ifft:Oa,rfft:Gs,irfft:Kc},uO={hammingWindow:sR,hannWindow:tv,frame:nv,stft:cR},cO={flipLeftRight:fR,resizeNearestNeighbor:lv,resizeBilinear:ov,rotateWithOffset:gR,cropAndResize:dR,nonMaxSuppression:yR,nonMaxSuppressionAsync:NR,nonMaxSuppressionWithScore:CR,nonMaxSuppressionWithScoreAsync:OR,nonMaxSuppressionPadded:DR,nonMaxSuppressionPaddedAsync:FR},hO={bandPart:zR,gramSchmidt:_R,qr:HR},dO={absoluteDifference:GR,computeWeightedLoss:Xn,cosineDistance:KR,hingeLoss:$R,huberLoss:JR,logLoss:QR,meanSquaredError:tO,sigmoidCrossEntropy:rO,softmaxCrossEntropy:oO};var di=function(n){qn(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return zt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return gg(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&zt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:we(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(km);Object.defineProperty(di,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Zc=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;ft(function(){var d=fe(J(c,i.rho),J(Ye(u),1-i.rho)),p=J(Ie(_t(fe(h,i.epsilon)),_t(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ye(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(zt(this.accumulatedGrads.map(function(e){return e.variable})),zt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(di);ui(Zc);var Qc=function(n){qn(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:ft(function(){return Oc(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;ft(function(){var h=fe(c,Ye(u));c.assign(h);var d=fe(J(Ie(u,_t(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&zt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(di);ui(Qc);var eh=function(n){qn(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],ft(function(){s.accBeta1=we(i).variable(),s.accBeta2=we(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:ft(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:ft(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ye(h),1-i.beta2)),g=Ie(f,a),v=Ie(m,s);d.assign(f),p.assign(m);var b=fe(J(Ie(g,fe(_t(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&zt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),ft(function(){a.accBeta1.assign(jn(a.beta1,a.iterations_+1)),a.accBeta2.assign(jn(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(di);ui(eh);var th=function(n){qn(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],ft(function(){o.iteration=we(0).variable(),o.accBeta1=we(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=Ie(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=qt(h),v=kr(m,g);d.assign(f),p.assign(v);var b=fe(J(Ie(s,a),Ie(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&zt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&zt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(di);ui(th);var eo=function(n){qn(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];ft(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=Um(we(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(di);ui(eo);var nh=function(n){qn(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=we(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;ft(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&zt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(eo);ui(nh);var ih=function(n){qn(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;ft(function(){var d=fe(J(c,i.decay),J(Ye(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Ie(J(u,i.learningRate),_t(ge(d,fe(Ye(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ye(u),1-i.decay)),g=fe(J(h,i.momentum),Ie(J(u,i.learningRate),_t(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&zt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&zt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&zt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(di);ui(ih);var $i=function(){function n(){}return n.sgd=function(t){return new eo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new nh(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new ih(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new eh(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Zc(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new th(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Qc(t,e)},n}();var pO={sgd:$i.sgd,momentum:$i.momentum,adadelta:$i.adadelta,adagrad:$i.adagrad,rmsprop:$i.rmsprop,adamax:$i.adamax,adam:$i.adam};var fO=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function mO(){return new Promise(function(n){return fO(function(){return n()})})}function gO(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function vO(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s<a;++s)r=r.concat([n[s+1]/t[s],t[s]]);r=r.concat(n.slice(a+1))}return r}function yO(n,t,e){e===void 0&&(e=!0);var i=[];if(e){i.push(t);for(var r=t+1;r<n;++r)r<=2*t?(i.push(r),i.push(r-(t+1))):i.push(r)}else{for(var a=[],s=[],r=1;r<n;++r)r>=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function bO(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a<n.length;++a)a<=t.length?i?r.push(t[a-1]*n[a]):r.push(n[a]/t[a-1]):r.push(n[a]);return r}function wO(n,t){for(var e=[0],i=0;i<t;++i)e.push(n[i][0]);return e}function SO(n,t,e){for(var i=n.slice(0,1),r=0;r<e;++r)i.push(n[r+1]-t[r][0]-t[r][1]);return i}var cv=1.7580993408473768,hv=1.0507009873554805;var LO=.3275911,IO=.254829592,AO=-.284496736,TO=1.421413741,NO=-1.453152027,xO=1.061405429;function CO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.warn.apply(console,n)}function RO(){for(var n=[],t=0;t<arguments.length;t++)n[t]=arguments[t];He().getBool("IS_TEST")||console.log.apply(console,n)}function OO(n,t){if(n.length!==t.length)throw new Error("Cannot merge real and imag arrays of different lengths. real:"+(n.length+", imag: "+t.length+"."));for(var e=new Float32Array(n.length*2),i=0;i<e.length;i+=2)e[i]=n[i/2],e[i+1]=t[i/2];return e}function EO(n){for(var t=new Float32Array(n.length/2),e=new Float32Array(n.length/2),i=0;i<n.length;i+=2)t[i/2]=n[i],e[i/2]=n[i+1];return{real:t,imag:e}}function DO(n){for(var t=Math.ceil(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=0;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function kO(n){for(var t=Math.floor(n.length/4),e=new Float32Array(t),i=new Float32Array(t),r=2;r<n.length;r+=4)e[Math.floor(r/4)]=n[r],i[Math.floor(r/4)]=n[r+1];return{real:e,imag:i}}function FO(n,t){var e=n[t*2],i=n[t*2+1];return{real:e,imag:i}}function WO(n,t,e,i){n[i*2]=t,n[i*2+1]=e}function UO(n,t){for(var e=new Float32Array(n/2),i=new Float32Array(n/2),r=0;r<Math.ceil(n/2);r++){var a=(t?2:-2)*Math.PI*(r/n);e[r]=Math.cos(a),i[r]=Math.sin(a)}return{real:e,imag:i}}function BO(n,t,e){var i=(e?2:-2)*Math.PI*(n/t),r=Math.cos(i),a=Math.sin(i);return{real:r,imag:a}}function zO(n,t,e){if(t==="complex64"){if(n.dtype==="complex64")return n.clone();var i=Kn(n.shape),r=ue(n,"float32"),a=e.complex(r,i);return i.dispose(),r.dispose(),a}if(!_f(n.dtype,t))return z.makeTensorFromDataId(n.dataId,n.shape,t);if(n.dtype==="complex64"){var s=e.real(n),a=ue(s,t);return s.dispose(),a}if(t==="int32")return e.int(n);if(t==="bool"){var o=we(0,n.dtype),a=e.notEqual(n,o);return o.dispose(),a}else throw new Error("Error in Cast: failed to cast "+n.dtype+" to "+t)}function PO(n,t){return z.makeTensorFromDataId(n.dataId,t,n.dtype)}function _O(n,t,e){var i=(t-n)/(e-1),r=Lr(e,"float32");r[0]=n;for(var a=1;a<r.length;a++)r[a]=r[a-1]+i;return Ur(r,"float32")}var MO={__proto__:null,slice_util:Dm,segment_util:M1,castTensor:zO,reshapeTensor:PO,linspaceImpl:_O,upcastType:Ls,axesAreInnerMostDims:Lc,combineLocations:Pm,computeOutAndReduceShapes:_m,expandShapeToKeepDim:Qt,assertAxesAreInnerMostDims:vT,getAxesPermutation:en,getUndoAxesPermutation:xs,getInnerMostAxes:En,getBroadcastDims:A1,getReductionAxes:mt,assertAndGetBroadcastShape:et,assertParamsConsistent:Xm,computeOutShape:Jm,computeDilation2DInfo:xT,computePool2DInfo:Rr,computePool3DInfo:Ia,computeConv2DInfo:Dn,computeConv3DInfo:La,computeDefaultPad:Ac,tupleValuesAreOne:ci,eitherStridesOrDilationsAreOne:Pt,convertConv2DDataFormat:Sa,getFusedDyActivation:Xs,getFusedBiasGradient:Js,applyActivation:Zs,shouldFuse:Qs,PARALLELIZE_THRESHOLD:Ec,computeOptimalWindowSize:z1,getImageCenter:gO,getReshaped:vO,getPermuted:yO,getReshapedPermuted:bO,getSliceBeginCoords:wO,getSliceSize:SO,prepareAndValidate:bm,validateUpdateShape:mc,validateInput:gc,calculateShapes:wm,SELU_SCALEALPHA:cv,SELU_SCALE:hv,ERF_P:LO,ERF_A1:IO,ERF_A2:AO,ERF_A3:TO,ERF_A4:NO,ERF_A5:xO,warn:CO,log:RO,mergeRealAndImagArrays:OO,splitRealAndImagArrays:EO,complexWithEvenIndex:DO,complexWithOddIndex:kO,getComplexWithIndex:FO,assignToTypedArray:WO,exponents:UO,exponent:BO,prepareSplitSize:Rg};function HO(n,t,e){var i=new Array(n.rank).fill(0),r=n.shape.slice();return t.map(function(a){var s=r.slice();s[e]=a;var o=ze(n,i,s);return i[e]+=a,o})}function VO(n,t){for(var e=new Array(n.rank),i=0;i<e.length;i++)e[i]=n.shape[i]*t[i];for(var r=On(e,n.dtype),i=0;i<r.values.length;++i){for(var a=r.indexToLoc(i),s=new Array(n.rank),o=0;o<s.length;o++)s[o]=a[o]%n.shape[o];var l=n.locToIndex(s);r.values[i]=n.values[l]}return r.toTensor()}function qO(n,t,e,i,r){for(var a=t[t.length-1],s=[n.length/a,a],o=s[0],l=s[1],u=gs(e,o*i),c=gs("int32",o*i),h=0;h<o;h++){for(var d=h*l,p=n.subarray(d,d+l),f=[],m=0;m<p.length;m++)f.push({value:p[m],index:m});f.sort(function(S,L){return L.value-S.value});for(var g=h*i,v=u.subarray(g,g+i),b=c.subarray(g,g+i),m=0;m<i;m++)v[m]=f[m].value,b[m]=f[m].index}var w=t.slice();return w[w.length-1]=i,[si(u,w,e),si(c,w,"int32")]}var GO={__proto__:null,nonMaxSuppressionV3Impl:iv,nonMaxSuppressionV4Impl:rv,nonMaxSuppressionV5Impl:av,split:HO,tile:VO,topkImpl:qO,whereImpl:Fg};var YO={kernelName:il,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,zr(ue(e,"float32"),-1))}}}};var KO={kernelName:rl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Ye(ue(e,"float32")),r=_t(ge(we(1),i));return gt(Ie(n,r))}}}};var jO={kernelName:al,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(ge(Ye(ue(e,"float32")),1));return Ie(n,i)}}}};var $O={kernelName:ps,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,i.shape)};return{a,b:s}}};var XO={kernelName:sl,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var JO={kernelName:ol,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var ZO={kernelName:ll,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var QO={kernelName:ul,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,_t(ge(we(1),Ye(ue(e,"float32")))))}}}};var eE={kernelName:cl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=_t(fe(we(1),Ye(ue(e,"float32"))));return Ie(n,i)}}}};var tE={kernelName:pl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=fe(Ye(e),Ye(i)),l=J(n,Ie(i,o)),u=mt(e.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,e.shape)},s=function(){var o=fe(Ye(e),Ye(i)),l=gt(J(n,Ie(e,o))),u=mt(i.shape,r);return u.length>0&&(l=Ae(l,u)),V(l,i.shape)};return{a,b:s}}};var nE={kernelName:hl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(Ye(ue(e,"float32")),1))}}}};var iE={kernelName:dl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ge(we(1),Ye(ue(e,"float32"))))}}}};function rE(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(Pt(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(rt(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=Ia(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,zp,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var aE=U({avgPool3dBackprop_:rE});var sE={kernelName:ml,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return aE(n,i,a,s,c,l,u)}}}};function oE(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=Rr(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,Bp,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var lE=U({avgPoolBackprop_:oE});var uE={kernelName:fl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return lE(n,i,a,s,o)}}}};var cE={kernelName:gl,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return We(n,a,!1,!0)},b:function(){return We(r,n,!0,!1)}}:!o&&l?{a:function(){return We(n,a,!1,!1)},b:function(){return We(n,r,!0,!1)}}:o&&!l?{a:function(){return We(a,n,!1,!0)},b:function(){return We(r,n,!1,!1)}}:{a:function(){return We(a,n,!0,!0)},b:function(){return We(n,r,!0,!0)}}}};var hE={kernelName:vl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return Vs(n,r,a)}}}};var dE={kernelName:yl,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o<s.length;o++)s[o]>1&&l.push(o);return{x:function(){return Ae(n,l,!0)}}}};var pE={kernelName:fs,gradFunc:function(n){return{x:function(){return n.clone()}}}};var fE={kernelName:bl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var mE={kernelName:wl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return dn(qi(_i(i,a),Mi(i,s)),n,Ee(n))}}}};var gE={kernelName:Sl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Qe(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=Br(n,s,a);return o.map(function(l){return function(){return l}})}};var vE={kernelName:Ll,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(ci(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return Nc(r.shape,n,a,l,u,c)},filter:function(){return Xc(r,n,a.shape,l,u,c)}}}};var yE={kernelName:Il,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Er(n,a,o,l,u,1,c)},filter:function(){return Xc(n,r,a.shape,o,l,u,c)}}}};function bE(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=La(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,Mp,u)}var wE=U({conv3DBackpropFilter_:bE});var SE={kernelName:Al,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(ci(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return ng(o.shape,n,l,a,s)},filter:function(){return wE(o,n,l.shape,a,s)}}}};var LE={kernelName:Tl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(gt(Gc(ue(e,"float32"))),n)}}}};var IE={kernelName:Nl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Yc(ue(e,"float32")),n)}}}};var AE={kernelName:xl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=en([a],i.rank),u=Cc(n,a,s,!o);return l!=null&&(u=ut(u,l)),u}}}};var TE={kernelName:Cl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(ci(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(Pt(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be  1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(rt(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Dn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return ev(c.shape,n,h,d)},filter:function(){return Qg(c,n,h.shape,d)}}}};var NE={kernelName:Rl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(jp,s,e)},filter:function(){return z.runKernel($p,o,e)}}}};var xE={kernelName:Ol,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var CE={kernelName:El,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Xp)}}}};var RE={kernelName:Dl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(pn(gt(Ye(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var OE={kernelName:kl,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var EE={kernelName:Fl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,pn(e))}}}};var DE={kernelName:Wl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var kE={kernelName:Ul,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=Ie(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ae(o,l),i.shape));var u=Ye(i);return gt(Ie(o,ue(u,"float32")))};return{a,b:s}}};var FE={kernelName:Bl,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??we(1),u=mt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h<r.shape.length-1;++h)c.push(r.shape[h]);c.push(1)}var d=ge(r,a),p=J(n,l),f=qc(fe(s,we(i))),m=J(J(J(f,f),f),we(-.5)),g=function(){return a.rank===1?V(J(J(n,Dr(V(f,[1,1,1,a.shape[0]]),c)),l),r.shape):V(J(J(n,f),l),r.shape)},v=function(){var L=J(J(f,we(-1)),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},b=function(){var L=J(J(m,d),p);return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)},w=function(){var L=J(d,f),x=J(n,L);return a.rank===1&&(x=Ae(x,u)),V(x,a.shape)},S=function(){var L=n;return a.rank===1&&(L=Ae(L,u)),V(L,a.shape)};return{x:g,mean:v,variance:b,scale:w,offset:S}}};var WE={kernelName:zl,inputsToSave:["x","indices"],gradFunc:function(n,t,e){var i=t[0],r=t[1],a=e.axis,s=Qe(a,i.shape)[0],o=function(){var l=i.shape,u=r.size,c=l.slice(0,s),h=c.length,d=l.slice(a,l.length).slice(1),p=d.length,f=dv(0,h),m=dv(h+1,h+1+p),g=pv([c,[u],d]),v=V(n,g),b=V(r,[u]),w=pv([[h],f,m]),S=ut(v,w),L=jc(S,b,i.shape[s]),x=xs(w);return L=ut(L,x),L};return{x:o,indices:function(){return r}}}};function dv(n,t){for(var e=[],i=n;i<t;++i)e.push(i);return e}function pv(n){for(var t=[],e=0;e<n.length;++e)for(var i=0;i<n[e].length;++i)t.push(n[e][i]);return t}var UE={kernelName:Pl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1];return{a:function(){return Ee(e)},b:function(){return Ee(i)}}}};var BE={kernelName:_l,gradFunc:function(n){return{x:function(){return ue(n,"float32")}}}};var zE={kernelName:Ml,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var PE={kernelName:Hl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var _E={kernelName:Vl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ME={kernelName:Gl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,fe(e,1))}}}};var HE={kernelName:ql,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,ue(e,"float32"))}}}};var VE={kernelName:Yl,inputsToSave:[],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.axis;return{logits:function(){var a=!0,s=pn(i);return ge(n,J(Ae(n,r,a),s))}}}};function qE(n,t,e,i,r,a,s){i===void 0&&(i=5),r===void 0&&(r=1),a===void 0&&(a=1),s===void 0&&(s=.5);var o=function(c){return c.LRNGrad(e,n,t,i,r,a,s)},l={x:n,y:t,dy:e},u={depthRadius:i,bias:r,alpha:a,beta:s};return z.runKernelFunc(o,l,null,df,u)}var GE=U({localResponseNormalizationBackprop_:qE});var YE={kernelName:Kl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.depthRadius,l=s.bias,u=s.alpha,c=s.beta;return{x:function(){return GE(r,a,n,o,l,u,c)}}}};function fv(n,t,e,i,r){return t.rank<e.rank&&(t=V(t,Qt(t.shape,i))),n.rank<e.rank&&(n=V(n,Qt(n.shape,i))),{x:function(){var a=J(n,ue(Na(e,t),n.dtype));return r==null?a:ut(a,r)}}}var mv={kernelName:jl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.reductionIndices,a=t[0],s=t[1],o=Qe(r,a.shape),l=en(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var KE={kernelName:$l,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(_i(e,i),"float32"))},a=function(){return J(n,ue(Bs(e,i),"float32"))};return{a:r,b:a}}};function jE(n,t,e,i,r,a,s,o){a===void 0&&(a=[1,1,1]);var l=O(n,"dy","maxPool3dBackprop"),u=O(t,"input","maxPool3dBackprop"),c=O(e,"output","maxPool3dBackprop"),h=l,d=u,p=c,f=!1;u.rank===4&&(f=!0,h=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=V(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),p=V(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),E(h.rank===5,function(){return"Error in maxPool3dBackprop: dy must be rank 5 but got rank "+(h.rank+".")}),E(d.rank===5,function(){return"Error in maxPool3dBackprop: input must be rank 5 but got rank "+(d.rank+".")}),E(p.rank===5,function(){return"Error in maxPool3dBackprop: output must be rank 5 but got rank "+(p.rank+".")}),E(Pt(r,a),function(){return"Error in maxPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+r+" and dilations '"+a+"'")}),o!=null&&E(rt(s),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+o+" but got pad "+s+".")});var m=function(w){var S=Ia(d.shape,i,r,a,s,o);return w.maxPool3dBackprop(h,d,p,S)},g={dy:h,input:d,output:p},v={filterSize:i,strides:r,dilations:a,pad:s,dimRoundingMode:o},b=z.runKernelFunc(m,g,null,ff,v);return f?V(b,[b.shape[1],b.shape[2],b.shape[3],b.shape[4]]):b}var $E=U({maxPool3dBackprop_:jE});var XE={kernelName:Jl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.dilations,c=s.pad,h=s.dimRoundingMode,d=u??[1,1,1];return{x:function(){return $E(n,r,a,o,l,d,c,h)}}}};function JE(n,t,e,i,r,a,s){var o=O(n,"dy","maxPoolBackprop"),l=O(t,"input","maxPoolBackprop"),u=O(e,"output","maxPoolBackprop");E(l.rank===o.rank,function(){return"Rank of input ("+l.rank+") does not match rank of dy "+("("+o.rank+")")}),E(o.rank===4,function(){return"Error in maxPoolBackprop: dy must be rank 4 but got rank "+(o.rank+".")}),E(l.rank===4,function(){return"Error in maxPoolBackprop: input must be rank 4 but got rank "+(l.rank+".")}),s!=null&&E(rt(a),function(){return"Error in maxPoolBackprop: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+a+".")});var c=function(p){var f=Rr(l.shape,i,r,1,a,s);return p.maxPoolBackprop(o,l,u,f)},h={dy:o,input:l,output:u},d={filterSize:i,strides:r,pad:a,dimRoundingMode:s};return z.runKernelFunc(c,h,null,pf,d)}var ZE=U({maxPoolBackprop_:JE});var QE={kernelName:Xl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.filterSize,l=s.strides,u=s.pad;return{x:function(){return ZE(n,r,a,o,l,u)}}}};var eD={kernelName:Zl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:function(n,t,e){var i=e,r=i.axis,a=t[0],s=t[1],o=Qe(r,a.shape),l=en(o,a.rank),u=fv(n,s,a,o,l);return{x:function(){var c=u.x();return l!=null&&(c=ut(c)),c}}}};var tD={kernelName:Ql,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=function(){return J(n,ue(Mi(e,i),"float32"))},a=function(){return J(n,ue(hi(e,i),"float32"))};return{a:r,b:a}}};var nD={kernelName:eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=mt(e.shape,r);return o.length>0?V(Ae(n,o),e.shape):n},s=function(){var o=J(n,gt(Fs(Ie(e,i)))),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var iD={kernelName:tu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ae(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);return l.length>0?V(Ae(o,l),i.shape):o};return{a,b:s}}};var rD={kernelName:nu,gradFunc:function(n){return{x:function(){return gt(n)}}}};var aD={kernelName:ru,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return Kn(e.shape,"float32")}}}};var sD={kernelName:iu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var gv={kernelName:au,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var oD={kernelName:su,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=et(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,jn(a,ge(c,we(1))))),d=mt(a.shape,o);return d.length>0&&(h=Ae(h,d)),V(h,a.shape)},u=function(){var c=hi(a,0),h=dn(c,Hi(a),Ee(a)),d=J(n,J(r,h)),p=mt(s.shape,o);return p.length>0&&(d=Ae(d,p)),V(d,s.shape)};return{a:l,b:u}}};var lD={kernelName:ou,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=hi(e,0);return{x:function(){return dn(r,n,J(n,i))},alpha:function(){var a=dn(r,Ee(n),J(n,e)),s=mt(i.shape,n.shape);return s.length>0&&(a=Ae(a,s)),V(a,i.shape)}}}};var uD={kernelName:lu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,gt(Ye(e)))}}}};var cD={kernelName:pu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(Mi(e,6),zr(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var hD={kernelName:uu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue(zr(e),"float32"))}}}};var dD={kernelName:cu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var pD={kernelName:du,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,Af,e)};return{images:s}}};var fD={kernelName:hu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,If,e)};return{images:s}}};var mD={kernelName:fu,gradFunc:function(n,t,e){var i=e.dims,r=Qe(i,n.shape);return{x:function(){return $n(n,r)}}}};var gD={kernelName:mu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var vD={kernelName:gu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return gt(Ie(n,J(jn(e,1.5),2)))}}}};var yD={kernelName:vu,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(Ps(e),n.dtype))}}}};var bD={kernelName:yu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=hi(e,we(0)),r=we(cv),a=we(hv),s=J(n,a),o=J(J(n,r),pn(ue(e,"float32")));return dn(i,s,o)}}}};var wD={kernelName:Iu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(we(1),e)))}}}};var SD={kernelName:Lu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var LD={kernelName:wu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ks(ue(e,"float32")),n)}}}};var ID={kernelName:Su,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(xc(ue(e,"float32")),n)}}}};var AD={kernelName:bu,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=vc(i,a,s),u=l[0],c=l[1],h=[],d=0;d<n.rank;d++)h.push([u[d],o[d]-u[d]-c[d]]);return{x:function(){return Gi(n,h)}}}};var TD={kernelName:Ru,outputsToSave:[!0],gradFunc:function(n,t,e){var i=t[0],r=e.dim,a=!0,s=J(n,i);return{logits:function(){return ge(s,J(Ae(s,[r],a),i))}}}};var ND={kernelName:Au,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,Pi(e))}}}};var vv={kernelName:xu,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.paddings;return{x:function(){return Os(n,r,a)}}}};var yv={kernelName:Cu,gradFunc:function(n,t,e){var i=e.axis;return{x:function(){return xt(n,i)}}}};var xD={kernelName:Tu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,J(_t(ue(e,"float32")),2))}}}};var CD={kernelName:Nf,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(ue(e,"float32"),2))}}}};var RD={kernelName:Ou,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=we(2),a=function(){return J(n,J(r,ge(e,i)))},s=function(){return J(n,J(r,ge(i,e)))};return{a,b:s}}};var OD={kernelName:Pu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var ED={kernelName:Eu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=et(e.shape,i.shape),a=function(){var o=n,l=mt(e.shape,r);return l.length>0&&(o=Ae(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ae(o,l)),V(gt(o),i.shape)};return{a,b:s}}};var DD={kernelName:Nu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Qe(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Fr(i.shape,"float32"));return{x:function(){return l}}}};var kD={kernelName:Du,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ie(n,Ye(ks(e)))}}}};var FD={kernelName:ku,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(we(1),Ye(e)),n)}}}};var WD={kernelName:Fu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o<r[0];++o)s=fe(s,ze(n,[o*i.shape[0]],[i.shape[0]]));else if(i.rank===2)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1]],[i.shape[0],i.shape[1]]));else if(i.rank===3)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2]],[i.shape[0],i.shape[1],i.shape[2]]));else if(i.rank===4)for(var o=0;o<r[0];++o)for(var l=0;l<r[1];++l)for(var u=0;u<r[2];++u)for(var c=0;c<r[3];++c)s=fe(s,ze(n,[o*i.shape[0],l*i.shape[1],u*i.shape[2],c*i.shape[3]],[i.shape[0],i.shape[1],i.shape[2],i.shape[3]]));else throw new Error("Gradient for tile operation is not implemented for rank-"+(i.rank+" tensors yet."));return s};return{x:a}}};var UD={kernelName:Wu,gradFunc:function(n,t,e){var i=e,r=i.perm,a=xs(r);return{x:function(){return ut(n,a)}}}};var BD={kernelName:Uu,gradFunc:function(n,t,e){var i=e,r=i.axis;return{value:function(){return ji(n,r)}}}};var PD={kernelName:Bu,inputsToSave:["segmentIds"],gradFunc:function(n,t){var e=t[0],i=function(){return zD(n,e)};return{x:i}}};function zD(n,t){for(var e=kr(t,Ee(t)),i=Ws(n,e),r=_i(t,we(0,"int32")),a=i.rank-r.rank,s=0;s<a;++s)r=fn(r,s+1);r=qi(r,Fr(i.shape,"bool"));var o=Ee(i);return dn(r,i,o)}var _D={kernelName:zu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var MD=[YO,KO,jO,$O,XO,JO,ZO,QO,eE,tE,nE,iE,sE,uE,cE,hE,dE,pE,fE,mE,gE,yE,vE,SE,LE,IE,AE,TE,NE,xE,CE,RE,OE,EE,kE,DE,FE,WE,UE,BE,zE,PE,_E,ME,HE,VE,YE,mv,mv,KE,XE,QE,eD,tD,nD,iD,rD,aD,sD,gv,gv,oD,lD,uD,cD,hD,dD,pD,fD,mD,gD,vD,yD,bD,wD,SD,LD,ID,AD,TD,ND,vv,vv,yv,yv,xD,RD,CD,OD,ED,DD,kD,FD,WD,UD,BD,PD,_D];for(var rh=0,bv=MD;rh<bv.length;rh++){var HD=bv[rh];kf(HD)}Y.prototype.abs=function(){return this.throwIfDisposed(),qt(this)};Y.prototype.acos=function(){return this.throwIfDisposed(),Bm(this)};Y.prototype.acosh=function(){return this.throwIfDisposed(),zm(this)};Y.prototype.addStrict=function(n){return this.throwIfDisposed(),Hg(this,n)};Y.prototype.add=function(n){return this.throwIfDisposed(),fe(this,n)};Y.prototype.all=function(n,t){return this.throwIfDisposed(),Mm(this,n,t)};Y.prototype.any=function(n,t){return this.throwIfDisposed(),Hm(this,n,t)};Y.prototype.argMax=function(n){return this.throwIfDisposed(),Vm(this,n)};Y.prototype.argMin=function(n){return this.throwIfDisposed(),qm(this,n)};Y.prototype.asScalar=function(){return this.throwIfDisposed(),E(this.size===1,function(){return"The array must have only 1 element."}),V(this,[])};Y.prototype.asType=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.as1D=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.as2D=function(n,t){return this.throwIfDisposed(),V(this,[n,t])};Y.prototype.as3D=function(n,t,e){return this.throwIfDisposed(),V(this,[n,t,e])};Y.prototype.as4D=function(n,t,e,i){return this.throwIfDisposed(),V(this,[n,t,e,i])};Y.prototype.as5D=function(n,t,e,i,r){return this.throwIfDisposed(),V(this,[n,t,e,i,r])};Y.prototype.asin=function(){return this.throwIfDisposed(),Gm(this)};Y.prototype.asinh=function(){return this.throwIfDisposed(),Ym(this)};Y.prototype.atan=function(){return this.throwIfDisposed(),Km(this)};Y.prototype.atan2=function(n){return this.throwIfDisposed(),jm(this,n)};Y.prototype.atanh=function(){return this.throwIfDisposed(),$m(this)};Y.prototype.avgPool=function(n,t,e,i){return this.throwIfDisposed(),Tc(this,n,t,e,i)};Y.prototype.batchToSpaceND=function(n,t){return this.throwIfDisposed(),Os(this,n,t)};Y.prototype.batchNorm=function(n,t,e,i,r){return this.throwIfDisposed(),Aa(this,n,t,e,i,r)};Y.prototype.broadcastTo=function(n){return this.throwIfDisposed(),Ds(this,n)};Y.prototype.cast=function(n){return this.throwIfDisposed(),ue(this,n)};Y.prototype.ceil=function(){return this.throwIfDisposed(),Zm(this)};Y.prototype.clipByValue=function(n,t){return this.throwIfDisposed(),Qm(this,n,t)};Y.prototype.concat=function(n,t){return this.throwIfDisposed(),n instanceof Y&&(n=[n]),xt([this].concat(n),t)};Y.prototype.conv1d=function(n,t,e,i,r,a){return this.throwIfDisposed(),eg(this,n,t,e,i,r,a)};Y.prototype.conv2dTranspose=function(n,t,e,i,r){return this.throwIfDisposed(),tg(this,n,t,e,i,r)};Y.prototype.conv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Er(this,n,t,e,i,r,a)};Y.prototype.cos=function(){return this.throwIfDisposed(),ks(this)};Y.prototype.cosh=function(){return this.throwIfDisposed(),xc(this)};Y.prototype.cumsum=function(n,t,e){return this.throwIfDisposed(),Cc(this,n,t,e)};Y.prototype.depthToSpace=function(n,t){return this.throwIfDisposed(),ig(this,n,t)};Y.prototype.depthwiseConv2D=function(n,t,e,i,r,a){return It("depthwiseConv2D is deprecated, use depthwiseConv2d instead"),this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.depthwiseConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ta(this,n,t,e,i,r,a)};Y.prototype.dilation2d=function(n,t,e,i,r){return this.throwIfDisposed(),rg(this,n,t,e,i,r)};Y.prototype.divNoNan=function(n){return this.throwIfDisposed(),ag(this,n)};Y.prototype.divStrict=function(n){return this.throwIfDisposed(),Vg(this,n)};Y.prototype.div=function(n){return this.throwIfDisposed(),Ie(this,n)};Y.prototype.dot=function(n){return this.throwIfDisposed(),sg(this,n)};Y.prototype.elu=function(){return this.throwIfDisposed(),Rc(this)};Y.prototype.equalStrict=function(n){return this.throwIfDisposed(),Ug(this,n)};Y.prototype.equal=function(n){return this.throwIfDisposed(),Na(this,n)};Y.prototype.erf=function(){return this.throwIfDisposed(),og(this)};Y.prototype.exp=function(){return this.throwIfDisposed(),pn(this)};Y.prototype.expandDims=function(n){return this.throwIfDisposed(),fn(this,n)};Y.prototype.expm1=function(){return this.throwIfDisposed(),lg(this)};Y.prototype.fft=function(){return this.throwIfDisposed(),qs(this)};Y.prototype.flatten=function(){return this.throwIfDisposed(),V(this,[this.size])};Y.prototype.floor=function(){return this.throwIfDisposed(),Fs(this)};Y.prototype.floorDiv=function(n){return this.throwIfDisposed(),Sc(this,n)};Y.prototype.gather=function(n,t){return this.throwIfDisposed(),Ws(this,n,t)};Y.prototype.greaterEqualStrict=function(n){return this.throwIfDisposed(),Bg(this,n)};Y.prototype.greaterEqual=function(n){return this.throwIfDisposed(),_i(this,n)};Y.prototype.greaterStrict=function(n){return this.throwIfDisposed(),zg(this,n)};Y.prototype.greater=function(n){return this.throwIfDisposed(),hi(this,n)};Y.prototype.ifft=function(){return this.throwIfDisposed(),Oa(this)};Y.prototype.irfft=function(){return this.throwIfDisposed(),Kc(this)};Y.prototype.isFinite=function(){return this.throwIfDisposed(),hg(this)};Y.prototype.isInf=function(){return this.throwIfDisposed(),dg(this)};Y.prototype.isNaN=function(){return this.throwIfDisposed(),pg(this)};Y.prototype.leakyRelu=function(n){return this.throwIfDisposed(),fg(this,n)};Y.prototype.lessEqualStrict=function(n){return this.throwIfDisposed(),Pg(this,n)};Y.prototype.lessEqual=function(n){return this.throwIfDisposed(),Mi(this,n)};Y.prototype.lessStrict=function(n){return this.throwIfDisposed(),_g(this,n)};Y.prototype.less=function(n){return this.throwIfDisposed(),Bs(this,n)};Y.prototype.localResponseNormalization=function(n,t,e,i){return this.throwIfDisposed(),mg(this,n,t,e,i)};Y.prototype.logSigmoid=function(){return this.throwIfDisposed(),vg(this)};Y.prototype.logSoftmax=function(n){return this.throwIfDisposed(),yg(this,n)};Y.prototype.logSumExp=function(n,t){return this.throwIfDisposed(),Fc(this,n,t)};Y.prototype.log=function(){return this.throwIfDisposed(),Hi(this)};Y.prototype.log1p=function(){return this.throwIfDisposed(),Dc(this)};Y.prototype.logicalAnd=function(n){return this.throwIfDisposed(),qi(this,n)};Y.prototype.logicalNot=function(){return this.throwIfDisposed(),Ps(this)};Y.prototype.logicalOr=function(n){return this.throwIfDisposed(),Wc(this,n)};Y.prototype.logicalXor=function(n){return this.throwIfDisposed(),bg(this,n)};Y.prototype.matMul=function(n,t,e){return this.throwIfDisposed(),We(this,n,t,e)};Y.prototype.maxPool=function(n,t,e,i){return this.throwIfDisposed(),Uc(this,n,t,e,i)};Y.prototype.max=function(n,t){return this.throwIfDisposed(),Vi(this,n,t)};Y.prototype.maximumStrict=function(n){return this.throwIfDisposed(),qg(this,n)};Y.prototype.maximum=function(n){return this.throwIfDisposed(),kr(this,n)};Y.prototype.mean=function(n,t){return this.throwIfDisposed(),xa(this,n,t)};Y.prototype.min=function(n,t){return this.throwIfDisposed(),_s(this,n,t)};Y.prototype.minimumStrict=function(n){return this.throwIfDisposed(),Gg(this,n)};Y.prototype.minimum=function(n){return this.throwIfDisposed(),Ms(this,n)};Y.prototype.modStrict=function(n){return this.throwIfDisposed(),Yg(this,n)};Y.prototype.mod=function(n){return this.throwIfDisposed(),Bc(this,n)};Y.prototype.mulStrict=function(n){return this.throwIfDisposed(),Kg(this,n)};Y.prototype.mul=function(n){return this.throwIfDisposed(),J(this,n)};Y.prototype.neg=function(){return this.throwIfDisposed(),gt(this)};Y.prototype.norm=function(n,t,e){return this.throwIfDisposed(),$s(this,n,t,e)};Y.prototype.notEqualStrict=function(n){return this.throwIfDisposed(),Mg(this,n)};Y.prototype.notEqual=function(n){return this.throwIfDisposed(),Hs(this,n)};Y.prototype.oneHot=function(n,t,e){return t===void 0&&(t=1),e===void 0&&(e=0),this.throwIfDisposed(),Ts(this,n,t,e)};Y.prototype.onesLike=function(){return this.throwIfDisposed(),zc(this)};Y.prototype.pad=function(n,t){return this.throwIfDisposed(),Gi(this,n,t)};Y.prototype.pool=function(n,t,e,i,r){return this.throwIfDisposed(),wg(this,n,t,e,i,r)};Y.prototype.powStrict=function(n){return this.throwIfDisposed(),jg(this,n)};Y.prototype.pow=function(n){return this.throwIfDisposed(),jn(this,n)};Y.prototype.prelu=function(n){return this.throwIfDisposed(),Pc(this,n)};Y.prototype.prod=function(n,t){return this.throwIfDisposed(),Sg(this,n,t)};Y.prototype.reciprocal=function(){return this.throwIfDisposed(),Ig(this)};Y.prototype.relu=function(){return this.throwIfDisposed(),Ra(this)};Y.prototype.relu6=function(){return this.throwIfDisposed(),Vc(this)};Y.prototype.reshapeAs=function(n){return this.throwIfDisposed(),V(this,n.shape)};Y.prototype.reshape=function(n){return this.throwIfDisposed(),V(this,n)};Y.prototype.resizeBilinear=function(n,t){return this.throwIfDisposed(),ov(this,n,t)};Y.prototype.resizeNearestNeighbor=function(n,t){return this.throwIfDisposed(),lv(this,n,t)};Y.prototype.reverse=function(n){return this.throwIfDisposed(),$n(this,n)};Y.prototype.rfft=function(){return this.throwIfDisposed(),Gs(this)};Y.prototype.round=function(){return this.throwIfDisposed(),Ag(this)};Y.prototype.rsqrt=function(){return this.throwIfDisposed(),qc(this)};Y.prototype.selu=function(){return this.throwIfDisposed(),Tg(this)};Y.prototype.separableConv2d=function(n,t,e,i,r,a){return this.throwIfDisposed(),Ng(this,n,t,e,i,r,a)};Y.prototype.sigmoid=function(){return this.throwIfDisposed(),Pi(this)};Y.prototype.sign=function(){return this.throwIfDisposed(),xg(this)};Y.prototype.sin=function(){return this.throwIfDisposed(),Gc(this)};Y.prototype.sinh=function(){return this.throwIfDisposed(),Yc(this)};Y.prototype.slice=function(n,t){return this.throwIfDisposed(),ze(this,n,t)};Y.prototype.softmax=function(n){return this.throwIfDisposed(),Cg(this,n)};Y.prototype.softplus=function(){return this.throwIfDisposed(),kc(this)};Y.prototype.spaceToBatchND=function(n,t){return this.throwIfDisposed(),Vs(this,n,t)};Y.prototype.split=function(n,t){return this.throwIfDisposed(),Br(this,n,t)};Y.prototype.sqrt=function(){return this.throwIfDisposed(),_t(this)};Y.prototype.square=function(){return this.throwIfDisposed(),Ye(this)};Y.prototype.squaredDifference=function(n){return this.throwIfDisposed(),Ys(this,n)};Y.prototype.squaredDifferenceStrict=function(n){return this.throwIfDisposed(),$g(this,n)};Y.prototype.squeeze=function(n){return this.throwIfDisposed(),Ks(this,n)};Y.prototype.stack=function(n,t){this.throwIfDisposed();var e=n instanceof Y?[this,n]:[this].concat(n);return ji(e,t)};Y.prototype.step=function(n){return this.throwIfDisposed(),zr(this,n)};Y.prototype.stridedSlice=function(n,t,e,i,r,a,s,o){return this.throwIfDisposed(),Og(this,n,t,e,i,r,a,s,o)};Y.prototype.subStrict=function(n){return this.throwIfDisposed(),Xg(this,n)};Y.prototype.sub=function(n){return this.throwIfDisposed(),ge(this,n)};Y.prototype.sum=function(n,t){return this.throwIfDisposed(),Ae(this,n,t)};Y.prototype.tan=function(){return this.throwIfDisposed(),Eg(this)};Y.prototype.tanh=function(){return this.throwIfDisposed(),Rs(this)};Y.prototype.tile=function(n){return this.throwIfDisposed(),Dr(this,n)};Y.prototype.toBool=function(){return this.throwIfDisposed(),ue(this,"bool")};Y.prototype.toFloat=function(){return this.throwIfDisposed(),ue(this,"float32")};Y.prototype.toInt=function(){return this.throwIfDisposed(),ue(this,"int32")};Y.prototype.topk=function(n,t){return this.throwIfDisposed(),Dg(this,n,t)};Y.prototype.transpose=function(n){return this.throwIfDisposed(),ut(this,n)};Y.prototype.unique=function(n){return this.throwIfDisposed(),kg(this,n)};Y.prototype.unsortedSegmentSum=function(n,t){return this.throwIfDisposed(),jc(this,n,t)};Y.prototype.unstack=function(n){return this.throwIfDisposed(),js(this,n)};Y.prototype.where=function(n,t){return this.throwIfDisposed(),dn(n,this,t)};Y.prototype.zerosLike=function(){return this.throwIfDisposed(),Ee(this)};I.Abs=il;I.Acos=rl;I.Acosh=al;I.AdadeltaOptimizer=Zc;I.AdagradOptimizer=Qc;I.AdamOptimizer=eh;I.AdamaxOptimizer=th;I.Add=ps;I.AddN=sl;I.All=Wp;I.Any=Up;I.ArgMax=ol;I.ArgMin=ll;I.Asin=ul;I.Asinh=cl;I.Atan=hl;I.Atan2=pl;I.Atanh=dl;I.AvgPool=fl;I.AvgPool3D=ml;I.AvgPool3DBackprop=zp;I.AvgPoolBackprop=Bp;I.BatchMatMul=gl;I.BatchToSpaceND=vl;I.BroadcastTo=yl;I.Cast=fs;I.Ceil=bl;I.ClipByValue=wl;I.Complex=Pp;I.Concat=Sl;I.Conv2D=Ll;I.Conv2DBackpropFilter=_p;I.Conv2DBackpropInput=Il;I.Conv3D=Al;I.Conv3DBackpropFilterV2=Mp;I.Conv3DBackpropInputV2=Hp;I.Cos=Tl;I.Cosh=Nl;I.CropAndResize=Vp;I.Cumsum=xl;I.DataStorage=EI;I.DepthToSpace=qp;I.DepthwiseConv2dNative=Cl;I.DepthwiseConv2dNativeBackpropFilter=Gp;I.DepthwiseConv2dNativeBackpropInput=Yp;I.Diag=Kp;I.Dilation2D=Rl;I.Dilation2DBackpropFilter=$p;I.Dilation2DBackpropInput=jp;I.Div=Ol;I.Elu=El;I.EluGrad=Xp;I.Environment=Dp;I.Equal=Jp;I.Erf=Dl;I.Exp=kl;I.Expm1=Fl;I.FFT=Zp;I.Fill=Qp;I.FlipLeftRight=ef;I.Floor=Wl;I.FloorDiv=Ul;I.FromPixels=_u;I.FusedBatchNorm=Bl;I.FusedConv2D=Hu;I.FusedDepthwiseConv2D=Vu;I.GatherNd=tf;I.GatherV2=zl;I.Greater=nf;I.GreaterEqual=Pl;I.IFFT=rf;I.Identity=_l;I.Imag=af;I.IsFinite=Ml;I.IsInf=Hl;I.IsNan=Vl;I.KernelBackend=Op;I.LRN=Kl;I.LRNBackprop=df;I.Less=sf;I.LessEqual=of;I.LinSpace=lf;I.Log=ql;I.Log1p=Gl;I.LogSoftmax=Yl;I.LogicalAnd=uf;I.LogicalNot=cf;I.LogicalOr=hf;I.Max=jl;I.MaxPool=Xl;I.MaxPool3D=Jl;I.MaxPool3DBackprop=ff;I.MaxPoolBackprop=pf;I.MaxPoolWithArgmax=mf;I.Maximum=$l;I.Mean=BI;I.Min=Zl;I.Minimum=Ql;I.Mod=eu;I.MomentumOptimizer=nh;I.Multiply=tu;I.Negate=nu;I.NonMaxSuppressionV3=vf;I.NonMaxSuppressionV4=yf;I.NonMaxSuppressionV5=bf;I.NotEqual=gf;I.OP_SCOPE_SUFFIX=tm;I.OneHot=ru;I.OnesLike=iu;I.Optimizer=di;I.PadV2=au;I.Pool=zI;I.Pow=su;I.Prelu=ou;I.Prod=wf;I.RMSPropOptimizer=ih;I.Range=Sf;I.Real=Lf;I.Reciprocal=lu;I.Relu=uu;I.Relu6=pu;I.Reshape=cu;I.ResizeBilinear=du;I.ResizeBilinearGrad=Af;I.ResizeNearestNeighbor=hu;I.ResizeNearestNeighborGrad=If;I.Reverse=fu;I.RotateWithOffset=Ef;I.Round=mu;I.Rsqrt=gu;I.SGDOptimizer=eo;I.ScatterNd=Tf;I.SelectV2=vu;I.Selu=yu;I.Sigmoid=Iu;I.Sign=Lu;I.Sin=wu;I.Sinh=Su;I.Slice=bu;I.Softmax=Ru;I.Softplus=Au;I.SpaceToBatchND=xu;I.SparseToDense=xf;I.SplitV=Cu;I.Sqrt=Tu;I.Square=Nf;I.SquaredDifference=Ou;I.Step=Pu;I.StridedSlice=Cf;I.Sub=Eu;I.Sum=Nu;I.Tan=Du;I.Tanh=ku;I.Tensor=Y;I.TensorBuffer=Ss;I.Tile=Fu;I.TopK=Rf;I.Transpose=Wu;I.Unique=Of;I.Unpack=Uu;I.UnsortedSegmentSum=Bu;I.Variable=ya;I.ZerosLike=zu;I._FusedMatMul=Mu;I.abs=qt;I.acos=Bm;I.acosh=zm;I.add=fe;I.addN=gT;I.addStrict=Hg;I.all=Mm;I.any=Hm;I.argMax=Vm;I.argMin=qm;I.asin=Gm;I.asinh=Ym;I.atan=Km;I.atan2=jm;I.atanh=$m;I.avgPool=Tc;I.avgPool3d=FT;I.backend=sT;I.backend_util=MO;I.basicLSTMCell=_T;I.batchNorm=Aa;I.batchNorm2d=GT;I.batchNorm3d=KT;I.batchNorm4d=$T;I.batchToSpaceND=Os;I.booleanMaskAsync=yC;I.broadcastTo=Ds;I.browser=OA;I.buffer=On;I.cast=ue;I.ceil=Zm;I.clipByValue=Qm;I.clone=Bi;I.complex=ri;I.concat=xt;I.concat1d=e1;I.concat2d=n1;I.concat3d=r1;I.concat4d=s1;I.conv1d=eg;I.conv2d=Er;I.conv2dTranspose=tg;I.conv3d=d1;I.conv3dTranspose=m1;I.copyRegisteredKernels=MI;I.cos=ks;I.cosh=xc;I.cosineWindow=$c;I.cumsum=Cc;I.customGrad=kn;I.deprecationWarn=It;I.depthToSpace=ig;I.depthwiseConv2d=Ta;I.device_util=I2;I.diag=L1;I.dilation2d=rg;I.disableDeprecationWarnings=KA;I.dispose=zt;I.disposeVariables=jA;I.div=Ie;I.divNoNan=ag;I.divStrict=Vg;I.dot=sg;I.dropout=YC;I.elu=Rc;I.enableDebugMode=YA;I.enableProdMode=GA;I.enclosingPowerOfTwo=Zg;I.engine=$A;I.env=He;I.equal=Na;I.equalStrict=Ug;I.erf=og;I.exp=pn;I.expandDims=fn;I.expm1=lg;I.eye=ug;I.fft=qs;I.fill=Oc;I.findBackend=iT;I.findBackendFactory=rT;I.floor=Fs;I.floorDiv=Sc;I.fused=rR;I.gather=Ws;I.gatherND=VC;I.gather_util=EA;I.getBackend=tT;I.getGradient=Yu;I.getKernel=Gu;I.getKernelsForBackend=ms;I.grad=iN;I.grads=rN;I.greater=hi;I.greaterEqual=_i;I.greaterEqualStrict=Bg;I.greaterStrict=zg;I.ifft=Oa;I.imag=Us;I.image=cO;I.inTopKAsync=jC;I.io=bA;I.irfft=Kc;I.isFinite=hg;I.isInf=dg;I.isNaN=pg;I.keep=Um;I.kernel_impls=GO;I.leakyRelu=fg;I.less=Bs;I.lessEqual=Mi;I.lessEqualStrict=Pg;I.lessStrict=_g;I.linalg=hO;I.linspace=Q1;I.localResponseNormalization=mg;I.log=Hi;I.log1p=Dc;I.logSigmoid=vg;I.logSoftmax=yg;I.logSumExp=Fc;I.logicalAnd=qi;I.logicalNot=Ps;I.logicalOr=Wc;I.logicalXor=bg;I.losses=dO;I.matMul=We;I.math=NA;I.max=Vi;I.maxPool=Uc;I.maxPool3d=SN;I.maxPoolWithArgmax=IN;I.maximum=kr;I.maximumStrict=qg;I.mean=xa;I.memory=XA;I.min=_s;I.minimum=Ms;I.minimumStrict=Gg;I.mod=Bc;I.modStrict=Yg;I.moments=ON;I.movingAverage=UC;I.mul=J;I.mulStrict=Kg;I.multiRNNCell=DN;I.multinomial=FN;I.neg=gt;I.nextFrame=mO;I.norm=$s;I.notEqual=Hs;I.notEqualStrict=Mg;I.oneHot=Ts;I.ones=Fr;I.onesLike=zc;I.op=U;I.outerProduct=PN;I.pad=Gi;I.pad1d=HN;I.pad2d=qN;I.pad3d=YN;I.pad4d=jN;I.pool=wg;I.pow=jn;I.powStrict=jg;I.prelu=Pc;I.print=hm;I.prod=Sg;I.profile=JA;I.rand=ix;I.randomGamma=fx;I.randomNormal=gx;I.randomUniform=Lg;I.range=Hc;I.ready=eT;I.real=Ca;I.reciprocal=Ig;I.registerBackend=aT;I.registerGradient=kf;I.registerKernel=Df;I.relu=Ra;I.relu6=Vc;I.removeBackend=nT;I.reshape=V;I.reverse=$n;I.reverse1d=Ix;I.reverse2d=Tx;I.reverse3d=xx;I.reverse4d=Rx;I.rfft=Gs;I.round=Ag;I.rsqrt=qc;I.scalar=we;I.scatterND=zC;I.scatter_util=DA;I.selu=Tg;I.separableConv2d=Ng;I.serialization=WA;I.setBackend=QA;I.setPlatform=oT;I.setdiff1dAsync=Wx;I.sigmoid=Pi;I.sign=xg;I.signal=uO;I.sin=Gc;I.sinh=Yc;I.slice=ze;I.slice1d=_x;I.slice2d=Hx;I.slice3d=qx;I.slice4d=Yx;I.slice_util=Dm;I.softmax=Cg;I.softplus=kc;I.spaceToBatchND=Vs;I.sparseToDense=MC;I.spectral=lO;I.split=Br;I.sqrt=_t;I.square=Ye;I.squaredDifference=Ys;I.squaredDifferenceStrict=$g;I.squeeze=Ks;I.stack=ji;I.step=zr;I.stridedSlice=Og;I.sub=ge;I.subStrict=Xg;I.sum=Ae;I.sumOutType=f2;I.tan=Eg;I.tanh=Rs;I.tensor=si;I.tensor1d=Ur;I.tensor2d=Ea;I.tensor3d=ym;I.tensor4d=sC;I.tensor5d=oC;I.tensor6d=lC;I.tensor_util=v2;I.test_util=VA;I.tidy=ft;I.tile=Dr;I.time=ZA;I.topk=Dg;I.train=pO;I.transpose=ut;I.truncatedNormal=hC;I.unique=kg;I.unregisterGradient=_I;I.unregisterKernel=PI;I.unsortedSegmentSum=jc;I.unstack=js;I.upcastType=Ls;I.util=i2;I.valueAndGrad=aN;I.valueAndGrads=sN;I.variable=mC;I.variableGrads=gg;I.version_core=qA;I.where=dn;I.whereAsync=Wg;I.zeros=Kn;I.zerosLike=Ee});var Yb=be(Xe=>{"use strict";Object.defineProperty(Xe,"__esModule",{value:!0});var y=Xi();var ah=function(n,t){return ah=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},ah(n,t)};function Q(n,t){ah(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var Gt=function(){return Gt=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t},Gt.apply(this,arguments)};function VD(n,t){var e={};for(var i in n)Object.prototype.hasOwnProperty.call(n,i)&&t.indexOf(i)<0&&(e[i]=n[i]);if(n!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,i=Object.getOwnPropertySymbols(n);r<i.length;r++)t.indexOf(i[r])<0&&(e[i[r]]=n[i[r]]);return e}function Le(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function ve(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var sh;function vt(){return sh==null&&(sh=y.backend().epsilon()),sh}function mn(){return"channelsLast"}var pi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Fn=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),M=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),Te=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),qD=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error),SH=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return Object.setPrototypeOf(i,t.prototype),i}return t}(Error);function Ji(n,t){if(Array.isArray(n)){for(var e=[],i=0;i<t;i++)e=e.concat(n);return e}else{var e=new Array(t);return e.fill(n),e}}function Wn(n,t){if(!n)throw new qD(t)}function wv(n,t){for(var e=0,i=0,r=n;i<r.length;i++){var a=r[i];a===t&&e++}return e}function Mt(n){return n.length===1?n[0]:n}function Je(n){return Array.isArray(n)?n:[n]}function Jn(n){var t=n.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2"),e=t.replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function Zi(n){return n.length<=1||n.indexOf("_")===-1?n:n.replace(/[_]+(\w|$)/g,function(t,e){return e.toUpperCase()})}var ln={};function oh(n){if(n==null)return null;var t={};return t.className=n.getClassName(),t.config=n.getConfig(),t}function lh(n){if(n==null||typeof n!="object")return;if(Array.isArray(n))n.forEach(function(s){return lh(s)});else for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e],a=n[r];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?n[r]=a.value:lh(a))}}function Da(n,t,e,i,r){var a,s,o;if(t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i="object"),r===void 0&&(r=!1),typeof n=="string"){var l=n,u=void 0;if(l in e)u=e[l];else if(l in ln)u=ln[l];else if(u=t[l],u==null)throw new M("Unknown "+i+": "+n+`. This may be due to one of the following reasons:
 `+("1. The "+i+" is defined in Python, in which ")+`case it needs to be ported to TensorFlow.js or your JavaScript code.
 `+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");return u}else{var c=n;if(c.className==null||c.config==null)throw new M(i+": Improper config format: "+(JSON.stringify(c)+`.
-`)+"'className' and 'config' must set.");var h=c.className,d=void 0,p=void 0;if(h in e?(a=e[h],d=a[0],p=a[1]):h in un?(s=un.className,d=s[0],p=s[1]):h in t&&(o=t[h],d=o[0],p=o[1]),d==null)throw new M("Unknown "+i+": "+h+`. This may be due to one of the following reasons:
+`)+"'className' and 'config' must set.");var h=c.className,d=void 0,p=void 0;if(h in e?(a=e[h],d=a[0],p=a[1]):h in ln?(s=ln.className,d=s[0],p=s[1]):h in t&&(o=t[h],d=o[0],p=o[1]),d==null)throw new M("Unknown "+i+": "+h+`. This may be due to one of the following reasons:
 `+("1. The "+i+" is defined in Python, in which ")+`case it needs to be ported to TensorFlow.js or your JavaScript code.
-`+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");if(p!=null){for(var f={},m=0,g=Object.keys(un);m<g.length;m++){var v=g[m];f[v]=un[v]}for(var b=0,w=Object.keys(e);b<w.length;b++){var v=w[b];f[v]=e[v]}var S=c.config;S.customObjects=f;for(var L=Yt({},un),x=0,C=Object.keys(e);x<C.length;x++){var v=C[x];un[v]=e[v]}oh(c.config);var R=p(d,c.config,e,r);return un=Yt({},L),R}else{for(var L=Yt({},un),D=0,k=Object.keys(e);D<k.length;D++){var v=k[D];un[v]=e[v]}var R=new d(c.config);return un=Yt({},L),R}}}function GD(n,t){return n<t?-1:n>t?1:0}function to(n,t){return-1*GD(n,t)}function fi(n){if(n==null)return n;for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.indexOf(r)===-1&&t.push(r)}return t}function YD(n){if(n==null)throw new M("Invalid value in obj: "+JSON.stringify(n));for(var t in n)if(n.hasOwnProperty(t))return!1;return!0}function _r(n,t,e){if(e==null)return;if(n.indexOf(e)<0)throw new M(e+" is not a valid "+t+".  Valid values are "+n+" or null/undefined.")}function lh(n,t,e,i){return e===void 0&&(e=0),i===void 0&&(i=Infinity),Wn(e>=0),Wn(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function At(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return At(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(Sv(n)+".")})}function Sv(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return Sv(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function KD(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];var o=y.util.now();return o-e<t||(e=o,i=n.apply(void 0,a)),i};return r}function Lv(n){return n==="relu"?"relu":n==="linear"?"linear":n==="elu"?"elu":null}function uh(n,t){return y.tidy(function(){return y.sqrt(y.sum(y.mul(n,n),t,!0))})}var ka=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),Iv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMaxValue=2,i.defaultAxis=0,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=uh(e,i.axis),a=y.clipByValue(r,0,i.maxValue);return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{maxValue:this.maxValue,axis:this.axis}},t.className="MaxNorm",t}(ka);y.serialization.registerClass(Iv);var Av=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultAxis=0,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){return y.div(e,y.add(vt(),uh(e,i.axis)))})},t.prototype.getConfig=function(){return{axis:this.axis}},t.className="UnitNorm",t}(ka);y.serialization.registerClass(Av);var Tv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="NonNeg",t}(ka);y.serialization.registerClass(Tv);var Nv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMinValue=0,i.defaultMaxValue=1,i.defaultRate=1,i.defaultAxis=0,i.minValue=e.minValue!=null?e.minValue:i.defaultMinValue,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.rate=e.rate!=null?e.rate:i.defaultRate,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=uh(e,i.axis),a=y.add(y.mul(i.rate,y.clipByValue(r,i.minValue,i.maxValue)),y.mul(1-i.rate,r));return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}},t.className="MinMaxNorm",t}(ka);y.serialization.registerClass(Nv);var xv={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function yt(n){return sh(n)}function Cv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"constraint")}function bt(n){if(n==null)return null;if(typeof n=="string"){var t=n in xv?xv[n]:n,e={className:t,config:{}};return Cv(e)}else return n instanceof ka?n:Cv(n)}function jD(n){return new Iv(n)}function $D(n){return new Av(n)}function XD(){return new Tv}function JD(n){return new Nv(n)}var ZD={__proto__:null,maxNorm:jD,unitNorm:$D,nonNeg:XD,minMaxNorm:JD};var QD=["channelsFirst","channelsLast"],ek=["valid","same","causal"],tk=["max","avg"],nk=["sum","mul","concat","ave"];var Mr=new Map;function ct(n){_r(QD,"DataFormat",n)}function nn(n){_r(ek,"PaddingMode",n)}function Rv(n){_r(tk,"PoolMode",n)}var Fa=[],Ov="/";function Qi(n,t){Fa.push(n);try{var e=t();return Fa.pop(),e}catch(i){throw Fa.pop(),i}}function ik(){return Fa.length===0?"":Fa.join(Ov)+Ov}function Dv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");return ik()+n}function kv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");Mr.has(n)||Mr.set(n,0);var t=Mr.get(n);if(Mr.set(n,Mr.get(n)+1),t>0){var e=n+"_"+t;return Mr.set(e,1),e}else return n}var rk=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function Ev(n){return!!n.match(rk)}function ak(n){return n===parseInt(n.toString(),10)}function mi(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r<e;++r)i*=n[r];return i}function Fv(n){return n=Array.isArray(n)?new Float32Array(n):n,y.tensor1d(n)}function Hr(n){return y.min(Fv(n)).dataSync()[0]}function gi(n){return y.max(Fv(n)).dataSync()[0]}function vn(n,t){if(t<n)throw new M("end ("+t+") < begin ("+n+") is forbidden.");for(var e=[],i=n;i<t;++i)e.push(i);return e}function Wa(n,t){return n.asType(t)}function Ua(n,t){t===void 0&&(t=-1);var e=n.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),n.reshape(e)}function sk(n,t){return y.tidy(function(){if(n.shape.length!==2)throw new M("repeat() expects a rank-2 tensor, but received a "+("rank-"+n.shape.length+" tensor."));var e=Ua(n,1);return ch(e,[1,t,1])})}function ok(n){var t=[mi(n.shape)];return n.reshape(t)}function lk(n){if(n.rank<=1)throw new M("batchFlatten requires a minimum rank of 2. Got rank: "+n.rank+".");var t=[n.shape[0],mi(n.shape,1)];return n.reshape(t)}function er(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[t,0],[e,n.shape[1]]);case 3:return y.slice3d(n,[t,0,0],[e,n.shape[1],n.shape[2]]);case 4:return y.slice4d(n,[t,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3]]);case 5:return y.slice(n,[t,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4]]);case 6:return y.slice(n,[t,0,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4],n.shape[5]]);default:throw new M("sliceAlongFirstAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function hh(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[0,t],[n.shape[0],e]);case 3:return y.slice3d(n,[0,0,t],[n.shape[0],n.shape[1],e]);case 4:return y.slice4d(n,[0,0,0,t],[n.shape[0],n.shape[1],n.shape[2],e]);default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function no(n,t,e,i){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:switch(i){case 1:return er(n,t,e);case 2:return hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 3:switch(i){case 1:return er(n,t,e);case 2:return y.slice3d(n,[0,t,0],[n.shape[0],e,n.shape[2]]);case 3:return hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 4:switch(i){case 1:return er(n,t,e);case 2:return y.slice4d(n,[0,t,0,0],[n.shape[0],e,n.shape[2],n.shape[3]]);case 3:return y.slice4d(n,[0,0,t,0],[n.shape[0],n.shape[1],e,n.shape[3]]);case 4:return hh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function dh(n,t){t===void 0&&(t=-1);var e;return t<0&&(e=n[0].rank,e!==0?t=e:t=0),t===n[0].rank&&(t=-1),y.concat(n,t)}function Wv(n,t){switch(n.rank){case 1:return y.concat1d([n,t]);case 2:return y.concat2d([n,t],0);case 3:return y.concat3d([n,t],0);case 4:return y.concat4d([n,t],0);default:throw new M("concatAlongFirstAxis() received an unsupported "+("tensor rank: "+n.rank))}}function ch(n,t){if(Array.isArray(t)||(t=[t]),n.rank!==t.length)throw new M("The length of input n ("+t.length+") does not match "+("the number of dimensions in input x ("+n.rank+")"));return y.tile(n,t)}function io(n,t,e,i,r){return t===void 0&&(t=0),e===void 0&&(e=1),y.randomNormal(n,t,e,i,r)}function Un(n,t,e,i){if(n.rank<2||t.rank<2)throw new Te("dot requires both inputs to be rank >= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?ph(n.rank,i,gn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?ph(n.rank,i,gn()):null,activation:e}).reshape(f)}}function Uv(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function Ba(n){return y.mul(n,n)}function ph(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Bn(n,t,e){return y.tidy(function(){return e==null&&(e=gn()),ct(e),n.add(ph(n.rank,t,e))})}function uk(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function ck(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function Bv(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function hk(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function za(n,t,e){return e===void 0&&(e=!1),e?n():t()}var dk=["fanIn","fanOut","fanAvg"],pk=["normal","uniform","truncatedNormal"];function fk(n){_r(dk,"FanMode",n)}function mk(n){_r(pk,"Distribution",n)}var cn=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),zv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(cn);y.serialization.registerClass(zv);var fh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(cn);y.serialization.registerClass(fh);var Pv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(cn);y.serialization.registerClass(Pv);var _v=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(cn);y.serialization.registerClass(_v);var Mv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return io(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(cn);y.serialization.registerClass(Mv);var Hv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(cn);y.serialization.registerClass(Hv);var Vv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(cn);y.serialization.registerClass(Vv);function gk(n,t){t===void 0&&(t="channelsLast");var e,i;if(ct(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=mi(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=mi(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=mi(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var Kt=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,fk(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,mk(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=gk(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(cn);y.serialization.registerClass(Kt);var mh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="GlorotUniform",t}(Kt);y.serialization.registerClass(mh);var gh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="GlorotNormal",t}(Kt);y.serialization.registerClass(gh);var vh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="HeNormal",t}(Kt);y.serialization.registerClass(vh);var yh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="HeUniform",t}(Kt);y.serialization.registerClass(yh);var bh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="LeCunNormal",t}(Kt);y.serialization.registerClass(bh);var wh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="LeCunNormal",t}(Kt);y.serialization.registerClass(wh);var qv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=io(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(cn);y.serialization.registerClass(qv);var Gv={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function Yv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function st(n){return sh(n)}function tt(n){if(typeof n=="string"){var t=n in Gv?Gv[n]:n;if(t==="GlorotNormal")return new gh;if(t==="GlorotUniform")return new mh;if(t==="HeNormal")return new vh;if(t==="HeUniform")return new yh;if(t==="LeCunNormal")return new bh;if(t==="LeCunUniform")return new wh;var e={};return e.className=t,e.config={},Yv(e)}else return n instanceof cn?n:Yv(n)}function vk(){return new zv}function yk(){return new fh}function bk(n){return new Pv(n)}function wk(n){return new _v(n)}function Sk(n){return new Mv(n)}function Lk(n){return new Hv(n)}function Ik(n){return new Vv(n)}function Ak(n){return new Kt(n)}function Tk(n){return new mh(n)}function Nk(n){return new gh(n)}function xk(n){return new vh(n)}function Ck(n){return new yh(n)}function Rk(n){return new bh(n)}function Ok(n){return new wh(n)}function Ek(n){return new qv(n)}var Dk={__proto__:null,zeros:vk,ones:yk,constant:bk,randomUniform:wk,randomNormal:Sk,truncatedNormal:Lk,identity:Ik,varianceScaling:Ak,glorotUniform:Tk,glorotNormal:Nk,heNormal:xk,heUniform:Ck,leCunNormal:Rk,leCunUniform:Ok,orthogonal:Ek};var kk=0;function Kv(){return kk++}var ro={};function ao(n){return n===void 0&&(n=""),n in ro||(ro[n]=0),ro[n]+=1,n+ro[n].toString()}function Sh(n){return Array.isArray(n)&&Array.isArray(n[0])}function so(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function Ke(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function oo(n){for(var t=0,e=0,i=n;e<i.length;e++){var r=i[e];r.shape.length===0?t+=1:t+=r.shape.reduce(function(a,s){return a*s})}return t}var jv="Variable",$v=function(){function n(t,e,i,r,a){e===void 0&&(e="float32"),i===void 0&&(i=jv),r===void 0&&(r=!0),a===void 0&&(a=null),this.dtype=e??"float32",this.shape=t.shape,this.id=Kv(),i=i??jv,this.originalName=Dv(i),this.name=kv(this.originalName),this.trainable_=r,this.constraint=a,this.val=y.variable(t,this.trainable_,this.name,this.dtype)}return n.prototype.read=function(){return this.assertNotDisposed(),this.val},n.prototype.write=function(t){return this.assertNotDisposed(),Fk(this.val,t),this.val.id!==t.id&&(this.val.assign(t),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this},n.prototype.dispose=function(){this.assertNotDisposed(),this.val.dispose()},n.prototype.assertNotDisposed=function(){if(this.val.isDisposed)throw new Error("LayersVariable "+this.name+" is already disposed.")},Object.defineProperty(n.prototype,"trainable",{get:function(){return this.trainable_},set:function(t){this.trainable_=t,this.val.trainable=t},enumerable:!0,configurable:!0}),n}();function Fk(n,t){if(n.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(n.shape)+" vs. "+JSON.stringify(t.shape))}function Lh(n){return n.map(function(t){return t.read()})}function Ih(n){n.forEach(function(t){var e=t[0];e.write(t[1])})}var Tt=function(){function n(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}return n}(),yn=function(){function n(t,e,i,r,a,s,o){this.dtype=t,this.shape=e,this.sourceLayer=i,this.inputs=r,this.callArgs=a,this.outputTensorIndex=o,this.id=Kv(),s!=null&&(this.originalName=Dv(s),this.name=kv(this.originalName)),this.rank=e.length}return n}(),Wk=0,lo=function(){function n(t,e){this.callArgs=e,this.id=Wk++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(var i=0,r=t.inboundLayers;i<r.length;i++){var a=r[i];a!=null&&a.outboundNodes.push(this)}t.outboundLayer.inboundNodes.push(this)}return n.prototype.getConfig=function(){for(var t=[],e=0,i=this.inboundLayers;e<i.length;e++){var r=i[e];r!=null?t.push(r.name):t.push(null)}return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}},n}(),Uk=0,De=function(n){Q(t,n);function t(e){e===void 0&&(e={});var i=n.call(this)||this;i._callHook=null,i._addedWeightNames=[],i._stateful=!1,i.id=Uk++,i.activityRegularizer=null,i.inputSpec=null,i.supportsMasking=!1,i._trainableWeights=[],i._nonTrainableWeights=[],i._losses=[],i._updates=[],i._built=!1,i.inboundNodes=[],i.outboundNodes=[];var r=e.name;if(!r){var a=i.getClassName();r=Jn(a)+"_"+ao(a)}if(i.name=r,i.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){var s=void 0;if(e.batchInputShape!=null)s=e.batchInputShape;else if(e.inputShape!=null){var o=null;e.batchSize!=null&&(o=e.batchSize),s=[o].concat(e.inputShape)}i.batchInputShape=s;var l=e.dtype;l==null&&(l=e.inputDType),l==null&&(l="float32"),i.dtype=l}return e.weights!=null?i.initialWeights=e.weights:i.initialWeights=null,i._refCount=null,i.fastWeightInitDuringBuild=!1,i}return t.nodeKey=function(e,i){return e.name+"_ib-"+i.toString()},t.prototype.getNodeAtIndex=function(e,i){if(this.inboundNodes.length===0)throw new Fn("The layer has never been called "+("and thus has no defined "+i+"."));if(this.inboundNodes.length<=e)throw new M("Asked to get "+i+" at node "+e+", "+("but the layer has only "+this.inboundNodes.length+" inbound nodes."));return this.inboundNodes[e]},t.prototype.getInputAt=function(e){return Mt(this.getNodeAtIndex(e,"input").inputTensors)},t.prototype.getOutputAt=function(e){return Mt(this.getNodeAtIndex(e,"output").outputTensors)},Object.defineProperty(t.prototype,"input",{get:function(){if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" is not connected, no input to return.");return Mt(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return Mt(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Je(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Je(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r<e.length;r++){var a=e[r],s=i[r];if(s==null)continue;var o=a.rank;if(s.ndim!=null&&o!==s.ndim)throw new M("Input "+r+" is incompatible with layer "+this.name+": "+("expected ndim="+s.ndim+", found ndim="+o));if(s.maxNDim!=null&&o>s.maxNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected max_ndim="+s.maxNDim+", found ndim="+o));if(s.minNDim!=null&&o<s.minNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected min_ndim="+s.minNDim+", found ndim="+o+"."));if(s.dtype!=null&&a.dtype!==s.dtype)throw new M("Input "+r+" is incompatible with layer "+this.name+" "+(": expected dtype="+s.dtype+", found dtype="+a.dtype+"."));if(s.axes){var l=a.shape;for(var u in s.axes){var c=Number(u),h=s.axes[u],d=c>=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p<s.shape.length;++p){var f=s.shape[p],m=a.shape[p];if(f!=null&&m!=null&&f!==m)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected shape="+s.shape+", ")+("found shape="+a.shape+"."))}}},t.prototype.call=function(e,i){return e},t.prototype.invokeCallHook=function(e,i){this._callHook!=null&&this._callHook(e,i)},t.prototype.setCallHook=function(e){this._callHook=e},t.prototype.clearCallHook=function(){this._callHook=null},t.prototype.apply=function(e,i){var r=this;i=i||{},this.assertNotDisposed();for(var a=Je(e),s=!0,o=0,l=a;o<l.length;o++){var u=l[o];if(!(u instanceof yn)){s=!1;break}}for(var c=!0,h=0,d=a;h<d.length;h++){var u=d[h];if(u instanceof yn){c=!1;break}}if(s===c)throw new M("Arguments to apply() must be all SymbolicTensors or all Tensors");return Qi(this.name,function(){if(!r.built){r.assertInputCompatibility(e);for(var p=[],f=0,m=Je(e);f<m.length;f++){var g=m[f];p.push(g.shape)}r.build(Mt(p)),r.built=!0,r.initialWeights&&r.setWeights(r.initialWeights),r._refCount===null&&c&&(r._refCount=1)}if(r.assertInputCompatibility(e),c){for(var v=r.call(e,i),b=Je(v),w=[],S=0,L=b;S<L.length;S++){var x=L[S];a.indexOf(x)!==-1&&(x=x.clone()),w.push(x)}if(v=Mt(w),r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}else{var C=Bk(e),R=r.computeOutputShape(C),v=void 0,D=zk(e);if(r.warnOnIncompatibleInputShape(Array.isArray(e)?C[0]:C),R!=null&&R.length>0&&Array.isArray(R[0])?v=R.map(function(W,F){return new yn(D,W,r,Je(e),i,r.name,F)}):v=new yn(D,R,r,Je(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new pi("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i<r.length;i++){var a=r[i],s=JSON.stringify(a.outputShapes);e.indexOf(s)===-1&&e.push(s)}if(e.length===1){var o=this.inboundNodes[0].outputShapes;return Array.isArray(o)&&Array.isArray(o[0])&&o.length===1?o[0]:o}else throw new pi("The layer "+this.name+' has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.')},enumerable:!0,configurable:!0}),t.prototype.countParams=function(){if(!this.built)throw new Fn("You tried to call countParams() on "+this.name+", but the layer is not built yet. Build it first by calling build(batchInputShape).");return oo(this.weights)},t.prototype.build=function(e){this.built=!0},t.prototype.getWeights=function(e){return e===void 0&&(e=!1),Lh(e?this.trainableWeights:this.weights)},t.prototype.setWeights=function(e){var i=this;y.tidy(function(){var r=i.weights;if(r.length!==e.length)throw new M('You called setWeights(weights) on layer "'+i.name+'" '+("with a weight list of length "+e.length+", ")+("but the layer was expecting "+r.length+" weights. ")+("Provided weights: "+e+"..."));if(r.length===0)return;for(var a=[],s=Lh(r),o=0;o<s.length;++o){var l=s[o],u=r[o],c=e[o];if(!y.util.arraysEqual(l.shape,c.shape))throw new M("Layer weight shape "+l.shape+" "+("not compatible with provided weight shape "+c.shape));a.push([u,c])}Ih(a)})},t.prototype.addWeight=function(e,i,r,a,s,o,l){if(this._addedWeightNames.indexOf(e)!==-1)throw new M("Duplicate weight name "+e+" for layer "+this.name);this._addedWeightNames.push(e),r==null&&(r="float32"),this.fastWeightInitDuringBuild&&(a=tt("zeros"));var u=a.apply(i,r),c=new $v(u,r,e,o,l);return u.dispose(),s!=null&&this.addLoss(function(){return s.apply(c.read())}),o==null&&(o=!0),o?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c},t.prototype.setFastWeightInitDuringBuild=function(e){this.fastWeightInitDuringBuild=e},t.prototype.addLoss=function(e){var i;if(e==null||Array.isArray(e)&&e.length===0)return;e=Je(e),this._losses!==void 0&&this._losses!==null&&(i=this.losses).push.apply(i,e)},t.prototype.computeOutputShape=function(e){return e},t.prototype.computeMask=function(e,i){var r=this;if(!this.supportsMasking){if(i!=null)if(Array.isArray(i))i.forEach(function(a){if(a!=null)throw new TypeError("Layer "+r.name+" does not support masking, but was passed an inputMask.")});else throw new TypeError("Layer "+this.name+" does not support masking, but was passed an inputMask.");return null}return i},t.prototype.addInboundNode=function(e,i,r,a,s,o,l){l===void 0&&(l=null);var u=Je(e);i=Je(i),r=Je(r),a=Je(a),s=so(s),o=so(o);for(var c=[],h=[],d=[],p=0,f=u;p<f.length;p++){var m=f[p];c.push(m.sourceLayer),h.push(m.nodeIndex),d.push(m.tensorIndex)}new lo({outboundLayer:this,inboundLayers:c,nodeIndices:h,tensorIndices:d,inputTensors:u,outputTensors:i,inputMasks:r,outputMasks:a,inputShapes:s,outputShapes:o},l);for(var g=0;g<i.length;g++)i[g].sourceLayer=this,i[g].nodeIndex=this.inboundNodes.length-1,i[g].tensorIndex=g},t.prototype.getConfig=function(){var e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e},t.prototype.disposeWeights=function(){return this.weights.forEach(function(e){return e.dispose()}),this.weights.length},t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Layer '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){if(!this.built)throw new Error("Cannot dispose Layer "+this.name+" because it has not been built yet.");if(this._refCount===null)throw new Error("Cannot dispose Layer "+this.name+" because it has not been used yet.");this.assertNotDisposed();var e=0;return--this._refCount===0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}},t}(y.serialization.Serializable);function Bk(n){n=Je(n);for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.push(r.shape)}return Mt(t)}function zk(n){return"float32"}function Xv(n,t,e){if((t==null||e!=null&&e>0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a<i.inboundLayers.length;a++)for(var s=i.inputTensors[a],o=i.inboundLayers[a],l=i.nodeIndices[a],u=Xv(s,o,l),c=0,h=u;c<h.length;c++){var d=h[c];r.indexOf(d)===-1&&r.push(d)}return r}var Pa=function(n){Q(t,n);function t(e){var i=n.call(this,{dtype:e.dtype,name:e.name!=null?e.name:ao("input").toString()})||this;if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),i.trainable=!1,i.built=!0,i.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new M("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");var r=e.batchInputShape;if(r==null){if(e.inputShape==null)throw new M("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");r=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new M("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");var a=e.dtype||"float32";i.batchInputShape=r,i.dtype=a,i.inputSpec=[{shape:r}];var s=new yn(i.dtype,i.batchInputShape,i,[],{},i.name);return s.nodeIndex=0,s.tensorIndex=0,new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[r],outputShapes:[r]}),i}return t.prototype.apply=function(e,i){throw new M("Cannot pass any input to an "+("InputLayer's apply() method. InputLayer name: "+this.name))},t.prototype.dispose=function(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}},t.prototype.getConfig=function(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}},t.className="InputLayer",t}(De);y.serialization.registerClass(Pa);function Jv(n){if(n.batchShape==null&&n.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(n.batchShape!=null&&n.shape!=null)throw new M("Please provide either a `shape` or `batchShape` argument to Input, but not both.");var t=n.batchShape;n.shape!=null&&t==null&&(t=[null].concat(n.shape));var e=n.dtype;e==null&&(e="float32");var i=new Pa({batchInputShape:t,name:n.name,dtype:e,sparse:n.sparse}),r=i.inboundNodes[0].outputTensors;return r[0]}function vi(n){return Le(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(n==null)return[2];t=[],e=[],i=[];for(r in n)a=n[r],typeof a!="number"&&(s=a,t.push(s.data()),e.push(r),i.push(s));return t.length>0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l<o.length;++l)n[e[l]]=o[l][0];y.dispose(i),u.label=2;case 2:return[2]}})})}function Zv(n){if(n==null)return;for(var t in n){var e=n[t];typeof e!="number"&&e.dispose()}}var Qv;(function(n){n[n.SILENT=0]="SILENT",n[n.VERBOSE=1]="VERBOSE"})(Qv||(Qv={}));var Pk=125,Vr=function(){function n(){this.validationData=null}return n.prototype.setParams=function(t){this.params=t},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.setModel=function(t){},n}(),ey=function(){function n(t,e){e===void 0&&(e=10),t==null&&(t=[]),this.callbacks=t,this.queueLength=e}return n.prototype.append=function(t){this.callbacks.push(t)},n.prototype.setParams=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setParams(t)}},n.prototype.setModel=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setModel(t)}},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainBegin(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainEnd(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n}(),_k=function(n){Q(t,n);function t(){return n.call(this)||this}return t.prototype.onEpochBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.seen=0,this.totals={},[2]})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l=this;return ve(this,function(u){i==null&&(i={}),r=i.size==null?0:i.size,this.seen+=r,a=function(c){var h=i[c];if(typeof h=="number")s.totals.hasOwnProperty(c)||(s.totals[c]=0),s.totals[c]=s.totals[c]+h*r;else{var d=void 0;c in s.totals?d=s.totals[c]:s.totals[c]=0;var p=y.tidy(function(){return y.add(l.totals[c],y.mul(h,r))});s.totals[c]=p,d!=null&&d.dispose()}},s=this;for(o in i)a(o);return[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u=this;return ve(this,function(c){if(i!=null)for(r=function(h){if(a.totals[h]==null)return"continue";typeof a.totals[h]=="number"?i[h]=a.totals[h]/a.seen:y.tidy(function(){var d=y.mul(y.div(1,u.seen),u.totals[h]);i[h]=d,u.totals[h].dispose(),y.keep(i[h])})},a=this,s=0,o=this.params.metrics;s<o.length;s++)l=o[s],r(l);return[2]})})},t}(Vr),ty=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.epoch=[],this.history={},[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){i==null&&(i={}),this.epoch.push(e);for(r in i)this.history[r]==null&&(this.history[r]=[]),this.history[r].push(i[r]);return[2]})})},t.prototype.syncData=function(){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:e=[],i=[],r=[];for(a in this.history)for(s=this.history[a],o=0;o<s.length;++o)typeof s[o]!="number"&&(l=s[o],e.push(l.data()),i.push(a),r.push(o));return[4,Promise.all(e)];case 1:for(u=d.sent(),c=0;c<u.length;++c)h=this.history[i[c]][r[c]],h.dispose(),this.history[i[c]][r[c]]=u[c][0];return[2]}})})},t}(Vr),ny=function(n){Q(t,n);function t(e,i){var r=n.call(this)||this;if(r.currentEpoch=0,r.yieldEvery=i||"auto",r.yieldEvery==="auto"&&(r.yieldEvery=Pk),r.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");return y.util.isNumber(r.yieldEvery)&&(r.maybeWait=KD(r.maybeWait.bind(r),r.yieldEvery)),r.trainBegin=e.onTrainBegin,r.trainEnd=e.onTrainEnd,r.epochBegin=e.onEpochBegin,r.epochEnd=e.onEpochEnd,r.batchBegin=e.onBatchBegin,r.batchEnd=e.onBatchEnd,r.yield=e.onYield,r}return t.prototype.maybeWait=function(e,i,r){return Le(this,void 0,void 0,function(){var a;return ve(this,function(s){switch(s.label){case 0:return a=[],this.yield!=null?[4,vi(r)]:[3,2];case 1:s.sent(),a.push(this.yield(e,i,r)),s.label=2;case 2:return a.push(y.nextFrame()),[4,Promise.all(a)];case 3:return s.sent(),[2]}})})},t.prototype.onEpochBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.currentEpoch=e,this.epochBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.epochBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.epochEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.epochEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="epoch"&&r.push(y.nextFrame()),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onBatchBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.batchBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.batchBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.batchEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.batchEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="batch"?r.push(y.nextFrame()):y.util.isNumber(this.yieldEvery)&&r.push(this.maybeWait(this.currentEpoch,e,i)),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainBegin!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainBegin(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainEnd!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainEnd(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t}(Vr);function iy(n,t){if(n==null&&(n={}),n instanceof Vr)return[n];if(Array.isArray(n)&&n[0]instanceof Vr)return n;var e=Je(n);return e.map(function(i){return new ny(i,t)})}var ry=function(){function n(){}return n.registerCallbackConstructor=function(t,e){y.util.assert(t>=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function ay(n,t,e,i,r,a,s,o,l){var u=new ty,c=[new _k].concat(ry.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new ey(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function bn(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function uo(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(Ba(n),t,!0),i=y.fill(e.shape,vt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function tr(n,t){return y.tidy(function(){return y.mean(Ba(y.sub(t,n)),-1)})}function co(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function qr(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),vt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function Mk(n,t){return y.tidy(function(){var e=y.clipByValue(t,vt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,vt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(Ba(y.sub(i,a)),-1)})}function Hk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(Ba(e),-1)})}function Vk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function qk(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function Gk(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function _a(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,vt(),1-vt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function ho(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(ok(n)).toInt();t=y.clipByValue(t,vt(),1-vt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return _a(a,t,e)})}function Yk(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function po(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,vt(),1-vt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(Yk(n,e),-1)})}function Kk(n,t){return y.tidy(function(){var e=y.clipByValue(n,vt(),1),i=y.clipByValue(t,vt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function jk(n,t){return y.tidy(function(){var e=y.log(y.add(vt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function Ah(n,t){return y.tidy(function(){var e=uo(n,-1),i=uo(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var fo={meanSquaredError:tr,meanAbsoluteError:co,meanAbsolutePercentageError:qr,meanSquaredLogarithmicError:Mk,squaredHinge:Hk,hinge:Vk,categoricalHinge:qk,logcosh:Gk,categoricalCrossentropy:_a,sparseCategoricalCrossentropy:ho,binaryCrossentropy:po,kullbackLeiblerDivergence:Kk,poisson:jk,cosineProximity:Ah};function Th(n){if(typeof n=="string"){if(n in fo)return fo[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function Nh(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=Wa(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function xh(n,t){return y.tidy(function(){return Wa(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function sy(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function $k(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function Xk(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function oy(n,t){return y.tidy(function(){var e=sy(n,t),i=Xk(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Jk(n,t){return y.tidy(function(){var e=sy(n,t),i=$k(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function ly(n,t){return po(n,t)}function uy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var Zk=tr,Qk=tr,e3=co,t3=co,n3=qr,i3=qr,Ch=_a,r3=Ah,cy=ho,mo={binaryAccuracy:Nh,categoricalAccuracy:xh,precision:oy,categoricalCrossentropy:Ch,sparseCategoricalCrossentropy:cy,mse:Zk,MSE:Qk,mae:e3,MAE:t3,mape:n3,MAPE:i3,cosine:r3};function a3(n){if(typeof n=="string"&&n in mo)return mo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function go(n){if(Wn(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(fo);e<i.length;e++){var r=i[e];if(fo[r]===n){t=r;break}}if(t!==void 0)return t;for(var a=0,s=Object.keys(mo);a<s.length;a++){var r=s[a];if(mo[r]===n){t=r;break}}return t!==void 0?t:n.name}function s3(n){var t={Adagrad:function(){return y.train.adagrad(.01)},Adadelta:function(){return y.train.adadelta(1,.95,vt())},Adam:function(){return y.train.adam(.001,.9,.999,vt())},Adamax:function(){return y.train.adamax(.002,.9,.999,vt(),0)},RMSProp:function(){return y.train.rmsprop(.001,.9,0,vt())},SGD:function(){return y.train.sgd(.01)}};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,n in t)return t[n]();throw new M("Unknown Optimizer "+n)}var hy=1*1024*1024;function dy(n,t,e){if(e===void 0&&(e=!1),n==null||typeof n!="object"||Object.getPrototypeOf(n)!==Object.prototype||!Rh(n))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){var i=JSON.stringify(n);i.length>hy&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+(hy+"."))}}function Rh(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e];if(typeof r!="string")return!1;if(!Rh(n[r]))return!1}return!0}else if(Array.isArray(n)){for(var a=0,s=n;a<s.length;a++){var o=s[a];if(!Rh(o))return!1}return!0}else return!1;else{var l=typeof n;return l==="string"||l==="number"||l==="boolean"}}function h3(n,t,e,i){i===void 0&&(i=console.log);var r=l3(n),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,e=e||[.45,.85,1]):(t=t||98,e=e||[.33,.55,.67,1]),e[e.length-1]<=1&&(e=e.map(function(d){return Math.floor(t*d)}));var s;if(!r){a.push("Receives inputs"),s=[];for(var o in n.nodesByDepth)s.push.apply(s,n.nodesByDepth[o])}i("_".repeat(t)),vo(a,e,i),i("=".repeat(t));for(var l=n.layers,u=0;u<l.length;++u)r?u3(l[u],e,i):c3(l[u],e,s,i),i((u===l.length-1?"=":"_").repeat(t));n.checkTrainableWeightsConsistency();var c=o3(n),h=oo(n.nonTrainableWeights);i("Total params: "+(c+h)),i("Trainable params: "+c),i("Non-trainable params: "+h),i("_".repeat(t))}function o3(n){var t;return n.collectedTrainableWeights!=null?t=oo(n.collectedTrainableWeights):t=oo(n.trainableWeights),t}function l3(n){var t=!0,e=[],i=[];for(var r in n.nodesByDepth)e.push(n.nodesByDepth[r]);for(var a=0,s=e;a<s.length;a++){var o=s[a];if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l<u.length;l++){for(var c=u[l],h=!1,d=0,p=c.inboundNodes;d<p.length;d++){var f=p[d];if(i.indexOf(f)!==-1)if(h){t=!1;break}else h=!0}if(!t)break}return t}function vo(n,t,e){e===void 0&&(e=console.log);for(var i="",r=0;r<n.length;++r)r>0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function u3(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];vo(s,t,e)}function c3(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s<o.length;s++){var l=o[s];if(e!=null&&e.length>0&&e.indexOf(l)===-1)continue;for(var u=0;u<l.inboundLayers.length;++u){var c=l.inboundLayers[u].name,h=l.nodeIndices[u],d=l.tensorIndices[u];a.push(c+"["+h+"]["+d+"]")}}var p=n.name,f=n.getClassName(),m=a.length===0?"":a[0],g=[p+" ("+f+")",r,n.countParams().toString(),m];vo(g,t,i);for(var u=1;u<a.length;++u)vo(["","","",a[u]],t,i)}function py(n,t,e){return(n==="inboundNodes"||n==="outputLayers"||n==="inputLayers")&&t===0&&typeof e=="string"}function Ma(n,t){if(n===null)return null;if(typeof n=="string")return Zi(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Ma(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u];if(u==="name"&&typeof c=="string")s[u]=c;else{var h=Zi(u);s[h]=Ma(c,h)}}return s}}function Oh(n,t){if(n==null)return null;if(typeof n=="string")return Jn(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Oh(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u],h=Jn(u);(u==="name"||u==="className")&&typeof c=="string"?s[h]=c:s[h]=Oh(c,u)}return s}}var Eh="2.6.0";function d3(n,t){if(n.dtype==null||n.dtype===t.dtype)return t;try{return y.cast(t,n.dtype)}catch(e){throw new M("The dtype of the feed ("+t.dtype+") can not be cast to the dtype "+("of the key '"+n.name+"' ("+n.dtype+")."))}}var Gr=function(){function n(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof n)for(var e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(var i=0,r=t;i<r.length;i++){var a=r[i];this.add(a.key,a.value)}}}return n.prototype.add=function(t,e,i){if(this.id2Value[t.id]==null)this.id2Value[t.id]=d3(t,e),this.name2Id[t.name]=t.id,i!=null&&(this.id2Mask[t.id]=i);else throw new M("Duplicate key: name="+t.name+", id="+t.id);return this},n.prototype.addFeed=function(t){this.add(t.key,t.value)},n.prototype.hasKey=function(t){return this.id2Value[t.id]!=null},n.prototype.names=function(){return Object.keys(this.name2Id)},n.prototype.getValue=function(t){if(t instanceof yn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Value[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Value[e]}},n.prototype.getMask=function(t){if(t instanceof yn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Mask[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Mask[e]}},n.prototype.disposeMasks=function(){this.id2Mask!=null&&y.dispose(this.id2Mask)},n}(),Dh={},fy={};function Ha(n,t,e,i){for(var r=e==null?!1:e.training,a=Array.isArray(n),s=a?n:[n],o=s.map(function(X){return X.name}),l=[],u=t.names(),c=0,h=o;c<h.length;c++){var d=h[c];u.indexOf(d)!==-1?l.push(t.getValue(d)):l.push(null)}i!=null&&(i.maxNumTensors=-Infinity,i.minNumTensors=Infinity);var p=o.join(",")+"|"+t.names().join(","),f,m;if(Dh[p]==null){var g=p3(s,t);f=g.sorted,m=g.recipientCounts,Dh[p]=f,fy[p]=m}f=Dh[p],m={},r||Object.assign(m,fy[p]);for(var v=new Gr(t),b=0;b<f.length;++b){if(i!=null){var w=y.memory().numTensors;w>i.maxNumTensors&&(i.maxNumTensors=w),w<i.minNumTensors&&(i.minNumTensors=w)}var S=f[b],L=S.sourceLayer;if(L instanceof Pa)continue;for(var x=[],C=[],R=[],D=!1,k=0,W=S.inputs;k<W.length;k++){var F=W[k],P=v.getValue(F),H=v.getMask(F);x.push(P),C.push(H),H!=null&&(D=!0),r||(m[F.name]--,m[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!P.isDisposed&&F.sourceLayer.stateful!==!0&&R.push(P))}D&&(e=e||{},e.mask=C[0]);var _=Je(L.apply(x,e)),K=null;L.supportsMasking&&(K=L.computeMask(x,C));for(var j=f3(S),q=Array.isArray(j)?j:[j],G=0;G<q.length;++G){v.hasKey(q[G])||v.add(q[G],_[G],Array.isArray(K)?K[0]:K);var Z=o.indexOf(q[G].name);Z!==-1&&(l[Z]=_[G])}r||y.dispose(R)}return v.disposeMasks(),a?l:l[0]}function p3(n,t){y.util.assert(n!=null&&n.length>0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=my(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s<o.length;s++){for(var l=o[s],u=my(l,t),c=u.sorted,h=u.recipientMap,d=0,p=c;d<p.length;d++){var f=p[d];a.has(f.name)||(e.push(f),a.add(f.name))}var m=function(v){i[v]==null&&(i[v]=new Set),h[v].forEach(function(b){return i[v].add(b)})};for(var g in h)m(g)}return{sorted:e,recipientCounts:m3(i)}}function m3(n){var t={};for(var e in n)t[e]=n[e].size;return t}function my(n,t){for(var e=new Set,i=[],r={},a=0,s=t.names();a<s.length;a++){var o=s[a];e.add(o)}var l=[],u=[];for(l.push(n);l.length>0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d<p.length;d++){var f=p[d];if(r[f.name]==null&&(r[f.name]=new Set),r[f.name].add(c.name),e.has(f.name))continue;l.push(f)}}}return{sorted:i,recipientMap:r}}function f3(n){var t;if(n.sourceLayer.inboundNodes.length===1)t=n.sourceLayer.output;else{for(var e=null,i=0;i<n.sourceLayer.inboundNodes.length;++i)for(var r=0,a=n.sourceLayer.inboundNodes[i].outputTensors;r<a.length;r++){var s=a[r];if(s.id===n.id){e=i;break}}t=n.sourceLayer.getOutputAt(e)}return t}var g3=function(n){Q(t,n);function t(e){var i=n.call(this,{})||this;if(i.containerNodes=new Set,i.name=e.name,i.name==null){var r=i.getClassName().toLowerCase();i.name=ao(r)}if(i.supportsMasking=!1,i.trainable_=!0,Array.isArray(e.inputs)?i.inputs=e.inputs.slice():i.inputs=[e.inputs],Array.isArray(e.outputs)?i.outputs=e.outputs.slice():i.outputs=[e.outputs],fi(i.inputs).length!==i.inputs.length)throw new M("The list of inputs passed to the model is redundant. All inputs should only appear once. Found: "+(""+i.inputs.map(function(Ce){return Ce.name})));fi(i.outputs).length!==i.outputs.length&&console.warn("The list of outputs passed to the model is redundant. All outputs should only appear once. Found: "+(""+i.outputs.map(function(Ce){return Ce.name}))),i.inputLayers=[],i.inputLayersNodeIndices=[],i.inputLayersTensorIndices=[],i.outputLayers=[],i.outputLayersNodeIndices=[],i.outputLayersTensorIndices=[],i.layers=[],i.internalContainerRefs=[];for(var a=0,s=i.outputs;a<s.length;a++){var o=s[a],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;i.outputLayers.push(l),i.outputLayersNodeIndices.push(u),i.outputLayersTensorIndices.push(c)}for(var h=0,d=i.inputs;h<d.length;h++){var o=d[h],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;Wn(u===0,"input layer has >1 nodes"),Wn(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p<i.inputLayers.length;p++){var l=i.inputLayers[p];if(!(l instanceof Pa))throw new TypeError("Input layers to a LayersModel must be InputLayer objects. "+("Received inputs: "+e.inputs+". ")+("Input "+p+" (0-based) originates ")+("from layer type "+l.getClassName()+"."));i.inputNames.push(l.name),i.feedInputShapes.push(l.batchInputShape),i.feedInputNames.push(l.name)}for(var f=0,m=i.outputLayers;f<m.length;f++){var l=m[f];i.outputNames.push(l.name)}i.internalInputShapes=i.inputs.map(function(Ce){return Ce.shape}),i.internalOutputShapes=i.outputs.map(function(Ce){return Ce.shape});for(var g={},v={},b={},w={},S={},L=[],x=function(Ce,kt,pt,it,on,hs){(it==null||on==null||hs==null)&&(it=Ce.sourceLayer,on=Ce.nodeIndex,hs=Ce.tensorIndex);var Bt=it.inboundNodes[on];if(pt.indexOf(Bt)!==-1)throw new Fn("The tensor "+Ce.name+' at layer "'+it.name+'" is part of a cycle.');if(kt.indexOf(Bt)!==-1)return;i.containerNodes.add(t.nodeKey(it,on)),it.id in S||(S[it.id]=Object.keys(S).length),pt.indexOf(Bt)===-1&&pt.push(Bt);for(var II=Bt.inboundLayers.length,yr=0;yr<II;yr++){var AI=Bt.inputTensors[yr],TI=Bt.inboundLayers[yr],NI=Bt.nodeIndices[yr],xI=Bt.tensorIndices[yr];x(AI,kt,pt,TI,NI,xI)}for(kt.push(Bt);pt.indexOf(Bt)>=0;)pt.splice(pt.indexOf(Bt),1);L.push(Bt)},C=[],R=[],D=0,k=i.outputs;D<k.length;D++){var o=k[D];x(o,C,R)}for(var W=L.slice().reverse(),F=0,P=W;F<P.length;F++){var H=P[F];v[H.id]=H,H.id in g||(g[H.id]=0);var _=g[H.id],K=b[H.outboundLayer.id]==null?0:b[H.outboundLayer.id];_=Math.max(_,K),b[H.outboundLayer.id]=_,w[H.outboundLayer.id]=H.outboundLayer,g[H.id]=_;for(var p=0;p<H.inboundLayers.length;p++){var j=H.inboundLayers[p],u=H.nodeIndices[p],q=j.inboundNodes[u],G=g[q.id]==null?0:g[q.id];g[q.id]=Math.max(_+1,G),v[q.id]=q}}var Z={};for(var X in g){var _=g[X];_ in Z||(Z[_]=[]),Z[_].push(v[X])}var ee={};for(var ne in b){var _=b[ne];_ in ee||(ee[_]=[]),ee[_].push(w[ne])}var ie=Object.keys(ee).map(function(Ce){return parseInt(Ce,10)}).sort(to);i.layers=[];for(var te=0,re=ie;te<re.length;te++){var _=re[te],le=ee[_];le.sort(function(kt,pt){var it=S[kt.id],on=S[pt.id];return it<on?-1:it>on?1:0});for(var he=0,ye=le;he<ye.length;he++){var l=ye[he];l instanceof t&&i.internalContainerRefs.push(l),i.layers.push(l)}}i.layersByDepth=ee,ie=Object.keys(Z).map(function(Ce){return parseInt(Ce,10)}).sort(to);for(var Oe=i.inputs.slice(),Fe=[],_e=0,Me=ie;_e<Me.length;_e++)for(var _=Me[_e],ht=0,St=Z[_];ht<St.length;ht++){var H=St[ht],l=H.outboundLayer;if(l!=null){for(var Ge=0,Ue=H.inputTensors;Ge<Ue.length;Ge++){var o=Ue[Ge];if(Oe.indexOf(o)===-1)throw new Fn("Graph disconnected: cannot obtain value for tensor "+o+(' at layer "'+l.name+'". ')+"The following previous layers were accessed without "+("issue: "+Fe))}for(var me=0,Dt=H.outputTensors;me<Dt.length;me++){var o=Dt[me];Oe.push(o)}Fe.push(l.name)}}i.nodesByDepth=Z;for(var lt=i.layers.map(function(Ce){return Ce.name}),dt=function(Ce){var kt=lt.filter(function(pt){return pt===Ce}).length;if(kt!==1)throw new Fn('The name "'+Ce+'" is used '+kt+" times in the model. All layer names should be unique. Layer names: "+JSON.stringify(lt))},xn=0,Rt=lt;xn<Rt.length;xn++){var Lt=Rt[xn];dt(Lt)}return i.outboundNodes=[],i.inboundNodes=[],new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:i.inputs,outputTensors:i.outputs,inputMasks:i.inputs.map(function(Ce){return null}),outputMasks:i.outputs.map(function(Ce){return null}),inputShapes:i.inputs.map(function(Ce){return Ce.shape}),outputShapes:i.outputs.map(function(Ce){return Ce.shape})}),i.built=!0,i._refCount=1,i}return t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Container '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){this.assertNotDisposed();var e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(var i=0,r=this.layers;i<r.length;i++){var a=r[i];e.numDisposedVariables+=a.dispose().numDisposedVariables}for(var s=0,o=this.internalContainerRefs;s<o.length;s++){var l=o[s];e.numDisposedVariables+=l.dispose().numDisposedVariables}}return e.refCountAfterDispose=this._refCount,e},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this.layers.forEach(function(i){i._trainableWeights.forEach(function(r){return r.trainable=e})}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(this._trainableWeights.length>0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e=e.concat(a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.layers;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.loadWeights=function(e,i){i===void 0&&(i=!0);for(var r={},a=0,s=0,o=this.layers;s<o.length;s++)for(var l=o[s],u=0,c=l.weights;u<c.length;u++){var h=c[u];if(r[h.originalName]!=null)throw new M("Duplicate weight name: "+h.originalName);r[h.originalName]=h,a++}var d=[];for(var p in e){var f=p;if(r[p]==null){var m=p.split("/"),g=m.slice(0,-2).concat([m[m.length-1]]);f=g.join("/")}if(r[f]!=null)d.push([r[f],e[p]]);else if(i)throw new M("Provided weight data has no target variable: "+p);delete r[f]}if(i){var v=[];for(var b in r)v.push(b);if(v.length>0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}Ih(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+Eh,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=Oh(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Je(e);for(var a=new Gr,s=0;s<r.inputs.length;++s)a.add(r.inputs[s],e[s]);return Ha(r.outputs,a,i)})},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){e=Je(e);var a;return i==null?a=Ji(null,e.length):a=Je(i),r.runInternalGraph(e,a)[1]})},t.prototype.computeOutputShape=function(e){var i=so(e);if(i.length!==this.inputLayers.length)throw new M("Invalid inputShape argument "+e+": "+("model has "+this.inputLayers.length+" tensor inputs."));for(var r={},a=0;a<i.length;a++){var s=this.inputLayers[a],o=i[a],l=s.name+"_0_0";r[l]=o}var u=Object.keys(this.nodesByDepth).map(function(P){return parseInt(P,10)}).sort(to);if(u.length>1)for(var c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){var g=m[f],s=g.outboundLayer;if(this.inputLayers.map(function(H){return H.id}).indexOf(s.id)!==-1)continue;for(var v=[],b=0;b<g.inboundLayers.length;b++){var w=g.inboundLayers[b],S=g.nodeIndices[b],L=g.tensorIndices[b],l=w.name+"_"+S+"_"+L,x=r[l];v.push(x)}for(var C=s.computeOutputShape(Mt(v)),R=so(C),D=s.inboundNodes.indexOf(g),b=0;b<R.length;b++){var l=s.name+"_"+D+"_"+b;r[l]=R[b]}}for(var k=[],W=[],a=0;a<this.outputLayers.length;a++){var s=this.outputLayers[a],D=this.outputLayersNodeIndices[a],L=this.outputLayersTensorIndices[a],l=s.name+"_"+D+"_"+L;W.push(l)}for(var a=0;a<W.length;a++){var F=W[a];Wn(F in r),k.push(r[F])}return Mt(k)},t.prototype.runInternalGraph=function(e,i){i==null&&(i=Ji(null,e.length));for(var r={},a=0;a<this.inputs.length;++a){var s=this.inputs[a],o=e[a],l=i[a];r[s.id]=[o,l]}for(var u=Object.keys(this.nodesByDepth).map(function(ee){return parseInt(ee,10)}).sort(to),c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){for(var g=m[f],v=g.outboundLayer,b=g.inputTensors,w=g.outputTensors,S=new Array,L=0,x=b;L<x.length;L++){var s=x[L];s.id in r&&S.push(r[s.id])}if(S.length===b.length){var C={},R=void 0,D=void 0,k=void 0,W=void 0;if(g.callArgs!=null&&(C=g.callArgs),S.length===1){var F=S[0],P=F[0],H=F[1];C.mask==null&&(C.mask=H),k=Je(v.call(P,C)),W=Je(v.computeMask(P,H)),R=[P],D=[H]}else R=S.map(function(ee){return ee[0]}),D=S.map(function(ee){return ee[1]}),C.mask==null&&(C.mask=D),k=Je(v.call(R,C)),W=Je(v.computeMask(R,D));if(v.activityRegularizer)throw new Te("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(var a=0;a<w.length;++a){var s=w[a],o=k[a],l=W[a];r[s.id]=[o,l]}}}for(var _=[],K=[],j=[],q=0,G=this.outputs;q<G.length;q++){var s=G[q];Wn(s.id in r,"Could not compute output "+s.name+" : "+s.id);var Z=r[s.id],X=Z[0],l=Z[1];j.push(X.shape),_.push(X),K.push(l)}return[_,K,j]},t.prototype.buildNodeConversionMap=function(e){for(var i={},r,a=0,s=this.layers;a<s.length;a++){var o=s[a];r=o instanceof t?1:0;for(var l=0;l<o.inboundNodes.length;l++){var u=t.nodeKey(o,l);this.containerNodes.has(u)&&(i[u]=r,r+=1)}}return i},t.prototype.getLayer=function(e,i){if(i!=null){if(this.layers.length<=i)throw new M("Was asked to retrieve layer at index "+i+", but model only "+("has "+this.layers.length+" layer(s)."));return this.layers[i]}else if(e==null)throw new M("Provide either a layer name or layer index");for(var r=0,a=this.layers;r<a.length;r++){var s=a[r];if(s.name===e)return s}throw new M("No such layer: "+e)},t.prototype.calculateLosses=function(){var e=this;return y.tidy(function(){for(var i=[],r=0,a=e.layers;r<a.length;r++)for(var s=a[r],o=0;o<s.inboundNodes.length;++o){var l=t.nodeKey(s,o);e.containerNodes.has(l)&&i.push.apply(i,s.calculateLosses())}return i})},t.prototype.getConfig=function(){for(var e={name:this.name},i=this.buildNodeConversionMap(this.layers),r=[],a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=o.getClassName(),u=o.getConfig(),c=[],h=0;h<o.inboundNodes.length;h++){var d=o.inboundNodes[h],p=t.nodeKey(o,h),f={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),f=d.callArgs}catch(D){console.warn("Layer "+o.name+" was passed non-serializable keyword arguments: "+(d.callArgs+". They will not be included ")+"in the serialized model (and thus will be missing at deserialization time)."),f={}}if(d.inboundLayers.length>0){for(var m=[],g=0;g<d.inboundLayers.length;g++){var v=d.inboundLayers[g],b=d.nodeIndices[g],w=d.tensorIndices[g],S=t.nodeKey(v,b),L=i[S];L==null&&(L=0),m.push([v.name,L,w,f])}c.push(m)}}}var x={};x.name=o.name,x.className=l,x.config=u,x.inboundNodes=c,r.push(x)}e.layers=r;for(var C=[],g=0;g<this.inputLayers.length;g++){var o=this.inputLayers[g],b=this.inputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.inputLayersTensorIndices[g];C.push([o.name,L,w])}e.inputLayers=C;for(var R=[],g=0;g<this.outputLayers.length;g++){var o=this.outputLayers[g],b=this.outputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.outputLayersTensorIndices[g];R.push([o.name,L,w])}return e.outputLayers=R,e},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s={},o={};function l(G,Z){G.name in o?o[G.name].push(Z):o[G.name]=[Z]}function u(G,Z){for(var X=[],ee,ne=0,ie=Z;ne<ie.length;ne++){var te=ie[ne],re=te[0],le=te[1],he=te[2];if(ee=te[3]==null?{}:te[3],!(re in s)){l(G,Z);return}var ye=s[re];if(ye.inboundNodes.length<=le){l(G,Z);return}var Oe=ye.inboundNodes[le];X.push(Oe.outputTensors[he])}X.length>0&&G.apply(Mt(X),ee)}function c(G){var Z=G.name,X=bn(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p<f.length;p++){var m=f[p];c(m)}for(;!YD(o);)for(var g=0,v=d;g<v.length;g++){var m=v[g],b=s[m.name];if(b.name in o){var w=o[b.name];delete o[b.name];for(var S=0,L=w;S<L.length;S++){var x=L[S];u(b,x)}}}for(var C=[],R=[],D=i.inputLayers,k=0,W=D;k<W.length;k++){var m=W[k],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;C.push(_[H])}for(var K=i.outputLayers,j=0,q=K;j<q.length;j++){var m=q[j],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;R.push(_[H])}return new e({inputs:C,outputs:R,name:h})},Object.defineProperty(t.prototype,"stateful",{get:function(){if(this._stateful)throw new M("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(var e=0,i=this.layers;e<i.length;e++){var r=i[e];if(r.stateful)return!0}return!1},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){var e=this;y.tidy(function(){e.layers.forEach(function(i){i.stateful&&i.resetStates()})})},t}(De);function v3(n,t,e){var i=t.length;if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(a){return null});if(i===1)return Array.isArray(n)&&n.length===1?n:typeof n=="object"&&t[0]in n?[n[t[0]]]:[n];if(Array.isArray(n)){if(n.length!==i)throw new Error("Provided "+e+" is an array of "+n.length+" "+("element(s), but the model has "+i+" outputs. ")+"Make sure a set of weights is provided for each model output.");return n}else if(typeof n=="object"&&Object.keys(n).length>0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function gy(n,t){return v3(n,t,"classWeight")}function vy(n,t,e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function y3(n,t){return y.mul(n,t)}var b3=32;function by(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor.  The provided Dataset instead generates "+(""+t)});var a=yy("input",n.inputNames,e),s=yy("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs.  (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs.  (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},u=0;u<a.length;u++)l(u);for(var c=function(d){y.util.assert(s[d].shape[0]===o,function(){return"Batch size mismatch: output "+(n.outputNames[d]+" has "+s[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},h=0;h<s.length;h++)c(h);return{xs:a,ys:s}}function yy(n,t,e){if(e instanceof y.Tensor)return[e];if(Array.isArray(e))return y.util.assert(e.length===t.length,function(){return"Received an array of "+e.length+" Tensors, but expected "+t.length+" to match the "+n+" keys "+t+"."}),e;for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r];if(e[s]==null)throw new M("The feature data generated by the dataset lacks the required "+(n+" key '"+s+"'."));i.push(e[s])}return i}function w3(n){if(n.length===3)throw new Te("Validation with sample weights is not implemented yet.");return{xs:n[0],ys:n[1]}}function L3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k,W,F,P,H,_,K,F,j,q,G,Z,F;return ve(this,function(X){switch(X.label){case 0:if(i=e.batchesPerEpoch!=null,y.util.assert(n.optimizer!=null,function(){return"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."}),y.util.assert(e!=null,function(){return"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."}),y.util.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(wy(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=w3(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=iy(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=ay(h,d,e.epochs,null,null,S3(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g<e.epochs?(b={},[4,f.onEpochBegin(g)]):[3,23];case 5:return X.sent(),w=0,S=0,i?[3,7]:[4,t.iterator()];case 6:v=X.sent(),X.label=7;case 7:return(i?w<e.batchesPerEpoch:!0)?[4,v.next()]:[3,21];case 8:return L=X.sent(),i&&L.done?(console.warn("You provided `batchesPerEpoch` as "+(e.batchesPerEpoch+", ")+"but your dataset iterator ran out of data after "+(w+" batches; ")+"interrupting training. Make sure that your dataset can generate at least `batchesPerEpoch * epochs` batches (in this case, "+(e.batchesPerEpoch*e.epochs+" batches). ")+"You may need to use the repeat() function when building your dataset."),[3,21]):L.value!=null?(x=by(n,L.value),C=x.xs,R=x.ys,D={},D.batch=S,D.size=C[0].shape[0],[4,f.onBatchBegin(S,D)]):[3,15];case 9:if(X.sent(),k=[],!(e.classWeight!=null))return[3,13];W=gy(e.classWeight,n.outputNames),F=0,X.label=10;case 10:return F<W.length?(H=(P=k).push,[4,vy(R[F],null,W[F])]):[3,13];case 11:H.apply(P,[X.sent()]),X.label=12;case 12:return++F,[3,10];case 13:for(_=C.concat(R).concat(k),K=l(_),y.dispose(_),F=0;F<u.length;++F)j=u[F],q=K[F],D[j]=q,y.keep(q);return[4,f.onBatchEnd(S,D)];case 14:X.sent(),Zv(D),S++,w++,X.label=15;case 15:return(i?w>=e.batchesPerEpoch:L.done)?r?(G=void 0,wy(e.validationData)?(Z=Je,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Je(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?b3:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F<n.metricsNames.length;++F)b["val_"+n.metricsNames[F]]=G[F];X.label=19;case 19:return[3,21];case 20:return n.stopTraining_?[3,21]:[3,7];case 21:return[4,f.onEpochEnd(g,b)];case 22:return X.sent(),g++,n.stopTraining_?[3,23]:[3,4];case 23:return[4,f.onTrainEnd()];case 24:return X.sent(),[4,n.history.syncData()];case 25:return X.sent(),[2,n.history];case 26:return n.isTraining=!1,[7];case 27:return[2]}})})}function S3(n,t){var e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(n.size)&&(e=n.size),e}function wy(n){return typeof n.iterator=="function"}function I3(n){return typeof n.next=="function"}function A3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:if(e=e||{},i=e.batches!=null,r=n.testFunction,a=[],e.verbose>0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),I3(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=by(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var x=0;x<L.length;++x)a.push(y.scalar(0));for(var C=S[0].shape[0],R=function(D){var k=L[D],W=a[D];a[D]=y.tidy(function(){return y.add(a[D],y.mul(C,k))}),u>0&&y.dispose(W)},x=0;x<L.length;++x)R(x);y.dispose(L),l+=C,++u}return a}),m.done?(i&&console.warn("Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least `batches` "+("batches (in this case, "+e.batches+" batches). ")+"You may need to use the repeat() function when building your dataset."),[2,"break"]):[2]}})},f.label=4;case 4:return(i?u<e.batches:!0)?[5,c()]:[3,6];case 5:return h=f.sent(),h==="break"?[3,6]:[3,4];case 6:for(d=0;d<a.length;++d)p=a[d],a[d]=y.div(a[d],l),y.dispose(p);return[2,Mt(a)]}})})}function kh(n){y.util.assert(n>0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function Va(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return er(i,t,e-t)}):er(n,t,e-t)}function Fh(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return Fh(e,t)}):Uv(n,t.dtype==="int32"?t:t.toInt())})}function Wh(n,t){for(var e=[],i=0,r=null;i<n;)r=i+t,r>=n&&(r=n),e.push([i,r]),i=r;return e}function T3(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Le(this,void 0,void 0,function(){var m,g,v,b,w,S,L,x,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=vn(0,g)),s==null&&(s=1),b=ay(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(K){switch(K.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(K.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=Wh(g,r),P=function(j){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(j,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[j][0],X=F[j][1],ee=er(W,Z,X-Z);q.batch=j,q.size=X-Z;for(var ne=Fh(e,ee),ie=t(ne),te=0;te<i.length;++te){var re=i[te],le=ie[te];q[re]=le,y.keep(le)}if(j===F.length-1&&m)for(var he=n.testLoop(l,u,r),te=0;te<i.length;++te){var re=i[te],le=he[te];y.keep(le),k["val_"+re]=le}}),[4,w.onBatchEnd(j,q)];case 2:return G.sent(),Zv(q),n.stopTraining_?[2,"break"]:[2]}})},H=0,K.label=3;case 3:return H<F.length?[5,P(H)]:[3,6];case 4:if(_=K.sent(),_==="break")return[3,6];K.label=5;case 5:return++H,[3,3];case 6:W.dispose(),K.label=7;case 7:return[4,w.onEpochEnd(D,k)];case 8:return K.sent(),n.stopTraining_?[2,"break"]:[2]}})},x=d,R.label=2;case 2:return x<a?[5,L(x)]:[3,5];case 3:if(C=R.sent(),C==="break")return[3,5];R.label=4;case 4:return++x,[3,2];case 5:return[4,w.onTrainEnd()];case 6:return R.sent(),[4,n.history.syncData()];case 7:return R.sent(),[2,n.history]}})})}function N3(n,t,e,i){return i===void 0&&(i={}),Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k;return ve(this,function(W){switch(W.label){case 0:if(n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,W.label=1;case 1:return W.trys.push([1,,7,8]),h=i.batchSize==null?32:i.batchSize,kh(h),d=!1,[4,n.standardizeUserData(t,e,i.sampleWeight,i.classWeight,d,h)];case 2:if(p=W.sent(),r=p[0],a=p[1],c=p[2],f=!1,m=void 0,!(i.validationData!=null&&i.validationData.length>0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=Va(r,b,w),r=Va(r,0,b),u=Va(a,b,w),a=Va(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),x=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=x.slice().concat(x.map(function(F){return"val_"+F}))):(C=null,m=[],R=x.slice()),D=iy(i.callbacks,i.yieldEvery),[4,T3(n,L,S,x,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,nr(r,t),nr(a,e),nr(l,s),nr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function Sy(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e<n.length;++e){var i=n[e];if(i.rank===1)t.push(Ua(i,1));else{if(i.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(i)}}return t}function nr(n,t){if(n==null)return;var e=[];if(t instanceof y.Tensor)e.push(t.id);else if(Array.isArray(t))t.forEach(function(l){return e.push(l.id)});else if(t!=null)for(var i in t){var r=t[i];e.push(r.id)}var a=[];if(n instanceof y.Tensor)e.indexOf(n.id)===-1&&a.push(n);else if(Array.isArray(n))n.forEach(function(l){e.indexOf(l.id)===-1&&a.push(l)});else if(n!=null)for(var s in n){var o=n[s];e.indexOf(o.id)===-1&&a.push(o)}a.forEach(function(l){l.isDisposed||l.dispose()})}function x3(n){return n instanceof y.Tensor}function Uh(n){return Array.isArray(n)}function Ly(n){return!x3(n)&&!Uh(n)}function Iy(n,t,e,i,r){if(i===void 0&&(i=!0),r===void 0&&(r=""),t==null||t.length===0){if(n!=null){var a=!1;if(Uh(n)&&n.length>0)a=!0;else if(Ly(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(Ly(n)){n=n,o=[];for(var l=0,u=t;l<u.length;l++){var c=u[l];if(n[c]==null)throw new M('No data provided for "'+c+'". Need data for each key in: '+(""+t));o.push(n[c])}}else if(Uh(n)){if(n=n,n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("model expected. Expected to see "+t.length+" Tensor(s), but ")+("instead got the following list of Tensor(s): "+n));o=n}else{if(n=n,t.length>1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=Sy(o),e!=null)for(var h=0;h<t.length;++h){if(e[h]==null)continue;var d=o[h];if(d.shape.length!==e[h].length)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have "+e[h].length+" dimension(s). but got array with ")+("shape "+d.shape));for(var p=0;p<e[h].length;++p){if(p===0&&!i)continue;var f=d.shape[p],m=e[h][p];if(m!=null&&m>=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function C3(n,t,e){var i=fi(n.map(function(a){return a.shape[0]}));i.sort();var r=fi(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function R3(n,t,e){for(var i=[tr,po,_a],r=0;r<n.length;++r){var a=n[r],s=t[r],o=e[r];if(s==null)continue;if(s===_a&&a.shape[a.shape.length-1]===1)throw new M("You are passing a target array of shape "+a.shape+" while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].");if(i.indexOf(s)!==-1)for(var l=a.shape.slice(1),u=o.slice(1),c=0;c<l.length;++c){var h=l[c],d=u[c];if(d!=null&&h!==d)throw new M("A target Tensor with shape "+a.shape+" was passed for an "+("output of shape "+o+", while using a loss function that ")+"expects targets to have the same shape as the output.")}}}function Ay(n,t,e,i,r){i===void 0&&(i=!0),r===void 0&&(r="");var a;if(Array.isArray(n)){if(n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("the model expected. Expected to see "+t.length+" Tensor(s),")+(" but instead got "+n.length+" Tensors(s)."));a=n}else{if(t.length>1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s<t.length;++s){if(e[s]==null)continue;var o=a[s];if(o.shape.length!==e[s].length)throw new M("Error when checking "+r+": expected "+t[s]+" "+("to have "+e[s].length+" dimension(s), but got array with ")+("shape "+JSON.stringify(o.shape)));for(var l=0;l<e[s].length;++l){if(l===0&&!i)continue;var u=o.shape[l],c=e[s][l];if(c!=null&&c!==u)throw new M("Error when checking "+r+": expected "+(t[s]+" to have shape "+JSON.stringify(e[s])+" but ")+("got array with shape "+JSON.stringify(o.shape)+"."))}}}function O3(n,t){if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(l){return[]});var e;if(typeof n=="string"||typeof n=="function")e=[n];else if(Array.isArray(n)||typeof n=="object")e=n;else throw new TypeError("Type of metrics argument not understood. Expected an string,"+("function, Array, or Object, found: "+n));if(Array.isArray(e))return t.map(function(l){return e});for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r],o=e.hasOwnProperty(s)?e[s]:[];Array.isArray(o)||(o=[o]),i.push(o)}return i}var E3="layers-model",yi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.isTraining=!1,i}return t.prototype.summary=function(e,i,r){if(r===void 0&&(r=console.log),!this.built)throw new M("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");h3(this,e,i,r)},t.prototype.compile=function(e){var i=this;if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=s3(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof y.Optimizer))throw new M("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}var r=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(var a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new M('Unknown entry in loss dictionary: "'+a+'". '+("Only expected the following keys: "+this.outputNames));for(var s=0,o=this.outputNames;s<o.length;s++){var l=o[s];e.loss[l]==null&&console.warn('Output "'+l+'" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data '+("to be passed to "+l+" during training")),r.push(Th(e.loss[l]))}}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new M("When passing an Array as loss, it should have one entry per "+("model output. The model has "+this.outputs.length+" output(s), ")+("but you passed loss="+e.loss+"."));var u=e.loss;r=u.map(function(v){return Th(v)})}else{var c=Th(e.loss);this.outputs.forEach(function(v){r.push(c)})}this.lossFunctions=r,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(var h=0;h<this.outputs.length;++h){var d=this.internalOutputShapes[h],p=this.outputNames[h];this.feedOutputNames.push(p),this.feedOutputShapes.push(d),this.feedLossFns.push(this.lossFunctions[h])}var f=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],Qi("loss",function(){for(var v=0;v<i.outputs.length;++v){if(f.indexOf(v)!==-1)continue;var b=i.lossFunctions[v];i.outputs.length>1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=O3(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};Qi("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(x){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var K=i.internalOutputShapes[w];K[K.length-1]===1||i.lossFunctions[w]===po?["accuracy","acc"].indexOf(_)!==-1?D=Nh:["crossentropy","ce"].indexOf(_)!==-1&&(D=ly):i.lossFunctions[w]===ho?["accuracy","acc"].indexOf(_)!==-1?D=uy:["crossentropy","ce"].indexOf(_)!==-1&&(D=cy):["accuracy","acc"].indexOf(_)!==-1?D=xh:["crossentropy","ce"].indexOf(_)!==-1&&(D=Ch);var j=void 0;["accuracy","acc"].indexOf(_)!==-1?j="acc":["crossentropy","ce"].indexOf(_)!==-1&&(j="ce"),k=D,R=C+j}else{var q=a3(_);k=q,R=C+go(_)}var G;Qi(R,function(){G=k}),g(w,R,G)},F=0,P=x;F<P.length;F++){var H=P[F];W(H)}};L(S)},b=0;b<i.outputs.length;++b)v(b)}),this.collectedTrainableWeights=this.trainableWeights},t.prototype.checkTrainableWeightsConsistency=function(){if(this.collectedTrainableWeights==null)return;this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")},t.prototype.evaluate=function(e,i,r){r===void 0&&(r={});var a=r.batchSize==null?32:r.batchSize;kh(a);var s=!0,o=this.standardizeUserDataXY(e,i,s,a);try{var l=o[0].concat(o[1]);this.makeTestFunction();var u=this.testFunction,c=this.testLoop(u,l,a,r.verbose,r.steps);return Mt(c)}finally{nr(o[0],e),nr(o[1],i)}},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return this.makeTestFunction(),[2,A3(this,e,i)]})})},t.prototype.checkNumSamples=function(e,i,r,a){a===void 0&&(a="steps");var s;if(r!=null){if(s=null,i!=null)throw new M("If "+a+" is set, batchSize must be null or undefined."+("Got batchSize = "+i))}else if(e!=null)Array.isArray(e)?s=e[0].shape[0]:s=e.shape[0];else throw new M("Either the input data should have a defined shape, or "+(a+" shoud be specified."));return s},t.prototype.execute=function(e,i){if(Array.isArray(i)&&i.length===0)throw new M("`outputs` is an empty Array, which is not allowed.");var r=Array.isArray(i),a=r?i:[i],s=this.retrieveSymbolicTensors(a),o=new Gr;if(e instanceof y.Tensor&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new M("The number of inputs provided ("+e.length+") does not match the number of inputs of this model "+("("+this.inputs.length+")."));for(var l=0;l<this.inputs.length;++l)o.add(this.inputs[l],e[l])}else for(var u=0,c=this.inputs;u<c.length;u++){var h=c[u],d=e[h.name];if(d==null)throw new M("No value is provided for the model's input "+h.name);o.add(h,d)}var p=Ha(s,o);return r?p:p[0]},t.prototype.retrieveSymbolicTensors=function(e){for(var i=Ji(null,e.length),r=e.length,a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=Array.isArray(o.output)?o.output:[o.output],u=l.map(function(p){return p.name}),c=0;c<e.length;++c){var h=u.indexOf(e[c]);if(h!==-1&&(i[c]=l[h],r--),r===0)break}if(r===0)break}if(r>0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=Wh(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=Va(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v<m.length;++v)g.push({key:a.inputs[v],value:m[v]});else g.push({key:a.inputs[0],value:m});var b=new Gr(g);return Ha(a.outputs,b)});d.forEach(function(p,f){return l[f].push(p)})},c=0;c<o.length;++c)u(c);return Mt(l.map(function(h){return y.concat(h,0)}))})},t.prototype.predict=function(e,i){i===void 0&&(i={});var r=Sy(e);Ay(r,this.inputNames,this.feedInputShapes,!1);try{var a=i.batchSize==null?32:i.batchSize;return kh(a),this.predictLoop(r,a)}finally{nr(r,e)}},t.prototype.predictOnBatch=function(e){Ay(e,this.inputNames,this.feedInputShapes,!0);var i=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,i)},t.prototype.standardizeUserDataXY=function(e,i,r,a){if(this.optimizer_==null)throw new Fn("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");for(var s=[],o=0;o<this.feedOutputShapes.length;++o){var l=this.feedOutputShapes[o],u=this.feedLossFns[o];u===ho?s.push(l.slice(0,l.length-1).concat([1])):s.push(l)}if(e=Iy(e,this.feedInputNames,this.feedInputShapes,!1,"input"),i=Iy(i,this.feedOutputNames,s,!1,"target"),C3(e,i),R3(i,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Le(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=gy(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p<d.length?(m=(f=h).push,[4,vy(c[p],null,d[p])]):[3,4];case 2:m.apply(f,[g.sent()]),g.label=3;case 3:return++p,[3,1];case 4:return[2,[u,c,h]]}})})},t.prototype.testLoop=function(e,i,r,a,s){var o=this;return a===void 0&&(a=0),y.tidy(function(){var l=o.checkNumSamples(i,r,s,"steps"),u=[];if(a>0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=Wh(l,r),h=y.tensor1d(vn(0,l)),d=0;d<c.length;++d){var p=c[d][0],f=c[d][1],m=er(h,p,f-p),g=Fh(i,m),v=e(g);if(d===0)for(var b=0;b<v.length;++b)u.push(y.scalar(0));for(var b=0;b<v.length;++b){var w=v[b];u[b]=y.add(u[b],y.mul(f-p,w))}}for(var b=0;b<u.length;++b)u[b]=y.div(u[b],l);return u})},t.prototype.getDedupedMetricsNames=function(){for(var e=this.metricsNames,i=[],r=0;r<e.length;++r){var a=e[r],s=a;if(wv(e,a)>1){var o=wv(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f<e.inputs.length;++f)p.push({key:e.inputs[f],value:a[f]});for(var m=new Gr(p),g=Ha(e.outputs,m,{training:!0}),v,f=0;f<e.lossFunctions.length;++f){var b=e.lossFunctions[f],w=b(s[f],g[f]);o[f]!=null&&(w=y3(w,o[f]));var S=y.mean(w);r.push(S),f===0?v=w:v=y.add(v,w)}for(var f=0;f<e.metricsTensors.length;++f){var L=void 0;if(e.outputs.length>1&&f<e.outputs.length)L=r[f];else{var x=e.metricsTensors[f][0],C=e.metricsTensors[f][1];L=y.mean(x(s[C],g[C]))}y.keep(L),l.push(L)}return v=y.mean(v),e.calculateLosses().forEach(function(R){v=y.add(v,R)}),v},c=e.collectedTrainableWeights.map(function(p){return p.read()}),h=!0,d=e.optimizer_.minimize(u,h,c);return[d].concat(l)}},t.prototype.makeTestFunction=function(){var e=this;this.testFunction=function(i){return y.tidy(function(){for(var r=[],a,s=i.slice(0,e.inputs.length),o=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),l=[],u=0;u<e.inputs.length;++u)l.push({key:e.inputs[u],value:s[u]});for(var c=new Gr(l),h=Ha(e.outputs,c),u=0;u<e.lossFunctions.length;++u){var d=e.lossFunctions[u],p=y.mean(d(o[u],h[u]));u===0?a=p:a=y.add(a,p),r.push(a)}for(var u=0;u<e.metricsTensors.length;++u){var f=e.metricsTensors[u][0],m=e.metricsTensors[u][1],g=y.mean(f(o[m],h[m]));r.push(g)}return r})}},t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){return[2,N3(this,e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,L3(this,e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:return[4,this.standardizeUserData(e,i)];case 1:r=f.sent(),a=r[0],s=r[1],o=this.makeTrainFunction(),l=o(a.concat(s)),u=[],c=0,h=l,f.label=2;case 2:return c<h.length?(d=h[c],[4,d.data()]):[3,5];case 3:p=f.sent(),u.push(p[0]),f.label=4;case 4:return c++,[3,2];case 5:return y.dispose(l),[2,Mt(u)]}})})},t.prototype.getNamedWeights=function(e){for(var i=[],r=e!=null&&e.trainableOnly,a=r?this.trainableWeights:this.weights,s=this.getWeights(r),o=0;o<a.length;++o){if(r&&!a[o].trainable)continue;i.push({name:a[o].originalName,tensor:s[o]})}return i},Object.defineProperty(t.prototype,"stopTraining",{get:function(){return this.stopTraining_},set:function(e){this.stopTraining_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.optimizer_},set:function(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)},enumerable:!0,configurable:!0}),t.prototype.dispose=function(){var e=n.prototype.dispose.call(this);if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){var i=y.memory().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=i-y.memory().numTensors}return e},t.prototype.getLossIdentifiers=function(){var e;if(typeof this.loss=="string")e=Jn(this.loss);else if(Array.isArray(this.loss)){for(var i=0,r=this.loss;i<r.length;i++){var a=r[i];if(typeof a!="string")throw new Error("Serialization of non-string loss is not supported.")}e=this.loss.map(function(h){return Jn(h)})}else{var s=Object.keys(this.loss);e={};for(var o=this.loss,l=0,u=s;l<u.length;l++){var c=u[l];if(typeof o[c]=="string")e[c]=Jn(o[c]);else throw new Error("Serialization of non-string loss is not supported.")}}return e},t.prototype.getMetricIdentifiers=function(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[Jn(go(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(function(r){return Jn(go(r))});var e={};for(var i in this.metrics)e[i]=Jn(go(this.metrics[i]));return e},t.prototype.getTrainingConfig=function(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}},t.prototype.loadTrainingConfig=function(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");var i=Ma(e.optimizer_config),r=bn(i),a;if(typeof e.loss=="string")a=Zi(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(function(l){return Zi(l)});else if(e.loss!=null){a={};for(var s in e.loss)a[s]=Zi(e.loss[s])}var o;if(Array.isArray(e.metrics))o=e.metrics.map(function(l){return Zi(l)});else if(e.metrics!=null){o={};for(var s in e.metrics)o[s]=Zi(e.metrics[s])}this.compile({loss:a,metrics:o,optimizer:r})},t.prototype.save=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b;return ve(this,function(w){switch(w.label){case 0:if(typeof e=="string"){if(r=y.io.getSaveHandlers(e),r.length===0)throw new M("Cannot find any save handlers for URL '"+e+"'");if(r.length>1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:E3,generatedBy:"TensorFlow.js tfjs-layers v"+Eh,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,dy(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){dy(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(g3);y.serialization.registerClass(yi);var D3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(yi);y.serialization.registerClass(D3);function k3(n,t){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=Ma(e),r=bn(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o<l.length;o++)u=l[o],s[u.originalName]=a[u.originalName];r.loadWeights(s),y.dispose(a),c.label=2;case 2:return[2,r]}})})}function W3(n,t){return Le(this,void 0,void 0,function(){var e;return ve(this,function(i){if(t==null&&(t={}),typeof n=="string"){if(e=y.io.getLoadHandlers(n,t),e.length===0)e.push(y.io.browserHTTPRequest(n,t));else if(e.length>1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,F3(n,void 0,t)]})})}function F3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=bn(Ma(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=U3(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function U3(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var Bh=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:ao("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r<a.length;r++){var s=a[r];i.add(s)}return i}return t.prototype.checkShape=function(e){var i=e.inboundNodes[0].outputTensors[0].shape;if(i.some(function(r){return r<0}))throw new M("Negative dimension size caused by adding layer "+(e.name+" with input shape [")+(e.inboundNodes[0].inputTensors[0].shape+"]"))},t.prototype.add=function(e){var i=e instanceof t||e instanceof yi,r;if(i){if(r=e,r.outputs.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(r.inputs.length!==1)throw new M("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new M("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");var a=Jv({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(i)this.outputs=r.outputs,this.inputs=r.inputs;else{if(e.inboundNodes.length!==1)throw new M("A layer added to a Sequential model must not already be "+("connected somewhere else. LayersModel received layer "+e.name+" ")+("which has "+e.inboundNodes.length+" pre-existing inbound ")+"connections.");if(e.inboundNodes[0].outputTensors.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Xv(this.outputs[0])}this.inboundNodes=[],new lo({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:Ji(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(function(o){return o.shape}),outputShapes:this.outputs[0].shape})}else{var s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1},t.prototype.pop=function(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{var e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}},t.prototype.call=function(e,i){return this.model==null&&this.build(),this.model.call(e,i)},t.prototype.build=function(e){if(Ke(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new yi({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0},t.prototype.countParams=function(){return this.built||this.build(),n.prototype.countParams.call(this)},t.prototype.summary=function(e,i,r){r===void 0&&(r=console.log),this.built||this.build(),n.prototype.summary.call(this,e,i,r)},t.prototype.setWeights=function(e){this.model==null&&this.build(),this.model.setWeights(e)},t.prototype.evaluate=function(e,i,r){if(r===void 0&&(r={}),!this.built)throw new Fn("The model needs to be compiled before being used.");return this.model.evaluate(e,i,r)},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.evaluateDataset(e,i)]})})},t.prototype.predict=function(e,i){return i===void 0&&(i={}),this.model==null&&this.build(),this.model.predict(e,i)},t.prototype.predictOnBatch=function(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)},t.prototype.compile=function(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames},Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.model==null?void 0:this.model.optimizer},set:function(e){this.model.optimizer=e},enumerable:!0,configurable:!0}),t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fit(e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fitDataset(e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,this.model.trainOnBatch(e,i)]})})},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s,o={};if(i instanceof Array){if(!(i[0].className!=null)||i[0].className==="Merge")throw new M("Legacy serialization format not supported yet.");s=i}else y.util.assert(i.layers!=null,function(){return"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."}),s=i.layers,delete i.layers,o=i;var l=new e(o);if(!(l instanceof t))throw new Te("Sequential.fromConfig called on non-Sequential input: "+l);for(var u=0,c=s;u<c.length;u++){var h=c[u],d=void 0,p=bn(h,d,a);a&&p.setFastWeightInitDuringBuild(!0),l.add(p)}return l},Object.defineProperty(t.prototype,"stopTraining",{get:function(){if(this.model==null)throw new M("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining},set:function(e){if(this.model==null)throw new M("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e},enumerable:!0,configurable:!0}),t.prototype.getConfig=function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i],s={};s.className=a.getClassName(),s.config=a.getConfig(),e.push(s)}return{name:this.name,layers:e}},t.className="Sequential",t}(yi);y.serialization.registerClass(Bh);function B3(n){return new yi(n)}function z3(n){return new Bh(n)}function P3(n,t){return t==null&&(t={}),W3(n,t)}function Ty(n){return Jv(n)}function _3(n,t){ry.registerCallbackConstructor(n,t)}var jt=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),M3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),uk(e,i)},t.className="elu",t}(jt);y.serialization.registerClass(M3);var H3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.selu(e)},t.className="selu",t}(jt);y.serialization.registerClass(H3);var V3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="relu",t}(jt);y.serialization.registerClass(V3);var q3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tidy(function(){return y.minimum(6,y.relu(e))})},t.className="relu6",t}(jt);y.serialization.registerClass(q3);var G3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return e},t.className="linear",t}(jt);y.serialization.registerClass(G3);var Y3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.sigmoid(e)},t.className="sigmoid",t}(jt);y.serialization.registerClass(Y3);var K3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return hk(e)},t.className="hardSigmoid",t}(jt);y.serialization.registerClass(K3);var j3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.softplus(e)},t.className="softplus",t}(jt);y.serialization.registerClass(j3);var $3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return ck(e)},t.className="softsign",t}(jt);y.serialization.registerClass($3);var X3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tanh(e)},t.className="tanh",t}(jt);y.serialization.registerClass(X3);var Ny=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.softmax(e,i)},t.className="softmax",t}(jt);y.serialization.registerClass(Ny);var J3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.logSoftmax(e,i)},t.className="logSoftmax",t}(jt);y.serialization.registerClass(J3);var Z3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),y.tidy(function(){return y.sigmoid(e.mul(i)).mul(e)})},t.className="swish",t}(jt);y.serialization.registerClass(Z3);function bi(n){return n.getClassName()}function zh(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"activation")}function wi(n){if(n==null){var t={};return t.className="linear",t.config={},zh(t)}if(typeof n=="string"){var t={};return t.className=n,t.config={},zh(t)}else return n instanceof jt?n:zh(n)}function Ph(n){if(n!=null&&typeof n!="object")throw new Error("Argument to L1L2 regularizer's constructor is expected to be an "+("object, but received: "+n))}var xy=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(y.serialization.Serializable),yo=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return Ph(e),i.l1=e==null||e.l1==null?.01:e.l1,i.l2=e==null||e.l2==null?.01:e.l2,i.hasL1=i.l1!==0,i.hasL2=i.l2!==0,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=y.zeros([1]);return i.hasL1&&(r=y.add(r,y.sum(y.mul(i.l1,y.abs(e))))),i.hasL2&&(r=y.add(r,y.sum(y.mul(i.l2,Ba(e))))),r.asScalar()})},t.prototype.getConfig=function(){return{l1:this.l1,l2:this.l2}},t.fromConfig=function(e,i){return new e({l1:i.l1,l2:i.l2})},t.className="L1L2",t}(xy);y.serialization.registerClass(yo);function Q3(n){return Ph(n),new yo({l1:n!=null?n.l1:null,l2:0})}function e4(n){return Ph(n),new yo({l2:n!=null?n.l2:null,l1:0})}var Cy={l1l2:"L1L2"};function je(n){return sh(n)}function Ry(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"regularizer")}function nt(n){if(n==null)return null;if(typeof n=="string"){var t=n in Cy?Cy[n]:n,e={className:t,config:{}};return Ry(e)}else return n instanceof xy?n:Ry(n)}var Oy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null&&(i.maxValue=e.maxValue),i}return t.prototype.call=function(e,i){e=xe(e);var r=y.relu(e);return this.maxValue!=null&&(r=y.clipByValue(r,0,this.maxValue)),r},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={maxValue:this.maxValue},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ReLU",t}(De);y.serialization.registerClass(Oy);var Ey=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_ALPHA=.3,e==null&&(e={}),i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.leakyRelu(r,this.alpha)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LeakyReLU",t}(De);y.serialization.registerClass(Ey);var Dy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),i.supportsMasking=!0,i.alphaInitializer=tt(e.alphaInitializer||i.DEFAULT_ALPHA_INITIALIZER),i.alphaRegularizer=nt(e.alphaRegularizer),i.alphaConstraint=bt(e.alphaConstraint),e.sharedAxes==null)i.sharedAxes=null;else if(Array.isArray(e.sharedAxes))i.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")i.sharedAxes=[e.sharedAxes];else throw new M("Expected sharedAxes to be a number or an array of numbers, "+("but got "+e.sharedAxes));return i}return t.prototype.build=function(e){e=Ke(e);var i=e.slice(1);if(this.sharedAxes!=null)for(var r=0,a=this.sharedAxes;r<a.length;r++){var s=a[r];i[s-1]=1}this.alpha=this.addWeight("alpha",i,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);var o={};if(this.sharedAxes!=null)for(var s=1;s<e.length;++s)o[s]=e[s];this.inputSpec=[new Tt({ndim:e.length,axes:o})],this.built=!0},t.prototype.call=function(e,i){return e=xe(e),y.prelu(e,this.alpha.read())},t.prototype.getConfig=function(){var e={alphaInitializer:st(this.alphaInitializer),alphaRegularizer:je(this.alphaRegularizer),alphaConstraint:yt(this.alphaConstraint),sharedAxes:this.sharedAxes},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="PReLU",t}(De);y.serialization.registerClass(Dy);var ky=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==i.DEFAULT_ALPHA)throw new Te("Non-default alpha value ("+e.alpha+") is not supported by the ELU layer yet.");return i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.elu(r)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ELU",t}(De);y.serialization.registerClass(ky);var Fy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_THETA=1,e==null&&(e={}),i.theta=e.theta==null?i.DEFAULT_THETA:e.theta,i}return t.prototype.call=function(e,i){var r=xe(e);return r.mul(Wa(r.greater(this.theta),"float32"))},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={theta:this.theta},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ThresholdedReLU",t}(De);y.serialization.registerClass(Fy);var Wy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_AXIS=1,e==null&&(e={}),i.softmax=new Ny().apply,i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i}return t.prototype.call=function(e,i){var r=xe(e);return this.softmax(r,this.axis)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Softmax",t}(De);y.serialization.registerClass(Wy);function Yr(n,t,e){if(typeof n=="number")return Ji(n,t);if(n.length!==t)throw new M("The "+e+" argument must be an integer or tuple of "+t+" integers."+(" Received: "+n.length+" elements."));for(var i=0;i<t;++i){var r=n[i];if(!ak(r))throw new M("The "+e+" argument must be an integer or tuple of "+t+(" integers. Received: "+JSON.stringify(n)+" including a")+(" non-integer number "+r))}return n}function wn(n,t,e,i,r){if(r===void 0&&(r=1),n==null)return n;var a=t+(t-1)*(r-1),s;return e==="same"?s=n:s=n-a+1,Math.floor((s+i-1)/i)}function bo(n,t,e,i){if(n==null)return null;if(i==="valid")n=n*t+gi([e-t,0]);else if(i==="same")n=n*t;else throw new M("Unsupport padding mode: "+i+".");return n}function _h(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,1]):n})}function Uy(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,4,1]):n})}function t4(n,t,e,i,r,a,s){return i===void 0&&(i=1),r===void 0&&(r="valid"),s===void 0&&(s=1),y.tidy(function(){if(a==null&&(a=gn()),ct(a),n.shape.length!==3)throw new M("The input of a conv1dWithBias operation should be 3, but is "+(n.shape.length+" instead."));if(t.shape.length!==3)throw new M("The kernel for a conv1dWithBias operation should be 3, but is "+(t.shape.length+" instead"));if(e!=null&&e.shape.length!==1)throw new M("The bias for a conv1dWithBias operation should be 1, but is "+(t.shape.length+" instead"));if(a==="channelsFirst"&&(n=y.transpose(n,[0,2,1])),r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");var o=y.conv1d(n,t,i,r==="same"?"same":"valid","NWC",s);return e!=null&&(o=Bn(o,e)),o})}function By(n,t,e,i,r,a,s,o){return i===void 0&&(i=[1,1]),r===void 0&&(r="valid"),o===void 0&&(o=null),y.tidy(function(){if(a==null&&(a=gn()),ct(a),n.rank!==3&&n.rank!==4)throw new M("conv2dWithBiasActivation expects input to be of rank 3 or 4, "+("but received "+n.rank+"."));if(t.rank!==3&&t.rank!==4)throw new M("conv2dWithBiasActivation expects kernel to be of rank 3 or 4, "+("but received "+n.rank+"."));var l=_h(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=y.fused.conv2d({x:l,filter:t,strides:i,pad:r==="same"?"same":"valid",dilations:s,dataFormat:"NHWC",bias:e,activation:o}),a==="channelsFirst"&&(l=y.transpose(l,[0,3,1,2])),l})}function n4(n,t,e,i,r,a,s){return i===void 0&&(i=[1,1,1]),r===void 0&&(r="valid"),y.tidy(function(){if(a==null&&(a=gn()),ct(a),n.rank!==4&&n.rank!==5)throw new M("conv3dWithBias expects input to be of rank 4 or 5, but received "+(n.rank+"."));if(t.rank!==4&&t.rank!==5)throw new M("conv3dWithBias expects kernel to be of rank 4 or 5, but received "+(n.rank+"."));var o=Uy(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=y.conv3d(o,t,i,r==="same"?"same":"valid","NDHWC",s),e!=null&&(o=Bn(o,e)),a==="channelsFirst"&&(o=y.transpose(o,[0,4,1,2,3])),o})}var zy=function(n){Q(t,n);function t(e,i){var r=n.call(this,i)||this;if(r.bias=null,r.DEFAULT_KERNEL_INITIALIZER="glorotNormal",r.DEFAULT_BIAS_INITIALIZER="zeros",t.verifyArgs(i),r.rank=e,At(r.rank,"rank"),r.rank!==1&&r.rank!==2&&r.rank!==3)throw new Te("Convolution layer for rank other than 1, 2, or 3 ("+r.rank+") is not implemented yet.");if(r.kernelSize=Yr(i.kernelSize,e,"kernelSize"),r.strides=Yr(i.strides==null?1:i.strides,e,"strides"),r.padding=i.padding==null?"valid":i.padding,nn(r.padding),r.dataFormat=i.dataFormat==null?"channelsLast":i.dataFormat,ct(r.dataFormat),r.activation=wi(i.activation),r.useBias=i.useBias==null?!0:i.useBias,r.biasInitializer=tt(i.biasInitializer||r.DEFAULT_BIAS_INITIALIZER),r.biasConstraint=bt(i.biasConstraint),r.biasRegularizer=nt(i.biasRegularizer),r.activityRegularizer=nt(i.activityRegularizer),r.dilationRate=Yr(i.dilationRate==null?1:i.dilationRate,e,"dilationRate"),r.rank===1&&Array.isArray(r.dilationRate)&&r.dilationRate.length!==1)throw new M("dilationRate must be a number or an array of a single number for 1D convolution, but received "+(""+JSON.stringify(r.dilationRate)));if(r.rank===2){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==2)throw new M("dilationRate must be a number or array of two numbers for 2D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}else if(r.rank===3){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==3)throw new M("dilationRate must be a number or array of three numbers for 3D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}return r}return t.verifyArgs=function(e){if(Wn("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!lh(e.kernelSize,"number",1,3))throw new M("BaseConv expects config.kernelSize to be number or number[] with "+("length 1, 2, or 3, but received "+JSON.stringify(e.kernelSize)+"."))},t.prototype.getConfig=function(){var e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:bi(this.activation),useBias:this.useBias,biasInitializer:st(this.biasInitializer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),wo=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;return r.kernel=null,t.verifyArgs(i),r.filters=i.filters,At(r.filters,"filters"),r.kernelInitializer=tt(i.kernelInitializer||r.DEFAULT_KERNEL_INITIALIZER),r.kernelConstraint=bt(i.kernelConstraint),r.kernelRegularizer=nt(i.kernelRegularizer),r}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=this.kernelSize.concat([a,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:(i={},i[r]=a,i)}],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a,s=r.bias==null?null:r.bias.read(),o=Lv(r.activation.getClassName());if(o!=null&&r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate,o);else{if(r.rank===1)a=t4(e,r.kernel.read(),s,r.strides[0],r.padding,r.dataFormat,r.dilationRate[0]);else if(r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else if(r.rank===3)a=n4(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else throw new Te("convolutions greater than 3D are not implemented yet.");r.activation!=null&&(a=r.activation.apply(a))}return a})},t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=[],r=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2),a=0;a<r.length;++a){var s=wn(r[a],this.kernelSize[a],this.padding,this.strides[a],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[a]);i.push(s)}var o=[e[0]];return this.dataFormat==="channelsLast"?(o=o.concat(i),o.push(this.filters)):(o.push(this.filters),o=o.concat(i)),o},t.prototype.getConfig=function(){var e={filters:this.filters,kernelInitializer:st(this.kernelInitializer),kernelRegularizer:je(this.kernelRegularizer),kernelConstraint:yt(this.kernelConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.verifyArgs=function(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new M("Convolution layer expected config.filters to be a 'number' > 0 "+("but got "+JSON.stringify(e.filters)))},t}(zy),Mh=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!lh(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(wo);y.serialization.registerClass(Mh);var Py=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(wo);y.serialization.registerClass(Py);var _y=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Tt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Tt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=bo(c,f,d,r.padding),v=bo(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Bn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=bo(i[a],u,o,this.padding),i[s]=bo(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(Mh);y.serialization.registerClass(_y);var i4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=nt(i.depthwiseRegularizer),r.depthwiseConstraint=bt(i.depthwiseConstraint),r.pointwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=nt(i.pointwiseRegularizer),r.pointwiseConstraint=bt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length<this.rank+2)throw new M("Inputs to SeparableConv"+this.rank+"D should have rank "+(this.rank+2+", but received input shape: ")+(""+JSON.stringify(e)));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null||e[r]<0)throw new M("The channel dimension of the inputs should be defined, "+("but found "+JSON.stringify(e[r])));for(var a=e[r],s=this.kernelSize.concat([a,this.depthMultiplier]),o=[],l=0;l<this.rank;++l)o.push(1);o.push(a*this.depthMultiplier,this.filters);var u=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,u,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",o,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,u,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,u,this.biasConstraint):this.bias=null,this.inputSpec=[new Tt({ndim:this.rank+2,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a;if(r.rank===1)throw new Te("1D separable convolution is not implemented yet.");return r.rank===2&&(r.dataFormat==="channelsFirst"&&(e=y.transpose(e,[0,2,3,1])),a=y.separableConv2d(e,r.depthwiseKernel.read(),r.pointwiseKernel.read(),r.strides,r.padding,r.dilationRate,"NHWC")),r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),r.dataFormat==="channelsFirst"&&(a=y.transpose(a,[0,3,1,2])),a})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=st(this.depthwiseInitializer),e.pointwiseInitializer=st(this.pointwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.pointwiseRegularizer=je(this.pointwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseConstraint),e.pointwiseConstraint=yt(this.pointwiseConstraint),e},t.className="SeparableConv",t}(wo),My=function(n){Q(t,n);function t(e){return n.call(this,2,e)||this}return t.className="SeparableConv2D",t}(i4);y.serialization.registerClass(My);var Hy=function(n){Q(t,n);function t(e){var i=n.call(this,1,e)||this;return t.verifyArgs(e),i.inputSpec=[{ndim:3}],i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.dataFormat,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!lh(e.kernelSize,"number",1,1))throw new M("Conv1D expects config.kernelSize to be number or number[] with "+("length 1, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv1D",t}(wo);y.serialization.registerClass(Hy);var Vy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return typeof e.cropping=="number"?i.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?i.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:i.cropping=e.cropping,i.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,i.inputSpec=[{ndim:4}],i}return t.prototype.computeOutputShape=function(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=xe(e),r.dataFormat==="channelsLast"){var a=no(e,r.cropping[0][0],e.shape[1]-r.cropping[0][0]-r.cropping[0][1],2);return no(a,r.cropping[1][0],e.shape[2]-r.cropping[1][1]-r.cropping[1][0],3)}else{var a=no(e,r.cropping[0][0],e.shape[2]-r.cropping[0][0]-r.cropping[0][1],3);return no(a,r.cropping[1][0],e.shape[3]-r.cropping[1][1]-r.cropping[1][0],4)}})},t.prototype.getConfig=function(){var e={cropping:this.cropping,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Cropping2D",t}(De);y.serialization.registerClass(Vy);var qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_SIZE=[2,2],i.inputSpec=[{ndim:4}],i.size=e.size==null?i.DEFAULT_SIZE:e.size,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,i}return t.prototype.computeOutputShape=function(e){if(this.dataFormat==="channelsFirst"){var i=e[2]==null?null:this.size[0]*e[2],r=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],i,r]}else{var i=e[1]==null?null:this.size[0]*e[1],r=e[2]==null?null:this.size[1]*e[2];return[e[0],i,r,e[3]]}},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e),s=a.shape;if(r.dataFormat==="channelsFirst"){a=y.transpose(a,[0,2,3,1]);var o=r.size[0]*s[2],l=r.size[1]*s[3],u=a.resizeNearestNeighbor([o,l]);return y.transpose(u,[0,3,1,2])}else{var o=r.size[0]*s[1],l=r.size[1]*s[2];return a.resizeNearestNeighbor([o,l])}})},t.prototype.getConfig=function(){var e={size:this.size,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="UpSampling2D",t}(De);y.serialization.registerClass(qy);function r4(n,t,e,i,r,a){return e===void 0&&(e=[1,1]),i===void 0&&(i="valid"),y.tidy(function(){r==null&&(r=gn()),ct(r);var s=_h(n,r);if(n.rank!==4)throw new M("Input for depthwiseConv2d is required to be 4-D, but is instead "+(n.rank+"-D"));if(t.rank!==4)throw new M("depthwiseKernel is required to be 4-D, but is instead "+(t.rank+"-D"));return s=y.depthwiseConv2d(s,t,e,i==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}var Gy=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return i.depthwiseKernel=null,i.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,i.depthwiseInitializer=tt(e.depthwiseInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.depthwiseConstraint=bt(e.depthwiseConstraint),i.depthwiseRegularizer=nt(e.depthwiseRegularizer),i}return t.prototype.build=function(e){if(e=Ke(e),e.length<4)throw new M("Inputs to DepthwiseConv2D should have rank 4. "+("Received input shape: "+JSON.stringify(e)+"."));var i=this.dataFormat==="channelsFirst"?1:3;if(e[i]==null||e[i]<0)throw new M("The channel dimension of the inputs to DepthwiseConv2D should "+("be defined, but is not ("+e[i]+")."));var r=e[i],a=[this.kernelSize[0],this.kernelSize[1],r,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[r*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=r4(e,r.depthwiseKernel.read(),r.strides,r.padding,r.dataFormat,null);return r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),a})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,s=wn(i,this.kernelSize[0],this.padding,this.strides[0]),o=wn(r,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,s,o]:[e[0],s,o,a]},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=st(this.depthwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseRegularizer),e},t.className="DepthwiseConv2D",t}(zy);y.serialization.registerClass(Gy);function Yy(n,t,e,i){if(Array.isArray(n)){if(t!=null||e!=null)throw new M("When inputs is an array, neither initialState or constants should be provided");i!=null&&(e=n.slice(n.length-i,n.length),n=n.slice(0,n.length-i)),n.length>1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Ky(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(vn(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],x=y.tidy(function(){return n(L,d)});if(r==null)h=x[0],d=x[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=x[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return x[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v<p;++v)g(v);var b;if(o){var w=1;b=y.stack(c,w)}return[h,b,d]})}var Si=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r;if(e.cell==null)throw new M("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?r=new Hh({cells:e.cell}):r=e.cell,r.stateSize==null)throw new M("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");return i.cell=r,i.returnSequences=e.returnSequences==null?!1:e.returnSequences,i.returnState=e.returnState==null?!1:e.returnState,i.goBackwards=e.goBackwards==null?!1:e.goBackwards,i._stateful=e.stateful==null?!1:e.stateful,i.unroll=e.unroll==null?!1:e.unroll,i.supportsMasking=!0,i.inputSpec=[new Tt({ndim:3})],i.stateSpec=null,i.states_=null,i.numConstants=null,i.keptStates=[],i}return t.prototype.getStates=function(){if(this.states_==null){var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return vn(0,e).map(function(i){return null})}else return this.states_},t.prototype.setStates=function(e){this.states_=e},t.prototype.computeOutputShape=function(e){Sh(e)&&(e=e[0]),e=e;var i=this.cell.stateSize;Array.isArray(i)||(i=[i]);var r=i[0],a;if(this.returnSequences?a=[e[0],e[1],r]:a=[e[0],r],this.returnState){for(var s=[],o=0,l=i;o<l.length;o++){var u=l[o];s.push([e[0],u])}return[a].concat(s)}else return a},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){Array.isArray(i)&&(i=i[0]);var a=r.returnSequences?i:null;if(r.returnState){var s=r.states.map(function(o){return null});return[a].concat(s)}else return a})},Object.defineProperty(t.prototype,"states",{get:function(){if(this.states_==null){for(var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,i=[],r=0;r<e;++r)i.push(null);return i}else return this.states_},set:function(e){this.states_=e},enumerable:!0,configurable:!0}),t.prototype.build=function(e){if(this.numConstants!=null)throw new Te("Constants support is not implemented in RNN yet.");Sh(e)&&(e=e[0]),e=e;var i=this.stateful?e[0]:null,r=e.slice(2);this.inputSpec[0]=new Tt({shape:[i,null].concat(r)});var a=[e[0]].concat(e.slice(2));this.cell.build(a);var s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!y.util.arraysEqual(this.stateSpec.map(function(o){return o.shape[o.shape.length-1]}),s))throw new M("An initialState was passed that is not compatible with "+("cell.stateSize. Received stateSpec="+this.stateSpec+"; ")+("However cell.stateSize is "+this.cell.stateSize))}else this.stateSpec=s.map(function(o){return new Tt({shape:[null,o]})});this.stateful&&this.resetStates()},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape[0];if(a==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.states_==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_=[y.zeros([a,r.cell.stateSize])];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_[0]=y.zeros([a,r.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i===!0?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var s=0;s<r.states_.length;++s){var o=e[s],l=Array.isArray(r.cell.stateSize)?r.cell.stateSize[s]:r.cell.stateSize,u=[a,l];if(!y.util.arraysEqual(o.shape,u))throw new M("State "+s+" is incompatible with layer "+r.name+": "+("expected shape="+u+", received shape="+o.shape));r.states_[s]=o}}r.states_=r.states_.map(function(c){return y.keep(c.clone())})})},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);e=s.inputs,r=s.initialState,a=s.constants;var o=[],l=[];if(r!=null){i.initialState=r,o=o.concat(r),this.stateSpec=[];for(var u=0,c=r;u<c.length;u++){var h=c[u];this.stateSpec.push(new Tt({shape:h.shape}))}l=l.concat(this.stateSpec)}a!=null&&(i.constants=a,o=o.concat(a),this.numConstants=a.length);var d=o[0]instanceof yn;if(d){var p=[e].concat(o),f=this.inputSpec.concat(l),m=this.inputSpec;this.inputSpec=f;var g=n.prototype.apply.call(this,p,i);return this.inputSpec=m,g}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;e=xe(e),o==null&&(r.stateful?o=r.states_:o=r.getInitialState(e));var l=Array.isArray(r.cell.stateSize)?r.cell.stateSize.length:1;if(o.length!==l)throw new M("RNN Layer has "+l+" state(s) but was passed "+(o.length+" initial state(s)."));r.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");var u={training:s},c=function(g,v){var b=r.cell.call([g].concat(v),u);return[b[0],b.slice(1)]},h=Ky(c,e,o,r.goBackwards,a,null,r.unroll,r.returnSequences),d=h[0],p=h[1],f=h[2];r.stateful&&r.resetStates(f,s);var m=r.returnSequences?p:d;return r.returnState?[m].concat(f):m})},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=y.zeros(e.shape);return r=y.sum(r,[1,2]),r=Ua(r),Array.isArray(i.cell.stateSize)?i.cell.stateSize.map(function(a){return a>1?ch(r,[1,a]):r}):i.cell.stateSize>1?[ch(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),Yt({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=bn(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Si);var Kr=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),Vh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:s})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(a)},rate:r.recurrentDropout,training:s}));var o,l=r.dropoutMask,u=r.recurrentDropoutMask;l!=null?o=Un(y.mul(e,l),r.kernel.read()):o=Un(e,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),u!=null&&(a=y.mul(a,u));var c=y.add(o,Un(a,r.recurrentKernel.read()));return r.activation!=null&&(c=r.activation.apply(c)),[c,c]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Yt({},e,i)},t.className="SimpleRNNCell",t}(Kr);y.serialization.registerClass(Vh);var jy=function(n){Q(t,n);function t(e){var i=this;return e.cell=new Vh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return new e(i)},t.className="SimpleRNN",t}(Si);y.serialization.registerClass(jy);var qh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new M("GRUCell does not support reset_after parameter set to true.");return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e);var i=e[e.length-1];this.kernel=this.addWeight("kernel",[i,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("GRUCell expects 2 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training==null?!1:i.training,s=e[1];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:3})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:3}));var o=r.dropoutMask,l=r.recurrentDropoutMask,u,c,h;0<r.dropout&&r.dropout<1&&(e=y.mul(e,o[0]));var d=Un(e,r.kernel.read());r.useBias&&(d=Bn(d,r.bias.read())),0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,l[0]));var p=r.recurrentKernel.read(),f=y.split(p,[2*r.units,r.units],p.rank-1),m=f[0],g=f[1],v=Un(s,m),b=y.split(d,3,d.rank-1),w=b[0],S=b[1],L=b[2],x=y.split(v,2,v.rank-1),C=x[0],R=x[1];u=r.recurrentActivation.apply(y.add(w,C)),c=r.recurrentActivation.apply(y.add(S,R));var D=Un(y.mul(c,s),g);h=r.activation.apply(y.add(L,D));var k=y.add(y.mul(u,s),y.mul(y.add(1,y.neg(u)),h));return[k,k]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Yt({},e,i)},t.className="GRUCell",t}(Kr);y.serialization.registerClass(qh);var $y=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new qh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="GRU",t}(Si);y.serialization.registerClass($y);var So=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.unitForgetBias=e.unitForgetBias,i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=[i.units,i.units],i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel=this.addWeight("kernel",[r,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);var a;if(this.useBias){if(this.unitForgetBias){var s=this.biasInitializer,o=this.units;a=new(i=function(l){Q(u,l);function u(){return l!==null&&l.apply(this,arguments)||this}return u.prototype.apply=function(c,h){var d=s.apply([o]),p=new fh().apply([o]),f=s.apply([o*2]);return Wv(Wv(d,p),f)},u}(cn),i.className="CustomInit",i)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training;if(e=e,e.length!==3)throw new M("LSTMCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var s=e[1],o=e[2];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:4})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:4}));var l=r.dropoutMask,u=r.recurrentDropoutMask,c,h,d,p;0<r.dropout&&r.dropout<1&&(e=y.mul(e,l[0]));var f=Un(e,r.kernel.read());0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,u[0])),f=y.add(f,Un(s,r.recurrentKernel.read())),r.useBias&&(f=Bn(f,r.bias.read()));var m=y.split(f,4,f.rank-1),g=m[0],v=m[1],b=m[2],w=m[3];c=r.recurrentActivation.apply(g),h=r.recurrentActivation.apply(v),d=y.add(y.mul(h,o),y.mul(c,r.activation.apply(b))),p=r.recurrentActivation.apply(w);var S=y.mul(p,r.activation.apply(d));return[S,S,d]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Yt({},e,i)},t.className="LSTMCell",t}(Kr);y.serialization.registerClass(So);var Xy=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new So(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="LSTM",t}(Si);y.serialization.registerClass(Xy);var Hh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.cells=e.cells,i}return Object.defineProperty(t.prototype,"stateSize",{get:function(){for(var e=[],i=0,r=this.cells.slice().reverse();i<r.length;i++){var a=r[i];Array.isArray(a.stateSize)?e.push.apply(e,a.stateSize):e.push(a.stateSize)}return e},enumerable:!0,configurable:!0}),t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=e;for(var a=e.slice(1),s=[],o=0,l=r.cells.slice().reverse();o<l.length;o++){var u=l[o];Array.isArray(u.stateSize)?s.push(a.splice(0,u.stateSize.length)):s.push(a.splice(0,1))}s.reverse();for(var c=[],h,d=0;d<r.cells.length;++d){var u=r.cells[d];a=s[d],d===0?h=[e[0]].concat(a):h=[h[0]].concat(a),h=u.call(h,i),c.push(h.slice(1))}a=[];for(var p=0,f=c.slice().reverse();p<f.length;p++){var m=f[p];a.push.apply(a,m)}return[h[0]].concat(a)})},t.prototype.build=function(e){Sh(e)&&(e=e[0]),e=e;var i;this.cells.forEach(function(r,a){Qi("RNNCell_"+a,function(){r.build(e),Array.isArray(r.stateSize)?i=r.stateSize[0]:i=r.stateSize,e=[e[0],i]})}),this.built=!0},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=function(s){return{className:s.getClassName(),config:s.getConfig()}},r=this.cells.map(i),a={cells:r};return Yt({},e,a)},t.fromConfig=function(e,i,r){r===void 0&&(r={});for(var a=[],s=0,o=i.cells;s<o.length;s++){var l=o[s];a.push(bn(l,r))}return new e({cells:a})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(!this.trainable)return[];for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.cells;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.weights)}return Lh(e)},t.prototype.setWeights=function(e){for(var i=[],r=0,a=this.cells;r<a.length;r++)for(var s=a[r],o=s.weights.length,l=e.splice(o),u=0;u<s.weights.length;++u)i.push([s.weights[u],l[u]]);Ih(i)},t.className="StackedRNNCells",t}(Kr);y.serialization.registerClass(Hh);function Li(n){var t=n.ones,e=n.rate,i=n.training,r=i===void 0?!1:i,a=n.count,s=a===void 0?1:a,o=function(){return Bv(t(),e)},l=function(){return za(o,t,r)};if(!s||s<=1)return y.keep(l().clone());var u=Array(s).fill(void 0).map(l);return u.map(function(c){return y.keep(c.clone())})}var LH=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(Kr),a4=function(n){Q(t,n);function t(e){var i=this;if(e.unroll)throw new Te("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Te("It is not possible at the moment to stack convolutional cells.");return i=n.call(this,e)||this,i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null),i&&i.constants)throw new M("ConvRNN2D cell does not support constants");var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.prototype.computeOutputShape=function(e){var i=this.computeSingleOutputShape(e);return this.returnSequences||(i=[i[0]].concat(i.slice(2))),this.returnState&&(i=[i].concat(Array(2).fill([e[0]].concat(i.slice(-3))))),i},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=i.cell.stateSize,a=e.shape,s=i.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=y.zeros(o);return Array.isArray(r)?Array(r.length).fill(l):[l]})},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape,s=r.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=a[0];if(l==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.getStates()==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_=[y.zeros(o)];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_[0]=y.zeros(o);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var u=0;u<r.states_.length;++u){var c=e[u],h=o;if(!y.util.arraysEqual(c.shape,h))throw new M("State "+u+" is incompatible with layer "+r.name+": "+("expected shape="+h+", received shape="+c.shape));r.states_[u]=c}}r.states_=r.states_.map(function(d){return y.keep(d.clone())})})},t.prototype.computeSingleOutputShape=function(e){var i=this.cell,r=i.dataFormat,a=i.filters,s=i.kernelSize,o=i.padding,l=i.strides,u=i.dilationRate,c=r==="channelsFirst",h=e[c?3:2],d=e[c?4:3],p=wn(h,s[0],o,l[0],u[0]),f=wn(d,s[1],o,l[1],u[1]),m=e.slice(0,2).concat(c?[a,p,f]:[p,f,a]);return m},t.className="ConvRNN2D",t}(Si),Gh=function(n){Q(t,n);function t(e){var i=this,r=e.filters,a=e.kernelSize,s=e.strides,o=e.padding,l=e.dataFormat,u=e.dilationRate;return i=n.call(this,Yt({},e,{units:r}))||this,i.filters=r,At(i.filters,"filters"),i.kernelSize=Yr(a,2,"kernelSize"),i.kernelSize.forEach(function(c){return At(c,"kernelSize")}),i.strides=Yr(s||1,2,"strides"),i.strides.forEach(function(c){return At(c,"strides")}),i.padding=o||"valid",nn(i.padding),i.dataFormat=l||"channelsLast",ct(i.dataFormat),i.dilationRate=Yr(u||1,2,"dilationRate"),i.dilationRate.forEach(function(c){return At(c,"dilationRate")}),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=4,o=this.kernelSize.concat([a,this.filters*s]);this.kernel=this.addWeight("kernel",o,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);var l=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",l,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){var u=void 0;if(this.unitForgetBias){var c=this.biasInitializer,h=this.filters;u=new(i=function(d){Q(p,d);function p(){return d!==null&&d.apply(this,arguments)||this}return p.prototype.apply=function(f,m){var g=c.apply([h]),v=y.ones([h]),b=c.apply([h*2]);return dh([g,v,b])},p}(cn),i.className="CustomInit",i)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e.length!==3)throw new M("ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training||!1,s=e[0],o=e[1],l=e[2],u=4;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.dropout,training:a,count:u}));var c=r.dropoutMask,h=function(te,re,le){return!re||!re[le]?te:y.mul(re[le],te)},d=h(s,c,0),p=h(s,c,1),f=h(s,c,2),m=h(s,c,3);0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(o)},rate:r.recurrentDropout,training:a,count:u}));var g=r.recurrentDropoutMask,v=h(o,g,0),b=h(o,g,1),w=h(o,g,2),S=h(o,g,3),L=3,x=y.split(r.kernel.read(),u,L),C=x[0],R=x[1],D=x[2],k=x[3],W=r.useBias?y.split(r.bias.read(),u):[null,null,null,null],F=W[0],P=W[1],H=W[2],_=W[3];d=r.inputConv(d,C,F,r.padding),p=r.inputConv(p,R,P,r.padding),f=r.inputConv(f,D,H,r.padding),m=r.inputConv(m,k,_,r.padding);var K=y.split(r.recurrentKernel.read(),u,L),j=K[0],q=K[1],G=K[2],Z=K[3];v=r.recurrentConv(v,j),b=r.recurrentConv(b,q),w=r.recurrentConv(w,G),S=r.recurrentConv(S,Z);var X=r.recurrentActivation.apply(y.add(d,v)),ee=r.recurrentActivation.apply(y.add(p,b)),ne=y.add(y.mul(ee,l),y.mul(X,r.activation.apply(y.add(f,w)))),ie=y.mul(r.recurrentActivation.apply(y.add(m,S)),r.activation.apply(ne));return[ie,ie,ne]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=e.units,r=VD(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Yt({},r,a)},t.prototype.inputConv=function(e,i,r,a){var s=y.conv2d(e,i,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return r?Bn(s,r,this.dataFormat):s},t.prototype.recurrentConv=function(e,i){var r=1;return y.conv2d(e,i,r,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")},t.className="ConvLSTM2DCell",t}(So);y.serialization.registerClass(Gh);var Jy=function(n){Q(t,n);function t(e){var i=this,r=new Gh(e);return i=n.call(this,Yt({},e,{cell:r}))||this,i}return t.fromConfig=function(e,i){return new e(i)},t.className="ConvLSTM2D",t}(a4);y.serialization.registerClass(Jy);var Yh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.rate=Math.max(Math.min(e.rate,1),0),i.noiseShape=e.noiseShape,i.seed=e.seed,i.supportsMasking=!0,i}return t.prototype.getNoiseShape=function(e){if(this.noiseShape==null)return this.noiseShape;for(var i=e.shape,r=[],a=0;a<this.noiseShape.length;++a)r.push(this.noiseShape[a]==null?i[a]:this.noiseShape[a]);return r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(0<r.rate&&r.rate<1){var s=i.training==null?!1:i.training,o=r.getNoiseShape(a),l=za(function(){return Bv(a,r.rate,o,r.seed)},function(){return a},s);return l}return e})},t.prototype.getConfig=function(){var e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.dispose=function(){return n.prototype.dispose.call(this)},t.className="Dropout",t}(De);y.serialization.registerClass(Yh);var Zy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[{ndim:3}],i}return t.prototype.getNoiseShape=function(e){var i=e.shape;return[i[0],1,i[2]]},t.className="SpatialDropout1D",t}(Yh);y.serialization.registerClass(Zy);var Qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.activation=null,i.useBias=!0,i.kernel=null,i.bias=null,i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){var r=null;e.batchSize!=null&&(r=e.batchSize),i.batchInputShape=[r,e.inputDim]}return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation),e.useBias!=null&&(i.useBias=e.useBias),i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelConstraint=bt(e.kernelConstraint),i.biasConstraint=bt(e.biasConstraint),i.kernelRegularizer=nt(e.kernelRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.supportsMasking=!0,i.inputSpec=[{minNDim:2}],i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[r,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:(i={},i[-1]=r,i)}],this.built=!0},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return i[i.length-1]=this.units,i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=Lv(r.activation.getClassName()),o;return s!=null?o=Un(a,r.kernel.read(),s,r.bias?r.bias.read():null):(o=Un(a,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),r.activation!=null&&(o=r.activation.apply(o))),o})},t.prototype.getConfig=function(){var e={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dense",t}(De);y.serialization.registerClass(Qy);var eb=function(n){Q(t,n);function t(e){var i=this;return e=e||{},i=n.call(this,e)||this,i.inputSpec=[{minNDim:3}],i.dataFormat=e.dataFormat,i}return t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=0,r=e.slice(1);i<r.length;i++){var a=r[i];if(a==null)throw new M('The shape of the input to "Flatten" is not fully defined '+("(got "+e.slice(1)+"). Make sure to pass a complete ")+'"input_shape" or "batch_input_shape" argument to the first layer in your model.')}return[e[0],mi(e,1)]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.dataFormat==="channelsFirst"&&a.rank>1){for(var s=[0],o=2;o<a.rank;++o)s.push(o);s.push(1),a=a.transpose(s)}return lk(a)})},t.prototype.getConfig=function(){var e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);var i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Flatten",t}(De);y.serialization.registerClass(eb);var tb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.activation=wi(e.activation),i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);return r.activation.apply(a)})},t.prototype.getConfig=function(){var e={activation:bi(this.activation)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Activation",t}(De);y.serialization.registerClass(tb);var nb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.n=e.n,i.inputSpec=[{ndim:2}],i}return t.prototype.computeOutputShape=function(e){return[e[0],this.n,e[1]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return e=xe(e),sk(e,r.n)})},t.prototype.getConfig=function(){var e={n:this.n},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="RepeatVector",t}(De);y.serialization.registerClass(nb);var ib=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;i.targetShape=e.targetShape;for(var r=0;r<i.targetShape.length;++r)i.isUnknown(i.targetShape[r])&&(i.targetShape[r]=null);return i}return t.prototype.isUnknown=function(e){return e<0||e==null},t.prototype.fixUnknownDimension=function(e,i){for(var r="Total size of new array must be unchanged.",a=i.slice(),s=1,o=null,l=0;l<a.length;++l){var u=a[l];if(this.isUnknown(u))if(o===null)o=l;else throw new M("Can only specifiy one unknown dimension.");else s*=u}var c=mi(e);if(o!==null){if(s===0||c%s!==0)throw new M(r);a[o]=c/s}else if(c!==s)throw new M(r);return a},t.prototype.computeOutputShape=function(e){for(var i=!1,r=0;r<e.length;++r)if(this.isUnknown(e[r])){i=!0;break}return i?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=a.shape,o=s.slice(0,1).concat(r.fixUnknownDimension(s.slice(1),r.targetShape));return a.reshape(o)})},t.prototype.getConfig=function(){var e={targetShape:this.targetShape},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Reshape",t}(De);y.serialization.registerClass(ib);var rb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error("Permute constructor requires `dims` to be an Array, but received "+(e.dims+" instead."));var r=vn(1,e.dims.length+1);if(!y.util.arraysEqual(e.dims.slice().sort(),r))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");return i.dims=e.dims,i.dimsIncludingBatch=[0].concat(i.dims),i.inputSpec=[new Tt({ndim:i.dims.length+1})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return this.dims.forEach(function(r,a){i[a+1]=e[r]}),i},t.prototype.call=function(e,i){return y.transpose(xe(e),this.dimsIncludingBatch)},t.prototype.getConfig=function(){var e={dims:this.dims},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Permute",t}(De);y.serialization.registerClass(rb);var ab=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null?i.maskValue=e.maskValue==null?0:e.maskValue:i.maskValue=0,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={maskValue:this.maskValue};return Object.assign(i,e),i},t.prototype.computeMask=function(e,i){var r=xe(e),a=-1;return y.any(y.notEqual(r,this.maskValue),a)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=-1,o=!0,l=y.any(y.notEqual(a,r.maskValue),s,o),u=a.mul(l.asType(a.dtype));return u})},t.className="Masking",t}(De);y.serialization.registerClass(ab);var sb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.embeddings=null,i.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){var r=null;e.batchSize!=null&&(r=e.batchSize),e.inputLength==null?i.batchInputShape=[r,null]:i.batchInputShape=[r].concat(Je(e.inputLength))}return i.inputDim=e.inputDim,At(i.inputDim,"inputDim"),i.outputDim=e.outputDim,At(i.outputDim,"outputDim"),i.embeddingsInitializer=tt(e.embeddingsInitializer||i.DEFAULT_EMBEDDINGS_INITIALIZER),i.embeddingsRegularizer=nt(e.embeddingsRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.embeddingsConstraint=bt(e.embeddingsConstraint),i.maskZero=e.maskZero,i.supportsMasking=e.maskZero,i.inputLength=e.inputLength,i}return t.prototype.build=function(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0},t.prototype.warnOnIncompatibleInputShape=function(e){},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){return r.maskZero?(e=xe(e),y.notEqual(e,y.zerosLike(e))):null})},t.prototype.computeOutputShape=function(e){if(e=Ke(e),this.inputLength==null)return e.concat([this.outputDim]);var i=Je(this.inputLength);if(i.length!==e.length-1)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));for(var r=0,a=0;a<i.length;++a){var s=i[a],o=e[a+1];if(s!=null&&o!=null&&s!==o)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));s==null&&(i[r]=o),r++}return[e[0]].concat(i,[this.outputDim])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);a.dtype!=="int32"&&(a=Wa(a,"int32"));var s=Uv(r.embeddings.read(),a.as1D());return s.reshape(Ke(r.computeOutputShape(a.shape)))})},t.prototype.getConfig=function(){var e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:st(this.embeddingsInitializer),embeddingsRegularizer:je(this.embeddingsRegularizer),activityRegularizer:je(this.activityRegularizer),embeddingsConstraint:yt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Embedding",t}(De);y.serialization.registerClass(sb);var ir=function(n){Q(t,n);function t(e){var i=n.call(this,e||{})||this;return i.supportsMasking=!0,i}return t.prototype.mergeFunction=function(e){throw new Te},t.prototype.computeElementwiseOpOutputShape=function(e,i){if(e==null||i==null)return null;if(e.length<i.length)return this.computeElementwiseOpOutputShape(i,e);if(i.length===0)return e;for(var r=e.slice(0,e.length-i.length),a=0;a<i.length;++a){var s=e[e.length-i.length+a],o=i[a];if(s==null||o==null||s<0||o<0)r.push(null);else if(s===1)r.push(o);else if(o===1)r.push(s);else{if(s!==o)throw new M("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(i));r.push(s)}}return r},t.prototype.build=function(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[Ke(e)]),e=e,e.length<2)throw new M("A merge layer should be called on an Array of at least 2 inputs."+(" Got "+e.length+" input(s)."));for(var i=[],r=0,a=e;r<a.length;r++){var s=a[r];s!=null&&s[0]!==null&&i.push(s[0])}if(i=fi(i),i.length>1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l<e.length;++l){var s=e[l]==null?null:e[l].slice(1);o=this.computeElementwiseOpOutputShape(o,s)}var u=e.map(function(c){return c.length});e.indexOf(null)===-1&&fi(u).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,r.reshapeRequired){var a=[],s=e.map(function(D){return D.rank});if(s.indexOf(null)===-1){for(var o=gi(s),l=0,u=e;l<u.length;l++){for(var c=u[l],h=c.rank,d=0;d<o-h;++d)c=Ua(c,1);a.push(c)}return r.mergeFunction(a)}else{for(var p=!1,f=0,m=e;f<m.length;f++){var c=m[f],h=c.rank;if(h==null){var g=c.shape,v=g[0],b=g.slice(1).concat([v]),w=c.reshape([v].concat(mi(g.slice(1))));w=y.transpose(w,[1,0]),w=w.reshape(b),a.push(w),p=!0}else if(h>1){var S=vn(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),x=L.rank;if(p){if(x==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(x>1){var S=[x-1].concat(vn(0,x-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r<e.length;++r){var a=e[r]==null?null:e[r].slice(1);i=this.computeElementwiseOpOutputShape(i,a)}for(var s=[],o=0,l=e;o<l.length;o++){var a=l[o];a!=null&&a[0]!==null&&s.push(a[0])}return s=fi(s),s.length===1?i=s.concat(i):i=[null].concat(i),i},t.prototype.computeMask=function(e,i){return y.tidy(function(){if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an Array");if(!Array.isArray(e))throw new M("`inputs` should be an Array");if(i.length!==e.length)throw new M("The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths "+("("+e.length+" vs "+i.length+")"));if(i.every(function(s){return s==null}))return null;i=i.map(function(s){return s==null?s:y.expandDims(s,0)});for(var r=i[0],a=1;a<i.length-1;++a)r=y.logicalAnd(r,i[a]);return r})},t}(De),ob=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return i})},t.className="Add",t}(ir);y.serialization.registerClass(ob);var lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.mul(i,e[r]);return i})},t.className="Multiply",t}(ir);y.serialization.registerClass(lb);var ub=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return y.mul(1/e.length,i)})},t.className="Average",t}(ir);y.serialization.registerClass(ub);var cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.maximum(i,e[r]);return i})},t.className="Maximum",t}(ir);y.serialization.registerClass(cb);var hb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.minimum(i,e[r]);return i})},t.className="Minimum",t}(ir);y.serialization.registerClass(hb);var db=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_AXIS=-1,e==null&&(e={}),i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new M("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;for(var i=!0,r=0,a=e;r<a.length;r++){var s=a[r];if(s!=null){i=!1;break}}if(i)return;for(var o=[],l=0;l<e.length;++l){var u=e[l].slice();u.splice(this.axis,1);for(var c=!1,h=0,d=o;h<d.length;h++){var s=d[h];if(y.util.arraysEqual(s,u)){c=!0;break}}c||o.push(u)}if(o.length>1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return dh(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s<o.length;s++){var l=o[s];if(r[a]==null||l[a]==null){r[a]=null;break}r[a]+=l[a]}return r},t.prototype.computeMask=function(e,i){var r=this;if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new M("`inputs` should be an array for Concatenate");if(i.length!==e.length)throw new M("Mismatch in the length of mask ("+i.length+") "+("and the legnth of inputs ("+e.length+")"));return y.tidy(function(){var a=!0;if(i.forEach(function(u){if(u!=null){a=!1;return}}),a)return null;for(var s=[],o=0;o<e.length;++o)i[o]==null?s.push(y.onesLike(e[o]).asType("bool")):i[o].rank<e[o].rank?s.push(y.expandDims(i[o],-1)):s.push(i[o]);var l=y.concat(s,r.axis);return y.all(l,-1,!1)})},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Concatenate",t}(ir);y.serialization.registerClass(db);function qa(n,t){for(;n<0;)n+=t;return n}function s4(n,t,e){if(n.shape.length>3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;l<s;++l)o.push(1);t=t.reshape(t.shape.concat(o))}else if(r>i){s=r-i;for(var o=[],l=0;l<s;++l)o.push(1);n=n.reshape(n.shape.concat(o))}else s=0;var u;if(n.shape.length===2&&t.shape.length===2)a[0]===a[1]?u=n.mul(t).sum(a[0]):u=n.transpose([1,0]).mul(t).sum(a[1]);else{var c=a[0]!==n.shape.length-1,h=a[1]===t.shape.length-1;u=n.matMul(t,c,h)}if(s>0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l<d+s;++l)p.push(l);u=u.squeeze(p)}return u.shape.length===1&&(u=u.expandDims(1)),u})}var pb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.axes=e.axes,i.normalize=e.normalize==null?!1:e.normalize,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0],r=e[1];if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return qa(s,e[o].shape.length)}):a=[qa(this.axes,i.shape.length),qa(this.axes,r.shape.length)],this.normalize&&(i=uo(i,a[0]),r=uo(r,a[1])),s4(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[qa(this.axes,e.length),qa(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(ir);y.serialization.registerClass(pb);var fb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return io(a.shape,0,r.stddev).add(a)},o=za(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass(fb);var mb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(io(a.shape,1,o))};return za(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(mb);var gb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=Wa(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return za(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(gb);function Ga(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function o4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=Ga(n,s,o,e,t,r);return[l,s,o]})}function l4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=vn(0,n.rank);u<c.length;u++){var h=c[u];i.indexOf(h)!==-1?l.push(1):l.push(n.shape[h])}var d=s.reshape(l),p=o.reshape(l),f=t==null?null:t.reshape(l),m=e==null?null:e.reshape(l),g=Ga(n,d,p,m,f,r);return[g,s,o]})}function u4(n,t,e,i,r){return r===void 0&&(r=.001),y.util.arraysEqual(i.slice().sort(),vn(0,n.rank-1))?o4(n,t,e,i,r):l4(n,t,e,i,r)}var vb=function(n){Q(t,n);function t(e){var i=this;return e==null&&(e={}),i=n.call(this,e)||this,i.supportsMasking=!0,i.axis=e.axis==null?-1:e.axis,i.momentum=e.momentum==null?.99:e.momentum,i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.movingMeanInitializer=tt(e.movingMeanInitializer||"zeros"),i.movingVarianceInitializer=tt(e.movingVarianceInitializer||"ones"),i.betaConstraint=bt(e.betaConstraint),i.gammaConstraint=bt(e.gammaConstraint),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.axis>=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Tt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=vn(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=Ji(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,vn(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),x=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return Ga(s,L,x,C,R,r.epsilon)}else return Ga(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=u4(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,x,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(x).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),movingMeanInitializer:st(this.movingMeanInitializer),movingVarianceInitializer:st(this.movingVarianceInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer),betaConstraint:yt(this.betaConstraint),gammaConstraint:yt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(vb);var yb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r<a.length;r++){var s=a[r];if(!Number.isInteger(s))throw new Error("Expected axis to be an array of integers, "+("but received "+JSON.stringify(i.axis)))}else throw new Error("Expected axis to be an integer or an array of integers, "+("but received "+JSON.stringify(i.axis)));return i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i.supportsMasking=!0,i}return t.prototype.build=function(e){e=Ke(e);var i=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(var r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=i);for(var a=0,s=this.axis;a<s.length;a++){var o=s[a];if(o<0||o>=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==fi(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=Ji(1,o),p=0,f=r.axis;p<f.length;p++){var m=f[p];d[m]=s[m]}for(var g=function(x){return x!=null&&x.shape.length!==o&&r.axis!==[o-1]?x.reshape(d):x},v=g(r.gamma.read()),b=g(r.beta.read()),w=[],S=[],L=0;L<o;++L)r.axis.indexOf(L)!==-1?(w.push(s[L]),S.push(1)):(w.push(1),S.push(s[L]));return c=c.tile(w),h=h.tile(w),v=v.tile(S),b=b.tile(S),Ga(a,c,h,b,v,r.epsilon)})},t.prototype.getConfig=function(){var e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LayerNormalization",t}(De);y.serialization.registerClass(yb);function c4(n,t,e){return y.tidy(function(){if(n.rank!==4)throw new M("temporalPadding expects input tensor to be 4-D, but received a "+(n.rank+"-D tensor."));if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new M("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(e==null&&(e=gn()),e!=="channelsLast"&&e!=="channelsFirst")throw new M("Unknown data format: "+e+". Supported data formats are 'channelsLast' and 'channelsFirst.");var i;return e==="channelsFirst"?i=[[0,0],[0,0],t[0],t[1]]:i=[[0,0],t[0],t[1],[0,0]],y.pad(n,i)})}var bb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.dataFormat=e.dataFormat==null?gn():e.dataFormat,e.padding==null)i.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")i.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new M("ZeroPadding2D expects padding to be a length-2 array, but "+("received a length-"+e.padding.length+" array."));var r=void 0,a=void 0;if(typeof e.padding[0]=="number")r=[e.padding[0],e.padding[0]],a=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new M("ZeroPadding2D expects height padding to be a length-2 array, "+("but received a length-"+e.padding[0].length+" array."));if(r=e.padding[0],e.padding[1].length!==2)throw new M("ZeroPadding2D expects width padding to be a length-2 array, "+("but received a length-"+e.padding[1].length+" array."));a=e.padding[1]}i.padding=[r,a]}return i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i,r;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return c4(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(bb);function Lo(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),nn(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=gn()),a==null&&(a="max"),n=_h(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function wb(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),nn(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=gn()),a==null&&(a="max"),n=Uy(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var Sb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(At(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,nn(i.padding),i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=wn(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=Ua(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(Sb);y.serialization.registerClass(Lb);var Ib=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(Sb);y.serialization.registerClass(Ib);var Ab=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),nn(i.padding),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=wn(i,this.poolSize[0],this.padding,this.strides[0]),r=wn(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Tb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(Ab);y.serialization.registerClass(Tb);var Nb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(Ab);y.serialization.registerClass(Nb);var xb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),nn(i.padding),i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=wn(i,this.poolSize[0],this.padding,this.strides[0]),r=wn(r,this.poolSize[1],this.padding,this.strides[1]),a=wn(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),wb(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(xb);y.serialization.registerClass(Cb);var Rb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),nn(a),wb(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(xb);y.serialization.registerClass(Rb);var Ob=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),Eb=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(Ob);y.serialization.registerClass(Eb);var Db=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(Ob);y.serialization.registerClass(Db);var kb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Fb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(kb);y.serialization.registerClass(Fb);var Wb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(kb);y.serialization.registerClass(Wb);var Ub=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=bn(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),Bb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=Ke(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Ky(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(Ub);y.serialization.registerClass(Bb);function h4(n){_r(nk,"BidirectionalMergeMode",n)}var d4="concat",zb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=bn(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=bn(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?d4:e.mergeMode,h4(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):Mt(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Tt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof yn,d=0,p=o;d<p.length;d++){var f=p[d];if(f instanceof yn!==h)throw new M("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors")}if(h){var m=[e].concat(o),g=this.inputSpec.concat(l),v=this.inputSpec;this.inputSpec=g;var b=n.prototype.apply.call(this,m,i);return this.inputSpec=v,b}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.initialState,s,o;if(a==null)s=r.forwardLayer.call(e,i),o=r.backwardLayer.call(e,i);else{var l=a.slice(0,a.length/2),u=a.slice(a.length/2);s=r.forwardLayer.call(e,Object.assign(i,{initialState:l})),o=r.backwardLayer.call(e,Object.assign(i,{initialState:u}))}var c;r.returnState&&(Array.isArray(s)&&(c=s.slice(1).concat(o.slice(1))),s=s[0],o=o[0]),r.returnSequences&&(o=y.reverse(o,1));var h;return r.mergeMode==="concat"?h=dh([s,o]):r.mergeMode==="sum"?h=y.add(s,o):r.mergeMode==="ave"?h=y.mul(.5,y.add(s,o)):r.mergeMode==="mul"?h=y.mul(s,o):r.mergeMode==null&&(h=[s,o]),r.returnState?r.mergeMode==null?h.concat(c):[h].concat(c):h})},t.prototype.resetStates=function(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()},t.prototype.build=function(e){var i=this;Qi(this.forwardLayer.name,function(){i.forwardLayer.build(e)}),Qi(this.backwardLayer.name,function(){i.backwardLayer.build(e)}),this.built=!0},t.prototype.computeMask=function(e,i){Array.isArray(i)&&(i=i[0]);var r;if(this.returnSequences?this.mergeMode==null?r=[i,i]:r=i:this.mergeMode==null?r=[null,null]:r=null,this.returnState){var a=this.forwardLayer.states,s=a.map(function(o){return null});return Array.isArray(r)?r.concat(s).concat(s):[r].concat(s).concat(s)}else return r},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e={mergeMode:this.mergeMode},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.fromConfig=function(e,i){var r=bn(i.layer);if(delete i.layer,i.numConstants!=null)throw new Te("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");var a=i;return a.layer=r,new e(a)},t.className="Bidirectional",t}(Ub);y.serialization.registerClass(zb);function p4(n){return new Pa(n)}function f4(n){return new ky(n)}function m4(n){return new Oy(n)}function g4(n){return new Ey(n)}function v4(n){return new Dy(n)}function y4(n){return new Wy(n)}function b4(n){return new Fy(n)}function w4(n){return new Hy(n)}function S4(n){return new Mh(n)}function L4(n){return new _y(n)}function I4(n){return new Py(n)}function A4(n){return new My(n)}function T4(n){return new Vy(n)}function N4(n){return new qy(n)}function x4(n){return new Gy(n)}function C4(n){return new tb(n)}function R4(n){return new Qy(n)}function O4(n){return new Yh(n)}function E4(n){return new Zy(n)}function D4(n){return new eb(n)}function k4(n){return new nb(n)}function F4(n){return new ib(n)}function W4(n){return new rb(n)}function U4(n){return new sb(n)}function B4(n){return new ob(n)}function z4(n){return new ub(n)}function P4(n){return new db(n)}function _4(n){return new cb(n)}function M4(n){return new hb(n)}function H4(n){return new lb(n)}function V4(n){return new pb(n)}function q4(n){return new vb(n)}function G4(n){return new yb(n)}function Y4(n){return new bb(n)}function Kh(n){return new Ib(n)}function K4(n){return Kh(n)}function j4(n){return Kh(n)}function jh(n){return new Nb(n)}function $4(n){return jh(n)}function X4(n){return jh(n)}function $h(n){return new Rb(n)}function J4(n){return $h(n)}function Z4(n){return $h(n)}function Q4(n){return new Eb(n)}function eF(n){return new Fb(n)}function Pb(n){return new Db(n)}function _b(n){return new Wb(n)}function Mb(n){return new Lb(n)}function Hb(n){return new Tb(n)}function tF(n){return new Cb(n)}function nF(n){return new $y(n)}function iF(n){return new qh(n)}function rF(n){return new Xy(n)}function aF(n){return new So(n)}function sF(n){return new jy(n)}function oF(n){return new Vh(n)}function lF(n){return new Jy(n)}function uF(n){return new Gh(n)}function cF(n){return new Si(n)}function hF(n){return new Hh(n)}function dF(n){return new zb(n)}function pF(n){return new Bb(n)}var fF=Pb,mF=_b,gF=Mb,vF=Hb;function yF(n){return new fb(n)}function bF(n){return new mb(n)}function wF(n){return new gb(n)}function SF(n){return new ab(n)}var LF={__proto__:null,inputLayer:p4,elu:f4,reLU:m4,leakyReLU:g4,prelu:v4,softmax:y4,thresholdedReLU:b4,conv1d:w4,conv2d:S4,conv2dTranspose:L4,conv3d:I4,separableConv2d:A4,cropping2D:T4,upSampling2d:N4,depthwiseConv2d:x4,activation:C4,dense:R4,dropout:O4,spatialDropout1d:E4,flatten:D4,repeatVector:k4,reshape:F4,permute:W4,embedding:U4,add:B4,average:z4,concatenate:P4,maximum:_4,minimum:M4,multiply:H4,dot:V4,batchNormalization:q4,layerNormalization:G4,zeroPadding2d:Y4,averagePooling1d:Kh,avgPool1d:K4,avgPooling1d:j4,averagePooling2d:jh,avgPool2d:$4,avgPooling2d:X4,averagePooling3d:$h,avgPool3d:J4,avgPooling3d:Z4,globalAveragePooling1d:Q4,globalAveragePooling2d:eF,globalMaxPooling1d:Pb,globalMaxPooling2d:_b,maxPooling1d:Mb,maxPooling2d:Hb,maxPooling3d:tF,gru:nF,gruCell:iF,lstm:rF,lstmCell:aF,simpleRNN:sF,simpleRNNCell:oF,convLstm2d:lF,convLstm2dCell:uF,rnn:cF,stackedRNNCells:hF,bidirectional:dF,timeDistributed:pF,globalMaxPool1d:fF,globalMaxPool2d:mF,maxPool1d:gF,maxPool2d:vF,Layer:De,RNN:Si,RNNCell:Kr,input:Ty,gaussianNoise:yF,gaussianDropout:bF,alphaDropout:wF,masking:SF};function IF(n,t){return Nh(n,t)}function AF(n,t){return ly(n,t)}function TF(n,t){return uy(n,t)}function NF(n,t){return xh(n,t)}function xF(n,t){return Ch(n,t)}function CF(n,t){return oy(n,t)}function RF(n,t){return Jk(n,t)}function OF(n,t){return Ah(n,t)}function EF(n,t){return co(n,t)}function DF(n,t){return qr(n,t)}function kF(n,t){return qr(n,t)}function FF(n,t){return qr(n,t)}function WF(n,t){return tr(n,t)}function UF(n,t){return tr(n,t)}function BF(n,t){return tr(n,t)}var zF={__proto__:null,binaryAccuracy:IF,binaryCrossentropy:AF,sparseCategoricalAccuracy:TF,categoricalAccuracy:NF,categoricalCrossentropy:xF,precision:CF,recall:RF,cosineProximity:OF,meanAbsoluteError:EF,meanAbsolutePercentageError:DF,MAPE:kF,mape:FF,meanSquaredError:WF,MSE:UF,mse:BF};var PF={__proto__:null,modelFromJSON:k3};function _F(n){return new yo(n)}function MF(n){return Q3(n)}function HF(n){return e4(n)}var VF={__proto__:null,l1l2:_F,l1:MF,l2:HF};var Vb=function(n){Q(t,n);function t(){var e=n!==null&&n.apply(this,arguments)||this;return e.model=null,e}return t.prototype.setModel=function(e){if(!(e instanceof yi))throw new Error("model must be a LayersModel, not some other Container");this.model=e},t}(Vr);function Io(n,t){return n<t}function qb(n,t){return n>t}var Gb=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=Io:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=qb:i.monitorFunc=Io,i.monitorFunc===Io&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===Io?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,vi(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(Vb);function qF(n){return new Gb(n)}var GF={earlyStopping:qF};Xe.Callback=Vb;Xe.CallbackList=ey;Xe.CustomCallback=ny;Xe.EarlyStopping=Gb;Xe.History=ty;Xe.InputSpec=Tt;Xe.LayerVariable=$v;Xe.LayersModel=yi;Xe.RNN=Si;Xe.Sequential=Bh;Xe.SymbolicTensor=yn;Xe.callbacks=GF;Xe.constraints=ZD;Xe.initializers=Dk;Xe.input=Ty;Xe.layers=LF;Xe.loadLayersModel=P3;Xe.metrics=zF;Xe.model=B3;Xe.models=PF;Xe.registerCallbackConstructor=_3;Xe.regularizers=VF;Xe.sequential=z3;Xe.version_layers=Eh});var ow=be(rr=>{"use strict";Object.defineProperty(rr,"__esModule",{value:!0});var B=Xi();var Kb=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t};function zn(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function Sn(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=r[u[0]&2?"return":u[0]?"throw":"next"])&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[0,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Ln;(function(n){n[n.DT_INVALID=0]="DT_INVALID",n[n.DT_FLOAT=1]="DT_FLOAT",n[n.DT_DOUBLE=2]="DT_DOUBLE",n[n.DT_INT32=3]="DT_INT32",n[n.DT_UINT8=4]="DT_UINT8",n[n.DT_INT16=5]="DT_INT16",n[n.DT_INT8=6]="DT_INT8",n[n.DT_STRING=7]="DT_STRING",n[n.DT_COMPLEX64=8]="DT_COMPLEX64",n[n.DT_INT64=9]="DT_INT64",n[n.DT_BOOL=10]="DT_BOOL",n[n.DT_QINT8=11]="DT_QINT8",n[n.DT_QUINT8=12]="DT_QUINT8",n[n.DT_QINT32=13]="DT_QINT32",n[n.DT_BFLOAT16=14]="DT_BFLOAT16",n[n.DT_FLOAT_REF=101]="DT_FLOAT_REF",n[n.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",n[n.DT_INT32_REF=103]="DT_INT32_REF",n[n.DT_UINT8_REF=104]="DT_UINT8_REF",n[n.DT_INT16_REF=105]="DT_INT16_REF",n[n.DT_INT8_REF=106]="DT_INT8_REF",n[n.DT_STRING_REF=107]="DT_STRING_REF",n[n.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",n[n.DT_INT64_REF=109]="DT_INT64_REF",n[n.DT_BOOL_REF=110]="DT_BOOL_REF",n[n.DT_QINT8_REF=111]="DT_QINT8_REF",n[n.DT_QUINT8_REF=112]="DT_QUINT8_REF",n[n.DT_QINT32_REF=113]="DT_QINT32_REF",n[n.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(Ln||(Ln={}));var jb;(function(n){var t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=n.CheckpointFormatVersion||(n.CheckpointFormatVersion={}))})(jb||(jb={}));var Xh={};function YF(n,t){var e={tfOpName:n,category:"custom",inputs:[],attrs:[],customExecutor:t};Xh[n]=e}function $b(n){return Xh[n]}function KF(n){delete Xh[n]}function A(n,t,e,i){var r=t.inputParams[n];if(r&&r.inputIndexStart!==void 0){var a=r.inputIndexStart,s=r.inputIndexEnd===0?void 0:r.inputIndexEnd===void 0?a+1:r.inputIndexEnd;if(r.type==="tensor")return Ht(t.inputNames[r.inputIndexStart],e,i);if(r.type==="tensors"){var o=t.inputNames.slice(a,s);return o.map(function(h){return Ht(h,e,i)})}var l=Ht(t.inputNames.slice(a)[0],e,i),u=l.dataSync();return r.type==="number"?u[0]:B.util.toNestedArray(l.shape,u)}var c=t.attrParams[n];return c&&c.value}function Ht(n,t,e){var i=$t(n),r=i[0],a=i[1],s=e.currentContextIds.find(function(o){return!!t[Ao(r,o)]});return s!==void 0?t[Ao(r,s)][a]:void 0}function jF(n,t,e){return t[Ao(n,e.currentContextId)]}function Zn(n,t){var e=$t(n),i=e[0],r=e[1];return[Ao(i,t&&t.currentContextId),r]}function Ao(n,t){return t?n+"-"+t:n}function $t(n){var t=n.split(":");if(t.length===1)return[n,0];var e=t[0];return[e,Number(t[t.length-1])]}function To(n,t,e){var i=A("pad",n,t,e);if(i==="explicit"){i=A("explicitPaddings",n,t,e);for(var r=[[0,0],[0,0],[0,0],[0,0]],a=0;a<4;a++)r[a][0]=i[a*2],r[a][1]=i[a*2+1];return r}return i}function Qn(n){return n.kept?n:B.clone(n)}var $F=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],XF={__proto__:null,json:$F};var JF=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"clip_value_min",name:"clipValueMin",type:"number"},{tfName:"clip_value_max",name:"clipValueMax",type:"number"}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"clipValueMin",name:"clipValueMin",type:"number",defaultValue:0},{tfName:"clipValueMax",name:"clipValueMax",type:"number",defaultValue:6}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],ZF={__proto__:null,json:JF};var QF=[{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],eW={__proto__:null,json:QF};var tW=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],nW={__proto__:null,json:tW};var iW=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],rW={__proto__:null,json:iW};var aW=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],sW={__proto__:null,json:aW};var oW=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],lW={__proto__:null,json:oW};var uW=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],cW={__proto__:null,json:uW};var hW=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],dW={__proto__:null,json:hW};var pW=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],fW={__proto__:null,json:pW};var mW=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],gW={__proto__:null,json:mW};var vW=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],yW={__proto__:null,json:vW};var bW=[{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],wW={__proto__:null,json:bW};var SW=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool",notSupported:!0}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],LW={__proto__:null,json:SW};var IW=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],AW={__proto__:null,json:IW};var TW=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]}],NW={__proto__:null,json:TW};var Jb=function(){function n(){var t=[XF,ZF,eW,nW,rW,sW,lW,fW,dW,cW,gW,yW,wW,LW,AW,NW],e=[].concat.apply([],t.map(function(i){return i.json}));this.opMappers=e.reduce(function(i,r){return i[r.tfOpName]=r,i},{})}return Object.defineProperty(n,"Instance",{get:function(){return this._instance||(this._instance=new this)},enumerable:!0,configurable:!0}),n.prototype.transformGraph=function(t,e){var i=this;e===void 0&&(e={});var r=t.node,a=[],s=[],o=[],l=r.reduce(function(g,v){return g[v.name]=i.mapNode(v),v.op.startsWith("Placeholder")?a.push(g[v.name]):v.op==="Const"?s.push(g[v.name]):(v.input==null||v.input.length===0)&&o.push(g[v.name]),g},{}),u=[],c=[],h={},d={};e!=null&&(h=this.mapSignatureEntries(e.inputs),d=this.mapSignatureEntries(e.outputs));var p=Object.keys(l);p.forEach(function(g){var v=l[g];v.inputNames.forEach(function(b){var w=Zn(b)[0];v.inputs.push(l[w]),l[w].children.push(v)})}),Object.keys(d).length===0?p.forEach(function(g){var v=l[g];v.children.length===0&&c.push(v)}):Object.keys(d).forEach(function(g){var v=Zn(g)[0],b=l[v];b!=null&&(b.signatureKey=d[g],c.push(b))}),Object.keys(h).length>0?Object.keys(h).forEach(function(g){var v=Zn(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=$b(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Jh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Jh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=ad(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=ad(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=Qh(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=Qh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Zh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Zh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=od(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=id(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=id(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=sd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=sd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=td(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=td(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=nd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=nd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Xb(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Xb(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=Zn(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:ed(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=Zn(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=Zn(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function xW(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function Zb(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):xW(n);return t?e:e.toLowerCase()}function Jh(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?Zb(r.s,i):e}function Zh(n,t,e){var i=n[t];return i?i.b:e}function Qh(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function ed(n){typeof n=="string"&&(n=Ln[n]);switch(n){case Ln.DT_FLOAT:return"float32";case Ln.DT_INT32:case Ln.DT_INT64:case Ln.DT_INT8:case Ln.DT_UINT8:return"int32";case Ln.DT_BOOL:return"bool";case Ln.DT_DOUBLE:return"float32";case Ln.DT_STRING:return"string";default:return null}}function Xb(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function td(n,t,e){var i=n[t];return i&&i.type?ed(i.type):e}function nd(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return ed(r)}):e}function Qb(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function id(n,t,e){var i=n[t];return i&&i.shape?Qb(i.shape):e}function rd(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function ad(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return Zb(a,i)}):e}function sd(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Qb(r)}):e}function od(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var CW=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Ht(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Ht(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return Qh(this.node.rawAttrs,t,e);if(i.s!=null)return Jh(this.node.rawAttrs,t,e);if(i.b!=null)return Zh(this.node.rawAttrs,t,e);if(i.shape!=null)return id(this.node.rawAttrs,t,e);if(i.type!=null)return td(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return rd(this.node.rawAttrs,t,e);if(i.list.s!=null)return ad(this.node.rawAttrs,t,e);if(i.list.shape!=null)return sd(this.node.rawAttrs,t,e);if(i.list.b!=null)return od(this.node.rawAttrs,t,e);if(i.list.type!=null)return nd(this.node.rawAttrs,t,e)}return e},n}();var RW=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var OW=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Ht(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function hn(n,t,e){e===void 0&&(e=""),B.util.assert(EW(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function EW(n,t){if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==-1&&t[e]!==-1&&n[e]!==t[e])return!1;return!0}var DW=function(){function n(t,e,i,r,a,s,o){this.name=t,this.dtype=e,this.maxSize=i,this.elementShape=r,this.identicalElementShapes=a,this.dynamicSize=s,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=B.scalar(0),B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"closed",{get:function(){return this.closed_},enumerable:!0,configurable:!0}),n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.read=function(t){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||t>=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`,
-          because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),hn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i<this.size();i++)t.push(i)}if(t.length===0)return B.tensor([],[0].concat(this.elementShape));var r=this.readMany(t);return hn(this.elementShape,r[0].shape,"TensorArray shape mismatch: "),B.stack(r,0)},n.prototype.concat=function(t){if(!!t&&t!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but concat requested dtype "+t);if(this.size()===0)return B.tensor([],[0].concat(this.elementShape));for(var e=[],i=0;i<this.size();i++)e.push(i);var r=this.readMany(e);return hn(this.elementShape,r[0].shape,"TensorArray shape mismatch: tensor array shape ("+this.elementShape+") vs first tensor shape ("+r[0].shape+")"),B.concat(r,0)},n.prototype.scatter=function(t,e){if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);if(t.length!==e.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+e.shape[0]);var i=Math.max.apply(Math,t);if(!this.dynamicSize&&i>=this.maxSize)throw new Error("Max index must be < array size ("+i+"  vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
+`+("2. The custom "+i+" is defined in JavaScript, ")+"but is not registered properly with tf.serialization.registerClass().");if(p!=null){for(var f={},m=0,g=Object.keys(ln);m<g.length;m++){var v=g[m];f[v]=ln[v]}for(var b=0,w=Object.keys(e);b<w.length;b++){var v=w[b];f[v]=e[v]}var S=c.config;S.customObjects=f;for(var L=Gt({},ln),x=0,C=Object.keys(e);x<C.length;x++){var v=C[x];ln[v]=e[v]}lh(c.config);var R=p(d,c.config,e,r);return ln=Gt({},L),R}else{for(var L=Gt({},ln),D=0,k=Object.keys(e);D<k.length;D++){var v=k[D];ln[v]=e[v]}var R=new d(c.config);return ln=Gt({},L),R}}}function GD(n,t){return n<t?-1:n>t?1:0}function to(n,t){return-1*GD(n,t)}function fi(n){if(n==null)return n;for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.indexOf(r)===-1&&t.push(r)}return t}function YD(n){if(n==null)throw new M("Invalid value in obj: "+JSON.stringify(n));for(var t in n)if(n.hasOwnProperty(t))return!1;return!0}function _r(n,t,e){if(e==null)return;if(n.indexOf(e)<0)throw new M(e+" is not a valid "+t+".  Valid values are "+n+" or null/undefined.")}function uh(n,t,e,i){return e===void 0&&(e=0),i===void 0&&(i=Infinity),Wn(e>=0),Wn(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function At(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return At(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(Sv(n)+".")})}function Sv(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return Sv(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function KD(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s<arguments.length;s++)a[s]=arguments[s];var o=y.util.now();return o-e<t||(e=o,i=n.apply(void 0,a)),i};return r}function Lv(n){return n==="relu"?"relu":n==="linear"?"linear":n==="elu"?"elu":null}function ch(n,t){return y.tidy(function(){return y.sqrt(y.sum(y.mul(n,n),t,!0))})}var ka=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),Iv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMaxValue=2,i.defaultAxis=0,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=ch(e,i.axis),a=y.clipByValue(r,0,i.maxValue);return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{maxValue:this.maxValue,axis:this.axis}},t.className="MaxNorm",t}(ka);y.serialization.registerClass(Iv);var Av=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultAxis=0,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){return y.div(e,y.add(vt(),ch(e,i.axis)))})},t.prototype.getConfig=function(){return{axis:this.axis}},t.className="UnitNorm",t}(ka);y.serialization.registerClass(Av);var Tv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="NonNeg",t}(ka);y.serialization.registerClass(Tv);var Nv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.defaultMinValue=0,i.defaultMaxValue=1,i.defaultRate=1,i.defaultAxis=0,i.minValue=e.minValue!=null?e.minValue:i.defaultMinValue,i.maxValue=e.maxValue!=null?e.maxValue:i.defaultMaxValue,i.rate=e.rate!=null?e.rate:i.defaultRate,i.axis=e.axis!=null?e.axis:i.defaultAxis,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=ch(e,i.axis),a=y.add(y.mul(i.rate,y.clipByValue(r,i.minValue,i.maxValue)),y.mul(1-i.rate,r));return y.mul(e,y.div(a,y.add(vt(),r)))})},t.prototype.getConfig=function(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}},t.className="MinMaxNorm",t}(ka);y.serialization.registerClass(Nv);var xv={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function yt(n){return oh(n)}function Cv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"constraint")}function bt(n){if(n==null)return null;if(typeof n=="string"){var t=n in xv?xv[n]:n,e={className:t,config:{}};return Cv(e)}else return n instanceof ka?n:Cv(n)}function jD(n){return new Iv(n)}function $D(n){return new Av(n)}function XD(){return new Tv}function JD(n){return new Nv(n)}var ZD={__proto__:null,maxNorm:jD,unitNorm:$D,nonNeg:XD,minMaxNorm:JD};var QD=["channelsFirst","channelsLast"],ek=["valid","same","causal"],tk=["max","avg"],nk=["sum","mul","concat","ave"];var Mr=new Map;function ct(n){_r(QD,"DataFormat",n)}function tn(n){_r(ek,"PaddingMode",n)}function Rv(n){_r(tk,"PoolMode",n)}var Fa=[],Ov="/";function Qi(n,t){Fa.push(n);try{var e=t();return Fa.pop(),e}catch(i){throw Fa.pop(),i}}function ik(){return Fa.length===0?"":Fa.join(Ov)+Ov}function Dv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");return ik()+n}function kv(n){if(!Ev(n))throw new Error("Not a valid tensor name: '"+n+"'");Mr.has(n)||Mr.set(n,0);var t=Mr.get(n);if(Mr.set(n,Mr.get(n)+1),t>0){var e=n+"_"+t;return Mr.set(e,1),e}else return n}var rk=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function Ev(n){return!!n.match(rk)}function ak(n){return n===parseInt(n.toString(),10)}function mi(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r<e;++r)i*=n[r];return i}function Fv(n){return n=Array.isArray(n)?new Float32Array(n):n,y.tensor1d(n)}function Hr(n){return y.min(Fv(n)).dataSync()[0]}function gi(n){return y.max(Fv(n)).dataSync()[0]}function gn(n,t){if(t<n)throw new M("end ("+t+") < begin ("+n+") is forbidden.");for(var e=[],i=n;i<t;++i)e.push(i);return e}function Wa(n,t){return n.asType(t)}function Ua(n,t){t===void 0&&(t=-1);var e=n.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),n.reshape(e)}function sk(n,t){return y.tidy(function(){if(n.shape.length!==2)throw new M("repeat() expects a rank-2 tensor, but received a "+("rank-"+n.shape.length+" tensor."));var e=Ua(n,1);return hh(e,[1,t,1])})}function ok(n){var t=[mi(n.shape)];return n.reshape(t)}function lk(n){if(n.rank<=1)throw new M("batchFlatten requires a minimum rank of 2. Got rank: "+n.rank+".");var t=[n.shape[0],mi(n.shape,1)];return n.reshape(t)}function er(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[t,0],[e,n.shape[1]]);case 3:return y.slice3d(n,[t,0,0],[e,n.shape[1],n.shape[2]]);case 4:return y.slice4d(n,[t,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3]]);case 5:return y.slice(n,[t,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4]]);case 6:return y.slice(n,[t,0,0,0,0,0],[e,n.shape[1],n.shape[2],n.shape[3],n.shape[4],n.shape[5]]);default:throw new M("sliceAlongFirstAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function dh(n,t,e){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:return y.slice2d(n,[0,t],[n.shape[0],e]);case 3:return y.slice3d(n,[0,0,t],[n.shape[0],n.shape[1],e]);case 4:return y.slice4d(n,[0,0,0,t],[n.shape[0],n.shape[1],n.shape[2],e]);default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function no(n,t,e,i){return y.tidy(function(){switch(n.rank){case 1:return y.slice1d(n,t,e);case 2:switch(i){case 1:return er(n,t,e);case 2:return dh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 3:switch(i){case 1:return er(n,t,e);case 2:return y.slice3d(n,[0,t,0],[n.shape[0],e,n.shape[2]]);case 3:return dh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}case 4:switch(i){case 1:return er(n,t,e);case 2:return y.slice4d(n,[0,t,0,0],[n.shape[0],e,n.shape[2],n.shape[3]]);case 3:return y.slice4d(n,[0,0,t,0],[n.shape[0],n.shape[1],e,n.shape[3]]);case 4:return dh(n,t,e);default:throw new M("The axis is not within the rank of the tensor "+(""+i))}default:throw new M("sliceAlongLastAxis() received an unsupported tensor rank: "+(""+n.rank))}})}function ph(n,t){t===void 0&&(t=-1);var e;return t<0&&(e=n[0].rank,e!==0?t=e:t=0),t===n[0].rank&&(t=-1),y.concat(n,t)}function Wv(n,t){switch(n.rank){case 1:return y.concat1d([n,t]);case 2:return y.concat2d([n,t],0);case 3:return y.concat3d([n,t],0);case 4:return y.concat4d([n,t],0);default:throw new M("concatAlongFirstAxis() received an unsupported "+("tensor rank: "+n.rank))}}function hh(n,t){if(Array.isArray(t)||(t=[t]),n.rank!==t.length)throw new M("The length of input n ("+t.length+") does not match "+("the number of dimensions in input x ("+n.rank+")"));return y.tile(n,t)}function io(n,t,e,i,r){return t===void 0&&(t=0),e===void 0&&(e=1),y.randomNormal(n,t,e,i,r)}function Un(n,t,e,i){if(n.rank<2||t.rank<2)throw new Te("dot requires both inputs to be rank >= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?fh(n.rank,i,mn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?fh(n.rank,i,mn()):null,activation:e}).reshape(f)}}function Uv(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function Ba(n){return y.mul(n,n)}function fh(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Bn(n,t,e){return y.tidy(function(){return e==null&&(e=mn()),ct(e),n.add(fh(n.rank,t,e))})}function uk(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function ck(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function Bv(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function hk(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function za(n,t,e){return e===void 0&&(e=!1),e?n():t()}var dk=["fanIn","fanOut","fanAvg"],pk=["normal","uniform","truncatedNormal"];function fk(n){_r(dk,"FanMode",n)}function mk(n){_r(pk,"Distribution",n)}var un=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),zv=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(un);y.serialization.registerClass(zv);var mh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(un);y.serialization.registerClass(mh);var Pv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(un);y.serialization.registerClass(Pv);var _v=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(un);y.serialization.registerClass(_v);var Mv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return io(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(un);y.serialization.registerClass(Mv);var Hv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(un);y.serialization.registerClass(Hv);var Vv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(un);y.serialization.registerClass(Vv);function gk(n,t){t===void 0&&(t="channelsLast");var e,i;if(ct(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=mi(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=mi(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=mi(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var Yt=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,fk(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,mk(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=gk(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(un);y.serialization.registerClass(Yt);var gh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="GlorotUniform",t}(Yt);y.serialization.registerClass(gh);var vh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="GlorotNormal",t}(Yt);y.serialization.registerClass(vh);var yh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="HeNormal",t}(Yt);y.serialization.registerClass(yh);var bh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="HeUniform",t}(Yt);y.serialization.registerClass(bh);var wh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="LeCunNormal",t}(Yt);y.serialization.registerClass(wh);var Sh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Yt.className},t.className="LeCunNormal",t}(Yt);y.serialization.registerClass(Sh);var qv=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=io(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(un);y.serialization.registerClass(qv);var Gv={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function Yv(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function st(n){return oh(n)}function tt(n){if(typeof n=="string"){var t=n in Gv?Gv[n]:n;if(t==="GlorotNormal")return new vh;if(t==="GlorotUniform")return new gh;if(t==="HeNormal")return new yh;if(t==="HeUniform")return new bh;if(t==="LeCunNormal")return new wh;if(t==="LeCunUniform")return new Sh;var e={};return e.className=t,e.config={},Yv(e)}else return n instanceof un?n:Yv(n)}function vk(){return new zv}function yk(){return new mh}function bk(n){return new Pv(n)}function wk(n){return new _v(n)}function Sk(n){return new Mv(n)}function Lk(n){return new Hv(n)}function Ik(n){return new Vv(n)}function Ak(n){return new Yt(n)}function Tk(n){return new gh(n)}function Nk(n){return new vh(n)}function xk(n){return new yh(n)}function Ck(n){return new bh(n)}function Rk(n){return new wh(n)}function Ok(n){return new Sh(n)}function Ek(n){return new qv(n)}var Dk={__proto__:null,zeros:vk,ones:yk,constant:bk,randomUniform:wk,randomNormal:Sk,truncatedNormal:Lk,identity:Ik,varianceScaling:Ak,glorotUniform:Tk,glorotNormal:Nk,heNormal:xk,heUniform:Ck,leCunNormal:Rk,leCunUniform:Ok,orthogonal:Ek};var kk=0;function Kv(){return kk++}var ro={};function ao(n){return n===void 0&&(n=""),n in ro||(ro[n]=0),ro[n]+=1,n+ro[n].toString()}function Lh(n){return Array.isArray(n)&&Array.isArray(n[0])}function so(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function Ke(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function oo(n){for(var t=0,e=0,i=n;e<i.length;e++){var r=i[e];r.shape.length===0?t+=1:t+=r.shape.reduce(function(a,s){return a*s})}return t}var jv="Variable",$v=function(){function n(t,e,i,r,a){e===void 0&&(e="float32"),i===void 0&&(i=jv),r===void 0&&(r=!0),a===void 0&&(a=null),this.dtype=e??"float32",this.shape=t.shape,this.id=Kv(),i=i??jv,this.originalName=Dv(i),this.name=kv(this.originalName),this.trainable_=r,this.constraint=a,this.val=y.variable(t,this.trainable_,this.name,this.dtype)}return n.prototype.read=function(){return this.assertNotDisposed(),this.val},n.prototype.write=function(t){return this.assertNotDisposed(),Fk(this.val,t),this.val.id!==t.id&&(this.val.assign(t),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this},n.prototype.dispose=function(){this.assertNotDisposed(),this.val.dispose()},n.prototype.assertNotDisposed=function(){if(this.val.isDisposed)throw new Error("LayersVariable "+this.name+" is already disposed.")},Object.defineProperty(n.prototype,"trainable",{get:function(){return this.trainable_},set:function(t){this.trainable_=t,this.val.trainable=t},enumerable:!0,configurable:!0}),n}();function Fk(n,t){if(n.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(n.shape)+" vs. "+JSON.stringify(t.shape))}function Ih(n){return n.map(function(t){return t.read()})}function Ah(n){n.forEach(function(t){var e=t[0];e.write(t[1])})}var Tt=function(){function n(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}return n}(),vn=function(){function n(t,e,i,r,a,s,o){this.dtype=t,this.shape=e,this.sourceLayer=i,this.inputs=r,this.callArgs=a,this.outputTensorIndex=o,this.id=Kv(),s!=null&&(this.originalName=Dv(s),this.name=kv(this.originalName)),this.rank=e.length}return n}(),Wk=0,lo=function(){function n(t,e){this.callArgs=e,this.id=Wk++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(var i=0,r=t.inboundLayers;i<r.length;i++){var a=r[i];a!=null&&a.outboundNodes.push(this)}t.outboundLayer.inboundNodes.push(this)}return n.prototype.getConfig=function(){for(var t=[],e=0,i=this.inboundLayers;e<i.length;e++){var r=i[e];r!=null?t.push(r.name):t.push(null)}return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}},n}(),Uk=0,De=function(n){Q(t,n);function t(e){e===void 0&&(e={});var i=n.call(this)||this;i._callHook=null,i._addedWeightNames=[],i._stateful=!1,i.id=Uk++,i.activityRegularizer=null,i.inputSpec=null,i.supportsMasking=!1,i._trainableWeights=[],i._nonTrainableWeights=[],i._losses=[],i._updates=[],i._built=!1,i.inboundNodes=[],i.outboundNodes=[];var r=e.name;if(!r){var a=i.getClassName();r=Jn(a)+"_"+ao(a)}if(i.name=r,i.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){var s=void 0;if(e.batchInputShape!=null)s=e.batchInputShape;else if(e.inputShape!=null){var o=null;e.batchSize!=null&&(o=e.batchSize),s=[o].concat(e.inputShape)}i.batchInputShape=s;var l=e.dtype;l==null&&(l=e.inputDType),l==null&&(l="float32"),i.dtype=l}return e.weights!=null?i.initialWeights=e.weights:i.initialWeights=null,i._refCount=null,i.fastWeightInitDuringBuild=!1,i}return t.nodeKey=function(e,i){return e.name+"_ib-"+i.toString()},t.prototype.getNodeAtIndex=function(e,i){if(this.inboundNodes.length===0)throw new Fn("The layer has never been called "+("and thus has no defined "+i+"."));if(this.inboundNodes.length<=e)throw new M("Asked to get "+i+" at node "+e+", "+("but the layer has only "+this.inboundNodes.length+" inbound nodes."));return this.inboundNodes[e]},t.prototype.getInputAt=function(e){return Mt(this.getNodeAtIndex(e,"input").inputTensors)},t.prototype.getOutputAt=function(e){return Mt(this.getNodeAtIndex(e,"output").outputTensors)},Object.defineProperty(t.prototype,"input",{get:function(){if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" is not connected, no input to return.");return Mt(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new pi("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new pi("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return Mt(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Je(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Je(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r<e.length;r++){var a=e[r],s=i[r];if(s==null)continue;var o=a.rank;if(s.ndim!=null&&o!==s.ndim)throw new M("Input "+r+" is incompatible with layer "+this.name+": "+("expected ndim="+s.ndim+", found ndim="+o));if(s.maxNDim!=null&&o>s.maxNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected max_ndim="+s.maxNDim+", found ndim="+o));if(s.minNDim!=null&&o<s.minNDim)throw new M("Input "+r+" is incompatible with layer "+this.name+(": expected min_ndim="+s.minNDim+", found ndim="+o+"."));if(s.dtype!=null&&a.dtype!==s.dtype)throw new M("Input "+r+" is incompatible with layer "+this.name+" "+(": expected dtype="+s.dtype+", found dtype="+a.dtype+"."));if(s.axes){var l=a.shape;for(var u in s.axes){var c=Number(u),h=s.axes[u],d=c>=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p<s.shape.length;++p){var f=s.shape[p],m=a.shape[p];if(f!=null&&m!=null&&f!==m)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected shape="+s.shape+", ")+("found shape="+a.shape+"."))}}},t.prototype.call=function(e,i){return e},t.prototype.invokeCallHook=function(e,i){this._callHook!=null&&this._callHook(e,i)},t.prototype.setCallHook=function(e){this._callHook=e},t.prototype.clearCallHook=function(){this._callHook=null},t.prototype.apply=function(e,i){var r=this;i=i||{},this.assertNotDisposed();for(var a=Je(e),s=!0,o=0,l=a;o<l.length;o++){var u=l[o];if(!(u instanceof vn)){s=!1;break}}for(var c=!0,h=0,d=a;h<d.length;h++){var u=d[h];if(u instanceof vn){c=!1;break}}if(s===c)throw new M("Arguments to apply() must be all SymbolicTensors or all Tensors");return Qi(this.name,function(){if(!r.built){r.assertInputCompatibility(e);for(var p=[],f=0,m=Je(e);f<m.length;f++){var g=m[f];p.push(g.shape)}r.build(Mt(p)),r.built=!0,r.initialWeights&&r.setWeights(r.initialWeights),r._refCount===null&&c&&(r._refCount=1)}if(r.assertInputCompatibility(e),c){for(var v=r.call(e,i),b=Je(v),w=[],S=0,L=b;S<L.length;S++){var x=L[S];a.indexOf(x)!==-1&&(x=x.clone()),w.push(x)}if(v=Mt(w),r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}else{var C=Bk(e),R=r.computeOutputShape(C),v=void 0,D=zk(e);if(r.warnOnIncompatibleInputShape(Array.isArray(e)?C[0]:C),R!=null&&R.length>0&&Array.isArray(R[0])?v=R.map(function(W,F){return new vn(D,W,r,Je(e),i,r.name,F)}):v=new vn(D,R,r,Je(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new pi("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i<r.length;i++){var a=r[i],s=JSON.stringify(a.outputShapes);e.indexOf(s)===-1&&e.push(s)}if(e.length===1){var o=this.inboundNodes[0].outputShapes;return Array.isArray(o)&&Array.isArray(o[0])&&o.length===1?o[0]:o}else throw new pi("The layer "+this.name+' has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.')},enumerable:!0,configurable:!0}),t.prototype.countParams=function(){if(!this.built)throw new Fn("You tried to call countParams() on "+this.name+", but the layer is not built yet. Build it first by calling build(batchInputShape).");return oo(this.weights)},t.prototype.build=function(e){this.built=!0},t.prototype.getWeights=function(e){return e===void 0&&(e=!1),Ih(e?this.trainableWeights:this.weights)},t.prototype.setWeights=function(e){var i=this;y.tidy(function(){var r=i.weights;if(r.length!==e.length)throw new M('You called setWeights(weights) on layer "'+i.name+'" '+("with a weight list of length "+e.length+", ")+("but the layer was expecting "+r.length+" weights. ")+("Provided weights: "+e+"..."));if(r.length===0)return;for(var a=[],s=Ih(r),o=0;o<s.length;++o){var l=s[o],u=r[o],c=e[o];if(!y.util.arraysEqual(l.shape,c.shape))throw new M("Layer weight shape "+l.shape+" "+("not compatible with provided weight shape "+c.shape));a.push([u,c])}Ah(a)})},t.prototype.addWeight=function(e,i,r,a,s,o,l){if(this._addedWeightNames.indexOf(e)!==-1)throw new M("Duplicate weight name "+e+" for layer "+this.name);this._addedWeightNames.push(e),r==null&&(r="float32"),this.fastWeightInitDuringBuild&&(a=tt("zeros"));var u=a.apply(i,r),c=new $v(u,r,e,o,l);return u.dispose(),s!=null&&this.addLoss(function(){return s.apply(c.read())}),o==null&&(o=!0),o?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c},t.prototype.setFastWeightInitDuringBuild=function(e){this.fastWeightInitDuringBuild=e},t.prototype.addLoss=function(e){var i;if(e==null||Array.isArray(e)&&e.length===0)return;e=Je(e),this._losses!==void 0&&this._losses!==null&&(i=this.losses).push.apply(i,e)},t.prototype.computeOutputShape=function(e){return e},t.prototype.computeMask=function(e,i){var r=this;if(!this.supportsMasking){if(i!=null)if(Array.isArray(i))i.forEach(function(a){if(a!=null)throw new TypeError("Layer "+r.name+" does not support masking, but was passed an inputMask.")});else throw new TypeError("Layer "+this.name+" does not support masking, but was passed an inputMask.");return null}return i},t.prototype.addInboundNode=function(e,i,r,a,s,o,l){l===void 0&&(l=null);var u=Je(e);i=Je(i),r=Je(r),a=Je(a),s=so(s),o=so(o);for(var c=[],h=[],d=[],p=0,f=u;p<f.length;p++){var m=f[p];c.push(m.sourceLayer),h.push(m.nodeIndex),d.push(m.tensorIndex)}new lo({outboundLayer:this,inboundLayers:c,nodeIndices:h,tensorIndices:d,inputTensors:u,outputTensors:i,inputMasks:r,outputMasks:a,inputShapes:s,outputShapes:o},l);for(var g=0;g<i.length;g++)i[g].sourceLayer=this,i[g].nodeIndex=this.inboundNodes.length-1,i[g].tensorIndex=g},t.prototype.getConfig=function(){var e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e},t.prototype.disposeWeights=function(){return this.weights.forEach(function(e){return e.dispose()}),this.weights.length},t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Layer '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){if(!this.built)throw new Error("Cannot dispose Layer "+this.name+" because it has not been built yet.");if(this._refCount===null)throw new Error("Cannot dispose Layer "+this.name+" because it has not been used yet.");this.assertNotDisposed();var e=0;return--this._refCount===0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}},t}(y.serialization.Serializable);function Bk(n){n=Je(n);for(var t=[],e=0,i=n;e<i.length;e++){var r=i[e];t.push(r.shape)}return Mt(t)}function zk(n){return"float32"}function Xv(n,t,e){if((t==null||e!=null&&e>0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a<i.inboundLayers.length;a++)for(var s=i.inputTensors[a],o=i.inboundLayers[a],l=i.nodeIndices[a],u=Xv(s,o,l),c=0,h=u;c<h.length;c++){var d=h[c];r.indexOf(d)===-1&&r.push(d)}return r}var Pa=function(n){Q(t,n);function t(e){var i=n.call(this,{dtype:e.dtype,name:e.name!=null?e.name:ao("input").toString()})||this;if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),i.trainable=!1,i.built=!0,i.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new M("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");var r=e.batchInputShape;if(r==null){if(e.inputShape==null)throw new M("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");r=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new M("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");var a=e.dtype||"float32";i.batchInputShape=r,i.dtype=a,i.inputSpec=[{shape:r}];var s=new vn(i.dtype,i.batchInputShape,i,[],{},i.name);return s.nodeIndex=0,s.tensorIndex=0,new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[r],outputShapes:[r]}),i}return t.prototype.apply=function(e,i){throw new M("Cannot pass any input to an "+("InputLayer's apply() method. InputLayer name: "+this.name))},t.prototype.dispose=function(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}},t.prototype.getConfig=function(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}},t.className="InputLayer",t}(De);y.serialization.registerClass(Pa);function Jv(n){if(n.batchShape==null&&n.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(n.batchShape!=null&&n.shape!=null)throw new M("Please provide either a `shape` or `batchShape` argument to Input, but not both.");var t=n.batchShape;n.shape!=null&&t==null&&(t=[null].concat(n.shape));var e=n.dtype;e==null&&(e="float32");var i=new Pa({batchInputShape:t,name:n.name,dtype:e,sparse:n.sparse}),r=i.inboundNodes[0].outputTensors;return r[0]}function vi(n){return Le(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(n==null)return[2];t=[],e=[],i=[];for(r in n)a=n[r],typeof a!="number"&&(s=a,t.push(s.data()),e.push(r),i.push(s));return t.length>0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l<o.length;++l)n[e[l]]=o[l][0];y.dispose(i),u.label=2;case 2:return[2]}})})}function Zv(n){if(n==null)return;for(var t in n){var e=n[t];typeof e!="number"&&e.dispose()}}var Qv;(function(n){n[n.SILENT=0]="SILENT",n[n.VERBOSE=1]="VERBOSE"})(Qv||(Qv={}));var Pk=125,Vr=function(){function n(){this.validationData=null}return n.prototype.setParams=function(t){this.params=t},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return[2]})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){return ve(this,function(e){return[2]})})},n.prototype.setModel=function(t){},n}(),ey=function(){function n(t,e){e===void 0&&(e=10),t==null&&(t=[]),this.callbacks=t,this.queueLength=e}return n.prototype.append=function(t){this.callbacks.push(t)},n.prototype.setParams=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setParams(t)}},n.prototype.setModel=function(t){for(var e=0,i=this.callbacks;e<i.length;e++){var r=i[e];r.setModel(t)}},n.prototype.onEpochBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onEpochEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onEpochEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchBegin=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchBegin(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onBatchEnd=function(t,e){return Le(this,void 0,void 0,function(){var i,r,a;return ve(this,function(s){switch(s.label){case 0:e==null&&(e={}),i=0,r=this.callbacks,s.label=1;case 1:return i<r.length?(a=r[i],[4,a.onBatchEnd(t,e)]):[3,4];case 2:s.sent(),s.label=3;case 3:return i++,[3,1];case 4:return[2]}})})},n.prototype.onTrainBegin=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainBegin(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n.prototype.onTrainEnd=function(t){return Le(this,void 0,void 0,function(){var e,i,r;return ve(this,function(a){switch(a.label){case 0:t==null&&(t={}),e=0,i=this.callbacks,a.label=1;case 1:return e<i.length?(r=i[e],[4,r.onTrainEnd(t)]):[3,4];case 2:a.sent(),a.label=3;case 3:return e++,[3,1];case 4:return[2]}})})},n}(),_k=function(n){Q(t,n);function t(){return n.call(this)||this}return t.prototype.onEpochBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.seen=0,this.totals={},[2]})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l=this;return ve(this,function(u){i==null&&(i={}),r=i.size==null?0:i.size,this.seen+=r,a=function(c){var h=i[c];if(typeof h=="number")s.totals.hasOwnProperty(c)||(s.totals[c]=0),s.totals[c]=s.totals[c]+h*r;else{var d=void 0;c in s.totals?d=s.totals[c]:s.totals[c]=0;var p=y.tidy(function(){return y.add(l.totals[c],y.mul(h,r))});s.totals[c]=p,d!=null&&d.dispose()}},s=this;for(o in i)a(o);return[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u=this;return ve(this,function(c){if(i!=null)for(r=function(h){if(a.totals[h]==null)return"continue";typeof a.totals[h]=="number"?i[h]=a.totals[h]/a.seen:y.tidy(function(){var d=y.mul(y.div(1,u.seen),u.totals[h]);i[h]=d,u.totals[h].dispose(),y.keep(i[h])})},a=this,s=0,o=this.params.metrics;s<o.length;s++)l=o[s],r(l);return[2]})})},t}(Vr),ty=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.epoch=[],this.history={},[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){i==null&&(i={}),this.epoch.push(e);for(r in i)this.history[r]==null&&(this.history[r]=[]),this.history[r].push(i[r]);return[2]})})},t.prototype.syncData=function(){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:e=[],i=[],r=[];for(a in this.history)for(s=this.history[a],o=0;o<s.length;++o)typeof s[o]!="number"&&(l=s[o],e.push(l.data()),i.push(a),r.push(o));return[4,Promise.all(e)];case 1:for(u=d.sent(),c=0;c<u.length;++c)h=this.history[i[c]][r[c]],h.dispose(),this.history[i[c]][r[c]]=u[c][0];return[2]}})})},t}(Vr),ny=function(n){Q(t,n);function t(e,i){var r=n.call(this)||this;if(r.currentEpoch=0,r.yieldEvery=i||"auto",r.yieldEvery==="auto"&&(r.yieldEvery=Pk),r.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");return y.util.isNumber(r.yieldEvery)&&(r.maybeWait=KD(r.maybeWait.bind(r),r.yieldEvery)),r.trainBegin=e.onTrainBegin,r.trainEnd=e.onTrainEnd,r.epochBegin=e.onEpochBegin,r.epochEnd=e.onEpochEnd,r.batchBegin=e.onBatchBegin,r.batchEnd=e.onBatchEnd,r.yield=e.onYield,r}return t.prototype.maybeWait=function(e,i,r){return Le(this,void 0,void 0,function(){var a;return ve(this,function(s){switch(s.label){case 0:return a=[],this.yield!=null?[4,vi(r)]:[3,2];case 1:s.sent(),a.push(this.yield(e,i,r)),s.label=2;case 2:return a.push(y.nextFrame()),[4,Promise.all(a)];case 3:return s.sent(),[2]}})})},t.prototype.onEpochBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.currentEpoch=e,this.epochBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.epochBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.epochEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.epochEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="epoch"&&r.push(y.nextFrame()),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onBatchBegin=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){switch(r.label){case 0:return this.batchBegin!=null?[4,vi(i)]:[3,3];case 1:return r.sent(),[4,this.batchBegin(e,i)];case 2:r.sent(),r.label=3;case 3:return[2]}})})},t.prototype.onBatchEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return r=[],this.batchEnd!=null?[4,vi(i)]:[3,2];case 1:a.sent(),r.push(this.batchEnd(e,i)),a.label=2;case 2:return this.yieldEvery==="batch"?r.push(y.nextFrame()):y.util.isNumber(this.yieldEvery)&&r.push(this.maybeWait(this.currentEpoch,e,i)),[4,Promise.all(r)];case 3:return a.sent(),[2]}})})},t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainBegin!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainBegin(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){switch(i.label){case 0:return this.trainEnd!=null?[4,vi(e)]:[3,3];case 1:return i.sent(),[4,this.trainEnd(e)];case 2:i.sent(),i.label=3;case 3:return[2]}})})},t}(Vr);function iy(n,t){if(n==null&&(n={}),n instanceof Vr)return[n];if(Array.isArray(n)&&n[0]instanceof Vr)return n;var e=Je(n);return e.map(function(i){return new ny(i,t)})}var ry=function(){function n(){}return n.registerCallbackConstructor=function(t,e){y.util.assert(t>=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function ay(n,t,e,i,r,a,s,o,l){var u=new ty,c=[new _k].concat(ry.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new ey(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function yn(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function uo(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(Ba(n),t,!0),i=y.fill(e.shape,vt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function tr(n,t){return y.tidy(function(){return y.mean(Ba(y.sub(t,n)),-1)})}function co(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function qr(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),vt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function Mk(n,t){return y.tidy(function(){var e=y.clipByValue(t,vt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,vt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(Ba(y.sub(i,a)),-1)})}function Hk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(Ba(e),-1)})}function Vk(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function qk(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function Gk(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function _a(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,vt(),1-vt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function ho(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(ok(n)).toInt();t=y.clipByValue(t,vt(),1-vt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return _a(a,t,e)})}function Yk(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function po(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,vt(),1-vt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(Yk(n,e),-1)})}function Kk(n,t){return y.tidy(function(){var e=y.clipByValue(n,vt(),1),i=y.clipByValue(t,vt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function jk(n,t){return y.tidy(function(){var e=y.log(y.add(vt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function Th(n,t){return y.tidy(function(){var e=uo(n,-1),i=uo(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var fo={meanSquaredError:tr,meanAbsoluteError:co,meanAbsolutePercentageError:qr,meanSquaredLogarithmicError:Mk,squaredHinge:Hk,hinge:Vk,categoricalHinge:qk,logcosh:Gk,categoricalCrossentropy:_a,sparseCategoricalCrossentropy:ho,binaryCrossentropy:po,kullbackLeiblerDivergence:Kk,poisson:jk,cosineProximity:Th};function Nh(n){if(typeof n=="string"){if(n in fo)return fo[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function xh(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=Wa(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function Ch(n,t){return y.tidy(function(){return Wa(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function sy(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function $k(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function Xk(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function oy(n,t){return y.tidy(function(){var e=sy(n,t),i=Xk(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Jk(n,t){return y.tidy(function(){var e=sy(n,t),i=$k(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function ly(n,t){return po(n,t)}function uy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var Zk=tr,Qk=tr,e3=co,t3=co,n3=qr,i3=qr,Rh=_a,r3=Th,cy=ho,mo={binaryAccuracy:xh,categoricalAccuracy:Ch,precision:oy,categoricalCrossentropy:Rh,sparseCategoricalCrossentropy:cy,mse:Zk,MSE:Qk,mae:e3,MAE:t3,mape:n3,MAPE:i3,cosine:r3};function a3(n){if(typeof n=="string"&&n in mo)return mo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function go(n){if(Wn(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(fo);e<i.length;e++){var r=i[e];if(fo[r]===n){t=r;break}}if(t!==void 0)return t;for(var a=0,s=Object.keys(mo);a<s.length;a++){var r=s[a];if(mo[r]===n){t=r;break}}return t!==void 0?t:n.name}function s3(n){var t={Adagrad:function(){return y.train.adagrad(.01)},Adadelta:function(){return y.train.adadelta(1,.95,vt())},Adam:function(){return y.train.adam(.001,.9,.999,vt())},Adamax:function(){return y.train.adamax(.002,.9,.999,vt(),0)},RMSProp:function(){return y.train.rmsprop(.001,.9,0,vt())},SGD:function(){return y.train.sgd(.01)}};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,n in t)return t[n]();throw new M("Unknown Optimizer "+n)}var hy=1*1024*1024;function dy(n,t,e){if(e===void 0&&(e=!1),n==null||typeof n!="object"||Object.getPrototypeOf(n)!==Object.prototype||!Oh(n))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){var i=JSON.stringify(n);i.length>hy&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+(hy+"."))}}function Oh(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e<i.length;e++){var r=i[e];if(typeof r!="string")return!1;if(!Oh(n[r]))return!1}return!0}else if(Array.isArray(n)){for(var a=0,s=n;a<s.length;a++){var o=s[a];if(!Oh(o))return!1}return!0}else return!1;else{var l=typeof n;return l==="string"||l==="number"||l==="boolean"}}function h3(n,t,e,i){i===void 0&&(i=console.log);var r=l3(n),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,e=e||[.45,.85,1]):(t=t||98,e=e||[.33,.55,.67,1]),e[e.length-1]<=1&&(e=e.map(function(d){return Math.floor(t*d)}));var s;if(!r){a.push("Receives inputs"),s=[];for(var o in n.nodesByDepth)s.push.apply(s,n.nodesByDepth[o])}i("_".repeat(t)),vo(a,e,i),i("=".repeat(t));for(var l=n.layers,u=0;u<l.length;++u)r?u3(l[u],e,i):c3(l[u],e,s,i),i((u===l.length-1?"=":"_").repeat(t));n.checkTrainableWeightsConsistency();var c=o3(n),h=oo(n.nonTrainableWeights);i("Total params: "+(c+h)),i("Trainable params: "+c),i("Non-trainable params: "+h),i("_".repeat(t))}function o3(n){var t;return n.collectedTrainableWeights!=null?t=oo(n.collectedTrainableWeights):t=oo(n.trainableWeights),t}function l3(n){var t=!0,e=[],i=[];for(var r in n.nodesByDepth)e.push(n.nodesByDepth[r]);for(var a=0,s=e;a<s.length;a++){var o=s[a];if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l<u.length;l++){for(var c=u[l],h=!1,d=0,p=c.inboundNodes;d<p.length;d++){var f=p[d];if(i.indexOf(f)!==-1)if(h){t=!1;break}else h=!0}if(!t)break}return t}function vo(n,t,e){e===void 0&&(e=console.log);for(var i="",r=0;r<n.length;++r)r>0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function u3(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];vo(s,t,e)}function c3(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s<o.length;s++){var l=o[s];if(e!=null&&e.length>0&&e.indexOf(l)===-1)continue;for(var u=0;u<l.inboundLayers.length;++u){var c=l.inboundLayers[u].name,h=l.nodeIndices[u],d=l.tensorIndices[u];a.push(c+"["+h+"]["+d+"]")}}var p=n.name,f=n.getClassName(),m=a.length===0?"":a[0],g=[p+" ("+f+")",r,n.countParams().toString(),m];vo(g,t,i);for(var u=1;u<a.length;++u)vo(["","","",a[u]],t,i)}function py(n,t,e){return(n==="inboundNodes"||n==="outputLayers"||n==="inputLayers")&&t===0&&typeof e=="string"}function Ma(n,t){if(n===null)return null;if(typeof n=="string")return Zi(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Ma(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u];if(u==="name"&&typeof c=="string")s[u]=c;else{var h=Zi(u);s[h]=Ma(c,h)}}return s}}function Eh(n,t){if(n==null)return null;if(typeof n=="string")return Jn(n);if(typeof n=="number"||typeof n=="boolean")return n;if(n instanceof Array){for(var e=[],i=n.length,r=0;r<i;++r){var a=n[r];py(t,r,a)?e.push(a):e.push(Eh(a,t))}return e}else{for(var s={},o=0,l=Object.keys(n);o<l.length;o++){var u=l[o],c=n[u],h=Jn(u);(u==="name"||u==="className")&&typeof c=="string"?s[h]=c:s[h]=Eh(c,u)}return s}}var Dh="2.6.0";function d3(n,t){if(n.dtype==null||n.dtype===t.dtype)return t;try{return y.cast(t,n.dtype)}catch(e){throw new M("The dtype of the feed ("+t.dtype+") can not be cast to the dtype "+("of the key '"+n.name+"' ("+n.dtype+")."))}}var Gr=function(){function n(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof n)for(var e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(var i=0,r=t;i<r.length;i++){var a=r[i];this.add(a.key,a.value)}}}return n.prototype.add=function(t,e,i){if(this.id2Value[t.id]==null)this.id2Value[t.id]=d3(t,e),this.name2Id[t.name]=t.id,i!=null&&(this.id2Mask[t.id]=i);else throw new M("Duplicate key: name="+t.name+", id="+t.id);return this},n.prototype.addFeed=function(t){this.add(t.key,t.value)},n.prototype.hasKey=function(t){return this.id2Value[t.id]!=null},n.prototype.names=function(){return Object.keys(this.name2Id)},n.prototype.getValue=function(t){if(t instanceof vn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Value[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Value[e]}},n.prototype.getMask=function(t){if(t instanceof vn){if(this.id2Value[t.id]==null)throw new M("Nonexistent key: "+t.name);return this.id2Mask[t.id]}else{var e=this.name2Id[t];if(e==null)throw new M("Feed dict has no SymbolicTensor name: "+t);return this.id2Mask[e]}},n.prototype.disposeMasks=function(){this.id2Mask!=null&&y.dispose(this.id2Mask)},n}(),kh={},fy={};function Ha(n,t,e,i){for(var r=e==null?!1:e.training,a=Array.isArray(n),s=a?n:[n],o=s.map(function(X){return X.name}),l=[],u=t.names(),c=0,h=o;c<h.length;c++){var d=h[c];u.indexOf(d)!==-1?l.push(t.getValue(d)):l.push(null)}i!=null&&(i.maxNumTensors=-Infinity,i.minNumTensors=Infinity);var p=o.join(",")+"|"+t.names().join(","),f,m;if(kh[p]==null){var g=p3(s,t);f=g.sorted,m=g.recipientCounts,kh[p]=f,fy[p]=m}f=kh[p],m={},r||Object.assign(m,fy[p]);for(var v=new Gr(t),b=0;b<f.length;++b){if(i!=null){var w=y.memory().numTensors;w>i.maxNumTensors&&(i.maxNumTensors=w),w<i.minNumTensors&&(i.minNumTensors=w)}var S=f[b],L=S.sourceLayer;if(L instanceof Pa)continue;for(var x=[],C=[],R=[],D=!1,k=0,W=S.inputs;k<W.length;k++){var F=W[k],P=v.getValue(F),H=v.getMask(F);x.push(P),C.push(H),H!=null&&(D=!0),r||(m[F.name]--,m[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!P.isDisposed&&F.sourceLayer.stateful!==!0&&R.push(P))}D&&(e=e||{},e.mask=C[0]);var _=Je(L.apply(x,e)),K=null;L.supportsMasking&&(K=L.computeMask(x,C));for(var j=f3(S),q=Array.isArray(j)?j:[j],G=0;G<q.length;++G){v.hasKey(q[G])||v.add(q[G],_[G],Array.isArray(K)?K[0]:K);var Z=o.indexOf(q[G].name);Z!==-1&&(l[Z]=_[G])}r||y.dispose(R)}return v.disposeMasks(),a?l:l[0]}function p3(n,t){y.util.assert(n!=null&&n.length>0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=my(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s<o.length;s++){for(var l=o[s],u=my(l,t),c=u.sorted,h=u.recipientMap,d=0,p=c;d<p.length;d++){var f=p[d];a.has(f.name)||(e.push(f),a.add(f.name))}var m=function(v){i[v]==null&&(i[v]=new Set),h[v].forEach(function(b){return i[v].add(b)})};for(var g in h)m(g)}return{sorted:e,recipientCounts:m3(i)}}function m3(n){var t={};for(var e in n)t[e]=n[e].size;return t}function my(n,t){for(var e=new Set,i=[],r={},a=0,s=t.names();a<s.length;a++){var o=s[a];e.add(o)}var l=[],u=[];for(l.push(n);l.length>0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d<p.length;d++){var f=p[d];if(r[f.name]==null&&(r[f.name]=new Set),r[f.name].add(c.name),e.has(f.name))continue;l.push(f)}}}return{sorted:i,recipientMap:r}}function f3(n){var t;if(n.sourceLayer.inboundNodes.length===1)t=n.sourceLayer.output;else{for(var e=null,i=0;i<n.sourceLayer.inboundNodes.length;++i)for(var r=0,a=n.sourceLayer.inboundNodes[i].outputTensors;r<a.length;r++){var s=a[r];if(s.id===n.id){e=i;break}}t=n.sourceLayer.getOutputAt(e)}return t}var g3=function(n){Q(t,n);function t(e){var i=n.call(this,{})||this;if(i.containerNodes=new Set,i.name=e.name,i.name==null){var r=i.getClassName().toLowerCase();i.name=ao(r)}if(i.supportsMasking=!1,i.trainable_=!0,Array.isArray(e.inputs)?i.inputs=e.inputs.slice():i.inputs=[e.inputs],Array.isArray(e.outputs)?i.outputs=e.outputs.slice():i.outputs=[e.outputs],fi(i.inputs).length!==i.inputs.length)throw new M("The list of inputs passed to the model is redundant. All inputs should only appear once. Found: "+(""+i.inputs.map(function(Ce){return Ce.name})));fi(i.outputs).length!==i.outputs.length&&console.warn("The list of outputs passed to the model is redundant. All outputs should only appear once. Found: "+(""+i.outputs.map(function(Ce){return Ce.name}))),i.inputLayers=[],i.inputLayersNodeIndices=[],i.inputLayersTensorIndices=[],i.outputLayers=[],i.outputLayersNodeIndices=[],i.outputLayersTensorIndices=[],i.layers=[],i.internalContainerRefs=[];for(var a=0,s=i.outputs;a<s.length;a++){var o=s[a],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;i.outputLayers.push(l),i.outputLayersNodeIndices.push(u),i.outputLayersTensorIndices.push(c)}for(var h=0,d=i.inputs;h<d.length;h++){var o=d[h],l=o.sourceLayer,u=o.nodeIndex,c=o.tensorIndex;Wn(u===0,"input layer has >1 nodes"),Wn(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p<i.inputLayers.length;p++){var l=i.inputLayers[p];if(!(l instanceof Pa))throw new TypeError("Input layers to a LayersModel must be InputLayer objects. "+("Received inputs: "+e.inputs+". ")+("Input "+p+" (0-based) originates ")+("from layer type "+l.getClassName()+"."));i.inputNames.push(l.name),i.feedInputShapes.push(l.batchInputShape),i.feedInputNames.push(l.name)}for(var f=0,m=i.outputLayers;f<m.length;f++){var l=m[f];i.outputNames.push(l.name)}i.internalInputShapes=i.inputs.map(function(Ce){return Ce.shape}),i.internalOutputShapes=i.outputs.map(function(Ce){return Ce.shape});for(var g={},v={},b={},w={},S={},L=[],x=function(Ce,kt,pt,it,sn,hs){(it==null||sn==null||hs==null)&&(it=Ce.sourceLayer,sn=Ce.nodeIndex,hs=Ce.tensorIndex);var Bt=it.inboundNodes[sn];if(pt.indexOf(Bt)!==-1)throw new Fn("The tensor "+Ce.name+' at layer "'+it.name+'" is part of a cycle.');if(kt.indexOf(Bt)!==-1)return;i.containerNodes.add(t.nodeKey(it,sn)),it.id in S||(S[it.id]=Object.keys(S).length),pt.indexOf(Bt)===-1&&pt.push(Bt);for(var II=Bt.inboundLayers.length,yr=0;yr<II;yr++){var AI=Bt.inputTensors[yr],TI=Bt.inboundLayers[yr],NI=Bt.nodeIndices[yr],xI=Bt.tensorIndices[yr];x(AI,kt,pt,TI,NI,xI)}for(kt.push(Bt);pt.indexOf(Bt)>=0;)pt.splice(pt.indexOf(Bt),1);L.push(Bt)},C=[],R=[],D=0,k=i.outputs;D<k.length;D++){var o=k[D];x(o,C,R)}for(var W=L.slice().reverse(),F=0,P=W;F<P.length;F++){var H=P[F];v[H.id]=H,H.id in g||(g[H.id]=0);var _=g[H.id],K=b[H.outboundLayer.id]==null?0:b[H.outboundLayer.id];_=Math.max(_,K),b[H.outboundLayer.id]=_,w[H.outboundLayer.id]=H.outboundLayer,g[H.id]=_;for(var p=0;p<H.inboundLayers.length;p++){var j=H.inboundLayers[p],u=H.nodeIndices[p],q=j.inboundNodes[u],G=g[q.id]==null?0:g[q.id];g[q.id]=Math.max(_+1,G),v[q.id]=q}}var Z={};for(var X in g){var _=g[X];_ in Z||(Z[_]=[]),Z[_].push(v[X])}var ee={};for(var ne in b){var _=b[ne];_ in ee||(ee[_]=[]),ee[_].push(w[ne])}var ie=Object.keys(ee).map(function(Ce){return parseInt(Ce,10)}).sort(to);i.layers=[];for(var te=0,re=ie;te<re.length;te++){var _=re[te],le=ee[_];le.sort(function(kt,pt){var it=S[kt.id],sn=S[pt.id];return it<sn?-1:it>sn?1:0});for(var he=0,ye=le;he<ye.length;he++){var l=ye[he];l instanceof t&&i.internalContainerRefs.push(l),i.layers.push(l)}}i.layersByDepth=ee,ie=Object.keys(Z).map(function(Ce){return parseInt(Ce,10)}).sort(to);for(var Oe=i.inputs.slice(),Fe=[],_e=0,Me=ie;_e<Me.length;_e++)for(var _=Me[_e],ht=0,St=Z[_];ht<St.length;ht++){var H=St[ht],l=H.outboundLayer;if(l!=null){for(var Ge=0,Ue=H.inputTensors;Ge<Ue.length;Ge++){var o=Ue[Ge];if(Oe.indexOf(o)===-1)throw new Fn("Graph disconnected: cannot obtain value for tensor "+o+(' at layer "'+l.name+'". ')+"The following previous layers were accessed without "+("issue: "+Fe))}for(var me=0,Dt=H.outputTensors;me<Dt.length;me++){var o=Dt[me];Oe.push(o)}Fe.push(l.name)}}i.nodesByDepth=Z;for(var lt=i.layers.map(function(Ce){return Ce.name}),dt=function(Ce){var kt=lt.filter(function(pt){return pt===Ce}).length;if(kt!==1)throw new Fn('The name "'+Ce+'" is used '+kt+" times in the model. All layer names should be unique. Layer names: "+JSON.stringify(lt))},xn=0,Rt=lt;xn<Rt.length;xn++){var Lt=Rt[xn];dt(Lt)}return i.outboundNodes=[],i.inboundNodes=[],new lo({outboundLayer:i,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:i.inputs,outputTensors:i.outputs,inputMasks:i.inputs.map(function(Ce){return null}),outputMasks:i.outputs.map(function(Ce){return null}),inputShapes:i.inputs.map(function(Ce){return Ce.shape}),outputShapes:i.outputs.map(function(Ce){return Ce.shape})}),i.built=!0,i._refCount=1,i}return t.prototype.assertNotDisposed=function(){if(this._refCount===0)throw new Error("Container '"+this.name+"' is already disposed.")},t.prototype.dispose=function(){this.assertNotDisposed();var e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(var i=0,r=this.layers;i<r.length;i++){var a=r[i];e.numDisposedVariables+=a.dispose().numDisposedVariables}for(var s=0,o=this.internalContainerRefs;s<o.length;s++){var l=o[s];e.numDisposedVariables+=l.dispose().numDisposedVariables}}return e.refCountAfterDispose=this._refCount,e},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this.layers.forEach(function(i){i._trainableWeights.forEach(function(r){return r.trainable=e})}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(this._trainableWeights.length>0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e=e.concat(a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.layers;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.loadWeights=function(e,i){i===void 0&&(i=!0);for(var r={},a=0,s=0,o=this.layers;s<o.length;s++)for(var l=o[s],u=0,c=l.weights;u<c.length;u++){var h=c[u];if(r[h.originalName]!=null)throw new M("Duplicate weight name: "+h.originalName);r[h.originalName]=h,a++}var d=[];for(var p in e){var f=p;if(r[p]==null){var m=p.split("/"),g=m.slice(0,-2).concat([m[m.length-1]]);f=g.join("/")}if(r[f]!=null)d.push([r[f],e[p]]);else if(i)throw new M("Provided weight data has no target variable: "+p);delete r[f]}if(i){var v=[];for(var b in r)v.push(b);if(v.length>0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}Ah(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+Dh,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=Eh(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Je(e);for(var a=new Gr,s=0;s<r.inputs.length;++s)a.add(r.inputs[s],e[s]);return Ha(r.outputs,a,i)})},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){e=Je(e);var a;return i==null?a=Ji(null,e.length):a=Je(i),r.runInternalGraph(e,a)[1]})},t.prototype.computeOutputShape=function(e){var i=so(e);if(i.length!==this.inputLayers.length)throw new M("Invalid inputShape argument "+e+": "+("model has "+this.inputLayers.length+" tensor inputs."));for(var r={},a=0;a<i.length;a++){var s=this.inputLayers[a],o=i[a],l=s.name+"_0_0";r[l]=o}var u=Object.keys(this.nodesByDepth).map(function(P){return parseInt(P,10)}).sort(to);if(u.length>1)for(var c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){var g=m[f],s=g.outboundLayer;if(this.inputLayers.map(function(H){return H.id}).indexOf(s.id)!==-1)continue;for(var v=[],b=0;b<g.inboundLayers.length;b++){var w=g.inboundLayers[b],S=g.nodeIndices[b],L=g.tensorIndices[b],l=w.name+"_"+S+"_"+L,x=r[l];v.push(x)}for(var C=s.computeOutputShape(Mt(v)),R=so(C),D=s.inboundNodes.indexOf(g),b=0;b<R.length;b++){var l=s.name+"_"+D+"_"+b;r[l]=R[b]}}for(var k=[],W=[],a=0;a<this.outputLayers.length;a++){var s=this.outputLayers[a],D=this.outputLayersNodeIndices[a],L=this.outputLayersTensorIndices[a],l=s.name+"_"+D+"_"+L;W.push(l)}for(var a=0;a<W.length;a++){var F=W[a];Wn(F in r),k.push(r[F])}return Mt(k)},t.prototype.runInternalGraph=function(e,i){i==null&&(i=Ji(null,e.length));for(var r={},a=0;a<this.inputs.length;++a){var s=this.inputs[a],o=e[a],l=i[a];r[s.id]=[o,l]}for(var u=Object.keys(this.nodesByDepth).map(function(ee){return parseInt(ee,10)}).sort(to),c=0,h=u;c<h.length;c++)for(var d=h[c],p=this.nodesByDepth[d],f=0,m=p;f<m.length;f++){for(var g=m[f],v=g.outboundLayer,b=g.inputTensors,w=g.outputTensors,S=new Array,L=0,x=b;L<x.length;L++){var s=x[L];s.id in r&&S.push(r[s.id])}if(S.length===b.length){var C={},R=void 0,D=void 0,k=void 0,W=void 0;if(g.callArgs!=null&&(C=g.callArgs),S.length===1){var F=S[0],P=F[0],H=F[1];C.mask==null&&(C.mask=H),k=Je(v.call(P,C)),W=Je(v.computeMask(P,H)),R=[P],D=[H]}else R=S.map(function(ee){return ee[0]}),D=S.map(function(ee){return ee[1]}),C.mask==null&&(C.mask=D),k=Je(v.call(R,C)),W=Je(v.computeMask(R,D));if(v.activityRegularizer)throw new Te("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(var a=0;a<w.length;++a){var s=w[a],o=k[a],l=W[a];r[s.id]=[o,l]}}}for(var _=[],K=[],j=[],q=0,G=this.outputs;q<G.length;q++){var s=G[q];Wn(s.id in r,"Could not compute output "+s.name+" : "+s.id);var Z=r[s.id],X=Z[0],l=Z[1];j.push(X.shape),_.push(X),K.push(l)}return[_,K,j]},t.prototype.buildNodeConversionMap=function(e){for(var i={},r,a=0,s=this.layers;a<s.length;a++){var o=s[a];r=o instanceof t?1:0;for(var l=0;l<o.inboundNodes.length;l++){var u=t.nodeKey(o,l);this.containerNodes.has(u)&&(i[u]=r,r+=1)}}return i},t.prototype.getLayer=function(e,i){if(i!=null){if(this.layers.length<=i)throw new M("Was asked to retrieve layer at index "+i+", but model only "+("has "+this.layers.length+" layer(s)."));return this.layers[i]}else if(e==null)throw new M("Provide either a layer name or layer index");for(var r=0,a=this.layers;r<a.length;r++){var s=a[r];if(s.name===e)return s}throw new M("No such layer: "+e)},t.prototype.calculateLosses=function(){var e=this;return y.tidy(function(){for(var i=[],r=0,a=e.layers;r<a.length;r++)for(var s=a[r],o=0;o<s.inboundNodes.length;++o){var l=t.nodeKey(s,o);e.containerNodes.has(l)&&i.push.apply(i,s.calculateLosses())}return i})},t.prototype.getConfig=function(){for(var e={name:this.name},i=this.buildNodeConversionMap(this.layers),r=[],a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=o.getClassName(),u=o.getConfig(),c=[],h=0;h<o.inboundNodes.length;h++){var d=o.inboundNodes[h],p=t.nodeKey(o,h),f={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),f=d.callArgs}catch(D){console.warn("Layer "+o.name+" was passed non-serializable keyword arguments: "+(d.callArgs+". They will not be included ")+"in the serialized model (and thus will be missing at deserialization time)."),f={}}if(d.inboundLayers.length>0){for(var m=[],g=0;g<d.inboundLayers.length;g++){var v=d.inboundLayers[g],b=d.nodeIndices[g],w=d.tensorIndices[g],S=t.nodeKey(v,b),L=i[S];L==null&&(L=0),m.push([v.name,L,w,f])}c.push(m)}}}var x={};x.name=o.name,x.className=l,x.config=u,x.inboundNodes=c,r.push(x)}e.layers=r;for(var C=[],g=0;g<this.inputLayers.length;g++){var o=this.inputLayers[g],b=this.inputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.inputLayersTensorIndices[g];C.push([o.name,L,w])}e.inputLayers=C;for(var R=[],g=0;g<this.outputLayers.length;g++){var o=this.outputLayers[g],b=this.outputLayersNodeIndices[g],p=t.nodeKey(o,b);if(!this.containerNodes.has(p))continue;var L=i[p];L==null&&(L=0);var w=this.outputLayersTensorIndices[g];R.push([o.name,L,w])}return e.outputLayers=R,e},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s={},o={};function l(G,Z){G.name in o?o[G.name].push(Z):o[G.name]=[Z]}function u(G,Z){for(var X=[],ee,ne=0,ie=Z;ne<ie.length;ne++){var te=ie[ne],re=te[0],le=te[1],he=te[2];if(ee=te[3]==null?{}:te[3],!(re in s)){l(G,Z);return}var ye=s[re];if(ye.inboundNodes.length<=le){l(G,Z);return}var Oe=ye.inboundNodes[le];X.push(Oe.outputTensors[he])}X.length>0&&G.apply(Mt(X),ee)}function c(G){var Z=G.name,X=yn(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p<f.length;p++){var m=f[p];c(m)}for(;!YD(o);)for(var g=0,v=d;g<v.length;g++){var m=v[g],b=s[m.name];if(b.name in o){var w=o[b.name];delete o[b.name];for(var S=0,L=w;S<L.length;S++){var x=L[S];u(b,x)}}}for(var C=[],R=[],D=i.inputLayers,k=0,W=D;k<W.length;k++){var m=W[k],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;C.push(_[H])}for(var K=i.outputLayers,j=0,q=K;j<q.length;j++){var m=q[j],F=m[0],P=m[1],H=m[2];Wn(F in s);var b=s[F],_=b.inboundNodes[P].outputTensors;R.push(_[H])}return new e({inputs:C,outputs:R,name:h})},Object.defineProperty(t.prototype,"stateful",{get:function(){if(this._stateful)throw new M("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(var e=0,i=this.layers;e<i.length;e++){var r=i[e];if(r.stateful)return!0}return!1},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){var e=this;y.tidy(function(){e.layers.forEach(function(i){i.stateful&&i.resetStates()})})},t}(De);function v3(n,t,e){var i=t.length;if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(a){return null});if(i===1)return Array.isArray(n)&&n.length===1?n:typeof n=="object"&&t[0]in n?[n[t[0]]]:[n];if(Array.isArray(n)){if(n.length!==i)throw new Error("Provided "+e+" is an array of "+n.length+" "+("element(s), but the model has "+i+" outputs. ")+"Make sure a set of weights is provided for each model output.");return n}else if(typeof n=="object"&&Object.keys(n).length>0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function gy(n,t){return v3(n,t,"classWeight")}function vy(n,t,e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function y3(n,t){return y.mul(n,t)}var b3=32;function by(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor.  The provided Dataset instead generates "+(""+t)});var a=yy("input",n.inputNames,e),s=yy("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs.  (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs.  (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},u=0;u<a.length;u++)l(u);for(var c=function(d){y.util.assert(s[d].shape[0]===o,function(){return"Batch size mismatch: output "+(n.outputNames[d]+" has "+s[d].shape[0]+"; ")+("expected  "+o+" based on input "+n.inputNames[0]+".")})},h=0;h<s.length;h++)c(h);return{xs:a,ys:s}}function yy(n,t,e){if(e instanceof y.Tensor)return[e];if(Array.isArray(e))return y.util.assert(e.length===t.length,function(){return"Received an array of "+e.length+" Tensors, but expected "+t.length+" to match the "+n+" keys "+t+"."}),e;for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r];if(e[s]==null)throw new M("The feature data generated by the dataset lacks the required "+(n+" key '"+s+"'."));i.push(e[s])}return i}function w3(n){if(n.length===3)throw new Te("Validation with sample weights is not implemented yet.");return{xs:n[0],ys:n[1]}}function L3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k,W,F,P,H,_,K,F,j,q,G,Z,F;return ve(this,function(X){switch(X.label){case 0:if(i=e.batchesPerEpoch!=null,y.util.assert(n.optimizer!=null,function(){return"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."}),y.util.assert(e!=null,function(){return"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."}),y.util.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(wy(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=w3(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=iy(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=ay(h,d,e.epochs,null,null,S3(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g<e.epochs?(b={},[4,f.onEpochBegin(g)]):[3,23];case 5:return X.sent(),w=0,S=0,i?[3,7]:[4,t.iterator()];case 6:v=X.sent(),X.label=7;case 7:return(i?w<e.batchesPerEpoch:!0)?[4,v.next()]:[3,21];case 8:return L=X.sent(),i&&L.done?(console.warn("You provided `batchesPerEpoch` as "+(e.batchesPerEpoch+", ")+"but your dataset iterator ran out of data after "+(w+" batches; ")+"interrupting training. Make sure that your dataset can generate at least `batchesPerEpoch * epochs` batches (in this case, "+(e.batchesPerEpoch*e.epochs+" batches). ")+"You may need to use the repeat() function when building your dataset."),[3,21]):L.value!=null?(x=by(n,L.value),C=x.xs,R=x.ys,D={},D.batch=S,D.size=C[0].shape[0],[4,f.onBatchBegin(S,D)]):[3,15];case 9:if(X.sent(),k=[],!(e.classWeight!=null))return[3,13];W=gy(e.classWeight,n.outputNames),F=0,X.label=10;case 10:return F<W.length?(H=(P=k).push,[4,vy(R[F],null,W[F])]):[3,13];case 11:H.apply(P,[X.sent()]),X.label=12;case 12:return++F,[3,10];case 13:for(_=C.concat(R).concat(k),K=l(_),y.dispose(_),F=0;F<u.length;++F)j=u[F],q=K[F],D[j]=q,y.keep(q);return[4,f.onBatchEnd(S,D)];case 14:X.sent(),Zv(D),S++,w++,X.label=15;case 15:return(i?w>=e.batchesPerEpoch:L.done)?r?(G=void 0,wy(e.validationData)?(Z=Je,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Je(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?b3:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F<n.metricsNames.length;++F)b["val_"+n.metricsNames[F]]=G[F];X.label=19;case 19:return[3,21];case 20:return n.stopTraining_?[3,21]:[3,7];case 21:return[4,f.onEpochEnd(g,b)];case 22:return X.sent(),g++,n.stopTraining_?[3,23]:[3,4];case 23:return[4,f.onTrainEnd()];case 24:return X.sent(),[4,n.history.syncData()];case 25:return X.sent(),[2,n.history];case 26:return n.isTraining=!1,[7];case 27:return[2]}})})}function S3(n,t){var e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(n.size)&&(e=n.size),e}function wy(n){return typeof n.iterator=="function"}function I3(n){return typeof n.next=="function"}function A3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:if(e=e||{},i=e.batches!=null,r=n.testFunction,a=[],e.verbose>0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),I3(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=by(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var x=0;x<L.length;++x)a.push(y.scalar(0));for(var C=S[0].shape[0],R=function(D){var k=L[D],W=a[D];a[D]=y.tidy(function(){return y.add(a[D],y.mul(C,k))}),u>0&&y.dispose(W)},x=0;x<L.length;++x)R(x);y.dispose(L),l+=C,++u}return a}),m.done?(i&&console.warn("Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least `batches` "+("batches (in this case, "+e.batches+" batches). ")+"You may need to use the repeat() function when building your dataset."),[2,"break"]):[2]}})},f.label=4;case 4:return(i?u<e.batches:!0)?[5,c()]:[3,6];case 5:return h=f.sent(),h==="break"?[3,6]:[3,4];case 6:for(d=0;d<a.length;++d)p=a[d],a[d]=y.div(a[d],l),y.dispose(p);return[2,Mt(a)]}})})}function Fh(n){y.util.assert(n>0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function Va(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return er(i,t,e-t)}):er(n,t,e-t)}function Wh(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return Wh(e,t)}):Uv(n,t.dtype==="int32"?t:t.toInt())})}function Uh(n,t){for(var e=[],i=0,r=null;i<n;)r=i+t,r>=n&&(r=n),e.push([i,r]),i=r;return e}function T3(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Le(this,void 0,void 0,function(){var m,g,v,b,w,S,L,x,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=gn(0,g)),s==null&&(s=1),b=ay(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(K){switch(K.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(K.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=Uh(g,r),P=function(j){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(j,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[j][0],X=F[j][1],ee=er(W,Z,X-Z);q.batch=j,q.size=X-Z;for(var ne=Wh(e,ee),ie=t(ne),te=0;te<i.length;++te){var re=i[te],le=ie[te];q[re]=le,y.keep(le)}if(j===F.length-1&&m)for(var he=n.testLoop(l,u,r),te=0;te<i.length;++te){var re=i[te],le=he[te];y.keep(le),k["val_"+re]=le}}),[4,w.onBatchEnd(j,q)];case 2:return G.sent(),Zv(q),n.stopTraining_?[2,"break"]:[2]}})},H=0,K.label=3;case 3:return H<F.length?[5,P(H)]:[3,6];case 4:if(_=K.sent(),_==="break")return[3,6];K.label=5;case 5:return++H,[3,3];case 6:W.dispose(),K.label=7;case 7:return[4,w.onEpochEnd(D,k)];case 8:return K.sent(),n.stopTraining_?[2,"break"]:[2]}})},x=d,R.label=2;case 2:return x<a?[5,L(x)]:[3,5];case 3:if(C=R.sent(),C==="break")return[3,5];R.label=4;case 4:return++x,[3,2];case 5:return[4,w.onTrainEnd()];case 6:return R.sent(),[4,n.history.syncData()];case 7:return R.sent(),[2,n.history]}})})}function N3(n,t,e,i){return i===void 0&&(i={}),Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R,D,k;return ve(this,function(W){switch(W.label){case 0:if(n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,W.label=1;case 1:return W.trys.push([1,,7,8]),h=i.batchSize==null?32:i.batchSize,Fh(h),d=!1,[4,n.standardizeUserData(t,e,i.sampleWeight,i.classWeight,d,h)];case 2:if(p=W.sent(),r=p[0],a=p[1],c=p[2],f=!1,m=void 0,!(i.validationData!=null&&i.validationData.length>0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=Va(r,b,w),r=Va(r,0,b),u=Va(a,b,w),a=Va(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),x=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=x.slice().concat(x.map(function(F){return"val_"+F}))):(C=null,m=[],R=x.slice()),D=iy(i.callbacks,i.yieldEvery),[4,T3(n,L,S,x,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,nr(r,t),nr(a,e),nr(l,s),nr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function Sy(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e<n.length;++e){var i=n[e];if(i.rank===1)t.push(Ua(i,1));else{if(i.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(i)}}return t}function nr(n,t){if(n==null)return;var e=[];if(t instanceof y.Tensor)e.push(t.id);else if(Array.isArray(t))t.forEach(function(l){return e.push(l.id)});else if(t!=null)for(var i in t){var r=t[i];e.push(r.id)}var a=[];if(n instanceof y.Tensor)e.indexOf(n.id)===-1&&a.push(n);else if(Array.isArray(n))n.forEach(function(l){e.indexOf(l.id)===-1&&a.push(l)});else if(n!=null)for(var s in n){var o=n[s];e.indexOf(o.id)===-1&&a.push(o)}a.forEach(function(l){l.isDisposed||l.dispose()})}function x3(n){return n instanceof y.Tensor}function Bh(n){return Array.isArray(n)}function Ly(n){return!x3(n)&&!Bh(n)}function Iy(n,t,e,i,r){if(i===void 0&&(i=!0),r===void 0&&(r=""),t==null||t.length===0){if(n!=null){var a=!1;if(Bh(n)&&n.length>0)a=!0;else if(Ly(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(Ly(n)){n=n,o=[];for(var l=0,u=t;l<u.length;l++){var c=u[l];if(n[c]==null)throw new M('No data provided for "'+c+'". Need data for each key in: '+(""+t));o.push(n[c])}}else if(Bh(n)){if(n=n,n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("model expected. Expected to see "+t.length+" Tensor(s), but ")+("instead got the following list of Tensor(s): "+n));o=n}else{if(n=n,t.length>1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=Sy(o),e!=null)for(var h=0;h<t.length;++h){if(e[h]==null)continue;var d=o[h];if(d.shape.length!==e[h].length)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have "+e[h].length+" dimension(s). but got array with ")+("shape "+d.shape));for(var p=0;p<e[h].length;++p){if(p===0&&!i)continue;var f=d.shape[p],m=e[h][p];if(m!=null&&m>=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function C3(n,t,e){var i=fi(n.map(function(a){return a.shape[0]}));i.sort();var r=fi(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function R3(n,t,e){for(var i=[tr,po,_a],r=0;r<n.length;++r){var a=n[r],s=t[r],o=e[r];if(s==null)continue;if(s===_a&&a.shape[a.shape.length-1]===1)throw new M("You are passing a target array of shape "+a.shape+" while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].");if(i.indexOf(s)!==-1)for(var l=a.shape.slice(1),u=o.slice(1),c=0;c<l.length;++c){var h=l[c],d=u[c];if(d!=null&&h!==d)throw new M("A target Tensor with shape "+a.shape+" was passed for an "+("output of shape "+o+", while using a loss function that ")+"expects targets to have the same shape as the output.")}}}function Ay(n,t,e,i,r){i===void 0&&(i=!0),r===void 0&&(r="");var a;if(Array.isArray(n)){if(n.length!==t.length)throw new M("Error when checking model "+r+": the Array of Tensors that you are passing to your model is not the size the "+("the model expected. Expected to see "+t.length+" Tensor(s),")+(" but instead got "+n.length+" Tensors(s)."));a=n}else{if(t.length>1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s<t.length;++s){if(e[s]==null)continue;var o=a[s];if(o.shape.length!==e[s].length)throw new M("Error when checking "+r+": expected "+t[s]+" "+("to have "+e[s].length+" dimension(s), but got array with ")+("shape "+JSON.stringify(o.shape)));for(var l=0;l<e[s].length;++l){if(l===0&&!i)continue;var u=o.shape[l],c=e[s][l];if(c!=null&&c!==u)throw new M("Error when checking "+r+": expected "+(t[s]+" to have shape "+JSON.stringify(e[s])+" but ")+("got array with shape "+JSON.stringify(o.shape)+"."))}}}function O3(n,t){if(n==null||Array.isArray(n)&&n.length===0)return t.map(function(l){return[]});var e;if(typeof n=="string"||typeof n=="function")e=[n];else if(Array.isArray(n)||typeof n=="object")e=n;else throw new TypeError("Type of metrics argument not understood. Expected an string,"+("function, Array, or Object, found: "+n));if(Array.isArray(e))return t.map(function(l){return e});for(var i=[],r=0,a=t;r<a.length;r++){var s=a[r],o=e.hasOwnProperty(s)?e[s]:[];Array.isArray(o)||(o=[o]),i.push(o)}return i}var E3="layers-model",yi=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.isTraining=!1,i}return t.prototype.summary=function(e,i,r){if(r===void 0&&(r=console.log),!this.built)throw new M("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");h3(this,e,i,r)},t.prototype.compile=function(e){var i=this;if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=s3(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof y.Optimizer))throw new M("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}var r=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(var a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new M('Unknown entry in loss dictionary: "'+a+'". '+("Only expected the following keys: "+this.outputNames));for(var s=0,o=this.outputNames;s<o.length;s++){var l=o[s];e.loss[l]==null&&console.warn('Output "'+l+'" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data '+("to be passed to "+l+" during training")),r.push(Nh(e.loss[l]))}}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new M("When passing an Array as loss, it should have one entry per "+("model output. The model has "+this.outputs.length+" output(s), ")+("but you passed loss="+e.loss+"."));var u=e.loss;r=u.map(function(v){return Nh(v)})}else{var c=Nh(e.loss);this.outputs.forEach(function(v){r.push(c)})}this.lossFunctions=r,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(var h=0;h<this.outputs.length;++h){var d=this.internalOutputShapes[h],p=this.outputNames[h];this.feedOutputNames.push(p),this.feedOutputShapes.push(d),this.feedLossFns.push(this.lossFunctions[h])}var f=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],Qi("loss",function(){for(var v=0;v<i.outputs.length;++v){if(f.indexOf(v)!==-1)continue;var b=i.lossFunctions[v];i.outputs.length>1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=O3(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};Qi("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(x){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var K=i.internalOutputShapes[w];K[K.length-1]===1||i.lossFunctions[w]===po?["accuracy","acc"].indexOf(_)!==-1?D=xh:["crossentropy","ce"].indexOf(_)!==-1&&(D=ly):i.lossFunctions[w]===ho?["accuracy","acc"].indexOf(_)!==-1?D=uy:["crossentropy","ce"].indexOf(_)!==-1&&(D=cy):["accuracy","acc"].indexOf(_)!==-1?D=Ch:["crossentropy","ce"].indexOf(_)!==-1&&(D=Rh);var j=void 0;["accuracy","acc"].indexOf(_)!==-1?j="acc":["crossentropy","ce"].indexOf(_)!==-1&&(j="ce"),k=D,R=C+j}else{var q=a3(_);k=q,R=C+go(_)}var G;Qi(R,function(){G=k}),g(w,R,G)},F=0,P=x;F<P.length;F++){var H=P[F];W(H)}};L(S)},b=0;b<i.outputs.length;++b)v(b)}),this.collectedTrainableWeights=this.trainableWeights},t.prototype.checkTrainableWeightsConsistency=function(){if(this.collectedTrainableWeights==null)return;this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")},t.prototype.evaluate=function(e,i,r){r===void 0&&(r={});var a=r.batchSize==null?32:r.batchSize;Fh(a);var s=!0,o=this.standardizeUserDataXY(e,i,s,a);try{var l=o[0].concat(o[1]);this.makeTestFunction();var u=this.testFunction,c=this.testLoop(u,l,a,r.verbose,r.steps);return Mt(c)}finally{nr(o[0],e),nr(o[1],i)}},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return this.makeTestFunction(),[2,A3(this,e,i)]})})},t.prototype.checkNumSamples=function(e,i,r,a){a===void 0&&(a="steps");var s;if(r!=null){if(s=null,i!=null)throw new M("If "+a+" is set, batchSize must be null or undefined."+("Got batchSize = "+i))}else if(e!=null)Array.isArray(e)?s=e[0].shape[0]:s=e.shape[0];else throw new M("Either the input data should have a defined shape, or "+(a+" shoud be specified."));return s},t.prototype.execute=function(e,i){if(Array.isArray(i)&&i.length===0)throw new M("`outputs` is an empty Array, which is not allowed.");var r=Array.isArray(i),a=r?i:[i],s=this.retrieveSymbolicTensors(a),o=new Gr;if(e instanceof y.Tensor&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new M("The number of inputs provided ("+e.length+") does not match the number of inputs of this model "+("("+this.inputs.length+")."));for(var l=0;l<this.inputs.length;++l)o.add(this.inputs[l],e[l])}else for(var u=0,c=this.inputs;u<c.length;u++){var h=c[u],d=e[h.name];if(d==null)throw new M("No value is provided for the model's input "+h.name);o.add(h,d)}var p=Ha(s,o);return r?p:p[0]},t.prototype.retrieveSymbolicTensors=function(e){for(var i=Ji(null,e.length),r=e.length,a=0,s=this.layers;a<s.length;a++){for(var o=s[a],l=Array.isArray(o.output)?o.output:[o.output],u=l.map(function(p){return p.name}),c=0;c<e.length;++c){var h=u.indexOf(e[c]);if(h!==-1&&(i[c]=l[h],r--),r===0)break}if(r===0)break}if(r>0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=Uh(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=Va(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v<m.length;++v)g.push({key:a.inputs[v],value:m[v]});else g.push({key:a.inputs[0],value:m});var b=new Gr(g);return Ha(a.outputs,b)});d.forEach(function(p,f){return l[f].push(p)})},c=0;c<o.length;++c)u(c);return Mt(l.map(function(h){return y.concat(h,0)}))})},t.prototype.predict=function(e,i){i===void 0&&(i={});var r=Sy(e);Ay(r,this.inputNames,this.feedInputShapes,!1);try{var a=i.batchSize==null?32:i.batchSize;return Fh(a),this.predictLoop(r,a)}finally{nr(r,e)}},t.prototype.predictOnBatch=function(e){Ay(e,this.inputNames,this.feedInputShapes,!0);var i=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,i)},t.prototype.standardizeUserDataXY=function(e,i,r,a){if(this.optimizer_==null)throw new Fn("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");for(var s=[],o=0;o<this.feedOutputShapes.length;++o){var l=this.feedOutputShapes[o],u=this.feedLossFns[o];u===ho?s.push(l.slice(0,l.length-1).concat([1])):s.push(l)}if(e=Iy(e,this.feedInputNames,this.feedInputShapes,!1,"input"),i=Iy(i,this.feedOutputNames,s,!1,"target"),C3(e,i),R3(i,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Le(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=gy(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p<d.length?(m=(f=h).push,[4,vy(c[p],null,d[p])]):[3,4];case 2:m.apply(f,[g.sent()]),g.label=3;case 3:return++p,[3,1];case 4:return[2,[u,c,h]]}})})},t.prototype.testLoop=function(e,i,r,a,s){var o=this;return a===void 0&&(a=0),y.tidy(function(){var l=o.checkNumSamples(i,r,s,"steps"),u=[];if(a>0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=Uh(l,r),h=y.tensor1d(gn(0,l)),d=0;d<c.length;++d){var p=c[d][0],f=c[d][1],m=er(h,p,f-p),g=Wh(i,m),v=e(g);if(d===0)for(var b=0;b<v.length;++b)u.push(y.scalar(0));for(var b=0;b<v.length;++b){var w=v[b];u[b]=y.add(u[b],y.mul(f-p,w))}}for(var b=0;b<u.length;++b)u[b]=y.div(u[b],l);return u})},t.prototype.getDedupedMetricsNames=function(){for(var e=this.metricsNames,i=[],r=0;r<e.length;++r){var a=e[r],s=a;if(wv(e,a)>1){var o=wv(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f<e.inputs.length;++f)p.push({key:e.inputs[f],value:a[f]});for(var m=new Gr(p),g=Ha(e.outputs,m,{training:!0}),v,f=0;f<e.lossFunctions.length;++f){var b=e.lossFunctions[f],w=b(s[f],g[f]);o[f]!=null&&(w=y3(w,o[f]));var S=y.mean(w);r.push(S),f===0?v=w:v=y.add(v,w)}for(var f=0;f<e.metricsTensors.length;++f){var L=void 0;if(e.outputs.length>1&&f<e.outputs.length)L=r[f];else{var x=e.metricsTensors[f][0],C=e.metricsTensors[f][1];L=y.mean(x(s[C],g[C]))}y.keep(L),l.push(L)}return v=y.mean(v),e.calculateLosses().forEach(function(R){v=y.add(v,R)}),v},c=e.collectedTrainableWeights.map(function(p){return p.read()}),h=!0,d=e.optimizer_.minimize(u,h,c);return[d].concat(l)}},t.prototype.makeTestFunction=function(){var e=this;this.testFunction=function(i){return y.tidy(function(){for(var r=[],a,s=i.slice(0,e.inputs.length),o=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),l=[],u=0;u<e.inputs.length;++u)l.push({key:e.inputs[u],value:s[u]});for(var c=new Gr(l),h=Ha(e.outputs,c),u=0;u<e.lossFunctions.length;++u){var d=e.lossFunctions[u],p=y.mean(d(o[u],h[u]));u===0?a=p:a=y.add(a,p),r.push(a)}for(var u=0;u<e.metricsTensors.length;++u){var f=e.metricsTensors[u][0],m=e.metricsTensors[u][1],g=y.mean(f(o[m],h[m]));r.push(g)}return r})}},t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){return[2,N3(this,e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,L3(this,e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p;return ve(this,function(f){switch(f.label){case 0:return[4,this.standardizeUserData(e,i)];case 1:r=f.sent(),a=r[0],s=r[1],o=this.makeTrainFunction(),l=o(a.concat(s)),u=[],c=0,h=l,f.label=2;case 2:return c<h.length?(d=h[c],[4,d.data()]):[3,5];case 3:p=f.sent(),u.push(p[0]),f.label=4;case 4:return c++,[3,2];case 5:return y.dispose(l),[2,Mt(u)]}})})},t.prototype.getNamedWeights=function(e){for(var i=[],r=e!=null&&e.trainableOnly,a=r?this.trainableWeights:this.weights,s=this.getWeights(r),o=0;o<a.length;++o){if(r&&!a[o].trainable)continue;i.push({name:a[o].originalName,tensor:s[o]})}return i},Object.defineProperty(t.prototype,"stopTraining",{get:function(){return this.stopTraining_},set:function(e){this.stopTraining_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.optimizer_},set:function(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)},enumerable:!0,configurable:!0}),t.prototype.dispose=function(){var e=n.prototype.dispose.call(this);if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){var i=y.memory().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=i-y.memory().numTensors}return e},t.prototype.getLossIdentifiers=function(){var e;if(typeof this.loss=="string")e=Jn(this.loss);else if(Array.isArray(this.loss)){for(var i=0,r=this.loss;i<r.length;i++){var a=r[i];if(typeof a!="string")throw new Error("Serialization of non-string loss is not supported.")}e=this.loss.map(function(h){return Jn(h)})}else{var s=Object.keys(this.loss);e={};for(var o=this.loss,l=0,u=s;l<u.length;l++){var c=u[l];if(typeof o[c]=="string")e[c]=Jn(o[c]);else throw new Error("Serialization of non-string loss is not supported.")}}return e},t.prototype.getMetricIdentifiers=function(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[Jn(go(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(function(r){return Jn(go(r))});var e={};for(var i in this.metrics)e[i]=Jn(go(this.metrics[i]));return e},t.prototype.getTrainingConfig=function(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}},t.prototype.loadTrainingConfig=function(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");var i=Ma(e.optimizer_config),r=yn(i),a;if(typeof e.loss=="string")a=Zi(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(function(l){return Zi(l)});else if(e.loss!=null){a={};for(var s in e.loss)a[s]=Zi(e.loss[s])}var o;if(Array.isArray(e.metrics))o=e.metrics.map(function(l){return Zi(l)});else if(e.metrics!=null){o={};for(var s in e.metrics)o[s]=Zi(e.metrics[s])}this.compile({loss:a,metrics:o,optimizer:r})},t.prototype.save=function(e,i){return Le(this,void 0,void 0,function(){var r,a,s,o,l,u,c,h,d,p,f,m,g,v,b;return ve(this,function(w){switch(w.label){case 0:if(typeof e=="string"){if(r=y.io.getSaveHandlers(e),r.length===0)throw new M("Cannot find any save handlers for URL '"+e+"'");if(r.length>1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:E3,generatedBy:"TensorFlow.js tfjs-layers v"+Dh,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,dy(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){dy(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(g3);y.serialization.registerClass(yi);var D3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(yi);y.serialization.registerClass(D3);function k3(n,t){return Le(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=Ma(e),r=yn(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o<l.length;o++)u=l[o],s[u.originalName]=a[u.originalName];r.loadWeights(s),y.dispose(a),c.label=2;case 2:return[2,r]}})})}function W3(n,t){return Le(this,void 0,void 0,function(){var e;return ve(this,function(i){if(t==null&&(t={}),typeof n=="string"){if(e=y.io.getLoadHandlers(n,t),e.length===0)e.push(y.io.browserHTTPRequest(n,t));else if(e.length>1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,F3(n,void 0,t)]})})}function F3(n,t,e){return Le(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=yn(Ma(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=U3(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function U3(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var zh=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:ao("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r<a.length;r++){var s=a[r];i.add(s)}return i}return t.prototype.checkShape=function(e){var i=e.inboundNodes[0].outputTensors[0].shape;if(i.some(function(r){return r<0}))throw new M("Negative dimension size caused by adding layer "+(e.name+" with input shape [")+(e.inboundNodes[0].inputTensors[0].shape+"]"))},t.prototype.add=function(e){var i=e instanceof t||e instanceof yi,r;if(i){if(r=e,r.outputs.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(r.inputs.length!==1)throw new M("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new M("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");var a=Jv({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(i)this.outputs=r.outputs,this.inputs=r.inputs;else{if(e.inboundNodes.length!==1)throw new M("A layer added to a Sequential model must not already be "+("connected somewhere else. LayersModel received layer "+e.name+" ")+("which has "+e.inboundNodes.length+" pre-existing inbound ")+"connections.");if(e.inboundNodes[0].outputTensors.length!==1)throw new M("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Xv(this.outputs[0])}this.inboundNodes=[],new lo({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:Ji(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(function(o){return o.shape}),outputShapes:this.outputs[0].shape})}else{var s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1},t.prototype.pop=function(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{var e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}},t.prototype.call=function(e,i){return this.model==null&&this.build(),this.model.call(e,i)},t.prototype.build=function(e){if(Ke(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new yi({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0},t.prototype.countParams=function(){return this.built||this.build(),n.prototype.countParams.call(this)},t.prototype.summary=function(e,i,r){r===void 0&&(r=console.log),this.built||this.build(),n.prototype.summary.call(this,e,i,r)},t.prototype.setWeights=function(e){this.model==null&&this.build(),this.model.setWeights(e)},t.prototype.evaluate=function(e,i,r){if(r===void 0&&(r={}),!this.built)throw new Fn("The model needs to be compiled before being used.");return this.model.evaluate(e,i,r)},t.prototype.evaluateDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.evaluateDataset(e,i)]})})},t.prototype.predict=function(e,i){return i===void 0&&(i={}),this.model==null&&this.build(),this.model.predict(e,i)},t.prototype.predictOnBatch=function(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)},t.prototype.compile=function(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames},Object.defineProperty(t.prototype,"optimizer",{get:function(){return this.model==null?void 0:this.model.optimizer},set:function(e){this.model.optimizer=e},enumerable:!0,configurable:!0}),t.prototype.fit=function(e,i,r){return r===void 0&&(r={}),Le(this,void 0,void 0,function(){return ve(this,function(a){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fit(e,i,r)]})})},t.prototype.fitDataset=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){if(!this.built)throw new Fn("The model needs to be compiled before being used.");return[2,this.model.fitDataset(e,i)]})})},t.prototype.trainOnBatch=function(e,i){return Le(this,void 0,void 0,function(){return ve(this,function(r){return[2,this.model.trainOnBatch(e,i)]})})},t.fromConfig=function(e,i,r,a){a===void 0&&(a=!1);var s,o={};if(i instanceof Array){if(!(i[0].className!=null)||i[0].className==="Merge")throw new M("Legacy serialization format not supported yet.");s=i}else y.util.assert(i.layers!=null,function(){return"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."}),s=i.layers,delete i.layers,o=i;var l=new e(o);if(!(l instanceof t))throw new Te("Sequential.fromConfig called on non-Sequential input: "+l);for(var u=0,c=s;u<c.length;u++){var h=c[u],d=void 0,p=yn(h,d,a);a&&p.setFastWeightInitDuringBuild(!0),l.add(p)}return l},Object.defineProperty(t.prototype,"stopTraining",{get:function(){if(this.model==null)throw new M("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining},set:function(e){if(this.model==null)throw new M("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e},enumerable:!0,configurable:!0}),t.prototype.getConfig=function(){for(var e=[],i=0,r=this.layers;i<r.length;i++){var a=r[i],s={};s.className=a.getClassName(),s.config=a.getConfig(),e.push(s)}return{name:this.name,layers:e}},t.className="Sequential",t}(yi);y.serialization.registerClass(zh);function B3(n){return new yi(n)}function z3(n){return new zh(n)}function P3(n,t){return t==null&&(t={}),W3(n,t)}function Ty(n){return Jv(n)}function _3(n,t){ry.registerCallbackConstructor(n,t)}var Kt=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),M3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),uk(e,i)},t.className="elu",t}(Kt);y.serialization.registerClass(M3);var H3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.selu(e)},t.className="selu",t}(Kt);y.serialization.registerClass(H3);var V3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.relu(e)},t.className="relu",t}(Kt);y.serialization.registerClass(V3);var q3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tidy(function(){return y.minimum(6,y.relu(e))})},t.className="relu6",t}(Kt);y.serialization.registerClass(q3);var G3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return e},t.className="linear",t}(Kt);y.serialization.registerClass(G3);var Y3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.sigmoid(e)},t.className="sigmoid",t}(Kt);y.serialization.registerClass(Y3);var K3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return hk(e)},t.className="hardSigmoid",t}(Kt);y.serialization.registerClass(K3);var j3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.softplus(e)},t.className="softplus",t}(Kt);y.serialization.registerClass(j3);var $3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return ck(e)},t.className="softsign",t}(Kt);y.serialization.registerClass($3);var X3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e){return y.tanh(e)},t.className="tanh",t}(Kt);y.serialization.registerClass(X3);var Ny=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.softmax(e,i)},t.className="softmax",t}(Kt);y.serialization.registerClass(Ny);var J3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=-1),y.logSoftmax(e,i)},t.className="logSoftmax",t}(Kt);y.serialization.registerClass(J3);var Z3=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return i===void 0&&(i=1),y.tidy(function(){return y.sigmoid(e.mul(i)).mul(e)})},t.className="swish",t}(Kt);y.serialization.registerClass(Z3);function bi(n){return n.getClassName()}function Ph(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"activation")}function wi(n){if(n==null){var t={};return t.className="linear",t.config={},Ph(t)}if(typeof n=="string"){var t={};return t.className=n,t.config={},Ph(t)}else return n instanceof Kt?n:Ph(n)}function _h(n){if(n!=null&&typeof n!="object")throw new Error("Argument to L1L2 regularizer's constructor is expected to be an "+("object, but received: "+n))}var xy=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(y.serialization.Serializable),yo=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return _h(e),i.l1=e==null||e.l1==null?.01:e.l1,i.l2=e==null||e.l2==null?.01:e.l2,i.hasL1=i.l1!==0,i.hasL2=i.l2!==0,i}return t.prototype.apply=function(e){var i=this;return y.tidy(function(){var r=y.zeros([1]);return i.hasL1&&(r=y.add(r,y.sum(y.mul(i.l1,y.abs(e))))),i.hasL2&&(r=y.add(r,y.sum(y.mul(i.l2,Ba(e))))),r.asScalar()})},t.prototype.getConfig=function(){return{l1:this.l1,l2:this.l2}},t.fromConfig=function(e,i){return new e({l1:i.l1,l2:i.l2})},t.className="L1L2",t}(xy);y.serialization.registerClass(yo);function Q3(n){return _h(n),new yo({l1:n!=null?n.l1:null,l2:0})}function e4(n){return _h(n),new yo({l2:n!=null?n.l2:null,l1:0})}var Cy={l1l2:"L1L2"};function je(n){return oh(n)}function Ry(n,t){return t===void 0&&(t={}),Da(n,y.serialization.SerializationMap.getMap().classNameMap,t,"regularizer")}function nt(n){if(n==null)return null;if(typeof n=="string"){var t=n in Cy?Cy[n]:n,e={className:t,config:{}};return Ry(e)}else return n instanceof xy?n:Ry(n)}var Oy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null&&(i.maxValue=e.maxValue),i}return t.prototype.call=function(e,i){e=xe(e);var r=y.relu(e);return this.maxValue!=null&&(r=y.clipByValue(r,0,this.maxValue)),r},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={maxValue:this.maxValue},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ReLU",t}(De);y.serialization.registerClass(Oy);var Ey=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_ALPHA=.3,e==null&&(e={}),i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.leakyRelu(r,this.alpha)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LeakyReLU",t}(De);y.serialization.registerClass(Ey);var Dy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),i.supportsMasking=!0,i.alphaInitializer=tt(e.alphaInitializer||i.DEFAULT_ALPHA_INITIALIZER),i.alphaRegularizer=nt(e.alphaRegularizer),i.alphaConstraint=bt(e.alphaConstraint),e.sharedAxes==null)i.sharedAxes=null;else if(Array.isArray(e.sharedAxes))i.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")i.sharedAxes=[e.sharedAxes];else throw new M("Expected sharedAxes to be a number or an array of numbers, "+("but got "+e.sharedAxes));return i}return t.prototype.build=function(e){e=Ke(e);var i=e.slice(1);if(this.sharedAxes!=null)for(var r=0,a=this.sharedAxes;r<a.length;r++){var s=a[r];i[s-1]=1}this.alpha=this.addWeight("alpha",i,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);var o={};if(this.sharedAxes!=null)for(var s=1;s<e.length;++s)o[s]=e[s];this.inputSpec=[new Tt({ndim:e.length,axes:o})],this.built=!0},t.prototype.call=function(e,i){return e=xe(e),y.prelu(e,this.alpha.read())},t.prototype.getConfig=function(){var e={alphaInitializer:st(this.alphaInitializer),alphaRegularizer:je(this.alphaRegularizer),alphaConstraint:yt(this.alphaConstraint),sharedAxes:this.sharedAxes},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="PReLU",t}(De);y.serialization.registerClass(Dy);var ky=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;if(i.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==i.DEFAULT_ALPHA)throw new Te("Non-default alpha value ("+e.alpha+") is not supported by the ELU layer yet.");return i.alpha=e.alpha==null?i.DEFAULT_ALPHA:e.alpha,i}return t.prototype.call=function(e,i){var r=xe(e);return y.elu(r)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={alpha:this.alpha},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ELU",t}(De);y.serialization.registerClass(ky);var Fy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_THETA=1,e==null&&(e={}),i.theta=e.theta==null?i.DEFAULT_THETA:e.theta,i}return t.prototype.call=function(e,i){var r=xe(e);return r.mul(Wa(r.greater(this.theta),"float32"))},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={theta:this.theta},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ThresholdedReLU",t}(De);y.serialization.registerClass(Fy);var Wy=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.DEFAULT_AXIS=1,e==null&&(e={}),i.softmax=new Ny().apply,i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i}return t.prototype.call=function(e,i){var r=xe(e);return this.softmax(r,this.axis)},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Softmax",t}(De);y.serialization.registerClass(Wy);function Yr(n,t,e){if(typeof n=="number")return Ji(n,t);if(n.length!==t)throw new M("The "+e+" argument must be an integer or tuple of "+t+" integers."+(" Received: "+n.length+" elements."));for(var i=0;i<t;++i){var r=n[i];if(!ak(r))throw new M("The "+e+" argument must be an integer or tuple of "+t+(" integers. Received: "+JSON.stringify(n)+" including a")+(" non-integer number "+r))}return n}function bn(n,t,e,i,r){if(r===void 0&&(r=1),n==null)return n;var a=t+(t-1)*(r-1),s;return e==="same"?s=n:s=n-a+1,Math.floor((s+i-1)/i)}function bo(n,t,e,i){if(n==null)return null;if(i==="valid")n=n*t+gi([e-t,0]);else if(i==="same")n=n*t;else throw new M("Unsupport padding mode: "+i+".");return n}function Mh(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,1]):n})}function Uy(n,t){return y.tidy(function(){return ct(t),t==="channelsFirst"?y.transpose(n,[0,2,3,4,1]):n})}function t4(n,t,e,i,r,a,s){return i===void 0&&(i=1),r===void 0&&(r="valid"),s===void 0&&(s=1),y.tidy(function(){if(a==null&&(a=mn()),ct(a),n.shape.length!==3)throw new M("The input of a conv1dWithBias operation should be 3, but is "+(n.shape.length+" instead."));if(t.shape.length!==3)throw new M("The kernel for a conv1dWithBias operation should be 3, but is "+(t.shape.length+" instead"));if(e!=null&&e.shape.length!==1)throw new M("The bias for a conv1dWithBias operation should be 1, but is "+(t.shape.length+" instead"));if(a==="channelsFirst"&&(n=y.transpose(n,[0,2,1])),r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");var o=y.conv1d(n,t,i,r==="same"?"same":"valid","NWC",s);return e!=null&&(o=Bn(o,e)),o})}function By(n,t,e,i,r,a,s,o){return i===void 0&&(i=[1,1]),r===void 0&&(r="valid"),o===void 0&&(o=null),y.tidy(function(){if(a==null&&(a=mn()),ct(a),n.rank!==3&&n.rank!==4)throw new M("conv2dWithBiasActivation expects input to be of rank 3 or 4, "+("but received "+n.rank+"."));if(t.rank!==3&&t.rank!==4)throw new M("conv2dWithBiasActivation expects kernel to be of rank 3 or 4, "+("but received "+n.rank+"."));var l=Mh(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=y.fused.conv2d({x:l,filter:t,strides:i,pad:r==="same"?"same":"valid",dilations:s,dataFormat:"NHWC",bias:e,activation:o}),a==="channelsFirst"&&(l=y.transpose(l,[0,3,1,2])),l})}function n4(n,t,e,i,r,a,s){return i===void 0&&(i=[1,1,1]),r===void 0&&(r="valid"),y.tidy(function(){if(a==null&&(a=mn()),ct(a),n.rank!==4&&n.rank!==5)throw new M("conv3dWithBias expects input to be of rank 4 or 5, but received "+(n.rank+"."));if(t.rank!==4&&t.rank!==5)throw new M("conv3dWithBias expects kernel to be of rank 4 or 5, but received "+(n.rank+"."));var o=Uy(n,a);if(r==="causal")throw new Te("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=y.conv3d(o,t,i,r==="same"?"same":"valid","NDHWC",s),e!=null&&(o=Bn(o,e)),a==="channelsFirst"&&(o=y.transpose(o,[0,4,1,2,3])),o})}var zy=function(n){Q(t,n);function t(e,i){var r=n.call(this,i)||this;if(r.bias=null,r.DEFAULT_KERNEL_INITIALIZER="glorotNormal",r.DEFAULT_BIAS_INITIALIZER="zeros",t.verifyArgs(i),r.rank=e,At(r.rank,"rank"),r.rank!==1&&r.rank!==2&&r.rank!==3)throw new Te("Convolution layer for rank other than 1, 2, or 3 ("+r.rank+") is not implemented yet.");if(r.kernelSize=Yr(i.kernelSize,e,"kernelSize"),r.strides=Yr(i.strides==null?1:i.strides,e,"strides"),r.padding=i.padding==null?"valid":i.padding,tn(r.padding),r.dataFormat=i.dataFormat==null?"channelsLast":i.dataFormat,ct(r.dataFormat),r.activation=wi(i.activation),r.useBias=i.useBias==null?!0:i.useBias,r.biasInitializer=tt(i.biasInitializer||r.DEFAULT_BIAS_INITIALIZER),r.biasConstraint=bt(i.biasConstraint),r.biasRegularizer=nt(i.biasRegularizer),r.activityRegularizer=nt(i.activityRegularizer),r.dilationRate=Yr(i.dilationRate==null?1:i.dilationRate,e,"dilationRate"),r.rank===1&&Array.isArray(r.dilationRate)&&r.dilationRate.length!==1)throw new M("dilationRate must be a number or an array of a single number for 1D convolution, but received "+(""+JSON.stringify(r.dilationRate)));if(r.rank===2){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==2)throw new M("dilationRate must be a number or array of two numbers for 2D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}else if(r.rank===3){if(typeof r.dilationRate=="number")r.dilationRate=[r.dilationRate,r.dilationRate,r.dilationRate];else if(r.dilationRate.length!==3)throw new M("dilationRate must be a number or array of three numbers for 3D "+("convolution, but received "+JSON.stringify(r.dilationRate)))}return r}return t.verifyArgs=function(e){if(Wn("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!uh(e.kernelSize,"number",1,3))throw new M("BaseConv expects config.kernelSize to be number or number[] with "+("length 1, 2, or 3, but received "+JSON.stringify(e.kernelSize)+"."))},t.prototype.getConfig=function(){var e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:bi(this.activation),useBias:this.useBias,biasInitializer:st(this.biasInitializer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),wo=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;return r.kernel=null,t.verifyArgs(i),r.filters=i.filters,At(r.filters,"filters"),r.kernelInitializer=tt(i.kernelInitializer||r.DEFAULT_KERNEL_INITIALIZER),r.kernelConstraint=bt(i.kernelConstraint),r.kernelRegularizer=nt(i.kernelRegularizer),r}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=this.kernelSize.concat([a,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:(i={},i[r]=a,i)}],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a,s=r.bias==null?null:r.bias.read(),o=Lv(r.activation.getClassName());if(o!=null&&r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate,o);else{if(r.rank===1)a=t4(e,r.kernel.read(),s,r.strides[0],r.padding,r.dataFormat,r.dilationRate[0]);else if(r.rank===2)a=By(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else if(r.rank===3)a=n4(e,r.kernel.read(),s,r.strides,r.padding,r.dataFormat,r.dilationRate);else throw new Te("convolutions greater than 3D are not implemented yet.");r.activation!=null&&(a=r.activation.apply(a))}return a})},t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=[],r=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2),a=0;a<r.length;++a){var s=bn(r[a],this.kernelSize[a],this.padding,this.strides[a],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[a]);i.push(s)}var o=[e[0]];return this.dataFormat==="channelsLast"?(o=o.concat(i),o.push(this.filters)):(o.push(this.filters),o=o.concat(i)),o},t.prototype.getConfig=function(){var e={filters:this.filters,kernelInitializer:st(this.kernelInitializer),kernelRegularizer:je(this.kernelRegularizer),kernelConstraint:yt(this.kernelConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.verifyArgs=function(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new M("Convolution layer expected config.filters to be a 'number' > 0 "+("but got "+JSON.stringify(e.filters)))},t}(zy),Hh=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!uh(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(wo);y.serialization.registerClass(Hh);var Py=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(wo);y.serialization.registerClass(Py);var _y=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Tt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Tt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=bo(c,f,d,r.padding),v=bo(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Bn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=bo(i[a],u,o,this.padding),i[s]=bo(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(Hh);y.serialization.registerClass(_y);var i4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=nt(i.depthwiseRegularizer),r.depthwiseConstraint=bt(i.depthwiseConstraint),r.pointwiseInitializer=tt(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=nt(i.pointwiseRegularizer),r.pointwiseConstraint=bt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=Ke(e),e.length<this.rank+2)throw new M("Inputs to SeparableConv"+this.rank+"D should have rank "+(this.rank+2+", but received input shape: ")+(""+JSON.stringify(e)));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null||e[r]<0)throw new M("The channel dimension of the inputs should be defined, "+("but found "+JSON.stringify(e[r])));for(var a=e[r],s=this.kernelSize.concat([a,this.depthMultiplier]),o=[],l=0;l<this.rank;++l)o.push(1);o.push(a*this.depthMultiplier,this.filters);var u=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,u,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",o,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,u,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,u,this.biasConstraint):this.bias=null,this.inputSpec=[new Tt({ndim:this.rank+2,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a;if(r.rank===1)throw new Te("1D separable convolution is not implemented yet.");return r.rank===2&&(r.dataFormat==="channelsFirst"&&(e=y.transpose(e,[0,2,3,1])),a=y.separableConv2d(e,r.depthwiseKernel.read(),r.pointwiseKernel.read(),r.strides,r.padding,r.dilationRate,"NHWC")),r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),r.dataFormat==="channelsFirst"&&(a=y.transpose(a,[0,3,1,2])),a})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=st(this.depthwiseInitializer),e.pointwiseInitializer=st(this.pointwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.pointwiseRegularizer=je(this.pointwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseConstraint),e.pointwiseConstraint=yt(this.pointwiseConstraint),e},t.className="SeparableConv",t}(wo),My=function(n){Q(t,n);function t(e){return n.call(this,2,e)||this}return t.className="SeparableConv2D",t}(i4);y.serialization.registerClass(My);var Hy=function(n){Q(t,n);function t(e){var i=n.call(this,1,e)||this;return t.verifyArgs(e),i.inputSpec=[{ndim:3}],i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,delete e.dataFormat,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!uh(e.kernelSize,"number",1,1))throw new M("Conv1D expects config.kernelSize to be number or number[] with "+("length 1, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv1D",t}(wo);y.serialization.registerClass(Hy);var Vy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return typeof e.cropping=="number"?i.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?i.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:i.cropping=e.cropping,i.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,i.inputSpec=[{ndim:4}],i}return t.prototype.computeOutputShape=function(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=xe(e),r.dataFormat==="channelsLast"){var a=no(e,r.cropping[0][0],e.shape[1]-r.cropping[0][0]-r.cropping[0][1],2);return no(a,r.cropping[1][0],e.shape[2]-r.cropping[1][1]-r.cropping[1][0],3)}else{var a=no(e,r.cropping[0][0],e.shape[2]-r.cropping[0][0]-r.cropping[0][1],3);return no(a,r.cropping[1][0],e.shape[3]-r.cropping[1][1]-r.cropping[1][0],4)}})},t.prototype.getConfig=function(){var e={cropping:this.cropping,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Cropping2D",t}(De);y.serialization.registerClass(Vy);var qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_SIZE=[2,2],i.inputSpec=[{ndim:4}],i.size=e.size==null?i.DEFAULT_SIZE:e.size,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,i}return t.prototype.computeOutputShape=function(e){if(this.dataFormat==="channelsFirst"){var i=e[2]==null?null:this.size[0]*e[2],r=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],i,r]}else{var i=e[1]==null?null:this.size[0]*e[1],r=e[2]==null?null:this.size[1]*e[2];return[e[0],i,r,e[3]]}},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e),s=a.shape;if(r.dataFormat==="channelsFirst"){a=y.transpose(a,[0,2,3,1]);var o=r.size[0]*s[2],l=r.size[1]*s[3],u=a.resizeNearestNeighbor([o,l]);return y.transpose(u,[0,3,1,2])}else{var o=r.size[0]*s[1],l=r.size[1]*s[2];return a.resizeNearestNeighbor([o,l])}})},t.prototype.getConfig=function(){var e={size:this.size,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="UpSampling2D",t}(De);y.serialization.registerClass(qy);function r4(n,t,e,i,r,a){return e===void 0&&(e=[1,1]),i===void 0&&(i="valid"),y.tidy(function(){r==null&&(r=mn()),ct(r);var s=Mh(n,r);if(n.rank!==4)throw new M("Input for depthwiseConv2d is required to be 4-D, but is instead "+(n.rank+"-D"));if(t.rank!==4)throw new M("depthwiseKernel is required to be 4-D, but is instead "+(t.rank+"-D"));return s=y.depthwiseConv2d(s,t,e,i==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}var Gy=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return i.depthwiseKernel=null,i.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,i.depthwiseInitializer=tt(e.depthwiseInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.depthwiseConstraint=bt(e.depthwiseConstraint),i.depthwiseRegularizer=nt(e.depthwiseRegularizer),i}return t.prototype.build=function(e){if(e=Ke(e),e.length<4)throw new M("Inputs to DepthwiseConv2D should have rank 4. "+("Received input shape: "+JSON.stringify(e)+"."));var i=this.dataFormat==="channelsFirst"?1:3;if(e[i]==null||e[i]<0)throw new M("The channel dimension of the inputs to DepthwiseConv2D should "+("be defined, but is not ("+e[i]+")."));var r=e[i],a=[this.kernelSize[0],this.kernelSize[1],r,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[r*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=r4(e,r.depthwiseKernel.read(),r.strides,r.padding,r.dataFormat,null);return r.useBias&&(a=Bn(a,r.bias.read(),r.dataFormat)),r.activation!=null&&(a=r.activation.apply(a)),a})},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,s=bn(i,this.kernelSize[0],this.padding,this.strides[0]),o=bn(r,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,s,o]:[e[0],s,o,a]},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=st(this.depthwiseInitializer),e.depthwiseRegularizer=je(this.depthwiseRegularizer),e.depthwiseConstraint=yt(this.depthwiseRegularizer),e},t.className="DepthwiseConv2D",t}(zy);y.serialization.registerClass(Gy);function Yy(n,t,e,i){if(Array.isArray(n)){if(t!=null||e!=null)throw new M("When inputs is an array, neither initialState or constants should be provided");i!=null&&(e=n.slice(n.length-i,n.length),n=n.slice(0,n.length-i)),n.length>1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Ky(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(gn(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],x=y.tidy(function(){return n(L,d)});if(r==null)h=x[0],d=x[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=x[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return x[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v<p;++v)g(v);var b;if(o){var w=1;b=y.stack(c,w)}return[h,b,d]})}var Si=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r;if(e.cell==null)throw new M("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?r=new Vh({cells:e.cell}):r=e.cell,r.stateSize==null)throw new M("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");return i.cell=r,i.returnSequences=e.returnSequences==null?!1:e.returnSequences,i.returnState=e.returnState==null?!1:e.returnState,i.goBackwards=e.goBackwards==null?!1:e.goBackwards,i._stateful=e.stateful==null?!1:e.stateful,i.unroll=e.unroll==null?!1:e.unroll,i.supportsMasking=!0,i.inputSpec=[new Tt({ndim:3})],i.stateSpec=null,i.states_=null,i.numConstants=null,i.keptStates=[],i}return t.prototype.getStates=function(){if(this.states_==null){var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return gn(0,e).map(function(i){return null})}else return this.states_},t.prototype.setStates=function(e){this.states_=e},t.prototype.computeOutputShape=function(e){Lh(e)&&(e=e[0]),e=e;var i=this.cell.stateSize;Array.isArray(i)||(i=[i]);var r=i[0],a;if(this.returnSequences?a=[e[0],e[1],r]:a=[e[0],r],this.returnState){for(var s=[],o=0,l=i;o<l.length;o++){var u=l[o];s.push([e[0],u])}return[a].concat(s)}else return a},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){Array.isArray(i)&&(i=i[0]);var a=r.returnSequences?i:null;if(r.returnState){var s=r.states.map(function(o){return null});return[a].concat(s)}else return a})},Object.defineProperty(t.prototype,"states",{get:function(){if(this.states_==null){for(var e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,i=[],r=0;r<e;++r)i.push(null);return i}else return this.states_},set:function(e){this.states_=e},enumerable:!0,configurable:!0}),t.prototype.build=function(e){if(this.numConstants!=null)throw new Te("Constants support is not implemented in RNN yet.");Lh(e)&&(e=e[0]),e=e;var i=this.stateful?e[0]:null,r=e.slice(2);this.inputSpec[0]=new Tt({shape:[i,null].concat(r)});var a=[e[0]].concat(e.slice(2));this.cell.build(a);var s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!y.util.arraysEqual(this.stateSpec.map(function(o){return o.shape[o.shape.length-1]}),s))throw new M("An initialState was passed that is not compatible with "+("cell.stateSize. Received stateSpec="+this.stateSpec+"; ")+("However cell.stateSize is "+this.cell.stateSize))}else this.stateSpec=s.map(function(o){return new Tt({shape:[null,o]})});this.stateful&&this.resetStates()},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape[0];if(a==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.states_==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_=[y.zeros([a,r.cell.stateSize])];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(c){return y.zeros([a,c])}):r.states_[0]=y.zeros([a,r.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i===!0?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var s=0;s<r.states_.length;++s){var o=e[s],l=Array.isArray(r.cell.stateSize)?r.cell.stateSize[s]:r.cell.stateSize,u=[a,l];if(!y.util.arraysEqual(o.shape,u))throw new M("State "+s+" is incompatible with layer "+r.name+": "+("expected shape="+u+", received shape="+o.shape));r.states_[s]=o}}r.states_=r.states_.map(function(c){return y.keep(c.clone())})})},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);e=s.inputs,r=s.initialState,a=s.constants;var o=[],l=[];if(r!=null){i.initialState=r,o=o.concat(r),this.stateSpec=[];for(var u=0,c=r;u<c.length;u++){var h=c[u];this.stateSpec.push(new Tt({shape:h.shape}))}l=l.concat(this.stateSpec)}a!=null&&(i.constants=a,o=o.concat(a),this.numConstants=a.length);var d=o[0]instanceof vn;if(d){var p=[e].concat(o),f=this.inputSpec.concat(l),m=this.inputSpec;this.inputSpec=f;var g=n.prototype.apply.call(this,p,i);return this.inputSpec=m,g}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;e=xe(e),o==null&&(r.stateful?o=r.states_:o=r.getInitialState(e));var l=Array.isArray(r.cell.stateSize)?r.cell.stateSize.length:1;if(o.length!==l)throw new M("RNN Layer has "+l+" state(s) but was passed "+(o.length+" initial state(s)."));r.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");var u={training:s},c=function(g,v){var b=r.cell.call([g].concat(v),u);return[b[0],b.slice(1)]},h=Ky(c,e,o,r.goBackwards,a,null,r.unroll,r.returnSequences),d=h[0],p=h[1],f=h[2];r.stateful&&r.resetStates(f,s);var m=r.returnSequences?p:d;return r.returnState?[m].concat(f):m})},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=y.zeros(e.shape);return r=y.sum(r,[1,2]),r=Ua(r),Array.isArray(i.cell.stateSize)?i.cell.stateSize.map(function(a){return a>1?hh(r,[1,a]):r}):i.cell.stateSize>1?[hh(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),Gt({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=yn(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Si);var Kr=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),qh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:s})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(a)},rate:r.recurrentDropout,training:s}));var o,l=r.dropoutMask,u=r.recurrentDropoutMask;l!=null?o=Un(y.mul(e,l),r.kernel.read()):o=Un(e,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),u!=null&&(a=y.mul(a,u));var c=y.add(o,Un(a,r.recurrentKernel.read()));return r.activation!=null&&(c=r.activation.apply(c)),[c,c]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Gt({},e,i)},t.className="SimpleRNNCell",t}(Kr);y.serialization.registerClass(qh);var jy=function(n){Q(t,n);function t(e){var i=this;return e.cell=new qh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return new e(i)},t.className="SimpleRNN",t}(Si);y.serialization.registerClass(jy);var Gh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new M("GRUCell does not support reset_after parameter set to true.");return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ke(e);var i=e[e.length-1];this.kernel=this.addWeight("kernel",[i,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("GRUCell expects 2 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training==null?!1:i.training,s=e[1];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:3})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:3}));var o=r.dropoutMask,l=r.recurrentDropoutMask,u,c,h;0<r.dropout&&r.dropout<1&&(e=y.mul(e,o[0]));var d=Un(e,r.kernel.read());r.useBias&&(d=Bn(d,r.bias.read())),0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,l[0]));var p=r.recurrentKernel.read(),f=y.split(p,[2*r.units,r.units],p.rank-1),m=f[0],g=f[1],v=Un(s,m),b=y.split(d,3,d.rank-1),w=b[0],S=b[1],L=b[2],x=y.split(v,2,v.rank-1),C=x[0],R=x[1];u=r.recurrentActivation.apply(y.add(w,C)),c=r.recurrentActivation.apply(y.add(S,R));var D=Un(y.mul(c,s),g);h=r.activation.apply(y.add(L,D));var k=y.add(y.mul(u,s),y.mul(y.add(1,y.neg(u)),h));return[k,k]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Gt({},e,i)},t.className="GRUCell",t}(Kr);y.serialization.registerClass(Gh);var $y=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Gh(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="GRU",t}(Si);y.serialization.registerClass($y);var So=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation===void 0?i.DEFAULT_ACTIVATION:e.activation),i.recurrentActivation=wi(e.recurrentActivation===void 0?i.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=tt(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.unitForgetBias=e.unitForgetBias,i.kernelRegularizer=nt(e.kernelRegularizer),i.recurrentRegularizer=nt(e.recurrentRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Hr([1,gi([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Hr([1,gi([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.implementation=e.implementation,i.stateSize=[i.units,i.units],i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel=this.addWeight("kernel",[r,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);var a;if(this.useBias){if(this.unitForgetBias){var s=this.biasInitializer,o=this.units;a=new(i=function(l){Q(u,l);function u(){return l!==null&&l.apply(this,arguments)||this}return u.prototype.apply=function(c,h){var d=s.apply([o]),p=new mh().apply([o]),f=s.apply([o*2]);return Wv(Wv(d,p),f)},u}(un),i.className="CustomInit",i)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training;if(e=e,e.length!==3)throw new M("LSTMCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var s=e[1],o=e[2];e=e[0],0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(e)},rate:r.dropout,training:a,count:4})),0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.recurrentDropout,training:a,count:4}));var l=r.dropoutMask,u=r.recurrentDropoutMask,c,h,d,p;0<r.dropout&&r.dropout<1&&(e=y.mul(e,l[0]));var f=Un(e,r.kernel.read());0<r.recurrentDropout&&r.recurrentDropout<1&&(s=y.mul(s,u[0])),f=y.add(f,Un(s,r.recurrentKernel.read())),r.useBias&&(f=Bn(f,r.bias.read()));var m=y.split(f,4,f.rank-1),g=m[0],v=m[1],b=m[2],w=m[3];c=r.recurrentActivation.apply(g),h=r.recurrentActivation.apply(v),d=y.add(y.mul(h,o),y.mul(c,r.activation.apply(b))),p=r.recurrentActivation.apply(w);var S=y.mul(p,r.activation.apply(d));return[S,S,d]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={units:this.units,activation:bi(this.activation),recurrentActivation:bi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),recurrentInitializer:st(this.recurrentInitializer),biasInitializer:st(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:je(this.kernelRegularizer),recurrentRegularizer:je(this.recurrentRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),recurrentConstraint:yt(this.recurrentConstraint),biasConstraint:yt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Gt({},e,i)},t.className="LSTMCell",t}(Kr);y.serialization.registerClass(So);var Xy=function(n){Q(t,n);function t(e){var i=this;return e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new So(e),i=n.call(this,e)||this,i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null);var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.fromConfig=function(e,i){return i.implmentation===0&&(i.implementation=1),new e(i)},t.className="LSTM",t}(Si);y.serialization.registerClass(Xy);var Vh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.cells=e.cells,i}return Object.defineProperty(t.prototype,"stateSize",{get:function(){for(var e=[],i=0,r=this.cells.slice().reverse();i<r.length;i++){var a=r[i];Array.isArray(a.stateSize)?e.push.apply(e,a.stateSize):e.push(a.stateSize)}return e},enumerable:!0,configurable:!0}),t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=e;for(var a=e.slice(1),s=[],o=0,l=r.cells.slice().reverse();o<l.length;o++){var u=l[o];Array.isArray(u.stateSize)?s.push(a.splice(0,u.stateSize.length)):s.push(a.splice(0,1))}s.reverse();for(var c=[],h,d=0;d<r.cells.length;++d){var u=r.cells[d];a=s[d],d===0?h=[e[0]].concat(a):h=[h[0]].concat(a),h=u.call(h,i),c.push(h.slice(1))}a=[];for(var p=0,f=c.slice().reverse();p<f.length;p++){var m=f[p];a.push.apply(a,m)}return[h[0]].concat(a)})},t.prototype.build=function(e){Lh(e)&&(e=e[0]),e=e;var i;this.cells.forEach(function(r,a){Qi("RNNCell_"+a,function(){r.build(e),Array.isArray(r.stateSize)?i=r.stateSize[0]:i=r.stateSize,e=[e[0],i]})}),this.built=!0},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=function(s){return{className:s.getClassName(),config:s.getConfig()}},r=this.cells.map(i),a={cells:r};return Gt({},e,a)},t.fromConfig=function(e,i,r){r===void 0&&(r={});for(var a=[],s=0,o=i.cells;s<o.length;s++){var l=o[s];a.push(yn(l,r))}return new e({cells:a})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){if(!this.trainable)return[];for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.trainableWeights)}return e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.nonTrainableWeights)}if(!this.trainable){for(var s=[],o=0,l=this.cells;o<l.length;o++){var a=l[o];s.push.apply(s,a.trainableWeights)}return s.concat(e)}return e},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){for(var e=[],i=0,r=this.cells;i<r.length;i++){var a=r[i];e.push.apply(e,a.weights)}return Ih(e)},t.prototype.setWeights=function(e){for(var i=[],r=0,a=this.cells;r<a.length;r++)for(var s=a[r],o=s.weights.length,l=e.splice(o),u=0;u<s.weights.length;++u)i.push([s.weights[u],l[u]]);Ah(i)},t.className="StackedRNNCells",t}(Kr);y.serialization.registerClass(Vh);function Li(n){var t=n.ones,e=n.rate,i=n.training,r=i===void 0?!1:i,a=n.count,s=a===void 0?1:a,o=function(){return Bv(t(),e)},l=function(){return za(o,t,r)};if(!s||s<=1)return y.keep(l().clone());var u=Array(s).fill(void 0).map(l);return u.map(function(c){return y.keep(c.clone())})}var LH=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(Kr),a4=function(n){Q(t,n);function t(e){var i=this;if(e.unroll)throw new Te("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Te("It is not possible at the moment to stack convolutional cells.");return i=n.call(this,e)||this,i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.cell.dropoutMask!=null&&(y.dispose(r.cell.dropoutMask),r.cell.dropoutMask=null),r.cell.recurrentDropoutMask!=null&&(y.dispose(r.cell.recurrentDropoutMask),r.cell.recurrentDropoutMask=null),i&&i.constants)throw new M("ConvRNN2D cell does not support constants");var a=i==null?null:i.mask,s=i==null?null:i.training,o=i==null?null:i.initialState;return n.prototype.call.call(r,e,{mask:a,training:s,initialState:o})})},t.prototype.computeOutputShape=function(e){var i=this.computeSingleOutputShape(e);return this.returnSequences||(i=[i[0]].concat(i.slice(2))),this.returnState&&(i=[i].concat(Array(2).fill([e[0]].concat(i.slice(-3))))),i},t.prototype.getInitialState=function(e){var i=this;return y.tidy(function(){var r=i.cell.stateSize,a=e.shape,s=i.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=y.zeros(o);return Array.isArray(r)?Array(r.length).fill(l):[l]})},t.prototype.resetStates=function(e,i){var r=this;i===void 0&&(i=!1),y.tidy(function(){if(!r.stateful)throw new pi("Cannot call resetStates() on an RNN Layer that is not stateful.");var a=r.inputSpec[0].shape,s=r.computeSingleOutputShape(a),o=[s[0]].concat(s.slice(2)),l=a[0];if(l==null)throw new M("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(r.getStates()==null)Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_=[y.zeros(o)];else if(e==null)y.dispose(r.states_),r.keptStates!=null&&(y.dispose(r.keptStates),r.keptStates=[]),Array.isArray(r.cell.stateSize)?r.states_=r.cell.stateSize.map(function(){return y.zeros(o)}):r.states_[0]=y.zeros(o);else{if(Array.isArray(e)||(e=[e]),e.length!==r.states_.length)throw new M("Layer "+r.name+" expects "+r.states_.length+" state(s), "+("but it received "+e.length+" state value(s). Input ")+("received: "+e));i?r.keptStates.push(r.states_.slice()):y.dispose(r.states_);for(var u=0;u<r.states_.length;++u){var c=e[u],h=o;if(!y.util.arraysEqual(c.shape,h))throw new M("State "+u+" is incompatible with layer "+r.name+": "+("expected shape="+h+", received shape="+c.shape));r.states_[u]=c}}r.states_=r.states_.map(function(d){return y.keep(d.clone())})})},t.prototype.computeSingleOutputShape=function(e){var i=this.cell,r=i.dataFormat,a=i.filters,s=i.kernelSize,o=i.padding,l=i.strides,u=i.dilationRate,c=r==="channelsFirst",h=e[c?3:2],d=e[c?4:3],p=bn(h,s[0],o,l[0],u[0]),f=bn(d,s[1],o,l[1],u[1]),m=e.slice(0,2).concat(c?[a,p,f]:[p,f,a]);return m},t.className="ConvRNN2D",t}(Si),Yh=function(n){Q(t,n);function t(e){var i=this,r=e.filters,a=e.kernelSize,s=e.strides,o=e.padding,l=e.dataFormat,u=e.dilationRate;return i=n.call(this,Gt({},e,{units:r}))||this,i.filters=r,At(i.filters,"filters"),i.kernelSize=Yr(a,2,"kernelSize"),i.kernelSize.forEach(function(c){return At(c,"kernelSize")}),i.strides=Yr(s||1,2,"strides"),i.strides.forEach(function(c){return At(c,"strides")}),i.padding=o||"valid",tn(i.padding),i.dataFormat=l||"channelsLast",ct(i.dataFormat),i.dilationRate=Yr(u||1,2,"dilationRate"),i.dilationRate.forEach(function(c){return At(c,"dilationRate")}),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the input should be defined. "+("Found "+e[r]));var a=e[r],s=4,o=this.kernelSize.concat([a,this.filters*s]);this.kernel=this.addWeight("kernel",o,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);var l=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",l,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){var u=void 0;if(this.unitForgetBias){var c=this.biasInitializer,h=this.filters;u=new(i=function(d){Q(p,d);function p(){return d!==null&&d.apply(this,arguments)||this}return p.prototype.apply=function(f,m){var g=c.apply([h]),v=y.ones([h]),b=c.apply([h*2]);return ph([g,v,b])},p}(un),i.className="CustomInit",i)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e.length!==3)throw new M("ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got "+(e.length+"."));var a=i.training||!1,s=e[0],o=e[1],l=e[2],u=4;0<r.dropout&&r.dropout<1&&r.dropoutMask==null&&(r.dropoutMask=Li({ones:function(){return y.onesLike(s)},rate:r.dropout,training:a,count:u}));var c=r.dropoutMask,h=function(te,re,le){return!re||!re[le]?te:y.mul(re[le],te)},d=h(s,c,0),p=h(s,c,1),f=h(s,c,2),m=h(s,c,3);0<r.recurrentDropout&&r.recurrentDropout<1&&r.recurrentDropoutMask==null&&(r.recurrentDropoutMask=Li({ones:function(){return y.onesLike(o)},rate:r.recurrentDropout,training:a,count:u}));var g=r.recurrentDropoutMask,v=h(o,g,0),b=h(o,g,1),w=h(o,g,2),S=h(o,g,3),L=3,x=y.split(r.kernel.read(),u,L),C=x[0],R=x[1],D=x[2],k=x[3],W=r.useBias?y.split(r.bias.read(),u):[null,null,null,null],F=W[0],P=W[1],H=W[2],_=W[3];d=r.inputConv(d,C,F,r.padding),p=r.inputConv(p,R,P,r.padding),f=r.inputConv(f,D,H,r.padding),m=r.inputConv(m,k,_,r.padding);var K=y.split(r.recurrentKernel.read(),u,L),j=K[0],q=K[1],G=K[2],Z=K[3];v=r.recurrentConv(v,j),b=r.recurrentConv(b,q),w=r.recurrentConv(w,G),S=r.recurrentConv(S,Z);var X=r.recurrentActivation.apply(y.add(d,v)),ee=r.recurrentActivation.apply(y.add(p,b)),ne=y.add(y.mul(ee,l),y.mul(X,r.activation.apply(y.add(f,w)))),ie=y.mul(r.recurrentActivation.apply(y.add(m,S)),r.activation.apply(ne));return[ie,ie,ne]})},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i=e.units,r=VD(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Gt({},r,a)},t.prototype.inputConv=function(e,i,r,a){var s=y.conv2d(e,i,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return r?Bn(s,r,this.dataFormat):s},t.prototype.recurrentConv=function(e,i){var r=1;return y.conv2d(e,i,r,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")},t.className="ConvLSTM2DCell",t}(So);y.serialization.registerClass(Yh);var Jy=function(n){Q(t,n);function t(e){var i=this,r=new Yh(e);return i=n.call(this,Gt({},e,{cell:r}))||this,i}return t.fromConfig=function(e,i){return new e(i)},t.className="ConvLSTM2D",t}(a4);y.serialization.registerClass(Jy);var Kh=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.rate=Math.max(Math.min(e.rate,1),0),i.noiseShape=e.noiseShape,i.seed=e.seed,i.supportsMasking=!0,i}return t.prototype.getNoiseShape=function(e){if(this.noiseShape==null)return this.noiseShape;for(var i=e.shape,r=[],a=0;a<this.noiseShape.length;++a)r.push(this.noiseShape[a]==null?i[a]:this.noiseShape[a]);return r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(0<r.rate&&r.rate<1){var s=i.training==null?!1:i.training,o=r.getNoiseShape(a),l=za(function(){return Bv(a,r.rate,o,r.seed)},function(){return a},s);return l}return e})},t.prototype.getConfig=function(){var e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.dispose=function(){return n.prototype.dispose.call(this)},t.className="Dropout",t}(De);y.serialization.registerClass(Kh);var Zy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[{ndim:3}],i}return t.prototype.getNoiseShape=function(e){var i=e.shape;return[i[0],1,i[2]]},t.className="SpatialDropout1D",t}(Kh);y.serialization.registerClass(Zy);var Qy=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.activation=null,i.useBias=!0,i.kernel=null,i.bias=null,i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){var r=null;e.batchSize!=null&&(r=e.batchSize),i.batchInputShape=[r,e.inputDim]}return i.units=e.units,At(i.units,"units"),i.activation=wi(e.activation),e.useBias!=null&&(i.useBias=e.useBias),i.kernelInitializer=tt(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.biasInitializer=tt(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelConstraint=bt(e.kernelConstraint),i.biasConstraint=bt(e.biasConstraint),i.kernelRegularizer=nt(e.kernelRegularizer),i.biasRegularizer=nt(e.biasRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.supportsMasking=!0,i.inputSpec=[{minNDim:2}],i}return t.prototype.build=function(e){var i;e=Ke(e);var r=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[r,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:(i={},i[-1]=r,i)}],this.built=!0},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return i[i.length-1]=this.units,i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=Lv(r.activation.getClassName()),o;return s!=null?o=Un(a,r.kernel.read(),s,r.bias?r.bias.read():null):(o=Un(a,r.kernel.read()),r.bias!=null&&(o=Bn(o,r.bias.read())),r.activation!=null&&(o=r.activation.apply(o))),o})},t.prototype.getConfig=function(){var e={units:this.units,activation:bi(this.activation),useBias:this.useBias,kernelInitializer:st(this.kernelInitializer),biasInitializer:st(this.biasInitializer),kernelRegularizer:je(this.kernelRegularizer),biasRegularizer:je(this.biasRegularizer),activityRegularizer:je(this.activityRegularizer),kernelConstraint:yt(this.kernelConstraint),biasConstraint:yt(this.biasConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dense",t}(De);y.serialization.registerClass(Qy);var eb=function(n){Q(t,n);function t(e){var i=this;return e=e||{},i=n.call(this,e)||this,i.inputSpec=[{minNDim:3}],i.dataFormat=e.dataFormat,i}return t.prototype.computeOutputShape=function(e){e=Ke(e);for(var i=0,r=e.slice(1);i<r.length;i++){var a=r[i];if(a==null)throw new M('The shape of the input to "Flatten" is not fully defined '+("(got "+e.slice(1)+"). Make sure to pass a complete ")+'"input_shape" or "batch_input_shape" argument to the first layer in your model.')}return[e[0],mi(e,1)]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.dataFormat==="channelsFirst"&&a.rank>1){for(var s=[0],o=2;o<a.rank;++o)s.push(o);s.push(1),a=a.transpose(s)}return lk(a)})},t.prototype.getConfig=function(){var e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);var i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Flatten",t}(De);y.serialization.registerClass(eb);var tb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.activation=wi(e.activation),i}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);return r.activation.apply(a)})},t.prototype.getConfig=function(){var e={activation:bi(this.activation)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Activation",t}(De);y.serialization.registerClass(tb);var nb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.n=e.n,i.inputSpec=[{ndim:2}],i}return t.prototype.computeOutputShape=function(e){return[e[0],this.n,e[1]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return e=xe(e),sk(e,r.n)})},t.prototype.getConfig=function(){var e={n:this.n},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="RepeatVector",t}(De);y.serialization.registerClass(nb);var ib=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;i.targetShape=e.targetShape;for(var r=0;r<i.targetShape.length;++r)i.isUnknown(i.targetShape[r])&&(i.targetShape[r]=null);return i}return t.prototype.isUnknown=function(e){return e<0||e==null},t.prototype.fixUnknownDimension=function(e,i){for(var r="Total size of new array must be unchanged.",a=i.slice(),s=1,o=null,l=0;l<a.length;++l){var u=a[l];if(this.isUnknown(u))if(o===null)o=l;else throw new M("Can only specifiy one unknown dimension.");else s*=u}var c=mi(e);if(o!==null){if(s===0||c%s!==0)throw new M(r);a[o]=c/s}else if(c!==s)throw new M(r);return a},t.prototype.computeOutputShape=function(e){for(var i=!1,r=0;r<e.length;++r)if(this.isUnknown(e[r])){i=!0;break}return i?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=a.shape,o=s.slice(0,1).concat(r.fixUnknownDimension(s.slice(1),r.targetShape));return a.reshape(o)})},t.prototype.getConfig=function(){var e={targetShape:this.targetShape},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Reshape",t}(De);y.serialization.registerClass(ib);var rb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error("Permute constructor requires `dims` to be an Array, but received "+(e.dims+" instead."));var r=gn(1,e.dims.length+1);if(!y.util.arraysEqual(e.dims.slice().sort(),r))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");return i.dims=e.dims,i.dimsIncludingBatch=[0].concat(i.dims),i.inputSpec=[new Tt({ndim:i.dims.length+1})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=e.slice();return this.dims.forEach(function(r,a){i[a+1]=e[r]}),i},t.prototype.call=function(e,i){return y.transpose(xe(e),this.dimsIncludingBatch)},t.prototype.getConfig=function(){var e={dims:this.dims},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Permute",t}(De);y.serialization.registerClass(rb);var ab=function(n){Q(t,n);function t(e){var i=n.call(this,e??{})||this;return i.supportsMasking=!0,e!=null?i.maskValue=e.maskValue==null?0:e.maskValue:i.maskValue=0,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={maskValue:this.maskValue};return Object.assign(i,e),i},t.prototype.computeMask=function(e,i){var r=xe(e),a=-1;return y.any(y.notEqual(r,this.maskValue),a)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=-1,o=!0,l=y.any(y.notEqual(a,r.maskValue),s,o),u=a.mul(l.asType(a.dtype));return u})},t.className="Masking",t}(De);y.serialization.registerClass(ab);var sb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.embeddings=null,i.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){var r=null;e.batchSize!=null&&(r=e.batchSize),e.inputLength==null?i.batchInputShape=[r,null]:i.batchInputShape=[r].concat(Je(e.inputLength))}return i.inputDim=e.inputDim,At(i.inputDim,"inputDim"),i.outputDim=e.outputDim,At(i.outputDim,"outputDim"),i.embeddingsInitializer=tt(e.embeddingsInitializer||i.DEFAULT_EMBEDDINGS_INITIALIZER),i.embeddingsRegularizer=nt(e.embeddingsRegularizer),i.activityRegularizer=nt(e.activityRegularizer),i.embeddingsConstraint=bt(e.embeddingsConstraint),i.maskZero=e.maskZero,i.supportsMasking=e.maskZero,i.inputLength=e.inputLength,i}return t.prototype.build=function(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0},t.prototype.warnOnIncompatibleInputShape=function(e){},t.prototype.computeMask=function(e,i){var r=this;return y.tidy(function(){return r.maskZero?(e=xe(e),y.notEqual(e,y.zerosLike(e))):null})},t.prototype.computeOutputShape=function(e){if(e=Ke(e),this.inputLength==null)return e.concat([this.outputDim]);var i=Je(this.inputLength);if(i.length!==e.length-1)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));for(var r=0,a=0;a<i.length;++a){var s=i[a],o=e[a+1];if(s!=null&&o!=null&&s!==o)throw new M('"inputLength" is '+this.inputLength+", but received "+("input shape has shape "+e));s==null&&(i[r]=o),r++}return[e[0]].concat(i,[this.outputDim])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);a.dtype!=="int32"&&(a=Wa(a,"int32"));var s=Uv(r.embeddings.read(),a.as1D());return s.reshape(Ke(r.computeOutputShape(a.shape)))})},t.prototype.getConfig=function(){var e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:st(this.embeddingsInitializer),embeddingsRegularizer:je(this.embeddingsRegularizer),activityRegularizer:je(this.activityRegularizer),embeddingsConstraint:yt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Embedding",t}(De);y.serialization.registerClass(sb);var ir=function(n){Q(t,n);function t(e){var i=n.call(this,e||{})||this;return i.supportsMasking=!0,i}return t.prototype.mergeFunction=function(e){throw new Te},t.prototype.computeElementwiseOpOutputShape=function(e,i){if(e==null||i==null)return null;if(e.length<i.length)return this.computeElementwiseOpOutputShape(i,e);if(i.length===0)return e;for(var r=e.slice(0,e.length-i.length),a=0;a<i.length;++a){var s=e[e.length-i.length+a],o=i[a];if(s==null||o==null||s<0||o<0)r.push(null);else if(s===1)r.push(o);else if(o===1)r.push(s);else{if(s!==o)throw new M("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(i));r.push(s)}}return r},t.prototype.build=function(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[Ke(e)]),e=e,e.length<2)throw new M("A merge layer should be called on an Array of at least 2 inputs."+(" Got "+e.length+" input(s)."));for(var i=[],r=0,a=e;r<a.length;r++){var s=a[r];s!=null&&s[0]!==null&&i.push(s[0])}if(i=fi(i),i.length>1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l<e.length;++l){var s=e[l]==null?null:e[l].slice(1);o=this.computeElementwiseOpOutputShape(o,s)}var u=e.map(function(c){return c.length});e.indexOf(null)===-1&&fi(u).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,r.reshapeRequired){var a=[],s=e.map(function(D){return D.rank});if(s.indexOf(null)===-1){for(var o=gi(s),l=0,u=e;l<u.length;l++){for(var c=u[l],h=c.rank,d=0;d<o-h;++d)c=Ua(c,1);a.push(c)}return r.mergeFunction(a)}else{for(var p=!1,f=0,m=e;f<m.length;f++){var c=m[f],h=c.rank;if(h==null){var g=c.shape,v=g[0],b=g.slice(1).concat([v]),w=c.reshape([v].concat(mi(g.slice(1))));w=y.transpose(w,[1,0]),w=w.reshape(b),a.push(w),p=!0}else if(h>1){var S=gn(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),x=L.rank;if(p){if(x==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(x>1){var S=[x-1].concat(gn(0,x-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r<e.length;++r){var a=e[r]==null?null:e[r].slice(1);i=this.computeElementwiseOpOutputShape(i,a)}for(var s=[],o=0,l=e;o<l.length;o++){var a=l[o];a!=null&&a[0]!==null&&s.push(a[0])}return s=fi(s),s.length===1?i=s.concat(i):i=[null].concat(i),i},t.prototype.computeMask=function(e,i){return y.tidy(function(){if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an Array");if(!Array.isArray(e))throw new M("`inputs` should be an Array");if(i.length!==e.length)throw new M("The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths "+("("+e.length+" vs "+i.length+")"));if(i.every(function(s){return s==null}))return null;i=i.map(function(s){return s==null?s:y.expandDims(s,0)});for(var r=i[0],a=1;a<i.length-1;++a)r=y.logicalAnd(r,i[a]);return r})},t}(De),ob=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return i})},t.className="Add",t}(ir);y.serialization.registerClass(ob);var lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.mul(i,e[r]);return i})},t.className="Multiply",t}(ir);y.serialization.registerClass(lb);var ub=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0].clone(),r=1;r<e.length;++r)i=y.add(i,e[r]);return y.mul(1/e.length,i)})},t.className="Average",t}(ir);y.serialization.registerClass(ub);var cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.maximum(i,e[r]);return i})},t.className="Maximum",t}(ir);y.serialization.registerClass(cb);var hb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.mergeFunction=function(e){return y.tidy(function(){for(var i=e[0],r=1;r<e.length;++r)i=y.minimum(i,e[r]);return i})},t.className="Minimum",t}(ir);y.serialization.registerClass(hb);var db=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_AXIS=-1,e==null&&(e={}),i.axis=e.axis==null?i.DEFAULT_AXIS:e.axis,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new M("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;for(var i=!0,r=0,a=e;r<a.length;r++){var s=a[r];if(s!=null){i=!1;break}}if(i)return;for(var o=[],l=0;l<e.length;++l){var u=e[l].slice();u.splice(this.axis,1);for(var c=!1,h=0,d=o;h<d.length;h++){var s=d[h];if(y.util.arraysEqual(s,u)){c=!0;break}}c||o.push(u)}if(o.length>1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return ph(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s<o.length;s++){var l=o[s];if(r[a]==null||l[a]==null){r[a]=null;break}r[a]+=l[a]}return r},t.prototype.computeMask=function(e,i){var r=this;if(i==null)return null;if(!Array.isArray(i))throw new M("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new M("`inputs` should be an array for Concatenate");if(i.length!==e.length)throw new M("Mismatch in the length of mask ("+i.length+") "+("and the legnth of inputs ("+e.length+")"));return y.tidy(function(){var a=!0;if(i.forEach(function(u){if(u!=null){a=!1;return}}),a)return null;for(var s=[],o=0;o<e.length;++o)i[o]==null?s.push(y.onesLike(e[o]).asType("bool")):i[o].rank<e[o].rank?s.push(y.expandDims(i[o],-1)):s.push(i[o]);var l=y.concat(s,r.axis);return y.all(l,-1,!1)})},t.prototype.getConfig=function(){var e={axis:this.axis},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Concatenate",t}(ir);y.serialization.registerClass(db);function qa(n,t){for(;n<0;)n+=t;return n}function s4(n,t,e){if(n.shape.length>3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;l<s;++l)o.push(1);t=t.reshape(t.shape.concat(o))}else if(r>i){s=r-i;for(var o=[],l=0;l<s;++l)o.push(1);n=n.reshape(n.shape.concat(o))}else s=0;var u;if(n.shape.length===2&&t.shape.length===2)a[0]===a[1]?u=n.mul(t).sum(a[0]):u=n.transpose([1,0]).mul(t).sum(a[1]);else{var c=a[0]!==n.shape.length-1,h=a[1]===t.shape.length-1;u=n.matMul(t,c,h)}if(s>0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l<d+s;++l)p.push(l);u=u.squeeze(p)}return u.shape.length===1&&(u=u.expandDims(1)),u})}var pb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.axes=e.axes,i.normalize=e.normalize==null?!1:e.normalize,i.supportsMasking=!0,i.reshapeRequired=!1,i}return t.prototype.build=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0],r=e[1];if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return qa(s,e[o].shape.length)}):a=[qa(this.axes,i.shape.length),qa(this.axes,r.shape.length)],this.normalize&&(i=uo(i,a[0]),r=uo(r,a[1])),s4(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[qa(this.axes,e.length),qa(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(ir);y.serialization.registerClass(pb);var fb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return io(a.shape,0,r.stddev).add(a)},o=za(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass(fb);var mb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(io(a.shape,1,o))};return za(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(mb);var gb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=Wa(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return za(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(gb);function Ga(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function o4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=Ga(n,s,o,e,t,r);return[l,s,o]})}function l4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=gn(0,n.rank);u<c.length;u++){var h=c[u];i.indexOf(h)!==-1?l.push(1):l.push(n.shape[h])}var d=s.reshape(l),p=o.reshape(l),f=t==null?null:t.reshape(l),m=e==null?null:e.reshape(l),g=Ga(n,d,p,m,f,r);return[g,s,o]})}function u4(n,t,e,i,r){return r===void 0&&(r=.001),y.util.arraysEqual(i.slice().sort(),gn(0,n.rank-1))?o4(n,t,e,i,r):l4(n,t,e,i,r)}var vb=function(n){Q(t,n);function t(e){var i=this;return e==null&&(e={}),i=n.call(this,e)||this,i.supportsMasking=!0,i.axis=e.axis==null?-1:e.axis,i.momentum=e.momentum==null?.99:e.momentum,i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.movingMeanInitializer=tt(e.movingMeanInitializer||"zeros"),i.movingVarianceInitializer=tt(e.movingVarianceInitializer||"ones"),i.betaConstraint=bt(e.betaConstraint),i.gammaConstraint=bt(e.gammaConstraint),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i}return t.prototype.build=function(e){var i;e=Ke(e);var r=this.axis>=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Tt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=gn(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=Ji(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,gn(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),x=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return Ga(s,L,x,C,R,r.epsilon)}else return Ga(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=u4(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,x,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(x).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),movingMeanInitializer:st(this.movingMeanInitializer),movingVarianceInitializer:st(this.movingVarianceInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer),betaConstraint:yt(this.betaConstraint),gammaConstraint:yt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(vb);var yb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r<a.length;r++){var s=a[r];if(!Number.isInteger(s))throw new Error("Expected axis to be an array of integers, "+("but received "+JSON.stringify(i.axis)))}else throw new Error("Expected axis to be an integer or an array of integers, "+("but received "+JSON.stringify(i.axis)));return i.epsilon=e.epsilon==null?.001:e.epsilon,i.center=e.center==null?!0:e.center,i.scale=e.scale==null?!0:e.scale,i.betaInitializer=tt(e.betaInitializer||"zeros"),i.gammaInitializer=tt(e.gammaInitializer||"ones"),i.betaRegularizer=nt(e.betaRegularizer),i.gammaRegularizer=nt(e.gammaRegularizer),i.supportsMasking=!0,i}return t.prototype.build=function(e){e=Ke(e);var i=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(var r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=i);for(var a=0,s=this.axis;a<s.length;a++){var o=s[a];if(o<0||o>=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==fi(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=Ji(1,o),p=0,f=r.axis;p<f.length;p++){var m=f[p];d[m]=s[m]}for(var g=function(x){return x!=null&&x.shape.length!==o&&r.axis!==[o-1]?x.reshape(d):x},v=g(r.gamma.read()),b=g(r.beta.read()),w=[],S=[],L=0;L<o;++L)r.axis.indexOf(L)!==-1?(w.push(s[L]),S.push(1)):(w.push(1),S.push(s[L]));return c=c.tile(w),h=h.tile(w),v=v.tile(S),b=b.tile(S),Ga(a,c,h,b,v,r.epsilon)})},t.prototype.getConfig=function(){var e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:st(this.betaInitializer),gammaInitializer:st(this.gammaInitializer),betaRegularizer:je(this.betaRegularizer),gammaRegularizer:je(this.gammaRegularizer)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="LayerNormalization",t}(De);y.serialization.registerClass(yb);function c4(n,t,e){return y.tidy(function(){if(n.rank!==4)throw new M("temporalPadding expects input tensor to be 4-D, but received a "+(n.rank+"-D tensor."));if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new M("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(e==null&&(e=mn()),e!=="channelsLast"&&e!=="channelsFirst")throw new M("Unknown data format: "+e+". Supported data formats are 'channelsLast' and 'channelsFirst.");var i;return e==="channelsFirst"?i=[[0,0],[0,0],t[0],t[1]]:i=[[0,0],t[0],t[1],[0,0]],y.pad(n,i)})}var bb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.dataFormat=e.dataFormat==null?mn():e.dataFormat,e.padding==null)i.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")i.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new M("ZeroPadding2D expects padding to be a length-2 array, but "+("received a length-"+e.padding.length+" array."));var r=void 0,a=void 0;if(typeof e.padding[0]=="number")r=[e.padding[0],e.padding[0]],a=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new M("ZeroPadding2D expects height padding to be a length-2 array, "+("but received a length-"+e.padding[0].length+" array."));if(r=e.padding[0],e.padding[1].length!==2)throw new M("ZeroPadding2D expects width padding to be a length-2 array, "+("but received a length-"+e.padding[1].length+" array."));a=e.padding[1]}i.padding=[r,a]}return i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i,r;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return c4(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(bb);function Lo(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),tn(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=mn()),a==null&&(a="max"),n=Mh(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function wb(n,t,e,i,r,a){return y.tidy(function(){ct(r),Rv(a),tn(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=mn()),a==null&&(a="max"),n=Uy(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var Sb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(At(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,tn(i.padding),i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=bn(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=Ua(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Lb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(Sb);y.serialization.registerClass(Lb);var Ib=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(Sb);y.serialization.registerClass(Ib);var Ab=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),tn(i.padding),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=bn(i,this.poolSize[0],this.padding,this.strides[0]),r=bn(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Tb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(Ab);y.serialization.registerClass(Tb);var Nb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),Lo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(Ab);y.serialization.registerClass(Nb);var xb=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),tn(i.padding),i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=Ke(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=bn(i,this.poolSize[0],this.padding,this.strides[0]),r=bn(r,this.poolSize[1],this.padding,this.strides[1]),a=bn(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Cb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),wb(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(xb);y.serialization.registerClass(Cb);var Rb=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),tn(a),wb(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(xb);y.serialization.registerClass(Rb);var Ob=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),Eb=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(Ob);y.serialization.registerClass(Eb);var Db=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(Ob);y.serialization.registerClass(Db);var kb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),Fb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(kb);y.serialization.registerClass(Fb);var Wb=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(kb);y.serialization.registerClass(Wb);var Ub=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=yn(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),Bb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=Ke(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=Ke(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Ky(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(Ub);y.serialization.registerClass(Bb);function h4(n){_r(nk,"BidirectionalMergeMode",n)}var d4="concat",zb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=yn(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=yn(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?d4:e.mergeMode,h4(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):Mt(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=Yy(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Tt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof vn,d=0,p=o;d<p.length;d++){var f=p[d];if(f instanceof vn!==h)throw new M("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors")}if(h){var m=[e].concat(o),g=this.inputSpec.concat(l),v=this.inputSpec;this.inputSpec=g;var b=n.prototype.apply.call(this,m,i);return this.inputSpec=v,b}else return n.prototype.apply.call(this,e,i)},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.initialState,s,o;if(a==null)s=r.forwardLayer.call(e,i),o=r.backwardLayer.call(e,i);else{var l=a.slice(0,a.length/2),u=a.slice(a.length/2);s=r.forwardLayer.call(e,Object.assign(i,{initialState:l})),o=r.backwardLayer.call(e,Object.assign(i,{initialState:u}))}var c;r.returnState&&(Array.isArray(s)&&(c=s.slice(1).concat(o.slice(1))),s=s[0],o=o[0]),r.returnSequences&&(o=y.reverse(o,1));var h;return r.mergeMode==="concat"?h=ph([s,o]):r.mergeMode==="sum"?h=y.add(s,o):r.mergeMode==="ave"?h=y.mul(.5,y.add(s,o)):r.mergeMode==="mul"?h=y.mul(s,o):r.mergeMode==null&&(h=[s,o]),r.returnState?r.mergeMode==null?h.concat(c):[h].concat(c):h})},t.prototype.resetStates=function(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()},t.prototype.build=function(e){var i=this;Qi(this.forwardLayer.name,function(){i.forwardLayer.build(e)}),Qi(this.backwardLayer.name,function(){i.backwardLayer.build(e)}),this.built=!0},t.prototype.computeMask=function(e,i){Array.isArray(i)&&(i=i[0]);var r;if(this.returnSequences?this.mergeMode==null?r=[i,i]:r=i:this.mergeMode==null?r=[null,null]:r=null,this.returnState){var a=this.forwardLayer.states,s=a.map(function(o){return null});return Array.isArray(r)?r.concat(s).concat(s):[r].concat(s).concat(s)}else return r},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e={mergeMode:this.mergeMode},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.fromConfig=function(e,i){var r=yn(i.layer);if(delete i.layer,i.numConstants!=null)throw new Te("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");var a=i;return a.layer=r,new e(a)},t.className="Bidirectional",t}(Ub);y.serialization.registerClass(zb);function p4(n){return new Pa(n)}function f4(n){return new ky(n)}function m4(n){return new Oy(n)}function g4(n){return new Ey(n)}function v4(n){return new Dy(n)}function y4(n){return new Wy(n)}function b4(n){return new Fy(n)}function w4(n){return new Hy(n)}function S4(n){return new Hh(n)}function L4(n){return new _y(n)}function I4(n){return new Py(n)}function A4(n){return new My(n)}function T4(n){return new Vy(n)}function N4(n){return new qy(n)}function x4(n){return new Gy(n)}function C4(n){return new tb(n)}function R4(n){return new Qy(n)}function O4(n){return new Kh(n)}function E4(n){return new Zy(n)}function D4(n){return new eb(n)}function k4(n){return new nb(n)}function F4(n){return new ib(n)}function W4(n){return new rb(n)}function U4(n){return new sb(n)}function B4(n){return new ob(n)}function z4(n){return new ub(n)}function P4(n){return new db(n)}function _4(n){return new cb(n)}function M4(n){return new hb(n)}function H4(n){return new lb(n)}function V4(n){return new pb(n)}function q4(n){return new vb(n)}function G4(n){return new yb(n)}function Y4(n){return new bb(n)}function jh(n){return new Ib(n)}function K4(n){return jh(n)}function j4(n){return jh(n)}function $h(n){return new Nb(n)}function $4(n){return $h(n)}function X4(n){return $h(n)}function Xh(n){return new Rb(n)}function J4(n){return Xh(n)}function Z4(n){return Xh(n)}function Q4(n){return new Eb(n)}function eF(n){return new Fb(n)}function Pb(n){return new Db(n)}function _b(n){return new Wb(n)}function Mb(n){return new Lb(n)}function Hb(n){return new Tb(n)}function tF(n){return new Cb(n)}function nF(n){return new $y(n)}function iF(n){return new Gh(n)}function rF(n){return new Xy(n)}function aF(n){return new So(n)}function sF(n){return new jy(n)}function oF(n){return new qh(n)}function lF(n){return new Jy(n)}function uF(n){return new Yh(n)}function cF(n){return new Si(n)}function hF(n){return new Vh(n)}function dF(n){return new zb(n)}function pF(n){return new Bb(n)}var fF=Pb,mF=_b,gF=Mb,vF=Hb;function yF(n){return new fb(n)}function bF(n){return new mb(n)}function wF(n){return new gb(n)}function SF(n){return new ab(n)}var LF={__proto__:null,inputLayer:p4,elu:f4,reLU:m4,leakyReLU:g4,prelu:v4,softmax:y4,thresholdedReLU:b4,conv1d:w4,conv2d:S4,conv2dTranspose:L4,conv3d:I4,separableConv2d:A4,cropping2D:T4,upSampling2d:N4,depthwiseConv2d:x4,activation:C4,dense:R4,dropout:O4,spatialDropout1d:E4,flatten:D4,repeatVector:k4,reshape:F4,permute:W4,embedding:U4,add:B4,average:z4,concatenate:P4,maximum:_4,minimum:M4,multiply:H4,dot:V4,batchNormalization:q4,layerNormalization:G4,zeroPadding2d:Y4,averagePooling1d:jh,avgPool1d:K4,avgPooling1d:j4,averagePooling2d:$h,avgPool2d:$4,avgPooling2d:X4,averagePooling3d:Xh,avgPool3d:J4,avgPooling3d:Z4,globalAveragePooling1d:Q4,globalAveragePooling2d:eF,globalMaxPooling1d:Pb,globalMaxPooling2d:_b,maxPooling1d:Mb,maxPooling2d:Hb,maxPooling3d:tF,gru:nF,gruCell:iF,lstm:rF,lstmCell:aF,simpleRNN:sF,simpleRNNCell:oF,convLstm2d:lF,convLstm2dCell:uF,rnn:cF,stackedRNNCells:hF,bidirectional:dF,timeDistributed:pF,globalMaxPool1d:fF,globalMaxPool2d:mF,maxPool1d:gF,maxPool2d:vF,Layer:De,RNN:Si,RNNCell:Kr,input:Ty,gaussianNoise:yF,gaussianDropout:bF,alphaDropout:wF,masking:SF};function IF(n,t){return xh(n,t)}function AF(n,t){return ly(n,t)}function TF(n,t){return uy(n,t)}function NF(n,t){return Ch(n,t)}function xF(n,t){return Rh(n,t)}function CF(n,t){return oy(n,t)}function RF(n,t){return Jk(n,t)}function OF(n,t){return Th(n,t)}function EF(n,t){return co(n,t)}function DF(n,t){return qr(n,t)}function kF(n,t){return qr(n,t)}function FF(n,t){return qr(n,t)}function WF(n,t){return tr(n,t)}function UF(n,t){return tr(n,t)}function BF(n,t){return tr(n,t)}var zF={__proto__:null,binaryAccuracy:IF,binaryCrossentropy:AF,sparseCategoricalAccuracy:TF,categoricalAccuracy:NF,categoricalCrossentropy:xF,precision:CF,recall:RF,cosineProximity:OF,meanAbsoluteError:EF,meanAbsolutePercentageError:DF,MAPE:kF,mape:FF,meanSquaredError:WF,MSE:UF,mse:BF};var PF={__proto__:null,modelFromJSON:k3};function _F(n){return new yo(n)}function MF(n){return Q3(n)}function HF(n){return e4(n)}var VF={__proto__:null,l1l2:_F,l1:MF,l2:HF};var Vb=function(n){Q(t,n);function t(){var e=n!==null&&n.apply(this,arguments)||this;return e.model=null,e}return t.prototype.setModel=function(e){if(!(e instanceof yi))throw new Error("model must be a LayersModel, not some other Container");this.model=e},t}(Vr);function Io(n,t){return n<t}function qb(n,t){return n>t}var Gb=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=Io:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=qb:i.monitorFunc=Io,i.monitorFunc===Io&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===Io?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Le(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,vi(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Le(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(Vb);function qF(n){return new Gb(n)}var GF={earlyStopping:qF};Xe.Callback=Vb;Xe.CallbackList=ey;Xe.CustomCallback=ny;Xe.EarlyStopping=Gb;Xe.History=ty;Xe.InputSpec=Tt;Xe.LayerVariable=$v;Xe.LayersModel=yi;Xe.RNN=Si;Xe.Sequential=zh;Xe.SymbolicTensor=vn;Xe.callbacks=GF;Xe.constraints=ZD;Xe.initializers=Dk;Xe.input=Ty;Xe.layers=LF;Xe.loadLayersModel=P3;Xe.metrics=zF;Xe.model=B3;Xe.models=PF;Xe.registerCallbackConstructor=_3;Xe.regularizers=VF;Xe.sequential=z3;Xe.version_layers=Dh});var ow=be(rr=>{"use strict";Object.defineProperty(rr,"__esModule",{value:!0});var B=Xi();var Kb=Object.assign||function(t){for(var e,i=1,r=arguments.length;i<r;i++){e=arguments[i];for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&(t[a]=e[a])}return t};function zn(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function wn(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=r[u[0]&2?"return":u[0]?"throw":"next"])&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[0,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var Sn;(function(n){n[n.DT_INVALID=0]="DT_INVALID",n[n.DT_FLOAT=1]="DT_FLOAT",n[n.DT_DOUBLE=2]="DT_DOUBLE",n[n.DT_INT32=3]="DT_INT32",n[n.DT_UINT8=4]="DT_UINT8",n[n.DT_INT16=5]="DT_INT16",n[n.DT_INT8=6]="DT_INT8",n[n.DT_STRING=7]="DT_STRING",n[n.DT_COMPLEX64=8]="DT_COMPLEX64",n[n.DT_INT64=9]="DT_INT64",n[n.DT_BOOL=10]="DT_BOOL",n[n.DT_QINT8=11]="DT_QINT8",n[n.DT_QUINT8=12]="DT_QUINT8",n[n.DT_QINT32=13]="DT_QINT32",n[n.DT_BFLOAT16=14]="DT_BFLOAT16",n[n.DT_FLOAT_REF=101]="DT_FLOAT_REF",n[n.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",n[n.DT_INT32_REF=103]="DT_INT32_REF",n[n.DT_UINT8_REF=104]="DT_UINT8_REF",n[n.DT_INT16_REF=105]="DT_INT16_REF",n[n.DT_INT8_REF=106]="DT_INT8_REF",n[n.DT_STRING_REF=107]="DT_STRING_REF",n[n.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",n[n.DT_INT64_REF=109]="DT_INT64_REF",n[n.DT_BOOL_REF=110]="DT_BOOL_REF",n[n.DT_QINT8_REF=111]="DT_QINT8_REF",n[n.DT_QUINT8_REF=112]="DT_QUINT8_REF",n[n.DT_QINT32_REF=113]="DT_QINT32_REF",n[n.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(Sn||(Sn={}));var jb;(function(n){var t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=n.CheckpointFormatVersion||(n.CheckpointFormatVersion={}))})(jb||(jb={}));var Jh={};function YF(n,t){var e={tfOpName:n,category:"custom",inputs:[],attrs:[],customExecutor:t};Jh[n]=e}function $b(n){return Jh[n]}function KF(n){delete Jh[n]}function A(n,t,e,i){var r=t.inputParams[n];if(r&&r.inputIndexStart!==void 0){var a=r.inputIndexStart,s=r.inputIndexEnd===0?void 0:r.inputIndexEnd===void 0?a+1:r.inputIndexEnd;if(r.type==="tensor")return Ht(t.inputNames[r.inputIndexStart],e,i);if(r.type==="tensors"){var o=t.inputNames.slice(a,s);return o.map(function(h){return Ht(h,e,i)})}var l=Ht(t.inputNames.slice(a)[0],e,i),u=l.dataSync();return r.type==="number"?u[0]:B.util.toNestedArray(l.shape,u)}var c=t.attrParams[n];return c&&c.value}function Ht(n,t,e){var i=jt(n),r=i[0],a=i[1],s=e.currentContextIds.find(function(o){return!!t[Ao(r,o)]});return s!==void 0?t[Ao(r,s)][a]:void 0}function jF(n,t,e){return t[Ao(n,e.currentContextId)]}function Zn(n,t){var e=jt(n),i=e[0],r=e[1];return[Ao(i,t&&t.currentContextId),r]}function Ao(n,t){return t?n+"-"+t:n}function jt(n){var t=n.split(":");if(t.length===1)return[n,0];var e=t[0];return[e,Number(t[t.length-1])]}function To(n,t,e){var i=A("pad",n,t,e);if(i==="explicit"){i=A("explicitPaddings",n,t,e);for(var r=[[0,0],[0,0],[0,0],[0,0]],a=0;a<4;a++)r[a][0]=i[a*2],r[a][1]=i[a*2+1];return r}return i}function Qn(n){return n.kept?n:B.clone(n)}var $F=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],XF={__proto__:null,json:$F};var JF=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"clip_value_min",name:"clipValueMin",type:"number"},{tfName:"clip_value_max",name:"clipValueMax",type:"number"}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"clipValueMin",name:"clipValueMin",type:"number",defaultValue:0},{tfName:"clipValueMax",name:"clipValueMax",type:"number",defaultValue:6}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],ZF={__proto__:null,json:JF};var QF=[{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],eW={__proto__:null,json:QF};var tW=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],nW={__proto__:null,json:tW};var iW=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],rW={__proto__:null,json:iW};var aW=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],sW={__proto__:null,json:aW};var oW=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],lW={__proto__:null,json:oW};var uW=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],cW={__proto__:null,json:uW};var hW=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],dW={__proto__:null,json:hW};var pW=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],fW={__proto__:null,json:pW};var mW=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],gW={__proto__:null,json:mW};var vW=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],yW={__proto__:null,json:vW};var bW=[{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],wW={__proto__:null,json:bW};var SW=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool",notSupported:!0}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],LW={__proto__:null,json:SW};var IW=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],AW={__proto__:null,json:IW};var TW=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]}],NW={__proto__:null,json:TW};var Jb=function(){function n(){var t=[XF,ZF,eW,nW,rW,sW,lW,fW,dW,cW,gW,yW,wW,LW,AW,NW],e=[].concat.apply([],t.map(function(i){return i.json}));this.opMappers=e.reduce(function(i,r){return i[r.tfOpName]=r,i},{})}return Object.defineProperty(n,"Instance",{get:function(){return this._instance||(this._instance=new this)},enumerable:!0,configurable:!0}),n.prototype.transformGraph=function(t,e){var i=this;e===void 0&&(e={});var r=t.node,a=[],s=[],o=[],l=r.reduce(function(g,v){return g[v.name]=i.mapNode(v),v.op.startsWith("Placeholder")?a.push(g[v.name]):v.op==="Const"?s.push(g[v.name]):(v.input==null||v.input.length===0)&&o.push(g[v.name]),g},{}),u=[],c=[],h={},d={};e!=null&&(h=this.mapSignatureEntries(e.inputs),d=this.mapSignatureEntries(e.outputs));var p=Object.keys(l);p.forEach(function(g){var v=l[g];v.inputNames.forEach(function(b){var w=Zn(b)[0];v.inputs.push(l[w]),l[w].children.push(v)})}),Object.keys(d).length===0?p.forEach(function(g){var v=l[g];v.children.length===0&&c.push(v)}):Object.keys(d).forEach(function(g){var v=Zn(g)[0],b=l[v];b!=null&&(b.signatureKey=d[g],c.push(b))}),Object.keys(h).length>0?Object.keys(h).forEach(function(g){var v=Zn(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=$b(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Zh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Zh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=sd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=sd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=ed(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=ed(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=ad(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=ad(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Qh(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Qh(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=ld(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=ld(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=od(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=nd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=nd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=id(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=id(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Xb(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Xb(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=Zn(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:td(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=Zn(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=Zn(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function xW(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function Zb(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):xW(n);return t?e:e.toLowerCase()}function Zh(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?Zb(r.s,i):e}function Qh(n,t,e){var i=n[t];return i?i.b:e}function ed(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function td(n){typeof n=="string"&&(n=Sn[n]);switch(n){case Sn.DT_FLOAT:return"float32";case Sn.DT_INT32:case Sn.DT_INT64:case Sn.DT_INT8:case Sn.DT_UINT8:return"int32";case Sn.DT_BOOL:return"bool";case Sn.DT_DOUBLE:return"float32";case Sn.DT_STRING:return"string";default:return null}}function Xb(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function nd(n,t,e){var i=n[t];return i&&i.type?td(i.type):e}function id(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return td(r)}):e}function Qb(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function rd(n,t,e){var i=n[t];return i&&i.shape?Qb(i.shape):e}function ad(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function sd(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return Zb(a,i)}):e}function od(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Qb(r)}):e}function ld(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var CW=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Ht(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Ht(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return ed(this.node.rawAttrs,t,e);if(i.s!=null)return Zh(this.node.rawAttrs,t,e);if(i.b!=null)return Qh(this.node.rawAttrs,t,e);if(i.shape!=null)return rd(this.node.rawAttrs,t,e);if(i.type!=null)return nd(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return ad(this.node.rawAttrs,t,e);if(i.list.s!=null)return sd(this.node.rawAttrs,t,e);if(i.list.shape!=null)return od(this.node.rawAttrs,t,e);if(i.list.b!=null)return ld(this.node.rawAttrs,t,e);if(i.list.type!=null)return id(this.node.rawAttrs,t,e)}return e},n}();var RW=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var OW=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Ht(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function cn(n,t,e){e===void 0&&(e=""),B.util.assert(EW(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function EW(n,t){if(n.length!==t.length)return!1;for(var e=0;e<n.length;e++)if(n[e]!==-1&&t[e]!==-1&&n[e]!==t[e])return!1;return!0}var DW=function(){function n(t,e,i,r,a,s,o){this.name=t,this.dtype=e,this.maxSize=i,this.elementShape=r,this.identicalElementShapes=a,this.dynamicSize=s,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=B.scalar(0),B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"closed",{get:function(){return this.closed_},enumerable:!0,configurable:!0}),n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.read=function(t){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||t>=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`,
+          because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),cn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i<this.size();i++)t.push(i)}if(t.length===0)return B.tensor([],[0].concat(this.elementShape));var r=this.readMany(t);return cn(this.elementShape,r[0].shape,"TensorArray shape mismatch: "),B.stack(r,0)},n.prototype.concat=function(t){if(!!t&&t!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but concat requested dtype "+t);if(this.size()===0)return B.tensor([],[0].concat(this.elementShape));for(var e=[],i=0;i<this.size();i++)e.push(i);var r=this.readMany(e);return cn(this.elementShape,r[0].shape,"TensorArray shape mismatch: tensor array shape ("+this.elementShape+") vs first tensor shape ("+r[0].shape+")"),B.concat(r,0)},n.prototype.scatter=function(t,e){if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);if(t.length!==e.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+e.shape[0]);var i=Math.max.apply(Math,t);if(!this.dynamicSize&&i>=this.maxSize)throw new Error("Max index must be < array size ("+i+"  vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
           tensor.shape[0], but sum of lengths is
-        `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;c<t.length;++c){var h=c===0?0:a[c-1],d=[0,h,0],p=[1,t[c],s];o[c]=B.reshape(B.slice(e,d,p),i.elementShape)}return o});for(var l=[],u=0;u<t.length;u++)l[u]=u;this.writeMany(l,o)},n}();var No=function(){function n(t,e,i,r){r===void 0&&(r=-1),this.tensors=t,this.elementShape=e,this.elementDtype=i,t!=null&&t.forEach(function(a){if(i!==a.dtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+a.dtype);hn(e,a.shape,"TensorList shape mismatch: "),B.keep(a)}),this.idTensor=B.scalar(0),this.maxNumElements=r,B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),n.prototype.copy=function(){return new n(this.tensors.slice(),this.elementShape,this.elementDtype)},n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.stack=function(t,e,i){var r=this;if(i===void 0&&(i=-1),e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(i!==-1&&this.tensors.length!==i)throw new Error("Operation expected a list with "+i+" elements but got a list with "+this.tensors.length+" elements.");return hn(t,this.elementShape,"TensorList shape mismatch: "),B.tidy(function(){var a=r.tensors.map(function(s){return B.reshape(s,t)});return B.stack(a,0)})},n.prototype.popBack=function(t,e){if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(this.size()===0)throw new Error("Trying to pop from an empty list.");var i=this.tensors.pop();return hn(i.shape,t,"TensorList shape mismatch: "),B.reshape(i,t)},n.prototype.pushBack=function(t){if(t.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+t.dtype+", but list elements "+this.elementDtype);if(hn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");B.keep(t),this.tensors.push(t)},n.prototype.resize=function(t){if(t<0)throw new Error("TensorListResize expects size to be non-negative. Got: "+t);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return hn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");hn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return hn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return hn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function kW(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);hn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new No(a,t,i)}function FW(n,t,e){return new No([],n,t,e)}function WW(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+"  vs. "+i+")");var a=new No([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function UW(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to
+        `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;c<t.length;++c){var h=c===0?0:a[c-1],d=[0,h,0],p=[1,t[c],s];o[c]=B.reshape(B.slice(e,d,p),i.elementShape)}return o});for(var l=[],u=0;u<t.length;u++)l[u]=u;this.writeMany(l,o)},n}();var No=function(){function n(t,e,i,r){r===void 0&&(r=-1),this.tensors=t,this.elementShape=e,this.elementDtype=i,t!=null&&t.forEach(function(a){if(i!==a.dtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+a.dtype);cn(e,a.shape,"TensorList shape mismatch: "),B.keep(a)}),this.idTensor=B.scalar(0),this.maxNumElements=r,B.keep(this.idTensor)}return Object.defineProperty(n.prototype,"id",{get:function(){return this.idTensor.id},enumerable:!0,configurable:!0}),n.prototype.copy=function(){return new n(this.tensors.slice(),this.elementShape,this.elementDtype)},n.prototype.clearAndClose=function(t){this.tensors.forEach(function(e){(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()},n.prototype.size=function(){return this.tensors.length},n.prototype.stack=function(t,e,i){var r=this;if(i===void 0&&(i=-1),e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(i!==-1&&this.tensors.length!==i)throw new Error("Operation expected a list with "+i+" elements but got a list with "+this.tensors.length+" elements.");return cn(t,this.elementShape,"TensorList shape mismatch: "),B.tidy(function(){var a=r.tensors.map(function(s){return B.reshape(s,t)});return B.stack(a,0)})},n.prototype.popBack=function(t,e){if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);if(this.size()===0)throw new Error("Trying to pop from an empty list.");var i=this.tensors.pop();return cn(i.shape,t,"TensorList shape mismatch: "),B.reshape(i,t)},n.prototype.pushBack=function(t){if(t.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+t.dtype+", but list elements "+this.elementDtype);if(cn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");B.keep(t),this.tensors.push(t)},n.prototype.resize=function(t){if(t<0)throw new Error("TensorListResize expects size to be non-negative. Got: "+t);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return cn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");cn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return cn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return cn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function kW(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);cn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new No(a,t,i)}function FW(n,t,e){return new No([],n,t,e)}function WW(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+"  vs. "+i+")");var a=new No([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function UW(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to
           tensor.shape[0], but sum of lengths is
-        `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c<t.length;++c){var h=c===0?0:r[c-1],d=[0,h,0],p=[1,t[c],a];u[c]=B.reshape(B.slice(n,d,p),e)}return n.dispose(),u}),o=new No([],e,n.dtype,t.length),l=0;l<s.length;l++)o.setItem(l,s[l]);return o}var BW=void 0,zW=function(n,t,e){return zn(BW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,o,h,d,l,p,f,m,m,g,v,g,b,g,g,g,w,S,L,x,C,R,D,k,W,F,P,H,_,K,j,q,G,Z,X,ee,ne,ie,te,re,le,he,ye,Oe,Fe,_e,Me,ht,St,Ge,Ue,F,P,me,Ue,K,L,Dt,me,ne,ie,L,lt,me,L,dt,lt,me,q,G,L,dt,me,Ue,L,dt,lt,me,xn,L,dt,me,re,me,he,L,Ue,P,me,Ue,L,Dt,me,Oe,L,Fe,me;return Sn(this,function(Rt){switch(Rt.label){case 0:i=n.op;switch(i){case"If":return[3,1];case"StatelessIf":return[3,1];case"While":return[3,3];case"StatelessWhile":return[3,3];case"LoopCond":return[3,9];case"Switch":return[3,10];case"Merge":return[3,12];case"Enter":return[3,13];case"Exit":return[3,14];case"NextIteration":return[3,15];case"TensorArrayV3":return[3,16];case"TensorArrayWriteV3":return[3,17];case"TensorArrayReadV3":return[3,18];case"TensorArrayGatherV3":return[3,19];case"TensorArrayScatterV3":return[3,20];case"TensorArrayConcatV3":return[3,21];case"TensorArraySplitV3":return[3,22];case"TensorArraySizeV3":return[3,23];case"TensorArrayCloseV3":return[3,24];case"TensorListSetItem":return[3,25];case"TensorListGetItem":return[3,26];case"TensorListScatterV2":return[3,27];case"TensorListScatter":return[3,27];case"TensorListReserve":return[3,28];case"TensorListGather":return[3,29];case"TensorListStack":return[3,30];case"TensorListFromTensor":return[3,31];case"TensorListConcat":return[3,32];case"TensorListPushBack":return[3,33];case"TensorListPopBack":return[3,34];case"TensorListSplit":return[3,35]}return[3,36];case 1:return r=A("thenBranch",n,t,e),a=A("elseBranch",n,t,e),s=A("cond",n,t,e),o=A("args",n,t,e),[4,s.data()];case 2:return l=Rt.sent(),l[0]?[2,e.functionMap[r].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)]:[2,e.functionMap[a].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 3:return u=A("body",n,t,e),c=A("cond",n,t,e),o=A("args",n,t,e),[4,e.functionMap[c].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 4:return h=Rt.sent(),d=o.map(function(Lt){return Lt.id}),[4,h[0].data()];case 5:l=Rt.sent(),h.forEach(function(Lt){!Lt.kept&&d.indexOf(Lt.id)===-1&&Lt.dispose()}),p=o,f=function(){var Lt,Ce,kt;return Sn(this,function(pt){switch(pt.label){case 0:return Lt=p,[4,e.functionMap[u].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 1:return p=pt.sent(),Ce=p.map(function(it){return it.id}),Lt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[4,e.functionMap[c].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 2:return kt=pt.sent(),[4,kt[0].data()];case 3:return l=pt.sent(),kt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[2]}})},Rt.label=6;case 6:return l[0]?[5,f()]:[3,8];case 7:return Rt.sent(),[3,6];case 8:return[2,p];case 9:return m=A("pred",n,t,e),[2,[Qn(m)]];case 10:return m=A("pred",n,t,e),g=A("data",n,t,e),g.kept||(g=Qn(g)),[4,m.data()];case 11:return[2,Rt.sent()[0]?[void 0,g]:[g,void 0]];case 12:return v=n.inputNames.find(function(Lt){return Ht(Lt,t,e)!==void 0}),v?(g=Ht(v,t,e),[2,[Qn(g)]]):[2,void 0];case 13:return b=A("frameName",n,t,e),g=A("tensor",n,t,e),e.enterFrame(b),[2,[Qn(g)]];case 14:return g=A("tensor",n,t,e),e.exitFrame(),[2,[Qn(g)]];case 15:return g=A("tensor",n,t,e),e.nextIteration(),[2,[Qn(g)]];case 16:return w=A("size",n,t,e),S=A("dtype",n,t,e),L=A("elementShape",n,t,e),x=A("dynamicSize",n,t,e),C=A("clearAfterRead",n,t,e),R=A("identicalElementShapes",n,t,e),D=A("name",n,t,e),k=new DW(D,S,w,L,R,x,C),e.addTensorArray(k),[2,[k.idTensor,B.scalar(1)]];case 17:return W=A("tensorArrayId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),H=e.getTensorArray(W.id),H.write(F,P),[2,[H.idTensor]];case 18:return _=A("tensorArrayId",n,t,e),K=A("index",n,t,e),j=e.getTensorArray(_.id),[2,[j.read(K)]];case 19:return q=A("tensorArrayId",n,t,e),G=A("indices",n,t,e),Z=A("dtype",n,t,e),X=e.getTensorArray(q.id),[2,[X.gather(G,Z)]];case 20:return ee=A("tensorArrayId",n,t,e),ne=A("indices",n,t,e),ie=A("tensor",n,t,e),te=e.getTensorArray(ee.id),te.scatter(ne,ie),[2,[te.idTensor]];case 21:return re=A("tensorArrayId",n,t,e),le=e.getTensorArray(re.id),he=A("dtype",n,t,e),[2,[le.concat(he)]];case 22:return ye=A("tensorArrayId",n,t,e),Oe=A("tensor",n,t,e),Fe=A("lengths",n,t,e),_e=e.getTensorArray(ye.id),_e.split(Fe,Oe),[2,[_e.idTensor]];case 23:return Me=A("tensorArrayId",n,t,e),ht=e.getTensorArray(Me.id),[2,[B.scalar(ht.size(),"int32")]];case 24:return St=A("tensorArrayId",n,t,e),Ge=e.getTensorArray(St.id),Ge.clearAndClose(),[2,[Ge.idTensor]];case 25:return Ue=A("tensorListId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.setItem(F,P),[2,[me.idTensor]];case 26:return Ue=A("tensorListId",n,t,e),K=A("index",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.getItem(K,L,Dt)]];case 27:return ne=A("indices",n,t,e),ie=A("tensor",n,t,e),L=A("elementShape",n,t,e),lt=A("numElements",n,t,e),me=WW(ie,ne,L,lt),e.addTensorList(me),[2,[me.idTensor]];case 28:return L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=FW(L,dt,lt),e.addTensorList(me),[2,[me.idTensor]];case 29:return q=A("tensorListId",n,t,e),G=A("indices",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=e.getTensorList(q.id),[2,[me.gather(G,dt,L)]];case 30:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=e.getTensorList(Ue.id),[2,[me.stack(L,dt,lt)]];case 31:return xn=A("tensor",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=kW(xn,L,dt),e.addTensorList(me),[2,[me.idTensor]];case 32:return re=A("tensorListId",n,t,e),me=e.getTensorList(re.id),he=A("dtype",n,t,e),L=A("elementShape",n,t,e),[2,[me.concat(he,L)]];case 33:return Ue=A("tensorListId",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.pushBack(P),[2,[me.idTensor]];case 34:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.popBack(L,Dt)]];case 35:return Oe=A("tensor",n,t,e),L=A("elementShape",n,t,e),Fe=A("lengths",n,t,e),me=UW(Oe,Fe,L),e.addTensorList(me),[2,[me.idTensor]];case 36:throw TypeError("Node type "+n.op+" is not implemented")}})})};function ew(n,t,e){var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=r==="fusedbatchnorm",u=A("numArgs",n,t,e);if(s){if(o&&u!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&u!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");var c=A("strides",n,t,e),h=To(n,t,e),d=A("dataFormat",n,t,e).toUpperCase(),p=A("dilations",n,t,e),f=A("args",n,t,e),m=f[0],g=f[1];return{stride:c,pad:h,dataFormat:d,dilations:p,biasArg:m,preluArg:g,activationFunc:a}}var PW=function(n,t,e){switch(n.op){case"Conv1D":{var i=A("stride",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),s=A("dilation",n,t,e);return[B.conv1d(A("x",n,t,e),A("filter",n,t,e),i,r,a,s)]}case"Conv2D":{var i=A("strides",n,t,e),r=To(n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv2d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{var l=ew(n,t,e),i=l.stride,r=l.pad,a=l.dataFormat,o=l.dilations,u=l.biasArg,c=l.preluArg,h=l.activationFunc;return[B.fused.conv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"FusedDepthwiseConv2dNative":{var d=ew(n,t,e),i=d.stride,r=d.pad,a=d.dataFormat,o=d.dilations,u=d.biasArg,c=d.preluArg,h=d.activationFunc;return[B.fused.depthwiseConv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{var p=A("outputShape",n,t,e),i=A("strides",n,t,e),r=To(n,t,e);return[B.conv2dTranspose(A("x",n,t,e),A("filter",n,t,e),p,[i[1],i[2]],r)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{var i=A("strides",n,t,e),r=To(n,t,e),o=A("dilations",n,t,e),a=A("dataFormat",n,t,e).toUpperCase();return[B.depthwiseConv2d(A("input",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"Conv3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv3d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2],i[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPoolWithArgmax":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e),m=A("includeBatchInIndex",n,t,e),g=B.maxPoolWithArgmax(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r,m),v=g.result,b=g.indexes;return[v,b]}case"AvgPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"MaxPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"Dilation2D":{var w=A("strides",n,t,e),r=A("pad",n,t,e),o=A("dilations",n,t,e),S=w[1],L=w[2],x=o[1],C=o[2];return[B.dilation2d(A("x",n,t,e),A("filter",n,t,e),[S,L],r,[x,C],"NHWC")]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var _W=function(n,t,e){switch(n.op){case"Fill":{var i=A("shape",n,t,e),r=A("dtype",n,t,e),a=A("value",n,t,e);return[B.fill(i,a,r)]}case"LinSpace":{var s=A("start",n,t,e),o=A("stop",n,t,e),l=A("num",n,t,e);return[B.linspace(s,o,l)]}case"Multinomial":{var u=A("logits",n,t,e),c=A("numSamples",n,t,e),h=A("seed",n,t,e);return[B.multinomial(u,c,h)]}case"OneHot":{var d=A("indices",n,t,e),p=A("depth",n,t,e),f=A("onValue",n,t,e),m=A("offValue",n,t,e);return[B.oneHot(d,p,f,m)]}case"Ones":return[B.ones(A("shape",n,t,e),A("dtype",n,t,e))];case"OnesLike":return[B.onesLike(A("x",n,t,e))];case"RandomUniform":return[B.randomUniform(A("shape",n,t,e),A("minval",n,t,e),A("maxval",n,t,e),A("dtype",n,t,e))];case"Range":{var s=A("start",n,t,e),g=A("stop",n,t,e),v=A("step",n,t,e);return[B.range(s,g,v,A("dtype",n,t,e))]}case"TruncatedNormal":{var i=A("shape",n,t,e),b=A("mean",n,t,e),w=A("stdDev",n,t,e),h=A("seed",n,t,e);return[B.truncatedNormal(i,b,w,A("dtype",n,t,e),h)]}case"Zeros":return[B.zeros(A("shape",n,t,e),A("dtype",n,t,e))];case"ZerosLike":return[B.zerosLike(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var MW=void 0;function ld(n,t,e){var i=A("boxes",n,t,e),r=A("scores",n,t,e),a=A("maxOutputSize",n,t,e),s=A("iouThreshold",n,t,e),o=A("scoreThreshold",n,t,e),l=A("softNmsSigma",n,t,e);return{boxes:i,scores:r,maxOutputSize:a,iouThreshold:s,scoreThreshold:o,softNmsSigma:l}}var HW=function(n,t,e){return zn(MW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,a,s,o,l,u,p,h,f,a,s,o,l,u,m,h;return Sn(this,function(g){switch(g.label){case 0:i=n.op;switch(i){case"NonMaxSuppressionV5":return[3,1];case"NonMaxSuppressionV4":return[3,3];case"NonMaxSuppressionV3":return[3,5];case"NonMaxSuppressionV2":return[3,5];case"Where":return[3,7];case"ListDiff":return[3,9]}return[3,10];case 1:return r=ld(n,t,e),a=r.boxes,s=r.scores,o=r.maxOutputSize,l=r.iouThreshold,u=r.scoreThreshold,c=r.softNmsSigma,[4,B.image.nonMaxSuppressionWithScoreAsync(a,s,o,l,u,c)];case 2:return h=g.sent(),[2,[h.selectedIndices,h.selectedScores]];case 3:return d=ld(n,t,e),a=d.boxes,s=d.scores,o=d.maxOutputSize,l=d.iouThreshold,u=d.scoreThreshold,p=A("padToMaxOutputSize",n,t,e),[4,B.image.nonMaxSuppressionPaddedAsync(a,s,o,l,u,p)];case 4:return h=g.sent(),[2,[h.selectedIndices,h.validOutputs]];case 5:return f=ld(n,t,e),a=f.boxes,s=f.scores,o=f.maxOutputSize,l=f.iouThreshold,u=f.scoreThreshold,[4,B.image.nonMaxSuppressionAsync(a,s,o,l,u)];case 6:return[2,[g.sent()]];case 7:return m=B.cast(A("condition",n,t,e),"bool"),[4,B.whereAsync(m)];case 8:return h=[g.sent()],m.dispose(),[2,h];case 9:return[2,B.setdiff1dAsync(A("x",n,t,e),A("y",n,t,e))];case 10:throw TypeError("Node type "+n.op+" is not implemented")}})})};var VW=function(n,t,e){switch(n.op){case"TopKV2":{var i=A("x",n,t,e),r=A("k",n,t,e),a=A("sorted",n,t,e),s=B.topk(i,r,a);return[s.values,s.indices]}case"Unique":{var i=A("x",n,t,e),s=B.unique(i);return[s.values,s.indices]}case"UniqueV2":{var i=A("x",n,t,e),o=A("axis",n,t,e),s=B.unique(i,o);return[s.values,s.indices]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var qW=function(n,t,e){switch(n.op){case"Const":return t[n.name];case"PlaceholderWithDefault":var i=A("default",n,t,e);return[Ht(n.name,t,e)||i];case"Placeholder":return[Ht(n.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{var r=A("x",n,t,e);return[Qn(r)]}case"IdentityN":return A("x",n,t,e).map(function(h){return Qn(h)});case"Snapshot":var a=A("x",n,t,e);return[Qn(a)];case"Shape":return[B.tensor1d(A("x",n,t,e).shape,"int32")];case"ShapeN":return A("x",n,t,e).map(function(h){return B.tensor1d(h.shape)});case"Size":return[B.scalar(A("x",n,t,e).size,"int32")];case"Rank":return[B.scalar(A("x",n,t,e).rank,"int32")];case"NoOp":return[B.scalar(1)];case"Print":var s=A("x",n,t,e),o=A("data",n,t,e),l=A("message",n,t,e),u=A("summarize",n,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(l);for(var c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,u));return[s];default:throw TypeError("Node type "+n.op+" is not implemented")}};var GW=function(n,t,e){switch(n.op){case"ResizeBilinear":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeBilinear(i,[r[0],r[1]],a)]}case"ResizeNearestNeighbor":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeNearestNeighbor(i,[r[0],r[1]],a)]}case"CropAndResize":{var s=A("image",n,t,e),o=A("boxes",n,t,e),l=A("boxInd",n,t,e),u=A("cropSize",n,t,e),c=A("method",n,t,e),h=A("extrapolationValue",n,t,e);return[B.image.cropAndResize(s,o,l,u,c,h)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var YW=function(n,t,e){switch(n.op){case"Equal":return[B.equal(A("a",n,t,e),A("b",n,t,e))];case"NotEqual":return[B.notEqual(A("a",n,t,e),A("b",n,t,e))];case"Greater":return[B.greater(A("a",n,t,e),A("b",n,t,e))];case"GreaterEqual":return[B.greaterEqual(A("a",n,t,e),A("b",n,t,e))];case"Less":return[B.less(A("a",n,t,e),A("b",n,t,e))];case"LessEqual":return[B.lessEqual(A("a",n,t,e),A("b",n,t,e))];case"LogicalAnd":return[B.logicalAnd(A("a",n,t,e),A("b",n,t,e))];case"LogicalNot":return[B.logicalNot(A("a",n,t,e))];case"LogicalOr":return[B.logicalOr(A("a",n,t,e),A("b",n,t,e))];case"Select":case"SelectV2":return[B.where(A("condition",n,t,e),A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var KW=function(n,t,e){switch(n.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[B.matMul(A("a",n,t,e),A("b",n,t,e),A("transposeA",n,t,e),A("transposeB",n,t,e))];case"Transpose":return[B.transpose(A("x",n,t,e),A("perm",n,t,e))];case"_FusedMatMul":var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=A("numArgs",n,t,e);if(s){if(o&&l!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&l!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}var u=A("args",n,t,e),c=u[0],h=u[1];return[B.fused.matMul({a:A("a",n,t,e),b:A("b",n,t,e),transposeA:A("transposeA",n,t,e),transposeB:A("transposeB",n,t,e),bias:c,activation:a,preluActivationWeights:h})];default:throw TypeError("Node type "+n.op+" is not implemented")}};var jW=function(n,t,e){switch(n.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"FusedBatchNormV3":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"LRN":return[B.localResponseNormalization(A("x",n,t,e),A("radius",n,t,e),A("bias",n,t,e),A("alpha",n,t,e),A("beta",n,t,e))];case"Softmax":return[B.softmax(A("x",n,t,e))];case"LogSoftmax":return[B.logSoftmax(A("x",n,t,e))];case"SparseToDense":return[B.sparseToDense(A("sparseIndices",n,t,e),A("outputShape",n,t,e),A("sparseValues",n,t,e),A("defaultValue",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var $W=function(n,t,e){switch(n.op){case"Max":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.max(A("x",n,t,e),i,r)]}case"Mean":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.mean(A("x",n,t,e),i,r)]}case"Min":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.min(A("x",n,t,e),i,r)]}case"Sum":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.sum(A("x",n,t,e),i,r)]}case"All":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.all(A("x",n,t,e),i,r)]}case"Any":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.any(A("x",n,t,e),i,r)]}case"ArgMax":{var i=A("axis",n,t,e);return[B.argMax(A("x",n,t,e),i)]}case"ArgMin":{var i=A("axis",n,t,e);return[B.argMin(A("x",n,t,e),i)]}case"Prod":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.prod(A("x",n,t,e),i,r)]}case"Cumsum":{var i=A("axis",n,t,e),a=A("exclusive",n,t,e),s=A("reverse",n,t,e);return[B.cumsum(A("x",n,t,e),i,a,s)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var XW=function(n,t,e){switch(n.op){case"ConcatV2":case"Concat":{var i=A("n",n,t,e),r=A("axis",n,t,e),a=A("tensors",n,t,e);return a=a.slice(0,i),[B.concat(a,r)]}case"GatherV2":case"Gather":{var r=A("axis",n,t,e),s=A("x",n,t,e),o=A("indices",n,t,e);return[B.gather(s,B.cast(o,"int32"),r)]}case"ReverseV2":case"Reverse":{var r=A("axis",n,t,e),s=A("x",n,t,e);return[B.reverse(s,r)]}case"Slice":{var l=A("begin",n,t,e),u=A("size",n,t,e);return[B.slice(A("x",n,t,e),l,u)]}case"StridedSlice":{var l=A("begin",n,t,e),c=A("end",n,t,e),h=A("strides",n,t,e),d=A("beginMask",n,t,e),p=A("endMask",n,t,e),f=A("ellipsisMask",n,t,e),m=A("newAxisMask",n,t,e),g=A("shrinkAxisMask",n,t,e),v=A("x",n,t,e);return[B.stridedSlice(v,l,c,h,d,p,f,m,g)]}case"Pack":return B.tidy(function(){var D=A("axis",n,t,e),k=A("tensors",n,t,e),W=k[0].shape,F=B.squeeze(k[0]).shape,P=k.map(function(H){var _=B.util.arraysEqual(H.shape,W);if(!_&&!B.util.arraysEqual(B.squeeze(H).shape,F))throw new Error("the input tensors shape does not match");return _?H:B.reshape(H,W)});return[B.stack(P,D)]});case"Unpack":{var r=A("axis",n,t,e),v=A("tensor",n,t,e);return B.unstack(v,r)}case"Tile":{var b=A("reps",n,t,e);return[B.tile(A("x",n,t,e),b)]}case"Split":case"SplitV":{var r=A("axis",n,t,e),w=A("numOrSizeSplits",n,t,e),v=A("x",n,t,e);return B.split(v,w,r)}case"ScatterNd":{var o=A("indices",n,t,e),S=A("values",n,t,e),L=A("shape",n,t,e);return[B.scatterND(o,S,L)]}case"GatherNd":{var x=A("x",n,t,e),o=A("indices",n,t,e);return[B.gatherND(x,o)]}case"SparseToDense":{var o=A("sparseIndices",n,t,e),L=A("outputShape",n,t,e),C=A("sparseValues",n,t,e),R=A("defaultValue",n,t,e);return[B.sparseToDense(o,C,L,C.dtype===R.dtype?R:B.cast(R,C.dtype))]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var JW=function(n,t,e){switch(n.op){case"FFT":return[B.fft(A("x",n,t,e))];case"IFFT":return[B.ifft(A("x",n,t,e))];case"RFFT":return[B.rfft(A("x",n,t,e))];case"IRFFT":return[B.irfft(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var ZW=function(n,t,e){switch(n.op){case"Cast":return[B.cast(A("x",n,t,e),A("dtype",n,t,e))];case"ExpandDims":{var i=A("axis",n,t,e);return[B.expandDims(A("x",n,t,e),i)]}case"Squeeze":{var i=A("axis",n,t,e);return[B.squeeze(A("x",n,t,e),i)]}case"Reshape":return[B.reshape(A("x",n,t,e),A("shape",n,t,e))];case"PadV2":case"Pad":return[B.pad(A("x",n,t,e),A("padding",n,t,e),A("constantValue",n,t,e))];case"SpaceToBatchND":{var r=A("blockShape",n,t,e),a=A("paddings",n,t,e);return[B.spaceToBatchND(A("x",n,t,e),r,a)]}case"BatchToSpaceND":{var r=A("blockShape",n,t,e),s=A("crops",n,t,e);return[B.batchToSpaceND(A("x",n,t,e),r,s)]}case"DepthToSpace":{var o=A("blockSize",n,t,e),l=A("dataFormat",n,t,e).toUpperCase();return[B.depthToSpace(A("x",n,t,e),o,l)]}case"BroadcastTo":return[B.broadcastTo(A("x",n,t,e),A("shape",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function tw(n,t,e){var i=function(r,a,s){switch(r.category){case"arithmetic":return B.tidy(function(){return RW(r,a,s)});case"basic_math":return B.tidy(function(){return OW(r,a,s)});case"control":return zW(r,a,s);case"convolution":return B.tidy(function(){return PW(r,a,s)});case"creation":return B.tidy(function(){return _W(r,a,s)});case"dynamic":return HW(r,a,s);case"evaluation":return B.tidy(function(){return VW(r,a,s)});case"image":return B.tidy(function(){return GW(r,a,s)});case"graph":return B.tidy(function(){return qW(r,a,s)});case"logical":return B.tidy(function(){return YW(r,a,s)});case"matrices":return B.tidy(function(){return KW(r,a,s)});case"normalization":return B.tidy(function(){return jW(r,a,s)});case"reduction":return B.tidy(function(){return $W(r,a,s)});case"slice_join":return B.tidy(function(){return XW(r,a,s)});case"spectral":return B.tidy(function(){return JW(r,a,s)});case"transformation":return B.tidy(function(){return ZW(r,a,s)});case"custom":var o=$b(r.op);if(o&&o.customExecutor)return o.customExecutor(new CW(r,a,s));throw TypeError("Custom op "+r.op+" is not registered.");default:throw TypeError("Unknown op '"+r.op+"'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()")}}(n,t,e);return i instanceof Promise?i.then(function(r){return[].concat(r)}):[].concat(i)}var nw=function(){function n(t,e,i,r){t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i={}),r===void 0&&(r={}),this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=i,this.functionMap=r,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}return n.prototype.newFrame=function(t,e){return{id:t,frameName:e,iterationId:0}},Object.defineProperty(n.prototype,"currentContext",{get:function(){return this.contexts},set:function(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextId",{get:function(){return this._currentContextIds[0]},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextIds",{get:function(){return this._currentContextIds},enumerable:!0,configurable:!0}),n.prototype.generateCurrentContextIds=function(){for(var t=[],e=0;e<this.contexts.length-1;e++){var i=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(i))}t.push(""),this._currentContextIds=t},n.prototype.contextIdforContexts=function(t){return t?t.map(function(e){return e.id===0&&e.iterationId===0?"":e.frameName+"-"+e.iterationId}).join("/"):""},n.prototype.enterFrame=function(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))},n.prototype.exitFrame=function(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function rw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return $t(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return $t(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((iw(d)||QW(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function eU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return $t(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var tU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],nU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function iw(n){return tU.indexOf(n.op)>=0}function QW(n){return nU.indexOf(n.op)>=0}var aw=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=rw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return eU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[$t(d)[0]]}),s=e.map(function(d){return $t(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new nw(i.weightMap,c,h,i.functionExecutorMap),p=Kb({},i.weightMap);Object.keys(t).forEach(function(w){var S=$t(w),L=S[0],x=S[1],C=[];C[x]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g<u.length;g++){var v=u[g];if(!p[v.name]){var b=tw(v,p,d);if(b instanceof Promise)throw new Error("The execution of the op '"+v.op+"' returned a promise. Please use model.executeAsync() instead.");p[v.name]=b,i.checkTensorForDisposal(v.name,v,p,d,f,s,m)}}return i.parent==null&&d.dispose(f),e.map(function(w){return Ht(w,p,d)})})},n.prototype.getFrozenTensorIds=function(t){var e=[].concat.apply([],Object.keys(t).map(function(i){return t[i]}).map(function(i){return i.map(function(r){return r.id})}));return new Set(e)},n.prototype.checkTensorForDisposal=function(t,e,i,r,a,s,o){if(e.category==="control"||s.indexOf(t)!==-1)return;i[t].forEach(function(l){l!=null&&(o[l.id]=(o[l.id]||0)+e.children.length)}),e.inputs.forEach(function(l){if(l.category!=="control"){var u=jF(l.name,i,r);u!=null&&u.forEach(function(c){if(c&&!a.has(c.id)){var h=o[c.id];h===1?(c.dispose(),delete o[c.id]):h!=null&&o[c.id]--}})}})},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){return Sn(this,function(i){return[2,this._executeAsync(t,e)]})})},n.prototype._executeAsync=function(t,e,i,r,a){return i===void 0&&(i=!1),r===void 0&&(r={}),a===void 0&&(a={}),zn(this,void 0,void 0,function(){var s,o,l,u,c,h;return Sn(this,function(d){switch(d.label){case 0:return i||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e)),s=new nw(this.weightMap,r,a,this.functionExecutorMap),[4,this.executeWithControlFlow(t,s,e,i)];case 1:return o=d.sent(),l=e.map(function(p){return Ht(p,o,s)}),u=l.map(function(p){return p.id}),c=Object.keys(t).map(function(p){return t[p].id}),h=new Set(u.concat(c,this.weightIds)),Object.keys(o).forEach(function(p){var f=o[p];f.forEach(function(m){m&&!m.isDisposed&&!h.has(m.id)&&m.dispose()})}),this.parent==null&&s.dispose(h),[2,l]}})})},n.prototype.executeFunctionAsync=function(t,e,i){return zn(this,void 0,void 0,function(){var r,a=this;return Sn(this,function(s){return r=t.reduce(function(o,l,u){return o[a.inputs[u].name]=l,o},{}),[2,this._executeAsync(r,this.outputNodes,!0,e,i)]})})},n.prototype.executeWithControlFlow=function(t,e,i,r){return zn(this,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x=this;return Sn(this,function(C){switch(C.label){case 0:a=Object.keys(t),s=a.map(function(R){return x.graph.nodes[$t(R)[0]]}),o=i.map(function(R){return $t(R)[0]}),l=o.map(function(R){return x.graph.nodes[R]}),u=rw(t,l,this.weightMap),c=u.usedNodes,h=u.missingInputs,d=u.dynamicNode,p=u.syncInputs,f=s.concat(this.graph.weights).map(function(R){return{node:R,contexts:e.currentContext}}),m=Kb({},this.weightMap),Object.keys(t).forEach(function(R){var D=$t(R),k=D[0],W=D[1],F=[];F[W]=t[R],m[k]=F}),g={},v=this.getFrozenTensorIds(m),b={},C.label=1;case 1:return f.length>0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!iw(R)&&!Ht(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=Zn(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=tw(f.node,r,i);m||(m=Zn(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=Zn(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=$t(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=$t(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=$t(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var iU="?tfjs-format=file",rU="model.json",sw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return zn(this,void 0,void 0,function(){var t;return Sn(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new aw(Jb.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Jb.Instance.transformGraph(t.modelInitializer);this.initializer=new aw(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return zn(this,void 0,void 0,function(){var i;return Sn(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){var i;return Sn(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function aU(n,t){return t===void 0&&(t={}),zn(this,void 0,void 0,function(){var e;return Sn(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+rU+iU)),e=new sw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var sU="2.6.0";rr.GraphModel=sw;rr.deregisterOp=KF;rr.loadGraphModel=aU;rr.registerOp=YF;rr.version_converter=sU});var lw=be(()=>{});var Cw=be(Vt=>{"use strict";Object.defineProperty(Vt,"__esModule",{value:!0});var Ne=Xi();var ud=function(n,t){return ud=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},ud(n,t)};function Ve(n,t){ud(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var jr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function ar(n,t){return t={exports:{}},n(t,t.exports),t.exports}var oU=ar(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(jr,n,!1)}),lU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(jr,n,!1)}),uU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(jr,n,!1)}),cU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(jr,n,!1)}),hU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(jr,n,!1)}),dU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(jr,n,!1)}),sr=ar(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});sr.alea=oU;sr.xor128=lU;sr.xorwow=uU;sr.xorshift7=cU;sr.xor4096=hU;sr.tychei=dU;var pU=sr,uw=pU.alea;function fU(n,t){return xo(n,t)}function xo(n,t,e,i){if(e===void 0&&(e=new Map),i===void 0&&(i=new Set),n==null)return null;if(i.has(n))throw new Error("Circular references are not supported.");if(e.has(n))return e.get(n);var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if($r(n)){var a=Array.isArray(n)?[]:{};i.add(n);for(var s in n){var o=n[s],l=xo(o,t,e,i);a[s]=l}return i.delete(n),a}else throw new Error("Can't recurse into non-iterable type: "+n);else return e.set(n,r.value),r.value}function mU(n,t){return t===void 0&&(t=hw),cw(n,t)}function cw(n,t,e){e===void 0&&(e=new Set);var i=n[0];if(e.has(i))throw new Error("Circular references are not supported.");var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if($r(i)){var a=Array.isArray(i)?[]:{};e.add(i);var s=function(l){var u=n.map(function(h){return h[l]}),c=cw(u,t,e);a[l]=c};for(var o in i)s(o);return e.delete(i),a}else throw new Error("Can't recurse into non-iterable type: "+i);else return r.value}function hw(n){return n===null?null:$r(n[0])?{value:null,recurse:!0}:{value:n,recurse:!1}}function dw(n,t){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:e=new Map,xo(n,t,e),i=0,r=Array.from(e.keys()),u.label=1;case 1:return i<r.length?(a=r[i],s=e.get(a),s instanceof Promise?[4,s]:[3,3]):[3,4];case 2:o=u.sent(),e.set(a,o),u.label=3;case 3:return i++,[3,1];case 4:return l=xo(n,t,e),[2,l]}})})}function $r(n){return n!=null&&!ArrayBuffer.isView(n)&&(Array.isArray(n)||typeof n=="object"&&!(n instanceof Ne.Tensor))}function vU(n){return n==null||gU(n)||Array.isArray(n)||typeof n=="object"&&n instanceof Ne.Tensor||Ne.util.isTypedArray(n)}function gU(n){return n===null||typeof n!="object"&&typeof n!="function"}function bU(n){return fU(n,yU)}function yU(n){return n instanceof Ne.Tensor?{value:n.clone(),recurse:!1}:$r(n)?{value:null,recurse:!0}:{value:n,recurse:!1}}var pw=function(){function n(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}return n.prototype.wrap=function(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity},n.prototype.get=function(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]},n.prototype.set=function(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e},n.prototype.length=function(){var t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t},n.prototype.isFull=function(){return this.length()===this.capacity},n.prototype.isEmpty=function(){return this.length()===0},n.prototype.push=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)},n.prototype.pushAll=function(t){for(var e=0,i=t;e<i.length;e++){var r=i[e];this.push(r)}},n.prototype.pop=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);var t=this.get(this.end);return this.set(this.end,void 0),t},n.prototype.unshift=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)},n.prototype.shift=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t},n.prototype.shuffleExcise=function(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var e=this.wrap(this.begin+t),i=this.get(e);return this.set(e,this.pop()),i},n}();var wU=function(n){Ve(t,n);function t(){return n.call(this,t.INITIAL_CAPACITY)||this}return t.prototype.isFull=function(){return!1},t.prototype.push=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.push.call(this,e)},t.prototype.unshift=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.unshift.call(this,e)},t.prototype.expand=function(){for(var e=this.capacity*2,i=new Array(e),r=this.length(),a=0;a<r;a++)i[a]=this.get(this.wrap(this.begin+a));this.data=i,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=r},t.INITIAL_CAPACITY=32,t}(pw);function fw(n){return new SU(n)}function cd(n){return new LU(n)}function IU(n,t){return new mw(n,t)}function TU(n,t){return t===void 0&&(t=Ii.FAIL),new AU(n,t)}var Nt=function(){function n(){}return n.prototype.toArray=function(){return se(this,void 0,void 0,function(){var t,e;return oe(this,function(i){switch(i.label){case 0:return t=[],[4,this.next()];case 1:e=i.sent(),i.label=2;case 2:return e.done?[3,4]:(t.push(e.value),[4,this.next()]);case 3:return e=i.sent(),[3,2];case 4:return[2,t]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){var t,e,i;return oe(this,function(r){switch(r.label){case 0:return t=this.prefetch(100),e=[],[4,t.next()];case 1:i=r.sent(),r.label=2;case 2:return i.done?[3,4]:(e.push(i.value),[4,t.next()]);case 3:return i=r.sent(),[3,2];case 4:return[2,e]}})})},n.prototype.resolveFully=function(){return se(this,void 0,void 0,function(){var t;return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:t=e.sent(),e.label=2;case 2:return t.done?[3,4]:[4,this.next()];case 3:return t=e.sent(),[3,2];case 4:return[2]}})})},n.prototype.resolveWhile=function(t){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:return[4,this.next()];case 1:e=r.sent(),i=t(e.value),r.label=2;case 2:return!e.done&&i?[4,this.next()]:[3,4];case 3:return e=r.sent(),i=t(e.value),[3,2];case 4:return[2]}})})},n.prototype.handleErrors=function(t){return new DU(this,t)},n.prototype.filter=function(t){return new OU(this,t)},n.prototype.map=function(t){return new EU(this,t)},n.prototype.mapAsync=function(t){return new gw(this,t)},n.prototype.serialMapAsync=function(t){return new gw(this,t).serial()},n.prototype.flatmap=function(t){return new kU(this,t)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.map(t).resolveFully()]})})},n.prototype.serialForEach=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.serialMapAsync(t).resolveWhile(function(i){return i===!0})]})})},n.prototype.rowMajorBatch=function(t,e){return e===void 0&&(e=!0),new RU(this,t,e)},n.prototype.columnMajorBatch=function(t,e,i){e===void 0&&(e=!0),i===void 0&&(i=hw);var r=this.rowMajorBatch(t,e);return r.map(function(a){return mU(a,i)})},n.prototype.concatenate=function(t,e){return new mw(fw([this,t]),e)},n.prototype.take=function(t){return t<0||t==null?this:new CU(this,t)},n.prototype.skip=function(t){return t<0||t==null?this:new xU(this,t)},n.prototype.prefetch=function(t){return new vw(this,t)},n.prototype.shuffle=function(t,e){return new FU(this,t,e)},n.prototype.serial=function(){return new NU(this)},n}(),SU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.items=e,i.trav=0,i}return t.prototype.summary=function(){return"Array of "+this.items.length+" items"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return this.trav>=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:bU(e),done:!1}])})})},t}(Nt),LU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(Nt),NU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(Nt),xU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++<this.maxCount?[4,this.upstream.next()]:[3,2];case 1:return e=i.sent(),e.done?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2,this.upstream.next()]}})})},t}(Nt),CU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(Nt),RU=function(n){Ve(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length<this.batchSize?[4,this.upstream.next()]:[3,3];case 2:return i=r.sent(),i.done?this.enableSmallLastBatch&&e.length>0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(Nt),OU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(Nt),EU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),DU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.handler=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(Nt),gw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),hd=function(n){Ve(t,n);function t(){var e=n.call(this)||this;return e.outputQueue=new wU,e.lastRead=Promise.resolve({value:null,done:!1}),e}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.outputQueue.length()===0?[4,this.pump()]:[3,2];case 1:return e.sent()?[3,0]:[2,{value:null,done:!0}];case 2:return[2,{value:this.outputQueue.shift(),done:!1}]}})})},t}(Nt),kU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,!0]}})})},t}(hd),mw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.baseErrorHandler=i,r.lastRead=null,r.iterator=null,r.moreIterators=e,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of chained summaries";return e+" -> Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(Nt),Ii;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Ii||(Ii={}));var AU=function(n){Ve(t,n);function t(e,i){i===void 0&&(i=Ii.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof Nt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,dw(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Ii.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Ii.SHORTEST:return[2,{value:null,done:!0}];case Ii.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(Nt),vw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new pw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(Nt),FU=function(n){Ve(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=uw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(vw);var Ya=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is
-      `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),Xt(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,WU)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,Xt(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,Xt(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=cd(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,IU(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?r=0:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().skip(t)]}})})},r)},n.prototype.shuffle=function(t,e,i){var r=this;if(i===void 0&&(i=!0),t==null||t<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError("`Dataset.shuffle()` requires bufferSize to be specified.  If your data fits in main memory (for regular JS objects), and/or GPU memory (for `tf.Tensor`s), consider setting "+("bufferSize to the dataset size ("+this.size+" elements)"));var a=this,s=uw(e||Ne.util.now().toString());return Xt(function(){return se(r,void 0,void 0,function(){var o;return oe(this,function(l){switch(l.label){case 0:return o=s.int32(),i&&(o+=s.int32()),[4,a.iterator()];case 1:return[2,l.sent().shuffle(t,o.toString())]}})})},this.size)},n.prototype.take=function(t){var e=this,i=this,r;return this.size!=null&&this.size>t?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function Xt(n,t){return t===void 0&&(t=null),new(function(e){Ve(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(Ya))}function UU(n){var t=this;return Xt(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,fw(n)]})})},n.length)}function BU(n){var t=this;if(!$r(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i<n.length;i++)e=e==null?n[i].size:Math.min(e,n[i].size);else if(n instanceof Object)for(var r in n)e=e==null?n[r].size:Math.min(e,n[r].size);return Xt(function(){return se(t,void 0,void 0,function(){var a;return oe(this,function(s){switch(s.label){case 0:return[4,dw(n,function(o){if(o instanceof Ya)return{value:o.iterator(),recurse:!1};if($r(o))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")})];case 1:return a=s.sent(),[2,TU(a,Ii.SHORTEST)]}})})},e)}function WU(n){if(n===null)return null;var t=n[0];if(vU(t)){var e=zU(n);return{value:e,recurse:!1}}return{value:null,recurse:!0}}function zU(n){if(n.length===0)throw new Error("Can't make a batch of zero elements.");return n[0]instanceof Ne.Tensor?Ne.stack(n):Ne.tensor(n)}var yw=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.input=e,i}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.input.iterator()];case 1:return e=a.sent(),i=e.decodeUTF8(),r=i.split(`
-`).map(function(s){return s.endsWith("\r")&&(s=s.slice(0,-1)),s}),[2,r]}})})},t}(Ya);var Co='"',Ka=Symbol("out"),bw=Symbol("field"),Ro=Symbol("quote"),dd=Symbol("quoteafterquote"),ww=Symbol("quoteinquote"),Sw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.input=e,r.hasHeader=!0,r.fullColumnNames=null,r.columnNamesValidated=!1,r.columnConfigs=null,r.configuredColumnsOnly=!1,r.delimiter=",",r.delimWhitespace=!1,r.base=new yw(e),i||(i={}),r.hasHeader=!(i.hasHeader===!1),r.fullColumnNames=i.columnNames,r.columnConfigs=i.columnConfigs,r.configuredColumnsOnly=i.configuredColumnsOnly,i.delimWhitespace?(Ne.util.assert(i.delimiter==null,function(){return"Delimiter should not be provided when delimWhitespace is true."}),r.delimWhitespace=!0,r.delimiter=" "):r.delimiter=i.delimiter?i.delimiter:",",r}return t.prototype.columnNames=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:e.sent(),e.label=2;case 2:return[2,this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames]}})})},t.prototype.setColumnNames=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u=this;return oe(this,function(c){switch(c.label){case 0:return[4,this.maybeReadHeaderLine()];case 1:if(e=c.sent(),!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");if(this.fullColumnNames&&e&&Ne.util.assert(e.length===this.fullColumnNames.length,function(){return"The length of provided columnNames ("+u.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."}),this.fullColumnNames||(this.fullColumnNames=e),i=this.fullColumnNames.reduce(function(h,d){return h[d]=h[d]+1||1,h},{}),r=Object.keys(i).filter(function(h){return i[h]>1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a<s.length;a++)if(o=s[a],l=this.fullColumnNames.indexOf(o),l===-1)throw new Error('The key "'+o+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}return this.columnNamesValidated=!0,[2]}})})},t.prototype.maybeReadHeaderLine=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return this.hasHeader?[4,this.base.iterator()]:[3,3];case 1:return e=s.sent(),[4,e.next()];case 2:if(i=s.sent(),i.done)throw new Error("No data was found for CSV parsing.");return r=i.value,a=this.parseRow(r,!1),[2,a];case 3:return[2,null]}})})},t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i=this;return oe(this,function(r){switch(r.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:r.sent(),r.label=2;case 2:return[4,this.base.iterator()];case 3:return e=r.sent(),this.hasHeader&&(e=e.skip(1)),[2,e.map(function(a){return i.makeDataElement(a)})]}})})},t.prototype.makeDataElement=function(e){for(var i=this.parseRow(e),r={},a={},s=0;s<this.fullColumnNames.length;s++){var o=this.fullColumnNames[s],l=this.columnConfigs?this.columnConfigs[o]:null;if(this.configuredColumnsOnly&&!l)continue;var u=i[s],c=null;if(u==="")if(l&&l.default!==void 0)c=l.default;else{if(l&&(l.required||l.isLabel))throw new Error("Required column "+o+" is empty in this line: "+e);c=void 0}else{var h=Number(u);if(isNaN(h))l&&l.dtype==="bool"?c=this.getBoolean(u):c=u;else if(!l||!l.dtype)c=h;else switch(l.dtype){case"float32":c=h;break;case"int32":c=Math.floor(h);break;case"bool":c=this.getBoolean(u);break;default:c=h}}l&&l.isLabel?a[o]=c:r[o]=c}return Object.keys(a).length===0?r:{xs:r,ys:a}},t.prototype.getBoolean=function(e){return e==="1"||e.toLowerCase()==="true"?1:0},t.prototype.parseRow=function(e,i){i===void 0&&(i=!0);for(var r=[],a=0,s=e.length,o=Ka,l=0;l<s;l++)switch(o){case Ka:switch(e.charAt(l)){case Co:a=l+1,o=Ro;break;case this.delimiter:if(a=l+1,this.delimiter===" "&&this.delimWhitespace)break;r.push(""),o=Ka;break;default:o=bw,a=l;break}break;case bw:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l)),o=Ka,a=l+1;break}break;case Ro:switch(e.charAt(l)){case Co:o=dd;break}break;case dd:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l-1)),o=Ka,a=l+1;break;case Co:o=Ro;break;default:o=ww;break}break;case ww:switch(e.charAt(l)){case Co:o=Ro;break}break}if(o===dd?r.push(e.substring(a,s-1)):r.push(e.substring(a)),i&&r.length!==this.fullColumnNames.length)throw new Error("Invalid row in csv file. Should have "+this.fullColumnNames.length+" elements in a row, but got "+r);return r},t}(Ya);var PU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;i.microphoneConfig=e,i.isClosed=!1,i.fftSize=e.fftSize||1024;var r=Math.log2(i.fftSize);if(i.fftSize<0||r<4||r>14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(Nt);var _U=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(Nt);var Lw=function(){function n(){}return n}();var Iw=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new MU(this,e)},t}(Nt),MU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new HU(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),HU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r<a.length;r++)s=a[r],this.outputQueue.push(s);return this.carryover=i[i.length-1],[2,!0]}})})},t}(hd);var qU=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.decodeUTF8=function(){return new VU(this)},t}(Nt),VU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.impl=new GU(e),i}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),GU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;if(i.upstream=e,Ne.env().get("IS_BROWSER"))i.decoder=new TextDecoder("utf-8");else{var r=lw().StringDecoder;i.decoder=new r("utf8")}return i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(hd);var Aw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(qU);function KU(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=YU(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new Aw(a,t)];case 3:throw new Error(r.statusText)}})})}var YU=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function Tw(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var Nw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return Tw(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new Aw(this.input,this.options)]})})},t}(Lw);var xw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return Tw(this.url)?[2,new Nw(this.url,this.fileOptions).iterator()]:[2,KU(this.url,this.fileOptions)]})})},t}(Lw);function jU(n,t){return t===void 0&&(t={}),new Sw(new xw(n),t)}function $U(n){var t=this,e=cd(n);return Xt(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function XU(n){var t=this;return Xt(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,cd(function(){return e.next()})]}})})})}function JU(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,_U.create(n,t)]})})}function ZU(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,PU.create(n)]})})}var QU="2.6.0";Vt.CSVDataset=Sw;Vt.Dataset=Ya;Vt.FileDataSource=Nw;Vt.TextLineDataset=yw;Vt.URLDataSource=xw;Vt.array=UU;Vt.csv=jU;Vt.func=$U;Vt.generator=XU;Vt.microphone=ZU;Vt.version_data=QU;Vt.webcam=JU;Vt.zip=BU});var Ow=be((Rw,pd)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c<u.length;c++){o+=u.charCodeAt(c);var h=.02519603282416938*o;o=h>>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(Rw,typeof pd=="object"&&pd,typeof define=="function"&&define)});var Dw=be((Ew,fd)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(Ew,typeof fd=="object"&&fd,typeof define=="function"&&define)});var Fw=be((kw,md)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,u==l.length&&(o.d=o.x<<10^o.x>>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(kw,typeof md=="object"&&md,typeof define=="function"&&define)});var Uw=be((Ww,gd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h<c.length;++h)p[h&7]=p[h&7]<<15^c.charCodeAt(h)+p[h+1&7]<<13;for(;p.length<8;)p.push(0);for(h=0;h<8&&p[h]===0;++h);for(h==8?d=p[7]=-1:d=p[h],u.x=p,u.i=0,h=256;h>0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(Ww,typeof gd=="object"&&gd,typeof define=="function"&&define)});var zw=be((Bw,vd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f<v;++f)c&&(d^=c.charCodeAt((f+32)%c.length)),f===0&&(m=d),d^=d<<10,d^=d>>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(Bw,typeof vd=="object"&&vd,typeof define=="function"&&define)});var _w=be((Pw,yd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u<l.length+20;u++)o.b^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.a=s.a,o.b=s.b,o.c=s.c,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(Pw,typeof yd=="object"&&yd,typeof define=="function"&&define)});var Mw=be((CH,Oo)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var x=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),x),R=new p(x),D=function(){for(var k=R.g(r),W=o,F=0;k<l;)k=(k+F)*i,W*=i,F=R.g(1);for(;k>=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,x=this,C=0,R=x.i=x.j=0,D=x.S=[];for(L||(w=[L++]);C<i;)D[C]=C++;for(C=0;C<i;C++)D[C]=D[R=c&R+w[C%L]+(S=D[C])],D[R]=S;(x.g=function(k){for(var W,F=0,P=x.i,H=x.j,_=x.S;k--;)W=_[P=c&P+1],F=F*i+_[c&(_[P]=_[H=c&H+W])+(_[H]=W)];return x.i=P,x.j=H,F})(i)}function f(w,S){return S.i=w.i,S.j=w.j,S.S=w.S.slice(),S}function m(w,S){var L=[],x=typeof w,C;if(S&&x=="object")for(C in w)try{L.push(m(w[C],S-1))}catch(R){}return L.length?L:x=="string"?w:w+"\0"}function g(w,S){for(var L=w+"",x,C=0;C<L.length;)S[c&C]=c&(x^=S[c&C]*19)+L.charCodeAt(C++);return b(S)}function v(){try{var w;return h&&(w=h.randomBytes)?w=w(i):(w=new Uint8Array(i),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(x){var S=e.navigator,L=S&&S.plugins;return[+new Date,e,L,e.screen,b(n)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),n),typeof Oo=="object"&&Oo.exports){Oo.exports=d;try{h=ds()}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return d})})([],Math)});var Vw=be((RH,Hw)=>{var eB=Ow(),tB=Dw(),nB=Fw(),iB=Uw(),rB=zw(),aB=_w(),or=Mw();or.alea=eB;or.xor128=tB;or.xorwow=nB;or.xorshift7=iB;or.xor4096=rB;or.tychei=aB;Hw.exports=or});var m0=be(ja=>{"use strict";Object.defineProperty(ja,"__esModule",{value:!0});var T=Xi(),sB=Vw();var bd=function(n,t){return bd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},bd(n,t)};function oB(n,t){bd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function qw(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function Gw(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}function ae(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&T.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the CPU backend."})})}var lB=T.kernel_impls.nonMaxSuppressionV3Impl,uB=T.kernel_impls.split,cB=T.kernel_impls.tile,hB=T.kernel_impls.topkImpl,dB=T.kernel_impls.whereImpl;function wd(n,t,e,i){if(e==="linear")return n.linear(t);if(e==="relu")return n.relu(t);if(e==="elu")return T.elu(t);if(e==="relu6")return n.relu6(t);if(e==="prelu")return n.prelu(t,i);throw new Error("Activation "+e+" has not been implemented for the CPU backend.")}var Yw=function(n){oB(t,n);function t(){var e=n.call(this)||this;return e.blockSize=48,e.firstUse=!0,e.data=new T.DataStorage(e,T.engine()),e}return t.prototype.write=function(e,i,r){this.firstUse&&(this.firstUse=!1,T.env().get("IS_NODE")&&T.backend_util.warn(`
+        `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c<t.length;++c){var h=c===0?0:r[c-1],d=[0,h,0],p=[1,t[c],a];u[c]=B.reshape(B.slice(n,d,p),e)}return n.dispose(),u}),o=new No([],e,n.dtype,t.length),l=0;l<s.length;l++)o.setItem(l,s[l]);return o}var BW=void 0,zW=function(n,t,e){return zn(BW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,o,h,d,l,p,f,m,m,g,v,g,b,g,g,g,w,S,L,x,C,R,D,k,W,F,P,H,_,K,j,q,G,Z,X,ee,ne,ie,te,re,le,he,ye,Oe,Fe,_e,Me,ht,St,Ge,Ue,F,P,me,Ue,K,L,Dt,me,ne,ie,L,lt,me,L,dt,lt,me,q,G,L,dt,me,Ue,L,dt,lt,me,xn,L,dt,me,re,me,he,L,Ue,P,me,Ue,L,Dt,me,Oe,L,Fe,me;return wn(this,function(Rt){switch(Rt.label){case 0:i=n.op;switch(i){case"If":return[3,1];case"StatelessIf":return[3,1];case"While":return[3,3];case"StatelessWhile":return[3,3];case"LoopCond":return[3,9];case"Switch":return[3,10];case"Merge":return[3,12];case"Enter":return[3,13];case"Exit":return[3,14];case"NextIteration":return[3,15];case"TensorArrayV3":return[3,16];case"TensorArrayWriteV3":return[3,17];case"TensorArrayReadV3":return[3,18];case"TensorArrayGatherV3":return[3,19];case"TensorArrayScatterV3":return[3,20];case"TensorArrayConcatV3":return[3,21];case"TensorArraySplitV3":return[3,22];case"TensorArraySizeV3":return[3,23];case"TensorArrayCloseV3":return[3,24];case"TensorListSetItem":return[3,25];case"TensorListGetItem":return[3,26];case"TensorListScatterV2":return[3,27];case"TensorListScatter":return[3,27];case"TensorListReserve":return[3,28];case"TensorListGather":return[3,29];case"TensorListStack":return[3,30];case"TensorListFromTensor":return[3,31];case"TensorListConcat":return[3,32];case"TensorListPushBack":return[3,33];case"TensorListPopBack":return[3,34];case"TensorListSplit":return[3,35]}return[3,36];case 1:return r=A("thenBranch",n,t,e),a=A("elseBranch",n,t,e),s=A("cond",n,t,e),o=A("args",n,t,e),[4,s.data()];case 2:return l=Rt.sent(),l[0]?[2,e.functionMap[r].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)]:[2,e.functionMap[a].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 3:return u=A("body",n,t,e),c=A("cond",n,t,e),o=A("args",n,t,e),[4,e.functionMap[c].executeFunctionAsync(o,e.tensorArrayMap,e.tensorListMap)];case 4:return h=Rt.sent(),d=o.map(function(Lt){return Lt.id}),[4,h[0].data()];case 5:l=Rt.sent(),h.forEach(function(Lt){!Lt.kept&&d.indexOf(Lt.id)===-1&&Lt.dispose()}),p=o,f=function(){var Lt,Ce,kt;return wn(this,function(pt){switch(pt.label){case 0:return Lt=p,[4,e.functionMap[u].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 1:return p=pt.sent(),Ce=p.map(function(it){return it.id}),Lt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[4,e.functionMap[c].executeFunctionAsync(p,e.tensorArrayMap,e.tensorListMap)];case 2:return kt=pt.sent(),[4,kt[0].data()];case 3:return l=pt.sent(),kt.forEach(function(it){!it.kept&&d.indexOf(it.id)===-1&&Ce.indexOf(it.id)===-1&&it.dispose()}),[2]}})},Rt.label=6;case 6:return l[0]?[5,f()]:[3,8];case 7:return Rt.sent(),[3,6];case 8:return[2,p];case 9:return m=A("pred",n,t,e),[2,[Qn(m)]];case 10:return m=A("pred",n,t,e),g=A("data",n,t,e),g.kept||(g=Qn(g)),[4,m.data()];case 11:return[2,Rt.sent()[0]?[void 0,g]:[g,void 0]];case 12:return v=n.inputNames.find(function(Lt){return Ht(Lt,t,e)!==void 0}),v?(g=Ht(v,t,e),[2,[Qn(g)]]):[2,void 0];case 13:return b=A("frameName",n,t,e),g=A("tensor",n,t,e),e.enterFrame(b),[2,[Qn(g)]];case 14:return g=A("tensor",n,t,e),e.exitFrame(),[2,[Qn(g)]];case 15:return g=A("tensor",n,t,e),e.nextIteration(),[2,[Qn(g)]];case 16:return w=A("size",n,t,e),S=A("dtype",n,t,e),L=A("elementShape",n,t,e),x=A("dynamicSize",n,t,e),C=A("clearAfterRead",n,t,e),R=A("identicalElementShapes",n,t,e),D=A("name",n,t,e),k=new DW(D,S,w,L,R,x,C),e.addTensorArray(k),[2,[k.idTensor,B.scalar(1)]];case 17:return W=A("tensorArrayId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),H=e.getTensorArray(W.id),H.write(F,P),[2,[H.idTensor]];case 18:return _=A("tensorArrayId",n,t,e),K=A("index",n,t,e),j=e.getTensorArray(_.id),[2,[j.read(K)]];case 19:return q=A("tensorArrayId",n,t,e),G=A("indices",n,t,e),Z=A("dtype",n,t,e),X=e.getTensorArray(q.id),[2,[X.gather(G,Z)]];case 20:return ee=A("tensorArrayId",n,t,e),ne=A("indices",n,t,e),ie=A("tensor",n,t,e),te=e.getTensorArray(ee.id),te.scatter(ne,ie),[2,[te.idTensor]];case 21:return re=A("tensorArrayId",n,t,e),le=e.getTensorArray(re.id),he=A("dtype",n,t,e),[2,[le.concat(he)]];case 22:return ye=A("tensorArrayId",n,t,e),Oe=A("tensor",n,t,e),Fe=A("lengths",n,t,e),_e=e.getTensorArray(ye.id),_e.split(Fe,Oe),[2,[_e.idTensor]];case 23:return Me=A("tensorArrayId",n,t,e),ht=e.getTensorArray(Me.id),[2,[B.scalar(ht.size(),"int32")]];case 24:return St=A("tensorArrayId",n,t,e),Ge=e.getTensorArray(St.id),Ge.clearAndClose(),[2,[Ge.idTensor]];case 25:return Ue=A("tensorListId",n,t,e),F=A("index",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.setItem(F,P),[2,[me.idTensor]];case 26:return Ue=A("tensorListId",n,t,e),K=A("index",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.getItem(K,L,Dt)]];case 27:return ne=A("indices",n,t,e),ie=A("tensor",n,t,e),L=A("elementShape",n,t,e),lt=A("numElements",n,t,e),me=WW(ie,ne,L,lt),e.addTensorList(me),[2,[me.idTensor]];case 28:return L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=FW(L,dt,lt),e.addTensorList(me),[2,[me.idTensor]];case 29:return q=A("tensorListId",n,t,e),G=A("indices",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=e.getTensorList(q.id),[2,[me.gather(G,dt,L)]];case 30:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),lt=A("numElements",n,t,e),me=e.getTensorList(Ue.id),[2,[me.stack(L,dt,lt)]];case 31:return xn=A("tensor",n,t,e),L=A("elementShape",n,t,e),dt=A("elementDType",n,t,e),me=kW(xn,L,dt),e.addTensorList(me),[2,[me.idTensor]];case 32:return re=A("tensorListId",n,t,e),me=e.getTensorList(re.id),he=A("dtype",n,t,e),L=A("elementShape",n,t,e),[2,[me.concat(he,L)]];case 33:return Ue=A("tensorListId",n,t,e),P=A("tensor",n,t,e),me=e.getTensorList(Ue.id),me.pushBack(P),[2,[me.idTensor]];case 34:return Ue=A("tensorListId",n,t,e),L=A("elementShape",n,t,e),Dt=A("elementDType",n,t,e),me=e.getTensorList(Ue.id),[2,[me.popBack(L,Dt)]];case 35:return Oe=A("tensor",n,t,e),L=A("elementShape",n,t,e),Fe=A("lengths",n,t,e),me=UW(Oe,Fe,L),e.addTensorList(me),[2,[me.idTensor]];case 36:throw TypeError("Node type "+n.op+" is not implemented")}})})};function ew(n,t,e){var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=r==="fusedbatchnorm",u=A("numArgs",n,t,e);if(s){if(o&&u!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&u!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");var c=A("strides",n,t,e),h=To(n,t,e),d=A("dataFormat",n,t,e).toUpperCase(),p=A("dilations",n,t,e),f=A("args",n,t,e),m=f[0],g=f[1];return{stride:c,pad:h,dataFormat:d,dilations:p,biasArg:m,preluArg:g,activationFunc:a}}var PW=function(n,t,e){switch(n.op){case"Conv1D":{var i=A("stride",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),s=A("dilation",n,t,e);return[B.conv1d(A("x",n,t,e),A("filter",n,t,e),i,r,a,s)]}case"Conv2D":{var i=A("strides",n,t,e),r=To(n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv2d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{var l=ew(n,t,e),i=l.stride,r=l.pad,a=l.dataFormat,o=l.dilations,u=l.biasArg,c=l.preluArg,h=l.activationFunc;return[B.fused.conv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"FusedDepthwiseConv2dNative":{var d=ew(n,t,e),i=d.stride,r=d.pad,a=d.dataFormat,o=d.dilations,u=d.biasArg,c=d.preluArg,h=d.activationFunc;return[B.fused.depthwiseConv2d({x:A("x",n,t,e),filter:A("filter",n,t,e),strides:[i[1],i[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:u,activation:h,preluActivationWeights:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{var p=A("outputShape",n,t,e),i=A("strides",n,t,e),r=To(n,t,e);return[B.conv2dTranspose(A("x",n,t,e),A("filter",n,t,e),p,[i[1],i[2]],r)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{var i=A("strides",n,t,e),r=To(n,t,e),o=A("dilations",n,t,e),a=A("dataFormat",n,t,e).toUpperCase();return[B.depthwiseConv2d(A("input",n,t,e),A("filter",n,t,e),[i[1],i[2]],r,a,[o[1],o[2]])]}case"Conv3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),a=A("dataFormat",n,t,e).toUpperCase(),o=A("dilations",n,t,e);return[B.conv3d(A("x",n,t,e),A("filter",n,t,e),[i[1],i[2],i[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPool":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r)]}case"MaxPoolWithArgmax":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e),m=A("includeBatchInIndex",n,t,e),g=B.maxPoolWithArgmax(A("x",n,t,e),[f[1],f[2]],[i[1],i[2]],r,m),v=g.result,b=g.indexes;return[v,b]}case"AvgPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.avgPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"MaxPool3D":{var i=A("strides",n,t,e),r=A("pad",n,t,e),f=A("kernelSize",n,t,e);return[B.maxPool3d(A("x",n,t,e),[f[1],f[2],f[3]],[i[1],i[2],i[3]],r)]}case"Dilation2D":{var w=A("strides",n,t,e),r=A("pad",n,t,e),o=A("dilations",n,t,e),S=w[1],L=w[2],x=o[1],C=o[2];return[B.dilation2d(A("x",n,t,e),A("filter",n,t,e),[S,L],r,[x,C],"NHWC")]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var _W=function(n,t,e){switch(n.op){case"Fill":{var i=A("shape",n,t,e),r=A("dtype",n,t,e),a=A("value",n,t,e);return[B.fill(i,a,r)]}case"LinSpace":{var s=A("start",n,t,e),o=A("stop",n,t,e),l=A("num",n,t,e);return[B.linspace(s,o,l)]}case"Multinomial":{var u=A("logits",n,t,e),c=A("numSamples",n,t,e),h=A("seed",n,t,e);return[B.multinomial(u,c,h)]}case"OneHot":{var d=A("indices",n,t,e),p=A("depth",n,t,e),f=A("onValue",n,t,e),m=A("offValue",n,t,e);return[B.oneHot(d,p,f,m)]}case"Ones":return[B.ones(A("shape",n,t,e),A("dtype",n,t,e))];case"OnesLike":return[B.onesLike(A("x",n,t,e))];case"RandomUniform":return[B.randomUniform(A("shape",n,t,e),A("minval",n,t,e),A("maxval",n,t,e),A("dtype",n,t,e))];case"Range":{var s=A("start",n,t,e),g=A("stop",n,t,e),v=A("step",n,t,e);return[B.range(s,g,v,A("dtype",n,t,e))]}case"TruncatedNormal":{var i=A("shape",n,t,e),b=A("mean",n,t,e),w=A("stdDev",n,t,e),h=A("seed",n,t,e);return[B.truncatedNormal(i,b,w,A("dtype",n,t,e),h)]}case"Zeros":return[B.zeros(A("shape",n,t,e),A("dtype",n,t,e))];case"ZerosLike":return[B.zerosLike(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var MW=void 0;function ud(n,t,e){var i=A("boxes",n,t,e),r=A("scores",n,t,e),a=A("maxOutputSize",n,t,e),s=A("iouThreshold",n,t,e),o=A("scoreThreshold",n,t,e),l=A("softNmsSigma",n,t,e);return{boxes:i,scores:r,maxOutputSize:a,iouThreshold:s,scoreThreshold:o,softNmsSigma:l}}var HW=function(n,t,e){return zn(MW,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,a,s,o,l,u,p,h,f,a,s,o,l,u,m,h;return wn(this,function(g){switch(g.label){case 0:i=n.op;switch(i){case"NonMaxSuppressionV5":return[3,1];case"NonMaxSuppressionV4":return[3,3];case"NonMaxSuppressionV3":return[3,5];case"NonMaxSuppressionV2":return[3,5];case"Where":return[3,7];case"ListDiff":return[3,9]}return[3,10];case 1:return r=ud(n,t,e),a=r.boxes,s=r.scores,o=r.maxOutputSize,l=r.iouThreshold,u=r.scoreThreshold,c=r.softNmsSigma,[4,B.image.nonMaxSuppressionWithScoreAsync(a,s,o,l,u,c)];case 2:return h=g.sent(),[2,[h.selectedIndices,h.selectedScores]];case 3:return d=ud(n,t,e),a=d.boxes,s=d.scores,o=d.maxOutputSize,l=d.iouThreshold,u=d.scoreThreshold,p=A("padToMaxOutputSize",n,t,e),[4,B.image.nonMaxSuppressionPaddedAsync(a,s,o,l,u,p)];case 4:return h=g.sent(),[2,[h.selectedIndices,h.validOutputs]];case 5:return f=ud(n,t,e),a=f.boxes,s=f.scores,o=f.maxOutputSize,l=f.iouThreshold,u=f.scoreThreshold,[4,B.image.nonMaxSuppressionAsync(a,s,o,l,u)];case 6:return[2,[g.sent()]];case 7:return m=B.cast(A("condition",n,t,e),"bool"),[4,B.whereAsync(m)];case 8:return h=[g.sent()],m.dispose(),[2,h];case 9:return[2,B.setdiff1dAsync(A("x",n,t,e),A("y",n,t,e))];case 10:throw TypeError("Node type "+n.op+" is not implemented")}})})};var VW=function(n,t,e){switch(n.op){case"TopKV2":{var i=A("x",n,t,e),r=A("k",n,t,e),a=A("sorted",n,t,e),s=B.topk(i,r,a);return[s.values,s.indices]}case"Unique":{var i=A("x",n,t,e),s=B.unique(i);return[s.values,s.indices]}case"UniqueV2":{var i=A("x",n,t,e),o=A("axis",n,t,e),s=B.unique(i,o);return[s.values,s.indices]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var qW=function(n,t,e){switch(n.op){case"Const":return t[n.name];case"PlaceholderWithDefault":var i=A("default",n,t,e);return[Ht(n.name,t,e)||i];case"Placeholder":return[Ht(n.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{var r=A("x",n,t,e);return[Qn(r)]}case"IdentityN":return A("x",n,t,e).map(function(h){return Qn(h)});case"Snapshot":var a=A("x",n,t,e);return[Qn(a)];case"Shape":return[B.tensor1d(A("x",n,t,e).shape,"int32")];case"ShapeN":return A("x",n,t,e).map(function(h){return B.tensor1d(h.shape)});case"Size":return[B.scalar(A("x",n,t,e).size,"int32")];case"Rank":return[B.scalar(A("x",n,t,e).rank,"int32")];case"NoOp":return[B.scalar(1)];case"Print":var s=A("x",n,t,e),o=A("data",n,t,e),l=A("message",n,t,e),u=A("summarize",n,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(l);for(var c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,u));return[s];default:throw TypeError("Node type "+n.op+" is not implemented")}};var GW=function(n,t,e){switch(n.op){case"ResizeBilinear":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeBilinear(i,[r[0],r[1]],a)]}case"ResizeNearestNeighbor":{var i=A("images",n,t,e),r=A("size",n,t,e),a=A("alignCorners",n,t,e);return[B.image.resizeNearestNeighbor(i,[r[0],r[1]],a)]}case"CropAndResize":{var s=A("image",n,t,e),o=A("boxes",n,t,e),l=A("boxInd",n,t,e),u=A("cropSize",n,t,e),c=A("method",n,t,e),h=A("extrapolationValue",n,t,e);return[B.image.cropAndResize(s,o,l,u,c,h)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var YW=function(n,t,e){switch(n.op){case"Equal":return[B.equal(A("a",n,t,e),A("b",n,t,e))];case"NotEqual":return[B.notEqual(A("a",n,t,e),A("b",n,t,e))];case"Greater":return[B.greater(A("a",n,t,e),A("b",n,t,e))];case"GreaterEqual":return[B.greaterEqual(A("a",n,t,e),A("b",n,t,e))];case"Less":return[B.less(A("a",n,t,e),A("b",n,t,e))];case"LessEqual":return[B.lessEqual(A("a",n,t,e),A("b",n,t,e))];case"LogicalAnd":return[B.logicalAnd(A("a",n,t,e),A("b",n,t,e))];case"LogicalNot":return[B.logicalNot(A("a",n,t,e))];case"LogicalOr":return[B.logicalOr(A("a",n,t,e),A("b",n,t,e))];case"Select":case"SelectV2":return[B.where(A("condition",n,t,e),A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var KW=function(n,t,e){switch(n.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[B.matMul(A("a",n,t,e),A("b",n,t,e),A("transposeA",n,t,e),A("transposeB",n,t,e))];case"Transpose":return[B.transpose(A("x",n,t,e),A("perm",n,t,e))];case"_FusedMatMul":var i=A("fusedOps",n,t,e),r=i[0],a=i[1],s=r==="biasadd",o=a==="prelu",l=A("numArgs",n,t,e);if(s){if(o&&l!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&l!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}var u=A("args",n,t,e),c=u[0],h=u[1];return[B.fused.matMul({a:A("a",n,t,e),b:A("b",n,t,e),transposeA:A("transposeA",n,t,e),transposeB:A("transposeB",n,t,e),bias:c,activation:a,preluActivationWeights:h})];default:throw TypeError("Node type "+n.op+" is not implemented")}};var jW=function(n,t,e){switch(n.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"FusedBatchNormV3":return[B.batchNorm(A("x",n,t,e),A("mean",n,t,e),A("variance",n,t,e),A("offset",n,t,e),A("scale",n,t,e),A("epsilon",n,t,e))];case"LRN":return[B.localResponseNormalization(A("x",n,t,e),A("radius",n,t,e),A("bias",n,t,e),A("alpha",n,t,e),A("beta",n,t,e))];case"Softmax":return[B.softmax(A("x",n,t,e))];case"LogSoftmax":return[B.logSoftmax(A("x",n,t,e))];case"SparseToDense":return[B.sparseToDense(A("sparseIndices",n,t,e),A("outputShape",n,t,e),A("sparseValues",n,t,e),A("defaultValue",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var $W=function(n,t,e){switch(n.op){case"Max":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.max(A("x",n,t,e),i,r)]}case"Mean":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.mean(A("x",n,t,e),i,r)]}case"Min":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.min(A("x",n,t,e),i,r)]}case"Sum":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.sum(A("x",n,t,e),i,r)]}case"All":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.all(A("x",n,t,e),i,r)]}case"Any":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.any(A("x",n,t,e),i,r)]}case"ArgMax":{var i=A("axis",n,t,e);return[B.argMax(A("x",n,t,e),i)]}case"ArgMin":{var i=A("axis",n,t,e);return[B.argMin(A("x",n,t,e),i)]}case"Prod":{var i=A("axis",n,t,e),r=A("keepDims",n,t,e);return[B.prod(A("x",n,t,e),i,r)]}case"Cumsum":{var i=A("axis",n,t,e),a=A("exclusive",n,t,e),s=A("reverse",n,t,e);return[B.cumsum(A("x",n,t,e),i,a,s)]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var XW=function(n,t,e){switch(n.op){case"ConcatV2":case"Concat":{var i=A("n",n,t,e),r=A("axis",n,t,e),a=A("tensors",n,t,e);return a=a.slice(0,i),[B.concat(a,r)]}case"GatherV2":case"Gather":{var r=A("axis",n,t,e),s=A("x",n,t,e),o=A("indices",n,t,e);return[B.gather(s,B.cast(o,"int32"),r)]}case"ReverseV2":case"Reverse":{var r=A("axis",n,t,e),s=A("x",n,t,e);return[B.reverse(s,r)]}case"Slice":{var l=A("begin",n,t,e),u=A("size",n,t,e);return[B.slice(A("x",n,t,e),l,u)]}case"StridedSlice":{var l=A("begin",n,t,e),c=A("end",n,t,e),h=A("strides",n,t,e),d=A("beginMask",n,t,e),p=A("endMask",n,t,e),f=A("ellipsisMask",n,t,e),m=A("newAxisMask",n,t,e),g=A("shrinkAxisMask",n,t,e),v=A("x",n,t,e);return[B.stridedSlice(v,l,c,h,d,p,f,m,g)]}case"Pack":return B.tidy(function(){var D=A("axis",n,t,e),k=A("tensors",n,t,e),W=k[0].shape,F=B.squeeze(k[0]).shape,P=k.map(function(H){var _=B.util.arraysEqual(H.shape,W);if(!_&&!B.util.arraysEqual(B.squeeze(H).shape,F))throw new Error("the input tensors shape does not match");return _?H:B.reshape(H,W)});return[B.stack(P,D)]});case"Unpack":{var r=A("axis",n,t,e),v=A("tensor",n,t,e);return B.unstack(v,r)}case"Tile":{var b=A("reps",n,t,e);return[B.tile(A("x",n,t,e),b)]}case"Split":case"SplitV":{var r=A("axis",n,t,e),w=A("numOrSizeSplits",n,t,e),v=A("x",n,t,e);return B.split(v,w,r)}case"ScatterNd":{var o=A("indices",n,t,e),S=A("values",n,t,e),L=A("shape",n,t,e);return[B.scatterND(o,S,L)]}case"GatherNd":{var x=A("x",n,t,e),o=A("indices",n,t,e);return[B.gatherND(x,o)]}case"SparseToDense":{var o=A("sparseIndices",n,t,e),L=A("outputShape",n,t,e),C=A("sparseValues",n,t,e),R=A("defaultValue",n,t,e);return[B.sparseToDense(o,C,L,C.dtype===R.dtype?R:B.cast(R,C.dtype))]}default:throw TypeError("Node type "+n.op+" is not implemented")}};var JW=function(n,t,e){switch(n.op){case"FFT":return[B.fft(A("x",n,t,e))];case"IFFT":return[B.ifft(A("x",n,t,e))];case"RFFT":return[B.rfft(A("x",n,t,e))];case"IRFFT":return[B.irfft(A("x",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var ZW=function(n,t,e){switch(n.op){case"Cast":return[B.cast(A("x",n,t,e),A("dtype",n,t,e))];case"ExpandDims":{var i=A("axis",n,t,e);return[B.expandDims(A("x",n,t,e),i)]}case"Squeeze":{var i=A("axis",n,t,e);return[B.squeeze(A("x",n,t,e),i)]}case"Reshape":return[B.reshape(A("x",n,t,e),A("shape",n,t,e))];case"PadV2":case"Pad":return[B.pad(A("x",n,t,e),A("padding",n,t,e),A("constantValue",n,t,e))];case"SpaceToBatchND":{var r=A("blockShape",n,t,e),a=A("paddings",n,t,e);return[B.spaceToBatchND(A("x",n,t,e),r,a)]}case"BatchToSpaceND":{var r=A("blockShape",n,t,e),s=A("crops",n,t,e);return[B.batchToSpaceND(A("x",n,t,e),r,s)]}case"DepthToSpace":{var o=A("blockSize",n,t,e),l=A("dataFormat",n,t,e).toUpperCase();return[B.depthToSpace(A("x",n,t,e),o,l)]}case"BroadcastTo":return[B.broadcastTo(A("x",n,t,e),A("shape",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function tw(n,t,e){var i=function(r,a,s){switch(r.category){case"arithmetic":return B.tidy(function(){return RW(r,a,s)});case"basic_math":return B.tidy(function(){return OW(r,a,s)});case"control":return zW(r,a,s);case"convolution":return B.tidy(function(){return PW(r,a,s)});case"creation":return B.tidy(function(){return _W(r,a,s)});case"dynamic":return HW(r,a,s);case"evaluation":return B.tidy(function(){return VW(r,a,s)});case"image":return B.tidy(function(){return GW(r,a,s)});case"graph":return B.tidy(function(){return qW(r,a,s)});case"logical":return B.tidy(function(){return YW(r,a,s)});case"matrices":return B.tidy(function(){return KW(r,a,s)});case"normalization":return B.tidy(function(){return jW(r,a,s)});case"reduction":return B.tidy(function(){return $W(r,a,s)});case"slice_join":return B.tidy(function(){return XW(r,a,s)});case"spectral":return B.tidy(function(){return JW(r,a,s)});case"transformation":return B.tidy(function(){return ZW(r,a,s)});case"custom":var o=$b(r.op);if(o&&o.customExecutor)return o.customExecutor(new CW(r,a,s));throw TypeError("Custom op "+r.op+" is not registered.");default:throw TypeError("Unknown op '"+r.op+"'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()")}}(n,t,e);return i instanceof Promise?i.then(function(r){return[].concat(r)}):[].concat(i)}var nw=function(){function n(t,e,i,r){t===void 0&&(t={}),e===void 0&&(e={}),i===void 0&&(i={}),r===void 0&&(r={}),this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=i,this.functionMap=r,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}return n.prototype.newFrame=function(t,e){return{id:t,frameName:e,iterationId:0}},Object.defineProperty(n.prototype,"currentContext",{get:function(){return this.contexts},set:function(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextId",{get:function(){return this._currentContextIds[0]},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"currentContextIds",{get:function(){return this._currentContextIds},enumerable:!0,configurable:!0}),n.prototype.generateCurrentContextIds=function(){for(var t=[],e=0;e<this.contexts.length-1;e++){var i=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(i))}t.push(""),this._currentContextIds=t},n.prototype.contextIdforContexts=function(t){return t?t.map(function(e){return e.id===0&&e.iterationId===0?"":e.frameName+"-"+e.iterationId}).join("/"):""},n.prototype.enterFrame=function(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))},n.prototype.exitFrame=function(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function rw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return jt(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return jt(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((iw(d)||QW(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function eU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return jt(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var tU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],nU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function iw(n){return tU.indexOf(n.op)>=0}function QW(n){return nU.indexOf(n.op)>=0}var aw=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=rw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return eU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[jt(d)[0]]}),s=e.map(function(d){return jt(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new nw(i.weightMap,c,h,i.functionExecutorMap),p=Kb({},i.weightMap);Object.keys(t).forEach(function(w){var S=jt(w),L=S[0],x=S[1],C=[];C[x]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g<u.length;g++){var v=u[g];if(!p[v.name]){var b=tw(v,p,d);if(b instanceof Promise)throw new Error("The execution of the op '"+v.op+"' returned a promise. Please use model.executeAsync() instead.");p[v.name]=b,i.checkTensorForDisposal(v.name,v,p,d,f,s,m)}}return i.parent==null&&d.dispose(f),e.map(function(w){return Ht(w,p,d)})})},n.prototype.getFrozenTensorIds=function(t){var e=[].concat.apply([],Object.keys(t).map(function(i){return t[i]}).map(function(i){return i.map(function(r){return r.id})}));return new Set(e)},n.prototype.checkTensorForDisposal=function(t,e,i,r,a,s,o){if(e.category==="control"||s.indexOf(t)!==-1)return;i[t].forEach(function(l){l!=null&&(o[l.id]=(o[l.id]||0)+e.children.length)}),e.inputs.forEach(function(l){if(l.category!=="control"){var u=jF(l.name,i,r);u!=null&&u.forEach(function(c){if(c&&!a.has(c.id)){var h=o[c.id];h===1?(c.dispose(),delete o[c.id]):h!=null&&o[c.id]--}})}})},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){return wn(this,function(i){return[2,this._executeAsync(t,e)]})})},n.prototype._executeAsync=function(t,e,i,r,a){return i===void 0&&(i=!1),r===void 0&&(r={}),a===void 0&&(a={}),zn(this,void 0,void 0,function(){var s,o,l,u,c,h;return wn(this,function(d){switch(d.label){case 0:return i||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e)),s=new nw(this.weightMap,r,a,this.functionExecutorMap),[4,this.executeWithControlFlow(t,s,e,i)];case 1:return o=d.sent(),l=e.map(function(p){return Ht(p,o,s)}),u=l.map(function(p){return p.id}),c=Object.keys(t).map(function(p){return t[p].id}),h=new Set(u.concat(c,this.weightIds)),Object.keys(o).forEach(function(p){var f=o[p];f.forEach(function(m){m&&!m.isDisposed&&!h.has(m.id)&&m.dispose()})}),this.parent==null&&s.dispose(h),[2,l]}})})},n.prototype.executeFunctionAsync=function(t,e,i){return zn(this,void 0,void 0,function(){var r,a=this;return wn(this,function(s){return r=t.reduce(function(o,l,u){return o[a.inputs[u].name]=l,o},{}),[2,this._executeAsync(r,this.outputNodes,!0,e,i)]})})},n.prototype.executeWithControlFlow=function(t,e,i,r){return zn(this,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x=this;return wn(this,function(C){switch(C.label){case 0:a=Object.keys(t),s=a.map(function(R){return x.graph.nodes[jt(R)[0]]}),o=i.map(function(R){return jt(R)[0]}),l=o.map(function(R){return x.graph.nodes[R]}),u=rw(t,l,this.weightMap),c=u.usedNodes,h=u.missingInputs,d=u.dynamicNode,p=u.syncInputs,f=s.concat(this.graph.weights).map(function(R){return{node:R,contexts:e.currentContext}}),m=Kb({},this.weightMap),Object.keys(t).forEach(function(R){var D=jt(R),k=D[0],W=D[1],F=[];F[W]=t[R],m[k]=F}),g={},v=this.getFrozenTensorIds(m),b={},C.label=1;case 1:return f.length>0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!iw(R)&&!Ht(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=Zn(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=tw(f.node,r,i);m||(m=Zn(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=Zn(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Ht(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=jt(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=jt(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=jt(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var iU="?tfjs-format=file",rU="model.json",sw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return zn(this,void 0,void 0,function(){var t;return wn(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new aw(Jb.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Jb.Instance.transformGraph(t.modelInitializer);this.initializer=new aw(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return zn(this,void 0,void 0,function(){var i;return wn(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return zn(this,void 0,void 0,function(){var i;return wn(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function aU(n,t){return t===void 0&&(t={}),zn(this,void 0,void 0,function(){var e;return wn(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+rU+iU)),e=new sw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var sU="2.6.0";rr.GraphModel=sw;rr.deregisterOp=KF;rr.loadGraphModel=aU;rr.registerOp=YF;rr.version_converter=sU});var lw=be(()=>{});var Cw=be(Vt=>{"use strict";Object.defineProperty(Vt,"__esModule",{value:!0});var Ne=Xi();var cd=function(n,t){return cd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},cd(n,t)};function Ve(n,t){cd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var jr=typeof globalThis!="undefined"?globalThis:typeof window!="undefined"?window:typeof global!="undefined"?global:typeof self!="undefined"?self:{};function ar(n,t){return t={exports:{}},n(t,t.exports),t.exports}var oU=ar(function(n){(function(t,e,i){function r(l){var u=this,c=o();u.next=function(){var h=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=h-(u.c=h|0)},u.c=1,u.s0=c(" "),u.s1=c(" "),u.s2=c(" "),u.s0-=c(l),u.s0<0&&(u.s0+=1),u.s1-=c(l),u.s1<0&&(u.s1+=1),u.s2-=c(l),u.s2<0&&(u.s2+=1),c=null}function a(l,u){return u.c=l.c,u.s0=l.s0,u.s1=l.s1,u.s2=l.s2,u}function s(l,u){var c=new r(l),h=u&&u.state,d=c.next;return d.int32=function(){return c.next()*4294967296|0},d.double=function(){return d()+(d()*2097152|0)*11102230246251565e-32},d.quick=d,h&&(typeof h=="object"&&a(h,c),d.state=function(){return a(c,{})}),d}function o(){var l=4022871197,u=function(c){c=c.toString();for(var h=0;h<c.length;h++){l+=c.charCodeAt(h);var d=.02519603282416938*l;l=d>>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(jr,n,!1)}),lU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(jr,n,!1)}),uU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c<u.length+64;c++)l.x^=u.charCodeAt(c)|0,c==u.length&&(l.d=l.x<<10^l.x>>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(jr,n,!1)}),cU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d<h.length;++d)f[d&7]=f[d&7]<<15^h.charCodeAt(d)+f[d+1&7]<<13;for(;f.length<8;)f.push(0);for(d=0;d<8&&f[d]===0;++d);for(d==8?p=f[7]=-1:p=f[d],c.x=f,c.i=0,d=256;d>0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(jr,n,!1)}),hU=ar(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m<b;++m)h&&(p^=h.charCodeAt((m+32)%h.length)),m===0&&(g=p),p^=p<<10,p^=p>>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(jr,n,!1)}),dU=ar(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c<u.length+20;c++)l.b^=u.charCodeAt(c)|0,l.next()}function a(o,l){return l.a=o.a,l.b=o.b,l.c=o.c,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(jr,n,!1)}),sr=ar(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,x){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W<u;)W=(W+P)*r,F*=r,P=D.g(1);for(;W>=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||x||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,x=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(x||(S=[x++]);R<r;)k[R]=R++;for(R=0;R<r;R++)k[R]=k[D=h&D+S[R%x]+(L=k[R])],k[D]=L;(C.g=function(W){for(var F,P=0,H=C.i,_=C.j,K=C.S;W--;)F=K[H=h&H+1],P=P*r+K[h&(K[H]=K[_=h&_+F])+(K[_]=F)];return C.i=H,C.j=_,P})(r)}function m(S,L){return L.i=S.i,L.j=S.j,L.S=S.S.slice(),L}function g(S,L){var x=[],C=typeof S,R;if(L&&C=="object")for(R in S)try{x.push(g(S[R],L-1))}catch(D){}return x.length?x:C=="string"?S:S+"\0"}function v(S,L){for(var x=S+"",C,R=0;R<x.length;)L[h&R]=h&(C^=L[h&R]*19)+x.charCodeAt(R++);return w(L)}function b(){try{var S;return d&&(S=d.randomBytes)?S=S(r):(S=new Uint8Array(r),(i.crypto||i.msCrypto).getRandomValues(S)),w(S)}catch(C){var L=i.navigator,x=L&&L.plugins;return[+new Date,i,x,i.screen,w(t)]}}function w(S){return String.fromCharCode.apply(0,S)}if(v(e.random(),t),n.exports){n.exports=p;try{d=ds()}catch(S){}}})([],Math)});sr.alea=oU;sr.xor128=lU;sr.xorwow=uU;sr.xorshift7=cU;sr.xor4096=hU;sr.tychei=dU;var pU=sr,uw=pU.alea;function fU(n,t){return xo(n,t)}function xo(n,t,e,i){if(e===void 0&&(e=new Map),i===void 0&&(i=new Set),n==null)return null;if(i.has(n))throw new Error("Circular references are not supported.");if(e.has(n))return e.get(n);var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if($r(n)){var a=Array.isArray(n)?[]:{};i.add(n);for(var s in n){var o=n[s],l=xo(o,t,e,i);a[s]=l}return i.delete(n),a}else throw new Error("Can't recurse into non-iterable type: "+n);else return e.set(n,r.value),r.value}function mU(n,t){return t===void 0&&(t=hw),cw(n,t)}function cw(n,t,e){e===void 0&&(e=new Set);var i=n[0];if(e.has(i))throw new Error("Circular references are not supported.");var r=t(n);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if($r(i)){var a=Array.isArray(i)?[]:{};e.add(i);var s=function(l){var u=n.map(function(h){return h[l]}),c=cw(u,t,e);a[l]=c};for(var o in i)s(o);return e.delete(i),a}else throw new Error("Can't recurse into non-iterable type: "+i);else return r.value}function hw(n){return n===null?null:$r(n[0])?{value:null,recurse:!0}:{value:n,recurse:!1}}function dw(n,t){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:e=new Map,xo(n,t,e),i=0,r=Array.from(e.keys()),u.label=1;case 1:return i<r.length?(a=r[i],s=e.get(a),s instanceof Promise?[4,s]:[3,3]):[3,4];case 2:o=u.sent(),e.set(a,o),u.label=3;case 3:return i++,[3,1];case 4:return l=xo(n,t,e),[2,l]}})})}function $r(n){return n!=null&&!ArrayBuffer.isView(n)&&(Array.isArray(n)||typeof n=="object"&&!(n instanceof Ne.Tensor))}function vU(n){return n==null||gU(n)||Array.isArray(n)||typeof n=="object"&&n instanceof Ne.Tensor||Ne.util.isTypedArray(n)}function gU(n){return n===null||typeof n!="object"&&typeof n!="function"}function bU(n){return fU(n,yU)}function yU(n){return n instanceof Ne.Tensor?{value:n.clone(),recurse:!1}:$r(n)?{value:null,recurse:!0}:{value:n,recurse:!1}}var pw=function(){function n(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}return n.prototype.wrap=function(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity},n.prototype.get=function(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]},n.prototype.set=function(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e},n.prototype.length=function(){var t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t},n.prototype.isFull=function(){return this.length()===this.capacity},n.prototype.isEmpty=function(){return this.length()===0},n.prototype.push=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)},n.prototype.pushAll=function(t){for(var e=0,i=t;e<i.length;e++){var r=i[e];this.push(r)}},n.prototype.pop=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);var t=this.get(this.end);return this.set(this.end,void 0),t},n.prototype.unshift=function(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)},n.prototype.shift=function(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t},n.prototype.shuffleExcise=function(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");var e=this.wrap(this.begin+t),i=this.get(e);return this.set(e,this.pop()),i},n}();var wU=function(n){Ve(t,n);function t(){return n.call(this,t.INITIAL_CAPACITY)||this}return t.prototype.isFull=function(){return!1},t.prototype.push=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.push.call(this,e)},t.prototype.unshift=function(e){n.prototype.isFull.call(this)&&this.expand(),n.prototype.unshift.call(this,e)},t.prototype.expand=function(){for(var e=this.capacity*2,i=new Array(e),r=this.length(),a=0;a<r;a++)i[a]=this.get(this.wrap(this.begin+a));this.data=i,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=r},t.INITIAL_CAPACITY=32,t}(pw);function fw(n){return new SU(n)}function hd(n){return new LU(n)}function IU(n,t){return new mw(n,t)}function TU(n,t){return t===void 0&&(t=Ii.FAIL),new AU(n,t)}var Nt=function(){function n(){}return n.prototype.toArray=function(){return se(this,void 0,void 0,function(){var t,e;return oe(this,function(i){switch(i.label){case 0:return t=[],[4,this.next()];case 1:e=i.sent(),i.label=2;case 2:return e.done?[3,4]:(t.push(e.value),[4,this.next()]);case 3:return e=i.sent(),[3,2];case 4:return[2,t]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){var t,e,i;return oe(this,function(r){switch(r.label){case 0:return t=this.prefetch(100),e=[],[4,t.next()];case 1:i=r.sent(),r.label=2;case 2:return i.done?[3,4]:(e.push(i.value),[4,t.next()]);case 3:return i=r.sent(),[3,2];case 4:return[2,e]}})})},n.prototype.resolveFully=function(){return se(this,void 0,void 0,function(){var t;return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:t=e.sent(),e.label=2;case 2:return t.done?[3,4]:[4,this.next()];case 3:return t=e.sent(),[3,2];case 4:return[2]}})})},n.prototype.resolveWhile=function(t){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:return[4,this.next()];case 1:e=r.sent(),i=t(e.value),r.label=2;case 2:return!e.done&&i?[4,this.next()]:[3,4];case 3:return e=r.sent(),i=t(e.value),[3,2];case 4:return[2]}})})},n.prototype.handleErrors=function(t){return new DU(this,t)},n.prototype.filter=function(t){return new OU(this,t)},n.prototype.map=function(t){return new EU(this,t)},n.prototype.mapAsync=function(t){return new gw(this,t)},n.prototype.serialMapAsync=function(t){return new gw(this,t).serial()},n.prototype.flatmap=function(t){return new kU(this,t)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.map(t).resolveFully()]})})},n.prototype.serialForEach=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.serialMapAsync(t).resolveWhile(function(i){return i===!0})]})})},n.prototype.rowMajorBatch=function(t,e){return e===void 0&&(e=!0),new RU(this,t,e)},n.prototype.columnMajorBatch=function(t,e,i){e===void 0&&(e=!0),i===void 0&&(i=hw);var r=this.rowMajorBatch(t,e);return r.map(function(a){return mU(a,i)})},n.prototype.concatenate=function(t,e){return new mw(fw([this,t]),e)},n.prototype.take=function(t){return t<0||t==null?this:new CU(this,t)},n.prototype.skip=function(t){return t<0||t==null?this:new xU(this,t)},n.prototype.prefetch=function(t){return new vw(this,t)},n.prototype.shuffle=function(t,e){return new FU(this,t,e)},n.prototype.serial=function(){return new NU(this)},n}(),SU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.items=e,i.trav=0,i}return t.prototype.summary=function(){return"Array of "+this.items.length+" items"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return this.trav>=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:bU(e),done:!1}])})})},t}(Nt),LU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(Nt),NU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(Nt),xU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++<this.maxCount?[4,this.upstream.next()]:[3,2];case 1:return e=i.sent(),e.done?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2,this.upstream.next()]}})})},t}(Nt),CU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(Nt),RU=function(n){Ve(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length<this.batchSize?[4,this.upstream.next()]:[3,3];case 2:return i=r.sent(),i.done?this.enableSmallLastBatch&&e.length>0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(Nt),OU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(Nt),EU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),DU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.handler=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(Nt),gw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,{value:r,done:!1}]}})})},t}(Nt),dd=function(n){Ve(t,n);function t(){var e=n.call(this)||this;return e.outputQueue=new wU,e.lastRead=Promise.resolve({value:null,done:!1}),e}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.outputQueue.length()===0?[4,this.pump()]:[3,2];case 1:return e.sent()?[3,0]:[2,{value:null,done:!0}];case 2:return[2,{value:this.outputQueue.shift(),done:!1}]}})})},t}(Nt),kU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s<o.length;s++)l=o[s],Ne.tensor_util.isTensorInList(l,a)||l.dispose();return[2,!0]}})})},t}(dd),mw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.baseErrorHandler=i,r.lastRead=null,r.iterator=null,r.moreIterators=e,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of chained summaries";return e+" -> Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(Nt),Ii;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Ii||(Ii={}));var AU=function(n){Ve(t,n);function t(e,i){i===void 0&&(i=Ii.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof Nt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,dw(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Ii.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Ii.SHORTEST:return[2,{value:null,done:!0}];case Ii.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(Nt),vw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new pw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(Nt),FU=function(n){Ve(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=uw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(vw);var Ya=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is
+      `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),$t(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,WU)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,$t(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,$t(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return $t(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return $t(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return $t(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,$t(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=hd(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,IU(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?r=0:r=null,$t(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().skip(t)]}})})},r)},n.prototype.shuffle=function(t,e,i){var r=this;if(i===void 0&&(i=!0),t==null||t<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError("`Dataset.shuffle()` requires bufferSize to be specified.  If your data fits in main memory (for regular JS objects), and/or GPU memory (for `tf.Tensor`s), consider setting "+("bufferSize to the dataset size ("+this.size+" elements)"));var a=this,s=uw(e||Ne.util.now().toString());return $t(function(){return se(r,void 0,void 0,function(){var o;return oe(this,function(l){switch(l.label){case 0:return o=s.int32(),i&&(o+=s.int32()),[4,a.iterator()];case 1:return[2,l.sent().shuffle(t,o.toString())]}})})},this.size)},n.prototype.take=function(t){var e=this,i=this,r;return this.size!=null&&this.size>t?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,$t(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function $t(n,t){return t===void 0&&(t=null),new(function(e){Ve(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(Ya))}function UU(n){var t=this;return $t(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,fw(n)]})})},n.length)}function BU(n){var t=this;if(!$r(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i<n.length;i++)e=e==null?n[i].size:Math.min(e,n[i].size);else if(n instanceof Object)for(var r in n)e=e==null?n[r].size:Math.min(e,n[r].size);return $t(function(){return se(t,void 0,void 0,function(){var a;return oe(this,function(s){switch(s.label){case 0:return[4,dw(n,function(o){if(o instanceof Ya)return{value:o.iterator(),recurse:!1};if($r(o))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")})];case 1:return a=s.sent(),[2,TU(a,Ii.SHORTEST)]}})})},e)}function WU(n){if(n===null)return null;var t=n[0];if(vU(t)){var e=zU(n);return{value:e,recurse:!1}}return{value:null,recurse:!0}}function zU(n){if(n.length===0)throw new Error("Can't make a batch of zero elements.");return n[0]instanceof Ne.Tensor?Ne.stack(n):Ne.tensor(n)}var yw=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.input=e,i}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.input.iterator()];case 1:return e=a.sent(),i=e.decodeUTF8(),r=i.split(`
+`).map(function(s){return s.endsWith("\r")&&(s=s.slice(0,-1)),s}),[2,r]}})})},t}(Ya);var Co='"',Ka=Symbol("out"),bw=Symbol("field"),Ro=Symbol("quote"),pd=Symbol("quoteafterquote"),ww=Symbol("quoteinquote"),Sw=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.input=e,r.hasHeader=!0,r.fullColumnNames=null,r.columnNamesValidated=!1,r.columnConfigs=null,r.configuredColumnsOnly=!1,r.delimiter=",",r.delimWhitespace=!1,r.base=new yw(e),i||(i={}),r.hasHeader=!(i.hasHeader===!1),r.fullColumnNames=i.columnNames,r.columnConfigs=i.columnConfigs,r.configuredColumnsOnly=i.configuredColumnsOnly,i.delimWhitespace?(Ne.util.assert(i.delimiter==null,function(){return"Delimiter should not be provided when delimWhitespace is true."}),r.delimWhitespace=!0,r.delimiter=" "):r.delimiter=i.delimiter?i.delimiter:",",r}return t.prototype.columnNames=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:e.sent(),e.label=2;case 2:return[2,this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames]}})})},t.prototype.setColumnNames=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u=this;return oe(this,function(c){switch(c.label){case 0:return[4,this.maybeReadHeaderLine()];case 1:if(e=c.sent(),!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");if(this.fullColumnNames&&e&&Ne.util.assert(e.length===this.fullColumnNames.length,function(){return"The length of provided columnNames ("+u.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."}),this.fullColumnNames||(this.fullColumnNames=e),i=this.fullColumnNames.reduce(function(h,d){return h[d]=h[d]+1||1,h},{}),r=Object.keys(i).filter(function(h){return i[h]>1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a<s.length;a++)if(o=s[a],l=this.fullColumnNames.indexOf(o),l===-1)throw new Error('The key "'+o+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}return this.columnNamesValidated=!0,[2]}})})},t.prototype.maybeReadHeaderLine=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return this.hasHeader?[4,this.base.iterator()]:[3,3];case 1:return e=s.sent(),[4,e.next()];case 2:if(i=s.sent(),i.done)throw new Error("No data was found for CSV parsing.");return r=i.value,a=this.parseRow(r,!1),[2,a];case 3:return[2,null]}})})},t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e,i=this;return oe(this,function(r){switch(r.label){case 0:return this.columnNamesValidated?[3,2]:[4,this.setColumnNames()];case 1:r.sent(),r.label=2;case 2:return[4,this.base.iterator()];case 3:return e=r.sent(),this.hasHeader&&(e=e.skip(1)),[2,e.map(function(a){return i.makeDataElement(a)})]}})})},t.prototype.makeDataElement=function(e){for(var i=this.parseRow(e),r={},a={},s=0;s<this.fullColumnNames.length;s++){var o=this.fullColumnNames[s],l=this.columnConfigs?this.columnConfigs[o]:null;if(this.configuredColumnsOnly&&!l)continue;var u=i[s],c=null;if(u==="")if(l&&l.default!==void 0)c=l.default;else{if(l&&(l.required||l.isLabel))throw new Error("Required column "+o+" is empty in this line: "+e);c=void 0}else{var h=Number(u);if(isNaN(h))l&&l.dtype==="bool"?c=this.getBoolean(u):c=u;else if(!l||!l.dtype)c=h;else switch(l.dtype){case"float32":c=h;break;case"int32":c=Math.floor(h);break;case"bool":c=this.getBoolean(u);break;default:c=h}}l&&l.isLabel?a[o]=c:r[o]=c}return Object.keys(a).length===0?r:{xs:r,ys:a}},t.prototype.getBoolean=function(e){return e==="1"||e.toLowerCase()==="true"?1:0},t.prototype.parseRow=function(e,i){i===void 0&&(i=!0);for(var r=[],a=0,s=e.length,o=Ka,l=0;l<s;l++)switch(o){case Ka:switch(e.charAt(l)){case Co:a=l+1,o=Ro;break;case this.delimiter:if(a=l+1,this.delimiter===" "&&this.delimWhitespace)break;r.push(""),o=Ka;break;default:o=bw,a=l;break}break;case bw:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l)),o=Ka,a=l+1;break}break;case Ro:switch(e.charAt(l)){case Co:o=pd;break}break;case pd:switch(e.charAt(l)){case this.delimiter:r.push(e.substring(a,l-1)),o=Ka,a=l+1;break;case Co:o=Ro;break;default:o=ww;break}break;case ww:switch(e.charAt(l)){case Co:o=Ro;break}break}if(o===pd?r.push(e.substring(a,s-1)):r.push(e.substring(a)),i&&r.length!==this.fullColumnNames.length)throw new Error("Invalid row in csv file. Should have "+this.fullColumnNames.length+" elements in a row, but got "+r);return r},t}(Ya);var PU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;i.microphoneConfig=e,i.isClosed=!1,i.fftSize=e.fftSize||1024;var r=Math.log2(i.fftSize);if(i.fftSize<0||r<4||r>14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(Nt);var _U=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(Nt);var Lw=function(){function n(){}return n}();var Iw=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new MU(this,e)},t}(Nt),MU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new HU(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),HU=function(n){Ve(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r<a.length;r++)s=a[r],this.outputQueue.push(s);return this.carryover=i[i.length-1],[2,!0]}})})},t}(dd);var qU=function(n){Ve(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.decodeUTF8=function(){return new VU(this)},t}(Nt),VU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.impl=new GU(e),i}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(Iw),GU=function(n){Ve(t,n);function t(e){var i=n.call(this)||this;if(i.upstream=e,Ne.env().get("IS_BROWSER"))i.decoder=new TextDecoder("utf-8");else{var r=lw().StringDecoder;i.decoder=new r("utf8")}return i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(dd);var Aw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(qU);function KU(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=YU(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new Aw(a,t)];case 3:throw new Error(r.statusText)}})})}var YU=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function Tw(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var Nw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return Tw(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new Aw(this.input,this.options)]})})},t}(Lw);var xw=function(n){Ve(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return Tw(this.url)?[2,new Nw(this.url,this.fileOptions).iterator()]:[2,KU(this.url,this.fileOptions)]})})},t}(Lw);function jU(n,t){return t===void 0&&(t={}),new Sw(new xw(n),t)}function $U(n){var t=this,e=hd(n);return $t(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function XU(n){var t=this;return $t(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,hd(function(){return e.next()})]}})})})}function JU(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,_U.create(n,t)]})})}function ZU(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,PU.create(n)]})})}var QU="2.6.0";Vt.CSVDataset=Sw;Vt.Dataset=Ya;Vt.FileDataSource=Nw;Vt.TextLineDataset=yw;Vt.URLDataSource=xw;Vt.array=UU;Vt.csv=jU;Vt.func=$U;Vt.generator=XU;Vt.microphone=ZU;Vt.version_data=QU;Vt.webcam=JU;Vt.zip=BU});var Ow=be((Rw,fd)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c<u.length;c++){o+=u.charCodeAt(c);var h=.02519603282416938*o;o=h>>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(Rw,typeof fd=="object"&&fd,typeof define=="function"&&define)});var Dw=be((Ew,md)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(Ew,typeof md=="object"&&md,typeof define=="function"&&define)});var Fw=be((kw,gd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u<l.length+64;u++)o.x^=l.charCodeAt(u)|0,u==l.length&&(o.d=o.x<<10^o.x>>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(kw,typeof gd=="object"&&gd,typeof define=="function"&&define)});var Uw=be((Ww,vd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h<c.length;++h)p[h&7]=p[h&7]<<15^c.charCodeAt(h)+p[h+1&7]<<13;for(;p.length<8;)p.push(0);for(h=0;h<8&&p[h]===0;++h);for(h==8?d=p[7]=-1:d=p[h],u.x=p,u.i=0,h=256;h>0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(Ww,typeof vd=="object"&&vd,typeof define=="function"&&define)});var zw=be((Bw,yd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f<v;++f)c&&(d^=c.charCodeAt((f+32)%c.length)),f===0&&(m=d),d^=d<<10,d^=d>>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(Bw,typeof yd=="object"&&yd,typeof define=="function"&&define)});var _w=be((Pw,bd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u<l.length+20;u++)o.b^=l.charCodeAt(u)|0,o.next()}function r(s,o){return o.a=s.a,o.b=s.b,o.c=s.c,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(Pw,typeof bd=="object"&&bd,typeof define=="function"&&define)});var Mw=be((CH,Oo)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var x=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),x),R=new p(x),D=function(){for(var k=R.g(r),W=o,F=0;k<l;)k=(k+F)*i,W*=i,F=R.g(1);for(;k>=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,x=this,C=0,R=x.i=x.j=0,D=x.S=[];for(L||(w=[L++]);C<i;)D[C]=C++;for(C=0;C<i;C++)D[C]=D[R=c&R+w[C%L]+(S=D[C])],D[R]=S;(x.g=function(k){for(var W,F=0,P=x.i,H=x.j,_=x.S;k--;)W=_[P=c&P+1],F=F*i+_[c&(_[P]=_[H=c&H+W])+(_[H]=W)];return x.i=P,x.j=H,F})(i)}function f(w,S){return S.i=w.i,S.j=w.j,S.S=w.S.slice(),S}function m(w,S){var L=[],x=typeof w,C;if(S&&x=="object")for(C in w)try{L.push(m(w[C],S-1))}catch(R){}return L.length?L:x=="string"?w:w+"\0"}function g(w,S){for(var L=w+"",x,C=0;C<L.length;)S[c&C]=c&(x^=S[c&C]*19)+L.charCodeAt(C++);return b(S)}function v(){try{var w;return h&&(w=h.randomBytes)?w=w(i):(w=new Uint8Array(i),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(x){var S=e.navigator,L=S&&S.plugins;return[+new Date,e,L,e.screen,b(n)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),n),typeof Oo=="object"&&Oo.exports){Oo.exports=d;try{h=ds()}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return d})})([],Math)});var Vw=be((RH,Hw)=>{var eB=Ow(),tB=Dw(),nB=Fw(),iB=Uw(),rB=zw(),aB=_w(),or=Mw();or.alea=eB;or.xor128=tB;or.xorwow=nB;or.xorshift7=iB;or.xor4096=rB;or.tychei=aB;Hw.exports=or});var m0=be(ja=>{"use strict";Object.defineProperty(ja,"__esModule",{value:!0});var T=Xi(),sB=Vw();var wd=function(n,t){return wd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},wd(n,t)};function oB(n,t){wd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function qw(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function Gw(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}function ae(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&T.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the CPU backend."})})}var lB=T.kernel_impls.nonMaxSuppressionV3Impl,uB=T.kernel_impls.split,cB=T.kernel_impls.tile,hB=T.kernel_impls.topkImpl,dB=T.kernel_impls.whereImpl;function Sd(n,t,e,i){if(e==="linear")return n.linear(t);if(e==="relu")return n.relu(t);if(e==="elu")return T.elu(t);if(e==="relu6")return n.relu6(t);if(e==="prelu")return n.prelu(t,i);throw new Error("Activation "+e+" has not been implemented for the CPU backend.")}var Yw=function(n){oB(t,n);function t(){var e=n.call(this)||this;return e.blockSize=48,e.firstUse=!0,e.data=new T.DataStorage(e,T.engine()),e}return t.prototype.write=function(e,i,r){this.firstUse&&(this.firstUse=!1,T.env().get("IS_NODE")&&T.backend_util.warn(`
 ============================
 Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed things up dramatically, install our node backend, which binds to TensorFlow C++, by running npm i @tensorflow/tfjs-node, or npm i @tensorflow/tfjs-node-gpu if you have CUDA. Then call require('@tensorflow/tfjs-node'); (-gpu suffix for CUDA) at the start of your program. Visit https://github.com/tensorflow/tfjs-node for more details.
-============================`));var a={};return this.data.set(a,{values:e,dtype:r,refCount:1}),a},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return{dataId:a,shape:e,dtype:i}},t.prototype.incRef=function(e){var i=this.data.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.data.has(e)){var i=this.data.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){this.data.set(e,{values:i,dtype:a,refCount:1})},t.prototype.numDataIds=function(){return this.data.numDataIds()},t.prototype.read=function(e){return qw(this,void 0,void 0,function(){return Gw(this,function(i){return[2,this.readSync(e)]})})},t.prototype.readSync=function(e){var i=this.data.get(e),r=i.dtype,a=i.complexTensorInfos;if(r==="complex64"){var s=this.readSync(a.real.dataId),o=this.readSync(a.imag.dataId);return T.backend_util.mergeRealAndImagArrays(s,o)}return this.data.get(e).values},t.prototype.bufferSync=function(e){var i=this.readSync(e.dataId),r=i;if(e.dtype==="string")try{r=i.map(function(a){return T.util.decodeString(a)})}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return T.buffer(e.shape,e.dtype,r)},t.prototype.makeOutput=function(e,i,r){var a=this.write(e,i,r);return T.engine().makeTensorFromDataId(a,i,r,this)},t.prototype.disposeData=function(e){if(this.data.has(e)){var i=this.data.get(e).complexTensorInfos;i!=null&&(this.disposeData(i.real.dataId),this.disposeData(i.imag.dataId)),this.data.delete(e)}},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.data.has(i)){var r=this.data.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.time=function(e){return qw(this,void 0,void 0,function(){var i,r;return Gw(this,function(a){return i=T.util.now(),e(),r=T.util.now()-i,[2,{kernelMs:r}]})})},t.prototype.memory=function(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}},t.prototype.stridedSlice=function(e,i,r,a){ae(e,"stridedSlice");var s=T.slice_util.computeOutShape(i,r,a);if(s.some(function(p){return p===0}))return T.tensor([],s);for(var o=T.buffer(s,e.dtype),l=this.bufferSync(e),u=0;u<o.size;u++){for(var c=o.indexToLoc(u),h=new Array(c.length),d=0;d<h.length;d++)h[d]=c[d]*a[d]+i[d];o.set.apply(o,[l.get.apply(l,h)].concat(c))}return o.toTensor()},t.prototype.diag=function(e){for(var i=this.readSync(e.dataId),r=T.buffer([e.size,e.size],e.dtype),a=r.values,s=0;s<i.length;s++)a[s*e.size+s]=i[s];return r.toTensor()},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=T.slice(e,l,u).reshape(a);return c},t.prototype.reverse=function(e,i){ae(e,"reverse");for(var r=T.buffer(e.shape,e.dtype),a=this.bufferSync(e),s=function(l){var u=r.indexToLoc(l),c=u.slice();i.forEach(function(h){return c[h]=e.shape[h]-1-c[h]}),r.set.apply(r,[a.get.apply(a,c)].concat(u))},o=0;o<r.size;o++)s(o);return r.toTensor()},t.prototype.neg=function(e){return ae(e,"neg"),T.mul(T.scalar(-1),e)},t.prototype.addN=function(e){var i=this;ae(e,"addN");for(var r=e.map(function(c){return i.readSync(c.dataId)}),a=T.buffer(e[0].shape,e[0].dtype),s=a.values,o=0;o<e.length;o++)for(var l=r[o],u=0;u<s.length;u++)s[u]+=l[u];return a.toTensor()},t.prototype.softmax=function(e,i){var r=T.util.parseAxisParam([i],e.shape),a=T.max(e,r),s=T.backend_util.expandShapeToKeepDim(a.shape,r),o=T.sub(e,a.reshape(s)),l=T.exp(o),u=this.sum(l,r).reshape(s);return T.div(l,u)},t.prototype.pow=function(e,i){return ae([e,i],"pow"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.pow(r,a)})},t.prototype.batchMatMul=function(e,i,r,a){ae([e,i],"matMul");for(var s=r?e.shape[1]:e.shape[2],o=r?e.shape[2]:e.shape[1],l=a?i.shape[1]:i.shape[2],u=e.shape[0],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=r?[e.strides[0],1,e.strides[1]]:[e.strides[0],e.strides[1],1],p=d[0],f=d[1],m=d[2],g=a?[1,i.strides[1],i.strides[0]]:[i.strides[1],1,i.strides[0]],v=g[0],b=g[1],w=g[2],S=o*l,L=T.buffer([u,o,l],e.dtype),x=L.values,C=this.blockSize,R=0;R<u;R++)for(var D=0;D<o;D+=C)for(var k=0;k<l;k+=C)for(var W=0;W<s;W+=C)for(var F=Math.min(D+C,o),P=Math.min(k+C,l),H=Math.min(W+C,s),_=D;_<F;_++)for(var K=k;K<P;K++){for(var j=0,q=W;q<H;q++)j+=c[R*p+_*f+q*m]*h[q*v+K*b+R*w];x[R*S+(_*l+K)]+=j}return L.toTensor()},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=this.batchMatMul(i,r,a,s);return o&&(c=T.add(c,o)),l&&(c=wd(this,c,l,u)),c},t.prototype.floorDiv=function(e,i){ae([e,i],"floorDiv");var r=function(s,o){return Math.floor(s/o)},a="int32";return this.broadcastedBinaryOp(e,i,a,r)},t.prototype.sum=function(e,i){ae(e,"sum"),T.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=0,m=0;m<u;++m)f+=h[p+m];c[d]=f}return l},t.prototype.prod=function(e,i){ae(e,"sum");for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=1,m=0;m<u;++m)f*=h[p+m];c[d]=f}return l},t.prototype.unsortedSegmentSum=function(e,i,r){ae(e,"unsortedSegmentSum");for(var a=[],s=e.rank-i.rank,o=0;o<s;++o)i=i.expandDims(o+1);for(var o=0;o<r;++o){var l=T.scalar(o,"int32"),u=T.equal(l,i).asType("float32"),c=u.mul(e).sum(0);a.push(c)}return T.stack(a)},t.prototype.argMin=function(e,i){ae(e,"argMin");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMin",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v<f&&(f=v,m=g)}c[d]=m}return l},t.prototype.argMax=function(e,i){ae(e,"argMax");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMax",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v>f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;d<u.length;d+=c)for(var p=0;p<c;p++){var f=h(d,p);if(p===0)l[f]=r?0:u[f];else{var m=h(d,p-1);l[f]=r?u[m]+l[m]:u[f]+l[m]}}return o},t.prototype.equal=function(e,i){return ae([e,i],"equal"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r===a?1:0})},t.prototype.notEqual=function(e,i){return ae([e,i],"notEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r!==a?1:0})},t.prototype.less=function(e,i){return ae([e,i],"less"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<a?1:0})},t.prototype.lessEqual=function(e,i){return ae([e,i],"lessEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<=a?1:0})},t.prototype.greater=function(e,i){return ae([e,i],"greater"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>a?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d<a.length;d++)for(var p=0;p<h;p++)a[d]===1?u[c++]=s[d]:u[c++]=o[d];return l},t.prototype.where=function(e){ae([e],"where");var i=this.readSync(e.dataId);return dB(e.shape,i)},t.prototype.topk=function(e,i,r){ae(e,"topk");var a=this.readSync(e.dataId);return hB(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){ae(e,"min"),T.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];m<p&&(p=m)}u[h]=p}return o},t.prototype.minimum=function(e,i){return ae([e,i],"minimum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.min(r,a)})},t.prototype.mod=function(e,i){return ae([e,i],"mod"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r%a;return r<0&&a<0||r>=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p&&m}u[h]=p}return o},t.prototype.any=function(e,i){ae(e,"any"),T.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p||m}u[h]=p}return o},t.prototype.squaredDifference=function(e,i){return ae([e,i],"squaredDifference"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r-a;return s*s})},t.prototype.linear=function(e){return e},t.prototype.relu=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.max(0,a[s]);return i},t.prototype.relu6=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.min(Math.max(0,a[s]),6);return i},t.prototype.prelu=function(e,i){return ae([e,i],"prelu"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return r<0?a*r:r})},t.prototype.eluDer=function(e,i){ae([e,i],"eluDer");for(var r=new Float32Array(i.size),a=this.readSync(i.dataId),s=this.readSync(e.dataId),o=0;o<a.length;++o){var l=a[o];l>=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=wd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),x=this.readSync(i.dataId),C=d.values,R=0;R<r.batchSize;++R)for(var D=R*p,k=R*v,W=0;W<r.outHeight;++W)for(var F=k+W*b,P=W*r.strideHeight-c,H=0;H<a;H++){var _=P+H*o;if(_<0||_>=r.inHeight)continue;for(var K=H*i.strides[0],j=D+_*f,q=0;q<r.outWidth;++q)for(var G=F+q*w,Z=q*r.strideWidth-u,X=0;X<s;X++){var ee=Z+X*l;if(ee<0||ee>=r.inWidth)continue;for(var ne=K+X*i.strides[1],ie=j+ee*m,te=ne,re=0;re<r.inChannels;++re){for(var le=L[ie+re*g],he=0;he<r.outChannels;++he)C[G+he*S]+=le*x[te+he];te+=r.outChannels}}}return d.toTensor()},t.prototype.conv3d=function(e,i,r){for(var a=r.filterDepth,s=r.filterHeight,o=r.filterWidth,l=r.dilationDepth,u=r.dilationHeight,c=r.dilationWidth,h=r.padInfo.front,d=r.padInfo.left,p=r.padInfo.top,f=T.buffer(r.outShape,e.dtype),m=this.readSync(e.dataId),g=this.readSync(i.dataId),v=f.values,b=0;b<r.batchSize;++b)for(var w=b*e.strides[0],S=b*f.strides[0],L=0;L<r.outDepth;++L)for(var x=S+L*f.strides[1],C=L*r.strideDepth-h,R=0;R<a;R++){var D=C+R*l;if(D<0||D>=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F<r.outHeight;++F)for(var P=x+F*f.strides[2],H=F*r.strideHeight-p,_=0;_<s;_++){var K=H+_*u;if(K<0||K>=r.inHeight)continue;for(var j=k+_*i.strides[1],q=W+K*e.strides[2],G=0;G<r.outWidth;++G)for(var Z=P+G*r.outChannels,X=G*r.strideWidth-d,ee=0;ee<o;ee++){var ne=X+ee*c;if(ne<0||ne>=r.inWidth)continue;for(var ie=j+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le<r.inChannels;++le){for(var he=m[te+le],ye=0;ye<r.outChannels;++ye)v[Z+ye]+=he*g[re+ye];re+=r.outChannels}}}}return f.toTensor()},t.prototype.conv2dDerInput=function(e,i,r){ae([e,i],"conv2dDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=this.readSync(e.dataId),l=this.readSync(i.dataId),u=i.strides,c=u[0],h=u[1],d=u[2],p=r.batchSize,f=r.filterHeight,m=r.filterWidth,g=r.inChannels,v=r.inHeight,b=r.inWidth,w=r.outChannels,S=r.outHeight,L=r.outWidth,x=r.strideHeight,C=r.strideWidth,R=r.dataFormat,D=f-1-r.padInfo.top,k=m-1-r.padInfo.left,W=R==="channelsLast",F=a.strides[0],P=W?a.strides[1]:a.strides[2],H=W?a.strides[2]:1,_=W?1:a.strides[1],K=e.strides[0],j=W?e.strides[1]:e.strides[2],q=W?e.strides[2]:1,G=W?1:e.strides[1],Z=0;Z<p;++Z)for(var X=0;X<g;++X)for(var ee=0;ee<v;++ee)for(var ne=ee-D,ie=Math.max(0,Math.ceil(ne/x)),te=Math.min(S,(f+ne)/x),re=0;re<b;++re){for(var le=re-k,he=Math.max(0,Math.ceil(le/C)),ye=Math.min(L,(m+le)/C),Oe=0,Fe=ie;Fe<te;++Fe)for(var _e=Fe*x-ne,Me=he;Me<ye;++Me)for(var ht=Me*C-le,St=K*Z+j*Fe+q*Me,Ge=c*(f-1-_e)+h*(m-1-ht)+d*X,Ue=0;Ue<w;++Ue){var me=o[St+G*Ue],Dt=l[Ge+Ue];Oe+=me*Dt}var lt=F*Z+P*ee+H*re+_*X;s[lt]=Oe}return a.toTensor()},t.prototype.conv3dDerInput=function(e,i,r){for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=o[3],d=this.readSync(e.dataId),p=e.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=r.batchSize,D=r.filterDepth,k=r.filterHeight,W=r.filterWidth,F=r.inChannels,P=r.inDepth,H=r.inHeight,_=r.inWidth,K=r.outChannels,j=r.outDepth,q=r.outHeight,G=r.outWidth,Z=r.strideDepth,X=r.strideHeight,ee=r.strideWidth,ne=D-1-r.padInfo.front,ie=k-1-r.padInfo.top,te=W-1-r.padInfo.left,re=0;re<R;++re)for(var le=0;le<F;++le)for(var he=0;he<P;++he)for(var ye=he-ne,Oe=Math.max(0,Math.ceil(ye/Z)),Fe=Math.min(j,(D+ye)/Z),_e=0;_e<H;++_e)for(var Me=_e-ie,ht=Math.max(0,Math.ceil(Me/X)),St=Math.min(q,(k+Me)/X),Ge=0;Ge<_;++Ge){for(var Ue=Ge-te,me=Math.max(0,Math.ceil(Ue/ee)),Dt=Math.min(G,(W+Ue)/ee),lt=0,dt=Oe;dt<Fe;++dt)for(var xn=dt*Z-ye,Rt=ht;Rt<St;++Rt)for(var Lt=Rt*X-Me,Ce=me;Ce<Dt;++Ce)for(var kt=Ce*ee-Ue,pt=f*re+m*dt+g*Rt+v*Ce,it=S*(D-1-xn)+L*(k-1-Lt)+x*(W-1-kt)+C*le,on=0;on<K;++on){var hs=d[pt+on],Bt=b[it+on];lt+=hs*Bt}s[l*re+u*he+c*_e+h*Ge+le]=lt}return a.toTensor()},t.prototype.conv2dDerFilter=function(e,i,r){ae([e,i],"conv2dDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=r.dataFormat==="channelsLast",c=T.buffer(r.filterShape,"float32"),h=r.padInfo.left,d=r.padInfo.top,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((d-m)/a)),v=Math.min(r.outHeight,(r.inHeight+d-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((h-b)/s)),S=Math.min(r.outWidth,(r.inWidth+h-b)/s),L=0;L<r.inChannels;++L)for(var x=0;x<r.outChannels;++x){for(var C=0,R=0;R<r.batchSize;++R)for(var D=g;D<v;++D)for(var k=m+D*a-d,W=w;W<S;++W){var F=b+W*s-h;u?C+=p.get(R,k,F,L)*f.get(R,D,W,x):C+=p.get(R,L,k,F)*f.get(R,x,D,W)}c.set(C,m,b,L,x)}return c.toTensor()},t.prototype.conv3dDerFilter=function(e,i,r){for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=T.buffer(r.filterShape,"float32"),d=h.values,p=h.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=this.readSync(e.dataId),D=e.strides,k=D[0],W=D[1],F=D[2],P=D[3],H=r.padInfo.front,_=r.padInfo.left,K=r.padInfo.top,j=0;j<l;++j)for(var q=Math.max(0,Math.ceil((H-j)/a)),G=Math.min(r.outDepth,(r.inDepth+H-j)/a),Z=j*f,X=0;X<u;++X)for(var ee=Math.max(0,Math.ceil((K-X)/s)),ne=Math.min(r.outHeight,(r.inHeight+K-X)/s),ie=X*m+Z,te=0;te<c;++te)for(var re=Math.max(0,Math.ceil((_-te)/o)),le=Math.min(r.outWidth,(r.inWidth+_-te)/o),he=te*g+ie,ye=0;ye<r.inChannels;++ye)for(var Oe=ye*v+he,Fe=0;Fe<r.outChannels;++Fe){for(var _e=0,Me=0;Me<r.batchSize;++Me)for(var ht=Me*k,St=Me*S,Ge=q;Ge<G;++Ge)for(var Ue=j+Ge*a-H,me=Ue*W+ht,Dt=Ge*L+St,lt=ee;lt<ne;++lt)for(var dt=X+lt*s-K,xn=dt*F+me,Rt=lt*x+Dt,Lt=re;Lt<le;++Lt){var Ce=te+Lt*o-_,kt=Ce*P+xn,pt=Lt*C+Rt;_e+=R[kt+ye]*b[pt+Fe]}d[Oe+Fe]=_e}return h.toTensor()},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.depthwiseConv2D(i,r,a);return s&&(u=T.add(u,s)),o&&(u=wd(this,u,o,l)),u},t.prototype.depthwiseConv2D=function(e,i,r){ae([e,i],"depthwiseConv2D");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.outChannels/r.inChannels,d=T.buffer(r.outShape,e.dtype),p=this.readSync(e.dataId),f=this.readSync(i.dataId),m=d.values,g=0;g<r.batchSize;++g)for(var v=g*e.strides[0],b=g*d.strides[0],w=0;w<r.outHeight;++w)for(var S=b+w*d.strides[1],L=w*r.strideHeight-u,x=0;x<a;++x){var C=L+x*o;if(C<0||C>=r.inHeight)continue;for(var R=x*i.strides[0],D=v+C*e.strides[1],k=0;k<r.outWidth;++k)for(var W=S+k*d.strides[2],F=k*r.strideWidth-c,P=0;P<s;++P){var H=F+P*l;if(H<0||H>=r.inWidth)continue;for(var _=R+P*i.strides[1],K=D+H*r.inChannels,j=W,q=_,G=0;G<r.inChannels;++G){for(var Z=p[K+G],X=0;X<h;++X)m[j+X]+=Z*f[q+X];j+=h,q+=h}}}return d.toTensor()},t.prototype.depthwiseConv2DDerInput=function(e,i,r){ae([e,i],"depthwiseConv2DDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=this.readSync(e.dataId),d=e.strides,p=d[0],f=d[1],m=d[2],g=this.readSync(i.dataId),v=i.strides,b=v[0],w=v[1],S=v[2],L=r.batchSize,x=r.filterHeight,C=r.filterWidth,R=r.inChannels,D=r.inHeight,k=r.inWidth,W=r.outChannels,F=r.outHeight,P=r.outWidth,H=r.strideHeight,_=r.strideWidth,K=x-1-r.padInfo.top,j=C-1-r.padInfo.left,q=W/R,G=0;G<L;++G)for(var Z=0;Z<R;++Z)for(var X=0;X<D;++X)for(var ee=X-K,ne=Math.max(0,Math.ceil(ee/H)),ie=Math.min(F,(x+ee)/H),te=0;te<k;++te){for(var re=te-j,le=Math.max(0,Math.ceil(re/_)),he=Math.min(P,(C+re)/_),ye=0,Oe=ne;Oe<ie;++Oe)for(var Fe=Oe*H-ee,_e=le;_e<he;++_e)for(var Me=_e*_-re,ht=p*G+f*Oe+m*_e,St=b*(x-1-Fe)+w*(C-1-Me)+S*Z,Ge=0;Ge<q;++Ge){var Ue=Z*q+Ge,me=h[ht+Ue],Dt=g[St+Ge];ye+=me*Dt}s[l*G+u*X+c*te+Z]=ye}return a.toTensor()},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){ae([e,i],"depthwiseConv2DDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=T.buffer(r.filterShape,"float32"),c=r.padInfo.left,h=r.padInfo.top,d=r.outChannels/r.inChannels,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((h-m)/a)),v=Math.min(r.outHeight,(r.inHeight+h-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((c-b)/s)),S=Math.min(r.outWidth,(r.inWidth+c-b)/s),L=0;L<r.outChannels;++L){for(var x=Math.trunc(L/d),C=L%d,R=0,D=0;D<r.batchSize;++D)for(var k=g;k<v;++k)for(var W=m+k*a-h,F=w;F<S;++F){var P=b+F*s-c;R+=p.get(D,W,P,x)*f.get(D,k,F,L)}u.set(R,m,b,x,C)}return u.toTensor()},t.prototype.tile=function(e,i){return ae(e,"tile"),cB(this.bufferSync(e),i)},t.prototype.gather=function(e,i,r){ae([e,i],"gather");var a=e.shape.slice(),s=this.readSync(i.dataId);a[r]=s.length;for(var o=T.buffer(a,e.dtype),l=this.bufferSync(e),u=0;u<o.size;++u){var c=o.indexToLoc(u),h=c.slice();h[r]=s[c[r]];var d=l.locToIndex(h);o.values[u]=l.values[d]}return o.toTensor()},t.prototype.batchToSpaceND=function(e,i,r){ae([e],"batchToSpaceND");var a=i.reduce(function(h,d){return h*d}),s=T.backend_util.getReshaped(e.shape,i,a),o=T.backend_util.getPermuted(s.length,i.length),l=T.backend_util.getReshapedPermuted(e.shape,i,a),u=T.backend_util.getSliceBeginCoords(r,i.length),c=T.backend_util.getSliceSize(l,r,i.length);return T.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.pool3d=function(e,i,r){ae(e,"pool3d");for(var a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=r==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,b=this.readSync(e.dataId),w=T.buffer(i.outShape,e.dtype),S=w.values,L=i.outShape[1]*i.outShape[2]*i.outShape[3]*i.outShape[4],x=i.outShape[2]*i.outShape[3]*i.outShape[4],C=i.outShape[3]*i.outShape[4],R=i.outShape[4],D=0;D<i.batchSize;++D)for(var k=D*L,W=D*e.strides[0],F=0;F<i.inChannels;++F)for(var P=0;P<i.outDepth;++P){for(var H=P*a-f,_=H;_<0;)_+=l;for(var K=Math.min(i.inDepth,h+H),j=k+P*x,q=0;q<i.outHeight;++q){for(var G=q*s-m,Z=G;Z<0;)Z+=u;for(var X=Math.min(i.inHeight,d+G),ee=j+q*C,ne=0;ne<i.outWidth;++ne){for(var ie=ne*o-g,te=ie;te<0;)te+=c;for(var re=Math.min(i.inWidth,p+ie),le=ee+ne*R,he=v,ye=0,Oe=0,Fe=_;Fe<K;Fe+=l){for(var _e=W+Fe*e.strides[1],Me=Z;Me<X;Me+=u){for(var ht=_e+Me*e.strides[2],St=te;St<re;St+=c){var Ge=ht+St*e.strides[3],Ue=b[Ge+F];if(r==="max"&&Ue>he?he=Ue:r==="avg"&&(ye+=Ue,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),x=this.bufferSync(e),C=0;C<r.batchSize;++C)for(var R=0;R<r.inChannels;++R)for(var D=0;D<r.inDepth;++D)for(var k=0;k<r.inHeight;++k)for(var W=0;W<r.inWidth;++W){for(var F=D-v,P=k-w,H=W-b,_=0,K=0;K<f;K+=h){var j=(F+K)/a;if(j<0||j>=r.outDepth||Math.floor(j)!==j)continue;for(var q=0;q<m;q+=d){var G=(P+q)/s;if(G<0||G>=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z<g;Z+=p){var X=(H+Z)/o;if(X<0||X>=r.outWidth||Math.floor(X)!==X)continue;var ee=x.get(C,j,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b<i.batchSize;++b)for(var w=0;w<i.inChannels;++w)for(var S=0;S<i.outDepth;++S){for(var L=S*a-f,x=L;x<0;)x+=l;for(var C=Math.min(i.inDepth,h+L),R=0;R<i.outHeight;++R){for(var D=R*s-m,k=D;k<0;)k+=u;for(var W=Math.min(i.inHeight,d+D),F=0;F<i.outWidth;++F){for(var P=F*o-g,H=P;H<0;)H+=c;for(var _=Math.min(i.inWidth,p+P),K=Number.NEGATIVE_INFINITY,j=-1,q=x;q<C;q+=l)for(var G=q-L,Z=k;Z<W;Z+=u)for(var X=Z-D,ee=H;ee<_;ee+=c){var ne=ee-P,ie=v.get(b,q,Z,ee,w);ie>=K&&(K=ie,j=G*d*p+X*d+ne)}r.set(j,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),x=0;x<a.batchSize;++x)for(var C=0;C<a.inChannels;++C)for(var R=0;R<a.inDepth;++R)for(var D=0;D<a.inHeight;++D)for(var k=0;k<a.inWidth;++k){for(var W=R-g,F=D-b,P=k-v,H=0,_=0;_<p;_+=c){var K=(W+_)/o;if(K<0||K>=a.outDepth||Math.floor(K)!==K)continue;for(var j=0;j<f;j+=h){var q=(F+j)/l;if(q<0||q>=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<m;G+=d){var Z=(P+G)/u;if(Z<0||Z>=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(x,K,q,Z,C),ee=_*f*m+j*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(x,K,q,Z,C);H+=ie*ne}}}w.set(H,x,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b<o;b++)for(var w=0;w<i;w++)for(var S=g*w,L=Math.floor(S),x=S-L,C=Math.min(l-1,Math.ceil(S)),R=b*e.strides[0]+L*e.strides[1],D=b*e.strides[0]+C*e.strides[1],k=0;k<r;k++)for(var W=v*k,F=Math.floor(W),P=W-F,H=Math.min(u-1,Math.ceil(W)),_=R+F*e.strides[2],K=D+F*e.strides[2],j=R+H*e.strides[2],q=D+H*e.strides[2],G=0;G<c;G++){var Z=h[_+G],X=h[K+G],ee=h[j+G],ne=h[q+G],ie=Z+(ee-Z)*P,te=X+(ne-X)*P,re=ie+(te-ie)*x;d[m++]=re}return T.tensor(d,[o,i,r,c])},t.prototype.resizeBilinearBackprop=function(e,i,r){ae([e,i],"resizeBilinearBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=[r&&h>1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S<s;S++)for(var L=S*i.strides[0],x=0;x<h;x++)for(var C=x*g,R=Math.floor(C),D=Math.min(Math.ceil(C),o-1),k=L+R*i.strides[1],W=L+D*i.strides[1],F=C-R,P=1-F,H=0;H<d;H++)for(var _=H*v,K=Math.floor(_),j=Math.min(Math.ceil(_),l-1),q=_-K,G=1-q,Z=k+K*i.strides[2],X=k+j*i.strides[2],ee=W+K*i.strides[2],ne=W+j*i.strides[2],ie=P*G,te=P*q,re=F*G,le=F*q,he=0;he<u;he++){var ye=b[w++];p[Z+he]+=ye*ie,p[X+he]+=ye*te,p[ee+he]+=ye*re,p[ne+he]+=ye*le}return T.tensor4d(p,[s,l,o,u],i.dtype)},t.prototype.resizeNearestNeighbor=function(e,i,r,a){ae(e,"resizeNearestNeighbor");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(o*i*r*c),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b<o;b++)for(var w=b*e.strides[0],S=0;S<i;S++)for(var L=m*S,x=Math.min(l-1,a?Math.round(L):Math.floor(L)),C=w+x*e.strides[1],R=0;R<r;R++)for(var D=g*R,k=Math.min(u-1,a?Math.round(D):Math.floor(D)),W=C+k*e.strides[2],F=0;F<c;F++){var P=h[W+F];d[v++]=P}return T.tensor(d,[o,i,r,c],e.dtype)},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){ae([e,i],"resizeNearestNeighborBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=this.readSync(e.dataId),m=[r&&h>1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,x=Math.ceil(S)*2+2,C=0;C<s;C++)for(var R=C*i.strides[0],D=0;D<o;D++)for(var k=R+D*i.strides[1],W=Math.floor(D*w),F=Math.floor(W-L/2),P=0;P<l;P++)for(var H=k+P*i.strides[2],_=Math.floor(P*S),K=Math.floor(_-x/2),j=0;j<u;j++){for(var q=0,G=0;G<L;G++){var Z=G+F;if(Z<0||Z>=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie<x;ie++){var te=ie+K;if(te<0||te>=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+j])}}p[H+j]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p<c;p++){var f=d(p),m=u[p]*Math.pow(r+a*f,-s);h[p]=m}return T.tensor4d(h,e.shape)},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){ae(e,"LRNGrad");for(var u=e.shape[3],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=this.readSync(r.dataId),p=new Float32Array(e.size),f=e.size,m=0;m<f;m++){for(var g=m%u,v=m-g+Math.max(0,g-a),b=m-g+Math.min(u,g+a+1),w=0,S=v;S<b;S++)w+=Math.pow(h[S],2);w=o*w+s;for(var S=v;S<b;S++){var L=-2*o*l*h[S]*d[m]/w;m===S&&(L+=Math.pow(w,-l)),L*=c[m],p[S]+=L}}return T.tensor4d(p,e.shape)},t.prototype.multinomial=function(e,i,r,a){ae(e,"multinomial");for(var s=i?e:T.softmax(e),o=s.shape[0],l=s.shape[1],u=T.zeros([o,r],"int32"),c=this.readSync(u.dataId),h=this.readSync(s.dataId),d=0;d<o;++d){var p=d*l,f=new Float32Array(l-1);f[0]=h[p];for(var m=1;m<f.length;++m)f[m]=f[m-1]+h[p+m];for(var g=sB.alea(a.toString()),v=d*r,b=0;b<r;++b){var w=g();c[v+b]=f.length;for(var S=0;S<f.length;S++)if(w<f[S]){c[v+b]=S;break}}}return u},t.prototype.oneHot=function(e,i,r,a){ae(e,"oneHot");var s=new Float32Array(e.size*i);s.fill(a);for(var o=this.readSync(e.dataId),l=0;l<e.size;++l)o[l]>=0&&o[l]<i&&(s[l*i+o[l]]=r);return T.tensor2d(s,[e.size,i],"int32")},t.prototype.nonMaxSuppression=function(e,i,r,a,s){ae(e,"nonMaxSuppression");var o=this.readSync(e.dataId),l=this.readSync(i.dataId);return lB(o,l,r,a,s)},t.prototype.depthToSpace=function(e,i,r){T.util.assert(r==="NHWC",function(){return"Only NHWC dataFormat supported on CPU for depthToSpace. Got "+r}),T.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m<a;++m)for(var g=0;g<u;++g)for(var v=Math.floor(g/i),b=g%i,w=0;w<c;++w)for(var S=Math.floor(w/i),L=w%i,x=(b*i+L)*h,C=0;C<h;++C){var R=C+x,D=R+l*(S+o*(v+s*m));p[f++]=d[D]}return T.tensor4d(p,[a,u,c,h])},t.prototype.broadcastedBinaryOp=function(e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(e.shape,i.shape),o=T.buffer(s,r),l=this.readSync(e.dataId),u=this.readSync(i.dataId),c=T.backend_util.getBroadcastDims(e.shape,s),h=T.backend_util.getBroadcastDims(i.shape,s),d=o.values;if(c.length+h.length===0)for(var p=0;p<d.length;++p)d[p]=a(l[p%l.length],u[p%u.length]);else for(var f=this.bufferSync(e),m=this.bufferSync(i),g=function(v){var b=o.indexToLoc(v),w=b.slice(-e.rank);c.forEach(function(C){return w[C]=0});var S=f.locToIndex(w),L=b.slice(-i.rank);h.forEach(function(C){return L[C]=0});var x=m.locToIndex(L);d[v]=a(l[S],u[x])},p=0;p<d.length;++p)g(p);return o.toTensor()},t.prototype.split=function(e,i,r){return uB(e,i,r)},t.prototype.dispose=function(){},t.prototype.floatPrecision=function(){return 32},t.prototype.epsilon=function(){return n.prototype.epsilon.call(this)},t.prototype.cropAndResize=function(e,i,r,a,s,o){for(var l=e.shape,u=l[0],c=l[1],h=l[2],d=l[3],p=i.shape[0],f=a[0],m=a[1],g=T.buffer([p,f,m,d],"float32"),v=this.readSync(i.dataId),b=this.readSync(r.dataId),w=this.readSync(e.dataId),S=e.strides,L=g.strides,x=0;x<p;x++){var C=x*4,R=v[C],D=v[C+1],k=v[C+2],W=v[C+3],F=b[x];if(F>=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_<f;_++){var K=f>1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(K<0||K>c-1){for(var j=0;j<m;j++)for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}if(s==="bilinear")for(var Z=Math.floor(K),X=Math.ceil(K),ee=K-Z,j=0;j<m;j++){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var ie=Math.floor(ne),te=Math.ceil(ne),re=ne-ie,q=0;q<d;q++){var G=q+ie*S[2]+Z*S[1]+F*S[0],le=w[G];G=q+te*S[2]+Z*S[1]+F*S[0];var he=w[G];G=q+ie*S[2]+X*S[1]+F*S[0];var ye=w[G];G=q+te*S[2]+X*S[1]+F*S[0];var Oe=w[G],Fe=le+(he-le)*re,_e=ye+(Oe-ye)*re;G=q+j*L[2]+_*L[1]+x*L[0],g.values[G]=Fe+(_e-Fe)*ee}}else for(var j=0;j<m;++j){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var Me=Math.round(ne),ht=Math.round(K),q=0;q<d;q++){var St=q+Me*S[2]+ht*S[1]+F*S[0],Ge=q+j*L[2]+_*L[1]+x*L[0];g.values[Ge]=w[St]}}}}return g.toTensor()},t.prototype.sparseToDense=function(e,i,r,a){var s=T.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.sliceSize,c=s.strides,h=s.outputSize,d=!1;return this.scatter(e,i,r,h,u,l,o,c,a,d)},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=T.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3];if(l===0)return T.tensor([],o,e.dtype);for(var h=new T.TensorBuffer([l,u],e.dtype),d=this.readSync(i.dataId),p=this.readSync(e.dataId),f=0;f<l;f++){for(var m=[],g=0,v=0;v<a;v++){var b=d[f*a+v];g+=b*c[v],m.push(b)}if(g<0||g>=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w<u;w++)h.values[f*u+w]=p[g*u+w]}return h.toTensor().reshape(o)},t.prototype.scatterND=function(e,i,r){var a=T.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=T.scalar(0),d=!0;return this.scatter(e,i,r,c,l,o,s,u,h,d)},t.prototype.fill=function(e,i,r){r=r||T.util.inferDtype(i);var a=T.util.getArrayFromDType(r,T.util.sizeFromShape(e));return a.fill(i),T.engine().makeTensor(a,e,r,this)},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported for string tensors");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){var i=T.util.getArrayFromDType(e.dtype,T.util.sizeFromShape(e.shape));return this.makeOutput(i,e.shape,e.dtype)},t.prototype.linspace=function(e,i,r){return T.backend_util.linspaceImpl(e,i,r)},t.prototype.scatter=function(e,i,r,a,s,o,l,u,c,h){var d=[a/s,s],p=this.readSync(e.dataId),f=this.readSync(i.dataId);if(a===0)return T.tensor([],r,i.dtype);var m=new T.TensorBuffer(d,i.dtype);m.values.fill(this.readSync(c.dataId)[0]);for(var g=0;g<o;g++){for(var v=[],b=0,w=0;w<l;w++){var S=p[g*l+w];v.push(S),b+=S*u[w]}if(b<0||b>=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;L<s;L++)h?m.values[b*s+L]+=f[g*s+L]:m.values[b*s+L]=i.rank===0?f[0]:f[g*s+L]}return m.toTensor().reshape(r)},t}(T.KernelBackend);function Kw(n){for(var t=new Float32Array(n.length),e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}var pB=function(n){var t=n.inputs.x,e=n.backend,i=new Float32Array(T.util.sizeFromShape(t.shape));if(t.dtype!=="complex64"){var r=e.data.get(t.dataId).values;i=Kw(r)}else for(var a=e.data.get(t.dataId),s=a.complexTensorInfos.real,o=a.complexTensorInfos.imag,l=e.data.get(s.dataId).values,u=e.data.get(o.dataId).values,c=0;c<l.length;c++){var h=l[c],d=u[c];i[c]=Math.hypot(h,d)}return e.makeOutput(i,t.shape,"float32")},fB={kernelName:T.Abs,backendName:"cpu",kernelFunc:pB};function lr(n){return function(t,e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=T.util.computeStrides(s),u=T.util.sizeFromShape(s),c=T.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=T.util.computeStrides(t),f=T.util.computeStrides(e),m=T.backend_util.getBroadcastDims(t,s),g=T.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(var v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(var b=function(w){var S=T.util.indexToLoc(w,o,l),L=S.slice(-h);m.forEach(function(D){return L[D]=0});var x=T.util.locToIndex(L,h,p),C=S.slice(-d);g.forEach(function(D){return C[D]=0});var R=T.util.locToIndex(C,d,f);c[w]=n(i[x],r[R])},v=0;v<c.length;++v)b(v);return[c,s]}}function In(n){var t=n.inputs,e=n.backend,i=t.real,r=t.imag,a=e.data.get(i.dataId).values,s=e.data.get(r.dataId).values,o=e.makeTensorInfo(i.shape,"complex64"),l=e.data.get(o.dataId);return l.complexTensorInfos={real:e.makeTensorInfo(i.shape,"float32",a),imag:e.makeTensorInfo(r.shape,"float32",s)},o}var mB={kernelName:T.Complex,backendName:"cpu",kernelFunc:In};function Xr(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var gB={kernelName:T.Identity,backendName:"cpu",kernelFunc:Xr};function $a(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.real,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var vB={kernelName:T.Real,backendName:"cpu",kernelFunc:$a};function Xa(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.dtype;if(a==="complex64"){if(r.dtype==="complex64")return Xr({inputs:{x:r},backend:e});var s=T.zeros(r.shape),o=Xa({inputs:{x:r},backend:e,attrs:{dtype:"float32"}}),l=In({inputs:{real:o,imag:s},backend:e});return s.dispose(),e.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){var u=$a({inputs:{input:r},backend:e}),l=Xa({inputs:{x:u},backend:e,attrs:{dtype:a}});return e.disposeIntermediateTensorInfo(u),l}if(!T.util.hasEncodingLoss(r.dtype,a)){var l=Xr({inputs:{x:r},backend:e});return{dataId:l.dataId,shape:l.shape,dtype:a}}if(a==="int32"){var c=e.data.get(r.dataId).values,h=Int32Array.from(c);return e.makeTensorInfo(r.shape,"int32",h)}if(a==="bool"){var d=e.data.get(r.dataId).values,p=T.util.toTypedArray([0],r.dtype),f=lr(function(v,b){return v!==b?1:0})(r.shape,[],d,p,"bool"),m=f[0],g=f[1];return e.makeTensorInfo(g,"bool",m)}throw new Error("Error in Cast: failed to cast "+r.dtype+" to "+a)}var yB={kernelName:T.Cast,backendName:"cpu",kernelFunc:Xa};function Jr(n,t,e,i){return e==null?function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;ae([l,u],n);var h=c.data.get(l.dataId).values,d=c.data.get(u.dataId).values,p=i||l.dtype,f=t(l.shape,u.shape,h,d,p),m=f[0],g=f[1];return c.makeTensorInfo(g,p,m)}:function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;if(l.dtype==="complex64"||u.dtype==="complex64"){var h=Xa({inputs:{x:l},backend:c,attrs:{dtype:"complex64"}}),d=c.data.get(h.dataId),p=d.complexTensorInfos.real,f=d.complexTensorInfos.imag,m=c.data.get(p.dataId).values,g=c.data.get(f.dataId).values,v=Xa({inputs:{x:u},backend:c,attrs:{dtype:"complex64"}}),b=c.data.get(v.dataId),w=b.complexTensorInfos.real,S=b.complexTensorInfos.imag,L=c.data.get(w.dataId).values,x=c.data.get(S.dataId).values,C=e(l.shape,u.shape,m,g,L,x),R=C[0],D=C[1],k=C[2],W=c.makeTensorInfo(k,"float32",R),F=c.makeTensorInfo(k,"float32",D),P=In({inputs:{real:W,imag:F},backend:c});return c.disposeIntermediateTensorInfo(h),c.disposeIntermediateTensorInfo(v),c.disposeIntermediateTensorInfo(W),c.disposeIntermediateTensorInfo(F),P}else{var H=c.data.get(l.dataId).values,_=c.data.get(u.dataId).values,K=i||l.dtype,j=t(l.shape,u.shape,H,_,K),q=j[0],k=j[1];return c.makeTensorInfo(k,K,q)}}}function Sd(n){return function(t,e,i,r,a,s){var o=T.backend_util.assertAndGetBroadcastShape(t,e),l=T.util.sizeFromShape(o),u=o.length,c=T.util.computeStrides(o),h=T.util.getTypedArrayFromDType("float32",l),d=T.util.getTypedArrayFromDType("float32",l),p=T.backend_util.getBroadcastDims(t,o),f=T.backend_util.getBroadcastDims(e,o),m=T.backend_util.mergeRealAndImagArrays(i,r),g=T.backend_util.mergeRealAndImagArrays(a,s),v=t.length,b=T.util.computeStrides(t),w=e.length,S=T.util.computeStrides(e);if(p.length+f.length===0)for(var L=0;L<h.length;L++){var x=L%m.length,C=L%g.length,R=n(m[x*2],m[x*2+1],g[C*2],g[C*2+1]);h[L]=R.real,d[L]=R.imag}else for(var D=function(k){var W=T.util.indexToLoc(k,u,c),F=W.slice(-v);p.forEach(function(j){return F[j]=0});var P=T.util.locToIndex(F,v,b),H=W.slice(-w);f.forEach(function(j){return H[j]=0});var _=T.util.locToIndex(H,w,S),K=n(m[P*2],m[P*2+1],g[_*2],g[_*2+1]);h[k]=K.real,d[k]=K.imag},L=0;L<h.length;L++)D(L);return[h,d,o]}}var jw=lr(function(n,t){return n+t}),bB=Sd(function(n,t,e,i){return{real:n+e,imag:t+i}}),$w=Jr(T.Add,jw,bB),wB={kernelName:T.Add,backendName:"cpu",kernelFunc:$w};function Zr(n){return function(t,e,i){for(var r=T.util.getTypedArrayFromDType(e,t.length),a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}function $e(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");for(var l=s,u=l.data.get(o.dataId).values,c=T.util.sizeFromShape(o.shape),h=e||o.dtype,d=T.util.getArrayFromDType(h,c),p=0;p<c;++p)d[p]=t(u[p],a);return l.makeTensorInfo(o.shape,h,d)}}function Qr(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");var l=s,u=l.data.get(o.dataId).values,c=e||o.dtype,h=t(u,c,a);return l.makeTensorInfo(o.shape,c,h)}}var Xw=Zr(function(n){return Math.ceil(n)}),SB=Qr(T.Ceil,Xw),LB={kernelName:T.Ceil,backendName:"cpu",kernelFunc:SB};var Jw=Zr(function(n){return Math.exp(n)}),IB=Qr(T.Exp,Jw),AB={kernelName:T.Exp,backendName:"cpu",kernelFunc:IB};var Zw=Zr(function(n){return Math.expm1(n)}),TB=Qr(T.Expm1,Zw),NB={kernelName:T.Expm1,backendName:"cpu",kernelFunc:TB};var Qw=Zr(function(n){return Math.floor(n)}),xB=Qr(T.Floor,Qw),CB={kernelName:T.Floor,backendName:"cpu",kernelFunc:xB};var e0=Zr(function(n){return Math.log(n)}),RB=Qr(T.Log,e0),OB={kernelName:T.Log,backendName:"cpu",kernelFunc:RB};function t0(n,t,e,i){for(var r=T.util.getTypedArrayFromDType(i,T.util.sizeFromShape(e)),a=0;a<r.length;++a){for(var s=a*t,o=n[s],l=0;l<t;++l){var u=n[s+l];u>o&&(o=u)}r[a]=o}return r}var n0=lr(function(n,t){return n*t}),EB=Sd(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),i0=Jr(T.Multiply,n0,EB),DB={kernelName:T.Multiply,backendName:"cpu",kernelFunc:i0};var r0=Zr(function(n){return 1/Math.sqrt(n)}),kB=Qr(T.Rsqrt,r0),FB={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:kB};function a0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;c<s;++c){var h=e.length,d=T.util.computeStrides(e),p=T.util.indexToLoc(c,h,d),f=p.map(function(g,v){return g+t[v]}),m=T.util.locToIndex(f,i.length,o);u[c]=n[m]}return u}function Ld(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.begin,s=i.size;ae(r,"slice");var o=T.slice_util.parseSliceParams(r,a,s),l=o[0],u=o[1];T.slice_util.assertParamsValid(r,l,u);var c=e.data.get(r.dataId).values,h=a0(c,l,u,r.shape,r.dtype);return e.makeTensorInfo(u,r.dtype,h)}var WB={kernelName:T.Slice,backendName:"cpu",kernelFunc:Ld};var s0=lr(function(n,t){return n-t}),UB=Sd(function(n,t,e,i){return{real:n-e,imag:t-i}}),o0=Jr(T.Sub,s0,UB),BB={kernelName:T.Sub,backendName:"cpu",kernelFunc:o0};function Id(n,t,e,i,r){for(var a=t.length,s=T.util.sizeFromShape(t),o=T.util.computeStrides(t),l=T.util.computeStrides(r),u=T.util.getTypedArrayFromDType(e,T.util.sizeFromShape(r)),c=0;c<s;++c){for(var h=T.util.indexToLoc(c,a,o),d=new Array(h.length),p=0;p<d.length;p++)d[p]=h[i[p]];var f=T.util.locToIndex(d,a,l);u[f]=n[c]}return u}function l0(n,t,e,i){for(var r=T.util.parseAxisParam(t,e)[0],a=[1,e[0],1],s=0;s<r;s++)a[0]*=e[s];a[1]=e[r];for(var s=r+1;s<e.length;s++)a[2]*=e[s];for(var o={},l=new Int32Array(e[r]),u=new T.TensorBuffer(a,i,n),c=[],h=a[0]===1&&a[2]===1,s=0;s<e[r];s++){var d=void 0;if(h)d=n[s].toString();else{for(var p=[],f=0;f<a[0];f++)for(var m=0;m<a[2];m++)p.push(u.get(f,s,m));d=p.join(",")}if(o[d]!==void 0)l[s]=o[d];else{var g=Object.keys(o).length;o[d]=g,l[s]=g,c.push(s)}}var v=a.slice();v[1]=Object.keys(o).length;var b=new T.TensorBuffer(v,i);c.forEach(function(S,L){for(var x=0;x<a[0];x++)for(var C=0;C<a[2];C++)b.set(u.get(x,S,C),x,L,C)});var w=e.slice();return w[r]=v[1],{outputValues:b.values,outputShape:w,indices:l}}var zB={__proto__:null,simpleAbsImpl:Kw,addImpl:jw,ceilImpl:Xw,expImpl:Jw,expm1Impl:Zw,floorImpl:Qw,logImpl:e0,maxImpl:t0,multiplyImpl:n0,rsqrtImpl:r0,sliceImpl:a0,subImpl:s0,transposeImpl:Id,uniqueImpl:l0};var PB="2.6.0";T.registerBackend("cpu",function(){return new Yw},1);var _B=$e(T.Acos,function(n){return Math.acos(n)}),MB={kernelName:T.Acos,backendName:"cpu",kernelFunc:_B};var HB=$e(T.Acosh,function(n){return Math.acosh(n)}),VB={kernelName:T.Acosh,backendName:"cpu",kernelFunc:HB};var qB=$e(T.Asin,function(n){return Math.asin(n)}),GB={kernelName:T.Asin,backendName:"cpu",kernelFunc:qB};var YB=$e(T.Asinh,function(n){return Math.asinh(n)}),KB={kernelName:T.Asinh,backendName:"cpu",kernelFunc:YB};var jB=$e(T.Atan,function(n){return Math.atan(n)}),$B={kernelName:T.Atan,backendName:"cpu",kernelFunc:jB};var XB=$e(T.Atanh,function(n){return Math.atanh(n)}),JB={kernelName:T.Atanh,backendName:"cpu",kernelFunc:XB};function Ad(n,t,e,i,r,a){for(var s=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,u=r.dilationWidth,c=r.effectiveFilterHeight,h=r.effectiveFilterWidth,d=r.padInfo.top,p=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=T.buffer(r.outShape,e),g=m.values,v=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],w=r.outShape[3],S=0;S<r.batchSize;++S)for(var L=S*v,x=S*i[0],C=0;C<r.inChannels;++C)for(var R=0;R<r.outHeight;++R)for(var D=R*s-d,k=Math.max(0,D),W=Math.min(r.inHeight,c+D),F=L+R*b,P=0;P<r.outWidth;++P){for(var H=P*o-p,_=Math.max(0,H),K=Math.min(r.inWidth,h+H),j=f,q=0,G=0,Z=k;Z<W;Z+=l){for(var X=x+Z*i[1],ee=_;ee<K;ee+=u){var ne=X+ee*i[2],ie=n[ne+C];a==="max"&&ie>j?j=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(j))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:j}return m}function u0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;g<i.batchSize;++g)for(var v=0;v<i.inChannels;++v)for(var b=0;b<i.outHeight;++b){for(var w=b*o-p,S=w;S<0;)S+=u;for(var L=Math.min(i.inHeight,h+w),x=0;x<i.outWidth;++x){for(var C=x*l-f,R=C;R<0;)R+=c;for(var D=Math.min(i.inWidth,d+C),k=Number.NEGATIVE_INFINITY,W=-1,F=S;F<L;F+=u)for(var P=F-w,H=R;H<D;H+=c){var _=H-C,K=m.get(g,F,H,v);K>k&&(k=K,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,x,v)}}return s}function ZB(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Ad(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var QB={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:ZB};function ez(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),x=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D<c.batchSize;++D)for(var k=0;k<c.inChannels;++k)for(var W=0;W<c.inHeight;++W)for(var F=0;F<c.inWidth;++F){for(var P=W-S,H=F-w,_=0,K=0;K<v;K+=m){var j=(P+K)/h;if(j<0||j>=c.outHeight||Math.floor(j)!==j)continue;for(var q=0;q<b;q+=g){var G=(H+q)/d;if(G<0||G>=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,j,G,k);_+=Z}}L.set(_*x,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var tz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:ez};function nz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,x=0,C=0,R=0;R<c.length;++R)m[R]=f[S++]+(c[R]-h[L++])*p[x++]/Math.sqrt(d[C++]+u),S>=g&&(S=0),L>=w&&(L=0),x>=v&&(x=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var iz={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:nz};var rz=$e(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n<e.clipValueMin?e.clipValueMin:n}),az={kernelName:T.ClipByValue,backendName:"cpu",kernelFunc:rz};function Eo(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.imag,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var sz={kernelName:T.Imag,backendName:"cpu",kernelFunc:Eo};function Ai(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=T.util.sizeFromShape(r.shape),o=T.util.inferFromImplicitShape(a,s),l=T.util.sizeFromShape(o);T.util.assert(s===l,function(){return"The new shape ("+o+") has "+l+" elements and the old "+("shape ("+r.shape+") has "+s+" elements. The new shape and old ")+"shape must have the same number of elements."}),e.incRef(r.dataId);var u=e.data.get(r.dataId);if(u.complexTensorInfos!=null){var c=u.complexTensorInfos.real,h=u.complexTensorInfos.imag;c.shape=o,h.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var oz={kernelName:T.Reshape,backendName:"cpu",kernelFunc:Ai};function Ja(n){var t=n.inputs,e=n.backend,i=n.attrs,r=i.axis,a=T.util.parseAxisParam(r,t[0].shape)[0],s=T.backend_util.computeOutShape(t.map(function(S){return S.shape}),a);if(T.util.sizeFromShape(s)===0)return e.makeTensorInfo(s,t[0].dtype,[]);var o=t.filter(function(S){return T.util.sizeFromShape(S.shape)>0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return $a({inputs:{input:S},backend:e})}),c=o.map(function(S){return Eo({inputs:{input:S},backend:e})}),h=Ja({inputs:u,backend:e,attrs:{axis:r}}),d=Ja({inputs:c,backend:e,attrs:{axis:r}}),p=In({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),x=[-1,L];return Ai({inputs:{x:S},backend:e,attrs:{shape:x}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,x=T.util.sizeFromShape(S.shape);m.set(L,g),g+=x})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,x=0,C=0;C<S.shape[0];++C)for(var R=C*s[1]+v,D=0;D<S.shape[1];++D)m[R+D]=L[x++];v+=S.shape[1]})}var b=T.backend_util.computeOutShape(o.map(function(S){return S.shape}),a),w=e.makeTensorInfo(b,t[0].dtype,m);return f.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),w}var lz={kernelName:T.Concat,backendName:"cpu",kernelFunc:Ja};var uz=$e(T.Cos,function(n){return Math.cos(n)}),cz={kernelName:T.Cos,backendName:"cpu",kernelFunc:uz};var hz=$e(T.Cosh,function(n){return Math.cosh(n)}),dz={kernelName:T.Cosh,backendName:"cpu",kernelFunc:hz};var pz={kernelName:T.Dilation2D,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=i,l=o.strides,u=o.pad,c=o.dilations,h=e,d=h.data.get(a.dataId).values,p=a.shape.length,f=h.data.get(s.dataId).values,m=s.shape.length,g=T.backend_util.computeDilation2DInfo(a.shape,s.shape,l,u,"NHWC",c),v=g.batchSize,b=g.inHeight,w=g.inWidth,S=g.inChannels,L=g.outHeight,x=g.outWidth,C=g.padInfo,R=g.strideHeight,D=g.strideWidth,k=g.filterHeight,W=g.filterWidth,F=g.dilationHeight,P=g.dilationWidth,H=g.outShape,_=T.util.sizeFromShape(H),K=H.length,j=T.util.getArrayFromDType(a.dtype,_),q=0;q<v;++q)for(var G=0;G<L;++G)for(var Z=G*R-C.top,X=0;X<x;++X)for(var ee=X*D-C.left,ne=0;ne<S;++ne){for(var ie=Number.MIN_SAFE_INTEGER,te=0;te<k;++te){var re=Z+te*F;if(re>=0&&re<b)for(var le=0;le<W;++le){var he=ee+le*P;if(he>=0&&he<w){var ye=T.util.locToIndex([q,re,he,ne],p,T.util.computeStrides(a.shape)),Oe=T.util.locToIndex([te,le,ne],m,T.util.computeStrides(s.shape)),Fe=d[ye]+f[Oe];Fe>ie&&(ie=Fe)}}}var _e=T.util.locToIndex([q,G,X,ne],K,T.util.computeStrides(H));j[_e]=ie}var Me=h.write(T.util.toTypedArray(j,a.dtype),H,a.dtype);return{dataId:Me,shape:H,dtype:a.dtype}}};var fz={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=0,ie=0,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var mz={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=q<0?0:q,ie=Z<0?0:Z,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=re,ie=he)}}}_[K][ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var gz=lr(function(n,t){return n/t}),vz=Jr(T.Div,gz),Td={kernelName:T.Div,backendName:"cpu",kernelFunc:vz};var yz=$e(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),bz={kernelName:T.Elu,backendName:"cpu",kernelFunc:yz};var wz=T.backend_util.ERF_P,Sz=T.backend_util.ERF_A1,Lz=T.backend_util.ERF_A2,Iz=T.backend_util.ERF_A3,Az=T.backend_util.ERF_A4,Tz=T.backend_util.ERF_A5,Nz=$e(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+wz*e);return t*(1-((((Tz*i+Az)*i+Iz)*i+Lz)*i+Sz)*i*Math.exp(-e*e))}),xz={kernelName:T.Erf,backendName:"cpu",kernelFunc:Nz};function c0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p<r;p++){for(var f=Ld({inputs:{x:o},backend:e,attrs:{begin:[p,0],size:[1,a]}}),m=Ld({inputs:{x:l},backend:e,attrs:{begin:[p,0],size:[1,a]}}),g=In({inputs:{real:f,imag:m},backend:e}),v=Cz(g,t,e),b=v.real,w=v.imag,S=T.backend_util.mergeRealAndImagArrays(b,w),L=0;L<a;L++){var x=T.backend_util.getComplexWithIndex(S,L);h[p*a+L]=x.real,d[p*a+L]=x.imag}e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(g)}var C=e.makeTensorInfo(u,"float32",h),R=e.makeTensorInfo(u,"float32",d),D=In({inputs:{real:C,imag:R},backend:e});return e.disposeIntermediateTensorInfo(C),e.disposeIntermediateTensorInfo(R),D}function Cz(n,t,e){var i=T.util.sizeFromShape(n.shape),r=e.data.get(n.dataId),a=e.data.get(r.complexTensorInfos.real.dataId).values,s=e.data.get(r.complexTensorInfos.imag.dataId).values;if(Rz(i)){var o=Nd(a,s,i,t,e),l=[n.shape[0],n.shape[1]];if(t){var u=e.makeTensorInfo(l,"float32",o.real),c=e.makeTensorInfo(l,"float32",o.imag),h=e.makeTensorInfo([],"float32",T.util.createScalarValue(i,"float32")),d=Xr({inputs:{x:h},backend:e}),p=Td.kernelFunc({inputs:{a:u,b:h},backend:e}),f=Td.kernelFunc({inputs:{a:c,b:d},backend:e}),m=e.data.get(p.dataId).values,g=e.data.get(f.dataId).values;return e.disposeIntermediateTensorInfo(u),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return o}else{var v=T.backend_util.mergeRealAndImagArrays(a,s),b=Oz(v,i,t);return T.backend_util.splitRealAndImagArrays(b)}}function Rz(n){return(n&n-1)===0}function Nd(n,t,e,i,r){if(e===1)return{real:n,imag:t};var a=T.backend_util.mergeRealAndImagArrays(n,t),s=e/2,o=T.backend_util.complexWithEvenIndex(a),l=o.real,u=o.imag,c=[l.length],h=r.makeTensorInfo(c,"float32",l),d=r.makeTensorInfo(c,"float32",u),p=In({inputs:{real:h,imag:d},backend:r}),f=T.backend_util.complexWithOddIndex(a),m=f.real,g=f.imag,v=[m.length],b=r.makeTensorInfo(v,"float32",m),w=r.makeTensorInfo(v,"float32",g),S=In({inputs:{real:b,imag:w},backend:r}),L=Nd(l,u,s,i,r),x=L.real,C=L.imag,R=[x.length],D=r.makeTensorInfo(R,"float32",x),k=r.makeTensorInfo(R,"float32",C),W=In({inputs:{real:D,imag:k},backend:r}),F=Nd(m,g,s,i,r),P=F.real,H=F.imag,_=[P.length],K=r.makeTensorInfo(_,"float32",P),j=r.makeTensorInfo(_,"float32",H),q=In({inputs:{real:K,imag:j},backend:r}),G=T.backend_util.exponents(e,i),Z=[G.real.length],X=r.makeTensorInfo(Z,"float32",G.real),ee=r.makeTensorInfo(Z,"float32",G.imag),ne=In({inputs:{real:X,imag:ee},backend:r}),ie=i0({inputs:{a:ne,b:q},backend:r}),te=$w({inputs:{a:W,b:ie},backend:r}),re=o0({inputs:{a:W,b:ie},backend:r}),le=$a({inputs:{input:te},backend:r}),he=$a({inputs:{input:re},backend:r}),ye=Eo({inputs:{input:te},backend:r}),Oe=Eo({inputs:{input:re},backend:r}),Fe=Ja({inputs:[le,he],backend:r,attrs:{axis:0}}),_e=Ja({inputs:[ye,Oe],backend:r,attrs:{axis:0}}),Me=r.data.get(Fe.dataId).values,ht=r.data.get(_e.dataId).values;return r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(w),r.disposeIntermediateTensorInfo(S),r.disposeIntermediateTensorInfo(D),r.disposeIntermediateTensorInfo(k),r.disposeIntermediateTensorInfo(W),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(q),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(ie),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(re),r.disposeIntermediateTensorInfo(le),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(he),r.disposeIntermediateTensorInfo(Oe),r.disposeIntermediateTensorInfo(Fe),r.disposeIntermediateTensorInfo(_e),{real:Me,imag:ht}}function Oz(n,t,e){for(var i=new Float32Array(t*2),r=0;r<t;r++){for(var a=0,s=0,o=0;o<t;o++){var l=T.backend_util.exponent(r*o,t,e),u=T.backend_util.getComplexWithIndex(n,o);a+=u.real*l.real-u.imag*l.imag,s+=u.real*l.imag+u.imag*l.real}e&&(a/=t,s/=t),T.backend_util.assignToTypedArray(i,a,s,r)}return i}function Ez(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!1,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Dz={kernelName:T.FFT,backendName:"cpu",kernelFunc:Ez};var kz={kernelName:T.FlipLeftRight,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=i,s=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),o=r.shape,l=o[0],u=o[1],c=o[2],h=o[3],d=a.data.get(r.dataId).values,p=0;p<l;p++)for(var f=p*c*u*h,m=0;m<u;m++)for(var g=m*(c*h),v=0;v<c;v++)for(var b=v*h,w=0;w<h;w++){var S=[l,m,v,w],L=S[2],x=Math.round(c-L),C=f+g+b+w,R=d[C];if(x>=0&&x<c){var D=x*h,k=f+g+D+w;R=d[k]}s[C]=R}var W=a.write(s,r.shape,r.dtype);return{dataId:W,shape:r.shape,dtype:r.dtype}}};function Fz(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!0,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Wz={kernelName:T.IFFT,backendName:"cpu",kernelFunc:Fz};var Uz=$e(T.IsFinite,function(n){return Number.isFinite(n)?1:0},"bool"),Bz={kernelName:T.IsFinite,backendName:"cpu",kernelFunc:Uz};var zz=$e(T.IsInf,function(n){return Math.abs(n)===Infinity?1:0},"bool"),Pz={kernelName:T.IsInf,backendName:"cpu",kernelFunc:zz};var _z=$e(T.IsNan,function(n){return Number.isNaN(n)?1:0},"bool"),Mz={kernelName:T.IsNan,backendName:"cpu",kernelFunc:_z};var Hz=$e(T.Log1p,function(n){return Math.log1p(n)}),Vz={kernelName:T.Log1p,backendName:"cpu",kernelFunc:Hz};var qz=$e(T.LogicalNot,function(n){return n?0:1},"bool"),Gz={kernelName:T.LogicalNot,backendName:"cpu",kernelFunc:qz};var Yz={kernelName:T.Max,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape,c=u.length,h=T.util.parseAxisParam(s,u),d=h,p=T.backend_util.getAxesPermutation(d,c),f=l.data.get(r.dataId).values;if(p!=null){for(var m=new Array(c),g=0;g<m.length;g++)m[g]=u[p[g]];f=Id(f,u,r.dtype,p,m),d=T.backend_util.getInnerMostAxes(d.length,c),u=m}ae(r,"max"),T.backend_util.assertAxesAreInnerMostDims("max",d,c);var v=T.backend_util.computeOutAndReduceShapes(u,d),b=v[0],w=v[1],S=T.util.sizeFromShape(w),L=t0(f,S,b,r.dtype),x=l.write(L,b,r.dtype),C=b;if(o){var m=T.backend_util.expandShapeToKeepDim(b,h);C=m}return{dataId:x,shape:C,dtype:r.dtype}}};function Kz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Ad(d,r.shape,r.dtype,p,c,"max");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var jz={kernelName:T.MaxPool,backendName:"cpu",kernelFunc:Kz};function $z(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ae([a,s],"maxPoolBackprop");for(var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=T.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=e.data.get(o.dataId).values,f=T.buffer(d.outShape,o.dtype,u0(p,o.shape,o.dtype,d).values),m=d.strideHeight,g=d.strideWidth,v=d.dilationHeight,b=d.dilationWidth,w=d.effectiveFilterHeight,S=d.effectiveFilterWidth,L=S-1-d.padInfo.left,x=w-1-d.padInfo.top,C=T.buffer(o.shape,"float32"),R=e.data.get(r.dataId).values,D=T.buffer(r.shape,"float32",R),k=0;k<d.batchSize;++k)for(var W=0;W<d.inChannels;++W)for(var F=0;F<d.inHeight;++F)for(var P=0;P<d.inWidth;++P){for(var H=F-x,_=P-L,K=0,j=0;j<w;j+=v){var q=(H+j)/m;if(q<0||q>=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<S;G+=b){var Z=(_+G)/g;if(Z<0||Z>=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=j*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);K+=ie*ne}}C.set(K,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var Xz={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:$z};function Jz(n,t,e,i,r){var a=T.util.computeStrides(t),s=Ad(n,t,e,a,r,"max"),o=u0(n,t,e,r,!0,i);return[s.values,o.values]}var Zz={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=Jz(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var Qz=T.kernel_impls.nonMaxSuppressionV4Impl,eP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=Qz(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var tP=T.kernel_impls.nonMaxSuppressionV5Impl,nP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=tP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var iP=lr(function(n,t){return n!==t?1:0}),rP=Jr(T.NotEqual,iP,null,"bool"),aP={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:rP};function sP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(x,C){return x[0]+r.shape[C]+x[1]}),l=a.map(function(x){return x[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v<c;v++){var b=T.util.indexToLoc(v,h,d),w=b.map(function(x,C){return x+l[C]}),S=T.util.locToIndex(w,f,m);g[S]=u[v]}var L=e.write(g,o,r.dtype);return{dataId:L,shape:o,dtype:r.dtype}}var h0={kernelName:T.PadV2,backendName:"cpu",kernelFunc:sP};var oP=$e(T.Reciprocal,function(n){return 1/n}),lP={kernelName:T.Reciprocal,backendName:"cpu",kernelFunc:oP};var uP={kernelName:T.RotateWithOffset,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),h=r.shape,d=h[0],p=h[1],f=h[2],m=h[3],g=T.backend_util.getImageCenter(l,p,f),v=g[0],b=g[1],w=255,S=Math.sin(s),L=Math.cos(s),x=u.data.get(r.dataId).values,C=0;C<d;C++)for(var R=C*f*p*m,D=0;D<p;D++)for(var k=D*(f*m),W=0;W<f;W++)for(var F=W*m,P=0;P<m;P++){var H=[d,D,W,P],_=H[2],K=H[1],j=(_-v)*L-(K-b)*S,q=(_-v)*S+(K-b)*L;j=Math.round(j+v),q=Math.round(q+b);var G=o;if(typeof o!="number"&&(P===3?G=w:G=o[P]),j>=0&&j<f&&q>=0&&q<p){var Z=q*(f*m),X=j*m,ee=R+Z+X+P;G=x[ee]}var ne=R+k+F+P;c[ne]=G}var ie=u.write(c,r.shape,r.dtype);return{dataId:ie,shape:r.shape,dtype:r.dtype}}};var cP=$e(T.Round,function(n){var t=Math.floor(n);return n-t<.5?Math.floor(n):n-t>.5?Math.ceil(n):t%2===0?t:t+1}),hP={kernelName:T.Round,backendName:"cpu",kernelFunc:cP};var dP=T.backend_util.SELU_SCALEALPHA,pP=T.backend_util.SELU_SCALE,fP=$e(T.Selu,function(n){return n>=0?pP*n:dP*(Math.exp(n)-1)}),mP={kernelName:T.Selu,backendName:"cpu",kernelFunc:fP};var gP=$e(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),vP={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:gP};var yP=$e(T.Sign,function(n){return n<0?-1:n>0?1:0}),bP={kernelName:T.Sign,backendName:"cpu",kernelFunc:yP};var wP=$e(T.Sin,function(n){return Math.sin(n)}),SP={kernelName:T.Sin,backendName:"cpu",kernelFunc:wP};var LP=$e(T.Sinh,function(n){return Math.sinh(n)}),IP={kernelName:T.Sinh,backendName:"cpu",kernelFunc:LP};var AP=11920928955078125e-23,d0=Math.log(AP)+2,TP=$e(T.Softplus,function(n){var t=n>-d0,e=n<d0,i=Math.exp(n),r;return e?r=i:t?r=n:r=Math.log(1+i),r}),NP={kernelName:T.Softplus,backendName:"cpu",kernelFunc:TP};function p0(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm;ae(r,"transpose");for(var s=r.shape.length,o=new Array(s),l=0;l<o.length;l++)o[l]=r.shape[a[l]];var u=i.data.get(r.dataId).values,c=Id(u,r.shape,r.dtype,a,o),h=i.write(c,o,r.dtype);return{dataId:h,shape:o,dtype:r.dtype}}var xP={kernelName:T.Transpose,backendName:"cpu",kernelFunc:p0};function CP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.blockShape,s=i.paddings;ae([r],"spaceToBatchND");var o=T.util.sizeFromShape(a),l=[[0,0]];l.push.apply(l,s);for(var u=1+a.length;u<r.shape.length;++u)l.push([0,0]);var c=h0.kernelFunc({inputs:{x:r},backend:e,attrs:{paddings:l,constantValue:0}}),h=T.backend_util.getReshaped(c.shape,a,o,!1),d=T.backend_util.getPermuted(h.length,a.length,!1),p=T.backend_util.getReshapedPermuted(c.shape,a,o,!1),f={x:c},m={shape:h},g=Ai({inputs:f,backend:e,attrs:m}),v={x:g},b={perm:d},w=p0({inputs:v,backend:e,attrs:b}),S={x:w},L={shape:p},x=Ai({inputs:S,backend:e,attrs:L});return e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(w),x}var RP={kernelName:T.SpaceToBatchND,backendName:"cpu",kernelFunc:CP};var OP=$e(T.Sqrt,function(n){return Math.sqrt(n)}),EP={kernelName:T.Sqrt,backendName:"cpu",kernelFunc:OP};var DP={kernelName:T.Square,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.x,r=e;ae(i,"square");for(var a=r.data.get(i.dataId).values,s=new Float32Array(a.length),o=0;o<a.length;++o){var l=a[o];s[o]=l*l}var u=r.write(s,i.shape,i.dtype);return{dataId:u,shape:i.shape,dtype:i.dtype}}};var kP=lr(function(n,t){var e=n-t;return e*e}),FP=Jr(T.SquaredDifference,kP),WP={kernelName:T.SquaredDifference,backendName:"cpu",kernelFunc:FP};var UP=$e(T.Step,function(n,t){var e=t;return isNaN(n)?NaN:n>0?1:e.alpha}),BP={kernelName:T.Step,backendName:"cpu",kernelFunc:UP};var zP=$e(T.Tan,function(n){return Math.tan(n)}),PP={kernelName:T.Tan,backendName:"cpu",kernelFunc:zP};var _P=$e(T.Tanh,function(n){return Math.tanh(n)}),MP={kernelName:T.Tanh,backendName:"cpu",kernelFunc:_P};function HP(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=l0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var VP={kernelName:T.Unique,backendName:"cpu",kernelFunc:HP};var qP=[fB,MB,VB,wB,GB,KB,$B,JB,QB,tz,iz,yB,LB,az,mB,lz,cz,dz,pz,mz,fz,Td,bz,xz,AB,NB,Dz,kz,CB,gB,Wz,sz,Bz,Pz,Mz,OB,Vz,Gz,jz,Xz,Zz,Yz,DB,eP,nP,aP,h0,vB,lP,oz,uP,hP,FB,mP,vP,bP,SP,IP,WB,NP,RP,EP,DP,WP,BP,BB,PP,MP,xP,VP];for(var xd=0,f0=qP;xd<f0.length;xd++){var GP=f0[xd];T.registerKernel(GP)}ja.MathBackendCPU=Yw;ja.shared=zB;ja.version_cpu=PB});var JS=be(Pn=>{"use strict";Object.defineProperty(Pn,"__esModule",{value:!0});var N=Xi();var Cd=function(n,t){return Cd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Cd(n,t)};function YP(n,t){Cd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Do(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function ko(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var ur={},Rd={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function g0(n,t){ur[n]=t}function _n(n){if(!(n in ur)){var t=KP(n);if(t!==null)ur[n]=t;else return console.log("Could not get context for WebGL version",n),null}var e=ur[n];return e.isContextLost()?(delete ur[n],_n(n)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),ur[n])}function jP(n){if(typeof OffscreenCanvas!="undefined"&&n===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function KP(n){if(n!==1&&n!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");var t=jP(n);return t.addEventListener("webglcontextlost",function(e){e.preventDefault(),delete ur[n]},!1),n===1?t.getContext("webgl",Rd)||t.getContext("experimental-webgl",Rd):t.getContext("webgl2",Rd)}var Za;(function(n){n[n.DENSE=0]="DENSE",n[n.SHARED_BATCH=1]="SHARED_BATCH"})(Za||(Za={}));var rn;(function(n){n[n.RENDER=0]="RENDER",n[n.UPLOAD=1]="UPLOAD",n[n.PIXELS=2]="PIXELS",n[n.DOWNLOAD=3]="DOWNLOAD"})(rn||(rn={}));var Ct;(function(n){n[n.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",n[n.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",n[n.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",n[n.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",n[n.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Ct||(Ct={}));function Qa(n,t){return[t,n]}function $P(n,t){return n*t}function es(n){var t=N.util.sizeFromShape(n),e=Math.ceil(t/4);return N.util.sizeToSquarishShape(e)}function ea(n,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(n/2))]}function XP(n,t){var e=ea(n,t),i=e[0],r=e[1];return i*r*4}function Od(n,t){var e=n,i,r,a,s,o,l,u,c,h,d;return N.env().getNumber("WEBGL_VERSION")===2?(i=e.R32F,r=e.R16F,a=e.RGBA16F,s=e.RGBA32F,o=e.RED,u=4,c=1,h=e.HALF_FLOAT,d=e.FLOAT):(i=n.RGBA,r=n.RGBA,a=n.RGBA,s=e.RGBA,o=n.RGBA,u=4,c=4,h=t!=null?t.HALF_FLOAT_OES:null,d=n.FLOAT),l=n.RGBA,{internalFormatFloat:i,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:s,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:u,defaultNumChannels:c,textureTypeHalfFloat:h,textureTypeFloat:d}}function ce(n,t){var e=t();return N.env().getBool("DEBUG")&&JP(n),e}function JP(n){var t=n.getError();if(t!==n.NO_ERROR)throw new Error("WebGL Error: "+v0(n,t))}var ZP=596e-10,QP=65504;function y0(n){return!!(N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||n===0||ZP<Math.abs(n)&&Math.abs(n)<QP)}function v0(n,t){switch(t){case n.NO_ERROR:return"NO_ERROR";case n.INVALID_ENUM:return"INVALID_ENUM";case n.INVALID_VALUE:return"INVALID_VALUE";case n.INVALID_OPERATION:return"INVALID_OPERATION";case n.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case n.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case n.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return"Unknown error code "+t}}function ts(n,t){return ei(n,function(){return n.getExtension(t)},'Extension "'+t+'" not supported on this browser.')}function b0(n,t){var e=ei(n,function(){return n.createShader(n.VERTEX_SHADER)},"Unable to create vertex WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw console.log(n.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function w0(n,t){var e=ei(n,function(){return n.createShader(n.FRAGMENT_SHADER)},"Unable to create fragment WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw e_(t,n.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var t_=/ERROR: [0-9]+:([0-9]+):/g;function e_(n,t){var e=t_.exec(t);if(e==null){console.log("Couldn't parse line number in error: "+t),console.log(n);return}for(var i=+e[1],r=n.split(`
+============================`));var a={};return this.data.set(a,{values:e,dtype:r,refCount:1}),a},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return{dataId:a,shape:e,dtype:i}},t.prototype.incRef=function(e){var i=this.data.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.data.has(e)){var i=this.data.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){this.data.set(e,{values:i,dtype:a,refCount:1})},t.prototype.numDataIds=function(){return this.data.numDataIds()},t.prototype.read=function(e){return qw(this,void 0,void 0,function(){return Gw(this,function(i){return[2,this.readSync(e)]})})},t.prototype.readSync=function(e){var i=this.data.get(e),r=i.dtype,a=i.complexTensorInfos;if(r==="complex64"){var s=this.readSync(a.real.dataId),o=this.readSync(a.imag.dataId);return T.backend_util.mergeRealAndImagArrays(s,o)}return this.data.get(e).values},t.prototype.bufferSync=function(e){var i=this.readSync(e.dataId),r=i;if(e.dtype==="string")try{r=i.map(function(a){return T.util.decodeString(a)})}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return T.buffer(e.shape,e.dtype,r)},t.prototype.makeOutput=function(e,i,r){var a=this.write(e,i,r);return T.engine().makeTensorFromDataId(a,i,r,this)},t.prototype.disposeData=function(e){if(this.data.has(e)){var i=this.data.get(e).complexTensorInfos;i!=null&&(this.disposeData(i.real.dataId),this.disposeData(i.imag.dataId)),this.data.delete(e)}},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.data.has(i)){var r=this.data.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.time=function(e){return qw(this,void 0,void 0,function(){var i,r;return Gw(this,function(a){return i=T.util.now(),e(),r=T.util.now()-i,[2,{kernelMs:r}]})})},t.prototype.memory=function(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}},t.prototype.stridedSlice=function(e,i,r,a){ae(e,"stridedSlice");var s=T.slice_util.computeOutShape(i,r,a);if(s.some(function(p){return p===0}))return T.tensor([],s);for(var o=T.buffer(s,e.dtype),l=this.bufferSync(e),u=0;u<o.size;u++){for(var c=o.indexToLoc(u),h=new Array(c.length),d=0;d<h.length;d++)h[d]=c[d]*a[d]+i[d];o.set.apply(o,[l.get.apply(l,h)].concat(c))}return o.toTensor()},t.prototype.diag=function(e){for(var i=this.readSync(e.dataId),r=T.buffer([e.size,e.size],e.dtype),a=r.values,s=0;s<i.length;s++)a[s*e.size+s]=i[s];return r.toTensor()},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=T.slice(e,l,u).reshape(a);return c},t.prototype.reverse=function(e,i){ae(e,"reverse");for(var r=T.buffer(e.shape,e.dtype),a=this.bufferSync(e),s=function(l){var u=r.indexToLoc(l),c=u.slice();i.forEach(function(h){return c[h]=e.shape[h]-1-c[h]}),r.set.apply(r,[a.get.apply(a,c)].concat(u))},o=0;o<r.size;o++)s(o);return r.toTensor()},t.prototype.neg=function(e){return ae(e,"neg"),T.mul(T.scalar(-1),e)},t.prototype.addN=function(e){var i=this;ae(e,"addN");for(var r=e.map(function(c){return i.readSync(c.dataId)}),a=T.buffer(e[0].shape,e[0].dtype),s=a.values,o=0;o<e.length;o++)for(var l=r[o],u=0;u<s.length;u++)s[u]+=l[u];return a.toTensor()},t.prototype.softmax=function(e,i){var r=T.util.parseAxisParam([i],e.shape),a=T.max(e,r),s=T.backend_util.expandShapeToKeepDim(a.shape,r),o=T.sub(e,a.reshape(s)),l=T.exp(o),u=this.sum(l,r).reshape(s);return T.div(l,u)},t.prototype.pow=function(e,i){return ae([e,i],"pow"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.pow(r,a)})},t.prototype.batchMatMul=function(e,i,r,a){ae([e,i],"matMul");for(var s=r?e.shape[1]:e.shape[2],o=r?e.shape[2]:e.shape[1],l=a?i.shape[1]:i.shape[2],u=e.shape[0],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=r?[e.strides[0],1,e.strides[1]]:[e.strides[0],e.strides[1],1],p=d[0],f=d[1],m=d[2],g=a?[1,i.strides[1],i.strides[0]]:[i.strides[1],1,i.strides[0]],v=g[0],b=g[1],w=g[2],S=o*l,L=T.buffer([u,o,l],e.dtype),x=L.values,C=this.blockSize,R=0;R<u;R++)for(var D=0;D<o;D+=C)for(var k=0;k<l;k+=C)for(var W=0;W<s;W+=C)for(var F=Math.min(D+C,o),P=Math.min(k+C,l),H=Math.min(W+C,s),_=D;_<F;_++)for(var K=k;K<P;K++){for(var j=0,q=W;q<H;q++)j+=c[R*p+_*f+q*m]*h[q*v+K*b+R*w];x[R*S+(_*l+K)]+=j}return L.toTensor()},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=this.batchMatMul(i,r,a,s);return o&&(c=T.add(c,o)),l&&(c=Sd(this,c,l,u)),c},t.prototype.floorDiv=function(e,i){ae([e,i],"floorDiv");var r=function(s,o){return Math.floor(s/o)},a="int32";return this.broadcastedBinaryOp(e,i,a,r)},t.prototype.sum=function(e,i){ae(e,"sum"),T.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=0,m=0;m<u;++m)f+=h[p+m];c[d]=f}return l},t.prototype.prod=function(e,i){ae(e,"sum");for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.upcastType(e.dtype,"int32"),l=T.zeros(a,o),u=T.util.sizeFromShape(s),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=1,m=0;m<u;++m)f*=h[p+m];c[d]=f}return l},t.prototype.unsortedSegmentSum=function(e,i,r){ae(e,"unsortedSegmentSum");for(var a=[],s=e.rank-i.rank,o=0;o<s;++o)i=i.expandDims(o+1);for(var o=0;o<r;++o){var l=T.scalar(o,"int32"),u=T.equal(l,i).asType("float32"),c=u.mul(e).sum(0);a.push(c)}return T.stack(a)},t.prototype.argMin=function(e,i){ae(e,"argMin");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMin",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v<f&&(f=v,m=g)}c[d]=m}return l},t.prototype.argMax=function(e,i){ae(e,"argMax");var r=[i];T.backend_util.assertAxesAreInnerMostDims("argMax",r,e.rank);for(var a=T.backend_util.computeOutAndReduceShapes(e.shape,r),s=a[0],o=a[1],l=T.zeros(s,"int32"),u=T.util.sizeFromShape(o),c=this.readSync(l.dataId),h=this.readSync(e.dataId),d=0;d<c.length;++d){for(var p=d*u,f=h[p],m=0,g=0;g<u;++g){var v=h[p+g];v>f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;d<u.length;d+=c)for(var p=0;p<c;p++){var f=h(d,p);if(p===0)l[f]=r?0:u[f];else{var m=h(d,p-1);l[f]=r?u[m]+l[m]:u[f]+l[m]}}return o},t.prototype.equal=function(e,i){return ae([e,i],"equal"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r===a?1:0})},t.prototype.notEqual=function(e,i){return ae([e,i],"notEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r!==a?1:0})},t.prototype.less=function(e,i){return ae([e,i],"less"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<a?1:0})},t.prototype.lessEqual=function(e,i){return ae([e,i],"lessEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r<=a?1:0})},t.prototype.greater=function(e,i){return ae([e,i],"greater"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>a?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d<a.length;d++)for(var p=0;p<h;p++)a[d]===1?u[c++]=s[d]:u[c++]=o[d];return l},t.prototype.where=function(e){ae([e],"where");var i=this.readSync(e.dataId);return dB(e.shape,i)},t.prototype.topk=function(e,i,r){ae(e,"topk");var a=this.readSync(e.dataId);return hB(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){ae(e,"min"),T.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];m<p&&(p=m)}u[h]=p}return o},t.prototype.minimum=function(e,i){return ae([e,i],"minimum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.min(r,a)})},t.prototype.mod=function(e,i){return ae([e,i],"mod"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r%a;return r<0&&a<0||r>=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p&&m}u[h]=p}return o},t.prototype.any=function(e,i){ae(e,"any"),T.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h<u.length;++h){for(var d=h*l,p=c[d],f=0;f<l;++f){var m=c[d+f];p=p||m}u[h]=p}return o},t.prototype.squaredDifference=function(e,i){return ae([e,i],"squaredDifference"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){var s=r-a;return s*s})},t.prototype.linear=function(e){return e},t.prototype.relu=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.max(0,a[s]);return i},t.prototype.relu6=function(e){ae(e,"relu");for(var i=T.zeros(e.shape,e.dtype),r=this.readSync(i.dataId),a=this.readSync(e.dataId),s=0;s<a.length;++s)r[s]=Math.min(Math.max(0,a[s]),6);return i},t.prototype.prelu=function(e,i){return ae([e,i],"prelu"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return r<0?a*r:r})},t.prototype.eluDer=function(e,i){ae([e,i],"eluDer");for(var r=new Float32Array(i.size),a=this.readSync(i.dataId),s=this.readSync(e.dataId),o=0;o<a.length;++o){var l=a[o];l>=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Sd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),x=this.readSync(i.dataId),C=d.values,R=0;R<r.batchSize;++R)for(var D=R*p,k=R*v,W=0;W<r.outHeight;++W)for(var F=k+W*b,P=W*r.strideHeight-c,H=0;H<a;H++){var _=P+H*o;if(_<0||_>=r.inHeight)continue;for(var K=H*i.strides[0],j=D+_*f,q=0;q<r.outWidth;++q)for(var G=F+q*w,Z=q*r.strideWidth-u,X=0;X<s;X++){var ee=Z+X*l;if(ee<0||ee>=r.inWidth)continue;for(var ne=K+X*i.strides[1],ie=j+ee*m,te=ne,re=0;re<r.inChannels;++re){for(var le=L[ie+re*g],he=0;he<r.outChannels;++he)C[G+he*S]+=le*x[te+he];te+=r.outChannels}}}return d.toTensor()},t.prototype.conv3d=function(e,i,r){for(var a=r.filterDepth,s=r.filterHeight,o=r.filterWidth,l=r.dilationDepth,u=r.dilationHeight,c=r.dilationWidth,h=r.padInfo.front,d=r.padInfo.left,p=r.padInfo.top,f=T.buffer(r.outShape,e.dtype),m=this.readSync(e.dataId),g=this.readSync(i.dataId),v=f.values,b=0;b<r.batchSize;++b)for(var w=b*e.strides[0],S=b*f.strides[0],L=0;L<r.outDepth;++L)for(var x=S+L*f.strides[1],C=L*r.strideDepth-h,R=0;R<a;R++){var D=C+R*l;if(D<0||D>=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F<r.outHeight;++F)for(var P=x+F*f.strides[2],H=F*r.strideHeight-p,_=0;_<s;_++){var K=H+_*u;if(K<0||K>=r.inHeight)continue;for(var j=k+_*i.strides[1],q=W+K*e.strides[2],G=0;G<r.outWidth;++G)for(var Z=P+G*r.outChannels,X=G*r.strideWidth-d,ee=0;ee<o;ee++){var ne=X+ee*c;if(ne<0||ne>=r.inWidth)continue;for(var ie=j+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le<r.inChannels;++le){for(var he=m[te+le],ye=0;ye<r.outChannels;++ye)v[Z+ye]+=he*g[re+ye];re+=r.outChannels}}}}return f.toTensor()},t.prototype.conv2dDerInput=function(e,i,r){ae([e,i],"conv2dDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=this.readSync(e.dataId),l=this.readSync(i.dataId),u=i.strides,c=u[0],h=u[1],d=u[2],p=r.batchSize,f=r.filterHeight,m=r.filterWidth,g=r.inChannels,v=r.inHeight,b=r.inWidth,w=r.outChannels,S=r.outHeight,L=r.outWidth,x=r.strideHeight,C=r.strideWidth,R=r.dataFormat,D=f-1-r.padInfo.top,k=m-1-r.padInfo.left,W=R==="channelsLast",F=a.strides[0],P=W?a.strides[1]:a.strides[2],H=W?a.strides[2]:1,_=W?1:a.strides[1],K=e.strides[0],j=W?e.strides[1]:e.strides[2],q=W?e.strides[2]:1,G=W?1:e.strides[1],Z=0;Z<p;++Z)for(var X=0;X<g;++X)for(var ee=0;ee<v;++ee)for(var ne=ee-D,ie=Math.max(0,Math.ceil(ne/x)),te=Math.min(S,(f+ne)/x),re=0;re<b;++re){for(var le=re-k,he=Math.max(0,Math.ceil(le/C)),ye=Math.min(L,(m+le)/C),Oe=0,Fe=ie;Fe<te;++Fe)for(var _e=Fe*x-ne,Me=he;Me<ye;++Me)for(var ht=Me*C-le,St=K*Z+j*Fe+q*Me,Ge=c*(f-1-_e)+h*(m-1-ht)+d*X,Ue=0;Ue<w;++Ue){var me=o[St+G*Ue],Dt=l[Ge+Ue];Oe+=me*Dt}var lt=F*Z+P*ee+H*re+_*X;s[lt]=Oe}return a.toTensor()},t.prototype.conv3dDerInput=function(e,i,r){for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=o[3],d=this.readSync(e.dataId),p=e.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=r.batchSize,D=r.filterDepth,k=r.filterHeight,W=r.filterWidth,F=r.inChannels,P=r.inDepth,H=r.inHeight,_=r.inWidth,K=r.outChannels,j=r.outDepth,q=r.outHeight,G=r.outWidth,Z=r.strideDepth,X=r.strideHeight,ee=r.strideWidth,ne=D-1-r.padInfo.front,ie=k-1-r.padInfo.top,te=W-1-r.padInfo.left,re=0;re<R;++re)for(var le=0;le<F;++le)for(var he=0;he<P;++he)for(var ye=he-ne,Oe=Math.max(0,Math.ceil(ye/Z)),Fe=Math.min(j,(D+ye)/Z),_e=0;_e<H;++_e)for(var Me=_e-ie,ht=Math.max(0,Math.ceil(Me/X)),St=Math.min(q,(k+Me)/X),Ge=0;Ge<_;++Ge){for(var Ue=Ge-te,me=Math.max(0,Math.ceil(Ue/ee)),Dt=Math.min(G,(W+Ue)/ee),lt=0,dt=Oe;dt<Fe;++dt)for(var xn=dt*Z-ye,Rt=ht;Rt<St;++Rt)for(var Lt=Rt*X-Me,Ce=me;Ce<Dt;++Ce)for(var kt=Ce*ee-Ue,pt=f*re+m*dt+g*Rt+v*Ce,it=S*(D-1-xn)+L*(k-1-Lt)+x*(W-1-kt)+C*le,sn=0;sn<K;++sn){var hs=d[pt+sn],Bt=b[it+sn];lt+=hs*Bt}s[l*re+u*he+c*_e+h*Ge+le]=lt}return a.toTensor()},t.prototype.conv2dDerFilter=function(e,i,r){ae([e,i],"conv2dDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=r.dataFormat==="channelsLast",c=T.buffer(r.filterShape,"float32"),h=r.padInfo.left,d=r.padInfo.top,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((d-m)/a)),v=Math.min(r.outHeight,(r.inHeight+d-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((h-b)/s)),S=Math.min(r.outWidth,(r.inWidth+h-b)/s),L=0;L<r.inChannels;++L)for(var x=0;x<r.outChannels;++x){for(var C=0,R=0;R<r.batchSize;++R)for(var D=g;D<v;++D)for(var k=m+D*a-d,W=w;W<S;++W){var F=b+W*s-h;u?C+=p.get(R,k,F,L)*f.get(R,D,W,x):C+=p.get(R,L,k,F)*f.get(R,x,D,W)}c.set(C,m,b,L,x)}return c.toTensor()},t.prototype.conv3dDerFilter=function(e,i,r){for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=T.buffer(r.filterShape,"float32"),d=h.values,p=h.strides,f=p[0],m=p[1],g=p[2],v=p[3],b=this.readSync(i.dataId),w=i.strides,S=w[0],L=w[1],x=w[2],C=w[3],R=this.readSync(e.dataId),D=e.strides,k=D[0],W=D[1],F=D[2],P=D[3],H=r.padInfo.front,_=r.padInfo.left,K=r.padInfo.top,j=0;j<l;++j)for(var q=Math.max(0,Math.ceil((H-j)/a)),G=Math.min(r.outDepth,(r.inDepth+H-j)/a),Z=j*f,X=0;X<u;++X)for(var ee=Math.max(0,Math.ceil((K-X)/s)),ne=Math.min(r.outHeight,(r.inHeight+K-X)/s),ie=X*m+Z,te=0;te<c;++te)for(var re=Math.max(0,Math.ceil((_-te)/o)),le=Math.min(r.outWidth,(r.inWidth+_-te)/o),he=te*g+ie,ye=0;ye<r.inChannels;++ye)for(var Oe=ye*v+he,Fe=0;Fe<r.outChannels;++Fe){for(var _e=0,Me=0;Me<r.batchSize;++Me)for(var ht=Me*k,St=Me*S,Ge=q;Ge<G;++Ge)for(var Ue=j+Ge*a-H,me=Ue*W+ht,Dt=Ge*L+St,lt=ee;lt<ne;++lt)for(var dt=X+lt*s-K,xn=dt*F+me,Rt=lt*x+Dt,Lt=re;Lt<le;++Lt){var Ce=te+Lt*o-_,kt=Ce*P+xn,pt=Lt*C+Rt;_e+=R[kt+ye]*b[pt+Fe]}d[Oe+Fe]=_e}return h.toTensor()},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.depthwiseConv2D(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Sd(this,u,o,l)),u},t.prototype.depthwiseConv2D=function(e,i,r){ae([e,i],"depthwiseConv2D");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.outChannels/r.inChannels,d=T.buffer(r.outShape,e.dtype),p=this.readSync(e.dataId),f=this.readSync(i.dataId),m=d.values,g=0;g<r.batchSize;++g)for(var v=g*e.strides[0],b=g*d.strides[0],w=0;w<r.outHeight;++w)for(var S=b+w*d.strides[1],L=w*r.strideHeight-u,x=0;x<a;++x){var C=L+x*o;if(C<0||C>=r.inHeight)continue;for(var R=x*i.strides[0],D=v+C*e.strides[1],k=0;k<r.outWidth;++k)for(var W=S+k*d.strides[2],F=k*r.strideWidth-c,P=0;P<s;++P){var H=F+P*l;if(H<0||H>=r.inWidth)continue;for(var _=R+P*i.strides[1],K=D+H*r.inChannels,j=W,q=_,G=0;G<r.inChannels;++G){for(var Z=p[K+G],X=0;X<h;++X)m[j+X]+=Z*f[q+X];j+=h,q+=h}}}return d.toTensor()},t.prototype.depthwiseConv2DDerInput=function(e,i,r){ae([e,i],"depthwiseConv2DDerInput");for(var a=T.buffer(r.inShape,"float32"),s=a.values,o=a.strides,l=o[0],u=o[1],c=o[2],h=this.readSync(e.dataId),d=e.strides,p=d[0],f=d[1],m=d[2],g=this.readSync(i.dataId),v=i.strides,b=v[0],w=v[1],S=v[2],L=r.batchSize,x=r.filterHeight,C=r.filterWidth,R=r.inChannels,D=r.inHeight,k=r.inWidth,W=r.outChannels,F=r.outHeight,P=r.outWidth,H=r.strideHeight,_=r.strideWidth,K=x-1-r.padInfo.top,j=C-1-r.padInfo.left,q=W/R,G=0;G<L;++G)for(var Z=0;Z<R;++Z)for(var X=0;X<D;++X)for(var ee=X-K,ne=Math.max(0,Math.ceil(ee/H)),ie=Math.min(F,(x+ee)/H),te=0;te<k;++te){for(var re=te-j,le=Math.max(0,Math.ceil(re/_)),he=Math.min(P,(C+re)/_),ye=0,Oe=ne;Oe<ie;++Oe)for(var Fe=Oe*H-ee,_e=le;_e<he;++_e)for(var Me=_e*_-re,ht=p*G+f*Oe+m*_e,St=b*(x-1-Fe)+w*(C-1-Me)+S*Z,Ge=0;Ge<q;++Ge){var Ue=Z*q+Ge,me=h[ht+Ue],Dt=g[St+Ge];ye+=me*Dt}s[l*G+u*X+c*te+Z]=ye}return a.toTensor()},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){ae([e,i],"depthwiseConv2DDerFilter");for(var a=r.strideHeight,s=r.strideWidth,o=r.filterHeight,l=r.filterWidth,u=T.buffer(r.filterShape,"float32"),c=r.padInfo.left,h=r.padInfo.top,d=r.outChannels/r.inChannels,p=this.bufferSync(e),f=this.bufferSync(i),m=0;m<o;++m)for(var g=Math.max(0,Math.ceil((h-m)/a)),v=Math.min(r.outHeight,(r.inHeight+h-m)/a),b=0;b<l;++b)for(var w=Math.max(0,Math.ceil((c-b)/s)),S=Math.min(r.outWidth,(r.inWidth+c-b)/s),L=0;L<r.outChannels;++L){for(var x=Math.trunc(L/d),C=L%d,R=0,D=0;D<r.batchSize;++D)for(var k=g;k<v;++k)for(var W=m+k*a-h,F=w;F<S;++F){var P=b+F*s-c;R+=p.get(D,W,P,x)*f.get(D,k,F,L)}u.set(R,m,b,x,C)}return u.toTensor()},t.prototype.tile=function(e,i){return ae(e,"tile"),cB(this.bufferSync(e),i)},t.prototype.gather=function(e,i,r){ae([e,i],"gather");var a=e.shape.slice(),s=this.readSync(i.dataId);a[r]=s.length;for(var o=T.buffer(a,e.dtype),l=this.bufferSync(e),u=0;u<o.size;++u){var c=o.indexToLoc(u),h=c.slice();h[r]=s[c[r]];var d=l.locToIndex(h);o.values[u]=l.values[d]}return o.toTensor()},t.prototype.batchToSpaceND=function(e,i,r){ae([e],"batchToSpaceND");var a=i.reduce(function(h,d){return h*d}),s=T.backend_util.getReshaped(e.shape,i,a),o=T.backend_util.getPermuted(s.length,i.length),l=T.backend_util.getReshapedPermuted(e.shape,i,a),u=T.backend_util.getSliceBeginCoords(r,i.length),c=T.backend_util.getSliceSize(l,r,i.length);return T.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.pool3d=function(e,i,r){ae(e,"pool3d");for(var a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=r==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,b=this.readSync(e.dataId),w=T.buffer(i.outShape,e.dtype),S=w.values,L=i.outShape[1]*i.outShape[2]*i.outShape[3]*i.outShape[4],x=i.outShape[2]*i.outShape[3]*i.outShape[4],C=i.outShape[3]*i.outShape[4],R=i.outShape[4],D=0;D<i.batchSize;++D)for(var k=D*L,W=D*e.strides[0],F=0;F<i.inChannels;++F)for(var P=0;P<i.outDepth;++P){for(var H=P*a-f,_=H;_<0;)_+=l;for(var K=Math.min(i.inDepth,h+H),j=k+P*x,q=0;q<i.outHeight;++q){for(var G=q*s-m,Z=G;Z<0;)Z+=u;for(var X=Math.min(i.inHeight,d+G),ee=j+q*C,ne=0;ne<i.outWidth;++ne){for(var ie=ne*o-g,te=ie;te<0;)te+=c;for(var re=Math.min(i.inWidth,p+ie),le=ee+ne*R,he=v,ye=0,Oe=0,Fe=_;Fe<K;Fe+=l){for(var _e=W+Fe*e.strides[1],Me=Z;Me<X;Me+=u){for(var ht=_e+Me*e.strides[2],St=te;St<re;St+=c){var Ge=ht+St*e.strides[3],Ue=b[Ge+F];if(r==="max"&&Ue>he?he=Ue:r==="avg"&&(ye+=Ue,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),x=this.bufferSync(e),C=0;C<r.batchSize;++C)for(var R=0;R<r.inChannels;++R)for(var D=0;D<r.inDepth;++D)for(var k=0;k<r.inHeight;++k)for(var W=0;W<r.inWidth;++W){for(var F=D-v,P=k-w,H=W-b,_=0,K=0;K<f;K+=h){var j=(F+K)/a;if(j<0||j>=r.outDepth||Math.floor(j)!==j)continue;for(var q=0;q<m;q+=d){var G=(P+q)/s;if(G<0||G>=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z<g;Z+=p){var X=(H+Z)/o;if(X<0||X>=r.outWidth||Math.floor(X)!==X)continue;var ee=x.get(C,j,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b<i.batchSize;++b)for(var w=0;w<i.inChannels;++w)for(var S=0;S<i.outDepth;++S){for(var L=S*a-f,x=L;x<0;)x+=l;for(var C=Math.min(i.inDepth,h+L),R=0;R<i.outHeight;++R){for(var D=R*s-m,k=D;k<0;)k+=u;for(var W=Math.min(i.inHeight,d+D),F=0;F<i.outWidth;++F){for(var P=F*o-g,H=P;H<0;)H+=c;for(var _=Math.min(i.inWidth,p+P),K=Number.NEGATIVE_INFINITY,j=-1,q=x;q<C;q+=l)for(var G=q-L,Z=k;Z<W;Z+=u)for(var X=Z-D,ee=H;ee<_;ee+=c){var ne=ee-P,ie=v.get(b,q,Z,ee,w);ie>=K&&(K=ie,j=G*d*p+X*d+ne)}r.set(j,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),x=0;x<a.batchSize;++x)for(var C=0;C<a.inChannels;++C)for(var R=0;R<a.inDepth;++R)for(var D=0;D<a.inHeight;++D)for(var k=0;k<a.inWidth;++k){for(var W=R-g,F=D-b,P=k-v,H=0,_=0;_<p;_+=c){var K=(W+_)/o;if(K<0||K>=a.outDepth||Math.floor(K)!==K)continue;for(var j=0;j<f;j+=h){var q=(F+j)/l;if(q<0||q>=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<m;G+=d){var Z=(P+G)/u;if(Z<0||Z>=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(x,K,q,Z,C),ee=_*f*m+j*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(x,K,q,Z,C);H+=ie*ne}}}w.set(H,x,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b<o;b++)for(var w=0;w<i;w++)for(var S=g*w,L=Math.floor(S),x=S-L,C=Math.min(l-1,Math.ceil(S)),R=b*e.strides[0]+L*e.strides[1],D=b*e.strides[0]+C*e.strides[1],k=0;k<r;k++)for(var W=v*k,F=Math.floor(W),P=W-F,H=Math.min(u-1,Math.ceil(W)),_=R+F*e.strides[2],K=D+F*e.strides[2],j=R+H*e.strides[2],q=D+H*e.strides[2],G=0;G<c;G++){var Z=h[_+G],X=h[K+G],ee=h[j+G],ne=h[q+G],ie=Z+(ee-Z)*P,te=X+(ne-X)*P,re=ie+(te-ie)*x;d[m++]=re}return T.tensor(d,[o,i,r,c])},t.prototype.resizeBilinearBackprop=function(e,i,r){ae([e,i],"resizeBilinearBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=[r&&h>1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S<s;S++)for(var L=S*i.strides[0],x=0;x<h;x++)for(var C=x*g,R=Math.floor(C),D=Math.min(Math.ceil(C),o-1),k=L+R*i.strides[1],W=L+D*i.strides[1],F=C-R,P=1-F,H=0;H<d;H++)for(var _=H*v,K=Math.floor(_),j=Math.min(Math.ceil(_),l-1),q=_-K,G=1-q,Z=k+K*i.strides[2],X=k+j*i.strides[2],ee=W+K*i.strides[2],ne=W+j*i.strides[2],ie=P*G,te=P*q,re=F*G,le=F*q,he=0;he<u;he++){var ye=b[w++];p[Z+he]+=ye*ie,p[X+he]+=ye*te,p[ee+he]+=ye*re,p[ne+he]+=ye*le}return T.tensor4d(p,[s,l,o,u],i.dtype)},t.prototype.resizeNearestNeighbor=function(e,i,r,a){ae(e,"resizeNearestNeighbor");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(o*i*r*c),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b<o;b++)for(var w=b*e.strides[0],S=0;S<i;S++)for(var L=m*S,x=Math.min(l-1,a?Math.round(L):Math.floor(L)),C=w+x*e.strides[1],R=0;R<r;R++)for(var D=g*R,k=Math.min(u-1,a?Math.round(D):Math.floor(D)),W=C+k*e.strides[2],F=0;F<c;F++){var P=h[W+F];d[v++]=P}return T.tensor(d,[o,i,r,c],e.dtype)},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){ae([e,i],"resizeNearestNeighborBackprop");for(var a=i.shape,s=a[0],o=a[1],l=a[2],u=a[3],c=e.shape,h=c[1],d=c[2],p=new Float32Array(s*o*l*u),f=this.readSync(e.dataId),m=[r&&h>1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,x=Math.ceil(S)*2+2,C=0;C<s;C++)for(var R=C*i.strides[0],D=0;D<o;D++)for(var k=R+D*i.strides[1],W=Math.floor(D*w),F=Math.floor(W-L/2),P=0;P<l;P++)for(var H=k+P*i.strides[2],_=Math.floor(P*S),K=Math.floor(_-x/2),j=0;j<u;j++){for(var q=0,G=0;G<L;G++){var Z=G+F;if(Z<0||Z>=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie<x;ie++){var te=ie+K;if(te<0||te>=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+j])}}p[H+j]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p<c;p++){var f=d(p),m=u[p]*Math.pow(r+a*f,-s);h[p]=m}return T.tensor4d(h,e.shape)},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){ae(e,"LRNGrad");for(var u=e.shape[3],c=this.readSync(e.dataId),h=this.readSync(i.dataId),d=this.readSync(r.dataId),p=new Float32Array(e.size),f=e.size,m=0;m<f;m++){for(var g=m%u,v=m-g+Math.max(0,g-a),b=m-g+Math.min(u,g+a+1),w=0,S=v;S<b;S++)w+=Math.pow(h[S],2);w=o*w+s;for(var S=v;S<b;S++){var L=-2*o*l*h[S]*d[m]/w;m===S&&(L+=Math.pow(w,-l)),L*=c[m],p[S]+=L}}return T.tensor4d(p,e.shape)},t.prototype.multinomial=function(e,i,r,a){ae(e,"multinomial");for(var s=i?e:T.softmax(e),o=s.shape[0],l=s.shape[1],u=T.zeros([o,r],"int32"),c=this.readSync(u.dataId),h=this.readSync(s.dataId),d=0;d<o;++d){var p=d*l,f=new Float32Array(l-1);f[0]=h[p];for(var m=1;m<f.length;++m)f[m]=f[m-1]+h[p+m];for(var g=sB.alea(a.toString()),v=d*r,b=0;b<r;++b){var w=g();c[v+b]=f.length;for(var S=0;S<f.length;S++)if(w<f[S]){c[v+b]=S;break}}}return u},t.prototype.oneHot=function(e,i,r,a){ae(e,"oneHot");var s=new Float32Array(e.size*i);s.fill(a);for(var o=this.readSync(e.dataId),l=0;l<e.size;++l)o[l]>=0&&o[l]<i&&(s[l*i+o[l]]=r);return T.tensor2d(s,[e.size,i],"int32")},t.prototype.nonMaxSuppression=function(e,i,r,a,s){ae(e,"nonMaxSuppression");var o=this.readSync(e.dataId),l=this.readSync(i.dataId);return lB(o,l,r,a,s)},t.prototype.depthToSpace=function(e,i,r){T.util.assert(r==="NHWC",function(){return"Only NHWC dataFormat supported on CPU for depthToSpace. Got "+r}),T.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m<a;++m)for(var g=0;g<u;++g)for(var v=Math.floor(g/i),b=g%i,w=0;w<c;++w)for(var S=Math.floor(w/i),L=w%i,x=(b*i+L)*h,C=0;C<h;++C){var R=C+x,D=R+l*(S+o*(v+s*m));p[f++]=d[D]}return T.tensor4d(p,[a,u,c,h])},t.prototype.broadcastedBinaryOp=function(e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(e.shape,i.shape),o=T.buffer(s,r),l=this.readSync(e.dataId),u=this.readSync(i.dataId),c=T.backend_util.getBroadcastDims(e.shape,s),h=T.backend_util.getBroadcastDims(i.shape,s),d=o.values;if(c.length+h.length===0)for(var p=0;p<d.length;++p)d[p]=a(l[p%l.length],u[p%u.length]);else for(var f=this.bufferSync(e),m=this.bufferSync(i),g=function(v){var b=o.indexToLoc(v),w=b.slice(-e.rank);c.forEach(function(C){return w[C]=0});var S=f.locToIndex(w),L=b.slice(-i.rank);h.forEach(function(C){return L[C]=0});var x=m.locToIndex(L);d[v]=a(l[S],u[x])},p=0;p<d.length;++p)g(p);return o.toTensor()},t.prototype.split=function(e,i,r){return uB(e,i,r)},t.prototype.dispose=function(){},t.prototype.floatPrecision=function(){return 32},t.prototype.epsilon=function(){return n.prototype.epsilon.call(this)},t.prototype.cropAndResize=function(e,i,r,a,s,o){for(var l=e.shape,u=l[0],c=l[1],h=l[2],d=l[3],p=i.shape[0],f=a[0],m=a[1],g=T.buffer([p,f,m,d],"float32"),v=this.readSync(i.dataId),b=this.readSync(r.dataId),w=this.readSync(e.dataId),S=e.strides,L=g.strides,x=0;x<p;x++){var C=x*4,R=v[C],D=v[C+1],k=v[C+2],W=v[C+3],F=b[x];if(F>=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_<f;_++){var K=f>1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(K<0||K>c-1){for(var j=0;j<m;j++)for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}if(s==="bilinear")for(var Z=Math.floor(K),X=Math.ceil(K),ee=K-Z,j=0;j<m;j++){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var ie=Math.floor(ne),te=Math.ceil(ne),re=ne-ie,q=0;q<d;q++){var G=q+ie*S[2]+Z*S[1]+F*S[0],le=w[G];G=q+te*S[2]+Z*S[1]+F*S[0];var he=w[G];G=q+ie*S[2]+X*S[1]+F*S[0];var ye=w[G];G=q+te*S[2]+X*S[1]+F*S[0];var Oe=w[G],Fe=le+(he-le)*re,_e=ye+(Oe-ye)*re;G=q+j*L[2]+_*L[1]+x*L[0],g.values[G]=Fe+(_e-Fe)*ee}}else for(var j=0;j<m;++j){var ne=m>1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q<d;q++){var G=q+j*L[2]+_*L[1]+x*L[0];g.values[G]=o}continue}for(var Me=Math.round(ne),ht=Math.round(K),q=0;q<d;q++){var St=q+Me*S[2]+ht*S[1]+F*S[0],Ge=q+j*L[2]+_*L[1]+x*L[0];g.values[Ge]=w[St]}}}}return g.toTensor()},t.prototype.sparseToDense=function(e,i,r,a){var s=T.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.sliceSize,c=s.strides,h=s.outputSize,d=!1;return this.scatter(e,i,r,h,u,l,o,c,a,d)},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=T.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3];if(l===0)return T.tensor([],o,e.dtype);for(var h=new T.TensorBuffer([l,u],e.dtype),d=this.readSync(i.dataId),p=this.readSync(e.dataId),f=0;f<l;f++){for(var m=[],g=0,v=0;v<a;v++){var b=d[f*a+v];g+=b*c[v],m.push(b)}if(g<0||g>=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w<u;w++)h.values[f*u+w]=p[g*u+w]}return h.toTensor().reshape(o)},t.prototype.scatterND=function(e,i,r){var a=T.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=T.scalar(0),d=!0;return this.scatter(e,i,r,c,l,o,s,u,h,d)},t.prototype.fill=function(e,i,r){r=r||T.util.inferDtype(i);var a=T.util.getArrayFromDType(r,T.util.sizeFromShape(e));return a.fill(i),T.engine().makeTensor(a,e,r,this)},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported for string tensors");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){var i=T.util.getArrayFromDType(e.dtype,T.util.sizeFromShape(e.shape));return this.makeOutput(i,e.shape,e.dtype)},t.prototype.linspace=function(e,i,r){return T.backend_util.linspaceImpl(e,i,r)},t.prototype.scatter=function(e,i,r,a,s,o,l,u,c,h){var d=[a/s,s],p=this.readSync(e.dataId),f=this.readSync(i.dataId);if(a===0)return T.tensor([],r,i.dtype);var m=new T.TensorBuffer(d,i.dtype);m.values.fill(this.readSync(c.dataId)[0]);for(var g=0;g<o;g++){for(var v=[],b=0,w=0;w<l;w++){var S=p[g*l+w];v.push(S),b+=S*u[w]}if(b<0||b>=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;L<s;L++)h?m.values[b*s+L]+=f[g*s+L]:m.values[b*s+L]=i.rank===0?f[0]:f[g*s+L]}return m.toTensor().reshape(r)},t}(T.KernelBackend);function Kw(n){for(var t=new Float32Array(n.length),e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}var pB=function(n){var t=n.inputs.x,e=n.backend,i=new Float32Array(T.util.sizeFromShape(t.shape));if(t.dtype!=="complex64"){var r=e.data.get(t.dataId).values;i=Kw(r)}else for(var a=e.data.get(t.dataId),s=a.complexTensorInfos.real,o=a.complexTensorInfos.imag,l=e.data.get(s.dataId).values,u=e.data.get(o.dataId).values,c=0;c<l.length;c++){var h=l[c],d=u[c];i[c]=Math.hypot(h,d)}return e.makeOutput(i,t.shape,"float32")},fB={kernelName:T.Abs,backendName:"cpu",kernelFunc:pB};function lr(n){return function(t,e,i,r,a){var s=T.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=T.util.computeStrides(s),u=T.util.sizeFromShape(s),c=T.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=T.util.computeStrides(t),f=T.util.computeStrides(e),m=T.backend_util.getBroadcastDims(t,s),g=T.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(var v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(var b=function(w){var S=T.util.indexToLoc(w,o,l),L=S.slice(-h);m.forEach(function(D){return L[D]=0});var x=T.util.locToIndex(L,h,p),C=S.slice(-d);g.forEach(function(D){return C[D]=0});var R=T.util.locToIndex(C,d,f);c[w]=n(i[x],r[R])},v=0;v<c.length;++v)b(v);return[c,s]}}function Ln(n){var t=n.inputs,e=n.backend,i=t.real,r=t.imag,a=e.data.get(i.dataId).values,s=e.data.get(r.dataId).values,o=e.makeTensorInfo(i.shape,"complex64"),l=e.data.get(o.dataId);return l.complexTensorInfos={real:e.makeTensorInfo(i.shape,"float32",a),imag:e.makeTensorInfo(r.shape,"float32",s)},o}var mB={kernelName:T.Complex,backendName:"cpu",kernelFunc:Ln};function Xr(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var gB={kernelName:T.Identity,backendName:"cpu",kernelFunc:Xr};function $a(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.real,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var vB={kernelName:T.Real,backendName:"cpu",kernelFunc:$a};function Xa(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.dtype;if(a==="complex64"){if(r.dtype==="complex64")return Xr({inputs:{x:r},backend:e});var s=T.zeros(r.shape),o=Xa({inputs:{x:r},backend:e,attrs:{dtype:"float32"}}),l=Ln({inputs:{real:o,imag:s},backend:e});return s.dispose(),e.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){var u=$a({inputs:{input:r},backend:e}),l=Xa({inputs:{x:u},backend:e,attrs:{dtype:a}});return e.disposeIntermediateTensorInfo(u),l}if(!T.util.hasEncodingLoss(r.dtype,a)){var l=Xr({inputs:{x:r},backend:e});return{dataId:l.dataId,shape:l.shape,dtype:a}}if(a==="int32"){var c=e.data.get(r.dataId).values,h=Int32Array.from(c);return e.makeTensorInfo(r.shape,"int32",h)}if(a==="bool"){var d=e.data.get(r.dataId).values,p=T.util.toTypedArray([0],r.dtype),f=lr(function(v,b){return v!==b?1:0})(r.shape,[],d,p,"bool"),m=f[0],g=f[1];return e.makeTensorInfo(g,"bool",m)}throw new Error("Error in Cast: failed to cast "+r.dtype+" to "+a)}var yB={kernelName:T.Cast,backendName:"cpu",kernelFunc:Xa};function Jr(n,t,e,i){return e==null?function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;ae([l,u],n);var h=c.data.get(l.dataId).values,d=c.data.get(u.dataId).values,p=i||l.dtype,f=t(l.shape,u.shape,h,d,p),m=f[0],g=f[1];return c.makeTensorInfo(g,p,m)}:function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s;if(l.dtype==="complex64"||u.dtype==="complex64"){var h=Xa({inputs:{x:l},backend:c,attrs:{dtype:"complex64"}}),d=c.data.get(h.dataId),p=d.complexTensorInfos.real,f=d.complexTensorInfos.imag,m=c.data.get(p.dataId).values,g=c.data.get(f.dataId).values,v=Xa({inputs:{x:u},backend:c,attrs:{dtype:"complex64"}}),b=c.data.get(v.dataId),w=b.complexTensorInfos.real,S=b.complexTensorInfos.imag,L=c.data.get(w.dataId).values,x=c.data.get(S.dataId).values,C=e(l.shape,u.shape,m,g,L,x),R=C[0],D=C[1],k=C[2],W=c.makeTensorInfo(k,"float32",R),F=c.makeTensorInfo(k,"float32",D),P=Ln({inputs:{real:W,imag:F},backend:c});return c.disposeIntermediateTensorInfo(h),c.disposeIntermediateTensorInfo(v),c.disposeIntermediateTensorInfo(W),c.disposeIntermediateTensorInfo(F),P}else{var H=c.data.get(l.dataId).values,_=c.data.get(u.dataId).values,K=i||l.dtype,j=t(l.shape,u.shape,H,_,K),q=j[0],k=j[1];return c.makeTensorInfo(k,K,q)}}}function Ld(n){return function(t,e,i,r,a,s){var o=T.backend_util.assertAndGetBroadcastShape(t,e),l=T.util.sizeFromShape(o),u=o.length,c=T.util.computeStrides(o),h=T.util.getTypedArrayFromDType("float32",l),d=T.util.getTypedArrayFromDType("float32",l),p=T.backend_util.getBroadcastDims(t,o),f=T.backend_util.getBroadcastDims(e,o),m=T.backend_util.mergeRealAndImagArrays(i,r),g=T.backend_util.mergeRealAndImagArrays(a,s),v=t.length,b=T.util.computeStrides(t),w=e.length,S=T.util.computeStrides(e);if(p.length+f.length===0)for(var L=0;L<h.length;L++){var x=L%m.length,C=L%g.length,R=n(m[x*2],m[x*2+1],g[C*2],g[C*2+1]);h[L]=R.real,d[L]=R.imag}else for(var D=function(k){var W=T.util.indexToLoc(k,u,c),F=W.slice(-v);p.forEach(function(j){return F[j]=0});var P=T.util.locToIndex(F,v,b),H=W.slice(-w);f.forEach(function(j){return H[j]=0});var _=T.util.locToIndex(H,w,S),K=n(m[P*2],m[P*2+1],g[_*2],g[_*2+1]);h[k]=K.real,d[k]=K.imag},L=0;L<h.length;L++)D(L);return[h,d,o]}}var jw=lr(function(n,t){return n+t}),bB=Ld(function(n,t,e,i){return{real:n+e,imag:t+i}}),$w=Jr(T.Add,jw,bB),wB={kernelName:T.Add,backendName:"cpu",kernelFunc:$w};function Zr(n){return function(t,e,i){for(var r=T.util.getTypedArrayFromDType(e,t.length),a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}function $e(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");for(var l=s,u=l.data.get(o.dataId).values,c=T.util.sizeFromShape(o.shape),h=e||o.dtype,d=T.util.getArrayFromDType(h,c),p=0;p<c;++p)d[p]=t(u[p],a);return l.makeTensorInfo(o.shape,h,d)}}function Qr(n,t,e){return function(i){var r=i.inputs,a=i.attrs,s=i.backend,o=r.x;if(ae(o,n),o.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");var l=s,u=l.data.get(o.dataId).values,c=e||o.dtype,h=t(u,c,a);return l.makeTensorInfo(o.shape,c,h)}}var Xw=Zr(function(n){return Math.ceil(n)}),SB=Qr(T.Ceil,Xw),LB={kernelName:T.Ceil,backendName:"cpu",kernelFunc:SB};var Jw=Zr(function(n){return Math.exp(n)}),IB=Qr(T.Exp,Jw),AB={kernelName:T.Exp,backendName:"cpu",kernelFunc:IB};var Zw=Zr(function(n){return Math.expm1(n)}),TB=Qr(T.Expm1,Zw),NB={kernelName:T.Expm1,backendName:"cpu",kernelFunc:TB};var Qw=Zr(function(n){return Math.floor(n)}),xB=Qr(T.Floor,Qw),CB={kernelName:T.Floor,backendName:"cpu",kernelFunc:xB};var e0=Zr(function(n){return Math.log(n)}),RB=Qr(T.Log,e0),OB={kernelName:T.Log,backendName:"cpu",kernelFunc:RB};function t0(n,t,e,i){for(var r=T.util.getTypedArrayFromDType(i,T.util.sizeFromShape(e)),a=0;a<r.length;++a){for(var s=a*t,o=n[s],l=0;l<t;++l){var u=n[s+l];u>o&&(o=u)}r[a]=o}return r}var n0=lr(function(n,t){return n*t}),EB=Ld(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),i0=Jr(T.Multiply,n0,EB),DB={kernelName:T.Multiply,backendName:"cpu",kernelFunc:i0};var r0=Zr(function(n){return 1/Math.sqrt(n)}),kB=Qr(T.Rsqrt,r0),FB={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:kB};function a0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;c<s;++c){var h=e.length,d=T.util.computeStrides(e),p=T.util.indexToLoc(c,h,d),f=p.map(function(g,v){return g+t[v]}),m=T.util.locToIndex(f,i.length,o);u[c]=n[m]}return u}function Id(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.begin,s=i.size;ae(r,"slice");var o=T.slice_util.parseSliceParams(r,a,s),l=o[0],u=o[1];T.slice_util.assertParamsValid(r,l,u);var c=e.data.get(r.dataId).values,h=a0(c,l,u,r.shape,r.dtype);return e.makeTensorInfo(u,r.dtype,h)}var WB={kernelName:T.Slice,backendName:"cpu",kernelFunc:Id};var s0=lr(function(n,t){return n-t}),UB=Ld(function(n,t,e,i){return{real:n-e,imag:t-i}}),o0=Jr(T.Sub,s0,UB),BB={kernelName:T.Sub,backendName:"cpu",kernelFunc:o0};function Ad(n,t,e,i,r){for(var a=t.length,s=T.util.sizeFromShape(t),o=T.util.computeStrides(t),l=T.util.computeStrides(r),u=T.util.getTypedArrayFromDType(e,T.util.sizeFromShape(r)),c=0;c<s;++c){for(var h=T.util.indexToLoc(c,a,o),d=new Array(h.length),p=0;p<d.length;p++)d[p]=h[i[p]];var f=T.util.locToIndex(d,a,l);u[f]=n[c]}return u}function l0(n,t,e,i){for(var r=T.util.parseAxisParam(t,e)[0],a=[1,e[0],1],s=0;s<r;s++)a[0]*=e[s];a[1]=e[r];for(var s=r+1;s<e.length;s++)a[2]*=e[s];for(var o={},l=new Int32Array(e[r]),u=new T.TensorBuffer(a,i,n),c=[],h=a[0]===1&&a[2]===1,s=0;s<e[r];s++){var d=void 0;if(h)d=n[s].toString();else{for(var p=[],f=0;f<a[0];f++)for(var m=0;m<a[2];m++)p.push(u.get(f,s,m));d=p.join(",")}if(o[d]!==void 0)l[s]=o[d];else{var g=Object.keys(o).length;o[d]=g,l[s]=g,c.push(s)}}var v=a.slice();v[1]=Object.keys(o).length;var b=new T.TensorBuffer(v,i);c.forEach(function(S,L){for(var x=0;x<a[0];x++)for(var C=0;C<a[2];C++)b.set(u.get(x,S,C),x,L,C)});var w=e.slice();return w[r]=v[1],{outputValues:b.values,outputShape:w,indices:l}}var zB={__proto__:null,simpleAbsImpl:Kw,addImpl:jw,ceilImpl:Xw,expImpl:Jw,expm1Impl:Zw,floorImpl:Qw,logImpl:e0,maxImpl:t0,multiplyImpl:n0,rsqrtImpl:r0,sliceImpl:a0,subImpl:s0,transposeImpl:Ad,uniqueImpl:l0};var PB="2.6.0";T.registerBackend("cpu",function(){return new Yw},1);var _B=$e(T.Acos,function(n){return Math.acos(n)}),MB={kernelName:T.Acos,backendName:"cpu",kernelFunc:_B};var HB=$e(T.Acosh,function(n){return Math.acosh(n)}),VB={kernelName:T.Acosh,backendName:"cpu",kernelFunc:HB};var qB=$e(T.Asin,function(n){return Math.asin(n)}),GB={kernelName:T.Asin,backendName:"cpu",kernelFunc:qB};var YB=$e(T.Asinh,function(n){return Math.asinh(n)}),KB={kernelName:T.Asinh,backendName:"cpu",kernelFunc:YB};var jB=$e(T.Atan,function(n){return Math.atan(n)}),$B={kernelName:T.Atan,backendName:"cpu",kernelFunc:jB};var XB=$e(T.Atanh,function(n){return Math.atanh(n)}),JB={kernelName:T.Atanh,backendName:"cpu",kernelFunc:XB};function Td(n,t,e,i,r,a){for(var s=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,u=r.dilationWidth,c=r.effectiveFilterHeight,h=r.effectiveFilterWidth,d=r.padInfo.top,p=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=T.buffer(r.outShape,e),g=m.values,v=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],w=r.outShape[3],S=0;S<r.batchSize;++S)for(var L=S*v,x=S*i[0],C=0;C<r.inChannels;++C)for(var R=0;R<r.outHeight;++R)for(var D=R*s-d,k=Math.max(0,D),W=Math.min(r.inHeight,c+D),F=L+R*b,P=0;P<r.outWidth;++P){for(var H=P*o-p,_=Math.max(0,H),K=Math.min(r.inWidth,h+H),j=f,q=0,G=0,Z=k;Z<W;Z+=l){for(var X=x+Z*i[1],ee=_;ee<K;ee+=u){var ne=X+ee*i[2],ie=n[ne+C];a==="max"&&ie>j?j=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(j))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:j}return m}function u0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;g<i.batchSize;++g)for(var v=0;v<i.inChannels;++v)for(var b=0;b<i.outHeight;++b){for(var w=b*o-p,S=w;S<0;)S+=u;for(var L=Math.min(i.inHeight,h+w),x=0;x<i.outWidth;++x){for(var C=x*l-f,R=C;R<0;)R+=c;for(var D=Math.min(i.inWidth,d+C),k=Number.NEGATIVE_INFINITY,W=-1,F=S;F<L;F+=u)for(var P=F-w,H=R;H<D;H+=c){var _=H-C,K=m.get(g,F,H,v);K>k&&(k=K,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,x,v)}}return s}function ZB(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Td(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var QB={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:ZB};function ez(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),x=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D<c.batchSize;++D)for(var k=0;k<c.inChannels;++k)for(var W=0;W<c.inHeight;++W)for(var F=0;F<c.inWidth;++F){for(var P=W-S,H=F-w,_=0,K=0;K<v;K+=m){var j=(P+K)/h;if(j<0||j>=c.outHeight||Math.floor(j)!==j)continue;for(var q=0;q<b;q+=g){var G=(H+q)/d;if(G<0||G>=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,j,G,k);_+=Z}}L.set(_*x,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var tz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:ez};function nz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,x=0,C=0,R=0;R<c.length;++R)m[R]=f[S++]+(c[R]-h[L++])*p[x++]/Math.sqrt(d[C++]+u),S>=g&&(S=0),L>=w&&(L=0),x>=v&&(x=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var iz={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:nz};var rz=$e(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n<e.clipValueMin?e.clipValueMin:n}),az={kernelName:T.ClipByValue,backendName:"cpu",kernelFunc:rz};function Eo(n){var t=n.inputs,e=n.backend,i=t.input,r=e.data.get(i.dataId).complexTensorInfos.imag,a=e.data.get(r.dataId).values;return e.makeTensorInfo(r.shape,r.dtype,a)}var sz={kernelName:T.Imag,backendName:"cpu",kernelFunc:Eo};function Ai(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=T.util.sizeFromShape(r.shape),o=T.util.inferFromImplicitShape(a,s),l=T.util.sizeFromShape(o);T.util.assert(s===l,function(){return"The new shape ("+o+") has "+l+" elements and the old "+("shape ("+r.shape+") has "+s+" elements. The new shape and old ")+"shape must have the same number of elements."}),e.incRef(r.dataId);var u=e.data.get(r.dataId);if(u.complexTensorInfos!=null){var c=u.complexTensorInfos.real,h=u.complexTensorInfos.imag;c.shape=o,h.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var oz={kernelName:T.Reshape,backendName:"cpu",kernelFunc:Ai};function Ja(n){var t=n.inputs,e=n.backend,i=n.attrs,r=i.axis,a=T.util.parseAxisParam(r,t[0].shape)[0],s=T.backend_util.computeOutShape(t.map(function(S){return S.shape}),a);if(T.util.sizeFromShape(s)===0)return e.makeTensorInfo(s,t[0].dtype,[]);var o=t.filter(function(S){return T.util.sizeFromShape(S.shape)>0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return $a({inputs:{input:S},backend:e})}),c=o.map(function(S){return Eo({inputs:{input:S},backend:e})}),h=Ja({inputs:u,backend:e,attrs:{axis:r}}),d=Ja({inputs:c,backend:e,attrs:{axis:r}}),p=Ln({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),x=[-1,L];return Ai({inputs:{x:S},backend:e,attrs:{shape:x}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,x=T.util.sizeFromShape(S.shape);m.set(L,g),g+=x})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,x=0,C=0;C<S.shape[0];++C)for(var R=C*s[1]+v,D=0;D<S.shape[1];++D)m[R+D]=L[x++];v+=S.shape[1]})}var b=T.backend_util.computeOutShape(o.map(function(S){return S.shape}),a),w=e.makeTensorInfo(b,t[0].dtype,m);return f.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),w}var lz={kernelName:T.Concat,backendName:"cpu",kernelFunc:Ja};var uz=$e(T.Cos,function(n){return Math.cos(n)}),cz={kernelName:T.Cos,backendName:"cpu",kernelFunc:uz};var hz=$e(T.Cosh,function(n){return Math.cosh(n)}),dz={kernelName:T.Cosh,backendName:"cpu",kernelFunc:hz};var pz={kernelName:T.Dilation2D,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=i,l=o.strides,u=o.pad,c=o.dilations,h=e,d=h.data.get(a.dataId).values,p=a.shape.length,f=h.data.get(s.dataId).values,m=s.shape.length,g=T.backend_util.computeDilation2DInfo(a.shape,s.shape,l,u,"NHWC",c),v=g.batchSize,b=g.inHeight,w=g.inWidth,S=g.inChannels,L=g.outHeight,x=g.outWidth,C=g.padInfo,R=g.strideHeight,D=g.strideWidth,k=g.filterHeight,W=g.filterWidth,F=g.dilationHeight,P=g.dilationWidth,H=g.outShape,_=T.util.sizeFromShape(H),K=H.length,j=T.util.getArrayFromDType(a.dtype,_),q=0;q<v;++q)for(var G=0;G<L;++G)for(var Z=G*R-C.top,X=0;X<x;++X)for(var ee=X*D-C.left,ne=0;ne<S;++ne){for(var ie=Number.MIN_SAFE_INTEGER,te=0;te<k;++te){var re=Z+te*F;if(re>=0&&re<b)for(var le=0;le<W;++le){var he=ee+le*P;if(he>=0&&he<w){var ye=T.util.locToIndex([q,re,he,ne],p,T.util.computeStrides(a.shape)),Oe=T.util.locToIndex([te,le,ne],m,T.util.computeStrides(s.shape)),Fe=d[ye]+f[Oe];Fe>ie&&(ie=Fe)}}}var _e=T.util.locToIndex([q,G,X,ne],K,T.util.computeStrides(H));j[_e]=ie}var Me=h.write(T.util.toTypedArray(j,a.dtype),H,a.dtype);return{dataId:Me,shape:H,dtype:a.dtype}}};var fz={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=0,ie=0,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var mz={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,x=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),K=0;K<g;++K)for(var j=0;j<S;++j)for(var q=j*C-x.top,G=0;G<L;++G)for(var Z=G*R-x.left,X=0;X<w;++X){for(var ee=Number.MIN_SAFE_INTEGER,ne=q<0?0:q,ie=Z<0?0:Z,te=0;te<D;++te){var re=q+te*W;if(re>=0&&re<v)for(var le=0;le<k;++le){var he=Z+le*F;if(he>=0&&he<b){var ye=p[K][re][he][X]+f[te][le][X];ye>ee&&(ee=ye,ne=re,ie=he)}}}_[K][ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var gz=lr(function(n,t){return n/t}),vz=Jr(T.Div,gz),Nd={kernelName:T.Div,backendName:"cpu",kernelFunc:vz};var yz=$e(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),bz={kernelName:T.Elu,backendName:"cpu",kernelFunc:yz};var wz=T.backend_util.ERF_P,Sz=T.backend_util.ERF_A1,Lz=T.backend_util.ERF_A2,Iz=T.backend_util.ERF_A3,Az=T.backend_util.ERF_A4,Tz=T.backend_util.ERF_A5,Nz=$e(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+wz*e);return t*(1-((((Tz*i+Az)*i+Iz)*i+Lz)*i+Sz)*i*Math.exp(-e*e))}),xz={kernelName:T.Erf,backendName:"cpu",kernelFunc:Nz};function c0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p<r;p++){for(var f=Id({inputs:{x:o},backend:e,attrs:{begin:[p,0],size:[1,a]}}),m=Id({inputs:{x:l},backend:e,attrs:{begin:[p,0],size:[1,a]}}),g=Ln({inputs:{real:f,imag:m},backend:e}),v=Cz(g,t,e),b=v.real,w=v.imag,S=T.backend_util.mergeRealAndImagArrays(b,w),L=0;L<a;L++){var x=T.backend_util.getComplexWithIndex(S,L);h[p*a+L]=x.real,d[p*a+L]=x.imag}e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(g)}var C=e.makeTensorInfo(u,"float32",h),R=e.makeTensorInfo(u,"float32",d),D=Ln({inputs:{real:C,imag:R},backend:e});return e.disposeIntermediateTensorInfo(C),e.disposeIntermediateTensorInfo(R),D}function Cz(n,t,e){var i=T.util.sizeFromShape(n.shape),r=e.data.get(n.dataId),a=e.data.get(r.complexTensorInfos.real.dataId).values,s=e.data.get(r.complexTensorInfos.imag.dataId).values;if(Rz(i)){var o=xd(a,s,i,t,e),l=[n.shape[0],n.shape[1]];if(t){var u=e.makeTensorInfo(l,"float32",o.real),c=e.makeTensorInfo(l,"float32",o.imag),h=e.makeTensorInfo([],"float32",T.util.createScalarValue(i,"float32")),d=Xr({inputs:{x:h},backend:e}),p=Nd.kernelFunc({inputs:{a:u,b:h},backend:e}),f=Nd.kernelFunc({inputs:{a:c,b:d},backend:e}),m=e.data.get(p.dataId).values,g=e.data.get(f.dataId).values;return e.disposeIntermediateTensorInfo(u),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return o}else{var v=T.backend_util.mergeRealAndImagArrays(a,s),b=Oz(v,i,t);return T.backend_util.splitRealAndImagArrays(b)}}function Rz(n){return(n&n-1)===0}function xd(n,t,e,i,r){if(e===1)return{real:n,imag:t};var a=T.backend_util.mergeRealAndImagArrays(n,t),s=e/2,o=T.backend_util.complexWithEvenIndex(a),l=o.real,u=o.imag,c=[l.length],h=r.makeTensorInfo(c,"float32",l),d=r.makeTensorInfo(c,"float32",u),p=Ln({inputs:{real:h,imag:d},backend:r}),f=T.backend_util.complexWithOddIndex(a),m=f.real,g=f.imag,v=[m.length],b=r.makeTensorInfo(v,"float32",m),w=r.makeTensorInfo(v,"float32",g),S=Ln({inputs:{real:b,imag:w},backend:r}),L=xd(l,u,s,i,r),x=L.real,C=L.imag,R=[x.length],D=r.makeTensorInfo(R,"float32",x),k=r.makeTensorInfo(R,"float32",C),W=Ln({inputs:{real:D,imag:k},backend:r}),F=xd(m,g,s,i,r),P=F.real,H=F.imag,_=[P.length],K=r.makeTensorInfo(_,"float32",P),j=r.makeTensorInfo(_,"float32",H),q=Ln({inputs:{real:K,imag:j},backend:r}),G=T.backend_util.exponents(e,i),Z=[G.real.length],X=r.makeTensorInfo(Z,"float32",G.real),ee=r.makeTensorInfo(Z,"float32",G.imag),ne=Ln({inputs:{real:X,imag:ee},backend:r}),ie=i0({inputs:{a:ne,b:q},backend:r}),te=$w({inputs:{a:W,b:ie},backend:r}),re=o0({inputs:{a:W,b:ie},backend:r}),le=$a({inputs:{input:te},backend:r}),he=$a({inputs:{input:re},backend:r}),ye=Eo({inputs:{input:te},backend:r}),Oe=Eo({inputs:{input:re},backend:r}),Fe=Ja({inputs:[le,he],backend:r,attrs:{axis:0}}),_e=Ja({inputs:[ye,Oe],backend:r,attrs:{axis:0}}),Me=r.data.get(Fe.dataId).values,ht=r.data.get(_e.dataId).values;return r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(w),r.disposeIntermediateTensorInfo(S),r.disposeIntermediateTensorInfo(D),r.disposeIntermediateTensorInfo(k),r.disposeIntermediateTensorInfo(W),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(q),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(ie),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(re),r.disposeIntermediateTensorInfo(le),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(he),r.disposeIntermediateTensorInfo(Oe),r.disposeIntermediateTensorInfo(Fe),r.disposeIntermediateTensorInfo(_e),{real:Me,imag:ht}}function Oz(n,t,e){for(var i=new Float32Array(t*2),r=0;r<t;r++){for(var a=0,s=0,o=0;o<t;o++){var l=T.backend_util.exponent(r*o,t,e),u=T.backend_util.getComplexWithIndex(n,o);a+=u.real*l.real-u.imag*l.imag,s+=u.real*l.imag+u.imag*l.real}e&&(a/=t,s/=t),T.backend_util.assignToTypedArray(i,a,s,r)}return i}function Ez(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!1,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Dz={kernelName:T.FFT,backendName:"cpu",kernelFunc:Ez};var kz={kernelName:T.FlipLeftRight,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=i,s=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),o=r.shape,l=o[0],u=o[1],c=o[2],h=o[3],d=a.data.get(r.dataId).values,p=0;p<l;p++)for(var f=p*c*u*h,m=0;m<u;m++)for(var g=m*(c*h),v=0;v<c;v++)for(var b=v*h,w=0;w<h;w++){var S=[l,m,v,w],L=S[2],x=Math.round(c-L),C=f+g+b+w,R=d[C];if(x>=0&&x<c){var D=x*h,k=f+g+D+w;R=d[k]}s[C]=R}var W=a.write(s,r.shape,r.dtype);return{dataId:W,shape:r.shape,dtype:r.dtype}}};function Fz(n){var t=n.inputs,e=n.backend,i=t.input,r=T.util.sizeFromShape(i.shape),a=i.shape[i.shape.length-1],s=r/a,o=Ai({inputs:{x:i},backend:e,attrs:{shape:[s,a]}}),l=c0(o,!0,e),u=Ai({inputs:{x:l},backend:e,attrs:{shape:i.shape}});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(l),u}var Wz={kernelName:T.IFFT,backendName:"cpu",kernelFunc:Fz};var Uz=$e(T.IsFinite,function(n){return Number.isFinite(n)?1:0},"bool"),Bz={kernelName:T.IsFinite,backendName:"cpu",kernelFunc:Uz};var zz=$e(T.IsInf,function(n){return Math.abs(n)===Infinity?1:0},"bool"),Pz={kernelName:T.IsInf,backendName:"cpu",kernelFunc:zz};var _z=$e(T.IsNan,function(n){return Number.isNaN(n)?1:0},"bool"),Mz={kernelName:T.IsNan,backendName:"cpu",kernelFunc:_z};var Hz=$e(T.Log1p,function(n){return Math.log1p(n)}),Vz={kernelName:T.Log1p,backendName:"cpu",kernelFunc:Hz};var qz=$e(T.LogicalNot,function(n){return n?0:1},"bool"),Gz={kernelName:T.LogicalNot,backendName:"cpu",kernelFunc:qz};var Yz={kernelName:T.Max,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape,c=u.length,h=T.util.parseAxisParam(s,u),d=h,p=T.backend_util.getAxesPermutation(d,c),f=l.data.get(r.dataId).values;if(p!=null){for(var m=new Array(c),g=0;g<m.length;g++)m[g]=u[p[g]];f=Ad(f,u,r.dtype,p,m),d=T.backend_util.getInnerMostAxes(d.length,c),u=m}ae(r,"max"),T.backend_util.assertAxesAreInnerMostDims("max",d,c);var v=T.backend_util.computeOutAndReduceShapes(u,d),b=v[0],w=v[1],S=T.util.sizeFromShape(w),L=t0(f,S,b,r.dtype),x=l.write(L,b,r.dtype),C=b;if(o){var m=T.backend_util.expandShapeToKeepDim(b,h);C=m}return{dataId:x,shape:C,dtype:r.dtype}}};function Kz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=Xr({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Td(d,r.shape,r.dtype,p,c,"max");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var jz={kernelName:T.MaxPool,backendName:"cpu",kernelFunc:Kz};function $z(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ae([a,s],"maxPoolBackprop");for(var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=T.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=e.data.get(o.dataId).values,f=T.buffer(d.outShape,o.dtype,u0(p,o.shape,o.dtype,d).values),m=d.strideHeight,g=d.strideWidth,v=d.dilationHeight,b=d.dilationWidth,w=d.effectiveFilterHeight,S=d.effectiveFilterWidth,L=S-1-d.padInfo.left,x=w-1-d.padInfo.top,C=T.buffer(o.shape,"float32"),R=e.data.get(r.dataId).values,D=T.buffer(r.shape,"float32",R),k=0;k<d.batchSize;++k)for(var W=0;W<d.inChannels;++W)for(var F=0;F<d.inHeight;++F)for(var P=0;P<d.inWidth;++P){for(var H=F-x,_=P-L,K=0,j=0;j<w;j+=v){var q=(H+j)/m;if(q<0||q>=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G<S;G+=b){var Z=(_+G)/g;if(Z<0||Z>=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=j*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);K+=ie*ne}}C.set(K,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var Xz={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:$z};function Jz(n,t,e,i,r){var a=T.util.computeStrides(t),s=Td(n,t,e,a,r,"max"),o=u0(n,t,e,r,!0,i);return[s.values,o.values]}var Zz={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=Jz(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var Qz=T.kernel_impls.nonMaxSuppressionV4Impl,eP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=Qz(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var tP=T.kernel_impls.nonMaxSuppressionV5Impl,nP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=tP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var iP=lr(function(n,t){return n!==t?1:0}),rP=Jr(T.NotEqual,iP,null,"bool"),aP={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:rP};function sP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(x,C){return x[0]+r.shape[C]+x[1]}),l=a.map(function(x){return x[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v<c;v++){var b=T.util.indexToLoc(v,h,d),w=b.map(function(x,C){return x+l[C]}),S=T.util.locToIndex(w,f,m);g[S]=u[v]}var L=e.write(g,o,r.dtype);return{dataId:L,shape:o,dtype:r.dtype}}var h0={kernelName:T.PadV2,backendName:"cpu",kernelFunc:sP};var oP=$e(T.Reciprocal,function(n){return 1/n}),lP={kernelName:T.Reciprocal,backendName:"cpu",kernelFunc:oP};var uP={kernelName:T.RotateWithOffset,backendName:"cpu",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=T.util.getTypedArrayFromDType(r.dtype,T.util.sizeFromShape(r.shape)),h=r.shape,d=h[0],p=h[1],f=h[2],m=h[3],g=T.backend_util.getImageCenter(l,p,f),v=g[0],b=g[1],w=255,S=Math.sin(s),L=Math.cos(s),x=u.data.get(r.dataId).values,C=0;C<d;C++)for(var R=C*f*p*m,D=0;D<p;D++)for(var k=D*(f*m),W=0;W<f;W++)for(var F=W*m,P=0;P<m;P++){var H=[d,D,W,P],_=H[2],K=H[1],j=(_-v)*L-(K-b)*S,q=(_-v)*S+(K-b)*L;j=Math.round(j+v),q=Math.round(q+b);var G=o;if(typeof o!="number"&&(P===3?G=w:G=o[P]),j>=0&&j<f&&q>=0&&q<p){var Z=q*(f*m),X=j*m,ee=R+Z+X+P;G=x[ee]}var ne=R+k+F+P;c[ne]=G}var ie=u.write(c,r.shape,r.dtype);return{dataId:ie,shape:r.shape,dtype:r.dtype}}};var cP=$e(T.Round,function(n){var t=Math.floor(n);return n-t<.5?Math.floor(n):n-t>.5?Math.ceil(n):t%2===0?t:t+1}),hP={kernelName:T.Round,backendName:"cpu",kernelFunc:cP};var dP=T.backend_util.SELU_SCALEALPHA,pP=T.backend_util.SELU_SCALE,fP=$e(T.Selu,function(n){return n>=0?pP*n:dP*(Math.exp(n)-1)}),mP={kernelName:T.Selu,backendName:"cpu",kernelFunc:fP};var gP=$e(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),vP={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:gP};var yP=$e(T.Sign,function(n){return n<0?-1:n>0?1:0}),bP={kernelName:T.Sign,backendName:"cpu",kernelFunc:yP};var wP=$e(T.Sin,function(n){return Math.sin(n)}),SP={kernelName:T.Sin,backendName:"cpu",kernelFunc:wP};var LP=$e(T.Sinh,function(n){return Math.sinh(n)}),IP={kernelName:T.Sinh,backendName:"cpu",kernelFunc:LP};var AP=11920928955078125e-23,d0=Math.log(AP)+2,TP=$e(T.Softplus,function(n){var t=n>-d0,e=n<d0,i=Math.exp(n),r;return e?r=i:t?r=n:r=Math.log(1+i),r}),NP={kernelName:T.Softplus,backendName:"cpu",kernelFunc:TP};function p0(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm;ae(r,"transpose");for(var s=r.shape.length,o=new Array(s),l=0;l<o.length;l++)o[l]=r.shape[a[l]];var u=i.data.get(r.dataId).values,c=Ad(u,r.shape,r.dtype,a,o),h=i.write(c,o,r.dtype);return{dataId:h,shape:o,dtype:r.dtype}}var xP={kernelName:T.Transpose,backendName:"cpu",kernelFunc:p0};function CP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.blockShape,s=i.paddings;ae([r],"spaceToBatchND");var o=T.util.sizeFromShape(a),l=[[0,0]];l.push.apply(l,s);for(var u=1+a.length;u<r.shape.length;++u)l.push([0,0]);var c=h0.kernelFunc({inputs:{x:r},backend:e,attrs:{paddings:l,constantValue:0}}),h=T.backend_util.getReshaped(c.shape,a,o,!1),d=T.backend_util.getPermuted(h.length,a.length,!1),p=T.backend_util.getReshapedPermuted(c.shape,a,o,!1),f={x:c},m={shape:h},g=Ai({inputs:f,backend:e,attrs:m}),v={x:g},b={perm:d},w=p0({inputs:v,backend:e,attrs:b}),S={x:w},L={shape:p},x=Ai({inputs:S,backend:e,attrs:L});return e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(w),x}var RP={kernelName:T.SpaceToBatchND,backendName:"cpu",kernelFunc:CP};var OP=$e(T.Sqrt,function(n){return Math.sqrt(n)}),EP={kernelName:T.Sqrt,backendName:"cpu",kernelFunc:OP};var DP={kernelName:T.Square,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.x,r=e;ae(i,"square");for(var a=r.data.get(i.dataId).values,s=new Float32Array(a.length),o=0;o<a.length;++o){var l=a[o];s[o]=l*l}var u=r.write(s,i.shape,i.dtype);return{dataId:u,shape:i.shape,dtype:i.dtype}}};var kP=lr(function(n,t){var e=n-t;return e*e}),FP=Jr(T.SquaredDifference,kP),WP={kernelName:T.SquaredDifference,backendName:"cpu",kernelFunc:FP};var UP=$e(T.Step,function(n,t){var e=t;return isNaN(n)?NaN:n>0?1:e.alpha}),BP={kernelName:T.Step,backendName:"cpu",kernelFunc:UP};var zP=$e(T.Tan,function(n){return Math.tan(n)}),PP={kernelName:T.Tan,backendName:"cpu",kernelFunc:zP};var _P=$e(T.Tanh,function(n){return Math.tanh(n)}),MP={kernelName:T.Tanh,backendName:"cpu",kernelFunc:_P};function HP(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=l0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var VP={kernelName:T.Unique,backendName:"cpu",kernelFunc:HP};var qP=[fB,MB,VB,wB,GB,KB,$B,JB,QB,tz,iz,yB,LB,az,mB,lz,cz,dz,pz,mz,fz,Nd,bz,xz,AB,NB,Dz,kz,CB,gB,Wz,sz,Bz,Pz,Mz,OB,Vz,Gz,jz,Xz,Zz,Yz,DB,eP,nP,aP,h0,vB,lP,oz,uP,hP,FB,mP,vP,bP,SP,IP,WB,NP,RP,EP,DP,WP,BP,BB,PP,MP,xP,VP];for(var Cd=0,f0=qP;Cd<f0.length;Cd++){var GP=f0[Cd];T.registerKernel(GP)}ja.MathBackendCPU=Yw;ja.shared=zB;ja.version_cpu=PB});var JS=be(Pn=>{"use strict";Object.defineProperty(Pn,"__esModule",{value:!0});var N=Xi();var Rd=function(n,t){return Rd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Rd(n,t)};function YP(n,t){Rd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Do(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function ko(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]<a[3])){e.label=u[1];break}if(u[0]===6&&e.label<a[1]){e.label=a[1],a=u;break}if(a&&e.label<a[2]){e.label=a[2],e.ops.push(u);break}a[2]&&e.ops.pop(),e.trys.pop();continue}u=t.call(n,e)}catch(c){u=[6,c],r=0}finally{i=a=0}if(u[0]&5)throw u[1];return{value:u[0]?u[1]:void 0,done:!0}}}var ur={},Od={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function g0(n,t){ur[n]=t}function _n(n){if(!(n in ur)){var t=KP(n);if(t!==null)ur[n]=t;else return console.log("Could not get context for WebGL version",n),null}var e=ur[n];return e.isContextLost()?(delete ur[n],_n(n)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),ur[n])}function jP(n){if(typeof OffscreenCanvas!="undefined"&&n===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function KP(n){if(n!==1&&n!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");var t=jP(n);return t.addEventListener("webglcontextlost",function(e){e.preventDefault(),delete ur[n]},!1),n===1?t.getContext("webgl",Od)||t.getContext("experimental-webgl",Od):t.getContext("webgl2",Od)}var Za;(function(n){n[n.DENSE=0]="DENSE",n[n.SHARED_BATCH=1]="SHARED_BATCH"})(Za||(Za={}));var nn;(function(n){n[n.RENDER=0]="RENDER",n[n.UPLOAD=1]="UPLOAD",n[n.PIXELS=2]="PIXELS",n[n.DOWNLOAD=3]="DOWNLOAD"})(nn||(nn={}));var Ct;(function(n){n[n.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",n[n.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",n[n.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",n[n.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",n[n.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Ct||(Ct={}));function Qa(n,t){return[t,n]}function $P(n,t){return n*t}function es(n){var t=N.util.sizeFromShape(n),e=Math.ceil(t/4);return N.util.sizeToSquarishShape(e)}function ea(n,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(n/2))]}function XP(n,t){var e=ea(n,t),i=e[0],r=e[1];return i*r*4}function Ed(n,t){var e=n,i,r,a,s,o,l,u,c,h,d;return N.env().getNumber("WEBGL_VERSION")===2?(i=e.R32F,r=e.R16F,a=e.RGBA16F,s=e.RGBA32F,o=e.RED,u=4,c=1,h=e.HALF_FLOAT,d=e.FLOAT):(i=n.RGBA,r=n.RGBA,a=n.RGBA,s=e.RGBA,o=n.RGBA,u=4,c=4,h=t!=null?t.HALF_FLOAT_OES:null,d=n.FLOAT),l=n.RGBA,{internalFormatFloat:i,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:s,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:u,defaultNumChannels:c,textureTypeHalfFloat:h,textureTypeFloat:d}}function ce(n,t){var e=t();return N.env().getBool("DEBUG")&&JP(n),e}function JP(n){var t=n.getError();if(t!==n.NO_ERROR)throw new Error("WebGL Error: "+v0(n,t))}var ZP=596e-10,QP=65504;function y0(n){return!!(N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||n===0||ZP<Math.abs(n)&&Math.abs(n)<QP)}function v0(n,t){switch(t){case n.NO_ERROR:return"NO_ERROR";case n.INVALID_ENUM:return"INVALID_ENUM";case n.INVALID_VALUE:return"INVALID_VALUE";case n.INVALID_OPERATION:return"INVALID_OPERATION";case n.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case n.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case n.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return"Unknown error code "+t}}function ts(n,t){return ei(n,function(){return n.getExtension(t)},'Extension "'+t+'" not supported on this browser.')}function b0(n,t){var e=ei(n,function(){return n.createShader(n.VERTEX_SHADER)},"Unable to create vertex WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw console.log(n.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function w0(n,t){var e=ei(n,function(){return n.createShader(n.FRAGMENT_SHADER)},"Unable to create fragment WebGLShader.");if(ce(n,function(){return n.shaderSource(e,t)}),ce(n,function(){return n.compileShader(e)}),n.getShaderParameter(e,n.COMPILE_STATUS)===!1)throw e_(t,n.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var t_=/ERROR: [0-9]+:([0-9]+):/g;function e_(n,t){var e=t_.exec(t);if(e==null){console.log("Couldn't parse line number in error: "+t),console.log(n);return}for(var i=+e[1],r=n.split(`
 `),a=r.length.toString().length+2,s=r.map(function(d,p){return N.util.rightPad((p+1).toString(),a)+d}),o=0,l=0;l<s.length;l++)o=Math.max(s[l].length,o);var u=s.slice(0,i-1),c=s.slice(i-1,i),h=s.slice(i);console.log(u.join(`
 `)),console.log(t.split(`
 `)[0]),console.log("%c "+N.util.rightPad(c[0],o),"border:1px solid red; background-color:#e3d2d2; color:#a61717"),console.log(h.join(`
-`))}function S0(n){return ei(n,function(){return n.createProgram()},"Unable to create WebGLProgram.")}function L0(n,t){if(ce(n,function(){return n.linkProgram(t)}),n.getProgramParameter(t,n.LINK_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function Fo(n,t){if(ce(n,function(){return n.validateProgram(t)}),n.getProgramParameter(t,n.VALIDATE_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function I0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function A0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ELEMENT_ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ELEMENT_ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function n_(){return N.env().getNumber("WEBGL_VERSION")===2?1:4}function T0(n){return ei(n,function(){return n.createTexture()},"Unable to create WebGLTexture.")}function N0(n,t){var e=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(n<=0||t<=0){var i="["+n+"x"+t+"]";throw new Error("Requested texture size "+i+" is invalid.")}if(n>e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function x0(n){return ei(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function Ed(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function R0(n,t,e){C0(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function i_(n,t){C0(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function O0(n,t,e){return ei(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function E0(n,t,e){return n.getUniformLocation(t,e)}function D0(n,t,e,i){ce(n,function(){return R0(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function r_(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function Wo(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function Dd(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function ns(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+k0(n,t))}function k0(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function ei(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function C0(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(i<n.TEXTURE0||i>e){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function cr(n,t){return t===void 0&&(t=2),N.util.sizeFromShape(n.slice(0,n.length-t))}function hr(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function Uo(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[cr(n)].concat(hr(n))),t}function F0(n,t){var e;t===void 0&&(t=!1);var i=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?N.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=N.util.squeezeShape(n);n=r.newShape}var a=N.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=cr(n),o=2,l=2;return n.length&&(e=hr(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),N.util.sizeToSquarishShape(a).map(function(u){return u*2})}return N.util.sizeToSquarishShape(a)}function Bo(n){return n%2===0}function is(n,t){if(n=n.slice(-2),t=t.slice(-2),N.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(Bo(e)&&Bo(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&Bo(n[0])&&Bo(t[0])}var zo,Po;function W0(n){if(zo==null){var t=_n(n);zo=t.getParameter(t.MAX_TEXTURE_SIZE)}return zo}function a_(){zo=null}function s_(){Po=null}function U0(n){if(Po==null){var t=_n(n);Po=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,Po)}function B0(n){if(n===0)return 0;var t,e=_n(n);return an(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:an(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function an(n,t){var e=n.getExtension(t);return e!=null}function kd(n){try{var t=_n(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function z0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!an(t,"OES_texture_float"))return!1}else if(!an(t,"EXT_color_buffer_float"))return!1;var e=Fd(t);return e}function P0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!an(t,"OES_texture_float"))return!1;if(!an(t,"WEBGL_color_buffer_float"))return!1}else{if(an(t,"EXT_color_buffer_float"))return Fd(t);var e="EXT_color_buffer_half_float";if(an(t,e)){var i=t.getExtension(e);return o_(t,i)}return!1}var r=Fd(t);return r}function Fd(n){var t=Od(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function o_(n,t){var e=Od(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function _0(n){if(n!==2)return!1;var t=_n(n),e=t.fenceSync!=null;return e}function ta(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&N.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var l_={__proto__:null,callAndCheck:ce,canBeRepresented:y0,getWebGLErrorMessage:v0,getExtensionOrThrow:ts,createVertexShader:b0,createFragmentShader:w0,createProgram:S0,linkProgram:L0,validateProgram:Fo,createStaticVertexBuffer:I0,createStaticIndexBuffer:A0,getNumChannels:n_,createTexture:T0,validateTextureSize:N0,createFramebuffer:x0,bindVertexBufferToProgramAttribute:Ed,bindTextureUnit:R0,unbindTextureUnit:i_,getProgramUniformLocationOrThrow:O0,getProgramUniformLocation:E0,bindTextureToProgramUniformSampler:D0,bindCanvasToFramebuffer:r_,bindColorTextureToFramebuffer:Wo,unbindColorTextureFromFramebuffer:Dd,validateFramebuffer:ns,getFramebufferErrorMessage:k0,getBatchDim:cr,getRowsCols:hr,getShapeAs3D:Uo,getTextureShapeFromLogicalShape:F0,isReshapeFree:is,getWebGLMaxTextureSize:W0,resetMaxTextureSize:a_,resetMaxTexturesInShader:s_,getMaxTexturesInShader:U0,getWebGLDisjointQueryTimerVersion:B0,hasExtension:an,isWebGLVersionEnabled:kd,isCapableOfRenderingToFloatTexture:z0,isDownloadFloatTextureEnabled:P0,isWebGLFenceEnabled:_0,assertNotComplex:ta};var Se=N.env();Se.registerFlag("HAS_WEBGL",function(){return Se.getNumber("WEBGL_VERSION")>0});Se.registerFlag("WEBGL_VERSION",function(){return kd(2)?2:kd(1)?1:0});Se.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});Se.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return Se.get("WEBGL_VERSION")===2});Se.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});Se.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});Se.registerFlag("WEBGL_PACK",function(){return Se.getBool("HAS_WEBGL")});Se.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_CLIP",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});Se.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_REDUCE",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_LAZILY_UNPACK",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_CONV_IM2COL",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return W0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return U0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=Se.getNumber("WEBGL_VERSION");return n===0?0:B0(n)});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return Se.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!N.device_util.isMobile()});Se.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return z0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return Se.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:Se.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});Se.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return P0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return _0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=Se.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});Se.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function u_(n){const t=new Float32Array(n.length);for(let e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}function Wd(n){return(t,e,i,r,a)=>{const s=N.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=N.util.computeStrides(s),u=N.util.sizeFromShape(s),c=N.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=N.util.computeStrides(t),f=N.util.computeStrides(e),m=N.backend_util.getBroadcastDims(t,s),g=N.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(let v=0;v<c.length;++v){const b=N.util.indexToLoc(v,o,l),w=b.slice(-h);m.forEach(C=>w[C]=0);const S=N.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const x=N.util.locToIndex(L,d,f);c[v]=n(i[S],r[x])}return[c,s]}}const c_=Wd((n,t)=>n+t);function na(n){return(t,e,i)=>{const r=N.util.getTypedArrayFromDType(e,t.length);for(let a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}const h_=na(n=>Math.ceil(n));const d_=na(n=>Math.exp(n));const p_=na(n=>Math.expm1(n));const f_=na(n=>Math.floor(n));const m_=na(n=>Math.log(n));function g_(n,t,e,i){const r=N.util.getTypedArrayFromDType(i,N.util.sizeFromShape(e));for(let a=0;a<r.length;++a){const s=a*t;let o=n[s];for(let l=0;l<t;++l){const u=n[s+l];u>o&&(o=u)}r[a]=o}return r}const v_=Wd((n,t)=>n*t);const y_=na(n=>1/Math.sqrt(n));function b_(n,t,e,i,r){const a=N.slice_util.isSliceContinous(i,t,e),s=N.util.sizeFromShape(e),o=N.util.computeStrides(i);if(a){const u=N.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=N.util.getTypedArrayFromDType(r,s);for(let u=0;u<s;++u){const c=e.length,h=N.util.computeStrides(e),d=N.util.indexToLoc(u,c,h),p=d.map((m,g)=>m+t[g]),f=N.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const w_=Wd((n,t)=>n-t);function S_(n,t,e,i,r){const a=t.length,s=N.util.sizeFromShape(t),o=N.util.computeStrides(t),l=N.util.computeStrides(r),u=N.util.getTypedArrayFromDType(e,N.util.sizeFromShape(r));for(let c=0;c<s;++c){const h=N.util.indexToLoc(c,a,o),d=new Array(h.length);for(let f=0;f<d.length;f++)d[f]=h[i[f]];const p=N.util.locToIndex(d,a,l);u[p]=n[c]}return u}function L_(n,t,e,i){const r=N.util.parseAxisParam(t,e)[0],a=[1,e[0],1];for(let f=0;f<r;f++)a[0]*=e[f];a[1]=e[r];for(let f=r+1;f<e.length;f++)a[2]*=e[f];const s={},o=new Int32Array(e[r]),l=new N.TensorBuffer(a,i,n),u=[],c=a[0]===1&&a[2]===1;for(let f=0;f<e[r];f++){let m;if(c)m=n[f].toString();else{const g=[];for(let v=0;v<a[0];v++)for(let b=0;b<a[2];b++)g.push(l.get(v,f,b));m=g.join(",")}if(s[m]!==void 0)o[f]=s[m];else{const g=Object.keys(s).length;s[m]=g,o[f]=g,u.push(f)}}const h=a.slice();h[1]=Object.keys(s).length;const d=new N.TensorBuffer(h,i);u.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let v=0;v<a[2];v++)d.set(l.get(g,f,v),g,m,v)});const p=e.slice();return p[r]=h[1],{outputValues:d.values,outputShape:p,indices:o}}var I_=u_,A_=c_,T_=h_,N_=d_,x_=p_,C_=f_,R_=m_,O_=g_,E_=v_,D_=y_,k_=b_,F_=w_,M0=S_,W_=L_;var U_=function(){function n(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map(function(a,s){return"T"+s});var i=[];this.variableNames.forEach(function(a){i.push("float v"+a+" = get"+a+"AtOutCoords();")});var r=this.variableNames.map(function(a){return"v"+a}).join(" + ");this.userCode=`
+`))}function S0(n){return ei(n,function(){return n.createProgram()},"Unable to create WebGLProgram.")}function L0(n,t){if(ce(n,function(){return n.linkProgram(t)}),n.getProgramParameter(t,n.LINK_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function Fo(n,t){if(ce(n,function(){return n.validateProgram(t)}),n.getProgramParameter(t,n.VALIDATE_STATUS)===!1)throw console.log(n.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function I0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function A0(n,t){var e=ei(n,function(){return n.createBuffer()},"Unable to create WebGLBuffer");return ce(n,function(){return n.bindBuffer(n.ELEMENT_ARRAY_BUFFER,e)}),ce(n,function(){return n.bufferData(n.ELEMENT_ARRAY_BUFFER,t,n.STATIC_DRAW)}),e}function n_(){return N.env().getNumber("WEBGL_VERSION")===2?1:4}function T0(n){return ei(n,function(){return n.createTexture()},"Unable to create WebGLTexture.")}function N0(n,t){var e=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(n<=0||t<=0){var i="["+n+"x"+t+"]";throw new Error("Requested texture size "+i+" is invalid.")}if(n>e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function x0(n){return ei(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function Dd(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function R0(n,t,e){C0(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function i_(n,t){C0(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function O0(n,t,e){return ei(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function E0(n,t,e){return n.getUniformLocation(t,e)}function D0(n,t,e,i){ce(n,function(){return R0(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function r_(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function Wo(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function kd(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function ns(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+k0(n,t))}function k0(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function ei(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function C0(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(i<n.TEXTURE0||i>e){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function cr(n,t){return t===void 0&&(t=2),N.util.sizeFromShape(n.slice(0,n.length-t))}function hr(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function Uo(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[cr(n)].concat(hr(n))),t}function F0(n,t){var e;t===void 0&&(t=!1);var i=N.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?N.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=N.util.squeezeShape(n);n=r.newShape}var a=N.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=cr(n),o=2,l=2;return n.length&&(e=hr(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),N.util.sizeToSquarishShape(a).map(function(u){return u*2})}return N.util.sizeToSquarishShape(a)}function Bo(n){return n%2===0}function is(n,t){if(n=n.slice(-2),t=t.slice(-2),N.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(Bo(e)&&Bo(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&Bo(n[0])&&Bo(t[0])}var zo,Po;function W0(n){if(zo==null){var t=_n(n);zo=t.getParameter(t.MAX_TEXTURE_SIZE)}return zo}function a_(){zo=null}function s_(){Po=null}function U0(n){if(Po==null){var t=_n(n);Po=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,Po)}function B0(n){if(n===0)return 0;var t,e=_n(n);return rn(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:rn(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function rn(n,t){var e=n.getExtension(t);return e!=null}function Fd(n){try{var t=_n(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function z0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!rn(t,"OES_texture_float"))return!1}else if(!rn(t,"EXT_color_buffer_float"))return!1;var e=Wd(t);return e}function P0(n){if(n===0)return!1;var t=_n(n);if(n===1){if(!rn(t,"OES_texture_float"))return!1;if(!rn(t,"WEBGL_color_buffer_float"))return!1}else{if(rn(t,"EXT_color_buffer_float"))return Wd(t);var e="EXT_color_buffer_half_float";if(rn(t,e)){var i=t.getExtension(e);return o_(t,i)}return!1}var r=Wd(t);return r}function Wd(n){var t=Ed(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function o_(n,t){var e=Ed(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function _0(n){if(n!==2)return!1;var t=_n(n),e=t.fenceSync!=null;return e}function ta(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&N.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var l_={__proto__:null,callAndCheck:ce,canBeRepresented:y0,getWebGLErrorMessage:v0,getExtensionOrThrow:ts,createVertexShader:b0,createFragmentShader:w0,createProgram:S0,linkProgram:L0,validateProgram:Fo,createStaticVertexBuffer:I0,createStaticIndexBuffer:A0,getNumChannels:n_,createTexture:T0,validateTextureSize:N0,createFramebuffer:x0,bindVertexBufferToProgramAttribute:Dd,bindTextureUnit:R0,unbindTextureUnit:i_,getProgramUniformLocationOrThrow:O0,getProgramUniformLocation:E0,bindTextureToProgramUniformSampler:D0,bindCanvasToFramebuffer:r_,bindColorTextureToFramebuffer:Wo,unbindColorTextureFromFramebuffer:kd,validateFramebuffer:ns,getFramebufferErrorMessage:k0,getBatchDim:cr,getRowsCols:hr,getShapeAs3D:Uo,getTextureShapeFromLogicalShape:F0,isReshapeFree:is,getWebGLMaxTextureSize:W0,resetMaxTextureSize:a_,resetMaxTexturesInShader:s_,getMaxTexturesInShader:U0,getWebGLDisjointQueryTimerVersion:B0,hasExtension:rn,isWebGLVersionEnabled:Fd,isCapableOfRenderingToFloatTexture:z0,isDownloadFloatTextureEnabled:P0,isWebGLFenceEnabled:_0,assertNotComplex:ta};var Se=N.env();Se.registerFlag("HAS_WEBGL",function(){return Se.getNumber("WEBGL_VERSION")>0});Se.registerFlag("WEBGL_VERSION",function(){return Fd(2)?2:Fd(1)?1:0});Se.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});Se.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return Se.get("WEBGL_VERSION")===2});Se.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});Se.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});Se.registerFlag("WEBGL_PACK",function(){return Se.getBool("HAS_WEBGL")});Se.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_CLIP",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});Se.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_PACK_REDUCE",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_LAZILY_UNPACK",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_CONV_IM2COL",function(){return Se.getBool("WEBGL_PACK")});Se.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return W0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return U0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=Se.getNumber("WEBGL_VERSION");return n===0?0:B0(n)});Se.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return Se.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!N.device_util.isMobile()});Se.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return z0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return Se.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:Se.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});Se.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return P0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return _0(Se.getNumber("WEBGL_VERSION"))});Se.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=Se.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});Se.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function u_(n){const t=new Float32Array(n.length);for(let e=0;e<n.length;++e)t[e]=Math.abs(n[e]);return t}function Ud(n){return(t,e,i,r,a)=>{const s=N.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=N.util.computeStrides(s),u=N.util.sizeFromShape(s),c=N.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=N.util.computeStrides(t),f=N.util.computeStrides(e),m=N.backend_util.getBroadcastDims(t,s),g=N.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;v<c.length;++v)c[v]=n(i[v%i.length],r[v%r.length]);else for(let v=0;v<c.length;++v){const b=N.util.indexToLoc(v,o,l),w=b.slice(-h);m.forEach(C=>w[C]=0);const S=N.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const x=N.util.locToIndex(L,d,f);c[v]=n(i[S],r[x])}return[c,s]}}const c_=Ud((n,t)=>n+t);function na(n){return(t,e,i)=>{const r=N.util.getTypedArrayFromDType(e,t.length);for(let a=0;a<t.length;++a)r[a]=n(t[a],i);return r}}const h_=na(n=>Math.ceil(n));const d_=na(n=>Math.exp(n));const p_=na(n=>Math.expm1(n));const f_=na(n=>Math.floor(n));const m_=na(n=>Math.log(n));function g_(n,t,e,i){const r=N.util.getTypedArrayFromDType(i,N.util.sizeFromShape(e));for(let a=0;a<r.length;++a){const s=a*t;let o=n[s];for(let l=0;l<t;++l){const u=n[s+l];u>o&&(o=u)}r[a]=o}return r}const v_=Ud((n,t)=>n*t);const y_=na(n=>1/Math.sqrt(n));function b_(n,t,e,i,r){const a=N.slice_util.isSliceContinous(i,t,e),s=N.util.sizeFromShape(e),o=N.util.computeStrides(i);if(a){const u=N.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=N.util.getTypedArrayFromDType(r,s);for(let u=0;u<s;++u){const c=e.length,h=N.util.computeStrides(e),d=N.util.indexToLoc(u,c,h),p=d.map((m,g)=>m+t[g]),f=N.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const w_=Ud((n,t)=>n-t);function S_(n,t,e,i,r){const a=t.length,s=N.util.sizeFromShape(t),o=N.util.computeStrides(t),l=N.util.computeStrides(r),u=N.util.getTypedArrayFromDType(e,N.util.sizeFromShape(r));for(let c=0;c<s;++c){const h=N.util.indexToLoc(c,a,o),d=new Array(h.length);for(let f=0;f<d.length;f++)d[f]=h[i[f]];const p=N.util.locToIndex(d,a,l);u[p]=n[c]}return u}function L_(n,t,e,i){const r=N.util.parseAxisParam(t,e)[0],a=[1,e[0],1];for(let f=0;f<r;f++)a[0]*=e[f];a[1]=e[r];for(let f=r+1;f<e.length;f++)a[2]*=e[f];const s={},o=new Int32Array(e[r]),l=new N.TensorBuffer(a,i,n),u=[],c=a[0]===1&&a[2]===1;for(let f=0;f<e[r];f++){let m;if(c)m=n[f].toString();else{const g=[];for(let v=0;v<a[0];v++)for(let b=0;b<a[2];b++)g.push(l.get(v,f,b));m=g.join(",")}if(s[m]!==void 0)o[f]=s[m];else{const g=Object.keys(s).length;s[m]=g,o[f]=g,u.push(f)}}const h=a.slice();h[1]=Object.keys(s).length;const d=new N.TensorBuffer(h,i);u.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let v=0;v<a[2];v++)d.set(l.get(g,f,v),g,m,v)});const p=e.slice();return p[r]=h[1],{outputValues:d.values,outputShape:p,indices:o}}var I_=u_,A_=c_,T_=h_,N_=d_,x_=p_,C_=f_,R_=m_,O_=g_,E_=v_,D_=y_,k_=b_,F_=w_,M0=S_,W_=L_;var U_=function(){function n(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map(function(a,s){return"T"+s});var i=[];this.variableNames.forEach(function(a){i.push("float v"+a+" = get"+a+"AtOutCoords();")});var r=this.variableNames.map(function(a){return"v"+a}).join(" + ");this.userCode=`
       void main() {
         `+i.join(`
         `)+`
@@ -69,7 +69,7 @@ Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed thi
         }
         setOutput(float(bestIndex));
       }
-    `}return n}();function H0(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function Jt(n,t){return t===1?[n]:H0(n,t)}function P_(n,t){if(n===1)return"rc";for(var e="",i=0;i<n;i++)e+=t[i],i<n-1&&(e+=",");return e}function Ft(){var n,t,e,i,r,a,s,o,l,u;return N.env().getNumber("WEBGL_VERSION")===2?(n="#version 300 es",t="in",e="out",i="in",r="texture",a="outputColor",s="out vec4 outputColor;",o=`
+    `}return n}();function H0(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function Xt(n,t){return t===1?[n]:H0(n,t)}function P_(n,t){if(n===1)return"rc";for(var e="",i=0;i<n;i++)e+=t[i],i<n-1&&(e+=",");return e}function Ft(){var n,t,e,i,r,a,s,o,l,u;return N.env().getNumber("WEBGL_VERSION")===2?(n="#version 300 es",t="in",e="out",i="in",r="texture",a="outputColor",s="out vec4 outputColor;",o=`
       bool isnan_custom(float val) {
         return (val > 0.0 || val < 0.0) ? false : val != 0.0;
       }
@@ -114,7 +114,7 @@ Hi there 👋. Looks like you are running TensorFlow.js in Node.js. To speed thi
       ivec4 round(vec4 value) {
         return ivec4(floor(value + vec4(0.5)));
       }
-    `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function dr(n,t,e){e===void 0&&(e="index");var i=N.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function Ud(n){var t=N.util.computeStrides(n).map(function(e){return e.toString()});return`
+    `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function dr(n,t,e){e===void 0&&(e="index");var i=N.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function Bd(n){var t=N.util.computeStrides(n).map(function(e){return e.toString()});return`
   int getFlatIndex(ivec3 coords) {
     return coords.x * `+t[0]+" + coords.y * "+t[1]+` + coords.z;
   }
@@ -715,7 +715,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
       `+d+`
       return get`+i+"("+f+`);
     }
-  `}function Ze(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function aa(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function sa(n,t){return t.map(function(e){return n[e]}).join(", ")}var LM=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,N.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Ze(l),c=Jt("coords",l),h,d;if(s===1){d=l+1;var p=Ze(d);h=`
+  `}function Ze(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function aa(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function sa(n,t){return t.map(function(e){return n[e]}).join(", ")}var LM=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,N.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Ze(l),c=Xt("coords",l),h,d;if(s===1){d=l+1;var p=Ze(d);h=`
         `+p+" sourceLocR = "+p+"("+c.join()+`, 0);
         ++`+c[l-1]+`;
         `+p+" sourceLocG = "+p+"("+c.join()+`, 0);
@@ -731,7 +731,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
         `+u+` sourceLocA = coords;
         --`+c[l-1]+`;
         `+u+` sourceLocB = coords;
-        --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=Jt("sourceLocR",d-1).concat("inIdx.r"),b=Jt("sourceLocG",d-1).concat("inIdx.g"),w=Jt("sourceLocB",d-1).concat("inIdx.b"),S=Jt("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",x=r?"":`
+        --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=Xt("sourceLocR",d-1).concat("inIdx.r"),b=Xt("sourceLocG",d-1).concat("inIdx.g"),w=Xt("sourceLocB",d-1).concat("inIdx.b"),S=Xt("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",x=r?"":`
           inIdx = round(vec4(getBestIndicesAChannel(`+v.join()+`),
                              getBestIndicesAChannel(`+b.join()+`),
                              getBestIndicesAChannel(`+w.join()+`),
@@ -891,7 +891,7 @@ vec2 packedUVfrom3D(int texNumR, int texNumC,
     `}return n}();var $0=`
   if (isnan(a)) return a;
   if (isnan(b)) return b;
-`,Bd="return a + b;",zd="return a - b;",X0="return a * b;",TM=`
+`,zd="return a + b;",Pd="return a - b;",X0="return a * b;",TM=`
   float s = sign(a) * sign(b);
   int ia = round(a);
   int ib = round(b);
@@ -1019,7 +1019,7 @@ return (round(mod(b, 2.0)) != 1) ?
             result.y = (coords + 1) >= `+this.outputShape[0]+` ? 0. : result.y;
             result.z = 0.;
             result.w = 0.;
-          `;else{var l=Jt("coords",a);s+=`
+          `;else{var l=Xt("coords",a);s+=`
             bool nextRowOutOfBounds =
               (`+l[a-2]+" + 1) >= "+this.outputShape[a-2]+`;
             bool nextColOutOfBounds =
@@ -1090,7 +1090,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+r.join(`
         `)+`
       }
-    `}return n}();var i5=function(){function n(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=N.backend_util.computeOutShape(t,e);var i=this.outputShape,r=i.length,a=Ze(r),s=Jt("coords",r),o=["x","y","z","w","u","v"].slice(0,r);this.variableNames=t.map(function(v,b){return"T"+b});var l=new Array(t.length-1);l[0]=t[0][e];for(var u=1;u<l.length;u++)l[u]=l[u-1]+t[u][e];for(var c=o[e],h=o.slice(-2),d=o.join(),p="if ("+c+" < "+l[0]+`) {
+    `}return n}();var i5=function(){function n(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=N.backend_util.computeOutShape(t,e);var i=this.outputShape,r=i.length,a=Ze(r),s=Xt("coords",r),o=["x","y","z","w","u","v"].slice(0,r);this.variableNames=t.map(function(v,b){return"T"+b});var l=new Array(t.length-1);l[0]=t[0][e];for(var u=1;u<l.length;u++)l[u]=l[u-1]+t[u][e];for(var c=o[e],h=o.slice(-2),d=o.join(),p="if ("+c+" < "+l[0]+`) {
         return getChannel(
             getT0(`+d+"), vec2("+h.join()+`));
         }`,u=1;u<l.length;u++){var f=l[u-1];p+=`
@@ -1965,14 +1965,14 @@ return (round(mod(b, 2.0)) != 1) ?
           float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;
           setOutput(val);
       }
-    `}return n}();var g5=function(){function n(t){this.variableNames=["A"],this.outTexUsage=rn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
+    `}return n}();var g5=function(){function n(t){this.variableNames=["A"],this.outTexUsage=nn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
       `+V0+`
 
       void main() {
         float x = getAAtOutCoords();
         `+e.output+` = encode_float(x);
       }
-    `}return n}();var v5=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=rn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
+    `}return n}();var v5=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=nn.DOWNLOAD;var e=Ft();this.outputShape=t,this.userCode=`
       `+V0+`
 
       void main() {
@@ -1981,7 +1981,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+e.output+` = encode_float(x);
       }
     `}return n}();var y5=function(){function n(t,e,i){i===void 0&&(i=!1),this.variableNames=["A"];var r=Ft(),a=e[0],s=e[1];this.outputShape=t;var o="result";i&&(o="floor(result * 255. + 0.5)"),this.userCode=`
-      `+Ud(t)+`
+      `+Bd(t)+`
 
       void main() {
         ivec3 coords = getOutputCoords();
@@ -2039,7 +2039,7 @@ return (round(mod(b, 2.0)) != 1) ?
             }
           }
         `}this.userCode=`
-      `+Ud(t)+`
+      `+Bd(t)+`
 
       void main() {
         ivec3 coords = getOutputCoords();
@@ -2118,7 +2118,7 @@ return (round(mod(b, 2.0)) != 1) ?
     void main() {
       gl_Position = vec4(clipSpacePos, 1);
       resultUV = uv;
-    }`;return b0(n,e)}function lS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return I0(n,t)}function uS(n){var t=new Uint16Array([0,1,2,2,1,3]);return A0(n,t)}function rs(n,t,e,i,r,a){N0(t,e);var s=T0(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function Pd(n){return n.internalFormatFloat}function cS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Pd(i),i.textureFormatFloat,n.FLOAT)}function _d(n){return n.internalFormatHalfFloat}function hS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,_d(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function Md(n){return n.downloadTextureFormat}function dS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Md(i),n.RGBA,n.UNSIGNED_BYTE)}function Hd(n){return n.internalFormatPackedFloat}function pS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,Hd(i),n.RGBA,n.FLOAT)}function Vd(n){return n.internalFormatPackedHalfFloat}function fS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,Vd(i),n.RGBA,i.textureTypeHalfFloat)}function mS(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=Ed(n,t,"clipSpacePos",e,3,a,i);return s&&Ed(n,t,"uv",e,2,a,r)}function gS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function vS(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function yS(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function bS(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function wS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array($P(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function SS(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(XP(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function LS(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var A5={__proto__:null,createVertexShader:oS,createVertexBuffer:lS,createIndexBuffer:uS,getInternalFormatForFloat32MatrixTexture:Pd,createFloat32MatrixTexture:cS,getInternalFormatForFloat16MatrixTexture:_d,createFloat16MatrixTexture:hS,getInternalFormatForUnsignedBytesMatrixTexture:Md,createUnsignedBytesMatrixTexture:dS,getInternalFormatForPackedMatrixTexture:Hd,createPackedMatrixTexture:pS,getInternalFormatForFloat16PackedMatrixTexture:Vd,createFloat16PackedMatrixTexture:fS,bindVertexProgramAttributeStreams:mS,uploadDenseMatrixToTexture:gS,uploadPixelDataToTexture:vS,createBufferFromOutputTexture:yS,downloadFloat32MatrixFromBuffer:bS,downloadByteEncodedFloatMatrixFromOutputTexture:wS,downloadPackedMatrixFromBuffer:SS,downloadMatrixFromPackedOutputTexture:LS};var IS=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=N.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,g0(e,t)):this.gl=_n(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(N.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ts(this.gl,a),an(this.gl,s))this.textureHalfFloatExtension=ts(this.gl,s);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),an(this.gl,r))this.colorBufferHalfFloatExtension=ts(this.gl,r);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",an(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(an(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=lS(this.gl),this.indexBuffer=uS(this.gl),this.framebuffer=x0(this.gl),this.textureConfig=Od(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return N.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture.  This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),cS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),hS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),dS(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),vS(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),gS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),fS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),pS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(Dd(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return wS(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return SS(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return bS(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=yS(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(N.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return LS(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=w0(e,t),r=oS(e),a=S0(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),L0(e,a),this.debug&&Fo(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=mS(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&Fo(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?O0(this.gl,t,e):E0(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),D0(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=ea(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&Fo(this.gl,this.program),ns(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ts(this.gl,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Do(this,void 0,void 0,function(){var e=this;return ko(this,function(i){switch(i.label){case 0:return[4,N.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=T5(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;N.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),Wo(this.gl,t,this.framebuffer),this.debug&&ns(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(Wo(this.gl,this.outputTexture,this.framebuffer),this.debug&&ns(this.gl)):Dd(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;Wo(r,t,this.framebuffer),this.debug&&ns(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function T5(n){for(var t=0;t<n.length;++t){var e=n[t]();if(!e)break}return t-1}function N5(n,t,e,i){var r=t.userCode,a=e.map(function(g,v){var b={logicalShape:g.shape,texShape:g.isUniform?null:g.texData.texShape,isUniform:g.isUniform,isPacked:g.isUniform?!1:g.texData.isPacked,flatOffset:null};return g.texData!=null&&g.texData.slice!=null&&g.texData.slice.flatOffset>0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=j_(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);N.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p<t.variableNames.length;p++){var f=t.variableNames[p],m=!1;d[f]=n.getUniformLocation(u,f,m),d["offset"+f]=n.getUniformLocation(u,"offset"+f,m)}return{program:t,source:l,webGLProgram:u,uniformLocations:d,inShapeInfos:s,outShapeInfo:o,infLoc:c,nanLoc:h}}function AS(n,t){if(n.length!==t.length)throw Error("Binary was compiled with "+n.length+" inputs, but "+("was executed with "+t.length+" inputs"));n.forEach(function(e,i){var r=e.logicalShape,a=t[i],s=a.shape;if(!N.util.arraysEqual(r,s))throw Error("Binary was compiled with different shapes than "+("the current args. Shapes "+r+" and "+s+" must match"));if(e.isUniform&&a.isUniform)return;var o=e.texShape,l=a.isUniform?null:a.texData.texShape;if(!N.util.arraysEqual(o,l))throw Error("Binary was compiled with different texture shapes than the"+(" current args. Shape "+o+" and "+l+" must match"))})}function x5(n,t,e,i,r){AS(t.inShapeInfos,e),AS([t.outShapeInfo],[i]);var a=i.texData.texture,s=i.texData.texShape;i.texData.isPacked?n.setOutputPackedMatrixTexture(a,s[0],s[1]):n.setOutputMatrixTexture(a,s[0],s[1]),n.setProgram(t.webGLProgram),N.env().getNumber("WEBGL_VERSION")===1&&(t.infLoc!==null&&n.gl.uniform1f(t.infLoc,Infinity)),t.nanLoc!==null&&n.gl.uniform1f(t.nanLoc,NaN),e.forEach(function(o,l){var u=t.program.variableNames[l],c=t.uniformLocations[u],h=t.uniformLocations["offset"+u];if(c==null)return;if(o.isUniform){if(N.util.sizeFromShape(o.shape)<2)n.gl.uniform1f(c,o.uniformValues[0]);else{var d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),n.gl.uniform1fv(c,d)}return}o.texData.slice!=null&&h!=null&&n.gl.uniform1i(h,o.texData.slice.flatOffset),n.setInputMatrixTexture(o.texData.texture,c,l)}),r!=null&&r(n,t.webGLProgram),n.executeProgram()}function C5(n,t,e){var i="";t.concat(e).forEach(function(s){var o=s.texData!=null&&s.texData.slice!=null&&s.texData.slice.flatOffset>0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var R5=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Ft(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var x=0;x<=1;x++)S+=`
+    }`;return b0(n,e)}function lS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return I0(n,t)}function uS(n){var t=new Uint16Array([0,1,2,2,1,3]);return A0(n,t)}function rs(n,t,e,i,r,a){N0(t,e);var s=T0(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function _d(n){return n.internalFormatFloat}function cS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,_d(i),i.textureFormatFloat,n.FLOAT)}function Md(n){return n.internalFormatHalfFloat}function hS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Md(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function Hd(n){return n.downloadTextureFormat}function dS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1];return rs(n,a,s,Hd(i),n.RGBA,n.UNSIGNED_BYTE)}function Vd(n){return n.internalFormatPackedFloat}function pS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,Vd(i),n.RGBA,n.FLOAT)}function qd(n){return n.internalFormatPackedHalfFloat}function fS(n,t,e,i){var r=ea(t,e),a=r[0],s=r[1];return rs(n,a,s,qd(i),n.RGBA,i.textureTypeHalfFloat)}function mS(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=Dd(n,t,"clipSpacePos",e,3,a,i);return s&&Dd(n,t,"uv",e,2,a,r)}function gS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function vS(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function yS(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function bS(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function wS(n,t,e,i){var r=Qa(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array($P(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function SS(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(XP(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function LS(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var A5={__proto__:null,createVertexShader:oS,createVertexBuffer:lS,createIndexBuffer:uS,getInternalFormatForFloat32MatrixTexture:_d,createFloat32MatrixTexture:cS,getInternalFormatForFloat16MatrixTexture:Md,createFloat16MatrixTexture:hS,getInternalFormatForUnsignedBytesMatrixTexture:Hd,createUnsignedBytesMatrixTexture:dS,getInternalFormatForPackedMatrixTexture:Vd,createPackedMatrixTexture:pS,getInternalFormatForFloat16PackedMatrixTexture:qd,createFloat16PackedMatrixTexture:fS,bindVertexProgramAttributeStreams:mS,uploadDenseMatrixToTexture:gS,uploadPixelDataToTexture:vS,createBufferFromOutputTexture:yS,downloadFloat32MatrixFromBuffer:bS,downloadByteEncodedFloatMatrixFromOutputTexture:wS,downloadPackedMatrixFromBuffer:SS,downloadMatrixFromPackedOutputTexture:LS};var IS=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=N.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,g0(e,t)):this.gl=_n(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(N.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ts(this.gl,a),rn(this.gl,s))this.textureHalfFloatExtension=ts(this.gl,s);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),rn(this.gl,r))this.colorBufferHalfFloatExtension=ts(this.gl,r);else if(N.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",rn(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(rn(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=lS(this.gl),this.indexBuffer=uS(this.gl),this.framebuffer=x0(this.gl),this.textureConfig=Ed(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return N.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture.  This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),cS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),hS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),dS(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),vS(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),gS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),fS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),pS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(kd(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return wS(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return SS(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return bS(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=yS(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(N.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return LS(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=w0(e,t),r=oS(e),a=S0(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),L0(e,a),this.debug&&Fo(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=mS(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&Fo(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?O0(this.gl,t,e):E0(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),D0(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=ea(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&Fo(this.gl,this.program),ns(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ts(this.gl,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Do(this,void 0,void 0,function(){var e=this;return ko(this,function(i){switch(i.label){case 0:return[4,N.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=T5(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;N.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),Wo(this.gl,t,this.framebuffer),this.debug&&ns(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(Wo(this.gl,this.outputTexture,this.framebuffer),this.debug&&ns(this.gl)):kd(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;Wo(r,t,this.framebuffer),this.debug&&ns(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function T5(n){for(var t=0;t<n.length;++t){var e=n[t]();if(!e)break}return t-1}function N5(n,t,e,i){var r=t.userCode,a=e.map(function(g,v){var b={logicalShape:g.shape,texShape:g.isUniform?null:g.texData.texShape,isUniform:g.isUniform,isPacked:g.isUniform?!1:g.texData.isPacked,flatOffset:null};return g.texData!=null&&g.texData.slice!=null&&g.texData.slice.flatOffset>0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=j_(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);N.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p<t.variableNames.length;p++){var f=t.variableNames[p],m=!1;d[f]=n.getUniformLocation(u,f,m),d["offset"+f]=n.getUniformLocation(u,"offset"+f,m)}return{program:t,source:l,webGLProgram:u,uniformLocations:d,inShapeInfos:s,outShapeInfo:o,infLoc:c,nanLoc:h}}function AS(n,t){if(n.length!==t.length)throw Error("Binary was compiled with "+n.length+" inputs, but "+("was executed with "+t.length+" inputs"));n.forEach(function(e,i){var r=e.logicalShape,a=t[i],s=a.shape;if(!N.util.arraysEqual(r,s))throw Error("Binary was compiled with different shapes than "+("the current args. Shapes "+r+" and "+s+" must match"));if(e.isUniform&&a.isUniform)return;var o=e.texShape,l=a.isUniform?null:a.texData.texShape;if(!N.util.arraysEqual(o,l))throw Error("Binary was compiled with different texture shapes than the"+(" current args. Shape "+o+" and "+l+" must match"))})}function x5(n,t,e,i,r){AS(t.inShapeInfos,e),AS([t.outShapeInfo],[i]);var a=i.texData.texture,s=i.texData.texShape;i.texData.isPacked?n.setOutputPackedMatrixTexture(a,s[0],s[1]):n.setOutputMatrixTexture(a,s[0],s[1]),n.setProgram(t.webGLProgram),N.env().getNumber("WEBGL_VERSION")===1&&(t.infLoc!==null&&n.gl.uniform1f(t.infLoc,Infinity)),t.nanLoc!==null&&n.gl.uniform1f(t.nanLoc,NaN),e.forEach(function(o,l){var u=t.program.variableNames[l],c=t.uniformLocations[u],h=t.uniformLocations["offset"+u];if(c==null)return;if(o.isUniform){if(N.util.sizeFromShape(o.shape)<2)n.gl.uniform1f(c,o.uniformValues[0]);else{var d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),n.gl.uniform1fv(c,d)}return}o.texData.slice!=null&&h!=null&&n.gl.uniform1i(h,o.texData.slice.flatOffset),n.setInputMatrixTexture(o.texData.texture,c,l)}),r!=null&&r(n,t.webGLProgram),n.executeProgram()}function C5(n,t,e){var i="";t.concat(e).forEach(function(s){var o=s.texData!=null&&s.texData.slice!=null&&s.texData.slice.flatOffset>0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var R5=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Ft(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var x=0;x<=1;x++)S+=`
           blockIndex = rc.y + `+x+`;
           pos = rc.x + `+L+`;
 
@@ -2408,7 +2408,7 @@ return (round(mod(b, 2.0)) != 1) ?
         }
         setOutput(dotProd);
       }
-    `}return n}();var qd=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) {
+    `}return n}();var Gd=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) {
           vec4 b = getPreluActivationWeightsAtOutCoords();
           `+s+`
         }`:f=`vec4 activation(vec4 x) {
@@ -2475,7 +2475,7 @@ return (round(mod(b, 2.0)) != 1) ?
         void main() {
           setOutput(vec4(getA(), 0., 0., 0.));
         }
-      `;else{var i=Jt("rc",e),r=Ze(e),a=B5(e,t,i),s=z5(e,t[t.length-1],t[t.length-2],i),o=P5(t,i);this.userCode=`
+      `;else{var i=Xt("rc",e),r=Ze(e),a=B5(e,t,i),s=z5(e,t[t.length-1],t[t.length-2],i),o=P5(t,i);this.userCode=`
         void main() {
           `+r+` rc = getOutputCoords();
 
@@ -2525,7 +2525,7 @@ return (round(mod(b, 2.0)) != 1) ?
           setOutput(getX(`+l+`));
         }
       }
-    `}return n}();var V5=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Ze(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=Jt("rc",r),u=Jt("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1;
+    `}return n}();var V5=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Ze(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=Xt("rc",r),u=Xt("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1;
        if(`+c+`) {
       `,r===1?"":`}
        rc = outputLoc;
@@ -2690,7 +2690,7 @@ return (round(mod(b, 2.0)) != 1) ?
         }
         setOutput(`+L+`);
       }
-    `}return n}(),Gd=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=`
+    `}return n}(),Yd=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=`
         const ivec3 strides =
             ivec3(`+o+", "+l+", "+u+`);
         const ivec3 pads = ivec3(`+g+", "+v+", "+b+`);
@@ -2951,7 +2951,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+(r>0?"}":"")+`
       `}this.userCode=`
       `+q5(e)+`
-      `+Ud(t)+`
+      `+Bd(t)+`
 
       void main() {
         ivec3 rc = getOutputCoords();
@@ -3265,7 +3265,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+s+` coords = getOutputCoords();
         setOutput(getX(`+a+`));
       }
-    `}return n}();var J5=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=Jt("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Ze(i);i===1?this.userCode=`
+    `}return n}();var J5=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=Xt("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Ze(i);i===1?this.userCode=`
         void main(){
           int rc = getOutputCoords();
           vec4 result = vec4(0.);
@@ -3429,7 +3429,7 @@ return (round(mod(b, 2.0)) != 1) ?
           setOutput(getB(`+a+`));
         }
       }
-    `}return n}();var t9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i="uniform int start["+this.rank+"];",r=e9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+Yd[l]+" = start["+l+"] + coords."+Yd[l]+";"});a=`
+    `}return n}();var t9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i="uniform int start["+this.rank+"];",r=e9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+Kd[l]+" = start["+l+"] + coords."+Kd[l]+";"});a=`
         `+e+` sourceLoc;
         `+e+` coords = getOutputCoords();
         `+s.join(`
@@ -3440,7 +3440,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+a+`
         setOutput(getSource(`+r+`));
       }
-    `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),Yd=["x","y","z","w","u","v"];function e9(n){if(n===1)return"sourceLoc";if(n<=6)return Yd.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var n9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i=Jt("coords",this.rank),r=Jt("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=`
+    `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),Kd=["x","y","z","w","u","v"];function e9(n){if(n===1)return"sourceLoc";if(n<=6)return Kd.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var n9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Ze(this.rank),i=Xt("coords",this.rank),r=Xt("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=`
       result.x = `+s+`;
       if (++`+i[this.rank-1]+" < "+t[this.rank-1]+`) {
         ++`+r[this.rank-1]+`;
@@ -3478,7 +3478,7 @@ return (round(mod(b, 2.0)) != 1) ?
         `+s+` coords = getOutputCoords();
         setOutput(getX(`+o+`));
       }
-    `}return n}();var r9=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=RS(e,i),a=OS(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=CS(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Ct.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Ct.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Ct.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=RS(i,r),s=OS(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=CS(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=N.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function a9(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function CS(n,t,e,i,r){var a=s9(t,i),s;if(r){var o=ea(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=Qa(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=a9(e,a);return s*p}function s9(n,t){switch(n){case Ct.PACKED_2X2_FLOAT32:return Hd(t);case Ct.PACKED_2X2_FLOAT16:return Vd(t);case Ct.UNPACKED_FLOAT32:return Pd(t);case Ct.UNPACKED_FLOAT16:return _d(t);case Ct.PACKED_4X1_UNSIGNED_BYTE:return Md(t);default:throw new Error("Unknown physical texture type "+n)}}function o9(n){return N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Ct.PACKED_2X2_FLOAT32:Ct.UNPACKED_FLOAT32:n?Ct.PACKED_2X2_FLOAT16:Ct.UNPACKED_FLOAT16}function RS(n,t){if(n===rn.UPLOAD)return Ct.PACKED_2X2_FLOAT32;if(n===rn.RENDER||n==null)return o9(t);if(n===rn.DOWNLOAD||n===rn.PIXELS)return Ct.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function OS(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var u9=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[r]*e[r];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=l9(t);this.userCode=`
+    `}return n}();var r9=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=RS(e,i),a=OS(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=CS(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Ct.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Ct.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Ct.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Ct.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=RS(i,r),s=OS(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=CS(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=N.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function a9(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function CS(n,t,e,i,r){var a=s9(t,i),s;if(r){var o=ea(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=Qa(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=a9(e,a);return s*p}function s9(n,t){switch(n){case Ct.PACKED_2X2_FLOAT32:return Vd(t);case Ct.PACKED_2X2_FLOAT16:return qd(t);case Ct.UNPACKED_FLOAT32:return _d(t);case Ct.UNPACKED_FLOAT16:return Md(t);case Ct.PACKED_4X1_UNSIGNED_BYTE:return Hd(t);default:throw new Error("Unknown physical texture type "+n)}}function o9(n){return N.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Ct.PACKED_2X2_FLOAT32:Ct.UNPACKED_FLOAT32:n?Ct.PACKED_2X2_FLOAT16:Ct.UNPACKED_FLOAT16}function RS(n,t){if(n===nn.UPLOAD)return Ct.PACKED_2X2_FLOAT32;if(n===nn.RENDER||n==null)return o9(t);if(n===nn.DOWNLOAD||n===nn.PIXELS)return Ct.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function OS(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var u9=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[r]*e[r];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=l9(t);this.userCode=`
       void main() {
         `+a+` resRC = getOutputCoords();
         setOutput(getA(`+s+`));
@@ -3634,14 +3634,14 @@ return (round(mod(b, 2.0)) != 1) ?
 
         setOutput(y);
       }
-    `}return n}();var P9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=Jt("rc",e),r=Ze(e),a=P_(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=`
+    `}return n}();var P9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=Xt("rc",e),r=Ze(e),a=P_(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=`
       void main() {
         `+r+` rc = getOutputCoords();
         vec4 packedInput = getA(`+a+`);
 
         setOutput(getChannel(packedInput, `+o+`));
       }
-    `}return n}();var VS=N.backend_util.segment_util,_9=N.kernel_impls.split,M9=N.kernel_impls.tile,H9=N.kernel_impls.topkImpl,V9=N.kernel_impls.whereImpl,q9=1e-7,G9=1e-4,Vo={};function Y9(n){return n in Vo||(Vo[n]={}),Vo[n]}function qo(n,t){if(t===void 0&&(t=!1),n==="linear")return t?B9:c9;if(n==="relu")return t?_S:DS;if(n==="elu")return t?HS:FS;if(n==="relu6")return t?MS:kS;if(n==="prelu")return t?Z0:J0;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var K9=128,j9=600;function $9(){return N.env().global.screen==null?1024:N.env().global.screen.height*N.env().global.screen.width*window.devicePixelRatio*j9/1024/1024}var qS=1e3,GS=function(n){YP(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!N.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=_n(N.env().getNumber("WEBGL_VERSION"));i.binaryCache=Y9(N.env().getNumber("WEBGL_VERSION")),i.gpgpu=new IS(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new r9(i.gpgpu),i.numMBBeforeWarning=$9(),i.texData=new N.DataStorage(i,N.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((N.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||N.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:rn.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(N.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:rn.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new ss(l,Ho):c=new Re(l,Ho);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=N.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=N.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=N.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R;return ko(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new ss(s,Ho):h=new Re(s,Ho),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&N.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&N.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(es(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=N.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=N.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),x=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(x)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,x]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i<e.length;i++){var r=e[i];if(!y0(r))throw N.env().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error("The value "+r+" cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'"):Error("The value "+r+" cannot be represented on this device.")}},t.prototype.getValuesFromTexture=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.isPacked,l=N.util.sizeFromShape(a);if(N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){var u=this.decode(e),c=this.texData.get(u.dataId),h=(i=this.gpgpu).downloadMatrixFromPackedTexture.apply(i,[c.texture].concat(es(a))).subarray(0,l);return this.disposeIntermediateTensorInfo(u),h}var d=N.env().getBool("WEBGL_PACK")&&o===!0,p=d?Uo(a):a,f=d?new v5(p):new g5(p),m=this.runWebGLProgram(f,[{shape:p,dtype:s,dataId:e}],"float32"),g=this.texData.get(m.dataId),v=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(g.texture,g.texShape[0],g.texShape[1]).subarray(0,l);return this.disposeIntermediateTensorInfo(m),v},t.prototype.time=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u;return ko(this,function(c){switch(c.label){case 0:return i=this.activeTimers,r=[],a=!1,this.programTimersStack==null?(this.programTimersStack=r,a=!0):this.activeTimers.push(r),this.activeTimers=r,e(),s=N.util.flatten(this.activeTimers.map(function(h){return h.query})).filter(function(h){return h!=null}),o=N.util.flatten(this.activeTimers.map(function(h){return h.name})).filter(function(h){return h!=null}),this.activeTimers=i,a&&(this.programTimersStack=null),l={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=N.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:N.util.now(),endMs:null}},t.prototype.endTimer=function(e){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=N.util.now(),e)},t.prototype.getQueryTime=function(e){return Do(this,void 0,void 0,function(){var i;return ko(this,function(r){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return N.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=N.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=K9);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&N.util.sizeFromShape(s.shape)<i})},t.prototype.getGPGPUContext=function(){return this.gpgpu},t.prototype.complex=function(e,i){var r=this.makeOutput(e.shape,"complex64"),a=this.texData.get(r.dataId);return a.complexTensors={real:N.engine().keep(e.clone()),imag:N.engine().keep(i.clone())},r},t.prototype.real=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.real.clone()},t.prototype.imag=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.imag.clone()},t.prototype.slice=function(e,i,r){if(this.shouldExecuteOnCPU([e])){var a=k_(this.texData.get(e.dataId).values,i,r,e.shape,e.dtype);return this.makeOutput(r,e.dtype,a)}if(N.util.sizeFromShape(r)===0)return N.tensor([],r,e.dtype);var s=this.texData.get(e.dataId).isPacked,o=N.slice_util.isSliceContinous(e.shape,i,r);if(s||!o){var l=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new n9(r):new t9(r),u=l.getCustomSetupFunc(i);return this.compileAndRun(l,[e],null,u)}return this.uploadToGPU(e.dataId),this.shallowSlice(e,i,r)},t.prototype.shallowSlice=function(e,i,r){var a=this.texData.get(e.dataId),s=this.makeOutput(r,e.dtype),o=this.texData.get(s.dataId);Object.assign(o,a),o.shape=r,o.dtype=e.dtype;var l=N.slice_util.computeFlatOffset(i,e.strides);a.slice&&(l+=a.slice.flatOffset),o.slice={flatOffset:l,origDataId:a.slice&&a.slice.origDataId||e.dataId};var u=this.dataRefCount.get(o.slice.origDataId)||1;return this.dataRefCount.set(o.slice.origDataId,u+1),s},t.prototype.stridedSlice=function(e,i,r,a){var s=this,o=this.tryRunOnCpuOrThrow([e],function(){return s.cpuBackend.stridedSlice(e,i,r,a)});if(o)return o;var l=N.slice_util.computeOutShape(i,r,a);if(l.some(function(c){return c===0}))return N.tensor([],l);var u=new i9(i,a,l);return this.compileAndRun(u,[e])},t.prototype.reverse=function(e,i){var r=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new J5(e.shape,i):new X5(e.shape,i);return this.compileAndRun(r,[e])},t.prototype.concat=function(e,i){if(e[0].dtype==="complex64"){var r=e.map(function(f){return N.real(f)}),a=e.map(function(f){return N.imag(f)});return N.complex(this.concat(r,i),this.concat(a,i))}if(e.length===1)return e[0];if(e.length>N.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new i5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=N.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,N.util.sizeFromShape(f.shape.slice(i)))}),d=new n5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,WS,e.dtype);var a=new Re(e.shape,WS);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>qS){r&&(e=N.transpose(e,[0,2,1])),a&&(i=N.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=N.upcastType(e.dtype,i.dtype),m=new qd(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=N.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?qo(l,!0):null,b=new qd(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new j0(K0.REAL,e.shape,i.shape),o=new j0(K0.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=E_(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,X0,e.dtype);var g=new wt(X0,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=N.env().getBool("WEBGL_PACK_NORMALIZATION")?new D5(e.shape,i,r,a,s):new O5(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new E5(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return N.util.decodeString(l)}),s=N.buffer(e.shape,e.dtype,a);return M9(s,i)}var o=new u9(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new V5(e.shape,i,r):new H5(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new L5(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=N.backend_util.getReshaped(e.shape,i,a),o=N.backend_util.getPermuted(s.length,i.length),l=N.backend_util.getReshapedPermuted(e.shape,i,a),u=N.backend_util.getSliceBeginCoords(r,i.length),c=N.backend_util.getSliceSize(l,r,i.length);return N.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;o<e.shape.length;++o)s.push([0,0]);var l=e.pad(s),u=N.backend_util.getReshaped(l.shape,i,a,!1),c=N.backend_util.getPermuted(u.length,i.length,!1),h=N.backend_util.getReshapedPermuted(l.shape,i,a,!1),d=N.transpose(l.reshape(u),c);return N.reshape(d,h)},t.prototype.reduce=function(e,i,r){var a=e.shape[0],s=e.shape[1],o=N.backend_util.computeOptimalWindowSize(s),l=Math.ceil(s/o),u={windowSize:o,inSize:s,batchSize:a,outSize:l},c=new TS(u,i),h=this.compileAndRun(c,[e],r);return h.shape[1]===1?h:this.reduce(h,i,r)},t.prototype.argReduce=function(e,i,r){r===void 0&&(r=null);var a=e.shape[0],s=e.shape[1];r!=null&&(a=r.shape[0],s=r.shape[1]);var o=N.backend_util.computeOptimalWindowSize(s),l={windowSize:o,inSize:s,batchSize:a,outSize:Math.ceil(s/o)},u=new z_(l,i,r==null),c=[e];r!=null&&c.push(r);var h=this.compileAndRun(u,c,"int32");return h.shape[1]===1?h:this.argReduce(e,i,h)},t.prototype.argReducePacked=function(e,i,r){r===void 0&&(r=null);var a=r!=null?r.shape:e.shape,s=a[a.length-1],o=N.backend_util.computeOptimalWindowSize(s),l=new LM(a,o,i,r==null),u=r==null?[e]:[e,r],c=this.compileAndRun(l,u,"int32");return c.rank===e.rank?this.argReducePacked(e,i,c):c},t.prototype.sum=function(e,i){N.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o),u=N.sumOutType(e.dtype);return this.reduce(l,"sum",u).reshape(a)},t.prototype.prod=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e],function(){return r.cpuBackend.prod(e,i)});if(a)return a;var s=N.backend_util.computeOutAndReduceShapes(e.shape,i),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u),h=N.sumOutType(e.dtype);return this.reduce(c,"prod",h).reshape(o)},t.prototype.unsortedSegmentSum=function(e,i,r){var a=0,s=N.backend_util.getAxesPermutation([a],e.rank),o=e;s!=null&&(o=N.transpose(e,s),a=N.backend_util.getInnerMostAxes(1,e.rank)[0]);var l=VS.computeOutShape(o.shape,a,r),u=N.util.sizeFromShape([o.shape[a]]),c=o.as2D(-1,u),h=N.sumOutType(e.dtype),d=this.segOpCompute(c,"unsortedSegmentSum",i,h,r).reshape(l);return s!=null&&(d=N.transpose(d,N.backend_util.getUndoAxesPermutation(s))),d},t.prototype.segOpCompute=function(e,i,r,a,s){var o=e.shape[0],l=e.shape[1],u=VS.segOpComputeOptimalWindowSize(l,s),c={windowSize:u,inSize:l,batchSize:o,numSegments:s},h=new Z5(c,i),d=this.compileAndRun(h,[e,r],a);return d.shape[1]===s?d:(r=N.range(0,s).tile([l/u]),this.segOpCompute(d,i,r,a,s))},t.prototype.argMinMaxReduce=function(e,i,r){var a=[i];if(N.backend_util.assertAxesAreInnerMostDims("arg"+r.charAt(0).toUpperCase()+r.slice(1),a,e.rank),!N.env().getBool("WEBGL_PACK_REDUCE")||e.rank<=2){var s=N.backend_util.computeOutAndReduceShapes(e.shape,a),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u);return this.argReduce(c,r).reshape(o)}return this.argReducePacked(e,r)},t.prototype.argMin=function(e,i){return this.argMinMaxReduce(e,i,"min")},t.prototype.argMax=function(e,i){return this.argMinMaxReduce(e,i,"max")},t.prototype.cumsum=function(e,i,r,a){if(i!==e.rank-1)throw new Error("WebGL cumsum shader expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=e.shape[i],o=e,l=0;l<=Math.ceil(Math.log2(s))-1;l++){var u=new rS(e.shape,!1,a),c=u.getCustomSetupFunc(l),h=o;o=this.compileAndRun(u,[o],o.dtype,c),h.dispose()}if(r){var u=new rS(e.shape,r,a),h=o;o=this.compileAndRun(u,[o]),h.dispose()}return o},t.prototype.equal=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,HM,"bool");var r=new wt(xM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.notEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,VM,"bool");var r=new wt(CM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.less=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.less(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,qM,"bool");var s=new wt(RM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.lessEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,GM,"bool");var r=new wt(OM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.greater=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.greater(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,YM,"bool");var s=new wt(EM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.greaterEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,KM,"bool");var r=new wt(DM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalNot=function(e){var i=new Re(e.shape,W9);return this.compileAndRun(i,[e])},t.prototype.logicalAnd=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,jM,"bool");var r=new wt(kM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalOr=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,$M,"bool");var r=new wt(FM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.select=function(e,i,r){var a=new Q5(e.rank,i.shape,i.rank);return this.compileAndRun(a,[e,i,r],N.upcastType(i.dtype,r.dtype))},t.prototype.where=function(e){N.backend_util.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");var i=e.dataSync();return V9(e.shape,i)},t.prototype.topk=function(e,i,r){var a=e.dataSync();return H9(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){N.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"min",l.dtype).reshape(a)},t.prototype.minimum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.minimum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(JM,e.shape,i.shape):new wt(UM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.mod=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(ZM,e.shape,i.shape):new wt(BM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.maximum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.maximum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(XM,e.shape,i.shape):new wt(WM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.all=function(e,i){N.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"all",l.dtype).reshape(a)},t.prototype.any=function(e,i){N.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"any",l.dtype).reshape(a)},t.prototype.floorDiv=function(e,i){var r=TM,a="int32";if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,PM,a);var s=new wt(r,e.shape,i.shape);return this.compileAndRun(s,[e,i],a)},t.prototype.add=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,Bd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=A_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,Bd,r);var c=new wt(Bd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.packedUnaryOp=function(e,i,r){var a=new ss(e.shape,i);return this.compileAndRun(a,[e],r)},t.prototype.packedBinaryOp=function(e,i,r,a,s){s===void 0&&(s=!1);var o=new Ti(r,e.shape,i.shape,s);return this.compileAndRun(o,[e,i],a)},t.prototype.complexSeparableBinaryOp=function(e,i,r){var a=this,s=this.texData.get(e.dataId),o=this.texData.get(i.dataId),l=[[s.complexTensors.real,o.complexTensors.real],[s.complexTensors.imag,o.complexTensors.imag]].map(function(d){var p=d[0],f=d[1],m=a.makeComplexComponentTensorInfo(e,p),g=a.makeComplexComponentTensorInfo(i,f),v=new wt(r,e.shape,i.shape);return a.compileAndRun(v,[m,g],N.upcastType(p.dtype,f.dtype))}),u=l[0],c=l[1],h=this.complex(u,c);return u.dispose(),c.dispose(),h},t.prototype.makeComplexComponentTensorInfo=function(e,i){return{dataId:i.dataId,dtype:i.dtype,shape:e.shape}},t.prototype.addN=function(e){if(e.length===1)return e[0];if(e.length>N.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return N.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=N.env().getBool("WEBGL_PACK"),u=l?new B_(e[0].shape,o):new U_(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,zd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=F_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,zd,e.dtype);var c=new wt(zd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new Ti(_M,e.shape,i.shape):new wt(NM,e.shape,i.shape),s=N.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=T_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,US,e.dtype);var r=new Re(e.shape,US);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=C_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,BS,e.dtype);var r=new Re(e.shape,BS);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,p9);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,f9);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,m9);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,g9);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,v9);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=N_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,zS,e.dtype);var r=new Re(e.shape,zS);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=x_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,PS,e.dtype);var r=new Re(e.shape,PS);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=N.util.parseAxisParam([i],e.shape),a=N.max(e,r),s=N.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return N.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=R_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,z9,e.dtype);var r=new Re(e.shape,y9);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,b9);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,w9);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=D_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,S9);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,F9);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,_S):i=new Re(e.shape,DS),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,MS):i=new Re(e.shape,kS),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(Z0,e.shape,i.shape):new wt(J0,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,HS,e.dtype);var i=new Re(e.shape,FS);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(MM,e.shape,i.shape):new wt(zM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,h9);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,U9);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;N.env().getBool("WEBGL_PACK_CLIP")?a=new e5(e.shape):a=new QM(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=I_(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,ES,e.dtype);var r=new Re(e.shape,ES);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new t5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,L9);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,I9);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,A9);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,T9);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,N9);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,x9);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,C9);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,R9);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,O9);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,E9);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,D9);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,k9);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,d9(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>qS,v=l[2]%2!==0&&!!u.isPacked;if(g||!N.env().getBool("WEBGL_LAZILY_UNPACK")||!N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=N.reshape(e,[1,b,r.inChannels]),S=N.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return N.reshape(L,r.outShape)}var x=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,x,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,N.util.assert(is(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=N.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return N.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,N.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),x=new R5(v,S.shape,r),C=this.compileAndRun(x,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?qo(s,!0):null,W=new qd(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(N.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?qo(o,!1):null,d=new Q0(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(N.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new Q0(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new a5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new r5(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?qo(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new tS(a,d,c,p),this.compileAndRun(f,h)):(f=new eS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new tS(r),this.compileAndRun(a,[e,i])):(a=new eS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new u5(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new l5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new c5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new o5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new s5(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return N.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=this.slice(e,l,u).reshape(a);return c},t.prototype.avgPool3d=function(e,i){var r=new Gd(i,"avg",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.avgPool3dBackprop=function(e,i,r){var a=new AM(r);return this.compileAndRun(a,[e],i.dtype)},t.prototype.maxPool3d=function(e,i){var r=new Gd(i,"max",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.maxPool3dBackprop=function(e,i,r,a){var s=!0,o=new Gd(a,"max",s),l=this.compileAndRun(o,[i]),u=new F5(a),c=this.compileAndRun(u,[e,l],i.dtype);return l.dispose(),c},t.prototype.resizeBilinear=function(e,i,r,a){var s=N.env().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new K5(e.shape,i,r,a):new Y5(e.shape,i,r,a);return this.compileAndRun(s,[e],"float32")},t.prototype.resizeBilinearBackprop=function(e,i,r){var a=new G5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.resizeNearestNeighbor=function(e,i,r,a){var s=new $5(e.shape,i,r,a);return this.compileAndRun(s,[e])},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){var a=new j5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.multinomial=function(e,i,r,a){var s=i?e:N.softmax(e),o=s.shape[0],l=s.shape[1],u=new W5(o,l,r),c=u.getCustomSetupFunc(a);return this.compileAndRun(u,[s],"int32",c)},t.prototype.oneHot=function(e,i,r,a){var s=new U5(e.size,i,r,a);return this.compileAndRun(s,[e])},t.prototype.diag=function(e){var i=new m5(e.size);return this.compileAndRun(i,[e])},t.prototype.cropAndResize=function(e,i,r,a,s,o){var l=new h5(e.shape,i.shape,a,s,o);return this.compileAndRun(l,[e,i,r],"float32")},t.prototype.depthToSpace=function(e,i,r){N.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new f5(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return _9(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=N.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return N.backend_util.reshapeTensor(N.tensor([]),r);var f=N.scalar(0),m=new xS(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=N.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new xS(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new sS(aS.REAL,e.shape,i),s=new sS(aS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=N.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new I5(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||N.util.inferDtype(i),r==="string"){var a=N.util.getArrayFromDType(r,N.util.sizeFromShape(e));return a.fill(i),N.engine().makeTensor(a,e,r,this)}else{var s=new w5(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return N.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return N.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new P9(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new _5(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[cr(e.shape)].concat(hr(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[cr(i)].concat(hr(i)),o=new NS(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=Uo(a),l;r?l=new p5(o):l=new d5(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===Za.DENSE){var c=es(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),N.util.sizeFromShape(l.shape)===0)return u.values=N.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&N.util.sizeFromShape(w.shape)<=N.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!is(S.shape,w.shape)){var L=w,x=w.shape;w.shape=S.shape,w=o.packedReshape(w,x),h.push(w),S=o.texData.get(w.dataId),L.shape=x}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=C5(e,d,p),m=this.getAndSaveBinary(f,function(){return N5(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),x5(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!N.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return N.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!N.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=N.tidy(function(){if(!N.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=N.env().getBool("DEBUG");N.env().set("DEBUG",!1);var r=e.abs(N.scalar(1e-8)).dataSync()[0];if(N.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?q9:G9},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=N.util.now());var p=r.texShape;if(p==null&&(p=F0(a,c),r.texShape=p),o!=null){var f=Uo(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=ea(p[0],p[1]),g=i[0],v=i[1],m=new b5(f,[v,g],b)):m=new y5(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=rn.PIXELS:this.texData.get(w.dataId).usage=rn.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),x=this.texData.get(L.dataId);r.texture=x.texture,r.texShape=x.texShape,r.isPacked=x.isPacked,r.usage=x.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=N.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=X9(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*N.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(N.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(N.KernelBackend);function X9(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i<e.length;++i)e[i]=Math.round(n[i]);return e}else throw new Error("Unknown dtype "+t)}var J9="2.6.0";function YS(){N.env().set("WEBGL_FORCE_F16_TEXTURES",!0)}N.device_util.isBrowser()&&N.registerBackend("webgl",function(){return new GS},2);var Z9={forceHalfFloat:YS},KS="if (isnan(x)) return x;",Q9=`
+    `}return n}();var VS=N.backend_util.segment_util,_9=N.kernel_impls.split,M9=N.kernel_impls.tile,H9=N.kernel_impls.topkImpl,V9=N.kernel_impls.whereImpl,q9=1e-7,G9=1e-4,Vo={};function Y9(n){return n in Vo||(Vo[n]={}),Vo[n]}function qo(n,t){if(t===void 0&&(t=!1),n==="linear")return t?B9:c9;if(n==="relu")return t?_S:DS;if(n==="elu")return t?HS:FS;if(n==="relu6")return t?MS:kS;if(n==="prelu")return t?Z0:J0;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var K9=128,j9=600;function $9(){return N.env().global.screen==null?1024:N.env().global.screen.height*N.env().global.screen.width*window.devicePixelRatio*j9/1024/1024}var qS=1e3,GS=function(n){YP(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!N.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=_n(N.env().getNumber("WEBGL_VERSION"));i.binaryCache=Y9(N.env().getNumber("WEBGL_VERSION")),i.gpgpu=new IS(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new r9(i.gpgpu),i.numMBBeforeWarning=$9(),i.texData=new N.DataStorage(i,N.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((N.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||N.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:nn.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(N.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:nn.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new ss(l,Ho):c=new Re(l,Ho);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=N.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=N.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=N.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,x,C,R;return ko(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new ss(s,Ho):h=new Re(s,Ho),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&N.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&N.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(es(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=N.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=N.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),x=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(x)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,x]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i<e.length;i++){var r=e[i];if(!y0(r))throw N.env().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error("The value "+r+" cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'"):Error("The value "+r+" cannot be represented on this device.")}},t.prototype.getValuesFromTexture=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.isPacked,l=N.util.sizeFromShape(a);if(N.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){var u=this.decode(e),c=this.texData.get(u.dataId),h=(i=this.gpgpu).downloadMatrixFromPackedTexture.apply(i,[c.texture].concat(es(a))).subarray(0,l);return this.disposeIntermediateTensorInfo(u),h}var d=N.env().getBool("WEBGL_PACK")&&o===!0,p=d?Uo(a):a,f=d?new v5(p):new g5(p),m=this.runWebGLProgram(f,[{shape:p,dtype:s,dataId:e}],"float32"),g=this.texData.get(m.dataId),v=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(g.texture,g.texShape[0],g.texShape[1]).subarray(0,l);return this.disposeIntermediateTensorInfo(m),v},t.prototype.time=function(e){return Do(this,void 0,void 0,function(){var i,r,a,s,o,l,u;return ko(this,function(c){switch(c.label){case 0:return i=this.activeTimers,r=[],a=!1,this.programTimersStack==null?(this.programTimersStack=r,a=!0):this.activeTimers.push(r),this.activeTimers=r,e(),s=N.util.flatten(this.activeTimers.map(function(h){return h.query})).filter(function(h){return h!=null}),o=N.util.flatten(this.activeTimers.map(function(h){return h.name})).filter(function(h){return h!=null}),this.activeTimers=i,a&&(this.programTimersStack=null),l={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=N.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:N.util.now(),endMs:null}},t.prototype.endTimer=function(e){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=N.util.now(),e)},t.prototype.getQueryTime=function(e){return Do(this,void 0,void 0,function(){var i;return ko(this,function(r){return N.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return N.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=N.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=K9);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&N.util.sizeFromShape(s.shape)<i})},t.prototype.getGPGPUContext=function(){return this.gpgpu},t.prototype.complex=function(e,i){var r=this.makeOutput(e.shape,"complex64"),a=this.texData.get(r.dataId);return a.complexTensors={real:N.engine().keep(e.clone()),imag:N.engine().keep(i.clone())},r},t.prototype.real=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.real.clone()},t.prototype.imag=function(e){var i=this.texData.get(e.dataId);return i.complexTensors.imag.clone()},t.prototype.slice=function(e,i,r){if(this.shouldExecuteOnCPU([e])){var a=k_(this.texData.get(e.dataId).values,i,r,e.shape,e.dtype);return this.makeOutput(r,e.dtype,a)}if(N.util.sizeFromShape(r)===0)return N.tensor([],r,e.dtype);var s=this.texData.get(e.dataId).isPacked,o=N.slice_util.isSliceContinous(e.shape,i,r);if(s||!o){var l=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new n9(r):new t9(r),u=l.getCustomSetupFunc(i);return this.compileAndRun(l,[e],null,u)}return this.uploadToGPU(e.dataId),this.shallowSlice(e,i,r)},t.prototype.shallowSlice=function(e,i,r){var a=this.texData.get(e.dataId),s=this.makeOutput(r,e.dtype),o=this.texData.get(s.dataId);Object.assign(o,a),o.shape=r,o.dtype=e.dtype;var l=N.slice_util.computeFlatOffset(i,e.strides);a.slice&&(l+=a.slice.flatOffset),o.slice={flatOffset:l,origDataId:a.slice&&a.slice.origDataId||e.dataId};var u=this.dataRefCount.get(o.slice.origDataId)||1;return this.dataRefCount.set(o.slice.origDataId,u+1),s},t.prototype.stridedSlice=function(e,i,r,a){var s=this,o=this.tryRunOnCpuOrThrow([e],function(){return s.cpuBackend.stridedSlice(e,i,r,a)});if(o)return o;var l=N.slice_util.computeOutShape(i,r,a);if(l.some(function(c){return c===0}))return N.tensor([],l);var u=new i9(i,a,l);return this.compileAndRun(u,[e])},t.prototype.reverse=function(e,i){var r=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new J5(e.shape,i):new X5(e.shape,i);return this.compileAndRun(r,[e])},t.prototype.concat=function(e,i){if(e[0].dtype==="complex64"){var r=e.map(function(f){return N.real(f)}),a=e.map(function(f){return N.imag(f)});return N.complex(this.concat(r,i),this.concat(a,i))}if(e.length===1)return e[0];if(e.length>N.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new i5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=N.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,N.util.sizeFromShape(f.shape.slice(i)))}),d=new n5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,WS,e.dtype);var a=new Re(e.shape,WS);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>qS){r&&(e=N.transpose(e,[0,2,1])),a&&(i=N.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=N.upcastType(e.dtype,i.dtype),m=new Gd(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=N.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?qo(l,!0):null,b=new Gd(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new j0(K0.REAL,e.shape,i.shape),o=new j0(K0.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=E_(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,X0,e.dtype);var g=new wt(X0,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=N.env().getBool("WEBGL_PACK_NORMALIZATION")?new D5(e.shape,i,r,a,s):new O5(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new E5(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return N.util.decodeString(l)}),s=N.buffer(e.shape,e.dtype,a);return M9(s,i)}var o=new u9(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new V5(e.shape,i,r):new H5(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new L5(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=N.backend_util.getReshaped(e.shape,i,a),o=N.backend_util.getPermuted(s.length,i.length),l=N.backend_util.getReshapedPermuted(e.shape,i,a),u=N.backend_util.getSliceBeginCoords(r,i.length),c=N.backend_util.getSliceSize(l,r,i.length);return N.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){N.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;o<e.shape.length;++o)s.push([0,0]);var l=e.pad(s),u=N.backend_util.getReshaped(l.shape,i,a,!1),c=N.backend_util.getPermuted(u.length,i.length,!1),h=N.backend_util.getReshapedPermuted(l.shape,i,a,!1),d=N.transpose(l.reshape(u),c);return N.reshape(d,h)},t.prototype.reduce=function(e,i,r){var a=e.shape[0],s=e.shape[1],o=N.backend_util.computeOptimalWindowSize(s),l=Math.ceil(s/o),u={windowSize:o,inSize:s,batchSize:a,outSize:l},c=new TS(u,i),h=this.compileAndRun(c,[e],r);return h.shape[1]===1?h:this.reduce(h,i,r)},t.prototype.argReduce=function(e,i,r){r===void 0&&(r=null);var a=e.shape[0],s=e.shape[1];r!=null&&(a=r.shape[0],s=r.shape[1]);var o=N.backend_util.computeOptimalWindowSize(s),l={windowSize:o,inSize:s,batchSize:a,outSize:Math.ceil(s/o)},u=new z_(l,i,r==null),c=[e];r!=null&&c.push(r);var h=this.compileAndRun(u,c,"int32");return h.shape[1]===1?h:this.argReduce(e,i,h)},t.prototype.argReducePacked=function(e,i,r){r===void 0&&(r=null);var a=r!=null?r.shape:e.shape,s=a[a.length-1],o=N.backend_util.computeOptimalWindowSize(s),l=new LM(a,o,i,r==null),u=r==null?[e]:[e,r],c=this.compileAndRun(l,u,"int32");return c.rank===e.rank?this.argReducePacked(e,i,c):c},t.prototype.sum=function(e,i){N.backend_util.assertAxesAreInnerMostDims("sum",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o),u=N.sumOutType(e.dtype);return this.reduce(l,"sum",u).reshape(a)},t.prototype.prod=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e],function(){return r.cpuBackend.prod(e,i)});if(a)return a;var s=N.backend_util.computeOutAndReduceShapes(e.shape,i),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u),h=N.sumOutType(e.dtype);return this.reduce(c,"prod",h).reshape(o)},t.prototype.unsortedSegmentSum=function(e,i,r){var a=0,s=N.backend_util.getAxesPermutation([a],e.rank),o=e;s!=null&&(o=N.transpose(e,s),a=N.backend_util.getInnerMostAxes(1,e.rank)[0]);var l=VS.computeOutShape(o.shape,a,r),u=N.util.sizeFromShape([o.shape[a]]),c=o.as2D(-1,u),h=N.sumOutType(e.dtype),d=this.segOpCompute(c,"unsortedSegmentSum",i,h,r).reshape(l);return s!=null&&(d=N.transpose(d,N.backend_util.getUndoAxesPermutation(s))),d},t.prototype.segOpCompute=function(e,i,r,a,s){var o=e.shape[0],l=e.shape[1],u=VS.segOpComputeOptimalWindowSize(l,s),c={windowSize:u,inSize:l,batchSize:o,numSegments:s},h=new Z5(c,i),d=this.compileAndRun(h,[e,r],a);return d.shape[1]===s?d:(r=N.range(0,s).tile([l/u]),this.segOpCompute(d,i,r,a,s))},t.prototype.argMinMaxReduce=function(e,i,r){var a=[i];if(N.backend_util.assertAxesAreInnerMostDims("arg"+r.charAt(0).toUpperCase()+r.slice(1),a,e.rank),!N.env().getBool("WEBGL_PACK_REDUCE")||e.rank<=2){var s=N.backend_util.computeOutAndReduceShapes(e.shape,a),o=s[0],l=s[1],u=N.util.sizeFromShape(l),c=e.as2D(-1,u);return this.argReduce(c,r).reshape(o)}return this.argReducePacked(e,r)},t.prototype.argMin=function(e,i){return this.argMinMaxReduce(e,i,"min")},t.prototype.argMax=function(e,i){return this.argMinMaxReduce(e,i,"max")},t.prototype.cumsum=function(e,i,r,a){if(i!==e.rank-1)throw new Error("WebGL cumsum shader expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=e.shape[i],o=e,l=0;l<=Math.ceil(Math.log2(s))-1;l++){var u=new rS(e.shape,!1,a),c=u.getCustomSetupFunc(l),h=o;o=this.compileAndRun(u,[o],o.dtype,c),h.dispose()}if(r){var u=new rS(e.shape,r,a),h=o;o=this.compileAndRun(u,[o]),h.dispose()}return o},t.prototype.equal=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,HM,"bool");var r=new wt(xM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.notEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,VM,"bool");var r=new wt(CM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.less=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.less(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,qM,"bool");var s=new wt(RM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.lessEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,GM,"bool");var r=new wt(OM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.greater=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.greater(e,i)});if(a)return a;if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,YM,"bool");var s=new wt(EM,e.shape,i.shape);return this.compileAndRun(s,[e,i],"bool")},t.prototype.greaterEqual=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,KM,"bool");var r=new wt(DM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalNot=function(e){var i=new Re(e.shape,W9);return this.compileAndRun(i,[e])},t.prototype.logicalAnd=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,jM,"bool");var r=new wt(kM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.logicalOr=function(e,i){if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,$M,"bool");var r=new wt(FM,e.shape,i.shape);return this.compileAndRun(r,[e,i],"bool")},t.prototype.select=function(e,i,r){var a=new Q5(e.rank,i.shape,i.rank);return this.compileAndRun(a,[e,i,r],N.upcastType(i.dtype,r.dtype))},t.prototype.where=function(e){N.backend_util.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");var i=e.dataSync();return V9(e.shape,i)},t.prototype.topk=function(e,i,r){var a=e.dataSync();return H9(a,e.shape,e.dtype,i,r)},t.prototype.min=function(e,i){N.backend_util.assertAxesAreInnerMostDims("min",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"min",l.dtype).reshape(a)},t.prototype.minimum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.minimum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(JM,e.shape,i.shape):new wt(UM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.mod=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(ZM,e.shape,i.shape):new wt(BM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.maximum=function(e,i){var r=this,a=this.tryRunOnCpuOrThrow([e,i],function(){return r.cpuBackend.maximum(e,i)});if(a)return a;var s=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(XM,e.shape,i.shape):new wt(WM,e.shape,i.shape);return this.compileAndRun(s,[e,i])},t.prototype.all=function(e,i){N.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"all",l.dtype).reshape(a)},t.prototype.any=function(e,i){N.backend_util.assertAxesAreInnerMostDims("any",i,e.rank);var r=N.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=N.util.sizeFromShape(s),l=e.as2D(-1,o);return this.reduce(l,"any",l.dtype).reshape(a)},t.prototype.floorDiv=function(e,i){var r=TM,a="int32";if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,PM,a);var s=new wt(r,e.shape,i.shape);return this.compileAndRun(s,[e,i],a)},t.prototype.add=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,zd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=A_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,zd,r);var c=new wt(zd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.packedUnaryOp=function(e,i,r){var a=new ss(e.shape,i);return this.compileAndRun(a,[e],r)},t.prototype.packedBinaryOp=function(e,i,r,a,s){s===void 0&&(s=!1);var o=new Ti(r,e.shape,i.shape,s);return this.compileAndRun(o,[e,i],a)},t.prototype.complexSeparableBinaryOp=function(e,i,r){var a=this,s=this.texData.get(e.dataId),o=this.texData.get(i.dataId),l=[[s.complexTensors.real,o.complexTensors.real],[s.complexTensors.imag,o.complexTensors.imag]].map(function(d){var p=d[0],f=d[1],m=a.makeComplexComponentTensorInfo(e,p),g=a.makeComplexComponentTensorInfo(i,f),v=new wt(r,e.shape,i.shape);return a.compileAndRun(v,[m,g],N.upcastType(p.dtype,f.dtype))}),u=l[0],c=l[1],h=this.complex(u,c);return u.dispose(),c.dispose(),h},t.prototype.makeComplexComponentTensorInfo=function(e,i){return{dataId:i.dataId,dtype:i.dtype,shape:e.shape}},t.prototype.addN=function(e){if(e.length===1)return e[0];if(e.length>N.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return N.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=N.env().getBool("WEBGL_PACK"),u=l?new B_(e[0].shape,o):new U_(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,Pd);var r=N.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=F_(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,Pd,e.dtype);var c=new wt(Pd,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new Ti(_M,e.shape,i.shape):new wt(NM,e.shape,i.shape),s=N.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=T_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,US,e.dtype);var r=new Re(e.shape,US);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=C_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,BS,e.dtype);var r=new Re(e.shape,BS);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,p9);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,f9);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,m9);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,g9);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,v9);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=N_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,zS,e.dtype);var r=new Re(e.shape,zS);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=x_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,PS,e.dtype);var r=new Re(e.shape,PS);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=N.util.parseAxisParam([i],e.shape),a=N.max(e,r),s=N.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return N.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=R_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,z9,e.dtype);var r=new Re(e.shape,y9);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,b9);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,w9);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=D_(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,S9);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,F9);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,_S):i=new Re(e.shape,DS),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return N.env().getBool("WEBGL_PACK")?i=new ss(e.shape,MS):i=new Re(e.shape,kS),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(Z0,e.shape,i.shape):new wt(J0,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,HS,e.dtype);var i=new Re(e.shape,FS);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(MM,e.shape,i.shape):new wt(zM,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,h9);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,U9);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;N.env().getBool("WEBGL_PACK_CLIP")?a=new e5(e.shape):a=new QM(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=I_(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(N.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,ES,e.dtype);var r=new Re(e.shape,ES);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new t5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,L9);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,I9);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,A9);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,T9);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,N9);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,x9);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,C9);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,R9);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,O9);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,E9);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,D9);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,k9);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,d9(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>qS,v=l[2]%2!==0&&!!u.isPacked;if(g||!N.env().getBool("WEBGL_LAZILY_UNPACK")||!N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=N.reshape(e,[1,b,r.inChannels]),S=N.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return N.reshape(L,r.outShape)}var x=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,x,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,N.util.assert(is(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=N.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return N.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,N.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),x=new R5(v,S.shape,r),C=this.compileAndRun(x,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?qo(s,!0):null,W=new Gd(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(N.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?qo(o,!1):null,d=new Q0(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(N.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new Q0(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new a5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new r5(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?qo(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new tS(a,d,c,p),this.compileAndRun(f,h)):(f=new eS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return N.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new tS(r),this.compileAndRun(a,[e,i])):(a=new eS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new u5(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new l5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new c5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new o5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new s5(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return N.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o<e.rank;o++)o!==i&&(a[s++]=e.shape[o]);var l=new Array(e.rank).fill(0),u=e.shape.slice();u[i]=1;for(var c=new Array(r),o=0;o<c.length;o++)l[i]=o,c[o]=this.slice(e,l,u).reshape(a);return c},t.prototype.avgPool3d=function(e,i){var r=new Yd(i,"avg",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.avgPool3dBackprop=function(e,i,r){var a=new AM(r);return this.compileAndRun(a,[e],i.dtype)},t.prototype.maxPool3d=function(e,i){var r=new Yd(i,"max",!1);return this.compileAndRun(r,[e],"float32")},t.prototype.maxPool3dBackprop=function(e,i,r,a){var s=!0,o=new Yd(a,"max",s),l=this.compileAndRun(o,[i]),u=new F5(a),c=this.compileAndRun(u,[e,l],i.dtype);return l.dispose(),c},t.prototype.resizeBilinear=function(e,i,r,a){var s=N.env().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new K5(e.shape,i,r,a):new Y5(e.shape,i,r,a);return this.compileAndRun(s,[e],"float32")},t.prototype.resizeBilinearBackprop=function(e,i,r){var a=new G5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.resizeNearestNeighbor=function(e,i,r,a){var s=new $5(e.shape,i,r,a);return this.compileAndRun(s,[e])},t.prototype.resizeNearestNeighborBackprop=function(e,i,r){var a=new j5(e,i,r);return this.compileAndRun(a,[e])},t.prototype.multinomial=function(e,i,r,a){var s=i?e:N.softmax(e),o=s.shape[0],l=s.shape[1],u=new W5(o,l,r),c=u.getCustomSetupFunc(a);return this.compileAndRun(u,[s],"int32",c)},t.prototype.oneHot=function(e,i,r,a){var s=new U5(e.size,i,r,a);return this.compileAndRun(s,[e])},t.prototype.diag=function(e){var i=new m5(e.size);return this.compileAndRun(i,[e])},t.prototype.cropAndResize=function(e,i,r,a,s,o){var l=new h5(e.shape,i.shape,a,s,o);return this.compileAndRun(l,[e,i,r],"float32")},t.prototype.depthToSpace=function(e,i,r){N.util.assert(i>1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new f5(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return _9(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=N.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return N.backend_util.reshapeTensor(N.tensor([]),r);var f=N.scalar(0),m=new xS(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=N.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new xS(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new sS(aS.REAL,e.shape,i),s=new sS(aS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=N.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new I5(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||N.util.inferDtype(i),r==="string"){var a=N.util.getArrayFromDType(r,N.util.sizeFromShape(e));return a.fill(i),N.engine().makeTensor(a,e,r,this)}else{var s=new w5(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return N.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return N.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new P9(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new _5(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[cr(e.shape)].concat(hr(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[cr(i)].concat(hr(i)),o=new NS(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=Uo(a),l;r?l=new p5(o):l=new d5(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===Za.DENSE){var c=es(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),N.util.sizeFromShape(l.shape)===0)return u.values=N.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&N.util.sizeFromShape(w.shape)<=N.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!is(S.shape,w.shape)){var L=w,x=w.shape;w.shape=S.shape,w=o.packedReshape(w,x),h.push(w),S=o.texData.get(w.dataId),L.shape=x}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=C5(e,d,p),m=this.getAndSaveBinary(f,function(){return N5(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),x5(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!N.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return N.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!N.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=N.tidy(function(){if(!N.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=N.env().getBool("DEBUG");N.env().set("DEBUG",!1);var r=e.abs(N.scalar(1e-8)).dataSync()[0];if(N.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?q9:G9},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=N.util.now());var p=r.texShape;if(p==null&&(p=F0(a,c),r.texShape=p),o!=null){var f=Uo(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=ea(p[0],p[1]),g=i[0],v=i[1],m=new b5(f,[v,g],b)):m=new y5(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=nn.PIXELS:this.texData.get(w.dataId).usage=nn.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),x=this.texData.get(L.dataId);r.texture=x.texture,r.texShape=x.texShape,r.isPacked=x.isPacked,r.usage=x.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=N.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=X9(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*N.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(N.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(N.KernelBackend);function X9(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i<e.length;++i)e[i]=Math.round(n[i]);return e}else throw new Error("Unknown dtype "+t)}var J9="2.6.0";function YS(){N.env().set("WEBGL_FORCE_F16_TEXTURES",!0)}N.device_util.isBrowser()&&N.registerBackend("webgl",function(){return new GS},2);var Z9={forceHalfFloat:YS},KS="if (isnan(x)) return x;",Q9=`
   if (isnan(a)) return a;
   if (isnan(b)) return b;
 `,e6=`
@@ -3649,14 +3649,14 @@ return (round(mod(b, 2.0)) != 1) ?
   result.g = isNaN.g > 0. ? NAN : result.g;
   result.b = isNaN.b > 0. ? NAN : result.b;
   result.a = isNaN.a > 0. ? NAN : result.a;
-`;function Go(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function Kd(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(t,l.shape,u.shape,!!e):new wt(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var t6=Q9+`
+`;function Go(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function jd(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=N.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new Ti(t,l.shape,u.shape,!!e):new wt(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var t6=Q9+`
   return atan(a, b);
 `,n6=`
   vec4 result = atan(a, b);
   vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
   `+e6+`
   return result;
-`,i6=Kd(t6,n6),r6={kernelName:N.Atan2,backendName:"webgl",kernelFunc:i6};function jd(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var a6={kernelName:N.Identity,backendName:"webgl",kernelFunc:jd};function s6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return jd({inputs:{x:r},backend:e});var h=new as(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var o6={kernelName:N.AvgPool,backendName:"webgl",kernelFunc:s6};function l6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ta([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=N.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new IM(c);return e.runWebGLProgram(h,[r],s.dtype)}var u6={kernelName:N.AvgPoolBackprop,backendName:"webgl",kernelFunc:l6};var c6=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],N.backend_util.assertAndGetBroadcastShape(t,e),N.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(N.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(N.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
+`,i6=jd(t6,n6),r6={kernelName:N.Atan2,backendName:"webgl",kernelFunc:i6};function $d(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var a6={kernelName:N.Identity,backendName:"webgl",kernelFunc:$d};function s6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return $d({inputs:{x:r},backend:e});var h=new as(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var o6={kernelName:N.AvgPool,backendName:"webgl",kernelFunc:s6};function l6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ta([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=N.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new IM(c);return e.runWebGLProgram(h,[r],s.dtype)}var u6={kernelName:N.AvgPoolBackprop,backendName:"webgl",kernelFunc:l6};var c6=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],N.backend_util.assertAndGetBroadcastShape(t,e),N.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(N.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(N.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
       void main() {
         float x = getXAtOutCoords();
         float mean = getMeanAtOutCoords();
@@ -3703,7 +3703,7 @@ return a / b;`,y6=`
   }
 
   return result;
-`,b6=Kd(v6,y6,!0),w6={kernelName:N.Div,backendName:"webgl",kernelFunc:b6};var S6=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=`
+`,b6=jd(v6,y6,!0),w6={kernelName:N.Div,backendName:"webgl",kernelFunc:b6};var S6=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=`
         void main() {
           ivec4 coords = getOutputCoords();
           int x = coords[2];
@@ -3773,7 +3773,7 @@ return a / b;`,y6=`
 
         `+e.output+` = result;
       }
-    `}return n}();var N6={kernelName:N.FromPixels,backendName:"webgl",kernelFunc:T6},oa;function T6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(oa==null&&(oa=document.createElement("canvas").getContext("2d")),oa.canvas.width=u,oa.canvas.height=c,oa.drawImage(r,0,0,u,c),r=oa.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=rn.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=N.env().getBool("WEBGL_PACK")?new A6(d):new I6(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function x6(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=N.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function C6(n,t,e,i){for(var r=x6(n.shape),a=n,s=0;s<r.length;s++){var o=r[s],l=o.inSize,u=o.windowSize,c=o.outSize,h=new TS({windowSize:u,inSize:l,batchSize:n.shape[0],outSize:c},e),d=a;a=i.runWebGLProgram(h,[a],t),d.dataId!==n.dataId&&i.disposeData(d.dataId)}return a}function R6(n,t,e){var i=[cr(n.shape)].concat(hr(n.shape)),r={dtype:n.dtype,shape:i,dataId:n.dataId},a=[cr(t)].concat(hr(t)),s=new NS(a,i),o=!0,l=e.runWebGLProgram(s,[r],n.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function $d(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=e,o=N.util.sizeFromShape(r.shape),l=N.util.inferFromImplicitShape(a,o),u=N.util.sizeFromShape(l);N.util.assert(o===u,function(){return"The new shape ("+l+") has "+u+" elements and the old "+("shape ("+r.shape+") has "+o+" elements. The new shape and old ")+"shape must have the same number of elements."});var c=s.texData.get(r.dataId);return c.isPacked&&!is(r.shape,l)&&!(c.texture!==null&&is(c.shape,l))?R6(r,l,s):(s.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var O6={kernelName:N.Reshape,backendName:"webgl",kernelFunc:$d};function E6(n,t,e,i){var r=N.util.sizeFromShape(t),a=N.util.sizeFromShape(n.shape),s=a/r,o=$d({inputs:{x:n},attrs:{shape:[s,r]},backend:i}),l=C6(o,n.dtype,"max",i),u=$d({inputs:{x:l},attrs:{shape:e},backend:i});return i.disposeIntermediateTensorInfo(o),i.disposeIntermediateTensorInfo(l),u}var k6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[e[r]];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=D6(e);this.userCode=`
+    `}return n}();var N6={kernelName:N.FromPixels,backendName:"webgl",kernelFunc:T6},oa;function T6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(oa==null&&(oa=document.createElement("canvas").getContext("2d")),oa.canvas.width=u,oa.canvas.height=c,oa.drawImage(r,0,0,u,c),r=oa.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=nn.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=N.env().getBool("WEBGL_PACK")?new A6(d):new I6(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function x6(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=N.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function C6(n,t,e,i){for(var r=x6(n.shape),a=n,s=0;s<r.length;s++){var o=r[s],l=o.inSize,u=o.windowSize,c=o.outSize,h=new TS({windowSize:u,inSize:l,batchSize:n.shape[0],outSize:c},e),d=a;a=i.runWebGLProgram(h,[a],t),d.dataId!==n.dataId&&i.disposeData(d.dataId)}return a}function R6(n,t,e){var i=[cr(n.shape)].concat(hr(n.shape)),r={dtype:n.dtype,shape:i,dataId:n.dataId},a=[cr(t)].concat(hr(t)),s=new NS(a,i),o=!0,l=e.runWebGLProgram(s,[r],n.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function Xd(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.shape,s=e,o=N.util.sizeFromShape(r.shape),l=N.util.inferFromImplicitShape(a,o),u=N.util.sizeFromShape(l);N.util.assert(o===u,function(){return"The new shape ("+l+") has "+u+" elements and the old "+("shape ("+r.shape+") has "+o+" elements. The new shape and old ")+"shape must have the same number of elements."});var c=s.texData.get(r.dataId);return c.isPacked&&!is(r.shape,l)&&!(c.texture!==null&&is(c.shape,l))?R6(r,l,s):(s.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var O6={kernelName:N.Reshape,backendName:"webgl",kernelFunc:Xd};function E6(n,t,e,i){var r=N.util.sizeFromShape(t),a=N.util.sizeFromShape(n.shape),s=a/r,o=Xd({inputs:{x:n},attrs:{shape:[s,r]},backend:i}),l=C6(o,n.dtype,"max",i),u=Xd({inputs:{x:l},attrs:{shape:e},backend:i});return i.disposeIntermediateTensorInfo(o),i.disposeIntermediateTensorInfo(l),u}var k6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r<i.length;r++)i[r]=t[e[r]];this.outputShape=i,this.rank=i.length;var a=Ze(this.rank),s=D6(e);this.userCode=`
     void main() {
       `+a+` resRC = getOutputCoords();
       setOutput(getA(`+s+`));
@@ -3795,7 +3795,7 @@ return a / b;`,y6=`
       }
       setOutput(result);
     }
-    `}return n}();function jS(n,t,e){var i=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new F6(n.shape,t):new k6(n.shape,t);return e.runWebGLProgram(i,[n],n.dtype)}var W6={kernelName:N.Max,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape.length,c=N.util.parseAxisParam(s,r.shape),h=c,d=N.backend_util.getAxesPermutation(h,u),p=d!=null,f=l.shouldExecuteOnCPU([r]),m=r;if(p){if(f){for(var g=l.texData.get(m.dataId),v=g.values,b=new Array(u),w=0;w<b.length;w++)b[w]=r.shape[d[w]];var S=M0(v,r.shape,r.dtype,d,b);m=l.makeTensorInfo(b,r.dtype);var L=l.texData.get(m.dataId);L.values=S}else m=jS(r,d,l);h=N.backend_util.getInnerMostAxes(h.length,u)}N.backend_util.assertAxesAreInnerMostDims("max",h,u);var x=N.backend_util.computeOutAndReduceShapes(m.shape,h),C=x[0],R=x[1],D=C;o&&(D=N.backend_util.expandShapeToKeepDim(C,c));var k;if(f){var g=l.texData.get(m.dataId),v=g.values,W=O_(v,N.util.sizeFromShape(R),D,r.dtype);k=l.makeTensorInfo(D,r.dtype);var F=l.texData.get(k.dataId);F.values=W}else k=E6(m,R,D,l);return p&&l.disposeIntermediateTensorInfo(m),k}};function U6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return jd({inputs:{x:r},backend:e});var h=new as(c,"max",!1);return e.runWebGLProgram(h,[r],r.dtype)}var B6={kernelName:N.MaxPool,backendName:"webgl",kernelFunc:U6};function z6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ta([a,s],"maxPoolBackprop");var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=N.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=!0,f=new as(d,"max",p),m=e.runWebGLProgram(f,[o],o.dtype),g=new k5(d),v=e.runWebGLProgram(g,[r,m],o.dtype);return e.disposeIntermediateTensorInfo(m),v}var P6={kernelName:N.MaxPoolBackprop,backendName:"webgl",kernelFunc:z6};function _6(n,t,e,i){var r=new as(e,"max",!1),a=i.runWebGLProgram(r,[n],"float32");r=new as(e,"max",!0,!0,t);var s=i.runWebGLProgram(r,[n],"float32");return[a,s]}var M6={kernelName:N.MaxPoolWithArgmax,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;N.util.assert(r.shape.length===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+r.shape.length+"."});var h=[1,1];N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(o,h),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+o+" and dilations '"+h+"'")});var d=N.backend_util.computePool2DInfo(r.shape,s,o,h,l),p=_6(r,u,d,c),f=p[0],m=p[1];return[f,m]}};var H6={kernelName:N.NonMaxSuppressionV3,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=e,d=h.readSync(a.dataId),p=h.readSync(s.dataId),f=l,m=u,g=c;return N.kernel_impls.nonMaxSuppressionV3Impl(d,p,f,m,g)}};var V6=N.kernel_impls.nonMaxSuppressionV4Impl,q6={kernelName:N.NonMaxSuppressionV4,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=V6(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var G6=N.kernel_impls.nonMaxSuppressionV5Impl,Y6={kernelName:N.NonMaxSuppressionV5,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=l,g=u,v=c,b=h,w=G6(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var K6=function(){function n(t,e,i,r){this.variableNames=["Image"],this.outputShape=[];var a=t[1],s=t[2],o=Math.sin(e).toFixed(3),l=Math.cos(e).toFixed(3);this.outputShape=t;var u=N.backend_util.getImageCenter(r,a,s),c=u[0],h=u[1],d=c.toFixed(3),p=h.toFixed(3),f="";typeof i=="number"?f="float outputValue = "+i.toFixed(2)+";":f=`
+    `}return n}();function jS(n,t,e){var i=N.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new F6(n.shape,t):new k6(n.shape,t);return e.runWebGLProgram(i,[n],n.dtype)}var W6={kernelName:N.Max,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.reductionIndices,o=a.keepDims,l=i,u=r.shape.length,c=N.util.parseAxisParam(s,r.shape),h=c,d=N.backend_util.getAxesPermutation(h,u),p=d!=null,f=l.shouldExecuteOnCPU([r]),m=r;if(p){if(f){for(var g=l.texData.get(m.dataId),v=g.values,b=new Array(u),w=0;w<b.length;w++)b[w]=r.shape[d[w]];var S=M0(v,r.shape,r.dtype,d,b);m=l.makeTensorInfo(b,r.dtype);var L=l.texData.get(m.dataId);L.values=S}else m=jS(r,d,l);h=N.backend_util.getInnerMostAxes(h.length,u)}N.backend_util.assertAxesAreInnerMostDims("max",h,u);var x=N.backend_util.computeOutAndReduceShapes(m.shape,h),C=x[0],R=x[1],D=C;o&&(D=N.backend_util.expandShapeToKeepDim(C,c));var k;if(f){var g=l.texData.get(m.dataId),v=g.values,W=O_(v,N.util.sizeFromShape(R),D,r.dtype);k=l.makeTensorInfo(D,r.dtype);var F=l.texData.get(k.dataId);F.values=W}else k=E6(m,R,D,l);return p&&l.disposeIntermediateTensorInfo(m),k}};function U6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ta(r,"maxPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=N.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&N.util.arraysEqual(c.inShape,c.outShape))return $d({inputs:{x:r},backend:e});var h=new as(c,"max",!1);return e.runWebGLProgram(h,[r],r.dtype)}var B6={kernelName:N.MaxPool,backendName:"webgl",kernelFunc:U6};function z6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=t.output,o=a;ta([a,s],"maxPoolBackprop");var l=i.filterSize,u=i.strides,c=i.pad,h=i.dimRoundingMode,d=N.backend_util.computePool2DInfo(o.shape,l,u,1,c,h),p=!0,f=new as(d,"max",p),m=e.runWebGLProgram(f,[o],o.dtype),g=new k5(d),v=e.runWebGLProgram(g,[r,m],o.dtype);return e.disposeIntermediateTensorInfo(m),v}var P6={kernelName:N.MaxPoolBackprop,backendName:"webgl",kernelFunc:z6};function _6(n,t,e,i){var r=new as(e,"max",!1),a=i.runWebGLProgram(r,[n],"float32");r=new as(e,"max",!0,!0,t);var s=i.runWebGLProgram(r,[n],"float32");return[a,s]}var M6={kernelName:N.MaxPoolWithArgmax,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;N.util.assert(r.shape.length===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+r.shape.length+"."});var h=[1,1];N.util.assert(N.backend_util.eitherStridesOrDilationsAreOne(o,h),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+o+" and dilations '"+h+"'")});var d=N.backend_util.computePool2DInfo(r.shape,s,o,h,l),p=_6(r,u,d,c),f=p[0],m=p[1];return[f,m]}};var H6={kernelName:N.NonMaxSuppressionV3,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=e,d=h.readSync(a.dataId),p=h.readSync(s.dataId),f=l,m=u,g=c;return N.kernel_impls.nonMaxSuppressionV3Impl(d,p,f,m,g)}};var V6=N.kernel_impls.nonMaxSuppressionV4Impl,q6={kernelName:N.NonMaxSuppressionV4,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=V6(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var G6=N.kernel_impls.nonMaxSuppressionV5Impl,Y6={kernelName:N.NonMaxSuppressionV5,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs;N.backend_util.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");var r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e,p=d.readSync(a.dataId),f=d.readSync(s.dataId),m=l,g=u,v=c,b=h,w=G6(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var K6=function(){function n(t,e,i,r){this.variableNames=["Image"],this.outputShape=[];var a=t[1],s=t[2],o=Math.sin(e).toFixed(3),l=Math.cos(e).toFixed(3);this.outputShape=t;var u=N.backend_util.getImageCenter(r,a,s),c=u[0],h=u[1],d=c.toFixed(3),p=h.toFixed(3),f="";typeof i=="number"?f="float outputValue = "+i.toFixed(2)+";":f=`
         vec3 fill = vec3(`+i.join(",")+`);
         float outputValue = fill[coords[3]];`,this.userCode=`
         void main() {
@@ -3814,7 +3814,7 @@ return a / b;`,y6=`
         }
     `}return n}();var j6={kernelName:N.RotateWithOffset,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=new K6(r.shape,s,o,l),h=u.runWebGLProgram(c,[r],r.dtype);return h}};var $6=KS+`
   return sin(x);
-`,X6=Go($6),J6={kernelName:N.Sin,backendName:"webgl",kernelFunc:X6};var Z6="return x * x;",Q6=Go(Z6),e8={kernelName:N.Square,backendName:"webgl",kernelFunc:Q6};var $S="return (a - b) * (a - b);",t8=Kd($S,$S),n8={kernelName:N.SquaredDifference,backendName:"webgl",kernelFunc:t8};var i8="return tan(x);",r8=Go(i8),a8={kernelName:N.Tan,backendName:"webgl",kernelFunc:r8};var s8={kernelName:N.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u<l.length;u++)l[u]=r.shape[a[u]];var c;if(s.shouldExecuteOnCPU([r])){var h=s.texData.get(r.dataId),d=h.values,p=M0(d,r.shape,r.dtype,a,l);c=s.makeTensorInfo(l,r.dtype);var f=s.texData.get(c.dataId);f.values=p}else c=jS(r,a,s);return c}};function o8(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ta(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");var s=i.readSync(a.dataId),o=W_(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var l8={kernelName:N.Unique,backendName:"webgl",kernelFunc:o8};var u8=[r6,o6,u6,p6,g6,w6,L6,N6,a6,W6,B6,P6,M6,H6,q6,Y6,O6,j6,J6,e8,n8,a8,s8,l8];for(var Xd=0,XS=u8;Xd<XS.length;Xd++){var c8=XS[Xd];N.registerKernel(c8)}Pn.GPGPUContext=IS;Pn.MathBackendWebGL=GS;Pn.forceHalfFloat=YS;Pn.gpgpu_util=A5;Pn.setWebGLContext=g0;Pn.version_webgl=J9;Pn.webgl=Z9;Pn.webgl_util=l_});var Wt=be(fr=>{"use strict";Object.defineProperty(fr,"__esModule",{value:!0});var Jd=Xi(),Zd=Yb(),Qd=ow(),ZS=Cw(),h8=m0(),d8=JS();var p8="2.6.0";var f8={"tfjs-core":Jd.version_core,"tfjs-backend-cpu":h8.version_cpu,"tfjs-backend-webgl":d8.version_webgl,"tfjs-data":ZS.version_data,"tfjs-layers":Zd.version_layers,"tfjs-converter":Qd.version_converter,tfjs:p8};Object.keys(Jd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Jd[n]}})});Object.keys(Zd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Zd[n]}})});Object.keys(Qd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Qd[n]}})});fr.data=ZS;fr.version=f8});var iL=be(Yo=>{const ke=Wt(),QS=6;function m8(n){const t={strides:[n/16,n/8],anchors:[2,6]},e=[];for(let i=0;i<t.strides.length;i++){const r=t.strides[i],a=Math.floor((n+r-1)/r),s=Math.floor((n+r-1)/r),o=t.anchors[i];for(let l=0;l<a;l++){const u=r*(l+.5);for(let c=0;c<s;c++){const h=r*(c+.5);for(let d=0;d<o;d++)e.push([h,u])}}}return e}const eL=n=>{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},tL=n=>({startEndTensor:n,startPoint:ke.slice(n,[0,0],[-1,2]),endPoint:ke.slice(n,[0,2],[-1,2])}),g8=(n,t)=>{const e=ke.mul(n.startPoint,t),i=ke.mul(n.endPoint,t),r=ke.concat2d([e,i],1);return tL(r)};function v8(n,t,e){const i=ke.slice(n,[0,1],[-1,2]),r=ke.add(i,t),a=ke.slice(n,[0,3],[-1,2]),s=ke.div(a,e),o=ke.div(r,e),l=ke.div(s,2),u=ke.sub(o,l),c=ke.add(o,l),h=ke.mul(u,e),d=ke.mul(c,e),p=1;return ke.concat2d([h,d],p)}function y8(n,t){return ke.tidy(()=>{const e=n.box?n.box:n;return g8(e,t).startEndTensor.squeeze()})}class nL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=m8(t.detector.inputSize),this.anchors=ke.tensor2d(this.anchorsData),this.inputSize=ke.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){if(!n||n.isDisposedInternal||n.shape.length!==4||n.shape[1]<1||n.shape[2]<1)return null;const[t,e,i]=ke.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=ke.mul(ke.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u);let h;if(Array.isArray(c)){const m=c.sort((w,S)=>w.size-S.size),g=ke.concat([m[0],m[2]],2),v=ke.concat([m[1],m[3]],2),b=ke.concat([v,g],1);h=b.squeeze(0)}else h=c.squeeze();const d=v8(h,this.anchors,this.inputSize),p=ke.slice(h,[0,0],[-1,1]),f=ke.sigmoid(p).squeeze();return[h,d,f]}),r=await ke.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>ke.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l<s.length;l++){const u=s[l],c=ke.tidy(()=>{const h=tL(u),d=a[l],p=this.anchorsData[d],f=ke.slice(t,[d,QS-1],[1,-1]).squeeze().reshape([QS,-1]),m=ke.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.width,n.shape[1]/this.height]}}async estimateFaces(n){const t=ke.tidy(()=>(n instanceof ke.Tensor||(n=ke.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:e,scaleFactor:i}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(e.map(async r=>{const a=y8(r,i),[s,o,l]=await Promise.all([r.landmarks,a,r.probability].map(async f=>f.array())),u=r.anchor,[c,h]=i,d=s.map(f=>[(f[0]+u[0])*c,(f[1]+u[1])*h]),p={topLeft:o.slice(0,2),bottomRight:o.slice(2),landmarks:d,probability:l};return eL(r.box),r.landmarks.dispose(),r.probability.dispose(),a.dispose(),p}))}}async function b8(n){const t=await ke.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new nL(t,n);return e}Yo.load=b8;Yo.BlazeFaceModel=nL;Yo.disposeBox=eL});var tp=be(ep=>{ep.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};ep.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var rL=be(mr=>{const w8=Wt();function S8(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}mr.scaleBoxCoordinates=S8;function np(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}mr.getBoxSize=np;function ip(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}mr.getBoxCenter=ip;function L8(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return w8.image.cropAndResize(t,a,[0],e)}mr.cutBoxFromImageAndResize=L8;function I8(n,t=1.5){const e=ip(n),i=np(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.enlargeBox=I8;function A8(n){const t=ip(n),e=np(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.squarifyBox=A8});var uL=be(An=>{An.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function aL(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}An.normalizeRadians=aL;function T8(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return aL(e)}An.computeRotation=T8;function N8(n){return n*180/Math.PI}An.radToDegrees=N8;function sL(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function la(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}An.dot=la;function oL(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}An.getColumnFrom2DArr=oL;function lL(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(la(n[r],oL(t,a)))}return e}function x8(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=sL(t[0],t[1]),s=lL(a,r),o=sL(-t[0],-t[1]);return lL(s,o)}An.buildRotationMatrix=x8;function C8(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-la(t[0],e),-la(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}An.invertTransformMatrix=C8;function R8(n,t){return[la(n,t[0]),la(n,t[1])]}An.rotatePoint=R8;function O8(n,t){return Math.sqrt((n[0]-t[0])**2+(n[1]-t[1])**2)}An.xyDistanceBetweenPoints=O8});var hL=be(cL=>{const Ni=Wt(),Mn=rL(),xi=tp(),Ci=uL(),E8=468,D8=.25,k8=13,F8=[k8,xi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],W8=3,U8=2,B8=[W8,U8],rp=xi.MESH_ANNOTATIONS.leftEyeLower0,ap=[rp[0],rp[rp.length-1]],sp=xi.MESH_ANNOTATIONS.rightEyeLower0,op=[sp[0],sp[sp.length-1]],z8=3,P8=4,_8=71,lp=76;function Ko(n,t,e,i){for(let r=0;r<xi.MESH_TO_IRIS_INDICES_MAP.length;r++){const{key:a,indices:s}=xi.MESH_TO_IRIS_INDICES_MAP[r],o=xi.MESH_ANNOTATIONS[`${e}${a}`],l=i==null;if(l||i.includes(a))for(let u=0;u<s.length;u++){const c=s[u];n[o[u]]=[t[c][0],t[c][1],(t[c][2]+n[o[u]][2])/2]}}}class M8{constructor(n,t,e,i){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.irisModel=e,this.meshWidth=i.mesh.inputSize,this.meshHeight=i.mesh.inputSize,this.irisSize=i.iris.inputSize,this.irisEnlarge=i.iris.enlargeFactor}transformRawCoords(n,t,e,i){const r=Mn.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Ci.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Ci.rotatePoint(d,o),d[2]]),u=Ci.invertTransformMatrix(i),c=[...Mn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Ci.dot(c,u[0]),Ci.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[ap[0]][2],e=n[op[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Mn.squarifyBox(Mn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Mn.getBoxSize(a);let o=Ni.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Ni.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a<lp;a++){const s=n[a*3],o=n[a*3+1],l=n[a*3+2];r.push([(i?1-s/this.irisSize:s/this.irisSize)*e[0]+t.startPoint[0],o/this.irisSize*e[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(_8)}}getAdjustedIrisCoords(n,t,e){const i=n[xi.MESH_ANNOTATIONS[`${e}EyeUpper0`][z8]][2],r=n[xi.MESH_ANNOTATIONS[`${e}EyeLower0`][P8]][2],a=(i+r)/2;return t.map((s,o)=>{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Mn.scaleBoxCoordinates(o,r),u=Mn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Ni.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=E8;let[o,l]=F8;s===!1&&([o,l]=B8),a=Ci.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Mn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Ci.IDENTITY_MATRIX;a!==0&&(h=Ni.image.rotateWithOffset(n,a,0,c),d=Ci.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Mn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Ni.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:x,boxSize:C,crop:R}=this.getEyeBox(b,f,ap[0],ap[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,op[0],op[1]),F=this.irisModel.predict(Ni.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,lp*3),{rawCoords:_,iris:K}=this.getEyeCoords(H,x,C,!0),j=P.slice(lp*3),{rawCoords:q,iris:G}=this.getEyeCoords(j,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(Ko(b,_,"left"),Ko(b,q,"right")):Z<1?Ko(b,_,"left",["EyeUpper0","EyeLower0"]):Ko(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,K,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Ni.dispose(b);const S=Mn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const x=Ni.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:x.arraySync()};const C={coords:x,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t<n.length;t++){const e=n[t],i=this.regionsOfInterest[t];let r=0;if(i&&i.startPoint){const[a,s]=e.startPoint,[o,l]=e.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}r<D8&&(this.regionsOfInterest[t]=e)}this.regionsOfInterest=this.regionsOfInterest.slice(0,n.length)}clearRegionOfInterest(n){this.regionsOfInterest[n]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,n),...this.regionsOfInterest.slice(n+1)])}shouldUpdateRegionsOfInterest(){const n=this.regionsOfInterest.length,t=n===0;return this.maxFaces===1||t?t:n!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}cL.Pipeline=M8});var pL=be(dL=>{dL.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var fL=be(H8=>{xp(H8,{default:()=>V8});var V8=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var vL=be(os=>{const gr=Wt(),q8=iL(),mL=tp(),G8=hL(),Y8=pL(),K8=fL().default;class gL{constructor(n,t,e,i){this.pipeline=new G8.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=gr.tidy(()=>(n instanceof gr.Tensor||(n=gr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);gr.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in mL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=mL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?gr.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function j8(n){const t=await Promise.all([q8.load(n),gr.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),gr.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new gL(t[0],t[1],t[2],n);return e}os.load=j8;os.MediaPipeFaceMesh=gL;os.uv_coords=Y8;os.triangulation=K8});var bL=be(jo=>{const sn=Wt(),Ri={};let yL={age:0,gender:""},up=0;async function $8(n,t){const e=sn.browser.fromPixels(n),i=sn.image.resizeBilinear(e,[t,t]),r=sn.cast(sn.expandDims(i,0),"float32");return r}async function X8(n){return Ri.age||(Ri.age=await sn.loadGraphModel(n.face.age.modelPath)),Ri.age}async function J8(n){return Ri.gender||(Ri.gender=await sn.loadGraphModel(n.face.gender.modelPath)),Ri.gender}async function Z8(n,t){if(up+=1,up>=t.face.age.skipFrames)return up=0,yL;let e;if(n instanceof sn.Tensor){const r=sn.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=sn.mul(r,[255]),sn.dispose(r)}else e=await $8(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await Ri.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,sn.dispose(r)}if(t.face.gender.enabled){const r=await Ri.gender.predict(e),a=await r.data();i.gender=Math.trunc(100*a[0])<50?"female":"male",sn.dispose(r)}return sn.dispose(e),yL=i,i}jo.predict=Z8;jo.loadAge=X8;jo.loadGender=J8});var LL=be(cp=>{const Ut=Wt(),Q8=["angry","discust","fear","happy","sad","surpise","neutral"],$o={};let wL=[],hp=0;const SL=1.5;function e7(n,t){const e=Ut.tidy(()=>{const i=Ut.browser.fromPixels(n,1),r=Ut.image.resizeBilinear(i,[t,t]),a=Ut.cast(Ut.expandDims(r,0),"float32");return a});return e}async function t7(n){return $o.emotion||($o.emotion=await Ut.loadGraphModel(n.face.emotion.modelPath)),$o.emotion}async function n7(n,t){if(hp+=1,hp>=t.face.emotion.skipFrames)return hp=0,wL;const e=Ut.tidy(()=>{if(n instanceof Ut.Tensor){const r=Ut.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Ut.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Ut.mul(a,[.2989]),u=Ut.mul(s,[.587]),c=Ut.mul(o,[.114]),h=Ut.addN([l,u,c]);return h}return s}return e7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await $o.emotion.predict(e),a=await r.data();for(let s=0;s<a.length;s++)SL*a[s]>t.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*SL*a[s])/100),emotion:Q8[s]});i.sort((s,o)=>o.score-s.score),Ut.dispose(r)}return Ut.dispose(e),wL=i,i}cp.predict=n7;cp.load=t7});var TL=be(IL=>{const AL=Wt();class i7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;AL.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return AL.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}IL.BaseModel=i7});var dp=be(NL=>{const xL=Wt(),r7=TL();class a7 extends r7.BaseModel{preprocessInput(n){return xL.tidy(()=>xL.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}NL.MobileNet=a7});var RL=be(CL=>{function pp(n){return Math.floor(n/2)}class s7{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(pp(n),n);)this.exchange(n,pp(n)),n=pp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(n,t))break;this.exchange(n,t),n=t}}getValueAt(n){return this.getElementValue(this.priorityQueue[n])}less(n,t){return this.getValueAt(n)<this.getValueAt(t)}exchange(n,t){const e=this.priorityQueue[n];this.priorityQueue[n]=this.priorityQueue[t],this.priorityQueue[t]=e}}CL.MaxHeap=s7});var EL=be(OL=>{const o7=RL();function l7(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;h<c;++h){const d=Math.max(i-r,0),p=Math.min(i+r+1,o);for(let f=d;f<p;++f)if(a.get(h,f,n)>t){l=!1;break}if(!l)break}return l}function u7(n,t,e){const[i,r,a]=e.shape,s=new o7.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o<i;++o)for(let l=0;l<r;++l)for(let u=0;u<a;++u){const c=e.get(o,l,u);if(c<n)continue;l7(u,c,o,l,t,e)&&s.enqueue({score:c,part:{heatmapY:o,heatmapX:l,id:u}})}return s}OL.buildPartWithScoreQueue=u7});var ls=be(Tn=>{Tn.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];Tn.NUM_KEYPOINTS=Tn.partNames.length;Tn.partIds=Tn.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const c7=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];Tn.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];Tn.connectedPartIndices=c7.map(([n,t])=>[Tn.partIds[n],Tn.partIds[t]]);Tn.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var mp=be(Oi=>{const h7=ls();function DL(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+h7.NUM_KEYPOINTS)}}Oi.getOffsetPoint=DL;function d7(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=DL(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Oi.getImageCoords=d7;function p7(n,t){const e=new Array(t);for(let i=0;i<t;i++)e[i]=n;return e}Oi.fillArray=p7;function fp(n,t,e){return n<t?t:n>e?e:n}Oi.clamp=fp;function f7(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Oi.squaredDistance=f7;function m7(n,t){return{x:n.x+t.x,y:n.y+t.y}}Oi.addVectors=m7;function g7(n,t,e){return{y:fp(n.y,t,e),x:fp(n.x,t,e)}}Oi.clampVector=g7});var BL=be(kL=>{const us=ls(),ua=mp(),FL=us.poseChain.map(([n,t])=>[us.partIds[n],us.partIds[t]]),gp=FL.map(([,n])=>n),WL=FL.map(([n])=>n);function v7(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function vp(n,t,e,i){return{y:ua.clamp(Math.round(n.y/t),0,e-1),x:ua.clamp(Math.round(n.x/t),0,i-1)}}function UL(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=vp(t.position,a,l,u),h=v7(n,c,s),d=ua.addVectors(t.position,h);let p=d;for(let g=0;g<o;g++){const v=vp(p,a,l,u),b=ua.getOffsetPoint(v.y,v.x,e,r);p=ua.addVectors({x:v.x*a,y:v.y*a},{x:b.x,y:b.y})}const f=vp(p,a,l,u),m=i.get(f.y,f.x,e);return{position:p,part:us.partNames[e],score:m}}function y7(n,t,e,i,r,a){const s=t.shape[2],o=gp.length,l=new Array(s),{part:u,score:c}=n,h=ua.getImageCoords(u,i,e);l[u.id]={score:c,part:us.partNames[u.id],position:h};for(let d=o-1;d>=0;--d){const p=gp[d],f=WL[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,a))}for(let d=0;d<o;++d){const p=WL[d],f=gp[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,r))}return l}kL.decodePose=y7});var yp=be(zL=>{const b7=EL(),w7=BL(),PL=mp();function _L(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return PL.squaredDistance(i,e,s.y,s.x)<=t})}function S7(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(_L(n,t,a,o)||(r+=s),r),0);return i/e.length}const L7=1;function I7(n,t,e,i,r,a,s=.5,o=20){const l=[],u=b7.buildPartWithScoreQueue(s,L7,n),c=o*o;for(;l.length<a&&!u.empty();){const h=u.dequeue(),d=PL.getImageCoords(h.part,r,t);if(_L(l,c,d,h.part.id))continue;const p=w7.decodePose(h,n,t,r,e,i),f=S7(l,c,p);l.push({keypoints:p,score:f})}return l}zL.decodeMultiplePoses=I7});var wp=be(dn=>{const ca=Wt(),A7=ls();function T7(n,t,e){return n<e||t<e}function N7(n,t){return A7.connectedPartIndices.reduce((e,[i,r])=>(T7(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}dn.getAdjacentKeyPoints=N7;const{NEGATIVE_INFINITY:ML,POSITIVE_INFINITY:HL}=Number;function VL(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:ML,maxY:ML,minX:HL,minY:HL})}dn.getBoundingBox=VL;function x7(n){const{minX:t,minY:e,maxX:i,maxY:r}=VL(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}dn.getBoundingBoxPoints=x7;async function C7(n){return Promise.all(n.map(t=>t.buffer()))}dn.toTensorBuffers3D=C7;function qL(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}dn.scalePose=qL;function GL(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>qL(a,t,e,i,r))}dn.scalePoses=GL;function YL(n){return n instanceof ca.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}dn.getInputTensorDimensions=YL;function bp(n){return n instanceof ca.Tensor?n:ca.browser.fromPixels(n)}dn.toInputTensor=bp;function R7(n,t,e){return ca.tidy(()=>{const i=bp(n);return i.resizeBilinear([t,e])})}dn.toResizedInputTensor=R7;function O7(n,[t,e]){const[i,r]=YL(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s<a?(o=0,l=0,u=Math.round(.5*(a*i-r)),c=Math.round(.5*(a*i-r))):(o=Math.round(.5*(1/a*r-i)),l=Math.round(.5*(1/a*r-i)),u=0,c=0);const h=ca.tidy(()=>{let d=bp(n);return d=ca.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}dn.padAndResizeTo=O7;function E7(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=GL(n,s,o,-a.top,-a.left);return l}dn.scaleAndFlipPoses=E7});var jL=be(Sp=>{const D7=Wt(),k7=dp(),F7=yp(),Xo=wp();class KL{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=Xo.getInputTensorDimensions(n),{resized:a,padding:s}=Xo.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await Xo.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await F7.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=Xo.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}Sp.PoseNet=KL;async function W7(n){const t=await D7.loadGraphModel(n.modelPath),e=new k7.MobileNet(t,n.outputStride);return new KL(e)}async function U7(n){return W7(n)}Sp.load=U7});var XL=be(Zt=>{const B7=dp(),$L=jL(),z7=yp(),Jo=ls(),cs=wp();Zt.load=$L.load;Zt.PoseNet=$L.PoseNet;Zt.MobileNet=B7.MobileNet;Zt.decodeMultiplePoses=z7.decodeMultiplePoses;Zt.partChannels=Jo.partChannels;Zt.partIds=Jo.partIds;Zt.partNames=Jo.partNames;Zt.poseChain=Jo.poseChain;Zt.getAdjacentKeyPoints=cs.getAdjacentKeyPoints;Zt.getBoundingBox=cs.getBoundingBox;Zt.getBoundingBoxPoints=cs.getBoundingBoxPoints;Zt.scaleAndFlipPoses=cs.scaleAndFlipPoses;Zt.scalePose=cs.scalePose});var Ap=be(Ei=>{const P7=Wt();function Lp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Ei.getBoxSize=Lp;function Ip(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Ei.getBoxCenter=Ip;function _7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return P7.image.cropAndResize(t,a,[0],e)}Ei.cutBoxFromImageAndResize=_7;function M7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}Ei.scaleBoxCoordinates=M7;function H7(n,t=1.5){const e=Ip(n),i=Lp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.enlargeBox=H7;function V7(n){const t=Ip(n),e=Lp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.squarifyBox=V7;function q7(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}Ei.shiftBox=q7});var ZL=be(JL=>{const qe=Wt(),G7=Ap();class Y7{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=qe.tensor2d(this.anchors),this.inputSizeTensor=qe.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=qe.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return qe.tidy(()=>{const t=qe.slice(n,[0,0],[-1,2]),e=qe.slice(n,[0,2],[-1,2]),i=qe.add(qe.div(t,this.inputSizeTensor),this.anchorsTensor),r=qe.div(e,this.doubleInputSizeTensor),a=qe.mul(qe.sub(i,r),this.inputSizeTensor),s=qe.mul(qe.add(i,r),this.inputSizeTensor);return qe.concat2d([a,s],1)})}normalizeLandmarks(n,t){return qe.tidy(()=>{const e=qe.add(qe.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return qe.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=qe.tidy(()=>qe.mul(qe.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=qe.tidy(()=>qe.sigmoid(qe.slice(i,[0,0],[-1,1])).squeeze()),a=qe.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await qe.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=qe.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=qe.slice(s,[p,0],[1,-1]),m=qe.slice(i,[p,5],[1,14]),g=qe.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=qe.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(G7.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}JL.HandDetector=Y7});var eI=be(QL=>{QL.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var aI=be(Di=>{function tI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Di.normalizeRadians=tI;function K7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return tI(e)}Di.computeRotation=K7;const nI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function ha(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Di.dot=ha;function iI(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Di.getColumnFrom2DArr=iI;function rI(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(ha(n[r],iI(t,a)))}return e}function j7(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=nI(t[0],t[1]),s=rI(a,r),o=nI(-t[0],-t[1]);return rI(s,o)}Di.buildRotationMatrix=j7;function $7(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-ha(t[0],e),-ha(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Di.invertTransformMatrix=$7;function X7(n,t){return[ha(n,t[0]),ha(n,t[1])]}Di.rotatePoint=X7});var uI=be(sI=>{const oI=Wt(),Hn=Ap(),ki=aI(),J7=.8,Z7=[0,-.4],Q7=[0,-.1],eH=1.65,lI=[0,5,9,13,17,1,2],tH=0,nH=2;class iH{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return ki.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(i,Z7)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(t,Q7)),eH),i=[];for(let r=0;r<lI.length;r++)i.push(n[lI[r]].slice(0,2));return e.palmLandmarks=i,e}transformRawCoords(n,t,e,i){const r=Hn.getBoxSize(t),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=ki.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=ki.rotatePoint(d,o);return[...p,d[2]]}),u=ki.invertTransformMatrix(i),c=[...Hn.getBoxCenter(t),1],h=[ki.dot(c,u[0]),ki.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=ki.computeRotation(a.palmLandmarks[tH],a.palmLandmarks[nH]),o=Hn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=oI.image.rotateWithOffset(n,s,0,l),c=ki.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=Hn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),v<t.minConfidence)return g.dispose(),this.regionsOfInterest[r]=[],i;const b=oI.reshape(g,[-1,3]),w=await b.array();g.dispose(),b.dispose();const S=this.transformRawCoords(w,h,s,c),L=this.getBoxForHandLandmarks(S);this.updateRegionsOfInterest(L,!1,r);const x={landmarks:S,confidence:v,box:{topLeft:L.startPoint,bottomRight:L.endPoint}};i.push(x)}return i}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>J7?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}sI.HandPipeline=iH});var dI=be(Tp=>{const vr=Wt(),rH=ZL(),cI=eI(),aH=uI();class hI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=vr.tidy(()=>(n instanceof vr.Tensor||(n=vr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(cI.MESH_ANNOTATIONS))s[o]=cI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}Tp.HandPose=hI;async function sH(n){if(vr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return vr.util.fetch(n).then(t=>t.json())}async function oH(n){const[t,e,i]=await Promise.all([sH(n.detector.anchors),vr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),vr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new rH.HandDetector(e,t,n),a=new aH.HandPipeline(r,i,n),s=new hI(a);return s}Tp.load=oH});var pI=be(lH=>{xp(lH,{default:()=>uH});var uH={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:5,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:2,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var mI=be((oV,fI)=>{fI.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});var LI=be(Nn=>{const qt=Wt(),gI=vL(),Zo=bL(),vI=LL(),yI=XL(),bI=dI(),wI=pI().default,cH=mI();let Pe;const Et={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},Vn=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),Qo=(...n)=>{Pe.console&&console.log(...n)};function SI(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=SI(a,s):e[r]=s}),e),{})}function hH(n){if(!n)return"input is not defined";const t=n.naturalWidth||n.videoWidth||n.width||n.shape&&n.shape[1]>0;if(!t||t===0)return"input is empty";if(n.readyState&&n.readyState<=2)return"input is not ready";try{qt.getBackend()}catch{return"backend not loaded"}return null}async function dH(n,t){Pe=SI(wI,t);const e=hH(n);return e?(Qo(e,n),{error:e}):new Promise(async i=>{const r=Object.values(Et).filter(h=>h).length;r===0&&Qo("Human library starting"),qt.getBackend()!==Pe.backend&&(Qo("Human library setting backend:",Pe.backend),await qt.setBackend(Pe.backend),await qt.ready());let a;qt.getBackend()==="webgl"&&(a=qt.env().get("WEBGL_PACK_DEPTHWISECONV"),qt.env().set("WEBGL_PACK_DEPTHWISECONV",!0)),Pe.face.enabled&&!Et.facemesh&&(Et.facemesh=await gI.load(Pe.face)),Pe.body.enabled&&!Et.posenet&&(Et.posenet=await yI.load(Pe.body)),Pe.hand.enabled&&!Et.handpose&&(Et.handpose=await bI.load(Pe.hand)),Pe.face.enabled&&Pe.face.age.enabled&&!Et.age&&(Et.age=await Zo.loadAge(Pe)),Pe.face.enabled&&Pe.face.gender.enabled&&!Et.gender&&(Et.gender=await Zo.loadGender(Pe)),Pe.face.enabled&&Pe.face.emotion.enabled&&!Et.emotion&&(Et.emotion=await vI.load(Pe));const s={};let o;o=Vn(),qt.engine().startScope();const l=Pe.body.enabled?await Et.posenet.estimatePoses(n,Pe.body):[];qt.engine().endScope(),s.body=Math.trunc(Vn()-o),o=Vn(),qt.engine().startScope();const u=Pe.hand.enabled?await Et.handpose.estimateHands(n,Pe.hand):[];qt.engine().endScope(),s.hand=Math.trunc(Vn()-o);const c=[];if(Pe.face.enabled){o=Vn(),qt.engine().startScope();const h=await Et.facemesh.estimateFaces(n,Pe.face);s.face=Math.trunc(Vn()-o);for(const d of h){if(!d.image||d.image.isDisposedInternal){Qo("face object is disposed:",d.image);continue}o=Vn();const p=Pe.face.age.enabled||Pe.face.gender.enabled?await Zo.predict(d.image,Pe):{};s.agegender=Math.trunc(Vn()-o),o=Vn();const f=Pe.face.emotion.enabled?await vI.predict(d.image,Pe):{};s.emotion=Math.trunc(Vn()-o),d.image.dispose();const m=d.annotations.leftEyeIris&&d.annotations.rightEyeIris?Math.max(d.annotations.leftEyeIris[3][0]-d.annotations.leftEyeIris[1][0],d.annotations.rightEyeIris[3][0]-d.annotations.rightEyeIris[1][0]):0;c.push({confidence:d.confidence,box:d.box,mesh:d.mesh,annotations:d.annotations,age:p.age,gender:p.gender,emotion:f,iris:m!==0?Math.trunc(100*11.7/m)/100:0})}qt.engine().endScope()}qt.env().set("WEBGL_PACK_DEPTHWISECONV",a),s.total=Object.values(s).reduce((h,d)=>h+d),i({face:c,body:l,hand:u,performance:s})})}Nn.detect=dH;Nn.defaults=wI;Nn.config=Pe;Nn.models=Et;Nn.facemesh=gI;Nn.ssrnet=Zo;Nn.posenet=yI;Nn.handpose=bI;Nn.tf=qt;Nn.version=cH.version});return LI();})();
+`,X6=Go($6),J6={kernelName:N.Sin,backendName:"webgl",kernelFunc:X6};var Z6="return x * x;",Q6=Go(Z6),e8={kernelName:N.Square,backendName:"webgl",kernelFunc:Q6};var $S="return (a - b) * (a - b);",t8=jd($S,$S),n8={kernelName:N.SquaredDifference,backendName:"webgl",kernelFunc:t8};var i8="return tan(x);",r8=Go(i8),a8={kernelName:N.Tan,backendName:"webgl",kernelFunc:r8};var s8={kernelName:N.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u<l.length;u++)l[u]=r.shape[a[u]];var c;if(s.shouldExecuteOnCPU([r])){var h=s.texData.get(r.dataId),d=h.values,p=M0(d,r.shape,r.dtype,a,l);c=s.makeTensorInfo(l,r.dtype);var f=s.texData.get(c.dataId);f.values=p}else c=jS(r,a,s);return c}};function o8(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ta(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");var s=i.readSync(a.dataId),o=W_(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var l8={kernelName:N.Unique,backendName:"webgl",kernelFunc:o8};var u8=[r6,o6,u6,p6,g6,w6,L6,N6,a6,W6,B6,P6,M6,H6,q6,Y6,O6,j6,J6,e8,n8,a8,s8,l8];for(var Jd=0,XS=u8;Jd<XS.length;Jd++){var c8=XS[Jd];N.registerKernel(c8)}Pn.GPGPUContext=IS;Pn.MathBackendWebGL=GS;Pn.forceHalfFloat=YS;Pn.gpgpu_util=A5;Pn.setWebGLContext=g0;Pn.version_webgl=J9;Pn.webgl=Z9;Pn.webgl_util=l_});var Wt=be(fr=>{"use strict";Object.defineProperty(fr,"__esModule",{value:!0});var Zd=Xi(),Qd=Yb(),ep=ow(),ZS=Cw(),h8=m0(),d8=JS();var p8="2.6.0";var f8={"tfjs-core":Zd.version_core,"tfjs-backend-cpu":h8.version_cpu,"tfjs-backend-webgl":d8.version_webgl,"tfjs-data":ZS.version_data,"tfjs-layers":Qd.version_layers,"tfjs-converter":ep.version_converter,tfjs:p8};Object.keys(Zd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Zd[n]}})});Object.keys(Qd).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return Qd[n]}})});Object.keys(ep).forEach(function(n){n!=="default"&&Object.defineProperty(fr,n,{enumerable:!0,get:function(){return ep[n]}})});fr.data=ZS;fr.version=f8});var iL=be(Yo=>{const ke=Wt(),QS=6;function m8(n){const t={strides:[n/16,n/8],anchors:[2,6]},e=[];for(let i=0;i<t.strides.length;i++){const r=t.strides[i],a=Math.floor((n+r-1)/r),s=Math.floor((n+r-1)/r),o=t.anchors[i];for(let l=0;l<a;l++){const u=r*(l+.5);for(let c=0;c<s;c++){const h=r*(c+.5);for(let d=0;d<o;d++)e.push([h,u])}}}return e}const eL=n=>{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},tL=n=>({startEndTensor:n,startPoint:ke.slice(n,[0,0],[-1,2]),endPoint:ke.slice(n,[0,2],[-1,2])}),g8=(n,t)=>{const e=ke.mul(n.startPoint,t),i=ke.mul(n.endPoint,t),r=ke.concat2d([e,i],1);return tL(r)};function v8(n,t,e){const i=ke.slice(n,[0,1],[-1,2]),r=ke.add(i,t),a=ke.slice(n,[0,3],[-1,2]),s=ke.div(a,e),o=ke.div(r,e),l=ke.div(s,2),u=ke.sub(o,l),c=ke.add(o,l),h=ke.mul(u,e),d=ke.mul(c,e),p=1;return ke.concat2d([h,d],p)}function y8(n,t){return ke.tidy(()=>{const e=n.box?n.box:n;return g8(e,t).startEndTensor.squeeze()})}class nL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=m8(t.detector.inputSize),this.anchors=ke.tensor2d(this.anchorsData),this.inputSize=ke.tensor1d([this.width,this.height]),this.iouThreshold=t.detector.iouThreshold,this.scaleFaces=.8,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){if(!n||n.isDisposedInternal||n.shape.length!==4||n.shape[1]<1||n.shape[2]<1)return null;const[t,e,i]=ke.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=ke.mul(ke.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u);let h;if(Array.isArray(c)){const m=c.sort((w,S)=>w.size-S.size),g=ke.concat([m[0],m[2]],2),v=ke.concat([m[1],m[3]],2),b=ke.concat([v,g],1);h=b.squeeze(0)}else h=c.squeeze();const d=v8(h,this.anchors,this.inputSize),p=ke.slice(h,[0,0],[-1,1]),f=ke.sigmoid(p).squeeze();return[h,d,f]}),r=await ke.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>ke.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l<s.length;l++){const u=s[l],c=ke.tidy(()=>{const h=tL(u),d=a[l],p=this.anchorsData[d],f=ke.slice(t,[d,QS-1],[1,-1]).squeeze().reshape([QS,-1]),m=ke.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.width,n.shape[1]/this.height]}}async estimateFaces(n){const t=ke.tidy(()=>(n instanceof ke.Tensor||(n=ke.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:e,scaleFactor:i}=await this.getBoundingBoxes(t);return t.dispose(),Promise.all(e.map(async r=>{const a=y8(r,i),[s,o,l]=await Promise.all([r.landmarks,a,r.probability].map(async f=>f.array())),u=r.anchor,[c,h]=i,d=s.map(f=>[(f[0]+u[0])*c,(f[1]+u[1])*h]),p={topLeft:o.slice(0,2),bottomRight:o.slice(2),landmarks:d,probability:l};return eL(r.box),r.landmarks.dispose(),r.probability.dispose(),a.dispose(),p}))}}async function b8(n){const t=await ke.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new nL(t,n);return e}Yo.load=b8;Yo.BlazeFaceModel=nL;Yo.disposeBox=eL});var np=be(tp=>{tp.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};tp.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var rL=be(mr=>{const w8=Wt();function S8(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}mr.scaleBoxCoordinates=S8;function ip(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}mr.getBoxSize=ip;function rp(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}mr.getBoxCenter=rp;function L8(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return w8.image.cropAndResize(t,a,[0],e)}mr.cutBoxFromImageAndResize=L8;function I8(n,t=1.5){const e=rp(n),i=ip(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.enlargeBox=I8;function A8(n){const t=rp(n),e=ip(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}mr.squarifyBox=A8});var uL=be(In=>{In.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function aL(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}In.normalizeRadians=aL;function T8(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return aL(e)}In.computeRotation=T8;function N8(n){return n*180/Math.PI}In.radToDegrees=N8;function sL(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function la(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}In.dot=la;function oL(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}In.getColumnFrom2DArr=oL;function lL(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(la(n[r],oL(t,a)))}return e}function x8(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=sL(t[0],t[1]),s=lL(a,r),o=sL(-t[0],-t[1]);return lL(s,o)}In.buildRotationMatrix=x8;function C8(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-la(t[0],e),-la(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}In.invertTransformMatrix=C8;function R8(n,t){return[la(n,t[0]),la(n,t[1])]}In.rotatePoint=R8;function O8(n,t){return Math.sqrt((n[0]-t[0])**2+(n[1]-t[1])**2)}In.xyDistanceBetweenPoints=O8});var hL=be(cL=>{const Ni=Wt(),Mn=rL(),xi=np(),Ci=uL(),E8=468,D8=.25,k8=13,F8=[k8,xi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],W8=3,U8=2,B8=[W8,U8],ap=xi.MESH_ANNOTATIONS.leftEyeLower0,sp=[ap[0],ap[ap.length-1]],op=xi.MESH_ANNOTATIONS.rightEyeLower0,lp=[op[0],op[op.length-1]],z8=3,P8=4,_8=71,up=76;function Ko(n,t,e,i){for(let r=0;r<xi.MESH_TO_IRIS_INDICES_MAP.length;r++){const{key:a,indices:s}=xi.MESH_TO_IRIS_INDICES_MAP[r],o=xi.MESH_ANNOTATIONS[`${e}${a}`],l=i==null;if(l||i.includes(a))for(let u=0;u<s.length;u++){const c=s[u];n[o[u]]=[t[c][0],t[c][1],(t[c][2]+n[o[u]][2])/2]}}}class M8{constructor(n,t,e,i){this.regionsOfInterest=[],this.runsWithoutFaceDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.irisModel=e,this.meshWidth=i.mesh.inputSize,this.meshHeight=i.mesh.inputSize,this.irisSize=i.iris.inputSize,this.irisEnlarge=i.iris.enlargeFactor}transformRawCoords(n,t,e,i){const r=Mn.getBoxSize({startPoint:t.startPoint,endPoint:t.endPoint}),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Ci.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Ci.rotatePoint(d,o),d[2]]),u=Ci.invertTransformMatrix(i),c=[...Mn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Ci.dot(c,u[0]),Ci.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[sp[0]][2],e=n[lp[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Mn.squarifyBox(Mn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Mn.getBoxSize(a);let o=Ni.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Ni.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a<up;a++){const s=n[a*3],o=n[a*3+1],l=n[a*3+2];r.push([(i?1-s/this.irisSize:s/this.irisSize)*e[0]+t.startPoint[0],o/this.irisSize*e[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(_8)}}getAdjustedIrisCoords(n,t,e){const i=n[xi.MESH_ANNOTATIONS[`${e}EyeUpper0`][z8]][2],r=n[xi.MESH_ANNOTATIONS[`${e}EyeLower0`][P8]][2],a=(i+r)/2;return t.map((s,o)=>{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Mn.scaleBoxCoordinates(o,r),u=Mn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Ni.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=E8;let[o,l]=F8;s===!1&&([o,l]=B8),a=Ci.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Mn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Ci.IDENTITY_MATRIX;a!==0&&(h=Ni.image.rotateWithOffset(n,a,0,c),d=Ci.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Mn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Ni.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:x,boxSize:C,crop:R}=this.getEyeBox(b,f,sp[0],sp[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,lp[0],lp[1]),F=this.irisModel.predict(Ni.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,up*3),{rawCoords:_,iris:K}=this.getEyeCoords(H,x,C,!0),j=P.slice(up*3),{rawCoords:q,iris:G}=this.getEyeCoords(j,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(Ko(b,_,"left"),Ko(b,q,"right")):Z<1?Ko(b,_,"left",["EyeUpper0","EyeLower0"]):Ko(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,K,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Ni.dispose(b);const S=Mn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const x=Ni.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:x.arraySync()};const C={coords:x,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t<n.length;t++){const e=n[t],i=this.regionsOfInterest[t];let r=0;if(i&&i.startPoint){const[a,s]=e.startPoint,[o,l]=e.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}r<D8&&(this.regionsOfInterest[t]=e)}this.regionsOfInterest=this.regionsOfInterest.slice(0,n.length)}clearRegionOfInterest(n){this.regionsOfInterest[n]!=null&&(this.regionsOfInterest=[...this.regionsOfInterest.slice(0,n),...this.regionsOfInterest.slice(n+1)])}shouldUpdateRegionsOfInterest(){const n=this.regionsOfInterest.length,t=n===0;return this.maxFaces===1||t?t:n!==this.maxFaces&&this.runsWithoutFaceDetector>=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}cL.Pipeline=M8});var pL=be(dL=>{dL.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var fL=be(H8=>{xp(H8,{default:()=>V8});var V8=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var vL=be(os=>{const gr=Wt(),q8=iL(),mL=np(),G8=hL(),Y8=pL(),K8=fL().default;class gL{constructor(n,t,e,i){this.pipeline=new G8.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=gr.tidy(()=>(n instanceof gr.Tensor||(n=gr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);gr.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in mL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=mL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?gr.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function j8(n){const t=await Promise.all([q8.load(n),gr.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),gr.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new gL(t[0],t[1],t[2],n);return e}os.load=j8;os.MediaPipeFaceMesh=gL;os.uv_coords=Y8;os.triangulation=K8});var bL=be(jo=>{const an=Wt(),Ri={};let yL={age:0,gender:""},$o=0;async function $8(n,t){const e=an.browser.fromPixels(n),i=an.image.resizeBilinear(e,[t,t]),r=an.cast(an.expandDims(i,0),"float32");return r}async function X8(n){return Ri.age||(Ri.age=await an.loadGraphModel(n.face.age.modelPath)),Ri.age}async function J8(n){return Ri.gender||(Ri.gender=await an.loadGraphModel(n.face.gender.modelPath)),Ri.gender}async function Z8(n,t){if($o>t.face.age.skipFrames?$o=0:$o+=1,$o===0)return yL;let e;if(n instanceof an.Tensor){const r=an.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=an.mul(r,[255]),an.dispose(r)}else e=await $8(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await Ri.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,an.dispose(r)}if(t.face.gender.enabled){const r=await Ri.gender.predict(e),a=await r.data(),s=Math.trunc(Math.abs(1.9*100*(a[0]-.5)))/100;s>t.face.gender.minConfidence&&(i.gender=a[0]<=.5?"female":"male",i.confidence=s),an.dispose(r)}return an.dispose(e),yL=i,i}jo.predict=Z8;jo.loadAge=X8;jo.loadGender=J8});var LL=be(cp=>{const Ut=Wt(),Q8=["angry","discust","fear","happy","sad","surpise","neutral"],Xo={};let wL=[],hp=0;const SL=1.5;function e7(n,t){const e=Ut.tidy(()=>{const i=Ut.browser.fromPixels(n,1),r=Ut.image.resizeBilinear(i,[t,t]),a=Ut.cast(Ut.expandDims(r,0),"float32");return a});return e}async function t7(n){return Xo.emotion||(Xo.emotion=await Ut.loadGraphModel(n.face.emotion.modelPath)),Xo.emotion}async function n7(n,t){if(hp+=1,hp>=t.face.emotion.skipFrames)return hp=0,wL;const e=Ut.tidy(()=>{if(n instanceof Ut.Tensor){const r=Ut.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Ut.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Ut.mul(a,[.2989]),u=Ut.mul(s,[.587]),c=Ut.mul(o,[.114]),h=Ut.addN([l,u,c]);return h}return s}return e7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await Xo.emotion.predict(e),a=await r.data();for(let s=0;s<a.length;s++)SL*a[s]>t.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*SL*a[s])/100),emotion:Q8[s]});i.sort((s,o)=>o.score-s.score),Ut.dispose(r)}return Ut.dispose(e),wL=i,i}cp.predict=n7;cp.load=t7});var TL=be(IL=>{const AL=Wt();class i7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;AL.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return AL.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}IL.BaseModel=i7});var dp=be(NL=>{const xL=Wt(),r7=TL();class a7 extends r7.BaseModel{preprocessInput(n){return xL.tidy(()=>xL.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}NL.MobileNet=a7});var RL=be(CL=>{function pp(n){return Math.floor(n/2)}class s7{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(pp(n),n);)this.exchange(n,pp(n)),n=pp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t<this.numberOfElements&&this.less(t,t+1)&&t++,!this.less(n,t))break;this.exchange(n,t),n=t}}getValueAt(n){return this.getElementValue(this.priorityQueue[n])}less(n,t){return this.getValueAt(n)<this.getValueAt(t)}exchange(n,t){const e=this.priorityQueue[n];this.priorityQueue[n]=this.priorityQueue[t],this.priorityQueue[t]=e}}CL.MaxHeap=s7});var EL=be(OL=>{const o7=RL();function l7(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;h<c;++h){const d=Math.max(i-r,0),p=Math.min(i+r+1,o);for(let f=d;f<p;++f)if(a.get(h,f,n)>t){l=!1;break}if(!l)break}return l}function u7(n,t,e){const[i,r,a]=e.shape,s=new o7.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o<i;++o)for(let l=0;l<r;++l)for(let u=0;u<a;++u){const c=e.get(o,l,u);if(c<n)continue;l7(u,c,o,l,t,e)&&s.enqueue({score:c,part:{heatmapY:o,heatmapX:l,id:u}})}return s}OL.buildPartWithScoreQueue=u7});var ls=be(An=>{An.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];An.NUM_KEYPOINTS=An.partNames.length;An.partIds=An.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const c7=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];An.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];An.connectedPartIndices=c7.map(([n,t])=>[An.partIds[n],An.partIds[t]]);An.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var mp=be(Oi=>{const h7=ls();function DL(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+h7.NUM_KEYPOINTS)}}Oi.getOffsetPoint=DL;function d7(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=DL(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Oi.getImageCoords=d7;function p7(n,t){const e=new Array(t);for(let i=0;i<t;i++)e[i]=n;return e}Oi.fillArray=p7;function fp(n,t,e){return n<t?t:n>e?e:n}Oi.clamp=fp;function f7(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Oi.squaredDistance=f7;function m7(n,t){return{x:n.x+t.x,y:n.y+t.y}}Oi.addVectors=m7;function g7(n,t,e){return{y:fp(n.y,t,e),x:fp(n.x,t,e)}}Oi.clampVector=g7});var BL=be(kL=>{const us=ls(),ua=mp(),FL=us.poseChain.map(([n,t])=>[us.partIds[n],us.partIds[t]]),gp=FL.map(([,n])=>n),WL=FL.map(([n])=>n);function v7(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function vp(n,t,e,i){return{y:ua.clamp(Math.round(n.y/t),0,e-1),x:ua.clamp(Math.round(n.x/t),0,i-1)}}function UL(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=vp(t.position,a,l,u),h=v7(n,c,s),d=ua.addVectors(t.position,h);let p=d;for(let g=0;g<o;g++){const v=vp(p,a,l,u),b=ua.getOffsetPoint(v.y,v.x,e,r);p=ua.addVectors({x:v.x*a,y:v.y*a},{x:b.x,y:b.y})}const f=vp(p,a,l,u),m=i.get(f.y,f.x,e);return{position:p,part:us.partNames[e],score:m}}function y7(n,t,e,i,r,a){const s=t.shape[2],o=gp.length,l=new Array(s),{part:u,score:c}=n,h=ua.getImageCoords(u,i,e);l[u.id]={score:c,part:us.partNames[u.id],position:h};for(let d=o-1;d>=0;--d){const p=gp[d],f=WL[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,a))}for(let d=0;d<o;++d){const p=WL[d],f=gp[d];l[p]&&!l[f]&&(l[f]=UL(d,l[p],f,t,e,i,r))}return l}kL.decodePose=y7});var yp=be(zL=>{const b7=EL(),w7=BL(),PL=mp();function _L(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return PL.squaredDistance(i,e,s.y,s.x)<=t})}function S7(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(_L(n,t,a,o)||(r+=s),r),0);return i/e.length}const L7=1;function I7(n,t,e,i,r,a,s=.5,o=20){const l=[],u=b7.buildPartWithScoreQueue(s,L7,n),c=o*o;for(;l.length<a&&!u.empty();){const h=u.dequeue(),d=PL.getImageCoords(h.part,r,t);if(_L(l,c,d,h.part.id))continue;const p=w7.decodePose(h,n,t,r,e,i),f=S7(l,c,p);l.push({keypoints:p,score:f})}return l}zL.decodeMultiplePoses=I7});var wp=be(hn=>{const ca=Wt(),A7=ls();function T7(n,t,e){return n<e||t<e}function N7(n,t){return A7.connectedPartIndices.reduce((e,[i,r])=>(T7(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}hn.getAdjacentKeyPoints=N7;const{NEGATIVE_INFINITY:ML,POSITIVE_INFINITY:HL}=Number;function VL(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:ML,maxY:ML,minX:HL,minY:HL})}hn.getBoundingBox=VL;function x7(n){const{minX:t,minY:e,maxX:i,maxY:r}=VL(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}hn.getBoundingBoxPoints=x7;async function C7(n){return Promise.all(n.map(t=>t.buffer()))}hn.toTensorBuffers3D=C7;function qL(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}hn.scalePose=qL;function GL(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>qL(a,t,e,i,r))}hn.scalePoses=GL;function YL(n){return n instanceof ca.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}hn.getInputTensorDimensions=YL;function bp(n){return n instanceof ca.Tensor?n:ca.browser.fromPixels(n)}hn.toInputTensor=bp;function R7(n,t,e){return ca.tidy(()=>{const i=bp(n);return i.resizeBilinear([t,e])})}hn.toResizedInputTensor=R7;function O7(n,[t,e]){const[i,r]=YL(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s<a?(o=0,l=0,u=Math.round(.5*(a*i-r)),c=Math.round(.5*(a*i-r))):(o=Math.round(.5*(1/a*r-i)),l=Math.round(.5*(1/a*r-i)),u=0,c=0);const h=ca.tidy(()=>{let d=bp(n);return d=ca.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}hn.padAndResizeTo=O7;function E7(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=GL(n,s,o,-a.top,-a.left);return l}hn.scaleAndFlipPoses=E7});var jL=be(Sp=>{const D7=Wt(),k7=dp(),F7=yp(),Jo=wp();class KL{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=Jo.getInputTensorDimensions(n),{resized:a,padding:s}=Jo.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await Jo.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await F7.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=Jo.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}Sp.PoseNet=KL;async function W7(n){const t=await D7.loadGraphModel(n.modelPath),e=new k7.MobileNet(t,n.outputStride);return new KL(e)}async function U7(n){return W7(n)}Sp.load=U7});var XL=be(Jt=>{const B7=dp(),$L=jL(),z7=yp(),Zo=ls(),cs=wp();Jt.load=$L.load;Jt.PoseNet=$L.PoseNet;Jt.MobileNet=B7.MobileNet;Jt.decodeMultiplePoses=z7.decodeMultiplePoses;Jt.partChannels=Zo.partChannels;Jt.partIds=Zo.partIds;Jt.partNames=Zo.partNames;Jt.poseChain=Zo.poseChain;Jt.getAdjacentKeyPoints=cs.getAdjacentKeyPoints;Jt.getBoundingBox=cs.getBoundingBox;Jt.getBoundingBoxPoints=cs.getBoundingBoxPoints;Jt.scaleAndFlipPoses=cs.scaleAndFlipPoses;Jt.scalePose=cs.scalePose});var Ap=be(Ei=>{const P7=Wt();function Lp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Ei.getBoxSize=Lp;function Ip(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Ei.getBoxCenter=Ip;function _7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return P7.image.cropAndResize(t,a,[0],e)}Ei.cutBoxFromImageAndResize=_7;function M7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}Ei.scaleBoxCoordinates=M7;function H7(n,t=1.5){const e=Ip(n),i=Lp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.enlargeBox=H7;function V7(n){const t=Ip(n),e=Lp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}Ei.squarifyBox=V7;function q7(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}Ei.shiftBox=q7});var ZL=be(JL=>{const qe=Wt(),G7=Ap();class Y7{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=qe.tensor2d(this.anchors),this.inputSizeTensor=qe.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=qe.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return qe.tidy(()=>{const t=qe.slice(n,[0,0],[-1,2]),e=qe.slice(n,[0,2],[-1,2]),i=qe.add(qe.div(t,this.inputSizeTensor),this.anchorsTensor),r=qe.div(e,this.doubleInputSizeTensor),a=qe.mul(qe.sub(i,r),this.inputSizeTensor),s=qe.mul(qe.add(i,r),this.inputSizeTensor);return qe.concat2d([a,s],1)})}normalizeLandmarks(n,t){return qe.tidy(()=>{const e=qe.add(qe.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return qe.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=qe.tidy(()=>qe.mul(qe.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=qe.tidy(()=>qe.sigmoid(qe.slice(i,[0,0],[-1,1])).squeeze()),a=qe.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await qe.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=qe.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=qe.slice(s,[p,0],[1,-1]),m=qe.slice(i,[p,5],[1,14]),g=qe.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=qe.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(G7.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}JL.HandDetector=Y7});var eI=be(QL=>{QL.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var aI=be(Di=>{function tI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Di.normalizeRadians=tI;function K7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return tI(e)}Di.computeRotation=K7;const nI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function ha(n,t){let e=0;for(let i=0;i<n.length;i++)e+=n[i]*t[i];return e}Di.dot=ha;function iI(n,t){const e=[];for(let i=0;i<n.length;i++)e.push(n[i][t]);return e}Di.getColumnFrom2DArr=iI;function rI(n,t){const e=[],i=n.length;for(let r=0;r<i;r++){e.push([]);for(let a=0;a<i;a++)e[r].push(ha(n[r],iI(t,a)))}return e}function j7(n,t){const e=Math.cos(n),i=Math.sin(n),r=[[e,-i,0],[i,e,0],[0,0,1]],a=nI(t[0],t[1]),s=rI(a,r),o=nI(-t[0],-t[1]);return rI(s,o)}Di.buildRotationMatrix=j7;function $7(n){const t=[[n[0][0],n[1][0]],[n[0][1],n[1][1]]],e=[n[0][2],n[1][2]],i=[-ha(t[0],e),-ha(t[1],e)];return[t[0].concat(i[0]),t[1].concat(i[1]),[0,0,1]]}Di.invertTransformMatrix=$7;function X7(n,t){return[ha(n,t[0]),ha(n,t[1])]}Di.rotatePoint=X7});var uI=be(sI=>{const oI=Wt(),Hn=Ap(),ki=aI(),J7=.8,Z7=[0,-.4],Q7=[0,-.1],eH=1.65,lI=[0,5,9,13,17,1,2],tH=0,nH=2;class iH{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return ki.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(i,Z7)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=Hn.enlargeBox(Hn.squarifyBox(Hn.shiftBox(t,Q7)),eH),i=[];for(let r=0;r<lI.length;r++)i.push(n[lI[r]].slice(0,2));return e.palmLandmarks=i,e}transformRawCoords(n,t,e,i){const r=Hn.getBoxSize(t),a=[r[0]/this.meshWidth,r[1]/this.meshHeight],s=n.map(d=>[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=ki.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=ki.rotatePoint(d,o);return[...p,d[2]]}),u=ki.invertTransformMatrix(i),c=[...Hn.getBoxCenter(t),1],h=[ki.dot(c,u[0]),ki.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=ki.computeRotation(a.palmLandmarks[tH],a.palmLandmarks[nH]),o=Hn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=oI.image.rotateWithOffset(n,s,0,l),c=ki.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=Hn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),v<t.minConfidence)return g.dispose(),this.regionsOfInterest[r]=[],i;const b=oI.reshape(g,[-1,3]),w=await b.array();g.dispose(),b.dispose();const S=this.transformRawCoords(w,h,s,c),L=this.getBoxForHandLandmarks(S);this.updateRegionsOfInterest(L,!1,r);const x={landmarks:S,confidence:v,box:{topLeft:L.startPoint,bottomRight:L.endPoint}};i.push(x)}return i}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>J7?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}sI.HandPipeline=iH});var dI=be(Tp=>{const vr=Wt(),rH=ZL(),cI=eI(),aH=uI();class hI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=vr.tidy(()=>(n instanceof vr.Tensor||(n=vr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(cI.MESH_ANNOTATIONS))s[o]=cI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}Tp.HandPose=hI;async function sH(n){if(vr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return vr.util.fetch(n).then(t=>t.json())}async function oH(n){const[t,e,i]=await Promise.all([sH(n.detector.anchors),vr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),vr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new rH.HandDetector(e,t,n),a=new aH.HandPipeline(r,i,n),s=new hI(a);return s}Tp.load=oH});var pI=be(lH=>{xp(lH,{default:()=>uH});var uH={backend:"webgl",console:!0,face:{enabled:!0,detector:{modelPath:"../models/blazeface/back/model.json",inputSize:256,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,minConfidence:.8,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:10,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:10,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});var mI=be((oV,fI)=>{fI.exports={name:"@vladmandic/human",version:"0.3.3",description:"human: 3D Face Detection, Iris Tracking and Age & Gender Prediction",sideEffects:!1,main:"dist/human-nobundle.cjs",module:"dist/human.esm.js",browser:"dist/human.esm.js",author:"Vladimir Mandic <mandic00@live.com>",bugs:{url:"https://github.com/vladmandic/human/issues"},homepage:"https://github.com/vladmandic/human#readme",license:"MIT",engines:{node:">=14.0.0"},repository:{type:"git",url:"git+https://github.com/vladmandic/human.git"},dependencies:{},peerDependencies:{},devDependencies:{"@tensorflow/tfjs":"^2.6.0","@tensorflow/tfjs-node":"^2.6.0",esbuild:"^0.7.15",eslint:"^7.10.0","eslint-config-airbnb-base":"^14.2.0","eslint-plugin-import":"^2.22.1","eslint-plugin-json":"^2.1.2","eslint-plugin-node":"^11.1.0","eslint-plugin-promise":"^4.2.1",rimraf:"^3.0.2"},scripts:{start:"node --trace-warnings --trace-uncaught --no-deprecation demo/demo-node.js",lint:"eslint src/*.js demo/*.js","build-iife":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=iife --minify --external:fs --global-name=human --outfile=dist/human.js src/index.js","build-esm-bundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:fs --outfile=dist/human.esm.js src/index.js","build-esm-nobundle":"esbuild --bundle --platform=browser --sourcemap --target=esnext --format=esm --minify --external:@tensorflow --external:fs --outfile=dist/human.esm-nobundle.js src/index.js","build-node-bundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --minify --outfile=dist/human.cjs src/index.js","build-node-nobundle":"esbuild --bundle --platform=node --sourcemap --target=esnext --format=cjs --external:@tensorflow --outfile=dist/human-nobundle.cjs src/index.js",build:"rimraf dist/* && npm run build-iife && npm run build-esm-bundle && npm run build-esm-nobundle && npm run build-node-bundle && npm run build-node-nobundle && ls -l dist/",update:"npm update --depth 20 && npm dedupe && npm prune && npm audit"},keywords:["tensorflowjs","face-detection","face-geometry","body-tracking","hand-tracking","iris-tracking","age-estimation","emotion-detection","gender-prediction","gesture-recognition"]}});var LI=be(Tn=>{const Nn=Wt(),gI=vL(),Qo=bL(),vI=LL(),yI=XL(),bI=dI(),wI=pI().default,cH=mI();let Pe;const Et={facemesh:null,posenet:null,handpose:null,iris:null,age:null,gender:null,emotion:null},Vn=()=>typeof performance!="undefined"?performance.now():parseInt(Number(process.hrtime.bigint())/1e3/1e3),el=(...n)=>{Pe.console&&console.log(...n)};function SI(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=SI(a,s):e[r]=s}),e),{})}function hH(n){if(!n)return"input is not defined";const t=n.naturalWidth||n.videoWidth||n.width||n.shape&&n.shape[1]>0;if(!t||t===0)return"input is empty";if(n.readyState&&n.readyState<=2)return"input is not ready";try{Nn.getBackend()}catch{return"backend not loaded"}return null}async function dH(n,t){Pe=SI(wI,t);const e=hH(n);return e?(el(e,n),{error:e}):new Promise(async i=>{const r=Object.values(Et).filter(c=>c).length;r===0&&el("Human library starting"),Nn.getBackend()!==Pe.backend&&(el("Human library setting backend:",Pe.backend),await Nn.setBackend(Pe.backend),await Nn.ready()),Pe.face.enabled&&!Et.facemesh&&(Et.facemesh=await gI.load(Pe.face)),Pe.body.enabled&&!Et.posenet&&(Et.posenet=await yI.load(Pe.body)),Pe.hand.enabled&&!Et.handpose&&(Et.handpose=await bI.load(Pe.hand)),Pe.face.enabled&&Pe.face.age.enabled&&!Et.age&&(Et.age=await Qo.loadAge(Pe)),Pe.face.enabled&&Pe.face.gender.enabled&&!Et.gender&&(Et.gender=await Qo.loadGender(Pe)),Pe.face.enabled&&Pe.face.emotion.enabled&&!Et.emotion&&(Et.emotion=await vI.load(Pe));const a={};let s;s=Vn(),Nn.engine().startScope();const o=Pe.body.enabled?await Et.posenet.estimatePoses(n,Pe.body):[];Nn.engine().endScope(),a.body=Math.trunc(Vn()-s),s=Vn(),Nn.engine().startScope();const l=Pe.hand.enabled?await Et.handpose.estimateHands(n,Pe.hand):[];Nn.engine().endScope(),a.hand=Math.trunc(Vn()-s);const u=[];if(Pe.face.enabled){s=Vn(),Nn.engine().startScope();const c=await Et.facemesh.estimateFaces(n,Pe.face);a.face=Math.trunc(Vn()-s);for(const h of c){if(!h.image||h.image.isDisposedInternal){el("face object is disposed:",h.image);continue}s=Vn();const d=Pe.face.age.enabled||Pe.face.gender.enabled?await Qo.predict(h.image,Pe):{};a.agegender=Math.trunc(Vn()-s),s=Vn();const p=Pe.face.emotion.enabled?await vI.predict(h.image,Pe):{};a.emotion=Math.trunc(Vn()-s),h.image.dispose();const f=h.annotations.leftEyeIris&&h.annotations.rightEyeIris?Math.max(h.annotations.leftEyeIris[3][0]-h.annotations.leftEyeIris[1][0],h.annotations.rightEyeIris[3][0]-h.annotations.rightEyeIris[1][0]):0;u.push({confidence:h.confidence,box:h.box,mesh:h.mesh,annotations:h.annotations,age:d.age,gender:d.gender,agConfidence:d.confidence,emotion:p,iris:f!==0?Math.trunc(100*11.7/f)/100:0})}Nn.engine().endScope()}a.total=Object.values(a).reduce((c,h)=>c+h),i({face:u,body:o,hand:l,performance:a})})}Tn.detect=dH;Tn.defaults=wI;Tn.config=Pe;Tn.models=Et;Tn.facemesh=gI;Tn.ssrnet=Qo;Tn.posenet=yI;Tn.handpose=bI;Tn.tf=Nn;Tn.version=cH.version});return LI();})();
 /*! *****************************************************************************
 Copyright (c) Microsoft Corporation. All rights reserved.
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use
diff --git a/dist/human.js.map b/dist/human.js.map
index 6ee75b86..ca72ead2 100644
--- a/dist/human.js.map
+++ b/dist/human.js.map
@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["empty:/home/vlado/dev/human/node_modules/node-fetch/browser.js", "empty:util", "empty:crypto", "../node_modules/@tensorflow/tfjs-core/src/backends/backend.ts", "../node_modules/@tensorflow/tfjs-core/src/environment.ts", "../node_modules/@tensorflow/tfjs-core/src/global_util.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_names.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/util.ts", "../node_modules/@tensorflow/tfjs-core/src/profiler.ts", "../node_modules/@tensorflow/tfjs-core/src/tape.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_format.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/types.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util.ts", "../node_modules/@tensorflow/tfjs-core/src/engine.ts", "../node_modules/@tensorflow/tfjs-core/src/device_util.ts", "../node_modules/@tensorflow/tfjs-core/src/flags.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util_env.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/operation.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/complex.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/io/types.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/io/router_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/io/indexed_db.ts", "../node_modules/@tensorflow/tfjs-core/src/io/local_storage.ts", "../node_modules/@tensorflow/tfjs-core/src/io/model_management.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_browser.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_node.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/buffer.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clone.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/print.ts", "../node_modules/@tensorflow/tfjs-core/src/base_side_effects.ts", "../node_modules/@tensorflow/tfjs-core/src/io/browser_files.ts", "../node_modules/@tensorflow/tfjs-core/src/io/progress.ts", "../node_modules/@tensorflow/tfjs-core/src/io/weights_loader.ts", "../node_modules/@tensorflow/tfjs-core/src/io/http.ts", "../node_modules/@tensorflow/tfjs-core/src/io/passthrough.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/confusion_matrix.ts", "../node_modules/@tensorflow/tfjs-core/src/math.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/browser.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice_util.ts", "../node_modules/@tensorflow/tfjs-core/src/serialization.ts", "../node_modules/@tensorflow/tfjs-core/src/test_util.ts", "../node_modules/@tensorflow/tfjs-core/src/version.ts", "../node_modules/@tensorflow/tfjs-core/src/globals.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add_n.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/axis_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/basic_lstm_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/diag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/eye.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fill.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reduce_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/segment_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/imag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linspace.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_with_argmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moments.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multi_rnn_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multinomial.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/real.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/outer_product.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rand.ts", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-core/src/ops/rand_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_gamma.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_uniform.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/range.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/setdiff1d_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor5d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor6d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/truncated_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/variable.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/where_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/boolean_mask.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/compare.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/binary_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moving_average.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/in_top_k.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hamming_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hann_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/frame.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/stft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/crop_and_resize.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/flip_left_right.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/rotate_with_offset.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/nonmax_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/array_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/non_max_suppression_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/band_part.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/gram_schmidt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/qr.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/loss_ops_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/compute_weighted_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/absolute_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/cosine_distance.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/hinge_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/huber_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/log_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/mean_squared_error.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/sigmoid_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/softmax_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ops.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adadelta_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adagrad_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adam_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adamax_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/sgd_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/momentum_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/rmsprop_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer_constructors.ts", "../node_modules/@tensorflow/tfjs-core/src/train.ts", "../node_modules/@tensorflow/tfjs-core/src/browser_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rotate_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/array_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf_util.ts", "../node_modules/@tensorflow/tfjs-core/src/log.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/complex_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/backend_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/split_shared.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/tile_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/topk_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/kernel_impls.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Abs_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Add_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AddN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchMatMul_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchToSpaceND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BroadcastTo_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cast_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Ceil_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ClipByValue_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Concat_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2DBackpropInput_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cumsum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/DepthwiseConv2dNative_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Dilation2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Div_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Elu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Erf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Exp_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Expm1_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Floor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FloorDiv_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FusedBatchNorm_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GatherV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GreaterEqual_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Identity_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsFinite_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsInf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsNan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log1p_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LogSoftmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LRN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/min_max_grad_util.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Max_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Maximum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Min_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Minimum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Mod_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Multiply_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Negate_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OneHot_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OnesLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/PadV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Pow_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Prelu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reciprocal_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu6_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reshape_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeBilinear_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeNearestNeighbor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reverse_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Round_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Rsqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SelectV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Selu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sigmoid_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sign_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Slice_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softplus_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SpaceToBatchND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SplitV_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Square_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SquaredDifference_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Step_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sub_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tile_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Transpose_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Unpack_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/UnsortedSegmentSum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ZerosLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/register_all_gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_type.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as3d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as4d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as5d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2D_deprecated.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/flatten.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape_as.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_bool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_float.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_int.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/errors.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/generic_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/math_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/tfjs_backend.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/initializer_config.ts", "../node_modules/@tensorflow/tfjs-layers/src/initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/state.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/types_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/variable_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/variables.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/topology.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/input_layer.ts", "../node_modules/@tensorflow/tfjs-layers/src/logs.ts", "../node_modules/@tensorflow/tfjs-layers/src/base_callbacks.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/serialization.ts", "../node_modules/@tensorflow/tfjs-layers/src/losses.ts", "../node_modules/@tensorflow/tfjs-layers/src/metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/optimizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/user_defined_metadata.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/layer_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/serialization_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/version.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/executor.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/container.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_dataset.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_tensors.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training.ts", "../node_modules/@tensorflow/tfjs-layers/src/models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports.ts", "../node_modules/@tensorflow/tfjs-layers/src/activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/advanced_activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/conv_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_depthwise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/core.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/embeddings.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/merge.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/noise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/normalization.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/padding.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/pooling.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/wrappers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_layers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/callbacks.ts", "../node_modules/@tensorflow/tfjs-converter/src/data/compiled_api.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/register.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/arithmetic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/basic_math.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/control.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/convolution.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/creation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/dynamic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/evaluation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/graph.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/image.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/logical.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/matrices.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/normalization.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/reduction.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/slice_join.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/spectral.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/transformation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_mapper.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/node_value_impl.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/arithmetic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/basic_math_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_array.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_list.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/control_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/convolution_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/creation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/dynamic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/evaluation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/image_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/logical_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/matrices_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/normalization_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/reduction_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/slice_join_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/spectral_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/transformation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/execution_context.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/model_analysis.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_model.ts", "../node_modules/@tensorflow/tfjs-converter/src/version.ts", "empty:/home/vlado/dev/human/node_modules/string_decoder/lib/string_decoder.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-data/src/util/deep_map.ts", "../node_modules/@tensorflow/tfjs-data/src/util/deep_clone.ts", "../node_modules/@tensorflow/tfjs-data/src/util/ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/util/growing_ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/lazy_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/text_line_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/csv_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/microphone_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/webcam_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/datasource.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/string_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/byte_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/file_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/url_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/util/source_util.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/file_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/url_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/readers.ts", "../node_modules/@tensorflow/tfjs-data/src/version.ts", "../node_modules/seedrandom/lib/alea.js", "../node_modules/seedrandom/lib/xor128.js", "../node_modules/seedrandom/lib/xorwow.js", "../node_modules/seedrandom/lib/xorshift7.js", "../node_modules/seedrandom/lib/xor4096.js", "../node_modules/seedrandom/lib/tychei.js", "../node_modules/seedrandom/seedrandom.js", "../node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-backend-cpu/src/cpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/backend_cpu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Abs.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/binary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Complex.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Real.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cast.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/kernel_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Add.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Ceil.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Exp.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Expm1.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Floor.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Multiply.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Rsqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Slice.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sub.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/shared.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/pool_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Clip.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Imag.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Concat.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2D.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropFilter.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropInput.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Elu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Erf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/fft_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IFFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsFinite.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsInf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsNaN.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log1p.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/LogicalNot.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NotEqual.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/PadV2.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reciprocal.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Round.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Selu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sigmoid.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sign.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Softplus.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SpaceToBatchND.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Step.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/canvas_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tex_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flags_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Abs.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/binary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Add.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/unary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Ceil.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Exp.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Expm1.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Floor.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Log.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Max_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Multiply.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Rsqrt.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Slice.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Sub.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Transpose_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Unique_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/shared.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/packing_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/glsl_version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/avg_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_complex_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/complex_abs_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_packed_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/crop_and_resize_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/cumsum_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/depth_to_space_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/diag_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fft_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fill_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_nd_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_context.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_math.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/im2col_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_grad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/max_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/mulmat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/multinomial_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/onehot_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pool_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reduce_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reshape_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/scatter_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/segment_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/select_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/strided_slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/texture_manager.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tile_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unpack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/backend_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/kernel_funcs_utils.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Atan2.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flip_left_right_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reduce.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV3.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/rotate_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs/src/version.ts", "../node_modules/@tensorflow/tfjs/src/index.ts", "../src/facemesh/blazeface.js", "../src/facemesh/keypoints.js", "../src/facemesh/box.js", "../src/facemesh/util.js", "../src/facemesh/pipeline.js", "../src/facemesh/uvcoords.js", "../src/facemesh/triangulation.js", "../src/facemesh/facemesh.js", "../src/ssrnet/ssrnet.js", "../src/emotion/emotion.js", "../src/posenet/modelBase.js", "../src/posenet/modelMobileNet.js", "../src/posenet/heapSort.js", "../src/posenet/buildParts.js", "../src/posenet/keypoints.js", "../src/posenet/vectors.js", "../src/posenet/decodePose.js", "../src/posenet/decodeMultiple.js", "../src/posenet/util.js", "../src/posenet/modelPoseNet.js", "../src/posenet/posenet.js", "../src/handpose/box.js", "../src/handpose/handdetector.js", "../src/handpose/keypoints.js", "../src/handpose/util.js", "../src/handpose/pipeline.js", "../src/handpose/handpose.js", "../src/config.js", "../src/index.js"],
-  "sourcesContent": ["", "", "", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n  kernelMs: number|{error: string};\n  getExtraProfileInfo?(): string;  // a field for additional timing information\n                                   // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n  disposeData(dataId: DataId): void;\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void;\n  memory(): {unreliable: boolean;};  // Backend-specific information.\n  /** Returns number of data ids currently in the storage. */\n  numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage<T> {\n  private data = new WeakMap<DataId, T>();\n  private dataIdsCount = 0;\n\n  constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n  get(dataId: DataId) {\n    if (!this.data.has(dataId)) {\n      this.dataMover.moveData(this.backend, dataId);\n    }\n    return this.data.get(dataId);\n  }\n\n  set(dataId: DataId, value: T): void {\n    this.dataIdsCount++;\n    this.data.set(dataId, value);\n  }\n\n  has(dataId: DataId): boolean {\n    return this.data.has(dataId);\n  }\n\n  delete(dataId: DataId): boolean {\n    this.dataIdsCount--;\n    return this.data.delete(dataId);\n  }\n\n  numDataIds(): number {\n    return this.dataIdsCount;\n  }\n}\n\nexport interface DataMover {\n  /**\n   * To be called by backends whenever they see a dataId that they don't own.\n   * Upon calling this method, the mover will fetch the tensor from another\n   * backend and register it with the current active backend.\n   */\n  moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo>;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo> {\n    return notYetImplemented('time');\n  }\n  read(dataId: object): Promise<BackendValues> {\n    return notYetImplemented('read');\n  }\n  readSync(dataId: object): BackendValues {\n    return notYetImplemented('readSync');\n  }\n  numDataIds(): number {\n    return notYetImplemented('numDataIds');\n  }\n  disposeData(dataId: object): void {\n    return notYetImplemented('disposeData');\n  }\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    return notYetImplemented('write');\n  }\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    return notYetImplemented('move');\n  }\n  memory(): {unreliable: boolean; reasons?: string[]} {\n    return notYetImplemented('memory');\n  }\n  /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n  floatPrecision(): 16|32 {\n    return notYetImplemented('floatPrecision');\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    return notYetImplemented('batchMatMul');\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          FusedBatchMatMulConfig): Tensor3D {\n    return notYetImplemented('fusedBatchMatMul');\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    return notYetImplemented('slice');\n  }\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    return notYetImplemented('stridedSlice');\n  }\n  unstack(x: Tensor, axis: number): Tensor[] {\n    return notYetImplemented('unstack');\n  }\n  reverse<T extends Tensor>(a: T, axis: number[]): T {\n    return notYetImplemented('reverse');\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    return notYetImplemented('concat');\n  }\n\n  neg<T extends Tensor>(a: T): T {\n    return notYetImplemented('neg');\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('add');\n  }\n  addN<T extends Tensor>(tensors: T[]): T {\n    return notYetImplemented('addN');\n  }\n  subtract(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('subtract');\n  }\n  multiply(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('multiply');\n  }\n  realDivide(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('realDivide');\n  }\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('floorDiv');\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('sum');\n  }\n  prod(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('prod');\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    return notYetImplemented('unsortedSegmentSum');\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMin');\n  }\n  argMax(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMax');\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('equal');\n  }\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('notEqual');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('less');\n  }\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('lessEqual');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greater');\n  }\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greaterEqual');\n  }\n\n  logicalNot<T extends Tensor>(a: T): T {\n    return notYetImplemented('logicalNot');\n  }\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalAnd');\n  }\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalOr');\n  }\n\n  where(condition: Tensor): Tensor2D {\n    return notYetImplemented('where');\n  }\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('select');\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    return notYetImplemented('topk');\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('min');\n  }\n  minimum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('minimum');\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('mod');\n  }\n\n  max(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('max');\n  }\n  maximum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('maximum');\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('all');\n  }\n  any(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('any');\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('squaredDifference');\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    return notYetImplemented('ceil');\n  }\n  floor<T extends Tensor>(x: T): T {\n    return notYetImplemented('floor');\n  }\n  round<T extends Tensor>(x: T): T {\n    return notYetImplemented('round');\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    return notYetImplemented('sign');\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    return notYetImplemented('isNaN');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    return notYetImplemented('isInf');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    return notYetImplemented('isFinite');\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    return notYetImplemented('pow');\n  }\n  exp<T extends Tensor>(x: T): T {\n    return notYetImplemented('exp');\n  }\n  expm1<T extends Tensor>(x: T): T {\n    return notYetImplemented('expm1');\n  }\n  softmax<T extends Tensor>(x: T, dim: number): T {\n    return notYetImplemented('softmax');\n  }\n  log<T extends Tensor>(x: T): T {\n    return notYetImplemented('log');\n  }\n  log1p<T extends Tensor>(x: T): T {\n    return notYetImplemented('log1p');\n  }\n  sqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('sqrt');\n  }\n  rsqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('rsqrt');\n  }\n  square<T extends Tensor>(x: T): T {\n    return notYetImplemented('square');\n  }\n  reciprocal<T extends Tensor>(x: T): T {\n    return notYetImplemented('reciprocal');\n  }\n  relu<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu');\n  }\n  relu6<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu6');\n  }\n  prelu<T extends Tensor>(x: T, a: T): T {\n    return notYetImplemented('prelu');\n  }\n  elu<T extends Tensor>(x: T): T {\n    return notYetImplemented('elu');\n  }\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    return notYetImplemented('eluDer');\n  }\n  selu<T extends Tensor>(x: T): T {\n    return notYetImplemented('selu');\n  }\n  int<T extends Tensor>(x: T): T {\n    return notYetImplemented('int');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    return notYetImplemented('clip');\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    return notYetImplemented('abs');\n  }\n  complexAbs<T extends Tensor>(x: T): T {\n    return notYetImplemented('complexAbs');\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    return notYetImplemented('sigmoid');\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    return notYetImplemented('softplus');\n  }\n\n  sin<T extends Tensor>(x: T): T {\n    return notYetImplemented('sin');\n  }\n  cos<T extends Tensor>(x: T): T {\n    return notYetImplemented('cos');\n  }\n  tan<T extends Tensor>(x: T): T {\n    return notYetImplemented('tan');\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    return notYetImplemented('asin');\n  }\n  acos<T extends Tensor>(x: T): T {\n    return notYetImplemented('acos');\n  }\n  atan<T extends Tensor>(x: T): T {\n    return notYetImplemented('atan');\n  }\n  atan2<T extends Tensor>(a: T, b: T): T {\n    return notYetImplemented('atan2');\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('sinh');\n  }\n  cosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('cosh');\n  }\n  tanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('tanh');\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('asinh');\n  }\n  acosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('acosh');\n  }\n  atanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('atanh');\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    return notYetImplemented('erf');\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    return notYetImplemented('step');\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedConv2d');\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2d');\n  }\n  conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('conv2dDerInput');\n  }\n  conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2dDerFilter');\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedDepthwiseConv2D');\n  }\n\n  depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2D');\n  }\n  depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerInput');\n  }\n  depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerFilter');\n  }\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3d');\n  }\n  conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n      Tensor5D {\n    return notYetImplemented('conv3dDerInput');\n  }\n  conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3dDerFilter');\n  }\n  maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('maxPool');\n  }\n  maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('maxPoolBackprop');\n  }\n  avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPool');\n  }\n  avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPoolBackprop');\n  }\n  avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3d');\n  }\n  avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3dBackprop');\n  }\n  maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3d');\n  }\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3dBackprop');\n  }\n\n  reshape<T extends Tensor, R extends Rank>(x: T, shape: ShapeMap[R]):\n      Tensor<R> {\n    return notYetImplemented('reshape');\n  }\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return notYetImplemented('cast');\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    return notYetImplemented('tile');\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    return notYetImplemented('pad');\n  }\n\n  transpose<T extends Tensor>(x: T, perm: number[]): T {\n    return notYetImplemented('transpose');\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    return notYetImplemented('gather');\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    return notYetImplemented('gatherND');\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    return notYetImplemented('scatterND');\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    return notYetImplemented('batchToSpaceND');\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: number[][]): T {\n    return notYetImplemented('spaceToBatchND');\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeBilinear');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    return notYetImplemented('resizeBilinearBackprop');\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHEight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighbor');\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighborBackprop');\n  }\n\n  batchNorm(\n      x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n      offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n      varianceEpsilon?: number): Tensor4D {\n    return notYetImplemented('batchNorm');\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    return notYetImplemented('localResponseNormalization4D');\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n      bias: number, alpha: number, beta: number): Tensor4D {\n    return notYetImplemented('LRNGrad');\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    return notYetImplemented('multinomial');\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    return notYetImplemented('oneHot');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    return notYetImplemented('cumsum');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold?: number): Tensor1D {\n    return notYetImplemented('nonMaxSuppression');\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('fft');\n  }\n  ifft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('ifft');\n  }\n  complex<T extends Tensor>(real: T, imag: T): T {\n    return notYetImplemented('complex');\n  }\n  real<T extends Tensor>(input: T): T {\n    return notYetImplemented('real');\n  }\n  imag<T extends Tensor>(input: T): T {\n    return notYetImplemented('imag');\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    return notYetImplemented('cropAndResize');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n    return notYetImplemented('depthToSpace');\n  }\n\n  // Aligns with the \"SplitV\" kernel in TensorFlow.\n  split<T extends Tensor>(value: T, sizeSplits: number[], axis: number): T[] {\n    return notYetImplemented('split');\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    return notYetImplemented('sparseToDense');\n  }\n\n  diag(x: Tensor): Tensor {\n    return notYetImplemented('diag');\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    return notYetImplemented('fill');\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('onesLike');\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('zerosLike');\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return notYetImplemented('linspace');\n  }\n\n  dispose(): void {\n    return notYetImplemented('dispose');\n  }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n  throw new Error(\n      `'${kernelName}' not yet implemented or not found in the registry. ` +\n      `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise<FlagValue>);\nexport type Flags = {\n  [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n  evaluationFn: FlagEvaluationFn;\n  setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n  private flags: Flags = {};\n  private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n  private urlFlags: Flags = {};\n\n  platformName: string;\n  platform: Platform;\n\n  // tslint:disable-next-line: no-any\n  constructor(public global: any) {\n    this.populateURLFlags();\n  }\n\n  setPlatform(platformName: string, platform: Platform) {\n    if (this.platform != null) {\n      console.warn(\n          `Platform ${this.platformName} has already been set. ` +\n          `Overwriting the platform with ${platform}.`);\n    }\n    this.platformName = platformName;\n    this.platform = platform;\n  }\n\n  registerFlag(\n      flagName: string, evaluationFn: FlagEvaluationFn,\n      setHook?: (value: FlagValue) => void) {\n    this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n    // Override the flag value from the URL. This has to happen here because the\n    // environment is initialized before flags get registered.\n    if (this.urlFlags[flagName] != null) {\n      const flagValue = this.urlFlags[flagName];\n      console.warn(\n          `Setting feature override from URL ${flagName}: ${flagValue}.`);\n      this.set(flagName, flagValue);\n    }\n  }\n\n  async getAsync(flagName: string): Promise<FlagValue> {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    this.flags[flagName] = await this.evaluateFlag(flagName);\n    return this.flags[flagName];\n  }\n\n  get(flagName: string): FlagValue {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    const flagValue = this.evaluateFlag(flagName);\n    if (flagValue instanceof Promise) {\n      throw new Error(\n          `Flag ${flagName} cannot be synchronously evaluated. ` +\n          `Please use getAsync() instead.`);\n    }\n\n    this.flags[flagName] = flagValue;\n\n    return this.flags[flagName];\n  }\n\n  getNumber(flagName: string): number {\n    return this.get(flagName) as number;\n  }\n\n  getBool(flagName: string): boolean {\n    return this.get(flagName) as boolean;\n  }\n\n  getFlags(): Flags {\n    return this.flags;\n  }\n  // For backwards compatibility.\n  get features(): Flags {\n    return this.flags;\n  }\n\n  set(flagName: string, value: FlagValue): void {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot set flag ${flagName} as it has not been registered.`);\n    }\n    this.flags[flagName] = value;\n    if (this.flagRegistry[flagName].setHook != null) {\n      this.flagRegistry[flagName].setHook(value);\n    }\n  }\n\n  private evaluateFlag(flagName: string): FlagValue|Promise<FlagValue> {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n    }\n    return this.flagRegistry[flagName].evaluationFn();\n  }\n\n  setFlags(flags: Flags) {\n    this.flags = Object.assign({}, flags);\n  }\n\n  reset() {\n    this.flags = {};\n    this.urlFlags = {};\n    this.populateURLFlags();\n  }\n\n  private populateURLFlags(): void {\n    if (typeof this.global === 'undefined' ||\n        typeof this.global.location === 'undefined' ||\n        typeof this.global.location.search === 'undefined') {\n      return;\n    }\n\n    const urlParams = getQueryParams(this.global.location.search);\n    if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n      const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n      keyValues.forEach(keyValue => {\n        const [key, value] = keyValue.split(':') as [string, string];\n        this.urlFlags[key] = parseValue(key, value);\n      });\n    }\n  }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n  const params = {};\n  queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n    decodeParam(params, t[0], t[1]);\n    return t.join('=');\n  });\n  return params;\n}\n\nfunction decodeParam(\n    params: {[key: string]: string}, name: string, value?: string) {\n  params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n  value = value.toLowerCase();\n  if (value === 'true' || value === 'false') {\n    return value === 'true';\n  } else if (`${+ value}` === value) {\n    return +value;\n  }\n  throw new Error(\n      `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n  return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n  ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map<string, any>};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map<string, any>} {\n  if (globalNameSpace == null) {\n    // tslint:disable-next-line:no-any\n    let ns: any;\n    if (typeof (window) !== 'undefined') {\n      ns = window;\n    } else if (typeof (global) !== 'undefined') {\n      ns = global;\n    } else if (typeof (process) !== 'undefined') {\n      ns = process;\n    } else if (typeof (self) !== 'undefined') {\n      ns = self;\n    } else {\n      throw new Error('Could not find a global object');\n    }\n    globalNameSpace = ns;\n  }\n  return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map<string, any> {\n  const ns = getGlobalNamespace();\n  if (ns._tfGlobals == null) {\n    ns._tfGlobals = new Map();\n  }\n  return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n *             the first time it is fetched.\n */\nexport function getGlobal<T>(key: string, init: () => T): T {\n  const globalMap = getGlobalMap();\n  if (globalMap.has(key)) {\n    return globalMap.get(key);\n  } else {\n    const singleton = init();\n    globalMap.set(key, singleton);\n    return globalMap.get(key);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AllAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AnyAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMaxAttrs {\n  axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMinAttrs {\n  axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\nexport interface BatchMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BatchToSpaceNDAttrs {\n  blockShape: number[];\n  crops: number[][];\n}\n\nexport type BinaryInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BroadCastToAttrs {\n  shape: number[];\n  inputShape: number[];  // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n  dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n  clipValueMin: number;\n  clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick<NamedTensorInfoMap, 'real'|'imag'>;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n  axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'dy'>;\nexport interface Conv2DBackpropFilterAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'|'filter'>;\nexport interface Conv2DBackpropInputAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv3DAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'y'>;\n\nexport interface Conv3DBackpropFilterAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'>;\nexport interface Conv3DBackpropInputAttrs {\n  pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface CumsumAttrs {\n  axis: number;\n  exclusive: boolean;\n  reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n    Pick<NamedTensorInfoMap, 'image'|'boxes'|'boxInd'>;\nexport interface CropAndResizeAttrs {\n  cropSize: [number, number];\n  method: 'bilinear'|'nearest';\n  extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface DepthToSpaceAttrs {\n  blockSize: number;\n  dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface DepthwiseConv2dNativeAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n    'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'dy'>;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n    'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'dy'>;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Dilation2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick<NamedTensorInfoMap, 'dy'|'y'>;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n  shape: number[];\n  value: number|string;\n  dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick<NamedTensorInfoMap, 'image'>;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n    Pick<NamedTensorInfoMap, 'x'|'scale'|'offset'|'mean'|'variance'>;\nexport interface FusedBatchNormAttrs {\n  varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick<NamedTensorInfoMap, 'x'|'indices'>;\nexport interface GatherV2Attrs {\n  axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick<NamedTensorInfoMap, 'params'|'indices'>;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n  start: number;\n  stop: number;\n  num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface LogSoftmaxAttrs {\n  axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface LRNAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick<NamedTensorInfoMap, 'x'|'y'|'dy'>;\nexport interface LRNBackpropAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxAttrs {\n  reductionIndices: number|number[];\n  keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolWithArgmaxAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MeanAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MinAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV3Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV4Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV5Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick<NamedTensorInfoMap, 'indices'>;\nexport interface OneHotAttrs {\n  depth: number;\n  onValue: number;\n  offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface PadV2Attrs {\n  paddings: Array<[number, number]>;\n  constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick<NamedTensorInfoMap, 'x'|'alpha'>;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ProdAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n  start: number;\n  stop: number;\n  step: number;\n  dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReshapeAttrs {\n  shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeNearestNeighborAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n    Pick<NamedTensorInfoMap, 'images'>;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeBilinearAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick<NamedTensorInfoMap, 'images'>;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReverseAttrs {\n  dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick<NamedTensorInfoMap, 'indices'|'updates'>;\nexport interface ScatterNdAttrs {\n  shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick<NamedTensorInfoMap, 'condition'|'t'|'e'>;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SliceAttrs {\n  begin: number|number[];\n  size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SumAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SpaceToBatchNDAttrs {\n  blockShape: number[];\n  paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SplitVAttrs {\n  numOrSizeSplits: number[]|number;\n  axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface SoftmaxAttrs {\n  dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n    Pick<NamedTensorInfoMap, 'sparseIndices'|'sparseValues'|'defaultValue'>;\nexport interface SparseToDenseAttrs {\n  outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface StridedSliceAttrs {\n  begin: number[];\n  end: number[];\n  strides: number[];\n  beginMask: number;\n  endMask: number;\n  ellipsisMask: number;\n  newAxisMask: number;\n  shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TileAttrs {\n  reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TopKAttrs {\n  k: number;\n  sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TransposeAttrs {\n  perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface UniqueAttrs {\n  axis: number;\n}\n\nexport type UnaryInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick<NamedTensorInfoMap, 'value'>;\nexport interface UnpackAttrs {\n  axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n    Pick<NamedTensorInfoMap, 'x'|'segmentIds'>;\nexport interface UnsortedSegmentSumAttrs {\n  numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n  alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n  pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n      HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n  numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick<NamedTensorInfoMap, 'image'>;\nexport interface RotateWithOffsetAttrs {\n  radians: number;\n  fillValue: number|[number, number, number];\n  center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n  a: TensorInfo;\n  b: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n  activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n    getGlobal('kernelRegistry', () => new Map<string, KernelConfig>());\nconst gradRegistry =\n    getGlobal('gradRegistry', () => new Map<string, GradConfig>());\n\nexport type DataId = object;\n\ntype AttributeValue =\n    number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray<AttributeValue>;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n  inputs: NamedTensorInfoMap,\n  backend: {},\n  attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n    (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n        NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n  kernelName: string;\n  backendName: string;\n  kernelFunc: KernelFunc;\n  setupFunc?: KernelSetupFunc;\n  disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n  kernelName: string;\n  inputsToSave?: string[];\n  // When saveAllInputs is true, all inputs will be saved. Only use this flag\n  // if inputs is an array of Tensors.\n  saveAllInputs?: boolean;\n  outputsToSave?: boolean[];\n  gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n  dataId: DataId;\n  shape: number[];\n  dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n  [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n  [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n    kernelName: string, backendName: string): KernelConfig {\n  const key = makeKey(kernelName, backendName);\n  return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n  return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n  const it = kernelRegistry.entries();\n  const result: KernelConfig[] = [];\n\n  while (true) {\n    const {done, value} = it.next();\n    if (done) {\n      break;\n    }\n    const [key, config] = value;\n    const [backend, ] = key.split('_');\n    if (backend === backendName) {\n      result.push(config);\n    }\n  }\n  return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n  const {kernelName, backendName} = config;\n  const key = makeKey(kernelName, backendName);\n  if (kernelRegistry.has(key)) {\n    console.warn(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is already registered`);\n  }\n  kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n  const {kernelName} = config;\n\n  if (gradRegistry.has(kernelName)) {\n    // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n    // to debug mode.\n    if (env().getBool('DEBUG')) {\n      console.warn(`Overriding the gradient for '${kernelName}'`);\n    }\n  }\n  gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n    kernelName: string, backendName: string): void {\n  const key = makeKey(kernelName, backendName);\n  if (!kernelRegistry.has(key)) {\n    throw new Error(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is not registered`);\n  }\n  kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n  if (!gradRegistry.has(kernelName)) {\n    throw new Error(\n        `The gradient '${kernelName}' for backend is not registered`);\n  }\n  gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n    registeredBackendName: string, newBackendName: string): void {\n  const kernels = getKernelsForBackend(registeredBackendName);\n  kernels.forEach(kernelConfig => {\n    const newKernelConfig =\n        Object.assign({}, kernelConfig, {backendName: newBackendName});\n    registerKernel(newKernelConfig);\n  });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n  return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n                        Float32Array): void {\n  let counter = array.length;\n  let temp = 0;\n  let index = 0;\n  // While there are elements in the array\n  while (counter > 0) {\n    // Pick a random index\n    index = (Math.random() * counter) | 0;\n    // Decrease counter by 1\n    counter--;\n    // And swap the last element with it\n    temp = array[counter];\n    array[counter] = array[index];\n    array[index] = temp;\n  }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n  return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n  return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n  let sum = 0;\n  for (let i = 0; i < arr.length; i++) {\n    sum += arr[i];\n  }\n  return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n  const r = Math.random();\n  return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n  let result = 0;\n  for (let i = 0; i < a.length; i++) {\n    const diff = Number(a[i]) - Number(b[i]);\n    result += diff * diff;\n  }\n  return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n *     error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n  if (!expr) {\n    throw new Error(typeof msg === 'string' ? msg : msg());\n  }\n}\n\nexport function assertShapesMatch(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  assert(\n      arraysEqual(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n  assert(\n      a != null,\n      () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten<number>().\n/**\n *  Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n *  @param arr The nested array to flatten.\n *  @param result The destination array which holds the elements.\n *  @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n *      to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten<T extends number|boolean|string|Promise<number>|TypedArray>(\n    arr: T|RecursiveArray<T>, result: T[] = [], skipTypedArray = false): T[] {\n  if (result == null) {\n    result = [];\n  }\n  if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n    for (let i = 0; i < arr.length; ++i) {\n      flatten(arr[i], result, skipTypedArray);\n    }\n  } else {\n    result.push(arr as T);\n  }\n  return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n  if (shape.length === 0) {\n    // Scalar.\n    return 1;\n  }\n  let size = shape[0];\n  for (let i = 1; i < shape.length; i++) {\n    size *= shape[i];\n  }\n  return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n  return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n  if (n1 === n2) {\n    return true;\n  }\n  if (n1 == null || n2 == null) {\n    return false;\n  }\n\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function isInt(a: number): boolean {\n  return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n  // tslint:disable-next-line:no-any\n  if ((Math as any).tanh != null) {\n    // tslint:disable-next-line:no-any\n    return (Math as any).tanh(x);\n  }\n  if (x === Infinity) {\n    return 1;\n  } else if (x === -Infinity) {\n    return -1;\n  } else {\n    const e2x = Math.exp(2 * x);\n    return (e2x - 1) / (e2x + 1);\n  }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n  const width = Math.ceil(Math.sqrt(size));\n  return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n  const shuffledIndices = new Uint32Array(n);\n  for (let i = 0; i < n; ++i) {\n    shuffledIndices[i] = i;\n  }\n  shuffle(shuffledIndices);\n  return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n  if (size <= a.length) {\n    return a;\n  }\n  return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n    checkFn: () => boolean, delayFn = (counter: number) => 0,\n    maxCounter?: number): Promise<void> {\n  return new Promise<void>((resolve, reject) => {\n    let tryCount = 0;\n\n    const tryFn = () => {\n      if (checkFn()) {\n        resolve();\n        return;\n      }\n\n      tryCount++;\n\n      const nextBackoff = delayFn(tryCount);\n\n      if (maxCounter != null && tryCount >= maxCounter) {\n        reject();\n        return;\n      }\n      setTimeout(tryFn, nextBackoff);\n    };\n\n    tryFn();\n  });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n    shape: number[], size: number): number[] {\n  let shapeProd = 1;\n  let implicitIdx = -1;\n\n  for (let i = 0; i < shape.length; ++i) {\n    if (shape[i] >= 0) {\n      shapeProd *= shape[i];\n    } else if (shape[i] === -1) {\n      if (implicitIdx !== -1) {\n        throw Error(\n            `Shapes can only have 1 implicit size. ` +\n            `Found -1 at dim ${implicitIdx} and dim ${i}`);\n      }\n      implicitIdx = i;\n    } else if (shape[i] < 0) {\n      throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n    }\n  }\n\n  if (implicitIdx === -1) {\n    if (size > 0 && size !== shapeProd) {\n      throw Error(`Size(${size}) must match the product of shape ${shape}`);\n    }\n    return shape;\n  }\n\n  if (shapeProd === 0) {\n    throw Error(\n        `Cannot infer the missing size in [${shape}] when ` +\n        `there are 0 elements`);\n  }\n  if (size % shapeProd !== 0) {\n    throw Error(\n        `The implicit shape can't be a fractional number. ` +\n        `Got ${size} / ${shapeProd}`);\n  }\n\n  const newShape = shape.slice();\n  newShape[implicitIdx] = size / shapeProd;\n  return newShape;\n}\n\nexport function parseAxisParam(\n    axis: number|number[], shape: number[]): number[] {\n  const rank = shape.length;\n\n  // Normalize input\n  axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n  // Check for valid range\n  assert(\n      axis.every(ax => ax >= -rank && ax < rank),\n      () =>\n          `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n          `got axis ${axis}`);\n\n  // Check for only integers\n  assert(\n      axis.every(ax => isInt(ax)),\n      () => `All values in axis param must be integers but ` +\n          `got axis ${axis}`);\n\n  // Handle negative axis.\n  return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n    {newShape: number[], keptDims: number[]} {\n  const newShape: number[] = [];\n  const keptDims: number[] = [];\n  const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n  const axes = (axis == null || isEmptyArray) ?\n      null :\n      parseAxisParam(axis, shape).sort();\n  let j = 0;\n  for (let i = 0; i < shape.length; ++i) {\n    if (axes != null) {\n      if (axes[j] === i && shape[i] !== 1) {\n        throw new Error(\n            `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n      }\n      if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n        newShape.push(shape[i]);\n        keptDims.push(i);\n      }\n      if (axes[j] <= i) {\n        j++;\n      }\n    }\n    if (shape[i] !== 1) {\n      newShape.push(shape[i]);\n      keptDims.push(i);\n    }\n  }\n  return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType<D extends NumericDataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType<D extends DataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else if (dtype === 'string') {\n    values = new Array<'string'>(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors<D extends DataType>(\n    vals: DataTypeMap[D]|number[], dtype: D): void {\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n    }\n  }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n  return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n      dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n  if (newType === 'complex64') {\n    return false;\n  }\n  if (newType === 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'bool' && oldType === 'bool') {\n    return false;\n  }\n  return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n  return a instanceof Float32Array || a instanceof Int32Array ||\n      a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n  if (dtype === 'float32' || dtype === 'int32') {\n    return 4;\n  } else if (dtype === 'complex64') {\n    return 8;\n  } else if (dtype === 'bool') {\n    return 1;\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n  if (arr == null) {\n    return 0;\n  }\n  let bytes = 0;\n  arr.forEach(x => bytes += x.length);\n  return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n  return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n  return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n  return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n  if (Array.isArray(values)) {\n    return inferDtype(values[0]);\n  }\n  if (values instanceof Float32Array) {\n    return 'float32';\n  } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n    return 'int32';\n  } else if (isNumber(values)) {\n    return 'float32';\n  } else if (isString(values)) {\n    return 'string';\n  } else if (isBoolean(values)) {\n    return 'bool';\n  }\n  return 'float32';\n}\n\nexport function isFunction(f: Function) {\n  return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n  for (let i = start; i < size; ++i) {\n    if (size % i === 0) {\n      return i;\n    }\n  }\n  return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n  const rank = shape.length;\n  if (rank < 2) {\n    return [];\n  }\n\n  // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n  // strides.\n  const strides = new Array(rank - 1);\n  strides[rank - 2] = shape[rank - 1];\n  for (let i = rank - 3; i >= 0; --i) {\n    strides[i] = strides[i + 1] * shape[i + 1];\n  }\n  return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n    value: DataType, dtype: DataType): BackendValues {\n  if (dtype === 'string') {\n    return encodeString(value);\n  }\n\n  return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n  if (dtype === 'string') {\n    throw new Error('Cannot convert a string[] to a TypedArray');\n  }\n  if (Array.isArray(a)) {\n    a = flatten(a);\n  }\n\n  if (env().getBool('DEBUG')) {\n    checkConversionForErrors(a as number[], dtype);\n  }\n  if (noConversionNeeded(a, dtype)) {\n    return a as TypedArray;\n  }\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(a as number[]);\n  } else if (dtype === 'int32') {\n    return new Int32Array(a as number[]);\n  } else if (dtype === 'bool') {\n    const bool = new Uint8Array((a as number[]).length);\n    for (let i = 0; i < bool.length; ++i) {\n      if (Math.round((a as number[])[i]) !== 0) {\n        bool[i] = 1;\n      }\n    }\n    return bool;\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n  const ret = new Array();\n  if (shape.length === 1) {\n    const d = shape[0];\n    for (let i = 0; i < d; i++) {\n      ret[i] = a[offset + i];\n    }\n  } else {\n    const d = shape[0];\n    const rest = shape.slice(1);\n    const len = rest.reduce((acc, c) => acc * c);\n    for (let i = 0; i < d; i++) {\n      ret[i] = createNestedArray(offset + i * len, rest, a);\n    }\n  }\n  return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n  if (shape.length === 0) {\n    // Scalar type should return a single number.\n    return a[0];\n  }\n  const size = shape.reduce((acc, c) => acc * c);\n  if (size === 0) {\n    // A tensor with shape zero should be turned into empty list.\n    return [];\n  }\n  if (size !== a.length) {\n    throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n  }\n\n  return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n  return (a instanceof Float32Array && dtype === 'float32') ||\n      (a instanceof Int32Array && dtype === 'int32') ||\n      (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  const array = makeZerosTypedArray(size, dtype);\n  for (let i = 0; i < array.length; i++) {\n    array[i] = 1;\n  }\n  return array;\n}\n\nexport function makeZerosTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'int32') {\n    return new Int32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'bool') {\n    return new Uint8Array(size) as DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray<D extends DataType>(\n    shape: number[], dtype: D) {\n  const size = shape.reduce((prev, curr) => prev * curr, 1);\n  if (dtype == null || dtype === 'float32') {\n    return toNestedArray(shape, new Float32Array(size));\n  } else if (dtype === 'int32') {\n    return toNestedArray(shape, new Int32Array(size));\n  } else if (dtype === 'bool') {\n    return toNestedArray(shape, new Uint8Array(size));\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n  return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n  shape.forEach(dimSize => {\n    assert(\n        Number.isInteger(dimSize) && dimSize >= 0,\n        () =>\n            `Tensor must have a shape comprised of positive integers but got ` +\n            `shape [${shape}].`);\n  });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n    path: string, requestInits?: RequestInit): Promise<Response> {\n  return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n  encoding = encoding || 'utf-8';\n  return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n  encoding = encoding || 'utf-8';\n  return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n    locs: number[], rank: number, strides: number[]): number {\n  if (rank === 0) {\n    return 0;\n  } else if (rank === 1) {\n    return locs[0];\n  }\n  let index = locs[locs.length - 1];\n  for (let i = 0; i < locs.length - 1; ++i) {\n    index += strides[i] * locs[i];\n  }\n  return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n    index: number, rank: number, strides: number[]): number[] {\n  if (rank === 0) {\n    return [];\n  } else if (rank === 1) {\n    return [index];\n  }\n  const locs: number[] = new Array(rank);\n  for (let i = 0; i < locs.length - 1; ++i) {\n    locs[i] = Math.floor(index / strides[i]);\n    index -= locs[i] * strides[i];\n  }\n  locs[locs.length - 1] = index;\n  return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n  kernelName: string,\n  outputs: Tensor[],\n  inputs: NamedTensorMap,\n  timeMs: Promise<number|{error: string}>,\n  extraInfo: Promise<string>\n};\n\nexport class Profiler {\n  constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n    if (logger == null) {\n      this.logger = new Logger();\n    }\n  }\n\n  profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n      KernelProfile {\n    let outputs: Tensor[];\n    const holdResultWrapperFn = () => {\n      outputs = f();\n    };\n    const timer = this.backendTimer.time(holdResultWrapperFn);\n\n    for (let i = 0; i < outputs.length; i++) {\n      const output = outputs[i];\n      // Dangling promise here because we don't want to propagate up\n      // asynchronicity.\n      output.data().then(tensorVals => {\n        checkComputationForErrors(tensorVals, output.dtype, kernelName);\n      });\n    }\n\n    const kernelProfile = {\n      kernelName,\n      outputs,\n      inputs,\n      timeMs: timer.then(timing => timing.kernelMs),\n      extraInfo: timer.then(\n          timing => timing.getExtraProfileInfo != null ?\n              timing.getExtraProfileInfo() :\n              '')\n    };\n    return kernelProfile;\n  }\n\n  logKernelProfile(kernelProfile: KernelProfile): void {\n    const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n    outputs.forEach(result => {\n      Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n        this.logger.logKernelProfile(\n            kernelName, result, valueContainer[0], valueContainer[1], inputs,\n            valueContainer[2]);\n      });\n    });\n  }\n}\n\nexport function checkComputationForErrors<D extends DataType>(\n    vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n  if (dtype !== 'float32') {\n    // Only floating point computations will generate NaN values\n    return false;\n  }\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      // Throwing custom exception so behavior is testable.\n      console.warn(`Found ${num} in the result of '${kernelName}'`);\n      return true;\n    }\n  }\n  return false;\n}\n\nexport class Logger {\n  logKernelProfile(\n      name: string, result: Tensor, vals: TypedArray,\n      timeMs: number|{error: string}, inputs: NamedTensorMap,\n      extraInfo?: string) {\n    const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n                                              timeMs['error'];\n    const paddedName = util.rightPad(name, 25);\n    const rank = result.rank;\n    const size = result.size;\n    const shape = util.rightPad(result.shape.toString(), 14);\n    let inputShapesDescription = '';\n\n    for (const name in inputs) {\n      const input = inputs[name];\n      if (input != null) {\n        // The input might be a non-tensor (e.g HTMLImageElement), in which case\n        // we claim the output shape as input shape.\n        const inputShape = input.shape || result.shape;\n        const inputRank = inputShape.length;\n        inputShapesDescription +=\n            `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n      }\n    }\n\n    console.log(\n        `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n            inputShapesDescription}\\t%c${extraInfo}`,\n        'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n        'color: green', 'color: steelblue');\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n  id: number;\n  kernelName: string;\n  outputs: Tensor[];\n  inputs: NamedTensorMap;\n  // Optional params, defined only for ops with gradient impl.\n  gradient?: (dys: Tensor[]) => NamedGradientMap;\n  saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n  [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n    tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n  // Forward pass to compute all the nodes and Tensors that are transitively a\n  // function of x.\n  const tensorsFromX: {[tensorId: number]: boolean} = {};\n  const nodesFromX: {[nodeId: number]: boolean} = {};\n  for (let i = 0; i < xs.length; i++) {\n    tensorsFromX[xs[i].id] = true;\n  }\n\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n    for (const inputName in nodeInputs) {\n      const input = nodeInputs[inputName];\n\n      let anyInputFromX = false;\n      for (let j = 0; j < xs.length; j++) {\n        if (tensorsFromX[input.id]) {\n          node.outputs.forEach(output => tensorsFromX[output.id] = true);\n          anyInputFromX = true;\n          nodesFromX[node.id] = true;\n          break;\n        }\n      }\n\n      if (anyInputFromX) {\n        break;\n      }\n    }\n  }\n\n  // Backward pass to find all of the nodes and Tensors that lead to y.\n  const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n  tensorsLeadToY[y.id] = true;\n  const nodesToY: {[nodeId: number]: boolean} = {};\n\n  for (let i = tape.length - 1; i >= 0; i--) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n\n    // If any of the outputs lead to y, mark all of the inputs as leading to y.\n    for (let j = 0; j < node.outputs.length; j++) {\n      if (tensorsLeadToY[node.outputs[j].id]) {\n        for (const inputName in nodeInputs) {\n          tensorsLeadToY[nodeInputs[inputName].id] = true;\n          nodesToY[node.id] = true;\n        }\n        break;\n      }\n    }\n  }\n\n  // Return the paths that come from x and lead to y.\n  const filteredTape: TapeNode[] = [];\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n\n    if (nodesFromX[node.id] && nodesToY[node.id]) {\n      // Prune the inputs from the node that aren't a function of x.\n      const prunedInputs: {[inputName: string]: Tensor} = {};\n      for (const inputName in node.inputs) {\n        const nodeInput = node.inputs[inputName];\n        if (tensorsFromX[nodeInput.id]) {\n          prunedInputs[inputName] = nodeInput;\n        }\n      }\n\n      // Copy the node and overwrite inputsAndArgs to the pruned version.\n      const prunedNode = Object.assign({}, node);\n      prunedNode.inputs = prunedInputs;\n      prunedNode.outputs = node.outputs;\n\n      filteredTape.push(prunedNode);\n    }\n  }\n\n  return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n    tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n    filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n    add: (a: Tensor, b: Tensor) => Tensor) {\n  // Walk the tape backward and keep a map of Tensor to its gradient.\n  for (let i = filteredTape.length - 1; i >= 0; i--) {\n    const node = filteredTape[i];\n\n    const dys: Tensor[] = [];\n    node.outputs.forEach(o => {\n      const gradTensor = tensorAccumulatedGradientMap[o.id];\n      if (gradTensor != null) {\n        dys.push(gradTensor);\n      } else {\n        // This particular output is not in the back-propagation subgraph, so it\n        // does not affect the final output, thus we put null for its dy.\n        dys.push(null);\n      }\n    });\n\n    if (node.gradient == null) {\n      throw new Error(\n          `Cannot compute gradient: gradient function not found ` +\n          `for ${node.kernelName}.`);\n    }\n\n    // Backprop dy through this node and accumulate gradients over the inputs.\n    const inputGradients = node.gradient(dys);\n\n    for (const inputName in node.inputs) {\n      if (!(inputName in inputGradients)) {\n        throw new Error(\n            `Cannot backprop through input ${inputName}. ` +\n            `Available gradients found: ${Object.keys(inputGradients)}.`);\n      }\n\n      // Call the gradient function.\n      const dx = tidy(() => inputGradients[inputName]());\n      if (dx.dtype !== 'float32') {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n      }\n      const x = node.inputs[inputName];\n      if (!util.arraysEqual(dx.shape, x.shape)) {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `'${inputName}' has shape '${dx.shape}', which does not match ` +\n            `the shape of the input '${x.shape}'`);\n      }\n\n      if (tensorAccumulatedGradientMap[x.id] == null) {\n        tensorAccumulatedGradientMap[x.id] = dx;\n      } else {\n        const curGradient = tensorAccumulatedGradientMap[x.id];\n        tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n        curGradient.dispose();\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    verbose: boolean) {\n  const strides = computeStrides(shape);\n  const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n  const rank = shape.length;\n  const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n  const lines = ['Tensor'];\n  if (verbose) {\n    lines.push(`  dtype: ${dtype}`);\n    lines.push(`  rank: ${rank}`);\n    lines.push(`  shape: [${shape}]`);\n    lines.push(`  values:`);\n  }\n  lines.push(valsLines.map(l => '    ' + l).join('\\n'));\n  return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[]): number[] {\n  const n = sizeFromShape(shape);\n  const numCols = strides[strides.length - 1];\n  const padPerCol = new Array(numCols).fill(0);\n  const rank = shape.length;\n  const valuesOrTuples =\n      dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n  if (rank > 1) {\n    for (let row = 0; row < n / numCols; row++) {\n      const offset = row * numCols;\n      for (let j = 0; j < numCols; j++) {\n        padPerCol[j] = Math.max(\n            padPerCol[j],\n            valToString(valuesOrTuples[offset + j], 0, dtype).length);\n      }\n    }\n  }\n  return padPerCol;\n}\n\nfunction valToString(\n    val: number|string|[number, number], pad: number, dtype: DataType) {\n  let valStr: string;\n  if (Array.isArray(val)) {\n    valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n        `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n  } else if (isString(val)) {\n    valStr = `'${val}'`;\n  } else if (dtype === 'bool') {\n    valStr = boolNumToString(val);\n  } else {\n    valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n  }\n\n  return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n  return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[], padPerCol: number[], isLast = true): string[] {\n  const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n  const size = shape[0];\n  const rank = shape.length;\n  if (rank === 0) {\n    if (dtype === 'complex64') {\n      const complexTuple = createComplexTuples(vals);\n      return [valToString(complexTuple[0], 0, dtype)];\n    }\n    if (dtype === 'bool') {\n      return [boolNumToString(vals[0] as number)];\n    }\n    return [vals[0].toString()];\n  }\n\n  if (rank === 1) {\n    if (size > FORMAT_LIMIT_NUM_VALS) {\n      const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n      let firstVals = Array.from<number|string|[number, number]>(\n          vals.slice(0, firstValsSize));\n      let lastVals = Array.from<number|string|[number, number]>(vals.slice(\n          (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n          size * storagePerElement));\n      if (dtype === 'complex64') {\n        firstVals = createComplexTuples(firstVals);\n        lastVals = createComplexTuples(lastVals);\n      }\n      return [\n        '[' +\n        firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n            .join(', ') +\n        ', ..., ' +\n        lastVals\n            .map(\n                (x, i) => valToString(\n                    x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n            .join(', ') +\n        ']'\n      ];\n    }\n    const displayVals: Array<number|string|[number, number]> =\n        dtype === 'complex64' ? createComplexTuples(vals) :\n                                Array.from<number|string>(vals);\n\n    return [\n      '[' +\n      displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n          .join(', ') +\n      ']'\n    ];\n  }\n\n  // The array is rank 2 or more.\n  const subshape = shape.slice(1);\n  const substrides = strides.slice(1);\n  const stride = strides[0] * storagePerElement;\n  const lines: string[] = [];\n  if (size > FORMAT_LIMIT_NUM_VALS) {\n    for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          false /* isLast */));\n    }\n    lines.push('...');\n    for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  } else {\n    for (let i = 0; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  }\n  const sep = rank === 2 ? ',' : '';\n  lines[0] = '[' + lines[0] + sep;\n  for (let i = 1; i < lines.length - 1; i++) {\n    lines[i] = ' ' + lines[i] + sep;\n  }\n  let newLineSep = ',\\n';\n  for (let i = 2; i < rank; i++) {\n    newLineSep += '\\n';\n  }\n  lines[lines.length - 1] =\n      ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n  return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n                             TypedArray): Array<[number, number]> {\n  const complexTuples: Array<[number, number]> = [];\n  for (let i = 0; i < vals.length; i += 2) {\n    complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n  }\n  return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData<D extends DataType> {\n  dataId?: DataId;\n  values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer<R extends Rank, D extends DataType = 'float32'> {\n  size: number;\n  shape: ShapeMap[R];\n  strides: number[];\n  values: DataTypeMap[D];\n\n  constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.size = util.sizeFromShape(shape);\n\n    if (values != null) {\n      const n = values.length;\n      util.assert(\n          n === this.size,\n          () => `Length of values '${n}' does not match the size ` +\n              `inferred by the shape '${this.size}'.`);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `complex64 dtype TensorBuffers are not supported. Please create ` +\n          `a TensorBuffer for the real and imaginary parts separately and ` +\n          `call tf.complex(real, imag).`);\n    }\n    this.values = values || util.getArrayFromDType(dtype, this.size);\n    this.strides = computeStrides(shape);\n  }\n\n  /**\n   * Sets a value in the buffer at a given location.\n   *\n   * @param value The value to set.\n   * @param locs  The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  set(value: SingleValueMap[D], ...locs: number[]): void {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    util.assert(\n        locs.length === this.rank,\n        () => `The number of provided coordinates (${locs.length}) must ` +\n            `match the rank (${this.rank})`);\n\n    const index = this.locToIndex(locs);\n    this.values[index] = value as number;\n  }\n\n  /**\n   * Returns the value in the buffer at the provided location.\n   *\n   * @param locs The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  get(...locs: number[]): SingleValueMap[D] {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    let i = 0;\n    for (const loc of locs) {\n      if (loc < 0 || loc >= this.shape[i]) {\n        const msg = `Requested out of range element at ${locs}. ` +\n            `  Buffer shape=${this.shape}`;\n        throw new Error(msg);\n      }\n      i++;\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return this.values[index] as SingleValueMap[D];\n  }\n\n  locToIndex(locs: number[]): number {\n    if (this.rank === 0) {\n      return 0;\n    } else if (this.rank === 1) {\n      return locs[0];\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return index;\n  }\n\n  indexToLoc(index: number): number[] {\n    if (this.rank === 0) {\n      return [];\n    } else if (this.rank === 1) {\n      return [index];\n    }\n    const locs: number[] = new Array(this.shape.length);\n    for (let i = 0; i < locs.length - 1; ++i) {\n      locs[i] = Math.floor(index / this.strides[i]);\n      index -= locs[i] * this.strides[i];\n    }\n    locs[locs.length - 1] = index;\n    return locs;\n  }\n\n  get rank() {\n    return this.shape.length;\n  }\n\n  /**\n   * Creates an immutable `tf.Tensor` object from the buffer.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  toTensor(): Tensor<R> {\n    return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n        Tensor<R>;\n  }\n}\n\nexport interface TensorTracker {\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: Backend): Tensor;\n  makeVariable(\n      initialValue: Tensor, trainable?: boolean, name?: string,\n      dtype?: DataType): Variable;\n  incRef(a: Tensor, backend: Backend): void;\n  disposeTensor(t: Tensor): void;\n  disposeVariable(v: Variable): void;\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n  cast<T extends Tensor>(x: T, dtype: DataType): T;\n  buffer<R extends Rank, D extends DataType>(\n      shape: ShapeMap[R], dtype: D,\n      values?: DataTypeMap[D]): TensorBuffer<R, D>;\n  print<T extends Tensor>(x: T, verbose: boolean): void;\n  clone<T extends Tensor>(x: T): T;\n  // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n  trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n  opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n  deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object;  // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor<R extends Rank = Rank> {\n  /** Unique id of this tensor. */\n  readonly id: number;\n  /**\n   * Id of the bucket holding the data for this tensor. Multiple arrays can\n   * point to the same bucket (e.g. when calling array.reshape()).\n   */\n  dataId: DataId;\n  /** The shape of the tensor. */\n  readonly shape: ShapeMap[R];\n  /** Number of elements in the tensor. */\n  readonly size: number;\n  /** The data type for the array. */\n  readonly dtype: DataType;\n  /** The rank type for the array (see `Rank` enum). */\n  readonly rankType: R;\n\n  /** Whether this tensor has been globally kept. */\n  kept = false;\n  /** The id of the scope this tensor is being tracked in. */\n  scopeId: number;\n\n  /**\n   * Number of elements to skip in each dimension when indexing. See\n   * https://docs.scipy.org/doc/numpy/reference/generated/\\\n   * numpy.ndarray.strides.html\n   */\n  readonly strides: number[];\n\n  constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.dtype = dtype || 'float32';\n    this.size = util.sizeFromShape(shape);\n    this.strides = computeStrides(shape);\n    this.dataId = dataId;\n    this.id = id;\n    this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n  }\n\n  get rank(): number {\n    return this.shape.length;\n  }\n\n  /**\n   * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async buffer<D extends DataType = 'float32'>(): Promise<TensorBuffer<R, D>> {\n    const vals = await this.data<D>();\n    return opHandler.buffer(this.shape, this.dtype as D, vals);\n  }\n\n  /**\n   * Returns a `tf.TensorBuffer` that holds the underlying data.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  bufferSync<D extends DataType = 'float32'>(): TensorBuffer<R, D> {\n    return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * asynchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async array(): Promise<ArrayMap[R]> {\n    const vals = await this.data();\n    return toNestedArray(this.shape, vals) as ArrayMap[R];\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * synchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  arraySync(): ArrayMap[R] {\n    return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n  }\n\n  /**\n   * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n   * promise of `TypedArray` that resolves when the computation has finished.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]> {\n    this.throwIfDisposed();\n    const data = trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      const bytes = await data as Uint8Array[];\n      try {\n        return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as Promise<DataTypeMap[D]>;\n  }\n\n  /**\n   * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n   * UI thread until the values are ready, which can cause performance issues.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D] {\n    this.throwIfDisposed();\n    const data = trackerFn().readSync(this.dataId);\n    if (this.dtype === 'string') {\n      try {\n        return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n            DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as DataTypeMap[D];\n  }\n\n  /** Returns the underlying bytes of the tensor's data. */\n  async bytes(): Promise<Uint8Array[]|Uint8Array> {\n    this.throwIfDisposed();\n    const data = await trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      return data as Uint8Array[];\n    } else {\n      return new Uint8Array((data as TypedArray).buffer);\n    }\n  }\n\n  /**\n   * Disposes `tf.Tensor` from memory.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dispose(): void {\n    if (this.isDisposed) {\n      return;\n    }\n    trackerFn().disposeTensor(this);\n    this.isDisposedInternal = true;\n  }\n\n  protected isDisposedInternal = false;\n  get isDisposed(): boolean {\n    return this.isDisposedInternal;\n  }\n\n  throwIfDisposed() {\n    if (this.isDisposed) {\n      throw new Error(`Tensor is disposed.`);\n    }\n  }\n\n  /**\n   * Prints the `tf.Tensor`. See `tf.print` for details.\n   *\n   * @param verbose Whether to print verbose information about the tensor,\n   *    including dtype and size.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  print(verbose = false): void {\n    return opHandler.print(this, verbose);\n  }\n\n  /**\n   * Returns a copy of the tensor. See `tf.clone` for details.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  clone<T extends Tensor>(this: T): T {\n    this.throwIfDisposed();\n    return opHandler.clone(this);\n  }\n\n  /**\n   * Returns a human-readable description of the tensor. Useful for logging.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  toString(verbose = false): string {\n    const vals = this.dataSync();\n    return tensorToString(vals, this.shape, this.dtype, verbose);\n  }\n\n  cast<T extends this>(dtype: DataType): T {\n    this.throwIfDisposed();\n    return opHandler.cast(this as T, dtype);\n  }\n  variable(trainable = true, name?: string, dtype?: DataType): Variable<R> {\n    this.throwIfDisposed();\n    return trackerFn().makeVariable(this, trainable, name, dtype) as\n        Variable<R>;\n  }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n  value: (instance: Tensor) => {\n    // Implementation note: we should use properties of the object that will be\n    // defined before the constructor body has finished executing (methods).\n    // This is because when this code is transpiled by babel, babel will call\n    // classCallCheck before the constructor body is run.\n    // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n    return !!instance && instance.data != null && instance.dataSync != null &&\n        instance.throwIfDisposed != null;\n  }\n});\n\nexport interface NumericTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: NumericDataType;\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D];\n  data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]>;\n}\n\nexport interface StringTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: 'string';\n  dataSync<D extends DataType = 'string'>(): DataTypeMap[D];\n  data<D extends DataType = 'string'>(): Promise<DataTypeMap[D]>;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor<Rank.R0>;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor<Rank.R1>;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor<Rank.R2>;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor<Rank.R3>;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor<Rank.R4>;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor<Rank.R5>;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor<Rank.R6>;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable<R extends Rank = Rank> extends Tensor<R> {\n  name: string;\n\n  constructor(\n      initialValue: Tensor<R>, public trainable: boolean, name: string,\n      tensorId: number) {\n    super(\n        initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n    this.name = name;\n  }\n\n  /**\n   * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n   * the same shape and dtype as the old `tf.Tensor`.\n   *\n   * @param newValue New tensor to be assigned to this variable.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  assign(newValue: Tensor<R>): void {\n    if (newValue.dtype !== this.dtype) {\n      throw new Error(\n          `dtype of the new value (${newValue.dtype}) and ` +\n          `previous value (${this.dtype}) must match`);\n    }\n    if (!util.arraysEqual(newValue.shape, this.shape)) {\n      throw new Error(\n          `shape of the new value (${newValue.shape}) and ` +\n          `previous value (${this.shape}) must match`);\n    }\n    trackerFn().disposeTensor(this);\n    this.dataId = newValue.dataId;\n    trackerFn().incRef(this, null /* backend */);\n  }\n\n  dispose(): void {\n    trackerFn().disposeVariable(this);\n    this.isDisposedInternal = true;\n  }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n  value: (instance: Variable) => {\n    return instance instanceof Tensor && instance.assign != null &&\n        instance.assign instanceof Function;\n  }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n  R0: number[];\n  R1: [number];\n  R2: [number, number];\n  R3: [number, number, number];\n  R4: [number, number, number, number];\n  R5: [number, number, number, number, number];\n  R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n  R0: number;\n  R1: number[];\n  R2: number[][];\n  R3: number[][][];\n  R4: number[][][][];\n  R5: number[][][][][];\n  R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n  float32: Float32Array;\n  int32: Int32Array;\n  bool: Uint8Array;\n  complex64: Float32Array;\n  string: string[];\n}\n\nexport interface SingleValueMap {\n  bool: boolean;\n  int32: number;\n  float32: number;\n  complex64: number;\n  string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n  R0 = 'R0',\n  R1 = 'R1',\n  R2 = 'R2',\n  R3 = 'R3',\n  R4 = 'R4',\n  R5 = 'R5',\n  R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray<T> =\n    T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray<T extends any> {\n  [index: number]: T|RecursiveArray<T>;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'int32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'bool',\n  'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n  'float32' = 'float32',\n  'int32' = 'float32',\n  'bool' = 'float32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n  'float32' = 'complex64',\n  'int32' = 'complex64',\n  'bool' = 'complex64',\n  'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n  'float32': UpcastFloat32AndMap,\n  'int32': UpcastInt32AndMap,\n  'bool': UpcastBoolAndMap,\n  'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n  if (typeA === 'string' || typeB === 'string') {\n    if (typeA === 'string' && typeB === 'string') {\n      return 'string';\n    }\n    throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n  }\n  return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n  return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n    TypedArray|number|boolean|string|RecursiveArray<number|number[]|TypedArray>|\n    RecursiveArray<boolean>|RecursiveArray<string>|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n    string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n    boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n    boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n    TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n    string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n    TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n    string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n  width: number;\n  height: number;\n  data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch<T extends Tensor>(a: T, b: T): [T, T] {\n  if (a.dtype === b.dtype) {\n    return [a, b];\n  }\n  const dtype = upcastType(a.dtype, b.dtype);\n  return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n  assert(\n      a.dtype === b.dtype,\n      () => `The dtypes of the first(${a.dtype}) and` +\n          ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n  return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n *   `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n *   is safe to pass any object here, except that `Promise`s are not\n *   supported.\n * @returns An array of `Tensors` found within the passed object. If the\n *   argument is simply a `Tensor', a list containing that `Tensor` is\n *   returned. If the object is not a `Tensor` or does not\n *   contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n  const list: Tensor[] = [];\n  const seen = new Set<{}|void>();\n  walkTensorContainer(result, list, seen);\n  return list;\n}\n\nfunction walkTensorContainer(\n    container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n  if (container == null) {\n    return;\n  }\n  if (container instanceof Tensor) {\n    list.push(container);\n    return;\n  }\n  if (!isIterable(container)) {\n    return;\n  }\n  // Iteration over keys works also for arrays.\n  const iterable = container as TensorContainerArray;\n  for (const k in iterable) {\n    const val = iterable[k];\n    if (!seen.has(val)) {\n      seen.add(val);\n      walkTensorContainer(val, list, seen);\n    }\n  }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n  return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc<T> = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n *   value: Tensor,\n *   gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc<T extends Tensor> =\n    (...inputs: Array<Tensor|GradSaveFunc>) => {\n      value: T;\n      gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n    };\n\nexport type MemoryInfo = {\n  numTensors: number; numDataBuffers: number; numBytes: number;\n  unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n  name: string; bytesAdded: number; totalBytesSnapshot: number;\n  tensorsAdded: number;\n  totalTensorsSnapshot: number;\n  inputShapes: number[][];\n  outputShapes: number[][];\n  kernelTimeMs: number | {error: string} | Promise<number|{error: string}>;\n  extraInfo: string | Promise<string>;\n};\n\nexport type ProfileInfo = {\n  newBytes: number; newTensors: number; peakBytes: number;\n  kernels: KernelInfo[];\n  result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n  wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn<T extends TensorContainer> = () => T;\n\ninterface ScopeState {\n  track: Tensor[];\n  name: string;\n  id: number;\n}\n\nclass EngineState {\n  // Public since optimizers will use it.\n  registeredVariables: NamedVariableMap = {};\n\n  nextTapeNodeId = 0;\n  numBytes = 0;\n  numTensors = 0;\n  numStringTensors = 0;\n  numDataBuffers = 0;\n\n  activeTape: TapeNode[];\n  // Number of nested tf.grad() statements when computing higher-order\n  // gradients. E.g. `1` for first-order gradients and `2` for second-order\n  // gradients. Used to track if the tape should be removed after a backprop.\n  gradientDepth = 0;\n  // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n  // off the tape.\n  kernelDepth = 0;\n\n  // Keep Tensors that parallel the tapes.\n  activeScope: ScopeState;\n  scopeStack: ScopeState[] = [];\n  /**\n   * Keeps track of the number of data moves during a kernel execution. We\n   * maintain a stack since kernels can call other kernels, recursively.\n   */\n  numDataMovesStack: number[] = [];\n  nextScopeId = 0;\n\n  tensorInfo = new WeakMap<DataId, {\n    backend: KernelBackend,\n    bytes: number,\n    dtype: DataType,\n    shape: number[],\n    refCount: number\n  }>();\n\n  profiling = false;\n  activeProfile: ProfileInfo =\n      {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n  dispose() {\n    for (const variableName in this.registeredVariables) {\n      this.registeredVariables[variableName].dispose();\n    }\n  }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n  state: EngineState;\n  backendName: string;\n  registry: {[id: string]: KernelBackend} = {};\n  registryFactory: {\n    [id: string]: {\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority: number\n    }\n  } = {};\n\n  private profiler: Profiler;\n  private backendInstance: KernelBackend;\n  private pendingBackendInit: Promise<boolean>;\n  private pendingBackendInitId = 0;\n\n  constructor(public ENV: Environment) {\n    this.state = new EngineState();\n  }\n\n  async ready(): Promise<void> {\n    if (this.pendingBackendInit != null) {\n      return this.pendingBackendInit.then(() => {});\n    }\n    if (this.backendInstance != null) {\n      return;\n    }\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const success = await this.initializeBackend(backendName).success;\n      if (success) {\n        await this.setBackend(backendName);\n        return;\n      }\n    }\n\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  get backend(): KernelBackend {\n    if (this.pendingBackendInit != null) {\n      throw new Error(\n          `Backend '${this.backendName}' has not yet been initialized. Make ` +\n          `sure to await tf.ready() or await tf.setBackend() before calling ` +\n          `other methods`);\n    }\n    if (this.backendInstance == null) {\n      const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n      if (asyncInit) {\n        throw new Error(\n            `The highest priority backend '${name}' has not yet been ` +\n            `initialized. Make sure to await tf.ready() or ` +\n            `await tf.setBackend() before calling other methods`);\n      }\n      this.setBackend(name);\n    }\n    return this.backendInstance;\n  }\n\n  backendNames(): string[] {\n    return Object.keys(this.registryFactory);\n  }\n\n  findBackend(backendName: string): KernelBackend {\n    if (!(backendName in this.registry)) {\n      // If the backend hasn't been initialized but we have a registry entry for\n      // it, initialize it and return it.\n      if (backendName in this.registryFactory) {\n        const {asyncInit} = this.initializeBackend(backendName);\n        if (asyncInit) {\n          // Backend is not ready yet.\n          return null;\n        }\n      } else {\n        return null;\n      }\n    }\n    return this.registry[backendName];\n  }\n\n  findBackendFactory(backendName: string):\n      () => KernelBackend | Promise<KernelBackend> {\n    if (!(backendName in this.registryFactory)) {\n      return null;\n    }\n    return this.registryFactory[backendName].factory;\n  }\n\n  registerBackend(\n      backendName: string,\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority = 1): boolean {\n    if (backendName in this.registryFactory) {\n      console.warn(\n          `${backendName} backend was already registered. ` +\n          `Reusing existing backend factory.`);\n      return false;\n    }\n    this.registryFactory[backendName] = {factory, priority};\n    return true;\n  }\n\n  async setBackend(backendName: string): Promise<boolean> {\n    if (this.registryFactory[backendName] == null) {\n      throw new Error(`Backend name '${backendName}' not found in registry`);\n    }\n    this.backendName = backendName;\n    if (this.registry[backendName] == null) {\n      this.backendInstance = null;\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      const result = asyncInit ? await success : success;\n      if (!result) {\n        return false;\n      }\n    }\n    this.backendInstance = this.registry[backendName];\n    this.setupRegisteredKernels();\n    // Reset the profiler.\n    this.profiler = new Profiler(this.backendInstance);\n\n    return true;\n  }\n\n  private setupRegisteredKernels(): void {\n    const kernels = getKernelsForBackend(this.backendName);\n    kernels.forEach(kernel => {\n      if (kernel.setupFunc != null) {\n        kernel.setupFunc(this.backendInstance);\n      }\n    });\n  }\n\n  private disposeRegisteredKernels(backendName: string): void {\n    const kernels = getKernelsForBackend(backendName);\n    kernels.forEach(kernel => {\n      if (kernel.disposeFunc != null) {\n        kernel.disposeFunc(this.registry[backendName]);\n      }\n    });\n  }\n\n  /**\n   * Initializes a backend by looking up the backend name in the factory\n   * registry and calling the factory method. Returns a boolean representing\n   * whether the initialization of the backend suceeded. Throws an error if\n   * there is no backend in the factory registry.\n   */\n  private initializeBackend(backendName: string):\n      {success: boolean|Promise<boolean>, asyncInit: boolean} {\n    const registryFactoryEntry = this.registryFactory[backendName];\n    if (registryFactoryEntry == null) {\n      throw new Error(\n          `Cannot initialize backend ${backendName}, no registration found.`);\n    }\n\n    try {\n      const backend = registryFactoryEntry.factory();\n      /* Test if the factory returns a promise.\n      Done in a more liberal way than\n      previous 'Promise.resolve(backend)===backend'\n      as we needed to account for custom Promise\n      implementations (e.g. Angular) */\n      if (backend && !(backend instanceof KernelBackend)\n          && typeof backend.then === 'function') {\n        const promiseId = ++this.pendingBackendInitId;\n        const success =\n            backend\n                .then(backendInstance => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.registry[backendName] = backendInstance;\n                  this.pendingBackendInit = null;\n                  return true;\n                })\n                .catch(err => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.pendingBackendInit = null;\n                  console.warn(\n                      `Initialization of backend ${backendName} failed`);\n                  console.warn(err.stack || err.message);\n                  return false;\n                });\n        this.pendingBackendInit = success;\n        return {success, asyncInit: true};\n      } else {\n        this.registry[backendName] = backend as KernelBackend;\n        return {success: true, asyncInit: false};\n      }\n    } catch (err) {\n      console.warn(`Initialization of backend ${backendName} failed`);\n      console.warn(err.stack || err.message);\n      return {success: false, asyncInit: false};\n    }\n  }\n\n  removeBackend(backendName: string): void {\n    if (!(backendName in this.registryFactory)) {\n      throw new Error(`${backendName} backend not found in registry`);\n    }\n    if (this.backendName === backendName && this.pendingBackendInit != null) {\n      // There is a pending promise of the backend we want to remove. Make it\n      // obsolete.\n      this.pendingBackendInitId++;\n    }\n\n    if (backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n\n    delete this.registryFactory[backendName];\n\n    // Unset the backend if it is active.\n    if (this.backendName === backendName) {\n      this.pendingBackendInit = null;\n      this.backendName = null;\n      this.backendInstance = null;\n    }\n  }\n\n  private getSortedBackends(): string[] {\n    if (Object.keys(this.registryFactory).length === 0) {\n      throw new Error('No backend found in registry.');\n    }\n    return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n      // Highest priority comes first.\n      return this.registryFactory[b].priority -\n          this.registryFactory[a].priority;\n    });\n  }\n\n  private initializeBackendsAndReturnBest():\n      {name: string, asyncInit: boolean} {\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      if (asyncInit || success) {\n        return {name: backendName, asyncInit};\n      }\n    }\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  moveData(backend: KernelBackend, dataId: DataId) {\n    const info = this.state.tensorInfo.get(dataId);\n    const srcBackend = info.backend;\n    const values = this.readSync(dataId);\n    // Delete the tensor from the old backend and move it to the new\n    // backend.\n    srcBackend.disposeData(dataId);\n    info.backend = backend;\n    backend.move(dataId, values, info.shape, info.dtype);\n    if (this.shouldCheckForMemLeaks()) {\n      // Track the number of moves during a kernel execution to correctly\n      // detect memory leaks.\n      this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n    }\n  }\n\n  tidy<T extends TensorContainer>(nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>):\n      T {\n    let name: string = null;\n    if (fn == null) {\n      // Called with only 1 argument.\n      if (typeof nameOrFn !== 'function') {\n        throw new Error('Please provide a function to tidy()');\n      }\n      fn = nameOrFn;\n    } else {\n      // Called with 2 arguments.\n      if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n        throw new Error(\n            'When calling with two arguments, the first argument ' +\n            'to tidy() must be a string');\n      }\n      if (typeof fn !== 'function') {\n        throw new Error(\n            'When calling with two arguments, the 2nd argument ' +\n            'to tidy() must be a function');\n      }\n      name = nameOrFn as string;\n      // TODO(nsthorat,smilkov): Do operation logging and performance\n      // profiling.\n    }\n    let result: T;\n    return this.scopedRun(\n        () => this.startScope(name), () => this.endScope(result), () => {\n          result = fn();\n          if (result instanceof Promise) {\n            console.error('Cannot return a Promise inside of tidy.');\n          }\n          return result;\n        });\n  }\n\n  private scopedRun<T>(start: () => void, end: () => void, f: () => T): T {\n    start();\n    try {\n      const res = f();\n      end();\n      return res;\n    } catch (ex) {\n      end();\n      throw ex;\n    }\n  }\n\n  private static nextTensorId = 0;\n  private nextTensorId(): number {\n    return Engine.nextTensorId++;\n  }\n\n  private static nextVariableId = 0;\n  private nextVariableId(): number {\n    return Engine.nextVariableId++;\n  }\n\n  /**\n   * This method is called instead of the public-facing tensor.clone() when\n   * saving a tensor for backwards pass. It makes sure to add the clone\n   * operation to the tape regardless of being called inside a kernel\n   * execution.\n   *\n   * This method will go away once all kernels are modularized since we won't\n   * need to turn off the tape inside runKernel().\n   */\n  private clone(x: Tensor): Tensor {\n    const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n    const inputs = {x};\n    const grad = (dy: Tensor) => ({\n      x: () => {\n        const dtype = 'float32';\n        const gradInputs = {x: dy};\n        const attrs = {dtype};\n\n        return ENGINE.runKernelFunc(\n            backend => backend.cast(dy, dtype),\n            gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n            attrs as {} as NamedAttrMap);\n      }\n    });\n    const saved: Tensor[] = [];\n    this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n    return y;\n  }\n\n  /**\n   * Execute a kernel with the given name and return the output tensor.\n   *\n   * @param kernelName The name of the kernel to execute.\n   * @param inputs A map of input names to tensors.\n   * @param attrs A map of attribute names to their values. An attribute is a\n   *     primitive (non-tensor) input to the kernel.\n   * @param inputsToSave A list of tensors, inputs to save for the backprop\n   *     computation.\n   * @param outputsToSave A list of booleans, specifying which output to save\n   *     for the backprop computation. These are booleans since the output\n   * tensors are not visible to the user.\n   */\n  runKernel(\n      kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n      inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n    const forwardFunc: null = null;\n    const backwardsFunc: null = null;\n    // Call runKernel as a stop-gap until we modularize all kernels.\n    // Once we modularize all kernels, we will remove the existing\n    // `runKernelFunc`.\n    return this.runKernelFunc(\n        forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n        outputsToSave);\n  }\n\n  private shouldCheckForMemLeaks(): boolean {\n    return this.ENV.getBool('IS_TEST');\n  }\n\n  private checkKernelForMemLeak(\n      kernelName: string, numDataIdsBefore: number,\n      outInfos: TensorInfo[]): void {\n    const numDataIdsAfter = this.backend.numDataIds();\n\n    // Count the number of data ids associated with the result of the kernel.\n    let numOutputDataIds = 0;\n    outInfos.forEach(info => {\n      // Complex numbers allocate 3 data ids, one for 'real', one for\n      // 'imaginary', and one for the container that holds the former two.\n      numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n    });\n\n    // Account for the number of moves during kernel execution. A \"data move\"\n    // can happen in the middle of a kernel execution, placing a new (key,value)\n    // pair in the data storage. Since data moves have net zero effect (we\n    // always remove the data from the old backend), we have to cancel them out\n    // when detecting memory leaks.\n    const numMoves =\n        this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n    const dataIdsLeaked =\n        numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n    if (dataIdsLeaked > 0) {\n      throw new Error(\n          `Backend '${this.backendName}' has an internal memory leak ` +\n          `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n    }\n  }\n\n  /**\n   * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n   *     only for kernels that are not yet fully modularized.\n   */\n  runKernelFunc<T extends Tensor|Tensor[], I extends NamedTensorMap>(\n      forwardFunc: ForwardFunc<T>, inputs: I,\n      backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n      kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n      outputsToSave?: boolean[]): T {\n    let outputs: Tensor[];\n    let saved: Tensor[] = [];\n    const isTapeOn = this.isTapeOn();\n    if (kernelName == null) {\n      kernelName =\n          this.state.activeScope != null ? this.state.activeScope.name : '';\n    }\n\n    const startingBytecount = this.state.numBytes;\n    const startingNumTensors = this.state.numTensors;\n\n    if (this.shouldCheckForMemLeaks()) {\n      this.state.numDataMovesStack.push(0);\n    }\n\n    let kernelFunc: () => Tensor[];\n    const kernel = getKernel(kernelName, this.backendName);\n    let out: TensorInfo|TensorInfo[];\n    if (kernel != null) {\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n        const outInfos = Array.isArray(out) ? out : [out];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n        }\n        const outTensors = outInfos.map(\n            ({dataId, shape, dtype}) =>\n                this.makeTensorFromDataId(dataId, shape, dtype));\n\n        // Save the inputs and outputs.\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (isTapeOn) {\n          let tensorsToSave =\n              this.getTensorsForGradient(kernelName, inputs, outTensors);\n          if (tensorsToSave == null) {\n            // Fallback for ops that call runKernelFunc and pass in\n            // inputsToSave and outputsToSave. Currently this is the set of ops\n            // with kernel support in the WASM backend. Once those ops and\n            // respective gradients are modularised we can remove this path.\n            if (outputsToSave == null) {\n              outputsToSave = [];\n            }\n            const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n            tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n          }\n          saved = this.saveTensorsForBackwardMode(tensorsToSave);\n        }\n        return outTensors;\n      };\n    } else {\n      const saveFunc: GradSaveFunc = (tensors) => {\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (!isTapeOn) {\n          return;\n        }\n        saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n      };\n\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n        const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n        }\n        return outs;\n      };\n    }\n\n    // Stop recording to a tape when running a kernel.\n    let kernelProfile: KernelProfile;\n    this.scopedRun(\n        () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n          if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n            outputs = kernelFunc();\n          } else {\n            kernelProfile = this.profiler.profileKernel(\n                kernelName, inputs, () => kernelFunc());\n            if (this.ENV.getBool('DEBUG')) {\n              this.profiler.logKernelProfile(kernelProfile);\n            }\n            outputs = kernelProfile.outputs;\n          }\n        });\n\n    if (isTapeOn) {\n      this.addTapeNode(\n          kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n    }\n\n    if (this.state.profiling) {\n      this.state.activeProfile.kernels.push({\n        name: kernelName,\n        bytesAdded: this.state.numBytes - startingBytecount,\n        totalBytesSnapshot: this.state.numBytes,\n        tensorsAdded: this.state.numTensors - startingNumTensors,\n        totalTensorsSnapshot: this.state.numTensors,\n        inputShapes: Object.keys(inputs).map(\n            key => inputs[key] != null ? inputs[key].shape : null),\n        outputShapes: outputs.map(item => item.shape),\n        kernelTimeMs: kernelProfile.timeMs,\n        extraInfo: kernelProfile.extraInfo\n      });\n    }\n    return (Array.isArray(out) ? outputs : outputs[0]) as T;\n  }\n\n  /**\n   * Saves tensors used in forward mode for use in backward mode.\n   *\n   * @param tensors the list of tensors to save.\n   */\n  private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n    const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n    return saved;\n  }\n\n  /**\n   * Returns a list of tensors to save for a given gradient calculation.\n   *\n   * Returns undefined if their is no registered gradient for this kernel in the\n   * gradient registry.\n   *\n   * @param kernelName name of kernel to look up gradient for.\n   * @param inputs a map of input tensors.\n   * @param outputs an array of output tensors from forward mode of kernel.\n   */\n  private getTensorsForGradient(\n      kernelName: string, inputs: NamedTensorMap,\n      outputs: Tensor[]): Tensor[]|null {\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      const inputsToSave: string[] = gradConfig.inputsToSave || [];\n      const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n      // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n      // specified in inputsToSave will be saved.\n      let inputTensorsToSave: Tensor[];\n      if (gradConfig.saveAllInputs) {\n        util.assert(\n            Array.isArray(inputs),\n            () => 'saveAllInputs is true, expected inputs to be an array.');\n\n        inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n      } else {\n        inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n      }\n\n      const outputTensorsToSave: Tensor[] =\n          outputs.filter((_, i) => outputsToSave[i]);\n\n      return inputTensorsToSave.concat(outputTensorsToSave);\n    }\n    // TODO(yassogba) throw exception here once all runkernelFunc calls with\n    // inputsToSave/outputsToSave are removed\n    return null;\n  }\n\n  /**\n   * Internal method used by public APIs for tensor creation. Makes a new\n   * tensor with the provided shape, dtype and values. It always\n   * creates a new data id and writes the values to the underlying backend.\n   */\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    if (values == null) {\n      throw new Error('Values passed to engine.makeTensor() are null');\n    }\n    dtype = dtype || 'float32';\n    backend = backend || this.backend;\n    let backendVals = values as BackendValues;\n    if (dtype === 'string' && util.isString(values[0])) {\n      backendVals = (values as string[]).map(d => util.encodeString(d));\n    }\n    const dataId = backend.write(backendVals, shape, dtype);\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n\n    // Count bytes for string tensors.\n    if (dtype === 'string') {\n      const info = this.state.tensorInfo.get(dataId);\n      const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n      this.state.numBytes += newBytes - info.bytes;\n      info.bytes = newBytes;\n    }\n    return t;\n  }\n\n  /**\n   * Internal method used by backends. Makes a new tensor\n   * that is a wrapper around an existing data id. It doesn't create\n   * a new data id, only increments the ref count used in memory tracking.\n   */\n  makeTensorFromDataId(\n      dataId: DataId, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    dtype = dtype || 'float32';\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n    return t;\n  }\n\n  makeVariable(\n      initialValue: Tensor, trainable = true, name?: string,\n      dtype?: DataType): Variable {\n    name = name || this.nextVariableId().toString();\n    if (dtype != null && dtype !== initialValue.dtype) {\n      initialValue = initialValue.cast(dtype);\n    }\n    const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n    if (this.state.registeredVariables[v.name] != null) {\n      throw new Error(`Variable with name ${v.name} was already registered`);\n    }\n    this.state.registeredVariables[v.name] = v;\n    this.incRef(v, this.backend);\n    return v;\n  }\n\n  incRef(a: Tensor, backend: KernelBackend): void {\n    const refCount = this.state.tensorInfo.has(a.dataId) ?\n        this.state.tensorInfo.get(a.dataId).refCount :\n        0;\n    this.state.numTensors++;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors++;\n    }\n    if (refCount === 0) {\n      this.state.numDataBuffers++;\n\n      // Bytes for complex numbers are counted by their components. Bytes for\n      // string tensors are counted when writing values.\n      let bytes = 0;\n      if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n        bytes = a.size * util.bytesPerElement(a.dtype);\n      }\n      this.state.tensorInfo.set(a.dataId, {\n        backend: backend || this.backend,\n        dtype: a.dtype,\n        shape: a.shape,\n        bytes,\n        refCount: 0\n      });\n      this.state.numBytes += bytes;\n    }\n\n    this.state.tensorInfo.get(a.dataId).refCount++;\n\n    if (!(a instanceof Variable)) {\n      this.track(a);\n    }\n  }\n\n  disposeTensor(a: Tensor): void {\n    if (!this.state.tensorInfo.has(a.dataId)) {\n      return;\n    }\n\n    this.state.numTensors--;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors--;\n    }\n    const info = this.state.tensorInfo.get(a.dataId);\n    const refCount = info.refCount;\n\n    if (refCount <= 1) {\n      // Don't count bytes for complex numbers as they are counted by their\n      // components.\n      if (a.dtype !== 'complex64') {\n        this.state.numBytes -= info.bytes;\n      }\n      this.state.numDataBuffers--;\n\n      info.backend.disposeData(a.dataId);\n      this.state.tensorInfo.delete(a.dataId);\n    } else {\n      this.state.tensorInfo.get(a.dataId).refCount--;\n    }\n    // TODO(nsthorat): Construct an error and save the stack trace for\n    // debugging when in debug mode. Creating a stack trace is too expensive\n    // to do unconditionally.\n  }\n\n  disposeVariables(): void {\n    for (const varName in this.state.registeredVariables) {\n      const v = this.state.registeredVariables[varName];\n      this.disposeVariable(v);\n    }\n  }\n\n  disposeVariable(v: Variable): void {\n    this.disposeTensor(v);\n    if (this.state.registeredVariables[v.name] != null) {\n      delete this.state.registeredVariables[v.name];\n    }\n  }\n\n  memory(): MemoryInfo {\n    const info = this.backend.memory() as MemoryInfo;\n    info.numTensors = this.state.numTensors;\n    info.numDataBuffers = this.state.numDataBuffers;\n    info.numBytes = this.state.numBytes;\n    if (this.state.numStringTensors > 0) {\n      info.unreliable = true;\n      if (info.reasons == null) {\n        info.reasons = [];\n      }\n      info.reasons.push(\n          'Memory usage by string tensors is approximate ' +\n          '(2 bytes per character)');\n    }\n    return info;\n  }\n\n  async profile(query: () => (TensorContainer | Promise<TensorContainer>)):\n      Promise<ProfileInfo> {\n    this.state.profiling = true;\n\n    const startBytes = this.state.numBytes;\n    const startNumTensors = this.state.numTensors;\n\n    this.state.activeProfile.kernels = [];\n    this.state.activeProfile.result = await query();\n\n    this.state.profiling = false;\n\n    this.state.activeProfile.peakBytes = Math.max(\n        ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n    this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n    this.state.activeProfile.newTensors =\n        this.state.numTensors - startNumTensors;\n    for (const kernel of this.state.activeProfile.kernels) {\n      kernel.kernelTimeMs = await kernel.kernelTimeMs;\n      kernel.extraInfo = await kernel.extraInfo;\n    }\n    return this.state.activeProfile;\n  }\n\n  isTapeOn(): boolean {\n    return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n  }\n\n  private addTapeNode(\n      kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n      gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n    const tapeNode: TapeNode =\n        {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      gradientsFunc = gradConfig.gradFunc;\n    }\n    if (gradientsFunc != null) {\n      tapeNode.gradient = (dys: Tensor[]) => {\n        // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n        // the backprop graph to the user as null instead of zeros\n        dys = dys.map((dy, i) => {\n          if (dy == null) {\n            const output = outputs[i];\n            const vals = util.makeZerosTypedArray(output.size, output.dtype);\n            return this.makeTensor(vals, output.shape, output.dtype);\n          }\n          return dy;\n        });\n        // Grad functions of ops with single outputs expect a dy, while ops\n        // with multiple outputs expect dys (array of dy).\n        return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n      };\n    }\n    this.state.activeTape.push(tapeNode);\n  }\n\n  keep<T extends Tensor>(result: T): T {\n    result.kept = true;\n    return result;\n  }\n\n  private startTape() {\n    if (this.state.gradientDepth === 0) {\n      this.state.activeTape = [];\n    }\n    this.state.gradientDepth++;\n  }\n\n  private endTape() {\n    this.state.gradientDepth--;\n  }\n\n  /**\n   * Start a scope. Use this with endScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  startScope(name?: string) {\n    const scopeInfo: ScopeState = {\n      track: [],\n      name: 'unnamed scope',\n      id: this.state.nextScopeId++\n    };\n    if (name) {\n      scopeInfo.name = name;\n    }\n    this.state.scopeStack.push(scopeInfo);\n    this.state.activeScope = scopeInfo;\n  }\n\n  /**\n   * End a scope. Use this with startScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  endScope(result?: TensorContainer) {\n    const tensorsToTrackInParent = getTensorsInContainer(result);\n    const tensorsToTrackInParentSet =\n        new Set(tensorsToTrackInParent.map(t => t.id));\n\n    // Dispose the arrays tracked in this scope.\n    for (let i = 0; i < this.state.activeScope.track.length; i++) {\n      const tensor = this.state.activeScope.track[i];\n      if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n        tensor.dispose();\n      }\n    }\n\n    const oldScope = this.state.scopeStack.pop();\n    this.state.activeScope = this.state.scopeStack.length === 0 ?\n        null :\n        this.state.scopeStack[this.state.scopeStack.length - 1];\n\n    // Track the current result in the parent scope.\n    tensorsToTrackInParent.forEach(tensor => {\n      // Only track the tensor if was allocated in the inner scope and is not\n      // globally kept.\n      if (!tensor.kept && tensor.scopeId === oldScope.id) {\n        this.track(tensor);\n      }\n    });\n  }\n\n  /**\n   * Returns gradients of `f` with respect to each of the `xs`. The gradients\n   * returned are of the same length as `xs`, but some might be null if `f`\n   * was not a function of that `x`. It also takes optional dy to multiply the\n   * gradient, which defaults to `1`.\n   */\n  gradients<T extends Tensor>(\n      f: () => T, xs: Tensor[], dy?: T,\n      allowNoGradients = false): {value: T, grads: Tensor[]} {\n    util.assert(\n        xs.length > 0, () => 'gradients() received an empty list of xs.');\n    if (dy != null && dy.dtype !== 'float32') {\n      throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n    }\n\n    const y = this.scopedRun(\n        () => this.startTape(), () => this.endTape(),\n        () => this.tidy('forward', f));\n\n    util.assert(\n        y instanceof Tensor,\n        () => 'The result y returned by f() must be a tensor.');\n    // Filter out the nodes that don't connect x => y.\n    const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n    if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n      throw new Error(\n          'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n          'that the f you passed encloses all operations that lead from x ' +\n          'to y.');\n    }\n\n    return this.tidy('backward', () => {\n      const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n      accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n      // Backprop gradients through the filtered nodes.\n      backpropagateGradients(\n          accumulatedGradientMap, filteredTape,\n          // Pass the tidy function to avoid circular dep with `tape.ts`.\n          f => this.tidy(f as ScopeFn<Tensor>),\n          // Pass an add function to avoide a circular dep with `tape.ts`.\n          add);\n      const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n      if (this.state.gradientDepth === 0) {\n        // This means that we are not computing higher-order gradients\n        // and can clean up the tape.\n        this.state.activeTape.forEach(node => {\n          for (const tensor of node.saved) {\n            tensor.dispose();\n          }\n        });\n        this.state.activeTape = null;\n      }\n      return {value: y, grads};\n    });\n  }\n\n  customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n      (...args: Array<Tensor|GradSaveFunc>) => T {\n    util.assert(\n        util.isFunction(f),\n        () => 'The f passed in customGrad(f) must be a function.');\n    return (...inputs: Tensor[]): T => {\n      util.assert(\n          inputs.every(t => t instanceof Tensor),\n          () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n              'tensors');\n\n      let res: {\n        value: T,\n        gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n      };\n      const inputMap: NamedTensorMap = {};\n      inputs.forEach((input, i) => {\n        inputMap[i] = input;\n      });\n      return this.runKernelFunc(\n          (_, save) => {\n            res = f(...[...inputs, save]);\n            util.assert(\n                res.value instanceof Tensor,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.value` is a tensor');\n            util.assert(\n                util.isFunction(res.gradFunc),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function.');\n            return res.value;\n          },\n          inputMap,\n          (dy: T, saved: Tensor[]) => {\n            const gradRes = res.gradFunc(dy, saved);\n            const grads: Tensor[] =\n                Array.isArray(gradRes) ? gradRes : [gradRes];\n            util.assert(\n                grads.length === inputs.length,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'the same number of tensors as inputs passed to f(...).');\n            util.assert(\n                grads.every(t => t instanceof Tensor),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'a list of only tensors.');\n            const gradMap: {[key: string]: () => Tensor} = {};\n            grads.forEach((grad, i) => {\n              gradMap[i] = () => grad;\n            });\n            return gradMap;\n          });\n    };\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.readSync(dataId);\n  }\n  read(dataId: DataId): Promise<BackendValues> {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.read(dataId);\n  }\n\n  async time(query: () => void): Promise<TimingInfo> {\n    const start = now();\n    const timingInfo = await this.backend.time(query) as TimingInfo;\n    timingInfo.wallMs = now() - start;\n    return timingInfo;\n  }\n\n  /**\n   * Tracks a Tensor in the current scope to be automatically cleaned up\n   * when the current scope ends, and returns the value.\n   *\n   * @param result The Tensor to track in the current scope.\n   */\n  private track<T extends Tensor>(result: T): T {\n    if (this.state.activeScope != null) {\n      result.scopeId = this.state.activeScope.id;\n      this.state.activeScope.track.push(result);\n    }\n\n    return result;\n  }\n\n  get registeredVariables(): NamedVariableMap {\n    return this.state.registeredVariables;\n  }\n\n  /**\n   * Resets the engine state. Removes all backends but does not remove\n   * registered backend factories.\n   */\n  reset(): void {\n    // Make any pending promise obsolete.\n    this.pendingBackendInitId++;\n\n    this.state.dispose();\n    this.ENV.reset();\n    this.state = new EngineState();\n\n    for (const backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n    this.backendName = null;\n    this.backendInstance = null;\n    this.pendingBackendInit = null;\n  }\n}\n\nfunction ones(shape: number[]): Tensor {\n  const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n  return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n  const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n  if (ns._tfengine == null) {\n    const environment = new Environment(ns);\n    ns._tfengine = new Engine(environment);\n  }\n  setEnvironmentGlobal(ns._tfengine.ENV);\n\n  // Tell the current tensor interface that the global engine is responsible\n  // for tracking.\n  setTensorTracker(() => ns._tfengine);\n  return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n  // We duplicate Add here to avoid a circular dependency with add.ts.\n  const inputs = {a, b};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.add(a, b);\n    save([a, b]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n  return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n  if (_isNavigatorDefined()) {\n    // tslint:disable-next-line:no-any\n    const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n    // tslint:disable-next-line:max-line-length\n    return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n               .test(a) ||\n        // tslint:disable-next-line:max-line-length\n        /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n            .test(a.substr(0, 4));\n  }\n  return false;\n}\n\nexport function isBrowser(): boolean {\n  return (typeof window !== 'undefined' && window.document != null) ||\n      //@ts-ignore\n      (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n  if (debugValue) {\n    console.warn(\n        'Debugging mode is ON. The output of every math call will ' +\n        'be downloaded to CPU and checked for NaNs. ' +\n        'This significantly impacts performance.');\n  }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n    'IS_NODE',\n    () => (typeof process !== 'undefined') &&\n        (typeof process.versions !== 'undefined') &&\n        (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n    'IS_CHROME',\n    () => typeof navigator !== 'undefined' && navigator != null &&\n        navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n        /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n    'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n  let firstElem: typeof val = val;\n\n  if (isTypedArray(val)) {\n    return dtype === 'string' ? [] : [val.length];\n  }\n  if (!Array.isArray(val)) {\n    return [];  // Scalar.\n  }\n  const shape: number[] = [];\n\n  while (Array.isArray(firstElem) ||\n         isTypedArray(firstElem) && dtype !== 'string') {\n    shape.push(firstElem.length);\n    firstElem = firstElem[0];\n  }\n  if (Array.isArray(val) &&\n      env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n    deepAssertShapeConsistency(val, shape, []);\n  }\n\n  return shape;\n}\n\nfunction deepAssertShapeConsistency(\n    val: TensorLike, shape: number[], indices: number[]) {\n  indices = indices || [];\n  if (!(Array.isArray(val)) && !isTypedArray(val)) {\n    assert(\n        shape.length === 0,\n        () => `Element arr[${indices.join('][')}] is a primitive, ` +\n            `but should be an array/TypedArray of ${shape[0]} elements`);\n    return;\n  }\n  assert(\n      shape.length > 0,\n      () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n          `but is an array of ${val.length} elements`);\n  assert(\n      val.length === shape[0],\n      () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n          `elements, but has ${val.length} elements`);\n  const subShape = shape.slice(1);\n  for (let i = 0; i < val.length; ++i) {\n    deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n  }\n}\n\nfunction assertDtype(\n    expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n    functionName: string) {\n  if (expectedDtype == null) {\n    return;\n  }\n  if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n      expectedDtype === 'numeric' && actualDType === 'string') {\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must ` +\n        `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n  }\n}\n\nexport function convertToTensor<T extends Tensor>(\n    x: T|TensorLike, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T {\n  if (x instanceof Tensor) {\n    assertDtype(parseAsDtype, x.dtype, argName, functionName);\n    return x;\n  }\n  let inferredDtype = inferDtype(x);\n  // If the user expects a bool/int/float, use that info to update the\n  // inferredDtype when it is not a string.\n  if (inferredDtype !== 'string' &&\n      ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n    inferredDtype = parseAsDtype as DataType;\n  }\n  assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n  if ((x == null) ||\n      (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n       typeof x !== 'boolean' && typeof x !== 'string')) {\n    const type = x == null ? 'null' : (x as {}).constructor.name;\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must be a ` +\n        `Tensor or TensorLike, but got '${type}'`);\n  }\n  const inferredShape = inferShape(x, inferredDtype);\n  if (!isTypedArray(x) && !Array.isArray(x)) {\n    x = [x] as number[];\n  }\n  const skipTypedArray = true;\n  const values = inferredDtype !== 'string' ?\n      toTypedArray(x, inferredDtype as DataType) :\n      flatten(x as string[], [], skipTypedArray) as string[];\n  return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray<T extends Tensor>(\n    arg: Array<T|TensorLike>, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n  if (!Array.isArray(arg)) {\n    throw new Error(\n        `Argument ${argName} passed to ${functionName} must be a ` +\n        '`Tensor[]` or `TensorLike[]`');\n  }\n  const tensors = arg as T[];\n  return tensors.map(\n      (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n      parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op<T extends Function>(f: {[name: string]: T}): T {\n  const keys = Object.keys(f);\n  if (keys.length !== 1) {\n    throw new Error(\n        `Please provide an object with a single key ` +\n        `(operation name) mapping to a function. Got an object with ` +\n        `${keys.length} keys.`);\n  }\n\n  let opName = keys[0];\n  const fn = f[opName];\n\n  // Strip the underscore from the end of the function name.\n  if (opName.endsWith('_')) {\n    opName = opName.substring(0, opName.length - 1);\n  }\n\n  // add an __op suffix to distinguish ops from kernels in tf.profile\n  opName = opName + OP_SCOPE_SUFFIX;\n\n  // tslint:disable-next-line:no-any\n  const f2 = (...args: any[]) => {\n    ENGINE.startScope(opName);\n    try {\n      const result = fn(...args);\n      if (result instanceof Promise) {\n        console.error('Cannot return a Promise inside of tidy.');\n      }\n      ENGINE.endScope(result);\n      return result;\n    } catch (ex) {\n      ENGINE.endScope(null);\n      throw ex;\n    }\n  };\n  Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n  // tslint:disable-next-line:no-any\n  return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_<T extends Tensor>(real: T|TensorLike, imag: T|TensorLike): T {\n  const $real = convertToTensor(real, 'real', 'complex');\n  const $imag = convertToTensor(imag, 'imag', 'complex');\n  util.assertShapesMatch(\n      $real.shape, $imag.shape,\n      `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n          `must match in call to tf.complex().`);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.complex($real, $imag);\n  };\n  const inputs: ComplexInputs = {real: $real, imag: $imag};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n    values: TensorLike, shape: number[], inferredShape: number[],\n    dtype?: DataType): Tensor {\n  if (dtype == null) {\n    dtype = inferDtype(values);\n  }\n  if (dtype === 'complex64') {\n    throw new Error(\n        `Cannot construct a complex64 tensor directly. ` +\n        `Please use tf.complex(real, imag).`);\n  }\n  if (!isTypedArray(values) && !Array.isArray(values) &&\n      typeof values !== 'number' && typeof values !== 'boolean' &&\n      typeof values !== 'string') {\n    throw new Error(\n        'values passed to tensor(values) must be a number/boolean/string or ' +\n        'an array of numbers/booleans/strings, or a TypedArray');\n  }\n  if (shape != null) {\n    assertNonNegativeIntegerDimensions(shape);\n\n    const providedSize = sizeFromShape(shape);\n    const inferredSize = sizeFromShape(inferredShape);\n    assert(\n        providedSize === inferredSize,\n        () =>\n            `Based on the provided shape, [${shape}], the tensor should have ` +\n            `${providedSize} values but has ${inferredSize}`);\n\n    for (let i = 0; i < inferredShape.length; ++i) {\n      const inferred = inferredShape[i];\n      const flatDimsDontMatch = i === inferredShape.length - 1 ?\n          inferred !== sizeFromShape(shape.slice(i)) :\n          true;\n      assert(\n          inferredShape[i] === shape[i] || !flatDimsDontMatch,\n          () => `Error creating a new Tensor. Inferred shape ` +\n              `(${inferredShape}) does not match the provided ` +\n              `shape (${shape}). `);\n    }\n  }\n\n  if (!isTypedArray(values) && !Array.isArray(values)) {\n    values = [values] as number[];\n  }\n\n  shape = shape || inferredShape;\n  values = dtype !== 'string' ?\n      toTypedArray(values, dtype) :\n      flatten(values as string[], [], true) as string[];\n  return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`. If the values are strings,\n *     they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor<R extends Rank>(\n    values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor<R> {\n  const inferredShape = inferShape(values, dtype);\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n  'float32': 4,\n  'float16': 2,\n  'int32': 4,\n  'uint16': 2,\n  'uint8': 1,\n  'bool': 1,\n  'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n  /**\n   * An ordered list of paths.\n   *\n   * Paths are intentionally abstract in order to be general. For example, they\n   * can be relative URL paths or relative paths on the file system.\n   */\n  paths: string[];\n\n  /**\n   * Specifications of the weights stored in the paths.\n   */\n  weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n  /**\n   * Name of the weight, e.g., 'Dense_1/bias'\n   */\n  name: string;\n\n  /**\n   * Shape of the weight.\n   */\n  shape: number[];\n\n  /**\n   * Data type of the weight.\n   */\n  dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n  /**\n   * Type of the weight.\n   *\n   * Optional.\n   *\n   * The value 'optimizer' indicates the weight belongs to an optimizer\n   * (i.e., used only during model training and not during inference).\n   */\n  group?: WeightGroup;\n\n  /**\n   * Information for dequantization of the weight.\n   */\n  quantization?: {\n    scale?: number,  // The scaling constant to multiply by.\n    min?: number,    // The (possibly nudged) minimum weight to add.\n       dtype: 'uint16'|'uint8'|'float16'  // The dtype of the quantized weights.\n  };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n  /**\n   * Whether to save only the trainable weights of the model, ignoring the\n   * non-trainable ones.\n   */\n  trainableOnly?: boolean;\n\n  /**\n   * Whether the optimizer will be saved (if exists).\n   *\n   * Default: `false`.\n   */\n  includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n  /**\n   * Information about the model artifacts saved.\n   */\n  modelArtifactsInfo: ModelArtifactsInfo;\n\n  /**\n   * HTTP responses from the server that handled the model-saving request (if\n   * any). This is applicable only to server-based saving routes.\n   */\n  responses?: Response[];\n\n  /**\n   * Error messages and related data (if any).\n   */\n  errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n  /**\n   * Timestamp for when the model is saved.\n   */\n  dateSaved: Date;\n\n  /**\n   * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n   * come in a JSON format and none of our IOHandlers support a non json\n   * format. We could conder replacing this with 'Binary' if we want to\n   * allow future handlers to save to non json formats (though they will\n   * probably want more information than 'Binary').\n   * Type of the model topology\n   *\n   * Type of the model topology\n   *\n   * Possible values:\n   *   - JSON: JSON config (human-readable, e.g., Keras JSON).\n   *   - GraphDef: TensorFlow\n   *     [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n   *     protocol buffer (binary).\n   */\n  modelTopologyType: 'JSON'|'GraphDef';\n\n  /**\n   * Size of model topology (Keras JSON or GraphDef), in bytes.\n   */\n  modelTopologyBytes?: number;\n\n  /**\n   * Size of weight specification or manifest, in bytes.\n   */\n  weightSpecsBytes?: number;\n\n  /**\n   * Size of weight value data, in bytes.\n   */\n  weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  // See\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n  /** Optimizer used for the model training. */\n  optimizer_config: {};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Loss function(s) for the model's output(s). */\n  loss: string|string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Metric function(s) for the model's output(s). */\n  metrics?: string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  weighted_metrics?: string[];\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  sample_weight_mode?: string;\n\n  loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers.  For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology?: {}|ArrayBuffer;\n\n  /**\n   * Serialized configuration for the model's training.\n   */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weight specifications.\n   *\n   * This corresponds to the weightsData below.\n   */\n  weightSpecs?: WeightsManifestEntry[];\n\n  /**\n   * Binary buffer for all weight values concatenated in the order specified\n   * by `weightSpecs`.\n   */\n  weightData?: ArrayBuffer;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology: {};\n\n  /** Model training configuration. */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weights manifest.\n   *\n   * The weights manifest consists of an ordered list of weight-manifest\n   * groups. Each weight-manifest group consists of a number of weight values\n   * stored in a number of paths. See the documentation of\n   * `WeightsManifestConfig` for more details.\n   */\n  weightsManifest: WeightsManifestConfig;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise<ModelArtifacts>;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n    Promise<SaveResult>;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n  save?: SaveHandler;\n  load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n  /**\n   * List all models in the model store.\n   *\n   * @returns A dictionary mapping paths of existing models to their\n   *   model artifacts info. Model artifacts info include type of the model's\n   *   topology, byte sizes of the topology, weights, etc.\n   */\n  listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n  /**\n   * Remove a model specified by `path`.\n   *\n   * @param path\n   * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n   *   is successful).\n   * @throws Error if deletion fails, e.g., if no model exists at `path`.\n   */\n  removeModel(path: string): Promise<ModelArtifactsInfo>;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n  /**\n   * RequestInit (options) for HTTP requests.\n   *\n   * For detailed information on the supported fields, see\n   * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n   *     https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n   */\n  requestInit?: RequestInit;\n\n  /**\n   * Progress callback.\n   */\n  onProgress?: OnProgressCallback;\n\n  /**\n   * A function used to override the `window.fetch` function.\n   */\n  fetchFunc?: Function;\n\n  /**\n   * Strict loading model: whether extraneous weights or missing\n   * weights should trigger an `Error`.\n   *\n   * If `true`, require that the provided weights exactly match those\n   * required by the layers. `false` means that both extra weights\n   * and missing weights will be silently ignored.\n   *\n   * Default: `true`.\n   */\n  strict?: boolean;\n\n  /**\n   * Path prefix for weight files, by default this is calculated from the\n   * path of the model JSON file.\n   *\n   * For instance, if the path to the model JSON file is\n   * `http://localhost/foo/model.json`, then the default path prefix will be\n   * `http://localhost/foo/`. If a weight file has the path value\n   * `group1-shard1of2` in the weight manifest, then the weight file will be\n   * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n   * if you provide a `weightPathPrefix` value of\n   * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n   * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n   */\n  weightPathPrefix?: string;\n\n  /**\n   * Whether the module or model is to be loaded from TF Hub.\n   *\n   * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n   * standard model file name and the query parameters.\n   *\n   * Default: `false`.\n   */\n  fromTFHub?: boolean;\n\n  /**\n   * An async function to convert weight file name to URL. The weight file\n   * names are stored in model.json's weightsManifest.paths field. By default we\n   * consider weight files are colocated with the model.json file. For example:\n   *     model.json URL: https://www.google.com/models/1/model.json\n   *     group1-shard1of1.bin url:\n   *        https://www.google.com/models/1/group1-shard1of1.bin\n   *\n   * With this func you can convert the weight file name to any URL.\n   */\n  weightUrlConverter?: (weightFileName: string) => Promise<string>;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n  /**\n   * Is this request for a binary file (as opposed to a json file)\n   */\n  isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n *     concatenated.\n *   - An `Array` of `WeightManifestEntry`s, carrying information including\n *     tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n    tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n    Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n  // TODO(adarob, cais): Support quantization.\n  const specs: WeightsManifestEntry[] = [];\n  const dataPromises: Array<Promise<TypedArray>> = [];\n\n  const names: string[] = Array.isArray(tensors) ?\n      tensors.map(tensor => tensor.name) :\n      Object.keys(tensors);\n\n  for (let i = 0; i < names.length; ++i) {\n    const name = names[i];\n    const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n    if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n        t.dtype !== 'string' && t.dtype !== 'complex64') {\n      throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n    }\n    const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n    if (t.dtype === 'string') {\n      const utf8bytes = new Promise<TypedArray>(async resolve => {\n        const vals = await t.bytes() as Uint8Array[];\n        const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n            NUM_BYTES_STRING_LENGTH * vals.length;\n        const bytes = new Uint8Array(totalNumBytes);\n        let offset = 0;\n        for (let i = 0; i < vals.length; i++) {\n          const val = vals[i];\n          const bytesOfLength =\n              new Uint8Array(new Uint32Array([val.length]).buffer);\n          bytes.set(bytesOfLength, offset);\n          offset += NUM_BYTES_STRING_LENGTH;\n          bytes.set(val, offset);\n          offset += val.length;\n        }\n        resolve(bytes);\n      });\n      dataPromises.push(utf8bytes);\n    } else {\n      dataPromises.push(t.data());\n    }\n    if (group != null) {\n      spec.group = group;\n    }\n    specs.push(spec);\n  }\n\n  const tensorValues = await Promise.all(dataPromises);\n  return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n *   concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n *   whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n *   to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n    buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n  // TODO(adarob, cais): Support quantization.\n  const out: NamedTensorMap = {};\n  let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n  let offset = 0;\n  for (const spec of specs) {\n    const name = spec.name;\n    const dtype = spec.dtype;\n    const shape = spec.shape;\n    const size = sizeFromShape(shape);\n    let values: TypedArray|string[]|Uint8Array[];\n\n    if ('quantization' in spec) {\n      const quantization = spec.quantization;\n      if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n        if (!('min' in quantization && 'scale' in quantization)) {\n          throw new Error(\n              `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n              `doesn't have corresponding metadata min and scale.`);\n        }\n      } else if (quantization.dtype === 'float16') {\n        if (dtype !== 'float32') {\n          throw new Error(\n              `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n              `which only supports weights of type float32 not ${dtype}.`);\n        }\n      } else {\n        throw new Error(\n            `Weight ${spec.name} has unknown ` +\n            `quantization dtype ${quantization.dtype}. ` +\n            `Supported quantization dtypes are: ` +\n            `'uint8', 'uint16', and 'float16'.`);\n      }\n      const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n      const byteBuffer =\n          buffer.slice(offset, offset + size * quantizationSizeFactor);\n      const quantizedArray = (quantization.dtype === 'uint8') ?\n          new Uint8Array(byteBuffer) :\n          new Uint16Array(byteBuffer);\n      if (dtype === 'float32') {\n        if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n          values = new Float32Array(quantizedArray.length);\n          for (let i = 0; i < quantizedArray.length; i++) {\n            const v = quantizedArray[i];\n            values[i] = v * quantization.scale + quantization.min;\n          }\n        } else if (quantization.dtype === 'float16') {\n          if (float16Decode === undefined) {\n            float16Decode = getFloat16Decoder();\n          }\n          values = float16Decode(quantizedArray as Uint16Array);\n        } else {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type float32.`);\n        }\n      } else if (dtype === 'int32') {\n        if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type int32.`);\n        }\n        values = new Int32Array(quantizedArray.length);\n        for (let i = 0; i < quantizedArray.length; i++) {\n          const v = quantizedArray[i];\n          values[i] = Math.round(v * quantization.scale + quantization.min);\n        }\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * quantizationSizeFactor;\n    } else if (dtype === 'string') {\n      const size = sizeFromShape(spec.shape);\n      values = [];\n      for (let i = 0; i < size; i++) {\n        const byteLength = new Uint32Array(\n            buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n        offset += NUM_BYTES_STRING_LENGTH;\n        const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n        (values as Uint8Array[]).push(bytes);\n        offset += byteLength;\n      }\n    } else {\n      const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n      const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n      if (dtype === 'float32') {\n        values = new Float32Array(byteBuffer);\n      } else if (dtype === 'int32') {\n        values = new Int32Array(byteBuffer);\n      } else if (dtype === 'bool') {\n        values = new Uint8Array(byteBuffer);\n      } else if (dtype === 'complex64') {\n        values = new Float32Array(byteBuffer);\n        const real = new Float32Array(values.length / 2);\n        const image = new Float32Array(values.length / 2);\n        for (let i = 0; i < real.length; i++) {\n          real[i] = values[i * 2];\n          image[i] = values[i * 2 + 1];\n        }\n        const realTensor = tensor(real, shape, 'float32');\n        const imageTensor = tensor(image, shape, 'float32');\n        out[name] = complex(realTensor, imageTensor);\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * dtypeFactor;\n    }\n    if (dtype !== 'complex64') {\n      out[name] = tensor(values, shape, dtype);\n    }\n  }\n  return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n  // TODO(adarob, cais): Support quantization.\n  if (xs === null) {\n    throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n  }\n\n  let totalByteLength = 0;\n\n  // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n  // can have a different byte length from that of the `TypedArray` itself,\n  // for example, when the `TypedArray` is created from an offset in an\n  // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n  // the `TypedArray` in byte length. If an element of `xs` does not show\n  // this property, a new `TypedArray` that satisfy this property will be\n  // constructed and pushed into `normalizedXs`.\n  const normalizedXs: TypedArray[] = [];\n  xs.forEach((x: TypedArray) => {\n    totalByteLength += x.byteLength;\n    // tslint:disable:no-any\n    normalizedXs.push(\n        x.byteLength === x.buffer.byteLength ? x :\n                                               new (x.constructor as any)(x));\n    if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n          x as any instanceof Uint8Array)) {\n      throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n    }\n    // tslint:enable:no-any\n  });\n\n  const y = new Uint8Array(totalByteLength);\n  let offset = 0;\n  normalizedXs.forEach((x: TypedArray) => {\n    y.set(new Uint8Array(x.buffer), offset);\n    offset += x.byteLength;\n  });\n\n  return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n    (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n     typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n  if (useNodeBuffer) {\n    return Buffer.byteLength(str);\n  }\n  return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n  if (useNodeBuffer) {\n    return Buffer.from(buffer).toString('base64');\n  }\n  const buf = new Uint8Array(buffer);\n  let s = '';\n  for (let i = 0, l = buf.length; i < l; i++) {\n    s += String.fromCharCode(buf[i]);\n  }\n  return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n  if (useNodeBuffer) {\n    const buf = Buffer.from(str, 'base64');\n    return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n  }\n  const s = atob(str);\n  const buffer = new Uint8Array(s.length);\n  for (let i = 0; i < s.length; ++i) {\n    buffer.set([s.charCodeAt(i)], i);\n  }\n  return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n  if (buffers.length === 1) {\n    return buffers[0];\n  }\n\n  let totalByteLength = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    totalByteLength += buffer.byteLength;\n  });\n\n  const temp = new Uint8Array(totalByteLength);\n  let offset = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    temp.set(new Uint8Array(buffer), offset);\n    offset += buffer.byteLength;\n  });\n  return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n  const SEPARATOR = '/';\n  path = path.trim();\n  while (path.endsWith(SEPARATOR)) {\n    path = path.slice(0, path.length - 1);\n  }\n  const items = path.split(SEPARATOR);\n  return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n    ModelArtifactsInfo {\n  if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n    throw new Error('Expected JSON model topology, received ArrayBuffer.');\n  }\n\n  return {\n    dateSaved: new Date(),\n    modelTopologyType: 'JSON',\n    modelTopologyBytes: modelArtifacts.modelTopology == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n    weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n    weightDataBytes: modelArtifacts.weightData == null ?\n        0 :\n        modelArtifacts.weightData.byteLength,\n  };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n  const convertMantissa = (i: number): number => {\n    let m = i << 13;\n    let e = 0;\n\n    while ((m & 0x00800000) === 0) {\n      e -= 0x00800000;\n      m <<= 1;\n    }\n    m &= ~0x00800000;\n    e += 0x38800000;\n\n    return m | e;\n  };\n\n  const mantisaTable = new Uint32Array(2048);\n\n  mantisaTable[0] = 0;\n  for (let i = 1; i < 1024; i++) {\n    mantisaTable[i] = convertMantissa(i);\n  }\n  for (let i = 1024; i < 2048; i++) {\n    mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n  }\n\n  return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n  const exponentTable = new Uint32Array(64);\n\n  exponentTable[0] = 0;\n  exponentTable[31] = 0x47800000;\n  exponentTable[32] = 0x80000000;\n  exponentTable[63] = 0xc7800000;\n  for (let i = 1; i < 31; i++) {\n    exponentTable[i] = i << 23;\n  }\n  for (let i = 33; i < 63; i++) {\n    exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n  }\n\n  return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n  const offsetTable = new Uint32Array(64);\n\n  for (let i = 0; i < 64; i++) {\n    offsetTable[i] = 1024;\n  }\n  offsetTable[0] = offsetTable[32] = 0;\n\n  return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n *          the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n  // Algorithm is based off of\n  // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n  // Cache lookup tables\n  const mantisaTable = computeFloat16MantisaTable();\n  const exponentTable = computeFloat16ExponentTable();\n  const offsetTable = computeFloat16OffsetTable();\n\n  return (quantizedArray: Uint16Array) => {\n    const buffer = new ArrayBuffer(4 * quantizedArray.length);\n    const bufferUint32View = new Uint32Array(buffer);\n    for (let index = 0; index < quantizedArray.length; index++) {\n      const float16Bits = quantizedArray[index];\n      const float32Bits =\n          mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n          exponentTable[float16Bits >> 10];\n      bufferUint32View[index] = float32Bits;\n    }\n    return new Float32Array(buffer);\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n    IOHandler;\n\nexport class IORouterRegistry {\n  // Singleton instance.\n  private static instance: IORouterRegistry;\n\n  private saveRouters: IORouter[];\n  private loadRouters: IORouter[];\n\n  private constructor() {\n    this.saveRouters = [];\n    this.loadRouters = [];\n  }\n\n  private static getInstance(): IORouterRegistry {\n    if (IORouterRegistry.instance == null) {\n      IORouterRegistry.instance = new IORouterRegistry();\n    }\n    return IORouterRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerSaveRouter(saveRouter: IORouter) {\n    IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n  }\n\n  /**\n   * Register a load-handler router.\n   *\n   * @param loadRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `load` method defined or `null`.\n   */\n  static registerLoadRouter(loadRouter: IORouter) {\n    IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n  }\n\n  /**\n   * Look up IOHandler for saving, given a URL-like string.\n   *\n   * @param url\n   * @returns If only one match is found, an instance of IOHandler with the\n   * `save` method defined. If no match is found, `null`.\n   * @throws Error, if more than one match is found.\n   */\n  static getSaveHandlers(url: string|string[]): IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'save');\n  }\n\n  /**\n   * Look up IOHandler for loading, given a URL-like string.\n   *\n   * @param url\n   * @param loadOptions Optional, custom load options.\n   * @returns All valid handlers for `url`, given the currently registered\n   *   handler routers.\n   */\n  static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n      IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n  }\n\n  private static getHandlers(\n      url: string|string[], handlerType: 'save'|'load',\n      loadOptions?: LoadOptions): IOHandler[] {\n    const validHandlers: IOHandler[] = [];\n    const routers = handlerType === 'load' ?\n        IORouterRegistry.getInstance().loadRouters :\n        IORouterRegistry.getInstance().saveRouters;\n    routers.forEach(router => {\n      const handler = router(url, loadOptions);\n      if (handler !== null) {\n        validHandlers.push(handler);\n      }\n    });\n    return validHandlers;\n  }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n    IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n    (url: string|string[], loadOptions?: LoadOptions) =>\n        IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n//    as the type of topology (JSON vs binary), byte size of the topology, byte\n//    size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise<void> {\n  const idbFactory = getIndexedDBFactory();\n\n  return new Promise<void>((resolve, reject) => {\n    const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n    deleteRequest.onsuccess = () => resolve();\n    deleteRequest.onerror = error => reject(error);\n  });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n  if (!env().getBool('IS_BROWSER')) {\n    // TODO(cais): Add more info about what IOHandler subtypes are available.\n    //   Maybe point to a doc page on the web and/or automatically determine\n    //   the available IOHandlers and print them in the error message.\n    throw new Error(\n        'Failed to obtain IndexedDB factory because the current environment' +\n        'is not a web browser.');\n  }\n  // tslint:disable-next-line:no-any\n  const theWindow: any = typeof window === 'undefined' ? self : window;\n  const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n      theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n      theWindow.shimIndexedDB;\n  if (factory == null) {\n    throw new Error(\n        'The current browser does not appear to support IndexedDB.');\n  }\n  return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n  const db = openRequest.result as IDBDatabase;\n  db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n  db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n  protected readonly indexedDB: IDBFactory;\n  protected readonly modelPath: string;\n\n  static readonly URL_SCHEME = 'indexeddb://';\n\n  constructor(modelPath: string) {\n    this.indexedDB = getIndexedDBFactory();\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For IndexedDB, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    // TODO(cais): Support saving GraphDef models.\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    return this.databaseAction(this.modelPath, modelArtifacts) as\n        Promise<SaveResult>;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    return this.databaseAction(this.modelPath) as Promise<ModelArtifacts>;\n  }\n\n  /**\n   * Perform database action to put model artifacts into or read model artifacts\n   * from IndexedDB object store.\n   *\n   * Whether the action is put or get depends on whether `modelArtifacts` is\n   * specified. If it is specified, the action will be put; otherwise the action\n   * will be get.\n   *\n   * @param modelPath A unique string path for the model.\n   * @param modelArtifacts If specified, it will be the model artifacts to be\n   *   stored in IndexedDB.\n   * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n   *   of `ModelArtifacts`, if the action is get.\n   */\n  private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n      Promise<ModelArtifacts|SaveResult> {\n    return new Promise<ModelArtifacts|SaveResult>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n\n        if (modelArtifacts == null) {\n          // Read model out from object store.\n          const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n          const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n          const getRequest = modelStore.get(this.modelPath);\n          getRequest.onsuccess = () => {\n            if (getRequest.result == null) {\n              db.close();\n              return reject(new Error(\n                  `Cannot find model with path '${this.modelPath}' ` +\n                  `in IndexedDB.`));\n            } else {\n              resolve(getRequest.result.modelArtifacts);\n            }\n          };\n          getRequest.onerror = error => {\n            db.close();\n            return reject(getRequest.error);\n          };\n          modelTx.oncomplete = () => db.close();\n        } else {\n          // Put model into object store.\n          const modelArtifactsInfo: ModelArtifactsInfo =\n              getModelArtifactsInfoForJSON(modelArtifacts);\n          // First, put ModelArtifactsInfo into info store.\n          const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n          let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n          const putInfoRequest =\n              infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n          let modelTx: IDBTransaction;\n          putInfoRequest.onsuccess = () => {\n            // Second, put model data into model store.\n            modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n            const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n            const putModelRequest = modelStore.put({\n              modelPath: this.modelPath,\n              modelArtifacts,\n              modelArtifactsInfo\n            });\n            putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n            putModelRequest.onerror = error => {\n              // If the put-model request fails, roll back the info entry as\n              // well.\n              infoStore = infoTx.objectStore(INFO_STORE_NAME);\n              const deleteInfoRequest = infoStore.delete(this.modelPath);\n              deleteInfoRequest.onsuccess = () => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n              deleteInfoRequest.onerror = error => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n            };\n          };\n          putInfoRequest.onerror = error => {\n            db.close();\n            return reject(putInfoRequest.error);\n          };\n          infoTx.oncomplete = () => {\n            if (modelTx == null) {\n              db.close();\n            } else {\n              modelTx.oncomplete = () => db.close();\n            }\n          };\n        }\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n      return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n *   which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n  return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n      key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n      key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n  private indexedDB: IDBFactory;\n\n  constructor() {\n    this.indexedDB = getIndexedDBFactory();\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    return new Promise<{[path: string]: ModelArtifactsInfo}>(\n        (resolve, reject) => {\n          const openRequest =\n              this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n          openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n          openRequest.onsuccess = () => {\n            const db = openRequest.result;\n            const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n            const store = tx.objectStore(INFO_STORE_NAME);\n            // tslint:disable:max-line-length\n            // Need to cast `store` as `any` here because TypeScript's DOM\n            // library does not have the `getAll()` method even though the\n            // method is supported in the latest version of most mainstream\n            // browsers:\n            // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n            // tslint:enable:max-line-length\n            // tslint:disable-next-line:no-any\n            const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n            getAllInfoRequest.onsuccess = () => {\n              const out: {[path: string]: ModelArtifactsInfo} = {};\n              for (const item of getAllInfoRequest.result) {\n                out[item.modelPath] = item.modelArtifactsInfo;\n              }\n              resolve(out);\n            };\n            getAllInfoRequest.onerror = error => {\n              db.close();\n              return reject(getAllInfoRequest.error);\n            };\n            tx.oncomplete = () => db.close();\n          };\n          openRequest.onerror = error => reject(openRequest.error);\n        });\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    return new Promise<ModelArtifactsInfo>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n        const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n        const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n        const getInfoRequest = infoStore.get(path);\n        let modelTx: IDBTransaction;\n        getInfoRequest.onsuccess = () => {\n          if (getInfoRequest.result == null) {\n            db.close();\n            return reject(new Error(\n                `Cannot find model with path '${path}' ` +\n                `in IndexedDB.`));\n          } else {\n            // First, delete the entry in the info store.\n            const deleteInfoRequest = infoStore.delete(path);\n            const deleteModelData = () => {\n              // Second, delete the entry in the model store.\n              modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n              const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n              const deleteModelRequest = modelStore.delete(path);\n              deleteModelRequest.onsuccess = () =>\n                  resolve(getInfoRequest.result.modelArtifactsInfo);\n              deleteModelRequest.onerror = error =>\n                  reject(getInfoRequest.error);\n            };\n            // Proceed with deleting model data regardless of whether deletion\n            // of info data succeeds or not.\n            deleteInfoRequest.onsuccess = deleteModelData;\n            deleteInfoRequest.onerror = error => {\n              deleteModelData();\n              db.close();\n              return reject(getInfoRequest.error);\n            };\n          }\n        };\n        getInfoRequest.onerror = error => {\n          db.close();\n          return reject(getInfoRequest.error);\n        };\n\n        infoTx.oncomplete = () => {\n          if (modelTx == null) {\n            db.close();\n          } else {\n            modelTx.oncomplete = () => db.close();\n          }\n        };\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n  if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n      typeof window.localStorage === 'undefined') {\n    throw new Error(\n        'purgeLocalStorageModels() cannot proceed because local storage is ' +\n        'unavailable in the current environment.');\n  }\n  const LS = window.localStorage;\n  const purgedModelPaths: string[] = [];\n  for (let i = 0; i < LS.length; ++i) {\n    const key = LS.key(i);\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    if (key.startsWith(prefix) && key.length > prefix.length) {\n      LS.removeItem(key);\n      const modelName = getModelPathFromKey(key);\n      if (purgedModelPaths.indexOf(modelName) === -1) {\n        purgedModelPaths.push(modelName);\n      }\n    }\n  }\n  return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n} {\n  return {\n    info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n    topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n    weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n    weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n    modelMetadata:\n        [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n  };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n  const items = key.split(PATH_SEPARATOR);\n  if (items.length < 3) {\n    throw new Error(`Invalid key format: ${key}`);\n  }\n  return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n      key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n      key;\n}\n\ndeclare type LocalStorageKeys = {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n  protected readonly LS: Storage;\n  protected readonly modelPath: string;\n  protected readonly keys: LocalStorageKeys;\n\n  static readonly URL_SCHEME = 'localstorage://';\n\n  constructor(modelPath: string) {\n    if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n        typeof window.localStorage === 'undefined') {\n      // TODO(cais): Add more info about what IOHandler subtypes are\n      // available.\n      //   Maybe point to a doc page on the web and/or automatically determine\n      //   the available IOHandlers and print them in the error message.\n      throw new Error(\n          'The current environment does not support local storage.');\n    }\n    this.LS = window.localStorage;\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For local storage, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n    this.keys = getModelKeys(this.modelPath);\n  }\n\n  /**\n   * Save model artifacts to browser local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @param modelArtifacts The model artifacts to be stored.\n   * @returns An instance of SaveResult.\n   */\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const topology = JSON.stringify(modelArtifacts.modelTopology);\n      const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n      const modelArtifactsInfo: ModelArtifactsInfo =\n          getModelArtifactsInfoForJSON(modelArtifacts);\n\n      try {\n        this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n        this.LS.setItem(this.keys.topology, topology);\n        this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n        this.LS.setItem(\n            this.keys.weightData,\n            arrayBufferToBase64String(modelArtifacts.weightData));\n        this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n          format: modelArtifacts.format,\n          generatedBy: modelArtifacts.generatedBy,\n          convertedBy: modelArtifacts.convertedBy,\n          userDefinedMetadata: modelArtifacts.userDefinedMetadata\n        }));\n\n        return {modelArtifactsInfo};\n      } catch (err) {\n        // If saving failed, clean up all items saved so far.\n        this.LS.removeItem(this.keys.info);\n        this.LS.removeItem(this.keys.topology);\n        this.LS.removeItem(this.keys.weightSpecs);\n        this.LS.removeItem(this.keys.weightData);\n        this.LS.removeItem(this.keys.modelMetadata);\n\n        throw new Error(\n            `Failed to save model '${this.modelPath}' to local storage: ` +\n            `size quota being exceeded is a possible cause of this failure: ` +\n            `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n            `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n            `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n      }\n    }\n  }\n\n  /**\n   * Load a model from local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const info =\n        JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n    if (info == null) {\n      throw new Error(\n          `In local storage, there is no model with name '${this.modelPath}'`);\n    }\n\n    if (info.modelTopologyType !== 'JSON') {\n      throw new Error(\n          'BrowserLocalStorage does not support loading non-JSON model ' +\n          'topology yet.');\n    }\n\n    const out: ModelArtifacts = {};\n\n    // Load topology.\n    const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n    if (topology == null) {\n      throw new Error(\n          `In local storage, the topology of model '${this.modelPath}' ` +\n          `is missing.`);\n    }\n    out.modelTopology = topology;\n\n    // Load weight specs.\n    const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n    if (weightSpecs == null) {\n      throw new Error(\n          `In local storage, the weight specs of model '${this.modelPath}' ` +\n          `are missing.`);\n    }\n    out.weightSpecs = weightSpecs;\n\n    // Load meta-data fields.\n    const metadataString = this.LS.getItem(this.keys.modelMetadata);\n    if (metadataString != null) {\n      const metadata = JSON.parse(metadataString) as ModelArtifacts;\n      out.format = metadata['format'];\n      out.generatedBy = metadata['generatedBy'];\n      out.convertedBy = metadata['convertedBy'];\n      out.userDefinedMetadata = metadata['userDefinedMetadata'];\n    }\n\n    // Load weight data.\n    const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n    if (weightDataBase64 == null) {\n      throw new Error(\n          `In local storage, the binary weight values of model ` +\n          `'${this.modelPath}' are missing.`);\n    }\n    out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n    return out;\n  }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n      return browserLocalStorage(\n          url.slice(BrowserLocalStorage.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n *   - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n *     model, such as date saved, type of the topology, size in bytes, etc.\n *   - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n *     style models, this is a stringized JSON.\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n *     model, can be used to decode the saved binary weight values (see\n *     item below).\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n *     weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n *   `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n  return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n  private readonly LS: Storage;\n\n  constructor() {\n    assert(\n        env().getBool('IS_BROWSER'),\n        () => 'Current environment is not a web browser');\n    assert(\n        typeof window === 'undefined' ||\n            typeof window.localStorage !== 'undefined',\n        () => 'Current browser does not appear to support localStorage');\n    this.LS = window.localStorage;\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    const out: {[path: string]: ModelArtifactsInfo} = {};\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n    for (let i = 0; i < this.LS.length; ++i) {\n      const key = this.LS.key(i);\n      if (key.startsWith(prefix) && key.endsWith(suffix)) {\n        const modelPath = getModelPathFromKey(key);\n        out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n      }\n    }\n    return out;\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    const keys = getModelKeys(path);\n    if (this.LS.getItem(keys.info) == null) {\n      throw new Error(`Cannot find model at path '${path}'`);\n    }\n    const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n    this.LS.removeItem(keys.info);\n    this.LS.removeItem(keys.topology);\n    this.LS.removeItem(keys.weightSpecs);\n    this.LS.removeItem(keys.weightData);\n    return info;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n *   string.\n * - Moving or copying model from one path to another in the same medium or from\n *   one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n  // Singleton instance.\n  private static instance: ModelStoreManagerRegistry;\n\n  private managers: {[scheme: string]: ModelStoreManager};\n\n  private constructor() {\n    this.managers = {};\n  }\n\n  private static getInstance(): ModelStoreManagerRegistry {\n    if (ModelStoreManagerRegistry.instance == null) {\n      ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n    }\n    return ModelStoreManagerRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerManager(scheme: string, manager: ModelStoreManager) {\n    assert(scheme != null, () => 'scheme must not be undefined or null.');\n    if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n      scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n    }\n    assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n    const registry = ModelStoreManagerRegistry.getInstance();\n    assert(\n        registry.managers[scheme] == null,\n        () => `A model store manager is already registered for scheme '${\n            scheme}'.`);\n    registry.managers[scheme] = manager;\n  }\n\n  static getManager(scheme: string): ModelStoreManager {\n    const manager = this.getInstance().managers[scheme];\n    if (manager == null) {\n      throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n    }\n    return manager;\n  }\n\n  static getSchemes(): string[] {\n    return Object.keys(this.getInstance().managers);\n  }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n *   Scheme: e.g., 'localstorage' in the example above.\n *   Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n  if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n    throw new Error(\n        `The url string provided does not contain a scheme. ` +\n        `Supported schemes are: ` +\n        `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n  }\n  return {\n    scheme: url.split(URL_SCHEME_SUFFIX)[0],\n    path: url.split(URL_SCHEME_SUFFIX)[1],\n  };\n}\n\nasync function cloneModelInternal(\n    sourceURL: string, destURL: string,\n    deleteSource = false): Promise<ModelArtifactsInfo> {\n  assert(\n      sourceURL !== destURL,\n      () => `Old path and new path are the same: '${sourceURL}'`);\n\n  const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n  assert(\n      loadHandlers.length > 0,\n      () => `Copying failed because no load handler is found for source URL ${\n          sourceURL}.`);\n  assert(\n      loadHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `load handlers for source URL ${sourceURL}.`);\n  const loadHandler = loadHandlers[0];\n\n  const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n  assert(\n      saveHandlers.length > 0,\n      () => `Copying failed because no save handler is found for destination ` +\n          `URL ${destURL}.`);\n  assert(\n      saveHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `save handlers for destination URL ${destURL}.`);\n  const saveHandler = saveHandlers[0];\n\n  const sourceScheme = parseURL(sourceURL).scheme;\n  const sourcePath = parseURL(sourceURL).path;\n  const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n  const modelArtifacts = await loadHandler.load();\n\n  // If moving within the same storage medium, remove the old model as soon as\n  // the loading is done. Without doing this, it is possible that the combined\n  // size of the two models will cause the cloning to fail.\n  if (deleteSource && sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  const saveResult = await saveHandler.save(modelArtifacts);\n\n  // If moving between mediums, the deletion is done after the save succeeds.\n  // This guards against the case in which saving to the destination medium\n  // fails.\n  if (deleteSource && !sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n *   'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n  const schemes = ModelStoreManagerRegistry.getSchemes();\n  const out: {[url: string]: ModelArtifactsInfo} = {};\n  for (const scheme of schemes) {\n    const schemeOut =\n        await ModelStoreManagerRegistry.getManager(scheme).listModels();\n    for (const path in schemeOut) {\n      const url = scheme + URL_SCHEME_SUFFIX + path;\n      out[url] = schemeOut[path];\n    }\n  }\n  return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n *   'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n *   is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise<ModelArtifactsInfo> {\n  const schemeAndPath = parseURL(url);\n  const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n  return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function copyModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = false;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function moveModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = true;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n  // According to the spec, the built-in encoder can do only UTF-8 encoding.\n  // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n  private textEncoder: TextEncoder;\n\n  fetch(path: string, init?: RequestInit): Promise<Response> {\n    return fetch(path, init);\n  }\n\n  now(): number {\n    return performance.now();\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Browser's encoder only supports utf-8, but got ${encoding}`);\n    }\n    if (this.textEncoder == null) {\n      this.textEncoder = new TextEncoder();\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    return new TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_BROWSER')) {\n  env().setPlatform('browser', new PlatformBrowser());\n\n  // Register LocalStorage IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n  } catch (err) {\n  }\n\n  // Register IndexedDB IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n  } catch (err) {\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n  // tslint:disable-next-line:no-require-imports\n  importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise<Response>;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n  systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n  systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n  return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n  private textEncoder: TextEncoder;\n  // tslint:disable-next-line:no-any\n  util: any;\n\n  constructor() {\n    // tslint:disable-next-line:no-require-imports\n    this.util = require('util');\n    // According to the spec, the built-in encoder can do only UTF-8 encoding.\n    // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n    this.textEncoder = new this.util.TextEncoder();\n  }\n\n  fetch(path: string, requestInits?: RequestInit): Promise<Response> {\n    if (env().global.fetch != null) {\n      return env().global.fetch(path, requestInits);\n    }\n\n    if (systemFetch == null) {\n      systemFetch = getNodeFetch.importFetch();\n    }\n    return systemFetch(path, requestInits);\n  }\n\n  now(): number {\n    const time = process.hrtime();\n    return time[0] * 1000 + time[1] / 1000000;\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Node built-in encoder only supports utf-8, but got ${encoding}`);\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    if (bytes.length === 0) {\n      return '';\n    }\n    return new this.util.TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_NODE')) {\n  env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer<R extends Rank, D extends DataType = 'float32'>(\n    shape: ShapeMap[R], dtype: D = 'float32' as D,\n    values?: DataTypeMap[D]): TensorBuffer<R, D> {\n  dtype = dtype || 'float32' as D;\n  util.assertNonNegativeIntegerDimensions(shape);\n  return new TensorBuffer<R, D>(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_<T extends Tensor>(x: T|TensorLike, dtype: DataType): T {\n  const $x = convertToTensor(x, 'x', 'cast');\n\n  // Sanity checks.\n  if (!util.isValidDtype(dtype)) {\n    throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n  }\n  if (dtype === 'string' && $x.dtype !== 'string' ||\n      dtype !== 'string' && $x.dtype === 'string') {\n    throw new Error('Only strings can be casted to strings');\n  }\n\n  const inputs: CastInputs = {x: $x};\n  const attrs: CastAttrs = {dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n      null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'clone', null);\n  const forward = () =>\n      ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n  const inputs: IdentityInputs = {x: $x};\n\n  // Note this op is called tf.identity in python. Hence the kernel name used\n  // here.\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print<T extends Tensor>(x: T, verbose = false): void {\n  console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n  buffer,\n  cast,\n  clone,\n  print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer<T>(f: () => T): Promise<T> {\n  return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n  private readonly modelTopologyFileName: string;\n  private readonly weightDataFileName: string;\n  private readonly jsonAnchor: HTMLAnchorElement;\n  private readonly weightDataAnchor: HTMLAnchorElement;\n\n  static readonly URL_SCHEME = 'downloads://';\n\n  constructor(fileNamePrefix?: string) {\n    if (!env().getBool('IS_BROWSER')) {\n      // TODO(cais): Provide info on what IOHandlers are available under the\n      //   current environment.\n      throw new Error(\n          'browserDownloads() cannot proceed because the current environment ' +\n          'is not a browser.');\n    }\n\n    if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n      fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n    }\n    if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n      fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n    }\n\n    this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n    this.weightDataFileName =\n        fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (typeof (document) === 'undefined') {\n      throw new Error(\n          'Browser downloads are not supported in ' +\n          'this environment since `document` is not present');\n    }\n    const weightsURL = window.URL.createObjectURL(new Blob(\n        [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserDownloads.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const weightsManifest: WeightsManifestConfig = [{\n        paths: ['./' + this.weightDataFileName],\n        weights: modelArtifacts.weightSpecs\n      }];\n      const modelTopologyAndWeightManifest: ModelJSON = {\n        modelTopology: modelArtifacts.modelTopology,\n        format: modelArtifacts.format,\n        generatedBy: modelArtifacts.generatedBy,\n        convertedBy: modelArtifacts.convertedBy,\n        weightsManifest\n      };\n      const modelTopologyAndWeightManifestURL =\n          window.URL.createObjectURL(new Blob(\n              [JSON.stringify(modelTopologyAndWeightManifest)],\n              {type: 'application/json'}));\n\n      // If anchor elements are not provided, create them without attaching them\n      // to parents, so that the downloaded file names can be controlled.\n      const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n                                                   this.jsonAnchor;\n      jsonAnchor.download = this.modelTopologyFileName;\n      jsonAnchor.href = modelTopologyAndWeightManifestURL;\n      // Trigger downloads by evoking a click event on the download anchors.\n      // When multiple downloads are started synchronously, Firefox will only\n      // save the last one.\n      await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n      if (modelArtifacts.weightData != null) {\n        const weightDataAnchor = this.weightDataAnchor == null ?\n            document.createElement('a') :\n            this.weightDataAnchor;\n        weightDataAnchor.download = this.weightDataFileName;\n        weightDataAnchor.href = weightsURL;\n        await defer(\n            () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n      }\n\n      return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n    }\n  }\n}\n\nclass BrowserFiles implements IOHandler {\n  private readonly files: File[];\n\n  constructor(files: File[]) {\n    if (files == null || files.length < 1) {\n      throw new Error(\n          `When calling browserFiles, at least 1 file is required, ` +\n          `but received ${files}`);\n    }\n    this.files = files;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    const jsonFile = this.files[0];\n    const weightFiles = this.files.slice(1);\n\n    return new Promise<ModelArtifacts>((resolve, reject) => {\n      const jsonReader = new FileReader();\n      jsonReader.onload = (event: Event) => {\n        // tslint:disable-next-line:no-any\n        const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n        const modelTopology = modelJSON.modelTopology;\n        if (modelTopology == null) {\n          reject(new Error(\n              `modelTopology field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        if (weightFiles.length === 0) {\n          resolve({modelTopology});\n        }\n\n        const weightsManifest = modelJSON.weightsManifest;\n        if (weightsManifest == null) {\n          reject(new Error(\n              `weightManifest field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        let pathToFile: {[path: string]: File};\n        try {\n          pathToFile =\n              this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n        } catch (err) {\n          reject(err);\n          return;\n        }\n\n        const weightSpecs: WeightsManifestEntry[] = [];\n        const paths: string[] = [];\n        const perFileBuffers: ArrayBuffer[] = [];\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            paths.push(path);\n            perFileBuffers.push(null);\n          });\n          weightSpecs.push(...weightsGroup.weights);\n        });\n\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            const weightFileReader = new FileReader();\n            weightFileReader.onload = (event: Event) => {\n              // tslint:disable-next-line:no-any\n              const weightData = (event.target as any).result as ArrayBuffer;\n              const index = paths.indexOf(path);\n              perFileBuffers[index] = weightData;\n              if (perFileBuffers.indexOf(null) === -1) {\n                resolve({\n                  modelTopology,\n                  weightSpecs,\n                  weightData: concatenateArrayBuffers(perFileBuffers),\n                  format: modelJSON.format,\n                  generatedBy: modelJSON.generatedBy,\n                  convertedBy: modelJSON.convertedBy,\n                  userDefinedMetadata: modelJSON.userDefinedMetadata\n                });\n              }\n            };\n            weightFileReader.onerror = error =>\n                reject(`Failed to weights data from file of path '${path}'.`);\n            weightFileReader.readAsArrayBuffer(pathToFile[path]);\n          });\n        });\n      };\n      jsonReader.onerror = error => reject(\n          `Failed to read model topology and weights manifest JSON ` +\n          `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n          `Keras-style tf.Model artifacts only.`);\n      jsonReader.readAsText(jsonFile);\n    });\n  }\n\n  /**\n   * Check the compatibility between weights manifest and weight files.\n   */\n  private checkManifestAndWeightFiles(\n      manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n    const basenames: string[] = [];\n    const fileNames = files.map(file => basename(file.name));\n    const pathToFile: {[path: string]: File} = {};\n    for (const group of manifest) {\n      group.paths.forEach(path => {\n        const pathBasename = basename(path);\n        if (basenames.indexOf(pathBasename) !== -1) {\n          throw new Error(\n              `Duplicate file basename found in weights manifest: ` +\n              `'${pathBasename}'`);\n        }\n        basenames.push(pathBasename);\n        if (fileNames.indexOf(pathBasename) === -1) {\n          throw new Error(\n              `Weight file with basename '${pathBasename}' is not provided.`);\n        } else {\n          pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n        }\n      });\n    }\n\n    if (basenames.length !== files.length) {\n      throw new Error(\n          `Mismatch in the number of files in weights manifest ` +\n          `(${basenames.length}) and the number of weight files provided ` +\n          `(${files.length}).`);\n    }\n    return pathToFile;\n  }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n      return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * //   'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n *   `tf.Model`, `fileNamePrefix` should follow either of the following two\n *   formats:\n *   1. `null` or `undefined`, in which case the default file\n *      names will be used:\n *      - 'model.json' for the JSON file containing the model topology and\n *        weights manifest.\n *      - 'model.weights.bin' for the binary file containing the binary weight\n *        values.\n *   2. A single string or an Array of a single string, as the file name prefix.\n *      For example, if `'foo'` is provided, the downloaded JSON\n *      file and binary weights file will be named 'foo.json' and\n *      'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n  return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * //   elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (`<input type=\"file\" ...>`)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n *   which an `Array` of `File`s is expected (in that order):\n *   - A JSON file containing the model topology and weight manifest.\n *   - Optionally, One or more binary files containing the binary weights.\n *     These files must have names that match the paths in the `weightsManifest`\n *     contained by the aforementioned JSON file, or errors will be thrown\n *     during loading. These weights files have the same format as the ones\n *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n *     Python PIP package. If no weights files are provided, only the model\n *     topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n  return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n    promises: Array<Promise<{}|void>>, onProgress: OnProgressCallback,\n    startFraction?: number, endFraction?: number) {\n  checkPromises(promises);\n  startFraction = startFraction == null ? 0 : startFraction;\n  endFraction = endFraction == null ? 1 : endFraction;\n  checkFraction(startFraction, endFraction);\n  let resolvedPromise = 0;\n\n  const registerMonitor = (promise: Promise<{}>) => {\n    promise.then(value => {\n      const fraction = startFraction +\n          ++resolvedPromise / promises.length * (endFraction - startFraction);\n      // pass fraction as parameter to callback function.\n      onProgress(fraction);\n      return value;\n    });\n    return promise;\n  };\n\n  function checkPromises(promises: Array<Promise<{}|void>>): void {\n    assert(\n        promises != null && Array.isArray(promises) && promises.length > 0,\n        () => 'promises must be a none empty array');\n  }\n\n  function checkFraction(startFraction: number, endFraction: number): void {\n    assert(\n        startFraction >= 0 && startFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got startFraction ${startFraction}`);\n    assert(\n        endFraction >= 0 && endFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got endFraction ${endFraction}`);\n    assert(\n        endFraction >= startFraction,\n        () => `startFraction must be no more than endFraction, but ` +\n            `got startFraction ${startFraction} and endFraction ` +\n            `${endFraction}`);\n  }\n\n  return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n *   before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n *   length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n    fetchURLs: string[], loadOptions?: LoadOptions): Promise<ArrayBuffer[]> {\n  if (loadOptions == null) {\n    loadOptions = {};\n  }\n\n  const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n                                                    loadOptions.fetchFunc;\n\n  // Create the requests for all of the weights in parallel.\n  const requests = fetchURLs.map(\n      fetchURL =>\n          fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n  const fetchStartFraction = 0;\n  const fetchEndFraction = 0.5;\n\n  const responses = loadOptions.onProgress == null ?\n      await Promise.all(requests) :\n      await monitorPromisesProgress(\n          requests, loadOptions.onProgress, fetchStartFraction,\n          fetchEndFraction);\n\n  const bufferPromises = responses.map(response => response.arrayBuffer());\n\n  const bufferStartFraction = 0.5;\n  const bufferEndFraction = 1;\n\n  const buffers = loadOptions.onProgress == null ?\n      await Promise.all(bufferPromises) :\n      await monitorPromisesProgress(\n          bufferPromises, loadOptions.onProgress, bufferStartFraction,\n          bufferEndFraction);\n  return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n *     Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n    manifest: WeightsManifestConfig, filePathPrefix = '',\n    weightNames?: string[],\n    requestInit?: RequestInit): Promise<NamedTensorMap> {\n  // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n  // single weight from a group, the whole group will be fetched. At a future\n  // date, we should support fetching only the individual shards within a\n  // group that are needed to reconstruct the requested weight.\n  // TODO(cais): Use `decodeWeights` for implementation.\n\n  const fetchWeights = (fetchUrls: string[]) =>\n      loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n  const loadWeights = weightsLoaderFactory(fetchWeights);\n\n  return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n *   filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n *   fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n    fetchWeightsFunction: (fetchUrls: string[]) => Promise<ArrayBuffer[]>):\n    (manifest: WeightsManifestConfig, filePathPrefix?: string,\n     weightNames?: string[]) => Promise<NamedTensorMap> {\n  return async(\n             manifest: WeightsManifestConfig, filePathPrefix = '',\n             weightNames?: string[]): Promise<NamedTensorMap> => {\n    // Collect all the groups, weights, and their relative offsets to be\n    // fetched.\n    const groupIndicesToFetchMap = manifest.map(() => false);\n    const groupWeightsToFetch: {\n      [group: number]: Array<{\n        manifestEntry: WeightsManifestEntry; groupOffset: number;\n        sizeBytes: number;\n      }>\n    } = {};\n    const weightsFound =\n        weightNames != null ? weightNames.map(() => false) : [];\n    const allManifestWeightNames: string[] = [];\n    manifest.forEach((manifestGroupConfig, groupIndex) => {\n      let groupOffset = 0;\n      manifestGroupConfig.weights.forEach(weightsEntry => {\n        const rawDtype = ('quantization' in weightsEntry) ?\n            weightsEntry.quantization.dtype :\n            weightsEntry.dtype;\n\n        const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n            util.sizeFromShape(weightsEntry.shape);\n\n        const enqueueWeightsForFetchingFn = () => {\n          groupIndicesToFetchMap[groupIndex] = true;\n          if (groupWeightsToFetch[groupIndex] == null) {\n            groupWeightsToFetch[groupIndex] = [];\n          }\n\n          groupWeightsToFetch[groupIndex].push({\n            manifestEntry: weightsEntry,\n            groupOffset,\n            sizeBytes: weightsBytes\n          });\n        };\n\n        if (weightNames != null) {\n          weightNames.forEach((weightName, weightIndex) => {\n            if (weightName === weightsEntry.name) {\n              enqueueWeightsForFetchingFn();\n              weightsFound[weightIndex] = true;\n            }\n          });\n        } else {\n          enqueueWeightsForFetchingFn();\n        }\n\n        allManifestWeightNames.push(weightsEntry.name);\n        groupOffset += weightsBytes;\n      });\n    });\n\n    if (!weightsFound.every(found => found)) {\n      const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n      throw new Error(\n          `Could not find weights in manifest with names: ` +\n          `${weightsNotFound.join(', ')}. \\n` +\n          `Manifest JSON has weights with names: ` +\n          `${allManifestWeightNames.join(', ')}.`);\n    }\n\n    // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n    // IDs.\n    const groupIndicesToFetch =\n        groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n          if (shouldFetch) {\n            accumulator.push(i);\n          }\n          return accumulator;\n        }, []);\n\n    const fetchUrls: string[] = [];\n    groupIndicesToFetch.forEach(i => {\n      manifest[i].paths.forEach(filepath => {\n        const fetchUrl = filePathPrefix +\n            (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n        fetchUrls.push(fetchUrl);\n      });\n    });\n    const buffers = await fetchWeightsFunction(fetchUrls);\n\n    const weightsTensorMap: NamedTensorMap = {};\n    let bufferIndexOffset = 0;\n    groupIndicesToFetch.forEach(i => {\n      const numBuffers = manifest[i].paths.length;\n\n      let groupBytes = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        groupBytes += buffers[bufferIndexOffset + i].byteLength;\n      }\n\n      // Create a buffer for the whole group.\n      const groupBuffer = new ArrayBuffer(groupBytes);\n      const groupByteBuffer = new Uint8Array(groupBuffer);\n      let groupBufferOffset = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n        groupByteBuffer.set(buffer, groupBufferOffset);\n        groupBufferOffset += buffer.byteLength;\n      }\n\n      const weightsEntries = groupWeightsToFetch[i];\n      weightsEntries.forEach(weightsEntry => {\n        const byteBuffer = groupBuffer.slice(\n            weightsEntry.groupOffset,\n            weightsEntry.groupOffset + weightsEntry.sizeBytes);\n        const nameToTensorMap =\n            decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n        for (const name in nameToTensorMap) {\n          weightsTensorMap[name] = nameToTensorMap[name];\n        }\n      });\n\n      bufferIndexOffset += numBuffers;\n    });\n\n    return weightsTensorMap;\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n  protected readonly path: string;\n  protected readonly requestInit: RequestInit;\n\n  private readonly fetch: Function;\n  private readonly weightUrlConverter: (weightName: string) => Promise<string>;\n\n  readonly DEFAULT_METHOD = 'POST';\n\n  static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n  private readonly weightPathPrefix: string;\n  private readonly onProgress: OnProgressCallback;\n\n  constructor(path: string, loadOptions?: LoadOptions) {\n    if (loadOptions == null) {\n      loadOptions = {};\n    }\n    this.weightPathPrefix = loadOptions.weightPathPrefix;\n    this.onProgress = loadOptions.onProgress;\n    this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n    if (loadOptions.fetchFunc != null) {\n      assert(\n          typeof loadOptions.fetchFunc === 'function',\n          () => 'Must pass a function that matches the signature of ' +\n              '`fetch` (see ' +\n              'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n      this.fetch = loadOptions.fetchFunc;\n    } else {\n      this.fetch = env().platform.fetch;\n    }\n\n    assert(\n        path != null && path.length > 0,\n        () => 'URL path for http must not be null, undefined or ' +\n            'empty.');\n\n    if (Array.isArray(path)) {\n      assert(\n          path.length === 2,\n          () => 'URL paths for http must have a length of 2, ' +\n              `(actual length is ${path.length}).`);\n    }\n    this.path = path;\n\n    if (loadOptions.requestInit != null &&\n        loadOptions.requestInit.body != null) {\n      throw new Error(\n          'requestInit is expected to have no pre-existing body, but has one.');\n    }\n    this.requestInit = loadOptions.requestInit || {};\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserHTTPRequest.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n    init.body = new FormData();\n\n    const weightsManifest: WeightsManifestConfig = [{\n      paths: ['./model.weights.bin'],\n      weights: modelArtifacts.weightSpecs,\n    }];\n    const modelTopologyAndWeightManifest: ModelJSON = {\n      modelTopology: modelArtifacts.modelTopology,\n      format: modelArtifacts.format,\n      generatedBy: modelArtifacts.generatedBy,\n      convertedBy: modelArtifacts.convertedBy,\n      userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n      weightsManifest\n    };\n\n    init.body.append(\n        'model.json',\n        new Blob(\n            [JSON.stringify(modelTopologyAndWeightManifest)],\n            {type: JSON_TYPE}),\n        'model.json');\n\n    if (modelArtifacts.weightData != null) {\n      init.body.append(\n          'model.weights.bin',\n          new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n          'model.weights.bin');\n    }\n\n    const response = await this.fetch(this.path, init);\n\n    if (response.ok) {\n      return {\n        modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n        responses: [response],\n      };\n    } else {\n      throw new Error(\n          `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n          `${response.status}.`);\n    }\n  }\n\n  /**\n   * Load model artifacts via HTTP request(s).\n   *\n   * See the documentation to `tf.io.http` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model artifacts (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n    if (!modelConfigRequest.ok) {\n      throw new Error(\n          `Request to ${this.path} failed with status code ` +\n          `${modelConfigRequest.status}. Please verify this URL points to ` +\n          `the model JSON of the model to load.`);\n    }\n    let modelConfig: ModelJSON;\n    try {\n      modelConfig = await modelConfigRequest.json();\n    } catch (e) {\n      let message = `Failed to parse model JSON of response from ${this.path}.`;\n      // TODO(nsthorat): Remove this after some time when we're comfortable that\n      // .pb files are mostly gone.\n      if (this.path.endsWith('.pb')) {\n        message += ' Your path contains a .pb file extension. ' +\n            'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n            'in favor of .json models. You can re-convert your Python ' +\n            'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n            'or you can convert your.pb models with the \\'pb2json\\'' +\n            'NPM script in the tensorflow/tfjs-converter repository.';\n      } else {\n        message += ' Please make sure the server is serving valid ' +\n            'JSON for this request.';\n      }\n      throw new Error(message);\n    }\n    const modelTopology = modelConfig.modelTopology;\n    const weightsManifest = modelConfig.weightsManifest;\n    const generatedBy = modelConfig.generatedBy;\n    const convertedBy = modelConfig.convertedBy;\n    const format = modelConfig.format;\n    const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n    // We do not allow both modelTopology and weightsManifest to be missing.\n    if (modelTopology == null && weightsManifest == null) {\n      throw new Error(\n          `The JSON from HTTP path ${this.path} contains neither model ` +\n          `topology or manifest for weights.`);\n    }\n\n    let weightSpecs: WeightsManifestEntry[];\n    let weightData: ArrayBuffer;\n    if (weightsManifest != null) {\n      const results = await this.loadWeights(weightsManifest);\n      [weightSpecs, weightData] = results;\n    }\n\n    const artifacts: ModelArtifacts = {\n      modelTopology,\n      weightSpecs,\n      weightData,\n      userDefinedMetadata,\n      generatedBy,\n      convertedBy,\n      format\n    };\n\n    const initializer = modelConfig.modelInitializer;\n    if (initializer) {\n      artifacts.modelInitializer = initializer;\n    }\n\n    return artifacts;\n  }\n\n  private async loadWeights(weightsManifest: WeightsManifestConfig):\n      Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n    const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n    const [prefix, suffix] = parseUrl(weightPath);\n    const pathPrefix = this.weightPathPrefix || prefix;\n\n    const weightSpecs = [];\n    for (const entry of weightsManifest) {\n      weightSpecs.push(...entry.weights);\n    }\n\n    const fetchURLs: string[] = [];\n    const urlPromises: Array<Promise<string>> = [];\n    for (const weightsGroup of weightsManifest) {\n      for (const path of weightsGroup.paths) {\n        if (this.weightUrlConverter != null) {\n          urlPromises.push(this.weightUrlConverter(path));\n        } else {\n          fetchURLs.push(pathPrefix + path + suffix);\n        }\n      }\n    }\n\n    if (this.weightUrlConverter) {\n      fetchURLs.push(...await Promise.all(urlPromises));\n    }\n\n    const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n      requestInit: this.requestInit,\n      fetchFunc: this.fetch,\n      onProgress: this.onProgress\n    });\n    return [weightSpecs, concatenateArrayBuffers(buffers)];\n  }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n  const lastSlash = url.lastIndexOf('/');\n  const lastSearchParam = url.lastIndexOf('?');\n  const prefix = url.substring(0, lastSlash);\n  const suffix =\n      lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n  return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n  return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n    (url: string, loadOptions?: LoadOptions) => {\n      if (typeof fetch === 'undefined' &&\n          (loadOptions == null || loadOptions.fetchFunc == null)) {\n        // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n        // an environment that is not the browser or node they have to setup a\n        // global fetch polyfill.\n        return null;\n      } else {\n        let isHTTP = true;\n        if (Array.isArray(url)) {\n          isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n        } else {\n          isHTTP = isHTTPScheme(url);\n        }\n        if (isHTTP) {\n          return http(url, loadOptions);\n        }\n      }\n      return null;\n    };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n *   - A JSON file consisting of `modelTopology` and `weightsManifest`.\n *   - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n *     'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n *   Can be an absolute HTTP path (e.g.,\n *   'http://localhost:8000/model-upload)') or a relative path (e.g.,\n *   './model-upload').\n * @param requestInit Request configurations to be used when sending\n *    HTTP request to server using `fetch`. It can contain fields such as\n *    `method`, `credentials`, `headers`, `mode`, etc. See\n *    https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n *    for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n *   following fields:\n *   - weightPathPrefix Optional, this specifies the path prefix for weight\n *     files, by default this is calculated from the path param.\n *   - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n *     the `fetch` from node-fetch can be used here.\n *   - onProgress Optional, progress callback function, fired periodically\n *     before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n  return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n    path: string, loadOptions?: LoadOptions): IOHandler {\n  return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n  constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n  async load(): Promise<ModelArtifacts> {\n    return this.modelArtifacts;\n  }\n}\n\nclass PassthroughSaver implements IOHandler {\n  constructor(\n      private readonly saveHandler:\n          (artifacts: ModelArtifacts) => Promise<SaveResult>) {}\n\n  async save(modelArtifacts: ModelArtifacts) {\n    return this.saveHandler(modelArtifacts);\n  }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n *     modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n *   the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n *   names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n *   concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n    modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n    weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n  if (arguments.length === 1) {\n    const isModelArtifacts =\n        (modelArtifacts as ModelArtifacts).modelTopology != null ||\n        (modelArtifacts as ModelArtifacts).weightSpecs != null;\n    if (isModelArtifacts) {\n      return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n    } else {\n      // Legacy support: with only modelTopology.\n      // TODO(cais): Remove this deprecated API.\n      console.warn(\n          'Please call tf.io.fromMemory() with only one argument. ' +\n          'The argument should be of type ModelArtifacts. ' +\n          'The multi-argument signature of tf.io.fromMemory() has been ' +\n          'deprecated and will be removed in a future release.');\n      return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n    }\n  } else {\n    // Legacy support.\n    // TODO(cais): Remove this deprecated API.\n    console.warn(\n        'Please call tf.io.fromMemory() with only one argument. ' +\n        'The argument should be of type ModelArtifacts. ' +\n        'The multi-argument signature of tf.io.fromMemory() has been ' +\n        'deprecated and will be removed in a future release.');\n    return new PassthroughLoader({\n      modelTopology: modelArtifacts as {},\n      weightSpecs,\n      weightData,\n      trainingConfig\n    });\n  }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n *   // ... do something with the artifacts ...\n *   return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n *     `SaveResult`.\n */\nexport function withSaveHandler(\n    saveHandler: (artifacts: ModelArtifacts) =>\n        Promise<SaveResult>): IOHandler {\n  return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n  browserFiles,\n  browserHTTPRequest,\n  concatenateArrayBuffers,\n  decodeWeights,\n  encodeWeights,\n  fromMemory,\n  getLoadHandlers,\n  getModelArtifactsInfoForJSON,\n  getSaveHandlers,\n  http,\n  IOHandler,\n  isHTTPScheme,\n  LoadHandler,\n  LoadOptions,\n  loadWeights,\n  ModelArtifacts,\n  ModelArtifactsInfo,\n  ModelJSON,\n  ModelStoreManager,\n  OnProgressCallback,\n  registerLoadRouter,\n  registerSaveRouter,\n  RequestDetails,\n  SaveConfig,\n  SaveHandler,\n  SaveResult,\n  WeightGroup,\n  weightsLoaderFactory,\n  WeightsManifestConfig,\n  WeightsManifestEntry,\n  withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  const $x = convertToTensor(x, 'x', 'reshape', null);\n\n  const inputs: ReshapeInputs = {x: $x};\n  const attrs: ReshapeAttrs = {shape};\n  const forward: ForwardFunc<\n      Tensor<R>> = (backend: KernelBackend, save: GradSaveFunc) => {\n    shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n    util.assert(\n        $x.size === util.sizeFromShape(shape),\n        () => 'new shape and old shape must have the same number of elements.');\n    save([$x]);\n    return backend.reshape($x, shape);\n  };\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n      attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print();  // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike, transposeA = false,\n    transposeB = false): T {\n  let $a = convertToTensor(a, 'a', 'matMul');\n  let $b = convertToTensor(b, 'b', 'matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n          `got ranks ${$a.rank} and ${$b.rank}.`);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n                           reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n                           reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([a3D, b3D]);\n\n    return backend.batchMatMul(\n        a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n  };\n\n  const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n  const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n      attrs as {} as NamedAttrMap);\n\n  return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n *     not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n    indices: Tensor|TensorLike, depth: number, onValue = 1,\n    offValue = 0): Tensor {\n  if (depth < 2) {\n    throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n  }\n  const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n  const outShape = [...$indices.shape, depth];\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$indices]);\n    return reshape(\n        backend.oneHot(\n            reshape($indices, [$indices.size]), depth, onValue, offValue),\n        outShape);\n  };\n\n  const inputs: OneHotInputs = {indices: $indices};\n  const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print();  // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_<T extends Tensor>(x: T|TensorLike, perm?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'transpose');\n\n  if (perm == null) {\n    perm = $x.shape.map((s, i) => i).reverse();\n  }\n  util.assert(\n      $x.rank === perm.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of perm ${perm}.`);\n  perm.forEach(axis => {\n    util.assert(\n        axis >= 0 && axis < $x.rank,\n        () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n            ` but got ${perm}`);\n  });\n\n  if ($x.rank <= 1) {\n    return $x.clone();\n  }\n\n  const inputs: TransposeInputs = {x: $x};\n  const attrs: TransposeAttrs = {perm};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n      null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * //  [0, 1, 1],\n * //  [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n *   for the classes. The shape is `[numExamples]`, where\n *   `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n *   0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n *   Its value must be larger than the largest element in `labels` and\n *   `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n *   row `r` and column `c` is the number of times examples of actual class\n *   `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n    labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n    numClasses: number): Tensor2D {\n  const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n  util.assert(\n      numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n      () => `If provided, numClasses must be a positive integer, ` +\n          `but got ${numClasses}`);\n  util.assert(\n      $labels.rank === 1,\n      () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n  util.assert(\n      $predictions.rank === 1,\n      () => `Expected the rank of predictions to be 1, ` +\n          `but got ${$predictions.rank}`);\n  util.assert(\n      $labels.shape[0] === $predictions.shape[0],\n      () => `Mismatch in the number of examples: ` +\n          `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n          `Labels and predictions should have the same number of elements.`);\n  util.assert(\n      numClasses > 0 && Number.isInteger(numClasses),\n      () => `numClasses is required to be a positive integer, but got ` +\n          `${numClasses}`);\n  // TODO(cais): In the future, if oneHot supports tensors inputs for\n  //   `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n  const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n  const oneHotPredictions =\n      oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n  const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n  return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided,  it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n    values: TensorLike3D, shape?: [number, number, number],\n    dtype?: DataType): Tensor3D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 3) {\n    throw new Error('tensor3d() requires shape to have three numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor3d() requires values to be number[][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor3d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement,\n    numChannels = 3): Tensor3D {\n  // Sanity checks.\n  if (numChannels > 4) {\n    throw new Error(\n        'Cannot construct Tensor with more than 4 channels from pixels.');\n  }\n  if (pixels == null) {\n    throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n  }\n  let isPixelData = false;\n  let isImageData = false;\n  let isVideo = false;\n  let isImage = false;\n  let isCanvasLike = false;\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    isPixelData = true;\n  } else if (\n      typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n    isImageData = true;\n  } else if (\n      typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement) {\n    isVideo = true;\n  } else if (\n      typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement) {\n    isImage = true;\n    // tslint:disable-next-line: no-any\n  } else if ((pixels as any).getContext != null) {\n    isCanvasLike = true;\n  } else {\n    throw new Error(\n        'pixels passed to tf.browser.fromPixels() must be either an ' +\n        `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n        `in browser, or OffscreenCanvas, ImageData in webworker` +\n        ` or {data: Uint32Array, width: number, height: number}, ` +\n        `but was ${(pixels as {}).constructor.name}`);\n  }\n  if (isVideo) {\n    const HAVE_CURRENT_DATA_READY_STATE = 2;\n    if (isVideo &&\n        (pixels as HTMLVideoElement).readyState <\n            HAVE_CURRENT_DATA_READY_STATE) {\n      throw new Error(\n          'The video element has not loaded data yet. Please wait for ' +\n          '`loadeddata` event on the <video> element.');\n    }\n  }\n  // If the current backend has 'FromPixels' registered, it has a more\n  // efficient way of handling pixel uploads, so we call that.\n  const kernel = getKernel(FromPixels, ENGINE.backendName);\n  if (kernel != null) {\n    const inputs: FromPixelsInputs = {pixels};\n    const attrs: FromPixelsAttrs = {numChannels};\n    return ENGINE.runKernel(\n               FromPixels, inputs as {} as NamedTensorMap,\n               attrs as {} as NamedAttrMap) as Tensor3D;\n  }\n\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n  let vals: Uint8ClampedArray|Uint8Array;\n\n  if (isCanvasLike) {\n    vals =\n        // tslint:disable-next-line:no-any\n        (pixels as any).getContext('2d').getImageData(0, 0, width, height).data;\n  } else if (isImageData || isPixelData) {\n    vals = (pixels as PixelData | ImageData).data;\n  } else if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement, 0, 0, width, height);\n    vals = fromPixels2DContext.getImageData(0, 0, width, height).data;\n  }\n  let values: Int32Array;\n  if (numChannels === 4) {\n    values = new Int32Array(vals);\n  } else {\n    const numPixels = width * height;\n    values = new Int32Array(numPixels * numChannels);\n    for (let i = 0; i < numPixels; i++) {\n      for (let channel = 0; channel < numChannels; ++channel) {\n        values[i * numChannels + channel] = vals[i * 4 + channel];\n      }\n    }\n  }\n  const outShape: [number, number, number] = [height, width, numChannels];\n  return tensor3d(values, outShape, 'int32');\n}\n\n/**\n * Draws a `tf.Tensor` of pixel values to a byte array or optionally a\n * canvas.\n *\n * When the dtype of the input is 'float32', we assume values in the range\n * [0-1]. Otherwise, when input is 'int32', we assume values in the range\n * [0-255].\n *\n * Returns a promise that resolves when the canvas has been drawn to.\n *\n * @param img A rank-2 or rank-3 tensor. If rank-2, draws grayscale. If\n *     rank-3, must have depth of 1, 3 or 4. When depth of 1, draws\n * grayscale. When depth of 3, we draw with the first three components of\n * the depth dimension corresponding to r, g, b and alpha = 1. When depth of\n * 4, all four components of the depth dimension correspond to r, g, b, a.\n * @param canvas The canvas to draw to.\n *\n * @doc {heading: 'Browser', namespace: 'browser'}\n */\nexport async function toPixels(\n    img: Tensor2D|Tensor3D|TensorLike,\n    canvas?: HTMLCanvasElement): Promise<Uint8ClampedArray> {\n  let $img = convertToTensor(img, 'img', 'toPixels');\n  if (!(img instanceof Tensor)) {\n    // Assume int32 if user passed a native array.\n    const originalImgTensor = $img;\n    $img = cast(originalImgTensor, 'int32');\n    originalImgTensor.dispose();\n  }\n  if ($img.rank !== 2 && $img.rank !== 3) {\n    throw new Error(\n        `toPixels only supports rank 2 or 3 tensors, got rank ${$img.rank}.`);\n  }\n  const [height, width] = $img.shape.slice(0, 2);\n  const depth = $img.rank === 2 ? 1 : $img.shape[2];\n\n  if (depth > 4 || depth === 2) {\n    throw new Error(\n        `toPixels only supports depth of size ` +\n        `1, 3 or 4 but got ${depth}`);\n  }\n\n  if ($img.dtype !== 'float32' && $img.dtype !== 'int32') {\n    throw new Error(\n        `Unsupported type for toPixels: ${$img.dtype}.` +\n        ` Please use float32 or int32 tensors.`);\n  }\n\n  const data = await $img.data();\n  const multiplier = $img.dtype === 'float32' ? 255 : 1;\n  const bytes = new Uint8ClampedArray(width * height * 4);\n\n  for (let i = 0; i < height * width; ++i) {\n    const rgba = [0, 0, 0, 255];\n\n    for (let d = 0; d < depth; d++) {\n      const value = data[i * depth + d];\n\n      if ($img.dtype === 'float32') {\n        if (value < 0 || value > 1) {\n          throw new Error(\n              `Tensor values for a float32 Tensor must be in the ` +\n              `range [0 - 1] but encountered ${value}.`);\n        }\n      } else if ($img.dtype === 'int32') {\n        if (value < 0 || value > 255) {\n          throw new Error(\n              `Tensor values for a int32 Tensor must be in the ` +\n              `range [0 - 255] but encountered ${value}.`);\n        }\n      }\n\n      if (depth === 1) {\n        rgba[0] = value * multiplier;\n        rgba[1] = value * multiplier;\n        rgba[2] = value * multiplier;\n      } else {\n        rgba[d] = value * multiplier;\n      }\n    }\n\n    const j = i * 4;\n    bytes[j + 0] = Math.round(rgba[0]);\n    bytes[j + 1] = Math.round(rgba[1]);\n    bytes[j + 2] = Math.round(rgba[2]);\n    bytes[j + 3] = Math.round(rgba[3]);\n  }\n\n  if (canvas != null) {\n    canvas.width = width;\n    canvas.height = height;\n    const ctx = canvas.getContext('2d');\n    const imageData = new ImageData(bytes, width, height);\n    ctx.putImageData(imageData, 0, 0);\n  }\n  if ($img !== img) {\n    $img.dispose();\n  }\n  return bytes;\n}\n\nexport const fromPixels = op({fromPixels_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\nimport {computeStrides} from '../util';\n\n/**\n * Validate gather nd inputs.\n *\n * @param tensor The tensor contains the source values.\n * @param indices The tensor contains the indices to slice the source.\n *\n * @returns [resultShape, numUpdates, sliceSize, strides]\n */\nexport function prepareAndValidate(\n    tensor: Tensor, indices: Tensor): [number[], number, number, number[]] {\n  if (tensor.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the input to be rank 1 or higher,' +\n        ` but the rank was ${tensor.rank}.`);\n  }\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(\n        'tf.gatherND() expects the indices to be int32 type,' +\n        ` but the dtype was ${indices.dtype}.`);\n  }\n  if (indices.shape[indices.rank - 1] > tensor.rank) {\n    throw new Error(\n        'index innermost dimension length must be <= tensor rank; saw: ' +\n        `${indices.shape[indices.rank - 1]} vs. ${tensor.rank}`);\n  }\n\n  if (tensor.size === 0) {\n    throw new Error(\n        'Requested more than 0 entries, but input is empty.' +\n        ` Input shape: ${tensor.shape}.`);\n  }\n\n  const indicesShape = indices.shape;\n  const sliceRank = indicesShape[indicesShape.length - 1];\n\n  // The result shape is\n  //   indices.shape[:-1] + params.shape[indices.shape[-1]:]\n  let nResult = 1;\n  for (let i = 0; i < indicesShape.length - 1; ++i) {\n    nResult *= indicesShape[i];\n  }\n\n  const inputShape = tensor.shape;\n\n  const resultShape = indicesShape.slice();\n  resultShape.pop();\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < tensor.rank; ++i) {\n    sliceSize *= inputShape[i];\n    resultShape.push(inputShape[i]);\n  }\n\n  const strides =\n      [...computeStrides(tensor.shape).map(stride => stride / sliceSize),\n       1].slice(0, sliceRank);\n\n  return [resultShape, nResult, sliceSize, strides];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {computeStrides, sizeFromShape} from '../util';\n\n/**\n * Check whether updates.shape = indices.shape[:batchDim] +\n * shape[sliceDim:]\n *\n * @param x The input tensor.\n */\nexport function validateUpdateShape(\n    shape: number[], indices: Tensor, updates: Tensor) {\n  const sliceDim = (indices.rank > 1) ? indices.shape[indices.rank - 1] : 1;\n  const batchDim = (indices.rank > 1) ? indices.rank - 1 : 1;\n\n  const shapeError = 'Must have updates.shape = indices.shape[:batchDim] + ' +\n      `shape[sliceDim:], got updates.shape: ${updates.shape}` +\n      `, indices.shape: ${indices.shape}, shape: ${shape}` +\n      `, sliceDim: ${sliceDim}, and batchDim: ${batchDim}.`;\n\n  if (updates.rank < batchDim) {\n    throw new Error(shapeError + ` update.rank < ${batchDim}. `);\n  }\n  if (shape.length < sliceDim + (updates.rank - batchDim)) {\n    throw new Error(\n        shapeError +\n        ` Output shape length < ${sliceDim + (updates.rank - batchDim)}`);\n  }\n  if (updates.rank !== batchDim + shape.length - sliceDim) {\n    throw new Error(\n        shapeError + ` update.rank != ${batchDim + shape.length - sliceDim}`);\n  }\n  for (let d = 0; d < batchDim; ++d) {\n    if (updates.shape[d] !== indices.shape[d]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d}] (${updates.shape[d]}) != indices.shape[${d}] (${\n              indices.shape[d]}).`);\n    }\n  }\n  for (let d = 0; d < updates.rank - batchDim; ++d) {\n    if (updates.shape[d + batchDim] !== shape[d + sliceDim]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d + batchDim}] (${\n              updates.shape[d + batchDim]}) != shape[${d + batchDim}] (${\n              shape[d + batchDim]})`);\n    }\n  }\n}\n\nexport interface ScatterShapeInfo {\n  sliceRank: number;\n  numUpdates: number;\n  sliceSize: number;\n  strides: number[];\n  outputSize: number;\n}\n/**\n * Validate scatter nd inputs.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n */\nexport function validateInput(\n    updates: Tensor, indices: Tensor, shape: number[]) {\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (updates.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the updates to be rank 1 or higher,' +\n        ` but the rank was ${updates.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${\n        indices.dtype}`);\n  }\n  if (shape.length < 1) {\n    throw new Error(\n        `Output rank must be greater or equal to 1, but got shape: ${shape}`);\n  }\n\n  if (shape.length === 0) {\n    if (indices.size === 0) {\n      throw new Error(`Indices specified for empty output. indices shape: ${\n          indices.shape}`);\n    }\n    if (updates.size === 0) {\n      throw new Error(`Updates specified for empty output. updates shape: ${\n          updates.shape}`);\n    }\n  }\n\n  validateUpdateShape(shape, indices, updates);\n}\n\n/**\n * Calculate the shape information for the output.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n *\n * @returns ScatterShapeInfo\n */\nexport function calculateShapes(\n    updates: TensorInfo, indices: TensorInfo,\n    shape: number[]): ScatterShapeInfo {\n  // Calculate the number of dimensions in indices\n  const indicesRank = indices.shape.length;\n  const sliceRank = (indicesRank > 1) ? indices.shape[indicesRank - 1] : 1;\n\n  // Calculate the number of elements that make up each slice of our updated\n  // tensor. This allows us to work with flattened tensors and copy over whole\n  // slices at a time.\n  const totalNd = shape.length;\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < totalNd; ++i) {\n    sliceSize *= shape[i];\n  }\n\n  const safeSliceDim = (sliceRank < 1) ? 1 : sliceRank;\n  const numUpdates = sizeFromShape(indices.shape) / safeSliceDim;\n\n  const strides = [...computeStrides(shape.slice(0, sliceRank)), 1];\n  const outputSize = sizeFromShape(shape);\n  return {sliceRank, numUpdates, sliceSize, strides, outputSize};\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '../kernel_registry';\nimport * as util from '../util';\n\nexport function assertParamsValid(\n    input: TensorInfo, begin: number[], size: number[]): void {\n  const inputRank = input.shape.length;\n  util.assert(\n      inputRank === begin.length,\n      () => `Error in slice${inputRank}D: Length of begin ${begin} must ` +\n          `match the rank of the array (${inputRank}).`);\n  util.assert(\n      inputRank === size.length,\n      () => `Error in slice${inputRank}D: Length of size ${size} must ` +\n          `match the rank of the array (${inputRank}).`);\n\n  for (let i = 0; i < inputRank; ++i) {\n    util.assert(\n        begin[i] + size[i] <= input.shape[i],\n        () => `Error in slice${inputRank}D: begin[${i}] + size[${i}] ` +\n            `(${begin[i] + size[i]}) would overflow input.shape[${i}] (${\n                  input.shape[i]})`);\n  }\n}\n\n/** Converts a binary mask to an array of axes. Used in stridedSlice(). */\nexport function maskToAxes(mask: number): number[] {\n  const axes = [];\n  let axis = 0;\n  while (mask > 0) {\n    if (mask & 1) {\n      axes.push(axis);\n    }\n    mask /= 2;\n    axis++;\n  }\n  return axes;\n}\n\n/** Computes the output shape given the strided slice params. */\nexport function computeOutShape(\n    begin: number[], end: number[], strides: number[]): number[] {\n  const size = [];\n  for (let axis = 0; axis < begin.length; axis++) {\n    size[axis] = Math.ceil((end[axis] - begin[axis]) / strides[axis]);\n  }\n  return size;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stride value. Otherwise, insert.\nexport function stridesWithElidedDims(\n    strides: number[], ellipsisInsertionIndex: number, numElidedAxes: number,\n    inputShape: number[]): number[] {\n  const newStrides = [...strides];\n  for (let i = newStrides.length; i < inputShape.length; i++) {\n    newStrides.push(1);\n  }\n  for (let i = 0; i < numElidedAxes; i++) {\n    if (i === 0) {\n      newStrides[ellipsisInsertionIndex] = 1;\n    } else {\n      newStrides.splice(\n          ellipsisInsertionIndex, 0 /* num elements to delete */,\n          1 /* element to add */);\n      newStrides.pop();\n    }\n  }\n  return newStrides;\n}\n\nfunction unnormalizeAxis(\n    ellipsisInsertionIndex: number, numElidedAxes: number,\n    normalizedAxis: number): number {\n  if (normalizedAxis <= ellipsisInsertionIndex) {\n    return normalizedAxis;\n  }\n\n  return normalizedAxis - (numElidedAxes - 1);\n}\n\nfunction getElidedAxes(numElidedAxes: number, ellipsisInsertionIndex: number) {\n  const elidedAxes = [];\n  for (let i = 0; i < numElidedAxes; i++) {\n    elidedAxes.push(ellipsisInsertionIndex + i);\n  }\n  return elidedAxes;\n}\n\n// Normalize the start, end and strides.\nexport function getNormalizedAxes(\n    inputShape: number[], ellipsisAxes: number[], numInterpolatedAxes: number,\n    begin: number[], end: number[], strides: number[], beginMask: number,\n    endMask: number,\n    ellipsisMask: number): {begin: number[], end: number[], strides: number[]} {\n  const inputRank = inputShape.length;\n  let normalizedBegin = new Array(inputRank),\n      normalizedEnd = new Array(inputRank),\n      normalizedStrides = new Array(inputRank);\n  if (ellipsisAxes.length && numInterpolatedAxes > 0) {\n    const fullIndex = ellipsisAxes[0];\n\n    // The ellipsis applies to the masked index as well as any dimensions\n    // that are interpolated.\n    const numElidedAxes = numInterpolatedAxes + 1;\n    normalizedBegin = startIndicesWithElidedDims(\n        beginMask, fullIndex, numElidedAxes, begin, inputShape);\n    normalizedEnd = stopIndicesWithElidedDims(\n        endMask, fullIndex, numElidedAxes, end, inputShape);\n    normalizedStrides =\n        stridesWithElidedDims(strides, fullIndex, numElidedAxes, inputShape);\n  } else {\n    for (let axis = 0; axis < inputRank; axis++) {\n      normalizedBegin[axis] = startForAxis(\n          beginMask, begin, strides, inputShape, axis, ellipsisMask);\n      normalizedEnd[axis] =\n          stopForAxis(endMask, end, strides, inputShape, axis, ellipsisMask);\n      normalizedStrides[axis] = stridesForAxis(strides, axis, ellipsisMask);\n    }\n  }\n\n  return {\n    begin: normalizedBegin,\n    end: normalizedEnd,\n    strides: normalizedStrides\n  };\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current start value. Otherwise, insert.\nexport function startIndicesWithElidedDims(\n    beginMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalBegin: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = 0;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalBegin[originalAxis];\n      if (beginMask & 1 << originalAxis) {\n        originalValue = 0;\n      }\n\n      newIndices[axis] = originalValue;\n    }\n  }\n  return newIndices;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stop value. Otherwise, insert.\nexport function stopIndicesWithElidedDims(\n    endMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalEnd: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = Number.MAX_SAFE_INTEGER;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalEnd[originalAxis];\n      if (endMask & 1 << originalAxis) {\n        originalValue = Number.MAX_SAFE_INTEGER;\n      }\n      newIndices[axis] = originalValue;\n    }\n  }\n\n  for (let i = 0; i < newIndices.length; i++) {\n    // Handle negative indices\n    const axisSize = inputShape[i];\n    if (newIndices[i] < 0) {\n      newIndices[i] += axisSize;\n    }\n    newIndices[i] = util.clamp(0, newIndices[i], inputShape[i]);\n  }\n  return newIndices;\n}\n\nexport function stridesForAxis(\n    strides: number[], axis: number, ellipsisMask: number): number {\n  let stride = strides[axis];\n  if (ellipsisMask & (1 << axis) || stride == null) {\n    stride = 1;\n  }\n\n  return stride;\n}\n\nexport function startForAxis(\n    beginMask: number, startIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let start = startIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or the begin index is not set\n  // for the axis.\n  if (beginMask & 1 << axis || ellipsisMask & 1 << axis || start == null) {\n    if (stride > 0) {\n      // Forward iteration - use the first element. These values will get\n      // clamped below (Note: We could have set them to 0 and axis_size-1, but\n      // use lowest() and max() to maintain symmetry with StopForAxis())\n      start = Number.MIN_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the last element.\n      start = Number.MAX_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (start < 0) {\n    start += axisSize;\n  }\n\n  // Clamping\n  start = util.clamp(0, start, axisSize - 1);\n\n  return start;\n}\n\nexport function stopForAxis(\n    endMask: number, stopIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let stop = stopIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or if the stop index is not\n  // set for this axis.\n  if (endMask & (1 << axis) || ellipsisMask & (1 << axis) || stop == null) {\n    if (stride > 0) {\n      // Forward iteration - use the last element. These values will get\n      // clamped below\n      stop = Number.MAX_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the first element.\n      stop = Number.MIN_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (stop < 0) {\n    stop += axisSize;\n  }\n\n  // Clamping\n  // Because the end index points one past the last element, we need slightly\n  // different clamping ranges depending on the direction.\n  if (stride > 0) {\n    // Forward iteration\n    stop = util.clamp(0, stop, axisSize);\n  } else {\n    // Backward iteration\n    stop = util.clamp(-1, stop, axisSize - 1);\n  }\n\n  return stop;\n}\n\n/**\n * Returns true if the slice occupies a continous set of elements in the\n * 'flat' space.\n */\nexport function isSliceContinous(\n    shape: number[], begin: number[], size: number[]) {\n  // Index of the first axis that has size > 1.\n  let firstNonOneAxis = size.length;\n  for (let i = 0; i < size.length; i++) {\n    if (size[i] > 1) {\n      firstNonOneAxis = i;\n      break;\n    }\n  }\n\n  for (let i = firstNonOneAxis + 1; i < size.length; i++) {\n    if (begin[i] > 0 || size[i] !== shape[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function computeFlatOffset(begin: number[], strides: number[]): number {\n  let flatOffset = begin.length > 0 ? begin[begin.length - 1] : 1;\n  for (let i = 0; i < begin.length - 1; i++) {\n    flatOffset += begin[i] * strides[i];\n  }\n  return flatOffset;\n}\n\nexport function parseSliceParams(\n    x: TensorInfo, begin: number|number[], size?: number|number[]) {\n  // The following logic allows for more ergonomic calls.\n  let begin_: number[];\n  const xRank = x.shape.length;\n  if (typeof begin === 'number') {\n    begin_ = [begin, ...new Array(xRank - 1).fill(0)];\n  } else if (begin.length < xRank) {\n    begin_ = begin.concat(new Array(xRank - begin.length).fill(0));\n  } else {\n    begin_ = begin.slice();\n  }\n  begin_.forEach(d => {\n    util.assert(\n        d !== -1, () => 'slice() does not support negative begin indexing.');\n  });\n  let size_: number[];\n  if (size == null) {\n    size_ = new Array(xRank).fill(-1);\n  } else if (typeof size === 'number') {\n    size_ = [size, ...new Array(xRank - 1).fill(-1)];\n  } else if (size.length < xRank) {\n    size_ = size.concat(new Array(xRank - size.length).fill(-1));\n  } else {\n    size_ = size;\n  }\n  size_ = size_.map((d, i) => {\n    if (d >= 0) {\n      return d;\n    } else {\n      util.assert(\n          d === -1,\n          () => `Negative size values should be exactly -1 but got ` +\n              `${d} for the slice() size at index ${i}.`);\n      return x.shape[i] - begin_[i];\n    }\n  });\n  return [begin_, size_];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from './util';\n\n/**\n * Types to support JSON-esque data structures internally.\n *\n * Internally ConfigDict's use camelCase keys and values where the\n * values are class names to be instantiated.  On the python side, these\n * will be snake_case.  Internally we allow Enums into the values for better\n * type safety, but these need to be converted to raw primitives (usually\n * strings) for round-tripping with python.\n *\n * toConfig returns the TS-friendly representation. model.toJSON() returns\n * the pythonic version as that's the portable format.  If you need to\n * python-ify a non-model level toConfig output, you'll need to use a\n * convertTsToPythonic from serialization_utils in -Layers.\n *\n */\nexport declare type ConfigDictValue =\n    boolean | number | string | null | ConfigDictArray | ConfigDict;\nexport declare interface ConfigDict {\n  [key: string]: ConfigDictValue;\n}\nexport declare interface ConfigDictArray extends Array<ConfigDictValue> {}\n\n/**\n * Type to represent the class-type of Serializable objects.\n *\n * Ie the class prototype with access to the constructor and any\n * static members/methods. Instance methods are not listed here.\n *\n * Source for this idea: https://stackoverflow.com/a/43607255\n */\nexport declare type SerializableConstructor<T extends Serializable> = {\n  // tslint:disable-next-line:no-any\n  new (...args: any[]): T; className: string; fromConfig: FromConfigMethod<T>;\n};\nexport declare type FromConfigMethod<T extends Serializable> =\n    (cls: SerializableConstructor<T>, config: ConfigDict) => T;\n\n/**\n * Serializable defines the serialization contract.\n *\n * TFJS requires serializable classes to return their className when asked\n * to avoid issues with minification.\n */\nexport abstract class Serializable {\n  /**\n   * Return the class name for this class to use in serialization contexts.\n   *\n   * Generally speaking this will be the same thing that constructor.name\n   * would have returned.  However, the class name needs to be robust\n   * against minification for serialization/deserialization to work properly.\n   *\n   * There's also places such as initializers.VarianceScaling, where\n   * implementation details between different languages led to different\n   * class hierarchies and a non-leaf node is used for serialization purposes.\n   */\n  getClassName(): string {\n    return (this.constructor as SerializableConstructor<Serializable>)\n        .className;\n  }\n\n  /**\n   * Return all the non-weight state needed to serialize this object.\n   */\n  abstract getConfig(): ConfigDict;\n\n  /**\n   * Creates an instance of T from a ConfigDict.\n   *\n   * This works for most descendants of serializable.  A few need to\n   * provide special handling.\n   * @param cls A Constructor for the class to instantiate.\n   * @param config The Configuration for the object.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config);\n  }\n}\n\n/**\n * Maps string keys to class constructors.\n *\n * Used during (de)serialization from the cross-language JSON format, which\n * requires the class name in the serialization format matches the class\n * names as used in Python, should it exist.\n */\nexport class SerializationMap {\n  private static instance: SerializationMap;\n  classNameMap: {\n    [className: string]:\n        [SerializableConstructor<Serializable>, FromConfigMethod<Serializable>]\n  };\n\n  private constructor() {\n    this.classNameMap = {};\n  }\n\n  /**\n   * Returns the singleton instance of the map.\n   */\n  static getMap(): SerializationMap {\n    if (SerializationMap.instance == null) {\n      SerializationMap.instance = new SerializationMap();\n    }\n    return SerializationMap.instance;\n  }\n\n  /**\n   * Registers the class as serializable.\n   */\n  static register<T extends Serializable>(cls: SerializableConstructor<T>) {\n    SerializationMap.getMap().classNameMap[cls.className] =\n        [cls, cls.fromConfig];\n  }\n}\n\n/**\n * Register a class with the serialization map of TensorFlow.js.\n *\n * This is often used for registering custom Layers, so they can be\n * serialized and deserialized.\n *\n * Example:\n *\n * ```js\n * class MyCustomLayer extends tf.layers.Layer {\n *   static className = 'MyCustomLayer';\n *\n *   constructor(config) {\n *     super(config);\n *   }\n * }\n * tf.serialization.registerClass(MyCustomLayer);\n * ```\n *\n * @param cls The class to be registered. It must have a public static member\n *   called `className` defined and the value must be a non-empty string.\n *\n * @doc {heading: 'Models', subheading: 'Serialization', ignoreCI: true}\n */\nexport function registerClass<T extends Serializable>(\n    cls: SerializableConstructor<T>) {\n  assert(\n      cls.className != null,\n      () => `Class being registered does not have the static className ` +\n          `property defined.`);\n  assert(\n      typeof cls.className === 'string',\n      () => `className is required to be a string, but got type ` +\n          typeof cls.className);\n  assert(\n      cls.className.length > 0,\n      () => `Class being registered has an empty-string as its className, ` +\n          `which is disallowed.`);\n\n  SerializationMap.register(cls);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {inferShape} from './tensor_util_env';\nimport {RecursiveArray, TensorLike, TypedArray} from './types';\nimport {arraysEqual, flatten, isString, isTypedArray} from './util';\n\nconst TEST_EPSILON_FLOAT32 = 1e-3;\nexport const TEST_EPSILON_FLOAT16 = 1e-1;\n\nexport function expectArraysClose(\n    actual: TypedArray|number|RecursiveArray<number>,\n    expected: TypedArray|number|RecursiveArray<number>, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, epsilon));\n}\n\nexport function testEpsilon() {\n  return ENGINE.backend.floatPrecision() === 32 ? TEST_EPSILON_FLOAT32 :\n                                                  TEST_EPSILON_FLOAT16;\n}\n\nfunction expectArraysPredicate(\n    actual: TensorLike, expected: TensorLike,\n    predicate: (a: {}, b: {}) => boolean) {\n  let checkClassType = true;\n  if (isTypedArray(actual) || isTypedArray(expected)) {\n    checkClassType = false;\n  }\n  if (isTypedArray(actual) && isTypedArray(expected)) {\n    checkClassType = true;\n  }\n  if (checkClassType) {\n    const aType = actual.constructor.name;\n    const bType = expected.constructor.name;\n\n    if (aType !== bType) {\n      throw new Error(\n          `Arrays are of different type. Actual: ${aType}. ` +\n          `Expected: ${bType}`);\n    }\n  }\n\n  if (Array.isArray(actual) && Array.isArray(expected)) {\n    const actualShape = inferShape(actual);\n    const expectedShape = inferShape(expected);\n    if (!arraysEqual(actualShape, expectedShape)) {\n      throw new Error(\n          `Arrays have different shapes. ` +\n          `Actual: [${actualShape}]. Expected: [${expectedShape}]`);\n    }\n  }\n\n  const actualFlat =\n      isTypedArray(actual) ? actual : flatten(actual as RecursiveArray<number>);\n  const expectedFlat = isTypedArray(expected) ?\n      expected :\n      flatten(expected as RecursiveArray<number>);\n\n  if (actualFlat.length !== expectedFlat.length) {\n    throw new Error(\n        `Arrays have different lengths actual: ${actualFlat.length} vs ` +\n        `expected: ${expectedFlat.length}.\\n` +\n        `Actual:   ${actualFlat}.\\n` +\n        `Expected: ${expectedFlat}.`);\n  }\n  for (let i = 0; i < expectedFlat.length; ++i) {\n    const a = actualFlat[i];\n    const e = expectedFlat[i];\n\n    if (!predicate(a, e)) {\n      throw new Error(\n          `Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${e}.\\n` +\n          `Actual:   ${actualFlat}.\\n` +\n          `Expected: ${expectedFlat}.`);\n    }\n  }\n}\n\nexport interface DoneFn {\n  (): void;\n  fail: (message?: Error|string) => void;\n}\n\nexport function expectPromiseToFail(fn: () => Promise<{}>, done: DoneFn): void {\n  fn().then(() => done.fail(), () => done());\n}\n\nexport function expectArraysEqual(actual: TensorLike, expected: TensorLike) {\n  const exp = typeof expected === 'string' || typeof expected === 'number' ||\n          typeof expected === 'boolean' ?\n      [expected] as number[] :\n      expected as number[];\n  if (isString(actual) || isString((actual as string[])[0]) ||\n      isString(expected) || isString((expected as string[])[0])) {\n    // tslint:disable-next-line: triple-equals\n    return expectArraysPredicate(actual, exp, (a, b) => a == b);\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, 0));\n}\n\nexport function expectNumbersClose(a: number, e: number, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  if (!areClose(a, e, epsilon)) {\n    throw new Error(`Numbers differ: actual === ${a}, expected === ${e}`);\n  }\n}\n\nfunction areClose(a: number, e: number, epsilon: number): boolean {\n  if (!isFinite(a) && !isFinite(e)) {\n    return true;\n  }\n  if (isNaN(a) || isNaN(e) || Math.abs(a - e) > epsilon) {\n    return false;\n  }\n  return true;\n}\n\nexport function expectValuesInRange(\n    actual: TypedArray|number[], low: number, high: number) {\n  for (let i = 0; i < actual.length; i++) {\n    if (actual[i] < low || actual[i] > high) {\n      throw new Error(\n          `Value out of range:${actual[i]} low: ${low}, high: ${high}`);\n    }\n  }\n}\n\nexport function expectArrayBuffersEqual(\n    actual: ArrayBuffer, expected: ArrayBuffer) {\n  // Safari & Jasmine don't like comparing ArrayBuffers directly. Wrapping in\n  // a Float32Array solves this issue.\n  expect(new Float32Array(actual)).toEqual(new Float32Array(expected));\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from './backends/backend';\nimport {ENGINE, Engine, MemoryInfo, ProfileInfo, ScopeFn, TimingInfo} from './engine';\nimport {env} from './environment';\n\nimport {Platform} from './platforms/platform';\nimport {setDeprecationWarningFn, Tensor} from './tensor';\nimport {TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\n\n/**\n * Enables production mode which disables correctness checks in favor of\n * performance.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableProdMode(): void {\n  env().set('PROD', true);\n}\n\n/**\n * Enables debug mode which will log information about all executed kernels:\n * the elapsed time of the kernel execution, as well as the rank, shape, and\n * size of the output tensor.\n *\n * Debug mode will significantly slow down your application as it will\n * download the result of every operation to the CPU. This should not be used in\n * production. Debug mode does not affect the timing information of the kernel\n * execution as we do not measure download time in the kernel execution time.\n *\n * See also: `tf.profile`, `tf.memory`.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableDebugMode(): void {\n  env().set('DEBUG', true);\n}\n\n/** Globally disables deprecation warnings */\nexport function disableDeprecationWarnings(): void {\n  env().set('DEPRECATION_WARNINGS_ENABLED', false);\n  console.warn(`TensorFlow.js deprecation warnings have been disabled.`);\n}\n\n/** Warn users about deprecated functionality. */\nexport function deprecationWarn(msg: string) {\n  if (env().getBool('DEPRECATION_WARNINGS_ENABLED')) {\n    console.warn(\n        msg + ' You can disable deprecation warnings with ' +\n        'tf.disableDeprecationWarnings().');\n  }\n}\nsetDeprecationWarningFn(deprecationWarn);\n\n/**\n * Dispose all variables kept in backend engine.\n *\n * @doc {heading: 'Environment'}\n */\nexport function disposeVariables(): void {\n  ENGINE.disposeVariables();\n}\n\n/**\n * It returns the global engine that keeps track of all tensors and backends.\n *\n * @doc {heading: 'Environment'}\n */\nexport function engine(): Engine {\n  return ENGINE;\n}\n\n/**\n * Returns memory info at the current time in the program. The result is an\n * object with the following properties:\n *\n * - `numBytes`: Number of bytes allocated (undisposed) at this time.\n * - `numTensors`: Number of unique tensors allocated.\n * - `numDataBuffers`: Number of unique data buffers allocated\n *   (undisposed) at this time, which is \u2264 the number of tensors\n *   (e.g. `a.reshape(newShape)` makes a new Tensor that shares the same\n *   data buffer with `a`).\n * - `unreliable`: True if the memory usage is unreliable. See `reasons` when\n *    `unreliable` is true.\n * - `reasons`: `string[]`, reasons why the memory is unreliable, present if\n *    `unreliable` is true.\n *\n * WebGL Properties:\n * - `numBytesInGPU`: Number of bytes allocated (undisposed) in the GPU only at\n *     this time.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function memory(): MemoryInfo {\n  return ENGINE.memory();\n}\n\n/**\n * Executes the provided function `f()` and returns a promise that resolves\n * with information about the function's memory use:\n * - `newBytes`: the number of new bytes allocated\n * - `newTensors`: the number of new tensors created\n * - `peakBytes`: the peak number of bytes allocated\n * - `kernels`: an array of objects for each kernel involved that reports\n * their input and output shapes, number of bytes used, and number of new\n * tensors created.\n *\n * ```js\n * const profile = await tf.profile(() => {\n *   const x = tf.tensor1d([1, 2, 3]);\n *   let x2 = x.square();\n *   x2.dispose();\n *   x2 = x.square();\n *   x2.dispose();\n *   return x;\n * });\n *\n * console.log(`newBytes: ${profile.newBytes}`);\n * console.log(`newTensors: ${profile.newTensors}`);\n * console.log(`byte usage over all kernels: ${profile.kernels.map(k =>\n * k.totalBytesSnapshot)}`);\n * ```\n *\n *\n * @doc {heading: 'Performance', subheading: 'Profile'}\n */\nexport function profile(f: () => (TensorContainer | Promise<TensorContainer>)):\n    Promise<ProfileInfo> {\n  return ENGINE.profile(f);\n}\n\n/**\n * Executes the provided function `fn` and after it is executed, cleans up all\n * intermediate tensors allocated by `fn` except those returned by `fn`.\n * `fn` must not return a Promise (async functions not allowed). The returned\n * result can be a complex object.\n *\n * Using this method helps avoid memory leaks. In general, wrap calls to\n * operations in `tf.tidy` for automatic memory cleanup.\n *\n * NOTE: Variables do *not* get cleaned up when inside a tidy(). If you want to\n * dispose variables, please use `tf.disposeVariables` or call dispose()\n * directly on variables.\n *\n * ```js\n * // y = 2 ^ 2 + 1\n * const y = tf.tidy(() => {\n *   // a, b, and one will be cleaned up when the tidy ends.\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *   const b = a.square();\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * y.print();\n * ```\n *\n * @param nameOrFn The name of the closure, or the function to execute.\n *     If a name is provided, the 2nd argument should be the function.\n *     If debug mode is on, the timing and the memory usage of the function\n *     will be tracked and displayed on the console using the provided name.\n * @param fn The function to execute.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function tidy<T extends TensorContainer>(\n    nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>): T {\n  return ENGINE.tidy(nameOrFn, fn);\n}\n\n/**\n * Disposes any `tf.Tensor`s found within the provided object.\n *\n * @param container an object that may be a `tf.Tensor` or may directly\n *     contain `tf.Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. If\n *     the object is not a `tf.Tensor` or does not contain `Tensors`, nothing\n *     happens. In general it is safe to pass any object here, except that\n *     `Promise`s are not supported.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function dispose(container: TensorContainer) {\n  const tensors = getTensorsInContainer(container);\n  tensors.forEach(tensor => tensor.dispose());\n}\n\n/**\n * Keeps a `tf.Tensor` generated inside a `tf.tidy` from being disposed\n * automatically.\n *\n * ```js\n * let b;\n * const y = tf.tidy(() => {\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *\n *   // b will not be cleaned up by the tidy. a and one will be cleaned up\n *   // when the tidy ends.\n *   b = tf.keep(a.square());\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * console.log('y:');\n * y.print();\n * console.log('b:');\n * b.print();\n * ```\n *\n * @param result The tensor to keep from being disposed.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function keep<T extends Tensor>(result: T): T {\n  return ENGINE.keep(result);\n}\n\n/**\n * Executes `f()` and returns a promise that resolves with timing\n * information.\n *\n * The result is an object with the following properties:\n *\n * - `wallMs`: Wall execution time.\n * - `kernelMs`: Kernel execution time, ignoring data transfer. If using the\n * WebGL backend and the query timer extension is not available, this will\n * return an error object.\n * - On `WebGL` The following additional properties exist:\n *   - `uploadWaitMs`: CPU blocking time on texture uploads.\n *   - `downloadWaitMs`: CPU blocking time on texture downloads (readPixels).\n *\n * ```js\n * const x = tf.randomNormal([20, 20]);\n * const time = await tf.time(() => x.matMul(x));\n *\n * console.log(`kernelMs: ${time.kernelMs}, wallTimeMs: ${time.wallMs}`);\n * ```\n *\n * @param f The function to execute and time.\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nexport function time(f: () => void): Promise<TimingInfo> {\n  return ENGINE.time(f);\n}\n\n/**\n * Sets the backend (cpu, webgl, wasm, etc) responsible for creating tensors and\n * executing operations on those tensors. Returns a promise that resolves\n * to a boolean if the backend initialization was successful.\n *\n * Note this disposes the current backend, if any, as well as any tensors\n * associated with it. A new backend is initialized, even if it is of the\n * same type as the previous one.\n *\n * @param backendName The name of the backend. Currently supports\n *     `'webgl'|'cpu'` in the browser, `'tensorflow'` under node.js\n *     (requires tfjs-node), and `'wasm'` (requires tfjs-backend-wasm).\n *\n * @doc {heading: 'Backends'}\n */\nexport function setBackend(backendName: string): Promise<boolean> {\n  return ENGINE.setBackend(backendName);\n}\n\n/**\n * Returns a promise that resolves when the currently selected backend (or the\n * highest priority one) has initialized. Await this promise when you are using\n * a backend that has async initialization.\n *\n * @doc {heading: 'Backends'}\n */\nexport function ready(): Promise<void> {\n  return ENGINE.ready();\n}\n\n/**\n * Returns the current backend name (cpu, webgl, etc). The backend is\n * responsible for creating tensors and executing operations on those tensors.\n *\n * @doc {heading: 'Backends'}\n */\nexport function getBackend(): string {\n  return ENGINE.backendName;\n}\n\n/**\n * Removes a backend and the registered factory.\n *\n * @doc {heading: 'Backends'}\n */\nexport function removeBackend(name: string): void {\n  ENGINE.removeBackend(name);\n}\n\n/**\n * Finds the backend registered under the provided name. Returns null if the\n * name is not in the registry, or the registration hasn't finished yet.\n */\nexport function findBackend(name: string): KernelBackend {\n  return ENGINE.findBackend(name);\n}\n\n/**\n * Finds the backend factory registered under the provided name. Returns a\n * function that produces a new backend when called. Returns null if the name\n * is not in the registry.\n */\nexport function findBackendFactory(name: string): () =>\n    KernelBackend | Promise<KernelBackend> {\n  return ENGINE.findBackendFactory(name);\n}\n\n/**\n * Registers a global backend. The registration should happen when importing\n * a module file (e.g. when importing `backend_webgl.ts`), and is used for\n * modular builds (e.g. custom tfjs bundle with only webgl support).\n *\n * @param factory The backend factory function. When called, it should\n * return a backend instance, or a promise of an instance.\n * @param priority The priority of the backend (higher = more important).\n *     In case multiple backends are registered, the priority is used to find\n *     the best backend. Defaults to 1.\n * @return False if there is already a registered backend under this name, true\n *     if not.\n *\n * @doc {heading: 'Backends'}\n */\nexport function registerBackend(\n    name: string, factory: () => KernelBackend | Promise<KernelBackend>,\n    priority = 1): boolean {\n  return ENGINE.registerBackend(name, factory, priority);\n}\n\n/**\n * Gets the current backend. If no backends have been initialized, this will\n * attempt to initialize the best backend. Will throw an error if the highest\n * priority backend has async initialization, in which case, you should call\n * 'await tf.ready()' before running other code.\n *\n * @doc {heading: 'Backends'}\n */\nexport function backend(): KernelBackend {\n  return ENGINE.backend;\n}\n\n/**\n * Sets the global platform.\n *\n * @param platformName The name of this platform.\n * @param platform A platform implementation.\n */\nexport function setPlatform(platformName: string, platform: Platform) {\n  env().setPlatform(platformName, platform);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Add, AddInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n *\n * ```js\n * // Broadcast add a with b.\n * const a = tf.scalar(5);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n * @param a The first `tf.Tensor` to add.\n * @param b The second `tf.Tensor` to add. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction add_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'add');\n  let $b = convertToTensor(b, 'b', 'add');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.add($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: AddInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Add) as T;\n}\n\nexport const add = op({add_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FloorDiv, FloorDivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n * The result is rounded with floor function.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.floorDiv(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.floorDiv(b).print();  // or tf.floorDiv(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction floorDiv_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'floorDiv');\n  let $b = convertToTensor(b, 'b', 'floorDiv');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.floorDiv($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: FloorDivInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             FloorDiv) as T;\n}\n\nexport const floorDiv = op({floorDiv_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Div, DivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {floorDiv} from './floorDiv';\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction div_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'int32' && $b.dtype === 'int32') {\n    return floorDiv($a, $b);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.realDivide($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: DivInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Div,\n             attrs) as T;\n}\n\nexport const div = op({div_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Multiply, MultiplyInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Multiplies two `tf.Tensor`s element-wise, A * B. Supports broadcasting.\n *\n * We also expose `tf.mulStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([2, 3, 4, 5]);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n *\n * ```js\n * // Broadcast mul a with b.\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.scalar(5);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n * @param a The first tensor to multiply.\n * @param b The second tensor to multiply. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mul_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mul');\n  let $b = convertToTensor(b, 'b', 'mul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.multiply($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MultiplyInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Multiply) as T;\n}\nexport const mul = op({mul_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Abs, AbsInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes absolute value element-wise: `abs(x)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.abs().print();  // or tf.abs(x)\n * ```\n * @param x The input `tf.Tensor`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction abs_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'abs');\n\n  const inputs: AbsInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    if ($x.dtype === 'complex64') {\n      return backend.complexAbs($x);\n    }\n\n    return backend.abs($x);\n  }, inputs as {} as NamedTensorMap, null /* grad */, Abs);\n}\n\nexport const abs = op({abs_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Acos, AcosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes acos of the input `tf.Tensor` element-wise: `acos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.acos().print();  // or tf.acos(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acos');\n  const inputs: AcosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acos);\n}\nexport const acos = op({acos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Acosh, AcoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the inverse hyperbolic cos of the input `tf.Tensor` element-wise:\n * `acosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([10, 1, 3, 5.7]);\n *\n * x.acosh().print();  // or tf.acosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acosh');\n  const inputs: AcoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acosh);\n}\nexport const acosh = op({acosh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AddN, AddNInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Adds a list of `tf.Tensor`s element-wise, each with the same shape and dtype.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n *\n * tf.addN([a, b, c]).print();\n * ```\n * @param tensors A list of tensors with the same shape and dtype.\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction addN_<T extends Tensor>(tensors: Array<T|TensorLike>): T {\n  util.assert(\n      Array.isArray(tensors),\n      () => 'The argument passed to tf.addN() must be a list of tensors');\n  util.assert(\n      tensors.length >= 1,\n      () => `Must pass at least one tensor to tf.addN(), but got ` +\n          `${tensors.length}`);\n\n  const $tensors =\n      tensors.map((t, i) => convertToTensor(t, `tensors${i}`, 'addN'));\n\n  const firstTensor = $tensors[0];\n  $tensors.forEach(t => {\n    if (t.dtype !== firstTensor.dtype) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same dtype');\n    }\n  });\n\n  $tensors.forEach(t => {\n    if (!util.arraysEqual(t.shape, firstTensor.shape)) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same shape');\n    }\n  });\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.addN($tensors);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: AddNInputs = $tensors;\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, AddN) as\n      T;\n}\n\nexport const addN = op({addN_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\n/**\n * Returns true if the axis specifies the inner most dimensions of the\n * array.\n */\nexport function axesAreInnerMostDims(axes: number[], rank: number): boolean {\n  for (let i = 0; i < axes.length; ++i) {\n    if (axes[axes.length - i - 1] !== rank - 1 - i) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function combineLocations(\n    outputLoc: number[], reduceLoc: number[], axes: number[]): number[] {\n  const rank = outputLoc.length + reduceLoc.length;\n  const loc = [];\n  let outIdx = 0;\n  let reduceIdx = 0;\n  \u00A0 for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      loc.push(outputLoc[outIdx++]);\n    } else {\n      loc.push(reduceLoc[reduceIdx++]);\n    }\n  }\n  return loc;\n}\n\nexport function computeOutAndReduceShapes(\n    aShape: number[], axes: number[]): [number[], number[]] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      outShape.push(aShape[dim]);\n    }\n  }\n  const reduceShape = axes.map(dim => aShape[dim]);\n  return [outShape, reduceShape];\n}\n\nexport function expandShapeToKeepDim(\n    shape: number[], axes: number[]): number[] {\n  const reduceSubShape = axes.map(x => 1);\n  return combineLocations(shape, reduceSubShape, axes);\n}\n\nexport function assertAxesAreInnerMostDims(\n    msg: string, axes: number[], rank: number): void {\n  util.assert(\n      axesAreInnerMostDims(axes, rank),\n      () => `${msg} supports only inner-most axes for now. ` +\n          `Got axes ${axes} and rank-${rank} input.`);\n}\n\n/**\n * Returns the axes permutation to be used with `tf.transpose`, if such\n * permutation is necessary. Otherwise it returns null. This method is used by\n * operations that operate only on inner-most axes.\n */\nexport function getAxesPermutation(axes: number[], rank: number): number[]|\n    null {\n  if (axesAreInnerMostDims(axes, rank)) {\n    return null;\n  }\n  const result: number[] = [];\n  for (let i = 0; i < rank; ++i) {\n    if (axes.indexOf(i) === -1) {\n      result.push(i);\n    }\n  }\n  axes.forEach(axis => result.push(axis));\n  return result;\n}\n\n/** Returns the axes permutation that undoes the original permutation. */\nexport function getUndoAxesPermutation(axes: number[]): number[] {\n  return axes.map((axis, i) => [i, axis])\n      .sort((a, b) => a[1] - b[1])\n      .map(x => x[0]);\n}\n\nexport function getInnerMostAxes(numAxes: number, rank: number): number[] {\n  const res: number[] = [];\n  for (let i = rank - numAxes; i < rank; ++i) {\n    res.push(i);\n  }\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {All, AllAttrs, AllInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical and of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.all().print();  // or tf.all(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.all(axis).print();  // or tf.all(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction all_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'all', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.all($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AllInputs = {x: $x};\n  const attrs: AllAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, All,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const all = op({all_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Any, AnyAttrs, AnyInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical or of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.any().print();  // or tf.any(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.any(axis).print();  // or tf.any(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction any_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'any', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.any($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AnyInputs = {x: $x};\n  const attrs: AnyAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Any,\n             attrs as {} as NamedAttrMap) as T;\n}\n\n// tslint:disable-next-line:variable-name\nexport const any = op({any_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMax, ArgMaxAttrs, ArgMaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the maximum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMax().print();  // or tf.argMax(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMax(axis).print();  // or tf.argMax(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMax_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMax');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMax($x, axes[0]);\n  };\n\n  const inputs: ArgMaxInputs = {x: $x};\n  const attrs: ArgMaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMax,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMax = op({argMax_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMin, ArgMinAttrs, ArgMinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the minimum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMin().print();  // or tf.argMin(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMin(axis).print();  // or tf.argMin(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMin_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMin');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if (axis == null) {\n      axis = 0;\n    }\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMin($x, axes[0]);\n  };\n\n  const inputs: ArgMinInputs = {x: $x};\n  const attrs: ArgMinAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMin,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMin = op({argMin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asin, AsinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes asin of the input `tf.Tensor` element-wise: `asin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asin().print();  // or tf.asin(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asin');\n  const inputs: AsinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asin);\n}\nexport const asin = op({asin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asinh, AsinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic sin of the input `tf.Tensor` element-wise:\n * `asinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asinh().print();  // or tf.asinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asinh');\n\n  const inputs: AsinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asinh);\n}\nexport const asinh = op({asinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atan, AtanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes atan of the input `tf.Tensor` element-wise: `atan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.atan().print();  // or tf.atan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atan');\n\n  const inputs: AtanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atan);\n}\nexport const atan = op({atan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Atan2, Atan2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes arctangent of `tf.Tensor`s a / b element-wise: `atan2(a, b)`.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1.0, 1.0, -1.0, .7]);\n * const b = tf.tensor1d([2.0, 13.0, 3.5, .21]);\n *\n * tf.atan2(a, b).print()\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan2_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'atan2');\n  let $b = convertToTensor(b, 'b', 'atan2');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.atan2($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: Atan2Inputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Atan2) as T;\n}\n\nexport const atan2 = op({atan2_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atanh, AtanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic tan of the input `tf.Tensor` element-wise:\n * `atanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.atanh().print();  // or tf.atanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atanh');\n\n  const inputs: AtanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atanh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atanh);\n}\nexport const atanh = op({atanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\ntype PadType = 'SAME'|'VALID'|'NUMBER'|'EXPLICIT';\n\n// For NHWC should be in the following form:\n//  [[0, 0], [pad_top,pad_bottom], [pad_left, pad_right], [0, 0]]\n// For NCHW should be in the following form:\n//  [[0, 0], [0, 0], [pad_top,pad_bottom], [pad_left, pad_right]]\n// Reference: https://www.tensorflow.org/api_docs/python/tf/nn/conv2d\nexport type ExplicitPadding =\n    [[number, number], [number, number], [number, number], [number, number]];\n\nexport type PadInfo = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  type: PadType\n};\n\nexport type PadInfo3D = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  front: number,\n  back: number,\n  type: PadType\n};\n\n/**\n * Information about the forward pass of a convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv2DInfo = {\n  batchSize: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideHeight: number,\n  strideWidth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo,\n  inShape: [number, number, number, number],\n  outShape: [number, number, number, number],\n  filterShape: [number, number, number, number]\n};\n\n/**\n *\n * @param inputShape Input tensor shape is of the following dimensions:\n *     `[batch, height, width, inChannels]`.\n * @param filterShape The filter shape is of the following dimensions:\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat The data format of the input and output data.\n *     Defaults to 'NHWC'.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.\n *     Defaults to `[1, 1]`. If `dilations` is a single number, then\n *     `dilationHeight == dilationWidth`.\n */\nexport function computeDilation2DInfo(\n    inputShape: [number, number, number, number],\n    filterShape: [number, number, number], strides: number|[number, number],\n    pad: 'same'|'valid'|number, dataFormat: 'NHWC' = 'NHWC',\n    dilations: number|[number, number]) {\n  // `computerConv2DInfo` require filterShape to be in the dimension of:\n  // `[filterHeight, filterWidth, depth, outDepth]`, dilation2d doesn't have\n  // outDepth, it should have the same depth as the input.\n  // Input shape: [batch, height, width, inChannels]\n  const inputChannels = inputShape[3];\n  const $filterShape =\n      [...filterShape, inputChannels] as [number, number, number, number];\n  const $dataFormat = convertConv2DDataFormat(dataFormat);\n\n  return computeConv2DInfo(\n      inputShape, $filterShape, strides, dilations, pad,\n      null /* roundingMode */, null /* depthWise */, $dataFormat);\n}\n\nexport function computePool2DInfo(\n    inShape: [number, number, number, number],\n    filterSize: [number, number]|number, strides: number|[number, number],\n    dilations: number|[number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  const [filterHeight, filterWidth] = parseTupleParam(filterSize);\n\n  let filterShape: [number, number, number, number];\n  if (dataFormat === 'channelsLast') {\n    filterShape = [filterHeight, filterWidth, inShape[3], inShape[3]];\n  } else if (dataFormat === 'channelsFirst') {\n    filterShape = [filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv2DInfo(\n      inShape, filterShape, strides, dilations, pad, roundingMode, false,\n      dataFormat);\n}\n\n/**\n * Computes the information for a forward pass of a pooling3D operation.\n */\nexport function computePool3DInfo(\n    inShape: [number, number, number, number, number],\n    filterSize: number|[number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC'): Conv3DInfo {\n  const [filterDepth, filterHeight, filterWidth] = parse3TupleParam(filterSize);\n\n  let filterShape: [number, number, number, number, number];\n  let $dataFormat: 'channelsFirst'|'channelsLast';\n  if (dataFormat === 'NDHWC') {\n    $dataFormat = 'channelsLast';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[4], inShape[4]];\n  } else if (dataFormat === 'NCDHW') {\n    $dataFormat = 'channelsFirst';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv3DInfo(\n      inShape, filterShape, strides, dilations, pad, false, $dataFormat,\n      roundingMode);\n}\n\n/**\n * Computes the information for a forward pass of a convolution/pooling\n * operation.\n */\nexport function computeConv2DInfo(\n    inShape: [number, number, number, number],\n    filterShape: [number, number, number, number],\n    strides: number|[number, number], dilations: number|[number, number],\n    pad: 'same'|'valid'|number|ExplicitPadding,\n    roundingMode?: 'floor'|'round'|'ceil', depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  let [batchSize, inHeight, inWidth, inChannels] = [-1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterHeight, filterWidth, , filterChannels] = filterShape;\n  const [strideHeight, strideWidth] = parseTupleParam(strides);\n  const [dilationHeight, dilationWidth] = parseTupleParam(dilations);\n\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outHeight, outWidth} = getPadAndOutInfo(\n      pad, inHeight, inWidth, strideHeight, strideWidth, effectiveFilterHeight,\n      effectiveFilterWidth, roundingMode, dataFormat);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inHeight,\n    inWidth,\n    inChannels,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideHeight,\n    strideWidth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\n/**\n * Information about the forward pass of a 3D convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv3DInfo = {\n  batchSize: number,\n  inDepth: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideDepth: number,\n  strideHeight: number,\n  strideWidth: number,\n  dilationDepth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterDepth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterDepth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo3D,\n  inShape: [number, number, number, number, number],\n  outShape: [number, number, number, number, number],\n  filterShape: [number, number, number, number, number]\n};\n\n/**\n * Computes the information for a forward pass of a 3D convolution/pooling\n * operation.\n */\nexport function computeConv3DInfo(\n    inShape: [number, number, number, number, number],\n    filterShape: [number, number, number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast',\n    roundingMode?: 'floor'|'round'|'ceil'): Conv3DInfo {\n  let [batchSize, inDepth, inHeight, inWidth, inChannels] =\n      [-1, -1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inDepth, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inDepth, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterDepth, filterHeight, filterWidth, , filterChannels] =\n      filterShape;\n  const [strideDepth, strideHeight, strideWidth] = parse3TupleParam(strides);\n  const [dilationDepth, dilationHeight, dilationWidth] =\n      parse3TupleParam(dilations);\n\n  const effectiveFilterDepth =\n      getEffectiveFilterSize(filterDepth, dilationDepth);\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outDepth, outHeight, outWidth} = get3DPadAndOutInfo(\n      pad, inDepth, inHeight, inWidth, strideDepth, strideHeight, strideWidth,\n      effectiveFilterDepth, effectiveFilterHeight, effectiveFilterWidth,\n      roundingMode);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outDepth, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outDepth, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inDepth,\n    inHeight,\n    inWidth,\n    inChannels,\n    outDepth,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideDepth,\n    strideHeight,\n    strideWidth,\n    filterDepth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterDepth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationDepth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\nfunction computeOutputShape2D(\n    inShape: [number, number], fieldSize: number, stride: number,\n    zeroPad?: number, roundingMode?: 'floor'|'round'|'ceil'): [number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputRows = inShape[0];\n  const inputCols = inShape[1];\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputRows, outputCols];\n}\n\nfunction computeOutputShape4D(\n    inShape: [number, number, number, number], fieldSize: number,\n    outChannels: number, stride: number, zeroPad?: number,\n    roundingMode?: 'floor'|'round'|'ceil'): [number, number, number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputDepth = inShape[0];\n  const inputRows = inShape[1];\n  const inputCols = inShape[2];\n\n  const outputDepths = conditionalRound(\n      (inputDepth - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputDepths),\n      () => `The output # of depths (${outputDepths}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputDepths, outputRows, outputCols, outChannels];\n}\n\nexport function computeDefaultPad(\n    inputShape: [number, number]|[number, number, number, number],\n    fieldSize: number, stride: number, dilation = 1): number {\n  const effectiveFieldSize = getEffectiveFilterSize(fieldSize, dilation);\n  return Math.floor(\n      (inputShape[0] * (stride - 1) - stride + effectiveFieldSize) / 2);\n}\n\nfunction parseTupleParam(param: number|number[]): [number, number, number] {\n  if (typeof param === 'number') {\n    return [param, param, param];\n  }\n  if (param.length === 2) {\n    return [param[0], param[1], 1];\n  }\n  return param as [number, number, number];\n}\n\nfunction parse3TupleParam(param: number|[number, number, number]):\n    [number, number, number] {\n  return typeof param === 'number' ? [param, param, param] : param;\n}\n\n/* See https://www.tensorflow.org/api_docs/python/tf/nn/atrous_conv2d\n * Atrous convolution is equivalent to standard convolution with upsampled\n * filters with effective_filter_height =\n * filter_height + (filter_height - 1) * (dilation - 1)\n * and effective_filter_width =\n * filter_width + (filter_width - 1) * (dilation - 1),\n * produced by inserting dilation - 1 zeros along consecutive elements across\n * the filters' spatial dimensions.\n * When there is a dilation, this converts a filter dimension to the\n * effective filter dimension, so it can be used in a standard convolution.\n */\nfunction getEffectiveFilterSize(filterSize: number, dilation: number) {\n  if (dilation <= 1) {\n    return filterSize;\n  }\n\n  return filterSize + (filterSize - 1) * (dilation - 1);\n}\n\nfunction getPadAndOutInfo(\n    pad: 'same'|'valid'|number|ExplicitPadding, inHeight: number,\n    inWidth: number, strideHeight: number, strideWidth: number,\n    filterHeight: number, filterWidth: number,\n    roundingMode: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|\n    'channelsLast'): {padInfo: PadInfo, outHeight: number, outWidth: number} {\n  let padInfo: PadInfo;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {top: pad, bottom: pad, left: pad, right: pad, type: padType};\n    const outShape = computeOutputShape2D(\n        [inHeight, inWidth], filterHeight, strideHeight, pad, roundingMode);\n    outHeight = outShape[0];\n    outWidth = outShape[1];\n  } else if (pad === 'same') {\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongHeight =\n        Math.max(0, (outHeight - 1) * strideHeight + filterHeight - inHeight);\n    const padAlongWidth =\n        Math.max(0, (outWidth - 1) * strideWidth + filterWidth - inWidth);\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n    padInfo = {top, bottom, left, right, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {top: 0, bottom: 0, left: 0, right: 0, type: 'VALID'};\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else if (typeof pad === 'object') {\n    const top = dataFormat === 'channelsLast' ? pad[1][0] : pad[2][0];\n    const bottom = dataFormat === 'channelsLast' ? pad[1][1] : pad[2][1];\n    const left = dataFormat === 'channelsLast' ? pad[2][0] : pad[3][0];\n    const right = dataFormat === 'channelsLast' ? pad[2][1] : pad[3][1];\n    const padType = (top === 0 && bottom === 0 && left === 0 && right === 0) ?\n        'VALID' :\n        'EXPLICIT';\n    padInfo = {top, bottom, left, right, type: padType};\n    outHeight = conditionalRound(\n        (inHeight - filterHeight + top + bottom) / strideHeight + 1,\n        roundingMode);\n    outWidth = conditionalRound(\n        (inWidth - filterWidth + left + right) / strideWidth + 1, roundingMode);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outHeight, outWidth};\n}\n\nfunction get3DPadAndOutInfo(\n    pad: 'same'|'valid'|number, inDepth: number, inHeight: number,\n    inWidth: number, strideDepth: number, strideHeight: number,\n    strideWidth: number, filterDepth: number, filterHeight: number,\n    filterWidth: number, roundingMode?: 'floor'|'round'|'ceil'): {\n  padInfo: PadInfo3D,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number\n} {\n  let padInfo: PadInfo3D;\n  let outDepth: number;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {\n      top: pad,\n      bottom: pad,\n      left: pad,\n      right: pad,\n      front: pad,\n      back: pad,\n      type: padType\n    };\n    const outShape = computeOutputShape4D(\n        [inDepth, inHeight, inWidth, 1], filterDepth, 1, strideDepth, pad,\n        roundingMode);\n    outDepth = outShape[0];\n    outHeight = outShape[1];\n    outWidth = outShape[2];\n  } else if (pad === 'same') {\n    outDepth = Math.ceil(inDepth / strideDepth);\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongDepth = (outDepth - 1) * strideDepth + filterDepth - inDepth;\n    const padAlongHeight =\n        (outHeight - 1) * strideHeight + filterHeight - inHeight;\n    const padAlongWidth = (outWidth - 1) * strideWidth + filterWidth - inWidth;\n    const front = Math.floor(padAlongDepth / 2);\n    const back = padAlongDepth - front;\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n\n    padInfo = {top, bottom, left, right, front, back, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {\n      top: 0,\n      bottom: 0,\n      left: 0,\n      right: 0,\n      front: 0,\n      back: 0,\n      type: 'VALID'\n    };\n    outDepth = Math.ceil((inDepth - filterDepth + 1) / strideDepth);\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outDepth, outHeight, outWidth};\n}\n\n/**\n * Rounds a value depending on the rounding mode\n * @param value\n * @param roundingMode\n */\nfunction conditionalRound(\n    value: number, roundingMode?: 'floor'|'round'|'ceil') {\n  if (!roundingMode) {\n    return value;\n  }\n  switch (roundingMode) {\n    case 'round':\n      // used for Caffe Conv\n      return Math.round(value);\n    case 'ceil':\n      // used for Caffe Pool\n      return Math.ceil(value);\n    case 'floor':\n      return Math.floor(value);\n    default:\n      throw new Error(`Unknown roundingMode ${roundingMode}`);\n  }\n}\n\nexport function tupleValuesAreOne(param: number|number[]): boolean {\n  const [dimA, dimB, dimC] = parseTupleParam(param);\n  return dimA === 1 && dimB === 1 && dimC === 1;\n}\n\nexport function eitherStridesOrDilationsAreOne(\n    strides: number|number[], dilations: number|number[]): boolean {\n  return tupleValuesAreOne(strides) || tupleValuesAreOne(dilations);\n}\n\n/**\n * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to\n *    'channelsLast'|'channelsFirst'\n * @param dataFormat in 'NHWC'|'NCHW' mode\n * @return dataFormat in 'channelsLast'|'channelsFirst' mode\n * @throws unknown dataFormat\n */\nexport function convertConv2DDataFormat(dataFormat: 'NHWC'|'NCHW'):\n    'channelsLast'|'channelsFirst' {\n  if (dataFormat === 'NHWC') {\n    return 'channelsLast';\n  } else if (dataFormat === 'NCHW') {\n    return 'channelsFirst';\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D average pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction avgPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'avgPool', 'float32');\n  const dilations = 1;\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in avgPool: x must be rank 4 but got rank ${x4D.rank}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    save([x4D]);\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      return x4D.clone();\n    }\n\n    return backend.avgPool(x4D, convInfo);\n  };\n\n  const inputs: AvgPoolInputs = {x: x4D};\n\n  const attrs: AvgPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, $x.dtype);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool = op({avgPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {AvgPool3D, AvgPool3DAttrs, AvgPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D average pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.avgPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates:\n *     `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction avgPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'avgPool3d', 'float32');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in avgPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in avgPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n\n    save([x5D]);\n\n    return backend.avgPool3d(x5D, convInfo);\n  };\n\n  const inputs: AvgPool3DInputs = {x: x5D};\n\n  const attrs: AvgPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool3D,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, x5D.dtype);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3d = op({avgPool3d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\nexport function assertParamsConsistent(shapes: number[][], axis: number) {\n  const rank = shapes[0].length;\n  shapes.forEach((shape, i) => {\n    util.assert(\n        shape.length === rank,\n        () =>\n            `Error in concat${rank}D: rank of tensors[${i}] must be the same ` +\n            `as the rank of the rest (${rank})`);\n  });\n\n  util.assert(\n      axis >= 0 && axis < rank,\n      () => `Error in concat${rank}D: axis must be between 0 and ${rank - 1}.`);\n\n  const firstShape = shapes[0];\n  shapes.forEach((shape, i) => {\n    for (let r = 0; r < rank; r++) {\n      util.assert(\n          (r === axis) || (shape[r] === firstShape[r]),\n          () => `Error in concat${rank}D: Shape of tensors[${i}] (${shape}) ` +\n              `does not match the shape of the rest (${firstShape}) ` +\n              `along the non-concatenated axis ${i}.`);\n    }\n  });\n}\n\nexport function computeOutShape(shapes: number[][], axis: number): number[] {\n  const outputShape = shapes[0].slice();\n  for (let i = 1; i < shapes.length; i++) {\n    outputShape[axis] += shapes[i][axis];\n  }\n  return outputShape;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Concat, ConcatAttrs, ConcatInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, parseAxisParam, sizeFromShape} from '../util';\n\nimport {assertParamsConsistent, computeOutShape} from './concat_util';\nimport {op} from './operation';\nimport {tensor} from './tensor';\n\n/**\n * Concatenates a list of `tf.Tensor`s along a given axis.\n *\n * The tensors ranks and types must match, and their sizes must match in all\n * dimensions except `axis`.\n *\n * Also available are stricter rank-specific methods that assert that\n * `tensors` are of the given rank:\n *   - `tf.concat1d`\n *   - `tf.concat2d`\n *   - `tf.concat3d`\n *   - `tf.concat4d`\n *\n * Except `tf.concat1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * a.concat(b).print();  // or a.concat(b)\n * ```\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.concat([a, b, c]).print();\n * ```\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [10, 20]]);\n * const b = tf.tensor2d([[3, 4], [30, 40]]);\n * const axis = 1;\n * tf.concat([a, b], axis).print();\n * ```\n * @param tensors A list of tensors to concatenate.\n * @param axis The axis to concate along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction concat_<T extends Tensor>(tensors: Array<T|TensorLike>, axis = 0): T {\n  assert(tensors.length >= 1, () => 'Pass at least one tensor to concat');\n\n  let $tensors = convertToTensorArray(tensors, 'tensors', 'concat');\n\n  if ($tensors[0].dtype === 'complex64') {\n    $tensors.forEach(tensor => {\n      if (tensor.dtype !== 'complex64') {\n        throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${tensor.dtype}. `);\n      }\n    });\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $axis = parseAxisParam(axis, $tensors[0].shape)[0];\n    const outShape = computeOutShape($tensors.map(t => t.shape), $axis);\n    if (sizeFromShape(outShape) === 0) {\n      return tensor([], outShape) as T;\n    }\n    // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n    $tensors = $tensors.filter(t => t.size > 0);\n    if ($tensors.length === 1) {\n      return $tensors[0];\n    }\n\n    const shapes = $tensors.map(t => t.shape);\n    assertParamsConsistent(shapes, $axis);\n\n    const res = backend.concat($tensors, $axis);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: ConcatInputs = $tensors;\n  const attr: ConcatAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Concat,\n             attr as {} as NamedAttrMap) as T;\n}\n\nexport const concat = op({concat_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sigmoid, SigmoidInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sigmoid element-wise, `1 / (1 + exp(-x))`\n *\n * ```js\n * const x = tf.tensor1d([0, -1, 2, -3]);\n *\n * x.sigmoid().print();  // or tf.sigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sigmoid');\n\n  const inputs: SigmoidInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sigmoid($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sigmoid);\n}\nexport const sigmoid = op({sigmoid_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Slice, SliceAttrs, SliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as slice_util from './slice_util';\n\n/**\n * Extracts a slice from a `tf.Tensor` starting at coordinates `begin`\n * and is of size `size`.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `x` is of the given rank:\n *   - `tf.slice1d`\n *   - `tf.slice2d`\n *   - `tf.slice3d`\n *   - `tf.slice4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.slice([1], [2]).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * x.slice([1, 0], [1, 2]).print();\n * ```\n * @param x The input `tf.Tensor` to slice from.\n * @param begin The coordinates to start the slice from. The length can be\n *     less than the rank of x - the rest of the axes will have implicit 0 as\n *     start. Can also be a single number, in which case it specifies the\n *     first axis.\n * @param size The size of the slice. The length can be less than the rank of\n *     x - the rest of the axes will have implicit -1. A value of -1 requests\n *     the rest of the dimensions in the axis. Can also be a single number,\n *     in which case it specifies the size of the first axis.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction slice_<R extends Rank, T extends Tensor<R>>(\n    x: T|TensorLike, begin: number|number[], size?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'slice');\n\n  if ($x.rank === 0) {\n    throw new Error('Slicing scalar is not possible');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const [begin_, size_] = slice_util.parseSliceParams($x, begin, size);\n    slice_util.assertParamsValid($x, begin_, size_);\n    save([$x]);\n    return backend.slice($x, begin_, size_);\n  };\n\n  const inputs: SliceInputs = {x: $x};\n  const attrs: SliceAttrs = {begin, size};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Slice,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const slice = op({slice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tanh, TanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic tangent of the input `tf.Tensor` element-wise: `tanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, 70]);\n *\n * x.tanh().print();  // or tf.tanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tanh');\n\n  const inputs: TanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const y = backend.tanh($x);\n    save([y]);\n    return y;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tanh);\n}\nexport const tanh = op({tanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {add} from './add';\nimport {concat} from './concat';\nimport {matMul} from './mat_mul';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {slice} from './slice';\nimport {tanh} from './tanh';\n\n/**\n * Computes the next state and output of a BasicLSTMCell.\n *\n * Returns `[newC, newH]`.\n *\n * Derived from tf.contrib.rnn.BasicLSTMCell.\n *\n * @param forgetBias Forget bias for the cell.\n * @param lstmKernel The weights for the cell.\n * @param lstmBias The bias for the cell.\n * @param data The input to the cell.\n * @param c Previous cell state.\n * @param h Previous cell output.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction basicLSTMCell_(\n    forgetBias: Scalar|TensorLike, lstmKernel: Tensor2D|TensorLike,\n    lstmBias: Tensor1D|TensorLike, data: Tensor2D|TensorLike,\n    c: Tensor2D|TensorLike, h: Tensor2D|TensorLike): [Tensor2D, Tensor2D] {\n  const $forgetBias =\n      convertToTensor(forgetBias, 'forgetBias', 'basicLSTMCell');\n  const $lstmKernel =\n      convertToTensor(lstmKernel, 'lstmKernel', 'basicLSTMCell');\n  const $lstmBias = convertToTensor(lstmBias, 'lstmBias', 'basicLSTMCell');\n  const $data = convertToTensor(data, 'data', 'basicLSTMCell');\n  const $c = convertToTensor(c, 'c', 'basicLSTMCell');\n  const $h = convertToTensor(h, 'h', 'basicLSTMCell');\n\n  const combined = concat([$data, $h], 1);\n  const weighted = matMul(combined, $lstmKernel);\n  const res: Tensor2D = add(weighted, $lstmBias);\n\n  // i = input_gate, j = new_input, f = forget_gate, o = output_gate\n  const batchSize = res.shape[0];\n  const sliceCols = res.shape[1] / 4;\n  const sliceSize: [number, number] = [batchSize, sliceCols];\n  const i = slice(res, [0, 0], sliceSize);\n  const j = slice(res, [0, sliceCols], sliceSize);\n  const f = slice(res, [0, sliceCols * 2], sliceSize);\n  const o = slice(res, [0, sliceCols * 3], sliceSize);\n\n  const newC: Tensor2D =\n      add(mul(sigmoid(i), tanh(j)),\n          mul($c, sigmoid(add($forgetBias, f)) as Tensor2D));\n  const newH: Tensor2D = mul(tanh(newC), sigmoid(o));\n  return [newC, newH];\n}\n\nexport const basicLSTMCell = op({basicLSTMCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchToSpaceND, BatchToSpaceNDAttrs, BatchToSpaceNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation reshapes the \"batch\" dimension 0 into `M + 1` dimensions of\n * shape `blockShape + [batch]`, interleaves these blocks back into the grid\n * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with\n * the same rank as the input. The spatial dimensions of this intermediate\n * result are then optionally cropped according to `crops` to produce the\n * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise\n * description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);\n * const blockShape = [2, 2];\n * const crops = [[0, 0], [0, 0]];\n *\n * x.batchToSpaceND(blockShape, crops).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param crops A 2-D array.  Must have shape `[M, 2]`, all values must be >= 0.\n * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input\n * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required\n * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,\n * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,\n * x.shape[N-1]]`\n *\n * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /\n * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`\n * according to `crops` to produce the output of shape: `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],\n * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -\n * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction batchToSpaceND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], crops: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'batchToSpaceND');\n  const prod = blockShape.reduce((a, b) => a * b);\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank is ${$x.rank} but should be > than blockShape.length ${\n          blockShape.length}`);\n\n  util.assert(\n      crops.length === blockShape.length,\n      () => `crops.length is ${\n          crops.length} but should be equal to blockShape.length  ${\n          blockShape.length}`);\n\n  util.assert(\n      $x.shape[0] % prod === 0,\n      () => `input tensor batch is ${\n                $x.shape[0]} but is not divisible by the product of ` +\n          `the elements of blockShape ${blockShape.join(' * ')} === ${prod}`);\n\n  const forward: ForwardFunc<T> = backend => {\n    return backend.batchToSpaceND($x, blockShape, crops);\n  };\n\n  const inputs: BatchToSpaceNDInputs = {x: $x};\n  const attrs: BatchToSpaceNDAttrs = {blockShape, crops};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      BatchToSpaceND, attrs as {} as NamedAttrMap);\n}\n\nexport const batchToSpaceND = op({batchToSpaceND_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, Tensor4D} from '../tensor';\nimport {Rank} from '../types';\nimport {reshape} from './reshape';\n\nexport function xAs4D<R extends Rank>(x: Tensor<R>) {\n  let x4D: Tensor4D;\n  if (x.rank === 0 || x.rank === 1) {\n    x4D = reshape(x, [1, 1, 1, x.size]);\n  } else if (x.rank === 2) {\n    x4D = reshape(x, [1, 1, x.shape[0], x.shape[1]]);\n  } else if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  } else {\n    x4D = x as Tensor4D;\n  }\n\n  return x4D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {xAs4D} from './batchnorm_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Batch normalization.\n *\n * As described in\n * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).\n *\n * Mean, variance, scale, and offset can be of two shapes:\n *   - The same shape as the input.\n *   - In the common case, the depth dimension is the last dimension of x, so\n *     the values would be an `tf.Tensor1D` of shape [depth].\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that parameters passed are of given rank\n *   - `tf.batchNorm2d`\n *   - `tf.batchNorm3d`\n *   - `tf.batchNorm4d`\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction batchNorm_<R extends Rank>(\n    x: Tensor<R>|TensorLike, mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor<R>|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor<R>|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n\n  util.assert(\n      $mean.rank === $variance.rank,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      $offset == null || $mean.rank === $offset.rank,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      $scale == null || $mean.rank === $scale.rank,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  const x4D: Tensor4D = xAs4D($x);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([x4D, $mean, $variance, $scale]);\n\n    return backend.batchNorm(\n        x4D, as1DOr4D($mean), as1DOr4D($variance), as1DOr4D($offset),\n        as1DOr4D($scale), varianceEpsilon);\n  };\n\n  const inputs: FusedBatchNormInputs = {\n    x: x4D,\n    scale: $scale,\n    offset: $offset,\n    mean: $mean,\n    variance: $variance\n  };\n\n  const attrs: FusedBatchNormAttrs = {varianceEpsilon};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      FusedBatchNorm, attrs as {} as NamedAttrMap);\n\n  return reshape(res, $x.shape);\n}\n\nfunction as1DOr4D(x: Tensor): Tensor4D|Tensor1D {\n  if (x == null) {\n    return null;\n  }\n  if (x.rank === 0) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [x.size]) as Tensor1D;\n  } else if (x.rank === 1) {\n    return x as Tensor1D;\n  } else if (x.rank === 2) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, 1, x.shape[0], x.shape[1]]) as Tensor4D;\n  } else if (x.rank === 3) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]) as Tensor4D;\n  }\n  return x as Tensor4D;\n}\n\nexport const batchNorm = op({batchNorm_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 2D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm2d_(\n    x: Tensor2D|TensorLike, mean: Tensor2D|Tensor1D|TensorLike,\n    variance: Tensor2D|Tensor1D|TensorLike,\n    offset?: Tensor2D|Tensor1D|TensorLike, scale?: Tensor2D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor2D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor2D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 2,\n      () => `Error in batchNorm2D: x must be rank 2 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 2 || $mean.rank === 1,\n      () => `Error in batchNorm2D: mean must be rank 2 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 2 || $variance.rank === 1,\n      () => `Error in batchNorm2D: variance must be rank 2 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 2 || $scale.rank === 1,\n        () => `Error in batchNorm2D: scale must be rank 2 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 2 || $offset.rank === 1,\n        () => `Error in batchNorm2D: offset must be rank 2 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm2d = op({batchNorm2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 3D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm3d_(\n    x: Tensor3D|TensorLike, mean: Tensor3D|Tensor1D|TensorLike,\n    variance: Tensor3D|Tensor1D|TensorLike,\n    offset?: Tensor3D|Tensor1D|TensorLike, scale?: Tensor3D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor3D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor3D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 3,\n      () => `Error in batchNorm3D: x must be rank 3 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 3 || $mean.rank === 1,\n      () => `Error in batchNorm3D: mean must be rank 3 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 3 || $variance.rank === 1,\n      () => `Error in batchNorm3D: variance must be rank 3 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 3 || $scale.rank === 1,\n        () => `Error in batchNorm3D: scale must be rank 3 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 3 || $offset.rank === 1,\n        () => `Error in batchNorm3D: offset must be rank 3 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm3d = op({batchNorm3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 4D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm4d_(\n    x: Tensor4D|TensorLike, mean: Tensor4D|Tensor1D|TensorLike,\n    variance: Tensor4D|Tensor1D|TensorLike,\n    offset?: Tensor4D|Tensor1D|TensorLike, scale?: Tensor4D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor4D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor4D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 4,\n      () => `Error in batchNorm4D: x must be rank 4 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 4 || $mean.rank === 1,\n      () => `Error in batchNorm4D: mean must be rank 4 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 4 || $variance.rank === 1,\n      () => `Error in batchNorm4D: variance must be rank 4 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 4 || $scale.rank === 1,\n        () => `Error in batchNorm4D: scale must be rank 4 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 4 || $offset.rank === 1,\n        () => `Error in batchNorm4D: offset must be rank 4 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm4d = op({batchNorm4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE} from '../engine';\nimport {BroadcastTo, BroadCastToAttrs, BroadcastToInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Broadcast an array to a compatible shape NumPy-style.\n *\n * The tensor's shape is compared to the broadcast shape from end to beginning.\n * Ones are prepended to the tensor's shape until is has the same length as\n * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is\n * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then\n * the input tensor is tiled N times along that axis (using tf.tile).\n *\n * @param input The tensor that is to be broadcasted.\n * @param shape The input is to be broadcast to this shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction broadcastTo_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  let input = convertToTensor(x, 'broadcastTo', 'x');\n  const xShape = input.shape;\n\n  if (shape.some(d => !(d > 0) || d % 1 !== 0)) {\n    throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);\n  }\n\n  if (shape.length < input.rank) {\n    throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${\n        input.rank}.`);\n  }\n\n  if (shape.length > input.rank) {\n    const newShape = input.shape.slice();\n    while (newShape.length < shape.length) {\n      newShape.unshift(1);\n    }\n    input = reshape(input, newShape);\n  }\n\n  const inputShape = input.shape;\n  const reps: number[] = Array.from(shape);\n  for (let i = shape.length - 1; i >= 0; i--) {\n    if (inputShape[i] === shape[i]) {\n      reps[i] = 1;\n    } else if (input.shape[i] !== 1) {\n      throw new Error(\n          `broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);\n    }\n  }\n  const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);\n\n  if (axes.length === 0) {\n    return clone(input) as Tensor<R>;\n  }\n\n  const forward = (backend: KernelBackend) => backend.tile(input, reps);\n\n  const inputs: BroadcastToInputs = {x: input};\n  const attrs: BroadCastToAttrs = {shape, inputShape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             BroadcastTo, attrs as unknown as NamedAttrMap) as Tensor<R>;\n}\n\nexport const broadcastTo = op({broadcastTo_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Ceil, CeilInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes ceiling of input `tf.Tensor` element-wise: `ceil(x)`\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.ceil().print();  // or tf.ceil(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction ceil_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'ceil');\n\n  const inputs: CeilInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.ceil($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Ceil);\n}\nexport const ceil = op({ceil_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {ClipByValue, ClipByValueAttrs, ClipByValueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)\n * ```\n * @param x The input tensor.\n * @param clipValueMin Lower-bound of range to be clipped to.\n * @param clipValueMax Upper-bound of range to be clipped to.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction clipByValue_<T extends Tensor>(\n    x: T|TensorLike, clipValueMin: number, clipValueMax: number): T {\n  const $x = convertToTensor(x, 'x', 'clipByValue');\n  util.assert(\n      (clipValueMin <= clipValueMax),\n      () => `Error in clip: min (${clipValueMin}) must be ` +\n          `less than or equal to max (${clipValueMax}).`);\n\n  const inputs: ClipByValueInputs = {x: $x};\n  const attrs: ClipByValueAttrs = {clipValueMin, clipValueMax};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const res = backend.clip($x, clipValueMin, clipValueMax);\n        save([$x]);\n        return res;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, ClipByValue,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const clipByValue = op({clipByValue_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor1D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(3) = |r1, g1, b1|\n * B: shape(2) = |r2, g2|\n * C = tf.concat1d([A, B]) == |r1, g1, b1, r2, g2|\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @return The concatenated array.\n */\nfunction concat1d_(tensors: Array<Tensor1D|TensorLike>): Tensor1D {\n  return concat(tensors, 0 /* axis */);\n}\n\nexport const concat1d = op({concat1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor2D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *\n * B: shape(2, 3) = | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * C = tf.concat2d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *                  | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * if axis = 1:\n * C = shape(2, 6) = | r1, g1, b1, r3, g3, b3 |\n *                   | r2, g2, b2, r4, g4, b4 |\n *\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concatenate along.\n * @return The concatenated array.\n */\nfunction concat2d_(\n    tensors: Array<Tensor2D|TensorLike>, axis: number): Tensor2D {\n  return concat(tensors, axis);\n}\n\nexport const concat2d = op({concat2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor3D`s along an axis.\n * See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *\n * B: shape(2, 1, 3) = | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * C = tf.concat3d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *                     | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * if axis = 1:\n * C: shape(2, 2, 3) = | r1, g1, b1, r3, g3, b3 |\n *                     | r2, g2, b2, r4, g4, b4 |\n *\n * if axis = 2:\n * C = shape(2, 1, 6) = | r1, g1, b1, r3, g3, b3 |\n *                      | r2, g2, b2, r4, g4, b4 |\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat3d_(\n    tensors: Array<Tensor3D|TensorLike>, axis: number): Tensor3D {\n  return concat(tensors, axis);\n}\n\nexport const concat3d = op({concat3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor4D`s along an axis.\n * See `concat` for details.\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat4d_(\n    tensors: Array<Tensor4D|TensorLike>, axis: number): Tensor4D {\n  return concat(tensors, axis);\n}\n\nexport const concat4d = op({concat4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2D, Conv2DAttrs, Conv2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 2D convolution over the input x.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2d: input must be rank 4, but got rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  util.assert(\n      inDepth === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${inDepth}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        false, $dataFormat);\n    const res = backend.conv2d(x4D, $filter, convInfo);\n\n    save([x4D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DInputs = {x: x4D, filter: $filter};\n  const attrs:\n      Conv2DAttrs = {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv2D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2d = op({conv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D, Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 1D convolution over the input x.\n *\n * @param x The input tensor, of rank 3 or rank 2, of shape\n *     `[batch, width, inChannels]`. If rank 2, batch of 1 is assumed.\n * @param filter The filter, rank 3, of shape\n *     `[filterWidth, inDepth, outDepth]`.\n * @param stride The number of entries by which the filter is moved right at\n *     each step.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from \"NWC\", \"NCW\". Defaults to \"NWC\",\n *     the data is stored in the order of [batch, in_width, in_channels]. Only\n *     \"NWC\" is currently supported.\n * @param dilation The dilation rate in which we sample input values in\n *     atrous convolution. Defaults to `1`. If it is greater than 1, then\n *     stride must be `1`.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv1d_<T extends Tensor2D|Tensor3D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike, stride: number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv1d');\n  const $filter = convertToTensor(filter, 'filter', 'conv1d');\n\n  let x3D = $x as Tensor3D;\n  let reshapedTo3D = false;\n  if ($x.rank === 2) {\n    reshapedTo3D = true;\n    x3D = reshape($x, [1, $x.shape[0], $x.shape[1]]);\n  }\n\n  util.assert(\n      x3D.rank === 3,\n      () => `Error in conv1d: input must be rank 3, but got rank ${x3D.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in conv1d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv1d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x3D.shape[2] === $filter.shape[1],\n      () => `Error in conv1d: depth of input (${x3D.shape[2]}) must match ` +\n          `input depth for filter ${$filter.shape[1]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(stride, dilation),\n      () => 'Error in conv1D: Either stride or dilation must be 1. ' +\n          `Got stride ${stride} and dilation '${dilation}'`);\n  util.assert(\n      dataFormat === 'NWC',\n      () => `Error in conv1d: got dataFormat of ${\n          dataFormat} but only NWC is currently supported.`);\n\n  const filter4D = reshape(\n      $filter, [1, $filter.shape[0], $filter.shape[1], $filter.shape[2]]);\n  const input4D = reshape(x3D, [x3D.shape[0], 1, x3D.shape[1], x3D.shape[2]]);\n  const strides: [number, number] = [1, stride];\n  const dilations: [number, number] = [1, dilation];\n\n  const conv2dDataFormat = 'NHWC';\n\n  const res = conv2d(\n      (input4D as Tensor4D), (filter4D as Tensor4D), strides, pad,\n      conv2dDataFormat, dilations, dimRoundingMode);\n\n  if (reshapedTo3D) {\n    return reshape(res, [res.shape[2], res.shape[3]]) as T;\n  }\n\n  return reshape(res, [res.shape[0], res.shape[2], res.shape[3]]) as T;\n}\n\nexport const conv1d = op({conv1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs, Conv2DBackpropInputInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 2D convolution.\n *\n * @param xShape The shape of the input: [batch, height, width, inDepth].\n * If length of 3, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 4 or rank 3 of shape\n *   `[batch, outHeight, outWidth, outDepth]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction conv2DBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape4D = xShape as [number, number, number, number];\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n    xShape4D = [1, xShape[0], xShape[1], xShape[2]];\n  }\n\n  util.assert(\n      xShape4D.length === 4,\n      () =>\n          `Error in conv2dDerInput: inShape must be length 4, but got length ` +\n          `${xShape4D.length}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerInput: dy must be rank 4, but got ` +\n          `rank ${dy4D.rank}`);\n  util.assert(\n      filter.rank === 4,\n      () => `Error in conv2dDerInput: filter must be rank 4, but got ` +\n          `rank ${filter.rank}`);\n  const inDepth = dataFormat === 'NHWC' ? xShape4D[3] : xShape4D[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filter.shape[2],\n      () => `Error in conv2dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[2]}.`);\n  util.assert(\n      outDepth === filter.shape[3],\n      () => `Error in conv2dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[3]}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerInput: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const dilations = 1;\n\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        xShape4D, filter.shape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    const res = backend.conv2dDerInput(dy4D, filter, convInfo);\n\n    save([dy4D, filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DBackpropInputInputs = {dy: dy4D, filter};\n\n  const attrs: Conv2DBackpropInputAttrs =\n      {strides, pad, dataFormat, dimRoundingMode, inputShape: xShape4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      Conv2DBackpropInput, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2DBackpropInput = op({conv2DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv2DBackpropInput} from './conv2d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 2D convolution of an image, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 4 or rank 3, of shape\n *   `[batch, height, width, inDepth]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 4 or rank 3:\n *     `[batch, height, width, outDepth]`. If rank 3, batch of 1 is assumed.\n * @param strides The strides of the original convolution:\n *     `[strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n * @param dimRoundingMode The rounding mode used when computing output\n *    dimensions if pad is a number. If none is provided, it will not round\n *    and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2dTranspose_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv2dTranspose');\n\n  return conv2DBackpropInput(\n      outputShape, $x, $filter, strides, pad, 'NHWC', dimRoundingMode);\n}\n\nexport const conv2dTranspose = op({conv2dTranspose_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3D, Conv3DAttrs, Conv3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {eitherStridesOrDilationsAreOne} from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 3D convolution over the input x.\n *\n * @param x The input tensor, of rank 5 or rank 4, of shape\n *     `[batch, depth, height, width, channels]`. If rank 4,\n * batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inChannels, outChannels]`.\n *      inChannels must match between input and filter.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationDepth, dilationHeight,\n *     dilationWidth]` in which we sample input values across the height\n *     and width dimensions in atrous convolution. Defaults to `[1, 1, 1]`.\n *     If `dilations` is a single number, then\n *     `dilationDepth == dilationHeight == dilationWidth`. If it is greater\n *     than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    strides: [number, number, number]|number, pad: 'valid'|'same',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations: [number, number, number]|number = [1, 1, 1]): T {\n  const $x = convertToTensor(x, 'x', 'conv3d');\n  const $filter = convertToTensor(filter, 'filter', 'conv3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3d: input must be rank 5, but got rank ${x5D.rank}.`);\n  util.assert(\n      $filter.rank === 5,\n      () => `Error in conv3d: filter must be rank 5, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x5D.shape[4] === $filter.shape[3],\n      () => `Error in conv3d: depth of input (${x5D.shape[4]}) must match ` +\n          `input depth for filter ${$filter.shape[3]}.`);\n  util.assert(\n      eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv3D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in conv3d: got dataFormat of ${\n          dataFormat} but only NDHWC is currently supported.`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, $filter.shape, strides, dilations, pad);\n    const res = backend.conv3d(x5D, $filter, convInfo);\n\n    save([x5D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv3DInputs = {x: x5D, filter: $filter};\n\n  const attrs: Conv3DAttrs = {strides, pad, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3d = op({conv3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropInputAttrs, Conv3DBackpropInputInputs, Conv3DBackpropInputV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 3D convolution.\n *\n * @param xShape The shape of the input: [batch, depth, height, width,\n * in_channels]. If length of 4, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 5 or rank 4 of shape\n *   `[batch, outDepth, outHeight, outWidth, in_channels]`.\n * If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n */\nfunction conv3DBackpropInput_<T extends Tensor4D|Tensor5D>(\n    xShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    dy: T, filter: Tensor5D, strides: [number, number, number]|number,\n    pad: 'valid'|'same'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape5D = xShape as [number, number, number, number, number];\n  let dy5D = dy as Tensor5D;\n  let reshapedTo5D = false;\n  if (dy.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n    xShape5D = [1, xShape[0], xShape[1], xShape[2], xShape[3]];\n  }\n\n  const inDepth = xShape5D[4];\n  const outDepth = dy5D.shape[4];\n  util.assert(\n      xShape5D.length === 5,\n      () =>\n          `Error in conv3dDerInput: inShape must be length 5, but got length ` +\n          `${xShape5D.length}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerInput: dy must be rank 5, but got ` +\n          `rank ${dy5D.rank}`);\n  util.assert(\n      filter.rank === 5,\n      () => `Error in conv3dDerInput: filter must be rank 5, but got ` +\n          `rank ${filter.rank}`);\n  util.assert(\n      inDepth === filter.shape[3],\n      () => `Error in conv3dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[3]}.`);\n  util.assert(\n      outDepth === filter.shape[4],\n      () => `Error in conv3dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[4]}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        xShape5D, filter.shape, strides, dilations, pad);\n\n    return backend.conv3dDerInput(dy5D, filter, convInfo);\n  };\n\n  const inputs: Conv3DBackpropInputInputs = {dy: dy5D};\n\n  const attrs: Conv3DBackpropInputAttrs = {pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, Conv3DBackpropInputV2,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3DBackpropInput = op({conv3DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D, Tensor5D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv3DBackpropInput} from './conv3d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 3D convolution of a volume, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 5 or rank 4, of shape\n *   `[batch, depth, height, width, inDepth]`. If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[depth, filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 5 or rank 4:\n *     `[batch, depth, height, width, outDepth]`. If rank 3, batch of 1 is\n *    assumed.\n * @param strides The strides of the original convolution:\n *     `[strideDepth, strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3dTranspose_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    outputShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): T {\n  const $x = convertToTensor(x, 'x', 'conv3dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv3dTranspose');\n\n  return conv3DBackpropInput(outputShape, $x, $filter, strides, pad);\n}\n\nexport const conv3dTranspose = op({conv3dTranspose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cos, CosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes cos of the input `tf.Tensor` element-wise: `cos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.cos().print();  // or tf.cos(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cos');\n\n  const inputs: CosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cos);\n}\nexport const cos = op({cos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cosh, CoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic cos of the input `tf.Tensor` element-wise: `cosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.cosh().print();  // or tf.cosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cosh');\n  const inputs: CoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cosh);\n}\nexport const cosh = op({cosh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Cumsum, CumsumAttrs, CumsumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {getAxesPermutation, getInnerMostAxes, getUndoAxesPermutation} from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the cumulative sum of a `tf.Tensor` along `axis`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4]);\n * x.cumsum().print();\n * ```\n * ```js\n * const x = tf.tensor([[1, 2], [3, 4]]);\n * x.cumsum().print();\n * ```\n *\n * @param x The input tensor to be summed.\n * @param axis The axis along which to sum. Optional. Defaults to 0.\n * @param exclusive Whether to perform exclusive cumulative sum. Optional.\n *     Defaults to false. If set to true then the sum of each tensor entry\n *     does not include its own value, but only the values previous to it\n *     along the specified axis.\n * @param reverse Whether to sum in the opposite direction. Optional.\n *     Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Scan'}\n */\nfunction cumsum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis = 0, exclusive = false, reverse = false): T {\n  const $x = convertToTensor(x, 'x', 'cumsum');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const permutation = getAxesPermutation([axis], $x.rank);\n        let permutedX = $x;\n        if (permutation != null) {\n          permutedX = transpose($x, permutation);\n        }\n        const permutedAxis = getInnerMostAxes(1, $x.rank)[0];\n        let value = backend.cumsum(permutedX, permutedAxis, exclusive, reverse);\n        save([$x]);\n\n        if (permutation != null) {\n          const reversePermutation = getUndoAxesPermutation(permutation);\n          value = transpose(value, reversePermutation);\n        }\n        return value;\n      };\n\n  const inputs: CumsumInputs = {x: $x};\n  const attrs: CumsumAttrs = {axis, exclusive, reverse};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Cumsum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const cumsum = op({cumsum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthToSpace, DepthToSpaceAttrs, DepthToSpaceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Rearranges data from depth into blocks of spatial data. More specifically,\n * this op outputs a copy of the input tensor where values from the `depth`\n * dimension are moved in spatial blocks to the `height` and `width` dimensions.\n * The attr `blockSize` indicates the input block size and how the data is\n * moved.\n *\n *  - Chunks of data of size `blockSize * blockSize` from depth are rearranged\n * into non-overlapping blocks of size `blockSize x blockSize`\n *\n *  - The width the output tensor is `inputWidth * blockSize`, whereas the\n * height is `inputHeight * blockSize`\n *\n *  - The Y, X coordinates within each block of the output image are determined\n * by the high order component of the input channel index\n *\n *  - The depth of the input tensor must be divisible by `blockSize *\n * blockSize`\n *\n * The `dataFormat` attr specifies the layout of the input and output tensors\n * with the following options: \"NHWC\": [ `batch, height, width, channels` ]\n * \"NCHW\": [ `batch, channels, height, width` ]\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 1, 1, 4]);\n * const blockSize = 2;\n * const dataFormat = \"NHWC\";\n *\n * tf.depthToSpace(x, blockSize, dataFormat).print();\n * ```\n *\n * @param x The input tensor of rank 4\n * @param blockSIze  An `int` that is `>= 2`. The size of the spatial block\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\"\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction depthToSpace_(\n    x: Tensor4D|TensorLike4D, blockSize: number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'depthToSpace') as Tensor4D;\n\n  const inputHeight = (dataFormat === 'NHWC') ? $x.shape[1] : $x.shape[2];\n  const inputWidth = (dataFormat === 'NHWC') ? $x.shape[2] : $x.shape[3];\n  const inputDepth = (dataFormat === 'NHWC') ? $x.shape[3] : $x.shape[1];\n\n  util.assert(\n      inputHeight * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputHeight} and ${blockSize}  for depthToSpace with input shape\n    ${$x.shape}`);\n\n  util.assert(\n      inputWidth * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputWidth} and ${blockSize} for depthToSpace with input shape\n        ${$x.shape}`);\n\n  util.assert(\n      (inputDepth % (blockSize * blockSize) === 0),\n      () => `Dimension size must be evenly divisible by ${\n          blockSize * blockSize} but is ${\n          inputDepth} for depthToSpace with input shape ${$x.shape}`);\n\n  const forward: ForwardFunc<Tensor4D> = backend =>\n      backend.depthToSpace($x, blockSize, dataFormat);\n\n  const inputs: DepthToSpaceInputs = {x: $x};\n  const attrs: DepthToSpaceAttrs = {blockSize, dataFormat};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      DepthToSpace, attrs as {} as NamedAttrMap);\n}\n\nexport const depthToSpace = op({depthToSpace_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs, DepthwiseConv2dNativeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Depthwise 2D convolution.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction depthwiseConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in depthwiseConv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in depthwiseConv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1];\n    }\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in depthwiseConv2d: Either strides or dilations must be ' +\n            `1. Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        true /* depthwise */);\n    const res = backend.depthwiseConv2D(x4D, $filter, convInfo);\n    save([x4D, $filter]);\n    return res;\n  };\n\n  const inputs: DepthwiseConv2dNativeInputs = {x: x4D, filter: $filter};\n  const attrs: DepthwiseConv2dNativeAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      DepthwiseConv2dNative, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2d = op({depthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Diag, DiagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a diagonal tensor with a given diagonal values.\n *\n * Given a diagonal, this operation returns a tensor with the diagonal and\n * everything else padded with zeros.\n *\n * Assume the input has dimensions `[D1,..., Dk]`, then the output is a tensor\n * of rank 2k with dimensions `[D1,..., Dk, D1,..., Dk]`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * tf.diag(x).print()\n * ```\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4, 5, 6, 6, 8], [4, 2])\n *\n * tf.diag(x).print()\n * ```\n * @param x The input tensor.\n */\nfunction diag_(x: Tensor): Tensor {\n  const $x = convertToTensor(x, 'x', 'diag');\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const flat = reshape($x, [$x.size]);\n    const result = backend.diag(flat);\n    const outShape = [...x.shape, ...x.shape];\n\n    return reshape(result, outShape);\n  };\n\n  const inputs: DiagInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Diag);\n}\n\nexport const diag = op({diag_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DAttrs, Dilation2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the grayscale dilation over the input `x`.\n *\n * @param x The input tensor, rank 3 or rank 4 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter tensor, rank 3, of shape\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat Specify the data format of the input and output data.\n *      Defaults to 'NHWC'. Only 'NHWC' is currently supported. With the\n *      default format \"NHWC\", the data is stored in the order of: [batch,\n *      height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     for atrous morphological dilation. Defaults to `[1, 1]`. If `dilations`\n *     is a single number, then `dilationHeight == dilationWidth`. If it is\n *     greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction dilation2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same',\n    dilations: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'dilation2d');\n  const $filter = convertToTensor(filter, 'filter', 'dilation2d');\n\n  util.assert(\n      $x.rank === 3 || $x.rank === 4,\n      () => `Error in dilation2d: input must be rank 3 or 4, but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in dilation2d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in dilation2d: Only NHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n    reshapedTo4D = true;\n  }\n\n  const inputs: Dilation2DInputs = {x: x4D, filter: $filter};\n  const attrs: Dilation2DAttrs = {strides, pad, dilations};\n\n  const res = ENGINE.runKernel(\n                  Dilation2D, inputs as {} as NamedTensorMap,\n                  attrs as {} as NamedAttrMap) as T;\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res;\n}\n\nexport const dilation2d = op({dilation2d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Returns the dimensions in the input shape that are broadcasted to\n * produce the provided output shape.\n *\n * The returned dimensions are 0-indexed and sorted. An example:\n * inShape = [4, 1, 3]\n * outShape = [5, 4, 3, 3]\n * result = [1]. Dimension 1 (2nd dimension of input) gets broadcasted 1 => 3.\n */\nexport function getBroadcastDims(\n    inShape: number[], outShape: number[]): number[] {\n  const inRank = inShape.length;\n  const dims: number[] = [];\n  for (let i = 0; i < inRank; i++) {\n    const dim = inRank - 1 - i;\n    const a = inShape[dim] || 1;\n    const b = outShape[outShape.length - 1 - i] || 1;\n    if (b > 1 && a === 1) {\n      dims.unshift(dim);\n    }\n  }\n  return dims;\n}\n\n/**\n * Returns the axes in the output space that should be reduced to produce\n * the input space.\n */\nexport function getReductionAxes(\n    inShape: number[], outShape: number[]): number[] {\n  const result: number[] = [];\n  for (let i = 0; i < outShape.length; i++) {\n    const inDim = inShape[inShape.length - i - 1];\n    const outAxis = outShape.length - i - 1;\n    const outDim = outShape[outAxis];\n    if (inDim == null || (inDim === 1 && outDim > 1)) {\n      result.unshift(outAxis);\n    }\n  }\n  return result;\n}\n\nexport function assertAndGetBroadcastShape(\n    shapeA: number[], shapeB: number[]): number[] {\n  const result: number[] = [];\n  const l = Math.max(shapeA.length, shapeB.length);\n\n  for (let i = 0; i < l; i++) {\n    let a = shapeA[shapeA.length - i - 1];\n    if (a == null) {\n      a = 1;\n    }\n    let b = shapeB[shapeB.length - i - 1];\n    if (b == null) {\n      b = 1;\n    }\n    if (a === 1) {\n      result.unshift(b);\n    } else if (b === 1) {\n      result.unshift(a);\n    } else if (a !== b) {\n      const errMsg = `Operands could not be broadcast together with shapes ` +\n          `${shapeA} and ${shapeB}.`;\n      throw Error(errMsg);\n    } else {\n      result.unshift(a);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Equal, EqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a == b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.equal(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction equal_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'equal');\n  let $b = convertToTensor(b, 'b', 'equal');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.equal($a, $b);\n\n  const inputs: EqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, Equal) as T;\n}\n\nexport const equal = op({equal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SelectV2, SelectV2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch} from '../util';\n\nimport {broadcastTo} from './broadcast_to';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the elements, either `a` or `b` depending on the `condition`.\n *\n * If the condition is true, select from `a`, otherwise select from `b`.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const a = tf.tensor1d([1 , 2, 3]);\n * const b = tf.tensor1d([-1, -2, -3]);\n *\n * a.where(cond, b).print();\n * ```\n *\n * @param condition The input condition. Must be of dtype bool.\n * @param a If `condition` is rank 1, `a` may have a higher rank but\n *     its first dimension must match the size of `condition`.\n * @param b A tensor with the same dtype as `a` and with shape that is\n *     compatible with `a`.\n * @return A tensor with same dtype as `a` and `b`, and shape that is\n *     broadcastable from `a` and `b`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction where_<T extends Tensor>(\n    condition: Tensor|TensorLike, a: T|TensorLike, b: T|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'where');\n  const $b = convertToTensor(b, 'b', 'where');\n  const $condition = convertToTensor(condition, 'condition', 'where', 'bool');\n  // TODO: move this logic to forward function when the broadcastTo op is\n  // implemented in WASM.\n  // Find the broadcastable shape for $a and $b.\n  const broadcastShape = assertAndGetBroadcastShape($a.shape, $b.shape);\n  const $broadcastedA = broadcastTo($a, broadcastShape);\n  const $broadcastedB = broadcastTo($b, broadcastShape);\n  if ($condition.rank === 1) {\n    // If condition rank is 1, then the first dimension must match the size of\n    // condition.\n    assert(\n        $condition.shape[0] === $a.shape[0],\n        () => 'The first dimension of `a` must match the size of `condition`.');\n  }\n\n  if ($condition.rank !== 1) {\n    // A must have the same shape as condition.\n    assertShapesMatch(\n        $condition.shape, $broadcastedB.shape, 'Error in where: ');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.select($condition, $broadcastedA, $broadcastedB);\n    save([$condition]);\n    return res;\n  };\n  const inputs: SelectV2Inputs = {\n    condition: $condition,\n    t: $broadcastedA,\n    e: $broadcastedB\n  };\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* gradient */,\n             SelectV2) as T;\n}\n\nexport const where = op({where_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {ZerosLike, ZerosLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.zerosLike(x).print();\n * ```\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction zerosLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'zerosLike');\n  const inputs: ZerosLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             backend => backend.zerosLike($x), inputs as {} as NamedTensorMap,\n             null /* grad */, ZerosLike) as T;\n}\nexport const zerosLike = op({zerosLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {div} from './div';\nimport {equal} from './equal';\nimport {op} from './operation';\nimport {where} from './where';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting. Return 0\n * if denominator is 0.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n * const c = tf.tensor1d([0, 0, 0, 0]);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n * const c = tf.scalar(0);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction divNoNan_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  // TODO: Make this into its own kernel.\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const divResult = div($a, $b);\n  const zeros = zerosLike(divResult);\n  const bEqualsZero = equal($b, zeros);\n  return where(bEqualsZero, zeros, divResult) as T;\n}\n\nexport const divNoNan = op({divNoNan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor,} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices and/or vectors, `t1` and `t2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor2d([[1, 2], [3, 4]]);\n * const c = tf.tensor2d([[1, 2, 3], [4, 5, 6]]);\n *\n * a.dot(b).print();  // or tf.dot(a, b)\n * b.dot(a).print();\n * b.dot(c).print();\n * ```\n * @param t1 The first tensor in the dot operation.\n * @param t2 The second tensor in the dot operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction dot_(t1: Tensor|TensorLike, t2: Tensor|TensorLike): Tensor {\n  const $t1 = convertToTensor(t1, 't1', 'dot');\n  const $t2 = convertToTensor(t2, 't2', 'dot');\n\n  util.assert(\n      ($t1.rank === 1 || $t1.rank === 2) && ($t2.rank === 1 || $t2.rank === 2),\n      () => `Error in dot: inputs must all be rank 1 or 2, but got ranks ` +\n          `${$t1.rank} and ${$t2.rank}.`);\n\n  const t1Inner = ($t1.rank === 1 ? $t1.size : $t1.shape[1]);\n  const t2Inner = ($t2.rank === 1 ? $t2.size : $t2.shape[0]);\n\n  util.assert(\n      t1Inner === t2Inner,\n      () => `Error in dot: inner dimensions of inputs must match, but got ` +\n          `${t1Inner} and ${t2Inner}.`);\n\n  if ($t1.rank === 1 && $t2.rank === 1) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, []);\n  } else if ($t1.rank === 1 && $t2.rank === 2) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else if ($t1.rank === 2 && $t2.rank === 1) {\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul($t1, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else {\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul($t1, t22D);\n    return t1t2;\n  }\n}\n\nexport const dot = op({dot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential linear element-wise: `x > 0 ? e ^ x - 1 : 0`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 1, -3, 2]);\n *\n * x.elu().print();  // or tf.elu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction elu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'elu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.elu($x);\n    save([y]);\n    return y;\n  };\n\n  const inputs: EluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Elu) as\n      T;\n}\n\nexport const elu = op({elu_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Erf, ErfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes gause error function of the input `tf.Tensor` element-wise:\n * `erf(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.erf().print(); // or tf.erf(x);\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction erf_<T extends Tensor>(x: T|TensorLike): T {\n  let $x = convertToTensor(x, 'x', 'erf');\n  util.assert(\n      $x.dtype === 'int32' || $x.dtype === 'float32',\n      () => 'Input dtype must be `int32` or `float32`.');\n\n  if ($x.dtype === 'int32') {\n    $x = cast($x, 'float32');\n  }\n\n  const inputs: ErfInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.erf($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Erf);\n}\nexport const erf = op({erf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Exp, ExpInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.exp().print();  // or tf.exp(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction exp_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'exp');\n\n  const inputs: ExpInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.exp($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Exp);\n}\nexport const exp = op({exp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension\n * into the tensor's shape.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const axis = 1;\n * x.expandDims(axis).print();\n * ```\n *\n * @param x The input tensor whose dimensions to be expanded.\n * @param axis The dimension index at which to insert shape of `1`. Defaults\n *     to 0 (the first dimension).\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction expandDims_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'expandDims', parseAs);\n\n  util.assert(axis <= $x.rank, () => 'Axis must be <= rank of the tensor');\n  const newShape = $x.shape.slice();\n  if (axis < 0) {\n    // Negative value is counted from the tail of rank.\n    util.assert(\n        -($x.rank + 1) <= axis,\n        () => `Axis must be in the interval [${- ($x.rank + 1)}, ${$x.rank}]`);\n    axis = $x.rank + axis + 1;\n  }\n  newShape.splice(axis, 0, 1);\n  return reshape($x, newShape) as T;\n}\n\nexport const expandDims = op({expandDims_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Expm1, Expm1Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` minus one element-wise.\n * `e ^ x - 1`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.expm1().print();  // or tf.expm1(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction expm1_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'expm1');\n\n  const inputs: Expm1Inputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.expm1($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Expm1);\n}\nexport const expm1 = op({expm1_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Tile, TileAttrs, TileInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Construct a tensor by repeating it the number of times given by reps.\n *\n * This operation creates a new tensor by replicating `input` `reps`\n * times. The output tensor's i'th dimension has `input.shape[i] *\n * reps[i]` elements, and the values of `input` are replicated\n * `reps[i]` times along the i'th dimension. For example, tiling\n * `[a, b, c, d]` by `[2]` produces `[a, b, c, d, a, b, c, d]`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n *\n * a.tile([2]).print();    // or a.tile([2])\n * ```\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.tile([1, 2]).print();  // or a.tile([1, 2])\n * ```\n * @param x The tensor to tile.\n * @param reps Determines the number of replications per dimension.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction tile_<T extends Tensor>(x: T|TensorLike, reps: number[]): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'tile', parseAs);\n  util.assert(\n      $x.rank === reps.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of reps ${reps}.`);\n\n  const forward: ForwardFunc<T> = (backend, save) => {\n    const res = backend.tile($x, reps);\n    save([$x]);\n    return res;\n  };\n\n  const inputsToSave = [$x];\n  const inputs: TileInputs = {x: $x};\n  const attrs: TileAttrs = {reps};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, Tile,\n      attrs as unknown as NamedAttrMap, inputsToSave);\n}\n\nexport const tile = op({tile_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {DataType} from '../types';\n\nimport {buffer} from './buffer';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {tile} from './tile';\n\n/**\n * Create an identity matrix.\n *\n * @param numRows Number of rows.\n * @param numColumns Number of columns. Defaults to `numRows`.\n * @param batchShape If provided, will add the batch shape to the beginning\n *   of the shape of the returned `tf.Tensor` by repeating the identity\n *   matrix.\n * @param dtype Data type.\n * @returns Identity matrix of the specified size and data type, possibly\n *   with batch repetition if `batchShape` is specified.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction eye_(\n    numRows: number, numColumns?: number,\n    batchShape?:\n        [\n          number\n        ]|[number,\n           number]|[number, number, number]|[number, number, number, number],\n    dtype: DataType = 'float32'): Tensor2D {\n  if (numColumns == null) {\n    numColumns = numRows;\n  }\n  const buff = buffer([numRows, numColumns], dtype);\n  const n = numRows <= numColumns ? numRows : numColumns;\n  for (let i = 0; i < n; ++i) {\n    buff.set(1, i, i);\n  }\n  const out: Tensor2D = reshape(buff.toTensor(), [numRows, numColumns]);\n  if (batchShape == null) {\n    return out;\n  } else {\n    if (batchShape.length === 1) {\n      return tile(expandDims(out, 0), [batchShape[0], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 2) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(\n                 expandDims(expandDims(out, 0), 0),\n                 [batchShape[0], batchShape[1], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 3) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(expandDims(expandDims(expandDims(out, 0), 0), 0), [\n               batchShape[0], batchShape[1], batchShape[2], 1, 1\n             ]) as Tensor2D;\n    } else {\n      throw new Error(\n          `eye() currently supports only 1D and 2D ` +\n          // tslint:disable-next-line:no-any\n          `batchShapes, but received ${(batchShape as any).length}D.`);\n    }\n  }\n}\n\nexport const eye = op({eye_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Fill, FillAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\n/**\n * Creates a `tf.Tensor` filled with a scalar value.\n *\n * ```js\n * tf.fill([2, 2], 4).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param value The scalar value to fill the tensor with.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n * 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction fill<R extends Rank>(\n    shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n  const attrs: FillAttrs = {shape, value, dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.fill(shape, value, dtype), {}, null, Fill,\n      attrs as {} as NamedAttrMap);\n}\n\nexport {fill};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Floor, FloorInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes floor of input `tf.Tensor` element-wise: `floor(x)`.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.floor().print();  // or tf.floor(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction floor_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'floor');\n\n  const inputs: FloorInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.floor($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Floor);\n}\nexport const floor = op({floor_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inputs of size above this threshold will be parallelized by calling multiple\n * shader programs.\n */\nimport {nearestDivisor} from '../util';\n\nexport const PARALLELIZE_THRESHOLD = 30;\n\nexport interface ReduceInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  outSize: number;\n}\n\nexport function computeOptimalWindowSize(inSize: number): number {\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    return inSize;\n  }\n  return nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {nearestDivisor} from '../util';\n\nimport {PARALLELIZE_THRESHOLD} from './reduce_util';\n\nexport interface SegOpInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  numSegments: number;\n}\n\nexport function segOpComputeOptimalWindowSize(\n    inSize: number, numSegments: number): number {\n  let done = false;\n  let res;\n\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    res = inSize;\n    done = true;\n  } else {\n    res = nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n  }\n\n  while (!done) {\n    if (res > numSegments || res === inSize) {\n      done = true;\n    } else {\n      res = nearestDivisor(inSize, res + 1);\n    }\n  }\n  return res;\n}\n\nexport function computeOutShape(\n    aShape: number[], axis: number, numSegments: number): number[] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (dim !== axis) {\n      outShape.push(aShape[dim]);\n    } else {\n      outShape.push(numSegments);\n    }\n  }\n  return outShape;\n}\n\nexport interface GatherOpShapeInfo {\n  batchSize: number;\n  sliceSize: number;\n  dimSize: number;\n  outputShape: number[];\n}\nexport function collectGatherOpShapeInfo(\n    x: Tensor, indices: Tensor, axis: number): GatherOpShapeInfo {\n  const dimSize = x.shape[axis];\n\n  const outputShape: number[] = [];\n  let batchSize = 1;\n  let sliceSize = 1;\n  for (let i = 0; i < axis; i++) {\n    outputShape.push(x.shape[i]);\n    batchSize *= x.shape[i];\n  }\n\n  for (let i = 0; i < indices.rank; i++) {\n    outputShape.push(indices.shape[i]);\n  }\n\n  for (let i = axis + 1; i < x.rank; i++) {\n    outputShape.push(x.shape[i]);\n    sliceSize *= x.shape[i];\n  }\n\n  return {batchSize, sliceSize, dimSize, outputShape};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherV2, GatherV2Attrs, GatherV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {collectGatherOpShapeInfo} from './segment_util';\n\n/**\n * Gather slices from tensor `x`'s axis `axis` according to `indices`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const indices = tf.tensor1d([1, 3, 3], 'int32');\n *\n * x.gather(indices).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const indices = tf.tensor1d([1, 1, 0], 'int32');\n *\n * x.gather(indices).print();\n * ```\n * @param x The input tensor whose slices to be gathered.\n * @param indices The indices of the values to extract.\n * @param axis The axis over which to select values. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction gather_<T extends Tensor>(\n    x: T|TensorLike, indices: Tensor|TensorLike, axis = 0): T {\n  const $x = convertToTensor(x, 'x', 'gather');\n  const $indices = convertToTensor(indices, 'indices', 'gather', 'int32');\n\n  const inputs: GatherV2Inputs = {x: $x, indices: $indices};\n  const attrs: GatherV2Attrs = {axis};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const parsedAxis = parseAxisParam(axis, $x.shape)[0];\n    const shapeInfo = collectGatherOpShapeInfo($x, $indices, parsedAxis);\n\n    const res =\n        backend.gather($x, reshape($indices, [$indices.size]), parsedAxis);\n    save([$x, $indices]);\n\n    return reshape(res, shapeInfo.outputShape);\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, GatherV2,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const gather = op({gather_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Greater, GreaterInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a > b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greater(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greater_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greater');\n  let $b = convertToTensor(b, 'b', 'greater');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.greater($a, $b);\n\n  const inputs: GreaterInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             Greater) as T;\n}\n\nexport const greater = op({greater_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GreaterEqual, GreaterEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a >= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greaterEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greaterEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greaterEqual');\n  let $b = convertToTensor(b, 'b', 'greaterEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.greaterEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: GreaterEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             GreaterEqual) as T;\n}\n\nexport const greaterEqual = op({greaterEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Imag, ImagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n/**\n * Returns the imaginary part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the imaginary part of each element in input considered as a complex number.\n * If input is real, a tensor of all zeros is returned.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.imag(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction imag_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'imag');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.imag($input);\n  };\n\n  const inputs: ImagInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Imag) as T;\n}\n\nexport const imag = op({imag_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsFinite, IsFiniteInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are finite.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isFinite().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isFinite_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isFinite');\n\n  const inputs: IsFiniteInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isFinite($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsFinite);\n}\nexport const isFinite = op({isFinite_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsInf, IsInfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are Infinity or -Infinity.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isInf().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isInf_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isInf');\n\n  const inputs: IsInfInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isInf($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsInf);\n}\nexport const isInf = op({isInf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsNan, IsNanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * RReturns which elements of x are NaN.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isNaN().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isNaN_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isNaN');\n  const inputs: IsNanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.isNaN($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsNan);\n}\nexport const isNaN = op({isNaN_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Maximum, MaximumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the max of a and b (`a > b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `tf.maximumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * ```js\n * // Broadcast maximum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction maximum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'maximum');\n  let $b = convertToTensor(b, 'b', 'maximum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.maximum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: MaximumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Maximum) as T;\n}\n\nexport const maximum = op({maximum_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar} from '../tensor';\nimport {DataType} from '../types';\nimport {isTypedArray} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-0 `tf.Tensor` (scalar) with the provided value and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.scalar` as it makes the code more readable.\n *\n * ```js\n * tf.scalar(3.14).print();\n * ```\n *\n * @param value The value of the scalar.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function scalar(\n    value: number|boolean|string|Uint8Array, dtype?: DataType): Scalar {\n  if (((isTypedArray(value) && dtype !== 'string') || Array.isArray(value)) &&\n      dtype !== 'complex64') {\n    throw new Error(\n        'Error creating a new Scalar: value must be a primitive ' +\n        '(number|boolean|string)');\n  }\n  if (dtype === 'string' && isTypedArray(value) &&\n      !(value instanceof Uint8Array)) {\n    throw new Error(\n        'When making a scalar from encoded string, ' +\n        'the value must be `Uint8Array`.');\n  }\n  const shape: number[] = [];\n  const inferredShape: number[] = [];\n  return makeTensor(value, shape, inferredShape, dtype) as Scalar;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {maximum} from './maximum';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {scalar} from './scalar';\n\n/**\n * Computes leaky rectified linear element-wise.\n *\n * See\n * [http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf](\n *     http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf)\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.leakyRelu(0.1).print();  // or tf.leakyRelu(x, 0.1)\n * ```\n * @param x The input tensor.\n * @param alpha The scaling factor for negative values, defaults to 0.2.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction leakyRelu_<T extends Tensor>(x: T|TensorLike, alpha = 0.2): T {\n  const $x = convertToTensor(x, 'x', 'leakyRelu');\n  return maximum(mul(scalar(alpha), $x), $x);\n}\n\nexport const leakyRelu = op({leakyRelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Less, LessInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a < b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.less(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction less_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'less');\n  let $b = convertToTensor(b, 'b', 'less');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.less($a, $b);\n\n  const inputs: LessInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Less) as\n      T;\n}\n\nexport const less = op({less_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LessEqual, LessEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a <= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.lessEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction lessEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'lessEqual');\n  let $b = convertToTensor(b, 'b', 'lessEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.lessEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: LessEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LessEqual) as T;\n}\n\nexport const lessEqual = op({lessEqual_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LinSpace, LinSpaceAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor1D} from '../tensor';\n\n/**\n * Return an evenly spaced sequence of numbers over the given interval.\n *\n * ```js\n * tf.linspace(0, 9, 10).print();\n * ```\n * @param start The start value of the sequence.\n * @param stop The end value of the sequence.\n * @param num The number of values to generate.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function linspace(start: number, stop: number, num: number): Tensor1D {\n  if (num <= 0) {\n    throw new Error('The number of values should be positive.');\n  }\n\n  const attrs: LinSpaceAttrs = {start, stop, num};\n  return ENGINE.runKernelFunc(\n      backend => backend.linspace(start, stop, num), {} /* inputs */,\n      null /* grad */, LinSpace, attrs as {} as NamedAttrMap);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRN, LRNAttrs, LRNInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Normalizes the activation of a local neighborhood across or within\n * channels.\n *\n * @param x The input tensor. The 4-D input tensor is treated as a 3-D array\n *     of 1D vectors (along the last dimension), and each vector is\n *     normalized independently.\n * @param depthRadius The number of adjacent channels in the 1D normalization\n *     window.\n * @param bias A constant bias term for the basis.\n * @param alpha A scale factor, usually positive.\n * @param beta An exponent.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction localResponseNormalization_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const $x = convertToTensor(x, 'x', 'localResponseNormalization');\n  util.assert(\n      $x.rank === 4 || $x.rank === 3,\n      () => `Error in localResponseNormalization: x must be rank 3 or 4 but got\n               rank ${$x.rank}.`);\n  util.assert(\n      util.isInt(depthRadius),\n      () => `Error in localResponseNormalization: depthRadius must be an ` +\n          `integer but got depthRadius ${depthRadius}.`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.localResponseNormalization4D(\n        x4D, depthRadius, bias, alpha, beta);\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: LRNInputs = {x: x4D};\n\n  const attrs: LRNAttrs = {depthRadius, bias, alpha, beta};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, LRN,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  } else {\n    return res as T;\n  }\n}\n\nexport const localResponseNormalization = op({localResponseNormalization_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log, LogInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` element-wise: `ln(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E]);\n *\n * x.log().print();  // or tf.log(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log');\n\n  const inputs: LogInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log);\n}\nexport const log = op({log_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log1p, Log1pInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` plus one\n * element-wise: `ln(1 + x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E - 1]);\n *\n * x.log1p().print();  // or tf.log1p(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log1p_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log1p');\n\n  const inputs: Log1pInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log1p($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log1p);\n}\nexport const log1p = op({log1p_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {CustomGradientFunc, ENGINE} from './engine';\nimport {Scalar, Tensor, Variable} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {convertToTensor, convertToTensorArray} from './tensor_util_env';\nimport {TensorLike} from './types';\nimport * as util from './util';\n\n/**\n * Provided `f(x)`, returns another function `g(x, dy?)`, which gives the\n * gradient of `f(x)` with respect to `x`.\n *\n * If `dy` is provided, the gradient of `f(x).mul(dy).sum()` with respect to\n * `x` is computed instead. `f(x)` must take a single tensor `x` and return a\n * single tensor `y`. If `f()` takes multiple inputs, use `tf.grads` instead.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.grad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * g(x).print();\n * ```\n *\n * ```js\n * // f(x) = x ^ 3\n * const f = x => x.pow(tf.scalar(3, 'int32'));\n * // f'(x) = 3x ^ 2\n * const g = tf.grad(f);\n * // f''(x) = 6x\n * const gg = tf.grad(g);\n *\n * const x = tf.tensor1d([2, 3]);\n * gg(x).print();\n * ```\n *\n * @param f The function f(x), to compute gradient for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grad(f: (x: Tensor) => Tensor): (\n    x: TensorLike|Tensor, dy?: TensorLike|Tensor) => Tensor {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grad(f) must be a function');\n  return (x: TensorLike|Tensor, dy?: TensorLike|Tensor): Tensor => {\n    // x can be of any dtype, thus null as the last argument.\n    const $x = convertToTensor(x, 'x', 'tf.grad', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grad') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f($x), [$x], $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grad(f)(x, dy) must match the shape ' +\n                'returned by f(x)');\n      }\n      checkGrads(grads);\n      return grads[0];\n    });\n  };\n}\n\n/**\n * Provided `f(x1, x2,...)`, returns another function `g([x1, x2,...], dy?)`,\n * which gives an array of gradients of `f()` with respect to each input\n * [`x1`,`x2`,...].\n *\n * If `dy` is passed when calling `g()`, the gradient of\n * `f(x1,...).mul(dy).sum()` with respect to each input is computed instead.\n * The provided `f` must take one or more tensors and return a single tensor\n * `y`. If `f()` takes a single input, we recommend using `tf.grad` instead.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df / da = b, df / db = a\n * const g = tf.grads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const [da, db] = g([a, b]);\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @param f The function `f(x1, x2,...)` to compute gradients for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grads(f: (...args: Tensor[]) => Tensor): (\n    args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike) => Tensor[] {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grads(f) must be a function');\n  return (args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike): Tensor[] => {\n    util.assert(\n        Array.isArray(args),\n        () => 'The args passed in grads(f)(args) must be an array ' +\n            'of `Tensor`s or `TensorLike`s');\n    // args can be of any dtype, thus null as the last argument.\n    const $args = convertToTensorArray(args, 'args', 'tf.grads', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grads') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f(...$args), $args, $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grads(f)([x1,...], dy) must ' +\n                'match the shape returned by f([x1,...])');\n      }\n      checkGrads(grads);\n      return grads;\n    });\n  };\n}\n\n/**\n * Like `tf.grad`, but also returns the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grad: The gradient of `f(x)` w.r.t `x` (result of `tf.grad`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.valueAndGrad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * const {value, grad} = g(x);\n *\n * console.log('value');\n * value.print();\n * console.log('grad');\n * grad.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrad<I extends Tensor, O extends Tensor>(f: (x: I) => O): (\n    x: I, dy?: O) => {\n  value: O;\n  grad: I;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrad(f) must be a function');\n  return (x: I, dy?: O) => {\n    util.assert(\n        x instanceof Tensor,\n        () => 'The x passed in valueAndGrad(f)(x) must be a tensor');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrad(f)(x, dy) must be a tensor');\n    const {grads, value} = ENGINE.gradients(() => f(x), [x], dy);\n    checkGrads(grads);\n    return {grad: grads[0] as I, value};\n  };\n}\n\n/**\n * Like `tf.grads`, but returns also the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grads: The gradients of `f()` w.r.t each input (result of `tf.grads`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df/da = b, df/db = a\n * const g = tf.valueAndGrads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const {value, grads} = g([a, b]);\n *\n * const [da, db] = grads;\n *\n * console.log('value');\n * value.print();\n *\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrads<O extends Tensor>(f: (...args: Tensor[]) => O): (\n    args: Tensor[], dy?: O) => {\n  grads: Tensor[];\n  value: O;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrads(f) must be a function');\n  return (args: Tensor[], dy?: O) => {\n    util.assert(\n        Array.isArray(args) && args.every(arg => arg instanceof Tensor),\n        () => 'The args passed in valueAndGrads(f)(args) must be array of ' +\n            'tensors');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrads(f)(args, dy) must be a tensor');\n    const res = ENGINE.gradients(() => f(...args), args, dy);\n    if (dy != null) {\n      util.assertShapesMatch(\n          res.value.shape, dy.shape,\n          'The shape of dy passed in valueAndGrads(f)([x1,...], dy) must ' +\n              'match the shape returned by f([x1,...])');\n    }\n    checkGrads(res.grads);\n    return res;\n  };\n}\n\n/**\n * Computes and returns the gradient of f(x) with respect to the list of\n * trainable variables provided by `varList`. If no list is provided, it\n * defaults to all trainable variables.\n *\n * ```js\n * const a = tf.variable(tf.tensor1d([3, 4]));\n * const b = tf.variable(tf.tensor1d([5, 6]));\n * const x = tf.tensor1d([1, 2]);\n *\n * // f(a, b) = a * x ^ 2 + b * x\n * const f = () => a.mul(x.square()).add(b.mul(x)).sum();\n * // df/da = x ^ 2, df/db = x\n * const {value, grads} = tf.variableGrads(f);\n *\n * Object.keys(grads).forEach(varName => grads[varName].print());\n * ```\n *\n * @param f The function to execute. f() should return a scalar.\n * @param varList The list of variables to compute the gradients with respect\n *     to. Defaults to all trainable variables.\n * @returns An object with the following keys and values:\n *   - `value`: The value of the function `f`.\n *   - `grads`: A map from the names of the variables to the gradients.\n *     If the `varList` argument is provided explicitly and contains a subset of\n *     non-trainable variables, this map in the return value will contain keys\n *     that map the names of the non-trainable variables to `null`.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction variableGrads(f: () => Scalar, varList?: Variable[]):\n    {value: Scalar, grads: NamedTensorMap} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in variableGrads(f) must be a function');\n  util.assert(\n      varList == null ||\n          Array.isArray(varList) && varList.every(v => v instanceof Variable),\n      () =>\n          'The varList passed in variableGrads(f, varList) must be an array ' +\n          'of variables');\n\n  const specifiedVarList = varList != null;\n  if (!specifiedVarList) {\n    // Get all of the trainable variables.\n    varList = [];\n    for (const varName in ENGINE.registeredVariables) {\n      varList.push(ENGINE.registeredVariables[varName]);\n    }\n  }\n\n  const specifiedNonTrainable: Variable[] =\n      specifiedVarList ? varList.filter(variable => !variable.trainable) : null;\n\n  // Prune non-trainable variables.\n  const originalVarCount = varList.length;\n  varList = varList.filter(variable => variable.trainable);\n  util.assert(\n      varList.length > 0,\n      () => `variableGrads() expects at least one of the input variables to ` +\n          `be trainable, but none of the ${originalVarCount} variables is ` +\n          `trainable.`);\n\n  const allowNoGradients = true;\n  const {value, grads} = ENGINE.gradients(f, varList, null, allowNoGradients);\n\n  util.assert(\n      grads.some(g => g != null),\n      () => 'Cannot find a connection between any variable and the result of ' +\n          'the loss function y=f(x). Please make sure the operations that ' +\n          'use variables are inside the function f passed to minimize().');\n  util.assert(\n      value.rank === 0,\n      () => `The f passed in variableGrads(f) must return a scalar, but it ` +\n          `returned a rank-${value.rank} tensor`);\n\n  const namedGrads: NamedTensorMap = {};\n  varList.forEach((v, i) => {\n    if (grads[i] != null) {\n      namedGrads[v.name] = grads[i];\n    }\n  });\n  if (specifiedNonTrainable != null) {\n    // If varList is explicitly provided and contains non-trainable values,\n    // add them to the returned gradients with `null` values.\n    specifiedNonTrainable.forEach(v => namedGrads[v.name] = null);\n  }\n  return {value, grads: namedGrads};\n}\n\n/**\n * Overrides the gradient computation of a function `f`.\n *\n * Takes a function\n * `f(...inputs, save) => {value: Tensor, gradFunc: (dy, saved) => Tensor[]}`\n * and returns another function `g(...inputs)` which takes the same inputs as\n * `f`. When called, `g` returns `f().value`. In backward mode, custom gradients\n * with respect to each input of `f` are computed using `f().gradFunc`.\n *\n * The `save` function passsed to `f` should be used for saving tensors needed\n * in the gradient. And the `saved` passed to the `gradFunc` is a\n * `NamedTensorMap`, which contains those saved tensor.\n *\n * ```js\n * const customOp = tf.customGrad((x, save) => {\n *   // Save x to make sure it's available later for the gradient.\n *   save([x]);\n *   // Override gradient of our custom x ^ 2 op to be dy * abs(x);\n *   return {\n *     value: x.square(),\n *     // Note `saved.x` which points to the `x` we saved earlier.\n *     gradFunc: (dy, saved) => [dy.mul(saved[0].abs())]\n *   };\n * });\n *\n * const x = tf.tensor1d([-1, -2, 3]);\n * const dx = tf.grad(x => customOp(x));\n *\n * console.log(`f(x):`);\n * customOp(x).print();\n * console.log(`f'(x):`);\n * dx(x).print();\n * ```\n *\n * @param f The function to evaluate in forward mode, which should return\n *     `{value: Tensor, gradFunc: (dy, saved) => Tensor[]}`, where `gradFunc`\n *     returns the custom gradients of `f` with respect to its inputs.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n    (...args: Tensor[]) => T {\n  return ENGINE.customGrad(f);\n}\n\nfunction checkGrads(grads: Tensor[]) {\n  const numNullGradients = grads.filter(g => g == null).length;\n  if (numNullGradients > 0) {\n    throw new Error(\n        `Cannot compute gradient of y=f(x) with respect to x. Make sure that\n    the f you passed encloses all operations that lead from x to y.`);\n  }\n}\n\nexport {\n  customGrad,\n  variableGrads,\n  valueAndGrad,\n  valueAndGrads,\n  grad,\n  grads,\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Negate, NegateInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes `-1 * x` element-wise.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, -2, 0], [2, 2]);\n *\n * x.neg().print();  // or tf.neg(x)\n * ```\n *\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction neg_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'neg');\n\n  const inputs: NegateInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.neg($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Negate);\n}\nexport const neg = op({neg_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softplus, SoftplusInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes softplus of the input `tf.Tensor` element-wise: `log(exp(x) + 1)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.softplus().print();  // or tf.softplus(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction softplus_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'softplus');\n\n  const inputs: SoftplusInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.softplus($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Softplus);\n}\nexport const softplus = op({softplus_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {mul} from './mul';\nimport {neg} from './neg';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {softplus} from './softplus';\n\n/**\n * Computes log sigmoid of the input `tf.Tensor` element-wise:\n * `logSigmoid(x)`. For numerical stability, we use `-tf.softplus(-x)`.\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.logSigmoid().print();  // or tf.logSigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction logSigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logSigmoid');\n\n  // Use a custom gradient to maintain previous implementation.\n  // There is no LogSigmoid kernel in TF so we can't use engine.runKernel\n  // directly\n  const customOp = customGrad((x: Tensor) => {\n    // TODO(yassogba) we can remove the chained softplus call here only\n    // after backends have modualrized softplus at which point we can call\n    // engine runKernel(..., Sotfplus, ...) directly.\n    const value = neg(softplus(neg(x)));\n\n    const gradFunc = (dy: T) => {\n      const derX = mul(dy, sigmoid(neg(x)));\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\nexport const logSigmoid = op({logSigmoid_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Max, MaxAttrs, MaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the maximum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.max().print();  // or tf.max(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.max(axis).print();  // or tf.max(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction max_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'max');\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = util.parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let maxInput = $x;\n        if (permutedAxes != null) {\n          maxInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, maxInput.rank);\n        }\n\n        const y = backend.max(maxInput, axes);\n        if (permutedAxes != null) {\n          maxInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape = axis_util.expandShapeToKeepDim(\n              res.shape, util.parseAxisParam(axis, $x.shape));\n          res = reshape(res, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MaxInputs = {x: $x};\n  const attrs: MaxAttrs = {reductionIndices: axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Max,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const max = op({max_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sub, SubInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Subtracts two `tf.Tensor`s element-wise, A - B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n *\n * ```js\n * // Broadcast subtract a with b.\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.scalar(5);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n * @param a The first `tf.Tensor` to subtract from.\n * @param b The second `tf.Tensor` to be subtracted. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction sub_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'sub');\n  let $b = convertToTensor(b, 'b', 'sub');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.subtract($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SubInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sub) as\n      T;\n}\n\nexport const sub = op({sub_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sum, SumAttrs, SumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the sum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If axes has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.sum().print();  // or tf.sum(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.sum(axis).print();  // or tf.sum(x, axis)\n * ```\n *\n * @param x The input tensor to compute the sum over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction sum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'sum');\n  if ($x.dtype === 'bool') {\n    $x = cast($x, 'int32');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.sum(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n    return value;\n  };\n\n  const inputs: SumInputs = {x: $x};\n  const attrs: SumAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const sum = op({sum_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LogSoftmax, LogSoftmaxAttrs, LogSoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log softmax.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param axis The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction logSoftmax_<T extends Tensor>(logits: T|TensorLike, axis = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'logSoftmax');\n\n  if (axis === -1) {\n    axis = $logits.rank - 1;\n  }\n  if (axis !== $logits.rank - 1) {\n    throw Error(\n        'Log Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and axis was ${axis}`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const keepDims = true;\n    const xMax = max(logits, axis, true);\n    const shifted = sub(logits, xMax);\n    const value =\n        sub(cast(shifted, 'float32'), log(sum(exp(shifted), axis, keepDims)));\n    save([value]);\n    return value;\n  };\n\n  const inputs: LogSoftmaxInputs = {logits: $logits};\n  const attrs: LogSoftmaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LogSoftmax, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const logSoftmax = op({logSoftmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {add} from './add';\nimport {expandShapeToKeepDim} from './axis_util';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log(sum(exp(elements across the reduction dimensions)).\n *\n * Reduces the input along the dimensions given in `axis`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.logSumExp().print();  // or tf.logSumExp(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.logSumExp(axis).print();  // or tf.logSumExp(a, axis)\n * ```\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. If null (the default),\n *     reduces all dimensions.\n * @param keepDims If true, retains reduced dimensions with length\n *     of 1. Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction logSumExp_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'logSumExp');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const xMax = max($x, axes, true /* keepDims */);\n  const a = sub($x, xMax);\n  const b = exp(a);\n  const c = sum(b, axes);\n  const d = log(c);\n  const res = add(reshape(xMax, d.shape), d);\n\n  if (keepDims) {\n    const newShape = expandShapeToKeepDim(res.shape, axes);\n    return reshape(res, newShape) as T;\n  }\n  return res as T;\n}\n\nexport const logSumExp = op({logSumExp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {LogicalAnd, LogicalAndInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a AND b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalAnd(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalAnd_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalAnd', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalAnd', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalAndInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalAnd($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalAnd) as T;\n}\n\nexport const logicalAnd = op({logicalAnd_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalNot, LogicalNotInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `NOT x` element-wise.\n *\n * ```js\n * const a = tf.tensor1d([false, true], 'bool');\n *\n * a.logicalNot().print();\n * ```\n *\n * @param x The input tensor. Must be of dtype 'bool'.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalNot_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logicalNot', 'bool');\n  const inputs: LogicalNotInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.logicalNot($x), inputs as {} as NamedTensorMap,\n      null /* grad */, LogicalNot);\n}\n\nexport const logicalNot = op({logicalNot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalOr, LogicalOrInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a OR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalOr(b).print();\n * ```\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalOr_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalOr', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalOr', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalOrInputs = {a: $a, b: $b};\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalOr($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalOr) as T;\n}\nexport const logicalOr = op({logicalOr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {logicalAnd} from './logical_and';\nimport {logicalNot} from './logical_not';\nimport {logicalOr} from './logical_or';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a XOR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalXor(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalXor_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalXor', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalXor', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  // x ^ y = (x | y) & ~(x & y)\n  return logicalAnd(logicalOr(a, b), logicalNot(logicalAnd(a, b)));\n}\n\nexport const logicalXor = op({logicalXor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool, MaxPoolAttrs, MaxPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D max pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction maxPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'maxPool');\n  const dilations = 1;\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in maxPool: input must be rank 4 but got rank ${x4D.rank}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    let y;\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      y = x4D.clone();\n    } else {\n      y = backend.maxPool(x4D, convInfo);\n    }\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: MaxPoolInputs = {x: x4D};\n\n  const attrs: MaxPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const maxPool = op({maxPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {MaxPool3D, MaxPool3DAttrs, MaxPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D max pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.maxPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number = [1, 1, 1],\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'maxPool3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in maxPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in maxPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n    const y = backend.maxPool3d(x5D, convInfo);\n    save([x5D, y]);\n    return y;\n  };\n\n  const inputs: MaxPool3DInputs = {x: x5D};\n\n  const attrs: MaxPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3d = op({maxPool3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the 2D max pooling of an image with Argmax index.\n * The indices in argmax are flattened, so that a maximum value at position `[b,\n * y, x, c]` becomes flattened index: `(y * width + x) * channels + c` if\n * include_batch_in_index is False; `((b * height + y) * width + x) * channels\n * +c` if include_batch_in_index is True.\n *\n * The indices returned are always in `[0, height) x [0, width)` before\n * flattening.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param includeBatchIndex Defaults to False. Whether to include batch\n *    dimension in flattened index of argmax.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPoolWithArgmax_<T extends Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    includeBatchInIndex = false): NamedTensorMap {\n  const $x = convertToTensor(x, 'x', 'maxPoolWithArgmax');\n\n  const inputs: MaxPoolWithArgmaxInputs = {x: $x};\n  const attrs:\n      MaxPoolWithArgmaxAttrs = {filterSize, strides, pad, includeBatchInIndex};\n\n  const result = ENGINE.runKernel(\n                     MaxPoolWithArgmax, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {result: result[0], indexes: result[1]};\n}\n\nexport const maxPoolWithArgmax = op({maxPoolWithArgmax_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeZerosTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0.\n *\n * ```js\n * tf.zeros([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Can\n *     be 'float32', 'int32' or 'bool'. Defaults to 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function zeros<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = zeros(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeZerosTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeOnesTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\nimport {zeros} from './zeros';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1.\n *\n * ```js\n * tf.ones([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n *     'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function ones<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = ones(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeOnesTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam, sizeFromShape} from '../util';\n\nimport {computeOutAndReduceShapes} from './axis_util';\nimport {cast} from './cast';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {ones} from './ones';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sum} from './sum';\n\n/**\n * Computes the mean of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces `x` along the dimensions given in `axis`. Unless `keepDims` is\n * true, the rank of the `tf.Tensor` is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and a `tf.Tensor` with\n * a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.mean().print();  // or tf.mean(a)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.mean(axis).print();  // or tf.mean(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction mean_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'mean');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const shapes = computeOutAndReduceShapes($x.shape, axes);\n  const reduceShape = shapes[1];\n  const reduceSize = sizeFromShape(reduceShape);\n\n  // Use a custom gradient to bypass 2 gradient backprops since mean is used\n  // extremely often.\n  const customOp = customGrad((x: Tensor) => {\n    const reduceSizeScalar = scalar(reduceSize);\n    // Cast if needed.\n    const xReduce = reduceSizeScalar.dtype === x.dtype ?\n        x :\n        cast(x, reduceSizeScalar.dtype);\n    const res = div(xReduce, reduceSizeScalar);\n    const value = sum(res, axis, keepDims);\n\n    const gradFunc = (dy: Tensor) => {\n      const expandedDyShape = x.shape.slice();\n      axes.forEach(axis => {\n        expandedDyShape[axis] = 1;\n      });\n      const expandedDy = reshape(dy, expandedDyShape);\n      const derX = div(mul(expandedDy, ones(x.shape, 'float32')), reduceSize);\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\n\nexport const mean = op({mean_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Min, MinAttrs, MinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the minimum value from the input.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axes`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axes` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.min().print();  // or tf.min(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.min(axis).print();  // or tf.min(x, axis)\n * ```\n *\n * @param x The input Tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction min_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'min');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let minInput = $x;\n        if (permutedAxes != null) {\n          minInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n        }\n\n        const y = backend.min(minInput, axes);\n        if (permutedAxes != null) {\n          minInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape =\n              axis_util.expandShapeToKeepDim(res.shape, origAxes);\n          res = reshape(y, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MinInputs = {x: $x};\n  const attrs: MinAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Min,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const min = op({min_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Minimum, MinimumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the min of a and b (`a < b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `minimumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * ```js\n * // Broadcast minimum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction minimum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'minimum');\n  let $b = convertToTensor(b, 'b', 'minimum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.minimum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MinimumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Minimum) as T;\n}\n\nexport const minimum = op({minimum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Mod, ModInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns the mod of a and b element-wise.\n * `floor(x / y) * y + mod(x, y) = x`\n * Supports broadcasting.\n *\n * We also expose `tf.modStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * ```js\n * // Broadcast a mod b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mod_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mod');\n  let $b = convertToTensor(b, 'b', 'mod');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.mod($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: ModInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Mod) as T;\n}\n\nexport const mod = op({mod_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Computes square of `x` element-wise: `x ^ 2`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.sqrt(2), -1]);\n *\n * x.square().print();  // or tf.square(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction square_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'square');\n  const attrs = {};\n  const inputsToSave = [$x];\n  const outputsToSave: boolean[] = [];\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    return backend.square($x);\n  }, {x: $x}, null /* grad */, 'Square', attrs, inputsToSave, outputsToSave);\n}\n\nexport const square = op({square_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim} from './axis_util';\nimport {cast} from './cast';\nimport {mean} from './mean';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {square} from './square';\nimport {sub} from './sub';\n\n/**\n * Calculates the mean and variance of `x`. The mean and variance are\n * calculated by aggregating the contents of `x` across `axes`. If `x` is\n * 1-D and `axes = [0]` this is just the mean and variance of a vector.\n *\n * @param x The input tensor.\n * @param axis The dimension(s) along with to compute mean and\n *     variance. By default it reduces all dimensions.\n * @param keepDims If true, the moments have the same dimensionality as the\n *     input.\n * @return An object with two keys: `mean` and `variance`.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction moments_(\n    x: Tensor|TensorLike, axis: number|number[] = null,\n    keepDims = false): {mean: Tensor, variance: Tensor} {\n  x = convertToTensor(x, 'x', 'moments');\n  const axes = parseAxisParam(axis, x.shape);\n  const xMean = mean(x, axes, keepDims);\n  let keepDimsShape = xMean.shape;\n  if (!keepDims) {\n    keepDimsShape = expandShapeToKeepDim(xMean.shape, axes);\n  }\n  const devSquared =\n      square(sub(cast(x, 'float32'), reshape(xMean, keepDimsShape)));\n  const variance = mean(devSquared, axes, keepDims);\n  return {mean: xMean, variance};\n}\n\nexport const moments = op({moments_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor, convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * @docalias (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D]\n */\nexport type LSTMCellFunc = {\n  (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D];\n};\n\n/**\n * Computes the next states and outputs of a stack of LSTMCells.\n *\n * Each cell output is used as input to the next cell.\n *\n * Returns `[cellState, cellOutput]`.\n *\n * Derived from tf.contrib.rn.MultiRNNCell.\n *\n * @param lstmCells Array of LSTMCell functions.\n * @param data The input to the cell.\n * @param c Array of previous cell states.\n * @param h Array of previous cell outputs.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction multiRNNCell_(\n    lstmCells: LSTMCellFunc[], data: Tensor2D|TensorLike,\n    c: Array<Tensor2D|TensorLike>,\n    h: Array<Tensor2D|TensorLike>): [Tensor2D[], Tensor2D[]] {\n  const $data = convertToTensor(data, 'data', 'multiRNNCell');\n  const $c = convertToTensorArray(c, 'c', 'multiRNNCell');\n  const $h = convertToTensorArray(h, 'h', 'multiRNNCell');\n\n  let input = $data;\n  const newStates = [];\n  for (let i = 0; i < lstmCells.length; i++) {\n    const output = lstmCells[i](input, $c[i], $h[i]);\n    newStates.push(output[0]);\n    newStates.push(output[1]);\n    input = output[1];\n  }\n  const newC: Tensor2D[] = [];\n  const newH: Tensor2D[] = [];\n  for (let i = 0; i < newStates.length; i += 2) {\n    newC.push(newStates[i]);\n    newH.push(newStates[i + 1]);\n  }\n  return [newC, newH];\n}\nexport const multiRNNCell = op({multiRNNCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a `tf.Tensor` with values drawn from a multinomial distribution.\n *\n * ```js\n * const probs = tf.tensor([.75, .25]);\n * tf.multinomial(probs, 3).print();\n * ```\n *\n * @param logits 1D array with unnormalized log-probabilities, or\n *     2D array of shape `[batchSize, numOutcomes]`. See the `normalized`\n *     parameter.\n * @param numSamples Number of samples to draw for each row slice.\n * @param seed The seed number.\n * @param normalized Whether the provided `logits` are normalized true\n *     probabilities (sum to 1). Defaults to false.\n * @return 1D array of shape `[numSamples]`, or 2D array of shape\n *     `[batchSize, numSamples]`, depending on the rank of the input.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction multinomial_(\n    logits: Tensor1D|Tensor2D|TensorLike, numSamples: number, seed?: number,\n    normalized = false): Tensor1D|Tensor2D {\n  const $logits = convertToTensor(logits, 'logits', 'multinomial');\n  const numOutcomes = $logits.size;\n  const origRank = $logits.rank;\n  if (numOutcomes < 2) {\n    throw new Error(\n        `Error in multinomial: you need at least 2 outcomes, but got ` +\n        `${numOutcomes}.`);\n  }\n  if (origRank > 2) {\n    throw new Error(`Rank of probabilities must be 1 or 2, but is ${origRank}`);\n  }\n  seed = seed || Math.random();\n  const logits2D: Tensor2D =\n      origRank === 1 ? reshape($logits, [1, -1]) : $logits as Tensor2D;\n  const res = ENGINE.runKernelFunc(\n      backend => backend.multinomial(logits2D, normalized, numSamples, seed),\n      {logits2D});\n  // tslint:disable-next-line:no-unnecessary-type-assertion\n  return origRank === 1 ? reshape(res, [res.size]) as Tensor1D : res;\n}\n\nexport const multinomial = op({multinomial_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {NotEqual, NotEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a != b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([0, 2, 3]);\n *\n * a.notEqual(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction notEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'notEqual');\n  let $b = convertToTensor(b, 'b', 'notEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => backend.notEqual($a, $b);\n\n  const inputs: NotEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             NotEqual) as T;\n}\n\nexport const notEqual = op({notEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Real, RealInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the real part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the real part of each element in input considered as a complex number.\n *\n * If the input is real, it simply makes a clone.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.real(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction real_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'real');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.real($input);\n  };\n\n  const inputs: RealInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Real) as T;\n}\n\nexport const real = op({real_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OnesLike, OnesLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {complex} from './complex';\nimport {imag} from './imag';\nimport {op} from './operation';\nimport {real} from './real';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.onesLike(x).print();\n * ```\n * @param x A tensor.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction onesLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'onesLike');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if ($x.dtype === 'complex64') {\n      const r = onesLike(real($x));\n      const i = zerosLike(imag($x));\n      return complex(r, i);\n    }\n\n    return backend.onesLike($x);\n  };\n\n  const inputs: OnesLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             OnesLike) as T;\n}\n\nexport const onesLike = op({onesLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the outer product of two vectors, `v1` and `v2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([3, 4, 5]);\n *\n * tf.outerProduct(a, b).print();\n * ```\n * @param v1 The first vector in the outer product operation.\n * @param v2 The second vector in the outer product operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction outerProduct_(\n    v1: Tensor1D|TensorLike, v2: Tensor1D|TensorLike): Tensor2D {\n  const $v1 = convertToTensor(v1, 'v1', 'outerProduct');\n  const $v2 = convertToTensor(v2, 'v2', 'outerProduct');\n\n  util.assert(\n      $v1.rank === 1 && $v2.rank === 1,\n      () => `Error in outerProduct: inputs must be rank 1, but got ranks ` +\n          `${$v1.rank} and ${$v2.rank}.`);\n\n  const v12D = reshape($v1, [-1, 1]);\n  const v22D = reshape($v2, [1, -1]);\n  return matMul(v12D, v22D) as Tensor2D;\n}\n\nexport const outerProduct = op({outerProduct_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {PadV2, PadV2Attrs, PadV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Pads a `tf.Tensor` with a given value and paddings.\n *\n * This operation currently only implements the `CONSTANT` mode.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `paddings` is of given length.\n *   - `tf.pad1d`\n *   - `tf.pad2d`\n *   - `tf.pad3d`\n *   - `tf.pad4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.pad([[1, 2]]).print();\n * ```\n * @param x The tensor to pad.\n * @param paddings An array of length `R` (the rank of the tensor), where\n * each element is a length-2 tuple of ints `[padBefore, padAfter]`,\n * specifying how much to pad along each dimension of the tensor.\n * @param constantValue The pad value to use. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction pad_<T extends Tensor>(\n    x: T|TensorLike, paddings: Array<[number, number]>, constantValue = 0): T {\n  const $x = convertToTensor(x, 'x', 'pad');\n  if ($x.rank === 0) {\n    throw new Error('pad(scalar) is not defined. Pass non-scalar to pad');\n  }\n  const forward: ForwardFunc<T> = (backend, save) => {\n    save([$x]);\n    return backend.pad($x, paddings, constantValue);\n  };\n\n  const attrs: PadV2Attrs = {paddings, constantValue};\n  const inputs: PadV2Inputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, PadV2,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const pad = op({pad_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor1D` with a given value and paddings. See `pad` for details.\n */\nfunction pad1d_(\n    x: Tensor1D|TensorLike, paddings: [number, number],\n    constantValue = 0): Tensor1D {\n  assert(\n      paddings.length === 2,\n      () => 'Invalid number of paddings. Must be length of 2.');\n  return pad(x, [paddings], constantValue);\n}\n\nexport const pad1d = op({pad1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor2D` with a given value and paddings. See `pad` for details.\n */\nfunction pad2d_(\n    x: Tensor2D|TensorLike, paddings: [[number, number], [number, number]],\n    constantValue = 0): Tensor2D {\n  assert(\n      paddings.length === 2 && paddings[0].length === 2 &&\n          paddings[1].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad2d = op({pad2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor3D` with a given value and paddings. See `pad` for details.\n */\nfunction pad3d_(\n    x: Tensor3D|TensorLike,\n    paddings: [[number, number], [number, number], [number, number]],\n    constantValue = 0): Tensor3D {\n  assert(\n      paddings.length === 3 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad3d = op({pad3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor4D` with a given value and paddings. See `pad` for details.\n */\nfunction pad4d_(\n    x: Tensor4D|TensorLike,\n    paddings:\n        [\n          [number, number], [number, number], [number, number], [number, number]\n        ],\n    constantValue = 0): Tensor4D {\n  assert(\n      paddings.length === 4 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2 &&\n          paddings[3].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad4d = op({pad4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation divides \"spatial\" dimensions `[1, ..., M]` of the input into\n * a grid of blocks of shape `blockShape`, and interleaves these blocks with\n * the \"batch\" dimension (0) such that in the output, the spatial\n * dimensions `[1, ..., M]` correspond to the position within the grid,\n * and the batch dimension combines both the position within a spatial block\n * and the original batch position. Prior to division into blocks,\n * the spatial dimensions of the input are optionally zero padded\n * according to `paddings`. See below for a precise description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);\n * const blockShape = [2, 2];\n * const paddings = [[0, 0], [0, 0]];\n *\n * x.spaceToBatchND(blockShape, paddings).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param paddings A 2-D array. Must have shape `[M, 2]`, all values must be >=\n *     0. `paddings[i] = [padStart, padEnd]` specifies the amount to zero-pad\n * from input dimension `i + 1`, which corresponds to spatial dimension `i`. It\n * is required that\n * `(inputShape[i + 1] + padStart + padEnd) % blockShape[i] === 0`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Zero-pad the start and end of dimensions `[1, ..., M]` of the input\n * according to `paddings` to produce `padded` of shape paddedShape.\n *\n * 2. Reshape `padded` to `reshapedPadded` of shape:\n * `[batch] + [paddedShape[1] / blockShape[0], blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1], blockShape[M-1]] + remainingShape`\n *\n * 3. Permute dimensions of `reshapedPadded` to produce `permutedReshapedPadded`\n * of shape: `blockShape + [batch] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * 4. Reshape `permutedReshapedPadded` to flatten `blockShape` into the\n * batch dimension, producing an output tensor of shape:\n * `[batch * prod(blockShape)] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction spaceToBatchND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], paddings: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'spaceToBatchND');\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank ${$x.rank} should be > than [blockShape] ${\n          blockShape.length}`);\n\n  util.assert(\n      paddings.length === blockShape.length,\n      () => `paddings.shape[0] ${\n          paddings.length} must be equal to [blockShape] ${blockShape.length}`);\n\n  util.assert(\n      $x.shape.reduce(\n          (a, b, i) => {\n            if (i > 0 && i <= blockShape.length) {\n              return a &&\n                  ((b + paddings[i - 1][0] + paddings[i - 1][1]) %\n                       blockShape[i - 1] ===\n                   0);\n            }\n            return a;\n          },\n          true),\n      () => `input spatial dimensions ${$x.shape.slice(1)} with paddings ${\n          paddings.toString()} must be divisible by blockShapes ${\n          blockShape.toString()}`);\n\n  const forward: ForwardFunc<T> = backend =>\n      backend.spaceToBatchND($x, blockShape, paddings);\n\n  const inputs: SpaceToBatchNDInputs = {x: $x};\n  const attrs: SpaceToBatchNDAttrs = {blockShape, paddings};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      SpaceToBatchND, attrs as {} as NamedAttrMap);\n}\n\nexport const spaceToBatchND = op({spaceToBatchND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {avgPool} from './avg_pool';\nimport {batchToSpaceND} from './batch_to_space_nd';\nimport * as conv_util from './conv_util';\nimport {maxPool} from './max_pool';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {spaceToBatchND} from './space_to_batch_nd';\n\n/**\n * Performs an N-D pooling operation\n *\n * @param input The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param windowShape The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param poolingType The type of pooling, either 'max' or 'avg'.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilationRate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction pool_<T extends Tensor3D|Tensor4D>(\n    input: T|TensorLike, windowShape: [number, number]|number,\n    poolingType: 'avg'|'max', pad: 'valid'|'same'|number,\n    dilations?: [number, number]|number, strides?: [number, number]|number) {\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  if (strides == null) {\n    strides = 1;\n  }\n  if (pad === 0) {\n    pad = 'valid';\n  }\n\n  const $x = convertToTensor(input, 'x', 'maxPool');\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in pool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = conv_util.computePool2DInfo(\n      x4D.shape, windowShape, strides, dilations, pad);\n  const dilation: [number, number] =\n      [convInfo.dilationHeight, convInfo.dilationWidth];\n\n  // The following implementation does batchToSpace(pool(spaceToBatch(x)))\n  // whenever dilation > 1 since the TF kernels do not support dilation > 1.\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L1037\n\n  let basePadding: number[][];\n  if (pad === 'same') {\n    basePadding = withSpaceToBatchBasePaddings(\n        [convInfo.filterHeight, convInfo.filterWidth], dilation);\n  } else {\n    basePadding = [[0, 0], [0, 0]];\n  }\n\n  const isDilationOne = dilation[0] === 1 && dilation[1] === 1;\n  const [adjustedPadding, adjustedCrops] = requiredSpaceToBatchPaddings(\n      [convInfo.inHeight, convInfo.inWidth], dilation, basePadding);\n  const convertedPad = isDilationOne ? pad : 'valid';\n  const convertedX =\n      isDilationOne ? x4D : spaceToBatchND(x4D, dilation, adjustedPadding);\n\n  const forwardOp = poolingType === 'avg' ?\n      () => avgPool(convertedX, windowShape, strides, convertedPad) :\n      () => maxPool(convertedX, windowShape, strides, convertedPad);\n  const y = forwardOp();\n\n  const res = isDilationOne ? y : batchToSpaceND(y, dilation, adjustedCrops);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\n// Helper function to compute crops and paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/array_ops.py#L2184\nfunction requiredSpaceToBatchPaddings(\n    inputShape: [number, number], blockShape: [number, number],\n    basePadding: number[][]) {\n  const padStart = basePadding.map(b => b[0]);\n  const origPadEnd = basePadding.map(b => b[1]);\n  const fullInputShape = inputShape.concat(padStart, origPadEnd);\n  const padEndExtra = blockShape.map((b, i) => (b - fullInputShape[i] % b) % b);\n  const padEnd = origPadEnd.map((s, i) => s + padEndExtra[i]);\n  const paddings = blockShape.map((_, i) => [padStart[i], padEnd[i]]);\n  const crops = blockShape.map((_, i) => [0, padEndExtra[i]]);\n  return [paddings, crops];\n}\n\n// Helper function to compute base paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L524\nfunction withSpaceToBatchBasePaddings(\n    filterShape: [number, number], dilation: [number, number]) {\n  // Spatial dimensions of the filters and the upsampled filters in which we\n  // introduce (rate - 1) zeros between consecutive filter values.\n  const dilatedFilterShape = filterShape.map((s, i) => {\n    return s + (s - 1) * (dilation[i] - 1);\n  });\n  const padExtraShape = dilatedFilterShape.map(s => s - 1);\n\n  // When padding is odd, we pad more at end, following the same\n  // convention as conv2d.\n  const padExtraStart = padExtraShape.map(s => Math.floor(s / 2));\n  const padExtraEnd = padExtraShape.map((s, i) => s - padExtraStart[i]);\n  return padExtraShape.map((_, i) => {\n    return [padExtraStart[i], padExtraEnd[i]];\n  });\n}\n\nexport const pool = op({pool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Pow, PowInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the power of one `tf.Tensor` to another. Supports broadcasting.\n *\n * Given a `tf.Tensor` x and a `tf.Tensor` y, this operation computes x^y for\n * corresponding elements in x and y. The result's dtype will be the upcasted\n * type of the `base` and `exp` dtypes.\n *\n * ```js\n * const a = tf.tensor([[2, 3], [4, 5]])\n * const b = tf.tensor([[1, 2], [3, 0]]).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n *\n * ```js\n * const a = tf.tensor([[1, 2], [3, 4]])\n * const b = tf.tensor(2).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n * We also expose `powStrict` which has the same signature as this op and\n * asserts that `base` and `exp` are the same shape (does not broadcast).\n *\n * @param base The base `tf.Tensor` to pow element-wise.\n * @param exp The exponent `tf.Tensor` to pow element-wise.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction pow_<T extends Tensor>(\n    base: Tensor|TensorLike, exp: Tensor|TensorLike): T {\n  let $base = convertToTensor(base, 'base', 'pow');\n  let $exp = convertToTensor(exp, 'exp', 'pow');\n  [$base, $exp] = makeTypesMatch($base, $exp);\n\n  const inputs: PowInputs = {a: $base, b: $exp};\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.pow($base, $exp);\n    save([$base, $exp, y]);\n    return y;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Pow) as T;\n}\n\nexport const pow = op({pow_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prelu, PreluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes leaky rectified linear element-wise with parametric alphas.\n *\n * `x < 0 ? alpha * x : f(x) = x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n * const alpha = tf.scalar(0.1);\n *\n * x.prelu(alpha).print();  // or tf.prelu(x, alpha)\n * ```\n * @param x The input tensor.\n * @param alpha Scaling factor for negative values.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction prelu_<T extends Tensor>(x: T|TensorLike, alpha: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'prelu');\n  const $alpha = convertToTensor(alpha, 'alpha', 'prelu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.prelu($x, $alpha);\n    save([$x, $alpha]);\n    return res;\n  };\n\n  const inputs: PreluInputs = {x: $x, alpha: $alpha};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prelu) as\n      T;\n}\n\nexport const prelu = op({prelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prod, ProdAttrs, ProdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the product of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.prod().print();  // or tf.prod(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.prod(axis).print();  // or tf.prod(x, axis)\n * ```\n *\n * @param x The input tensor to compute the product over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction prod_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'prod');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if ($x.dtype === 'bool') {\n      $x = cast($x, 'int32');\n    }\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.prod(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n\n    return value as T;\n  };\n\n  const inputs: ProdInputs = {x: $x};\n  const attrs: ProdAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prod,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const prod = op({prod_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a random number generator\n * function defined by the user.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param randFunction A random number generator function which is called\n * for each element in the output tensor.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n */\nfunction rand_<R extends Rank>(\n    shape: ShapeMap[R], randFunction: () => number,\n    dtype?: DataType): Tensor<R> {\n  const size = sizeFromShape(shape);\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  for (let i = 0; i < size; i++) {\n    values[i] = randFunction();\n  }\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n\nexport const rand = op({rand_});\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as seedrandom from 'seedrandom';\n\nimport {expectNumbersClose, testEpsilon} from '../test_util';\nimport {TypedArray} from '../types';\n\nexport interface RandomBase {\n  nextValue(): number;\n}\n\nexport interface RandomGamma {\n  nextValue(): number;\n}\n\nexport interface RandNormalDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\nexport interface RandGammaDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\n// https://en.wikipedia.org/wiki/Marsaglia_polar_method\nexport class MPRandGauss implements RandomBase {\n  private mean: number;\n  private stdDev: number;\n  private nextVal: number;\n  private dtype?: keyof RandNormalDataTypes;\n  private truncated?: boolean;\n  private upper?: number;\n  private lower?: number;\n  private random: seedrandom.prng;\n\n  constructor(\n      mean: number, stdDeviation: number, dtype?: keyof RandNormalDataTypes,\n      truncated?: boolean, seed?: number) {\n    this.mean = mean;\n    this.stdDev = stdDeviation;\n    this.dtype = dtype;\n    this.nextVal = NaN;\n    this.truncated = truncated;\n    if (this.truncated) {\n      this.upper = this.mean + this.stdDev * 2;\n      this.lower = this.mean - this.stdDev * 2;\n    }\n    const seedValue = seed ? seed : Math.random();\n    this.random = seedrandom.alea(seedValue.toString());\n  }\n\n  /** Returns next sample from a Gaussian distribution. */\n  public nextValue(): number {\n    if (!isNaN(this.nextVal)) {\n      const value = this.nextVal;\n      this.nextVal = NaN;\n      return value;\n    }\n\n    let resultX: number, resultY: number;\n    let isValid = false;\n    while (!isValid) {\n      let v1: number, v2: number, s: number;\n      do {\n        v1 = 2 * this.random() - 1;\n        v2 = 2 * this.random() - 1;\n        s = v1 * v1 + v2 * v2;\n      } while (s >= 1 || s === 0);\n\n      const mul = Math.sqrt(-2.0 * Math.log(s) / s);\n      resultX = this.mean + this.stdDev * v1 * mul;\n      resultY = this.mean + this.stdDev * v2 * mul;\n\n      if (!this.truncated || this.isValidTruncated(resultX)) {\n        isValid = true;\n      }\n    }\n\n    if (!this.truncated || this.isValidTruncated(resultY)) {\n      this.nextVal = this.convertValue(resultY);\n    }\n    return this.convertValue(resultX);\n  }\n\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype == null || this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  /** Returns true if less than 2-standard-deviations from the mean. */\n  private isValidTruncated(value: number): boolean {\n    return value <= this.upper && value >= this.lower;\n  }\n}\n\n// Marsaglia, George, and Wai Wan Tsang. 2000. \"A Simple Method for Generating\n// Gamma Variables.\"\nexport class RandGamma implements RandomGamma {\n  private alpha: number;\n  private beta: number;\n  private d: number;\n  private c: number;\n  private dtype?: keyof RandGammaDataTypes;\n  private randu: seedrandom.prng;\n  private randn: MPRandGauss;\n\n  constructor(\n      alpha: number, beta: number, dtype: keyof RandGammaDataTypes,\n      seed?: number) {\n    this.alpha = alpha;\n    this.beta = 1 / beta;  // convert rate to scale parameter\n    this.dtype = dtype;\n\n    const seedValue = seed ? seed : Math.random();\n    this.randu = seedrandom.alea(seedValue.toString());\n    this.randn = new MPRandGauss(0, 1, dtype, false, this.randu());\n\n    if (alpha < 1) {\n      this.d = alpha + (2 / 3);\n    } else {\n      this.d = alpha - (1 / 3);\n    }\n    this.c = 1 / Math.sqrt(9 * this.d);\n  }\n\n  /** Returns next sample from a gamma distribution. */\n  public nextValue(): number {\n    let x2: number, v0: number, v1: number, x: number, u: number, v: number;\n    while (true) {\n      do {\n        x = this.randn.nextValue();\n        v = 1 + (this.c * x);\n      } while (v <= 0);\n      v *= v * v;\n      x2 = x * x;\n      v0 = 1 - (0.331 * x2 * x2);\n      v1 = (0.5 * x2) + (this.d * (1 - v + Math.log(v)));\n      u = this.randu();\n      if (u < v0 || Math.log(u) < v1) {\n        break;\n      }\n    }\n    v = (1 / this.beta) * this.d * v;\n    if (this.alpha < 1) {\n      v *= Math.pow(this.randu(), 1 / this.alpha);\n    }\n    return this.convertValue(v);\n  }\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n}\n\nexport class UniformRandom implements RandomBase {\n  private min: number;\n  private range: number;\n  private random: seedrandom.prng;\n  private dtype?: keyof RandNormalDataTypes;\n\n  constructor(\n      min = 0, max = 1, dtype?: keyof RandNormalDataTypes,\n      seed?: string|number) {\n    this.min = min;\n    this.range = max - min;\n    this.dtype = dtype;\n    if (seed == null) {\n      seed = Math.random();\n    }\n    if (typeof seed === 'number') {\n      seed = seed.toString();\n    }\n\n    if (!this.canReturnFloat() && this.range <= 1) {\n      throw new Error(\n          `The difference between ${min} - ${max} <= 1 and dtype is not float`);\n    }\n    this.random = seedrandom.alea(seed);\n  }\n\n  /** Handles proper rounding for non floating point numbers. */\n  private canReturnFloat = () =>\n      (this.dtype == null || this.dtype === 'float32');\n\n  private convertValue(value: number): number {\n    if (this.canReturnFloat()) {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  nextValue() {\n    return this.convertValue(this.min + this.range * this.random());\n  }\n}\n\nexport function jarqueBeraNormalityTest(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test\n  const n = values.length;\n  const s = skewness(values);\n  const k = kurtosis(values);\n  const jb = n / 6 * (Math.pow(s, 2) + 0.25 * Math.pow(k - 3, 2));\n  // JB test requires 2-degress of freedom from Chi-Square @ 0.95:\n  // http://www.itl.nist.gov/div898/handbook/eda/section3/eda3674.htm\n  const CHI_SQUARE_2DEG = 5.991;\n  if (jb > CHI_SQUARE_2DEG) {\n    throw new Error(`Invalid p-value for JB: ${jb}`);\n  }\n}\n\nexport function expectArrayInMeanStdRange(\n    actual: TypedArray|number[], expectedMean: number, expectedStdDev: number,\n    epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  const actualMean = mean(actual);\n  expectNumbersClose(actualMean, expectedMean, epsilon);\n  expectNumbersClose(\n      standardDeviation(actual, actualMean), expectedStdDev, epsilon);\n}\n\nfunction mean(values: TypedArray|number[]) {\n  let sum = 0;\n  for (let i = 0; i < values.length; i++) {\n    sum += values[i];\n  }\n  return sum / values.length;\n}\n\nfunction standardDeviation(values: TypedArray|number[], mean: number) {\n  let squareDiffSum = 0;\n  for (let i = 0; i < values.length; i++) {\n    const diff = values[i] - mean;\n    squareDiffSum += diff * diff;\n  }\n  return Math.sqrt(squareDiffSum / values.length);\n}\n\nfunction kurtosis(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Kurtosis\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum4 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum4 += Math.pow(v, 4);\n  }\n  return (1 / n) * sum4 / Math.pow((1 / n) * sum2, 2);\n}\n\nfunction skewness(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Skewness\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum3 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum3 += Math.pow(v, 3);\n  }\n  return (1 / n) * sum3 / Math.pow((1 / (n - 1)) * sum2, 3 / 2);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {RandGamma} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a gamma distribution.\n *\n * ```js\n * tf.randomGamma([2, 2], 1).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param alpha The shape parameter of the gamma distribution.\n * @param beta The inverse scale parameter of the gamma distribution. Defaults\n *     to 1.\n * @param dtype The data type of the output. Defaults to float32.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomGamma_<R extends Rank>(\n    shape: ShapeMap[R], alpha: number, beta = 1,\n    dtype: 'float32'|'int32' = 'float32', seed?: number): Tensor<R> {\n  if (beta == null) {\n    beta = 1;\n  }\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const rgamma = new RandGamma(alpha, beta, dtype, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = rgamma.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomGamma = op({randomGamma_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a normal distribution.\n *\n * ```js\n * tf.randomNormal([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, false /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomNormal = op({randomNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {UniformRandom} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a uniform distribution.\n *\n * The generated values follow a uniform distribution in the range [minval,\n * maxval). The lower bound minval is included in the range, while the upper\n * bound maxval is excluded.\n *\n * ```js\n * tf.randomUniform([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param minval The lower bound on the range of random values to generate.\n *   Defaults to 0.\n * @param maxval The upper bound on the range of random values to generate.\n *   Defaults to 1.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomUniform_<R extends Rank>(\n    shape: ShapeMap[R], minval = 0, maxval = 1, dtype: DataType = 'float32',\n    seed?: number|string): Tensor<R> {\n  const res = buffer(shape, dtype);\n  const random = new UniformRandom(minval, maxval, null, seed);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = random.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomUniform = op({randomUniform_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike1D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-1 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor1d` as it makes the code more readable.\n *\n * ```js\n * tf.tensor1d([1, 2, 3]).print();\n * ```\n *\n * @param values The values of the tensor. Can be array of numbers,\n *     or a `TypedArray`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor1d(values: TensorLike1D, dtype?: DataType): Tensor1D {\n  assertNonNull(values);\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 1) {\n    throw new Error('tensor1d() requires values to be a flat/TypedArray');\n  }\n  const shape: number[] = null;\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Range, RangeAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {makeZerosTypedArray} from '../util';\n\nimport {tensor1d} from './tensor1d';\nimport {zeros} from './zeros';\n\n/**\n * Creates a new `tf.Tensor1D` filled with the numbers in the range provided.\n *\n * The tensor is a is half-open interval meaning it includes start, but\n * excludes stop. Decrementing ranges and negative step values are also\n * supported.sv\n *\n *\n * ```js\n * tf.range(0, 9, 2).print();\n * ```\n *\n * @param start An integer start value\n * @param stop An integer stop value\n * @param step An integer increment (will default to 1 or -1)\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function range(\n    start: number, stop: number, step = 1,\n    dtype: 'float32'|'int32' = 'float32'): Tensor1D {\n  if (step === 0) {\n    throw new Error('Cannot have a step of zero');\n  }\n\n  const forward: ForwardFunc<Tensor> = () => {\n    const sameStartStop = start === stop;\n    const increasingRangeNegativeStep = start < stop && step < 0;\n    const decreasingRangePositiveStep = stop < start && step > 1;\n\n    if (sameStartStop || increasingRangeNegativeStep ||\n        decreasingRangePositiveStep) {\n      return zeros([0], dtype);\n    }\n\n    const numElements = Math.abs(Math.ceil((stop - start) / step));\n    const values = makeZerosTypedArray(numElements, dtype);\n\n    if (stop < start && step === 1) {\n      // Auto adjust the step's sign if it hasn't been set\n      // (or was set to 1)\n      step = -1;\n    }\n\n    values[0] = start;\n    for (let i = 1; i < values.length; i++) {\n      values[i] = values[i - 1] + step;\n    }\n\n    return tensor1d(values, dtype);\n  };\n\n  const attrs: RangeAttrs = {start, stop, step, dtype};\n\n  return ENGINE.runKernelFunc(\n             forward, {} /* inputs */, null /* grad */, Range,\n             attrs as {} as NamedAttrMap) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Reciprocal, ReciprocalInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of x element-wise: `1 / x`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, 2]);\n *\n * x.reciprocal().print();  // or tf.reciprocal(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction reciprocal_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'reciprocal');\n\n  const inputs: ReciprocalInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.reciprocal($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Reciprocal);\n}\nexport const reciprocal = op({reciprocal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu, ReluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport { cast } from './cast';\n\n/**\n * Computes rectified linear element-wise: `max(x, 0)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.relu().print();  // or tf.relu(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu($x);\n  };\n\n  const inputs: ReluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu) as\n      T;\n}\n\nexport const relu = op({relu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu6, Relu6Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes rectified linear 6 element-wise: `min(max(x, 0), 6)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 8]);\n *\n * x.relu6().print();  // or tf.relu6(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu6_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu6');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu6($x);\n  };\n\n  const inputs: Relu6Inputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu6) as\n      T;\n}\n\nexport const relu6 = op({relu6_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reverse, ReverseAttrs, ReverseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Reverses a `tf.Tensor` along a specified axis.\n *\n * Also available are stricter rank-specific methods that assert that `x` is\n * of the given rank:\n *   - `tf.reverse1d`\n *   - `tf.reverse2d`\n *   - `tf.reverse3d`\n *   - `tf.reverse4d`\n *\n * Except `tf.reverse1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.reverse().print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.reverse(axis).print();\n * ```\n * @param x The input tensor to be reversed.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction reverse_<T extends Tensor>(\n    x: T|TensorLike, axis?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'reverse');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const axes = parseAxisParam(axis, $x.shape);\n    if ($x.rank === 0) {\n      return clone($x);\n    }\n    const res = backend.reverse($x, axes);\n    return reshape(res, $x.shape);\n  };\n\n  const inputs: ReverseInputs = {x: $x};\n  const attrs: ReverseAttrs = {dims: axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Reverse, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const reverse = op({reverse_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor1D`.\n *\n * @param x The input tensor.\n */\nfunction reverse1d_(x: Tensor1D|TensorLike): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 1,\n      () => `Error in reverse1D: x must be rank 1 but got rank ${$x.rank}.`);\n  return reverse($x, 0);\n}\n\nexport const reverse1d = op({reverse1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor2D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse2d_(x: Tensor2D|TensorLike, axis?: number|number[]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 2,\n      () => `Error in reverse2D: x must be rank 2 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse2d = op({reverse2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor3D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse3d_(x: Tensor3D|TensorLike, axis?: number|number[]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 3,\n      () => `Error in reverse3D: x must be rank 3 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse3d = op({reverse3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor4D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse4d_(x: Tensor4D|TensorLike, axis?: number|number[]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 4,\n      () => `Error in reverse4D: x must be rank 4 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse4d = op({reverse4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Round, RoundInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes round of input `tf.Tensor` element-wise: `round(x)`.\n * It implements banker's rounding.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.round().print();  // or tf.round(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction round_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'round');\n  const inputs: RoundInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.round($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Round);\n}\n\nexport const round = op({round_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Rsqrt, RsqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of square root of the input `tf.Tensor` element-wise:\n * `y = 1 / sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.rsqrt().print();  // or tf.rsqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction rsqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'rsqrt');\n\n  const inputs: RsqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.rsqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Rsqrt);\n}\nexport const rsqrt = op({rsqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Selu, SeluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes scaled exponential linear element-wise.\n *\n * `x < 0 ? scale * alpha * (exp(x) - 1) : x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.selu().print();  // or tf.selu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction selu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'selu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.selu($x);\n    save([$x]);\n    return res;\n  };\n\n  const inputs: SeluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Selu) as\n      T;\n}\n\nexport const selu = op({selu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport {depthwiseConv2d} from './depthwise_conv2d';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * 2-D convolution with separable filters.\n *\n * Performs a depthwise convolution that acts separately on channels followed\n * by a pointwise convolution that mixes channels. Note that this is\n * separability between dimensions [1, 2] and 3, not spatial separability\n * between dimensions 1 and 2.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param depthwiseFilter The depthwise filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`. This is\n *     the filter used in the first step.\n * @param pointwiseFilter The pointwise filter tensor, rank 4, of shape\n *     `[1, 1, inChannels * channelMultiplier, outChannels]`. This is\n *     the filter used in the second step.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction separableConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthwiseFilter: Tensor4D|TensorLike,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'separableConv2d');\n  const $depthwiseFilter =\n      convertToTensor(depthwiseFilter, 'depthwiseFilter', 'separableConv2d');\n  const $pointwiseFilter =\n      convertToTensor(pointwiseFilter, 'pointwiseFilter', 'separableConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  if (dataFormat === 'NCHW') {\n    throw new Error(\n        'separableConv2d currently does not support dataFormat NCHW; only ' +\n        'NHWC is supported');\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in separableConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $depthwiseFilter.rank === 4,\n      () => `Error in separableConv2d: depthwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.rank === 4,\n      () => `Error in separableConv2d: pointwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.shape[0] === 1,\n      () =>\n          `Error in separableConv2d: the first dimension of pointwise filter ` +\n          ` must be 1, but got ${$pointwiseFilter.shape[0]}.`);\n  util.assert(\n      $pointwiseFilter.shape[1] === 1,\n      () => `Error in separableConv2d: the second dimension of pointwise ` +\n          `filter must be 1, but got ${$pointwiseFilter.shape[1]}.`);\n\n  const inChannels = $depthwiseFilter.shape[2];\n  const channelMultiplier = $depthwiseFilter.shape[3];\n  util.assert(\n      $pointwiseFilter.shape[2] === inChannels * channelMultiplier,\n      () =>\n          `Error in separableConv2d: the third dimension of pointwise filter ` +\n          `must be ${inChannels * channelMultiplier}, ` +\n          `but got ${$pointwiseFilter.shape[2]}.`);\n\n  const depthwise = depthwiseConv2d(\n      x4D, $depthwiseFilter, strides, pad, dataFormat, dilation);\n  const pointwiseStride = 1;\n  const res =\n      conv2d(depthwise, $pointwiseFilter, pointwiseStride, 'valid', dataFormat);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const separableConv2d = op({separableConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\n/**\n * Computes the difference between two lists of numbers.\n *\n * Given a Tensor `x` and a Tensor `y`, this operation returns a Tensor `out`\n * that represents all values that are in `x` but not in `y`. The returned\n * Tensor `out` is sorted in the same order that the numbers appear in `x`\n * (duplicates are preserved). This operation also returns a Tensor indices that\n * represents the position of each out element in `x`. In other words:\n *\n * `out[i] = x[idx[i]] for i in [0, 1, ..., out.length - 1]`\n *\n * ```js\n * const x = [1, 2, 3, 4, 5, 6];\n * const y = [1, 3, 5];\n *\n * const [out, indices] = await tf.setdiff1dAsync(x, y);\n * out.print(); // [2, 4, 6]\n * indices.print(); // [1, 3, 5]\n * ```\n *\n * @param x 1-D Tensor. Values to keep.\n * @param y 1-D Tensor. Must have the same type as x. Values to exclude in the\n *     output.\n * @returns Promise of Tensor tuple [out, indices].\n *  out: Tensor with the same type as x.\n *  indices: A Tensor of type int32.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nasync function setdiff1dAsync_(\n    x: Tensor|TensorLike, y: Tensor|TensorLike): Promise<[Tensor, Tensor]> {\n  const $x = convertToTensor(x, 'x', 'setdiff1d');\n  const $y = convertToTensor(y, 'y', 'setdiff1d');\n\n  util.assert(\n      $x.dtype === $y.dtype,\n      () => `x and y should have the same dtype, but got x (${\n          $x.dtype}) and y (${$y.dtype}).`);\n\n  util.assert(\n      $x.rank === 1, () => `x should be 1D tensor, but got x (${$x.shape}).`);\n\n  util.assert(\n      $y.rank === 1, () => `y should be 1D tensor, but got y (${$y.shape}).`);\n\n  const xVals = await $x.data();\n  const yVals = await $y.data();\n  const ySet = new Set(yVals);\n\n  let outputSize = 0;\n  for (let i = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      outputSize++;\n    }\n  }\n\n  const buffer = new TensorBuffer([outputSize], $x.dtype);\n  const indices = new TensorBuffer([outputSize], 'int32');\n  for (let i = 0, p = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      buffer.values[p] = xVals[i];\n      indices.values[p] = i;\n      p++;\n    }\n  }\n  return [buffer.toTensor(), indices.toTensor()];\n}\nexport const setdiff1dAsync = setdiff1dAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sign, SignInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns an element-wise indication of the sign of a number.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3, NaN, 0]);\n *\n * x.sign().print();  // or tf.sign(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sign_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sign');\n  const inputs: SignInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.sign($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Sign);\n}\nexport const sign = op({sign_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sin, SinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sin of the input Tensor element-wise: `sin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.sin().print();  // or tf.sin(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sin');\n\n  const inputs: SinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sin);\n}\nexport const sin = op({sin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sinh, SinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic sin of the input `tf.Tensor` element-wise: `sinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.sinh().print();  // or tf.sinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sinh');\n  const inputs: SinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sinh);\n}\nexport const sinh = op({sinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 1D slice from 1D array starting at coordinates `begin` and is\n * of length `size`. See `slice` for details.\n */\nfunction slice1d_(\n    x: Tensor1D|TensorLike, begin: number, size: number): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'slice1d');\n  util.assert(\n      $x.rank === 1,\n      () =>\n          `slice1d expects a rank-1 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, [begin], [size]);\n}\nexport const slice1d = op({slice1d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 2D slice from a 2D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice2d_(\n    x: Tensor2D|TensorLike, begin: [number, number],\n    size: [number, number]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'slice2d');\n  util.assert(\n      $x.rank === 2,\n      () =>\n          `slice2d expects a rank-2 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice2d = op({slice2d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 3D slice from a 3D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice3d_(\n    x: Tensor3D|TensorLike, begin: [number, number, number],\n    size: [number, number, number]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'slice3d');\n  util.assert(\n      $x.rank === 3,\n      () =>\n          `slice3d expects a rank-3 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice3d = op({slice3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 4D slice from a 4D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice4d_(\n    x: Tensor4D|TensorLike, begin: [number, number, number, number],\n    size: [number, number, number, number]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'slice4d');\n  util.assert(\n      $x.rank === 4,\n      () =>\n          `slice4d expects a rank-4 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice4d = op({slice4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softmax, SoftmaxAttrs, SoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the softmax normalized vector given the logits.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction softmax_<T extends Tensor>(logits: T|TensorLike, dim = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'softmax', 'float32');\n\n  if (dim === -1) {\n    dim = $logits.rank - 1;\n  }\n  if (dim !== $logits.rank - 1) {\n    throw Error(\n        'Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and dim was ${dim}`);\n  }\n\n  const inputs: SoftmaxInputs = {logits: $logits};\n  const attrs: SoftmaxAttrs = {dim};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const y = backend.softmax($logits, dim);\n        save([y]);\n        return y;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, Softmax,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const softmax = op({softmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FFT, FFTInputs} from '../../kernel_names';\nimport {Tensor} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\n\n/**\n * Fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the inner-most\n * dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.fft().print();  // tf.spectral.fft(x).print();\n * ```\n * @param input The complex input to compute an fft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction fft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.fft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: FFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D = input.as2D(batch, innerDimensionSize);\n    const result = backend.fft(input2D);\n    return result.reshape(input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, FFT);\n}\n\nexport const fft = op({fft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {IFFT, IFFTInputs} from '../../kernel_names';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Inverse fast Fourier transform.\n *\n * Computes the inverse 1-dimensional discrete Fourier transform over the\n * inner-most dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.ifft().print();  // tf.spectral.ifft(x).print();\n * ```\n * @param input The complex input to compute an ifft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction ifft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.ifft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: IFFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D: Tensor2D = reshape(input, [batch, innerDimensionSize]);\n    const result = backend.ifft(input2D);\n    return reshape(result, input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, IFFT);\n}\n\nexport const ifft = op({ifft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {reverse} from '../reverse';\nimport {scalar} from '../scalar';\nimport {slice} from '../slice';\n\nimport {ifft} from './ifft';\n\n/**\n * Inversed real value input fast Fourier transform.\n *\n * Computes the 1-dimensional inversed discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([0, 0, 0]);\n * const x = tf.complex(real, imag);\n *\n * x.irfft().print();\n * ```\n * @param input The real value input to compute an irfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction irfft_(input: Tensor): Tensor {\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n  let ret: Tensor;\n  if (innerDimensionSize <= 2) {\n    const complexInput = reshape(input, [batch, innerDimensionSize]);\n    ret = ifft(complexInput);\n  } else {\n    // The length of unique components of the DFT of a real-valued signal\n    // is 2 * (input_len - 1)\n    const outputShape = [batch, 2 * (innerDimensionSize - 1)];\n    const realInput = reshape(real(input), [batch, innerDimensionSize]);\n    const imagInput = reshape(imag(input), [batch, innerDimensionSize]);\n\n    const realConjugate =\n        reverse(slice(realInput, [0, 1], [batch, innerDimensionSize - 2]), 1);\n    const imagConjugate: Tensor2D = mul(\n        reverse(slice(imagInput, [0, 1], [batch, innerDimensionSize - 2]), 1),\n        scalar(-1));\n\n    const r = concat([realInput, realConjugate], 1);\n    const i = concat([imagInput, imagConjugate], 1);\n    const complexInput =\n        reshape(complex(r, i), [outputShape[0], outputShape[1]]);\n    ret = ifft(complexInput);\n  }\n  ret = real(ret);\n  // reshape the result if the input is 3D tensor.\n  if (input.rank === 3 && input.shape[0] !== 0) {\n    const temp = ret;\n    const batch = input.shape[0];\n    ret = reshape(ret, [batch, ret.shape[0] / batch, ret.shape[1]]);\n    temp.dispose();\n  }\n  return ret;\n}\n\nexport const irfft = op({irfft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {assert} from '../util';\n\n/**\n * Prepare the split size array. When the input is a number, the axis is evenly\n * divided among the split size. When the input contains the negative value, the\n * rest of the axis is allocated toward that.\n */\nexport function prepareSplitSize(\n    x: Tensor|TensorInfo, numOrSizeSplits: number[]|number,\n    axis = 0): number[] {\n  let splitSizes = [];\n  if (typeof (numOrSizeSplits) === 'number') {\n    assert(\n        x.shape[axis] % numOrSizeSplits === 0,\n        () => 'Number of splits must evenly divide the axis.');\n    splitSizes =\n        new Array(numOrSizeSplits).fill(x.shape[axis] / numOrSizeSplits);\n  } else {\n    const numOfNegs = numOrSizeSplits.reduce((count, value) => {\n      if (value === -1) {\n        count += 1;\n      }\n      return count;\n    }, 0);\n    assert(\n        numOfNegs <= 1,\n        () => 'There should be only one negative value in split array.');\n    const negIndex = numOrSizeSplits.indexOf(-1);\n    // Allow the number of split array to be -1, which indicates the rest\n    // of dimension is allocated to that split.\n    if (negIndex !== -1) {\n      const total = numOrSizeSplits.reduce((a, b) => b > 0 ? a + b : a);\n      numOrSizeSplits[negIndex] = x.shape[axis] - total;\n    }\n    assert(\n        x.shape[axis] === numOrSizeSplits.reduce((a, b) => a + b),\n        () => 'The sum of sizes must match the size of the axis dimension.');\n    splitSizes = numOrSizeSplits;\n  }\n\n  return splitSizes;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SplitV, SplitVAttrs, SplitVInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {prepareSplitSize} from './split_util';\n\n/**\n * Splits a `tf.Tensor` into sub tensors.\n *\n * If `numOrSizeSplits` is a number, splits `x` along dimension `axis`\n * into `numOrSizeSplits` smaller tensors.\n * Requires that `numOrSizeSplits` evenly divides `x.shape[axis]`.\n *\n * If `numOrSizeSplits` is a number array, splits `x` into\n * `numOrSizeSplits.length` pieces. The shape of the `i`-th piece has the\n * same size as `x` except along dimension `axis` where the size is\n * `numOrSizeSplits[i]`.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4, 5, 6, 7, 8], [2, 4]);\n * const [a, b] = tf.split(x, 2, 1);\n * a.print();\n * b.print();\n *\n * const [c, d, e] = tf.split(x, [1, 2, 1], 1);\n * c.print();\n * d.print();\n * e.print();\n * ```\n *\n * @param x The input tensor to split.\n * @param numOrSizeSplits Either an integer indicating the number of\n * splits along the axis or an array of integers containing the sizes of\n * each output tensor along the axis. If a number then it must evenly divide\n * `x.shape[axis]`; otherwise the sum of sizes must match `x.shape[axis]`.\n * Can contain one -1 indicating that dimension is to be inferred.\n * @param axis The dimension along which to split. Defaults to 0 (the first\n * dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction split_<T extends Tensor>(\n    x: Tensor|TensorLike, numOrSizeSplits: number[]|number, axis = 0): T[] {\n  const $x = convertToTensor(x, 'x', 'split');\n\n  const forward: ForwardFunc<Tensor> = (backend, _) => {\n    const $axis = parseAxisParam(axis, $x.shape)[0];\n    const splitSizes = prepareSplitSize($x, numOrSizeSplits, $axis);\n    return backend.split($x, splitSizes, $axis) as {} as T;\n  };\n\n  const inputs: SplitVInputs = {x: $x};\n  const attr: SplitVAttrs = {numOrSizeSplits, axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, SplitV,\n             attr as {} as NamedAttrMap) as {} as T[];\n}\n\nexport const split = op({split_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {assert} from '../../util';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {split} from '../split';\nimport {zeros} from '../zeros';\nimport {zerosLike} from '../zeros_like';\n\nimport {fft} from './fft';\n\n/**\n * Real value input fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n *\n * real.rfft().print();\n * ```\n * @param input The real value input to compute an rfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction rfft_(input: Tensor, fftLength?: number): Tensor {\n  assert(\n      input.dtype === 'float32',\n      () => `The dtype for rfft() must be real value but got ${input.dtype}`);\n\n  let innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n\n  let adjustedInput: Tensor;\n  if (fftLength != null && fftLength < innerDimensionSize) {\n    // Need to crop\n    const begin = input.shape.map(v => 0);\n    const size = input.shape.map(v => v);\n    size[input.shape.length - 1] = fftLength;\n    adjustedInput = slice(input, begin, size);\n    innerDimensionSize = fftLength;\n  } else if (fftLength != null && fftLength > innerDimensionSize) {\n    // Need to pad with zeros\n    const zerosShape = input.shape.map(v => v);\n    zerosShape[input.shape.length - 1] = fftLength - innerDimensionSize;\n    adjustedInput = concat([input, zeros(zerosShape)], input.shape.length - 1);\n    innerDimensionSize = fftLength;\n  } else {\n    adjustedInput = input;\n  }\n\n  // Complement the input with zero imaginary numbers.\n  const zerosInput = zerosLike(adjustedInput);\n  const complexInput =\n      reshape(complex(adjustedInput, zerosInput), [batch, innerDimensionSize]);\n\n  const ret = fft(complexInput);\n\n  // Exclude complex conjugations. These conjugations are put symmetrically.\n  const half = Math.floor(innerDimensionSize / 2) + 1;\n  const realValues = real(ret);\n  const imagValues = imag(ret);\n  const realComplexConjugate = split(\n      realValues, [half, innerDimensionSize - half],\n      realValues.shape.length - 1);\n  const imagComplexConjugate = split(\n      imagValues, [half, innerDimensionSize - half],\n      imagValues.shape.length - 1);\n\n  const outputShape = adjustedInput.shape.slice();\n  outputShape[adjustedInput.shape.length - 1] = half;\n\n  return reshape(\n      complex(realComplexConjugate[0], imagComplexConjugate[0]), outputShape);\n}\n\nexport const rfft = op({rfft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sqrt, SqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes square root of the input `tf.Tensor` element-wise: `y = sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.sqrt().print();  // or tf.sqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sqrt');\n\n  const inputs: SqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sqrt);\n}\nexport const sqrt = op({sqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SquaredDifference, SquaredDifferenceInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns (a - b) * (a - b) element-wise.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * ```js\n * // Broadcast squared difference  a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction squaredDifference_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'squaredDifference');\n  let $b = convertToTensor(b, 'b', 'squaredDifference');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.squaredDifference($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SquaredDifferenceInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             SquaredDifference, attrs) as T;\n}\n\nexport const squaredDifference = op({squaredDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {squeezeShape} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Removes dimensions of size 1 from the shape of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4], [1, 1, 4]);\n * x.squeeze().print();\n * ```\n *\n * @param x The input tensor to be squeezed.\n * @param axis An optional list of numbers. If specified, only\n *     squeezes the dimensions listed. The dimension index starts at 0. It\n * is an error to squeeze a dimension that is not 1.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction squeeze_<T extends Tensor>(x: Tensor|TensorLike, axis?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'squeeze');\n  return reshape($x, squeezeShape($x.shape, axis).newShape) as T;\n}\n\nexport const squeeze = op({squeeze_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {concat} from './concat';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\n\n/**\n * Stacks a list of rank-`R` `tf.Tensor`s into one rank-`(R+1)` `tf.Tensor`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.stack([a, b, c]).print();\n * ```\n *\n * @param tensors A list of tensor objects with the same shape and dtype.\n * @param axis The axis to stack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction stack_<T extends Tensor>(\n    tensors: Array<T|TensorLike>, axis = 0): Tensor {\n  const $tensors = convertToTensorArray(tensors, 'tensors', 'stack');\n\n  util.assert(\n      $tensors.length >= 1, () => 'Pass at least one tensor to tf.stack');\n\n  if ($tensors.length === 1) {\n    return expandDims($tensors[0], axis);\n  }\n\n  const rank = $tensors[0].rank;\n  const shape = $tensors[0].shape;\n  const dtype = $tensors[0].dtype;\n\n  util.assert(axis <= rank, () => 'Axis must be <= rank of the tensor');\n\n  $tensors.forEach(t => {\n    util.assertShapesMatch(\n        shape, t.shape,\n        'All tensors passed to stack must have matching shapes');\n    util.assert(\n        dtype === t.dtype,\n        () => 'All tensors passed to stack must have matching dtypes');\n  });\n\n  const expandedTensors = $tensors.map(t => expandDims(t, axis));\n  // Stack exists in the TensorFlow C++ API\n  // (https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/stack) but not\n  // in\n  // https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/core/ops/ops.pbtxt.\n  // Therefore we are treating it like a high-level op rather than\n  // creating a dedicated stack kernel.\n  return concat(expandedTensors, axis);\n}\n\nexport const stack = op({stack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Step, StepAttrs, StepInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes step of the input `tf.Tensor` element-wise: `x > 0 ? 1 : alpha * x`\n *\n * ```js\n * const x = tf.tensor1d([0, 2, -1, -3]);\n *\n * x.step(.5).print();  // or tf.step(x, .5)\n * ```\n * @param x The input tensor.\n * @param alpha The gradient when input is negative.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction step_<T extends Tensor>(x: T|TensorLike, alpha = 0.0): T {\n  const $x = convertToTensor(x, 'x', 'step');\n\n  const inputs: StepInputs = {x: $x};\n  const attrs: StepAttrs = {alpha};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.step($x, alpha), inputs as {} as NamedTensorMap,\n      null /* grad */, Step, attrs as {} as NamedAttrMap);\n}\nexport const step = op({step_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {StridedSlice, StridedSliceAttrs, StridedSliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {slice} from './slice';\nimport {computeOutShape, getNormalizedAxes, maskToAxes} from './slice_util';\n\n/**\n * Extracts a strided slice of a tensor.\n *\n * Roughly speaking, this op extracts a slice of size (end-begin)/stride from\n * the given input tensor (x). Starting at the location specified by begin the\n * slice continues by adding stride to the index until all dimensions are not\n * less than end. Note that a stride can be negative, which causes a reverse\n * slice.\n *\n * ```js\n * const t = tf.tensor3d([1, 1, 1 ,2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6],\n *    [3, 2, 3]);\n * t.stridedSlice([1, 0, 0], [2, 1, 3], [1, 1, 1]).print()  // [[[3, 3, 3]]]\n * t.stridedSlice([1, 0, 0], [2, 2, 3], [1, 1, 1]).print()  // [[[3, 3, 3],\n *                                                     // [4, 4, 4]]]\n * t.stridedSlice([1, -1, 0], [2, -3, 3], [1, -1, 1]).print() // [[[4, 4, 4],\n *                                                     // [3, 3, 3]]]\n * ```\n *\n * @param x The tensor to stride slice.\n * @param begin The coordinates to start the slice from.\n * @param end: The coordinates to end the slice at.\n * @param strides: The size of the slice.\n * @param beginMask: If the ith bit of beginMask is set, begin[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param endMask: If the ith bit of endMask is set, end[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param shrinkAxisMask: a bitmask where bit i implies that\n * the ith specification should shrink the dimensionality. begin and end must\n * imply a slice of size 1 in the dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction stridedSlice_(\n    x: Tensor|TensorLike, begin: number[], end: number[], strides?: number[],\n    beginMask = 0, endMask = 0, ellipsisMask = 0, newAxisMask = 0,\n    shrinkAxisMask = 0): Tensor {\n  let $x = convertToTensor(x, 'x', 'stridedSlice');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if (strides == null) {\n      strides = new Array(begin.length);\n    }\n\n    const ellipsisAxes = maskToAxes(ellipsisMask);\n    if (ellipsisAxes.length > 1) {\n      throw new Error('Multiple ellipses in slice is not allowed.');\n    }\n\n    if (ellipsisMask !== 0 && newAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and newAxisMask is not yet supported.');\n    }\n\n    if (ellipsisMask !== 0 && shrinkAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and shrinkAxisMask is not yet supported.');\n    }\n\n    const numInterpolatedAxes = $x.rank - begin.length;\n\n    // Expand the dims of x based on the newAxisMask.\n    const expandAxes = maskToAxes(newAxisMask);\n    const newShape = $x.shape.slice();\n    expandAxes.forEach(axis => {\n      begin[axis] = 0;\n      end[axis] = 1;\n      newShape.splice(axis, 0, 1);\n    });\n    $x = reshape($x, newShape);\n\n    const {\n      begin: normalizedBegin,\n      end: normalizedEnd,\n      strides: normalizedStrides\n    } =\n        getNormalizedAxes(\n            $x.shape, ellipsisAxes, numInterpolatedAxes, begin, end, strides,\n            beginMask, endMask, ellipsisMask);\n    begin = normalizedBegin;\n    end = normalizedEnd;\n    strides = normalizedStrides;\n\n    const shrinkAxes = maskToAxes(shrinkAxisMask);\n    // Adjust the ends based on the shrink mask.\n    shrinkAxes.forEach(axis => {\n      end[axis] = begin[axis] + 1;\n      strides[axis] = 1;\n    });\n\n    // Figure out the output shape.\n    const size = computeOutShape(begin, end, strides);\n    // Remove the axes based on shrinkMask.\n    const outShape = size.filter((_, axis) => shrinkAxes.indexOf(axis) === -1);\n\n    const nonStrided = strides.every(v => v === 1);\n    if (nonStrided) {\n      return reshape(slice($x, begin, size), outShape);\n    }\n\n    const res = backend.stridedSlice($x, begin, end, strides);\n    return reshape(res, outShape);\n  };\n\n  const inputs: StridedSliceInputs = {x: $x};\n  const attrs: StridedSliceAttrs = {\n    begin,\n    end,\n    strides,\n    beginMask,\n    endMask,\n    ellipsisMask,\n    newAxisMask,\n    shrinkAxisMask\n  };\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, StridedSlice,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const stridedSlice = op({stridedSlice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tan, TanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes tan of the input `tf.Tensor` element-wise, `tan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.tan().print();  // or tf.tan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tan');\n\n  const inputs: TanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.tan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tan);\n}\nexport const tan = op({tan_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike2D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-2 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor2d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor2d([[1, 2], [3, 4]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor2d(\n    values: TensorLike2D, shape?: [number, number],\n    dtype?: DataType): Tensor2D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 2) {\n    throw new Error('tensor2d() requires shape to have two numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 2 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor2d() requires values to be number[][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor2d() requires shape to be provided when `values` ' +\n        'are a flat/TypedArray');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-4 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor4d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor4d([[[[1], [2]], [[3], [4]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor4d(\n    values: TensorLike4D, shape?: [number, number, number, number],\n    dtype?: DataType): Tensor4D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 4) {\n    throw new Error('tensor4d() requires shape to have four numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 4 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor4d() requires values to be number[][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor4d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor4D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor5D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike5D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-5 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor5d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor5d([[[[[1], [2]], [[3], [4]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor5d(\n    values: TensorLike5D, shape?: [number, number, number, number, number],\n    dtype?: DataType): Tensor5D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 5) {\n    throw new Error('tensor5d() requires shape to have five numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 5 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor5d() requires values to be ' +\n        'number[][][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor5d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor5D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor6D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike6D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-6 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor6d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor6d([[[[[[1],[2]],[[3],[4]]],[[[5],[6]],[[7],[8]]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor6d([1, 2, 3, 4, 5, 6, 7, 8], [1, 1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor6d(\n    values: TensorLike6D,\n    shape?: [number, number, number, number, number, number],\n    dtype?: DataType): Tensor6D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 6) {\n    throw new Error('tensor6d() requires shape to have six numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 6 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor6d() requires values to be number[][][][][][] or ' +\n        'flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor6d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  shape = shape ||\n      inferredShape as [number, number, number, number, number, number];\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor6D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {TopK, TopKAttrs, TopKInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {NumericTensor, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Finds the values and indices of the `k` largest entries along the last\n * dimension.\n *\n * If the input is a vector (rank=1), finds the k largest entries in the vector\n * and outputs their values and indices as vectors. Thus values[j] is the j-th\n * largest entry in input, and its index is indices[j].\n * For higher rank inputs, computes the top k entries along the last dimension.\n *\n * If two elements are equal, the lower-index element appears first.\n *\n * ```js\n * const a = tf.tensor2d([[1, 5], [4, 3]]);\n * const {values, indices} = tf.topk(a);\n * values.print();\n * indices.print();\n * ```\n * @param x 1-D or higher `tf.Tensor` with last dimension being at least `k`.\n * @param k Number of top elements to look for along the last dimension.\n * @param sorted If true, the resulting `k` elements will be sorted by the\n *     values in descending order.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction topk_<T extends Tensor>(\n    x: T|TensorLike, k = 1, sorted = true): {values: T, indices: T} {\n  const $x = convertToTensor(x, 'x', 'topk');\n  if ($x.rank === 0) {\n    throw new Error('topk() expects the input to be of rank 1 or higher');\n  }\n  const lastDim = $x.shape[$x.shape.length - 1];\n  if (k > lastDim) {\n    throw new Error(\n        `'k' passed to topk() must be <= the last dimension (${lastDim}) ` +\n        `but got ${k}`);\n  }\n\n  const inputs: TopKInputs = {x: $x};\n  const attrs: TopKAttrs = {k, sorted};\n\n  const [values, indices] = ENGINE.runKernelFunc(\n      b => b.topk($x as NumericTensor, k, sorted),\n      inputs as {} as NamedTensorMap, null /* grad */, TopK,\n      attrs as {} as NamedAttrMap);\n\n  return {values, indices} as {values: T, indices: T};\n}\n\nexport const topk = op({topk_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a truncated normal\n * distribution.\n *\n * ```js\n * tf.truncatedNormal([2, 2]).print();\n * ```\n *\n * The generated values follow a normal distribution with specified mean and\n * standard deviation, except that values whose magnitude is more than 2\n * standard deviations from the mean are dropped and re-picked.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output tensor.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction truncatedNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type $ { dtype }`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, true /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const truncatedNormal = op({truncatedNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Unique, UniqueAttrs, UniqueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Finds unique elements along an axis of a tensor.\n *\n * It returns a tensor `values` containing all of the unique elements along the\n * `axis` of the given tensor `x` in the same order that they occur along the\n * `axis` in `x`; `x` does not need to be sorted. It also returns a tensor\n * `indices` the same size as the number of the elements in `x` along the `axis`\n * dimension. It contains the index in the unique output `values`.\n *\n * ```js\n * // A 1-D tensor\n * const a = tf.tensor1d([1, 1, 2, 4, 4, 4, 7, 8, 8]);\n * const {values, indices} = tf.unique(a);\n * values.print();   // [1, 2, 4, 7, 8,]\n * indices.print();  // [0, 0, 1, 2, 2, 2, 3, 4, 4]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=0\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 0)\n * values.print();   // [[1, 0, 0],\n *                   //  [2, 0, 0]]\n * indices.print();  // [0, 0, 1]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=1\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 1)\n * values.print();   // [[1, 0],\n *                   //  [1, 0],\n *                   //  [2, 0]]\n * indices.print();  // [0, 1, 1]\n * ```\n * @param x A tensor (int32, string, bool).\n * @param axis The axis of the tensor to find the unique elements.\n * @returns [uniqueElements, indices] (see above for details)\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction unique_<T extends Tensor>(\n    x: T|TensorLike, axis = 0): {values: T, indices: Tensor1D} {\n  // x can be of any dtype, thus null as the last argument.\n  const $x = convertToTensor(x, 'x', 'unique', null);\n  assert($x.rank > 0, () => 'The input tensor must be at least 1D');\n\n  const inputs: UniqueInputs = {x: $x};\n  const attrs: UniqueAttrs = {axis};\n  const [values, indices] = ENGINE.runKernel(\n                                Unique, inputs as {} as NamedTensorMap,\n                                attrs as {} as NamedAttrMap) as [T, Tensor1D];\n  return {values, indices};\n}\n\nexport const unique = op({unique_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {UnsortedSegmentSum, UnsortedSegmentSumAttrs, UnsortedSegmentSumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, isInt} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Computes the sum along segments of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const segmentIds = tf.tensor1d([1, 2, 0, 1], 'int32');\n * const numSegments = 3;\n *\n * x.unsortedSegmentSum(segmentIds, numSegments).print()\n * //or tf.unsortedSegmentSum(x, segmentIds, numSegments)\n * ```\n * @param x The `tf.Tensor` that will be summed along its segments.\n * @param segmentIds A `tf.Tensor1D` whose rank is equal to the rank of `x`'s\n * dimension along the `axis`.  Maps each element of `x` to a segment.\n * @param numSegments The number of distinct `segmentIds`.\n *\n * @doc {heading: 'Operations', subheading: 'Segment'}\n */\nfunction unsortedSegmentSum_<T extends Tensor>(\n    x: T|TensorLike, segmentIds: Tensor1D|TensorLike, numSegments: number): T {\n  const $x = convertToTensor(x, 'x', 'unsortedSegmentSum');\n  const $segmentIds =\n      convertToTensor(segmentIds, 'segmentIds', 'unsortedSegmentSum', 'int32');\n  assert(isInt(numSegments), () => 'numSegments must be of dtype int');\n\n  const inputs: UnsortedSegmentSumInputs = {x: $x, segmentIds: $segmentIds};\n  const attrs: UnsortedSegmentSumAttrs = {numSegments};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.unsortedSegmentSum($x, $segmentIds, numSegments);\n    save([$segmentIds]);\n    return res;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             UnsortedSegmentSum, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const unsortedSegmentSum = op({unsortedSegmentSum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Unpack, UnpackAttrs, UnpackInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Unstacks a `tf.Tensor` of rank-`R` into a list of rank-`(R-1)` `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * tf.unstack(a).forEach(tensor => tensor.print());\n * ```\n *\n * @param x A tensor object.\n * @param axis The axis to unstack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction unstack_(x: Tensor|TensorLike, axis = 0): Tensor[] {\n  const $x = convertToTensor(x, 'x', 'unstack');\n  util.assert(\n      axis >= -$x.shape.length && axis < $x.shape.length,\n      () =>\n          `Axis = ${axis} is not in [-${$x.shape.length}, ${$x.shape.length})`);\n  if (axis < 0) {\n    axis += $x.shape.length;\n  }\n  const inputs: UnpackInputs = {value: $x};\n  const attrs: UnpackAttrs = {axis};\n  const forward: ForwardFunc<Tensor[]> = (backend: KernelBackend) =>\n      backend.unstack($x, axis);\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Unpack,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const unstack = op({unstack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor, Variable} from '../tensor';\nimport {DataType, Rank} from '../types';\n\n/**\n * Creates a new variable with the provided initial value.\n * ```js\n * const x = tf.variable(tf.tensor([1, 2, 3]));\n * x.assign(tf.tensor([4, 5, 6]));\n *\n * x.print();\n * ```\n *\n * @param initialValue Initial value for the tensor.\n * @param trainable If true, optimizers are allowed to update it.\n * @param name Name of the variable. Defaults to a unique id.\n * @param dtype If set, initialValue will be converted to the given type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function variable<R extends Rank>(\n    initialValue: Tensor<R>, trainable = true, name?: string,\n    dtype?: DataType): Variable<R> {\n  return ENGINE.makeVariable(initialValue, trainable, name, dtype) as\n      Variable<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the Where kernel shared between cpu and webgl */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor2D} from '../tensor';\nimport {TypedArray} from '../types';\n\nexport function whereImpl(condShape: number[], condVals: TypedArray): Tensor2D {\n  const indices = [];\n  for (let i = 0; i < condVals.length; i++) {\n    if (condVals[i]) {\n      indices.push(i);\n    }\n  }\n\n  const inBuffer = buffer(condShape, 'int32');\n\n  const out = buffer([indices.length, condShape.length], 'int32');\n  for (let i = 0; i < indices.length; i++) {\n    const loc = inBuffer.indexToLoc(indices[i]);\n    const offset = i * condShape.length;\n    out.values.set(loc, offset);\n  }\n  return out.toTensor() as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {whereImpl} from '../backends/where_impl';\nimport {Tensor, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\n/**\n * Returns the coordinates of true elements of condition.\n *\n * The coordinates are returned in a 2-D tensor where the first dimension (rows)\n * represents the number of true elements, and the second dimension (columns)\n * represents the coordinates of the true elements. Keep in mind, the shape of\n * the output tensor can vary depending on how many true values there are in\n * input. Indices are output in row-major order. The resulting tensor has the\n * shape `[numTrueElems, condition.rank]`.\n *\n * This is analogous to calling the python `tf.where(cond)` without an x or y.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const result = await tf.whereAsync(cond);\n * result.print();\n * ```\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nasync function whereAsync_(condition: Tensor|TensorLike): Promise<Tensor2D> {\n  const $condition =\n      convertToTensor(condition, 'condition', 'whereAsync', 'bool');\n  const vals = await $condition.data();\n  const res = whereImpl($condition.shape, vals);\n  if (condition !== $condition) {\n    $condition.dispose();\n  }\n  return res;\n}\n\nexport const whereAsync = whereAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {gather} from './gather';\nimport {reshape} from './reshape';\nimport {squeeze} from './squeeze';\nimport {whereAsync} from './where_async';\n\n/**\n * Apply boolean mask to tensor.\n *\n * ```js\n * const tensor = tf.tensor2d([1, 2, 3, 4, 5, 6], [3, 2]);\n * const mask = tf.tensor1d([1, 0, 1], 'bool');\n * const result = await tf.booleanMaskAsync(tensor, mask);\n * result.print();\n * ```\n *\n * @param tensor N-D tensor.\n * @param mask K-D boolean tensor, K <= N and K must be known statically.\n * @param axis A 0-D int Tensor representing the axis in tensor to mask from.\n *     By default, axis is 0 which will mask from the first dimension.\n *     Otherwise K + axis <= N.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nasync function booleanMaskAsync_(\n    tensor: Tensor|TensorLike, mask: Tensor|TensorLike,\n    axis?: number): Promise<Tensor> {\n  const $tensor = convertToTensor(tensor, 'tensor', 'boolMask');\n  const $mask = convertToTensor(mask, 'mask', 'boolMask', 'bool');\n\n  const axisFrom = axis == null ? 0 : axis;\n  const maskDim = $mask.rank;\n  const tensorShape = $tensor.shape;\n\n  util.assert(maskDim > 0, () => 'mask cannot be scalar');\n  util.assertShapesMatch(\n      tensorShape.slice(axisFrom, axisFrom + maskDim), $mask.shape,\n      `mask's shape must match the first K dimensions of tensor's shape,`);\n\n  let leadingSize = 1;\n  for (let i = axisFrom; i < axisFrom + maskDim; i++) {\n    leadingSize *= tensorShape[i];\n  }\n  const targetTensorShape =\n      tensorShape.slice(0, axisFrom)\n          .concat([leadingSize], tensorShape.slice(axisFrom + maskDim));\n  const reshapedTensor = reshape($tensor, targetTensorShape);\n  const reshapedMask = reshape($mask, [-1]);\n  const positivePositions = await whereAsync(reshapedMask);\n  const indices = squeeze(positivePositions, [1]);\n\n  const res = gather(reshapedTensor, indices, axisFrom);\n\n  // Ensure no memory leak.\n  if (tensor !== $tensor) {\n    $tensor.dispose();\n  }\n  if (mask !== $mask) {\n    $mask.dispose();\n  }\n  indices.dispose();\n  reshapedTensor.dispose();\n  reshapedMask.dispose();\n  positivePositions.dispose();\n\n  return res;\n}\n\nexport const booleanMaskAsync = booleanMaskAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertShapesMatch} from '../util';\n\nimport {equal} from './equal';\nimport {greater} from './greater';\nimport {greaterEqual} from './greater_equal';\nimport {less} from './less';\nimport {lessEqual} from './less_equal';\nimport {notEqual} from './not_equal';\nimport {op} from './operation';\n\n/**\n * @deprecated\n * Strict version of `tf.notEqual` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction notEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'notEqualStrict');\n  const $b = convertToTensor(b, 'b', 'notEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in notEqualStrict: ');\n  return notEqual($a, $b);\n}\n\n/**\n * @deprecated\n * Strict version of `tf.less` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction lessStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessStrict');\n  const $b = convertToTensor(b, 'b', 'lessStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessStrict: ');\n  return less($a, $b);\n}\n\nfunction equalStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'equalStrict');\n  const $b = convertToTensor(b, 'b', 'equalStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in equalStrict: ');\n  return equal($a, $b);\n}\n\nfunction lessEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessEqualStrict');\n  const $b = convertToTensor(b, 'b', 'lessEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessEqualStrict: ');\n  return lessEqual($a, $b);\n}\n\nfunction greaterStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterStrict');\n  const $b = convertToTensor(b, 'b', 'greaterStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterStrict: ');\n  return greater($a, $b);\n}\n\nfunction greaterEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterEqualStrict');\n  const $b = convertToTensor(b, 'b', 'greaterEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterEqualStrict: ');\n  return greaterEqual($a, $b);\n}\n\nexport const equalStrict = op({equalStrict_});\nexport const greaterEqualStrict = op({greaterEqualStrict_});\nexport const greaterStrict = op({greaterStrict_});\nexport const lessEqualStrict = op({lessEqualStrict_});\nexport const lessStrict = op({lessStrict_});\nexport const notEqualStrict = op({notEqualStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {maximum} from './maximum';\nimport {minimum} from './minimum';\nimport {mod} from './mod';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {squaredDifference} from './squared_difference';\nimport {sub} from './sub';\n\n/**\n * @deprecated\n * Adds two `tf.Tensor`s element-wise, A + B.\n *\n * Inputs must be the same shape. For broadcasting support, use add() instead.\n *\n * @param a The first Tensor to add element-wise.\n * @param b The second Tensor to add element-wise.\n */\nfunction addStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'addStrict');\n  const $b = convertToTensor(b, 'b', 'addStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in addStrict: ');\n  return add($a, $b);\n}\n\n/**\n * @deprecated\n * Subtracts two `tf.Tensor`s element-wise, A - B. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.sub` instead.\n *\n * @param a The first Tensor to subtract element-wise.\n * @param b The second Tensor to subtract element-wise.\n */\nfunction subStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'subStrict');\n  const $b = convertToTensor(b, 'b', 'subStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in subStrict: ');\n  return sub($a, $b);\n}\n\n/**\n * @deprecated\n * Computes the power of one `tf.Tensor` to another. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.pow` instead.\n *\n * @param base The base tensor to pow element-wise.\n * @param exp The exponent tensor to pow element-wise.\n */\nfunction powStrict_<T extends Tensor>(base: T, exp: Tensor): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  util.assertShapesMatch(base.shape, exp.shape, 'Error in powStrict: ');\n  return pow(base, exp);\n}\n\n/**\n * @deprecated\n * Multiplies two `tf.Tensor`s element-wise, A * B.\n *\n * Inputs must be the same shape. For broadcasting support, use `tf.mul`.\n *\n * @param a The first tensor to multiply.\n * @param b The first tensor to multiply. Must have the same\n *    dtype as `a`.\n */\nfunction mulStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'mul');\n  const $b = convertToTensor(b, 'b', 'mul');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in multiplyStrict: ');\n  return mul($a, $b);\n}\n\n/**\n * @deprecated\n * Divides two `tf.Tensor`s element-wise, A / B. Inputs must\n * be the same shape.\n *\n * @param a The first tensor as the numerator for element-wise division.\n * @param b The second tensor as the denominator for element-wise division.\n */\nfunction divStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'div');\n  const $b = convertToTensor(b, 'b', 'div');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in divideStrict: ');\n  return div($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the mod of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use mod().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction modStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'modStrict');\n  const $b = convertToTensor(b, 'b', 'modStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in modStrict: ');\n  return mod($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the min of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use minimum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction minimumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'minimumStrict');\n  const $b = convertToTensor(b, 'b', 'minimumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in minimumStrict: ');\n  return minimum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the max of a and b (`a > b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use maximum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction maximumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'maximumStrict');\n  const $b = convertToTensor(b, 'b', 'maximumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in maximumStrict: ');\n  return maximum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns (a - b) * (a - b) element-wise.\n *\n * Inputs must be the same shape. For broadcasting support, use\n * `tf.squaredDifference` instead.\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n */\nfunction squaredDifferenceStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'squaredDifferenceStrict');\n  const $b = convertToTensor(b, 'b', 'squaredDifferenceStrict');\n  util.assertShapesMatch(\n      $a.shape, $b.shape, 'Error in squaredDifferenceStrict: ');\n  return squaredDifference($a, $b);\n}\n\nexport const addStrict = op({addStrict_});\nexport const divStrict = op({divStrict_});\nexport const maximumStrict = op({maximumStrict_});\nexport const minimumStrict = op({minimumStrict_});\nexport const modStrict = op({modStrict_});\nexport const mulStrict = op({mulStrict_});\nexport const powStrict = op({powStrict_});\nexport const squaredDifferenceStrict = op({squaredDifferenceStrict_});\nexport const subStrict = op({subStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {abs} from './abs';\nimport * as axis_util from './axis_util';\nimport {max} from './max';\nimport {min} from './min';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sqrt} from './sqrt';\nimport {square} from './square';\nimport {sum} from './sum';\n\n/**\n * Computes the norm of scalar, vectors, and matrices.\n * This function can compute several different vector norms (the 1-norm, the\n * Euclidean or 2-norm, the inf-norm, and in general the p-norm for p > 0)\n * and matrix norms (Frobenius, 1-norm, and inf-norm).\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.norm().print();  // or tf.norm(x)\n * ```\n *\n * @param x The input array.\n * @param ord Optional. Order of the norm. Supported norm types are\n * following:\n *\n *  | ord        | norm for matrices         | norm for vectors\n *  |------------|---------------------------|---------------------\n *  |'euclidean' |Frobenius norm             |2-norm\n *  |'fro'       |Frobenius norm\t           |\n *  |Infinity    |max(sum(abs(x), axis=1))   |max(abs(x))\n *  |-Infinity   |min(sum(abs(x), axis=1))   |min(abs(x))\n *  |1           |max(sum(abs(x), axis=0))   |sum(abs(x))\n *  |2           |                           |sum(abs(x)^2)^1/2*\n *\n * @param axis Optional. If axis is null (the default), the input is\n * considered a vector and a single vector norm is computed over the entire\n * set of values in the Tensor, i.e. norm(x, ord) is equivalent\n * to norm(x.reshape([-1]), ord). If axis is a integer, the input\n * is considered a batch of vectors, and axis determines the axis in x\n * over which to compute vector norms. If axis is a 2-tuple of integer it is\n * considered a batch of matrices and axis determines the axes in NDArray\n * over which to compute a matrix norm.\n * @param keepDims Optional. If true, the norm have the same dimensionality\n * as the input.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction norm_(\n    x: Tensor|TensorLike, ord: number|'euclidean'|'fro' = 'euclidean',\n    axis: number|number[] = null, keepDims = false): Tensor {\n  x = convertToTensor(x, 'x', 'norm');\n\n  const norm = normImpl(x, ord, axis);\n  let keepDimsShape = norm.shape;\n  if (keepDims) {\n    const axes = parseAxisParam(axis, x.shape);\n    keepDimsShape = axis_util.expandShapeToKeepDim(norm.shape, axes);\n  }\n  return reshape(norm, keepDimsShape);\n}\n\nfunction normImpl(\n    x: Tensor, p: number|string, axis: number|number[] = null): Tensor {\n  if (x.rank === 0) {\n    return abs(x);\n  }\n\n  // consider vector when no axis is specified\n  if (x.rank !== 1 && axis === null) {\n    return normImpl(reshape(x, [-1]), p, axis);\n  }\n\n  // vector\n  if (x.rank === 1 || typeof axis === 'number' ||\n      Array.isArray(axis) && axis.length === 1) {\n    if (p === 1) {\n      return sum(abs(x), axis);\n    }\n    if (p === Infinity) {\n      return max(abs(x), axis);\n    }\n    if (p === -Infinity) {\n      return min(abs(x), axis);\n    }\n    if (p === 'euclidean' || p === 2) {\n      // norm(x, 2) = sum(abs(xi) ^ 2) ^ 1/2\n      return sqrt(sum(pow(abs(x), scalar(2, 'int32')), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  // matrix (assumption axis[0] < axis[1])\n  if (Array.isArray(axis) && axis.length === 2) {\n    if (p === 1) {\n      return max(sum(abs(x), axis[0]), axis[1] - 1);\n    }\n    if (p === Infinity) {\n      return max(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === -Infinity) {\n      return min(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === 'fro' || p === 'euclidean') {\n      // norm(x) = sqrt(sum(pow(x, 2)))\n      return sqrt(sum(square(x), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  throw new Error(`Error in norm: invalid axis: ${axis}`);\n}\n\nexport const norm = op({norm_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor} from '../tensor';\nimport {assertTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {scalar} from './scalar';\nimport {sub} from './sub';\n\n/**\n * Compute the moving average of a variable.\n *\n * Without zeroDebias, the moving average operation is defined by:\n *   `v += delta`\n * where\n *   `delta = (1 - decay) * (x - v)`\n *\n * With zeroDebias (default), the `delta` term is scaled to debias the\n * effect of the (assumed) zero-initialization of `v`.\n *   `delta /= (1 - decay ^ step)`\n *\n * For more details on the zero-debiasing algorithm, see:\n *   https://arxiv.org/abs/1412.6980\n *\n * Note that this function is completely stateless and does not keep track of\n * step count. The step count needs to be maintained by the caller and passed\n * in as `step`.\n *\n * @param v The current moving average value.\n * @param x New input value, must have the same shape and dtype as `v`.\n * @param decay The decay factor. Typical values are 0.95 and 0.99.\n * @param step Step count.\n * @param zeroDebias: Whether zeroDebias is to be performed (default: `true`).\n * @returns The new moving average value.\n *\n * @doc {heading: 'Operations', subheading: 'Moving Average'}\n */\nfunction movingAverage_<T extends Tensor>(\n    v: T|TensorLike, x: T|TensorLike, decay: number|Scalar,\n    step?: number|Scalar, zeroDebias = true): T {\n  const $v = convertToTensor(v, 'v', 'movingAverage');\n  const $x = convertToTensor(x, 'x', 'movingAverage');\n  const $decay = convertToTensor(decay, 'decay', 'movingAverage');\n\n  assertTypesMatch($v, $x);\n  util.assert(\n      util.arraysEqual($v.shape, $x.shape), () => 'Shape mismatch in v and x');\n\n  const one = scalar(1);\n  const oneMinusDecay = sub(one, $decay);\n\n  let update = mul(sub($x, $v), oneMinusDecay);\n  if (zeroDebias) {\n    util.assert(\n        step != null, () => 'When using zeroDebias: true, step is required.');\n    const $step = convertToTensor(step, 'step', 'movingAverage');\n    update = div(update, sub(one, pow($decay, $step)));\n  }\n  return add($v, update);\n}\n\nexport const movingAverage = op({movingAverage_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ScatterNd, ScatterNdAttrs, ScatterNdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as scatter_nd_util from './scatter_nd_util';\n\n/**\n * Creates a new tensor by applying sparse updates to individual\n * values or slices within a zero tensor of the given shape tensor according to\n * indices. This operator is the inverse of the `tf.gatherND` operator which\n * extracts values or slices from a given tensor.\n *\n * ```js\n * const indices = tf.tensor2d([4, 3, 1, 7], [4, 1], 'int32');\n * const updates = tf.tensor1d([9, 10, 11, 12]);\n * const shape = [8];\n * tf.scatterND(indices, updates, shape).print() //[0, 11, 0, 10, 9, 0, 0, 12]\n * ```\n *\n * @param indices The tensor contains the indices into the output tensor.\n * @param updates The tensor contains the value for the indices.\n * @param shape: The shape of the output tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction scatterND_<R extends Rank>(\n    indices: Tensor|TensorLike, updates: Tensor|TensorLike,\n    shape: ShapeMap[R]): Tensor<R> {\n  const $indices = convertToTensor(indices, 'indices', 'scatterND', 'int32');\n  const $updates = convertToTensor(updates, 'updates', 'scatterND');\n  scatter_nd_util.validateInput($updates, $indices, shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.scatterND($indices, $updates, shape);\n  };\n\n  const inputs: ScatterNdInputs = {indices: $indices, updates: $updates};\n  const attrs: ScatterNdAttrs = {shape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             ScatterNd, attrs as {} as NamedAttrMap) as Tensor<R>;\n}\n\nexport const scatterND = op({scatterND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\n\n/**\n * Validate sparseToDense inputs.\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape number[]. Shape of the dense output tensor.\n * @param validateIndices boolean. indice validation is not supported, error\n * will be thrown if it is set.\n */\nexport function validateInput(\n    sparseIndices: Tensor, sparseValues: Tensor, outputShape: number[],\n    defaultValues: Tensor) {\n  if (sparseIndices.dtype !== 'int32') {\n    throw new Error(\n        'tf.sparseToDense() expects the indices to be int32 type,' +\n        ` but the dtype was ${sparseIndices.dtype}.`);\n  }\n  if (sparseIndices.rank > 2) {\n    throw new Error(\n        'sparseIndices should be a scalar, vector, or matrix,' +\n        ` but got shape ${sparseIndices.shape}.`);\n  }\n\n  const numElems = sparseIndices.rank > 0 ? sparseIndices.shape[0] : 1;\n  const numDims = sparseIndices.rank > 1 ? sparseIndices.shape[1] : 1;\n\n  if (outputShape.length !== numDims) {\n    throw new Error(\n        'outputShape has incorrect number of elements:,' +\n        ` ${outputShape.length}, should be: ${numDims}.`);\n  }\n\n  const numValues = sparseValues.size;\n  if (!(sparseValues.rank === 0 ||\n        sparseValues.rank === 1 && numValues === numElems)) {\n    throw new Error(\n        'sparseValues has incorrect shape ' +\n        `${sparseValues.shape}, should be [] or [${numElems}]`);\n  }\n\n  if (sparseValues.dtype !== defaultValues.dtype) {\n    throw new Error('sparseValues.dtype must match defaultValues.dtype');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {SparseToDense, SparseToDenseAttrs, SparseToDenseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport * as sparse_to_dense from '../ops/sparse_to_dense_util';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ScalarLike, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Converts a sparse representation into a dense tensor.\n *\n * Builds an array dense with shape outputShape such that:\n *\n * // If sparseIndices is scalar\n * dense[i] = (i == sparseIndices ? sparseValues : defaultValue)\n *\n * // If sparseIndices is a vector, then for each i\n * dense[sparseIndices[i]] = sparseValues[i]\n *\n * // If sparseIndices is an n by d matrix, then for each i in [0, n)\n * dense[sparseIndices[i][0], ..., sparseIndices[i][d-1]] = sparseValues[i]\n * All other values in dense are set to defaultValue. If sparseValues is a\n * scalar, all sparse indices are set to this single value.\n *\n * If indices are repeated the final value is summed over all values for those\n * indices.\n *\n * ```js\n * const indices = tf.tensor1d([4, 5, 6, 1, 2, 3], 'int32');\n * const values = tf.tensor1d([10, 11, 12, 13, 14, 15], 'float32');\n * const shape = [8];\n * tf.sparseToDense(indices, values, shape).print();\n * ```\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape Shape of the dense output tensor. the type is inferred.\n * @param defaultValue Scalar. Value to set for indices not specified in\n * sparseIndices. Defaults to zero.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction sparseToDense_<R extends Rank>(\n    sparseIndices: Tensor|TensorLike, sparseValues: Tensor|TensorLike,\n    outputShape: ShapeMap[R], defaultValue: Scalar|ScalarLike = 0): Tensor<R> {\n  const $sparseIndices =\n      convertToTensor(sparseIndices, 'sparseIndices', 'sparseToDense', 'int32');\n  const $sparseValues =\n      convertToTensor(sparseValues, 'sparseValues', 'sparseToDense');\n  const $defaultValue = convertToTensor(\n      defaultValue, 'defaultValue', 'sparseToDense', $sparseValues.dtype);\n\n  sparse_to_dense.validateInput(\n      $sparseIndices, $sparseValues, outputShape, $defaultValue);\n\n  const inputs: SparseToDenseInputs = {\n    sparseIndices: $sparseIndices,\n    sparseValues: $sparseValues,\n    defaultValue: $defaultValue\n  };\n\n  const attrs: SparseToDenseAttrs = {outputShape};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.sparseToDense(\n          $sparseIndices, $sparseValues, outputShape, $defaultValue),\n      inputs as {} as NamedTensorMap, null /* grad */, SparseToDense,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const sparseToDense = op({sparseToDense_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherNd, GatherNdInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Gather slices from input tensor into a Tensor with shape specified by\n * `indices`.\n *\n * `indices` is an K-dimensional integer tensor, best thought of as a\n * (K-1)-dimensional tensor of indices into input, where each element defines a\n * slice of input:\n * output[\\\\(i_0, ..., i_{K-2}\\\\)] = input[indices[\\\\(i_0, ..., i_{K-2}\\\\)]]\n *\n * Whereas in `tf.gather`, `indices` defines slices into the first dimension of\n * input, in `tf.gatherND`, `indices` defines slices into the first N dimensions\n * of input, where N = indices.shape[-1].\n *\n * The last dimension of indices can be at most the rank of input:\n * indices.shape[-1] <= input.rank\n *\n * The last dimension of `indices` corresponds to elements\n * (if indices.shape[-1] == input.rank) or slices\n * (if indices.shape[-1] < input.rank) along dimension indices.shape[-1] of\n * input.\n * The output tensor has shape\n * indices.shape[:-1] + input.shape[indices.shape[-1]:]\n *\n * Note that on CPU, if an out of bound index is found, an error is returned. On\n * GPU, if an out of bound index is found, a 0 is stored in the corresponding\n * output value.\n *\n * ```js\n * const indices = tf.tensor2d([0, 1, 1, 0], [2,2], 'int32');\n * const input = tf.tensor2d([9, 10, 11, 12], [2, 2]);\n * tf.gatherND(input, indices).print() // [10, 11]\n * ```\n *\n * @param x The tensor from which to gather values.\n * @param indices Index tensor, must be of type int32.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction gatherND_(x: Tensor|TensorLike, indices: Tensor|TensorLike): Tensor {\n  const $indices = convertToTensor(indices, 'indices', 'gatherND', 'int32');\n  const $x = convertToTensor(x, 'x', 'gatherND');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.gatherND($x, $indices);\n  };\n\n  const inputs: GatherNdInputs = {params: $x, indices: $indices};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */, GatherNd);\n}\n\nexport const gatherND = op({gatherND_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\n/**\n * Normalize noise shape based on provided tensor and noise shape.\n *\n * @param x Tensor.\n * @param noiseShape The shape for the randomly generated keep/drop flags, as\n *   an array of numbers. Optional.\n * @returns Normalized noise shape.\n */\nexport function getNoiseShape(x: Tensor, noiseShape?: number[]): number[] {\n  if (noiseShape == null) {\n    return x.shape.slice();\n  }\n  if (util.arraysEqual(x.shape, noiseShape)) {\n    return noiseShape;\n  }\n  if (x.shape.length === noiseShape.length) {\n    const newDimension: number[] = [];\n    for (let i = 0; i < x.shape.length; i++) {\n      if (noiseShape[i] == null && x.shape[i] != null) {\n        newDimension.push(x.shape[i]);\n      } else {\n        newDimension.push(noiseShape[i]);\n      }\n    }\n    return newDimension;\n  }\n\n  return noiseShape;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {getNoiseShape} from './dropout_util';\nimport {floor} from './floor';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {randomUniform} from './random_uniform';\n\n/**\n * Computes dropout.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 2, 1]);\n * const rate = 0.75;\n * const output = tf.dropout(x, rate);\n * output.print();\n * ```\n *\n * @param x A floating point Tensor or TensorLike.\n * @param rate A float in the range [0, 1). The probability that each element\n *   of x is discarded.\n * @param noiseShape An array of numbers of type int32, representing the\n * shape for randomly generated keep/drop flags. If the noiseShape has null\n * value, it will be automatically replaced with the x's relative dimension\n * size. Optional.\n * @param seed Used to create random seeds. Optional.\n * @returns A Tensor of the same shape of x.\n *\n * @doc {heading: 'Operations', subheading: 'Dropout'}\n */\nfunction dropout_(\n    x: Tensor|TensorLike, rate: number, noiseShape?: number[],\n    seed?: number|string): Tensor {\n  const $x = convertToTensor(x, 'x', 'dropout');\n\n  util.assert(\n      $x.dtype === 'float32',\n      () => `x has to be a floating point tensor since it's going to be ` +\n          `scaled, but got a ${$x.dtype} tensor instead.`);\n  util.assert(\n      rate >= 0 && rate < 1,\n      () => `rate must be a float in the range [0, 1), but got ${rate}.`);\n\n  if (rate === 0) {\n    return x instanceof Tensor ? $x.clone() : $x;\n  }\n\n  const $noiseShape = getNoiseShape($x, noiseShape);\n  const keepProb = 1 - rate;\n  const multiplier = div(\n      floor(add(randomUniform($noiseShape, 0, 1, 'float32', seed), keepProb)),\n      keepProb);\n\n  return mul($x, multiplier);\n}\n\nexport const dropout = op({dropout_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {tensor1d} from './tensor1d';\n\nexport function enclosingPowerOfTwo(value: number) {\n  // Return 2**N for integer N such that 2**N >= value.\n  return Math.floor(Math.pow(2, Math.ceil(Math.log(value) / Math.log(2.0))));\n}\n\nexport function cosineWindow(\n    windowLength: number, a: number, b: number): Tensor1D {\n  const even = 1 - windowLength % 2;\n  const newValues = new Float32Array(windowLength);\n  for (let i = 0; i < windowLength; ++i) {\n    const cosArg = (2.0 * Math.PI * i) / (windowLength + even - 1);\n    newValues[i] = a - b * Math.cos(cosArg);\n  }\n  return tensor1d(newValues, 'float32');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch, getTypedArrayFromDType} from '../util';\nimport {tensor} from './tensor';\n\n/**\n * Returns whether the targets are in the top K predictions.\n *\n * ```js\n * const predictions = tf.tensor2d([[20, 10, 40, 30], [30, 50, -20, 10]]);\n * const targets = tf.tensor1d([2, 0]);\n * const precision = await tf.inTopKAsync(predictions, targets);\n * precision.print();\n * ```\n * @param predictions 2-D or higher `tf.Tensor` with last dimension being\n *     at least `k`.\n * @param targets 1-D or higher `tf.Tensor`.\n * @param k Optional Number of top elements to look at for computing precision,\n *     default to 1.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nasync function inTopKAsync_<T extends Tensor, U extends Tensor>(\n    predictions: T|TensorLike, targets: U|TensorLike, k = 1): Promise<U> {\n  const $predictions = convertToTensor(predictions, 'predictions', 'inTopK');\n  const $targets = convertToTensor(targets, 'targets', 'inTopK');\n\n  assert(\n      $predictions.rank > 1,\n      () => 'inTopK() expects the predictions to be of rank 2 or higher, ' +\n          `but got ${$predictions.rank}`);\n  assert(\n      $predictions.rank - 1 === $targets.rank,\n      () => `predictions rank should be 1 larger than ` +\n          `targets rank, but got predictions rank ` +\n          `${$predictions.rank} and targets rank ${$targets.rank}`);\n  assertShapesMatch(\n      $predictions.shape.slice(0, $predictions.shape.length - 1),\n      $targets.shape,\n      `predictions's shape should be align with the targets' shape, ` +\n          'except the last dimension.');\n  const lastDim = $predictions.shape[$predictions.shape.length - 1];\n  assert(\n      k > 0 && k <= lastDim,\n      () => `'k' passed to inTopK() must be > 0 && <= the predictions last ` +\n          `dimension (${lastDim}), but got ${k}`);\n\n  const predictionsVals = await $predictions.data();\n  const targetsVals = await $targets.data();\n\n  // Reshape predictionsVals into a 2d tensor [batch, lastDim]\n  // and look up topK along lastDim.\n  const [batch, size] = [predictionsVals.length / lastDim, lastDim];\n  const precision = getTypedArrayFromDType('bool', batch);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = predictionsVals.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    precision[b] = 0;\n    for (let i = 0; i < k; i++) {\n      if (valAndInd[i].index === targetsVals[b]) {\n        precision[b] = 1;\n        break;\n      }\n    }\n  }\n\n  if (predictions !== $predictions) {\n    $predictions.dispose();\n  }\n  if (targets !== $targets) {\n    $targets.dispose();\n  }\n\n  // Output precision has the same shape as targets.\n  return tensor(precision, $targets.shape, 'bool') as U;\n}\n\nexport const inTopKAsync = inTopKAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropFilter, Conv2DBackpropFilterAttrs, Conv2DBackpropFilterInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 2D convolution.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     [batch, height, width, inChannels]. If rank 3, batch of 1 is assumed.\n * @param dy The dy image, of rank 4 or rank 3, of shape\n *     [batch, height, width, outDepth]. If rank 3, batch of 1 is assumed.\n * @param filterShape The shape of the filter, length 4,\n *     [filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction conv2DBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2dDerFilter: input must be rank 4, but got shape ` +\n          `${x4D.shape}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerFilter: dy must be rank 4, but got shape ` +\n          `${dy4D.shape}.`);\n  util.assert(\n      filterShape.length === 4,\n      () => `Error in conv2dDerFilter: filterShape must be length 4, but got ` +\n          `${filterShape}.`);\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filterShape[2],\n      () => `Error in conv2dDerFilter: depth of input ${inDepth}) must ` +\n          `match input depth in filter (${filterShape[2]}.`);\n  util.assert(\n      outDepth === filterShape[3],\n      () => `Error in conv2dDerFilter: depth of dy (${outDepth}) must ` +\n          `match output depth for filter (${filterShape[3]}).`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerFilter: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, filterShape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    return backend.conv2dDerFilter(x4D, dy4D, convInfo);\n  };\n\n  const inputs: Conv2DBackpropFilterInputs = {x: x4D, dy: dy4D};\n  const attrs:\n      Conv2DBackpropFilterAttrs = {strides, pad, dataFormat, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv2DBackpropFilter, attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const conv2DBackpropFilter = op({conv2DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\nimport * as broadcast_util from './broadcast_util';\nimport {elu} from './elu';\nimport {Activation} from './fused_types';\nimport {mul} from './mul';\nimport {prelu} from './prelu';\nimport {relu} from './relu';\nimport {relu6} from './relu6';\nimport {reshape} from './reshape';\nimport {step} from './step';\nimport {sum} from './sum';\n\n// Returns gradient for fused activation.\nexport function getFusedDyActivation(\n    dy: Tensor, y: Tensor, activation: Activation): Tensor {\n  if (activation == null || activation === 'linear') {\n    return dy;\n  }\n  if (activation === 'relu') {\n    return mul(dy, step(y));\n  }\n  throw new Error(\n      `Cannot compute gradient for fused activation ${activation}.`);\n}\n\n// Returns gradient for fused bias.\nexport function getFusedBiasGradient(\n    bias: Tensor, dyActivation: Tensor): Tensor {\n  let res = dyActivation;\n  const reduceAxes =\n      broadcast_util.getReductionAxes(bias.shape, dyActivation.shape);\n  if (reduceAxes.length > 0) {\n    res = sum(res, reduceAxes);\n  }\n  return reshape(res, bias.shape);\n}\n\nexport function applyActivation(\n    x: Tensor, activation: Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return x;\n  } else if (activation === 'relu') {\n    return relu(x);\n  } else if (activation === 'elu') {\n    return elu(x);\n  } else if (activation === 'relu6') {\n    return relu6(x);\n  } else if (activation === 'prelu') {\n    return prelu(x, preluActivationWeights);\n  }\n  throw new Error(`Unknown fused activation ${activation}.`);\n}\n\n// Whether we should call fused ops.\nexport const shouldFuse = (gradientDepth: number, activation: Activation) => {\n  const gradientMode = gradientDepth > 0;\n  return !gradientMode || activation === 'linear';\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedConv2D, FusedConv2DAttrs, FusedConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {conv2d as unfusedConv2d} from '../conv2d';\nimport {conv2DBackpropFilter} from '../conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../conv2d_backprop_input';\nimport * as conv_util from '../conv_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes a 2D convolution over the input x, optionally fused with adding a\n * bias and applying an activation.\n *\n * ```js\n * const inputDepth = 2;\n * const inShape = [2, 2, 2, inputDepth];\n * const outputDepth = 2;\n * const fSize = 1;\n * const pad = 0;\n * const strides = 1;\n *\n * const x = tf.tensor4d( [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,\n * 16], inShape);\n * const w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth,\n * outputDepth]);\n *\n * tf.fused.conv2d({ x, filter: w, strides, pad, dataFormat: 'NHWC',\n * dilations: [1, 1], bias: tf.scalar(5), activation: 'relu' }).print();\n * ```\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid` output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`) to be\n *     applied\n *      after biasAdd.\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  activation = activation || 'linear';\n\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused conv2d: input must be rank 4, but got rank ` +\n          `${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${x4D.shape[3]}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in conv2d: got dataFormat of ${\n          dataFormat} but only NHWC is currently supported.`);\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused conv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    const [$filter, x4D, y, $bias] =\n        saved as [Tensor4D, Tensor4D, Tensor4D, Tensor];\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused conv2D: ' +\n            `dilation rates greater than 1 ` +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const xDer =\n        conv2DBackpropInput(x4D.shape, dyActivation, $filter, strides, pad);\n    const filterDer =\n        conv2DBackpropFilter(x4D, dyActivation, $filter.shape, strides, pad);\n    const der: Tensor[] = [xDer, filterDer];\n\n    if ($bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      der.push(biasDer);\n    }\n    return der;\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedConv2d({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n\n  const attrs: FusedConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const conv2d = op({fusedConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropFilterInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    convInfo: conv_util.Conv2DInfo): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerFilter(x4D, dy4D, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropFilterInputs = {x: x4D, dy: dy4D};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             DepthwiseConv2dNativeBackpropFilter) as Tensor4D;\n}\n\nexport const depthwiseConv2dNativeBackpropFilter =\n    op({depthwiseConv2dNativeBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropInput, DepthwiseConv2dNativeBackpropInputInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, convInfo: conv_util.Conv2DInfo): T {\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerInput(dy4D, filter, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropInputInputs = {dy: dy4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null,\n      DepthwiseConv2dNativeBackpropInput);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2dNativeBackpropInput =\n    op({depthwiseConv2dNativeBackpropInput_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedDepthwiseConv2D, FusedDepthwiseConv2DAttrs, FusedDepthwiseConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport * as conv_util from '../conv_util';\nimport {depthwiseConv2d as unfusedDepthwiseConv2d} from '../depthwise_conv2d';\nimport {depthwiseConv2dNativeBackpropFilter} from '../depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../depthwise_conv2d_native_backprop_input';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes depthwise 2D convolution, optionally fused with adding a\n * bias and applying an activation.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`).\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedDepthwiseConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedDepthwiseConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused depthwiseConv2d: filter must be rank 4, ` +\n          `but got rank ${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in fused depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () =>\n          'Error in fused depthwiseConv2d: Either strides or dilations must ' +\n          `be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused depthwiseConv2d: pad must be an integer when ` +\n            `using dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n      true /* depthwise */);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused depthwiseConv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused depthwiseConv2d: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${dilations}'`);\n    const [$filter, x4D, y, bias] = saved;\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    const xDer = depthwiseConv2dNativeBackpropInput(\n        (x4D as Tensor4D).shape, dyActivation, $filter as Tensor4D, convInfo);\n    const filterDer = depthwiseConv2dNativeBackpropFilter(\n        x4D as Tensor4D, dyActivation, ($filter as Tensor4D).shape, convInfo);\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [xDer, filterDer, biasDer];\n    }\n    return [xDer, filterDer];\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedDepthwiseConv2D({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedDepthwiseConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: FusedDepthwiseConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const depthwiseConv2d = op({fusedDepthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {_FusedMatMul, _FusedMatMulAttrs, _FusedMatMulInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {matMul as unfusedMatMul} from '../mat_mul';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes the dot product of two matrices with optional activation and bias.\n *\n * ```js\n * const a = tf.tensor2d([-1, -2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const bias = tf.tensor2d([1, 2], [1, 2]);\n *\n * tf.fused.matMul({a, b, bias, activation: 'relu'}).print();\n * ```\n *\n * @param obj An object with the following properties:\n * - `a` First matrix in dot product operation.\n * - `b` Second matrix in dot product operation.\n * - `transposeA` If true, `a` is transposed before multiplication.\n * - `transposeB` If true, `b` is transposed before multiplication.\n * - `bias` Matrix to be added to the result.\n * - `activation` Name of activation kernel (defaults to `linear`).\n * - `preluActivationWeights` Tensor of prelu weights.\n */\nfunction fusedMatMul_<T extends Tensor>({\n  a,\n  b,\n  transposeA = false,\n  transposeB = false,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  a: T|TensorLike,\n  b: T|TensorLike,\n  transposeA?: boolean,\n  transposeB?: boolean,\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedMatMul(a, b, transposeA, transposeB);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  let $a = convertToTensor(a, 'a', 'fused matMul');\n  let $b = convertToTensor(b, 'b', 'fused matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () =>\n          `Error in fused matMul: inputs must have the same rank of at least ` +\n          `2, got ranks ${$a.rank} and ${$b.rank}.`);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in fused matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in fused matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D: Tensor3D = transposeA ?\n      reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n      reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D: Tensor3D = transposeB ?\n      reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n      reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused matMul');\n    [$bias] = makeTypesMatch($bias, $a);\n\n    broadcast_util.assertAndGetBroadcastShape(outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused matMul');\n  }\n\n  const grad = (dy: Tensor3D, saved: Tensor[]) => {\n    const [a3D, b3D, y, $bias] = saved;\n    // we reshape dy because the result of the forward is not\n    // necessarily going to be a 3d tensor due to a reshape done at the end of\n    // the customOp.\n    const dyActivation =\n        getFusedDyActivation(reshape(dy, y.shape), y, activation);\n    let aDer: Tensor;\n    let bDer: Tensor;\n\n    if (!transposeA && !transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, true, false);\n    } else if (!transposeA && transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, false);\n      bDer = unfusedMatMul(dyActivation, a3D, true, false);\n    } else if (transposeA && !transposeB) {\n      aDer = unfusedMatMul(b3D, dyActivation, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, false, false);\n    } else {\n      aDer = unfusedMatMul(b3D, dyActivation, true, true);\n      bDer = unfusedMatMul(dyActivation, a3D, true, true);\n    }\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [aDer, bDer, biasDer];\n    } else {\n      return [aDer, bDer];\n    }\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const y = backend.fusedBatchMatMul({\n      a: a3D,\n      b: b3D,\n      transposeA,\n      transposeB,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return y;\n  };\n\n  const inputs: _FusedMatMulInputs = {\n    a: a3D,\n    b: b3D,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: _FusedMatMulAttrs = {transposeA, transposeB, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((a3D: Tensor3D, b3D: Tensor3D, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n    return customOp(a3D, b3D) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (a3D: Tensor3D, b3D: Tensor3D, $bias: Tensor, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res, $bias]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n\n    return customOpWithBias(a3D, b3D, $bias) as T;\n  }\n}\n\nexport const matMul = op({fusedMatMul_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {conv2d} from './fused/conv2d';\nimport {depthwiseConv2d} from './fused/depthwise_conv2d';\nimport {matMul} from './fused/mat_mul';\nimport {Activation} from './fused_types';\n\nexport {Activation, conv2d, depthwiseConv2d, matMul};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a hamming window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hammingWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hammingWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.54, 0.46);\n}\nexport const hammingWindow = op({hammingWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a Hann window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hannWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hannWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.5, 0.5);\n}\n\nexport const hannWindow = op({hannWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {fill} from '../fill';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {tensor2d} from '../tensor2d';\n\n/**\n * Expands input into frames of frameLength.\n * Slides a window size with frameStep.\n *\n * ```js\n * tf.signal.frame([1, 2, 3], 2, 1).print();\n * ```\n * @param signal The input tensor to be expanded\n * @param frameLength Length of each frame\n * @param frameStep The frame hop size in samples.\n * @param padEnd Whether to pad the end of signal with padValue.\n * @param padValue An number to use where the input signal does\n *     not exist when padEnd is True.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction frame_(\n    signal: Tensor1D, frameLength: number, frameStep: number, padEnd = false,\n    padValue = 0): Tensor {\n  let start = 0;\n  const output: Tensor[] = [];\n  while (start + frameLength <= signal.size) {\n    output.push(slice(signal, start, frameLength));\n    start += frameStep;\n  }\n\n  if (padEnd) {\n    while (start < signal.size) {\n      const padLen = (start + frameLength) - signal.size;\n      const pad = concat([\n        slice(signal, start, frameLength - padLen), fill([padLen], padValue)\n      ]);\n      output.push(pad);\n      start += frameStep;\n    }\n  }\n\n  if (output.length === 0) {\n    return tensor2d([], [0, frameLength]);\n  }\n\n  return reshape(concat(output), [output.length, frameLength]);\n}\nexport const frame = op({frame_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {enclosingPowerOfTwo} from '../signal_ops_util';\nimport {slice} from '../slice';\nimport {rfft} from '../spectral/rfft';\n\nimport {frame} from './frame';\nimport {hannWindow} from './hann_window';\n\n/**\n * Computes the Short-time Fourier Transform of signals\n * See: https://en.wikipedia.org/wiki/Short-time_Fourier_transform\n *\n * ```js\n * const input = tf.tensor1d([1, 1, 1, 1, 1])\n * tf.signal.stft(input, 3, 1).print();\n * ```\n * @param signal 1-dimensional real value tensor.\n * @param frameLength The window length of samples.\n * @param frameStep The number of samples to step.\n * @param fftLength The size of the FFT to apply.\n * @param windowFn A callable that takes a window length and returns 1-d tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction stft_(\n    signal: Tensor1D, frameLength: number, frameStep: number,\n    fftLength?: number,\n    windowFn: (length: number) => Tensor1D = hannWindow): Tensor {\n  if (fftLength == null) {\n    fftLength = enclosingPowerOfTwo(frameLength);\n  }\n  const framedSignal = frame(signal, frameLength, frameStep);\n  const windowedSignal = mul(framedSignal, windowFn(frameLength));\n  const output: Tensor[] = [];\n  for (let i = 0; i < framedSignal.shape[0]; i++) {\n    output.push(\n        rfft(slice(windowedSignal, [i, 0], [1, frameLength]), fftLength));\n  }\n  return concat(output);\n}\nexport const stft = op({stft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {CropAndResize, CropAndResizeAttrs, CropAndResizeInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor1D, Tensor2D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Extracts crops from the input image tensor and resizes them using bilinear\n * sampling or nearest neighbor sampling (possibly with aspect ratio change)\n * to a common output size specified by crop_size.\n *\n * @param image 4d tensor of shape `[batch,imageHeight,imageWidth, depth]`,\n *     where imageHeight and imageWidth must be positive, specifying the\n *     batch of images from which to take crops\n * @param boxes 2d float32 tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the normalized\n *     coordinates of the box in the boxInd[i]'th image in the batch\n * @param boxInd 1d int32 tensor of shape `[numBoxes]` with values in range\n *     `[0, batch)` that specifies the image that the `i`-th box refers to.\n * @param cropSize 1d int32 tensor of 2 elements `[cropHeigh, cropWidth]`\n *     specifying the size to which all crops are resized to.\n * @param method Optional string from `'bilinear' | 'nearest'`,\n *     defaults to bilinear, which specifies the sampling method for resizing\n * @param extrapolationValue A threshold for deciding when to remove boxes based\n *     on score. Defaults to 0.\n * @return A 4D tensor of the shape `[numBoxes,cropHeight,cropWidth,depth]`\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction cropAndResize_(\n    image: Tensor4D|TensorLike,\n    boxes: Tensor2D|TensorLike,\n    boxInd: Tensor1D|TensorLike,\n    cropSize: [number, number],\n    method?: 'bilinear'|'nearest',\n    extrapolationValue?: number,\n    ): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'cropAndResize');\n  const $boxes = convertToTensor(boxes, 'boxes', 'cropAndResize', 'float32');\n  const $boxInd = convertToTensor(boxInd, 'boxInd', 'cropAndResize', 'int32');\n  method = method || 'bilinear';\n  extrapolationValue = extrapolationValue || 0;\n\n  const numBoxes = $boxes.shape[0];\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in cropAndResize: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n  util.assert(\n      $boxes.rank === 2 && $boxes.shape[1] === 4,\n      () => `Error in cropAndResize: boxes must be have size [${numBoxes},4] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      $boxInd.rank === 1 && $boxInd.shape[0] === numBoxes,\n      () => `Error in cropAndResize: boxInd must be have size [${numBoxes}] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      cropSize.length === 2,\n      () => `Error in cropAndResize: cropSize must be of length 2, but got ` +\n          `length ${cropSize.length}.`);\n  util.assert(\n      cropSize[0] >= 1 && cropSize[1] >= 1,\n      () => `cropSize must be atleast [1,1], but was ${cropSize}`);\n  util.assert(\n      method === 'bilinear' || method === 'nearest',\n      () => `method must be bilinear or nearest, but was ${method}`);\n\n  const forward: ForwardFunc<Tensor4D> = (backend) => backend.cropAndResize(\n      $image, $boxes, $boxInd, cropSize, method, extrapolationValue);\n\n  const inputs:\n      CropAndResizeInputs = {image: $image, boxes: $boxes, boxInd: $boxInd};\n  const attrs: CropAndResizeAttrs = {method, extrapolationValue, cropSize};\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, CropAndResize,\n      attrs as {} as NamedAttrMap);\n  return res;\n}\n\nexport const cropAndResize = op({cropAndResize_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FlipLeftRight, FlipLeftRightInputs} from '../../kernel_names';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {op} from '../operation';\n\n/**\n * Flips the image left to right. Currently available in the CPU, WebGL, and\n * WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n */\n/** @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'} */\nfunction flipLeftRight_(image: Tensor4D|TensorLike): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'flipLeftRight', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in flipLeftRight: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: FlipLeftRightInputs = {image: $image};\n  const res =\n      ENGINE.runKernel(FlipLeftRight, inputs as {} as NamedTensorMap, {});\n  return res as Tensor4D;\n}\n\nexport const flipLeftRight = op({flipLeftRight_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Rotates the input image tensor counter-clockwise with an optional offset\n * center of rotation. Currently available in the CPU, WebGL, and WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n * @param radians The amount of rotation.\n * @param fillValue The value to fill in the empty space leftover\n *     after rotation. Can be either a single grayscale value (0-255), or an\n *     array of three numbers `[red, green, blue]` specifying the red, green,\n *     and blue channels. Defaults to `0` (black).\n * @param center The center of rotation. Can be either a single value (0-1), or\n *     an array of two numbers `[centerX, centerY]`. Defaults to `0.5` (rotates\n *     the image around its center).\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction rotateWithOffset_(\n    image: Tensor4D|TensorLike, radians: number,\n    fillValue: number|[number, number, number] = 0,\n    center: number|[number, number] = 0.5): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'rotateWithOffset', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in rotateWithOffset: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: RotateWithOffsetInputs = {image: $image};\n  const attrs: RotateWithOffsetAttrs = {radians, fillValue, center};\n  const res = ENGINE.runKernel(\n      RotateWithOffset, inputs as {} as NamedTensorMap,\n      attrs as {} as NamedAttrMap);\n  return res as Tensor4D;\n}\n\nexport const rotateWithOffset = op({rotateWithOffset_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport * as util from '../util';\n\nfunction nonMaxSuppSanityCheck(\n    boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma?: number): {\n  maxOutputSize: number,\n  iouThreshold: number,\n  scoreThreshold: number,\n  softNmsSigma: number\n} {\n  if (iouThreshold == null) {\n    iouThreshold = 0.5;\n  }\n  if (scoreThreshold == null) {\n    scoreThreshold = Number.NEGATIVE_INFINITY;\n  }\n  if (softNmsSigma == null) {\n    softNmsSigma = 0.0;\n  }\n\n  const numBoxes = boxes.shape[0];\n  maxOutputSize = Math.min(maxOutputSize, numBoxes);\n\n  util.assert(\n      0 <= iouThreshold && iouThreshold <= 1,\n      () => `iouThreshold must be in [0, 1], but was '${iouThreshold}'`);\n  util.assert(\n      boxes.rank === 2,\n      () => `boxes must be a 2D tensor, but was of rank '${boxes.rank}'`);\n  util.assert(\n      boxes.shape[1] === 4,\n      () =>\n          `boxes must have 4 columns, but 2nd dimension was ${boxes.shape[1]}`);\n  util.assert(scores.rank === 1, () => 'scores must be a 1D tensor');\n  util.assert(\n      scores.shape[0] === numBoxes,\n      () => `scores has incompatible shape with boxes. Expected ${numBoxes}, ` +\n          `but was ${scores.shape[0]}`);\n  util.assert(\n      0 <= softNmsSigma && softNmsSigma <= 1,\n      () => `softNmsSigma must be in [0, 1], but was '${softNmsSigma}'`);\n  return {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n}\n\nexport {nonMaxSuppSanityCheck};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV3} from '../../kernel_names';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\nfunction nonMaxSuppression_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Tensor1D {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const attrs = {maxOutputSize, iouThreshold, scoreThreshold};\n  return ENGINE.runKernelFunc(\n      b => b.nonMaxSuppression(\n          $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold),\n      {boxes: $boxes, scores: $scores}, null /* grad */, NonMaxSuppressionV3,\n      attrs);\n}\n\nexport const nonMaxSuppression = op({nonMaxSuppression_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inserts a value into a sorted array. This method allows duplicate, meaning it\n * allows inserting duplicate value, in which case, the element will be inserted\n * at the lowest index of the value.\n * @param arr The array to modify.\n * @param element The element to insert.\n * @param comparator Optional. If no comparator is specified, elements are\n * compared using array_util.defaultComparator, which is suitable for Strings\n * and Numbers in ascending arrays. If the array contains multiple instances of\n * the target value, the left-most instance will be returned. To provide a\n * comparator, it should take 2 arguments to compare and return a negative,\n * zero, or a positive number.\n */\nexport function binaryInsert<T>(\n    arr: T[], element: T, comparator?: (a: T, b: T) => number) {\n  const index = binarySearch(arr, element, comparator);\n  const insertionPoint = index < 0 ? -(index + 1) : index;\n  arr.splice(insertionPoint, 0, element);\n}\n\n/**\n * Searches the array for the target using binary search, returns the index\n * of the found element, or position to insert if element not found. If no\n * comparator is specified, elements are compared using array_\n * util.defaultComparator, which is suitable for Strings and Numbers in\n * ascending arrays. If the array contains multiple instances of the target\n * value, the left-most instance will be returned.\n * @param arr The array to be searched in.\n * @param target The target to be searched for.\n * @param comparator Should take 2 arguments to compare and return a negative,\n *    zero, or a positive number.\n * @return Lowest index of the target value if found, otherwise the insertion\n *    point where the target should be inserted, in the form of\n *    (-insertionPoint - 1).\n */\nexport function binarySearch<T>(\n    arr: T[], target: T, comparator?: (a: T, b: T) => number) {\n  return binarySearch_(arr, target, comparator || defaultComparator);\n}\n\n/**\n * Compares its two arguments for order.\n * @param a The first element to be compared.\n * @param b The second element to be compared.\n * @return A negative number, zero, or a positive number as the first\n *     argument is less than, equal to, or greater than the second.\n */\nfunction defaultComparator<T>(a: T, b: T): number {\n  return a > b ? 1 : a < b ? -1 : 0;\n}\n\nfunction binarySearch_<T>(\n    arr: T[], target: T, comparator: (a: T, b: T) => number) {\n  let left = 0;\n  let right = arr.length;\n  let middle = 0;\n  let found = false;\n  while (left < right) {\n    middle = left + ((right - left) >>> 1);\n    const compareResult = comparator(target, arr[middle]);\n    if (compareResult > 0) {\n      left = middle + 1;\n    } else {\n      right = middle;\n      // If compareResult is 0, the value is found. We record it is found,\n      // and then keep looking because there may be duplicate.\n      found = !compareResult;\n    }\n  }\n\n  return found ? left : -left - 1;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Implementation of the NonMaxSuppression kernel shared between webgl and cpu.\n */\n\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\n\nimport {binaryInsert} from './array_util';\n\ninterface Candidate {\n  score: number;\n  boxIndex: number;\n  suppressBeginIndex: number;\n}\n\nexport function nonMaxSuppressionV3Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number): Tensor1D {\n  return nonMaxSuppressionImpl_(\n             boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n             0 /* softNmsSigma */)\n             .selectedIndices as Tensor1D;\n}\n\nexport function nonMaxSuppressionV4Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    padToMaxOutputSize: boolean): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n      0 /* softNmsSigma */, false /* returnScoresTensor */,\n      padToMaxOutputSize /* padToMaxOutputSize */, true\n      /* returnValidOutputs */);\n}\n\nexport function nonMaxSuppressionV5Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    softNmsSigma: number): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma,\n      true /* returnScoresTensor */);\n}\n\nfunction nonMaxSuppressionImpl_(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma: number,\n    returnScoresTensor = false, padToMaxOutputSize = false,\n    returnValidOutputs = false): NamedTensorMap {\n  // The list is sorted in ascending order, so that we can always pop the\n  // candidate with the largest score in O(1) time.\n  const candidates = [];\n\n  for (let i = 0; i < scores.length; i++) {\n    if (scores[i] > scoreThreshold) {\n      candidates.push({score: scores[i], boxIndex: i, suppressBeginIndex: 0});\n    }\n  }\n\n  candidates.sort(ascendingComparator);\n\n  // If softNmsSigma is 0, the outcome of this algorithm is exactly same as\n  // before.\n  const scale = softNmsSigma > 0 ? (-0.5 / softNmsSigma) : 0.0;\n\n  const selectedIndices: number[] = [];\n  const selectedScores: number[] = [];\n\n  while (selectedIndices.length < maxOutputSize && candidates.length > 0) {\n    const candidate = candidates.pop();\n    const {score: originalScore, boxIndex, suppressBeginIndex} = candidate;\n\n    if (originalScore < scoreThreshold) {\n      break;\n    }\n\n    // Overlapping boxes are likely to have similar scores, therefore we\n    // iterate through the previously selected boxes backwards in order to\n    // see if candidate's score should be suppressed. We use\n    // suppressBeginIndex to track and ensure a candidate can be suppressed\n    // by a selected box no more than once. Also, if the overlap exceeds\n    // iouThreshold, we simply ignore the candidate.\n    let ignoreCandidate = false;\n    for (let j = selectedIndices.length - 1; j >= suppressBeginIndex; --j) {\n      const iou = intersectionOverUnion(boxes, boxIndex, selectedIndices[j]);\n\n      if (iou >= iouThreshold) {\n        ignoreCandidate = true;\n        break;\n      }\n\n      candidate.score =\n          candidate.score * suppressWeight(iouThreshold, scale, iou);\n\n      if (candidate.score <= scoreThreshold) {\n        break;\n      }\n    }\n\n    // At this point, if `candidate.score` has not dropped below\n    // `scoreThreshold`, then we know that we went through all of the\n    // previous selections and can safely update `suppressBeginIndex` to the\n    // end of the selected array. Then we can re-insert the candidate with\n    // the updated score and suppressBeginIndex back in the candidate list.\n    // If on the other hand, `candidate.score` has dropped below the score\n    // threshold, we will not add it back to the candidates list.\n    candidate.suppressBeginIndex = selectedIndices.length;\n\n    if (!ignoreCandidate) {\n      // Candidate has passed all the tests, and is not suppressed, so\n      // select the candidate.\n      if (candidate.score === originalScore) {\n        selectedIndices.push(boxIndex);\n        selectedScores.push(candidate.score);\n      } else if (candidate.score > scoreThreshold) {\n        // Candidate's score is suppressed but is still high enough to be\n        // considered, so add back to the candidates list.\n        binaryInsert(candidates, candidate, ascendingComparator);\n      }\n    }\n  }\n\n  // NonMaxSuppressionV4 feature: padding output to maxOutputSize.\n  const validOutputs = selectedIndices.length;\n  const elemsToPad = maxOutputSize - validOutputs;\n\n  if (padToMaxOutputSize && elemsToPad > 0) {\n    selectedIndices.push(...new Array(elemsToPad).fill(0));\n    selectedScores.push(...new Array(elemsToPad).fill(0.0));\n  }\n\n  const result:\n      NamedTensorMap = {selectedIndices: tensor1d(selectedIndices, 'int32')};\n\n  if (returnScoresTensor) {\n    result['selectedScores'] = tensor1d(selectedScores, 'float32');\n  }\n\n  if (returnValidOutputs) {\n    result['validOutputs'] = scalar(validOutputs, 'int32');\n  }\n\n  return result;\n}\n\nfunction intersectionOverUnion(boxes: TypedArray, i: number, j: number) {\n  const iCoord = boxes.subarray(i * 4, i * 4 + 4);\n  const jCoord = boxes.subarray(j * 4, j * 4 + 4);\n  const yminI = Math.min(iCoord[0], iCoord[2]);\n  const xminI = Math.min(iCoord[1], iCoord[3]);\n  const ymaxI = Math.max(iCoord[0], iCoord[2]);\n  const xmaxI = Math.max(iCoord[1], iCoord[3]);\n  const yminJ = Math.min(jCoord[0], jCoord[2]);\n  const xminJ = Math.min(jCoord[1], jCoord[3]);\n  const ymaxJ = Math.max(jCoord[0], jCoord[2]);\n  const xmaxJ = Math.max(jCoord[1], jCoord[3]);\n  const areaI = (ymaxI - yminI) * (xmaxI - xminI);\n  const areaJ = (ymaxJ - yminJ) * (xmaxJ - xminJ);\n  if (areaI <= 0 || areaJ <= 0) {\n    return 0.0;\n  }\n  const intersectionYmin = Math.max(yminI, yminJ);\n  const intersectionXmin = Math.max(xminI, xminJ);\n  const intersectionYmax = Math.min(ymaxI, ymaxJ);\n  const intersectionXmax = Math.min(xmaxI, xmaxJ);\n  const intersectionArea = Math.max(intersectionYmax - intersectionYmin, 0.0) *\n      Math.max(intersectionXmax - intersectionXmin, 0.0);\n  return intersectionArea / (areaI + areaJ - intersectionArea);\n}\n\n// A Gaussian penalty function, this method always returns values in [0, 1].\n// The weight is a function of similarity, the more overlap two boxes are, the\n// smaller the weight is, meaning highly overlapping boxe will be significantly\n// penalized. On the other hand, a non-overlapping box will not be penalized.\nfunction suppressWeight(iouThreshold: number, scale: number, iou: number) {\n  const weight = Math.exp(scale * iou * iou);\n  return iou <= iouThreshold ? weight : 0.0;\n}\n\nfunction ascendingComparator(c1: Candidate, c2: Candidate) {\n  // For objects with same scores, we make the object with the larger index go\n  // first. In an array that pops from the end, this means that the object with\n  // the smaller index will be popped first. This ensures the same output as\n  // the TensorFlow python version.\n  return (c1.score - c2.score) ||\n      ((c1.score === c2.score) && (c2.boxIndex - c1.boxIndex));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {nonMaxSuppressionV3Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This is the async version of `nonMaxSuppression`\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @return A 1D tensor with the selected box indices.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Promise<Tensor1D> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV3Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionAsync = nonMaxSuppressionAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionWithScore_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const inputs: NonMaxSuppressionV5Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV5Attrs =\n      {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV5, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], selectedScores: result[1]};\n}\n\nexport const nonMaxSuppressionWithScore = op({nonMaxSuppressionWithScore_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV5Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionWithScoreAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV5Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionWithScoreAsync = nonMaxSuppressionWithScoreAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionPadded_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const inputs: NonMaxSuppressionV4Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV4Attrs = {\n    maxOutputSize: $maxOutputSize,\n    iouThreshold: $iouThreshold,\n    scoreThreshold: $scoreThreshold,\n    padToMaxOutputSize\n  };\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV4, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], validOutputs: result[1]};\n}\n\nexport const nonMaxSuppressionPadded = op({nonMaxSuppressionPadded_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV4Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionPaddedAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const [boxesVals, scoresVals] =\n      await Promise.all([$boxes.data(), $scores.data()]);\n\n  // We call a cpu based impl directly with the typedarray data here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV4Impl(\n      boxesVals, scoresVals, $maxOutputSize, $iouThreshold, $scoreThreshold,\n      padToMaxOutputSize);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionPaddedAsync = nonMaxSuppressionPaddedAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Bilinear resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeBilinear_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeBilinear');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeBilinear: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () => `Error in resizeBilinear: new shape must 2D, but got shape ` +\n          `${size}.`);\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n\n  const [newHeight, newWidth] = size;\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeBilinear(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const inputs: ResizeBilinearInputs = {images: batchImages};\n  const attrs: ResizeBilinearAttrs = {alignCorners, size};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeBilinear, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeBilinear = op({resizeBilinear_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * NearestNeighbor resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeNearestNeighbor_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeNearestNeighbor');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeNearestNeighbor: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () =>\n          `Error in resizeNearestNeighbor: new shape must 2D, but got shape ` +\n          `${size}.`);\n  util.assert(\n      $images.dtype === 'float32' || $images.dtype === 'int32',\n      () => '`images` must have `int32` or `float32` as dtype');\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n  const [newHeight, newWidth] = size;\n\n  const inputs: ResizeNearestNeighborInputs = {images: batchImages};\n  const attrs: ResizeNearestNeighborAttrs = {alignCorners, size};\n\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeNearestNeighbor(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeNearestNeighbor, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeNearestNeighbor = op({resizeNearestNeighbor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assert} from '../../util';\n\nimport {greaterEqual} from '../greater_equal';\nimport {lessEqual} from '../less_equal';\nimport {logicalAnd} from '../logical_and';\nimport {op} from '../operation';\nimport {range} from '../range';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\nimport {zeros} from '../zeros';\n\n/**\n * Copy a tensor setting everything outside a central band in each innermost\n * matrix to zero.\n *\n * The band part is computed as follows: Assume input has `k` dimensions\n * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where\n * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`.\n * The indicator function\n * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))`\n * `&& (num_upper < 0 || (n-m) <= num_upper)`\n *\n * ```js\n * const x = tf.tensor2d([[ 0,  1,  2, 3],\n *                        [-1,  0,  1, 2],\n *                        [-2, -1,  0, 1],\n *                        [-3, -2, -1, 0]]);\n * let y = tf.linalg.bandPart(x, 1, -1);\n * y.print(); // [[ 0,  1,  2, 3],\n *            //  [-1,  0,  1, 2],\n *            //  [ 0, -1,  0, 1],\n *            //  [ 0, 0 , -1, 0]]\n * let z = tf.linalg.bandPart(x, 2, 1);\n * z.print(); // [[ 0,  1,  0, 0],\n *            //  [-1,  0,  1, 0],\n *            //  [-2, -1,  0, 1],\n *            //  [ 0, -2, -1, 0]]\n * ```\n *\n * @param x Rank `k` tensor\n * @param numLower Number of subdiagonals to keep.\n *   If negative, keep entire lower triangle.\n * @param numUpper Number of subdiagonals to keep.\n *   If negative, keep entire upper triangle.\n * @returns Rank `k` tensor of the same shape as input.\n *   The extracted banded tensor.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction bandPart_<T extends Tensor>(\n    a: T|TensorLike, numLower: number, numUpper: number): T {\n  assert(\n      numLower % 1 === 0,\n      () => `bandPart(): numLower must be an integer, got ${numLower}.`);\n  assert(\n      numUpper % 1 === 0,\n      () => `bandPart(): numUpper must be an integer, got ${numUpper}.`);\n\n  const $a = convertToTensor(a, 'a', 'bandPart');\n\n  assert(\n      $a.rank >= 2,\n      () => `bandPart(): Rank must be at least 2, got ${$a.rank}.`);\n\n  const shape = $a.shape;\n  const [M, N] = $a.shape.slice(-2);\n\n  if (!(numLower <= M)) {\n    throw new Error(\n        `bandPart(): numLower (${numLower})` +\n        ` must not be greater than the number of rows (${M}).`);\n  }\n  if (!(numUpper <= N)) {\n    throw new Error(\n        `bandPart(): numUpper (${numUpper})` +\n        ` must not be greater than the number of columns (${N}).`);\n  }\n\n  if (numLower < 0) {\n    numLower = M;\n  }\n  if (numUpper < 0) {\n    numUpper = N;\n  }\n\n  const i = reshape(range(0, M, 1, 'int32'), [-1, 1]);\n  const j = range(0, N, 1, 'int32');\n  const ij = sub(i, j);\n\n  const inBand = logicalAnd(\n      lessEqual(ij, scalar(+numLower, 'int32')),\n      greaterEqual(ij, scalar(-numUpper, 'int32')));\n\n  const zero = zeros([M, N], $a.dtype);\n\n  return reshape(\n             stack(unstack(reshape($a, [-1, M, N]))\n                       .map(mat => where(inBand, mat, zero))),\n             shape) as T;\n}\n\nexport const bandPart = op({bandPart_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {div} from '../div';\nimport {mul} from '../mul';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {split} from '../split';\nimport {squeeze} from '../squeeze';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\n/**\n * Gram-Schmidt orthogonalization.\n *\n * ```js\n * const x = tf.tensor2d([[1, 2], [3, 4]]);\n * let y = tf.linalg.gramSchmidt(x);\n * y.print();\n * console.log('Othogonalized:');\n * y.dot(y.transpose()).print();  // should be nearly the identity matrix.\n * console.log('First row direction maintained:');\n * const data = await y.array();\n * console.log(data[0][1] / data[0][0]);  // should be nearly 2.\n * ```\n *\n * @param xs The vectors to be orthogonalized, in one of the two following\n *   formats:\n *   - An Array of `tf.Tensor1D`.\n *   - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows\n *     of `xs`.\n *   In each case, all the vectors must have the same length and the length\n *   must be greater than or equal to the number of vectors.\n * @returns The orthogonalized and normalized vectors or matrix.\n *   Orthogonalization means that the vectors or the rows of the matrix\n *   are orthogonal (zero inner products). Normalization means that each\n *   vector or each row of the matrix has an L2 norm that equals `1`.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D {\n  let inputIsTensor2D: boolean;\n  if (Array.isArray(xs)) {\n    inputIsTensor2D = false;\n    assert(\n        xs != null && xs.length > 0,\n        () => 'Gram-Schmidt process: input must not be null, undefined, or ' +\n            'empty');\n    const dim = xs[0].shape[0];\n    for (let i = 1; i < xs.length; ++i) {\n      assert(\n          xs[i].shape[0] === dim,\n          () =>\n              'Gram-Schmidt: Non-unique lengths found in the input vectors: ' +\n              `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`);\n    }\n  } else {\n    inputIsTensor2D = true;\n    xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0]));\n  }\n\n  assert(\n      xs.length <= xs[0].shape[0],\n      () => `Gram-Schmidt: Number of vectors (${\n                (xs as Tensor1D[]).length}) exceeds ` +\n          `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`);\n\n  const ys: Tensor1D[] = [];\n  const xs1d = xs;\n  for (let i = 0; i < xs.length; ++i) {\n    ys.push(ENGINE.tidy(() => {\n      let x = xs1d[i];\n      if (i > 0) {\n        for (let j = 0; j < i; ++j) {\n          const proj = mul(sum(mul(ys[j], x)), ys[j]);\n          x = sub(x, proj);\n        }\n      }\n      return div(x, norm(x, 'euclidean'));\n    }));\n  }\n\n  if (inputIsTensor2D) {\n    return stack(ys, 0) as Tensor2D;\n  } else {\n    return ys;\n  }\n}\n\nexport const gramSchmidt = op({gramSchmidt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../../engine';\nimport {dispose} from '../../globals';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {clone} from '../clone';\nimport {concat} from '../concat';\nimport {div} from '../div';\nimport {eye} from '../eye';\nimport {greater} from '../greater';\nimport {matMul} from '../mat_mul';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {tensor2d} from '../tensor2d';\nimport {transpose} from '../transpose';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\n\n/**\n * Compute QR decomposition of m-by-n matrix using Householder transformation.\n *\n * Implementation based on\n *   [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf]\n * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf)\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [3, 4]]);\n * let [q, r] = tf.linalg.qr(a);\n * console.log('Q');\n * q.print();\n * console.log('R');\n * r.print();\n * console.log('Orthogonalized');\n * q.dot(q.transpose()).print()  // should be nearly the identity matrix.\n * console.log('Reconstructed');\n * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]];\n * ```\n *\n * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose\n *   it has the shape `[..., M, N]`.\n * @param fullMatrices An optional boolean parameter. Defaults to `false`.\n *   If `true`, compute full-sized `Q`. If `false` (the default),\n *   compute only the leading N columns of `Q` and `R`.\n * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix,\n *   i.e., its columns all have unit norm and are mutually orthogonal.\n *   If `M >= N`,\n *     If `fullMatrices` is `false` (default),\n *       - `Q` has a shape of `[..., M, N]`,\n *       - `R` has a shape of `[..., N, N]`.\n *     If `fullMatrices` is `true` (default),\n *       - `Q` has a shape of `[..., M, M]`,\n *       - `R` has a shape of `[..., M, N]`.\n *   If `M < N`,\n *     - `Q` has a shape of `[..., M, M]`,\n *     - `R` has a shape of `[..., M, N]`.\n * @throws If the rank of `x` is less than 2.\n *\n * @doc {heading:'Operations',\n *       subheading:'Linear Algebra',\n *       namespace:'linalg'}\n */\nfunction qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] {\n  assert(\n      x.rank >= 2,\n      () => `qr() requires input tensor to have a rank >= 2, but got rank ${\n          x.rank}`);\n\n  if (x.rank === 2) {\n    return qr2d(x as Tensor2D, fullMatrices);\n  } else {\n    // Rank > 2.\n    // TODO(cais): Below we split the input into individual 2D tensors,\n    //   perform QR decomposition on them and then stack the results back\n    //   together. We should explore whether this can be parallelized.\n    const outerDimsProd = x.shape.slice(0, x.shape.length - 2)\n                              .reduce((value, prev) => value * prev);\n    const x2ds = unstack(\n        reshape(\n            x,\n            [\n              outerDimsProd, x.shape[x.shape.length - 2],\n              x.shape[x.shape.length - 1]\n            ]),\n        0);\n    const q2ds: Tensor2D[] = [];\n    const r2ds: Tensor2D[] = [];\n    x2ds.forEach(x2d => {\n      const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices);\n      q2ds.push(q2d);\n      r2ds.push(r2d);\n    });\n    const q = reshape(stack(q2ds, 0), x.shape);\n    const r = reshape(stack(r2ds, 0), x.shape);\n    return [q, r];\n  }\n}\n\nfunction qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] {\n  return ENGINE.tidy(() => {\n    assert(\n        x.shape.length === 2,\n        () => `qr2d() requires a 2D Tensor, but got a ${\n            x.shape.length}D Tensor.`);\n\n    const m = x.shape[0];\n    const n = x.shape[1];\n\n    let q = eye(m);    // Orthogonal transform so far.\n    let r = clone(x);  // Transformed matrix so far.\n\n    const one2D = tensor2d([[1]], [1, 1]);\n    let w: Tensor2D = clone(one2D);\n\n    const iters = m >= n ? n : m;\n    for (let j = 0; j < iters; ++j) {\n      // This tidy within the for-loop ensures we clean up temporary\n      // tensors as soon as they are no longer needed.\n      const rTemp = r;\n      const wTemp = w;\n      const qTemp = q;\n      [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => {\n        // Find H = I - tau * w * w', to put zeros below R(j, j).\n        const rjEnd1 = slice(r, [j, j], [m - j, 1]);\n        const normX = norm(rjEnd1);\n        const rjj = slice(r, [j, j], [1, 1]);\n\n        // The sign() function returns 0 on 0, which causes division by zero.\n        const s = where(greater(rjj, 0), tensor2d([[-1]]), tensor2d([[1]]));\n\n        const u1 = sub(rjj, mul(s, normX));\n        const wPre = div(rjEnd1, u1);\n        if (wPre.shape[0] === 1) {\n          w = clone(one2D);\n        } else {\n          w = concat(\n              [\n                one2D,\n                slice(wPre, [1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as\n                    Tensor2D\n              ],\n              0);\n        }\n        const tau = neg(div(matMul(s, u1), normX)) as Tensor2D;\n\n        // -- R := HR, Q := QH.\n        const rjEndAll = slice(r, [j, 0], [m - j, n]);\n        const tauTimesW: Tensor2D = mul(tau, w);\n        const wT: Tensor2D = transpose(w);\n        if (j === 0) {\n          r = sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n        } else {\n          const rTimesTau: Tensor2D =\n              sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n          r = concat([slice(r, [0, 0], [j, n]), rTimesTau], 0);\n        }\n        const tawTimesWT: Tensor2D = transpose(tauTimesW);\n        const qAllJEnd = slice(q, [0, j], [m, q.shape[1] - j]);\n        if (j === 0) {\n          q = sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n        } else {\n          const qTimesTau: Tensor2D =\n              sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n          q = concat([slice(q, [0, 0], [m, j]), qTimesTau], 1);\n        }\n        return [w, r, q];\n      });\n      dispose([rTemp, wTemp, qTemp]);\n    }\n\n    if (!fullMatrices && m > n) {\n      q = slice(q, [0, 0], [m, n]);\n      r = slice(r, [0, 0], [n, n]);\n    }\n\n    return [q, r];\n  }) as [Tensor2D, Tensor2D];\n}\n\nexport const qr = op({qr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport enum Reduction {\n  NONE,\n  MEAN,\n  SUM,\n  SUM_BY_NONZERO_WEIGHTS\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {Reduction} from '../loss_ops_utils';\nimport {mean} from '../mean';\nimport {mul} from '../mul';\nimport {notEqual} from '../not_equal';\nimport {ones} from '../ones';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sum} from '../sum';\n\n/**\n * Computes the weighted loss between two tensors.\n *\n * @param losses Tensor of shape `[batch_size, d1, ... dN]`.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `losses`, and must be broadcastable to `losses` (i.e., all\n *    dimensions must be either `1`, or the same as the corresponding\n *    `losses` dimension).\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction computeWeightedLoss_<T extends Tensor, O extends Tensor>(\n    losses: T|TensorLike, weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $losses = convertToTensor(losses, 'losses', 'computeWeightedLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'computeWeightedLoss');\n  }\n\n  const weightedLoss = ($weights == null) ? $losses : mul($losses, $weights);\n\n  if (reduction === Reduction.NONE) {\n    return weightedLoss as O;\n  }\n  if (reduction === Reduction.SUM) {\n    return sum(weightedLoss);\n  }\n  if (reduction === Reduction.MEAN) {\n    if ($weights == null) {\n      return mean(weightedLoss);\n    } else {\n      const broadcastFactor = $losses.size / $weights.size;\n      const result = div(sum(weightedLoss), sum($weights));\n      return broadcastFactor > 1 ? div(result, scalar(broadcastFactor)) :\n                                   result as O;\n    }\n  }\n  if (reduction === Reduction.SUM_BY_NONZERO_WEIGHTS) {\n    if ($weights == null) {\n      return div(sum(weightedLoss), scalar($losses.size));\n    } else {\n      const broadcastedWeights = mul($weights, ones($losses.shape));\n\n      const numNonZeros =\n          cast(sum(notEqual(broadcastedWeights, scalar(0))), 'float32');\n      return div(sum(weightedLoss), numNonZeros);\n    }\n  }\n\n  throw Error(`Unknown reduction: ${reduction}`);\n}\nexport const computeWeightedLoss = op({computeWeightedLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the absolute difference loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction absoluteDifference_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'absoluteDifference');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'absoluteDifference');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'absoluteDifference');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in absoluteDifference: ');\n\n  const losses = abs(sub($labels, $predictions));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const absoluteDifference = op({absoluteDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the cosine distance loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param axis The dimension along which the cosine distance is computed.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction cosineDistance_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike, axis: number,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'cosineDistance');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'cosineDistance');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'cosineDistance');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in cosineDistance: ');\n\n  const one = scalar(1);\n  const losses = sub(one, sum(mul($labels, $predictions), axis, true));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const cosineDistance = op({cosineDistance_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the Hinge loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction hingeLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $labels = convertToTensor(labels, 'labels', 'hingeLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'hingeLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'hingeLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in hingeLoss: ');\n\n  const one = scalar(1);\n  // Convert binary labels to (-1, 1)\n  $labels = sub(mul(scalar(2), $labels), one);\n  const losses = relu(sub(one, mul($labels, $predictions)));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const hingeLoss = op({hingeLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {Reduction} from '../loss_ops_utils';\nimport {minimum} from '../minimum';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {square} from '../square';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the huber loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param delta Point where huber loss changes from quadratic to linear.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`.\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction huberLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, delta = 1.0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'huberLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'huberLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'huberLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in huberLoss: ');\n\n  const deltaScalar = scalar(delta);\n  const error = abs(sub($predictions, $labels));\n  const quadratic = minimum(error, deltaScalar);\n  const linear = sub(error, quadratic);\n\n  const losses =\n      add(mul(scalar(0.5), square(quadratic)), mul(deltaScalar, linear));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const huberLoss = op({huberLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {log} from '../log';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the log loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param epsilon A small increment to avoid taking log of zero\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction logLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, epsilon = 1e-7,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'logLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'logLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'logLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in logLoss: ');\n\n  const one = scalar(1);\n  const epsilonScalar = scalar(epsilon);\n\n  const l1 = neg(mul($labels, log(add($predictions, epsilonScalar))));\n  const l2 =\n      mul(sub(one, $labels), log(add(sub(one, $predictions), epsilonScalar)));\n  const losses = sub(l1, l2);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const logLoss = op({logLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {squaredDifference} from '../squared_difference';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the mean squared error between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction meanSquaredError_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'meanSquaredError');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'meanSquaredError');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'meanSquaredError');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in meanSquaredError: ');\n\n  const losses = squaredDifference($labels, $predictions);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const meanSquaredError = op({meanSquaredError_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {exp} from '../exp';\nimport {log1p} from '../log1p';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\nfunction sigmoidCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, logits: T|TensorLike): O {\n  const $labels =\n      convertToTensor(labels, 'labels', 'sigmoidCrossEntropyWithLogits');\n  const $logits =\n      convertToTensor(logits, 'logits', 'sigmoidCrossEntropyWithLogits');\n  assertShapesMatch(\n      $labels.shape, $logits.shape, 'Error in sigmoidCrossEntropyWithLogits: ');\n\n  /**\n   * Implementation Details:\n   *\n   * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n   *     z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n   *   = (1 - z) * x + log(1 + exp(-x))\n   *   = x - x * z + log(1 + exp(-x))\n   *\n   *   For x < 0, to avoid overflow in exp(-x), we reformulate the above\n   *     x - x * z + log(1 + exp(-x))\n   *   = log(exp(x)) - x * z + log(1 + exp(-x))\n   *   = - x * z + log(1 + exp(x))\n   *\n   * Hence, to ensure stability and avoid overflow, the implementation uses\n   * this equivalent formulation:\n   *     max(x, 0) - x * z + log(1 + exp(-abs(x)))\n   */\n  const maxOutput = relu($logits);\n  const outputXTarget = mul($logits, $labels);\n  const sigmoidOutput = log1p(exp(neg(abs($logits))));\n\n  return add(sub(maxOutput, outputXTarget), sigmoidOutput);\n}\n\n/**\n * Computes the sigmoid cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newMulticlassLabels = multiclassLabels * (1 - labelSmoothing)\n *                         + 0.5 * labelSmoothing\n *\n * @param multiClassLabels The ground truth output tensor of shape\n * [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction sigmoidCrossEntropy_<T extends Tensor, O extends Tensor>(\n    multiClassLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $multiClassLabels = convertToTensor(\n      multiClassLabels, 'multiClassLabels', 'sigmoidCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'sigmoidCrossEntropy');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'sigmoidCrossEntropy');\n  }\n  assertShapesMatch(\n      $multiClassLabels.shape, $logits.shape, 'Error in sigmoidCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const half = scalar(0.5);\n\n    $multiClassLabels =\n        add(mul($multiClassLabels, sub(one, labelSmoothingScalar)),\n            mul(half, labelSmoothingScalar));\n  }\n  const losses = sigmoidCrossEntropyWithLogits_($multiClassLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const sigmoidCrossEntropy = op({sigmoidCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../../gradients';\nimport {Tensor} from '../../tensor';\nimport {GradSaveFunc} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {expandShapeToKeepDim} from '../axis_util';\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {exp} from '../exp';\nimport {logSumExp} from '../log_sum_exp';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes softmax cross entropy between logits and labels.\n *\n * Measures the probability error in discrete classification tasks in which\n * the classes are mutually exclusive (each entry is in exactly one class).\n * For example, each CIFAR-10 image is labeled with one and only one label: an\n * image can be a dog or a truck, but not both.\n *\n * `NOTE`: While the classes are mutually exclusive, their probabilities need\n * not be. All that is required is that each row of labels is a valid\n * probability distribution. If they are not, the computation of the gradient\n * will be incorrect.\n *\n * `WARNING`: This op expects unscaled logits, since it performs a softmax on\n * logits internally for efficiency. Do not call this op with the output of\n * softmax, as it will produce incorrect results.\n *\n * logits and labels must have the same shape, e.g. [batch_size, num_classes]\n * and the same dtype.\n * @param labels The labels array.\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n */\nfunction softmaxCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T, logits: T, dim = -1): O {\n  if (dim === -1) {\n    dim = logits.rank - 1;\n  }\n\n  if (dim !== logits.rank - 1) {\n    throw Error(\n        `Softmax cross entropy along a non-last dimension is not yet ` +\n        `supported. Labels / logits was rank ${logits.rank} ` +\n        `and dim was ${dim}`);\n  }\n  // Use a custom gradient for numerical stability.\n  const customOp =\n      customGrad((labels: Tensor, logits: Tensor, save: GradSaveFunc) => {\n        // Reference:\n        //   1. http://cs231n.github.io/linear-classify/#softmax\n        //   2. https://blog.feedly.com/tricks-of-the-trade-logsumexp/\n        const keepDims = true;\n        const lse = logSumExp(logits, [dim], keepDims);\n        const logResult = sub(cast(logits, 'float32'), lse);\n        save([labels, logResult]);\n\n        const costVector = neg(mul(logResult, labels));\n        const value: O = sum(costVector, [dim]);\n\n        const gradFunc = (dy: O, saved: Tensor[]) => {\n          const [labels, logResult] = saved;\n          const dyShape = expandShapeToKeepDim(dy.shape, [dim]);\n          return [\n            mul(reshape(dy, dyShape),\n                sub(cast(labels, 'float32'), exp(logResult))),\n            mul(reshape(dy, dyShape),\n                sub(exp(logResult), cast(labels, 'float32'))),\n          ];\n        };\n        return {value, gradFunc};\n      });\n\n  return customOp(labels, logits);\n}\n\n/**\n * Computes the softmax cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newOnehotLabels = onehotLabels * (1 - labelSmoothing)\n *                         + labelSmoothing / numClasses\n *\n * @param onehotLabels One hot encoded labels\n *    [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or 1, and must be\n *    broadcastable to `loss`  of shape [batch_size]\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction softmaxCrossEntropy_<T extends Tensor, O extends Tensor>(\n    onehotLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $onehotLabels =\n      convertToTensor(onehotLabels, 'onehotLabels', 'softmaxCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'softmaxCrossEntropy');\n  let $weights: Tensor = null;\n\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'softmaxCrossEntropy');\n  }\n\n  assertShapesMatch(\n      $onehotLabels.shape, $logits.shape, 'Error in softmaxCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const numClasses = scalar($onehotLabels.shape[1]);\n\n    $onehotLabels =\n        add(mul($onehotLabels, sub(one, labelSmoothingScalar)),\n            div(labelSmoothingScalar, numClasses));\n  }\n\n  const losses = softmaxCrossEntropyWithLogits_($onehotLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const softmaxCrossEntropy = op({softmaxCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Modularized ops.\nexport {abs} from './abs';\nexport {acos} from './acos';\nexport {acosh} from './acosh';\nexport {add} from './add';\nexport {addN} from './add_n';\nexport {all} from './all';\nexport {any} from './any';\nexport {argMax} from './arg_max';\nexport {argMin} from './arg_min';\nexport {asin} from './asin';\nexport {asinh} from './asinh';\nexport {atan} from './atan';\nexport {atan2} from './atan2';\nexport {atanh} from './atanh';\nexport {avgPool} from './avg_pool';\nexport {avgPool3d} from './avg_pool_3d';\nexport {basicLSTMCell} from './basic_lstm_cell';\nexport {batchToSpaceND} from './batch_to_space_nd';\nexport {batchNorm} from './batchnorm';\nexport {batchNorm2d} from './batchnorm2d';\nexport {batchNorm3d} from './batchnorm3d';\nexport {batchNorm4d} from './batchnorm4d';\nexport {broadcastTo} from './broadcast_to';\nexport {buffer} from './buffer';\nexport {cast} from './cast';\nexport {ceil} from './ceil';\nexport {clipByValue} from './clip_by_value';\nexport {clone} from './clone';\nexport {complex} from './complex';\nexport {concat} from './concat';\nexport {concat1d} from './concat_1d';\nexport {concat2d} from './concat_2d';\nexport {concat3d} from './concat_3d';\nexport {concat4d} from './concat_4d';\nexport {conv1d} from './conv1d';\nexport {conv2d} from './conv2d';\nexport {conv2dTranspose} from './conv2d_transpose';\nexport {conv3d} from './conv3d';\nexport {conv3dTranspose} from './conv3d_transpose';\nexport {cos} from './cos';\nexport {cosh} from './cosh';\nexport {cumsum} from './cumsum';\nexport {depthToSpace} from './depth_to_space';\nexport {depthwiseConv2d} from './depthwise_conv2d';\nexport {diag} from './diag';\nexport {dilation2d} from './dilation2d';\nexport {div} from './div';\nexport {divNoNan} from './div_no_nan';\nexport {dot} from './dot';\nexport {elu} from './elu';\nexport {equal} from './equal';\nexport {erf} from './erf';\nexport {exp} from './exp';\nexport {expandDims} from './expand_dims';\nexport {expm1} from './expm1';\nexport {eye} from './eye';\nexport {fill} from './fill';\nexport {floor} from './floor';\nexport {floorDiv} from './floorDiv';\nexport {gather} from './gather';\nexport {greater} from './greater';\nexport {greaterEqual} from './greater_equal';\nexport {imag} from './imag';\nexport {isFinite} from './is_finite';\nexport {isInf} from './is_inf';\nexport {isNaN} from './is_nan';\nexport {leakyRelu} from './leaky_relu';\nexport {less} from './less';\nexport {lessEqual} from './less_equal';\nexport {linspace} from './linspace';\nexport {localResponseNormalization} from './local_response_normalization';\nexport {log} from './log';\nexport {log1p} from './log1p';\nexport {logSigmoid} from './log_sigmoid';\nexport {logSoftmax} from './log_softmax';\nexport {logSumExp} from './log_sum_exp';\nexport {logicalAnd} from './logical_and';\nexport {logicalNot} from './logical_not';\nexport {logicalOr} from './logical_or';\nexport {logicalXor} from './logical_xor';\nexport {matMul} from './mat_mul';\nexport {max} from './max';\nexport {maxPool} from './max_pool';\nexport {maxPool3d} from './max_pool_3d';\nexport {maxPoolWithArgmax} from './max_pool_with_argmax';\nexport {maximum} from './maximum';\nexport {mean} from './mean';\nexport {min} from './min';\nexport {minimum} from './minimum';\nexport {mod} from './mod';\nexport {moments} from './moments';\nexport {mul} from './mul';\nexport {LSTMCellFunc, multiRNNCell} from './multi_rnn_cell';\nexport {multinomial} from './multinomial';\nexport {neg} from './neg';\nexport {notEqual} from './not_equal';\nexport {oneHot} from './one_hot';\nexport {ones} from './ones';\nexport {onesLike} from './ones_like';\nexport {outerProduct} from './outer_product';\nexport {pad} from './pad';\nexport {pad1d} from './pad1d';\nexport {pad2d} from './pad2d';\nexport {pad3d} from './pad3d';\nexport {pad4d} from './pad4d';\nexport {pool} from './pool';\nexport {pow} from './pow';\nexport {prelu} from './prelu';\nexport {print} from './print';\nexport {prod} from './prod';\nexport {rand} from './rand';\nexport {randomGamma} from './random_gamma';\nexport {randomNormal} from './random_normal';\nexport {randomUniform} from './random_uniform';\nexport {range} from './range';\nexport {real} from './real';\nexport {reciprocal} from './reciprocal';\nexport {relu} from './relu';\nexport {relu6} from './relu6';\nexport {reshape} from './reshape';\nexport {reverse} from './reverse';\nexport {reverse1d} from './reverse_1d';\nexport {reverse2d} from './reverse_2d';\nexport {reverse3d} from './reverse_3d';\nexport {reverse4d} from './reverse_4d';\nexport {round} from './round';\nexport {rsqrt} from './rsqrt';\nexport {scalar} from './scalar';\nexport {selu} from './selu';\nexport {separableConv2d} from './separable_conv2d';\nexport {setdiff1dAsync} from './setdiff1d_async';\nexport {sigmoid} from './sigmoid';\nexport {sign} from './sign';\nexport {sin} from './sin';\nexport {sinh} from './sinh';\nexport {slice} from './slice';\nexport {slice1d} from './slice1d';\nexport {slice2d} from './slice2d';\nexport {slice3d} from './slice3d';\nexport {slice4d} from './slice4d';\nexport {softmax} from './softmax';\nexport {softplus} from './softplus';\nexport {spaceToBatchND} from './space_to_batch_nd';\nexport {fft} from './spectral/fft';\nexport {ifft} from './spectral/ifft';\nexport {irfft} from './spectral/irfft';\nexport {rfft} from './spectral/rfft';\nexport {split} from './split';\nexport {sqrt} from './sqrt';\nexport {square} from './square';\nexport {squaredDifference} from './squared_difference';\nexport {squeeze} from './squeeze';\nexport {stack} from './stack';\nexport {step} from './step';\nexport {stridedSlice} from './strided_slice';\nexport {sub} from './sub';\nexport {sum} from './sum';\nexport {tan} from './tan';\nexport {tanh} from './tanh';\nexport {tensor} from './tensor';\nexport {tensor1d} from './tensor1d';\nexport {tensor2d} from './tensor2d';\nexport {tensor3d} from './tensor3d';\nexport {tensor4d} from './tensor4d';\nexport {tensor5d} from './tensor5d';\nexport {tensor6d} from './tensor6d';\nexport {tile} from './tile';\nexport {topk} from './topk';\nexport {truncatedNormal} from './truncated_normal';\nexport {unique} from './unique';\nexport {unsortedSegmentSum} from './unsorted_segment_sum';\nexport {unstack} from './unstack';\nexport {variable} from './variable';\nexport {where} from './where';\nexport {whereAsync} from './where_async';\nexport {zeros} from './zeros';\nexport {zerosLike} from './zeros_like';\n\nexport * from './boolean_mask';\nexport * from './compare';\nexport * from './binary_ops';\nexport * from './transpose';\nexport * from './norm';\nexport * from './moving_average';\nexport * from './scatter_nd';\nexport * from './sparse_to_dense';\nexport * from './gather_nd';\nexport * from './dropout';\nexport * from './signal_ops_util';\nexport * from './in_top_k';\n\nexport {op, OP_SCOPE_SUFFIX} from './operation';\n\nimport {rfft} from './spectral/rfft';\nimport {fft} from './spectral/fft';\nimport {ifft} from './spectral/ifft';\nimport {irfft} from './spectral/irfft';\nconst spectral = {\n  fft,\n  ifft,\n  rfft,\n  irfft\n};\n\nimport * as fused from './fused_ops';\n\nimport {hammingWindow} from './signal/hamming_window';\nimport {hannWindow} from './signal/hann_window';\nimport {frame} from './signal/frame';\nimport {stft} from './signal/stft';\nconst signal = {\n  hammingWindow,\n  hannWindow,\n  frame,\n  stft,\n};\n\n// Image Ops namespace\nimport {cropAndResize} from './image/crop_and_resize';\nimport {flipLeftRight} from './image/flip_left_right';\nimport {rotateWithOffset} from './image/rotate_with_offset';\nimport {nonMaxSuppression} from './image/non_max_suppression';\nimport {nonMaxSuppressionAsync} from './image/non_max_suppression_async';\nimport {nonMaxSuppressionWithScore} from './image/non_max_suppression_with_score';\nimport {nonMaxSuppressionWithScoreAsync} from './image/non_max_suppression_with_score_async';\nimport {nonMaxSuppressionPadded} from './image/non_max_suppression_padded';\nimport {nonMaxSuppressionPaddedAsync} from './image/non_max_suppression_padded_async';\nimport {resizeBilinear} from './image/resize_bilinear';\nimport {resizeNearestNeighbor} from './image/resize_nearest_neighbor';\nconst image = {\n  flipLeftRight,\n  resizeNearestNeighbor,\n  resizeBilinear,\n  rotateWithOffset,\n  cropAndResize,\n  nonMaxSuppression,\n  nonMaxSuppressionAsync,\n  nonMaxSuppressionWithScore,\n  nonMaxSuppressionWithScoreAsync,\n  nonMaxSuppressionPadded,\n  nonMaxSuppressionPaddedAsync\n};\n\n// linalg namespace\nimport {bandPart} from './linalg/band_part';\nimport {gramSchmidt} from './linalg/gram_schmidt';\nimport {qr} from './linalg/qr';\nconst linalg = {\n  bandPart,\n  gramSchmidt,\n  qr\n};\n\n// losses namespace;\nimport {absoluteDifference} from './losses/absolute_difference';\nimport {computeWeightedLoss} from './losses/compute_weighted_loss';\nimport {cosineDistance} from './losses/cosine_distance';\nimport {hingeLoss} from './losses/hinge_loss';\nimport {huberLoss} from './losses/huber_loss';\nimport {logLoss} from './losses/log_loss';\nimport {meanSquaredError} from './losses/mean_squared_error';\nimport {sigmoidCrossEntropy} from './losses/sigmoid_cross_entropy';\nimport {softmaxCrossEntropy} from './losses/softmax_cross_entropy';\nconst losses = {\n  absoluteDifference,\n  computeWeightedLoss,\n  cosineDistance,\n  hingeLoss,\n  huberLoss,\n  logLoss,\n  meanSquaredError,\n  sigmoidCrossEntropy,\n  softmaxCrossEntropy\n};\n\n// Second level exports.\nexport {image, linalg, losses, spectral, fused, signal};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {dispose} from '../globals';\nimport {variableGrads} from '../gradients';\nimport {scalar} from '../ops/ops';\nimport {Serializable} from '../serialization';\nimport {Scalar, Variable} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\n/**\n * A variable that belongs to an optimizer.\n *\n * The `originalName` field is required for keeping track of the canonical\n * name of the variable, which is usually the name of the model weight that\n * the variable is related to plus a suffix, e.g., 'dense1/kernel/momentum'.\n * The name of the `Variable` object itself cannot be used directly due to\n * possible deduplication: Every `Variable` must have a unique name but more\n * than one optimizer objects of the same type may be created for the same model\n * or the same `Variable`.\n */\nexport interface OptimizerVariable {\n  originalName: string;\n  variable: Variable;\n}\n\n/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */\nexport abstract class Optimizer extends Serializable {\n  protected iterations_: number;\n\n  /**\n   * Executes `f()` and minimizes the scalar output of `f()` by computing\n   * gradients of y with respect to the list of trainable variables provided by\n   * `varList`. If no list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to minimize.\n   * @param returnCost Whether to return the scalar cost value produced by\n   * executing `f()`.\n   * @param varList An optional list of variables to update. If specified, only\n   * the trainable variables in varList will be updated by minimize. Defaults to\n   * all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  minimize(f: () => Scalar, returnCost = false, varList?: Variable[]): Scalar\n      |null {\n    const {value, grads} = this.computeGradients(f, varList);\n\n    if (varList != null) {\n      const gradArray: NamedTensor[] =\n          varList.map(v => ({name: v.name, tensor: grads[v.name]}));\n      this.applyGradients(gradArray);\n    } else {\n      this.applyGradients(grads);\n    }\n\n    // Dispose gradients.\n    dispose(grads);\n\n    if (returnCost) {\n      return value;\n    } else {\n      value.dispose();\n      return null;\n    }\n  }\n\n  /**\n   * The number of iterations that this optimizer instance has been invoked for.\n   */\n  get iterations(): number {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return this.iterations_;\n  }\n\n  protected incrementIterations() {\n    this.iterations_ = this.iterations + 1;\n  }\n\n  /**\n   * Executes f() and computes the gradient of the scalar output of f() with\n   * respect to the list of trainable variables provided by `varList`. If no\n   * list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to use for computing\n   * gradients with respect to variables.\n   * @param varList An optional list of variables to compute gradients with\n   * respect to. If specified, only the trainable variables in varList will have\n   * gradients computed with respect to. Defaults to all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  computeGradients(f: () => Scalar, varList?: Variable[]):\n      {value: Scalar, grads: NamedTensorMap} {\n    return variableGrads(f, varList);\n  }\n\n  /**\n   * Updates variables by using the computed gradients.\n   *\n   * @param variableGradients A mapping of variable name to its gradient value.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  abstract applyGradients(variableGradients: NamedTensorMap|\n                          NamedTensor[]): void;\n\n  /**\n   * Dispose the variables (if any) owned by this optimizer instance.\n   */\n  dispose(): void {\n    if (this.iterations_ != null) {\n      dispose(this.iterations_);\n    }\n  }\n\n  async saveIterations(): Promise<NamedTensor> {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return {\n      name: 'iter',  // Named for Python compatibility.\n      // TODO(cais): Use 'int64' type when available.\n      tensor: scalar(this.iterations_, 'int32')\n    };\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for this optimizer yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error(\n        `setWeights() is not implemented for this optimizer class ` +\n        `${this.getClassName()}`);\n  }\n\n  /**\n   * Extract the first element of the weight values and set it\n   * as the iterations counter variable of this instance of optimizer.\n   *\n   * @param weightValues\n   * @returns Weight values with the first element consumed and excluded.\n   */\n  protected async extractIterations(weightValues: NamedTensor[]):\n      Promise<NamedTensor[]> {\n    this.iterations_ = (await weightValues[0].tensor.data())[0];\n    return weightValues.slice(1);\n  }\n}\n\nObject.defineProperty(Optimizer, Symbol.hasInstance, {\n  value: (instance: Optimizer) => {\n    return instance.minimize != null && instance.computeGradients != null &&\n        instance.applyGradients != null;\n  }\n});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/ops';\nimport {square} from '../ops/square';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdadeltaOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adadelta';  // Name matters for Python compatibility.\n  private accumulatedGrads: OptimizerVariable[] = [];\n  private accumulatedUpdates: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected rho: number,\n      protected epsilon: number = null) {\n    super();\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedGrads[i] == null) {\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accum_grad`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedUpdates[i] == null) {\n        this.accumulatedUpdates[i] = {\n          originalName: `${name}/accum_var`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n      const accumulatedUpdate = this.accumulatedUpdates[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad =\n            add(mul(accumulatedGrad, this.rho),\n                mul(square(gradient), 1 - this.rho));\n\n        const updates =\n            mul(div(sqrt(add(accumulatedUpdate, this.epsilon)),\n                    sqrt(add(accumulatedGrad, this.epsilon))),\n                gradient);\n\n        const newAccumulatedUpdate =\n            add(mul(accumulatedUpdate, this.rho),\n                mul(square(updates), 1 - this.rho));\n\n        accumulatedGrad.assign(newAccumulatedGrad);\n        accumulatedUpdate.assign(newAccumulatedUpdate);\n\n        const newValue = add(mul(updates, -this.learningRate), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedUpdates != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n      dispose(this.accumulatedUpdates.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedGrads, ...this.accumulatedUpdates];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedGrads =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedUpdates =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'rho': this.rho,\n      'epsilon': this.epsilon\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['rho'], config['epsilon']);\n  }\n}\nregisterClass(AdadeltaOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {fill} from '../ops/fill';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdagradOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adagrad';  // Note: Name matters for Python compatibility.\n\n  private accumulatedGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private initialAccumulatorValue = 0.1) {\n    super();\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulatedGrads[i] == null) {\n        const trainable = false;\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accumulator`,\n          variable: tidy(\n              () => fill(value.shape, this.initialAccumulatorValue)\n                        .variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad = add(accumulatedGrad, square(gradient));\n        accumulatedGrad.assign(newAccumulatedGrad);\n\n        const newValue = add(\n            mul(div(gradient,\n                    sqrt(add(newAccumulatedGrad, ENGINE.backend.epsilon()))),\n                -this.learningRate),\n            value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedGrads != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulatedGrads.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulatedGrads = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'initialAccumulatorValue': this.initialAccumulatorValue,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['initialAccumulatorValue']);\n  }\n}\nregisterClass(AdagradOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adam';  // Note: Name matters for Python compatibility.\n  private accBeta1: Variable;\n  private accBeta2: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedSecondMoment: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null) {\n    super();\n    tidy(() => {\n      // accB* will be updated by batch.\n      this.accBeta1 = scalar(beta1).variable();\n      this.accBeta2 = scalar(beta2).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const oneMinusAccBeta2 = sub(1, this.accBeta2);\n\n      varNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n        if (this.accumulatedSecondMoment[i] == null) {\n          this.accumulatedSecondMoment[i] = {\n            originalName: `${name}/v`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const secondMoment = this.accumulatedSecondMoment[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n        const newSecondMoment =\n            add(mul(secondMoment, this.beta2),\n                mul(square(gradient), 1 - this.beta2));\n\n        const biasCorrectedFirstMoment = div(newFirstMoment, oneMinusAccBeta1);\n        const biasCorrectedSecondMoment =\n            div(newSecondMoment, oneMinusAccBeta2);\n\n        firstMoment.assign(newFirstMoment);\n        secondMoment.assign(newSecondMoment);\n\n        const newValue =\n            add(mul(div(biasCorrectedFirstMoment,\n                        add(sqrt(biasCorrectedSecondMoment), this.epsilon)),\n                    -this.learningRate),\n                value);\n        value.assign(newValue);\n      });\n\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n      this.accBeta2.assign(mul(this.accBeta2, this.beta2));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.accBeta2.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedSecondMoment != null) {\n      dispose(this.accumulatedSecondMoment.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedFirstMoment, ...this.accumulatedSecondMoment];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    tidy(() => {\n      this.accBeta1.assign(pow(this.beta1, this.iterations_ + 1));\n      this.accBeta2.assign(pow(this.beta2, this.iterations_ + 1));\n    });\n\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedFirstMoment =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedSecondMoment =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon']);\n  }\n}\nregisterClass(AdamOptimizer);\n", "\uFEFF/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {abs} from '../ops/abs';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {maximum} from '../ops/maximum';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamaxOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adamax';  // Note: Name matters for Python compatbility.\n  private accBeta1: Variable;\n  private iteration: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null,\n      protected decay = 0.0) {\n    super();\n\n    tidy(() => {\n      this.iteration = scalar(0).variable();\n      this.accBeta1 = scalar(beta1).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const lr =\n          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));\n\n      variableNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n        if (this.accumulatedWeightedInfNorm[i] == null) {\n          this.accumulatedWeightedInfNorm[i] = {\n            originalName: `${name}/v`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n\n        const ut0 = mul(weightedInfNorm, this.beta2);\n        const ut1 = abs(gradient);\n\n        const newWeightedInfNorm = maximum(ut0, ut1);\n\n        firstMoment.assign(newFirstMoment);\n        weightedInfNorm.assign(newWeightedInfNorm);\n\n        const newValue =\n            add(mul(div(lr, oneMinusAccBeta1),\n                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),\n                value);\n\n        value.assign(newValue);\n      });\n\n      this.iteration.assign(add(this.iteration, 1));\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.iteration.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedWeightedInfNorm != null) {\n      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for Adamax yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error('setWeights() is not implemented for Adamax yet.');\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n      'decay': this.decay\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon'], config['decay']);\n  }\n}\nregisterClass(AdamaxOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {keep, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer} from './optimizer';\n\n/** @doclink Optimizer */\nexport class SGDOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'SGD';  // Note: Name matters for Python compatibility.\n  protected c: Scalar;\n\n  constructor(protected learningRate: number) {\n    super();\n    this.setLearningRate(learningRate);\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    varNames.forEach((name, i) => {\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n      const value = ENGINE.registeredVariables[name];\n      tidy(() => {\n        const newValue = add(mul(this.c, gradient), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  /**\n   * Sets the learning rate of the optimizer.\n   */\n  setLearningRate(learningRate: number) {\n    this.learningRate = learningRate;\n    if (this.c != null) {\n      this.c.dispose();\n    }\n    this.c = keep(scalar(-learningRate));\n  }\n\n  dispose() {\n    this.c.dispose();\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    return [await this.saveIterations()];\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    if (weightValues.length !== 0) {\n      throw new Error('SGD optimizer does not have settable weights.');\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {'learningRate': this.learningRate};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate']);\n  }\n}\nregisterClass(SGDOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {OptimizerVariable} from './optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\n/** @doclink Optimizer */\nexport class MomentumOptimizer extends SGDOptimizer {\n  /** @nocollapse */\n  static className = 'Momentum';  // Name matters for Python compatibility.\n  private m: Scalar;\n  private accumulations: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private momentum: number,\n      private useNesterov = false) {\n    super(learningRate);\n    this.m = scalar(this.momentum);\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulations[i] == null) {\n        const trainable = false;\n        this.accumulations[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const accumulation = this.accumulations[i].variable;\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      tidy(() => {\n        let newValue: Tensor;\n        const newAccumulation = add(mul(this.m, accumulation), gradient);\n        if (this.useNesterov) {\n          newValue = add(\n              mul(this.c, add(gradient, mul(newAccumulation, this.m))), value);\n        } else {\n          newValue = add(mul(this.c, newAccumulation), value);\n        }\n        accumulation.assign(newAccumulation);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.m.dispose();\n    if (this.accumulations != null) {\n      dispose(this.accumulations.map(v => v.variable));\n    }\n  }\n\n  /**\n   * Sets the momentum of the optimizer.\n   *\n   * @param momentum\n   */\n  setMomentum(momentum: number) {\n    this.momentum = momentum;\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulations.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulations = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'momentum': this.momentum,\n      'useNesterov': this.useNesterov\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['momentum'], config['useNesterov']);\n  }\n}\nregisterClass(MomentumOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class RMSPropOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'RMSProp';  // Note: Name matters for Python compatibility.\n  private centered: boolean;\n\n  private accumulatedMeanSquares: OptimizerVariable[] = [];\n  private accumulatedMoments: OptimizerVariable[] = [];\n  private accumulatedMeanGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected decay = 0.9,\n      protected momentum = 0.0, protected epsilon: number = null,\n      centered = false) {\n    super();\n\n    this.centered = centered;\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n    if (learningRate == null) {\n      throw new Error(`learningRate for RMSPropOptimizer must be defined.`);\n    }\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedMeanSquares[i] == null) {\n        this.accumulatedMeanSquares[i] = {\n          originalName: `${name}/rms`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMoments[i] == null) {\n        this.accumulatedMoments[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMeanGrads[i] == null && this.centered) {\n        this.accumulatedMeanGrads[i] = {\n          originalName: `${name}/mg`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedMeanSquare = this.accumulatedMeanSquares[i].variable;\n      const accumulatedMoments = this.accumulatedMoments[i].variable;\n      tidy(() => {\n        const newAccumulatedMeanSquare =\n            add(mul(accumulatedMeanSquare, this.decay),\n                mul(square(gradient), 1 - this.decay));\n\n        if (this.centered) {\n          const accumulatedMeanGrad = this.accumulatedMeanGrads[i].variable;\n          // Centered gradient\n          const newAccumulatedMeanGrad =\n              add(mul(accumulatedMeanGrad, this.decay),\n                  mul(gradient, 1 - this.decay));\n\n          const gradContribution =\n              div(mul(gradient, this.learningRate),\n                  sqrt(\n                      sub(newAccumulatedMeanSquare,\n                          add(square(newAccumulatedMeanGrad), this.epsilon))));\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum), gradContribution);\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMeanGrad.assign(newAccumulatedMeanGrad);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        } else {\n          // Plain gradient\n          const newAccumulatedMeanSquare =\n              add(mul(accumulatedMeanSquare, this.decay),\n                  mul(square(gradient), 1 - this.decay));\n\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum),\n                  div(mul(gradient, this.learningRate),\n                      sqrt(add(newAccumulatedMeanSquare, this.epsilon))));\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        }\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedMeanSquares != null) {\n      dispose(this.accumulatedMeanSquares.map(v => v.variable));\n    }\n    if (this.accumulatedMeanGrads != null && this.centered) {\n      dispose(this.accumulatedMeanGrads.map(v => v.variable));\n    }\n    if (this.accumulatedMoments != null) {\n      dispose(this.accumulatedMoments.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedMeanSquares, ...this.accumulatedMoments];\n    if (this.centered) {\n      variables.push(...this.accumulatedMeanGrads);\n    }\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount =\n        this.centered ? weightValues.length / 3 : weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedMeanSquares =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedMoments =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n    if (this.centered) {\n      this.accumulatedMeanGrads =\n          weightValues.slice(variableCount * 2, variableCount * 3)\n              .map(v => ({\n                     originalName: v.name,\n                     variable: v.tensor.variable(trainable)\n                   }));\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'decay': this.decay,\n      'momentum': this.momentum,\n      'epsilon': this.epsilon,\n      'centered': this.centered\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['decay'], config['momentum'],\n        config['epsilon'], config['centered']);\n  }\n}\nregisterClass(RMSPropOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AdadeltaOptimizer} from './adadelta_optimizer';\nimport {AdagradOptimizer} from './adagrad_optimizer';\nimport {AdamOptimizer} from './adam_optimizer';\nimport {AdamaxOptimizer} from './adamax_optimizer';\nimport {MomentumOptimizer} from './momentum_optimizer';\nimport {RMSPropOptimizer} from './rmsprop_optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\nexport class OptimizerConstructors {\n  /**\n   * Constructs a `tf.SGDOptimizer` that uses stochastic gradient descent.\n   *\n   * ```js\n   * // Fit a quadratic function by learning the coefficients a, b, c.\n   * const xs = tf.tensor1d([0, 1, 2, 3]);\n   * const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);\n   *\n   * const a = tf.scalar(Math.random()).variable();\n   * const b = tf.scalar(Math.random()).variable();\n   * const c = tf.scalar(Math.random()).variable();\n   *\n   * // y = a * x^2 + b * x + c.\n   * const f = x => a.mul(x.square()).add(b.mul(x)).add(c);\n   * const loss = (pred, label) => pred.sub(label).square().mean();\n   *\n   * const learningRate = 0.01;\n   * const optimizer = tf.train.sgd(learningRate);\n   *\n   * // Train the model.\n   * for (let i = 0; i < 10; i++) {\n   *   optimizer.minimize(() => loss(f(xs), ys));\n   * }\n   *\n   * // Make predictions.\n   * console.log(\n   *     `a: ${a.dataSync()}, b: ${b.dataSync()}, c: ${c.dataSync()}`);\n   * const preds = f(xs).dataSync();\n   * preds.forEach((pred, i) => {\n   *   console.log(`x: ${i}, pred: ${pred}`);\n   * });\n   * ```\n   *\n   * @param learningRate The learning rate to use for the SGD algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static sgd(learningRate: number): SGDOptimizer {\n    return new SGDOptimizer(learningRate);\n  }\n\n  /**\n   * Constructs a `tf.MomentumOptimizer` that uses momentum gradient\n   * descent.\n   *\n   * See\n   * [http://proceedings.mlr.press/v28/sutskever13.pdf](\n   * http://proceedings.mlr.press/v28/sutskever13.pdf)\n   *\n   * @param learningRate The learning rate to use for the Momentum gradient\n   * descent algorithm.\n   * @param momentum The momentum to use for the momentum gradient descent\n   * algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static momentum(learningRate: number, momentum: number, useNesterov = false):\n      MomentumOptimizer {\n    return new MomentumOptimizer(learningRate, momentum, useNesterov);\n  }\n\n  /**\n   * Constructs a `tf.RMSPropOptimizer` that uses RMSProp gradient\n   * descent. This implementation uses plain momentum and is not centered\n   * version of RMSProp.\n   *\n   * See\n   * [http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf](\n   * http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)\n   *\n   * @param learningRate The learning rate to use for the RMSProp gradient\n   * descent algorithm.\n   * @param decay The discounting factor for the history/coming gradient.\n   * @param momentum The momentum to use for the RMSProp gradient descent\n   * algorithm.\n   * @param epsilon Small value to avoid zero denominator.\n   * @param centered If true, gradients are normalized by the estimated\n   * variance of the gradient.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static rmsprop(\n      learningRate: number, decay = .9, momentum = 0.0, epsilon: number = null,\n      centered = false): RMSPropOptimizer {\n    return new RMSPropOptimizer(\n        learningRate, decay, momentum, epsilon, centered);\n  }\n\n  /**\n   * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adam gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adam(\n      learningRate = 0.001, beta1 = 0.9, beta2 = 0.999,\n      epsilon: number = null): AdamOptimizer {\n    return new AdamOptimizer(learningRate, beta1, beta2, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.\n   * See [https://arxiv.org/abs/1212.5701](https://arxiv.org/abs/1212.5701)\n   *\n   * @param learningRate The learning rate to use for the Adadelta gradient\n   * descent algorithm.\n   * @param rho The learning rate decay over each update.\n   * @param epsilon A constant epsilon used to better condition the grad\n   * update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adadelta(learningRate = .001, rho = .95, epsilon: number = null):\n      AdadeltaOptimizer {\n    return new AdadeltaOptimizer(learningRate, rho, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdamaxOptimizer` that uses the Adamax algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adamax gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   * @param decay The learning rate decay over each update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adamax(\n      learningRate = 0.002, beta1 = 0.9, beta2 = 0.999, epsilon: number = null,\n      decay = 0.0): AdamaxOptimizer {\n    return new AdamaxOptimizer(learningRate, beta1, beta2, epsilon, decay);\n  }\n\n  /**\n   * Constructs a `tf.AdagradOptimizer` that uses the Adagrad algorithm.\n   * See\n   * [http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf](\n   * http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\n   * or\n   * [http://ruder.io/optimizing-gradient-descent/index.html#adagrad](\n   * http://ruder.io/optimizing-gradient-descent/index.html#adagrad)\n   *\n   * @param learningRate The learning rate to use for the Adagrad gradient\n   * descent algorithm.\n   * @param initialAccumulatorValue Starting value for the accumulators, must be\n   * positive.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adagrad(learningRate: number, initialAccumulatorValue = 0.1):\n      AdagradOptimizer {\n    return new AdagradOptimizer(learningRate, initialAccumulatorValue);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// So typings can propagate.\nimport {AdadeltaOptimizer} from './optimizers/adadelta_optimizer';\nimport {AdagradOptimizer} from './optimizers/adagrad_optimizer';\nimport {AdamOptimizer} from './optimizers/adam_optimizer';\nimport {AdamaxOptimizer} from './optimizers/adamax_optimizer';\nimport {MomentumOptimizer} from './optimizers/momentum_optimizer';\nimport {OptimizerConstructors} from './optimizers/optimizer_constructors';\nimport {RMSPropOptimizer} from './optimizers/rmsprop_optimizer';\nimport {SGDOptimizer} from './optimizers/sgd_optimizer';\n\n// tslint:disable-next-line:no-unused-expression\n[MomentumOptimizer, SGDOptimizer, AdadeltaOptimizer, AdagradOptimizer,\n RMSPropOptimizer, AdamaxOptimizer, AdamOptimizer];\n\nexport const train = {\n  sgd: OptimizerConstructors.sgd,\n  momentum: OptimizerConstructors.momentum,\n  adadelta: OptimizerConstructors.adadelta,\n  adagrad: OptimizerConstructors.adagrad,\n  rmsprop: OptimizerConstructors.rmsprop,\n  adamax: OptimizerConstructors.adamax,\n  adam: OptimizerConstructors.adam\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst delayCallback: Function = (() => {\n  if (typeof requestAnimationFrame !== 'undefined') {\n    return requestAnimationFrame;\n  } else if (typeof setImmediate !== 'undefined') {\n    return setImmediate;\n  }\n  return (f: Function) => f();  // no delays\n})();\n\n/**\n * Returns a promise that resolve when a requestAnimationFrame has completed.\n *\n * On Node.js this uses setImmediate instead of requestAnimationFrame.\n *\n * This is simply a sugar method so that users can do the following:\n * `await tf.nextFrame();`\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nfunction nextFrame(): Promise<void> {\n  return new Promise<void>(resolve => delayCallback(() => resolve()));\n}\n\nexport {nextFrame};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Returns the image center in pixels.\nexport function getImageCenter(\n    center: number|[number, number], imageHeight: number,\n    imageWidth: number): [number, number] {\n  const centerX =\n      imageWidth * (typeof center === 'number' ? center : center[0]);\n  const centerY =\n      imageHeight * (typeof center === 'number' ? center : center[1]);\n  return [centerX, centerY];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Gets the new shape of the input Tensor after it's been reshaped\n * to:\n * [blockShape[0], ..., blockShape[M-1], batch / prod(blockShape),\n * inputShape[1], ..., inputShape[N-1]]\n *\n * See step 1: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshaped(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  let reshaped: number[] = [];\n  if (batchToSpace) {\n    reshaped = reshaped.concat(blockShape.slice(0));\n    reshaped.push(inputShape[0] / prod);\n    reshaped = reshaped.concat(inputShape.slice(1));\n  } else {\n    reshaped = reshaped.concat(inputShape[0]);\n    const spatialLength = blockShape.length;\n    for (let i = 0; i < spatialLength; ++i) {\n      reshaped =\n          reshaped.concat([inputShape[i + 1] / blockShape[i], blockShape[i]]);\n    }\n    reshaped = reshaped.concat(inputShape.slice(spatialLength + 1));\n  }\n  return reshaped;\n}\n\n/**\n * Gets the permutation that will transpose the dimensions of the\n * reshaped tensor to shape:\n *\n * [batch / prod(block_shape),inputShape[1], blockShape[0], ...,\n * inputShape[M], blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * see step 2: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getPermuted(\n    reshapedRank: number, blockShapeRank: number,\n    batchToSpace = true): number[] {\n  const permuted = [];\n  if (batchToSpace) {\n    permuted.push(blockShapeRank);\n    for (let i = blockShapeRank + 1; i < reshapedRank; ++i) {\n      if (i <= 2 * blockShapeRank) {\n        permuted.push(i);\n        permuted.push(i - (blockShapeRank + 1));\n      } else {\n        permuted.push(i);\n      }\n    }\n  } else {\n    const permutedBeforeBatch = [];\n    const permutedAfterBatch = [];\n    for (let i = 1; i < reshapedRank; ++i) {\n      if (i >= blockShapeRank * 2 + 1 || i % 2 === 1) {\n        permutedAfterBatch.push(i);\n      } else {\n        permutedBeforeBatch.push(i);\n      }\n    }\n    permuted.push(...permutedBeforeBatch);\n    permuted.push(0);\n    permuted.push(...permutedAfterBatch);\n  }\n  return permuted;\n}\n\n/**\n * Gets the shape of the reshaped and permuted input Tensor before any cropping\n * is applied.  The new shape will be:\n *\n * [batch / prod(blockShape),inputShape[1] * blockShape[0], ...,\n * inputShape[M] * blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 3: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshapedPermuted(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  const reshapedPermuted = [];\n\n  if (batchToSpace) {\n    reshapedPermuted.push(inputShape[0] / prod);\n  } else {\n    reshapedPermuted.push(inputShape[0] * prod);\n  }\n\n  for (let i = 1; i < inputShape.length; ++i) {\n    if (i <= blockShape.length) {\n      if (batchToSpace) {\n        reshapedPermuted.push(blockShape[i - 1] * inputShape[i]);\n      } else {\n        reshapedPermuted.push(inputShape[i] / blockShape[i - 1]);\n      }\n    } else {\n      reshapedPermuted.push(inputShape[i]);\n    }\n  }\n\n  return reshapedPermuted;\n}\n\n/**\n * Converts the crops argument into the beginning coordinates of a slice\n * operation.\n */\nexport function getSliceBeginCoords(\n    crops: number[][], blockShape: number): number[] {\n  const sliceBeginCoords = [0];\n  for (let i = 0; i < blockShape; ++i) {\n    sliceBeginCoords.push(crops[i][0]);\n  }\n  return sliceBeginCoords;\n}\n\n/**\n * Converts the crops argument into the size of a slice operation.  When\n * combined with getSliceBeginCoords this function allows the reshaped and\n * permuted Tensor to be cropped to its final output shape of:\n *\n * inputShape[1] * blockShape[0] - crops[0,0] - crops[0,1], ...,\n * inputShape[M] * blockShape[M-1] -crops[M-1,0] -\n * crops[M-1,1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 4: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getSliceSize(\n    uncroppedShape: number[], crops: number[][], blockShape: number): number[] {\n  const sliceSize = uncroppedShape.slice(0, 1);\n  for (let i = 0; i < blockShape; ++i) {\n    sliceSize.push(uncroppedShape[i + 1] - crops[i][0] - crops[i][1]);\n  }\n\n  return sliceSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const SELU_SCALEALPHA = 1.7580993408473768599402175208123;\nexport const SELU_SCALE = 1.0507009873554804934193349852946;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const ERF_P = 0.3275911;\nexport const ERF_A1 = 0.254829592;\nexport const ERF_A2 = -0.284496736;\nexport const ERF_A3 = 1.421413741;\nexport const ERF_A4 = -1.453152027;\nexport const ERF_A5 = 1.061405429;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\n\nexport function warn(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.warn(...msg);\n  }\n}\n\nexport function log(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.log(...msg);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TypedArray} from '../types';\n/**\n * Merges real and imaginary Float32Arrays into a single complex Float32Array.\n *\n * The memory layout is interleaved as follows:\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n * complex: [r0, i0, r1, i1, r2, i2]\n *\n * This is the inverse of splitRealAndImagArrays.\n *\n * @param real The real values of the complex tensor values.\n * @param imag The imag values of the complex tensor values.\n * @returns A complex tensor as a Float32Array with merged values.\n */\nexport function mergeRealAndImagArrays(\n    real: Float32Array, imag: Float32Array): Float32Array {\n  if (real.length !== imag.length) {\n    throw new Error(\n        `Cannot merge real and imag arrays of different lengths. real:` +\n        `${real.length}, imag: ${imag.length}.`);\n  }\n  const result = new Float32Array(real.length * 2);\n  for (let i = 0; i < result.length; i += 2) {\n    result[i] = real[i / 2];\n    result[i + 1] = imag[i / 2];\n  }\n  return result;\n}\n\n/**\n * Splits a complex Float32Array into real and imag parts.\n *\n * The memory layout is interleaved as follows:\n * complex: [r0, i0, r1, i1, r2, i2]\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n *\n * This is the inverse of mergeRealAndImagArrays.\n *\n * @param complex The complex tensor values.\n * @returns An object with real and imag Float32Array components of the complex\n *     tensor.\n */\nexport function splitRealAndImagArrays(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(complex.length / 2);\n  const imag = new Float32Array(complex.length / 2);\n  for (let i = 0; i < complex.length; i += 2) {\n    real[i / 2] = complex[i];\n    imag[i / 2] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts even indexed complex values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithEvenIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.ceil(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 0; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts odd indexed comple values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithOddIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.floor(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 2; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Get the map representing a complex value in the given array.\n * @param complex The complex tensor values.\n * @param index An index of the target complex value.\n */\nexport function getComplexWithIndex(\n    complex: Float32Array, index: number): {real: number, imag: number} {\n  const real = complex[index * 2];\n  const imag = complex[index * 2 + 1];\n  return {real, imag};\n}\n\n/**\n * Insert a given complex value into the TypedArray.\n * @param data The array in which the complex value is inserted.\n * @param c The complex value to be inserted.\n * @param index An index of the target complex value.\n */\nexport function assignToTypedArray(\n    data: TypedArray, real: number, imag: number, index: number) {\n  data[index * 2] = real;\n  data[index * 2 + 1] = imag;\n}\n\n/**\n * Make the list of exponent terms used by FFT.\n */\nexport function exponents(\n    n: number, inverse: boolean): {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(n / 2);\n  const imag = new Float32Array(n / 2);\n  for (let i = 0; i < Math.ceil(n / 2); i++) {\n    const x = (inverse ? 2 : -2) * Math.PI * (i / n);\n    real[i] = Math.cos(x);\n    imag[i] = Math.sin(x);\n  }\n  return {real, imag};\n}\n\n/**\n * Make the exponent term used by FFT.\n */\nexport function exponent(\n    k: number, n: number, inverse: boolean): {real: number, imag: number} {\n  const x = (inverse ? 2 : -2) * Math.PI * (k / n);\n  const real = Math.cos(x);\n  const imag = Math.sin(x);\n  return {real, imag};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {cast} from '../ops/cast';\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\nimport {Rank} from '../types';\nimport {DataType, ShapeMap} from '../types';\nimport {hasEncodingLoss, makeZerosTypedArray} from '../util';\n\nimport {KernelBackend} from './backend';\n\n// Utilities needed by backend consumers of tf-core.\nexport * from '../ops/axis_util';\nexport * from '../ops/broadcast_util';\nexport * from '../ops/concat_util';\nexport * from '../ops/conv_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/fused_types';\nexport * from '../ops/reduce_util';\n\nimport * as slice_util from '../ops/slice_util';\nexport {slice_util};\n\nexport {BackendValues, TypedArray, upcastType, PixelData} from '../types';\nexport {MemoryInfo, TimingInfo} from '../engine';\nexport * from '../ops/rotate_util';\nexport * from '../ops/array_ops_util';\nexport * from '../ops/gather_nd_util';\nexport * from '../ops/scatter_nd_util';\nexport * from '../ops/selu_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/erf_util';\nexport * from '../log';\nexport * from '../backends/complex_util';\nexport * from '../ops/split_util';\n\nimport * as segment_util from '../ops/segment_util';\nexport {segment_util};\n\nexport function castTensor<T extends Tensor>(\n    x: T, dtype: DataType, backend: KernelBackend): T {\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return x.clone();\n    }\n    const zerosTensor = zeros(x.shape);\n    const floatX = cast(x, 'float32');\n    const result = backend.complex(floatX, zerosTensor);\n    zerosTensor.dispose();\n    floatX.dispose();\n    return result as T;\n  }\n\n  if (!hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    return ENGINE.makeTensorFromDataId(x.dataId, x.shape, dtype) as T;\n  }\n  if (x.dtype === 'complex64') {\n    const real = backend.real(x);\n    const result = cast(real, dtype);\n    real.dispose();\n    return result;\n  }\n  if (dtype === 'int32') {\n    return backend.int(x);\n  } else if (dtype === 'bool') {\n    const zero = scalar(0, x.dtype);\n    const result = backend.notEqual(x, zero) as T;\n    zero.dispose();\n    return result;\n  } else {\n    throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n  }\n}\n\nexport function reshapeTensor<T extends Tensor, R extends Rank>(\n    x: T, shape: ShapeMap[R]): Tensor<R> {\n  return ENGINE.makeTensorFromDataId(x.dataId, shape, x.dtype) as Tensor<R>;\n}\n\nexport function linspaceImpl(start: number, stop: number, num: number) {\n  const step = (stop - start) / (num - 1);\n\n  const values = makeZerosTypedArray(num, 'float32');\n  values[0] = start;\n  for (let i = 1; i < values.length; i++) {\n    values[i] = values[i - 1] + step;\n  }\n\n  return tensor1d(values, 'float32');\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\n// TODO(annxingyuan): Use this helper in WASM Split kernel once intermediate\n// kernels have been modularized in WebGL and CPU\n// https://github.com/tensorflow/tfjs/issues/2822.\n/** Shared implementation of the split kernel across WebGL and CPU. */\nexport function split<T extends Tensor>(\n    x: T, sizeSplits: number[], axis: number): T[] {\n  const begin = new Array(x.rank).fill(0);\n  const size = x.shape.slice();\n  return sizeSplits.map(s => {\n    const sliceSize = [...size];\n    sliceSize[axis] = s;\n    const sliceT = slice(x, begin, sliceSize);\n    begin[axis] += s;\n    return sliceT;\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * An implementation of the tile kernel shared between webgl and cpu for string\n * tensors only.\n */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {DataType, Rank} from '../types';\n\nexport function tile<R extends Rank>(\n    xBuf: TensorBuffer<R, DataType>, reps: number[]): Tensor<R> {\n  const newShape: number[] = new Array(xBuf.rank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = xBuf.shape[i] * reps[i];\n  }\n  const result = buffer(newShape, xBuf.dtype);\n  for (let i = 0; i < result.values.length; ++i) {\n    const newLoc = result.indexToLoc(i);\n\n    const originalLoc: number[] = new Array(xBuf.rank);\n    for (let j = 0; j < originalLoc.length; j++) {\n      originalLoc[j] = newLoc[j] % xBuf.shape[j];\n    }\n\n    const originalIndex = xBuf.locToIndex(originalLoc);\n\n    result.values[i] = xBuf.values[originalIndex];\n  }\n  return result.toTensor() as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the TopK kernel shared between webgl and cpu. */\n\nimport {tensor} from '../ops/tensor';\nimport {Tensor} from '../tensor';\nimport {NumericDataType, TypedArray} from '../types';\nimport {getTypedArrayFromDType} from '../util';\n\nexport function topkImpl<T extends Tensor>(\n    x: TypedArray, xShape: number[], xDtype: NumericDataType, k: number,\n    sorted: boolean): [T, T] {\n  // Reshape into a 2d tensor [batch, lastDim] and compute topk along lastDim.\n  const lastDim = xShape[xShape.length - 1];\n  const [batch, size] = [x.length / lastDim, lastDim];\n  const allTopKVals = getTypedArrayFromDType(xDtype, batch * k);\n  const allTopKIndices = getTypedArrayFromDType('int32', batch * k);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = x.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    const outOffset = b * k;\n    const topKVals = allTopKVals.subarray(outOffset, outOffset + k);\n    const topKIndices = allTopKIndices.subarray(outOffset, outOffset + k);\n    for (let i = 0; i < k; i++) {\n      topKVals[i] = valAndInd[i].value;\n      topKIndices[i] = valAndInd[i].index;\n    }\n  }\n  // Reshape back to the original input shape, except that the last\n  // dimension is k.\n  const outputShape = xShape.slice();\n  outputShape[outputShape.length - 1] = k;\n  return [\n    tensor(allTopKVals, outputShape, xDtype) as T,\n    tensor(allTopKIndices, outputShape, 'int32') as T\n  ];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport {nonMaxSuppressionV3Impl, nonMaxSuppressionV4Impl, nonMaxSuppressionV5Impl} from './non_max_suppression_impl';\nexport {split} from './split_shared';\nexport {tile} from './tile_impl';\nexport {topkImpl} from './topk_impl';\nexport {whereImpl} from './where_impl';\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const absGradConfig: GradConfig = {\n  kernelName: Abs,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, step(cast(x, 'float32'), -1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acosGradConfig: GradConfig = {\n  kernelName: Acos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = square(cast(x, 'float32'));\n        const b = sqrt(sub(scalar(1), a));\n        return neg(div(dy, b));\n      }\n\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acoshGradConfig: GradConfig = {\n  kernelName: Acosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(sub(square(cast(x, 'float32')), 1));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Add} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const addGradConfig: GradConfig = {\n  kernelName: Add,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AddN} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const addNGradConfig: GradConfig = {\n  kernelName: AddN,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const ders: {[key: string]: () => Tensor} = {};\n    saved.forEach((_, i) => {\n      ders[i] = () => dy.clone();\n    });\n    return ders;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMax} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMaxGradConfig: GradConfig = {\n  kernelName: ArgMax,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMinGradConfig: GradConfig = {\n  kernelName: ArgMin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const asinGradConfig: GradConfig = {\n  kernelName: Asin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, sqrt(sub(scalar(1), square(cast(x, 'float32')))))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const asinhGradConfig: GradConfig = {\n  kernelName: Asinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(add(scalar(1), square(cast(x, 'float32'))));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const atan2GradConfig: GradConfig = {\n  kernelName: Atan2,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const d = add(square(a), square(b));\n      let res = mul(dy, div(b, d));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      const d = add(square(a), square(b));\n      let res = neg(mul(dy, div(a, d)));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const atanGradConfig: GradConfig = {\n  kernelName: Atan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, add(square(cast(x, 'float32')), 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {scalar} from '../ops/scalar';\nimport {Tensor} from '../tensor';\n\nexport const atanhGradConfig: GradConfig = {\n  kernelName: Atanh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, sub(scalar(1), square(cast(x, 'float32'))))};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool3DBackprop, AvgPool3DBackpropAttrs, AvgPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d avg pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0. The dilation\n *     rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction avgPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in avgPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in avgPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.avgPool3dBackprop(dy5D, input5D, convInfo);\n  };\n\n  const inputs: AvgPool3DBackpropInputs = {dy: dy5D, input: input5D};\n\n  const attrs: AvgPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      AvgPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3dBackprop = op({avgPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool3D, AvgPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPool3dBackprop} from '../ops/avg_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const avgPool3DGradConfig: GradConfig = {\n  kernelName: AvgPool3D,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as AvgPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => avgPool3dBackprop(\n          dy as Tensor5D, x, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of an 2D avg pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The input image, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction avgPoolBackprop_<T extends Tensor3D|Tensor4D>(\n    dy: T|TensorLike, input: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy (${\n          $dy.rank})`);\n\n  let input4D = $input as Tensor4D;\n  let dy4D = $dy as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($input.rank === 3) {\n    reshapedTo4D = true;\n    input4D =\n        reshape($input, [1, $input.shape[0], $input.shape[1], $input.shape[2]]);\n    dy4D = reshape($dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2]]);\n  }\n\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in avgPoolBackprop: dy must be rank 4 but got rank ` +\n          `${dy4D.rank}.`);\n  util.assert(\n      input4D.rank === 4,\n      () => `Error in avgPoolBackprop: input must be rank 4 but got rank ` +\n          `${input4D.rank}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        input4D.shape, filterSize, strides, 1 /* dilations */, pad);\n\n    return backend.avgPoolBackprop(dy4D, input4D, convInfo);\n  };\n\n  const inputs: AvgPoolBackpropInputs = {dy: dy4D, input: input4D};\n\n  const attrs: AvgPoolBackpropAttrs = {filterSize, strides, pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, AvgPoolBackprop,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const avgPoolBackprop = op({avgPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool, AvgPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPoolBackprop} from '../ops/avg_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const avgPoolGradConfig: GradConfig = {\n  kernelName: AvgPool,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as AvgPoolAttrs;\n    return {\n      x: () => avgPoolBackprop(dy as Tensor4D, x, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchMatMul, BatchMatMulAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {matMul} from '../ops/mat_mul';\nimport {Tensor, Tensor3D} from '../tensor';\n\nexport const batchMatMulGradConfig: GradConfig = {\n  kernelName: BatchMatMul,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor3D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [a, b] = saved as Tensor3D[];\n\n    const {transposeA, transposeB} = attrs as {} as BatchMatMulAttrs;\n\n    if (!transposeA && !transposeB) {\n      return {\n        a: () => matMul(dy, b, false, true),\n        b: () => matMul(a, dy, true, false)\n      };\n    } else if (!transposeA && transposeB) {\n      return {\n        a: () => matMul(dy, b, false, false),\n        b: () => matMul(dy, a, true, false)\n      };\n    } else if (transposeA && !transposeB) {\n      return {\n        a: () => matMul(b, dy, false, true),\n        b: () => matMul(a, dy, false, false)\n      };\n    } else {\n      return {\n        a: () => matMul(b, dy, true, true),\n        b: () => matMul(dy, a, true, true)\n      };\n    }\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchToSpaceND, BatchToSpaceNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {spaceToBatchND} from '../ops/space_to_batch_nd';\nimport {Tensor} from '../tensor';\n\nexport const batchToSpaceNDGradConfig: GradConfig = {\n  kernelName: BatchToSpaceND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, crops} = attrs as {} as BatchToSpaceNDAttrs;\n    return {x: () => spaceToBatchND(dy, blockShape, crops)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BroadcastTo, BroadCastToAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const broadcastToGradConfig: GradConfig = {\n  kernelName: BroadcastTo,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const broadCastToAttrs: BroadCastToAttrs =\n        attrs as unknown as BroadCastToAttrs;\n\n    const inputShape = broadCastToAttrs.inputShape;\n    const outputShape = broadCastToAttrs.shape;\n\n    const reps: number[] = Array.from(outputShape);\n    for (let i = inputShape.length - 1; i >= 0; i--) {\n      if (inputShape[i] === outputShape[i]) {\n        reps[i] = 1;\n      } else if (inputShape[i] !== 1) {\n        throw new Error(`broadcastTo(): [${\n            inputShape}] cannot be broadcast to [${outputShape}].`);\n      }\n    }\n    const axes: number[] = [];\n    for (let i = 0; i < reps.length; i++) {\n      if (reps[i] > 1) {\n        axes.push(i);\n      }\n    }\n\n    return {x: () => sum(dy, axes, true /* keepDims */)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cast} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const castGradConfig: GradConfig = {\n  kernelName: Cast,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => dy.clone()};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const ceilGradConfig: GradConfig = {\n  kernelName: Ceil,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {lessEqual} from '../ops/less_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const clipByValueGradConfig: GradConfig = {\n  kernelName: ClipByValue,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {clipValueMin, clipValueMax} = attrs as {} as ClipByValueAttrs;\n    return {\n      x: () => where(\n          logicalAnd(greaterEqual(x, clipValueMin), lessEqual(x, clipValueMax)),\n          dy, zerosLike(dy)),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Concat, ConcatAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {split} from '../ops/split';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const concatGradConfig: GradConfig = {\n  kernelName: Concat,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const shapes = saved.map(t => t.shape);\n    const {axis} = attrs as {} as ConcatAttrs;\n    const $axis = parseAxisParam(axis, saved[0].shape)[0];\n    const sizeSplits = shapes.map(s => s[$axis]);\n    const derTensors = split(dy, sizeSplits, $axis);\n    return derTensors.map(t => () => t) as {};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2D, Conv2DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../ops/conv2d_backprop_input';\nimport * as conv_util from '../ops/conv_util';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv2DGradConfig: GradConfig = {\n  kernelName: Conv2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x4D, $filter] = saved as [Tensor4D, Tensor4D];\n    const {dilations, strides, pad, dataFormat} = attrs as {} as Conv2DAttrs;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of conv2D: dilation rates greater than 1 ' +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    return {\n      x: () =>\n          conv2DBackpropInput(x4D.shape, dy, $filter, strides, pad, dataFormat),\n      filter: () =>\n          conv2DBackpropFilter(x4D, dy, $filter.shape, strides, pad, dataFormat)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2d} from '../ops/conv2d';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const conv2DBackpropInputGradConfig: GradConfig = {\n  kernelName: Conv2DBackpropInput,\n  inputsToSave: ['dy', 'filter'],\n  gradFunc: (ddx: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [dy, filter] = saved as [Tensor4D, Tensor4D];\n\n    const {strides, pad, dataFormat, dimRoundingMode} =\n        attrs as {} as Conv2DBackpropInputAttrs;\n\n    return {\n      dy: () => conv2d(\n          ddx, filter, strides, pad, dataFormat, 1 /* dilations */,\n          dimRoundingMode),\n      filter: () => conv2DBackpropFilter(\n          ddx, dy, filter.shape, strides, pad, dataFormat, dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropFilterAttrs, Conv3DBackpropFilterInputs, Conv3DBackpropFilterV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 3D convolution.\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     [batch, depth, height, width, inChannels]. If rank 4, batch of 1 is\n *     assumed.\n * @param dy The dy image, of rank 5 or rank 4, of shape\n *     [batch, depth, height, width, outDepth]. If rank 4, batch of 1 is\n *     assumed.\n * @param filterShape The shape of the filter, length 5,\n *     [filterDepth, filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideDepth, strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction conv3DBackpropFilter_<T extends Tensor4D|Tensor5D>(\n    x: T, dy: T, filterShape: [number, number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): Tensor5D {\n  let x5D = x as Tensor5D;\n  if (x.rank === 4) {\n    x5D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2], x.shape[3]]);\n  }\n  let dy5D = dy as Tensor5D;\n  if (dy5D.rank === 4) {\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3dDerFilter: input must be rank 5, but got shape ` +\n          `${x5D.shape}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerFilter: dy must be rank 5, but got shape ` +\n          `${dy5D.shape}.`);\n  util.assert(\n      filterShape.length === 5,\n      () => `Error in conv3dDerFilter: filterShape must be length 5, but got ` +\n          `${filterShape}.`);\n  util.assert(\n      x5D.shape[4] === filterShape[3],\n      () => `Error in conv3dDerFilter: depth of input ${x5D.shape[4]}) must ` +\n          `match input depth in filter (${filterShape[3]}.`);\n  util.assert(\n      dy5D.shape[4] === filterShape[4],\n      () => `Error in conv3dDerFilter: depth of dy (${dy5D.shape[4]}) must ` +\n          `match output depth for filter (${filterShape[4]}).`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, filterShape, strides, dilations, pad);\n\n    return backend.conv3dDerFilter(x5D, dy5D, convInfo);\n  };\n\n  const inputs: Conv3DBackpropFilterInputs = {x: x5D, y: dy5D};\n\n  const attrs: Conv3DBackpropFilterAttrs = {strides, pad};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv3DBackpropFilterV2, attrs as {} as NamedAttrMap) as Tensor5D;\n}\n\nexport const conv3DBackpropFilter = op({conv3DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv3D, Conv3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv3DBackpropFilter} from '../ops/conv3d_backprop_filter';\nimport {conv3DBackpropInput} from '../ops/conv3d_backprop_input';\nimport {tupleValuesAreOne} from '../ops/conv_util';\nimport {Tensor, Tensor5D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv3DGradConfig: GradConfig = {\n  kernelName: Conv3D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor5D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad} = attrs as {} as Conv3DAttrs;\n    util.assert(\n        tupleValuesAreOne(dilations),\n        () =>\n            'Error in gradient of conv3D: dilation rates greater than 1 are ' +\n            `not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const [x5D, $filter] = saved;\n\n    return {\n      x: () => conv3DBackpropInput(\n          (x5D as Tensor5D).shape, dy, $filter as Tensor5D, strides, pad),\n      filter: () => conv3DBackpropFilter(\n          x5D as Tensor5D, dy, ($filter as Tensor5D).shape, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {sin} from '../ops/sin';\nimport {Tensor} from '../tensor';\n\nexport const cosGradConfig: GradConfig = {\n  kernelName: Cos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(neg(sin(cast(x, 'float32'))), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {sinh} from '../ops/sinh';\nimport {Tensor} from '../tensor';\n\nexport const coshGradConfig: GradConfig = {\n  kernelName: Cosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(sinh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cumsum, CumsumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getAxesPermutation} from '../ops/axis_util';\nimport {cumsum} from '../ops/cumsum';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const cumsumGradConfig: GradConfig = {\n  kernelName: Cumsum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {axis, exclusive, reverse}: CumsumAttrs = attrs as {} as CumsumAttrs;\n\n    return {\n      x: () => {\n        const permutation = getAxesPermutation([axis], x.rank);\n\n        let out = cumsum(dy, axis, exclusive, !reverse);\n\n        if (permutation != null) {\n          out = transpose(out, permutation);\n        }\n\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as conv_util from '../ops/conv_util';\nimport {depthwiseConv2dNativeBackpropFilter} from '../ops/depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../ops/depthwise_conv2d_native_backprop_input';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const depthwiseConv2dNativeGradConfig: GradConfig = {\n  kernelName: DepthwiseConv2dNative,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad, dimRoundingMode} =\n        attrs as {} as DepthwiseConv2dNativeAttrs;\n\n    const $dilations = dilations == null ? [1, 1] : dilations;\n\n    util.assert(\n        conv_util.tupleValuesAreOne($dilations),\n        () => 'Error in gradient of depthwiseConv2dNative: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${$dilations}'`);\n\n    const [x, filter] = saved as [Tensor4D, Tensor4D];\n\n    util.assert(\n        x.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: input must be ` +\n            `rank 4, but got rank ${x.rank}.`);\n    util.assert(\n        filter.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: filter must be ` +\n            `rank 4, but got rank ${filter.rank}.`);\n    util.assert(\n        x.shape[3] === filter.shape[2],\n        () => `Error in gradient of depthwiseConv2d: number of input ` +\n            `channels (${x.shape[3]}) must match the inChannels dimension ` +\n            `in filter ${filter.shape[2]}.`);\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, $dilations),\n        () => 'Error in gradient of depthwiseConv2d: Either strides or ' +\n            `dilations must be  1. Got strides ${strides} and dilations ` +\n            `'${$dilations}'.`);\n\n    if (dimRoundingMode != null) {\n      util.assert(\n          util.isInt(pad as number),\n          () =>\n              `Error in depthwiseConv2d: pad must be an integer when using, ` +\n              `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n    }\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x.shape, filter.shape, strides, $dilations as number | [number, number],\n        pad, dimRoundingMode, true /* depthwise */);\n\n    return {\n      x: () =>\n          depthwiseConv2dNativeBackpropInput(x.shape, dy, filter, convInfo),\n      filter: () =>\n          depthwiseConv2dNativeBackpropFilter(x, dy, filter.shape, convInfo),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DBackpropFilter, Dilation2DBackpropFilterInputs, Dilation2DBackpropInput, Dilation2DBackpropInputInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const dilation2dGradConfig: GradConfig = {\n  kernelName: Dilation2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, filter] = saved as [Tensor4D, Tensor3D];\n\n    const inputInputs: Dilation2DBackpropInputInputs = {x, filter, dy};\n    const filterInputs: Dilation2DBackpropFilterInputs = {x, filter, dy};\n\n    return {\n      x: () => ENGINE.runKernel(\n                   Dilation2DBackpropInput, inputInputs as {} as NamedTensorMap,\n                   attrs) as Tensor,\n      filter: () => ENGINE.runKernel(\n                        Dilation2DBackpropFilter,\n                        filterInputs as {} as NamedTensorMap, attrs) as Tensor\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const divGradConfig: GradConfig = {\n  kernelName: Div,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluGrad, EluGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const eluGradConfig: GradConfig = {\n  kernelName: Elu,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      return backend.eluDer(dy, y);\n    };\n\n    const inputs: EluGradInputs = {dy, y};\n\n    return {\n      x: () => ENGINE.runKernelFunc(\n          backPropKernelFunc, inputs as {} as NamedTensorMap, null /* grad */,\n          EluGrad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Erf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const erfGradConfig: GradConfig = {\n  kernelName: Erf,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const a = mul(exp(neg(square(x))), 2 / Math.sqrt(Math.PI));\n    return {x: () => mul(dy, a)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expGradConfig: GradConfig = {\n  kernelName: Exp,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n    return {x: () => mul(dy, y)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expm1GradConfig: GradConfig = {\n  kernelName: Expm1,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, exp(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const floorGradConfig: GradConfig = {\n  kernelName: Floor,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FloorDiv} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const floorDivGradConfig: GradConfig = {\n  kernelName: FloorDiv,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {FusedBatchNorm, FusedBatchNormAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {rsqrt} from '../ops/rsqrt';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {tile} from '../ops/tile';\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nexport const fusedBatchNormGradConfig: GradConfig = {\n  kernelName: FusedBatchNorm,\n  inputsToSave: ['x', 'mean', 'variance', 'scale'],\n  gradFunc: <R extends Rank>(\n      dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {varianceEpsilon} = attrs as {} as FusedBatchNormAttrs;\n    const [x, mean, variance, scale] = saved;\n\n    const scaleValue = scale == null ? scalar(1) : scale;\n    const reductionAxes = getReductionAxes(mean.shape, x.shape);\n    const tileShape: number[] = [];\n    if (mean.rank === 1) {\n      for (let i = 0; i < x.shape.length - 1; ++i) {\n        tileShape.push(x.shape[i]);\n      }\n      tileShape.push(1);\n    }\n\n    const xMinusMean = sub(x, mean);\n    const dyTimesScaleValue = mul(dy, scaleValue);\n    const oneOverSqrtVariance = rsqrt(add(variance, scalar(varianceEpsilon)));\n    const minusHalfRCube = mul(\n        mul(mul(oneOverSqrtVariance, oneOverSqrtVariance), oneOverSqrtVariance),\n        scalar(-0.5));\n\n    const derX = () => {\n      if (mean.rank === 1) {\n        return reshape(\n            mul(mul(dy,\n                    tile(\n                        reshape(oneOverSqrtVariance, [1, 1, 1, mean.shape[0]]),\n                        tileShape)),\n                scaleValue),\n            x.shape);\n      } else {\n        return reshape(mul(mul(dy, oneOverSqrtVariance), scaleValue), x.shape);\n      }\n    };\n    const derMean = () => {\n      let meanDer =\n          mul(mul(oneOverSqrtVariance, scalar(-1)), dyTimesScaleValue);\n      if (mean.rank === 1) {\n        meanDer = sum(meanDer, reductionAxes);\n      }\n      return reshape(meanDer, mean.shape as ShapeMap[R]);\n    };\n    const derVariance = () => {\n      let varianceDer = mul(mul(minusHalfRCube, xMinusMean), dyTimesScaleValue);\n\n      if (mean.rank === 1) {\n        varianceDer = sum(varianceDer, reductionAxes);\n      }\n      return reshape(varianceDer, mean.shape as ShapeMap[R]);\n    };\n    const derScale = () => {\n      const xMinusMean2TimesRsqrt = mul(xMinusMean, oneOverSqrtVariance);\n\n      let scaleDer = mul(dy, xMinusMean2TimesRsqrt);\n      if (mean.rank === 1) {\n        scaleDer = sum(scaleDer, reductionAxes);\n      }\n      return reshape(scaleDer, mean.shape as ShapeMap[R]);\n    };\n    const derOffset = () => {\n      let offsetDer = dy;\n      if (mean.rank === 1) {\n        offsetDer = sum(offsetDer, reductionAxes);\n      }\n      return reshape(offsetDer, mean.shape as ShapeMap[R]);\n    };\n\n    return {\n      x: derX,\n      mean: derMean,\n      variance: derVariance,\n      scale: derScale,\n      offset: derOffset\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GatherV2, GatherV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getUndoAxesPermutation} from '../ops/axis_util';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {unsortedSegmentSum} from '../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const gatherGradConfig: GradConfig = {\n  kernelName: GatherV2,\n  inputsToSave: ['x', 'indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, indices] = saved;\n    const {axis} = attrs as {} as GatherV2Attrs;\n\n    const parsedAxis = parseAxisParam(axis, x.shape)[0];\n\n    const derX = () => {\n      const paramsShape = x.shape;\n      const indicesSize = indices.size;\n\n      const outerShape = paramsShape.slice(0, parsedAxis);\n      const outerDims = outerShape.length;\n      const innerShape = paramsShape.slice(axis, paramsShape.length).slice(1);\n      const innerDims = innerShape.length;\n\n      const outerAxesIndices = arrayRange(0, outerDims);\n      const innerAxesIndices =\n          arrayRange(outerDims + 1, outerDims + 1 + innerDims);\n\n      const valuesShape = arrayConcat([outerShape, [indicesSize], innerShape]);\n\n      const values = reshape(dy, valuesShape);\n      const reshapedIndices = reshape(indices, [indicesSize]);\n\n      const transposeDims =\n          arrayConcat([[outerDims], outerAxesIndices, innerAxesIndices]);\n      const valuesTranspose = transpose(values, transposeDims);\n      let paramsGrad = unsortedSegmentSum(\n          valuesTranspose, reshapedIndices as Tensor1D, x.shape[parsedAxis]);\n\n      const invertTransposeDims = getUndoAxesPermutation(transposeDims);\n      paramsGrad = transpose(paramsGrad, invertTransposeDims);\n\n      return paramsGrad;\n    };\n    return {x: derX, indices: () => indices};\n  }\n};\n\nfunction arrayRange(start: number, stop: number): number[] {\n  const result = [];\n  for (let i = start; i < stop; ++i) {\n    result.push(i);\n  }\n  return result;\n}\n\nfunction arrayConcat(arrays: number[][]): number[] {\n  const result = [];\n  for (let i = 0; i < arrays.length; ++i) {\n    for (let j = 0; j < arrays[i].length; ++j) {\n      result.push(arrays[i][j]);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GreaterEqual} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const greaterEqualGradConfig: GradConfig = {\n  kernelName: GreaterEqual,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    return {a: () => zerosLike(a), b: () => zerosLike(b)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {Tensor} from '../tensor';\n\nexport const identityGradConfig: GradConfig = {\n  kernelName: Identity,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => cast(dy, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isFiniteGradConfig: GradConfig = {\n  kernelName: IsFinite,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isInfGradConfig: GradConfig = {\n  kernelName: IsInf,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isNanGradConfig: GradConfig = {\n  kernelName: IsNan,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log1p} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const log1pGradConfig: GradConfig = {\n  kernelName: Log1p,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, add(x, 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const logGradConfig: GradConfig = {\n  kernelName: Log,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, cast(x, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {LogSoftmax, LogSoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const logSoftmaxGradConfig: GradConfig = {\n  kernelName: LogSoftmax,\n  inputsToSave: [],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [value] = saved;\n    const {axis} = attrs as {} as LogSoftmaxAttrs;\n    return {\n      logits: () => {\n        const keepDims = true;\n        const softmax = exp(value);\n        return sub(dy, mul(sum(dy, axis, keepDims), softmax));\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRNBackprop, LRNBackpropAttrs, LRNBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport {op} from './operation';\n\nfunction localResponseNormalizationBackprop_<T extends Tensor4D>(\n    x: T, y: T, dy: T, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.LRNGrad(dy, x, y, depthRadius, bias, alpha, beta);\n\n  const inputs: LRNBackpropInputs = {x, y, dy};\n\n  const attrs: LRNBackpropAttrs = {depthRadius, bias, alpha, beta};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LRNBackprop, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const localResponseNormalizationBackprop =\n    op({localResponseNormalizationBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {LRN, LRNAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {localResponseNormalizationBackprop} from '../ops/local_response_normalization_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const lrnGradConfig: GradConfig = {\n  kernelName: LRN,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {depthRadius, bias, alpha, beta} = attrs as {} as LRNAttrs;\n\n    return {\n      x: () => localResponseNormalizationBackprop(\n          x, y, dy, depthRadius, bias, alpha, beta)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as axis_util from '../ops/axis_util';\nimport {cast} from '../ops/cast';\nimport {equal} from '../ops/equal';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\n/**\n * Gradient helper function for the min and max operations.\n */\nexport function gradForMinAndMax<T extends Tensor>(\n    dy: T, y: T, xOrig: Tensor, origAxes: number[], permutedAxes: number[]) {\n  if (y.rank < xOrig.rank) {\n    y = reshape(y, axis_util.expandShapeToKeepDim(y.shape, origAxes)) as T;\n  }\n  if (dy.rank < xOrig.rank) {\n    dy = reshape(dy, axis_util.expandShapeToKeepDim(dy.shape, origAxes)) as T;\n  }\n  return {\n    x: () => {\n      const dx = mul(dy, cast(equal(xOrig, y), dy.dtype));\n      return permutedAxes == null ? dx : transpose(dx, permutedAxes);\n    }\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const maxGradConfig: GradConfig = {\n  kernelName: Max,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const maxAttrs: MaxAttrs = attrs as {} as MaxAttrs;\n    const {reductionIndices} = maxAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const maxGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = maxGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Maximum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {less} from '../ops/less';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const maximumGradConfig: GradConfig = {\n  kernelName: Maximum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(greaterEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(less(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool3DBackprop, MaxPool3DBackpropAttrs, MaxPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d max pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank 4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param output The original output image, of rank 5 of shape\n *     [batchSize, outDepth, outHeight, outWidth, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike, output: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'maxPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPool3dBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let output5D = $output as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n    output5D = reshape($output, [\n      1, $output.shape[0], $output.shape[1], $output.shape[2], $output.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in maxPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in maxPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n  util.assert(\n      output5D.rank === 5,\n      () => `Error in maxPool3dBackprop: output must be rank 5 but got rank ` +\n          `${output5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.maxPool3dBackprop(dy5D, input5D, output5D, convInfo);\n  };\n\n  const inputs:\n      MaxPool3DBackpropInputs = {dy: dy5D, input: input5D, output: output5D};\n\n  const attrs: MaxPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      MaxPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3dBackprop = op({maxPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool3D, MaxPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPool3dBackprop} from '../ops/max_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const maxPool3DGradConfig: GradConfig = {\n  kernelName: MaxPool3D,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor5D, Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as MaxPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => maxPool3dBackprop(\n          dy as Tensor5D, x, y, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\n\n/**\n * Computes the backprop of a 2D max pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The original input image, of rank 4, of shape\n *     [batchSize, height, width, channels].\n * @param output The original output image, of rank 4, of shape\n *     [batchSize, outHeight, outWidth, channels].\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPoolBackprop_(\n    dy: Tensor4D|TensorLike, input: Tensor4D|TensorLike,\n    output: Tensor4D|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  const $dy = convertToTensor(dy, 'dy', 'maxPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPoolBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy ` +\n          `(${$dy.rank})`);\n\n  util.assert(\n      $dy.rank === 4,\n      () => `Error in maxPoolBackprop: dy must be rank 4 but got rank ` +\n          `${$dy.rank}.`);\n  util.assert(\n      $input.rank === 4,\n      () => `Error in maxPoolBackprop: input must be rank 4 but got rank ` +\n          `${$input.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPoolBackprop: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        $input.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    return backend.maxPoolBackprop($dy, $input, $output, convInfo);\n  };\n\n  const inputs:\n      MaxPoolBackpropInputs = {dy: $dy, input: $input, output: $output};\n\n  const attrs:\n      MaxPoolBackpropAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, MaxPoolBackprop,\n             attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const maxPoolBackprop = op({maxPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool, MaxPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPoolBackprop} from '../ops/max_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const maxPoolGradConfig: GradConfig = {\n  kernelName: MaxPool,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as MaxPoolAttrs;\n\n    return {\n      x: () => maxPoolBackprop(dy as Tensor4D, x, y, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Min, MinAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const minGradConfig: GradConfig = {\n  kernelName: Min,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const minAttrs: MinAttrs = attrs as {} as MinAttrs;\n    const {axis} = minAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(axis, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const minGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = minGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Minimum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const minimumGradConfig: GradConfig = {\n  kernelName: Minimum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(lessEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(greater(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Mod} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {floor} from '../ops/floor';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const modGradConfig: GradConfig = {\n  kernelName: Mod,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(dy, reduceAxes), a.shape);\n      }\n      return dy;\n    };\n    const derB = () => {\n      const res = mul(dy, neg(floor(div(a, b))));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Multiply} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const multiplyGradConfig: GradConfig = {\n  kernelName: Multiply,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = mul(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      const res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Negate} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {neg} from '../ops/neg';\nimport {Tensor} from '../tensor';\n\nexport const negateGradConfig: GradConfig = {\n  kernelName: Negate,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => neg(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OneHot} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\n\nexport const oneHotGradConfig: GradConfig = {\n  kernelName: OneHot,\n  inputsToSave: ['indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const indices = saved[0];\n    return {indices: () => zeros(indices.shape, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OnesLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const onesLikeGradConfig: GradConfig = {\n  kernelName: OnesLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PadV2, PadV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\nexport const padV2GradConfig: GradConfig = {\n  kernelName: PadV2,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    // Pad introduces values around the original tensor, so the gradient\n    // slices the original shape out of the gradient.\n    const x = saved[0];\n    const {paddings} = attrs as unknown as PadV2Attrs;\n    const begin = paddings.map(p => p[0]);\n    return {x: () => slice(dy, begin, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Pow} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {log} from '../ops/log';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {reshape} from '../ops/reshape';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const powGradConfig: GradConfig = {\n  kernelName: Pow,\n  inputsToSave: ['a', 'b'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b, y] = saved;\n    const base = a;\n    const exp = b;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(base.shape, exp.shape);\n\n    const derBase = () => {\n      const expFloat = cast(exp, 'float32');\n      let res = mul(dy, mul(expFloat, pow(base, sub(expFloat, scalar(1)))));\n      const reduceAxes = broadcast_util.getReductionAxes(base.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, base.shape);\n    };\n    const derExp = () => {\n      const condition = greater(base, 0);\n      const logBase = where(condition, log(base), zerosLike(base));\n      let res = mul(dy, mul(y, logBase));\n      const reduceAxes = broadcast_util.getReductionAxes(exp.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, exp.shape);\n    };\n    return {a: derBase, b: derExp};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Prelu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const preluGradConfig: GradConfig = {\n  kernelName: Prelu,\n  inputsToSave: ['x', 'alpha'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x, alpha] = saved;\n    const mask = greater(x, 0);\n\n    return {\n      x: () => where(mask, dy, mul(dy, alpha)),\n      alpha: () => {\n        let res = where(mask, zerosLike(dy), mul(dy, x));\n        const reduceAxes = getReductionAxes(alpha.shape, dy.shape);\n        if (reduceAxes.length > 0) {\n          res = sum(res, reduceAxes);\n        }\n        return reshape(res, alpha.shape);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Reciprocal} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const reciprocalGradConfig: GradConfig = {\n  kernelName: Reciprocal,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, neg(square(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu6} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const relu6GradConfig: GradConfig = {\n  kernelName: Relu6,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const mask = mul(lessEqual(x, 6), step(x));\n\n    return {x: () => mul(dy, cast(mask, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const reluGradConfig: GradConfig = {\n  kernelName: Relu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, cast(step(x), 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reshape} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\n\nexport const reshapeGradConfig: GradConfig = {\n  kernelName: Reshape,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => reshape(dy, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearGrad, ResizeBilinearGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeBilinearGradConfig: GradConfig = {\n  kernelName: ResizeBilinear,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeBilinearAttrs;\n      return backend.resizeBilinearBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeBilinearGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeBilinearGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborGrad, ResizeNearestNeighborGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeNearestNeighborGradConfig: GradConfig = {\n  kernelName: ResizeNearestNeighbor,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeNearestNeighborAttrs;\n      return backend.resizeNearestNeighborBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeNearestNeighborGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeNearestNeighborGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reverse, ReverseAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {reverse} from '../ops/reverse';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const reverseGradConfig: GradConfig = {\n  kernelName: Reverse,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dims} = attrs as {} as ReverseAttrs;\n    const axes = parseAxisParam(dims, dy.shape);\n    return {x: () => reverse(dy, axes)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Round} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const roundGradConfig: GradConfig = {\n  kernelName: Round,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rsqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {pow} from '../ops/pow';\nimport {Tensor} from '../tensor';\n\nexport const rsqrtGradConfig: GradConfig = {\n  kernelName: Rsqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => neg(div(dy, mul(pow(x, 1.5), 2)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SelectV2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {logicalNot} from '../ops/logical_not';\nimport {mul} from '../ops/mul';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const selectV2PoolGradConfig: GradConfig = {\n  kernelName: SelectV2,\n  inputsToSave: ['condition'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [condition] = saved;\n    return {\n      // TODO(julianoks): Return null for condition gradient\n      // when backprop supports it.\n      condition: () => cast(zerosLike(condition), 'float32'),\n      t: () => mul(dy, cast(condition, dy.dtype)),\n      e: () => mul(dy, cast(logicalNot(condition), dy.dtype))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Selu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {exp} from '../ops/exp';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {SELU_SCALE, SELU_SCALEALPHA} from '../ops/selu_util';\nimport {where} from '../ops/where';\nimport {Tensor} from '../tensor';\n\nexport const seluGradConfig: GradConfig = {\n  kernelName: Selu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {\n      x: () => {\n        const mask = greater(x, scalar(0));\n\n        const scaleAlpha = scalar(SELU_SCALEALPHA);\n        const scale = scalar(SELU_SCALE);\n\n        const greaterThanZeroDer = mul(dy, scale);\n        const lessEqualZeroDer =\n            mul(mul(dy, scaleAlpha), exp(cast(x, 'float32')));\n\n        return where(mask, greaterThanZeroDer, lessEqualZeroDer);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sigmoid} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const sigmoidGradConfig: GradConfig = {\n  kernelName: Sigmoid,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(dy, mul(y, sub(scalar(1), y)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sign} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const signGradConfig: GradConfig = {\n  kernelName: Sign,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cos} from '../ops/cos';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinGradConfig: GradConfig = {\n  kernelName: Sin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(cos(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cosh} from '../ops/cosh';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinhGradConfig: GradConfig = {\n  kernelName: Sinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(cosh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Slice, SliceAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {pad} from '../ops/pad';\nimport {parseSliceParams} from '../ops/slice_util';\nimport {Tensor} from '../tensor';\n\nexport const sliceGradConfig: GradConfig = {\n  kernelName: Slice,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {begin, size} = attrs as {} as SliceAttrs;\n\n    const inputShape = x.shape;\n    const [begin_, size_] = parseSliceParams(x, begin, size);\n\n    // Create an Nx2 padding where the first column represents how many\n    // zeros are prepended (at start) for each dimension, and the second\n    // column indicates how many zeros are appended (at end).\n\n    // The number of zeros to append is the shape of the input\n    // elementwise-subtracted by both the begin vector and sizes vector.\n    const paddings: Array<[number, number]> = [];\n    for (let i = 0; i < dy.rank; i++) {\n      paddings.push([begin_[i], inputShape[i] - begin_[i] - size_[i]]);\n    }\n    return {x: () => pad(dy, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softmax, SoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const softmaxGradConfig: GradConfig = {\n  kernelName: Softmax,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [y] = saved;\n    const {dim} = attrs as {} as SoftmaxAttrs;\n    const keepDims = true;\n\n    const dyTimesY = mul(dy, y);\n    return {\n      logits: () => sub(dyTimesY, mul(sum(dyTimesY, [dim], keepDims), y))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softplus} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sigmoid} from '../ops/sigmoid';\nimport {Tensor} from '../tensor';\n\nexport const softplusGradConfig: GradConfig = {\n  kernelName: Softplus,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(dy, sigmoid(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SpaceToBatchND, SpaceToBatchNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {batchToSpaceND} from '../ops/batch_to_space_nd';\nimport {Tensor} from '../tensor';\n\nexport const spaceToBatchNDGradConfig: GradConfig = {\n  kernelName: SpaceToBatchND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, paddings} = attrs as {} as SpaceToBatchNDAttrs;\n    return {x: () => batchToSpaceND(dy, blockShape, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {SplitV, SplitVAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {concat} from '../ops/concat';\nimport {Tensor} from '../tensor';\n\nexport const splitVGradConfig: GradConfig = {\n  kernelName: SplitV,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const {axis} = attrs as {} as SplitVAttrs;\n\n    return {x: () => concat(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {Tensor} from '../tensor';\n\nexport const sqrtGradConfig: GradConfig = {\n  kernelName: Sqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, mul(sqrt(cast(x, 'float32')), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const squareGradConfig: GradConfig = {\n  kernelName: Square,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, mul(cast(x, 'float32'), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SquaredDifference} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const squaredDifferenceGradConfig: GradConfig = {\n  kernelName: SquaredDifference,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const two = scalar(2);\n    const derA = () => mul(dy, mul(two, sub(a, b)));\n    const derB = () => mul(dy, mul(two, sub(b, a)));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Step} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const stepGradConfig: GradConfig = {\n  kernelName: Step,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports\n    // it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Sub} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const subGradConfig: GradConfig = {\n  kernelName: Sub,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(neg(res), b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sum, SumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {ones} from '../ops/ones';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const sumGradConfig: GradConfig = {\n  kernelName: Sum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const expandedDyShape = x.shape.slice();\n    const {axis} = attrs as {} as SumAttrs;\n\n    const axes = parseAxisParam(axis, x.shape);\n    axes.forEach(axis => {\n      expandedDyShape[axis] = 1;\n    });\n    const expandedDy = reshape(dy, expandedDyShape);\n    const derX = mul(expandedDy, ones(x.shape, 'float32'));\n\n    return {x: () => derX};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cos} from '../ops/cos';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const tanGradConfig: GradConfig = {\n  kernelName: Tan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, square(cos(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const tanhGradConfig: GradConfig = {\n  kernelName: Tanh,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(sub(scalar(1), square(y)), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tile, TileAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {slice} from '../ops/slice';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const tileGradConfig: GradConfig = {\n  kernelName: Tile,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {reps} = attrs as unknown as TileAttrs;\n\n    const derX = () => {\n      let xGrad = zerosLike(x);\n      // TODO(cais): Maybe reduce memory footprint by avoiding repeated\n      // slicing.\n      if (x.rank === 1) {\n        for (let i = 0; i < reps[0]; ++i) {\n          xGrad = add(xGrad, slice(dy, [i * x.shape[0]], [x.shape[0]]));\n        }\n      } else if (x.rank === 2) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            xGrad = add(xGrad, slice(dy, [i * x.shape[0], j * x.shape[1]], [\n                          x.shape[0], x.shape[1]\n                        ]));\n          }\n        }\n      } else if (x.rank === 3) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              xGrad =\n                  add(xGrad,\n                      slice(\n                          dy, [i * x.shape[0], j * x.shape[1], k * x.shape[2]],\n                          [x.shape[0], x.shape[1], x.shape[2]]));\n            }\n          }\n        }\n      } else if (x.rank === 4) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              for (let l = 0; l < reps[3]; ++l) {\n                xGrad =\n                    add(xGrad,\n                        slice(\n                            dy,\n                            [\n                              i * x.shape[0], j * x.shape[1], k * x.shape[2],\n                              l * x.shape[3]\n                            ],\n                            [x.shape[0], x.shape[1], x.shape[2], x.shape[3]]));\n              }\n            }\n          }\n        }\n      } else {\n        throw new Error(\n            `Gradient for tile operation is not implemented for rank-` +\n            `${x.rank} tensors yet.`);\n      }\n      return xGrad;\n    };\n    return {x: derX};\n  },\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Transpose, TransposeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const transposeGradConfig: GradConfig = {\n  kernelName: Transpose,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const transposeAttrs: TransposeAttrs = attrs as {} as TransposeAttrs;\n    const {perm} = transposeAttrs;\n    const undoPerm = axis_util.getUndoAxesPermutation(perm);\n    return {x: () => transpose(dy, undoPerm)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Unpack, UnpackAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {stack} from '../ops/stack';\nimport {Tensor} from '../tensor';\n\nexport const unpackGradConfig: GradConfig = {\n  kernelName: Unpack,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const unpackAttrs: UnpackAttrs = attrs as {} as UnpackAttrs;\n    const {axis} = unpackAttrs;\n    return {value: () => stack(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {UnsortedSegmentSum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {expandDims} from '../ops/expand_dims';\nimport {gather} from '../ops/gather';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {maximum} from '../ops/maximum';\nimport {ones} from '../ops/ones';\nimport {scalar} from '../ops/scalar';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor, Tensor1D} from '../tensor';\n\nexport const unsortedSegmentSumGradConfig: GradConfig = {\n  kernelName: UnsortedSegmentSum,\n  inputsToSave: ['segmentIds'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [segmentIds] = saved;\n\n    const derX = () => {\n      return gatherDropNegatives(dy, segmentIds as Tensor1D);\n    };\n    return {x: derX};\n  }\n};\n\nfunction gatherDropNegatives<T extends Tensor>(x: T, indices: Tensor1D) {\n  // Helper function for unsorted segment ops. Gathers params for\n  // positive segment ids and gathers 0 for inputs with negative segment id.\n  // Mirrors _GatherDropNegatives from tensorflow/python/ops/math_grad.py\n  const zeroClippedIndices = maximum(indices, zerosLike(indices));\n  const gathered = gather(x, zeroClippedIndices as Tensor1D);\n  let isPositive = greaterEqual(indices, scalar(0, 'int32'));\n  const numIters = gathered.rank - isPositive.rank;\n  for (let i = 0; i < numIters; ++i) {\n    isPositive = expandDims(isPositive, i + 1);\n  }\n  isPositive = logicalAnd(isPositive, ones(gathered.shape, 'bool'));\n  const zeroSlice = zerosLike(gathered);\n  return where(isPositive, gathered, zeroSlice);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ZerosLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const zerosLikeGradConfig: GradConfig = {\n  kernelName: ZerosLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {absGradConfig} from './gradients/Abs_grad';\nimport {acosGradConfig} from './gradients/Acos_grad';\nimport {acoshGradConfig} from './gradients/Acosh_grad';\nimport {addGradConfig} from './gradients/Add_grad';\nimport {addNGradConfig} from './gradients/AddN_grad';\nimport {argMaxGradConfig} from './gradients/ArgMax_grad';\nimport {argMinGradConfig} from './gradients/ArgMin_grad';\nimport {asinGradConfig} from './gradients/Asin_grad';\nimport {asinhGradConfig} from './gradients/Asinh_grad';\nimport {atan2GradConfig} from './gradients/Atan2_grad';\nimport {atanGradConfig} from './gradients/Atan_grad';\nimport {atanhGradConfig} from './gradients/Atanh_grad';\nimport {avgPool3DGradConfig} from './gradients/AvgPool3D_grad';\nimport {avgPoolGradConfig} from './gradients/AvgPool_grad';\nimport {batchMatMulGradConfig} from './gradients/BatchMatMul_grad';\nimport {batchToSpaceNDGradConfig} from './gradients/BatchToSpaceND_grad';\nimport {broadcastToGradConfig} from './gradients/BroadcastTo_grad';\nimport {castGradConfig} from './gradients/Cast_grad';\nimport {ceilGradConfig} from './gradients/Ceil_grad';\nimport {clipByValueGradConfig} from './gradients/ClipByValue_grad';\nimport {concatGradConfig} from './gradients/Concat_grad';\nimport {conv2DGradConfig} from './gradients/Conv2D_grad';\nimport {conv2DBackpropInputGradConfig} from './gradients/Conv2DBackpropInput_grad';\nimport {conv3DGradConfig} from './gradients/Conv3D_grad';\nimport {cosGradConfig} from './gradients/Cos_grad';\nimport {coshGradConfig} from './gradients/Cosh_grad';\nimport {cumsumGradConfig} from './gradients/Cumsum_grad';\nimport {depthwiseConv2dNativeGradConfig} from './gradients/DepthwiseConv2dNative_grad';\nimport {dilation2dGradConfig} from './gradients/Dilation2D_grad';\nimport {divGradConfig} from './gradients/Div_grad';\nimport {eluGradConfig} from './gradients/Elu_grad';\nimport {erfGradConfig} from './gradients/Erf_grad';\nimport {expGradConfig} from './gradients/Exp_grad';\nimport {expm1GradConfig} from './gradients/Expm1_grad';\nimport {floorGradConfig} from './gradients/Floor_grad';\nimport {floorDivGradConfig} from './gradients/FloorDiv_grad';\nimport {fusedBatchNormGradConfig} from './gradients/FusedBatchNorm_grad';\nimport {gatherGradConfig} from './gradients/GatherV2_grad';\nimport {greaterEqualGradConfig} from './gradients/GreaterEqual_grad';\nimport {identityGradConfig} from './gradients/Identity_grad';\nimport {isFiniteGradConfig} from './gradients/IsFinite_grad';\nimport {isInfGradConfig} from './gradients/IsInf_grad';\nimport {isNanGradConfig} from './gradients/IsNan_grad';\nimport {log1pGradConfig} from './gradients/Log1p_grad';\nimport {logGradConfig} from './gradients/Log_grad';\nimport {logSoftmaxGradConfig} from './gradients/LogSoftmax_grad';\nimport {lrnGradConfig} from './gradients/LRN_grad';\nimport {maxGradConfig} from './gradients/Max_grad';\nimport {maximumGradConfig} from './gradients/Maximum_grad';\nimport {maxPool3DGradConfig} from './gradients/MaxPool3D_grad';\nimport {maxPoolGradConfig} from './gradients/MaxPool_grad';\nimport {minGradConfig} from './gradients/Min_grad';\nimport {minimumGradConfig} from './gradients/Minimum_grad';\nimport {modGradConfig} from './gradients/Mod_grad';\nimport {multiplyGradConfig} from './gradients/Multiply_grad';\nimport {negateGradConfig} from './gradients/Negate_grad';\nimport {oneHotGradConfig} from './gradients/OneHot_grad';\nimport {onesLikeGradConfig} from './gradients/OnesLike_grad';\nimport {padV2GradConfig} from './gradients/PadV2_grad';\nimport {powGradConfig} from './gradients/Pow_grad';\nimport {preluGradConfig} from './gradients/Prelu_grad';\nimport {reciprocalGradConfig} from './gradients/Reciprocal_grad';\nimport {relu6GradConfig} from './gradients/Relu6_grad';\nimport {reluGradConfig} from './gradients/Relu_grad';\nimport {reshapeGradConfig} from './gradients/Reshape_grad';\nimport {resizeBilinearGradConfig} from './gradients/ResizeBilinear_grad';\nimport {resizeNearestNeighborGradConfig} from './gradients/ResizeNearestNeighbor_grad';\nimport {reverseGradConfig} from './gradients/Reverse_grad';\nimport {roundGradConfig} from './gradients/Round_grad';\nimport {rsqrtGradConfig} from './gradients/Rsqrt_grad';\nimport {selectV2PoolGradConfig} from './gradients/SelectV2_grad';\nimport {seluGradConfig} from './gradients/Selu_grad';\nimport {sigmoidGradConfig} from './gradients/Sigmoid_grad';\nimport {signGradConfig} from './gradients/Sign_grad';\nimport {sinGradConfig} from './gradients/Sin_grad';\nimport {sinhGradConfig} from './gradients/Sinh_grad';\nimport {sliceGradConfig} from './gradients/Slice_grad';\nimport {softmaxGradConfig} from './gradients/Softmax_grad';\nimport {softplusGradConfig} from './gradients/Softplus_grad';\nimport {spaceToBatchNDGradConfig} from './gradients/SpaceToBatchND_grad';\nimport {splitVGradConfig} from './gradients/SplitV_grad';\nimport {sqrtGradConfig} from './gradients/Sqrt_grad';\nimport {squareGradConfig} from './gradients/Square_grad';\nimport {squaredDifferenceGradConfig} from './gradients/SquaredDifference_grad';\nimport {stepGradConfig} from './gradients/Step_grad';\nimport {subGradConfig} from './gradients/Sub_grad';\nimport {sumGradConfig} from './gradients/Sum_grad';\nimport {tanGradConfig} from './gradients/Tan_grad';\nimport {tanhGradConfig} from './gradients/Tanh_grad';\nimport {tileGradConfig} from './gradients/Tile_grad';\nimport {transposeGradConfig} from './gradients/Transpose_grad';\nimport {unpackGradConfig} from './gradients/Unpack_grad';\nimport {unsortedSegmentSumGradConfig} from './gradients/UnsortedSegmentSum_grad';\nimport {zerosLikeGradConfig} from './gradients/ZerosLike_grad';\nimport {GradConfig} from './kernel_registry';\nimport {registerGradient} from './kernel_registry';\n\n// Export all kernel configs here so that the package can auto register them\nconst gradConfigs: GradConfig[] = [\n  absGradConfig,\n  acosGradConfig,\n  acoshGradConfig,\n  addGradConfig,\n  addNGradConfig,\n  argMaxGradConfig,\n  argMinGradConfig,\n  asinGradConfig,\n  asinhGradConfig,\n  atan2GradConfig,\n  atanGradConfig,\n  atanhGradConfig,\n  avgPool3DGradConfig,\n  avgPoolGradConfig,\n  batchMatMulGradConfig,\n  batchToSpaceNDGradConfig,\n  broadcastToGradConfig,\n  castGradConfig,\n  ceilGradConfig,\n  clipByValueGradConfig,\n  concatGradConfig,\n  conv2DBackpropInputGradConfig,\n  conv2DGradConfig,\n  conv3DGradConfig,\n  cosGradConfig,\n  coshGradConfig,\n  cumsumGradConfig,\n  depthwiseConv2dNativeGradConfig,\n  dilation2dGradConfig,\n  divGradConfig,\n  eluGradConfig,\n  erfGradConfig,\n  expGradConfig,\n  expm1GradConfig,\n  floorDivGradConfig,\n  floorGradConfig,\n  fusedBatchNormGradConfig,\n  gatherGradConfig,\n  greaterEqualGradConfig,\n  identityGradConfig,\n  isFiniteGradConfig,\n  isInfGradConfig,\n  isNanGradConfig,\n  log1pGradConfig,\n  logGradConfig,\n  logSoftmaxGradConfig,\n  lrnGradConfig,\n  maxGradConfig,\n  maxGradConfig,\n  maximumGradConfig,\n  maxPool3DGradConfig,\n  maxPoolGradConfig,\n  minGradConfig,\n  minimumGradConfig,\n  modGradConfig,\n  multiplyGradConfig,\n  negateGradConfig,\n  oneHotGradConfig,\n  onesLikeGradConfig,\n  padV2GradConfig,\n  padV2GradConfig,\n  powGradConfig,\n  preluGradConfig,\n  reciprocalGradConfig,\n  relu6GradConfig,\n  reluGradConfig,\n  reshapeGradConfig,\n  resizeBilinearGradConfig,\n  resizeNearestNeighborGradConfig,\n  reverseGradConfig,\n  roundGradConfig,\n  rsqrtGradConfig,\n  selectV2PoolGradConfig,\n  seluGradConfig,\n  sigmoidGradConfig,\n  signGradConfig,\n  sinGradConfig,\n  sinhGradConfig,\n  sliceGradConfig,\n  softmaxGradConfig,\n  softplusGradConfig,\n  spaceToBatchNDGradConfig,\n  spaceToBatchNDGradConfig,\n  splitVGradConfig,\n  splitVGradConfig,\n  sqrtGradConfig,\n  squaredDifferenceGradConfig,\n  squareGradConfig,\n  stepGradConfig,\n  subGradConfig,\n  sumGradConfig,\n  tanGradConfig,\n  tanhGradConfig,\n  tileGradConfig,\n  transposeGradConfig,\n  unpackGradConfig,\n  unsortedSegmentSumGradConfig,\n  zerosLikeGradConfig\n];\n\nfor (const gradientConfig of gradConfigs) {\n  registerGradient(gradientConfig);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {abs} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    abs<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.abs = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return abs(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acos = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acosh = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {addStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    addStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.addStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return addStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {add} from '../../ops/add';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    add<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.add = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return add(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {all} from '../../ops/all';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    all<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.all = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return all(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {any} from '../../ops/any';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    any<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.any = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return any(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMax} from '../../ops/arg_max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMax<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMax = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return argMax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMin} from '../../ops/arg_min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMin<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMin = function<T extends Tensor>(axis: number): T {\n  this.throwIfDisposed();\n  return argMin(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Scalar, Tensor} from '../../tensor';\nimport {Rank} from '../../types';\nimport {assert} from '../../util';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asScalar<T extends Tensor>(): Scalar;\n  }\n}\n\n/** Converts a size-1 `tf.Tensor` to a `tf.Scalar`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asScalar = function<T extends Tensor>(this: T): Scalar {\n  this.throwIfDisposed();\n  assert(this.size === 1, () => 'The array must have only 1 element.');\n  return reshape(this, []);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asType<T extends Tensor>(this: T, dtype: DataType): T;\n  }\n}\n\n/**\n * Casts a `tf.Tensor` to a specified dtype.\n *\n * @param dtype Data-type to cast the tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asType = function<T extends Tensor>(\n    this: T, dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast<T>(this, dtype);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as1D<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Converts a `tf.Tensor` to a `tf.Tensor1D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as1D = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as2D<T extends Tensor>(rows: number, columns: number): Tensor2D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor2D`.\n *\n * @param rows Number of rows in `tf.Tensor2D`.\n * @param columns Number of columns in `tf.Tensor2D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as2D = function<T extends Tensor>(\n    rows: number, columns: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as3D<T extends Tensor>(rows: number, columns: number, depth: number):\n        Tensor3D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor3D`.\n *\n * @param rows Number of rows in `tf.Tensor3D`.\n * @param columns Number of columns in `tf.Tensor3D`.\n * @param depth Depth of `tf.Tensor3D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as3D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as4D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number): Tensor4D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor4D`.\n *\n * @param rows Number of rows in `tf.Tensor4D`.\n * @param columns Number of columns in `tf.Tensor4D`.\n * @param depth Depth of `tf.Tensor4D`.\n * @param depth2 4th dimension of `tf.Tensor4D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as4D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as5D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number,\n        depth3: number): Tensor5D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor5D`.\n *\n * @param rows Number of rows in `tf.Tensor5D`.\n * @param columns Number of columns in `tf.Tensor5D`.\n * @param depth Depth of `tf.Tensor5D`.\n * @param depth2 4th dimension of `tf.Tensor5D`.\n * @param depth3 5th dimension of 'tf.Tensor5D'\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as5D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number,\n    depth3: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2, depth3]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {atan2} from '../../ops/atan2';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan2<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.atan2 = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return atan2(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {avgPool} from '../../ops/avg_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    avgPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.avgPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return avgPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchToSpaceND} from '../../ops/batch_to_space_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchToSpaceND<R extends Rank>(blockShape: number[], crops: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.batchToSpaceND = function<R extends Rank>(\n    blockShape: number[], crops: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return batchToSpaceND(this, blockShape, crops);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchNorm} from '../../ops/batchnorm';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchNorm<T extends Tensor>(\n        mean: Tensor<R>|Tensor1D|TensorLike,\n        variance: Tensor<R>|Tensor1D|TensorLike,\n        offset?: Tensor<R>|Tensor1D|TensorLike,\n        scale?: Tensor<R>|Tensor1D|TensorLike,\n        varianceEpsilon?: number): Tensor<R>;\n  }\n}\n\nTensor.prototype.batchNorm = function<R extends Rank>(\n    mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  this.throwIfDisposed();\n  return batchNorm(this, mean, variance, offset, scale, varianceEpsilon);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {broadcastTo} from '../../ops/broadcast_to';\nimport {Tensor} from '../../tensor';\nimport {Rank, ShapeMap} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    broadcastTo<R extends Rank>(shape: ShapeMap[R]): Tensor<R>;\n  }\n}\n\nTensor.prototype.broadcastTo = function<R extends Rank>(shape: ShapeMap[R]):\n    Tensor<R> {\n  this.throwIfDisposed();\n  return broadcastTo(this, shape);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cast<T extends Tensor>(dtype: DataType): T;\n  }\n}\n\nTensor.prototype.cast = function<T extends Tensor>(dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast(this, dtype) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ceil} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ceil<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.ceil = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return ceil(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {clipByValue} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    clipByValue<T extends Tensor>(min: number, max: number): Tensor;\n  }\n}\n\nTensor.prototype.clipByValue = function<T extends Tensor>(\n    min: number, max: number): T {\n  this.throwIfDisposed();\n  return clipByValue(this, min, max) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {concat} from '../../ops/concat';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    concat<T extends Tensor>(tensors: T|Array<T|TensorLike>, axis?: number): T;\n  }\n}\n\nTensor.prototype.concat = function<T extends Tensor>(\n    x: T|Array<T|TensorLike>, axis?: number): T {\n  this.throwIfDisposed();\n  if (x instanceof Tensor) {\n    x = [x];\n  }\n  return concat([this, ...x], axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv1d} from '../../ops/conv1d';\nimport {ExplicitPadding} from '../../ops/conv_util';\nimport {Tensor, Tensor2D, Tensor3D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv1d<T extends Tensor2D|Tensor3D>(\n        filter: Tensor3D|TensorLike3D, stride: number,\n        pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n        dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv1d = function<T extends Tensor2D|Tensor3D>(\n    filter: Tensor3D|TensorLike3D, stride: number,\n    pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n    dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv1d(\n             this, filter, stride, pad, dataFormat, dilation,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2dTranspose} from '../../ops/conv2d_transpose';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2dTranspose<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D,\n        outputShape: [number, number, number, number]|[number, number, number],\n        strides: [number, number]|number, pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2dTranspose = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2dTranspose(\n             this, filter, outputShape, strides, pad, dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2d} from '../../ops/conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cos = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cosh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {cumsum} from '../../ops/cumsum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cumsum<R extends Rank>(\n        axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R>;\n  }\n}\n\nTensor.prototype.cumsum = function<R extends Rank>(\n    axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R> {\n  this.throwIfDisposed();\n  return cumsum(this, axis, exclusive, reverse);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthToSpace} from '../../ops/depth_to_space';\nimport {Tensor, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthToSpace<T extends Tensor4D>(\n        blockSize: number, dataFormat: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.depthToSpace = function<T extends Tensor4D>(\n    blockSize: number, dataFormat: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return depthToSpace(this, blockSize, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../../globals';\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2D<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\n/**\n * @deprecated Use `depthwiseConv2d` instead.\n */\nTensor.prototype.depthwiseConv2D = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  deprecationWarn('depthwiseConv2D is deprecated, use depthwiseConv2d instead');\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.depthwiseConv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dilation2d} from '../../ops/dilation2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dilation2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n        pad: 'valid'|'same', dilations?: [number, number]|number,\n        dataFormat?: 'NHWC'): T;\n  }\n}\n\nTensor.prototype.dilation2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n    pad: 'valid'|'same', dilations?: [number, number]|number,\n    dataFormat?: 'NHWC'): T {\n  this.throwIfDisposed();\n  return dilation2d(this, filter, strides, pad, dilations, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {divNoNan} from '../../ops/div_no_nan';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divNoNan<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.divNoNan = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return divNoNan(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {divStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.divStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return divStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {div} from '../../ops/div';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    div<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.div = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return div(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dot} from '../../ops/dot';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dot<T extends Tensor>(b: Tensor|TensorLike): Tensor;\n  }\n}\n\nTensor.prototype.dot = function<T extends Tensor>(b: T|TensorLike): Tensor {\n  this.throwIfDisposed();\n  return dot(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {elu} from '../../ops/elu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    elu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.elu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return elu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {equalStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equalStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.equalStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return equalStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {equal} from '../../ops/equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equal<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.equal = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return equal(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {erf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    erf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.erf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return erf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {exp} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    exp<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.exp = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return exp(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {expandDims} from '../../ops/expand_dims';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expandDims<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.expandDims = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return expandDims(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {expm1} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expm1<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.expm1 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return expm1(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {fft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    fft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.fft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return fft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    flatten<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Flatten a Tensor to a 1D array.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.flatten = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {floor} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floor<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.floor = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return floor(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {floorDiv} from '../../ops/floorDiv';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floorDiv<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.floorDiv = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return floorDiv(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {gather} from '../../ops/gather';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    gather<T extends Tensor>(\n        this: T, indices: Tensor|TensorLike, axis?: number): T;\n  }\n}\n\nTensor.prototype.gather = function<T extends Tensor>(\n    this: T, indices: Tensor|TensorLike, axis?: number): T {\n  this.throwIfDisposed();\n  return gather(this, indices, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return greaterEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greaterEqual} from '../../ops/greater_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greaterEqual = function<T extends Tensor>(b: Tensor|\n                                                           TensorLike): T {\n  this.throwIfDisposed();\n  return greaterEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return greaterStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greater} from '../../ops/greater';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greater<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greater = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return greater(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ifft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ifft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.ifft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return ifft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {irfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    irfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.irfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return irfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isFinite} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isFinite<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isFinite = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isFinite(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isInf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isInf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isInf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isInf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isNaN} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isNaN<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isNaN = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isNaN(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {leakyRelu} from '../../ops/leaky_relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    leakyRelu<T extends Tensor>(alpha: number): T;\n  }\n}\n\nTensor.prototype.leakyRelu = function<T extends Tensor>(\n    this: T, alpha: number): T {\n  this.throwIfDisposed();\n  return leakyRelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.lessEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {lessEqual} from '../../ops/less_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessEqual = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return lessEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {less} from '../../ops/less';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    less<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.less = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return less(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {localResponseNormalization} from '../../ops/local_response_normalization';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    localResponseNormalization<T extends Tensor>(\n        depthRadius?: number, bias?: number, alpha?: number, beta?: number): T;\n  }\n}\n\nTensor.prototype.localResponseNormalization = function<T extends Tensor>(\n    depthRadius?: number, bias?: number, alpha?: number, beta?: number): T {\n  this.throwIfDisposed();\n  return localResponseNormalization(this, depthRadius, bias, alpha, beta) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.logSigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return logSigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSoftmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSoftmax<T extends Tensor>(this: T, axis?: number): T;\n  }\n}\n\nTensor.prototype.logSoftmax = function<T extends Tensor>(\n    this: T, axis?: number): T {\n  this.throwIfDisposed();\n  return logSoftmax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logSumExp} from '../../ops/log_sum_exp';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSumExp<T extends Tensor>(\n        this: T, axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.logSumExp = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return logSumExp(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log1p} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log1p<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log1p = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log1p(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalAnd} from '../../ops/logical_and';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalAnd<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalAnd = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalAnd(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalNot} from '../../ops/logical_not';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalNot<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.logicalNot = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return logicalNot(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalOr} from '../../ops/logical_or';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalOr<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalOr = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return logicalOr(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalXor} from '../../ops/logical_xor';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalXor<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalXor = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalXor(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {matMul} from '../../ops/mat_mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    matMul<T extends Tensor>(\n        b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T;\n  }\n}\n\nTensor.prototype.matMul = function<T extends Tensor>(\n    this: T, b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T {\n  this.throwIfDisposed();\n  return matMul(this, b, transposeA, transposeB);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maxPool} from '../../ops/max_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maxPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.maxPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return maxPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {max} from '../../ops/max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    max<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.max = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return max(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {maximumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.maximumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return maximumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maximum} from '../../ops/maximum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.maximum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return maximum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mean} from '../../ops/mean';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mean<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.mean = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return mean(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {min} from '../../ops/min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    min<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.min = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return min(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {minimumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.minimumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return minimumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {minimum} from '../../ops/minimum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.minimum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return minimum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {modStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    modStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.modStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return modStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mod} from '../../ops/mod';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mod<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mod = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mod(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {mulStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mulStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.mulStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return mulStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mul} from '../../ops/mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mul<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mul = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mul(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {neg} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    neg<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.neg = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return neg(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {norm} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    norm<T extends Tensor>(\n        ord?: number|'euclidean'|'fro', axis?: number|number[],\n        keepDims?: boolean): Tensor;\n  }\n}\n\nTensor.prototype.norm = function<T extends Tensor>(\n    ord?: number|'euclidean'|'fro', axis?: number|number[],\n    keepDims?: boolean) {\n  this.throwIfDisposed();\n  return norm(this, ord, axis, keepDims) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {notEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.notEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return notEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {notEqual} from '../../ops/not_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.notEqual = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return notEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {oneHot} from '../../ops/one_hot';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    oneHot(depth: number, onValue: number, offValue: number): Tensor;\n  }\n}\n\nTensor.prototype.oneHot = function(\n    depth: number, onValue = 1, offValue = 0): Tensor {\n  this.throwIfDisposed();\n  return oneHot(this, depth, onValue, offValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {onesLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    onesLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.onesLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return onesLike(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {pad} from '../../ops/pad';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pad<T extends Tensor>(\n        paddings: Array<[number, number]>, constantValue?: number): T;\n  }\n}\n\nTensor.prototype.pad = function<T extends Tensor>(\n    this: T, paddings: Array<[number, number]>, constantValue: number): T {\n  this.throwIfDisposed();\n  return pad(this, paddings, constantValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pool} from '../../ops/pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pool<T extends Tensor3D|Tensor4D>(\n        windowShape: [number, number]|number, poolingType: 'avg'|'max',\n        padding: 'valid'|'same'|number, diationRate?: [number, number]|number,\n        strides?: [number, number]|number): T;\n  }\n}\n\nTensor.prototype.pool = function<T extends Tensor3D|Tensor4D>(\n    this: T, windowShape: [number, number]|number, poolingType: 'max'|'avg',\n    padding: 'valid'|'same'|number, dilationRate?: [number, number]|number,\n    strides?: [number, number]|number): T {\n  this.throwIfDisposed();\n  return pool(this, windowShape, poolingType, padding, dilationRate, strides);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {powStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    powStrict<T extends Tensor>(exp: Tensor): Tensor;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.powStrict = function<T extends Tensor>(exp: Tensor): T {\n  this.throwIfDisposed();\n  return powStrict(this, exp) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pow} from '../../ops/pow';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pow<T extends Tensor>(exp: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.pow = function<T extends Tensor>(exp: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return pow(this, exp);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prelu} from '../../ops/prelu';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prelu<T extends Tensor>(alpha: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.prelu = function<T extends Tensor>(\n    this: T, alpha: T|TensorLike): T {\n  this.throwIfDisposed();\n  return prelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prod} from '../../ops/prod';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prod<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.prod = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return prod(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {reciprocal} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reciprocal<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.reciprocal = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return reciprocal(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu} from '../../ops/relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu6} from '../../ops/relu6';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu6<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu6 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu6(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshapeAs<T extends Tensor>(x: T): T;\n  }\n}\n\n/**\n * Reshapes the tensor into the shape of the provided tensor.\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.reshapeAs = function<T extends Tensor>(x: T): T {\n  this.throwIfDisposed();\n  return reshape(this, x.shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshape<T extends Tensor>(shape: number[]): T;\n  }\n}\n\nTensor.prototype.reshape = function<T extends Tensor>(shape: number[]): T {\n  this.throwIfDisposed();\n  return reshape(this, shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeBilinear} from '../../ops/image/resize_bilinear';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeBilinear<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeBilinear = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeBilinear(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeNearestNeighbor} from '../../ops/image/resize_nearest_neighbor';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeNearestNeighbor<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeNearestNeighbor = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeNearestNeighbor(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reverse} from '../../ops/reverse';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reverse<T extends Tensor>(this: T, axis?: number|number[]): T;\n  }\n}\n\nTensor.prototype.reverse = function<T extends Tensor>(\n    this: T, axis?: number|number[]): T {\n  this.throwIfDisposed();\n  return reverse(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.rfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return rfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {round} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    round<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.round = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return round(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rsqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rsqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.rsqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return rsqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {selu} from '../../ops/selu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    selu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.selu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return selu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {separableConv2d} from '../../ops/separable_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    separableConv2d<T extends Tensor3D|Tensor4D>(\n        depthwiseFilter: Tensor4D|TensorLike4D,\n        pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n        pad: 'valid'|'same', dilation?: [number, number]|number,\n        dataFormat?: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.separableConv2d = function<T extends Tensor3D|Tensor4D>(\n    depthwiseFilter: Tensor4D|TensorLike4D,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation?: [number, number]|number,\n    dataFormat?: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return separableConv2d(\n             this, depthwiseFilter, pointwiseFilter, strides, pad, dilation,\n             dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sign} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sign<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sign = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sign(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {slice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    slice<T extends Tensor>(\n        this: T, begin: number|number[], size?: number|number[]): T;\n  }\n}\n\nTensor.prototype.slice = function<T extends Tensor>(\n    this: T, begin: number|number[], size?: number|number[]): T {\n  this.throwIfDisposed();\n  return slice(this, begin, size);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softmax<T extends Tensor>(this: T, dim?: number): T;\n  }\n}\n\nTensor.prototype.softmax = function<T extends Tensor>(this: T, dim: number): T {\n  this.throwIfDisposed();\n  return softmax(this, dim);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softplus} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softplus<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.softplus = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return softplus(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {spaceToBatchND} from '../../ops/space_to_batch_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    spaceToBatchND<R extends Rank>(blockShape: number[], paddings: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.spaceToBatchND = function<R extends Rank>(\n    blockShape: number[], paddings: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return spaceToBatchND(this, blockShape, paddings);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {split} from '../../ops/split';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    split<T extends Tensor>(numOrSizeSplits: number[]|number, axis?: number):\n        T[];\n  }\n}\n\nTensor.prototype.split = function<T extends Tensor>(\n    numOrSizeSplits: number[]|number, axis?: number): T[] {\n  this.throwIfDisposed();\n  return split(this, numOrSizeSplits, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {square} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    square<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.square = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return square(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {squaredDifference} from '../../ops/squared_difference';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifference<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.squaredDifference = function<T extends Tensor>(b: Tensor|\n                                                                TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifference(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {squaredDifferenceStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifferenceStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.squaredDifferenceStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifferenceStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {squeeze} from '../../ops/squeeze';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squeeze<T extends Tensor>(axis?: number[]): T;\n  }\n}\n\nTensor.prototype.squeeze = function<T extends Tensor>(axis?: number[]): T {\n  this.throwIfDisposed();\n  return squeeze(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {stack} from '../../ops/stack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stack<T extends Tensor>(x: Tensor|Tensor[], axis?: number): T;\n  }\n}\n\nTensor.prototype.stack = function<T extends Tensor>(\n    x: Tensor|Tensor[], axis?: number): T {\n  this.throwIfDisposed();\n  const tensorsToBeStacked = x instanceof Tensor ? [this, x] : [this, ...x];\n  return stack(tensorsToBeStacked, axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {step} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    step<T extends Tensor>(this: T, alpha?: number): T;\n  }\n}\n\nTensor.prototype.step = function<T extends Tensor>(this: T, alpha?: number) {\n  this.throwIfDisposed();\n  return step(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {stridedSlice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stridedSlice<T extends Tensor>(\n        this: Tensor, begin: number[], end: number[], strides: number[],\n        beginMask?: number, endMask?: number, ellipsisMask?: number,\n        newAxisMask?: number, shrinkAxisMask?: number): Tensor;\n  }\n}\n\nTensor.prototype.stridedSlice = function<T extends Tensor>(\n    this: Tensor, begin: number[], end: number[], strides: number[],\n    beginMask?: number, endMask?: number, ellipsisMask?: number,\n    newAxisMask?: number, shrinkAxisMask?: number): T {\n  this.throwIfDisposed();\n  return stridedSlice(\n             this, begin, end, strides, beginMask, endMask, ellipsisMask,\n             newAxisMask, shrinkAxisMask) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {subStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    subStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.subStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return subStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sub} from '../../ops/sub';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sub<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.sub = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return sub(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sum} from '../../ops/sum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sum<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.sum = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return sum(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tile} from '../../ops/tile';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tile<T extends Tensor>(b: number[]): T;\n  }\n}\n\nTensor.prototype.tile = function<T extends Tensor>(reps: number[]): T {\n  this.throwIfDisposed();\n  return tile(this, reps) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toBool<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `bool`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toBool = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'bool');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toFloat<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `float32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toFloat = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'float32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toInt<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `int32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toInt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'int32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {topk} from '../../ops/topk';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    topk<T extends Tensor>(this: T, k?: number, sorted?: boolean):\n        {values: T, indices: T};\n  }\n}\n\nTensor.prototype.topk = function<T extends Tensor>(\n    this: T, k?: number, sorted?: boolean): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return topk(this, k, sorted) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {transpose} from '../../ops/transpose';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    transpose<T extends Tensor>(perm?: number[]): T;\n  }\n}\n\nTensor.prototype.transpose = function<T extends Tensor>(\n    this: T, perm?: number[]): T {\n  this.throwIfDisposed();\n  return transpose(this, perm);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unique} from '../../ops/unique';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unique<T extends Tensor>(this: T, axis?: number): {values: T, indices: T};\n  }\n}\n\nTensor.prototype.unique = function<T extends Tensor>(\n    this: T, axis?: number): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return unique(this, axis) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unsortedSegmentSum} from '../../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike1D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unsortedSegmentSum<T extends Tensor>(\n        this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T;\n  }\n}\n\nTensor.prototype.unsortedSegmentSum = function<T extends Tensor>(\n    this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T {\n  this.throwIfDisposed();\n  return unsortedSegmentSum(this, segmentIds, numSegments);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {unstack} from '../../ops/unstack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unstack<T extends Tensor>(axis?: number): T[];\n  }\n}\n\nTensor.prototype.unstack = function<T extends Tensor>(axis?: number): T[] {\n  this.throwIfDisposed();\n  return unstack(this, axis) as T[];\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {where} from '../../ops/where';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    where<T extends Tensor>(condition: Tensor|TensorLike, x: Tensor|TensorLike):\n        T;\n  }\n}\n\nTensor.prototype.where = function<T extends Tensor>(\n    condition: Tensor|TensorLike, x: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return where(condition, this, x) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {zerosLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    zerosLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.zerosLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return zerosLike(this);\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {backend} from '@tensorflow/tfjs-core';\nimport {DataFormat} from '../keras_format/common';\n\nlet _epsilon: number;\n\n/**\n * Returns the value of the fuzz factor used in numeric expressions.\n */\nexport function epsilon() {\n  if (_epsilon == null) {\n    _epsilon = backend().epsilon();\n  }\n  return _epsilon;\n}\n\n/**\n * Sets the value of the fuzz factor used in numeric expressions.\n * @param e New value of epsilon.\n */\nexport function setEpsilon(e: number) {\n  _epsilon = e;\n}\n\n/**\n * Returns the default image data format convention.\n */\nexport function imageDataFormat(): DataFormat {\n  return 'channelsLast';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Explicit error types.\n *\n * See the following link for more information about why the code includes\n * calls to setPrototypeOf:\n *\n * https://github.com/Microsoft/TypeScript-wiki/blob/master/Breaking-Changes.md#extending-built-ins-like-error-array-and-map-may-no-longer-work\n */\n// tslint:enable\n\n/**\n * Equivalent of Python's AttributeError.\n */\nexport class AttributeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AttributeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's RuntimeError.\n */\nexport class RuntimeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, RuntimeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's ValueError.\n */\nexport class ValueError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, ValueError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's NotImplementedError.\n */\nexport class NotImplementedError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, NotImplementedError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's AssertionError.\n */\nexport class AssertionError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AssertionError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's IndexError.\n */\nexport class IndexError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, IndexError.prototype);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {DataType, fused, serialization, util} from '@tensorflow/tfjs-core';\n\nimport {AssertionError, ValueError} from '../errors';\n\n// tslint:enable\n\n/**\n * If `value` is an Array, equivalent to Python's `value * numValues`.\n * If `value` is not an Array, equivalent to Python's `[value] * numValues`\n */\n// tslint:disable-next-line:no-any\nexport function pyListRepeat(value: any, numValues: number): any[] {\n  if (Array.isArray(value)) {\n    // tslint:disable-next-line:no-any\n    let newArray: any[] = [];\n    for (let i = 0; i < numValues; i++) {\n      newArray = newArray.concat(value);\n    }\n    return newArray;\n  } else {\n    const newArray = new Array(numValues);\n    newArray.fill(value);\n    return newArray;\n  }\n}\n\nexport function assert(val: boolean, message?: string): void {\n  if (!val) {\n    throw new AssertionError(message);\n  }\n}\n\n/**\n * Count the number of elements of the `array` that are equal to `reference`.\n */\nexport function count<T>(array: T[], refernce: T) {\n  let counter = 0;\n  for (const item of array) {\n    if (item === refernce) {\n      counter++;\n    }\n  }\n  return counter;\n}\n\n/**\n * If an array is of length 1, just return the first element. Otherwise, return\n * the full array.\n * @param tensors\n */\nexport function singletonOrArray<T>(xs: T[]): T|T[] {\n  if (xs.length === 1) {\n    return xs[0];\n  }\n  return xs;\n}\n\n/**\n * Normalizes a list/tensor into a list.\n *\n * If a tensor is passed, we return\n * a list of size 1 containing the tensor.\n *\n * @param x target object to be normalized.\n */\n// tslint:disable-next-line:no-any\nexport function toList(x: any): any[] {\n  if (Array.isArray(x)) {\n    return x;\n  }\n  return [x];\n}\n\n/**\n * Generate a UID for a list\n */\n// tslint:disable-next-line:no-any\nexport function objectListUid(objs: any|any[]): string {\n  const objectList = toList(objs);\n  let retVal = '';\n  for (const obj of objectList) {\n    if (obj.id == null) {\n      throw new ValueError(\n          `Object ${obj} passed to objectListUid without an id`);\n    }\n    if (retVal !== '') {\n      retVal = retVal + ', ';\n    }\n    retVal = `${retVal}${Math.abs(obj.id)}`;\n  }\n  return retVal;\n}\n/**\n * Converts string to snake-case.\n * @param name\n */\nexport function toSnakeCase(name: string): string {\n  const intermediate = name.replace(/(.)([A-Z][a-z0-9]+)/g, '$1_$2');\n  const insecure =\n      intermediate.replace(/([a-z])([A-Z])/g, '$1_$2').toLowerCase();\n  /*\n   If the class is private the name starts with \"_\" which is not secure\n   for creating scopes. We prefix the name with \"private\" in this case.\n   */\n  if (insecure[0] !== '_') {\n    return insecure;\n  }\n  return 'private' + insecure;\n}\n\nexport function toCamelCase(identifier: string): string {\n  // quick return for empty string or single character strings\n  if (identifier.length <= 1) {\n    return identifier;\n  }\n  // Check for the underscore indicating snake_case\n  if (identifier.indexOf('_') === -1) {\n    return identifier;\n  }\n  return identifier.replace(/[_]+(\\w|$)/g, (m, p1) => p1.toUpperCase());\n}\n\n// tslint:disable-next-line:no-any\nlet _GLOBAL_CUSTOM_OBJECTS = {} as {[objName: string]: any};\n\nexport function serializeKerasObject(instance: serialization.Serializable):\n    serialization.ConfigDictValue {\n  if (instance === null || instance === undefined) {\n    return null;\n  }\n  const dict: serialization.ConfigDictValue = {};\n  dict['className'] = instance.getClassName();\n  dict['config'] = instance.getConfig();\n  return dict;\n}\n\n/**\n * Replace ndarray-style scalar objects in serialization objects with numbers.\n *\n * Background: In some versions of tf.keras, certain scalar values in the HDF5\n * model save file can be serialized as: `{'type': 'ndarray', 'value': num}`,\n * where in `num` is a plain number. This method converts such serialization\n * to a `number`.\n *\n * @param config The keras-format serialization object to be processed\n *   (in place).\n */\nfunction convertNDArrayScalarsInConfig(config: serialization.ConfigDictValue):\n    void {\n  if (config == null || typeof config !== 'object') {\n    return;\n  } else if (Array.isArray(config)) {\n    config.forEach(configItem => convertNDArrayScalarsInConfig(configItem));\n  } else {\n    const fields = Object.keys(config);\n    for (const field of fields) {\n      const value = config[field];\n      if (value != null && typeof value === 'object') {\n        if (!Array.isArray(value) && value['type'] === 'ndarray' &&\n            typeof value['value'] === 'number') {\n          config[field] = value['value'];\n        } else {\n          convertNDArrayScalarsInConfig(value as serialization.ConfigDict);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Deserialize a saved Keras Object\n * @param identifier either a string ID or a saved Keras dictionary\n * @param moduleObjects a list of Python class names to object constructors\n * @param customObjects a list of Python class names to object constructors\n * @param printableModuleName debug text for the object being reconstituted\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns a TensorFlow.js Layers object\n */\n// tslint:disable:no-any\nexport function deserializeKerasObject(\n    identifier: string|serialization.ConfigDict,\n    moduleObjects = {} as {[objName: string]: any},\n    customObjects = {} as {[objName: string]: any},\n    printableModuleName = 'object', fastWeightInit = false): any {\n  // tslint:enable\n  if (typeof identifier === 'string') {\n    const functionName = identifier;\n    let fn;\n    if (functionName in customObjects) {\n      fn = customObjects[functionName];\n    } else if (functionName in _GLOBAL_CUSTOM_OBJECTS) {\n      fn = _GLOBAL_CUSTOM_OBJECTS[functionName];\n    } else {\n      fn = moduleObjects[functionName];\n      if (fn == null) {\n        throw new ValueError(\n            `Unknown ${printableModuleName}: ${identifier}. ` +\n            `This may be due to one of the following reasons:\\n` +\n            `1. The ${printableModuleName} is defined in Python, in which ` +\n            `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n            `code.\\n` +\n            `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n            `but is not registered properly with ` +\n            `tf.serialization.registerClass().`);\n        // TODO(cais): Add link to tutorial page on custom layers.\n      }\n    }\n    return fn;\n  } else {\n    // In this case we are dealing with a Keras config dictionary.\n    const config = identifier;\n    if (config['className'] == null || config['config'] == null) {\n      throw new ValueError(\n          `${printableModuleName}: Improper config format: ` +\n          `${JSON.stringify(config)}.\\n` +\n          `'className' and 'config' must set.`);\n    }\n    const className = config['className'] as string;\n    let cls, fromConfig;\n    if (className in customObjects) {\n      [cls, fromConfig] = customObjects[className];\n    } else if (className in _GLOBAL_CUSTOM_OBJECTS) {\n      [cls, fromConfig] = _GLOBAL_CUSTOM_OBJECTS['className'];\n    } else if (className in moduleObjects) {\n      [cls, fromConfig] = moduleObjects[className];\n    }\n    if (cls == null) {\n      throw new ValueError(\n          `Unknown ${printableModuleName}: ${className}. ` +\n          `This may be due to one of the following reasons:\\n` +\n          `1. The ${printableModuleName} is defined in Python, in which ` +\n          `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n          `code.\\n` +\n          `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n          `but is not registered properly with ` +\n          `tf.serialization.registerClass().`);\n      // TODO(cais): Add link to tutorial page on custom layers.\n    }\n    if (fromConfig != null) {\n      // Porting notes: Instead of checking to see whether fromConfig accepts\n      // customObjects, we create a customObjects dictionary and tack it on to\n      // config['config'] as config['config'].customObjects. Objects can use it,\n      // if they want.\n\n      // tslint:disable-next-line:no-any\n      const customObjectsCombined = {} as {[objName: string]: any};\n      for (const key of Object.keys(_GLOBAL_CUSTOM_OBJECTS)) {\n        customObjectsCombined[key] = _GLOBAL_CUSTOM_OBJECTS[key];\n      }\n      for (const key of Object.keys(customObjects)) {\n        customObjectsCombined[key] = customObjects[key];\n      }\n      // Add the customObjects to config\n      const nestedConfig = config['config'] as serialization.ConfigDict;\n      nestedConfig['customObjects'] = customObjectsCombined;\n\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      convertNDArrayScalarsInConfig(config['config']);\n      const returnObj =\n          fromConfig(cls, config['config'], customObjects, fastWeightInit);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n\n      return returnObj;\n    } else {\n      // Then `cls` may be a function returning a class.\n      // In this case by convention `config` holds\n      // the kwargs of the function.\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      // In python this is **config['config'], for tfjs-layers we require\n      // classes that use this fall-through construction method to take\n      // a config interface that mimics the expansion of named parameters.\n      const returnObj = new cls(config['config']);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n      return returnObj;\n    }\n  }\n}\n\n/**\n * Compares two numbers for sorting.\n * @param a\n * @param b\n */\nexport function numberCompare(a: number, b: number) {\n  return (a < b) ? -1 : ((a > b) ? 1 : 0);\n}\n\n/**\n * Comparison of two numbers for reverse sorting.\n * @param a\n * @param b\n */\nexport function reverseNumberCompare(a: number, b: number) {\n  return -1 * numberCompare(a, b);\n}\n\n/**\n * Convert a string into the corresponding DType.\n * @param dtype\n * @returns An instance of DType.\n */\nexport function stringToDType(dtype: string): DataType {\n  switch (dtype) {\n    case 'float32':\n      return 'float32';\n    default:\n      throw new ValueError(`Invalid dtype: ${dtype}`);\n  }\n}\n\n/**\n * Test the element-by-element equality of two Arrays of strings.\n * @param xs First array of strings.\n * @param ys Second array of strings.\n * @returns Wether the two arrays are all equal, element by element.\n */\nexport function stringsEqual(xs: string[], ys: string[]): boolean {\n  if (xs == null || ys == null) {\n    return xs === ys;\n  }\n  if (xs.length !== ys.length) {\n    return false;\n  }\n  for (let i = 0; i < xs.length; ++i) {\n    if (xs[i] !== ys[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Get the unique elements of an array.\n * @param xs Array.\n * @returns An Array consisting of the unique elements in `xs`.\n */\nexport function unique<T>(xs: T[]): T[] {\n  if (xs == null) {\n    return xs;\n  }\n  const out: T[] = [];\n  // TODO(cais): Maybe improve performance by sorting.\n  for (const x of xs) {\n    if (out.indexOf(x) === -1) {\n      out.push(x);\n    }\n  }\n  return out;\n}\n\n/**\n * Determine if an Object is empty (i.e., does not have own properties).\n * @param obj Object\n * @returns Whether the Object is empty.\n * @throws ValueError: If object is `null` or `undefined`.\n */\nexport function isObjectEmpty(obj: {}): boolean {\n  if (obj == null) {\n    throw new ValueError(`Invalid value in obj: ${JSON.stringify(obj)}`);\n  }\n  for (const key in obj) {\n    if (obj.hasOwnProperty(key)) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Helper function used to build type union/enum run-time checkers.\n * @param values The list of allowed values.\n * @param label A string name for the type\n * @param value The value to test.\n * @throws ValueError: If the value is not in values nor `undefined`/`null`.\n */\nexport function checkStringTypeUnionValue(\n    values: string[], label: string, value: string): void {\n  if (value == null) {\n    return;\n  }\n  if (values.indexOf(value) < 0) {\n    throw new ValueError(`${value} is not a valid ${label}.  Valid values are ${\n        values} or null/undefined.`);\n  }\n}\n\n/**\n * Helper function for verifying the types of inputs.\n *\n * Ensures that the elements of `x` are all of type `expectedType`.\n * Also verifies that the length of `x` is within bounds.\n *\n * @param x Object to test.\n * @param expectedType The string expected type of all of the elements in the\n * Array.\n * @param minLength Return false if x.length is less than this.\n * @param maxLength Return false if x.length is greater than this.\n * @returns true if and only if `x` is an `Array<expectedType>` with\n * length >= `minLength` and <= `maxLength`.\n */\n// tslint:disable:no-any\nexport function checkArrayTypeAndLength(\n    x: any, expectedType: string, minLength = 0,\n    maxLength = Infinity): boolean {\n  assert(minLength >= 0);\n  assert(maxLength >= minLength);\n  return (\n      Array.isArray(x) && x.length >= minLength && x.length <= maxLength &&\n      x.every(e => typeof e === expectedType));\n}\n// tslint:enable:no-any\n\n/**\n * Assert that a value or an array of value are positive integer.\n *\n * @param value The value being asserted on. May be a single number or an array\n *   of numbers.\n * @param name Name of the value, used to make the error message.\n */\nexport function assertPositiveInteger(value: number|number[], name: string) {\n  if (Array.isArray(value)) {\n    util.assert(\n        value.length > 0, () => `${name} is unexpectedly an empty array.`);\n    value.forEach(\n        (v, i) => assertPositiveInteger(v, `element ${i + 1} of ${name}`));\n  } else {\n    util.assert(\n        Number.isInteger(value) && value > 0,\n        () => `Expected ${name} to be a positive integer, but got ` +\n            `${formatAsFriendlyString(value)}.`);\n  }\n}\n\n/**\n * Format a value into a display-friendly, human-readable fashion.\n *\n * - `null` is formatted as `'null'`\n * - Strings are formated with flanking pair of quotes.\n * - Arrays are formatted with flanking pair of square brackets.\n *\n * @param value The value to display.\n * @return Formatted string.\n */\n// tslint:disable-next-line:no-any\nexport function formatAsFriendlyString(value: any): string {\n  if (value === null) {\n    return 'null';\n  } else if (Array.isArray(value)) {\n    return '[' + value.map(v => formatAsFriendlyString(v)).join(',') + ']';\n  } else if (typeof value === 'string') {\n    return `\"${value}\"`;\n  } else {\n    return `${value}`;\n  }\n}\n\n/**\n * Returns a function `f2` (decorator) which wraps the original function\n * `f`. `f2` guarantees that `f` can be called at most once\n * every `waitMs` ms. If `f2` is called more often, it will return\n * the last returned result of `f`.\n *\n * @param f The original function `f` to wrap.\n * @param waitMs The time between two consecutive calls to `f` in ms.\n */\nexport function debounce<T>(\n    f: (...args: Array<{}>) => T, waitMs: number): (...args: Array<{}>) => T {\n  let lastTime = util.now();\n  let lastResult: T;\n  const f2 = (...args: Array<{}>) => {\n    const now = util.now();\n    if (now - lastTime < waitMs) {\n      return lastResult;\n    }\n    lastTime = now;\n    lastResult = f(...args);\n    return lastResult;\n  };\n  return f2;\n}\n\n/**\n * Returns the fusable activation given a layers identifier.\n *\n * @param activationName The layers identifier string.\n * @return The name of the fusable activation.\n */\nexport function mapActivationToFusedKernel(activationName: string):\n    fused.Activation {\n  if (activationName === 'relu') {\n    return 'relu';\n  }\n  if (activationName === 'linear') {\n    return 'linear';\n  }\n  if (activationName === 'elu') {\n    return 'elu';\n  }\n  return null;\n}\n\ntype PossibleValues = Array<Array<boolean|string|number>>;\n\n/**\n * Returns the cartesian product of sets of values.\n * This works the same as itertools.product in Python.\n *\n * Example:\n *\n * filters = [128, 256, 512]\n * paddings = ['same', 'valid']\n *\n * product = [ [128, 'same'], [128, 'valid'], [256, 'same'], [256, 'valid'],\n * [512, 'same'], [512, 'valid']]\n *\n * @param arrayOfValues List/array of values.\n * @return The cartesian product.\n */\nexport function getCartesianProductOfValues(...arrayOfValues: PossibleValues):\n    PossibleValues {\n  assert(arrayOfValues.length > 0, 'arrayOfValues is empty');\n\n  for (const values of arrayOfValues) {\n    assert(Array.isArray(values), 'one of the values is not an array');\n    assert(values.length > 0, 'one of the values is empty');\n  }\n\n  return arrayOfValues.reduce((products, values) => {\n    if (products.length === 0) {\n      return values.map(value => [value]);\n    }\n\n    return values\n        .map(value => {\n          return products.map((prevValue) => [...prevValue, value]);\n        })\n        .reduce((flattenedProduct, unflattenedProduct) => {\n          return flattenedProduct.concat(unflattenedProduct);\n        }, []);\n  }, [] as PossibleValues);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/contraints.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport {epsilon} from './backend/common';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\n/**\n * Helper function used by many of the Constraints to find the L2Norms.\n */\nfunction calcL2Norms(w: Tensor, axis: number): Tensor {\n  return tidy(() => tfc.sqrt(tfc.sum(tfc.mul(w, w), axis, true)));\n}\n\n/**\n * Base class for functions that impose constraints on weight values\n *\n * @doc {\n *   heading: 'Constraints',\n *   subheading: 'Classes',\n *   namespace: 'constraints'\n * }\n */\nexport abstract class Constraint extends serialization.Serializable {\n  /* Porting note: was __call__, apply chosen to match other similar choices */\n  abstract apply(w: Tensor): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport interface MaxNormArgs {\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   *\n   *  For instance, in a `Dense` layer the weight matrix\n   *  has shape `[inputDim, outputDim]`,\n   *  set `axis` to `0` to constrain each weight vector\n   *  of length `[inputDim,]`.\n   *  In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   *  the weight tensor has shape\n   *  `[rows, cols, inputDepth, outputDepth]`,\n   *  set `axis` to `[0, 1, 2]`\n   *  to constrain the weights of each filter tensor of size\n   *  `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class MaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MaxNorm';\n  private maxValue: number;\n  private axis: number;\n  private readonly defaultMaxValue = 2;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MaxNormArgs) {\n    super();\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.clipByValue(norms, 0, this.maxValue);\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {maxValue: this.maxValue, axis: this.axis};\n  }\n}\nserialization.registerClass(MaxNorm);\n\nexport interface UnitNormArgs {\n  /**\n   * Axis along which to calculate norms.\n   *\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * [rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class UnitNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'UnitNorm';\n  private axis: number;\n  private readonly defaultAxis = 0;\n  constructor(args: UnitNormArgs) {\n    super();\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(\n        () => tfc.div(w, tfc.add(epsilon(), calcL2Norms(w, this.axis))));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {axis: this.axis};\n  }\n}\nserialization.registerClass(UnitNorm);\n\nexport class NonNeg extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'NonNeg';\n\n  apply(w: Tensor): Tensor {\n    return tfc.relu(w);\n  }\n}\nserialization.registerClass(NonNeg);\n\nexport interface MinMaxNormArgs {\n  /**\n   * Minimum norm for incoming weights\n   */\n  minValue?: number;\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * `[rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n  /**\n   * Rate for enforcing the constraint: weights will be rescaled to yield:\n   * `(1 - rate) * norm + rate * norm.clip(minValue, maxValue)`.\n   * Effectively, this means that rate=1.0 stands for strict\n   * enforcement of the constraint, while rate<1.0 means that\n   * weights will be rescaled at each step to slowly move\n   * towards a value inside the desired interval.\n   */\n  rate?: number;\n}\n\nexport class MinMaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MinMaxNorm';\n  private minValue: number;\n  private maxValue: number;\n  private rate: number;\n  private axis: number;\n  private readonly defaultMinValue = 0.0;\n  private readonly defaultMaxValue = 1.0;\n  private readonly defaultRate = 1.0;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MinMaxNormArgs) {\n    super();\n    this.minValue =\n        args.minValue != null ? args.minValue : this.defaultMinValue;\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.rate = args.rate != null ? args.rate : this.defaultRate;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.add(\n          tfc.mul(\n              this.rate, tfc.clipByValue(norms, this.minValue, this.maxValue)),\n          tfc.mul(1.0 - this.rate, norms));\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      minValue: this.minValue,\n      maxValue: this.maxValue,\n      rate: this.rate,\n      axis: this.axis\n    };\n  }\n}\nserialization.registerClass(MinMaxNorm);\n\n/** @docinline */\nexport type ConstraintIdentifier =\n    'maxNorm'|'minMaxNorm'|'nonNeg'|'unitNorm'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in ConstraintIdentifier]: string} = {\n      'maxNorm': 'MaxNorm',\n      'minMaxNorm': 'MinMaxNorm',\n      'nonNeg': 'NonNeg',\n      'unitNorm': 'UnitNorm'\n    };\n\nexport function serializeConstraint(constraint: Constraint):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeConstraint(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Constraint {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'constraint');\n}\n\nexport function getConstraint(identifier: ConstraintIdentifier|\n                              serialization.ConfigDict|Constraint): Constraint {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeConstraint(config);\n  } else if (identifier instanceof Constraint) {\n    return identifier;\n  } else {\n    return deserializeConstraint(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constraint, MaxNorm, MaxNormArgs, MinMaxNorm, MinMaxNormArgs, NonNeg, UnitNorm, UnitNormArgs} from './constraints';\n\n/**\n * MaxNorm weight constraint.\n *\n * Constrains the weights incident to each hidden unit\n * to have a norm less than or equal to a desired value.\n *\n * References\n *       - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting\n * Srivastava, Hinton, et al.\n * 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Constraints',namespace: 'constraints'}\n */\nexport function maxNorm(args: MaxNormArgs): Constraint {\n  return new MaxNorm(args);\n}\n\n/**\n * Constrains the weights incident to each hidden unit to have unit norm.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function unitNorm(args: UnitNormArgs): Constraint {\n  return new UnitNorm(args);\n}\n\n/**\n * Constains the weight to be non-negative.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function nonNeg(): Constraint {\n  return new NonNeg();\n}\n\n/** @doc {heading: 'Constraints', namespace: 'constraints'} */\nexport function minMaxNorm(config: MinMaxNormArgs): Constraint {\n  return new MinMaxNorm(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// TODO(huan): add layer-specific input shape types (see: https://github.com/tensorflow/tfjs-layers/pull/492)\n/** @docalias (null | number)[] */\nexport type Shape = Array<null | number>;\n\n// The tfjs-core version of DataType must stay synced with this.\nexport type DataType = 'float32'|'int32'|'bool'|'complex64'|'string';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type DataFormat = 'channelsFirst'|'channelsLast';\nexport const VALID_DATA_FORMAT_VALUES = ['channelsFirst', 'channelsLast'];\n\n// These constants have a snake vs. camel distinction.\nexport type DataFormatSerialization = 'channels_first'|'channels_last';\n\n/** @docinline */\nexport type PaddingMode = 'valid'|'same'|'causal';\nexport const VALID_PADDING_MODE_VALUES = ['valid', 'same', 'causal'];\n\n/** @docinline */\nexport type PoolMode = 'max'|'avg';\nexport const VALID_POOL_MODE_VALUES = ['max', 'avg'];\n\n/** @docinline */\nexport type BidirectionalMergeMode = 'sum'|'mul'|'concat'|'ave';\nexport const VALID_BIDIRECTIONAL_MERGE_MODES = ['sum', 'mul', 'concat', 'ave'];\n\n/** @docinline */\nexport type SampleWeightMode = 'temporal';\nexport const VALID_SAMPLE_WEIGHT_MODES = ['temporal'];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Common functions for TensorFlow.js Layers.\n */\nimport {VALID_DATA_FORMAT_VALUES, VALID_PADDING_MODE_VALUES, VALID_POOL_MODE_VALUES} from './keras_format/common';\nimport {checkStringTypeUnionValue} from './utils/generic_utils';\n\n// A map from the requested scoped name of a Tensor to the number of Tensors\n// wanting that name so far.  This allows enforcing name uniqueness by appending\n// an incrementing index, e.g. scope/name, scope/name_1, scope/name_2, etc.\nconst nameMap: Map<string, number> = new Map<string, number>();\n\nexport function checkDataFormat(value?: string): void {\n  checkStringTypeUnionValue(VALID_DATA_FORMAT_VALUES, 'DataFormat', value);\n}\n\nexport function checkPaddingMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_PADDING_MODE_VALUES, 'PaddingMode', value);\n}\n\nexport function checkPoolMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_POOL_MODE_VALUES, 'PoolMode', value);\n}\n\nconst _nameScopeStack: string[] = [];\nconst _nameScopeDivider = '/';\n\n/**\n * Enter namescope, which can be nested.\n */\nexport function nameScope<T>(name: string, fn: () => T): T {\n  _nameScopeStack.push(name);\n  try {\n    const val: T = fn();\n    _nameScopeStack.pop();\n    return val;\n  } catch (e) {\n    _nameScopeStack.pop();\n    throw e;\n  }\n}\n\n/**\n * Get the current namescope as a flat, concatenated string.\n */\nfunction currentNameScopePrefix(): string {\n  if (_nameScopeStack.length === 0) {\n    return '';\n  } else {\n    return _nameScopeStack.join(_nameScopeDivider) + _nameScopeDivider;\n  }\n}\n\n/**\n * Get the name a Tensor (or Variable) would have if not uniqueified.\n * @param tensorName\n * @return Scoped name string.\n */\nexport function getScopedTensorName(tensorName: string): string {\n  if (!isValidTensorName(tensorName)) {\n    throw new Error('Not a valid tensor name: \\'' + tensorName + '\\'');\n  }\n  return currentNameScopePrefix() + tensorName;\n}\n\n/**\n * Get unique names for Tensors and Variables.\n * @param scopedName The fully-qualified name of the Tensor, i.e. as produced by\n *  `getScopedTensorName()`.\n * @return A unique version of the given fully scoped name.\n *   If this is the first time that the scoped name is seen in this session,\n *   then the given `scopedName` is returned unaltered.  If the same name is\n *   seen again (producing a collision), an incrementing suffix is added to the\n *   end of the name, so it takes the form 'scope/name_1', 'scope/name_2', etc.\n */\nexport function getUniqueTensorName(scopedName: string): string {\n  if (!isValidTensorName(scopedName)) {\n    throw new Error('Not a valid tensor name: \\'' + scopedName + '\\'');\n  }\n  if (!nameMap.has(scopedName)) {\n    nameMap.set(scopedName, 0);\n  }\n  const index = nameMap.get(scopedName);\n  nameMap.set(scopedName, nameMap.get(scopedName) + 1);\n\n  if (index > 0) {\n    const result = `${scopedName}_${index}`;\n    // Mark the composed name as used in case someone wants\n    // to call getUniqueTensorName(\"name_1\").\n    nameMap.set(result, 1);\n    return result;\n  } else {\n    return scopedName;\n  }\n}\n\nconst tensorNameRegex = new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\\._\\/]*$/);\n\n/**\n * Determine whether a string is a valid tensor name.\n * @param name\n * @returns A Boolean indicating whether `name` is a valid tensor name.\n */\nexport function isValidTensorName(name: string): boolean {\n  return !!name.match(tensorNameRegex);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Math utility functions.\n *\n * This file contains some frequently used math function that operates on\n * number[] or Float32Array and return a number. Many of these functions are\n * not-so-thick wrappers around TF.js Core functions. But they offer the\n * convenience of\n * 1) not having to convert the inputs into Tensors,\n * 2) not having to convert the returned Tensors to numbers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar, Tensor1D, tensor1d} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\n\nexport type ArrayTypes = Uint8Array|Int32Array|Float32Array;\n\n/**\n * Determine if a number is an integer.\n */\nexport function isInteger(x: number): boolean {\n  return x === parseInt(x.toString(), 10);\n}\n\n/**\n * Calculate the product of an array of numbers.\n * @param array The array to calculate the product over.\n * @param begin Beginning index, inclusive.\n * @param end Ending index, exclusive.\n * @return The product.\n */\nexport function arrayProd(\n    array: number[]|ArrayTypes, begin?: number, end?: number): number {\n  if (begin == null) {\n    begin = 0;\n  }\n  if (end == null) {\n    end = array.length;\n  }\n\n  let prod = 1;\n  for (let i = begin; i < end; ++i) {\n    prod *= array[i];\n  }\n  return prod;\n}\n\n/**\n * A helper function transforms the two input types to an instance of Tensor1D,\n * so the return value can be fed directly into various TF.js Core functions.\n * @param array\n */\nfunction toArray1D(array: number[]|Float32Array): Tensor1D {\n  array = Array.isArray(array) ? new Float32Array(array) : array;\n  return tensor1d(array);\n}\n\n/**\n * Compute minimum value.\n * @param array\n * @return minimum value.\n */\nexport function min(array: number[]|Float32Array): number {\n  return tfc.min(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute maximum value.\n * @param array\n * @return maximum value\n */\nexport function max(array: number[]|Float32Array): number {\n  return tfc.max(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute sum of array.\n * @param array\n * @return The sum.\n */\nexport function sum(array: number[]|Float32Array): number {\n  return tfc.sum(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute mean of array.\n * @param array\n * @return The mean.\n */\nexport function mean(array: number[]|Float32Array): number {\n  return sum(array) / array.length;\n}\n\n/**\n * Compute variance of array.\n * @param array\n * @return The variance.\n */\nexport function variance(array: number[]|Float32Array): number {\n  const demeaned = tfc.sub(toArray1D(array), scalar(mean(array)));\n  const sumSquare = tfc.sum(tfc.mul(demeaned, demeaned)).dataSync()[0];\n  return sumSquare / array.length;\n}\n\n/**\n * Compute median of array.\n * @param array\n * @return The median value.\n */\nexport function median(array: number[]|Float32Array): number {\n  const arraySorted = array.slice().sort((a, b) => a - b);\n  const lowIdx = Math.floor((arraySorted.length - 1) / 2);\n  const highIdx = Math.ceil((arraySorted.length - 1) / 2);\n  if (lowIdx === highIdx) {\n    return arraySorted[lowIdx];\n  }\n  return (arraySorted[lowIdx] + arraySorted[highIdx]) / 2;\n}\n\n/**\n * Generate an array of integers in [begin, end).\n * @param begin Beginning integer, inclusive.\n * @param end Ending integer, exclusive.\n * @returns Range array.\n * @throws ValueError, iff `end` < `begin`.\n */\nexport function range(begin: number, end: number): number[] {\n  if (end < begin) {\n    throw new ValueError(`end (${end}) < begin (${begin}) is forbidden.`);\n  }\n  const out: number[] = [];\n  for (let i = begin; i < end; ++i) {\n    out.push(i);\n  }\n  return out;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * deeplearn.js backend.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {onesLike as coreOnesLike, scalar, Tensor, Tensor1D, tensor1d, Tensor2D, Tensor3D, Tensor4D, Tensor5D, tidy, where, zerosLike as coreZerosLike} from '@tensorflow/tfjs-core';\nimport {checkDataFormat} from '../common';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {HasShape} from '../types';\nimport * as math_utils from '../utils/math_utils';\n\nimport {imageDataFormat} from './common';\n\n// tslint:enable\n\n/* Setting and getting backend from deeplearn.js. */\n\n// Default deeplearn.js backend is WebGL (GPU).\nlet backend: 'cpu'|'webgl' = 'webgl';\n\nexport function setBackend(requestedBackend: 'cpu'|'webgl') {\n  tfc.setBackend(requestedBackend);\n  backend = requestedBackend;\n}\n\nexport function getBackend(): 'cpu'|'webgl' {\n  return backend;\n}\n\n/**\n * Indicates whether the backend is operating symbolically.\n *\n * This function will be used to determine how to interpret user code. If\n * it returns true, calls to the backend construct a symbolic graph; if\n * it returns false, calls to the backend execute immediately.\n */\nexport function isBackendSymbolic(): boolean {\n  return false;\n}\n\n/**\n * Get the number of elements in a Tensor.\n * @param x The Tensor.\n * @return Number of elements in `x`.\n */\nexport function countParams(x: HasShape): number {\n  const shape = x.shape;\n  if (shape.length > 0) {\n    return shape.reduce((a: number, b: number) => a * b);\n  } else {\n    // Scalar.\n    return 1;\n  }\n}\n\n/**\n * Casts a tensor to a different dtype and returns it.\n * @param x Input tensor.\n * @param dtype String: 'float32'|'int32'|'bool'.\n * @returns Tensor of the specified `dtype`.\n */\nexport function cast(x: Tensor, dtype: tfc.DataType): Tensor {\n  return x.asType(dtype);\n}\n\n/**\n * Adds a 1-sized dimension at index \"axis\".\n * @param x Input tensor.\n * @param axis Position where to add the new axis.\n * @returns Result of the dimension expansion.\n */\nexport function expandDims(x: Tensor, axis = -1): Tensor {\n  const outShape = x.shape.slice();\n  if (axis < 0) {\n    axis = outShape.length + axis + 1;\n  }\n  outShape.splice(axis, 0, 1);\n  return x.reshape(outShape);\n}\n\n/**\n * Repeats a 2D tensor.\n *\n * If `x` has shape `[samples, dim]` and `n` is 2, for example, the output\n * will have shape `[samples, 2, dim]`.\n *\n * @param x Input tensor.\n * @param n Integer, number of times to repeat.\n * @returns The result of the repeat operation.\n * @throws ValueError: If input tensor is not 2D.\n */\nexport function repeat(x: Tensor, n: number): Tensor {\n  return tidy(() => {\n    if (x.shape.length !== 2) {\n      throw new ValueError(\n          `repeat() expects a rank-2 tensor, but received a ` +\n          `rank-${x.shape.length} tensor.`);\n    }\n    const y = expandDims(x, 1);\n    return tile(y, [1, n, 1]);\n  });\n}\n\n/**\n * Flatten a Tensor into 1D.\n * @param x Input tensor.\n * @return The result of the flattening `x`.\n */\nexport function flatten(x: Tensor): Tensor {\n  const newShape = [math_utils.arrayProd(x.shape)];\n  return x.reshape(newShape);\n}\n\n/**\n * Turn a nD tensor into a 2D tensor with same 0th dimension.\n * In other words, it flattens each data samples of a batch.\n *\n * @param x The tensor to flatten. The rank of this tensor is required to be 2\n *   or higher.\n * @return The result of the flattening.\n */\nexport function batchFlatten(x: Tensor): Tensor {\n  if (x.rank <= 1) {\n    throw new ValueError(\n        `batchFlatten requires a minimum rank of 2. Got rank: ${x.rank}.`);\n  }\n  const newShape = [x.shape[0], math_utils.arrayProd(x.shape, 1)];\n  return x.reshape(newShape);\n}\n\n/**\n * Do slicing along the first axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the first axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongFirstAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [start, 0], [size, array.shape[1]]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [start, 0, 0],\n            [size, array.shape[1], array.shape[2]]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [start, 0, 0, 0],\n            [size, array.shape[1], array.shape[2], array.shape[3]]);\n      case 5:\n        return tfc.slice(array as Tensor5D, [start, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4]\n        ]);\n      case 6:\n        return tfc.slice(array, [start, 0, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4],\n          array.shape[5]\n        ]);\n      default:\n        throw new ValueError(\n            `sliceAlongFirstAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the last axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the last axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongLastAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [0, start], [array.shape[0], size]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [0, 0, start],\n            [array.shape[0], array.shape[1], size]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [0, 0, 0, start],\n            [array.shape[0], array.shape[1], array.shape[2], size]);\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the sepcified axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size of the slice along the chosen axis.\n * @param choose an axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongAxis(\n    array: Tensor, start: number, size: number, axis: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 3:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice3d(\n                array as Tensor3D, [0, start, 0],\n                [array.shape[0], size, array.shape[2]]);\n          case 3:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 4:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice4d(\n                array as Tensor4D, [0, start, 0, 0],\n                [array.shape[0], size, array.shape[2], array.shape[3]]);\n          case 3:\n            return tfc.slice4d(\n                array as Tensor4D, [0, 0, start, 0],\n                [array.shape[0], array.shape[1], size, array.shape[3]]);\n          case 4:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Concatenates a list of tensors alongside the specified axis.\n * @param tensors `Array` of tensors to concatenate.\n * @param axis Concatenation axis.\n * @returns The result of the concatenation.\n */\nexport function concatenate(tensors: Tensor[], axis = -1): Tensor {\n  let rank: number;\n  if (axis < 0) {\n    rank = tensors[0].rank;\n    if (rank !== 0) {\n      axis = rank;\n    } else {\n      axis = 0;\n    }\n  }\n  if (axis === tensors[0].rank) {\n    // Porting Note: This is necessary because tfc.concat() requires axis to be\n    //   in the interval [-rank, rank).\n    axis = -1;\n  }\n  // Porting Note: Sparse concat is not supported yet.\n  return tfc.concat(tensors, axis);\n}\n\n/**\n * Concatenate two arrays along the first dimension.\n * @param a The 1st `tf.Tensor` to concatenate.\n * @param b The 2nd `tf.Tensor` to concatenate.\n * @returns Result of the concatenation.\n * @throws ValueError: If `a` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function concatAlongFirstAxis(a: Tensor, b: Tensor): Tensor {\n  switch (a.rank) {\n    case 1:\n      return tfc.concat1d([a as Tensor1D, b as Tensor1D]);\n    case 2:\n      return tfc.concat2d([a as Tensor2D, b as Tensor2D], 0);\n    case 3:\n      return tfc.concat3d([a as Tensor3D, b as Tensor3D], 0);\n    case 4:\n      return tfc.concat4d([a as Tensor4D, b as Tensor4D], 0);\n    default:\n      throw new ValueError(\n          `concatAlongFirstAxis() received an unsupported ` +\n          `tensor rank: ${a.rank}`);\n  }\n}\n\n/**\n * Creates a tensor by tiling `x` by `n`.\n * @param x A tensor.\n * @param n An Array of integers or a single integer. If an Array, the length\n *   must be the same as the number of dimensions in `x`. If a single integer,\n *   it will be treated as an Array of length 1.\n */\nexport function tile(x: Tensor, n: number|number[]): Tensor {\n  if (!Array.isArray(n)) {\n    n = [n];\n  }\n  if (x.rank !== n.length) {\n    throw new ValueError(\n        `The length of input n (${n.length}) does not match ` +\n        `the number of dimensions in input x (${x.rank})`);\n  }\n  return tfc.tile(x, n);\n}\n\n/* Creation of random tensors. */\n\n/**\n * Get a tensor with normal distribution of values.\n *\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @return The normal tensor.\n */\nexport function randomNormal(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: 'float32'|'int32',\n    seed?: number): Tensor {\n  return tfc.randomNormal(shape, mean, stddev, dtype, seed);\n}\n\n/* Linear Algebra */\n\n/**\n * Multiply two tensors and returns the result as a tensor.\n *\n * For 2D tensors, this is equivalent to matrix multiplication (matMul).\n * For tensors of higher ranks, it follows the Theano behavior,\n * (e.g. `(2, 3) * (4, 3, 5) -> (2, 4, 5)`).  From the Theano documentation:\n *\n * For N dimensions it is a sum product over the last axis of x and the\n * second-to-last of y:\n *\n * @param a A tensor of at least rank 2.\n * @param b A tensor of at least rank 2.\n * @param activation (optional) A string identifying the activation\n *   function.\n * @return Result of the dot operation.\n */\nexport function dot(\n    a: Tensor, b: Tensor, activation?: tfc.fused.Activation,\n    bias?: Tensor): Tensor {\n  if ((a.rank < 2) || (b.rank < 2)) {\n    throw new NotImplementedError(\n        `dot requires both inputs to be rank >= 2` +\n        ` but got x shape = ${a.shape} and y shape = ${b.shape}`);\n  }\n  if (b.rank >= 3) {\n    const xLastDim = a.shape.slice(-1)[0];\n    const ySecondLastDim = b.shape.slice(-2)[0];\n    if (xLastDim !== ySecondLastDim) {\n      throw new NotImplementedError(\n          `If rank y >= 3, then the second last dim` +\n          ` of y must equal the last dim of x but got x shape = ${\n              a.shape} and ` +\n          ` y shape = ${b.shape}`);\n    }\n  }\n  // Handle basic 2D x 2D case.\n  if ((a.rank === 2) && (b.rank === 2)) {\n    const transposeA = false;\n    const transposeB = false;\n    // tfc.fused.matMul only fuses certain activation functions. Unsupported\n    // activation functions are treated as 'linear' activations, which is\n    // equivalent to a no-op.\n    return tfc.fused.matMul({\n      a,\n      b: b as Tensor2D,\n      transposeA,\n      transposeB,\n      bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n      activation\n    });\n  } else {\n    // Reshape x into the analogous 2D Tensor.\n    const aFirstDims = a.shape.slice();  // Holds all but the last dim of x.\n    const aLastDim = aFirstDims.pop();\n    a = a.reshape([-1, aLastDim]);\n\n    // Reshape y into the analogous 2D Tensor, and keep track of the\n    // required dimensions to reproduce the output shape.\n    const bShape = b.shape.slice();\n    const bLastDim = bShape.pop();\n    const ySecondLastDim = bShape.pop();\n    const yOtherDims = [...bShape, bLastDim];\n    // permutation should be like [r-2, 0, 1, 2, ... r-4, r-3, r-1]\n    // where r is the rank of y.\n    const perm = Array.from({length: b.rank}, (_, i) => {\n      if (i === 0) {\n        return b.rank - 2;\n      } else if (i <= b.rank - 2) {\n        return i - 1;\n      }\n      return i;\n    });\n    b = b.transpose(perm).reshape([ySecondLastDim, -1]);\n\n    // Multiply x and y as 2D Tensors, and then reshape back to original.\n    const outputShape = [...aFirstDims, ...yOtherDims];\n    const transposeA = false;\n    const transposeB = false;\n    return tfc.fused\n        .matMul({\n          a,\n          b,\n          transposeA,\n          transposeB,\n          bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n          activation\n        })\n        .reshape(outputShape);\n  }\n}\n\n/**\n * Compute the sign Tensor of an input Tensor.\n *\n * Elements of the input `tf.Tensor` that are === 0 are mapped to 0.\n * Elements of the input `tf.Tensor` that are > 0 are mapped to 1.\n * Elements of the input `tf.Tensor` that are < 0 are mapped to -1.\n *\n * @param x Input `tf.Tensor`.\n * @return The sign `tf.Tensor`.\n */\nexport function sign(x: Tensor): Tensor {\n  // TODO(cais): Move to the core.\n  return tidy(() => {\n    const zerosLikeX = coreZerosLike(x);\n    const onesLikeX = coreOnesLike(x);\n    return where(\n        tfc.equal(x, zerosLikeX), zerosLikeX,\n        where(\n            tfc.greater(x, coreZerosLike(x)), onesLikeX,\n            tfc.mul(-1, onesLikeX)));\n  });\n}\n\n/**\n * Computes the one-hot representation of an integer tensor.\n * @param indices nD integer tensor of shape\n *   `(batch_size, dim1, dim2, ... dim(n-1))`\n * @param numClasses Integer, number of classes to consider.\n * @returns (n + 1)D one hot representation of the input\n *   with shape `(batch_size, dim1, dim2, ... dim(n-1), num_classes)`\n */\nexport function oneHot(indices: Tensor, numClasses: number): Tensor {\n  return tidy(() => {\n    if (indices.rank !== 1) {\n      throw new Error(\n          'Only 1D one-hot tensors are supported in the ' +\n          'deeplearn backend, at present.');\n    }\n    indices = indices.toInt();\n    return tfc.oneHot(indices as Tensor1D, numClasses).toFloat();\n  });\n}\n\n/* Elementary math functions. */\n\n/**\n * Retrieves the elements of indices `indices` in the tensor `reference`.\n * @param reference A tensor.\n * @param indices An integer tensor of indices or an `Array` of integers.\n * @param axis Axis along which to perform the gather operation.\n * @returns The result of the gathering as a tensor.\n */\nexport function gather(\n    reference: Tensor, indices: number[]|Tensor1D, axis?: number): Tensor {\n  return tidy(() => {\n    if (Array.isArray(indices)) {\n      indices = tensor1d(indices, 'int32');\n    } else {\n      indices = indices.toInt();\n    }\n    return tfc.gather(reference, indices, axis);\n  });\n}\n\n/**\n * Element-wise square.\n * @param x Input tensor.\n * @return element-wise x^2\n */\nexport function square(x: Tensor): Tensor {\n  return tfc.mul(x, x);\n}\n\n/**\n * Element-wise exponentiation.\n *\n * Porting Note: In PyKeras, `a` (the exponent) is a Python integer, which\n *   takes advatnage of the backend's (e.g., TensorFlow's) automatic\n * conversion to tensor. Here we allow `a` to be either a number or a tensor.\n *\n * @param x The base tensor.\n * @param a The exponent, tensor or number. If a number, it is rounded to the\n *   nearest integer and converted to a tensor.\n * @returns A tensor of the same shape as `x`.\n */\nexport function pow(x: Tensor, a: Tensor|number): Tensor {\n  return tidy(() => {\n    if (typeof (a) === 'number') {\n      a = scalar(Math.round(a), 'int32');\n    }\n    if (a.dtype !== 'int32') {\n      throw new NotImplementedError(\n          `Non-int32 dtype (${a.dtype}) is not supported by pow() yet`);\n    }\n    return tfc.pow(x, a);\n  });\n}\n\n/**\n * Reshapes bias tensor according to rank of x.\n */\nfunction reshapeBias(xRank: number, bias: Tensor, dataFormat: string) {\n  const biasShape = bias.shape;\n\n  if (bias.rank !== 1 && bias.rank !== xRank) {\n    throw new ValueError(\n        `Unexpected bias dimensions: ${bias.rank}` +\n        `; expected it to be 1 or ${xRank}`);\n  }\n\n  if (xRank === 5) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1, 1]);\n      } else {\n        return bias.reshape(\n            [1, biasShape[3], biasShape[0], biasShape[1], biasShape[2]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 4) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1]);\n      } else {\n        return bias.reshape([1, biasShape[2], biasShape[0], biasShape[1]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 3) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1]);\n      } else {\n        return bias.reshape([1, biasShape[1], biasShape[0]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank < 3) {\n    return bias;\n  }\n  throw new ValueError(`Unsupported input rank by biasAdd: ${bias.rank}`);\n}\n\n/* Neural-network operations. */\n\n/**\n * Add a bias to a tensor.\n *\n * @param x The tensor to add the bias to.\n * @param bias The bias to add to `x`. Must be 1D or the same rank as `x`.\n * @return Result of the bias adding.\n * @throws ValueError: If the rank of `bias` is incorrect.\n */\nexport function biasAdd(\n    x: Tensor, bias: Tensor, dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n\n    return x.add(reshapeBias(x.rank, bias, dataFormat));\n  });\n}\n\n/**\n * Exponential linear unit (ELU).\n * @param x A tensor or variable to compute the activation function for.\n * @param alpha: A scalar, a scaling factor for the negative section.\n * @return Output of the ELU operation.\n */\nexport function elu(x: Tensor, alpha = 1): Tensor {\n  // TODO(cais): Add support for alpha values other than 1.\n  if (alpha !== 1) {\n    throw new NotImplementedError(\n        `Support for alpha values other than 1 (${alpha}) is not implemented ` +\n        `yet.`);\n  }\n  return tfc.elu(x);\n}\n\n/**\n * Softsign of a tensor.\n *\n * Defined as x / (abs(x) + 1), element-wise.\n *\n * @param x: Input.\n * @returns Output.\n */\nexport function softsign(x: Tensor): Tensor {\n  return tidy(() => tfc.div(x, tfc.abs(x).add(1)));\n}\n\n/**\n * Sets entries in `x` to zero at random, while scaling the entire tensor.\n *\n * @param x input tensor.\n * @param level fraction of the entries in the tensor that will be set to 0.\n * @param noiseShape shape of randomly generated keep/drop flags, must be\n *   broadcastable to the shape of `x`. Optional.\n * @param seed random seed to ensure determinism. Optional.\n * @returns Result of the dropout operation.\n */\nexport function dropout(\n    x: Tensor, level: number, noiseShape?: number[], seed?: number): Tensor {\n  return tidy(() => tfc.dropout(x, level, noiseShape, seed));\n}\n\n/**\n * Element-wise, segment-wise linear approximation of sigmoid.\n *\n * Returns `0.` if `x < -2.5`, `1.` if `x > 2.5`.\n * In `-2.5 <= x <= 2.5`, returns `0.2 * x + 0.5`.\n *\n * @param x Input tensor.\n * @returns Output tensor.\n */\nexport function hardSigmoid(x: Tensor): Tensor {\n  return tidy(() => {\n    const y = tfc.add(.5, tfc.mul(.2, x));\n    return tfc.clipByValue(y, 0, 1);\n  });\n}\n\n/**\n * Invoke `x` in the training phase, and `alt` otherwise.\n *\n * Porting Note: We do not create placeholder tensors for the `training`\n * boolean flag here, because there is no such thing in the TF.js imperative\n * backend.\n *\n * @param x The function to invoke iff `training` is `true`.\n * @param alt The function to invoke iff `training` is `false`.\n * @param training Boolean flag for whether training phase is active.\n * @returns The return value of `x()` if `training` is `true`, or the return\n *   value of `alt()` if `training` is `false`.\n */\nexport function inTrainPhase<T>(x: () => T, alt: () => T, training = false): T {\n  return training ? x() : alt();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {BaseSerialization} from './types';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type FanMode = 'fanIn'|'fanOut'|'fanAvg';\nexport const VALID_FAN_MODE_VALUES = ['fanIn', 'fanOut', 'fanAvg'];\n\n// These constants have a snake vs. camel distinction.\nexport type FanModeSerialization = 'fan_in'|'fan_out'|'fan_avg';\n\n/** @docinline */\nexport type Distribution = 'normal'|'uniform'|'truncatedNormal';\nexport const VALID_DISTRIBUTION_VALUES =\n    ['normal', 'uniform', 'truncatedNormal'];\n// These constants have a snake vs. camel distinction.\nexport type DistributionSerialization = 'normal'|'uniform'|'truncated_normal';\n\nexport type ZerosSerialization = BaseSerialization<'Zeros', {}>;\n\nexport type OnesSerialization = BaseSerialization<'Ones', {}>;\n\nexport type ConstantConfig = {\n  value: number;\n};\n\nexport type ConstantSerialization =\n    BaseSerialization<'Constant', ConstantConfig>;\n\nexport type RandomNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type RandomNormalSerialization =\n    BaseSerialization<'RandomNormal', RandomNormalConfig>;\n\nexport type RandomUniformConfig = {\n  minval?: number;\n  maxval?: number;\n  seed?: number;\n};\n\nexport type RandomUniformSerialization =\n    BaseSerialization<'RandomUniform', RandomUniformConfig>;\n\nexport type TruncatedNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type TruncatedNormalSerialization =\n    BaseSerialization<'TruncatedNormal', TruncatedNormalConfig>;\n\nexport type VarianceScalingConfig = {\n  scale?: number;\n\n  mode?: FanModeSerialization;\n  distribution?: DistributionSerialization;\n  seed?: number;\n};\n\nexport type VarianceScalingSerialization =\n    BaseSerialization<'VarianceScaling', VarianceScalingConfig>;\n\nexport type OrthogonalConfig = {\n  seed?: number;\n  gain?: number;\n};\n\nexport type OrthogonalSerialization =\n    BaseSerialization<'Orthogonal', OrthogonalConfig>;\n\nexport type IdentityConfig = {\n  gain?: number;\n};\n\nexport type IdentitySerialization =\n    BaseSerialization<'Identity', IdentityConfig>;\n\n// Update initializerClassNames below in concert with this.\nexport type InitializerSerialization = ZerosSerialization|OnesSerialization|\n    ConstantSerialization|RandomUniformSerialization|RandomNormalSerialization|\n    TruncatedNormalSerialization|IdentitySerialization|\n    VarianceScalingSerialization|OrthogonalSerialization;\n\nexport type InitializerClassName = InitializerSerialization['class_name'];\n\n// We can't easily extract a string[] from the string union type, but we can\n// recapitulate the list, enforcing at compile time that the values are valid\n// and that we have the right number of them.\n\n/**\n * A string array of valid Initializer class names.\n *\n * This is guaranteed to match the `InitializerClassName` union type.\n */\nexport const initializerClassNames: InitializerClassName[] = [\n  'Zeros', 'Ones', 'Constant', 'RandomNormal', 'RandomUniform',\n  'TruncatedNormal', 'VarianceScaling', 'Orthogonal', 'Identity'\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, eye, linalg, mul, ones, randomUniform, scalar, serialization, Tensor, Tensor2D, tidy, truncatedNormal, zeros} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {checkDataFormat} from './common';\nimport {NotImplementedError, ValueError} from './errors';\nimport {DataFormat, Shape} from './keras_format/common';\nimport {Distribution, FanMode, VALID_DISTRIBUTION_VALUES, VALID_FAN_MODE_VALUES} from './keras_format/initializer_config';\nimport {checkStringTypeUnionValue, deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\nimport {arrayProd} from './utils/math_utils';\n\nexport function checkFanMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_FAN_MODE_VALUES, 'FanMode', value);\n}\n\nexport function checkDistribution(value?: string): void {\n  checkStringTypeUnionValue(VALID_DISTRIBUTION_VALUES, 'Distribution', value);\n}\n\n/**\n * Initializer base class.\n *\n * @doc {\n *   heading: 'Initializers', subheading: 'Classes', namespace: 'initializers'}\n */\nexport abstract class Initializer extends serialization.Serializable {\n  public fromConfigUsesCustomObjects(): boolean {\n    return false;\n  }\n  /**\n   * Generate an initial value.\n   * @param shape\n   * @param dtype\n   * @return The init value.\n   */\n  abstract apply(shape: Shape, dtype?: DataType): Tensor;\n\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport class Zeros extends Initializer {\n  /** @nocollapse */\n  static className = 'Zeros';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return zeros(shape, dtype);\n  }\n}\nserialization.registerClass(Zeros);\n\nexport class Ones extends Initializer {\n  /** @nocollapse */\n  static className = 'Ones';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return ones(shape, dtype);\n  }\n}\nserialization.registerClass(Ones);\n\nexport interface ConstantArgs {\n  /** The value for each element in the variable. */\n  value: number;\n}\n\nexport class Constant extends Initializer {\n  /** @nocollapse */\n  static className = 'Constant';\n  private value: number;\n  constructor(args: ConstantArgs) {\n    super();\n    if (typeof args !== 'object') {\n      throw new ValueError(\n          `Expected argument of type ConstantConfig but got ${args}`);\n    }\n    if (args.value === undefined) {\n      throw new ValueError(`config must have value set but got ${args}`);\n    }\n    this.value = args.value;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => mul(scalar(this.value), ones(shape, dtype)));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      value: this.value,\n    };\n  }\n}\nserialization.registerClass(Constant);\n\nexport interface RandomUniformArgs {\n  /** Lower bound of the range of random values to generate. */\n  minval?: number;\n  /** Upper bound of the range of random values to generate. */\n  maxval?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomUniform extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomUniform';\n  readonly DEFAULT_MINVAL = -0.05;\n  readonly DEFAULT_MAXVAL = 0.05;\n  private minval: number;\n  private maxval: number;\n  private seed: number;\n\n  constructor(args: RandomUniformArgs) {\n    super();\n    this.minval = args.minval || this.DEFAULT_MINVAL;\n    this.maxval = args.maxval || this.DEFAULT_MAXVAL;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return randomUniform(shape, this.minval, this.maxval, dtype);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {minval: this.minval, maxval: this.maxval, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomUniform);\n\nexport interface RandomNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomNormal';\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: RandomNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `randomNormal does not support dType ${dtype}.`);\n    }\n\n    return K.randomNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomNormal);\n\nexport interface TruncatedNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class TruncatedNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'TruncatedNormal';\n\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: TruncatedNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `truncatedNormal does not support dType ${dtype}.`);\n    }\n    return truncatedNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(TruncatedNormal);\n\nexport interface IdentityArgs {\n  /**\n   * Multiplicative factor to apply to the identity matrix.\n   */\n  gain?: number;\n}\n\nexport class Identity extends Initializer {\n  /** @nocollapse */\n  static className = 'Identity';\n  private gain: number;\n  constructor(args: IdentityArgs) {\n    super();\n    this.gain = args.gain != null ? args.gain : 1.0;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length !== 2 || shape[0] !== shape[1]) {\n        throw new ValueError(\n            'Identity matrix initializer can only be used for' +\n            ' 2D square matrices.');\n      } else {\n        return mul(this.gain, eye(shape[0]));\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {gain: this.gain};\n  }\n}\nserialization.registerClass(Identity);\n\n/**\n * Computes the number of input and output units for a weight shape.\n * @param shape Shape of weight.\n * @param dataFormat data format to use for convolution kernels.\n *   Note that all kernels in Keras are standardized on the\n *   CHANNEL_LAST ordering (even when inputs are set to CHANNEL_FIRST).\n * @return An length-2 array: fanIn, fanOut.\n */\nfunction computeFans(\n    shape: Shape, dataFormat: DataFormat = 'channelsLast'): number[] {\n  let fanIn: number;\n  let fanOut: number;\n  checkDataFormat(dataFormat);\n  if (shape.length === 2) {\n    fanIn = shape[0];\n    fanOut = shape[1];\n  } else if ([3, 4, 5].indexOf(shape.length) !== -1) {\n    if (dataFormat === 'channelsFirst') {\n      const receptiveFieldSize = arrayProd(shape, 2);\n      fanIn = shape[1] * receptiveFieldSize;\n      fanOut = shape[0] * receptiveFieldSize;\n    } else if (dataFormat === 'channelsLast') {\n      const receptiveFieldSize = arrayProd(shape, 0, shape.length - 2);\n      fanIn = shape[shape.length - 2] * receptiveFieldSize;\n      fanOut = shape[shape.length - 1] * receptiveFieldSize;\n    }\n  } else {\n    const shapeProd = arrayProd(shape);\n    fanIn = Math.sqrt(shapeProd);\n    fanOut = Math.sqrt(shapeProd);\n  }\n\n  return [fanIn, fanOut];\n}\n\nexport interface VarianceScalingArgs {\n  /** Scaling factor (positive float). */\n  scale?: number;\n\n  /** Fanning mode for inputs and outputs. */\n  mode?: FanMode;\n\n  /** Probabilistic distribution of the values. */\n  distribution?: Distribution;\n\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class VarianceScaling extends Initializer {\n  /** @nocollapse */\n  static className = 'VarianceScaling';\n  private scale: number;\n  private mode: FanMode;\n  private distribution: Distribution;\n  private seed: number;\n\n  /**\n   * Constructor of VarianceScaling.\n   * @throws ValueError for invalid value in scale.\n   */\n  constructor(args: VarianceScalingArgs) {\n    super();\n    if (args.scale < 0.0) {\n      throw new ValueError(\n          `scale must be a positive float. Got: ${args.scale}`);\n    }\n    this.scale = args.scale == null ? 1.0 : args.scale;\n    this.mode = args.mode == null ? 'fanIn' : args.mode;\n    checkFanMode(this.mode);\n    this.distribution =\n        args.distribution == null ? 'normal' : args.distribution;\n    checkDistribution(this.distribution);\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    const fans = computeFans(shape);\n    const fanIn = fans[0];\n    const fanOut = fans[1];\n    let scale = this.scale;\n    if (this.mode === 'fanIn') {\n      scale /= Math.max(1, fanIn);\n    } else if (this.mode === 'fanOut') {\n      scale /= Math.max(1, fanOut);\n    } else {\n      scale /= Math.max(1, (fanIn + fanOut) / 2);\n    }\n\n    if (this.distribution === 'normal') {\n      const stddev = Math.sqrt(scale);\n      dtype = dtype || 'float32';\n      if (dtype !== 'float32' && dtype !== 'int32') {\n        throw new NotImplementedError(\n            `${this.getClassName()} does not support dType ${dtype}.`);\n      }\n      return truncatedNormal(shape, 0, stddev, dtype, this.seed);\n    } else {\n      const limit = Math.sqrt(3 * scale);\n      return randomUniform(shape, -limit, limit, dtype);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      scale: this.scale,\n      mode: this.mode,\n      distribution: this.distribution,\n      seed: this.seed\n    };\n  }\n}\nserialization.registerClass(VarianceScaling);\n\nexport interface SeedOnlyInitializerArgs {\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class GlorotUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotUniform';\n\n  /**\n   * Constructor of GlorotUniform\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotUniform);\n\nexport class GlorotNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotNormal';\n\n  /**\n   * Constructor of GlorotNormal.\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotNormal);\n\nexport class HeNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeNormal);\n\nexport class HeUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeUniform';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeUniform);\n\nexport class LeCunNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunNormal);\n\nexport class LeCunUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunUniform);\n\nexport interface OrthogonalArgs extends SeedOnlyInitializerArgs {\n  /**\n   * Multiplicative factor to apply to the orthogonal matrix. Defaults to 1.\n   */\n  gain?: number;\n}\n\nexport class Orthogonal extends Initializer {\n  /** @nocollapse */\n  static className = 'Orthogonal';\n  readonly DEFAULT_GAIN = 1;\n  protected readonly gain: number;\n  protected readonly seed: number;\n\n  constructor(args?: OrthogonalArgs) {\n    super();\n    this.gain = args.gain == null ? this.DEFAULT_GAIN : args.gain;\n    this.seed = args.seed;\n\n    if (this.seed != null) {\n      throw new NotImplementedError(\n          'Random seed is not implemented for Orthogonal Initializer yet.');\n    }\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length < 2) {\n        throw new NotImplementedError('Shape must be at least 2D.');\n      }\n      if (shape[0] * shape[1] > 2000) {\n        console.warn(\n            `Orthogonal initializer is being called on a matrix with more ` +\n            `than 2000 (${shape[0] * shape[1]}) elements: ` +\n            `Slowness may result.`);\n      }\n\n      // TODO(cais): Add seed support.\n      const normalizedShape =\n          shape[0] > shape[1] ? [shape[1], shape[0]] : shape;\n      const a = K.randomNormal(normalizedShape, 0, 1, 'float32') as Tensor2D;\n      let q = linalg.gramSchmidt(a) as Tensor2D;\n      if (shape[0] > shape[1]) {\n        q = q.transpose();\n      }\n      return mul(this.gain, q);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      gain: this.gain,\n      seed: this.seed,\n    };\n  }\n}\nserialization.registerClass(Orthogonal);\n\n/** @docinline */\nexport type InitializerIdentifier =\n    'constant'|'glorotNormal'|'glorotUniform'|'heNormal'|'heUniform'|'identity'|\n    'leCunNormal'|'leCunUniform'|'ones'|'orthogonal'|'randomNormal'|\n    'randomUniform'|'truncatedNormal'|'varianceScaling'|'zeros'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in InitializerIdentifier]: string} = {\n      'constant': 'Constant',\n      'glorotNormal': 'GlorotNormal',\n      'glorotUniform': 'GlorotUniform',\n      'heNormal': 'HeNormal',\n      'heUniform': 'HeUniform',\n      'identity': 'Identity',\n      'leCunNormal': 'LeCunNormal',\n      'leCunUniform': 'LeCunUniform',\n      'ones': 'Ones',\n      'orthogonal': 'Orthogonal',\n      'randomNormal': 'RandomNormal',\n      'randomUniform': 'RandomUniform',\n      'truncatedNormal': 'TruncatedNormal',\n      'varianceScaling': 'VarianceScaling',\n      'zeros': 'Zeros'\n    };\n\nfunction deserializeInitializer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Initializer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'initializer');\n}\n\nexport function serializeInitializer(initializer: Initializer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(initializer);\n}\n\nexport function getInitializer(identifier: InitializerIdentifier|Initializer|\n                               serialization.ConfigDict): Initializer {\n  if (typeof identifier === 'string') {\n    const className = identifier in INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    /* We have four 'helper' classes for common initializers that\n    all get serialized as 'VarianceScaling' and shouldn't go through\n    the deserializeInitializer pathway. */\n    if (className === 'GlorotNormal') {\n      return new GlorotNormal();\n    } else if (className === 'GlorotUniform') {\n      return new GlorotUniform();\n    } else if (className === 'HeNormal') {\n      return new HeNormal();\n    } else if (className === 'HeUniform') {\n      return new HeUniform();\n    } else if (className === 'LeCunNormal') {\n      return new LeCunNormal();\n    } else if (className === 'LeCunUniform') {\n      return new LeCunUniform();\n    } else {\n      const config: serialization.ConfigDict = {};\n      config['className'] = className;\n      config['config'] = {};\n      return deserializeInitializer(config);\n    }\n  } else if (identifier instanceof Initializer) {\n    return identifier;\n  } else {\n    return deserializeInitializer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constant, ConstantArgs, GlorotNormal, GlorotUniform, HeNormal, HeUniform, Identity, IdentityArgs, Initializer, LeCunNormal, LeCunUniform, Ones, Orthogonal, OrthogonalArgs, RandomNormal, RandomNormalArgs, RandomUniform, RandomUniformArgs, SeedOnlyInitializerArgs, TruncatedNormal, TruncatedNormalArgs, VarianceScaling, VarianceScalingArgs, Zeros} from './initializers';\n\n/**\n * Initializer that generates tensors initialized to 0.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function zeros(): Zeros {\n  return new Zeros();\n}\n\n/**\n * Initializer that generates tensors initialized to 1.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function ones(): Initializer {\n  return new Ones();\n}\n\n/**\n * Initializer that generates values initialized to some constant.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function constant(args: ConstantArgs): Initializer {\n  return new Constant(args);\n}\n\n/**\n * Initializer that generates random values initialized to a uniform\n * distribution.\n *\n * Values will be distributed uniformly between the configured minval and\n * maxval.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomUniform(args: RandomUniformArgs): Initializer {\n  return new RandomUniform(args);\n}\n\n/**\n * Initializer that generates random values initialized to a normal\n * distribution.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomNormal(args: RandomNormalArgs): Initializer {\n  return new RandomNormal(args);\n}\n\n/**\n * Initializer that generates random values initialized to a truncated normal.\n * distribution.\n *\n * These values are similar to values from a `RandomNormal` except that values\n * more than two standard deviations from the mean are discarded and re-drawn.\n * This is the recommended initializer for neural network weights and filters.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function truncatedNormal(args: TruncatedNormalArgs): Initializer {\n  return new TruncatedNormal(args);\n}\n\n/**\n * Initializer that generates the identity matrix.\n * Only use for square 2D matrices.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function identity(args: IdentityArgs): Initializer {\n  return new Identity(args);\n}\n\n/**\n * Initializer capable of adapting its scale to the shape of weights.\n * With distribution=NORMAL, samples are drawn from a truncated normal\n * distribution centered on zero, with `stddev = sqrt(scale / n)` where n is:\n *   - number of input units in the weight tensor, if mode = FAN_IN.\n *   - number of output units, if mode = FAN_OUT.\n *   - average of the numbers of input and output units, if mode = FAN_AVG.\n * With distribution=UNIFORM,\n * samples are drawn from a uniform distribution\n * within [-limit, limit], with `limit = sqrt(3 * scale / n)`.\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function varianceScaling(config: VarianceScalingArgs): Initializer {\n  return new VarianceScaling(config);\n}\n\n/**\n * Glorot uniform initializer, also called Xavier uniform initializer.\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotUniform(args);\n}\n\n/**\n * Glorot normal initializer, also called Xavier normal initializer.\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor.\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotNormal(args);\n}\n\n/**\n * He normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function heNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeNormal(args);\n}\n\n/**\n * He uniform initializer.\n *\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / fan_in)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function heUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeUniform(args);\n}\n\n/**\n * LeCun normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(1 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * References:\n *   [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *   [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunNormal(args);\n}\n\n/**\n * LeCun uniform initializer.\n *\n * It draws samples from a uniform distribution in the interval\n * `[-limit, limit]` with `limit = sqrt(3 / fanIn)`,\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunUniform(args);\n}\n\n/**\n * Initializer that generates a random orthogonal matrix.\n *\n * Reference:\n * [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function orthogonal(args: OrthogonalArgs): Initializer {\n  return new Orthogonal(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Utilities related to persistent state in the backend.\n */\n\n/**\n * An ID to track `tf.SymbolicTensor`s and derived classes.\n * Required in different places in engine/topology.ts to identify unique\n * tensors.\n */\nlet _nextUniqueTensorId = 0;\n\nexport function getNextUniqueTensorId(): number {\n  return _nextUniqueTensorId++;\n}\n\nconst _uidPrefixes: {[prefix: string]: number} = {};\n\n/**\n * Provides a unique UID given a string prefix.\n *\n * @param prefix\n */\nexport function getUid(prefix = ''): string {\n  if (!(prefix in _uidPrefixes)) {\n    _uidPrefixes[prefix] = 0;\n  }\n  _uidPrefixes[prefix] += 1;\n  return prefix + _uidPrefixes[prefix].toString();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\n// tslint:enable\n\n/**\n * Determine whether the input is an Array of Shapes.\n */\nexport function isArrayOfShapes(x: Shape|Shape[]): boolean {\n  return Array.isArray(x) && Array.isArray(x[0]);\n}\n\n/**\n * Special case of normalizing shapes to lists.\n *\n * @param x A shape or list of shapes to normalize into a list of Shapes.\n * @return A list of Shapes.\n */\nexport function normalizeShapeList(x: Shape|Shape[]): Shape[] {\n  if (x.length === 0) {\n    return [];\n  }\n  if (!Array.isArray(x[0])) {\n    return [x] as Shape[];\n  }\n  return x as Shape[];\n}\n\n/**\n * Helper function to obtain exactly one Tensor.\n * @param xs: A single `tf.Tensor` or an `Array` of `tf.Tensor`s.\n * @return A single `tf.Tensor`. If `xs` is an `Array`, return the first one.\n * @throws ValueError: If `xs` is an `Array` and its length is not 1.\n */\nexport function getExactlyOneTensor(xs: Tensor|Tensor[]): Tensor {\n  let x: Tensor;\n  if (Array.isArray(xs)) {\n    if (xs.length !== 1) {\n      throw new ValueError(`Expected Tensor length to be 1; got ${xs.length}`);\n    }\n    x = xs[0];\n  } else {\n    x = xs;\n  }\n  return x;\n}\n\n/**\n * Helper function to obtain exactly on instance of Shape.\n *\n * @param shapes Input single `Shape` or Array of `Shape`s.\n * @returns If input is a single `Shape`, return it unchanged. If the input is\n *   an `Array` containing exactly one instance of `Shape`, return the instance.\n *   Otherwise, throw a `ValueError`.\n * @throws ValueError: If input is an `Array` of `Shape`s, and its length is not\n *   1.\n */\nexport function getExactlyOneShape(shapes: Shape|Shape[]): Shape {\n  if (Array.isArray(shapes) && Array.isArray(shapes[0])) {\n    if (shapes.length === 1) {\n      shapes = shapes as Shape[];\n      return shapes[0];\n    } else {\n      throw new ValueError(`Expected exactly 1 Shape; got ${shapes.length}`);\n    }\n  } else {\n    return shapes as Shape;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {LayerVariable} from '../variables';\n\n/**\n * Count the elements in an Array of LayerVariables.\n *\n * @param weights: The LayerVariables of which the constituent numbers are to\n *   be counted.\n * @returns A count of the elements in all the LayerVariables\n */\nexport function countParamsInWeights(weights: LayerVariable[]): number {\n  let count = 0;\n  for (const weight of weights) {\n    if (weight.shape.length === 0) {\n      count += 1;\n    } else {\n      count += weight.shape.reduce((a, b) => a * b);\n    }\n  }\n  return count;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, Tensor, variableGrads} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId} from './backend/state';\nimport {getScopedTensorName, getUniqueTensorName} from './common';\nimport {Constraint} from './constraints';\nimport {NotImplementedError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {HasShape} from './types';\n\nconst DEFAULT_VARIABLE_NAME_PREFIX = 'Variable';\n\n/**\n * A `tf.layers.LayerVariable` is similar to a `tf.Tensor` in that it has a\n * dtype and shape, but its value is mutable.  The value is itself represented\n * as a`tf.Tensor`, and can be read with the `read()` method and updated with\n * the `write()` method.\n */\nexport class LayerVariable {\n  readonly dtype: DataType;\n  readonly shape: Shape;\n\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName: string;\n  private trainable_: boolean;\n\n  protected readonly val: tfc.Variable;\n  readonly constraint: Constraint;\n  /**\n   * Construct Variable from a `tf.Tensor`.\n   *\n   * If not explicitly named, the Variable will be given a name with the\n   * prefix 'Variable'. Variable names are unique. In the case of name\n   * collision, suffixies '_<num>' will be added to the name.\n   *\n   * @param val Initial value of the Variable.\n   * @param name Name of the variable. If `null` or `undefined` is provided, it\n   *   will default a name with the prefix 'Variable'.\n   * @param constraint Optional, projection function to be applied to the\n   * variable after optimize updates\n   * @throws ValueError if `name` is `null` or `undefined`.\n   */\n  constructor(\n      val: Tensor, dtype: DataType = 'float32',\n      name = DEFAULT_VARIABLE_NAME_PREFIX, trainable = true,\n      constraint: Constraint = null) {\n    this.dtype = dtype == null ? 'float32' : dtype;\n    this.shape = val.shape;\n    this.id = getNextUniqueTensorId();\n\n    name = name == null ? DEFAULT_VARIABLE_NAME_PREFIX : name;\n    this.originalName = getScopedTensorName(name);\n    this.name = getUniqueTensorName(this.originalName);\n\n    this.trainable_ = trainable;\n    this.constraint = constraint;\n\n    this.val = tfc.variable(val, this.trainable_, this.name, this.dtype);\n  }\n\n  /**\n   * Get a snapshot of the Variable's value.\n   *\n   * The returned value is a snapshot of the Variable's value at the time of\n   * the invocation. Future mutations in the value of the tensor will only\n   * be reflected by future calls to this method.\n   */\n  read(): Tensor {\n    this.assertNotDisposed();\n    return this.val;\n  }\n\n  /**\n   * Update the value of the Variable.\n   *\n   * @param newVal: The new value to update to. Must be consistent with the\n   *   dtype and shape of the Variable.\n   * @return This Variable.\n   */\n  write(newVal: Tensor) {\n    // TODO(cais): Once  TF.js Core supports Tensor.dtype, check dtype match.\n    this.assertNotDisposed();\n    checkShapesMatch(this.val, newVal);\n    // Skip updating if this is the exact same tensor.\n    if (this.val.id !== newVal.id) {\n      this.val.assign(newVal);\n      if (this.constraint != null) {\n        this.val.assign(this.constraint.apply(this.val));\n      }\n    }\n    return this;\n  }\n\n  /**\n   * Dispose this LayersVariable instance from memory.\n   */\n  dispose(): void {\n    this.assertNotDisposed();\n    this.val.dispose();\n  }\n\n  protected assertNotDisposed(): void {\n    if (this.val.isDisposed) {\n      throw new Error(`LayersVariable ${this.name} is already disposed.`);\n    }\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.trainable_ = trainable;\n    this.val.trainable = trainable;\n  }\n}\n\nfunction checkShapesMatch(x: HasShape, y: HasShape): void {\n  if (x.shape.toString() !== y.shape.toString()) {\n    throw new Error(\n        'Shape mismatch: ' + JSON.stringify(x.shape) + ' vs. ' +\n        JSON.stringify(y.shape));\n  }\n}\n\n/**\n * Create a Variable.\n * @param x The initial value of the `Variable`.\n * @param dtype optional, the type of the variable.\n * @param name optional, the name of the variable, default provided by\n * Variable.\n * @param constraint optional, a constraint to be applied after every update.\n * @return The newly instantiated `Variable`.\n */\nexport function variable(\n    x: Tensor, dtype?: DataType, name?: string,\n    constraint?: Constraint): LayerVariable {\n  return new LayerVariable(x, dtype, name, true, constraint);\n}\n\n/**\n * Instantiates an all-zeros Variable and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-zero Variable.\n */\nexport function zerosVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  return new LayerVariable(tfc.zeros(shape), dtype, name);\n}\n\n/**\n * Instantiates an all-zeros tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function zerosLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.zerosLike(x), dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-ones Variable.\n */\nexport function onesVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  const allocated = tfc.ones(shape);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function onesLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  const allocated = tfc.onesLike(x);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiate an identity matrix and returns it, as a Variable\n *\n * @param size Number of rows/columns.\n * @param dtype Data type of returned Variable.\n * @param name Name of returned Variable.\n * @return A Variable, an identity matrix.\n */\nexport function eyeVariable(\n    size: number, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.eye(size), dtype, name);\n}\n\n/**\n * Get a Variable with uniform distribution of values.\n * @param shape Shape of the tensor.\n * @param minval Lower bound of the uniform distribution.\n * @param maxval Upper bound of the uniform distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The uniform-random Variable.\n */\nexport function randomUniformVariable(\n    shape: Shape, minval: number, maxval: number, dtype?: DataType,\n    seed?: number, name = 'randomUniform'): LayerVariable {\n  return new LayerVariable(\n      tfc.randomUniform(shape, minval, maxval, dtype), dtype, name);\n}\n\n/**\n * Get a Variable with truncated-normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function truncatedNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'truncatedNormal'): LayerVariable {\n  // TODO(cais): Implement logic for dtype and seed once they are supported\n  // by deeplearn.js.\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormal does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.truncatedNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n/**\n * Get a Variable with normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function randomNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'randomNormal'): LayerVariable {\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormalVariable does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.randomNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n\n/**\n * Update the value of a Variable.\n * @param x The Variable to be updated.\n * @param xNew The new value to update to.\n * @return The Variable updated.\n */\nexport function update(x: LayerVariable, xNew: Tensor): LayerVariable {\n  return x.write(xNew);\n}\n\n/**\n * Update the value of a Variable by adding an increment.\n * @param x The Variable to be updated.\n * @param increment The incrment to add to `x`.\n * @return The Variable updated.\n */\nexport function updateAdd(x: LayerVariable, increment: Tensor): LayerVariable {\n  return x.write(tfc.add(x.read(), increment));\n}\n\n/**\n * Update the value of a Variable by subtracting a decrement.\n * @param x The Variable to be updated.\n * @param decrement The decrement to subtract from `x`.\n * @return The Variable updated.\n */\nexport function updateSub(x: LayerVariable, decrement: Tensor): LayerVariable {\n  return x.write(tfc.sub(x.read(), decrement));\n}\n\n/**\n * Get the values of an array of Variables.\n *\n * @param tensors An `Array` of `Variable`s to get the values of.\n * @return The values of the inputs, as an `Array` of`tf.Tensor`s.\n */\nexport function batchGetValue(xs: LayerVariable[]): Tensor[] {\n  return xs.map(x => x.read());\n}\n\n/**\n * Update the value of multiple Variables at once.\n *\n * @param variablesAndValues An `Array`, each element is of type\n *   [Variable, Tensor]. The first item is the\n *   `Variable` of which the value is to be updated. The second item\n *   carries the new value.\n */\nexport function batchSetValue(\n    variablesAndValues: Array<[LayerVariable, Tensor]>): void {\n  variablesAndValues.forEach(variableAndValue => {\n    const variable: LayerVariable = variableAndValue[0];\n    variable.write(variableAndValue[1]);\n  });\n}\n\n/**\n * Returns the gradients of `variables` w.r.t. the return value of `lossFn`.\n * @param lossFn A function which returns a Scalar to be used as the function\n *   value (i.e., numerator) for differentiation.\n * @param variables List of variables to be used as the independent variables\n *   (i.e., denominator) for differentiation.\n * @returns An Array of gradients tensors.\n */\nexport function gradients(\n    lossFn: () => tfc.Scalar, variables: LayerVariable[]): Tensor[] {\n  // TODO(cais): The return type signature can be simplified if deeplearn makes\n  //   the corresponding type public.\n  const variableList =\n      variables.map(variable => variable.read() as tfc.Variable);\n  const valudAndGrads = variableGrads(lossFn, variableList);\n  return variables.map(variable => valudAndGrads.grads[variable.name]);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {DataType, Scalar, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId, getUid} from '../backend/state';\nimport {getScopedTensorName, getUniqueTensorName, nameScope} from '../common';\nimport {Constraint} from '../constraints';\nimport {AttributeError, NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {getInitializer, Initializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs, RegularizerFn} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as types_utils from '../utils/types_utils';\nimport * as variable_utils from '../utils/variable_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\n\n// TODO(michaelterry): This is a stub until it's defined.\nexport type Op = (x: LayerVariable) => LayerVariable;\n\n/**\n * Constructor arguments for InputSpec.\n */\nexport interface InputSpecArgs {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n}\n\n/**\n * Specifies the ndim, dtype and shape of every input to a layer.\n *\n * Every layer should expose (if appropriate) an `inputSpec` attribute:\n * a list of instances of InputSpec (one per input tensor).\n *\n * A null entry in a shape is compatible with any dimension,\n * a null shape is compatible with any shape.\n */\nexport class InputSpec {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n\n  constructor(args: InputSpecArgs) {\n    this.dtype = args.dtype;\n    this.shape = args.shape;\n    /*\n      TODO(michaelterry): Could throw error if ndim and shape are both defined\n        (then backport).\n    */\n    if (args.shape != null) {\n      this.ndim = args.shape.length;\n    } else {\n      this.ndim = args.ndim;\n    }\n    this.maxNDim = args.maxNDim;\n    this.minNDim = args.minNDim;\n    this.axes = args.axes || {};\n  }\n}\n\n/**\n * `tf.SymbolicTensor` is a placeholder for a Tensor without any concrete value.\n *\n * They are most often encountered when building a graph of `Layer`s for a\n * a `tf.LayersModel` and the input data's shape, but not values are known.\n *\n * @doc {heading: 'Models', 'subheading': 'Classes'}\n */\nexport class SymbolicTensor {\n  /* A unique ID for the tensor to be able to differentiate tensors. */\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName?: string;\n  /**\n   * Rank/dimensionality of the tensor.\n   */\n  readonly rank: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  nodeIndex: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  tensorIndex: number;\n\n  /**\n   *\n   * @param dtype\n   * @param shape\n   * @param sourceLayer The Layer that produced this symbolic tensor.\n   * @param inputs The inputs passed to sourceLayer's __call__() method.\n   * @param nodeIndex\n   * @param tensorIndex\n   * @param callArgs The keyword arguments passed to the __call__() method.\n   * @param name\n   * @param outputTensorIndex The index of this tensor in the list of outputs\n   *   returned by apply().\n   */\n  constructor(\n      readonly dtype: DataType, readonly shape: Shape,\n      public sourceLayer: Layer, readonly inputs: SymbolicTensor[],\n      readonly callArgs: Kwargs, name?: string,\n      readonly outputTensorIndex?: number) {\n    this.id = getNextUniqueTensorId();\n    if (name != null) {\n      this.originalName = getScopedTensorName(name);\n      this.name = getUniqueTensorName(this.originalName);\n    }\n    this.rank = shape.length;\n  }\n}\n\n/**\n * Constructor arguments for Node.\n */\nexport interface NodeArgs {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`.\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n}\n\n/**\n * The type of the return value of Layer.dispose() and Container.dispose().\n */\nexport interface DisposeResult {\n  /**\n   * Reference count after the dispose call.\n   */\n  refCountAfterDispose: number;\n\n  /**\n   * Number of variables dispose in this dispose call.\n   */\n  numDisposedVariables: number;\n}\n\nlet _nextNodeID = 0;\n\n/**\n * A `Node` describes the connectivity between two layers.\n *\n * Each time a layer is connected to some new input,\n * a node is added to `layer.inboundNodes`.\n *\n * Each time the output of a layer is used by another layer,\n * a node is added to `layer.outboundNodes`.\n *\n * `nodeIndices` and `tensorIndices` are basically fine-grained coordinates\n * describing the origin of the `inputTensors`, verifying the following:\n *\n * `inputTensors[i] ==\n * inboundLayers[i].inboundNodes[nodeIndices[i]].outputTensors[\n *   tensorIndices[i]]`\n *\n * A node from layer A to layer B is added to:\n *     A.outboundNodes\n *     B.inboundNodes\n */\nexport class Node {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n\n  readonly id: number;\n\n  constructor(\n      args: NodeArgs,\n      // TODO(michaelterry): Define actual type for this.\n      public callArgs?: Kwargs) {\n    this.id = _nextNodeID++;\n    /*\n      Layer instance (NOT a list).\n      this is the layer that takes a list of input tensors\n      and turns them into a list of output tensors.\n      the current node will be added to\n      the inboundNodes of outboundLayer.\n    */\n    this.outboundLayer = args.outboundLayer;\n\n    /*\n        The following 3 properties describe where\n        the input tensors come from: which layers,\n        and for each layer, which node and which\n        tensor output of each node.\n    */\n\n    // List of layer instances.\n    this.inboundLayers = args.inboundLayers;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.nodeIndices = args.nodeIndices;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.tensorIndices = args.tensorIndices;\n\n    /*\n        Following 2 properties:\n        tensor inputs and outputs of outboundLayer.\n    */\n\n    // List of tensors. 1:1 mapping with inboundLayers.\n    this.inputTensors = args.inputTensors;\n    // List of tensors, created by outboundLayer.call().\n    this.outputTensors = args.outputTensors;\n\n    /*\n        Following 2 properties: input and output masks.\n        List of tensors, 1:1 mapping with inputTensor.\n    */\n    this.inputMasks = args.inputMasks;\n    // List of tensors, created by outboundLayer.computeMask().\n    this.outputMasks = args.outputMasks;\n\n    // Following 2 properties: input and output shapes.\n\n    // List of shape tuples, shapes of inputTensors.\n    this.inputShapes = args.inputShapes;\n    // List of shape tuples, shapes of outputTensors.\n    this.outputShapes = args.outputShapes;\n\n    // Add nodes to all layers involved.\n    for (const layer of args.inboundLayers) {\n      if (layer != null) {\n        layer.outboundNodes.push(this);\n      }\n    }\n    args.outboundLayer.inboundNodes.push(this);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const inboundNames: string[] = [];\n    for (const layer of this.inboundLayers) {\n      if (layer != null) {\n        inboundNames.push(layer.name);\n      } else {\n        inboundNames.push(null);\n      }\n    }\n    return {\n      outboundLayer: this.outboundLayer ? this.outboundLayer.name : null,\n      inboundLayers: inboundNames,\n      nodeIndices: this.nodeIndices,\n      tensorIndices: this.tensorIndices\n    };\n  }\n}\n\n/** Constructor arguments for Layer. */\nexport declare interface LayerArgs {\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  inputShape?: Shape;\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  batchInputShape?: Shape;\n  /**\n   * If `inputShape` is specified and `batchInputShape` is *not* specified,\n   * `batchSize` is used to construct the `batchInputShape`: `[batchSize,\n   * ...inputShape]`\n   */\n  batchSize?: number;\n  /**\n   * The data-type for this layer. Defaults to 'float32'.\n   * This argument is only applicable to input layers (the first layer of a\n   * model).\n   */\n  dtype?: DataType;\n  /** Name for this layer. */\n  name?: string;\n  /**\n   * Whether the weights of this layer are updatable by `fit`.\n   * Defaults to true.\n   */\n  trainable?: boolean;\n  /**\n   * Initial weight values of the layer.\n   */\n  weights?: Tensor[];\n  /** Legacy support. Do not use for new code. */\n  inputDType?: DataType;\n}\n\n// If necessary, add `output` arguments to the CallHook function.\n// This is currently used for testing only, but may be used for debugger-related\n// purposes in the future.\nexport type CallHook = (inputs: Tensor|Tensor[], kwargs: Kwargs) => void;\n\nlet _nextLayerID = 0;\n\n/**\n * A layer is a grouping of operations and weights that can be composed to\n * create a `tf.LayersModel`.\n *\n * Layers are constructed by using the functions under the\n * [tf.layers](#Layers-Basic) namespace.\n *\n * @doc {heading: 'Layers', subheading: 'Classes', namespace: 'layers'}\n */\nexport abstract class Layer extends serialization.Serializable {\n  /** Name for this layer. Must be unique within a model. */\n  name: string;\n  /**\n   * List of InputSpec class instances.\n   *\n   * Each entry describes one required input:\n   * - ndim\n   * - dtype\n   * A layer with `n` input tensors must have an `inputSpec` of length `n`.\n   */\n  inputSpec: InputSpec[];\n  supportsMasking: boolean;\n  /** Whether the layer weights will be updated during training. */\n  protected trainable_: boolean;\n  batchInputShape: Shape;\n  dtype: DataType;\n  initialWeights: Tensor[];\n\n  inboundNodes: Node[];\n  outboundNodes: Node[];\n\n  activityRegularizer: Regularizer;\n\n  protected _trainableWeights: LayerVariable[];\n  private _nonTrainableWeights: LayerVariable[];\n  private _losses: RegularizerFn[];\n  // TODO(cais): _updates is currently unused.\n  private _updates: Tensor[];\n  private _built: boolean;\n  private _callHook: CallHook = null;\n\n  private _addedWeightNames: string[] = [];\n\n  readonly id: number;\n\n  // Porting Notes: PyKeras does not have this property in this base Layer\n  //   class. Instead lets Layer subclass set it dynamically and checks the\n  //   value with `hasattr`. In tfjs-layers, we let this be a member of this\n  //   base class.\n  protected _stateful = false;\n\n  protected _refCount: number|null;\n\n  // A flag for whether fast (i.e., all-zero) weight initialization is to\n  // be used during `build()` call. This speeds up weight initialization\n  // by saving unnecessary calls to expensive initializers in cases where\n  // the initialized values will be overwritten by loaded weight values\n  // during model loading.\n  private fastWeightInitDuringBuild: boolean;\n\n  constructor(args: LayerArgs = {}) {\n    super();\n    this.id = _nextLayerID++;\n\n    this.activityRegularizer = null;\n\n    this.inputSpec = null;\n    this.supportsMasking = false;\n\n    // These properties will be set upon call of this.build()\n    this._trainableWeights = [];\n    this._nonTrainableWeights = [];\n    this._losses = [];\n    this._updates = [];\n    this._built = false;\n\n    /*\n      These lists will be filled via successive calls\n      to this.addInboundNode().\n     */\n    this.inboundNodes = [];\n    this.outboundNodes = [];\n\n    let name = args.name;\n    if (!name) {\n      const prefix = this.getClassName();\n      name = generic_utils.toSnakeCase(prefix) + '_' + getUid(prefix);\n    }\n    this.name = name;\n\n    this.trainable_ = args.trainable == null ? true : args.trainable;\n\n    if (args.inputShape != null || args.batchInputShape != null) {\n      /*\n        In this case we will later create an input layer\n        to insert before the current layer\n       */\n      let batchInputShape: Shape;\n      if (args.batchInputShape != null) {\n        batchInputShape = args.batchInputShape;\n      } else if (args.inputShape != null) {\n        let batchSize: number = null;\n        if (args.batchSize != null) {\n          batchSize = args.batchSize;\n        }\n        batchInputShape = [batchSize].concat(args.inputShape);\n      }\n      this.batchInputShape = batchInputShape;\n\n      // Set dtype.\n      let dtype = args.dtype;\n      if (dtype == null) {\n        dtype = args.inputDType;\n      }\n      if (dtype == null) {\n        dtype = 'float32';\n      }\n      this.dtype = dtype;\n    }\n\n    if (args.weights != null) {\n      this.initialWeights = args.weights;\n    } else {\n      this.initialWeights = null;\n    }\n\n    // The value of `_refCount` is initialized to null. When the layer is used\n    // in a symbolic way for the first time, it will be set to 1.\n    this._refCount = null;\n\n    this.fastWeightInitDuringBuild = false;\n  }\n\n  /**\n   * Converts a layer and its index to a unique (immutable type) name.\n   * This function is used internally with `this.containerNodes`.\n   * @param layer The layer.\n   * @param nodeIndex The layer's position (e.g. via enumerate) in a list of\n   *   nodes.\n   *\n   * @returns The unique name.\n   */\n  protected static nodeKey(layer: Layer, nodeIndex: number) {\n    return layer.name + '_ib-' + nodeIndex.toString();\n  }\n\n  /**\n   * Returns this.inboundNode at index nodeIndex.\n   *\n   * Porting note: This is a replacement for _get_node_attribute_at_index()\n   * @param nodeIndex\n   * @param attrName The name of the attribute related to request for this node.\n   */\n  private getNodeAtIndex(nodeIndex: number, attrName: string): Node {\n    if (this.inboundNodes.length === 0) {\n      throw new RuntimeError(\n          'The layer has never been called ' +\n          `and thus has no defined ${attrName}.`);\n    }\n    if (this.inboundNodes.length <= nodeIndex) {\n      throw new ValueError(\n          `Asked to get ${attrName} at node ${nodeIndex}, ` +\n          `but the layer has only ${this.inboundNodes.length} inbound nodes.`);\n    }\n    return this.inboundNodes[nodeIndex];\n  }\n\n  /**\n   * Retrieves the input tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple inputs).\n   */\n  getInputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple outputs).\n   */\n  getOutputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'output').outputTensors);\n  }\n\n  // Properties\n\n  /**\n   * Retrieves the input tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Input tensor or list of input tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get input(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer input\" ' +\n          'is ill-defined. ' +\n          'Use `getInputAt(nodeIndex)` instead.');\n    } else if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' is not connected, no input to return.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Output tensor or list of output tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get output(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has no inbound nodes.');\n    }\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer output\" ' +\n          'is ill-defined. ' +\n          'Use `getOutputAt(nodeIndex)` instead.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'output').outputTensors);\n  }\n\n  get losses(): RegularizerFn[] {\n    return this._losses;\n  }\n\n  /**\n   * Retrieves the Layer's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Layer.loss in PyKeras.\n    //   In PyKeras, Layer.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return this.losses.map(lossFn => lossFn());\n  }\n\n  get updates(): Tensor[] {\n    return this._updates;\n  }\n\n  get built(): boolean {\n    return this._built;\n  }\n\n  set built(built: boolean) {\n    this._built = built;\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this._trainableWeights.forEach(w => w.trainable = trainable);\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (this.trainable_) {\n      return this._trainableWeights.filter(w => w.trainable);\n    } else {\n      return [];\n    }\n  }\n\n  set trainableWeights(weights: LayerVariable[]) {\n    this._trainableWeights = weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    if (this.trainable) {\n      return this._trainableWeights.filter(w => !w.trainable)\n          .concat(this._nonTrainableWeights);\n    } else {\n      return this._trainableWeights.concat(this._nonTrainableWeights);\n    }\n  }\n\n  set nonTrainableWeights(weights: LayerVariable[]) {\n    this._nonTrainableWeights = weights;\n  }\n\n  /**\n   * The concatenation of the lists trainableWeights and nonTrainableWeights\n   * (in this order).\n   */\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  get stateful(): boolean {\n    return this._stateful;\n  }\n\n  /**\n   * Reset the states of the layer.\n   *\n   * This method of the base Layer class is essentially a no-op.\n   * Subclasses that are stateful (e.g., stateful RNNs) should override this\n   * method.\n   */\n  resetStates(): void {\n    if (!this.stateful) {\n      throw new Error(\n          'Cannot call the resetStates() method of a non-stateful Layer ' +\n          'object.');\n    }\n  }\n\n  /**\n   * Checks compatibility between the layer and provided inputs.\n   *\n   * This checks that the tensor(s) `input`\n   * verify the input assumptions of the layer\n   * (if any). If not, exceptions are raised.\n   *\n   * @param inputs Input tensor or list of input tensors.\n   *\n   * @exception ValueError in case of mismatch between\n   *   the provided inputs and the expectations of the layer.\n   */\n  protected assertInputCompatibility(inputs: Tensor|Tensor[]|SymbolicTensor|\n                                     SymbolicTensor[]): void {\n    inputs = generic_utils.toList(inputs);\n    if (this.inputSpec == null || this.inputSpec.length === 0) {\n      return;\n    }\n    const inputSpec = generic_utils.toList(this.inputSpec);\n    if (inputs.length !== inputSpec.length) {\n      throw new ValueError(\n          `Layer ${this.name} expects ${inputSpec.length} inputs, ` +\n          `but it received ${inputs.length} input tensors. ` +\n          `Input received: ${inputs}`);\n    }\n    for (let inputIndex = 0; inputIndex < inputs.length; inputIndex++) {\n      const x = inputs[inputIndex];\n      const spec: InputSpec = inputSpec[inputIndex];\n      if (spec == null) {\n        continue;\n      }\n\n      // Check ndim.\n      const ndim = x.rank;\n      if (spec.ndim != null) {\n        if (ndim !== spec.ndim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}: ` +\n              `expected ndim=${spec.ndim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.maxNDim != null) {\n        if (ndim > spec.maxNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected max_ndim=${spec.maxNDim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.minNDim != null) {\n        if (ndim < spec.minNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected min_ndim=${spec.minNDim}, found ndim=${ndim}.`);\n        }\n      }\n\n      // Check dtype.\n      if (spec.dtype != null) {\n        if (x.dtype !== spec.dtype) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name} ` +\n              `: expected dtype=${spec.dtype}, found dtype=${x.dtype}.`);\n        }\n      }\n\n      // Check specific shape axes.\n      if (spec.axes) {\n        const xShape = x.shape;\n        for (const key in spec.axes) {\n          const axis = Number(key);\n          const value = spec.axes[key];\n          // Perform Python-style slicing in case axis < 0;\n          // TODO(cais): Use https://github.com/alvivi/typescript-underscore to\n          // ensure type safety through Underscore calls.\n          const xShapeAtAxis =\n              axis >= 0 ? xShape[axis] : xShape[xShape.length + axis];\n          if (value != null && [value, null].indexOf(xShapeAtAxis) === -1) {\n            throw new ValueError(\n                `Input ${inputIndex} is incompatible with layer ` +\n                `${this.name}: expected axis ${axis} of input shape to ` +\n                `have value ${value} but got shape ${xShape}.`);\n          }\n        }\n      }\n\n      // Check shape.\n      if (spec.shape != null) {\n        for (let i = 0; i < spec.shape.length; ++i) {\n          const specDim = spec.shape[i];\n          const dim = x.shape[i];\n          if (specDim != null && dim != null) {\n            if (specDim !== dim) {\n              throw new ValueError(\n                  `Input ${inputIndex} is incompatible with layer ` +\n                  `${this.name}: expected shape=${spec.shape}, ` +\n                  `found shape=${x.shape}.`);\n            }\n          }\n        }\n      }\n    }\n  }\n\n  /**\n   * This is where the layer's logic lives.\n   *\n   * @param inputs Input tensor, or list/tuple of input tensors.\n   * @param kwargs Additional keyword arguments.\n   *\n   * @return A tensor or list/tuple of tensors.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return inputs;\n  }\n\n  protected invokeCallHook(inputs: Tensor|Tensor[], kwargs: Kwargs) {\n    if (this._callHook != null) {\n      this._callHook(inputs, kwargs);\n    }\n  }\n\n  /**\n   * Set call hook.\n   * This is currently used for testing only.\n   * @param callHook\n   */\n  setCallHook(callHook: CallHook) {\n    this._callHook = callHook;\n  }\n\n  /**\n   * Clear call hook.\n   * This is currently used for testing only.\n   */\n  clearCallHook() {\n    this._callHook = null;\n  }\n\n  /**\n   * Builds or executes a `Layer's logic.\n   *\n   * When called with `tf.Tensor`(s), execute the `Layer`s computation and\n   * return Tensor(s). For example:\n   *\n   * ```js\n   * const denseLayer = tf.layers.dense({\n   *   units: 1,\n   *   kernelInitializer: 'zeros',\n   *   useBias: false\n   * });\n   *\n   * // Invoke the layer's apply() method with a `tf.Tensor` (with concrete\n   * // numeric values).\n   * const input = tf.ones([2, 2]);\n   * const output = denseLayer.apply(input);\n   *\n   * // The output's value is expected to be [[0], [0]], due to the fact that\n   * // the dense layer has a kernel initialized to all-zeros and does not have\n   * // a bias.\n   * output.print();\n   * ```\n   *\n   * When called with `tf.SymbolicTensor`(s), this will prepare the layer for\n   * future execution.  This entails internal book-keeping on shapes of\n   * expected Tensors, wiring layers together, and initializing weights.\n   *\n   * Calling `apply` with `tf.SymbolicTensor`s are typically used during the\n   * building of non-`tf.Sequential` models. For example:\n   *\n   * ```js\n   * const flattenLayer = tf.layers.flatten();\n   * const denseLayer = tf.layers.dense({units: 1});\n   *\n   * // Use tf.layers.input() to obtain a SymbolicTensor as input to apply().\n   * const input = tf.input({shape: [2, 2]});\n   * const output1 = flattenLayer.apply(input);\n   *\n   * // output1.shape is [null, 4]. The first dimension is the undetermined\n   * // batch size. The second dimension comes from flattening the [2, 2]\n   * // shape.\n   * console.log(JSON.stringify(output1.shape));\n   *\n   * // The output SymbolicTensor of the flatten layer can be used to call\n   * // the apply() of the dense layer:\n   * const output2 = denseLayer.apply(output1);\n   *\n   * // output2.shape is [null, 1]. The first dimension is the undetermined\n   * // batch size. The second dimension matches the number of units of the\n   * // dense layer.\n   * console.log(JSON.stringify(output2.shape));\n   *\n   * // The input and output and be used to construct a model that consists\n   * // of the flatten and dense layers.\n   * const model = tf.model({inputs: input, outputs: output2});\n   * ```\n   *\n   * @param inputs a `tf.Tensor` or `tf.SymbolicTensor` or an Array of them.\n   * @param kwargs Additional keyword arguments to be passed to `call()`.\n   *\n   * @return Output of the layer's `call` method.\n   *\n   * @exception ValueError error in case the layer is missing shape information\n   *   for its `build` call.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  // Porting Note: This is a replacement for __call__() in Python.\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    kwargs = kwargs || {};\n\n    this.assertNotDisposed();\n\n    // Ensure inputs are all the same type.\n    const inputsList = generic_utils.toList(inputs);\n\n    let allAreSymbolic = true;\n    for (const input of inputsList) {\n      if (!(input instanceof SymbolicTensor)) {\n        allAreSymbolic = false;\n        break;\n      }\n    }\n    let noneAreSymbolic = true;\n    for (const input of inputsList) {\n      if (input instanceof SymbolicTensor) {\n        noneAreSymbolic = false;\n        break;\n      }\n    }\n\n    if (allAreSymbolic === noneAreSymbolic) {\n      throw new ValueError(\n          'Arguments to apply() must be all ' +\n          'SymbolicTensors or all Tensors');\n    }\n\n    // TODO(michaelterry): nameScope() may not be necessary.\n    return nameScope(this.name, () => {\n      // Handle laying building (weight creating, input spec locking).\n      if (!this.built) {\n        /*\n          Throw exceptions in case the input is not compatible\n          with the inputSpec specified in the layer constructor.\n         */\n        this.assertInputCompatibility(inputs);\n\n        // Collect input shapes to build layer.\n        const inputShapes: Shape[] = [];\n        for (const xElem of generic_utils.toList(inputs)) {\n          inputShapes.push(xElem.shape);\n        }\n        this.build(generic_utils.singletonOrArray(inputShapes));\n        this.built = true;\n\n        // Load weights that were specified at layer instantiation.\n        if (this.initialWeights) {\n          this.setWeights(this.initialWeights);\n        }\n\n        if (this._refCount === null && noneAreSymbolic) {\n          // The first use of this layer is a non-symbolic call, set ref count\n          // to 1 so the Layer can be properly disposed if its dispose() method\n          // is called.\n          this._refCount = 1;\n        }\n      }\n\n      /*\n        Throw exceptions in case the input is not compatible\n        with the inputSpec set at build time.\n      */\n      this.assertInputCompatibility(inputs);\n\n      // Handle mask propagation.\n      // TODO(michaelterry): Mask propagation not currently implemented.\n\n      // Actually call the layer, collecting output(s), mask(s), and shape(s).\n      if (noneAreSymbolic) {\n        let output = this.call(inputs as Tensor | Tensor[], kwargs);\n        // TODO(michaelterry): Compute the outputMask\n\n        // If the layer returns tensors from its inputs, unmodified,\n        // we copy them to avoid loss of tensor metadata.\n        const outputList: Tensor[] = generic_utils.toList(output);\n        const outputListCopy: Tensor[] = [];\n        // TODO(michaelterry): This copying may not be necessary given our eager\n        // backend.\n        for (let x of outputList) {\n          if (inputsList.indexOf(x) !== -1) {\n            x = x.clone();\n          }\n          outputListCopy.push(x);\n        }\n        output = generic_utils.singletonOrArray(outputListCopy);\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        // TODO(michaelterry): Call addInboundNode()?\n        return output;\n      } else {\n        const inputShape = collectInputShape(inputs);\n        const outputShape = this.computeOutputShape(inputShape);\n        let output: SymbolicTensor|SymbolicTensor[];\n        const outputDType = guessOutputDType(inputs);\n        this.warnOnIncompatibleInputShape(\n            Array.isArray(inputs) ? inputShape[0] as Shape :\n                                    inputShape as Shape);\n\n        if (outputShape != null && outputShape.length > 0 &&\n            Array.isArray(outputShape[0])) {\n          // We have multiple output shapes. Create multiple output tensors.\n          output = (outputShape as Shape[])\n                       .map(\n                           (shape, index) => new SymbolicTensor(\n                               outputDType, shape, this,\n                               generic_utils.toList(inputs), kwargs, this.name,\n                               index));\n        } else {\n          output = new SymbolicTensor(\n              outputDType, outputShape as Shape, this,\n              generic_utils.toList(inputs), kwargs, this.name);\n        }\n\n        /*\n          Add an inbound node to the layer, so that it keeps track\n          of the call and of all new variables created during the call.\n          This also updates the layer history of the output tensor(s).\n          If the input tensor(s) had no previous history,\n          this does nothing.\n        */\n        this.addInboundNode(\n            inputs as SymbolicTensor | SymbolicTensor[], output, null, null,\n            inputShape, outputShape, kwargs);\n        this._refCount++;\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        return output;\n      }\n    });\n  }\n\n  /**\n   * Check compatibility between input shape and this layer's batchInputShape.\n   *\n   * Print warning if any incompatibility is found.\n   *\n   * @param inputShape Input shape to be checked.\n   */\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {\n    if (this.batchInputShape == null) {\n      return;\n    } else if (inputShape.length !== this.batchInputShape.length) {\n      console.warn(\n          `The rank of the input tensor provided (shape: ` +\n          `${JSON.stringify(inputShape)}) does not match that of the ` +\n          `batchInputShape (${JSON.stringify(this.batchInputShape)}) ` +\n          `of the layer ${this.name}`);\n    } else {\n      let dimMismatch = false;\n      this.batchInputShape.forEach((dimension, i) => {\n        if (dimension != null && inputShape[i] != null &&\n            inputShape[i] !== dimension) {\n          dimMismatch = true;\n        }\n      });\n      if (dimMismatch) {\n        console.warn(\n            `The shape of the input tensor ` +\n            `(${JSON.stringify(inputShape)}) does not ` +\n            `match the expectation of layer ${this.name}: ` +\n            `${JSON.stringify(this.batchInputShape)}`);\n      }\n    }\n  }\n\n  /**\n   * Retrieves the output shape(s) of a layer.\n   *\n   * Only applicable if the layer has only one inbound node, or if all inbound\n   * nodes have the same output shape.\n   *\n   * @returns Output shape or shapes.\n   * @throws AttributeError: if the layer is connected to more than one incoming\n   *   nodes.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  get outputShape(): Shape|Shape[] {\n    if (this.inboundNodes == null || this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `The layer ${this.name} has never been called and thus has no ` +\n          `defined output shape.`);\n    }\n    const allOutputShapes: string[] = [];\n    for (const node of this.inboundNodes) {\n      const shapeString = JSON.stringify(node.outputShapes);\n      if (allOutputShapes.indexOf(shapeString) === -1) {\n        allOutputShapes.push(shapeString);\n      }\n    }\n    if (allOutputShapes.length === 1) {\n      const outputShapes = this.inboundNodes[0].outputShapes;\n      if (Array.isArray(outputShapes) && Array.isArray(outputShapes[0]) &&\n          outputShapes.length === 1) {\n        return (outputShapes as Shape[])[0];\n      } else {\n        return outputShapes;\n      }\n\n    } else {\n      throw new AttributeError(\n          `The layer ${this.name} has multiple inbound nodes with different ` +\n          `output shapes. Hence the notion of \"output shape\" is ill-defined ` +\n          `for the layer.`);\n      // TODO(cais): Implement getOutputShapeAt().\n    }\n  }\n\n  /**\n   * Counts the total number of numbers (e.g., float32, int32) in the\n   * weights.\n   *\n   * @returns An integer count.\n   * @throws RuntimeError: If the layer is not built yet (in which case its\n   *   weights are not defined yet.)\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  countParams(): number {\n    if (!this.built) {\n      throw new RuntimeError(\n          `You tried to call countParams() on ${this.name}, ` +\n          `but the layer is not built yet. Build it first by calling ` +\n          `build(batchInputShape).`);\n    }\n    return variable_utils.countParamsInWeights(this.weights);\n  }\n\n  /**\n   * Creates the layer weights.\n   *\n   * Must be implemented on all layers that have weights.\n   *\n   * Called when apply() is called to construct the weights.\n   *\n   * @param inputShape A `Shape` or array of `Shape` (unused).\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  build(inputShape: Shape|Shape[]) {\n    this.built = true;\n  }\n\n  /**\n   * Returns the current values of the weights of the layer.\n   *\n   * @param trainableOnly Whether to get the values of only trainable weights.\n   * @returns Weight values as an `Array` of `tf.Tensor`s.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getWeights(trainableOnly = false): Tensor[] {\n    return batchGetValue(trainableOnly ? this.trainableWeights : this.weights);\n  }\n\n  /**\n   * Sets the weights of the layer, from Tensors.\n   *\n   * @param weights a list of Tensors. The number of arrays and their shape\n   *   must match number of the dimensions of the weights of the layer (i.e.\n   *   it should match the output of `getWeights`).\n   *\n   * @exception ValueError If the provided weights list does not match the\n   *   layer's specifications.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  setWeights(weights: Tensor[]): void {\n    tidy(() => {\n      const params = this.weights;\n      if (params.length !== weights.length) {\n        // TODO(cais): Restore the following and use `providedWeights`, instead\n        // of `weights` in the error message, once the deeplearn.js bug is\n        // fixed: https://github.com/PAIR-code/deeplearnjs/issues/498 const\n        // providedWeights = JSON.stringify(weights).substr(0, 50);\n        throw new ValueError(\n            `You called setWeights(weights) on layer \"${this.name}\" ` +\n            `with a weight list of length ${weights.length}, ` +\n            `but the layer was expecting ${params.length} weights. ` +\n            `Provided weights: ${weights}...`);\n      }\n      if (params.length === 0) {\n        return;\n      }\n      const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n      const paramValues = batchGetValue(params);\n      for (let i = 0; i < paramValues.length; ++i) {\n        const pv = paramValues[i];\n        const p = params[i];\n        const w = weights[i];\n        if (!util.arraysEqual(pv.shape, w.shape)) {\n          throw new ValueError(\n              `Layer weight shape ${pv.shape} ` +\n              `not compatible with provided weight shape ${w.shape}`);\n        }\n        weightValueTuples.push([p, w]);\n      }\n      batchSetValue(weightValueTuples);\n    });\n  }\n\n  /**\n   * Adds a weight variable to the layer.\n   *\n   * @param name Name of the new weight variable.\n   * @param shape The shape of the weight.\n   * @param dtype The dtype of the weight.\n   * @param initializer An initializer instance.\n   * @param regularizer A regularizer instance.\n   * @param trainable Whether the weight should be trained via backprop or not\n   *   (assuming that the layer itself is also trainable).\n   * @param constraint An optional trainable.\n   * @return The created weight variable.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  protected addWeight(\n      name: string, shape: Shape, dtype?: DataType, initializer?: Initializer,\n      regularizer?: Regularizer, trainable?: boolean,\n      constraint?: Constraint): LayerVariable {\n    // Reject duplicate weight names.\n    if (this._addedWeightNames.indexOf(name) !== -1) {\n      throw new ValueError(\n          `Duplicate weight name ${name} for layer ${this.name}`);\n    }\n    this._addedWeightNames.push(name);\n\n    if (dtype == null) {\n      dtype = 'float32';\n    }\n\n    if (this.fastWeightInitDuringBuild) {\n      initializer = getInitializer('zeros');\n    }\n    const initValue = initializer.apply(shape, dtype);\n    const weight =\n        new LayerVariable(initValue, dtype, name, trainable, constraint);\n    initValue.dispose();\n    // Request backend not to dispose the weights of the model on scope() exit.\n    if (regularizer != null) {\n      this.addLoss(() => regularizer.apply(weight.read()));\n    }\n    if (trainable == null) {\n      trainable = true;\n    }\n    if (trainable) {\n      this._trainableWeights.push(weight);\n    } else {\n      this._nonTrainableWeights.push(weight);\n    }\n    return weight;\n  }\n\n  /**\n   * Set the fast-weight-initialization flag.\n   *\n   * In cases where the initialized weight values will be immediately\n   * overwritten by loaded weight values during model loading, setting\n   * the flag to `true` saves unnecessary calls to potentially expensive\n   * initializers and speeds up the loading process.\n   *\n   * @param value Target value of the flag.\n   */\n  setFastWeightInitDuringBuild(value: boolean) {\n    this.fastWeightInitDuringBuild = value;\n  }\n\n  /**\n   * Add losses to the layer.\n   *\n   * The loss may potentionally be conditional on some inputs tensors,\n   * for instance activity losses are conditional on the layer's inputs.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  addLoss(losses: RegularizerFn|RegularizerFn[]): void {\n    if (losses == null || Array.isArray(losses) && losses.length === 0) {\n      return;\n    }\n    // Update this.losses\n    losses = generic_utils.toList(losses);\n    if (this._losses !== undefined && this._losses !== null) {\n      this.losses.push(...losses);\n    }\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (!this.supportsMasking) {\n      if (mask != null) {\n        if (Array.isArray(mask)) {\n          mask.forEach(maskElement => {\n            if (maskElement != null) {\n              throw new TypeError(\n                  `Layer ${this.name} does not support masking, ` +\n                  'but was passed an inputMask.');\n            }\n          });\n        } else {\n          throw new TypeError(\n              `Layer ${this.name} does not support masking, ` +\n              'but was passed an inputMask.');\n        }\n      }\n      // masking not explicitly supported: return null as mask\n      return null;\n    }\n    // if masking is explictly supported, by default\n    // carry over the input mask\n    return mask;\n  }\n\n  /**\n   * Internal method to create an inbound node for the layer.\n   *\n   * @param inputTensors List of input tensors.\n   * @param outputTensors List of output tensors.\n   * @param inputMasks List of input masks (a mask can be a tensor, or null).\n   * @param outputMasks List of output masks (a mask can be a tensor, or null).\n   * @param inputShapes List of input shape tuples.\n   * @param outputShapes List of output shape tuples.\n   * @param kwargs Dictionary of keyword arguments that were passed to the\n   *   `call` method of the layer at the call that created the node.\n   */\n  private addInboundNode(\n      inputTensors: SymbolicTensor|SymbolicTensor[],\n      outputTensors: SymbolicTensor|SymbolicTensor[],\n      inputMasks: Tensor|Tensor[], outputMasks: Tensor|Tensor[],\n      inputShapes: Shape|Shape[], outputShapes: Shape|Shape[],\n      kwargs: {} = null): void {\n    const inputTensorList: SymbolicTensor[] =\n        generic_utils.toList(inputTensors);\n    outputTensors = generic_utils.toList(outputTensors);\n    inputMasks = generic_utils.toList(inputMasks);\n    outputMasks = generic_utils.toList(outputMasks);\n    inputShapes = types_utils.normalizeShapeList(inputShapes);\n    outputShapes = types_utils.normalizeShapeList(outputShapes);\n\n    // Collect input tensor(s) coordinates.\n    const inboundLayers: Layer[] = [];\n    const nodeIndices: number[] = [];\n    const tensorIndices: number[] = [];\n    for (const x of inputTensorList) {\n      /*\n       * TODO(michaelterry): Keras adds this value to tensors; it's not\n       * clear whether we'll use this or not.\n       */\n      inboundLayers.push(x.sourceLayer);\n      nodeIndices.push(x.nodeIndex);\n      tensorIndices.push(x.tensorIndex);\n    }\n\n    // Create node, add it to inbound nodes.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node(\n        {\n          outboundLayer: this,\n          inboundLayers,\n          nodeIndices,\n          tensorIndices,\n          inputTensors: inputTensorList,\n          outputTensors,\n          inputMasks,\n          outputMasks,\n          inputShapes,\n          outputShapes\n        },\n        kwargs);\n\n    // Update tensor history\n    for (let i = 0; i < outputTensors.length; i++) {\n      // TODO(michaelterry: _uses_learning_phase not tracked.\n      outputTensors[i].sourceLayer = this;\n      outputTensors[i].nodeIndex = this.inboundNodes.length - 1;\n      outputTensors[i].tensorIndex = i;\n    }\n  }\n\n  /**\n   * Returns the config of the layer.\n   *\n   * A layer config is a TS dictionary (serializable)\n   * containing the configuration of a layer.\n   * The same layer can be reinstantiated later\n   * (without its trained weights) from this configuration.\n   *\n   * The config of a layer does not include connectivity\n   * information, nor the layer class name.  These are handled\n   * by 'Container' (one layer of abstraction above).\n   *\n   * Porting Note: The TS dictionary follows TS naming standrds for\n   * keys, and uses tfjs-layers type-safe Enums.  Serialization methods\n   * should use a helper function to convert to the pythonic storage\n   * standard. (see serialization_utils.convertTsToPythonic)\n   *\n   * @returns TS dictionary of configuration.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getConfig(): serialization.ConfigDict {\n    const config:\n        serialization.ConfigDict = {name: this.name, trainable: this.trainable};\n    if (this.batchInputShape != null) {\n      config['batchInputShape'] = this.batchInputShape;\n    }\n    if (this.dtype != null) {\n      config['dtype'] = this.dtype;\n    }\n    return config;\n  }\n\n  /**\n   * Dispose the weight variables that this Layer instance holds.\n   *\n   * @returns {number} Number of disposed variables.\n   */\n  protected disposeWeights(): number {\n    this.weights.forEach(weight => weight.dispose());\n    return this.weights.length;\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Layer '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose layer's weights.\n   *\n   * This method decrease the reference count of the Layer object by 1.\n   *\n   * A Layer is reference-counted. Its reference count is incremented by 1\n   * the first item its `apply()` method is called and when it becomes a part\n   * of a new `Node` (through calling the `apply()`) method on a\n   * `tf.SymbolicTensor`).\n   *\n   * If the reference count of a Layer becomes 0, all the weights will be\n   * disposed and the underlying memory (e.g., the textures allocated in WebGL)\n   * will be freed.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * weights of the Layer will *not* be disposed.\n   *\n   * After a Layer is disposed, it cannot be used in calls such as `apply()`,\n   * `getWeights()` or `setWeights()` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the Container after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the layer has already\n   *   been disposed.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  dispose(): DisposeResult {\n    if (!this.built) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been ` +\n          `built yet.`);\n    }\n\n    if (this._refCount === null) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been used ` +\n          `yet.`);\n    }\n\n    this.assertNotDisposed();\n\n    let numDisposedVariables = 0;\n    if (--this._refCount === 0) {\n      numDisposedVariables = this.disposeWeights();\n    }\n\n    return {refCountAfterDispose: this._refCount, numDisposedVariables};\n  }\n}\n\n/**\n * Collects the input shape(s) of a list of `tf.Tensor`s or\n * `tf.SymbolicTensor`s.\n *\n * TODO(michaelterry): Update PyKeras docs (backport).\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return List of shape tuples (or single tuple), one tuple per input.\n */\nfunction collectInputShape(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                           Tensor[]): Shape|Shape[] {\n  inputTensors =\n      generic_utils.toList(inputTensors) as SymbolicTensor[] | Tensor[];\n  const shapes: Shape[] = [];\n  for (const x of inputTensors) {\n    shapes.push(x.shape);\n  }\n  return generic_utils.singletonOrArray(shapes);\n}\n\n/**\n * Guesses output dtype based on inputs.\n *\n * At present, just returns 'float32' for any input.\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return The guessed DType. At present, always returns 'float32'.\n */\nfunction guessOutputDType(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                          Tensor[]): DataType {\n  return 'float32';\n}\n\n/**\n * Returns the list of input tensors necessary to compute `tensor`.\n *\n * Output will always be a list of tensors (potentially with 1 element).\n *\n * @param tensor The tensor to start from.\n * @param layer Origin layer of the tensor.\n * @param nodeIndex Origin node index of the tensor.\n *\n * @return Array of input tensors.\n */\nexport function getSourceInputs(\n    tensor: SymbolicTensor, layer?: Layer,\n    nodeIndex?: number): SymbolicTensor[] {\n  if (layer == null || (nodeIndex != null && nodeIndex > 0)) {\n    layer = tensor.sourceLayer;\n    nodeIndex = tensor.nodeIndex;\n  }\n  if (layer.inboundNodes.length === 0) {\n    return [tensor];\n  } else {\n    const node = layer.inboundNodes[nodeIndex];\n    if (node.inboundLayers.length === 0) {\n      return node.inputTensors;\n    } else {\n      const sourceTensors: SymbolicTensor[] = [];\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const x = node.inputTensors[i];\n        const layer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const previousSources = getSourceInputs(x, layer, nodeIndex);\n        // Avoid input redundancy.\n        for (const x of previousSources) {\n          if (sourceTensors.indexOf(x) === -1) {\n            sourceTensors.push(x);\n          }\n        }\n      }\n      return sourceTensors;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\n\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/**\n * Constructor arguments for InputLayer.\n *\n * Note: You should provide only inputShape or batchInputShape (not both).\n * If only inputShape is provided, then the batchInputShape is determined by\n * the batchSize argument and the inputShape: [batchSize].concat(inputShape).\n */\nexport declare interface InputLayerArgs {\n  /** Input shape, not including the batch axis. */\n  inputShape?: Shape;\n  /** Optional input batch size (integer or null). */\n  batchSize?: number;\n  /** Batch input shape, including the batch axis. */\n  batchInputShape?: Shape;\n  /** Datatype of the input.  */\n  dtype?: DataType;\n  /**\n   * Whether the placeholder created is meant to be sparse.\n   */\n  sparse?: boolean;  // TODO(michaelterry): Not clear whether we'll need this.\n\n  /** Name of the layer. */\n  name?: string;\n}\n\nexport class InputLayer extends Layer {\n  /** @nocollapse */\n  static readonly className = 'InputLayer';\n  sparse: boolean;\n  constructor(args: InputLayerArgs) {\n    super({\n      dtype: args.dtype,\n      name: args.name != null ? args.name : getUid('input').toString()\n    });\n    // Normalize config.batchSize and config.sparse\n    if (args.batchSize == null) {\n      args.batchSize = null;\n    }\n    if (args.sparse == null) {\n      args.sparse = false;\n    }\n\n    this.trainable = false;\n    this.built = true;\n    this.sparse = args.sparse;\n\n    if (args.inputShape != null && args.batchInputShape != null) {\n      throw new ValueError(\n          'Only provide the inputShape OR ' +\n          'batchInputShape argument to inputLayer, not both at the same time.');\n    }\n    let batchInputShape = args.batchInputShape;\n    if (batchInputShape == null) {\n      if (args.inputShape == null) {\n        throw new ValueError(\n            'An InputLayer should be passed either a ' +\n            '`batchInputShape` or an `inputShape`.');\n      } else {\n        batchInputShape = [args.batchSize].concat(args.inputShape);\n      }\n    } else {\n      // TODO(michaelterry): Backport to PyKeras\n      if (args.batchSize != null) {\n        throw new ValueError(\n            'Cannot specify batchSize if batchInputShape is ' +\n            'specified when creating an InputLayer.');\n      }\n    }\n\n    const dtype = args.dtype || 'float32';\n\n    this.batchInputShape = batchInputShape;\n    this.dtype = dtype;\n    // TODO(michaelterry): Backport this to PyKeras?\n    this.inputSpec = [{shape: batchInputShape}];\n\n    const inputTensor = new SymbolicTensor(\n        this.dtype, this.batchInputShape, this, [], {}, this.name);\n    inputTensor.nodeIndex = 0;\n    inputTensor.tensorIndex = 0;\n\n    // Create an input node to add to this.outboundNode.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: [inputTensor],\n      outputTensors: [inputTensor],\n      inputMasks: [null],\n      outputMasks: [null],\n      inputShapes: [batchInputShape],\n      outputShapes: [batchInputShape]\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor {\n    throw new ValueError(\n        'Cannot pass any input to an ' +\n        `InputLayer's apply() method. InputLayer name: ${this.name}`);\n  }\n\n  dispose(): DisposeResult {\n    // dispose() for InputLayer is overridden as no-op.\n    return {refCountAfterDispose: this._refCount, numDisposedVariables: 0};\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      batchInputShape: this.batchInputShape,\n      dtype: this.dtype,\n      sparse: this.sparse,\n      name: this.name\n    };\n  }\n}\nserialization.registerClass(InputLayer);\n\n/**\n * Config for the Input function.\n *\n * Note: You should provide only shape or batchShape (not both).\n * If only shape is provided, then the batchShape becomes\n * [null].concat(inputShape).\n */\nexport interface InputConfig {\n  /**\n   * A shape, not including the batch size. For instance, `shape=[32]`\n   * indicates that the expected input will be batches of 32-dimensional\n   * vectors.\n   */\n  shape?: Shape;\n  /**\n   * A shape tuple (integer), including the batch size. For instance,\n   * `batchShape=[10, 32]` indicates that the expected input will be batches of\n   * 10 32-dimensional vectors. `batchShape=[null, 32]` indicates batches of an\n   * arbitrary number of 32-dimensional vectors.\n   */\n  batchShape?: Shape;\n  /**\n   * An optional name string for the layer. Should be unique in a model (do not\n   * reuse the same name twice). It will be autogenerated if it isn't provided.\n   */\n  name?: string;\n  dtype?: DataType;\n  /**\n   * A boolean specifying whether the placeholder to be created is sparse.\n   */\n  sparse?: boolean;\n}\n\nexport function Input(config: InputConfig): SymbolicTensor {\n  if (config.batchShape == null && config.shape == null) {\n    throw new Error(\n        'Please provide to Input either a `shape`' +\n        ' or a `batchShape` argument. Note that ' +\n        '`shape` does not include the batch ' +\n        'dimension.');\n  }\n  if (config.batchShape != null && config.shape != null) {\n    // TODO(michaelterry): Backport to PyKeras.\n    throw new ValueError(\n        'Please provide either a `shape` or `batchShape` ' +\n        'argument to Input, but not both.');\n  }\n  let batchShape = config.batchShape;\n  if (config.shape != null && batchShape == null) {\n    batchShape = [null].concat(config.shape);\n  }\n\n  let dtype = config.dtype;\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n\n  const inputLayer = new InputLayer({\n    batchInputShape: batchShape,\n    name: config.name,\n    dtype,\n    sparse: config.sparse\n  });\n\n  const outputs = inputLayer.inboundNodes[0].outputTensors;\n  return outputs[0];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, Scalar} from '@tensorflow/tfjs-core';\n\n/**\n * Logs in which values can be either numbers or Tensors (Scalars).\n *\n * Used internally.\n */\nexport type UnresolvedLogs = {\n  [key: string]: number|Scalar;\n};\n\n/**\n * Turn any Scalar values in a Logs object into actual number values.\n *\n * @param logs The `Logs` object to be resolved in place.\n */\nexport async function resolveScalarsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n  const keys: string[] = [];\n  const scalarsToDispose: Scalar[] = [];\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      const valueScalar = value;\n      promises.push(valueScalar.data());\n      keys.push(key);\n      scalarsToDispose.push(valueScalar);\n    }\n  }\n  if (promises.length > 0) {\n    const values = await Promise.all(promises);\n    for (let i = 0; i < values.length; ++i) {\n      logs[keys[i]] = values[i][0];\n    }\n    // Dispose the original scalar tensors.\n    dispose(scalarsToDispose);\n  }\n}\n\n/**\n * Dispose all Tensors in an UnresolvedLogs object.\n *\n * @param logs An `UnresolvedLogs` object potentially containing `tf.Tensor`s in\n *   places where the values can be `tf.Tensor` or `number`.\n */\nexport function disposeTensorsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      value.dispose();\n    }\n  }\n}\n\n/**\n * Logs in which values can only be numbers.\n *\n * Used when calling client-provided custom callbacks.\n */\nexport type Logs = {\n  [key: string]: number;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {add, div, keep, mul, nextFrame, Scalar, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Container} from './engine/container';\nimport {ValueError} from './errors';\nimport {Logs, resolveScalarsInLogs, UnresolvedLogs} from './logs';\nimport * as generic_utils from './utils/generic_utils';\n\n/** Verbosity logging level when fitting a model. */\nexport enum ModelLoggingVerbosity {\n  SILENT = 0,\n  VERBOSE = 1\n}\n\n/** How often to yield to the main thread when training (in ms). */\nexport const DEFAULT_YIELD_EVERY_MS = 125;\n\nexport type Params = {\n  [key: string]: number|string|boolean|number[]|string[]|boolean[];\n};\n\nexport type YieldEveryOptions = 'auto'|'batch'|'epoch'|'never'|number;\n\n/**\n * Abstract base class used to build new callbacks.\n *\n * The `logs` dictionary that callback methods take as argument will contain\n * keys for quantities relevant to the current batch or epoch.\n *\n * Currently, the `.fit()` method of the `Sequential` model class\n * will include the following quantities in the `logs` that\n * it passes to its callbacks:\n *\n * onEpochEnd: Logs include `acc` and `loss`, and optionally include `valLoss`\n *   (if validation is enabled in `fit`), and `valAcc` (if validation and\n *   accuracy monitoring are enabled).\n * onBatchBegin: Logs include `size`, the number of samples in the current\n *   batch.\n * onBatchEnd: Logs include `loss`, and optionally `acc` (if accuracy monitoring\n *   is enabled).\n */\nexport abstract class BaseCallback {\n  // TODO(michaelterry): This type is a best guess.\n  validationData: Tensor|Tensor[] = null;\n  /**\n   * Training parameters (eg. verbosity, batch size, number of epochs...).\n   */\n  params: Params;\n\n  setParams(params: Params): void {\n    this.params = params;\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {}\n\n  async onTrainBegin(logs?: UnresolvedLogs) {}\n\n  async onTrainEnd(logs?: UnresolvedLogs) {}\n\n  // LayersModel needs to call Callback.setModel(), but cannot actually depend\n  // on Callback because that creates a cyclic dependency.  Providing this no-op\n  // method on BaseCallback breaks the cycle: this way LayersModel can depend on\n  // BaseCallback but not on Callback.  The argument is typed as `Container`\n  // (the superclass of LayersModel) to avoid recapitulating the cycle. Callback\n  // overrides this method and enforces that the argument is really a\n  // LayersModel.\n  setModel(model: Container): void {\n    // Do nothing. Use Callback instead of BaseCallback to track the model.\n  }\n}\n\n/**\n * Container abstracting a list of callbacks.\n */\nexport class CallbackList {\n  callbacks: BaseCallback[];\n  queueLength: number;\n\n  // TODO(cais): When the need arises, uncomment the following lines and\n  // implement the queue for time values.\n  // private deltaTBatch: number;\n  // private deltaTsBatchBegin: Array<number>;\n  // private deltaTsBatchEnd: Array<number>;\n\n  /**\n   * Constructor of CallbackList.\n   * @param callbacks Array of `Callback` instances.\n   * @param queueLength Queue length for keeping running statistics over\n   *   callback execution time.\n   */\n  constructor(callbacks?: BaseCallback[], queueLength = 10) {\n    // TODO(cais): Make use of queueLength when implementing the queue for time\n    // values.\n    if (callbacks == null) {\n      callbacks = [];\n    }\n    this.callbacks = callbacks;\n    this.queueLength = queueLength;\n  }\n\n  append(callback: BaseCallback): void {\n    this.callbacks.push(callback);\n  }\n\n  setParams(params: Params): void {\n    for (const callback of this.callbacks) {\n      callback.setParams(params);\n    }\n  }\n\n  setModel(model: Container): void {\n    for (const callback of this.callbacks) {\n      callback.setModel(model);\n    }\n  }\n\n  /**\n   * Called at the start of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochBegin(epoch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochEnd(epoch, logs);\n    }\n  }\n\n  /**\n   * Called  right before processing a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchBegin(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchEnd(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the beginning of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainBegin(logs);\n    }\n  }\n\n  /**\n   * Called at the end of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainEnd(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainEnd(logs);\n    }\n  }\n}\n\n/**\n * Callback that accumulates epoch averages of metrics.\n *\n * This callback is automatically applied to every LayersModel.\n */\nexport class BaseLogger extends BaseCallback {\n  private seen: number;\n  private totals: UnresolvedLogs;\n\n  constructor() {\n    super();\n  }\n\n  async onEpochBegin(epoch: number) {\n    this.seen = 0;\n    this.totals = {};\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const batchSize = logs['size'] == null ? 0 : logs['size'] as number;\n    this.seen += batchSize;\n    for (const key in logs) {\n      const value = logs[key];\n      if (typeof value === 'number') {\n        if (!this.totals.hasOwnProperty(key)) {\n          this.totals[key] = 0;\n        }\n        this.totals[key] = this.totals[key] as number + value * batchSize;\n      } else {\n        let oldTotalsToDispose: Scalar;\n        if (key in this.totals) {\n          oldTotalsToDispose = this.totals[key] as Scalar;\n        } else {\n          this.totals[key] = 0;\n        }\n        const total: Scalar =\n            tidy(() => add((this.totals[key]), mul(value, batchSize)));\n        this.totals[key] = total;\n        if (oldTotalsToDispose != null) {\n          oldTotalsToDispose.dispose();\n        }\n      }\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs != null) {\n      for (const key of this.params['metrics'] as string[]) {\n        if (this.totals[key] == null) {\n          continue;\n        }\n        if (typeof this.totals[key] === 'number') {\n          logs[key] = this.totals[key] as number / this.seen;\n        } else {\n          tidy(() => {\n            const log: Scalar = mul(div(1, this.seen), this.totals[key]);\n            logs[key] = log;\n            (this.totals[key] as Tensor).dispose();\n            keep(logs[key] as Scalar);\n          });\n        }\n      }\n    }\n  }\n}\n\n/**\n * Callback that records events into a `History` object. This callback is\n * automatically applied to every TF.js Layers model. The `History` object\n * gets returned by the `fit` method of models.\n */\nexport class History extends BaseCallback {\n  epoch: number[];\n  history: {[key: string]: Array<number|Tensor>};\n\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    this.epoch = [];\n    this.history = {};\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    this.epoch.push(epoch);\n    for (const key in logs) {\n      if (this.history[key] == null) {\n        this.history[key] = [];\n      }\n      this.history[key].push(logs[key]);\n    }\n  }\n\n  /**\n   * Await the values of all losses and metrics.\n   */\n  async syncData() {\n    const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n    const keys: string[] = [];\n    const indices: number[] = [];\n    for (const key in this.history) {\n      const valueArray = this.history[key];\n      for (let i = 0; i < valueArray.length; ++i) {\n        if (typeof valueArray[i] !== 'number') {\n          const valueScalar = valueArray[i] as Tensor;\n          promises.push(valueScalar.data());\n          keys.push(key);\n          indices.push(i);\n        }\n      }\n    }\n    const values = await Promise.all(promises);\n    for (let n = 0; n < values.length; ++n) {\n      const tensorToDispose = this.history[keys[n]][indices[n]] as Tensor;\n      tensorToDispose.dispose();\n      this.history[keys[n]][indices[n]] = values[n][0];\n    }\n  }\n}\n\nexport interface CustomCallbackArgs {\n  onTrainBegin?: (logs?: Logs) => void | Promise<void>;\n  onTrainEnd?: (logs?: Logs) => void | Promise<void>;\n  onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n}\n\n/**\n * Custom callback for training.\n */\nexport class CustomCallback extends BaseCallback {\n  protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;\n  protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;\n  protected readonly epochBegin:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly epochEnd:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchBegin:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchEnd:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly yield:\n      (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n\n  private yieldEvery: YieldEveryOptions;\n  private currentEpoch = 0;\n\n  constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions) {\n    super();\n    this.yieldEvery = yieldEvery || 'auto';\n    if (this.yieldEvery === 'auto') {\n      this.yieldEvery = DEFAULT_YIELD_EVERY_MS;\n    }\n    if (this.yieldEvery === 'never' && args.onYield != null) {\n      throw new Error(\n          'yieldEvery is `never` but you provided an `onYield` callback. ' +\n          'Either change `yieldEvery` or remove the callback');\n    }\n    if (util.isNumber(this.yieldEvery)) {\n      // Decorate `maybeWait` so it will be called at most once every\n      // `yieldEvery` ms.\n      this.maybeWait = generic_utils.debounce(\n          this.maybeWait.bind(this), this.yieldEvery as number);\n    }\n    this.trainBegin = args.onTrainBegin;\n    this.trainEnd = args.onTrainEnd;\n    this.epochBegin = args.onEpochBegin;\n    this.epochEnd = args.onEpochEnd;\n    this.batchBegin = args.onBatchBegin;\n    this.batchEnd = args.onBatchEnd;\n    this.yield = args.onYield;\n  }\n\n  async maybeWait(epoch: number, batch: number, logs: UnresolvedLogs) {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.yield != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.yield(epoch, batch, logs as Logs));\n    }\n    ps.push(nextFrame());\n    await Promise.all(ps);\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    this.currentEpoch = epoch;\n    if (this.epochBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.epochBegin(epoch, logs as Logs);\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.epochEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.epochEnd(epoch, logs as Logs));\n    }\n    if (this.yieldEvery === 'epoch') {\n      ps.push(nextFrame());\n    }\n    await Promise.all(ps);\n  }\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    if (this.batchBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.batchBegin(batch, logs as Logs);\n    }\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.batchEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.batchEnd(batch, logs as Logs));\n    }\n    if (this.yieldEvery === 'batch') {\n      ps.push(nextFrame());\n    } else if (util.isNumber(this.yieldEvery)) {\n      ps.push(this.maybeWait(this.currentEpoch, batch, logs));\n    }\n    await Promise.all(ps);\n  }\n\n  async onTrainBegin(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainBegin(logs as Logs);\n    }\n  }\n\n  async onTrainEnd(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainEnd != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainEnd(logs as Logs);\n    }\n  }\n}\n\n/**\n * Standardize callbacks or configurations of them to an Array of callbacks.\n */\nexport function standardizeCallbacks(\n    callbacks: BaseCallback|BaseCallback[]|CustomCallbackArgs|\n    CustomCallbackArgs[],\n    yieldEvery: YieldEveryOptions): BaseCallback[] {\n  if (callbacks == null) {\n    callbacks = {} as BaseCallback;\n  }\n  if (callbacks instanceof BaseCallback) {\n    return [callbacks];\n  }\n  if (Array.isArray(callbacks) && callbacks[0] instanceof BaseCallback) {\n    return callbacks as BaseCallback[];\n  }\n  // Convert custom callback configs to custom callback objects.\n  const callbackConfigs =\n      generic_utils.toList(callbacks) as CustomCallbackArgs[];\n  return callbackConfigs.map(\n      callbackConfig => new CustomCallback(callbackConfig, yieldEvery));\n}\n\nexport declare type BaseCallbackConstructor = {\n  new (): BaseCallback\n};\n\n/**\n * A global registry for callback constructors to be used during\n * LayersModel.fit().\n */\nexport class CallbackConstructorRegistry {\n  private static constructors:\n      {[verbosityLevel: number]: BaseCallbackConstructor[]} = {};\n\n  /**\n   * Blocks public access to constructor.\n   */\n  private constructor() {}\n\n  /**\n   * Register a tf.LayersModel.fit() callback constructor.\n   *\n   * The registered callback constructor will be used to instantiate\n   * callbacks for every tf.LayersModel.fit() call afterwards.\n   *\n   * @param verbosityLevel Level of verbosity at which the `callbackConstructor`\n   *   is to be reigstered.\n   * @param callbackConstructor A no-arg constructor for `tf.Callback`.\n   * @throws Error, if the same callbackConstructor has been registered before,\n   *   either at the same or a different `verbosityLevel`.\n   */\n  static registerCallbackConstructor(\n      verbosityLevel: number, callbackConstructor: BaseCallbackConstructor) {\n    util.assert(\n        verbosityLevel >= 0 && Number.isInteger(verbosityLevel),\n        () => `Verbosity level is expected to be an integer >= 0, ` +\n            `but got ${verbosityLevel}`);\n    CallbackConstructorRegistry.checkForDuplicate(callbackConstructor);\n    if (CallbackConstructorRegistry.constructors[verbosityLevel] == null) {\n      CallbackConstructorRegistry.constructors[verbosityLevel] = [];\n    }\n    CallbackConstructorRegistry.constructors[verbosityLevel].push(\n        callbackConstructor);\n  }\n\n  private static checkForDuplicate(callbackConstructor:\n                                       BaseCallbackConstructor) {\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const constructors = CallbackConstructorRegistry.constructors[+levelName];\n      constructors.forEach(ctor => {\n        if (ctor === callbackConstructor) {\n          throw new ValueError('Duplicate callback constructor.');\n        }\n      });\n    }\n  }\n\n  /**\n   * Clear all registered callback constructors.\n   */\n  protected static clear() {\n    CallbackConstructorRegistry.constructors = {};\n  }\n\n  /**\n   * Create callbacks using the registered callback constructors.\n   *\n   * Given `verbosityLevel`, all constructors registered at that level or above\n   * will be called and the instantiated callbacks will be used.\n   *\n   * @param verbosityLevel: Level of verbosity.\n   */\n  static createCallbacks(verbosityLevel: number): BaseCallback[] {\n    const constructors: BaseCallbackConstructor[] = [];\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const level = +levelName;\n      if (verbosityLevel >= level) {\n        constructors.push(...CallbackConstructorRegistry.constructors[level]);\n      }\n    }\n    return constructors.map(ctor => new ctor());\n  }\n}\n\nexport function configureCallbacks(\n    callbacks: BaseCallback[], verbose: ModelLoggingVerbosity, epochs: number,\n    initialEpoch: number, numTrainSamples: number, stepsPerEpoch: number,\n    batchSize: number, doValidation: boolean,\n    callbackMetrics: string[]): {callbackList: CallbackList, history: History} {\n  const history = new History();\n  const actualCallbacks: BaseCallback[] = [\n    new BaseLogger(), ...CallbackConstructorRegistry.createCallbacks(verbose)\n  ];\n  if (callbacks != null) {\n    actualCallbacks.push(...callbacks);\n  }\n  actualCallbacks.push(history);\n  const callbackList = new CallbackList(actualCallbacks);\n\n  // TODO(cais): Figure out when this LayersModel instance can have a\n  // dynamically\n  //   set property called 'callback_model' as in PyKeras.\n\n  callbackList.setParams({\n    epochs,\n    initialEpoch,\n    samples: numTrainSamples,\n    steps: stepsPerEpoch,\n    batchSize,\n    verbose,\n    doValidation,\n    metrics: callbackMetrics,\n  });\n  return {callbackList, history};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source layers/__init__.py */\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {deserializeKerasObject} from '../utils/generic_utils';\n\n/**\n * Instantiate a layer from a config dictionary.\n * @param config dict of the form {class_name: str, config: dict}\n * @param customObjects dict mapping class names (or function names)\n *   of custom (non-Keras) objects to class/functions\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns Layer instance (may be LayersModel, Sequential, Layer...)\n */\nexport function deserialize(\n    config: serialization.ConfigDict,\n    customObjects = {} as serialization.ConfigDict,\n    fastWeightInit = false): serialization.Serializable {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'layer', fastWeightInit);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: losses.py */\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\nimport * as K from './backend/tfjs_backend';\nimport {ValueError} from './errors';\nimport {LossOrMetricFn} from './types';\n\n/**\n * Normalizes a tensor wrt the L2 norm alongside the specified axis.\n * @param x\n * @param axis Axis along which to perform normalization.\n */\nexport function l2Normalize(x: Tensor, axis?: number): Tensor {\n  return tidy(() => {\n    if (x.dtype !== 'float32') {\n      x = x.asType('float32');\n    }\n    const squareSum = tfc.sum(K.square(x), axis, true);\n    const epsilonTensor = tfc.fill(squareSum.shape, epsilon());\n    const norm = tfc.sqrt(tfc.maximum(squareSum, epsilonTensor));\n    return tfc.div(x, norm);\n  });\n}\n\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(K.square(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(tfc.abs(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const diff = tfc.sub(yTrue, yPred);\n    const clippedTrue =\n        tfc.clipByValue(tfc.abs(yTrue), epsilon(), Number.MAX_VALUE);\n    const absResult = tfc.abs(tfc.div(diff, clippedTrue));\n    return tfc.mul(100, tfc.mean(absResult, -1));\n  });\n}\n\nexport function meanSquaredLogarithmicError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), Number.MAX_VALUE);\n    const firstLog = tfc.log(tfc.add(1, clippedPred));\n\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), Number.MAX_VALUE);\n    const secondLog = tfc.log(tfc.add(1, clippedTrue));\n\n    return tfc.mean(K.square(tfc.sub(firstLog, secondLog)), -1);\n  });\n}\n\nexport function squaredHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(K.square(maxResult), -1);\n  });\n}\n\nexport function hinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(maxResult, -1);\n  });\n}\n\nexport function categoricalHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const pos = tfc.sum(tfc.mul(yTrue, yPred), -1);\n    const neg = tfc.max(tfc.mul(tfc.sub(1, yTrue), yPred), -1);\n    return tfc.maximum(0, tfc.add(1, tfc.sub(neg, pos)));\n  });\n}\n\n/**\n * Logarithm of the hyperbolic cosine of the prediction error.\n *\n * `log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and\n * to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly\n * like the mean squared error, but will not be so strongly affected by the\n * occasional wildly incorrect prediction.\n */\nexport function logcosh(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const log2 = Math.log(2);\n    const predictionDiff = tfc.sub(yPred, yTrue);\n    const logcoshResult = tfc.sub(\n        tfc.add(predictionDiff, tfc.softplus(tfc.mul(-2, predictionDiff))),\n        log2);\n    return tfc.mean(logcoshResult, -1);\n  });\n}\n\nexport function categoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    if (fromLogits) {\n      output = tfc.softmax(output);\n    } else {\n      // scale preds so that the class probabilities of each sample sum to 1.\n      const outputSum = tfc.sum(output, output.shape.length - 1, true);\n      output = tfc.div(output, outputSum);\n    }\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    return tfc.neg(tfc.sum(\n        tfc.mul(target.toFloat(), tfc.log(output)), output.shape.length - 1));\n  });\n}\n\n/**\n * Categorical crossentropy with integer targets.\n *\n * @param target An integer tensor.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n */\nexport function sparseCategoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    const flatTarget = tfc.floor(K.flatten(target)).toInt() as Tensor1D;\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    const outputShape = output.shape;\n    const oneHotTarget =\n        tfc.oneHot(flatTarget, outputShape[outputShape.length - 1])\n            .reshape(outputShape);\n    return categoricalCrossentropy(oneHotTarget, output, fromLogits);\n  });\n}\n\n/**\n * From TensorFlow's implementation in nn_impl.py:\n *\n * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n *      z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n *    = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n *    = (1 - z) * x + log(1 + exp(-x))\n *    = x - x * z + log(1 + exp(-x))\n * For x < 0, to avoid overflow in exp(-x), we reformulate the above\n *      x - x * z + log(1 + exp(-x))\n *    = log(exp(x)) - x * z + log(1 + exp(-x))\n *    = - x * z + log(1 + exp(x))\n * Hence, to ensure stability and avoid overflow, the implementation uses this\n * equivalent formulation\n *    max(x, 0) - x * z + log(1 + exp(-abs(x)))\n *\n * @param labels The labels.\n * @param logits The logits.\n */\nexport function sigmoidCrossEntropyWithLogits(\n    labels: Tensor, logits: Tensor): Tensor {\n  if (!util.arraysEqual(labels.shape, logits.shape)) {\n    throw new ValueError(\n        `logits and labels must have the same shape, but got shapes ` +\n        `${JSON.stringify(labels.shape)} and ${JSON.stringify(logits.shape)}`);\n  }\n  return tidy(() => {\n    // The logistic loss formula from above is\n    //   x - x * z + log(1 + exp(-x))\n    // For x < 0, a more numerically stable formula is\n    //   -x * z + log(1 + exp(x))\n    // Note that these two expressions can be combined into the following:\n    //   max(x, 0) - x * z + log(1 + exp(-abs(x)))\n    const reluLogits = logits.relu();\n    const negAbsLogits = logits.abs().neg();\n    return reluLogits.sub(logits.mul(labels)).add(negAbsLogits.exp().log1p());\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    let y: Tensor;\n    y = tfc.clipByValue(yPred, epsilon(), 1 - epsilon());\n    y = tfc.log(tfc.div(y, tfc.sub(1, y)));\n    return tfc.mean(sigmoidCrossEntropyWithLogits(yTrue, y), -1);\n  });\n}\n\nexport function kullbackLeiblerDivergence(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), 1);\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), 1);\n    return tfc.sum(\n        tfc.mul(yTrue, tfc.log(tfc.div(clippedTrue, clippedPred))), -1);\n  });\n}\n\nexport function poisson(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const logPred = tfc.log(tfc.add(epsilon(), yPred));\n    return tfc.mean(tfc.sub(yPred, tfc.mul(yTrue, logPred)), -1);\n  });\n}\n\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const trueNormalized = l2Normalize(yTrue, -1);\n    const predNormalized = l2Normalize(yPred, -1);\n    const trueXPred = tfc.mul(trueNormalized, predNormalized);\n    return tfc.neg(tfc.sum(trueXPred, -1));\n  });\n}\n\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const msle = meanSquaredLogarithmicError;\nexport const MSLE = meanSquaredLogarithmicError;\nexport const kld = kullbackLeiblerDivergence;\nexport const KLD = kullbackLeiblerDivergence;\nexport const cosine = cosineProximity;\n\n// TODO(michaelterry): Add deserialize() function.\n\nexport const lossesMap: {[functionName: string]: LossOrMetricFn} = {\n  meanSquaredError,\n  meanAbsoluteError,\n  meanAbsolutePercentageError,\n  meanSquaredLogarithmicError,\n  squaredHinge,\n  hinge,\n  categoricalHinge,\n  logcosh,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  binaryCrossentropy,\n  kullbackLeiblerDivergence,\n  poisson,\n  cosineProximity\n};\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function get(identifierOrFn: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifierOrFn === 'string') {\n    if (identifierOrFn in lossesMap) {\n      return lossesMap[identifierOrFn];\n    }\n    let errMsg = `Unknown loss ${identifierOrFn}`;\n    if (identifierOrFn.toLowerCase().includes('softmaxcrossentropy')) {\n      errMsg = `Unknown loss ${identifierOrFn}. ` +\n          'Use \"categoricalCrossentropy\" as the string name for ' +\n          'tf.losses.softmaxCrossEntropy';\n    }\n    throw new ValueError(errMsg);\n  } else {\n    return identifierOrFn;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Built-in metrics.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {NotImplementedError, ValueError} from './errors';\nimport {categoricalCrossentropy as categoricalCrossentropyLoss, cosineProximity, meanAbsoluteError, meanAbsolutePercentageError, meanSquaredError, sparseCategoricalCrossentropy as sparseCategoricalCrossentropyLoss} from './losses';\nimport {binaryCrossentropy as lossBinaryCrossentropy} from './losses';\nimport {lossesMap} from './losses';\nimport {LossOrMetricFn} from './types';\nimport * as util from './utils/generic_utils';\n\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const threshold = tfc.mul(.5, tfc.onesLike(yPred));\n    const yPredThresholded = K.cast(tfc.greater(yPred, threshold), yTrue.dtype);\n    return tfc.mean(tfc.equal(yTrue, yPredThresholded), -1);\n  });\n}\n\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(\n      () => K.cast(\n          tfc.equal(tfc.argMax(yTrue, -1), tfc.argMax(yPred, -1)), 'float32'));\n}\n\nfunction truePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nfunction falseNegatives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(0)).sum().cast('float32');\n  });\n}\n\nfunction falsePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(0), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fp = falsePositives(yTrue, yPred);\n\n    const denominator = tp.add(fp);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fn = falseNegatives(yTrue, yPred);\n\n    const denominator = tp.add(fn);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return lossBinaryCrossentropy(yTrue, yPred);\n}\n\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  if (yTrue.rank === yPred.rank) {\n    yTrue = yTrue.squeeze([yTrue.rank - 1]);\n  }\n  yPred = yPred.argMax(-1);\n  if (yPred.dtype !== yTrue.dtype) {\n    yPred = yPred.asType(yTrue.dtype);\n  }\n  return tfc.equal(yTrue, yPred).asType('float32');\n}\n\nexport function topKCategoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\nexport function sparseTopKCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\n// Aliases.\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const categoricalCrossentropy = categoricalCrossentropyLoss;\nexport const cosine = cosineProximity;\nexport const sparseCategoricalCrossentropy = sparseCategoricalCrossentropyLoss;\n\n// TODO(cais, nielsene): Add serialize().\n\nexport const metricsMap: {[functionName: string]: LossOrMetricFn} = {\n  binaryAccuracy,\n  categoricalAccuracy,\n  precision,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  mse,\n  MSE,\n  mae,\n  MAE,\n  mape,\n  MAPE,\n  cosine\n};\n\nexport function get(identifier: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifier === 'string' && identifier in metricsMap) {\n    return metricsMap[identifier];\n  } else if (typeof identifier !== 'string' && identifier != null) {\n    return identifier;\n  } else {\n    throw new ValueError(`Unknown metric ${identifier}`);\n  }\n}\n\n/**\n * Get the shortcut function name.\n *\n * If the fn name is a string,\n *   directly return the string name.\n * If the function is included in metricsMap or lossesMap,\n *   return key of the map.\n *   - If the function relative to multiple keys,\n *     return the first found key as the function name.\n *   - If the function exists in both lossesMap and metricsMap,\n *     search lossesMap first.\n * If the function is not included in metricsMap or lossesMap,\n *   return the function name.\n *\n * @param fn loss function, metric function, or short cut name.\n * @returns Loss or Metric name in string.\n */\nexport function getLossOrMetricName(fn: string|LossOrMetricFn): string {\n  util.assert(fn !== null, `Unknown LossOrMetricFn ${fn}`);\n  if (typeof fn === 'string') {\n    return fn;\n  } else {\n    let fnName;\n    for (const key of Object.keys(lossesMap)) {\n      if (lossesMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    for (const key of Object.keys(metricsMap)) {\n      if (metricsMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    return (fn as Function).name;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Optimizers.\n */\n\nimport {Optimizer, train} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\n\nimport {ValueError} from './errors';\n\n// Add (de)serialize()\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function getOptimizer(identifier: string): Optimizer {\n  const optimizerMap: {[optimizerName: string]: () => Optimizer} = {\n    'Adagrad': () => train.adagrad(0.01),\n    'Adadelta': () => train.adadelta(1, 0.95, epsilon()),\n    'Adam': () => train.adam(0.001, 0.9, 0.999, epsilon()),\n    'Adamax': () => train.adamax(0.002, 0.9, 0.999, epsilon(), 0),\n    'RMSProp': () => train.rmsprop(0.001, 0.9, 0, epsilon()),\n    'SGD': () => train.sgd(0.01)\n  };\n  optimizerMap['adagrad'] = optimizerMap['Adagrad'];\n  optimizerMap['adadelta'] = optimizerMap['Adadelta'];\n  optimizerMap['adam'] = optimizerMap['Adam'];\n  optimizerMap['adamax'] = optimizerMap['Adamax'];\n  optimizerMap['rmsprop'] = optimizerMap['RMSProp'];\n  optimizerMap['sgd'] = optimizerMap['SGD'];\n\n  if (identifier in optimizerMap) {\n    return optimizerMap[identifier]();\n  }\n  throw new ValueError(`Unknown Optimizer ${identifier}`);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/** Utility functions related to user-defined metadata. */\n\n// Maximum recommended serialized size for user-defined metadata.\n// Beyond this limit, a warning message will be printed during model loading and\n// saving.\nexport const MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH = 1 * 1024 * 1024;\n\n/**\n * Check validity of user-defined metadata.\n *\n * @param userDefinedMetadata\n * @param modelName Name of the model that the user-defined metadata belongs to.\n *   Used during construction of error messages.\n * @param checkSize Whether to check the size of the metadata is under\n *   recommended limit. Default: `false`. If `true`, will try stringify the\n *   JSON object and print a console warning if the serialzied size is above the\n *   limit.\n * @throws Error if `userDefinedMetadata` is not a plain JSON object.\n */\nexport function checkUserDefinedMetadata(\n    userDefinedMetadata: {}, modelName: string, checkSize = false): void {\n  if (userDefinedMetadata == null ||\n      typeof userDefinedMetadata !== 'object' ||\n      Object.getPrototypeOf(userDefinedMetadata) !== Object.prototype ||\n      !plainObjectCheck(userDefinedMetadata)) {\n    throw new Error(\n        'User-defined metadata is expected to be a JSON object, but is not.');\n  }\n\n  if (checkSize) {\n    const out = JSON.stringify(userDefinedMetadata);\n    if (out.length > MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH) {\n      console.warn(\n          `User-defined metadata of model \"${modelName}\" is too large in ` +\n          `size (length=${out.length} when serialized). It is not ` +\n          `recommended to store such large objects in user-defined metadata. ` +\n          `Please make sure its serialized length is <= ` +\n          `${MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH}.`);\n    }\n  }\n}\n\n/**\n * Check if an input is plain JSON object or any valid subfield of it.\n *\n * @param x The input to be checked.\n * @param assertObject Whether to assert `x` is a JSON object, i.e., reject\n *   cases of arrays and primitives.\n * @return Returns `true` if and only if `x` is a plain JSON object,\n *   a JSON-valid primitive including string, number, boolean and null,\n *   or an array of the said types.\n */\n// tslint:disable-next-line:no-any\nexport function plainObjectCheck(x: any): boolean {\n  if (x === null) {\n    // Note: typeof `null` is 'object', and `null` is valid in JSON.\n    return true;\n  } else if (typeof x === 'object') {\n    if (Object.getPrototypeOf(x) === Object.prototype) {\n      // `x` is a JavaScript object and its prototype is Object.\n      const keys = Object.keys(x);\n      for (const key of keys) {\n        if (typeof key !== 'string') {\n          // JSON keys must be strings.\n          return false;\n        }\n        if (!plainObjectCheck(x[key])) {  // Recursive call.\n          return false;\n        }\n      }\n      return true;\n    } else {\n      // `x` is a JavaScript object but its prototype is not Object.\n      if (Array.isArray(x)) {\n        // `x` is a JavaScript array.\n        for (const item of x) {\n          if (!plainObjectCheck(item)) {  // Recursive call.\n            return false;\n          }\n        }\n        return true;\n      } else {\n        // `x` is a JavaScript object and its prototype is not Object,\n        // and it's not an Array. I.e., it's a complex object such as\n        // `Error` and `Date`.\n        return false;\n      }\n    }\n  } else {\n    // `x` is not a JavaScript object or `null`.\n    const xType = typeof x;\n    return xType === 'string' || xType === 'number' || xType === 'boolean';\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {Container} from '../engine/container';\nimport {Layer, Node} from '../engine/topology';\nimport {countParamsInWeights} from './variable_utils';\n\n/**\n * Print the summary of a LayersModel object.\n *\n * @param model tf.LayersModel instance.\n * @param lineLength Total length of printed lines. Set this to adapt to the\n *   display to different terminal or console sizes.\n * @param positions Relative or absolute positions of log elements in each\n *   line. Each number corresponds to right-most (i.e., ending) position of a\n *   column.\n *   If not provided, defaults to `[0.45, 0.85, 1]` for sequential-like\n *   models and `[0.33, 0.55, 0.67, 1]` for non-sequential like models.\n * @param printFn Print function to use.\n *   It will be called on each line of the summary. You can provide a custom\n *   function in order to capture the string summary. Defaults to `console.log`.\n */\nexport function printSummary(\n    model: Container, lineLength?: number, positions?: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void =\n        console.log): void {\n  const sequentialLike = isModelSequentialLike(model);\n\n  // Header names for different log elements.\n  const toDisplay: string[] = ['Layer (type)', 'Output shape', 'Param #'];\n  if (sequentialLike) {\n    lineLength = lineLength || 65;\n    positions = positions || [0.45, 0.85, 1];\n  } else {\n    lineLength = lineLength || 98;\n    positions = positions || [0.33, 0.55, 0.67, 1];\n    // Header names for different log elements.\n  }\n\n  if (positions[positions.length - 1] <= 1) {\n    // `positions` is relative. Convert it to absolute positioning.\n    positions = positions.map(p => Math.floor(lineLength * p));\n  }\n\n  let relevantNodes: Node[];\n  if (!sequentialLike) {\n    toDisplay.push('Receives inputs');\n    relevantNodes = [];\n    for (const depth in model.nodesByDepth) {\n      relevantNodes.push(...model.nodesByDepth[depth]);\n    }\n  }\n\n  printFn('_'.repeat(lineLength));\n  printRow(toDisplay, positions, printFn);\n  printFn('='.repeat(lineLength));\n\n  const layers = model.layers;\n  for (let i = 0; i < layers.length; ++i) {\n    if (sequentialLike) {\n      printLayerSummary(layers[i], positions, printFn);\n    } else {\n      printLayerSummaryWithConnections(\n          layers[i], positions, relevantNodes, printFn);\n    }\n    printFn((i === layers.length - 1 ? '=' : '_').repeat(lineLength));\n  }\n\n  // tslint:disable-next-line:no-any\n  (model as any).checkTrainableWeightsConsistency();\n\n  const trainableCount = countTrainableParams(model);\n  const nonTrainableCount = countParamsInWeights(model.nonTrainableWeights);\n\n  printFn(`Total params: ${trainableCount + nonTrainableCount}`);\n  printFn(`Trainable params: ${trainableCount}`);\n  printFn(`Non-trainable params: ${nonTrainableCount}`);\n  printFn('_'.repeat(lineLength));\n}\n\nfunction countTrainableParams(model: Container): number {\n  let trainableCount: number;\n  // tslint:disable:no-any\n  if ((model as any).collectedTrainableWeights != null) {\n    trainableCount =\n        countParamsInWeights((model as any).collectedTrainableWeights);\n  } else {\n    trainableCount = countParamsInWeights(model.trainableWeights);\n  }\n  // tslint:enable:no-any\n  return trainableCount;\n}\n\nfunction isModelSequentialLike(model: Container): boolean {\n  let sequentialLike = true;\n  const nodesByDepth: Node[][] = [];\n  const nodes: Node[] = [];\n  for (const depth in model.nodesByDepth) {\n    nodesByDepth.push(model.nodesByDepth[depth]);\n  }\n  for (const depthNodes of nodesByDepth) {\n    if (depthNodes.length > 1 ||\n        depthNodes.length === 1 && depthNodes[0].inboundLayers.length > 1) {\n      sequentialLike = false;\n      break;\n    }\n    nodes.push(...depthNodes);\n  }\n  if (sequentialLike) {\n    // Search for shared layers.\n    for (const layer of model.layers) {\n      let flag = false;\n      for (const node of layer.inboundNodes) {\n        if (nodes.indexOf(node) !== -1) {\n          if (flag) {\n            sequentialLike = false;\n            break;\n          } else {\n            flag = true;\n          }\n        }\n      }\n      if (!sequentialLike) {\n        break;\n      }\n    }\n  }\n  return sequentialLike;\n}\n\nfunction printRow(\n    fields: string[], positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void = console.log) {\n  let line = '';\n  for (let i = 0; i < fields.length; ++i) {\n    if (i > 0) {\n      line = line.slice(0, line.length - 1) + ' ';\n    }\n    line += fields[i];\n    line = line.slice(0, positions[i]);\n    line += ' '.repeat(positions[i] - line.length);\n  }\n  printFn(line);\n}\n\n/**\n * Prints a summary for a single Layer, without connectivity information.\n *\n * @param layer: Layer instance to print.\n */\nfunction printLayerSummary(\n    layer: Layer, positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const name = layer.name;\n  const className = layer.getClassName();\n  const fields: string[] =\n      [`${name} (${className})`, outputShape, layer.countParams().toString()];\n  printRow(fields, positions, printFn);\n}\n\n/**\n * Prints a summary for a single Layer, with connectivity information.\n */\nfunction printLayerSummaryWithConnections(\n    layer: Layer, positions: number[], relevantNodes: Node[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const connections: string[] = [];\n  for (const node of layer.inboundNodes) {\n    if (relevantNodes != null && relevantNodes.length > 0 &&\n        relevantNodes.indexOf(node) === -1) {\n      continue;\n    }\n    for (let i = 0; i < node.inboundLayers.length; ++i) {\n      const inboundLayer = node.inboundLayers[i].name;\n      const inboundLayerIndex = node.nodeIndices[i];\n      const inboundTensorIndex = node.tensorIndices[i];\n      connections.push(\n          `${inboundLayer}[${inboundLayerIndex}][${inboundTensorIndex}]`);\n    }\n  }\n  const name = layer.name;\n  const className = layer.getClassName();\n  const firstConnection = connections.length === 0 ? '' : connections[0];\n  const fields: string[] = [\n    `${name} (${className})`, outputShape, layer.countParams().toString(),\n    firstConnection\n  ];\n\n  printRow(fields, positions, printFn);\n  for (let i = 1; i < connections.length; ++i) {\n    printRow(['', '', '', connections[i]], positions, printFn);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Porting note: This file doesn't exist in PyKeras.\n// Its purpose here is to centralize the boundary layer between\n// tfjs-layers's internal Config TS-Centric format and PyKeras's\n// serialized Python Config format.\n\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {PyJsonValue} from '../keras_format/types';\nimport * as generic_utils from '../utils/generic_utils';\n// tslint:enable\n\n/**\n * Test whether a value in an array is the name of a LayersModel or Layer.\n * @param key The key name that the value is found under. Note that the key\n *   may not be at the level immediately above the value, if the value is in a\n *   nested array.\n * @param index Index of the value in the Array that it is found in.\n * @param value The value object.\n * @returns A boolean indicating whether value is a name.\n */\nfunction isArrayItemInputOrOutputName<T>(\n    key: string, index: number, value: T): boolean {\n  return (key === 'inboundNodes' || key === 'outputLayers' ||\n          key === 'inputLayers') &&\n      index === 0 && typeof value === 'string';\n}\n\n/**\n * Convert a Pythonic config object to TypeScript config object.\n * @param pythonicConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertPythonicToTs(\n    pythonicConfig: PyJsonValue, key?: string): serialization.ConfigDictValue {\n  if (pythonicConfig === null) {\n    return null;\n  } else if (typeof pythonicConfig === 'string') {\n    return generic_utils.toCamelCase(pythonicConfig);\n  } else if (\n      (typeof pythonicConfig === 'number') ||\n      (typeof pythonicConfig === 'boolean')) {\n    return pythonicConfig;\n  } else if (pythonicConfig instanceof Array) {\n    const tsArray = [];\n    const arrayLength = pythonicConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = pythonicConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        tsArray.push(item);\n      } else {\n        tsArray.push(convertPythonicToTs(item, key));\n      }\n    }\n    return tsArray;\n  } else {\n    const tsDict: serialization.ConfigDict = {};\n    for (const pythonicKey of Object.keys(pythonicConfig)) {\n      const pythonicValue = pythonicConfig[pythonicKey];\n      if (pythonicKey === 'name' && typeof pythonicValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // camel-case conversion.\n        tsDict[pythonicKey] = pythonicValue;\n      } else {\n        const tsKey = generic_utils.toCamelCase(pythonicKey);\n        tsDict[tsKey] = convertPythonicToTs(pythonicValue, tsKey);\n      }\n    }\n    return tsDict;\n  }\n}\n\n/**\n * Convert a TypeScript config object to Python config object.\n * @param tsConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertTsToPythonic(\n    tsConfig: serialization.ConfigDictValue, key?: string): PyJsonValue {\n  if (tsConfig === null || tsConfig === undefined) {\n    return null;\n  } else if (typeof tsConfig === 'string') {\n    return generic_utils.toSnakeCase(tsConfig);\n  } else if (\n      (typeof tsConfig === 'number') || (typeof tsConfig === 'boolean')) {\n    return tsConfig;\n  } else if (tsConfig instanceof Array) {\n    const pyArray = [];\n    const arrayLength = tsConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = tsConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        pyArray.push(item);\n      } else {\n        pyArray.push(convertTsToPythonic(item, key));\n      }\n    }\n    return pyArray;\n  } else {\n    const pyDict: serialization.ConfigDict = {};\n    for (const tsKey of Object.keys(tsConfig)) {\n      const tsValue = tsConfig[tsKey];\n      const pyKey = generic_utils.toSnakeCase(tsKey);\n      if ((tsKey === 'name' || tsKey === 'className') &&\n          typeof tsValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // snake-case conversion.\n        pyDict[pyKey] = tsValue;\n      } else {\n        pyDict[pyKey] = convertTsToPythonic(tsValue, tsKey);\n      }\n    }\n    return pyDict;\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Executor: Evaluates SymbolicTensor based on feeds.\n */\n\nimport {cast, dispose, memory, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {ValueError} from '../errors';\nimport {Kwargs} from '../types';\nimport {toList} from '../utils/generic_utils';\n\nimport {InputLayer} from './input_layer';\nimport {SymbolicTensor} from './topology';\n\n/**\n * Helper function to check the dtype and shape compatibility of a feed value.\n */\nfunction assertFeedCompatibility(key: SymbolicTensor, val: Tensor): Tensor {\n  // Check dtype compatibility.\n  if (key.dtype == null || key.dtype === val.dtype) {\n    //  a.  If types match, return val tensor as is.\n    return val;\n  }\n  try {\n    //  b. Attempt to convert to expected type.\n    return cast(val, key.dtype);\n  } catch (err) {\n    //  c. If conversion fails, return helpful error.\n    throw new ValueError(\n        `The dtype of the feed (${val.dtype}) can not be cast to the dtype ` +\n        `of the key '${key.name}' (${key.dtype}).`);\n  }\n}\n\n/**\n * A concrete Tensor value for a symbolic tensor as the key.\n */\nexport interface Feed {\n  key: SymbolicTensor;\n  value: Tensor;\n}\n\n/**\n * FeedDict: A mapping from unique SymbolicTensors to feed values for them.\n * A feed value is a concrete value represented as an `Tensor`.\n */\nexport class FeedDict {\n  private id2Value: {[id: number]: Tensor} = {};\n  private id2Mask: {[id: number]: Tensor} = {};\n  private name2Id: {[name: string]: number} = {};\n\n  /**\n   * Constructor, optionally does copy-construction.\n   * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case\n   *   copy-construction will be performed.\n   */\n  constructor(feeds?: Feed[]|FeedDict) {\n    if (feeds instanceof FeedDict) {\n      for (const id in feeds.id2Value) {\n        this.id2Value[id] = feeds.id2Value[id];\n        if (id in feeds.id2Mask) {\n          this.id2Mask[id] = feeds.id2Mask[id];\n        }\n      }\n    } else {\n      if (feeds == null) {\n        return;\n      }\n      for (const feed of feeds) {\n        this.add(feed.key, feed.value);\n      }\n    }\n  }\n\n  /**\n   * Add a key-value pair to the FeedDict.\n   *\n   * @param key The key of the feed.\n   * @param value The value of the tensor feed.\n   * @param mask The value of the mask feed (optional).\n   * @returns This `FeedDict`.\n   * @throws ValueError: If the key `SymbolicTensor` already exists in the\n   *   `FeedDict`.\n   */\n  add(key: SymbolicTensor, value: Tensor, mask?: Tensor): FeedDict {\n    if (this.id2Value[key.id] == null) {\n      this.id2Value[key.id] = assertFeedCompatibility(key, value);\n      this.name2Id[key.name] = key.id;\n      if (mask != null) {\n        this.id2Mask[key.id] = mask;\n      }\n    } else {\n      throw new ValueError(`Duplicate key: name=${key.name}, id=${key.id}`);\n    }\n    return this;\n  }\n\n  /**\n   * Add a Feed to the FeedDict.\n   * @param feed The new `Feed` to add.\n   * @returns This `FeedDict`.\n   */\n  addFeed(feed: Feed) {\n    this.add(feed.key, feed.value);\n  }\n\n  /**\n   * Probe whether a key already exists in the FeedDict.\n   * @param key\n   */\n  hasKey(key: SymbolicTensor): boolean {\n    return this.id2Value[key.id] != null;\n  }\n\n  /**\n   * Get all the SymbolicTensor available in this FeedDict.\n   */\n  names(): string[] {\n    return Object.keys(this.name2Id);\n  }\n\n  /**\n   * Get the feed value for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed value.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getValue(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Value[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Value[id];\n    }\n  }\n\n  /**\n   * Get the feed mask for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed mask.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getMask(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Mask[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Mask[id];\n    }\n  }\n\n  /** Dispose all mask Tensors held by this object. */\n  disposeMasks() {\n    if (this.id2Mask != null) {\n      dispose(this.id2Mask);\n    }\n  }\n}\n\n// Cache for topologically sorted SymbolicTensors for given execution\n// targets (i.e., fetches).\nconst cachedSorted: {[concatFetchNames: string]: SymbolicTensor[]} = {};\n\n// Cache for recipient count maps for given execution targets (i.e., fetches).\nconst cachedRecipientCounts:\n    {[concatFetchNames: string]: {[fetchName: string]: number}} = {};\n\n/**\n * Interface for the optional object used for probing the memory\n * usage and other statistics during execution.\n */\nexport interface ExecutionProbe {\n  /**\n   * Maximum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  maxNumTensors?: number;\n\n  /**\n   * Minimum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  minNumTensors?: number;\n}\n\n/**\n * Execute a SymbolicTensor by using concrete feed values.\n *\n * A `SymbolicTensor` object is a node in a computation graph of TF.js\n * Layers. The object is backed by a source layer and input\n * `SymbolicTensor`s to the source layer. This method evaluates\n * the `call()` method of the source layer, using concrete values of the\n * inputs obtained from either\n * * `feedDict`, if the input key exists in `feedDict`, or else,\n * * a recursive call to `execute()` itself.\n *\n * @param x: The `SymbolicTensor` to execute.\n * @param feedDict: The feed values, as base condition of the recursion.\n *   execution.\n * @param kwargs: Optional keyword arguments.\n * @param probe: A probe object (of interface `ExecutionProbe`) used for\n *   testing memory footprint of `execute` calls.\n * @returns Result of the execution.\n * @throws ValueError: If any `SymbolicTensor`s from `InputLayer`s\n *   encountered during the execution lacks a feed value in `feedDict`.\n */\nexport function execute(\n    fetches: SymbolicTensor|SymbolicTensor[], feedDict: FeedDict,\n    kwargs?: Kwargs, probe?: ExecutionProbe): Tensor|\n    Tensor[]|[Tensor | Tensor[]] {\n  const training: boolean = kwargs == null ? false : kwargs['training'];\n\n  const arrayFetches = Array.isArray(fetches);\n  const fetchArray: SymbolicTensor[] =\n      arrayFetches ? fetches as SymbolicTensor[] : [fetches as SymbolicTensor];\n\n  const outputNames = fetchArray.map(t => t.name);\n  const finalOutputs: Tensor[] = [];\n  const feedNames = feedDict.names();\n  for (const outputName of outputNames) {\n    if (feedNames.indexOf(outputName) !== -1) {\n      finalOutputs.push(feedDict.getValue(outputName));\n    } else {\n      finalOutputs.push(null);\n    }\n  }\n\n  if (probe != null) {\n    // For optional probing of memory footprint during execution.\n    probe.maxNumTensors = -Infinity;\n    probe.minNumTensors = Infinity;\n  }\n\n  // Check cache.\n  const fetchAndFeedKey =\n      outputNames.join(',') + '|' + feedDict.names().join(',');\n  let sorted: SymbolicTensor[];\n  let recipientCounts: {[fetchName: string]: number};\n  if (cachedSorted[fetchAndFeedKey] == null) {\n    // Cache doesn't contain the desired combination of fetches. Compute\n    // topological sort for the combination for the first time.\n    const out = getTopologicalSortAndRecipientCounts(fetchArray, feedDict);\n    sorted = out.sorted;\n    recipientCounts = out.recipientCounts;\n\n    // Store results in cache for future use.\n    cachedSorted[fetchAndFeedKey] = sorted;\n    cachedRecipientCounts[fetchAndFeedKey] = recipientCounts;\n  }\n  sorted = cachedSorted[fetchAndFeedKey];\n  recipientCounts = {};\n  if (!training) {\n    Object.assign(recipientCounts, cachedRecipientCounts[fetchAndFeedKey]);\n  }\n\n  const internalFeedDict = new FeedDict(feedDict);\n\n  // Start iterative execution on the topologically-sorted SymbolicTensors.\n  for (let i = 0; i < sorted.length; ++i) {\n    if (probe != null) {\n      // For optional probing of memory usage during execution.\n      const numTensors = memory().numTensors;\n      if (numTensors > probe.maxNumTensors) {\n        probe.maxNumTensors = numTensors;\n      }\n      if (numTensors < probe.minNumTensors) {\n        probe.minNumTensors = numTensors;\n      }\n    }\n\n    const symbolic = sorted[i];\n    const srcLayer = symbolic.sourceLayer;\n    if (srcLayer instanceof InputLayer) {\n      continue;\n    }\n    const inputValues: Tensor[] = [];\n    const inputMasks: Tensor[] = [];\n    const tensorsToDispose: Tensor[] = [];\n\n    let maskExists = false;\n    for (const input of symbolic.inputs) {\n      const value = internalFeedDict.getValue(input);\n      const mask = internalFeedDict.getMask(input);\n      inputValues.push(value);\n      inputMasks.push(mask);\n      if (mask != null) {\n        maskExists = true;\n      }\n      if (!training) {\n        recipientCounts[input.name]--;\n        if (recipientCounts[input.name] === 0 && !feedDict.hasKey(input) &&\n            outputNames.indexOf(input.name) === -1 && !value.isDisposed &&\n            input.sourceLayer.stateful !== true) {\n          tensorsToDispose.push(value);\n        }\n      }\n    }\n\n    if (maskExists) {\n      kwargs = kwargs || {};\n      kwargs['mask'] = inputMasks[0];\n    }\n    const outputTensors =\n        toList(srcLayer.apply(inputValues, kwargs)) as Tensor[];\n    let outputMask: Tensor|Tensor[] = null;\n    if (srcLayer.supportsMasking) {\n      outputMask = srcLayer.computeMask(inputValues, inputMasks);\n    }\n    const layerOutputs = getNodeOutputs(symbolic);\n    const outputSymbolicTensors =\n        Array.isArray(layerOutputs) ? layerOutputs : [layerOutputs];\n    for (let i = 0; i < outputSymbolicTensors.length; ++i) {\n      if (!internalFeedDict.hasKey(outputSymbolicTensors[i])) {\n        internalFeedDict.add(\n            outputSymbolicTensors[i], outputTensors[i],\n            Array.isArray(outputMask) ? outputMask[0] : outputMask);\n      }\n      const index = outputNames.indexOf(outputSymbolicTensors[i].name);\n      if (index !== -1) {\n        finalOutputs[index] = outputTensors[i];\n      }\n    }\n\n    if (!training) {\n      // Clean up Tensors that are no longer needed.\n      dispose(tensorsToDispose);\n    }\n  }\n  // NOTE(cais): Unlike intermediate tensors, we don't discard mask\n  // tensors as we go, because these tensors are sometimes passed over a\n  // series of mutliple layers, i.e., not obeying the immediate input\n  // relations in the graph. If this becomes a memory-usage concern,\n  // we can improve this in the future.\n  internalFeedDict.disposeMasks();\n\n  return arrayFetches ? finalOutputs : finalOutputs[0];\n}\n\ntype RecipientCounts = {\n  [fetchName: string]: number\n};\n\nexport type RecipientMap = {\n  [fetchName: string]: Set<string>;\n};\n\n/**\n * Sort the `SymbolicTensor`s topologically, for an array of fetches.\n *\n * This function calls getTopologicalSortAndRecipientCountsForOneFetch and\n * merges their results.\n *\n * @param fetch The array of fetches requested. Must be a non-empty array.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientCounts: Recipient counts for all SymbolicTensors in `sorted`.\n */\nfunction getTopologicalSortAndRecipientCounts(\n    fetches: SymbolicTensor[], feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientCounts: RecipientCounts} {\n  util.assert(\n      fetches != null && fetches.length > 0,\n      () => `Expected at least one fetch, got none`);\n\n  let finalSorted: SymbolicTensor[] = [];\n  let finalRecipientMap: RecipientMap = {};\n  if (fetches.length === 1) {\n    // Special-casing 1 fetch for efficiency.\n    const out =\n        getTopologicalSortAndRecipientCountsForOneFetch(fetches[0], feedDict);\n    finalSorted = out.sorted;\n    finalRecipientMap = out.recipientMap;\n  } else {\n    const visited = new Set<string>();\n    for (const fetch of fetches) {\n      const {sorted, recipientMap} =\n          getTopologicalSortAndRecipientCountsForOneFetch(fetch, feedDict);\n\n      // Merge sorted SymbolicTensor Arrays.\n      for (const symbolicTensor of sorted) {\n        if (!visited.has(symbolicTensor.name)) {\n          finalSorted.push(symbolicTensor);\n          visited.add(symbolicTensor.name);\n        }\n      }\n\n      // Merge recipient maps.\n      for (const name in recipientMap) {\n        if (finalRecipientMap[name] == null) {\n          finalRecipientMap[name] = new Set<string>();\n        }\n        recipientMap[name].forEach(\n            recipient => finalRecipientMap[name].add(recipient));\n      }\n    }\n  }\n  return {\n    sorted: finalSorted,\n    recipientCounts: recipientMap2Counts(finalRecipientMap)\n  };\n}\n\nfunction recipientMap2Counts(recipientMap: RecipientMap): RecipientCounts {\n  const recipientCounts: RecipientCounts = {};\n  for (const name in recipientMap) {\n    recipientCounts[name] = recipientMap[name].size;\n  }\n  return recipientCounts;\n}\n\n/**\n * Sort the `SymbolicTensor`s topologically, for a single fetch.\n *\n * This helper function processes the upstream SymbolicTensors of a single\n * fetch.\n *\n * @param fetch The single fetch requested.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientMap: Recipient names for all SymbolicTensors in `sorted`.\n */\nexport function getTopologicalSortAndRecipientCountsForOneFetch(\n    fetch: SymbolicTensor, feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientMap: RecipientMap} {\n  const visited = new Set<string>();\n  const sorted: SymbolicTensor[] = [];\n  const recipientMap: RecipientMap = {};\n\n  // Put keys of the feedDict into visited first, so they don't have to be\n  // walked. This is needed in case where there are feeds for intermediate\n  // SymbolicTensors of the graph.\n  for (const key of feedDict.names()) {\n    visited.add(key);\n  }\n\n  const stack: SymbolicTensor[] = [];\n  const marks: number[] = [];\n\n  // Initial population of stack and marks.\n  stack.push(fetch);\n\n  while (stack.length > 0) {\n    const top = stack[stack.length - 1];\n    if (visited.has(top.name)) {\n      stack.pop();\n      continue;\n    }\n    const topIsMarked = marks[marks.length - 1] === stack.length - 1;\n    if (top.inputs.length === 0 || topIsMarked) {\n      // Input SymbolicTensor or all children have been visited.\n      stack.pop();\n      sorted.push(top);\n      visited.add(top.name);\n      if (topIsMarked) {\n        marks.pop();\n      }\n    } else {\n      // A non-input SymbolicTensor whose upstream SymbolicTensors haven't\n      // been visited yet. Push them onto the stack.\n      marks.push(stack.length - 1);\n      for (const input of top.inputs) {\n        // Increment the recipient count. Note that this needs to happen\n        // regardless of whether the SymbolicTensor has been visited before.\n        if (recipientMap[input.name] == null) {\n          recipientMap[input.name] = new Set<string>();\n        }\n        recipientMap[input.name].add(top.name);\n\n        if (visited.has(input.name)) {\n          continue;  // Avoid repeated visits to the same SymbolicTensor.\n        }\n        stack.push(input);\n      }\n    }\n  }\n  return {sorted, recipientMap};\n}\n\n/**\n * Get the symbolic output tensors of the node to which a given fetch belongs.\n * @param fetch The fetched symbolic tensor.\n * @returns The Array of symbolic tensors output by the node to which `fetch`\n *   belongs.\n */\nfunction getNodeOutputs(fetch: SymbolicTensor): SymbolicTensor|\n    SymbolicTensor[] {\n  let layerOutputs: SymbolicTensor|SymbolicTensor[];\n  if (fetch.sourceLayer.inboundNodes.length === 1) {\n    layerOutputs = fetch.sourceLayer.output;\n  } else {\n    let nodeIndex: number = null;\n    for (let i = 0; i < fetch.sourceLayer.inboundNodes.length; ++i) {\n      for (const outputTensor of fetch.sourceLayer.inboundNodes[i]\n               .outputTensors) {\n        if (outputTensor.id === fetch.id) {\n          nodeIndex = i;\n          break;\n        }\n      }\n    }\n    layerOutputs = fetch.sourceLayer.getOutputAt(nodeIndex);\n  }\n  return layerOutputs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {NamedTensorMap, Scalar, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {TensorKeyWithArgsArray} from '../keras_format/node_config';\nimport {PyJsonDict} from '../keras_format/types';\nimport {deserialize as deserializeLayer} from '../layers/serialization';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {convertTsToPythonic} from '../utils/serialization_utils';\nimport * as types_utils from '../utils/types_utils';\nimport {batchSetValue, LayerVariable} from '../variables';\nimport {version as layersVersion} from '../version';\n\nimport {execute, FeedDict} from './executor';\nimport {InputLayer} from './input_layer';\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/** Constructor config for Container. */\nexport interface ContainerArgs {\n  inputs: SymbolicTensor|SymbolicTensor[];\n  outputs: SymbolicTensor|SymbolicTensor[];\n  name?: string;\n}\n\n/**\n * A Container is a directed acyclic graph of layers.\n *\n * It is the topological form of a \"model\". A LayersModel\n * is simply a Container with added training routines.\n *\n */\nexport abstract class Container extends Layer {\n  inputs: SymbolicTensor[];\n  outputs: SymbolicTensor[];\n\n  inputLayers: Layer[];\n  inputLayersNodeIndices: number[];\n  inputLayersTensorIndices: number[];\n\n  outputLayers: Layer[];\n  outputLayersNodeIndices: number[];\n  outputLayersTensorIndices: number[];\n\n  layers: Layer[];\n  layersByDepth: {[depth: string]: Layer[]};\n  nodesByDepth: {[depth: string]: Node[]};\n\n  internalContainerRefs: Container[];\n\n  containerNodes = new Set<string>();\n\n  // TODO(michaelterry): Add cache support\n  // private outputMaskCache: any;\n  // private outputTensorCache: any;\n  // private outputShapeCache: any;\n\n  inputNames: string[];\n  outputNames: string[];\n  feedInputShapes: Shape[];\n\n  protected internalInputShapes: Shape[];\n  protected internalOutputShapes: Shape[];\n  // TODO(cais): Maybe 'feed' should not in the names of these variables,\n  //   due to the fact that our backend is not symbolic.\n  protected feedInputNames: string[];\n  protected feedOutputNames: string[];\n\n  constructor(args: ContainerArgs) {\n    // No args passed to super's constructor.\n    super({});\n    this.name = args.name;\n    if (this.name == null) {\n      const prefix = this.getClassName().toLowerCase();\n      this.name = getUid(prefix);\n    }\n\n    this.supportsMasking = false;\n    this.trainable_ = true;\n\n    // TODO(michaelterry): Initialize perInputLosses/Updates here.\n\n    // Container-specific properties.\n    if (Array.isArray(args.inputs)) {\n      this.inputs = args.inputs.slice();\n    } else {\n      this.inputs = [args.inputs];\n    }\n    if (Array.isArray(args.outputs)) {\n      this.outputs = args.outputs.slice();\n    } else {\n      this.outputs = [args.outputs];\n    }\n\n    // Check for redundancy in inputs.\n    if (generic_utils.unique(this.inputs).length !== this.inputs.length) {\n      throw new ValueError(\n          'The list of inputs passed to the model is ' +\n          'redundant. All inputs should only appear once. Found: ' +\n          `${this.inputs.map(x => x.name)}`);\n    }\n\n    // Check for redundancy in outputs.\n    if (generic_utils.unique(this.outputs).length !== this.outputs.length) {\n      console.warn(\n          'The list of outputs passed to the model is redundant. ' +\n          'All outputs should only appear once. Found: ' +\n          `${this.outputs.map(x => x.name)}`);\n    }\n\n    /*\n      List of initial layers (1 to 1 mapping with this.inputs, hence the same\n      layer might appear twice)\n    */\n    this.inputLayers = [];\n    this.inputLayersNodeIndices = [];\n    this.inputLayersTensorIndices = [];\n    /*\n      List of layers (1 to 1 mapping with this.outputs, hence the same layer\n      might appear twice)\n    */\n    this.outputLayers = [];\n    this.outputLayersNodeIndices = [];\n    this.outputLayersTensorIndices = [];\n    /*\n      All layers in order of horizontal graph traversal. Entries are unique.\n      Includes input and output layers.\n    */\n    this.layers = [];\n\n    /*\n      References to container layers that were constructed internally. We need\n      these to properly dispose of tensors from nested containers.\n    */\n    this.internalContainerRefs = [];\n\n    // TODO(michaelterry): Determine if caching still needed with eager\n    // backend.\n    /*\n      This is for performance optimization when calling the Container on new\n      inputs. Every time the Container is called on a set on input tensors,\n      we compute the output tensors, output masks and output shapes in one pass,\n      then cache them here. When one of these outputs is queried later,\n      we retrieve it from there instead of recomputing it.\n    */\n    // this.outputTensorCache = {};\n    // this.outputShapeCache = {};\n\n    // Build this.outputLayers:\n    for (const x of this.outputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      this.outputLayers.push(layer);\n      this.outputLayersNodeIndices.push(nodeIndex);\n      this.outputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // TODO(michaelterry): Add output mask cache code.\n\n    // Build this.inputLayers:\n    for (const x of this.inputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      /*\n        It's supposed to be an input layer, so only one node\n        and one tensor output.\n      */\n      generic_utils.assert(nodeIndex === 0, 'input layer has >1 nodes');\n      generic_utils.assert(tensorIndex === 0, 'input layer has >1 tensors');\n      this.inputLayers.push(layer);\n      this.inputLayersNodeIndices.push(nodeIndex);\n      this.inputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // Build this.inputNames and this.outputNames.\n    this.inputNames = [];\n    this.outputNames = [];\n    this.feedInputShapes = [];\n    this.feedInputNames = [];\n    this.feedOutputNames = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      // Check that layer is an InputLayer.\n      if (!(layer instanceof InputLayer)) {\n        throw new TypeError(\n            'Input layers to a LayersModel must be InputLayer objects. ' +\n            `Received inputs: ${args.inputs}. ` +\n            `Input ${i} (0-based) originates ` +\n            `from layer type ${layer.getClassName()}.`);\n      }\n      this.inputNames.push(layer.name);\n      this.feedInputShapes.push(layer.batchInputShape);\n\n      this.feedInputNames.push(layer.name);\n    }\n    for (const layer of this.outputLayers) {\n      this.outputNames.push(layer.name);\n    }\n\n    this.internalInputShapes = this.inputs.map(x => x.shape);\n    this.internalOutputShapes = this.outputs.map(x => x.shape);\n\n    /*\n      Container_nodes: set of nodes included in the graph (not all nodes\n      included in the layers are relevant to the current graph).\n    */\n    // ids of all nodes relevant to the Container:\n    const nodesDepths: {[nodeID: string]: number} = {};\n    // To recover nodes from their ID.\n    const nodeIDToNode: {[nodeID: string]: Node} = {};\n    const layersDepths: {[layerID: string]: number} = {};\n    // To layers from their ID.\n    const layerIDToLayer: {[layerID: string]: Layer} = {};\n    const layerIndices: {[layerID: string]: number} = {};\n    const nodesInDecreasingDepth: Node[] = [];\n\n    /**\n     * Builds a map of the graph of layers.\n     *\n     * This recursively updates the map `layerIndices`,\n     * the list `nodesInDecreasingDepth` and the set `containerNodes`.\n     *\n     * @param tensor Some tensor in a graph.\n     * @param finishedNodes Set of nodes whose subgraphs have been traversed\n     *         completely. Useful to prevent duplicated work.\n     * @param nodesInProgress Set of nodes that are currently active on the\n     *         recursion stack. Useful to detect cycles.\n     * @param layer Layer from which `tensor` comes from. If not provided,\n     *   will be obtained from tensor.sourceLayer.\n     * @param nodeIndex Node index from which `tensor` comes from.\n     * @param tensorIndex TensorIndex from which `tensor` comes from.\n     *\n     * @exception RuntimeError if a cycle is detected.\n     */\n    const buildMapOfGraph =\n        (tensor: SymbolicTensor, finishedNodes: Node[], nodesInProgress: Node[],\n         layer?: Layer, nodeIndex?: number, tensorIndex?: number) => {\n          if (layer == null || nodeIndex == null || tensorIndex == null) {\n            layer = tensor.sourceLayer;\n            nodeIndex = tensor.nodeIndex;\n            tensorIndex = tensor.tensorIndex;\n          }\n          const node = layer.inboundNodes[nodeIndex];\n\n          // Prevent cycles.\n          if (nodesInProgress.indexOf(node) !== -1) {\n            throw new RuntimeError(\n                `The tensor ${tensor.name} at layer \"${layer.name}\" ` +\n                'is part of a cycle.');\n          }\n\n          // Don't repeat work for shared subgraphs\n          if (finishedNodes.indexOf(node) !== -1) {\n            return;\n          }\n\n          // Update containerNodes.\n          this.containerNodes.add(Container.nodeKey(layer, nodeIndex));\n\n          // Store the traversal order for layer sorting.\n          if (!(layer.id in layerIndices)) {\n            layerIndices[layer.id] = Object.keys(layerIndices).length;\n          }\n\n          if (nodesInProgress.indexOf(node) === -1) {\n            nodesInProgress.push(node);\n          }\n\n          // Propagate to all previous tensors connected to this node.\n          const numInboundLayers = node.inboundLayers.length;\n          for (let i = 0; i < numInboundLayers; i++) {\n            const x = node.inputTensors[i];\n            const layer = node.inboundLayers[i];\n            const nodeIndex = node.nodeIndices[i];\n            const tensorIndex = node.tensorIndices[i];\n            buildMapOfGraph(\n                x, finishedNodes, nodesInProgress, layer, nodeIndex,\n                tensorIndex);\n          }\n          finishedNodes.push(node);\n          while (nodesInProgress.indexOf(node) >= 0) {\n            nodesInProgress.splice(nodesInProgress.indexOf(node), 1);\n          }\n          nodesInDecreasingDepth.push(node);\n        };\n\n    const finishedNodes: Node[] = [];\n    const nodesInProgress: Node[] = [];\n    for (const x of this.outputs) {\n      buildMapOfGraph(x, finishedNodes, nodesInProgress);\n    }\n\n    const reversedNodesInDecreasingDepth =\n        nodesInDecreasingDepth.slice().reverse();\n    for (const node of reversedNodesInDecreasingDepth) {\n      nodeIDToNode[node.id] = node;\n      // If the depth is not set, the node has no outbound nodes (depth 0).\n      if (!(node.id in nodesDepths)) {\n        nodesDepths[node.id] = 0;\n      }\n      let depth = nodesDepths[node.id];\n\n      // Update the depth of the corresponding layer\n      const previousDepth =\n          (layersDepths[node.outboundLayer.id] == null ?\n               0 :\n               layersDepths[node.outboundLayer.id]);\n\n      /*\n        If we've seen this layer before at a higher depth, we should use that\n        depth instead of the node depth.  This is necessary for shared layers\n        that have inputs at different depth levels in the graph.\n      */\n      depth = Math.max(depth, previousDepth);\n      layersDepths[node.outboundLayer.id] = depth;\n      layerIDToLayer[node.outboundLayer.id] = node.outboundLayer;\n      nodesDepths[node.id] = depth;\n\n      // Update the depth of inbound nodes.\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const inboundLayer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const inboundNode = inboundLayer.inboundNodes[nodeIndex];\n        const previousDepth =\n            (nodesDepths[inboundNode.id] == null ? 0 :\n                                                   nodesDepths[inboundNode.id]);\n        nodesDepths[inboundNode.id] = Math.max(depth + 1, previousDepth);\n        nodeIDToNode[inboundNode.id] = inboundNode;\n      }\n    }\n\n    // Build a dict {depth: list of nodes with this depth}\n    const nodesByDepth: {[depth: string]: Node[]} = {};\n    for (const nodeID in nodesDepths) {\n      const depth = nodesDepths[nodeID];\n      if (!(depth in nodesByDepth)) {\n        nodesByDepth[depth] = [];\n      }\n      nodesByDepth[depth].push(nodeIDToNode[nodeID]);\n    }\n\n    // Build a dict {depth: list of layers with this depth}\n    const layersByDepth: {[depth: string]: Layer[]} = {};\n    for (const layerID in layersDepths) {\n      const depth = layersDepths[layerID];\n      if (!(depth in layersByDepth)) {\n        layersByDepth[depth] = [];\n      }\n      layersByDepth[depth].push(layerIDToLayer[layerID]);\n    }\n\n    // Get sorted list of layer depths.\n    let depthKeys = Object.keys(layersByDepth)\n                        .map(x => parseInt(x, 10))\n                        .sort(generic_utils.reverseNumberCompare);\n\n    // Set this.layers and this.layersByDepth.\n    this.layers = [];\n    for (const depth of depthKeys) {\n      const layersForDepth = layersByDepth[depth];\n      // Container.layers needs to have a deterministic order:\n      // here we order them by traversal order.\n      layersForDepth.sort((a, b) => {\n        const aIndex = layerIndices[a.id];\n        const bIndex = layerIndices[b.id];\n        if (aIndex < bIndex) {\n          return -1;\n        }\n        if (aIndex > bIndex) {\n          return 1;\n        }\n        return 0;\n      });\n      for (const layer of layersForDepth) {\n        if (layer instanceof Container) {\n          this.internalContainerRefs.push(layer);\n        }\n        this.layers.push(layer);\n      }\n    }\n    this.layersByDepth = layersByDepth;\n\n    // Get sorted list of node depths;\n    depthKeys = Object.keys(nodesByDepth)\n                    .map(x => parseInt(x, 10))\n                    .sort(generic_utils.reverseNumberCompare);\n\n    // Check that all tensors required are computable.\n    // computable_tensors: all tensors in the graph\n    // that can be computed from the inputs provided.\n    const computableTensors = this.inputs.slice();\n\n    // To provide a better error msg.\n    const layersWithCompleteInput: string[] = [];\n    for (const depth of depthKeys) {\n      for (const node of nodesByDepth[depth]) {\n        const layer = node.outboundLayer;\n        if (layer != null) {\n          for (const x of node.inputTensors) {\n            if (computableTensors.indexOf(x) === -1) {\n              throw new RuntimeError(\n                  `Graph disconnected: cannot obtain value for tensor ${x}` +\n                  ` at layer \"${layer.name}\". ` +\n                  'The following previous layers were accessed without ' +\n                  `issue: ${layersWithCompleteInput}`);\n            }\n          }\n          for (const x of node.outputTensors) {\n            computableTensors.push(x);\n          }\n          layersWithCompleteInput.push(layer.name);\n        }\n      }\n    }\n\n    // Set this.containerNodes and this.nodesByDepth.\n    this.nodesByDepth = nodesByDepth;\n\n    // Ensure name unicity, which will be crucial for serialization\n    // (since serialized nodes refer to layers by their name).\n    const allNames = this.layers.map(x => x.name);\n    for (const name of allNames) {\n      const numOccurrences = allNames.filter(x => x === name).length;\n      if (numOccurrences !== 1) {\n        throw new RuntimeError(\n            `The name \"${name}\" is used ${numOccurrences} times ` +\n            'in the model. All layer names should be unique. Layer names: ' +\n            JSON.stringify(allNames));\n      }\n    }\n\n    // Layer parameters.\n    // The new container starts with a single inbound node\n    // for its inputs, and no outbound nodes.\n    // Will be appended to by future calls to apply().\n    this.outboundNodes = [];\n    // Will be appended to below, and by future calls to apply().\n    this.inboundNodes = [];\n\n    // Create the node linking internal inputs to internal outputs.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: this.inputs,\n      outputTensors: this.outputs,\n      inputMasks: this.inputs.map(x => null),\n      outputMasks: this.outputs.map(x => null),\n      inputShapes: this.inputs.map(x => x.shape),\n      outputShapes: this.outputs.map(x => x.shape)\n    });\n    this.built = true;\n    this._refCount = 1;  // The ref count of a container always start at 1.\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Container '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose a LayersModel's weights.\n   *\n   * This method decrease the reference count of the LayersModel object by 1.\n   *\n   * A LayersModel is reference-counted. Its reference count is incremented by 1\n   * when it is first constructed and when it is used as a Layer of another\n   * LayersModel.\n   *\n   * If the reference count of a LayersModel becomes 0, the `dispose` method of\n   * all its constituent `Layer`s will be called.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * `dispose` method of its constituent `Layer`s will *not* be called.\n   *\n   * After a LayersModel is disposed, it cannot be used in calls such as\n   * 'predict`, `evaluate` or `fit` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the LayersModel after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the LayersModel has\n   *   already been disposed.\n   */\n  dispose(): DisposeResult {\n    this.assertNotDisposed();\n    const result:\n        DisposeResult = {refCountAfterDispose: null, numDisposedVariables: 0};\n    if (--this._refCount === 0) {\n      for (const layer of this.layers) {\n        result.numDisposedVariables += layer.dispose().numDisposedVariables;\n      }\n\n      // Call dispose on each internally created container layer again to ensure\n      // their refCounts hit zero and their tensors are subsequently deleted.\n      for (const container of this.internalContainerRefs) {\n        result.numDisposedVariables += container.dispose().numDisposedVariables;\n      }\n    }\n    result.refCountAfterDispose = this._refCount;\n    return result;\n  }\n\n  get trainable() {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.layers.forEach(layer => {\n      // tslint:disable-next-line:no-any\n      ((layer as any)._trainableWeights as LayerVariable[])\n          .forEach(w => w.trainable = trainable);\n    });\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    // Porting Note: This check below is to prevent errors where the\n    //   _trainableWeights inherited from the parent class (Layer) gets\n    //   inadvertently used.\n    if (this._trainableWeights.length > 0) {\n      throw new ValueError(\n          'Container instance unexpectedly contains _trainableWeights.' +\n          'The trainable weights of a Container are a union of the ' +\n          'trainable weights of its consituent Layers. Its own ' +\n          '_trainableWeights must remain an empty Array.');\n    }\n\n    if (!this.trainable) {\n      return [];\n    }\n    let weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights = weights.concat(layer.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights.push(...layer.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const layer of this.layers) {\n        trainableWeights.push(...layer.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  /**\n   * Loads all layer weights from a JSON object.\n   *\n   * Porting Note: HDF5 weight files cannot be directly loaded in JavaScript /\n   *   TypeScript. The utility script at `scripts/pykeras.py` offers means\n   *   to convert them into JSON strings compatible with this method.\n   * Porting Note: TensorFlow.js Layers supports only loading by name currently.\n   *\n   * @param weights A JSON mapping weight names to weight values as nested\n   *   arrays of numbers, or a `NamedTensorMap`, i.e., a JSON mapping weight\n   *   names to `tf.Tensor` objects.\n   * @param strict Require that the provided weights exactly match those\n   *   required by the container.  Default: `true`.  Passing `false` means that\n   *   extra weights and missing weights will be silently ignored.\n   */\n  loadWeights(weights: NamedTensorMap, strict = true) {\n    const nameToWeight: {[name: string]: LayerVariable} = {};\n    let totalWeightsCount = 0;\n    for (const layer of this.layers) {\n      for (const weight of layer.weights) {\n        if (nameToWeight[weight.originalName] != null) {\n          throw new ValueError(`Duplicate weight name: ${weight.originalName}`);\n        }\n        nameToWeight[weight.originalName] = weight;\n        totalWeightsCount++;\n      }\n    }\n\n    const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n    for (const name in weights) {\n      // TF 2.2.0 added cell name to the weight name in the format of\n      // layer_name/cell_name/weight_name, we need to remove\n      // the inner cell name.\n      let validatedName = name;\n      if (nameToWeight[name] == null) {\n        const tokens = name.split('/');\n        const shortenNameArray =\n            tokens.slice(0, -2).concat([tokens[tokens.length - 1]]);\n        validatedName = shortenNameArray.join('/');\n      }\n      if (nameToWeight[validatedName] != null) {\n        weightValueTuples.push([nameToWeight[validatedName], weights[name]]);\n      } else if (strict) {\n        throw new ValueError(\n            `Provided weight data has no target variable: ${name}`);\n      }\n      delete nameToWeight[validatedName];\n    }\n\n    if (strict) {\n      // Check that all weights are set.\n      const unsetNames: string[] = [];\n      for (const name in nameToWeight) {\n        unsetNames.push(name);\n      }\n      if (unsetNames.length > 0) {\n        throw new ValueError(\n            `${unsetNames.length} of ${\n                totalWeightsCount} weights are not set: ` +\n            `${unsetNames}`);\n      }\n    }\n\n    batchSetValue(weightValueTuples);\n  }\n\n  /**\n   * Util shared between different serialization methods.\n   * @returns LayersModel config with Keras version information added.\n   */\n  protected updatedConfig(): serialization.ConfigDict {\n    const theConfig = this.getConfig();\n    const modelConfig: serialization.ConfigDict = {};\n    modelConfig['className'] = this.getClassName();\n    modelConfig['config'] = theConfig;\n    modelConfig['kerasVersion'] = `tfjs-layers ${layersVersion}`;\n    // TODO(nielsene): Replace something like K.backend() once\n    // possible.\n    modelConfig['backend'] = 'TensorFlow.js';\n    return modelConfig;\n  }\n\n  /**\n   * Returns a JSON string containing the network configuration.\n   *\n   * To load a network from a JSON save file, use\n   * models.modelFromJSON(jsonString);\n   * @param extraJsonArgs Unused in tfjs-layers, maintained for PyKeras\n   * @param returnString Whether the return value should be stringified\n   *    (default: `true`).\n   * @returns a JSON string if `returnString` (default), or a JSON object if\n   *   `!returnString`.\n   */\n  // tslint:disable-next-line:no-any\n  toJSON(unused?: any, returnString = true): string|PyJsonDict {\n    const modelConfig = convertTsToPythonic(this.updatedConfig()) as PyJsonDict;\n    return returnString ? JSON.stringify(modelConfig) : modelConfig;\n  }\n\n  /**\n   * Call the model on new inputs.\n   *\n   * In this case `call` just reapplies all ops in the graph to the new inputs\n   * (e.g. build a new computational graph from the provided inputs).\n   *\n   * @param inputs A tensor or list of tensors.\n   * @param mask A mask or list of masks. A mask can be either a tensor or null\n   *   (no mask).\n   *\n   * @return A tensor if there is a single output, or a list of tensors if there\n   *   are more than one outputs.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      const feedDict = new FeedDict();\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n      return execute(this.outputs, feedDict, kwargs) as Tensor | Tensor[];\n    });\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      let masks: Tensor[];\n      if (mask == null) {\n        masks = generic_utils.pyListRepeat(null, inputs.length);\n      } else {\n        masks = generic_utils.toList(mask);\n      }\n      // TODO(michaelterry): Add support for mask caching.\n      return this.runInternalGraph(inputs, masks)[1];\n    });\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    const inputShapes = types_utils.normalizeShapeList(inputShape);\n    if (inputShapes.length !== this.inputLayers.length) {\n      throw new ValueError(\n          `Invalid inputShape argument ${inputShape}: ` +\n          `model has ${this.inputLayers.length} tensor inputs.`);\n    }\n\n    // TODO(michaelterry): Add caching\n    const layersToOutputShapes: {[shapeKey: string]: Shape} = {};\n    for (let i = 0; i < inputShapes.length; i++) {\n      const layer = this.inputLayers[i];\n      const inputShape = inputShapes[i];\n      // It's an input layer: computeOutputShape is identity,\n      // and there is only one node and one tensor output.\n      const shapeKey = layer.name + '_0_0';\n      layersToOutputShapes[shapeKey] = inputShape;\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    // Iterate over nodes, by depth level.\n    if (depthKeys.length > 1) {\n      for (const depth of depthKeys) {\n        const nodes = this.nodesByDepth[depth];\n        for (const node of nodes) {\n          // This is always a single layer, never a list.\n          const layer = node.outboundLayer;\n          if (this.inputLayers.map(x => x.id).indexOf(layer.id) !== -1) {\n            // We've already covered the input layers a few lines above.\n            continue;\n          }\n          // Potentially redundant list, same size of node.inputTensors.\n          const inputShapes: Shape[] = [];\n          for (let j = 0; j < node.inboundLayers.length; j++) {\n            const inboundLayer = node.inboundLayers[j];\n            const nodeIndex = node.nodeIndices[j];\n            const tensorIndex = node.tensorIndices[j];\n            const shapeKey = `${inboundLayer.name}_${nodeIndex}_${tensorIndex}`;\n            const inputShape = layersToOutputShapes[shapeKey];\n            inputShapes.push(inputShape);\n          }\n\n          const outputShape = layer.computeOutputShape(\n              generic_utils.singletonOrArray(inputShapes));\n\n          const outputShapes = types_utils.normalizeShapeList(outputShape);\n          const nodeIndex = layer.inboundNodes.indexOf(node);\n          for (let j = 0; j < outputShapes.length; j++) {\n            const shapeKey = `${layer.name}_${nodeIndex}_${j}`;\n            layersToOutputShapes[shapeKey] = outputShapes[j];\n          }\n        }\n      }\n    }\n\n    // Read final output shapes from layersToOutputShapes.\n    const outputShapes: Shape[] = [];\n    const outputShapeKeys: string[] = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      const shapeKey = `${layer.name}_${nodeIndex}_${tensorIndex}`;\n      outputShapeKeys.push(shapeKey);\n    }\n\n    for (let i = 0; i < outputShapeKeys.length; i++) {\n      const key = outputShapeKeys[i];\n      generic_utils.assert(key in layersToOutputShapes);\n      outputShapes.push(layersToOutputShapes[key]);\n    }\n\n    // TODO(michaelterry): Update cache\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  /**\n   * Computes output tensors for new inputs.\n   *\n   * Note:\n   *   - Expects `inputs` to be a list (potentially with 1 element).\n   *\n   * @param inputs List of tensors\n   * @param masks List of masks (tensors or null).\n   * @return Three lists: outputTensors, outputMasks, outputShapes\n   */\n  protected runInternalGraph(inputs: Tensor[], masks?: Tensor[]):\n      [Tensor[], Tensor[], Shape[]] {\n    if (masks == null) {\n      masks = generic_utils.pyListRepeat(null, inputs.length);\n    }\n\n    // Dictionary mapping reference tensors to tuples\n    // (computed tensor, compute mask)\n    // we assume a 1:1 mapping from tensor to mask\n    // TODO: raise exception when a `.computeMask()` call\n    // does not return a list the same size as `call`\n    const tensorMap: {[tensorID: string]: [Tensor, Tensor]} = {};\n    for (let i = 0; i < this.inputs.length; ++i) {\n      const x = this.inputs[i];\n      const y = inputs[i];\n      const mask = masks[i];\n      tensorMap[x.id] = [y, mask];\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    for (const depth of depthKeys) {\n      const nodes = this.nodesByDepth[depth];\n      for (const node of nodes) {\n        // This is always a single layer, never a list.\n        const layer = node.outboundLayer;\n        const referenceInputTensors = node.inputTensors;\n        const referenceOutputTensors = node.outputTensors;\n\n        // If all previous input tensors are available in tensorMap,\n        // then call node.inboundLayer on them.\n        // List of tuples [input, mask]:\n        const computedData = new Array<[Tensor, Tensor]>();\n        for (const x of referenceInputTensors) {\n          if (x.id in tensorMap) {\n            computedData.push(tensorMap[x.id]);\n          }\n        }\n        if (computedData.length === referenceInputTensors.length) {\n          // TODO(michaelterry): Add K.name_scope here, if we need it.\n          let kwargs: Kwargs = {};\n          let computedTensors: Tensor[];\n          let computedMasks: Tensor[];\n          let outputTensors: Tensor[];\n          let outputMasks: Tensor[];\n          // call layer\n          if (node.callArgs != null) {\n            kwargs = node.callArgs;\n          }\n          if (computedData.length === 1) {\n            const [computedTensor, computedMask] = computedData[0];\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMask;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensor, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensor, computedMask));\n            computedTensors = [computedTensor];\n            computedMasks = [computedMask];\n          } else {\n            computedTensors = computedData.map(x => x[0]);\n            computedMasks = computedData.map(x => x[1]);\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMasks;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensors, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensors, computedMasks));\n          }\n\n          if (layer.activityRegularizer) {\n            throw new NotImplementedError(\n                'LayersModel invocation with concrete Tensor value(s) in the ' +\n                'presence of activity regularizer(s) is not supported yet.');\n          }\n          // TODO(michaelterry): Add model updates and losses\n\n          // Update tensor map.\n          for (let i = 0; i < referenceOutputTensors.length; ++i) {\n            const x = referenceOutputTensors[i];\n            const y = outputTensors[i];\n            const mask = outputMasks[i];\n            tensorMap[x.id] = [y, mask];\n          }\n        }\n      }\n    }\n\n    const outputTensors: Tensor[] = [];\n    const outputMasks: Tensor[] = [];\n    const outputShapes: Shape[] = [];\n    for (const x of this.outputs) {\n      generic_utils.assert(\n          x.id in tensorMap, `Could not compute output ${x.name} : ${x.id}`);\n      const [tensor, mask] = tensorMap[x.id];\n      outputShapes.push(tensor.shape);\n      outputTensors.push(tensor);\n      outputMasks.push(mask);\n    }\n\n    // TODO(michaelterry): Add support for caches.\n    return [outputTensors, outputMasks, outputShapes];\n  }\n\n  /**\n   * Builds a map of internal node keys to node ordering.\n   * Used in serializaion a node orderings may change as unused nodes are\n   * dropped. Porting Note:  This helper method was pulled out of getConfig to\n   * improve readability.\n   * @param layers An array of Layers in the model.\n   * @returns Map of Node Keys to index order within the layer.\n   */\n  private buildNodeConversionMap(layers: Layer[]): {[nodeKey: string]: number} {\n    const nodeConversionMap: {[nodeKey: string]: number} = {};\n    let keptNodes: number;\n    for (const layer of this.layers) {\n      keptNodes = layer instanceof Container ? 1 : 0;\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        if (this.containerNodes.has(nodeKey)) {\n          // i.e. we mark it to be saved\n          nodeConversionMap[nodeKey] = keptNodes;\n          keptNodes += 1;\n        }\n      }\n    }\n    return nodeConversionMap;\n  }\n\n  /**\n   * Retrieves a layer based on either its name (unique) or index.\n   *\n   * Indices are based on order of horizontal graph traversal (bottom-up).\n   *\n   * If both `name` and `index` are specified, `index` takes precedence.\n   *\n   * @param name Name of layer.\n   * @param index Index of layer.\n   * @returns A Layer instance.\n   * @throws ValueError: In case of invalid layer name or index.\n   *\n   * @doc {\n   *    heading: 'Layers',\n   *    subheading: 'Classes',\n   *    namespace: 'layers',\n   *    subclasses: ['LayersModel']\n   * }\n   */\n  getLayer(name?: string, index?: number): Layer {\n    if (index != null) {\n      if (this.layers.length <= index) {\n        throw new ValueError(\n            `Was asked to retrieve layer at index ${index}, but model only ` +\n            `has ${this.layers.length} layer(s).`);\n      } else {\n        return this.layers[index];\n      }\n    } else {\n      if (name == null) {\n        throw new ValueError('Provide either a layer name or layer index');\n      }\n    }\n\n    for (const layer of this.layers) {\n      if (layer.name === name) {\n        return layer;\n      }\n    }\n    throw new ValueError(`No such layer: ${name}`);\n  }\n\n  /**\n   * Retrieves the Container's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Container.loss in PyKeras.\n    //   In PyKeras, Container.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return tidy(() => {\n      const losses: Scalar[] = [];\n      for (const layer of this.layers) {\n        for (let nodeIndex = 0; nodeIndex < layer.inboundNodes.length;\n             ++nodeIndex) {\n          const nodeKey = Container.nodeKey(layer, nodeIndex);\n          if (this.containerNodes.has(nodeKey)) {\n            losses.push(...layer.calculateLosses());\n          }\n        }\n      }\n      // TODO(cais): Add any unconditional model-level losses?\n      return losses;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {name: this.name};\n\n    // Build a map from layer unique name (self._node_key)\n    // to the index of the nodes that are saved in the config.\n    // Only nodes in container_nodes are saved.\n    const nodeConversionMap: {[nodeKey: string]: number} =\n        this.buildNodeConversionMap(this.layers);\n\n    // Serialize and save the layers in layerConfigs\n    const layerConfigs = [];\n    for (const layer of this.layers) {\n      const layerClassName = layer.getClassName();\n      const layerConfig = layer.getConfig();\n      const filteredInboundNodes = [];\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const node = layer.inboundNodes[originalNodeIndex];\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        let kwargs = {};\n        if (this.containerNodes.has(nodeKey)) {\n          // The node is relevant to the model:\n          // add to filteredInboundNodes.\n          if (node.callArgs) {\n            try {\n              JSON.stringify(node.callArgs);\n              kwargs = node.callArgs;\n            } catch (err) {\n              console.warn(\n                  `Layer ${layer.name} was passed ` +\n                  `non-serializable keyword arguments: ` +\n                  `${node.callArgs}. They will not be included ` +\n                  `in the serialized model (and thus will be ` +\n                  `missing at deserialization time).`);\n              kwargs = {};\n            }\n          }\n          if (node.inboundLayers.length > 0) {\n            const nodeData = [];\n            for (let i = 0; i < node.inboundLayers.length; i++) {\n              const inboundLayer = node.inboundLayers[i];\n              const nodeIndex = node.nodeIndices[i];\n              const tensorIndex = node.tensorIndices[i];\n              const nodeKey = Container.nodeKey(inboundLayer, nodeIndex);\n              let newNodeIndex = nodeConversionMap[nodeKey];\n              if (newNodeIndex == null) {\n                newNodeIndex = 0;\n              }\n              nodeData.push(\n                  [inboundLayer.name, newNodeIndex, tensorIndex, kwargs]);\n            }\n            filteredInboundNodes.push(nodeData);\n          }\n        }\n      }\n      const dict: serialization.ConfigDict = {};\n      dict['name'] = layer.name;\n      dict['className'] = layerClassName;\n      dict['config'] = layerConfig;\n      dict['inboundNodes'] = filteredInboundNodes;\n      layerConfigs.push(dict);\n    }\n    config['layers'] = layerConfigs;\n    // Gather info about inputs and outputs\n    const modelInputs = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      const nodeIndex = this.inputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.inputLayersTensorIndices[i];\n      modelInputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['inputLayers'] = modelInputs;\n\n    const modelOutputs = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      modelOutputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['outputLayers'] = modelOutputs;\n    return config;\n  }\n\n  /**\n   * Instantiates a LayersModel from its config (output of `get_config()`).\n   * @param cls the class to create\n   * @param config LayersModel config dictionary.\n   * @param customObjects An optional dictionary of custom objects.\n   * @param fastWeightInit Optional flag to use fast weight initialization\n   *   during deserialization. This is applicable to cases in which\n   *   the initialization will be immediately overwritten by loaded weight\n   *   values. Default: `false`.\n   * @returns A LayersModel instance.\n   * @throws ValueError: In case of improperly formatted config dict.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    // Layer instances created during\n    // the graph reconstruction process\n    const createdLayers: {[layerName: string]: Layer} = {};\n\n    // Dictionary mapping layer instances to\n    // node data that specifies a layer call.\n    // It acts as a queue that maintains any unprocessed\n    // layer call until it becomes possible to process it\n    // (i.e. until the input tensors to the call all exist).\n    const unprocessedNodes: {[layer: string]: TensorKeyWithArgsArray[][]} = {};\n    function addUnprocessedNode(\n        layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      if (!(layer.name in unprocessedNodes)) {\n        unprocessedNodes[layer.name] = [nodeData];\n      } else {\n        unprocessedNodes[layer.name].push(nodeData);\n      }\n    }\n\n    function processNode(layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      const inputTensors: SymbolicTensor[] = [];\n      let kwargs;\n      for (const inputData of nodeData) {\n        const inboundLayerName = inputData[0];\n        const inboundNodeIndex = inputData[1];\n        const inboundTensorIndex = inputData[2];\n\n        kwargs = inputData[3] == null ?\n            {} :\n            inputData[3] as serialization.ConfigDict;\n        if (!(inboundLayerName in createdLayers)) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundLayer = createdLayers[inboundLayerName];\n        if (inboundLayer.inboundNodes.length <= inboundNodeIndex) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundNode = inboundLayer.inboundNodes[inboundNodeIndex];\n        inputTensors.push(inboundNode.outputTensors[inboundTensorIndex]);\n      }\n      // Call layer on its inputs, thus creating the node\n      // and building the layer if needed.\n      // Note: This has Eager vs Graph Implications.\n      if (inputTensors.length > 0) {\n        layer.apply(\n            generic_utils.singletonOrArray(inputTensors),\n            kwargs);  // was ** kwargs\n      }\n    }\n\n    /**\n     * Deserialize a layer, then call it on appropriate inputs.\n     * @param layerData: layer config dict.\n     * @throws ValueError: In case of improperly formatted `layer_data`\n     * dict.\n     */\n    function processLayer(layerData: serialization.ConfigDict|null) {\n      const layerName = layerData['name'] as string;\n      // Instantiate layer.\n      const layer =\n          deserializeLayer(\n              layerData,\n              config['customObjects'] != null ?\n                  config['customObjects'] as serialization.ConfigDict :\n                  {}) as Layer;\n      layer.setFastWeightInitDuringBuild(fastWeightInit);\n      createdLayers[layerName] = layer;\n      // Gather layer inputs.\n      const inboundNodesData =\n          layerData['inboundNodes'] as TensorKeyWithArgsArray[][];\n      inboundNodesData.forEach(nodeData => {\n        if (!(nodeData instanceof Array)) {\n          throw new ValueError(\n              `Corrupted configuration, expected array for nodeData: ${\n                  nodeData}`);\n        }\n        // We don't process nodes (i.e. make layer calls)\n        // on the fly because the inbound node may not yet exist,\n        // in case of layer shared at different topological depths\n        // (e.g.a model such as A(B(A(B(x)))))\n        addUnprocessedNode(layer, nodeData);\n      });\n    }\n\n    // First, we create all layers and enqueue nodes to be processed.\n    const name = config['name'];\n    const layersFromConfig = config['layers'] as serialization.ConfigDict[];\n    for (const layerData of layersFromConfig) {\n      processLayer(layerData);\n    }\n\n    // Then we process nodes in order of layer depth.\n    // Nodes that cannot yet be processed(if the inbound node\n    // does not yet exist) are re - enqueued, and the process\n    // is repeated until all nodes are processed.\n    while (!generic_utils.isObjectEmpty(unprocessedNodes)) {\n      for (const layerData of layersFromConfig) {\n        const layer = createdLayers[layerData['name'] as string];\n        if (layer.name in unprocessedNodes) {\n          const currentUnprocessedNodesForLayer = unprocessedNodes[layer.name];\n          delete unprocessedNodes[layer.name];\n          for (const nodeData of currentUnprocessedNodesForLayer) {\n            processNode(layer, nodeData);\n          }\n        }\n      }\n    }\n\n    const inputTensors: SymbolicTensor[] = [];\n    const outputTensors: SymbolicTensor[] = [];\n    const inputLayersFromConfig =\n        config['inputLayers'] as serialization.ConfigDict[];\n    for (const layerData of inputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      inputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    const outputLayersFromConfig =\n        config['outputLayers'] as serialization.ConfigDict[];\n    for (const layerData of outputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      outputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    return new cls({inputs: inputTensors, outputs: outputTensors, name});\n  }\n\n  /**\n   * Determine whether the container is stateful.\n   *\n   * Porting Note: this is the equivalent of the stateful @property of\n   *   the Container class in PyKeras.\n   */\n  get stateful(): boolean {\n    // Porting Note: This check is to prevent inadvertent setting of the\n    //   _stateful property of the Container instance.\n    if (this._stateful) {\n      throw new ValueError(\n          'Container instance unexpectedly has _stateful = true. The ' +\n          'statefulness of a Container is determined by the Layers it ' +\n          'contains. Its _stateful property must remain the default false.');\n    }\n    for (const layer of this.layers) {\n      if (layer.stateful) {\n        return true;\n      }\n    }\n    return false;\n  }\n\n  /**\n   * Reset the state of all stateful constituent layers (if any).\n   *\n   * Examples of stateful layers include RNN layers whose `stateful` property\n   * is set as `true`.\n   */\n  resetStates() {\n    tidy(() => {\n      this.layers.forEach(layer => {\n        // tslint:disable:no-any\n        if (layer.stateful) {\n          layer.resetStates();\n        }\n        // tslint:enable:no-any\n      });\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, mul, Tensor, Tensor1D, tensor1d, tidy} from '@tensorflow/tfjs-core';\n\n/**\n * For multi-class classification problems, this object is designed to store a\n * mapping from class index to the \"weight\" of the class, where higher weighted\n * classes have larger impact on loss, accuracy, and other metrics.\n *\n * This is useful for cases in which you want the model to \"pay more attention\"\n * to examples from an under-represented class, e.g., in unbalanced datasets.\n */\nexport type ClassWeight = {\n  [classIndex: number]: number\n};\n\n/**\n * Class weighting for a model with multiple outputs.\n *\n * This object maps each output name to a class-weighting object.\n */\nexport type ClassWeightMap = {\n  [outputName: string]: ClassWeight\n};\n\nfunction standardizeSampleOrClassWeights(\n    xWeight: ClassWeight|ClassWeight[]|ClassWeightMap, outputNames: string[],\n    weightType: 'sampleWeight'|'classWeight'): ClassWeight[] {\n  const numOutputs = outputNames.length;\n  if (xWeight == null || (Array.isArray(xWeight) && xWeight.length === 0)) {\n    return outputNames.map(name => null);\n  }\n  if (numOutputs === 1) {\n    if (Array.isArray(xWeight) && xWeight.length === 1) {\n      return xWeight;\n    } else if (typeof xWeight === 'object' && outputNames[0] in xWeight) {\n      return [(xWeight as ClassWeightMap)[outputNames[0]]];\n    } else {\n      return [xWeight as ClassWeight];\n    }\n  }\n  if (Array.isArray(xWeight)) {\n    if (xWeight.length !== numOutputs) {\n      throw new Error(\n          `Provided ${weightType} is an array of ${xWeight.length} ` +\n          `element(s), but the model has ${numOutputs} outputs. ` +\n          `Make sure a set of weights is provided for each model output.`);\n    }\n    return xWeight;\n  } else if (\n      typeof xWeight === 'object' && Object.keys(xWeight).length > 0 &&\n      typeof (xWeight as ClassWeightMap)[Object.keys(xWeight)[0]] ===\n          'object') {\n    const output: ClassWeight[] = [];\n    outputNames.forEach(outputName => {\n      if (outputName in xWeight) {\n        output.push((xWeight as ClassWeightMap)[outputName]);\n      } else {\n        output.push(null);\n      }\n    });\n    return output;\n  } else {\n    throw new Error(\n        `The model has multiple (${numOutputs}) outputs, ` +\n        `so ${weightType} must be either an array with ` +\n        `${numOutputs} elements or an object with ${outputNames} keys. ` +\n        `Provided ${weightType} not understood: ${JSON.stringify(xWeight)}`);\n  }\n}\n\n/**\n * Standardize class weighting objects.\n *\n * This function takes a single class-weighting object, an array of them,\n * or a map from output name to class-weighting object. It compares it to the\n * output name(s) of the model, base on which it outputs an array of\n * class-weighting objects of which the length matches the number of outputs.\n *\n * @param classWeight Input class-weighting object(s).\n * @param outputNames All output name(s) of the model.\n * @return An array of class-weighting objects. The length of the array matches\n *   the model's number of outputs.\n */\nexport function standardizeClassWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'classWeight');\n}\n\nexport function standardizeSampleWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'sampleWeight');\n}\n\n/**\n * Standardize by-sample and/or by-class weights for training.\n *\n * Note that this function operates on one model output at a time. For a model\n * with multiple outputs, you must call this function multiple times.\n *\n * @param y The target tensor that the by-sample and/or by-class weight is for.\n *     The values of y are assumed to encode the classes, either directly\n *     as an integer index, or as one-hot encoding.\n * @param sampleWeight By-sample weights.\n * @param classWeight By-class weights: an object mapping class indices\n *     (integers) to a weight (float) to apply to the model's loss for the\n *     samples from this class during training. This can be useful to tell the\n *     model to \"pay more attention\" to samples from an under-represented class.\n * @param sampleWeightMode The mode for the sample weights.\n * @return A Promise of weight tensor, of which the size of the first dimension\n *     matches that of `y`.\n */\nexport async function standardizeWeights(\n    y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight,\n    sampleWeightMode?: 'temporal'): Promise<Tensor> {\n  if (sampleWeight != null || sampleWeightMode != null) {\n    // TODO(cais): Once 'temporal' mode is implemented, document it in the doc\n    // string.\n    throw new Error('Support sampleWeight is not implemented yet');\n  }\n\n  if (classWeight != null) {\n    // Apply class weights per sample.\n    const yClasses: Tensor1D = tidy(() => {\n      if (y.shape.length === 1) {\n        // Assume class indices.\n        return y.clone() as Tensor1D;\n      } else if (y.shape.length === 2) {\n        if (y.shape[1] > 1) {\n          // Assume one-hot encoding of classes.\n          const axis = 1;\n          return y.argMax(axis);\n        } else if (y.shape[1] === 1) {\n          // Class index.\n          return y.reshape([y.shape[0]]);\n        } else {\n          throw new Error(\n              `Encountered unexpected last-dimension size (${y.shape[1]}) ` +\n              `during handling of class weights. The size is expected to be ` +\n              `>= 1.`);\n        }\n      } else {\n        throw new Error(\n            `Unexpected rank of target (y) tensor (${y.rank}) during ` +\n            `handling of class weights. The rank is expected to be 1 or 2.`);\n      }\n    });\n\n    const yClassIndices = Array.from(await yClasses.data());\n    dispose(yClasses);\n    const classSampleWeight: number[] = [];\n    yClassIndices.forEach(classIndex => {\n      if (classWeight[classIndex] == null) {\n        throw new Error(\n            `classWeight must contain all classes in the training data. ` +\n            `The class ${classIndex} exists in the data but not in ` +\n            `classWeight`);\n      } else {\n        classSampleWeight.push(classWeight[classIndex]);\n      }\n    });\n\n    return tensor1d(classSampleWeight, 'float32');\n  } else {\n    return null;\n  }\n}\n\n/**\n * Apply per-sample weights on the loss values from a number of samples.\n *\n * @param losses Loss tensor of shape `[batchSize]`.\n * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.\n * @returns Tensor of the same shape as`losses`.\n */\nexport function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor) {\n  return mul(losses, sampleWeights);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using TensorFlow.js datasets.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar} from '@tensorflow/tfjs-core';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {TensorOrArrayOrMap} from '../types';\nimport {singletonOrArray, toList} from '../utils/generic_utils';\n\nimport {Dataset, LazyIterator} from './dataset_stub';\nimport {ClassWeight, ClassWeightMap, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Interface for configuring model training based on a dataset object.\n */\nexport interface ModelFitDatasetArgs<T> {\n  /**\n   * (Optional) Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. It should\n   * typically be equal to the number of samples of your dataset divided by\n   * the batch size, so that `fitDataset`() call can utilize the entire dataset.\n   * If it is not provided, use `done` return value in `iterator.next()` as\n   * signal to finish an epoch.\n   */\n  batchesPerEpoch?: number;\n\n  /**\n   * Integer number of times to iterate over the training dataset.\n   */\n  epochs: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be any of the following:\n   *\n   *   - An array `[xVal, yVal]`, where the two values may be `tf.Tensor`,\n   *     an array of Tensors, or a map of string to Tensor.\n   *   - Similarly, an array ` [xVal, yVal, valSampleWeights]`\n   *     (not implemented yet).\n   *   - a `Dataset` object with elements of the form `{xs: xVal, ys: yVal}`,\n   *     where `xs` and `ys` are the feature and label tensors, respectively.\n   *\n   * If `validationData` is an Array of Tensor objects, each `tf.Tensor` will be\n   * sliced into batches during validation, using the parameter\n   * `validationBatchSize` (which defaults to 32). The entirety of the\n   * `tf.Tensor` objects will be used in the validation.\n   *\n   * If `validationData` is a dataset object, and the `validationBatches`\n   * parameter is specified, the validation will use `validationBatches` batches\n   * drawn from the dataset object. If `validationBatches` parameter is not\n   * specified, the validation will stop when the dataset is exhausted.\n   *\n   * The model will not be trained on this data.\n   */\n  validationData?: [\n    TensorOrArrayOrMap, TensorOrArrayOrMap\n  ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|Dataset<T>;\n\n  /**\n   * Optional batch size for validation.\n   *\n   * Used only if `validationData` is an array of `tf.Tensor` objects, i.e., not\n   * a dataset object.\n   *\n   * If not specified, its value defaults to 32.\n   */\n  validationBatchSize?: number;\n\n  /**\n   * (Optional) Only relevant if `validationData` is specified and is a dataset\n   * object.\n   *\n   * Total number of batches of samples to draw from `validationData` for\n   * validation purpose before stopping at the end of every epoch. If not\n   * specified, `evaluateDataset` will use `iterator.next().done` as signal to\n   * stop validation.\n   */\n  validationBatches?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - a `number`: Will yield every `number` milliseconds.\n   *   - `'never'`: never yield. (But yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n}\n\nexport interface FitDatasetElement {\n  xs: TensorOrArrayOrMap;\n  ys: TensorOrArrayOrMap;\n}\n\n/**\n * Interface for configuring model evaluation based on a dataset object.\n */\nexport interface ModelEvaluateDatasetArgs {\n  /**\n   * Number of batches to draw from the dataset object before ending the\n   * evaluation.\n   */\n  batches?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n}\n\n// Default batch size used during tensor-based validation.\nconst DEFAULT_VALIDATION_BATCH_SIZE = 32;\n\n/**\n * Standardize the output of a dataset iterator for use by\n * LayersModel.fitDataset().\n *\n * @param model: A `tf.LayersModel` object.\n * @param iteratorOut The output of a dataset iterator. It is required to be\n *   an object of the form `{xs: TensorOrArrayOrMap, ys:\n * TensorOrArrayOrMap}`, where `TensorOrArrayOrMap` is a single `tf.Tensor`,\n * a `tf.Tensor[]`, or a flat map from string names to `tf.Tensor`s.\n * @returns A flat array of `tf.Tensor` objects: the input `tf.Tensor`s\n *   followed by the target `tf.Tensor`s.  When `tf.Tensor`s are provided\n *   as a map, the order in the resulting array is taken from the `inputNames`\n *   and `outputNames` of the model.\n */\nfunction standardizeDataIteratorOutput(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, iteratorOut: {}): {xs: tfc.Tensor[], ys: tfc.Tensor[]} {\n  let xs: TensorOrArrayOrMap;\n  let ys: TensorOrArrayOrMap;\n\n  const iteratorOutObj = iteratorOut as FitDatasetElement;\n  xs = iteratorOutObj['xs'];\n  ys = iteratorOutObj['ys'];\n  tfc.util.assert(\n      xs != null && ys != null,\n      () => 'A Dataset iterator for fitDataset() is expected to generate ' +\n          'objects of the form `{xs: xVal, ys: yVal}`, where the two ' +\n          'values may be `tf.Tensor`, an array of Tensors, or a map of ' +\n          'string to Tensor.  The provided Dataset instead generates ' +\n          `${iteratorOut}`);\n\n  const flattenedXs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('input', model.inputNames, xs);\n  const flattenedYs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('output', model.outputNames, ys);\n\n  const batchSize: number = flattenedXs[0].shape[0];\n\n  tfc.util.assert(\n      flattenedXs.length === model.inputs.length,\n      () => `LayersModel has ${model.inputs.length} inputs, but the dataset ` +\n          `provides ${flattenedXs.length} inputs.  (Expected input keys: ` +\n          `${JSON.stringify(model.inputNames)})`);\n\n  tfc.util.assert(\n      flattenedYs.length === model.outputs.length,\n      () =>\n          `LayersModel has ${model.outputs.length} outputs, but the dataset ` +\n          `provides ${flattenedYs.length} outputs.  (Expected output keys: ` +\n          `${JSON.stringify(model.outputNames)})`);\n\n  for (let xIndex = 0; xIndex < flattenedXs.length; xIndex++) {\n    tfc.util.assert(\n        flattenedXs[xIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: input ` +\n            `${model.inputNames[xIndex]} has ${\n                  flattenedXs[xIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  for (let yIndex = 0; yIndex < flattenedYs.length; yIndex++) {\n    tfc.util.assert(\n        flattenedYs[yIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: output ` +\n            `${model.outputNames[yIndex]} has ${\n                  flattenedYs[yIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  return {xs: flattenedXs, ys: flattenedYs};\n}\n\nfunction flattenTensorOrArrayOrMap(\n    inputOrOutput: string, names: string[], values: TensorOrArrayOrMap) {\n  if (values instanceof tfc.Tensor) {\n    return [values];\n  } else if (Array.isArray(values)) {\n    tfc.util.assert(\n        values.length === names.length,\n        () => `Received an array of ${values.length} Tensors, but expected ${\n            names.length} to match the ${inputOrOutput} keys ${names}.`);\n    return values;\n  } else {\n    const result: tfc.Tensor[] = [];\n    // Check that all the required keys are available.\n    for (const name of names) {\n      if (values[name] == null) {\n        throw new ValueError(\n            `The feature data generated by the dataset lacks the required ` +\n            `${inputOrOutput} key '${name}'.`);\n      }\n      result.push(values[name]);\n    }\n    return result;\n  }\n}\n\nfunction standardizeTensorValidationData<T>(\n    data:\n        [\n          tfc.Tensor|tfc.Tensor[], tfc.Tensor|tfc.Tensor[]\n        ]|[tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n           tfc.Tensor | tfc.Tensor[]]):\n    {xs: tfc.Tensor|tfc.Tensor[], ys: tfc.Tensor|tfc.Tensor[]} {\n  if (data.length === 3) {\n    throw new NotImplementedError(\n        'Validation with sample weights is not implemented yet.');\n  }\n  return {xs: data[0], ys: data[1]};\n}\n\nexport async function fitDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>,\n    args: ModelFitDatasetArgs<T>): Promise<History> {\n  const hasBatchesPerEpoch = args.batchesPerEpoch != null;\n  tfc.util.assert(\n      model.optimizer != null,\n      () => 'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileConfig).');\n\n  tfc.util.assert(\n      args != null,\n      () => `For fitDataset(), the 2nd argument (config) is required, ` +\n          `but it is not provided in this call.`);\n  tfc.util.assert(\n      args.epochs != null && args.epochs > 0 && Number.isInteger(args.epochs),\n      () => `For fitDataset(), config.epochs is expected to be a positive ` +\n          `integer, but got ${args.epochs}`);\n  tfc.util.assert(\n      !hasBatchesPerEpoch ||\n          (args.batchesPerEpoch > 0 && Number.isInteger(args.batchesPerEpoch)),\n      () => `For fitDataset(), config.batchesPerEpoch is expected to be a ` +\n          `positive integer if specified, but got ${args.batchesPerEpoch}`);\n  tfc.util.assert(\n      // tslint:disable-next-line:no-any\n      (args as any)['validationSplit'] == null,\n      () => '`validationSplit` is not supported by `fitDataset()`. ' +\n          'Use validationData instead.');\n\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n\n  try {\n    const doValidation = args.validationData != null;\n    let valXs: tfc.Tensor|tfc.Tensor[];\n    let valYs: tfc.Tensor|tfc.Tensor[];\n    if (doValidation) {\n      if (isDatasetObject(args.validationData)) {\n        tfc.util.assert(\n            args.validationBatches == null ||\n                (args.validationBatches > 0 &&\n                 Number.isInteger(args.validationBatches)),\n            () => `For fitDataset() with dataset-based validation, ` +\n                `config.validationBatches is expected not to be provided, ` +\n                `or to be a positive integer, ` +\n                `but got ${args.validationBatches}`);\n      } else {\n        const validationData = standardizeTensorValidationData(\n            args.validationData as\n                    [tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[]] |\n            [\n              tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n              tfc.Tensor | tfc.Tensor[]\n            ]);\n        valXs = validationData.xs;\n        valYs = validationData.ys;\n      }\n    }\n\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let callbackMetrics: string[];\n    if (doValidation) {\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const verbose = args.verbose == null ? 1 : args.verbose;\n    const {callbackList, history} = configureCallbacks(\n        callbacks, verbose, args.epochs, null, null,\n        getStepsPerEpoch(dataset, args),\n        null,  // Batch size determined by the dataset itself.\n        doValidation, callbackMetrics);\n    callbackList.setModel(model);\n    model.history = history;\n\n    await callbackList.onTrainBegin();\n    model.stopTraining_ = false;\n    let epoch = args.initialEpoch == null ? 0 : args.initialEpoch;\n\n    let dataIterator = await dataset.iterator();\n    while (epoch < args.epochs) {\n      const epochLogs: UnresolvedLogs = {};\n      await callbackList.onEpochBegin(epoch);\n      let stepsDone = 0;\n      let batchIndex = 0;\n      if (!hasBatchesPerEpoch) {\n        dataIterator = await dataset.iterator();\n      }\n      while (hasBatchesPerEpoch ? stepsDone < args.batchesPerEpoch : true) {\n        const iteratorOut = await dataIterator.next();\n\n        // If `batchesPerEpoch` is specified, the dataset should not be\n        // exhausted until all epoches are done.\n        if (hasBatchesPerEpoch && iteratorOut.done) {\n          console.warn(\n              'You provided `batchesPerEpoch` as ' +\n              `${args.batchesPerEpoch}, ` +\n              'but your dataset iterator ran out of data after ' +\n              `${stepsDone} batches; ` +\n              'interrupting training. Make sure that your ' +\n              'dataset can generate at least `batchesPerEpoch * epochs` ' +\n              'batches (in this case, ' +\n              `${args.batchesPerEpoch * args.epochs} batches). ` +\n              'You may need to use the repeat() function when building ' +\n              'your dataset.');\n          break;\n        }\n\n        if (iteratorOut.value != null) {\n          const {xs, ys} =\n              standardizeDataIteratorOutput(model, iteratorOut.value);\n          const batchLogs: UnresolvedLogs = {};\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = xs[0].shape[0];\n\n          await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n          const sampleWeights: tfc.Tensor[] = [];\n          if (args.classWeight != null) {\n            const standardClassWeights =\n                standardizeClassWeights(args.classWeight, model.outputNames);\n            for (let i = 0; i < standardClassWeights.length; ++i) {\n              sampleWeights.push(await standardizeWeights(\n                  ys[i], null, standardClassWeights[i]));\n            }\n          }\n\n          // Train on batch.\n          const ins = xs.concat(ys).concat(sampleWeights);\n          const outs = trainFunction(ins);\n          tfc.dispose(ins);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n          }\n\n          await callbackList.onBatchEnd(batchIndex, batchLogs);\n          disposeTensorsInLogs(batchLogs);\n\n          batchIndex++;\n          stepsDone++;\n        }\n\n        if (hasBatchesPerEpoch ? stepsDone >= args.batchesPerEpoch :\n                                 iteratorOut.done) {\n          // Epoch finished. Perform validation.\n          if (doValidation) {\n            let valOuts: tfc.Scalar[];\n            if (isDatasetObject(args.validationData)) {\n              valOuts = toList(await model.evaluateDataset(\n                  args.validationData, {batches: args.validationBatches}));\n            } else {\n              valOuts = toList(model.evaluate(valXs, valYs, {\n                batchSize: args.validationBatchSize == null ?\n                    DEFAULT_VALIDATION_BATCH_SIZE :\n                    args.validationBatchSize,\n                verbose: 0\n              }));\n            }\n            for (let i = 0; i < model.metricsNames.length; ++i) {\n              epochLogs[`val_${model.metricsNames[i]}`] = valOuts[i];\n            }\n          }\n          // Call `break` to exit one epoch lopp after validation is done. If\n          // config.batchesPerEpoch is specified, an epoch while loop will\n          // stop when `stepsDone >= config.batchesPerEpoch`. When\n          // config.batchesPerEpoch is not provided, the following `break` is\n          // required to exit the while lopp after dataset is exhausted.\n          break;\n        }\n\n        if (model.stopTraining_) {\n          break;\n        }\n      }\n      await callbackList.onEpochEnd(epoch, epochLogs);\n      epoch++;\n      if (model.stopTraining_) {\n        break;\n      }\n    }\n    await callbackList.onTrainEnd();\n    await model.history.syncData();\n    return model.history;\n  } finally {\n    model.isTraining = false;\n  }\n}\n\n/** Helper function that determines number of steps (batches) per epoch. */\nfunction getStepsPerEpoch<T>(\n    dataset: Dataset<T>, args: ModelFitDatasetArgs<T>): number {\n  // Attempt to determine # of batches in an epoch.\n  let stepsPerEpoch: number = null;\n  if (args.batchesPerEpoch != null) {\n    stepsPerEpoch = args.batchesPerEpoch;\n  } else if (Number.isFinite(dataset.size)) {\n    stepsPerEpoch = dataset.size;\n  }\n  return stepsPerEpoch;\n}\n\n// Check if provided object is a Dataset object by checking its .iterator\n// element.\nfunction isDatasetObject<T>(\n    dataset:\n        [\n          TensorOrArrayOrMap, TensorOrArrayOrMap\n        ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|\n    Dataset<T>): boolean {\n  return (typeof (dataset as Dataset<T>).iterator === 'function');\n}\n\n// Check if provided object is a LazyIterator object by checking it's .next\n// element.\nfunction isLazyIteratorObject<T>(iterator: Dataset<T>|\n                                 LazyIterator<T>): boolean {\n  return (typeof (iterator as LazyIterator<T>).next === 'function');\n}\n\nexport async function evaluateDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>|LazyIterator<T>,\n    args: ModelEvaluateDatasetArgs): Promise<tfc.Scalar|tfc.Scalar[]> {\n  args = args || {};\n  const hasBatches = args.batches != null;\n  const f = model.testFunction;\n  let outs: tfc.Scalar[] = [];\n  if (args.verbose > 0) {\n    throw new NotImplementedError('Verbose mode is not implemented yet.');\n  }\n\n  tfc.util.assert(\n      !hasBatches || (args.batches > 0 && Number.isInteger(args.batches)),\n      () => 'Test loop expects `batches` to be a positive integer, but ' +\n          `received ${JSON.stringify(args.batches)}`);\n  const dataIterator = isLazyIteratorObject(dataset) ?\n      dataset as LazyIterator<T>:\n      await (dataset as Dataset<T>).iterator();\n  // Keeps track of number of examples used in this evaluation.\n  let numExamples = 0;\n  let batch = 0;\n\n  while (hasBatches ? batch < args.batches : true) {\n    const iteratorOut = await dataIterator.next();\n    outs = tfc.tidy(() => {\n      if (iteratorOut.value) {\n        // TODO(cais): Once real dataset is available, use\n        //   `map(x => standardizeDataIteratorOutput(model, x).map(f)`.\n        const {xs, ys} =\n            standardizeDataIteratorOutput(model, iteratorOut.value);\n        const xsAndYs = xs.concat(ys);\n        const batchOuts = tfc.tidy(() => f(xsAndYs));\n        tfc.dispose(xsAndYs);\n\n        if (batch === 0) {\n          for (let i = 0; i < batchOuts.length; ++i) {\n            outs.push(scalar(0));\n          }\n        }\n\n        const batchSize = xsAndYs[0].shape[0];\n        for (let i = 0; i < batchOuts.length; ++i) {\n          const batchOut = batchOuts[i];\n          const oldScalar = outs[i];\n          outs[i] =\n              tfc.tidy(() => tfc.add(outs[i], tfc.mul(batchSize, batchOut)));\n          if (batch > 0) {\n            tfc.dispose(oldScalar);\n          }\n        }\n        tfc.dispose(batchOuts);\n        numExamples += batchSize;\n\n        ++batch;\n      }\n      return outs;\n    });\n\n    if (iteratorOut.done) {\n      if (hasBatches) {\n        console.warn(\n            'Your dataset iterator ran out of data during evaluateDataset(). ' +\n            'Interrupting evalution. Make sure that your ' +\n            'dataset can generate at least `batches` ' +\n            `batches (in this case, ${args.batches} batches). ` +\n            'You may need to use the repeat() function when building ' +\n            'your dataset.');\n      }\n      break;\n    }\n  }\n\n  for (let i = 0; i < outs.length; ++i) {\n    const oldScalar = outs[i];\n    outs[i] = tfc.div(outs[i], numExamples);\n    tfc.dispose(oldScalar);\n  }\n\n  return singletonOrArray(outs);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using tf.Tensor objects.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Scalar, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport {expandDims, gather, sliceAlongFirstAxis} from '../backend/tfjs_backend';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {range} from '../utils/math_utils';\nimport {ClassWeight, ClassWeightMap} from './training_utils';\n\n/**\n * Interface configuration model training based on data as `tf.Tensor`s.\n */\nexport interface ModelFitArgs {\n  /**\n   * Number of samples per gradient update. If unspecified, it\n   * will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Integer number of times to iterate over the training data arrays.\n   */\n  epochs?: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Float between 0 and 1: fraction of the training data\n   * to be used as validation data. The model will set apart this fraction of\n   * the training data, will not train on it, and will evaluate the loss and\n   * any model metrics on this data at the end of each epoch.\n   * The validation data is selected from the last samples in the `x` and `y`\n   * data provided, before shuffling.\n   */\n  validationSplit?: number;\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal,\n   * valSampleWeights]. The model will not be trained on this data.\n   * `validationData` will override `validationSplit`.\n   */\n  validationData?: [\n    Tensor|Tensor[], Tensor|Tensor[]\n  ]|[Tensor | Tensor[], Tensor|Tensor[], Tensor|Tensor[]];\n\n  /**\n   * Whether to shuffle the training data before each epoch. Has\n   * no effect when `stepsPerEpoch` is not `null`.\n   */\n  shuffle?: boolean;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n\n  /**\n   * Optional array of the same length as x, containing\n   * weights to apply to the model's loss for each sample. In the case of\n   * temporal data, you can pass a 2D array with shape (samples,\n   * sequenceLength), to apply a different weight to every timestep of every\n   * sample. In this case you should make sure to specify\n   * sampleWeightMode=\"temporal\" in compile().\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. When training\n   * with Input Tensors such as TensorFlow data tensors, the default `null` is\n   * equal to the number of unique samples in your dataset divided by the\n   * batch size, or 1 if that cannot be determined.\n   */\n  stepsPerEpoch?: number;\n\n  /**\n   * Only relevant if `stepsPerEpoch` is specified. Total number of steps\n   * (batches of samples) to validate before stopping.\n   */\n  validationSteps?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - any `number`: yield every `number` milliseconds.\n   *   - `'never'`: never yield. (yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n}\n\nexport function checkBatchSize(batchSize: number) {\n  tfc.util.assert(\n      batchSize > 0 && Number.isInteger(batchSize),\n      () => `batchSize is required to be a positive integer, but got ${\n          batchSize}`);\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by start and stop indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArraysByIndices()` together.\n *\n * @param arrays: the input.\n * @param start: the starting index (inclusive).\n * @param stop: the stopping index (exclusive).\n * @returns The result of the slicing. If `arrays` is an `Array` of\n *   `tf.Tensor`s, the slicing will be applied to all elements of the `Array`\n *   in the same way.\n */\nexport function sliceArrays(\n    arrays: Tensor|Tensor[], start: number, stop: number): Tensor|Tensor[] {\n  if (arrays == null) {\n    return [null];\n  } else if (Array.isArray(arrays)) {\n    return arrays.map(array => sliceAlongFirstAxis(array, start, stop - start));\n  } else {  // Tensor.\n    return sliceAlongFirstAxis(arrays, start, stop - start);\n  }\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by random-order indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArrays()` together.\n *\n * @param arrays The input `tf.Tensor` or `Array` of `tf.Tensor`s to slice.\n *   If an `Array` of `tf.Tensor`s, all `tf.Tensor`s will be sliced in the\n *   same fashion.\n * @param indices The indices to use for slicing along the first (batch)\n *   dimension.\n * @returns Result(s) of the slicing.\n */\nexport function sliceArraysByIndices(\n    arrays: Tensor|Tensor[], indices: Tensor1D): Tensor|Tensor[] {\n  return tfc.tidy(() => {\n    if (arrays == null) {\n      return null;\n    } else if (Array.isArray(arrays)) {\n      return arrays.map(\n          array => (sliceArraysByIndices(array, indices) as Tensor));\n    } else {\n      // TODO(cais): indices should be a pre-constructed Tensor1D to avoid\n      //   tensor1d() calls.\n      return gather(\n          arrays, indices.dtype === 'int32' ? indices : indices.toInt());\n    }\n  });\n}\n\n/**\n * Returns a list of batch indices (tuples of indices).\n * @param size: Integer, total size of the data to slice into batches.\n * @param batchSize: Integer, batch size.\n * @returns An Array of [batchStart, batchEnd] tuples. batchStart is\n *   inclusive; batchEnd is exclusive. I.e., each batch consists of indices x\n *   that satisfy batchStart <= x < batchEnd.\n */\nexport function makeBatches(\n    size: number, batchSize: number): Array<[number, number]> {\n  const output: Array<[number, number]> = [];\n  let batchStart = 0;\n  let batchEnd: number = null;\n  while (batchStart < size) {\n    batchEnd = batchStart + batchSize;\n    if (batchEnd >= size) {\n      batchEnd = size;\n    }\n    output.push([batchStart, batchEnd]);\n    batchStart = batchEnd;\n  }\n  return output;\n}\n\n/**\n * Abstract fit function for `f(ins)`.\n * @param f A Function returning a list of tensors. For training, this\n *   function is expected to perform the updates to the variables.\n * @param ins List of tensors to be fed to `f`.\n * @param outLabels List of strings, display names of the outputs of `f`.\n * @param batchSize Integer batch size or `== null` if unknown. Default : 32.\n * @param epochs Number of times to iterate over the data. Default : 1.\n * @param verbose Verbosity mode: 0, 1, or 2. Default: 1.\n * @param callbacks List of callbacks to be called during training.\n * @param valF Function to call for validation.\n * @param valIns List of tensors to be fed to `valF`.\n * @param shuffle Whether to shuffle the data at the beginning of every\n * epoch. Default : true.\n * @param callbackMetrics List of strings, the display names of the metrics\n *   passed to the callbacks. They should be the concatenation of the\n *   display names of the outputs of `f` and the list of display names\n *   of the outputs of `valF`.\n * @param initialEpoch Epoch at which to start training (useful for\n *   resuming a previous training run). Default : 0.\n * @param stepsPerEpoch Total number of steps (batches on samples) before\n *   declaring one epoch finished and starting the next epoch. Ignored with\n *   the default value of `undefined` or `null`.\n * @param validationSteps Number of steps to run validation for (only if\n *   doing validation from data tensors). Not applicable for tfjs-layers.\n * @returns A `History` object.\n */\nasync function fitLoop(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, f: (data: Tensor[]) => Scalar[], ins: Tensor[],\n    outLabels?: string[], batchSize?: number, epochs?: number, verbose?: number,\n    callbacks?: BaseCallback[], valF?: (data: Tensor[]) => Scalar[],\n    valIns?: Tensor[], shuffle?: boolean|string, callbackMetrics?: string[],\n    initialEpoch?: number, stepsPerEpoch?: number,\n    validationSteps?: number): Promise<History> {\n  if (batchSize == null) {\n    batchSize = 32;\n  }\n  if (epochs == null) {\n    epochs = 1;\n  }\n  if (shuffle == null) {\n    shuffle = true;\n  }\n  if (initialEpoch == null) {\n    initialEpoch = 0;\n  }\n\n  // TODO(cais): Change const to let below when implementing validation.\n  let doValidation = false;\n  if (valF != null && valIns != null) {\n    doValidation = true;\n    // TODO(cais): verbose message.\n  }\n  if (validationSteps != null) {\n    doValidation = true;\n    if (stepsPerEpoch == null) {\n      throw new ValueError(\n          'Can only use `validationSteps` when doing step-wise training, ' +\n          'i.e., `stepsPerEpoch` must be set.');\n    }\n  }\n\n  const numTrainSamples =\n      model.checkNumSamples(ins, batchSize, stepsPerEpoch, 'steps_per_epoch');\n  let indexArray: number[];\n  if (numTrainSamples != null) {\n    indexArray = range(0, numTrainSamples);\n  }\n\n  if (verbose == null) {\n    verbose = 1;\n  }\n\n  const {callbackList, history} = configureCallbacks(\n      callbacks, verbose, epochs, initialEpoch, numTrainSamples, stepsPerEpoch,\n      batchSize, doValidation, callbackMetrics);\n  callbackList.setModel(model);\n  model.history = history;\n  await callbackList.onTrainBegin();\n  model.stopTraining_ = false;\n  // TODO(cais): Take care of callbacks.validation_data as in PyKeras.\n  // TODO(cais): Pre-convert feeds for performance as in PyKeras.\n\n  for (let epoch = initialEpoch; epoch < epochs; ++epoch) {\n    await callbackList.onEpochBegin(epoch);\n    const epochLogs: UnresolvedLogs = {};\n    if (stepsPerEpoch != null) {\n      throw new NotImplementedError(\n          'stepsPerEpoch mode is not implemented yet.');\n    } else {\n      if (shuffle === 'batch') {\n        throw new NotImplementedError('batch shuffling is not implemneted yet');\n      } else if (shuffle) {\n        util.shuffle(indexArray);\n      }\n      // Convert the potentially shuffled indices to Tensor1D, to avoid the\n      // cost of repeated creation of Array1Ds later on.\n      const epochIndexArray1D = tensor1d(indexArray);\n\n      const batches = makeBatches(numTrainSamples, batchSize);\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchLogs: UnresolvedLogs = {};\n        await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n        tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds = sliceAlongFirstAxis(\n                               epochIndexArray1D, batchStart,\n                               batchEnd - batchStart) as Tensor1D;\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = batchEnd - batchStart;\n\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Tensor[];\n          const outs = f(insBatch);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n            // TODO(cais): Use scope() to avoid ownership.\n          }\n\n          if (batchIndex === batches.length - 1) {  // Last batch.\n            if (doValidation) {\n              const valOuts = model.testLoop(valF, valIns, batchSize);\n              // Porting Notes: In tfjs-layers, valOuts is always an Array.\n              for (let i = 0; i < outLabels.length; ++i) {\n                const label = outLabels[i];\n                const out = valOuts[i];\n                tfc.keep(out);\n                // TODO(cais): Use scope() to avoid ownership.\n                epochLogs['val_' + label] = out;\n              }\n            }\n          }\n        });\n\n        await callbackList.onBatchEnd(batchIndex, batchLogs);\n        disposeTensorsInLogs(batchLogs);\n\n        if (model.stopTraining_) {\n          break;\n        }\n        // TODO(cais): return outs as list of Tensor.\n      }\n\n      epochIndexArray1D.dispose();\n    }\n    // TODO(cais): Run validation at the end of the epoch.\n    await callbackList.onEpochEnd(epoch, epochLogs);\n    if (model.stopTraining_) {\n      break;\n    }\n  }\n  await callbackList.onTrainEnd();\n\n  await model.history.syncData();\n  return model.history;\n}\n\nexport async function fitTensors(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    args: ModelFitArgs = {}): Promise<History> {\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n  let inputs: Tensor[];\n  let targets: Tensor[];\n  let inputValX: Tensor|Tensor[];\n  let inputValY: Tensor|Tensor[];\n  let valX: Tensor|Tensor[];\n  let valY: Tensor|Tensor[];\n  let sampleWeights: Tensor[];\n  try {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // Validate user data.\n    // TODO(cais): Support sampleWeight.\n    const checkBatchAxis = false;\n    const standardizedOuts =\n        await model.standardizeUserData(\n            x, y, args.sampleWeight, args.classWeight, checkBatchAxis,\n            batchSize) as [Tensor[], Tensor[], Tensor[]];\n    inputs = standardizedOuts[0];\n    targets = standardizedOuts[1];\n    sampleWeights = standardizedOuts[2];\n\n    // Prepare validation data.\n    let doValidation = false;\n    let valIns: Tensor[];\n    if (args.validationData != null && args.validationData.length > 0) {\n      doValidation = true;\n      if (args.validationData.length === 2) {\n        // config.validationData consists of valX and valY.\n        inputValX = args.validationData[0];\n        inputValY = args.validationData[1];\n      } else if (args.validationData.length === 3) {\n        throw new NotImplementedError(\n            'validationData including sample weights is not supported yet.');\n      } else {\n        throw new ValueError(\n            `When passing validation data, it must contain 2 (valX, valY) ` +\n            `or 3 (valX, valY, valSampleWeight) items; ` +\n            `${args.validationData} is invalid.`);\n      }\n\n      const checkBatchAxis = true;\n      const valStandardized =\n          await model.standardizeUserData(\n              inputValX, inputValY, null, /** Unused sample weights. */\n              null,                       /** Unused class weights. */\n              checkBatchAxis, batchSize) as [Tensor[], Tensor[], Tensor[]];\n      valX = valStandardized[0];\n      valY = valStandardized[1];\n      valIns = valX.concat(valY);\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (\n        args.validationSplit != null && args.validationSplit > 0 &&\n        args.validationSplit < 1) {\n      doValidation = true;\n      // Porting Note: In tfjs-layers, inputs[0] is always a Tensor.\n      const splitAt =\n          Math.floor(inputs[0].shape[0] * (1 - args.validationSplit));\n      const originalBatchSize = inputs[0].shape[0];\n      valX = sliceArrays(inputs, splitAt, originalBatchSize) as Tensor[];\n      inputs = sliceArrays(inputs, 0, splitAt) as Tensor[];\n      valY = sliceArrays(targets, splitAt, originalBatchSize) as Tensor[];\n      targets = sliceArrays(targets, 0, splitAt) as Tensor[];\n      // TODO(cais): Once sampleWeights becomes available, slice it to get\n      //   valSampleWeights.\n      valIns = valX.concat(valY);\n\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (args.validationSteps != null) {\n      doValidation = true;\n      // TODO(cais): Add useLearningPhase.\n    }\n\n    const ins = inputs.concat(targets).concat(sampleWeights);\n\n    model.checkTrainableWeightsConsistency();\n\n    // TODO(cais): Handle use_learning_phase and learning_phase?\n\n    // Porting Note: Here we see a key deviation of tfjs-layers from\n    // Keras.\n    //  Due to the imperative nature of tfjs-layers' backend (tfjs-core),\n    //  we do not construct symbolic computation graphs to embody the\n    //  training process. Instead, we define a function that performs the\n    //  training action. In PyKeras, the data (inputs and targets) are fed\n    //  through graph placeholders. In tfjs-layers, the data are fed as\n    //  function arguments. Since the function are defined below in the\n    //  scope, we don't have equivalents of PyKeras's\n    //  `_make_train_funciton`.\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let valFunction: (data: Tensor[]) => Scalar[];\n    let callbackMetrics: string[];\n    if (doValidation) {\n      model.makeTestFunction();\n      valFunction = model.testFunction;\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      valFunction = null;\n      valIns = [];\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const out = await fitLoop(\n        model, trainFunction, ins, outLabels, batchSize, args.epochs,\n        args.verbose, callbacks, valFunction, valIns, args.shuffle,\n        callbackMetrics, args.initialEpoch, null, null);\n    return out;\n  } finally {\n    model.isTraining = false;\n    // Memory clean up.\n    disposeNewTensors(inputs, x);\n    disposeNewTensors(targets, y);\n    disposeNewTensors(valX as Tensor[], inputValX);\n    disposeNewTensors(valY as Tensor[], inputValY);\n    if (sampleWeights != null) {\n      tfc.dispose(sampleWeights);\n    }\n  }\n  // TODO(cais): Add value to outLabels.\n}\n\n/**\n * Ensure tensors all have a rank of at least 2.\n *\n * If a tensor has a rank of 1, it is dimension-expanded to rank 2.\n * If any tensor has a rank of 0 (i.e., is a scalar), an error will be thrown.\n */\nexport function ensureTensorsRank2OrHigher(tensors: Tensor|Tensor[]): Tensor[] {\n  const outs: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    tensors = [tensors];\n  }\n\n  // Make Tensors at least 2D.\n  for (let i = 0; i < tensors.length; ++i) {\n    const tensor = tensors[i];\n    if (tensor.rank === 1) {\n      outs.push(expandDims(tensor, 1));\n    } else if (tensor.rank === 0) {\n      throw new Error(\n          'Expected tensor to be at least 1D, but received a 0D tensor ' +\n          '(scalar).');\n    } else {\n      outs.push(tensor);\n    }\n  }\n  return outs;\n}\n\n/**\n * Compare a set of tensors with a reference (old) set, discard the ones\n * in the new set that are not present in the reference set.\n *\n * This method is used for memory clenaup during calls such as\n * LayersModel.fit().\n *\n * @param tensors New set which may contain Tensors not present in\n *   `refTensors`.\n * @param refTensors Reference Tensor set.\n */\n// TODO(cais, kangyizhang): Deduplicate with tfjs-data.\nexport function disposeNewTensors(\n    tensors: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    refTensors: Tensor|Tensor[]|{[inputName: string]: Tensor}): void {\n  if (tensors == null) {\n    return;\n  }\n  const oldTensorIds: number[] = [];\n  if (refTensors instanceof Tensor) {\n    oldTensorIds.push(refTensors.id);\n  } else if (Array.isArray(refTensors)) {\n    refTensors.forEach(t => oldTensorIds.push(t.id));\n  } else if (refTensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in refTensors) {\n      const oldTensor = refTensors[name];\n      oldTensorIds.push(oldTensor.id);\n    }\n  }\n\n  const tensorsToDispose: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    if (oldTensorIds.indexOf(tensors.id) === -1) {\n      tensorsToDispose.push(tensors);\n    }\n  } else if (Array.isArray(tensors)) {\n    tensors.forEach(t => {\n      if (oldTensorIds.indexOf(t.id) === -1) {\n        tensorsToDispose.push(t);\n      }\n    });\n  } else if (tensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in tensors) {\n      const tensor = tensors[name];\n      if (oldTensorIds.indexOf(tensor.id) === -1) {\n        tensorsToDispose.push(tensor);\n      }\n    }\n  }\n\n  tensorsToDispose.forEach(t => {\n    if (!t.isDisposed) {\n      t.dispose();\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: engine/training.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {io, ModelPredictConfig as ModelPredictArgs, NamedTensorMap, Optimizer, Scalar, scalar, serialization, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {History, ModelLoggingVerbosity} from '../base_callbacks';\nimport {nameScope} from '../common';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {LossIdentifier} from '../keras_format/loss_config';\nimport {OptimizerSerialization} from '../keras_format/optimizer_config';\nimport {MetricsIdentifier, TrainingConfig} from '../keras_format/training_config';\nimport {deserialize} from '../layers/serialization';\nimport * as losses from '../losses';\nimport * as Metrics from '../metrics';\nimport * as optimizers from '../optimizers';\nimport {LossOrMetricFn, NamedTensor} from '../types';\nimport {checkUserDefinedMetadata} from '../user_defined_metadata';\nimport {count, pyListRepeat, singletonOrArray, toCamelCase, toSnakeCase, unique} from '../utils/generic_utils';\nimport {printSummary} from '../utils/layer_utils';\nimport {range} from '../utils/math_utils';\nimport {convertPythonicToTs} from '../utils/serialization_utils';\nimport {LayerVariable} from '../variables';\nimport {version} from '../version';\n\nimport {Container, ContainerArgs} from './container';\nimport {Dataset} from './dataset_stub';\nimport {execute, FeedDict} from './executor';\nimport {DisposeResult, SymbolicTensor} from './topology';\nimport {evaluateDataset, fitDataset, ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './training_dataset';\nimport {checkBatchSize, disposeNewTensors, ensureTensorsRank2OrHigher, fitTensors, makeBatches, ModelFitArgs, sliceArrays, sliceArraysByIndices} from './training_tensors';\nimport {ClassWeight, ClassWeightMap, computeWeightedLoss, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Helper function for polymorphic input data: 1. singleton Tensor.\n */\nexport function isDataTensor(x: Tensor|Tensor[]|{[inputName: string]: Tensor}|\n                             {[inputName: string]: Tensor[]}): boolean {\n  return x instanceof Tensor;\n}\n\n/**\n * Helper function for polymorphic input data: 2. Array of Tensor.\n */\nexport function isDataArray(x: Tensor|Tensor[]|\n                            {[inputName: string]: Tensor}): boolean {\n  return Array.isArray(x);\n}\n\n/**\n * Helper function for polymorphic input data: 3. \"dict\" of Tensor.\n */\nexport function isDataDict(x: Tensor|Tensor[]|\n                           {[inputName: string]: Tensor}): boolean {\n  return !isDataTensor(x) && !isDataArray(x);\n}\n\n/**\n * Normalizes inputs and targets provided by users.\n * @param data User-provided input data (polymorphic).\n * @param names An Array of expected Tensor names.\n * @param shapes Optional Array of expected Tensor shapes.\n * @param checkBatchAxis Whether to check that the batch axis of the arrays\n *   match  the expected value found in `shapes`.\n * @param exceptionPrefix String prefix used for exception formatting.\n * @returns List of standardized input Tensors (one Tensor per model input).\n * @throws ValueError: in case of improperly formatted user data.\n */\nexport function standardizeInputData(\n    data: Tensor|Tensor[]|{[inputName: string]: Tensor}, names: string[],\n    shapes?: Shape[], checkBatchAxis = true, exceptionPrefix = ''): Tensor[] {\n  if (names == null || names.length === 0) {\n    // Check for the case where the model expected no data, but some data got\n    // sent.\n    if (data != null) {\n      let gotUnexpectedData = false;\n      if (isDataArray(data) && (data as Tensor[]).length > 0) {\n        gotUnexpectedData = true;\n      } else if (isDataDict(data)) {\n        for (const key in data) {\n          if (data.hasOwnProperty(key)) {\n            gotUnexpectedData = true;\n            break;\n          }\n        }\n      } else {\n        // `data` is a singleton Tensor in this case.\n        gotUnexpectedData = true;\n      }\n      if (gotUnexpectedData) {\n        throw new ValueError(\n            `Error when checking model ${exceptionPrefix} expected no data, ` +\n            `but got ${data}`);\n      }\n    }\n    return [];\n  }\n  if (data == null) {\n    return names.map(name => null);\n  }\n\n  let arrays: Tensor[];\n  if (isDataDict(data)) {\n    data = data as {[inputName: string]: Tensor};\n    arrays = [];\n    for (const name of names) {\n      if (data[name] == null) {\n        throw new ValueError(\n            `No data provided for \"${name}\". Need data for each key in: ` +\n            `${names}`);\n      }\n      arrays.push(data[name]);\n    }\n  } else if (isDataArray(data)) {\n    data = data as Tensor[];\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `model expected. Expected to see ${names.length} Tensor(s), but ` +\n          `instead got the following list of Tensor(s): ${data}`);\n    }\n    arrays = data;\n  } else {\n    data = data as Tensor;\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model ${exceptionPrefix} expects ${names.length} Tensor(s), ` +\n          `but only received one Tensor. Found: Tensor with shape ${\n              data.shape}`);\n    }\n    arrays = [data];\n  }\n\n  arrays = ensureTensorsRank2OrHigher(arrays);\n\n  // Check shape compatibility.\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s). but got array with ` +\n            `shape ${array.shape}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          // Skip the first (batch) axis.\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null && refDim >= 0 && dim !== refDim) {\n          throw new ValueError(\n              `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n              `to have shape [${shapes[i]}], but got array with shape ` +\n              `[${array.shape}].`);\n        }\n      }\n    }\n  }\n  return arrays;\n}\n\n/**\n * User input validation for Tensors.\n * @param inputs `Array` of `tf.Tensor`s for inputs.\n * @param targets `Array` of `tf.Tensor`s for targets.\n * @param weights Optional `Array` of `tf.Tensor`s for sample weights.\n * @throws ValueError: in case of incorrectly formatted data.\n */\nexport function checkArrayLengths(\n    inputs: Tensor[], targets: Tensor[], weights?: Tensor[]) {\n  const setX = unique(inputs.map(input => input.shape[0]));\n  setX.sort();\n  const setY = unique(targets.map(target => target.shape[0]));\n  setY.sort();\n  // TODO(cais): Check `weights` as well.\n  if (setX.length > 1) {\n    throw new ValueError(\n        `All input Tensors (x) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(inputs.map(input => input.shape))}`);\n  }\n  if (setY.length > 1) {\n    throw new ValueError(\n        `All target Tensors (y) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(targets.map(target => target.shape))}`);\n  }\n  if (setX.length > 0 && setY.length > 0 && !util.arraysEqual(setX, setY)) {\n    throw new ValueError(\n        `Input Tensors should have the same number of samples as target ` +\n        `Tensors. Found ${setX[0]} input sample(s) and ${setY[0]} target ` +\n        `sample(s).`);\n  }\n}\n\n/**\n * Validation on the compatibility of targes and loss functions.\n *\n * This helps prevent users from using loss functions incorrectly.\n *\n * @param targets `Array` of `tf.Tensor`s of targets.\n * @param lossFns `Array` of loss functions.\n * @param outputShapes `Array` of shapes of model outputs.\n */\nfunction checkLossAndTargetCompatibility(\n    targets: Tensor[], lossFns: LossOrMetricFn[], outputShapes: Shape[]) {\n  // TODO(cais): Dedicated test coverage?\n  const keyLosses = [\n    losses.meanSquaredError, losses.binaryCrossentropy,\n    losses.categoricalCrossentropy\n  ];\n  for (let i = 0; i < targets.length; ++i) {\n    const y = targets[i];\n    const loss = lossFns[i];\n    const shape = outputShapes[i];\n    if (loss == null) {\n      continue;\n    }\n    if (loss === losses.categoricalCrossentropy) {\n      if (y.shape[y.shape.length - 1] === 1) {\n        throw new ValueError(\n            `You are passing a target array of shape ${y.shape} while using ` +\n            `a loss 'categorical_crossentropy'. 'categorical_crossentropy'` +\n            `expects targets to be binary matrices (1s and 0s) of shape ` +\n            `[samples, classes].`);\n        // TODO(cais): Example code in error message.\n      }\n    }\n    if (keyLosses.indexOf(loss) !== -1) {\n      const slicedYShape = y.shape.slice(1);\n      const slicedShape = shape.slice(1);\n      for (let j = 0; j < slicedYShape.length; ++j) {\n        const targetDim = slicedYShape[j];\n        const outDim = slicedShape[j];\n        if (outDim != null && targetDim !== outDim) {\n          throw new ValueError(\n              `A target Tensor with shape ${y.shape} was passed for an ` +\n              `output of shape ${shape}, while using a loss function that ` +\n              `expects targets to have the same shape as the output.`);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Check inputs provided by the user.\n *\n * Porting Note: This corresponds to _standardize_input_data() in Python\n *   Keras. Because of the strong typing in TF.js, we do not need to convert\n *   the data. Specifically:\n *   1) in PyKeras, `data` can be `DataFrame` instances from pandas, for\n *      example. We don't need to worry about that here because there is no\n *      widely popular javascript/typesdcript equivalent of pandas (so far).\n *      If one becomes available in the future, we can add support.\n *   2) in PyKeras, inputs can be Python dict. But here we are stipulating\n * that the data is either a single `tf.Tensor` or an Array of `tf.Tensor`s. We\n * may add support for `Object` data inputs in the future when the need\n * arises.\n *\n * Instead, we perform basic checks for number of parameters and shapes.\n *\n * @param data: The input data.\n * @param names: Name for the inputs, from the model.\n * @param shapes: Expected shapes for the input data, from the model.\n * @param checkBatchAxis: Whether the size along the batch axis (i.e., the\n *   first dimension) will be checked for matching.\n * @param exceptionPrefix: Execption prefix message, used in generating error\n *   messages.\n * @throws ValueError: on incorrect number of inputs or mismatches in shapes.\n */\nfunction checkInputData(\n    data: Tensor|Tensor[], names: string[], shapes?: Shape[],\n    checkBatchAxis = true, exceptionPrefix = '') {\n  let arrays: Tensor[];\n  if (Array.isArray(data)) {\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `the model expected. Expected to see ${names.length} Tensor(s),` +\n          ` but instead got ${data.length} Tensors(s).`);\n    }\n    arrays = data;\n  } else {\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model expects ${names.length} ${exceptionPrefix} Tensors, ` +\n          `but only received one Tensor. Found: array with shape ` +\n          `${JSON.stringify(data.shape)}.`);\n    }\n    arrays = [data];\n  }\n\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s), but got array with ` +\n            `shape ${JSON.stringify(array.shape)}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null) {\n          if (refDim !== dim) {\n            throw new ValueError(\n                `Error when checking ${exceptionPrefix}: expected ` +\n                `${names[i]} to have shape ${JSON.stringify(shapes[i])} but ` +\n                `got array with shape ${JSON.stringify(array.shape)}.`);\n          }\n        }\n      }\n    }\n  }\n}\n\n/**\n * Maps metric functions to model outputs.\n * @param metrics An shortcut strings name, metric function, `Array` or dict\n *   (`Object`) of metric functions.\n * @param outputNames An `Array` of the names of model outputs.\n * @returns An `Array` (one entry per model output) of `Array` of metric\n *   functions. For instance, if the model has 2 outputs, and for the first\n *   output we want to compute `binaryAccuracy` and `binaryCrossentropy`,\n *   and just `binaryAccuracy` for the second output, the `Array` would look\n *   like:\n *     `[[binaryAccuracy, binaryCrossentropy],  [binaryAccuracy]]`\n * @throws TypeError: incompatible metrics format.\n */\nexport function collectMetrics(\n    metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n    {[outputName: string]: string | LossOrMetricFn},\n    outputNames: string[]): Array<Array<string|LossOrMetricFn>> {\n  if (metrics == null || Array.isArray(metrics) && metrics.length === 0) {\n    return outputNames.map(name => []);\n  }\n\n  let wrappedMetrics: Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  if (typeof metrics === 'string' || typeof metrics === 'function') {\n    wrappedMetrics = [metrics];\n  } else if (Array.isArray(metrics) || typeof metrics === 'object') {\n    wrappedMetrics = metrics as Array<string|LossOrMetricFn>|\n        {[outputName: string]: string} | {[outputName: string]: LossOrMetricFn};\n  } else {\n    throw new TypeError(\n        'Type of metrics argument not understood. Expected an string,' +\n        `function, Array, or Object, found: ${metrics}`);\n  }\n\n  if (Array.isArray(wrappedMetrics)) {\n    // We then apply all metrics to all outputs.\n    return outputNames.map(\n        name => wrappedMetrics as Array<string|LossOrMetricFn>);\n  } else {\n    // In this case, metrics is a dict.\n    const nestedMetrics: Array<Array<string|LossOrMetricFn>> = [];\n    for (const name of outputNames) {\n      let outputMetrics: string|LossOrMetricFn|Array<string|LossOrMetricFn> =\n          wrappedMetrics.hasOwnProperty(name) ? wrappedMetrics[name] : [];\n      if (!Array.isArray(outputMetrics)) {\n        outputMetrics = [outputMetrics];\n      }\n      nestedMetrics.push(outputMetrics);\n    }\n    return nestedMetrics;\n  }\n}\n\nexport interface ModelEvaluateArgs {\n  /**\n   * Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * Tensor of weights to weight the contribution of different samples to the\n   * loss and metrics.\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * integer: total number of steps (batches of samples)\n   * before declaring the evaluation round finished. Ignored with the default\n   * value of `undefined`.\n   */\n  steps?: number;\n}\n\n/**\n * Configuration for calls to `LayersModel.compile()`.\n */\nexport interface ModelCompileArgs {\n  /**\n   * An instance of `tf.train.Optimizer` or a string name for an Optimizer.\n   */\n  optimizer: string|Optimizer;\n\n  /**\n   * Object function(s) or name(s) of object function(s).\n   * If the model has multiple outputs, you can use a different loss\n   * on each output by passing a dictionary or an Array of losses.\n   * The loss value that will be minimized by the model will then be the sum\n   * of all individual losses.\n   */\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n\n  /**\n   * List of metrics to be evaluated by the model during training and testing.\n   * Typically you will use `metrics=['accuracy']`.\n   * To specify different metrics for different outputs of a multi-output\n   * model, you could also pass a dictionary.\n   */\n  metrics?: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n\n  // TODO(cais): Add lossWeights, sampleWeightMode, weightedMetrics, and\n  //   targetTensors.\n}\n\nconst LAYERS_MODEL_FORMAT_NAME = 'layers-model';\n\n/**\n * A `tf.LayersModel` is a directed, acyclic graph of `tf.Layer`s plus methods\n * for training, evaluation, prediction and saving.\n *\n * `tf.LayersModel` is the basic unit of training, inference and evaluation in\n * TensorFlow.js. To create a `tf.LayersModel`, use `tf.LayersModel`.\n *\n * See also:\n *   `tf.Sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class LayersModel extends Container implements tfc.InferenceModel {\n  // The class name is 'Model' rather than 'LayersModel' for backwards\n  // compatibility since this class name shows up in the serialization format.\n  /** @nocollapse */\n  static className = 'Model';\n  protected optimizer_: Optimizer;\n  // Whether the model instance owns the optimizer: `true` if and only if\n  // `optimizer` is created from a string parameter during `compile()` call.\n  protected isOptimizerOwned: boolean;\n\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n  lossFunctions: LossOrMetricFn[];\n\n  // TODO(cais): These private variables should probably not have the string\n  //   'feed' in their names, because we are not dealing with a symbolic\n  //   backend.\n  private feedOutputShapes: Shape[];\n  private feedLossFns: LossOrMetricFn[];\n  private collectedTrainableWeights: LayerVariable[];\n  private testFunction: (data: Tensor[]) => Scalar[];\n  history: History;\n\n  // A public property that can be set by Callbacks to order early stopping\n  // during `fit()` calls.\n  protected stopTraining_: boolean;\n  protected isTraining: boolean;\n\n  metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  metricsNames: string[];\n  // Porting Note: `metrics_tensors` in PyKeras is a symbolic tensor. But given\n  //   the imperative nature of tfjs-core, `metricsTensors` is a\n  //   TypeScript function here.\n  //   Also note that due to the imperative nature of tfjs-core, `metricsTensor`\n  //   here needs an output index to keep track of which output of the\n  //   LayersModel a metric belongs to. This is unlike `metrics_tensors` in\n  //   PyKeras, which is a `list` of symbolic tensors, each of which has\n  //   implicit \"knowledge\" of the outputs it depends on.\n  metricsTensors: Array<[LossOrMetricFn, number]>;\n\n  // User defind metadata (if any).\n  private userDefinedMetadata: {};\n\n  constructor(args: ContainerArgs) {\n    super(args);\n    this.isTraining = false;\n  }\n\n  /**\n   * Print a text summary of the model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - If the model has non-sequential-like topology, the inputs each layer\n   *   receives\n   * - The total number of trainable and non-trainable parameters of the model.\n   *\n   * ```js\n   * const input1 = tf.input({shape: [10]});\n   * const input2 = tf.input({shape: [20]});\n   * const dense1 = tf.layers.dense({units: 4}).apply(input1);\n   * const dense2 = tf.layers.dense({units: 8}).apply(input2);\n   * const concat = tf.layers.concatenate().apply([dense1, dense2]);\n   * const output =\n   *     tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat);\n   *\n   * const model = tf.model({inputs: [input1, input2], outputs: output});\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      throw new ValueError(\n          `This model has never been called, thus its weights have not been ` +\n          `created yet. So no summary can be displayed. Build the model ` +\n          `first (e.g., by calling it on some test data).`);\n    }\n    printSummary(this, lineLength, positions, printFn);\n  }\n\n  /**\n   * Configures and prepares the model for training and evaluation.  Compiling\n   * outfits the model with an optimizer, loss, and/or metrics.  Calling `fit`\n   * or `evaluate` on an un-compiled model will throw an error.\n   *\n   * @param args a `ModelCompileArgs` specifying the loss, optimizer, and\n   * metrics to be used for fitting and evaluating this model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  compile(args: ModelCompileArgs): void {\n    if (args.loss == null) {\n      args.loss = [];\n    }\n    this.loss = args.loss;\n\n    if (typeof args.optimizer === 'string') {\n      this.optimizer_ = optimizers.getOptimizer(args.optimizer);\n      this.isOptimizerOwned = true;\n    } else {\n      if (!(args.optimizer instanceof Optimizer)) {\n        throw new ValueError(\n            `User-defined optimizer must be an instance of tf.Optimizer.`);\n      }\n      this.optimizer_ = args.optimizer;\n      this.isOptimizerOwned = false;\n    }\n\n    // TODO(cais): Add lossWeights.\n    // TODO(cais): Add sampleWeightMode.\n\n    // Prepare loss functions.\n    let lossFunctions: LossOrMetricFn[] = [];\n    if (!Array.isArray(args.loss) && typeof args.loss !== 'string' &&\n        typeof args.loss !== 'function') {\n      args.loss = args.loss as {[outputName: string]: string};\n      for (const name in args.loss) {\n        if (this.outputNames.indexOf(name) === -1) {\n          throw new ValueError(\n              `Unknown entry in loss dictionary: \"${name}\". ` +\n              `Only expected the following keys: ${this.outputNames}`);\n        }\n      }\n      for (const name of this.outputNames) {\n        if (args.loss[name] == null) {\n          console.warn(\n              `Output \"${name}\" is missing from loss dictionary. We assume ` +\n              `this was done on purpose, and we will not be expecting data ` +\n              `to be passed to ${name} during training`);\n        }\n        lossFunctions.push(losses.get(args.loss[name]));\n      }\n    } else if (Array.isArray(args.loss)) {\n      if (args.loss.length !== this.outputs.length) {\n        throw new ValueError(\n            `When passing an Array as loss, it should have one entry per ` +\n            `model output. The model has ${this.outputs.length} output(s), ` +\n            `but you passed loss=${args.loss}.`);\n      }\n      const theLosses = args.loss as Array<string|LossOrMetricFn>;\n      lossFunctions = theLosses.map(l => losses.get(l));\n    } else {\n      const lossFunction = losses.get(args.loss);\n      this.outputs.forEach(_ => {\n        lossFunctions.push(lossFunction);\n      });\n    }\n\n    this.lossFunctions = lossFunctions;\n\n    this.feedOutputNames = [];\n    this.feedOutputShapes = [];\n    this.feedLossFns = [];\n    for (let i = 0; i < this.outputs.length; ++i) {\n      // TODO(cais): Logic for skipping target(s).\n      const shape = this.internalOutputShapes[i];\n      const name = this.outputNames[i];\n      this.feedOutputNames.push(name);\n      this.feedOutputShapes.push(shape);\n      this.feedLossFns.push(this.lossFunctions[i]);\n    }\n\n    // TODO(cais): Add logic for output masks.\n    // TODO(cais): Add logic for sample weights.\n    const skipTargetIndices: number[] = [];\n\n    // Prepare metrics.\n    this.metrics = args.metrics;\n    // TODO(cais): Add weightedMetrics.\n    this.metricsNames = ['loss'];\n    this.metricsTensors = [];\n\n    // Compute total loss.\n    // Porting Note: In PyKeras, metrics_tensors are symbolic tensor objects.\n    //   Here, metricsTensors are TypeScript functions. This difference is due\n    //   to the difference in symbolic/imperative property of the backends.\n    nameScope('loss', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        // TODO(cais): Add weightedLoss, sampleWeight and mask.\n        //   The following line should be weightedLoss\n        const weightedLoss = this.lossFunctions[i];\n        if (this.outputs.length > 1) {\n          this.metricsTensors.push([weightedLoss, i]);\n          this.metricsNames.push(this.outputNames[i] + '_loss');\n        }\n      }\n\n      // Porting Note: Due to the imperative nature of the backend, we calculate\n      //   the regularizer penalties in the totalLossFunction, instead of here.\n    });\n\n    const nestedMetrics = collectMetrics(args.metrics, this.outputNames);\n    // TODO(cais): Add nestedWeightedMetrics.\n\n    /**\n     * Helper function used in loop below.\n     */\n    const appendMetric =\n        (outputIndex: number, metricName: string,\n         metricTensor: LossOrMetricFn) => {\n          if (this.outputNames.length > 1) {\n            metricName = this.outputNames[outputIndex] + '_' + metricName;\n          }\n          this.metricsNames.push(metricName);\n          this.metricsTensors.push([metricTensor, outputIndex]);\n        };\n\n    nameScope('metric', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        const outputMetrics = nestedMetrics[i];\n        // TODO(cais): Add weights and outputWeightedMetrics.\n\n        // TODO(cais): Add optional arg `weights` to the following function.\n        const handleMetrics = (metrics: Array<string|LossOrMetricFn>) => {\n          const metricNamePrefix = '';\n          let metricName: string;\n          let accFn: LossOrMetricFn;\n          let weightedMetricFn: LossOrMetricFn;\n          //  TODO(cais): Use 'weights_' for weighted metrics.\n\n          for (const metric of metrics) {\n            if (typeof metric === 'string' &&\n                ['accuracy', 'acc', 'crossentropy', 'ce'].indexOf(metric) !==\n                    -1) {\n              const outputShape = this.internalOutputShapes[i];\n\n              if (outputShape[outputShape.length - 1] === 1 ||\n                  this.lossFunctions[i] === losses.binaryCrossentropy) {\n                // case: binary accuracy/crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryCrossentropy;\n                }\n              } else if (\n                  this.lossFunctions[i] ===\n                  losses.sparseCategoricalCrossentropy) {\n                // case: categorical accuracy / crossentropy with sparse\n                // targets.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalCrossentropy;\n                }\n              } else {\n                // case: categorical accuracy / crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalCrossentropy;\n                }\n              }\n              let suffix: string;\n              if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                suffix = 'acc';\n              } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                suffix = 'ce';\n              }\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = accFn;\n              metricName = metricNamePrefix + suffix;\n            } else {\n              const metricFn = Metrics.get(metric);\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = metricFn;\n              metricName =\n                  metricNamePrefix + Metrics.getLossOrMetricName(metric);\n            }\n\n            // TODO(cais): Add weighting and masking to metricResult.\n            let metricResult: LossOrMetricFn;\n            nameScope(metricName, () => {\n              metricResult = weightedMetricFn;\n            });\n            appendMetric(i, metricName, metricResult);\n          }\n        };\n\n        handleMetrics(outputMetrics);\n        // TODO(cais): Call handleMetrics with weights.\n      }\n    });\n\n    // Porting Notes: Given the imperative backend of tfjs-core,\n    //   there is no need for constructing the symbolic graph and placeholders.\n    this.collectedTrainableWeights = this.trainableWeights;\n  }\n\n  /**\n   * Check trainable weights count consistency.\n   *\n   * This will raise a warning if `this.trainableWeights` and\n   * `this.collectedTrainableWeights` are inconsistent (i.e., have different\n   * numbers of parameters).\n   * Inconsistency will typically arise when one modifies `model.trainable`\n   * without calling `model.compile()` again.\n   */\n  protected checkTrainableWeightsConsistency(): void {\n    if (this.collectedTrainableWeights == null) {\n      return;\n    }\n    if (this.trainableWeights.length !==\n        this.collectedTrainableWeights.length) {\n      console.warn(\n          'Discrepancy between trainableweights and collected trainable ' +\n          'weights. Did you set `model.trainable` without calling ' +\n          '`model.compile()` afterwards?');\n    }\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(\n   *     tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4});\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateArgs`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // TODO(cais): Standardize `config.sampleWeights` as well.\n    // Validate user data.\n    const checkBatchAxis = true;\n    const standardizedOuts =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    try {\n      // TODO(cais): If uses `useLearningPhase`, set the corresponding element\n      // of the input to 0.\n      const ins = standardizedOuts[0].concat(standardizedOuts[1]);\n      this.makeTestFunction();\n      const f = this.testFunction;\n      const testOuts =\n          this.testLoop(f, ins, batchSize, args.verbose, args.steps);\n      return singletonOrArray(testOuts);\n    } finally {\n      disposeNewTensors(standardizedOuts[0], x);\n      disposeNewTensors(standardizedOuts[1], y);\n    }\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args?: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    this.makeTestFunction();\n    return evaluateDataset(this, dataset, args);\n  }\n\n  /**\n   * Get number of samples provided for training, evaluation or prediction.\n   *\n   * @param ins Input `tf.Tensor`.\n   * @param batchSize Integer batch size, optional.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring loop finished. Optional.\n   * @param stepsName The public API's parameter name for `steps`.\n   * @returns Number of samples provided.\n   */\n  private checkNumSamples(\n      ins: Tensor|Tensor[], batchSize?: number, steps?: number,\n      stepsName = 'steps'): number {\n    let numSamples: number;\n    if (steps != null) {\n      numSamples = null;\n      if (batchSize != null) {\n        throw new ValueError(\n            `If ${stepsName} is set, batchSize must be null or undefined.` +\n            `Got batchSize = ${batchSize}`);\n      }\n    } else if (ins != null) {\n      if (Array.isArray(ins)) {\n        numSamples = ins[0].shape[0];\n      } else {\n        numSamples = ins.shape[0];\n      }\n    } else {\n      throw new ValueError(\n          `Either the input data should have a defined shape, or ` +\n          `${stepsName} shoud be specified.`);\n    }\n    return numSamples;\n  }\n\n  /**\n   * Execute internal tensors of the model with input data feed.\n   * @param inputs Input data feed. Must match the inputs of the model.\n   * @param outputs Names of the output tensors to be fetched. Must match\n   *   names of the SymbolicTensors that belong to the graph.\n   * @returns Fetched values for `outputs`.\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs: string|string[]):\n      Tensor|Tensor[] {\n    if (Array.isArray(outputs) && outputs.length === 0) {\n      throw new ValueError(\n          '`outputs` is an empty Array, which is not allowed.');\n    }\n\n    const outputsIsArray = Array.isArray(outputs);\n    const outputNames =\n        (outputsIsArray ? outputs as string[] : [outputs as string]);\n    const outputSymbolicTensors = this.retrieveSymbolicTensors(outputNames);\n\n    // Format the input into a FeedDict.\n    const feedDict = new FeedDict();\n    if (inputs instanceof Tensor) {\n      inputs = [inputs];\n    }\n    if (Array.isArray(inputs)) {\n      if (inputs.length !== this.inputs.length) {\n        throw new ValueError(\n            `The number of inputs provided (${inputs.length}) ` +\n            `does not match the number of inputs of this model ` +\n            `(${this.inputs.length}).`);\n      }\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n    } else {\n      for (const input of this.inputs) {\n        const tensorValue = inputs[input.name];\n        if (tensorValue == null) {\n          throw new ValueError(\n              `No value is provided for the model's input ${input.name}`);\n        }\n        feedDict.add(input, tensorValue);\n      }\n    }\n\n    // Run execution.\n    const executeOutputs = execute(outputSymbolicTensors, feedDict) as Tensor[];\n    return outputsIsArray ? executeOutputs : executeOutputs[0];\n  }\n\n  /**\n   * Retrieve the model's internal symbolic tensors from symbolic-tensor names.\n   */\n  private retrieveSymbolicTensors(symbolicTensorNames: string[]):\n      SymbolicTensor[] {\n    const outputSymbolicTensors: SymbolicTensor[] =\n        pyListRepeat(null, symbolicTensorNames.length);\n    let outputsRemaining = symbolicTensorNames.length;\n    for (const layer of this.layers) {\n      const layerOutputs: SymbolicTensor[] =\n          Array.isArray(layer.output) ? layer.output : [layer.output];\n      const layerOutputNames = layerOutputs.map(output => output.name);\n      for (let i = 0; i < symbolicTensorNames.length; ++i) {\n        const index = layerOutputNames.indexOf(symbolicTensorNames[i]);\n        if (index !== -1) {\n          outputSymbolicTensors[i] = layerOutputs[index];\n          outputsRemaining--;\n        }\n        if (outputsRemaining === 0) {\n          break;\n        }\n      }\n      if (outputsRemaining === 0) {\n        break;\n      }\n    }\n\n    if (outputsRemaining > 0) {\n      const remainingNames: string[] = [];\n      outputSymbolicTensors.forEach((tensor, i) => {\n        if (tensor == null) {\n          remainingNames.push(symbolicTensorNames[i]);\n        }\n      });\n      throw new ValueError(\n          `Cannot find SymbolicTensors for output name(s): ` +\n          `${JSON.stringify(remainingNames)}`);\n    }\n    return outputSymbolicTensors;\n  }\n\n  /**\n   * Helper method to loop over some data in batches.\n   *\n   * Porting Note: Not using the functional approach in the Python equivalent\n   *   due to the imperative backend.\n   * Porting Note: Does not support step mode currently.\n   *\n   * @param ins: input data\n   * @param batchSize: integer batch size.\n   * @param verbose: verbosity model\n   * @returns: Predictions as `tf.Tensor` (if a single output) or an `Array` of\n   *   `tf.Tensor` (if multipe outputs).\n   */\n  private predictLoop(ins: Tensor|Tensor[], batchSize = 32, verbose = false):\n      Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins);\n      if (verbose) {\n        throw new NotImplementedError(\n            'Verbose predictLoop() is not implemented yet.');\n      }\n\n      // Sample-based predictions.\n      // Porting Note: Tensor currently does not support sliced assignments as\n      //   in numpy, e.g., x[1:3] = y. Therefore we use concatenation while\n      //   iterating over the batches.\n\n      const batches = makeBatches(numSamples, batchSize);\n      const outsBatches: Tensor[][] = this.outputs.map(output => []);\n\n      // TODO(cais): Can the scope() be pushed down inside the for loop?\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchOuts = tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          // TODO(cais): Take care of the case of the last element is a flag for\n          //   training/test.\n          const insBatch = sliceArrays(ins, batchStart, batchEnd);\n\n          // Construct the feeds for execute();\n          const feeds = [];\n          if (Array.isArray(insBatch)) {\n            for (let i = 0; i < insBatch.length; ++i) {\n              feeds.push({key: this.inputs[i], value: insBatch[i]});\n            }\n          } else {\n            feeds.push({key: this.inputs[0], value: insBatch});\n          }\n          const feedDict = new FeedDict(feeds);\n          return execute(this.outputs, feedDict) as Tensor[];\n        });\n        batchOuts.forEach((batchOut, i) => outsBatches[i].push(batchOut));\n      }\n      return singletonOrArray(\n          outsBatches.map(batches => tfc.concat(batches, 0)));\n    });\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFlow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([8, 10]), {batchSize: 4}).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param args A `ModelPredictArgs` object containing optional fields.\n   *\n   * @return Prediction results as a `tf.Tensor`(s).\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    const xsRank2OrHigher = ensureTensorsRank2OrHigher(x);\n    checkInputData(\n        xsRank2OrHigher, this.inputNames, this.feedInputShapes, false);\n    try {\n      // TODO(cais): Take care of stateful models.\n      //   if (this.stateful) ...\n      // TODO(cais): Take care of the learning_phase boolean flag.\n      //   if (this.useLearningPhase) ...\n      const batchSize = args.batchSize == null ? 32 : args.batchSize;\n      checkBatchSize(batchSize);\n      return this.predictLoop(xsRank2OrHigher, batchSize);\n    } finally {\n      disposeNewTensors(xsRank2OrHigher, x);\n    }\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predictOnBatch(tf.ones([8, 10])).print();\n   * ```\n   * @param x: Input samples, as a Tensor (for models with exactly one\n   *   input) or an array of Tensors (for models with more than one input).\n   * @return Tensor(s) of predictions\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predictOnBatch(x: Tensor|Tensor[]): Tensor|Tensor[] {\n    checkInputData(x, this.inputNames, this.feedInputShapes, true);\n    // TODO(cais): Take care of the learning_phase boolean flag.\n    //   if (this.useLearningPhase) ...\n    const batchSize = (Array.isArray(x) ? x[0] : x).shape[0];\n    return this.predictLoop(x, batchSize);\n  }\n\n  protected standardizeUserDataXY(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor}, checkBatchAxis = true,\n      batchSize?: number): [Tensor[], Tensor[]] {\n    // TODO(cais): Add sampleWeight, classWeight\n    if (this.optimizer_ == null) {\n      throw new RuntimeError(\n          'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileArgs).');\n    }\n    const outputShapes: Shape[] = [];\n    for (let i = 0; i < this.feedOutputShapes.length; ++i) {\n      const outputShape = this.feedOutputShapes[i];\n      const lossFn = this.feedLossFns[i];\n      if (lossFn === losses.sparseCategoricalCrossentropy) {\n        outputShapes.push(\n            outputShape.slice(0, outputShape.length - 1).concat([1]));\n      } else {\n        // Porting Note: Because of strong typing `lossFn` must be a function.\n        outputShapes.push(outputShape);\n      }\n    }\n    x = standardizeInputData(\n        x, this.feedInputNames, this.feedInputShapes, false, 'input');\n    y = standardizeInputData(\n        y, this.feedOutputNames, outputShapes, false, 'target');\n    // TODO(cais): Standardize sampleWeights & classWeights.\n    checkArrayLengths(x, y, null);\n    // TODO(cais): Check sampleWeights as well.\n    checkLossAndTargetCompatibility(y, this.feedLossFns, this.feedOutputShapes);\n    if (this.stateful && batchSize != null && batchSize > 0) {\n      if (x[0].shape[0] % batchSize !== 0) {\n        throw new ValueError(\n            `In a stateful network, you should only pass inputs with a ` +\n            `number of samples that is divisible by the batch size ` +\n            `${batchSize}. Found: ${x[0].shape[0]} sample(s).`);\n      }\n    }\n    return [x, y];\n  }\n\n  protected async standardizeUserData(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      sampleWeight?: Tensor|Tensor[]|{[outputName: string]: Tensor},\n      classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap,\n      checkBatchAxis = true,\n      batchSize?: number): Promise<[Tensor[], Tensor[], Tensor[]]> {\n    const [standardXs, standardYs] =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    // TODO(cais): Handle sampleWeights.\n    if (sampleWeight != null) {\n      throw new Error('sample weight is not supported yet.');\n    }\n\n    let standardSampleWeights: Tensor[] = null;\n    if (classWeight != null) {\n      const classWeights =\n          standardizeClassWeights(classWeight, this.outputNames);\n      standardSampleWeights = [];\n      for (let i = 0; i < classWeights.length; ++i) {\n        standardSampleWeights.push(\n            await standardizeWeights(standardYs[i], null, classWeights[i]));\n      }\n    }\n\n    // TODO(cais): Deal with the case of model.stateful == true.\n    return [standardXs, standardYs, standardSampleWeights];\n  }\n\n  /**\n   * Loop over some test data in batches.\n   * @param f A Function returning a list of tensors.\n   * @param ins Array of tensors to be fed to `f`.\n   * @param batchSize Integer batch size or `null` / `undefined`.\n   * @param verbose verbosity mode.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring test finished. Ignored with the default value of `null` /\n   * `undefined`.\n   * @returns Array of Scalars.\n   */\n  private testLoop(\n      f: (data: Tensor[]) => Scalar[], ins: Tensor[], batchSize?: number,\n      verbose = 0, steps?: number): Scalar[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins, batchSize, steps, 'steps');\n      const outs: Scalar[] = [];\n      if (verbose > 0) {\n        throw new NotImplementedError('Verbose mode is not implemented yet.');\n      }\n      // TODO(cais): Use `indicesForConversionToDense' to prevent slow down.\n      if (steps != null) {\n        throw new NotImplementedError(\n            'steps mode in testLoop() is not implemented yet');\n      } else {\n        const batches = makeBatches(numSamples, batchSize);\n        const indexArray = tensor1d(range(0, numSamples));\n        for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds =\n              K.sliceAlongFirstAxis(\n                  indexArray, batchStart, batchEnd - batchStart) as Tensor1D;\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Scalar[];\n          const batchOuts = f(insBatch);\n          if (batchIndex === 0) {\n            for (let i = 0; i < batchOuts.length; ++i) {\n              outs.push(scalar(0));\n            }\n          }\n          for (let i = 0; i < batchOuts.length; ++i) {\n            const batchOut = batchOuts[i];\n            outs[i] =\n                tfc.add(outs[i], tfc.mul(batchEnd - batchStart, batchOut));\n          }\n        }\n        for (let i = 0; i < outs.length; ++i) {\n          outs[i] = tfc.div(outs[i], numSamples);\n        }\n      }\n      return outs;\n    });\n  }\n\n  protected getDedupedMetricsNames(): string[] {\n    const outLabels = this.metricsNames;\n    // Rename duplicated metrics names (can happen with an output layer\n    // shared among multiple dataflows).\n    const dedupedOutLabels = [];\n    for (let i = 0; i < outLabels.length; ++i) {\n      const label = outLabels[i];\n      let newLabel = label;\n      if (count(outLabels, label) > 1) {\n        const dupIndex = count(outLabels.slice(0, i), label);\n        newLabel += `_${dupIndex}`;\n      }\n      dedupedOutLabels.push(newLabel);\n    }\n    return dedupedOutLabels;\n  }\n\n  /**\n   * Creates a function that performs the following actions:\n   *\n   * 1. computes the losses\n   * 2. sums them to get the total loss\n   * 3. call the optimizer computes the gradients of the LayersModel's\n   *    trainable weights w.r.t. the total loss and update the variables\n   * 4. calculates the metrics\n   * 5. returns the values of the losses and metrics.\n   */\n  protected makeTrainFunction(): (data: Tensor[]) => Scalar[] {\n    return (data: Tensor[]) => {\n      const lossValues: Scalar[] = [];\n\n      const inputs = data.slice(0, this.inputs.length);\n      const targets = data.slice(\n          this.inputs.length, this.inputs.length + this.outputs.length);\n      const sampleWeights = data.slice(\n          this.inputs.length + this.outputs.length,\n          this.inputs.length + this.outputs.length * 2);\n\n      const metricsValues: Scalar[] = [];\n\n      // Create a function that computes the total loss based on the\n      // inputs. This function is used for obtaining gradients through\n      // backprop.\n      const totalLossFunction = () => {\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs =\n            execute(this.outputs, feedDict, {'training': true}) as Tensor[];\n        // TODO(cais): Take care of the case of multiple outputs from a\n        //   single layer?\n\n        let totalLoss: Tensor;\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          let loss = lossFunction(targets[i], outputs[i]);\n          if (sampleWeights[i] != null) {\n            loss = computeWeightedLoss(loss, sampleWeights[i]);\n          }\n\n          // TODO(cais): push Scalar instead.\n          const meanLoss: Scalar = tfc.mean(loss);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          lossValues.push(meanLoss);\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n        }\n\n        // Compute the metrics.\n        // TODO(cais): These should probably be calculated outside\n        //   totalLossFunction to benefit speed?\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          let weightedMetric: Scalar;\n\n          if (this.outputs.length > 1 && i < this.outputs.length) {\n            weightedMetric = lossValues[i];\n          } else {\n            const metric = this.metricsTensors[i][0];\n            const outputIndex = this.metricsTensors[i][1];\n            weightedMetric =\n                tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          }\n\n          tfc.keep(weightedMetric);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          metricsValues.push(weightedMetric);\n        }\n\n        totalLoss = tfc.mean(totalLoss);\n\n        // Add regularizer penalties.\n        this.calculateLosses().forEach(regularizerLoss => {\n          totalLoss = tfc.add(totalLoss, regularizerLoss);\n        });\n\n        return totalLoss as Scalar;\n      };\n\n      const variables = this.collectedTrainableWeights.map(\n          param => param.read() as tfc.Variable);\n      const returnCost = true;\n      const totalLossValue =\n          this.optimizer_.minimize(totalLossFunction, returnCost, variables);\n\n      return [totalLossValue].concat(metricsValues);\n    };\n  }\n\n  /**\n   * Create a function which, when invoked with an array of `tf.Tensor`s as a\n   * batch of inputs, returns the prespecified loss and metrics of the model\n   * under the batch of input data.\n   */\n  private makeTestFunction() {\n    this.testFunction = (data: Tensor[]) => {\n      return tfc.tidy(() => {\n        const valOutputs: Scalar[] = [];\n        let totalLoss: Scalar;\n        const inputs = data.slice(0, this.inputs.length);\n        const targets = data.slice(\n            this.inputs.length, this.inputs.length + this.outputs.length);\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs = execute(this.outputs, feedDict) as Tensor[];\n        // Compute total loss.\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          // TODO(cais): Add sample weighting and replace the simple\n          // averaging.\n          const loss: Scalar = tfc.mean(lossFunction(targets[i], outputs[i]));\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n          valOutputs.push(totalLoss);\n        }\n        // Compute the metrics.\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          const metric = this.metricsTensors[i][0];\n          const outputIndex = this.metricsTensors[i][1];\n          // TODO(cais): Replace K.mean() with a proper weighting function.\n          const meanMetric =\n              tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          valOutputs.push(meanMetric as Scalar);\n        }\n        return valOutputs;\n      });\n    };\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a\n   * dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *     layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * for (let i = 1; i < 5 ; ++i) {\n   *   const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *       batchSize: 4,\n   *       epochs: 3\n   *   });\n   *   console.log(\"Loss after Epoch \" + i + \" : \" + h.history.loss[0]);\n   * }\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you\n   * can also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named,\n   * you can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args A `ModelFitArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input\n   * data and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    return fitTensors(this, x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below when it's possible to instantiate\n  //   actual dataset objects.\n  /**\n   * Trains the model using a dataset object.\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for training. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs.\n   *   Of the two items in the array, the first is the input feature(s) and\n   *   the second is the output target(s).\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    return fitDataset(this, dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    // TODO(cais): Support sampleWeight and classWeight.\n    // TODO(cais): Support Dataset objects.\n    const standardizeOut = await this.standardizeUserData(x, y);\n    const inputs = standardizeOut[0];\n    const targets = standardizeOut[1];\n    const trainFunction = this.makeTrainFunction();\n    const losses = trainFunction(inputs.concat(targets));\n    const lossValues: number[] = [];\n    for (const loss of losses) {\n      const v = await loss.data();\n      lossValues.push(v[0]);\n    }\n    tfc.dispose(losses);\n    return singletonOrArray(lossValues);\n  }\n\n  /**\n   * Extract weight values of the model.\n   *\n   * @param config: An instance of `io.SaveConfig`, which specifies\n   * model-saving options such as whether only trainable weights are to be\n   * saved.\n   * @returns A `NamedTensorMap` mapping original weight names (i.e.,\n   *   non-uniqueified weight names) to their values.\n   */\n  protected getNamedWeights(config?: io.SaveConfig): NamedTensor[] {\n    const namedWeights: NamedTensor[] = [];\n\n    const trainableOnly = config != null && config.trainableOnly;\n    const weights = trainableOnly ? this.trainableWeights : this.weights;\n    const weightValues = this.getWeights(trainableOnly);\n    for (let i = 0; i < weights.length; ++i) {\n      if (trainableOnly && !weights[i].trainable) {\n        // Optionally skip non-trainable weights.\n        continue;\n      }\n      namedWeights.push(\n          {name: weights[i].originalName, tensor: weightValues[i]});\n    }\n    return namedWeights;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const input = tf.input({shape: [10]});\n   * const output = tf.layers.dense({units: 1}).apply(input);\n   * const model = tf.model({inputs: [input], outputs: [output]});\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10\n   * values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    this.stopTraining_ = stop;\n  }\n\n  get stopTraining(): boolean {\n    return this.stopTraining_;\n  }\n\n  get optimizer(): Optimizer {\n    return this.optimizer_;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    if (this.optimizer_ !== optimizer) {\n      this.optimizer_ = optimizer;\n      this.isOptimizerOwned = false;\n    }\n  }\n\n  dispose(): DisposeResult {\n    const result = super.dispose();\n    if (result.refCountAfterDispose === 0 && this.optimizer != null &&\n        this.isOptimizerOwned) {\n      const numTensorsBeforeOptmizerDisposal = tfc.memory().numTensors;\n      this.optimizer_.dispose();\n      result.numDisposedVariables +=\n          numTensorsBeforeOptmizerDisposal - tfc.memory().numTensors;\n    }\n    return result;\n  }\n\n  private getLossIdentifiers(): LossIdentifier|LossIdentifier[]|\n      {[outputName: string]: LossIdentifier} {\n    let lossNames: LossIdentifier|LossIdentifier[]|\n        {[outputName: string]: LossIdentifier};\n    if (typeof this.loss === 'string') {\n      lossNames = toSnakeCase(this.loss) as LossIdentifier;\n    } else if (Array.isArray(this.loss)) {\n      for (const loss of this.loss) {\n        if (typeof loss !== 'string') {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n      lossNames = (this.loss as string[]).map(name => toSnakeCase(name)) as\n          LossIdentifier[];\n    } else {\n      const outputNames = Object.keys(this.loss);\n      lossNames = {} as {[outputName: string]: LossIdentifier};\n      const losses =\n          this.loss as {[outputName: string]: LossOrMetricFn | string};\n      for (const outputName of outputNames) {\n        if (typeof losses[outputName] === 'string') {\n          lossNames[outputName] =\n              toSnakeCase(losses[outputName] as string) as LossIdentifier;\n        } else {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n    }\n    return lossNames;\n  }\n\n  private getMetricIdentifiers(): MetricsIdentifier[]|\n      {[key: string]: MetricsIdentifier} {\n    if (typeof this.metrics === 'string' ||\n        typeof this.metrics === 'function') {\n      return [toSnakeCase(Metrics.getLossOrMetricName(this.metrics))];\n    } else if (Array.isArray(this.metrics)) {\n      return this.metrics.map(\n          metric => toSnakeCase(Metrics.getLossOrMetricName(metric)));\n    } else {\n      const metricsIdentifiers: {[key: string]: MetricsIdentifier} = {};\n      for (const key in this.metrics) {\n        metricsIdentifiers[key] =\n            toSnakeCase(Metrics.getLossOrMetricName(this.metrics[key]));\n      }\n      return metricsIdentifiers;\n    }\n  }\n\n  protected getTrainingConfig(): TrainingConfig {\n    return {\n      loss: this.getLossIdentifiers(),\n      metrics: this.getMetricIdentifiers(),\n      optimizer_config: {\n        class_name: this.optimizer.getClassName(),\n        config: this.optimizer.getConfig()\n      } as OptimizerSerialization\n    };\n    // TODO(cais): Add weight_metrics when they are supported.\n    // TODO(cais): Add sample_weight_mode when it's supported.\n    // TODO(cais): Add loss_weights when it's supported.\n  }\n\n  loadTrainingConfig(trainingConfig: TrainingConfig) {\n    if (trainingConfig.weighted_metrics != null) {\n      throw new Error('Loading weight_metrics is not supported yet.');\n    }\n    if (trainingConfig.loss_weights != null) {\n      throw new Error('Loading loss_weights is not supported yet.');\n    }\n    if (trainingConfig.sample_weight_mode != null) {\n      throw new Error('Loading sample_weight_mode is not supported yet.');\n    }\n\n    const tsConfig = convertPythonicToTs(trainingConfig.optimizer_config) as\n        serialization.ConfigDict;\n    const optimizer = deserialize(tsConfig) as Optimizer;\n\n    let loss;\n    if (typeof trainingConfig.loss === 'string') {\n      loss = toCamelCase(trainingConfig.loss);\n    } else if (Array.isArray(trainingConfig.loss)) {\n      loss = trainingConfig.loss.map(lossEntry => toCamelCase(lossEntry));\n    } else if (trainingConfig.loss != null) {\n      loss = {} as {[outputName: string]: LossIdentifier};\n      for (const key in trainingConfig.loss) {\n        loss[key] = toCamelCase(trainingConfig.loss[key]) as LossIdentifier;\n      }\n    }\n\n    let metrics;\n    if (Array.isArray(trainingConfig.metrics)) {\n      metrics = trainingConfig.metrics.map(metric => toCamelCase(metric));\n    } else if (trainingConfig.metrics != null) {\n      metrics = {} as {[outputName: string]: MetricsIdentifier};\n      for (const key in trainingConfig.metrics) {\n        metrics[key] = toCamelCase(trainingConfig.metrics[key]);\n      }\n    }\n\n    this.compile({loss, metrics, optimizer});\n  }\n\n  /**\n   * Save the configuration and/or weights of the LayersModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 2. Saving `model`'s topology and weights to browser\n   * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('indexeddb://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 3. Saving `model`'s topology and weights as two files\n   * (`my-model-1.json` and `my-model-1.weights.bin`) downloaded from\n   * browser.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('downloads://my-model-1');\n   * ```\n   *\n   * Example 4. Send  `model`'s topology and weights to an HTTP server.\n   * See the documentation of `tf.io.http` for more details\n   * including specifying request parameters and implementation of the\n   * server.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('http://my-server/model/upload');\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new ValueError(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new ValueError(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new ValueError(\n          'LayersModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    const weightDataAndSpecs =\n        await io.encodeWeights(this.getNamedWeights(config));\n\n    const returnString = false;\n    const unusedArg: {} = null;\n    const modelConfig = this.toJSON(unusedArg, returnString);\n    const modelArtifacts: io.ModelArtifacts = {\n      modelTopology: modelConfig,\n      format: LAYERS_MODEL_FORMAT_NAME,\n      generatedBy: `TensorFlow.js tfjs-layers v${version}`,\n      convertedBy: null,\n    };\n\n    const includeOptimizer = config == null ? false : config.includeOptimizer;\n    if (includeOptimizer && this.optimizer != null) {\n      modelArtifacts.trainingConfig = this.getTrainingConfig();\n      const weightType = 'optimizer';\n      const {data: optimizerWeightData, specs: optimizerWeightSpecs} =\n          await io.encodeWeights(await this.optimizer.getWeights(), weightType);\n      weightDataAndSpecs.specs.push(...optimizerWeightSpecs);\n      weightDataAndSpecs.data = io.concatenateArrayBuffers(\n          [weightDataAndSpecs.data, optimizerWeightData]);\n    }\n\n    if (this.userDefinedMetadata != null) {\n      // Check serialized size of user-defined metadata.\n      const checkSize = true;\n      checkUserDefinedMetadata(this.userDefinedMetadata, this.name, checkSize);\n      modelArtifacts.userDefinedMetadata = this.userDefinedMetadata;\n    }\n\n    modelArtifacts.weightData = weightDataAndSpecs.data;\n    modelArtifacts.weightSpecs = weightDataAndSpecs.specs;\n    return handlerOrURL.save(modelArtifacts);\n  }\n\n  /**\n   * Set user-defined metadata.\n   *\n   * The set metadata will be serialized together with the topology\n   * and weights of the model during `save()` calls.\n   *\n   * @param setUserDefinedMetadata\n   */\n  setUserDefinedMetadata(userDefinedMetadata: {}): void {\n    checkUserDefinedMetadata(userDefinedMetadata, this.name);\n    this.userDefinedMetadata = userDefinedMetadata;\n  }\n\n  /**\n   * Get user-defined metadata.\n   *\n   * The metadata is supplied via one of the two routes:\n   *   1. By calling `setUserDefinedMetadata()`.\n   *   2. Loaded during model loading (if the model is constructed\n   *      via `tf.loadLayersModel()`.)\n   *\n   * If no user-defined metadata is available from either of the\n   * two routes, this function will return `undefined`.\n   */\n  getUserDefinedMetadata(): {} {\n    return this.userDefinedMetadata;\n  }\n}\nserialization.registerClass(LayersModel);\n\n/**\n * A `tf.Functional` is an alias to `tf.LayersModel`.\n *\n * See also:\n *   `tf.LayersModel`, `tf.Sequential`, `tf.loadLayersModel`.\n */\n/** @doc {heading: 'Models', subheading: 'Classes'} */\nexport class Functional extends LayersModel {\n  static className = 'Functional';\n}\nserialization.registerClass(Functional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source keras/models.py */\n\nimport {dispose, io, NamedTensorMap, Optimizer, Scalar, serialization, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {getUid} from './backend/state';\nimport {History} from './base_callbacks';\nimport {Dataset} from './engine/dataset_stub';\nimport {Input} from './engine/input_layer';\nimport {getSourceInputs, Layer, Node, SymbolicTensor} from './engine/topology';\nimport {LayersModel, ModelCompileArgs, ModelEvaluateArgs} from './engine/training';\nimport {ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './engine/training_dataset';\nimport {ModelFitArgs} from './engine/training_tensors';\nimport {NotImplementedError, RuntimeError, ValueError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {TrainingConfig} from './keras_format/training_config';\nimport {PyJsonDict} from './keras_format/types';\nimport {deserialize} from './layers/serialization';\nimport {Kwargs, NamedTensor} from './types';\nimport * as generic_utils from './utils/generic_utils';\nimport {convertPythonicToTs} from './utils/serialization_utils';\nimport {getExactlyOneShape} from './utils/types_utils';\n\n/**\n * Parses a JSON model configuration file and returns a model instance.\n *\n * ```js\n * // This example shows how to serialize a model using `toJSON()` and\n * // deserialize it as another model using `tf.models.modelFromJSON()`.\n * // Note: this example serializes and deserializes only the topology\n * // of the model; the weights of the loaded model will be different\n * // from those of the the original model, due to random weight\n * // initialization.\n * // To load the topology and weights of a model, use `tf.loadLayersModel()`.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.repeatVector({inputShape: [2], n: 4}));\n * // Serialize `model1` as a JSON object.\n * const model1JSON = model1.toJSON(null, false);\n * model1.summary();\n *\n * const model2 = await tf.models.modelFromJSON(model1JSON);\n * model2.summary();\n * ```\n *\n *  @param modelAndWeightsConfig JSON object or string encoding a model and\n *       weights configuration. It can also be only the topology JSON of the\n *       model, in which case the weights will not be loaded.\n *  @param custom_objects Optional dictionary mapping names\n *       (strings) to custom classes or functions to be\n *       considered during deserialization.\n * @returns A TensorFlow.js Layers `tf.LayersModel` instance (uncompiled).\n */\nexport async function modelFromJSON(\n    modelAndWeightsConfig: ModelAndWeightsConfig|PyJsonDict,\n    customObjects?: serialization.ConfigDict): Promise<LayersModel> {\n  if (!('modelTopology' in modelAndWeightsConfig)) {\n    modelAndWeightsConfig = {modelTopology: modelAndWeightsConfig};\n  }\n  modelAndWeightsConfig = modelAndWeightsConfig as ModelAndWeightsConfig;\n\n  let modelTopology = modelAndWeightsConfig.modelTopology;\n  if (modelTopology['model_config'] != null) {\n    // If the model-topology JSON contains a 'model_config' field, then it is\n    // a full model JSON (e.g., from `keras.Model.save()`), which contains\n    // not only the model's architecture in its 'model_config' field, but\n    // additional information such as the model's optimizer. We use only the\n    // 'model_config' field currently.\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n  const tsConfig =\n      convertPythonicToTs(modelTopology) as serialization.ConfigDict;\n  const model = deserialize(tsConfig, customObjects) as LayersModel;\n\n  if (modelAndWeightsConfig.weightsManifest != null) {\n    // Load the weight values keyed by the original tensor names in the model\n    // file that was loaded.  These should match the keys of the weight\n    // manifest.\n    const weightValues = await io.loadWeights(\n        modelAndWeightsConfig.weightsManifest, modelAndWeightsConfig.pathPrefix,\n        model.weights.map(weight => weight.originalName));\n\n    // Map the weights to the unique tensor names generated during model loading\n    const uniqueWeightValues: NamedTensorMap = {};\n    for (const weight of model.weights) {\n      uniqueWeightValues[weight.originalName] =\n          weightValues[weight.originalName];\n    }\n\n    model.loadWeights(uniqueWeightValues);\n    // Dispose temporary weight values.\n    dispose(weightValues);\n  }\n  return model;\n}\n\n/**\n * Options for loading a saved mode in TensorFlow.js format.\n */\nexport interface ModelAndWeightsConfig {\n  /**\n   * A JSON object or JSON string containing the model config.\n   *\n   * This can be either of the following two formats:\n   *   - A model archiecture-only config,  i.e., a format consistent with the\n   *     return value of`keras.Model.to_json()`.\n   *   - A full model config, containing not only model architecture, but also\n   *     training options and state, i.e., a format consistent with the return\n   *     value of `keras.models.save_model()`.\n   */\n  modelTopology: PyJsonDict;\n\n  /**\n   * A weights manifest in TensorFlow.js format.\n   */\n  weightsManifest?: io.WeightsManifestConfig;\n\n  /**\n   * Path to prepend to the paths in `weightManifest` before fetching.\n   *\n   * The path may optionally end in a slash ('/').\n   */\n  pathPrefix?: string;\n}\n\n// TODO(nielsene): Remove after: https://github.com/tensorflow/tfjs/issues/400\nexport interface ModelPredictArgs {\n  /**\n   * Optional. Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Optional. Verbosity mode. Defaults to false.\n   */\n  verbose?: boolean;\n}\n\n/**\n * Load a model, including its topology and optionally weights.  See the\n * Tutorial named \"How to import a Keras Model\" for usage examples.\n *\n * Example 1: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 2. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * Example 4. Load a model from an HTTP server.\n *\n * ```js\n * const model = await\n *     tf.loadLayersModel('https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. This path will be\n *      interpreted as a relative HTTP path, to which `fetch` will be used to\n *      request the model topology and weight manifest JSON.\n *      The content of the JSON file is assumed to be a JSON object with the\n *      following fields and values:\n *      - 'modelTopology': A JSON object that can be either of:\n *        1. a model architecture JSON consistent with the format of the return\n *            value of `keras.Model.to_json()`\n *        2. a full model JSON in the format of `keras.models.save_model()`.\n *      - 'weightsManifest': A TensorFlow.js weights manifest.\n *      See the Python converter function `save_model()` for more details.\n *      It is also assumed that model weights can be accessed from relative\n *      paths described by the `paths` fields in weights manifest.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A progress callback of the form:\n *     `(fraction: number) => void`. This callback can be used to monitor the\n *     model-loading process.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n */\nexport async function loadLayersModelInternal(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (typeof pathOrIOHandler === 'string') {\n    const handlers = io.getLoadHandlers(pathOrIOHandler, options);\n    if (handlers.length === 0) {\n      // For backward compatibility: if no load handler can be found,\n      // assume it is a relative http path.\n      // TODO(cais): Reformat the args into a single `LoadOptions` once the core\n      // is refactored.\n      handlers.push(io.browserHTTPRequest(pathOrIOHandler, options));\n    } else if (handlers.length > 1) {\n      throw new ValueError(\n          `Found more than one (${handlers.length}) load handlers for ` +\n          `URL '${pathOrIOHandler}'`);\n    }\n    pathOrIOHandler = handlers[0];\n  }\n  return loadLayersModelFromIOHandler(pathOrIOHandler, undefined, options);\n}\n\n/**\n * Load a model and optionally its weights, using an IOHandler object.\n *\n * @param handler The instance of `IOHandler` to be used during the model\n *   loading.\n * @param customObjects Any optional custom objects to be used during model\n *   loading.\n * @param strict Whether the weight loading will be done in strict mode.\n *   Default: `true`.\n */\nexport async function loadLayersModelFromIOHandler(\n    handler: io.IOHandler, customObjects?: serialization.ConfigDict,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (handler.load == null) {\n    throw new ValueError(\n        'Cannot proceed with model loading because the IOHandler provided ' +\n        'does not have the `load` method implemented.');\n  }\n  const artifacts = await handler.load();\n  let modelTopology = artifacts.modelTopology as PyJsonDict;\n  if (modelTopology['model_config'] != null) {\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n\n  const strict = options.strict == null ? true : options.strict;\n  // If weights are provided and the weight-loading mode is strict, use\n  // fast weight initialization. This skips costly initializers such as\n  // 'orthogonal' and saves unnecessary computation in cases where\n  // the initialized weight values will immediately be overwritten by\n  // loaded weight values.\n  const fastWeightInit =\n      artifacts.weightData != null && artifacts.weightSpecs != null && strict;\n  const model =\n      deserialize(\n          convertPythonicToTs(modelTopology) as serialization.ConfigDict,\n          customObjects, fastWeightInit) as LayersModel;\n\n  const trainingConfig = artifacts.trainingConfig as TrainingConfig;\n  if (trainingConfig != null) {\n    model.loadTrainingConfig(trainingConfig);\n  }\n  if (artifacts.userDefinedMetadata != null) {\n    model.setUserDefinedMetadata(artifacts.userDefinedMetadata);\n  }\n\n  // If weightData is present, load the weights into the model.\n  if (artifacts.weightData != null) {\n    // Loading weights requires weightSpecs.\n    if (artifacts.weightSpecs == null) {\n      throw new ValueError(\n          'LayersModel artifacts contains weight data, but not weight specs. ' +\n          'Therefore loading of weights cannot proceed.');\n    }\n\n    const {modelWeights, optimizerWeights} = decodeModelAndOptimizerWeights(\n        artifacts.weightData, artifacts.weightSpecs);\n    model.loadWeights(modelWeights, strict);\n\n    if (model.optimizer != null && optimizerWeights.length > 0) {\n      await model.optimizer.setWeights(optimizerWeights);\n    }\n\n    // Dispose temporary weight values.\n    dispose(modelWeights);\n    dispose(optimizerWeights.map(w => w.tensor));\n  }\n  return model;\n}\n\nfunction decodeModelAndOptimizerWeights(\n    buffer: ArrayBuffer, specs: io.WeightsManifestEntry[]):\n    {modelWeights: NamedTensorMap, optimizerWeights: NamedTensor[]} {\n  const name2Tensor = io.decodeWeights(buffer, specs);\n  const modelWeights: NamedTensorMap = {};\n  const optimizerWeights: NamedTensor[] = [];\n  specs.forEach(spec => {\n    if (spec.group === 'optimizer') {\n      optimizerWeights.push({name: spec.name, tensor: name2Tensor[spec.name]});\n    } else {\n      modelWeights[spec.name] = name2Tensor[spec.name];\n    }\n  });\n  return {modelWeights, optimizerWeights};\n}\n\n/**\n * Configuration for a Sequential model.\n */\nexport interface SequentialArgs {\n  /** Stack of layers for the model. */\n  layers?: Layer[];\n\n  /** The name of this model. */\n  name?: string;\n}\n\n/**\n * A model with a stack of layers, feeding linearly from one to the next.\n *\n * `tf.sequential` is a factory function that creates an instance of\n * `tf.Sequential`.\n *\n * ```js\n *  // Define a model for linear regression.\n *  const model = tf.sequential();\n *  model.add(tf.layers.dense({units: 1, inputShape: [1]}));\n *\n *  // Prepare the model for training: Specify the loss and the optimizer.\n *  model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n *\n *  // Generate some synthetic data for training.\n *  const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);\n *  const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);\n *\n *  // Train the model using the data then do inference on a data point the\n *  // model hasn't seen:\n *  await model.fit(xs, ys);\n *  model.predict(tf.tensor2d([5], [1, 1])).print();\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class Sequential extends LayersModel {\n  /** @nocollapse */\n  static className = 'Sequential';\n  private model: LayersModel;\n  constructor(args?: SequentialArgs) {\n    super({inputs: [], outputs: []});\n    args = args || {};\n\n    this.trainable = true;\n    this.built = false;\n\n    // Set model name.\n    this.name = (args.name != null) ? args.name : getUid('sequential_');\n\n    // Add to the model any layers passed to the constructor.\n    if (args.layers != null) {\n      for (const layer of args.layers) {\n        this.add(layer);\n      }\n    }\n  }\n\n  // Helper function to Sequential.add  Throws if the new output shape will be\n  // invalid.\n  private checkShape(layer: Layer) {\n    const shape = layer.inboundNodes[0].outputTensors[0].shape;\n    if (shape.some(x => x < 0)) {\n      throw new ValueError(\n          'Negative dimension size caused by adding layer ' +\n          `${layer.name} with input shape [` +\n          `${layer.inboundNodes[0].inputTensors[0].shape}]`);\n    }\n  }\n\n  /**\n   * Adds a layer instance on top of the layer stack.\n   *\n   * ```js\n   *  const model = tf.sequential();\n   *  model.add(tf.layers.dense({units: 8, inputShape: [1]}));\n   *  model.add(tf.layers.dense({units: 4, activation: 'relu6'}));\n   *  model.add(tf.layers.dense({units: 1, activation: 'relu6'}));\n   *  // Note that the untrained model is random at this point.\n   *  model.predict(tf.randomNormal([10, 1])).print();\n   * ```\n   * @param layer Layer instance.\n   *\n   * @exception ValueError In case the `layer` argument does not know its\n   * input shape.\n   * @exception ValueError In case the `layer` argument has multiple output\n   *   tensors, or is already connected somewhere else (forbidden in\n   *   `Sequential` models).\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  add(layer: Layer): void {\n    const isLayerModelInstance =\n        layer instanceof Sequential || layer instanceof LayersModel;\n    let modelLayer: LayersModel;\n    if (isLayerModelInstance) {\n      modelLayer = layer as LayersModel;\n      if (modelLayer.outputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      if (modelLayer.inputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single input tensor. ' +\n            'For multi-input layers, ' +\n            'use the functional API.');\n      }\n    }\n\n    if (this.outputs.length === 0) {\n      // first layer in model: check that it is an input layer\n      if (layer.inboundNodes.length === 0) {\n        // create an input layer\n        if (layer.batchInputShape == null) {\n          throw new ValueError(\n              'The first layer in a Sequential model must ' +\n              'get an `inputShape` or `batchInputShape` argument.');\n        }\n        // Instantiate the input layer.\n        const x = Input({\n          batchShape: layer.batchInputShape,\n          dtype: layer.dtype,\n          name: layer.name + '_input'\n        });\n        // This will build the current layer and create the node connecting\n        // the current layer to the input layer we just created.\n        layer.apply(x);\n      }\n\n      if (isLayerModelInstance) {\n        this.outputs = modelLayer.outputs;\n        this.inputs = modelLayer.inputs;\n      } else {\n        if (layer.inboundNodes.length !== 1) {\n          throw new ValueError(\n              'A layer added to a Sequential model must not already be ' +\n              `connected somewhere else. LayersModel received layer ${\n                  layer.name} ` +\n              `which has ${layer.inboundNodes.length} pre-existing inbound ` +\n              'connections.');\n        }\n\n        if (layer.inboundNodes[0].outputTensors.length !== 1) {\n          throw new ValueError(\n              'All layers in a Sequential model ' +\n              'should have a single output tensor. ' +\n              'For multi-output layers, ' +\n              'use the functional API.');\n        }\n        this.checkShape(layer);\n        this.outputs = [layer.inboundNodes[0].outputTensors[0]];\n        this.inputs = getSourceInputs(this.outputs[0]);\n      }\n\n      this.inboundNodes = [];\n      // We create an input node, which we will keep updated\n      // as we add more layers.\n      // (This call has side effects.)\n      // tslint:disable-next-line:no-unused-expression\n      new Node({\n        outboundLayer: this,\n        inboundLayers: [],\n        nodeIndices: [],\n        tensorIndices: [],\n        inputTensors: this.inputs,\n        outputTensors: this.outputs,\n        // no model-level masking for now\n        inputMasks: generic_utils.pyListRepeat(null, this.inputs.length),\n        outputMasks: [null],\n        inputShapes: this.inputs.map(x => x.shape),\n        outputShapes: this.outputs[0].shape\n      });\n    } else {\n      const outputTensor = layer.apply(this.outputs[0]);\n      if (Array.isArray(outputTensor)) {\n        throw new TypeError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      this.checkShape(layer);\n      this.outputs = [outputTensor as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n\n    this.layers.push(layer);\n    this.built = false;\n  }\n\n  /**\n   * Removes the last layer in the model.\n   *\n   * @exception TypeError if there are no layers in the model.\n   */\n  pop(): void {\n    if (this.layers.length === 0) {\n      throw new TypeError('There are no layers in the model.');\n    }\n\n    this.layers.pop();\n    if (this.layers.length === 0) {\n      this.outputs = [];\n      this.inboundNodes = [];\n      this.outboundNodes = [];\n    } else {\n      const lastLayerIndex = this.layers.length - 1;\n      this.layers[lastLayerIndex].outboundNodes = [];\n      this.outputs = [this.layers[lastLayerIndex].output as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.call(inputs, kwargs);\n  }\n\n  build(inputShape?: Shape|Shape[]) {\n    // Call `getExactlyOneShape` without using its return value,\n    // to verify that exactly one input shape is provided.\n    getExactlyOneShape(inputShape);\n\n    if (this.inputs.length === 0 || this.outputs.length === 0) {\n      throw new TypeError(\n          'Sequential model cannot be built: model is empty.' +\n          ' Add some layers first.');\n    }\n    // actually create the model\n    this.model = new LayersModel({\n      inputs: this.inputs,\n      outputs: this.outputs[0],\n      name: this.name + '_model'\n    });\n    this.model.trainable = this.trainable;\n\n    // mirror model attributes\n    this.supportsMasking = this.model.supportsMasking;\n    // TODO(michaelterry): Add caches\n    this.inputLayers = this.model.inputLayers;\n    this.inputLayersNodeIndices = this.model.inputLayersNodeIndices;\n    this.inputLayersTensorIndices = this.model.inputLayersTensorIndices;\n    this.outputLayers = this.model.outputLayers;\n    this.outputLayersNodeIndices = this.model.outputLayersNodeIndices;\n    this.outputLayersTensorIndices = this.model.outputLayersTensorIndices;\n    this.nodesByDepth = this.model.nodesByDepth;\n    this.containerNodes = this.model.containerNodes;\n    this.outputNames = this.model.outputNames;\n    this.inputNames = this.model.inputNames;\n    // TODO(michaelterry): Add feedInputNames, feedInputs, if needed.\n    // TODO(michaelterry): Add callbackModel if needed.\n    this.built = true;\n  }\n\n  countParams(): number {\n    if (!this.built) {\n      this.build();\n    }\n    return super.countParams();\n  }\n\n  /**\n   * Print a text summary of the Sequential model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - The total number of trainable and non-trainable parameters of the\n   * model.\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(\n   *     tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'}));\n   * model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n   *\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      this.build();\n    }\n    super.summary(lineLength, positions, printFn);\n  }\n\n  /**\n   * Sets the weights of the model.\n   *\n   * @param weights Should be a list of Tensors with shapes and types matching\n   *   the output of `model.getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    if (this.model == null) {\n      this.build();\n    }\n    this.model.setWeights(weights);\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   * });\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateConfig`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluate(x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluateDataset(dataset, args);\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([2, 10])).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param conifg A `ModelPredictConfig` object containing optional fields.\n   *\n   * @return `tf.Tensor`(s) of predictions.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predict(x, args);\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * @param x: Input samples, as a Tensor, or list of Tensors (if the model\n   *   has multiple inputs).\n   * @return Tensor(s) of predictions\n   */\n  predictOnBatch(x: Tensor): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predictOnBatch(x);\n  }\n\n  /**\n   * See `LayersModel.compile`.\n   *\n   * @param args\n   */\n  compile(args: ModelCompileArgs): void {\n    this.build();\n    this.model.compile(args);\n    this.optimizer_ = this.model.optimizer;\n    // tslint:disable-next-line:no-any\n    this.isOptimizerOwned = (this.model as any).isOptimizerOwned;\n    this.loss = this.model.loss;\n    this.metrics = this.model.metrics;\n    // TODO(cais): Add this.lossWeights, this.sampleWeightMode,\n    //   this.weightedMetrics, this.targets.\n    this.metricsTensors = this.model.metricsTensors;\n    this.metricsNames = this.model.metricsNames;\n    // TODO(cais): Add sampleWeights.\n  }\n\n  get optimizer(): Optimizer {\n    return this.model == null ? undefined : this.model.optimizer;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    this.model.optimizer = optimizer;\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   *   epochs: 3\n   * });\n   * console.log(history.history.loss[0]);\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you can\n   * also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named, you\n   *  can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args  A `ModelFitConfig`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fit(x, y, args);\n  }\n\n  /**\n   * Trains the model using a dataset object.\n   *\n   * ```js\n   * const xArray = [\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   * ];\n   * const yArray = [1, 1, 1, 1];\n   * // Create a dataset from the JavaScript array.\n   * const xDataset = tf.data.array(xArray);\n   * const yDataset = tf.data.array(yArray);\n   * // Zip combines the `x` and `y` Datasets into a single Dataset, the\n   * // iterator of which will return an object containing of two tensors,\n   * // corresponding to `x` and `y`.  The call to `batch(4)` will bundle\n   * // four such samples into a single object, with the same keys now pointing\n   * // to tensors that hold 4 examples, organized along the batch dimension.\n   * // The call to `shuffle(4)` causes each iteration through the dataset to\n   * // happen in a different order.  The size of the shuffle window is 4.\n   * const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset})\n   *     .batch(4)\n   *     .shuffle(4);\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [9]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fitDataset(xyDataset, {\n   *   epochs: 4,\n   *   callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)}\n   * });\n   * ```\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected to\n   *   generate a dataset iterator object, the `next()` method of which is\n   *   expected to produce data batches for evaluation. The return value of the\n   *   `next()` call ought to contain a boolean `done` field and a `value`\n   *   field.\n   *\n   *   The `value` field is expected to be an object of with fields\n   *   `xs` and `ys`, which point to the feature tensor and the target tensor,\n   *   respectively. This case is for models with exactly one input and one\n   *   output (e.g.. a sequential model). For example:\n   *   ```js\n   *   {value: {xs: xsTensor, ys: ysTensor}, done: false}\n   *   ```\n   *\n   *   If the model has multiple inputs, the `xs` field of `value` should\n   *   be an object mapping input names to their respective feature tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: {\n   *         input_1: xsTensor1,\n   *         input_2: xsTensor2\n   *       },\n   *       ys: ysTensor\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   *   If the model has multiple outputs, the `ys` field of `value` should\n   *   be an object mapping output names to their respective target tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: xsTensor,\n   *       ys: {\n   *         output_1: ysTensor1,\n   *         output_2: ysTensor2\n   *       },\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fitDataset(dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    return this.model.trainOnBatch(x, y);\n  }\n\n  /* See parent class for JsDoc */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    let configArray: serialization.ConfigDictArray;\n    let extraModelConfig: serialization.ConfigDict = {};\n    if (config instanceof Array) {\n      if (!(config[0].className != null) ||\n          config[0]['className'] === 'Merge') {\n        throw new ValueError('Legacy serialization format not supported yet.');\n      }\n      configArray = config;\n    } else {\n      util.assert(\n          config['layers'] != null,\n          () =>\n              `When the config data for a Sequential model is not an Array, ` +\n              `it must be an Object that contains the 'layers' field.`);\n      configArray = config['layers'] as serialization.ConfigDictArray;\n      delete config['layers'];\n      extraModelConfig = config;\n    }\n\n    const model = new cls(extraModelConfig);\n    if (!(model instanceof Sequential)) {\n      throw new NotImplementedError(\n          `Sequential.fromConfig called on non-Sequential input: ${model}`);\n    }\n    for (const conf of configArray) {\n      const customObjects: serialization.ConfigDict = undefined;\n      const layer = deserialize(\n                        conf as serialization.ConfigDict, customObjects,\n                        fastWeightInit) as Layer;\n      if (fastWeightInit) {\n        layer.setFastWeightInitDuringBuild(true);\n      }\n      model.add(layer);\n    }\n    return model;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(tf.layers.dense({units: 1, inputShape: [10]}));\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10 values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    // TODO(cais): When refactoring to remove the composition pattern happens,\n    // remove this method overriding.\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot set the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    this.model.stopTraining = stop;\n  }\n\n  get stopTraining(): boolean {\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot get the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    return this.model.stopTraining;\n  }\n\n  // TODO(cais): Override get trainableWeights() here\n\n  // tslint:disable-next-line:no-any\n  getConfig(): any {\n    // NOTE(cais): We override the return type of getConfig() to `any` here,\n    //   because the `Sequential` class is a special case among `Container`\n    //   subtypes in that its getConfig() method returns an Array (not a\n    //   dict).\n    const layers: serialization.ConfigDict[] = [];\n    for (const layer of this.layers) {\n      const dict: serialization.ConfigDict = {};\n      dict['className'] = layer.getClassName();\n      dict['config'] = layer.getConfig();\n      layers.push(dict);\n    }\n    return {name: this.name, layers};\n  }\n}\nserialization.registerClass(Sequential);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Exported functions.\n */\n\nimport {io} from '@tensorflow/tfjs-core';\n\nimport {BaseCallbackConstructor, CallbackConstructorRegistry} from './base_callbacks';\nimport {ContainerArgs} from './engine/container';\nimport {Input, InputConfig,} from './engine/input_layer';\nimport {SymbolicTensor} from './engine/topology';\nimport {LayersModel} from './engine/training';\nimport {loadLayersModelInternal, Sequential, SequentialArgs} from './models';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// LayersModel and related factory methods.\n\n/**\n * A model is a data structure that consists of `Layers` and defines inputs\n * and outputs.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.model` is more generic, supporting an arbitrary graph (without\n * cycles) of layers. `tf.sequential` is less generic and supports only a linear\n * stack of layers.\n *\n * When creating a `tf.LayersModel`, specify its input(s) and output(s). Layers\n * are used to wire input(s) to output(s).\n *\n * For example, the following code snippet defines a model consisting of\n * two `dense` layers, with 10 and 4 units, respectively.\n *\n * ```js\n * // Define input, which has a size of 5 (not including batch dimension).\n * const input = tf.input({shape: [5]});\n *\n * // First dense layer uses relu activation.\n * const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});\n * // Second dense layer uses softmax activation.\n * const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});\n *\n * // Obtain the output symbolic tensor by applying the layers on the input.\n * const output = denseLayer2.apply(denseLayer1.apply(input));\n *\n * // Create the model based on the inputs.\n * const model = tf.model({inputs: input, outputs: output});\n *\n * // The model can be used for training, evaluation and prediction.\n * // For example, the following line runs prediction with the model on\n * // some fake data.\n * model.predict(tf.ones([2, 5])).print();\n * ```\n * See also:\n *   `tf.sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function model(args: ContainerArgs): LayersModel {\n  return new LayersModel(args);\n}\n\n/**\n * Creates a `tf.Sequential` model.  A sequential model is any model where the\n * outputs of one layer are the inputs to the next layer, i.e. the model\n * topology is a simple 'stack' of layers, with no branching or skipping.\n *\n * This means that the first layer passed to a `tf.Sequential` model should have\n * a defined input shape. What that means is that it should have received an\n * `inputShape` or `batchInputShape` argument, or for some type of layers\n * (recurrent, Dense...) an `inputDim` argument.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.sequential` is less generic, supporting only a linear stack of layers.\n * `tf.model` is more generic and supports an arbitrary graph (without\n * cycles) of layers.\n *\n * Examples:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have an input shape defined.\n * model.add(tf.layers.dense({units: 32, inputShape: [50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output, which equals\n * // `[null, 4]`. The 1st dimension is the undetermined batch dimension; the\n * // 2nd is the output size of the model's last layer.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * It is also possible to specify a batch size (with potentially undetermined\n * batch dimension, denoted by \"null\") for the first layer using the\n * `batchInputShape` key. The following example is equivalent to the above:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have a defined input shape\n * model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * You can also use an `Array` of already-constructed `Layer`s to create\n * a `tf.Sequential` model:\n *\n * ```js\n * const model = tf.sequential({\n *   layers: [tf.layers.dense({units: 32, inputShape: [50]}),\n *            tf.layers.dense({units: 4})]\n * });\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function sequential(config?: SequentialArgs): Sequential {\n  return new Sequential(config);\n}\n\n/**\n * Load a model composed of Layer objects, including its topology and optionally\n * weights. See the Tutorial named \"How to import a Keras Model\" for usage\n * examples.\n *\n * This method is applicable to:\n *\n * 1. Models created with the `tf.layers.*`, `tf.sequential`, and\n * `tf.model` APIs of TensorFlow.js and later saved with the\n * `tf.LayersModel.save` method.\n * 2. Models converted from Keras or TensorFlow tf.keras using the\n * [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).\n *\n * This mode is *not* applicable to TensorFlow `SavedModel`s or their converted\n * forms. For those models, use `tf.loadGraphModel`.\n *\n * Example 1. Load a model from an HTTP server.\n *\n * ```js\n * const model = await tf.loadLayersModel(\n *     'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * Example 2: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 4. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. For file://\n *      (tfjs-node-only), http:// and https:// schemas, the path can be\n *      either absolute or relative.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A function of the signature `(fraction: number) => void',\n *     that can be used as the progress callback for the model loading.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport function loadLayersModel(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  return loadLayersModelInternal(pathOrIOHandler, options);\n}\n\n/**\n * Used to instantiate an input to a model as a `tf.SymbolicTensor`.\n *\n * Users should call the `input` factory function for\n * consistency with other generator functions.\n *\n * Example:\n *\n * ```js\n * // Defines a simple logistic regression model with 32 dimensional input\n * // and 3 dimensional output.\n * const x = tf.input({shape: [32]});\n * const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x);\n * const model = tf.model({inputs: x, outputs: y});\n * model.predict(tf.ones([2, 32])).print();\n * ```\n *\n * Note: `input` is only necessary when using `model`. When using\n * `sequential`, specify `inputShape` for the first layer or use `inputLayer`\n * as the first layer.\n *\n * @doc {heading: 'Models', subheading: 'Inputs'}\n */\nexport function input(config: InputConfig): SymbolicTensor {\n  return Input(config);\n}\n\nexport function registerCallbackConstructor(\n    verbosityLevel: number,\n    callbackConstructor: BaseCallbackConstructor): void {\n  CallbackConstructorRegistry.registerCallbackConstructor(\n      verbosityLevel, callbackConstructor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Layer activation functions\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {ActivationIdentifier} from './keras_format/activation_config';\nimport {deserializeKerasObject} from './utils/generic_utils';\n\n/**\n * Base class for Activations.\n *\n * Special note: due to cross-language compatibility reasons, the\n * static readonly className field in this family of classes must be set to\n * the initialLowerCamelCase name of the activation.\n */\nexport abstract class Activation extends serialization.Serializable {\n  abstract apply(tensor: Tensor, axis?: number): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\n/**\n * Exponential linear unit (ELU).\n * Reference: https://arxiv.org/abs/1511.07289\n */\nexport class Elu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'elu';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x: Input.\n   * @param alpha: Scaling factor the negative section.\n   * @return Output of the ELU activation.\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return K.elu(x, alpha);\n  }\n}\nserialization.registerClass(Elu);\n\n/**\n * Scaled Exponential Linear Unit. (Klambauer et al., 2017).\n * Reference: Self-Normalizing Neural Networks, https://arxiv.org/abs/1706.02515\n * Notes:\n *   - To be used together with the initialization \"lecunNormal\".\n *   - To be used together with the dropout variant \"AlphaDropout\".\n */\nexport class Selu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'selu';\n  apply(x: Tensor): Tensor {\n    return tfc.selu(x);\n  }\n}\nserialization.registerClass(Selu);\n\n/**\n *  Rectified linear unit\n */\nexport class Relu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu';\n  apply(x: Tensor): Tensor {\n    return tfc.relu(x);\n  }\n}\nserialization.registerClass(Relu);\n\n/**\n * Rectified linear unit activation maxing out at 6.0.\n */\nexport class Relu6 extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu6';\n  apply(x: Tensor): Tensor {\n    return tidy(() => tfc.minimum(6.0, tfc.relu(x)));\n  }\n}\nserialization.registerClass(Relu6);\n\n//* Linear activation (no-op) */\nexport class Linear extends Activation {\n  /** @nocollapse */\n  static readonly className = 'linear';\n  apply(x: Tensor): Tensor {\n    return x;\n  }\n}\nserialization.registerClass(Linear);\n\n/**\n * Sigmoid activation function.\n */\nexport class Sigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'sigmoid';\n  apply(x: Tensor): Tensor {\n    return tfc.sigmoid(x);\n  }\n}\nserialization.registerClass(Sigmoid);\n\n/**\n * Segment-wise linear approximation of sigmoid.\n */\nexport class HardSigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'hardSigmoid';\n  apply(x: Tensor): Tensor {\n    return K.hardSigmoid(x);\n  }\n}\nserialization.registerClass(HardSigmoid);\n\n/**\n * Softplus activation function.\n */\nexport class Softplus extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softplus';\n  apply(x: Tensor): Tensor {\n    return tfc.softplus(x);\n  }\n}\nserialization.registerClass(Softplus);\n\n/**\n * Softsign activation function.\n */\nexport class Softsign extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softsign';\n  apply(x: Tensor): Tensor {\n    return K.softsign(x);\n  }\n}\nserialization.registerClass(Softsign);\n\n/**\n * Hyperbolic tangent function.\n */\nexport class Tanh extends Activation {\n  /** @nocollapse */\n  static readonly className = 'tanh';\n  apply(x: Tensor): Tensor {\n    return tfc.tanh(x);\n  }\n}\nserialization.registerClass(Tanh);\n\n/**\n * Softmax activation function\n */\nexport class Softmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softmax';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.softmax(x, axis);\n  }\n}\nserialization.registerClass(Softmax);\n\n/**\n * Log softmax activation function\n */\nexport class LogSoftmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'logSoftmax';\n  /**\n   * Calculate the activation function of log softmax:\n   * log( exp(x_i) / sum(exp(x)) )\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.logSoftmax(x, axis);\n  }\n}\nserialization.registerClass(LogSoftmax);\n\n/**\n * Swish activation function\n */\nexport class Swish extends Activation {\n  /** @nocollapse */\n  static readonly className = 'swish';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param alpha Scaling factor for the sigmoid function.\n   * @returns a Tensor of the same shape as x\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return tidy(() => tfc.sigmoid(x.mul(alpha)).mul(x));\n  }\n}\nserialization.registerClass(Swish);\n\nexport function serializeActivation(activation: Activation): string {\n  return activation.getClassName();\n}\n\nexport function deserializeActivation(\n   config: serialization.ConfigDict,\n   customObjects: serialization.ConfigDict = {}): Activation {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'activation');\n}\n\nexport function getActivation(identifier: ActivationIdentifier|\n                              serialization.ConfigDict|Activation): Activation {\n  if (identifier == null) {\n    const config: serialization.ConfigDict = {};\n    config['className'] = 'linear';\n    config['config'] = {};\n    return deserializeActivation(config);\n  }\n  if (typeof identifier === 'string') {\n    const config: serialization.ConfigDict = {};\n    config['className'] = identifier;\n    config['config'] = {};\n    return deserializeActivation(config);\n  } else if (identifier instanceof Activation) {\n    return identifier;\n  } else {\n    return deserializeActivation(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* original source: keras/regularizers.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {abs, add, Scalar, serialization, sum, Tensor, tidy, zeros} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\nfunction assertObjectArgs(args: L1Args | L2Args | L1L2Args): void {\n  if (args != null && typeof args !== 'object') {\n    throw new Error(\n        `Argument to L1L2 regularizer's constructor is expected to be an ` +\n        `object, but received: ${args}`);\n  }\n}\n\n/**\n * Regularizer base class.\n */\nexport abstract class Regularizer extends serialization.Serializable {\n  abstract apply(x: Tensor): Scalar;\n}\n\nexport interface L1L2Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1?: number;\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2?: number;\n}\n\nexport interface L1Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1: number;\n}\n\nexport interface L2Args {\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2: number;\n}\n\nexport class L1L2 extends Regularizer {\n  /** @nocollapse */\n  static className = 'L1L2';\n\n  private readonly l1: number;\n  private readonly l2: number;\n  private readonly hasL1: boolean;\n  private readonly hasL2: boolean;\n  constructor(args?: L1L2Args) {\n    super();\n\n    assertObjectArgs(args);\n\n    this.l1 = args == null || args.l1 == null ? 0.01 : args.l1;\n    this.l2 = args == null || args.l2 == null ? 0.01 : args.l2;\n    this.hasL1 = this.l1 !== 0;\n    this.hasL2 = this.l2 !== 0;\n  }\n\n  /**\n   * Porting note: Renamed from __call__.\n   * @param x Variable of which to calculate the regularization score.\n   */\n  apply(x: Tensor): Scalar {\n    return tidy(() => {\n      let regularization: Tensor = zeros([1]);\n      if (this.hasL1) {\n        regularization = add(regularization, sum(tfc.mul(this.l1, abs(x))));\n      }\n      if (this.hasL2) {\n        regularization =\n            add(regularization, sum(tfc.mul(this.l2, K.square(x))));\n      }\n      return regularization.asScalar();\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {'l1': this.l1, 'l2': this.l2};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls({l1: config['l1'] as number, l2: config['l2'] as number});\n  }\n}\nserialization.registerClass(L1L2);\n\nexport function l1(args?: L1Args) {\n  assertObjectArgs(args);\n  return new L1L2({l1: args != null ? args.l1 : null, l2: 0});\n}\n\nexport function l2(args: L2Args) {\n  assertObjectArgs(args);\n  return new L1L2({l2: args != null ? args.l2 : null, l1: 0});\n}\n\n/** @docinline */\nexport type RegularizerIdentifier = 'l1l2'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding keras symbols.\nexport const REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in RegularizerIdentifier]: string} = {\n      'l1l2': 'L1L2'\n    };\n\nexport function serializeRegularizer(constraint: Regularizer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeRegularizer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Regularizer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'regularizer');\n}\n\nexport function getRegularizer(identifier: RegularizerIdentifier|\n                               serialization.ConfigDict|\n                               Regularizer): Regularizer {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeRegularizer(config);\n  } else if (identifier instanceof Regularizer) {\n    return identifier;\n  } else {\n    return deserializeRegularizer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n *  Advanced activation layers.\n */\n\nimport {clipByValue, elu, leakyRelu, prelu, relu, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {Softmax as softmaxActivation} from '../activations';\nimport {cast} from '../backend/tfjs_backend';\nimport {Constraint, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface ReLULayerArgs extends LayerArgs {\n  /**\n   * Float, the maximum output value.\n   */\n  maxValue?: number;\n}\n\nexport class ReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ReLU';\n  maxValue: number;\n\n  constructor(args?: ReLULayerArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maxValue = args.maxValue;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    let output = relu(inputs);\n    if (this.maxValue != null) {\n      output = clipByValue(output, 0, this.maxValue);\n    }\n    return output;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {maxValue: this.maxValue};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ReLU);\n\nexport declare interface LeakyReLULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `0.3`.\n   */\n  alpha?: number;\n}\n\nexport class LeakyReLU extends Layer {\n  /** @nocollapse */\n  static className = 'LeakyReLU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 0.3;\n\n  constructor(args?: LeakyReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return leakyRelu(x, this.alpha);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LeakyReLU);\n\nexport declare interface PReLULayerArgs extends LayerArgs {\n  /**\n   * Initializer for the learnable alpha.\n   */\n  alphaInitializer?: Initializer|InitializerIdentifier;\n\n  /**\n   * Regularizer for the learnable alpha.\n   */\n  alphaRegularizer?: Regularizer;\n\n  /**\n   * Constraint for the learnable alpha.\n   */\n  alphaConstraint?: Constraint;\n\n  /**\n   * The axes along which to share learnable parameters for the activation\n   * function. For example, if the incoming feature maps are from a 2D\n   * convolution with output shape `[numExamples, height, width, channels]`,\n   * and you wish to share parameters across space (height and width) so that\n   * each filter channels has only one set of parameters, set\n   * `shared_axes: [1, 2]`.\n   */\n  sharedAxes?: number|number[];\n}\n\nexport class PReLU extends Layer {\n  /** @nocollapse */\n  static className = 'PReLU';\n  private readonly alphaInitializer: Initializer;\n  private readonly alphaRegularizer: Regularizer;\n  private readonly alphaConstraint: Constraint;\n  private readonly sharedAxes: number[];\n  private alpha: LayerVariable;\n\n  readonly DEFAULT_ALPHA_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args?: PReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.supportsMasking = true;\n    this.alphaInitializer =\n        getInitializer(args.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER);\n    this.alphaRegularizer = getRegularizer(args.alphaRegularizer);\n    this.alphaConstraint = getConstraint(args.alphaConstraint);\n    if (args.sharedAxes == null) {\n      this.sharedAxes = null;\n    } else if (Array.isArray(args.sharedAxes)) {\n      this.sharedAxes = args.sharedAxes;\n    } else if (typeof args.sharedAxes === 'number') {\n      this.sharedAxes = [args.sharedAxes];\n    } else {\n      throw new ValueError(\n          `Expected sharedAxes to be a number or an array of numbers, ` +\n          `but got ${args.sharedAxes}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]) {\n    inputShape = getExactlyOneShape(inputShape);\n    const paramShape: Shape = inputShape.slice(1);\n    if (this.sharedAxes != null) {\n      for (const i of this.sharedAxes) {\n        paramShape[i - 1] = 1;\n      }\n    }\n    this.alpha = this.addWeight(\n        'alpha', paramShape, 'float32', this.alphaInitializer,\n        this.alphaRegularizer, true, this.alphaConstraint);\n    // Set input spec.\n    const axes: {[axis: number]: number} = {};\n    if (this.sharedAxes != null) {\n      for (let i = 1; i < inputShape.length; ++i) {\n        axes[i] = inputShape[i];\n      }\n    }\n    this.inputSpec = [new InputSpec({\n      ndim: inputShape.length,\n      axes,\n    })];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    return prelu(inputs, this.alpha.read());\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      alphaInitializer: serializeInitializer(this.alphaInitializer),\n      alphaRegularizer: serializeRegularizer(this.alphaRegularizer),\n      alphaConstraint: serializeConstraint(this.alphaConstraint),\n      sharedAxes: this.sharedAxes\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(PReLU);\n\nexport declare interface ELULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `1.0`.\n   */\n  alpha?: number;\n}\n\nexport class ELU extends Layer {\n  /** @nocollapse */\n  static className = 'ELU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 1.0;\n\n  constructor(args?: ELULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    if (args.alpha != null && args.alpha !== this.DEFAULT_ALPHA) {\n      throw new NotImplementedError(\n          `Non-default alpha value (${args.alpha}) is not supported by the ` +\n          `ELU layer yet.`);\n    }\n\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return elu(x);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ELU);\n\nexport declare interface ThresholdedReLULayerArgs extends LayerArgs {\n  /**\n   * Float >= 0. Threshold location of activation.\n   */\n  theta?: number;\n}\n\nexport class ThresholdedReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ThresholdedReLU';\n  readonly theta: number;\n\n  readonly DEFAULT_THETA = 1.0;\n\n  constructor(args?: ThresholdedReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.theta = args.theta == null ? this.DEFAULT_THETA : args.theta;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return x.mul(cast(x.greater(this.theta), 'float32'));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {theta: this.theta};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ThresholdedReLU);\n\nexport declare interface SoftmaxLayerArgs extends LayerArgs {\n  /**\n   * Integer, axis along which the softmax normalization is applied.\n   * Defaults to `-1` (i.e., the last axis).\n   */\n  axis?: number;\n}\n\nexport class Softmax extends Layer {\n  /** @nocollapse */\n  static className = 'Softmax';\n  readonly axis: number;\n  readonly softmax: (t: Tensor, a?: number) => Tensor;\n  readonly DEFAULT_AXIS = 1.0;\n\n  constructor(args?: SoftmaxLayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.softmax = new softmaxActivation().apply;\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return this.softmax(x, this.axis);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {axis: this.axis};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Softmax);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {ValueError} from '../errors';\nimport {PaddingMode} from '../keras_format/common';\n\nimport {pyListRepeat} from './generic_utils';\nimport {isInteger, max} from './math_utils';\n\n/**\n * Transforms a single number of array of numbers into an array of numbers.\n * @param value\n * @param n: The size of the tuple to be returned.\n * @param name: Name of the parameter, used for generating error messages.\n * @returns An array of numbers.\n */\nexport function normalizeArray(\n    value: number|number[], n: number, name: string): number[] {\n  if (typeof value === 'number') {\n    return pyListRepeat(value, n);\n  } else {\n    if (value.length !== n) {\n      throw new ValueError(\n          `The ${name} argument must be an integer or tuple of ${n} integers.` +\n          ` Received: ${value.length} elements.`);\n    }\n    for (let i = 0; i < n; ++i) {\n      const singleValue = value[i];\n      if (!isInteger(singleValue)) {\n        throw new ValueError(\n            `The ${name} argument must be an integer or tuple of ${n}` +\n            ` integers. Received: ${JSON.stringify(value)} including a` +\n            ` non-integer number ${singleValue}`);\n      }\n    }\n    return value;\n  }\n}\n\n/**\n * Determines output length of a convolution given input length.\n * @param inputLength\n * @param filterSize\n * @param padding\n * @param stride\n * @param dilation: dilation rate.\n */\nexport function convOutputLength(\n    inputLength: number, filterSize: number, padding: PaddingMode,\n    stride: number, dilation = 1): number {\n  if (inputLength == null) {\n    return inputLength;\n  }\n  const dilatedFilterSize = filterSize + (filterSize - 1) * (dilation - 1);\n  let outputLength: number;\n  if (padding === 'same') {\n    outputLength = inputLength;\n  } else {  // VALID\n    outputLength = inputLength - dilatedFilterSize + 1;\n  }\n  return Math.floor((outputLength + stride - 1) / stride);\n}\n\nexport function deconvLength(\n    dimSize: number, strideSize: number, kernelSize: number,\n    padding: PaddingMode): number {\n  if (dimSize == null) {\n    return null;\n  }\n\n  if (padding === 'valid') {\n    dimSize = dimSize * strideSize + max([kernelSize - strideSize, 0]);\n  } else if (padding === 'same') {\n    dimSize = dimSize * strideSize;\n  } else {\n    throw new ValueError(`Unsupport padding mode: ${padding}.`);\n  }\n  return dimSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {fused, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, PaddingMode, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, deconvLength, normalizeArray} from '../utils/conv_utils';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Transpose and cast the input before the conv2d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv2DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  // TODO(cais): Cast type to float32 if not.\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 1]);  // NCHW -> NHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * Transpose and cast the input before the conv3d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv3DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 4, 1]);  // NCDHW -> NDHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * 1D-convolution with bias added.\n *\n * Porting Note: This function does not exist in the Python Keras backend.\n *   It is exactly the same as `conv2d`, except the added `bias`.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.\n * @param bias Bias, rank-3, of shape `[outDepth]`.\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    // Check the ranks of x, kernel and bias.\n    if (x.shape.length !== 3) {\n      throw new ValueError(\n          `The input of a conv1dWithBias operation should be 3, but is ` +\n          `${x.shape.length} instead.`);\n    }\n    if (kernel.shape.length !== 3) {\n      throw new ValueError(\n          `The kernel for a conv1dWithBias operation should be 3, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    if (bias != null && bias.shape.length !== 1) {\n      throw new ValueError(\n          `The bias for a conv1dWithBias operation should be 1, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    // TODO(cais): Support CAUSAL padding mode.\n    if (dataFormat === 'channelsFirst') {\n      x = tfc.transpose(x, [0, 2, 1]);  // NCW -> NWC.\n    }\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    let y: Tensor = tfc.conv1d(\n        x as Tensor2D | Tensor3D, kernel as Tensor3D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias);\n    }\n    return y;\n  });\n}\n\n/**\n * 1D-convolution.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.s\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1d(\n    x: Tensor, kernel: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv1dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 2D pooling.\n */\nexport function conv2d(\n    x: Tensor, kernel: Tensor, strides = [1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv2dWithBiasActivation(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution with an added bias and optional activation.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv2d`, except the added `bias`.\n */\nexport function conv2dWithBiasActivation(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat, dilationRate?: [number, number],\n    activation: fused.Activation = null): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 3 && x.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects input to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    if (kernel.rank !== 3 && kernel.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects kernel to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.fused.conv2d({\n      x: y as Tensor3D | Tensor4D,\n      filter: kernel as Tensor4D,\n      strides: strides as [number, number],\n      pad: padding === 'same' ? 'same' : 'valid',\n      dilations: dilationRate,\n      dataFormat: 'NHWC',\n      bias,\n      activation\n    });\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\n/**\n * 3D Convolution.\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 3D convolution.\n */\nexport function conv3d(\n    x: Tensor, kernel: Tensor, strides = [1, 1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv3dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 3D Convolution with an added bias.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv3d`, except the added `bias`.\n */\nexport function conv3dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 4 && x.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects input to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    if (kernel.rank !== 4 && kernel.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects kernel to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    let y = preprocessConv3DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv3dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.conv3d(\n        y as Tensor4D | tfc.Tensor<tfc.Rank.R5>,\n        kernel as tfc.Tensor<tfc.Rank.R5>, strides as [number, number, number],\n        padding === 'same' ? 'same' : 'valid', 'NDHWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias as Tensor1D);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);\n    }\n    return y;\n  });\n}\n\n/**\n * Base LayerConfig for depthwise and non-depthwise convolutional layers.\n */\nexport declare interface BaseConvLayerArgs extends LayerArgs {\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number]|[number, number, number];\n\n  /**\n   * Activation function of the layer.\n   *\n   * If you don't specify the activation, none is applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector. Defaults to `true`.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the convolutional kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the convolutional kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\n/**\n * LayerConfig for non-depthwise convolutional layers.\n * Applies to non-depthwise convolution of all ranks (e.g, Conv1D, Conv2D,\n * Conv3D).\n */\nexport declare interface ConvLayerArgs extends BaseConvLayerArgs {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n}\n\n/**\n * Abstract convolution layer.\n */\nexport abstract class BaseConv extends Layer {\n  protected readonly rank: number;\n  protected readonly kernelSize: number[];\n  protected readonly strides: number[];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n  protected readonly activation: Activation;\n  protected readonly useBias: boolean;\n  protected readonly dilationRate: number[];\n\n  // Bias-related members are here because all convolution subclasses use the\n  // same configuration parmeters to control bias.  Kernel-related members\n  // are in subclass `Conv` because some subclasses use different parameters to\n  // control kernel properties, for instance, `DepthwiseConv2D` uses\n  // `depthwiseInitializer` instead of `kernelInitializer`.\n  protected readonly biasInitializer?: Initializer;\n  protected readonly biasConstraint?: Constraint;\n  protected readonly biasRegularizer?: Regularizer;\n\n  protected bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(rank: number, args: BaseConvLayerArgs) {\n    super(args as LayerArgs);\n    BaseConv.verifyArgs(args);\n    this.rank = rank;\n    generic_utils.assertPositiveInteger(this.rank, 'rank');\n    if (this.rank !== 1 && this.rank !== 2 && this.rank !== 3) {\n      throw new NotImplementedError(\n          `Convolution layer for rank other than 1, 2, or 3 (${\n              this.rank}) is ` +\n          `not implemented yet.`);\n    }\n    this.kernelSize = normalizeArray(args.kernelSize, rank, 'kernelSize');\n    this.strides = normalizeArray(\n        args.strides == null ? 1 : args.strides, rank, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.activation = getActivation(args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.dilationRate = normalizeArray(\n        args.dilationRate == null ? 1 : args.dilationRate, rank,\n        'dilationRate');\n    if (this.rank === 1 &&\n        (Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)) {\n      throw new ValueError(\n          `dilationRate must be a number or an array of a single number ` +\n          `for 1D convolution, but received ` +\n          `${JSON.stringify(this.dilationRate)}`);\n    } else if (this.rank === 2) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate = [this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 2) {\n        throw new ValueError(\n            `dilationRate must be a number or array of two numbers for 2D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    } else if (this.rank === 3) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate =\n            [this.dilationRate, this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 3) {\n        throw new ValueError(\n            `dilationRate must be a number or array of three numbers for 3D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    }\n  }\n\n  protected static verifyArgs(args: BaseConvLayerArgs) {\n    // Check config.kernelSize type and shape.\n    generic_utils.assert(\n        'kernelSize' in args, `required key 'kernelSize' not in config`);\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 3)) {\n      throw new ValueError(\n          `BaseConv expects config.kernelSize to be number or number[] with ` +\n          `length 1, 2, or 3, but received ${\n              JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      kernelSize: this.kernelSize,\n      strides: this.strides,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\n/**\n * Abstract nD convolution layer.  Ancestor of convolution layers which reduce\n * across channels, i.e., Conv1D and Conv2D, but not DepthwiseConv2D.\n */\nexport abstract class Conv extends BaseConv {\n  protected readonly filters: number;\n\n  protected kernel: LayerVariable = null;\n\n  // Bias-related properties are stored in the superclass `BaseConv` because all\n  // convolution subclasses use the same configuration parameters to control\n  // bias. Kernel-related properties are defined here rather than in the\n  // superclass because some convolution subclasses use different names and\n  // configuration parameters for their internal kernel state.\n  protected readonly kernelInitializer?: Initializer;\n  protected readonly kernelConstraint?: Constraint;\n  protected readonly kernelRegularizer?: Regularizer;\n\n  constructor(rank: number, args: ConvLayerArgs) {\n    super(rank, args as BaseConvLayerArgs);\n    Conv.verifyArgs(args);\n    this.filters = args.filters;\n    generic_utils.assertPositiveInteger(this.filters, 'filters');\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n    const inputDim = inputShape[channelAxis];\n\n    const kernelShape = this.kernelSize.concat([inputDim, this.filters]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.inputSpec = [{ndim: this.rank + 2, axes: {[channelAxis]: inputDim}}];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs: Tensor;\n      const biasValue = this.bias == null ? null : this.bias.read();\n      const fusedActivationName = generic_utils.mapActivationToFusedKernel(\n                                            this.activation.getClassName());\n\n      if (fusedActivationName != null && this.rank === 2) {\n        outputs = conv2dWithBiasActivation(\n            inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n            this.dataFormat, this.dilationRate as [number, number],\n            fusedActivationName);\n      } else {\n        if (this.rank === 1) {\n          outputs = conv1dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides[0],\n              this.padding, this.dataFormat, this.dilationRate[0]);\n        } else if (this.rank === 2) {\n          // TODO(cais): Move up to constructor.\n          outputs = conv2dWithBiasActivation(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number]);\n        } else if (this.rank === 3) {\n          outputs = conv3dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number, number]);\n        } else {\n          throw new NotImplementedError(\n              'convolutions greater than 3D are not implemented yet.');\n        }\n\n        if (this.activation != null) {\n          outputs = this.activation.apply(outputs);\n        }\n      }\n\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const newSpace: number[] = [];\n    const space = (this.dataFormat === 'channelsLast') ?\n        inputShape.slice(1, inputShape.length - 1) :\n        inputShape.slice(2);\n    for (let i = 0; i < space.length; ++i) {\n      const newDim = convOutputLength(\n          space[i], this.kernelSize[i], this.padding, this.strides[i],\n          typeof this.dilationRate === 'number' ? this.dilationRate :\n                                                  this.dilationRate[i]);\n      newSpace.push(newDim);\n    }\n\n    let outputShape = [inputShape[0]];\n    if (this.dataFormat === 'channelsLast') {\n      outputShape = outputShape.concat(newSpace);\n      outputShape.push(this.filters);\n    } else {\n      outputShape.push(this.filters);\n      outputShape = outputShape.concat(newSpace);\n    }\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      filters: this.filters,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // Check config.filters type, shape, and value.\n    if (!('filters' in args) || typeof args.filters !== 'number' ||\n        args.filters < 1) {\n      throw new ValueError(\n          `Convolution layer expected config.filters to be a 'number' > 0 ` +\n          `but got ${JSON.stringify(args.filters)}`);\n    }\n  }\n}\n\nexport class Conv2D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv2D';\n  constructor(args: ConvLayerArgs) {\n    super(2, args);\n    Conv2D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if ((typeof args.kernelSize !== 'number') &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 2)) {\n      throw new ValueError(\n          `Conv2D expects config.kernelSize to be number or number[] with ` +\n          `length 1 or 2, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv2D);\n\nexport class Conv3D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv3D';\n  constructor(args: ConvLayerArgs) {\n    super(3, args);\n    Conv3D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number') {\n      if (!(Array.isArray(args.kernelSize) &&\n            (args.kernelSize.length === 1 || args.kernelSize.length === 3))) {\n        throw new ValueError(\n            `Conv3D expects config.kernelSize to be number or` +\n            ` [number, number, number], but received ${\n                JSON.stringify(args.kernelSize)}.`);\n      }\n    }\n  }\n}\nserialization.registerClass(Conv3D);\n\nexport class Conv2DTranspose extends Conv2D {\n  /** @nocollapse */\n  static className = 'Conv2DTranspose';\n  inputSpec: InputSpec[];\n\n  constructor(args: ConvLayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n\n    if (this.padding !== 'same' && this.padding !== 'valid') {\n      throw new ValueError(\n          `Conv2DTranspose currently supports only padding modes 'same' ` +\n          `and 'valid', but received padding mode ${this.padding}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    if (inputShape.length !== 4) {\n      throw new ValueError(\n          'Input should have rank 4; Received input shape: ' +\n          JSON.stringify(inputShape));\n    }\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          'The channel dimension of the inputs should be defined. ' +\n          'Found `None`.');\n    }\n    const inputDim = inputShape[channelAxis];\n    const kernelShape = this.kernelSize.concat([this.filters, inputDim]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, 'float32', this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    // Set input spec.\n    this.inputSpec =\n        [new InputSpec({ndim: 4, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs);\n      if (input.shape.length !== 4) {\n        throw new ValueError(\n            `Conv2DTranspose.call() expects input tensor to be rank-4, but ` +\n            `received a tensor of rank-${input.shape.length}`);\n      }\n\n      const inputShape = input.shape;\n      const batchSize = inputShape[0];\n\n      let hAxis: number;\n      let wAxis: number;\n      if (this.dataFormat === 'channelsFirst') {\n        hAxis = 2;\n        wAxis = 3;\n      } else {\n        hAxis = 1;\n        wAxis = 2;\n      }\n\n      const height = inputShape[hAxis];\n      const width = inputShape[wAxis];\n      const kernelH = this.kernelSize[0];\n      const kernelW = this.kernelSize[1];\n      const strideH = this.strides[0];\n      const strideW = this.strides[1];\n\n      // Infer the dynamic output shape.\n      const outHeight = deconvLength(height, strideH, kernelH, this.padding);\n      const outWidth = deconvLength(width, strideW, kernelW, this.padding);\n\n      // Porting Note: We don't branch based on `this.dataFormat` here,\n      // because\n      //   the tjfs-core function `conv2dTranspose` called below always\n      //   assumes channelsLast.\n      const outputShape: [number, number, number, number] =\n          [batchSize, outHeight, outWidth, this.filters];\n\n      if (this.dataFormat !== 'channelsLast') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n      }\n      let outputs = tfc.conv2dTranspose(\n          input as Tensor4D, this.kernel.read() as Tensor4D, outputShape,\n          this.strides as [number, number], this.padding as 'same' | 'valid');\n      if (this.dataFormat !== 'channelsLast') {\n        outputs = tfc.transpose(outputs, [0, 3, 1, 2]);\n      }\n\n      if (this.bias != null) {\n        outputs =\n            K.biasAdd(outputs, this.bias.read(), this.dataFormat) as Tensor4D;\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs) as Tensor4D;\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n\n    let channelAxis: number;\n    let heightAxis: number;\n    let widthAxis: number;\n    if (this.dataFormat === 'channelsFirst') {\n      channelAxis = 1;\n      heightAxis = 2;\n      widthAxis = 3;\n    } else {\n      channelAxis = 3;\n      heightAxis = 1;\n      widthAxis = 2;\n    }\n\n    const kernelH = this.kernelSize[0];\n    const kernelW = this.kernelSize[1];\n    const strideH = this.strides[0];\n    const strideW = this.strides[1];\n\n    outputShape[channelAxis] = this.filters;\n    outputShape[heightAxis] =\n        deconvLength(outputShape[heightAxis], strideH, kernelH, this.padding);\n    outputShape[widthAxis] =\n        deconvLength(outputShape[widthAxis], strideW, kernelW, this.padding);\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['dilationRate'];\n    return config;\n  }\n}\nserialization.registerClass(Conv2DTranspose);\n\nexport declare interface SeparableConvLayerArgs extends ConvLayerArgs {\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal\n   * to `filtersIn * depthMultiplier`. Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the pointwise kernel matrix.\n   */\n  pointwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the pointwise kernel matrix.\n   */\n  pointwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the pointwise kernel matrix.\n   */\n  pointwiseConstraint?: ConstraintIdentifier|Constraint;\n}\n\nexport class SeparableConv extends Conv {\n  /** @nocollapse */\n  static className = 'SeparableConv';\n\n  readonly depthMultiplier: number;\n\n  protected readonly depthwiseInitializer?: Initializer;\n  protected readonly depthwiseRegularizer?: Regularizer;\n  protected readonly depthwiseConstraint?: Constraint;\n  protected readonly pointwiseInitializer?: Initializer;\n  protected readonly pointwiseRegularizer?: Regularizer;\n  protected readonly pointwiseConstraint?: Constraint;\n\n  readonly DEFAULT_DEPTHWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n  readonly DEFAULT_POINTWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n\n  protected depthwiseKernel: LayerVariable = null;\n  protected pointwiseKernel: LayerVariable = null;\n\n  constructor(rank: number, config?: SeparableConvLayerArgs) {\n    super(rank, config);\n\n    if (config.filters == null) {\n      throw new ValueError(\n          'The `filters` configuration field is required by SeparableConv, ' +\n          'but is unspecified.');\n    }\n    if (config.kernelInitializer != null || config.kernelRegularizer != null ||\n        config.kernelConstraint != null) {\n      throw new ValueError(\n          'Fields kernelInitializer, kernelRegularizer and kernelConstraint ' +\n          'are invalid for SeparableConv2D. Use depthwiseInitializer, ' +\n          'depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, ' +\n          'pointwiseRegularizer and pointwiseConstraint instead.');\n    }\n    if (config.padding != null && config.padding !== 'same' &&\n        config.padding !== 'valid') {\n      throw new ValueError(\n          `SeparableConv${this.rank}D supports only padding modes: ` +\n          `'same' and 'valid', but received ${JSON.stringify(config.padding)}`);\n    }\n\n    this.depthMultiplier =\n        config.depthMultiplier == null ? 1 : config.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER);\n    this.depthwiseRegularizer = getRegularizer(config.depthwiseRegularizer);\n    this.depthwiseConstraint = getConstraint(config.depthwiseConstraint);\n    this.pointwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER);\n    this.pointwiseRegularizer = getRegularizer(config.pointwiseRegularizer);\n    this.pointwiseConstraint = getConstraint(config.pointwiseConstraint);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < this.rank + 2) {\n      throw new ValueError(\n          `Inputs to SeparableConv${this.rank}D should have rank ` +\n          `${this.rank + 2}, but received input shape: ` +\n          `${JSON.stringify(inputShape)}`);\n    }\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          `The channel dimension of the inputs should be defined, ` +\n          `but found ${JSON.stringify(inputShape[channelAxis])}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape =\n        this.kernelSize.concat([inputDim, this.depthMultiplier]);\n    const pointwiseKernelShape = [];\n    for (let i = 0; i < this.rank; ++i) {\n      pointwiseKernelShape.push(1);\n    }\n    pointwiseKernelShape.push(inputDim * this.depthMultiplier, this.filters);\n\n    const trainable = true;\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, 'float32',\n        this.depthwiseInitializer, this.depthwiseRegularizer, trainable,\n        this.depthwiseConstraint);\n    this.pointwiseKernel = this.addWeight(\n        'pointwise_kernel', pointwiseKernelShape, 'float32',\n        this.pointwiseInitializer, this.pointwiseRegularizer, trainable,\n        this.pointwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, trainable, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n\n    this.inputSpec =\n        [new InputSpec({ndim: this.rank + 2, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      let output: Tensor;\n      if (this.rank === 1) {\n        throw new NotImplementedError(\n            '1D separable convolution is not implemented yet.');\n      } else if (this.rank === 2) {\n        if (this.dataFormat === 'channelsFirst') {\n          inputs = tfc.transpose(inputs, [0, 2, 3, 1]);  // NCHW -> NHWC.\n        }\n\n        output = tfc.separableConv2d(\n            inputs as Tensor4D, this.depthwiseKernel.read() as Tensor4D,\n            this.pointwiseKernel.read() as Tensor4D,\n            this.strides as [number, number], this.padding as 'same' | 'valid',\n            this.dilationRate as [number, number], 'NHWC');\n      }\n\n      if (this.useBias) {\n        output = K.biasAdd(output, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      if (this.dataFormat === 'channelsFirst') {\n        output = tfc.transpose(output, [0, 3, 1, 2]);  // NHWC -> NCHW.\n      }\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['kernelInitializer'];\n    delete config['kernelRegularizer'];\n    delete config['kernelConstraint'];\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['pointwiseInitializer'] =\n        serializeInitializer(this.pointwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['pointwiseRegularizer'] =\n        serializeRegularizer(this.pointwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseConstraint);\n    config['pointwiseConstraint'] =\n        serializeConstraint(this.pointwiseConstraint);\n    return config;\n  }\n}\n\nexport class SeparableConv2D extends SeparableConv {\n  /** @nocollapse */\n  static className = 'SeparableConv2D';\n  constructor(args?: SeparableConvLayerArgs) {\n    super(2, args);\n  }\n}\nserialization.registerClass(SeparableConv2D);\n\nexport class Conv1D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv1D';\n  constructor(args: ConvLayerArgs) {\n    super(1, args);\n    Conv1D.verifyArgs(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['dataFormat'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 1)) {\n      throw new ValueError(\n          `Conv1D expects config.kernelSize to be number or number[] with ` +\n          `length 1, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv1D);\n\nexport declare interface Cropping2DLayerArgs extends LayerArgs {\n  /**\n   * Dimension of the cropping along the width and the height.\n   * - If integer: the same symmetric cropping\n   *  is applied to width and height.\n   * - If list of 2 integers:\n   *   interpreted as two different\n   *   symmetric cropping values for height and width:\n   *   `[symmetric_height_crop, symmetric_width_crop]`.\n   * - If a list of 2 list of 2 integers:\n   *   interpreted as\n   *   `[[top_crop, bottom_crop], [left_crop, right_crop]]`\n   */\n  cropping: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   * `channels_first` corresponds to inputs with shape\n   *   `(batch, channels, ...)`\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class Cropping2D extends Layer {\n  /** @nocollapse */\n  static className = 'Cropping2D';\n  protected readonly cropping: [[number, number], [number, number]];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Cropping2DLayerArgs) {\n    super(args);\n    if (typeof args.cropping === 'number') {\n      this.cropping =\n          [[args.cropping, args.cropping], [args.cropping, args.cropping]];\n    } else if (typeof args.cropping[0] === 'number') {\n      this.cropping = [\n        [args.cropping[0], args.cropping[0]],\n        [args.cropping[1] as number, args.cropping[1] as number]\n      ];\n    } else {\n      this.cropping = args.cropping as [[number, number], [number, number]];\n    }\n    this.dataFormat =\n        args.dataFormat === undefined ? 'channelsLast' : args.dataFormat;\n    this.inputSpec = [{ndim: 4}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      return [\n        inputShape[0], inputShape[1],\n        inputShape[2] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[3] - this.cropping[1][0] - this.cropping[1][1]\n      ];\n    } else {\n      return [\n        inputShape[0],\n        inputShape[1] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[2] - this.cropping[1][0] - this.cropping[1][1], inputShape[3]\n      ];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      if (this.dataFormat === 'channelsLast') {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[1] - this.cropping[0][0] - this.cropping[0][1], 2);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[2] - this.cropping[1][1] - this.cropping[1][0], 3);\n      } else {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[2] - this.cropping[0][0] - this.cropping[0][1], 3);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[3] - this.cropping[1][1] - this.cropping[1][0], 4);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {cropping: this.cropping, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Cropping2D);\n\nexport declare interface UpSampling2DLayerArgs extends LayerArgs {\n  /**\n   * The upsampling factors for rows and columns.\n   *\n   * Defaults to `[2, 2]`.\n   */\n  size?: number[];\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `\"channelsLast\"` corresponds to inputs with shape\n   *   `[batch, ..., channels]`\n   *\n   *  `\"channelsFirst\"` corresponds to inputs with shape `[batch, channels,\n   * ...]`.\n   *\n   * Defaults to `\"channelsLast\"`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class UpSampling2D extends Layer {\n  /** @nocollapse */\n  static className = 'UpSampling2D';\n  protected readonly DEFAULT_SIZE = [2, 2];\n  protected readonly size: number[];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: UpSampling2DLayerArgs) {\n    super(args);\n    this.inputSpec = [{ndim: 4}];\n    this.size = args.size == null ? this.DEFAULT_SIZE : args.size;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      const height =\n          inputShape[2] == null ? null : this.size[0] * inputShape[2];\n      const width = inputShape[3] == null ? null : this.size[1] * inputShape[3];\n      return [inputShape[0], inputShape[1], height, width];\n    } else {\n      const height =\n          inputShape[1] == null ? null : this.size[0] * inputShape[1];\n      const width = inputShape[2] == null ? null : this.size[1] * inputShape[2];\n      return [inputShape[0], height, width, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs) as Tensor4D;\n      const inputShape = input.shape;\n\n      if (this.dataFormat === 'channelsFirst') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n        const height = this.size[0] * inputShape[2];\n        const width = this.size[1] * inputShape[3];\n        const resized = input.resizeNearestNeighbor([height, width]);\n        return tfc.transpose(resized, [0, 3, 1, 2]);\n      } else {\n        const height = this.size[0] * inputShape[1];\n        const width = this.size[1] * inputShape[2];\n        return input.resizeNearestNeighbor([height, width]);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {size: this.size, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(UpSampling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Depthwise Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {BaseConv, BaseConvLayerArgs, ConvLayerArgs, preprocessConv2DInput} from './convolutional';\n\n/**\n * 2D convolution with separable filters.\n * @param x Input tensor.\n * @param depthwiseKernel Convolution kernel for depthwise convolution.\n * @param strides Strides (Array of two integers).\n * @param padding Padding model.\n * @param dataFormat Data format.\n * @param dilationRate Array of two integers, dilation rates for the separable\n *   convolution.\n * @returns Output tensor.\n * @throws ValueError If depthwiseKernel is not a 4D array.\n */\nexport function depthwiseConv2d(\n    x: Tensor, depthwiseKernel: Tensor, strides: [number, number] = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `Input for depthwiseConv2d is required to be 4-D, but is instead ` +\n          `${x.rank}-D`);\n    }\n    if (depthwiseKernel.rank !== 4) {\n      throw new ValueError(\n          `depthwiseKernel is required to be 4-D, but is instead ` +\n          `${depthwiseKernel.rank}-D`);\n    }\n    y = tfc.depthwiseConv2d(\n        y as Tensor4D, depthwiseKernel as Tensor4D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NHWC', dilationRate);\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\nexport declare interface DepthwiseConv2DLayerArgs extends BaseConvLayerArgs {\n  /**\n   * An integer or Array of 2 integers, specifying the width and height of the\n   * 2D convolution window. Can be a single integer to specify the same value\n   * for all spatial dimensions.\n   */\n  kernelSize: number|[number, number];\n\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal to\n   * `filtersIn * depthMultiplier`.\n   * Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   * Default: GlorotNormal.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regulzarizer function for the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class DepthwiseConv2D extends BaseConv {\n  /** @nocollapse */\n  static className = 'DepthwiseConv2D';\n  private readonly depthMultiplier: number;\n  private readonly depthwiseInitializer: Initializer;\n  private readonly depthwiseConstraint: Constraint;\n  private readonly depthwiseRegularizer: Regularizer;\n\n  private depthwiseKernel: LayerVariable = null;\n\n  constructor(args: DepthwiseConv2DLayerArgs) {\n    super(2, args as ConvLayerArgs);\n    this.depthMultiplier =\n        args.depthMultiplier == null ? 1 : args.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        args.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.depthwiseConstraint = getConstraint(args.depthwiseConstraint);\n    this.depthwiseRegularizer = getRegularizer(args.depthwiseRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 4) {\n      throw new ValueError(\n          `Inputs to DepthwiseConv2D should have rank 4. ` +\n          `Received input shape: ${JSON.stringify(inputShape)}.`);\n    }\n    const channelAxis = this.dataFormat === 'channelsFirst' ? 1 : 3;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          'The channel dimension of the inputs to DepthwiseConv2D should ' +\n          `be defined, but is not (${inputShape[channelAxis]}).`);\n    }\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape: Shape = [\n      this.kernelSize[0], this.kernelSize[1], inputDim, this.depthMultiplier\n    ];\n\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, null,\n        this.depthwiseInitializer, this.depthwiseRegularizer, true,\n        this.depthwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [inputDim * this.depthMultiplier], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs = depthwiseConv2d(\n          inputs, this.depthwiseKernel.read(), this.strides as [number, number],\n          this.padding, this.dataFormat, null);\n      // TODO(cais): Add support for dilation.\n      if (this.useBias) {\n        outputs = K.biasAdd(outputs, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs);\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    const cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    const outFilters = this.dataFormat === 'channelsFirst' ?\n        inputShape[1] * this.depthMultiplier :\n        inputShape[3] * this.depthMultiplier;\n    const outRows = convOutputLength(\n        rows, this.kernelSize[0], this.padding, this.strides[0]);\n    const outCols = convOutputLength(\n        cols, this.kernelSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], outFilters, outRows, outCols];\n    } else {\n      // In this case, assume 'channelsLast'.\n      return [inputShape[0], outRows, outCols, outFilters];\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    config['depthMultiplier'] = this.depthMultiplier;\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseRegularizer);\n    return config;\n  }\n}\nserialization.registerClass(DepthwiseConv2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Recurrent Neural Network Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, SymbolicTensor} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, Ones, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs, RnnStepFunction} from '../types';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor, isArrayOfShapes} from '../utils/types_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\nimport {deserialize} from './serialization';\n\n/**\n * Standardize `apply()` args to a single list of tensor inputs.\n *\n * When running a model loaded from file, the input tensors `initialState` and\n * `constants` are passed to `RNN.apply()` as part of `inputs` instead of the\n * dedicated kwargs fields. `inputs` consists of\n * `[inputs, initialState0, initialState1, ..., constant0, constant1]` in this\n * case.\n * This method makes sure that arguments are\n * separated and that `initialState` and `constants` are `Array`s of tensors\n * (or None).\n *\n * @param inputs Tensor or `Array` of  tensors.\n * @param initialState Tensor or `Array` of tensors or `null`/`undefined`.\n * @param constants Tensor or `Array` of tensors or `null`/`undefined`.\n * @returns An object consisting of\n *   inputs: A tensor.\n *   initialState: `Array` of tensors or `null`.\n *   constants: `Array` of tensors or `null`.\n * @throws ValueError, if `inputs` is an `Array` but either `initialState` or\n *   `constants` is provided.\n */\nexport function standardizeArgs(\n    inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    initialState: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    constants: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    numConstants?: number): {\n  inputs: Tensor|SymbolicTensor,\n  initialState: Tensor[]|SymbolicTensor[],\n  constants: Tensor[]|SymbolicTensor[]\n} {\n  if (Array.isArray(inputs)) {\n    if (initialState != null || constants != null) {\n      throw new ValueError(\n          'When inputs is an array, neither initialState or constants ' +\n          'should be provided');\n    }\n    if (numConstants != null) {\n      constants = inputs.slice(inputs.length - numConstants, inputs.length);\n      inputs = inputs.slice(0, inputs.length - numConstants);\n    }\n    if (inputs.length > 1) {\n      initialState = inputs.slice(1, inputs.length);\n    }\n    inputs = inputs[0];\n  }\n\n  function toListOrNull(x: Tensor|Tensor[]|SymbolicTensor|\n                        SymbolicTensor[]): Tensor[]|SymbolicTensor[] {\n    if (x == null || Array.isArray(x)) {\n      return x as Tensor[] | SymbolicTensor[];\n    } else {\n      return [x] as Tensor[] | SymbolicTensor[];\n    }\n  }\n\n  initialState = toListOrNull(initialState);\n  constants = toListOrNull(constants);\n\n  return {inputs, initialState, constants};\n}\n\n/**\n * Iterates over the time dimension of a tensor.\n *\n * @param stepFunction RNN step function.\n *   Parameters:\n *     inputs: tensor with shape `[samples, ...]` (no time dimension),\n *       representing input for the batch of samples at a certain time step.\n *     states: an Array of tensors.\n *   Returns:\n *     outputs: tensor with shape `[samples, outputDim]` (no time dimension).\n *     newStates: list of tensors, same length and shapes as `states`. The first\n *       state in the list must be the output tensor at the previous timestep.\n * @param inputs Tensor of temporal data of shape `[samples, time, ...]` (at\n *   least 3D).\n * @param initialStates Tensor with shape `[samples, outputDim]` (no time\n *   dimension), containing the initial values of the states used in the step\n *   function.\n * @param goBackwards If `true`, do the iteration over the time dimension in\n *   reverse order and return the reversed sequence.\n * @param mask Binary tensor with shape `[sample, time, 1]`, with a zero for\n *   every element that is masked.\n * @param constants An Array of constant values passed at each step.\n * @param unroll Whether to unroll the RNN or to use a symbolic loop. *Not*\n *   applicable to this imperative deeplearn.js backend. Its value is ignored.\n * @param needPerStepOutputs Whether the per-step outputs are to be\n *   concatenated into a single tensor and returned (as the second return\n *   value). Default: `false`. This arg is included so that the relatively\n *   expensive concatenation of the stepwise outputs can be omitted unless\n *   the stepwise outputs need to be kept (e.g., for an LSTM layer of which\n *   `returnSequence` is `true`.)\n * @returns An Array: `[lastOutput, outputs, newStates]`.\n *   lastOutput: the lastest output of the RNN, of shape `[samples, ...]`.\n *   outputs: tensor with shape `[samples, time, ...]` where each entry\n *     `output[s, t]` is the output of the step function at time `t` for sample\n *     `s`. This return value is provided if and only if the\n *     `needPerStepOutputs` is set as `true`. If it is set as `false`, this\n *     return value will be `undefined`.\n *   newStates: Array of tensors, latest states returned by the step function,\n *      of shape `(samples, ...)`.\n * @throws ValueError If input dimension is less than 3.\n *\n * TODO(nielsene): This needs to be tidy-ed.\n */\nexport function rnn(\n    stepFunction: RnnStepFunction, inputs: Tensor, initialStates: Tensor[],\n    goBackwards = false, mask?: Tensor, constants?: Tensor[], unroll = false,\n    needPerStepOutputs = false): [Tensor, Tensor, Tensor[]] {\n  return tfc.tidy(() => {\n    const ndim = inputs.shape.length;\n    if (ndim < 3) {\n      throw new ValueError(`Input should be at least 3D, but is ${ndim}D.`);\n    }\n\n    // Transpose to time-major, i.e., from [batch, time, ...] to [time, batch,\n    // ...].\n    const axes = [1, 0].concat(math_utils.range(2, ndim));\n    inputs = tfc.transpose(inputs, axes);\n\n    if (constants != null) {\n      throw new NotImplementedError(\n          'The rnn() functoin of the deeplearn.js backend does not support ' +\n          'constants yet.');\n    }\n\n    // Porting Note: the unroll option is ignored by the imperative backend.\n    if (unroll) {\n      console.warn(\n          'Backend rnn(): the unroll = true option is not applicable to the ' +\n          'imperative deeplearn.js backend.');\n    }\n\n    if (mask != null) {\n      mask = mask.asType('bool').asType('float32');\n      if (mask.rank === ndim - 1) {\n        mask = tfc.expandDims(mask, -1);\n      }\n      mask = tfc.transpose(mask, axes);\n    }\n\n    if (goBackwards) {\n      inputs = tfc.reverse(inputs, 0);\n      if (mask != null) {\n        mask = tfc.reverse(mask, 0);\n      }\n    }\n\n    // Porting Note: PyKeras with TensorFlow backend uses a symbolic loop\n    //   (tf.while_loop). But for the imperative deeplearn.js backend, we just\n    //   use the usual TypeScript control flow to iterate over the time steps in\n    //   the inputs.\n    // Porting Note: PyKeras patches a \"_use_learning_phase\" attribute to\n    // outputs.\n    //   This is not idiomatic in TypeScript. The info regarding whether we are\n    //   in a learning (i.e., training) phase for RNN is passed in a different\n    //   way.\n\n    const perStepOutputs: Tensor[] = [];\n    let lastOutput: Tensor;\n    let states = initialStates;\n    const timeSteps = inputs.shape[0];\n    const perStepInputs = tfc.unstack(inputs);\n    let perStepMasks: Tensor[];\n    if (mask != null) {\n      perStepMasks = tfc.unstack(mask);\n    }\n\n    for (let t = 0; t < timeSteps; ++t) {\n      const currentInput = perStepInputs[t];\n      const stepOutputs = tfc.tidy(() => stepFunction(currentInput, states));\n\n      if (mask == null) {\n        lastOutput = stepOutputs[0];\n        states = stepOutputs[1];\n      } else {\n        const maskedOutputs = tfc.tidy(() => {\n          const stepMask = perStepMasks[t];\n          const negStepMask = tfc.onesLike(stepMask).sub(stepMask);\n          // TODO(cais): Would tfc.where() be better for performance?\n          const output =\n              stepOutputs[0].mul(stepMask).add(states[0].mul(negStepMask));\n          const newStates = states.map((state, i) => {\n            return stepOutputs[1][i].mul(stepMask).add(state.mul(negStepMask));\n          });\n          return {output, newStates};\n        });\n        lastOutput = maskedOutputs.output;\n        states = maskedOutputs.newStates;\n      }\n\n      if (needPerStepOutputs) {\n        perStepOutputs.push(lastOutput);\n      }\n    }\n    let outputs: Tensor;\n    if (needPerStepOutputs) {\n      const axis = 1;\n      outputs = tfc.stack(perStepOutputs, axis);\n    }\n    return [lastOutput, outputs, states] as [Tensor, Tensor, Tensor[]];\n  });\n}\n\nexport declare interface BaseRNNLayerArgs extends LayerArgs {\n  /**\n   * A RNN cell instance. A RNN cell is a class that has:\n   *   - a `call()` method, which takes `[Tensor, Tensor]` as the\n   *     first input argument. The first item is the input at time t, and\n   *     second item is the cell state at time t.\n   *     The `call()` method returns `[outputAtT, statesAtTPlus1]`.\n   *     The `call()` method of the cell can also take the argument `constants`,\n   *     see section \"Note on passing external constants\" below.\n   *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,\n   *       which are Layer subtypes, to accept two arguments. tfjs-layers does\n   *       not do such overriding. Instead we preseve the `call()` signature,\n   *       which due to its `Tensor|Tensor[]` argument and return value, is\n   *       flexible enough to handle the inputs and states.\n   *   - a `stateSize` attribute. This can be a single integer (single state)\n   *     in which case it is the size of the recurrent state (which should be\n   *     the same as the size of the cell output). This can also be an Array of\n   *     integers (one size per state). In this case, the first entry\n   *     (`stateSize[0]`) should be the same as the size of the cell output.\n   * It is also possible for `cell` to be a list of RNN cell instances, in which\n   * case the cells get stacked on after the other in the RNN, implementing an\n   * efficient stacked RNN.\n   */\n  cell?: RNNCell|RNNCell[];\n\n  /**\n   * Whether to return the last output in the output sequence, or the full\n   * sequence.\n   */\n  returnSequences?: boolean;\n\n  /**\n   * Whether to return the last state in addition to the output.\n   */\n  returnState?: boolean;\n\n  /**\n   * If `true`, process the input sequence backwards and return the reversed\n   * sequence (default: `false`).\n   */\n  goBackwards?: boolean;\n\n  /**\n   * If `true`, the last state for each sample at index i in a batch will be\n   * used as initial state of the sample of index i in the following batch\n   * (default: `false`).\n   *\n   * You can set RNN layers to be \"stateful\", which means that the states\n   * computed for the samples in one batch will be reused as initial states\n   * for the samples in the next batch. This assumes a one-to-one mapping\n   * between samples in different successive batches.\n   *\n   * To enable \"statefulness\":\n   *   - specify `stateful: true` in the layer constructor.\n   *   - specify a fixed batch size for your model, by passing\n   *     - if sequential model:\n   *       `batchInputShape: [...]` to the first layer in your model.\n   *     - else for functional model with 1 or more Input layers:\n   *       `batchShape: [...]` to all the first layers in your model.\n   *     This is the expected shape of your inputs\n   *     *including the batch size*.\n   *     It should be a tuple of integers, e.g., `[32, 10, 100]`.\n   *   - specify `shuffle: false` when calling `LayersModel.fit()`.\n   *\n   * To reset the state of your model, call `resetStates()` on either the\n   * specific layer or on the entire model.\n   */\n  stateful?: boolean;\n  // TODO(cais): Explore whether we can warn users when they fail to set\n  //   `shuffle: false` when training a model consisting of stateful RNNs\n  //   and any stateful Layers in general.\n\n  /**\n   * If `true`, the network will be unrolled, else a symbolic loop will be\n   * used. Unrolling can speed-up a RNN, although it tends to be more memory-\n   * intensive. Unrolling is only suitable for short sequences (default:\n   * `false`).\n   * Porting Note: tfjs-layers has an imperative backend. RNNs are executed with\n   *   normal TypeScript control flow. Hence this property is inapplicable and\n   *   ignored in tfjs-layers.\n   */\n  unroll?: boolean;\n\n  /**\n   * Dimensionality of the input (integer).\n   *   This option (or alternatively, the option `inputShape`) is required when\n   *   this layer is used as the first layer in a model.\n   */\n  inputDim?: number;\n\n  /**\n   * Length of the input sequences, to be specified when it is constant.\n   * This argument is required if you are going to connect `Flatten` then\n   * `Dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed). Note that if the recurrent layer is not the first layer in\n   * your model, you would need to specify the input length at the level of the\n   * first layer (e.g., via the `inputShape` option).\n   */\n  inputLength?: number;\n}\n\nexport class RNN extends Layer {\n  /** @nocollapse */\n  static className = 'RNN';\n  public readonly cell: RNNCell;\n  public readonly returnSequences: boolean;\n  public readonly returnState: boolean;\n  public readonly goBackwards: boolean;\n  public readonly unroll: boolean;\n\n  public stateSpec: InputSpec[];\n  protected states_: Tensor[];\n\n  // NOTE(cais): For stateful RNNs, the old states cannot be disposed right\n  // away when new states are set, because the old states may need to be used\n  // later for backpropagation through time (BPTT) and other purposes. So we\n  // keep them here for final disposal when the state is reset completely\n  // (i.e., through no-arg call to `resetStates()`).\n  protected keptStates: Tensor[][];\n\n  private numConstants: number;\n\n  constructor(args: RNNLayerArgs) {\n    super(args);\n    let cell: RNNCell;\n    if (args.cell == null) {\n      throw new ValueError(\n          'cell property is missing for the constructor of RNN.');\n    } else if (Array.isArray(args.cell)) {\n      cell = new StackedRNNCells({cells: args.cell});\n    } else {\n      cell = args.cell;\n    }\n    if (cell.stateSize == null) {\n      throw new ValueError(\n          'The RNN cell should have an attribute `stateSize` (tuple of ' +\n          'integers, one integer per RNN state).');\n    }\n    this.cell = cell;\n    this.returnSequences =\n        args.returnSequences == null ? false : args.returnSequences;\n    this.returnState = args.returnState == null ? false : args.returnState;\n    this.goBackwards = args.goBackwards == null ? false : args.goBackwards;\n    this._stateful = args.stateful == null ? false : args.stateful;\n    this.unroll = args.unroll == null ? false : args.unroll;\n\n    this.supportsMasking = true;\n    this.inputSpec = [new InputSpec({ndim: 3})];\n    this.stateSpec = null;\n    this.states_ = null;\n    // TODO(cais): Add constantsSpec and numConstants.\n    this.numConstants = null;\n    // TODO(cais): Look into the use of initial_state in the kwargs of the\n    //   constructor.\n\n    this.keptStates = [];\n  }\n\n  // Porting Note: This is the equivalent of `RNN.states` property getter in\n  //   PyKeras.\n  getStates(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      return math_utils.range(0, numStates).map(x => null);\n    } else {\n      return this.states_;\n    }\n  }\n\n  // Porting Note: This is the equivalent of the `RNN.states` property setter in\n  //   PyKeras.\n  setStates(states: Tensor[]): void {\n    this.states_ = states;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    // TODO(cais): Remove the casting once stacked RNN cells become supported.\n    let stateSize = this.cell.stateSize;\n    if (!Array.isArray(stateSize)) {\n      stateSize = [stateSize];\n    }\n    const outputDim = stateSize[0];\n    let outputShape: Shape|Shape[];\n    if (this.returnSequences) {\n      outputShape = [inputShape[0], inputShape[1], outputDim];\n    } else {\n      outputShape = [inputShape[0], outputDim];\n    }\n\n    if (this.returnState) {\n      const stateShape: Shape[] = [];\n      for (const dim of stateSize) {\n        stateShape.push([inputShape[0], dim]);\n      }\n      return [outputShape].concat(stateShape);\n    } else {\n      return outputShape;\n    }\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tfc.tidy(() => {\n      if (Array.isArray(mask)) {\n        mask = mask[0];\n      }\n      const outputMask = this.returnSequences ? mask : null;\n\n      if (this.returnState) {\n        const stateMask = this.states.map(s => null);\n        return [outputMask].concat(stateMask);\n      } else {\n        return outputMask;\n      }\n    });\n  }\n\n  /**\n   * Get the current state tensors of the RNN.\n   *\n   * If the state hasn't been set, return an array of `null`s of the correct\n   * length.\n   */\n  get states(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      const output: Tensor[] = [];\n      for (let i = 0; i < numStates; ++i) {\n        output.push(null);\n      }\n      return output;\n    } else {\n      return this.states_;\n    }\n  }\n\n  set states(s: Tensor[]) {\n    this.states_ = s;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    // Note inputShape will be an Array of Shapes of initial states and\n    // constants if these are passed in apply().\n    const constantShape: Shape[] = null;\n    if (this.numConstants != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    }\n\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    const batchSize: number = this.stateful ? inputShape[0] : null;\n    const inputDim = inputShape.slice(2);\n    this.inputSpec[0] = new InputSpec({shape: [batchSize, null, ...inputDim]});\n\n    // Allow cell (if RNNCell Layer) to build before we set or validate\n    // stateSpec.\n    const stepInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (constantShape != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    } else {\n      this.cell.build(stepInputShape);\n    }\n\n    // Set or validate stateSpec.\n    let stateSize: number[];\n    if (Array.isArray(this.cell.stateSize)) {\n      stateSize = this.cell.stateSize;\n    } else {\n      stateSize = [this.cell.stateSize];\n    }\n\n    if (this.stateSpec != null) {\n      if (!util.arraysEqual(\n              this.stateSpec.map(spec => spec.shape[spec.shape.length - 1]),\n              stateSize)) {\n        throw new ValueError(\n            `An initialState was passed that is not compatible with ` +\n            `cell.stateSize. Received stateSpec=${this.stateSpec}; ` +\n            `However cell.stateSize is ${this.cell.stateSize}`);\n      }\n    } else {\n      this.stateSpec =\n          stateSize.map(dim => new InputSpec({shape: [null, dim]}));\n    }\n    if (this.stateful) {\n      this.resetStates();\n    }\n  }\n\n  /**\n   * Reset the state tensors of the RNN.\n   *\n   * If the `states` argument is `undefined` or `null`, will set the\n   * state tensor(s) of the RNN to all-zero tensors of the appropriate\n   * shape(s).\n   *\n   * If `states` is provided, will set the state tensors of the RNN to its\n   * value.\n   *\n   * @param states Optional externally-provided initial states.\n   * @param training Whether this call is done during training. For stateful\n   *   RNNs, this affects whether the old states are kept or discarded. In\n   *   particular, if `training` is `true`, the old states will be kept so\n   *   that subsequent backpropgataion through time (BPTT) may work properly.\n   *   Else, the old states will be discarded.\n   */\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n      const batchSize = this.inputSpec[0].shape[0];\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n      // Initialize state if null.\n      if (this.states_ == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_ = [tfc.zeros([batchSize, this.cell.stateSize])];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_[0] = tfc.zeros([batchSize, this.cell.stateSize]);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training === true) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n          const dim = Array.isArray(this.cell.stateSize) ?\n              this.cell.stateSize[index] :\n              this.cell.stateSize;\n          const expectedShape = [batchSize, dim];\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n          this.states_[index] = value;\n        }\n      }\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    // TODO(cais): Figure out whether initialState is in kwargs or inputs.\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    // If any of `initial_state` or `constants` are specified and are\n    // `tf.SymbolicTensor`s, then add them to the inputs and temporarily modify\n    // the input_spec to include them.\n\n    let additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    let additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      kwargs['initialState'] = initialState;\n      additionalInputs = additionalInputs.concat(initialState);\n      this.stateSpec = [];\n      for (const state of initialState) {\n        this.stateSpec.push(new InputSpec({shape: state.shape}));\n      }\n      // TODO(cais): Use the following instead.\n      // this.stateSpec = initialState.map(state => new InputSpec({shape:\n      // state.shape}));\n      additionalSpecs = additionalSpecs.concat(this.stateSpec);\n    }\n    if (constants != null) {\n      kwargs['constants'] = constants;\n      additionalInputs = additionalInputs.concat(constants);\n      // TODO(cais): Add this.constantsSpec.\n      this.numConstants = constants.length;\n    }\n\n    const isTensor = additionalInputs[0] instanceof SymbolicTensor;\n    if (isTensor) {\n      // Compute full input spec, including state and constants.\n      const fullInput =\n          [inputs].concat(additionalInputs) as Tensor[] | SymbolicTensor[];\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call with temporarily replaced inputSpec.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output = super.apply(fullInput, kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  // tslint:disable-next-line:no-any\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    // Input shape: `[samples, time (padded with zeros), input_dim]`.\n    // Note that the .build() method of subclasses **must** define\n    // this.inputSpec and this.stateSpec owith complete input shapes.\n    return tidy(() => {\n      const mask = kwargs == null ? null : kwargs['mask'] as Tensor;\n      const training = kwargs == null ? null : kwargs['training'];\n      let initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      inputs = getExactlyOneTensor(inputs);\n      if (initialState == null) {\n        if (this.stateful) {\n          initialState = this.states_;\n        } else {\n          initialState = this.getInitialState(inputs);\n        }\n      }\n\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      if (initialState.length !== numStates) {\n        throw new ValueError(\n            `RNN Layer has ${numStates} state(s) but was passed ` +\n            `${initialState.length} initial state(s).`);\n      }\n      if (this.unroll) {\n        console.warn(\n            'Ignoring unroll = true for RNN layer, due to imperative backend.');\n      }\n\n      const cellCallKwargs: Kwargs = {training};\n\n      // TODO(cais): Add support for constants.\n      const step = (inputs: Tensor, states: Tensor[]) => {\n        // `inputs` and `states` are concatenated to form a single `Array` of\n        // `tf.Tensor`s as the input to `cell.call()`.\n        const outputs =\n            this.cell.call([inputs].concat(states), cellCallKwargs) as Tensor[];\n        // Marshall the return value into output and new states.\n        return [outputs[0], outputs.slice(1)] as [Tensor, Tensor[]];\n      };\n\n      // TODO(cais): Add support for constants.\n\n      const rnnOutputs =\n          rnn(step, inputs, initialState, this.goBackwards, mask, null,\n              this.unroll, this.returnSequences);\n      const lastOutput = rnnOutputs[0];\n      const outputs = rnnOutputs[1];\n      const states = rnnOutputs[2];\n\n      if (this.stateful) {\n        this.resetStates(states, training);\n      }\n\n      const output = this.returnSequences ? outputs : lastOutput;\n\n      // TODO(cais): Porperty set learning phase flag.\n\n      if (this.returnState) {\n        return [output].concat(states);\n      } else {\n        return output;\n      }\n    });\n  }\n\n  getInitialState(inputs: Tensor): Tensor[] {\n    return tidy(() => {\n      // Build an all-zero tensor of shape [samples, outputDim].\n      // [Samples, timeSteps, inputDim].\n      let initialState = tfc.zeros(inputs.shape);\n      // [Samples].\n      initialState = tfc.sum(initialState, [1, 2]);\n      initialState = K.expandDims(initialState);  // [Samples, 1].\n\n      if (Array.isArray(this.cell.stateSize)) {\n        return this.cell.stateSize.map(\n            dim => dim > 1 ? K.tile(initialState, [1, dim]) : initialState);\n      } else {\n        return this.cell.stateSize > 1 ?\n            [K.tile(initialState, [1, this.cell.stateSize])] :\n            [initialState];\n      }\n    });\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    return this.cell.trainableWeights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    if (!this.trainable) {\n      return this.cell.weights;\n    }\n    return this.cell.nonTrainableWeights;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.cell != null) {\n      this.cell.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      returnSequences: this.returnSequences,\n      returnState: this.returnState,\n      goBackwards: this.goBackwards,\n      stateful: this.stateful,\n      unroll: this.unroll,\n    };\n\n    if (this.numConstants != null) {\n      config['numConstants'] = this.numConstants;\n    }\n\n    const cellConfig = this.cell.getConfig();\n\n    if (this.getClassName() === RNN.className) {\n      config['cell'] = {\n        'className': this.cell.getClassName(),\n        'config': cellConfig,\n      } as serialization.ConfigDictValue;\n    }\n\n    // this order is necessary, to prevent cell name from replacing layer name\n    return {...cellConfig, ...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cellConfig = config['cell'] as serialization.ConfigDict;\n    const cell = deserialize(cellConfig, customObjects) as RNNCell;\n    return new cls(Object.assign(config, {cell}));\n  }\n}\nserialization.registerClass(RNN);\n\n// Porting Note: This is a common parent class for RNN cells. There is no\n// equivalent of this in PyKeras. Having a common parent class forgoes the\n//  need for `has_attr(cell, ...)` checks or its TypeScript equivalent.\n/**\n * An RNNCell layer.\n *\n * @doc {heading: 'Layers', subheading: 'Classes'}\n */\nexport abstract class RNNCell extends Layer {\n  /**\n   * Size(s) of the states.\n   * For RNN cells with only a single state, this is a single integer.\n   */\n  public stateSize: number|number[];\n  public dropoutMask: Tensor|Tensor[];\n  public recurrentDropoutMask: Tensor|Tensor[];\n}\n\nexport declare interface SimpleRNNCellLayerArgs extends LayerArgs {\n  /**\n   * units: Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   * Default: hyperbolic tangent ('tanh').\n   * If you pass `null`,  'linear' activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the `kernel` weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the `recurrent_kernel` weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the `kernel` weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the `recurrentKernel` weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraintfunction applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\nexport class SimpleRNNCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'SimpleRNNCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args: SimpleRNNCellLayerArgs) {\n    super(args);\n    this.units = args.units;\n    assertPositiveInteger(this.units, `units`);\n    this.activation = getActivation(\n        args.activation == null ? this.DEFAULT_ACTIVATION : args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    // TODO(cais): Use regularizer.\n    this.kernel = this.addWeight(\n        'kernel', [inputShape[inputShape.length - 1], this.units], null,\n        this.kernelInitializer, this.kernelRegularizer, true,\n        this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  // Porting Note: PyKeras' equivalent of this method takes two tensor inputs:\n  //   `inputs` and `states`. Here, the two tensors are combined into an\n  //   `Tensor[]` Array as the first input argument.\n  //   Similarly, PyKeras' equivalent of this method returns two values:\n  //    `output` and `[output]`. Here the two are combined into one length-2\n  //    `Tensor[]`, consisting of `output` repeated.\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `SimpleRNNCell expects 2 input Tensors, got ${inputs.length}.`);\n      }\n      let prevOutput = inputs[1];\n      inputs = inputs[0];\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training\n                           }) as Tensor;\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(prevOutput),\n                                      rate: this.recurrentDropout,\n                                      training\n                                    }) as Tensor;\n      }\n      let h: Tensor;\n      const dpMask: Tensor = this.dropoutMask as Tensor;\n      const recDpMask: Tensor = this.recurrentDropoutMask as Tensor;\n      if (dpMask != null) {\n        h = K.dot(tfc.mul(inputs, dpMask), this.kernel.read());\n      } else {\n        h = K.dot(inputs, this.kernel.read());\n      }\n      if (this.bias != null) {\n        h = K.biasAdd(h, this.bias.read());\n      }\n      if (recDpMask != null) {\n        prevOutput = tfc.mul(prevOutput, recDpMask);\n      }\n      let output = tfc.add(h, K.dot(prevOutput, this.recurrentKernel.read()));\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      // TODO(cais): Properly set learning phase on output tensor?\n      return [output, output];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(SimpleRNNCell);\n\nexport declare interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {\n  /**\n   * Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   *\n   * Defaults to  hyperbolic tangent (`tanh`)\n   *\n   * If you pass `null`, no activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the recurrentKernel weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the kernel weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the recurrentKernel weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\n/**\n * RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the\n * `cell` property required. This interface is to be used with constructors\n * of concrete RNN layer subtypes.\n */\nexport declare interface RNNLayerArgs extends BaseRNNLayerArgs {\n  cell: RNNCell|RNNCell[];\n}\n\nexport class SimpleRNN extends RNN {\n  /** @nocollapse */\n  static className = 'SimpleRNN';\n  constructor(args: SimpleRNNLayerArgs) {\n    args.cell = new SimpleRNNCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\nserialization.registerClass(SimpleRNN);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n\n  /**\n   * GRU convention (whether to apply reset gate after or before matrix\n   * multiplication). false = \"before\", true = \"after\" (only false is\n   * supported).\n   */\n  resetAfter?: boolean;\n}\n\nexport class GRUCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'GRUCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION: ActivationIdentifier = 'hardSigmoid';\n\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: GRUCellLayerArgs) {\n    super(args);\n    if (args.resetAfter) {\n      throw new ValueError(\n          `GRUCell does not support reset_after parameter set to true.`);\n    }\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 3], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 3], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units * 3], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `GRUCell expects 2 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      let hTMinus1 = inputs[1];  // Previous memory state.\n      inputs = inputs[0];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2, regardless of the actual value of\n      // config.implementation.\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 3\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 3\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor];\n      const recDpMask = this.recurrentDropoutMask as [Tensor, Tensor, Tensor];\n      let z: Tensor;\n      let r: Tensor;\n      let hh: Tensor;\n\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let matrixX = K.dot(inputs, this.kernel.read());\n      if (this.useBias) {\n        matrixX = K.biasAdd(matrixX, this.bias.read());\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n\n      const recurrentKernelValue = this.recurrentKernel.read();\n      const [rk1, rk2] = tfc.split(\n          recurrentKernelValue, [2 * this.units, this.units],\n          recurrentKernelValue.rank - 1);\n      const matrixInner = K.dot(hTMinus1, rk1);\n\n      const [xZ, xR, xH] = tfc.split(matrixX, 3, matrixX.rank - 1);\n      const [recurrentZ, recurrentR] =\n          tfc.split(matrixInner, 2, matrixInner.rank - 1);\n      z = this.recurrentActivation.apply(tfc.add(xZ, recurrentZ));\n      r = this.recurrentActivation.apply(tfc.add(xR, recurrentR));\n\n      const recurrentH = K.dot(tfc.mul(r, hTMinus1), rk2);\n      hh = this.activation.apply(tfc.add(xH, recurrentH));\n\n      const h =\n          tfc.add(tfc.mul(z, hTMinus1), tfc.mul(tfc.add(1, tfc.neg(z)), hh));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n      resetAfter: false\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(GRUCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface GRULayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   * smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class GRU extends RNN {\n  /** @nocollapse */\n  static className = 'GRU';\n  constructor(args: GRULayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new GRUCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(GRU);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class LSTMCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'LSTMCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n  readonly unitForgetBias: boolean;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number[];\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION = 'hardSigmoid';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n\n  readonly DEFAULT_BIAS_INITIALIZER = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: LSTMCellLayerArgs) {\n    super(args);\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.unitForgetBias = args.unitForgetBias;\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = [this.units, this.units];\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 4], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 4], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    let biasInitializer: Initializer;\n    if (this.useBias) {\n      if (this.unitForgetBias) {\n        const capturedBiasInit = this.biasInitializer;\n        const capturedUnits = this.units;\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): Tensor {\n            // TODO(cais): More informative variable names?\n            const bI = capturedBiasInit.apply([capturedUnits]);\n            const bF = (new Ones()).apply([capturedUnits]);\n            const bCAndH = capturedBiasInit.apply([capturedUnits * 2]);\n            return K.concatAlongFirstAxis(\n                K.concatAlongFirstAxis(bI, bF), bCAndH);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n      this.bias = this.addWeight(\n          'bias', [this.units * 4], null, biasInitializer, this.biasRegularizer,\n          true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `LSTMCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n      let hTMinus1 = inputs[1];    // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n      inputs = inputs[0];\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 4\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 4\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor, Tensor];\n      const recDpMask =\n          this.recurrentDropoutMask as [Tensor, Tensor, Tensor, Tensor];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2 regardless of the actual value of\n      // config.implementation.\n      let i: Tensor;\n      let f: Tensor;\n      let c: Tensor;\n      let o: Tensor;\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let z = K.dot(inputs, this.kernel.read());\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n      z = tfc.add(z, K.dot(hTMinus1, this.recurrentKernel.read()));\n      if (this.useBias) {\n        z = K.biasAdd(z, this.bias.read());\n      }\n\n      const [z0, z1, z2, z3] = tfc.split(z, 4, z.rank - 1);\n\n      i = this.recurrentActivation.apply(z0);\n      f = this.recurrentActivation.apply(z1);\n      c = tfc.add(tfc.mul(f, cTMinus1), tfc.mul(i, this.activation.apply(z2)));\n      o = this.recurrentActivation.apply(z3);\n\n      const h = tfc.mul(o, this.activation.apply(c));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      unitForgetBias: this.unitForgetBias,\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(LSTMCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface LSTMLayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *   Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions, whereas mode 2 will\n   *   batch them into fewer, larger operations. These modes will\n   *   have different performance profiles on different hardware and\n   *   for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this config field.\n   */\n  implementation?: number;\n}\n\nexport class LSTM extends RNN {\n  /** @nocollapse */\n  static className = 'LSTM';\n  constructor(args: LSTMLayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new LSTMCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(LSTM);\n\nexport declare interface StackedRNNCellsArgs extends LayerArgs {\n  /**\n   * A `Array` of `RNNCell` instances.\n   */\n  cells: RNNCell[];\n}\n\nexport class StackedRNNCells extends RNNCell {\n  /** @nocollapse */\n  static className = 'StackedRNNCells';\n  protected cells: RNNCell[];\n\n  constructor(args: StackedRNNCellsArgs) {\n    super(args);\n    this.cells = args.cells;\n  }\n\n  get stateSize(): number[] {\n    // States are a flat list in reverse order of the cell stack.\n    // This allows perserving the requirement `stack.statesize[0] ===\n    // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,\n    // assuming one LSTM has states `[h, c]`.\n    const stateSize: number[] = [];\n    for (const cell of this.cells.slice().reverse()) {\n      if (Array.isArray(cell.stateSize)) {\n        stateSize.push(...cell.stateSize);\n      } else {\n        stateSize.push(cell.stateSize);\n      }\n    }\n    return stateSize;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      let states = inputs.slice(1);\n\n      // Recover per-cell states.\n      const nestedStates: Tensor[][] = [];\n      for (const cell of this.cells.slice().reverse()) {\n        if (Array.isArray(cell.stateSize)) {\n          nestedStates.push(states.splice(0, cell.stateSize.length));\n        } else {\n          nestedStates.push(states.splice(0, 1));\n        }\n      }\n      nestedStates.reverse();\n\n      // Call the cells in order and store the returned states.\n      const newNestedStates: Tensor[][] = [];\n      let callInputs: Tensor[];\n      for (let i = 0; i < this.cells.length; ++i) {\n        const cell = this.cells[i];\n        states = nestedStates[i];\n        // TODO(cais): Take care of constants.\n        if (i === 0) {\n          callInputs = [inputs[0]].concat(states);\n        } else {\n          callInputs = [callInputs[0]].concat(states);\n        }\n        callInputs = cell.call(callInputs, kwargs) as Tensor[];\n        newNestedStates.push(callInputs.slice(1));\n      }\n\n      // Format the new states as a flat list in reverse cell order.\n      states = [];\n      for (const cellStates of newNestedStates.slice().reverse()) {\n        states.push(...cellStates);\n      }\n      return [callInputs[0]].concat(states);\n    });\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    if (isArrayOfShapes(inputShape)) {\n      // TODO(cais): Take care of input constants.\n      // const constantShape = inputShape.slice(1);\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n    let outputDim: number;\n    this.cells.forEach((cell, i) => {\n      nameScope(`RNNCell_${i}`, () => {\n        // TODO(cais): Take care of input constants.\n\n        cell.build(inputShape);\n        if (Array.isArray(cell.stateSize)) {\n          outputDim = cell.stateSize[0];\n        } else {\n          outputDim = cell.stateSize;\n        }\n        inputShape = [inputShape[0], outputDim] as Shape;\n      });\n    });\n    this.built = true;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const getCellConfig = (cell: RNNCell) => {\n      return {\n        'className': cell.getClassName(),\n        'config': cell.getConfig(),\n      };\n    };\n\n    const cellConfigs = this.cells.map(getCellConfig);\n\n    const config = {'cells': cellConfigs};\n\n    return {...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cells: RNNCell[] = [];\n    for (const cellConfig of (config['cells'] as serialization.ConfigDict[])) {\n      cells.push(deserialize(cellConfig, customObjects) as RNNCell);\n    }\n    return new cls({cells});\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const cell of this.cells) {\n        trainableWeights.push(...cell.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  /**\n   * Retrieve the weights of a the model.\n   *\n   * @returns A flat `Array` of `tf.Tensor`s.\n   */\n  getWeights(): Tensor[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.weights);\n    }\n    return batchGetValue(weights);\n  }\n\n  /**\n   * Set the weights of the model.\n   *\n   * @param weights An `Array` of `tf.Tensor`s with shapes and types matching\n   *     the output of `getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    const tuples: Array<[LayerVariable, Tensor]> = [];\n    for (const cell of this.cells) {\n      const numParams = cell.weights.length;\n      const inputWeights = weights.splice(numParams);\n      for (let i = 0; i < cell.weights.length; ++i) {\n        tuples.push([cell.weights[i], inputWeights[i]]);\n      }\n    }\n    batchSetValue(tuples);\n  }\n\n  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.\n}\nserialization.registerClass(StackedRNNCells);\n\nexport function generateDropoutMask(args: {\n  ones: () => tfc.Tensor,\n  rate: number,\n  training?: boolean,\n  count?: number,\n}): tfc.Tensor|tfc.Tensor[] {\n  const {ones, rate, training = false, count = 1} = args;\n\n  const droppedInputs = () => K.dropout(ones(), rate);\n\n  const createMask = () => K.inTrainPhase(droppedInputs, ones, training);\n\n  // just in case count is provided with null or undefined\n  if (!count || count <= 1) {\n    return tfc.keep(createMask().clone());\n  }\n\n  const masks = Array(count).fill(undefined).map(createMask);\n\n  return masks.map(m => tfc.keep(m.clone()));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {Activation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint} from '../constraints';\nimport {InputSpec} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {Initializer} from '../initializers';\nimport {DataFormat, DataType, PaddingMode, Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, normalizeArray} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\nimport {BaseRNNLayerArgs, generateDropoutMask, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNNCellLayerArgs} from './recurrent';\n\ndeclare interface ConvRNN2DCellArgs extends\n    Omit<SimpleRNNCellLayerArgs, 'units'> {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number];\n}\n\nabstract class ConvRNN2DCell extends RNNCell {\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n}\n\ndeclare interface ConvRNN2DLayerArgs extends BaseRNNLayerArgs,\n                                             ConvRNN2DCellArgs {}\n\n/**\n * Base class for convolutional-recurrent layers.\n */\nclass ConvRNN2D extends RNN {\n  /** @nocollapse */\n  static className = 'ConvRNN2D';\n\n  readonly cell: ConvRNN2DCell;\n\n  constructor(args: ConvRNN2DLayerArgs) {\n    if (args.unroll) {\n      throw new NotImplementedError(\n          'Unrolling is not possible with convolutional RNNs.');\n    }\n\n    if (Array.isArray(args.cell)) {\n      throw new NotImplementedError(\n          'It is not possible at the moment to stack convolutional cells.');\n    }\n\n    super(args as RNNLayerArgs);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n\n        this.cell.dropoutMask = null;\n      }\n\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n\n        this.cell.recurrentDropoutMask = null;\n      }\n\n      if (kwargs && kwargs['constants']) {\n        throw new ValueError('ConvRNN2D cell does not support constants');\n      }\n\n      const mask = kwargs == null ? null : kwargs['mask'];\n\n      const training = kwargs == null ? null : kwargs['training'];\n\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  computeOutputShape(inputShape: Shape): Shape|Shape[] {\n    let outShape: Shape = this.computeSingleOutputShape(inputShape);\n\n    if (!this.returnSequences) {\n      outShape = [outShape[0], ...outShape.slice(2)];\n    }\n\n    if (this.returnState) {\n      outShape =\n          [outShape, ...Array(2).fill([inputShape[0], ...outShape.slice(-3)])];\n    }\n\n    return outShape;\n  }\n\n  getInitialState(inputs: tfc.Tensor): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      const {stateSize} = this.cell;\n\n      const inputShape = inputs.shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const initialState = tfc.zeros(stateShape);\n\n      if (Array.isArray(stateSize)) {\n        return Array(stateSize.length).fill(initialState);\n      }\n\n      return [initialState];\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tfc.tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n\n      const inputShape = this.inputSpec[0].shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const batchSize = inputShape[0];\n\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n\n      // Initialize state if null.\n      if (this.getStates() == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_ = [tfc.zeros(stateShape)];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_[0] = tfc.zeros(stateShape);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n\n          const expectedShape = stateShape;\n\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n\n          this.states_[index] = value;\n        }\n      }\n\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  protected computeSingleOutputShape(inputShape: Shape): Shape {\n    const {dataFormat, filters, kernelSize, padding, strides, dilationRate} =\n        this.cell;\n\n    const isChannelsFirst = dataFormat === 'channelsFirst';\n\n    const h = inputShape[isChannelsFirst ? 3 : 2];\n    const w = inputShape[isChannelsFirst ? 4 : 3];\n\n    const hOut = convOutputLength(\n        h, kernelSize[0], padding, strides[0], dilationRate[0]);\n    const wOut = convOutputLength(\n        w, kernelSize[1], padding, strides[1], dilationRate[1]);\n\n    const outShape: Shape = [\n      ...inputShape.slice(0, 2),\n      ...(isChannelsFirst ? [filters, hOut, wOut] : [hOut, wOut, filters])\n    ];\n\n    return outShape;\n  }\n}\n\nexport declare interface ConvLSTM2DCellArgs extends\n    Omit<LSTMCellLayerArgs, 'units'>, ConvRNN2DCellArgs {}\n\nexport class ConvLSTM2DCell extends LSTMCell implements ConvRNN2DCell {\n  /** @nocollapse */\n  static className = 'ConvLSTM2DCell';\n\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  constructor(args: ConvLSTM2DCellArgs) {\n    const {\n      filters,\n      kernelSize,\n      strides,\n      padding,\n      dataFormat,\n      dilationRate,\n    } = args;\n\n    super({...args, units: filters});\n\n    this.filters = filters;\n    assertPositiveInteger(this.filters, 'filters');\n\n    this.kernelSize = normalizeArray(kernelSize, 2, 'kernelSize');\n    this.kernelSize.forEach(size => assertPositiveInteger(size, 'kernelSize'));\n\n    this.strides = normalizeArray(strides || 1, 2, 'strides');\n    this.strides.forEach(stride => assertPositiveInteger(stride, 'strides'));\n\n    this.padding = padding || 'valid';\n    checkPaddingMode(this.padding);\n\n    this.dataFormat = dataFormat || 'channelsLast';\n    checkDataFormat(this.dataFormat);\n\n    this.dilationRate = normalizeArray(dilationRate || 1, 2, 'dilationRate');\n    this.dilationRate.forEach(\n        rate => assertPositiveInteger(rate, 'dilationRate'));\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n\n    const numOfKernels = 4;\n\n    const kernelShape =\n        this.kernelSize.concat([inputDim, this.filters * numOfKernels]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n\n    const recurrentKernelShape =\n        this.kernelSize.concat([this.filters, this.filters * numOfKernels]);\n\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', recurrentKernelShape, null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n\n    if (this.useBias) {\n      let biasInitializer: Initializer;\n\n      if (this.unitForgetBias) {\n        const init = this.biasInitializer;\n\n        const filters = this.filters;\n\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): tfc.Tensor {\n            const biasI = init.apply([filters]);\n            const biasF = tfc.ones([filters]);\n            const biasCAndO = init.apply([filters * 2]);\n            return K.concatenate([biasI, biasF, biasCAndO]);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n\n      this.bias = this.addWeight(\n          'bias', [this.filters * numOfKernels], null, biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: tfc.Tensor[], kwargs: Kwargs): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] || false;\n\n      const x = inputs[0];         // Current input\n      const hTMinus1 = inputs[1];  // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n\n      const numOfKernels = 4;\n\n      type DropoutMasks = [tfc.Tensor, tfc.Tensor, tfc.Tensor, tfc.Tensor];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(x),\n                             rate: this.dropout,\n                             training,\n                             count: numOfKernels\n                           }) as tfc.Tensor[];\n      }\n\n      const dropoutMask = this.dropoutMask as DropoutMasks;\n\n      const applyDropout =\n          (x: tfc.Tensor, mask: tfc.Tensor[], index: number) => {\n            if (!mask || !mask[index]) {\n              return x;\n            }\n\n            return tfc.mul(mask[index], x);\n          };\n\n      let xI = applyDropout(x, dropoutMask, 0);\n      let xF = applyDropout(x, dropoutMask, 1);\n      let xC = applyDropout(x, dropoutMask, 2);\n      let xO = applyDropout(x, dropoutMask, 3);\n\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: numOfKernels\n                                    }) as tfc.Tensor[];\n      }\n\n      const recDropoutMask = this.recurrentDropoutMask as DropoutMasks;\n\n      let hI = applyDropout(hTMinus1, recDropoutMask, 0);\n      let hF = applyDropout(hTMinus1, recDropoutMask, 1);\n      let hC = applyDropout(hTMinus1, recDropoutMask, 2);\n      let hO = applyDropout(hTMinus1, recDropoutMask, 3);\n\n      const kernelChannelAxis = 3;\n\n      const [kernelI, kernelF, kernelC, kernelO]: tfc.Tensor[] =\n          tfc.split(this.kernel.read(), numOfKernels, kernelChannelAxis);\n\n      const [biasI, biasF, biasC, biasO]: tfc.Tensor[] = this.useBias ?\n          tfc.split(this.bias.read(), numOfKernels) :\n          [null, null, null, null];\n\n      xI = this.inputConv(xI, kernelI, biasI, this.padding);\n      xF = this.inputConv(xF, kernelF, biasF, this.padding);\n      xC = this.inputConv(xC, kernelC, biasC, this.padding);\n      xO = this.inputConv(xO, kernelO, biasO, this.padding);\n\n      const [recKernelI, recKernelF, recKernelC, recKernelO]: tfc.Tensor[] =\n          tfc.split(\n              this.recurrentKernel.read(), numOfKernels, kernelChannelAxis);\n\n      hI = this.recurrentConv(hI, recKernelI);\n      hF = this.recurrentConv(hF, recKernelF);\n      hC = this.recurrentConv(hC, recKernelC);\n      hO = this.recurrentConv(hO, recKernelO);\n\n      const i = this.recurrentActivation.apply(tfc.add(xI, hI));\n      const f = this.recurrentActivation.apply(tfc.add(xF, hF));\n      const c = tfc.add(\n          tfc.mul(f, cTMinus1),\n          tfc.mul(i, this.activation.apply(tfc.add(xC, hC))));\n      const h = tfc.mul(\n          this.recurrentActivation.apply(tfc.add(xO, hO)),\n          this.activation.apply(c));\n\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): tfc.serialization.ConfigDict {\n    const {'units': _, ...baseConfig} = super.getConfig();\n\n    const config: tfc.serialization.ConfigDict = {\n      filters: this.filters,\n      kernelSize: this.kernelSize,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      strides: this.strides,\n    };\n\n    return {...baseConfig, ...config};\n  }\n\n  inputConv(x: Tensor, w: Tensor, b?: Tensor, padding?: PaddingMode) {\n    const out = tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, this.strides as [number, number],\n        (padding || 'valid') as 'same' | 'valid',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC',\n        this.dilationRate as [number, number]);\n\n    if (b) {\n      return K.biasAdd(out, b, this.dataFormat) as tfc.Tensor3D;\n    }\n\n    return out;\n  }\n\n  recurrentConv(x: Tensor, w: Tensor) {\n    const strides = 1;\n\n    return tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, strides, 'same',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC');\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2DCell);\n\nexport declare interface ConvLSTM2DArgs extends\n    Omit<LSTMLayerArgs, 'units'|'cell'>, ConvRNN2DLayerArgs {}\n\nexport class ConvLSTM2D extends ConvRNN2D {\n  /** @nocollapse */\n  static className = 'ConvLSTM2D';\n\n  constructor(args: ConvLSTM2DArgs) {\n    const cell = new ConvLSTM2DCell(args);\n\n    super({...args, cell} as ConvRNN2DLayerArgs);\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends tfc.serialization.Serializable>(\n      cls: tfc.serialization.SerializableConstructor<T>,\n      config: tfc.serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Basic Layers.\n */\n\nimport {any, notEqual, serialization, Tensor, tidy, transpose, util} from '@tensorflow/tfjs-core';\n\nimport {Activation as ActivationFn, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {DisposeResult, InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {LayerConfig} from '../keras_format/topology_config';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {assertPositiveInteger, mapActivationToFusedKernel} from '../utils/generic_utils';\nimport {arrayProd, range} from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface DropoutLayerArgs extends LayerArgs {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /**\n   * Integer array representing the shape of the binary dropout mask that will\n   * be multiplied with the input.\n   *\n   * For instance, if your inputs have shape `(batchSize, timesteps, features)`\n   * and you want the dropout mask to be the same for all timesteps, you can use\n   * `noise_shape=(batch_size, 1, features)`.\n   */\n  noiseShape?: number[];\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class Dropout extends Layer {\n  /** @nocollapse */\n  static className = 'Dropout';\n  private readonly rate: number;\n  private readonly noiseShape: number[];\n  private readonly seed: number;\n\n  constructor(args: DropoutLayerArgs) {\n    super(args);\n    this.rate = Math.max(Math.min(args.rate, 1), 0);\n    // So that the scalar doesn't get tidied up between executions.\n    this.noiseShape = args.noiseShape;\n    this.seed = args.seed;\n    this.supportsMasking = true;\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    if (this.noiseShape == null) {\n      return this.noiseShape;\n    }\n    const inputShape = input.shape;\n    const noiseShape: Shape = [];\n    for (let i = 0; i < this.noiseShape.length; ++i) {\n      noiseShape.push(\n          this.noiseShape[i] == null ? inputShape[i] : this.noiseShape[i]);\n    }\n    return noiseShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      if (0 < this.rate && this.rate < 1) {\n        const training =\n            kwargs['training'] == null ? false : kwargs['training'];\n        const noiseShape = this.getNoiseShape(input);\n        const output = K.inTrainPhase(\n            () => K.dropout(input, this.rate, noiseShape, this.seed),\n            () => input, training);\n        return output;\n      }\n      return inputs;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      rate: this.rate,\n      noiseShape: this.noiseShape,\n      seed: this.seed,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  dispose(): DisposeResult {\n    return super.dispose();\n  }\n}\nserialization.registerClass(Dropout);\n\nexport declare interface DenseLayerArgs extends LayerArgs {\n  /** Positive integer, dimensionality of the output space. */\n  units: number;\n  /**\n   * Activation function to use.\n   *\n   * If unspecified, no activation is applied.\n   */\n  activation?: ActivationIdentifier;\n  /** Whether to apply a bias. */\n  useBias?: boolean;\n  /**\n   * Initializer for the dense kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * If specified, defines inputShape as `[inputDim]`.\n   */\n  inputDim?: number;\n\n  /**\n   * Constraint for the kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the dense kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport interface SpatialDropout1DLayerConfig extends LayerConfig {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class SpatialDropout1D extends Dropout {\n  /** @nocollapse */\n  static className = 'SpatialDropout1D';\n\n  constructor(args: SpatialDropout1DLayerConfig) {\n    super(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    const inputShape = input.shape;\n    return [inputShape[0], 1, inputShape[2]];\n  }\n}\nserialization.registerClass(SpatialDropout1D);\n\nexport class Dense extends Layer {\n  /** @nocollapse */\n  static className = 'Dense';\n  private units: number;\n  // Default activation: Linear (none).\n  private activation: ActivationFn = null;\n  private useBias = true;\n  private kernelInitializer: Initializer;\n  private biasInitializer: Initializer;\n  private kernel: LayerVariable = null;\n  private bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n  private readonly kernelConstraint?: Constraint;\n  private readonly biasConstraint?: Constraint;\n  private readonly kernelRegularizer?: Regularizer;\n  private readonly biasRegularizer?: Regularizer;\n\n  constructor(args: DenseLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null &&\n        args.inputDim != null) {\n      // This logic is copied from Layer's constructor, since we can't\n      // do exactly what the Python constructor does for Dense().\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      this.batchInputShape = [batchSize, args.inputDim];\n    }\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(args.activation);\n    if (args.useBias != null) {\n      this.useBias = args.useBias;\n    }\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.supportsMasking = true;\n\n    this.inputSpec = [{minNDim: 2}];\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputLastDim = inputShape[inputShape.length - 1];\n    if (this.kernel == null) {\n      this.kernel = this.addWeight(\n          'kernel', [inputLastDim, this.units], null, this.kernelInitializer,\n          this.kernelRegularizer, true, this.kernelConstraint);\n      if (this.useBias) {\n        this.bias = this.addWeight(\n            'bias', [this.units], null, this.biasInitializer,\n            this.biasRegularizer, true, this.biasConstraint);\n      }\n    }\n\n    this.inputSpec = [{minNDim: 2, axes: {[-1]: inputLastDim}}];\n    this.built = true;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    outputShape[outputShape.length - 1] = this.units;\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Dense layer accepts only a single input.\n      const input = getExactlyOneTensor(inputs);\n      const fusedActivationName =\n          mapActivationToFusedKernel(this.activation.getClassName());\n      let output: Tensor;\n\n      if (fusedActivationName != null) {\n        output = K.dot(\n            input, this.kernel.read(), fusedActivationName,\n            this.bias ? this.bias.read() : null);\n      } else {\n        output = K.dot(input, this.kernel.read());\n        if (this.bias != null) {\n          output = K.biasAdd(output, this.bias.read());\n        }\n        if (this.activation != null) {\n          output = this.activation.apply(output);\n        }\n      }\n\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dense);\n\nexport declare interface FlattenLayerArgs extends LayerArgs {\n  /** Image data format: channeLast (default) or channelFirst. */\n  dataFormat?: DataFormat;\n}\n\nexport class Flatten extends Layer {\n  private dataFormat: DataFormat;\n\n  /** @nocollapse */\n  static className = 'Flatten';\n  constructor(args?: FlattenLayerArgs) {\n    args = args || {};\n    super(args);\n    this.inputSpec = [{minNDim: 3}];\n    this.dataFormat = args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    for (const dim of inputShape.slice(1)) {\n      if (dim == null) {\n        throw new ValueError(\n            `The shape of the input to \"Flatten\" is not fully defined ` +\n            `(got ${inputShape.slice(1)}). Make sure to pass a complete ` +\n            `\"input_shape\" or \"batch_input_shape\" argument to the first ` +\n            `layer in your model.`);\n      }\n    }\n    return [inputShape[0], arrayProd(inputShape, 1)];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n\n      let input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsFirst' && input.rank > 1) {\n        const permutation: number[] = [0];\n        for (let i = 2; i < input.rank; ++i) {\n          permutation.push(i);\n        }\n        permutation.push(1);\n        input = input.transpose(permutation);\n      }\n\n      return K.batchFlatten(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {};\n    if (this.dataFormat != null) {\n      config['dataFormat'] = this.dataFormat;\n    }\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Flatten);\n\nexport declare interface ActivationLayerArgs extends LayerArgs {\n  /**\n   * Name of the activation function to use.\n   */\n  activation: ActivationIdentifier;\n}\n\nexport class Activation extends Layer {\n  /** @nocollapse */\n  static className = 'Activation';\n  activation: ActivationFn;\n\n  constructor(args: ActivationLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n    this.activation = getActivation(args.activation);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      return this.activation.apply(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {activation: serializeActivation(this.activation)};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Activation);\n\nexport declare interface ReshapeLayerArgs extends LayerArgs {\n  /** The target shape. Does not include the batch axis. */\n  targetShape: Shape;\n}\n\nexport declare interface RepeatVectorLayerArgs extends LayerArgs {\n  /**\n   * The integer number of times to repeat the input.\n   */\n  n: number;\n}\n\nexport class RepeatVector extends Layer {\n  /** @nocollapse */\n  static className = 'RepeatVector';\n  readonly n: number;\n\n  constructor(args: RepeatVectorLayerArgs) {\n    super(args);\n    this.n = args.n;\n    this.inputSpec = [{ndim: 2}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], this.n, inputShape[1]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      return K.repeat(inputs, this.n);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      n: this.n,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(RepeatVector);\n\nexport class Reshape extends Layer {\n  /** @nocollapse */\n  static className = 'Reshape';\n  private targetShape: Shape;\n\n  constructor(args: ReshapeLayerArgs) {\n    super(args);\n    this.targetShape = args.targetShape;\n\n    // Make sure that all unknown dimensions are represented as `null`.\n    for (let i = 0; i < this.targetShape.length; ++i) {\n      if (this.isUnknown(this.targetShape[i])) {\n        this.targetShape[i] = null;\n      }\n    }\n  }\n\n  private isUnknown(dim: number): boolean {\n    return dim < 0 || dim == null;\n  }\n\n  /**\n   * Finds and replaces a missing dimension in output shape.\n   *\n   * This is a near direct port of the internal Numpy function\n   * `_fix_unknown_dimension` in `numpy/core/src/multiarray/shape.c`.\n   *\n   * @param inputShape: Original shape of array begin reshape.\n   * @param outputShape: Target shape of the array, with at most a single\n   * `null` or negative number, which indicates an underdetermined dimension\n   * that should be derived from `inputShape` and the known dimensions of\n   *   `outputShape`.\n   * @returns: The output shape with `null` replaced with its computed value.\n   * @throws: ValueError: If `inputShape` and `outputShape` do not match.\n   */\n  private fixUnknownDimension(inputShape: Shape, outputShape: Shape): Shape {\n    const errorMsg = 'Total size of new array must be unchanged.';\n    const finalShape = outputShape.slice();\n    let known = 1;\n    let unknown = null;\n    for (let i = 0; i < finalShape.length; ++i) {\n      const dim = finalShape[i];\n      if (this.isUnknown(dim)) {\n        if (unknown === null) {\n          unknown = i;\n        } else {\n          throw new ValueError('Can only specifiy one unknown dimension.');\n        }\n      } else {\n        known *= dim;\n      }\n    }\n\n    const originalSize = arrayProd(inputShape);\n    if (unknown !== null) {\n      if (known === 0 || originalSize % known !== 0) {\n        throw new ValueError(errorMsg);\n      }\n      finalShape[unknown] = originalSize / known;\n    } else if (originalSize !== known) {\n      throw new ValueError(errorMsg);\n    }\n\n    return finalShape;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    let anyUnknownDims = false;\n    for (let i = 0; i < inputShape.length; ++i) {\n      if (this.isUnknown(inputShape[i])) {\n        anyUnknownDims = true;\n        break;\n      }\n    }\n\n    if (anyUnknownDims) {\n      return inputShape.slice(0, 1).concat(this.targetShape);\n    } else {\n      return inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const outputShape = inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n      return input.reshape(outputShape);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      targetShape: this.targetShape,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Reshape);\n\nexport declare interface PermuteLayerArgs extends LayerArgs {\n  /**\n   * Array of integers. Permutation pattern. Does not include the\n   * sample (batch) dimension. Index starts at 1.\n   * For instance, `[2, 1]` permutes the first and second dimensions\n   * of the input.\n   */\n  dims: number[];\n}\n\nexport class Permute extends Layer {\n  /** @nocollapse */\n  static className = 'Permute';\n  readonly dims: number[];\n  private readonly dimsIncludingBatch: number[];\n\n  constructor(args: PermuteLayerArgs) {\n    super(args);\n    if (args.dims == null) {\n      throw new Error(\n          'Required configuration field `dims` is missing during Permute ' +\n          'constructor call.');\n    }\n    if (!Array.isArray(args.dims)) {\n      throw new Error(\n          'Permute constructor requires `dims` to be an Array, but received ' +\n          `${args.dims} instead.`);\n    }\n\n    // Check the validity of the permutation indices.\n    const expectedSortedIndices = range(1, args.dims.length + 1);\n    if (!util.arraysEqual(args.dims.slice().sort(), expectedSortedIndices)) {\n      throw new Error(\n          'Invalid permutation `dims`: ' + JSON.stringify(args.dims) +\n          ' `dims` must contain consecutive integers starting from 1.');\n    }\n\n    this.dims = args.dims;\n    this.dimsIncludingBatch = [0].concat(this.dims);\n    this.inputSpec = [new InputSpec({ndim: this.dims.length + 1})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    this.dims.forEach((dim: number, i: number) => {\n      outputShape[i + 1] = (inputShape as Shape)[dim];\n    });\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return transpose(getExactlyOneTensor(inputs), this.dimsIncludingBatch);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      dims: this.dims,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Permute);\n\nexport declare interface MaskingArgs extends LayerArgs {\n  /**\n   * Masking Value. Defaults to `0.0`.\n   */\n  maskValue?: number;\n}\n\nexport class Masking extends Layer {\n  /** @nocollapse */\n  static className = 'Masking';\n  maskValue: number;\n\n  constructor(args?: MaskingArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maskValue = args.maskValue == null ? 0 : args.maskValue;\n    } else {\n      this.maskValue = 0;\n    }\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig() {\n    const baseConfig = super.getConfig();\n    const config = {maskValue: this.maskValue};\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    const input = getExactlyOneTensor(inputs);\n    const axis = -1;\n    return any(notEqual(input, this.maskValue), axis);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const axis = -1;\n      const keepDims = true;\n      const booleanMask = any(notEqual(input, this.maskValue), axis, keepDims);\n      const output = input.mul(booleanMask.asType(input.dtype));\n      return output;\n    });\n  }\n}\nserialization.registerClass(Masking);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Embedding Layer.\n *\n * Original source: keras/constraints.py\n */\nimport {notEqual, serialization, Tensor, tidy, zerosLike} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface EmbeddingLayerArgs extends LayerArgs {\n  /**\n   * Integer > 0. Size of the vocabulary, i.e. maximum integer index + 1.\n   */\n  inputDim: number;\n  /**\n   * Integer >= 0. Dimension of the dense embedding.\n   */\n  outputDim: number;\n  /**\n   * Initializer for the `embeddings` matrix.\n   */\n  embeddingsInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Regularizer function applied to the `embeddings` matrix.\n   */\n  embeddingsRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Constraint function applied to the `embeddings` matrix.\n   */\n  embeddingsConstraint?: ConstraintIdentifier|Constraint;\n  /**\n   * Whether the input value 0 is a special \"padding\" value that should be\n   * masked out. This is useful when using recurrent layers which may take\n   * variable length input.\n   *\n   * If this is `True` then all subsequent layers in the model need to support\n   * masking or an exception will be raised. If maskZero is set to `True`, as a\n   * consequence, index 0 cannot be used in the vocabulary (inputDim should\n   * equal size of vocabulary + 1).\n   */\n  maskZero?: boolean;\n  /**\n   * Length of input sequences, when it is constant.\n   *\n   * This argument is required if you are going to connect `flatten` then\n   * `dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed).\n   */\n  inputLength?: number|number[];\n}\n\nexport class Embedding extends Layer {\n  /** @nocollapse */\n  static className = 'Embedding';\n  private inputDim: number;\n  private outputDim: number;\n  private embeddingsInitializer: Initializer;\n  private maskZero: boolean;\n  private inputLength: number|number[];\n\n  private embeddings: LayerVariable = null;\n\n  readonly DEFAULT_EMBEDDINGS_INITIALIZER: InitializerIdentifier =\n      'randomUniform';\n  private readonly embeddingsRegularizer?: Regularizer;\n  private readonly embeddingsConstraint?: Constraint;\n\n  constructor(args: EmbeddingLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null) {\n      // Porting Note: This logic is copied from Layer's constructor, since we\n      // can't do exactly what the Python constructor does for Embedding().\n      // Specifically, the super constructor can not be called after the\n      // mutation of the `config` argument.\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      if (args.inputLength == null) {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (None, )\n        this.batchInputShape = [batchSize, null];\n      } else {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (config.inputLength, )\n        this.batchInputShape =\n            [batchSize].concat(generic_utils.toList(args.inputLength));\n      }\n    }\n    this.inputDim = args.inputDim;\n    generic_utils.assertPositiveInteger(this.inputDim, 'inputDim');\n    this.outputDim = args.outputDim;\n    generic_utils.assertPositiveInteger(this.outputDim, 'outputDim');\n    this.embeddingsInitializer = getInitializer(\n        args.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER);\n    this.embeddingsRegularizer = getRegularizer(args.embeddingsRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.embeddingsConstraint = getConstraint(args.embeddingsConstraint);\n    this.maskZero = args.maskZero;\n    this.supportsMasking = args.maskZero;\n    this.inputLength = args.inputLength;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    this.embeddings = this.addWeight(\n        'embeddings', [this.inputDim, this.outputDim], this.dtype,\n        this.embeddingsInitializer, this.embeddingsRegularizer, true,\n        this.embeddingsConstraint);\n    this.built = true;\n  }\n\n  // Override warnOnIncompatibleInputShape because an embedding layer allows\n  // the input to have varying ranks.\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {}\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tidy(() => {\n      if (!this.maskZero) {\n        return null;\n      } else {\n        inputs = getExactlyOneTensor(inputs);\n        return notEqual(inputs, zerosLike(inputs));\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    if (this.inputLength == null) {\n      return [...inputShape, this.outputDim];\n    }\n    // inputLength can be an array if input is 3D or higher.\n    const inLens: number[] = generic_utils.toList(this.inputLength);\n    if (inLens.length !== inputShape.length - 1) {\n      throw new ValueError(\n          `\"inputLength\" is ${this.inputLength}, but received ` +\n          `input shape has shape ${inputShape}`);\n    } else {\n      let i = 0;\n      for (let k = 0; k < inLens.length; ++k) {\n        const s1 = inLens[k];\n        const s2 = inputShape[k + 1];\n        if ((s1 != null) && (s2 != null) && (s1 !== s2)) {\n          throw new ValueError(\n              `\"inputLength\" is ${this.inputLength}, but received ` +\n              `input shape has shape ${inputShape}`);\n        } else if (s1 == null) {\n          inLens[i] = s2;\n        }\n        i++;\n      }\n    }\n    return [inputShape[0], ...inLens, this.outputDim];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Embedding layer accepts only a single input.\n      let input = getExactlyOneTensor(inputs);\n      if (input.dtype !== 'int32') {\n        input = K.cast(input, 'int32');\n      }\n      const output = K.gather(this.embeddings.read(), input.as1D());\n      return output.reshape(\n          getExactlyOneShape(this.computeOutputShape(input.shape)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      inputDim: this.inputDim,\n      outputDim: this.outputDim,\n      embeddingsInitializer: serializeInitializer(this.embeddingsInitializer),\n      embeddingsRegularizer: serializeRegularizer(this.embeddingsRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      embeddingsConstraint: serializeConstraint(this.embeddingsConstraint),\n      maskZero: this.maskZero,\n      inputLength: this.inputLength\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Embedding);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Merge Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {l2Normalize} from '../losses';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as mathUtils from '../utils/math_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\n/**\n * Generic Merge layer for element-wise merge functions.\n *\n * Used to implement `Sum`, `Average`, `Concatenate`, etc.\n */\nexport abstract class Merge extends Layer {\n  protected reshapeRequired: boolean;\n\n  constructor(args?: LayerArgs) {\n    super(args || {});\n    this.supportsMasking = true;\n  }\n\n  /**\n   * Logic for merging multiple tensors, to be overridden by subclasses.\n   * @param inputs\n   */\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    throw new NotImplementedError();\n  }\n\n  /**\n   * Computes the shape of the result of an elementwise operation.\n   *\n   * @param shape1: Shape of the first tensor.\n   * @param shape2: Shape of the second tensor.\n   * @returns Expected output shape when an elementwise operation is carried\n   *   out on 2 tensors with shapes `shape1` and `shape2`.\n   * @throws ValueError: If `shape1` and `shape2` are not compatible for\n   *   element-wise operations.\n   */\n  private computeElementwiseOpOutputShape(shape1: Shape, shape2: Shape): Shape {\n    if (shape1 == null || shape2 == null) {\n      return null;\n    } else if (shape1.length < shape2.length) {\n      return this.computeElementwiseOpOutputShape(shape2, shape1);\n    } else if (shape2.length === 0) {\n      return shape1;\n    }\n    const outputShape: Shape = shape1.slice(0, shape1.length - shape2.length);\n    for (let k = 0; k < shape2.length; ++k) {\n      const i = shape1[shape1.length - shape2.length + k];\n      const j = shape2[k];\n      if (i == null || j == null || i < 0 || j < 0) {\n        outputShape.push(null);\n      } else if (i === 1) {\n        outputShape.push(j);\n      } else if (j === 1) {\n        outputShape.push(i);\n      } else {\n        if (i !== j) {\n          throw new ValueError(\n              'Operands could not be broadcast together with shapes ' +\n              JSON.stringify(shape1) + ' ' + JSON.stringify(shape2));\n        }\n        outputShape.push(i);\n      }\n    }\n    return outputShape;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.\n    if (Array.isArray(inputShape) && !Array.isArray(inputShape[0])) {\n      // Make sure that inputShape is an Array of shape.\n      inputShape = [getExactlyOneShape(inputShape)];\n    }\n    inputShape = inputShape as Shape[];\n    if (inputShape.length < 2) {\n      throw new ValueError(\n          'A merge layer should be called on an Array of at least 2 inputs.' +\n          ` Got ${inputShape.length} input(s).`);\n    }\n\n    // Make sure that there is at most one unique batch size among the input\n    // shapes.\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length > 1) {\n      throw new ValueError(\n          `Can not merge tensors with different batch sizes. ` +\n          `Got tensors with shapes: ${JSON.stringify(inputShape)}.`);\n    }\n\n    let outputShape: Shape =\n        inputShape[0] == null ? null : inputShape[0].slice(1);\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n    // If the inputs have different ranks, we have to reshape them to make them\n    // broadcastable.\n    const allRanks = inputShape.map(shape => shape.length);\n    if (inputShape.indexOf(null) === -1 &&\n        generic_utils.unique(allRanks).length === 1) {\n      this.reshapeRequired = false;\n    } else {\n      this.reshapeRequired = true;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (this.reshapeRequired) {\n        const reshapedInputs: Tensor[] = [];\n        const inputDims = inputs.map(input => input.rank);\n        if (inputDims.indexOf(null) === -1) {\n          // If ranks of all inputs are available, we simply expand each of them\n          // at axis=1 until all of them have the same rank.\n          const maxNDim = mathUtils.max(inputDims);\n          for (let x of inputs) {\n            const xNDim = x.rank;\n            for (let k = 0; k < maxNDim - xNDim; ++k) {\n              x = K.expandDims(x, 1);\n            }\n            reshapedInputs.push(x);\n          }\n          return this.mergeFunction(reshapedInputs);\n        } else {\n          // Transpose all inputs so that batch size is the last dimension.\n          // [batchSize, dim1, dim2, ...] -> [dim1, dim2, ..., batchSize]\n          let transposed = false;\n          for (const x of inputs) {\n            const xNDim = x.rank;\n            if (xNDim == null) {\n              const xShape = x.shape;\n              const batchSize = xShape[0];\n              const newShape = xShape.slice(1).concat([batchSize]);\n              let xTransposed = x.reshape(\n                  [batchSize].concat(mathUtils.arrayProd(xShape.slice(1))));\n              xTransposed = tfc.transpose(xTransposed, [1, 0]);\n              xTransposed = xTransposed.reshape(newShape);\n              reshapedInputs.push(xTransposed);\n              transposed = true;\n            } else if (xNDim > 1) {\n              const dims = mathUtils.range(1, xNDim).concat([0]);\n              reshapedInputs.push(tfc.transpose(x, dims));\n              transposed = true;\n            } else {\n              // We don't transpose inputs if they are 1D vectors or scalars.\n              reshapedInputs.push(x);\n            }\n          }\n          let y = this.mergeFunction(reshapedInputs);\n          const yNDim = y.rank;\n          if (transposed) {\n            // If inputs have been transposed, we have to transpose the output\n            // too.\n            if (yNDim == null) {\n              const yShape = y.shape;\n              const yNDim = yShape.length;\n              const batchSize = yShape[yNDim - 1];\n              const newShape =\n                  [batchSize].concat(yShape.slice(0, yShape.length - 1));\n              y = tfc.transpose(y.reshape([-1, batchSize]), [1, 0])\n                      .reshape(newShape);\n            } else if (yNDim > 1) {\n              const dims = [yNDim - 1].concat(mathUtils.range(0, yNDim - 1));\n              y = tfc.transpose(y, dims);\n            }\n          }\n          return y;\n        }\n      } else {\n        return this.mergeFunction(inputs);\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape[];\n    let outputShape: Shape;\n    if (inputShape[0] == null) {\n      outputShape = null;\n    } else {\n      outputShape = inputShape[0].slice(1);\n    }\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length === 1) {\n      outputShape = batchSizes.concat(outputShape);\n    } else {\n      outputShape = [null].concat(outputShape);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tfc.tidy(() => {\n      if (mask == null) {\n        return null;\n      }\n      if (!Array.isArray(mask)) {\n        throw new ValueError('`mask` should be an Array');\n      }\n      if (!Array.isArray(inputs)) {\n        throw new ValueError('`inputs` should be an Array');\n      }\n      if (mask.length !== inputs.length) {\n        throw new ValueError(\n            `The Array 'inputs' and 'mask' are expected to have the same ` +\n            `length, but have different lengths ` +\n            `(${inputs.length} vs ${mask.length})`);\n      }\n      if (mask.every(m => m == null)) {\n        return null;\n      }\n      mask = mask.map(m => m == null ? m : tfc.expandDims(m, 0));\n      let output = mask[0];\n      for (let i = 1; i < mask.length - 1; ++i) {\n        output = tfc.logicalAnd(output, mask[i]);\n      }\n      return output;\n    });\n  }\n}\n\nexport class Add extends Merge {\n  /** @nocollapse */\n  static className = 'Add';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Add);\n\n/**\n * Calculate the element-wise sum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Add` layer, by using no input argument\n *    or a single configuration argument. The resultant `Add` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const addLayer = tf.layers.add();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = addLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.add([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.add([input1, input2]).print();\n * // Gives [[11, 22], [33, 44]].\n *\n */\nexport function add(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Add({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Add(config);\n  }\n}\n\nexport class Multiply extends Merge {\n  /** @nocollapse */\n  static className = 'Multiply';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.mul(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Multiply);\n\n/**\n * Calculate the element-wise product of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Multiply` layer, by using no input argument\n *    or a single configuration argument. The resultant `Multiply` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const multiplyLayer = tf.layers.multiply();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = multiplyLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.multiply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.multiply([input1, input2]).print();\n * // Gives [[10, 40], [90, 160]].\n *\n */\nexport function multiply(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Multiply({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Multiply(config);\n  }\n}\n\nexport class Average extends Merge {\n  /** @nocollapse */\n  static className = 'Average';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return tfc.mul(1 / inputs.length, output);\n    });\n  }\n}\nserialization.registerClass(Average);\n\n/**\n * Calculate the element-wise arithmetic mean of inputs, which all have the same\n * shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Average` layer, by using no input argument\n *    or a single configuration argument. The resultant `Average` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const averageLayer = tf.layers.average();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = averageLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.average([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.average([input1, input2]).print();\n * // Gives [[5.5, 11], [16.5, 22]].\n *\n */\nexport function average(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Average({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Average(config);\n  }\n}\n\nexport class Maximum extends Merge {\n  /** @nocollapse */\n  static className = 'Maximum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.maximum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Maximum);\n\n/**\n * Calculate the element-wise maximum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Maximum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Maximum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const maximumLayer = tf.layers.maximum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = maximumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.maximum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.maximum([input1, input2]).print();\n * // Gives [[10, 20], [30, 40]].\n *\n */\nexport function maximum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Maximum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Maximum(config);\n  }\n}\n\nexport class Minimum extends Merge {\n  /** @nocollapse */\n  static className = 'Minimum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.minimum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Minimum);\n\n/**\n * Calculate the element-wise minimum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Minimum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Minimum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const minimumLayer = tf.layers.minimum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = minimumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.minimum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.minimum([input1, input2]).print();\n * // Gives [[1, 2], [3, 4]].\n *\n */\nexport function minimum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Minimum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Minimum(config);\n  }\n}\n\nexport declare interface ConcatenateLayerArgs extends LayerArgs {\n  /**\n   * Axis along which to concatenate.\n   */\n  axis?: number;\n}\n\nexport class Concatenate extends Merge {\n  /** @nocollapse */\n  static className = 'Concatenate';\n  readonly DEFAULT_AXIS = -1;\n  private readonly axis: number;\n\n  constructor(args?: ConcatenateLayerArgs) {\n    super(args);\n    if (args == null) {\n      args = {};\n    }\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.]\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0])) ||\n        inputShape.length === 1) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of at least 2 ' +\n          'inputs');\n    }\n    inputShape = inputShape as Shape[];\n\n    let allNoneShape = true;\n    for (const shape of inputShape) {\n      if (shape != null) {\n        allNoneShape = false;\n        break;\n      }\n    }\n    if (allNoneShape) {\n      return;\n    }\n\n    const shapeSet: Shape[] = [];\n    for (let i = 0; i < inputShape.length; ++i) {\n      const shapeWithoutConcatAxis = inputShape[i].slice();\n      shapeWithoutConcatAxis.splice(this.axis, 1);\n      let exists = false;\n      for (const shape of shapeSet) {\n        if (util.arraysEqual(shape, shapeWithoutConcatAxis)) {\n          exists = true;\n          break;\n        }\n      }\n      if (!exists) {\n        shapeSet.push(shapeWithoutConcatAxis);\n      }\n    }\n    if (shapeSet.length > 1) {\n      throw new ValueError(\n          'A `Concatenate` layer requires inputs with matching shapes ' +\n          'except for the concat axis. Got input shapes: ' +\n          JSON.stringify(inputShape));\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      return K.concatenate(inputs, this.axis);\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0]))) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of inputs.');\n    }\n    const inputShapes = inputShape as Shape[];\n    const outputShape = inputShapes[0].slice();\n    const axis = this.axis < 0 ? outputShape.length + this.axis : this.axis;\n    // Porting Note: the line above is because TypeScript doesn't support\n    //   negative indices.\n    for (const shape of inputShapes.slice(1)) {\n      if (outputShape[axis] == null || shape[axis] == null) {\n        outputShape[axis] = null;\n        break;\n      }\n      outputShape[axis] += shape[axis];\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    if (mask == null) {\n      return null;\n    }\n    if (!Array.isArray(mask)) {\n      throw new ValueError('`mask` should be an array for Concatenate');\n    }\n    if (!Array.isArray(inputs)) {\n      throw new ValueError('`inputs` should be an array for Concatenate');\n    }\n    if (mask.length !== inputs.length) {\n      throw new ValueError(\n          `Mismatch in the length of mask (${mask.length}) ` +\n          `and the legnth of inputs (${inputs.length})`);\n    }\n    return tfc.tidy(() => {\n      let allNullMasks = true;\n      mask.forEach(m => {\n        if (m != null) {\n          allNullMasks = false;\n          return;\n        }\n      });\n      if (allNullMasks) {\n        return null;\n      }\n      const outputMasks: Tensor[] = [];\n      for (let i = 0; i < inputs.length; ++i) {\n        if (mask[i] == null) {\n          // Input is unmasked. Append all 1's to masks.\n          outputMasks.push(tfc.onesLike(inputs[i]).asType('bool'));\n        } else if (mask[i].rank < inputs[i].rank) {\n          // Mask is smaller than the input, expand it.\n          outputMasks.push(tfc.expandDims(mask[i], -1));\n        } else {\n          outputMasks.push(mask[i]);\n        }\n      }\n      const concatenatedMasks = tfc.concat(outputMasks, this.axis);\n      return tfc.all(concatenatedMasks, -1, false);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axis': this.axis,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Concatenate);\n\n/**\n * Concatenate an `Array` of inputs.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Concatenate` layer, by using no input argument\n *    or a single configuration argument. The resultant `Concatenate` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const concatLayer = tf.layers.concatenate();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = concatLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 7], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = tf.layers.concatenate([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([[1, 2], [3, 4]], [2, 2]);\n * const input2 = tf.tensor2d([[10, 20], [30, 40]], [2, 2]);\n * tf.layers.concatenate([input1, input2]).print();\n * // Gives [[1, 2, 10, 20], [3, 4, 30, 40]].\n *\n */\nexport function concatenate(config?: SymbolicTensor[]|Tensor[]|\n                            ConcatenateLayerArgs): Layer|SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Concatenate({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Concatenate(config);\n  }\n}\n\nexport declare interface DotLayerArgs extends LayerArgs {\n  /**\n   * Axis or axes along which the dot product will be taken.\n   *\n   * Integer or an Array of integers.\n   */\n  axes: number|[number, number];\n\n  /**\n   * Whether to L2-normalize samples along the dot product axis\n   * before taking the dot product.\n   *\n   * If set to `true`, the output of the dot product isthe cosine\n   * proximity between the two samples.\n   */\n  normalize?: boolean;\n}\n\n/**\n * Interpretable potentially negative axis index.\n *\n * For example, given axis = -1, and dim = 3, this function will return 2.\n *\n * @param axis The axis index, may be a positive, zero or negative integer.\n * @param dim Total number of dimensions, a positive integer.\n * @returns A non-negative axis index equivalent to the input `axis`.\n */\nfunction interpretAxis(axis: number, dim: number): number {\n  while (axis < 0) {\n    axis += dim;\n  }\n  return axis;\n}\n\nfunction batchDot(x: Tensor, y: Tensor, axes: number|[number, number]): Tensor {\n  if (x.shape.length > 3 || y.shape.length > 3) {\n    throw new NotImplementedError(\n        'batchDot is not implemented for tensors of 4D or higher rank yet');\n  }\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of x to be >= 2, ` +\n          `but got ${x.shape.length}`);\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of y to be >= 2, ` +\n          `but got ${y.shape.length}`);\n\n  if (typeof axes === 'number') {\n    axes = [axes, axes];\n  }\n\n  if (x.dtype === 'complex64' || y.dtype === 'complex64') {\n    throw new NotImplementedError(\n        'batchDot is not implemented for complex64-type Tensors yet.');\n  }\n\n  const xNDim = x.shape.length;\n  const yNDim = y.shape.length;\n  if (axes == null) {\n    // Behave like batchMatmul by default.\n    axes = [xNDim - 1, yNDim - 2];\n  }\n  const axesArray = axes as [number, number];\n\n  return tfc.tidy(() => {\n    let diff: number;\n    if (xNDim > yNDim) {\n      diff = xNDim - yNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      y = y.reshape(y.shape.concat(diffShape));\n    } else if (yNDim > xNDim) {\n      diff = yNDim - xNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      x = x.reshape(x.shape.concat(diffShape));\n    } else {\n      diff = 0;\n    }\n\n    let out: Tensor;\n    if (x.shape.length === 2 && y.shape.length === 2) {\n      if (axesArray[0] === axesArray[1]) {\n        out = x.mul(y).sum(axesArray[0]);\n      } else {\n        out = x.transpose([1, 0]).mul(y).sum(axesArray[1]);\n      }\n    } else {\n      const adjX = axesArray[0] !== x.shape.length - 1;\n      const adjY = axesArray[1] === y.shape.length - 1;\n      out = x.matMul(y, adjX, adjY);\n    }\n\n    if (diff > 0) {\n      let idx: number;\n      if (xNDim > yNDim) {\n        idx = xNDim + yNDim - 3;\n      } else {\n        idx = xNDim - 1;\n      }\n      const squeezeAxes: number[] = [];\n      for (let i = idx; i < idx + diff; ++i) {\n        squeezeAxes.push(i);\n      }\n      out = out.squeeze(squeezeAxes);\n    }\n    if (out.shape.length === 1) {\n      out = out.expandDims(1);\n    }\n    return out;\n  });\n}\n\nexport class Dot extends Merge {\n  /** @nocollapse */\n  static className = 'Dot';\n\n  private axes: number|[number, number];\n  private normalize: boolean;\n\n  constructor(args: DotLayerArgs) {\n    super(args);\n    this.axes = args.axes;\n    this.normalize = args.normalize == null ? false : args.normalize;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = inputShape[0] as Shape;\n    const shape2 = inputShape[1] as Shape;\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    if (shape1[axes[0]] !== shape2[axes[1]]) {\n      throw new ValueError(\n          `Dimension incompatibility: ` +\n          `${shape1[axes[0]]} !== ${shape2[axes[1]]}`);\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    if (inputs.length !== 2) {\n      throw new ValueError(\n          'A `Dot` layer must be called on exactly 2 inputs, ' +\n          `but received ${inputs.length} input(s).`);\n    }\n\n    let x1 = inputs[0];\n    let x2 = inputs[1];\n    let axes: [number, number];\n    if (!Array.isArray(this.axes)) {\n      axes = [\n        interpretAxis(this.axes, x1.shape.length),\n        interpretAxis(this.axes, x2.shape.length)\n      ];\n    } else {\n      axes = this.axes.map(\n                 (axis, i) => interpretAxis(\n                     axis, inputs[i].shape.length)) as [number, number];\n    }\n    if (this.normalize) {\n      x1 = l2Normalize(x1, axes[0]);\n      x2 = l2Normalize(x2, axes[1]);\n    }\n    return batchDot(x1, x2, axes);\n  }\n\n  private interpretAxes(shape1: Shape, shape2: Shape): number[] {\n    let axes: number[];\n    if (!Array.isArray(this.axes)) {\n      // `this.axes` is a single integer.\n      axes = [\n        interpretAxis(this.axes, shape1.length),\n        interpretAxis(this.axes, shape2.length)\n      ];\n    } else {\n      // `this.axes` is an Array of integers.\n      axes = this.axes;\n    }\n    return axes;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = (inputShape[0] as Shape).slice();\n    const shape2 = (inputShape[1] as Shape).slice();\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    shape1.splice(axes[0], 1);\n    shape2.splice(axes[1], 1);\n    shape2.splice(0, 1);\n    const outputShape = shape1.concat(shape2);\n    if (outputShape.length === 1) {\n      outputShape.push(1);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return null;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axes': this.axes,\n      'normalize': this.normalize\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dot);\n\n// TODO(cais): Add functional interfaces for the merge layers.\n", "/**\r\n * @license\r\n * Copyright 2018 Google LLC\r\n *\r\n * Use of this source code is governed by an MIT-style\r\n * license that can be found in the LICENSE file or at\r\n * https://opensource.org/licenses/MIT.\r\n * =============================================================================\r\n */\r\n\r\n/**\r\n * TensorFlow.js Layers: Noise Layers.\r\n */\r\n\r\nimport {greaterEqual, randomUniform, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\r\n\r\nimport * as K from '../backend/tfjs_backend';\r\nimport {Layer, LayerArgs} from '../engine/topology';\r\nimport {Shape} from '../keras_format/common';\r\nimport {Kwargs} from '../types';\r\nimport {getExactlyOneTensor} from '../utils/types_utils';\r\n\r\nexport declare interface GaussianNoiseArgs extends LayerArgs {\r\n  /** Standard Deviation.  */\r\n  stddev: number;\r\n}\r\n\r\nexport class GaussianNoise extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianNoise';\r\n  readonly stddev: number;\r\n\r\n  constructor(args: GaussianNoiseArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.stddev = args.stddev;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {stddev: this.stddev};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      const noised = () =>\r\n          K.randomNormal(input.shape, 0, this.stddev).add(input);\r\n      const output =\r\n          K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      return output;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianNoise);\r\n\r\nexport declare interface GaussianDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n}\r\n\r\nexport class GaussianDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianDropout';\r\n  readonly rate: number;\r\n\r\n  constructor(args: GaussianDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      if (this.rate > 0 && this.rate < 1) {\r\n        const noised = () => {\r\n          const stddev = Math.sqrt(this.rate / (1 - this.rate));\r\n          return input.mul(K.randomNormal(input.shape, 1, stddev));\r\n        };\r\n        return K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      }\r\n      return input;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianDropout);\r\n\r\nexport declare interface AlphaDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n  /**\r\n   * A 1-D `Tensor` of type `int32`, representing the\r\n   * shape for randomly generated keep/drop flags.\r\n   */\r\n  noiseShape?: Shape;\r\n}\r\n\r\n/**\r\n * Applies Alpha Dropout to the input.\r\n *\r\n * As it is a regularization layer, it is only active at training time.\r\n *\r\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\r\n * to their original values, in order to ensure the self-normalizing property\r\n * even after this dropout.\r\n * Alpha Dropout fits well to Scaled Exponential Linear Units\r\n * by randomly setting activations to the negative saturation value.\r\n *\r\n * Arguments:\r\n *   - `rate`: float, drop probability (as with `Dropout`).\r\n *     The multiplicative noise will have\r\n *     standard deviation `sqrt(rate / (1 - rate))`.\r\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\r\n *     shape for randomly generated keep/drop flags.\r\n *\r\n * Input shape:\r\n *   Arbitrary. Use the keyword argument `inputShape`\r\n *   (tuple of integers, does not include the samples axis)\r\n *   when using this layer as the first layer in a model.\r\n *\r\n * Output shape:\r\n *   Same shape as input.\r\n *\r\n * References:\r\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\r\n */\r\nexport class AlphaDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'AlphaDropout';\r\n  readonly rate: number;\r\n  readonly noiseShape: Shape;\r\n\r\n  constructor(args: AlphaDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n    this.noiseShape = args.noiseShape;\r\n  }\r\n\r\n  _getNoiseShape(inputs: Tensor|Tensor[]) {\r\n    return this.noiseShape || getExactlyOneTensor(inputs).shape;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      if (this.rate < 1 && this.rate > 0) {\r\n        const noiseShape = this._getNoiseShape(inputs);\r\n\r\n        const droppedInputs = () => {\r\n          const input = getExactlyOneTensor(inputs);\r\n\r\n          const alpha = 1.6732632423543772848170429916717;\r\n          const scale = 1.0507009873554804934193349852946;\r\n\r\n          const alphaP = -alpha * scale;\r\n\r\n          let keptIdx = greaterEqual(randomUniform(noiseShape), this.rate);\r\n\r\n          keptIdx = K.cast(keptIdx, 'float32');  // get default dtype.\r\n\r\n          // Get affine transformation params.\r\n          const a = ((1 - this.rate) * (1 + this.rate * alphaP ** 2)) ** -0.5;\r\n          const b = -a * alphaP * this.rate;\r\n\r\n          // Apply mask.\r\n          const x = input.mul(keptIdx).add(keptIdx.add(-1).mul(alphaP));\r\n\r\n          return x.mul(a).add(b);\r\n        };\r\n        return K.inTrainPhase(\r\n            droppedInputs, () => getExactlyOneTensor(inputs),\r\n            kwargs['training'] || false);\r\n      }\r\n      return inputs;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(AlphaDropout);\r\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Normalization layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {moments, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Applies batch normalization on x given mean, var, beta and gamma.\n *\n * I.e. returns:\n *   `output = (x - mean) / (sqrt(var) + epsilon) * gamma + beta`\n *\n * @param x Input tensor.\n * @param mean Mean of batch.\n * @param variance Variance of batch.\n * @param beta Tensor with which to center the input.\n * @param gamma Tensor by which to scale the input.\n * @param epsilon Fuzz factor.\n * @returns The result of the batch normalization.\n */\nexport function batchNormalization(\n    x: Tensor, mean: Tensor, variance: Tensor, beta?: Tensor, gamma?: Tensor,\n    epsilon = 1e-3): Tensor {\n  let out: Tensor;\n  if (x.rank === 2) {\n    out = tfc.batchNorm2d(\n        x as Tensor2D, mean as Tensor2D | Tensor1D,\n        variance as Tensor2D | Tensor1D, beta as Tensor2D | Tensor1D,\n        gamma as Tensor2D | Tensor1D, epsilon);\n  } else if (x.rank === 3) {\n    // TODO(cais): Check rank; give proper error message.\n    out = tfc.batchNorm3d(\n        x as Tensor3D, mean as Tensor3D | Tensor1D,\n        variance as Tensor3D | Tensor1D, beta as Tensor3D | Tensor1D,\n        gamma as Tensor3D | Tensor1D, epsilon);\n  } else if (x.rank === 4) {\n    out = tfc.batchNorm4d(\n        x as Tensor4D, mean as Tensor4D | Tensor1D,\n        variance as Tensor4D | Tensor1D, beta as Tensor4D | Tensor1D,\n        gamma as Tensor4D | Tensor1D, epsilon);\n  } else {\n    throw new NotImplementedError(\n        `batchNormalization is not implemented for array of rank ${x.rank} ` +\n        `yet`);\n  }\n  return out;\n}\n\n/**\n * Non-broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction regularNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const normed =\n               batchNormalization(x, mean, variance, beta, gamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction broadcastNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const targetShape: number[] = [];\n           for (const axis of math_utils.range(0, x.rank)) {\n             if (reductionAxes.indexOf(axis) !== -1) {\n               targetShape.push(1);\n             } else {\n               targetShape.push(x.shape[axis]);\n             }\n           }\n           const broadcastMean = mean.reshape(targetShape);\n           const broadcastVariance = variance.reshape(targetShape);\n           const broadcastGamma =\n               gamma == null ? null : gamma.reshape(targetShape);\n           const broadcastBeta =\n               beta == null ? null : beta.reshape(targetShape);\n           const normed = batchNormalization(\n               x, broadcastMean, broadcastVariance, broadcastBeta,\n               broadcastGamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Batch normalization for use in training (not inference).\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nexport function normalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  if (util.arraysEqual(\n          reductionAxes.slice().sort(), math_utils.range(0, x.rank - 1))) {\n    return regularNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  } else {\n    return broadcastNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  }\n}\n\nexport declare interface BatchNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The integer axis that should be normalized (typically the features axis).\n   * Defaults to -1.\n   *\n   * For instance, after a `Conv2D` layer with `data_format=\"channels_first\"`,\n   * set `axis=1` in `batchNormalization`.\n   */\n  axis?: number;\n\n  /**\n   * Momentum of the moving average. Defaults to 0.99.\n   */\n  momentum?: number;\n\n  /**\n   * Small float added to the variance to avoid dividing by zero. Defaults to\n   * 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Defaults to `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear (also e.g. `nn.relu`),\n   * this can be disabled since the scaling will be done by the next layer.\n   * Defaults to `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   *  Defaults to 'zeros'.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   *  Defaults to `ones`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving mean.\n   * Defaults to `zeros`\n   */\n  movingMeanInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving variance.\n   *  Defaults to 'Ones'.\n   */\n  movingVarianceInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the beta weight.\n   */\n  betaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for gamma weight.\n   */\n  gammaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer for the beta weight.\n   */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer for the gamma weight.\n   */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class BatchNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'BatchNormalization';\n  private readonly axis: number;\n  private readonly momentum: number;\n  private readonly epsilon: number;\n  private readonly center: boolean;\n  private readonly scale: boolean;\n  private readonly betaInitializer: Initializer;\n  private readonly gammaInitializer: Initializer;\n  private readonly movingMeanInitializer: Initializer;\n  private readonly movingVarianceInitializer: Initializer;\n  private readonly betaConstraint: Constraint;\n  private readonly gammaConstraint: Constraint;\n  private readonly betaRegularizer: Regularizer;\n  private readonly gammaRegularizer: Regularizer;\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n  private movingMean: LayerVariable;\n  private movingVariance: LayerVariable;\n\n  constructor(args?: BatchNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.supportsMasking = true;\n    this.axis = args.axis == null ? -1 : args.axis;\n    this.momentum = args.momentum == null ? 0.99 : args.momentum;\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.movingMeanInitializer =\n        getInitializer(args.movingMeanInitializer || 'zeros');\n    this.movingVarianceInitializer =\n        getInitializer(args.movingVarianceInitializer || 'ones');\n    this.betaConstraint = getConstraint(args.betaConstraint);\n    this.gammaConstraint = getConstraint(args.gammaConstraint);\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const axis = this.axis >= 0 ? this.axis : (this.axis + inputShape.length);\n    const dim = inputShape[axis];\n    if (dim == null) {\n      throw new ValueError(\n          `Axis ${axis} of input tensor should have a defined dimension but ` +\n          `the layer received an input with shape ` +\n          `${JSON.stringify(inputShape)}.`);\n    }\n    this.inputSpec =\n        [new InputSpec({ndim: inputShape.length, axes: {[axis]: dim}})];\n    const shape = [dim];\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', shape, null, this.gammaInitializer, this.gammaRegularizer,\n          true, this.gammaConstraint);\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', shape, null, this.betaInitializer, this.betaRegularizer, true,\n          this.betaConstraint);\n    }\n    this.movingMean = this.addWeight(\n        'moving_mean', shape, null, this.movingMeanInitializer, null, false);\n    this.movingVariance = this.addWeight(\n        'moving_variance', shape, null, this.movingVarianceInitializer, null,\n        false);\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const ndim = inputShape.length;\n      const reductionAxes = math_utils.range(0, ndim);\n      const axis = this.axis >= 0 ? this.axis : (this.axis + ndim);\n      reductionAxes.splice(axis, 1);\n      const broadcastShape = generic_utils.pyListRepeat(1, ndim);\n      broadcastShape[axis] = inputShape[axis];\n\n      const sortedReductionAxes = reductionAxes.slice();\n      sortedReductionAxes.sort();\n      const needsBroadcasting = !util.arraysEqual(\n          sortedReductionAxes, math_utils.range(0, ndim).slice(0, ndim - 1));\n\n      const normalizeInference: () => Tensor = () => {\n        if (needsBroadcasting) {\n          const broadcastMovingMean =\n              this.movingMean.read().reshape(broadcastShape);\n          const broadcastMovingVariance =\n              this.movingVariance.read().reshape(broadcastShape);\n          const broadcastBeta =\n              this.center ? this.beta.read().reshape(broadcastShape) : null;\n          const broadcastGamma =\n              this.scale ? this.gamma.read().reshape(broadcastShape) : null;\n          return batchNormalization(\n              input, broadcastMovingMean, broadcastMovingVariance,\n              broadcastBeta, broadcastGamma, this.epsilon);\n        } else {\n          return batchNormalization(\n              input, this.movingMean.read(), this.movingVariance.read(),\n              this.beta == null ? null : this.beta.read(),\n              this.gamma == null ? null : this.gamma.read(), this.epsilon);\n        }\n      };\n\n      if (!training) {\n        return normalizeInference();\n      }\n\n      const [normedTraining, mean, variance] = normalizeBatchInTraining(\n          input, this.gamma.read(), this.beta.read(), reductionAxes,\n          this.epsilon);\n\n      const doMovingAverage =\n          (variable: LayerVariable, value: Tensor, momentum: number): void => {\n            tfc.tidy(() => {\n              const decay = 1 - momentum;\n              const origValue = variable.read();\n              const updateDelta = origValue.sub(value).mul(decay);\n              variable.write(origValue.sub(updateDelta));\n            });\n          };\n\n      // Perform updates to moving mean and moving variance for training.\n      // Porting Note: In PyKeras, these updates to `movingMean` and\n      //   `movingAverage` are done as a deferred Graph, added to the `Layer`'s\n      //   `update`s using the `add_update()` method. Here we do it imperatively\n      //   and encapsulate the updates in a function that is invoked\n      //   immediately.\n      const updateMovingMeanAndVariance = () => {\n        doMovingAverage(this.movingMean, mean, this.momentum);\n        doMovingAverage(this.movingVariance, variance, this.momentum);\n      };\n      updateMovingMeanAndVariance();\n\n      return normedTraining;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      momentum: this.momentum,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      movingMeanInitializer: serializeInitializer(this.movingMeanInitializer),\n      movingVarianceInitializer:\n          serializeInitializer(this.movingVarianceInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer),\n      betaConstraint: serializeConstraint(this.betaConstraint),\n      gammaConstraint: serializeConstraint(this.gammaConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(BatchNormalization);\n\nexport interface LayerNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The axis or axes that should be normalized (typically, the feature axis.)\n   * Defaults to -1 (the last axis.)\n   */\n  axis?: number|number[];\n\n  /**\n   * A small positive float added to variance to avoid divison by zero.\n   * Defaults to 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Default: `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply output by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear, this can be disabled since scaling will\n   * be done by the next layer.\n   * Default: `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   * Default: `'zeros'`.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   * Default: `'ones'`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /** Regularizer for the beta weight. */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /** Regularizer for the gamma weight. */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class LayerNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'LayerNormalization';\n\n  private axis: number|number[];\n  readonly epsilon: number;\n  readonly center: boolean;\n  readonly scale: boolean;\n  readonly betaInitializer: Initializer;\n  readonly gammaInitializer: Initializer;\n  readonly betaRegularizer: Regularizer;\n  readonly gammaRegularizer: Regularizer;\n\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n\n  constructor(args?: LayerNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.axis = args.axis == null ? -1 : args.axis;\n    if (typeof this.axis === 'number') {\n      if (!Number.isInteger(this.axis)) {\n        throw new Error(\n            `Expected axis to be an integer, but received ${this.axis}`);\n      }\n    } else if (Array.isArray(this.axis)) {\n      for (const axis of this.axis) {\n        if (!Number.isInteger(axis)) {\n          throw new Error(\n              `Expected axis to be an array of integers, ` +\n              `but received ${JSON.stringify(this.axis)}`);\n        }\n      }\n    } else {\n      throw new Error(\n          `Expected axis to be an integer or an array of integers, ` +\n          `but received ${JSON.stringify(this.axis)}`);\n    }\n\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n\n    this.supportsMasking = true;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const nDims = inputShape.length;\n\n    // Convert axis to array and resolve negatives.\n    if (typeof this.axis === 'number') {\n      this.axis = [this.axis];\n    }\n    for (let i = 0; i < this.axis.length; ++i) {\n      if (this.axis[i] < 0) {\n        this.axis[i] += nDims;\n      }\n    }\n\n    // Further validate axes.\n    for (const axis of this.axis) {\n      if (axis < 0 || axis >= nDims) {\n        throw new Error(`Invalid axis: ${axis}`);\n      }\n    }\n    if (this.axis.length !== generic_utils.unique(this.axis).length) {\n      throw new Error(`Found duplicate axes in: ${this.axis}`);\n    }\n\n    const paramShape = this.axis.map(axis => inputShape[axis]) as number[];\n\n    const trainable = true;\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', paramShape, 'float32', this.gammaInitializer,\n          this.gammaRegularizer, trainable);\n    } else {\n      this.gamma = null;\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', paramShape, 'float32', this.betaInitializer,\n          this.betaRegularizer, trainable);\n    } else {\n      this.beta = null;\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const input = getExactlyOneTensor(inputs);\n    const inputShape = input.shape;\n    const nDims = inputShape.length;\n\n    return tidy(() => {\n      const keepDims = true;\n      let {mean, variance} = moments(input, this.axis, keepDims);\n      const broadcastShape = generic_utils.pyListRepeat(1, nDims);\n      for (const dim of this.axis as number[]) {\n        broadcastShape[dim] = inputShape[dim];\n      }\n\n      const broadcast = (v: Tensor) => {\n        if (v != null && v.shape.length !== nDims &&\n            this.axis !== [nDims - 1]) {\n          return v.reshape(broadcastShape);\n        } else {\n          return v;\n        }\n      };\n\n      let scale = broadcast(this.gamma.read());\n      let offset = broadcast(this.beta.read());\n\n      // TODO(https://github.com/tensorflow/tfjs/issues/2120): The tiling below\n      // is a workaround for the limitation of core's batchNormalization?d don't\n      // support broadcasting in their gradients. In addition, the tiling is\n      // necessary to ensure correctness on the browser CPU backend regardless\n      // of forward or backward computation. Remove this workaround once the\n      // limitation is addressed. See .\n      const momentsTiling: number[] = [];\n      const scaleOffsetTiling: number[] = [];\n      for (let i = 0; i < nDims; ++i) {\n        if ((this.axis as number[]).indexOf(i) !== -1) {\n          momentsTiling.push(inputShape[i]);\n          scaleOffsetTiling.push(1);\n        } else {\n          momentsTiling.push(1);\n          scaleOffsetTiling.push(inputShape[i]);\n        }\n      }\n      mean = mean.tile(momentsTiling);\n      variance = variance.tile(momentsTiling);\n      scale = scale.tile(scaleOffsetTiling);\n      offset = offset.tile(scaleOffsetTiling);\n\n      return batchNormalization(\n          input, mean, variance, offset, scale, this.epsilon);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LayerNormalization);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Padding Layers.\n */\n\n// Porting Note: In Python Keras, the padding layers are in convolutional.py,\n//   but we decided to put them in a separate file (padding.ts) for clarity.\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\n/**\n * Pads the middle dimension of a 3D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of 2 integers, how many zeros to add at the start and\n *   end of the middle dimension (i.e., dimension 1).\n * @return A padded 3D `tf.Tensor`.\n */\nexport function temporalPadding(x: Tensor, padding?: [number, number]): Tensor {\n  return tidy(() => {\n    if (x.rank !== 3) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 3-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [1, 1];\n    }\n    if (padding.length !== 2) {\n      throw new ValueError(\n          `temporalPadding expects input padding pattern to be a length-2 ` +\n          `array, but received a length-${padding.length} array.`);\n    }\n\n    const pattern: Array<[number, number]> = [[0, 0], padding, [0, 0]];\n    return tfc.pad(x, pattern);\n  });\n}\n\n/**\n * Pads the 2nd and 3rd dimensions of a 4D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of two `Array`s, each of which is an `Array` of two\n *   integers. The amount of padding at the beginning and end of the 2nd and 3rd\n *   dimensions, respectively.\n * @param dataFormat 'channelsLast' (default) or 'channelsFirst'.\n * @return Padded 4D `tf.Tensor`.\n */\nexport function spatial2dPadding(\n    x: Tensor, padding?: [[number, number], [number, number]],\n    dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 4-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [[1, 1], [1, 1]];\n    }\n    if (padding.length !== 2 || padding[0].length !== 2 ||\n        padding[1].length !== 2) {\n      throw new ValueError(\n          'spatial2dPadding expects `padding` to be an Array of two Arrays, ' +\n          'each of which is an Array of two integers.');\n    }\n\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (dataFormat !== 'channelsLast' && dataFormat !== 'channelsFirst') {\n      throw new ValueError(\n          `Unknown data format: ${dataFormat}. ` +\n          `Supported data formats are 'channelsLast' and 'channelsFirst.`);\n    }\n\n    let pattern: Array<[number, number]>;\n    if (dataFormat === 'channelsFirst') {\n      pattern = [[0, 0], [0, 0], padding[0], padding[1]];\n    } else {\n      pattern = [[0, 0], padding[0], padding[1], [0, 0]];\n    }\n\n    return tfc.pad(x, pattern);\n  });\n}\n\nexport declare interface ZeroPadding2DLayerArgs extends LayerArgs {\n  /**\n   * Integer, or `Array` of 2 integers, or `Array` of 2 `Array`s, each of\n   * which is an `Array` of 2 integers.\n   * - If integer, the same symmetric padding is applied to width and height.\n   * - If Array` of 2 integers, interpreted as two different symmetric values\n   *   for height and width:\n   *   `[symmetricHeightPad, symmetricWidthPad]`.\n   * - If `Array` of 2 `Array`s, interpreted as:\n   *   `[[topPad, bottomPad], [leftPad, rightPad]]`.\n   */\n  padding?: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * One of `'channelsLast'` (default) and `'channelsFirst'`.\n   *\n   * The ordering of the dimensions in the inputs.\n   * `channelsLast` corresponds to inputs with shape\n   * `[batch, height, width, channels]` while `channelsFirst`\n   * corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class ZeroPadding2D extends Layer {\n  /** @nocollapse */\n  static className = 'ZeroPadding2D';\n  readonly dataFormat: DataFormat;\n  readonly padding: [[number, number], [number, number]];\n\n  constructor(args?: ZeroPadding2DLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.dataFormat =\n        args.dataFormat == null ? imageDataFormat() : args.dataFormat;\n    // TODO(cais): Maybe refactor the following logic surrounding `padding`\n    //   into a helper method.\n    if (args.padding == null) {\n      this.padding = [[1, 1], [1, 1]];\n    } else if (typeof args.padding === 'number') {\n      this.padding =\n          [[args.padding, args.padding], [args.padding, args.padding]];\n    } else {\n      args.padding = args.padding;\n      if (args.padding.length !== 2) {\n        throw new ValueError(\n            `ZeroPadding2D expects padding to be a length-2 array, but ` +\n            `received a length-${args.padding.length} array.`);\n      }\n\n      let heightPadding: [number, number];\n      let widthPadding: [number, number];\n      if (typeof args.padding[0] === 'number') {\n        heightPadding = [args.padding[0], args.padding[0]];\n        widthPadding = [args.padding[1] as number, args.padding[1] as number];\n      } else {\n        args.padding = args.padding as [[number, number], [number, number]];\n\n        if (args.padding[0].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects height padding to be a length-2 array, ` +\n              `but received a length-${args.padding[0].length} array.`);\n        }\n        heightPadding = args.padding[0] as [number, number];\n\n        if (args.padding[1].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects width padding to be a length-2 array, ` +\n              `but received a length-${args.padding[1].length} array.`);\n        }\n        widthPadding = args.padding[1] as [number, number];\n      }\n      this.padding = [heightPadding, widthPadding];\n    }\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n\n    let rows: number;\n    let cols: number;\n    if (this.dataFormat === 'channelsFirst') {\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        rows = inputShape[2] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[3] != null && inputShape[3] >= 0) {\n        cols = inputShape[3] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      if (inputShape[1] != null && inputShape[1] >= 0) {\n        rows = inputShape[1] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        cols = inputShape[2] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(\n        () => spatial2dPadding(\n            getExactlyOneTensor(inputs), this.padding, this.dataFormat));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ZeroPadding2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Pooling Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor3D, Tensor4D, Tensor5D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode, checkPoolMode} from '../common';\nimport {InputSpec} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, PaddingMode, PoolMode, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\nimport {preprocessConv2DInput, preprocessConv3DInput} from './convolutional';\n\n/**\n * 2D pooling.\n * @param x\n * @param poolSize\n * @param stridesdes strides. Defaults to [1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 2D pooling.\n */\nexport function pool2d(\n    x: Tensor, poolSize: [number, number], strides?: [number, number],\n    padding?: PaddingMode, dataFormat?: DataFormat,\n    poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // TODO(cais): Remove the preprocessing step once deeplearn.js supports\n    // dataFormat as an input argument.\n    x = preprocessConv2DInput(x, dataFormat);  // x is NHWC after preprocessing.\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      // TODO(cais): Rank check?\n      y = tfc.maxPool(x as Tensor4D, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      // TODO(cais): Check the dtype and rank of x and give clear error message\n      //   if those are incorrect.\n      y = tfc.avgPool(\n          // TODO(cais): Rank check?\n          x as Tensor3D | Tensor4D, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);  // NHWC -> NCHW.\n    }\n    return y;\n  });\n}\n\n/**\n * 3D pooling.\n * @param x\n * @param poolSize. Default to [1, 1, 1].\n * @param strides strides. Defaults to [1, 1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 3D pooling.\n */\nexport function pool3d(\n    x: Tensor5D, poolSize: [number, number, number],\n    strides?: [number, number, number], padding?: PaddingMode,\n    dataFormat?: DataFormat, poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // x is NDHWC after preprocessing.\n    x = preprocessConv3DInput(x as Tensor, dataFormat) as Tensor5D;\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      y = tfc.maxPool3d(x, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      y = tfc.avgPool3d(x, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);  // NDHWC -> NCDHW.\n    }\n    return y;\n  });\n}\n\nexport declare interface Pooling1DLayerArgs extends LayerArgs {\n  /**\n   * Size of the window to pool over, should be an integer.\n   */\n  poolSize?: number|[number];\n  /**\n   * Period at which to sample the pooled values.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number];\n  /** How to fill in data that's not an integer multiple of poolSize. */\n  padding?: PaddingMode;\n}\n\n/**\n * Abstract class for different pooling 1D layers.\n */\nexport abstract class Pooling1D extends Layer {\n  protected readonly poolSize: [number];\n  protected readonly strides: [number];\n  protected readonly padding: PaddingMode;\n\n  /**\n   *\n   * @param args Parameters for the Pooling layer.\n   *\n   * config.poolSize defaults to 2.\n   */\n  constructor(args: Pooling1DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = 2;\n    }\n    super(args);\n    if (typeof args.poolSize === 'number') {\n      this.poolSize = [args.poolSize];\n    } else if (\n        Array.isArray(args.poolSize) &&\n        (args.poolSize as number[]).length === 1 &&\n        typeof (args.poolSize as number[])[0] === 'number') {\n      this.poolSize = args.poolSize;\n    } else {\n      throw new ValueError(\n          `poolSize for 1D convolutional layer must be a number or an ` +\n          `Array of a single number, but received ` +\n          `${JSON.stringify(args.poolSize)}`);\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else {\n      if (typeof args.strides === 'number') {\n        this.strides = [args.strides];\n      } else if (\n          Array.isArray(args.strides) &&\n          (args.strides as number[]).length === 1 &&\n          typeof (args.strides as number[])[0] === 'number') {\n        this.strides = args.strides;\n      } else {\n        throw new ValueError(\n            `strides for 1D convolutional layer must be a number or an ` +\n            `Array of a single number, but received ` +\n            `${JSON.stringify(args.strides)}`);\n      }\n    }\n    assertPositiveInteger(this.strides, 'strides');\n\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const length = convOutputLength(\n        inputShape[1], this.poolSize[0], this.padding, this.strides[0]);\n    return [inputShape[0], length, inputShape[2]];\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Add dummy last dimension.\n      inputs = K.expandDims(getExactlyOneTensor(inputs), 2);\n      const output = this.poolingFunction(\n          getExactlyOneTensor(inputs), [this.poolSize[0], 1],\n          [this.strides[0], 1], this.padding, 'channelsLast');\n      // Remove dummy last dimension.\n      return tfc.squeeze(output, [2]);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'MaxPooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling1D);\n\nexport class AveragePooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'AveragePooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling1D);\n\nexport declare interface Pooling2DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [vertical, horizontal].\n   * Expects an integer or an array of 2 integers.\n   *\n   * For example, `[2, 2]` will halve the input in both spatial dimension.\n   * If only one integer is specified, the same window length\n   * will be used for both dimensions.\n   */\n  poolSize?: number|[number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 2 integers. Integer, tuple of 2 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 2D layers.\n */\nexport abstract class Pooling2D extends Layer {\n  protected readonly poolSize: [number, number];\n  protected readonly strides: [number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling2DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 2) {\n        throw new ValueError(\n            `If the strides property of a 2D pooling layer is an Array, ` +\n            `it is expected to have a length of 2, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    rows =\n        convOutputLength(rows, this.poolSize[0], this.padding, this.strides[0]);\n    cols =\n        convOutputLength(cols, this.poolSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'MaxPooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling2D);\n\nexport class AveragePooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'AveragePooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling2D);\n\nexport declare interface Pooling3DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [depth, height, width].\n   * Expects an integer or an array of 3 integers.\n   *\n   * For example, `[2, 2, 2]` will halve the input in three dimensions.\n   * If only one integer is specified, the same window length\n   * will be used for all dimensions.\n   */\n  poolSize?: number|[number, number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 3 integers. Integer, tuple of 3 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 3D layers.\n */\nexport abstract class Pooling3D extends Layer {\n  protected readonly poolSize: [number, number, number];\n  protected readonly strides: [number, number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling3DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 3) {\n        throw new ValueError(\n            `If the strides property of a 3D pooling layer is an Array, ` +\n            `it is expected to have a length of 3, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let depths =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[4] : inputShape[3];\n    depths = convOutputLength(\n        depths, this.poolSize[0], this.padding, this.strides[0]);\n    rows =\n        convOutputLength(rows, this.poolSize[1], this.padding, this.strides[1]);\n    cols =\n        convOutputLength(cols, this.poolSize[2], this.padding, this.strides[2]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], depths, rows, cols];\n    } else {\n      return [inputShape[0], depths, rows, cols, inputShape[4]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'MaxPooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling3D);\n\nexport class AveragePooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'AveragePooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling3D);\n\n/**\n * Abstract class for different global pooling 1D layers.\n */\nexport abstract class GlobalPooling1D extends Layer {\n  constructor(args: LayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], inputShape[2]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n}\n\nexport class GlobalAveragePooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling1D';\n  constructor(args?: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.mean(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling1D);\n\nexport class GlobalMaxPooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling1D';\n  constructor(args: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.max(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling1D);\n\nexport declare interface GlobalPooling2DLayerArgs extends LayerArgs {\n  /**\n   * One of `CHANNEL_LAST` (default) or `CHANNEL_FIRST`.\n   *\n   * The ordering of the dimensions in the inputs. `CHANNEL_LAST` corresponds\n   * to inputs with shape `[batch, height, width, channels[` while\n   * `CHANNEL_FIRST` corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different global pooling 2D layers.\n */\nexport abstract class GlobalPooling2D extends Layer {\n  protected dataFormat: DataFormat;\n  constructor(args: GlobalPooling2DLayerArgs) {\n    super(args);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape;\n    if (this.dataFormat === 'channelsLast') {\n      return [inputShape[0], inputShape[3]];\n    } else {\n      return [inputShape[0], inputShape[1]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class GlobalAveragePooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.mean(input, [1, 2]);\n      } else {\n        return tfc.mean(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling2D);\n\nexport class GlobalMaxPooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.max(input, [1, 2]);\n      } else {\n        return tfc.max(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Layers that augment the functionality of a base layer.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {InputSpec, Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {BidirectionalMergeMode, Shape, VALID_BIDIRECTIONAL_MERGE_MODES} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {RegularizerFn, RnnStepFunction} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {rnn, RNN, standardizeArgs} from './recurrent';\nimport {deserialize} from './serialization';\n\nexport declare interface WrapperLayerArgs extends LayerArgs {\n  /**\n   * The layer to be wrapped.\n   */\n  layer: Layer;\n}\n\n/**\n * Abstract wrapper base class.\n *\n * Wrappers take another layer and augment it in various ways.\n * Do not use this class as a layer, it is only an abstract base class.\n * Two usable wrappers are the `TimeDistributed` and `Bidirectional` wrappers.\n */\nexport abstract class Wrapper extends Layer {\n  readonly layer: Layer;\n\n  constructor(args: WrapperLayerArgs) {\n    // Porting Note: In PyKeras, `self.layer` is set prior to the calling\n    //   `super()`. But we can't do that here due to TypeScript's restriction.\n    //   See: https://github.com/Microsoft/TypeScript/issues/8277\n    //   As a result, we have to add checks in `get trainable()` and\n    //   `set trainable()` below in order to prevent using `this.layer` when\n    //   its value is `undefined`. The super constructor does use the getter\n    //   and the setter of `this.layer`.\n    super(args);\n    this.layer = args.layer;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    this.built = true;\n  }\n\n  // TODO(cais): Implement activityRegularizer getter.\n\n  get trainable(): boolean {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      return this.layer.trainable;\n    } else {\n      return false;\n    }\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      this.layer.trainable = value;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.layer.trainableWeights;\n  }\n  // TODO(cais): Implement setter for trainableWeights.\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.layer.nonTrainableWeights;\n  }\n  // TODO(cais): Implement setter for nonTrainableWeights.\n\n  get updates(): Tensor[] {\n    // tslint:disable-next-line:no-any\n    return (this.layer as any)._updates;\n  }\n\n  // TODO(cais): Implement getUpdatesFor().\n\n  get losses(): RegularizerFn[] {\n    return this.layer.losses;\n  }\n\n  // TODO(cais): Implement getLossesFor().\n\n  getWeights(): Tensor[] {\n    return this.layer.getWeights();\n  }\n\n  setWeights(weights: Tensor[]): void {\n    this.layer.setWeights(weights);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'layer': {\n        'className': this.layer.getClassName(),\n        'config': this.layer.getConfig(),\n      }\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.layer != null) {\n      this.layer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const layerConfig = config['layer'] as serialization.ConfigDict;\n    const layer = deserialize(layerConfig, customObjects) as Layer;\n    delete config['layer'];\n    const newConfig = {layer};\n    Object.assign(newConfig, config);\n    return new cls(newConfig);\n  }\n}\n\nexport class TimeDistributed extends Wrapper {\n  /** @nocollapse */\n  static className = 'TimeDistributed';\n  constructor(args: WrapperLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 3) {\n      throw new ValueError(\n          `TimeDistributed layer expects an input shape >= 3D, but received ` +\n          `input shape ${JSON.stringify(inputShape)}`);\n    }\n    this.inputSpec = [{shape: inputShape}];\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (!this.layer.built) {\n      this.layer.build(childInputShape);\n      this.layer.built = true;\n    }\n    super.build(inputShape);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    const childOutputShape =\n        this.layer.computeOutputShape(childInputShape) as Shape;\n    const timesteps = inputShape[1];\n    return [childOutputShape[0], timesteps].concat(childOutputShape.slice(1));\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      // TODO(cais): Add 'training' and 'useLearningPhase' to kwargs.\n      inputs = getExactlyOneTensor(inputs);\n      // Porting Note: In tfjs-layers, `inputs` are always concrete tensor\n      // values. Hence the inputs can't have an undetermined first (batch)\n      // dimension, which is why we always use the K.rnn approach here.\n      const step: RnnStepFunction = (inputs: Tensor, states: Tensor[]) => {\n        // TODO(cais): Add useLearningPhase.\n        // NOTE(cais): `layer.call` may return a length-1 array of Tensor in\n        //   some cases (e.g., `layer` is a `Sequential` instance), which is\n        //   why `getExactlyOneTensor` is used below.\n        const output = getExactlyOneTensor(this.layer.call(inputs, kwargs));\n        return [output, []];\n      };\n      const rnnOutputs =\n          rnn(step, inputs, [], false /* goBackwards */, null /* mask */,\n              null /* constants */, false /* unroll */,\n              true /* needPerStepOutputs */);\n      const y = rnnOutputs[1];\n      // TODO(cais): Add activity regularization.\n      // TODO(cais): Add useLearningPhase.\n      return y;\n    });\n  }\n\n  // TODO(cais): Implement detailed computeMask() logic.\n}\nserialization.registerClass(TimeDistributed);\n\nexport function checkBidirectionalMergeMode(value?: string): void {\n  generic_utils.checkStringTypeUnionValue(\n      VALID_BIDIRECTIONAL_MERGE_MODES, 'BidirectionalMergeMode', value);\n}\n\nexport declare interface BidirectionalLayerArgs extends WrapperLayerArgs {\n  /**\n   * The instance of an `RNN` layer to be wrapped.\n   */\n  layer: RNN;\n\n  /**\n   * Mode by which outputs of the forward and backward RNNs are\n   * combined. If `null` or `undefined`, the output will not be\n   * combined, they will be returned as an `Array`.\n   *\n   * If `undefined` (i.e., not provided), defaults to `'concat'`.\n   */\n  mergeMode?: BidirectionalMergeMode;\n}\n\nconst DEFAULT_BIDIRECTIONAL_MERGE_MODE: BidirectionalMergeMode = 'concat';\n\nexport class Bidirectional extends Wrapper {\n  /** @nocollapse */\n  static className = 'Bidirectional';\n  mergeMode: BidirectionalMergeMode;\n  private forwardLayer: RNN;\n  private backwardLayer: RNN;\n  private returnSequences: boolean;\n  private returnState: boolean;\n  private numConstants?: number;\n  private _trainable: boolean;\n\n  constructor(args: BidirectionalLayerArgs) {\n    super(args);\n\n    // Note: When creating `this.forwardLayer`, the original Layer object\n    //   (`config.layer`) ought to be cloned. This is why we call\n    //   `getConfig()` followed by `deserialize()`. Without this cloning,\n    //   the layer names saved during serialization will incorrectly contain\n    //   the 'forward_' prefix. In Python Keras, this is done using\n    //   `copy.copy` (shallow copy), which does not have a simple equivalent\n    //   in JavaScript. JavaScript's `Object.assign()` does not copy\n    //   methods.\n    const layerConfig = args.layer.getConfig();\n    const forwDict: serialization.ConfigDict = {};\n    forwDict['className'] = args.layer.getClassName();\n    forwDict['config'] = layerConfig;\n    this.forwardLayer = deserialize(forwDict) as RNN;\n    layerConfig['goBackwards'] =\n        layerConfig['goBackwards'] === true ? false : true;\n    const backDict: serialization.ConfigDict = {};\n    backDict['className'] = args.layer.getClassName();\n    backDict['config'] = layerConfig;\n    this.backwardLayer = deserialize(backDict) as RNN;\n    this.forwardLayer.name = 'forward_' + this.forwardLayer.name;\n    this.backwardLayer.name = 'backward_' + this.backwardLayer.name;\n\n    this.mergeMode = args.mergeMode === undefined ?\n        DEFAULT_BIDIRECTIONAL_MERGE_MODE :\n        args.mergeMode;\n    checkBidirectionalMergeMode(this.mergeMode);\n    if (args.weights) {\n      throw new NotImplementedError(\n          'weights support is not implemented for Bidirectional layer yet.');\n    }\n    this._stateful = args.layer.stateful;\n    this.returnSequences = args.layer.returnSequences;\n    this.returnState = args.layer.returnState;\n    this.supportsMasking = true;\n    this._trainable = true;\n    this.inputSpec = args.layer.inputSpec;\n    this.numConstants = null;\n  }\n\n  get trainable(): boolean {\n    return this._trainable;\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    this._trainable = value;\n    if (this.forwardLayer != null) {\n      this.forwardLayer.trainable = value;\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.trainable = value;\n    }\n  }\n\n  getWeights(): Tensor[] {\n    return this.forwardLayer.getWeights().concat(\n        this.backwardLayer.getWeights());\n  }\n\n  setWeights(weights: Tensor[]): void {\n    const numWeights = weights.length;\n    const numeightsOver2 = Math.floor(numWeights / 2);\n    this.forwardLayer.setWeights(weights.slice(0, numeightsOver2));\n    this.backwardLayer.setWeights(weights.slice(numeightsOver2));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    let layerShapes: Shape|Shape[] =\n        this.forwardLayer.computeOutputShape(inputShape);\n    if (!(Array.isArray(layerShapes) && Array.isArray(layerShapes[0]))) {\n      layerShapes = [layerShapes as Shape];\n    }\n    layerShapes = layerShapes as Shape[];\n\n    let outputShape: Shape;\n    let outputShapes: Shape[];\n    let stateShape: Shape[];\n    if (this.returnState) {\n      stateShape = layerShapes.slice(1);\n      outputShape = layerShapes[0];\n    } else {\n      outputShape = layerShapes[0];\n    }\n    outputShape = outputShape;\n    if (this.mergeMode === 'concat') {\n      outputShape[outputShape.length - 1] *= 2;\n      outputShapes = [outputShape];\n    } else if (this.mergeMode == null) {\n      outputShapes = [outputShape, outputShape.slice()];\n    } else {\n      outputShapes = [outputShape];\n    }\n\n    if (this.returnState) {\n      if (this.mergeMode == null) {\n        return outputShapes.concat(stateShape).concat(stateShape.slice());\n      }\n      return [outputShape].concat(stateShape).concat(stateShape.slice());\n    }\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    if (Array.isArray(inputs)) {\n      initialState = (inputs as Tensor[] | SymbolicTensor[]).slice(1);\n      inputs = (inputs as Tensor[] | SymbolicTensor[])[0];\n    }\n\n    if ((initialState == null || initialState.length === 0) &&\n        constants == null) {\n      return super.apply(inputs, kwargs);\n    }\n    const additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    const additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      const numStates = initialState.length;\n      if (numStates % 2 > 0) {\n        throw new ValueError(\n            'When passing `initialState` to a Bidrectional RNN, ' +\n            'the state should be an Array containing the states of ' +\n            'the underlying RNNs.');\n      }\n      kwargs['initialState'] = initialState;\n      additionalInputs.push(...initialState);\n      const stateSpecs = (initialState as Array<Tensor|SymbolicTensor>)\n                             .map(state => new InputSpec({shape: state.shape}));\n      this.forwardLayer.stateSpec = stateSpecs.slice(0, numStates / 2);\n      this.backwardLayer.stateSpec = stateSpecs.slice(numStates / 2);\n      additionalSpecs.push(...stateSpecs);\n    }\n    if (constants != null) {\n      throw new NotImplementedError(\n          'Support for constants in Bidirectional layers is not ' +\n          'implemented yet.');\n    }\n\n    const isSymbolicTensor = additionalInputs[0] instanceof SymbolicTensor;\n    for (const tensor of additionalInputs) {\n      if (tensor instanceof SymbolicTensor !== isSymbolicTensor) {\n        throw new ValueError(\n            'The initial state of a Bidirectional layer cannot be ' +\n            'specified as a mix of symbolic and non-symbolic tensors');\n      }\n    }\n\n    if (isSymbolicTensor) {\n      // Compute the full input and specs, including the states.\n      const fullInput = [inputs].concat(additionalInputs);\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call temporarily and replace inputSpec.\n      // Note: with initial states symbolic calls and non-symbolic calls to\n      // this method differ in how the initial states are passed. For\n      // symbolic calls, the initial states are passed in the first arg, as\n      // an Array of SymbolicTensors; for non-symbolic calls, they are\n      // passed in the second arg as a part of the kwargs. Hence the need to\n      // temporarily modify inputSpec here.\n      // TODO(cais): Make refactoring so that this hacky code below is no\n      // longer needed.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output =\n          super.apply(fullInput as Tensor[] | SymbolicTensor[], kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const initialState = kwargs['initialState'];\n\n      let y: Tensor|Tensor[];\n      let yRev: Tensor|Tensor[];\n      if (initialState == null) {\n        y = this.forwardLayer.call(inputs, kwargs);\n        yRev = this.backwardLayer.call(inputs, kwargs);\n      } else {\n        const forwardState = initialState.slice(0, initialState.length / 2);\n        const backwardState = initialState.slice(initialState.length / 2);\n        y = this.forwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: forwardState}));\n        yRev = this.backwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: backwardState}));\n      }\n\n      let states: Tensor[];\n      if (this.returnState) {\n        if (Array.isArray(y)) {\n          states = y.slice(1).concat((yRev as Tensor[]).slice(1));\n        } else {\n        }\n        y = (y as Tensor[])[0];\n        yRev = (yRev as Tensor[])[0];\n      }\n\n      if (this.returnSequences) {\n        yRev = tfc.reverse(yRev as Tensor, 1);\n      }\n\n      let output: Tensor|Tensor[];\n      if (this.mergeMode === 'concat') {\n        output = K.concatenate([y as Tensor, yRev as Tensor]);\n      } else if (this.mergeMode === 'sum') {\n        output = tfc.add(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode === 'ave') {\n        output = tfc.mul(.5, tfc.add(y as Tensor, yRev as Tensor));\n      } else if (this.mergeMode === 'mul') {\n        output = tfc.mul(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode == null) {\n        output = [y as Tensor, yRev as Tensor];\n      }\n\n      // TODO(cais): Properly set learning phase.\n      if (this.returnState) {\n        if (this.mergeMode == null) {\n          return (output as Tensor[]).concat(states);\n        }\n        return [output as Tensor].concat(states);\n      }\n      return output;\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[]): void {\n    this.forwardLayer.resetStates();\n    this.backwardLayer.resetStates();\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    nameScope(this.forwardLayer.name, () => {\n      this.forwardLayer.build(inputShape);\n    });\n    nameScope(this.backwardLayer.name, () => {\n      this.backwardLayer.build(inputShape);\n    });\n    this.built = true;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (Array.isArray(mask)) {\n      mask = mask[0];\n    }\n    let outputMask: Tensor|Tensor[];\n    if (this.returnSequences) {\n      if (this.mergeMode == null) {\n        outputMask = [mask, mask];\n      } else {\n        outputMask = mask;\n      }\n    } else {\n      if (this.mergeMode == null) {\n        outputMask = [null, null];\n      } else {\n        outputMask = null;\n      }\n    }\n    if (this.returnState) {\n      const states = this.forwardLayer.states;\n      const stateMask: Tensor[] = states.map(state => null);\n      if (Array.isArray(outputMask)) {\n        return outputMask.concat(stateMask).concat(stateMask);\n      } else {\n        return [outputMask].concat(stateMask).concat(stateMask);\n      }\n    } else {\n      return outputMask;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.forwardLayer.trainableWeights.concat(\n        this.backwardLayer.trainableWeights);\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.forwardLayer.nonTrainableWeights.concat(\n        this.backwardLayer.nonTrainableWeights);\n  }\n\n  // TODO(cais): Implement constraints().\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.forwardLayer != null) {\n      this.forwardLayer.setFastWeightInitDuringBuild(value);\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'mergeMode': this.mergeMode,\n    };\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    const rnnLayer =\n        deserialize(config['layer'] as serialization.ConfigDict) as RNN;\n    delete config['layer'];\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    if (config['numConstants'] != null) {\n      throw new NotImplementedError(\n          `Deserialization of a Bidirectional layer with numConstants ` +\n          `present is not supported yet.`);\n    }\n    // tslint:disable-next-line:no-any\n    const newConfig: {[key: string]: any} = config;\n    newConfig['layer'] = rnnLayer;\n    return new cls(newConfig);\n  }\n}\nserialization.registerClass(Bidirectional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {InputLayer, InputLayerArgs} from './engine/input_layer';\nimport {Layer, LayerArgs} from './engine/topology';\nimport {input} from './exports';\nimport {ELU, ELULayerArgs, LeakyReLU, LeakyReLULayerArgs, PReLU, PReLULayerArgs, ReLU, ReLULayerArgs, Softmax, SoftmaxLayerArgs, ThresholdedReLU, ThresholdedReLULayerArgs} from './layers/advanced_activations';\nimport {Conv1D, Conv2D, Conv2DTranspose, Conv3D, ConvLayerArgs, Cropping2D, Cropping2DLayerArgs, SeparableConv2D, SeparableConvLayerArgs, UpSampling2D, UpSampling2DLayerArgs} from './layers/convolutional';\nimport {DepthwiseConv2D, DepthwiseConv2DLayerArgs} from './layers/convolutional_depthwise';\nimport {ConvLSTM2D, ConvLSTM2DArgs, ConvLSTM2DCell, ConvLSTM2DCellArgs} from './layers/convolutional_recurrent';\nimport {Activation, ActivationLayerArgs, Dense, DenseLayerArgs, Dropout, DropoutLayerArgs, Flatten, FlattenLayerArgs, Masking, MaskingArgs, Permute, PermuteLayerArgs, RepeatVector, RepeatVectorLayerArgs, Reshape, ReshapeLayerArgs, SpatialDropout1D, SpatialDropout1DLayerConfig} from './layers/core';\nimport {Embedding, EmbeddingLayerArgs} from './layers/embeddings';\nimport {Add, Average, Concatenate, ConcatenateLayerArgs, Dot, DotLayerArgs, Maximum, Minimum, Multiply} from './layers/merge';\nimport {AlphaDropout, AlphaDropoutArgs, GaussianDropout, GaussianDropoutArgs, GaussianNoise, GaussianNoiseArgs} from './layers/noise';\nimport {BatchNormalization, BatchNormalizationLayerArgs, LayerNormalization, LayerNormalizationLayerArgs} from './layers/normalization';\nimport {ZeroPadding2D, ZeroPadding2DLayerArgs} from './layers/padding';\nimport {AveragePooling1D, AveragePooling2D, AveragePooling3D, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalPooling2DLayerArgs, MaxPooling1D, MaxPooling2D, MaxPooling3D, Pooling1DLayerArgs, Pooling2DLayerArgs, Pooling3DLayerArgs} from './layers/pooling';\nimport {GRU, GRUCell, GRUCellLayerArgs, GRULayerArgs, LSTM, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNN, SimpleRNNCell, SimpleRNNCellLayerArgs, SimpleRNNLayerArgs, StackedRNNCells, StackedRNNCellsArgs} from './layers/recurrent';\nimport {Bidirectional, BidirectionalLayerArgs, TimeDistributed, WrapperLayerArgs} from './layers/wrappers';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// Input Layer.\n/**\n * An input layer is an entry point into a `tf.LayersModel`.\n *\n * `InputLayer` is generated automatically for `tf.Sequential`` models by\n * specifying the `inputshape` or `batchInputShape` for the first layer.  It\n * should not be specified explicitly. However, it can be useful sometimes,\n * e.g., when constructing a sequential model from a subset of another\n * sequential model's layers. Like the code snippet below shows.\n *\n * ```js\n * // Define a model which simply adds two inputs.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));\n * model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n * model1.summary();\n * model1.predict(tf.zeros([1, 4])).print();\n *\n * // Construct another model, reusing the second layer of `model1` while\n * // not using the first layer of `model1`. Note that you cannot add the second\n * // layer of `model` directly as the first layer of the new sequential model,\n * // because doing so will lead to an error related to the fact that the layer\n * // is not an input layer. Instead, you need to create an `inputLayer` and add\n * // it to the new sequential model before adding the reused layer.\n * const model2 = tf.sequential();\n * // Use an inputShape that matches the input shape of `model1`'s second\n * // layer.\n * model2.add(tf.layers.inputLayer({inputShape: [3]}));\n * model2.add(model1.layers[1]);\n * model2.summary();\n * model2.predict(tf.zeros([1, 3])).print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}\n */\nexport function inputLayer(args: InputLayerArgs): Layer {\n  return new InputLayer(args);\n}\n\n// Advanced Activation Layers.\n\n/**\n * Exponetial Linear Unit (ELU).\n *\n * It follows:\n * `f(x) =  alpha * (exp(x) - 1.) for x < 0`,\n * `f(x) = x for x >= 0`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Fast and Accurate Deep Network Learning by Exponential Linear Units\n * (ELUs)](https://arxiv.org/abs/1511.07289v1)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function elu(args?: ELULayerArgs): Layer {\n  return new ELU(args);\n}\n\n/**\n * Rectified Linear Unit activation function.\n *\n * Input shape:\n *   Arbitrary. Use the config field `inputShape` (Array of integers, does\n *   not include the sample axis) when using this layer as the first layer\n *   in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function reLU(args?: ReLULayerArgs): Layer {\n  return new ReLU(args);\n}\n\n/**\n * Leaky version of a rectified linear unit.\n *\n * It allows a small gradient when the unit is not active:\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function leakyReLU(args?: LeakyReLULayerArgs): Layer {\n  return new LeakyReLU(args);\n}\n\n/**\n * Parameterized version of a leaky rectified linear unit.\n *\n * It follows\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n * wherein `alpha` is a trainable weight.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function prelu(args?: PReLULayerArgs): Layer {\n  return new PReLU(args);\n}\n\n/**\n * Softmax activation layer.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function softmax(args?: SoftmaxLayerArgs): Layer {\n  return new Softmax(args);\n}\n\n/**\n * Thresholded Rectified Linear Unit.\n *\n * It follows:\n * `f(x) = x for x > theta`,\n * `f(x) = 0 otherwise`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Zero-Bias Autoencoders and the Benefits of Co-Adapting\n * Features](http://arxiv.org/abs/1402.3337)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function thresholdedReLU(args?: ThresholdedReLULayerArgs): Layer {\n  return new ThresholdedReLU(args);\n}\n\n// Convolutional Layers.\n\n/**\n * 1D convolution layer (e.g., temporal convolution).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input over a single spatial (or temporal) dimension\n * to produce a tensor of outputs.\n *\n * If `use_bias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model, provide an\n * `inputShape` argument `Array` or `null`.\n *\n * For example, `inputShape` would be:\n * - `[10, 128]` for sequences of 10 vectors of 128-dimensional vectors\n * - `[null, 128]` for variable-length sequences of 128-dimensional vectors.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional',  namespace: 'layers'}\n */\nexport function conv1d(args: ConvLayerArgs): Layer {\n  return new Conv1D(args);\n}\n\n/**\n * 2D convolution layer (e.g. spatial convolution over images).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 3]` for 128x128 RGB pictures\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2d(args: ConvLayerArgs): Layer {\n  return new Conv2D(args);\n}\n\n/**\n * Transposed convolutional layer (sometimes called Deconvolution).\n *\n * The need for transposed convolutions generally arises\n * from the desire to use a transformation going in the opposite direction of\n * a normal convolution, i.e., from something that has the shape of the output\n * of some convolution to something that has the shape of its input while\n * maintaining a connectivity pattern that is compatible with said\n * convolution.\n *\n * When using this layer as the first layer in a model, provide the\n * configuration `inputShape` (`Array` of integers, does not include the\n * sample axis), e.g., `inputShape: [128, 128, 3]` for 128x128 RGB pictures in\n * `dataFormat: 'channelsLast'`.\n *\n * Input shape:\n *   4D tensor with shape:\n *   `[batch, channels, rows, cols]` if `dataFormat` is `'channelsFirst'`.\n *   or 4D tensor with shape\n *   `[batch, rows, cols, channels]` if `dataFormat` is `'channelsLast`.\n *\n * Output shape:\n *   4D tensor with shape:\n *   `[batch, filters, newRows, newCols]` if `dataFormat` is\n * `'channelsFirst'`. or 4D tensor with shape:\n *   `[batch, newRows, newCols, filters]` if `dataFormat` is `'channelsLast'`.\n *\n * References:\n *   - [A guide to convolution arithmetic for deep\n * learning](https://arxiv.org/abs/1603.07285v1)\n *   - [Deconvolutional\n * Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2dTranspose(args: ConvLayerArgs): Layer {\n  return new Conv2DTranspose(args);\n}\n\n/**\n * 3D convolution layer (e.g. spatial convolution over volumes).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 128, 1]` for 128x128x128 grayscale volumes\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv3d(args: ConvLayerArgs): Layer {\n  return new Conv3D(args);\n}\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Separable convolution consists of first performing\n * a depthwise spatial convolution\n * (which acts on each input channel separately)\n * followed by a pointwise convolution which mixes together the resulting\n * output channels. The `depthMultiplier` argument controls how many\n * output channels are generated per input channel in the depthwise step.\n *\n * Intuitively, separable convolutions can be understood as\n * a way to factorize a convolution kernel into two smaller kernels,\n * or as an extreme version of an Inception block.\n *\n * Input shape:\n *   4D tensor with shape:\n *     `[batch, channels, rows, cols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, rows, cols, channels]` if data_format='channelsLast'.\n *\n * Output shape:\n *   4D tensor with shape:\n *     `[batch, filters, newRows, newCols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, newRows, newCols, filters]` if data_format='channelsLast'.\n *     `rows` and `cols` values might have changed due to padding.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function separableConv2d(args: SeparableConvLayerArgs): Layer {\n  return new SeparableConv2D(args);\n}\n\n/**\n * Cropping layer for 2D input (e.g., image).\n *\n * This layer can crop an input\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, croppedRows, croppedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, croppedRows, croppedCols]`.\n *\n * Examples\n * ```js\n *\n * const model = tf.sequential();\n * model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]],\n *                                inputShape: [128, 128, 3]}));\n * //now output shape is [batch, 124, 124, 3]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function cropping2D(args: Cropping2DLayerArgs): Layer {\n  return new Cropping2D(args);\n}\n\n/**\n * Upsampling layer for 2D inputs.\n *\n * Repeats the rows and columns of the data\n * by size[0] and size[1] respectively.\n *\n *\n * Input shape:\n *    4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *         `[batch, rows, cols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *        `[batch, channels, rows, cols]`\n *\n * Output shape:\n *     4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *        `[batch, upsampledRows, upsampledCols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *         `[batch, channels, upsampledRows, upsampledCols]`\n *\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function upSampling2d(args: UpSampling2DLayerArgs): Layer {\n  return new UpSampling2D(args);\n}\n\n// Convolutional(depthwise) Layers.\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Depthwise Separable convolutions consists in performing just the first step\n * in a depthwise spatial convolution (which acts on each input channel\n * separately). The `depthMultplier` argument controls how many output channels\n * are generated per input channel in the depthwise step.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function depthwiseConv2d(args: DepthwiseConv2DLayerArgs): Layer {\n  return new DepthwiseConv2D(args);\n}\n\n// Basic Layers.\n\n/**\n * Applies an activation function to an output.\n *\n * This layer applies element-wise activation function.  Other layers, notably\n * `dense` can also apply activation functions.  Use this isolated activation\n * function to extract the values before and after the\n * activation. For instance:\n *\n * ```js\n * const input = tf.input({shape: [5]});\n * const denseLayer = tf.layers.dense({units: 1});\n * const activationLayer = tf.layers.activation({activation: 'relu6'});\n *\n * // Obtain the output symbolic tensors by applying the layers in order.\n * const denseOutput = denseLayer.apply(input);\n * const activationOutput = activationLayer.apply(denseOutput);\n *\n * // Create the model based on the inputs.\n * const model = tf.model({\n *     inputs: input,\n *     outputs: [denseOutput, activationOutput]\n * });\n *\n * // Collect both outputs and print separately.\n * const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));\n * denseOut.print();\n * activationOut.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function activation(args: ActivationLayerArgs): Layer {\n  return new Activation(args);\n}\n\n/**\n * Creates a dense (fully connected) layer.\n *\n * This layer implements the operation:\n *   `output = activation(dot(input, kernel) + bias)`\n *\n * `activation` is the element-wise activation function\n *   passed as the `activation` argument.\n *\n * `kernel` is a weights matrix created by the layer.\n *\n * `bias` is a bias vector created by the layer (only applicable if `useBias`\n * is `true`).\n *\n * **Input shape:**\n *\n *   nD `tf.Tensor` with shape: `(batchSize, ..., inputDim)`.\n *\n *   The most common situation would be\n *   a 2D input with shape `(batchSize, inputDim)`.\n *\n * **Output shape:**\n *\n *   nD tensor with shape: `(batchSize, ..., units)`.\n *\n *   For instance, for a 2D input with shape `(batchSize, inputDim)`,\n *   the output would have shape `(batchSize, units)`.\n *\n * Note: if the input to the layer has a rank greater than 2, then it is\n * flattened prior to the initial dot product with the kernel.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dense(args: DenseLayerArgs): Layer {\n  return new Dense(args);\n}\n\n/**\n * Applies\n * [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to\n * the input.\n *\n * Dropout consists in randomly setting a fraction `rate` of input units to 0 at\n * each update during training time, which helps prevent overfitting.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dropout(args: DropoutLayerArgs): Layer {\n  return new Dropout(args);\n}\n\n/**\n * Spatial 1D version of Dropout.\n *\n * This Layer type performs the same function as the Dropout layer, but it drops\n * entire 1D feature maps instead of individual elements. For example, if an\n * input example consists of 3 timesteps and the feature map for each timestep\n * has a size of 4, a `spatialDropout1d` layer may zero out the feature maps\n * of the 1st timesteps and 2nd timesteps completely while sparing all feature\n * elements of the 3rd timestep.\n *\n * If adjacent frames (timesteps) are strongly correlated (as is normally the\n * case in early convolution layers), regular dropout will not regularize the\n * activation and will otherwise just result in merely an effective learning\n * rate decrease. In this case, `spatialDropout1d` will help promote\n * independence among feature maps and should be used instead.\n *\n * **Arguments:**\n *   rate: A floating-point number >=0 and <=1. Fraction of the input elements\n *     to drop.\n *\n * **Input shape:**\n *   3D tensor with shape `(samples, timesteps, channels)`.\n *\n * **Output shape:**\n *   Same as the input shape.\n *\n * References:\n *   - [Efficient Object Localization Using Convolutional\n *      Networks](https://arxiv.org/abs/1411.4280)\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function spatialDropout1d(args: SpatialDropout1DLayerConfig): Layer {\n  return new SpatialDropout1D(args);\n}\n\n/**\n * Flattens the input. Does not affect the batch size.\n *\n * A `Flatten` layer flattens each batch in its inputs to 1D (making the output\n * 2D).\n *\n * For example:\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const flattenLayer = tf.layers.flatten();\n * // Inspect the inferred output shape of the flatten layer, which\n * // equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the\n * // flattening. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(flattenLayer.apply(input).shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function flatten(args?: FlattenLayerArgs): Layer {\n  return new Flatten(args);\n}\n\n/**\n * Repeats the input n times in a new dimension.\n *\n * ```js\n *  const model = tf.sequential();\n *  model.add(tf.layers.repeatVector({n: 4, inputShape: [2]}));\n *  const x = tf.tensor2d([[10, 20]]);\n *  // Use the model to do inference on a data point the model hasn't see\n *  model.predict(x).print();\n *  // output shape is now [batch, 2, 4]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function repeatVector(args: RepeatVectorLayerArgs): Layer {\n  return new RepeatVector(args);\n}\n\n/**\n * Reshapes an input to a certain shape.\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const reshapeLayer = tf.layers.reshape({targetShape: [2, 6]});\n * // Inspect the inferred output shape of the Reshape layer, which\n * // equals `[null, 2, 6]`. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(reshapeLayer.apply(input).shape));\n * ```\n *\n * Input shape:\n *   Arbitrary, although all dimensions in the input shape must be fixed.\n *   Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n *\n * Output shape:\n *   [batchSize, targetShape[0], targetShape[1], ...,\n *    targetShape[targetShape.length - 1]].\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function reshape(args: ReshapeLayerArgs): Layer {\n  return new Reshape(args);\n}\n\n/**\n * Permutes the dimensions of the input according to a given pattern.\n *\n * Useful for, e.g., connecting RNNs and convnets together.\n *\n * Example:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.permute({\n *   dims: [2, 1],\n *   inputShape: [10, 64]\n * }));\n * console.log(model.outputShape);\n * // Now model's output shape is [null, 64, 10], where null is the\n * // unpermuted sample (batch) dimension.\n * ```\n *\n * Input shape:\n *   Arbitrary. Use the configuration field `inputShape` when using this\n *   layer as the first layer in a model.\n *\n * Output shape:\n *   Same rank as the input shape, but with the dimensions re-ordered (i.e.,\n *   permuted) according to the `dims` configuration of this layer.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function permute(args: PermuteLayerArgs): Layer {\n  return new Permute(args);\n}\n\n/**\n * Maps positive integers (indices) into dense vectors of fixed size.\n * eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]\n *\n * **Input shape:** 2D tensor with shape: `[batchSize, sequenceLength]`.\n *\n * **Output shape:** 3D tensor with shape: `[batchSize, sequenceLength,\n * outputDim]`.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function embedding(args: EmbeddingLayerArgs): Layer {\n  return new Embedding(args);\n}\n\n// Merge Layers.\n\n/**\n * Layer that performs element-wise addition on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). The inputs are specified as an\n * `Array` when the `apply` method of the `Add` layer instance is called. For\n * example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const addLayer = tf.layers.add();\n * const sum = addLayer.apply([input1, input2]);\n * console.log(JSON.stringify(sum.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function add(args?: LayerArgs): Layer {\n  return new Add(args);\n}\n\n/**\n * Layer that performs element-wise averaging on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const averageLayer = tf.layers.average();\n * const average = averageLayer.apply([input1, input2]);\n * console.log(JSON.stringify(average.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function average(args?: LayerArgs): Layer {\n  return new Average(args);\n}\n\n/**\n * Layer that concatenates an `Array` of inputs.\n *\n * It takes a list of tensors, all of the same shape except for the\n * concatenation axis, and returns a single tensor, the concatenation\n * of all inputs. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 3]});\n * const concatLayer = tf.layers.concatenate();\n * const output = concatLayer.apply([input1, input2]);\n * console.log(JSON.stringify(output.shape));\n * // You get [null, 2, 5], with the first dimension as the undetermined batch\n * // dimension. The last dimension (5) is the result of concatenating the\n * // last dimensions of the inputs (2 and 3).\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function concatenate(args?: ConcatenateLayerArgs): Layer {\n  return new Concatenate(args);\n}\n\n/**\n * Layer that computes the element-wise maximum an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const maxLayer = tf.layers.maximum();\n * const max = maxLayer.apply([input1, input2]);\n * console.log(JSON.stringify(max.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function maximum(args?: LayerArgs): Layer {\n  return new Maximum(args);\n}\n\n/**\n * Layer that computes the element-wise minimum of an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const minLayer = tf.layers.minimum();\n * const min = minLayer.apply([input1, input2]);\n * console.log(JSON.stringify(min.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function minimum(args?: LayerArgs): Layer {\n  return new Minimum(args);\n}\n\n/**\n * Layer that multiplies (element-wise) an `Array` of inputs.\n *\n * It takes as input an Array of tensors, all of the same\n * shape, and returns a single tensor (also of the same shape).\n * For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const input3 = tf.input({shape: [2, 2]});\n * const multiplyLayer = tf.layers.multiply();\n * const product = multiplyLayer.apply([input1, input2, input3]);\n * console.log(product.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function multiply(args?: LayerArgs): Layer {\n  return new Multiply(args);\n}\n\n/**\n * Layer that computes a dot product between samples in two tensors.\n *\n * E.g., if applied to a list of two tensors `a` and `b` both of shape\n * `[batchSize, n]`, the output will be a tensor of shape `[batchSize, 1]`,\n * where each entry at index `[i, 0]` will be the dot product between\n * `a[i, :]` and `b[i, :]`.\n *\n * Example:\n *\n * ```js\n * const dotLayer = tf.layers.dot({axes: -1});\n * const x1 = tf.tensor2d([[10, 20], [30, 40]]);\n * const x2 = tf.tensor2d([[-1, -2], [-3, -4]]);\n *\n * // Invoke the layer's apply() method in eager (imperative) mode.\n * const y = dotLayer.apply([x1, x2]);\n * y.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function dot(args: DotLayerArgs): Layer {\n  return new Dot(args);\n}\n\n// Normalization Layers.\n\n/**\n * Batch normalization layer (Ioffe and Szegedy, 2014).\n *\n * Normalize the activations of the previous layer at each batch,\n * i.e. applies a transformation that maintains the mean activation\n * close to 0 and the activation standard deviation close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape` (Array of integers, does\n *   not include the sample axis) when calling the constructor of this class,\n *   if this layer is used as a first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Batch Normalization: Accelerating Deep Network Training by Reducing\n * Internal Covariate Shift](https://arxiv.org/abs/1502.03167)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function batchNormalization(args?: BatchNormalizationLayerArgs): Layer {\n  return new BatchNormalization(args);\n}\n\n/**\n * Layer-normalization layer (Ba et al., 2016).\n *\n * Normalizes the activations of the previous layer for each given example in a\n * batch independently, instead of across a batch like in `batchNormalization`.\n * In other words, this layer applies a transformation that maintanis the mean\n * activation within each example close to0 and activation variance close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the argument `inputShape` when using this layer as the first\n *   layer in a model.\n *\n * Output shape:\n *   Same as input.\n *\n * References:\n *   - [Layer Normalization](https://arxiv.org/abs/1607.06450)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function layerNormalization(args?: LayerNormalizationLayerArgs): Layer {\n  return new LayerNormalization(args);\n}\n\n// Padding Layers.\n\n/**\n * Zero-padding layer for 2D input (e.g., image).\n *\n * This layer can add rows and columns of zeros\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, paddedRows, paddedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, paddedRows, paddedCols]`.\n *\n * @doc {heading: 'Layers', subheading: 'Padding', namespace: 'layers'}\n */\nexport function zeroPadding2d(args?: ZeroPadding2DLayerArgs): Layer {\n  return new ZeroPadding2D(args);\n}\n\n// Pooling Layers.\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape: `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * `tf.avgPool1d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling1d(args: Pooling1DLayerArgs): Layer {\n  return new AveragePooling1D(args);\n}\nexport function avgPool1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape:\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, rows, cols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, pooleRows, pooledCols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, pooleRows, pooledCols]`\n *\n * `tf.avgPool2d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling2d(args: Pooling2DLayerArgs): Layer {\n  return new AveragePooling2D(args);\n}\nexport function avgPool2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n\n/**\n * Average pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      4D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling3d(args: Pooling3DLayerArgs): Layer {\n  return new AveragePooling3D(args);\n}\nexport function avgPool3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n\n/**\n * Global average pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling1d(args?: LayerArgs): Layer {\n  return new GlobalAveragePooling1D(args);\n}\n\n/**\n * Global average pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalAveragePooling2D(args);\n}\n\n/**\n * Global max pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling1d(args?: LayerArgs): Layer {\n  return new GlobalMaxPooling1D(args);\n}\n\n/**\n * Global max pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalMaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for temporal data.\n *\n * Input shape:  `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling1d(args: Pooling1DLayerArgs): Layer {\n  return new MaxPooling1D(args);\n}\n\n/**\n * Max pooling operation for spatial data.\n *\n * Input shape\n *   - If `dataFormat === CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, rows, cols, channels]`\n *   - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *       `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, pooleRows, pooledCols, channels]`\n *   - If `dataFormat=CHANNEL_FIRST`:\n *       4D tensor with shape:\n *       `[batchSize, channels, pooleRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling2d(args: Pooling2DLayerArgs): Layer {\n  return new MaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      5D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling3d(args: Pooling3DLayerArgs): Layer {\n  return new MaxPooling3D(args);\n}\n\n// Recurrent Layers.\n\n/**\n * Gated Recurrent Unit - Cho et al. 2014.\n *\n * This is an `RNN` layer consisting of one `GRUCell`. However, unlike\n * the underlying `GRUCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.gru({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `GRUCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gru(args: GRULayerArgs): Layer {\n  return new GRU(args);\n}\n\n/**\n * Cell class for `GRU`.\n *\n * `GRUCell` is distinct from the `RNN` subclass `GRU` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `GRU` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.gruCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `GRUCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.gruCell({units: 4}),\n *   tf.layers.gruCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `gruCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `GRUCell`, use the\n * `tf.layers.gru`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gruCell(args: GRUCellLayerArgs): RNNCell {\n  return new GRUCell(args);\n}\n\n/**\n * Long-Short Term Memory layer - Hochreiter 1997.\n *\n * This is an `RNN` layer consisting of one `LSTMCell`. However, unlike\n * the underlying `LSTMCell`, the `apply` method of `LSTM` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const lstm = tf.layers.lstm({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = lstm.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `LSTMCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstm(args: LSTMLayerArgs): Layer {\n  return new LSTM(args);\n}\n\n/**\n * Cell class for `LSTM`.\n *\n * `LSTMCell` is distinct from the `RNN` subclass `LSTM` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `LSTM` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.lstmCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `LSTMCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.lstmCell({units: 4}),\n *   tf.layers.lstmCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `lstmCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `LSTMCell`, use the\n * `tf.layers.lstm`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstmCell(args: LSTMCellLayerArgs): RNNCell {\n  return new LSTMCell(args);\n}\n\n/**\n * Fully-connected RNN where the output is to be fed back to input.\n *\n * This is an `RNN` layer consisting of one `SimpleRNNCell`. However, unlike\n * the underlying `SimpleRNNCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.simpleRNN({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `SimpleRNNCell`'s number of units.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNN(args: SimpleRNNLayerArgs): Layer {\n  return new SimpleRNN(args);\n}\n\n/**\n * Cell class for `SimpleRNN`.\n *\n * `SimpleRNNCell` is distinct from the `RNN` subclass `SimpleRNN` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `SimpleRNN` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.simpleRNNCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `SimpleRNNCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.simpleRNNCell({units: 4}),\n *   tf.layers.simpleRNNCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `SimpleRNNCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `SimpleRNNCell`, use the\n * `tf.layers.simpleRNN`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNNCell(args: SimpleRNNCellLayerArgs): RNNCell {\n  return new SimpleRNNCell(args);\n}\n\n/**\n * Convolutional LSTM layer - Xingjian Shi 2015.\n *\n * This is an `ConvRNN2D` layer consisting of one `ConvLSTM2DCell`. However,\n * unlike the underlying `ConvLSTM2DCell`, the `apply` method of `ConvLSTM2D`\n * operates on a sequence of inputs. The shape of the input (not including the\n * first, batch dimension) needs to be 4-D, with the first dimension being time\n * steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const batchSize = 4;\n * const sequenceLength = 2;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [batchSize, sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const layer = tf.layers.convLstm2d({filters, kernelSize});\n *\n * const output = layer.apply(input);\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2d(args: ConvLSTM2DArgs): ConvLSTM2D {\n  return new ConvLSTM2D(args);\n}\n\n/**\n * Cell class for `ConvLSTM2D`.\n *\n * `ConvLSTM2DCell` is distinct from the `ConvRNN2D` subclass `ConvLSTM2D` in\n * that its `call` method takes the input data of only a single time step and\n * returns the cell's output at the time step, while `ConvLSTM2D` takes the\n * input data over a number of time steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const sequenceLength = 1;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const cell = tf.layers.convLstm2dCell({filters, kernelSize});\n *\n * cell.build(input.shape);\n *\n * const outputSize = size - kernelSize + 1;\n * const outShape = [sequenceLength, outputSize, outputSize, filters];\n *\n * const initialH = tf.zeros(outShape);\n * const initialC = tf.zeros(outShape);\n *\n * const [o, h, c] = cell.call([input, initialH, initialC], {});\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2dCell(args: ConvLSTM2DCellArgs): ConvLSTM2DCell {\n  return new ConvLSTM2DCell(args);\n}\n\n/**\n * Base class for recurrent layers.\n *\n * Input shape:\n *   3D tensor with shape `[batchSize, timeSteps, inputDim]`.\n *\n * Output shape:\n *   - if `returnState`, an Array of tensors (i.e., `tf.Tensor`s). The first\n *     tensor is the output. The remaining tensors are the states at the\n *     last time step, each with shape `[batchSize, units]`.\n *   - if `returnSequences`, the output will have shape\n *     `[batchSize, timeSteps, units]`.\n *   - else, the output will have shape `[batchSize, units]`.\n *\n * Masking:\n *   This layer supports masking for input data with a variable number\n *   of timesteps. To introduce masks to your data,\n *   use an embedding layer with the `mask_zero` parameter\n *   set to `True`.\n *\n * Notes on using statefulness in RNNs:\n *   You can set RNN layers to be 'stateful', which means that the states\n *   computed for the samples in one batch will be reused as initial states\n *   for the samples in the next batch. This assumes a one-to-one mapping\n *   between samples in different successive batches.\n *\n *   To enable statefulness:\n *     - specify `stateful: true` in the layer constructor.\n *     - specify a fixed batch size for your model, by passing\n *       if sequential model:\n *         `batchInputShape=[...]` to the first layer in your model.\n *       else for functional model with 1 or more Input layers:\n *         `batchShape=[...]` to all the first layers in your model.\n *       This is the expected shape of your inputs *including the batch size*.\n *       It should be a tuple of integers, e.g. `(32, 10, 100)`.\n *     - specify `shuffle=False` when calling fit().\n *\n *   To reset the states of your model, call `.resetStates()` on either\n *   a specific layer, or on your entire model.\n *\n * Note on specifying the initial state of RNNs\n *   You can specify the initial state of RNN layers symbolically by\n *   calling them with the option `initialState`. The value of\n *   `initialState` should be a tensor or list of tensors representing\n *   the initial state of the RNN layer.\n *\n *   You can specify the initial state of RNN layers numerically by\n *   calling `resetStates` with the keyword argument `states`. The value of\n *   `states` should be a numpy array or list of numpy arrays representing\n *   the initial state of the RNN layer.\n *\n * Note on passing external constants to RNNs\n *   You can pass \"external\" constants to the cell using the `constants`\n *   keyword argument of `RNN.call` method. This requires that the `cell.call`\n *   method accepts the same keyword argument `constants`. Such constants\n *   can be used to conditon the cell transformation on additional static inputs\n *   (not changing over time), a.k.a an attention mechanism.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function rnn(args: RNNLayerArgs): Layer {\n  return new RNN(args);\n}\n\n/**\n * Wrapper allowing a stack of RNN cells to behave as a single cell.\n *\n * Used to implement efficient stacked RNNs.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function stackedRNNCells(args: StackedRNNCellsArgs): RNNCell {\n  return new StackedRNNCells(args);\n}\n\n// Wrapper Layers.\n\n/** @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'} */\nexport function bidirectional(args: BidirectionalLayerArgs): Bidirectional {\n  return new Bidirectional(args);\n}\n\n/**\n * This wrapper applies a layer to every temporal slice of an input.\n *\n * The input should be at least 3D,  and the dimension of the index `1` will be\n * considered to be the temporal dimension.\n *\n * Consider a batch of 32 samples, where each sample is a sequence of 10 vectors\n * of 16 dimensions. The batch input shape of the layer is then `[32,  10,\n * 16]`, and the `inputShape`, not including the sample dimension, is\n * `[10, 16]`.\n *\n * You can then use `TimeDistributed` to apply a `Dense` layer to each of the 10\n * timesteps, independently:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.dense({units: 8}),\n *   inputShape: [10, 16],\n * }));\n *\n * // Now model.outputShape = [null, 10, 8].\n * // The output will then have shape `[32, 10, 8]`.\n *\n * // In subsequent layers, there is no need for `inputShape`:\n * model.add(tf.layers.timeDistributed({layer: tf.layers.dense({units: 32})}));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * // Now model.outputShape = [null, 10, 32].\n * ```\n *\n * The output will then have shape `[32, 10, 32]`.\n *\n * `TimeDistributed` can be used with arbitrary layers, not just `Dense`, for\n * instance a `Conv2D` layer.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.conv2d({filters: 64, kernelSize: [3, 3]}),\n *   inputShape: [10, 299, 299, 3],\n * }));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}\n */\nexport function timeDistributed(args: WrapperLayerArgs): Layer {\n  return new TimeDistributed(args);\n}\n\n// Aliases for pooling.\nexport const globalMaxPool1d = globalMaxPooling1d;\nexport const globalMaxPool2d = globalMaxPooling2d;\nexport const maxPool1d = maxPooling1d;\nexport const maxPool2d = maxPooling2d;\n\nexport {Layer, RNN, RNNCell, input /* alias for tf.input */};\n\n/**\n * Apply additive zero-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * This is useful to mitigate overfitting\n * (you could see it as a form of random data augmentation).\n * Gaussian Noise (GS) is a natural choice as corruption process\n * for real valued inputs.\n *\n * # Arguments\n *     stddev: float, standard deviation of the noise distribution.\n *\n * # Input shape\n *         Arbitrary. Use the keyword argument `input_shape`\n *         (tuple of integers, does not include the samples axis)\n *         when using this layer as the first layer in a model.\n *\n * # Output shape\n *         Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianNoise(args: GaussianNoiseArgs): GaussianNoise {\n  return new GaussianNoise(args);\n}\n\n/**\n * Apply multiplicative 1-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting](\n *      http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianDropout(args: GaussianDropoutArgs): GaussianDropout {\n  return new GaussianDropout(args);\n}\n\n/**\n * Applies Alpha Dropout to the input.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\n * to their original values, in order to ensure the self-normalizing property\n * even after this dropout.\n * Alpha Dropout fits well to Scaled Exponential Linear Units\n * by randomly setting activations to the negative saturation value.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\n *     shape for randomly generated keep/drop flags.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function alphaDropout(args: AlphaDropoutArgs): AlphaDropout {\n  return new AlphaDropout(args);\n}\n\n/**\n * Masks a sequence by using a mask value to skip timesteps.\n *\n * If all features for a given sample timestep are equal to `mask_value`,\n * then the sample timestep will be masked (skipped) in all downstream layers\n * (as long as they support masking).\n *\n * If any downstream layer does not support masking yet receives such\n * an input mask, an exception will be raised.\n *\n * Arguments:\n *   - `maskValue`: Either None or mask value to skip.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Mask', namespace: 'layers'}\n */\nexport function masking(args?: MaskingArgs): Layer {\n  return new Masking(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport * as losses from './losses';\nimport * as metrics from './metrics';\n\n/**\n * Binary accuracy metric function.\n *\n * `yTrue` and `yPred` can have 0-1 values. Example:\n * ```js\n * const x = tf.tensor2d([[1, 1, 1, 1], [0, 0, 0, 0]], [2, 4]);\n * const y = tf.tensor2d([[1, 0, 1, 0], [0, 0, 0, 1]], [2, 4]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * `yTrue` and `yPred` can also have floating-number values between 0 and 1, in\n * which case the values will be thresholded at 0.5 to yield 0-1 values (i.e.,\n * a value >= 0.5 and <= 1.0 is interpreted as 1.\n * )\n * Example:\n * ```js\n * const x = tf.tensor1d([1, 1, 1, 1, 0, 0, 0, 0]);\n * const y = tf.tensor1d([0.2, 0.4, 0.6, 0.8, 0.2, 0.3, 0.4, 0.7]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryAccuracy(yTrue, yPred);\n}\n\n/**\n * Binary crossentropy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0], [1], [1], [1]]);\n * const y = tf.tensor2d([[0], [0], [0.5], [1]]);\n * const crossentropy = tf.metrics.binaryCrossentropy(x, y);\n * crossentropy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction, probabilities for the `1` case.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryCrossentropy(yTrue, yPred);\n}\n\n/**\n * Sparse categorical accuracy metric function.\n *\n * Example:\n * ```js\n *\n * const yTrue = tf.tensor1d([1, 1, 2, 2, 0]);\n * const yPred = tf.tensor2d(\n *      [[0, 1, 0], [1, 0, 0], [0, 0.4, 0.6], [0, 0.6, 0.4], [0.7, 0.3, 0]]);\n * const crossentropy = tf.metrics.sparseCategoricalAccuracy(yTrue, yPred);\n * crossentropy.print();\n * ```\n *\n * @param yTrue True labels: indices.\n * @param yPred Predicted probabilities or logits.\n * @returns Accuracy tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.sparseCategoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical accuracy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0, 0, 0, 1], [0, 0, 0, 1]]);\n * const y = tf.tensor2d([[0.1, 0.8, 0.05, 0.05], [0.1, 0.05, 0.05, 0.8]]);\n * const accuracy = tf.metrics.categoricalAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth: one-hot encoding of categories.\n * @param yPred Binary Tensor of prediction: probabilities or logits for the\n *   same categories as in `yTrue`.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical crossentropy between an output tensor and a target tensor.\n *\n * @param target A tensor of the same shape as `output`.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalCrossentropy(yTrue, yPred);\n}\n\n/**\n * Computes the precision of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const precision = tf.metrics.precision(x, y);\n * precision.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Precision Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.precision(yTrue, yPred);\n}\n\n/**\n * Computes the recall of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const recall = tf.metrics.recall(x, y);\n * recall.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Recall Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.recall(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Cosine proximity.\n *\n * Mathematically, cosine proximity is defined as:\n *   `-sum(l2Normalize(yTrue) * l2Normalize(yPred))`,\n * wherein `l2Normalize()` normalizes the L2 norm of the input to 1 and `*`\n * represents element-wise multiplication.\n *\n * ```js\n * const yTrue = tf.tensor2d([[1, 0], [1, 0]]);\n * const yPred = tf.tensor2d([[1 / Math.sqrt(2), 1 / Math.sqrt(2)], [0, 1]]);\n * const proximity = tf.metrics.cosineProximity(yTrue, yPred);\n * proximity.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Cosine proximity Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.cosineProximity(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute error.\n *\n * Mathematically, mean absolute error is defined as:\n *   `mean(abs(yPred - yTrue))`,\n * wherein the `mean` is applied over feature dimensions.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [0, 0], [2, 3]]);\n * const yPred = tf.tensor2d([[0, 1], [0, 1], [-2, -3]]);\n * const mse = tf.metrics.meanAbsoluteError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsoluteError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute percentage error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [10, 20]]);\n * const yPred = tf.tensor2d([[0, 1], [11, 24]]);\n * const mse = tf.metrics.meanAbsolutePercentageError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MAPE`, `tf.metrics.mape`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute percentage error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function MAPE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function mape(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean squared error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [3, 4]]);\n * const yPred = tf.tensor2d([[0, 1], [-3, -4]]);\n * const mse = tf.metrics.meanSquaredError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MSE`, `tf.metrics.mse`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean squared error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function MSE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function mse(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nexport {modelFromJSON} from './models';\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport * as regularizers from './regularizers';\n// tslint:disable-next-line:max-line-length\nimport {L1Args, L1L2, L1L2Args, L2Args, Regularizer} from './regularizers';\n\n/**\n * Regularizer for L1 and L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x)) + sum(l2 * x^2)\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1l2(config?: L1L2Args): Regularizer {\n  return new L1L2(config);\n}\n\n/**\n * Regularizer for L1 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x))\n * @param args l1 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1(config?: L1Args): Regularizer {\n  return regularizers.l1(config);\n}\n\n/**\n * Regularizer for L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l2 * x^2)\n * @param args l2 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l2(config?: L2Args): Regularizer {\n  return regularizers.l2(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {BaseCallback} from './base_callbacks';\nimport {Container} from './engine/container';\nimport {LayersModel} from './engine/training';\nimport {NotImplementedError} from './errors';\nimport {Logs, resolveScalarsInLogs} from './logs';\n\nexport abstract class Callback extends BaseCallback {\n  /** Instance of `keras.models.Model`. Reference of the model being trained. */\n  model: LayersModel = null;\n\n  setModel(model: Container): void {\n    if (!(model instanceof LayersModel)) {\n      throw new Error('model must be a LayersModel, not some other Container');\n    }\n    this.model = model;\n  }\n}\n\nexport interface EarlyStoppingCallbackArgs {\n  /**\n   * Quantity to be monitored.\n   *\n   * Defaults to 'val_loss'.\n   */\n  monitor?: string;\n\n  /**\n   * Minimum change in the monitored quantity to qualify as improvement,\n   * i.e., an absolute change of less than `minDelta` will count as no\n   * improvement.\n   *\n   * Defaults to 0.\n   */\n  minDelta?: number;\n\n  /**\n   * Number of epochs with no improvement after which training will be stopped.\n   *\n   * Defaults to 0.\n   */\n  patience?: number;\n\n  /** Verbosity mode. */\n  verbose?: number;\n\n  /**\n   * Mode: one of 'min', 'max', and 'auto'.\n   * - In 'min' mode, training will be stopped when the quantity monitored has\n   *   stopped decreasing.\n   * - In 'max' mode, training will be stopped when the quantity monitored has\n   *   stopped increasing.\n   * - In 'auto' mode, the direction is inferred automatically from the name of\n   *   the monitored quantity.\n   *\n   * Defaults to 'auto'.\n   */\n  mode?: 'auto'|'min'|'max';\n\n  /**\n   * Baseline value of the monitored quantity.\n   *\n   * If specified, training will be stopped if the model doesn't show\n   * improvement over the baseline.\n   */\n  baseline?: number;\n\n  /**\n   * Whether to restore model weights from the epoch with the best value\n   * of the monitored quantity. If `False`, the model weights obtained at the\n   * at the last step of training are used.\n   *\n   * **`True` is not supported yet.**\n   */\n  restoreBestWeights?: boolean;\n}\n\nfunction less(currVal: number, prevVal: number) {\n  return currVal < prevVal;\n}\n\nfunction greater(currVal: number, prevVal: number) {\n  return currVal > prevVal;\n}\n\n/**\n * A Callback that stops training when a monitored quantity has stopped\n * improving.\n */\nexport class EarlyStopping extends Callback {\n  protected readonly monitor: string;\n  protected readonly minDelta: number;\n  protected readonly patience: number;\n  protected readonly baseline: number;\n  protected readonly verbose: number;\n  protected readonly mode: 'auto'|'min'|'max';\n\n  protected monitorFunc: (currVal: number, prevVal: number) => boolean;\n\n  private wait: number;\n  private stoppedEpoch: number;\n  private best: number;\n\n  constructor(args?: EarlyStoppingCallbackArgs) {\n    super();\n    if (args == null) {\n      args = {};\n    }\n    if (args.restoreBestWeights) {\n      throw new NotImplementedError(\n          'restoreBestWeights = True is not implemented in EarlyStopping yet.');\n    }\n\n    this.monitor = args.monitor || 'val_loss';\n    this.minDelta = Math.abs(args.minDelta || 0);\n    this.patience = args.patience || 0;\n    this.verbose = args.verbose || 0;\n    this.mode = args.mode || 'auto';\n    this.baseline = args.baseline;\n\n    if (['auto', 'min', 'max'].indexOf(this.mode) === -1) {\n      console.warn(\n          `EarlyStopping mode '${this.mode}' is invalid. ` +\n          `Falling back to mode 'auto'.`);\n      this.mode = 'auto';\n    }\n\n    if (this.mode === 'min') {\n      this.monitorFunc = less;\n    } else if (this.mode === 'max') {\n      this.monitorFunc = greater;\n    } else {\n      // For mode === 'auto'.\n      if (this.monitor.indexOf('acc') !== -1) {\n        this.monitorFunc = greater;\n      } else {\n        this.monitorFunc = less;\n      }\n    }\n\n    if (this.monitorFunc === less) {\n      this.minDelta *= -1;\n    }\n  }\n\n  async onTrainBegin(logs?: Logs) {\n    this.wait = 0;\n    this.stoppedEpoch = 0;\n    if (this.baseline != null) {\n      this.best = this.baseline;\n    } else {\n      this.best = this.monitorFunc === less ? Infinity : -Infinity;\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: Logs) {\n    await resolveScalarsInLogs(logs);\n    const current = this.getMonitorValue(logs);\n    if (current == null) {\n      return;\n    }\n\n    if (this.monitorFunc(current - this.minDelta, this.best)) {\n      this.best = current;\n      this.wait = 0;\n      // TODO(cais): Logic for restoreBestWeights.\n    } else {\n      this.wait++;\n      if (this.wait >= this.patience) {\n        this.stoppedEpoch = epoch;\n        this.model.stopTraining = true;\n      }\n      // TODO(cais): Logic for restoreBestWeights.\n    }\n  }\n\n  async onTrainEnd(logs?: Logs) {\n    if (this.stoppedEpoch > 0 && this.verbose) {\n      console.log(`Epoch ${this.stoppedEpoch}: early stopping.`);\n    }\n  }\n\n  private getMonitorValue(logs: Logs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const monitorValue = logs[this.monitor];\n    if (monitorValue == null) {\n      console.warn(\n          `Metric for EarlyStopping ${this.monitor} is not available. ` +\n          `Available metrics are: ${Object.keys(logs)}`);\n    }\n    return monitorValue;\n  }\n}\n\n/**\n * Factory function for a Callback that stops training when a monitored\n * quantity has stopped improving.\n *\n * Early stopping is a type of regularization, and protects model against\n * overfitting.\n *\n * The following example based on fake data illustrates how this callback\n * can be used during `tf.LayersModel.fit()`:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense({\n *   units: 3,\n *   activation: 'softmax',\n *   kernelInitializer: 'ones',\n *   inputShape: [2]\n * }));\n * const xs = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const ys = tf.tensor2d([[1, 0, 0], [0, 1, 0]], [2, 3]);\n * const xsVal = tf.tensor2d([4, 3, 2, 1], [2, 2]);\n * const ysVal = tf.tensor2d([[0, 0, 1], [0, 1, 0]], [2, 3]);\n * model.compile(\n *     {loss: 'categoricalCrossentropy', optimizer: 'sgd', metrics: ['acc']});\n *\n * // Without the EarlyStopping callback, the val_acc value would be:\n * //   0.5, 0.5, 0.5, 0.5, ...\n * // With val_acc being monitored, training should stop after the 2nd epoch.\n * const history = await model.fit(xs, ys, {\n *   epochs: 10,\n *   validationData: [xsVal, ysVal],\n *   callbacks: tf.callbacks.earlyStopping({monitor: 'val_acc'})\n * });\n *\n * // Expect to see a length-2 array.\n * console.log(history.history.val_acc);\n * ```\n *\n * @doc {\n *   heading: 'Callbacks',\n *   namespace: 'callbacks'\n * }\n */\nexport function earlyStopping(args?: EarlyStoppingCallbackArgs) {\n  return new EarlyStopping(args);\n}\n\nexport const callbacks = {earlyStopping};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/* tslint:disable */\n\n/** Properties of an Any. */\nexport declare interface IAny {\n  /** Any typeUrl */\n  typeUrl?: (string|null);\n\n  /** Any value */\n  value?: (Uint8Array|null);\n}\n\n/** DataType enum. */\nexport enum DataType {\n  'DT_INVALID' = 0,\n  'DT_FLOAT' = 1,\n  'DT_DOUBLE' = 2,\n  'DT_INT32' = 3,\n  'DT_UINT8' = 4,\n  'DT_INT16' = 5,\n  'DT_INT8' = 6,\n  'DT_STRING' = 7,\n  'DT_COMPLEX64' = 8,\n  'DT_INT64' = 9,\n  'DT_BOOL' = 10,\n  'DT_QINT8' = 11,\n  'DT_QUINT8' = 12,\n  'DT_QINT32' = 13,\n  'DT_BFLOAT16' = 14,\n  'DT_FLOAT_REF' = 101,\n  'DT_DOUBLE_REF' = 102,\n  'DT_INT32_REF' = 103,\n  'DT_UINT8_REF' = 104,\n  'DT_INT16_REF' = 105,\n  'DT_INT8_REF' = 106,\n  'DT_STRING_REF' = 107,\n  'DT_COMPLEX64_REF' = 108,\n  'DT_INT64_REF' = 109,\n  'DT_BOOL_REF' = 110,\n  'DT_QINT8_REF' = 111,\n  'DT_QUINT8_REF' = 112,\n  'DT_QINT32_REF' = 113,\n  'DT_BFLOAT16_REF' = 114\n}\n\n/** Properties of a TensorShape. */\nexport declare interface ITensorShape {\n  /** TensorShape dim */\n  dim?: (TensorShape.IDim[]|null);\n\n  /** TensorShape unknownRank */\n  unknownRank?: (boolean|null);\n}\n\nexport namespace TensorShape {\n  /** Properties of a Dim. */\n  export declare interface IDim {\n    /** Dim size */\n    size?: (number|string|null);\n\n    /** Dim name */\n    name?: (string|null);\n  }\n}\n\n/** Properties of a Tensor. */\nexport declare interface ITensor {\n  /** Tensor dtype */\n  dtype?: (DataType|null);\n\n  /** Tensor tensorShape */\n  tensorShape?: (ITensorShape|null);\n\n  /** Tensor versionNumber */\n  versionNumber?: (number|null);\n\n  /** Tensor tensorContent */\n  tensorContent?: (Uint8Array|null);\n\n  /** Tensor floatVal */\n  floatVal?: (number[]|null);\n\n  /** Tensor doubleVal */\n  doubleVal?: (number[]|null);\n\n  /** Tensor intVal */\n  intVal?: (number[]|null);\n\n  /** Tensor stringVal */\n  stringVal?: (Uint8Array[]|null);\n\n  /** Tensor scomplexVal */\n  scomplexVal?: (number[]|null);\n\n  /** Tensor int64Val */\n  int64Val?: ((number | string)[]|null);\n\n  /** Tensor boolVal */\n  boolVal?: (boolean[]|null);\n\n  /** Tensor uint32Val */\n  uint32Val?: (number[]|null);\n\n  /** Tensor uint64Val */\n  uint64Val?: ((number | string)[]|null);\n}\n\n/** Properties of an AttrValue. */\nexport declare interface IAttrValue {\n  /** AttrValue list */\n  list?: (AttrValue.IListValue|null);\n\n  /** AttrValue s */\n  s?: (string|null);\n\n  /** AttrValue i */\n  i?: (number|string|null);\n\n  /** AttrValue f */\n  f?: (number|null);\n\n  /** AttrValue b */\n  b?: (boolean|null);\n\n  /** AttrValue type */\n  type?: (DataType|null);\n\n  /** AttrValue shape */\n  shape?: (ITensorShape|null);\n\n  /** AttrValue tensor */\n  tensor?: (ITensor|null);\n\n  /** AttrValue placeholder */\n  placeholder?: (string|null);\n\n  /** AttrValue func */\n  func?: (INameAttrList|null);\n}\n\nexport namespace AttrValue {\n  /** Properties of a ListValue. */\n  export declare interface IListValue {\n    /** ListValue s */\n    s?: (string[]|null);\n\n    /** ListValue i */\n    i?: ((number | string)[]|null);\n\n    /** ListValue f */\n    f?: (number[]|null);\n\n    /** ListValue b */\n    b?: (boolean[]|null);\n\n    /** ListValue type */\n    type?: (DataType[]|null);\n\n    /** ListValue shape */\n    shape?: (ITensorShape[]|null);\n\n    /** ListValue tensor */\n    tensor?: (ITensor[]|null);\n\n    /** ListValue func */\n    func?: (INameAttrList[]|null);\n  }\n}\n\n/** Properties of a NameAttrList. */\nexport declare interface INameAttrList {\n  /** NameAttrList name */\n  name?: (string|null);\n\n  /** NameAttrList attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a NodeDef. */\nexport declare interface INodeDef {\n  /** NodeDef name */\n  name?: (string|null);\n\n  /** NodeDef op */\n  op?: (string|null);\n\n  /** NodeDef input */\n  input?: (string[]|null);\n\n  /** NodeDef device */\n  device?: (string|null);\n\n  /** NodeDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a VersionDef. */\nexport declare interface IVersionDef {\n  /** VersionDef producer */\n  producer?: (number|null);\n\n  /** VersionDef minConsumer */\n  minConsumer?: (number|null);\n\n  /** VersionDef badConsumers */\n  badConsumers?: (number[]|null);\n}\n\n/** Properties of a GraphDef. */\nexport declare interface IGraphDef {\n  /** GraphDef node */\n  node?: (INodeDef[]|null);\n\n  /** GraphDef versions */\n  versions?: (IVersionDef|null);\n\n  /** GraphDef library */\n  library?: (IFunctionDefLibrary|null);\n}\n\n/** Properties of a CollectionDef. */\nexport declare interface ICollectionDef {\n  /** CollectionDef nodeList */\n  nodeList?: (CollectionDef.INodeList|null);\n\n  /** CollectionDef bytesList */\n  bytesList?: (CollectionDef.IBytesList|null);\n\n  /** CollectionDef int64List */\n  int64List?: (CollectionDef.IInt64List|null);\n\n  /** CollectionDef floatList */\n  floatList?: (CollectionDef.IFloatList|null);\n\n  /** CollectionDef anyList */\n  anyList?: (CollectionDef.IAnyList|null);\n}\n\nexport namespace CollectionDef {\n  /** Properties of a NodeList. */\n  export declare interface INodeList {\n    /** NodeList value */\n    value?: (string[]|null);\n  }\n\n  /** Properties of a BytesList. */\n  export declare interface IBytesList {\n    /** BytesList value */\n    value?: (Uint8Array[]|null);\n  }\n\n  /** Properties of an Int64List. */\n  export declare interface IInt64List {\n    /** Int64List value */\n    value?: ((number | string)[]|null);\n  }\n\n  /** Properties of a FloatList. */\n  export declare interface IFloatList {\n    /** FloatList value */\n    value?: (number[]|null);\n  }\n\n  /** Properties of an AnyList. */\n  export declare interface IAnyList {\n    /** AnyList value */\n    value?: (IAny[]|null);\n  }\n}\n\n/** Properties of a SaverDef. */\nexport declare interface ISaverDef {\n  /** SaverDef filenameTensorName */\n  filenameTensorName?: (string|null);\n\n  /** SaverDef saveTensorName */\n  saveTensorName?: (string|null);\n\n  /** SaverDef restoreOpName */\n  restoreOpName?: (string|null);\n\n  /** SaverDef maxToKeep */\n  maxToKeep?: (number|null);\n\n  /** SaverDef sharded */\n  sharded?: (boolean|null);\n\n  /** SaverDef keepCheckpointEveryNHours */\n  keepCheckpointEveryNHours?: (number|null);\n\n  /** SaverDef version */\n  version?: (SaverDef.CheckpointFormatVersion|null);\n}\n\nexport namespace SaverDef {\n  /** CheckpointFormatVersion enum. */\n  export enum CheckpointFormatVersion {'LEGACY' = 0, 'V1' = 1, 'V2' = 2}\n}\n\n/** Properties of a TensorInfo. */\nexport declare interface ITensorInfo {\n  /** TensorInfo name */\n  name?: (string|null);\n\n  /** TensorInfo cooSparse */\n  cooSparse?: (TensorInfo.ICooSparse|null);\n\n  /** TensorInfo dtype */\n  dtype?: (DataType|null);\n\n  /** TensorInfo tensorShape */\n  tensorShape?: (ITensorShape|null);\n}\n\nexport namespace TensorInfo {\n  /** Properties of a CooSparse. */\n  export declare interface ICooSparse {\n    /** CooSparse valuesTensorName */\n    valuesTensorName?: (string|null);\n\n    /** CooSparse indicesTensorName */\n    indicesTensorName?: (string|null);\n\n    /** CooSparse denseShapeTensorName */\n    denseShapeTensorName?: (string|null);\n  }\n}\n\n/** Properties of a SignatureDef. */\nexport declare interface ISignatureDef {\n  /** SignatureDef inputs */\n  inputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef outputs */\n  outputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef methodName */\n  methodName?: (string|null);\n}\n\n/** Properties of an AssetFileDef. */\nexport declare interface IAssetFileDef {\n  /** AssetFileDef tensorInfo */\n  tensorInfo?: (ITensorInfo|null);\n\n  /** AssetFileDef filename */\n  filename?: (string|null);\n}\n\n/** Properties of an OpDef. */\nexport declare interface IOpDef {\n  /** OpDef name */\n  name?: (string|null);\n\n  /** OpDef inputArg */\n  inputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef outputArg */\n  outputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef attr */\n  attr?: (OpDef.IAttrDef[]|null);\n\n  /** OpDef deprecation */\n  deprecation?: (OpDef.IOpDeprecation|null);\n\n  /** OpDef summary */\n  summary?: (string|null);\n\n  /** OpDef description */\n  description?: (string|null);\n\n  /** OpDef isCommutative */\n  isCommutative?: (boolean|null);\n\n  /** OpDef isAggregate */\n  isAggregate?: (boolean|null);\n\n  /** OpDef isStateful */\n  isStateful?: (boolean|null);\n\n  /** OpDef allowsUninitializedInput */\n  allowsUninitializedInput?: (boolean|null);\n}\n\nexport namespace OpDef {\n  /** Properties of an ArgDef. */\n  export declare interface IArgDef {\n    /** ArgDef name */\n    name?: (string|null);\n\n    /** ArgDef description */\n    description?: (string|null);\n\n    /** ArgDef type */\n    type?: (DataType|null);\n\n    /** ArgDef typeAttr */\n    typeAttr?: (string|null);\n\n    /** ArgDef numberAttr */\n    numberAttr?: (string|null);\n\n    /** ArgDef typeListAttr */\n    typeListAttr?: (string|null);\n\n    /** ArgDef isRef */\n    isRef?: (boolean|null);\n  }\n\n  /** Properties of an AttrDef. */\n  export declare interface IAttrDef {\n    /** AttrDef name */\n    name?: (string|null);\n\n    /** AttrDef type */\n    type?: (string|null);\n\n    /** AttrDef defaultValue */\n    defaultValue?: (IAttrValue|null);\n\n    /** AttrDef description */\n    description?: (string|null);\n\n    /** AttrDef hasMinimum */\n    hasMinimum?: (boolean|null);\n\n    /** AttrDef minimum */\n    minimum?: (number|string|null);\n\n    /** AttrDef allowedValues */\n    allowedValues?: (IAttrValue|null);\n  }\n\n  /** Properties of an OpDeprecation. */\n  export declare interface IOpDeprecation {\n    /** OpDeprecation version */\n    version?: (number|null);\n\n    /** OpDeprecation explanation */\n    explanation?: (string|null);\n  }\n}\n\n/** Properties of an OpList. */\nexport declare interface IOpList {\n  /** OpList op */\n  op?: (IOpDef[]|null);\n}\n\n/** Properties of a MetaGraphDef. */\nexport declare interface IMetaGraphDef {\n  /** MetaGraphDef metaInfoDef */\n  metaInfoDef?: (MetaGraphDef.IMetaInfoDef|null);\n\n  /** MetaGraphDef graphDef */\n  graphDef?: (IGraphDef|null);\n\n  /** MetaGraphDef saverDef */\n  saverDef?: (ISaverDef|null);\n\n  /** MetaGraphDef collectionDef */\n  collectionDef?: ({[k: string]: ICollectionDef}|null);\n\n  /** MetaGraphDef signatureDef */\n  signatureDef?: ({[k: string]: ISignatureDef}|null);\n\n  /** MetaGraphDef assetFileDef */\n  assetFileDef?: (IAssetFileDef[]|null);\n}\n\nexport namespace MetaGraphDef {\n  /** Properties of a MetaInfoDef. */\n  export declare interface IMetaInfoDef {\n    /** MetaInfoDef metaGraphVersion */\n    metaGraphVersion?: (string|null);\n\n    /** MetaInfoDef strippedOpList */\n    strippedOpList?: (IOpList|null);\n\n    /** MetaInfoDef anyInfo */\n    anyInfo?: (IAny|null);\n\n    /** MetaInfoDef tags */\n    tags?: (string[]|null);\n\n    /** MetaInfoDef tensorflowVersion */\n    tensorflowVersion?: (string|null);\n\n    /** MetaInfoDef tensorflowGitVersion */\n    tensorflowGitVersion?: (string|null);\n  }\n}\n\n/** Properties of a SavedModel. */\nexport declare interface ISavedModel {\n  /** SavedModel savedModelSchemaVersion */\n  savedModelSchemaVersion?: (number|string|null);\n\n  /** SavedModel metaGraphs */\n  metaGraphs?: (IMetaGraphDef[]|null);\n}\n\n/** Properties of a FunctionDefLibrary. */\nexport declare interface IFunctionDefLibrary {\n  /** FunctionDefLibrary function */\n  'function'?: (IFunctionDef[]|null);\n\n  /** FunctionDefLibrary gradient */\n  gradient?: (IGradientDef[]|null);\n}\n\n/** Properties of a FunctionDef. */\nexport declare interface IFunctionDef {\n  /** FunctionDef signature */\n  signature?: (IOpDef|null);\n\n  /** FunctionDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n\n  /** FunctionDef nodeDef */\n  nodeDef?: (INodeDef[]|null);\n\n  /** FunctionDef ret */\n  ret?: ({[k: string]: string}|null);\n}\n\n/** Properties of a GradientDef. */\nexport declare interface IGradientDef {\n  /** GradientDef functionName */\n  functionName?: (string|null);\n\n  /** GradientDef gradientFunc */\n  gradientFunc?: (string|null);\n}\n", "\n/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpExecutor, OpMapper} from '../types';\n\nconst CUSTOM_OPS: {[key: string]: OpMapper} = {};\n\n/**\n * Register an Op for graph model executor. This allow you to register\n * TensorFlow custom op or override existing op.\n *\n * Here is an example of registering a new MatMul Op.\n * ```js\n * const customMatmul = (node) =>\n *    tf.matMul(\n *        node.inputs[0], node.inputs[1],\n *        node.attrs['transpose_a'], node.attrs['transpose_b']);\n *\n * tf.registerOp('MatMul', customMatmul);\n * ```\n * The inputs and attrs of the node object is based on the TensorFlow op\n * registry.\n *\n * @param name The Tensorflow Op name.\n * @param opFunc An op function which is called with the current graph node\n * during execution and needs to return a tensor or a list of tensors. The node\n * has the following attributes:\n *    - attr: A map from attribute name to its value\n *    - inputs: A list of input tensors\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function registerOp(name: string, opFunc: OpExecutor) {\n  const opMapper: OpMapper = {\n    tfOpName: name,\n    category: 'custom',\n    inputs: [],\n    attrs: [],\n    customExecutor: opFunc\n  };\n\n  CUSTOM_OPS[name] = opMapper;\n}\n\n/**\n * Retrieve the OpMapper object for the registered op.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function getRegisteredOp(name: string): OpMapper {\n  return CUSTOM_OPS[name];\n}\n\n/**\n * Deregister the Op for graph model executor.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function deregisterOp(name: string) {\n  delete CUSTOM_OPS[name];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {clone, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {Node, ValueType} from '../types';\n\nexport function getParamValue(\n    paramName: string, node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  const inputParam = node.inputParams[paramName];\n  if (inputParam && inputParam.inputIndexStart !== undefined) {\n    const start = inputParam.inputIndexStart;\n    const end = inputParam.inputIndexEnd === 0 ?\n        undefined :\n        (inputParam.inputIndexEnd === undefined ? start + 1 :\n                                                  inputParam.inputIndexEnd);\n    if (inputParam.type === 'tensor') {\n      return getTensor(\n          node.inputNames[inputParam.inputIndexStart], tensorMap, context);\n    }\n    if (inputParam.type === 'tensors') {\n      const inputs = node.inputNames.slice(start, end);\n\n      return inputs.map(name => getTensor(name, tensorMap, context));\n    }\n    const tensor =\n        getTensor(node.inputNames.slice(start)[0], tensorMap, context);\n    const data = tensor.dataSync();\n    return inputParam.type === 'number' ?\n        data[0] :\n        util.toNestedArray(tensor.shape, data);\n  }\n  const attrParam = node.attrParams[paramName];\n  return attrParam && attrParam.value;\n}\n\n/**\n * Retrieve the tensor based on input name by extracting the node name and\n * output index information.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensor(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor {\n  const [nodeName, index] = parseNodeName(name);\n  const contextId = context.currentContextIds.find(contextId => {\n    return !!tensorsMap[getNodeNameWithContextId(nodeName, contextId)];\n  });\n\n  return contextId !== undefined ?\n      tensorsMap[getNodeNameWithContextId(nodeName, contextId)][index] :\n      undefined;\n}\n\n/**\n * Retrieve the tensors based on input name for current context.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensorsForCurrentContenxt(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor[] {\n  return tensorsMap[getNodeNameWithContextId(name, context.currentContextId)];\n}\n\n/**\n * Returns the node name and index from the Node input name.\n * @param inputName The input name of the node, in format of\n * node_name:output_index, i.e. MatMul:0, if the output_index is not set, it is\n * default to 0.\n */\nexport function getNodeNameAndIndex(\n    inputName: string, context?: ExecutionContext): [string, number] {\n  const [nodeName, index] = parseNodeName(inputName);\n\n  return [\n    getNodeNameWithContextId(nodeName, context && context.currentContextId),\n    index\n  ];\n}\n\nfunction getNodeNameWithContextId(name: string, contextId?: string): string {\n  return !!contextId ? `${name}-${contextId}` : name;\n}\n\nexport function parseNodeName(name: string): [string, number] {\n  const parts = name.split(':');\n  if (parts.length === 1) {\n    return [name, 0];\n  }\n\n  const nodeName = parts[0];\n  return [nodeName, Number(parts[parts.length - 1])];\n}\n\nexport function split(arr: number[], size: number) {\n  const res = [];\n  for (let i = 0; i < arr.length; i += size) {\n    res.push(arr.slice(i, i + size));\n  }\n  return res;\n}\nexport function getPadding(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  let pad = getParamValue('pad', node, tensorMap, context);\n  if (pad === 'explicit') {\n    // This is 1d array, we need to convert it to 2d array\n    pad = getParamValue('explicitPaddings', node, tensorMap, context);\n    const explicitPadding: [\n      [number, number], [number, number], [number, number], [number, number]\n    ] = [[0, 0], [0, 0], [0, 0], [0, 0]];\n    for (let i = 0; i < 4; i++) {\n      explicitPadding[i][0] = (pad as number[])[i * 2];\n      explicitPadding[i][1] = (pad as number[])[i * 2 + 1];\n    }\n    return explicitPadding;\n  }\n  return pad;\n}\n\n/**\n *  Reuse the tensor if it is marked as keep, otherwise clone the tensor to\n *  avoid disposal. This is important for TensorArray and TensorList ops, since\n *  internally they use a tensor as the id for TensorArray and TensorList, and\n * to simplify lookup, they also use Tensor.id as the key to the internal map.\n * These id tensors have been marked as kept in the backend, we need avoid clone\n * them in order to create new Tensor.id.\n * @param tensor\n */\nexport function cloneTensor(tensor: Tensor): Tensor {\n  return tensor.kept ? tensor : clone(tensor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Add',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddV2',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddN',\n    'category': 'arithmetic',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'BiasAdd',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sub',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'RealDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Div',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'DivNoNan',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mul',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Maximum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Minimum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Pow',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SquaredDifference',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorMod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Abs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan2',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ceil',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ClipByValue',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'clip_value_min', 'name': 'clipValueMin', 'type': 'number'},\n      {'tfName': 'clip_value_max', 'name': 'clipValueMax', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Complex',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'real', 'type': 'tensor'},\n      {'start': 1, 'name': 'imag', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ComplexAbs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Elu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Exp',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Floor',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Imag',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Neg',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Real',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Prelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'alpha', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu6',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'clipValueMin',\n        'name': 'clipValueMin',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'clipValueMax',\n        'name': 'clipValueMax',\n        'type': 'number',\n        'defaultValue': 6\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Selu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sigmoid',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Rsqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Square',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sign',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Round',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Expm1',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log1p',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Reciprocal',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Softplus',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Erf',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axes', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'keep_dims',\n        'name': 'keepDims',\n        'type': 'bool',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LeakyRelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 0.2\n      },\n      {\n        'tfName': 'T',\n        'name': 'dtype',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'LoopCond',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'pred', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Switch',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'data', 'type': 'tensor'},\n      {'start': 1, 'name': 'pred', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Merge',\n    'category': 'control',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Enter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'frame_name', 'name': 'frameName', 'type': 'string'},\n      {'tfName': 'is_constant', 'name': 'isConstant', 'type': 'bool'}\n    ]\n  },\n  {\n    'tfOpName': 'Exit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NextIteration',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'size', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'dynamic_size', 'name': 'dynamicSize', 'type': 'bool'},\n      {'tfName': 'clear_after_read', 'name': 'clearAfterRead', 'type': 'bool'},\n      {\n        'tfName': 'identical_element_shapes',\n        'name': 'identicalElementShapes',\n        'type': 'bool'\n      },\n      {'tfName': 'tensor_array_name', 'name': 'name', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayWriteV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayReadV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{\n      'tfName': 'dtype',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  },\n  {\n    'tfOpName': 'TensorArrayGatherV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayScatterV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArrayConcatV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}, {\n        'tfName': 'element_shape_except0',\n        'name': 'elementShapeExcept0',\n        'type': 'shape',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'TensorArraySplitV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArraySizeV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayCloseV3',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'StatelessIf',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'If',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'StatelessWhile',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'While',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListScatter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListScatterV2',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 3, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGather',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListSetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListReserve',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 1, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListFromTensor',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListStack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs': [\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'},\n      {'tfName': 'num_elements', 'name': 'numElements', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListSplit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListConcat',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListPopBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListPushBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n];\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'AvgPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPoolWithArgmax',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'}, {\n        'tfName': 'include_batch_in_index',\n        'name': 'includeBatchInIndex',\n        'type': 'bool'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AvgPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Conv1D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'stride', 'name': 'stride', 'type': 'number'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NWC'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'dilation',\n        'name': 'dilation',\n        'type': 'number',\n        'defaultValue': 1\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'useCudnnOnGpu', 'name': 'useCudnnOnGpu', 'type': 'bool'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': '_FusedConv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'use_cudnn_on_gpu',\n        'name': 'useCudnnOnGpu',\n        'type': 'bool',\n        'defaultValue': true\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n    ]\n  },\n  {\n    'tfOpName': 'Conv2DBackpropInput',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 2, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 0, 'name': 'outputShape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2d',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'FusedDepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ],\n  },\n  {\n    'tfOpName': 'Dilation2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'rates', 'name': 'dilations', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Fill',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n      {'start': 1, 'name': 'value', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'LinSpace',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'num', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'OneHot',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'depth', 'type': 'number'},\n      {'start': 2, 'name': 'onValue', 'type': 'number', 'defaultValue': 1},\n      {'start': 3, 'name': 'offValue', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [\n      {\n        'tfName': 'axis',\n        'name': 'axis',\n        'type': 'number',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ones',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'OnesLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'RandomUniform',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'minval',\n        'name': 'minval',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'maxval',\n        'name': 'maxval',\n        'type': 'number',\n        'defaultValue': 1\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Range',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'step', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [{'tfName': 'Tidx', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TruncatedNormal',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'means',\n        'name': 'mean',\n        'type': 'number',\n        'defaultValue': 0.0\n      },\n      {\n        'tfName': 'stddev',\n        'name': 'stdDev',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Zeros',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'ZerosLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'Multinomial',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'logits', 'type': 'tensor'},\n      {'start': 1, 'name': 'numSamples', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'},\n      {'tfName': 'seed2', 'name': 'seed2', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'output_dtype', 'name': 'output_dtype', 'type': 'dtype'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'NonMaxSuppressionV2',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV3',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV4',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'T_threshold',\n        'name': 'threshold',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {\n        'tfName': 'pad_to_max_output_size',\n        'name': 'padToMaxOutputSize',\n        'type': 'bool'\n      }\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV5',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'},\n      {'start': 5, 'name': 'softNmsSigma', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Where',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ListDiff',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'TopKV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'k', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'sorted', 'name': 'sorted', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Unique',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n  },\n  {\n    'tfOpName': 'UniqueV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n  },\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'PlaceholderWithDefault',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'default', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'Placeholder',\n    'category': 'graph',\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {'tfOpName': 'Const', 'category': 'graph'}, {\n    'tfOpName': 'Identity',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IdentityN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Snapshot',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Rank',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Size',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Shape',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'ShapeN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Print',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'data', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'message', 'name': 'message', 'type': 'string'}, {\n        'tfName': 'first_n',\n        'name': 'firstN',\n        'type': 'number',\n        'notSupported': true\n      },\n      {\n        'tfName': 'summarize',\n        'name': 'summarize',\n        'type': 'number',\n        'defaultValue': 3\n      }\n    ]\n  },\n  {'tfOpName': 'NoOp', 'category': 'graph', 'inputs': []}, {\n    'tfOpName': 'StopGradient',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'FakeQuantWithMinMaxVars',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'min', 'name': 'min', 'type': 'number'},\n      {'tfName': 'max', 'name': 'max', 'type': 'number'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ResizeBilinear',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ResizeNearestNeighbor',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'CropAndResize',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'image', 'type': 'tensor'},\n      {'start': 1, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 2, 'name': 'boxInd', 'type': 'tensor'},\n      {'start': 3, 'name': 'cropSize', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'method', 'name': 'method', 'type': 'string'}, {\n        'tfName': 'extrapolation_value',\n        'name': 'extrapolationValue',\n        'type': 'number'\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Equal',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NotEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Greater',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'GreaterEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Less',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LessEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalAnd',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalNot',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalOr',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Select',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SelectV2',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': '_FusedMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'}, {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMulV2',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Transpose',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'perm', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FusedBatchNorm',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV2',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV3',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'LRN',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'depth_radius',\n        'name': 'radius',\n        'type': 'number',\n        'defaultValue': 5\n      },\n      {'tfName': 'bias', 'name': 'bias', 'type': 'number', 'defaultValue': 1.0},\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {\n        'tfName': 'beta',\n        'name': 'beta',\n        'type': 'number',\n        'defaultValue': 0.5\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Softmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'LogSoftmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': true,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Max',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Mean',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Min',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Sum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'All',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Any',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'ArgMax',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'ArgMin',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Cumsum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'exclusive', 'name': 'exclusive', 'type': 'bool'},\n      {'tfName': 'reverse', 'name': 'reverse', 'type': 'bool'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ConcatV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': -1, 'name': 'tensors', 'type': 'tensors'},\n      {'start': -1, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs':\n        [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n  },\n  {\n    'tfOpName': 'Concat',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 1, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n      {'start': 0, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs': [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n\n  },\n  {\n    'tfOpName': 'GatherV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Gather',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'validate_indices',\n        'name': 'validateIndices',\n        'type': 'bool',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reverse',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'dims', 'type': 'bool', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ReverseV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Slice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'size', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'StridedSlice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'end', 'type': 'number[]'},\n      {'start': 3, 'name': 'strides', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'begin_mask',\n        'name': 'beginMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'end_mask',\n        'name': 'endMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'new_axis_mask',\n        'name': 'newAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'ellipsis_mask',\n        'name': 'ellipsisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'shrink_axis_mask',\n        'name': 'shrinkAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Pack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Unpack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'num',\n        'name': 'num',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Tile',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'reps', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Split',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'axis', 'type': 'number', 'defaultValue': 0},\n      {'start': 1, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'num_split',\n      'name': 'numOrSizeSplits',\n      'type': 'number',\n      'defaultValue': 1\n    }]\n  },\n  {\n    'tfOpName': 'SplitV',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'numOrSizeSplits', 'type': 'number[]'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'ScatterNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'values', 'type': 'tensor'},\n      {'start': 2, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'GatherNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': false,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IFFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'RFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'IRFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Cast',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'SrcT',\n        'name': 'sdtype',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {'tfName': 'DstT', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'ExpandDims',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Pad',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'constant_value',\n      'name': 'constantValue',\n      'type': 'number',\n      'defaultValue': 0\n    }]\n  },\n  {\n    'tfOpName': 'PadV2',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'}, {\n        'start': 2,\n        'name': 'constantValue',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reshape',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Squeeze',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'axis',\n      'tfDeprecatedName': 'squeeze_dims',\n      'name': 'axis',\n      'type': 'number[]'\n    }]\n  },\n  {\n    'tfOpName': 'SpaceToBatchND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'paddings', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'BatchToSpaceND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'crops', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthToSpace',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'block_size', 'name': 'blockSize', 'type': 'number'},\n      {'tfName': 'data_format', 'name': 'dataFormat', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'BroadcastTo',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': []\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, env} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\n\nimport {getRegisteredOp} from './custom_op/register';\nimport {getNodeNameAndIndex} from './executors/utils';\nimport * as arithmetic from './op_list/arithmetic';\nimport * as basicMath from './op_list/basic_math';\nimport * as control from './op_list/control';\nimport * as convolution from './op_list/convolution';\nimport * as creation from './op_list/creation';\nimport * as dynamic from './op_list/dynamic';\nimport * as evaluation from './op_list/evaluation';\nimport * as graph from './op_list/graph';\nimport * as image from './op_list/image';\nimport * as logical from './op_list/logical';\nimport * as matrices from './op_list/matrices';\nimport * as normalization from './op_list/normalization';\nimport * as reduction from './op_list/reduction';\nimport * as sliceJoin from './op_list/slice_join';\nimport * as spectral from './op_list/spectral';\nimport * as transformation from './op_list/transformation';\nimport {Graph, InputParamValue, Node, OpMapper, ParamValue} from './types';\n\nexport class OperationMapper {\n  private static _instance: OperationMapper;\n\n  private opMappers: {[key: string]: OpMapper};\n\n  // Singleton instance for the mapper\n  public static get Instance() {\n    return this._instance || (this._instance = new this());\n  }\n\n  // Loads the op mapping from the JSON file.\n  private constructor() {\n    const ops = [\n      arithmetic, basicMath, control, convolution, creation, dynamic,\n      evaluation, logical, image, graph, matrices, normalization, reduction,\n      sliceJoin, spectral, transformation\n    ];\n    const mappersJson: OpMapper[] = [].concat(...ops.map(op => op.json));\n\n    this.opMappers = mappersJson.reduce<{[key: string]: OpMapper}>(\n        (map, mapper: OpMapper) => {\n          map[mapper.tfOpName] = mapper;\n          return map;\n        },\n        {});\n  }\n\n  // Converts the model inference graph from Tensorflow GraphDef to local\n  // representation for TensorFlow.js API\n  transformGraph(\n      graph: tensorflow.IGraphDef,\n      signature: tensorflow.ISignatureDef = {}): Graph {\n    const tfNodes = graph.node;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    const initNodes: Node[] = [];\n    const nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n      map[node.name] = this.mapNode(node);\n      if (node.op.startsWith('Placeholder')) {\n        placeholders.push(map[node.name]);\n      } else if (node.op === 'Const') {\n        weights.push(map[node.name]);\n      } else if (node.input == null || node.input.length === 0) {\n        initNodes.push(map[node.name]);\n      }\n      return map;\n    }, {});\n\n    let inputs: Node[] = [];\n    const outputs: Node[] = [];\n    let inputNodeNameToKey: {[key: string]: string} = {};\n    let outputNodeNameToKey: {[key: string]: string} = {};\n    if (signature != null) {\n      inputNodeNameToKey = this.mapSignatureEntries(signature.inputs);\n      outputNodeNameToKey = this.mapSignatureEntries(signature.outputs);\n    }\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    // if signature has not outputs set, add any node that does not have\n    // outputs.\n    if (Object.keys(outputNodeNameToKey).length === 0) {\n      allNodes.forEach(key => {\n        const node = nodes[key];\n        if (node.children.length === 0) {\n          outputs.push(node);\n        }\n      });\n    } else {\n      Object.keys(outputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node != null) {\n          node.signatureKey = outputNodeNameToKey[name];\n          outputs.push(node);\n        }\n      });\n    }\n\n    if (Object.keys(inputNodeNameToKey).length > 0) {\n      Object.keys(inputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node) {\n          node.signatureKey = inputNodeNameToKey[name];\n          inputs.push(node);\n        }\n      });\n    } else {\n      inputs = placeholders;\n    }\n\n    let functions = {};\n    if (graph.library != null && graph.library.function != null) {\n      functions = graph.library.function.reduce((functions, func) => {\n        functions[func.signature.name] = this.mapFunction(func);\n        return functions;\n      }, {} as {[key: string]: Graph});\n    }\n\n    const result: Graph =\n        {nodes, inputs, outputs, weights, placeholders, signature, functions};\n\n    if (initNodes.length > 0) {\n      result.initNodes = initNodes;\n    }\n\n    return result;\n  }\n\n  private mapSignatureEntries(entries: {[k: string]: tensorflow.ITensorInfo}) {\n    return Object.keys(entries || {})\n        .reduce<{[key: string]: string}>((prev, curr) => {\n          prev[entries[curr].name] = curr;\n          return prev;\n        }, {});\n  }\n\n  private mapNode(node: tensorflow.INodeDef): Node {\n    // Unsupported ops will cause an error at run-time (not parse time), since\n    // they may not be used by the actual execution subgraph.\n    const mapper =\n        getRegisteredOp(node.op) || this.opMappers[node.op] || {} as OpMapper;\n    if (node.attr == null) {\n      node.attr = {};\n    }\n\n    const newNode: Node = {\n      name: node.name,\n      op: node.op,\n      category: mapper.category,\n      inputNames:\n          (node.input ||\n           []).map(input => input.startsWith('^') ? input.substr(1) : input),\n      inputs: [],\n      children: [],\n      inputParams: {},\n      attrParams: {},\n      rawAttrs: node.attr\n    };\n\n    if (mapper.inputs != null) {\n      newNode.inputParams =\n          mapper.inputs.reduce<{[key: string]: InputParamValue}>(\n              (map, param) => {\n                map[param.name] = {\n                  type: param.type,\n                  inputIndexStart: param.start,\n                  inputIndexEnd: param.end\n                };\n                return map;\n              },\n              {});\n    }\n    if (mapper.attrs != null) {\n      newNode.attrParams =\n          mapper.attrs.reduce<{[key: string]: ParamValue}>((map, param) => {\n            const type = param.type;\n            let value = undefined;\n            switch (param.type) {\n              case 'string':\n                value = getStringParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'string[]':\n                value = getStringArrayParam(\n                    node.attr, param.tfName, param.defaultValue as string[]);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string[]);\n                }\n                break;\n              case 'number':\n                value = getNumberParam(\n                    node.attr, param.tfName,\n                    (param.defaultValue || 0) as number);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumberParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number);\n                }\n                break;\n              case 'number[]':\n                value = getNumericArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumericArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'bool':\n                value = getBoolParam(\n                    node.attr, param.tfName, param.defaultValue as boolean);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean);\n                }\n                break;\n              case 'bool[]':\n                value = getBoolArrayParam(\n                    node.attr, param.tfName, param.defaultValue as boolean[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean[]);\n                }\n                break;\n              case 'shape':\n                value = getTensorShapeParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'shape[]':\n                value = getTensorShapeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[][]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[][]);\n                }\n                break;\n              case 'dtype':\n                value = getDtypeParam(\n                    node.attr, param.tfName, param.defaultValue as DataType);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType);\n                }\n                break;\n              case 'dtype[]':\n                value = getDtypeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as DataType[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType[]);\n                }\n                break;\n              case 'func':\n                value = getFuncParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getFuncParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'tensor':\n              case 'tensors':\n                break;\n              default:\n                throw new Error(\n                    `Unsupported param type: ${param.type} for op: ${node.op}`);\n            }\n            map[param.name] = {value, type};\n            return map;\n          }, {});\n    }\n    return newNode;\n  }\n\n  // map the TFunctionDef to TFJS graph object\n  private mapFunction(functionDef: tensorflow.IFunctionDef): Graph {\n    const tfNodes = functionDef.nodeDef;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    let nodes: {[key: string]: Node} = {};\n    if (tfNodes != null) {\n      nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n        map[node.name] = this.mapNode(node);\n        if (node.op === 'Const') {\n          weights.push(map[node.name]);\n        }\n        return map;\n      }, {});\n    }\n    const inputs: Node[] = [];\n    const outputs: Node[] = [];\n\n    functionDef.signature.inputArg.forEach(arg => {\n      const [nodeName, ] = getNodeNameAndIndex(arg.name);\n      const node: Node = {\n        name: nodeName,\n        op: 'Placeholder',\n        inputs: [],\n        inputNames: [],\n        category: 'graph',\n        inputParams: {},\n        attrParams: {dtype: {value: parseDtypeParam(arg.type), type: 'dtype'}},\n        children: []\n      };\n      node.signatureKey = arg.name;\n      inputs.push(node);\n      nodes[nodeName] = node;\n    });\n\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    const returnNodeMap = functionDef.ret;\n\n    functionDef.signature.outputArg.forEach(output => {\n      const [nodeName, index] = getNodeNameAndIndex(returnNodeMap[output.name]);\n      const node = nodes[nodeName];\n      if (node != null) {\n        node.defaultOutput = index;\n        outputs.push(node);\n      }\n    });\n\n    const signature = this.mapArgsToSignature(functionDef);\n    return {nodes, inputs, outputs, weights, placeholders, signature};\n  }\n\n  private mapArgsToSignature(functionDef: tensorflow.IFunctionDef):\n      tensorflow.ISignatureDef {\n    return {\n      methodName: functionDef.signature.name,\n      inputs: functionDef.signature.inputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n      outputs: functionDef.signature.outputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg, functionDef.ret);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n    };\n  }\n\n  private mapArgToTensorInfo(\n      arg: tensorflow.OpDef.IArgDef,\n      nameMap?: {[key: string]: string}): tensorflow.ITensorInfo {\n    let name = arg.name;\n    if (nameMap != null) {\n      name = nameMap[name];\n    }\n    return {name, dtype: arg.type};\n  }\n}\n\nexport function decodeBase64(text: string): string {\n  const global = env().global;\n  if (typeof global.atob !== 'undefined') {\n    return global.atob(text);\n  } else if (typeof Buffer !== 'undefined') {\n    return new Buffer(text, 'base64').toString();\n  } else {\n    throw new Error(\n        'Unable to decode base64 in this environment. ' +\n        'Missing built-in atob() or Buffer()');\n  }\n}\n\nexport function parseStringParam(s: []|string, keepCase: boolean): string {\n  const value =\n      Array.isArray(s) ? String.fromCharCode.apply(null, s) : decodeBase64(s);\n  return keepCase ? value : value.toLowerCase();\n}\n\nexport function getStringParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string,\n    keepCase = false): string {\n  const param = attrs[name];\n  if (param != null) {\n    return parseStringParam(param.s, keepCase);\n  }\n  return def;\n}\n\nexport function getBoolParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean): boolean {\n  const param = attrs[name];\n  return param ? param.b : def;\n}\n\nexport function getNumberParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number): number {\n  const param = attrs[name] || {};\n  const value =\n      param['i'] != null ? param['i'] : (param['f'] != null ? param['f'] : def);\n  return (typeof value === 'number') ? value : parseInt(value, 10);\n}\n\nexport function parseDtypeParam(value: string|tensorflow.DataType): DataType {\n  if (typeof (value) === 'string') {\n    // tslint:disable-next-line:no-any\n    value = tensorflow.DataType[value as any];\n  }\n  switch (value) {\n    case tensorflow.DataType.DT_FLOAT:\n      return 'float32';\n    case tensorflow.DataType.DT_INT32:\n    case tensorflow.DataType.DT_INT64:\n    case tensorflow.DataType.DT_INT8:\n    case tensorflow.DataType.DT_UINT8:\n      return 'int32';\n    case tensorflow.DataType.DT_BOOL:\n      return 'bool';\n    case tensorflow.DataType.DT_DOUBLE:\n      return 'float32';\n    case tensorflow.DataType.DT_STRING:\n      return 'string';\n    default:\n      // Unknown dtype error will happen at runtime (instead of parse time),\n      // since these nodes might not be used by the actual subgraph execution.\n      return null;\n  }\n}\n\nexport function getFuncParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: string): string {\n  const param = attrs[name];\n  if (param && param.func) {\n    return param.func.name;\n  }\n  return def;\n}\n\nexport function getDtypeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType): DataType {\n  const param = attrs[name];\n  if (param && param.type) {\n    return parseDtypeParam(param.type);\n  }\n  return def;\n}\n\nexport function getDtypeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType[]): DataType[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.type) {\n    return param.list.type.map(v => parseDtypeParam(v));\n  }\n  return def;\n}\n\nexport function parseTensorShapeParam(shape: tensorflow.ITensorShape): number[]|\n    undefined {\n  if (shape.unknownRank) {\n    return undefined;\n  }\n  if (shape.dim != null) {\n    return shape.dim.map(\n        dim =>\n            (typeof dim.size === 'number') ? dim.size : parseInt(dim.size, 10));\n  }\n  return [];\n}\n\nexport function getTensorShapeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def?: number[]): number[]|undefined {\n  const param = attrs[name];\n  if (param && param.shape) {\n    return parseTensorShapeParam(param.shape);\n  }\n  return def;\n}\n\nexport function getNumericArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[]): number[] {\n  const param = attrs[name];\n  if (param) {\n    return ((param.list.f && param.list.f.length ? param.list.f :\n                                                   param.list.i) ||\n            [])\n        .map(v => (typeof v === 'number') ? v : parseInt(v, 10));\n  }\n  return def;\n}\n\nexport function getStringArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string[],\n    keepCase = false): string[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.s) {\n    return param.list.s.map((v) => {\n      return parseStringParam(v, keepCase);\n    });\n  }\n  return def;\n}\n\nexport function getTensorShapeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[][]): number[][] {\n  const param = attrs[name];\n  if (param && param.list && param.list.shape) {\n    return param.list.shape.map((v) => {\n      return parseTensorShapeParam(v);\n    });\n  }\n  return def;\n}\n\nexport function getBoolArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean[]): boolean[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.b) {\n    return param.list.b;\n  }\n  return def;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {getTensor} from '../executors/utils';\nimport {getBoolArrayParam, getBoolParam, getDtypeArrayParam, getDtypeParam, getNumberParam, getNumericArrayParam, getStringArrayParam, getStringParam, getTensorShapeArrayParam, getTensorShapeParam} from '../operation_mapper';\nimport {GraphNode, Node, ValueType} from '../types';\n\n/**\n * Helper class for lookup inputs and params for nodes in the model graph.\n */\nexport class NodeValueImpl implements GraphNode {\n  public readonly inputs: Tensor[] = [];\n  public readonly attrs: {[key: string]: ValueType} = {};\n  constructor(\n      private node: Node, private tensorMap: NamedTensorsMap,\n      private context: ExecutionContext) {\n    this.inputs = node.inputNames.map(name => this.getInput(name));\n    if (node.rawAttrs != null) {\n      this.attrs = Object.keys(node.rawAttrs)\n                       .reduce((attrs: {[key: string]: ValueType}, key) => {\n                         attrs[key] = this.getAttr(key);\n                         return attrs;\n                       }, {});\n    }\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getInput(name: string): Tensor {\n    return getTensor(name, this.tensorMap, this.context);\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getAttr(name: string, defaultValue?: ValueType): ValueType {\n    const value = this.node.rawAttrs[name];\n    if (value.tensor != null) {\n      return getTensor(name, this.tensorMap, this.context);\n    }\n    if (value.i != null || value.f != null) {\n      return getNumberParam(this.node.rawAttrs, name, defaultValue as number);\n    }\n    if (value.s != null) {\n      return getStringParam(this.node.rawAttrs, name, defaultValue as string);\n    }\n    if (value.b != null) {\n      return getBoolParam(this.node.rawAttrs, name, defaultValue as boolean);\n    }\n    if (value.shape != null) {\n      return getTensorShapeParam(\n          this.node.rawAttrs, name, defaultValue as number[]);\n    }\n    if (value.type != null) {\n      return getDtypeParam(this.node.rawAttrs, name, defaultValue as DataType);\n    }\n    if (value.list != null) {\n      if (value.list.i != null || value.list.f != null) {\n        return getNumericArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[]);\n      }\n      if (value.list.s != null) {\n        return getStringArrayParam(\n            this.node.rawAttrs, name, defaultValue as string[]);\n      }\n      if (value.list.shape != null) {\n        return getTensorShapeArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[][]);\n      }\n      if (value.list.b != null) {\n        return getBoolArrayParam(\n            this.node.rawAttrs, name, defaultValue as boolean[]);\n      }\n      if (value.list.type != null) {\n        return getDtypeArrayParam(\n            this.node.rawAttrs, name, defaultValue as DataType[]);\n      }\n    }\n\n    return defaultValue;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BiasAdd':\n        case 'AddV2':\n        case 'Add': {\n          return [tfOps.add(\n              (getParamValue('a', node, tensorMap, context) as Tensor),\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'AddN': {\n          return [tfOps.addN((\n              getParamValue('tensors', node, tensorMap, context) as Tensor[]))];\n        }\n        case 'FloorMod':\n        case 'Mod':\n          return [tfOps.mod(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'Mul':\n          return [tfOps.mul(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'RealDiv':\n        case 'Div': {\n          return [tfOps.div(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'DivNoNan': {\n          return [tfOps.divNoNan(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'FloorDiv': {\n          return [tfOps.floorDiv(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sub': {\n          return [tfOps.sub(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Minimum': {\n          return [tfOps.minimum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Maximum': {\n          return [tfOps.maximum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Pow': {\n          return [tfOps.pow(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'SquaredDifference': {\n          return [tfOps.squaredDifference(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'arithmetic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Abs':\n        case 'ComplexAbs':\n          return [tfOps.abs(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acos':\n          return [tfOps.acos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acosh':\n          return [tfOps.acosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asin':\n          return [tfOps.asin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asinh':\n          return [tfOps.asinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan':\n          return [tfOps.atan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan2':\n          return [tfOps.atan2(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('y', node, tensorMap, context) as Tensor)];\n        case 'Atanh':\n          return [tfOps.atanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Ceil':\n          return [tfOps.ceil(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Complex':\n          return [tfOps.complex(\n              getParamValue('real', node, tensorMap, context) as Tensor,\n              getParamValue('imag', node, tensorMap, context) as Tensor)];\n        case 'Cos':\n          return [tfOps.cos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Cosh':\n          return [tfOps.cosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Elu':\n          return [tfOps.elu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Erf':\n          return [tfOps.erf(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Exp':\n          return [tfOps.exp(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Expm1': {\n          return [tfOps.expm1(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Floor':\n          return [tfOps.floor(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log':\n          return [tfOps.log(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log1p': {\n          return [tfOps.log1p(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Imag':\n          return [tfOps.imag(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n\n        case 'Neg':\n          return [tfOps.neg(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Reciprocal': {\n          return [tfOps.reciprocal(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Real':\n          return [tfOps.real(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu':\n          return [tfOps.relu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Round': {\n          return [tfOps.round(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Selu':\n          return [tfOps.selu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sigmoid':\n          return [tfOps.sigmoid(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sin':\n          return [tfOps.sin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sign': {\n          return [tfOps.sign(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sinh': {\n          return [tfOps.sinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Softplus': {\n          return [tfOps.softplus(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sqrt': {\n          return [tfOps.sqrt(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Square': {\n          return [tfOps.square(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tanh': {\n          return [tfOps.tanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tan':\n          return [tfOps.tan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu6':\n        case 'ClipByValue':\n          return [tfOps.clipByValue(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('clipValueMin', node, tensorMap, context) as number,\n              getParamValue('clipValueMax', node, tensorMap, context) as\n                  number)];\n        case 'Rsqrt':\n          return [tfOps.rsqrt(\n              getTensor(node.inputNames[0], tensorMap, context))];\n        case 'Prod':\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('axes', node, tensorMap, context) as number[])];\n        case 'LeakyRelu':\n          return [tfOps.leakyRelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as number)];\n        case 'Prelu':\n          return [tfOps.prelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as Tensor)];\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'basic_math';\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n/**\n * This differs from util.assertShapesMatch in that it allows values of\n * negative one, an undefined size of a dimensinon, in a shape to match\n * anything.\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function assertShapesMatchAllowUndefinedSize(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  util.assert(\n      shapesEqualAllowUndefinedSize(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function shapesEqualAllowUndefinedSize(n1: number[], n2: number[]) {\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== -1 && n2[i] !== -1 && n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\nexport interface TensorWithState {\n  tensor?: Tensor;\n  written?: boolean;\n  read?: boolean;\n  cleared?: boolean;\n}\n/**\n * The TensorArray object keeps an array of Tensors.  It\n * allows reading from the array and writing to the array.\n */\nexport class TensorArray {\n  private tensors: TensorWithState[] = [];\n  private closed_ = false;\n  readonly idTensor: Tensor;\n  constructor(\n      readonly name: string, readonly dtype: DataType, private maxSize: number,\n      private elementShape: number[], readonly identicalElementShapes: boolean,\n      readonly dynamicSize: boolean, readonly clearAfterRead: boolean) {\n    this.idTensor = scalar(0);\n    keep(this.idTensor);\n  }\n\n  get id() {\n    return this.idTensor.id;\n  }\n\n  get closed() {\n    return this.closed_;\n  }\n\n  /**\n   * Dispose the tensors and idTensor and mark the TensoryArray as closed.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.tensor.id)) {\n        tensor.tensor.dispose();\n      }\n    });\n    this.tensors = [];\n    this.closed_ = true;\n    this.idTensor.dispose();\n  }\n\n  size(): number {\n    return this.tensors.length;\n  }\n\n  /**\n   * Read the value at location index in the TensorArray.\n   * @param index Number the index to read from.\n   */\n  read(index: number): Tensor {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || index >= this.size()) {\n      throw new Error(`Tried to read from index ${index}, but array size is: ${\n          this.size()}`);\n    }\n\n    const tensorWithState = this.tensors[index];\n    if (tensorWithState.cleared) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not read index ${\n              index} twice because it was cleared after a previous read ` +\n          `(perhaps try setting clear_after_read = false?).`);\n    }\n\n    if (this.clearAfterRead) {\n      tensorWithState.cleared = true;\n    }\n\n    tensorWithState.read = true;\n    return tensorWithState.tensor;\n  }\n\n  /**\n   * Helper method to read multiple tensors from the specified indices.\n   */\n  readMany(indices: number[]): Tensor[] {\n    return indices.map(index => this.read(index));\n  }\n\n  /**\n   * Write value into the index of the TensorArray.\n   * @param index number the index to write to.\n   * @param tensor\n   */\n  write(index: number, tensor: Tensor) {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || !this.dynamicSize && index >= this.maxSize) {\n      throw new Error(`Tried to write to index ${\n          index}, but array is not resizeable and size is: ${this.maxSize}`);\n    }\n\n    const t = this.tensors[index] || {};\n\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray ${\n          this.name}: Could not write to TensorArray index ${index},\n          because the value dtype is ${\n          tensor.dtype}, but TensorArray dtype is ${this.dtype}.`);\n    }\n\n    // Set the shape for the first time write to unknow shape tensor array\n    if (this.size() === 0 &&\n        (this.elementShape == null || this.elementShape.length === 0)) {\n      this.elementShape = tensor.shape;\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape,\n        `TensorArray ${this.name}: Could not write to TensorArray index ${\n            index}.`);\n\n    if (t.read) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been read.`);\n    }\n\n    if (t.written) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been written.`);\n    }\n\n    t.tensor = tensor;\n    keep(tensor);\n    t.written = true;\n\n    this.tensors[index] = t;\n  }\n\n  /**\n   * Helper method to write multiple tensors to the specified indices.\n   */\n  writeMany(indices: number[], tensors: Tensor[]) {\n    if (indices.length !== tensors.length) {\n      throw new Error(\n          `TensorArray ${this.name}: could not write multiple tensors,` +\n          `because the index size: ${\n              indices.length} is not the same as tensors size: ${\n              tensors.length}.`);\n    }\n\n    indices.forEach((i, index) => this.write(i, tensors[index]));\n  }\n\n  /**\n   * Return selected values in the TensorArray as a packed Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param [indices] number[] Optional. Taking values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size(). If not specified returns\n   *    all tensors in the original order.\n   * @param [dtype]\n   */\n  gather(indices?: number[], dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but gather requested dtype ${dtype}`);\n    }\n\n    if (!indices) {\n      indices = [];\n      for (let i = 0; i < this.size(); i++) {\n        indices.push(i);\n      }\n    } else {\n      indices = indices.slice(0, this.size());\n    }\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    // Read all the PersistentTensors into a vector to keep track of\n    // their memory.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape, 'TensorArray shape mismatch: ');\n\n    return stack(tensors, 0);\n  }\n\n  /**\n   * Return the values in the TensorArray as a concatenated Tensor.\n   */\n  concat(dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but concat requested dtype ${dtype}`);\n    }\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    const indices = [];\n    for (let i = 0; i < this.size(); i++) {\n      indices.push(i);\n    }\n    // Collect all the tensors from the tensors array.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape,\n        `TensorArray shape mismatch: tensor array shape (${\n            this.elementShape}) vs first tensor shape (${tensors[0].shape})`);\n\n    return concat(tensors, 0);\n  }\n\n  /**\n   * Scatter the values of a Tensor in specific indices of a TensorArray.\n   * @param indices nummber[] values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size().\n   * @param tensor Tensor input tensor.\n   */\n  scatter(indices: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n\n    if (indices.length !== tensor.shape[0]) {\n      throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n          indices.length} vs. ${tensor.shape[0]}`);\n    }\n\n    const maxIndex = Math.max(...indices);\n\n    if (!this.dynamicSize && maxIndex >= this.maxSize) {\n      throw new Error(\n          `Max index must be < array size (${maxIndex}  vs. ${this.maxSize})`);\n    }\n\n    this.writeMany(indices, unstack(tensor, 0));\n  }\n\n  /**\n   * Split the values of a Tensor into the TensorArray.\n   * @param length number[] with the lengths to use when splitting value along\n   *    its first dimension.\n   * @param tensor Tensor, the tensor to split.\n   */\n  split(length: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n    let totalLength = 0;\n    const cumulativeLengths = length.map(len => {\n      totalLength += len;\n      return totalLength;\n    });\n\n    if (totalLength !== tensor.shape[0]) {\n      throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n    }\n\n    if (!this.dynamicSize && length.length !== this.maxSize) {\n      throw new Error(\n          `TensorArray's size is not equal to the size of lengths (${\n              this.maxSize} vs. ${length.length}), ` +\n          'and the TensorArray is not marked as dynamically resizeable');\n    }\n\n    const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n    const tensors: Tensor[] = [];\n    tidy(() => {\n      tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n      for (let i = 0; i < length.length; ++i) {\n        const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n        const indices = [0, previousLength, 0];\n        const sizes = [1, length[i], elementPerRow];\n        tensors[i] = reshape(slice(tensor, indices, sizes), this.elementShape);\n      }\n      return tensors;\n    });\n    const indices = [];\n    for (let i = 0; i < length.length; i++) {\n      indices[i] = i;\n    }\n    this.writeMany(indices, tensors);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\n/**\n * TensorList stores a container of `tf.Tensor` objects, which are accessible\n * via tensors field.\n *\n * In order to get a copy of the underlying list, use the copy method:\n * ```\n *    TensorList b = a.copy();\n *    b.tensors().pushBack(t);  // This does not modify a.tensors().\n * ```\n *\n * Note that this is not a deep copy: the memory locations of the underlying\n * tensors will still point to the same locations of the corresponding tensors\n * in the original.\n */\n\nexport class TensorList {\n  readonly idTensor: Tensor;\n  maxNumElements: number;\n\n  get id() {\n    return this.idTensor.id;\n  }\n  /**\n   *\n   * @param tensors list of tensors\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param maxNumElements The maximum allowed size of `tensors`. Defaults to -1\n   *   meaning that the size of `tensors` is unbounded.\n   */\n  constructor(\n      readonly tensors: Tensor[], readonly elementShape: number[],\n      readonly elementDtype: DataType, maxNumElements = -1) {\n    if (tensors != null) {\n      tensors.forEach(tensor => {\n        if (elementDtype !== tensor.dtype) {\n          throw new Error(`Invalid data types; op elements ${\n              elementDtype}, but list elements ${tensor.dtype}`);\n        }\n        assertShapesMatchAllowUndefinedSize(\n            elementShape, tensor.shape, 'TensorList shape mismatch: ');\n\n        keep(tensor);\n      });\n    }\n    this.idTensor = scalar(0);\n    this.maxNumElements = maxNumElements;\n    keep(this.idTensor);\n  }\n\n  /**\n   * Get a new TensorList containing a copy of the underlying tensor container.\n   */\n  copy(): TensorList {\n    return new TensorList(\n        [...this.tensors], this.elementShape, this.elementDtype);\n  }\n\n  /**\n   * Dispose the tensors and idTensor and clear the tensor list.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.id)) {\n        tensor.dispose();\n      }\n    });\n    this.tensors.length = 0;\n    this.idTensor.dispose();\n  }\n  /**\n   * The size of the tensors in the tensor list.\n   */\n  size() {\n    return this.tensors.length;\n  }\n\n  /**\n   * Return a tensor that stacks a list of rank-R tf.Tensors into one rank-(R+1)\n   * tf.Tensor.\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param numElements the number of elements to stack\n   */\n  stack(elementShape: number[], elementDtype: DataType, numElements = -1):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (numElements !== -1 && this.tensors.length !== numElements) {\n      throw new Error(`Operation expected a list with ${\n          numElements} elements but got a list with ${\n          this.tensors.length} elements.`);\n    }\n    assertShapesMatchAllowUndefinedSize(\n        elementShape, this.elementShape, 'TensorList shape mismatch: ');\n    return tidy(() => {\n      const reshapedTensors =\n          this.tensors.map(tensor => reshape(tensor, elementShape));\n      return stack(reshapedTensors, 0);\n    });\n  }\n\n  /**\n   * Pop a tensor from the end of the list.\n   * @param elementShape shape of the tensor\n   * @param elementDtype data type of the tensor\n   */\n  popBack(elementShape: number[], elementDtype: DataType): Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (this.size() === 0) {\n      throw new Error('Trying to pop from an empty list.');\n    }\n\n    const tensor = this.tensors.pop();\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, elementShape, 'TensorList shape mismatch: ');\n    return reshape(tensor, elementShape);\n  }\n\n  /**\n   * Push a tensor to the end of the list.\n   * @param tensor Tensor to be pushed.\n   */\n  pushBack(tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, this.elementShape, 'TensorList shape mismatch: ');\n\n    if (this.maxNumElements === this.size()) {\n      throw new Error(`Trying to push element into a full list.`);\n    }\n    keep(tensor);\n    this.tensors.push(tensor);\n  }\n\n  /**\n   * Update the size of the list.\n   * @param size the new size of the list.\n   */\n  resize(size: number) {\n    if (size < 0) {\n      throw new Error(\n          `TensorListResize expects size to be non-negative. Got: ${size}`);\n    }\n\n    if (this.maxNumElements !== -1 && size > this.maxNumElements) {\n      throw new Error(`TensorListResize input size ${\n          size} is greater maxNumElement ${this.maxNumElements}.`);\n    }\n    this.tensors.length = size;\n  }\n\n  /**\n   * Retrieve the element at the provided index\n   * @param elementShape shape of the tensor\n   * @param elementDtype dtype of the tensor\n   * @param elementIndex index of the tensor\n   */\n  getItem(elementIndex: number, elementShape: number[], elementDtype: DataType):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (elementIndex < 0 || elementIndex > this.tensors.length) {\n      throw new Error(`Trying to access element ${\n          elementIndex} in a list with ${this.tensors.length} elements.`);\n    }\n\n    if (this.tensors[elementIndex] == null) {\n      throw new Error(`element at index ${elementIndex} is null.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.tensors[elementIndex].shape, elementShape,\n        'TensorList shape mismatch: ');\n\n    return this.tensors[elementIndex];\n  }\n\n  /**\n   * Set the tensor at the index\n   * @param elementIndex index of the tensor\n   * @param tensor the tensor to be inserted into the list\n   */\n  setItem(elementIndex: number, tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (elementIndex < 0 ||\n        this.maxNumElements !== -1 && elementIndex >= this.maxNumElements) {\n      throw new Error(`Trying to set element ${\n          elementIndex} in a list with max ${this.maxNumElements} elements.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape, 'TensorList shape mismatch: ');\n    keep(tensor);\n    this.tensors[elementIndex] = tensor;\n  }\n\n  /**\n   * Return selected values in the TensorList as a stacked Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param indices indices of tensors to gather\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  gather(indices: number[], elementDtype: DataType, elementShape: number[]):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    // When indices is greater than the size of the list, indices beyond the\n    // size of the list are ignored.\n    indices = indices.slice(0, this.size());\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = indices.map(i => reshape(this.tensors[i], elementShape));\n      return stack(tensors, 0);\n    });\n  }\n\n  /**\n   * Return the values in the TensorList as a concatenated Tensor.\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  concat(elementDtype: DataType, elementShape: number[]): Tensor {\n    if (!!elementDtype && elementDtype !== this.elementDtype) {\n      throw new Error(`TensorList dtype is ${\n          this.elementDtype} but concat requested dtype ${elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = this.tensors.map(t => reshape(t, elementShape));\n      return concat(tensors, 0);\n    });\n  }\n}\n\n/**\n * Creates a TensorList which, when stacked, has the value of tensor.\n * @param tensor from tensor\n * @param elementShape output tensor element shape\n */\nexport function fromTensor(\n    tensor: Tensor, elementShape: number[], elementDtype: DataType) {\n  const dtype = tensor.dtype;\n  if (tensor.shape.length < 1) {\n    throw new Error(\n        `Tensor must be at least a vector, but saw shape: ${tensor.shape}`);\n  }\n  if (tensor.dtype !== elementDtype) {\n    throw new Error(`Invalid data types; op elements ${\n        tensor.dtype}, but list elements ${elementDtype}`);\n  }\n  const outputShape = tensor.shape.slice(1);\n  assertShapesMatchAllowUndefinedSize(\n      outputShape, elementShape, 'TensorList shape mismatch: ');\n\n  const tensorList: Tensor[] = unstack(tensor);\n  return new TensorList(tensorList, elementShape, dtype);\n}\n\n/**\n * Return a TensorList of the given size with empty elements.\n * @param elementShape the shape of the future elements of the list\n * @param elementDtype the desired type of elements in the list\n * @param numElements the number of elements to reserve\n */\nexport function reserve(\n    elementShape: number[], elementDtype: DataType, numElements: number) {\n  return new TensorList([], elementShape, elementDtype, numElements);\n}\n\n/**\n * Put tensors at specific indices of a stacked tensor into a TensorList.\n * @param indices list of indices on how to scatter the tensor.\n * @param tensor input tensor.\n * @param elementShape the shape of the future elements of the list\n * @param numElements the number of elements to scatter\n */\nexport function scatter(\n    tensor: Tensor, indices: number[], elementShape: number[],\n    numElements?: number): TensorList {\n  if (indices.length !== tensor.shape[0]) {\n    throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n        indices.length} vs. ${tensor.shape[0]}`);\n  }\n\n  const maxIndex = Math.max(...indices);\n\n  if (numElements != null && numElements !== -1 && maxIndex >= numElements) {\n    throw new Error(\n        `Max index must be < array size (${maxIndex}  vs. ${numElements})`);\n  }\n\n  const list = new TensorList([], elementShape, tensor.dtype, numElements);\n  const tensors = unstack(tensor, 0);\n  indices.forEach((value, index) => {\n    list.setItem(value, tensors[index]);\n  });\n  return list;\n}\n\n/**\n * Split the values of a Tensor into a TensorList.\n * @param length the lengths to use when splitting value along\n *    its first dimension.\n * @param tensor the tensor to split.\n * @param elementShape the shape of the future elements of the list\n */\nexport function split(\n    tensor: Tensor, length: number[], elementShape: number[]) {\n  let totalLength = 0;\n  const cumulativeLengths = length.map(len => {\n    totalLength += len;\n    return totalLength;\n  });\n\n  if (totalLength !== tensor.shape[0]) {\n    throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n  }\n\n  const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n  const tensors: Tensor[] = tidy(() => {\n    const tensors = [];\n    tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n    for (let i = 0; i < length.length; ++i) {\n      const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n      const indices = [0, previousLength, 0];\n      const sizes = [1, length[i], elementPerRow];\n      tensors[i] = reshape(slice(tensor, indices, sizes), elementShape);\n    }\n    tensor.dispose();\n    return tensors;\n  });\n\n  const list = new TensorList([], elementShape, tensor.dtype, length.length);\n\n  for (let i = 0; i < tensors.length; i++) {\n    list.setItem(i, tensors[i]);\n  }\n  return list;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, scalar, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {TensorArray} from '../../executor/tensor_array';\nimport {fromTensor, reserve, scatter, split} from '../../executor/tensor_list';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'If':\n    case 'StatelessIf': {\n      const thenFunc =\n          getParamValue('thenBranch', node, tensorMap, context) as string;\n      const elseFunc =\n          getParamValue('elseBranch', node, tensorMap, context) as string;\n      const cond = getParamValue('cond', node, tensorMap, context) as Tensor;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n      const condValue = await cond.data();\n      if (condValue[0]) {\n        return context.functionMap[thenFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      } else {\n        return context.functionMap[elseFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      }\n    }\n    case 'While':\n    case 'StatelessWhile': {\n      const bodyFunc =\n          getParamValue('body', node, tensorMap, context) as string;\n      const condFunc =\n          getParamValue('cond', node, tensorMap, context) as string;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n\n      // Calculate the condition of the loop\n      const condResult =\n          (await context.functionMap[condFunc].executeFunctionAsync(\n              args, context.tensorArrayMap, context.tensorListMap));\n      const argIds = args.map(tensor => tensor.id);\n      let condValue = await condResult[0].data();\n      // Dispose the intermediate tensors for condition function\n      condResult.forEach(tensor => {\n        if (!tensor.kept && argIds.indexOf(tensor.id) === -1) {\n          tensor.dispose();\n        }\n      });\n\n      let result: Tensor[] = args;\n\n      while (condValue[0]) {\n        // Record the previous result for intermediate tensor tracking\n        const origResult = result;\n        // Execution the body of the loop\n        result = await context.functionMap[bodyFunc].executeFunctionAsync(\n            result, context.tensorArrayMap, context.tensorListMap);\n        const resultIds = result.map(tensor => tensor.id);\n\n        // Dispose the intermediate tensor for body function that is not global\n        // kept, not input/output of the body function\n        origResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n\n        // Recalcuate the condition of the loop using the latest results.\n        const condResult =\n            (await context.functionMap[condFunc].executeFunctionAsync(\n                result, context.tensorArrayMap, context.tensorListMap));\n        condValue = await condResult[0].data();\n        // Dispose the intermediate tensors for condition function\n        condResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n      }\n      return result;\n    }\n    case 'LoopCond': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      return [cloneTensor(pred)];\n    }\n    case 'Switch': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      let data = getParamValue('data', node, tensorMap, context) as Tensor;\n      if (!data.kept) {\n        data = cloneTensor(data);\n      }\n      // Outputs nodes :0 => false, :1 => true\n      return (await pred.data())[0] ? [undefined, data] : [data, undefined];\n    }\n    case 'Merge': {\n      const inputName = node.inputNames.find(\n          name => getTensor(name, tensorMap, context) !== undefined);\n      if (inputName) {\n        const data = getTensor(inputName, tensorMap, context);\n        return [cloneTensor(data)];\n      }\n      return undefined;\n    }\n    case 'Enter': {\n      const frameId =\n          getParamValue('frameName', node, tensorMap, context) as string;\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.enterFrame(frameId);\n      return [cloneTensor(data)];\n    }\n    case 'Exit': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.exitFrame();\n      return [cloneTensor(data)];\n    }\n    case 'NextIteration': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.nextIteration();\n      return [cloneTensor(data)];\n    }\n    case 'TensorArrayV3': {\n      const size = getParamValue('size', node, tensorMap, context) as number;\n      const dtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const dynamicSize =\n          getParamValue('dynamicSize', node, tensorMap, context) as boolean;\n      const clearAfterRead =\n          getParamValue('clearAfterRead', node, tensorMap, context) as boolean;\n      const identicalElementShapes =\n          getParamValue('identicalElementShapes', node, tensorMap, context) as\n          boolean;\n      const name = getParamValue('name', node, tensorMap, context) as string;\n      const tensorArray = new TensorArray(\n          name, dtype, size, elementShape, identicalElementShapes, dynamicSize,\n          clearAfterRead);\n      context.addTensorArray(tensorArray);\n      return [tensorArray.idTensor, scalar(1.0)];\n    }\n    case 'TensorArrayWriteV3': {\n      const id =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const writeTensorArray = context.getTensorArray(id.id);\n      writeTensorArray.write(index, writeTensor);\n      return [writeTensorArray.idTensor];\n    }\n    case 'TensorArrayReadV3': {\n      const readId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const readTensorArray = context.getTensorArray(readId.id);\n      return [readTensorArray.read(readIndex)];\n    }\n    case 'TensorArrayGatherV3': {\n      const gatherId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const gatherDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const gatherTensorArray = context.getTensorArray(gatherId.id);\n      return [gatherTensorArray.gather(gatherIndices, gatherDtype)];\n    }\n    case 'TensorArrayScatterV3': {\n      const scatterId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const scatterTensorArray = context.getTensorArray(scatterId.id);\n      scatterTensorArray.scatter(scatterIndices, scatterTensor);\n      return [scatterTensorArray.idTensor];\n    }\n    case 'TensorArrayConcatV3': {\n      const concatId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const concatTensorArray = context.getTensorArray(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      return [concatTensorArray.concat(concatDtype)];\n    }\n    case 'TensorArraySplitV3': {\n      const splitId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n      const splitTensorArray = context.getTensorArray(splitId.id);\n      splitTensorArray.split(lengths, splitTensor);\n      return [splitTensorArray.idTensor];\n    }\n    case 'TensorArraySizeV3': {\n      const sizeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const sizeTensorArray = context.getTensorArray(sizeId.id);\n      return [scalar(sizeTensorArray.size(), 'int32')];\n    }\n    case 'TensorArrayCloseV3': {\n      const closeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const closeTensorArray = context.getTensorArray(closeId.id);\n      closeTensorArray.clearAndClose();\n      return [closeTensorArray.idTensor];\n    }\n    case 'TensorListSetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.setItem(index, writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.getItem(readIndex, elementShape, elementDType)];\n    }\n    case 'TensorListScatterV2':\n    case 'TensorListScatter': {\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList =\n          scatter(scatterTensor, scatterIndices, elementShape, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListReserve': {\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = reserve(elementShape, elementDtype, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGather': {\n      const gatherId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(gatherId.id);\n      return [tensorList.gather(gatherIndices, elementDtype, elementShape)];\n    }\n    case 'TensorListStack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.stack(elementShape, elementDtype, numElements)];\n    }\n    case 'TensorListFromTensor': {\n      const tensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = fromTensor(tensor, elementShape, elementDtype);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListConcat': {\n      const concatId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      return [tensorList.concat(concatDtype, elementShape)];\n    }\n    case 'TensorListPushBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.pushBack(writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListPopBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.popBack(elementShape, elementDType)];\n    }\n    case 'TensorListSplit': {\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n\n      const tensorList = split(splitTensor, lengths, elementShape);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'control';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rank, Tensor, Tensor3D, Tensor4D, Tensor5D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getPadding, getParamValue} from './utils';\n\nfunction fusedConvAndDepthWiseParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const [extraOp, activationFunc] =\n      (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n  const isBiasAdd = extraOp === 'biasadd';\n  const isPrelu = activationFunc === 'prelu';\n  const isBatchNorm = extraOp === 'fusedbatchnorm';\n\n  const numArgs =\n      (getParamValue('numArgs', node, tensorMap, context) as number);\n  if (isBiasAdd) {\n    if (isPrelu && numArgs !== 2) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu ' +\n          'must have two extra arguments: bias and alpha.');\n    }\n    if (!isPrelu && numArgs !== 1) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd must have ' +\n          'one extra argument: bias.');\n    }\n  }\n  if (isBatchNorm) {\n    throw new Error(\n        'FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.');\n  }\n  const stride = getParamValue('strides', node, tensorMap, context) as number[];\n  const pad = getPadding(node, tensorMap, context);\n  const dataFormat =\n      (getParamValue('dataFormat', node, tensorMap, context) as string)\n          .toUpperCase();\n  const dilations =\n      getParamValue('dilations', node, tensorMap, context) as number[];\n  const [biasArg, preluArg] =\n      getParamValue('args', node, tensorMap, context) as Tensor[];\n\n  return {\n    stride,\n    pad,\n    dataFormat,\n    dilations,\n    biasArg,\n    preluArg,\n    activationFunc\n  };\n}\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Conv1D': {\n          const stride =\n              getParamValue('stride', node, tensorMap, context) as number;\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilation =\n              getParamValue('dilation', node, tensorMap, context) as number;\n          return [tfOps.conv1d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              stride, pad as 'valid' | 'same', dataFormat as 'NWC' | 'NCW',\n              dilation)];\n        }\n        case 'Conv2D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case '_FusedConv2D': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.conv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n\n        case 'FusedDepthwiseConv2dNative': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.depthwiseConv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n        case 'Conv2DBackpropInput':\n        case 'Conv2dTranspose': {\n          const shape = getParamValue(\n                            'outputShape', node, tensorMap,\n                            context) as [number, number, number] |\n              [number, number, number, number];\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          return [tfOps.conv2dTranspose(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              shape, [stride[1], stride[2]], pad as 'valid' | 'same')];\n        }\n        case 'DepthwiseConv2dNative':\n        case 'DepthwiseConv2d': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n\n          return [tfOps.depthwiseConv2d(\n              getParamValue('input', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case 'Conv3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv3d(\n              getParamValue('x', node, tensorMap, context) as Tensor4D |\n                  Tensor<Rank.R5>,\n              getParamValue('filter', node, tensorMap, context) as\n                  Tensor<Rank.R5>,\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same',\n              dataFormat as 'NDHWC' | 'NCDHW',\n              [dilations[1], dilations[2], dilations[3]])];\n        }\n        case 'AvgPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPoolWithArgmax': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n          const includeBatchInIndex =\n              getParamValue('includeBatchInIndex', node, tensorMap, context) as\n              boolean;\n          const {result, indexes} = tfOps.maxPoolWithArgmax(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same', includeBatchInIndex);\n          return [result, indexes];\n        }\n        case 'AvgPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'MaxPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'Dilation2D': {\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n\n          // strides: [1, stride_height, stride_width, 1].\n          const strideHeight = strides[1];\n          const strideWidth = strides[2];\n\n          // dilations: [1, dilation_height, dilation_width, 1].\n          const dilationHeight = dilations[1];\n          const dilationWidth = dilations[2];\n\n          return [tfOps.dilation2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              [strideHeight, strideWidth], pad as 'valid' | 'same',\n              [dilationHeight, dilationWidth], 'NHWC' /* dataFormat */)];\n        }\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'convolution';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor, Tensor1D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Fill': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const dtype =\n              getParamValue('dtype', node, tensorMap, context) as DataType;\n          const value =\n              getParamValue('value', node, tensorMap, context) as number;\n          return [tfOps.fill(shape, value, dtype)];\n        }\n        case 'LinSpace': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const num = getParamValue('num', node, tensorMap, context) as number;\n          return [tfOps.linspace(start, stop, num)];\n        }\n        case 'Multinomial': {\n          const logits =\n              getParamValue('logits', node, tensorMap, context) as Tensor1D;\n          const numSamples =\n              getParamValue('numSamples', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.multinomial(logits, numSamples, seed)];\n        }\n        case 'OneHot': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          const depth =\n              getParamValue('depth', node, tensorMap, context) as number;\n          const onValue =\n              getParamValue('onValue', node, tensorMap, context) as number;\n          const offValue =\n              getParamValue('offValue', node, tensorMap, context) as number;\n          return [tfOps.oneHot(indices, depth, onValue, offValue)];\n        }\n        case 'Ones': {\n          return [tfOps.ones(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'OnesLike': {\n          return [tfOps.onesLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'RandomUniform': {\n          return [tfOps.randomUniform(\n              // tslint:disable-next-line:no-any\n              getParamValue('shape', node, tensorMap, context) as any,\n              getParamValue('minval', node, tensorMap, context) as number,\n              getParamValue('maxval', node, tensorMap, context) as number,\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'Range': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const step =\n              getParamValue('step', node, tensorMap, context) as number;\n          return [tfOps.range(\n              start, stop, step,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32')];\n        }\n        case 'TruncatedNormal': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const mean =\n              getParamValue('mean', node, tensorMap, context) as number;\n          const stdDev =\n              getParamValue('stdDev', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.truncatedNormal(\n              shape, mean, stdDev,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32',\n              seed)];\n        }\n        case 'Zeros': {\n          return [tfOps.zeros(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'ZerosLike': {\n          return [tfOps.zerosLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'creation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nfunction nmsParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const boxes = getParamValue('boxes', node, tensorMap, context) as Tensor;\n  const scores = getParamValue('scores', node, tensorMap, context) as Tensor;\n  const maxOutputSize =\n      getParamValue('maxOutputSize', node, tensorMap, context) as number;\n  const iouThreshold =\n      getParamValue('iouThreshold', node, tensorMap, context) as number;\n  const scoreThreshold =\n      getParamValue('scoreThreshold', node, tensorMap, context) as number;\n  const softNmsSigma =\n      getParamValue('softNmsSigma', node, tensorMap, context) as number;\n\n  return {\n    boxes,\n    scores,\n    maxOutputSize,\n    iouThreshold,\n    scoreThreshold,\n    softNmsSigma\n  };\n}\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'NonMaxSuppressionV5': {\n      const {\n        boxes,\n        scores,\n        maxOutputSize,\n        iouThreshold,\n        scoreThreshold,\n        softNmsSigma\n      } = nmsParams(node, tensorMap, context);\n\n      const result = await tfOps.image.nonMaxSuppressionWithScoreAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, softNmsSigma);\n\n      return [result.selectedIndices, result.selectedScores];\n    }\n    case 'NonMaxSuppressionV4': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      const padToMaxOutputSize =\n          getParamValue('padToMaxOutputSize', node, tensorMap, context) as\n          boolean;\n\n      const result = await tfOps.image.nonMaxSuppressionPaddedAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, padToMaxOutputSize);\n\n      return [result.selectedIndices, result.validOutputs];\n    }\n    case 'NonMaxSuppressionV3':\n    case 'NonMaxSuppressionV2': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      return [await tfOps.image.nonMaxSuppressionAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold)];\n    }\n    case 'Where': {\n      const condition = tfOps.cast(\n          (getParamValue('condition', node, tensorMap, context) as Tensor),\n          'bool');\n      const result = [await tfOps.whereAsync(condition)];\n      condition.dispose();\n      return result;\n    }\n    case 'ListDiff': {\n      return tfOps.setdiff1dAsync(\n          getParamValue('x', node, tensorMap, context) as Tensor,\n          getParamValue('y', node, tensorMap, context) as Tensor);\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'dynamic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'TopKV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const k = getParamValue('k', node, tensorMap, context) as number;\n              const sorted =\n                  getParamValue('sorted', node, tensorMap, context) as boolean;\n              const result = tfOps.topk(x, k, sorted);\n              return [result.values, result.indices];\n            }\n            case 'Unique': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const result = tfOps.unique(x);\n              return [result.values, result.indices];\n            }\n            case 'UniqueV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const axis =\n                  getParamValue('axis', node, tensorMap, context) as number;\n              const result = tfOps.unique(x, axis);\n              return [result.values, result.indices];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'evaluation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Const': {\n          return tensorMap[node.name];\n        }\n        case 'PlaceholderWithDefault':\n          const def =\n              getParamValue('default', node, tensorMap, context) as Tensor;\n          return [getTensor(node.name, tensorMap, context) || def];\n        case 'Placeholder':\n          return [getTensor(node.name, tensorMap, context)];\n        case 'Identity':\n        case 'StopGradient':\n        case 'FakeQuantWithMinMaxVars': {  // This op is currently ignored.\n          const data = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [cloneTensor(data)];\n        }\n        case 'IdentityN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => cloneTensor(t));\n        case 'Snapshot':\n          const snapshot =\n              (getParamValue('x', node, tensorMap, context) as Tensor);\n          return [cloneTensor(snapshot)];\n        case 'Shape':\n          return [tfOps.tensor1d(\n              (getParamValue('x', node, tensorMap, context) as Tensor).shape,\n              'int32')];\n        case 'ShapeN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => tfOps.tensor1d(t.shape));\n        case 'Size':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).size,\n              'int32')];\n        case 'Rank':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).rank,\n              'int32')];\n        case 'NoOp':\n          return [tfOps.scalar(1)];\n        case 'Print':\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const data =\n              getParamValue('data', node, tensorMap, context) as Tensor[];\n          const message =\n              getParamValue('message', node, tensorMap, context) as string;\n          const summarize =\n              getParamValue('summarize', node, tensorMap, context) as number;\n          console.warn(\n              'The graph has a tf.print() operation,' +\n              'usually used for debugging, which slows down performance.');\n          console.log(message);\n          for (let i = 0; i < data.length; i++) {\n            console.log(Array.prototype.slice.call(data[i].dataSync())\n                            .slice(0, summarize));\n          }\n          return [input];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'graph';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ResizeBilinear': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeBilinear(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'ResizeNearestNeighbor': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeNearestNeighbor(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'CropAndResize': {\n          const image =\n              getParamValue('image', node, tensorMap, context) as Tensor;\n          const boxes =\n              getParamValue('boxes', node, tensorMap, context) as Tensor;\n          const boxInd =\n              getParamValue('boxInd', node, tensorMap, context) as Tensor;\n          const cropSize =\n              getParamValue('cropSize', node, tensorMap, context) as number[];\n          const method =\n              getParamValue('method', node, tensorMap, context) as string;\n          const extrapolationValue =\n              getParamValue('extrapolationValue', node, tensorMap, context) as\n              number;\n          return [tfOps.image.cropAndResize(\n              image as Tensor4D, boxes as Tensor2D, boxInd as Tensor1D,\n              cropSize as [number, number], method as 'bilinear' | 'nearest',\n              extrapolationValue)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'image';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Equal': {\n          return [tfOps.equal(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'NotEqual': {\n          return [tfOps.notEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Greater': {\n          return [tfOps.greater(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'GreaterEqual': {\n          return [tfOps.greaterEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Less': {\n          return [tfOps.less(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LessEqual': {\n          return [tfOps.lessEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalAnd': {\n          return [tfOps.logicalAnd(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalNot': {\n          return [tfOps.logicalNot(\n              getParamValue('a', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalOr': {\n          return [tfOps.logicalOr(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Select':\n        case 'SelectV2': {\n          return [tfOps.where(\n              getParamValue('condition', node, tensorMap, context) as Tensor,\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'logical';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BatchMatMul':\n        case 'BatchMatMulV2':\n        case 'MatMul':\n          return [tfOps.matMul(\n              getParamValue('a', node, tensorMap, context) as Tensor2D,\n              getParamValue('b', node, tensorMap, context) as Tensor2D,\n              getParamValue('transposeA', node, tensorMap, context) as boolean,\n              getParamValue('transposeB', node, tensorMap, context) as\n                  boolean)];\n\n        case 'Transpose':\n          return [tfOps.transpose(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('perm', node, tensorMap, context) as number[])];\n\n        case '_FusedMatMul':\n          const [extraOp, activationFunc] =\n              (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n          const isBiasAdd = extraOp === 'biasadd';\n          const isPrelu = activationFunc === 'prelu';\n\n          const numArgs =\n              (getParamValue('numArgs', node, tensorMap, context) as number);\n          if (isBiasAdd) {\n            if (isPrelu && numArgs !== 2) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd and Prelu must have two ' +\n                  'extra arguments: bias and alpha.');\n            }\n            if (!isPrelu && numArgs !== 1) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd must have one extra argument: bias.');\n            }\n          }\n          const [biasArg, preluArg] =\n              getParamValue('args', node, tensorMap, context) as Tensor[];\n          return [tfOps.fused.matMul({\n            a: getParamValue('a', node, tensorMap, context) as Tensor2D,\n            b: getParamValue('b', node, tensorMap, context) as Tensor2D,\n            transposeA: getParamValue('transposeA', node, tensorMap, context) as\n                boolean,\n            transposeB: getParamValue('transposeB', node, tensorMap, context) as\n                boolean,\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'matrices';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'FusedBatchNorm':\n        case 'FusedBatchNormV2': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'FusedBatchNormV3': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'LRN': {\n          return [tfOps.localResponseNormalization(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('radius', node, tensorMap, context) as number,\n              getParamValue('bias', node, tensorMap, context) as number,\n              getParamValue('alpha', node, tensorMap, context) as number,\n              getParamValue('beta', node, tensorMap, context) as number)];\n        }\n        case 'Softmax': {\n          return [tfOps.softmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogSoftmax': {\n          return [tfOps.logSoftmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'SparseToDense': {\n          return [tfOps.sparseToDense(\n              getParamValue('sparseIndices', node, tensorMap, context) as\n                  Tensor,\n              getParamValue('outputShape', node, tensorMap, context) as Tensor,\n              getParamValue('sparseValues', node, tensorMap, context) as\n                  number[],\n              getParamValue('defaultValue', node, tensorMap, context) as\n                  Scalar)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'normalization';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Max': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.max(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Mean': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.mean(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Min': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.min(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Sum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.sum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'All': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.all(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Any': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.any(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'ArgMax': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMax(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'ArgMin': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMin(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Prod': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Cumsum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const exclusive =\n              getParamValue('exclusive', node, tensorMap, context) as boolean;\n          const reverse =\n              getParamValue('reverse', node, tensorMap, context) as boolean;\n          return [tfOps.cumsum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              exclusive, reverse)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'reduction';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ConcatV2':\n        case 'Concat': {\n          const n = getParamValue('n', node, tensorMap, context) as number;\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          let inputs =\n              getParamValue('tensors', node, tensorMap, context) as Tensor[];\n          inputs = inputs.slice(0, n);\n          return [tfOps.concat(inputs, axis)];\n        }\n        case 'GatherV2':\n        case 'Gather': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          return [tfOps.gather(input, tfOps.cast(indices, 'int32'), axis)];\n        }\n        case 'ReverseV2':\n        case 'Reverse': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [tfOps.reverse(input, axis)];\n        }\n        case 'Slice': {\n          // tslint:disable-next-line:no-any\n          const begin = getParamValue('begin', node, tensorMap, context) as any;\n          // tslint:disable-next-line:no-any\n          const size = getParamValue('size', node, tensorMap, context) as any;\n          return [tfOps.slice(\n              getParamValue('x', node, tensorMap, context) as Tensor, begin,\n              size)];\n        }\n        case 'StridedSlice': {\n          const begin =\n              getParamValue('begin', node, tensorMap, context) as number[];\n          const end =\n              getParamValue('end', node, tensorMap, context) as number[];\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const beginMask =\n              getParamValue('beginMask', node, tensorMap, context) as number;\n          const endMask =\n              getParamValue('endMask', node, tensorMap, context) as number;\n          const ellipsisMask =\n              getParamValue('ellipsisMask', node, tensorMap, context) as number;\n          const newAxisMask =\n              getParamValue('newAxisMask', node, tensorMap, context) as number;\n          const shrinkAxisMask =\n              getParamValue('shrinkAxisMask', node, tensorMap, context) as\n              number;\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return [tfOps.stridedSlice(\n              tensor, begin, end, strides, beginMask, endMask, ellipsisMask,\n              newAxisMask, shrinkAxisMask)];\n        }\n        case 'Pack': {\n          return tidy(() => {\n            const axis =\n                getParamValue('axis', node, tensorMap, context) as number;\n            const tensors =\n                getParamValue('tensors', node, tensorMap, context) as Tensor[];\n            // Reshape the tensors to the first tensor's shape if they don't\n            // match.\n            const shape = tensors[0].shape;\n            const squeezedShape = tfOps.squeeze(tensors[0]).shape;\n            const mapped = tensors.map(tensor => {\n              const sameShape = util.arraysEqual(tensor.shape, shape);\n              if (!sameShape &&\n                  !util.arraysEqual(\n                      tfOps.squeeze(tensor).shape, squeezedShape)) {\n                throw new Error('the input tensors shape does not match');\n              }\n              return sameShape ? tensor : tfOps.reshape(tensor, shape);\n            });\n            return [tfOps.stack(mapped, axis)];\n          });\n        }\n        case 'Unpack': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const tensor =\n              getParamValue('tensor', node, tensorMap, context) as Tensor;\n          return tfOps.unstack(tensor, axis);\n        }\n        case 'Tile': {\n          const reps =\n              getParamValue('reps', node, tensorMap, context) as number[];\n          return [tfOps.tile(\n              getParamValue('x', node, tensorMap, context) as Tensor, reps)];\n        }\n        case 'Split':\n        case 'SplitV': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const numOrSizeSplits =\n              getParamValue('numOrSizeSplits', node, tensorMap, context) as\n                  number |\n              number[];\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return tfOps.split(tensor, numOrSizeSplits, axis);\n        }\n        case 'ScatterNd': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          const values =\n              getParamValue('values', node, tensorMap, context) as Tensor;\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          return [tfOps.scatterND(indices, values, shape)];\n        }\n        case 'GatherNd': {\n          const x = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          return [tfOps.gatherND(x, indices)];\n        }\n        case 'SparseToDense': {\n          const indices =\n              getParamValue('sparseIndices', node, tensorMap, context) as\n              Tensor;\n          const shape =\n              getParamValue('outputShape', node, tensorMap, context) as\n              number[];\n          const sparseValues =\n              getParamValue('sparseValues', node, tensorMap, context) as Tensor;\n          const defaultValue =\n              getParamValue('defaultValue', node, tensorMap, context) as Scalar;\n          return [tfOps.sparseToDense(\n              indices, sparseValues, shape,\n              sparseValues.dtype === defaultValue.dtype ?\n                  defaultValue :\n                  tfOps.cast(defaultValue, sparseValues.dtype))];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'slice_join';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'FFT': {\n              return [tfOps.fft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IFFT': {\n              return [tfOps.ifft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'RFFT': {\n              return [tfOps.rfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IRFFT': {\n              return [tfOps.irfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'spectral';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Cast': {\n          return [tfOps.cast(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('dtype', node, tensorMap, context) as 'int32' |\n                  'float32' | 'bool')];\n        }\n        case 'ExpandDims': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.expandDims(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Squeeze': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          return [tfOps.squeeze(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n\n        case 'Reshape': {\n          return [tfOps.reshape(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        case 'PadV2':\n        case 'Pad': {\n          return [tfOps.pad(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('padding', node, tensorMap, context) as\n                  Array<[number, number]>,\n              getParamValue('constantValue', node, tensorMap, context) as\n                  number)];\n        }\n        case 'SpaceToBatchND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const paddings =\n              getParamValue('paddings', node, tensorMap, context) as number[][];\n          return [tfOps.spaceToBatchND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, paddings)];\n        }\n        case 'BatchToSpaceND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const crops =\n              getParamValue('crops', node, tensorMap, context) as number[][];\n          return [tfOps.batchToSpaceND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, crops)];\n        }\n        case 'DepthToSpace': {\n          const blockSize =\n              getParamValue('blockSize', node, tensorMap, context) as number;\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as\n               string).toUpperCase() as 'NHWC' |\n              'NCHW';\n          return [tfOps.depthToSpace(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              blockSize, dataFormat)];\n        }\n        case 'BroadcastTo': {\n          return [tfOps.broadcastTo(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'transformation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {ExecutionContext} from '../executor/execution_context';\n\nimport {NodeValueImpl} from './custom_op/node_value_impl';\nimport {getRegisteredOp} from './custom_op/register';\nimport * as arithmetic from './executors/arithmetic_executor';\nimport * as basicMath from './executors/basic_math_executor';\nimport * as control from './executors/control_executor';\nimport * as convolution from './executors/convolution_executor';\nimport * as creation from './executors/creation_executor';\nimport * as dynamic from './executors/dynamic_executor';\nimport * as evaluation from './executors/evaluation_executor';\nimport * as graph from './executors/graph_executor';\nimport * as image from './executors/image_executor';\nimport * as logical from './executors/logical_executor';\nimport * as matrices from './executors/matrices_executor';\nimport * as normalization from './executors/normalization_executor';\nimport * as reduction from './executors/reduction_executor';\nimport * as sliceJoin from './executors/slice_join_executor';\nimport * as spectral from './executors/spectral_executor';\nimport * as transformation from './executors/transformation_executor';\nimport {Node} from './types';\n\n/**\n * Executes the op defined by the node object.\n * @param node\n * @param tensorMap contains tensors for executed nodes and weights\n */\nexport function executeOp(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): tfc.Tensor[]|Promise<tfc.Tensor[]> {\n  const value =\n      ((node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) => {\n        switch (node.category) {\n          case 'arithmetic':\n            return tfc.tidy(\n                () => arithmetic.executeOp(node, tensorMap, context));\n          case 'basic_math':\n            return tfc.tidy(\n                () => basicMath.executeOp(node, tensorMap, context));\n          case 'control':\n            return control.executeOp(node, tensorMap, context);\n          case 'convolution':\n            return tfc.tidy(\n                () => convolution.executeOp(node, tensorMap, context));\n          case 'creation':\n            return tfc.tidy(() => creation.executeOp(node, tensorMap, context));\n          case 'dynamic':\n            return dynamic.executeOp(node, tensorMap, context);\n          case 'evaluation':\n            return tfc.tidy(\n                () => evaluation.executeOp(node, tensorMap, context));\n          case 'image':\n            return tfc.tidy(() => image.executeOp(node, tensorMap, context));\n          case 'graph':\n            return tfc.tidy(() => graph.executeOp(node, tensorMap, context));\n          case 'logical':\n            return tfc.tidy(() => logical.executeOp(node, tensorMap, context));\n          case 'matrices':\n            return tfc.tidy(() => matrices.executeOp(node, tensorMap, context));\n          case 'normalization':\n            return tfc.tidy(\n                () => normalization.executeOp(node, tensorMap, context));\n          case 'reduction':\n            return tfc.tidy(\n                () => reduction.executeOp(node, tensorMap, context));\n          case 'slice_join':\n            return tfc.tidy(\n                () => sliceJoin.executeOp(node, tensorMap, context));\n          case 'spectral':\n            return tfc.tidy(() => spectral.executeOp(node, tensorMap, context));\n          case 'transformation':\n            return tfc.tidy(\n                () => transformation.executeOp(node, tensorMap, context));\n          case 'custom':\n            const opMapper = getRegisteredOp(node.op);\n            if (opMapper && opMapper.customExecutor) {\n              return opMapper.customExecutor(\n                  new NodeValueImpl(node, tensorMap, context));\n            } else {\n              throw TypeError(`Custom op ${node.op} is not registered.`);\n            }\n          default:\n            throw TypeError(\n                `Unknown op '${node.op}'. File an issue at ` +\n                `https://github.com/tensorflow/tfjs/issues so we can add it` +\n                `, or register a custom execution with tf.registerOp()`);\n        }\n      })(node, tensorMap, context);\n  if (value instanceof Promise) {\n    return value.then((data) => [].concat(data));\n  }\n  return [].concat(value);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap, TensorArrayMap, TensorListMap} from '../data/types';\n\nimport {TensorArray} from './tensor_array';\nimport {TensorList} from './tensor_list';\nimport {FunctionExecutor} from './types';\n\nexport interface ExecutionContextInfo {\n  id: number;           // the unique id of the context info\n  frameName: string;    // The frame name of the loop, this comes from\n                        // the TensorFlow NodeDef.\n  iterationId: number;  // The iteration id of the loop\n}\n\n/**\n * ExecutionContext captures the runtime environment of the node. It keeps\n * track of the current frame and iteration for the control flow ops.\n *\n * For example, typical Dynamic RNN model may contain loops, for which\n * TensorFlow will generate graphs with Enter/Exit nodes to control the\n * current execution frame, and NextIteration Nodes for iteration id increment.\n * For model with branch logic, TensorFLow will generate Switch/Merge ops.\n */\nexport class ExecutionContext {\n  private rootContext = {id: 0, frameName: '', iterationId: 0};\n  private contexts: ExecutionContextInfo[] = [this.rootContext];\n  private lastId = 0;\n  private _currentContextIds: string[];\n\n  constructor(\n      readonly weightMap: NamedTensorsMap = {},\n      readonly tensorArrayMap: TensorArrayMap = {},\n      readonly tensorListMap: TensorListMap = {},\n      readonly functionMap: {[key: string]: FunctionExecutor} = {}) {\n    this.generateCurrentContextIds();\n  }\n\n  private newFrame(id: number, frameName: string) {\n    return {id, frameName, iterationId: 0};\n  }\n\n  /**\n   * Set the current context\n   * @param contexts: ExecutionContextInfo[] the current path of execution\n   * frames\n   */\n  set currentContext(contexts: ExecutionContextInfo[]) {\n    if (this.contexts !== contexts) {\n      this.contexts = contexts;\n      this.generateCurrentContextIds();\n    }\n  }\n\n  get currentContext(): ExecutionContextInfo[] {\n    return this.contexts;\n  }\n\n  /**\n   * Returns the current context in string format.\n   */\n  get currentContextId(): string {\n    return this._currentContextIds[0];\n  }\n\n  /**\n   * Returns the current context and all parent contexts in string format.\n   * This allow access to the nodes in the current and parent frames.\n   */\n  get currentContextIds(): string[] {\n    return this._currentContextIds;\n  }\n\n  private generateCurrentContextIds() {\n    const names = [];\n    for (let i = 0; i < this.contexts.length - 1; i++) {\n      const contexts = this.contexts.slice(0, this.contexts.length - i);\n      names.push(this.contextIdforContexts(contexts));\n    }\n    names.push('');\n    this._currentContextIds = names;\n  }\n\n  private contextIdforContexts(contexts: ExecutionContextInfo[]) {\n    return contexts ?\n        contexts\n            .map(\n                context => (context.id === 0 && context.iterationId === 0) ?\n                    '' :\n                    `${context.frameName}-${context.iterationId}`)\n            .join('/') :\n        '';\n  }\n\n  /**\n   * Enter a new frame, a new context is pushed on the current context list.\n   * @param frameId new frame id\n   */\n  enterFrame(frameId: string) {\n    if (this.contexts) {\n      this.lastId++;\n      this.contexts = this.contexts.slice();\n      this.contexts.push(this.newFrame(this.lastId, frameId));\n      this._currentContextIds.unshift(this.contextIdforContexts(this.contexts));\n    }\n  }\n\n  /**\n   * Exit the current frame, the last context is removed from the current\n   * context list.\n   */\n  exitFrame() {\n    if (this.contexts && this.contexts.length > 1) {\n      this.contexts = this.contexts.slice();\n      this.contexts.splice(-1);\n      this.currentContextIds.shift();\n    } else {\n      throw new Error('Cannot exit frame, the context is empty');\n    }\n  }\n\n  /**\n   * Enter the next iteration of a loop, the iteration id of last context is\n   * increased.\n   */\n  nextIteration() {\n    if (this.contexts && this.contexts.length > 0) {\n      this.contexts = this.contexts.slice();\n      this.lastId++;\n      const context =\n          Object.assign({}, this.contexts[this.contexts.length - 1]);\n      context.iterationId += 1;\n      context.id = this.lastId;\n      this.contexts.splice(-1, 1, context);\n      this._currentContextIds.splice(\n          0, 1, this.contextIdforContexts(this.contexts));\n    } else {\n      throw new Error('Cannot increase frame iteration, the context is empty');\n    }\n  }\n\n  getWeight(name: string): Tensor[] {\n    return this.weightMap[name];\n  }\n\n  addTensorArray(tensorArray: TensorArray) {\n    this.tensorArrayMap[tensorArray.id] = tensorArray;\n  }\n\n  getTensorArray(id: number): TensorArray {\n    return this.tensorArrayMap[id];\n  }\n\n  addTensorList(tensorList: TensorList) {\n    this.tensorListMap[tensorList.id] = tensorList;\n  }\n\n  getTensorList(id: number): TensorList {\n    return this.tensorListMap[id];\n  }\n\n  dispose(keepIds: Set<number>) {\n    for (const key in this.tensorArrayMap) {\n      this.tensorArrayMap[key].clearAndClose(keepIds);\n    }\n\n    for (const key in this.tensorListMap) {\n      this.tensorListMap[key].clearAndClose(keepIds);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NamedTensorMap} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {parseNodeName} from '../operations/executors/utils';\nimport {Graph, Node} from '../operations/types';\n\nexport interface ExecutionInfo {\n  inputs: NamedTensorMap;\n  outputs: Node[];\n  usedNodes: Set<string>;\n  missingInputs: string[];\n  dynamicNode: Node;\n  syncInputs: string[];\n}\n\n/**\n * Given graph inputs and desired outputs, find the minimal set of nodes\n * to execute in order to compute the outputs. In addition return other useful\n * info such:\n * - Missing inputs needed to compute the output.\n * - Whether the subgraph contains dynamic ops (control flow, dynamic shape).\n * - Alternative inputs in order to avoid async (dynamic op) execution.\n */\nexport function getExecutionSubgraph(\n    inputs: NamedTensorMap, outputs: Node[], weightMap: NamedTensorsMap,\n    initNodes?: Node[]): ExecutionInfo {\n  const usedNodes = new Set<string>();\n  const missingInputs: string[] = [];\n  let dynamicNode: Node = null;\n  let syncInputs: string[] = null;\n\n  // Start with the outputs, going backwards and find all the nodes that are\n  // needed to compute those outputs.\n  const seen = new Set<string>();\n  const inputNodeNames =\n      Object.keys(inputs).map(name => parseNodeName(name)[0]);\n\n  let initNodeNames: string[] = [];\n  if (initNodes != null) {\n    initNodeNames = initNodes.map(node => parseNodeName(node.name)[0]);\n  }\n\n  const frontier = [...outputs];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    if (isControlFlow(node) || isDynamicShape(node)) {\n      if (dynamicNode == null) {\n        dynamicNode = node;\n        syncInputs = dynamicNode.children.map(child => child.name)\n                         .filter(name => usedNodes.has(name));\n      }\n    }\n    usedNodes.add(node.name);\n\n    // Weights are dead end since we already have their values.\n    if (weightMap[node.name] != null) {\n      continue;\n    }\n    // This node is a dead end since it's one of the user-provided inputs.\n    if (inputNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    // This node is a dead end since it doesn't have any inputs.\n    if (initNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    if (node.inputs.length === 0) {\n      missingInputs.push(node.name);\n      continue;\n    }\n    node.inputs.forEach(input => {\n      // Don't add to the frontier if it is already there.\n      if (seen.has(input.name)) {\n        return;\n      }\n      seen.add(input.name);\n      frontier.push(input);\n    });\n  }\n  return {inputs, outputs, usedNodes, missingInputs, dynamicNode, syncInputs};\n}\n\n/**\n * Given the execution info, return a list of nodes in topological order that\n * need to be executed to compute the output.\n */\nexport function getNodesInTopologicalOrder(\n    graph: Graph, weightMap: NamedTensorsMap,\n    executionInfo: ExecutionInfo): Node[] {\n  const {usedNodes, inputs} = executionInfo;\n  const frontier: Node[] = [];\n  const inputNodes = Object.keys(inputs)\n                         .map(name => parseNodeName(name)[0])\n                         .map(name => graph.nodes[name]);\n  const initNodes = graph.initNodes;\n\n  inputNodes.forEach(input => {\n    if (usedNodes.has(input.name)) {\n      frontier.push(input);\n    }\n  });\n  graph.weights.forEach(weight => {\n    if (usedNodes.has(weight.name)) {\n      frontier.push(weight);\n    }\n  });\n  if (initNodes != null) {\n    initNodes.forEach(node => {\n      if (usedNodes.has(node.name)) {\n        frontier.push(node);\n      }\n    });\n  }\n  const seen = new Set<string>();\n  const orderedNodes: Node[] = [];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    seen.add(node.name);\n    if (!weightMap[node.name]) {\n      orderedNodes.push(node);\n    }\n    node.children.forEach(child => {\n      if (!seen.has(child.name) && usedNodes.has(child.name) &&\n          child.inputs.every(input => seen.has(input.name))) {\n        frontier.push(child);\n      }\n    });\n  }\n  return orderedNodes;\n}\n\nconst CONTROL_FLOW_OPS = [\n  'Switch', 'Merge', 'Enter', 'Exit', 'NextIteration', 'StatelessIf',\n  'StatelessWhile', 'if', 'While'\n];\nconst DYNAMIC_SHAPE_OPS = [\n  'NonMaxSuppressionV2', 'NonMaxSuppressionV3', 'NonMaxSuppressionV5', 'Where'\n];\n\nexport function isControlFlow(node: Node) {\n  return CONTROL_FLOW_OPS.indexOf(node.op) >= 0;\n}\n\nexport function isDynamicShape(node: Node) {\n  return DYNAMIC_SHAPE_OPS.indexOf(node.op) >= 0;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NamedTensorMap, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {ISignatureDef} from '../data/compiled_api';\nimport {NamedTensorsMap, TensorArrayMap, TensorInfo, TensorListMap} from '../data/types';\nimport {getNodeNameAndIndex, getParamValue, getTensor, getTensorsForCurrentContenxt, parseNodeName} from '../operations/executors/utils';\nimport {executeOp} from '../operations/operation_executor';\nimport {Graph, Node} from '../operations/types';\n\nimport {ExecutionContext, ExecutionContextInfo} from './execution_context';\nimport {getExecutionSubgraph, getNodesInTopologicalOrder, isControlFlow} from './model_analysis';\nimport {FunctionExecutor} from './types';\n\ninterface NodeWithContexts {\n  contexts: ExecutionContextInfo[];\n  node: Node;\n}\n\nexport class GraphExecutor implements FunctionExecutor {\n  private compiledMap: Map<string, Node[]> = new Map();\n  private _weightMap: NamedTensorsMap = {};\n  private _weightIds: number[];\n  private _signature: ISignatureDef;\n  private _inputs: Node[];\n  private _outputs: Node[];\n  private _initNodes: Node[];\n  private SEPERATOR = ',';\n  private _functions: {[key: string]: Graph} = {};\n  private _functionExecutorMap: {[key: string]: FunctionExecutor} = {};\n\n  get weightIds(): number[] {\n    return this.parent ? this.parent.weightIds : this._weightIds;\n  }\n\n  get functionExecutorMap(): {[key: string]: FunctionExecutor} {\n    return this.parent ? this.parent.functionExecutorMap :\n                         this._functionExecutorMap;\n  }\n\n  get weightMap(): NamedTensorsMap {\n    return this.parent ? this.parent.weightMap : this._weightMap;\n  }\n\n  set weightMap(weightMap: NamedTensorsMap) {\n    const weightIds = Object.keys(weightMap).map(\n        key => weightMap[key].map(tensor => tensor.id));\n    this._weightIds = [].concat(...weightIds);\n    this._weightMap = weightMap;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this._inputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get outputs(): TensorInfo[] {\n    return this._outputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get inputNodes(): string[] {\n    return this._inputs.map(node => node.signatureKey || node.name);\n  }\n\n  get outputNodes(): string[] {\n    return this._outputs.map((node) => {\n      const name = node.signatureKey || node.name;\n      return node.defaultOutput ? (`${name}:${node.defaultOutput}`) : name;\n    });\n  }\n\n  get functions(): {[key: string]: ISignatureDef} {\n    return Object.keys(this._functions).reduce((map, key) => {\n      map[key] = this._functions[key].signature;\n      return map;\n    }, {} as {[key: string]: ISignatureDef});\n  }\n\n  /**\n   *\n   * @param graph Graph the model or function graph to be executed.\n   * @param parent When building function exector you need to set the parent\n   * executor. Since the weights and function executor maps are set at parant\n   * level, that function executor can access the function maps and weight maps\n   * through the parent.\n   */\n  constructor(private graph: Graph, private parent?: GraphExecutor) {\n    this._outputs = graph.outputs;\n    this._inputs = graph.inputs;\n    this._initNodes = graph.initNodes;\n    this._signature = graph.signature;\n    this._functions = graph.functions;\n    // create sub-graph executors\n    if (graph.functions != null) {\n      Object.keys(graph.functions).forEach(name => {\n        this._functionExecutorMap[name] =\n            new GraphExecutor(graph.functions[name], this);\n      });\n    }\n  }\n\n  private getCompilationKey(inputs: Node[], outputs: Node[]): string {\n    const sortedInputs = inputs.map(node => node.name).sort();\n    const sortedOutputs = outputs.map(node => node.name).sort();\n    return sortedInputs.join(this.SEPERATOR) + '--' +\n        sortedOutputs.join(this.SEPERATOR);\n  }\n\n  /**\n   * Compiles the inference graph and returns the minimal set of nodes that are\n   * required for execution, in the correct execution order.\n   */\n  private compile(inputs: NamedTensorMap, outputs: Node[]): Node[] {\n    const executionInfo =\n        getExecutionSubgraph(inputs, outputs, this.weightMap, this._initNodes);\n    const {missingInputs, dynamicNode, syncInputs} = executionInfo;\n    if (dynamicNode != null) {\n      throw new Error(\n          `This execution contains the node '${dynamicNode.name}', which has ` +\n          `the dynamic op '${dynamicNode.op}'. Please use ` +\n          `model.executeAsync() instead. Alternatively, to avoid the ` +\n          `dynamic ops, specify the inputs [${syncInputs}]`);\n    }\n\n    if (missingInputs.length > 0) {\n      const outNames = outputs.map(n => n.name);\n      const inNames = Object.keys(inputs);\n      throw new Error(\n          `Cannot compute the outputs [${outNames}] from the provided inputs ` +\n          `[${inNames}]. Missing the following inputs: [${missingInputs}]`);\n    }\n\n    return getNodesInTopologicalOrder(\n        this.graph, this.weightMap, executionInfo);\n  }\n\n  /**\n   * Executes the inference for given input tensors.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  execute(inputs: NamedTensorMap, outputs: string[]): Tensor[] {\n    inputs = this.mapInputs(inputs);\n    const names = Object.keys(inputs).sort();\n    this.checkInputs(inputs);\n    this.checkInputShapeAndType(inputs);\n    outputs = this.mapOutputs(outputs);\n    this.checkOutputs(outputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputs.map(name => parseNodeName(name)[0]);\n    let outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n\n    // If no outputs are specified, then use the default outputs of the model.\n    if (outputNodes.length === 0) {\n      outputNodes = this._outputs;\n    }\n\n    const compilationKey = this.getCompilationKey(inputNodes, outputNodes);\n\n    // Do nothing if the compiled graph cache contains the input.\n    let orderedNodes = this.compiledMap.get(compilationKey);\n    if (orderedNodes == null) {\n      orderedNodes = this.compile(inputs, outputNodes);\n      this.compiledMap.set(compilationKey, orderedNodes);\n    }\n\n    const tensorArrayMap: TensorArrayMap = {};\n    const tensorListMap: TensorListMap = {};\n    return tidy(() => {\n      const context = new ExecutionContext(\n          this.weightMap, tensorArrayMap, tensorListMap,\n          this.functionExecutorMap);\n      const tensorsMap: NamedTensorsMap = {...this.weightMap};\n\n      Object.keys(inputs).forEach(name => {\n        const [nodeName, index] = parseNodeName(name);\n        const tensors: Tensor[] = [];\n        tensors[index] = inputs[name];\n        tensorsMap[nodeName] = tensors;\n      });\n\n      const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n      const intermediateTensorConsumerCount: {[key: number]: number} = {};\n      for (let i = 0; i < orderedNodes.length; i++) {\n        const node = orderedNodes[i];\n        if (!tensorsMap[node.name]) {\n          const tensors = executeOp(node, tensorsMap, context) as Tensor[];\n          if (tensors instanceof Promise) {\n            throw new Error(\n                `The execution of the op '${node.op}' returned a promise. ` +\n                `Please use model.executeAsync() instead.`);\n          }\n          tensorsMap[node.name] = tensors;\n          this.checkTensorForDisposal(\n              node.name, node, tensorsMap, context, tensorsToKeep,\n              outputNodeNames, intermediateTensorConsumerCount);\n        }\n      }\n      // dispose the context for the root executor\n      if (this.parent == null) {\n        context.dispose(tensorsToKeep);\n      }\n      return outputs.map(name => getTensor(name, tensorsMap, context));\n    });\n  }\n\n  private getFrozenTensorIds(tensorMap: NamedTensorsMap): Set<number> {\n    const ids = [].concat.apply(\n        [],\n        Object.keys(tensorMap)\n            .map(key => tensorMap[key])\n            .map(tensors => tensors.map(tensor => tensor.id)));\n    return new Set(ids);\n  }\n  private checkTensorForDisposal(\n      nodeName: string, node: Node, tensorMap: NamedTensorsMap,\n      context: ExecutionContext, tensorsToKeep: Set<number>,\n      outputNames: string[],\n      intermediateTensorConsumerCount: {[key: string]: number}) {\n    // Skip output nodes and any control flow nodes, since its dependency is\n    // tricky to track correctly.\n    if (node.category === 'control' || outputNames.indexOf(nodeName) !== -1) {\n      return;\n    }\n\n    tensorMap[nodeName].forEach(tensor => {\n      if (tensor != null) {\n        intermediateTensorConsumerCount[tensor.id] =\n            (intermediateTensorConsumerCount[tensor.id] || 0) +\n            node.children.length;\n      }\n    });\n    node.inputs.forEach(input => {\n      // Skip any control flow nodes, since its dependency is tricky to track\n      // correctly.\n      if (input.category !== 'control') {\n        const tensors =\n            getTensorsForCurrentContenxt(input.name, tensorMap, context);\n        if (tensors != null) {\n          tensors.forEach(tensor => {\n            if (tensor && !tensorsToKeep.has(tensor.id)) {\n              const count = intermediateTensorConsumerCount[tensor.id];\n              if (count === 1) {\n                tensor.dispose();\n                delete intermediateTensorConsumerCount[tensor.id];\n              } else if (count != null) {\n                // only intermediate nodes has count set, inputs and weights are\n                // not.\n                intermediateTensorConsumerCount[tensor.id]--;\n              }\n            }\n          });\n        }\n      }\n    });\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  async executeAsync(inputs: NamedTensorMap, outputs: string[]):\n      Promise<Tensor[]> {\n    return this._executeAsync(inputs, outputs);\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   * @param isFunctionExecution Flag for executing a function.\n   * @param tensorArrayMap Optional, global TensorArray map by id. Used for\n   * function execution.\n   * @param tensorArrayMap Optinal global TensorList map by id. Used for\n   * function execution.\n   */\n  private async _executeAsync(\n      inputs: NamedTensorMap, outputs: string[], isFunctionExecution = false,\n      tensorArrayMap: TensorArrayMap = {},\n      tensorListMap: TensorListMap = {}): Promise<Tensor[]> {\n    if (!isFunctionExecution) {\n      inputs = this.mapInputs(inputs);\n      this.checkInputs(inputs);\n      this.checkInputShapeAndType(inputs);\n      outputs = this.mapOutputs(outputs);\n      this.checkOutputs(outputs);\n    }\n\n    const context = new ExecutionContext(\n        this.weightMap, tensorArrayMap, tensorListMap,\n        this.functionExecutorMap);\n\n    // Graph with control flow op requires runtime evaluation of the execution\n    // order, while without control flow the execution order is pre-determined\n    // in the compile method.\n    const tensorMap = await this.executeWithControlFlow(\n        inputs, context, outputs, isFunctionExecution);\n    const results = outputs.map(name => getTensor(name, tensorMap, context));\n\n    // dispose all the intermediate tensors\n    const outputIds = results.map(t => t.id);\n    const inputIds = Object.keys(inputs).map(name => inputs[name].id);\n    const keepIds =\n        new Set<number>([...outputIds, ...inputIds, ...this.weightIds]);\n    Object.keys(tensorMap).forEach(key => {\n      const tensorArray = tensorMap[key];\n      tensorArray.forEach(tensor => {\n        if (tensor && !tensor.isDisposed && !keepIds.has(tensor.id)) {\n          tensor.dispose();\n        }\n      });\n    });\n    // dispose the context for the root executor\n    if (this.parent == null) {\n      context.dispose(keepIds);\n    }\n\n    return results;\n  }\n\n  async executeFunctionAsync(\n      inputs: Tensor[], tensorArrayMap: TensorArrayMap,\n      tensorListMap: TensorListMap): Promise<Tensor[]> {\n    const mappedInputs = inputs.reduce((map, tensor, index) => {\n      map[this.inputs[index].name] = tensor;\n      return map;\n    }, {} as NamedTensorMap);\n\n    return this._executeAsync(\n        mappedInputs, this.outputNodes, true, tensorArrayMap, tensorListMap);\n  }\n  /**\n   * When there are control flow nodes in the graph, the graph execution use\n   * ExecutionContext to keep track of the frames and loop iterators.\n   * @param inputs placeholder tensors for the graph.\n   * @param context the execution context object for current execution.\n   * @param isFunctionExecution Flag for executing a function.\n   */\n  private async executeWithControlFlow(\n      inputs: NamedTensorMap, context: ExecutionContext, outputNames: string[],\n      isFunctionExecution: boolean): Promise<NamedTensorsMap> {\n    const names = Object.keys(inputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputNames.map(name => parseNodeName(name)[0]);\n    const outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n    const {usedNodes, missingInputs, dynamicNode, syncInputs} =\n        getExecutionSubgraph(inputs, outputNodes, this.weightMap);\n\n    const stack: NodeWithContexts[] =\n        [...inputNodes, ...this.graph.weights].map(node => {\n          return {node, contexts: context.currentContext};\n        });\n    const tensorsMap: NamedTensorsMap = {...this.weightMap};\n    Object.keys(inputs).forEach(name => {\n      const [nodeName, index] = parseNodeName(name);\n      const tensors: Tensor[] = [];\n      tensors[index] = inputs[name];\n      tensorsMap[nodeName] = tensors;\n    });\n    const intermediateTensorConsumerCount: {[key: number]: number} = {};\n    const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n    const added: {[key: string]: boolean} = {};\n    while (stack.length > 0) {\n      const promises = this.processStack(\n          inputNodes, stack, context, tensorsMap, added, tensorsToKeep,\n          outputNodeNames, intermediateTensorConsumerCount, usedNodes);\n      await Promise.all(promises);\n    }\n    if (dynamicNode == null && !isFunctionExecution) {\n      console.warn(\n          `This model execution did not contain any nodes with control flow ` +\n          `or dynamic output shapes. You can use model.execute() instead.`);\n    }\n    const missingOutputs =\n        outputNodes\n            .filter(\n                node => !isControlFlow(node) &&\n                    !getTensor(node.name, tensorsMap, context))\n            .map(node => node.name);\n    if (missingOutputs.length > 0) {\n      let alternativeMsg = '';\n      if (dynamicNode != null) {\n        alternativeMsg =\n            `Alternatively, to avoid the dynamic ops, use model.execute() ` +\n            `and specify the inputs [${syncInputs}]`;\n      }\n      throw new Error(\n          `Cannot compute the outputs [${missingOutputs}] from the provided ` +\n          `inputs [${names}]. Consider providing the following inputs: ` +\n          `[${missingInputs}]. ${alternativeMsg}`);\n    }\n    return tensorsMap;\n  }\n\n  private processStack(\n      inputNodes: Node[], stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      tensorsToKeep: Set<number>, outputNames: string[],\n      intermediateTensorConsumerCount: {[key: number]: number},\n      usedNodes: Set<string>) {\n    const promises: Array<Promise<Tensor[]>> = [];\n    while (stack.length > 0) {\n      const item = stack.pop();\n      context.currentContext = item.contexts;\n      let nodeName = '';\n      // The tensor of the Enter op with isConstant set should be set\n      // in the parent scope, so it will be available as constant for the\n      // whole loop.\n      if (item.node.op === 'Enter' &&\n          getParamValue('isConstant', item.node, tensorMap, context)) {\n        [nodeName] = getNodeNameAndIndex(item.node.name, context);\n      }\n\n      // only process nodes that are not provided as input nodes.\n      if (inputNodes.indexOf(item.node) === -1) {\n        const tensors = executeOp(item.node, tensorMap, context);\n        if (!nodeName) {\n          [nodeName] = getNodeNameAndIndex(item.node.name, context);\n        }\n        const currentContext = context.currentContext;\n        if (tensors instanceof Promise) {\n          promises.push(tensors.then(t => {\n            tensorMap[nodeName] = t;\n            context.currentContext = currentContext;\n            this.checkTensorForDisposal(\n                nodeName, item.node, tensorMap, context, tensorsToKeep,\n                outputNames, intermediateTensorConsumerCount);\n            this.processChildNodes(\n                item.node, stack, context, tensorMap, added, usedNodes);\n            return t;\n          }));\n        } else {\n          tensorMap[nodeName] = tensors;\n          this.checkTensorForDisposal(\n              nodeName, item.node, tensorMap, context, tensorsToKeep,\n              outputNames, intermediateTensorConsumerCount);\n          this.processChildNodes(\n              item.node, stack, context, tensorMap, added, usedNodes);\n        }\n      } else {\n        this.processChildNodes(\n            item.node, stack, context, tensorMap, added, usedNodes);\n      }\n    }\n    return promises;\n  }\n\n  private processChildNodes(\n      node: Node, stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      usedNodes: Set<string>) {\n    node.children.forEach((childNode) => {\n      const [nodeName, ] = getNodeNameAndIndex(childNode.name, context);\n      if (added[nodeName] || !usedNodes.has(childNode.name)) {\n        return;\n      }\n      // Merge op can be pushed if any of its inputs has value.\n      if (childNode.op === 'Merge') {\n        if (childNode.inputNames.some(name => {\n              return !!getTensor(name, tensorMap, context);\n            })) {\n          added[nodeName] = true;\n          stack.push({contexts: context.currentContext, node: childNode});\n        }\n      } else  // Otherwise all inputs must to have value.\n          if (childNode.inputNames.every(name => {\n                return !!getTensor(name, tensorMap, context);\n              })) {\n        added[nodeName] = true;\n        stack.push({contexts: context.currentContext, node: childNode});\n      }\n    });\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   */\n  dispose() {\n    Object.keys(this.weightMap)\n        .forEach(\n            key => this.weightMap[key].forEach(tensor => tensor.dispose()));\n  }\n\n  private checkInputShapeAndType(inputs: NamedTensorMap) {\n    Object.keys(inputs).forEach(name => {\n      const input = inputs[name];\n      const [nodeName, ] = parseNodeName(name);\n      const node = this.graph.nodes[nodeName];\n      if (node.attrParams['shape'] && node.attrParams['shape'].value) {\n        const shape = node.attrParams['shape'].value as number[];\n        const match = shape.length === input.shape.length &&\n            input.shape.every(\n                (dim, index) => shape[index] === -1 || shape[index] === dim);\n        util.assert(\n            match,\n            () => `The shape of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be [${shape}], but was ` +\n                `[${input.shape}]`);\n      }\n      if (node.attrParams['dtype'] && node.attrParams['dtype'].value) {\n        util.assert(\n            input.dtype === node.attrParams['dtype'].value as string,\n            () => `The dtype of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be ` +\n                `${node.attrParams['dtype'].value}, but was ${input.dtype}`);\n      }\n    });\n  }\n\n  private mapInputs(inputs: NamedTensorMap) {\n    const result: NamedTensorMap = {};\n    for (const inputName in inputs) {\n      if (this._signature != null && this._signature.inputs != null &&\n          this._signature.inputs[inputName] != null) {\n        const tensor = this._signature.inputs[inputName];\n        result[tensor.name] = inputs[inputName];\n      } else {\n        result[inputName] = inputs[inputName];\n      }\n    }\n    return result;\n  }\n\n  private checkInputs(inputs: NamedTensorMap) {\n    const notInGraph = Object.keys(inputs).filter(name => {\n      const [nodeName] = parseNodeName(name);\n      return this.graph.nodes[nodeName] == null;\n    });\n    if (notInGraph.length > 0) {\n      throw new Error(\n          `The dict provided in model.execute(dict) has ` +\n          `keys: [${notInGraph}] that are not part of graph`);\n    }\n  }\n\n  private mapOutputs(outputs: string[]) {\n    return outputs.map(name => {\n      if (this._signature != null && this._signature.outputs != null &&\n          this._signature.outputs[name] != null) {\n        const tensor = this._signature.outputs[name];\n        return tensor.name;\n      }\n      return name;\n    }, {});\n  }\n\n  private checkOutputs(outputs: string[]): void {\n    outputs.forEach(name => {\n      const [normalizedName] = parseNodeName(name);\n      if (!this.graph.nodes[normalizedName]) {\n        throw new Error(`The output '${name}' is not found in the graph`);\n      }\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {InferenceModel, io, ModelPredictConfig, NamedTensorMap, Tensor} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\nimport {NamedTensorsMap, TensorInfo} from '../data/types';\nimport {OperationMapper} from '../operations/operation_mapper';\n\nimport {GraphExecutor} from './graph_executor';\n\nexport const TFHUB_SEARCH_PARAM = '?tfjs-format=file';\nexport const DEFAULT_MODEL_NAME = 'model.json';\n/**\n * A `tf.GraphModel` is a directed, acyclic graph built from a\n * SavedModel GraphDef and allows inference execution.\n *\n * A `tf.GraphModel` can only be created by loading from a model converted from\n * a [TensorFlow SavedModel](https://www.tensorflow.org/guide/saved_model) using\n * the command line converter tool and loaded via `tf.loadGraphModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class GraphModel implements InferenceModel {\n  private executor: GraphExecutor;\n  private version = 'n/a';\n  private handler: io.IOHandler;\n  private artifacts: io.ModelArtifacts;\n  private initializer: GraphExecutor;\n\n  // Returns the version information for the tensorflow model GraphDef.\n  get modelVersion(): string {\n    return this.version;\n  }\n\n  get inputNodes(): string[] {\n    return this.executor.inputNodes;\n  }\n\n  get outputNodes(): string[] {\n    return this.executor.outputNodes;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this.executor.inputs;\n  }\n\n  get outputs(): TensorInfo[] {\n    return this.executor.outputs;\n  }\n\n  get weights(): NamedTensorsMap {\n    return this.executor.weightMap;\n  }\n\n  /**\n   * @param modelUrl url for the model, or an `io.IOHandler`.\n   * @param weightManifestUrl url for the weight file generated by\n   * scripts/convert.py script.\n   * @param requestOption options for Request, which allows to send credentials\n   * and custom headers.\n   * @param onProgress Optional, progress callback function, fired periodically\n   * before the load is completed.\n   */\n  constructor(\n      private modelUrl: string|io.IOHandler,\n      private loadOptions: io.LoadOptions = {}) {\n    if (loadOptions == null) {\n      this.loadOptions = {};\n    }\n  }\n\n  private findIOHandler() {\n    const path = this.modelUrl;\n    if ((path as io.IOHandler).load != null) {\n      // Path is an IO Handler.\n      this.handler = path as io.IOHandler;\n    } else if (this.loadOptions.requestInit != null) {\n      this.handler = io.browserHTTPRequest(path as string, this.loadOptions);\n    } else {\n      const handlers = io.getLoadHandlers(path as string, this.loadOptions);\n      if (handlers.length === 0) {\n        // For backward compatibility: if no load handler can be found,\n        // assume it is a relative http path.\n        handlers.push(io.browserHTTPRequest(path as string, this.loadOptions));\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) load handlers for ` +\n            `URL '${[path]}'`);\n      }\n      this.handler = handlers[0];\n    }\n  }\n\n  /**\n   * Loads the model and weight files, construct the in memory weight map and\n   * compile the inference graph.\n   */\n  async load(): Promise<boolean> {\n    this.findIOHandler();\n    if (this.handler.load == null) {\n      throw new Error(\n          'Cannot proceed with model loading because the IOHandler provided ' +\n          'does not have the `load` method implemented.');\n    }\n    const artifacts = await this.handler.load();\n\n    return this.loadSync(artifacts);\n  }\n\n  /**\n   * Synchronously construct the in memory weight map and\n   * compile the inference graph. Also initialize hashtable if any.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  loadSync(artifacts: io.ModelArtifacts) {\n    this.artifacts = artifacts;\n    const graph = this.artifacts.modelTopology as tensorflow.IGraphDef;\n    let signature = {};\n    if (this.artifacts.userDefinedMetadata != null) {\n      signature =  // tslint:disable-next-line:no-any\n          (this.artifacts.userDefinedMetadata as any).signature as\n          tensorflow.ISignatureDef;\n    }\n\n    this.version = `${graph.versions.producer}.${graph.versions.minConsumer}`;\n    const weightMap =\n        io.decodeWeights(this.artifacts.weightData, this.artifacts.weightSpecs);\n    this.executor = new GraphExecutor(\n        OperationMapper.Instance.transformGraph(graph, signature));\n    this.executor.weightMap = this.convertTensorMapToTensorsMap(weightMap);\n\n    if (artifacts.modelInitializer != null) {\n      const initializer =\n          OperationMapper.Instance.transformGraph(artifacts.modelInitializer);\n      this.initializer = new GraphExecutor(initializer);\n      this.initializer.weightMap = this.executor.weightMap;\n      this.initializer.execute({}, []);\n    }\n\n    return true;\n  }\n\n  /**\n   * Save the configuration and/or weights of the GraphModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const modelUrl =\n   *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n   * const model = await tf.loadGraphModel(modelUrl);\n   * const zeros = tf.zeros([1, 224, 224, 3]);\n   * model.predict(zeros).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * model.predict(zeros).print();\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new Error(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new Error(\n          'GraphModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    return handlerOrURL.save(this.artifacts);\n  }\n\n  /**\n   * Execute the inference for the input tensors.\n   *\n   * @param input The input tensors, when there is single input for the model,\n   * inputs param should be a `tf.Tensor`. For models with mutliple inputs,\n   * inputs params should be in either `tf.Tensor`[] if the input order is\n   * fixed, or otherwise NamedTensorMap format.\n   *\n   * For model with multiple inputs, we recommend you use NamedTensorMap as the\n   * input type, if you use `tf.Tensor`[], the order of the array needs to\n   * follow the\n   * order of inputNodes array. @see {@link GraphModel.inputNodes}\n   *\n   * You can also feed any intermediate nodes using the NamedTensorMap as the\n   * input type. For example, given the graph\n   *    InputNode => Intermediate => OutputNode,\n   * you can execute the subgraph Intermediate => OutputNode by calling\n   *    model.execute('IntermediateNode' : tf.tensor(...));\n   *\n   * This is useful for models that uses tf.dynamic_rnn, where the intermediate\n   * state needs to be fed manually.\n   *\n   * For batch inference execution, the tensors for each input need to be\n   * concatenated together. For example with mobilenet, the required input shape\n   * is [1, 244, 244, 3], which represents the [batch, height, width, channel].\n   * If we are provide a batched data of 100 images, the input tensor should be\n   * in the shape of [100, 244, 244, 3].\n   *\n   * @param config Prediction configuration for specifying the batch size and\n   * output node names. Currently the batch size option is ignored for graph\n   * model.\n   *\n   * @returns Inference result tensors. The output would be single `tf.Tensor`\n   * if model has single output node, otherwise Tensor[] or NamedTensorMap[]\n   * will be returned for model with multiple outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(inputs: Tensor|Tensor[]|NamedTensorMap, config?: ModelPredictConfig):\n      Tensor|Tensor[]|NamedTensorMap {\n    return this.execute(inputs, this.outputNodes);\n  }\n\n  private normalizeInputs(inputs: Tensor|Tensor[]|\n                          NamedTensorMap): NamedTensorMap {\n    if (!(inputs instanceof Tensor) && !Array.isArray(inputs)) {\n      // The input is already a NamedTensorMap.\n      return inputs;\n    }\n    inputs = Array.isArray(inputs) ? inputs : [inputs];\n    if (inputs.length !== this.inputNodes.length) {\n      throw new Error(\n          'Input tensor count mismatch,' +\n          `the graph model has ${this.inputNodes.length} placeholders, ` +\n          `while there are ${inputs.length} input tensors.`);\n    }\n    return this.inputNodes.reduce((map, inputName, i) => {\n      map[inputName] = (inputs as Tensor[])[i];\n      return map;\n    }, {} as NamedTensorMap);\n  }\n\n  private normalizeOutputs(outputs: string|string[]): string[] {\n    outputs = outputs || this.outputNodes;\n    return !Array.isArray(outputs) ? [outputs] : outputs;\n  }\n\n  /**\n   * Executes inference for the model for given input tensors.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no\n   * outputs are specified, the default outputs of the model would be used.\n   * You can inspect intermediate nodes of the model by adding them to the\n   * outputs array.\n   *\n   * @returns A single tensor if provided with a single output or no outputs\n   * are provided and there is only one default output, otherwise return a\n   * tensor array. The order of the tensor array is the same as the outputs\n   * if provided, otherwise the order of outputNodes attribute of the model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs?: string|string[]):\n      Tensor|Tensor[] {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = this.executor.execute(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n  /**\n   * Executes inference for the model for given input tensors in async\n   * fashion, use this method when your model contains control flow ops.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   *\n   * @returns A Promise of single tensor if provided with a single output or\n   * no outputs are provided and there is only one default output, otherwise\n   * return a tensor map.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async executeAsync(\n      inputs: Tensor|Tensor[]|NamedTensorMap,\n      outputs?: string|string[]): Promise<Tensor|Tensor[]> {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = await this.executor.executeAsync(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n\n  private convertTensorMapToTensorsMap(map: NamedTensorMap): NamedTensorsMap {\n    return Object.keys(map).reduce((newMap: NamedTensorsMap, key) => {\n      newMap[key] = [map[key]];\n      return newMap;\n    }, {});\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  dispose() {\n    this.executor.dispose();\n\n    if (this.initializer) {\n      this.initializer.dispose();\n    }\n  }\n}\n\n/**\n * Load a graph model given a URL to the model definition.\n *\n * Example of loading MobileNetV2 from a URL and making a prediction with a\n * zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n * const model = await tf.loadGraphModel(modelUrl);\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n *\n * Example of loading MobileNetV2 from a TF Hub URL and making a prediction with\n * a zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/2';\n * const model = await tf.loadGraphModel(modelUrl, {fromTFHub: true});\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n * @param modelUrl The url or an `io.IOHandler` that loads the model.\n * @param options Options for the HTTP request, which allows to send credentials\n *    and custom headers.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport async function loadGraphModel(\n    modelUrl: string|io.IOHandler,\n    options: io.LoadOptions = {}): Promise<GraphModel> {\n  if (modelUrl == null) {\n    throw new Error(\n        'modelUrl in loadGraphModel() cannot be null. Please provide a url ' +\n        'or an IOHandler that loads the model');\n  }\n  if (options == null) {\n    options = {};\n  }\n\n  if (options.fromTFHub) {\n    if ((modelUrl as io.IOHandler).load == null) {\n      if (!(modelUrl as string).endsWith('/')) {\n        modelUrl = (modelUrl as string) + '/';\n      }\n      modelUrl = `${modelUrl}${DEFAULT_MODEL_NAME}${TFHUB_SEARCH_PARAM}`;\n    }\n  }\n  const model = new GraphModel(modelUrl, options);\n  await model.load();\n  return model;\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\n\n// tslint:disable:no-any\n\n/**\n * A return value for a mapping function that can be applied via deepMap.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapResult = {\n  value: any,\n  recurse: boolean\n};\n\n/**\n * Apply a mapping function to a nested structure in a recursive manner.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapResult`.  The `DeepMapResult` either provides a\n *   replacement value for that node (i.e., replacing the subtree), or indicates\n *   that the node should be processed recursively.\n */\nexport function deepMap(input: any, mapFn: (x: any) => DeepMapResult): any|\n    any[] {\n  return deepMapInternal(input, mapFn);\n}\n\n/**\n * @param seen: A Map of known object mappings (i.e., memoized results of\n *   `mapFn()`)\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepMapInternal(\n    input: any, mapFn: (x: any) => DeepMapResult,\n    seen: Map<any, any> = new Map(), containedIn: Set<{}> = new Set()): any|\n    any[] {\n  if (input == null) {\n    return null;\n  }\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  if (seen.has(input)) {\n    return seen.get(input);\n  }\n  const result = mapFn(input);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep map function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    seen.set(input, result.value);\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const child = input[k];\n      const childResult = deepMapInternal(child, mapFn, seen, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// TODO(soergel, kangyizhang) Reconsider naming of deepZip() to avoid confusion\n// with zip()\n\n/**\n * Zip nested structures together in a recursive manner.\n *\n * This has the effect of transposing or pivoting data, e.g. converting it from\n * a row-major representation to a column-major representation.\n *\n * For example, `deepZip([{a: 1, b: 2}, {a: 3, b: 4}])` returns\n * `{a: [1, 3], b: [2, 4]}`.\n *\n * The inputs should all have the same nested structure (i.e., of arrays and\n * dicts).  The result is a single object with the same nested structure, where\n * the leaves are arrays collecting the values of the inputs at that location\n * (or, optionally, the result of a custom function applied to those arrays).\n *\n * @param inputs: An array of the objects to zip together.\n * @param zipFn: (optional) A function that expects an array of elements at a\n *   single node of the object tree, and returns a `DeepMapResult`.  The\n *   `DeepMapResult` either provides a result value for that node (i.e.,\n *   representing the subtree), or indicates that the node should be processed\n *   recursively.  The default zipFn recurses as far as possible and places\n *   arrays at the leaves.\n */\nexport function deepZip(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult = zipToList): any|any[] {\n  return deepZipInternal(inputs, zipFn);\n}\n\n/**\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepZipInternal(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult,\n    containedIn: Set<{}> = new Set()): any|any[] {\n  // The recursion follows the structure of input 0; it's assumed that all the\n  // other inputs have the same structure.\n  const input = inputs[0];\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  const result = zipFn(inputs);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep zip function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const children = inputs.map(x => x[k]);\n      const childResult = deepZipInternal(children, zipFn, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// tslint:disable-next-line:no-any\nexport function zipToList(x: any[]): DeepMapResult {\n  if (x === null) {\n    return null;\n  }\n  // TODO(soergel): validate array type?\n\n  if (isIterable(x[0])) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: x, recurse: false};\n  }\n}\n\n/**\n * A return value for an async map function for use with deepMapAndAwaitAll.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapAsyncResult = {\n  value: Promise<any>,\n  recurse: boolean\n};\n\n/**\n * Apply an async mapping function to a nested structure in a recursive manner.\n *\n * This first creates a nested structure of Promises, and then awaits all of\n * those, resulting in a single Promise for a resolved nested structure.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapAsyncResult`.  The `DeepMapAsyncResult` either provides\n *   a `Promise` for a replacement value for that node (i.e., replacing the\n *   subtree), or indicates that the node should be processed recursively.  Note\n *   that the decision whether or not to recurse must be made immediately; only\n *   the mapped value may be promised.\n */\nexport async function deepMapAndAwaitAll(\n    input: any, mapFn: (x: any) => DeepMapAsyncResult): Promise<any|any[]> {\n  const seen: Map<any, Promise<any>> = new Map();\n\n  // First do a normal deepMap, collecting Promises in 'seen' as a side effect.\n  deepMapInternal(input, mapFn, seen);\n\n  // Replace the Promises in 'seen' in place.\n  // Note TypeScript provides no async map iteration, and regular map iteration\n  // is broken too, so sadly we have to do Array.from() to make it work.\n  // (There's no advantage to Promise.all(), and that would be tricky anyway.)\n  for (const key of Array.from(seen.keys())) {\n    const value = seen.get(key);\n    if (value instanceof Promise) {\n      const mappedValue = await value;\n      seen.set(key, mappedValue);\n    }\n  }\n\n  // Normal deepMap again, this time filling in the resolved values.\n  // It's unfortunate that we have to do two passes.\n  // TODO(soergel): test performance and think harder about a fast solution.\n  const result = deepMapInternal(input, mapFn, seen);\n  return result;\n}\n\n/**\n * Determine whether the argument is iterable.\n *\n * @returns true if the argument is an array or any non-Tensor object.\n */\n// tslint:disable-next-line:no-any\nexport function isIterable(obj: any): boolean {\n  return obj != null && (!ArrayBuffer.isView(obj)) &&\n      (Array.isArray(obj) ||\n       (typeof obj === 'object' && !(obj instanceof tf.Tensor)));\n}\n\n/**\n * Determine whether the argument can be converted to Tensor.\n *\n * Tensors, primitives, arrays, and TypedArrays all qualify; anything else does\n * not.\n *\n * @returns true if the argument can be converted to Tensor.\n */\n// tslint:disable-next-line:no-any\nexport function canTensorify(obj: any): boolean {\n  return obj == null || isPrimitive(obj) || Array.isArray(obj) ||\n      (typeof obj === 'object' && (obj instanceof tf.Tensor)) ||\n      tf.util.isTypedArray(obj);\n}\n\n/**\n * Returns true if the given `value` is a primitive type. Otherwise returns\n * false. This is equivalant to node util.isPrimitive\n */\nfunction isPrimitive(value: any): boolean {\n  return (\n      value === null ||\n      (typeof value !== 'object' && typeof value !== 'function'));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {deepMap, DeepMapResult, isIterable} from './deep_map';\n\nexport function deepClone<T>(container: T): T {\n  return deepMap(container, cloneIfTensor);\n}\n\n// tslint:disable-next-line: no-any\nfunction cloneIfTensor(item: any): DeepMapResult {\n  if (item instanceof tf.Tensor) {\n    return ({value: item.clone(), recurse: false});\n  } else if (isIterable(item)) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: item, recurse: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/**\n * A ring buffer, providing O(1) FIFO, LIFO, and related operations.\n */\nexport class RingBuffer<T> {\n  // Note we store the indices in the range 0 <= index < 2*capacity.\n  // This allows us to distinguish the full from the empty case.\n  // See https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/\n  protected begin = 0;  // inclusive\n  protected end = 0;    // exclusive\n  protected doubledCapacity: number;\n\n  protected data: T[];\n\n  /**\n   * Constructs a `RingBuffer`.\n   * @param capacity The number of items that the buffer can accomodate.\n   */\n  constructor(public capacity: number) {\n    if (capacity == null) {\n      throw new RangeError('Can\\'t create a ring buffer of unknown capacity.');\n    }\n    if (capacity < 1) {\n      throw new RangeError('Can\\'t create ring buffer of capacity < 1.');\n    }\n    this.data = new Array<T>(capacity);\n    this.doubledCapacity = 2 * capacity;\n  }\n\n  /**\n   * Map any index into the range 0 <= index < 2*capacity.\n   */\n  protected wrap(index: number) {\n    // don't trust % on negative numbers\n    while (index < 0) {\n      index += this.doubledCapacity;\n    }\n    return index % this.doubledCapacity;\n  }\n\n  protected get(index: number) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t get item at a negative index.');\n    }\n    return this.data[index % this.capacity];\n  }\n\n  protected set(index: number, value: T) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t set item at a negative index.');\n    }\n    this.data[index % this.capacity] = value;\n  }\n\n  /**\n   * Returns the current number of items in the buffer.\n   */\n  length(): number {\n    let length = this.end - this.begin;\n    if (length < 0) {\n      length = this.doubledCapacity + length;\n    }\n    return length;\n  }\n\n  /**\n   * Reports whether the buffer is full.\n   * @returns true if the number of items in the buffer equals its capacity, and\n   *   false otherwise.\n   */\n  isFull() {\n    return this.length() === this.capacity;\n  }\n\n  /**\n   * Reports whether the buffer is empty.\n   * @returns true if the number of items in the buffer equals zero, and\n   *   false otherwise.\n   */\n  isEmpty() {\n    return this.length() === 0;\n  }\n\n  /**\n   * Adds an item to the end of the buffer.\n   */\n  push(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.set(this.end, value);\n    this.end = this.wrap(this.end + 1);\n  }\n\n  /**\n   * Adds many items to the end of the buffer, in order.\n   */\n  pushAll(values: T[]) {\n    for (const value of values) {\n      this.push(value);\n    }\n  }\n\n  /**\n   * Removes and returns the last item in the buffer.\n   */\n  pop(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    this.end = this.wrap(this.end - 1);\n    const result = this.get(this.end);\n    this.set(this.end, undefined);\n    return result;\n  }\n\n  /**\n   * Adds an item to the beginning of the buffer.\n   */\n  unshift(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.begin = this.wrap(this.begin - 1);\n    this.set(this.begin, value);\n  }\n\n  /**\n   * Removes and returns the first item in the buffer.\n   */\n  shift(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const result = this.get(this.begin);\n    this.set(this.begin, undefined);\n    this.begin = this.wrap(this.begin + 1);\n    return result;\n  }\n\n  /**\n   * Removes and returns a specific item in the buffer, and moves the last item\n   * to the vacated slot.  This is useful for implementing a shuffling stream.\n   * Note that this operation necessarily scrambles the original order.\n   *\n   * @param relativeIndex: the index of the item to remove, relative to the\n   *   first item in the buffer (e.g., hiding the ring nature of the underlying\n   *   storage).\n   */\n  shuffleExcise(relativeIndex: number): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const index = this.wrap(this.begin + relativeIndex);\n    const result = this.get(index);\n    this.set(index, this.pop());\n    return result;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {RingBuffer} from './ring_buffer';\n\nexport class GrowingRingBuffer<T> extends RingBuffer<T> {\n  private static INITIAL_CAPACITY = 32;\n\n  /**\n   * Constructs a `GrowingRingBuffer`.\n   */\n  constructor() {\n    super(GrowingRingBuffer.INITIAL_CAPACITY);\n  }\n\n  isFull() {\n    return false;\n  }\n\n  push(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.push(value);\n  }\n\n  unshift(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.unshift(value);\n  }\n\n  /**\n   * Doubles the capacity of the buffer.\n   */\n  private expand() {\n    const newCapacity = this.capacity * 2;\n    const newData = new Array<T>(newCapacity);\n    const len = this.length();\n\n    // Rotate the buffer to start at index 0 again, since we can't just\n    // allocate more space at the end.\n    for (let i = 0; i < len; i++) {\n      newData[i] = this.get(this.wrap(this.begin + i));\n    }\n\n    this.data = newData;\n    this.capacity = newCapacity;\n    this.doubledCapacity = 2 * this.capacity;\n    this.begin = 0;\n    this.end = len;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {Container} from '../types';\nimport {deepClone} from '../util/deep_clone';\nimport {deepMapAndAwaitAll, DeepMapAsyncResult, DeepMapResult, deepZip, zipToList} from '../util/deep_map';\nimport {GrowingRingBuffer} from '../util/growing_ring_buffer';\nimport {RingBuffer} from '../util/ring_buffer';\n\n/**\n * A nested structure of LazyIterators, used as the input to zip().\n */\nexport type IteratorContainer = Container<LazyIterator<tf.TensorContainer>>;\n\n// Here we implement a simple asynchronous iterator.\n// This lets us avoid using either third-party stream libraries or\n// recent TypeScript language support requiring polyfills.\n\n/**\n * Create a `LazyIterator` from an array of items.\n */\nexport function iteratorFromItems<T>(items: T[]): LazyIterator<T> {\n  return new ArrayIterator(items);\n}\n\n/**\n * Create a `LazyIterator` of incrementing integers.\n */\nexport function iteratorFromIncrementing(start: number): LazyIterator<number> {\n  let i = start;\n  return iteratorFromFunction(() => ({value: i++, done: false}));\n}\n\n/**\n * Create a `LazyIterator` from a function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * await iter.forEachAsync(e => console.log(e));\n * ```\n *\n * @param func A function that produces data on each call.\n */\nexport function iteratorFromFunction<T>(\n    func: () =>\n        IteratorResult<T>| Promise<IteratorResult<T>>): LazyIterator<T> {\n  return new FunctionCallIterator(func);\n}\n\n/**\n * Create a `LazyIterator` by concatenating underlying streams, which are\n * themselves provided as a stream.\n *\n * This can also be thought of as a \"stream flatten\" operation.\n *\n * @param baseIterators A stream of streams to be concatenated.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenated<T>(\n    baseIterators: LazyIterator<LazyIterator<T>>,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return new ChainedIterator(baseIterators, baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by concatenating streams produced by calling a\n * stream-generating function a given number of times.\n *\n * Since a `LazyIterator` is read-once, it cannot be repeated, but this\n * function can be used to achieve a similar effect:\n *\n *   LazyIterator.ofConcatenatedFunction(() => new MyIterator(), 6);\n *\n * @param iteratorFunc: A function that produces a new stream on each call.\n * @param count: The number of times to call the function.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenatedFunction<T>(\n    iteratorFunc: () => IteratorResult<LazyIterator<T>>, count: number,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return iteratorFromConcatenated(\n      iteratorFromFunction(iteratorFunc).take(count), baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by zipping together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nexport function iteratorFromZipped<O extends tf.TensorContainer>(\n    iterators: IteratorContainer,\n    mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL): LazyIterator<O> {\n  return new ZipIterator<O>(iterators, mismatchMode);\n}\n\n/**\n * An asynchronous iterator, providing lazy access to a potentially\n * unbounded stream of elements.\n *\n * Iterator can be obtained from a dataset:\n * `const iter = await dataset.iterator();`\n */\nexport abstract class LazyIterator<T> {\n  // This class implements AsyncIterator<T>, but we have not yet set the\n  // TypeScript --downlevelIteration flag to enable that.\n\n  abstract summary(): string;\n\n  /**\n   * Returns a `Promise` for the next element in the stream.\n   *\n   * When an item can be provided successfully, the return value is\n   * `{value:T, done:false}`.\n   *\n   * Calling next() on a closed stream returns `{value:null, done:true}`.\n   */\n  abstract async next(): Promise<IteratorResult<T>>;\n\n  /**\n   * Collect all remaining elements of a bounded stream into an array.\n   * Obviously this will succeed only for small streams that fit in memory.\n   * Useful for testing.\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArray(): Promise<T[]> {\n    const result: T[] = [];\n    let x = await this.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await this.next();\n    }\n    return result;\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArrayForTest(): Promise<T[]> {\n    const stream = this.prefetch(100);\n    const result: T[] = [];\n    let x = await stream.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await stream.next();\n    }\n    return result;\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveFully(): Promise<void> {\n    let x = await this.next();\n    while (!x.done) {\n      x = await this.next();\n    }\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted, or a predicate fails.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveWhile(predicate: (r: T) => boolean): Promise<void> {\n    let x = await this.next();\n    let shouldContinue = predicate(x.value);\n    while ((!x.done) && shouldContinue) {\n      x = await this.next();\n      shouldContinue = predicate(x.value);\n    }\n  }\n\n  /**\n   * Handles errors thrown on this stream using a provided handler function.\n   *\n   * @param handler A function that handles any `Error` thrown during a `next()`\n   *   call and returns true if the stream should continue (dropping the failed\n   *   call) or false if the stream should quietly terminate.  If the handler\n   *   itself throws (or rethrows) an `Error`, that will be propagated.\n   *\n   * @returns A `LazyIterator` of elements passed through from upstream,\n   *   possibly filtering or terminating on upstream `next()` calls that\n   *   throw an `Error`.\n   */\n  handleErrors(handler: (error: Error) => boolean): LazyIterator<T> {\n    return new ErrorHandlingLazyIterator(this, handler);\n  }\n\n  // TODO(soergel): Implement reduce() etc.\n\n  /**\n   * Filters this stream according to `predicate`.\n   *\n   * @param predicate A function mapping a stream element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `LazyIterator` of elements for which the predicate was true.\n   */\n  filter(predicate: (value: T) => boolean): LazyIterator<T> {\n    return new FilterIterator(this, predicate);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  map<O>(transform: (value: T) => O): LazyIterator<O> {\n    return new MapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through an async 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a `Promise` for a\n   *   transformed stream element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  mapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform, forcing serial execution.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  serialMapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform).serial();\n  }\n\n  /**\n   * Maps this stream through a 1-to-many transform.\n   *\n   * @param transform A function mapping a stream element to an array of\n   *   transformed elements.\n   *\n   * @returns A `DataStream` of transformed elements.\n   */\n  flatmap<O>(transform: (value: T) => O[]): LazyIterator<O> {\n    return new FlatmapIterator(this, transform);\n  }\n\n  /**\n   * Apply a function to every element of the stream.\n   *\n   * @param f A function to apply to each stream element.\n   */\n  async forEachAsync(f: (value: T) => void): Promise<void> {\n    return this.map(f).resolveFully();\n  }\n\n  /**\n   * Apply a function to every element of the stream, forcing serial execution.\n   *\n   * @param f A function to apply to each stream element.  Should return 'true'\n   *   to indicate that the stream should continue, or 'false' to cause it to\n   *   terminate.\n   */\n  async serialForEach(f: (value: T) => Promise<boolean>): Promise<void> {\n    return this.serialMapAsync(f).resolveWhile(x => (x === true));\n  }\n\n  /**\n   * Groups elements into batches, represented as arrays of elements.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"Row-major\" means that the resulting batch is simply a collection of\n   * rows: `[row1, row2, row3, ...]`.  This is contrast to the column-major\n   * form, which is needed for vectorized computation.\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `LazyIterator` of batches of elements, represented as arrays\n   *   of the original element type.\n   */\n  rowMajorBatch(batchSize: number, smallLastBatch = true): LazyIterator<T[]> {\n    return new RowMajorBatchIterator(this, batchSize, smallLastBatch);\n  }\n\n  /**\n   * Groups elements into batches, represented in column-major form.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"column-major\" means that the resulting batch is a (potentially\n   * nested) structure representing the columns.  Each column entry, then,\n   * contains a collection of the values found in that column for a range of\n   * input elements.  This representation allows for vectorized computation, in\n   * contrast to the row-major form.\n   *\n   * The inputs should all have the same nested structure (i.e., of arrays and\n   * dicts).  The result is a single object with the same nested structure,\n   * where the leaves are arrays collecting the values of the inputs at that\n   * location (or, optionally, the result of a custom function applied to those\n   * arrays).\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @param zipFn: (optional) A function that expects an array of elements at a\n   *   single node of the object tree, and returns a `DeepMapResult`.  The\n   *   `DeepMapResult` either provides a result value for that node (i.e.,\n   *   representing the subtree), or indicates that the node should be processed\n   *   recursively.  The default zipFn recurses as far as possible and places\n   *   arrays at the leaves.\n   * @returns A `LazyIterator` of batches of elements, represented as an object\n   *   with collections at the leaves.\n   */\n  columnMajorBatch(\n      batchSize: number, smallLastBatch = true,\n      // tslint:disable-next-line:no-any\n      zipFn: (xs: any[]) => DeepMapResult = zipToList):\n      LazyIterator<tf.TensorContainer> {\n    // First collect the desired number of input elements as a row-major batch.\n    const rowBatches = this.rowMajorBatch(batchSize, smallLastBatch);\n    // Now 'rotate' or 'pivot' the data, collecting all values from each column\n    // in the batch (i.e., for each key within the elements) into an array.\n    return rowBatches.map(x => deepZip(x, zipFn));\n  }\n\n  /**\n   * Concatenate this `LazyIterator` with another.\n   *\n   * @param iterator A `LazyIterator` to be concatenated onto this one.\n   * @param baseErrorHandler An optional function that can intercept `Error`s\n   *   raised during a `next()` call on the base stream.  This function can\n   *   decide whether the error should be propagated, whether the error should\n   *   be ignored, or whether the base stream should be terminated.\n   * @returns A `LazyIterator`.\n   */\n  concatenate(\n      iterator: LazyIterator<T>,\n      baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n    return new ChainedIterator(\n        iteratorFromItems([this, iterator]), baseErrorHandler);\n  }\n\n  /**\n   * Limits this stream to return at most `count` items.\n   *\n   * @param count The maximum number of items to provide from the stream. If\n   * a negative or undefined value is given, the entire stream is returned\n   *   unaltered.\n   */\n  take(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new TakeIterator(this, count);\n  }\n\n  /**\n   * Skips the first `count` items in this stream.\n   *\n   * @param count The number of items to skip.  If a negative or undefined\n   * value is given, the entire stream is returned unaltered.\n   */\n  skip(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new SkipIterator(this, count);\n  }\n\n  /**\n   * Prefetch the first `bufferSize` items in this stream.\n   *\n   * Note this prefetches Promises, but makes no guarantees about when those\n   * Promises resolve.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   */\n  prefetch(bufferSize: number): LazyIterator<T> {\n    return new PrefetchIterator(this, bufferSize);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  /**\n   * Randomly shuffles the elements of this stream.\n   *\n   * @param bufferSize: An integer specifying the number of elements from\n   * this stream from which the new stream will sample.\n   * @param seed: (Optional.) An integer specifying the random seed that\n   * will be used to create the distribution.\n   */\n  shuffle(windowSize: number, seed?: string): LazyIterator<T> {\n    return new ShuffleIterator(this, windowSize, seed);\n  }\n\n  /**\n   * Force an iterator to execute serially: each next() call will await the\n   * prior one, so that they cannot execute concurrently.\n   */\n  serial(): LazyIterator<T> {\n    return new SerialIterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on LazyIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// Iterators that just extend LazyIterator directly\n// ============================================================================\n\nclass ArrayIterator<T> extends LazyIterator<T> {\n  private trav = 0;\n  constructor(protected items: T[]) {\n    super();\n  }\n\n  summary() {\n    return `Array of ${this.items.length} items`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.trav >= this.items.length) {\n      return {value: null, done: true};\n    }\n    const item = this.items[this.trav];\n    this.trav++;\n    return {value: deepClone(item), done: false};\n  }\n}\n\nclass FunctionCallIterator<T> extends LazyIterator<T> {\n  constructor(\n      protected nextFn: () => IteratorResult<T>| Promise<IteratorResult<T>>) {\n    super();\n  }\n\n  summary() {\n    return `Function call`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    try {\n      return this.nextFn();\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message =\n          `Error thrown while iterating through a dataset: ${e.message}`;\n      throw e;\n    }\n  }\n}\n\nclass SerialIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(protected upstream: LazyIterator<T>) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Serial`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    return this.upstream.next();\n  }\n}\n\nclass SkipIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  count = 0;\n\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Skip`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider tradeoffs of reading in parallel, eg.\n    // collecting next() promises in an Array and then waiting for\n    // Promise.all() of those. Benefit: pseudo-parallel execution.  Drawback:\n    // maybe delayed GC.\n    while (this.count++ < this.maxCount) {\n      const skipped = await this.upstream.next();\n      // short-circuit if upstream is already empty\n      if (skipped.done) {\n        return skipped;\n      }\n      tf.dispose(skipped.value as {});\n    }\n    return this.upstream.next();\n  }\n}\n\nclass TakeIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Take`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.count++ >= this.maxCount) {\n      return {value: null, done: true};\n    }\n    return this.upstream.next();\n  }\n}\n\n// Note this batch just groups items into row-wise element arrays.\n// Rotating these to a column-wise representation happens only at the dataset\n// level.\nclass RowMajorBatchIterator<T> extends LazyIterator<T[]> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T[]>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected batchSize: number,\n      protected enableSmallLastBatch = true) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> RowMajorBatch`;\n  }\n\n  async next(): Promise<IteratorResult<T[]>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T[]>> {\n    const batch: T[] = [];\n    while (batch.length < this.batchSize) {\n      const item = await this.upstream.next();\n      if (item.done) {\n        if (this.enableSmallLastBatch && batch.length > 0) {\n          return {value: batch, done: false};\n        }\n        return {value: null, done: true};\n      }\n      batch.push(item.value);\n    }\n    return {value: batch, done: false};\n  }\n}\n\nclass FilterIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected predicate: (value: T) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Filter`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      const item = await this.upstream.next();\n      if (item.done || this.predicate(item.value)) {\n        return item;\n      }\n      tf.dispose(item.value as {});\n    }\n  }\n}\n\nclass MapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Map`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\nclass ErrorHandlingLazyIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected handler: (error: Error) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> handleErrors`;\n  }\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      try {\n        return await this.upstream.next();\n      } catch (e) {\n        if (!this.handler(e)) {\n          return {value: null, done: true};\n        }\n        // If the handler returns true, loop and fetch the next upstream item.\n\n        // If the upstream iterator throws an endless stream of errors, and if\n        // the handler says to ignore them, then we loop forever here.  That is\n        // the correct behavior-- it's up to the handler to decide when to stop.\n      }\n    }\n  }\n}\n\nclass AsyncMapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => Promise<O>) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> AsyncMap`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = await this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\n// Iterators that maintain a queue of pending items\n// ============================================================================\n\n/**\n * A base class for transforming streams that operate by maintaining an\n * output queue of elements that are ready to return via next().  This is\n * commonly required when the transformation is 1-to-many:  A call to next()\n * may trigger a call to the underlying stream, which will produce many\n * mapped elements of this stream-- of which we need to return only one, so\n * we have to queue the rest.\n */\nexport abstract class OneToManyIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  protected outputQueue: RingBuffer<T>;\n\n  constructor() {\n    super();\n    this.outputQueue = new GrowingRingBuffer<T>();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  /**\n   * Read one or more chunks from upstream and process them, possibly\n   * reading or writing a carryover, and adding processed items to the\n   * output queue.  Note it's possible that no items are added to the queue\n   * on a given pump() call, even if the upstream stream is not closed\n   * (e.g., because items are filtered).\n   *\n   * @return `true` if any action was taken, i.e. fetching items from the\n   *   upstream source OR adding items to the output queue.  `false` if the\n   *   upstream source is exhausted AND nothing was added to the queue\n   * (i.e., any remaining carryover).\n   */\n  protected abstract async pump(): Promise<boolean>;\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // Fetch so that the queue contains at least one item if possible.\n    // If the upstream source is exhausted, AND there are no items left in\n    // the output queue, then this stream is also exhausted.\n    while (this.outputQueue.length() === 0) {\n      // TODO(soergel): consider parallel reads.\n      if (!await this.pump()) {\n        return {value: null, done: true};\n      }\n    }\n    return {value: this.outputQueue.shift(), done: false};\n  }\n}\nclass FlatmapIterator<I, O> extends OneToManyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O[]) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Flatmap`;\n  }\n\n  async pump(): Promise<boolean> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return false;\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // that's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying any\n    // intermediate Tensors.  Here we are concerned only about the inputs.\n    const mappedArray = this.transform(item.value);\n    const outputTensors =\n        tf.tensor_util.getTensorsInContainer(mappedArray as {});\n    this.outputQueue.pushAll(mappedArray);\n\n    // TODO(soergel) faster intersection, and deduplicate outputTensors\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n\n    return true;\n  }\n}\n\n/**\n * Provides a `LazyIterator` that concatenates a stream of underlying\n * streams.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n */\nexport class ChainedIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>> = null;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private iterator: LazyIterator<T> = null;\n  private moreIterators: LazyIterator<LazyIterator<T>>;\n\n  constructor(\n      iterators: LazyIterator<LazyIterator<T>>,\n      private readonly baseErrorHandler?: (e: Error) => boolean) {\n    super();\n    this.moreIterators = iterators;\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of chained summaries';\n    return `${upstreamSummaries} -> Chained`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    this.lastRead = this.readFromChain(this.lastRead);\n    return this.lastRead;\n  }\n\n  private async readFromChain(lastRead: Promise<IteratorResult<T>>):\n      Promise<IteratorResult<T>> {\n    // Must await on the previous read since the previous read may have advanced\n    // the stream of streams, from which we need to read.\n    // This is unfortunate since we can't parallelize reads. Which means\n    // prefetching of chained streams is a no-op.\n    // One solution is to prefetch immediately upstream of this.\n    await lastRead;\n    if (this.iterator == null) {\n      const iteratorResult = await this.moreIterators.next();\n      if (iteratorResult.done) {\n        // No more streams to stream from.\n        return {value: null, done: true};\n      }\n      this.iterator = iteratorResult.value;\n      if (this.baseErrorHandler != null) {\n        this.iterator = this.iterator.handleErrors(this.baseErrorHandler);\n      }\n    }\n    const itemResult = await this.iterator.next();\n    if (itemResult.done) {\n      this.iterator = null;\n      return this.readFromChain(lastRead);\n    }\n    return itemResult;\n  }\n}\n\nexport enum ZipMismatchMode {\n  FAIL,      // require zipped streams to have the same length\n  SHORTEST,  // terminate zip when the first stream is exhausted\n  LONGEST    // use nulls for exhausted streams; use up the longest stream.\n}\n\n/**\n * Provides a `LazyIterator` that zips together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nclass ZipIterator<O extends tf.TensorContainer> extends LazyIterator<O> {\n  private count = 0;\n  private currentPromise: Promise<IteratorResult<O>> = null;\n\n  constructor(\n      protected readonly iterators: IteratorContainer,\n      protected readonly mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL) {\n    super();\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of zip summaries';\n    return `{${upstreamSummaries}} -> Zip`;\n  }\n\n  private async nextState(afterState: Promise<IteratorResult<O>>):\n      Promise<IteratorResult<O>> {\n    // This chaining ensures that the underlying next() are not even called\n    // before the previous ones have resolved.\n    await afterState;\n\n    // Collect underlying iterator \"done\" signals as a side effect in\n    // getNext()\n    let numIterators = 0;\n    let iteratorsDone = 0;\n\n    function getNext(container: IteratorContainer): DeepMapAsyncResult {\n      if (container instanceof LazyIterator) {\n        const result = container.next();\n        return {\n          value: result.then(x => {\n            numIterators++;\n            if (x.done) {\n              iteratorsDone++;\n            }\n            return x.value;\n          }),\n          recurse: false\n        };\n      } else {\n        return {value: null, recurse: true};\n      }\n    }\n\n    const mapped: O = await deepMapAndAwaitAll(this.iterators, getNext);\n\n    if (numIterators === iteratorsDone) {\n      // The streams have all ended.\n      return {value: null, done: true};\n    }\n    if (iteratorsDone > 0) {\n      switch (this.mismatchMode) {\n        case ZipMismatchMode.FAIL:\n          throw new Error(\n              'Zipped streams should have the same length. ' +\n              `Mismatched at element ${this.count}.`);\n        case ZipMismatchMode.SHORTEST:\n          return {value: null, done: true};\n        case ZipMismatchMode.LONGEST:\n        default:\n          // Continue.  The exhausted streams already produced value: null.\n      }\n    }\n\n    this.count++;\n    return {value: mapped, done: false};\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    this.currentPromise = this.nextState(this.currentPromise);\n    return this.currentPromise;\n  }\n}\n\n// Iterators that maintain a ring buffer of pending promises\n// ============================================================================\n\n/**\n * A stream that prefetches a given number of items from an upstream source,\n * returning them in FIFO order.\n *\n * Note this prefetches Promises, but makes no guarantees about when those\n * Promises resolve.\n */\nexport class PrefetchIterator<T> extends LazyIterator<T> {\n  protected buffer: RingBuffer<Promise<IteratorResult<T>>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected bufferSize: number) {\n    super();\n    this.buffer = new RingBuffer<Promise<IteratorResult<T>>>(bufferSize);\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Prefetch`;\n  }\n\n  /**\n   * Refill the prefetch buffer.  Returns only after the buffer is full, or\n   * the upstream source is exhausted.\n   */\n  protected refill() {\n    while (!this.buffer.isFull()) {\n      const v = this.upstream.next();\n      this.buffer.push(v);\n    }\n  }\n\n  next(): Promise<IteratorResult<T>> {\n    this.refill();\n    // This shift will never throw an error because the buffer is always\n    // full after a refill. If the stream is exhausted, the buffer will be\n    // full of Promises that will resolve to the end-of-stream signal.\n    return this.buffer.shift();\n  }\n}\n\n/**\n * A stream that performs a sliding-window random shuffle on an upstream\n * source. This is like a `PrefetchIterator` except that the items are\n * returned in randomized order.  Mixing naturally improves as the buffer\n * size increases.\n */\nexport class ShuffleIterator<T> extends PrefetchIterator<T> {\n  private readonly random: seedrandom.prng;\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private upstreamExhausted = false;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected windowSize: number,\n      seed?: string) {\n    super(upstream, windowSize);\n    this.random = seedrandom.alea(seed || tf.util.now().toString());\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private randomInt(max: number) {\n    return Math.floor(this.random() * max);\n  }\n\n  protected chooseIndex(): number {\n    return this.randomInt(this.buffer.length());\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider performance\n    if (!this.upstreamExhausted) {\n      this.refill();\n    }\n    while (!this.buffer.isEmpty()) {\n      const chosenIndex = this.chooseIndex();\n      const result = await this.buffer.shuffleExcise(chosenIndex);\n      if (result.done) {\n        this.upstreamExhausted = true;\n      } else {\n        this.refill();\n        return result;\n      }\n    }\n    return {value: null, done: true};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {TensorContainer, TensorLike} from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {iteratorFromConcatenated, iteratorFromFunction, iteratorFromItems, iteratorFromZipped, LazyIterator, ZipMismatchMode} from './iterators/lazy_iterator';\nimport {Container} from './types';\nimport {canTensorify, deepMapAndAwaitAll, DeepMapResult, isIterable} from './util/deep_map';\n\n/**\n * A nested structure of Datasets, used as the input to zip().\n */\nexport type DatasetContainer = Container<Dataset<TensorContainer>>;\n\n// TODO(soergel): consider vectorized operations within the pipeline.\n\n/**\n * Represents a potentially large list of independent data elements (typically\n * 'samples' or 'examples').\n *\n * A 'data example' may be a primitive, an array, a map from string keys to\n * values, or any nested structure of these.\n *\n * A `Dataset` represents an ordered collection of elements, together with a\n * chain of transformations to be performed on those elements. Each\n * transformation is a method of `Dataset` that returns another `Dataset`, so\n * these may be chained, e.g.\n * `const processedDataset = rawDataset.filter(...).map(...).batch(...)`.\n *\n * Data loading and transformation is done in a lazy, streaming fashion.  The\n * dataset may be iterated over multiple times; each iteration starts the data\n * loading anew and recapitulates the transformations.\n *\n * A `Dataset` is typically processed as a stream of unbatched examples --i.e.,\n * its transformations are applied one example at a time. Batching produces a\n * new `Dataset` where each element is a batch. Batching should usually come\n * last in a pipeline, because data transformations are easier to express on a\n * per-example basis than on a per-batch basis.\n *\n * The following code examples are calling `await dataset.forEachAsync(...)` to\n * iterate once over the entire dataset in order to print out the data.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport abstract class Dataset<T extends tf.TensorContainer> {\n  /*\n   * Provide a new stream of elements.  Note this will also start new streams\n   * from any underlying `Dataset`s.\n   *\n   * CAUTION: Any Tensors contained within the elements returned from\n   * this stream *must* be manually disposed to avoid a GPU memory leak.\n   * The tf.tidy() approach cannot be used in an asynchronous context.\n   */\n  abstract async iterator(): Promise<LazyIterator<T>>;\n\n  readonly size: number = null;\n\n  // TODO(soergel): Make Datasets report whether repeated iterator() calls\n  // produce the same result (e.g., reading from a file) or different results\n  // (e.g., from the webcam).  Currently we don't make this distinction but it\n  // could be important for the user to know.\n  // abstract isDeterministic(): boolean;\n\n  /**\n   * Groups elements into batches.\n   *\n   * It is assumed that each of the incoming dataset elements has the same\n   * structure-- i.e. the same set of keys at each location in an object\n   * hierarchy.  For each key, the resulting `Dataset` provides a batched\n   * element collecting all of the incoming values for that key.\n   *\n   *  * Incoming primitives are grouped into a 1-D Tensor.\n   *  * Incoming Tensors are grouped into a new Tensor where the 0'th axis is\n   *    the batch dimension.\n   *  * Incoming arrays are converted to Tensor and then batched.\n   *  * A nested array is interpreted as an n-D Tensor, so the batched result\n   *    has n+1 dimensions.\n   *  * An array that cannot be converted to Tensor produces an error.\n   *\n   * If an array should not be batched as a unit, it should first be converted\n   * to an object with integer keys.\n   *\n   * Here are a few examples:\n   *\n   * Batch a dataset of numbers:\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8]).batch(4);\n   * await a.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of arrays:\n   * ```js\n   * const b = tf.data.array([[1], [2], [3], [4], [5], [6], [7], [8]]).batch(4);\n   * await b.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of objects:\n   * ```js\n   * const c = tf.data.array([{a: 1, b: 11}, {a: 2, b: 12}, {a: 3, b: 13},\n   *   {a: 4, b: 14}, {a: 5, b: 15}, {a: 6, b: 16}, {a: 7, b: 17},\n   *   {a: 8, b: 18}]).batch(4);\n   * await c.forEachAsync(e => {\n   *   console.log('{');\n   *   for(var key in e) {\n   *     console.log(key+':');\n   *     e[key].print();\n   *   }\n   *   console.log('}');\n   * })\n   * ```\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `Dataset`, from which a stream of batches can be obtained.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  batch(batchSize: number, smallLastBatch = true): Dataset<tf.TensorContainer> {\n    const base = this;\n    tf.util.assert(\n        batchSize > 0, () => `batchSize needs to be positive, but it is\n      ${batchSize}`);\n    let size;\n    if (this.size === Infinity || this.size == null) {\n      // If the size of this dataset is infinity or null, the new size keeps the\n      // same.\n      size = this.size;\n    } else if (smallLastBatch) {\n      // If the size of this dataset is known and include small last batch, the\n      // new size is full batch count plus last batch.\n      size = Math.ceil(this.size / batchSize);\n    } else {\n      // If the size of this dataset is known and not include small last batch,\n      // the new size is full batch count.\n      size = Math.floor(this.size / batchSize);\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator())\n          .columnMajorBatch(batchSize, smallLastBatch, deepBatchConcat);\n    }, size);\n  }\n\n  /**\n   * Concatenates this `Dataset` with another.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * const b = tf.data.array([4, 5, 6]);\n   * const c = a.concatenate(b);\n   * await c.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param dataset A `Dataset` to be concatenated onto this one.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  concatenate(dataset: Dataset<T>): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity || dataset.size === Infinity) {\n      // If the size of any of these two dataset is infinity, new size is\n      // infinity.\n      size = Infinity;\n    } else if (this.size != null && dataset.size != null) {\n      // If the size of both datasets are known and not infinity, new size is\n      // sum the size of these two datasets.\n      size = this.size + dataset.size;\n    } else {\n      // If neither of these two datasets has infinite size and any of these two\n      // datasets' size is null, the new size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () =>\n            (await base.iterator()).concatenate(await dataset.iterator()),\n        size);\n  }\n\n  /**\n   * Filters this dataset according to `predicate`.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])\n   *   .filter(x => x%2 === 0);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param predicate A function mapping a dataset element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `Dataset` of elements for which the predicate was true.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  filter(predicate: (value: T) => boolean): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity) {\n      // If the size of this dataset is infinity, new size is infinity\n      size = Infinity;\n    } else {\n      // If this dataset has limited elements, new size is null because it might\n      // exhausted randomly.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).filter(x => tf.tidy(() => predicate(x)));\n    }, size);\n  }\n\n  /**\n   * Apply a function to every element of the dataset.\n   *\n   * After the function is applied to a dataset element, any Tensors contained\n   * within that element are disposed.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param f A function to apply to each dataset element.\n   * @returns A `Promise` that resolves after all elements have been processed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async forEachAsync(f: (input: T) => void): Promise<void> {\n    return (await this.iterator()).forEachAsync(f);\n  }\n\n  /**\n   * Maps this dataset through a 1-to-1 transform.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).map(x => x*x);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a transformed\n   *   dataset element.\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  map<O extends tf.TensorContainer>(transform: (value: T) => O): Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).map(x => tf.tidy(() => transform(x)));\n    }, this.size);\n  }\n\n  /**\n   * Maps this dataset through an async 1-to-1 transform.\n   *\n   * ```js\n   * const a =\n   *  tf.data.array([1, 2, 3]).mapAsync(x => new Promise(function(resolve){\n   *    setTimeout(() => {\n   *      resolve(x * x);\n   *    }, Math.random()*1000 + 500);\n   *  }));\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a `Promise` for a\n   *   transformed dataset element.  This transform is responsible for disposing\n   *   any intermediate `Tensor`s, i.e. by wrapping its computation in\n   *   `tf.tidy()`; that cannot be automated here (as it is in the synchronous\n   *   `map()` case).\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  mapAsync<O extends tf.TensorContainer>(transform: (value: T) => Promise<O>):\n      Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).mapAsync(transform);\n    }, this.size);\n  }\n\n  /**\n   *  Creates a `Dataset` that prefetches elements from this dataset.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  prefetch(bufferSize: number): Dataset<T> {\n    if (bufferSize == null) {\n      throw new RangeError(\n          '`Dataset.prefetch()` requires bufferSize to be specified.');\n    }\n\n    const base = this;\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).prefetch(bufferSize), this.size);\n  }\n\n  /**\n   * Repeats this dataset `count` times.\n   *\n   * NOTE: If this dataset is a function of global state (e.g. a random number\n   * generator), then different repetitions may produce different elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).repeat(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: (Optional) An integer, representing the number of times\n   *   the dataset should be repeated. The default behavior (if `count` is\n   *   `undefined` or negative) is for the dataset be repeated indefinitely.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  repeat(count?: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count > 0) {\n      // If this dataset has size and count is positive, new size is current\n      // size multiply count. This also covers the case that current size is\n      // infinity.\n      size = this.size * count;\n    } else if (count === 0) {\n      // If count is 0, new size is 0.\n      size = 0;\n    } else if (this.size != null && (count === undefined || count < 0)) {\n      // If this dataset has size and count is undefined or negative, the\n      // dataset will be repeated indefinitely and new size is infinity.\n      size = Infinity;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      const iteratorIterator = iteratorFromFunction(\n          async () => ({value: await base.iterator(), done: false}));\n      return iteratorFromConcatenated(iteratorIterator.take(count));\n    }, size);\n  }\n\n  /**\n   * Creates a `Dataset` that skips `count` initial elements from this dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).skip(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be skipped\n   *   to form the new dataset.  If `count` is greater than the size of this\n   *   dataset, the new dataset will contain no elements.  If `count`\n   *   is `undefined` or negative, skips the entire dataset.\n   *\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  skip(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count >= 0 && this.size >= count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is current size minus skipped size.This also covers the case that\n      // current size is infinity.\n      size = this.size - count;\n    } else if (\n        this.size != null &&\n        (this.size < count || count === undefined || count < 0)) {\n      // If the size of this dataset is smaller than count, or count is\n      // undefined or negative, skips the entire dataset and the new size is 0.\n      size = 0;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).skip(count), size);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  static readonly MAX_BUFFER_SIZE = 10000;\n\n  /**\n   * Pseudorandomly shuffles the elements of this dataset. This is done in a\n   * streaming manner, by sampling from a given number of prefetched elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).shuffle(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param bufferSize: An integer specifying the number of elements from this\n   *   dataset from which the new dataset will sample.\n   * @param seed: (Optional) An integer specifying the random seed that will\n   *   be used to create the distribution.\n   * @param reshuffleEachIteration: (Optional) A boolean, which if true\n   *   indicates that the dataset should be pseudorandomly reshuffled each time\n   *   it is iterated over. If false, elements will be returned in the same\n   *   shuffled order on each iteration. (Defaults to `true`.)\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  shuffle(bufferSize: number, seed?: string, reshuffleEachIteration = true):\n      Dataset<T> {\n    if (bufferSize == null || bufferSize < 0) {\n      if (this.size == null) {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.');\n      } else {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.  ' +\n            'If your data fits in main memory (for regular JS objects), ' +\n            'and/or GPU memory (for `tf.Tensor`s), consider setting ' +\n            `bufferSize to the dataset size (${this.size} elements)`);\n      }\n    }\n    const base = this;\n    const random = seedrandom.alea(seed || tf.util.now().toString());\n    return datasetFromIteratorFn(async () => {\n      let seed2 = random.int32();\n      if (reshuffleEachIteration) {\n        seed2 += random.int32();\n      }\n      return (await base.iterator()).shuffle(bufferSize, seed2.toString());\n    }, this.size);\n  }\n\n  /**\n   * Creates a `Dataset` with at most `count` initial elements from this\n   * dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).take(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be taken\n   *   to form the new dataset.  If `count` is `undefined` or negative, or if\n   *   `count` is greater than the size of this dataset, the new dataset will\n   *   contain all elements of this dataset.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  take(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && this.size > count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is count.\n      size = count;\n    } else if (this.size != null && this.size <= count) {\n      // If the size of this dataset is equal or smaller than count, the new\n      // dataset's size is the size of this dataset.\n      size = this.size;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).take(count), size);\n  }\n\n  /**\n   * Collect all elements of this dataset into an array.\n   *\n   * Obviously this will succeed only for small datasets that fit in memory.\n   * Useful for testing and generally should be avoided if possible.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]);\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async toArray() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArray();\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   */\n  async toArrayForTest() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArrayForTest();\n  }\n}\n\n/**\n * Create a `Dataset` defined by a provided iterator() function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * const ds = tf.data.datasetFromIteratorFn(iter);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n */\nexport function datasetFromIteratorFn<T extends tf.TensorContainer>(\n    iteratorFn: () => Promise<LazyIterator<T>>,\n    size: number = null): Dataset<T> {\n  return new class extends Dataset<T> {\n    size = size;\n\n    /*\n     * Provide a new stream of elements.  Note this will also start new streams\n     * from any underlying `Dataset`s.\n     */\n    async iterator(): Promise<LazyIterator<T>> {\n      return iteratorFn();\n    }\n  }\n  ();\n}\n\n/**\n * Create a `Dataset` from an array of elements.\n *\n * Create a Dataset from an array of objects:\n * ```js\n * const a = tf.data.array([{'item': 1}, {'item': 2}, {'item': 3}]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n *\n * Create a Dataset from an array of numbers:\n * ```js\n * const a = tf.data.array([4, 5, 6]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n * @param items An array of elements that will be parsed as items in a dataset.\n *\n * @doc {heading: 'Data', subheading: 'Creation', namespace: 'data'}\n */\nexport function array<T extends tf.TensorContainer>(items: T[]): Dataset<T> {\n  return datasetFromIteratorFn(\n      async () => iteratorFromItems(items), items.length);\n}\n\n/**\n * Create a `Dataset` by zipping together an array, dict, or nested\n * structure of `Dataset`s (and perhaps additional constants).\n * The underlying datasets must provide elements in a consistent order such that\n * they correspond.\n *\n * The number of elements in the resulting dataset is the same as the size of\n * the smallest dataset in datasets.\n *\n * The nested structure of the `datasets` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Note this means that, given an array of two datasets that produce dict\n * elements, the result is a dataset that produces elements that are arrays\n * of two dicts:\n *\n * Zip an array of datasets:\n * ```js\n * console.log('Zip two datasets of objects:');\n * const ds1 = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const ds2 = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const ds3 = tf.data.zip([ds1, ds2]);\n * await ds3.forEachAsync(e => console.log(JSON.stringify(e)));\n *\n * // If the goal is to merge the dicts in order to produce elements like\n * // {a: ..., b: ...}, this requires a second step such as:\n * console.log('Merge the objects:');\n * const ds4 = ds3.map(x => {return {a: x[0].a, b: x[1].b}});\n * await ds4.forEachAsync(e => console.log(e));\n * ```\n *\n * Zip a dict of datasets:\n * ```js\n * const a = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const b = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const c = tf.data.zip({c: a, d: b});\n * await c.forEachAsync(e => console.log(JSON.stringify(e)));\n * ```\n *\n * @doc {heading: 'Data', subheading: 'Operations', namespace: 'data'}\n */\nexport function zip<O extends tf.TensorContainer>(datasets: DatasetContainer):\n    Dataset<O> {\n  // manually type-check the argument for JS users\n  if (!isIterable(datasets)) {\n    throw new Error('The argument to zip() must be an object or array.');\n  }\n  let size;\n  if (Array.isArray(datasets)) {\n    for (let i = 0; i < datasets.length; i++) {\n      size = size == null ? (datasets[i] as Dataset<O>).size :\n                            Math.min(size, (datasets[i] as Dataset<O>).size);\n    }\n  } else if (datasets instanceof Object) {\n    for (const ds in datasets) {\n      size = size == null ? (datasets[ds] as Dataset<O>).size :\n                            Math.min(size, (datasets[ds] as Dataset<O>).size);\n    }\n  }\n  return datasetFromIteratorFn<O>(async () => {\n    const streams = await deepMapAndAwaitAll(datasets, d => {\n      if (d instanceof Dataset) {\n        return {value: d.iterator(), recurse: false};\n      } else if (isIterable(d)) {\n        return {value: null, recurse: true};\n      } else {\n        throw new Error(\n            'Leaves of the structure passed to zip() must be Datasets, ' +\n            'not primitives.');\n      }\n    });\n    return iteratorFromZipped<O>(streams, ZipMismatchMode.SHORTEST);\n  }, size);\n}\n\n/**\n * A zip function for use with deepZip, passed via the columnMajorBatch call.\n *\n * Accepts an array of identically-structured nested elements and either batches\n * them (if they are primitives, numeric arrays, or Tensors) or requests\n * recursion (if not).\n */\n// tslint:disable-next-line:no-any\nfunction deepBatchConcat(rows: any[]): DeepMapResult {\n  if (rows === null) {\n    return null;\n  }\n\n  // use the first item to decide whether to recurse or batch here.\n  const exampleRow = rows[0];\n\n  if (canTensorify(exampleRow)) {\n    // rows is an array of primitives, Tensors, or arrays.  Batch them.\n    const value = batchConcat(rows);\n    return {value, recurse: false};\n  }\n\n  // the example row is an object, so recurse into it.\n  return {value: null, recurse: true};\n}\n\n/**\n * Assembles a list of same-shaped numbers, number arrays, or Tensors\n * into a single new Tensor where axis 0 is the batch dimension.\n */\nfunction batchConcat<T extends(TensorLike | tf.Tensor)>(arrays: T[]):\n    tf.Tensor {\n  if (arrays.length === 0) {\n    // We can't return an empty Tensor because we don't know the element shape.\n    throw new Error('Can\\'t make a batch of zero elements.');\n  }\n\n  if (arrays[0] instanceof tf.Tensor) {\n    // Input is an array of Tensors\n    return tf.stack(arrays as tf.Tensor[]);\n  } else {\n    // Input is a possibly-nested array of numbers.\n    return tf.tensor(arrays as TensorLike);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\n\n/**\n * Represents a potentially large collection of text lines.\n *\n * The results are not batched.\n */\nexport class TextLineDataset extends Dataset<string> {\n  /**\n   * Create a `TextLineDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   */\n  constructor(protected readonly input: DataSource) {\n    super();\n  }\n\n  async iterator(): Promise<LazyIterator<string>> {\n    const inputIterator = await this.input.iterator();\n    const utf8Iterator = inputIterator.decodeUTF8();\n    const lineIterator = utf8Iterator.split('\\n').map(line => {\n      // Windows/DOS format text file has extra line breaker at the end of line.\n      if (line.endsWith('\\r')) {\n        line = line.slice(0, -1);\n      }\n      return line;\n    });\n    return lineIterator;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\nimport {ColumnConfig, CSVConfig} from '../types';\nimport {TextLineDataset} from './text_line_dataset';\n\nconst CODE_QUOTE = '\"';\nconst STATE_OUT = Symbol('out');\nconst STATE_FIELD = Symbol('field');\nconst STATE_QUOTE = Symbol('quote');\nconst STATE_QUOTE_AFTER_QUOTE = Symbol('quoteafterquote');\nconst STATE_WITHIN_QUOTE_IN_QUOTE = Symbol('quoteinquote');\n\n/**\n * Represents a potentially large collection of delimited text records.\n *\n * The produced `TensorContainer`s each contain one key-value pair for\n * every column of the table.  When a field is empty in the incoming data, the\n * resulting value is `undefined`, or throw error if it is required.  Values\n * that can be parsed as numbers are emitted as type `number`, other values\n * are parsed as `string`.\n *\n * The results are not batched.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport class CSVDataset extends Dataset<TensorContainer> {\n  base: TextLineDataset;\n  private hasHeader = true;\n  private fullColumnNames: string[] = null;\n  private columnNamesValidated = false;\n  private columnConfigs: {[key: string]: ColumnConfig} = null;\n  private configuredColumnsOnly = false;\n  private delimiter = ',';\n  private delimWhitespace = false;\n\n  /**\n   * Returns column names of the csv dataset. If `configuredColumnsOnly` is\n   * true, return column names in `columnConfigs`. If `configuredColumnsOnly` is\n   * false and `columnNames` is provided, `columnNames`. If\n   * `configuredColumnsOnly` is false and `columnNames` is not provided, return\n   * all column names parsed from the csv file. For example usage please go to\n   * `tf.data.csv`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async columnNames() {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    return this.configuredColumnsOnly ? Object.keys(this.columnConfigs) :\n                                        this.fullColumnNames;\n  }\n\n  /* 1) If `columnNames` is provided as string[], use this string[] as output\n   * keys in corresponding order. The length must match the number of inferred\n   * columns if `hasHeader` is true .\n   * 2) If `columnNames` is not provided, parse header line as `columnNames` if\n   * hasHeader is true. If `hasHeader` is false, throw an error.\n   * 3) If `columnConfigs` is provided, all the keys in `columnConfigs` must\n   * exist in parsed `columnNames`.\n   */\n  private async setColumnNames() {\n    const columnNamesFromFile = await this.maybeReadHeaderLine();\n    if (!this.fullColumnNames && !columnNamesFromFile) {\n      // Throw an error if columnNames is not provided and no header line.\n      throw new Error(\n          'Column names must be provided if there is no header line.');\n    } else if (this.fullColumnNames && columnNamesFromFile) {\n      // Check provided columnNames match header line.\n      util.assert(\n          columnNamesFromFile.length === this.fullColumnNames.length,\n          () => 'The length of provided columnNames (' +\n              this.fullColumnNames.length.toString() +\n              ') does not match the length of the header line read from ' +\n              'file (' + columnNamesFromFile.length.toString() + ').');\n    }\n    if (!this.fullColumnNames) {\n      this.fullColumnNames = columnNamesFromFile;\n    }\n    // Check if there are duplicate column names.\n    const counts: {[key: string]: number} = this.fullColumnNames.reduce(\n        (countAcc: {[key: string]: number}, name) => {\n          countAcc[name] = (countAcc[name] + 1) || 1;\n          return countAcc;\n        },\n        {});\n    const duplicateNames =\n        Object.keys(counts).filter((name) => (counts[name] > 1));\n    util.assert(\n        duplicateNames.length === 0,\n        () => 'Duplicate column names found: ' + duplicateNames.toString());\n    // Check if keys in columnConfigs match columnNames.\n    if (this.columnConfigs) {\n      for (const key of Object.keys(this.columnConfigs)) {\n        const index = this.fullColumnNames.indexOf(key);\n        if (index === -1) {\n          throw new Error(\n              'The key \"' + key +\n              '\" provided in columnConfigs does not match any of the column ' +\n              'names (' + this.fullColumnNames.toString() + ').');\n        }\n      }\n    }\n    this.columnNamesValidated = true;\n  }\n\n  private async maybeReadHeaderLine() {\n    if (this.hasHeader) {\n      const iter = await this.base.iterator();\n      const firstElement = await iter.next();\n      if (firstElement.done) {\n        throw new Error('No data was found for CSV parsing.');\n      }\n      const firstLine: string = firstElement.value;\n      const headers = this.parseRow(firstLine, false);\n      return headers;\n    } else {\n      return null;\n    }\n  }\n\n  /**\n   * Create a `CSVDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   * @param csvConfig (Optional) A CSVConfig object that contains configurations\n   *     of reading and decoding from CSV file(s).\n   *\n   *     hasHeader: (Optional) A boolean value that indicates whether the first\n   *     row of provided CSV file is a header line with column names, and should\n   *     not be included in the data. Defaults to `true`.\n   *\n   *     columnNames: (Optional) A list of strings that corresponds to\n   *     the CSV column names, in order. If provided, it ignores the column\n   *     names inferred from the header row. If not provided, infers the column\n   *     names from the first row of the records. If hasHeader is false and\n   *     columnNames is not provided, this method throws an error.\n   *\n   *     columnConfigs: (Optional) A dictionary whose key is column names, value\n   *     is an object stating if this column is required, column's data type,\n   *     default value, and if this column is label. If provided, keys must\n   *     correspond to names provided in columnNames or inferred from the file\n   *     header lines. If isLabel is true any column, returns an array of two\n   *     items: the first item is a dict of features key/value pairs, the second\n   *     item is a dict of labels key/value pairs. If no feature is marked as\n   *     label, returns a dict of features only.\n   *\n   *     configuredColumnsOnly (Optional) If true, only columns provided in\n   *     columnConfigs will be parsed and provided during iteration.\n   *\n   *     delimiter (Optional) The string used to parse each line of the input\n   *     file. Defaults to `,`.\n   */\n  constructor(protected readonly input: DataSource, csvConfig?: CSVConfig) {\n    super();\n    this.base = new TextLineDataset(input);\n    if (!csvConfig) {\n      csvConfig = {};\n    }\n    this.hasHeader = csvConfig.hasHeader === false ? false : true;\n    this.fullColumnNames = csvConfig.columnNames;\n    this.columnConfigs = csvConfig.columnConfigs;\n    this.configuredColumnsOnly = csvConfig.configuredColumnsOnly;\n    if (csvConfig.delimWhitespace) {\n      util.assert(\n          csvConfig.delimiter == null,\n          () =>\n              'Delimiter should not be provided when delimWhitespace is true.');\n      this.delimWhitespace = true;\n      this.delimiter = ' ';\n    } else {\n      this.delimiter = csvConfig.delimiter ? csvConfig.delimiter : ',';\n    }\n  }\n\n  async iterator(): Promise<LazyIterator<TensorContainer>> {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    let lines = await this.base.iterator();\n    if (this.hasHeader) {\n      // We previously read the first line to get the columnNames.\n      // Now that we're providing data, skip it.\n      lines = lines.skip(1);\n    }\n    return lines.map(x => this.makeDataElement(x));\n  }\n\n  makeDataElement(line: string): TensorContainer {\n    const values = this.parseRow(line);\n    const features: {[key: string]: TensorContainer} = {};\n    const labels: {[key: string]: TensorContainer} = {};\n\n    for (let i = 0; i < this.fullColumnNames.length; i++) {\n      const key = this.fullColumnNames[i];\n      const config = this.columnConfigs ? this.columnConfigs[key] : null;\n      if (this.configuredColumnsOnly && !config) {\n        // This column is not selected.\n        continue;\n      } else {\n        const value = values[i];\n        let parsedValue = null;\n        if (value === '') {\n          // If default value is provided, use it. If default value is not\n          // provided, set as undefined.\n          if (config && config.default !== undefined) {\n            parsedValue = config.default;\n          } else if (config && (config.required || config.isLabel)) {\n            throw new Error(\n                `Required column ${key} is empty in this line: ${line}`);\n          } else {\n            parsedValue = undefined;\n          }\n        } else {\n          // A value is present, so parse it based on type\n          const valueAsNum = Number(value);\n          if (isNaN(valueAsNum)) {\n            // The value is a string and this column is declared as boolean\n            // in config, parse it as boolean.\n            if (config && config.dtype === 'bool') {\n              parsedValue = this.getBoolean(value);\n            } else {\n              // Set value as string\n              parsedValue = value;\n            }\n          } else if (!config || !config.dtype) {\n            // If this value is a number and no type config is provided, return\n            // it as number.\n            parsedValue = valueAsNum;\n          } else {\n            // If this value is a number and data type is provided, parse it\n            // according to provided data type.\n            switch (config.dtype) {\n              case 'float32':\n                parsedValue = valueAsNum;\n                break;\n              case 'int32':\n                parsedValue = Math.floor(valueAsNum);\n                break;\n              case 'bool':\n                parsedValue = this.getBoolean(value);\n                break;\n              default:\n                parsedValue = valueAsNum;\n            }\n          }\n        }\n        // Check if this column is label.\n        (config && config.isLabel) ? labels[key] = parsedValue :\n                                     features[key] = parsedValue;\n      }\n    }\n    // If label exists, return an object of features and labels as {xs:features,\n    // ys:labels}, otherwise return features only.\n    if (Object.keys(labels).length === 0) {\n      return features;\n\n    } else {\n      return {xs: features, ys: labels};\n    }\n  }\n\n  private getBoolean(value: string): number {\n    if (value === '1' || value.toLowerCase() === 'true') {\n      return 1;\n    } else {\n      return 0;\n    }\n  }\n\n  // adapted from https://beta.observablehq.com/@mbostock/streaming-csv\n  private parseRow(line: string, validateElementCount = true): string[] {\n    const result: string[] = [];\n    let readOffset = 0;\n    const readLength = line.length;\n    let currentState = STATE_OUT;\n    // Goes through the line to parse quote.\n    for (let i = 0; i < readLength; i++) {\n      switch (currentState) {\n        // Before enter a new field\n        case STATE_OUT:\n          switch (line.charAt(i)) {\n            // Enter a quoted field\n            case CODE_QUOTE:\n              readOffset = i + 1;\n              currentState = STATE_QUOTE;\n              break;\n            // Read an empty field\n            case this.delimiter:\n              readOffset = i + 1;\n              // If delimiter is white space and configured to collapse\n              // multiple white spaces, ignore this white space.\n              if (this.delimiter === ' ' && this.delimWhitespace) {\n                break;\n              }\n              result.push('');\n              currentState = STATE_OUT;\n              break;\n            // Enter an unquoted field\n            default:\n              currentState = STATE_FIELD;\n              readOffset = i;\n              break;\n          }\n          break;\n        // In an unquoted field\n        case STATE_FIELD:\n          switch (line.charAt(i)) {\n            // Exit an unquoted field, add it to result\n            case this.delimiter:\n              result.push(line.substring(readOffset, i));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            default:\n          }\n          break;\n        // In a quoted field\n        case STATE_QUOTE:\n          switch (line.charAt(i)) {\n            // Read a quote after a quote\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE_AFTER_QUOTE;\n              break;\n            default:\n          }\n          break;\n        // This state means it's right after a second quote in a field\n        case STATE_QUOTE_AFTER_QUOTE:\n          switch (line.charAt(i)) {\n            // Finished a quoted field\n            case this.delimiter:\n              result.push(line.substring(readOffset, i - 1));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            // Finished a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            // In a quoted part in a quoted field\n            default:\n              currentState = STATE_WITHIN_QUOTE_IN_QUOTE;\n              break;\n          }\n          break;\n        case STATE_WITHIN_QUOTE_IN_QUOTE:\n          switch (line.charAt(i)) {\n            // Exit a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            default:\n          }\n          break;\n        default:\n      }\n    }\n    // Adds last item based on if it is quoted.\n    if (currentState === STATE_QUOTE_AFTER_QUOTE) {\n      result.push(line.substring(readOffset, readLength - 1));\n    } else {\n      result.push(line.substring(readOffset));\n    }\n    // Check if each row has the same number of elements as column names.\n    if (validateElementCount && result.length !== this.fullColumnNames.length) {\n      throw new Error(`Invalid row in csv file. Should have ${\n          this.fullColumnNames.length} elements in a row, but got ${result}`);\n    }\n    return result;\n  }\n}\n\n// TODO(soergel): add more basic datasets for parity with tf.data\n// tf.data.FixedLengthRecordDataset()\n// tf.data.TFRecordDataset()\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env, Tensor, tensor, Tensor2D, Tensor3D, TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {MicrophoneConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of tensors from microphone audio stream. The tensors are\n * representing audio data as frequency-domain spectrogram generated with\n * browser's native FFT. Tensors representing time-domain waveform is available\n * based on configuration. Only works in browser environment.\n */\nexport class MicrophoneIterator extends LazyIterator<TensorContainer> {\n  private isClosed = false;\n  private stream: MediaStream;\n  private readonly fftSize: number;\n  private readonly columnTruncateLength: number;\n  private freqData: Float32Array;\n  private timeData: Float32Array;\n  private readonly numFrames: number;\n  private analyser: AnalyserNode;\n  private audioContext: AudioContext;\n  private sampleRateHz: number;\n  private readonly audioTrackConstraints: MediaTrackConstraints;\n  private readonly smoothingTimeConstant: number;\n  private readonly includeSpectrogram: boolean;\n  private readonly includeWaveform: boolean;\n\n  private constructor(protected readonly microphoneConfig: MicrophoneConfig) {\n    super();\n    this.fftSize = microphoneConfig.fftSize || 1024;\n    const fftSizeLog2 = Math.log2(this.fftSize);\n    if (this.fftSize < 0 || fftSizeLog2 < 4 || fftSizeLog2 > 14 ||\n        !Number.isInteger(fftSizeLog2)) {\n      throw new Error(\n          `Invalid fftSize: it must be a power of 2 between ` +\n          `2 to 4 and 2 to 14, but got ${this.fftSize}`);\n    }\n\n    this.numFrames = microphoneConfig.numFramesPerSpectrogram || 43;\n    this.sampleRateHz = microphoneConfig.sampleRateHz;\n    this.columnTruncateLength =\n        microphoneConfig.columnTruncateLength || this.fftSize;\n    this.audioTrackConstraints = microphoneConfig.audioTrackConstraints;\n    this.smoothingTimeConstant = microphoneConfig.smoothingTimeConstant || 0;\n\n    this.includeSpectrogram =\n        microphoneConfig.includeSpectrogram === false ? false : true;\n    this.includeWaveform =\n        microphoneConfig.includeWaveform === true ? true : false;\n    if (!this.includeSpectrogram && !this.includeWaveform) {\n      throw new Error(\n          'Both includeSpectrogram and includeWaveform are false. ' +\n          'At least one type of data should be returned.');\n    }\n  }\n\n  summary() {\n    return `microphone`;\n  }\n\n  // Construct a MicrophoneIterator and start the audio stream.\n  static async create(microphoneConfig: MicrophoneConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'microphone API is only supported in browser environment.');\n    }\n\n    const microphoneIterator = new MicrophoneIterator(microphoneConfig);\n\n    // Call async function start() to initialize the audio stream.\n    await microphoneIterator.start();\n\n    return microphoneIterator;\n  }\n\n  // Start the audio stream and FFT.\n  async start(): Promise<void> {\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        audio: this.audioTrackConstraints == null ? true :\n                                                    this.audioTrackConstraints,\n        video: false\n      });\n    } catch (e) {\n      throw new Error(\n          `Error thrown while initializing video stream: ${e.message}`);\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain audio from microphone.');\n    }\n\n    const ctxConstructor =\n        // tslint:disable-next-line:no-any\n        (window as any).AudioContext || (window as any).webkitAudioContext;\n    this.audioContext = new ctxConstructor();\n\n    if (!this.sampleRateHz) {\n      // If sample rate is not provided, use the available sample rate on\n      // device.\n      this.sampleRateHz = this.audioContext.sampleRate;\n    } else if (this.audioContext.sampleRate !== this.sampleRateHz) {\n      throw new Error(\n          `Mismatch in sampling rate: ` +\n          `Expected: ${this.sampleRateHz}; ` +\n          `Actual: ${this.audioContext.sampleRate}`);\n    }\n\n    const streamSource = this.audioContext.createMediaStreamSource(this.stream);\n    this.analyser = this.audioContext.createAnalyser();\n    this.analyser.fftSize = this.fftSize * 2;\n    this.analyser.smoothingTimeConstant = this.smoothingTimeConstant;\n    streamSource.connect(this.analyser);\n    this.freqData = new Float32Array(this.fftSize);\n    this.timeData = new Float32Array(this.fftSize);\n    return;\n  }\n\n  async next(): Promise<IteratorResult<TensorContainer>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let spectrogramTensor: Tensor;\n    let waveformTensor: Tensor;\n\n    const audioDataQueue = await this.getAudioData();\n    if (this.includeSpectrogram) {\n      const freqData = this.flattenQueue(audioDataQueue.freqDataQueue);\n      spectrogramTensor = this.getTensorFromAudioDataArray(\n          freqData, [this.numFrames, this.columnTruncateLength, 1]);\n    }\n    if (this.includeWaveform) {\n      const timeData = this.flattenQueue(audioDataQueue.timeDataQueue);\n      waveformTensor = this.getTensorFromAudioDataArray(\n          timeData, [this.numFrames * this.fftSize, 1]);\n    }\n\n    return {\n      value: {'spectrogram': spectrogramTensor, 'waveform': waveformTensor},\n      done: false\n    };\n  }\n\n  // Capture one result from the audio stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<{spectrogram: Tensor3D, waveform: Tensor2D}> {\n    return (await this.next()).value as\n        {spectrogram: Tensor3D, waveform: Tensor2D};\n  }\n\n  private async getAudioData():\n      Promise<{freqDataQueue: Float32Array[], timeDataQueue: Float32Array[]}> {\n    const freqDataQueue: Float32Array[] = [];\n    const timeDataQueue: Float32Array[] = [];\n    let currentFrames = 0;\n    return new Promise(resolve => {\n      const intervalID = setInterval(() => {\n        if (this.includeSpectrogram) {\n          this.analyser.getFloatFrequencyData(this.freqData);\n          // If the audio stream is initializing, return empty queue.\n          if (this.freqData[0] === -Infinity) {\n            resolve({freqDataQueue, timeDataQueue});\n          }\n          freqDataQueue.push(this.freqData.slice(0, this.columnTruncateLength));\n        }\n        if (this.includeWaveform) {\n          this.analyser.getFloatTimeDomainData(this.timeData);\n          timeDataQueue.push(this.timeData.slice());\n        }\n\n        // Clean interval and return when all frames have been collected\n        if (++currentFrames === this.numFrames) {\n          clearInterval(intervalID);\n          resolve({freqDataQueue, timeDataQueue});\n        }\n      }, this.fftSize / this.sampleRateHz * 1e3);\n    });\n  }\n\n  // Stop the audio stream and pause the iterator.\n  stop(): void {\n    if (!this.isClosed) {\n      this.isClosed = true;\n      this.analyser.disconnect();\n      this.audioContext.close();\n      if (this.stream != null && this.stream.getTracks().length > 0) {\n        this.stream.getTracks()[0].stop();\n      }\n    }\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor[]> {\n    throw new Error('Can not convert infinite audio stream to array.');\n  }\n\n  // Return audio sampling rate in Hz\n  getSampleRate(): number {\n    return this.sampleRateHz;\n  }\n\n  private flattenQueue(queue: Float32Array[]): Float32Array {\n    const frameSize = queue[0].length;\n    const freqData = new Float32Array(queue.length * frameSize);\n    queue.forEach((data, i) => freqData.set(data, i * frameSize));\n    return freqData;\n  }\n\n  private getTensorFromAudioDataArray(freqData: Float32Array, shape: number[]):\n      Tensor {\n    const vals = new Float32Array(util.sizeFromShape(shape));\n    // If the data is less than the output shape, the rest is padded with zeros.\n    vals.set(freqData, vals.length - freqData.length);\n    return tensor(vals, shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {browser, env, image, tensor1d, Tensor1D, tensor2d, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\nimport {WebcamConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of image tensors from webcam video stream. Only works in\n * browser environment.\n */\nexport class WebcamIterator extends LazyIterator<Tensor3D> {\n  private isClosed = true;\n  private stream: MediaStream;\n  private resize = false;\n  private cropSize: [number, number];\n  private cropBox: Tensor2D;\n  private cropBoxInd: Tensor1D;\n\n  private constructor(\n      protected readonly webcamVideoElement: HTMLVideoElement,\n      protected readonly webcamConfig: WebcamConfig) {\n    super();\n    if (this.needToResize()) {\n      this.resize = true;\n      this.cropSize =\n          [this.webcamConfig.resizeHeight, this.webcamConfig.resizeWidth];\n      this.cropBoxInd = tensor1d([0], 'int32');\n      if (this.webcamConfig.centerCrop) {\n        // Calculate the box based on resizing shape.\n        const widthCroppingRatio =\n            this.webcamConfig.resizeWidth * 1.0 / this.webcamVideoElement.width;\n        const heightCroppingRatio = this.webcamConfig.resizeHeight * 1.0 /\n            this.webcamVideoElement.height;\n        const widthCropStart = (1 - widthCroppingRatio) / 2;\n        const heightCropStart = (1 - heightCroppingRatio) / 2;\n        const widthCropEnd = widthCropStart + widthCroppingRatio;\n        const heightCropEnd = heightCroppingRatio + heightCropStart;\n        this.cropBox = tensor2d(\n            [heightCropStart, widthCropStart, heightCropEnd, widthCropEnd],\n            [1, 4]);\n      } else {\n        this.cropBox = tensor2d([0, 0, 1, 1], [1, 4]);\n      }\n    }\n  }\n\n  summary() {\n    return `webcam`;\n  }\n\n  // Construct a WebcamIterator and start it's video stream.\n  static async create(\n      webcamVideoElement?: HTMLVideoElement, webcamConfig: WebcamConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'tf.data.webcam is only supported in browser environment.');\n    }\n\n    if (!webcamVideoElement) {\n      // If webcam video element is not provided, create a hidden video element\n      // with provided width and height.\n      webcamVideoElement = document.createElement('video');\n      if (!webcamConfig.resizeWidth || !webcamConfig.resizeHeight) {\n        throw new Error(\n            'Please provide webcam video element, or resizeWidth and ' +\n            'resizeHeight to create a hidden video element.');\n      }\n      webcamVideoElement.width = webcamConfig.resizeWidth;\n      webcamVideoElement.height = webcamConfig.resizeHeight;\n    }\n    const webcamIterator = new WebcamIterator(webcamVideoElement, webcamConfig);\n\n    // Call async function to initialize the video stream.\n    await webcamIterator.start();\n\n    return webcamIterator;\n  }\n\n  // Async function to start video stream.\n  async start(): Promise<void> {\n    if (this.webcamConfig.facingMode) {\n      util.assert(\n          (this.webcamConfig.facingMode === 'user') ||\n              (this.webcamConfig.facingMode === 'environment'),\n          () =>\n              `Invalid webcam facing mode: ${this.webcamConfig.facingMode}. ` +\n              `Please provide 'user' or 'environment'`);\n    }\n\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        video: {\n          deviceId: this.webcamConfig.deviceId,\n          facingMode: this.webcamConfig.facingMode ?\n              this.webcamConfig.facingMode :\n              'user',\n          width: this.webcamVideoElement.width,\n          height: this.webcamVideoElement.height\n        }\n      });\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message = `Error thrown while initializing video stream: ${e.message}`;\n      throw e;\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain video from webcam.');\n    }\n\n    // Older browsers may not have srcObject\n    try {\n      this.webcamVideoElement.srcObject = this.stream;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = window.URL.createObjectURL(this.stream);\n    }\n    // Start the webcam video stream\n    this.webcamVideoElement.play();\n\n    this.isClosed = false;\n\n    return new Promise<void>(resolve => {\n      // Add event listener to make sure the webcam has been fully initialized.\n      this.webcamVideoElement.onloadedmetadata = () => {\n        resolve();\n      };\n    });\n  }\n\n  async next(): Promise<IteratorResult<Tensor3D>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let img;\n    try {\n      img = browser.fromPixels(this.webcamVideoElement);\n    } catch (e) {\n      throw new Error(\n          `Error thrown converting video to pixels: ${JSON.stringify(e)}`);\n    }\n    if (this.resize) {\n      try {\n        return {value: this.cropAndResizeFrame(img), done: false};\n      } catch (e) {\n        throw new Error(`Error thrown cropping the video: ${e.message}`);\n      } finally {\n        img.dispose();\n      }\n    } else {\n      return {value: img, done: false};\n    }\n  }\n\n  private needToResize() {\n    // If resizeWidth and resizeHeight are provided, and different from the\n    // width and height of original HTMLVideoElement, then resizing and cropping\n    // is required.\n    if (this.webcamConfig.resizeWidth && this.webcamConfig.resizeHeight &&\n        (this.webcamVideoElement.width !== this.webcamConfig.resizeWidth ||\n         this.webcamVideoElement.height !== this.webcamConfig.resizeHeight)) {\n      return true;\n    }\n    return false;\n  }\n\n  // Cropping and resizing each frame based on config\n  cropAndResizeFrame(img: Tensor3D): Tensor3D {\n    return tidy(() => {\n      const expandedImage: Tensor4D = img.toFloat().expandDims(0);\n      let resizedImage;\n      resizedImage = image.cropAndResize(\n          expandedImage, this.cropBox, this.cropBoxInd, this.cropSize,\n          'bilinear');\n      // Extract image from batch cropping.\n      const shape = resizedImage.shape;\n      return resizedImage.reshape(shape.slice(1) as [number, number, number]);\n    });\n  }\n\n  // Capture one frame from the video stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<Tensor3D> {\n    return (await this.next()).value;\n  }\n\n  // Stop the video stream and pause webcam iterator.\n  stop(): void {\n    const tracks = this.stream.getTracks();\n\n    tracks.forEach(track => track.stop());\n\n    try {\n      this.webcamVideoElement.srcObject = null;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = null;\n    }\n    this.isClosed = true;\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor3D[]> {\n    throw new Error('Can not convert infinite video stream to array.');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {ByteChunkIterator} from './iterators/byte_chunk_iterator';\n\n/**\n * Represents a data source readable as a stream of binary data chunks.\n *\n * Because `Dataset`s can be read repeatedly (via `Dataset.iterator()`), this\n * provides a means to repeatedly create streams from the underlying data\n * sources.\n */\nexport abstract class DataSource {\n  /**\n   * Obtain a new stream of binary data chunks.\n   *\n   * Starts the new stream from the beginning of the data source, even if other\n   * streams have been obtained previously.\n   */\n  abstract async iterator(): Promise<ByteChunkIterator>;\n\n  // TODO(soergel): consider chainable Dataset construction here\n}\n\n// TODO(soergel): consider convenience factory functions here\n// in combination with chainable source->dataset above, e.g.:\n// tf.data.url(...).asCsvDataset().shuffle().batch()\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\n\nexport abstract class StringIterator extends LazyIterator<string> {\n  /**\n   * Splits a string stream on a given separator.\n   *\n   * It is assumed that the incoming chunk boundaries have no semantic meaning,\n   * so conceptually the incoming stream is treated simply as the concatenation\n   * of its elements.\n   *\n   * The outgoing stream provides chunks corresponding to the results of the\n   * standard string split() operation (even if such a chunk spanned incoming\n   * chunks).  The separators are not included.\n   *\n   * A typical usage is to split a text file (represented as a stream with\n   * arbitrary chunk boundaries) into lines.\n   *\n   * @param upstream A readable stream of strings that can be treated as\n   *   concatenated.\n   * @param separator A character to split on.\n   */\n  split(separator: string): StringIterator {\n    return new SplitIterator(this, separator);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on StringIterator.  Unfortunately they can't be placed in separate files, due\n// to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class SplitIterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass SplitIterator extends StringIterator {\n  private impl: SplitIteratorImpl;\n\n  constructor(protected upstream: LazyIterator<string>, separator: string) {\n    super();\n    this.impl = new SplitIteratorImpl(upstream, separator);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\nclass SplitIteratorImpl extends OneToManyIterator<string> {\n  // A partial string at the end of an upstream chunk\n  carryover = '';\n\n  constructor(\n      protected upstream: LazyIterator<string>, protected separator: string) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Split('${this.separator}')`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    if (chunkResult.done) {\n      if (this.carryover === '') {\n        return false;\n      }\n\n      // Pretend that the pump succeeded in order to emit the small last batch.\n      // The next pump() call will actually fail.\n      this.outputQueue.push(this.carryover);\n      this.carryover = '';\n      return true;\n    }\n    const lines = chunkResult.value.split(this.separator);\n    // Note the behavior: \" ab \".split(' ') === ['', 'ab', '']\n    // Thus the carryover may be '' if the separator falls on a chunk\n    // boundary; this produces the correct result.\n\n    lines[0] = this.carryover + lines[0];\n    for (const line of lines.slice(0, -1)) {\n      this.outputQueue.push(line);\n    }\n    this.carryover = lines[lines.length - 1];\n\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\nimport {StringIterator} from './string_iterator';\n\nexport abstract class ByteChunkIterator extends LazyIterator<Uint8Array> {\n  /**\n   * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n   *\n   * The byte arrays producetd from the ByteChunkIterator on which this is\n   * called will be interpreted as concatenated.  No assumptions are made about\n   * the boundaries of the incoming chunks, so a multi-byte UTF8 encoding of a\n   * character may span the boundary between chunks.  This naturally happens,\n   * for instance, when reading fixed-size byte arrays from a file.\n   */\n  decodeUTF8(): StringIterator {\n    return new Utf8Iterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on ByteChunkIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class Utf8Iterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass Utf8Iterator extends StringIterator {\n  private impl: Utf8IteratorImpl;\n\n  constructor(protected upstream: LazyIterator<Uint8Array>) {\n    super();\n    this.impl = new Utf8IteratorImpl(upstream);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\n/**\n * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n *\n * This is tricky because the incoming byte array boundaries may disrupt a\n * multi-byte UTF8 character. Thus any incomplete character data at the end of\n * a chunk must be carried over and prepended to the next chunk before\n * decoding. Luckily with native decoder, TextDecoder in browser and\n * string_decoder in node, byte array boundaries are handled automatically.\n *\n * In the context of an input pipeline for machine learning, UTF8 decoding is\n * needed to parse text files containing training examples or prediction\n * requests (e.g., formatted as CSV or JSON). We cannot use the built-in\n * decoding provided by FileReader.readAsText() because here we are in a\n * streaming context, which FileReader does not support.\n *\n * @param upstream A `LazyIterator` of `Uint8Arrays` containing UTF8-encoded\n *   text, which should be interpreted as concatenated.  No assumptions are\n *   made about the boundaries of the incoming chunks, so a multi-byte UTF8\n *   encoding of a character may span the boundary between chunks.  This\n *   naturally happens, for instance, when reading fixed-size byte arrays from a\n *   file.\n */\nclass Utf8IteratorImpl extends OneToManyIterator<string> {\n  // `decoder` as `any` here to dynamically assign value based on the\n  // environment.\n  // tslint:disable-next-line:no-any\n  decoder: any;\n\n  constructor(protected readonly upstream: LazyIterator<Uint8Array>) {\n    super();\n    if (env().get('IS_BROWSER')) {\n      this.decoder = new TextDecoder('utf-8');\n    } else {\n      // tslint:disable-next-line:no-require-imports\n      const {StringDecoder} = require('string_decoder');\n      this.decoder = new StringDecoder('utf8');\n    }\n  }\n  summary() {\n    return `${this.upstream.summary()} -> Utf8`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    let chunk;\n    if (chunkResult.done) {\n      return false;\n    } else {\n      chunk = chunkResult.value;\n    }\n\n    let text: string;\n    if (env().get('IS_BROWSER')) {\n      text = this.decoder.decode(chunk, {stream: true});\n    } else {\n      text = this.decoder.write(Buffer.from(chunk.buffer));\n    }\n    this.outputQueue.push(text);\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// inspired by https://github.com/maxogden/filereader-stream\nimport {env, util} from '@tensorflow/tfjs-core';\nimport {FileElement} from '../types';\nimport {ByteChunkIterator} from './byte_chunk_iterator';\n\nexport interface FileChunkIteratorOptions {\n  /** The byte offset at which to begin reading the File or Blob. Default 0. */\n  offset?: number;\n  /** The number of bytes to read at a time. Default 1MB. */\n  chunkSize?: number;\n}\n\n/**\n * Provide a stream of chunks from a File, Blob, or Uint8Array.\n * @param file The source File, Blob or Uint8Array.\n * @param options Optional settings controlling file reading.\n * @returns a lazy Iterator of Uint8Arrays containing sequential chunks of the\n *   input File, Blob or Uint8Array.\n */\nexport class FileChunkIterator extends ByteChunkIterator {\n  offset: number;\n  chunkSize: number;\n\n  constructor(\n      protected file: FileElement,\n      protected options: FileChunkIteratorOptions = {}) {\n    super();\n    util.assert(\n        (file instanceof Uint8Array) ||\n            (env().get('IS_BROWSER') ?\n                 (file instanceof File || file instanceof Blob) :\n                 false),\n        () => 'FileChunkIterator only supports File, Blob and Uint8Array ' +\n            'right now.');\n    this.offset = options.offset || 0;\n    // default 1MB chunk has tolerable perf on large files\n    this.chunkSize = options.chunkSize || 1024 * 1024;\n  }\n\n  summary() {\n    return `FileChunks ${this.file}`;\n  }\n\n  async next(): Promise<IteratorResult<Uint8Array>> {\n    if (this.offset >= ((this.file instanceof Uint8Array) ?\n                            this.file.byteLength :\n                            this.file.size)) {\n      return {value: null, done: true};\n    }\n    const chunk = new Promise<Uint8Array>((resolve, reject) => {\n      const end = this.offset + this.chunkSize;\n      if (this.file instanceof Uint8Array) {\n        // Note if end > this.uint8Array.byteLength, we just get a small last\n        // chunk.\n        resolve(new Uint8Array(this.file.slice(this.offset, end)));\n      } else {\n        // This branch assumes that this.file type is File or Blob, which\n        // means it is in the browser environment.\n\n        // TODO(soergel): is this a performance issue?\n        const fileReader = new FileReader();\n        fileReader.onload = (event) => {\n          let data: string|ArrayBuffer|Uint8Array = fileReader.result;\n          // Not sure we can trust the return type of\n          // FileReader.readAsArrayBuffer See e.g.\n          // https://github.com/node-file-api/FileReader/issues/2\n          if (data instanceof ArrayBuffer) {\n            data = new Uint8Array(data);\n          }\n          if (!(data instanceof Uint8Array)) {\n            return reject(new TypeError('FileReader returned unknown type.'));\n          }\n          resolve(data);\n        };\n        fileReader.onabort = (event) => {\n          return reject(new Error('Aborted'));\n        };\n        fileReader.onerror = (event) => {\n          return reject(new Error(event.type));\n        };\n        // TODO(soergel): better handle onabort, onerror\n        // Note if end > this.file.size, we just get a small last chunk.\n        const slice = this.file.slice(this.offset, end);\n        // We can't use readAsText here (even if we know the file is text)\n        // because the slice boundary may fall within a multi-byte character.\n        fileReader.readAsArrayBuffer(slice);\n      }\n      this.offset = end;\n    });\n    return {value: (await chunk), done: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\nimport {FileChunkIterator, FileChunkIteratorOptions} from './file_chunk_iterator';\n\n/**\n * Provide a stream of chunks from a URL.\n *\n * Note this class first downloads the entire file into memory before providing\n * the first element from the stream.  This is because the Fetch API does not\n * yet reliably provide a reader stream for the response body.\n */\nexport async function urlChunkIterator(\n    url: RequestInfo, options: FileChunkIteratorOptions = {}) {\n  let urlString;\n  let requestInit;\n  if ((typeof url) === 'string') {\n    urlString = url as string;\n  } else {\n    urlString = (url as Request).url;\n    requestInit = getRequestInitFromRequest(url as Request);\n  }\n  const response = await util.fetch(urlString, requestInit);\n  if (response.ok) {\n    const uint8Array = new Uint8Array(await response.arrayBuffer());\n    return new FileChunkIterator(uint8Array, options);\n  } else {\n    throw new Error(response.statusText);\n  }\n}\n\n// Generate RequestInit from Request to match tf.util.fetch signature.\nconst getRequestInitFromRequest = (request: Request) => {\n  const init = {\n    method: request.method,\n    headers: request.headers,\n    body: request.body,\n    mode: request.mode,\n    credentials: request.credentials,\n    cache: request.cache,\n    redirect: request.redirect,\n    referrer: request.referrer,\n    integrity: request.integrity,\n  };\n  return init;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// Skip tslint any type check cause this method is aiming to check type of\n// input.\n// tslint:disable-next-line:no-any\nexport function isLocalPath(source: any): boolean {\n  return (typeof source === 'string') && source.substr(0, 7) === 'file://';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIterator, FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {FileElement} from '../types';\nimport {isLocalPath} from '../util/source_util';\n\n/**\n * Represents a file, blob, or Uint8Array readable as a stream of binary data\n * chunks.\n */\nexport class FileDataSource extends DataSource {\n  /**\n   * Create a `FileDataSource`.\n   *\n   * @param input Local file path, or `File`/`Blob`/`Uint8Array` object to\n   *     read. Local file only works in node environment.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected input: FileElement|string,\n      protected readonly options: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.input) && env().get('IS_NODE')) {\n      // tslint:disable-next-line:no-require-imports\n      const fs = require('fs');\n      this.input = fs.readFileSync((this.input as string).substr(7));\n    }\n    // TODO(kangyizhang): Add LocalFileChunkIterator to split local streaming\n    // with file in browser.\n    return new FileChunkIterator(this.input as FileElement, this.options);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {urlChunkIterator} from '../iterators/url_chunk_iterator';\nimport {isLocalPath} from '../util/source_util';\nimport {FileDataSource} from './file_data_source';\n\n/*\n * Represents a URL readable as a stream of binary data chunks.\n */\nexport class URLDataSource extends DataSource {\n  /**\n   * Create a `URLDataSource`.\n   *\n   * @param url A source URL string, or a `Request` object.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected readonly url: RequestInfo,\n      protected readonly fileOptions: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  // TODO(soergel): provide appropriate caching options.  Currently this\n  // will download the URL anew for each call to iterator().  Since we have\n  // to treat the downloaded file as a blob/buffer anyway, we may as well retain\n  // it-- but that raises GC issues.  Also we may want a persistent disk cache.\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.url)) {\n      return (new FileDataSource(this.url as string, this.fileOptions))\n          .iterator();\n    } else {\n      return urlChunkIterator(this.url, this.fileOptions);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer} from '@tensorflow/tfjs-core';\nimport {Dataset, datasetFromIteratorFn} from './dataset';\nimport {CSVDataset} from './datasets/csv_dataset';\nimport {iteratorFromFunction} from './iterators/lazy_iterator';\nimport {MicrophoneIterator} from './iterators/microphone_iterator';\nimport {WebcamIterator} from './iterators/webcam_iterator';\nimport {URLDataSource} from './sources/url_data_source';\nimport {CSVConfig, MicrophoneConfig, WebcamConfig} from './types';\n\n/**\n * Create a `CSVDataset` by reading and decoding CSV file(s) from provided URL\n * or local path if it's in Node environment.\n *\n * Note: If isLabel in columnConfigs is `true` for at least one column, the\n * element in returned `CSVDataset` will be an object of\n * `{xs:features, ys:labels}`: xs is a dict of features key/value pairs, ys\n * is a dict of labels key/value pairs. If no column is marked as label,\n * returns a dict of features only.\n *\n * ```js\n * const csvUrl =\n * 'https://storage.googleapis.com/tfjs-examples/multivariate-linear-regression/data/boston-housing-train.csv';\n *\n * async function run() {\n *   // We want to predict the column \"medv\", which represents a median value of\n *   // a home (in $1000s), so we mark it as a label.\n *   const csvDataset = tf.data.csv(\n *     csvUrl, {\n *       columnConfigs: {\n *         medv: {\n *           isLabel: true\n *         }\n *       }\n *     });\n *\n *   // Number of features is the number of column names minus one for the label\n *   // column.\n *   const numOfFeatures = (await csvDataset.columnNames()).length - 1;\n *\n *   // Prepare the Dataset for training.\n *   const flattenedDataset =\n *     csvDataset\n *     .map(({xs, ys}) =>\n *       {\n *         // Convert xs(features) and ys(labels) from object form (keyed by\n *         // column name) to array form.\n *         return {xs:Object.values(xs), ys:Object.values(ys)};\n *       })\n *     .batch(10);\n *\n *   // Define the model.\n *   const model = tf.sequential();\n *   model.add(tf.layers.dense({\n *     inputShape: [numOfFeatures],\n *     units: 1\n *   }));\n *   model.compile({\n *     optimizer: tf.train.sgd(0.000001),\n *     loss: 'meanSquaredError'\n *   });\n *\n *   // Fit the model using the prepared Dataset\n *   return model.fitDataset(flattenedDataset, {\n *     epochs: 10,\n *     callbacks: {\n *       onEpochEnd: async (epoch, logs) => {\n *         console.log(epoch + ':' + logs.loss);\n *       }\n *     }\n *   });\n * }\n *\n * await run();\n * ```\n *\n * @param source URL or local path to get CSV file. If it's a local path, it\n * must have prefix `file://` and it only works in node environment.\n * @param csvConfig (Optional) A CSVConfig object that contains configurations\n *     of reading and decoding from CSV file(s).\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function csv(\n    source: RequestInfo, csvConfig: CSVConfig = {}): CSVDataset {\n  return new CSVDataset(new URLDataSource(source), csvConfig);\n}\n\n/**\n * Create a `Dataset` that produces each element by calling a provided function.\n *\n * Note that repeated iterations over this `Dataset` may produce different\n * results, because the function will be called anew for each element of each\n * iteration.\n *\n * Also, beware that the sequence of calls to this function may be out of order\n * in time with respect to the logical order of the Dataset. This is due to the\n * asynchronous lazy nature of stream processing, and depends on downstream\n * transformations (e.g. .shuffle()). If the provided function is pure, this is\n * no problem, but if it is a closure over a mutable state (e.g., a traversal\n * pointer), then the order of the produced elements may be scrambled.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const ds = tf.data.func(func);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param f A function that produces one data element on each call.\n */\nexport function func<T extends TensorContainer>(\n    f: () => IteratorResult<T>| Promise<IteratorResult<T>>): Dataset<T> {\n  const iter = iteratorFromFunction(f);\n  return datasetFromIteratorFn(async () => iter);\n}\n\n/**\n * Create a `Dataset` that produces each element from provided JavaScript\n * generator, which is a function*\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions),\n * or a function that returns an\n * iterator\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions).\n *\n * The returned iterator should have `.next()` function that returns element in\n * format of `{value: TensorContainer, done:boolean}`.\n *\n * Example of creating a dataset from an iterator factory:\n * ```js\n * function makeIterator() {\n *   const numElements = 10;\n *   let index = 0;\n *\n *   const iterator = {\n *     next: () => {\n *       let result;\n *       if (index < numElements) {\n *         result = {value: index, done: false};\n *         index++;\n *         return result;\n *       }\n *       return {value: index, done: true};\n *     }\n *   };\n *   return iterator;\n * }\n * const ds = tf.data.generator(makeIterator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * Example of creating a dataset from a generator:\n * ```js\n * function* dataGenerator() {\n *   const numElements = 10;\n *   let index = 0;\n *   while (index < numElements) {\n *     const x = index;\n *     index++;\n *     yield x;\n *   }\n * }\n *\n * const ds = tf.data.generator(dataGenerator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param generator A Javascript generator function that returns a JavaScript\n *     iterator.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function generator<T extends TensorContainer>(\n    generator: () => Iterator<T>| Promise<Iterator<T>>): Dataset<T> {\n  return datasetFromIteratorFn(async () => {\n    const gen = await generator();\n    return iteratorFromFunction(() => gen.next());\n  });\n}\n\n/**\n * Create an iterator that generate `Tensor`s from webcam video stream. This API\n * only works in Browser environment when the device has webcam.\n *\n * Note: this code snippet only works when the device has a webcam. It will\n * request permission to open the webcam when running.\n * ```js\n * const videoElement = document.createElement('video');\n * videoElement.width = 100;\n * videoElement.height = 100;\n * const cam = await tf.data.webcam(videoElement);\n * const img = await cam.capture();\n * img.print();\n * cam.stop();\n * ```\n *\n * @param webcamVideoElement A `HTMLVideoElement` used to play video from\n *     webcam. If this element is not provided, a hidden `HTMLVideoElement` will\n *     be created. In that case, `resizeWidth` and `resizeHeight` must be\n *     provided to set the generated tensor shape.\n * @param webcamConfig A `WebcamConfig` object that contains configurations of\n *     reading and manipulating data from webcam video stream.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function webcam(\n    webcamVideoElement?: HTMLVideoElement,\n    webcamConfig?: WebcamConfig): Promise<WebcamIterator> {\n  return WebcamIterator.create(webcamVideoElement, webcamConfig);\n}\n\n/**\n * Create an iterator that generate frequency-domain spectrogram `Tensor`s from\n * microphone audio stream with browser's native FFT. This API only works in\n * browser environment when the device has microphone.\n *\n * Note: this code snippet only works when the device has a microphone. It will\n * request permission to open the microphone when running.\n * ```js\n * const mic = await tf.data.microphone({\n *   fftSize: 1024,\n *   columnTruncateLength: 232,\n *   numFramesPerSpectrogram: 43,\n *   sampleRateHz:44100,\n *   includeSpectrogram: true,\n *   includeWaveform: true\n * });\n * const audioData = await mic.capture();\n * const spectrogramTensor = audioData.spectrogram;\n * spectrogramTensor.print();\n * const waveformTensor = audioData.waveform;\n * waveformTensor.print();\n * mic.stop();\n * ```\n *\n * @param microphoneConfig A `MicrophoneConfig` object that contains\n *     configurations of reading audio data from microphone.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function microphone(microphoneConfig?: MicrophoneConfig):\n    Promise<MicrophoneIterator> {\n  return MicrophoneIterator.create(microphoneConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${\n              opName} does not support complex64 tensors in the CPU backend.`);\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {backend_util, BackendTimingInfo, DataStorage, DataType, DataValues, engine, env, kernel_impls, KernelBackend, max, NumericDataType, Rank, Scalar, ShapeMap, slice_util, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, TensorInfo, TypedArray, upcastType, util} from '@tensorflow/tfjs-core';\n\nconst nonMaxSuppressionV3Impl = kernel_impls.nonMaxSuppressionV3Impl;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\nimport * as seedrandom from 'seedrandom';\nimport {assertNotComplex} from './cpu_util';\n\ninterface DataId {}\n\nfunction mapActivation(\n    backend: MathBackendCPU, x: Tensor, activation: backend_util.Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return backend.linear(x);\n  } else if (activation === 'relu') {\n    return backend.relu(x);\n  } else if (activation === 'elu') {\n    return tf.elu(x);\n  } else if (activation === 'relu6') {\n    return backend.relu6(x);\n  } else if (activation === 'prelu') {\n    return backend.prelu(x, preluActivationWeights);\n  }\n  throw new Error(\n      `Activation ${activation} has not been implemented for the CPU backend.`);\n}\n\nexport interface TensorData<D extends DataType> {\n  values?: backend_util.BackendValues;\n  dtype: D;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensorInfos field.\n  complexTensorInfos?: {real: TensorInfo, imag: TensorInfo};\n  // refCount keeps track of how many tensors reference it. Used for memory\n  // management.\n  refCount: number;\n}\n\nexport class MathBackendCPU extends KernelBackend {\n  public blockSize = 48;\n\n  data: DataStorage<TensorData<DataType>>;\n  private firstUse = true;\n\n  constructor() {\n    super();\n    this.data = new DataStorage(this, engine());\n  }\n\n  write(values: backend_util.BackendValues, shape: number[], dtype: DataType):\n      DataId {\n    if (this.firstUse) {\n      this.firstUse = false;\n      if (env().get('IS_NODE')) {\n        backend_util.warn(\n            '\\n============================\\n' +\n            'Hi there \uD83D\uDC4B. Looks like you are running TensorFlow.js in ' +\n            'Node.js. To speed things up dramatically, install our node ' +\n            'backend, which binds to TensorFlow C++, by running ' +\n            'npm i @tensorflow/tfjs-node, ' +\n            'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +\n            'Then call require(\\'@tensorflow/tfjs-node\\'); (-gpu ' +\n            'suffix for CUDA) at the start of your program. ' +\n            'Visit https://github.com/tensorflow/tfjs-node for more details.' +\n            '\\n============================');\n      }\n    }\n    const dataId = {};\n\n    this.data.set(dataId, {values, dtype, refCount: 1});\n\n    return dataId;\n  }\n\n  /**\n   * Create a data bucket in cpu backend.\n   * @param shape Shape of the `TensorInfo`.\n   * @param dtype DType of the `TensorInfo`.\n   * @param values The value of the `TensorInfo` stored as a flattened array.\n   */\n  makeTensorInfo(\n      shape: number[], dtype: DataType,\n      values?: backend_util.BackendValues): TensorInfo {\n    const outId = this.write(values, shape, dtype);\n\n    return {dataId: outId, shape, dtype};\n  }\n\n  /** Increase refCount of a `TensorData`. */\n  incRef(dataId: DataId): void {\n    const tensorData = this.data.get(dataId);\n    tensorData.refCount++;\n  }\n\n  /** Decrease refCount of a `TensorData`. */\n  decRef(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n      tensorData.refCount--;\n    }\n  }\n\n  move(\n      dataId: DataId, values: backend_util.BackendValues, shape: number[],\n      dtype: DataType): void {\n    this.data.set(dataId, {values, dtype, refCount: 1});\n  }\n\n  numDataIds(): number {\n    return this.data.numDataIds();\n  }\n\n  async read(dataId: DataId): Promise<backend_util.BackendValues> {\n    return this.readSync(dataId);\n  }\n  readSync(dataId: DataId): backend_util.BackendValues {\n    const {dtype, complexTensorInfos} = this.data.get(dataId);\n\n    if (dtype === 'complex64') {\n      const realValues =\n          this.readSync(complexTensorInfos.real.dataId) as Float32Array;\n      const imagValues =\n          this.readSync(complexTensorInfos.imag.dataId) as Float32Array;\n      return backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    }\n\n    return this.data.get(dataId).values;\n  }\n\n  private bufferSync<R extends Rank>(t: Tensor<R>): TensorBuffer<R> {\n    const data = this.readSync(t.dataId);\n    let decodedData = data as DataValues;\n    if (t.dtype === 'string') {\n      try {\n        // Decode the bytes into string.\n        decodedData = (data as Uint8Array[]).map(d => util.decodeString(d));\n      } catch {\n        throw new Error('Failed to decode encoded string bytes into utf-8');\n      }\n    }\n    return tf.buffer(t.shape, t.dtype, decodedData) as TensorBuffer<R>;\n  }\n\n  makeOutput<T extends Tensor>(\n      values: backend_util.BackendValues, shape: number[], dtype: DataType): T {\n    const dataId = this.write(values, shape, dtype);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  disposeData(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const {complexTensorInfos} = this.data.get(dataId);\n\n      if (complexTensorInfos != null) {\n        this.disposeData(complexTensorInfos.real.dataId);\n        this.disposeData(complexTensorInfos.imag.dataId);\n      }\n\n      this.data.delete(dataId);\n    }\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n\n      tensorData.refCount--;\n\n      if (tensorData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  async time(f: () => void): Promise<BackendTimingInfo> {\n    const start = util.now();\n    f();\n    const kernelMs = util.now() - start;\n    return {kernelMs};\n  }\n\n  memory() {\n    return {\n      // Unreliable due to automatic gc. The numbers above are cumulative.\n      unreliable: true,\n      reasons:\n          ['The reported memory is an upper bound. Due to automatic garbage ' +\n           'collection, the true allocated memory may be less.']\n    };\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    assertNotComplex(x, 'stridedSlice');\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tf.tensor([], outShape) as T;\n    }\n\n    const buffer = tf.buffer(outShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n    for (let i = 0; i < buffer.size; i++) {\n      const loc = buffer.indexToLoc(i);\n\n      const newLoc: number[] = new Array(loc.length);\n      for (let j = 0; j < newLoc.length; j++) {\n        newLoc[j] = loc[j] * strides[j] + begin[j];\n      }\n      buffer.set(xBuf.get(...newLoc), ...loc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  diag(x: Tensor): Tensor {\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const buffer = tf.buffer([x.size, x.size], x.dtype);\n    const vals = buffer.values;\n    for (let i = 0; i < xVals.length; i++) {\n      vals[i * x.size + i] = xVals[i];\n    }\n    return buffer.toTensor();\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = tf.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    assertNotComplex(x, 'reverse');\n\n    const buffer = tf.buffer(x.shape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < buffer.size; i++) {\n      const outLoc = buffer.indexToLoc(i);\n      const inLoc = outLoc.slice();\n      axis.forEach(ax => inLoc[ax] = x.shape[ax] - 1 - inLoc[ax]);\n      buffer.set(xBuf.get(...inLoc), ...outLoc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'neg');\n\n    // TODO(lina128): Use mul directly once neg is modularized.\n    return tf.mul(tf.scalar(-1), x);\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    assertNotComplex(tensors, 'addN');\n\n    const vals = tensors.map(t => this.readSync(t.dataId) as TypedArray);\n    const result = tf.buffer(tensors[0].shape, tensors[0].dtype as 'float32');\n    const resultVals = result.values;\n    for (let i = 0; i < tensors.length; i++) {\n      const currVals = vals[i];\n      for (let j = 0; j < resultVals.length; j++) {\n        resultVals[j] += currVals[j];\n      }\n    }\n    return result.toTensor() as T;\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n\n    // TODO(lina128): Use sub directly once softmax is modularized.\n    const a = tf.sub(logits, maxLogit.reshape(expandedShape));\n    const b = tf.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax\n    // kernel modularization.\n    return tf.div(b, sumExp);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    assertNotComplex([a, b], 'pow');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.pow(aValue, bValue)) as\n        T;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    assertNotComplex([a, b], 'matMul');\n\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const leftDim = transposeA ? a.shape[2] : a.shape[1];\n    const rightDim = transposeB ? b.shape[1] : b.shape[2];\n    const batchDim = a.shape[0];\n\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const [aBatch, aOuterStep, aInnerStep] = transposeA ?\n        [a.strides[0], 1, a.strides[1]] :\n        [a.strides[0], a.strides[1], 1];\n    const [bInnerStep, bOuterStep, bBatch] = transposeB ?\n        [1, b.strides[1], b.strides[0]] :\n        [b.strides[1], 1, b.strides[0]];\n\n    const size = leftDim * rightDim;\n    const result = tf.buffer([batchDim, leftDim, rightDim], a.dtype);\n    const resVals = result.values as TypedArray;\n    const blockSize = this.blockSize;\n\n    for (let b = 0; b < batchDim; b++) {\n      for (let i0 = 0; i0 < leftDim; i0 += blockSize) {\n        for (let j0 = 0; j0 < rightDim; j0 += blockSize) {\n          for (let k0 = 0; k0 < sharedDim; k0 += blockSize) {\n            // for when blockSize doesn't evenly divide the input\n            const iBlock = Math.min(i0 + blockSize, leftDim);\n            const jBlock = Math.min(j0 + blockSize, rightDim);\n            const kBlock = Math.min(k0 + blockSize, sharedDim);\n\n            for (let i = i0; i < iBlock; i++) {\n              for (let j = j0; j < jBlock; j++) {\n                let sum = 0.0;\n\n                for (let k = k0; k < kBlock; k++) {\n                  sum += aValues[b * aBatch + i * aOuterStep + k * aInnerStep] *\n                      bValues[k * bInnerStep + j * bOuterStep + b * bBatch];\n                }\n                resVals[b * size + (i * rightDim + j)] += sum;\n              }\n            }\n          }\n        }\n      }\n    }\n    return result.toTensor() as Tensor3D;\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    let result = this.batchMatMul(a, b, transposeA, transposeB);\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedBatchMatMul is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor3D;\n    }\n\n    return result;\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'floorDiv');\n\n    const op = (a: number, b: number) => Math.floor(a / b);\n    const outputDtype = 'int32';\n    return this.broadcastedBinaryOp(a, b, outputDtype, op);\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let sum = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        sum += aVals[offset + j];\n      }\n      vals[i] = sum;\n    }\n    return result;\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let prod = 1;\n      for (let j = 0; j < reduceSize; ++j) {\n        prod *= aVals[offset + j];\n      }\n      vals[i] = prod;\n    }\n    return result;\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    assertNotComplex(x, 'unsortedSegmentSum');\n\n    const res = [];\n\n    // Reshape the segment id's so that they can be broadcast with\n    // x. The new shape should be [segmentIds.shape, 1, ..., 1]\n    const numIters = x.rank - segmentIds.rank;\n    for (let i = 0; i < numIters; ++i) {\n      segmentIds = segmentIds.expandDims(i + 1);\n    }\n\n    for (let i = 0; i < numSegments; ++i) {\n      const segmentId = tf.scalar(i, 'int32');\n      const mask = tf.equal(segmentId, segmentIds).asType('float32');\n      const sum = mask.mul(x).sum(0);\n      res.push(sum);\n    }\n\n    return tf.stack(res);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMin');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMin', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      let minIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n          minIndex = j;\n        }\n      }\n      vals[i] = minIndex;\n    }\n    return result;\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMax');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMax', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let max = aVals[offset];\n      let maxIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value > max) {\n          max = value;\n          maxIndex = j;\n        }\n      }\n      vals[i] = maxIndex;\n    }\n    return result;\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    assertNotComplex(x, 'cumsum');\n\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `backend.cumsum in CPU expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(x.shape, resultDtype);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    const finalDim = x.shape[x.rank - 1];\n    const indexAdjuster = reverse ?\n        (i: number, j: number) => i + finalDim - j - 1 :\n        (i: number, j: number) => i + j;\n    for (let i = 0; i < aVals.length; i += finalDim) {\n      for (let j = 0; j < finalDim; j++) {\n        const idx = indexAdjuster(i, j);\n        if (j === 0) {\n          vals[idx] = exclusive ? 0 : aVals[idx];\n        } else {\n          const prevIdx = indexAdjuster(i, j - 1);\n          vals[idx] = exclusive ? aVals[prevIdx] + vals[prevIdx] :\n                                  aVals[idx] + vals[prevIdx];\n        }\n      }\n    }\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'equal');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal === bVal) ? 1 : 0;\n    });\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'notEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal !== bVal) ? 1 : 0;\n    });\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'less');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal < bVal) ? 1 : 0;\n    });\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'lessEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal <= bVal) ? 1 : 0;\n    });\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greater');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal > bVal) ? 1 : 0;\n    });\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greaterEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal >= bVal) ? 1 : 0;\n    });\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalAnd');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal && bVal;\n    });\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalOr');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal || bVal;\n    });\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([condition, a, b], 'select');\n\n    const values = this.readSync(condition.dataId) as TypedArray;\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const result = tf.zeros(a.shape, upcastType(a.dtype, b.dtype));\n    const newValues = this.readSync(result.dataId) as TypedArray;\n    let index = 0;\n    const offset = condition.rank === 0 || condition.rank > 1 || a.rank === 1 ?\n        1 :\n        util.sizeFromShape(a.shape.slice(1));\n\n    for (let i = 0; i < values.length; i++) {\n      for (let j = 0; j < offset; j++) {\n        if (values[i] === 1) {\n          newValues[index++] = aValues[i];\n        } else {\n          newValues[index++] = bValues[i];\n        }\n      }\n    }\n\n    return result;\n  }\n\n  where(condition: Tensor): Tensor2D {\n    assertNotComplex([condition], 'where');\n\n    const condVals = this.readSync(condition.dataId) as TypedArray;\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    assertNotComplex(x, 'topk');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'min');\n\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n        }\n      }\n      vals[i] = min;\n    }\n    return result;\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'minimum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.min(aVal, bVal));\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'mod');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const rem = aVal % bVal;\n      if ((aVal < 0 && bVal < 0) || (aVal >= 0 && bVal >= 0)) {\n        return rem;\n      } else {\n        return (rem + bVal) % bVal;\n      }\n    });\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'maximum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.max(aVal, bVal));\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'all');\n\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let all = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        all = all && value;\n      }\n      vals[i] = all;\n    }\n    return result;\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'any');\n\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let anyVal = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        anyVal = anyVal || value;\n      }\n      vals[i] = anyVal;\n    }\n    return result;\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'squaredDifference');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const diff = aVal - bVal;\n      return diff * diff;\n    });\n  }\n\n  linear<T extends Tensor>(x: T): T {\n    return x;\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.max(0, inVals[i]);\n    }\n    return res as T;\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.min(Math.max(0, inVals[i]), 6);\n    }\n    return res as T;\n  }\n\n  prelu<T extends Tensor>(x: T, a: T): T {\n    assertNotComplex([x, a], 'prelu');\n\n    return this.broadcastedBinaryOp(\n               x, a, x.dtype,\n               (xValue, aValue) => xValue < 0 ? aValue * xValue : xValue) as T;\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    assertNotComplex([dy, y], 'eluDer');\n\n    const resultValues = new Float32Array(y.size);\n    const values = this.readSync(y.dataId) as TypedArray;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    for (let i = 0; i < values.length; ++i) {\n      const v = values[i];\n      if (v >= 1) {\n        resultValues[i] = dyValues[i];\n      } else {\n        resultValues[i] = dyValues[i] * (v + 1);\n      }\n    }\n    return this.makeOutput(resultValues, y.shape, 'float32');\n  }\n\n  atan2<T extends Tensor>(a: T, b: T): T {\n    assertNotComplex([a, b], 'atan2');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.atan2(aValue, bValue)) as\n        T;\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.conv2d(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedConv2d is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, filter], 'conv2d');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n\n    const xBatchStride = x.strides[0];\n    const xRowStride = isChannelsLast ? x.strides[1] : x.strides[2];\n    const xColStride = isChannelsLast ? x.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : x.strides[1];\n    const yBatchStride = y.strides[0];\n    const yRowStride = isChannelsLast ? y.strides[1] : y.strides[2];\n    const yColStride = isChannelsLast ? y.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : y.strides[1];\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * xBatchStride;\n      const yOffset1 = b * yBatchStride;\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * yRowStride;\n        const xRCorner = yR * convInfo.strideHeight - padTop;\n        for (let wR = 0; wR < filterHeight; wR++) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * xRowStride;\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * yColStride;\n            const xCCorner = yC * convInfo.strideWidth - padLeft;\n            for (let wC = 0; wC < filterWidth; wC++) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * xColStride;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1 * xChannelStride];\n                for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                  yVals[yOffset3 + d2 * yChannelStride] +=\n                      xVal * wVals[wOffset3 + d2];\n                }\n                wOffset3 += convInfo.outChannels;\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor() as Tensor4D;\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padFront = convInfo.padInfo.front;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const y = tf.buffer<Rank.R5>(convInfo.outShape, x.dtype as 'float32');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yF = 0; yF < convInfo.outDepth; ++yF) {\n        const yOffset2 = yOffset1 + yF * y.strides[1];\n        const xFCorner = yF * convInfo.strideDepth - padFront;\n        for (let wF = 0; wF < filterDepth; wF++) {\n          const xF = xFCorner + wF * dilationDepth;\n          if (xF < 0 || xF >= convInfo.inDepth) {\n            continue;\n          }\n          const wOffset1 = wF * filter.strides[0];\n          const xOffset2 = xOffset1 + xF * x.strides[1];\n\n          for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n            const yOffset3 = yOffset2 + yR * y.strides[2];\n            const xRCorner = yR * convInfo.strideHeight - padTop;\n            for (let wR = 0; wR < filterHeight; wR++) {\n              const xR = xRCorner + wR * dilationHeight;\n              if (xR < 0 || xR >= convInfo.inHeight) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wR * filter.strides[1];\n              const xOffset3 = xOffset2 + xR * x.strides[2];\n              for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n                const yOffset4 = yOffset3 + yC * convInfo.outChannels;\n                const xCCorner = yC * convInfo.strideWidth - padLeft;\n                for (let wC = 0; wC < filterWidth; wC++) {\n                  const xC = xCCorner + wC * dilationWidth;\n                  if (xC < 0 || xC >= convInfo.inWidth) {\n                    continue;\n                  }\n                  const wOffset3 = wOffset2 + wC * filter.strides[2];\n                  const xOffset4 = xOffset3 + xC * convInfo.inChannels;\n                  let wOffset4 = wOffset3;\n                  for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                    const xVal = xVals[xOffset4 + d1];\n                    for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                      yVals[yOffset4 + d2] += xVal * wVals[wOffset4 + d2];\n                    }\n                    wOffset4 += convInfo.outChannels;\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor();\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'conv2dDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth,\n      dataFormat\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const xBatchStride = dx.strides[0];\n    const xRowStride = isChannelsLast ? dx.strides[1] : dx.strides[2];\n    const xColStride = isChannelsLast ? dx.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : dx.strides[1];\n    const yBatchStride = dy.strides[0];\n    const yRowStride = isChannelsLast ? dy.strides[1] : dy.strides[2];\n    const yColStride = isChannelsLast ? dy.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : dy.strides[1];\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset =\n                    yBatchStride * b + yRowStride * yR + yColStride * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let d2 = 0; d2 < outChannels; ++d2) {\n                  const pixel = dyValues[dyOffset + yChannelStride * d2];\n                  const weight = fltValues[fltOffset + d2];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            const dxOffset = xBatchStride * b + xRowStride * xR +\n                xColStride * xC + xChannelStride * d1;\n            dxValues[dxOffset] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const dx = tf.buffer<Rank.R5>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2, dxS3] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2, fltS3] = filter.strides;\n    const {\n      batchSize,\n      filterDepth,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inDepth,\n      inHeight,\n      inWidth,\n      outChannels,\n      outDepth,\n      outHeight,\n      outWidth,\n      strideDepth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const frontPad = filterDepth - 1 - convInfo.padInfo.front;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        // Frames of depth\n        for (let xF = 0; xF < inDepth; ++xF) {\n          const xFCorner = xF - frontPad;\n          const xFMin = Math.max(0, Math.ceil(xFCorner / strideDepth));\n          const yFMax =\n              Math.min(outDepth, (filterDepth + xFCorner) / strideDepth);\n\n          // Rows as per standard 2d matrix notation\n          for (let xR = 0; xR < inHeight; ++xR) {\n            const xRCorner = xR - topPad;\n            const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n            const yRMax =\n                Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n            // Columns as per standard 2d matrix notation\n            for (let xC = 0; xC < inWidth; ++xC) {\n              const xCCorner = xC - leftPad;\n              const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n              const yCMax =\n                  Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n              let dotProd = 0;\n              for (let yF = xFMin; yF < yFMax; ++yF) {\n                const wF = yF * strideDepth - xFCorner;\n\n                for (let yR = xRMin; yR < yRMax; ++yR) {\n                  const wR = yR * strideHeight - xRCorner;\n\n                  for (let yC = xCMin; yC < yCMax; ++yC) {\n                    const wC = yC * strideWidth - xCCorner;\n                    const dyOffset =\n                        dyS0 * b + dyS1 * yF + dyS2 * yR + dyS3 * yC;\n                    const fltOffset = fltS0 * (filterDepth - 1 - wF) +\n                        fltS1 * (filterHeight - 1 - wR) +\n                        fltS2 * (filterWidth - 1 - wC) + fltS3 * d1;\n\n                    for (let d2 = 0; d2 < outChannels; ++d2) {\n                      const pixel = dyValues[dyOffset + d2];\n                      const weight = fltValues[fltOffset + d2];\n                      dotProd += pixel * weight;\n                    }\n                  }\n                }\n              }\n              dxValues[dxS0 * b + dxS1 * xF + dxS2 * xR + dxS3 * xC + d1] =\n                  dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, dy], 'conv2dDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n          for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n            // Need to convolve.\n            let dotProd = 0;\n            for (let b = 0; b < convInfo.batchSize; ++b) {\n              for (let yR = yRMin; yR < yRMax; ++yR) {\n                const xR = wR + yR * strideHeight - topPad;\n                for (let yC = yCMin; yC < yCMax; ++yC) {\n                  const xC = wC + yC * strideWidth - leftPad;\n                  if (isChannelsLast) {\n                    dotProd +=\n                        xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n                  } else {\n                    dotProd +=\n                        xBuf.get(b, d1, xR, xC) * dyBuf.get(b, d2, yR, yC);\n                  }\n                }\n              }\n            }\n            dW.set(dotProd, wR, wC, d1, d2);\n          }\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const dw = tf.buffer<Rank.R5>(convInfo.filterShape, 'float32');\n    const dwValues = dw.values;\n    const [dwS0, dwS1, dwS2, dwS3] = dw.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const [xS0, xS1, xS2, xS3] = x.strides;\n\n    const frontPad = convInfo.padInfo.front;\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n\n    for (let wF = 0; wF < filterDepth; ++wF) {\n      const yFMin = Math.max(0, Math.ceil((frontPad - wF) / strideDepth));\n      const yFMax = Math.min(\n          convInfo.outDepth, (convInfo.inDepth + frontPad - wF) / strideDepth);\n      const wOffset1 = wF * dwS0;\n\n      for (let wR = 0; wR < filterHeight; ++wR) {\n        const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n        const yRMax = Math.min(\n            convInfo.outHeight,\n            (convInfo.inHeight + topPad - wR) / strideHeight);\n        const wOffset2 = wR * dwS1 + wOffset1;\n\n        for (let wC = 0; wC < filterWidth; ++wC) {\n          const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n          const yCMax = Math.min(\n              convInfo.outWidth,\n              (convInfo.inWidth + leftPad - wC) / strideWidth);\n          const wOffset3 = wC * dwS2 + wOffset2;\n\n          for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n            const wOffset4 = d1 * dwS3 + wOffset3;\n\n            for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n              let dotProd = 0;\n              for (let b = 0; b < convInfo.batchSize; ++b) {\n                const xOffset1 = b * xS0;\n                const yOffset1 = b * dyS0;\n\n                for (let yF = yFMin; yF < yFMax; ++yF) {\n                  const xF = wF + yF * strideDepth - frontPad;\n                  const xOffset2 = xF * xS1 + xOffset1;\n                  const yOffset2 = yF * dyS1 + yOffset1;\n\n                  for (let yR = yRMin; yR < yRMax; ++yR) {\n                    const xR = wR + yR * strideHeight - topPad;\n                    const xOffset3 = xR * xS2 + xOffset2;\n                    const yOffset3 = yR * dyS2 + yOffset2;\n\n                    for (let yC = yCMin; yC < yCMax; ++yC) {\n                      const xC = wC + yC * strideWidth - leftPad;\n                      const xOffset4 = xC * xS3 + xOffset3;\n                      const yOffset4 = yC * dyS3 + yOffset3;\n\n                      dotProd +=\n                          xValues[xOffset4 + d1] * dyValues[yOffset4 + d2];\n                    }\n                  }\n                }\n              }\n              dwValues[wOffset4 + d2] = dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dw.toTensor();\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.depthwiseConv2D(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedDepthwiseConv2D is\n      // modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, filter], 'depthwiseConv2D');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * y.strides[1];\n        const xRCorner = yR * convInfo.strideHeight - padLeft;\n        for (let wR = 0; wR < filterHeight; ++wR) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * x.strides[1];\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * y.strides[2];\n            const xCCorner = yC * convInfo.strideWidth - padTop;\n            for (let wC = 0; wC < filterWidth; ++wC) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * convInfo.inChannels;\n              let yOffset4 = yOffset3;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1];\n                for (let q = 0; q < chMul; ++q) {\n                  yVals[yOffset4 + q] += xVal * wVals[wOffset3 + q];\n                }\n                yOffset4 += chMul;\n                wOffset3 += chMul;\n              }\n            }\n          }\n        }\n      }\n    }\n\n    return y.toTensor() as Tensor4D;\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'depthwiseConv2DDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n    const chMul = outChannels / inChannels;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset = dyS0 * b + dyS1 * yR + dyS2 * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let dm = 0; dm < chMul; ++dm) {\n                  const d2 = d1 * chMul + dm;\n                  const pixel = dyValues[dyOffset + d2];\n                  const weight = fltValues[fltOffset + dm];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            dxValues[dxS0 * b + dxS1 * xR + dxS2 * xC + d1] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, dy], 'depthwiseConv2DDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n          const d1 = Math.trunc(d2 / chMul);\n          const dm = d2 % chMul;\n\n          let dotProd = 0;\n          for (let b = 0; b < convInfo.batchSize; ++b) {\n            for (let yR = yRMin; yR < yRMax; ++yR) {\n              const xR = wR + yR * strideHeight - topPad;\n              for (let yC = yCMin; yC < yCMax; ++yC) {\n                const xC = wC + yC * strideWidth - leftPad;\n                dotProd += xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n              }\n            }\n          }\n          dW.set(dotProd, wR, wC, d1, dm);\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    assertNotComplex(x, 'tile');\n    return tile(this.bufferSync(x), reps) as T;\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    assertNotComplex([x, indices], 'gather');\n\n    const newShape: number[] = x.shape.slice();\n    const indicesValues = this.readSync(indices.dataId) as TypedArray;\n    newShape[axis] = indicesValues.length;\n    const result = tf.buffer(newShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < result.size; ++i) {\n      const newLoc = result.indexToLoc(i);\n\n      const originalLoc: number[] = newLoc.slice();\n      originalLoc[axis] = indicesValues[newLoc[axis]];\n\n      const originalIndex = xBuf.locToIndex(originalLoc);\n      result.values[i] = xBuf.values[originalIndex];\n    }\n    return result.toTensor() as T;\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    assertNotComplex([x], 'batchToSpaceND');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return tf.transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  private pool3d(\n      x: Tensor5D, convInfo: backend_util.Conv3DInfo,\n      poolType: 'max'|'avg'): Tensor5D {\n    assertNotComplex(x, 'pool3d');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const initialValue =\n        (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                              Number.POSITIVE_INFINITY);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = tf.buffer(convInfo.outShape, x.dtype);\n    const outputVals = output.values;\n\n    const outputBatchStrides = convInfo.outShape[1] * convInfo.outShape[2] *\n        convInfo.outShape[3] * convInfo.outShape[4];\n    const outputDepthStrides =\n        convInfo.outShape[2] * convInfo.outShape[3] * convInfo.outShape[4];\n    const outputRowStrides = convInfo.outShape[3] * convInfo.outShape[4];\n    const outputColStrides = convInfo.outShape[4];\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      const outputBatchOffset = batch * outputBatchStrides;\n      const inputBatchOffset = batch * x.strides[0];\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          const outputDepthOffset =\n              outputBatchOffset + yDepth * outputDepthStrides;\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            const outputRowOffset = outputDepthOffset + yRow * outputRowStrides;\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n              // Shader code begins\n              const outputColOffset = outputRowOffset + yCol * outputColStrides;\n              let minMaxValue = initialValue;\n              let avgValue = 0;\n              let count = 0;\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const xDepthOffset = inputBatchOffset + xDepth * x.strides[1];\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const xRowOffset = xDepthOffset + xRow * x.strides[2];\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const xColOffset = xRowOffset + xCol * x.strides[3];\n                    const pixel = xValues[xColOffset + channel];\n                    if ((poolType === 'max' && pixel > minMaxValue)) {\n                      minMaxValue = pixel;\n                    } else if (poolType === 'avg') {\n                      avgValue += pixel;\n                      count++;\n                    }\n                    if (isNaN(minMaxValue)) {\n                      break;\n                    }\n                  }\n                  if (isNaN(minMaxValue)) {\n                    break;\n                  }\n                }\n                if (isNaN(minMaxValue)) {\n                  break;\n                }\n              }\n              const outputOffset = outputColOffset + channel;\n              outputVals[outputOffset] =\n                  poolType === 'avg' ? avgValue / count : minMaxValue;\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor5D;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'avgPool3d');\n\n    return this.pool3d(x, convInfo, 'avg').toFloat();\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([dy, x], 'avgPool3dBackprop');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins.\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel;\n                  }\n                }\n              }\n              dx.set(\n                  dotProd * avgMultiplier, batch, dxDepth, dxRow, dxCol,\n                  channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'maxPool3d');\n\n    return this.pool3d(x, convInfo, 'max').toFloat();\n  }\n\n  private maxPool3dPositions(x: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const maxPositions = tf.buffer(convInfo.outShape, 'int32');\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const xBuf = this.bufferSync(x);\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n\n              // Shader code begins\n              let maxValue = Number.NEGATIVE_INFINITY;\n              let maxPosition = -1;\n\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const wDepth = xDepth - xDepthCorner;\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const wRow = xRow - xRowCorner;\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const wCol = xCol - xColCorner;\n                    const pixel = xBuf.get(batch, xDepth, xRow, xCol, channel);\n                    if (pixel >= maxValue) {\n                      maxValue = pixel;\n                      maxPosition = wDepth * effectiveFilterHeight *\n                              effectiveFilterWidth +\n                          wRow * effectiveFilterHeight + wCol;\n                    }\n                  }\n                }\n              }\n\n              maxPositions.set(maxPosition, batch, yDepth, yRow, yCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return maxPositions.toTensor() as Tensor5D;\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([x, y], 'maxPool3dBackprop');\n\n    const maxPositions = this.maxPool3dPositions(x, convInfo);\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const maxPosBuf = this.bufferSync(maxPositions);\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const maxPos = effectiveFilterDepth *\n                            effectiveFilterHeight * effectiveFilterWidth -\n                        1 -\n                        maxPosBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    const curPos =\n                        wDepth * effectiveFilterHeight * effectiveFilterWidth +\n                        wRow * effectiveFilterWidth + wCol;\n\n                    const mask = maxPos === curPos ? 1 : 0;\n                    if (mask === 0) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel * mask;\n                  }\n                }\n              }\n              dx.set(dotProd, batch, dxDepth, dxRow, dxCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeBilinear');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result = new Float32Array(\n        util.sizeFromShape([batch, newHeight, newWidth, numChannels]));\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n    let outputIdx = 0;\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n    for (let b = 0; b < batch; b++) {\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceRowFloor = Math.floor(sourceFracRow);\n        const rowFrac = sourceFracRow - sourceRowFloor;\n        const sourceRowCeil = Math.min(oldHeight - 1, Math.ceil(sourceFracRow));\n        const topRowOffset = b * x.strides[0] + sourceRowFloor * x.strides[1];\n        const botRowOffset = b * x.strides[0] + sourceRowCeil * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceColFloor = Math.floor(sourceFracCol);\n          const colFrac = sourceFracCol - sourceColFloor;\n          const sourceColCeil =\n              Math.min(oldWidth - 1, Math.ceil(sourceFracCol));\n          const topLeftOffest = topRowOffset + sourceColFloor * x.strides[2];\n          const botLeftOffset = botRowOffset + sourceColFloor * x.strides[2];\n          const topRightOffset = topRowOffset + sourceColCeil * x.strides[2];\n          const botRightOffest = botRowOffset + sourceColCeil * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n\n            // Compute the fractional index of the source.\n            const topLeft = xValues[topLeftOffest + d];\n            const bottomLeft = xValues[botLeftOffset + d];\n            const topRight = xValues[topRightOffset + d];\n            const bottomRight = xValues[botRightOffest + d];\n\n            const top = topLeft + (topRight - topLeft) * colFrac;\n            const bottom = bottomLeft + (bottomRight - bottomLeft) * colFrac;\n            const newValue = top + (bottom - top) * rowFrac;\n\n            result[outputIdx++] = newValue;\n          }\n        }\n      }\n    }\n    return tf.tensor(result, [batch, newHeight, newWidth, numChannels]);\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeBilinearBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass and add the\n    // corresponding coefficient from dy to the gradient (with some\n    // interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/3039375c86a5bbc9610c7725dcaa95d635f87ba2/tensorflow/core/kernels/resize_bilinear_op.cc#L275\n\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    let offset = 0;\n    for (let b = 0; b < batch; b++) {\n      const bOffset = b * x.strides[0];\n      for (let r = 0; r < yHeight; r++) {\n        const dxR = r * heightScale;\n        const topDxRIndex = Math.floor(dxR);\n        const bottomDxRIndex = Math.min(Math.ceil(dxR), xHeight - 1);\n\n        const topDxROffset = bOffset + topDxRIndex * x.strides[1];\n        const bottomDxROffset = bOffset + bottomDxRIndex * x.strides[1];\n\n        const dxRLerp = dxR - topDxRIndex;\n        const inverseDxRLerp = 1.0 - dxRLerp;\n        for (let c = 0; c < yWidth; c++) {\n          const dxC = c * widthScale;\n          const leftDxCIndex = Math.floor(dxC);\n          const rightDxCIndex = Math.min(Math.ceil(dxC), xWidth - 1);\n          const dxCLerp = dxC - leftDxCIndex;\n          const inverseDxCLerp = 1.0 - dxCLerp;\n\n          const topLeftRCOffset = topDxROffset + leftDxCIndex * x.strides[2];\n          const topRightRCOffset = topDxROffset + rightDxCIndex * x.strides[2];\n          const bottomLeftRCOffset =\n              bottomDxROffset + leftDxCIndex * x.strides[2];\n          const bottomRightRCOffset =\n              bottomDxROffset + rightDxCIndex * x.strides[2];\n\n          const inverseDxRLerpTimesInverseDxCLerp =\n              inverseDxRLerp * inverseDxCLerp;\n          const inverseDxRLerpTimesDxCLerp = inverseDxRLerp * dxCLerp;\n          const dxRLerpTimesInverseDxCLerp = dxRLerp * inverseDxCLerp;\n          const dxRLerpTimesDxCLerp = dxRLerp * dxCLerp;\n          for (let d = 0; d < depth; d++) {\n            const dyVal = dyValues[offset++];\n            output[topLeftRCOffset + d] +=\n                dyVal * inverseDxRLerpTimesInverseDxCLerp;\n            output[topRightRCOffset + d] += dyVal * inverseDxRLerpTimesDxCLerp;\n            output[bottomLeftRCOffset + d] +=\n                dyVal * dxRLerpTimesInverseDxCLerp;\n            output[bottomRightRCOffset + d] += dyVal * dxRLerpTimesDxCLerp;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, [batch, xWidth, xHeight, depth], x.dtype);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeNearestNeighbor');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = new Float32Array(batch * newHeight * newWidth * numChannels);\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n\n    let outputOffset = 0;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceNearestRow = Math.min(\n            oldHeight - 1,\n            alignCorners ? Math.round(sourceFracRow) :\n                           Math.floor(sourceFracRow));\n        const rowOffset = batchOffset + sourceNearestRow * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceNearestCol = Math.min(\n              oldWidth - 1,\n              alignCorners ? Math.round(sourceFracCol) :\n                             Math.floor(sourceFracCol));\n          const colOffset = rowOffset + sourceNearestCol * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n            // Compute the fractional index of the source.\n            const newVal = xValues[colOffset + d];\n            output[outputOffset++] = newVal;\n          }\n        }\n      }\n    }\n    return tf.tensor(\n        output, [batch, newHeight, newWidth, numChannels], x.dtype);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeNearestNeighborBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    // Loop over the output space.\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < xHeight; r++) {\n        const rowOffset = batchOffset + r * x.strides[1];\n\n        // Compute bounds for where in dy we will look\n        const startRLerp = Math.floor(r * invHeightScale);\n        const startDyR = Math.floor(startRLerp - (winHeight / 2));\n        for (let c = 0; c < xWidth; c++) {\n          const colOffset = rowOffset + c * x.strides[2];\n\n          // Compute bounds for where in dy we will look\n          const startCLerp = Math.floor(c * invWidthScale);\n          const startDyC = Math.floor(startCLerp - (winWidth / 2));\n\n          for (let d = 0; d < depth; d++) {\n            let accum = 0;\n            // loop over dy\n\n            for (let dyRIndex = 0; dyRIndex < winHeight; dyRIndex++) {\n              const dyR = dyRIndex + startDyR;\n              // Guard against the window exceeding the bounds of dy\n              if (dyR < 0 || dyR >= yHeight) {\n                continue;\n              }\n\n              const dyROffset = batchOffset + dyR * dy.strides[1];\n              const sourceFracRow = dyR * heightScale;\n              const sourceNearestRow = Math.min(\n                  xHeight - 1,\n                  alignCorners ? Math.round(sourceFracRow) :\n                                 Math.floor(sourceFracRow));\n              if (r !== sourceNearestRow) {\n                continue;\n              }\n              for (let dyCIndex = 0; dyCIndex < winWidth; dyCIndex++) {\n                const dyC = dyCIndex + startDyC;\n                // Guard against the window exceeding the bounds of dy\n                if (dyC < 0 || dyC >= yWidth) {\n                  continue;\n                }\n\n                const dyCOffset = dyROffset + dyC * dy.strides[2];\n                const sourceFracCol = dyC * widthScale;\n                const sourceNearestCol = Math.min(\n                    xWidth - 1,\n                    alignCorners ? Math.round(sourceFracCol) :\n                                   Math.floor(sourceFracCol));\n\n                if (c === sourceNearestCol) {\n                  accum += dyValues[dyCOffset + d];\n                }\n              }\n            }\n            output[colOffset + d] = accum;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, x.shape, x.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(x, 'localResponseNormalization4D');\n\n    const channels = x.shape[3];\n    const maxD = channels - 1;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const size = x.size;\n    const result = new Float32Array(size);\n\n    function sumAcrossChannels(offset: number) {\n      const currentChannel = offset % channels;\n      let beginSumOffset =\n          offset - currentChannel + Math.max(0, currentChannel - depthRadius);\n      const endSumOffset = offset - currentChannel +\n          Math.min(currentChannel + depthRadius, maxD);\n\n      let sum = 0.0;\n      for (; beginSumOffset <= endSumOffset; beginSumOffset++) {\n        const z = xValues[beginSumOffset];\n        sum += z * z;\n      }\n      return sum;\n    }\n\n    for (let offset = 0; offset < size; offset++) {\n      const sum = sumAcrossChannels(offset);\n      const val = xValues[offset] * Math.pow(bias + alpha * sum, -beta);\n      result[offset] = val;\n    }\n\n    return tf.tensor4d(result, x.shape);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(dy, 'LRNGrad');\n    const channels = dy.shape[3];\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const inputImageValues = this.readSync(inputImage.dataId) as TypedArray;\n    const outputImageValues = this.readSync(outputImage.dataId) as TypedArray;\n    const result = new Float32Array(dy.size);\n    const size = dy.size;\n\n    for (let offset = 0; offset < size; offset++) {\n      const currentChannel = offset % channels;\n      const depthBegin =\n          (offset - currentChannel) + Math.max(0, currentChannel - depthRadius);\n      const depthEnd = (offset - currentChannel) +\n          Math.min(channels, currentChannel + depthRadius + 1);\n\n      let norm = 0;\n      for (let k = depthBegin; k < depthEnd; k++) {\n        norm += Math.pow(inputImageValues[k], 2);\n      }\n      norm = alpha * norm + bias;\n\n      for (let k = depthBegin; k < depthEnd; k++) {\n        let dyi = -2 * alpha * beta * inputImageValues[k] *\n            outputImageValues[offset] / norm;\n        if (offset === k) {\n          dyi += Math.pow(norm, -beta);\n        }\n        dyi *= dyValues[offset];\n        result[k] += dyi;\n      }\n    }\n    return tf.tensor4d(result, dy.shape);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    assertNotComplex(logits, 'multinomial');\n\n    const probabilities = normalized ? logits : tf.softmax(logits);\n    const batchSize = probabilities.shape[0];\n    const numEvents = probabilities.shape[1];\n    const res = tf.zeros<Rank.R2>([batchSize, numSamples], 'int32');\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const probVals = this.readSync(probabilities.dataId) as TypedArray;\n\n    for (let b = 0; b < batchSize; ++b) {\n      const offset = b * numEvents;\n      // The cdf won't include the last event. It will be implicit if no other\n      // event happened.\n      const cdf = new Float32Array(numEvents - 1);\n      cdf[0] = probVals[offset];\n      for (let event = 1; event < cdf.length; ++event) {\n        cdf[event] = cdf[event - 1] + probVals[offset + event];\n      }\n\n      const random = seedrandom.alea(seed.toString());\n      const outOffset = b * numSamples;\n      for (let sampleId = 0; sampleId < numSamples; ++sampleId) {\n        const r = random();\n\n        // Assume last event happened by default.\n        resVals[outOffset + sampleId] = cdf.length;\n\n        for (let event = 0; event < cdf.length; event++) {\n          if (r < cdf[event]) {\n            resVals[outOffset + sampleId] = event;\n            break;\n          }\n        }\n      }\n    }\n    return res;\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    assertNotComplex(indices, 'oneHot');\n\n    const res = new Float32Array(indices.size * depth);\n    res.fill(offValue);\n    const indicesVal = this.readSync(indices.dataId) as TypedArray;\n\n    for (let event = 0; event < indices.size; ++event) {\n      if (indicesVal[event] >= 0 && indicesVal[event] < depth) {\n        res[event * depth + indicesVal[event]] = onValue;\n      }\n    }\n    return tf.tensor2d(res, [indices.size, depth], 'int32');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold: number): Tensor1D {\n    assertNotComplex(boxes, 'nonMaxSuppression');\n\n    const boxesVals = this.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = this.readSync(scores.dataId) as TypedArray;\n    return nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        dataFormat === 'NHWC',\n        () => `Only NHWC dataFormat supported on CPU for depthToSpace. Got ${\n            dataFormat}`);\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = x.shape[1];\n    const inputWidth = x.shape[2];\n    const inputDepth = x.shape[3];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result =\n        new Float32Array(batchSize * outputHeight * outputWidth * outputDepth);\n\n    let outputIdx = 0;\n    for (let b = 0; b < batchSize; ++b) {\n      for (let h = 0; h < outputHeight; ++h) {\n        const inH = Math.floor(h / blockSize);\n        const offsetH = (h % blockSize);\n        for (let w = 0; w < outputWidth; ++w) {\n          const inW = Math.floor(w / blockSize);\n          const offsetW = (w % blockSize);\n          const offsetD = (offsetH * blockSize + offsetW) * outputDepth;\n          for (let d = 0; d < outputDepth; ++d) {\n            const inD = d + offsetD;\n            const inputIdx =\n                inD + inputDepth * (inW + inputWidth * (inH + inputHeight * b));\n            result[outputIdx++] = xValues[inputIdx];\n          }\n        }\n      }\n    }\n    return tf.tensor4d(\n        result, [batchSize, outputHeight, outputWidth, outputDepth]);\n  }\n\n  private broadcastedBinaryOp(\n      a: Tensor, b: Tensor, dtype: DataType,\n      op: (a: number, b: number) => number): Tensor {\n    const newShape = backend_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const result = tf.buffer(newShape, dtype);\n    const aVals = this.readSync(a.dataId) as TypedArray;\n    const bVals = this.readSync(b.dataId) as TypedArray;\n    const aBroadcastDims = backend_util.getBroadcastDims(a.shape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(b.shape, newShape);\n\n    const resVals = result.values;\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resVals.length; ++i) {\n        resVals[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      const aBuf = this.bufferSync(a);\n      const bBuf = this.bufferSync(b);\n      for (let i = 0; i < resVals.length; ++i) {\n        const loc = result.indexToLoc(i);\n\n        const aLoc = loc.slice(-a.rank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = aBuf.locToIndex(aLoc);\n\n        const bLoc = loc.slice(-b.rank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = bBuf.locToIndex(bLoc);\n\n        resVals[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n    return result.toTensor();\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  dispose() {}\n\n  floatPrecision(): 16|32 {\n    return 32;\n  }\n\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return super.epsilon();\n  }\n\n  cropAndResize(\n      images: Tensor4D,\n      boxes: Tensor2D,\n      boxIndex: Tensor1D,\n      cropSize: [number, number],\n      method: string,\n      extrapolationValue: number,\n  ) {\n    const [batch, imageHeight, imageWidth, numChannels] = images.shape;\n    const numBoxes = boxes.shape[0];\n\n    const [cropHeight, cropWidth] = cropSize;\n    const output =\n        tf.buffer([numBoxes, cropHeight, cropWidth, numChannels], 'float32');\n\n    const boxVals = this.readSync(boxes.dataId) as TypedArray;\n    const boxIndVals = this.readSync(boxIndex.dataId) as TypedArray;\n    const imageVals = this.readSync(images.dataId) as TypedArray;\n\n    const inStride = images.strides;   // to calculate flat indexes into image\n    const outStride = output.strides;  // to calculate flat indexes into output\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op.cc\n    for (let b = 0; b < numBoxes; b++) {\n      const startInd = b * 4;\n      const y1 = boxVals[startInd];\n      const x1 = boxVals[startInd + 1];\n      const y2 = boxVals[startInd + 2];\n      const x2 = boxVals[startInd + 3];\n\n      const bInd: number = boxIndVals[b];\n      if (bInd >= batch) {\n        continue;\n      }\n\n      const heightScale = (cropHeight > 1) ?\n          (y2 - y1) * (imageHeight - 1) / (cropHeight - 1) :\n          0;\n      const widthScale =\n          (cropWidth > 1) ? (x2 - x1) * (imageWidth - 1) / (cropWidth - 1) : 0;\n\n      for (let y = 0; y < cropHeight; y++) {\n        const yInd: number = (cropHeight > 1) ?\n            y1 * (imageHeight - 1) + y * (heightScale) :\n            0.5 * (y1 + y2) * (imageHeight - 1);\n\n        if (yInd < 0 || yInd > imageHeight - 1) {\n          for (let x = 0; x < cropWidth; x++) {\n            for (let c = 0; c < numChannels; c++) {\n              const ind =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = extrapolationValue;\n            }\n          }\n          continue;\n        }\n\n        if (method === 'bilinear') {\n          const topInd = Math.floor(yInd);\n          const bottomInd = Math.ceil(yInd);\n          const yLerp = yInd - topInd;\n\n          for (let x = 0; x < cropWidth; x++) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const leftInd = Math.floor(xInd);\n            const rightInd = Math.ceil(xInd);\n            const xLerp = xInd - leftInd;\n\n            for (let c = 0; c < numChannels; c++) {\n              let ind = c + leftInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topRight = imageVals[ind];\n\n              ind = c + leftInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomRight = imageVals[ind];\n\n              const top = topLeft + (topRight - topLeft) * xLerp;\n              const bottom = bottomLeft + (bottomRight - bottomLeft) * xLerp;\n\n              ind = c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = top + ((bottom - top) * yLerp);\n            }\n          }\n        } else {  // method == \"nearest\"\n          for (let x = 0; x < cropWidth; ++x) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const closestX = Math.round(xInd);\n            const closestY = Math.round(yInd);\n            for (let c = 0; c < numChannels; c++) {\n              const inInd = c + closestX * inStride[2] +\n                  closestY * inStride[1] + bInd * inStride[0];\n              const outInd =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[outInd] = imageVals[inInd];\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor4D;\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n    const sumDupeIndices = false;\n    return this.scatter(\n        sparseIndices, sparseValues, outputShape, outputSize, sliceSize,\n        numUpdates, sliceRank, strides, defaultValue, sumDupeIndices);\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n    if (numSlices === 0) {\n      return tf.tensor([], resultShape, x.dtype);\n    }\n\n    const buffer = new TensorBuffer([numSlices, sliceSize], x.dtype);\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const xData = this.readSync(x.dataId) as TypedArray;\n\n    for (let i = 0; i < numSlices; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        flattenIndex += dim * strides[j];\n        index.push(dim);\n      }\n      if (flattenIndex < 0 || flattenIndex >= x.size / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${x.shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        buffer.values[i * sliceSize + k] = xData[flattenIndex * sliceSize + k];\n      }\n    }\n    return buffer.toTensor().reshape(resultShape);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n    const defaultValue = tf.scalar(0);\n    const sumDupeIndices = true;\n    return this.scatter(\n        indices, updates, shape, outputSize, sliceSize, numUpdates, sliceRank,\n        strides, defaultValue, sumDupeIndices);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n    const values =\n        util.getArrayFromDType(dtype, util.sizeFromShape(shape)) as TypedArray;\n    values.fill(value as number);\n    return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported for string tensors');\n    } else {\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    const values = util.getArrayFromDType(\n                       x.dtype, util.sizeFromShape(x.shape)) as TypedArray;\n    return this.makeOutput(values, x.shape, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  private scatter<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R], outputSize: number,\n      sliceSize: number, numUpdates: number, sliceRank: number,\n      strides: number[], defaultValue: Scalar,\n      sumDupeIndices: boolean): Tensor<R> {\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const updatesData = this.readSync(updates.dataId) as TypedArray;\n\n    if (outputSize === 0) {\n      return tf.tensor([], shape, updates.dtype);\n    }\n\n    const buffer = new TensorBuffer(flattenShape, updates.dtype as 'float32');\n    buffer.values.fill((this.readSync(defaultValue.dataId) as TypedArray)[0]);\n\n    for (let i = 0; i < numUpdates; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        index.push(dim);\n        flattenIndex += dim * strides[j];\n      }\n\n      if (flattenIndex < 0 || flattenIndex >= outputSize / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        if (sumDupeIndices) {\n          buffer.values[flattenIndex * sliceSize + k] +=\n              updatesData[i * sliceSize + k];\n        } else {\n          buffer.values[flattenIndex * sliceSize + k] = updates.rank === 0 ?\n              updatesData[0] :\n              updatesData[i * sliceSize + k];\n        }\n      }\n    }\n    return buffer.toTensor().reshape(shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs, AbsInputs, KernelConfig, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function simpleAbsImpl(vals: TypedArray): Float32Array {\n  const resultValues = new Float32Array(vals.length);\n  for (let i = 0; i < vals.length; ++i) {\n    resultValues[i] = Math.abs(vals[i]);\n  }\n  return resultValues;\n}\n\nexport const absKernelFunc =\n    (args: {inputs: AbsInputs, backend: MathBackendCPU}) => {\n      const {x} = args.inputs;\n      const cpuBackend = args.backend;\n      let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n      if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n        resultValues = simpleAbsImpl(values);\n      } else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals =\n            cpuBackend.data.get(real.dataId).values as Float32Array;\n        const imagVals =\n            cpuBackend.data.get(imag.dataId).values as Float32Array;\n        for (let i = 0; i < realVals.length; i++) {\n          const real = realVals[i];\n          const imag = imagVals[i];\n          resultValues[i] = Math.hypot(real, imag);\n        }\n      }\n      return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n    };\n\nexport const absConfig: KernelConfig = {\n  kernelName: Abs,\n  backendName: 'cpu',\n  kernelFunc: absKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleBinaryKernelImpl, SimpleBinaryOperation} from './binary_types';\n\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op: SimpleBinaryOperation):\n    SimpleBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aVals: TypedArray,\n          bVals: TypedArray, dtype: DataType): [TypedArray, number[]] => {\n    const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    const resultRank = newShape.length;\n    const resultStrides = util.computeStrides(newShape);\n    const resultSize = util.sizeFromShape(newShape);\n\n    const result =\n        util.getTypedArrayFromDType(dtype as NumericDataType, resultSize);\n\n    const aRank = aShape.length;\n    const bRank = bShape.length;\n\n    const aStrides = util.computeStrides(aShape);\n    const bStrides = util.computeStrides(bShape);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < result.length; ++i) {\n        result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      for (let i = 0; i < result.length; ++i) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        result[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n\n    return [result, newShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Complex, ComplexInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function complex(args: {inputs: ComplexInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {real, imag} = inputs;\n\n  const realVals = backend.data.get(real.dataId).values as TypedArray;\n  const imagVals = backend.data.get(imag.dataId).values as TypedArray;\n\n  const complexInfo = backend.makeTensorInfo(real.shape, 'complex64');\n\n  const complex = backend.data.get(complexInfo.dataId);\n\n  // The complex tensor owns the underlying real and imag tensorInfos, only the\n  // complex tensor tracks refCount, when complexData is disposed the\n  // underlying tensorData will be disposed.\n  complex.complexTensorInfos = {\n    real: backend.makeTensorInfo(real.shape, 'float32', realVals),\n    imag: backend.makeTensorInfo(imag.shape, 'float32', imagVals)\n  };\n\n  return complexInfo;\n}\n\nexport const complexConfig: KernelConfig = {\n  kernelName: Complex,\n  backendName: 'cpu',\n  kernelFunc: complex as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendCPU}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'cpu',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Real, RealInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function real(args: {inputs: RealInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const real = backend.data.get(input.dataId).complexTensorInfos.real;\n  const realVal = backend.data.get(real.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the real value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(real.shape, real.dtype, realVal);\n}\n\nexport const realConfig: KernelConfig = {\n  kernelName: Real,\n  backendName: 'cpu',\n  kernelFunc: real as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport * as tf from '@tensorflow/tfjs-core';\nimport {Cast, CastAttrs, CastInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\n\nimport {complex} from './Complex';\nimport {identity} from './Identity';\nimport {real} from './Real';\n\nexport function cast(\n    args: {inputs: CastInputs, backend: MathBackendCPU, attrs: CastAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {dtype} = attrs;\n\n  // Casting to complex64.\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return identity({inputs: {x}, backend});\n    }\n\n    // TODO(lina128): Import kernel function once zeros is modularized.\n    const zerosTensor = tf.zeros(x.shape);\n    const floatX = cast({inputs: {x}, backend, attrs: {dtype: 'float32'}});\n\n    const result =\n        complex({inputs: {real: floatX, imag: zerosTensor}, backend});\n\n    zerosTensor.dispose();\n    backend.disposeIntermediateTensorInfo(floatX);\n\n    return result;\n  }\n\n  // Casting from complex64\n  if (x.dtype === 'complex64') {\n    const realPart = real({inputs: {input: x}, backend});\n    const result = cast({inputs: {x: realPart}, backend, attrs: {dtype}});\n\n    backend.disposeIntermediateTensorInfo(realPart);\n\n    return result;\n  }\n\n  if (!util.hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    const result = identity({inputs: {x}, backend});\n    return {dataId: result.dataId, shape: result.shape, dtype};\n  }\n\n  if (dtype === 'int32') {\n    const values = backend.data.get(x.dataId).values as TypedArray;\n    const resultValues = Int32Array.from(values);\n    return backend.makeTensorInfo(x.shape, 'int32', resultValues);\n  }\n\n  if (dtype === 'bool') {\n    // This is essentially the result of notEqual(x, 0). We avoid using\n    // kernel notEqual to avoid circular dependency, i.e. binary_utils ->\n    // cast -> notEqual -> binary_utils.\n    const xVals = backend.data.get(x.dataId).values as TypedArray;\n    const zero = util.toTypedArray([0], x.dtype);\n\n    const [resultData, resultShape] = createSimpleBinaryKernelImpl(\n        (a, b) => (a !== b) ? 1 : 0)(x.shape, [], xVals, zero, 'bool');\n\n    return backend.makeTensorInfo(resultShape, 'bool', resultData);\n  }\n\n  throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n}\n\nexport const castConfig: KernelConfig = {\n  kernelName: Cast,\n  backendName: 'cpu',\n  kernelFunc: cast as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, BinaryInputs, DataType, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {cast} from '../kernels/Cast';\nimport {complex} from '../kernels/Complex';\n\nimport {ComplexBinaryKernelImpl, ComplexBinaryOperation, SimpleBinaryKernelImpl} from './binary_types';\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param name Kernel name.\n * @param binaryKernelImpl A `SimpleBinaryKernelImpl` for the kernel.\n * @param binaryKernelComplexImpl Optional. If exists, represents a\n *     `ComplexBinaryKernelImpl` for the kernel, will be used when input dtype\n *     is `complex64`.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    name: string, simpleImpl: SimpleBinaryKernelImpl,\n    complexImpl?: ComplexBinaryKernelImpl, dtype?: DataType): KernelFunc {\n  if (complexImpl == null) {\n    return ({inputs, backend}) => {\n      const {a, b} = inputs as BinaryInputs;\n      const cpuBackend = backend as MathBackendCPU;\n\n      assertNotComplex([a, b], name);\n\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    };\n  }\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    if (a.dtype === 'complex64' || b.dtype === 'complex64') {\n      const $aComplex = cast(\n          {inputs: {x: a}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $aComplexVals = cpuBackend.data.get($aComplex.dataId);\n\n      const aReal = $aComplexVals.complexTensorInfos.real;\n      const aImag = $aComplexVals.complexTensorInfos.imag;\n\n      const aRealVals =\n          cpuBackend.data.get(aReal.dataId).values as Float32Array;\n      const aImagVals =\n          cpuBackend.data.get(aImag.dataId).values as Float32Array;\n\n      const $bComplex = cast(\n          {inputs: {x: b}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $bComplexVals = cpuBackend.data.get($bComplex.dataId);\n\n      const bReal = $bComplexVals.complexTensorInfos.real;\n      const bImag = $bComplexVals.complexTensorInfos.imag;\n\n      const bRealVals =\n          cpuBackend.data.get(bReal.dataId).values as Float32Array;\n      const bImagVals =\n          cpuBackend.data.get(bImag.dataId).values as Float32Array;\n\n      const [resultRealData, resultImagData, resultShape] = complexImpl(\n          a.shape, b.shape, aRealVals, aImagVals, bRealVals, bImagVals);\n\n      const resultReal =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultRealData);\n\n      const resultImag =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultImagData);\n\n      const result = complex(\n          {inputs: {real: resultReal, imag: resultImag}, backend: cpuBackend});\n\n      cpuBackend.disposeIntermediateTensorInfo($aComplex);\n      cpuBackend.disposeIntermediateTensorInfo($bComplex);\n      cpuBackend.disposeIntermediateTensorInfo(resultReal);\n      cpuBackend.disposeIntermediateTensorInfo(resultImag);\n\n      return result;\n    } else {\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    }\n  };\n}\n\n/**\n * Template that creates the complex type implementation for binary ops.\n * Supports broadcast.\n */\nexport function createComplexBinaryKernelImpl(op: ComplexBinaryOperation):\n    ComplexBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aRealVals: Float32Array,\n          aImagVals: Float32Array, bRealVals: Float32Array,\n          bImagVals: Float32Array): [TypedArray, TypedArray, number[]] => {\n    const resultShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const resultSize = util.sizeFromShape(resultShape);\n    const resultRank = resultShape.length;\n    const resultStrides = util.computeStrides(resultShape);\n\n    const resultRealVals = util.getTypedArrayFromDType('float32', resultSize);\n    const resultImagVals = util.getTypedArrayFromDType('float32', resultSize);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, resultShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, resultShape);\n\n    const aVals = backend_util.mergeRealAndImagArrays(aRealVals, aImagVals);\n    const bVals = backend_util.mergeRealAndImagArrays(bRealVals, bImagVals);\n\n    const aRank = aShape.length;\n    const aStrides = util.computeStrides(aShape);\n\n    const bRank = bShape.length;\n    const bStrides = util.computeStrides(bShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const aIdx = i % aVals.length;\n        const bIdx = i % bVals.length;\n\n        const result =\n            op(aVals[aIdx * 2], aVals[aIdx * 2 + 1], bVals[bIdx * 2],\n               bVals[bIdx * 2 + 1]);\n\n        resultRealVals[i] = result.real;\n        resultImagVals[i] = result.imag;\n      }\n    } else {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        const opResult =\n            op(aVals[aIndex * 2], aVals[aIndex * 2 + 1], bVals[bIndex * 2],\n               bVals[bIndex * 2 + 1]);\n\n        resultRealVals[i] = opResult.real;\n        resultImagVals[i] = opResult.imag;\n      }\n    }\n    return [resultRealVals, resultImagVals, resultShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Add, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal + bReal, imag: aImag + bImag};\n    }));\n\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\n\nexport const addConfig: KernelConfig = {\n  kernelName: Add,\n  backendName: 'cpu',\n  kernelFunc: add\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NumericDataType, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op: SimpleUnaryOperation):\n    SimpleUnaryImpl {\n  return (values, dtype, attrs) => {\n    const newValues =\n        util.getTypedArrayFromDType(dtype as NumericDataType, values.length);\n    for (let i = 0; i < values.length; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return newValues;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelFunc, TypedArray, UnaryInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param name Kernel name.\n * @param op A `SimpleUnaryOperation` for the kernel.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFunc(\n    name: string, op: SimpleUnaryOperation, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xSize = util.sizeFromShape(x.shape);\n    const $dtype = dtype || x.dtype;\n    const newValues = util.getArrayFromDType($dtype, xSize);\n    for (let i = 0; i < xSize; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for unary ops from the given\n * `SimpleUnaryImpl`..\n * @param name Kernel name.\n * @param unaryImpl A `SimpleUnaryImpl` that implements the op.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFuncFromImpl(\n    name: string, unaryImpl: SimpleUnaryImpl, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const $dtype = dtype || x.dtype;\n    const newValues = unaryImpl(values, $dtype, attrs);\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\n\nexport const ceilConfig: KernelConfig = {\n  kernelName: Ceil,\n  backendName: 'cpu',\n  kernelFunc: ceilKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\n\nexport const expConfig: KernelConfig = {\n  kernelName: Exp,\n  backendName: 'cpu',\n  kernelFunc: expKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\n\nexport const expm1Config: KernelConfig = {\n  kernelName: Expm1,\n  backendName: 'cpu',\n  kernelFunc: expm1KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\n\nexport const floorConfig: KernelConfig = {\n  kernelName: Floor,\n  backendName: 'cpu',\n  kernelFunc: floorKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\n\nexport const logConfig: KernelConfig = {\n  kernelName: Log,\n  backendName: 'cpu',\n  kernelFunc: logKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function maxImpl(\n    aVals: TypedArray, reduceSize: number, outShape: number[],\n    dtype: DataType): TypedArray {\n  const vals = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(outShape));\n\n  for (let i = 0; i < vals.length; ++i) {\n    const offset = i * reduceSize;\n    let max = aVals[offset];\n    for (let j = 0; j < reduceSize; ++j) {\n      const value = aVals[offset + j];\n      if (value > max) {\n        max = value;\n      }\n    }\n    vals[i] = max;\n  }\n  return vals;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Multiply} from '@tensorflow/tfjs-core';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const multiplyImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n      };\n    }));\n\nexport const multiply =\n    binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\n\nexport const multiplyConfig: KernelConfig = {\n  kernelName: Multiply,\n  backendName: 'cpu',\n  kernelFunc: multiply\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Rsqrt} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\n\nexport const rsqrtConfig: KernelConfig = {\n  kernelName: Rsqrt,\n  backendName: 'cpu',\n  kernelFunc: rsqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelConfig, KernelFunc, NumericDataType, Slice, slice_util, SliceAttrs, SliceInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function sliceImpl(\n    vals: TypedArray, begin: number[], size: number[], shape: number[],\n    dtype: DataType): TypedArray {\n  const isContinous = slice_util.isSliceContinous(shape, begin, size);\n  const length = util.sizeFromShape(size);\n  const xStrides = util.computeStrides(shape);\n\n  if (isContinous) {\n    const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n    return vals.subarray(flatOffset, flatOffset + length);\n  }\n\n  const outVals = util.getTypedArrayFromDType(dtype as NumericDataType, length);\n  for (let i = 0; i < length; ++i) {\n    const rank = size.length;\n    const strides = util.computeStrides(size);\n    const loc = util.indexToLoc(i, rank, strides);\n    const xLoc = loc.map((idx: number, j) => idx + begin[j]);\n    const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n    outVals[i] = vals[xIndex];\n  }\n  return outVals;\n}\n\nexport function slice(\n    args: {inputs: SliceInputs, backend: MathBackendCPU, attrs: SliceAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {begin, size} = attrs;\n\n  assertNotComplex(x, 'slice');\n\n  const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n  slice_util.assertParamsValid(x, $begin, $size);\n\n  const vals = backend.data.get(x.dataId).values as TypedArray;\n  const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n  return backend.makeTensorInfo($size, x.dtype, outVals);\n}\n\nexport const sliceConfig: KernelConfig = {\n  kernelName: Slice,\n  backendName: 'cpu',\n  kernelFunc: slice as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sub} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const subImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal - bReal, imag: aImag - bImag};\n    }));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\n\nexport const subConfig: KernelConfig = {\n  kernelName: Sub,\n  backendName: 'cpu',\n  kernelFunc: sub\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function transposeImpl(\n    xVals: TypedArray, xShape: number[], dtype: DataType, perm: number[],\n    newShape: number[]): TypedArray {\n  const xRank = xShape.length;\n  const xSize = util.sizeFromShape(xShape);\n  const xStrides = util.computeStrides(xShape);\n  const newStrides = util.computeStrides(newShape);\n\n  const result = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(newShape));\n\n  for (let i = 0; i < xSize; ++i) {\n    const loc = util.indexToLoc(i, xRank, xStrides);\n\n    // Permute location.\n    const newLoc: number[] = new Array(loc.length);\n    for (let i = 0; i < newLoc.length; i++) {\n      newLoc[i] = loc[perm[i]];\n    }\n\n    const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n    result[newIndex] = xVals[i];\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendValues, DataType, TensorBuffer, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function uniqueImpl(\n    values: BackendValues, axis: number, shape: number[], dtype: DataType): {\n  outputValues: BackendValues,\n  outputShape: number[],\n  indices: BackendValues\n} {\n  // Normalize and validate axis.\n  const $axis = util.parseAxisParam(axis, shape)[0];\n\n  // Calculate the new shape that is suitable for extracting data along the\n  // given axis.\n  //\n  // The rank is 3.\n  // The size of the 1st dimension is the size of all the axes < the given axis.\n  // The size of the 2nd dimension is the same as the size of the given axis.\n  // The size of the 3rd dimension is the size of all the axes > the given axis.\n  //\n  // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n  // newShape would be: [2*3, 5, 4].\n  //\n  // Note that this is not the final output shape. This will be the shape for an\n  // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n  // values along the given axis. To demonstrate how it works, consider the\n  // following example:\n  //\n  // Input: a 3D tensor, with shape [1, 2, 3]\n  // [\n  //   [\n  //      [1,2,3],\n  //      [4,5,6]\n  //   ]\n  // ]\n  // Axis: 2 (the last axis).\n  // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n  //\n  // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n  // 1*2. The re-shaped data would look like:\n  //\n  // [\n  //   [\n  //     [1], [2], [3]\n  //   ],\n  //   [\n  //     [4], [5], [6]\n  //   ]\n  // ]\n  //\n  // Then, we can construct a 3-level nested loop by the following dimension\n  // order to extract the values along the axis (dimension1):\n  // i: dimension1       // 0,1,2 (newShape[1])\n  //   m: dimension0     // 0,1   (newShape[0])\n  //     n: dimension2   // 0     (newShape[2])\n  //\n  //                       m, i, n\n  //                      ---------\n  // Iteration 0: data at [0, 0, 0] => \"1\"\n  // Iteration 1: data at [1, 0, 0] => \"4\"\n  // We got [1,4].\n  // Iteration 2: data at [0, 1, 0] => \"2\"\n  // Iteration 3: data at [1, 1, 0] => \"5\"\n  // We got [2,5].\n  // Iteration 4: data at [0, 2, 0] => \"3\"\n  // Iteration 5: data at [1, 2, 0] => \"6\"\n  // We got [3,6].\n  const newShape = [1, shape[0], 1];\n  for (let i = 0; i < $axis; i++) {\n    newShape[0] *= shape[i];\n  }\n  newShape[1] = shape[$axis];\n  for (let i = $axis + 1; i < shape.length; i++) {\n    newShape[2] *= shape[i];\n  }\n\n  // A map from unique elements (their string representations) to their values\n  // in \"indices\" (below).\n  const uniqueElements: {[key: string]: number} = {};\n  // The indices of each unique element in the original tensor along the given\n  // axis. It is 1D and has the same size as the given axis.\n  const indices = new Int32Array(shape[$axis]);\n  // Create a buffer so we can easily extract value at a given location.\n  const inputBuffer = new TensorBuffer(newShape, dtype, values as TypedArray);\n  // The indices along the given axis that have unique elements. This is a\n  // de-duped version of \"indices\" above.\n  const uniqueIndices: number[] = [];\n  const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n  for (let i = 0; i < shape[$axis]; i++) {\n    // Extract values along the axis.\n    let element: string;\n    if (is1DTensor) {\n      // Fast path for 1D tensor input.\n      element = values[i].toString();\n    } else {\n      const axisValues = [];\n      for (let m = 0; m < newShape[0]; m++) {\n        for (let n = 0; n < newShape[2]; n++) {\n          axisValues.push(inputBuffer.get(m, i, n));\n        }\n      }\n      element = axisValues.join(',');\n    }\n\n    // Dedup and update various indices.\n    if (uniqueElements[element] !== undefined) {\n      indices[i] = uniqueElements[element];\n    } else {\n      const uniqueIndex = Object.keys(uniqueElements).length;\n      uniqueElements[element] = uniqueIndex;\n      indices[i] = uniqueIndex;\n      uniqueIndices.push(i);\n    }\n  }\n\n  // Now we know where each of the unique elements are located along the axis\n  // (uniqueIndices). Extract them from input buffer and store them in the\n  // output buffer.\n  const outputTmpShape = newShape.slice();\n  outputTmpShape[1] = Object.keys(uniqueElements).length;\n  const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n  uniqueIndices.forEach((uniqueElementIndex, i) => {\n    for (let m = 0; m < newShape[0]; m++) {\n      for (let n = 0; n < newShape[2]; n++) {\n        outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n      }\n    }\n  });\n\n  // The output shape can be calculated from the input shape with the size of\n  // the given axis replaced by the number of unique elements along that axis.\n  const outputShape = shape.slice();\n  outputShape[$axis] = outputTmpShape[1];\n\n  return {\n    outputValues: outputBuffer.values as BackendValues,\n    outputShape,\n    indices,\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Shared kernel impls for use in other backends.\nexport {simpleAbsImpl} from './kernels/Abs';\nexport {addImpl} from './kernels/Add';\nexport {ceilImpl} from './kernels/Ceil';\nexport {expImpl} from './kernels/Exp';\nexport {expm1Impl} from './kernels/Expm1';\nexport {floorImpl} from './kernels/Floor';\nexport {logImpl} from './kernels/Log';\nexport {maxImpl} from './kernels/Max_impl';\nexport {multiplyImpl} from './kernels/Multiply';\nexport {rsqrtImpl} from './kernels/Rsqrt';\nexport {sliceImpl} from './kernels/Slice';\nexport {subImpl} from './kernels/Sub';\nexport {transposeImpl} from './kernels/Transpose_impl';\nexport {uniqueImpl} from './kernels/Unique_impl';\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/*\n * base.ts contains all the exports from tfjs-backend-cpu\n * without auto-kernel registration\n */\nimport {registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from './backend_cpu';\nimport * as shared from './shared';\n\nexport {MathBackendCPU} from './backend_cpu';\nexport {version as version_cpu} from './version';\nexport {shared};\n\n// Side effects for default initialization of MathBackendCPU\nregisterBackend('cpu', () => new MathBackendCPU(), 1 /* priority */);\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acosKernelFunc = unaryKernelFunc(Acos, (xi) => Math.acos(xi));\n\nexport const acosConfig: KernelConfig = {\n  kernelName: Acos,\n  backendName: 'cpu',\n  kernelFunc: acosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acoshKernelFunc = unaryKernelFunc(Acosh, (xi) => Math.acosh(xi));\n\nexport const acoshConfig: KernelConfig = {\n  kernelName: Acosh,\n  backendName: 'cpu',\n  kernelFunc: acoshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinKernelFunc = unaryKernelFunc(Asin, (xi) => Math.asin(xi));\n\nexport const asinConfig: KernelConfig = {\n  kernelName: Asin,\n  backendName: 'cpu',\n  kernelFunc: asinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinhKernelFunc = unaryKernelFunc(Asinh, (xi) => Math.asinh(xi));\n\nexport const asinhConfig: KernelConfig = {\n  kernelName: Asinh,\n  backendName: 'cpu',\n  kernelFunc: asinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanKernelFunc = unaryKernelFunc(Atan, (xi) => Math.atan(xi));\n\nexport const atanConfig: KernelConfig = {\n  kernelName: Atan,\n  backendName: 'cpu',\n  kernelFunc: atanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanhKernelFunc = unaryKernelFunc(Atanh, (xi) => Math.atanh(xi));\n\nexport const atanhConfig: KernelConfig = {\n  kernelName: Atanh,\n  backendName: 'cpu',\n  kernelFunc: atanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, buffer, DataType, Rank, TensorBuffer, TypedArray} from '@tensorflow/tfjs-core';\n\nexport function pool(\n    xValues: TypedArray, xShape: number[], dtype: DataType, strides: number[],\n    convInfo: backend_util.Conv2DInfo,\n    poolType: 'max'|'avg'): TensorBuffer<Rank, DataType> {\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const initialValue =\n      (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                            Number.POSITIVE_INFINITY);\n\n  const output = buffer(convInfo.outShape, dtype);\n  const outputVals = output.values;\n\n  const outputBatchStrides =\n      convInfo.outShape[1] * convInfo.outShape[2] * convInfo.outShape[3];\n  const outputRowStrides = convInfo.outShape[2] * convInfo.outShape[3];\n  const outputColStrides = convInfo.outShape[3];\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    const outputBatchOffset = b * outputBatchStrides;\n    const inputBatchOffset = b * strides[0];\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        const outputRowOffset = outputBatchOffset + yR * outputRowStrides;\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          const xCMin = Math.max(0, xCCorner);\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let minMaxValue = initialValue;\n          let avgValue = 0;\n          let count = 0;\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const xROffset = inputBatchOffset + xR * strides[1];\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const xCOffset = xROffset + xC * strides[2];\n              const pixel = xValues[xCOffset + d];\n              if ((poolType === 'max' && pixel > minMaxValue)) {\n                minMaxValue = pixel;\n              } else if (poolType === 'avg') {\n                avgValue += pixel;\n                count++;\n              }\n            }\n            if (isNaN(minMaxValue)) {\n              break;\n            }\n          }\n          const outputOffset = outputRowOffset + yC * outputColStrides + d;\n          outputVals[outputOffset] =\n              poolType === 'avg' ? avgValue / count : minMaxValue;\n        }\n      }\n    }\n  }\n  return output;\n}\n\nexport function maxPoolPositions(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    convInfo: backend_util.Conv2DInfo, flattenPositions = false,\n    includeBatchInIndex = false): TensorBuffer<Rank, 'int32'> {\n  const maxPositions = buffer(convInfo.outShape, 'int32');\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const xBuf = buffer(xShape, dtype, xValues);\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        let xRMin = xRCorner;\n        while (xRMin < 0) {\n          xRMin += dilationHeight;\n        }\n        // const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          let xCMin = xCCorner;\n          while (xCMin < 0) {\n            xCMin += dilationWidth;\n          }\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let maxValue = Number.NEGATIVE_INFINITY;\n          let maxPosition = -1;\n\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const wR = xR - xRCorner;\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const wC = xC - xCCorner;\n              const pixel = xBuf.get(b, xR, xC, d);\n              if (pixel > maxValue) {\n                maxValue = pixel as number;\n                if (flattenPositions) {\n                  maxPosition = includeBatchInIndex ?\n                      ((b * convInfo.inHeight + xR) * convInfo.inWidth + xC) *\n                              convInfo.inChannels +\n                          d :\n                      (xR * convInfo.inWidth + xC) * convInfo.inChannels + d;\n                } else {\n                  maxPosition = wR * effectiveFilterWidth + wC;\n                }\n              }\n            }\n          }\n          maxPositions.set(maxPosition, b, yR, yC, d);\n        }\n      }\n    }\n  }\n  return maxPositions;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function avgPool(\n    args:\n        {inputs: AvgPoolInputs, backend: MathBackendCPU, attrs: AvgPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'avg');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'cpu',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, buffer, KernelConfig, KernelFunc, Rank, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const filterHeight = convInfo.filterHeight;\n  const filterWidth = convInfo.filterWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel;\n            }\n          }\n          dx.set(dotProd * avgMultiplier, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function batchNormKernelFunc(args: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendCPU,\n  attrs: FusedBatchNormAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x, scale, offset, mean, variance} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  assertNotComplex([x, mean, variance, scale, offset], 'batchNorm');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const mVals = backend.data.get(mean.dataId).values as TypedArray;\n  const varVals = backend.data.get(variance.dataId).values as TypedArray;\n  const sVals = scale ? backend.data.get(scale.dataId).values as TypedArray :\n                        new Float32Array([1]);\n  const offVals = offset ?\n      backend.data.get(offset.dataId).values as TypedArray :\n      new Float32Array([0]);\n  const outVals = new Float32Array(xVals.length);\n\n  const offValsLength = offVals.length;\n  const sValsLength = sVals.length;\n  const varValsLength = varVals.length;\n  const mValsLength = mVals.length;\n\n  let offi = 0;\n  let mi = 0;\n  let si = 0;\n  let vi = 0;\n  for (let i = 0; i < xVals.length; ++i) {\n    outVals[i] = offVals[offi++] +\n        (xVals[i] - mVals[mi++]) * sVals[si++] /\n            Math.sqrt(varVals[vi++] + varianceEpsilon);\n    if (offi >= offValsLength) {\n      offi = 0;\n    }\n    if (mi >= mValsLength) {\n      mi = 0;\n    }\n    if (si >= sValsLength) {\n      si = 0;\n    }\n    if (vi >= varValsLength) {\n      vi = 0;\n    }\n  }\n  return backend.makeTensorInfo(x.shape, x.dtype, outVals);\n}\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'cpu',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const clipKernelFunc = unaryKernelFunc(ClipByValue, (xi, attrs) => {\n  const clipAttrs = attrs as {} as ClipByValueAttrs;\n  if (xi > clipAttrs.clipValueMax) {\n    return clipAttrs.clipValueMax;\n  }\n  return xi < clipAttrs.clipValueMin ? clipAttrs.clipValueMin : xi;\n});\n\nexport const clipConfig: KernelConfig = {\n  kernelName: ClipByValue,\n  backendName: 'cpu',\n  kernelFunc: clipKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Imag, ImagInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function imag(args: {inputs: ImagInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const imag = backend.data.get(input.dataId).complexTensorInfos.imag;\n  const imagVal = backend.data.get(imag.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the imag value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(imag.shape, imag.dtype, imagVal);\n}\n\nexport const imagConfig: KernelConfig = {\n  kernelName: Imag,\n  backendName: 'cpu',\n  kernelFunc: imag as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function reshape(\n    args:\n        {inputs: ReshapeInputs, backend: MathBackendCPU, attrs: ReshapeAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  backend.incRef(x.dataId);\n\n  const xData = backend.data.get(x.dataId);\n\n  if (xData.complexTensorInfos != null) {\n    const real = xData.complexTensorInfos.real;\n    const imag = xData.complexTensorInfos.imag;\n\n    real.shape = $shape;\n    imag.shape = $shape;\n  }\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'cpu',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Concat, ConcatAttrs, ConcatInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nimport {complex} from './Complex';\nimport {imag} from './Imag';\nimport {real} from './Real';\nimport {reshape} from './Reshape';\n\nexport function concat(\n    args: {inputs: ConcatInputs, backend: MathBackendCPU, attrs: ConcatAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {axis} = attrs;\n\n  const $axis = util.parseAxisParam(axis, inputs[0].shape)[0];\n  let outShape = backend_util.computeOutShape(inputs.map(t => t.shape), $axis);\n\n  if (util.sizeFromShape(outShape) === 0) {\n    return backend.makeTensorInfo(outShape, inputs[0].dtype, []);\n  }\n\n  // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n  const $inputs = inputs.filter(t => util.sizeFromShape(t.shape) > 0);\n  if ($inputs.length === 1) {\n    return $inputs[0];\n  }\n\n  const shapes = $inputs.map(t => t.shape);\n  backend_util.assertParamsConsistent(shapes, $axis);\n\n  if ($inputs[0].dtype === 'complex64') {\n    const reals = $inputs.map((t) => real({inputs: {input: t}, backend}));\n    const imags = $inputs.map((t) => imag({inputs: {input: t}, backend}));\n\n    const realConcated = concat({inputs: reals, backend, attrs: {axis}});\n    const imagConcated = concat({inputs: imags, backend, attrs: {axis}});\n\n    const result =\n        complex({inputs: {real: realConcated, imag: imagConcated}, backend});\n\n    reals.forEach(r => backend.disposeIntermediateTensorInfo(r));\n    imags.forEach(i => backend.disposeIntermediateTensorInfo(i));\n    backend.disposeIntermediateTensorInfo(realConcated);\n    backend.disposeIntermediateTensorInfo(imagConcated);\n\n    return result;\n  }\n\n  // Any concat of n-dimensional tensors across any axis can be reduced to\n  // a concatenation of two-dimensional tensors across the axis 1 by first\n  // partitioning the axes of the original tensors into those less than the\n  // axis to be concatenated and the rest. Then reshape the tensors\n  // into a two-dimensional tensor by collapsing these two sets of axes and\n  // concatenate the resulting matrices across the axis 1, finally reshaping\n  // the result to have the proper shape.\n  const inputs2D = $inputs.map(t => {\n    const innerSize = util.sizeFromShape(t.shape.slice($axis));\n    const shape = [-1, innerSize];\n    return reshape({inputs: {x: t}, backend, attrs: {shape}});\n  });\n\n  // Concats 2d tensors along axis=1.\n  outShape =\n      backend_util.computeOutShape(inputs2D.map(t => t.shape), 1 /* axis */);\n\n  const outVals = util.getTypedArrayFromDType(\n      $inputs[0].dtype as 'float32', util.sizeFromShape(outShape));\n\n  if (inputs2D[0].shape[0] === 1) {\n    // Use built-in TypedArray.set() method for speed.\n    let offset = 0;\n    inputs2D.forEach(t => {\n      const val = backend.data.get(t.dataId).values as TypedArray;\n      const size = util.sizeFromShape(t.shape);\n\n      outVals.set(val, offset);\n      offset += size;\n    });\n  } else {\n    let colOffset = 0;\n\n    inputs2D.forEach(t => {\n      const tVals = backend.data.get(t.dataId).values as TypedArray;\n\n      let tIdx = 0;\n\n      for (let row = 0; row < t.shape[0]; ++row) {\n        const resIdx = row * outShape[1] + colOffset;\n        for (let col = 0; col < t.shape[1]; ++col) {\n          outVals[resIdx + col] = tVals[tIdx++];\n        }\n      }\n\n      colOffset += t.shape[1];\n    });\n  }\n\n  const finalOutShape =\n      backend_util.computeOutShape($inputs.map(t => t.shape), $axis);\n\n  const outInfo =\n      backend.makeTensorInfo(finalOutShape, inputs[0].dtype, outVals);\n\n  inputs2D.forEach(t => backend.disposeIntermediateTensorInfo(t));\n\n  return outInfo;\n}\n\nexport const concatConfig: KernelConfig = {\n  kernelName: Concat,\n  backendName: 'cpu',\n  kernelFunc: concat as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const cosKernelFunc = unaryKernelFunc(Cos, (xi) => Math.cos(xi));\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'cpu',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const coshKernelFunc = unaryKernelFunc(Cosh, (xi) => Math.cosh(xi));\n\nexport const coshConfig: KernelConfig = {\n  kernelName: Cosh,\n  backendName: 'cpu',\n  kernelFunc: coshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2D, Dilation2DAttrs, Dilation2DInputs, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dConfig: KernelConfig = {\n  kernelName: Dilation2D,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter} = inputs as Dilation2DInputs;\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xRank = x.shape.length;\n\n    const filterVals = cpuBackend.data.get(filter.dataId).values as TypedArray;\n    const filterRank = filter.shape.length;\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    const outSize = util.sizeFromShape(outShape);\n    const outRank = outShape.length;\n    const outputVals = util.getArrayFromDType(x.dtype, outSize);\n\n    // Upsampling the input by fill in `dilation size - 1` values between each\n    // input value.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const xIndex = util.locToIndex(\n                        [b, hIn, wIn, d], xRank, util.computeStrides(x.shape));\n                    const filterIndex = util.locToIndex(\n                        [h, w, d], filterRank,\n                        util.computeStrides(filter.shape));\n                    const val = xVals[xIndex] + filterVals[filterIndex];\n                    if (val > curVal) {\n                      curVal = val;\n                    }\n                  }\n                }\n              }\n            }\n            const outputIndex = util.locToIndex(\n                [b, hOut, wOut, d], outRank, util.computeStrides(outShape));\n            outputVals[outputIndex] = curVal;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(outputVals, x.dtype), outShape, x.dtype);\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropFilter, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropFilterConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropFilter,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropFilter}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed filter gradients has the same dimensions as the filter:\n    // [filterHeight, filterWidth, depth]\n    const gradients = util.makeZerosNestedTypedArray(\n                          filter.shape, filter.dtype) as number[][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hMax = 0;\n            let wMax = 0;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hMax = h;\n                      wMax = w;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[hMax][wMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), filter.shape, filter.dtype);\n\n    return {dataId, shape: filter.shape, dtype: filter.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropInput, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropInputConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropInput,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropInput}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed gradients has the same dimensions as the input:\n    // [batch, inputHeight, inputCols, inChannel]\n    const gradients =\n        util.makeZerosNestedTypedArray(x.shape, x.dtype) as number[][][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hInMax = (hBeg < 0) ? 0 : hBeg;\n            let wInMax = (wBeg < 0) ? 0 : wBeg;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hInMax = hIn;\n                      wInMax = wIn;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[b][hInMax][wInMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), x.shape, x.dtype);\n\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const divImpl =\n    createSimpleBinaryKernelImpl((a: number, b: number) => a / b);\nexport const div = binaryKernelFunc(Div, divImpl);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'cpu',\n  kernelFunc: div\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Elu, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const eluKernelFunc =\n    unaryKernelFunc(Elu, (xi) => xi >= 0 ? xi : (Math.exp(xi) - 1));\n\nexport const eluConfig: KernelConfig = {\n  kernelName: Elu,\n  backendName: 'cpu',\n  kernelFunc: eluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Erf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst p = backend_util.ERF_P;\nconst a1 = backend_util.ERF_A1;\nconst a2 = backend_util.ERF_A2;\nconst a3 = backend_util.ERF_A3;\nconst a4 = backend_util.ERF_A4;\nconst a5 = backend_util.ERF_A5;\n\nexport const erfKernelFunc = unaryKernelFunc(\n    Erf,\n    (xi) => {\n      const sign = Math.sign(xi);\n      const v = Math.abs(xi);\n      const t = 1.0 / (1.0 + p * v);\n      return sign *\n          (1.0 -\n           (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t *\n               Math.exp(-v * v));\n    },\n);\n\nexport const erfConfig: KernelConfig = {\n  kernelName: Erf,\n  backendName: 'cpu',\n  kernelFunc: erfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Tensor, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {add} from '../kernels/Add';\nimport {complex} from '../kernels/Complex';\nimport {concat} from '../kernels/Concat';\nimport {divConfig} from '../kernels/Div';\nimport {identity} from '../kernels/Identity';\nimport {imag} from '../kernels/Imag';\nimport {multiply} from '../kernels/Multiply';\nimport {real} from '../kernels/Real';\nimport {slice} from '../kernels/Slice';\nimport {sub} from '../kernels/Sub';\n\n/**\n * Calculate FFT of inner most elements of batch tensor.\n */\nexport function fftBatch(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): TensorInfo {\n  const inputShape = input.shape;\n  const batch = inputShape[0];\n  const innerDim = inputShape[1];\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const real2D = inputVals.complexTensorInfos.real;\n  const imag2D = inputVals.complexTensorInfos.imag;\n\n  // Collects real and imaginary values separately.\n  const resultShape = [batch, innerDim];\n  const resultSize = util.sizeFromShape(resultShape);\n  const resultReal = util.getTypedArrayFromDType('float32', resultSize);\n  const resultImag = util.getTypedArrayFromDType('float32', resultSize);\n\n  for (let b = 0; b < batch; b++) {\n    // TODO: Support slice ops for complex type.\n    const r = slice({\n      inputs: {x: real2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n    const i = slice({\n      inputs: {x: imag2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n\n    const input = complex({inputs: {real: r, imag: i}, backend: cpuBackend});\n\n    // Run FFT by batch element.\n    const {real, imag} = fftImpl(input, inverse, cpuBackend);\n    const res = backend_util.mergeRealAndImagArrays(real, imag);\n\n    for (let d = 0; d < innerDim; d++) {\n      const c = backend_util.getComplexWithIndex(res, d);\n      resultReal[b * innerDim + d] = c.real;\n      resultImag[b * innerDim + d] = c.imag;\n    }\n\n    cpuBackend.disposeIntermediateTensorInfo(r);\n    cpuBackend.disposeIntermediateTensorInfo(i);\n    cpuBackend.disposeIntermediateTensorInfo(input);\n  }\n\n  const $realInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultReal);\n  const $imagInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultImag);\n\n  const result = complex(\n      {inputs: {real: $realInfo, imag: $imagInfo}, backend: cpuBackend});\n\n  cpuBackend.disposeIntermediateTensorInfo($realInfo);\n  cpuBackend.disposeIntermediateTensorInfo($imagInfo);\n\n  return result;\n}\n\nexport function fftImpl(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  const inputSize = util.sizeFromShape(input.shape);\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const realVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.real.dataId).values as\n      Float32Array;\n\n  const imagVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.imag.dataId).values as\n      Float32Array;\n\n  if (isExponentOf2(inputSize)) {\n    const result =\n        fftRadix2(realVals, imagVals, inputSize, inverse, cpuBackend);\n\n    const resultShape = [input.shape[0], input.shape[1]];\n\n    if (inverse) {\n      const realInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.real);\n      const imagInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.imag);\n\n      const sizeInfo: TensorInfo = cpuBackend.makeTensorInfo(\n          [], 'float32',\n          util.createScalarValue(inputSize as {} as 'float32', 'float32'));\n      const sizeInfoCopy =\n          identity({inputs: {x: sizeInfo}, backend: cpuBackend});\n\n      const divRealInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: realInfo, b: sizeInfo}, backend: cpuBackend}) as\n          TensorInfo;\n      const divImagInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: imagInfo, b: sizeInfoCopy}, backend: cpuBackend}) as\n          TensorInfo;\n\n      const divRealVals =\n          cpuBackend.data.get(divRealInfo.dataId).values as Float32Array;\n      const divImagVals =\n          cpuBackend.data.get(divImagInfo.dataId).values as Float32Array;\n\n      cpuBackend.disposeIntermediateTensorInfo(realInfo);\n      cpuBackend.disposeIntermediateTensorInfo(imagInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfoCopy);\n      cpuBackend.disposeIntermediateTensorInfo(divRealInfo);\n      cpuBackend.disposeIntermediateTensorInfo(divImagInfo);\n\n      return {real: divRealVals, imag: divImagVals};\n    }\n\n    return result;\n  } else {\n    const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n    const rawOutput =\n        fourierTransformByMatmul(data, inputSize, inverse) as Float32Array;\n\n    return backend_util.splitRealAndImagArrays(rawOutput);\n  }\n}\n\nfunction isExponentOf2(size: number): boolean {\n  return (size & size - 1) === 0;\n}\n\n// FFT using Cooley-Tukey algorithm on radix 2 dimensional input.\nfunction fftRadix2(\n    realVals: Float32Array, imagVals: Float32Array, size: number,\n    inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  if (size === 1) {\n    return {real: realVals, imag: imagVals};\n  }\n\n  const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n  const half = size / 2;\n\n  const evenComplex = backend_util.complexWithEvenIndex(data);\n\n  const evenRealVals = evenComplex.real;\n  const evenImagVals = evenComplex.imag;\n\n  const evenShape = [evenRealVals.length];\n\n  const evenRealInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenRealVals);\n  const evenImagInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenImagVals);\n\n  const evenTensorInfo = complex(\n      {inputs: {real: evenRealInfo, imag: evenImagInfo}, backend: cpuBackend});\n\n  const oddComplex = backend_util.complexWithOddIndex(data);\n\n  const oddRealVals = oddComplex.real;\n  const oddImagVals = oddComplex.imag;\n\n  const oddShape = [oddRealVals.length];\n\n  const oddRealInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddRealVals);\n  const oddImagInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddImagVals);\n\n  const oddTensorInfo = complex(\n      {inputs: {real: oddRealInfo, imag: oddImagInfo}, backend: cpuBackend});\n\n  // Recursive call for half part of original input.\n  const $evenComplex =\n      fftRadix2(evenRealVals, evenImagVals, half, inverse, cpuBackend);\n\n  const $evenRealVals = $evenComplex.real;\n  const $evenImagVals = $evenComplex.imag;\n\n  const $evenShape = [$evenRealVals.length];\n\n  const $evenRealInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenRealVals);\n  const $evenImagInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenImagVals);\n\n  const $evenTensorInfo = complex({\n    inputs: {real: $evenRealInfo, imag: $evenImagInfo},\n    backend: cpuBackend\n  });\n\n  const $oddComplex =\n      fftRadix2(oddRealVals, oddImagVals, half, inverse, cpuBackend);\n\n  const $oddRealVals = $oddComplex.real;\n  const $oddImagVals = $oddComplex.imag;\n\n  const $oddShape = [$oddRealVals.length];\n\n  const $oddRealInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddRealVals);\n  const $oddImagInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddImagVals);\n\n  const $oddTensorInfo = complex(\n      {inputs: {real: $oddRealInfo, imag: $oddImagInfo}, backend: cpuBackend});\n\n  const e = backend_util.exponents(size, inverse);\n  const eShape = [e.real.length];\n\n  const eRealInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.real);\n  const eImagInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.imag);\n\n  const complexInfo = complex(\n      {inputs: {real: eRealInfo, imag: eImagInfo}, backend: cpuBackend});\n\n  const exponentInfo =\n      multiply(\n          {inputs: {a: complexInfo, b: $oddTensorInfo}, backend: cpuBackend}) as\n      TensorInfo;\n\n  const addPart = add({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n  const subPart = sub({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n\n  const addPartReal = real({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartReal = real({inputs: {input: subPart}, backend: cpuBackend});\n\n  const addPartImag = imag({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartImag = imag({inputs: {input: subPart}, backend: cpuBackend});\n\n  const $real = concat({\n    inputs: [addPartReal as Tensor, subPartReal as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n  const $imag = concat({\n    inputs: [addPartImag as Tensor, subPartImag as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n\n  const $realVals = cpuBackend.data.get($real.dataId).values as Float32Array;\n  const $imagVals = cpuBackend.data.get($imag.dataId).values as Float32Array;\n\n  cpuBackend.disposeIntermediateTensorInfo(evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(complexInfo);\n  cpuBackend.disposeIntermediateTensorInfo(exponentInfo);\n  cpuBackend.disposeIntermediateTensorInfo(addPart);\n  cpuBackend.disposeIntermediateTensorInfo(subPart);\n  cpuBackend.disposeIntermediateTensorInfo(addPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(addPartImag);\n  cpuBackend.disposeIntermediateTensorInfo(subPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(subPartImag);\n  cpuBackend.disposeIntermediateTensorInfo($real);\n  cpuBackend.disposeIntermediateTensorInfo($imag);\n\n  return {real: $realVals, imag: $imagVals};\n}\n\n// Calculate fourier transform by multplying sinusoid matrix.\nfunction fourierTransformByMatmul(\n    data: TypedArray, size: number, inverse: boolean): TypedArray {\n  const ret = new Float32Array(size * 2);\n  // TODO: Use matmul instead once it supports complex64 type.\n  for (let r = 0; r < size; r++) {\n    let real = 0.0;\n    let imag = 0.0;\n    for (let c = 0; c < size; c++) {\n      const e = backend_util.exponent(r * c, size, inverse);\n      const term = backend_util.getComplexWithIndex(data as Float32Array, c);\n      real += term.real * e.real - term.imag * e.imag;\n      imag += term.real * e.imag + term.imag * e.real;\n    }\n    if (inverse) {\n      real /= size;\n      imag /= size;\n    }\n    backend_util.assignToTypedArray(ret, real, imag, r);\n  }\n  return ret;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FFT, FFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function fft(args: {inputs: FFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, false, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const fftConfig: KernelConfig = {\n  kernelName: FFT,\n  backendName: 'cpu',\n  kernelFunc: fft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n\n            const coordX = Math.round(imageWidth - x);\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n\n            let outputValue = imageVals[outIdx];\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth) {\n              // set the output to the image value at the coordinate position.\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n            output[outIdx] = outputValue;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IFFT, IFFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function ifft(args: {inputs: IFFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, true, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const ifftConfig: KernelConfig = {\n  kernelName: IFFT,\n  backendName: 'cpu',\n  kernelFunc: ifft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isFiniteKernelFunc =\n    unaryKernelFunc(IsFinite, (xi) => Number.isFinite(xi) ? 1 : 0, 'bool');\n\nexport const isFiniteConfig: KernelConfig = {\n  kernelName: IsFinite,\n  backendName: 'cpu',\n  kernelFunc: isFiniteKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isInfKernelFunc =\n    unaryKernelFunc(IsInf, (xi) => Math.abs(xi) === Infinity ? 1 : 0, 'bool');\n\nexport const isInfConfig: KernelConfig = {\n  kernelName: IsInf,\n  backendName: 'cpu',\n  kernelFunc: isInfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isNaNKernelFunc =\n    unaryKernelFunc(IsNan, (xi) => Number.isNaN(xi) ? 1 : 0, 'bool');\n\nexport const isNaNConfig: KernelConfig = {\n  kernelName: IsNan,\n  backendName: 'cpu',\n  kernelFunc: isNaNKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log1p} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const log1pKernelFunc = unaryKernelFunc(Log1p, (xi) => Math.log1p(xi));\n\nexport const log1pConfig: KernelConfig = {\n  kernelName: Log1p,\n  backendName: 'cpu',\n  kernelFunc: log1pKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, LogicalNot} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const logicalNotKernelFunc =\n    unaryKernelFunc(LogicalNot, (xi) => xi ? 0 : 1, 'bool');\n\nexport const logicalNotConfig: KernelConfig = {\n  kernelName: LogicalNot,\n  backendName: 'cpu',\n  kernelFunc: logicalNotKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig} from '@tensorflow/tfjs-core';\nimport {TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    let xShape = x.shape;\n    const xRank = xShape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, xShape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    let xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    if (permutedAxes != null) {\n      const newShape: number[] = new Array(xRank);\n      for (let i = 0; i < newShape.length; i++) {\n        newShape[i] = xShape[permutedAxes[i]];\n      }\n\n      xVals = transposeImpl(xVals, xShape, x.dtype, permutedAxes, newShape);\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n\n      xShape = newShape;\n    }\n\n    assertNotComplex(x, 'max');\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(xShape, axes);\n\n    const reduceSize = util.sizeFromShape(reduceShape);\n\n    const result = maxImpl(xVals, reduceSize, maxOutShape, x.dtype);\n    const dataId = cpuBackend.write(result, maxOutShape, x.dtype);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // reshape\n      const newShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n      outShape = newShape;\n    }\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function maxPool(\n    args:\n        {inputs: MaxPoolInputs, backend: MathBackendCPU, attrs: MaxPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'max');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'cpu',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, buffer, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, Rank, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {maxPoolPositions} from '../utils/pool_utils';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const xValues = backend.data.get(x.dataId).values as TypedArray;\n  const maxPosBuf = buffer(\n      convInfo.outShape, x.dtype,\n      maxPoolPositions(xValues, x.shape, x.dtype, convInfo).values);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n              const maxPos = effectiveFilterHeight * effectiveFilterWidth - 1 -\n                  (maxPosBuf.get(b, dyR, dyC, d) as number);\n              const curPos = wR * effectiveFilterWidth + wC;\n\n              const mask = maxPos === curPos ? 1 : 0;\n              if (mask === 0) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel * mask;\n            }\n          }\n          dx.set(dotProd, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, DataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {maxPoolPositions, pool} from '../utils/pool_utils';\nexport function maxPoolWithArgmaxImpl(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    includeBatchInIndex: boolean, convInfo: backend_util.Conv2DInfo) {\n  const strides = util.computeStrides(xShape);\n  const maxPools = pool(xValues, xShape, dtype, strides, convInfo, 'max');\n  const maxPositions = maxPoolPositions(\n      xValues, xShape, dtype, convInfo, true, includeBatchInIndex);\n\n  return [maxPools.values, maxPositions.values];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'MaxPoolWithArgmax');\n\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        [1, 1], pad);\n    const [pooled, indexes] = maxPoolWithArgmaxImpl(\n        values, x.shape, x.dtype, includeBatchInIndex, convInfo);\n\n    const pooledDataId =\n        cpuBackend.write(pooled as Float32Array, convInfo.outShape, x.dtype);\n    const indexesDataId =\n        cpuBackend.write(indexes as Int32Array, convInfo.outShape, x.dtype);\n    return [\n      {dataId: pooledDataId, shape: convInfo.outShape, dtype: x.dtype},\n      {dataId: indexesDataId, shape: convInfo.outShape, dtype: 'int32'}\n    ];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionPadded');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionWithScore');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NotEqual} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const notEqualImpl =\n    createSimpleBinaryKernelImpl(((a, b) => (a !== b) ? 1 : 0));\nexport const notEqual =\n    binaryKernelFunc(NotEqual, notEqualImpl, null /* complexOp */, 'bool');\n\nexport const notEqualConfig: KernelConfig = {\n  kernelName: NotEqual,\n  backendName: 'cpu',\n  kernelFunc: notEqual\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, NumericDataType, PadV2, PadV2Attrs, PadV2Inputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function padV2(\n    args: {inputs: PadV2Inputs, backend: MathBackendCPU, attrs: PadV2Attrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {paddings, constantValue} = attrs;\n\n  assertNotComplex(x, 'pad');\n\n  const outShape = paddings.map(\n      (p, i) => p[0] /* beforePad */ + x.shape[i] + p[1] /* afterPad */);\n\n  const start = paddings.map(p => p[0]);\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const xSize = util.sizeFromShape(x.shape);\n  const xRank = x.shape.length;\n  const xStrides = util.computeStrides(x.shape);\n\n  const resultSize = util.sizeFromShape(outShape);\n  const resultRank = outShape.length;\n  const resultStrides = util.computeStrides(outShape);\n  const resVals =\n      util.getTypedArrayFromDType(x.dtype as NumericDataType, resultSize);\n\n  if (constantValue !== 0) {\n    resVals.fill(constantValue);\n  }\n\n  for (let i = 0; i < xSize; i++) {\n    const coords = util.indexToLoc(i, xRank, xStrides);\n    const outCoords = coords.map((c, i) => c + start[i]);\n    const outIndex = util.locToIndex(outCoords, resultRank, resultStrides);\n\n    resVals[outIndex] = xVals[i];\n  }\n\n  const outId = backend.write(resVals, outShape, x.dtype);\n\n  return {dataId: outId, shape: outShape, dtype: x.dtype};\n}\n\nexport const padV2Config: KernelConfig = {\n  kernelName: PadV2,\n  backendName: 'cpu',\n  kernelFunc: padV2 as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Reciprocal} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const reciprocalKernelFunc = unaryKernelFunc(Reciprocal, (xi) => 1 / xi);\n\nexport const reciprocalConfig: KernelConfig = {\n  kernelName: Reciprocal,\n  backendName: 'cpu',\n  kernelFunc: reciprocalKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {backend_util, RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const fullOpacityValue = 255;\n\n    const sinFactor = Math.sin(radians);\n    const cosFactor = Math.cos(radians);\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n            const y = coords[1];\n\n            // coordX/coordY are the result of rotating and translating x/y.\n            let coordX = (x - centerX) * cosFactor - (y - centerY) * sinFactor;\n            let coordY = (x - centerX) * sinFactor + (y - centerY) * cosFactor;\n            coordX = Math.round(coordX + centerX);\n            coordY = Math.round(coordY + centerY);\n\n            let outputValue = fillValue;\n            if (typeof fillValue !== 'number') {\n              if (channel === 3) {\n                outputValue = fullOpacityValue;\n              } else {\n                outputValue = fillValue[channel];\n              }\n            }\n\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth && coordY >= 0 &&\n                coordY < imageHeight) {\n              // set the output to the image value at the coordinate position.\n              const rotatedRowOffset = coordY * (imageWidth * numChannels);\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rotatedRowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n            output[outIdx] = outputValue as number;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Round} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const roundKernelFunc = unaryKernelFunc(Round, (xi) => {\n  // The algorithm is based on banker's rounding.\n  const base = Math.floor(xi);\n  if (xi - base < 0.5) {\n    return Math.floor(xi);\n  } else if (xi - base > 0.5) {\n    return Math.ceil(xi);\n  } else {\n    if (base % 2.0 === 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n});\n\nexport const roundConfig: KernelConfig = {\n  kernelName: Round,\n  backendName: 'cpu',\n  kernelFunc: roundKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, Selu} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst scaleAlpha = backend_util.SELU_SCALEALPHA;\nconst scale = backend_util.SELU_SCALE;\n\nexport const seluKernelFunc = unaryKernelFunc(Selu, (xi) => {\n  if (xi >= 0) {\n    return scale * xi;\n  } else {\n    return scaleAlpha * (Math.exp(xi) - 1);\n  }\n});\n\nexport const seluConfig: KernelConfig = {\n  kernelName: Selu,\n  backendName: 'cpu',\n  kernelFunc: seluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sigmoid} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sigmoidKernelFunc =\n    unaryKernelFunc(Sigmoid, (xi) => 1 / (1 + Math.exp(-xi)));\n\nexport const sigmoidConfig: KernelConfig = {\n  kernelName: Sigmoid,\n  backendName: 'cpu',\n  kernelFunc: sigmoidKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sign} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const signKernelFunc = unaryKernelFunc(Sign, (xi) => {\n  if (xi < 0) {\n    return -1;\n  } else if (xi > 0) {\n    return 1;\n  } else {\n    return 0;\n  }\n});\n\nexport const signConfig: KernelConfig = {\n  kernelName: Sign,\n  backendName: 'cpu',\n  kernelFunc: signKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinKernelFunc = unaryKernelFunc(Sin, (xi) => Math.sin(xi));\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'cpu',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sinh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinhKernelFunc = unaryKernelFunc(Sinh, (xi) => Math.sinh(xi));\n\nexport const sinhConfig: KernelConfig = {\n  kernelName: Sinh,\n  backendName: 'cpu',\n  kernelFunc: sinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Softplus} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\n// mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n\n// epsilon is the difference between 1.0 and the next representable float.\n// For a single precision 32 bit float this should be 2^-23, see:\n// https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\nconst epsilon = 1.1920928955078125e-7;\nconst threshold = Math.log(epsilon) + 2.0;\n\nexport const softplusKernelFunc = unaryKernelFunc(Softplus, (xi) => {\n  // Value above which exp(x) may overflow, but softplus(x) == x\n  // is within machine epsilon.\n  const tooLarge = xi > -threshold;\n\n  // Value below which exp(x) may underflow, but softplus(x) == exp(x)\n  // is within machine epsilon.\n  const tooSmall = xi < threshold;\n\n  const expX = Math.exp(xi);\n  let result;\n\n  if (tooSmall) {\n    result = expX;\n  } else if (tooLarge) {\n    result = xi;\n  } else {\n    result = Math.log(1.0 + expX);\n  }\n  return result;\n});\n\nexport const softplusConfig: KernelConfig = {\n  kernelName: Softplus,\n  backendName: 'cpu',\n  kernelFunc: softplusKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {transposeImpl} from './Transpose_impl';\n\nexport function transpose(args: {\n  inputs: TransposeInputs,\n  attrs: TransposeAttrs,\n  backend: MathBackendCPU\n}): TensorInfo {\n  const {inputs, attrs, backend} = args;\n  const {x} = inputs;\n  const {perm} = attrs;\n\n  assertNotComplex(x, 'transpose');\n\n  const xRank = x.shape.length;\n\n  const newShape: number[] = new Array(xRank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = x.shape[perm[i]];\n  }\n\n  const values = backend.data.get(x.dataId).values as TypedArray;\n  const result = transposeImpl(values, x.shape, x.dtype, perm, newShape);\n\n  const dataId = backend.write(result, newShape, x.dtype);\n  return {dataId, shape: newShape, dtype: x.dtype};\n}\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'cpu',\n  kernelFunc: transpose as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, KernelFunc, ReshapeAttrs, ReshapeInputs, SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs, TensorInfo, TransposeAttrs, TransposeInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {padV2Config} from './PadV2';\nimport {reshape} from './Reshape';\nimport {transpose} from './Transpose';\n\nexport function spaceToBatchND(args: {\n  inputs: SpaceToBatchNDInputs,\n  backend: MathBackendCPU,\n  attrs: SpaceToBatchNDAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {blockShape, paddings} = attrs;\n\n  assertNotComplex([x], 'spaceToBatchND');\n\n  const prod = util.sizeFromShape(blockShape);\n\n  const completePaddings: Array<[number, number]> = [[0, 0]];\n  completePaddings.push(...(paddings as Array<[number, number]>));\n\n  for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n    completePaddings.push([0, 0]);\n  }\n\n  const paddedX = padV2Config.kernelFunc({\n    inputs: {x},\n    backend,\n    attrs: {paddings: completePaddings, constantValue: 0}\n  }) as TensorInfo;\n\n  const reshapedPaddedShape =\n      backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n  const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n      reshapedPaddedShape.length, blockShape.length, false);\n\n  const flattenShape =\n      backend_util.getReshapedPermuted(paddedX.shape, blockShape, prod, false);\n\n  const reshapeInputs: ReshapeInputs = {x: paddedX};\n  const reshapeAttrs: ReshapeAttrs = {shape: reshapedPaddedShape};\n  const paddedXReshaped =\n      reshape({inputs: reshapeInputs, backend, attrs: reshapeAttrs});\n\n  const transposeInputs: TransposeInputs = {x: paddedXReshaped};\n  const transposeAttrs:\n      TransposeAttrs = {perm: permutedReshapedPaddedPermutation};\n  const paddedXT =\n      transpose({inputs: transposeInputs, backend, attrs: transposeAttrs});\n\n  const resultReshapeInputs: ReshapeInputs = {x: paddedXT};\n  const resultReshapeAttrs: ReshapeAttrs = {shape: flattenShape};\n  const result = reshape(\n      {inputs: resultReshapeInputs, backend, attrs: resultReshapeAttrs});\n\n  backend.disposeIntermediateTensorInfo(paddedX);\n  backend.disposeIntermediateTensorInfo(paddedXReshaped);\n  backend.disposeIntermediateTensorInfo(paddedXT);\n\n  return result;\n}\n\nexport const spaceToBatchNDConfig: KernelConfig = {\n  kernelName: SpaceToBatchND,\n  backendName: 'cpu',\n  kernelFunc: spaceToBatchND as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sqrt} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sqrtKernelFunc = unaryKernelFunc(Sqrt, (xi) => Math.sqrt(xi));\n\nexport const sqrtConfig: KernelConfig = {\n  kernelName: Sqrt,\n  backendName: 'cpu',\n  kernelFunc: sqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square, SquareInputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend}) => {\n    const {x} = inputs as SquareInputs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'square');\n\n    const values = cpuBackend.data.get(x.dataId).values as Float32Array;\n    const newValues = new Float32Array(values.length);\n    for (let i = 0; i < values.length; ++i) {\n      const value = values[i];\n      newValues[i] = value * value;\n    }\n    const dataId = cpuBackend.write(newValues, x.shape, x.dtype);\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const squaredDifferenceImpl = createSimpleBinaryKernelImpl(((a, b) => {\n  const diff = a - b;\n  return diff * diff;\n}));\nexport const squaredDifference =\n    binaryKernelFunc(SquaredDifference, squaredDifferenceImpl);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'cpu',\n  kernelFunc: squaredDifference\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Step, StepAttrs} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const stepKernelFunc = unaryKernelFunc(Step, (xi, attrs) => {\n  const stepAttrs = attrs as {} as StepAttrs;\n  if (isNaN(xi)) {\n    return NaN;\n  } else {\n    return xi > 0 ? 1 : stepAttrs.alpha;\n  }\n});\n\nexport const stepConfig: KernelConfig = {\n  kernelName: Step,\n  backendName: 'cpu',\n  kernelFunc: stepKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanKernelFunc = unaryKernelFunc(Tan, (xi) => Math.tan(xi));\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'cpu',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tanh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanhKernelFunc = unaryKernelFunc(Tanh, (xi) => Math.tanh(xi));\n\nexport const tanhConfig: KernelConfig = {\n  kernelName: Tanh,\n  backendName: 'cpu',\n  kernelFunc: tanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {uniqueImpl} from './Unique_impl';\n\nexport function unique(\n    args: {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendCPU}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  const values = backend.data.get(x.dataId).values;\n  const {outputValues, outputShape, indices} =\n      uniqueImpl(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'cpu',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// We explicitly import the modular kernels so they get registered in the\n// global registry when we compile the library. A modular build would replace\n// the contents of this file and import only the kernels that are needed.\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {absConfig} from './kernels/Abs';\nimport {acosConfig} from './kernels/Acos';\nimport {acoshConfig} from './kernels/Acosh';\nimport {addConfig} from './kernels/Add';\nimport {asinConfig} from './kernels/Asin';\nimport {asinhConfig} from './kernels/Asinh';\nimport {atanConfig} from './kernels/Atan';\nimport {atanhConfig} from './kernels/Atanh';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {castConfig} from './kernels/Cast';\nimport {ceilConfig} from './kernels/Ceil';\nimport {clipConfig} from './kernels/Clip';\nimport {complexConfig} from './kernels/Complex';\nimport {concatConfig} from './kernels/Concat';\nimport {cosConfig} from './kernels/Cos';\nimport {coshConfig} from './kernels/Cosh';\nimport {dilation2dConfig} from './kernels/Dilation2D';\nimport {dilation2dBackpropFilterConfig} from './kernels/Dilation2DBackpropFilter';\nimport {dilation2dBackpropInputConfig} from './kernels/Dilation2DBackpropInput';\nimport {divConfig} from './kernels/Div';\nimport {eluConfig} from './kernels/Elu';\nimport {erfConfig} from './kernels/Erf';\nimport {expConfig} from './kernels/Exp';\nimport {expm1Config} from './kernels/Expm1';\nimport {fftConfig} from './kernels/FFT';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {floorConfig} from './kernels/Floor';\nimport {identityConfig} from './kernels/Identity';\nimport {ifftConfig} from './kernels/IFFT';\nimport {imagConfig} from './kernels/Imag';\nimport {isFiniteConfig} from './kernels/IsFinite';\nimport {isInfConfig} from './kernels/IsInf';\nimport {isNaNConfig} from './kernels/IsNaN';\nimport {logConfig} from './kernels/Log';\nimport {log1pConfig} from './kernels/Log1p';\nimport {logicalNotConfig} from './kernels/LogicalNot';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {multiplyConfig} from './kernels/Multiply';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {notEqualConfig} from './kernels/NotEqual';\nimport {padV2Config} from './kernels/PadV2';\nimport {realConfig} from './kernels/Real';\nimport {reciprocalConfig} from './kernels/Reciprocal';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {roundConfig} from './kernels/Round';\nimport {rsqrtConfig} from './kernels/Rsqrt';\nimport {seluConfig} from './kernels/Selu';\nimport {sigmoidConfig} from './kernels/Sigmoid';\nimport {signConfig} from './kernels/Sign';\nimport {sinConfig} from './kernels/Sin';\nimport {sinhConfig} from './kernels/Sinh';\nimport {sliceConfig} from './kernels/Slice';\nimport {softplusConfig} from './kernels/Softplus';\nimport {spaceToBatchNDConfig} from './kernels/SpaceToBatchND';\nimport {sqrtConfig} from './kernels/Sqrt';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {stepConfig} from './kernels/Step';\nimport {subConfig} from './kernels/Sub';\nimport {tanConfig} from './kernels/Tan';\nimport {tanhConfig} from './kernels/Tanh';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  absConfig,\n  acosConfig,\n  acoshConfig,\n  addConfig,\n  asinConfig,\n  asinhConfig,\n  atanConfig,\n  atanhConfig,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  castConfig,\n  ceilConfig,\n  clipConfig,\n  complexConfig,\n  concatConfig,\n  cosConfig,\n  coshConfig,\n  dilation2dConfig,\n  dilation2dBackpropInputConfig,\n  dilation2dBackpropFilterConfig,\n  divConfig,\n  eluConfig,\n  erfConfig,\n  expConfig,\n  expm1Config,\n  fftConfig,\n  flipLeftRightConfig,\n  floorConfig,\n  identityConfig,\n  ifftConfig,\n  imagConfig,\n  isFiniteConfig,\n  isInfConfig,\n  isNaNConfig,\n  logConfig,\n  log1pConfig,\n  logicalNotConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  maxConfig,\n  multiplyConfig,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  notEqualConfig,\n  padV2Config,\n  realConfig,\n  reciprocalConfig,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  roundConfig,\n  rsqrtConfig,\n  seluConfig,\n  sigmoidConfig,\n  signConfig,\n  sinConfig,\n  sinhConfig,\n  sliceConfig,\n  softplusConfig,\n  spaceToBatchNDConfig,\n  sqrtConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  stepConfig,\n  subConfig,\n  tanConfig,\n  tanhConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst contexts: {[key: string]: WebGLRenderingContext} = {};\n\nconst WEBGL_ATTRIBUTES: WebGLContextAttributes = {\n  alpha: false,\n  antialias: false,\n  premultipliedAlpha: false,\n  preserveDrawingBuffer: false,\n  depth: false,\n  stencil: false,\n  failIfMajorPerformanceCaveat: true\n};\n\nexport function setWebGLContext(\n    webGLVersion: number, gl: WebGLRenderingContext) {\n  contexts[webGLVersion] = gl;\n}\n\nexport function getWebGLContext(webGLVersion: number): WebGLRenderingContext {\n  if (!(webGLVersion in contexts)) {\n    const newCtx = getWebGLRenderingContext(webGLVersion);\n    if (newCtx !== null) {\n      contexts[webGLVersion] = newCtx;\n    } else {\n      console.log('Could not get context for WebGL version', webGLVersion);\n      return null;\n    }\n  }\n  const gl = contexts[webGLVersion];\n  if (gl.isContextLost()) {\n    delete contexts[webGLVersion];\n    return getWebGLContext(webGLVersion);\n  }\n\n  gl.disable(gl.DEPTH_TEST);\n  gl.disable(gl.STENCIL_TEST);\n  gl.disable(gl.BLEND);\n  gl.disable(gl.DITHER);\n  gl.disable(gl.POLYGON_OFFSET_FILL);\n  gl.disable(gl.SAMPLE_COVERAGE);\n  gl.enable(gl.SCISSOR_TEST);\n  gl.enable(gl.CULL_FACE);\n  gl.cullFace(gl.BACK);\n\n  return contexts[webGLVersion];\n}\n\nfunction createCanvas(webGLVersion: number) {\n  if (typeof OffscreenCanvas !== 'undefined' && webGLVersion === 2) {\n    return new OffscreenCanvas(300, 150);\n  } else if (typeof document !== 'undefined') {\n    return document.createElement('canvas');\n  } else {\n    throw new Error('Cannot create a canvas in this context');\n  }\n}\n\nfunction getWebGLRenderingContext(webGLVersion: number): WebGLRenderingContext {\n  if (webGLVersion !== 1 && webGLVersion !== 2) {\n    throw new Error('Cannot get WebGL rendering context, WebGL is disabled.');\n  }\n  const canvas = createCanvas(webGLVersion);\n\n  canvas.addEventListener('webglcontextlost', (ev: Event) => {\n    ev.preventDefault();\n    delete contexts[webGLVersion];\n  }, false);\n  if (webGLVersion === 1) {\n    return (canvas.getContext('webgl', WEBGL_ATTRIBUTES) ||\n            canvas.getContext('experimental-webgl', WEBGL_ATTRIBUTES)) as\n        WebGLRenderingContext;\n  }\n  return canvas.getContext('webgl2', WEBGL_ATTRIBUTES) as WebGLRenderingContext;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {backend_util, DataId, DataType, Tensor, util} from '@tensorflow/tfjs-core';\n\nexport enum PackingScheme {\n  /**\n   * All values in a single texel are densely packed without any constraints.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 4]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   010|011   020|021\n   * -------   -------   -------\n   * 002|003   012|013   022|023\n   *\n   * 100|101   110|111   120|121\n   * -------   -------   -------\n   * 102|103   112|113   122|123\n   *\n   */\n  DENSE,\n\n  /**\n   * Single texels contain only values from the same batch, and from adjacent\n   * rows and columns.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 5]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   002|003   004|xxx   020|021   022|023   024|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   * 100|101   102|103   104|xxx   120|121   122|123   124|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   */\n  SHARED_BATCH\n}\n\nexport enum TextureUsage {\n  RENDER,\n  UPLOAD,\n  PIXELS,\n  DOWNLOAD\n}\n\nexport enum PhysicalTextureType {\n  UNPACKED_FLOAT16,\n  UNPACKED_FLOAT32,\n  PACKED_4X1_UNSIGNED_BYTE,\n  PACKED_2X2_FLOAT32,\n  PACKED_2X2_FLOAT16\n}\n\nexport interface TextureData {\n  // Required.\n  shape: number[];\n  dtype: DataType;\n\n  // Optional.\n  values?: backend_util.BackendValues;\n  texture?: WebGLTexture;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensors field. When this is defined, texture will be null.\n  complexTensors?: {real: Tensor, imag: Tensor};\n  /** [rows, columns] shape of the texture. */\n  texShape?: [number, number];\n  usage?: TextureUsage;\n  isPacked?: boolean;\n\n  refCount: number;\n\n  // Available when the tensor has been sliced.\n  slice?: {\n    // Offset in the 'flat index' space.\n    flatOffset: number;\n    // Used for counting how many sliced tensors point to the same texture.\n    origDataId: DataId;\n  };\n}\n\nexport function getUnpackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns, rows];\n}\n\nexport function getUnpackedArraySizeFromMatrixSize(\n    matrixSize: number, channelsPerTexture: number): number {\n  return matrixSize * channelsPerTexture;\n}\n\nexport function getColorMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns * 4, rows];\n}\n\n/**\n * Get shape for densely packed RGBA texture.\n */\nexport function getDenseTexShape(shape: number[]): [number, number] {\n  const size = util.sizeFromShape(shape);\n  const texelsNeeded = Math.ceil(size / 4);\n  return util.sizeToSquarishShape(texelsNeeded);\n}\n\nexport function getMatrixSizeFromUnpackedArraySize(\n    unpackedSize: number, channelsPerTexture: number): number {\n  if (unpackedSize % channelsPerTexture !== 0) {\n    throw new Error(\n        `unpackedSize (${unpackedSize}) must be a multiple of ` +\n        `${channelsPerTexture}`);\n  }\n  return unpackedSize / channelsPerTexture;\n}\n\nexport function decodeMatrixFromUnpackedColorRGBAArray(\n    unpackedArray: Float32Array, matrix: Float32Array, channels: number) {\n  const requiredSize = unpackedArray.length * channels / 4;\n  if (matrix.length < requiredSize) {\n    throw new Error(\n        `matrix length (${matrix.length}) must be >= ${requiredSize}`);\n  }\n  let dst = 0;\n  for (let src = 0; src < unpackedArray.length; src += 4) {\n    for (let c = 0; c < channels; c++) {\n      matrix[dst++] = unpackedArray[src + c];\n    }\n  }\n}\n\nexport function getPackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [\n    Math.max(1, Math.ceil(columns / 2)), Math.max(1, Math.ceil(rows / 2))\n  ];\n}\n\nexport function getPackedRGBAArraySizeFromMatrixShape(\n    rows: number, columns: number): number {\n  const [w, h] = getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return w * h * 4;\n}\n\nexport interface TextureConfig {\n  internalFormatFloat: number;\n  textureFormatFloat: number;\n  internalFormatPackedHalfFloat: number;\n  internalFormatHalfFloat: number;\n  internalFormatPackedFloat: number;\n\n  // The format to use during a gl.readPixels call.\n  downloadTextureFormat: number;\n  // How many channels need to be unpacked after a gl.readPixels call.\n  downloadUnpackNumChannels: number;\n\n  defaultNumChannels: number;\n  textureTypeHalfFloat: number;\n  textureTypeFloat: number;\n}\n\nexport function getTextureConfig(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension?: any): TextureConfig {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n\n  let internalFormatFloat: number;\n  let internalFormatHalfFloat: number;\n  let internalFormatPackedHalfFloat: number;\n  let internalFormatPackedFloat: number;\n  let textureFormatFloat: number;\n\n  let downloadTextureFormat: number;\n  let downloadUnpackNumChannels: number;\n\n  let defaultNumChannels: number;\n  let textureTypeHalfFloat: number;\n  let textureTypeFloat: number;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    internalFormatFloat = glany.R32F;\n    internalFormatHalfFloat = glany.R16F;\n    internalFormatPackedHalfFloat = glany.RGBA16F;\n    internalFormatPackedFloat = glany.RGBA32F;\n    textureFormatFloat = glany.RED;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 1;\n    textureTypeHalfFloat = glany.HALF_FLOAT;\n    textureTypeFloat = glany.FLOAT;\n  } else {\n    internalFormatFloat = gl.RGBA;\n    internalFormatHalfFloat = gl.RGBA;\n    internalFormatPackedHalfFloat = gl.RGBA;\n    internalFormatPackedFloat = glany.RGBA;\n    textureFormatFloat = gl.RGBA;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 4;\n    textureTypeHalfFloat = textureHalfFloatExtension != null ?\n        textureHalfFloatExtension.HALF_FLOAT_OES :\n        null;\n    textureTypeFloat = gl.FLOAT;\n  }\n  downloadTextureFormat = gl.RGBA;\n\n  return {\n    internalFormatFloat,\n    internalFormatHalfFloat,\n    internalFormatPackedHalfFloat,\n    internalFormatPackedFloat,\n    textureFormatFloat,\n    downloadTextureFormat,\n    downloadUnpackNumChannels,\n    defaultNumChannels,\n    textureTypeHalfFloat,\n    textureTypeFloat\n  };\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext} from './canvas_util';\nimport {getTextureConfig} from './tex_util';\n\nexport function callAndCheck<T>(gl: WebGLRenderingContext, func: () => T): T {\n  const returnValue = func();\n  if (env().getBool('DEBUG')) {\n    checkWebGLError(gl);\n  }\n  return returnValue;\n}\n\nfunction checkWebGLError(gl: WebGLRenderingContext) {\n  const error = gl.getError();\n  if (error !== gl.NO_ERROR) {\n    throw new Error('WebGL Error: ' + getWebGLErrorMessage(gl, error));\n  }\n}\n\n// https://en.wikipedia.org/wiki/Half-precision_floating-point_format\nconst MIN_FLOAT16 = 5.96e-8;\nconst MAX_FLOAT16 = 65504;\n\nexport function canBeRepresented(num: number): boolean {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED') || num === 0 ||\n      (MIN_FLOAT16 < Math.abs(num) && Math.abs(num) < MAX_FLOAT16)) {\n    return true;\n  }\n  return false;\n}\n\nexport function getWebGLErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.NO_ERROR:\n      return 'NO_ERROR';\n    case gl.INVALID_ENUM:\n      return 'INVALID_ENUM';\n    case gl.INVALID_VALUE:\n      return 'INVALID_VALUE';\n    case gl.INVALID_OPERATION:\n      return 'INVALID_OPERATION';\n    case gl.INVALID_FRAMEBUFFER_OPERATION:\n      return 'INVALID_FRAMEBUFFER_OPERATION';\n    case gl.OUT_OF_MEMORY:\n      return 'OUT_OF_MEMORY';\n    case gl.CONTEXT_LOST_WEBGL:\n      return 'CONTEXT_LOST_WEBGL';\n    default:\n      return `Unknown error code ${status}`;\n  }\n}\n\nexport function getExtensionOrThrow(\n    gl: WebGLRenderingContext, extensionName: string): {} {\n  return throwIfNull<{}>(\n      gl, () => gl.getExtension(extensionName),\n      'Extension \"' + extensionName + '\" not supported on this browser.');\n}\n\nexport function createVertexShader(\n    gl: WebGLRenderingContext, vertexShaderSource: string): WebGLShader {\n  const vertexShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.VERTEX_SHADER),\n      'Unable to create vertex WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(vertexShader, vertexShaderSource));\n  callAndCheck(gl, () => gl.compileShader(vertexShader));\n  if (gl.getShaderParameter(vertexShader, gl.COMPILE_STATUS) === false) {\n    console.log(gl.getShaderInfoLog(vertexShader));\n    throw new Error('Failed to compile vertex shader.');\n  }\n  return vertexShader;\n}\n\nexport function createFragmentShader(\n    gl: WebGLRenderingContext, fragmentShaderSource: string): WebGLShader {\n  const fragmentShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.FRAGMENT_SHADER),\n      'Unable to create fragment WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(fragmentShader, fragmentShaderSource));\n  callAndCheck(gl, () => gl.compileShader(fragmentShader));\n  if (gl.getShaderParameter(fragmentShader, gl.COMPILE_STATUS) === false) {\n    logShaderSourceAndInfoLog(\n        fragmentShaderSource, gl.getShaderInfoLog(fragmentShader));\n    throw new Error('Failed to compile fragment shader.');\n  }\n  return fragmentShader;\n}\n\nconst lineNumberRegex = /ERROR: [0-9]+:([0-9]+):/g;\nfunction logShaderSourceAndInfoLog(\n    shaderSource: string, shaderInfoLog: string) {\n  const lineNumberRegexResult = lineNumberRegex.exec(shaderInfoLog);\n  if (lineNumberRegexResult == null) {\n    console.log(`Couldn't parse line number in error: ${shaderInfoLog}`);\n    console.log(shaderSource);\n    return;\n  }\n\n  const lineNumber = +lineNumberRegexResult[1];\n\n  const shaderLines = shaderSource.split('\\n');\n  const pad = shaderLines.length.toString().length + 2;\n  const linesWithLineNumbers = shaderLines.map(\n      (line, lineNumber) =>\n          util.rightPad((lineNumber + 1).toString(), pad) + line);\n  let maxLineLength = 0;\n  for (let i = 0; i < linesWithLineNumbers.length; i++) {\n    maxLineLength = Math.max(linesWithLineNumbers[i].length, maxLineLength);\n  }\n\n  const beforeErrorLines = linesWithLineNumbers.slice(0, lineNumber - 1);\n  const errorLine = linesWithLineNumbers.slice(lineNumber - 1, lineNumber);\n  const afterErrorLines = linesWithLineNumbers.slice(lineNumber);\n\n  console.log(beforeErrorLines.join('\\n'));\n  console.log(shaderInfoLog.split('\\n')[0]);\n  console.log(\n      `%c ${util.rightPad(errorLine[0], maxLineLength)}`,\n      'border:1px solid red; background-color:#e3d2d2; color:#a61717');\n  console.log(afterErrorLines.join('\\n'));\n}\n\nexport function createProgram(gl: WebGLRenderingContext): WebGLProgram {\n  return throwIfNull<WebGLProgram>(\n      gl, () => gl.createProgram(), 'Unable to create WebGLProgram.');\n}\n\nexport function linkProgram(gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.linkProgram(program));\n  if (gl.getProgramParameter(program, gl.LINK_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Failed to link vertex and fragment shaders.');\n  }\n}\n\nexport function validateProgram(\n    gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.validateProgram(program));\n  if (gl.getProgramParameter(program, gl.VALIDATE_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Shader program validation failed.');\n  }\n}\n\nexport function createStaticVertexBuffer(\n    gl: WebGLRenderingContext, data: Float32Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(gl, () => gl.bufferData(gl.ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function createStaticIndexBuffer(\n    gl: WebGLRenderingContext, data: Uint16Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl, () => gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function getNumChannels(): number {\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    return 1;\n  }\n  return 4;\n}\n\nexport function createTexture(gl: WebGLRenderingContext): WebGLTexture {\n  return throwIfNull<WebGLTexture>(\n      gl, () => gl.createTexture(), 'Unable to create WebGLTexture.');\n}\n\nexport function validateTextureSize(width: number, height: number) {\n  const maxTextureSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if ((width <= 0) || (height <= 0)) {\n    const requested = `[${width}x${height}]`;\n    throw new Error('Requested texture size ' + requested + ' is invalid.');\n  }\n  if ((width > maxTextureSize) || (height > maxTextureSize)) {\n    const requested = `[${width}x${height}]`;\n    const max = `[${maxTextureSize}x${maxTextureSize}]`;\n    throw new Error(\n        'Requested texture size ' + requested +\n        ' greater than WebGL maximum on this browser / GPU ' + max + '.');\n  }\n}\n\nexport function createFramebuffer(gl: WebGLRenderingContext): WebGLFramebuffer {\n  return throwIfNull<WebGLFramebuffer>(\n      gl, () => gl.createFramebuffer(), 'Unable to create WebGLFramebuffer.');\n}\n\nexport function bindVertexBufferToProgramAttribute(\n    gl: WebGLRenderingContext, program: WebGLProgram, attribute: string,\n    buffer: WebGLBuffer, arrayEntriesPerItem: number, itemStrideInBytes: number,\n    itemOffsetInBytes: number): boolean {\n  const loc = gl.getAttribLocation(program, attribute);\n  if (loc === -1) {\n    // The GPU compiler decided to strip out this attribute because it's unused,\n    // thus no need to bind.\n    return false;\n  }\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl,\n      () => gl.vertexAttribPointer(\n          loc, arrayEntriesPerItem, gl.FLOAT, false, itemStrideInBytes,\n          itemOffsetInBytes));\n  callAndCheck(gl, () => gl.enableVertexAttribArray(loc));\n  return true;\n}\n\nexport function bindTextureUnit(\n    gl: WebGLRenderingContext, texture: WebGLTexture, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n}\n\nexport function unbindTextureUnit(\n    gl: WebGLRenderingContext, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function getProgramUniformLocationOrThrow(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return throwIfNull<WebGLUniformLocation>(\n      gl, () => gl.getUniformLocation(program, uniformName),\n      'uniform \"' + uniformName + '\" not present in program.');\n}\n\nexport function getProgramUniformLocation(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return gl.getUniformLocation(program, uniformName);\n}\n\nexport function bindTextureToProgramUniformSampler(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    uniformSamplerLocation: WebGLUniformLocation, textureUnit: number) {\n  callAndCheck(gl, () => bindTextureUnit(gl, texture, textureUnit));\n  callAndCheck(gl, () => gl.uniform1i(uniformSamplerLocation, textureUnit));\n}\n\nexport function bindCanvasToFramebuffer(gl: WebGLRenderingContext) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n  callAndCheck(gl, () => gl.viewport(0, 0, gl.canvas.width, gl.canvas.height));\n  callAndCheck(gl, () => gl.scissor(0, 0, gl.canvas.width, gl.canvas.height));\n}\n\nexport function bindColorTextureToFramebuffer(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0));\n}\n\nexport function unbindColorTextureFromFramebuffer(\n    gl: WebGLRenderingContext, framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, null, 0));\n}\n\nexport function validateFramebuffer(gl: WebGLRenderingContext) {\n  const status = gl.checkFramebufferStatus(gl.FRAMEBUFFER);\n  if (status !== gl.FRAMEBUFFER_COMPLETE) {\n    throw new Error(\n        'Error binding framebuffer: ' + getFramebufferErrorMessage(gl, status));\n  }\n}\n\nexport function getFramebufferErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:\n      return 'FRAMEBUFFER_INCOMPLETE_DIMENSIONS';\n    case gl.FRAMEBUFFER_UNSUPPORTED:\n      return 'FRAMEBUFFER_UNSUPPORTED';\n    default:\n      return `unknown error ${status}`;\n  }\n}\n\nfunction throwIfNull<T>(\n    gl: WebGLRenderingContext, returnTOrNull: () => T | null,\n    failureMessage: string): T {\n  const tOrNull: T|null = callAndCheck(gl, () => returnTOrNull());\n  if (tOrNull == null) {\n    throw new Error(failureMessage);\n  }\n  return tOrNull;\n}\n\nfunction validateTextureUnit(gl: WebGLRenderingContext, textureUnit: number) {\n  const maxTextureUnit = gl.MAX_COMBINED_TEXTURE_IMAGE_UNITS - 1;\n  const glTextureUnit = textureUnit + gl.TEXTURE0;\n  if (glTextureUnit < gl.TEXTURE0 || glTextureUnit > maxTextureUnit) {\n    const textureUnitRange = `[gl.TEXTURE0, gl.TEXTURE${maxTextureUnit}]`;\n    throw new Error(`textureUnit must be in ${textureUnitRange}.`);\n  }\n}\n\nexport function getBatchDim(shape: number[], dimsToSkip = 2): number {\n  return util.sizeFromShape(shape.slice(0, shape.length - dimsToSkip));\n}\n\nexport function getRowsCols(shape: number[]): [number, number] {\n  if (shape.length === 0) {\n    throw Error('Cannot get rows and columns of an empty shape array.');\n  }\n\n  return [\n    shape.length > 1 ? shape[shape.length - 2] : 1, shape[shape.length - 1]\n  ];\n}\n\nexport function getShapeAs3D(shape: number[]): [number, number, number] {\n  let shapeAs3D: [number, number, number] = [1, 1, 1];\n  const isScalar = shape.length === 0 || (shape.length === 1 && shape[0] === 1);\n  if (!isScalar) {\n    shapeAs3D =\n        [getBatchDim(shape), ...getRowsCols(shape)] as [number, number, number];\n  }\n  return shapeAs3D;\n}\n\nexport function getTextureShapeFromLogicalShape(\n    logShape: number[], isPacked = false): [number, number] {\n  let maxTexSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if (isPacked) {\n    maxTexSize = maxTexSize * 2;\n\n    // This logic ensures we accurately count the number of packed texels needed\n    // to accommodate the tensor. We can only pack values in the same texel if\n    // they are from adjacent pairs of rows/cols within the same batch. So if a\n    // tensor has 3 rows, we pretend it has 4 rows in order to account for the\n    // fact that the texels containing the third row are half empty.\n    logShape = logShape.map(\n        (d, i) => i >= logShape.length - 2 ?\n            util.nearestLargerEven(logShape[i]) :\n            logShape[i]);\n\n    // Packed texture height is at least 2 (the channel height of a single\n    // texel).\n    if (logShape.length === 1) {\n      logShape = [2, logShape[0]];\n    }\n  }\n\n  // If logical shape is 2, we don't squeeze, since we want to match physical.\n  if (logShape.length !== 2) {\n    const squeezeResult = util.squeezeShape(logShape);\n    logShape = squeezeResult.newShape;\n  }\n\n  let size = util.sizeFromShape(logShape);\n  if (logShape.length <= 1 && size <= maxTexSize) {\n    return [1, size];\n  } else if (\n      logShape.length === 2 && logShape[0] <= maxTexSize &&\n      logShape[1] <= maxTexSize) {\n    return logShape as [number, number];\n  } else if (\n      logShape.length === 3 && logShape[0] * logShape[1] <= maxTexSize &&\n      logShape[2] <= maxTexSize) {\n    return [logShape[0] * logShape[1], logShape[2]];\n  } else if (\n      logShape.length === 3 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2]];\n  } else if (\n      logShape.length === 4 &&\n      logShape[0] * logShape[1] * logShape[2] <= maxTexSize &&\n      logShape[3] <= maxTexSize) {\n    return [logShape[0] * logShape[1] * logShape[2], logShape[3]];\n  } else if (\n      logShape.length === 4 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] * logShape[3] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2] * logShape[3]];\n  } else {\n    if (isPacked) {\n      // For packed textures size equals the number of channels required to\n      // accommodate the texture data. However in order to squarify such that\n      // inner dimensions stay even, we rewrite size to equal the number of\n      // texels. Then in the return statement we rehydrate the squarified\n      // dimensions to channel units.\n\n      const batchDim = getBatchDim(logShape);\n      let rows = 2, cols = 2;\n      if (logShape.length) {\n        [rows, cols] = getRowsCols(logShape);\n      }\n      size = batchDim * (rows / 2) * (cols / 2);\n      return util.sizeToSquarishShape(size).map(d => d * 2) as [number, number];\n    }\n    return util.sizeToSquarishShape(size);\n  }\n}\n\nfunction isEven(n: number): boolean {\n  return n % 2 === 0;\n}\n\n/**\n * This determines whether reshaping a packed texture requires rearranging\n * the data within the texture, assuming 2x2 packing.\n */\nexport function isReshapeFree(shape1: number[], shape2: number[]): boolean {\n  shape1 = shape1.slice(-2);\n  shape2 = shape2.slice(-2);\n\n  if (util.arraysEqual(shape1, shape2)) {\n    return true;\n  }\n\n  if (!shape1.length || !shape2.length) {  // One of the shapes is a scalar.\n    return true;\n  }\n\n  if (shape1[0] === 0 || shape1[1] === 0 || shape2[0] === 0 ||\n      shape2[1] === 0) {\n    return true;\n  }\n\n  if (shape1.length !== shape2.length) {  // One of the shapes is a vector.\n    const shape1Cols = shape1.slice(-1)[0];\n    const shape2Cols = shape2.slice(-1)[0];\n    if (shape1Cols === shape2Cols) {\n      return true;\n    }\n\n    if (isEven(shape1Cols) && isEven(shape2Cols) &&\n        (shape1[0] === 1 || shape2[0] === 1)) {\n      return true;\n    }\n  }\n  return shape1[1] === shape2[1] && isEven(shape1[0]) && isEven(shape2[0]);\n}\n\n// We cache webgl params because the environment gets reset between\n// unit tests and we don't want to constantly query the WebGLContext for\n// MAX_TEXTURE_SIZE.\nlet MAX_TEXTURE_SIZE: number;\nlet MAX_TEXTURES_IN_SHADER: number;\n\nexport function getWebGLMaxTextureSize(webGLVersion: number): number {\n  if (MAX_TEXTURE_SIZE == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURE_SIZE = gl.getParameter(gl.MAX_TEXTURE_SIZE);\n  }\n  return MAX_TEXTURE_SIZE;\n}\n\nexport function resetMaxTextureSize() {\n  MAX_TEXTURE_SIZE = null;\n}\nexport function resetMaxTexturesInShader() {\n  MAX_TEXTURES_IN_SHADER = null;\n}\n\nexport function getMaxTexturesInShader(webGLVersion: number): number {\n  if (MAX_TEXTURES_IN_SHADER == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURES_IN_SHADER = gl.getParameter(gl.MAX_TEXTURE_IMAGE_UNITS);\n  }\n  // We cap at 16 to avoid spurious runtime \"memory exhausted\" error.\n  return Math.min(16, MAX_TEXTURES_IN_SHADER);\n}\n\nexport function getWebGLDisjointQueryTimerVersion(webGLVersion: number):\n    number {\n  if (webGLVersion === 0) {\n    return 0;\n  }\n\n  let queryTimerVersion: number;\n  const gl = getWebGLContext(webGLVersion);\n\n  if (hasExtension(gl, 'EXT_disjoint_timer_query_webgl2') &&\n      webGLVersion === 2) {\n    queryTimerVersion = 2;\n  } else if (hasExtension(gl, 'EXT_disjoint_timer_query')) {\n    queryTimerVersion = 1;\n  } else {\n    queryTimerVersion = 0;\n  }\n  return queryTimerVersion;\n}\n\nexport function hasExtension(gl: WebGLRenderingContext, extensionName: string) {\n  const ext = gl.getExtension(extensionName);\n  return ext != null;\n}\n\nexport function isWebGLVersionEnabled(webGLVersion: 1|2) {\n  try {\n    const gl = getWebGLContext(webGLVersion);\n    if (gl != null) {\n      return true;\n    }\n  } catch (e) {\n    console.log('Error when getting WebGL context: ', e);\n    return false;\n  }\n  return false;\n}\n\nexport function isCapableOfRenderingToFloatTexture(webGLVersion: number):\n    boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n  } else {\n    if (!hasExtension(gl, 'EXT_color_buffer_float')) {\n      return false;\n    }\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\n/**\n * Check if we can download values from a float/half-float texture.\n *\n * Note that for performance reasons we use binding a texture to a framebuffer\n * as a proxy for ability to download float values later using readPixels. The\n * texture params of this texture will not match those in readPixels exactly\n * but if we are unable to bind some kind of float texture to the frameBuffer\n * then we definitely will not be able to read float values from it.\n */\nexport function isDownloadFloatTextureEnabled(webGLVersion: number): boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n    if (!hasExtension(gl, 'WEBGL_color_buffer_float')) {\n      return false;\n    }\n  } else {\n    if (hasExtension(gl, 'EXT_color_buffer_float')) {\n      return createFloatTextureAndBindToFramebuffer(gl);\n    }\n\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (hasExtension(gl, COLOR_BUFFER_HALF_FLOAT)) {\n      const textureHalfFloatExtension =\n          gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      return createHalfFloatTextureAndBindToFramebuffer(\n          gl, textureHalfFloatExtension);\n    }\n\n    return false;\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\nfunction createFloatTextureAndBindToFramebuffer(gl: WebGLRenderingContext):\n    boolean {\n  const texConfig = getTextureConfig(gl);\n\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nfunction createHalfFloatTextureAndBindToFramebuffer(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension: any): boolean {\n  const texConfig = getTextureConfig(gl, textureHalfFloatExtension);\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatHalfFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeHalfFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nexport function isWebGLFenceEnabled(webGLVersion: number) {\n  if (webGLVersion !== 2) {\n    return false;\n  }\n  const gl = getWebGLContext(webGLVersion);\n\n  // tslint:disable-next-line:no-any\n  const isEnabled = (gl as any).fenceSync != null;\n  return isEnabled;\n}\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${opName} does not support complex64 tensors ` +\n              'in the WebGL backend.');\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {device_util, env} from '@tensorflow/tfjs-core';\n\nimport {getMaxTexturesInShader, getWebGLDisjointQueryTimerVersion, getWebGLMaxTextureSize, isCapableOfRenderingToFloatTexture, isDownloadFloatTextureEnabled, isWebGLFenceEnabled, isWebGLVersionEnabled} from './webgl_util';\n\nconst ENV = env();\n\n/**\n * This file contains WebGL-specific flag registrations.\n */\n\n/**\n * True if WebGL is supported.\n */\nENV.registerFlag('HAS_WEBGL', () => ENV.getNumber('WEBGL_VERSION') > 0);\n\n/** 0: No WebGL, 1: WebGL 1.0, 2: WebGL 2.0. */\nENV.registerFlag('WEBGL_VERSION', () => {\n  if (isWebGLVersionEnabled(2)) {\n    return 2;\n  } else if (isWebGLVersionEnabled(1)) {\n    return 1;\n  }\n  return 0;\n});\n\n/** Whether to check for numerical representation problems. */\nENV.registerFlag('WEBGL_CHECK_NUMERICAL_PROBLEMS', () => false);\n\nENV.registerFlag(\n    'WEBGL_BUFFER_SUPPORTED', () => ENV.get('WEBGL_VERSION') === 2);\n\n/** Whether the WebGL backend will sometimes forward ops to the CPU. */\nENV.registerFlag('WEBGL_CPU_FORWARD', () => true);\n\n/** Whether the WebGL backend will always use f16 textures for rendering. */\nENV.registerFlag('WEBGL_FORCE_F16_TEXTURES', () => false);\n\n/** Whether to turn all packing related flags on. */\nENV.registerFlag('WEBGL_PACK', () => ENV.getBool('HAS_WEBGL'));\n\n/** Whether we will pack the batchnormalization op. */\nENV.registerFlag('WEBGL_PACK_NORMALIZATION', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the clip op. */\nENV.registerFlag('WEBGL_PACK_CLIP', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the depthwise conv op. */\n// TODO: https://github.com/tensorflow/tfjs/issues/1679\nENV.registerFlag('WEBGL_PACK_DEPTHWISECONV', () => false);\n\n/** Whether we will pack binary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_BINARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack unary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_UNARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack array ops. */\nENV.registerFlag(\n    'WEBGL_PACK_ARRAY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack image ops. */\nENV.registerFlag(\n    'WEBGL_PACK_IMAGE_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack reduce ops. */\nENV.registerFlag('WEBGL_PACK_REDUCE', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether packed WebGL kernels lazily unpack their outputs. */\nENV.registerFlag('WEBGL_LAZILY_UNPACK', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will use the im2col algorithm to speed up convolutions. */\nENV.registerFlag('WEBGL_CONV_IM2COL', () => ENV.getBool('WEBGL_PACK'));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURE_SIZE',\n    () => getWebGLMaxTextureSize(ENV.getNumber('WEBGL_VERSION')));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURES_IN_SHADER',\n    () => getMaxTexturesInShader(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * The disjoint_query_timer extension version.\n * 0: disabled, 1: EXT_disjoint_timer_query, 2:\n * EXT_disjoint_timer_query_webgl2.\n * In Firefox with WebGL 2.0,\n * EXT_disjoint_timer_query_webgl2 is not available, so we must use the\n * WebGL 1.0 extension.\n */\nENV.registerFlag('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION', () => {\n  const webGLVersion = ENV.getNumber('WEBGL_VERSION');\n\n  if (webGLVersion === 0) {\n    return 0;\n  }\n  return getWebGLDisjointQueryTimerVersion(webGLVersion);\n});\n\n/**\n * Whether the timer object from the disjoint_query_timer extension gives\n * timing information that is reliable.\n */\nENV.registerFlag(\n    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE',\n    () => ENV.getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0 &&\n        !device_util.isMobile());\n\n/**\n * Whether the device is physically capable of rendering to float32 textures.\n */\nENV.registerFlag(\n    'WEBGL_RENDER_FLOAT32_CAPABLE',\n    () => isCapableOfRenderingToFloatTexture(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Whether rendering to float32 textures is enabled. If disabled, renders to\n * float16 textures.\n */\nENV.registerFlag('WEBGL_RENDER_FLOAT32_ENABLED', () => {\n  return ENV.getBool('WEBGL_FORCE_F16_TEXTURES') ?\n      false :\n      ENV.getBool('WEBGL_RENDER_FLOAT32_CAPABLE');\n});\n\n/**\n * Whether downloading float textures is enabled (16 or 32 bit). If disabled,\n * uses IEEE 754 encoding of the float32 values to 4 uint8 when downloading.\n */\nENV.registerFlag(\n    'WEBGL_DOWNLOAD_FLOAT_ENABLED',\n    () => isDownloadFloatTextureEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/** Whether the fence API is available. */\nENV.registerFlag(\n    'WEBGL_FENCE_API_ENABLED',\n    () => isWebGLFenceEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Tensors with size <= than this will be uploaded as uniforms, not textures.\n */\nENV.registerFlag('WEBGL_SIZE_UPLOAD_UNIFORM', () => {\n  // Use uniform uploads only when 32bit floats are supported. In\n  // 16bit\n  // environments there are problems with comparing a 16bit texture value\n  // with a 32bit uniform value.\n  const useUniforms = ENV.getBool('WEBGL_RENDER_FLOAT32_ENABLED');\n  return useUniforms ? 4 : 0;\n});\n\n/**\n * If the total number of bytes allocated on the GPU is greater than this\n * number, we will aggressively delete textures upon disposal with\n * gl.deleteMatrixTexture, rather than making them available for reuse.\n *\n * Default value -1 indicates that we will never aggressively delete textures.\n */\nENV.registerFlag(\n    'WEBGL_DELETE_TEXTURE_THRESHOLD',\n    () => {\n      return -1;\n    },\n    threshold => {\n      if (threshold < 0 && threshold !== -1) {\n        throw new Error(\n            `WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never ` +\n            `delete) or at least 0, but got ${threshold}.`);\n      }\n    });\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Abs, util } from '@tensorflow/tfjs-core';\nexport function simpleAbsImpl(vals) {\n    const resultValues = new Float32Array(vals.length);\n    for (let i = 0; i < vals.length; ++i) {\n        resultValues[i] = Math.abs(vals[i]);\n    }\n    return resultValues;\n}\nexport const absKernelFunc = (args) => {\n    const { x } = args.inputs;\n    const cpuBackend = args.backend;\n    let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n    if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values;\n        resultValues = simpleAbsImpl(values);\n    }\n    else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals = cpuBackend.data.get(real.dataId).values;\n        const imagVals = cpuBackend.data.get(imag.dataId).values;\n        for (let i = 0; i < realVals.length; i++) {\n            const real = realVals[i];\n            const imag = imagVals[i];\n            resultValues[i] = Math.hypot(real, imag);\n        }\n    }\n    return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n};\nexport const absConfig = {\n    kernelName: Abs,\n    backendName: 'cpu',\n    kernelFunc: absKernelFunc,\n};\n//# sourceMappingURL=Abs.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { backend_util, util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op) {\n    return (aShape, bShape, aVals, bVals, dtype) => {\n        const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n        const resultRank = newShape.length;\n        const resultStrides = util.computeStrides(newShape);\n        const resultSize = util.sizeFromShape(newShape);\n        const result = util.getTypedArrayFromDType(dtype, resultSize);\n        const aRank = aShape.length;\n        const bRank = bShape.length;\n        const aStrides = util.computeStrides(aShape);\n        const bStrides = util.computeStrides(bShape);\n        const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n        const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n        if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n            for (let i = 0; i < result.length; ++i) {\n                result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n            }\n        }\n        else {\n            for (let i = 0; i < result.length; ++i) {\n                const loc = util.indexToLoc(i, resultRank, resultStrides);\n                const aLoc = loc.slice(-aRank);\n                aBroadcastDims.forEach(d => aLoc[d] = 0);\n                const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n                const bLoc = loc.slice(-bRank);\n                bBroadcastDims.forEach(d => bLoc[d] = 0);\n                const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n                result[i] = op(aVals[aIndex], bVals[bIndex]);\n            }\n        }\n        return [result, newShape];\n    };\n}\n//# sourceMappingURL=binary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Add } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal + bReal, imag: aImag + bImag };\n}));\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\nexport const addConfig = {\n    kernelName: Add,\n    backendName: 'cpu',\n    kernelFunc: add\n};\n//# sourceMappingURL=Add.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op) {\n    return (values, dtype, attrs) => {\n        const newValues = util.getTypedArrayFromDType(dtype, values.length);\n        for (let i = 0; i < values.length; ++i) {\n            newValues[i] = op(values[i], attrs);\n        }\n        return newValues;\n    };\n}\n//# sourceMappingURL=unary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Ceil } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\nexport const ceilConfig = {\n    kernelName: Ceil,\n    backendName: 'cpu',\n    kernelFunc: ceilKernelFunc,\n};\n//# sourceMappingURL=Ceil.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Exp } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\nexport const expConfig = {\n    kernelName: Exp,\n    backendName: 'cpu',\n    kernelFunc: expKernelFunc,\n};\n//# sourceMappingURL=Exp.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Expm1 } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\nexport const expm1Config = {\n    kernelName: Expm1,\n    backendName: 'cpu',\n    kernelFunc: expm1KernelFunc,\n};\n//# sourceMappingURL=Expm1.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Floor } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\nexport const floorConfig = {\n    kernelName: Floor,\n    backendName: 'cpu',\n    kernelFunc: floorKernelFunc,\n};\n//# sourceMappingURL=Floor.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Log } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\nexport const logConfig = {\n    kernelName: Log,\n    backendName: 'cpu',\n    kernelFunc: logKernelFunc,\n};\n//# sourceMappingURL=Log.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function maxImpl(aVals, reduceSize, outShape, dtype) {\n    const vals = util.getTypedArrayFromDType(dtype, util.sizeFromShape(outShape));\n    for (let i = 0; i < vals.length; ++i) {\n        const offset = i * reduceSize;\n        let max = aVals[offset];\n        for (let j = 0; j < reduceSize; ++j) {\n            const value = aVals[offset + j];\n            if (value > max) {\n                max = value;\n            }\n        }\n        vals[i] = max;\n    }\n    return vals;\n}\n//# sourceMappingURL=Max_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Multiply } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const multiplyImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n    };\n}));\nexport const multiply = binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\nexport const multiplyConfig = {\n    kernelName: Multiply,\n    backendName: 'cpu',\n    kernelFunc: multiply\n};\n//# sourceMappingURL=Multiply.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Rsqrt } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\nexport const rsqrtConfig = {\n    kernelName: Rsqrt,\n    backendName: 'cpu',\n    kernelFunc: rsqrtKernelFunc,\n};\n//# sourceMappingURL=Rsqrt.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Slice, slice_util, util } from '@tensorflow/tfjs-core';\nimport { assertNotComplex } from '../cpu_util';\nexport function sliceImpl(vals, begin, size, shape, dtype) {\n    const isContinous = slice_util.isSliceContinous(shape, begin, size);\n    const length = util.sizeFromShape(size);\n    const xStrides = util.computeStrides(shape);\n    if (isContinous) {\n        const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n        return vals.subarray(flatOffset, flatOffset + length);\n    }\n    const outVals = util.getTypedArrayFromDType(dtype, length);\n    for (let i = 0; i < length; ++i) {\n        const rank = size.length;\n        const strides = util.computeStrides(size);\n        const loc = util.indexToLoc(i, rank, strides);\n        const xLoc = loc.map((idx, j) => idx + begin[j]);\n        const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n        outVals[i] = vals[xIndex];\n    }\n    return outVals;\n}\nexport function slice(args) {\n    const { inputs, backend, attrs } = args;\n    const { x } = inputs;\n    const { begin, size } = attrs;\n    assertNotComplex(x, 'slice');\n    const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n    slice_util.assertParamsValid(x, $begin, $size);\n    const vals = backend.data.get(x.dataId).values;\n    const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n    return backend.makeTensorInfo($size, x.dtype, outVals);\n}\nexport const sliceConfig = {\n    kernelName: Slice,\n    backendName: 'cpu',\n    kernelFunc: slice\n};\n//# sourceMappingURL=Slice.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Sub } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const subImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal - bReal, imag: aImag - bImag };\n}));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\nexport const subConfig = {\n    kernelName: Sub,\n    backendName: 'cpu',\n    kernelFunc: sub\n};\n//# sourceMappingURL=Sub.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function transposeImpl(xVals, xShape, dtype, perm, newShape) {\n    const xRank = xShape.length;\n    const xSize = util.sizeFromShape(xShape);\n    const xStrides = util.computeStrides(xShape);\n    const newStrides = util.computeStrides(newShape);\n    const result = util.getTypedArrayFromDType(dtype, util.sizeFromShape(newShape));\n    for (let i = 0; i < xSize; ++i) {\n        const loc = util.indexToLoc(i, xRank, xStrides);\n        // Permute location.\n        const newLoc = new Array(loc.length);\n        for (let i = 0; i < newLoc.length; i++) {\n            newLoc[i] = loc[perm[i]];\n        }\n        const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n        result[newIndex] = xVals[i];\n    }\n    return result;\n}\n//# sourceMappingURL=Transpose_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { TensorBuffer, util } from '@tensorflow/tfjs-core';\nexport function uniqueImpl(values, axis, shape, dtype) {\n    // Normalize and validate axis.\n    const $axis = util.parseAxisParam(axis, shape)[0];\n    // Calculate the new shape that is suitable for extracting data along the\n    // given axis.\n    //\n    // The rank is 3.\n    // The size of the 1st dimension is the size of all the axes < the given axis.\n    // The size of the 2nd dimension is the same as the size of the given axis.\n    // The size of the 3rd dimension is the size of all the axes > the given axis.\n    //\n    // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n    // newShape would be: [2*3, 5, 4].\n    //\n    // Note that this is not the final output shape. This will be the shape for an\n    // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n    // values along the given axis. To demonstrate how it works, consider the\n    // following example:\n    //\n    // Input: a 3D tensor, with shape [1, 2, 3]\n    // [\n    //   [\n    //      [1,2,3],\n    //      [4,5,6]\n    //   ]\n    // ]\n    // Axis: 2 (the last axis).\n    // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n    //\n    // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n    // 1*2. The re-shaped data would look like:\n    //\n    // [\n    //   [\n    //     [1], [2], [3]\n    //   ],\n    //   [\n    //     [4], [5], [6]\n    //   ]\n    // ]\n    //\n    // Then, we can construct a 3-level nested loop by the following dimension\n    // order to extract the values along the axis (dimension1):\n    // i: dimension1       // 0,1,2 (newShape[1])\n    //   m: dimension0     // 0,1   (newShape[0])\n    //     n: dimension2   // 0     (newShape[2])\n    //\n    //                       m, i, n\n    //                      ---------\n    // Iteration 0: data at [0, 0, 0] => \"1\"\n    // Iteration 1: data at [1, 0, 0] => \"4\"\n    // We got [1,4].\n    // Iteration 2: data at [0, 1, 0] => \"2\"\n    // Iteration 3: data at [1, 1, 0] => \"5\"\n    // We got [2,5].\n    // Iteration 4: data at [0, 2, 0] => \"3\"\n    // Iteration 5: data at [1, 2, 0] => \"6\"\n    // We got [3,6].\n    const newShape = [1, shape[0], 1];\n    for (let i = 0; i < $axis; i++) {\n        newShape[0] *= shape[i];\n    }\n    newShape[1] = shape[$axis];\n    for (let i = $axis + 1; i < shape.length; i++) {\n        newShape[2] *= shape[i];\n    }\n    // A map from unique elements (their string representations) to their values\n    // in \"indices\" (below).\n    const uniqueElements = {};\n    // The indices of each unique element in the original tensor along the given\n    // axis. It is 1D and has the same size as the given axis.\n    const indices = new Int32Array(shape[$axis]);\n    // Create a buffer so we can easily extract value at a given location.\n    const inputBuffer = new TensorBuffer(newShape, dtype, values);\n    // The indices along the given axis that have unique elements. This is a\n    // de-duped version of \"indices\" above.\n    const uniqueIndices = [];\n    const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n    for (let i = 0; i < shape[$axis]; i++) {\n        // Extract values along the axis.\n        let element;\n        if (is1DTensor) {\n            // Fast path for 1D tensor input.\n            element = values[i].toString();\n        }\n        else {\n            const axisValues = [];\n            for (let m = 0; m < newShape[0]; m++) {\n                for (let n = 0; n < newShape[2]; n++) {\n                    axisValues.push(inputBuffer.get(m, i, n));\n                }\n            }\n            element = axisValues.join(',');\n        }\n        // Dedup and update various indices.\n        if (uniqueElements[element] !== undefined) {\n            indices[i] = uniqueElements[element];\n        }\n        else {\n            const uniqueIndex = Object.keys(uniqueElements).length;\n            uniqueElements[element] = uniqueIndex;\n            indices[i] = uniqueIndex;\n            uniqueIndices.push(i);\n        }\n    }\n    // Now we know where each of the unique elements are located along the axis\n    // (uniqueIndices). Extract them from input buffer and store them in the\n    // output buffer.\n    const outputTmpShape = newShape.slice();\n    outputTmpShape[1] = Object.keys(uniqueElements).length;\n    const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n    uniqueIndices.forEach((uniqueElementIndex, i) => {\n        for (let m = 0; m < newShape[0]; m++) {\n            for (let n = 0; n < newShape[2]; n++) {\n                outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n            }\n        }\n    });\n    // The output shape can be calculated from the input shape with the size of\n    // the given axis replaced by the number of unique elements along that axis.\n    const outputShape = shape.slice();\n    outputShape[$axis] = outputTmpShape[1];\n    return {\n        outputValues: outputBuffer.values,\n        outputShape,\n        indices,\n    };\n}\n//# sourceMappingURL=Unique_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import shared functionality from tfjs-backend-cpu without triggering\n// side effects.\n// tslint:disable-next-line: no-imports-from-dist\nimport * as shared from '@tensorflow/tfjs-backend-cpu/dist/shared';\n\nconst {\n  simpleAbsImpl: simpleAbsImplCPU,\n  addImpl: addImplCPU,\n  ceilImpl: ceilImplCPU,\n  expImpl: expImplCPU,\n  expm1Impl: expm1ImplCPU,\n  floorImpl: floorImplCPU,\n  logImpl: logImplCPU,\n  maxImpl: maxImplCPU,\n  multiplyImpl: multiplyImplCPU,\n  rsqrtImpl: rsqrtImplCPU,\n  sliceImpl: sliceImplCPU,\n  subImpl: subImplCPU,\n  transposeImpl: transposeImplCPU,\n  uniqueImpl: uniqueImplCPU,\n} = shared;\n\nexport {\n  simpleAbsImplCPU,\n  addImplCPU,\n  ceilImplCPU,\n  expImplCPU,\n  expm1ImplCPU,\n  logImplCPU,\n  multiplyImplCPU,\n  sliceImplCPU,\n  subImplCPU,\n  floorImplCPU,\n  maxImplCPU,\n  rsqrtImplCPU,\n  transposeImplCPU,\n  uniqueImplCPU,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`float v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        float result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`vec4 v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        vec4 result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ArgMinMaxProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo, op: 'max'|'min',\n      firstPass: boolean) {\n    const {windowSize, batchSize, outSize} = reduceInfo;\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    this.outputShape = [batchSize, outSize];\n    const compOp = (op === 'max') ? '>' : '<';\n    const indexSnippet = firstPass ?\n        'inOffset + i;' :\n        'round(getBestIndicesA(batch, inOffset + i));';\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        int bestIndex = inOffset;\n        float bestValue = getA(batch, bestIndex);\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          int inIdx = ${indexSnippet};\n          float candidate = getA(batch, inIdx);\n          if (candidate ${compOp} bestValue) {\n            bestValue = candidate;\n            bestIndex = inIdx;\n          }\n        }\n        setOutput(float(bestIndex));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport function getVecChannels(name: string, rank: number): string[] {\n  return ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank).map(d => `${name}.${d}`);\n}\n\nexport function getChannels(name: string, rank: number): string[] {\n  if (rank === 1) {\n    return [name];\n  }\n  return getVecChannels(name, rank);\n}\n\nexport function getSourceCoords(rank: number, dims: string[]): string {\n  if (rank === 1) {\n    return 'rc';\n  }\n\n  let coords = '';\n  for (let i = 0; i < rank; i++) {\n    coords += dims[i];\n    if (i < rank - 1) {\n      coords += ',';\n    }\n  }\n  return coords;\n}", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '@tensorflow/tfjs-core';\n\nexport type GLSL = {\n  version: string,\n  attribute: string,\n  varyingVs: string,\n  varyingFs: string,\n  texture2D: string,\n  output: string,\n  defineOutput: string,\n  defineSpecialNaN: string,\n  defineSpecialInf: string,\n  defineRound: string\n};\n\nexport function getGlslDifferences(): GLSL {\n  let version: string;\n  let attribute: string;\n  let varyingVs: string;\n  let varyingFs: string;\n  let texture2D: string;\n  let output: string;\n  let defineOutput: string;\n  let defineSpecialNaN: string;\n  let defineSpecialInf: string;\n  let defineRound: string;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    version = '#version 300 es';\n    attribute = 'in';\n    varyingVs = 'out';\n    varyingFs = 'in';\n    texture2D = 'texture';\n    output = 'outputColor';\n    defineOutput = 'out vec4 outputColor;';\n\n    // Use custom isnan definition to work across differences between\n    // implementations on various platforms. While this should happen in ANGLE\n    // we still see differences between android and windows (on chrome) when\n    // using isnan directly.\n    defineSpecialNaN = `\n      bool isnan_custom(float val) {\n        return (val > 0.0 || val < 0.0) ? false : val != 0.0;\n      }\n\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan_custom(val.x),\n          isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));\n      }\n\n      #define isnan(value) isnan_custom(value)\n    `;\n    // In webgl 2 we do not need to specify a custom isinf so there is no\n    // need for a special INFINITY constant.\n    defineSpecialInf = ``;\n    defineRound = `\n      #define round(value) newRound(value)\n      int newRound(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 newRound(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  } else {\n    version = '';\n    attribute = 'attribute';\n    varyingVs = 'varying';\n    varyingFs = 'varying';\n    texture2D = 'texture2D';\n    output = 'gl_FragColor';\n    defineOutput = '';\n    // WebGL1 has no built in isnan so we define one here.\n    defineSpecialNaN = `\n      #define isnan(value) isnan_custom(value)\n      bool isnan_custom(float val) {\n        return (val > 0. || val < 1. || val == 0.) ? false : true;\n      }\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));\n      }\n    `;\n    defineSpecialInf = `\n      uniform float INFINITY;\n\n      bool isinf(float val) {\n        return abs(val) == INFINITY;\n      }\n      bvec4 isinf(vec4 val) {\n        return equal(abs(val), vec4(INFINITY));\n      }\n    `;\n    defineRound = `\n      int round(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 round(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  }\n\n  return {\n    version,\n    attribute,\n    varyingVs,\n    varyingFs,\n    texture2D,\n    output,\n    defineOutput,\n    defineSpecialNaN,\n    defineSpecialInf,\n    defineRound\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\n/**\n * Produces GLSL code that derives logical coordinates from a flat\n * index. The code performs integer division with each stride and decrements\n * the index until the index equals the final dimension coordinate.\n */\nexport function getLogicalCoordinatesFromFlatIndex(\n    coords: string[], shape: number[], index = 'index'): string {\n  const strides = util.computeStrides(shape);\n  return strides\n      .map((stride, i) => {\n        const line1 = `int ${coords[i]} = ${index} / ${stride}`;\n        const line2 = i === strides.length - 1 ?\n            `int ${coords[i + 1]} = ${index} - ${coords[i]} * ${stride}` :\n            `index -= ${coords[i]} * ${stride}`;\n        return `${line1}; ${line2};`;\n      })\n      .join('');\n}\n\nfunction buildVec(x: string[]): string {\n  if (x.length === 1) {\n    return `${x[0]}`;\n  }\n  return `vec${x.length}(${x.join(',')})`;\n}\n\n/**\n * Produces GLSL code that computes the dot product of the input x and y\n * vectors. Handles splitting inputs into increments of vec4s when necessary.\n */\nexport function dotify(x: string[], y: string[]): string {\n  if (x.length !== y.length) {\n    throw new Error(\n        `Vectors to be dotted must be of the same length -` +\n        `got ${x.length} and ${y.length}`);\n  }\n\n  const slices: string[] = [];\n  const nearestVec4 = Math.floor(x.length / 4);\n  const nearestVec4Remainder = x.length % 4;\n\n  for (let i = 0; i < nearestVec4; i++) {\n    const xSlice = x.slice(i * 4, i * 4 + 4);\n    const ySlice = y.slice(i * 4, i * 4 + 4);\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  if (nearestVec4Remainder !== 0) {\n    let xSlice = x.slice(nearestVec4 * 4);\n    let ySlice = y.slice(nearestVec4 * 4);\n    if (xSlice.length === 1) {\n      xSlice = xSlice.map(d => `float(${d})`);\n      ySlice = ySlice.map(d => `float(${d})`);\n    }\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  return slices.map((d, i) => `dot(${d})`).join('+');\n}\n\n/**\n * Produces GLSL that computes the flat index from 3D coordinates.\n */\nexport function getFlatIndexFrom3D(shape: [number, number, number]): string {\n  const strides = util.computeStrides(shape).map(d => d.toString());\n\n  return `\n  int getFlatIndex(ivec3 coords) {\n    return coords.x * ${strides[0]} + coords.y * ${strides[1]} + coords.z;\n  }\n`;\n}\n\nexport const ENCODE_FLOAT_SNIPPET = `\n  const float FLOAT_MAX = 1.70141184e38;\n  const float FLOAT_MIN = 1.17549435e-38;\n\n  lowp vec4 encode_float(highp float v) {\n    if (isnan(v)) {\n      return vec4(255, 255, 255, 255);\n    }\n\n    highp float av = abs(v);\n\n    if(av < FLOAT_MIN) {\n      return vec4(0.0, 0.0, 0.0, 0.0);\n    } else if(v > FLOAT_MAX) {\n      return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;\n    } else if(v < -FLOAT_MAX) {\n      return vec4(0.0, 0.0,  128.0, 255.0) / 255.0;\n    }\n\n    highp vec4 c = vec4(0,0,0,0);\n\n    highp float e = floor(log2(av));\n    highp float m = exp2(fract(log2(av))) - 1.0;\n\n    c[2] = floor(128.0 * m);\n    m -= c[2] / 128.0;\n    c[1] = floor(32768.0 * m);\n    m -= c[1] / 32768.0;\n    c[0] = floor(8388608.0 * m);\n\n    highp float ebias = e + 127.0;\n    c[3] = floor(ebias / 2.0);\n    ebias -= c[3] * 2.0;\n    c[2] += floor(ebias) * 128.0;\n\n    c[3] += 128.0 * step(0.0, -v);\n\n    return c / 255.0;\n  }\n`;\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\nconst {getBroadcastDims} = backend_util;\nimport {getGlslDifferences, GLSL} from './glsl_version';\nimport * as shader_util from './shader_compiler_util';\n\nexport type ShapeInfo = {\n  logicalShape: number[],\n  texShape: [number, number],\n  isUniform: boolean,\n  isPacked: boolean,\n  flatOffset: number\n};\n\nexport type InputInfo = {\n  name: string,\n  shapeInfo: ShapeInfo\n};\n\nexport function makeShader(\n    inputsInfo: InputInfo[], outputShape: ShapeInfo, userCode: string,\n    usesPackedTextures: boolean): string {\n  const prefixSnippets: string[] = [];\n  inputsInfo.forEach(x => {\n    const size = util.sizeFromShape(x.shapeInfo.logicalShape);\n\n    // Snippet when we decided to upload the values as uniform.\n    if (x.shapeInfo.isUniform) {\n      prefixSnippets.push(\n          `uniform float ${x.name}${size > 1 ? `[${size}]` : ''};`);\n    } else {\n      prefixSnippets.push(`uniform sampler2D ${x.name};`);\n      prefixSnippets.push(`uniform int offset${x.name};`);\n    }\n  });\n  const inputPrefixSnippet = prefixSnippets.join('\\n');\n\n  const inputSamplingSnippet =\n      inputsInfo\n          .map(x => getInputSamplingSnippet(x, outputShape, usesPackedTextures))\n          .join('\\n');\n  const outTexShape = outputShape.texShape;\n  const glsl = getGlslDifferences();\n  const floatTextureSampleSnippet = getFloatTextureSampleSnippet(glsl);\n  let outputSamplingSnippet: string;\n  let floatTextureSetOutputSnippet: string;\n  let shaderPrefix = getShaderPrefix(glsl);\n\n  if (outputShape.isPacked) {\n    outputSamplingSnippet =\n        getPackedOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRGBASnippet(glsl);\n  } else {\n    outputSamplingSnippet =\n        getOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRSnippet(glsl);\n  }\n\n  if (usesPackedTextures) {\n    shaderPrefix += SHADER_PACKED_PREFIX;\n  }\n\n  const source = [\n    shaderPrefix, floatTextureSampleSnippet, floatTextureSetOutputSnippet,\n    inputPrefixSnippet, outputSamplingSnippet, inputSamplingSnippet, userCode\n  ].join('\\n');\n  return source;\n}\n\nfunction getSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getSamplerScalar(inInfo);\n    case 1:\n      return getSampler1D(inInfo);\n    case 2:\n      return getSampler2D(inInfo);\n    case 3:\n      return getSampler3D(inInfo);\n    case 4:\n      return getSampler4D(inInfo);\n    case 5:\n      return getSampler5D(inInfo);\n    case 6:\n      return getSampler6D(inInfo);\n    default:\n      throw new Error(\n          `${shape.length}-D input sampling` +\n          ` is not yet supported`);\n  }\n}\n\nfunction getPackedSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getPackedSamplerScalar(inInfo);\n    case 1:\n      return getPackedSampler1D(inInfo);\n    case 2:\n      return getPackedSampler2D(inInfo);\n    case 3:\n      return getPackedSampler3D(inInfo);\n    default:\n      return getPackedSamplerND(inInfo);\n  }\n}\n\nfunction getInputSamplingSnippet(\n    inInfo: InputInfo, outShapeInfo: ShapeInfo,\n    usesPackedTextures = false): string {\n  let res = '';\n  if (usesPackedTextures) {\n    res += getPackedSamplerFromInInfo(inInfo);\n  } else {\n    res += getSamplerFromInInfo(inInfo);\n  }\n\n  const inShape = inInfo.shapeInfo.logicalShape;\n  const outShape = outShapeInfo.logicalShape;\n  if (inShape.length <= outShape.length) {\n    if (usesPackedTextures) {\n      res += getPackedSamplerAtOutputCoords(inInfo, outShapeInfo);\n    } else {\n      res += getSamplerAtOutputCoords(inInfo, outShapeInfo);\n    }\n  }\n  return res;\n}\n\nfunction getPackedOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutputPacked1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutputPacked2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutputPacked3DCoords(\n          outShape as [number, number, number], outTexShape);\n    default:\n      return getOutputPackedNDCoords(outShape, outTexShape);\n  }\n}\n\nfunction getOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutput1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutput2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutput3DCoords(\n          outShape as [number, number, number], outTexShape);\n    case 4:\n      return getOutput4DCoords(\n          outShape as [number, number, number, number], outTexShape);\n    case 5:\n      return getOutput5DCoords(\n          outShape as [number, number, number, number, number], outTexShape);\n    case 6:\n      return getOutput6DCoords(\n          outShape as [number, number, number, number, number, number],\n          outTexShape);\n    default:\n      throw new Error(\n          `${outShape.length}-D output sampling is not yet supported`);\n  }\n}\n\nfunction getFloatTextureSampleSnippet(glsl: GLSL): string {\n  return `\n    float sampleTexture(sampler2D textureSampler, vec2 uv) {\n      return ${glsl.texture2D}(textureSampler, uv).r;\n    }\n  `;\n}\n\nfunction getFloatTextureSetRSnippet(glsl: GLSL): string {\n  return `\n    void setOutput(float val) {\n      ${glsl.output} = vec4(val, 0, 0, 0);\n    }\n  `;\n}\n\nfunction getFloatTextureSetRGBASnippet(glsl: GLSL): string {\n  return `\n    void setOutput(vec4 val) {\n      ${glsl.output} = val;\n    }\n  `;\n}\n\nfunction getShaderPrefix(glsl: GLSL): string {\n  const SHADER_PREFIX = `${glsl.version}\n    precision highp float;\n    precision highp int;\n    precision highp sampler2D;\n    ${glsl.varyingFs} vec2 resultUV;\n    ${glsl.defineOutput}\n    const vec2 halfCR = vec2(0.5, 0.5);\n\n    struct ivec5\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n    };\n\n    struct ivec6\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n      int v;\n    };\n\n    uniform float NAN;\n    ${glsl.defineSpecialNaN}\n    ${glsl.defineSpecialInf}\n    ${glsl.defineRound}\n\n    int imod(int x, int y) {\n      return x - y * (x / y);\n    }\n\n    int idiv(int a, int b, float sign) {\n      int res = a / b;\n      int mod = imod(a, b);\n      if (sign < 0. && mod != 0) {\n        res -= 1;\n      }\n      return res;\n    }\n\n    //Based on the work of Dave Hoskins\n    //https://www.shadertoy.com/view/4djSRW\n    #define HASHSCALE1 443.8975\n    float random(float seed){\n      vec2 p = resultUV * seed;\n      vec3 p3  = fract(vec3(p.xyx) * HASHSCALE1);\n      p3 += dot(p3, p3.yzx + 19.19);\n      return fract((p3.x + p3.y) * p3.z);\n    }\n\n    ${SAMPLE_1D_SNIPPET}\n    ${SAMPLE_2D_SNIPPET}\n    ${SAMPLE_3D_SNIPPET}\n  `;\n\n  return SHADER_PREFIX;\n}\n\nconst SAMPLE_1D_SNIPPET = `\nvec2 uvFromFlat(int texNumR, int texNumC, int index) {\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\nvec2 packedUVfrom1D(int texNumR, int texNumC, int index) {\n  int texelIndex = index / 2;\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_2D_SNIPPET = `\nvec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,\n  int texNumC, int row, int col) {\n  int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_3D_SNIPPET = `\nvec2 packedUVfrom3D(int texNumR, int texNumC,\n    int texelsInBatch, int texelsInLogicalRow, int b,\n    int row, int col) {\n  int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SHADER_PACKED_PREFIX = `\n  float getChannel(vec4 frag, vec2 innerDims) {\n    vec2 modCoord = mod(innerDims, 2.);\n    return modCoord.x == 0. ?\n      (modCoord.y == 0. ? frag.r : frag.g) :\n      (modCoord.y == 0. ? frag.b : frag.a);\n  }\n  float getChannel(vec4 frag, int dim) {\n    float modCoord = mod(float(dim), 2.);\n    return modCoord == 0. ? frag.r : frag.g;\n  }\n`;\n\nfunction getOutputScalarCoords() {\n  return `\n    int getOutputCoords() {\n      return 0;\n    }\n  `;\n}\n\nfunction getOutputPacked1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (packedTexShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.x * ${packedTexShape[1]}.0);\n      }\n    `;\n  }\n\n  if (packedTexShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.y * ${packedTexShape[0]}.0);\n      }\n    `;\n  }\n\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      return 2 * (resTexRC.x * ${packedTexShape[1]} + resTexRC.y);\n    }\n  `;\n}\n\nfunction getOutput1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  if (texShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.x * ${texShape[1]}.0);\n      }\n    `;\n  }\n  if (texShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.y * ${texShape[0]}.0);\n      }\n    `;\n  }\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      return resTexRC.x * ${texShape[1]} + resTexRC.y;\n    }\n  `;\n}\n\nfunction getOutputPacked3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texelsInLogicalRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = texelsInLogicalRow * Math.ceil(shape[1] / 2);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec3(b, r, c);\n    }\n  `;\n}\n\nfunction getOutput3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n\nfunction getOutputPackedNDCoords(\n    shape: number[], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  const texelsInLogicalRow = Math.ceil(shape[shape.length - 1] / 2);\n  const texelsInBatch =\n      texelsInLogicalRow * Math.ceil(shape[shape.length - 2] / 2);\n  let texelsInBatchN = texelsInBatch;\n  let batches = ``;\n  let coords = 'b, r, c';\n\n  for (let b = 2; b < shape.length - 1; b++) {\n    texelsInBatchN *= shape[shape.length - b - 1];\n    batches = `\n      int b${b} = index / ${texelsInBatchN};\n      index -= b${b} * ${texelsInBatchN};\n    ` + batches;\n    coords = `b${b}, ` + coords;\n  }\n\n  return `\n    ivec${shape.length} getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      ${batches}\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec${shape.length}(${coords});\n    }\n  `;\n}\n\nfunction getOutput4DCoords(\n    shape: [number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2'], shape);\n\n  return `\n    ivec4 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec4(r, c, d, d2);\n    }\n  `;\n}\n\nfunction getOutput5DCoords(\n    shape: [number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3'], shape);\n\n  return `\n    ivec5 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx * vec2(${texShape[0]},\n                             ${texShape[1]}));\n\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec5 outShape = ivec5(r, c, d, d2, d3);\n      return outShape;\n    }\n  `;\n}\n\nfunction getOutput6DCoords(\n    shape: [number, number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3', 'd4'], shape);\n\n  return `\n    ivec6 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec6 result = ivec6(r, c, d, d2, d3, d4);\n      return result;\n    }\n  `;\n}\n\nfunction getOutputPacked2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return 2 * ivec2(resultUV.yx * vec2(${packedTexShape[0]}, ${\n        packedTexShape[1]}));\n      }\n    `;\n  }\n\n  // texels needed to accommodate a logical row\n  const texelsInLogicalRow = Math.ceil(shape[1] / 2);\n\n  /**\n   * getOutputCoords\n   *\n   * resTexRC: The rows and columns of the texels. If you move over one\n   * texel to the right in the packed texture, you are moving over one column\n   * (not two).\n   *\n   * index: The texel index\n   */\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getOutput2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return ivec2(resultUV.yx * vec2(${texShape[0]}, ${texShape[1]}));\n      }\n    `;\n  }\n  if (shape[1] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(index, 0);\n      }\n    `;\n  }\n  if (shape[0] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(0, index);\n      }\n    `;\n  }\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      int r = index / ${shape[1]};\n      int c = index - r * ${shape[1]};\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getFlatOffsetUniformName(texName: string): string {\n  return `offset${texName}`;\n}\n\nfunction getPackedSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}() {\n      return ${glsl.texture2D}(${texName}, halfCR);\n    }\n  `;\n}\n\nfunction getSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  if (inputInfo.shapeInfo.isUniform) {\n    return `float ${funcName}() {return ${texName};}`;\n  }\n  const [texNumR, texNumC] = inputInfo.shapeInfo.texShape;\n  if (texNumR === 1 && texNumC === 1) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n\n  const [tNumR, tNumC] = inputInfo.shapeInfo.texShape;\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}() {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int index) {\n      vec2 uv = packedUVfrom1D(\n        ${packedTexShape[0]}, ${packedTexShape[1]}, index);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int index) {\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const tNumR = texShape[0];\n  const tNumC = texShape[1];\n\n  if (tNumC === 1 && tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  if (tNumC === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2(0.5, (float(index + ${offset}) + 0.5) / ${tNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2((float(index + ${offset}) + 0.5) / ${tNumC}.0, 0.5);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  return `\n    float ${funcName}(int index) {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const glsl = getGlslDifferences();\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    return `\n      vec4 ${funcName}(int row, int col) {\n        vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n\n        return ${glsl.texture2D}(${texName}, uv);\n      }\n    `;\n  }\n\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const valuesPerRow = Math.ceil(shape[1] / 2);\n\n  return `\n    vec4 ${funcName}(int row, int col) {\n      vec2 uv = packedUVfrom2D(${valuesPerRow}, ${packedTexShape[0]}, ${\n      packedTexShape[1]}, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    const texNumR = texShape[0];\n    const texNumC = texShape[1];\n    return `\n    float ${funcName}(int row, int col) {\n      vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col) {\n        int index = round(dot(vec2(row, col), vec2(${shape[1]}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const offset = getFlatOffsetUniformName(texName);\n  if (texNumC === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2(0.5, (index + 0.5) / ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n  if (texNumR === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2((index + 0.5) / ${texNumC}.0, 0.5);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  return `\n  float ${funcName}(int row, int col) {\n    // Explicitly use integer operations as dot() only works on floats.\n    int index = row * ${shape[1]} + col + ${offset};\n    vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n    return sampleTexture(${texName}, uv);\n  }\n`;\n}\n\nfunction getPackedSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  if (shape[0] === 1) {\n    const squeezedShape = shape.slice(1);\n    const keptDims = [1, 2];\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['b', 'row', 'col'];\n    return `\n        ${getPackedSamplerFromInInfo(newInputInfo)}\n        vec4 ${funcName}(int b, int row, int col) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = valuesPerRow * Math.ceil(shape[1] / 2);\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int b, int row, int col) {\n      vec2 uv = packedUVfrom3D(\n        ${texNumR}, ${texNumC}, ${texelsInBatch}, ${valuesPerRow}, b, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride0 = shape[1] * shape[2];\n  const stride1 = shape[2];\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col', 'depth'];\n    return `\n        ${getSamplerFromInInfo(newInputInfo)}\n        float ${funcName}(int row, int col, int depth) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth) {\n        int index = round(dot(vec3(row, col, depth),\n                          vec3(${stride0}, ${stride1}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n        float ${funcName}(int row, int col, int depth) {\n          float texR = float(row);\n          float texC = dot(vec2(col, depth), vec2(${stride1}, 1));\n          vec2 uv = (vec2(texC, texR) + halfCR) /\n                     vec2(${texNumC}.0, ${texNumR}.0);\n          return sampleTexture(${texName}, uv);\n        }\n      `;\n  }\n\n  if (texNumC === stride1 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col, int depth) {\n      float texR = dot(vec2(row, col), vec2(${shape[1]}, 1));\n      float texC = float(depth);\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n      float ${funcName}(int row, int col, int depth) {\n        // Explicitly use integer operations as dot() only works on floats.\n        int index = row * ${stride0} + col * ${stride1} + depth + ${offset};\n        vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n        return sampleTexture(${texName}, uv);\n      }\n  `;\n}\n\nfunction getPackedSamplerND(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const rank = shape.length;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[rank - 1] / 2);\n  let texelsInBatch = valuesPerRow * Math.ceil(shape[rank - 2] / 2);\n  let params = `int b, int row, int col`;\n  let index = `b * ${texelsInBatch} + (row / 2) * ${valuesPerRow} + (col / 2)`;\n  for (let b = 2; b < rank - 1; b++) {\n    params = `int b${b}, ` + params;\n    texelsInBatch *= shape[rank - b - 1];\n    index = `b${b} * ${texelsInBatch} + ` + index;\n  }\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}(${params}) {\n      int index = ${index};\n      int texR = index / ${texNumC};\n      int texC = index - texR * ${texNumC};\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}, ${texNumR});\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler4D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride2 = shape[3];\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        int index = round(dot(vec4(row, col, depth, depth2),\n                          vec4(${stride0}, ${stride1}, ${stride2}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = float(row);\n        float texC =\n            dot(vec3(col, depth, depth2),\n                vec3(${stride1}, ${stride2}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride2 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = dot(vec3(row, col, depth),\n                         vec3(${shape[1] * shape[2]}, ${shape[2]}, 1));\n        float texC = float(depth2);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} +\n          depth * ${stride2} + depth2;\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler5D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride3 = shape[4];\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float index = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          depth3;\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n                         vec4(${stride1}, ${stride2}, ${stride3}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  if (texNumC === stride3 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float texR = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3]},\n               ${shape[2] * shape[3]}, ${shape[3]}, 1));\n        int texC = depth3;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler6D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3', 'depth4'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  const stride4 = shape[5];\n  const stride3 = shape[4] * stride4;\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n        int index = round(dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          dot(\n            vec2(depth3, depth4),\n            vec2(${stride4}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n          vec4(${stride1}, ${stride2}, ${stride3}, ${stride4})) +\n               float(depth4);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride4 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        float texR = dot(vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3] * shape[4]},\n               ${shape[2] * shape[3] * shape[4]},\n               ${shape[3] * shape[4]},\n               ${shape[4]})) + float(depth3);\n        int texC = depth4;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 * ${stride4} + depth4 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getUniformSampler(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n\n  if (inSize < 2) {\n    return `return ${texName};`;\n  }\n  return `\n    for (int i = 0; i < ${inSize}; i++) {\n      if (i == index) {\n        return ${texName}[i];\n      }\n    }\n  `;\n}\n\nfunction getPackedSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n\n  const type = getCoordsDataType(outRank);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  let output = `return outputValue;`;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n  const isInputScalar = inSize === 1;\n  const outSize = util.sizeFromShape(outShapeInfo.logicalShape);\n  const isOutputScalar = outSize === 1;\n\n  if (inRank === 1 && !isInputScalar && !isOutputScalar) {\n    output = `\n      return vec4(outputValue.xy, outputValue.xy);\n    `;\n  } else if (isInputScalar && !isOutputScalar) {\n    if (outRank === 1) {\n      output = `\n        return vec4(outputValue.x, outputValue.x, 0., 0.);\n      `;\n    } else {\n      output = `\n        return vec4(outputValue.x);\n      `;\n    }\n  } else if (broadcastDims.length) {\n    const rows = inRank - 2;\n    const cols = inRank - 1;\n\n    if (broadcastDims.indexOf(rows) > -1 && broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.x);`;\n    } else if (broadcastDims.indexOf(rows) > -1) {\n      output = `return vec4(outputValue.x, outputValue.y, ` +\n          `outputValue.x, outputValue.y);`;\n    } else if (broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.xx, outputValue.zz);`;\n    }\n  }\n\n  return `\n    vec4 ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      vec4 outputValue = get${texFuncSnippet}(${unpackedCoordsSnippet});\n      ${output}\n    }\n  `;\n}\n\nfunction getSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const outTexShape = outShapeInfo.texShape;\n  const inTexShape = inputInfo.shapeInfo.texShape;\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  if (!inputInfo.shapeInfo.isUniform && inRank === outRank &&\n      inputInfo.shapeInfo.flatOffset == null &&\n      util.arraysEqual(inTexShape, outTexShape)) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, resultUV);\n      }\n    `;\n  }\n\n  const type = getCoordsDataType(outRank);\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  return `\n    float ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      return get${texFuncSnippet}(${unpackedCoordsSnippet});\n    }\n  `;\n}\n\nexport function getCoordsDataType(rank: number): string {\n  if (rank <= 1) {\n    return 'int';\n  } else if (rank === 2) {\n    return 'ivec2';\n  } else if (rank === 3) {\n    return 'ivec3';\n  } else if (rank === 4) {\n    return 'ivec4';\n  } else if (rank === 5) {\n    return 'ivec5';\n  } else if (rank === 6) {\n    return 'ivec6';\n  } else {\n    throw Error(`GPU for rank ${rank} is not yet supported`);\n  }\n}\n\n/** Returns a new input info (a copy) that has a squeezed logical shape. */\nfunction squeezeInputInfo(\n    inInfo: InputInfo, squeezedShape: number[]): InputInfo {\n  // Deep copy.\n  const newInputInfo: InputInfo = JSON.parse(JSON.stringify(inInfo));\n  newInputInfo.shapeInfo.logicalShape = squeezedShape;\n  return newInputInfo;\n}\n\nfunction getSqueezedParams(params: string[], keptDims: number[]): string {\n  return keptDims.map(d => params[d]).join(', ');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ArgMinMaxPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      shape: number[], windowSize: number, op: 'max'|'min',\n      firstPass: boolean) {\n    util.assert(\n        shape.length > 2,\n        () => `Packed arg${\n            op.charAt(0).toUpperCase() +\n            op.slice(1)} supports only inputs with rank above 2.`);\n    const inSize = shape[shape.length - 1];\n    const outSize = Math.ceil(inSize / windowSize);\n    this.outputShape = shape.slice(0, -1);\n    if (outSize > 1) {\n      this.outputShape.push(outSize);\n    }\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    const outShape = this.outputShape;\n    const rank = outShape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n\n    let sourceLocSetup;\n    let sourceRank;\n    if (outSize === 1) {\n      sourceRank = rank + 1;\n      const sourceLocDType = getCoordsDataType(sourceRank);\n      sourceLocSetup = `\n        ${sourceLocDType} sourceLocR = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 1]};\n        ${sourceLocDType} sourceLocG = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 2]};\n        ${sourceLocDType} sourceLocA = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 1]};\n        ${sourceLocDType} sourceLocB = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 2]};`;\n    } else {\n      sourceRank = rank;\n      sourceLocSetup = `\n        ${dtype} sourceLocR = coords;\n        ++${coords[rank - 1]};\n        ${dtype} sourceLocG = coords;\n        ++${coords[rank - 2]};\n        ${dtype} sourceLocA = coords;\n        --${coords[rank - 1]};\n        ${dtype} sourceLocB = coords;\n        --${coords[rank - 2]};`;\n    }\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, sourceRank);\n    const inChannel = '.' + channels[sourceRank - 1];  // e.g. \".b\" for rank 3.\n    const intChannels = channels.map(x => 'int ' + x);\n    const srcRCoords =\n        getChannels('sourceLocR', sourceRank - 1).concat('inIdx.r');\n    const srcGCoords =\n        getChannels('sourceLocG', sourceRank - 1).concat('inIdx.g');\n    const srcBCoords =\n        getChannels('sourceLocB', sourceRank - 1).concat('inIdx.b');\n    const srcACoords =\n        getChannels('sourceLocA', sourceRank - 1).concat('inIdx.a');\n\n    const compOp = (op === 'max') ? 'greaterThan' : 'lessThan';\n    const fetchCandidateIdx = firstPass ? '' : `\n          inIdx = round(vec4(getBestIndicesAChannel(${srcRCoords.join()}),\n                             getBestIndicesAChannel(${srcGCoords.join()}),\n                             getBestIndicesAChannel(${srcBCoords.join()}),\n                             getBestIndicesAChannel(${srcACoords.join()})));`;\n\n    const fetchValue = `vec4(\n            getAChannel(${srcRCoords.join()}),\n            hasNextCol ? getAChannel(${srcGCoords.join()}) : 0.,\n            hasNextRow ? getAChannel(${srcBCoords.join()}) : 0.,\n            hasNextRow && hasNextCol ? getAChannel(${srcACoords.join()}) : 0.)`;\n\n    const getBestIndicesAChannelSnippet = firstPass ? '' : `\n      float getBestIndicesAChannel(${intChannels.join()}) {\n        return getChannel(getBestIndicesA(${channels.join()}),\n                                          vec2(${channels.slice(-2).join()}));\n      }`;\n\n    this.userCode = `\n      float getAChannel(${intChannels.join()}) {\n        return getChannel(getA(${channels.join()}),\n                               vec2(${channels.slice(-2).join()}));\n      }\n      ${getBestIndicesAChannelSnippet}\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        bool hasNextCol = ${coords[rank - 1]} < ${outShape[rank - 1] - 1};\n        bool hasNextRow = ${coords[rank - 2]} < ${outShape[rank - 2] - 1};\n        ${sourceLocSetup}\n        ivec4 srcIdx = ivec4(sourceLocR${inChannel}, sourceLocG${inChannel},\n          sourceLocB${inChannel}, sourceLocA${inChannel}) * ${windowSize};\n        ivec4 inIdx = srcIdx;\n        vec4 bestIndex = vec4(inIdx);\n        vec4 bestValue = ${fetchValue};\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          inIdx = srcIdx;\n          ${fetchCandidateIdx}\n          vec4 candidate = ${fetchValue};\n          bvec4 nan = isnan(candidate);\n          bvec4 replace = bvec4(\n            vec4(${compOp}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));\n\n          bestValue = vec4(replace.x  ? candidate.x : bestValue.x,\n                           replace.y  ? candidate.y : bestValue.y,\n                           replace.z  ? candidate.z : bestValue.z,\n                           replace.w  ? candidate.w : bestValue.w);\n          bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));\n          srcIdx++;\n        }\n        setOutput(bestIndex);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AvgPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth};\n            wC+= ${dilationWidth}) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n\n            dotProd += dyValue * avgMultiplier;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class AvgPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n\n              dotProd += dyValue * avgMultiplier;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\n// (Ar + Ai)(Br + Bi) =\n// ArBr + ArBi + AiBr + AiBi = ArBr - AB + ArBi + AiBr\n// Yr = ArBr - AB\n// Yi = ArBi + AiBr\nexport const COMPLEX_MULTIPLY = {\n  REAL: 'return areal * breal - aimag * bimag;',\n  IMAG: 'return areal * bimag + aimag * breal;'\n};\n\nexport class BinaryOpComplexProgram implements GPGPUProgram {\n  variableNames = ['AReal', 'AImag', 'BReal', 'BImag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    this.userCode = `\n      float binaryOpComplex(\n          float areal, float aimag, float breal, float bimag) {\n        ${op}\n      }\n\n      void main() {\n        float areal = getARealAtOutCoords();\n        float aimag = getAImagAtOutCoords();\n        float breal = getBRealAtOutCoords();\n        float bimag = getBImagAtOutCoords();\n        setOutput(binaryOpComplex(areal, aimag, breal, bimag));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nconst CHECK_NAN_SNIPPET = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const ADD = 'return a + b;';\nexport const SUB = 'return a - b;';\nexport const MUL = 'return a * b;';\n\n// We use native integer division to deal with floating point imprecision. Since\n// we implement floor division and glsl implements truncated division, we\n// correct for this by subtracting 1 from result when the result is negative and\n// there is a remainder.\nexport const INT_DIV = `\n  float s = sign(a) * sign(b);\n  int ia = round(a);\n  int ib = round(b);\n  if (ib != 0) {\n    // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n    return float(idiv(ia, ib, s));\n  } else {\n    return NAN;\n  }\n`;\n\nexport const POW = `\nif(a < 0.0 && floor(b) < b){\n  return NAN;\n}\nif (b == 0.0) {\n  return 1.0;\n}\nreturn (round(mod(b, 2.0)) != 1) ?\n    pow(abs(a), b) : sign(a) * pow(abs(a), b);\n`;\nexport const SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const EQUAL = `return float(a == b);`;\n\nexport const NOT_EQUAL = `return float(a != b);`;\n\nexport const LESS = `return float(a < b);`;\n\nexport const LESS_EQUAL = `return float(a <= b);`;\n\nexport const GREATER = `return float(a > b);`;\n\nexport const GREATER_EQUAL = `return float(a >= b);`;\n\nexport const LOGICAL_AND = `return float(a >= 1.0 && b >= 1.0);`;\n\nexport const LOGICAL_OR = `return float(a >= 1.0 || b >= 1.0);`;\n\nexport const MAX = CHECK_NAN_SNIPPET + `\n  return max(a, b);\n`;\nexport const MIN = CHECK_NAN_SNIPPET + `\n  return min(a, b);\n`;\nexport const MOD = `if (b == 0.0) return NAN;\n  return mod(a, b);`;\n\nexport const ELU_DER = `return (b >= 1.0) ? a : a * (b + 1.0);`;\n\nexport const PRELU = `return (a < 0.) ? b * a : a;`;\n\nexport class BinaryOpProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    this.userCode = `\n      float binaryOperation(float a, float b) {\n        ${op}\n      }\n\n      void main() {\n        float a = getAAtOutCoords();\n        float b = getBAtOutCoords();\n        setOutput(binaryOperation(a, b));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nconst CHECK_NAN_SNIPPET = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\nexport const INT_DIV = `\n  ivec4 ia = round(a);\n  ivec4 ib = round(b);\n  bvec4 cond = notEqual(ib, ivec4(0));\n  ivec4 result = ivec4(0);\n  vec4 s = sign(a) * sign(b);\n\n  // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n  if (cond[0]) {\n    result[0] = idiv(ia[0], ib[0], s[0]);\n  }\n  if (cond[1]) {\n    result[1] = idiv(ia[1], ib[1], s[1]);\n  }\n  if (cond[2]) {\n    result[2] = idiv(ia[2], ib[2], s[2]);\n  }\n  if (cond[3]) {\n    result[3] = idiv(ia[3], ib[3], s[3]);\n  }\n  return vec4(result);\n`;\n\nexport const POW = `\n  // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.\n  vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));\n  vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);\n  vec4 result = multiplier * pow(abs(a), b);\n\n  // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS\n  bvec4 isExpZero = equal(b, vec4(0.0));\n  result.r = isExpZero.r ? 1.0 : result.r;\n  result.g = isExpZero.g ? 1.0 : result.g;\n  result.b = isExpZero.b ? 1.0 : result.b;\n  result.a = isExpZero.a ? 1.0 : result.a;\n\n  vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const PRELU = `\n  vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));\n  return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);\n`;\n\nexport const ELU_DER = `\n  vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));\n  return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));\n`;\n\nexport const EQUAL = `\n  return vec4(equal(a, b));\n`;\n\nexport const NOT_EQUAL = `\n  return vec4(notEqual(a, b));\n`;\n\nexport const LESS = `\n  return vec4(lessThan(a, b));\n`;\n\nexport const LESS_EQUAL = `\n  return vec4(lessThanEqual(a, b));\n`;\n\nexport const GREATER = `\n  return vec4(greaterThan(a, b));\n`;\n\nexport const GREATER_EQUAL = `\n  return vec4(greaterThanEqual(a, b));\n`;\n\nexport const LOGICAL_AND = `\n  return vec4(\n    vec4(greaterThanEqual(a, vec4(1.0))) *\n    vec4(greaterThanEqual(b, vec4(1.0))));\n`;\n\nexport const LOGICAL_OR = `\n  return min(\n    vec4(greaterThanEqual(a, vec4(1.0))) +\n    vec4(greaterThanEqual(b, vec4(1.0))),\n    vec4(1.0));\n`;\n\nexport const MAX = `\n  vec4 result = vec4(max(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MIN = `\n  vec4 result = vec4(min(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MOD = `\n  vec4 result = mod(a, b);\n  vec4 isNaN = vec4(equal(b, vec4(0.0)));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport class BinaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n  supportsBroadcasting = true;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      op: string, aShape: number[], bShape: number[],\n      checkOutOfBounds = false) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const rank = this.outputShape.length;\n    let checkOutOfBoundsString = '';\n    if (checkOutOfBounds) {\n      if (rank === 0 || util.sizeFromShape(this.outputShape) === 1) {\n        checkOutOfBoundsString = `\n          result.y = 0.;\n          result.z = 0.;\n          result.w = 0.;\n        `;\n      } else {\n        const dtype = getCoordsDataType(rank);\n        checkOutOfBoundsString = `\n          ${dtype} coords = getOutputCoords();\n        `;\n        if (rank === 1) {\n          checkOutOfBoundsString += `\n            result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;\n            result.z = 0.;\n            result.w = 0.;\n          `;\n        } else {\n          const channels = getChannels('coords', rank);\n          checkOutOfBoundsString += `\n            bool nextRowOutOfBounds =\n              (${channels[rank - 2]} + 1) >= ${this.outputShape[rank - 2]};\n            bool nextColOutOfBounds =\n              (${channels[rank - 1]} + 1) >= ${this.outputShape[rank - 1]};\n            result.y = nextColOutOfBounds ? 0. : result.y;\n            result.z = nextRowOutOfBounds ? 0. : result.z;\n            result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;\n          `;\n        }\n      }\n    }\n\n    this.userCode = `\n      vec4 binaryOperation(vec4 a, vec4 b) {\n        ${op}\n      }\n\n      void main() {\n        vec4 a = getAAtOutCoords();\n        vec4 b = getBAtOutCoords();\n\n        vec4 result = binaryOperation(a, b);\n        ${checkOutOfBoundsString}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        float value = getAAtOutCoords();\n        if (isnan(value)) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, minVal, maxVal));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        vec4 value = getAAtOutCoords();\n\n        if (any(isnan(value))) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, vec4(minVal), vec4(maxVal)));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ComplexAbsProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(shape: number[]) {\n    this.outputShape = shape;\n    this.userCode = `\n      void main() {\n        float re = abs(getRealAtOutCoords());\n        float im = abs(getImagAtOutCoords());\n        float mx = max(re, im);\n\n        // sadly the length function in glsl is not underflow-safe\n        // (at least not on Intel GPUs). So the safe solution is\n        // to ensure underflow-safety in all cases.\n        setOutput(\n          mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))\n        );\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ConcatProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  // Concats 2d tensors along axis=1. See comments in MathBackendWebGL.concat().\n  constructor(shapes: Array<[number, number]>) {\n    this.outputShape = backend_util.computeOutShape(shapes, 1 /* axis */);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][1];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][1];\n    }\n\n    const snippets = [`if (yC < ${offsets[0]}) setOutput(getT0(yR, yC));`];\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      snippets.push(\n          `else if (yC < ${offsets[i]}) ` +\n          `setOutput(getT${i}(yR, yC-${shift}));`);\n    }\n    const lastIndex = offsets.length;\n    const lastShift = offsets[offsets.length - 1];\n    snippets.push(`else setOutput(getT${lastIndex}(yR, yC-${lastShift}));`);\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int yR = coords.x;\n        int yC = coords.y;\n\n        ${snippets.join('\\n        ')}\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ConcatPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(shapes: number[][], axis: number) {\n    this.outputShape = backend_util.computeOutShape(shapes, axis);\n    const shape = this.outputShape;\n    const rank = shape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][axis];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][axis];\n    }\n\n    const channel = channels[axis];\n    const lastChannels = channels.slice(-2);\n    const allChannels = channels.join();\n\n    let getValueSnippet = `if (${channel} < ${offsets[0]}) {\n        return getChannel(\n            getT0(${allChannels}), vec2(${lastChannels.join()}));\n        }`;\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      // Note: the >= comparison below may seem unnecessary given the check\n      // above but is needed to workaround branch execution issues on some\n      // devices. It makes all the conditions exclusive without relying on\n      // execution order.\n      getValueSnippet += `\n        if (${channel} < ${offsets[i]}  && ${channel} >= ${offsets[i - 1]}) {\n          return getChannel(\n            getT${i}(${shiftedChannels(channels, channel, shift)}),\n            vec2(${shiftedChannels(lastChannels, channel, shift)}));\n        }`;\n    }\n    const lastIndex = offsets.length;\n    const shift = offsets[offsets.length - 1];\n    getValueSnippet += `\n        return getChannel(\n          getT${lastIndex}(${shiftedChannels(channels, channel, shift)}),\n          vec2(${shiftedChannels(lastChannels, channel, shift)}));`;\n\n    this.userCode = `\n      float getValue(${channels.map(x => 'int ' + x)}) {\n        ${getValueSnippet}\n      }\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        vec4 result = vec4(getValue(${coords}), 0., 0., 0.);\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} + 1;\n        if (${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.g = getValue(${coords});\n        }\n\n        ${coords[rank - 2]} = ${coords[rank - 2]} + 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]}) {\n          result.a = getValue(${coords});\n        }\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} - 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]} &&\n            ${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.b = getValue(${coords});\n        }\n        setOutput(result);\n      }\n    `;\n  }\n}\n\n/**\n * Return an expression for coordinates into a vector where a given channel\n * will be offset by [shift].\n *\n * @param channels the channels to consider\n * @param channel the channel we want shifted\n * @param shift  the amount to subtract from the channel.\n *\n * @returns a string of the form 'x, y-[shift], z' where any one channel can\n * have the shift applied.\n */\nfunction shiftedChannels(channels: string[], channel: string, shift: number) {\n  const channelIdx = channels.indexOf(channel);\n  const res = channels.map((c, idx) => {\n    if (idx === channelIdx) {\n      return `${c} - ${shift}`;\n    } else {\n      return c;\n    }\n  });\n  return res.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int d2 = coords.w;\n\n        // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              if (${isChannelsLast}) {\n                float dyValue = getDy(b, yR, yC, d2);\n                float xValue = getX(b, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              } else {\n                float dyValue = getDy(b, d2, yR, yC);\n                float xValue = getX(b, d1, xR, xC);\n                dotProd += (xValue * dyValue);\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[${channelDim}];\n\n        ivec2 dyCorner = ivec2(coords[${rowDim}], coords[${colDim}]) - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n\n              if (${isChannelsLast}) {\n                float xValue = getDy(batch, idyR, idyC, d2);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              } else {\n                float xValue = getDy(batch, d2, idyR, idyC);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    this.userCode = `\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int wF = coords.x;\n        int wR = coords.y;\n        int wC = coords.z;\n        int d1 = coords.w;\n        int d2 = coords.u;\n\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yF = 0; yF < ${convInfo.outDepth}; yF++) {\n            int xF = wF + yF * ${strideDepth} - ${padFront};\n\n            if (xF < 0 || xF >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n              int xR = wR + yR * ${strideHeight} - ${padTop};\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n                int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float dyValue = getDy(b, yF, yR, yC, d2);\n                float xValue = getX(b, xF, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padFront = filterDepth - 1 - convInfo.padInfo.front;\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d1 = coords.u;\n\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyFCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          float dyF = float(dyFCorner + wF) / ${strideDepth}.0;\n\n          if (dyF < 0.0 || dyF >= ${convInfo.outDepth}.0 || fract(dyF) > 0.0) {\n            continue;\n          }\n          int idyF = int(dyF);\n\n          int wFPerm = ${filterDepth} - 1 - wF;\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n              fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            int wRPerm = ${filterHeight} - 1 - wR;\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              int wCPerm = ${filterWidth} - 1 - wC;\n\n              for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n                float xValue = getDy(batch, idyF, idyR, idyC, d2);\n                float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int dm = coords.w;\n        int d2 = d1 * ${channelMul} + dm;\n\n        float dotProd = 0.0;\n\n        // TO DO: Vec4 over the batch size\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float dyValue = getDy(b, yR, yC, d2);\n              float xValue = getX(b, xR, xC, d1);\n              dotProd += (xValue * dyValue);\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class DepthwiseConv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[3];\n        ivec2 dyCorner = coords.yz - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        float dotProd = 0.0;\n\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            // TO DO: Vec4 over the channelMul\n            for (int dm = 0; dm < ${channelMul}; dm++) {\n              int d2 = d1 * ${channelMul} + dm;\n              float xValue = getDy(batch, idyR, idyC, d2);\n              float wValue = getW(wRPerm, wCPerm, d1, dm);\n              dotProd += xValue * wValue;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivationWeights = false) {\n    this.outputShape = convInfo.outShape;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivationWeights) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivationWeights) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d2 = coords[${channelDim}];\n\n        ivec2 xRCCorner =\n            ivec2(coords[${rowDim}], coords[${colDim}]) * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${convInfo.inWidth}) {\n              continue;\n            }\n\n            for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n              vec4 wValues = vec4(\n                getW(wR, wC, d1, d2),\n                getW(wR, wC, d1 + 1, d2),\n                getW(wR, wC, d1 + 2, d2),\n                getW(wR, wC, d1 + 3, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec4 xValues = vec4(\n                  getX(batch, xR, xC, d1),\n                  getX(batch, xR, xC, d1 + 1),\n                  getX(batch, xR, xC, d1 + 2),\n                  getX(batch, xR, xC, d1 + 3)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec4 xValues = vec4(\n                  getX(batch, d1, xR, xC),\n                  getX(batch, d1 + 1, xR, xC),\n                  getX(batch, d1 + 2, xR, xC),\n                  getX(batch, d1 + 3, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n\n            if (${inputDepthVec4Remainder === 1}) {\n\n              if (${isChannelsLast}) {\n                dotProd +=\n                    getX(batch, xR, xC, ${inputDepthNearestVec4}) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              } else {\n                dotProd +=\n                    getX(batch, ${inputDepthNearestVec4}, xR, xC) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              }\n\n            } else if (${inputDepthVec4Remainder === 2}) {\n              vec2 wValues = vec2(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec2 xValues = vec2(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec2 xValues = vec2(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            } else if (${inputDepthVec4Remainder === 3}) {\n              vec3 wValues = vec3(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 2, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec3 xValues = vec3(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec3 xValues = vec3(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 2, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            }\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nexport class Conv3DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.outShape;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n\n    this.userCode = `\n      const ivec3 strides = ivec3(${strideDepth}, ${strideHeight}, ${\n        strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d2 = coords.u;\n\n        ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xFCorner = xFRCCorner.x;\n        int xRCorner = xFRCCorner.y;\n        int xCCorner = xFRCCorner.z;\n\n        // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get\n        // y(yF, yR, yC, d2). ? = to be determined. : = across all\n        // values in that axis.\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          int xF = xFCorner + wF * ${dilationDepth};\n\n          if (xF < 0 || xF >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            int xR = xRCorner + wR * ${dilationHeight};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n                vec4 xValues = vec4(\n                  getX(batch, xF, xR, xC, d1),\n                  getX(batch, xF, xR, xC, d1 + 1),\n                  getX(batch, xF, xR, xC, d1 + 2),\n                  getX(batch, xF, xR, xC, d1 + 3)\n                );\n                vec4 wValues = vec4(\n                  getW(wF, wR, wC, d1, d2),\n                  getW(wF, wR, wC, d1 + 1, d2),\n                  getW(wF, wR, wC, d1 + 2, d2),\n                  getW(wF, wR, wC, d1 + 3, d2)\n                );\n\n                dotProd += dot(xValues, wValues);\n              }\n\n              if (${inputDepthVec4Remainder === 1}) {\n                dotProd +=\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}) *\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2);\n              } else if (${inputDepthVec4Remainder === 2}) {\n                vec2 xValues = vec2(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                vec2 wValues = vec2(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else if (${inputDepthVec4Remainder === 3}) {\n                vec3 xValues = vec3(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                vec3 wValues = vec3(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 2, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2 / ${channelMul};\n        int q = d2 - d1 * ${channelMul};\n\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${xNumRows}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${xNumCols}) {\n              continue;\n            }\n\n            float xVal = getX(batch, xR, xC, d1);\n            float wVal = getW(wR, wC, d1, q);\n            dotProd += xVal * wVal;\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConvPacked2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const texelsAcross = filterWidth;\n\n    let mainLoop = `int xR; int xC; int xCOffset;`;\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `\n          vec4 xTexelR${r}C${c * 2} = vec4(0.);\n          vec4 wR${r}C${c} = vec4(0.);\n          vec4 xR${r}C${c} = vec4(0.);`;\n      }\n    }\n\n    /**\n     * This vectorized implementation works by gathering the values needed for\n     * each output channel's dot product into vec4's and then multiplying them\n     * all together (this happens in the final double for-loop below). Most of\n     * the main loop consists of constructing these vec4's with the minimum\n     * number of texture2D calls, which means making use of all four returned\n     * values from a texture2D call at once.\n     */\n    for (let r = 0; r < filterHeight; r++) {\n      for (let texelC = 0; texelC < texelsAcross; texelC++) {\n        const c = texelC * 2;\n\n        mainLoop += `\n          xR = xRCorner + ${r * dilationHeight};\n          xC = xCCorner + ${c * dilationWidth};\n        `;\n\n        if (strideWidth === 1) {\n          if (c < filterWidth) {\n            // If padding is odd, the outer texels have to be composed.\n            if (padLeft % 2 === 1) {\n              // TODO: Ensure vec4 previous does not result in redundant sample,\n              // and avoid setting xTexelRC's that exceed the boundary in the\n              // first place rather than resetting them to vec4(0)).\n\n              // To compute xCOffset:\n              // - If padding is odd, we must add 1 to ensure we ask for an\n              // even-numbered row.\n              // - We subtract 2 to access the previous texel.\n\n              mainLoop += `\n                xCOffset = xC + 1;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    xTexelR${r}C${c}.zw = vec2(0.);\n                  }\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + 1 - 2;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  vec4 previous = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    previous.zw = vec2(0.);\n                  }\n\n                  xR${r}C${c} = vec4(previous.zw, xTexelR${r}C${c}.xy);\n                } else {\n                  xR${r}C${c} = vec4(0, 0, xTexelR${r}C${c}.xy);\n                }\n              `;\n            } else {\n              // Padding is even, so xRC corresponds to a single texel.\n              mainLoop += `\n                if(xR >= 0 && xR < ${xNumRows} && xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xR${r}C${c} = xTexelR${r}C${c};\n              `;\n            }\n\n            if (c + 1 < filterWidth) {\n              // If dilation is even, the second entry should match the first\n              // (either both are composed or both are single samples). But if\n              // dilation is odd, then the second entry should be the opposite\n              // of the first (if the first is composed, the second is a single\n              // sample, and vice versa.)\n\n              const nextTexelOffset = padLeft % 2 === 0 ?\n                  util.nearestLargerEven(dilationWidth) :\n                  dilationWidth;\n\n              if ((dilationWidth % 2 === 0 && padLeft % 2 === 1) ||\n                  (dilationWidth % 2 !== 0 && padLeft % 2 !== 1)) {\n                mainLoop += `\n                  xCOffset = xC + ${padLeft % 2} + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n                `;\n\n                // If dilation > 1 then the xRC's will not be able to share any\n                // values, so each xRC will require two unique calls to getX.\n                if (dilationWidth > 1) {\n                  mainLoop += `\n                    xCOffset -= 2;\n                    if(xR >= 0 && xR < ${xNumRows} &&\n                      xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                      xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                    } else {\n                      xTexelR${r}C${c} = vec4(0.);\n                    }\n                  `;\n                }\n\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.xy);\n                `;\n              } else {\n                mainLoop += `\n                  xCOffset = xC + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n\n                  xR${r}C${c + 1} = xTexelR${r}C${c + 2};\n                `;\n              }\n            }\n          }\n        } else {  // stride > 1\n          if (c < filterWidth) {\n            mainLoop += `\n              if(xR >= 0 && xR < ${xNumRows}) {\n            `;\n\n            // Depending on whether padLeft is even or odd, we want either the\n            // xy or zw channels from X texels for xR${r}C${c}. If padLeft is\n            // even, xR${r}C${c + 1} is simply the zw channels of texels we've\n            // already sampled. But if padLeft is odd, xR${r}C{$c + 1}.zw will\n            // need to come from the xy channels of a new texel, hence the `vec4\n            // final` initialized below.\n            if (padLeft % 2 === 1) {\n              mainLoop += `\n                xCOffset = xC + 1 - ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                if(xC + 1 >= 0 && xC + 1 < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xC + 1, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  vec4 final = vec4(0.);\n                  xCOffset = xC + 1 + ${strideWidth};\n                  if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    final = getX(batch, xR, xCOffset, d1);\n                  }\n                  xR${r}C${c + 1} = vec4(xTexelR${r}C${c + 2}.xy, final.xy);\n                `;\n              }\n            } else {\n              mainLoop += `\n                if(xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.xy, xTexelR${r}C${c + 2}.xy);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n                `;\n              }\n            }\n\n            mainLoop += `}`;\n          }\n        }\n\n        if (c < filterWidth) {\n          mainLoop += `\n            vec4 wTexelR${r}C${c} = getW(${r}, ${c}, d1, q);\n            wR${r}C${c} = vec4(wTexelR${r}C${c}.xz, wTexelR${r}C${c}.xz);\n          `;\n\n          if (c + 1 < filterWidth) {\n            mainLoop += `\n              vec4 wTexelR${r}C${c + 1} = getW(${r}, ${c + 1}, d1, q);\n              wR${r}C${c + 1} =\n                vec4(wTexelR${r}C${c + 1}.xz, wTexelR${r}C${c + 1}.xz);`;\n          }\n        }\n      }\n    }\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `dotProd += xR${r}C${c} * wR${r}C${c};`;\n      }\n    }\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2;\n        int q = 0;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        vec4 dotProd = vec4(0.);\n\n        ${mainLoop}\n\n        vec4 result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class CropAndResizeProgram implements GPGPUProgram {\n  variableNames = ['Image', 'Boxes', 'BoxInd'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], boxShape: [number, number],\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number) {\n    const [batch, imageHeight, imageWidth, depth] = imageShape;\n    const [numBoxes, ] = boxShape;\n    const [cropHeight, cropWidth] = cropSize;\n    this.outputShape = [numBoxes, cropHeight, cropWidth, depth];\n    const methodId = method === 'bilinear' ? 1 : 0;\n\n    const [inputHeightFloat, inputWidthFloat] =\n        [`${imageHeight - 1}.0`, `${imageWidth - 1}.0`];\n\n    const [heightRatio, heightScale, inY] = cropHeight > 1 ?\n        [\n          `${(imageHeight - 1) / (cropHeight - 1)}`,\n          '(y2-y1) * height_ratio',\n          `y1*${inputHeightFloat} + float(y)*(height_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (y1+y2) * ${inputHeightFloat}`,\n        ];\n    const [widthRatio, widthScale, inX] = cropWidth > 1 ?\n        [\n          `${(imageWidth - 1) / (cropWidth - 1)}`,\n          '(x2-x1) * width_ratio',\n          `x1*${inputWidthFloat} + float(x)*(width_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (x1+x2) * ${inputWidthFloat}`,\n        ];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc\n    this.userCode = `\n      const float height_ratio = float(${heightRatio});\n      const float width_ratio = float(${widthRatio});\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int y = coords[1];\n        int x = coords[2];\n        int d = coords[3];\n\n        // get box vals\n        float y1 = getBoxes(b,0);\n        float x1 = getBoxes(b,1);\n        float y2 = getBoxes(b,2);\n        float x2 = getBoxes(b,3);\n\n        // get image in batch index\n        int bInd = round(getBoxInd(b));\n        if(bInd < 0 || bInd >= ${batch}) {\n          return;\n        }\n\n        float height_scale = ${heightScale};\n        float width_scale = ${widthScale};\n\n        float in_y = ${inY};\n        if( in_y < 0.0 || in_y > ${inputHeightFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n        float in_x = ${inX};\n        if( in_x < 0.0 || in_x > ${inputWidthFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n\n        vec2 sourceFracIndexCR = vec2(in_x,in_y);\n        if(${methodId} == 1) {\n          // Compute the four integer indices.\n          ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);\n          ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));\n\n          float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);\n          float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);\n          float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);\n          float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);\n\n          vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);\n\n          float top = topLeft + (topRight - topLeft) * fracCR.x;\n          float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;\n          float newValue = top + (bottom - top) * fracCR.y;\n          setOutput(newValue);\n        } else {\n          // Compute the coordinators of nearest neighbor point.\n          ivec2 sourceNearestCR = ivec2(floor(\n            sourceFracIndexCR + vec2(0.5,0.5)));\n          float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);\n          setOutput(newValue);\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class CumSumProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  index: WebGLUniformLocation;\n\n  constructor(shape: number[], exclusive: boolean, reverse: boolean) {\n    this.outputShape = shape;\n    const rank = shape.length;\n    const val = exclusive ? '0.0' : `getX(${getCoords(rank, 'coords')})`;\n    const length = shape[shape.length - 1];\n    let condition = '';\n    let idxString = '';\n    // When exclusive is set, the cumsum op becomes roll op that copies the\n    // value from the previous index based on the direction specified by the\n    // reverse flag.\n    if (exclusive) {\n      condition = reverse ? `end != ${length - 1}` : 'end != 0';\n      idxString = reverse ? 'end + 1' : 'end - 1';\n    } else {\n      condition = reverse ? `end + pow2 < ${length}` : 'end >= pow2';\n      idxString = (reverse ? 'end + pow2' : 'end - pow2');\n    }\n\n    this.userCode = `\n      uniform float index;\n      void main() {\n        ${getCoordsDataType(rank)} coords = getOutputCoords();\n        int end = ${getFinalCoord(rank, 'coords')};\n        float val = ${val};\n        int pow2 = int(pow(2.0, index));\n        if (${condition}) {\n          int idx = ${idxString};\n          ${getFinalCoord(rank, 'coords')} = idx;\n          val += getX(${getCoords(rank, 'coords')});\n        }\n        setOutput(val);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(index: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.index == null) {\n        this.index = gpgpu.getUniformLocation(webGLProgram, 'index');\n      }\n      gpgpu.gl.uniform1f(this.index, index);\n    };\n  }\n}\n\nfunction getCoords(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.x, ${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.x, ${name}.y, ${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.x, ${name}.y, ${name}.z, ${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n\nfunction getFinalCoord(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: [number, number, number];\n  packedInputs = false;\n  packedOutput = true;\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getA(rc.x, rc.y, rc.z);\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: [number, number, number];\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthToSpaceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  blockSize: number;\n  dataFormat: string;\n\n  constructor(\n      outputShape: number[], blockSize: number, dataFormat: 'NHWC'|'NCHW') {\n    this.outputShape = outputShape;\n    this.blockSize = blockSize;\n    this.dataFormat = dataFormat;\n    this.userCode = `\n    void main() {\n      ivec4 coords = getOutputCoords();\n      int b = coords[0];\n      int h = ${this.getHeightCoordString()};\n      int w = ${this.getWidthCoordString()};\n      int d = ${this.getDepthCoordString()};\n\n      int in_h = h / ${blockSize};\n      int offset_h = imod(h, ${blockSize});\n      int in_w = w / ${blockSize};\n      int offset_w = imod(w, ${blockSize});\n      int offset_d = (offset_h * ${blockSize} + offset_w) *\n        ${this.getOutputDepthSize()};\n      int in_d = d + offset_d;\n\n      float result = ${this.getInputSamplingString()};\n      setOutput(result);\n    }\n  `;\n  }\n\n  private getHeightCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[1]`;\n    } else {\n      return `coords[2]`;\n    }\n  }\n\n  private getWidthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[2]`;\n    } else {\n      return `coords[3]`;\n    }\n  }\n\n  private getDepthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[3]`;\n    } else {\n      return `coords[1]`;\n    }\n  }\n\n  private getOutputDepthSize(): number {\n    if (this.dataFormat === 'NHWC') {\n      return this.outputShape[3];\n    } else {\n      return this.outputShape[1];\n    }\n  }\n\n  private getInputSamplingString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `getX(b, in_h, in_w, in_d)`;\n    } else {\n      return `getX(b, in_d, in_h, in_w)`;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DiagProgram implements GPGPUProgram {\n  variableNames = ['X'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(size: number) {\n    this.outputShape = [size, size];\n    this.userCode = `\n      void main() {\n          ivec2 coords = getOutputCoords();\n          float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;\n          setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        float x = getAAtOutCoords();\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = false;\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: [number, number, number]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n        float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class EncodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let output = `result`;\n    if (inputIsUnsignedByte) {\n      output = `floor(result * 255. + 0.5)`;\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        int flatIndex = getFlatIndex(coords);\n        int offset = imod(flatIndex, 4);\n\n        flatIndex = idiv(flatIndex, 4, 1.);\n\n        int r = flatIndex / ${width};\n        int c = imod(flatIndex, ${width});\n        vec2 uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n        vec4 values = ${glsl.texture2D}(A, uv);\n\n        float result;\n\n        if(offset == 0) {\n          result = values[0];\n        } else if(offset == 1) {\n          result = values[1];\n        } else if(offset == 2) {\n          result = values[2];\n        } else {\n          result = values[3];\n        }\n\n        ${glsl.output} = vec4(${output}, 0., 0., 0.);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\n/*\nThis is how the shader encodes a tensor with shape = [2, 3, 5]\n(indices are [batch, row, col]).\n\n000|001   002|003   004|xxx   020|021   022|023   024|xxx\n-------   -------   -------   -------   -------   -------\n010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\n100|101   102|103   104|xxx   120|121   122|123   124|xxx\n-------   -------   -------   -------   -------   -------\n110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\nSingle texels contain only values from the same batch, and from adjacent rows\nand columns.\n */\n\nexport class EncodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let mainLoop = '';\n    let output = 'result';\n    if (inputIsUnsignedByte) {\n      output = 'floor(result * 255. + 0.5)';\n    }\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        const channel = row * 2 + col;\n\n        mainLoop += `\n          localCoords = coords;\n          if(localCoords[2] + ${col} < ${outputShape[2]}) {\n            localCoords[2] += ${col};\n            if(localCoords[1] + ${row} < ${outputShape[1]}) {\n              localCoords[1] += ${row};\n\n              flatIndex = getFlatIndex(localCoords);\n              offset = imod(flatIndex, 4);\n\n              flatIndex = idiv(flatIndex, 4, 1.);\n\n              r = flatIndex / ${width};\n              c = imod(flatIndex, ${width});\n              uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n              values = ${glsl.texture2D}(A, uv);\n\n              if(offset == 0) {\n                result[${channel}] = values[0];\n              } else if(offset == 1) {\n                result[${channel}] = values[1];\n              } else if(offset == 2) {\n                result[${channel}] = values[2];\n              } else {\n                result[${channel}] = values[3];\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        vec4 result = vec4(0.);\n        int flatIndex, r, c, offset;\n        ivec3 localCoords;\n        vec2 uv;\n        vec4 values;\n\n        ${mainLoop}\n\n        ${glsl.output} = ${output};\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const COMPLEX_FFT = {\n  REAL: 'return real * expR - imag * expI;',\n  IMAG: 'return real * expI + imag * expR;'\n};\n\nexport class FFTProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, inputShape: [number, number], inverse: boolean) {\n    const innerDim = inputShape[1];\n    this.outputShape = inputShape;\n\n    const exponentMultiplierSnippet =\n        inverse ? `2.0 * ${Math.PI}` : `-2.0 * ${Math.PI}`;\n    const resultDenominator = inverse ? `${innerDim}.0` : '1.0';\n\n    this.userCode = `\n      const float exponentMultiplier = ${exponentMultiplierSnippet};\n\n      float unaryOpComplex(float real, float expR, float imag, float expI) {\n        ${op}\n      }\n\n      float mulMatDFT(int batch, int index) {\n        float indexRatio = float(index) / float(${innerDim});\n        float exponentMultiplierTimesIndexRatio =\n            exponentMultiplier * indexRatio;\n\n        float result = 0.0;\n\n        for (int i = 0; i < ${innerDim}; i++) {\n          // x = (-2|2 * PI / N) * index * i;\n          float x = exponentMultiplierTimesIndexRatio * float(i);\n          float expR = cos(x);\n          float expI = sin(x);\n          float real = getReal(batch, i);\n          float imag = getImag(batch, i);\n\n          result +=\n              unaryOpComplex(real, expR, imag, expI) / ${resultDenominator};\n        }\n\n        return result;\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        setOutput(mulMatDFT(coords[0], coords[1]));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FillProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  valueLoc: WebGLUniformLocation;\n\n  constructor(shape: number[], value: number) {\n    this.variableNames = ['x'];\n    this.outputShape = shape;\n\n    this.userCode = `\n      uniform float value;\n      void main() {\n        // Input can be obtained from uniform value.\n        setOutput(value);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(value: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.valueLoc == null) {\n        this.valueLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'value');\n      }\n      gpgpu.gl.uniform1f(this.valueLoc, value);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherProgram implements GPGPUProgram {\n  variableNames = ['A', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], indicesLength: number, axis: number) {\n    const outputShape: number[] = aShape.slice();\n    outputShape[axis] = indicesLength;\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape, axis);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[], axis: number): string {\n  const rank = aShape.length;\n  if (rank > 4) {\n    throw Error(`Gather for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `int(getIndices(resRC))`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    if (i === axis) {\n      sourceCoords.push(`int(getIndices(${currentCoords[i]}))`);\n    } else {\n      sourceCoords.push(`${currentCoords[i]}`);\n    }\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherNDProgram implements GPGPUProgram {\n  variableNames = ['x', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  constructor(\n      private sliceDim: number, private strides: number[], shape: number[]) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    const strideString = this.sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${this.strides});\n         void main() {\n          ${dtype} coords = getOutputCoords();\n          int flattenIndex = 0;\n          for (int j = 0; j < ${this.sliceDim}; j++) {\n            int index = round(getIndices(coords[0], j));\n            flattenIndex += index * ${strideString};\n          }\n          setOutput(getX(flattenIndex, coords[1]));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PixelData, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {getGlslDifferences} from './glsl_version';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport * as webgl_util from './webgl_util';\n\nexport function createVertexShader(gl: WebGLRenderingContext): WebGLShader {\n  const glsl = getGlslDifferences();\n  const vertexShaderSource = `${glsl.version}\n    precision highp float;\n    ${glsl.attribute} vec3 clipSpacePos;\n    ${glsl.attribute} vec2 uv;\n    ${glsl.varyingVs} vec2 resultUV;\n\n    void main() {\n      gl_Position = vec4(clipSpacePos, 1);\n      resultUV = uv;\n    }`;\n  return webgl_util.createVertexShader(gl, vertexShaderSource);\n}\n\nexport function createVertexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // [x y z u v] * [upper-left, lower-left, upper-right, lower-right]\n  const vertexArray = new Float32Array(\n      [-1, 1, 0, 0, 1, -1, -1, 0, 0, 0, 1, 1, 0, 1, 1, 1, -1, 0, 1, 0]);\n  return webgl_util.createStaticVertexBuffer(gl, vertexArray);\n}\n\nexport function createIndexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // OpenGL (and WebGL) have \"CCW == front\" winding\n  const triangleVertexIndices = new Uint16Array([0, 1, 2, 2, 1, 3]);\n  return webgl_util.createStaticIndexBuffer(gl, triangleVertexIndices);\n}\n\nfunction createAndConfigureTexture(\n    gl: WebGLRenderingContext, width: number, height: number,\n    internalFormat: number, textureFormat: number,\n    textureType: number): WebGLTexture {\n  webgl_util.validateTextureSize(width, height);\n  const texture = webgl_util.createTexture(gl);\n\n  const tex2d = gl.TEXTURE_2D;\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(tex2d, texture));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MIN_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MAG_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          tex2d, 0, internalFormat, width, height, 0, textureFormat,\n          textureType, null));\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n  return texture;\n}\n\nexport function getInternalFormatForFloat32MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatFloat;\n}\n\nexport function createFloat32MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat32MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatHalfFloat;\n}\n\nexport function createFloat16MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, textureConfig.textureTypeHalfFloat);\n}\n\nexport function getInternalFormatForUnsignedBytesMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.downloadTextureFormat;\n}\n\nexport function createUnsignedBytesMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForUnsignedBytesMatrixTexture(textureConfig), gl.RGBA,\n      gl.UNSIGNED_BYTE);\n}\n\nexport function getInternalFormatForPackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedFloat;\n}\n\nexport function createPackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height, getInternalFormatForPackedMatrixTexture(textureConfig),\n      gl.RGBA, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16PackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedHalfFloat;\n}\n\nexport function createFloat16PackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16PackedMatrixTexture(textureConfig), gl.RGBA,\n      textureConfig.textureTypeHalfFloat);\n}\n\nexport function bindVertexProgramAttributeStreams(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    vertexBuffer: WebGLBuffer): boolean {\n  const posOffset = 0;               // x is the first buffer element\n  const uvOffset = 3 * 4;            // uv comes after [x y z]\n  const stride = (3 * 4) + (2 * 4);  // xyz + uv, each entry is 4-byte float.\n  webgl_util.callAndCheck(\n      gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, vertexBuffer));\n  const success = webgl_util.bindVertexBufferToProgramAttribute(\n      gl, program, 'clipSpacePos', vertexBuffer, 3, stride, posOffset);\n  return success &&\n      webgl_util.bindVertexBufferToProgramAttribute(\n          gl, program, 'uv', vertexBuffer, 2, stride, uvOffset);\n}\n\nexport function uploadDenseMatrixToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture, width: number,\n    height: number, data: TypedArray, textureConfig: TextureConfig) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n\n  let dataForUpload: TypedArray, texelDataType: number, internalFormat: number;\n  if (data instanceof Uint8Array) {\n    dataForUpload = new Uint8Array(width * height * 4);\n    texelDataType = gl.UNSIGNED_BYTE;\n    internalFormat = gl.RGBA;\n  } else {\n    dataForUpload = new Float32Array(width * height * 4);\n    texelDataType = gl.FLOAT;\n    internalFormat = textureConfig.internalFormatPackedFloat;\n  }\n\n  dataForUpload.set(data);\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          gl.TEXTURE_2D, 0, internalFormat, width, height, 0, gl.RGBA,\n          texelDataType, dataForUpload));\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function uploadPixelDataToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, pixels.width, pixels.height, 0, gl.RGBA,\n            gl.UNSIGNED_BYTE, (pixels as PixelData).data));\n  } else {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, gl.RGBA, gl.UNSIGNED_BYTE,\n            pixels as ImageData | HTMLImageElement | HTMLCanvasElement |\n                HTMLVideoElement));\n  }\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function createBufferFromOutputTexture(\n    gl2: WebGL2RenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLBuffer {\n  // Create and bind the buffer.\n  const buffer = gl2.createBuffer();\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer));\n\n  // Initialize the buffer to the size of the texture in bytes.\n  const bytesPerFloat = 4;\n  const valuesPerTexel = 4;\n  const bufferSizeBytes = bytesPerFloat * valuesPerTexel * rows * columns;\n\n  webgl_util.callAndCheck(\n      gl2,\n      () => gl2.bufferData(\n          gl2.PIXEL_PACK_BUFFER, bufferSizeBytes, gl2.STREAM_READ));\n\n  // Enqueue a command on the GPU command queue to copy of texture into the\n  // buffer.\n  webgl_util.callAndCheck(\n      gl2, () => gl2.readPixels(0, 0, columns, rows, gl2.RGBA, gl2.FLOAT, 0));\n\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null));\n\n  return buffer;\n}\n\nexport function downloadFloat32MatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer,\n    size: number): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget = new Float32Array(size);\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadByteEncodedFloatMatrixFromOutputTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig) {\n  const [w, h] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n\n  const numChannels = 4;\n  const downloadTarget = new Uint8Array(\n      tex_util.getUnpackedArraySizeFromMatrixSize(rows * columns, numChannels));\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, w, h, textureConfig.downloadTextureFormat, gl.UNSIGNED_BYTE,\n          downloadTarget));\n\n  // By wrapping the buffer in a Float32Array, we use native browser IEEE 754\n  // decoding of the 4 bytes that back each 32 bit float.\n  return new Float32Array(downloadTarget.buffer);\n}\n\nexport function downloadPackedMatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer, batch: number, rows: number,\n    cols: number, physicalRows: number, physicalCols: number,\n    textureConfig: TextureConfig): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget =\n      new Float32Array(tex_util.getPackedRGBAArraySizeFromMatrixShape(\n          physicalRows, physicalCols));\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadMatrixFromPackedOutputTexture(\n    gl: WebGLRenderingContext, physicalRows: number,\n    physicalCols: number): Float32Array {\n  const packedRGBA = new Float32Array(physicalRows * physicalCols * 4);\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, physicalCols, physicalRows, gl.RGBA, gl.FLOAT, packedRGBA));\n\n  return packedRGBA;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, PixelData, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext, setWebGLContext} from './canvas_util';\nimport * as gpgpu_util from './gpgpu_util';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport {WebGL1DisjointQueryTimerExtension, WebGL2DisjointQueryTimerExtension} from './webgl_types';\nimport * as webgl_util from './webgl_util';\n\nexport interface FenceContext {\n  query: WebGLQuery|WebGLSync;\n  isFencePassed(): boolean;\n}\n\nexport class GPGPUContext {\n  gl: WebGLRenderingContext;\n  textureFloatExtension: {};\n  textureHalfFloatExtension: {};\n  colorBufferFloatExtension: {};\n  colorBufferHalfFloatExtension: {};\n  disjointQueryTimerExtension: WebGL2DisjointQueryTimerExtension|\n      WebGL1DisjointQueryTimerExtension;\n  vertexBuffer: WebGLBuffer;\n  indexBuffer: WebGLBuffer;\n  framebuffer: WebGLFramebuffer;\n  outputTexture: WebGLTexture|null = null;\n  program: WebGLProgram|null = null;\n  private disposed = false;\n  private disjoint: boolean;\n  textureConfig: TextureConfig;\n\n  constructor(gl?: WebGLRenderingContext) {\n    const glVersion = env().getNumber('WEBGL_VERSION');\n    if (gl != null) {\n      this.gl = gl;\n      setWebGLContext(glVersion, gl);\n    } else {\n      this.gl = getWebGLContext(glVersion);\n    }\n    // WebGL 2.0 enables texture floats without an extension.\n    let COLOR_BUFFER_FLOAT = 'WEBGL_color_buffer_float';\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (env().getNumber('WEBGL_VERSION') === 1) {\n      const TEXTURE_FLOAT = 'OES_texture_float';\n      const TEXTURE_HALF_FLOAT = 'OES_texture_half_float';\n\n      this.textureFloatExtension =\n          webgl_util.getExtensionOrThrow(this.gl, TEXTURE_FLOAT);\n      if (webgl_util.hasExtension(this.gl, TEXTURE_HALF_FLOAT)) {\n        this.textureHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, TEXTURE_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support half float textures, yet the ' +\n            'environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n\n      this.colorBufferFloatExtension = this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, COLOR_BUFFER_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support color renderable half floats, yet ' +\n            'the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n    } else {\n      COLOR_BUFFER_FLOAT = 'EXT_color_buffer_float';\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_FLOAT)) {\n        this.colorBufferFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      } else if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      } else {\n        throw new Error('GL context does not support color renderable floats');\n      }\n    }\n\n    this.vertexBuffer = gpgpu_util.createVertexBuffer(this.gl);\n    this.indexBuffer = gpgpu_util.createIndexBuffer(this.gl);\n    this.framebuffer = webgl_util.createFramebuffer(this.gl);\n\n    this.textureConfig =\n        tex_util.getTextureConfig(this.gl, this.textureHalfFloatExtension);\n  }\n\n  private get debug(): boolean {\n    return env().getBool('DEBUG');\n  }\n\n  public dispose() {\n    if (this.disposed) {\n      return;\n    }\n    if (this.program != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound WebGLProgram.' +\n          ' This is probably a resource leak, delete the program with ' +\n          'GPGPUContext.deleteProgram before disposing.');\n    }\n    if (this.outputTexture != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound output matrix ' +\n          'texture.  This is probably a resource leak, delete the output ' +\n          'matrix texture with GPGPUContext.deleteMatrixTexture before ' +\n          'disposing.');\n    }\n    const gl = this.gl;\n    webgl_util.callAndCheck(gl, () => gl.finish());\n    webgl_util.callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteFramebuffer(this.framebuffer));\n    webgl_util.callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(\n        gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteBuffer(this.indexBuffer));\n    this.disposed = true;\n  }\n\n  public createFloat32MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat32MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createFloat16MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createUnsignedBytesMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createUnsignedBytesMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public uploadPixelDataToTexture(\n      texture: WebGLTexture,\n      pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadPixelDataToTexture(this.gl, texture, pixels);\n  }\n\n  public uploadDenseMatrixToTexture(\n      texture: WebGLTexture, width: number, height: number, data: TypedArray) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadDenseMatrixToTexture(\n        this.gl, texture, width, height, data, this.textureConfig);\n  }\n\n  public createFloat16PackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16PackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createPackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createPackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public deleteMatrixTexture(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    if (this.outputTexture === texture) {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n      this.outputTexture = null;\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.deleteTexture(texture));\n  }\n\n  public downloadByteEncodedFloatMatrixFromOutputTexture(\n      texture: WebGLTexture, rows: number, columns: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadByteEncodedFloatMatrixFromOutputTexture(\n            this.gl, rows, columns, this.textureConfig));\n  }\n\n  public downloadPackedMatrixFromBuffer(\n      buffer: WebGLBuffer, batch: number, rows: number, columns: number,\n      physicalRows: number, physicalCols: number): Float32Array {\n    return gpgpu_util.downloadPackedMatrixFromBuffer(\n        this.gl, buffer, batch, rows, columns, physicalRows, physicalCols,\n        this.textureConfig);\n  }\n\n  public downloadFloat32MatrixFromBuffer(buffer: WebGLBuffer, size: number):\n      Float32Array {\n    return gpgpu_util.downloadFloat32MatrixFromBuffer(this.gl, buffer, size);\n  }\n\n  public createBufferFromTexture(\n      texture: WebGLTexture, rows: number, columns: number): WebGLBuffer {\n    this.bindTextureToFrameBuffer(texture);\n    const result = gpgpu_util.createBufferFromOutputTexture(\n        this.gl as WebGL2RenderingContext, rows, columns, this.textureConfig);\n    this.unbindTextureToFrameBuffer();\n    return result;\n  }\n\n  public createAndWaitForFence(): Promise<void> {\n    const fenceContext = this.createFence(this.gl);\n    return this.pollFence(fenceContext);\n  }\n\n  private createFence(gl: WebGLRenderingContext): FenceContext {\n    let query: WebGLQuery|WebGLSync;\n    let isFencePassed: () => boolean;\n\n    if (env().getBool('WEBGL_FENCE_API_ENABLED')) {\n      const gl2 = gl as WebGL2RenderingContext;\n\n      const sync = gl2.fenceSync(gl2.SYNC_GPU_COMMANDS_COMPLETE, 0);\n      gl.flush();\n\n      isFencePassed = () => {\n        const status = gl2.clientWaitSync(sync, 0, 0);\n        return status === gl2.ALREADY_SIGNALED ||\n            status === gl2.CONDITION_SATISFIED;\n      };\n\n      query = sync;\n    } else if (\n        env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0) {\n      query = this.beginQuery();\n      this.endQuery();\n      isFencePassed = () => this.isQueryAvailable(\n          query,\n          env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n    } else {\n      // If we have no way to fence, return true immediately. This will fire in\n      // WebGL 1.0 when there is no disjoint query timer. In this case, because\n      // the fence passes immediately, we'll immediately ask for a download of\n      // the texture, which will cause the UI thread to hang.\n      isFencePassed = () => true;\n    }\n\n    return {query, isFencePassed};\n  }\n\n  public downloadMatrixFromPackedTexture(\n      texture: WebGLTexture, physicalRows: number,\n      physicalCols: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadMatrixFromPackedOutputTexture(\n            this.gl, physicalRows, physicalCols));\n  }\n\n  private vertexAttrsAreBound = false;\n\n  public createProgram(fragmentShaderSource: string): WebGLProgram {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    const fragmentShader: WebGLShader =\n        webgl_util.createFragmentShader(gl, fragmentShaderSource);\n    const vertexShader: WebGLShader = gpgpu_util.createVertexShader(gl);\n    const program: WebGLProgram = webgl_util.createProgram(gl);\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, vertexShader));\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, fragmentShader));\n    webgl_util.linkProgram(gl, program);\n    if (this.debug) {\n      webgl_util.validateProgram(gl, program);\n    }\n    if (!this.vertexAttrsAreBound) {\n      this.setProgram(program);\n      this.vertexAttrsAreBound = gpgpu_util.bindVertexProgramAttributeStreams(\n          gl, this.program, this.vertexBuffer);\n    }\n    return program;\n  }\n\n  public deleteProgram(program: WebGLProgram) {\n    this.throwIfDisposed();\n    if (program === this.program) {\n      this.program = null;\n    }\n    if (program != null) {\n      webgl_util.callAndCheck(this.gl, () => this.gl.deleteProgram(program));\n    }\n  }\n\n  public setProgram(program: WebGLProgram|null) {\n    this.throwIfDisposed();\n    this.program = program;\n    if ((this.program != null) && this.debug) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.useProgram(program));\n  }\n\n  public getUniformLocation(\n      program: WebGLProgram, uniformName: string,\n      shouldThrow = true): WebGLUniformLocation {\n    this.throwIfDisposed();\n    if (shouldThrow) {\n      return webgl_util.getProgramUniformLocationOrThrow(\n          this.gl, program, uniformName);\n    } else {\n      return webgl_util.getProgramUniformLocation(\n          this.gl, program, uniformName);\n    }\n  }\n\n  public getAttributeLocation(program: WebGLProgram, attribute: string):\n      number {\n    this.throwIfDisposed();\n    return webgl_util.callAndCheck(\n        this.gl, () => this.gl.getAttribLocation(program, attribute));\n  }\n\n  public getUniformLocationNoThrow(program: WebGLProgram, uniformName: string):\n      WebGLUniformLocation {\n    this.throwIfDisposed();\n    return this.gl.getUniformLocation(program, uniformName);\n  }\n\n  public setInputMatrixTexture(\n      inputMatrixTexture: WebGLTexture, uniformLocation: WebGLUniformLocation,\n      textureUnit: number) {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    webgl_util.bindTextureToProgramUniformSampler(\n        this.gl, inputMatrixTexture, uniformLocation, textureUnit);\n  }\n\n  public setOutputMatrixTexture(\n      outputMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.setOutputMatrixTextureDriver(outputMatrixTexture, columns, rows);\n  }\n\n  public setOutputPackedMatrixTexture(\n      outputPackedMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.throwIfDisposed();\n    const [width, height] =\n        tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n    this.setOutputMatrixTextureDriver(outputPackedMatrixTexture, width, height);\n  }\n\n  public setOutputMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    this.setOutputMatrixWriteRegionDriver(\n        startColumn, startRow, numColumns, numRows);\n  }\n\n  public setOutputPackedMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    throw new Error('setOutputPackedMatrixWriteRegion not implemented.');\n  }\n\n  public debugValidate() {\n    if (this.program != null) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.validateFramebuffer(this.gl);\n  }\n\n  public executeProgram() {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    const gl = this.gl;\n    if (this.debug) {\n      this.debugValidate();\n    }\n    webgl_util.callAndCheck(\n        gl, () => gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_SHORT, 0));\n  }\n\n  public blockUntilAllProgramsCompleted() {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(this.gl, () => this.gl.finish());\n  }\n\n  private getQueryTimerExtension(): WebGL1DisjointQueryTimerExtension\n      |WebGL2DisjointQueryTimerExtension {\n    if (this.disjointQueryTimerExtension == null) {\n      this.disjointQueryTimerExtension =\n          webgl_util.getExtensionOrThrow(\n              this.gl,\n              env().getNumber(\n                  'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2 ?\n                  'EXT_disjoint_timer_query_webgl2' :\n                  'EXT_disjoint_timer_query') as\n              WebGL1DisjointQueryTimerExtension |\n          WebGL2DisjointQueryTimerExtension;\n    }\n    return this.disjointQueryTimerExtension;\n  }\n\n  private getQueryTimerExtensionWebGL2(): WebGL2DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension();\n  }\n\n  private getQueryTimerExtensionWebGL1(): WebGL1DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension() as WebGL1DisjointQueryTimerExtension;\n  }\n\n  beginQuery(): WebGLQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const query = gl2.createQuery();\n      gl2.beginQuery(ext.TIME_ELAPSED_EXT, query);\n      return query;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    const query = ext.createQueryEXT() as WebGLQuery;\n    ext.beginQueryEXT(ext.TIME_ELAPSED_EXT, query);\n    return query;\n  }\n\n  endQuery() {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n      gl2.endQuery(ext.TIME_ELAPSED_EXT);\n      return;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    ext.endQueryEXT(ext.TIME_ELAPSED_EXT);\n  }\n\n  public async waitForQueryAndGetTime(query: WebGLQuery): Promise<number> {\n    await util.repeatedTry(\n        () => this.disposed ||  // while testing contexts are created / disposed\n                                // in rapid succession, so without this check we\n                                // may poll for the query timer indefinitely\n            this.isQueryAvailable(\n                query,\n                env().getNumber(\n                    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION')));\n    return this.getQueryTime(\n        query, env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n  }\n\n  private getQueryTime(query: WebGLQuery, queryTimerVersion: number): number {\n    if (queryTimerVersion === 0) {\n      return null;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n\n      const timeElapsedNanos = gl2.getQueryParameter(query, gl2.QUERY_RESULT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const timeElapsedNanos =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_EXT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    }\n  }\n\n  private isQueryAvailable(query: WebGLQuery, queryTimerVersion: number):\n      boolean {\n    if (queryTimerVersion === 0) {\n      return true;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const available =\n          gl2.getQueryParameter(query, gl2.QUERY_RESULT_AVAILABLE);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const available =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_AVAILABLE_EXT);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    }\n  }\n\n  pollFence(fenceContext: FenceContext) {\n    return new Promise<void>(resolve => {\n      this.addItemToPoll(() => fenceContext.isFencePassed(), () => resolve());\n    });\n  }\n\n  private itemsToPoll: PollItem[] = [];\n\n  pollItems(): void {\n    // Find the last query that has finished.\n    const index = linearSearchLastTrue(this.itemsToPoll.map(x => x.isDoneFn));\n    for (let i = 0; i <= index; ++i) {\n      const {resolveFn} = this.itemsToPoll[i];\n      resolveFn();\n    }\n    this.itemsToPoll = this.itemsToPoll.slice(index + 1);\n  }\n\n  private addItemToPoll(isDoneFn: () => boolean, resolveFn: () => void) {\n    this.itemsToPoll.push({isDoneFn, resolveFn});\n    if (this.itemsToPoll.length > 1) {\n      // We already have a running loop that polls.\n      return;\n    }\n    // Start a new loop that polls.\n    util.repeatedTry(() => {\n      this.pollItems();\n      // End the loop if no more items to poll.\n      return this.itemsToPoll.length === 0;\n    });\n  }\n\n  private bindTextureToFrameBuffer(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    webgl_util.bindColorTextureToFramebuffer(\n        this.gl, texture, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(this.gl);\n    }\n  }\n\n  private unbindTextureToFrameBuffer() {\n    if (this.outputTexture != null) {\n      webgl_util.bindColorTextureToFramebuffer(\n          this.gl, this.outputTexture, this.framebuffer);\n      if (this.debug) {\n        webgl_util.validateFramebuffer(this.gl);\n      }\n    } else {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n    }\n  }\n\n  private downloadMatrixDriver(\n      texture: WebGLTexture,\n      downloadAndDecode: () => Float32Array): Float32Array {\n    this.bindTextureToFrameBuffer(texture);\n    const result = downloadAndDecode();\n    this.unbindTextureToFrameBuffer();\n\n    return result;\n  }\n\n  private setOutputMatrixTextureDriver(\n      outputMatrixTextureMaybePacked: WebGLTexture, width: number,\n      height: number) {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    webgl_util.bindColorTextureToFramebuffer(\n        gl, outputMatrixTextureMaybePacked, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(gl);\n    }\n    this.outputTexture = outputMatrixTextureMaybePacked;\n    webgl_util.callAndCheck(gl, () => gl.viewport(0, 0, width, height));\n    webgl_util.callAndCheck(gl, () => gl.scissor(0, 0, width, height));\n  }\n\n  private setOutputMatrixWriteRegionDriver(\n      x: number, y: number, width: number, height: number) {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(\n        this.gl, () => this.gl.scissor(x, y, width, height));\n  }\n\n  private throwIfDisposed() {\n    if (this.disposed) {\n      throw new Error('Attempted to use disposed GPGPUContext.');\n    }\n  }\n\n  private throwIfNoProgram() {\n    if (this.program == null) {\n      throw new Error('No GPU program is currently set.');\n    }\n  }\n}\n\ntype PollItem = {\n  isDoneFn: () => boolean,\n  resolveFn: () => void\n};\n\n/**\n * Finds the index of the last true element using linear search.\n * Note: We can't do binary search because Chrome expects us to explicitly\n * test all fences before download:\n * https://github.com/tensorflow/tfjs/issues/1145\n */\nexport function linearSearchLastTrue(arr: Array<() => boolean>): number {\n  let i = 0;\n  for (; i < arr.length; ++i) {\n    const isDone = arr[i]();\n    if (!isDone) {\n      break;\n    }\n  }\n  return i - 1;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, Tensor, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport * as shader_compiler from './shader_compiler';\nimport {InputInfo, ShapeInfo} from './shader_compiler';\nimport {PackingScheme, TextureData, TextureUsage} from './tex_util';\n\nexport interface GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  /** If true, this program expects packed input textures. Defaults to false. */\n  packedInputs?: boolean;\n  /** If true, this program produces a packed texture. Defaults to false. */\n  packedOutput?: boolean;\n  /**\n   * Affects what type of texture we allocate for the output. Defaults to\n   * `TextureUsage.RENDER`.\n   */\n  outTexUsage?: TextureUsage;\n  /**\n   * The type of scheme to use when packing texels for the output values.\n   * See `PackingScheme` for details. Defaults to `PackingScheme.SHARED_BATCH`.\n   */\n  outPackingScheme?: PackingScheme;\n}\n\nexport interface GPGPUBinary {\n  webGLProgram: WebGLProgram;\n  program: GPGPUProgram;\n  uniformLocations: {[name: string]: WebGLUniformLocation};\n  source: string;\n  inShapeInfos: ShapeInfo[];\n  outShapeInfo: ShapeInfo;\n  infLoc: WebGLUniformLocation;\n  nanLoc: WebGLUniformLocation;\n}\n\nexport interface TensorData {\n  shape: number[];\n  texData: TextureData;\n  isUniform: boolean;\n  // Available when we decide to upload as uniform instead of texture.\n  uniformValues?: TypedArray;\n}\n\nexport function compileProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, program: GPGPUProgram, inputs: TensorData[],\n    output: TensorData): GPGPUBinary {\n  const userCode = program.userCode;\n  const inputInfos: InputInfo[] = inputs.map((input, i) => {\n    const shapeInfo: ShapeInfo = {\n      logicalShape: input.shape,\n      texShape: input.isUniform ? null : input.texData.texShape,\n      isUniform: input.isUniform,\n      isPacked: input.isUniform ? false : input.texData.isPacked,\n      flatOffset: null\n    };\n    if (input.texData != null && input.texData.slice != null &&\n        input.texData.slice.flatOffset > 0) {\n      shapeInfo.flatOffset = input.texData.slice.flatOffset;\n    }\n    return {name: program.variableNames[i], shapeInfo};\n  });\n  const inShapeInfos = inputInfos.map(x => x.shapeInfo);\n  const outShapeInfo: ShapeInfo = {\n    logicalShape: output.shape,\n    texShape: output.texData.texShape,\n    isUniform: false,\n    isPacked: output.texData.isPacked,\n    flatOffset: null\n  };\n  const source = shader_compiler.makeShader(\n      inputInfos, outShapeInfo, userCode, program.packedInputs);\n\n  const webGLProgram = gpgpu.createProgram(source);\n\n  // Add special uniforms (NAN, INFINITY)\n  let infLoc: WebGLUniformLocation = null;\n  const nanLoc = gpgpu.getUniformLocation(webGLProgram, 'NAN', false);\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    infLoc = gpgpu.getUniformLocation(webGLProgram, 'INFINITY', false);\n  }\n\n  // Add user-defined uniforms\n  const uniformLocations: {[name: string]: WebGLUniformLocation} = {};\n  for (let i = 0; i < program.variableNames.length; i++) {\n    const varName = program.variableNames[i];\n    const shouldThrow = false;\n    uniformLocations[varName] =\n        gpgpu.getUniformLocation(webGLProgram, varName, shouldThrow);\n    uniformLocations[`offset${varName}`] =\n        gpgpu.getUniformLocation(webGLProgram, `offset${varName}`, shouldThrow);\n  }\n\n  return {\n    program,\n    source,\n    webGLProgram,\n    uniformLocations,\n    inShapeInfos,\n    outShapeInfo,\n    infLoc,\n    nanLoc,\n  };\n}\n\nfunction validateBinaryAndProgram(\n    shapeInfos: ShapeInfo[], inputs: TensorData[]) {\n  if (shapeInfos.length !== inputs.length) {\n    throw Error(\n        `Binary was compiled with ${shapeInfos.length} inputs, but ` +\n        `was executed with ${inputs.length} inputs`);\n  }\n\n  shapeInfos.forEach((s, i) => {\n    const shapeA = s.logicalShape;\n    const input = inputs[i];\n    const shapeB = input.shape;\n\n    if (!util.arraysEqual(shapeA, shapeB)) {\n      throw Error(\n          `Binary was compiled with different shapes than ` +\n          `the current args. Shapes ${shapeA} and ${shapeB} must match`);\n    }\n    // The input is uploaded as uniform.\n    if (s.isUniform && input.isUniform) {\n      return;\n    }\n\n    const texShapeA = s.texShape;\n    const texShapeB = input.isUniform ? null : input.texData.texShape;\n    if (!util.arraysEqual(texShapeA, texShapeB)) {\n      throw Error(\n          `Binary was compiled with different texture shapes than the` +\n          ` current args. Shape ${texShapeA} and ${texShapeB} must match`);\n    }\n  });\n}\n\nexport function runProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, binary: GPGPUBinary, inputs: TensorData[],\n    output: TensorData,\n    customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) =>\n        void): void {\n  validateBinaryAndProgram(binary.inShapeInfos, inputs);\n  validateBinaryAndProgram([binary.outShapeInfo], [output]);\n\n  const outTex = output.texData.texture;\n  const outTexShape = output.texData.texShape;\n  if (output.texData.isPacked) {\n    gpgpu.setOutputPackedMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  } else {\n    gpgpu.setOutputMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  }\n  gpgpu.setProgram(binary.webGLProgram);\n\n  // Set special uniforms (NAN, INFINITY)\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    if (binary.infLoc !== null) {\n      gpgpu.gl.uniform1f(binary.infLoc, Infinity);\n    }\n  }\n  if (binary.nanLoc !== null) {\n    gpgpu.gl.uniform1f(binary.nanLoc, NaN);\n  }\n\n  // Set user-defined inputs\n  inputs.forEach((input, i) => {\n    const varName = binary.program.variableNames[i];\n    const varLoc = binary.uniformLocations[varName];\n    const varOffsetLoc = binary.uniformLocations[`offset${varName}`];\n\n    if (varLoc == null) {\n      // The compiler inferred that this variable is not used in this shader.\n      return;\n    }\n\n    if (input.isUniform) {\n      // Upload the values of the tensor as uniform.\n      if (util.sizeFromShape(input.shape) < 2) {\n        gpgpu.gl.uniform1f(varLoc, input.uniformValues[0]);\n      } else {\n        let vals = input.uniformValues;\n        if (!(vals instanceof Float32Array)) {\n          vals = new Float32Array(vals);\n        }\n        gpgpu.gl.uniform1fv(varLoc, vals);\n      }\n      return;\n    }\n\n    // If the input was sliced, upload the flat offset index.\n    if (input.texData.slice != null && varOffsetLoc != null) {\n      gpgpu.gl.uniform1i(varOffsetLoc, input.texData.slice.flatOffset);\n    }\n\n    gpgpu.setInputMatrixTexture(input.texData.texture, varLoc, i);\n  });\n\n  if (customSetup != null) {\n    customSetup(gpgpu, binary.webGLProgram);\n  }\n  gpgpu.executeProgram();\n}\n\nexport function makeShaderKey(\n    program: GPGPUProgram, inputs: TensorData[], output: TensorData): string {\n  let keyInputs = '';\n  inputs.concat(output).forEach(x => {\n    const hasOffset = x.texData != null && x.texData.slice != null &&\n        x.texData.slice.flatOffset > 0;\n    const texShape = x.isUniform ? 'uniform' : x.texData.texShape;\n    keyInputs += `${x.shape}_${texShape}_${hasOffset}`;\n  });\n  const keyUserCode = program.userCode;\n  let key = program.constructor.name;\n  // Fast string concat. See https://jsperf.com/string-concatenation/14.\n  key += '_' + keyInputs + '_' + keyUserCode;\n  return key;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Im2ColPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      outputShape: number[], inputShape: number[],\n      convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = outputShape;\n\n    const {\n      filterWidth,\n      inChannels,\n      strideWidth,\n      strideHeight,\n      padInfo,\n      outWidth,\n      dilationWidth,\n      dilationHeight,\n      dataFormat\n    } = convInfo;\n    const {left, top} = padInfo;\n    const itemsPerBlockRow = inChannels * filterWidth;\n    const glsl = getGlslDifferences();\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const rowDim = isChannelsLast ? 0 : 1;\n    const colDim = isChannelsLast ? 1 : 2;\n\n    let unrolled = ``;\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        unrolled += `\n          blockIndex = rc.y + ${col};\n          pos = rc.x + ${row};\n\n          if(blockIndex < ${outputShape[1]} && pos < ${outputShape[0]}) {\n            offsetY = int(blockIndex / (${outWidth})) * ${strideHeight} - ${\n            top};\n            d0 = offsetY + ${dilationHeight} * (pos / ${itemsPerBlockRow});\n\n            if(d0 < ${inputShape[rowDim]} && d0 >= 0) {\n\n              offsetX = int(mod(float(blockIndex), ${outWidth}.) * ${\n            strideWidth}. - ${left}.);\n              d1 = offsetX + ${dilationWidth} * (int(mod(float(pos), ${\n            itemsPerBlockRow}.) / ${inChannels}.));\n\n              if(d1 < ${inputShape[colDim]} && d1 >= 0) {\n\n                ch = int(mod(float(pos), ${inChannels}.));\n\n                if (${isChannelsLast}) {\n                  innerDims = vec2(d1, ch);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(d0, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                } else {\n                  innerDims = vec2(d0, d1);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(ch, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                }\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      void main() {\n        ivec2 rc = getOutputCoords();\n\n        vec4 result = vec4(0);\n\n        int blockIndex, pos, offsetY, d0, offsetX, d1, ch;\n        vec2 innerDims;\n\n        ${unrolled}\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n        int d = coords[3];\n        float x = getX(b, r, c, d);\n        float sum = 0.0;\n        for (int j = -${rad}; j <= ${rad}; j++) {\n          int idx = d + j;\n          if (idx >= 0 && idx <=  ${maxD}) {\n            float z = getX(b, r, c, idx);\n            sum += z * z;\n          }\n        }\n        float val = x * ${powOperator};\n        setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNGradProgram implements GPGPUProgram {\n  variableNames = ['inputImage', 'outputImage', 'dy'];\n  outputShape: number[] = [];\n  userCode: string;\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n  depth: number;\n\n  constructor(\n      inputShape: number[], depthRadius: number, bias: number, alpha: number,\n      beta: number) {\n    this.outputShape = inputShape;\n    this.depth = inputShape[3];\n    this.depthRadius = depthRadius;\n    this.bias = bias;\n    this.alpha = alpha;\n    this.beta = beta;\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n\n        float result = 0.0;\n        for (int d = 0; d < ${this.depth}; ++d) {\n          int depthBegin = int(max(0.0, float(d - ${depthRadius})));\n          int depthEnd = int(min(float(${this.depth}),\n              float(d + ${depthRadius} + 1)));\n\n          const int MIN_DEPTH_BEGIN = 0;\n          const int MAX_DEPTH_END = ${this.depth};\n\n          float norm = 0.0;\n          for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd) {\n              norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);\n            }\n            else {\n              break;\n            }\n          }\n\n          norm = float(${alpha}) * norm + float(${bias});\n\n          for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd){\n              float dyi = -2.0 * float(${alpha})\n                * float(${beta})\n                * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)\n                / norm;\n              if (k == d) {\n                dyi += pow(norm, -1.0 * ${beta});\n              }\n              if (k == coords[3]) {\n                dyi *= getDy(b, r, c, d);\n                result += dyi;\n              }\n            }\n            else {\n              break;\n            }\n          }\n      }\n      setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords.x;\n        int r = coords.y;\n        int c = coords.z;\n        int d = coords.w;\n\n        bool hasNextCol = d < ${this.outputShape[3]};\n        bool hasNextRow = c < ${this.outputShape[2]};\n\n        vec4 sum = vec4(0.);\n        vec4 xFragAtOutputCoords = getX(b, r, c, d);\n\n        vec4 xAtOutputCoords = vec4(\n          getChannel(xFragAtOutputCoords, vec2(c, d)),\n          hasNextCol ?\n            getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,\n          hasNextRow ?\n            getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0\n        );\n\n        int firstChannel = d - ${rad};\n        vec2 cache = vec2(0.);\n        if(firstChannel >= 0){\n          vec4 firstChannelFrag = getX(b, r, c, firstChannel);\n          cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));\n            if(hasNextRow){\n              cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));\n            }\n        }\n\n        ivec2 depth = ivec2(d, d + 1);\n        for (int j = - ${rad}; j <= ${rad}; j++) {\n          ivec2 idx = depth + j;\n          bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));\n          bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${maxD}));\n\n          bool depthInRange = aboveLowerBound.x && belowUpperBound.x;\n          bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;\n\n          if(depthInRange || depthPlusOneInRange){\n            vec4 z = vec4(0.);\n            vec4 xFragAtCurrentDepth;\n            z.xz = cache.xy;\n            if(depthPlusOneInRange && hasNextCol){\n              xFragAtCurrentDepth = idx.y != d ?\n                getX(b, r, c, idx.y) : xFragAtOutputCoords;\n              z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));\n              if(hasNextRow){\n                z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));\n              }\n            }\n            cache.xy = z.yw;\n            sum += z * z;\n          }\n        }\n        vec4 result = xAtOutputCoords * ${powOperator};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MaxPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex = effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n          wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n            int maxPosValue = ${lastIndex} - int(getMaxPos(b, idyR, idyC, d));\n\n            // Get the current value, check it against the value from the\n            // position matrix.\n            int curPosValue = wR * ${effectiveFilterWidth} + wC;\n            float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n            dotProd += dyValue * mask;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class MaxPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex =\n        effectiveFilterDepth * effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n           wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n              int maxPosValue = ${lastIndex} -\n                  int(getMaxPos(batch, idyD, idyR, idyC, ch));\n\n              // Get the current value, check it against the value from the\n              // position matrix.\n              int curPosValue =\n                  wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                  wR * ${effectiveFilterWidth} + wC;\n              float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n              dotProd += dyValue * mask;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MatMulPackedProgram implements GPGPUProgram {\n  variableNames = ['matrixA', 'matrixB'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      aShape: [number, number, number], outputShape: [number, number, number],\n      transposeA = false, transposeB = false, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = outputShape;\n\n    const sharedDim = transposeA ? aShape[1] : aShape[2];\n    const sharedDimensionPacked = Math.ceil(sharedDim / 2);\n\n    const aSample = transposeA ? 'i * 2, rc.y' : 'rc.y, i * 2';\n    const bSample = transposeB ? 'rc.z, i * 2' : 'i * 2, rc.z';\n    const aSwizzle = transposeA ? ['a.xxyy', 'a.zzww'] : ['a.xxzz', 'a.yyww'];\n    const bSwizzle = transposeB ? ['b.xzxz', 'b.ywyw'] : ['b.xyxy', 'b.zwzw'];\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const float sharedDimension = ${sharedDimensionPacked}.0;\n\n      vec4 dot2x2ARowBCol(ivec3 rc) {\n        vec4 result = vec4(0);\n        for (int i = 0; i < ${sharedDimensionPacked}; i++) {\n          vec4 a = getMatrixA(rc.x, ${aSample});\n          vec4 b = getMatrixB(rc.x, ${bSample});\n\n          // These swizzled products need to be separately added.\n          // See: https://github.com/tensorflow/tfjs/issues/1735\n          result += (${aSwizzle[0]} * ${bSwizzle[0]});\n          result += (${aSwizzle[1]} * ${bSwizzle[1]});\n        }\n        return result;\n      }\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n        vec4 result = dot2x2ARowBCol(rc);\n\n        ${addBiasSnippet}\n\n        ${applyActivationSnippet}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MultinomialProgram implements GPGPUProgram {\n  variableNames = ['probs'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(batchSize: number, numOutcomes: number, numSamples: number) {\n    this.outputShape = [batchSize, numSamples];\n\n    this.userCode = `\n      uniform float seed;\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n\n        float r = random(seed);\n        float cdf = 0.0;\n\n        for (int i = 0; i < ${numOutcomes - 1}; i++) {\n          cdf += getProbs(batch, i);\n\n          if (r < cdf) {\n            setOutput(float(i));\n            return;\n          }\n        }\n\n        // If no other event happened, last event happened.\n        setOutput(float(${numOutcomes - 1}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(seed: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.seedLoc == null) {\n        this.seedLoc = gpgpu.getUniformLocation(webGLProgram, 'seed');\n      }\n      gpgpu.gl.uniform1f(this.seedLoc, seed);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class OneHotProgram implements GPGPUProgram {\n  variableNames = ['indices'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(\n      numIndices: number, depth: number, onValue: number, offValue: number) {\n    this.outputShape = [numIndices, depth];\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int index = round(getIndices(coords.x));\n        setOutput(mix(float(${offValue}), float(${onValue}),\n                      float(index == coords.y)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape:\n          number[]) {  // TODO(https://github.com/tensorflow/tfjs/issues/893):\n                       // Only input / output 3D tensors.\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    if (rank === 0) {\n      this.userCode = `\n        void main() {\n          setOutput(vec4(getA(), 0., 0., 0.));\n        }\n      `;\n    } else {\n      const channels = getChannels('rc', rank);\n      const dtype = getCoordsDataType(rank);\n      const outOfBoundsCondition =\n          getOutOfBoundsCondition(rank, outputShape, channels);\n      const setup = getSetup(\n          rank, outputShape[outputShape.length - 1],\n          outputShape[outputShape.length - 2], channels);\n      const output = getOutput(outputShape, channels);\n\n      this.userCode = `\n        void main() {\n          ${dtype} rc = getOutputCoords();\n\n          if(${outOfBoundsCondition}) {\n            setOutput(vec4(0));\n          } else {\n            ${setup}\n\n            setOutput(vec4(${output}));\n          }\n        }\n      `;\n    }\n  }\n}\n\nfunction getSourceCoordsArr(rank: number, dims: string[]): string[] {\n  const coords = [];\n\n  for (let row = 0; row <= 1; row++) {\n    for (let col = 0; col <= 1; col++) {\n      let coord = `${row === 0 ? 'r' : 'rp1'}, ${col === 0 ? 'c' : 'cp1'}`;\n\n      for (let d = 2; d < rank; d++) {\n        coord = `${dims[dims.length - 1 - d]},` + coord;\n      }\n\n      coords.push(coord);\n    }\n  }\n  return coords;\n}\n\nfunction getOutOfBoundsCondition(\n    rank: number, shape: number[], dims: string[]): string {\n  if (rank === 1) {\n    return `rc > ${shape[0]}`;\n  }\n\n  let cond = '';\n  for (let i = rank - 2; i < rank; i++) {\n    cond += `${dims[i]} >= ${shape[i]}`;\n    if (i < rank - 1) {\n      cond += '||';\n    }\n  }\n\n  return cond;\n}\n\nfunction getSetup(\n    rank: number, cols: number, rows: number, dims: string[]): string {\n  if (rank === 1) {\n    return '';\n  }\n\n  const innerDims = dims.slice(-2);\n\n  return `\n    int r = ${innerDims[0]};\n    int c = ${innerDims[1]};\n    int rp1 = r + 1;\n    int cp1 = c + 1;\n\n    bool cEdge = cp1 >= ${cols};\n    bool rEdge = rp1 >= ${rows};\n  `;\n}\n\nfunction getOutput(shape: number[], dims: string[]): string {\n  const rank = shape.length;\n  const sourceCoords = getSourceCoordsArr(rank, dims);\n  if (rank === 1) {\n    return `getA(rc),\n            rc + 1 >= ${shape[0]} ? 0. : getA(rc + 1),\n            0, 0`;\n  }\n\n  return `getA(${sourceCoords[0]}),\n          cEdge ? 0. : getA(${sourceCoords[1]}),\n          rEdge ? 0. : getA(${sourceCoords[2]}),\n          rEdge || cEdge ? 0. : getA(${sourceCoords[3]})`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const type = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const unpackedCoords =\n        ['coords[0]', 'coords[1]', 'coords[2]', 'coords[3]'].slice(0, rank);\n\n    if (rank === 1) {\n      this.userCode = `\n        int start = ${start};\n        int end = ${end};\n\n        void main() {\n          int outC = getOutputCoords();\n          if (outC < start || outC >= end) {\n            setOutput(float(${constantValue}));\n          } else {\n            setOutput(getX(outC - start));\n          }\n        }\n      `;\n      return;\n    }\n    this.userCode = `\n      ${type} start = ${type}(${start});\n      ${type} end = ${type}(${end});\n\n      void main() {\n        ${type} outC = getOutputCoords();\n        if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {\n          setOutput(float(${constantValue}));\n        } else {\n          ${type} coords = outC - start;\n          setOutput(getX(${unpackedCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const dtype = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const coords = getChannels('rc', rank);\n    const source = getChannels('source', rank);\n    const cLimit = `${coords[rank - 1]} < ${this.outputShape[rank - 1]}`;\n    const innerDims =\n        rank === 1 ? 'source' : `vec2(${source.slice(-2).join()})`;\n\n    const componentSetup = [\n      `${dtype} rc = outputLoc;`, `${coords[rank - 1]} += 1;\n       if(${cLimit}) {\n      `,\n      rank === 1 ? '' : `}\n       rc = outputLoc;\n       ${coords[rank - 2]} += 1;\n       if(${coords[rank - 2]} < ${this.outputShape[rank - 2]}) {`,\n      rank === 1 ? '' : `  ${coords[rank - 1]} += 1;\n         if(${cLimit}) {`\n    ];\n\n    const paddingArea = rank === 1 ?\n        'rc < start || rc >= end' :\n        'any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))';\n    let mainLoop = '';\n    for (let i = 0, j = rank === 1 ? 2 : 4; i < j; i++) {\n      mainLoop += `\n        ${componentSetup[i]}\n        if (${paddingArea}) {\n          result[${i}] = float(${constantValue});\n        } else {\n          ${dtype} source = rc - start;\n          result[${i}] = getChannel(getX(${source.join()}), ${innerDims});\n        }\n      `;\n    }\n    mainLoop += (rank === 1 ? `} ` : `}}`);\n\n    this.userCode = `\n      const ${dtype} start = ${dtype}(${start});\n      const ${dtype} end = ${dtype}(${end});\n\n      void main() {\n        ${dtype} outputLoc = getOutputCoords();\n        vec4 result = vec4(0.);\n        ${mainLoop}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Pool2DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n    const batchFlattenPositionStr = `((batch  * ${convInfo.inHeight} + xR) * ${\n        convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n    const flattenPositionStr =\n        `(xR * ${convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n        const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int batch = coords[0];\n          int d = coords[3];\n\n          ivec2 xRCCorner = coords.yz * strides - pads;\n          int xRCorner = xRCCorner.x;\n          int xCCorner = xRCCorner.y;\n\n          // max/min x(?, ?, d) to get y(yR, yC, d).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n          float avgValue = 0.0;\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              int xC = xCCorner + wC;\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float value = getX(batch, xR, xC, d);\n\n              // If a min / max value has already been found, use it. If not,\n              // use the current value.\n              float currMinMaxValue = mix(\n                  value, minMaxValue, minMaxValueFound);\n              if (value ${compareOp} currMinMaxValue) {\n                minMaxValue = value;\n                minMaxValueFound = 1.0;\n                minMaxPosition = ${\n          flattenPositions ? (includeBatchInIndex ? batchFlattenPositionStr :\n                                                    flattenPositionStr) :\n                             `wR * ${effectiveFilterWidth} + wC`};\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xR, int xC, int d) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xR, xC, d);\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d = coords[3];\n\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // max/min x(?, ?, d) to get y(yR, yC, d).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          int xR = xRCorner + wR;\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              getValue(batch, xR, xC + 3 * ${dilationWidth}, d)\n            );\n\n            ${updateSnippet}\n          }\n\n          int xC = xCCorner + ${filterWidthNearestVec4};\n          if (${filterWidthVec4Remainder === 1}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              initializationValue,\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 2}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 3}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              initializationValue\n            );\n\n            ${updateSnippet}\n          }\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n\nexport class Pool3DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv3DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec3 strides =\n            ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n        const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n        void main() {\n          ivec5 coords = getOutputCoords();\n          int batch = coords.x;\n          int ch = coords.u;\n\n          ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n          int xDCorner = xCorner.x;\n          int xRCorner = xCorner.y;\n          int xCCorner = xCorner.z;\n\n          // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n\n          for (int wD = 0; wD < ${effectiveFilterDepth};\n              wD += ${dilationDepth}) {\n            int xD = xDCorner + wD;\n\n            if (xD < 0 || xD >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int wR = 0; wR < ${effectiveFilterHeight};\n                wR += ${dilationHeight}) {\n              int xR = xRCorner + wR;\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int wC = 0; wC < ${effectiveFilterWidth};\n                  wC += ${dilationWidth}) {\n                int xC = xCCorner + wC;\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float value = getX(batch, xD, xR, xC, ch);\n\n                // If a min / max value has already been found, use it. If not,\n                // use the current value.\n                float currMinMaxValue = mix(\n                    value, minMaxValue, minMaxValueFound);\n                if (value ${compareOp} currMinMaxValue) {\n                  minMaxValue = value;\n                  minMaxValueFound = 1.0;\n                  minMaxPosition = ${\n          flattenPositions ?\n              (includeBatchInIndex ?\n                   `(((batch * ${convInfo.inDepth} + xD) * ${\n                       convInfo.inHeight} + xR) * ${convInfo.inWidth} + xC) * ${\n                       convInfo.inChannels} + ch` :\n                   `((xD * ${convInfo.inHeight} + xR) * ${\n                       convInfo.inWidth} + xC) * ${convInfo.inChannels} + ch`) :\n              `wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                      wR * ${effectiveFilterWidth} + wC`};\n                }\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec3 strides =\n        ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xD, int xR, int xC, int ch) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xD, xR, xC, ch);\n      }\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xDCorner = xCorner.x;\n        int xRCorner = xCorner.y;\n        int xCCorner = xCorner.z;\n\n        // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          int xD = xDCorner + wD;\n\n          if (xD < 0 || xD >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 3 * ${dilationWidth}, ch)\n              );\n\n              ${updateSnippet}\n            }\n\n            int xC = xCCorner + ${filterWidthNearestVec4};\n            if (${filterWidthVec4Remainder === 1}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                initializationValue,\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 2}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 3}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                initializationValue\n              );\n\n              ${updateSnippet}\n            }\n          }\n          setOutput(${returnValue});\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ReduceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo,\n      reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod') {\n    const {windowSize, batchSize, inSize, outSize} = reduceInfo;\n    this.outputShape = [batchSize, outSize];\n\n    let initializationValue = '0.0';\n    let compareOp = ``;\n\n    if (reduceType === 'prod') {\n      initializationValue = '1.0';\n    } else if (reduceType === 'min') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '1.0 / 1e-20';\n      compareOp = `min`;\n    } else if (reduceType === 'max') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n      compareOp = `max`;\n    }\n\n    let returnValue = `${reduceType}(${reduceType}(${reduceType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n\n    if (reduceType === 'sum') {\n      returnValue = `sumValue`;\n    } else if (reduceType === 'prod') {\n      returnValue = `prodValue`;\n    } else if (reduceType === 'all') {\n      returnValue = `allValue`;\n    } else if (reduceType === 'any') {\n      returnValue = `anyValue`;\n    }\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    let updateSnippet = `\n      if (${reduceType === 'sum'}) {\n        sumValue += dot(values, ones);\n      } else if (${reduceType === 'prod'}) {\n        vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);\n        prodValue *= tmp[0] * tmp[1];\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    let vecType = `vec4`;\n\n    if (reduceType === 'all') {\n      initializationValue = '1.0';\n      updateSnippet = `\n        bool reducedAllValue = all(values);\n        float floatedReducedAllValue = float(reducedAllValue);\n        allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    } else if (reduceType === 'any') {\n      initializationValue = '0.0';\n      updateSnippet = `\n        bool reducedAnyValue = any(values);\n        float floatedReducedAnyValue = float(reducedAnyValue);\n        anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    }\n\n    let checkOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float getValue(int batch, int inIdx) {\n        ${checkOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        vec4 minMaxValue = vec4(${initializationValue});\n        float prodValue = 1.0;\n        float sumValue = 0.0;\n        float allValue = 1.0;\n        float anyValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class ReshapePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: [number, number, number], inputShape: [\n    number, number, number\n  ]) {\n    this.outputShape = outputShape;\n\n    let mainLoop = ``;\n    for (let i = 0; i < 4; i++) {\n      let thisRC = `thisRC = rc;`;\n      if (i % 2 === 1) {\n        thisRC += `thisRC.z += 1;`;\n      }\n      if (i > 1) {\n        thisRC += `thisRC.y += 1;`;\n      }\n\n      mainLoop += `\n        ${thisRC}\n        ${i > 0 ? `if(thisRC.y < rows && thisRC.z < cols){` : ''}\n          int flatIndex = getFlatIndex(thisRC);\n\n          ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);\n          vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));\n\n          result[${i}] =\n            getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);\n        ${i > 0 ? '}' : ''}\n      `;\n    }\n\n    this.userCode = `\n      ${getReshapedInputCoords(inputShape)}\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n\n        vec4 result = vec4(0.);\n\n        ivec3 thisRC;\n        int rows = ${outputShape[1]};\n        int cols = ${outputShape[2]};\n\n        ${mainLoop}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nfunction getReshapedInputCoords(shape: [number, number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 inputCoordsFromReshapedOutCoords(int index) {\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(startRLerp - float(winHeight / 2));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(startCLerp - float(winWidth / 2));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float dxR = float(dyR) * heightScale;\n            int topDxRIndex = int(floor(dxR));\n            int bottomDxRIndex = int(min(ceil(dxR), ${xHeight - 1}.0));\n            float dxRLerp = dxR - float(topDxRIndex);\n            float inverseDxRLerp = 1.0 - dxRLerp;\n\n            float dxC = float(dyC) * widthScale;\n            int leftDxCIndex = int(floor(dxC));\n            int rightDxCIndex = int(min(ceil(dxC), ${xWidth - 1}.0));\n            float dxCLerp = dxC - float(leftDxCIndex);\n            float inverseDxCLerp = 1.0 - dxCLerp;\n\n            if (r == topDxRIndex && c == leftDxCIndex) {\n              // topLeft\n              accumulator +=\n                getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;\n            }\n\n            if (r == topDxRIndex && c == rightDxCIndex) {\n              // topRight\n              accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == leftDxCIndex) {\n              // bottomLeft\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == rightDxCIndex) {\n              // bottomRight\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec2 sourceFloorRC = ivec2(sourceFracIndexRC);\n        ivec2 sourceCeilRC = ivec2(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);\n        float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);\n        float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);\n        float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);\n\n        vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);\n\n        float top = topLeft + (topRight - topLeft) * fracRC.y;\n        float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;\n        float newValue = top + (bottom - top) * fracRC.x;\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec3 effectiveInputOverOutputRatioRC = vec3(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec3 inputShapeRC = vec3(${oldHeight}.0, ${oldWidth}.0,\n                                     ${oldWidth}.0);\n\n      float getAValue(int b, int r, int c, int d) {\n        return getChannel(getA(b, r, c, d), vec2(c, d));\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        // Calculate values for next column in yRC.z.\n        ivec3 yRC = coords.yzz + ivec3(0, 0, 1);\n\n        // Fractional source index.\n        vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec3 sourceFloorRC = ivec3(sourceFracIndexRC);\n        ivec3 sourceCeilRC = ivec3(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        // Should we calculate next column and row elements in 2x2 packed cell.\n        bool hasNextCol = d < ${depth - 1};\n        bool hasNextRow = coords.z < ${newWidth - 1};\n\n        // In parallel, construct four corners for all four components in\n        // packed 2x2 cell.\n        vec4 topLeft = vec4(\n          getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 bottomLeft = vec4(\n          getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 topRight = vec4(\n          getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec4 bottomRight = vec4(\n          getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);\n\n        vec4 top = mix(topLeft, topRight, fracRC.yyzz);\n        vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);\n        vec4 newValue = mix(top, bottom, fracRC.x);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeigborBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(floor(startRLerp - float(winHeight / 2)));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(floor(startCLerp - float(winWidth / 2)));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float sourceFracRow =\n              float(${effectiveXSize[0]}) *\n                (float(dyR) / float(${effectiveYSize[0]}));\n\n            float sourceFracCol =\n                float(${effectiveXSize[1]}) *\n                  (float(dyC) / float(${effectiveYSize[1]}));\n\n            int sourceNearestRow = int(min(\n                float(int(${xHeight}) - 1),\n                ${alignCorners} ? float(round(sourceFracRow)) :\n                                  float(floor(sourceFracRow))));\n\n            int sourceNearestCol = int(min(\n                float(int(${xWidth}) - 1),\n                ${alignCorners} ? float(round(sourceFracCol)) :\n                                  float(floor(sourceFracCol))));\n\n            if (r == sourceNearestRow && c == sourceNearestCol) {\n              accumulator += getDy(b, dyR, dyC, d);\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeighborProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    // When align corners is false, we rounds the value with floor.\n    const roundBase = alignCorners ? '0.5' : '0.0';\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the coordinators of nearest neighbor point.\n        ivec2 sourceNearestRC = ivec2(\n          min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${roundBase})));\n\n        float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReverseProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n\n    if (rank === 1) {\n      this.userCode = `\n        void main() {\n          int coord = getOutputCoords();\n          setOutput(getX(${xShape[0]} - coord - 1));\n        }\n      `;\n      return;\n    }\n    const getInCoord = (i: number) => {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - coords[${i}] - 1`;\n      }\n      return `coords[${i}]`;\n    };\n    const inCoords = xShape.map((_, i) => getInCoord(i)).join(',');\n    const type = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${type} coords = getOutputCoords();\n        setOutput(getX(${inCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReversePackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n    const channels = getChannels('rc', rank);\n    const nextColumn =\n        `${channels[rank - 1]} + 1 < ${this.outputShape[rank - 1]}`;\n    const nextRow = `${channels[rank - 2]} + 1 < ${this.outputShape[rank - 2]}`;\n    const type = getCoordsDataType(rank);\n    if (rank === 1) {\n      this.userCode = `\n        void main(){\n          int rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = getChannel(getX(${xShape[0]} - rc - 1),\n            ${xShape[0]} - rc - 1);\n          if(${nextColumn}){\n              result.g = getChannel(getX(${xShape[0]} - (rc  + 1) - 1),\n                ${xShape[0]} - (rc  + 1) - 1);\n          }\n          setOutput(result);\n        }\n      `;\n    } else {\n      this.userCode = `\n        void main() {\n          ${type} rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = ${getR(channels.slice())};\n          if(${nextColumn}){\n            result.g = ${getG(channels.slice())};\n          }\n          if(${nextRow}) {\n            result.b = ${getB(channels.slice())};\n            if(${nextColumn}) {\n              result.a = ${getA(channels.slice())};\n            }\n          }\n          setOutput(result);\n        }\n    `;\n    }\n\n    function getR(channels: string[]): string {\n      return getChannel(channels);\n    }\n\n    function getG(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getB(channels: string[]): string {\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getA(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getChannel(channels: string[]): string {\n      const inCoordsArray = xShape.map((_, i) => getInCoord(i, channels));\n      const inCoords = inCoordsArray.join(',');\n      const innerDims = inCoordsArray.slice(-2).join(',');\n      return `getChannel(getX(${inCoords}), vec2(${innerDims}))`;\n    }\n\n    function getInCoord(i: number, channels1: string[]): string {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - ${channels1[i]} - 1`;\n      } else {\n        return `${channels1[i]}`;\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ScatterProgram implements GPGPUProgram {\n  variableNames = ['updates', 'indices', 'defaultValue'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      updateSize: number, sliceDim: number, indicesRank: number,\n      updatesRank: number, strides: number[], shape: number[],\n      summingDupeIndex = true) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    let indicesString = '';\n    if (indicesRank === 1) {\n      indicesString = 'i';\n    } else if (indicesRank === 2) {\n      indicesString = 'i, j';\n    }\n    const indicesSnippet = `getIndices(${indicesString})`;\n\n    let updatesString = '';\n    if (updatesRank === 1) {\n      updatesString = 'i';\n    } else if (updatesRank === 2) {\n      updatesString = 'i, coords[1]';\n    }\n    const updatesSnippet = `getUpdates(${updatesString})`;\n\n    const strideString = sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${strides});\n\n        void main() {\n          ${dtype} coords = getOutputCoords();\n          float sum = 0.0;\n          bool found = false;\n          for (int i = 0; i < ${updateSize}; i++) {\n            int flattenedIndex = 0;\n            for (int j = 0; j < ${sliceDim}; j++) {\n              int index = round(${indicesSnippet});\n              flattenedIndex += index * ${strideString};\n            }\n            if (flattenedIndex == coords[0]) {\n              sum += ${updatesSnippet};\n              found = true;\n            }\n          }\n          setOutput(mix(getDefaultValue(), sum, float(found)));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class SegmentOpProgram implements GPGPUProgram {\n  variableNames = ['x', 'segmentIds'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      segOpInfo: backend_util.segment_util.SegOpInfo,\n      segOpType: 'unsortedSegmentSum') {\n    const windowSize = segOpInfo.windowSize;\n    const batchSize = segOpInfo.batchSize;\n    const inSize = segOpInfo.inSize;\n    const numSegments = segOpInfo.numSegments;\n    const outSize = numSegments * Math.ceil(inSize / windowSize);\n    this.outputShape = [batchSize, outSize];\n\n    const initializationValue = '0.0';\n    const returnValue = `sumValue`;\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    const updateSnippet = `\n        sumValue += dot(values, segFilter);\n    `;\n\n    let checkValueOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkValueOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n\n    let checkSegmentIdOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkSegmentIdOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return -1.0;\n        }\n      `;\n    }\n\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n\n      float getValue(int batch, int inIdx) {\n        ${checkValueOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      float getSegmentIdAtIndex(int inIdx) {\n        ${checkSegmentIdOutOfBounds}\n        return getSegmentIds(inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = int(floor(float(outIdx) / float(\n          ${numSegments})) * float(${windowSize}));\n        int currentSeg = int(mod(float(outIdx), float(${numSegments})));\n\n        float sumValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          int inIdxSeg = int(getSegmentIdAtIndex(inIdx));\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            0,\n            0,\n            0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n              0,\n              0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            0\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SelectProgram implements GPGPUProgram {\n  variableNames = ['c', 'a', 'b'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(cRank: number, shape: number[], rank: number) {\n    this.outputShape = shape;\n\n    let cCoords;\n    let abCoords;\n    if (rank > 4) {\n      throw Error(`Where for rank ${rank} is not yet supported`);\n    }\n\n    if (rank === 1) {\n      abCoords = `resRC`;\n      cCoords = `resRC`;\n    } else {\n      const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n      const cCoordVars = [];\n      const abCoordVars = [];\n      for (let i = 0; i < shape.length; i++) {\n        abCoordVars.push(`${currentCoords[i]}`);\n        if (i < cRank) {\n          cCoordVars.push(`${currentCoords[i]}`);\n        }\n      }\n      cCoords = cCoordVars.join();\n      abCoords = abCoordVars.join();\n    }\n\n    const dtype = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        float cVal = getC(${cCoords});\n        if (cVal >= 1.0) {\n          setOutput(getA(${abCoords}));\n        } else {\n          setOutput(getB(${abCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SliceProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const uniformPart = `uniform int start[${this.rank}];`;\n    const sourceCoords = getCoords(this.rank);\n\n    let body: string;\n    const coordSum = destSize.map((_, i) => {\n      return `sourceLoc.${coords[i]} = start[${i}] + coords.${coords[i]};`;\n    });\n    body = `\n        ${dtype} sourceLoc;\n        ${dtype} coords = getOutputCoords();\n        ${coordSum.join('\\n')}\n      `;\n    this.userCode = `\n      ${uniformPart}\n      void main() {\n        ${body}\n        setOutput(getSource(${sourceCoords}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n\nconst coords = ['x', 'y', 'z', 'w', 'u', 'v'];\n\nfunction getCoords(rank: number): string {\n  if (rank === 1) {\n    return 'sourceLoc';\n  } else if (rank <= 6) {\n    return coords.slice(0, rank).map(x => 'sourceLoc.' + x).join(',');\n  } else {\n    throw Error(`Slicing for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SlicePackedProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const coords = getChannels('coords', this.rank);\n    const sourceLoc = getChannels('sourceLoc', this.rank);\n\n    const innerDims =\n        this.rank === 1 ? 'sourceLoc' : `vec2(${sourceLoc.slice(-2).join()})`;\n    const getChannel =\n        `getChannel(getSource(${sourceLoc.join()}), ${innerDims})`;\n    const upperRow = `\n      result.x = ${getChannel};\n      if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n        ++${sourceLoc[this.rank - 1]};\n        result.y = ${getChannel};\n        --${sourceLoc[this.rank - 1]};\n      }\n    `;\n    const lowerRow = this.rank === 1 ? '' : `\n      --${coords[this.rank - 1]};\n      if (++${coords[this.rank - 2]} < ${destSize[this.rank - 2]}) {\n        ++${sourceLoc[this.rank - 2]};\n        result.z = ${getChannel};\n        if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n          ++${sourceLoc[this.rank - 1]};\n          result.w = ${getChannel};\n        }\n      }\n    `;\n\n    const sourceLocSetup = this.rank <= 4 ?\n        `sourceLoc = coords +\n            ${dtype}(${destSize.map((_, i) => `start[${i}]`).join()});` :\n        destSize.map((_, i) => `${sourceLoc[i]} = ${coords[i]} + start[${i}];`)\n            .join('\\n');\n    this.userCode = `\n      uniform int start[${this.rank}];\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        ${dtype} sourceLoc;\n        ${sourceLocSetup}\n        vec4 result = vec4(0.);\n        ${upperRow}\n        ${lowerRow}\n        setOutput(result);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class StridedSliceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(begin: number[], strides: number[], size: number[]) {\n    this.outputShape = size;\n    const rank = size.length;\n    const inputDtype = getCoordsDataType(size.length);\n    const dtype = getCoordsDataType(size.length);\n\n    let newCoords = '';\n    if (rank === 1) {\n      newCoords = 'coords * strides + begin';\n    } else {\n      let outputAxis = 0;\n      newCoords =\n          size.map((_, i) => {\n                outputAxis++;\n                return size.length === 1 ?\n                    `coords * strides[${i}] + begin[${i}]` :\n                    `coords[${outputAxis - 1}] * strides[${i}] + begin[${i}]`;\n              })\n              .join(',');\n    }\n\n    this.userCode = `\n      ${inputDtype} begin = ${inputDtype}(${begin});\n      ${inputDtype} strides = ${inputDtype}(${strides});\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        setOutput(getX(${newCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {getInternalFormatForFloat16MatrixTexture, getInternalFormatForFloat16PackedMatrixTexture, getInternalFormatForFloat32MatrixTexture, getInternalFormatForPackedMatrixTexture, getInternalFormatForUnsignedBytesMatrixTexture} from './gpgpu_util';\nimport {getPackedMatrixTextureShapeWidthHeight, getUnpackedMatrixTextureShapeWidthHeight, PhysicalTextureType, TextureConfig, TextureUsage} from './tex_util';\n\nexport class TextureManager {\n  private numUsedTextures = 0;\n  private numFreeTextures = 0;\n  private _numBytesAllocated = 0;\n  private _numBytesFree = 0;  // How many bytes that have been allocated\n                              // are available for reuse.\n  private freeTextures: {[shape: string]: WebGLTexture[]} = {};\n  private logEnabled = false;\n  private usedTextures: {[shape: string]: WebGLTexture[]} = {};\n\n  constructor(private gpgpu: GPGPUContext) {}\n\n  acquireTexture(\n      shapeRC: [number, number], usage: TextureUsage,\n      isPacked: boolean): WebGLTexture {\n    const physicalTexType = getPhysicalFromLogicalTextureType(usage, isPacked);\n\n    const shapeKey = getKeyFromTextureShape(shapeRC, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n    if (!(shapeKey in this.usedTextures)) {\n      this.usedTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shapeRC, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n\n    if (this.freeTextures[shapeKey].length > 0) {\n      this.numFreeTextures--;\n      this.numUsedTextures++;\n      this._numBytesFree -= texBytes;\n      this.log();\n      const newTexture = this.freeTextures[shapeKey].shift();\n      this.usedTextures[shapeKey].push(newTexture);\n      return newTexture;\n    }\n\n    let newTexture: WebGLTexture;\n    if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT32) {\n      newTexture = this.gpgpu.createPackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16PackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT32) {\n      newTexture =\n          this.gpgpu.createFloat32MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (\n        physicalTexType === PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE) {\n      newTexture =\n          this.gpgpu.createUnsignedBytesMatrixTexture(shapeRC[0], shapeRC[1]);\n    }\n    this.usedTextures[shapeKey].push(newTexture);\n\n    this.numUsedTextures++;\n    this._numBytesAllocated += texBytes;\n    this.log();\n\n    return newTexture;\n  }\n\n  releaseTexture(\n      texture: WebGLTexture, shape: [number, number],\n      logicalTexType: TextureUsage, isPacked: boolean): void {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    const physicalTexType =\n        getPhysicalFromLogicalTextureType(logicalTexType, isPacked);\n    const shapeKey = getKeyFromTextureShape(shape, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shape, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n    const deleteTexThreshold = env().get('WEBGL_DELETE_TEXTURE_THRESHOLD');\n    if (deleteTexThreshold !== -1 &&\n        this._numBytesAllocated > deleteTexThreshold) {\n      this.gpgpu.deleteMatrixTexture(texture);\n      this._numBytesAllocated -= texBytes;\n    } else {\n      this.freeTextures[shapeKey].push(texture);\n      this.numFreeTextures++;\n      this._numBytesFree += texBytes;\n    }\n\n    this.numUsedTextures--;\n\n    const texList = this.usedTextures[shapeKey];\n    const texIndex = texList.indexOf(texture);\n    if (texIndex < 0) {\n      throw new Error(\n          'Cannot release a texture that was never provided by this ' +\n          'texture manager');\n    }\n    texList.splice(texIndex, 1);\n    this.log();\n  }\n\n  private log() {\n    if (!this.logEnabled) {\n      return;\n    }\n    const total = this.numFreeTextures + this.numUsedTextures;\n    console.log(\n        'Free/Used', `${this.numFreeTextures} / ${this.numUsedTextures}`,\n        `(${total})`);\n    const freeRatio = this._numBytesFree / this._numBytesAllocated;\n    console.log(`Bytes allocated: ${this._numBytesAllocated}`);\n    console.log(`Bytes unused: ${this._numBytesFree} (${\n        Math.round(100 * freeRatio)}%)`);\n  }\n\n  get numBytesAllocated(): number {\n    return this._numBytesAllocated;\n  }\n\n  get numBytesFree(): number {\n    return this._numBytesFree;\n  }\n\n  getNumUsedTextures(): number {\n    return this.numUsedTextures;\n  }\n\n  getNumFreeTextures(): number {\n    return this.numFreeTextures;\n  }\n\n  dispose() {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    for (const texShape in this.freeTextures) {\n      this.freeTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    for (const texShape in this.usedTextures) {\n      this.usedTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    this.freeTextures = null;\n    this.usedTextures = null;\n    this.numUsedTextures = 0;\n    this.numFreeTextures = 0;\n    this._numBytesAllocated = 0;\n    this._numBytesFree = 0;\n  }\n}\n\nfunction numBytesForInternalFormat(\n    gl: WebGLRenderingContext, internalFormat: number): number {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n  if (internalFormat === glany.R32F) {\n    return 4;\n  } else if (internalFormat === glany.R16F) {\n    return 2;\n  } else if (internalFormat === glany.RGBA32F) {\n    return 16;\n  } else if (internalFormat === gl.RGBA) {\n    return 16;\n  } else if (internalFormat === glany.RGBA16F) {\n    return 8;\n  }\n  throw new Error(`Unknown internal format ${internalFormat}`);\n}\n\nexport function computeBytes(\n    shape: [number, number], physicalTexType: PhysicalTextureType,\n    gl: WebGLRenderingContext, textureConfig: TextureConfig,\n    isPacked: boolean): number {\n  // It is not possible to infer packed status from the texture type because\n  // depending on the textureConfig, different  texture types may resolve to the\n  // same internal format (e.g. in WebGL1, the internal format for\n  // UNPACKED_FLOAT16 textures is gl.RGBA). Therefore we pass in `isPacked`\n  // explicitly.\n  const internalFormat =\n      internalFormatForPhysicalTexType(physicalTexType, textureConfig);\n\n  let numElements: number;\n  if (isPacked) {\n    const [packedWidth, packedHeight] =\n        getPackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = packedWidth * packedHeight;\n\n  } else {\n    const [width, height] =\n        getUnpackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = width * height;\n  }\n\n  const bytesPerElement = numBytesForInternalFormat(gl, internalFormat);\n  return numElements * bytesPerElement;\n}\n\nfunction internalFormatForPhysicalTexType(\n    physicalTexType: PhysicalTextureType,\n    textureConfig: TextureConfig): number {\n  switch (physicalTexType) {\n    case PhysicalTextureType.PACKED_2X2_FLOAT32:\n      return getInternalFormatForPackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_2X2_FLOAT16:\n      return getInternalFormatForFloat16PackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT32:\n      return getInternalFormatForFloat32MatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT16:\n      return getInternalFormatForFloat16MatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE:\n      return getInternalFormatForUnsignedBytesMatrixTexture(textureConfig);\n    default:\n      throw new Error(`Unknown physical texture type ${physicalTexType}`);\n  }\n}\n\nfunction getPhysicalTextureForRendering(isPacked: boolean):\n    PhysicalTextureType {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED')) {\n    if (isPacked) {\n      return PhysicalTextureType.PACKED_2X2_FLOAT32;\n    }\n    return PhysicalTextureType.UNPACKED_FLOAT32;\n  }\n\n  if (isPacked) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT16;\n  }\n  return PhysicalTextureType.UNPACKED_FLOAT16;\n}\n\nfunction getPhysicalFromLogicalTextureType(\n    logicalTexType: TextureUsage, isPacked: boolean): PhysicalTextureType {\n  if (logicalTexType === TextureUsage.UPLOAD) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT32;\n  } else if (logicalTexType === TextureUsage.RENDER || logicalTexType == null) {\n    return getPhysicalTextureForRendering(isPacked);\n  } else if (\n      logicalTexType === TextureUsage.DOWNLOAD ||\n      logicalTexType === TextureUsage.PIXELS) {\n    return PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE;\n  }\n  throw new Error(`Unknown logical texture type ${logicalTexType}`);\n}\n\nfunction getKeyFromTextureShape(\n    shapeRowsCol: [number, number], physicalTexType: PhysicalTextureType,\n    isPacked: boolean): string {\n  return `${shapeRowsCol[0]}_${shapeRowsCol[1]}_${physicalTexType}_${isPacked}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TileProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], reps: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[i] * reps[i];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[]): string {\n  const rank = aShape.length;\n  if (rank > 5) {\n    throw Error(`Tile for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `imod(resRC, ${aShape[0]})`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    sourceCoords.push(`imod(${currentCoords[i]}, ${aShape[i]})`);\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class UnaryOpProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      float unaryOperation(float x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        float x = getAAtOutCoords();\n        float y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n\nconst CHECK_NAN_SNIPPET = `if (isnan(x)) return x;`;\n\nexport const LINEAR = `return x;`;\n\nexport const ABS = `return abs(x);`;\n\nexport const RELU = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : x;\n`;\n\nexport const RELU6 = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : min(6.0, x);\n`;\n\nexport const ELU = `return (x >= 0.0) ? x : (exp(x) - 1.0);`;\n\nexport const SELU = `\n  // Stable and Attracting Fixed Point (0, 1) for Normalized Weights.\n  // see: https://arxiv.org/abs/1706.02515\n  float scaleAlpha = ${backend_util.SELU_SCALEALPHA};\n  float scale = ${backend_util.SELU_SCALE};\n  return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);\n`;\n\nexport function STEP(alpha = 0.0) {\n  return CHECK_NAN_SNIPPET + `\n    return x > 0.0 ? 1.0 : float(${alpha});\n  `;\n}\n\nexport const NEG = `return -x;`;\n\nexport const CEIL = `return ceil(x);`;\n\nexport const FLOOR = `return floor(x);`;\n\nexport const SIGN = `\n  if (isnan(x)) { return 0.0; }\n  return sign(x);\n`;\n\nexport const IS_NAN = `return float(isnan(x));`;\n\nexport const IS_INF = `return float(isinf(x));`;\n\nexport const IS_FINITE = `return float(!isnan(x) && !isinf(x));`;\n\nexport const ROUND = `\n  // OpenGL ES does not support round function.\n  // The algorithm is based on banker's rounding.\n  float base = floor(x);\n  if ((x - base) < 0.5) {\n    return floor(x);\n  } else if ((x - base) > 0.5) {\n    return ceil(x);\n  } else {\n    if (mod(base, 2.0) == 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n`;\n\nexport const EXP = `return exp(x);`;\n\nexport const EXPM1 = `return exp(x) - 1.0;`;\n\nexport const LOG = `if (x < 0.0) return NAN;\n  return log(x);`;\n\nexport const LOG1P = `return log(1.0 + x);`;\n\nexport const SQRT = `return sqrt(x);`;\n\nexport const RSQRT = `return inversesqrt(x);`;\n\nexport const SIGMOID = `return 1.0 / (1.0 + exp(-1.0 * x));`;\n\n/**\n * mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n *\n * epsilon is the difference between 1.0 and the next representable\n * float. For a single precision 32 bit float this should be 2^-23, see:\n * https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\n *\n * too_large = (x > -threshold) is value above which exp(x) may overflow\n * but softplus(x) == x is within machine epsilon\n *\n * too_small = (x < threshold) is value below which exp(x) may underflow,\n * but softplus(x) == exp(x) is within machine epsilon.\n */\nexport const SOFTPLUS = `\n  float epsilon = 1.1920928955078125e-7;\n  float threshold = log(epsilon) + 2.0;\n\n  bool too_large = x > -threshold;\n  bool too_small = x < threshold;\n\n  float result;\n  float exp_x = exp(x);\n\n  if (too_large){\n    result = x;\n  }\n  else if (too_small){\n    result = exp_x;\n  }\n  else{\n    result = log(exp_x + 1.0);\n  }\n  return result;\n`;\n\nexport const ASIN = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return asin(x);\n`;\n\nexport const ACOS = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return acos(x);\n`;\n\nexport const ATAN = CHECK_NAN_SNIPPET + `\n  return atan(x);\n`;\n\nexport const SINH = `\n  float e2x = exp(x);\n  return (e2x - 1.0 / e2x) / 2.0;\n`;\n\nexport const COSH = `\n  float e2x = exp(-x);\n  return (e2x + 1.0 / e2x) / 2.0;\n`;\n\nexport const TANH = `\n  float e2x = exp(-2.0 * abs(x));\n  return sign(x) * (1.0 - e2x) / (1.0 + e2x);\n`;\n\nexport const ASINH = CHECK_NAN_SNIPPET + `return log(x + sqrt(x * x + 1.0));`;\n\nexport const ACOSH = CHECK_NAN_SNIPPET + `\n  if (x < 1.0) return NAN;\n  return log(x + sqrt(x * x - 1.0));`;\n\nexport const ATANH = CHECK_NAN_SNIPPET + `\n  if ((x < -1.0) || (x > 1.0)) return NAN;\n  return (log(1.0 + x) - log(1.0 - x)) / 2.0;`;\n\nexport const ERF = `\n  // Error function is calculated approximately with elementary function.\n  // See \"Handbook of Mathematical Functions with Formulas,\n  // Graphs, and Mathematical Tables\", Abramowitz and Stegun.\n  float p = ${backend_util.ERF_P};\n  float a1 = ${backend_util.ERF_A1};\n  float a2 = ${backend_util.ERF_A2};\n  float a3 = ${backend_util.ERF_A3};\n  float a4 = ${backend_util.ERF_A4};\n  float a5 = ${backend_util.ERF_A5};\n\n  float sign = sign(x);\n  x = abs(x);\n  float t = 1.0 / (1.0 + p * x);\n  return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));\n`;\n\nexport const RECIPROCAL = `return 1.0 / x;`;\n\nexport const LOGICAL_NOT = `return float(!(x >= 1.0));`;\n\nexport const TO_INT = `return float(int(x));`;\n\nexport const CLONE = 'return x;';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const LINEAR = `return x;`;\n\nexport const LOG = `\n  vec4 result = log(x);\n  vec4 isNaN = vec4(lessThan(x, vec4(0.0)));\n  result.r = isNaN.r == 1.0 ? NAN : result.r;\n  result.g = isNaN.g == 1.0 ? NAN : result.g;\n  result.b = isNaN.b == 1.0 ? NAN : result.b;\n  result.a = isNaN.a == 1.0 ? NAN : result.a;\n\n  return result;\n`;\n\nexport const RELU = `\n  vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const RELU6 = `\n  vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const ELU = `\n  vec4 result;\n\n  result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);\n  result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);\n  result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);\n  result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);\n\n  return result;\n`;\n\nexport class UnaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      vec4 unaryOperation(vec4 x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        vec4 x = getAAtOutCoords();\n        vec4 y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels, getSourceCoords} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class UnpackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = false;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: number[]) {\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    const channels = getChannels('rc', rank);\n    const dtype = getCoordsDataType(rank);\n    const sourceCoords = getSourceCoords(rank, channels);\n    const innerDims = channels.slice(-2);\n    const coords = rank <= 1 ? 'rc' : `vec2(${innerDims.join(',')})`;\n\n    this.userCode = `\n      void main() {\n        ${dtype} rc = getOutputCoords();\n        vec4 packedInput = getA(${sourceCoords});\n\n        setOutput(getChannel(packedInput, ${coords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import webgl flags.\nimport './flags_webgl';\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {complex, DataId, div, engine, env, imag, max, MemoryInfo, range, real, RecursiveArray, reshape, scalar, softmax, tensor, tidy, TimingInfo, transpose} from '@tensorflow/tfjs-core';\nimport {backend_util, buffer, kernel_impls, slice_util, util} from '@tensorflow/tfjs-core';\nimport {DataStorage, DataType, KernelBackend, NumericDataType, Rank, Scalar, ShapeMap, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorInfo, TypedArray, upcastType} from '@tensorflow/tfjs-core';\n\nimport {addImplCPU, ceilImplCPU, expImplCPU, expm1ImplCPU, floorImplCPU, logImplCPU, multiplyImplCPU, rsqrtImplCPU, simpleAbsImplCPU, sliceImplCPU, subImplCPU} from './kernel_utils/shared';\n\nconst {segment_util} = backend_util;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\n\nimport {AddNProgram} from './addn_gpu';\nimport {AddNPackedProgram} from './addn_packed_gpu';\nimport {ArgMinMaxProgram} from './argminmax_gpu';\nimport {ArgMinMaxPackedProgram} from './argminmax_packed_gpu';\nimport {AvgPool3DBackpropProgram} from './avg_pool_backprop_gpu';\nimport * as binaryop_complex_gpu from './binaryop_complex_gpu';\nimport {BinaryOpComplexProgram} from './binaryop_complex_gpu';\nimport * as binaryop_gpu from './binaryop_gpu';\nimport {BinaryOpProgram} from './binaryop_gpu';\nimport * as binaryop_packed_gpu from './binaryop_packed_gpu';\nimport {BinaryOpPackedProgram} from './binaryop_packed_gpu';\nimport {getWebGLContext} from './canvas_util';\nimport {ClipProgram} from './clip_gpu';\nimport {ClipPackedProgram} from './clip_packed_gpu';\nimport {ComplexAbsProgram} from './complex_abs_gpu';\nimport {ConcatProgram} from './concat_gpu';\nimport {ConcatPackedProgram} from './concat_packed_gpu';\nimport {Conv2DDerFilterProgram, Conv2DDerInputProgram, Conv3DDerFilterProgram, Conv3DDerInputProgram} from './conv_backprop_gpu';\nimport {DepthwiseConv2DDerFilterProgram, DepthwiseConv2DDerInputProgram} from './conv_backprop_gpu_depthwise';\nimport {Conv2DProgram, Conv3DProgram} from './conv_gpu';\nimport {DepthwiseConv2DProgram} from './conv_gpu_depthwise';\nimport {DepthwiseConvPacked2DProgram} from './conv_packed_gpu_depthwise';\nimport {CropAndResizeProgram} from './crop_and_resize_gpu';\nimport {CumSumProgram} from './cumsum_gpu';\nimport {DecodeMatrixProgram} from './decode_matrix_gpu';\nimport {DecodeMatrixPackedProgram} from './decode_matrix_packed_gpu';\nimport {DepthToSpaceProgram} from './depth_to_space_gpu';\nimport {DiagProgram} from './diag_gpu';\nimport {EncodeFloatProgram} from './encode_float_gpu';\nimport {EncodeFloatPackedProgram} from './encode_float_packed_gpu';\nimport {EncodeMatrixProgram} from './encode_matrix_gpu';\nimport {EncodeMatrixPackedProgram} from './encode_matrix_packed_gpu';\nimport * as fft_gpu from './fft_gpu';\nimport {FFTProgram} from './fft_gpu';\nimport {FillProgram} from './fill_gpu';\nimport {GatherProgram} from './gather_gpu';\nimport {GatherNDProgram} from './gather_nd_gpu';\nimport {GPGPUContext} from './gpgpu_context';\nimport * as gpgpu_math from './gpgpu_math';\nimport {GPGPUBinary, GPGPUProgram, TensorData} from './gpgpu_math';\nimport {Im2ColPackedProgram} from './im2col_packed_gpu';\nimport {LRNProgram} from './lrn_gpu';\nimport {LRNGradProgram} from './lrn_grad_gpu';\nimport {LRNPackedProgram} from './lrn_packed_gpu';\nimport {MaxPool3DBackpropProgram} from './max_pool_backprop_gpu';\nimport {MatMulPackedProgram} from './mulmat_packed_gpu';\nimport {MultinomialProgram} from './multinomial_gpu';\nimport {OneHotProgram} from './onehot_gpu';\nimport {PackProgram} from './pack_gpu';\nimport {PadProgram} from './pad_gpu';\nimport {PadPackedProgram} from './pad_packed_gpu';\nimport {Pool3DProgram} from './pool_gpu';\nimport {ReduceProgram} from './reduce_gpu';\nimport {ReshapePackedProgram} from './reshape_packed_gpu';\nimport {ResizeBilinearBackpropProgram} from './resize_bilinear_backprop_gpu';\nimport {ResizeBilinearProgram} from './resize_bilinear_gpu';\nimport {ResizeBilinearPackedProgram} from './resize_bilinear_packed_gpu';\nimport {ResizeNearestNeigborBackpropProgram} from './resize_nearest_neighbor_backprop_gpu';\nimport {ResizeNearestNeighborProgram} from './resize_nearest_neighbor_gpu';\nimport {ReverseProgram} from './reverse_gpu';\nimport {ReversePackedProgram} from './reverse_packed_gpu';\nimport {ScatterProgram} from './scatter_gpu';\nimport {SegmentOpProgram} from './segment_gpu';\nimport {SelectProgram} from './select_gpu';\nimport {SliceProgram} from './slice_gpu';\nimport {SlicePackedProgram} from './slice_packed_gpu';\nimport {StridedSliceProgram} from './strided_slice_gpu';\nimport * as tex_util from './tex_util';\nimport {TextureData, TextureUsage} from './tex_util';\nimport {TextureManager} from './texture_manager';\nimport {TileProgram} from './tile_gpu';\nimport * as unary_op from './unaryop_gpu';\nimport {UnaryOpProgram} from './unaryop_gpu';\nimport * as unary_packed_op from './unaryop_packed_gpu';\nimport {UnaryOpPackedProgram} from './unaryop_packed_gpu';\nimport {UnpackProgram} from './unpack_gpu';\nimport * as webgl_util from './webgl_util';\nimport {BackendValues} from '@tensorflow/tfjs-core';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\ntype KernelInfo = {\n  name: string; query: Promise<number>;\n};\n\nexport type TimerNode = RecursiveArray<KernelInfo>|KernelInfo;\nexport interface CPUTimerQuery {\n  startMs: number;\n  endMs?: number;\n}\n\nexport interface WebGLMemoryInfo extends MemoryInfo {\n  numBytesInGPU: number;\n  // Tracks the total number of bytes allocated on the GPU, accounting for the\n  // physical texture type.\n  numBytesInGPUAllocated: number;\n  // Tracks byte size of textures that were created and then made available for\n  // reuse (disposed).\n  numBytesInGPUFree: number;\n  unreliable: boolean;\n}\n\nexport interface WebGLTimingInfo extends TimingInfo {\n  uploadWaitMs: number;\n  downloadWaitMs: number;\n}\n\nconst binaryCaches: {[webGLVersion: string]: {[key: string]: GPGPUBinary}} = {};\n\nexport function getBinaryCache(webGLVersion: number) {\n  if (webGLVersion in binaryCaches) {\n    return binaryCaches[webGLVersion];\n  }\n  binaryCaches[webGLVersion] = {};\n  return binaryCaches[webGLVersion];\n}\n\nfunction mapActivationToShaderProgram(\n    activation: backend_util.Activation, packed = false): string {\n  if (activation === 'linear') {\n    if (packed) {\n      return unary_packed_op.LINEAR;\n    }\n    return unary_op.LINEAR;\n  } else if (activation === 'relu') {\n    if (packed) {\n      return unary_packed_op.RELU;\n    }\n    return unary_op.RELU;\n  } else if (activation === 'elu') {\n    if (packed) {\n      return unary_packed_op.ELU;\n    }\n    return unary_op.ELU;\n  } else if (activation === 'relu6') {\n    if (packed) {\n      return unary_packed_op.RELU6;\n    }\n    return unary_op.RELU6;\n  } else if (activation === 'prelu') {\n    if (packed) {\n      return binaryop_packed_gpu.PRELU;\n    }\n    return binaryop_gpu.PRELU;\n  }\n  throw new Error(`Activation ${\n      activation} has not been implemented for the WebGL backend.`);\n}\n\n// Empirically determined constant used to determine size threshold for handing\n// off execution to the CPU.\nconst CPU_HANDOFF_SIZE_THRESHOLD = 128;\n\n// Empirically determined constant used to decide the number of MB on GPU\n// before we warn about high memory use. The MB are this constant * screen area\n// * dpi / 1024 / 1024.\nconst BEFORE_PAGING_CONSTANT = 600;\nfunction numMBBeforeWarning(): number {\n  if (env().global.screen == null) {\n    return 1024;  // 1 GB.\n  }\n  return (env().global.screen.height * env().global.screen.width *\n          window.devicePixelRatio) *\n      BEFORE_PAGING_CONSTANT / 1024 / 1024;\n}\n\n// Empirically determined minimal shared dimension in matmul before we forward\n// to a.mul(b).sum() in order to take advantage of GPU parallelism. See\n// https://github.com/tensorflow/tfjs-core/pull/1379 for benchmarks.\nexport const MATMUL_SHARED_DIM_THRESHOLD = 1000;\n\nexport class MathBackendWebGL extends KernelBackend {\n  texData: DataStorage<TextureData>;\n  gpgpu: GPGPUContext;\n\n  // Maps data ids that have a pending read operation, to list of subscribers.\n  private pendingRead = new WeakMap<DataId, Array<(arr: TypedArray) => void>>();\n  // List of data ids that are scheduled for disposal, but are waiting on a\n  // pending read operation.\n  private pendingDisposal = new WeakSet<DataId>();\n  // Used to count the number of 'shallow' sliced tensors that point to the\n  // same data id.\n  private dataRefCount = new WeakMap<DataId, number>();\n  private numBytesInGPU = 0;\n\n  private canvas: HTMLCanvasElement|OffscreenCanvas;\n\n  private programTimersStack: TimerNode[];\n  private activeTimers: TimerNode[];\n  // Accumulated time spent (including blocking) in uploading data to webgl.\n  private uploadWaitMs = 0;\n  // Accumulated time spent (including blocking in downloading data from webgl.\n  private downloadWaitMs = 0;\n  private cpuBackend: KernelBackend;\n\n  // Number of bits of precision of this backend.\n  private floatPrecisionValue: 32|16;\n\n  private textureManager: TextureManager;\n  private binaryCache: {[key: string]: GPGPUBinary};\n  private gpgpuCreatedLocally: boolean;\n  private numMBBeforeWarning: number;\n  private warnedAboutMemory = false;\n  private warnedAboutCPUBackend = false;\n\n  constructor(gpgpu?: GPGPUContext) {\n    super();\n    if (!env().getBool('HAS_WEBGL')) {\n      throw new Error('WebGL is not supported on this device');\n    }\n\n    if (gpgpu == null) {\n      const gl = getWebGLContext(env().getNumber('WEBGL_VERSION'));\n      this.binaryCache = getBinaryCache(env().getNumber('WEBGL_VERSION'));\n      this.gpgpu = new GPGPUContext(gl);\n      this.canvas = gl.canvas;\n      this.gpgpuCreatedLocally = true;\n    } else {\n      this.gpgpu = gpgpu;\n      this.binaryCache = {};\n      this.gpgpuCreatedLocally = false;\n      this.canvas = gpgpu.gl.canvas;\n    }\n    this.textureManager = new TextureManager(this.gpgpu);\n    this.numMBBeforeWarning = numMBBeforeWarning();\n\n    this.texData = new DataStorage(this, engine());\n  }\n\n  numDataIds() {\n    return this.texData.numDataIds() +\n        (this.cpuBackend ? this.cpuBackend.numDataIds() : 0) -\n        this.pendingDeletes;\n  }\n\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    if (env().getBool('WEBGL_CHECK_NUMERICAL_PROBLEMS') ||\n        env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64' && values != null) {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    const dataId = {};\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n    return dataId;\n  }\n\n  /** Increase refCount of a `TextureData`. */\n  incRef(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    texData.refCount++;\n  }\n\n  /** Decrease refCount of a `TextureData`. */\n  decRef(dataId: DataId): void {\n    if (this.texData.has(dataId)) {\n      const texData = this.texData.get(dataId);\n      texData.refCount--;\n    }\n  }\n\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    if (env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.texData.has(dataId)) {\n      const textureData = this.texData.get(dataId);\n\n      textureData.refCount--;\n\n      if (textureData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    const texData = this.texData.get(dataId);\n    const {values, dtype, complexTensors, slice, shape, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.readSync(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n    if (dtype === 'string') {\n      return values;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let result: Float32Array;\n    if (dtype === 'complex64') {\n      const realValues = complexTensors.real.dataSync() as Float32Array;\n      const imagValues = complexTensors.imag.dataSync() as Float32Array;\n      result = backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    } else {\n      result = this.getValuesFromTexture(dataId);\n    }\n\n    if (shouldTimeProgram) {\n      this.downloadWaitMs += util.now() - start;\n    }\n    return this.convertAndCacheOnCPU(dataId, result);\n  }\n\n  async read(dataId: DataId): Promise<BackendValues> {\n    if (this.pendingRead.has(dataId)) {\n      const subscribers = this.pendingRead.get(dataId);\n      return new Promise<TypedArray>(resolve => subscribers.push(resolve));\n    }\n    const texData = this.texData.get(dataId);\n    const {values, shape, slice, dtype, complexTensors, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.read(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n\n    if (!env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED') &&\n        env().getNumber('WEBGL_VERSION') === 2) {\n      throw new Error(\n          `tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and ` +\n          `WEBGL_VERSION=2 not yet supported.`);\n    }\n\n    let buffer = null;\n    let tmpDownloadTarget: TensorInfo;\n\n    if (dtype !== 'complex64' && env().get('WEBGL_BUFFER_SUPPORTED')) {\n      // Possibly copy the texture into a buffer before inserting a fence.\n      tmpDownloadTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpDownloadTarget.dataId);\n\n      buffer = this.gpgpu.createBufferFromTexture(\n          tmpData.texture, ...tex_util.getDenseTexShape(shape));\n    }\n\n    this.pendingRead.set(dataId, []);\n\n    if (dtype !== 'complex64') {\n      // Create a fence and wait for it to resolve.\n      await this.gpgpu.createAndWaitForFence();\n    }\n\n    // Download the values from the GPU.\n    let vals: Float32Array;\n    if (dtype === 'complex64') {\n      const ps = await Promise.all(\n          [complexTensors.real.data(), complexTensors.imag.data()]);\n      const realValues = ps[0];\n      const imagValues = ps[1];\n      vals = backend_util.mergeRealAndImagArrays(\n          realValues as Float32Array, imagValues as Float32Array);\n    } else if (buffer == null) {\n      vals = this.getValuesFromTexture(dataId);\n    } else {\n      const size = util.sizeFromShape(shape);\n      vals = this.gpgpu.downloadFloat32MatrixFromBuffer(buffer, size);\n    }\n    if (tmpDownloadTarget != null) {\n      this.disposeIntermediateTensorInfo(tmpDownloadTarget);\n    }\n    const dTypeVals = this.convertAndCacheOnCPU(dataId, vals);\n\n    const subscribers = this.pendingRead.get(dataId);\n    this.pendingRead.delete(dataId);\n\n    // Notify all pending reads.\n    subscribers.forEach(resolve => resolve(dTypeVals));\n    if (this.pendingDisposal.has(dataId)) {\n      this.pendingDisposal.delete(dataId);\n      this.disposeData(dataId);\n      this.pendingDeletes--;\n    }\n    return dTypeVals;\n  }\n\n  private checkNumericalProblems(values: BackendValues): void {\n    if (values == null) {\n      return;\n    }\n    for (let i = 0; i < values.length; i++) {\n      const num = values[i] as number;\n      if (!webgl_util.canBeRepresented(num)) {\n        if (env().getBool('WEBGL_RENDER_FLOAT32_CAPABLE')) {\n          throw Error(\n              `The value ${num} cannot be represented with your ` +\n              `current settings. Consider enabling float32 rendering: ` +\n              `'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`);\n        }\n        throw Error(`The value ${num} cannot be represented on this device.`);\n      }\n    }\n  }\n\n  private getValuesFromTexture(dataId: DataId): Float32Array {\n    const {shape, dtype, isPacked} = this.texData.get(dataId);\n    const size = util.sizeFromShape(shape);\n    if (env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED')) {\n      const tmpTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpTarget.dataId);\n      const vals = this.gpgpu\n                       .downloadMatrixFromPackedTexture(\n                           tmpData.texture, ...tex_util.getDenseTexShape(shape))\n                       .subarray(0, size);\n\n      this.disposeIntermediateTensorInfo(tmpTarget);\n\n      return vals;\n    }\n\n    const shouldUsePackedProgram =\n        env().getBool('WEBGL_PACK') && isPacked === true;\n    const outputShape =\n        shouldUsePackedProgram ? webgl_util.getShapeAs3D(shape) : shape;\n    const program = shouldUsePackedProgram ?\n        new EncodeFloatPackedProgram(outputShape as [number, number, number]) :\n        new EncodeFloatProgram(outputShape);\n    const output = this.runWebGLProgram(\n        program, [{shape: outputShape, dtype, dataId}], 'float32');\n    const tmpData = this.texData.get(output.dataId);\n    const vals =\n        this.gpgpu\n            .downloadByteEncodedFloatMatrixFromOutputTexture(\n                tmpData.texture, tmpData.texShape[0], tmpData.texShape[1])\n            .subarray(0, size);\n    this.disposeIntermediateTensorInfo(output);\n\n    return vals;\n  }\n\n  async time(f: () => void): Promise<WebGLTimingInfo> {\n    const oldActiveTimers = this.activeTimers;\n    const newActiveTimers: TimerNode[] = [];\n\n    let outerMostTime = false;\n    if (this.programTimersStack == null) {\n      this.programTimersStack = newActiveTimers;\n      outerMostTime = true;\n    } else {\n      this.activeTimers.push(newActiveTimers);\n    }\n    this.activeTimers = newActiveTimers;\n\n    f();\n\n    // needing to split these up because util.flatten only accepts certain types\n    const flattenedActiveTimerQueries =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.query))\n            .filter(d => d != null);\n    const flattenedActiveTimerNames =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.name))\n            .filter(d => d != null);\n\n    this.activeTimers = oldActiveTimers;\n\n    if (outerMostTime) {\n      this.programTimersStack = null;\n    }\n\n    const res: WebGLTimingInfo = {\n      uploadWaitMs: this.uploadWaitMs,\n      downloadWaitMs: this.downloadWaitMs,\n      kernelMs: null,\n      wallMs: null  // will be filled by the engine\n    };\n\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      const kernelMs = await Promise.all(flattenedActiveTimerQueries);\n\n      res['kernelMs'] = util.sum(kernelMs);\n      res['getExtraProfileInfo'] = () =>\n          kernelMs.map((d, i) => ({name: flattenedActiveTimerNames[i], ms: d}))\n              .map(d => `${d.name}: ${d.ms}`)\n              .join(', ');\n    } else {\n      res['kernelMs'] = {\n        error: 'WebGL query timers are not supported in this environment.'\n      };\n    }\n\n    this.uploadWaitMs = 0;\n    this.downloadWaitMs = 0;\n    return res;\n  }\n  memory(): WebGLMemoryInfo {\n    return {\n      unreliable: false,\n      numBytesInGPU: this.numBytesInGPU,\n      numBytesInGPUAllocated: this.textureManager.numBytesAllocated,\n      numBytesInGPUFree: this.textureManager.numBytesFree\n    } as WebGLMemoryInfo;\n  }\n\n  private startTimer(): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.beginQuery();\n    }\n    return {startMs: util.now(), endMs: null};\n  }\n\n  private endTimer(query: WebGLQuery|CPUTimerQuery): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      this.gpgpu.endQuery();\n      return query;\n    }\n    (query as CPUTimerQuery).endMs = util.now();\n    return query;\n  }\n\n  private async getQueryTime(query: WebGLQuery|CPUTimerQuery): Promise<number> {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.waitForQueryAndGetTime(query as WebGLQuery);\n    }\n    const timerQuery = query as CPUTimerQuery;\n    return timerQuery.endMs - timerQuery.startMs;\n  }\n\n  private pendingDeletes = 0;\n\n  disposeData(dataId: DataId): void {\n    if (this.pendingDisposal.has(dataId)) {\n      return;\n    }\n    if (this.pendingRead.has(dataId)) {\n      this.pendingDisposal.add(dataId);\n      this.pendingDeletes++;\n      return;\n    }\n    // No-op if already disposed.\n    if (!this.texData.has(dataId)) {\n      return;\n    }\n\n    this.releaseGPUData(dataId);\n    const {complexTensors} = this.texData.get(dataId);\n    if (complexTensors != null) {\n      complexTensors.real.dispose();\n      complexTensors.imag.dispose();\n    }\n    this.texData.delete(dataId);\n  }\n\n  private releaseGPUData(dataId: DataId): void {\n    const {texture, dtype, texShape, usage, isPacked, slice} =\n        this.texData.get(dataId);\n    const key = slice && slice.origDataId || dataId;\n    const refCount = this.dataRefCount.get(key);\n    if (refCount > 1) {\n      this.dataRefCount.set(key, refCount - 1);\n    } else {\n      this.dataRefCount.delete(key);\n      if (texture != null) {\n        this.numBytesInGPU -= this.computeBytes(texShape, dtype);\n        this.textureManager.releaseTexture(texture, texShape, usage, isPacked);\n      }\n    }\n    const texData = this.texData.get(dataId);\n    texData.texture = null;\n    texData.texShape = null;\n    texData.isPacked = false;\n    texData.slice = null;\n  }\n\n  getTexture(dataId: DataId): WebGLTexture {\n    this.uploadToGPU(dataId);\n    return this.texData.get(dataId).texture;\n  }\n\n  /**\n   * Returns internal information for the specific data bucket. Used in unit\n   * tests.\n   */\n  getDataInfo(dataId: DataId): TextureData {\n    return this.texData.get(dataId);\n  }\n\n  private getCPUBackend(): KernelBackend|null {\n    if (!env().getBool('WEBGL_CPU_FORWARD')) {\n      return null;\n    }\n\n    if (this.cpuBackend == null) {\n      this.cpuBackend = engine().findBackend('cpu');\n    }\n\n    return this.cpuBackend;\n  }\n\n  /*\n  Tests whether all the inputs to an op are small and on the CPU. This heuristic\n  determines when it would be faster to execute a kernel on the CPU. WebGL\n  kernels opt into running this check and forwarding when appropriate.\n  TODO(https://github.com/tensorflow/tfjs/issues/872): Develop a more\n  sustainable strategy for optimizing backend execution of ops.\n   */\n  shouldExecuteOnCPU(\n      inputs: TensorInfo[],\n      sizeThreshold = CPU_HANDOFF_SIZE_THRESHOLD): boolean {\n    const cpuBackend = this.getCPUBackend();\n    if (!this.warnedAboutCPUBackend && cpuBackend == null) {\n      console.warn(\n          'Your application contains ops that are small enough to be ' +\n          'executed on the CPU backend, however the CPU backend cannot ' +\n          'be found. Consider importing the CPU backend ' +\n          '(@tensorflow/tfjs-backend-cpu) for better performance.');\n\n      this.warnedAboutCPUBackend = true;\n    }\n\n    return cpuBackend != null &&\n        inputs.every(\n            input => this.texData.get(input.dataId).texture == null &&\n                util.sizeFromShape(input.shape) < sizeThreshold);\n  }\n\n  getGPGPUContext(): GPGPUContext {\n    return this.gpgpu;\n  }\n\n  complex<T extends Tensor>(real: T, imag: T): T {\n    const result = this.makeOutput(real.shape, 'complex64');\n    const resultData = this.texData.get(result.dataId);\n    // The backend owns the reference to the underlying real and imaginary\n    // clones. These will explicitly get disposed when the complex tensor is\n    // disposed.\n    resultData.complexTensors = {\n      real: engine().keep(real.clone()),\n      imag: engine().keep(imag.clone())\n    };\n\n    return result as T;\n  }\n  real<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.real.clone() as T;\n  }\n  imag<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.imag.clone() as T;\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = sliceImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, begin, size, x.shape,\n          x.dtype);\n      return this.makeOutput(size, x.dtype, outValues);\n    }\n    // Short-circuit computation if the slice is zero-sized.\n    if (util.sizeFromShape(size) === 0) {\n      return tensor([], size, x.dtype) as T;\n    }\n    const {isPacked} = this.texData.get(x.dataId);\n    const isContinous = slice_util.isSliceContinous(x.shape, begin, size);\n    if (isPacked || !isContinous) {\n      const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n          new SlicePackedProgram(size) :\n          new SliceProgram(size);\n      const customSetup = program.getCustomSetupFunc(begin);\n      return this.compileAndRun(program, [x], null, customSetup);\n    }\n    this.uploadToGPU(x.dataId);\n    return this.shallowSlice(x, begin, size) as T;\n  }\n\n  private shallowSlice(x: Tensor, begin: number[], size: number[]): Tensor {\n    const xTexData = this.texData.get(x.dataId);\n    const t = this.makeOutput(size, x.dtype);\n    const newTexData = this.texData.get(t.dataId);\n    // Copy texture data from the original tensor.\n    Object.assign(newTexData, xTexData);\n    newTexData.shape = size;\n    newTexData.dtype = x.dtype;\n    let flatOffset = slice_util.computeFlatOffset(begin, x.strides);\n    if (xTexData.slice) {\n      // We are slicing an already sliced tensor, so we have to accumulate\n      // the offset.\n      flatOffset += xTexData.slice.flatOffset;\n    }\n    newTexData.slice = {\n      flatOffset,\n      // Point to the original dataId, which is used to do ref counting.\n      origDataId: xTexData.slice && xTexData.slice.origDataId || x.dataId\n    };\n\n    // Increase the ref count for that data bucket.\n    const refCount = this.dataRefCount.get(newTexData.slice.origDataId) || 1;\n    this.dataRefCount.set(newTexData.slice.origDataId, refCount + 1);\n\n    return t;\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x], () => this.cpuBackend.stridedSlice(x, begin, end, strides));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tensor([], outShape) as T;\n    }\n\n    const program = new StridedSliceProgram(begin, strides, outShape);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new ReversePackedProgram(x.shape, axis) :\n        new ReverseProgram(x.shape, axis);\n    return this.compileAndRun(program, [x]);\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    if (tensors[0].dtype === 'complex64') {\n      const reals = tensors.map((t) => real(t));\n      const imags = tensors.map((t) => imag(t));\n      return complex(this.concat(reals, axis), this.concat(imags, axis));\n    }\n\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n    if (tensors.length > env().getNumber('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.concat(tensors.slice(0, midIndex), axis);\n      const rightSide = this.concat(tensors.slice(midIndex), axis);\n      return this.concat([leftSide, rightSide], axis);\n    }\n    if (env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') && tensors[0].rank > 1) {\n      const program = new ConcatPackedProgram(tensors.map(t => t.shape), axis);\n      return this.compileAndRun(program, tensors);\n    }\n    // Any concat of n-dimensional tensors across any axis can be reduced to\n    // a concatenation of two-dimensional tensors across the axis 1 by first\n    // partitioning the axes of the original tensors into those less than the\n    // axis to be concatenated and the rest. Then reshape the tensors\n    // into a two-dimensional tensor by collapsing these two sets of axes and\n    // concatenate the resulting matrices across the axis 1, finally reshaping\n    // the result to have the proper shape.\n    const outShape =\n        backend_util.computeOutShape(tensors.map(t => t.shape), axis);\n    const tensors2D =\n        tensors.map(t => t.as2D(-1, util.sizeFromShape(t.shape.slice(axis))));\n    const program = new ConcatProgram(tensors2D.map(t => t.shape));\n    const res: Tensor = this.compileAndRun(program, tensors2D);\n    return res.reshape(outShape);\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    const cpuRes = this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.neg(x));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.NEG, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.NEG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const [batch, , ] = a.shape;\n\n    // Since the matrices are vectors, it is faster to call mul().sum()\n    // because sum() is O(sqrt(N)) due to divide-and-conquer.\n    if ((outerShapeA === 1 || outerShapeB === 1) &&\n        sharedDim > MATMUL_SHARED_DIM_THRESHOLD) {\n      if (transposeA) {\n        a = transpose(a, [0, 2, 1]);\n      }\n      if (transposeB) {\n        b = transpose(b, [0, 2, 1]);\n      }\n\n      const a3D = outerShapeB === 1 ? a : a.as3D(batch, sharedDim, 1);\n      const axis = outerShapeB === 1 ? 2 : 1;\n      const b3D = outerShapeB === 1 ? b.as3D(batch, 1, sharedDim) : b;\n      return this.multiply(a3D, b3D).sum(axis, true /* keepDims */);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB);\n    return this.compileAndRun<Tensor3D>(program, [a, b], dtype);\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const [batch, , ] = a.shape;\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB,\n        hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [a, b];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun<Tensor3D>(program, inputs, dtype);\n  }\n\n  multiply(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64') {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n\n      const realProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.REAL, a.shape, b.shape);\n      const imagProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.IMAG, a.shape, b.shape);\n\n      const inputs = [\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.imag),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.imag)\n      ];\n      const real = this.compileAndRun<Tensor>(realProgram, inputs);\n      const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n\n      const complex = this.complex(real, imag);\n      real.dispose();\n      imag.dispose();\n      return complex;\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = multiplyImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.MUL, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.MUL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], a.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n        new LRNPackedProgram(x.shape, radius, bias, alpha, beta) :\n        new LRNProgram(x.shape, radius, bias, alpha, beta);\n    return this.compileAndRun(program, [x]);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program =\n        new LRNGradProgram(inputImage.shape, depthRadius, bias, alpha, beta);\n    return this.compileAndRun(program, [inputImage, outputImage, dy]);\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    if (x.dtype === 'string') {\n      const data = this.readSync(x.dataId) as Uint8Array[];\n      const decodedData = data.map(d => util.decodeString(d));\n      const buf = buffer(x.shape, x.dtype, decodedData);\n      return tile(buf, reps) as T;\n    }\n    const program = new TileProgram(x.shape, reps);\n    return this.compileAndRun(program, [x]);\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new PadPackedProgram(x.shape, paddings, constantValue) :\n        new PadProgram(x.shape, paddings, constantValue);\n    return this.compileAndRun(program, [x]);\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x, indices], () => this.cpuBackend.gather(x, indices, axis));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = new GatherProgram(x.shape, indices.size, axis);\n    return this.compileAndRun(program, [x, indices]);\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'batchToSpaceND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: Array<[number, number]>): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'spaceToBatchND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const completePaddings: Array<[number, number]> = [[0, 0]];\n    completePaddings.push(...paddings);\n    for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n      completePaddings.push([0, 0]);\n    }\n\n    const paddedX = x.pad(completePaddings);\n\n    const reshapedPaddedShape =\n        backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n    const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n        reshapedPaddedShape.length, blockShape.length, false);\n\n    const flattenShape = backend_util.getReshapedPermuted(\n        paddedX.shape, blockShape, prod, false);\n\n    const paddedXT = transpose(\n        paddedX.reshape(reshapedPaddedShape),\n        permutedReshapedPaddedPermutation);\n    return reshape(paddedXT, flattenShape) as T;\n  }\n\n  private reduce(\n      x: Tensor2D, reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod',\n      dtype: DataType): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const outSize = Math.ceil(inSize / windowSize);\n    const reduceInfo = {windowSize, inSize, batchSize, outSize};\n    const program = new ReduceProgram(reduceInfo, reduceType);\n    const output = this.compileAndRun<Tensor2D>(program, [x], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.reduce(output, reduceType, dtype);\n  }\n\n  private argReduce(\n      x: Tensor2D, reduceType: 'max'|'min',\n      bestIndicesA: Tensor2D = null): Tensor2D {\n    let batchSize = x.shape[0];\n    let inSize = x.shape[1];\n    if (bestIndicesA != null) {\n      batchSize = bestIndicesA.shape[0];\n      inSize = bestIndicesA.shape[1];\n    }\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const reduceInfo = {\n      windowSize,\n      inSize,\n      batchSize,\n      outSize: Math.ceil(inSize / windowSize)\n    };\n    const program =\n        new ArgMinMaxProgram(reduceInfo, reduceType, bestIndicesA == null);\n    const inputs = [x];\n    if (bestIndicesA != null) {\n      inputs.push(bestIndicesA);\n    }\n    const output = this.compileAndRun<Tensor2D>(program, inputs, 'int32');\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.argReduce(x, reduceType, output);\n  }\n\n  private argReducePacked(\n      x: Tensor, reduceType: 'max'|'min', bestIndicesA: Tensor = null): Tensor {\n    const inShape = bestIndicesA != null ? bestIndicesA.shape : x.shape;\n    const inSize = inShape[inShape.length - 1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const program = new ArgMinMaxPackedProgram(\n        inShape, windowSize, reduceType, bestIndicesA == null);\n    const inputs = bestIndicesA == null ? [x] : [x, bestIndicesA];\n    const output = this.compileAndRun<Tensor>(program, inputs, 'int32');\n    if (output.rank === x.rank) {\n      return this.argReducePacked(x, reduceType, output);\n    }\n    return output;\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'sum', outputDType).reshape(outShape);\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.prod(x, axes));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'prod', outputDType).reshape(outShape);\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    let axis = 0;\n    const permutation = backend_util.getAxesPermutation([axis], x.rank);\n    let permutedX = x;\n    if (permutation != null) {\n      permutedX = transpose(x, permutation);\n      axis = backend_util.getInnerMostAxes(1, x.rank)[0];\n    }\n\n    const outShape =\n        segment_util.computeOutShape(permutedX.shape, axis, numSegments);\n    const inSize = util.sizeFromShape([permutedX.shape[axis]]);\n    const a2D = permutedX.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    let result =\n        this.segOpCompute(\n                a2D, 'unsortedSegmentSum', segmentIds, outputDType, numSegments)\n            .reshape(outShape);\n    if (permutation != null) {\n      result =\n          transpose(result, backend_util.getUndoAxesPermutation(permutation));\n    }\n    return result;\n  }\n\n  private segOpCompute(\n      x: Tensor2D, segOpType: 'unsortedSegmentSum', segmentIds: Tensor1D,\n      dtype: DataType, numSegments: number): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize =\n        segment_util.segOpComputeOptimalWindowSize(inSize, numSegments);\n    const segOpInfo = {windowSize, inSize, batchSize, numSegments};\n    const program = new SegmentOpProgram(segOpInfo, segOpType);\n    const output =\n        this.compileAndRun<Tensor2D>(program, [x, segmentIds], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === numSegments) {\n      return output;\n    }\n    segmentIds = range(0, numSegments).tile([inSize / windowSize]);\n    return this.segOpCompute(output, segOpType, segmentIds, dtype, numSegments);\n  }\n\n  private argMinMaxReduce(x: Tensor, axis: number, reduceType: 'min'|'max'):\n      Tensor {\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims(\n        'arg' + reduceType.charAt(0).toUpperCase() + reduceType.slice(1), axes,\n        x.rank);\n    if (!env().getBool('WEBGL_PACK_REDUCE') || x.rank <= 2) {\n      const [outShape, reduceShape] =\n          backend_util.computeOutAndReduceShapes(x.shape, axes);\n      const inSize = util.sizeFromShape(reduceShape);\n      const a2D = x.as2D(-1, inSize);\n      return this.argReduce(a2D, reduceType).reshape(outShape);\n    }\n    return this.argReducePacked(x, reduceType);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'min');\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'max');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `WebGL cumsum shader expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const size = x.shape[axis];\n    let result = x;\n    // Use cumsum parallel algorithm, ref:\n    // https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-39-parallel-prefix-sum-scan-cuda\n    for (let i = 0; i <= Math.ceil(Math.log2(size)) - 1; i++) {\n      const program = new CumSumProgram(x.shape, false, reverse);\n      const customSetup = program.getCustomSetupFunc(i);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result], result.dtype, customSetup);\n      prevResult.dispose();\n    }\n    // For exclusive cumsum, shift the end result in the direction of sum and\n    // add 0 to the front index.\n    if (exclusive) {\n      const program = new CumSumProgram(x.shape, exclusive, reverse);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result]);\n      prevResult.dispose();\n    }\n\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.EQUAL, 'bool');\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.NOT_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.NOT_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.less(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.LESS, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LESS_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.greater(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.GREATER, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.GREATER, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.GREATER_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.GREATER_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalNot<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOGICAL_NOT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_AND, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_AND, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_OR, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_OR, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    const program = new SelectProgram(condition.rank, a.shape, a.rank);\n    return this.compileAndRun(\n        program, [condition, a, b], upcastType(a.dtype, b.dtype));\n  }\n\n  where(condition: Tensor): Tensor2D {\n    backend_util.warn(\n        'tf.where() in webgl locks the UI thread. ' +\n        'Call tf.whereAsync() instead');\n    const condVals = condition.dataSync();\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    const xVals = x.dataSync();\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'min', a2D.dtype).reshape(outShape);\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.minimum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MIN, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MIN, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MOD, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MOD, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.maximum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MAX, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MAX, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'all', a2D.dtype).reshape(outShape);\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'any', a2D.dtype).reshape(outShape);\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    const op = binaryop_gpu.INT_DIV;\n    const outputDtype = 'int32';\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.INT_DIV, outputDtype);\n    }\n    const program = new BinaryOpProgram(op, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], outputDtype);\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.ADD);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = addImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.ADD, dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.ADD, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  private packedUnaryOp(x: TensorInfo, op: string, dtype: DataType) {\n    const program = new UnaryOpPackedProgram(x.shape, op);\n    return this.compileAndRun<Tensor>(program, [x], dtype);\n  }\n\n  private packedBinaryOp(\n      a: TensorInfo, b: TensorInfo, op: string, dtype: DataType,\n      checkOutOfBounds = false) {\n    const program =\n        new BinaryOpPackedProgram(op, a.shape, b.shape, checkOutOfBounds);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  /**\n   * Computes a complex binary operation that can be decomposed into a simple\n   * binary operation on both the real and imagary parts.\n   */\n  private complexSeparableBinaryOp(a: Tensor, b: Tensor, op: string): Tensor {\n    const aData = this.texData.get(a.dataId);\n    const bData = this.texData.get(b.dataId);\n\n    const [real, imag] = [\n      [aData.complexTensors.real, bData.complexTensors.real],\n      [aData.complexTensors.imag, bData.complexTensors.imag]\n    ].map(complexParts => {\n      const [aPart, bPart] = complexParts;\n\n      const aHandle = this.makeComplexComponentTensorInfo(a, aPart);\n      const bHandle = this.makeComplexComponentTensorInfo(b, bPart);\n\n      const program = new BinaryOpProgram(op, a.shape, b.shape);\n      return this.compileAndRun<Tensor>(\n          program, [aHandle, bHandle], upcastType(aPart.dtype, bPart.dtype));\n    });\n\n    const complex = this.complex(real, imag);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  // Returns a TensorInfo with the complex shape and the dataId of the\n  // underlying part. We need to do this because a reshaped complex tensor is\n  // not reflected in its parts.\n  private makeComplexComponentTensorInfo(\n      complexTensor: Tensor, complexPart: Tensor): TensorInfo {\n    return {\n      dataId: complexPart.dataId,\n      dtype: complexPart.dtype,\n      shape: complexTensor.shape\n    };\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n\n    // Limit the number of uploaded textures for optimization.\n    if (tensors.length > env().get('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.addN(tensors.slice(0, midIndex));\n      const rightSide = this.addN(tensors.slice(midIndex));\n      return this.addN([leftSide, rightSide]);\n    }\n\n    const dtype =\n        tensors.map(t => t.dtype).reduce((d1, d2) => upcastType(d1, d2));\n    const shapes = tensors.map(t => t.shape);\n    // We can make sure shapes are identical in op level.\n    const usePackedOp = env().getBool('WEBGL_PACK');\n    const program = usePackedOp ?\n        new AddNPackedProgram(tensors[0].shape, shapes) :\n        new AddNProgram(tensors[0].shape, shapes);\n    return this.compileAndRun<T>(program, tensors, dtype);\n  }\n\n  subtract(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.SUB);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = subImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.SUB, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.SUB, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    const usePackedOp = env().getBool('WEBGL_PACK_BINARY_OPERATIONS');\n    const program = usePackedOp ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.POW, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.POW, a.shape, b.shape);\n    const dtype = upcastType(a.dtype, b.dtype);\n    return this.compileAndRun<T>(program, [a, b], dtype);\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          ceilImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.CEIL, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.CEIL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  floor<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = floorImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.FLOOR, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.FLOOR);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_NAN);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_INF);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_FINITE);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n\n  round<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ROUND);\n    return this.compileAndRun(program, [x]);\n  }\n\n  exp<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          expImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXP, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXP);\n    return this.compileAndRun(program, [x]);\n  }\n\n  expm1<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = expm1ImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXPM1, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXPM1);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n    const a = this.subtract(logits, maxLogit.reshape(expandedShape));\n    const b = this.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax kernel\n    // modularization.\n    return div(b, sumExp);\n  }\n\n  log<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          logImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.LOG, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  log1p<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG1P);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sqrt<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  rsqrt<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = rsqrtImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.RSQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reciprocal<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.RECIPROCAL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU6);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU6);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  prelu<T extends Tensor>(x: T, alpha: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.PRELU, x.shape, alpha.shape) :\n        new BinaryOpProgram(binaryop_gpu.PRELU, x.shape, alpha.shape);\n    return this.compileAndRun(program, [x, alpha]);\n  }\n\n  elu<T extends Tensor>(x: T): T {\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.ELU, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.ELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.ELU_DER, dy.shape, y.shape) :\n        new BinaryOpProgram(binaryop_gpu.ELU_DER, dy.shape, y.shape);\n    return this.compileAndRun(program, [dy, y]);\n  }\n\n  selu<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  int<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TO_INT);\n    return this.compileAndRun(program, [x], 'int32');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    let program;\n    if (env().getBool('WEBGL_PACK_CLIP')) {\n      program = new ClipPackedProgram(x.shape);\n    } else {\n      program = new ClipProgram(x.shape);\n    }\n    const customSetup = program.getCustomSetupFunc(min, max);\n    return this.compileAndRun(program, [x], null, customSetup);\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    // TODO: handle cases when x is complex.\n    if (this.shouldExecuteOnCPU([x]) && x.dtype !== 'complex64') {\n      const outValues =\n          simpleAbsImplCPU(this.texData.get(x.dataId).values as TypedArray);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.ABS, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.ABS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  complexAbs<T extends Tensor>(x: T): T {\n    const xData = this.texData.get(x.dataId);\n\n    const program = new ComplexAbsProgram(x.shape);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    return this.compileAndRun<Tensor>(program, inputs) as T;\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGMOID);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SOFTPLUS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASIN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acos<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atan<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATAN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.COSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  tanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ERF);\n    return this.compileAndRun(program, [x]);\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.STEP(alpha));\n    return this.compileAndRun(program, [x]);\n  }\n\n  private conv2dByMatMul(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Reshapes conv2D input to 2D tensors, uses matMul and then reshape the\n    // result from 2D to 4D.\n    const xShape = x.shape;\n    const xTexData = this.texData.get(x.dataId);\n    const sharedMatMulDim = convInfo.inChannels;\n    const outerShapeX = xShape[0] * xShape[1] * xShape[2];\n    const outerShapeFilter = convInfo.outChannels;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const transposeA = false;\n    const transposeB = false;\n\n    // TODO: Once reduction ops are packed, batchMatMul will always be packed\n    // and we can remove this condition.\n    const batchMatMulWillBeUnpacked =\n        (outerShapeX === 1 || outerShapeFilter === 1) &&\n        sharedMatMulDim > MATMUL_SHARED_DIM_THRESHOLD;\n    const reshapeWillBeExpensive = xShape[2] % 2 !== 0 && !!xTexData.isPacked;\n\n    if (batchMatMulWillBeUnpacked || !env().getBool('WEBGL_LAZILY_UNPACK') ||\n        !env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ||\n        !reshapeWillBeExpensive) {\n      const targetShape = isChannelsLast ? xShape[0] * xShape[1] * xShape[2] :\n                                           xShape[0] * xShape[2] * xShape[3];\n      const xReshaped = reshape(x, [1, targetShape, convInfo.inChannels]);\n      const filterReshaped =\n          reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n      const result = this.fusedBatchMatMul({\n        a: xReshaped as Tensor3D,\n        b: filterReshaped as Tensor3D,\n        transposeA,\n        transposeB,\n        bias,\n        activation,\n        preluActivationWeights\n      });\n      return reshape(result, convInfo.outShape);\n    }\n\n    // Following optimization is specific to packed |x| with odd row count\n    // (For example, in channelLast mode, 'row count' refers to x.shape[2]):\n    // we avoid expensive packed 2x2 reshape by padding row count to next,\n    // even number. When x.shape[2] is odd, the result of packed batchMatMul is\n    // the same (has the same texture layout and and values in the texture) as\n    // it is for even x.shape[2] + 1. We make the odd-rows tensor to look like\n    // even-rows tensor before the operation and, after the batchMatMul,\n    // fix the even-rows result to have odd number of rows.\n    const targetShape = isChannelsLast ?\n        xShape[0] * xShape[1] * (xShape[2] + 1) :\n        xShape[0] * xShape[2] * (xShape[3] + 1);\n    const xReshaped: TensorInfo = {\n      dataId: x.dataId,\n      shape: [1, targetShape, convInfo.inChannels],\n      dtype: x.dtype\n    };\n    // xTexData.shape gets referenced from GPGPUBinary.inShapeInfos.\n    // Decrementing row count, after batchMatMul->...->compileProgram leads to\n    // invalid row count within the reference in GPGPUBinary.inShapeInfos.\n    // Alternative fix would be to provide a copy to GPGPUBinary.inShapeInfos\n    // in compileProgram method, but that would affect compilation of all\n    // programs - instead, provide a copy here, with even row count, before\n    // calling batchMatMul->...->compileProgram and after that, the original\n    // xTexData.shape is restored.\n    const originalXTexDataShape = xTexData.shape;\n    xTexData.shape = xTexData.shape.slice();\n    xTexData.shape[xTexData.shape.length - 2]++;\n    util.assert(\n        webgl_util.isReshapeFree(xTexData.shape, xReshaped.shape),\n        () => `packed reshape ${xTexData.shape} to ${\n            xReshaped.shape} isn't free`);\n    const filterReshaped =\n        reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n    const pointwiseConv = this.fusedBatchMatMul({\n      a: xReshaped as Tensor3D,\n      b: filterReshaped as Tensor3D,\n      transposeA,\n      transposeB,\n      bias,\n      activation,\n      preluActivationWeights\n    });\n    const pointwiseConvTexData = this.texData.get(pointwiseConv.dataId);\n    util.assert(\n        pointwiseConvTexData.isPacked,\n        () => 'batchMatMul result is expected to be packed');\n    // Restore the input shape to original.\n    xTexData.shape = originalXTexDataShape;\n    // Set the output shape - there is no need for expensive reshape as data\n    // layout is already correct.\n    pointwiseConvTexData.shape = convInfo.outShape;\n    return engine().makeTensorFromDataId(\n               pointwiseConv.dataId, convInfo.outShape, pointwiseConv.dtype) as\n        Tensor4D;\n  }\n\n  private conv2dWithIm2Row(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Rearranges conv2d input so each block to be convolved over forms the\n    // column of a new matrix with shape [filterWidth * filterHeight *\n    // inChannels, outHeight * outWidth]. The filter is also rearranged so each\n    // output channel forms a row of a new matrix with shape [outChannels,\n    // filterWidth * filterHeight * inChannels]. The convolution is then\n    // computed by multiplying these matrices and reshaping the result.\n    const {\n      filterWidth,\n      filterHeight,\n      inChannels,\n      outWidth,\n      outHeight,\n      dataFormat\n    } = convInfo;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n\n    const sharedDim = filterWidth * filterHeight * inChannels;\n    const numCols = outHeight * outWidth;\n    const x2ColShape = [sharedDim, numCols];\n    const transposeA = true;\n    const transposeB = false;\n\n    const xSqueezed = x.squeeze([0]);\n    const w2Row = filter.reshape([1, sharedDim, -1]);\n\n    const im2ColProgram =\n        new Im2ColPackedProgram(x2ColShape, xSqueezed.shape, convInfo);\n    const im2Col: Tensor3D =\n        this.compileAndRun<Tensor2D>(im2ColProgram, [xSqueezed]).reshape([\n          1, x2ColShape[0], x2ColShape[1]\n        ]);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const matmulProgram = new MatMulPackedProgram(\n        im2Col.shape, [1, numCols, convInfo.outChannels], transposeA,\n        transposeB, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [im2Col, w2Row];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    const product = this.compileAndRun<Tensor4D>(matmulProgram, inputs);\n\n    if (isChannelsLast) {\n      return product.reshape([1, outHeight, outWidth, convInfo.outChannels]);\n    } else {\n      return product.reshape([1, convInfo.outChannels, outHeight, outWidth]);\n    }\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && input.shape[0] === 1) {\n      return this.conv2dWithIm2Row(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, false) : null;\n    const program = new Conv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [input, filter];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun(program, inputs);\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(x, filter, convInfo);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && x.shape[0] === 1) {\n      return this.conv2dWithIm2Row(x, filter, convInfo);\n    }\n    const program = new Conv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new Conv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    const program = new Conv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    const shouldPackDepthwiseConv = env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1;\n    const fusedActivation = activation ?\n        mapActivationToShaderProgram(activation, shouldPackDepthwiseConv) :\n        null;\n    const inputs: Tensor[] = [input, filter];\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    if (hasBias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (shouldPackDepthwiseConv) {\n      program = new DepthwiseConvPacked2DProgram(\n          convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n      return this.compileAndRun(program, inputs);\n    }\n\n    program = new DepthwiseConv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    return this.compileAndRun(program, inputs);\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1) {\n      program = new DepthwiseConvPacked2DProgram(convInfo);\n      return this.compileAndRun(program, [x, filter]);\n    }\n\n    program = new DepthwiseConv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Conv3DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return backend_util.castTensor(x, dtype, this);\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = this.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'avg', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const avgPool3dBackpropProgram = new AvgPool3DBackpropProgram(convInfo);\n    return this.compileAndRun(avgPool3dBackpropProgram, [dy], x.dtype);\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'max', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const getPositions = true;\n    const maxPool3dPositionsProgram =\n        new Pool3DProgram(convInfo, 'max', getPositions);\n    const maxPool3dPositions: Tensor5D =\n        this.compileAndRun(maxPool3dPositionsProgram, [x]);\n    const maxPool3dBackPropProgram = new MaxPool3DBackpropProgram(convInfo);\n    const result = this.compileAndRun(\n        maxPool3dBackPropProgram, [dy, maxPool3dPositions], x.dtype);\n    maxPool3dPositions.dispose();\n    return result as Tensor5D;\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_IMAGE_OPERATIONS') ?\n        new ResizeBilinearPackedProgram(\n            x.shape, newHeight, newWidth, alignCorners) :\n        new ResizeBilinearProgram(x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    const program = new ResizeBilinearBackpropProgram(dy, x, alignCorners);\n\n    return this.compileAndRun(program, [dy]);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = new ResizeNearestNeighborProgram(\n        x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x]);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    const program =\n        new ResizeNearestNeigborBackpropProgram(dy, x, alignCorners);\n    return this.compileAndRun(program, [dy]);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    const probs = normalized ? logits : softmax(logits);\n    const batchSize = probs.shape[0];\n    const numOutcomes = probs.shape[1];\n    const program = new MultinomialProgram(batchSize, numOutcomes, numSamples);\n    const customSetup = program.getCustomSetupFunc(seed);\n    return this.compileAndRun(program, [probs], 'int32', customSetup);\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    const program = new OneHotProgram(indices.size, depth, onValue, offValue);\n    return this.compileAndRun(program, [indices]);\n  }\n\n  diag(x: Tensor): Tensor {\n    const program = new DiagProgram(x.size);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    const program = new CropAndResizeProgram(\n        image.shape, boxes.shape, cropSize, method, extrapolationValue);\n    return this.compileAndRun(program, [image, boxes, boxIndex], 'float32');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = (dataFormat === 'NHWC') ? x.shape[1] : x.shape[2];\n    const inputWidth = (dataFormat === 'NHWC') ? x.shape[2] : x.shape[3];\n    const inputDepth = (dataFormat === 'NHWC') ? x.shape[3] : x.shape[1];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const outputShape = (dataFormat === 'NHWC') ?\n        [batchSize, outputHeight, outputWidth, outputDepth] :\n        [batchSize, outputDepth, outputHeight, outputWidth];\n\n    const program = new DepthToSpaceProgram(outputShape, blockSize, dataFormat);\n    return this.compileAndRun(program, [x]);\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n    const flattenIndices = indices.reshape([numUpdates, sliceRank]);\n    const flattenX = updates.reshape([numUpdates, sliceSize]);\n\n    if (outputSize === 0) {\n      return backend_util.reshapeTensor(tensor([]), shape);\n    }\n    const defaultValue = scalar(0);\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, flattenIndices.rank, flattenX.rank, strides,\n        flattenShape);\n    const res: Tensor =\n        this.compileAndRun(program, [flattenX, flattenIndices, defaultValue]);\n    return res.reshape(shape);\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n\n    const sumDupeIndices = false;\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, sparseIndices.rank, sparseValues.rank, strides,\n        [outputSize, 1], sumDupeIndices);\n    const res: Tensor = this.compileAndRun(\n        program, [sparseValues, sparseIndices, defaultValue]);\n    return res.reshape(outputShape);\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    const inverse = false;\n    return this.fftImpl(x, inverse);\n  }\n\n  ifft(x: Tensor2D): Tensor2D {\n    const inverse = true;\n    return this.fftImpl(x, inverse);\n  }\n\n  private fftImpl(x: Tensor2D, inverse: boolean): Tensor2D {\n    const xData = this.texData.get(x.dataId);\n\n    const realProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.REAL, x.shape, inverse);\n    const imagProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.IMAG, x.shape, inverse);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    const real = this.compileAndRun<Tensor>(realProgram, inputs);\n    const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n    const complex = this.complex(real, imag).as2D(x.shape[0], x.shape[1]);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n\n    const flattenIndices = indices.reshape([numSlices, sliceRank]);\n    const flattenX = x.reshape([x.size / sliceSize, sliceSize]);\n    const program =\n        new GatherNDProgram(sliceRank, strides, [numSlices, sliceSize]);\n    const res: Tensor = this.compileAndRun(program, [flattenX, flattenIndices]);\n    return res.reshape(resultShape);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n\n    if (dtype === 'string') {\n      // String type should be handled in CPU memory.\n      const values = util.getArrayFromDType(dtype, util.sizeFromShape(shape));\n      values.fill(value as string);\n      return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n    } else {\n      const program = new FillProgram(shape, value as number);\n      const customSetup = program.getCustomSetupFunc(value as number);\n      return this.compileAndRun(program, [], dtype, customSetup);\n    }\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported under string dtype');\n    } else {\n      // TODO(cais, smilkov): Add WebGL shader for onesLike:\n      //   https://github.com/tensorflow/tfjs/issues/1293\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return this.fill(x.shape, x.dtype === 'string' ? '' : 0, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    // TODO: Use CPU implementation due to the precision problem in Safari.\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  makeTensorInfo(shape: number[], dtype: DataType, values?: BackendValues):\n      TensorInfo {\n    const dataId = this.write(values, shape, dtype);\n    this.texData.get(dataId).usage = null;\n    return {dataId, shape, dtype};\n  }\n\n  private makeOutput<T extends Tensor>(\n      shape: number[], dtype: DataType, values?: BackendValues): T {\n    const {dataId} = this.makeTensorInfo(shape, dtype, values);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  private unpackTensor(input: TensorInfo): TensorInfo {\n    const program = new UnpackProgram(input.shape);\n    return this.runWebGLProgram(program, [input], input.dtype);\n  }\n\n  private packTensor(input: TensorInfo): TensorInfo {\n    const program = new PackProgram(input.shape);\n    const preventEagerUnpackingOutput = true;\n    return this.runWebGLProgram(\n        program, [input], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOutput);\n  }\n\n  private packedReshape(input: TensorInfo, afterShape: number[]): TensorInfo {\n    const input3DShape = [\n      webgl_util.getBatchDim(input.shape),\n      ...webgl_util.getRowsCols(input.shape)\n    ] as [number, number, number];\n    const input3D: TensorInfo = {\n      dtype: input.dtype,\n      shape: input3DShape,\n      dataId: input.dataId\n    };\n    const afterShapeAs3D = [\n      webgl_util.getBatchDim(afterShape), ...webgl_util.getRowsCols(afterShape)\n    ] as [number, number, number];\n\n    const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n    const preventEagerUnpackingOfOutput = true;\n    const output = this.runWebGLProgram(\n        program, [input3D], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOfOutput);\n    return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n  }\n\n  private decode(dataId: DataId): TensorInfo {\n    const texData = this.texData.get(dataId);\n    const {isPacked, shape, dtype} = texData;\n    const shapeAs3D =\n        webgl_util.getShapeAs3D(shape) as [number, number, number];\n    let program;\n    if (isPacked) {\n      program = new DecodeMatrixPackedProgram(shapeAs3D);\n    } else {\n      program = new DecodeMatrixProgram(shapeAs3D);\n    }\n    const preventEagerUnpackingOfOutput = true;\n    const out = this.runWebGLProgram(\n        program, [{shape: shapeAs3D, dtype, dataId}], dtype,\n        null /* customSetup */, preventEagerUnpackingOfOutput);\n    return {dtype, shape, dataId: out.dataId};\n  }\n\n  runWebGLProgram(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): TensorInfo {\n    const output = this.makeTensorInfo(program.outputShape, outputDtype);\n    const outData = this.texData.get(output.dataId);\n    if (program.packedOutput) {\n      outData.isPacked = true;\n    }\n    if (program.outPackingScheme === tex_util.PackingScheme.DENSE) {\n      const texelShape = tex_util.getDenseTexShape(program.outputShape);\n      // For a densely packed output, we explicitly set texShape\n      // so it doesn't get assigned later according to our typical packing\n      // scheme wherein a single texel can only contain values from adjacent\n      // rows/cols.\n      outData.texShape = texelShape.map(d => d * 2) as [number, number];\n    }\n    if (program.outTexUsage != null) {\n      outData.usage = program.outTexUsage;\n    }\n    if (util.sizeFromShape(output.shape) === 0) {\n      // Short-circuit the computation since the result is empty (has 0 in its\n      // shape).\n      outData.values =\n          util.getTypedArrayFromDType(output.dtype as 'float32', 0);\n      return output;\n    }\n\n    const dataToDispose: TensorInfo[] = [];\n    const inputsData: TensorData[] = inputs.map(input => {\n      if (input.dtype === 'complex64') {\n        throw new Error(\n            `GPGPUProgram does not support complex64 input. For complex64 ` +\n            `dtypes, please separate the program into real and imaginary ` +\n            `parts.`);\n      }\n\n      let texData = this.texData.get(input.dataId);\n\n      if (texData.texture == null) {\n        if (!program.packedInputs &&\n            util.sizeFromShape(input.shape) <=\n                env().getNumber('WEBGL_SIZE_UPLOAD_UNIFORM')) {\n          // Upload small tensors that live on the CPU as uniforms, not as\n          // textures. Do this only when the environment supports 32bit floats\n          // due to problems when comparing 16bit floats with 32bit floats.\n          // TODO(https://github.com/tensorflow/tfjs/issues/821): Make it\n          // possible for packed shaders to sample from uniforms.\n          return {\n            shape: input.shape,\n            texData: null,\n            isUniform: true,\n            uniformValues: texData.values as TypedArray\n          };\n        }\n\n        // This ensures that if a packed program's inputs have not yet been\n        // uploaded to the GPU, they get uploaded as packed right off the bat.\n        if (program.packedInputs) {\n          texData.isPacked = true;\n          texData.shape = input.shape;\n        }\n      } else if (!!texData.isPacked !== !!program.packedInputs) {\n        input = texData.isPacked ? this.unpackTensor(input) :\n                                   this.packTensor(input);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n      } else if (\n          texData.isPacked &&\n          !webgl_util.isReshapeFree(texData.shape, input.shape)) {\n        // This is a special case where a texture exists for a tensor\n        // but the shapes are incompatible (due to packing constraints) because\n        // the tensor did not have a chance to go through the packed reshape\n        // shader. This only happens when we reshape the *same* tensor to form\n        // *distinct* inputs to an op, e.g. dotting a vector with itself. This\n        // case will disappear once packed uploading is the default.\n\n        const savedInput = input;\n        const targetShape = input.shape;\n\n        input.shape = texData.shape;\n        input = this.packedReshape(input as Tensor, targetShape);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n\n        savedInput.shape = targetShape;\n      }\n\n      this.uploadToGPU(input.dataId);\n      return {shape: input.shape, texData, isUniform: false};\n    });\n\n    this.uploadToGPU(output.dataId);\n    const outputData:\n        TensorData = {shape: output.shape, texData: outData, isUniform: false};\n    const key = gpgpu_math.makeShaderKey(program, inputsData, outputData);\n    const binary = this.getAndSaveBinary(key, () => {\n      return gpgpu_math.compileProgram(\n          this.gpgpu, program, inputsData, outputData);\n    });\n    const shouldTimeProgram = this.activeTimers != null;\n    let query: WebGLQuery|CPUTimerQuery;\n    if (shouldTimeProgram) {\n      query = this.startTimer();\n    }\n\n    gpgpu_math.runProgram(\n        this.gpgpu, binary, inputsData, outputData, customSetup);\n\n    dataToDispose.forEach(info => this.disposeIntermediateTensorInfo(info));\n\n    if (shouldTimeProgram) {\n      query = this.endTimer(query);\n      this.activeTimers.push(\n          {name: program.constructor.name, query: this.getQueryTime(query)});\n    }\n\n    if (!env().getBool('WEBGL_LAZILY_UNPACK') && outData.isPacked &&\n        preventEagerUnpackingOfOutput === false) {\n      const unpacked = this.unpackTensor(output);\n      this.disposeIntermediateTensorInfo(output);\n      return unpacked;\n    }\n    return output;\n  }\n\n  compileAndRun<K extends TensorInfo>(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype?: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): K {\n    outputDtype = outputDtype || inputs[0].dtype;\n    const outInfo = this.runWebGLProgram(\n        program, inputs, outputDtype, customSetup,\n        preventEagerUnpackingOfOutput);\n    return engine().makeTensorFromDataId(\n               outInfo.dataId, outInfo.shape, outInfo.dtype) as {} as K;\n  }\n\n  private getAndSaveBinary(key: string, getBinary: () => GPGPUBinary):\n      GPGPUBinary {\n    if (!(key in this.binaryCache)) {\n      this.binaryCache[key] = getBinary();\n    }\n    return this.binaryCache[key];\n  }\n\n  getTextureManager(): TextureManager {\n    return this.textureManager;\n  }\n\n  private disposed = false;\n\n  dispose() {\n    if (this.disposed) {\n      return;\n    }\n    // Avoid disposing the compiled webgl programs during unit testing because\n    // it slows down test execution.\n    if (!env().getBool('IS_TEST')) {\n      const allKeys = Object.keys(this.binaryCache);\n      allKeys.forEach(key => {\n        this.gpgpu.deleteProgram(this.binaryCache[key].webGLProgram);\n        delete this.binaryCache[key];\n      });\n    }\n    this.textureManager.dispose();\n    if (this.canvas != null &&\n        (typeof (HTMLCanvasElement) !== 'undefined' &&\n         this.canvas instanceof HTMLCanvasElement)) {\n      this.canvas.remove();\n    } else {\n      this.canvas = null;\n    }\n    if (this.gpgpuCreatedLocally) {\n      this.gpgpu.program = null;\n      this.gpgpu.dispose();\n    }\n    this.disposed = true;\n  }\n\n  floatPrecision(): 16|32 {\n    if (this.floatPrecisionValue == null) {\n      this.floatPrecisionValue = tidy(() => {\n        if (!env().get('WEBGL_RENDER_FLOAT32_ENABLED')) {\n          // Momentarily switching DEBUG flag to false so we don't throw an\n          // error trying to upload a small value.\n          const debugFlag = env().getBool('DEBUG');\n          env().set('DEBUG', false);\n          const underflowCheckValue = this.abs(scalar(1e-8)).dataSync()[0];\n          env().set('DEBUG', debugFlag);\n\n          if (underflowCheckValue > 0) {\n            return 32;\n          }\n        }\n        return 16;\n      });\n    }\n    return this.floatPrecisionValue;\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  private uploadToGPU(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    const {shape, dtype, values, texture, usage, isPacked} = texData;\n\n    if (texture != null) {\n      // Array is already on GPU. No-op.\n      return;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let texShape = texData.texShape;\n    if (texShape == null) {\n      texShape = webgl_util.getTextureShapeFromLogicalShape(shape, isPacked);\n      texData.texShape = texShape;\n    }\n\n    if (values != null) {\n      const shapeAs3D = webgl_util.getShapeAs3D(shape);\n\n      let program;\n      let width = texShape[1], height = texShape[0];\n      const isByteArray = values instanceof Uint8Array;\n\n      if (isPacked) {\n        [width, height] = tex_util.getPackedMatrixTextureShapeWidthHeight(\n            texShape[0], texShape[1]);\n        program = new EncodeMatrixPackedProgram(\n            shapeAs3D, [height, width], isByteArray);\n      } else {\n        program =\n            new EncodeMatrixProgram(shapeAs3D, [height, width], isByteArray);\n      }\n\n      const tempDenseInputHandle = this.makeTensorInfo([height, width], dtype);\n      if (isByteArray) {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.PIXELS;\n      } else {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.UPLOAD;\n      }\n      this.gpgpu.uploadDenseMatrixToTexture(\n          this.getTexture(tempDenseInputHandle.dataId), width, height,\n          values as TypedArray);\n\n      // We want the output to remain packed regardless of the value of\n      // WEBGL_PACK.\n      const preventEagerUnpacking = true;\n      const encodedOutputTarget = this.runWebGLProgram(\n          program, [tempDenseInputHandle], dtype, null, preventEagerUnpacking);\n\n      // Have the original texture assume the identity of the encoded output.\n      const outputTexData = this.texData.get(encodedOutputTarget.dataId);\n      texData.texture = outputTexData.texture;\n      texData.texShape = outputTexData.texShape;\n      texData.isPacked = outputTexData.isPacked;\n      texData.usage = outputTexData.usage;\n\n      this.disposeIntermediateTensorInfo(tempDenseInputHandle);\n      this.texData.delete(encodedOutputTarget.dataId);\n\n      // Once uploaded, don't store the values on cpu.\n      texData.values = null;\n      if (shouldTimeProgram) {\n        this.uploadWaitMs += util.now() - start;\n      }\n    } else {\n      const newTexture = this.acquireTexture(texShape, usage, dtype, isPacked);\n      texData.texture = newTexture;\n    }\n  }\n\n  private convertAndCacheOnCPU(dataId: DataId, float32Values?: Float32Array):\n      TypedArray {\n    const texData = this.texData.get(dataId);\n    const {dtype} = texData;\n\n    this.releaseGPUData(dataId);\n\n    if (float32Values != null) {\n      texData.values = float32ToTypedArray(float32Values, dtype as 'float32');\n    }\n    return texData.values as TypedArray;\n  }\n\n  private acquireTexture(\n      texShape: [number, number], texType: TextureUsage, dtype: DataType,\n      isPacked: boolean): WebGLTexture {\n    this.numBytesInGPU += this.computeBytes(texShape, dtype);\n    if (!this.warnedAboutMemory &&\n        this.numBytesInGPU > this.numMBBeforeWarning * 1024 * 1024) {\n      const mb = (this.numBytesInGPU / 1024 / 1024).toFixed(2);\n      this.warnedAboutMemory = true;\n      console.warn(\n          `High memory usage in GPU: ${mb} MB, ` +\n          `most likely due to a memory leak`);\n    }\n    return this.textureManager.acquireTexture(texShape, texType, isPacked);\n  }\n\n  private computeBytes(shape: [number, number], dtype: DataType) {\n    return shape[0] * shape[1] * util.bytesPerElement(dtype);\n  }\n\n  private tryRunOnCpuOrThrow<T extends Tensor>(\n      inputs: TensorInfo[], fn: () => T): T|null {\n    if (this.shouldExecuteOnCPU(inputs)) {\n      try {\n        return fn();\n      } catch (e) {\n        if (env().getBool('IS_TEST')) {\n          throw new Error('CPU forwarding failed');\n        }\n      }\n    }\n    return null;\n  }\n}\n\nfunction float32ToTypedArray<D extends NumericDataType>(\n    a: Float32Array, dtype: D): tf.DataTypeMap[D] {\n  if (dtype === 'float32' || dtype === 'complex64') {\n    return a as tf.DataTypeMap[D];\n  } else if (dtype === 'int32' || dtype === 'bool') {\n    const result = (dtype === 'int32') ? new Int32Array(a.length) :\n                                         new Uint8Array(a.length);\n    for (let i = 0; i < result.length; ++i) {\n      result[i] = Math.round(a[i]);\n    }\n    return result as tf.DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport * as gpgpu_util from './gpgpu_util';\nimport * as webgl_util from './webgl_util';\n\nexport {MathBackendWebGL, WebGLMemoryInfo, WebGLTimingInfo} from './backend_webgl';\nexport {setWebGLContext} from './canvas_util';\nexport {GPGPUContext} from './gpgpu_context';\nexport {GPGPUProgram} from './gpgpu_math';\n// WebGL specific utils.\nexport {gpgpu_util, webgl_util};\n\n/**\n * Enforce use of half precision textures if available on the platform.\n *\n * @doc {heading: 'Environment', namespace: 'webgl'}\n */\nexport function forceHalfFloat(): void {\n  env().set('WEBGL_FORCE_F16_TEXTURES', true);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// base.ts is the webgl backend without auto kernel registration.\n\nimport {device_util, registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from './backend_webgl';\nexport {version as version_webgl} from './version';\n\nif (device_util.isBrowser()) {\n  registerBackend('webgl', () => new MathBackendWebGL(), 2 /* priority */);\n}\n\n// Export webgl utilities\nexport * from './webgl';\n\n// Export forceHalfFlost under webgl namespace for the union bundle.\nimport {forceHalfFloat} from './webgl';\nexport const webgl = {forceHalfFloat};\n", "import {BinaryInputs, DataType, env, KernelFunc, UnaryInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BinaryOpProgram} from '../binaryop_gpu';\nimport {BinaryOpPackedProgram} from '../binaryop_packed_gpu';\nimport {UnaryOpProgram} from '../unaryop_gpu';\n\nexport const CHECK_NAN_SNIPPET_UNARY = `if (isnan(x)) return x;`;\n\nexport const CHECK_NAN_SNIPPET_BINARY = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const CHECK_NAN_SNIPPET_BINARY_PACKED = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param opSnippets Op snippet to create `UnaryOpProgram`.\n */\nexport function unaryKernelFunc(opSnippet: string): KernelFunc {\n  return ({inputs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = new UnaryOpProgram(x.shape, opSnippet);\n    return webglBackend.runWebGLProgram(program, [x], x.dtype);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param opSnippet Op snippet to create `BinaryOpProgram`.\n * @param packedOpSnippet Op snippet to create `BinaryOpPackedProgram`.\n * @param checkOutOfBoundsForPackedProgram Whether to set checkOutOfBounds=true\n *     when creating BinaryOpPackedProgram.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    opSnippet: string, packedOpSnippet: string,\n    checkOutOfBoundsForPackedProgram?: boolean, dtype?: DataType): KernelFunc {\n  // TODO(jingjin): handle complex64.\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            packedOpSnippet, a.shape, b.shape,\n            !!checkOutOfBoundsForPackedProgram) :\n        new BinaryOpProgram(opSnippet, a.shape, b.shape);\n    const $dtype = dtype || a.dtype;\n    const output = webglBackend.runWebGLProgram(program, [a, b], $dtype);\n    return output;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc, CHECK_NAN_SNIPPET_BINARY, CHECK_NAN_SNIPPET_BINARY_PACKED} from '../kernel_utils/kernel_funcs_utils';\n\nconst ATAN2 = CHECK_NAN_SNIPPET_BINARY + `\n  return atan(a, b);\n`;\n\nconst ATAN2_PACKED = `\n  vec4 result = atan(a, b);\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET_BINARY_PACKED + `\n  return result;\n`;\n\nexport const atan2KernelFunc = binaryKernelFunc(ATAN2, ATAN2_PACKED);\n\nexport const atan2Config: KernelConfig = {\n  kernelName: Atan2,\n  backendName: 'webgl',\n  kernelFunc: atan2KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendWebGL}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'webgl',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function avgPool(args: {\n  inputs: AvgPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const avgPoolProgram = new Pool2DProgram(convInfo, 'avg', false);\n  return backend.runWebGLProgram(avgPoolProgram, [x], 'float32');\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'webgl',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {AvgPool2DBackpropProgram} from '../avg_pool_backprop_gpu';\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const avgPoolBackpropProgram = new AvgPool2DBackpropProgram(convInfo);\n  return backend.runWebGLProgram(avgPoolBackpropProgram, [dy], x.dtype);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = '0.0';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = '1.0';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        float x = getXAtOutCoords();\n        float mean = getMeanAtOutCoords();\n        float variance = getVarianceAtOutCoords();\n        float offset = ${offsetSnippet};\n        float scale = ${scaleSnippet};\n        float inv = scale * inversesqrt(variance + float(${varianceEpsilon}));\n        setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = 'vec4(0.0)';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = 'vec4(1.0)';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        vec4 offset = ${offsetSnippet};\n        vec4 scale = ${scaleSnippet};\n\n        vec4 x = getXAtOutCoords();\n        vec4 mean = getMeanAtOutCoords();\n        vec4 variance = getVarianceAtOutCoords();\n\n        vec4 inv = scale * inversesqrt(variance + vec4(${varianceEpsilon}));\n\n        setOutput((x - mean) * inv + offset);\n      }\n    `;\n  }\n}\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BatchNormProgram} from '../batchnorm_gpu';\nimport {BatchNormPackedProgram} from '../batchnorm_packed_gpu';\n\nexport const batchNormKernelFunc: (params: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendWebGL,\n  attrs: FusedBatchNormAttrs\n}) => TensorInfo = ({inputs, backend, attrs}) => {\n  const {x, mean, variance, offset, scale} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const finalInputs = [x, mean, variance];\n\n  let offsetShape = null;\n  if (offset != null) {\n    offsetShape = offset.shape;\n    finalInputs.push(offset);\n  }\n\n  let scaleShape = null;\n  if (scale != null) {\n    scaleShape = scale.shape;\n    finalInputs.push(scale);\n  }\n\n  const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n      new BatchNormPackedProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon) :\n      new BatchNormProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon);\n  const output =\n      backend.runWebGLProgram(program, finalInputs, finalInputs[0].dtype);\n\n  return output;\n};\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'webgl',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst COS = CHECK_NAN_SNIPPET_UNARY + `\n  return cos(x);\n`;\n\nexport const cosKernelFunc = unaryKernelFunc(COS);\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'webgl',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\n// Without the equality check div produces 0.9999 for a = b, which when\n// floored can cause errors.\nconst DIV = `\nif (a == b) {\n  return 1.0;\n};\nreturn a / b;`;\n\n// We do the same as in ./binaryop_gpu, with vec4 and ivec4.\n// On Linux, the vectorized implementation produces NaNs when a and b are 0.\nconst DIV_PACKED = `\n  // vec4 one = vec4(equal(a, b));\n  // return one + (vec4(1.0) - one) * a / b;\n  vec4 result = a / b;\n  if(a.x == b.x) {\n    result.x = 1.;\n  }\n  if(a.y == b.y) {\n    result.y = 1.;\n  }\n  if(a.z == b.z) {\n    result.z = 1.;\n  }\n  if(a.w == b.w) {\n    result.w = 1.;\n  }\n\n  return result;\n`;\n\nexport const divKernelFunc = binaryKernelFunc(\n    DIV, DIV_PACKED, true /* checkOutOfBoundsForPackedProgram */);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'webgl',\n  kernelFunc: divKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FlipLeftRightProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(imageShape: [number, number, number, number]) {\n    const imageWidth = imageShape[2];\n    this.outputShape = imageShape;\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n\n          int coordX = ${imageWidth} - x;\n          float outputValue;\n          if(coordX >= 0 && coordX < ${imageWidth}) {\n            outputValue = getImage(coords[0], coords[1], coordX, coords[3]);\n          } else {\n            outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {FlipLeftRightProgram} from '../flip_left_right_gpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new FlipLeftRightProgram((image as Tensor4D).shape);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n        vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${width}.0, ${height}.0);\n\n        vec4 values = ${glsl.texture2D}(A, uv);\n        float value;\n        if (depth == 0) {\n          value = values.r;\n        } else if (depth == 1) {\n          value = values.g;\n        } else if (depth == 2) {\n          value = values.b;\n        } else if (depth == 3) {\n          value = values.a;\n        }\n\n        setOutput(floor(value * 255.0 + 0.5));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n\n        vec4 result = vec4(0.);\n\n        for(int row=0; row<=1; row++) {\n          for(int col=0; col<=1; col++) {\n            texC = coords[1] + row;\n            depth = coords[2] + col;\n\n            vec2 uv = (vec2(texC, texR) + halfCR) /\n                       vec2(${width}.0, ${height}.0);\n            vec4 values = ${glsl.texture2D}(A, uv);\n            float value;\n            if (depth == 0) {\n              value = values.r;\n            } else if (depth == 1) {\n              value = values.g;\n            } else if (depth == 2) {\n              value = values.b;\n            } else if (depth == 3) {\n              value = values.a;\n            }\n\n            result[row * 2 + col] = floor(value * 255.0 + 0.5);\n          }\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {TextureUsage} from '../tex_util';\n\nimport {FromPixelsProgram} from './FromPixels_utils/from_pixels_gpu';\nimport {FromPixelsPackedProgram} from './FromPixels_utils/from_pixels_packed_gpu';\n\nexport const fromPixelsConfig: KernelConfig = {\n  kernelName: FromPixels,\n  backendName: 'webgl',\n  kernelFunc: fromPixels as {} as KernelFunc,\n};\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\nfunction fromPixels(args: {\n  inputs: FromPixelsInputs,\n  backend: MathBackendWebGL,\n  attrs: FromPixelsAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  let {pixels} = inputs;\n  const {numChannels} = attrs;\n\n  const isVideo = typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement;\n  const isImage = typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement;\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n\n  const texShape: [number, number] = [height, width];\n  const outShape = [height, width, numChannels];\n\n  if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement | HTMLImageElement, 0, 0, width, height);\n    pixels = fromPixels2DContext.canvas;\n  }\n\n  const tempPixelHandle = backend.makeTensorInfo(texShape, 'int32');\n  // This is a byte texture with pixels.\n  backend.texData.get(tempPixelHandle.dataId).usage = TextureUsage.PIXELS;\n  backend.gpgpu.uploadPixelDataToTexture(\n      backend.getTexture(tempPixelHandle.dataId), pixels as ImageData);\n  const program = env().getBool('WEBGL_PACK') ?\n      new FromPixelsPackedProgram(outShape) :\n      new FromPixelsProgram(outShape);\n  const res = backend.runWebGLProgram(program, [tempPixelHandle], 'int32');\n  backend.disposeData(tempPixelHandle.dataId);\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReduceProgram} from '../reduce_gpu';\n\ntype ReduceTypes = 'all'|'any'|'max'|'min'|'sum'|'prod';\n\n// Returns an array of configuration objects that describe each stage of the\n// reduction.\nfunction getReductionStages(inShape: number[]):\n    Array<{inSize: number, windowSize: number, outSize: number}> {\n  const stages = [];\n\n  while (stages.length === 0 || stages[stages.length - 1].outSize !== 1) {\n    const outSize: number =\n        stages.length ? stages[stages.length - 1].outSize : inShape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(outSize);\n    stages.push({\n      inSize: outSize,\n      windowSize,\n      outSize: Math.ceil(outSize / windowSize)\n    });\n  }\n\n  return stages;\n}\n\nexport function reduce(\n    x: TensorInfo, dtype: DataType, reductionType: ReduceTypes,\n    backend: MathBackendWebGL): TensorInfo {\n  const reductionStages = getReductionStages(x.shape);\n\n  let result = x;\n  for (let i = 0; i < reductionStages.length; i++) {\n    const {inSize, windowSize, outSize} = reductionStages[i];\n\n    const program = new ReduceProgram(\n        {windowSize, inSize, batchSize: x.shape[0], outSize}, reductionType);\n    const previousResult = result;\n    result = backend.runWebGLProgram(program, [result], dtype);\n\n    if (previousResult.dataId !== x.dataId) {\n      backend.disposeData(previousResult.dataId);\n    }\n  }\n\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReshapePackedProgram} from '../reshape_packed_gpu';\nimport {getBatchDim, getRowsCols} from '../webgl_util';\n\nexport function packedReshape(\n    input: TensorInfo, afterShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const input3DShape =\n      [getBatchDim(input.shape),\n       ...getRowsCols(input.shape)] as [number, number, number];\n  const input3D: TensorInfo = {\n    dtype: input.dtype,\n    shape: input3DShape,\n    dataId: input.dataId\n  };\n  const afterShapeAs3D =\n      [getBatchDim(afterShape),\n       ...getRowsCols(afterShape)] as [number, number, number];\n\n  const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n  const preventEagerUnpackingOfOutput = true;\n  const output = backend.runWebGLProgram(\n      program, [input3D], input.dtype, null /* customSetup */,\n      preventEagerUnpackingOfOutput);\n  return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {packedReshape} from '../kernel_utils/reshape';\nimport {isReshapeFree} from '../webgl_util';\n\nexport function reshape(args: {\n  inputs: ReshapeInputs,\n  backend: MathBackendWebGL,\n  attrs: ReshapeAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n  const webglBackend = backend;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  const xTexData = webglBackend.texData.get(x.dataId);\n  if (xTexData.isPacked && !isReshapeFree(x.shape, $shape) &&\n      !(xTexData.texture !== null && isReshapeFree(xTexData.shape, $shape))) {\n    return packedReshape(x, $shape, webglBackend);\n  }\n\n  webglBackend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'webgl',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {reduce} from '../kernel_utils/reduce';\nimport {reshape} from '../kernels/Reshape';\n\nexport function maxImpl(\n    x: TensorInfo, reduceShape: number[], outShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const inSize = util.sizeFromShape(reduceShape);\n  const xSize = util.sizeFromShape(x.shape);\n  const batchSize = xSize / inSize;\n  const reshapedInput =\n      reshape({inputs: {x}, attrs: {shape: [batchSize, inSize]}, backend});\n\n  const reduced = reduce(reshapedInput, x.dtype, 'max', backend);\n  const reshapedOutput =\n      reshape({inputs: {x: reduced}, attrs: {shape: outShape}, backend});\n\n  backend.disposeIntermediateTensorInfo(reshapedInput);\n  backend.disposeIntermediateTensorInfo(reduced);\n\n  return reshapedOutput;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposeProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const switched = getSwitchedCoords(newDim);\n\n    this.userCode = `\n    void main() {\n      ${dtype} resRC = getOutputCoords();\n      setOutput(getA(${switched}));\n    }\n    `;\n  }\n}\n\nfunction getSwitchedCoords(newDim: number[]): string {\n  const rank = newDim.length;\n  if (rank > 6) {\n    throw Error(`Transpose for rank ${rank} is not yet supported`);\n  }\n  const originalOrder =\n      ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u', 'resRC.v'];\n  const switchedCoords = new Array(rank);\n  for (let i = 0; i < newDim.length; i++) {\n    switchedCoords[newDim[i]] = originalOrder[i];\n  }\n  return switchedCoords.join();\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getVecChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    if (this.rank > 6) {\n      throw Error(\n          `Packed transpose for rank ${this.rank} is not yet supported.`);\n    }\n    const dtype = getCoordsDataType(this.rank);\n\n    const outputOrder = getVecChannels('rc', this.rank);\n    const switchedOrder = new Array(this.rank);\n    for (let i = 0; i < newDim.length; i++) {\n      switchedOrder[newDim[i]] = outputOrder[i];\n    }\n    const innerDims = `vec2(${switchedOrder.slice(-2).join()})`;\n    const nextColumn =\n        `++${outputOrder[this.rank - 1]} < ${outputShape[this.rank - 1]}`;\n    const getc = `getChannel(getA(${switchedOrder.join()}), ${innerDims})`;\n\n    this.userCode = `\n    void main() {\n      ${dtype} rc = getOutputCoords();\n      vec4 result = vec4(0.);\n      result[0] = ${getc};\n      if(${nextColumn}) {\n        result[1] = ${getc};\n      }\n      --${outputOrder[this.rank - 1]};\n      if(++${outputOrder[this.rank - 2]} < ${outputShape[this.rank - 2]}) {\n        result[2] = ${getc};\n        if(${nextColumn}) {\n          result[3] = ${getc};\n        }\n      }\n      setOutput(result);\n    }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {transposeImplCPU} from '../kernel_utils/shared';\nimport {TransposeProgram} from '../transpose_gpu';\nimport {TransposePackedProgram} from '../transpose_packed_gpu';\n\nexport function transposeImpl(\n    x: TensorInfo, perm: number[], backend: MathBackendWebGL): TensorInfo {\n  const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n      new TransposePackedProgram(x.shape, perm) :\n      new TransposeProgram(x.shape, perm);\n  return backend.runWebGLProgram(program, [x], x.dtype);\n}\n\nexport {transposeImplCPU};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {maxImplCPU} from '../kernel_utils/shared';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl, transposeImplCPU} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    const maxInputIsTransposed = permutedAxes != null;\n    const shouldExecuteOnCPU = webglBackend.shouldExecuteOnCPU([x]);\n\n    let maxInput = x;\n    if (maxInputIsTransposed) {\n      if (shouldExecuteOnCPU) {\n        const xTexData = webglBackend.texData.get(maxInput.dataId);\n        const values = xTexData.values as TypedArray;\n\n        const newShape: number[] = new Array(xRank);\n        for (let i = 0; i < newShape.length; i++) {\n          newShape[i] = x.shape[permutedAxes[i]];\n        }\n        const maxInputValues =\n            transposeImplCPU(values, x.shape, x.dtype, permutedAxes, newShape);\n\n        maxInput = webglBackend.makeTensorInfo(newShape, x.dtype);\n        const maxInputData = webglBackend.texData.get(maxInput.dataId);\n        maxInputData.values = maxInputValues;\n      } else {\n        maxInput = transposeImpl(x, permutedAxes, webglBackend);\n      }\n\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n    }\n\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(maxInput.shape, axes);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // rather than reshape at the end, set the target shape here.\n      outShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n    }\n\n    let out;\n    if (shouldExecuteOnCPU) {\n      const xTexData = webglBackend.texData.get(maxInput.dataId);\n      const values = xTexData.values as TypedArray;\n\n      const outValues = maxImplCPU(\n          values, util.sizeFromShape(reduceShape), outShape, x.dtype);\n\n      out = webglBackend.makeTensorInfo(outShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = maxImpl(maxInput, reduceShape, outShape, webglBackend);\n    }\n\n    if (maxInputIsTransposed) {\n      webglBackend.disposeIntermediateTensorInfo(maxInput);\n    }\n\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function maxPool(args: {\n  inputs: MaxPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const maxPoolProgram = new Pool2DProgram(convInfo, 'max', false);\n  return backend.runWebGLProgram(maxPoolProgram, [x], x.dtype);\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'webgl',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {MaxPool2DBackpropProgram} from '../max_pool_backprop_gpu';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const getPositions = true;\n  const maxPoolPositionsProgram =\n      new Pool2DProgram(convInfo, 'max', getPositions);\n  const maxPoolPositions: TensorInfo =\n      backend.runWebGLProgram(maxPoolPositionsProgram, [x], x.dtype);\n\n  const maxPoolBackPropProgram = new MaxPool2DBackpropProgram(convInfo);\n  const result = backend.runWebGLProgram(\n      maxPoolBackPropProgram, [dy, maxPoolPositions], x.dtype);\n  backend.disposeIntermediateTensorInfo(maxPoolPositions);\n  return result;\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\n\nexport function maxPoolWithArgmaxImpl(\n    x: TensorInfo, includeBatchInIndex: boolean,\n    convInfo: backend_util.Conv2DInfo,\n    backend: MathBackendWebGL): TensorInfo[] {\n  let program = new Pool2DProgram(convInfo, 'max', false);\n  const poolOutput = backend.runWebGLProgram(program, [x], 'float32');\n\n  program = new Pool2DProgram(convInfo, 'max', true, true, includeBatchInIndex);\n  const indexOutput = backend.runWebGLProgram(program, [x], 'float32');\n  return [poolOutput, indexOutput];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    util.assert(\n        x.shape.length === 4,\n        () => `Error in maxPool: input must be rank 4 but got rank ${\n            x.shape.length}.`);\n    const dilations: [number, number] = [1, 1];\n    util.assert(\n        backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n            `Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        dilations, pad);\n\n    const [result, indexes] =\n        maxPoolWithArgmaxImpl(x, includeBatchInIndex, convInfo, webglBackend);\n    return [result, indexes];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV3, NonMaxSuppressionV3Attrs, NonMaxSuppressionV3Inputs, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV3Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV3,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV3Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold} =\n        attrs as unknown as NonMaxSuppressionV3Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n\n    return kernel_impls.nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal);\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs, TypedArray} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class RotateProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], radians: number,\n      fillValue: number|[number, number, number],\n      center: number|[number, number]) {\n    const imageHeight = imageShape[1];\n    const imageWidth = imageShape[2];\n    const sinFactor = Math.sin(radians).toFixed(3);\n    const cosFactor = Math.cos(radians).toFixed(3);\n    this.outputShape = imageShape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const centerXString = centerX.toFixed(3);\n    const centerYString = centerY.toFixed(3);\n\n    let fillSnippet = '';\n    if (typeof fillValue === 'number') {\n      fillSnippet = `float outputValue = ${fillValue.toFixed(2)};`;\n    } else {\n      fillSnippet = `\n        vec3 fill = vec3(${fillValue.join(',')});\n        float outputValue = fill[coords[3]];`;\n    }\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n          int y = coords[1];\n          float coordXFloat = (float(x) - ${centerXString}) * ${\n        cosFactor} - (float(y) - ${centerYString}) * ${sinFactor};\n          float coordYFloat = (float(x) - ${centerXString}) * ${\n        sinFactor} + (float(y) - ${centerYString}) * ${cosFactor};\n          int coordX = int(round(coordXFloat + ${centerXString}));\n          int coordY = int(round(coordYFloat + ${centerYString}));\n          ${fillSnippet}\n          if(coordX >= 0 && coordX < ${imageWidth} && coordY >= 0 && coordY < ${\n        imageHeight}) {\n            outputValue = getImage(coords[0], coordY, coordX, coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {RotateProgram} from '../rotate_gpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new RotateProgram(\n        (image as Tensor4D).shape, radians, fillValue, center);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SIN = CHECK_NAN_SNIPPET_UNARY + `\n  return sin(x);\n`;\n\nexport const sinKernelFunc = unaryKernelFunc(SIN);\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'webgl',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Square} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARE = `return x * x;`;\n\nexport const squareKernelFunc = unaryKernelFunc(SQUARE);\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'webgl',\n  kernelFunc: squareKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const squaredDifferenceKernelFunc =\n    binaryKernelFunc(SQUARED_DIFFERENCE, SQUARED_DIFFERENCE);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'webgl',\n  kernelFunc: squaredDifferenceKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst TAN = `return tan(x);`;\n\nexport const tanKernelFunc = unaryKernelFunc(TAN);\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'webgl',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {transposeImpl} from './Transpose_impl';\nimport {transposeImplCPU as cpuTranspose} from './Transpose_impl';\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as TransposeInputs;\n    const {perm} = attrs as {} as TransposeAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const newShape: number[] = new Array(xRank);\n    for (let i = 0; i < newShape.length; i++) {\n      newShape[i] = x.shape[perm[i]];\n    }\n\n    let out: TensorInfo;\n    if (webglBackend.shouldExecuteOnCPU([x])) {\n      const xTexData = webglBackend.texData.get(x.dataId);\n      const values = xTexData.values as TypedArray;\n      const outValues = cpuTranspose(values, x.shape, x.dtype, perm, newShape);\n\n      out = webglBackend.makeTensorInfo(newShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = transposeImpl(x, perm, webglBackend);\n    }\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {uniqueImplCPU} from '../kernel_utils/shared';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function unique(\n    args:\n        {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendWebGL}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  // For now, always forward calculation to the CPU backend.\n  console.warn(\n      'WARNING: ',\n      'UI might be locked temporarily as data is being downloaded');\n  const values = backend.readSync(x.dataId);\n  const {outputValues, outputShape, indices} =\n      uniqueImplCPU(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'webgl',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {atan2Config} from './kernels/Atan2';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {cosConfig} from './kernels/Cos';\nimport {divConfig} from './kernels/Div';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {fromPixelsConfig} from './kernels/FromPixels';\nimport {identityConfig} from './kernels/Identity';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {nonMaxSuppressionV3Config} from './kernels/NonMaxSuppressionV3';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {sinConfig} from './kernels/Sin';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {tanConfig} from './kernels/Tan';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  atan2Config,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  cosConfig,\n  divConfig,\n  flipLeftRightConfig,\n  fromPixelsConfig,\n  identityConfig,\n  maxConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  nonMaxSuppressionV3Config,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  sinConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  tanConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport * from '@tensorflow/tfjs-core';\nexport * from '@tensorflow/tfjs-layers';\nexport * from '@tensorflow/tfjs-converter';\n\n// Export data api as tf.data\nimport * as data from '@tensorflow/tfjs-data';\nexport {data};\n\n// Import and register backends.\nimport '@tensorflow/tfjs-backend-cpu';\nimport '@tensorflow/tfjs-backend-webgl';\n\n// Import versions of all sub-packages.\nimport {version_core} from '@tensorflow/tfjs-core';\nimport {version_cpu} from '@tensorflow/tfjs-backend-cpu';\nimport {version_webgl} from '@tensorflow/tfjs-backend-webgl';\nimport {version_data} from '@tensorflow/tfjs-data';\nimport {version_layers} from '@tensorflow/tfjs-layers';\nimport {version_converter} from '@tensorflow/tfjs-converter';\nimport {version as version_union} from './version';\n\nexport const version = {\n  'tfjs-core': version_core,\n  'tfjs-backend-cpu': version_cpu,\n  'tfjs-backend-webgl': version_webgl,\n  'tfjs-data': version_data,\n  'tfjs-layers': version_layers,\n  'tfjs-converter': version_converter,\n  'tfjs': version_union\n};\n", "const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.age.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    obj.gender = Math.trunc(100 * data[0]) < 50 ? 'female' : 'male';\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "export default {\n  backend: 'webgl',\n  console: true,\n  face: {\n    enabled: true, // refers to detector, but since all other face modules rely on detector, it should be a global\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be blazeface-front or blazeface-back\n      inputSize: 256, // fixed value: 128 for front and tfhub and 256 for back\n      maxFaces: 10, // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10, // how many frames to go without running the bounding box detector\n      minConfidence: 0.5, // threshold for discarding a prediction\n      iouThreshold: 0.3, // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7, // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192, // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3, // empiric tuning\n      inputSize: 64, // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json',\n      inputSize: 64, // fixed value\n      skipFrames: 10,\n    },\n    gender: {\n      enabled: true,\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64, // fixed value\n      minConfidence: 0.5,\n      skipFrames: 10,\n      useGrayscale: true,\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257, // fixed value\n    outputStride: 16, // fixed value\n    maxDetections: 5,\n    scoreThreshold: 0.7,\n    nmsRadius: 20,\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256, // fixed value\n    skipFrames: 10,\n    minConfidence: 0.5,\n    iouThreshold: 0.3,\n    scoreThreshold: 0.7,\n    enlargeFactor: 1.65, // empiric tuning\n    maxHands: 2,\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    let savedWebglPackDepthwiseConvFlag;\n    if (tf.getBackend() === 'webgl') {\n      savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n      tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
-  "mappings": "oNAAA,kBCAA,kBCAA,0sDCAA,AAsBO,GAAM,IAAkB,KAClB,GAAkB,mBA0B7B,WAAoB,EAAgC,GAAhC,KAAA,QAAA,EAAgC,KAAA,UAAA,EAH5C,KAAA,KAAO,GAAI,SACX,KAAA,aAAe,EAIvB,SAAA,UAAA,IAAA,SAAI,GACF,MAAK,MAAK,KAAK,IAAI,IACjB,KAAK,UAAU,SAAS,KAAK,QAAS,GAEjC,KAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,IAAA,SAAI,EAAgB,GAClB,KAAK,eACL,KAAK,KAAK,IAAI,EAAQ,IAGxB,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,MAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,OAAA,SAAO,GACL,YAAK,eACE,KAAK,KAAK,OAAO,IAG1B,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,cAEhB,mBAqBA,cACE,SAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,WACE,MAAO,GAAkB,eAE3B,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,GAAkB,gBAE3B,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAC5C,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAE3D,MAAO,GAAkB,SAE3B,EAAA,UAAA,OAAA,WACE,MAAO,GAAkB,WAG3B,EAAA,UAAA,eAAA,WACE,MAAO,GAAkB,mBAG3B,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAG1D,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAEnD,MAAO,GAAkB,qBAG3B,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,MAAO,GAAkB,UAE3B,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,MAAO,GAAkB,iBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,GAAkB,WAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAG3B,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,uBAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAE3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAG3B,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,MAAO,GAAkB,iBAG3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,MAAO,GAAkB,WAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,MAAO,GAAkB,sBAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,GAAkB,WAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,MAAO,GAAkB,WAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAG3B,EAAA,UAAA,QAAA,SAA0B,GACxB,MAAO,GAAkB,YAG3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAG3B,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,yBAG3B,EAAA,UAAA,gBAAA,SAAgB,EAAiB,EAAkB,GAEjD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,wBAAA,SAAwB,EAAc,EAAkB,GAEtD,MAAO,GAAkB,4BAE3B,EAAA,UAAA,yBAAA,SAAyB,EAAa,EAAc,GAElD,MAAO,GAAkB,6BAE3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,EAAa,GAEtD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SAAkB,EAAc,EAAa,GAC3C,MAAO,GAAkB,sBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAAa,GAC1C,MAAO,GAAkB,sBAG3B,EAAA,UAAA,QAAA,SAA0C,EAAM,GAE9C,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,MAAO,GAAkB,QAG3B,EAAA,UAAA,UAAA,SAA4B,EAAM,GAChC,MAAO,GAAkB,cAG3B,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,MAAO,GAAkB,WAG3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GACpC,MAAO,GAAkB,cAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,mBAG3B,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,MAAO,GAAkB,2BAG3B,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,0BAG3B,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,MAAO,GAAkB,kCAG3B,EAAA,UAAA,UAAA,SACI,EAAa,EAAyB,EACtC,EAA4B,EAC5B,GACF,MAAO,GAAkB,cAG3B,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,MAAO,GAAkB,iCAG3B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EAAuB,EAC3D,EAAc,EAAe,GAC/B,MAAO,GAAkB,YAG3B,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,MAAO,GAAkB,WAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,MAAO,GAAkB,WAG3B,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,MAAO,GAAkB,sBAG3B,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAC3C,MAAO,GAAkB,iBAI3B,EAAA,UAAA,MAAA,SAAwB,EAAU,EAAsB,GACtD,MAAO,GAAkB,UAG3B,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,MAAO,GAAkB,SAG3B,EAAA,UAAA,SAAA,SAAyB,GACvB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,GAAkB,cAG3B,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAkB,aAG3B,EAAA,UAAA,QAAA,WACE,MAAO,GAAkB,YAE7B,KAEA,WAA2B,GACzB,KAAM,IAAI,OACN,IAAI,EAAU,4HCzpBpB,AAoBA,GAAM,IAA4B,0BA6BhC,WAAmB,GAAA,KAAA,OAAA,EATX,KAAA,MAAe,GACf,KAAA,aAAwD,GAExD,KAAA,SAAkB,GAOxB,KAAK,mBAGP,SAAA,UAAA,YAAA,SAAY,EAAsB,GAChC,AAAI,KAAK,UAAY,MACnB,QAAQ,KACJ,YAAY,KAAK,aAAY,0BAC7B,kCAAiC,EAAQ,MAE/C,KAAK,aAAe,EACpB,KAAK,SAAW,GAGlB,EAAA,UAAA,aAAA,SACI,EAAkB,EAClB,GAKF,GAJA,KAAK,aAAa,GAAY,CAAC,aAAY,EAAE,QAAO,GAIhD,KAAK,SAAS,IAAa,MAC7B,GAAM,GAAY,KAAK,SAAS,GAChC,QAAQ,KACJ,qCAAqC,EAAQ,KAAK,EAAS,KAC/D,KAAK,IAAI,EAAU,KAIjB,EAAA,UAAA,SAAN,SAAe,qGACb,MAAI,KAAY,MAAK,MACnB,CAAA,EAAO,KAAK,MAAM,IAGpB,GAAA,KAAK,MAAM,EAAA,EAAY,CAAA,EAAM,KAAK,aAAa,YAA/C,SAAA,GAAuB,EAAA,OACvB,CAAA,EAAO,KAAK,MAAM,UAGpB,EAAA,UAAA,IAAA,SAAI,GACF,GAAI,IAAY,MAAK,MACnB,MAAO,MAAK,MAAM,GAGpB,GAAM,GAAY,KAAK,aAAa,GACpC,GAAI,YAAqB,SACvB,KAAM,IAAI,OACN,QAAQ,EAAQ,sEAItB,YAAK,MAAM,GAAY,EAEhB,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,QAAA,SAAQ,GACN,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,OAGd,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,uCAGd,EAAA,UAAA,IAAA,SAAI,EAAkB,GACpB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,mBAAmB,EAAQ,mCAEjC,KAAK,MAAM,GAAY,EACvB,AAAI,KAAK,aAAa,GAAU,SAAW,MACzC,KAAK,aAAa,GAAU,QAAQ,IAIhC,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,yBAAyB,EAAQ,oCAEvC,MAAO,MAAK,aAAa,GAAU,gBAGrC,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,MAAQ,OAAO,OAAO,GAAI,IAGjC,EAAA,UAAA,MAAA,WACE,KAAK,MAAQ,GACb,KAAK,SAAW,GAChB,KAAK,oBAGC,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,GAAI,MAAO,MAAK,QAAW,aACvB,MAAO,MAAK,OAAO,UAAa,aAChC,MAAO,MAAK,OAAO,SAAS,QAAW,YACzC,OAGF,GAAM,GAAY,GAAe,KAAK,OAAO,SAAS,QACtD,GAAI,KAA6B,IAC/B,GAAM,GAAY,EAAU,IAA2B,MAAM,KAC7D,EAAU,QAAQ,SAAA,GACV,GAAA,GAAA,EAAA,MAAA,KAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,SAAS,GAAO,GAAW,EAAK,OAI7C,iBAE+B,GAC7B,GAAM,GAAS,GACf,SAAY,QAAQ,8BAA+B,SAAC,UAAG,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GACrD,UAAY,EAAQ,EAAE,GAAI,EAAE,IACrB,EAAE,KAAK,OAET,EAGT,YACI,EAAiC,EAAc,GACjD,EAAO,mBAAmB,IAAS,mBAAmB,GAAS,IAGjE,YAAoB,EAAkB,GAEpC,GADA,EAAQ,EAAM,cACV,IAAU,QAAU,IAAU,QAChC,MAAO,KAAU,OACZ,GAAI,GAAG,EAAE,IAAY,EAC1B,MAAO,CAAC,EAEV,KAAM,IAAI,OACN,oCAAoC,EAAK,aAAa,EAAQ,mBAYlE,MAAO,GAAA,IAGE,EAAA,IAAmB,iBACO,GACnC,EAAA,IAAM,EC9MR,AAqBA,GAAI,kBAGF,GAAI,IAAmB,MAErB,GAAI,GAAE,OACN,GAAI,MAAQ,SAAY,YACtB,EAAK,eACI,MAAQ,SAAY,YAC7B,EAAK,eACI,MAAQ,UAAa,YAC9B,EAAK,gBACI,MAAQ,OAAU,YAC3B,EAAK,SAEL,MAAM,IAAI,OAAM,kCAElB,GAAkB,EAEpB,MAAO,IAIT,cACE,GAAM,GAAK,KACX,MAAI,GAAG,YAAc,MACnB,GAAG,WAAa,GAAI,MAEf,EAAG,uBAUiB,EAAa,GACxC,GAAM,GAAY,KAClB,GAAI,EAAU,IAAI,GAChB,MAAO,GAAU,IAAI,GAErB,GAAM,GAAY,IAClB,SAAU,IAAI,EAAK,GACZ,EAAU,IAAI,MCzCZ,IAAM,MAGN,GAAO,OAGP,GAAQ,QAGR,GAAM,MAGN,GAAO,OAGP,GAAM,MAON,GAAM,MAON,GAAS,SAMT,GAAS,SAMT,GAAO,OAGP,GAAQ,QAGR,GAAO,OAGP,GAAQ,QAGR,GAAQ,QAGR,GAAU,UASV,GAAkB,kBAQlB,GAAY,YAWZ,GAAoB,oBAUpB,GAAc,cAOd,GAAiB,iBASjB,GAAc,cAOd,GAAO,OAMP,GAAO,OAGP,GAAc,cAOd,GAAU,UAGV,GAAS,SAMT,GAAS,SAUT,GAAuB,uBASvB,GAAsB,sBAUtB,GAAS,SAST,GAAyB,yBAQzB,GAAwB,wBAMxB,GAAM,MAGN,GAAO,OAGP,GAAS,SAQT,GAAgB,gBAShB,GAAe,eAOf,GAAwB,wBAWxB,GACT,sCAIS,GACT,qCAIS,GAAO,OAGP,GAAa,aAQb,GAA0B,0BAI1B,GAA2B,2BAI3B,GAAM,MAGN,GAAM,MAGN,GAAU,UAGV,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAO,OAOP,GAAgB,gBAGhB,GAAQ,QAGR,GAAW,WAGX,GAAiB,iBAOjB,GAAW,WAMX,GAAW,WAGX,GAAU,UAGV,GAAe,eAGf,GAAW,WAGX,GAAO,OAGP,GAAO,OAGP,GAAW,WAGX,GAAQ,QAGR,GAAQ,QAGR,GAAO,OAGP,GAAY,YAGZ,GAAW,WAMX,GAAM,MAGN,GAAQ,QAGR,GAAa,aAGb,GAAa,aAGb,GAAY,YAGZ,GAAa,aAMb,GAAM,MASN,GAAc,cASd,GAAM,MAON,GAAU,UAGV,GAAU,UASV,GAAkB,kBAUlB,GAAY,YAWZ,GAAoB,oBAWpB,GAAoB,oBASpB,GAAO,OAOP,GAAM,MAON,GAAU,UAGV,GAAM,MAGN,GAAW,WAGX,GAAS,SAGT,GAAW,WAGX,GAAsB,sBAStB,GAAsB,sBAUtB,GAAsB,sBAUtB,GAAW,WAGX,GAAS,SAQT,GAAQ,QAOR,GAAO,OAGP,GAAM,MAGN,GAAQ,QAGR,GAAO,OAOP,GAAQ,QAQR,GAAO,OAGP,GAAa,aAGb,GAAO,OAGP,GAAU,UAMV,GAAwB,wBAOxB,GAA4B,4BAI5B,GAAiB,iBAOjB,GAAqB,qBAGrB,GAAQ,QAGR,GAAU,UAMV,GAAQ,QAGR,GAAQ,QAGR,GAAY,YAMZ,GAAW,WAGX,GAAO,OAGP,GAAQ,QAMR,GAAM,MAGN,GAAO,OAGP,GAAO,OAGP,GAAU,UAGV,GAAW,WAGX,GAAO,OAGP,GAAM,MAON,GAAiB,iBAOjB,GAAS,SAOT,GAAU,UAMV,GAAoB,oBAGpB,GAAS,SAGT,GAAM,MAGN,GAAgB,gBAOhB,GAAe,eAaf,GAAM,MAGN,GAAO,OAGP,GAAO,OAMP,GAAO,OAOP,GAAY,YAMZ,GAAS,SAQT,GAAS,SAMT,GAAqB,qBAOrB,GAAY,YAMZ,GAAO,OAMP,GAAa,aASb,GAAmB,mBAQnB,GAAe,eAef,GAAc,cAgBd,GAAuB,uBC7zBpC,AAuBA,GAAM,IACF,GAAU,iBAAkB,WAAM,MAAA,IAAI,OACpC,GACF,GAAU,eAAgB,WAAM,MAAA,IAAI,mBAqEpC,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,MAAO,IAAe,IAAI,eAOA,GAC1B,MAAO,IAAa,IAAI,eAGW,GAInC,OAHM,GAAK,GAAe,UACpB,EAAyB,MAGvB,GAAA,GAAA,EAAA,OAAC,EAAA,EAAA,KAAM,EAAA,EAAA,MACb,GAAI,EACF,MAEK,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GACL,EAAA,EAAA,MAAA,KAAA,GACP,AAAI,IAAY,GACd,EAAO,KAAK,GAGhB,MAAO,eAcsB,GACtB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,YACb,EAAM,GAAQ,EAAY,GAChC,AAAI,GAAe,IAAI,IACrB,QAAQ,KACJ,eAAe,EAAU,iBACzB,KAAI,EAAW,4BAErB,GAAe,IAAI,EAAK,eAWO,GACxB,GAAA,GAAA,EAAA,WAEP,AAAI,GAAa,IAAI,IAGnB,CAAI,KAAM,QAAQ,UAChB,QAAQ,KAAK,gCAAgC,EAAU,MAG3D,GAAa,IAAI,EAAY,eAW3B,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,GAAI,CAAC,GAAe,IAAI,GACtB,KAAM,IAAI,OACN,eAAe,EAAU,iBACzB,KAAI,EAAW,wBAErB,GAAe,OAAO,eAIW,GACjC,GAAI,CAAC,GAAa,IAAI,GACpB,KAAM,IAAI,OACN,iBAAiB,EAAU,mCAEjC,GAAa,OAAO,eAUlB,EAA+B,GACjC,GAAM,GAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,GAAM,GACF,OAAO,OAAO,GAAI,EAAc,CAAC,YAAa,IAClD,GAAe,KAInB,YAAiB,EAAoB,GACnC,MAAU,GAAW,IAAI,ECrN3B,YAkCwB,GAMtB,OAJI,GAAU,EAAM,OAChB,EAAO,EACP,EAAQ,EAEL,EAAU,GAEf,EAAS,KAAK,SAAW,EAAW,EAEpC,IAEA,EAAO,EAAM,GACb,EAAM,GAAW,EAAM,GACvB,EAAM,GAAS,cAKG,EAAa,EAAW,GAC5C,MAAO,MAAK,IAAI,EAAK,KAAK,IAAI,EAAG,gBAGD,GAChC,MAAO,GAAM,IAAM,EAAI,EAAM,EAAM,cAGjB,GAElB,OADI,GAAM,EACD,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,GAAO,EAAI,GAEb,MAAO,eAUmB,EAAW,GACrC,GAAM,GAAI,KAAK,SACf,MAAQ,GAAI,EAAM,GAAI,GAAK,cAID,EAAe,GAEzC,OADI,GAAS,EACJ,EAAI,EAAG,EAAI,EAAE,OAAQ,KAC5B,GAAM,GAAO,OAAO,EAAE,IAAM,OAAO,EAAE,IACrC,GAAU,EAAO,EAEnB,MAAO,cAkBc,EAAe,GACpC,GAAI,CAAC,EACH,KAAM,IAAI,OAAM,MAAO,IAAQ,SAAW,EAAM,iBAKhD,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EACI,GAAY,EAAQ,GACpB,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlC,GAC5B,EACI,GAAK,KACL,WAAM,MAAA,8EAwBR,EAA0B,EAAkB,GAI9C,GAJ4B,AAAA,IAAA,QAAA,GAAA,IAAkB,AAAA,IAAA,QAAA,GAAA,IAC9C,AAAI,GAAU,MACZ,GAAS,IAEP,MAAM,QAAQ,IAAQ,GAAa,IAAQ,CAAC,EAC9C,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAAQ,EAAI,GAAI,EAAQ,OAG1B,GAAO,KAAK,GAEd,MAAO,eAcqB,GAC5B,GAAI,EAAM,SAAW,EAEnB,MAAO,GAGT,OADI,GAAO,EAAM,GACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,GAAQ,EAAM,GAEhB,MAAO,eAGqB,GAC5B,MAAO,GAAM,SAAW,cAGE,EAAgB,GAC1C,GAAI,IAAO,EACT,MAAO,GAET,GAAI,GAAM,MAAQ,GAAM,KACtB,MAAO,GAGT,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,EAAG,GACf,MAAO,GAGX,MAAO,eAGa,GACpB,MAAO,GAAI,IAAM,cAGE,GAEnB,GAAK,KAAa,MAAQ,KAExB,MAAQ,MAAa,KAAK,GAE5B,GAAI,IAAM,SACR,MAAO,GACF,GAAI,IAAM,UACf,MAAO,GAEP,GAAM,GAAM,KAAK,IAAI,EAAI,GACzB,MAAQ,GAAM,GAAM,GAAM,eAIM,GAClC,GAAM,GAAQ,KAAK,KAAK,KAAK,KAAK,IAClC,MAAO,CAAC,EAAO,KAAK,KAAK,EAAO,gBAeI,GAEpC,OADM,GAAkB,GAAI,aAAY,GAC/B,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAgB,GAAK,EAEvB,UAAQ,GACD,cAGgB,EAAW,GAClC,MAAI,IAAQ,EAAE,OACL,EAEF,EAAI,IAAI,OAAO,EAAO,EAAE,oBAI7B,EAAwB,EACxB,GADwB,MAAA,KAAA,QAAA,GAAA,SAAW,GAAoB,MAAA,KAElD,GAAI,SAAc,SAAC,EAAS,GACjC,GAAI,GAAW,EAET,EAAQ,WACZ,GAAI,KACF,IACA,OAGF,IAEA,GAAM,GAAc,EAAQ,GAE5B,GAAI,GAAc,MAAQ,GAAY,GACpC,IACA,OAEF,WAAW,EAAO,IAGpB,kBAcA,EAAiB,GAInB,OAHI,GAAY,EACZ,EAAc,GAET,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,GAAI,EAAM,IAAM,EACd,GAAa,EAAM,WACV,EAAM,KAAO,IACtB,GAAI,IAAgB,GAClB,KAAM,OACF,yCACA,oBAAmB,EAAW,YAAY,IAEhD,EAAc,UACL,EAAM,GAAK,EACpB,KAAM,OAAM,gCAAgC,EAAM,GAAE,WAAW,GAInE,GAAI,IAAgB,IAClB,GAAI,EAAO,GAAK,IAAS,EACvB,KAAM,OAAM,QAAQ,EAAI,qCAAqC,GAE/D,MAAO,GAGT,GAAI,IAAc,EAChB,KAAM,OACF,qCAAqC,EAAK,+BAGhD,GAAI,EAAO,IAAc,EACvB,KAAM,OACF,oDACA,QAAO,EAAI,MAAM,IAGvB,GAAM,GAAW,EAAM,QACvB,SAAS,GAAe,EAAO,EACxB,cAIL,EAAuB,GACzB,GAAM,GAAO,EAAM,OAGnB,SAAO,GAAQ,KAAO,EAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAK,GAAG,OAAO,GAGzD,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,CAAC,GAAQ,EAAK,IACrC,WACI,MAAA,+CAA+C,EAAI,KAAK,EAAI,SAC5D,aAAY,KAGpB,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,KACvB,WAAM,MAAA,iDACF,aAAY,KAGb,EAAK,IAAI,SAAA,GAAK,MAAA,GAAI,EAAI,EAAO,EAAI,gBAIb,EAAiB,GAS5C,OAPM,GAAqB,GACrB,EAAqB,GACrB,EAAe,GAAQ,MAAQ,MAAM,QAAQ,IAAS,EAAK,SAAW,EACtE,EAAQ,GAAQ,MAAQ,EAC1B,KACA,GAAe,EAAM,GAAO,OAC5B,EAAI,EACC,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,GAAQ,MACV,GAAI,EAAK,KAAO,GAAK,EAAM,KAAO,EAChC,KAAM,IAAI,OACN,sBAAsB,EAAC,mBAAmB,EAAM,GAAE,cAExD,AAAK,GAAK,IAAM,MAAQ,EAAK,GAAK,IAAM,EAAM,KAAO,GACnD,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAEhB,AAAI,EAAK,IAAM,GACb,IAGJ,AAAI,EAAM,KAAO,GACf,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAGlB,MAAO,CAAC,SAAQ,EAAE,SAAQ,eAIxB,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,SACnB,EAAS,GAAI,OAAgB,OAE7B,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAA+B,GACjC,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAC1B,KAAM,OAAM,oBAAoB,EAAK,4BAA4B,EAAG,kBAM7C,GAC3B,MAAO,KAAU,QAAU,IAAU,aAAe,IAAU,WAC1D,IAAU,SAAW,IAAU,qBAOL,EAAmB,GACjD,MAAI,KAAY,aAGZ,KAAY,WAAa,IAAY,aAGrC,IAAY,SAAW,IAAY,WAAa,IAAY,aALvD,GAQL,MAAY,QAAU,IAAY,oBAMX,GAC3B,MAAO,aAAa,eAAgB,YAAa,aAC7C,YAAa,wBAGa,GAC9B,GAAI,IAAU,WAAa,IAAU,QACnC,MAAO,GACF,GAAI,IAAU,YACnB,MAAO,GACF,GAAI,IAAU,OACnB,MAAO,GAEP,KAAM,IAAI,OAAM,iBAAiB,eAUA,GACnC,GAAI,GAAO,KACT,MAAO,GAET,GAAI,GAAQ,EACZ,SAAI,QAAQ,SAAA,GAAK,MAAA,IAAS,EAAE,SACrB,cAIgB,GACvB,MAAO,OAAO,IAAU,UAAY,YAAiB,oBAG7B,GACxB,MAAO,OAAO,IAAU,sBAGD,GACvB,MAAO,OAAO,IAAU,qBAGC,GACzB,MAAI,OAAM,QAAQ,GACT,GAAW,EAAO,IAEvB,YAAkB,cACb,UACE,YAAkB,aAAc,YAAkB,YACpD,QACE,GAAS,GACX,UACE,GAAS,GACX,SACE,GAAU,GACZ,OAEF,sBAGkB,GACzB,MAAO,CAAC,CAAE,IAAK,EAAE,aAAe,EAAE,MAAQ,EAAE,mBAGf,EAAc,GAC3C,OAAS,GAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,GAAI,EAAO,IAAM,EACf,MAAO,GAGX,MAAO,eAGsB,GAC7B,GAAM,GAAO,EAAM,OACnB,GAAI,EAAO,EACT,MAAO,GAKT,GAAM,GAAU,GAAI,OAAM,EAAO,GACjC,EAAQ,EAAO,GAAK,EAAM,EAAO,GACjC,OAAS,GAAI,EAAO,EAAG,GAAK,EAAG,EAAE,EAC/B,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAM,EAAI,GAE1C,MAAO,eAOL,EAAiB,GACnB,MAAI,KAAU,SACL,GAAa,GAGf,GAAa,CAAC,GAAQ,eAGF,EAAe,GAC1C,GAAI,IAAU,SACZ,KAAM,IAAI,OAAM,6CASlB,GAPA,AAAI,MAAM,QAAQ,IAChB,GAAI,GAAQ,IAGd,AAAI,KAAM,QAAQ,UAChB,GAAyB,EAAe,GAEtC,GAAmB,EAAG,GACxB,MAAO,GAET,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,QAEnB,OADM,GAAO,GAAI,YAAY,EAAe,QACnC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,AAAI,KAAK,MAAO,EAAe,MAAQ,GACrC,GAAK,GAAK,GAGd,MAAO,OAEP,MAAM,IAAI,OAAM,qBAAqB,GAIzC,YAA2B,EAAgB,EAAiB,GAC1D,GAAM,GAAM,GAAI,OAChB,GAAI,EAAM,SAAW,EAEnB,OADM,GAAI,EAAM,GACP,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,EAAE,EAAS,OAMtB,QAHM,GAAI,EAAM,GACV,EAAO,EAAM,MAAM,GACnB,EAAM,EAAK,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IACjC,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,GAAkB,EAAS,EAAI,EAAK,EAAM,GAGvD,MAAO,eAIqB,EAAiB,GAC7C,GAAI,EAAM,SAAW,EAEnB,MAAO,GAAE,GAEX,GAAM,GAAO,EAAM,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IAC5C,GAAI,IAAS,EAEX,MAAO,GAET,GAAI,IAAS,EAAE,OACb,KAAM,IAAI,OAAM,IAAI,EAAK,mCAAmC,EAAE,OAAM,KAGtE,MAAO,IAAkB,EAAG,EAAO,GAGrC,YAA4B,EAAe,GACzC,MAAQ,aAAa,eAAgB,IAAU,WAC1C,YAAa,aAAc,IAAU,SACrC,YAAa,aAAc,IAAU,mBAIxC,EAAc,GAEhB,OADM,GAAQ,GAAoB,EAAM,GAC/B,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAM,GAAK,EAEb,MAAO,eAIL,EAAc,GAChB,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,OACnB,MAAO,IAAI,YAAW,GAEtB,KAAM,IAAI,OAAM,qBAAqB,eAUrC,EAAiB,GACnB,GAAM,GAAO,EAAM,OAAO,SAAC,EAAM,GAAS,MAAA,GAAO,GAAM,GACvD,GAAI,GAAS,MAAQ,IAAU,UAC7B,MAAO,IAAc,EAAO,GAAI,cAAa,IACxC,GAAI,IAAU,QACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IACtC,GAAI,IAAU,OACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IAE3C,KAAM,IAAI,OAAM,qBAAqB,iBAgBvC,MAAO,MAAM,SAAS,kBAG2B,GACjD,EAAM,QAAQ,SAAA,GACZ,EACI,OAAO,UAAU,IAAY,GAAW,EACxC,WACI,MAAA,mEACA,WAAU,EAAK,sBAqBvB,EAAc,GAChB,MAAO,MAAM,SAAS,MAAM,EAAM,eAWP,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,SACtC,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAG,eAWL,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,SAC9C,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAO,eAYlC,EAAgB,EAAc,GAChC,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,EAAQ,GAAK,EAAK,GAE7B,MAAO,eAYL,EAAe,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,EAAQ,IACrC,GAAS,EAAK,GAAK,EAAQ,GAE7B,SAAK,EAAK,OAAS,GAAK,EACjB,22BC3xBT,AA+BA,GAAA,IAAA,WACE,WAAoB,EAAoC,GAApC,KAAA,aAAA,EAAoC,KAAA,OAAA,EACtD,AAAI,GAAU,MACZ,MAAK,OAAS,GAAI,KAItB,SAAA,UAAA,cAAA,SAAc,EAAoB,EAAwB,GAQxD,OANI,GACE,EAAsB,WAC1B,EAAU,KAEN,EAAQ,KAAK,aAAa,KAAK,cAE5B,GACP,GAAM,GAAS,EAAQ,GAGvB,EAAO,OAAO,KAAK,SAAA,GACjB,GAA0B,EAAY,EAAO,MAAO,MAL/C,EAAI,EAAG,EAAI,EAAQ,OAAQ,MAA3B,GAST,GAAM,GAAgB,CACpB,WAAU,EACV,QAAO,EACP,OAAM,EACN,OAAQ,EAAM,KAAK,SAAA,GAAU,MAAA,GAAO,WACpC,UAAW,EAAM,KACb,SAAA,GAAU,MAAA,GAAO,qBAAuB,KACpC,EAAO,sBACP,MAEV,MAAO,IAGT,EAAA,UAAA,iBAAA,SAAiB,GAAjB,GAAA,GAAA,KACS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,OAAQ,EAAA,EAAA,OAAQ,EAAA,EAAA,UAE5C,EAAQ,QAAQ,SAAA,GACd,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,IAAY,KAAK,SAAA,GACnD,EAAK,OAAO,iBACR,EAAY,EAAQ,EAAe,GAAI,EAAe,GAAI,EAC1D,EAAe,SAI3B,iBAGI,EAAsB,EAAU,GAClC,GAAI,IAAU,UAEZ,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAE1B,eAAQ,KAAK,SAAS,EAAG,sBAAsB,EAAU,KAClD,GAGX,MAAO,GAGT,GAAA,IAAA,WAAA,cACE,SAAA,UAAA,iBAAA,SACI,EAAc,EAAgB,EAC9B,EAAgC,EAChC,GACF,GAAM,GAAO,MAAO,IAAW,SAAW,GAAiB,EAAM,KAAM,GAC7B,EAAO,MAC3C,EAAa,GAAc,EAAM,IACjC,EAAO,EAAO,KACd,EAAO,EAAO,KACd,EAAQ,GAAc,EAAO,MAAM,WAAY,IACjD,EAAyB,GAE7B,OAAW,KAAQ,IACjB,GAAM,GAAQ,EAAO,GACrB,GAAI,GAAS,MAGX,GAAM,GAAa,EAAM,OAAS,EAAO,MACnC,EAAY,EAAW,OAC7B,GACO,EAAI,KAAK,EAAS,KAAK,GAAY,EAAI,EAAa,IAAE,KAIjE,QAAQ,IACJ,KAAK,EAAU,MAAO,EAAI,MAAO,EAAI,KAAK,EAAK,MAAO,EAAI,MACtD,EAAsB,MAAO,EACjC,mBAAoB,YAAa,aAAc,gBAC/C,eAAgB,qBAExB,KCjIA,YA4CI,EAAkB,EAAc,GAKlC,OAFM,GAA8C,GAC9C,EAA0C,GACvC,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,EAAa,EAAG,GAAG,IAAM,GAG3B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GACZ,EAAa,EAAK,OACxB,OAAW,KAAa,IAItB,OAHM,GAAQ,EAAW,GAErB,EAAgB,GACX,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAa,EAAM,KACrB,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAa,EAAO,IAAM,KACzD,EAAgB,GAChB,EAAW,EAAK,IAAM,GACtB,MAIJ,GAAI,EACF,OAMN,GAAM,GAAgD,GACtD,EAAe,EAAE,IAAM,GAGvB,OAFM,GAAwC,GAErC,EAAI,EAAK,OAAS,EAAG,GAAK,EAAG,IAKpC,OAJM,GAAO,EAAK,GACZ,EAAa,EAAK,OAGf,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,IACvC,GAAI,EAAe,EAAK,QAAQ,GAAG,KACjC,OAAW,KAAa,GACtB,EAAe,EAAW,GAAW,IAAM,GAC3C,EAAS,EAAK,IAAM,GAEtB,MAON,OADM,GAA2B,GACxB,EAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GAElB,GAAI,EAAW,EAAK,KAAO,EAAS,EAAK,KAEvC,GAAM,GAA8C,GACpD,OAAW,KAAa,GAAK,QAC3B,GAAM,GAAY,EAAK,OAAO,GAC9B,AAAI,EAAa,EAAU,KACzB,GAAa,GAAa,GAK9B,GAAM,GAAa,OAAO,OAAO,GAAI,GACrC,EAAW,OAAS,EACpB,EAAW,QAAU,EAAK,QAE1B,EAAa,KAAK,IAItB,MAAO,eAWL,EACA,EAA0B,EAC1B,GAEF,mBAAS,GACP,GAAM,GAAO,EAAa,GAEpB,EAAgB,GAYtB,GAXA,EAAK,QAAQ,QAAQ,SAAA,GACnB,GAAM,GAAa,EAA6B,EAAE,IAClD,AAAI,GAAc,KAChB,EAAI,KAAK,GAIT,EAAI,KAAK,QAIT,EAAK,UAAY,KACnB,KAAM,IAAI,OACN,wDACA,QAAO,EAAK,WAAU,MAI5B,GAAM,GAAiB,EAAK,SAAS,cAE1B,GACT,GAAI,CAAE,KAAa,IACjB,KAAM,IAAI,OACN,iCAAiC,EAAS,KAC1C,+BAA8B,OAAO,KAAK,GAAe,MAI/D,GAAM,GAAK,EAAK,WAAM,MAAA,GAAe,OACrC,GAAI,EAAG,QAAU,UACf,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BAChB,GAAS,wCAAwC,EAAG,MAAK,MAElE,GAAM,GAAI,EAAK,OAAO,GACtB,GAAI,CAAC,GAAiB,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BACnB,KAAI,EAAS,gBAAgB,EAAG,MAAK,4BACrC,4BAA2B,EAAE,MAAK,MAGxC,GAAI,EAA6B,EAAE,KAAO,KACxC,EAA6B,EAAE,IAAM,OAErC,GAAM,GAAc,EAA6B,EAAE,IACnD,EAA6B,EAAE,IAAM,EAAI,EAAa,GACtD,EAAY,YA7BhB,OAAW,KAAa,GAAK,SAAlB,IAxBJ,EAAI,EAAa,OAAS,EAAG,GAAK,EAAG,MAArC,GCvIX,AAqBA,GAAM,IAAwB,GAExB,GAA6B,EAE7B,GAAwB,cAG1B,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAU,GAAe,GACzB,EAAY,GAAwB,EAAM,EAAO,EAAO,GACxD,EAAO,EAAM,OACb,EAAY,GAAkB,EAAM,EAAO,EAAO,EAAS,GAC3D,EAAQ,CAAC,UACf,MAAI,IACF,GAAM,KAAK,YAAY,GACvB,EAAM,KAAK,WAAW,GACtB,EAAM,KAAK,aAAa,EAAK,KAC7B,EAAM,KAAK,cAEb,EAAM,KAAK,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,IAAG,KAAK;IACxC,EAAM,KAAK;GAGpB,YACI,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAI,GAAc,GAClB,EAAU,EAAQ,EAAQ,OAAS,GACnC,EAAY,GAAI,OAAM,GAAS,KAAK,GACpC,EAAO,EAAM,OACb,EACF,IAAU,YAAc,GAAoB,GAAQ,EAExD,GAAI,EAAO,EACT,OAAS,GAAM,EAAG,EAAM,EAAI,EAAS,IAEnC,OADM,GAAS,EAAM,EACZ,EAAI,EAAG,EAAI,EAAS,IAC3B,EAAU,GAAK,KAAK,IAChB,EAAU,GACV,GAAY,EAAe,EAAS,GAAI,EAAG,GAAO,QAI5D,MAAO,GAGT,YACI,EAAqC,EAAa,GACpD,GAAI,GACJ,MAAI,OAAM,QAAQ,GAChB,EAAY,WAAW,EAAI,GAAG,QAAQ,KAAuB,MACtD,YAAW,EAAI,GAAG,QAAQ,KAAuB,KACnD,AAAI,GAAS,GAClB,EAAS,IAAI,EAAG,IACX,AAAI,IAAU,OACnB,EAAS,GAAgB,GAEzB,EAAS,WAAW,EAAI,QAAQ,KAAwB,WAGnD,GAAS,EAAQ,GAG1B,YAAyB,GACvB,MAAO,KAAM,EAAI,QAAU,OAG7B,YACI,EAA2B,EAAiB,EAC5C,EAAmB,EAAqB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAM,GAAoB,IAAU,YAAc,EAAI,EAEhD,EAAO,EAAM,GACb,EAAO,EAAM,OACnB,GAAI,IAAS,GACX,GAAI,IAAU,aACZ,GAAM,GAAe,GAAoB,GACzC,MAAO,CAAC,GAAY,EAAa,GAAI,EAAG,IAE1C,MAAI,KAAU,OACL,CAAC,GAAgB,EAAK,KAExB,CAAC,EAAK,GAAG,YAGlB,GAAI,IAAS,GACX,GAAI,EAAO,IACT,GAAM,GAAgB,GAA6B,EAE/C,EAAY,MAAM,KAClB,EAAK,MAAM,EAAG,IACd,EAAW,MAAM,KAAqC,EAAK,MAC1D,GAAO,IAA8B,EACtC,EAAO,IACX,MAAI,KAAU,aACZ,GAAY,GAAoB,GAChC,EAAW,GAAoB,IAE1B,CACL,IACA,EAAU,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAChD,KAAK,MACV,UACA,EACK,IACG,SAAC,EAAG,GAAM,MAAA,IACN,EAAG,EAAU,EAAO,GAA6B,GAAI,KAC5D,KAAK,MACV,KAGJ,GAAM,GACF,IAAU,YAAc,GAAoB,GACpB,MAAM,KAAoB,GAEtD,MAAO,CACL,IACA,EAAY,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAClD,KAAK,MACV,KAKJ,GAAM,GAAW,EAAM,MAAM,GACvB,EAAa,EAAQ,MAAM,GAC3B,EAAS,EAAQ,GAAK,EACtB,EAAkB,GACxB,GAAI,EAAO,IACT,OAAS,GAAI,EAAG,EAAI,GAA4B,KAC9C,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,KAEN,EAAM,KAAK,OACX,OAAS,GAAI,EAAO,GAA4B,EAAI,EAAM,KACxD,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,SAGnB,QAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,IAGrB,GAAM,GAAM,IAAS,EAAI,IAAM,GAC/B,EAAM,GAAK,IAAM,EAAM,GAAK,EAC5B,OAAS,GAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,EAAM,GAAK,IAAM,EAAM,GAAK,EAG9B,OADI,GAAa;EACR,EAAI,EAAG,EAAI,EAAM,IACxB,GAAc;EAEhB,SAAM,EAAM,OAAS,GACjB,IAAM,EAAM,EAAM,OAAS,GAAK,IAAO,GAAS,GAAK,GAClD,EAGT,YAA6B,GAG3B,OADM,GAAyC,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,GAAK,EACpC,EAAc,KAAK,CAAC,EAAK,GAAI,EAAK,EAAI,KAExC,MAAO,GCnMT,kBA6CE,WAAY,EAA2B,EAAU,GAAjD,GAAA,GAAA,KAIE,GAJqC,KAAA,MAAA,EACrC,KAAK,MAAQ,EAAM,QACnB,KAAK,KAAO,GAAmB,GAE3B,GAAU,MACZ,GAAM,GAAI,EAAO,OACjB,EACI,IAAM,KAAK,KACX,WAAM,MAAA,qBAAqB,EAAC,6BACxB,2BAA0B,EAAK,KAAI,QAE7C,GAAI,IAAU,YACZ,KAAM,IAAI,OACN,8JAIN,KAAK,OAAS,GAAU,GAAuB,EAAO,KAAK,MAC3D,KAAK,QAAU,GAAe,GAWhC,SAAA,UAAA,IAAA,SAAI,UAAJ,GAAA,KAA8B,EAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GAC5B,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAEV,EACI,EAAK,SAAW,KAAK,KACrB,WAAM,MAAA,uCAAuC,EAAK,OAAM,UACpD,oBAAmB,EAAK,KAAI,OAEpC,GAAM,GAAQ,KAAK,WAAW,GAC9B,KAAK,OAAO,GAAS,GAUvB,EAAA,UAAA,IAAA,kBAAI,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACF,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAGV,OADI,GAAI,EACU,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,EAAM,GAAK,GAAO,KAAK,MAAM,IAC/B,GAAM,GAAM,qCAAqC,EAAI,KACjD,mBAAkB,KAAK,OAC3B,KAAM,IAAI,OAAM,GAElB,IAGF,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,MAAK,OAAO,IAGrB,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,IAGT,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,KAAK,MAAM,QACnC,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,KAAK,QAAQ,IAC1C,GAAS,EAAK,GAAK,KAAK,QAAQ,GAElC,SAAK,EAAK,OAAS,GAAK,EACjB,GAGT,OAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAY,WAAW,KAAK,OAAQ,KAAK,MAAO,KAAK,QAGhE,KA8BI,GAAiC,KAEjC,GAAuB,iBAaM,GAC/B,GAAY,cAQe,GAC3B,GAAY,mBA4DZ,WAAY,EAAoB,EAAiB,EAAgB,GAXjE,KAAA,KAAO,GAoIG,KAAA,mBAAqB,GAxH7B,KAAK,MAAQ,EAAM,QACnB,KAAK,MAAQ,GAAS,UACtB,KAAK,KAAO,GAAmB,GAC/B,KAAK,QAAU,GAAe,GAC9B,KAAK,OAAS,EACd,KAAK,GAAK,EACV,KAAK,SAAY,KAAK,KAAO,EAAI,KAAK,KAAK,WAAa,SAG1D,cAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQd,EAAA,UAAA,OAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAU,OAAO,KAAK,MAAO,KAAK,MAAY,UAOvD,EAAA,UAAA,WAAA,WACE,MAAO,IAAU,OAAO,KAAK,MAAO,KAAK,MAAY,KAAK,aAStD,EAAA,UAAA,MAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAc,KAAK,MAAO,UASnC,EAAA,UAAA,UAAA,WACE,MAAO,IAAc,KAAK,MAAO,KAAK,aASlC,EAAA,UAAA,KAAN,mHACE,MAAK,kBACC,EAAO,KAAY,KAAK,KAAK,QAC/B,KAAK,QAAU,SACH,CAAA,EAAM,GADlB,CAAA,EAAA,UACI,EAAQ,EAAA,OACd,IACE,MAAA,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,IAAkB,YACxC,GACA,KAAM,IAAI,OACN,kHAIR,MAAA,CAAA,EAAO,SAST,EAAA,UAAA,SAAA,WACE,KAAK,kBACL,GAAM,GAAO,KAAY,SAAS,KAAK,QACvC,GAAI,KAAK,QAAU,SACjB,IACE,MAAQ,GAAsB,IAAI,SAAA,GAAK,MAAA,IAAkB,WAEzD,GACA,KAAM,IAAI,OACN,iGAIR,MAAO,IAIH,EAAA,UAAA,MAAN,2GACE,YAAK,kBACQ,CAAA,EAAM,KAAY,KAAK,KAAK,gBACzC,MADM,GAAO,EAAA,OACT,KAAK,QAAU,SACjB,CAAA,EAAO,GAEP,CAAA,EAAO,GAAI,YAAY,EAAoB,eAS/C,EAAA,UAAA,QAAA,WACE,GAAI,KAAK,WACP,OAEF,KAAY,cAAc,MAC1B,KAAK,mBAAqB,IAI5B,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,oDAGd,EAAA,UAAA,gBAAA,WACE,GAAI,KAAK,WACP,KAAM,IAAI,OAAM,wBAYpB,EAAA,UAAA,MAAA,SAAM,GAAA,MAAA,KAAA,QAAA,GAAA,IACG,GAAU,MAAM,KAAM,IAO/B,EAAA,UAAA,MAAA,WACE,YAAK,kBACE,GAAU,MAAM,OAQzB,EAAA,UAAA,SAAA,SAAS,GAAA,AAAA,IAAA,QAAA,GAAA,IACP,GAAM,GAAO,KAAK,WAClB,MAAO,IAAe,EAAM,KAAK,MAAO,KAAK,MAAO,IAGtD,EAAA,UAAA,KAAA,SAAqB,GACnB,YAAK,kBACE,GAAU,KAAK,KAAW,IAEnC,EAAA,UAAA,SAAA,SAAS,EAAkB,EAAe,GAAjC,MAAA,KAAA,QAAA,GAAA,IACP,KAAK,kBACE,KAAY,aAAa,KAAM,EAAW,EAAM,IAG3D,KACA,OAAO,eAAe,EAAQ,OAAO,YAAa,CAChD,MAAO,SAAC,GAMN,MAAO,CAAC,CAAC,GAAY,EAAS,MAAQ,MAAQ,EAAS,UAAY,MAC/D,EAAS,iBAAmB,2BAoCiB,GAAA,EAAA,GAGnD,WACI,EAAgC,EAAoB,EACpD,GAFJ,GAAA,GAGE,EAAA,KAAA,KACI,EAAa,MAAO,EAAa,MAAO,EAAa,OAAQ,IAAS,KAHxC,SAAA,UAAA,EAIlC,EAAK,KAAO,IAWd,SAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAS,QAAU,KAAK,MAC1B,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,GAAI,CAAC,GAAiB,EAAS,MAAO,KAAK,OACzC,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,KAAY,cAAc,MAC1B,KAAK,OAAS,EAAS,OACvB,KAAY,OAAO,KAAM,OAG3B,EAAA,UAAA,QAAA,WACE,KAAY,gBAAgB,MAC5B,KAAK,mBAAqB,IAE9B,GAvCqD,GAyCrD,OAAO,eAAe,GAAU,OAAO,YAAa,CAClD,MAAO,SAAC,GACN,MAAO,aAAoB,IAAU,EAAS,QAAU,MACpD,EAAS,iBAAkB,aCnhBnC,AAgEA,AAAA,UAAY,GACV,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,OAPU,EAAA,MAAA,GAAA,KAAI,KAqBhB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,QACA,EAAA,UAAA,cAJG,IAAA,IAAiB,KAOtB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,OACA,EAAA,UAAA,cAJG,IAAA,IAAgB,KAOrB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,UACA,EAAA,KAAA,UACA,EAAA,UAAA,cAJG,IAAA,IAAmB,KAOxB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,YACA,EAAA,MAAA,YACA,EAAA,KAAA,YACA,EAAA,UAAA,cAJG,IAAA,IAAqB,KAO1B,GAAM,IAAgB,CACpB,QAAW,GACX,MAAS,GACT,KAAQ,GACR,UAAa,gBAGY,EAAiB,GAC1C,GAAI,IAAU,UAAY,IAAU,UAClC,GAAI,IAAU,UAAY,IAAU,SAClC,MAAO,SAET,KAAM,IAAI,OAAM,kBAAkB,EAAK,SAAS,GAElD,MAAO,IAAc,GAAO,eAIH,GACzB,MAAO,IAAW,EAAM,SCpI1B,YAsBiD,EAAM,GACrD,GAAI,EAAE,QAAU,EAAE,MAChB,MAAO,CAAC,EAAG,GAEb,GAAM,GAAQ,GAAW,EAAE,MAAO,EAAE,OACpC,MAAO,CAAC,EAAE,KAAK,GAAQ,EAAE,KAAK,gBAGC,EAAW,GAC1C,EACI,EAAE,QAAU,EAAE,MACd,WAAM,MAAA,2BAA2B,EAAE,MAAK,QACpC,YAAW,EAAE,MAAK,oCAGG,EAAgB,GAC7C,MAAO,GAAW,KAAK,SAAA,GAAK,MAAA,GAAE,KAAO,EAAO,iBAeR,GACpC,GAAM,GAAiB,GACjB,EAAO,GAAI,KACjB,UAAoB,EAAQ,EAAM,GAC3B,EAGT,YACI,EAA4B,EAAgB,GAC9C,GAAI,GAAa,KACf,OAEF,GAAI,YAAqB,IACvB,EAAK,KAAK,GACV,OAEF,GAAI,CAAC,GAAW,GACd,OAGF,GAAM,GAAW,EACjB,OAAW,KAAK,IACd,GAAM,GAAM,EAAS,GACrB,AAAK,EAAK,IAAI,IACZ,GAAK,IAAI,GACT,GAAoB,EAAK,EAAM,KAMrC,YAAoB,GAClB,MAAO,OAAM,QAAQ,IAAQ,MAAO,IAAQ,kHCrF9C,AAmFA,GAAA,IAAA,WAAA,aAEE,KAAA,oBAAwC,GAExC,KAAA,eAAiB,EACjB,KAAA,SAAW,EACX,KAAA,WAAa,EACb,KAAA,iBAAmB,EACnB,KAAA,eAAiB,EAMjB,KAAA,cAAgB,EAGhB,KAAA,YAAc,EAId,KAAA,WAA2B,GAK3B,KAAA,kBAA8B,GAC9B,KAAA,YAAc,EAEd,KAAA,WAAa,GAAI,SAQjB,KAAA,UAAY,GACZ,KAAA,cACI,CAAC,SAAU,EAAG,WAAY,EAAG,UAAW,EAAG,QAAS,GAAI,OAAQ,MAEpE,SAAA,UAAA,QAAA,WACE,OAAW,KAAgB,MAAK,oBAC9B,KAAK,oBAAoB,GAAc,WAG7C,KAEA,GAAA,WAgBE,WAAmB,GAAA,KAAA,IAAA,EAbnB,KAAA,SAA0C,GAC1C,KAAA,gBAKI,GAKI,KAAA,qBAAuB,EAG7B,KAAK,MAAQ,GAAI,IAGb,SAAA,UAAA,MAAN,iHACE,GAAI,KAAK,oBAAsB,KAC7B,MAAA,CAAA,EAAO,KAAK,mBAAmB,KAAK,eAEtC,GAAI,KAAK,iBAAmB,KAC1B,MAAA,CAAA,GAEI,EAAiB,KAAK,oBAEnB,EAAI,yBAAG,GAAI,EAAe,OAC3B,GAAc,EAAe,GACnB,CAAA,EAAM,KAAK,kBAAkB,GAAa,UAFnB,CAAA,EAAA,gBAEjC,GAAU,EAAA,OACZ,EACF,CAAA,EAAM,KAAK,WAAW,IADpB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,UALuC,wBAS3C,KAAM,IAAI,OACN,gFAIN,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,GAAI,KAAK,oBAAsB,KAC7B,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,uHAIlC,GAAI,KAAK,iBAAmB,MACpB,GAAA,GAAA,KAAA,kCAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UACb,GAAI,EACF,KAAM,IAAI,OACN,iCAAiC,EAAI,uHAI3C,KAAK,WAAW,GAElB,MAAO,MAAK,iDAGd,EAAA,UAAA,aAAA,WACE,MAAO,QAAO,KAAK,KAAK,kBAG1B,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,CAAE,KAAe,MAAK,UAGxB,GAAI,IAAe,MAAK,iBACf,GAAA,GAAA,KAAA,kBAAA,GAAA,UACP,GAAI,EAEF,MAAO,UAGT,OAAO,MAGX,MAAO,MAAK,SAAS,IAGvB,EAAA,UAAA,mBAAA,SAAmB,GAEjB,MAAM,KAAe,MAAK,gBAGnB,KAAK,gBAAgB,GAAa,QAFhC,MAKX,EAAA,UAAA,gBAAA,SACI,EACA,EACA,GACF,MADE,AAAA,KAAA,QAAA,GAAA,GACE,IAAe,MAAK,gBACtB,SAAQ,KACD,EAAW,sEAEX,IAET,MAAK,gBAAgB,GAAe,CAAC,QAAO,EAAE,SAAQ,GAC/C,KAGH,EAAA,UAAA,WAAN,SAAiB,2GACf,GAAI,KAAK,gBAAgB,IAAgB,KACvC,KAAM,IAAI,OAAM,iBAAiB,EAAW,iCAE9C,MAAK,YAAc,EACf,KAAK,SAAS,IAAgB,KAChC,MAAK,gBAAkB,KACjB,EAAuB,KAAK,kBAAkB,GAA7C,EAAO,EAAA,QAAE,EAAS,EAAA,UACV,EAAY,CAAA,EAAM,GAAlB,CAAA,EAAA,IAHb,CAAA,EAAA,UAGyB,SAAA,EAAA,oBAAgB,EAAA,mBAC3C,GADM,EAAM,EACR,CAAC,EACH,MAAA,CAAA,EAAO,qBAGX,YAAK,gBAAkB,KAAK,SAAS,GACrC,KAAK,yBAEL,KAAK,SAAW,GAAI,IAAS,KAAK,iBAElC,CAAA,EAAO,UAGD,EAAA,UAAA,uBAAR,WAAA,GAAA,GAAA,KACQ,EAAU,GAAqB,KAAK,aAC1C,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,WAAa,MACtB,EAAO,UAAU,EAAK,oBAKpB,EAAA,UAAA,yBAAR,SAAiC,GAAjC,GAAA,GAAA,KACQ,EAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,aAAe,MACxB,EAAO,YAAY,EAAK,SAAS,OAW/B,EAAA,UAAA,kBAAR,SAA0B,GAA1B,GAAA,GAAA,KAEQ,EAAuB,KAAK,gBAAgB,GAClD,GAAI,GAAwB,KAC1B,KAAM,IAAI,OACN,6BAA6B,EAAW,4BAG9C,IACE,GAAM,GAAU,EAAqB,UAMrC,GAAI,GAAW,CAAE,aAAmB,MAC7B,MAAO,GAAQ,MAAS,YAC7B,GAAM,GAAY,EAAE,KAAK,qBACnB,EACF,EACK,KAAK,SAAA,GAEJ,MAAI,GAAY,EAAK,qBACZ,GAET,GAAK,SAAS,GAAe,EAC7B,EAAK,mBAAqB,KACnB,MAER,MAAM,SAAA,GAEL,MAAI,GAAY,EAAK,sBAGrB,GAAK,mBAAqB,KAC1B,QAAQ,KACJ,6BAA6B,EAAW,WAC5C,QAAQ,KAAK,EAAI,OAAS,EAAI,UACvB,KAEjB,YAAK,mBAAqB,EACnB,CAAC,QAAO,EAAE,UAAW,QAE5B,aAAK,SAAS,GAAe,EACtB,CAAC,QAAS,GAAM,UAAW,UAE7B,GACP,eAAQ,KAAK,6BAA6B,EAAW,WACrD,QAAQ,KAAK,EAAI,OAAS,EAAI,SACvB,CAAC,QAAS,GAAO,UAAW,MAIvC,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAE,KAAe,MAAK,iBACxB,KAAM,IAAI,OAAS,EAAW,kCAEhC,AAAI,KAAK,cAAgB,GAAe,KAAK,oBAAsB,MAGjE,KAAK,uBAGP,AAAI,IAAe,MAAK,UACtB,MAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,IAGvB,MAAO,MAAK,gBAAgB,GAG5B,AAAI,KAAK,cAAgB,GACvB,MAAK,mBAAqB,KAC1B,KAAK,YAAc,KACnB,KAAK,gBAAkB,OAInB,EAAA,UAAA,kBAAR,WAAA,GAAA,GAAA,KACE,GAAI,OAAO,KAAK,KAAK,iBAAiB,SAAW,EAC/C,KAAM,IAAI,OAAM,iCAElB,MAAO,QAAO,KAAK,KAAK,iBAAiB,KAAK,SAAC,EAAW,GAExD,MAAO,GAAK,gBAAgB,GAAG,SAC3B,EAAK,gBAAgB,GAAG,YAIxB,EAAA,UAAA,gCAAR,WAIE,OAFM,GAAiB,KAAK,oBAEnB,EAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAc,EAAe,GAC7B,EAAA,KAAA,kBAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAChB,GAAI,GAAa,EACf,MAAO,CAAC,KAAM,EAAa,UAAS,GAGxC,KAAM,IAAI,OACN,2EAIN,EAAA,UAAA,SAAA,SAAS,EAAwB,GAC/B,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAa,EAAK,QAClB,EAAS,KAAK,SAAS,GAG7B,EAAW,YAAY,GACvB,EAAK,QAAU,EACf,EAAQ,KAAK,EAAQ,EAAQ,EAAK,MAAO,EAAK,OAC9C,AAAI,KAAK,0BAGP,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,MAIvE,EAAA,UAAA,KAAA,SAAgC,EAA6B,GAA7D,GAAA,GAAA,KAEM,EAAe,KACnB,GAAI,GAAM,MAER,GAAI,MAAO,IAAa,WACtB,KAAM,IAAI,OAAM,uCAElB,EAAK,OAGL,GAAI,MAAO,IAAa,UAAY,CAAE,aAAoB,SACxD,KAAM,IAAI,OACN,kFAGN,GAAI,MAAO,IAAO,WAChB,KAAM,IAAI,OACN,kFAGN,EAAO,EAIT,GAAI,GACJ,MAAO,MAAK,UACR,WAAM,MAAA,GAAK,WAAW,IAAO,WAAM,MAAA,GAAK,SAAS,IAAS,WACxD,SAAS,IACT,AAAI,YAAkB,UACpB,QAAQ,MAAM,2CAET,KAIP,EAAA,UAAA,UAAR,SAAqB,EAAmB,EAAiB,GACvD,IACA,IACE,GAAM,GAAM,IACZ,WACO,QACA,GACP,UACM,IAKF,EAAA,UAAA,aAAR,WACE,MAAO,GAAO,gBAIR,EAAA,UAAA,eAAR,WACE,MAAO,GAAO,kBAYR,EAAA,UAAA,MAAR,SAAc,GACZ,GAAM,GAAI,KAAK,qBAAqB,EAAE,OAAQ,EAAE,MAAO,EAAE,OACnD,EAAS,CAAC,EAAC,GACX,EAAO,SAAC,GAAe,MAAC,CAC5B,EAAG,WACD,GAAM,GAAQ,UACR,EAAa,CAAC,EAAG,GACjB,EAAQ,CAAC,MAAK,GAEpB,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAC5B,EAAoC,KAAiB,GACrD,MAGF,EAAkB,GACxB,YAAK,YAAY,KAAK,MAAM,YAAY,KAAM,EAAQ,CAAC,GAAI,EAAM,EAAO,IACjE,GAgBT,EAAA,UAAA,UAAA,SACI,EAAoB,EAAwB,EAC5C,EAAyB,GAC3B,GAAM,GAAoB,KACpB,EAAsB,KAI5B,MAAO,MAAK,cACR,EAAa,EAAQ,EAAe,EAAY,EAAO,EACvD,IAGE,EAAA,UAAA,uBAAR,WACE,MAAO,MAAK,IAAI,QAAQ,YAGlB,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAkB,KAAK,QAAQ,aAGjC,EAAmB,EACvB,EAAS,QAAQ,SAAA,GAGf,GAAqB,EAAK,QAAU,YAAc,EAAI,IAQxD,GAAM,GACF,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,GACjE,EACF,EAAkB,EAAmB,EAAmB,EAC5D,GAAI,EAAgB,EAClB,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,iCAC5B,KAAI,EAAa,6BAA6B,EAAU,OAQhE,EAAA,UAAA,cAAA,SACI,EAA6B,EAC7B,EACA,EAAqB,EAAsB,EAC3C,GAJJ,GAAA,GAAA,KAKM,EACA,EAAkB,GAChB,EAAW,KAAK,WACtB,AAAI,GAAc,MAChB,GACI,KAAK,MAAM,aAAe,KAAO,KAAK,MAAM,YAAY,KAAO,IAGrE,GAAM,GAAoB,KAAK,MAAM,SAC/B,EAAqB,KAAK,MAAM,WAEtC,AAAI,KAAK,0BACP,KAAK,MAAM,kBAAkB,KAAK,GAGpC,GAAI,GACE,EAAS,GAAU,EAAY,KAAK,aACtC,EACJ,GAAI,GAAU,KACZ,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAO,WAAW,CAAC,OAAM,EAAE,MAAK,EAAE,QAAS,EAAK,UACtD,GAAM,GAAW,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC7C,AAAI,EAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAE3D,GAAM,GAAa,EAAS,IACxB,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,MACb,MAAA,GAAK,qBAAqB,EAAQ,EAAO,KAMjD,GAAI,GACF,GAAI,GACA,EAAK,sBAAsB,EAAY,EAAQ,GACnD,GAAI,GAAiB,MAKnB,AAAI,GAAiB,MACnB,GAAgB,IAElB,GAAM,GAAa,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAC7D,EAAiB,IAAgB,IAAI,QAAQ,OAAO,GAEtD,EAAQ,EAAK,2BAA2B,GAE1C,MAAO,SAGT,GAAM,GAAyB,SAAC,GAI9B,GAAI,CAAC,EACH,OAEF,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,OAGrD,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAK,KAAK,WAAM,MAAA,GAAY,EAAK,QAAS,KAChD,GAAM,GAAQ,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC1C,MAAI,GAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAEpD,GAKX,GAAI,GACJ,YAAK,UACD,WAAM,MAAA,GAAK,MAAM,eAAe,WAAM,MAAA,GAAK,MAAM,eAAe,WAC9D,AAAI,CAAC,EAAK,IAAI,QAAQ,UAAY,CAAC,EAAK,MAAM,UAC5C,EAAU,IAEV,GAAgB,EAAK,SAAS,cAC1B,EAAY,EAAQ,WAAM,MAAA,OAC9B,AAAI,EAAK,IAAI,QAAQ,UACnB,EAAK,SAAS,iBAAiB,GAEjC,EAAU,EAAc,WAIhC,AAAI,GACF,KAAK,YACD,EAAY,EAAQ,EAAS,EAAe,EAAO,GAGzD,AAAI,KAAK,MAAM,WACb,KAAK,MAAM,cAAc,QAAQ,KAAK,CACpC,KAAM,EACN,WAAY,KAAK,MAAM,SAAW,EAClC,mBAAoB,KAAK,MAAM,SAC/B,aAAc,KAAK,MAAM,WAAa,EACtC,qBAAsB,KAAK,MAAM,WACjC,YAAa,OAAO,KAAK,GAAQ,IAC7B,SAAA,GAAO,MAAA,GAAO,IAAQ,KAAO,EAAO,GAAK,MAAQ,OACrD,aAAc,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,QACvC,aAAc,EAAc,OAC5B,UAAW,EAAc,YAGrB,MAAM,QAAQ,GAAO,EAAU,EAAQ,IAQzC,EAAA,UAAA,2BAAR,SAAmC,GAAnC,GAAA,GAAA,KACQ,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,MACzD,MAAO,IAaD,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAa,GAAY,GAC/B,GAAI,GAAc,MAChB,GAAM,GAAyB,EAAW,cAAgB,GACpD,EAA2B,EAAW,eAAiB,GAIzD,EAAkB,OACtB,AAAI,EAAW,cACb,GACI,MAAM,QAAQ,GACd,WAAM,MAAA,2DAEV,EAAqB,OAAO,KAAK,GAAQ,IAAI,SAAC,GAAQ,MAAA,GAAO,MAE7D,EAAqB,EAAa,IAAI,SAAC,GAAc,MAAA,GAAO,KAG9D,GAAM,GACF,EAAQ,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAE3C,MAAO,GAAmB,OAAO,GAInC,MAAO,OAQT,EAAA,UAAA,WAAA,SACI,EAAoB,EAAiB,EACrC,GACF,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,iDAElB,EAAQ,GAAS,UACjB,EAAU,GAAW,KAAK,QAC1B,GAAI,GAAc,EAClB,AAAI,IAAU,UAAY,GAAc,EAAO,KAC7C,GAAe,EAAoB,IAAI,SAAA,GAAK,MAAA,IAAkB,MAEhE,GAAM,GAAS,EAAQ,MAAM,EAAa,EAAO,GAC3C,EAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAIhD,GAHA,KAAK,OAAO,EAAG,GAGX,IAAU,UACZ,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAW,GAAqB,GACtC,KAAK,MAAM,UAAY,EAAW,EAAK,MACvC,EAAK,MAAQ,EAEf,MAAO,IAQT,EAAA,UAAA,qBAAA,SACI,EAAgB,EAAiB,EACjC,GACF,EAAQ,GAAS,UACjB,GAAM,GAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAChD,YAAK,OAAO,EAAG,GACR,GAGT,EAAA,UAAA,aAAA,SACI,EAAsB,EAAkB,EACxC,GADsB,AAAA,IAAA,QAAA,GAAA,IAExB,EAAO,GAAQ,KAAK,iBAAiB,WACrC,AAAI,GAAS,MAAQ,IAAU,EAAa,OAC1C,GAAe,EAAa,KAAK,IAEnC,GAAM,GAAI,GAAI,IAAS,EAAc,EAAW,EAAM,KAAK,gBAC3D,GAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,KAC5C,KAAM,IAAI,OAAM,sBAAsB,EAAE,KAAI,2BAE9C,YAAK,MAAM,oBAAoB,EAAE,MAAQ,EACzC,KAAK,OAAO,EAAG,KAAK,SACb,GAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAM,GAAW,KAAK,MAAM,WAAW,IAAI,EAAE,QACzC,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,SACpC,EAKJ,GAJA,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAET,IAAa,GACf,KAAK,MAAM,iBAIX,GAAI,GAAQ,EACZ,AAAI,EAAE,QAAU,aAAe,EAAE,QAAU,UACzC,GAAQ,EAAE,KAAO,GAAqB,EAAE,QAE1C,KAAK,MAAM,WAAW,IAAI,EAAE,OAAQ,CAClC,QAAS,GAAW,KAAK,QACzB,MAAO,EAAE,MACT,MAAO,EAAE,MACT,MAAK,EACL,SAAU,IAEZ,KAAK,MAAM,UAAY,EAGzB,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,WAEpC,AAAM,YAAa,KACjB,KAAK,MAAM,IAIf,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAC,KAAK,MAAM,WAAW,IAAI,EAAE,QAC/B,OAGF,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAEb,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,EAAE,QACnC,EAAW,EAAK,SAEtB,AAAI,GAAY,EAGd,CAAI,EAAE,QAAU,aACd,MAAK,MAAM,UAAY,EAAK,OAE9B,KAAK,MAAM,iBAEX,EAAK,QAAQ,YAAY,EAAE,QAC3B,KAAK,MAAM,WAAW,OAAO,EAAE,SAE/B,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,YAOxC,EAAA,UAAA,iBAAA,WACE,OAAW,KAAW,MAAK,MAAM,qBAC/B,GAAM,GAAI,KAAK,MAAM,oBAAoB,GACzC,KAAK,gBAAgB,KAIzB,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,cAAc,GACnB,AAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,MAC5C,MAAO,MAAK,MAAM,oBAAoB,EAAE,OAI5C,EAAA,UAAA,OAAA,WACE,GAAM,GAAO,KAAK,QAAQ,SAC1B,SAAK,WAAa,KAAK,MAAM,WAC7B,EAAK,eAAiB,KAAK,MAAM,eACjC,EAAK,SAAW,KAAK,MAAM,SAC3B,AAAI,KAAK,MAAM,iBAAmB,GAChC,GAAK,WAAa,GAClB,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,IAEjB,EAAK,QAAQ,KACT,0EAGC,GAGH,EAAA,UAAA,QAAN,SAAc,iHAEZ,YAAK,MAAM,UAAY,GAEjB,EAAa,KAAK,MAAM,SACxB,EAAkB,KAAK,MAAM,WAEnC,KAAK,MAAM,cAAc,QAAU,GACnC,EAAA,KAAK,MAAM,cAAuB,CAAA,EAAM,YAAxC,EAAyB,OAAS,EAAA,OAElC,KAAK,MAAM,UAAY,GAEvB,KAAK,MAAM,cAAc,UAAY,KAAK,IAAG,MAAR,KAC9B,KAAK,MAAM,cAAc,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,sBACnD,KAAK,MAAM,cAAc,SAAW,KAAK,MAAM,SAAW,EAC1D,KAAK,MAAM,cAAc,WACrB,KAAK,MAAM,WAAa,MACP,EAAA,KAAK,MAAM,cAAc,+BAAzB,GAAA,EAAA,OAAV,GAAM,EAAA,GACf,EAAA,EAAsB,CAAA,EAAM,EAAO,eADgB,CAAA,EAAA,UACnD,SAAO,aAAe,EAAA,OACtB,EAAA,EAAmB,CAAA,EAAM,EAAO,kBAAhC,EAAO,UAAY,EAAA,wBAFA,wBAIrB,MAAA,CAAA,EAAO,KAAK,MAAM,qBAGpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,MAAM,cAAgB,GAAK,KAAK,MAAM,cAAgB,GAG5D,EAAA,UAAA,YAAR,SACI,EAAoB,EAAwB,EAC5C,EAAyB,EAAiB,GAF9C,GAAA,GAAA,KAGQ,EACF,CAAC,GAAI,KAAK,MAAM,iBAAkB,WAAU,EAAE,OAAM,EAAE,QAAO,EAAE,MAAK,GAElE,EAAa,GAAY,GAC/B,AAAI,GAAc,MAChB,GAAgB,EAAW,UAE7B,AAAI,GAAiB,MACnB,GAAS,SAAW,SAAC,GAGnB,SAAM,EAAI,IAAI,SAAC,EAAI,GACjB,GAAI,GAAM,MACR,GAAM,GAAS,EAAQ,GACjB,EAAO,GAAyB,EAAO,KAAM,EAAO,OAC1D,MAAO,GAAK,WAAW,EAAM,EAAO,MAAO,EAAO,OAEpD,MAAO,KAIF,EAAc,EAAI,OAAS,EAAI,EAAM,EAAI,GAAI,EAAO,KAG/D,KAAK,MAAM,WAAW,KAAK,IAG7B,EAAA,UAAA,KAAA,SAAuB,GACrB,SAAO,KAAO,GACP,GAGD,EAAA,UAAA,UAAR,WACE,AAAI,KAAK,MAAM,gBAAkB,GAC/B,MAAK,MAAM,WAAa,IAE1B,KAAK,MAAM,iBAGL,EAAA,UAAA,QAAR,WACE,KAAK,MAAM,iBAOb,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAwB,CAC5B,MAAO,GACP,KAAM,gBACN,GAAI,KAAK,MAAM,eAEjB,AAAI,GACF,GAAU,KAAO,GAEnB,KAAK,MAAM,WAAW,KAAK,GAC3B,KAAK,MAAM,YAAc,GAO3B,EAAA,UAAA,SAAA,SAAS,GAMP,OANF,GAAA,KACQ,EAAyB,GAAsB,GAC/C,EACF,GAAI,KAAI,EAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,MAGrC,EAAI,EAAG,EAAI,KAAK,MAAM,YAAY,MAAM,OAAQ,KACvD,GAAM,GAAS,KAAK,MAAM,YAAY,MAAM,GAC5C,AAAI,CAAC,EAAO,MAAQ,CAAC,EAA0B,IAAI,EAAO,KACxD,EAAO,UAIX,GAAM,GAAW,KAAK,MAAM,WAAW,MACvC,KAAK,MAAM,YAAc,KAAK,MAAM,WAAW,SAAW,EACtD,KACA,KAAK,MAAM,WAAW,KAAK,MAAM,WAAW,OAAS,GAGzD,EAAuB,QAAQ,SAAA,GAG7B,AAAI,CAAC,EAAO,MAAQ,EAAO,UAAY,EAAS,IAC9C,EAAK,MAAM,MAWjB,EAAA,UAAA,UAAA,SACI,EAAY,EAAc,EAC1B,GAFJ,GAAA,GAAA,KAKE,GAHE,AAAA,IAAA,QAAA,GAAA,IACF,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,8CACrB,GAAM,MAAQ,EAAG,QAAU,UAC7B,KAAM,IAAI,OAAM,0CAA0C,EAAG,MAAK,KAGpE,GAAM,GAAI,KAAK,UACX,WAAM,MAAA,GAAK,aAAa,WAAM,MAAA,GAAK,WACnC,WAAM,MAAA,GAAK,KAAK,UAAW,KAE/B,EACI,YAAa,GACb,WAAM,MAAA,mDAEV,GAAM,GAAe,GAAqB,KAAK,MAAM,WAAY,EAAI,GACrE,GAAI,CAAC,GAAoB,EAAa,SAAW,GAAK,EAAG,OAAS,EAChE,KAAM,IAAI,OACN,uIAKN,MAAO,MAAK,KAAK,WAAY,WAC3B,GAAM,GAAuD,GAC7D,EAAuB,EAAE,IAAO,GAAc,GAAK,EAAE,OAGrD,GACI,EAAwB,EAExB,SAAA,GAAK,MAAA,GAAK,KAAK,IAEf,IACJ,GAAM,GAAQ,EAAG,IAAI,SAAA,GAAK,MAAA,GAAuB,EAAE,MAEnD,MAAI,GAAK,MAAM,gBAAkB,GAG/B,GAAK,MAAM,WAAW,QAAQ,SAAA,GAC5B,OAAqB,GAAA,EAAA,EAAA,EAAK,MAAL,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,EAAO,aAGX,EAAK,MAAM,WAAa,MAEnB,CAAC,MAAO,EAAG,MAAK,MAI3B,EAAA,UAAA,WAAA,SAA6B,GAA7B,GAAA,GAAA,KAEE,SACI,GAAgB,GAChB,WAAM,MAAA,sDACH,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACN,EACI,EAAO,MAAM,SAAA,GAAK,MAAA,aAAa,KAC/B,WAAM,MAAA,qEAGV,GAAI,GAIE,EAA2B,GACjC,SAAO,QAAQ,SAAC,EAAO,GACrB,EAAS,GAAK,IAET,EAAK,cACR,SAAC,EAAG,GACF,SAAM,EAAC,MAAA,OAAQ,EAAM,OAAA,CAAE,KACvB,EACI,EAAI,gBAAiB,GACrB,WAAM,MAAA,+FAEV,EACI,GAAgB,EAAI,UACpB,WAAM,MAAA,qGAEH,EAAI,OAEb,EACA,SAAC,EAAO,GACN,GAAM,GAAU,EAAI,SAAS,EAAI,GAC3B,EACF,MAAM,QAAQ,GAAW,EAAU,CAAC,GACxC,EACI,EAAM,SAAW,EAAO,OACxB,WAAM,MAAA,wKAGV,EACI,EAAM,MAAM,SAAA,GAAK,MAAA,aAAa,KAC9B,WAAM,MAAA,yIAGV,GAAM,GAAyC,GAC/C,SAAM,QAAQ,SAAC,EAAM,GACnB,EAAQ,GAAK,WAAM,MAAA,MAEd,MAKjB,EAAA,UAAA,SAAA,SAAS,GAEP,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,SAAS,IAE/B,EAAA,UAAA,KAAA,SAAK,GAEH,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,KAAK,IAGrB,EAAA,UAAA,KAAN,SAAW,qGACH,SAAQ,KACK,CAAA,EAAM,KAAK,QAAQ,KAAK,WAArC,SAAa,EAAA,OACnB,EAAW,OAAS,KAAQ,EAC5B,CAAA,EAAO,SASD,EAAA,UAAA,MAAR,SAAgC,GAC9B,MAAI,MAAK,MAAM,aAAe,MAC5B,GAAO,QAAU,KAAK,MAAM,YAAY,GACxC,KAAK,MAAM,YAAY,MAAM,KAAK,IAG7B,GAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAOpB,EAAA,UAAA,MAAA,WAEE,KAAK,uBAEL,KAAK,MAAM,UACX,KAAK,IAAI,QACT,KAAK,MAAQ,GAAI,IAEjB,OAAW,KAAe,MAAK,SAC7B,KAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,GAEvB,KAAK,YAAc,KACnB,KAAK,gBAAkB,KACvB,KAAK,mBAAqB,MA/sBb,EAAA,aAAe,EAKf,EAAA,eAAiB,EA4sBlC,KAEA,YAAc,GACZ,GAAM,GAAS,GAAmB,GAAc,GAAQ,WACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,yBAIxC,GAAM,GAAK,KACX,GAAI,EAAG,WAAa,MAClB,GAAM,GAAc,GAAI,IAAY,GACpC,EAAG,UAAY,GAAI,IAAO,GAE5B,UAAqB,EAAG,UAAU,KAIlC,GAAiB,WAAM,MAAA,GAAG,YACnB,EAAG,UAGL,GAAM,GAAS,iBAQF,EAAW,GAE7B,GAAM,GAAS,CAAC,EAAC,EAAE,EAAC,GACpB,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,EAAG,GAC3B,SAAK,CAAC,EAAG,IACF,GACN,EAAgC,KAAqB,ICzrC1D,AAkBA,cACE,MAAO,OAAO,YAAc,aAAe,WAAa,mBAIxD,GAAI,MAEF,GAAM,GAAI,UAAU,WAAa,UAAU,QAAW,OAAe,MAErE,MAAO,2TACK,KAAK,IAEb,0kDACK,KAAK,EAAE,OAAO,EAAG,IAE5B,MAAO,iBAIP,MAAQ,OAAO,SAAW,aAAe,OAAO,UAAY,MAEvD,MAAO,oBAAsB,6DCvCpC,AAqBA,GAAM,IAAM,KAOZ,GAAI,aAAa,QAAS,WAAM,MAAA,IAAO,SAAA,GACrC,AAAI,GACF,QAAQ,KACJ,iJAOR,GAAI,aAAa,aAAc,WAAM,MAAA,QAGrC,GAAI,aACA,UACA,WAAM,MAAC,OAAO,UAAY,aACrB,MAAO,SAAQ,UAAa,aAC5B,MAAO,SAAQ,SAAS,MAAS,cAG1C,GAAI,aACA,YACA,WAAM,MAAA,OAAO,YAAc,aAAe,WAAa,MACnD,UAAU,WAAa,MAAQ,SAAS,KAAK,UAAU,YACvD,aAAa,KAAK,UAAU,UAMpC,GAAI,aAAa,OAAQ,WAAM,MAAA,KAM/B,GAAI,aACA,qCAAsC,WAAM,MAAA,IAAI,QAAQ,WAG5D,GAAI,aAAa,+BAAgC,WAAM,MAAA,KAGvD,GAAI,aAAa,UAAW,WAAM,MAAA,KCvElC,YAuB2B,EAAiB,GAC1C,GAAI,GAAwB,EAE5B,GAAI,GAAa,GACf,MAAO,KAAU,SAAW,GAAK,CAAC,EAAI,QAExC,GAAI,CAAC,MAAM,QAAQ,GACjB,MAAO,GAIT,OAFM,GAAkB,GAEjB,MAAM,QAAQ,IACd,GAAa,IAAc,IAAU,UAC1C,EAAM,KAAK,EAAU,QACrB,EAAY,EAAU,GAExB,MAAI,OAAM,QAAQ,IACd,KAAM,QAAQ,uCAChB,GAA2B,EAAK,EAAO,IAGlC,EAGT,YACI,EAAiB,EAAiB,GAEpC,GADA,EAAU,GAAW,GACjB,CAAE,MAAM,QAAQ,IAAS,CAAC,GAAa,IACzC,EACI,EAAM,SAAW,EACjB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,qBACnC,yCAAwC,EAAM,GAAE,eACxD,OAEF,EACI,EAAM,OAAS,EACf,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,4BACnC,uBAAsB,EAAI,OAAM,eACxC,EACI,EAAI,SAAW,EAAM,GACrB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,iBAAiB,EAAM,GAAE,IAC5D,sBAAqB,EAAI,OAAM,eAEvC,OADM,GAAW,EAAM,MAAM,GACpB,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAA2B,EAAI,GAAI,EAAU,EAAQ,OAAO,IAIhE,YACI,EAAmC,EAAuB,EAC1D,GACF,GAAI,GAAiB,KACnB,OAEF,GAAI,IAAkB,WAAa,IAAkB,GACjD,IAAkB,WAAa,IAAgB,SACjD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,UAChD,OAAM,EAAa,oBAAoB,EAAW,uBAKtD,EAAiB,EAAiB,EAClC,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,YAAa,GACf,UAAY,EAAc,EAAE,MAAO,EAAS,GACrC,EAET,GAAI,GAAgB,GAAW,GAS/B,GANA,AAAI,IAAkB,UAClB,CAAC,OAAQ,QAAS,WAAW,QAAQ,IAAiB,GACxD,GAAgB,GAElB,GAAY,EAAc,EAAe,EAAS,GAE7C,GAAK,MACL,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IAAM,MAAO,IAAM,UACtD,MAAO,IAAM,WAAa,MAAO,IAAM,UAC1C,GAAM,GAAO,GAAK,KAAO,OAAU,EAAS,YAAY,KACxD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,eAChD,mCAAkC,EAAI,MAE5C,GAAM,GAAgB,GAAW,EAAG,GACpC,AAAI,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IACrC,GAAI,CAAC,IAEP,GAAM,GAAiB,GACjB,EAAS,IAAkB,SAC7B,GAAa,EAAG,GAChB,GAAQ,EAAe,GAAI,GAC/B,MAAO,GAAO,WAAW,EAAQ,EAAe,eAI9C,EAA0B,EAAiB,EAC3C,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,OACN,YAAY,EAAO,cAAc,EAAY,2CAGnD,GAAM,GAAU,EAChB,MAAO,GAAQ,IACX,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAM,EAAO,IAAI,EAAC,IAAK,IACjD,GCnIN,GAkBa,IAAkB,kBAOQ,GACrC,GAAM,GAAO,OAAO,KAAK,GACzB,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,OACN,yGAEG,GAAK,OAAM,WAGpB,GAAI,GAAS,EAAK,GACZ,EAAK,EAAE,GAGb,AAAI,EAAO,SAAS,MAClB,GAAS,EAAO,UAAU,EAAG,EAAO,OAAS,IAI/C,EAAS,EAAS,GAGlB,GAAM,GAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,EAAO,WAAW,GAClB,IACE,GAAM,GAAS,EAAE,MAAA,OAAI,GACrB,MAAI,aAAkB,UACpB,QAAQ,MAAM,2CAEhB,EAAO,SAAS,GACT,QACA,GACP,QAAO,SAAS,MACV,IAGV,cAAO,eAAe,EAAI,OAAQ,CAAC,MAAO,EAAQ,aAAc,KAGzD,EC/DT,AA8CA,YAAoC,EAAoB,GACtD,GAAM,GAAQ,EAAgB,EAAM,OAAQ,WACtC,EAAQ,EAAgB,EAAM,OAAQ,WAC5C,GACI,EAAM,MAAO,EAAM,MACnB,yBAAyB,EAAM,MAAK,QAAQ,EAAM,MAAK,yCAG3D,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,QAAQ,EAAO,IAE1B,EAAwB,CAAC,KAAM,EAAO,KAAM,GAClD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC/DnC,YAyBI,EAAoB,EAAiB,EACrC,GAIF,GAHA,AAAI,GAAS,MACX,GAAQ,GAAW,IAEjB,IAAU,YACZ,KAAM,IAAI,OACN,oFAGN,GAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IACxC,MAAO,IAAW,UAAY,MAAO,IAAW,WAChD,MAAO,IAAW,SACpB,KAAM,IAAI,OACN,4HAGN,GAAI,GAAS,MACX,GAAmC,GAEnC,GAAM,GAAe,GAAc,GAC7B,EAAe,GAAc,GACnC,EACI,IAAiB,EACjB,WACI,MAAA,iCAAiC,EAAK,6BACnC,GAAY,mBAAmB,KAE1C,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,EAAE,GAC1C,GAAM,GAAW,EAAc,GACzB,EAAoB,IAAM,EAAc,OAAS,EACnD,IAAa,GAAc,EAAM,MAAM,IACvC,GACJ,EACI,EAAc,KAAO,EAAM,IAAM,CAAC,EAClC,WAAM,MAAA,+CACF,KAAI,EAAa,kCACjB,WAAU,EAAK,UAI3B,MAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IAC1C,GAAS,CAAC,IAGZ,EAAQ,GAAS,EACjB,EAAS,IAAU,SACf,GAAa,EAAQ,GACrB,GAAQ,EAAoB,GAAI,IAC7B,EAAO,WAAW,EAAsB,EAAO,GC1ExD,YAqDI,EAAoB,EAAqB,GAC3C,GAAM,GAAgB,GAAW,EAAQ,GACzC,MAAO,IAAW,EAAQ,EAAO,EAAe,GCvDlD,AAsBO,GAAM,IAAkD,CAC7D,QAAW,EACX,QAAW,EACX,MAAS,EACT,OAAU,EACV,MAAS,EACT,KAAQ,EACR,UAAa,GC7Bf,AA2BA,GAAM,IAA0B,cAoB5B,EAAuC,oHAUzC,IAPM,EAAgC,GAChC,EAA2C,GAE3C,EAAkB,MAAM,QAAQ,GAClC,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,OAC7B,OAAO,KAAK,cAEP,GACP,GAAM,GAAO,EAAM,GACb,EAAI,MAAM,QAAQ,GAAW,EAAQ,GAAG,OAAS,EAAQ,GAC/D,GAAI,EAAE,QAAU,WAAa,EAAE,QAAU,SAAW,EAAE,QAAU,QAC5D,EAAE,QAAU,UAAY,EAAE,QAAU,YACtC,KAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,EAAE,OAE9D,GAAM,GAA6B,CAAC,KAAI,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,OACnE,GAAI,EAAE,QAAU,UACd,GAAM,GAAY,GAAI,SAAoB,SAAM,GAAO,MAAA,IAAA,EAAA,OAAA,OAAA,+EACxC,MAAA,CAAA,EAAM,EAAE,gBAKrB,IALM,EAAO,EAAA,OACP,EAAgB,EAAK,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAE,QAAQ,GACtD,GAA0B,EAAK,OAC7B,EAAQ,GAAI,YAAW,GACzB,EAAS,EACJ,EAAI,EAAG,EAAI,EAAK,OAAQ,IACzB,EAAM,EAAK,GACX,EACF,GAAI,YAAW,GAAI,aAAY,CAAC,EAAI,SAAS,QACjD,EAAM,IAAI,EAAe,GACzB,GAAU,GACV,EAAM,IAAI,EAAK,GACf,GAAU,EAAI,OAEhB,SAAQ,cAEV,EAAa,KAAK,OAElB,GAAa,KAAK,EAAE,QAEtB,AAAI,GAAS,MACX,GAAK,MAAQ,GAEf,EAAM,KAAK,IAjCJ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,IAA3B,GAoCY,MAAA,CAAA,EAAM,QAAQ,IAAI,WAAjC,SAAe,EAAA,OACrB,CAAA,EAAO,CAAC,KAAM,GAAuB,GAAe,MAAK,qBAmBvD,EAAqB,GAKvB,OAHM,GAAsB,GACxB,EACA,EAAS,EACM,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAO,EAAK,KACZ,EAAQ,EAAK,MACb,EAAQ,EAAK,MACb,EAAO,GAAc,GACvB,EAAM,OAEV,GAAI,gBAAkB,IACpB,GAAM,GAAe,EAAK,aAC1B,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,GAAI,CAAE,QAAS,IAAgB,SAAW,IACxC,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,+DAGxD,EAAa,QAAU,WAChC,GAAI,IAAU,UACZ,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,IAC3D,oDAAmD,EAAK,UAG9D,MAAM,IAAI,OACN,UAAU,EAAK,KAAI,gBACnB,uBAAsB,EAAa,MAAK,MACxC,wEAGN,GAAM,GAAyB,GAAqB,EAAa,OAC3D,EACF,EAAO,MAAM,EAAQ,EAAS,EAAO,GACnC,EAAkB,EAAa,QAAU,QAC3C,GAAI,YAAW,GACf,GAAI,aAAY,GACpB,GAAI,IAAU,UACZ,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,EAAS,GAAI,cAAa,EAAe,QACzC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,EAAI,EAAa,MAAQ,EAAa,aAE3C,EAAa,QAAU,UAChC,AAAI,IAAkB,QACpB,GAAgB,MAElB,EAAS,EAAc,OAEvB,MAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,qCAGhD,IAAU,SACnB,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,SAC3D,KAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,2BAGzD,EAAS,GAAI,YAAW,EAAe,QACvC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,KAAK,MAAM,EAAI,EAAa,MAAQ,EAAa,UAG/D,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,UACR,IAAU,UACnB,GAAM,GAAO,GAAc,EAAK,OAChC,EAAS,GACT,OAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAa,GAAI,aACnB,EAAO,MAAM,EAAQ,EAAS,KAA0B,GAC5D,GAAU,GACV,GAAM,GAAQ,GAAI,YAAW,EAAO,MAAM,EAAQ,EAAS,IAC1D,EAAwB,KAAK,GAC9B,GAAU,QAGZ,GAAM,GAAc,GAAqB,GACnC,EAAa,EAAO,MAAM,EAAQ,EAAS,EAAO,GAExD,GAAI,IAAU,UACZ,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,aACnB,EAAS,GAAI,cAAa,GAG1B,OAFM,GAAO,GAAI,cAAa,EAAO,OAAS,GACxC,EAAQ,GAAI,cAAa,EAAO,OAAS,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAK,GAAK,EAAO,EAAI,GACrB,EAAM,GAAK,EAAO,EAAI,EAAI,GAE5B,GAAM,GAAa,GAAO,EAAM,EAAO,WACjC,EAAc,GAAO,EAAO,EAAO,WACzC,EAAI,GAAQ,GAAQ,EAAY,OAEhC,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,EAEnB,AAAI,IAAU,aACZ,GAAI,GAAQ,GAAO,EAAQ,EAAO,IAGtC,MAAO,eAM8B,GAErC,GAAI,IAAO,KACT,KAAM,IAAI,OAAM,wBAAwB,KAAK,UAAU,IAGzD,GAAI,GAAkB,EAShB,EAA6B,GACnC,EAAG,QAAQ,SAAC,GAMV,GALA,GAAmB,EAAE,WAErB,EAAa,KACT,EAAE,aAAe,EAAE,OAAO,WAAa,EACA,GAAK,GAAE,YAAoB,IAClE,CAAE,aAAoB,eAAgB,YAAoB,aACxD,YAAoB,aACxB,KAAM,IAAI,OAAM,mCAAmC,EAAE,YAAY,QAKrE,GAAM,GAAI,GAAI,YAAW,GACrB,EAAS,EACb,SAAa,QAAQ,SAAC,GACpB,EAAE,IAAI,GAAI,YAAW,EAAE,QAAS,GAChC,GAAU,EAAE,aAGP,EAAE,OAIX,GAAM,IAAgB,MAAO,SAAW,aACnC,OAAO,OAAS,aAAe,MAAO,OAAS,aAC/C,MAAO,OAAS,yBAWY,GAC/B,MAAI,IACK,OAAO,WAAW,GAEpB,GAAI,MAAK,CAAC,IAAM,iBASiB,GACxC,GAAI,GACF,MAAO,QAAO,KAAK,GAAQ,SAAS,UAItC,OAFM,GAAM,GAAI,YAAW,GACvB,EAAI,GACC,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAI,EAAG,IACrC,GAAK,OAAO,aAAa,EAAI,IAE/B,MAAO,MAAK,eAS4B,GACxC,GAAI,IACF,GAAM,GAAM,OAAO,KAAK,EAAK,UAC7B,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YAI/D,OAFM,GAAI,KAAK,GACT,EAAS,GAAI,YAAW,EAAE,QACvB,EAAI,EAAG,EAAI,EAAE,OAAQ,EAAE,EAC9B,EAAO,IAAI,CAAC,EAAE,WAAW,IAAK,GAEhC,MAAO,GAAO,mBASwB,GACtC,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAI,GAAkB,EACtB,EAAQ,QAAQ,SAAC,GACf,GAAmB,EAAO,aAG5B,GAAM,GAAO,GAAI,YAAW,GACxB,EAAS,EACb,SAAQ,QAAQ,SAAC,GACf,EAAK,IAAI,GAAI,YAAW,GAAS,GACjC,GAAU,EAAO,aAEZ,EAAK,mBAUW,GACvB,GAAM,GAAY,IAElB,IADA,EAAO,EAAK,OACL,EAAK,SAAS,IACnB,EAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAErC,GAAM,GAAQ,EAAK,MAAM,GACzB,MAAO,GAAM,EAAM,OAAS,eAQe,GAE3C,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OAAM,uDAGlB,MAAO,CACL,UAAW,GAAI,MACf,kBAAmB,OACnB,mBAAoB,EAAe,eAAiB,KAChD,EACA,GAAiB,KAAK,UAAU,EAAe,gBACnD,iBAAkB,EAAe,aAAe,KAC5C,EACA,GAAiB,KAAK,UAAU,EAAe,cACnD,gBAAiB,EAAe,YAAc,KAC1C,EACA,EAAe,WAAW,YAUlC,cACE,GAAM,GAAkB,SAAC,GAIvB,OAHI,GAAI,GAAK,GACT,EAAI,EAEA,GAAI,WAAgB,GAC1B,GAAK,QACL,IAAM,EAER,UAAK,CAAC,QACN,GAAK,UAEE,EAAI,GAGP,EAAe,GAAI,aAAY,MAErC,EAAa,GAAK,EAClB,OAAS,GAAI,EAAG,EAAI,KAAM,IACxB,EAAa,GAAK,EAAgB,GAEpC,OAAS,GAAI,KAAM,EAAI,KAAM,IAC3B,EAAa,GAAK,UAAe,GAAI,MAAS,IAGhD,MAAO,GAST,cACE,GAAM,GAAgB,GAAI,aAAY,IAEtC,EAAc,GAAK,EACnB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,OAAS,GAAI,EAAG,EAAI,GAAI,IACtB,EAAc,GAAK,GAAK,GAE1B,OAAS,GAAI,GAAI,EAAI,GAAI,IACvB,EAAc,GAAK,WAAe,GAAI,IAAO,IAG/C,MAAO,GAST,cAGE,OAFM,GAAc,GAAI,aAAY,IAE3B,EAAI,EAAG,EAAI,GAAI,IACtB,EAAY,GAAK,KAEnB,SAAY,GAAK,EAAY,IAAM,EAE5B,gBAeP,GAAM,GAAe,KACf,EAAgB,KAChB,EAAc,KAEpB,MAAO,UAAC,GAGN,OAFM,GAAS,GAAI,aAAY,EAAI,EAAe,QAC5C,EAAmB,GAAI,aAAY,GAChC,EAAQ,EAAG,EAAQ,EAAe,OAAQ,KACjD,GAAM,GAAc,EAAe,GAC7B,EACF,EAAa,EAAY,GAAe,IAAO,GAAc,OAC7D,EAAc,GAAe,IACjC,EAAiB,GAAS,EAE5B,MAAO,IAAI,cAAa,IChf5B,AAsBA,GAAA,IAAA,WAOE,aACE,KAAK,YAAc,GACnB,KAAK,YAAc,GAGN,SAAA,YAAf,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UASnB,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAS3C,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAW3C,EAAA,gBAAP,SAAuB,GACrB,MAAO,GAAiB,YAAY,EAAK,SAWpC,EAAA,gBAAP,SAAuB,EAAsB,GAE3C,MAAO,GAAiB,YAAY,EAAK,OAAQ,IAGpC,EAAA,YAAf,SACI,EAAsB,EACtB,GACF,GAAM,GAA6B,GAC7B,EAAU,IAAgB,OAC5B,EAAiB,cAAc,YAC/B,EAAiB,cAAc,YACnC,SAAQ,QAAQ,SAAA,GACd,GAAM,GAAU,EAAO,EAAK,GAC5B,AAAI,IAAY,MACd,EAAc,KAAK,KAGhB,GAEX,KAEa,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAkB,SAAC,GAC5B,MAAA,IAAiB,gBAAgB,IACxB,GACT,SAAC,EAAsB,GACnB,MAAA,IAAiB,gBAAgB,EAAK,IC/G9C,AAyBA,GAAM,IAAgB,eAChB,GAAmB,EAKnB,GAAmB,eAInB,GAAkB,mBAexB,cACE,GAAI,CAAC,KAAM,QAAQ,cAIjB,KAAM,IAAI,OACN,2FAIN,GAAM,GAAiB,MAAO,SAAW,YAAc,KAAO,OACxD,EAAU,EAAU,WAAa,EAAU,cAC7C,EAAU,iBAAmB,EAAU,aACvC,EAAU,cACd,GAAI,GAAW,KACb,KAAM,IAAI,OACN,6DAEN,MAAO,GAGT,YAAuB,GACrB,GAAM,GAAK,EAAY,OACvB,EAAG,kBAAkB,GAAkB,CAAC,QAAS,cACjD,EAAG,kBAAkB,GAAiB,CAAC,QAAS,cAQlD,GAAA,IAAA,WAME,WAAY,GAGV,GAFA,KAAK,UAAY,KAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,kEAEN,KAAK,UAAY,EAGb,SAAA,UAAA,KAAN,SAAW,sEAET,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAIN,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,UAAW,SAIvC,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,iBAiB1B,EAAA,UAAA,eAAR,SAAuB,EAAmB,GAA1C,GAAA,GAAA,KAEE,MAAO,IAAI,SAAmC,SAAC,EAAS,GACtD,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OAEvB,GAAI,GAAkB,MAEpB,GAAM,GAAU,EAAG,YAAY,GAAkB,YAC3C,EAAa,EAAQ,YAAY,IACjC,EAAa,EAAW,IAAI,EAAK,WACvC,EAAW,UAAY,WACrB,GAAI,EAAW,QAAU,KACvB,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAK,UAAS,oBAGlD,EAAQ,EAAW,OAAO,iBAG9B,EAAW,QAAU,SAAA,GACnB,SAAG,QACI,EAAO,EAAW,QAE3B,EAAQ,WAAa,WAAM,MAAA,GAAG,cAG9B,GAAM,GACF,GAA6B,GAE3B,EAAS,EAAG,YAAY,GAAiB,aAC3C,EAAY,EAAO,YAAY,IAC7B,EACF,EAAU,IAAI,CAAC,UAAW,EAAK,UAAW,mBAAkB,IAC5D,EACJ,EAAe,UAAY,WAEzB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAkB,EAAW,IAAI,CACrC,UAAW,EAAK,UAChB,eAAc,EACd,mBAAkB,IAEpB,EAAgB,UAAY,WAAM,MAAA,GAAQ,CAAC,mBAAkB,KAC7D,EAAgB,QAAU,SAAA,GAGxB,EAAY,EAAO,YAAY,IAC/B,GAAM,GAAoB,EAAU,OAAO,EAAK,WAChD,EAAkB,UAAY,WAC5B,SAAG,QACI,EAAO,EAAgB,QAEhC,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAgB,UAIpC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAE/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,YAKtC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,WAvHtC,EAAA,WAAa,eA0H/B,KAEa,GAA4B,SAAC,GACxC,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAmBH,GAC/B,MAAO,IAAI,IAAiB,GAG9B,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAiB,YACnC,EAAI,MAAM,GAAiB,WAAW,QACtC,EAGN,GAAA,IAAA,WAGE,aACE,KAAK,UAAY,KAGb,SAAA,UAAA,WAAN,yFACE,MAAA,CAAA,EAAO,GAAI,SACP,SAAC,EAAS,GACR,GAAM,GACF,EAAK,UAAU,KAAK,GAAe,IACvC,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAK,EAAG,YAAY,GAAiB,YACrC,EAAQ,EAAG,YAAY,IASvB,EAAqB,EAAc,SACzC,EAAkB,UAAY,WAE5B,OADM,GAA4C,GAC/B,EAAA,EAAA,EAAA,EAAkB,OAAlB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAI,EAAK,WAAa,EAAK,mBAE7B,EAAQ,IAEV,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAkB,QAElC,EAAG,WAAa,WAAM,MAAA,GAAG,UAE3B,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAIpD,EAAA,UAAA,YAAN,SAAkB,iFAChB,SAAO,GAAiB,GACxB,CAAA,EAAO,GAAI,SAA4B,SAAC,EAAS,GAC/C,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAS,EAAG,YAAY,GAAiB,aACzC,EAAY,EAAO,YAAY,IAE/B,EAAiB,EAAU,IAAI,GACjC,EACJ,EAAe,UAAY,WACzB,GAAI,EAAe,QAAU,KAC3B,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAI,oBAIxC,GAAM,GAAoB,EAAU,OAAO,GACrC,EAAkB,WAEtB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAqB,EAAW,OAAO,GAC7C,EAAmB,UAAY,WAC3B,MAAA,GAAQ,EAAe,OAAO,qBAClC,EAAmB,QAAU,SAAA,GACzB,MAAA,GAAO,EAAe,SAI5B,EAAkB,UAAY,EAC9B,EAAkB,QAAU,SAAA,GAC1B,WACA,EAAG,QACI,EAAO,EAAe,SAInC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAG/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,WAIpC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAGxD,KCjWA,AAyBA,GAAM,IAAiB,IACjB,GAAc,sBACd,GAAc,OACd,GAAwB,iBACxB,GAAsB,eACtB,GAAqB,cACrB,GAAwB,iBA8B9B,YAAsB,GAOpB,MAAO,CACL,KAAM,CAAC,GAAa,EAAM,IAAa,KAAK,IAC5C,SAAU,CAAC,GAAa,EAAM,IAAuB,KAAK,IAC1D,YAAa,CAAC,GAAa,EAAM,IAAqB,KAAK,IAC3D,WAAY,CAAC,GAAa,EAAM,IAAoB,KAAK,IACzD,cACI,CAAC,GAAa,EAAM,IAAuB,KAAK,KAWxD,YAA6B,GAC3B,GAAM,GAAQ,EAAI,MAAM,IACxB,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OAAM,uBAAuB,GAEzC,MAAO,GAAM,MAAM,EAAG,EAAM,OAAS,GAAG,KAAK,IAG/C,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAoB,YACtC,EAAI,MAAM,GAAoB,WAAW,QACzC,EAgBN,GAAA,IAAA,WAOE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,eAAiB,MAAO,SAAW,aAClD,MAAO,QAAO,cAAiB,YAKjC,KAAM,IAAI,OACN,2DAIN,GAFA,KAAK,GAAK,OAAO,aAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,sEAEN,KAAK,UAAY,EACjB,KAAK,KAAO,GAAa,KAAK,WAY1B,SAAA,UAAA,KAAN,SAAW,gFACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAGE,EAAW,KAAK,UAAU,EAAe,eACzC,EAAc,KAAK,UAAU,EAAe,aAE5C,EACF,GAA6B,GAEjC,IACE,YAAK,GAAG,QAAQ,KAAK,KAAK,KAAM,KAAK,UAAU,IAC/C,KAAK,GAAG,QAAQ,KAAK,KAAK,SAAU,GACpC,KAAK,GAAG,QAAQ,KAAK,KAAK,YAAa,GACvC,KAAK,GAAG,QACJ,KAAK,KAAK,WACV,GAA0B,EAAe,aAC7C,KAAK,GAAG,QAAQ,KAAK,KAAK,cAAe,KAAK,UAAU,CACtD,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,uBAGtC,CAAA,EAAO,CAAC,mBAAkB,UACnB,GAEP,WAAK,GAAG,WAAW,KAAK,KAAK,MAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,UAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,aAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,YAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,eAEvB,GAAI,OACN,yBAAyB,KAAK,UAAS,sFAEvC,uBAAsB,EAAmB,mBAAkB,MAC3D,qBAAoB,EAAmB,iBAAgB,MACvD,oBAAmB,EAAmB,gBAAe,qBAazD,EAAA,UAAA,KAAN,gGAGE,GAFM,EACF,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,OACrC,GAAQ,KACV,KAAM,IAAI,OACN,kDAAkD,KAAK,UAAS,KAGtE,GAAI,EAAK,oBAAsB,OAC7B,KAAM,IAAI,OACN,6EAQN,GAJM,EAAsB,GAGtB,EAAW,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,WAClD,GAAY,KACd,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAS,iBAOhE,GAJA,EAAI,cAAgB,EAGd,EAAc,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,cACrD,GAAe,KACjB,KAAM,IAAI,OACN,gDAAgD,KAAK,UAAS,kBAiBpE,GAdA,EAAI,YAAc,EAGZ,EAAiB,KAAK,GAAG,QAAQ,KAAK,KAAK,eACjD,AAAI,GAAkB,MACd,GAAW,KAAK,MAAM,GAC5B,EAAI,OAAS,EAAS,OACtB,EAAI,YAAc,EAAS,YAC3B,EAAI,YAAc,EAAS,YAC3B,EAAI,oBAAsB,EAAS,qBAI/B,EAAmB,KAAK,GAAG,QAAQ,KAAK,KAAK,YAC/C,GAAoB,KACtB,KAAM,IAAI,OACN,uDACA,KAAI,KAAK,UAAS,mBAExB,SAAI,WAAa,GAA0B,GAE3C,CAAA,EAAO,QAzIO,EAAA,WAAa,kBA2I/B,KAEa,GAA+B,SAAC,GAC3C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAoB,aACrD,GACH,EAAI,MAAM,GAAoB,WAAW,SAJxC,MAUX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBA0BA,GAClC,MAAO,IAAI,IAAoB,GAGjC,GAAA,IAAA,WAGE,aACE,EACI,KAAM,QAAQ,cACd,WAAM,MAAA,6CACV,EACI,MAAO,SAAW,aACd,MAAO,QAAO,cAAiB,YACnC,WAAM,MAAA,4DACV,KAAK,GAAK,OAAO,aAGb,SAAA,UAAA,WAAN,8FAIE,IAHM,EAA4C,GAC5C,EAAS,GAAc,GACvB,EAAS,GAAiB,GACvB,EAAI,EAAG,EAAI,KAAK,GAAG,OAAQ,EAAE,EAC9B,EAAM,KAAK,GAAG,IAAI,GACxB,AAAI,EAAI,WAAW,IAAW,EAAI,SAAS,IACnC,GAAY,GAAoB,GACtC,EAAI,GAAa,KAAK,MAAM,KAAK,GAAG,QAAQ,KAGhD,MAAA,CAAA,EAAO,QAGH,EAAA,UAAA,YAAN,SAAkB,8EAGhB,GAFA,EAAO,GAAiB,GAClB,EAAO,GAAa,GACtB,KAAK,GAAG,QAAQ,EAAK,OAAS,KAChC,KAAM,IAAI,OAAM,8BAA8B,EAAI,KAE9C,SAAO,KAAK,MAAM,KAAK,GAAG,QAAQ,EAAK,OAE7C,KAAK,GAAG,WAAW,EAAK,MACxB,KAAK,GAAG,WAAW,EAAK,UACxB,KAAK,GAAG,WAAW,EAAK,aACxB,KAAK,GAAG,WAAW,EAAK,YACxB,CAAA,EAAO,QAEX,KCvVA,AAiCA,GAAM,IAAoB,MAE1B,GAAA,WAME,aACE,KAAK,SAAW,GAGH,SAAA,YAAf,WACE,MAAI,GAA0B,UAAY,MACxC,GAA0B,SAAW,GAAI,IAEpC,EAA0B,UAS5B,EAAA,gBAAP,SAAuB,EAAgB,GACrC,EAAO,GAAU,KAAM,WAAM,MAAA,0CAC7B,AAAI,EAAO,SAAS,KAClB,GAAS,EAAO,MAAM,EAAG,EAAO,QAAQ,MAE1C,EAAO,EAAO,OAAS,EAAG,WAAM,MAAA,wCAChC,GAAM,GAAW,EAA0B,cAC3C,EACI,EAAS,SAAS,IAAW,KAC7B,WAAM,MAAA,2DACF,EAAM,OACd,EAAS,SAAS,GAAU,GAGvB,EAAA,WAAP,SAAkB,GAChB,GAAM,GAAU,KAAK,cAAc,SAAS,GAC5C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,yCAAyC,EAAM,KAEjE,MAAO,IAGF,EAAA,WAAP,WACE,MAAO,QAAO,KAAK,KAAK,cAAc,WAE1C,KAUA,YAAkB,GAChB,GAAI,EAAI,QAAQ,MAAuB,GACrC,KAAM,IAAI,OACN,6EAEA,IAAG,GAA0B,aAAa,KAAK,OAErD,MAAO,CACL,OAAQ,EAAI,MAAM,IAAmB,GACrC,KAAM,EAAI,MAAM,IAAmB,IAIvC,YACI,EAAmB,EACnB,GAAA,MAAA,KAAA,QAAA,GAAA,6GACF,SACI,IAAc,EACd,WAAM,MAAA,wCAAwC,EAAS,MAErD,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,kEACF,EAAS,MACjB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,iCAAgC,EAAS,OAC3C,EAAc,EAAa,GAE3B,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,mEACF,QAAO,EAAO,OACtB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,sCAAqC,EAAO,OAC9C,EAAc,EAAa,GAE3B,EAAe,GAAS,GAAW,OACnC,EAAa,GAAS,GAAW,KACjC,EAAa,IAAiB,GAAS,GAAW,OAEjC,CAAA,EAAM,EAAY,qBAAnC,GAAiB,EAAA,OAKnB,GAAgB,EAClB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIiB,MAAA,CAAA,EAAM,EAAY,KAAK,iBAApC,GAAa,EAAA,OAKf,GAAgB,CAAC,EACnB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIF,MAAA,CAAA,EAAO,EAAW,yBAsCpB,4HACQ,EAAU,GAA0B,aACpC,EAA2C,OAC5B,EAAA,yBAAA,GAAA,EAAA,OAAV,GAAM,EAAA,GAEX,CAAA,EAAM,GAA0B,WAAW,GAAQ,eAF7B,CAAA,EAAA,UACpB,EACF,EAAA,OACJ,IAAW,IAAQ,GACX,EAAM,EAAS,GAAoB,EACzC,EAAI,GAAO,EAAU,oBALJ,wBAQrB,MAAA,CAAA,EAAO,QAoCT,YAA2B,8EACnB,SAAgB,GAAS,GACzB,EAAU,GAA0B,WAAW,EAAc,QACnE,CAAA,EAAO,EAAQ,YAAY,EAAc,WAkD3C,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QAiDhD,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QChWhD,AA0BA,GAAA,IAAA,WAAA,cAKE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,OAAM,EAAM,IAGrB,EAAA,UAAA,IAAA,WACE,MAAO,aAAY,OAGrB,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,kDAAkD,GAExD,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAAI,cAElB,KAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,IAAI,aAAY,GAAU,OAAO,IAE5C,KAEA,GAAI,KAAM,IAAI,eACZ,KAAM,YAAY,UAAW,GAAI,KAGjC,IACE,GAA0B,gBACtB,GAAoB,WAAY,GAAI,WACjC,IAIT,IACE,GAA0B,gBACtB,GAAiB,WAAY,GAAI,WAC9B,KCpEX,AAqBO,GAAM,IAAe,CAE1B,YAAa,WAAM,MAAA,QAIjB,GAaJ,GAAA,WAKE,aAEE,KAAK,KAAO,KAGZ,KAAK,YAAc,GAAI,MAAK,KAAK,YAGnC,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAI,MAAM,OAAO,OAAS,KACjB,KAAM,OAAO,MAAM,EAAM,GAGlC,CAAI,IAAe,MACjB,IAAc,GAAa,eAEtB,GAAY,EAAM,KAG3B,EAAA,UAAA,IAAA,WACE,GAAM,GAAO,QAAQ,SACrB,MAAO,GAAK,GAAK,IAAO,EAAK,GAAK,KAGpC,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,sDAAsD,GAE5D,MAAO,MAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAI,GAAM,SAAW,EACZ,GAEF,GAAI,MAAK,KAAK,YAAY,GAAU,OAAO,IAEtD,KAEA,AAAI,KAAM,IAAI,YACZ,KAAM,YAAY,OAAQ,GAAI,KCrFhC,YAgDI,EAAoB,EACpB,GADoB,MAAA,KAAA,QAAA,GAAW,WAEjC,EAAQ,GAAS,UACjB,GAAwC,GACjC,GAAI,IAAmB,EAAO,EAAO,GCpD9C,AAuCA,YAAiC,EAAiB,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAGnC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,mCAAmC,GAErD,GAAI,IAAU,UAAY,EAAG,QAAU,UACnC,IAAU,UAAY,EAAG,QAAU,SACrC,KAAM,IAAI,OAAM,yCAGlB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAGhB,IAAO,EAAG,CAAC,MAAK,KC3D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,QAAS,MACtC,EAAU,WACZ,MAAA,GAAO,qBAAqB,EAAG,OAAQ,EAAG,MAAO,EAAG,QAElD,EAAyB,CAAC,EAAG,GAInC,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAQ,EAAG,CAAC,OAAM,KCrD/B,YAgCwC,EAAM,GAAA,AAAA,IAAA,QAAA,GAAA,IAC5C,QAAQ,IAAI,EAAE,SAAS,ICjCzB,AAqBA,KAcA,GAAM,IAAuB,CAC3B,OAAM,GACN,KAAI,GACJ,MAAK,GACL,MAAK,IAEP,GAAa,ICzCb,AA6BA,GAAM,IAA2B,QAC3B,GAA8B,QAC9B,GAAqC,eAE3C,YAAkB,GAChB,MAAO,IAAI,SAAQ,SAAA,GAAW,MAAA,YAAW,KAAU,KAAK,GAG1D,GAAA,IAAA,WAQE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,cAGjB,KAAM,IAAI,OACN,uFAIN,AAAI,EAAe,WAAW,EAAiB,aAC7C,GAAiB,EAAe,MAAM,EAAiB,WAAW,SAEpE,AAAI,IAAkB,MAAQ,EAAe,SAAW,IACtD,GAAiB,IAGnB,KAAK,sBAAwB,EAAiB,GAC9C,KAAK,mBACD,EAAiB,GAGjB,SAAA,UAAA,KAAN,SAAW,6GACT,GAAI,MAAQ,WAAc,YACxB,KAAM,IAAI,OACN,8FAGA,EAAa,OAAO,IAAI,gBAAgB,GAAI,MAC9C,CAAC,EAAe,YAAa,CAAC,KAAM,+BAEpC,GAAe,wBAAyB,cAAxC,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,OACN,gGAGE,SAAyC,CAAC,CAC9C,MAAO,CAAC,KAAO,KAAK,oBACpB,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,gBAAe,GAEX,EACF,OAAO,IAAI,gBAAgB,GAAI,MAC3B,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,sBAIT,EAAa,KAAK,YAAc,KAAO,SAAS,cAAc,KACvB,KAAK,WAClD,EAAW,SAAW,KAAK,sBAC3B,EAAW,KAAO,EAIlB,CAAA,EAAM,GAAM,WAAM,MAAA,GAAW,cAAc,GAAI,YAAW,0BAA1D,GAAA,OAEI,EAAe,YAAc,KACzB,GAAmB,KAAK,kBAAoB,KAC9C,SAAS,cAAc,KACvB,KAAK,iBACT,EAAiB,SAAW,KAAK,mBACjC,EAAiB,KAAO,EACxB,CAAA,EAAM,GACF,WAAM,MAAA,GAAiB,cAAc,GAAI,YAAW,cAPtD,CAAA,EAAA,UAMF,EAAA,wBAIF,MAAA,CAAA,EAAO,CAAC,mBAAoB,GAA6B,WA1E7C,EAAA,WAAa,eA6E/B,KAEA,GAAA,WAGE,WAAY,GACV,GAAI,GAAS,MAAQ,EAAM,OAAS,EAClC,KAAM,IAAI,OACN,2DACA,iBAAgB,IAEtB,KAAK,MAAQ,EAGT,SAAA,UAAA,KAAN,6FACQ,SAAW,KAAK,MAAM,GACtB,EAAc,KAAK,MAAM,MAAM,GAErC,CAAA,EAAO,GAAI,SAAwB,SAAC,EAAS,GAC3C,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GAEnB,GAAM,GAAY,KAAK,MAAO,EAAM,OAAe,QAC7C,EAAgB,EAAU,cAChC,GAAI,GAAiB,MACnB,EAAO,GAAI,OACP,4CAA4C,EAAS,OACzD,OAGF,AAAI,EAAY,SAAW,GACzB,EAAQ,CAAC,cAAa,IAGxB,GAAM,GAAkB,EAAU,gBAClC,GAAI,GAAmB,MACrB,EAAO,GAAI,OACP,6CAA6C,EAAS,OAC1D,OAGF,GAAI,GACJ,IACE,EACI,EAAK,4BAA4B,EAAiB,SAC/C,GACP,EAAO,GACP,OAGF,GAAM,GAAsC,GACtC,EAAkB,GAClB,EAAgC,GACtC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,EAAM,KAAK,GACX,EAAe,KAAK,QAEtB,EAAY,KAAI,MAAhB,EAAoB,EAAa,WAGnC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,GAAM,GAAmB,GAAI,YAC7B,EAAiB,OAAS,SAAC,GAEzB,GAAM,GAAc,EAAM,OAAe,OACnC,EAAQ,EAAM,QAAQ,GAC5B,EAAe,GAAS,EACxB,AAAI,EAAe,QAAQ,QAAU,IACnC,EAAQ,CACN,cAAa,EACb,YAAW,EACX,WAAY,GAAwB,GACpC,OAAQ,EAAU,OAClB,YAAa,EAAU,YACvB,YAAa,EAAU,YACvB,oBAAqB,EAAU,uBAIrC,EAAiB,QAAU,SAAA,GACvB,MAAA,GAAO,6CAA6C,EAAI,OAC5D,EAAiB,kBAAkB,EAAW,SAIpD,EAAW,QAAU,SAAA,GAAS,MAAA,GAC1B,2DACA,eAAc,EAAS,KAAI,qCAC3B,yCACJ,EAAW,WAAW,WAOlB,EAAA,UAAA,4BAAR,SACI,EAAiC,GAInC,OAHM,GAAsB,GACtB,EAAY,EAAM,IAAI,SAAA,GAAQ,MAAA,IAAS,EAAK,QAC5C,EAAqC,GACvB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAM,MAAM,QAAQ,SAAA,GAClB,GAAM,GAAe,GAAS,GAC9B,GAAI,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,sDACA,KAAI,EAAY,MAGtB,GADA,EAAU,KAAK,GACX,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,8BAA8B,EAAY,sBAE9C,EAAW,GAAQ,EAAM,EAAU,QAAQ,MAKjD,GAAI,EAAU,SAAW,EAAM,OAC7B,KAAM,IAAI,OACN,uDACA,KAAI,EAAU,OAAM,8CACpB,KAAI,EAAM,OAAM,OAEtB,MAAO,IAEX,KAEa,GAAmC,SAAC,GAC/C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,gBAyCH,GAAA,MAAA,KAAA,QAAA,GAAA,SACxB,GAAI,IAAiB,eA2CD,GAC3B,MAAO,IAAI,IAAa,GC5V1B,YA8BI,EAAmC,EACnC,EAAwB,GAC1B,EAAc,GACd,EAAgB,GAAwB,EACxC,EAAc,GAAsB,EACpC,EAAc,EAAe,GAC7B,GAAI,GAAkB,EAEhB,EAAkB,SAAC,GACvB,SAAQ,KAAK,SAAA,GACX,GAAM,GAAW,EACb,GAAE,EAAkB,EAAS,OAAU,GAAc,GAEzD,SAAW,GACJ,IAEF,GAGT,WAAuB,GACrB,EACI,GAAY,MAAQ,MAAM,QAAQ,IAAa,EAAS,OAAS,EACjE,WAAM,MAAA,wCAGZ,WAAuB,EAAuB,GAC5C,EACI,GAAiB,GAAK,GAAiB,EACvC,WAAM,MAAA,kDACF,sBAAqB,KAC7B,EACI,GAAe,GAAK,GAAe,EACnC,WAAM,MAAA,kDACF,oBAAmB,KAC3B,EACI,GAAe,EACf,WAAM,MAAA,uDACF,sBAAqB,EAAa,qBAClC,IAAG,KAGb,MAAO,SAAQ,IAAI,EAAS,IAAI,ICvElC,YAqCI,EAAqB,6HACvB,AAAI,IAAe,MACjB,GAAc,IAGV,EAAY,EAAY,WAAa,KAAO,KAAM,SAAS,MACf,EAAY,UAGxD,EAAW,EAAU,IACvB,SAAA,GACI,MAAA,GAAU,EAAU,EAAY,YAAa,CAAC,SAAU,OAE1D,EAAqB,EACrB,EAAmB,GAEP,EAAY,YAAc,KACxC,CAAA,EAAM,QAAQ,IAAI,IADJ,CAAA,EAAA,UACd,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAU,EAAY,WAAY,EAClC,WAFJ,EAAA,EAAA,8BAFE,GAAS,EAMT,EAAiB,EAAU,IAAI,SAAA,GAAY,MAAA,GAAS,gBAEpD,EAAsB,GACtB,EAAoB,EAEV,EAAY,YAAc,KACtC,CAAA,EAAM,QAAQ,IAAI,IADN,CAAA,EAAA,UACZ,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAgB,EAAY,WAAY,EACxC,WAFJ,EAAA,EAAA,wBAFE,SAAO,EAKb,CAAA,EAAO,oBAaL,EAAiC,EACjC,EACA,GAFiC,MAAA,KAAA,QAAA,GAAA,wEAS7B,SAAe,SAAC,GAClB,MAAA,IAAyB,EAAW,CAAC,YAAW,KAC9C,EAAc,GAAqB,GAEzC,CAAA,EAAO,EAAY,EAAU,EAAgB,oBA4B3C,GADJ,GAAA,GAAA,KAIE,MAAO,UACI,EAAiC,EACjC,GADiC,MAAA,KAAA,QAAA,GAAA,4GAqD1C,GAjDM,EAAyB,EAAS,IAAI,WAAM,MAAA,KAC5C,EAKF,GACE,EACF,GAAe,KAAO,EAAY,IAAI,WAAM,MAAA,KAAS,GACnD,EAAmC,GACzC,EAAS,QAAQ,SAAC,EAAqB,GACrC,GAAI,GAAc,EAClB,EAAoB,QAAQ,QAAQ,SAAA,GAClC,GAAM,GAAY,gBAAkB,GAChC,EAAa,aAAa,MAC1B,EAAa,MAEX,EAAe,GAAqB,GACtC,GAAmB,EAAa,OAE9B,EAA8B,WAClC,EAAuB,GAAc,GACrC,AAAI,EAAoB,IAAe,MACrC,GAAoB,GAAc,IAGpC,EAAoB,GAAY,KAAK,CACnC,cAAe,EACf,YAAW,EACX,UAAW,KAIf,AAAI,GAAe,KACjB,EAAY,QAAQ,SAAC,EAAY,GAC/B,AAAI,IAAe,EAAa,MAC9B,KACA,EAAa,GAAe,MAIhC,IAGF,EAAuB,KAAK,EAAa,MACzC,GAAe,MAIf,CAAC,EAAa,MAAM,SAAA,GAAS,MAAA,KACzB,QAAkB,EAAY,OAAO,SAAC,EAAG,GAAM,MAAA,CAAC,EAAa,KAC7D,GAAI,OACN,kDACG,GAAgB,KAAK,MAAK;GAC7B,yCACG,GAAuB,KAAK,MAAK,MAKpC,SACF,EAAuB,OAAO,SAAC,EAAa,EAAa,GACvD,MAAI,IACF,EAAY,KAAK,GAEZ,GACN,IAED,EAAsB,GAC5B,EAAoB,QAAQ,SAAA,GAC1B,EAAS,GAAG,MAAM,QAAQ,SAAA,GACxB,GAAM,GAAW,EACZ,CAAC,EAAe,SAAS,KAAa,GAAN,KAAY,EACjD,EAAU,KAAK,OAGH,CAAA,EAAM,EAAqB,WAArC,SAAU,EAAA,OAEV,EAAmC,GACrC,EAAoB,EACxB,EAAoB,QAAQ,SAAA,GAI1B,OAHM,GAAa,EAAS,GAAG,MAAM,OAEjC,EAAa,EACR,EAAI,EAAG,EAAI,EAAY,IAC9B,GAAc,EAAQ,EAAoB,GAAG,WAO/C,OAHM,GAAc,GAAI,aAAY,GAC9B,EAAkB,GAAI,YAAW,GACnC,EAAoB,EACf,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAS,GAAI,YAAW,EAAQ,EAAoB,IAC1D,EAAgB,IAAI,EAAQ,GAC5B,GAAqB,EAAO,WAG9B,GAAM,GAAiB,EAAoB,GAC3C,EAAe,QAAQ,SAAA,GACrB,GAAM,GAAa,EAAY,MAC3B,EAAa,YACb,EAAa,YAAc,EAAa,WACtC,EACF,GAAc,EAAY,CAAC,EAAa,gBAC5C,OAAW,KAAQ,GACjB,EAAiB,GAAQ,EAAgB,KAI7C,GAAqB,IAGvB,CAAA,EAAO,SCpPX,AA+BA,GAAM,IAAyB,2BACzB,GAAY,mBAClB,GAAA,WAcE,WAAY,EAAc,GAgCxB,GAvCO,KAAA,eAAiB,OAQxB,AAAI,GAAe,MACjB,GAAc,IAEhB,KAAK,iBAAmB,EAAY,iBACpC,KAAK,WAAa,EAAY,WAC9B,KAAK,mBAAqB,EAAY,mBAEtC,AAAI,EAAY,WAAa,KAC3B,GACI,MAAO,GAAY,WAAc,WACjC,WAAM,MAAA,gIAGV,KAAK,MAAQ,EAAY,WAEzB,KAAK,MAAQ,KAAM,SAAS,MAG9B,EACI,GAAQ,MAAQ,EAAK,OAAS,EAC9B,WAAM,MAAA,4DAGV,AAAI,MAAM,QAAQ,IAChB,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,+CACF,sBAAqB,EAAK,OAAM,QAE1C,KAAK,KAAO,EAER,EAAY,aAAe,MAC3B,EAAY,YAAY,MAAQ,KAClC,KAAM,IAAI,OACN,sEAEN,KAAK,YAAc,EAAY,aAAe,GAG1C,SAAA,UAAA,KAAN,SAAW,yGACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,2FAIA,SAAO,OAAO,OAAO,CAAC,OAAQ,KAAK,gBAAiB,KAAK,aAC/D,EAAK,KAAO,GAAI,UAEV,EAAyC,CAAC,CAC9C,MAAO,CAAC,uBACR,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,oBACpC,gBAAe,GAGjB,EAAK,KAAK,OACN,aACA,GAAI,MACA,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,KACX,cAEJ,AAAI,EAAe,YAAc,MAC/B,EAAK,KAAK,OACN,oBACA,GAAI,MAAK,CAAC,EAAe,YAAa,CAAC,KAAM,KAC7C,qBAGW,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,WAE7C,GAFM,EAAW,EAAA,OAEb,EAAS,GACX,MAAA,CAAA,EAAO,CACL,mBAAoB,GAA6B,GACjD,UAAW,CAAC,KAGd,KAAM,IAAI,OACN,gEACG,GAAS,OAAM,YAYpB,EAAA,UAAA,KAAN,uIAC6B,MAAA,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,KAAK,qBAE5D,GAFM,EAAqB,EAAA,OAEvB,CAAC,EAAmB,GACtB,KAAM,IAAI,OACN,cAAc,KAAK,KAAI,4BACpB,GAAmB,OAAM,uCAC5B,sFAIU,CAAA,EAAM,EAAmB,eAAvC,SAAc,EAAA,qCAEV,EAAU,+CAA+C,KAAK,KAAI,IAGtE,AAAI,KAAK,KAAK,SAAS,OACrB,GAAW,+UAOX,GAAW,uEAGP,GAAI,OAAM,UAUlB,GARM,EAAgB,EAAY,cAC5B,EAAkB,EAAY,gBAC9B,EAAc,EAAY,YAC1B,EAAc,EAAY,YAC1B,EAAS,EAAY,OACrB,EAAsB,EAAY,oBAGpC,GAAiB,MAAQ,GAAmB,KAC9C,KAAM,IAAI,OACN,2BAA2B,KAAK,KAAI,mEAMtC,IAAmB,KACL,CAAA,EAAM,KAAK,YAAY,IADrC,CAAA,EAAA,UACI,EAAU,EAAA,OACf,EAAA,EAAA,GAAa,EAAA,EAAA,oBAGV,SAA4B,CAChC,cAAa,EACb,YAAW,EACX,WAAU,EACV,oBAAmB,EACnB,YAAW,EACX,YAAW,EACX,OAAM,GAGF,EAAc,EAAY,iBAChC,AAAI,GACF,GAAU,iBAAmB,GAG/B,CAAA,EAAO,SAGK,EAAA,UAAA,YAAd,SAA0B,2IAOxB,IALM,EAAa,MAAM,QAAQ,KAAK,MAAQ,KAAK,KAAK,GAAK,KAAK,KAC5D,EAAmB,GAAS,GAA3B,EAAM,EAAA,GAAE,EAAM,EAAA,GACf,EAAa,KAAK,kBAAoB,EAEtC,EAAc,GACpB,EAAA,EAAoB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAT,EAAK,EAAA,GACd,EAAY,KAAI,MAAhB,EAAoB,EAAM,SAK5B,IAFM,EAAsB,GACtB,EAAsC,GAC5C,EAAA,EAA2B,EAAA,EAAA,EAAA,EAAA,OAAA,IACzB,IADS,EAAY,EAAA,GACrB,EAAA,EAAmB,EAAA,EAAa,MAAb,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,AAAI,KAAK,oBAAsB,KAC7B,EAAY,KAAK,KAAK,mBAAmB,IAEzC,EAAU,KAAK,EAAa,EAAO,SAKrC,MAAK,sBACP,GAAA,EAAU,MAAI,SAAd,GAAkB,CAAA,EAAM,QAAQ,IAAI,KADlC,CAAA,EAAA,UACF,EAAA,MAAA,EAAA,EAAA,OAAA,CAAkB,EAAA,2BAGJ,MAAA,CAAA,EAAM,GAAyB,EAAW,CACxD,YAAa,KAAK,YAClB,UAAW,KAAK,MAChB,WAAY,KAAK,qBAHb,SAAU,EAAA,OAKhB,CAAA,EAAO,CAAC,EAAa,GAAwB,WA5M/B,EAAA,iBAAmB,eA8MrC,iBAayB,GACvB,GAAM,GAAY,EAAI,YAAY,KAC5B,EAAkB,EAAI,YAAY,KAClC,EAAS,EAAI,UAAU,EAAG,GAC1B,EACF,EAAkB,EAAY,EAAI,UAAU,GAAmB,GACnE,MAAO,CAAC,EAAS,IAAK,eAGK,GAC3B,MAAO,GAAI,MAAM,GAAY,mBAAqB,KAG7C,GAAM,IACT,SAAC,EAAa,GACZ,GAAI,MAAO,QAAU,aAChB,IAAe,MAAQ,EAAY,WAAa,MAInD,MAAO,MAEP,GAAI,GAAS,GAMb,MALA,AAAI,OAAM,QAAQ,GAChB,EAAS,EAAI,MAAM,SAAA,GAAW,MAAA,IAAa,KAE3C,EAAS,GAAa,GAEpB,EACK,GAAK,EAAK,GAGd,MAEb,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAwEf,EAAc,GACjC,MAAO,IAAI,IAAY,EAAM,eAS3B,EAAc,GAChB,MAAO,IAAK,EAAM,GC3XpB,AAuBA,GAAA,IAAA,WACE,WAA6B,GAAA,KAAA,eAAA,EAEvB,SAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,qBAEhB,KAEA,GAAA,WACE,WACqB,GAAA,KAAA,YAAA,EAGf,SAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,YAAY,SAE5B,iBAwBI,EAAmC,EACnC,EAA0B,GAC5B,GAAI,UAAU,SAAW,GACvB,GAAM,GACD,EAAkC,eAAiB,MACnD,EAAkC,aAAe,KACtD,MAAI,GACK,GAAI,IAAkB,GAI7B,SAAQ,KACJ,yNAIG,GAAI,IAAkB,CAAC,cAAe,SAK/C,gBAAQ,KACJ,yNAIG,GAAI,IAAkB,CAC3B,cAAe,EACf,YAAW,EACX,WAAU,EACV,eAAc,gBAqBhB,GAEF,MAAO,IAAI,IAAiB,GCpH9B,oYCAA,AAuDA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MAExC,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,MAAK,GAC5B,EACW,SAAC,EAAwB,GACxC,SAAQ,GAA4B,EAAO,EAAG,MAC9C,EACI,EAAG,OAAS,GAAmB,GAC/B,WAAM,MAAA,mEACV,EAAK,CAAC,IACC,EAAQ,QAAQ,EAAI,IAE7B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAEO,GAAU,EAAG,CAAC,SAAQ,KC1EnC,AA6CA,YACI,EAAiB,EAAiB,EAClC,SADkC,AAAA,IAAA,QAAA,GAAA,IAClC,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAK,EAAgB,EAAG,IAAK,UACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WAAM,MAAA,kEACF,cAAa,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE3C,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,sCAAsC,EAAU,UAC/C,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,kCAAkC,EAAW,UAC5C,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACxD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAExD,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,IAEJ,EAAQ,YACX,EAAiB,EAAiB,EAAY,IAG9C,EAA4B,CAAC,EAAG,EAAK,EAAG,GAExC,EAA0B,CAAC,WAAU,EAAE,WAAU,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAO,GAAQ,EAAK,MAGT,IAAS,EAAG,CAAC,QAAO,KC9GjC,AA+CA,YACI,EAA4B,EAAe,EAC3C,GACF,GAF6C,AAAA,IAAA,QAAA,GAAA,GAC3C,AAAA,IAAA,QAAA,GAAA,GACE,EAAQ,EACV,KAAM,IAAI,OAAM,iDAAiD,GAEnE,GAAM,GAAW,EAAgB,EAAS,UAAW,SAAU,SACzD,EAAe,EAAS,MAAK,OAAA,CAAE,IAE/B,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,IACC,EACH,EAAQ,OACJ,EAAQ,EAAU,CAAC,EAAS,OAAQ,EAAO,EAAS,GACxD,IAGA,EAAuB,CAAC,QAAS,GACjC,EAAqB,CAAC,MAAK,EAAE,QAAO,EAAE,SAAQ,GAEpD,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAS,EAAG,CAAC,QAAO,KCxEjC,AA+CA,YAAsC,EAAiB,GACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAgBnC,GAdA,AAAI,GAAQ,MACV,GAAO,EAAG,MAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAG,WAEnC,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OACzC,EAAK,QAAQ,SAAA,GACX,EACI,GAAQ,GAAK,EAAO,EAAG,KACvB,WAAM,MAAA,+CAA+C,GAAG,KAAO,GAC3D,aAAY,OAGlB,EAAG,MAAQ,EACb,MAAO,GAAG,QAGZ,GAAM,GAA0B,CAAC,EAAG,GAC9B,EAAwB,CAAC,KAAI,GAEnC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IAAO,EACxC,KAAqB,GAAW,MAGzB,IAAY,EAAG,CAAC,WAAU,KC5EvC,YA0DI,EAA6B,EAC7B,GACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,mBAC5C,EACF,EAAgB,EAAa,cAAe,mBAEhD,EACI,GAAc,MAAQ,EAAa,GAAK,OAAO,UAAU,GACzD,WAAM,MAAA,uDACF,YAAW,KACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,gDAAgD,EAAQ,OAClE,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,6CACF,YAAW,EAAa,QAChC,EACI,EAAQ,MAAM,KAAO,EAAa,MAAM,GACxC,WAAM,MAAA,uCACC,GAAQ,MAAM,GAAE,QAAQ,EAAa,MAAM,GAAE,MAChD,oEACR,EACI,EAAa,GAAK,OAAO,UAAU,GACnC,WAAM,MAAA,4DACF,IAAG,KAIX,GAAM,GAAe,GAAO,GAAK,EAAS,SAAU,GAC9C,EACF,GAAO,GAAK,EAAc,SAAU,GAClC,EAA0B,GAAU,GAC1C,MAAO,IAAK,GAAO,EAAe,GAAoB,SAGjD,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KC9FnD,2CCAA,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,mDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,oEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,AA6BA,GAAI,IAyBJ,YACI,EAEA,GAEF,GAFE,AAAA,IAAA,QAAA,GAAA,GAEE,EAAc,EAChB,KAAM,IAAI,OACN,kEAEN,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,4DAElB,GAAI,GAAc,GACd,EAAc,GACd,EAAU,GACV,EAAU,GACV,EAAe,GACnB,GAAK,EAAqB,eAAgB,YACxC,EAAc,WAEZ,MAAQ,YAAe,aAAe,YAAkB,WAC1D,EAAc,WAEZ,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAER,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAEA,EAAe,YAAc,KACvC,EAAe,OAEf,MAAM,IAAI,OACN,6OAIA,YAAY,EAAc,YAAY,OAE5C,GAAI,GACF,GAAM,GAAgC,EACtC,GAAI,GACC,EAA4B,WACzB,EACN,KAAM,IAAI,OACN,yGAMR,GAAM,GAAS,GAAU,GAAY,EAAO,aAC5C,GAAI,GAAU,MACZ,GAAM,GAA2B,CAAC,OAAM,GAClC,EAAyB,CAAC,YAAW,GAC3C,MAAO,GAAO,UACH,GAAY,EACZ,GAGP,GAAA,GAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAMV,EAEJ,AAAI,EACF,EAEK,EAAe,WAAW,MAAM,aAAa,EAAG,EAAG,EAAO,GAAQ,KAClE,AAAI,GAAe,EACxB,EAAQ,EAAiC,KACpC,AAAI,IAAW,IACpB,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAEpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA4B,EAAG,EAAG,EAAO,GAC7C,EAAO,GAAoB,aAAa,EAAG,EAAG,EAAO,GAAQ,MAE/D,GAAI,GACJ,GAAI,IAAgB,EAClB,EAAS,GAAI,YAAW,QAExB,GAAM,GAAY,EAAQ,EAC1B,EAAS,GAAI,YAAW,EAAY,GACpC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAU,EAAG,EAAU,EAAa,EAAE,EAC7C,EAAO,EAAI,EAAc,GAAW,EAAK,EAAI,EAAI,GAIvD,GAAM,GAAqC,CAAC,EAAQ,EAAO,GAC3D,MAAO,IAAS,EAAQ,EAAU,qBAuBhC,EACA,iIAQF,GAPI,EAAO,EAAgB,EAAK,MAAO,YACvC,AAAM,YAAe,IAEb,GAAoB,EAC1B,EAAO,GAAK,EAAmB,SAC/B,EAAkB,WAEhB,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,OACN,wDAAwD,EAAK,KAAI,KAKvE,GAHM,EAAkB,EAAK,MAAM,MAAM,EAAG,GAArC,EAAM,EAAA,GAAE,EAAK,EAAA,GACd,EAAQ,EAAK,OAAS,EAAI,EAAI,EAAK,MAAM,GAE3C,EAAQ,GAAK,IAAU,EACzB,KAAM,IAAI,OACN,wCACA,sBAAqB,IAG3B,GAAI,EAAK,QAAU,WAAa,EAAK,QAAU,QAC7C,KAAM,IAAI,OACN,kCAAkC,EAAK,MAAK,0CAIrC,MAAA,CAAA,EAAM,EAAK,eAIxB,IAJM,EAAO,EAAA,OACP,EAAa,EAAK,QAAU,UAAY,IAAM,EAC9C,EAAQ,GAAI,mBAAkB,EAAQ,EAAS,GAE5C,EAAI,EAAG,EAAI,EAAS,EAAO,EAAE,GAGpC,IAFM,EAAO,CAAC,EAAG,EAAG,EAAG,KAEd,EAAI,EAAG,EAAI,EAAO,KAGzB,GAFM,EAAQ,EAAK,EAAI,EAAQ,GAE3B,EAAK,QAAU,WACjB,GAAI,EAAQ,GAAK,EAAQ,EACvB,KAAM,IAAI,OACN,qDACA,kCAAiC,EAAK,cAEnC,EAAK,QAAU,SACpB,GAAQ,GAAK,EAAQ,KACvB,KAAM,IAAI,OACN,mDACA,oCAAmC,EAAK,MAIhD,AAAI,IAAU,EACZ,GAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,GAElB,EAAK,GAAK,EAAQ,EAIhB,EAAI,EAAI,EACd,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAGjC,MAAI,IAAU,MACZ,GAAO,MAAQ,EACf,EAAO,OAAS,EACV,EAAM,EAAO,WAAW,MACxB,EAAY,GAAI,WAAU,EAAO,EAAO,GAC9C,EAAI,aAAa,EAAW,EAAG,IAEjC,AAAI,IAAS,GACX,EAAK,UAEP,CAAA,EAAO,QAGF,GAAM,IAAa,EAAG,CAAC,YAAW,+DCrOrC,EAAgB,GAClB,GAAI,EAAO,KAAO,EAChB,KAAM,IAAI,OACN,0DACA,sBAAqB,EAAO,KAAI,MAEtC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,4DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OACN,sDACA,uBAAsB,EAAQ,MAAK,MAEzC,GAAI,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAAO,KAC3C,KAAM,IAAI,OACN,iEACG,GAAQ,MAAM,EAAQ,KAAO,GAAE,QAAQ,EAAO,OAGvD,GAAI,EAAO,OAAS,EAClB,KAAM,IAAI,OACN,qDACA,kBAAiB,EAAO,MAAK,MASnC,OANM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAIjD,EAAU,EACL,EAAI,EAAG,EAAI,EAAa,OAAS,EAAG,EAAE,EAC7C,GAAW,EAAa,GAG1B,GAAM,GAAa,EAAO,MAEpB,EAAc,EAAa,QACjC,EAAY,MAGZ,OADI,GAAY,EACP,EAAI,EAAW,EAAI,EAAO,KAAM,EAAE,EACzC,GAAa,EAAW,GACxB,EAAY,KAAK,EAAW,IAG9B,GAAM,GACE,GAAe,EAAO,OAAO,IAAI,SAAA,GAAU,MAAA,GAAS,IAAU,OAAA,CACjE,IAAG,MAAM,EAAG,GAEjB,MAAO,CAAC,EAAa,EAAS,EAAW,6DCtDvC,EAAiB,EAAiB,GACpC,GAAM,GAAY,EAAQ,KAAO,EAAK,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAClE,EAAY,EAAQ,KAAO,EAAK,EAAQ,KAAO,EAAI,EAEnD,EAAa,wDACf,yCAAwC,EAAQ,OAChD,qBAAoB,EAAQ,MAAK,YAAY,GAC7C,gBAAe,EAAQ,mBAAmB,EAAQ,KAEtD,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OAAM,EAAa,mBAAkB,EAAQ,OAEzD,GAAI,EAAM,OAAS,EAAY,GAAQ,KAAO,GAC5C,KAAM,IAAI,OACN,EACA,2BAA0B,GAAY,GAAQ,KAAO,MAE3D,GAAI,EAAQ,OAAS,EAAW,EAAM,OAAS,EAC7C,KAAM,IAAI,OACN,EAAa,oBAAmB,GAAW,EAAM,OAAS,KAEhE,OAAS,GAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,GAAI,EAAQ,MAAM,KAAO,EAAQ,MAAM,GACrC,KAAM,IAAI,OACN,EACA,mBAAkB,EAAC,MAAM,EAAQ,MAAM,GAAE,sBAAsB,EAAC,MAC5D,EAAQ,MAAM,GAAE,OAG5B,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAO,EAAU,EAAE,EAC7C,GAAI,EAAQ,MAAM,EAAI,KAAc,EAAM,EAAI,GAC5C,KAAM,IAAI,OACN,EACA,mBAAkB,GAAI,GAAQ,MAC1B,EAAQ,MAAM,EAAI,GAAS,cAAc,GAAI,GAAQ,MACrD,EAAM,EAAI,GAAS,kBAoB7B,EAAiB,EAAiB,GACpC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OAAM,0DACZ,EAAQ,OAEd,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OACN,6DAA6D,GAGnE,GAAI,EAAM,SAAW,GACnB,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAEd,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAIhB,GAAoB,EAAO,EAAS,eAalC,EAAqB,EACrB,GAWF,OATM,GAAc,EAAQ,MAAM,OAC5B,EAAa,EAAc,EAAK,EAAQ,MAAM,EAAc,GAAK,EAKjE,EAAU,EAAM,OAElB,EAAY,EACP,EAAI,EAAW,EAAI,EAAS,EAAE,EACrC,GAAa,EAAM,GAGrB,GAAM,GAAgB,EAAY,EAAK,EAAI,EACrC,EAAa,GAAc,EAAQ,OAAS,EAE5C,EAAc,GAAe,EAAM,MAAM,EAAG,IAAW,OAAA,CAAE,IACzD,EAAa,GAAc,GACjC,MAAO,CAAC,UAAS,EAAE,WAAU,EAAE,UAAS,EAAE,QAAO,EAAE,WAAU,sFCnJ/D,YAqBI,EAAmB,EAAiB,GACtC,GAAM,GAAY,EAAM,MAAM,OAC9B,EACI,IAAc,EAAM,OACpB,WAAM,MAAA,iBAAiB,EAAS,sBAAsB,EAAK,SACvD,iCAAgC,EAAS,QACjD,EACI,IAAc,EAAK,OACnB,WAAM,MAAA,iBAAiB,EAAS,qBAAqB,EAAI,SACrD,iCAAgC,EAAS,QAEjD,mBAAS,GACP,EACI,EAAM,GAAK,EAAK,IAAM,EAAM,MAAM,GAClC,WAAM,MAAA,iBAAiB,EAAS,YAAY,EAAC,YAAY,EAAC,KACtD,KAAI,GAAM,GAAK,EAAK,IAAE,gCAAgC,EAAC,MACjD,EAAM,MAAM,GAAE,QALrB,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,eAUgB,GAGzB,OAFM,GAAO,GACT,EAAO,EACJ,EAAO,GACZ,AAAI,EAAO,GACT,EAAK,KAAK,GAEZ,GAAQ,EACR,IAEF,MAAO,eAKL,EAAiB,EAAe,GAElC,OADM,GAAO,GACJ,EAAO,EAAG,EAAO,EAAM,OAAQ,IACtC,EAAK,GAAQ,KAAK,KAAM,GAAI,GAAQ,EAAM,IAAS,EAAQ,IAE7D,MAAO,eAML,EAAmB,EAAgC,EACnD,GAEF,OADM,GAAiB,EAAO,QACrB,EAAI,EAAW,OAAQ,EAAI,EAAW,OAAQ,IACrD,EAAW,KAAK,GAElB,OAAS,GAAI,EAAG,EAAI,EAAe,IACjC,AAAI,IAAM,EACR,EAAW,GAA0B,EAErC,GAAW,OACP,EAAwB,EACxB,GACJ,EAAW,OAGf,MAAO,GAGT,YACI,EAAgC,EAChC,GACF,MAAI,IAAkB,EACb,EAGF,EAAkB,GAAgB,GAG3C,YAAuB,EAAuB,GAE5C,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAe,IACjC,EAAW,KAAK,EAAyB,GAE3C,MAAO,eAKL,EAAsB,EAAwB,EAC9C,EAAiB,EAAe,EAAmB,EACnD,EACA,GACF,GAAM,GAAY,EAAW,OACzB,EAAkB,GAAI,OAAM,GAC5B,EAAgB,GAAI,OAAM,GAC1B,EAAoB,GAAI,OAAM,GAClC,GAAI,EAAa,QAAU,EAAsB,GAC/C,GAAM,GAAY,EAAa,GAIzB,EAAgB,EAAsB,EAC5C,EAAkB,GACd,EAAW,EAAW,EAAe,EAAO,GAChD,EAAgB,GACZ,EAAS,EAAW,EAAe,EAAK,GAC5C,EACI,GAAsB,EAAS,EAAW,EAAe,OAE7D,QAAS,GAAO,EAAG,EAAO,EAAW,IACnC,EAAgB,GAAQ,GACpB,EAAW,EAAO,EAAS,EAAY,EAAM,GACjD,EAAc,GACV,GAAY,EAAS,EAAK,EAAS,EAAY,EAAM,GACzD,EAAkB,GAAQ,GAAe,EAAS,EAAM,GAI5D,MAAO,CACL,MAAO,EACP,IAAK,EACL,QAAS,eAOT,EAAmB,EAAgC,EACnD,EAAyB,GAI3B,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAEnB,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAc,GAClC,AAAI,EAAY,GAAK,GACnB,GAAgB,GAGlB,EAAW,GAAQ,EAGvB,MAAO,eAML,EAAiB,EAAgC,EACjD,EAAuB,GAIzB,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAAO,sBAE1B,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAY,GAChC,AAAI,EAAU,GAAK,GACjB,GAAgB,OAAO,kBAEzB,EAAW,GAAQ,EAIvB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,KAErC,GAAM,GAAW,EAAW,GAC5B,AAAI,EAAW,GAAK,GAClB,GAAW,IAAM,GAEnB,EAAW,GAAK,GAAW,EAAG,EAAW,GAAI,EAAW,IAE1D,MAAO,eAIL,EAAmB,EAAc,GACnC,GAAI,GAAS,EAAQ,GACrB,MAAI,GAAgB,GAAK,GAAS,GAAU,OAC1C,GAAS,GAGJ,cAIL,EAAmB,EAAwB,EAC3C,EAAsB,EAAc,GAEtC,GAAI,GAAQ,EAAa,GACnB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAY,GAAK,GAAQ,EAAe,GAAK,GAAQ,GAAS,OAChE,CAAI,EAAS,EAIX,EAAQ,OAAO,iBAGf,EAAQ,OAAO,kBAKnB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAQ,GACV,IAAS,GAIX,EAAQ,GAAW,EAAG,EAAO,EAAW,GAEjC,cAIL,EAAiB,EAAuB,EACxC,EAAsB,EAAc,GAEtC,GAAI,GAAO,EAAY,GACjB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAW,GAAK,GAAS,EAAgB,GAAK,GAAS,GAAQ,OACjE,CAAI,EAAS,EAGX,EAAO,OAAO,iBAGd,EAAO,OAAO,kBAKlB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAO,GACT,IAAQ,GAMV,AAAI,EAAS,EAEX,EAAO,GAAW,EAAG,EAAM,GAG3B,EAAO,GAAW,GAAI,EAAM,EAAW,GAGlC,cAQL,EAAiB,EAAiB,GAGpC,OADI,GAAkB,EAAK,OAClB,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,GAAI,EAAK,GAAK,GACZ,EAAkB,EAClB,MAIJ,OAAS,GAAI,EAAkB,EAAG,EAAI,EAAK,OAAQ,IACjD,GAAI,EAAM,GAAK,GAAK,EAAK,KAAO,EAAM,GACpC,MAAO,GAGX,MAAO,eAGyB,EAAiB,GAEjD,OADI,GAAa,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EACrD,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAc,EAAM,GAAK,EAAQ,GAEnC,MAAO,eAIL,EAAe,EAAwB,GAEzC,GAAI,GACE,EAAQ,EAAE,MAAM,OACtB,AAAI,MAAO,IAAU,SACnB,EAAM,CAAI,GAAK,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,IACzC,AAAI,EAAM,OAAS,EACxB,EAAS,EAAM,OAAO,GAAI,OAAM,EAAQ,EAAM,QAAQ,KAAK,IAE3D,EAAS,EAAM,QAEjB,EAAO,QAAQ,SAAA,GACb,EACI,IAAM,GAAI,WAAM,MAAA,wDAEtB,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAAI,OAAM,GAAO,KAAK,IACzB,AAAI,MAAO,IAAS,SACzB,EAAK,CAAI,GAAI,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,KACvC,AAAI,EAAK,OAAS,EACvB,EAAQ,EAAK,OAAO,GAAI,OAAM,EAAQ,EAAK,QAAQ,KAAK,KAExD,EAAQ,EAEV,EAAQ,EAAM,IAAI,SAAC,EAAG,GACpB,MAAI,IAAK,EACA,EAEP,GACI,IAAM,GACN,WAAM,MAAA,qDACC,GAAC,kCAAkC,EAAC,OACxC,EAAE,MAAM,GAAK,EAAO,MAGxB,CAAC,EAAQ,wSCjWlB,AA8DA,GAAA,IAAA,WAAA,cAYE,SAAA,UAAA,aAAA,WACE,MAAQ,MAAK,YACR,WAiBA,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,IAEnB,KASA,GAAA,WAOE,aACE,KAAK,aAAe,GAMf,SAAA,OAAP,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UAMnB,EAAA,SAAP,SAAwC,GACtC,EAAiB,SAAS,aAAa,EAAI,WACvC,CAAC,EAAK,EAAI,aAElB,iBA2BI,GACF,EACI,EAAI,WAAa,KACjB,WAAM,MAAA,gFAEV,EACI,MAAO,GAAI,WAAc,SACzB,WAAM,MAAA,sDACF,MAAO,GAAI,YACnB,EACI,EAAI,UAAU,OAAS,EACvB,WAAM,MAAA,sFAGV,GAAiB,SAAS,gFC/K5B,AAsBA,GAAM,IAAuB,KAChB,GAAuB,eAGhC,EACA,EAAoD,GACtD,MAAI,IAAW,MACb,GAAU,MAEL,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,mBAInE,MAAO,GAAO,QAAQ,mBAAqB,GAAK,GACA,GAGlD,YACI,EAAoB,EACpB,GACF,GAAI,GAAiB,GAOrB,GANA,AAAI,IAAa,IAAW,GAAa,KACvC,GAAiB,IAEnB,AAAI,GAAa,IAAW,GAAa,IACvC,GAAiB,IAEf,GACF,GAAM,GAAQ,EAAO,YAAY,KAC3B,EAAQ,EAAS,YAAY,KAEnC,GAAI,IAAU,EACZ,KAAM,IAAI,OACN,yCAAyC,EAAK,KAC9C,cAAa,IAIrB,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,IACzC,GAAM,GAAc,GAAW,GACzB,EAAgB,GAAW,GACjC,GAAI,CAAC,GAAY,EAAa,GAC5B,KAAM,IAAI,OACN,iCACA,aAAY,EAAW,iBAAiB,EAAa,MAI7D,GAAM,GACF,GAAa,GAAU,EAAS,GAAQ,GACtC,EAAe,GAAa,GAC9B,EACA,GAAQ,GAEZ,GAAI,EAAW,SAAW,EAAa,OACrC,KAAM,IAAI,OACN,yCAAyC,EAAW,OAAM,OAC1D,cAAa,EAAa,OAAM;GAChC,cAAa,EAAU;GACvB,cAAa,EAAY,MAE/B,OAAS,GAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAI,EAAW,GACf,EAAI,EAAa,GAEvB,GAAI,CAAC,EAAU,EAAG,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAC,OAAO,EAAC,cAAc,EAAC,OAAO,EAAC;EACzD,cAAa,EAAU;GACvB,cAAa,EAAY,mBAUC,EAAuB,GACzD,IAAK,KAAK,WAAM,MAAA,GAAK,QAAQ,WAAM,MAAA,mBAGH,EAAoB,GACpD,GAAM,GAAM,MAAO,IAAa,UAAY,MAAO,IAAa,UACxD,MAAO,IAAa,UACxB,CAAC,GACD,EACJ,MAAI,IAAS,IAAW,GAAU,EAAoB,KAClD,GAAS,IAAa,GAAU,EAAsB,IAEjD,GAAsB,EAAQ,EAAK,SAAC,EAAG,GAAM,MAAA,IAAK,IAEpD,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,iBAGlC,EAAW,EAAW,GAIvD,GAHA,AAAI,GAAW,MACb,GAAU,MAER,CAAC,GAAS,EAAG,EAAG,GAClB,KAAM,IAAI,OAAM,8BAA8B,EAAC,kBAAkB,GAIrE,YAAkB,EAAW,EAAW,GACtC,MAAI,CAAC,SAAS,IAAM,CAAC,SAAS,GACrB,GAEL,QAAM,IAAM,MAAM,IAAM,KAAK,IAAI,EAAI,GAAK,eAO5C,EAA6B,EAAa,GAC5C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,GAAI,EAAO,GAAK,GAAO,EAAO,GAAK,EACjC,KAAM,IAAI,OACN,sBAAsB,EAAO,GAAE,SAAS,EAAG,WAAW,eAM5D,EAAqB,GAGvB,OAAO,GAAI,cAAa,IAAS,QAAQ,GAAI,cAAa,4MCzJ5D,GAGM,IAAU,QCHhB,cAiCE,KAAM,IAAI,OAAQ,kBAkBlB,KAAM,IAAI,QAAS,kBAKnB,KAAM,IAAI,+BAAgC,IAC1C,QAAQ,KAAK,sEAIiB,GAC9B,AAAI,KAAM,QAAQ,iCAChB,QAAQ,KACJ,EAAM,6FAYZ,EAAO,iCASP,MAAO,iBAyBP,MAAO,GAAO,qBAgCQ,GAEtB,MAAO,GAAO,QAAQ,eA4CpB,EAA6B,GAC/B,MAAO,GAAO,KAAK,EAAU,eAcP,GACtB,GAAM,GAAU,GAAsB,GACtC,EAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,wBAmCI,GACrC,MAAO,GAAO,KAAK,eA4BA,GACnB,MAAO,GAAO,KAAK,eAkBM,GACzB,MAAO,GAAO,WAAW,iBAWzB,MAAO,GAAO,sBAUd,MAAO,GAAO,wBAQc,GAC5B,EAAO,cAAc,eAOK,GAC1B,MAAO,GAAO,YAAY,eAQO,GAEjC,MAAO,GAAO,mBAAmB,eAmB/B,EAAc,EACd,GAAA,MAAA,KAAA,QAAA,GAAA,GACK,EAAO,gBAAgB,EAAM,EAAS,iBAY7C,MAAO,GAAO,oBASY,EAAsB,GAChD,KAAM,YAAY,EAAc,GC5XlC,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCnE3B,AAqDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCvErC,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OAGjC,GAFA,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAED,EAAG,QAAU,SAAW,EAAG,QAAU,QACvC,MAAO,IAAS,EAAI,GAGtB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,WAAW,EAAI,GACnC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAC/B,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KC3E3B,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAEA,GAAM,EAAG,CAAC,KAAI,KCpE3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GAEpC,MADA,GAAK,CAAC,IACF,EAAG,QAAU,YACR,EAAQ,WAAW,GAGrB,EAAQ,IAAI,IAClB,EAAgC,KAAiB,OAGzC,IAAM,EAAG,CAAC,KAAI,KCrD3B,AAoCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAuCA,YAAiC,GAC/B,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,+DACV,EACI,EAAQ,QAAU,EAClB,WAAM,MAAA,uDACF,IAAG,EAAQ,UAEnB,GAAM,GACF,EAAQ,IAAI,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAG,UAAU,EAAK,UAEtD,EAAc,EAAS,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAI,EAAE,QAAU,EAAY,MAC1B,KAAM,IAAI,OACN,8DAIR,EAAS,QAAQ,SAAA,GACf,GAAI,CAAC,GAAiB,EAAE,MAAO,EAAY,OACzC,KAAM,IAAI,OACN,8DAIR,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,GACE,GAGH,EAAqB,EAE3B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC/E7B,YAuBqC,EAAgB,GACnD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,GAAI,EAAK,EAAK,OAAS,EAAI,KAAO,EAAO,EAAI,EAC3C,MAAO,GAGX,MAAO,eAIL,EAAqB,EAAqB,GAK1C,OAJI,GAAO,EAAU,OAAS,EAAU,OACpC,EAAM,GACR,EAAS,EACT,EAAY,EACL,EAAM,EAAG,EAAM,EAAM,IAC9B,AAAI,EAAK,QAAQ,KAAS,GACxB,EAAI,KAAK,EAAU,MAEnB,EAAI,KAAK,EAAU,MAGvB,MAAO,eAIL,EAAkB,GAGpB,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,EAAK,QAAQ,KAAS,IACxB,EAAS,KAAK,EAAO,IAGzB,GAAM,GAAc,EAAK,IAAI,SAAA,GAAO,MAAA,GAAO,KAC3C,MAAO,CAAC,EAAU,eAIhB,EAAiB,GACnB,GAAM,GAAiB,EAAK,IAAI,SAAA,GAAK,MAAA,KACrC,MAAO,IAAiB,EAAO,EAAgB,eAI7C,EAAa,EAAgB,GAC/B,EACI,GAAqB,EAAM,GAC3B,WAAM,MAAG,GAAG,2CACR,aAAY,EAAI,aAAa,EAAI,yBAQR,EAAgB,GAEjD,GAAI,GAAqB,EAAM,GAC7B,MAAO,MAGT,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,AAAI,EAAK,QAAQ,KAAO,IACtB,EAAO,KAAK,GAGhB,SAAK,QAAQ,SAAA,GAAQ,MAAA,GAAO,KAAK,KAC1B,cAI8B,GACrC,MAAO,GAAK,IAAI,SAAC,EAAM,GAAM,MAAA,CAAC,EAAG,KAC5B,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAE,KACxB,IAAI,SAAA,GAAK,MAAA,GAAE,iBAGe,EAAiB,GAEhD,OADM,GAAgB,GACb,EAAI,EAAO,EAAS,EAAI,EAAM,EAAE,EACvC,EAAI,KAAK,GAEX,MAAO,GC3GT,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KCxF3B,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAIA,IAAM,EAAG,CAAC,KAAI,KCzF3B,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC7EjC,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,AAAI,GAAQ,MACV,GAAO,GAET,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KChFjC,AAqCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC/C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA2CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAEH,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAQ,EAAG,CAAC,OAAM,KC7D/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,YAoGI,EACA,EAAuC,EACvC,EAA4B,EAC5B,GAD4B,AAAA,IAAA,QAAA,GAAA,QAM9B,GAAM,GAAgB,EAAW,GAC3B,EACE,EAAW,OAAA,CAAE,IACf,EAAc,GAAwB,GAE5C,MAAO,IACH,EAAY,EAAc,EAAS,EAAW,EAC9C,KAAyB,KAAsB,eAIjD,EACA,EAAqC,EACrC,EAAoC,EACpC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,gBACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GAEjB,EACJ,GAAI,IAAe,eACjB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACrD,IAAe,gBACxB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAE9D,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,EAAc,GAC7D,eAOF,EACA,EACA,EACA,EAA4C,EAC5C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,SACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAE9B,EACA,EACJ,GAAI,IAAe,QACjB,EAAc,eACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACxD,IAAe,QACxB,EAAc,gBACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAEjE,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,GAAO,EACtD,eAQF,EACA,EACA,EAAkC,EAClC,EACA,EAAuC,EACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACvC,AAAA,IAAA,QAAA,GAAA,gBACE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GACnC,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WACtB,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAU,EAAA,EAAA,OAElC,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAC9B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GACf,EAAA,GAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAEjB,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,SAIrB,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAW,GAC1C,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAW,EAAU,IAGvC,CACL,UAAS,EACT,WAAU,EACV,SAAQ,EACR,QAAO,EACP,WAAU,EACV,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,aAAY,EACZ,YAAW,EACX,aAAY,EACZ,YAAW,EACX,sBAAqB,EACrB,qBAAoB,EACpB,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,eA2CX,EACA,EACA,EACA,EAA4C,EAC5C,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,gBAEE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GAE5C,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WAC/B,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,OAE3C,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAE3C,EAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAC5B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAe,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGhC,EACF,GAAuB,EAAa,GAClC,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,SAAU,EAAA,EAAA,UAAW,EAAA,EAAA,SAK/B,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAU,EAAW,GACpD,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAU,EAAW,EAAU,IAGjD,CACL,UAAS,EACT,WAAU,EACV,QAAO,EACP,SAAQ,EACR,QAAO,EACP,WAAU,EACV,SAAQ,EACR,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,YAAW,EACX,aAAY,EACZ,YAAW,EACX,YAAW,EACX,aAAY,EACZ,YAAW,EACX,qBAAoB,EACpB,sBAAqB,EACrB,qBAAoB,EACpB,cAAa,EACb,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,GAIf,YACI,EAA2B,EAAmB,EAC9C,EAAkB,GACpB,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAY,GAGtB,YACI,EAA2C,EAC3C,EAAqB,EAAgB,EACrC,GACF,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAa,EAAQ,GACrB,EAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAe,GAChB,GAAa,EAAY,EAAI,GAAW,EAAS,EAAG,GACzD,EACI,GAAW,GACX,WAAM,MAAA,2BAA2B,EAAY,uEAGjD,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAc,EAAY,EAAY,eAI5C,EACA,EAAmB,EAAgB,GAAA,AAAA,IAAA,QAAA,GAAA,GACrC,GAAM,GAAqB,GAAuB,EAAW,GAC7D,MAAO,MAAK,MACP,GAAW,GAAM,GAAS,GAAK,EAAS,GAAsB,GAGrE,YAAyB,GACvB,MAAI,OAAO,IAAU,SACZ,CAAC,EAAO,EAAO,GAEpB,EAAM,SAAW,EACZ,CAAC,EAAM,GAAI,EAAM,GAAI,GAEvB,EAGT,YAA0B,GAExB,MAAO,OAAO,IAAU,SAAW,CAAC,EAAO,EAAO,GAAS,EAc7D,YAAgC,EAAoB,GAClD,MAAI,IAAY,EACP,EAGF,EAAc,GAAa,GAAM,GAAW,GAGrD,YACI,EAA4C,EAC5C,EAAiB,EAAsB,EACvC,EAAsB,EACtB,EACA,GAEF,GAAI,GACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CAAC,IAAK,EAAK,OAAQ,EAAK,KAAM,EAAK,MAAO,EAAK,KAAM,GAC/D,GAAM,GAAW,GACb,CAAC,EAAU,GAAU,EAAc,EAAc,EAAK,GAC1D,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GACF,KAAK,IAAI,EAAI,GAAY,GAAK,EAAe,EAAe,GAC1D,EACF,KAAK,IAAI,EAAI,GAAW,GAAK,EAAc,EAAc,GACvD,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAC9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,gBAClC,IAAQ,QACjB,EAAU,CAAC,IAAK,EAAG,OAAQ,EAAG,KAAM,EAAG,MAAO,EAAG,KAAM,SACvD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,WAC1C,MAAO,IAAQ,UACxB,GAAM,GAAM,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GACzD,EAAS,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC5D,EAAO,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC1D,EAAQ,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC3D,EAAW,IAAQ,GAAK,IAAW,GAAK,IAAS,GAAK,IAAU,EAClE,QACA,WACJ,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,GAC3C,EAAY,GACP,GAAW,EAAe,EAAM,GAAU,EAAe,EAC1D,GACJ,EAAW,GACN,GAAU,EAAc,EAAO,GAAS,EAAc,EAAG,OAE9D,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,UAAS,EAAE,SAAQ,GAGtC,YACI,EAA4B,EAAiB,EAC7C,EAAiB,EAAqB,EACtC,EAAqB,EAAqB,EAC1C,EAAqB,GAMvB,GAAI,GACA,EACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,GAER,GAAM,GAAW,GACb,CAAC,EAAS,EAAU,EAAS,GAAI,EAAa,EAAG,EAAa,EAC9D,GACJ,EAAW,EAAS,GACpB,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAW,KAAK,KAAK,EAAU,GAC/B,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EACD,GAAY,GAAK,EAAe,EAAe,EAC9C,EAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EAAQ,KAAK,MAAM,EAAgB,GACnC,EAAO,EAAgB,EACvB,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAE9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,MAAK,EAAE,KAAI,EAAE,KAAM,gBAC/C,IAAQ,QACjB,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,SAER,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,GACnD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,OAEnD,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,SAAQ,EAAE,UAAS,EAAE,SAAQ,GAQhD,YACI,EAAe,GACjB,GAAI,CAAC,EACH,MAAO,GAET,OAAQ,OACD,QAEH,MAAO,MAAK,MAAM,OACf,OAEH,MAAO,MAAK,KAAK,OACd,QACH,MAAO,MAAK,MAAM,WAElB,KAAM,IAAI,OAAM,wBAAwB,gBAIZ,GAC1B,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,MAAO,KAAS,GAAK,IAAS,GAAK,IAAS,cAI1C,EAA0B,GAC5B,MAAO,IAAkB,IAAY,GAAkB,eAUjB,GAEtC,GAAI,IAAe,OACjB,MAAO,eACF,GAAI,IAAe,OACxB,MAAO,gBAEP,KAAM,IAAI,OAAM,sBAAsB,GC5mB1C,AAoDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,WACxC,EAAY,EAElB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,mDAAmD,EAAI,KAAI,MAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAIJ,MAFA,GAAK,CAAC,IAEF,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UACvC,EAAI,QAGN,EAAQ,QAAQ,EAAK,IAGxB,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAElE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAG,OAEf,EACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KClHnC,AA6EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,WAE5C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAGrB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GAEJ,SAAK,CAAC,IAEC,EAAQ,UAAU,EAAK,IAG1B,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAEjE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAI,OAEhB,EACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KCzJvC,YAmBuC,EAAoB,GACzD,GAAM,GAAO,EAAO,GAAG,OACvB,EAAO,QAAQ,SAAC,EAAO,GACrB,EACI,EAAM,SAAW,EACjB,WACI,MAAA,kBAAkB,EAAI,sBAAsB,EAAC,sBAC7C,6BAA4B,EAAI,SAG1C,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,kBAAkB,EAAI,iCAAiC,GAAO,GAAC,MAEzE,GAAM,GAAa,EAAO,GAC1B,EAAO,QAAQ,SAAC,EAAO,GACrB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EACK,IAAM,GAAU,EAAM,KAAO,EAAW,GACzC,WAAM,MAAA,kBAAkB,EAAI,uBAAuB,EAAC,MAAM,EAAK,KAC3D,0CAAyC,EAAU,MACnD,oCAAmC,EAAC,qBAKlB,EAAoB,GAElD,OADM,GAAc,EAAO,GAAG,QACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAY,IAAS,EAAO,GAAG,GAEjC,MAAO,GClDT,AAqEA,YAAmC,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/D,EAAO,EAAQ,QAAU,EAAG,WAAM,MAAA,uCAElC,GAAI,GAAW,GAAqB,EAAS,UAAW,UAExD,AAAI,EAAS,GAAG,QAAU,aACxB,EAAS,QAAQ,SAAA,GACf,GAAI,EAAO,QAAU,YACnB,KAAM,IAAI,OAAM;uBACD,EAAO,MAAK,QAKjC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAS,GAAG,OAAO,GAChD,EAAW,GAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAC7D,GAAI,GAAc,KAAc,EAC9B,MAAO,IAAO,GAAI,GAIpB,GADA,EAAW,EAAS,OAAO,SAAA,GAAK,MAAA,GAAE,KAAO,IACrC,EAAS,SAAW,EACtB,MAAO,GAAS,GAGlB,GAAM,GAAS,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QACnC,GAAuB,EAAQ,GAE/B,GAAM,GAAM,EAAQ,OAAO,EAAU,GACrC,SAAK,GACE,GAGH,EAAuB,EACvB,EAAoB,CAAC,KAAI,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC/GjC,AAsCA,YAAoC,GAClC,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAAwB,CAAC,EAAG,GAElC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,QAAQ,GAC5B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAU,EAAG,CAAC,SAAQ,KCjDnC,AA8DA,YACI,EAAiB,EAAwB,GAC3C,GAAM,GAAK,EAAgB,EAAG,IAAK,SAEnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,kCAGlB,GAAM,GAA+B,SAAC,EAAS,GACvC,GAAA,GAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,UAA6B,EAAI,EAAQ,GACzC,EAAK,CAAC,IACC,EAAQ,MAAM,EAAI,EAAQ,IAG7B,EAAsB,CAAC,EAAG,GAC1B,EAAoB,CAAC,MAAK,EAAE,KAAI,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCrF/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAI,EAAQ,KAAK,GACvB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA8CA,YACI,EAA+B,EAC/B,EAA+B,EAC/B,EAAwB,GAC1B,GAAM,GACF,EAAgB,EAAY,aAAc,iBACxC,EACF,EAAgB,EAAY,aAAc,iBACxC,EAAY,EAAgB,EAAU,WAAY,iBAClD,EAAQ,EAAgB,EAAM,OAAQ,iBACtC,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAE7B,EAAW,GAAO,CAAC,EAAO,GAAK,GAC/B,EAAW,GAAO,EAAU,GAC5B,EAAgB,GAAI,EAAU,GAG9B,EAAY,EAAI,MAAM,GACtB,EAAY,EAAI,MAAM,GAAK,EAC3B,EAA8B,CAAC,EAAW,GAC1C,EAAI,GAAM,EAAK,CAAC,EAAG,GAAI,GACvB,EAAI,GAAM,EAAK,CAAC,EAAG,GAAY,GAC/B,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GACnC,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GAEnC,EACF,GAAI,EAAI,GAAQ,GAAI,GAAK,IACrB,EAAI,EAAI,GAAQ,GAAI,EAAa,MACnC,EAAiB,EAAI,GAAK,GAAO,GAAQ,IAC/C,MAAO,CAAC,EAAM,MAGH,IAAgB,EAAG,CAAC,eAAc,KC/E/C,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAE7C,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,iBAAiB,EAAG,KAAI,2CAC1B,EAAW,SAEnB,EACI,EAAM,SAAW,EAAW,OAC5B,WAAM,MAAA,mBACF,EAAM,OAAM,8CACZ,EAAW,SAEnB,EACI,EAAG,MAAM,GAAK,IAAS,EACvB,WAAM,MAAA,yBACI,EAAG,MAAM,GAAE,2CACjB,+BAA8B,EAAW,KAAK,OAAM,QAAQ,KAEpE,GAAM,GAA0B,SAAA,GAC9B,MAAO,GAAQ,eAAe,EAAI,EAAY,IAG1C,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,MAAK,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,iBC1FX,GACpC,GAAI,GACJ,MAAI,GAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAG,EAAE,OACxB,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACvC,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAErD,EAAM,EAGD,EChCT,AAwDA,YACI,EAAyB,EACzB,EACA,EACA,EACA,GACF,AAAI,GAAmB,MACrB,GAAkB,MAEpB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,AAAI,GAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAG9C,EACI,EAAM,OAAS,EAAU,KACzB,WAAM,MAAA,iFAEV,EACI,GAAW,MAAQ,EAAM,OAAS,EAAQ,KAC1C,WAAM,MAAA,+EAEV,EACI,GAAU,MAAQ,EAAM,OAAS,EAAO,KACxC,WAAM,MAAA,8EAGV,GAAM,GAAgB,GAAM,GAEtB,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,EAAO,EAAW,IAEtB,EAAQ,UACX,EAAK,GAAS,GAAQ,GAAS,GAAY,GAAS,GACpD,GAAS,GAAS,IAGlB,EAA+B,CACnC,EAAG,EACH,MAAO,EACP,OAAQ,EACR,KAAM,EACN,SAAU,GAGN,EAA6B,CAAC,gBAAe,GAE7C,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAO,GAAQ,EAAK,EAAG,OAGzB,YAAkB,GAChB,MAAI,IAAK,KACA,KAEL,EAAE,OAAS,EAEN,EAAQ,EAAG,CAAC,EAAE,OACZ,EAAE,OAAS,EACb,EACE,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACpC,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAEjD,KAGI,IAAY,EAAG,CAAC,WAAU,KCrGvC,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAE/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC9E3C,AA4CA,YACI,EAAsB,GACxB,GAAI,GAAQ,EAAgB,EAAG,cAAe,KACxC,EAAS,EAAM,MAErB,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,CAAE,GAAI,IAAM,EAAI,IAAM,IACxC,KAAM,IAAI,OAAM,2CAA2C,EAAK,MAGlE,GAAI,EAAM,OAAS,EAAM,KACvB,KAAM,IAAI,OAAM,+BAA+B,EAAM,OAAM,iBACvD,EAAM,KAAI,KAGhB,GAAI,EAAM,OAAS,EAAM,MAEvB,OADM,GAAW,EAAM,MAAM,QACtB,EAAS,OAAS,EAAM,QAC7B,EAAS,QAAQ,GAEnB,EAAQ,EAAQ,EAAO,GAKzB,OAFM,GAAa,EAAM,MACnB,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAM,OAAS,EAAG,GAAK,EAAG,IACrC,GAAI,EAAW,KAAO,EAAM,GAC1B,EAAK,GAAK,UACD,EAAM,MAAM,KAAO,EAC5B,KAAM,IAAI,OACN,mBAAmB,EAAM,6BAA6B,EAAK,MAGnE,GAAM,GAAO,EAAK,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,KAAI,OAAO,SAAA,GAAK,MAAA,IAAK,IAEjE,GAAI,EAAK,SAAW,EAClB,MAAO,IAAM,GAGf,GAAM,GAAU,SAAC,GAA2B,MAAA,GAAQ,KAAK,EAAO,IAE1D,EAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,MAAK,EAAE,WAAU,GAElD,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAa,MAGb,IAAc,EAAG,CAAC,aAAY,KC5F3C,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAyCA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,eACnC,EACK,GAAgB,EACjB,WAAM,MAAA,uBAAuB,EAAY,aACrC,+BAA8B,EAAY,QAElD,GAAM,GAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,aAAY,EAAE,aAAY,GAE3D,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAM,EAAQ,KAAK,EAAI,EAAc,GAC3C,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAc,EAAG,CAAC,aAAY,KC7B3C,YAAmB,GACjB,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCYrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCDrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KC5BrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCnCrC,AA8DA,YACI,EAAiB,EACjB,EACA,EACA,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GACjE,EACI,IAAY,EAAQ,MAAM,GAC1B,WAAM,MAAA,oCAAoC,EAAO,gBAC7C,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,GAAO,GACL,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GACxC,EACY,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAS,EAAG,CAAC,QAAO,KC3EjC,YACI,EAAiB,EAA6B,EAC9C,EACA,EAAiC,EACjC,GADA,AAAA,IAAA,QAAA,GAAA,OAAiC,AAAA,IAAA,QAAA,GAAA,GAEnC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,MAG9C,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAQ,GACjD,WAAM,MAAA,yDACF,eAAc,EAAM,kBAAkB,EAAQ,OACtD,EACI,IAAe,MACf,WAAM,MAAA,sCACF,EAAU,0CAElB,GAAM,GAAW,EACb,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,KAC7D,EAAU,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAA4B,CAAC,EAAG,GAChC,EAA8B,CAAC,EAAG,GAElC,EAAmB,OAEnB,EAAM,GACP,EAAuB,EAAuB,EAAS,EACxD,EAAkB,EAAW,GAEjC,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAGxC,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,QAGhD,IAAS,EAAG,CAAC,QAAO,KCpHjC,AAoDA,YACI,EAAmE,EACnE,EAAkB,EAClB,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KAC1D,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,KAG9C,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,GAAM,GAAU,IAAe,OAAS,EAAS,GAAK,EAAS,GACzD,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAY,EAEZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEE,EAAM,EAAQ,eAAe,EAAM,EAAQ,GAEjD,SAAK,CAAC,EAAM,IAEL,GAGH,EAAoC,CAAC,GAAI,EAAM,OAAM,GAErD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,EAAE,WAAY,GAEtD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAqB,GAEzB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxF3D,YACI,EAAiB,EACjB,EACA,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IACH,EAAa,EAAI,EAAS,EAAS,EAAK,OAAQ,MAGzC,IAAkB,EAAG,CAAC,iBAAgB,KCxDnD,AA8DA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAExE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAA+B,EAAS,GACxC,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,QACf,WAAM,MAAA,sCACF,EAAU,4CAElB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,GAC5C,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GAExC,EAAqB,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,GAEzD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,KAGI,IAAS,EAAG,CAAC,QAAO,KC3HjC,AA6CA,YACI,EAGA,EAAO,EAAkB,EACzB,GACF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACvE,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,EAAO,KAGzD,GAAM,GAAU,EAAS,GACnB,EAAW,EAAK,MAAM,GAC5B,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,GAEhD,MAAO,GAAQ,eAAe,EAAM,EAAQ,IAGxC,EAAoC,CAAC,GAAI,GAEzC,EAAkC,CAAC,IAAG,GAEtC,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxE3D,YACI,EAAiB,EACjB,EAGA,EAA0C,GAC5C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IAAoB,EAAa,EAAI,EAAS,EAAS,MAGnD,IAAkB,EAAG,CAAC,iBAAgB,KCtDnD,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AAqDA,YACI,EAAsB,EAAU,EAAmB,GAA7B,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,IAAmB,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAG,MAC9C,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,IAE5B,GAAM,GAAe,GAAiB,EAAG,EAAG,MAAM,GAC9C,EAAQ,EAAQ,OAAO,EAAW,EAAc,EAAW,GAG/D,GAFA,EAAK,CAAC,IAEF,GAAe,MACjB,GAAM,GAAqB,GAAuB,GAClD,EAAQ,GAAU,EAAO,GAE3B,MAAO,IAGP,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,EAAE,UAAS,EAAE,QAAO,GAEpD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KCnFjC,AAiEA,YACI,EAA0B,EAC1B,GAAA,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,gBAE7B,EAAe,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC/D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC9D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAEpE,EACI,EAAc,GAAa,EAC3B,WAAM,MAAA;MACN,EAAW,QAAQ,EAAS;MAC5B,EAAG,QAEP,EACI,EAAa,GAAa,EAC1B,WAAM,MAAA;MACN,EAAU,QAAQ,EAAS;UACvB,EAAG,QAEX,EACK,EAAc,GAAY,KAAe,EAC1C,WAAM,MAAA,8CACF,EAAY,EAAS,WACrB,EAAU,sCAAsC,EAAG,QAE3D,GAAM,GAAiC,SAAA,GACnC,MAAA,GAAQ,aAAa,EAAI,EAAW,IAElC,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAAC,UAAS,EAAE,WAAU,GAEvD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAc,MAGP,IAAe,EAAG,CAAC,cAAa,KCvG7C,AA2EA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,sDACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAElC,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAGlB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,iEACF,mBAAkB,EAAO,mBAAmB,EAAS,OAE7D,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IACE,EAAM,EAAQ,gBAAgB,EAAK,EAAS,GAClD,SAAK,CAAC,EAAK,IACJ,GAGH,EAAsC,CAAC,EAAG,EAAK,OAAQ,GACvD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KC/InD,AA+CA,YAAe,GACb,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAA,GACnC,GAAM,GAAO,EAAQ,EAAI,CAAC,EAAG,OACvB,EAAS,EAAQ,KAAK,GACtB,EAAe,EAAE,MAAK,OAAK,EAAE,OAEnC,MAAO,GAAQ,EAAQ,IAGnB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAO,EAAG,CAAC,MAAK,KChE7B,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,GADA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IACzC,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAU,EAAgB,EAAQ,SAAU,cAElD,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA,gEACC,GAAG,KAAI,OAClB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,4DACC,GAAQ,KAAI,OACvB,EACI,IAAe,OACf,WAAM,MAAA,0DACF,0BAAyB,KAEjC,GAAI,GAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACzD,EAAe,IAGjB,GAAM,GAA2B,CAAC,EAAG,EAAK,OAAQ,GAC5C,EAAyB,CAAC,QAAO,EAAE,IAAG,EAAE,UAAS,GAEjD,EAAM,EAAO,UACH,GAAY,EACZ,GAEhB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAa,EAAG,CAAC,YAAW,KCvGzC,YA2BI,EAAmB,GAGrB,OAFM,GAAS,EAAQ,OACjB,EAAiB,GACd,EAAI,EAAG,EAAI,EAAQ,KAC1B,GAAM,GAAM,EAAS,EAAI,EACnB,EAAI,EAAQ,IAAQ,EACpB,EAAI,EAAS,EAAS,OAAS,EAAI,IAAM,EAC/C,AAAI,EAAI,GAAK,IAAM,GACjB,EAAK,QAAQ,GAGjB,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAQ,EAAQ,EAAQ,OAAS,EAAI,GACrC,EAAU,EAAS,OAAS,EAAI,EAChC,EAAS,EAAS,GACxB,AAAI,IAAS,MAAS,IAAU,GAAK,EAAS,IAC5C,EAAO,QAAQ,GAGnB,MAAO,eAIL,EAAkB,GAIpB,OAHM,GAAmB,GACnB,EAAI,KAAK,IAAI,EAAO,OAAQ,EAAO,QAEhC,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GACnC,AAAI,GAAK,MACP,GAAI,GAEN,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GAInC,GAHA,AAAI,GAAK,MACP,GAAI,GAEF,IAAM,EACR,EAAO,QAAQ,WACN,IAAM,EACf,EAAO,QAAQ,WACN,IAAM,GACf,GAAM,GAAS,wDACR,GAAM,QAAQ,EAAM,KAC3B,KAAM,OAAM,OAEZ,GAAO,QAAQ,GAGnB,MAAO,GCrFT,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,MAAM,EAAI,IAE5D,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,OAG/C,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AAoDA,YACI,EAA8B,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAa,EAAgB,EAAW,YAAa,QAAS,QAI9D,EAAiB,GAA2B,EAAG,MAAO,EAAG,OACzD,EAAgB,GAAY,EAAI,GAChC,EAAgB,GAAY,EAAI,GACtC,AAAI,EAAW,OAAS,GAGtB,EACI,EAAW,MAAM,KAAO,EAAG,MAAM,GACjC,WAAM,MAAA,mEAGZ,AAAI,EAAW,OAAS,GAEtB,GACI,EAAW,MAAO,EAAc,MAAO,oBAG7C,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,OAAO,EAAY,EAAe,GACtD,SAAK,CAAC,IACC,GAEH,EAAyB,CAC7B,UAAW,EACX,EAAG,EACH,EAAG,GAEL,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,OAGA,IAAQ,EAAG,CAAC,OAAM,KC5F/B,AAuCA,YAAsC,GACpC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAA0B,CAAC,EAAG,GACpC,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,IAAK,EAClC,KAAiB,OAEjB,IAAY,EAAG,CAAC,WAAU,KC9CvC,AA0DA,YACI,EAAsB,SAEpB,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAAY,GAAI,EAAI,GACpB,EAAQ,GAAU,GAClB,EAAc,GAAM,EAAI,GAC9B,MAAO,IAAM,EAAa,EAAO,MAGtB,IAAW,EAAG,CAAC,UAAS,KCvErC,AA2CA,YAAc,EAAuB,GACnC,GAAM,GAAM,EAAgB,EAAI,KAAM,OAChC,EAAM,EAAgB,EAAI,KAAM,OAEtC,EACK,GAAI,OAAS,GAAK,EAAI,OAAS,IAAO,GAAI,OAAS,GAAK,EAAI,OAAS,GACtE,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GACjD,EAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GAOvD,GALA,EACI,IAAY,EACZ,WAAM,MAAA,gEACC,GAAO,QAAQ,EAAO,OAE7B,EAAI,OAAS,GAAK,EAAI,OAAS,GACjC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,YACZ,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,CAAC,EAAK,eAClB,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAK,GACzB,MAAO,GAAQ,EAAM,CAAC,EAAK,YAE3B,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAK,GACzB,MAAO,OAIE,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,GACtB,SAAK,CAAC,IACC,GAGH,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCtD3B,AAyCA,YAAgC,GAC9B,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,EACI,EAAG,QAAU,SAAW,EAAG,QAAU,UACrC,WAAM,MAAA,8CAEV,AAAI,EAAG,QAAU,SACf,GAAK,GAAK,EAAI,YAGhB,GAAM,GAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KC1D3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAyCA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GAC3D,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,aAAc,GAEjD,EAAY,GAAQ,EAAG,KAAM,WAAM,MAAA,uCACnC,GAAM,GAAW,EAAG,MAAM,QAC1B,MAAI,GAAO,GAET,GACI,CAAE,GAAG,KAAO,IAAM,EAClB,WAAM,MAAA,iCAAiC,CAAG,GAAG,KAAO,GAAE,KAAK,EAAG,KAAI,MACtE,EAAO,EAAG,KAAO,EAAO,GAE1B,EAAS,OAAO,EAAM,EAAG,GAClB,EAAQ,EAAI,MAGR,IAAa,EAAG,CAAC,YAAW,KC1DzC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAqDA,YAAiC,EAAiB,GAChD,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,OAAQ,GAC3C,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OAEzC,GAAM,GAA0B,SAAC,EAAS,GACxC,GAAM,GAAM,EAAQ,KAAK,EAAI,GAC7B,SAAK,CAAC,IACC,GAGH,EAAe,CAAC,GAChB,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,GAE9B,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,EAAkC,MAG3B,IAAO,EAAG,CAAC,MAAK,KC5E7B,AAwCA,YACI,EAAiB,EACjB,EAKA,GAAA,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAc,MAChB,GAAa,GAIf,OAFM,GAAO,GAAO,CAAC,EAAS,GAAa,GACrC,EAAI,GAAW,EAAa,EAAU,EACnC,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAK,IAAI,EAAG,EAAG,GAEjB,GAAM,GAAgB,EAAQ,EAAK,WAAY,CAAC,EAAS,IACzD,GAAI,GAAc,KAChB,MAAO,GAEP,GAAI,EAAW,SAAW,EACxB,MAAO,IAAK,GAAW,EAAK,GAAI,CAAC,EAAW,GAAI,EAAG,IAC9C,GAAI,EAAW,SAAW,EAE/B,MAAO,IACI,GAAW,GAAW,EAAK,GAAI,GAC/B,CAAC,EAAW,GAAI,EAAW,GAAI,EAAG,IACxC,GAAI,EAAW,SAAW,EAE/B,MAAO,IAAK,GAAW,GAAW,GAAW,EAAK,GAAI,GAAI,GAAI,CACrD,EAAW,GAAI,EAAW,GAAI,EAAW,GAAI,EAAG,IAGzD,KAAM,IAAI,OACN,2CAEA,8BAA8B,EAAmB,OAAM,UAKpD,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAqCA,YACI,EAAoB,EAAsB,GAC5C,GAAM,GAAmB,CAAC,MAAK,EAAE,MAAK,EAAE,MAAK,GAE7C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAO,EAAO,IAAQ,GAAI,KAAM,GACxD,GC3CN,AAqCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC7C/B,AAuBO,GAAM,IAAwB,eASI,GACvC,MAAI,IAAU,GACL,EAEF,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KCpCrD,YA8BI,EAAgB,GAClB,GAAI,GAAO,GACP,EASJ,IAPA,AAAI,GAAU,GACZ,GAAM,EACN,EAAO,IAEP,EAAM,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KAG7C,CAAC,GACN,AAAI,EAAM,GAAe,IAAQ,EAC/B,EAAO,GAEP,EAAM,GAAe,EAAQ,EAAM,GAGvC,MAAO,eAIL,EAAkB,EAAc,GAGlC,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,IAAQ,EACV,EAAS,KAAK,EAAO,IAErB,EAAS,KAAK,GAGlB,MAAO,eAUL,EAAW,EAAiB,GAM9B,OALM,GAAU,EAAE,MAAM,GAElB,EAAwB,GAC1B,EAAY,EACZ,EAAY,EACP,EAAI,EAAG,EAAI,EAAM,IACxB,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAM,IAChC,EAAY,KAAK,EAAQ,MAAM,IAGjC,OAAS,GAAI,EAAO,EAAG,EAAI,EAAE,KAAM,IACjC,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,MAAO,CAAC,UAAS,EAAE,UAAS,EAAE,QAAO,EAAE,YAAW,2GC5FpD,AAoDA,YACI,EAAiB,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAW,EAAgB,EAAS,UAAW,SAAU,SAEzD,EAAyB,CAAC,EAAG,EAAI,QAAS,GAC1C,EAAuB,CAAC,KAAI,GAE5B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAa,GAAe,EAAM,EAAG,OAAO,GAC5C,EAAY,GAAyB,EAAI,EAAU,GAEnD,EACF,EAAQ,OAAO,EAAI,EAAQ,EAAU,CAAC,EAAS,OAAQ,GAC3D,SAAK,CAAC,EAAI,IAEH,EAAQ,EAAK,EAAU,cAGhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC5EjC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,QAAQ,EAAI,IAE9D,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC3DnC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,aAAa,EAAI,GACrC,SAAK,CAAC,EAAI,IACH,GAGH,EAA6B,CAAC,EAAG,EAAI,EAAG,GAE9C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAe,EAAG,CAAC,cAAa,KC/D7C,AAsCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCnD7B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GAEnC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,SAAS,IAAK,EACnC,KAAiB,OAEV,IAAW,EAAG,CAAC,UAAS,KC/CrC,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC/C/B,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC9C/B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAEhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAGH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,YAsCI,EAAyC,GAC3C,GAAM,IAAa,IAAU,IAAU,UAAa,MAAM,QAAQ,KAC9D,IAAU,YACZ,KAAM,IAAI,OACN,kFAGN,GAAI,IAAU,UAAY,GAAa,IACnC,CAAE,aAAiB,aACrB,KAAM,IAAI,OACN,6EAGN,GAAM,GAAkB,GAClB,EAA0B,GAChC,MAAO,IAAW,EAAO,EAAO,EAAe,GCrDjD,AA2CA,YAAsC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aACnC,MAAO,IAAQ,EAAI,GAAO,GAAQ,GAAK,MAG5B,IAAY,EAAG,CAAC,WAAU,KChDvC,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,QAC7B,EAAK,EAAgB,EAAG,IAAK,QACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAE3D,EAAqB,CAAC,EAAG,EAAI,EAAG,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC1D7B,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,UAAU,EAAI,GAClC,SAAK,CAAC,EAAI,IACH,GAGH,EAA0B,CAAC,EAAG,EAAI,EAAG,GAE3C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAY,EAAG,CAAC,WAAU,KC/DvC,YAkCyB,EAAe,EAAc,GACpD,GAAI,GAAO,EACT,KAAM,IAAI,OAAM,4CAGlB,GAAM,GAAuB,CAAC,MAAK,EAAE,KAAI,EAAE,IAAG,GAC9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,SAAS,EAAO,EAAM,IAAM,GAC/C,KAAiB,GAAU,GC1CjC,AA4CA,YACI,EAAiB,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IACzD,GAAM,GAAK,EAAgB,EAAG,IAAK,8BACnC,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA;sBACU,EAAG,KAAI,MAC3B,EACI,GAAW,GACX,WAAM,MAAA,+DACF,gCAA+B,EAAW,OAElD,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,6BACd,EAAK,EAAa,EAAM,EAAO,GAEnC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAoB,CAAC,EAAG,GAExB,EAAkB,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEpD,KAIE,IAA6B,EAAG,CAAC,4BAA2B,KCvFzE,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AA0DA,YAAc,GAEZ,SACI,GAAgB,GAAI,WAAM,MAAA,+CACvB,SAAC,EAAsB,GAE5B,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MACxC,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,WAAa,KAC1D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,kFAGN,GAAW,GACJ,EAAM,MAkCnB,YAAe,GAEb,SACI,GAAgB,GAAI,WAAM,MAAA,gDACvB,SAAC,EAAgC,GACtC,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,qFAGV,GAAM,GAAQ,GAAqB,EAAM,OAAQ,WAAY,MACvD,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,YAAc,KAC3D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,MAAA,OAAA,IAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,iGAGN,GAAW,GACJ,KA8Bb,YAA0D,GAKxD,SACI,GAAgB,GAChB,WAAM,MAAA,uDACH,SAAC,EAAM,GACZ,EACI,YAAa,GACb,WAAM,MAAA,wDACV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,6DACJ,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,UAAW,GACJ,CAAC,KAAM,EAAM,GAAS,MAAK,IAmCtC,YAAyC,GAKvC,SACI,GAAgB,GAChB,WAAM,MAAA,wDACH,SAAC,EAAgB,GACtB,EACI,MAAM,QAAQ,IAAS,EAAK,MAAM,SAAA,GAAO,MAAA,aAAe,KACxD,WAAM,MAAA,uEAEV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,iEACV,GAAM,GAAM,EAAO,UAAU,WAAM,MAAA,GAAC,MAAA,OAAI,IAAO,EAAM,GACrD,MAAI,IAAM,MACR,GACI,EAAI,MAAM,MAAO,EAAG,MACpB,yGAGN,GAAW,EAAI,OACR,GAkCX,YAAuB,EAAiB,GAEtC,EACI,GAAgB,GAChB,WAAM,MAAA,wDACV,EACI,GAAW,MACP,MAAM,QAAQ,IAAY,EAAQ,MAAM,SAAA,GAAK,MAAA,aAAa,MAC9D,WACI,MAAA,kFAGR,GAAM,GAAmB,GAAW,KACpC,GAAI,CAAC,GAEH,EAAU,GACV,OAAW,KAAW,GAAO,oBAC3B,EAAQ,KAAK,EAAO,oBAAoB,IAI5C,GAAM,GACF,EAAmB,EAAQ,OAAO,SAAA,GAAY,MAAA,CAAC,EAAS,YAAa,KAGnE,EAAmB,EAAQ,OACjC,EAAU,EAAQ,OAAO,SAAA,GAAY,MAAA,GAAS,YAC9C,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,kEACF,kCAAiC,EAAgB,kBACjD,eAER,GAAM,GAAmB,GACnB,EAAA,EAAA,UAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,EACI,EAAM,KAAK,SAAA,GAAK,MAAA,IAAK,OACrB,WAAM,MAAA,iMAGV,EACI,EAAM,OAAS,EACf,WAAM,MAAA,iEACF,oBAAmB,EAAM,KAAI,aAErC,GAAM,GAA6B,GACnC,SAAQ,QAAQ,SAAC,EAAG,GAClB,AAAI,EAAM,IAAM,MACd,GAAW,EAAE,MAAQ,EAAM,MAG/B,AAAI,GAAyB,MAG3B,EAAsB,QAAQ,SAAA,GAAK,MAAA,GAAW,EAAE,MAAQ,OAEnD,CAAC,MAAK,EAAE,MAAO,GA2CxB,YAAsC,GAEpC,MAAO,GAAO,WAAW,GAG3B,YAAoB,GAClB,GAAM,GAAmB,EAAM,OAAO,SAAA,GAAK,MAAA,IAAK,OAAM,OACtD,GAAI,EAAmB,EACrB,KAAM,IAAI,OACN;sEC7XR,AAuCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAuB,CAAC,EAAG,GACjC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,IAAI,IAAK,EAC5B,KAAiB,OAEV,IAAM,EAAG,CAAC,KAAI,KC/C3B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,SAAS,GAC7B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAW,EAAG,CAAC,UAAS,KChDrC,AAyCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAK7B,EAAW,GAAW,SAAC,GAI3B,GAAM,GAAQ,GAAI,GAAS,GAAI,KAEzB,EAAW,SAAC,GAChB,GAAM,GAAO,EAAI,EAAI,GAAQ,GAAI,KACjC,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAEL,IAAa,EAAG,CAAC,YAAW,KC9DzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAoB,EAAM,EAAG,OAC1C,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAS,OAG1D,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GAAgB,GAClB,EAAI,MAAO,GAAoB,EAAM,EAAG,QAC5C,EAAM,EAAQ,EAAK,GACnB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,iBAAkB,EAAM,SAAQ,GAEzD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCnE3B,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAGhB,GAAM,GAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IACN,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,IAAI,EAAW,GACnC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAEzB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KC/F3B,AAsDA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAAU,EAAgB,EAAQ,SAAU,cAKlD,GAHA,AAAI,IAAS,IACX,GAAO,EAAQ,KAAO,GAEpB,IAAS,EAAQ,KAAO,EAC1B,KAAM,OACF,gEACA,oBAAmB,EAAQ,KAAI,iBAAiB,IAGtD,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACX,EAAO,GAAI,EAAQ,EAAM,IACzB,EAAU,GAAI,EAAQ,GACtB,EACF,GAAI,GAAK,EAAS,WAAY,GAAI,GAAI,GAAI,GAAU,EAAM,KAC9D,SAAK,CAAC,IACC,GAGH,EAA2B,CAAC,OAAQ,GACpC,EAAyB,CAAC,KAAI,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAY,MAGZ,IAAa,EAAG,CAAC,YAAW,KCpFzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAO,GAAI,EAAI,EAAM,IACrB,EAAI,GAAI,EAAI,GACZ,EAAI,GAAI,GACR,EAAI,GAAI,EAAG,GACX,EAAI,GAAI,GACR,EAAM,GAAI,EAAQ,EAAM,EAAE,OAAQ,GAExC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,MAGI,IAAY,EAAG,CAAC,WAAU,KChFvC,AAyCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA2B,CAAC,EAAG,EAAI,EAAG,GAE5C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,WAAW,EAAI,IAClC,EAAgC,KAAiB,OAGjD,IAAa,EAAG,CAAC,YAAW,KCtDzC,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,WAAW,IAAK,EACnC,KAAiB,OAGV,IAAa,EAAG,CAAC,YAAW,KC9CzC,AAwCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,QAC1C,EAAK,EAAgB,EAAG,IAAK,YAAa,QAChD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA0B,CAAC,EAAG,EAAI,EAAG,GAC3C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IACjC,EAAgC,KAAiB,OAEjD,IAAY,EAAG,CAAC,WAAU,KCnDvC,AA0CA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,UAA2B,EAAG,MAAO,EAAG,OAGjC,GAAW,GAAU,EAAG,GAAI,GAAW,GAAW,EAAG,QAGjD,IAAa,EAAG,CAAC,YAAW,KCpDzC,AAwDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAC7B,EAAY,EAEd,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAEA,EAEJ,MAAI,GAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UAC9C,EAAI,EAAI,QAER,EAAI,EAAQ,QAAQ,EAAK,GAG3B,EAAK,CAAC,EAAK,IAEJ,GAGH,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAEhE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KCrHnC,AA2EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GAJiB,AAAA,IAAA,QAAA,GAAA,CAA+C,EAAG,EAAG,IAGtE,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE/B,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAErB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GACE,EAAI,EAAQ,UAAU,EAAK,GACjC,SAAK,CAAC,EAAK,IACJ,GAGH,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAE/D,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KClJvC,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAK,EAAgB,EAAG,IAAK,qBAE7B,EAAkC,CAAC,EAAG,GACtC,EACuB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,oBAAmB,GAErE,EAAS,EAAO,UACH,GAAmB,EACnB,GAEnB,MAAO,CAAC,OAAQ,EAAO,GAAI,QAAS,EAAO,OAGhC,IAAoB,EAAG,CAAC,mBAAkB,KC7EvD,YAsCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAM,EAAO,WACpB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAoB,GAAc,GAAQ,GACzD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC7C1C,YAuCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAK,EAAO,WACnB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAmB,GAAc,GAAQ,GACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC9C1C,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAS,GAA0B,EAAG,MAAO,GAC7C,EAAc,EAAO,GACrB,EAAa,GAAc,GAI3B,EAAW,GAAW,SAAC,GAC3B,GAAM,GAAmB,GAAO,GAE1B,EAAU,EAAiB,QAAU,EAAE,MACzC,EACA,GAAK,EAAG,EAAiB,OACvB,EAAM,GAAI,EAAS,GACnB,EAAQ,GAAI,EAAK,EAAM,GAEvB,EAAW,SAAC,GAChB,GAAM,GAAkB,EAAE,MAAM,QAChC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,GAAI,EAAI,EAAY,GAAK,EAAE,MAAO,YAAa,GAC5D,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAGL,IAAO,EAAG,CAAC,MAAK,KCpC7B,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAGpD,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GACF,GAA+B,EAAI,MAAO,GAC9C,EAAM,EAAQ,EAAG,GACjB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAGhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAEH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,AAuDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAEH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AAmCA,YAAmC,GACjC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAQ,GACR,EAAe,CAAC,GAChB,EAA2B,GACjC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,SAAK,CAAC,IACC,EAAQ,OAAO,IACrB,CAAC,EAAG,GAAK,KAAiB,SAAU,EAAO,EAAc,MAGjD,IAAS,EAAG,CAAC,QAAO,KC9CjC,AA4CA,YACI,EAAsB,EACtB,GADsB,AAAA,IAAA,QAAA,GAAA,MACtB,AAAA,IAAA,QAAA,GAAA,IACF,EAAI,EAAgB,EAAG,IAAK,WAC5B,GAAM,GAAO,GAAe,EAAM,EAAE,OAC9B,EAAQ,GAAK,EAAG,EAAM,GACxB,EAAgB,EAAM,MAC1B,AAAK,GACH,GAAgB,GAAqB,EAAM,MAAO,IAEpD,GAAM,GACF,GAAO,GAAI,GAAK,EAAG,WAAY,EAAQ,EAAO,KAC5C,EAAW,GAAK,EAAY,EAAM,GACxC,MAAO,CAAC,KAAM,EAAO,SAAQ,MAGlB,IAAU,EAAG,CAAC,SAAQ,KChBnC,YACI,EAA2B,EAC3B,EACA,GAOF,OANM,GAAQ,EAAgB,EAAM,OAAQ,gBACtC,EAAK,GAAqB,EAAG,IAAK,gBAClC,EAAK,GAAqB,EAAG,IAAK,gBAEpC,EAAQ,EACN,EAAY,GACT,EAAI,EAAG,EAAI,EAAU,OAAQ,KACpC,GAAM,GAAS,EAAU,GAAG,EAAO,EAAG,GAAI,EAAG,IAC7C,EAAU,KAAK,EAAO,IACtB,EAAU,KAAK,EAAO,IACtB,EAAQ,EAAO,GAIjB,OAFM,GAAmB,GACnB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,GAAK,EACzC,EAAK,KAAK,EAAU,IACpB,EAAK,KAAK,EAAU,EAAI,IAE1B,MAAO,CAAC,EAAM,MAEH,IAAe,EAAG,CAAC,cAAa,KCpE7C,AA4CA,YACI,EAAsC,EAAoB,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,eAC5C,EAAc,EAAQ,KACtB,EAAW,EAAQ,KACzB,GAAI,EAAc,EAChB,KAAM,IAAI,OACN,+DACG,GAAW,MAEpB,GAAI,EAAW,EACb,KAAM,IAAI,OAAM,gDAAgD,GAElE,EAAO,GAAQ,KAAK,SACpB,GAAM,GACF,IAAa,EAAI,EAAQ,EAAS,CAAC,EAAG,KAAO,EAC3C,EAAM,EAAO,cACf,SAAA,GAAW,MAAA,GAAQ,YAAY,EAAU,EAAY,EAAY,IACjE,CAAC,SAAQ,IAEb,MAAO,KAAa,EAAI,EAAQ,EAAK,CAAC,EAAI,OAAqB,KAGpD,IAAc,EAAG,CAAC,aAAY,KCpE3C,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,GAAY,MAAA,GAAQ,SAAS,EAAI,IAEjE,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KC1DrC,AAwCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCrD7B,AA0CA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAI,EAAG,QAAU,aACf,GAAM,GAAI,GAAS,GAAK,IAClB,EAAI,GAAU,GAAK,IACzB,MAAO,IAAQ,EAAG,GAGpB,MAAO,GAAQ,SAAS,IAGpB,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCtBrC,YACI,EAAyB,GAC3B,GAAM,GAAM,EAAgB,EAAI,KAAM,gBAChC,EAAM,EAAgB,EAAI,KAAM,gBAEtC,EACI,EAAI,OAAS,GAAK,EAAI,OAAS,EAC/B,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,EAAQ,EAAK,CAAC,EAAG,KAC9B,MAAO,IAAO,EAAM,MAGT,IAAe,EAAG,CAAC,cAAa,KCtD7C,AAmDA,YACI,EAAiB,EAAmC,GAAA,AAAA,IAAA,QAAA,GAAA,GACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAA0B,SAAC,EAAS,GACxC,SAAK,CAAC,IACC,EAAQ,IAAI,EAAI,EAAU,IAG7B,EAAoB,CAAC,SAAQ,EAAE,cAAa,GAC5C,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAM,EAAG,CAAC,KAAI,KC5C3B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,qDACH,GAAI,EAAG,CAAC,GAAW,MAGf,IAAQ,EAAG,CAAC,OAAM,KCT/B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCV/B,YACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,EACvD,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCX/B,YACI,EACA,EAIA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,GACnD,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCxC/B,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAEnC,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,cAAc,EAAG,KAAI,kCACvB,EAAW,SAEnB,EACI,EAAS,SAAW,EAAW,OAC/B,WAAM,MAAA,qBACF,EAAS,OAAM,kCAAkC,EAAW,SAEpE,EACI,EAAG,MAAM,OACL,SAAC,EAAG,EAAG,GACL,MAAI,GAAI,GAAK,GAAK,EAAW,OACpB,GACD,GAAI,EAAS,EAAI,GAAG,GAAK,EAAS,EAAI,GAAG,IACtC,EAAW,EAAI,KACnB,EAEA,GAET,IACJ,WAAM,MAAA,4BAA4B,EAAG,MAAM,MAAM,GAAE,kBAC/C,EAAS,WAAU,qCACnB,EAAW,aAEnB,GAAM,GAA0B,SAAA,GAC5B,MAAA,GAAQ,eAAe,EAAI,EAAY,IAErC,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,SAAQ,GAExD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,KCrHjD,AAwDA,YACI,EAAqB,EACrB,EAA0B,EAC1B,EAAqC,GACvC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,AAAI,GAAW,MACb,GAAU,GAEZ,AAAI,IAAQ,GACV,GAAM,SAGR,GAAM,GAAK,EAAgB,EAAO,IAAK,WACnC,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,yDACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAC1C,EACF,CAAC,EAAS,eAAgB,EAAS,eAOnC,EACJ,AAAI,IAAQ,OACV,EAAc,GACV,CAAC,EAAS,aAAc,EAAS,aAAc,GAEnD,EAAc,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,IAG7B,GAAM,GAAgB,EAAS,KAAO,GAAK,EAAS,KAAO,EACrD,EAAA,GAAA,CAAA,EAAA,SAAA,EAAA,SAAA,EAAA,GAAC,EAAA,EAAA,GAAiB,EAAA,EAAA,GAElB,EAAe,EAAgB,EAAM,QACrC,EACF,EAAgB,EAAM,GAAe,EAAK,EAAU,GAElD,EAAY,IAAgB,MAC9B,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAChD,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAC9C,EAAI,IAEJ,EAAM,EAAgB,EAAI,GAAe,EAAG,EAAU,GAE5D,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,EAMT,YACI,EAA8B,EAC9B,GACF,GAAM,GAAW,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,EAAa,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KACpC,EAAiB,EAAW,OAAO,EAAU,GAC7C,EAAc,EAAW,IAAI,SAAC,EAAG,GAAM,MAAC,GAAI,EAAe,GAAK,GAAK,IACrE,EAAS,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAY,KAClD,EAAW,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAS,GAAI,EAAO,MACzD,EAAQ,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAG,EAAY,MACvD,MAAO,CAAC,EAAU,GAMpB,YACI,EAA+B,GAGjC,GAAM,GAAqB,EAAY,IAAI,SAAC,EAAG,GAC7C,MAAO,GAAK,GAAI,GAAM,GAAS,GAAK,KAEhC,EAAgB,EAAmB,IAAI,SAAA,GAAK,MAAA,GAAI,IAIhD,EAAgB,EAAc,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAI,KACtD,EAAc,EAAc,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAc,KAClE,MAAO,GAAc,IAAI,SAAC,EAAG,GAC3B,MAAO,CAAC,EAAc,GAAI,EAAY,SAI7B,IAAO,EAAG,CAAC,MAAK,KChK7B,AAsDA,YACI,EAAyB,SACvB,EAAQ,EAAgB,EAAM,OAAQ,OACtC,EAAO,EAAgB,EAAK,MAAO,OACvC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,GAAM,GAAoB,CAAC,EAAG,EAAO,EAAG,GAClC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,EAAO,GAC7B,SAAK,CAAC,EAAO,EAAM,IACZ,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AA0CA,YAAkC,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAS,EAAgB,EAAO,QAAS,SAEzC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAGH,EAAsB,CAAC,EAAG,EAAI,MAAO,GAC3C,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AA8DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,QAE3B,EAA+B,SAAC,GACpC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAEhB,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,KAAK,EAAW,GACpC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAGzB,MAAO,IAGH,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,EAAE,SAAQ,GAExC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAO,EAAG,CAAC,MAAK,KChG7B,AAiCA,YACI,EAAoB,EACpB,GACF,GAAM,GAAO,GAAc,GACvB,EAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,IAEd,MAAO,GAAO,WAAW,EAAQ,EAAO,MAG7B,IAAO,EAAG,CAAC,MAAK,6OC1B7B,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,cAEJ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,AAyCA,GAAA,IAAA,WAUE,WACI,EAAc,EAAsB,EACpC,EAAqB,GACvB,KAAK,KAAO,EACZ,KAAK,OAAS,EACd,KAAK,MAAQ,EACb,KAAK,QAAU,IACf,KAAK,UAAY,EACjB,AAAI,KAAK,WACP,MAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,EACvC,KAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,GAEzC,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,OAAS,GAAgB,EAAU,YAInC,SAAA,UAAA,UAAP,WACE,GAAI,CAAC,MAAM,KAAK,UACd,GAAM,GAAQ,KAAK,QACnB,YAAK,QAAU,IACR,EAKT,OAFI,GAAiB,EACjB,EAAU,GACP,CAAC,IACN,GAAI,GAAE,OAAU,EAAE,OAAU,EAAC,OAC7B,EACE,GAAK,EAAI,KAAK,SAAW,EACzB,EAAK,EAAI,KAAK,SAAW,EACzB,EAAI,EAAK,EAAK,EAAK,QACZ,GAAK,GAAK,IAAM,GAEzB,GAAM,GAAM,KAAK,KAAK,GAAO,KAAK,IAAI,GAAK,GAC3C,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EACzC,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EAEzC,AAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,GAAU,IAId,MAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,MAAK,QAAU,KAAK,aAAa,IAE5B,KAAK,aAAa,IAInB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,OAAS,MAAQ,KAAK,QAAU,UAChC,EAEF,KAAK,MAAM,IAIZ,EAAA,UAAA,iBAAR,SAAyB,GACvB,MAAO,IAAS,KAAK,OAAS,GAAS,KAAK,OAEhD,KAIA,GAAA,WASE,WACI,EAAe,EAAc,EAC7B,GACF,KAAK,MAAQ,EACb,KAAK,KAAO,EAAI,EAChB,KAAK,MAAQ,EAEb,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,MAAQ,GAAgB,EAAU,YACvC,KAAK,MAAQ,GAAI,IAAY,EAAG,EAAG,EAAO,GAAO,KAAK,SAEtD,AAAI,EAAQ,EACV,KAAK,EAAI,EAAS,EAAI,EAEtB,KAAK,EAAI,EAAS,EAAI,EAExB,KAAK,EAAI,EAAI,KAAK,KAAK,EAAI,KAAK,GAI3B,SAAA,UAAA,UAAP,WAEE,OADI,GAAY,EAAY,EAAY,EAAW,EAAW,KAE5D,EACE,GAAI,KAAK,MAAM,YACf,EAAI,EAAK,KAAK,EAAI,QACX,GAAK,GAMd,GALA,GAAK,EAAI,EACT,EAAK,EAAI,EACT,EAAK,EAAK,KAAQ,EAAK,EACvB,EAAM,GAAM,EAAO,KAAK,EAAK,GAAI,EAAI,KAAK,IAAI,IAC9C,EAAI,KAAK,QACL,EAAI,GAAM,KAAK,IAAI,GAAK,EAC1B,MAGJ,SAAK,EAAI,KAAK,KAAQ,KAAK,EAAI,EAC/B,AAAI,KAAK,MAAQ,GACf,IAAK,KAAK,IAAI,KAAK,QAAS,EAAI,KAAK,QAEhC,KAAK,aAAa,IAGnB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,QAAU,UACV,EAEF,KAAK,MAAM,IAEtB,KAEA,GAAA,WAME,WACI,EAAS,EAAS,EAClB,GAFJ,GAAA,GAAA,KAaE,GAZE,AAAA,IAAA,QAAA,GAAA,GAAS,AAAA,IAAA,QAAA,GAAA,GAoBL,KAAA,eAAiB,WACrB,MAAC,GAAK,OAAS,MAAQ,EAAK,QAAU,WAnBxC,KAAK,IAAM,EACX,KAAK,MAAQ,EAAM,EACnB,KAAK,MAAQ,EACb,AAAI,GAAQ,MACV,GAAO,KAAK,UAEd,AAAI,MAAO,IAAS,UAClB,GAAO,EAAK,YAGV,CAAC,KAAK,kBAAoB,KAAK,OAAS,EAC1C,KAAM,IAAI,OACN,0BAA0B,EAAG,MAAM,EAAG,gCAE5C,KAAK,OAAS,GAAgB,GAOxB,SAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,iBACA,EAEF,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,WACE,MAAO,MAAK,aAAa,KAAK,IAAM,KAAK,MAAQ,KAAK,WAE1D,KCxNA,AAwCA,YACI,EAAoB,EAAe,EACnC,EAAsC,GAOxC,GARqC,AAAA,IAAA,QAAA,GAAA,GACnC,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAQ,MACV,GAAO,GAET,AAAI,GAAS,MACX,GAAQ,WAEN,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,OAAM,yBAAyB,GAI3C,OAFM,GAAS,GAAI,IAAU,EAAO,EAAM,EAAO,GAC3C,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAc,EAAG,CAAC,aAAY,KC5D3C,AAuCA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,yBAAyB,GAK3C,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAuB,GAC1D,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAe,EAAG,CAAC,cAAa,KCtD7C,AA4CA,YACI,EAAoB,EAAY,EAAY,EAC5C,GADoB,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,WAI9C,OAFM,GAAM,GAAO,EAAO,GACpB,EAAS,GAAI,IAAc,EAAQ,EAAQ,KAAM,GAC9C,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAgB,EAAG,CAAC,eAAc,KCvD/C,YAwCyB,EAAsB,GAC7C,GAAc,GACd,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,EAC3B,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAkB,KACxB,MAAO,IAAW,EAAQ,EAAO,EAAe,GC/ClD,YA8CI,EAAe,EAAc,EAC7B,GACF,GAF+B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAA,WACE,IAAS,EACX,KAAM,IAAI,OAAM,8BAGlB,GAAM,GAA+B,WACnC,GAAM,GAAgB,IAAU,EAC1B,EAA8B,EAAQ,GAAQ,EAAO,EACrD,EAA8B,EAAO,GAAS,EAAO,EAE3D,GAAI,GAAiB,GACjB,EACF,MAAO,IAAM,CAAC,GAAI,GAGpB,GAAM,GAAc,KAAK,IAAI,KAAK,KAAM,GAAO,GAAS,IAClD,EAAS,GAAoB,EAAa,GAEhD,AAAI,EAAO,GAAS,IAAS,GAG3B,GAAO,IAGT,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,IAGpB,EAAoB,CAAC,MAAK,EAAE,KAAI,EAAE,KAAI,EAAE,MAAK,GAEnD,MAAO,GAAO,cACH,EAAS,GAAiB,KAAiB,GAC3C,GCnFb,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAE7B,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,WAAW,GAC/B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAa,EAAG,CAAC,YAAW,KChDzC,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,KAAK,IAGhB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,MAAM,IAGjB,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC5D/B,AA6DA,YACI,EAAiB,GACnB,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAA+B,SAAC,GACpC,GAAM,GAAO,GAAe,EAAM,EAAG,OACrC,GAAI,EAAG,OAAS,EACd,MAAO,IAAM,GAEf,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,MAAO,GAAQ,EAAK,EAAG,QAGnB,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,KAAM,GAEnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAS,MAGT,IAAU,EAAG,CAAC,SAAQ,KClFnC,AA6BA,YAAoB,GAClB,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCrCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAGV,IAAQ,EAAG,CAAC,OAAM,KChD/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GAGH,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KCe7B,YACI,EAAiB,EACjB,EAAsC,EACtC,EAAqB,EACrB,GADqB,AAAA,IAAA,QAAA,GAAA,CAAqC,EAAG,IAC7D,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EACF,EAAgB,EAAiB,kBAAmB,mBAClD,EACF,EAAgB,EAAiB,kBAAmB,mBAEpD,EAAM,EACN,EAAe,GAMnB,GALA,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGvD,IAAe,OACjB,KAAM,IAAI,OACN,sFAIN,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,MAAM,KAAO,EAC9B,WACI,MAAA,qEACA,wBAAuB,EAAiB,MAAM,GAAE,OACxD,EACI,EAAiB,MAAM,KAAO,EAC9B,WAAM,MAAA,+DACF,8BAA6B,EAAiB,MAAM,GAAE,OAE9D,GAAM,GAAa,EAAiB,MAAM,GACpC,EAAoB,EAAiB,MAAM,GACjD,EACI,EAAiB,MAAM,KAAO,EAAa,EAC3C,WACI,MAAA,qEACA,YAAW,EAAa,EAAiB,MACzC,YAAW,EAAiB,MAAM,GAAE,OAE5C,GAAM,GAAY,GACd,EAAK,EAAkB,EAAS,EAAK,EAAY,GAC/C,EAAkB,EAClB,EACF,GAAO,EAAW,EAAkB,EAAiB,QAAS,GAElE,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KCxFnD,YACI,EAAsB,uHAClB,SAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aAEnC,EACI,EAAG,QAAU,EAAG,MAChB,WAAM,MAAA,kDACF,EAAG,MAAK,YAAY,EAAG,MAAK,OAEpC,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAEtE,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAExD,CAAA,EAAM,EAAG,eAAjB,SAAQ,EAAA,OACA,CAAA,EAAM,EAAG,eAIvB,IAJM,EAAQ,EAAA,OACR,EAAO,GAAI,KAAI,GAEjB,EAAa,EACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,AAAK,EAAK,IAAI,EAAM,KAClB,IAMJ,IAFM,EAAS,GAAI,IAAa,CAAC,GAAa,EAAG,OAC3C,EAAU,GAAI,IAAa,CAAC,GAAa,SACtC,EAAI,EAAG,EAAI,EAAG,EAAI,EAAM,OAAQ,IACvC,AAAK,EAAK,IAAI,EAAM,KAClB,GAAO,OAAO,GAAK,EAAM,GACzB,EAAQ,OAAO,GAAK,EACpB,KAGJ,MAAA,CAAA,EAAO,CAAC,EAAO,WAAY,EAAQ,qBAExB,IAAiB,GCxF9B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC7C7B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AA6BA,YACI,EAAwB,EAAe,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,CAAC,GAAQ,CAAC,OAEhB,IAAU,EAAG,CAAC,SAAQ,KCtCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AAgDA,YAAoC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxD,GAAM,GAAU,EAAgB,EAAQ,SAAU,UAAW,WAK7D,GAHA,AAAI,IAAQ,IACV,GAAM,EAAQ,KAAO,GAEnB,IAAQ,EAAQ,KAAO,EACzB,KAAM,OACF,4DACA,oBAAmB,EAAQ,KAAI,gBAAgB,IAGrD,GAAM,GAAwB,CAAC,OAAQ,GACjC,EAAsB,CAAC,IAAG,GAEhC,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAI,EAAQ,QAAQ,EAAS,GACnC,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAU,EAAG,CAAC,SAAQ,KCzEnC,AAyCA,YAAc,GACZ,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,qDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAoB,CAAC,MAAK,GAEhC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAU,EAAM,KAAK,EAAO,GAC5B,EAAS,EAAQ,IAAI,GAC3B,MAAO,GAAO,QAAQ,EAAM,QAC3B,EAAgC,KAAqB,OAG7C,IAAM,EAAG,CAAC,KAAI,KC5D3B,AA0CA,YAAe,GACb,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,sDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAqB,CAAC,MAAK,GAEjC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAoB,EAAQ,EAAO,CAAC,EAAO,IAC3C,EAAS,EAAQ,KAAK,GAC5B,MAAO,GAAQ,EAAQ,EAAM,QAC5B,EAAgC,KAAqB,OAG7C,IAAO,EAAG,CAAC,MAAK,KC7D7B,AAgDA,YAAgB,GACd,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EACvB,EACJ,GAAI,GAAsB,GACxB,GAAM,GAAe,EAAQ,EAAO,CAAC,EAAO,IAC5C,EAAM,GAAK,QAIX,GAAM,GAAc,CAAC,EAAO,EAAK,GAAqB,IAChD,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IACzC,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IAEzC,EACF,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACjE,EAA0B,EAC5B,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACnE,GAAO,KAEL,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EACF,EAAQ,GAAQ,EAAG,GAAI,CAAC,EAAY,GAAI,EAAY,KACxD,EAAM,GAAK,GAIb,GAFA,EAAM,GAAK,GAEP,EAAM,OAAS,GAAK,EAAM,MAAM,KAAO,GACzC,GAAM,GAAO,EACP,EAAQ,EAAM,MAAM,GAC1B,EAAM,EAAQ,EAAK,CAAC,EAAO,EAAI,MAAM,GAAK,EAAO,EAAI,MAAM,KAC3D,EAAK,UAEP,MAAO,MAGI,IAAQ,EAAG,CAAC,OAAM,iBC3D3B,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAa,GACjB,GAAI,MAAQ,IAAqB,SAC/B,EACI,EAAE,MAAM,GAAQ,IAAoB,EACpC,WAAM,MAAA,kDACV,EACI,GAAI,OAAM,GAAiB,KAAK,EAAE,MAAM,GAAQ,QAEpD,GAAM,GAAY,EAAgB,OAAO,SAAC,EAAO,GAC/C,MAAI,KAAU,IACZ,IAAS,GAEJ,GACN,GACH,EACI,GAAa,EACb,WAAM,MAAA,4DACV,GAAM,GAAW,EAAgB,QAAQ,IAGzC,GAAI,IAAa,IACf,GAAM,GAAQ,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,EAAI,IAC/D,EAAgB,GAAY,EAAE,MAAM,GAAQ,EAE9C,EACI,EAAE,MAAM,KAAU,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IACvD,WAAM,MAAA,gEACV,EAAa,EAGf,MAAO,GC1DT,AA+DA,YACI,EAAsB,EAAkC,GAAA,AAAA,IAAA,QAAA,GAAA,GAC1D,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAG,OAAO,GACvC,EAAa,GAAiB,EAAI,EAAiB,GACzD,MAAO,GAAQ,MAAM,EAAI,EAAY,IAGjC,EAAuB,CAAC,EAAG,GAC3B,EAAoB,CAAC,gBAAe,EAAE,KAAI,GAEhD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCjF/B,AA+CA,YAAe,EAAe,GAC5B,EACI,EAAM,QAAU,UAChB,WAAM,MAAA,mDAAmD,EAAM,QAEnE,GAAI,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACpD,EAAQ,EAAM,KAAO,EAEvB,EACJ,GAAI,GAAa,MAAQ,EAAY,GAEnC,GAAM,GAAQ,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAC7B,EAAO,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAClC,EAAK,EAAM,MAAM,OAAS,GAAK,EAC/B,EAAgB,GAAM,EAAO,EAAO,GACpC,EAAqB,UACZ,GAAa,MAAQ,EAAY,GAE1C,GAAM,GAAa,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KACxC,EAAW,EAAM,MAAM,OAAS,GAAK,EAAY,EACjD,EAAgB,GAAO,CAAC,EAAO,GAAM,IAAc,EAAM,MAAM,OAAS,GACxE,EAAqB,MAErB,GAAgB,EAIlB,GAAM,GAAa,GAAU,GACvB,EACF,EAAQ,GAAQ,EAAe,GAAa,CAAC,EAAO,IAElD,EAAM,GAAI,GAGV,EAAO,KAAK,MAAM,EAAqB,GAAK,EAC5C,EAAa,GAAK,GAClB,EAAa,GAAK,GAClB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GACxB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GAExB,EAAc,EAAc,MAAM,QACxC,SAAY,EAAc,MAAM,OAAS,GAAK,EAEvC,EACH,GAAQ,EAAqB,GAAI,EAAqB,IAAK,MAGpD,IAAO,EAAG,CAAC,MAAK,KClG7B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AAoDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,qBAC7B,EAAK,EAAgB,EAAG,IAAK,qBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,kBAAkB,EAAI,GAC1C,SAAK,CAAC,EAAI,IACH,GAGH,EAAkC,CAAC,EAAG,EAAI,EAAG,GAC7C,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAmB,MAGnB,IAAoB,EAAG,CAAC,mBAAkB,KC1EvD,AAwCA,YAAoC,EAAsB,GACxD,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,MAAO,GAAQ,EAAI,GAAa,EAAG,MAAO,GAAM,aAGrC,IAAU,EAAG,CAAC,SAAQ,KC7CnC,AAyCA,YACI,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAChC,GAAM,GAAW,GAAqB,EAAS,UAAW,SAK1D,GAHA,EACI,EAAS,QAAU,EAAG,WAAM,MAAA,yCAE5B,EAAS,SAAW,EACtB,MAAO,IAAW,EAAS,GAAI,GAGjC,GAAM,GAAO,EAAS,GAAG,KACnB,EAAQ,EAAS,GAAG,MACpB,EAAQ,EAAS,GAAG,MAE1B,EAAY,GAAQ,EAAM,WAAM,MAAA,uCAEhC,EAAS,QAAQ,SAAA,GACf,GACI,EAAO,EAAE,MACT,yDACJ,EACI,IAAU,EAAE,MACZ,WAAM,MAAA,4DAGZ,GAAM,GAAkB,EAAS,IAAI,SAAA,GAAK,MAAA,IAAW,EAAG,KAOxD,MAAO,IAAO,EAAiB,MAGpB,IAAQ,EAAG,CAAC,OAAM,KC7E/B,AAwCA,YAAiC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAEhB,IAAO,EAAG,CAAC,MAAK,KClD7B,AA+DA,YACI,EAAsB,EAAiB,EAAe,EACtD,EAAe,EAAa,EAAkB,EAC9C,GADA,AAAA,IAAA,QAAA,GAAA,GAAe,AAAA,IAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAAkB,AAAA,IAAA,QAAA,GAAA,GAC9C,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAE3B,EAA+B,SAAC,GACpC,AAAI,GAAW,MACb,GAAU,GAAI,OAAM,EAAM,SAG5B,GAAM,GAAe,GAAW,GAChC,GAAI,EAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAI,IAAiB,GAAK,IAAgB,EACxC,KAAM,IAAI,OACN,iEAGN,GAAI,IAAiB,GAAK,IAAmB,EAC3C,KAAM,IAAI,OACN,oEAGN,GAAM,GAAsB,EAAG,KAAO,EAAM,OAGtC,EAAa,GAAW,GACxB,EAAW,EAAG,MAAM,QAC1B,EAAW,QAAQ,SAAA,GACjB,EAAM,GAAQ,EACd,EAAI,GAAQ,EACZ,EAAS,OAAO,EAAM,EAAG,KAE3B,EAAK,EAAQ,EAAI,GAEX,GAAA,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GACJ,EAAA,EAAA,MACA,EAAA,EAAA,IACA,EAAA,EAAA,QAKF,EAAQ,EACR,EAAM,EACN,EAAU,EAEV,GAAM,GAAa,GAAW,GAE9B,EAAW,QAAQ,SAAA,GACjB,EAAI,GAAQ,EAAM,GAAQ,EAC1B,EAAQ,GAAQ,IAIlB,GAAM,GAAO,GAAgB,EAAO,EAAK,GAEnC,EAAW,EAAK,OAAO,SAAC,EAAG,GAAS,MAAA,GAAW,QAAQ,KAAU,KAEjE,EAAa,EAAQ,MAAM,SAAA,GAAK,MAAA,KAAM,IAC5C,GAAI,EACF,MAAO,GAAQ,GAAM,EAAI,EAAO,GAAO,GAGzC,GAAM,GAAM,EAAQ,aAAa,EAAI,EAAO,EAAK,GACjD,MAAO,GAAQ,EAAK,IAGhB,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAC/B,MAAK,EACL,IAAG,EACH,QAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,YAAW,EACX,eAAc,GAGhB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAe,EAAG,CAAC,cAAa,KCvJ7C,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,kEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,gFAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,sEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,wEAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GCjElD,YAgDI,EACA,EACA,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,0EAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,SAAQ,GACJ,EACG,GAAW,EAAQ,EAAO,EAAe,GCpElD,AAmDA,YACI,EAAiB,EAAO,GAAP,AAAA,IAAA,QAAA,GAAA,GAAO,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAK,EAAgB,EAAG,IAAK,QACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAU,EAAG,MAAM,EAAG,MAAM,OAAS,GAC3C,GAAI,EAAI,EACN,KAAM,IAAI,OACN,uDAAuD,EAAO,KAC9D,YAAW,IAGjB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,EAAC,EAAE,OAAM,GAE7B,EAAA,EAAA,cAAA,SAAA,GAAA,MAAA,GAAA,KAAA,EAAA,EAAA,IAAA,EAAA,KAAA,GAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAKf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAO,EAAG,CAAC,MAAK,KC3E7B,AA4CA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,qCAKlB,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAsB,GACzD,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAkB,EAAG,CAAC,iBAAgB,KC3DnD,AA6EA,YACI,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAEnB,GAAM,GAAK,EAAgB,EAAG,IAAK,SAAU,MAC7C,EAAO,EAAG,KAAO,EAAG,WAAM,MAAA,yCAE1B,GAAM,GAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAC1B,EAAA,EAAA,UAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAS,EAAG,CAAC,QAAO,KC3FjC,AA8CA,YACI,EAAiB,EAAiC,GACpD,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EACF,EAAgB,EAAY,aAAc,qBAAsB,SACpE,EAAO,GAAM,GAAc,WAAM,MAAA,qCAEjC,GAAM,GAAmC,CAAC,EAAG,EAAI,WAAY,GACvD,EAAiC,CAAC,YAAW,GAE7C,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,mBAAmB,EAAI,EAAa,GACxD,SAAK,CAAC,IACC,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAoB,MAGpB,IAAqB,EAAG,CAAC,oBAAmB,KCnEzD,AA2CA,YAAkB,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACtC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,EACI,GAAQ,CAAC,EAAG,MAAM,QAAU,EAAO,EAAG,MAAM,OAC5C,WACI,MAAA,UAAU,EAAI,gBAAgB,EAAG,MAAM,OAAM,KAAK,EAAG,MAAM,OAAM,MACzE,AAAI,EAAO,GACT,IAAQ,EAAG,MAAM,QAEnB,GAAM,GAAuB,CAAC,MAAO,GAC/B,EAAqB,CAAC,KAAI,GAC1B,EAAiC,SAAC,GACpC,MAAA,GAAQ,QAAQ,EAAI,IACxB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAU,EAAG,CAAC,SAAQ,KC7DnC,YAsCI,EAAyB,EAAkB,EAC3C,GADyB,MAAA,KAAA,QAAA,GAAA,IAEpB,EAAO,aAAa,EAAc,EAAW,EAAM,GCxC5D,YAuB0B,EAAqB,GAE7C,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,AAAI,EAAS,IACX,EAAQ,KAAK,GAOjB,OAHM,GAAW,GAAO,EAAW,SAE7B,EAAM,GAAO,CAAC,EAAQ,OAAQ,EAAU,QAAS,SAC9C,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAM,EAAS,WAAW,EAAQ,IAClC,EAAS,EAAI,EAAU,OAC7B,EAAI,OAAO,IAAI,EAAK,GAEtB,MAAO,GAAI,WCEb,YAA2B,uGACnB,SACF,EAAgB,EAAW,YAAa,aAAc,QAC7C,CAAA,EAAM,EAAW,eAAxB,SAAO,EAAA,OACP,EAAM,GAAU,EAAW,MAAO,GACxC,AAAI,IAAc,GAChB,EAAW,UAEb,CAAA,EAAO,WAGI,IAAa,GCpD1B,AA6CA,YACI,EAA2B,EAC3B,2HAcF,IAbM,EAAU,EAAgB,EAAQ,SAAU,YAC5C,EAAQ,EAAgB,EAAM,OAAQ,WAAY,QAElD,EAAW,GAAe,EAC1B,EAAU,EAAM,KAChB,EAAc,EAAQ,MAE5B,EAAY,EAAU,EAAG,WAAM,MAAA,0BAC/B,GACI,EAAY,MAAM,EAAU,EAAW,GAAU,EAAM,MACvD,qEAEA,EAAc,EACT,EAAI,EAAU,EAAI,EAAW,EAAS,IAC7C,GAAe,EAAY,GAEvB,SACF,EAAY,MAAM,EAAG,GAChB,OAAO,CAAC,GAAc,EAAY,MAAM,EAAW,IACtD,EAAiB,EAAQ,EAAS,GAClC,EAAe,EAAQ,EAAO,CAAC,KACX,CAAA,EAAM,GAAW,WAArC,SAAoB,EAAA,OACpB,EAAU,GAAQ,EAAmB,CAAC,IAEtC,EAAM,GAAO,EAAgB,EAAS,GAG5C,AAAI,IAAW,GACb,EAAQ,UAEV,AAAI,IAAS,GACX,EAAM,UAER,EAAQ,UACR,EAAe,UACf,EAAa,UACb,EAAkB,UAElB,CAAA,EAAO,WAGI,IAAmB,GCzFhC,AAuCA,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAK,EAAgB,EAAG,IAAK,kBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,6BAC/B,GAAS,EAAI,GAYtB,YAAuC,EAAiB,GACtD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAK,EAAgB,EAAG,IAAK,cACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,yBAC/B,GAAK,EAAI,GAGlB,YAAwC,EAAiB,GACvD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,eAC7B,EAAK,EAAgB,EAAG,IAAK,eACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,0BAC/B,GAAM,EAAI,GAGnB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAK,EAAgB,EAAG,IAAK,mBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,8BAC/B,GAAU,EAAI,GAGvB,YAA0C,EAAiB,GACzD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,4BAC/B,GAAQ,EAAI,GAGrB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EAAK,EAAgB,EAAG,IAAK,sBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,iCAC/B,GAAa,EAAI,MAGb,IAAc,EAAG,CAAC,aAAY,KAC9B,GAAqB,EAAG,CAAC,oBAAmB,KAC5C,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAkB,EAAG,CAAC,iBAAgB,KACtC,GAAa,EAAG,CAAC,YAAW,KAC5B,GAAiB,EAAG,CAAC,gBAAe,KCpHjD,AA2CA,YAAsC,EAAiB,GACrD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAS,GAC7C,UACI,6EAGJ,GAAuB,EAAK,MAAO,EAAI,MAAO,wBACvC,GAAI,EAAM,GAanB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,6BACpC,EAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,2BACpC,GAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAWjB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAWrB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAarB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,2BAC7B,EAAK,EAAgB,EAAG,IAAK,2BACnC,UACI,EAAG,MAAO,EAAG,MAAO,sCACjB,GAAkB,EAAI,MAGlB,IAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAA0B,EAAG,CAAC,yBAAwB,KACtD,GAAY,EAAG,CAAC,WAAU,KC5NvC,AAwEA,YACI,EAAsB,EACtB,EAA8B,GADR,AAAA,IAAA,QAAA,GAAA,aACtB,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IAChC,EAAI,EAAgB,EAAG,IAAK,QAE5B,GAAM,GAAO,GAAS,EAAG,EAAK,GAC1B,EAAgB,EAAK,MACzB,GAAI,GACF,GAAM,GAAO,GAAe,EAAM,EAAE,OACpC,EAAgB,GAA+B,EAAK,MAAO,GAE7D,MAAO,GAAQ,EAAM,GAGvB,YACI,EAAW,EAAkB,GAC/B,GAD+B,AAAA,IAAA,QAAA,GAAA,MAC3B,EAAE,OAAS,EACb,MAAO,IAAI,GAIb,GAAI,EAAE,OAAS,GAAK,IAAS,KAC3B,MAAO,IAAS,EAAQ,EAAG,CAAC,KAAM,EAAG,GAIvC,GAAI,EAAE,OAAS,GAAK,MAAO,IAAS,UAChC,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,aAAe,IAAM,EAE7B,MAAO,IAAK,GAAI,GAAI,GAAI,GAAI,GAAO,EAAG,UAAW,IAGnD,KAAM,IAAI,OAAM,qCAAqC,GAIvD,GAAI,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,GAAK,GAE7C,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,OAAS,IAAM,YAEvB,MAAO,IAAK,GAAI,GAAO,GAAI,IAG7B,KAAM,IAAI,OAAM,qCAAqC,GAGvD,KAAM,IAAI,OAAM,gCAAgC,MAGrC,IAAO,EAAG,CAAC,MAAK,KC3I7B,AA2DA,YACI,EAAiB,EAAiB,EAClC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAS,EAAgB,EAAO,QAAS,iBAE/C,GAAiB,EAAI,GACrB,EACI,GAAiB,EAAG,MAAO,EAAG,OAAQ,WAAM,MAAA,8BAEhD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAI,EAAK,GAE3B,EAAS,EAAI,GAAI,EAAI,GAAK,GAC9B,GAAI,GACF,EACI,GAAQ,KAAM,WAAM,MAAA,mDACxB,GAAM,GAAQ,EAAgB,EAAM,OAAQ,iBAC5C,EAAS,GAAI,EAAQ,GAAI,EAAK,GAAI,EAAQ,KAE5C,MAAO,IAAI,EAAI,MAGJ,IAAgB,EAAG,CAAC,eAAc,KCnF/C,AA+CA,YACI,EAA4B,EAC5B,GACF,GAAM,GAAW,EAAgB,EAAS,UAAW,YAAa,SAC5D,EAAW,EAAgB,EAAS,UAAW,aACrD,GAA8B,EAAU,EAAU,GAElD,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,UAAU,EAAU,EAAU,IAGzC,EAA0B,CAAC,QAAS,EAAU,QAAS,GACvD,EAAwB,CAAC,MAAK,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAW,MAGX,IAAY,EAAG,CAAC,WAAU,iBClCnC,EAAuB,EAAsB,EAC7C,GACF,GAAI,EAAc,QAAU,QAC1B,KAAM,IAAI,OACN,2DACA,uBAAsB,EAAc,MAAK,MAE/C,GAAI,EAAc,KAAO,EACvB,KAAM,IAAI,OACN,uDACA,mBAAkB,EAAc,MAAK,MAG3C,GAAM,GAAW,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAC7D,EAAU,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAElE,GAAI,EAAY,SAAW,EACzB,KAAM,IAAI,OACN,iDACA,KAAI,EAAY,OAAM,gBAAgB,EAAO,MAGnD,GAAM,GAAY,EAAa,KAC/B,GAAI,CAAE,GAAa,OAAS,GACtB,EAAa,OAAS,GAAK,IAAc,GAC7C,KAAM,IAAI,OACN,oCACG,GAAa,MAAK,sBAAsB,EAAQ,MAGzD,GAAI,EAAa,QAAU,EAAc,MACvC,KAAM,IAAI,OAAM,qDC/DpB,AAkEA,YACI,EAAkC,EAClC,EAA0B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC5B,GAAM,GACF,EAAgB,EAAe,gBAAiB,gBAAiB,SAC/D,EACF,EAAgB,EAAc,eAAgB,iBAC5C,EAAgB,EAClB,EAAc,eAAgB,gBAAiB,EAAc,OAEjE,GACI,EAAgB,EAAe,EAAa,GAEhD,GAAM,GAA8B,CAClC,cAAe,EACf,aAAc,EACd,aAAc,GAGV,EAA4B,CAAC,YAAW,GAE9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,cACf,EAAgB,EAAe,EAAa,IAChD,EAAgC,KAAiB,GACjD,MAGO,IAAgB,EAAG,CAAC,eAAc,KC9F/C,AA8DA,YAAmB,EAAsB,GACvC,GAAM,GAAW,EAAgB,EAAS,UAAW,WAAY,SAC3D,EAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,GACpC,MAAO,GAAQ,SAAS,EAAI,IAGxB,EAAyB,CAAC,OAAQ,EAAI,QAAS,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KAAqB,OAGvD,IAAW,EAAG,CAAC,UAAS,KC5ErC,YA4B8B,EAAW,GACvC,GAAI,GAAc,KAChB,MAAO,GAAE,MAAM,QAEjB,GAAI,GAAiB,EAAE,MAAO,GAC5B,MAAO,GAET,GAAI,EAAE,MAAM,SAAW,EAAW,QAEhC,OADM,GAAyB,GACtB,EAAI,EAAG,EAAI,EAAE,MAAM,OAAQ,IAClC,AAAI,EAAW,IAAM,MAAQ,EAAE,MAAM,IAAM,KACzC,EAAa,KAAK,EAAE,MAAM,IAE1B,EAAa,KAAK,EAAW,IAGjC,MAAO,GAGT,MAAO,GC/CT,AAoDA,YACI,EAAsB,EAAc,EACpC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAUnC,GARA,EACI,EAAG,QAAU,UACb,WAAM,MAAA,8DACF,sBAAqB,EAAG,MAAK,sBACrC,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,qDAAqD,EAAI,MAE/D,IAAS,EACX,MAAO,aAAa,GAAS,EAAG,QAAU,EAG5C,GAAM,GAAc,GAAc,EAAI,GAChC,EAAW,EAAI,EACf,EAAa,GACf,GAAM,GAAI,GAAc,EAAa,EAAG,EAAG,UAAW,GAAO,IAC7D,GAEJ,MAAO,GAAI,EAAI,MAGJ,IAAU,EAAG,CAAC,SAAQ,KC9EnC,YAoBoC,GAElC,MAAO,MAAK,MAAM,KAAK,IAAI,EAAG,KAAK,KAAK,KAAK,IAAI,GAAS,KAAK,IAAI,kBAIjE,EAAsB,EAAW,GAGnC,OAFM,GAAO,EAAI,EAAe,EAC1B,EAAY,GAAI,cAAa,GAC1B,EAAI,EAAG,EAAI,EAAc,EAAE,GAClC,GAAM,GAAU,EAAM,KAAK,GAAK,EAAM,GAAe,EAAO,GAC5D,EAAU,GAAK,EAAI,EAAI,KAAK,IAAI,GAElC,MAAO,IAAS,EAAW,WCjC7B,AAwCA,YACI,EAA2B,EAAuB,GAAA,MAAA,KAAA,QAAA,GAAA,wHAC9C,SAAe,EAAgB,EAAa,cAAe,UAC3D,EAAW,EAAgB,EAAS,UAAW,UAErD,EACI,EAAa,KAAO,EACpB,WAAM,MAAA,+DACF,YAAW,EAAa,QAChC,EACI,EAAa,KAAO,IAAM,EAAS,KACnC,WAAM,MAAA,mFAEC,GAAa,KAAI,qBAAqB,EAAS,QAC1D,GACI,EAAa,MAAM,MAAM,EAAG,EAAa,MAAM,OAAS,GACxD,EAAS,MACT,2FAEE,EAAU,EAAa,MAAM,EAAa,MAAM,OAAS,GAC/D,EACI,EAAI,GAAK,GAAK,EACd,WAAM,MAAA,iEACF,eAAc,EAAO,cAAc,KAEnB,CAAA,EAAM,EAAa,eAArC,SAAkB,EAAA,OACJ,CAAA,EAAM,EAAS,eAOnC,IAPM,EAAc,EAAA,OAId,EAAgB,CAAC,EAAgB,OAAS,EAAS,GAAlD,EAAK,EAAA,GAAE,EAAI,EAAA,GACZ,EAAY,GAAuB,OAAQ,GAExC,EAAI,EAAG,EAAI,EAAO,KAIzB,IAHM,EAAS,EAAI,EACb,EAAO,EAAgB,SAAS,EAAQ,EAAS,GACjD,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAKzC,IAHA,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAErC,EAAU,GAAK,EACN,EAAI,EAAG,EAAI,EAAG,IACrB,GAAI,EAAU,GAAG,QAAU,EAAY,IACrC,EAAU,GAAK,EACf,OAKN,MAAI,KAAgB,GAClB,EAAa,UAEf,AAAI,IAAY,GACd,EAAS,UAIX,CAAA,EAAO,GAAO,EAAW,EAAS,MAAO,iBAG9B,IAAc,GCtG3B,AAiDA,YACI,EAAM,EAAO,EACb,EACA,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE5D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GAC3D,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAY,GACxB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAY,GAAE,OACtD,EACI,IAAa,EAAY,GACzB,WAAM,MAAA,0CAA0C,EAAQ,UACpD,mCAAkC,EAAY,GAAE,QACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EACZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,GAAI,GAClD,EAC0B,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,GAE1E,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAsB,GAG5B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC/G7D,YAgCI,EAAY,EAAW,GACzB,GAAI,GAAc,MAAQ,IAAe,SACvC,MAAO,GAET,GAAI,IAAe,OACjB,MAAO,GAAI,EAAI,GAAK,IAEtB,KAAM,IAAI,OACN,gDAAgD,EAAU,iBAK5D,EAAc,GAChB,GAAI,GAAM,EACJ,EACF,GAAgC,EAAK,MAAO,EAAa,OAC7D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,mBAIvB,EAAW,EACX,GACF,GAAI,IAAe,SACjB,MAAO,GACF,GAAI,IAAe,OACxB,MAAO,IAAK,GACP,GAAI,IAAe,MACxB,MAAO,IAAI,GACN,GAAI,IAAe,QACxB,MAAO,IAAM,GACR,GAAI,IAAe,QACxB,MAAO,IAAM,EAAG,GAElB,KAAM,IAAI,OAAM,4BAA4B,EAAU,KAIjD,GAAM,IAAa,SAAC,EAAuB,GAChD,GAAM,GAAe,EAAgB,EACrC,MAAO,CAAC,GAAgB,IAAe,UC3EzC,AA8FA,YAAmD,MACjD,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAeA,GAFA,EAAa,GAAc,SAEvB,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,6DACC,GAAI,KAAI,OACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,8DACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,6DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,OACf,WAAM,MAAA,sCACF,EAAU,2CAElB,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,GAEnD,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACpB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAGlB,EAAe,GAAqB,EAAI,EAAG,GAEjD,EACI,GAA4B,GAC5B,WAAM,MAAA,oEAEF,uDAAsD,EAAS,OAEvE,GAAM,GACF,GAAoB,EAAI,MAAO,EAAc,EAAS,EAAS,GAC7D,EACF,GAAqB,EAAK,EAAc,EAAQ,MAAO,EAAS,GAC9D,GAAgB,CAAC,EAAM,GAE7B,GAAI,GAAS,MACX,GAAM,IAAU,GAAqB,EAAO,GAC5C,GAAI,KAAK,IAEX,MAAO,KAGH,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,YAAY,CAC9B,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA4B,CAChC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAGpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAS,EAAG,CAAC,aAAY,KC5QtC,AAyBA,YACI,EAAM,EAAO,EACb,GACF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,yBAAyB,EAAK,EAAM,IAE1C,EAAoD,CAAC,EAAG,EAAK,GAAI,GAEvE,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,IAGN,GAAM,IACT,EAAG,CAAC,qCAAoC,KChD5C,AAyBA,YACI,EAAmE,EACnE,EAAkB,GACpB,GAAI,GAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,wBAAwB,EAAM,EAAQ,IAE5C,EAAmD,CAAC,GAAI,GAExD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,IAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IACT,EAAG,CAAC,oCAAmC,KCnD3C,AAyFA,YAA4D,MAC1D,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAaA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,0DACF,iBAAgB,EAAQ,KAAI,OACpC,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,4DACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAClC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,EACI,GAAyC,EAAS,GAClD,WACI,MAAA,oEACA,sBAAqB,EAAO,mBAAmB,EAAS,OAEhE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,0BAAyB,EAAe,gBAAgB,EAAG,OAGrE,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IAEA,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,0BAG/C,GAAM,GAAO,SAAC,EAAc,GAC1B,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAS,OACd,GAAA,GAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAElB,EAAe,GAAqB,EAAI,EAAG,GAE3C,EAAO,GACR,EAAiB,MAAO,EAAc,EAAqB,GAC1D,EAAY,GACd,EAAiB,EAAe,EAAqB,MAAO,GAEhE,GAAI,GAAQ,MACV,GAAM,GAAU,GAAqB,EAAO,GAC5C,MAAO,CAAC,EAAM,EAAW,GAE3B,MAAO,CAAC,EAAM,IAGV,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,qBAAqB,CACvC,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAAqC,CACzC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAEpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAkB,EAAG,CAAC,sBAAqB,KClQxD,AAwDA,YAAwC,SACtC,EAAA,EAAA,EACA,EAAA,EAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAUA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GAAc,EAAG,EAAG,EAAY,GAC7C,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WACI,MAAA,qEACA,iBAAgB,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE9C,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAU,UACrD,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,wCAAwC,EAAW,UAClD,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACnC,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAErC,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAU,EAAM,QAG5D,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACnB,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAAK,GAAA,EAAA,GAAG,GAAA,EAAA,GAId,GACF,GAAqB,EAAQ,EAAI,GAAE,OAAQ,GAAG,GAC9C,GACA,GAgBJ,GAdA,AAAI,CAAC,GAAc,CAAC,EAClB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAM,KACzC,AAAI,CAAC,GAAc,EACxB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,GAAc,EAAK,GAAM,KACzC,AAAI,GAAc,CAAC,EACxB,IAAO,GAAc,EAAK,GAAc,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAO,KAE/C,IAAO,GAAc,EAAK,GAAc,GAAM,IAC9C,GAAO,GAAc,GAAc,EAAK,GAAM,KAG5C,GAAQ,MACV,GAAM,IAAU,GAAqB,GAAO,IAC5C,MAAO,CAAC,GAAM,GAAM,QAEpB,OAAO,CAAC,GAAM,KAIZ,EAA+B,SAAC,GACpC,GAAM,GAAI,EAAQ,iBAAiB,CACjC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA6B,CACjC,EAAG,EACH,EAAG,EACH,KAAM,EACN,uBAAwB,GAEpB,EAA2B,CAAC,WAAU,EAAE,WAAU,EAAE,WAAU,GAIpE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAe,GACxC,GAAM,GAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,IAET,CAAC,MAAO,EAAQ,EAAK,GAAW,SAAU,KAEvD,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAe,EAAe,GAC5C,GAAM,IAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,GAAK,IAEd,CAAC,MAAO,EAAQ,GAAK,GAAW,SAAU,KAGvD,MAAO,GAAiB,EAAK,EAAK,IAI/B,GAAM,IAAS,EAAG,CAAC,aAAY,KClOtC,+DCAA,AAiCA,YAAwB,GACtB,MAAO,IAAa,EAAc,IAAM,KAEnC,GAAM,IAAgB,EAAG,CAAC,eAAc,KCpC/C,AAiCA,YAAqB,GACnB,MAAO,IAAa,EAAc,GAAK,IAGlC,GAAM,IAAa,EAAG,CAAC,YAAW,KCrCzC,AAyCA,YACI,EAAkB,EAAqB,EAAmB,EAC1D,GAD0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,GAGF,OAFI,GAAQ,EACN,EAAmB,GAClB,EAAQ,GAAe,EAAO,MACnC,EAAO,KAAK,GAAM,EAAQ,EAAO,IACjC,GAAS,EAGX,GAAI,EACF,KAAO,EAAQ,EAAO,OACpB,GAAM,GAAU,EAAQ,EAAe,EAAO,KACxC,EAAM,GAAO,CACjB,GAAM,EAAQ,EAAO,EAAc,GAAS,GAAK,CAAC,GAAS,KAE7D,EAAO,KAAK,GACZ,GAAS,EAIb,MAAI,GAAO,SAAW,EACb,GAAS,GAAI,CAAC,EAAG,IAGnB,EAAQ,GAAO,GAAS,CAAC,EAAO,OAAQ,IAE1C,GAAM,IAAQ,EAAG,CAAC,OAAM,KCpE/B,AA4CA,YACI,EAAkB,EAAqB,EACvC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,AAAI,GAAa,MACf,GAAY,GAAoB,IAKlC,OAHM,GAAe,GAAM,EAAQ,EAAa,GAC1C,EAAiB,EAAI,EAAc,EAAS,IAC5C,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAa,MAAM,GAAI,IACzC,EAAO,KACH,GAAK,GAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,EAAG,IAAe,IAE5D,MAAO,IAAO,GAET,GAAM,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAmDA,YACI,EACA,EACA,EACA,EACA,EACA,GAEF,GAAM,GAAS,EAAgB,EAAO,QAAS,iBACzC,EAAS,EAAgB,EAAO,QAAS,gBAAiB,WAC1D,EAAU,EAAgB,EAAQ,SAAU,gBAAiB,SACnE,EAAS,GAAU,WACnB,EAAqB,GAAsB,EAE3C,GAAM,GAAW,EAAO,MAAM,GAE9B,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OACnC,EACI,EAAO,OAAS,GAAK,EAAO,MAAM,KAAO,EACzC,WAAM,MAAA,oDAAoD,EAAQ,OAC9D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAQ,OAAS,GAAK,EAAQ,MAAM,KAAO,EAC3C,WAAM,MAAA,qDAAqD,EAAQ,KAC/D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,iEACF,WAAU,EAAS,OAAM,OACjC,EACI,EAAS,IAAM,GAAK,EAAS,IAAM,EACnC,WAAM,MAAA,2CAA2C,IACrD,EACI,IAAW,YAAc,IAAW,UACpC,WAAM,MAAA,+CAA+C,IAEzD,GAAM,GAAiC,SAAC,GAAY,MAAA,GAAQ,cACxD,EAAQ,EAAQ,EAAS,EAAU,EAAQ,IAEzC,EACoB,CAAC,MAAO,EAAQ,MAAO,EAAQ,OAAQ,GAC3D,EAA4B,CAAC,OAAM,EAAE,mBAAkB,EAAE,SAAQ,GACjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GACJ,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KCtG/C,AAiCA,YAAwB,GACtB,GAAM,GAAS,EAAgB,EAAO,QAAS,gBAAiB,WAEhE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAA8B,CAAC,MAAO,GACtC,EACF,EAAO,UAAU,GAAe,EAAgC,IACpE,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KC/C/C,AA4CA,YACI,EAA4B,EAC5B,EACA,GADA,AAAA,IAAA,QAAA,GAAA,GACA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,mBAAoB,WAEnE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,mDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAAiC,CAAC,MAAO,GACzC,EAA+B,CAAC,QAAO,EAAE,UAAS,EAAE,OAAM,GAC1D,EAAM,EAAO,UACf,GAAkB,EAClB,GACJ,MAAO,GAGF,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC/DrD,AAoBA,YACI,EAAiB,EAAkB,EACnC,EAAsB,EAAwB,GAMhD,AAAI,GAAgB,MAClB,GAAe,IAEjB,AAAI,GAAkB,MACpB,GAAiB,OAAO,mBAE1B,AAAI,GAAgB,MAClB,GAAe,GAGjB,GAAM,GAAW,EAAM,MAAM,GAC7B,SAAgB,KAAK,IAAI,EAAe,GAExC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAClE,EACI,EAAM,OAAS,EACf,WAAM,MAAA,+CAA+C,EAAM,KAAI,MACnE,EACI,EAAM,MAAM,KAAO,EACnB,WACI,MAAA,oDAAoD,EAAM,MAAM,KACxE,EAAY,EAAO,OAAS,EAAG,WAAM,MAAA,+BACrC,EACI,EAAO,MAAM,KAAO,EACpB,WAAM,MAAA,sDAAsD,EAAQ,KAChE,YAAW,EAAO,MAAM,MAChC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAC3D,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GC3DnE,AA0BA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBAC1B,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAExB,GAAM,GAAQ,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,GAC1D,MAAO,GAAO,cACV,SAAA,GAAK,MAAA,GAAE,kBACH,EAAQ,EAAS,EAAe,EAAc,IAClD,CAAC,MAAO,EAAQ,OAAQ,GAAU,KAAiB,GACnD,GAGC,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC/CvD,YA+BI,EAAU,EAAY,GACxB,GAAM,GAAQ,GAAa,EAAK,EAAS,GACnC,EAAiB,EAAQ,EAAI,CAAE,GAAQ,GAAK,EAClD,EAAI,OAAO,EAAgB,EAAG,eAmB5B,EAAU,EAAW,GACvB,MAAO,IAAc,EAAK,EAAQ,GAAc,IAUlD,YAA8B,EAAM,GAClC,MAAO,GAAI,EAAI,EAAI,EAAI,EAAI,GAAK,EAGlC,YACI,EAAU,EAAW,GAKvB,OAJI,GAAO,EACP,EAAQ,EAAI,OACZ,EAAS,EACT,EAAQ,GACL,EAAO,IACZ,EAAS,EAAS,GAAQ,IAAU,GACpC,GAAM,GAAgB,EAAW,EAAQ,EAAI,IAC7C,AAAI,EAAgB,EAClB,EAAO,EAAS,EAEhB,GAAQ,EAGR,EAAQ,CAAC,GAIb,MAAO,GAAQ,EAAO,CAAC,EAAO,ECvFhC,YAoCI,EAAmB,EAAoB,EACvC,EAAsB,GACxB,MAAO,IACI,EAAO,EAAQ,EAAe,EAAc,EAC5C,GACC,4BAIV,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAC5C,EAAsB,GACtB,EAA6C,gBAK/C,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAAgB,EAC5D,IAGN,YACI,EAAmB,EAAoB,EACvC,EAAsB,EAAwB,EAC9C,EAA4B,EAC5B,GADA,AAAA,IAAA,QAAA,GAAA,IAA4B,AAAA,IAAA,QAAA,GAAA,IAC5B,AAAA,IAAA,QAAA,GAAA,IAKF,OAFM,GAAa,GAEV,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,EAAO,GAAK,GACd,EAAW,KAAK,CAAC,MAAO,EAAO,GAAI,SAAU,EAAG,mBAAoB,IAIxE,EAAW,KAAK,IAShB,OALM,GAAQ,EAAe,EAAK,IAAO,EAAgB,EAEnD,EAA4B,GAC5B,EAA2B,GAE1B,EAAgB,OAAS,GAAiB,EAAW,OAAS,IACnE,GAAM,GAAY,EAAW,MACtB,EAAA,EAAA,MAAsB,EAAA,EAAA,SAAU,EAAA,EAAA,mBAEvC,GAAI,EAAgB,EAClB,MAUF,OADI,GAAkB,GACb,EAAI,EAAgB,OAAS,EAAG,GAAK,EAAoB,EAAE,GAClE,GAAM,GAAM,GAAsB,EAAO,EAAU,EAAgB,IAEnE,GAAI,GAAO,GACT,EAAkB,GAClB,MAMF,GAHA,EAAU,MACN,EAAU,MAAQ,GAAe,EAAc,EAAO,GAEtD,EAAU,OAAS,EACrB,MAWJ,EAAU,mBAAqB,EAAgB,OAE/C,AAAK,GAGH,CAAI,EAAU,QAAU,EACtB,GAAgB,KAAK,GACrB,EAAe,KAAK,EAAU,QACzB,AAAI,EAAU,MAAQ,GAG3B,GAAa,EAAY,EAAW,KAM1C,GAAM,GAAe,EAAgB,OAC/B,EAAa,EAAgB,EAEnC,AAAI,GAAsB,EAAa,GACrC,GAAgB,KAAI,MAApB,EAAwB,GAAI,OAAM,GAAY,KAAK,IACnD,EAAe,KAAI,MAAnB,EAAuB,GAAI,OAAM,GAAY,KAAK,KAGpD,GAAM,GACe,CAAC,gBAAiB,GAAS,EAAiB,UAEjE,MAAI,IACF,GAAO,eAAoB,GAAS,EAAgB,YAGtD,AAAI,GACF,GAAO,aAAkB,GAAO,EAAc,UAGzC,EAGT,YAA+B,EAAmB,EAAW,GAC3D,GAAM,GAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAS,GAAQ,GAAU,GAAQ,GACnC,EAAS,GAAQ,GAAU,GAAQ,GACzC,GAAI,GAAS,GAAK,GAAS,EACzB,MAAO,GAET,GAAM,GAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAmB,EAAkB,GACnE,KAAK,IAAI,EAAmB,EAAkB,GAClD,MAAO,GAAoB,GAAQ,EAAQ,GAO7C,YAAwB,EAAsB,EAAe,GAC3D,GAAM,GAAS,KAAK,IAAI,EAAQ,EAAM,GACtC,MAAO,IAAO,EAAe,EAAS,EAGxC,YAA6B,EAAe,GAK1C,MAAQ,GAAG,MAAQ,EAAG,OAChB,EAAG,QAAU,EAAG,OAAW,EAAG,SAAW,EAAG,SC7MpD,AA2CA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,wHACpB,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAED,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,GACxD,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAyB,GC1EtC,AAyDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,GACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEtB,GAAM,GAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EACF,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GAExD,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,eAAgB,EAAO,IAGtD,GAAM,IAA6B,EAAG,CAAC,4BAA2B,KChCzE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,wGACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEC,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,EACpD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAkC,GCxF/C,AAmDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAEzB,EAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EAAkC,CACtC,cAAe,EACf,aAAc,EACd,eAAgB,EAChB,mBAAkB,GAGd,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,aAAc,EAAO,IAGpD,GAAM,IAA0B,EAAG,CAAC,yBAAwB,KCnCnE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,+GACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAG3B,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBADxC,SACF,EAAA,OADG,EAAS,EAAA,GAAE,EAAU,EAAA,GAMtB,EAAM,GACR,EAAW,EAAY,EAAgB,EAAe,EACtD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAA+B,GChF5C,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,2DACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,6DACC,GAAI,OAEf,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAG9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GACZ,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,eACX,EAAa,EAAW,EAAU,IAGlC,EAA+B,CAAC,OAAQ,GACxC,EAA6B,CAAC,aAAY,EAAE,KAAI,GAEhD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCpFjD,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,yBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,kEACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WACI,MAAA,oEACG,GAAI,OACf,EACI,EAAQ,QAAU,WAAa,EAAQ,QAAU,QACjD,WAAM,MAAA,qDAEV,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAE9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GAEZ,EAAsC,CAAC,OAAQ,GAC/C,EAAoC,CAAC,aAAY,EAAE,KAAI,GAEvD,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,sBACX,EAAa,EAAW,EAAU,IAGlC,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAwB,EAAG,CAAC,uBAAsB,KCxF/D,AAyEA,YACI,EAAiB,EAAkB,GACrC,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAClE,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAElE,GAAM,GAAK,EAAgB,EAAG,IAAK,YAEnC,EACI,EAAG,MAAQ,EACX,WAAM,MAAA,4CAA4C,EAAG,KAAI,MAE7D,GAAM,GAAQ,EAAG,MACX,EAAA,EAAA,MAAA,MAAA,IAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,kDAAiD,EAAC,OAExD,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,qDAAoD,EAAC,OAG3D,AAAI,EAAW,GACb,GAAW,GAEb,AAAI,EAAW,GACb,GAAW,GAGb,GAAM,GAAI,EAAQ,GAAM,EAAG,EAAG,EAAG,SAAU,CAAC,GAAI,IAC1C,EAAI,GAAM,EAAG,EAAG,EAAG,SACnB,EAAK,GAAI,EAAG,GAEZ,EAAS,GACX,GAAU,EAAI,GAAO,CAAC,EAAU,UAChC,GAAa,EAAI,GAAO,CAAC,EAAU,WAEjC,EAAO,GAAM,CAAC,EAAG,GAAI,EAAG,OAE9B,MAAO,GACI,GAAM,GAAQ,EAAQ,EAAI,CAAC,GAAI,EAAG,KACvB,IAAI,SAAA,GAAO,MAAA,IAAM,EAAQ,EAAK,MACzC,GAGN,GAAM,IAAW,EAAG,CAAC,UAAS,KC7HrC,AA2DA,YAAsB,GACpB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,EAAkB,GAClB,EACI,GAAM,MAAQ,EAAG,OAAS,EAC1B,WAAM,MAAA,sEAGV,OADM,GAAM,EAAG,GAAG,MAAM,cACf,GACP,EACI,EAAG,GAAG,MAAM,KAAO,EACnB,WACI,MAAA,gEACA,KAAK,EAAkB,GAAG,MAAM,GAAE,QAAQ,EAAG,QAL9C,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,OAQT,GAAkB,GAClB,EAAK,GAAM,EAAI,EAAG,MAAM,GAAI,GAAG,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,CAAC,MAGtD,EACI,EAAG,QAAU,EAAG,GAAG,MAAM,GACzB,WAAM,MAAA,oCACK,EAAkB,OAAM,aAC/B,0BAA0B,EAAkB,GAAG,MAAM,GAAE,QAI/D,OAFM,GAAiB,GACjB,EAAO,aACJ,GACP,EAAG,KAAK,EAAO,KAAK,WAClB,GAAI,GAAI,EAAK,GACb,GAAI,EAAI,EACN,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAO,EAAI,GAAI,EAAI,EAAG,GAAI,IAAK,EAAG,IACxC,EAAI,GAAI,EAAG,GAGf,MAAO,IAAI,EAAG,GAAK,EAAG,kBATjB,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,GAaT,MAAI,GACK,GAAM,EAAI,GAEV,EAIJ,GAAM,IAAc,EAAG,CAAC,aAAY,KC5G3C,AAmFA,YAAa,EAAW,GAMtB,GANsB,AAAA,IAAA,QAAA,GAAA,IACtB,EACI,EAAE,MAAQ,EACV,WAAM,MAAA,gEACF,EAAE,OAEN,EAAE,OAAS,EACb,MAAO,IAAK,EAAe,GAM3B,GAAM,GAAgB,EAAE,MAAM,MAAM,EAAG,EAAE,MAAM,OAAS,GAC7B,OAAO,SAAC,EAAO,GAAS,MAAA,GAAQ,IACrD,EAAO,GACT,EACI,EACA,CACE,EAAe,EAAE,MAAM,EAAE,MAAM,OAAS,GACxC,EAAE,MAAM,EAAE,MAAM,OAAS,KAE/B,GACE,EAAmB,GACnB,EAAmB,GACzB,EAAK,QAAQ,SAAA,GACL,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,KAAK,GACV,EAAK,KAAK,KAEZ,GAAM,GAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OAC9B,EAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OACpC,MAAO,CAAC,EAAG,GAIf,YAAc,EAAa,GAAA,MAAA,KAAA,QAAA,GAAA,IAClB,EAAO,KAAK,WACjB,EACI,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,0CACF,EAAE,MAAM,OAAM,cAYtB,OAVM,GAAI,EAAE,MAAM,GACZ,EAAI,EAAE,MAAM,GAEd,EAAI,GAAI,GACR,EAAI,GAAM,GAER,EAAQ,GAAS,CAAC,CAAC,IAAK,CAAC,EAAG,IAC9B,EAAc,GAAM,GAElB,EAAQ,GAAK,EAAI,EAAI,aAClB,SAGD,EAAQ,EACR,EAAQ,EACR,EAAQ,EACd,EAAA,EAAA,KAAA,wfAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GA8CP,GAAQ,CAAC,EAAO,EAAO,KApDhB,EAAI,EAAG,EAAI,EAAO,EAAE,IAApB,GAuDT,MAAI,CAAC,GAAgB,EAAI,GACvB,GAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,IACzB,EAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,KAGpB,CAAC,EAAG,KAIR,GAAM,IAAK,EAAG,CAAC,IAAG,KCxMzB,AAiBA,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,IAAA,GAAA,MACA,EAAA,EAAA,uBAAA,GAAA,2BAJU,EAAA,WAAA,GAAA,UAAS,KCyBrB,YACI,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GAAM,GAAgB,GAAY,KAAQ,EAAU,EAAI,EAAS,GAEjE,GAAI,IAAc,EAAA,UAAU,KAC1B,MAAO,GAET,GAAI,IAAc,EAAA,UAAU,IAC1B,MAAO,IAAI,GAEb,GAAI,IAAc,EAAA,UAAU,MAC1B,GAAI,GAAY,KACd,MAAO,IAAK,GAEZ,GAAM,GAAkB,EAAQ,KAAO,EAAS,KAC1C,EAAS,GAAI,GAAI,GAAe,GAAI,IAC1C,MAAO,GAAkB,EAAI,GAAI,EAAQ,GAAO,IACnB,EAGjC,GAAI,IAAc,EAAA,UAAU,wBAC1B,GAAI,GAAY,KACd,MAAO,IAAI,GAAI,GAAe,GAAO,EAAQ,OAE7C,GAAM,GAAqB,EAAI,EAAU,GAAK,EAAQ,QAEhD,EACF,GAAK,GAAI,GAAS,EAAoB,GAAO,KAAM,WACvD,MAAO,IAAI,GAAI,GAAe,GAIlC,KAAM,OAAM,sBAAsB,GAE7B,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCnF3D,AA2CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,sBAC5C,EACF,EAAgB,EAAa,cAAe,sBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,uBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,iCAEvC,GAAM,GAAS,GAAI,GAAI,EAAS,IAChC,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAqB,EAAG,CAAC,oBAAmB,KChBzD,YACI,EAAsB,EAA2B,EACjD,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAC5C,EACF,EAAgB,EAAa,cAAe,kBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,mBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,6BAEvC,GAAM,GAAM,GAAO,GACb,EAAS,GAAI,EAAK,GAAI,EAAI,EAAS,GAAe,EAAM,KAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCnBjD,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAU,EAAgB,EAAQ,SAAU,aAC1C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAM,GAAO,GAEnB,EAAU,GAAI,EAAI,GAAO,GAAI,GAAU,GACvC,GAAM,GAAS,GAAK,GAAI,EAAK,EAAI,EAAS,KAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KC9DvC,AAiDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,aAC5C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAc,GAAO,GACrB,EAAQ,GAAI,GAAI,EAAc,IAC9B,EAAY,GAAQ,EAAO,GAC3B,EAAS,GAAI,EAAO,GAEpB,EACF,GAAI,EAAI,GAAO,IAAM,GAAO,IAAa,EAAI,EAAa,IAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KCtEvC,AAgDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,MAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,WAC5C,EAAe,EAAgB,EAAa,cAAe,WAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,YAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,sBAErD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAO,GAEvB,EAAK,GAAI,EAAI,EAAS,GAAI,GAAI,EAAc,MAC5C,EACF,EAAI,GAAI,EAAK,GAAU,GAAI,GAAI,GAAI,EAAK,GAAe,KACrD,EAAS,GAAI,EAAI,GACvB,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAU,EAAG,CAAC,SAAQ,KCrEnC,AA0CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,oBAC5C,EACF,EAAgB,EAAa,cAAe,oBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,qBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,+BAEvC,GAAM,GAAS,GAAkB,EAAS,GAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC3DrD,AAmCA,YACI,EAAsB,GACxB,GAAM,GACF,EAAgB,EAAQ,SAAU,iCAChC,EACF,EAAgB,EAAQ,SAAU,iCACtC,GACI,EAAQ,MAAO,EAAQ,MAAO,4CAsBlC,GAAM,GAAY,GAAK,GACjB,EAAgB,EAAI,EAAS,GAC7B,EAAgB,GAAM,GAAI,GAAI,GAAI,MAExC,MAAO,IAAI,GAAI,EAAW,GAAgB,GAwB5C,YACI,EAAgC,EAChC,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAoB,EACpB,EAAkB,mBAAoB,uBACpC,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KAOvB,GANA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAEjD,GACI,EAAkB,MAAO,EAAQ,MAAO,kCAExC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAO,GAAO,IAEpB,EACI,GAAI,EAAI,EAAmB,GAAI,EAAK,IAChC,EAAI,EAAM,IAEpB,GAAM,GAAS,GAA+B,EAAmB,GAEjE,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxH3D,AA+DA,YACI,EAAW,EAAW,GAKxB,GALwB,AAAA,IAAA,QAAA,GAAA,IACxB,AAAI,IAAQ,IACV,GAAM,EAAO,KAAO,GAGlB,IAAQ,EAAO,KAAO,EACxB,KAAM,OACF,+DACA,wCAAuC,EAAO,KAAI,KAClD,gBAAe,IAGrB,GAAM,GACF,GAAW,SAAC,EAAgB,EAAgB,GAI1C,GAAM,GAAW,GACX,EAAM,GAAU,EAAQ,CAAC,GAAM,GAC/B,EAAY,GAAI,GAAK,EAAQ,WAAY,GAC/C,EAAK,CAAC,EAAQ,IAEd,GAAM,GAAa,GAAI,EAAI,EAAW,IAChC,EAAW,GAAI,EAAY,CAAC,IAE5B,EAAW,SAAC,EAAO,GAChB,GAAA,GAAA,EAAA,GAAQ,EAAA,EAAA,GACT,EAAU,GAAqB,EAAG,MAAO,CAAC,IAChD,MAAO,CACL,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAK,EAAQ,WAAY,GAAI,KACrC,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAI,GAAY,GAAK,EAAQ,eAGzC,MAAO,CAAC,MAAK,EAAE,SAAQ,KAG7B,MAAO,GAAS,EAAQ,GAsB1B,YACI,EAA4B,EAC5B,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GACA,EAAgB,EAAc,eAAgB,uBAC5C,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KASvB,GAPA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GACI,EAAc,MAAO,EAAQ,MAAO,kCAEpC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAa,GAAO,EAAc,MAAM,IAE9C,EACI,GAAI,EAAI,EAAe,GAAI,EAAK,IAC5B,GAAI,EAAsB,IAGpC,GAAM,GAAS,GAA+B,EAAe,GAE7D,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KC3J3D,GAuNM,IAAW,CACf,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,MAAK,IASD,GAAS,CACb,cAAa,GACb,WAAU,GACV,MAAK,GACL,KAAI,IAeA,GAAQ,CACZ,cAAa,GACb,sBAAqB,GACrB,eAAc,GACd,iBAAgB,GAChB,cAAa,GACb,kBAAiB,GACjB,uBAAsB,GACtB,2BAA0B,GAC1B,gCAA+B,GAC/B,wBAAuB,GACvB,6BAA4B,IAOxB,GAAS,CACb,SAAQ,GACR,YAAW,GACX,GAAE,IAaE,GAAS,CACb,mBAAkB,GAClB,oBAAmB,GACnB,eAAc,GACd,UAAS,GACT,UAAS,GACT,QAAO,GACP,iBAAgB,GAChB,oBAAmB,GACnB,oBAAmB,IClSrB,mBAyCwC,GAAA,EAAA,GAAxC,4DAiBE,SAAA,UAAA,SAAA,SAAS,EAAiB,EAAoB,GAApB,AAAA,IAAA,QAAA,GAAA,IAElB,GAAA,GAAA,KAAA,iBAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,GAAI,GAAW,MACb,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,KAAM,OAAQ,EAAM,EAAE,SACrD,KAAK,eAAe,OAEpB,MAAK,eAAe,GAMtB,MAFA,IAAQ,GAEJ,EACK,EAEP,GAAM,UACC,OAOX,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAEd,KAAK,6CAGJ,EAAA,UAAA,oBAAV,WACE,KAAK,YAAc,KAAK,WAAa,GAgBvC,EAAA,UAAA,iBAAA,SAAiB,EAAiB,GAEhC,MAAO,IAAc,EAAG,IAgB1B,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,aAAe,MACtB,GAAQ,KAAK,cAIX,EAAA,UAAA,eAAN,8EACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAErB,CAAA,EAAO,CACL,KAAM,OAEN,OAAQ,GAAO,KAAK,YAAa,gBAI/B,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,gEAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OACN,4DACA,IAAG,KAAK,sBAUE,EAAA,UAAA,kBAAhB,SAAkC,mGAEhC,SAAA,KAAoB,CAAA,EAAM,EAAa,GAAG,OAAO,eAAjD,SAAK,YAAe,EAAA,OAAqC,GACzD,CAAA,EAAO,EAAa,MAAM,UAE9B,GA5HwC,IA8HxC,OAAO,eAAe,GAAW,OAAO,YAAa,CACnD,MAAO,SAAC,GACN,MAAO,GAAS,UAAY,MAAQ,EAAS,kBAAoB,MAC7D,EAAS,gBAAkB,QC1KnC,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAAgC,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,MAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,IAAA,EAChC,EAAA,QAAA,EALN,EAAA,iBAAwC,GACxC,EAAA,mBAA0C,GAOhD,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,cACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,aACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAC3C,EAAoB,EAAK,mBAAmB,GAAG,SAErD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAiB,EAAK,KAC1B,EAAI,GAAO,GAAW,EAAI,EAAK,MAEjC,EACF,EAAI,GAAI,GAAK,GAAI,EAAmB,EAAK,UACjC,GAAK,GAAI,EAAiB,EAAK,WACnC,GAEF,EACF,GAAI,EAAI,EAAmB,EAAK,KAC5B,EAAI,GAAO,GAAU,EAAI,EAAK,MAEtC,EAAgB,OAAO,GACvB,EAAkB,OAAO,GAEzB,GAAM,GAAW,GAAI,EAAI,EAAS,CAAC,EAAK,cAAe,GACvD,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,oBAAsB,MAC7B,IAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,YACzC,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,cAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,iBAAgB,OAAK,KAAK,oBAC/B,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,iBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,IAAO,KAAK,IACZ,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,IAAQ,EAAO,UAjHxD,EAAA,UAAY,WAmHrB,GArHuC,IAsHvC,GAAc,ICrJd,mBA+BsC,GAAA,EAAA,GAMpC,WACc,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,IAD5C,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,aAAA,EAA8B,EAAA,wBAAA,EAHpC,EAAA,iBAAwC,KAOhD,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAM,GAAY,GAClB,EAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,eACrB,SAAU,GACN,WAAM,MAAA,IAAK,EAAM,MAAO,EAAK,yBAClB,SAAS,MAI5B,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAEjD,GAAK,WACH,GAAM,GAAqB,GAAI,EAAiB,GAAO,IACvD,EAAgB,OAAO,GAEvB,GAAM,GAAW,GACb,EAAI,GAAI,EACA,GAAK,GAAI,EAAoB,EAAO,QAAQ,aAChD,CAAC,EAAK,cACV,GACJ,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,kBAAoB,MAC3B,GAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIvC,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,iBAAiB,IAC9D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,iBAAmB,EAAa,IACjC,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,wBAA2B,KAAK,0BAK7B,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,0BA/EzC,EAAA,UAAY,UAiFrB,GAnFsC,IAoFtC,GAAc,ICnHd,mBAkCmC,GAAA,EAAA,GASjC,WACc,EAAgC,EAChC,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MAFvC,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EAL/B,EAAA,uBAA8C,GAC9C,EAAA,wBAA+C,GAMrD,GAAK,WAEH,EAAK,SAAW,GAAO,GAAO,WAC9B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EAAmB,GAAI,EAAG,EAAK,UAErC,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,wBAAwB,IAAM,MACrC,GAAK,wBAAwB,GAAK,CAChC,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAe,EAAK,wBAAwB,GAAG,SAE/C,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QACvD,EACF,GAAI,EAAI,EAAc,EAAK,OACvB,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EAA2B,GAAI,EAAgB,GAC/C,EACF,GAAI,EAAiB,GAEzB,EAAY,OAAO,GACnB,EAAa,OAAO,GAEpB,GAAM,GACF,GAAI,EAAI,GAAI,EACA,GAAI,GAAK,GAA4B,EAAK,UAC9C,CAAC,EAAK,cACV,GACR,EAAM,OAAO,KAGf,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,QAC7C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,SAAS,UAEd,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,yBAA2B,MAClC,GAAQ,KAAK,wBAAwB,IAAI,SAAA,GAAK,MAAA,GAAE,aAI9C,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,yBACrC,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,4GACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACf,GAAK,WACH,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,IACxD,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,MAGpD,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,wBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,UA5IN,EAAA,UAAY,OA8IrB,GAhJmC,IAiJnC,GAAc,ICnLb,mBAiCoC,GAAA,EAAA,GASnC,WACc,EAAgC,EAChC,EAAyB,EACzB,GADyB,AAAA,IAAA,QAAA,GAAA,MACzB,AAAA,IAAA,QAAA,GAAA,GAHd,GAAA,GAIE,EAAA,KAAA,OAAO,KAHK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EACzB,EAAA,MAAA,EANN,EAAA,uBAA8C,GAC9C,EAAA,2BAAkD,GAQxD,GAAK,WACH,EAAK,UAAY,GAAO,GAAG,WAC3B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EACF,GAAI,CAAC,EAAK,aAAc,GAAI,EAAI,EAAK,UAAW,EAAK,OAAQ,IAEjE,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAGxC,AAAI,EAAK,2BAA2B,IAAM,MACxC,GAAK,2BAA2B,GAAK,CACnC,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAIxC,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAkB,EAAK,2BAA2B,GAAG,SAErD,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QAEvD,EAAM,EAAI,EAAiB,EAAK,OAChC,EAAM,GAAI,GAEV,EAAqB,GAAQ,EAAK,GAExC,EAAY,OAAO,GACnB,EAAgB,OAAO,GAEvB,GAAM,GACF,GAAI,EAAI,GAAI,EAAI,GACR,GAAI,EAAgB,GAAI,EAAoB,EAAK,WACrD,GAER,EAAM,OAAO,KAGf,EAAK,UAAU,OAAO,GAAI,EAAK,UAAW,IAC1C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,UAAU,UAEf,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,4BAA8B,MACrC,GAAQ,KAAK,2BAA2B,IAAI,SAAA,GAAK,MAAA,GAAE,aAIjD,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,wDAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OAAM,wDAGlB,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,QAChB,MAAS,KAAK,QAKX,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,QAAY,EAAO,QAvHzB,EAAA,UAAY,SAyHrB,GA3HqC,IA4HrC,GAAc,IC7Jd,mBA6BkC,GAAA,EAAA,GAKhC,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,aAAA,EAEpB,EAAK,gBAAgB,KAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAEF,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAK,WACH,GAAM,GAAW,GAAI,EAAI,EAAK,EAAG,GAAW,GAC5C,EAAM,OAAO,OAGjB,KAAK,uBAMP,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,aAAe,EACpB,AAAI,KAAK,GAAK,MACZ,KAAK,EAAE,UAET,KAAK,EAAI,GAAK,GAAO,CAAC,KAGxB,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,WAGH,EAAA,UAAA,WAAN,qGACU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAA,CAAQ,EAAA,eAGJ,EAAA,UAAA,WAAN,SAAiB,6FACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAC5C,GADA,EAAe,EAAA,OACX,EAAa,SAAW,EAC1B,KAAM,IAAI,OAAM,iEAIpB,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,aAAgB,KAAK,eAIxB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,eA7DjB,EAAA,UAAY,MA+DrB,GAjEkC,IAkElC,GAAc,IC/Fd,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAA8B,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFZ,GAAA,GAGE,EAAA,KAAA,KAAM,IAAa,KAFP,SAAA,aAAA,EAA8B,EAAA,SAAA,EAChC,EAAA,YAAA,EAJJ,EAAA,cAAqC,GAM3C,EAAK,EAAI,GAAO,EAAK,YAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,cAAc,IAAM,MAC3B,GAAM,GAAY,GAClB,EAAK,cAAc,GAAK,CACtB,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,MAInD,GAAM,GAAe,EAAK,cAAc,GAAG,SACrC,EAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAK,WACH,GAAI,GACE,EAAkB,GAAI,EAAI,EAAK,EAAG,GAAe,GACvD,AAAI,EAAK,YACP,EAAW,GACP,EAAI,EAAK,EAAG,GAAI,EAAU,EAAI,EAAiB,EAAK,KAAM,GAE9D,EAAW,GAAI,EAAI,EAAK,EAAG,GAAkB,GAE/C,EAAa,OAAO,GACpB,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,UACP,AAAI,KAAK,eAAiB,MACxB,GAAQ,KAAK,cAAc,IAAI,SAAA,GAAK,MAAA,GAAE,aAS1C,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,SAAW,GAGZ,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,cAAc,IAC3D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,cAAgB,EAAa,IAC9B,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,SAAY,KAAK,SACjB,YAAe,KAAK,cAKjB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,SAAa,EAAO,cA3FlD,EAAA,UAAY,WA6FrB,GA/FuC,IAgGvC,GAAc,IC/Hd,mBAgCsC,GAAA,EAAA,GASpC,WACc,EAAgC,EAChC,EAA0B,EACpC,GAF0C,AAAA,IAAA,QAAA,GAAA,IAChC,AAAA,IAAA,QAAA,GAAA,GAA0B,AAAA,IAAA,QAAA,GAAA,MACpC,AAAA,IAAA,QAAA,GAAA,IAHJ,GAAA,GAIE,EAAA,KAAA,OAAO,KAOP,GAVY,EAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,SAAA,EAA0B,EAAA,QAAA,EANhC,EAAA,uBAA8C,GAC9C,EAAA,mBAA0C,GAC1C,EAAA,qBAA4C,GAQlD,EAAK,SAAW,EAEhB,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,WAE5B,GAAgB,KAClB,KAAM,IAAI,OAAM,+DAIpB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,OACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,qBAAqB,IAAM,MAAQ,EAAK,UAC/C,GAAK,qBAAqB,GAAK,CAC7B,aAAiB,EAAI,MACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAwB,EAAK,uBAAuB,GAAG,SACvD,EAAqB,EAAK,mBAAmB,GAAG,SACtD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEvC,GAAI,EAAK,UACP,GAAM,GAAsB,EAAK,qBAAqB,GAAG,SAEnD,EACF,GAAI,EAAI,EAAqB,EAAK,OAC9B,EAAI,EAAU,EAAI,EAAK,QAEzB,EACF,GAAI,EAAI,EAAU,EAAK,cACnB,GACI,GAAI,EACA,GAAI,GAAO,GAAyB,EAAK,YACnD,EACF,GAAI,EAAI,EAAoB,EAAK,UAAW,GAEhD,EAAsB,OAAO,GAC7B,EAAoB,OAAO,GAC3B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAGb,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EACF,GAAI,EAAI,EAAoB,EAAK,UAC7B,GAAI,EAAI,EAAU,EAAK,cACnB,GAAK,GAAI,EAA0B,EAAK,YAEpD,EAAsB,OAAO,GAC7B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAInB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,sBAAwB,MAAQ,KAAK,UAC5C,GAAQ,KAAK,qBAAqB,IAAI,SAAA,GAAK,MAAA,GAAE,YAE/C,AAAI,KAAK,oBAAsB,MAC7B,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,oBAC7C,AAAI,KAAK,UACP,EAAU,KAAI,MAAd,EAAkB,KAAK,sBAEjB,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EACF,KAAK,SAAW,EAAa,OAAS,EAAI,EAAa,OAAS,EAC9D,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,MAE3C,AAAI,KAAK,UACP,MAAK,qBACD,EAAa,MAAM,EAAgB,EAAG,EAAgB,GACjD,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,iBAK/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,SAAY,KAAK,SACjB,QAAW,KAAK,QAChB,SAAY,KAAK,WAKd,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,SAChD,EAAO,QAAY,EAAO,WA5KzB,EAAA,UAAY,UA8KrB,GAhLsC,IAiLtC,GAAc,ICjNd,AAyBA,GAAA,IAAA,WAAA,cAsCS,SAAA,IAAP,SAAW,GACT,MAAO,IAAI,IAAa,IAkBnB,EAAA,SAAP,SAAgB,EAAsB,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE/C,GAAI,IAAkB,EAAc,EAAU,IAuBhD,EAAA,QAAP,SACI,EAAsB,EAAY,EAAgB,EAClD,GADsB,MAAA,KAAA,QAAA,GAAA,IAAY,AAAA,IAAA,QAAA,GAAA,GAAgB,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,IACK,GAAI,IACP,EAAc,EAAO,EAAU,EAAS,IAevC,EAAA,KAAP,SACI,EAAsB,EAAa,EACnC,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MACnC,AAAA,IAAA,QAAA,GAAA,MACK,GAAI,IAAc,EAAc,EAAO,EAAO,IAehD,EAAA,SAAP,SAAgB,EAAqB,EAAW,GAAhC,MAAA,KAAA,QAAA,GAAA,MAAqB,AAAA,IAAA,QAAA,GAAA,KAAW,AAAA,IAAA,QAAA,GAAA,MAEvC,GAAI,IAAkB,EAAc,EAAK,IAgB3C,EAAA,OAAP,SACI,EAAsB,EAAa,EAAe,EAClD,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MAAe,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,GACK,GAAI,IAAgB,EAAc,EAAO,EAAO,EAAS,IAmB3D,EAAA,QAAP,SAAe,EAAsB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE5B,GAAI,IAAiB,EAAc,IAE9C,KC5LA,GA+Ba,IAAQ,CACnB,IAAK,GAAsB,IAC3B,SAAU,GAAsB,SAChC,SAAU,GAAsB,SAChC,QAAS,GAAsB,QAC/B,QAAS,GAAsB,QAC/B,OAAQ,GAAsB,OAC9B,KAAM,GAAsB,MCtC9B,AAiBA,GAAM,IAA2B,WAC/B,MAAI,OAAO,wBAA0B,YAC5B,sBACE,MAAO,eAAiB,YAC1B,aAEF,SAAC,GAAgB,MAAA,SAa1B,cACE,MAAO,IAAI,SAAc,SAAA,GAAW,MAAA,IAAc,WAAM,MAAA,SCrC1D,YAmBI,EAAiC,EACjC,GACF,GAAM,GACF,EAAc,OAAO,IAAW,SAAW,EAAS,EAAO,IACzD,EACF,EAAe,OAAO,IAAW,SAAW,EAAS,EAAO,IAChE,MAAO,CAAC,EAAS,GCzBnB,YA0BI,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAqB,GACzB,GAAI,EACF,EAAW,EAAS,OAAO,EAAW,MAAM,IAC5C,EAAS,KAAK,EAAW,GAAK,GAC9B,EAAW,EAAS,OAAO,EAAW,MAAM,SAE5C,EAAW,EAAS,OAAO,EAAW,IAEtC,OADM,GAAgB,EAAW,OACxB,EAAI,EAAG,EAAI,EAAe,EAAE,EACnC,EACI,EAAS,OAAO,CAAC,EAAW,EAAI,GAAK,EAAW,GAAI,EAAW,KAErE,EAAW,EAAS,OAAO,EAAW,MAAM,EAAgB,IAE9D,MAAO,eAaL,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAW,GACjB,GAAI,GACF,EAAS,KAAK,GACd,OAAS,GAAI,EAAiB,EAAG,EAAI,EAAc,EAAE,EACnD,AAAI,GAAK,EAAI,EACX,GAAS,KAAK,GACd,EAAS,KAAK,EAAK,GAAiB,KAEpC,EAAS,KAAK,QAMlB,OAFM,GAAsB,GACtB,EAAqB,GAClB,EAAI,EAAG,EAAI,EAAc,EAAE,EAClC,AAAI,GAAK,EAAiB,EAAI,GAAK,EAAI,IAAM,EAC3C,EAAmB,KAAK,GAExB,EAAoB,KAAK,GAG7B,EAAS,KAAI,MAAb,EAAiB,GACjB,EAAS,KAAK,GACd,EAAS,KAAI,MAAb,EAAiB,GAEnB,MAAO,eAaL,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAmB,GAEzB,AAAI,EACF,EAAiB,KAAK,EAAW,GAAK,GAEtC,EAAiB,KAAK,EAAW,GAAK,GAGxC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,GAAK,EAAW,OAClB,AAAI,EACF,EAAiB,KAAK,EAAW,EAAI,GAAK,EAAW,IAErD,EAAiB,KAAK,EAAW,GAAK,EAAW,EAAI,IAGvD,EAAiB,KAAK,EAAW,IAIrC,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,CAAC,GACjB,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAiB,KAAK,EAAM,GAAG,IAEjC,MAAO,eAeL,EAA0B,EAAmB,GAE/C,OADM,GAAY,EAAe,MAAM,EAAG,GACjC,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAU,KAAK,EAAe,EAAI,GAAK,EAAM,GAAG,GAAK,EAAM,GAAG,IAGhE,MAAO,GCvJT,AAiBO,GAAM,IAAkB,mBAClB,GAAa,mBClB1B,AAiBO,GAAM,IAAQ,SACR,GAAS,WACT,GAAS,YACT,GAAS,YACT,GAAS,aACT,GAAS,YCtBtB,qBAmBqB,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACnB,AAAK,KAAM,QAAQ,YACjB,QAAQ,KAAI,MAAZ,QAAgB,wBAIA,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GAClB,AAAK,KAAM,QAAQ,YACjB,QAAQ,IAAG,MAAX,QAAe,GC3BnB,YAiCI,EAAoB,GACtB,GAAI,EAAK,SAAW,EAAK,OACvB,KAAM,IAAI,OACN,gEACG,GAAK,OAAM,WAAW,EAAK,OAAM,MAG1C,OADM,GAAS,GAAI,cAAa,EAAK,OAAS,GACrC,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EACtC,EAAO,GAAK,EAAK,EAAI,GACrB,EAAO,EAAI,GAAK,EAAK,EAAI,GAE3B,MAAO,eAiB8B,GAIrC,OAFM,GAAO,GAAI,cAAa,EAAQ,OAAS,GACzC,EAAO,GAAI,cAAa,EAAQ,OAAS,GACtC,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,EAAI,GAAK,EAAQ,GACtB,EAAK,EAAI,GAAK,EAAQ,EAAI,GAE5B,MAAO,CAAC,KAAI,EAAE,KAAI,eAOiB,GAKnC,OAHM,GAAM,KAAK,KAAK,EAAQ,OAAS,GACjC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAOgB,GAKlC,OAHM,GAAM,KAAK,MAAM,EAAQ,OAAS,GAClC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAShB,EAAuB,GACzB,GAAM,GAAO,EAAQ,EAAQ,GACvB,EAAO,EAAQ,EAAQ,EAAI,GACjC,MAAO,CAAC,KAAI,EAAE,KAAI,eAUhB,EAAkB,EAAc,EAAc,GAChD,EAAK,EAAQ,GAAK,EAClB,EAAK,EAAQ,EAAI,GAAK,cAOpB,EAAW,GAGb,OAFM,GAAO,GAAI,cAAa,EAAI,GAC5B,EAAO,GAAI,cAAa,EAAI,GACzB,EAAI,EAAG,EAAI,KAAK,KAAK,EAAI,GAAI,KACpC,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GAC9C,EAAK,GAAK,KAAK,IAAI,GACnB,EAAK,GAAK,KAAK,IAAI,GAErB,MAAO,CAAC,KAAI,EAAE,KAAI,eAOhB,EAAW,EAAW,GACxB,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GACxC,EAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GACtB,MAAO,CAAC,KAAI,EAAE,KAAI,GCvJpB,YAyDI,EAAM,EAAiB,GACzB,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,GAAE,QAEX,GAAM,GAAc,GAAM,EAAE,OACtB,EAAS,GAAK,EAAG,WACjB,EAAS,EAAQ,QAAQ,EAAQ,GACvC,SAAY,UACZ,EAAO,UACA,EAGT,GAAI,CAAC,GAAgB,EAAE,MAAO,GAG5B,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAE,MAAO,GAExD,GAAI,EAAE,QAAU,aACd,GAAM,GAAO,EAAQ,KAAK,GACpB,EAAS,GAAK,EAAM,GAC1B,SAAK,UACE,EAET,GAAI,IAAU,QACZ,MAAO,GAAQ,IAAI,GACd,GAAI,IAAU,QACnB,GAAM,GAAO,GAAO,EAAG,EAAE,OACnB,EAAS,EAAQ,SAAS,EAAG,GACnC,SAAK,UACE,MAEP,MAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,eAK/D,EAAM,GACR,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAO,EAAE,mBAG3B,EAAe,EAAc,GACxD,GAAM,GAAQ,GAAO,GAAU,GAAM,GAE/B,EAAS,GAAoB,EAAK,WACxC,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,qvCC3G1B,YAyBI,EAAM,EAAsB,GAC9B,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,MAAO,GAAW,IAAI,SAAA,GACpB,GAAM,GAAgB,EAAI,QAC1B,EAAU,GAAQ,EAClB,GAAM,GAAS,GAAM,EAAG,EAAO,GAC/B,SAAM,IAAS,EACR,ICjCX,YA2BI,EAAiC,GAEnC,OADM,GAAqB,GAAI,OAAM,EAAK,MACjC,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAK,MAAM,GAAK,EAAK,GAGrC,OADM,GAAS,GAAO,EAAU,EAAK,OAC5B,EAAI,EAAG,EAAI,EAAO,OAAO,OAAQ,EAAE,GAI1C,OAHM,GAAS,EAAO,WAAW,GAE3B,EAAwB,GAAI,OAAM,EAAK,MACpC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,MAAM,GAG1C,GAAM,GAAgB,EAAK,WAAW,GAEtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,WC7ChB,YAyBI,EAAe,EAAkB,EAAyB,EAC1D,GAOF,OALM,GAAU,EAAO,EAAO,OAAS,GACjC,EAAA,CAAA,EAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAc,GAAuB,EAAQ,EAAQ,GACrD,EAAiB,GAAuB,QAAS,EAAQ,GAEtD,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHM,GAAS,EAAI,EACb,EAAO,EAAE,SAAS,EAAQ,EAAS,GACnC,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAEzC,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAKrC,OAHM,GAAY,EAAI,EAChB,EAAW,EAAY,SAAS,EAAW,EAAY,GACvD,EAAc,EAAe,SAAS,EAAW,EAAY,GAC1D,EAAI,EAAG,EAAI,EAAG,IACrB,EAAS,GAAK,EAAU,GAAG,MAC3B,EAAY,GAAK,EAAU,GAAG,MAKlC,GAAM,GAAc,EAAO,QAC3B,SAAY,EAAY,OAAS,GAAK,EAC/B,CACL,GAAO,EAAa,EAAa,GACjC,GAAO,EAAgB,EAAa,UCxDxC,mJCAA,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,WAAY,SC7BtD,AA4BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAO,GAAK,EAAG,YACnB,EAAI,GAAK,GAAI,GAAO,GAAI,IAC9B,MAAO,IAAI,GAAI,EAAI,QCtC3B,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAK,EAAG,YAAa,IAC/C,MAAO,IAAI,EAAI,OCnCvB,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAGxB,MAAO,CAAC,EAAS,EAAG,KChDxB,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAsC,GAC5C,SAAM,QAAQ,SAAC,EAAG,GAChB,EAAK,GAAK,WAAM,MAAA,GAAG,WAEd,IC7BX,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,mBChChE,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,cAC7C,MAAO,IAAI,EAAI,OCpCvB,AA6BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,EAAI,EAAI,GAAI,EAAG,IACnB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,GAAI,EAAI,EAAI,GAAI,EAAG,KACvB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAExB,MAAO,CAAC,EAAS,EAAG,KCtDxB,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAK,EAAG,YAAa,QC/B7D,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,kBC/B3D,AA2DA,YACI,EAAkB,EAClB,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBAE3C,EAAO,EACP,EAAU,EACV,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,MAIvE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OAEvB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,IAG5C,EAAkC,CAAC,GAAI,EAAM,MAAO,GAEpD,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KChIvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAY,EAAS,EAAY,EACpD,OCpCV,AA8CA,YACI,EAAkB,EAAqB,EACvC,EAAkC,GACpC,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBAE/C,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,gCAC/B,EAAI,KAAI,MAEhB,GAAI,GAAU,EACV,EAAO,EACP,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EACI,EAAQ,EAAQ,CAAC,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KACvE,EAAO,EAAQ,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGhE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,4DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,+DACC,GAAQ,KAAI,OAEvB,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAmB,GAE3D,MAAO,GAAQ,gBAAgB,EAAM,EAAS,IAG1C,EAAgC,CAAC,GAAI,EAAM,MAAO,GAElD,EAA8B,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,GAEvD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAC5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAY,EAAS,OC7BvE,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,WAEnB,MAAI,CAAC,GAAc,CAAC,EACX,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,MAEtB,CAAC,GAAc,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,MAEtB,GAAc,CAAC,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,MAGzB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,KAC7B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,QChDrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,MACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAQtC,OAPM,GACF,EAEE,EAAa,EAAiB,WAC9B,EAAc,EAAiB,MAE/B,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAW,OAAS,EAAG,GAAK,EAAG,IAC1C,GAAI,EAAW,KAAO,EAAY,GAChC,EAAK,GAAK,UACD,EAAW,KAAO,EAC3B,KAAM,IAAI,OAAM,mBACZ,EAAU,6BAA6B,EAAW,MAI1D,OADM,GAAiB,GACd,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,EAAK,GAAK,GACZ,EAAK,KAAK,GAId,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAM,QC/CnC,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,GAAG,YCxBxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAET,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AA0BO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,aAAc,EAAA,EAAA,aACrB,MAAO,CACL,EAAG,WAAM,MAAA,IACL,GAAW,GAAa,EAAG,GAAe,GAAU,EAAG,IACvD,EAAI,GAAU,QCnCxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAS,EAAM,IAAI,SAAA,GAAK,MAAA,GAAE,QACzB,EAAA,EAAA,KACD,EAAQ,GAAe,EAAM,EAAM,GAAG,OAAO,GAC7C,EAAa,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAa,GAAM,EAAI,EAAY,GACzC,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,YAAM,MAAA,QC/BrC,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACN,EAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAEhC,SACI,GAA4B,GAC5B,WAAM,MAAA,8DACF,uDAAsD,EAAS,OAEhE,CACL,EAAG,WACC,MAAA,IAAoB,EAAI,MAAO,EAAI,EAAS,EAAS,EAAK,IAC9D,OAAQ,WACJ,MAAA,IAAqB,EAAK,EAAI,EAAQ,MAAO,EAAS,EAAK,OCxCrE,AAsBO,GAAM,IAA4C,CACvD,WAAY,GACZ,aAAc,CAAC,KAAM,UACrB,SAAU,SAAC,EAAe,EAAiB,GACnC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAAY,EAAA,EAAA,gBAGjC,MAAO,CACL,GAAI,WAAM,MAAA,IACN,EAAK,EAAQ,EAAS,EAAK,EAAY,EACvC,IACJ,OAAQ,WAAM,MAAA,IACV,EAAK,EAAI,EAAO,MAAO,EAAS,EAAK,EAAY,OCpC3D,AA2CA,YACI,EAAM,EAAO,EACb,EAA0C,GAC5C,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEnE,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAEzE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,EACI,EAAI,MAAM,KAAO,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAI,MAAM,GAAE,UAC1D,iCAAgC,EAAY,GAAE,OACtD,EACI,EAAK,MAAM,KAAO,EAAY,GAC9B,WAAM,MAAA,0CAA0C,EAAK,MAAM,GAAE,UACzD,mCAAkC,EAAY,GAAE,QAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAEhD,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,EAAG,GAEjD,EAAmC,CAAC,QAAO,EAAE,IAAG,GAEtD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAwB,GAG9B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC7F7D,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAC3B,EACI,GAAkB,GAClB,WACI,MAAA,kEACA,mDAAkD,EAAS,OAE5D,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAEZ,MAAO,CACL,EAAG,WAAM,MAAA,IACJ,EAAiB,MAAO,EAAI,EAAqB,EAAS,IAC/D,OAAQ,WAAM,MAAA,IACV,EAAiB,EAAK,EAAqB,MAAO,EAAS,OCzCrE,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAI,GAAK,EAAG,aAAc,OC/BvD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UAAW,EAAA,EAAA,QAExB,MAAO,CACL,EAAG,WACD,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAE,MAE7C,EAAM,GAAO,EAAI,EAAM,EAAW,CAAC,GAEvC,MAAI,IAAe,MACjB,GAAM,GAAU,EAAK,IAGhB,MCzCf,AAwBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAG1B,EAAa,GAAoB,CAAC,EAAG,GAE3C,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAU,OAEhB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,EACI,EAAE,OAAS,EACX,WAAM,MAAA,6DACF,yBAAwB,EAAE,KAAI,OACtC,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,8DACF,yBAAwB,EAAO,KAAI,OAC3C,EACI,EAAE,MAAM,KAAO,EAAO,MAAM,GAC5B,WAAM,MAAA,yDACF,cAAa,EAAE,MAAM,GAAE,0CACvB,cAAa,EAAO,MAAM,GAAE,OAEpC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,sCAAqC,EAAO,mBAC5C,KAAI,EAAU,QAEtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WACI,MAAA,gEACA,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAW,GACb,EAAE,MAAO,EAAO,MAAO,EAAS,EAChC,EAAK,EAAiB,IAE1B,MAAO,CACL,EAAG,WACC,MAAA,IAAmC,EAAE,MAAO,EAAI,EAAQ,IAC5D,OAAQ,WACJ,MAAA,IAAoC,EAAG,EAAI,EAAO,MAAO,OC7EnE,AAuBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAA6C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAC3D,EAA+C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAEnE,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,UACH,GAAyB,EACzB,IACb,OAAQ,WAAM,MAAA,GAAO,UACH,GACA,EAAsC,OCtC9D,AA6BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OACnD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAgC,EAAE,MAAO,GAC5D,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GAC/C,MAAO,GAAQ,OAAO,EAAI,IAGtB,EAAwB,CAAC,GAAE,EAAE,EAAC,GAEpC,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,cACZ,EAAoB,EAAgC,KACpD,QCrCV,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAI,EAAI,GAAI,GAAI,GAAO,KAAM,EAAI,KAAK,KAAK,KAAK,KACtD,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC/B7B,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC3B7B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAI,QC5BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AA6BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAiB,EAAE,MAAO,GAC7C,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AA8BO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,IAAK,OAAQ,WAAY,SACxC,SAAU,SACN,EAAY,EAAiB,GACxB,GAAA,GAAA,EAAA,gBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GAAM,EAAA,EAAA,GAAU,EAAA,EAAA,GAEpB,EAAa,GAAgB,GAAO,GACpC,EAAgB,GAAiB,EAAK,MAAO,EAAE,OAC/C,EAAsB,GAC5B,GAAI,EAAK,OAAS,GAChB,OAAS,GAAI,EAAG,EAAI,EAAE,MAAM,OAAS,EAAG,EAAE,EACxC,EAAU,KAAK,EAAE,MAAM,IAEzB,EAAU,KAAK,GAGjB,GAAM,GAAa,GAAI,EAAG,GACpB,EAAoB,EAAI,EAAI,GAC5B,EAAsB,GAAM,GAAI,EAAU,GAAO,KACjD,EAAiB,EACnB,EAAI,EAAI,EAAqB,GAAsB,GACnD,GAAO,MAEL,EAAO,WACX,MAAI,GAAK,OAAS,EACT,EACH,EAAI,EAAI,EACA,GACI,EAAQ,EAAqB,CAAC,EAAG,EAAG,EAAG,EAAK,MAAM,KAClD,IACR,GACJ,EAAE,OAEC,EAAQ,EAAI,EAAI,EAAI,GAAsB,GAAa,EAAE,QAG9D,EAAU,WACd,GAAI,GACA,EAAI,EAAI,EAAqB,GAAO,KAAM,GAC9C,MAAI,GAAK,OAAS,GAChB,GAAU,GAAI,EAAS,IAElB,EAAQ,EAAS,EAAK,QAEzB,EAAc,WAClB,GAAI,GAAc,EAAI,EAAI,EAAgB,GAAa,GAEvD,MAAI,GAAK,OAAS,GAChB,GAAc,GAAI,EAAa,IAE1B,EAAQ,EAAa,EAAK,QAE7B,EAAW,WACf,GAAM,GAAwB,EAAI,EAAY,GAE1C,EAAW,EAAI,EAAI,GACvB,MAAI,GAAK,OAAS,GAChB,GAAW,GAAI,EAAU,IAEpB,EAAQ,EAAU,EAAK,QAE1B,EAAY,WAChB,GAAI,GAAY,EAChB,MAAI,GAAK,OAAS,GAChB,GAAY,GAAI,EAAW,IAEtB,EAAQ,EAAW,EAAK,QAGjC,MAAO,CACL,EAAG,EACH,KAAM,EACN,SAAU,EACV,MAAO,EACP,OAAQ,KC1Gd,AA0BO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,WACpB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACH,EAAA,EAAA,KAED,EAAa,GAAe,EAAM,EAAE,OAAO,GAE3C,EAAO,WACX,GAAM,GAAc,EAAE,MAChB,EAAc,EAAQ,KAEtB,EAAa,EAAY,MAAM,EAAG,GAClC,EAAY,EAAW,OACvB,EAAa,EAAY,MAAM,EAAM,EAAY,QAAQ,MAAM,GAC/D,EAAY,EAAW,OAEvB,EAAmB,GAAW,EAAG,GACjC,EACF,GAAW,EAAY,EAAG,EAAY,EAAI,GAExC,EAAc,GAAY,CAAC,EAAY,CAAC,GAAc,IAEtD,EAAS,EAAQ,EAAI,GACrB,EAAkB,EAAQ,EAAS,CAAC,IAEpC,EACF,GAAY,CAAC,CAAC,GAAY,EAAkB,IAC1C,EAAkB,GAAU,EAAQ,GACtC,EAAa,GACb,EAAiB,EAA6B,EAAE,MAAM,IAEpD,EAAsB,GAAuB,GACnD,SAAa,GAAU,EAAY,GAE5B,GAET,MAAO,CAAC,EAAG,EAAM,QAAS,WAAM,MAAA,OAIpC,YAAoB,EAAe,GAEjC,OADM,GAAS,GACN,EAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,EAAO,KAAK,GAEd,MAAO,GAGT,YAAqB,GAEnB,OADM,GAAS,GACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,EACtC,EAAO,KAAK,EAAO,GAAG,IAG1B,MAAO,GCnFT,AAqBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,IAAI,EAAG,WAAM,MAAA,IAAU,OC1BtD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAK,EAAI,eCzB9B,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC5B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,EAAG,QC5BpC,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,EAAG,gBC5BrC,AAyBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,GACd,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KACP,MAAO,CACL,OAAQ,WACN,GAAM,GAAW,GACX,EAAU,GAAI,GACpB,MAAO,IAAI,EAAI,EAAI,GAAI,EAAI,EAAM,GAAW,QCpCpD,AAyBA,YACI,EAAM,EAAM,EAAO,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,QAAQ,EAAI,EAAG,EAAG,EAAa,EAAM,EAAO,IAElD,EAA4B,CAAC,EAAC,EAAE,EAAC,EAAE,GAAE,GAErC,EAA0B,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAE/D,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAa,GAGnB,GAAM,IACT,EAAG,CAAC,oCAAmC,KCxC3C,AAqBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,YAAa,EAAA,EAAA,KAAM,EAAA,EAAA,MAAO,EAAA,EAAA,KAEjC,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAG,EAAG,EAAI,EAAa,EAAM,EAAO,OC/B9C,YA6BI,EAAO,EAAM,EAAe,EAAoB,GAClD,MAAI,GAAE,KAAO,EAAM,MACjB,GAAI,EAAQ,EAAG,GAA+B,EAAE,MAAO,KAEzD,AAAI,EAAG,KAAO,EAAM,MAClB,GAAK,EAAQ,EAAI,GAA+B,EAAG,MAAO,KAErD,CACL,EAAG,WACD,GAAM,GAAK,EAAI,EAAI,GAAK,GAAM,EAAO,GAAI,EAAG,QAC5C,MAAO,IAAgB,KAAO,EAAK,GAAU,EAAI,KCvCvD,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,iBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAkB,EAAE,OACnD,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAa,EAAG,GAAI,aAC9C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,GAAI,aAC5C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA6DA,YACI,EAAkB,EAAqB,EACvC,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE9C,EAAO,EACP,EAAU,EACV,EAAW,EACX,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KAErE,EAAW,EAAQ,EAAS,CAC1B,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAI3E,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAS,OAAS,EAClB,WAAM,MAAA,kEACC,GAAS,KAAI,OAExB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OACrE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,EAAU,IAGtD,EACwB,CAAC,GAAI,EAAM,MAAO,EAAS,OAAQ,GAE3D,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC1IvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAG,EAAY,EAAS,EAAY,EACvD,OCrCV,AAkDA,YACI,EAAyB,EACzB,EAA6B,EAC7B,EAAkC,EAClC,GACF,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBACzC,EAAU,EAAgB,EAAQ,SAAU,mBAElD,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,+BAC/B,KAAI,EAAI,KAAI,OAEpB,EACI,EAAI,OAAS,EACb,WAAM,MAAA,4DACC,GAAI,KAAI,OACnB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,+DACC,GAAO,KAAI,OACtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAO,MAAO,EAAY,EAAS,EAAmB,EACtD,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAQ,EAAS,IAGjD,EACsB,CAAC,GAAI,EAAK,MAAO,EAAQ,OAAQ,GAEvD,EACqB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAErE,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,GAC/C,GAGN,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAE5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAG,EAAY,EAAS,OC/B1E,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,KACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAM,EAAE,OACvC,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAU,EAAG,GAAI,aAC3C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAQ,EAAG,GAAI,aAC/C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA4BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAI,GAAa,EAAE,OAEjC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAI,GAAM,GAAI,EAAG,MAC/B,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KClDxB,AA0BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,OCzBzB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,WACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAU,EAAM,GACtB,MAAO,CAAC,QAAS,WAAM,MAAA,IAAM,EAAQ,MAAO,eC3BhD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAGtC,GAAM,GAAI,EAAM,GACT,EAAA,EAAA,SACD,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAM,EAAI,EAAO,EAAE,WC/BxC,AAgCO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GACP,EAAO,EACP,EAAM,EACN,EACF,GAA0C,EAAK,MAAO,EAAI,OAExD,EAAU,WACd,GAAM,GAAW,GAAK,EAAK,WACvB,EAAM,EAAI,EAAI,EAAI,EAAU,GAAI,EAAM,GAAI,EAAU,GAAO,OACzD,EAAa,GAAgC,EAAK,MAAO,GAC/D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,QAErB,EAAS,WACb,GAAM,GAAY,GAAQ,EAAM,GAC1B,EAAU,GAAM,EAAW,GAAI,GAAO,GAAU,IAClD,EAAM,EAAI,EAAI,EAAI,EAAG,IACnB,EAAa,GAAgC,EAAI,MAAO,GAC9D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAI,QAE1B,MAAO,CAAC,EAAG,EAAS,EAAG,KC9D3B,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,SACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,GAAQ,EAAG,GAExB,MAAO,CACL,EAAG,WAAM,MAAA,IAAM,EAAM,EAAI,EAAI,EAAI,KACjC,MAAO,WACL,GAAI,GAAM,GAAM,EAAM,GAAU,GAAK,EAAI,EAAI,IACvC,EAAa,GAAiB,EAAM,MAAO,EAAG,OACpD,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAM,WC1ClC,AAwBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,SC7BxC,AAwBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAO,EAAI,GAAU,EAAG,GAAI,GAAK,IAEvC,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAM,gBC/BxC,AAuBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,GAAI,gBC5B3C,AAqBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAQ,EAAI,EAAE,WC1BnC,AAuBO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,uBACX,EAAI,EAAoB,IAGxB,EAAmC,CAAC,OAAM,GAC1C,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAAoB,IAExB,MAAO,CAAC,OAAQ,KCxCpB,AAuBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,8BACX,EAAI,EAAoB,IAGxB,EAA0C,CAAC,OAAM,GACjD,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAA2B,IAE/B,MAAO,CAAC,OAAQ,KCxCpB,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,KACD,EAAO,GAAe,EAAM,EAAG,OACrC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAQ,EAAI,OC3BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAyBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,GAAI,EAAI,EAAI,GAAI,EAAG,KAAM,SC9BlD,AAyBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,aACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAGL,UAAW,WAAM,MAAA,IAAK,GAAU,GAAY,YAC5C,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAW,EAAG,SACpC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAW,GAAY,EAAG,YCnCtD,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CACL,EAAG,WACD,GAAM,GAAO,GAAQ,EAAG,GAAO,IAEzB,EAAa,GAAO,IACpB,EAAQ,GAAO,IAEf,EAAqB,EAAI,EAAI,GAC7B,EACF,EAAI,EAAI,EAAI,GAAa,GAAI,GAAK,EAAG,aAEzC,MAAO,IAAM,EAAM,EAAoB,OC3C/C,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,EAAG,GAAI,GAAO,GAAI,SC9BnD,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAK,EAAG,YAAa,OC7BlD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GActC,OAbO,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,KAER,EAAa,EAAE,MACf,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAQT,EAAoC,GACjC,EAAI,EAAG,EAAI,EAAG,KAAM,IAC3B,EAAS,KAAK,CAAC,EAAO,GAAI,EAAW,GAAK,EAAO,GAAK,EAAM,KAE9D,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,OC3C7B,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,IACD,EAAW,GAEX,EAAW,EAAI,EAAI,GACzB,MAAO,CACL,OAAQ,WAAM,MAAA,IAAI,EAAU,EAAI,GAAI,EAAU,CAAC,GAAM,GAAW,QClCtE,AAuBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAQ,QC7BrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,SACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,KAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAO,EAAI,OC1BhC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAI,GAAK,GAAK,EAAG,YAAa,QC/B3D,AAuBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,GAAK,EAAG,WAAY,QC5BrD,AAwBO,GAAM,IAA0C,CACrD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAM,GAAO,GACb,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MACrC,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MAC3C,MAAO,CAAC,EAAS,EAAG,KChCxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,GAAI,GAAM,EAAE,QAG7B,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAkB,EAAE,MAAM,QACzB,EAAA,EAAA,KAED,EAAO,GAAe,EAAM,EAAE,OACpC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,EAAI,EAAY,GAAK,EAAE,MAAO,YAE3C,MAAO,CAAC,EAAG,WAAM,MAAA,OCxCrB,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAO,GAAI,SC9BxC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAO,GAAI,GAAO,IAAK,OC/BpD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KAED,EAAO,WACX,GAAI,GAAQ,GAAU,GAGtB,GAAI,EAAE,OAAS,EACb,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,IAAK,CAAC,EAAE,MAAM,cAEjD,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IAAK,CACjD,EAAE,MAAM,GAAI,EAAE,MAAM,cAI7B,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IACjD,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,cAI1C,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EACA,CACE,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAC5C,EAAI,EAAE,MAAM,IAEd,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,UAMjE,MAAM,IAAI,OACN,2DACG,GAAE,KAAI,kBAEf,MAAO,IAET,MAAO,CAAC,EAAG,KCpFf,AAuBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAiC,EAChC,EAAA,EAAA,KACD,EAAW,GAAiC,GAClD,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,EAAI,OC7BnC,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACxC,GAAM,GAA2B,EAC1B,EAAA,EAAA,KACP,MAAO,CAAC,MAAO,WAAM,MAAA,IAAM,EAAI,OC1BnC,AA8BO,GAAM,IAA2C,CACtD,WAAY,GACZ,aAAc,CAAC,cACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAAO,WACX,MAAO,IAAoB,EAAI,IAEjC,MAAO,CAAC,EAAG,KAIf,YAA+C,EAAM,GAQnD,OAJM,GAAqB,GAAQ,EAAS,GAAU,IAChD,EAAW,GAAO,EAAG,GACvB,EAAa,GAAa,EAAS,GAAO,EAAG,UAC3C,EAAW,EAAS,KAAO,EAAW,KACnC,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,GAAW,EAAY,EAAI,GAE1C,EAAa,GAAW,EAAY,GAAK,EAAS,MAAO,SACzD,GAAM,GAAY,GAAU,GAC5B,MAAO,IAAM,EAAY,EAAU,GCxDrC,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAkHA,GAAM,IAA4B,CAChC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA6B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAxB,GAAM,IAAc,GAAA,IACvB,GAAiB,ICxNnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA+BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACL,EAAO,KAAK,OAAS,EAAG,WAAM,MAAA,wCACvB,EAAQ,KAAM,KClCvB,AAmCA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAQ,KAAM,ICtCvB,AA8BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AAkCA,EAAO,UAAU,KAAO,SACpB,EAAc,GAChB,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,KCrC9B,AAoCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,GACjC,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,KCvCvC,AAqCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,GAChD,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,KCxC9C,AAwCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,EAC9C,GACF,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,EAAQ,KC5CtD,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA+BA,EAAO,UAAU,UAAY,SACzB,EACA,EACA,EACA,EACA,GACF,YAAK,kBACE,GAAU,KAAM,EAAM,EAAU,EAAQ,EAAO,ICtCxD,AA2BA,EAAO,UAAU,YAAc,SAAyB,GAEtD,YAAK,kBACE,GAAY,KAAM,IC9B3B,AA4BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,YAAc,SAC3B,EAAa,GACf,YAAK,kBACE,GAAY,KAAM,EAAK,IC/BhC,AA0BA,EAAO,UAAU,OAAS,SACtB,EAA0B,GAC5B,YAAK,kBACL,AAAI,YAAa,IACf,GAAI,CAAC,IAEA,GAAM,CAAE,MAAI,OAAK,GAAI,IChC9B,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4C,EAC5C,EAAmB,GACrB,YAAK,kBACE,GACI,KAAM,EAAQ,EAAQ,EAAK,EAAY,EACvC,ICrCb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EACA,EAAkC,EAClC,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAa,EAAS,EAAK,ICrCtD,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAqB,GACtC,YAAK,kBACE,GAAO,KAAM,EAAM,EAAW,IC9BvC,AA2BA,EAAO,UAAU,aAAe,SAC5B,EAAmB,GACrB,YAAK,kBACE,GAAa,KAAM,EAAW,IC9BvC,AAkCA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,UAAgB,8DAChB,KAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,IC3Cb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA6BA,EAAO,UAAU,WAAa,SAC1B,EAA+B,EAC/B,EAAqB,EACrB,GACF,YAAK,kBACE,GAAW,KAAM,EAAQ,EAAS,EAAK,EAAW,IClC3D,AA2BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC9BxB,AA4BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC/BzB,AA2BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC7BnB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC5Bb,AA+BA,EAAO,UAAU,YAAc,SAClB,GACX,YAAK,kBACE,GAAY,KAAM,IClC3B,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA0BA,EAAO,UAAU,WAAa,SAA2B,GACvD,YAAK,kBACE,GAAW,KAAM,IC5B1B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA4BA,EAAO,UAAU,OAAS,SACb,EAA4B,GACvC,YAAK,kBACE,GAAO,KAAM,EAAS,IC/B/B,AA+BA,EAAO,UAAU,mBAAqB,SACzB,GACX,YAAK,kBACE,GAAmB,KAAM,IClClC,AA0BA,EAAO,UAAU,aAAe,SAA2B,GAEzD,YAAK,kBACE,GAAa,KAAM,IC7B5B,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA+BA,EAAO,UAAU,gBAAkB,SACtB,GACX,YAAK,kBACE,GAAgB,KAAM,IClC/B,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA0BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC5BpB,AA2BA,EAAO,UAAU,2BAA6B,SAC1C,EAAsB,EAAe,EAAgB,GACvD,YAAK,kBACE,GAA2B,KAAM,EAAa,EAAM,EAAO,IC9BpE,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA2BA,EAAO,UAAU,UAAY,SAChB,EAAwB,GACnC,YAAK,kBACE,GAAU,KAAM,EAAM,IC9B/B,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA0BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC5BpB,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA2BA,EAAO,UAAU,OAAS,SACb,EAAiB,EAAsB,GAClD,YAAK,kBACE,GAAO,KAAM,EAAG,EAAY,IC9BrC,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,KAAO,SACpB,EAAwB,GAC1B,YAAK,kBACE,GAAK,KAAM,EAAM,IC7B1B,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,EAAI,KAAM,IC5BnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,KAAO,SACpB,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAK,EAAM,IClC/B,AA+BA,EAAO,UAAU,eAAiB,SACrB,GACX,YAAK,kBACE,GAAe,KAAM,IClC9B,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAa,GAAb,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAC9B,KAAK,kBACE,GAAO,KAAM,EAAO,EAAS,IC9BtC,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,IAAM,SACV,EAAmC,GAC9C,YAAK,kBACE,GAAI,KAAM,EAAU,IC/B7B,AA6BA,EAAO,UAAU,KAAO,SACX,EAAsC,EAC/C,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAa,EAAa,EAAS,EAAc,IClCrE,AA+BA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,GAAU,KAAM,ICjCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,MAAQ,SACZ,GACX,YAAK,kBACE,GAAM,KAAM,IC7BrB,AA2BA,EAAO,UAAU,KAAO,SACX,EAAwB,GACnC,YAAK,kBACE,GAAK,KAAM,EAAM,IC9B1B,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA0BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC5Bf,AAkCA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,EAAQ,KAAM,EAAE,QCpCzB,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,EAAQ,KAAM,IC5BvB,AA2BA,EAAO,UAAU,eAAiB,SACrB,EAA8B,GACzC,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,sBAAwB,SAC5B,EAA8B,GACzC,YAAK,kBACE,GAAsB,KAAM,EAAY,IC9BjD,AA0BA,EAAO,UAAU,QAAU,SACd,GACX,YAAK,kBACE,GAAQ,KAAM,IC7BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EAAsC,EACtC,EAAqB,EACrB,GACF,YAAK,kBACE,GACI,KAAM,EAAiB,EAAiB,EAAS,EAAK,EACtD,ICtCb,AA4BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,OC9BjB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA6BA,EAAO,UAAU,MAAQ,SACZ,EAAwB,GACnC,YAAK,kBACE,GAAM,KAAM,EAAO,IChC5B,AA4BA,EAAO,UAAU,QAAU,SAAoC,GAC7D,YAAK,kBACE,GAAQ,KAAM,IC9BvB,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,MAAQ,SACrB,EAAkC,GACpC,YAAK,kBACE,GAAM,KAAM,EAAiB,IC9BtC,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAO,OC9BhB,AA2BA,EAAO,UAAU,kBAAoB,SAA2B,GAE9D,YAAK,kBACE,GAAkB,KAAM,IC9BjC,AA+BA,EAAO,UAAU,wBAA0B,SAC9B,GACX,YAAK,kBACE,GAAwB,KAAM,IClCvC,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,MAAQ,SACrB,EAAoB,GACtB,KAAK,kBACL,GAAM,GAAqB,YAAa,GAAS,CAAC,KAAM,GAAE,CAAI,MAAI,OAAK,GACvE,MAAO,IAAM,EAAoB,IC9BnC,AA4BA,EAAO,UAAU,KAAO,SAAoC,GAC1D,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA+BA,EAAO,UAAU,aAAe,SACd,EAAiB,EAAe,EAC9C,EAAoB,EAAkB,EACtC,EAAsB,GACxB,YAAK,kBACE,GACI,KAAM,EAAO,EAAK,EAAS,EAAW,EAAS,EAC/C,EAAa,ICtC1B,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC7BzB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC7BpB,AAgCA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAQ,KAAM,SClCvB,AAgCA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,KAAM,YClCvB,AAgCA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAQ,KAAM,UClCvB,AA4BA,EAAO,UAAU,KAAO,SACX,EAAY,GACvB,YAAK,kBACE,GAAK,KAAM,EAAG,IC/BvB,AA2BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC9BzB,AA2BA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAO,KAAM,IC9BtB,AA4BA,EAAO,UAAU,mBAAqB,SACzB,EAAmC,GAC9C,YAAK,kBACE,GAAmB,KAAM,EAAY,IC/B9C,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,MAAQ,SACrB,EAA8B,GAChC,YAAK,kBACE,GAAM,EAAW,KAAM,IC/BhC,AA4BA,EAAO,UAAU,UAAY,WAC3B,YAAK,kBACE,GAAU,4vQC9BnB,AAaA,GAAI,kBAMF,MAAI,KAAY,MACd,IAAW,EAAA,UAAU,WAEhB,iBAeP,MAAO,eCrCT,AAuBA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAChC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAa,aAE7C,MAAA,IANkC,OAWlC,EAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OAWhC,GAAA,SAAA,GAAyC,EAAA,EAAA,GACvC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAoB,aAEpD,MAAA,IANyC,OAWzC,GAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OC9EhC,YAuB6B,EAAY,GACvC,GAAI,MAAM,QAAQ,IAGhB,OADI,GAAkB,GACb,EAAI,EAAG,EAAI,EAAW,IAC7B,EAAW,EAAS,OAAO,GAE7B,MAAO,QAEP,GAAM,GAAW,GAAI,OAAM,GAC3B,SAAS,KAAK,GACP,eAIY,EAAc,GACnC,GAAI,CAAC,EACH,KAAM,IAAI,IAAe,eAOJ,EAAY,GAEnC,OADI,GAAU,EACK,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,IAAS,GACX,IAGJ,MAAO,eAQ2B,GAClC,MAAI,GAAG,SAAW,EACT,EAAG,GAEL,cAYc,GACrB,MAAI,OAAM,QAAQ,GACT,EAEF,CAAC,eA0BkB,GAC1B,GAAM,GAAe,EAAK,QAAQ,uBAAwB,SACpD,EACF,EAAa,QAAQ,kBAAmB,SAAS,cAKrD,MAAI,GAAS,KAAO,IACX,EAEF,UAAY,cAGO,GAE1B,MAAI,GAAW,QAAU,GAIrB,EAAW,QAAQ,OAAS,GAHvB,EAMF,EAAW,QAAQ,cAAe,SAAC,EAAG,GAAO,MAAA,GAAG,gBAIzD,GAAI,IAAyB,eAEQ,GAEnC,GAAI,GAAa,KACf,MAAO,MAET,GAAM,GAAsC,GAC5C,SAAK,UAAe,EAAS,eAC7B,EAAK,OAAY,EAAS,YACnB,EAcT,YAAuC,GAErC,GAAI,GAAU,MAAQ,MAAO,IAAW,SACtC,OACK,GAAI,MAAM,QAAQ,GACvB,EAAO,QAAQ,SAAA,GAAc,MAAA,IAA8B,SAG3D,QADM,GAAS,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAO,GACrB,AAAI,GAAS,MAAQ,MAAO,IAAU,UACpC,CAAI,CAAC,MAAM,QAAQ,IAAU,EAAM,OAAY,WAC3C,MAAO,GAAM,OAAa,SAC5B,EAAO,GAAS,EAAM,MAEtB,GAA8B,iBAqBpC,EACA,EACA,EACA,EAAgC,aAElC,GAJE,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,UAAgC,AAAA,IAAA,QAAA,GAAA,IAE9B,MAAO,IAAe,UACxB,GAAM,GAAe,EACjB,EAAE,OACN,GAAI,IAAgB,GAClB,EAAK,EAAc,WACV,IAAgB,IACzB,EAAK,GAAuB,WAE5B,EAAK,EAAc,GACf,GAAM,KACR,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAU;EAE7C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAKR,MAAO,QAGP,GAAM,GAAS,EACf,GAAI,EAAO,WAAgB,MAAQ,EAAO,QAAa,KACrD,KAAM,IAAI,GACH,EAAmB,6BACnB,MAAK,UAAU,GAAO;GACzB,sCAEN,GAAM,GAAY,EAAO,UACrB,EAAG,OAAE,EAAU,OAQnB,GAPA,AAAI,IAAa,GACf,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,GAAA,UAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IAEJ,GAAO,KACT,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAS;EAE5C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAIN,GAAI,GAAc,MAQhB,OADM,GAAwB,GACZ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,GAAuB,GAEtD,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,EAAc,GAG7C,GAAM,GAAe,EAAO,OAC5B,EAAa,cAAmB,EAGhC,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAE9C,GAA8B,EAAO,QACrC,GAAM,GACF,EAAW,EAAK,EAAO,OAAW,EAAe,GACrD,UAAsB,GAAA,GAAO,GAEtB,OAMP,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAK9C,GAAM,GAAY,GAAI,GAAI,EAAO,QACjC,UAAsB,GAAA,GAAO,GACtB,gBAUiB,EAAW,GACvC,MAAQ,GAAI,EAAK,GAAO,EAAI,EAAK,EAAI,cAQF,EAAW,GAC9C,MAAO,GAAK,GAAc,EAAG,eA2CL,GACxB,GAAI,GAAM,KACR,MAAO,GAIT,OAFM,GAAW,GAED,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAI,QAAQ,KAAO,IACrB,EAAI,KAAK,GAGb,MAAO,eASqB,GAC5B,GAAI,GAAO,KACT,KAAM,IAAI,GAAW,yBAAyB,KAAK,UAAU,IAE/D,OAAW,KAAO,GAChB,GAAI,EAAI,eAAe,GACrB,MAAO,GAGX,MAAO,eAWL,EAAkB,EAAe,GACnC,GAAI,GAAS,KACX,OAEF,GAAI,EAAO,QAAQ,GAAS,EAC1B,KAAM,IAAI,GAAc,EAAK,mBAAmB,EAAK,uBACjD,EAAM,mCAoBV,EAAQ,EAAsB,EAC9B,GAD8B,MAAA,KAAA,QAAA,GAAA,GAC9B,AAAA,IAAA,QAAA,GAAA,UACF,GAAO,GAAa,GACpB,GAAO,GAAa,GAEhB,MAAM,QAAQ,IAAM,EAAE,QAAU,GAAa,EAAE,QAAU,GACzD,EAAE,MAAM,SAAA,GAAK,MAAA,OAAO,KAAM,gBAWM,EAAwB,GAC5D,AAAI,MAAM,QAAQ,GAChB,GAAA,KAAK,OACD,EAAM,OAAS,EAAG,WAAM,MAAG,GAAI,qCACnC,EAAM,QACF,SAAC,EAAG,GAAM,MAAA,IAAsB,EAAG,WAAW,GAAI,GAAC,OAAO,MAE9D,EAAA,KAAK,OACD,OAAO,UAAU,IAAU,EAAQ,EACnC,WAAM,MAAA,YAAY,EAAI,sCACf,IAAuB,GAAM,mBAeL,GACrC,MAAI,KAAU,KACL,OACE,MAAM,QAAQ,GAChB,IAAM,EAAM,IAAI,SAAA,GAAK,MAAA,IAAuB,KAAI,KAAK,KAAO,IAC1D,MAAO,IAAU,SACnB,IAAI,EAAK,IAET,GAAG,cAcV,EAA8B,GAChC,GAAI,GAAW,EAAA,KAAK,MAChB,EACE,EAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,GAAM,GAAM,EAAA,KAAK,MACjB,MAAI,GAAM,EAAW,GAGrB,GAAW,EACX,EAAa,EAAC,MAAA,OAAI,IACX,GAET,MAAO,eASkC,GAEzC,MAAI,KAAmB,OACd,OAEL,IAAmB,SACd,SAEL,IAAmB,MACd,MAEF,KCvgBT,AAoBA,YAAqB,EAAW,GAC9B,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAI,EAAM,OAY1D,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAGE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GANyC,EAAA,cAAc,cA8BvD,GAAA,SAAA,GAA6B,EAAA,EAAA,GAQ3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAJQ,SAAA,gBAAkB,EAClB,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,YAAgB,EAAO,EAAG,EAAK,UAC/C,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,SAAU,KAAK,SAAU,KAAM,KAAK,OAtB9B,EAAA,UAAY,UAwB9B,GA1B6B,IA2B7B,EAAA,cAAc,cAAc,IAoB5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAK5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAFQ,SAAA,YAAc,EAG7B,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,KAAW,GAAY,EAAG,EAAK,WAG9D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OAdL,EAAA,UAAY,WAgB9B,GAlB8B,IAmB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAIE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAHF,EAAA,UAAY,SAK9B,GAP4B,IAQ5B,EAAA,cAAc,cAAc,IAoC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAY9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KANQ,SAAA,gBAAkB,EAClB,EAAA,gBAAkB,EAClB,EAAA,YAAc,EACd,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,YACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,IACZ,EAAA,IACI,EAAK,KAAM,EAAA,YAAgB,EAAO,EAAK,SAAU,EAAK,WAC1D,EAAA,IAAQ,EAAM,EAAK,KAAM,IAC7B,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,SAAU,KAAK,SACf,SAAU,KAAK,SACf,KAAM,KAAK,KACX,KAAM,KAAK,OApCC,EAAA,UAAY,aAuC9B,GAzCgC,IA0ChC,EAAA,cAAc,cAAc,IAQrB,GAAM,IACwC,CAC/C,QAAW,UACX,WAAc,aACd,OAAU,SACV,SAAY,wBAGkB,GAElC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA0C,GAC1C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCpQjC,YAyBwB,GACtB,MAAO,IAAI,IAAQ,eAQI,GACvB,MAAO,IAAI,IAAS,iBASpB,MAAO,IAAI,gBAIc,GACzB,MAAO,IAAI,IAAW,0ECjDxB,AAqBO,GAAM,IAA2B,CAAC,gBAAiB,gBAO7C,GAA4B,CAAC,QAAS,OAAQ,UAI9C,GAAyB,CAAC,MAAO,OAIjC,GAAkC,CAAC,MAAO,MAAO,SAAU,OCpCxE,AAmBA,GAAM,IAA+B,GAAI,iBAET,GAC9B,GAA0B,GAA0B,aAAc,eAGnC,GAC/B,GAA0B,GAA2B,cAAe,eAGxC,GAC5B,GAA0B,GAAwB,WAAY,GAGhE,GAAM,IAA4B,GAC5B,GAAoB,gBAKG,EAAc,GACzC,GAAgB,KAAK,GACrB,IACE,GAAM,GAAS,IACf,UAAgB,MACT,QACA,GACP,SAAgB,MACV,GAOV,cACE,MAAI,IAAgB,SAAW,EACtB,GAEA,GAAgB,KAAK,IAAqB,eASjB,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,MAAO,MAA2B,cAaA,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,AAAK,GAAQ,IAAI,IACf,GAAQ,IAAI,EAAY,GAE1B,GAAM,GAAQ,GAAQ,IAAI,GAG1B,GAFA,GAAQ,IAAI,EAAY,GAAQ,IAAI,GAAc,GAE9C,EAAQ,GACV,GAAM,GAAY,EAAU,IAAI,EAGhC,UAAQ,IAAI,EAAQ,GACb,MAEP,OAAO,GAIX,GAAM,IAAkB,GAAI,QAAO,+CAOD,GAChC,MAAO,CAAC,CAAC,EAAK,MAAM,ICjHtB,YA8B0B,GACxB,MAAO,KAAM,SAAS,EAAE,WAAY,gBAWlC,EAA4B,EAAgB,GAC9C,AAAI,GAAS,MACX,GAAQ,GAEV,AAAI,GAAO,MACT,GAAM,EAAM,QAId,OADI,GAAO,EACF,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,GAAQ,EAAM,GAEhB,MAAO,GAQT,YAAmB,GACjB,SAAQ,MAAM,QAAQ,GAAS,GAAI,cAAa,GAAS,EAClD,EAAA,SAAS,eAQE,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAQ1B,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAsDxB,EAAe,GACnC,GAAI,EAAM,EACR,KAAM,IAAI,GAAW,QAAQ,EAAG,cAAc,EAAK,mBAGrD,OADM,GAAgB,GACb,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,EAAI,KAAK,GAEX,MAAO,GChJT,YAwEqB,EAAW,GAC9B,MAAO,GAAE,OAAO,eASS,EAAW,GAAA,AAAA,IAAA,QAAA,GAAA,IACpC,GAAM,GAAW,EAAE,MAAM,QACzB,MAAI,GAAO,GACT,GAAO,EAAS,OAAS,EAAO,GAElC,EAAS,OAAO,EAAM,EAAG,GAClB,EAAE,QAAQ,eAcI,EAAW,GAChC,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,oDACA,SAAQ,EAAE,MAAM,OAAM,aAE5B,GAAM,GAAI,GAAW,EAAG,GACxB,MAAO,IAAK,EAAG,CAAC,EAAG,EAAG,kBASF,GACtB,GAAM,GAAW,CAAC,GAAqB,EAAE,QACzC,MAAO,GAAE,QAAQ,eAWU,GAC3B,GAAI,EAAE,MAAQ,EACZ,KAAM,IAAI,GACN,wDAAwD,EAAE,KAAI,KAEpE,GAAM,GAAW,CAAC,EAAE,MAAM,GAAI,GAAqB,EAAE,MAAO,IAC5D,MAAO,GAAE,QAAQ,eAYf,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,GAAI,CAAC,EAAM,EAAM,MAAM,SACnD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,GAC9B,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,EAAG,GACjC,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,MAAU,EAAmB,CAAC,EAAO,EAAG,EAAG,EAAG,GAAI,CACvD,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SAEjE,GACH,MAAO,GAAA,MAAU,EAAO,CAAC,EAAO,EAAG,EAAG,EAAG,EAAG,GAAI,CAC9C,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAClE,EAAM,MAAM,aAGd,KAAM,IAAI,GACN,8DACA,IAAG,EAAM,sBAcjB,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,GAAQ,CAAC,EAAM,MAAM,GAAI,QACjD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,GAC1B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,QAClC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAG,GAC7B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,YAErD,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAejB,EAAe,EAAe,EAAc,GAC9C,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,GAC9B,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,EAAG,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAO,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpD,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,YAGX,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAWO,EAAmB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC7C,GAAI,GACJ,MAAI,GAAO,GACT,GAAO,EAAQ,GAAG,KAClB,AAAI,IAAS,EACX,EAAO,EAEP,EAAO,GAGX,AAAI,IAAS,EAAQ,GAAG,MAGtB,GAAO,IAGF,EAAA,OAAW,EAAS,eAUQ,EAAW,GAC9C,OAAQ,EAAE,UACH,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,QACjC,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,WAEpD,KAAM,IAAI,GACN,kDACA,iBAAgB,EAAE,oBAWP,EAAW,GAI9B,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAI,CAAC,IAEH,EAAE,OAAS,EAAE,OACf,KAAM,IAAI,GACN,0BAA0B,EAAE,OAAM,oBAClC,yCAAwC,EAAE,KAAI,MAEpD,MAAO,GAAA,KAAS,EAAG,eAgBjB,EAAc,EAAY,EAAc,EACxC,GADc,MAAA,KAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAErB,EAAA,aAAiB,EAAO,EAAM,EAAQ,EAAO,eAsBlD,EAAW,EAAW,EACtB,GACF,GAAK,EAAE,KAAO,GAAO,EAAE,KAAO,EAC5B,KAAM,IAAI,IACN,2CACA,uBAAsB,EAAE,MAAK,kBAAkB,EAAE,QAEvD,GAAI,EAAE,MAAQ,GACZ,GAAM,GAAW,EAAE,MAAM,MAAM,IAAI,GAC7B,EAAiB,EAAE,MAAM,MAAM,IAAI,GACzC,GAAI,IAAa,EACf,KAAM,IAAI,IACN,2CACA,yDACI,EAAE,MAAK,SACX,eAAc,EAAE,QAIxB,GAAK,EAAE,OAAS,GAAO,EAAE,OAAS,GAChC,GAAM,GAAa,GACb,EAAa,GAInB,MAAO,GAAA,MAAU,OAAO,CACtB,EAAC,EACD,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,SAIZ,GAAM,GAAa,EAAE,MAAM,QACrB,EAAW,EAAW,MAC5B,EAAI,EAAE,QAAQ,CAAC,GAAI,IAInB,GAAM,GAAS,EAAE,MAAM,QACjB,EAAW,EAAO,MAClB,EAAiB,EAAO,MACxB,EAAiB,EAAM,OAAA,CAAE,IAGzB,EAAO,MAAM,KAAK,CAAC,OAAQ,EAAE,MAAO,SAAC,EAAG,GAC5C,MAAI,KAAM,EACD,EAAE,KAAO,EACP,GAAK,EAAE,KAAO,EAChB,EAAI,EAEN,IAET,EAAI,EAAE,UAAU,GAAM,QAAQ,CAAC,EAAgB,KAG/C,GAAM,GAAkB,EAAU,OAAK,GACjC,EAAa,GACb,EAAa,GACnB,MAAO,GAAA,MACF,OAAO,CACN,EAAC,EACD,EAAC,EACD,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,IAEX,QAAQ,gBAyDb,EAAmB,EAA4B,GACjD,MAAO,GAAA,KAAK,WACV,MAAI,OAAM,QAAQ,GAChB,EAAU,EAAA,SAAS,EAAS,SAE5B,EAAU,EAAQ,QAEb,EAAA,OAAW,EAAW,EAAS,iBASnB,GACrB,MAAO,GAAA,IAAQ,EAAG,GA+BpB,YAAqB,EAAe,EAAc,GAChD,GAAM,GAAY,EAAK,MAEvB,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,GACN,+BAA+B,EAAK,KACpC,6BAA4B,IAGlC,GAAI,IAAU,GACZ,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,EAAG,IAErC,EAAK,QACR,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,GAAI,EAAU,KAEzD,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAG,EAAU,KAEpC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,IAElC,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,KAE3D,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAU,KAEjC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,IAE/B,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,KAE7C,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAU,KAE9B,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,EAAQ,EACjB,MAAO,GAET,KAAM,IAAI,GAAW,sCAAsC,EAAK,kBAc9D,EAAW,EAAc,GAC3B,MAAO,GAAA,KAAK,WACV,MAAI,IAAc,MAChB,GAAa,MAEf,GAAgB,GAET,EAAE,IAAI,GAAY,EAAE,KAAM,EAAM,kBAUvB,EAAW,GAE7B,GAF6B,AAAA,IAAA,QAAA,GAAA,GAEzB,IAAU,EACZ,KAAM,IAAI,IACN,0CAA0C,EAAK,6BAGrD,MAAO,GAAA,IAAQ,eAWQ,GACvB,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,GAAG,IAAI,kBAc1C,EAAW,EAAe,EAAuB,GACnD,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAG,EAAO,EAAY,iBAY1B,GAC1B,MAAO,GAAA,KAAK,WACV,GAAM,GAAI,EAAA,IAAQ,GAAI,EAAA,IAAQ,GAAI,IAClC,MAAO,GAAA,YAAgB,EAAG,EAAG,iBAiBD,EAAY,EAAc,GAAA,MAAA,KAAA,QAAA,GAAA,IACjD,EAAW,IAAM,IC7sB1B,AAgBO,GAAM,IAAwB,CAAC,QAAS,SAAU,UAO5C,GACT,CAAC,SAAU,UAAW,mBCxB1B,YAoB6B,GAC3B,GAA0B,GAAuB,UAAW,eAG5B,GAChC,GAA0B,GAA2B,eAAgB,GASvE,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DACS,SAAA,UAAA,4BAAP,WACE,MAAO,IAUT,EAAA,UAAA,UAAA,WACE,MAAO,IAEX,GAf0C,EAAA,cAAc,cAiBxD,GAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,MAAM,EAAO,IAHf,EAAA,UAAY,QAKrB,GAP2B,IAQ3B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,KAAK,EAAO,IAHd,EAAA,UAAY,OAKrB,GAP0B,IAQ1B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,GACN,oDAAoD,GAE1D,GAAI,EAAK,QAAU,OACjB,KAAM,IAAI,GAAW,sCAAsC,GAE7D,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAI,EAAA,OAAO,EAAK,OAAQ,EAAA,KAAK,EAAO,OAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,QApBT,EAAA,UAAY,WAuBrB,GAzB8B,IA0B9B,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GASjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,eAAiB,KACjB,EAAA,eAAiB,IAOxB,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,cAAc,EAAO,KAAK,OAAQ,KAAK,OAAQ,IAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,OAAQ,KAAK,OAAQ,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAnBxD,EAAA,UAAY,gBAqBrB,GAvBmC,IAwBnC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAShC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,uCAAuC,EAAK,KAGlD,MAAO,IAAe,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGnE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,eA2BrB,GA7BkC,IA8BlC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,0CAA0C,EAAK,KAErD,MAAO,GAAA,gBAAgB,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGpE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,kBA2BrB,GA7BqC,IA8BrC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,IAG9C,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,SAAW,GAAK,EAAM,KAAO,EAAM,GAC3C,KAAM,IAAI,GACN,wEAGJ,MAAO,GAAA,IAAI,EAAK,KAAM,EAAA,IAAI,EAAM,QAKtC,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OApBd,EAAA,UAAY,WAsBrB,GAxB8B,IAyB9B,EAAA,cAAc,cAAc,IAU5B,YACI,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,gBAChB,GAAI,GACA,EAEJ,GADA,GAAgB,GACZ,EAAM,SAAW,EACnB,EAAQ,EAAM,GACd,EAAS,EAAM,WACN,CAAC,EAAG,EAAG,GAAG,QAAQ,EAAM,UAAY,IAC7C,GAAI,IAAe,iBACjB,GAAM,GAAqB,GAAU,EAAO,GAC5C,EAAQ,EAAM,GAAK,EACnB,EAAS,EAAM,GAAK,UACX,IAAe,gBACxB,GAAM,GAAqB,GAAU,EAAO,EAAG,EAAM,OAAS,GAC9D,EAAQ,EAAM,EAAM,OAAS,GAAK,EAClC,EAAS,EAAM,EAAM,OAAS,GAAK,QAGrC,GAAM,GAAY,GAAU,GAC5B,EAAQ,KAAK,KAAK,GAClB,EAAS,KAAK,KAAK,GAGrB,MAAO,CAAC,EAAO,GAiBjB,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAYnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,EAAK,MAAQ,EACf,KAAM,IAAI,GACN,wCAAwC,EAAK,OAEnD,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAM,EAAK,MAC7C,EAAK,KAAO,EAAK,MAAQ,KAAO,QAAU,EAAK,KAC/C,GAAa,EAAK,MAClB,EAAK,aACD,EAAK,cAAgB,KAAO,SAAW,EAAK,aAChD,GAAkB,EAAK,cACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAO,GAAY,GACnB,EAAQ,EAAK,GACb,EAAS,EAAK,GAChB,EAAQ,KAAK,MASjB,GARA,AAAI,KAAK,OAAS,QAChB,GAAS,KAAK,IAAI,EAAG,GAChB,AAAI,KAAK,OAAS,SACvB,GAAS,KAAK,IAAI,EAAG,GAErB,GAAS,KAAK,IAAI,EAAI,GAAQ,GAAU,GAGtC,KAAK,eAAiB,UACxB,GAAM,GAAS,KAAK,KAAK,GAEzB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACH,KAAK,eAAc,2BAA2B,EAAK,KAE5D,MAAO,GAAA,gBAAgB,EAAO,EAAG,EAAQ,EAAO,KAAK,WAErD,GAAM,GAAQ,KAAK,KAAK,EAAI,GAC5B,MAAO,GAAA,cAAc,EAAO,CAAC,EAAO,EAAO,KAI/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,MACZ,KAAM,KAAK,KACX,aAAc,KAAK,aACnB,KAAM,KAAK,OAzDR,EAAA,UAAY,kBA4DrB,GA9DqC,IA+DrC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,gBAwBrB,GA1BmC,IA2BnC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAWhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,eAwBrB,GA1BkC,IA2BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,WAiBrB,GAnB8B,IAoB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAI7B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,YAiBrB,GAnB+B,IAoB/B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAI/B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBiC,IAoBjC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAIhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBkC,IAoBlC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAO9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GATO,EAAA,aAAe,EAMtB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,KAAO,EAAK,KAEb,EAAK,MAAQ,KACf,KAAM,IAAI,IACN,2EAIR,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,IAAoB,8BAEhC,AAAI,EAAM,GAAK,EAAM,GAAK,KACxB,QAAQ,KACJ,gEACA,eAAc,EAAM,GAAK,EAAM,GAAE,gBACjC,wBAIN,GAAM,GACF,EAAM,GAAK,EAAM,GAAK,CAAC,EAAM,GAAI,EAAM,IAAM,EAC3C,EAAI,GAAe,EAAiB,EAAG,EAAG,WAC5C,EAAI,EAAA,OAAO,YAAY,GAC3B,MAAI,GAAM,GAAK,EAAM,IACnB,GAAI,EAAE,aAED,EAAA,IAAI,EAAK,KAAM,MAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CACL,KAAM,KAAK,KACX,KAAM,KAAK,OA3CR,EAAA,UAAY,aA8CrB,GAhDgC,IAiDhC,EAAA,cAAc,cAAc,IAUrB,GAAM,IACyC,CAChD,SAAY,WACZ,aAAgB,eAChB,cAAiB,gBACjB,SAAY,WACZ,UAAa,YACb,SAAY,WACZ,YAAe,cACf,aAAgB,eAChB,KAAQ,OACR,WAAc,aACd,aAAgB,eAChB,cAAiB,gBACjB,gBAAmB,kBACnB,gBAAmB,kBACnB,MAAS,SAGf,YACI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGgB,GAEnC,MAAO,IAAqB,eAGC,GAE7B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EAIJ,GAAI,IAAc,eAChB,MAAO,IAAI,IACN,GAAI,IAAc,gBACvB,MAAO,IAAI,IACN,GAAI,IAAc,WACvB,MAAO,IAAI,IACN,GAAI,IAAc,YACvB,MAAO,IAAI,IACN,GAAI,IAAc,cACvB,MAAO,IAAI,IACN,GAAI,IAAc,eACvB,MAAO,IAAI,IAEX,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAuB,OAE3B,OAAI,aAAsB,IACxB,EAEA,GAAuB,GCxoBlC,cAkBE,MAAO,IAAI,kBASX,MAAO,IAAI,gBAQY,GACvB,MAAO,IAAI,IAAS,eAYQ,GAC5B,MAAO,IAAI,IAAc,eASE,GAC3B,MAAO,IAAI,IAAa,eAaM,GAC9B,MAAO,IAAI,IAAgB,eASJ,GACvB,MAAO,IAAI,IAAS,eAgBU,GAC9B,MAAO,IAAI,IAAgB,eAgBC,GAC5B,MAAO,IAAI,IAAc,eAgBE,GAC3B,MAAO,IAAI,IAAa,eAeD,GACvB,MAAO,IAAI,IAAS,eAeI,GACxB,MAAO,IAAI,IAAU,eAgBK,GAC1B,MAAO,IAAI,IAAY,eAYI,GAC3B,MAAO,IAAI,IAAa,eAWC,GACzB,MAAO,IAAI,IAAW,kPChNxB,AAmBA,GAAI,IAAsB,gBAGxB,MAAO,MAGT,GAAM,IAA2C,eAO1B,GAAA,MAAA,KAAA,QAAA,GAAA,IACrB,AAAM,IAAU,KACd,IAAa,GAAU,GAEzB,GAAa,IAAW,EACjB,EAAS,GAAa,GAAQ,WCrCvC,YAoBgC,GAC9B,MAAO,OAAM,QAAQ,IAAM,MAAM,QAAQ,EAAE,gBASV,GACjC,MAAI,GAAE,SAAW,EACR,GAEJ,MAAM,QAAQ,EAAE,IAGd,EAFE,CAAC,eAWwB,GAClC,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAG,SAAW,EAChB,KAAM,IAAI,GAAW,uCAAuC,EAAG,QAEjE,EAAI,EAAG,OAEP,GAAI,EAEN,MAAO,eAa0B,GACjC,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,EAAO,KAChD,GAAI,EAAO,SAAW,EACpB,SAAS,EACF,EAAO,GAEd,KAAM,IAAI,GAAW,iCAAiC,EAAO,YAG/D,OAAO,GC9EX,YAmBqC,GAEnC,OADI,GAAQ,EACS,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,AAAI,EAAO,MAAM,SAAW,EAC1B,GAAS,EAET,GAAS,EAAO,MAAM,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAG/C,MAAO,GC5BT,AAoBA,GAAM,IAA+B,yBAqCnC,WACI,EAAa,EACb,EAAqC,EACrC,GAFa,AAAA,IAAA,QAAA,GAAA,WACb,AAAA,IAAA,QAAA,GAAA,IAAqC,AAAA,IAAA,QAAA,GAAA,IACrC,AAAA,IAAA,QAAA,GAAA,MACF,KAAK,MAAQ,GAAgB,UAC7B,KAAK,MAAQ,EAAI,MACjB,KAAK,GAAK,KAEV,EAAO,GAAe,GACtB,KAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,cAErC,KAAK,WAAa,EAClB,KAAK,WAAa,EAElB,KAAK,IAAM,EAAA,SAAa,EAAK,KAAK,WAAY,KAAK,KAAM,KAAK,OAUhE,SAAA,UAAA,KAAA,WACE,YAAK,oBACE,KAAK,KAUd,EAAA,UAAA,MAAA,SAAM,GAEJ,YAAK,oBACL,GAAiB,KAAK,IAAK,GAE3B,AAAI,KAAK,IAAI,KAAO,EAAO,IACzB,MAAK,IAAI,OAAO,GAChB,AAAI,KAAK,YAAc,MACrB,KAAK,IAAI,OAAO,KAAK,WAAW,MAAM,KAAK,OAGxC,MAMT,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,KAAK,IAAI,WAGD,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,IAAI,WACX,KAAM,IAAI,OAAM,kBAAkB,KAAK,KAAI,0BAI/C,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,WAAa,EAClB,KAAK,IAAI,UAAY,mCAEzB,KAEA,YAA0B,EAAa,GACrC,GAAI,EAAE,MAAM,aAAe,EAAE,MAAM,WACjC,KAAM,IAAI,OACN,mBAAqB,KAAK,UAAU,EAAE,OAAS,QAC/C,KAAK,UAAU,EAAE,oBA0LK,GAC5B,MAAO,GAAG,IAAI,SAAA,GAAK,MAAA,GAAE,qBAYnB,GACF,EAAmB,QAAQ,SAAA,GACzB,GAAM,GAA0B,EAAiB,GACjD,EAAS,MAAM,EAAiB,MClVpC,kBAuEE,WAAY,GACV,KAAK,MAAQ,EAAK,MAClB,KAAK,MAAQ,EAAK,MAKlB,AAAI,EAAK,OAAS,KAChB,KAAK,KAAO,EAAK,MAAM,OAEvB,KAAK,KAAO,EAAK,KAEnB,KAAK,QAAU,EAAK,QACpB,KAAK,QAAU,EAAK,QACpB,KAAK,KAAO,EAAK,MAAQ,GAE7B,MAAA,oBA6CE,WACa,EAA0B,EAC5B,EAA6B,EAC3B,EAAkB,EAClB,GAHA,KAAA,MAAA,EAA0B,KAAA,MAAA,EAC5B,KAAA,YAAA,EAA6B,KAAA,OAAA,EAC3B,KAAA,SAAA,EACA,KAAA,kBAAA,EACX,KAAK,GAAK,KACV,AAAI,GAAQ,MACV,MAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,eAEvC,KAAK,KAAO,EAAM,OAEtB,MAAA,MA2DI,GAAc,EAsBlB,GAAA,WAwCE,WACI,EAEO,GAAA,KAAA,SAAA,EACT,KAAK,GAAK,KAQV,KAAK,cAAgB,EAAK,cAU1B,KAAK,cAAgB,EAAK,cAE1B,KAAK,YAAc,EAAK,YAExB,KAAK,cAAgB,EAAK,cAQ1B,KAAK,aAAe,EAAK,aAEzB,KAAK,cAAgB,EAAK,cAM1B,KAAK,WAAa,EAAK,WAEvB,KAAK,YAAc,EAAK,YAKxB,KAAK,YAAc,EAAK,YAExB,KAAK,aAAe,EAAK,aAGzB,OAAoB,GAAA,EAAA,EAAA,EAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MACX,EAAM,cAAc,KAAK,MAG7B,EAAK,cAAc,aAAa,KAAK,MAGvC,SAAA,UAAA,UAAA,WAEE,OADM,GAAyB,GACX,EAAA,EAAA,EAAA,KAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,KACX,EAAa,KAAK,EAAM,MAExB,EAAa,KAAK,MAGtB,MAAO,CACL,cAAe,KAAK,cAAgB,KAAK,cAAc,KAAO,KAC9D,cAAe,EACf,YAAa,KAAK,YAClB,cAAe,KAAK,gBAG1B,KAkDI,GAAe,EAWnB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAmDlC,WAAY,GAAA,AAAA,IAAA,QAAA,GAAA,IAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAtBD,EAAA,UAAsB,KAEtB,EAAA,kBAA8B,GAQ5B,EAAA,UAAY,GAapB,EAAK,GAAK,KAEV,EAAK,oBAAsB,KAE3B,EAAK,UAAY,KACjB,EAAK,gBAAkB,GAGvB,EAAK,kBAAoB,GACzB,EAAK,qBAAuB,GAC5B,EAAK,QAAU,GACf,EAAK,SAAW,GAChB,EAAK,OAAS,GAMd,EAAK,aAAe,GACpB,EAAK,cAAgB,GAErB,GAAI,GAAO,EAAK,KAChB,GAAI,CAAC,GACH,GAAM,GAAS,EAAK,eACpB,EAAO,GAA0B,GAAU,IAAM,GAAO,GAM1D,GAJA,EAAK,KAAO,EAEZ,EAAK,WAAa,EAAK,WAAa,KAAO,GAAO,EAAK,UAEnD,EAAK,YAAc,MAAQ,EAAK,iBAAmB,MAKrD,GAAI,GAAe,OACnB,GAAI,EAAK,iBAAmB,KAC1B,EAAkB,EAAK,wBACd,EAAK,YAAc,MAC5B,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAkB,CAAC,GAAW,OAAO,EAAK,YAE5C,EAAK,gBAAkB,EAGvB,GAAI,GAAQ,EAAK,MACjB,AAAI,GAAS,MACX,GAAQ,EAAK,YAEf,AAAI,GAAS,MACX,GAAQ,WAEV,EAAK,MAAQ,EAGf,MAAI,GAAK,SAAW,KAClB,EAAK,eAAiB,EAAK,QAE3B,EAAK,eAAiB,KAKxB,EAAK,UAAY,KAEjB,EAAK,0BAA4B,KAYlB,SAAA,QAAjB,SAAyB,EAAc,GACrC,MAAO,GAAM,KAAO,OAAS,EAAU,YAUjC,EAAA,UAAA,eAAR,SAAuB,EAAmB,GACxC,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,mCACA,4BAA2B,EAAQ,MAEzC,GAAI,KAAK,aAAa,QAAU,EAC9B,KAAM,IAAI,GACN,gBAAgB,EAAQ,YAAY,EAAS,KAC7C,2BAA0B,KAAK,aAAa,OAAM,oBAExD,MAAO,MAAK,aAAa,IAY3B,EAAA,UAAA,WAAA,SAAW,GACT,MAAO,IACH,KAAK,eAAe,EAAW,SAAS,eAY9C,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,IACH,KAAK,eAAe,EAAW,UAAU,gBAgB/C,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,uHAIC,GAAI,KAAK,aAAa,SAAW,EACtC,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0CAEN,MAAO,IACH,KAAK,eAAe,EAAG,SAAS,+CActC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0BAEN,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,yHAKN,MAAO,IACH,KAAK,eAAe,EAAG,UAAU,gDAGvC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAQd,EAAA,UAAA,gBAAA,WAKE,MAAO,MAAK,OAAO,IAAI,SAAA,GAAU,MAAA,QAGnC,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,0CAGd,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,MAAO,MAAK,YAGd,SAAU,GACR,KAAK,OAAS,mCAGhB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,kBAAkB,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,IAClD,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAI,MAAK,WACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,GAAE,YAErC,QAIX,SAAqB,GACnB,KAAK,kBAAoB,mCAG3B,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAI,MAAK,UACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,CAAC,EAAE,YACxC,OAAO,KAAK,sBAEV,KAAK,kBAAkB,OAAO,KAAK,2BAI9C,SAAwB,GACtB,KAAK,qBAAuB,mCAO9B,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAG3C,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,2CAUd,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,SACR,KAAM,IAAI,OACN,yEAiBE,EAAA,UAAA,yBAAV,SAAmC,GAGjC,GADA,EAAS,GAAqB,GAC1B,KAAK,WAAa,MAAQ,KAAK,UAAU,SAAW,EACtD,OAEF,GAAM,GAAY,GAAqB,KAAK,WAC5C,GAAI,EAAO,SAAW,EAAU,OAC9B,KAAM,IAAI,GACN,SAAS,KAAK,KAAI,YAAY,EAAU,OAAM,YAC9C,oBAAmB,EAAO,OAAM,oBAChC,oBAAmB,IAEzB,OAAS,GAAa,EAAG,EAAa,EAAO,OAAQ,KACnD,GAAM,GAAI,EAAO,GACX,EAAkB,EAAU,GAClC,GAAI,GAAQ,KACV,SAIF,GAAM,GAAO,EAAE,KACf,GAAI,EAAK,MAAQ,MACX,IAAS,EAAK,KAChB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,KAC3D,kBAAiB,EAAK,KAAI,gBAAgB,IAGlD,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,IAG3D,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,EAAI,MAK/D,GAAI,EAAK,OAAS,MACZ,EAAE,QAAU,EAAK,MACnB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,IAC3D,qBAAoB,EAAK,MAAK,iBAAiB,EAAE,MAAK,MAK9D,GAAI,EAAK,MACP,GAAM,GAAS,EAAE,MACjB,OAAW,KAAO,GAAK,MACrB,GAAM,GAAO,OAAO,GACd,EAAQ,EAAK,KAAK,GAIlB,EACF,GAAQ,EAAI,EAAO,GAAQ,EAAO,EAAO,OAAS,GACtD,GAAI,GAAS,MAAQ,CAAC,EAAO,MAAM,QAAQ,KAAkB,GAC3D,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,mBAAmB,EAAI,uBACnC,eAAc,EAAK,kBAAkB,EAAM,OAMrD,GAAI,EAAK,OAAS,KAChB,OAAS,GAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAU,EAAK,MAAM,GACrB,EAAM,EAAE,MAAM,GACpB,GAAI,GAAW,MAAQ,GAAO,MACxB,IAAY,EACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,oBAAoB,EAAK,MAAK,MAC1C,gBAAe,EAAE,MAAK,SAgBtC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,IAGC,EAAA,UAAA,eAAV,SAAyB,EAAyB,GAChD,AAAI,KAAK,WAAa,MACpB,KAAK,UAAU,EAAQ,IAS3B,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,UAAY,GAOnB,EAAA,UAAA,cAAA,WACE,KAAK,UAAY,MAwEnB,EAAA,UAAA,MAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAGE,EAAS,GAAU,GAEnB,KAAK,oBAML,OAHM,GAAa,GAAqB,GAEpC,EAAiB,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,CAAE,aAAiB,MACrB,EAAiB,GACjB,OAIJ,OADI,GAAkB,GACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,YAAiB,KACnB,EAAkB,GAClB,OAIJ,GAAI,IAAmB,EACrB,KAAM,IAAI,GACN,mEAKN,MAAO,IAAU,KAAK,KAAM,WAE1B,GAAI,CAAC,EAAK,OAKR,EAAK,yBAAyB,GAI9B,OADM,GAAuB,GACT,EAAA,EAAA,EAAA,GAAqB,GAArB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,KAAK,EAAM,OAEzB,EAAK,MAAM,GAA+B,IAC1C,EAAK,MAAQ,GAGb,AAAI,EAAK,gBACP,EAAK,WAAW,EAAK,gBAGvB,AAAI,EAAK,YAAc,MAAQ,GAI7B,GAAK,UAAY,GAcrB,GANA,EAAK,yBAAyB,GAM1B,GAUF,OATI,GAAS,EAAK,KAAK,EAA6B,GAK9C,EAAuB,GAAqB,GAC5C,EAA2B,GAGnB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAT,GAAI,GAAC,EAAA,GACR,AAAI,EAAW,QAAQ,KAAO,IAC5B,GAAI,EAAE,SAER,EAAe,KAAK,GAItB,GAFA,EAAS,GAA+B,GAEpC,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAKN,MAAO,QAEP,GAAM,GAAa,GAAkB,GAC/B,EAAc,EAAK,mBAAmB,GACxC,EAAM,OACJ,EAAc,GAAiB,GAgCrC,GA/BA,EAAK,6BACD,MAAM,QAAQ,GAAU,EAAW,GACX,GAE5B,AAAI,GAAe,MAAQ,EAAY,OAAS,GAC5C,MAAM,QAAQ,EAAY,IAE5B,EAAU,EACI,IACG,SAAC,EAAO,GAAU,MAAA,IAAI,IAClB,EAAa,EAAO,EACpB,GAAqB,GAAS,EAAQ,EAAK,KAC3C,KAErB,EAAS,GAAI,IACT,EAAa,EAAsB,EACnC,GAAqB,GAAS,EAAQ,EAAK,MAUjD,EAAK,eACD,EAA6C,EAAQ,KAAM,KAC3D,EAAY,EAAa,GAC7B,EAAK,YAED,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAIN,MAAO,OAYH,EAAA,UAAA,6BAAV,SAAuC,GACrC,GAAI,KAAK,iBAAmB,KAC1B,OACK,GAAI,EAAW,SAAW,KAAK,gBAAgB,OACpD,QAAQ,KACJ,iDACG,MAAK,UAAU,GAAW,iCAC7B,qBAAoB,KAAK,UAAU,KAAK,iBAAgB,MACxD,iBAAgB,KAAK,YAEzB,GAAI,GAAc,GAClB,KAAK,gBAAgB,QAAQ,SAAC,EAAW,GACvC,AAAI,GAAa,MAAQ,EAAW,IAAM,MACtC,EAAW,KAAO,GACpB,GAAc,MAGlB,AAAI,GACF,QAAQ,KACJ,iCACA,KAAI,KAAK,UAAU,GAAW,eAC9B,mCAAkC,KAAK,KAAI,MAC3C,IAAG,KAAK,UAAU,KAAK,qBAiBjC,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,GAAI,KAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,EAC5D,KAAM,IAAI,IACN,aAAa,KAAK,KAAI,gEAI5B,OADM,GAA4B,GACf,EAAA,EAAA,EAAA,KAAK,aAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAc,KAAK,UAAU,EAAK,cACxC,AAAI,EAAgB,QAAQ,KAAiB,IAC3C,EAAgB,KAAK,GAGzB,GAAI,EAAgB,SAAW,GAC7B,GAAM,GAAe,KAAK,aAAa,GAAG,aAC1C,MAAI,OAAM,QAAQ,IAAiB,MAAM,QAAQ,EAAa,KAC1D,EAAa,SAAW,EAClB,EAAyB,GAE1B,MAIT,MAAM,IAAI,IACN,aAAa,KAAK,KAAI,+JAiB9B,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,sCAAsC,KAAK,KAAI,uFAIrD,MAAO,IAAoC,KAAK,UAclD,EAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAWf,EAAA,UAAA,WAAA,SAAW,GAAA,MAAA,KAAA,QAAA,GAAA,IACF,GAAc,EAAgB,KAAK,iBAAmB,KAAK,UAepE,EAAA,UAAA,WAAA,SAAW,GAAX,GAAA,GAAA,KACE,EAAA,KAAK,WACH,GAAM,GAAS,EAAK,QACpB,GAAI,EAAO,SAAW,EAAQ,OAK5B,KAAM,IAAI,GACN,4CAA4C,EAAK,KAAI,KACrD,iCAAgC,EAAQ,OAAM,MAC9C,gCAA+B,EAAO,OAAM,cAC5C,sBAAqB,EAAO,QAElC,GAAI,EAAO,SAAW,EACpB,OAIF,OAFM,GAAoD,GACpD,EAAc,GAAc,GACzB,EAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,GACxC,GAAM,GAAK,EAAY,GACjB,EAAI,EAAO,GACX,EAAI,EAAQ,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,GACN,sBAAsB,EAAG,MAAK,IAC9B,8CAA6C,EAAE,QAErD,EAAkB,KAAK,CAAC,EAAG,IAE7B,GAAc,MAmBR,EAAA,UAAA,UAAV,SACI,EAAc,EAAc,EAAkB,EAC9C,EAA2B,EAC3B,GAEF,GAAI,KAAK,kBAAkB,QAAQ,KAAU,GAC3C,KAAM,IAAI,GACN,yBAAyB,EAAI,cAAc,KAAK,MAEtD,KAAK,kBAAkB,KAAK,GAE5B,AAAI,GAAS,MACX,GAAQ,WAGV,AAAI,KAAK,2BACP,GAAc,GAAe,UAE/B,GAAM,GAAY,EAAY,MAAM,EAAO,GACrC,EACF,GAAI,IAAc,EAAW,EAAO,EAAM,EAAW,GACzD,SAAU,UAEV,AAAI,GAAe,MACjB,KAAK,QAAQ,WAAM,MAAA,GAAY,MAAM,EAAO,UAE9C,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,EACF,KAAK,kBAAkB,KAAK,GAE5B,KAAK,qBAAqB,KAAK,GAE1B,GAaT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,KAAK,0BAA4B,GAWnC,EAAA,UAAA,QAAA,SAAQ,SACN,GAAI,GAAU,MAAQ,MAAM,QAAQ,IAAW,EAAO,SAAW,EAC/D,OAGF,EAAS,GAAqB,GAC9B,AAAI,KAAK,UAAY,QAAa,KAAK,UAAY,MACjD,GAAA,KAAK,QAAO,KAAI,MAAA,EAAI,IAexB,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAYT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,GAAI,CAAC,KAAK,iBACR,GAAI,GAAQ,KACV,GAAI,MAAM,QAAQ,GAChB,EAAK,QAAQ,SAAA,GACX,GAAI,GAAe,KACjB,KAAM,IAAI,WACN,SAAS,EAAK,KAAI,iEAK1B,MAAM,IAAI,WACN,SAAS,KAAK,KAAI,2DAK1B,MAAO,MAIT,MAAO,IAeD,EAAA,UAAA,eAAR,SACI,EACA,EACA,EAA6B,EAC7B,EAA4B,EAC5B,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAM,GACF,GAAqB,GACzB,EAAgB,GAAqB,GACrC,EAAa,GAAqB,GAClC,EAAc,GAAqB,GACnC,EAAc,GAA+B,GAC7C,EAAe,GAA+B,GAM9C,OAHM,GAAyB,GACzB,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GAKV,EAAc,KAAK,EAAE,aACrB,EAAY,KAAK,EAAE,WACnB,EAAc,KAAK,EAAE,aAMvB,GAAI,IACA,CACE,cAAe,KACf,cAAa,EACb,YAAW,EACX,cAAa,EACb,aAAc,EACd,cAAa,EACb,WAAU,EACV,YAAW,EACX,YAAW,EACX,aAAY,GAEd,GAGJ,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,IAExC,EAAc,GAAG,YAAc,KAC/B,EAAc,GAAG,UAAY,KAAK,aAAa,OAAS,EACxD,EAAc,GAAG,YAAc,GAyBnC,EAAA,UAAA,UAAA,WACE,GAAM,GACyB,CAAC,KAAM,KAAK,KAAM,UAAW,KAAK,WACjE,MAAI,MAAK,iBAAmB,MAC1B,GAAO,gBAAqB,KAAK,iBAEnC,AAAI,KAAK,OAAS,MAChB,GAAO,MAAW,KAAK,OAElB,GAQC,EAAA,UAAA,eAAV,WACE,YAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,YAC/B,KAAK,QAAQ,QAGZ,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,UAAU,KAAK,KAAI,2BAkCvC,EAAA,UAAA,QAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,uCAIvC,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,sCAIvC,KAAK,oBAEL,GAAI,GAAuB,EAC3B,MAAI,EAAE,KAAK,YAAc,GACvB,GAAuB,KAAK,kBAGvB,CAAC,qBAAsB,KAAK,UAAW,qBAAoB,IAEtE,GAxoCoC,EAAA,cAAc,cAopClD,YAA2B,GAEzB,EACI,GAAqB,GAEzB,OADM,GAAkB,GACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAO,KAAK,EAAE,OAEhB,MAAO,IAA+B,GAYxC,YAA0B,GAExB,MAAO,sBAeL,EAAwB,EACxB,GAKF,GAJA,AAAI,IAAS,MAAS,GAAa,MAAQ,EAAY,IACrD,GAAQ,EAAO,YACf,EAAY,EAAO,WAEjB,EAAM,aAAa,SAAW,EAChC,MAAO,CAAC,GAER,GAAM,GAAO,EAAM,aAAa,GAChC,GAAI,EAAK,cAAc,SAAW,EAChC,MAAO,GAAK,aAGZ,OADM,GAAkC,GAC/B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,IAM7C,OALM,GAAI,EAAK,aAAa,GACtB,EAAQ,EAAK,cAAc,GAC3B,EAAY,EAAK,YAAY,GAC7B,EAAkB,GAAgB,EAAG,EAAO,GAElC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAc,QAAQ,KAAO,IAC/B,EAAc,KAAK,GAIzB,MAAO,GCvmDb,AA4CA,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CACJ,MAAO,EAAK,MACZ,KAAM,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAO,SAAS,cACtD,KAaF,GAXA,AAAI,EAAK,WAAa,MACpB,GAAK,UAAY,MAEnB,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,IAGhB,EAAK,UAAY,GACjB,EAAK,MAAQ,GACb,EAAK,OAAS,EAAK,OAEf,EAAK,YAAc,MAAQ,EAAK,iBAAmB,KACrD,KAAM,IAAI,GACN,qGAGN,GAAI,GAAkB,EAAK,gBAC3B,GAAI,GAAmB,MACrB,GAAI,EAAK,YAAc,KACrB,KAAM,IAAI,GACN,iFAGJ,EAAkB,CAAC,EAAK,WAAW,OAAO,EAAK,oBAI7C,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,yFAKR,GAAM,GAAQ,EAAK,OAAS,UAE5B,EAAK,gBAAkB,EACvB,EAAK,MAAQ,EAEb,EAAK,UAAY,CAAC,CAAC,MAAO,IAE1B,GAAM,GAAc,GAAI,IACpB,EAAK,MAAO,EAAK,gBAAiB,EAAM,GAAI,GAAI,EAAK,MACzD,SAAY,UAAY,EACxB,EAAY,YAAc,EAK1B,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,CAAC,GACf,cAAe,CAAC,GAChB,WAAY,CAAC,MACb,YAAa,CAAC,MACd,YAAa,CAAC,GACd,aAAc,CAAC,OAInB,SAAA,UAAA,MAAA,SACI,EACA,GACF,KAAM,IAAI,GACN,+BACA,kDAAiD,KAAK,QAG5D,EAAA,UAAA,QAAA,WAEE,MAAO,CAAC,qBAAsB,KAAK,UAAW,qBAAsB,IAGtE,EAAA,UAAA,UAAA,WACE,MAAO,CACL,gBAAiB,KAAK,gBACtB,MAAO,KAAK,MACZ,OAAQ,KAAK,OACb,KAAM,KAAK,OAzFC,EAAA,UAAY,aA4F9B,GA9FgC,IA+FhC,EAAA,cAAc,cAAc,gBAmCN,GACpB,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAC/C,KAAM,IAAI,OACN,gIAKN,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAE/C,KAAM,IAAI,GACN,oFAGN,GAAI,GAAa,EAAO,WACxB,AAAI,EAAO,OAAS,MAAQ,GAAc,MACxC,GAAa,CAAC,MAAM,OAAO,EAAO,QAGpC,GAAI,GAAQ,EAAO,MACnB,AAAI,GAAS,MACX,GAAQ,WAGV,GAAM,GAAa,GAAI,IAAW,CAChC,gBAAiB,EACjB,KAAM,EAAO,KACb,MAAK,EACL,OAAQ,EAAO,SAGX,EAAU,EAAW,aAAa,GAAG,cAC3C,MAAO,GAAQ,GC9MjB,YA0B2C,iHACzC,GAAI,GAAQ,KACV,MAAA,CAAA,GAEI,EAA+D,GAC/D,EAAiB,GACjB,EAA6B,GACnC,IAAW,IAAO,GACV,EAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACb,GAAc,EACpB,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAiB,KAAK,UAGtB,GAAS,OAAS,EACL,CAAA,EAAM,QAAQ,IAAI,IAD/B,CAAA,EAAA,UAEF,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAK,EAAK,IAAM,EAAO,GAAG,GAG5B,EAAA,QAAQ,+CAUyB,GACnC,GAAI,GAAQ,KACV,OAEF,OAAW,KAAO,IAChB,GAAM,GAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACnB,EAAM,WCjEZ,AAoBA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,QAAA,GAAA,YAFU,IAAA,IAAqB,KAM1B,GAAM,IAAyB,IA0BtC,GAAA,WAAA,aAEE,KAAA,eAAkC,KAMlC,SAAA,UAAA,UAAA,SAAU,GACR,KAAK,OAAS,GAGV,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,qFAEb,EAAA,UAAA,WAAN,SAAiB,qFASjB,EAAA,UAAA,SAAA,SAAS,KAGX,mBAqBE,WAAY,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,IAGtC,AAAI,GAAa,MACf,GAAY,IAEd,KAAK,UAAY,EACjB,KAAK,YAAc,EAGrB,SAAA,UAAA,OAAA,SAAO,GACL,KAAK,UAAU,KAAK,IAGtB,EAAA,UAAA,UAAA,SAAU,GACR,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,UAAU,KAIvB,EAAA,UAAA,SAAA,SAAS,GACP,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,SAAS,KAShB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,aAAN,SAAmB,uGACjB,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,KADO,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,WAAN,SAAiB,uGACf,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,KADS,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAI3B,KAOA,GAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,mBACE,GAAA,KAAA,OAAO,KAGH,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,OAAS,YAGV,EAAA,UAAA,WAAN,SAAiB,EAAe,yFAC9B,AAAI,GAAQ,MACV,GAAO,IAEH,EAAY,EAAK,MAAW,KAAO,EAAI,EAAK,KAClD,KAAK,MAAQ,aACF,GACT,GAAM,GAAQ,EAAK,GACnB,GAAI,MAAO,IAAU,SACnB,AAAK,EAAK,OAAO,eAAe,IAC9B,GAAK,OAAO,GAAO,GAErB,EAAK,OAAO,GAAO,EAAK,OAAO,GAAiB,EAAQ,OAExD,GAAI,GAAkB,OACtB,AAAI,IAAO,GAAK,OACd,EAAqB,EAAK,OAAO,GAEjC,EAAK,OAAO,GAAO,EAErB,GAAM,GACF,EAAA,KAAK,WAAM,MAAA,GAAA,IAAK,EAAK,OAAO,GAAO,EAAA,IAAI,EAAO,MAClD,EAAK,OAAO,GAAO,EACnB,AAAI,GAAsB,MACxB,EAAmB,mBAlBzB,IAAW,IAAO,KAAP,kBAwBP,EAAA,UAAA,WAAN,SAAiB,EAAe,2FAC9B,GAAI,GAAQ,KACV,eAAW,GACT,GAAI,EAAK,OAAO,IAAQ,sBAGxB,AAAI,MAAO,GAAK,OAAO,IAAS,SAC9B,EAAK,GAAO,EAAK,OAAO,GAAiB,EAAK,KAE9C,EAAA,KAAK,WACH,GAAM,GAAc,EAAA,IAAI,EAAA,IAAI,EAAG,EAAK,MAAO,EAAK,OAAO,IACvD,EAAK,GAAO,EACX,EAAK,OAAO,GAAgB,UAC7B,EAAA,KAAK,EAAK,cAXhB,EAAA,EAAkB,EAAA,KAAK,OAAO,QAAZ,EAAA,EAAA,OAAA,IAAP,EAAG,EAAA,KAAH,kBAiBjB,GA9DgC,mBAqEH,EAAA,EAAA,GAA7B,4DAIQ,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,MAAQ,GACb,KAAK,QAAU,YAGX,EAAA,UAAA,WAAN,SAAiB,EAAe,4EAC9B,AAAI,GAAQ,MACV,GAAO,IAET,KAAK,MAAM,KAAK,GAChB,IAAW,IAAO,GAChB,AAAI,KAAK,QAAQ,IAAQ,MACvB,MAAK,QAAQ,GAAO,IAEtB,KAAK,QAAQ,GAAK,KAAK,EAAK,mBAO1B,EAAA,UAAA,SAAN,6HACQ,EAA+D,GAC/D,EAAiB,GACjB,EAAoB,GAC1B,IAAW,IAAO,MAAK,QAErB,IADM,EAAa,KAAK,QAAQ,GACvB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,MAAO,GAAW,IAAO,UACrB,GAAc,EAAW,GAC/B,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAQ,KAAK,IAIJ,MAAA,CAAA,EAAM,QAAQ,IAAI,WACjC,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EAC7B,EAAkB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IACtD,EAAgB,UAChB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IAAM,EAAO,GAAG,mBAGpD,GA/C6B,mBA8DO,EAAA,EAAA,GAiBlC,WAAY,EAA0B,GAAtC,GAAA,GACE,EAAA,KAAA,OAAO,KAKP,GARM,EAAA,aAAe,EAIrB,EAAK,WAAa,GAAc,OAChC,AAAI,EAAK,aAAe,QACtB,GAAK,WAAa,IAEhB,EAAK,aAAe,SAAW,EAAK,SAAW,KACjD,KAAM,IAAI,OACN,mHAGN,MAAI,GAAA,KAAK,SAAS,EAAK,aAGrB,GAAK,UAAY,GACb,EAAK,UAAU,KAAK,GAAO,EAAK,aAEtC,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,MAAQ,EAAK,UAGd,SAAA,UAAA,UAAN,SAAgB,EAAe,EAAe,yGACtC,GAAgC,GAClC,KAAK,OAAS,KAChB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,MAAM,EAAO,EAAO,qBAEnC,SAAG,KAAK,EAAA,aACR,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAChC,MAAK,aAAe,EAChB,KAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,SACtB,EAAG,KAAK,EAAA,aAEV,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAC5B,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,QACtB,EAAG,KAAK,EAAA,aACH,AAAI,EAAA,KAAK,SAAS,KAAK,aAC5B,EAAG,KAAK,KAAK,UAAU,KAAK,aAAc,EAAO,IAEnD,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,mGACb,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,WAAtB,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,mGACX,MAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,SAAS,WAApB,EAAA,wCAGN,GA3GoC,gBAiHhC,EAEA,GAIF,GAHA,AAAI,GAAa,MACf,GAAY,IAEV,YAAqB,IACvB,MAAO,CAAC,GAEV,GAAI,MAAM,QAAQ,IAAc,EAAU,YAAc,IACtD,MAAO,GAGT,GAAM,GACF,GAAqB,GACzB,MAAO,GAAgB,IACnB,SAAA,GAAkB,MAAA,IAAI,IAAe,EAAgB,KAW3D,GAAA,IAAA,WAOE,cAcO,SAAA,4BAAP,SACI,EAAwB,GAC1B,EAAA,KAAK,OACD,GAAkB,GAAK,OAAO,UAAU,GACxC,WAAM,MAAA,sDACF,YAAW,KACnB,EAA4B,kBAAkB,GAC9C,AAAI,EAA4B,aAAa,IAAmB,MAC9D,GAA4B,aAAa,GAAkB,IAE7D,EAA4B,aAAa,GAAgB,KACrD,IAGS,EAAA,kBAAf,SAAiC,GAE/B,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAe,EAA4B,aAAa,CAAC,GAC/D,EAAa,QAAQ,SAAA,GACnB,GAAI,IAAS,EACX,KAAM,IAAI,GAAW,uCASZ,EAAA,MAAjB,WACE,EAA4B,aAAe,IAWtC,EAAA,gBAAP,SAAuB,GACrB,GAAM,GAA0C,GAChD,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAQ,CAAC,EACf,AAAI,GAAkB,GACpB,EAAa,KAAI,MAAjB,EAAqB,EAA4B,aAAa,IAGlE,MAAO,GAAa,IAAI,SAAA,GAAQ,MAAA,IAAI,MArEvB,EAAA,aAC6C,GAsE9D,iBAGI,EAA2B,EAAgC,EAC3D,EAAsB,EAAyB,EAC/C,EAAmB,EACnB,GACF,GAAM,GAAU,GAAI,IACd,EAAe,CACnB,GAAI,YAAiB,GAA4B,gBAAgB,IAEnE,AAAI,GAAa,MACf,EAAgB,KAAI,MAApB,EAAwB,GAE1B,EAAgB,KAAK,GACrB,GAAM,GAAe,GAAI,IAAa,GAMtC,SAAa,UAAU,CACrB,OAAM,EACN,aAAY,EACZ,QAAS,EACT,MAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,QAAS,IAEJ,CAAC,aAAY,EAAE,QAAO,GCnlB/B,YA2BI,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,QAAS,GChC9B,YAwB4B,EAAW,GACrC,MAAO,GAAA,KAAK,WACV,AAAI,EAAE,QAAU,WACd,GAAI,EAAE,OAAO,YAEf,GAAM,GAAY,EAAA,IAAQ,GAAS,GAAI,EAAM,IACvC,EAAgB,EAAA,KAAS,EAAU,MAAO,MAC1C,EAAO,EAAA,KAAS,EAAA,QAAY,EAAW,IAC7C,MAAO,GAAA,IAAQ,EAAG,iBAIW,EAAe,GAC9C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAO,IAAS,kBAG5B,EAAe,GAC/C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAO,IAAS,kBAIzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,EAAA,IAAQ,EAAO,GACtB,EACF,EAAA,YAAgB,EAAA,IAAQ,GAAQ,KAAW,OAAO,WAChD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAM,IACxC,MAAO,GAAA,IAAQ,IAAK,EAAA,KAAS,EAAW,mBAKxC,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAE9B,EAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAErC,MAAO,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAU,IAAa,kBAI/B,EAAe,GAC1C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,GAAS,GAAY,kBAInB,EAAe,GACnC,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,EAAW,kBAIE,EAAe,GAC9C,MAAO,GAAA,KAAK,WACV,GAAM,GAAM,EAAA,IAAQ,EAAA,IAAQ,EAAO,GAAQ,IACrC,EAAM,EAAA,IAAQ,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAQ,GAAQ,IACvD,MAAO,GAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAK,mBAY1B,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,KAAK,IAAI,GAChB,EAAiB,EAAA,IAAQ,EAAO,GAChC,EAAgB,EAAA,IAClB,EAAA,IAAQ,EAAgB,EAAA,SAAa,EAAA,IAAQ,GAAI,KACjD,GACJ,MAAO,GAAA,KAAS,EAAe,kBAK/B,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAI,EACF,EAAS,EAAA,QAAY,QAGrB,GAAM,GAAY,EAAA,IAAQ,EAAQ,EAAO,MAAM,OAAS,EAAG,IAC3D,EAAS,EAAA,IAAQ,EAAQ,GAE3B,SAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MACzC,EAAA,IAAQ,EAAA,IACX,EAAA,IAAQ,EAAO,UAAW,EAAA,IAAQ,IAAU,EAAO,MAAM,OAAS,kBActE,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAM,GAAa,EAAA,MAAU,GAAU,IAAS,QAChD,EAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MAChD,GAAM,GAAc,EAAO,MACrB,EACF,EAAA,OAAW,EAAY,EAAY,EAAY,OAAS,IACnD,QAAQ,GACjB,MAAO,IAAwB,EAAc,EAAQ,iBA0BrD,EAAgB,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAO,MAAO,EAAO,OACzC,KAAM,IAAI,GACN,8DACG,MAAK,UAAU,EAAO,OAAM,QAAQ,KAAK,UAAU,EAAO,SAEnE,MAAO,GAAA,KAAK,WAOV,GAAM,GAAa,EAAO,OACpB,EAAe,EAAO,MAAM,MAClC,MAAO,GAAW,IAAI,EAAO,IAAI,IAAS,IAAI,EAAa,MAAM,uBAIlC,EAAe,GAChD,MAAO,GAAA,KAAK,WACV,GAAI,GACJ,SAAI,EAAA,YAAgB,EAAO,KAAW,EAAI,MAC1C,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAG,KAC3B,EAAA,KAAS,GAA8B,EAAO,GAAI,kBAKzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,GAChD,EAAc,EAAA,YAAgB,EAAO,KAAW,GACtD,MAAO,GAAA,IACH,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAA,IAAQ,EAAa,KAAgB,kBAI5C,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,EAAA,IAAQ,EAAA,IAAQ,KAAW,IAC3C,MAAO,GAAA,KAAS,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAO,IAAW,kBAI7B,EAAe,GAC7C,MAAO,GAAA,KAAK,WACV,GAAM,GAAiB,GAAY,EAAO,IACpC,EAAiB,GAAY,EAAO,IACpC,EAAY,EAAA,IAAQ,EAAgB,GAC1C,MAAO,GAAA,IAAQ,EAAA,IAAQ,EAAW,OAkB/B,GAAM,IAAsD,CACjE,iBAAgB,GAChB,kBAAiB,GACjB,4BAA2B,GAC3B,4BAA2B,GAC3B,aAAY,GACZ,MAAK,GACL,iBAAgB,GAChB,QAAO,GACP,wBAAuB,GACvB,8BAA6B,GAC7B,mBAAkB,GAClB,0BAAyB,GACzB,QAAO,GACP,gBAAe,gBAKG,GAClB,GAAI,MAAO,IAAmB,UAC5B,GAAI,IAAkB,IACpB,MAAO,IAAU,GAEnB,GAAI,GAAS,gBAAgB,EAC7B,KAAI,GAAe,cAAc,SAAS,wBACxC,GAAS,gBAAgB,EAAc,wFAInC,GAAI,GAAW,OAErB,OAAO,GC7QX,YAyB+B,EAAe,GAC5C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,IAAQ,GAAI,EAAA,SAAa,IACrC,EAAmB,GAAO,EAAA,QAAY,EAAO,GAAY,EAAM,OACrE,MAAO,GAAA,KAAS,EAAA,MAAU,EAAO,GAAmB,kBAIpB,EAAe,GACjD,MAAO,GAAA,KACH,WAAM,MAAA,IACF,EAAA,MAAU,EAAA,OAAW,EAAO,IAAK,EAAA,OAAW,EAAO,KAAM,aAGnE,YAAuB,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,yBAI3C,EAAe,GACvC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIS,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIqB,EAAe,GAChD,MAAO,IAAuB,EAAO,eAInC,EAAe,GACjB,MAAI,GAAM,OAAS,EAAM,MACvB,GAAQ,EAAM,QAAQ,CAAC,EAAM,KAAO,KAEtC,EAAQ,EAAM,OAAO,IACrB,AAAI,EAAM,QAAU,EAAM,OACxB,GAAQ,EAAM,OAAO,EAAM,QAEtB,EAAA,MAAU,EAAO,GAAO,OAAO,WAajC,GAAM,IAAM,GACN,GAAM,GACN,GAAM,GACN,GAAM,GACN,GAAO,GACP,GAAO,GACP,GAA0B,GAC1B,GAAS,GACT,GAAgC,GAIhC,GAAuD,CAClE,eAAc,GACd,oBAAmB,GACnB,UAAS,GACT,wBAAuB,GACvB,8BAA6B,GAC7B,IAAG,GACH,IAAG,GACH,IAAG,GACH,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,OAAM,gBAGY,GAClB,GAAI,MAAO,IAAe,UAAY,IAAc,IAClD,MAAO,IAAW,GACb,GAAI,MAAO,IAAe,UAAY,GAAc,KACzD,MAAO,GAEP,KAAM,IAAI,GAAW,kBAAkB,eAqBP,GAElC,GADA,GAAY,IAAO,KAAM,0BAA0B,GAC/C,MAAO,IAAO,SAChB,MAAO,GAGP,OADI,GAAM,OACQ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAU,KAAS,GACrB,EAAS,EACT,OAGJ,GAAI,IAAW,OACb,MAAO,GAET,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAW,KAAS,GACtB,EAAS,EACT,OAGJ,MAAI,KAAW,OACN,EAED,EAAgB,KCzL5B,YAwB6B,GAC3B,GAAM,GAA2D,CAC/D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,MAC/B,SAAY,WAAM,MAAA,GAAA,MAAM,SAAS,EAAG,IAAM,OAC1C,KAAQ,WAAM,MAAA,GAAA,MAAM,KAAK,KAAO,GAAK,KAAO,OAC5C,OAAU,WAAM,MAAA,GAAA,MAAM,OAAO,KAAO,GAAK,KAAO,KAAW,IAC3D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,KAAO,GAAK,EAAG,OAC9C,IAAO,WAAM,MAAA,GAAA,MAAM,IAAI,OASzB,GAPA,EAAa,QAAa,EAAa,QACvC,EAAa,SAAc,EAAa,SACxC,EAAa,KAAU,EAAa,KACpC,EAAa,OAAY,EAAa,OACtC,EAAa,QAAa,EAAa,QACvC,EAAa,IAAS,EAAa,IAE/B,IAAc,GAChB,MAAO,GAAa,KAEtB,KAAM,IAAI,GAAW,qBAAqB,GC3C5C,AAeO,GAAM,IAA8C,EAAI,KAAO,iBAelE,EAAyB,EAAmB,GAC9C,GAD8C,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAuB,MACvB,MAAO,IAAwB,UAC/B,OAAO,eAAe,KAAyB,OAAO,WACtD,CAAC,GAAiB,GACpB,KAAM,IAAI,OACN,sEAGN,GAAI,GACF,GAAM,GAAM,KAAK,UAAU,GAC3B,AAAI,EAAI,OAAS,IACf,QAAQ,KACJ,mCAAmC,EAAS,qBAC5C,iBAAgB,EAAI,OAAM,iCAC1B,kHAEG,IAA2C,mBAgBvB,GAC/B,GAAI,IAAM,KAER,MAAO,GACF,GAAI,MAAO,IAAM,SACtB,GAAI,OAAO,eAAe,KAAO,OAAO,WAGtC,OADM,GAAO,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,MAAO,IAAQ,SAEjB,MAAO,GAET,GAAI,CAAC,GAAiB,EAAE,IACtB,MAAO,GAGX,MAAO,WAGH,MAAM,QAAQ,IAEhB,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,GAAiB,GACpB,MAAO,GAGX,MAAO,OAKP,OAAO,QAKX,GAAM,GAAQ,MAAO,GACrB,MAAO,KAAU,UAAY,IAAU,UAAY,IAAU,WCrGjE,YA8BI,EAAkB,EAAqB,EAEvC,GAAA,AAAA,IAAA,QAAA,GACI,QAAQ,KACd,GAAM,GAAiB,GAAsB,GAGvC,EAAsB,CAAC,eAAgB,eAAgB,WAC7D,AAAI,EACF,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAEtC,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAAM,IAI9C,AAAI,EAAU,EAAU,OAAS,IAAM,GAErC,GAAY,EAAU,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAa,MAGzD,GAAI,GACJ,GAAI,CAAC,GACH,EAAU,KAAK,mBACf,EAAgB,GAChB,OAAW,KAAS,GAAM,aACxB,EAAc,KAAI,MAAlB,EAAsB,EAAM,aAAa,IAI7C,EAAQ,IAAI,OAAO,IACnB,GAAS,EAAW,EAAW,GAC/B,EAAQ,IAAI,OAAO,IAGnB,OADM,GAAS,EAAM,OACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EACF,GAAkB,EAAO,GAAI,EAAW,GAExC,GACI,EAAO,GAAI,EAAW,EAAe,GAE3C,EAAS,KAAM,EAAO,OAAS,EAAI,IAAM,KAAK,OAAO,IAItD,EAAc,mCAEf,GAAM,GAAiB,GAAqB,GACtC,EAAoB,GAAqB,EAAM,qBAErD,EAAQ,iBAAiB,GAAiB,IAC1C,EAAQ,qBAAqB,GAC7B,EAAQ,yBAAyB,GACjC,EAAQ,IAAI,OAAO,IAGrB,YAA8B,GAC5B,GAAI,GAEJ,MAAK,GAAc,2BAA6B,KAC9C,EACI,GAAsB,EAAc,2BAExC,EAAiB,GAAqB,EAAM,kBAGvC,EAGT,YAA+B,GAC7B,GAAI,GAAiB,GACf,EAAyB,GACzB,EAAgB,GACtB,OAAW,KAAS,GAAM,aACxB,EAAa,KAAK,EAAM,aAAa,IAEvC,OAAyB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,EAAW,OAAS,GACpB,EAAW,SAAW,GAAK,EAAW,GAAG,cAAc,OAAS,GAClE,EAAiB,GACjB,MAEF,EAAM,KAAI,MAAV,EAAc,GAEhB,GAAI,EAEF,OAAoB,GAAA,EAAA,EAAA,EAAM,OAAN,EAAA,EAAA,OAAA,KAElB,OAFS,GAAK,EAAA,GACV,EAAO,GACQ,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAM,QAAQ,KAAU,GAC1B,GAAI,GACF,EAAiB,GACjB,UAEA,GAAO,GAIb,GAAI,CAAC,EACH,MAIN,MAAO,GAGT,YACI,EAAkB,EAElB,GAAA,AAAA,IAAA,QAAA,GAA6D,QAAQ,KAEvE,OADI,GAAO,GACF,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAI,GACN,GAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAAK,KAE1C,GAAQ,EAAO,GACf,EAAO,EAAK,MAAM,EAAG,EAAU,IAC/B,GAAQ,IAAI,OAAO,EAAU,GAAK,EAAK,QAEzC,EAAQ,GAQV,YACI,EAAc,EAEd,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAGhB,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EACF,CAAI,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,YAChE,GAAS,EAAQ,EAAW,GAM9B,YACI,EAAc,EAAqB,EAEnC,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAIhB,OADM,GAAwB,GACX,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,GAAiB,MAAQ,EAAc,OAAS,GAChD,EAAc,QAAQ,KAAU,GAClC,SAEF,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAAG,KACrC,EAAoB,EAAK,YAAY,GACrC,EAAqB,EAAK,cAAc,GAC9C,EAAY,KACL,EAAY,IAAI,EAAiB,KAAK,EAAkB,MAGnE,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EAAkB,EAAY,SAAW,EAAI,GAAK,EAAY,GAC9D,EAAmB,CACpB,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,WAC3D,GAGF,GAAS,EAAQ,EAAW,GAC5B,OAAS,GAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,EACxC,GAAS,CAAC,GAAI,GAAI,GAAI,EAAY,IAAK,EAAW,GCvNtD,AA8BA,YACI,EAAa,EAAe,GAC9B,MAAQ,KAAQ,gBAAkB,IAAQ,gBAClC,IAAQ,gBACZ,IAAU,GAAK,MAAO,IAAU,qBAUlC,EAA6B,GAC/B,GAAI,IAAmB,KACrB,MAAO,MACF,GAAI,MAAO,IAAmB,SACnC,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAmB,UAC1B,MAAO,IAAmB,UAC7B,MAAO,GACF,GAAI,YAA0B,QAGnC,OAFM,GAAU,GACV,EAAc,EAAe,OAC1B,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAe,GAC5B,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACf,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAArB,GAAM,GAAW,EAAA,GACd,EAAgB,EAAe,GACrC,GAAI,IAAgB,QAAU,MAAO,IAAkB,SAIrD,EAAO,GAAe,OAEtB,GAAM,GAAQ,GAA0B,GACxC,EAAO,GAAS,GAAoB,EAAe,IAGvD,MAAO,gBAWP,EAAyC,GAC3C,GAAI,GAAa,KACf,MAAO,MACF,GAAI,MAAO,IAAa,SAC7B,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAa,UAAc,MAAO,IAAa,UACzD,MAAO,GACF,GAAI,YAAoB,QAG7B,OAFM,GAAU,GACV,EAAc,EAAS,OACpB,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAS,GACtB,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACrB,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAU,EAAS,GACnB,EAAQ,GAA0B,GACxC,AAAK,KAAU,QAAU,IAAU,cAC/B,MAAO,IAAY,SAIrB,EAAO,GAAS,EAEhB,EAAO,GAAS,GAAoB,EAAS,GAGjD,MAAO,IC7HX,GAGM,IAAU,QCHhB,AA0BA,YAAiC,EAAqB,GAEpD,GAAI,EAAI,OAAS,MAAQ,EAAI,QAAU,EAAI,MAEzC,MAAO,GAET,IAEE,MAAO,GAAA,KAAK,EAAK,EAAI,aACd,GAEP,KAAM,IAAI,GACN,0BAA0B,EAAI,MAAK,kCACnC,gBAAe,EAAI,KAAI,MAAM,EAAI,MAAK,QAgB9C,GAAA,IAAA,WAUE,WAAY,GACV,GAVM,KAAA,SAAmC,GACnC,KAAA,QAAkC,GAClC,KAAA,QAAoC,GAQtC,YAAiB,GACnB,OAAW,KAAM,GAAM,SACrB,KAAK,SAAS,GAAM,EAAM,SAAS,GACnC,AAAI,IAAM,GAAM,SACd,MAAK,QAAQ,GAAM,EAAM,QAAQ,SAIrC,GAAI,GAAS,KACX,OAEF,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,KAAK,IAAI,EAAK,IAAK,EAAK,SAe9B,SAAA,UAAA,IAAA,SAAI,EAAqB,EAAe,GACtC,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAK,SAAS,EAAI,IAAM,GAAwB,EAAK,GACrD,KAAK,QAAQ,EAAI,MAAQ,EAAI,GAC7B,AAAI,GAAQ,MACV,MAAK,QAAQ,EAAI,IAAM,OAGzB,MAAM,IAAI,GAAW,uBAAuB,EAAI,KAAI,QAAQ,EAAI,IAElE,MAAO,OAQT,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,IAAI,EAAK,IAAK,EAAK,QAO1B,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,MAAK,SAAS,EAAI,KAAO,MAMlC,EAAA,UAAA,MAAA,WACE,MAAO,QAAO,KAAK,KAAK,UAU1B,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,SAAS,EAAI,SAG3B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,SAAS,KAWzB,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,QAAQ,EAAI,SAG1B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,QAAQ,KAKxB,EAAA,UAAA,aAAA,WACE,AAAI,KAAK,SAAW,MAClB,EAAA,QAAQ,KAAK,UAGnB,KAIM,GAA+D,GAG/D,GAC4D,eA4C9D,EAA0C,EAC1C,EAAiB,GAWnB,OATM,GAAoB,GAAU,KAAO,GAAQ,EAAO,SAEpD,EAAe,MAAM,QAAQ,GAC7B,EACF,EAAe,EAA8B,CAAC,GAE5C,EAAc,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,OACpC,EAAyB,GACzB,EAAY,EAAS,QACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,AAAI,EAAU,QAAQ,KAAgB,GACpC,EAAa,KAAK,EAAS,SAAS,IAEpC,EAAa,KAAK,MAItB,AAAI,GAAS,MAEX,GAAM,cAAgB,UACtB,EAAM,cAAgB,UAIxB,GAAM,GACF,EAAY,KAAK,KAAO,IAAM,EAAS,QAAQ,KAAK,KACpD,EACA,EACJ,GAAI,GAAa,IAAoB,MAGnC,GAAM,GAAM,GAAqC,EAAY,GAC7D,EAAS,EAAI,OACb,EAAkB,EAAI,gBAGtB,GAAa,GAAmB,EAChC,GAAsB,GAAmB,EAE3C,EAAS,GAAa,GACtB,EAAkB,GAClB,AAAK,GACH,OAAO,OAAO,EAAiB,GAAsB,IAMvD,OAHM,GAAmB,GAAI,IAAS,GAG7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAI,GAAS,MAEX,GAAM,GAAa,EAAA,SAAS,WAC5B,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAExB,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAI1B,GAAM,GAAW,EAAO,GAClB,EAAW,EAAS,YAC1B,GAAI,YAAoB,IACtB,SAOF,OALM,GAAwB,GACxB,EAAuB,GACvB,EAA6B,GAE/B,EAAa,GACG,EAAA,EAAA,EAAA,EAAS,OAAT,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAiB,SAAS,GAClC,EAAO,EAAiB,QAAQ,GACtC,EAAY,KAAK,GACjB,EAAW,KAAK,GAChB,AAAI,GAAQ,MACV,GAAa,IAEf,AAAK,GACH,GAAgB,EAAM,QACtB,AAAI,EAAgB,EAAM,QAAU,GAAK,CAAC,EAAS,OAAO,IACtD,EAAY,QAAQ,EAAM,QAAU,IAAM,CAAC,EAAM,YACjD,EAAM,YAAY,WAAa,IACjC,EAAiB,KAAK,IAK5B,AAAI,GACF,GAAS,GAAU,GACnB,EAAO,KAAU,EAAW,IAE9B,GAAM,GACF,GAAO,EAAS,MAAM,EAAa,IACnC,EAA8B,KAClC,AAAI,EAAS,iBACX,GAAa,EAAS,YAAY,EAAa,IAKjD,OAHM,GAAe,GAAe,GAC9B,EACF,MAAM,QAAQ,GAAgB,EAAe,CAAC,GACzC,EAAI,EAAG,EAAI,EAAsB,OAAQ,EAAE,GAClD,AAAK,EAAiB,OAAO,EAAsB,KACjD,EAAiB,IACb,EAAsB,GAAI,EAAc,GACxC,MAAM,QAAQ,GAAc,EAAW,GAAK,GAElD,GAAM,GAAQ,EAAY,QAAQ,EAAsB,GAAG,MAC3D,AAAI,IAAU,IACZ,GAAa,GAAS,EAAc,IAIxC,AAAK,GAEH,EAAA,QAAQ,GAQZ,SAAiB,eAEV,EAAe,EAAe,EAAa,GAsBpD,YACI,EAA2B,GAE7B,EAAA,KAAK,OACD,GAAW,MAAQ,EAAQ,OAAS,EACpC,WAAM,MAAA,0CAEV,GAAI,GAAgC,GAChC,EAAkC,GACtC,GAAI,EAAQ,SAAW,GAErB,GAAM,GACF,GAAgD,EAAQ,GAAI,GAChE,EAAc,EAAI,OAClB,EAAoB,EAAI,iBAGxB,QADM,GAAU,GAAI,KACA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAKlB,OALS,GAAK,EAAA,GACR,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,aAIc,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAxB,GAAM,GAAc,EAAA,GACvB,AAAK,EAAQ,IAAI,EAAe,OAC9B,GAAY,KAAK,GACjB,EAAQ,IAAI,EAAe,sBAKpB,GACT,AAAI,EAAkB,IAAS,MAC7B,GAAkB,GAAQ,GAAI,MAEhC,EAAa,GAAM,QACf,SAAA,GAAa,MAAA,GAAkB,GAAM,IAAI,MAL/C,OAAW,KAAQ,KAAR,GASf,MAAO,CACL,OAAQ,EACR,gBAAiB,GAAoB,IAIzC,YAA6B,GAC3B,GAAM,GAAmC,GACzC,OAAW,KAAQ,GACjB,EAAgB,GAAQ,EAAa,GAAM,KAE7C,MAAO,eAeL,EAAuB,GASzB,OAPM,GAAU,GAAI,KACd,EAA2B,GAC3B,EAA6B,GAKjB,EAAA,EAAA,EAAA,EAAS,QAAT,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAQ,IAAI,GAGd,GAAM,GAA0B,GAC1B,EAAkB,GAKxB,IAFA,EAAM,KAAK,GAEJ,EAAM,OAAS,IACpB,GAAM,GAAM,EAAM,EAAM,OAAS,GACjC,GAAI,EAAQ,IAAI,EAAI,OAClB,EAAM,MACN,SAEF,GAAM,GAAc,EAAM,EAAM,OAAS,KAAO,EAAM,OAAS,EAC/D,GAAI,EAAI,OAAO,SAAW,GAAK,EAE7B,EAAM,MACN,EAAO,KAAK,GACZ,EAAQ,IAAI,EAAI,MAChB,AAAI,GACF,EAAM,WAKR,EAAM,KAAK,EAAM,OAAS,GAC1B,OAAoB,GAAA,EAAA,EAAA,EAAI,OAAJ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GAQd,GALA,AAAI,EAAa,EAAM,OAAS,MAC9B,GAAa,EAAM,MAAQ,GAAI,MAEjC,EAAa,EAAM,MAAM,IAAI,EAAI,MAE7B,EAAQ,IAAI,EAAM,MACpB,SAEF,EAAM,KAAK,KAIjB,MAAO,CAAC,OAAM,EAAE,aAAY,GAS9B,YAAwB,GAEtB,GAAI,GACJ,GAAI,EAAM,YAAY,aAAa,SAAW,EAC5C,EAAe,EAAM,YAAY,YAGjC,OADI,GAAoB,KACf,EAAI,EAAG,EAAI,EAAM,YAAY,aAAa,OAAQ,EAAE,EAC3D,OAA2B,GAAA,EAAA,EAAA,EAAM,YAAY,aAAa,GAChD,cADiB,EAAA,EAAA,OAAA,KAAtB,GAAM,GAAY,EAAA,GAErB,GAAI,EAAa,KAAO,EAAM,IAC5B,EAAY,EACZ,OAIN,EAAe,EAAM,YAAY,YAAY,GAE/C,MAAO,GCjhBT,AA6CA,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAoCtC,WAAY,GAAZ,GAAA,GAEE,EAAA,KAAA,KAAM,KAAG,KAET,GAtBF,EAAA,eAAiB,GAAI,KAqBnB,EAAK,KAAO,EAAK,KACb,EAAK,MAAQ,MACf,GAAM,GAAS,EAAK,eAAe,cACnC,EAAK,KAAO,GAAO,GAqBrB,GAlBA,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAKlB,AAAI,MAAM,QAAQ,EAAK,QACrB,EAAK,OAAS,EAAK,OAAO,QAE1B,EAAK,OAAS,CAAC,EAAK,QAEtB,AAAI,MAAM,QAAQ,EAAK,SACrB,EAAK,QAAU,EAAK,QAAQ,QAE5B,EAAK,QAAU,CAAC,EAAK,SAInB,GAAqB,EAAK,QAAQ,SAAW,EAAK,OAAO,OAC3D,KAAM,IAAI,GACN,mGAEA,IAAG,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,SAIhC,AAAI,GAAqB,EAAK,SAAS,SAAW,EAAK,QAAQ,QAC7D,QAAQ,KACJ,qGAEA,IAAG,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,SAOjC,EAAK,YAAc,GACnB,EAAK,uBAAyB,GAC9B,EAAK,yBAA2B,GAKhC,EAAK,aAAe,GACpB,EAAK,wBAA0B,GAC/B,EAAK,0BAA4B,GAKjC,EAAK,OAAS,GAMd,EAAK,sBAAwB,GAe7B,OAAgB,GAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YACtB,EAAK,aAAa,KAAK,GACvB,EAAK,wBAAwB,KAAK,GAClC,EAAK,0BAA0B,KAAK,GAMtC,OAAgB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YAKtB,GAAqB,IAAc,EAAG,4BACtC,GAAqB,IAAgB,EAAG,8BACxC,EAAK,YAAY,KAAK,GACtB,EAAK,uBAAuB,KAAK,GACjC,EAAK,yBAAyB,KAAK,GAIrC,EAAK,WAAa,GAClB,EAAK,YAAc,GACnB,EAAK,gBAAkB,GACvB,EAAK,eAAiB,GACtB,EAAK,gBAAkB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,EAAK,YAAY,GAE/B,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,WACN,6DACA,qBAAoB,EAAK,OAAM,MAC/B,UAAS,EAAC,0BACV,oBAAmB,EAAM,eAAc,MAE7C,EAAK,WAAW,KAAK,EAAM,MAC3B,EAAK,gBAAgB,KAAK,EAAM,iBAEhC,EAAK,eAAe,KAAK,EAAM,MAEjC,OAAoB,GAAA,EAAA,EAAA,EAAK,aAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,YAAY,KAAK,EAAM,MAG9B,EAAK,oBAAsB,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QAClD,EAAK,qBAAuB,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,QAwFpD,OAjFM,GAA0C,GAE1C,EAAyC,GACzC,EAA4C,GAE5C,EAA6C,GAC7C,EAA4C,GAC5C,EAAiC,GAoBjC,EACF,SAAC,GAAwB,GAAuB,GAC/C,GAAe,GAAoB,IAClC,AAAI,KAAS,MAAQ,IAAa,MAAQ,IAAe,OACvD,IAAQ,GAAO,YACf,GAAY,GAAO,UACnB,GAAc,GAAO,aAEvB,GAAM,IAAO,GAAM,aAAa,IAGhC,GAAI,GAAgB,QAAQ,MAAU,GACpC,KAAM,IAAI,IACN,cAAc,GAAO,KAAI,cAAc,GAAM,KAAI,yBAKvD,GAAI,GAAc,QAAQ,MAAU,GAClC,OAIF,EAAK,eAAe,IAAI,EAAU,QAAQ,GAAO,KAGjD,AAAM,GAAM,KAAM,IAChB,GAAa,GAAM,IAAM,OAAO,KAAK,GAAc,QAGrD,AAAI,GAAgB,QAAQ,MAAU,IACpC,GAAgB,KAAK,IAKvB,OADM,IAAmB,GAAK,cAAc,OACnC,GAAI,EAAG,GAAI,GAAkB,MACpC,GAAM,IAAI,GAAK,aAAa,IACtB,GAAQ,GAAK,cAAc,IAC3B,GAAY,GAAK,YAAY,IAC7B,GAAc,GAAK,cAAc,IACvC,EACI,GAAG,GAAe,GAAiB,GAAO,GAC1C,IAGN,IADA,GAAc,KAAK,IACZ,GAAgB,QAAQ,KAAS,GACtC,GAAgB,OAAO,GAAgB,QAAQ,IAAO,GAExD,EAAuB,KAAK,KAG5B,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAgB,EAAG,EAAe,GAKpC,OAFM,GACF,EAAuB,QAAQ,UAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAa,EAAK,IAAM,EAExB,AAAM,EAAK,KAAM,IACf,GAAY,EAAK,IAAM,GAEzB,GAAI,GAAQ,EAAY,EAAK,IAGvB,EACD,EAAa,EAAK,cAAc,KAAO,KACnC,EACA,EAAa,EAAK,cAAc,IAOzC,EAAQ,KAAK,IAAI,EAAO,GACxB,EAAa,EAAK,cAAc,IAAM,EACtC,EAAe,EAAK,cAAc,IAAM,EAAK,cAC7C,EAAY,EAAK,IAAM,EAGvB,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAa,aAAa,GACxC,EACD,EAAY,EAAY,KAAO,KAAO,EACA,EAAY,EAAY,IACnE,EAAY,EAAY,IAAM,KAAK,IAAI,EAAQ,EAAG,GAClD,EAAa,EAAY,IAAM,GAKnC,GAAM,GAA0C,GAChD,OAAW,KAAU,IACnB,GAAM,GAAQ,EAAY,GAC1B,AAAM,IAAS,IACb,GAAa,GAAS,IAExB,EAAa,GAAO,KAAK,EAAa,IAIxC,GAAM,IAA4C,GAClD,OAAW,MAAW,IACpB,GAAM,GAAQ,EAAa,IAC3B,AAAM,IAAS,KACb,IAAc,GAAS,IAEzB,GAAc,GAAO,KAAK,EAAe,KAI3C,GAAI,IAAY,OAAO,KAAK,IACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAG1B,EAAK,OAAS,GACd,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACR,GAAiB,GAAc,GAGrC,GAAe,KAAK,SAAC,GAAG,IACtB,GAAM,IAAS,EAAa,GAAE,IACxB,GAAS,EAAa,GAAE,IAC9B,MAAI,IAAS,GACJ,GAEL,GAAS,GACJ,EAEF,IAET,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACd,AAAI,YAAiB,IACnB,EAAK,sBAAsB,KAAK,GAElC,EAAK,OAAO,KAAK,IAGrB,EAAK,cAAgB,GAGrB,GAAY,OAAO,KAAK,GACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAStB,OAJM,IAAoB,EAAK,OAAO,QAGhC,GAAoC,GACtB,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,KAClB,OADS,GAAK,GAAA,IACK,GAAA,EAAA,GAAA,EAAa,GAAb,GAAA,GAAA,OAAA,MAAd,GAAM,GAAI,GAAA,IACP,EAAQ,EAAK,cACnB,GAAI,GAAS,MACX,OAAgB,IAAA,EAAA,GAAA,EAAK,aAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAI,GAAkB,QAAQ,KAAO,GACnC,KAAM,IAAI,IACN,sDAAsD,EACtD,eAAc,EAAM,KAAI,OACxB,uDACA,WAAU,KAGlB,OAAgB,IAAA,EAAA,GAAA,EAAK,cAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAkB,KAAK,GAEzB,GAAwB,KAAK,EAAM,OAMzC,EAAK,aAAe,EAKpB,OADM,IAAW,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,mBAC7B,IACT,GAAM,IAAiB,GAAS,OAAO,SAAA,IAAK,MAAA,MAAM,KAAM,OACxD,GAAI,KAAmB,EACrB,KAAM,IAAI,IACN,aAAa,GAAI,aAAa,GAAc,uEAE5C,KAAK,UAAU,MANJ,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAd,GAAM,IAAI,GAAA,OAAJ,IAcX,SAAK,cAAgB,GAErB,EAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,EAAK,OACnB,cAAe,EAAK,QACpB,WAAY,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,QACjC,YAAa,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,QACnC,YAAa,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QACpC,aAAc,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,UAExC,EAAK,MAAQ,GACb,EAAK,UAAY,IAGT,SAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,cAAc,KAAK,KAAI,2BA8B3C,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,GAAM,GACc,CAAC,qBAAsB,KAAM,qBAAsB,GACvE,GAAI,EAAE,KAAK,YAAc,GACvB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAO,sBAAwB,EAAM,UAAU,qBAKjD,OAAwB,GAAA,EAAA,EAAA,KAAK,sBAAL,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAO,sBAAwB,EAAU,UAAU,sBAGvD,SAAO,qBAAuB,KAAK,UAC5B,GAGT,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,OAAO,QAAQ,SAAA,GAEhB,EAAc,kBACX,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,MAElC,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WAIE,GAAI,KAAK,kBAAkB,OAAS,EAClC,KAAM,IAAI,GACN,wNAMN,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADI,GAA2B,GACX,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAU,EAAQ,OAAO,EAAM,kBAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACb,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAQ,KAAI,MAAZ,EAAgB,EAAM,qBAExB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACtB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAiB,KAAI,MAArB,EAAyB,EAAM,kBAEjC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAkB3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,IAGnC,OAFM,GAAgD,GAClD,EAAoB,EACJ,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACO,EAAA,EAAA,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,EAAa,EAAO,eAAiB,KACvC,KAAM,IAAI,GAAW,0BAA0B,EAAO,cAExD,EAAa,EAAO,cAAgB,EACpC,IAIJ,GAAM,GAAoD,GAC1D,OAAW,KAAQ,IAIjB,GAAI,GAAgB,EACpB,GAAI,EAAa,IAAS,MACxB,GAAM,GAAS,EAAK,MAAM,KACpB,EACF,EAAO,MAAM,EAAG,IAAI,OAAO,CAAC,EAAO,EAAO,OAAS,KACvD,EAAgB,EAAiB,KAAK,KAExC,GAAI,EAAa,IAAkB,KACjC,EAAkB,KAAK,CAAC,EAAa,GAAgB,EAAQ,aACpD,EACT,KAAM,IAAI,GACN,gDAAgD,GAEtD,MAAO,GAAa,GAGtB,GAAI,GAEF,GAAM,GAAuB,GAC7B,OAAW,KAAQ,GACjB,EAAW,KAAK,GAElB,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,GACH,EAAW,OAAM,OAChB,EAAiB,yBACrB,IAAG,IAIX,GAAc,IAON,EAAA,UAAA,cAAV,WACE,GAAM,GAAY,KAAK,YACjB,EAAwC,GAC9C,SAAY,UAAe,KAAK,eAChC,EAAY,OAAY,EACxB,EAAY,aAAkB,eAAe,GAG7C,EAAY,QAAa,gBAClB,GAeT,EAAA,UAAA,OAAA,SAAO,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IACnB,GAAM,GAAc,GAAoB,KAAK,iBAC7C,MAAO,GAAe,KAAK,UAAU,GAAe,GAgBtD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAE9B,OADM,GAAW,GAAI,IACZ,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,EAAK,OAAO,GAAI,EAAO,IAEtC,MAAO,IAAQ,EAAK,QAAS,EAAU,MAa3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAC9B,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAA2B,KAAM,EAAO,QAEhD,EAAQ,GAAqB,GAGxB,EAAK,iBAAiB,EAAQ,GAAO,MAahD,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAM,GAAc,GAA+B,GACnD,GAAI,EAAY,SAAW,KAAK,YAAY,OAC1C,KAAM,IAAI,GACN,+BAA+B,EAAU,KACzC,cAAa,KAAK,YAAY,OAAM,oBAK1C,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,EAAY,OAAQ,KACtC,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAa,EAAY,GAGzB,EAAW,EAAM,KAAO,OAC9B,EAAqB,GAAY,EAGnC,GAAM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,GAAK,MAAA,UAAS,EAAG,MACrB,KAAK,IAE5B,GAAI,EAAU,OAAS,EACrB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GAEP,EAAQ,EAAK,cACnB,GAAI,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,QAAQ,EAAM,MAAQ,GAExD,SAIF,OADM,GAAuB,GACpB,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAc,EAAa,KAAI,IAAI,EAAS,IAAI,EAChD,EAAa,EAAqB,GACxC,EAAY,KAAK,GAQnB,OALM,GAAc,EAAM,mBACtB,GAA+B,IAE7B,EAAe,GAA+B,GAC9C,EAAY,EAAM,aAAa,QAAQ,GACpC,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAqB,GAAY,EAAa,IAStD,OAFM,GAAwB,GACxB,EAA4B,GACzB,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GACzC,EAAc,KAAK,0BAA0B,GAC7C,EAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAgB,KAAK,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAgB,OAAQ,KAC1C,GAAM,GAAM,EAAgB,GAC5B,GAAqB,IAAO,IAC5B,EAAa,KAAK,EAAqB,IAIzC,MAAO,IAA+B,IAa9B,EAAA,UAAA,iBAAV,SAA2B,EAAkB,GAE3C,AAAI,GAAS,MACX,GAAQ,GAA2B,KAAM,EAAO,SASlD,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,GACxC,GAAM,GAAI,KAAK,OAAO,GAChB,EAAI,EAAO,GACX,EAAO,EAAM,GACnB,EAAU,EAAE,IAAM,CAAC,EAAG,GAMxB,OAHM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAUjB,OAVS,GAAI,EAAA,GAEP,EAAQ,EAAK,cACb,EAAwB,EAAK,aAC7B,EAAyB,EAAK,cAK9B,EAAe,GAAI,OACT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAE,KAAM,IACV,EAAa,KAAK,EAAU,EAAE,KAGlC,GAAI,EAAa,SAAW,EAAsB,QAEhD,GAAI,GAAiB,GACjB,EAAe,OACf,EAAa,OACb,EAAa,OACb,EAAW,OAKf,GAHA,AAAI,EAAK,UAAY,MACnB,GAAS,EAAK,UAEZ,EAAa,SAAW,GACpB,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GACvB,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAgB,IACpD,EAAc,GACV,EAAM,YAAY,EAAgB,IACtC,EAAkB,CAAC,GACnB,EAAgB,CAAC,OAEjB,GAAkB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KAC1C,EAAgB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KACxC,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAiB,IACrD,EAAc,GACV,EAAM,YAAY,EAAiB,IAGzC,GAAI,EAAM,oBACR,KAAM,IAAI,IACN,yHAMN,OAAS,GAAI,EAAG,EAAI,EAAuB,OAAQ,EAAE,GACnD,GAAM,GAAI,EAAuB,GAC3B,EAAI,EAAc,GAClB,EAAO,EAAY,GACzB,EAAU,EAAE,IAAM,CAAC,EAAG,KAS9B,OAHM,GAA0B,GAC1B,EAAwB,GACxB,EAAwB,GACd,EAAA,EAAA,EAAA,KAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,GACI,EAAE,KAAM,GAAW,4BAA4B,EAAE,KAAI,MAAM,EAAE,IAC3D,GAAA,GAAA,EAAA,EAAA,IAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAa,KAAK,EAAO,OACzB,EAAc,KAAK,GACnB,EAAY,KAAK,GAInB,MAAO,CAAC,EAAe,EAAa,IAW9B,EAAA,UAAA,uBAAR,SAA+B,GAG7B,OAFM,GAAiD,GACnD,EACgB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,YAAiB,GAAY,EAAI,EAC7C,OAAS,GAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,KAAK,eAAe,IAAI,IAE1B,GAAkB,GAAW,EAC7B,GAAa,IAInB,MAAO,IAsBT,EAAA,UAAA,SAAA,SAAS,EAAe,GACtB,GAAI,GAAS,MACX,GAAI,KAAK,OAAO,QAAU,EACxB,KAAM,IAAI,GACN,wCAAwC,EAAK,oBAC7C,QAAO,KAAK,OAAO,OAAM,eAE7B,MAAO,MAAK,OAAO,WAGjB,GAAQ,KACV,KAAM,IAAI,GAAW,8CAIzB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,OAAS,EACjB,MAAO,GAGX,KAAM,IAAI,GAAW,kBAAkB,IAQzC,EAAA,UAAA,gBAAA,WAAA,GAAA,GAAA,KAKE,MAAO,GAAA,KAAK,WAEV,OADM,GAAmB,GACL,EAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACL,EAAY,EAAG,EAAY,EAAM,aAAa,OAClD,EAAE,GACL,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,EAAK,eAAe,IAAI,IAC1B,EAAO,KAAI,MAAX,EAAe,EAAM,mBAK3B,MAAO,MAIX,EAAA,UAAA,UAAA,WAWE,OAVM,GAAmC,CAAC,KAAM,KAAK,MAK/C,EACF,KAAK,uBAAuB,KAAK,QAG/B,EAAe,GACD,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EAAiB,EAAM,eACvB,EAAc,EAAM,YACpB,EAAuB,GACpB,EAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAO,EAAM,aAAa,GAC1B,EAAU,EAAU,QAAQ,EAAO,GACrC,EAAS,GACb,GAAI,KAAK,eAAe,IAAI,IAG1B,GAAI,EAAK,SACP,IACE,KAAK,UAAU,EAAK,UACpB,EAAS,EAAK,eACP,GACP,QAAQ,KACJ,SAAS,EAAM,KAAI,mDAEhB,GAAK,SAAQ,gCAChB,+EAEJ,EAAS,GAGb,GAAI,EAAK,cAAc,OAAS,GAE9B,OADM,GAAW,GACR,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAU,EAAU,QAAQ,EAAc,GAC5C,EAAe,EAAkB,GACrC,AAAI,GAAgB,MAClB,GAAe,GAEjB,EAAS,KACL,CAAC,EAAa,KAAM,EAAc,EAAa,IAErD,EAAqB,KAAK,KAIhC,GAAM,GAAiC,GACvC,EAAK,KAAU,EAAM,KACrB,EAAK,UAAe,EACpB,EAAK,OAAY,EACjB,EAAK,aAAkB,EACvB,EAAa,KAAK,GAEpB,EAAO,OAAY,EAGnB,OADM,GAAc,GACX,EAAI,EAAG,EAAI,KAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAY,KAAK,uBAAuB,GAExC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,yBAAyB,GAClD,EAAY,KAAK,CAAC,EAAM,KAAM,EAAc,IAE9C,EAAO,YAAiB,EAGxB,OADM,GAAe,GACZ,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GAEzC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,0BAA0B,GACnD,EAAa,KAAK,CAAC,EAAM,KAAM,EAAc,IAE/C,SAAO,aAAkB,EAClB,GAgBF,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAA8C,GAO9C,EAAkE,GACxE,WACI,EAAc,GAChB,AAAM,EAAM,OAAQ,GAGlB,EAAiB,EAAM,MAAM,KAAK,GAFlC,EAAiB,EAAM,MAAQ,CAAC,GAMpC,WAAqB,EAAc,GAGjC,OAFM,GAAiC,GACnC,GACoB,GAAA,EAAA,GAAA,EAAA,GAAA,GAAA,OAAA,MAAnB,GAAM,IAAS,GAAA,IACZ,GAAmB,GAAU,GAC7B,GAAmB,GAAU,GAC7B,GAAqB,GAAU,GAKrC,GAHA,GAAS,GAAU,IAAM,KACrB,GACA,GAAU,GACV,CAAE,MAAoB,KACxB,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAe,EAAc,IACnC,GAAI,GAAa,aAAa,QAAU,IACtC,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAc,GAAa,aAAa,IAC9C,EAAa,KAAK,GAAY,cAAc,KAK9C,AAAI,EAAa,OAAS,GACxB,EAAM,MACF,GAA+B,GAC/B,IAUR,WAAsB,GACpB,GAAM,GAAY,EAAU,KAEtB,EACF,GACI,EACA,EAAO,eAAoB,KACvB,EAAO,cACP,IACZ,EAAM,6BAA6B,GACnC,EAAc,GAAa,EAE3B,GAAM,IACF,EAAU,aACd,GAAiB,QAAQ,SAAA,IACvB,GAAI,CAAE,cAAoB,QACxB,KAAM,IAAI,GACN,yDACI,IAMV,EAAmB,EAAO,MAO9B,OAFM,GAAO,EAAO,KACd,EAAmB,EAAO,OACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAa,GAOf,KAAO,CAAC,GAA4B,IAClC,OAAwB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAQ,EAAc,EAAU,MACtC,GAAI,EAAM,OAAQ,IAChB,GAAM,GAAkC,EAAiB,EAAM,MAC/D,MAAO,GAAiB,EAAM,MAC9B,OAAuB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAY,EAAO,KAU3B,OAJM,GAAiC,GACjC,EAAkC,GAClC,EACF,EAAO,YACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAa,KAAK,EAAmB,IAIvC,OAFM,GACF,EAAO,aACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAc,KAAK,EAAmB,IAExC,MAAO,IAAI,GAAI,CAAC,OAAQ,EAAc,QAAS,EAAe,KAAI,KASpE,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WAGE,GAAI,KAAK,UACP,KAAM,IAAI,GACN,wLAIN,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,SACR,MAAO,GAGX,MAAO,oCAST,EAAA,UAAA,YAAA,WAAA,GAAA,GAAA,KACE,EAAA,KAAK,WACH,EAAK,OAAO,QAAQ,SAAA,GAElB,AAAI,EAAM,UACR,EAAM,mBAMhB,GA3vCwC,IC7CxC,AAiCA,YACI,EAAmD,EACnD,GACF,GAAM,GAAa,EAAY,OAC/B,GAAI,GAAW,MAAS,MAAM,QAAQ,IAAY,EAAQ,SAAW,EACnE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,QAEjC,GAAI,IAAe,EACjB,MAAI,OAAM,QAAQ,IAAY,EAAQ,SAAW,EACxC,EACE,MAAO,IAAY,UAAY,EAAY,IAAM,GACnD,CAAE,EAA2B,EAAY,KAEzC,CAAC,GAGZ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAQ,SAAW,EACrB,KAAM,IAAI,OACN,YAAY,EAAU,mBAAmB,EAAQ,OAAM,IACvD,kCAAiC,EAAU,cAC3C,iEAEN,MAAO,WAEL,MAAO,IAAY,UAAY,OAAO,KAAK,GAAS,OAAS,GAC7D,MAAQ,GAA2B,OAAO,KAAK,GAAS,KACpD,UACN,GAAM,GAAwB,GAC9B,SAAY,QAAQ,SAAA,GAClB,AAAI,IAAc,GAChB,EAAO,KAAM,EAA2B,IAExC,EAAO,KAAK,QAGT,MAEP,MAAM,IAAI,OACN,2BAA2B,EAAU,cACrC,OAAM,EAAU,kCACb,GAAU,+BAA+B,EAAW,WACvD,aAAY,EAAU,oBAAoB,KAAK,UAAU,iBAkB7D,EACA,GACF,MAAO,IACH,EAAa,EAAa,2BA6B5B,EAAW,EAAuB,EAClC,2GACF,GAAI,GAAgB,MAAQ,GAAoB,KAG9C,KAAM,IAAI,OAAM,qDAGd,IAAe,KAEX,GAAqB,EAAA,KAAK,WAC9B,GAAI,EAAE,MAAM,SAAW,EAErB,MAAO,GAAE,QACJ,GAAI,EAAE,MAAM,SAAW,EAC5B,GAAI,EAAE,MAAM,GAAK,GAEf,GAAM,GAAO,EACb,MAAO,GAAE,OAAO,QACX,GAAI,EAAE,MAAM,KAAO,EAExB,MAAO,GAAE,QAAQ,CAAC,EAAE,MAAM,KAE1B,KAAM,IAAI,OACN,+CAA+C,EAAE,MAAM,GAAE,4EAK/D,MAAM,IAAI,OACN,yCAAyC,EAAE,KAAI,4EAKjC,EAAA,GAAA,OAAM,KAAK,CAAA,EAAM,EAAS,SA3B9C,CAAA,EAAA,UA2BI,SAAgB,EAAA,MAAA,EAAA,CAAW,EAAA,SACjC,EAAA,QAAQ,GACF,EAA8B,GACpC,EAAc,QAAQ,SAAA,GACpB,GAAI,EAAY,IAAe,KAC7B,KAAM,IAAI,OACN,8DACA,cAAa,EAAU,mCACvB,eAEJ,EAAkB,KAAK,EAAY,MAIvC,CAAA,EAAO,EAAA,SAAS,EAAmB,mBAEnC,MAAA,CAAA,EAAO,uBAWyB,EAAgB,GAClD,MAAO,GAAA,IAAI,EAAQ,GC5LrB,AAyLA,GAAM,IAAgC,GAgBtC,YAII,EAAY,GACd,GAAI,GACA,EAEE,EAAiB,EACvB,EAAK,EAAe,GACpB,EAAK,EAAe,GACpB,EAAA,KAAS,OACL,GAAM,MAAQ,GAAM,KACpB,WAAM,MAAA,+OAIF,IAAG,KAEX,GAAM,GACF,GAA0B,QAAS,EAAM,WAAY,GACnD,EACF,GAA0B,SAAU,EAAM,YAAa,GAErD,EAAoB,EAAY,GAAG,MAAM,GAE/C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,OAAO,OACpC,WAAM,MAAA,mBAAmB,EAAM,OAAO,OAAM,4BACxC,aAAY,EAAY,OAAM,oCAC3B,MAAK,UAAU,EAAM,YAAW,OAE3C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,QAAQ,OACrC,WACI,MAAA,mBAAmB,EAAM,QAAQ,OAAM,6BACvC,aAAY,EAAY,OAAM,sCAC3B,MAAK,UAAU,EAAM,aAAY,OAE5C,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,8BACC,GAAM,WAAW,GAAO,QACrB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,+BACC,GAAM,YAAY,GAAO,QACtB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,MAAO,CAAC,GAAI,EAAa,GAAI,GAG/B,YACI,EAAuB,EAAiB,GAC1C,GAAI,YAAkB,GAAA,OACpB,MAAO,CAAC,GACH,GAAI,MAAM,QAAQ,GACvB,SAAA,KAAS,OACL,EAAO,SAAW,EAAM,OACxB,WAAM,MAAA,wBAAwB,EAAO,OAAM,0BACvC,EAAM,OAAM,iBAAiB,EAAa,SAAS,EAAK,MACzD,EAIP,OAFM,GAAuB,GAEV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,IAAS,KAClB,KAAM,IAAI,GACN,gEACG,GAAa,SAAS,EAAI,OAEnC,EAAO,KAAK,EAAO,IAErB,MAAO,GAIX,YACI,GAMF,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,IACN,0DAEN,MAAO,CAAC,GAAI,EAAK,GAAI,GAAI,EAAK,gBAO5B,EAAY,EACZ,yKA0BF,GAzBM,EAAqB,EAAK,iBAAmB,KACnD,EAAA,KAAS,OACL,EAAM,WAAa,KACnB,WAAM,MAAA,mGAGV,EAAA,KAAS,OACL,GAAQ,KACR,WAAM,MAAA,kGAEV,EAAA,KAAS,OACL,EAAK,QAAU,MAAQ,EAAK,OAAS,GAAK,OAAO,UAAU,EAAK,QAChE,WAAM,MAAA,gEACF,qBAAoB,EAAK,UACjC,EAAA,KAAS,OACL,CAAC,GACI,EAAK,gBAAkB,GAAK,OAAO,UAAU,EAAK,iBACvD,WAAM,MAAA,gEACF,2CAA0C,EAAK,mBACvD,EAAA,KAAS,OAEJ,EAAa,iBAAsB,KACpC,WAAM,MAAA,sFAGN,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,mDAGX,EAAe,EAAK,gBAAkB,KACxC,EAAK,OACL,EAAK,OACT,AAAI,GACF,CAAI,GAAgB,EAAK,gBACvB,EAAA,KAAS,OACL,EAAK,mBAAqB,MACrB,EAAK,kBAAoB,GACzB,OAAO,UAAU,EAAK,mBAC3B,WAAM,MAAA,yIAGF,YAAW,EAAK,qBAElB,GAAiB,GACnB,EAAK,gBAMT,EAAQ,EAAe,GACvB,EAAQ,EAAe,KAIrB,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAe,OACnB,AAAI,EACF,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,IAAK,MAAA,OAAS,MAEzD,EAAkB,EAAU,QAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YACtD,EAAU,EAAK,SAAW,KAAO,EAAI,EAAK,QAC1C,EAA0B,GAC5B,EAAW,EAAS,EAAK,OAAQ,KAAM,KACvC,GAAiB,EAAS,GAC1B,KACA,EAAc,GAJX,EAAY,EAAA,aAAE,EAAA,EAAA,QAKrB,EAAa,SAAS,GACtB,EAAM,QAAU,EAEhB,CAAA,EAAM,EAAa,uBAAnB,SAAA,OACA,EAAM,cAAgB,GAClB,EAAQ,EAAK,cAAgB,KAAO,EAAI,EAAK,aAE9B,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BACZ,GAAQ,EAAK,OACZ,GAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,KAFR,CAAA,EAAA,iBAExB,GAAA,OACI,EAAY,EACZ,EAAa,EACZ,EAAD,CAAA,EAAA,GACa,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BAEV,GAAqB,EAAY,EAAK,gBAAkB,IACzC,CAAA,EAAM,EAAa,QAD0B,CAAA,EAAA,WAKjE,MAJM,GAAc,EAAA,OAIhB,GAAsB,EAAY,KACpC,SAAQ,KACJ,qCACG,GAAK,gBAAe,MACvB,mDACG,GAAS,cACZ,8HAGG,GAAK,gBAAkB,EAAK,OAAM,eACrC,yEAEJ,CAAA,EAAA,KAGE,EAAY,OAAS,KACjB,GACF,GAA8B,EAAO,EAAY,OAD9C,EAAE,EAAA,GAAE,EAAE,EAAA,GAEP,EAA4B,GAClC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAG,GAAG,MAAM,GAEhC,CAAA,EAAM,EAAa,aAAa,EAAY,KAP1C,CAAA,EAAA,cAOF,EAAA,OAEM,EAA8B,IAChC,GAAK,aAAe,MAApB,MAAA,CAAA,EAAA,IACI,EACF,GAAwB,EAAK,YAAa,EAAM,aAC3C,EAAI,2BAAG,GAAI,EAAqB,OACvC,GAAA,GAAA,GAAc,KAAK,CAAA,EAAM,GACrB,EAAG,GAAI,KAAM,EAAqB,MAFO,CAAA,EAAA,YAC7C,EAAA,MAAA,EAAA,CAAmB,EAAA,4BAD4B,QAAE,iBAUrD,IAHM,EAAM,EAAG,OAAO,GAAI,OAAO,GAC3B,EAAO,EAAc,GAC3B,EAAA,QAAY,GACH,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EAChC,EAAQ,EAAU,GAClB,EAAM,EAAK,GACjB,EAAU,GAAS,EACnB,EAAA,KAAS,GAGX,MAAA,CAAA,EAAM,EAAa,WAAW,EAAY,YAA1C,EAAA,OACA,GAAqB,GAErB,IACA,6BAGE,GAAqB,GAAa,EAAK,gBAClB,EAAY,MAE/B,EACE,GAAO,OACP,GAAgB,EAAK,gBACb,GAAA,GAAO,CAAA,EAAM,EAAM,gBACzB,EAAK,eAAgB,CAAC,QAAS,EAAK,sBAFtC,CAAA,EAAA,KAFF,CAAA,EAAA,IAHF,CAAA,EAAA,YAME,SAAU,EAAA,MAAA,OAAA,CAAO,EAAA,wBAGjB,EAAU,GAAO,EAAM,SAAS,EAAO,EAAO,CAC5C,UAAW,EAAK,qBAAuB,KACnC,GACA,EAAK,oBACT,QAAS,wBAGb,IAAS,EAAI,EAAG,EAAI,EAAM,aAAa,OAAQ,EAAE,EAC/C,EAAU,OAAO,EAAM,aAAa,IAAQ,EAAQ,sBAQxD,MAAA,CAAA,EAAA,YAGF,MAAI,GAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,YAErC,MAFA,GAAA,OACA,IACI,EAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,sBAAnB,SAAA,OACA,CAAA,EAAM,EAAM,QAAQ,oBAApB,SAAA,OACA,CAAA,EAAO,EAAM,iBAEb,SAAM,WAAa,8BAKvB,YACI,EAAqB,GAEvB,GAAI,GAAwB,KAC5B,MAAI,GAAK,iBAAmB,KAC1B,EAAgB,EAAK,gBAChB,AAAI,OAAO,SAAS,EAAQ,OACjC,GAAgB,EAAQ,MAEnB,EAKT,YACI,GAKF,MAAQ,OAAQ,GAAuB,UAAa,WAKtD,YAAiC,GAE/B,MAAQ,OAAQ,GAA6B,MAAS,uBAOpD,EAAY,EACZ,uHAKF,GAJA,EAAO,GAAQ,GACT,EAAa,EAAK,SAAW,KAC7B,EAAI,EAAM,aACZ,EAAqB,GACrB,EAAK,QAAU,EACjB,KAAM,IAAI,IAAoB,8CAGhC,GAAA,KAAS,OACL,CAAC,GAAe,EAAK,QAAU,GAAK,OAAO,UAAU,EAAK,SAC1D,WAAM,MAAA,6DACF,aAAY,KAAK,UAAU,EAAK,YACnB,GAAqB,GACtC,GAAA,SADiB,CAAA,EAAA,UAEjB,MAAA,CAAA,EAAO,EAAuB,mBAA9B,EAAA,EAAA,wBAFE,EAAY,EAId,EAAc,EACd,EAAQ,uEAGU,MAAA,CAAA,EAAM,EAAa,eAmCvC,MAnCM,GAAc,EAAA,OACpB,EAAO,EAAA,KAAS,WACd,GAAI,EAAY,OAGR,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAU,EAAG,OAAO,GACpB,EAAY,EAAA,KAAS,WAAM,MAAA,GAAE,KAGnC,GAFA,EAAA,QAAY,GAER,IAAU,EACZ,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAKrB,OADM,GAAY,EAAQ,GAAG,MAAM,cAC1B,GACP,GAAM,GAAW,EAAU,GACrB,EAAY,EAAK,GACvB,EAAK,GACD,EAAA,KAAS,WAAM,MAAA,GAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,MACvD,AAAI,EAAQ,GACV,EAAA,QAAY,IANP,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,IAA/B,GAST,EAAA,QAAY,GACZ,GAAe,EAEf,EAAE,EAEJ,MAAO,KAGL,EAAY,KACd,CAAI,GACF,QAAQ,KACJ,uJAGA,2BAA0B,EAAK,QAAO,eACtC,qHA3CH,GAAa,EAAQ,EAAK,QAAU,YAAI,CAAA,EAAA,2DAkD/C,IAAS,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC3B,EAAY,EAAK,GACvB,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAC3B,EAAA,QAAY,GAGd,MAAA,CAAA,EAAO,GAAiB,SCrmB1B,YAiK+B,GAC7B,EAAA,KAAS,OACL,EAAY,GAAK,OAAO,UAAU,GAClC,WAAM,MAAA,2DACF,gBAiBN,EAAyB,EAAe,GAC1C,MAAI,IAAU,KACL,CAAC,MACC,MAAM,QAAQ,GAChB,EAAO,IAAI,SAAA,GAAS,MAAA,IAAoB,EAAO,EAAO,EAAO,KAE7D,GAAoB,EAAQ,EAAO,EAAO,eAkBjD,EAAyB,GAC3B,MAAO,GAAA,KAAS,WACd,MAAI,IAAU,KACL,KACE,MAAM,QAAQ,GAChB,EAAO,IACV,SAAA,GAAS,MAAC,IAAqB,EAAO,KAInC,GACH,EAAQ,EAAQ,QAAU,QAAU,EAAU,EAAQ,uBAc5D,EAAc,GAIhB,OAHM,GAAkC,GACpC,EAAa,EACb,EAAmB,KAChB,EAAa,GAClB,EAAW,EAAa,EACxB,AAAI,GAAY,GACd,GAAW,GAEb,EAAO,KAAK,CAAC,EAAY,IACzB,EAAa,EAEf,MAAO,GA8BT,YAGI,EAAY,EAAiC,EAC7C,EAAsB,EAAoB,EAAiB,EAC3D,EAA4B,EAC5B,EAAmB,EAA0B,EAC7C,EAAuB,EACvB,mHAoBF,GAnBA,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,GAAU,MACZ,GAAS,GAEX,AAAI,GAAW,MACb,GAAU,IAEZ,AAAI,GAAgB,MAClB,GAAe,GAIb,EAAe,GACnB,AAAI,GAAQ,MAAQ,GAAU,MAC5B,GAAe,IAGb,GAAmB,MACrB,GAAe,GACX,GAAiB,MACnB,KAAM,IAAI,GACN,oGAKF,SACF,EAAM,gBAAgB,EAAK,EAAW,EAAe,mBAEzD,AAAI,GAAmB,MACrB,GAAa,GAAM,EAAG,IAGxB,AAAI,GAAW,MACb,GAAU,GAGN,EAA0B,GAC5B,EAAW,EAAS,EAAQ,EAAc,EAAiB,EAC3D,EAAW,EAAc,GAFtB,EAAY,EAAA,aAAE,EAAO,EAAA,QAG5B,EAAa,SAAS,GACtB,EAAM,QAAU,EAChB,CAAA,EAAM,EAAa,uBAAnB,EAAA,OACA,EAAM,cAAgB,cAIb,qEACP,MAAA,CAAA,EAAM,EAAa,aAAa,cAAhC,EAAA,OACM,EAA4B,IAC9B,IAAiB,MAAjB,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,IACN,qDAEJ,GAAI,IAAY,QACd,KAAM,IAAI,IAAoB,0CACzB,AAAI,GACT,EAAA,KAAK,QAAQ,GAIT,EAAoB,EAAA,SAAS,GAE7B,EAAU,GAAY,EAAiB,cACpC,2DACD,SAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,EAAY,WAA5C,SAAA,OAEA,EAAA,KAAS,WACP,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,GAAW,GACI,EAAmB,EACnB,EAAW,GAChC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAW,EAM/B,OAFM,IAAW,GAAqB,EAAK,IACrC,GAAO,EAAE,IACN,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAK,IACjB,EAAU,IAAS,GACnB,EAAA,KAAS,IAIX,GAAI,IAAe,EAAQ,OAAS,GAC9B,EAGF,OAFM,IAAU,EAAM,SAAS,EAAM,EAAQ,GAEpC,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAQ,IACpB,EAAA,KAAS,IAET,EAAU,OAAS,IAAS,MAMpC,CAAA,EAAM,EAAa,WAAW,EAAY,WAG1C,MAHA,GAAA,OACA,GAAqB,GAEjB,EAAM,kCA3CH,EAAa,yBAAG,GAAa,EAAQ,YAArC,IAA2C,CAAA,EAAA,iEAAE,QAAE,eAiDxD,EAAkB,2BAGpB,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,WACrC,MADA,GAAA,OACI,EAAM,kCAtEH,EAAQ,yBAAc,GAAQ,OAA9B,IAAoC,CAAA,EAAA,iEAAE,QAAE,eA0EjD,MAAA,CAAA,EAAM,EAAa,qBAAnB,SAAA,OAEA,CAAA,EAAM,EAAM,QAAQ,mBAApB,SAAA,OACA,CAAA,EAAO,EAAM,0BAMX,EAAY,EACZ,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,yIACF,GAAI,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,iDASX,EAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIT,EAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAG,EAAG,EAAK,aAAc,EAAK,YAAa,EAC3C,cAHF,EACF,EAAA,OAGJ,EAAS,EAAiB,GAC1B,EAAU,EAAiB,GAC3B,EAAgB,EAAiB,GAG7B,EAAe,GACf,EAAM,QACN,GAAK,gBAAkB,MAAQ,EAAK,eAAe,OAAS,GAA5D,MAAA,CAAA,EAAA,GAEF,GADA,EAAe,GACX,EAAK,eAAe,SAAW,EAEjC,EAAY,EAAK,eAAe,GAChC,EAAY,EAAK,eAAe,OAC3B,MAAI,GAAK,eAAe,SAAW,EAClC,GAAI,IACN,iEAEE,GAAI,GACN,0GAEG,GAAK,eAAc,iBAGtB,SAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAW,EAAW,KACtB,KACA,EAAgB,WAJlB,SACF,EAAA,OAIJ,EAAO,EAAgB,GACvB,EAAO,EAAgB,GACvB,EAAS,EAAK,OAAO,gBAEhB,AACH,EAAK,iBAAmB,MAAQ,EAAK,gBAAkB,GACvD,EAAK,gBAAkB,EACzB,GAAe,GAET,EACF,KAAK,MAAM,EAAO,GAAG,MAAM,GAAM,GAAI,EAAK,kBACxC,EAAoB,EAAO,GAAG,MAAM,GAC1C,EAAO,GAAY,EAAQ,EAAS,GACpC,EAAS,GAAY,EAAQ,EAAG,GAChC,EAAO,GAAY,EAAS,EAAS,GACrC,EAAU,GAAY,EAAS,EAAG,GAGlC,EAAS,EAAK,OAAO,IAGhB,AAAI,EAAK,iBAAmB,MACjC,GAAe,qBAIX,SAAM,EAAO,OAAO,GAAS,OAAO,GAE1C,EAAM,mCAcA,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAW,OACX,EAAe,OACnB,AAAI,EACF,GAAM,mBACN,EAAc,EAAM,aACpB,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,MAEzD,GAAc,KACd,EAAS,GACT,EAAkB,EAAU,SAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YAChD,CAAA,EAAM,GACd,EAAO,EAAe,EAAK,EAAW,EAAW,EAAK,OACtD,EAAK,QAAS,EAAW,EAAa,EAAQ,EAAK,QACnD,EAAiB,EAAK,aAAc,KAAM,cAHxC,SAAM,EAAA,OAIZ,CAAA,EAAO,UAEP,SAAM,WAAa,GAEnB,GAAkB,EAAQ,GAC1B,GAAkB,EAAS,GAC3B,GAAkB,EAAkB,GACpC,GAAkB,EAAkB,GACpC,AAAI,GAAiB,MACnB,EAAA,QAAY,yCAYyB,GACzC,GAAM,GAAiB,GACvB,AAAI,YAAmB,GAAA,QACrB,GAAU,CAAC,IAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAS,EAAQ,GACvB,GAAI,EAAO,OAAS,EAClB,EAAK,KAAK,GAAW,EAAQ,SACxB,GAAI,EAAO,OAAS,EACzB,KAAM,IAAI,OACN,yEAGJ,EAAK,KAAK,IAGd,MAAO,eAgBL,EACA,GACF,GAAI,GAAW,KACb,OAEF,GAAM,GAAyB,GAC/B,GAAI,YAAsB,GAAA,OACxB,EAAa,KAAK,EAAW,YACpB,MAAM,QAAQ,GACvB,EAAW,QAAQ,SAAA,GAAK,MAAA,GAAa,KAAK,EAAE,cACnC,GAAc,KAEvB,OAAW,KAAQ,IACjB,GAAM,GAAY,EAAW,GAC7B,EAAa,KAAK,EAAU,IAIhC,GAAM,GAA6B,GACnC,GAAI,YAAmB,GAAA,OACrB,AAAI,EAAa,QAAQ,EAAQ,MAAQ,IACvC,EAAiB,KAAK,WAEf,MAAM,QAAQ,GACvB,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAa,QAAQ,EAAE,MAAQ,IACjC,EAAiB,KAAK,aAGjB,GAAW,KAEpB,OAAW,KAAQ,IACjB,GAAM,GAAS,EAAQ,GACvB,AAAI,EAAa,QAAQ,EAAO,MAAQ,IACtC,EAAiB,KAAK,GAK5B,EAAiB,QAAQ,SAAA,GACvB,AAAK,EAAE,YACL,EAAE,YCrnBR,YA+C6B,GAE3B,MAAO,aAAa,GAAA,mBAMM,GAE1B,MAAO,OAAM,QAAQ,eAMI,GAEzB,MAAO,CAAC,GAAa,IAAM,CAAC,GAAY,eAetC,EAAqD,EACrD,EAAkB,EAAuB,GAC3C,GADoB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACvC,GAAS,MAAQ,EAAM,SAAW,GAGpC,GAAI,GAAQ,MACV,GAAI,GAAoB,GACxB,GAAI,GAAY,IAAU,EAAkB,OAAS,EACnD,EAAoB,WACX,GAAW,IACpB,OAAW,KAAO,GAChB,GAAI,EAAK,eAAe,IACtB,EAAoB,GACpB,WAKJ,GAAoB,GAEtB,GAAI,EACF,KAAM,IAAI,GACN,6BAA6B,EAAe,sBAC5C,YAAW,IAGnB,MAAO,GAET,GAAI,GAAQ,KACV,MAAO,GAAM,IAAI,SAAA,GAAQ,MAAA,QAG3B,GAAI,GACJ,GAAI,GAAW,IACb,EAAO,EACP,EAAS,GACT,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAK,IAAS,KAChB,KAAM,IAAI,GACN,yBAAyB,EAAI,iCAC7B,IAAG,IAET,EAAO,KAAK,EAAK,aAEV,GAAY,IAErB,GADA,EAAO,EACH,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,oCAAmC,EAAM,OAAM,oBAC/C,iDAAgD,IAEtD,EAAS,OAGT,GADA,EAAO,EACH,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,aAAa,EAAe,YAAY,EAAM,OAAM,eACpD,2DACI,EAAK,QAEf,EAAS,CAAC,GAMZ,GAHA,EAAS,GAA2B,GAGhC,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,EAAM,QAErB,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EAEd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MAAQ,GAAU,GAAK,IAAQ,EAC3C,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,mBAAkB,EAAO,GAAE,gCAC3B,KAAI,EAAM,MAAK,QAK3B,MAAO,eAWL,EAAkB,EAAmB,GACvC,GAAM,GAAO,GAAO,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,MAAM,MACpD,EAAK,OACL,GAAM,GAAO,GAAO,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,MAAM,MAGvD,GAFA,EAAK,OAED,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,mFAEA,IAAG,KAAK,UAAU,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,WAElD,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,oFAEA,IAAG,KAAK,UAAU,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,WAErD,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,CAAC,EAAA,KAAK,YAAY,EAAM,GAChE,KAAM,IAAI,GACN,kEACA,mBAAkB,EAAK,GAAE,wBAAwB,EAAK,GAAE,YACxD,cAaR,YACI,EAAmB,EAA2B,GAMhD,OAJM,GAAY,CAChB,GAAyB,GACzB,IAEO,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAI,EAAQ,GACZ,EAAO,EAAQ,GACf,EAAQ,EAAa,GAC3B,GAAI,GAAQ,KACV,SAEF,GAAI,IAAS,IACP,EAAE,MAAM,EAAE,MAAM,OAAS,KAAO,EAClC,KAAM,IAAI,GACN,2CAA2C,EAAE,MAAK,4JAO1D,GAAI,EAAU,QAAQ,KAAU,GAG9B,OAFM,GAAe,EAAE,MAAM,MAAM,GAC7B,EAAc,EAAM,MAAM,GACvB,EAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAY,EAAa,GACzB,EAAS,EAAY,GAC3B,GAAI,GAAU,MAAQ,IAAc,EAClC,KAAM,IAAI,GACN,8BAA8B,EAAE,MAAK,sBACrC,oBAAmB,EAAK,uCACxB,2DAiCd,YACI,EAAuB,EAAiB,EACxC,EAAuB,GAAvB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACzB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,wCAAuC,EAAM,OAAM,eACnD,qBAAoB,EAAK,OAAM,iBAErC,EAAS,OAET,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,qBAAqB,EAAM,OAAM,IAAI,EAAe,mEAEjD,MAAK,UAAU,EAAK,OAAM,MAEnC,EAAS,CAAC,GAGZ,GAAI,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,KAAK,UAAU,EAAM,SAEpC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EACd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MACR,IAAW,EACb,KAAM,IAAI,GACN,uBAAuB,EAAe,cACnC,GAAM,GAAE,kBAAkB,KAAK,UAAU,EAAO,IAAG,SACtD,yBAAwB,KAAK,UAAU,EAAM,OAAM,oBAsB/D,EAEA,GACF,GAAI,GAAW,MAAQ,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAClE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,KAGjC,GAAI,GAEJ,GAAI,MAAO,IAAY,UAAY,MAAO,IAAY,WACpD,EAAiB,CAAC,WACT,MAAM,QAAQ,IAAY,MAAO,IAAY,SACtD,EAAiB,MAGjB,MAAM,IAAI,WACN,+DACA,uCAAsC,IAG5C,GAAI,MAAM,QAAQ,GAEhB,MAAO,GAAY,IACf,SAAA,GAAQ,MAAA,KAIZ,OADM,GAAqD,GACxC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACT,EACA,EAAe,eAAe,GAAQ,EAAe,GAAQ,GACjE,AAAK,MAAM,QAAQ,IACjB,GAAgB,CAAC,IAEnB,EAAc,KAAK,GAErB,MAAO,GA6DX,GAAM,IAA2B,8BAcA,EAAA,EAAA,GA4C/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WAAa,KAsCpB,SAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAGF,GAHE,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC1D,CAAC,KAAK,MACR,KAAM,IAAI,GACN,gLAIN,GAAa,KAAM,EAAY,EAAW,IAa5C,EAAA,UAAA,QAAA,SAAQ,GAAR,GAAA,GAAA,KAME,GALA,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAEd,KAAK,KAAO,EAAK,KAEb,MAAO,GAAK,WAAc,SAC5B,KAAK,WAAa,GAAwB,EAAK,WAC/C,KAAK,iBAAmB,QAExB,GAAI,CAAE,GAAK,oBAAqB,GAAA,WAC9B,KAAM,IAAI,GACN,+DAEN,KAAK,WAAa,EAAK,UACvB,KAAK,iBAAmB,GAO1B,GAAI,GAAkC,GACtC,GAAI,CAAC,MAAM,QAAQ,EAAK,OAAS,MAAO,GAAK,MAAS,UAClD,MAAO,GAAK,MAAS,YACvB,EAAK,KAAO,EAAK,KACjB,OAAW,KAAQ,GAAK,KACtB,GAAI,KAAK,YAAY,QAAQ,KAAU,GACrC,KAAM,IAAI,GACN,sCAAsC,EAAI,MAC1C,sCAAqC,KAAK,cAGlD,OAAmB,GAAA,EAAA,EAAA,KAAK,YAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAK,KAAK,IAAS,MACrB,QAAQ,KACJ,WAAW,EAAI,4GAEf,oBAAmB,EAAI,qBAE7B,EAAc,KAAK,GAAW,EAAK,KAAK,cAEjC,MAAM,QAAQ,EAAK,OAC5B,GAAI,EAAK,KAAK,SAAW,KAAK,QAAQ,OACpC,KAAM,IAAI,GACN,+DACA,gCAA+B,KAAK,QAAQ,OAAM,gBAClD,wBAAuB,EAAK,KAAI,MAEtC,GAAM,GAAY,EAAK,KACvB,EAAgB,EAAU,IAAI,SAAA,GAAK,MAAA,IAAW,UAE9C,GAAM,GAAe,GAAW,EAAK,MACrC,KAAK,QAAQ,QAAQ,SAAA,GACnB,EAAc,KAAK,KAIvB,KAAK,cAAgB,EAErB,KAAK,gBAAkB,GACvB,KAAK,iBAAmB,GACxB,KAAK,YAAc,GACnB,OAAS,GAAI,EAAG,EAAI,KAAK,QAAQ,OAAQ,EAAE,GAEzC,GAAM,GAAQ,KAAK,qBAAqB,GAClC,EAAO,KAAK,YAAY,GAC9B,KAAK,gBAAgB,KAAK,GAC1B,KAAK,iBAAiB,KAAK,GAC3B,KAAK,YAAY,KAAK,KAAK,cAAc,IAK3C,GAAM,GAA8B,GAGpC,KAAK,QAAU,EAAK,QAEpB,KAAK,aAAe,CAAC,QACrB,KAAK,eAAiB,GAMtB,GAAU,OAAQ,WAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,GACzC,GAAI,EAAkB,QAAQ,KAAO,GACnC,SAIF,GAAM,GAAe,EAAK,cAAc,GACxC,AAAI,EAAK,QAAQ,OAAS,GACxB,GAAK,eAAe,KAAK,CAAC,EAAc,IACxC,EAAK,aAAa,KAAK,EAAK,YAAY,GAAK,aAQnD,GAAM,GAAgB,GAAe,EAAK,QAAS,KAAK,aAMlD,EACF,SAAC,EAAqB,EACrB,GACC,AAAI,EAAK,YAAY,OAAS,GAC5B,GAAa,EAAK,YAAY,GAAe,IAAM,GAErD,EAAK,aAAa,KAAK,GACvB,EAAK,eAAe,KAAK,CAAC,EAAc,KAG9C,GAAU,SAAU,WAClB,mBAAS,GACP,GAAI,EAAkB,QAAQ,KAAO,oBAGrC,GAAM,GAAgB,EAAc,GAI9B,EAAgB,SAAC,GAOrB,OANM,GAAmB,GACrB,EACA,EACA,aAGO,GACT,GAAI,MAAO,IAAW,UAClB,CAAC,WAAY,MAAO,eAAgB,MAAM,QAAQ,KAC9C,IACN,GAAM,GAAc,EAAK,qBAAqB,GAE9C,AAAI,EAAY,EAAY,OAAS,KAAO,GACxC,EAAK,cAAc,KAAO,GAE5B,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAEL,AACH,EAAK,cAAc,KACnB,GAGF,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAIV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAGZ,GAAI,GAAM,OACV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAS,MACJ,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAS,MAGX,EAAmB,EACnB,EAAa,EAAmB,OAEhC,GAAM,GAAW,GAAY,GAE7B,EAAmB,EACnB,EACI,EAAmB,GAA4B,GAIrD,GAAI,GACJ,GAAU,EAAY,WACpB,EAAe,IAEjB,EAAa,EAAG,EAAY,IAtDT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,KAAN,KA0Db,EAAc,IAzEP,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,IAAlC,KAgFX,KAAK,0BAA4B,KAAK,kBAY9B,EAAA,UAAA,iCAAV,WACE,GAAI,KAAK,2BAA6B,KACpC,OAEF,AAAI,KAAK,iBAAiB,SACtB,KAAK,0BAA0B,QACjC,QAAQ,KACJ,sJAqCR,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIf,GAAM,GAAiB,GACjB,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GACrD,IAGE,GAAM,GAAM,EAAiB,GAAG,OAAO,EAAiB,IACxD,KAAK,mBACL,GAAM,GAAI,KAAK,aACT,EACF,KAAK,SAAS,EAAG,EAAK,EAAW,EAAK,QAAS,EAAK,OACxD,MAAO,IAAiB,WAExB,GAAkB,EAAiB,GAAI,GACvC,GAAkB,EAAiB,GAAI,KA0BrC,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,YAAK,mBACL,CAAA,EAAO,GAAgB,KAAM,EAAS,SAahC,EAAA,UAAA,gBAAR,SACI,EAAsB,EAAoB,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,SACF,GAAI,GACJ,GAAI,GAAS,MAEX,GADA,EAAa,KACT,GAAa,KACf,KAAM,IAAI,GACN,MAAM,EAAS,gDACf,oBAAmB,YAEhB,GAAO,KAChB,AAAI,MAAM,QAAQ,GAChB,EAAa,EAAI,GAAG,MAAM,GAE1B,EAAa,EAAI,MAAM,OAGzB,MAAM,IAAI,GACN,yDACG,GAAS,yBAElB,MAAO,IAUT,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,GAAI,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAC/C,KAAM,IAAI,GACN,sDAGN,GAAM,GAAiB,MAAM,QAAQ,GAC/B,EACD,EAAiB,EAAsB,CAAC,GACvC,EAAwB,KAAK,wBAAwB,GAGrD,EAAW,GAAI,IAIrB,GAHA,AAAI,YAAkB,GAAA,QACpB,GAAS,CAAC,IAER,MAAM,QAAQ,IAChB,GAAI,EAAO,SAAW,KAAK,OAAO,OAChC,KAAM,IAAI,GACN,kCAAkC,EAAO,OAAM,uDAE/C,KAAI,KAAK,OAAO,OAAM,OAE5B,OAAS,GAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,KAAK,OAAO,GAAI,EAAO,QAGtC,QAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAc,EAAO,EAAM,MACjC,GAAI,GAAe,KACjB,KAAM,IAAI,GACN,8CAA8C,EAAM,MAE1D,EAAS,IAAI,EAAO,GAKxB,GAAM,GAAiB,GAAQ,EAAuB,GACtD,MAAO,GAAiB,EAAiB,EAAe,IAMlD,EAAA,UAAA,wBAAR,SAAgC,GAK9B,OAHM,GACF,GAAa,KAAM,EAAoB,QACvC,EAAmB,EAAoB,OACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EACF,MAAM,QAAQ,EAAM,QAAU,EAAM,OAAS,CAAC,EAAM,QAClD,EAAmB,EAAa,IAAI,SAAA,GAAU,MAAA,GAAO,OAClD,EAAI,EAAG,EAAI,EAAoB,OAAQ,EAAE,GAChD,GAAM,GAAQ,EAAiB,QAAQ,EAAoB,IAK3D,GAJA,AAAI,IAAU,IACZ,GAAsB,GAAK,EAAa,GACxC,KAEE,IAAqB,EACvB,MAGJ,GAAI,IAAqB,EACvB,MAIJ,GAAI,EAAmB,GACrB,GAAM,GAA2B,GACjC,QAAsB,QAAQ,SAAC,EAAQ,GACrC,AAAI,GAAU,MACZ,EAAe,KAAK,EAAoB,MAGtC,GAAI,GACN,mDACA,IAAG,KAAK,UAAU,KAExB,MAAO,IAgBD,EAAA,UAAA,YAAR,SAAoB,EAAsB,EAAgB,GAA1D,GAAA,GAAA,KAA0C,MAAA,KAAA,QAAA,GAAA,IAAgB,AAAA,IAAA,QAAA,GAAA,IAEjD,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,GACxC,GAAI,EACF,KAAM,IAAI,IACN,iDAYN,OAJM,GAAU,GAAY,EAAY,GAClC,EAA0B,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,gBAGlD,GACP,GAAM,GAAY,EAAA,KAAS,WACzB,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAG/B,EAAW,GAAY,EAAK,EAAY,GAGxC,EAAQ,GACd,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,EAAE,EACrC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAS,SAGnD,GAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,IAE1C,GAAM,GAAW,GAAI,IAAS,GAC9B,MAAO,IAAQ,EAAK,QAAS,KAE/B,EAAU,QAAQ,SAAC,EAAU,GAAM,MAAA,GAAY,GAAG,KAAK,MApBhD,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,IAA/C,GAsBT,MAAO,IACH,EAAY,IAAI,SAAA,GAAW,MAAA,GAAA,OAAW,EAAS,SA+BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAkB,GAA2B,GACnD,GACI,EAAiB,KAAK,WAAY,KAAK,gBAAiB,IAC5D,IAKE,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,UAAe,GACR,KAAK,YAAY,EAAiB,WAEzC,GAAkB,EAAiB,KAmBvC,EAAA,UAAA,eAAA,SAAe,GACb,GAAe,EAAG,KAAK,WAAY,KAAK,gBAAiB,IAGzD,GAAM,GAAa,OAAM,QAAQ,GAAK,EAAE,GAAK,GAAG,MAAM,GACtD,MAAO,MAAK,YAAY,EAAG,IAGnB,EAAA,UAAA,sBAAV,SACI,EACA,EAAkD,EAClD,GAEF,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,IACN,gGAIN,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,KAAK,iBAAiB,OAAQ,EAAE,GAClD,GAAM,GAAc,KAAK,iBAAiB,GACpC,EAAS,KAAK,YAAY,GAChC,AAAI,IAAW,GACb,EAAa,KACT,EAAY,MAAM,EAAG,EAAY,OAAS,GAAG,OAAO,CAAC,KAGzD,EAAa,KAAK,GAWtB,GARA,EAAI,GACA,EAAG,KAAK,eAAgB,KAAK,gBAAiB,GAAO,SACzD,EAAI,GACA,EAAG,KAAK,gBAAiB,EAAc,GAAO,UAElD,GAAkB,EAAG,GAErB,GAAgC,EAAG,KAAK,YAAa,KAAK,kBACtD,KAAK,UAAY,GAAa,MAAQ,EAAY,GAChD,EAAE,GAAG,MAAM,GAAK,IAAc,EAChC,KAAM,IAAI,GACN,mHAEG,GAAS,YAAY,EAAE,GAAG,MAAM,GAAE,gBAG7C,MAAO,CAAC,EAAG,IAGG,EAAA,UAAA,oBAAhB,SACI,EACA,EACA,EACA,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAA,2GAKF,GAHM,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GAD9C,EAAU,EAAA,GAAE,EAAU,EAAA,GAGzB,GAAgB,KAClB,KAAM,IAAI,OAAM,0CAGd,EAAkC,MAClC,IAAe,MAAf,MAAA,CAAA,EAAA,GACI,EACF,GAAwB,EAAa,KAAK,aAC9C,EAAwB,GACf,EAAI,yBAAG,GAAI,EAAa,OAC/B,GAAA,GAAA,GAAsB,KAClB,CAAA,EAAM,GAAmB,EAAW,GAAI,KAAM,EAAa,MAF1B,CAAA,EAAA,UACrC,EAAA,MAAA,EAAA,CACI,EAAA,0BAFmC,QAAE,eAO7C,MAAA,CAAA,EAAO,CAAC,EAAY,EAAY,UAc1B,EAAA,UAAA,SAAR,SACI,EAAiC,EAAe,EAChD,EAAa,GAFjB,GAAA,GAAA,KAEI,MAAA,KAAA,QAAA,GAAA,GACK,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,EAAK,EAAW,EAAO,SACzD,EAAiB,GACvB,GAAI,EAAU,EACZ,KAAM,IAAI,IAAoB,wCAGhC,GAAI,GAAS,KACX,KAAM,IAAI,IACN,mDAIJ,OAFM,GAAU,GAAY,EAAY,GAClC,EAAa,EAAA,SAAS,GAAM,EAAG,IAC5B,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,GACtD,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,EACF,GACI,EAAY,EAAY,EAAW,GAGrC,EAAW,GAAqB,EAAK,GACrC,EAAY,EAAE,GACpB,GAAI,IAAe,EACjB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAGrB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAW,EAAU,GAC3B,EAAK,GACD,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,EAAY,KAGxD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAG/B,MAAO,MAID,EAAA,UAAA,uBAAV,WAKE,OAJM,GAAY,KAAK,aAGjB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAQ,EAAU,GACpB,EAAW,EACf,GAAI,GAAM,EAAW,GAAS,GAC5B,GAAM,GAAW,GAAM,EAAU,MAAM,EAAG,GAAI,GAC9C,GAAY,IAAI,EAElB,EAAiB,KAAK,GAExB,MAAO,IAaC,EAAA,UAAA,kBAAV,WAAA,GAAA,GAAA,KACE,MAAO,UAAC,GACN,GAAM,GAAuB,GAEvB,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAgB,EAAK,MACvB,EAAK,OAAO,OAAS,EAAK,QAAQ,OAClC,EAAK,OAAO,OAAS,EAAK,QAAQ,OAAS,GAEzC,EAA0B,GAK1B,EAAoB,WAExB,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KASjD,OAPM,GAAW,GAAI,IAAS,GACxB,EACF,GAAQ,EAAK,QAAS,EAAU,CAAC,SAAY,KAI7C,EACK,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GACpC,EAAO,EAAa,EAAQ,GAAI,EAAQ,IAC5C,AAAI,EAAc,IAAM,MACtB,GAAO,GAAoB,EAAM,EAAc,KAIjD,GAAM,GAAmB,EAAA,KAAS,GAElC,EAAW,KAAK,GAChB,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAOnC,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAI,GAAc,OAElB,GAAI,EAAK,QAAQ,OAAS,GAAK,EAAI,EAAK,QAAQ,OAC9C,EAAiB,EAAW,QAE5B,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAC3C,EACI,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAGpD,EAAA,KAAS,GAET,EAAc,KAAK,GAGrB,SAAY,EAAA,KAAS,GAGrB,EAAK,kBAAkB,QAAQ,SAAA,GAC7B,EAAY,EAAA,IAAQ,EAAW,KAG1B,GAGH,EAAY,EAAK,0BAA0B,IAC7C,SAAA,GAAS,MAAA,GAAM,SACb,EAAa,GACb,EACF,EAAK,WAAW,SAAS,EAAmB,EAAY,GAE5D,MAAO,CAAC,GAAgB,OAAO,KAS3B,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,KAAK,aAAe,SAAC,GACnB,MAAO,GAAA,KAAS,WAOd,OANM,GAAuB,GACzB,EACE,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KAKjD,OAHM,GAAW,GAAI,IAAS,GACxB,EAAU,GAAQ,EAAK,QAAS,GAE7B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAGlC,EAAe,EAAA,KAAS,EAAa,EAAQ,GAAI,EAAQ,KAC/D,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAEjC,EAAW,KAAK,GAGlB,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAErC,EACF,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAClD,EAAW,KAAK,GAElB,MAAO,OAuCP,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,MAAA,CAAA,EAAO,GAAW,KAAM,EAAG,EAAG,SA0B1B,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,MAAA,CAAA,EAAO,GAAW,KAAM,EAAS,SA0B7B,EAAA,UAAA,aAAN,SACI,EACA,qHAIqB,MAAA,CAAA,EAAM,KAAK,oBAAoB,EAAG,WAAnD,EAAiB,EAAA,OACjB,EAAS,EAAe,GACxB,EAAU,EAAe,GACzB,EAAgB,KAAK,oBACrB,EAAS,EAAc,EAAO,OAAO,IACrC,EAAuB,OACV,EAAA,yBAAA,GAAA,EAAA,OAAR,GAAI,EAAA,GACH,CAAA,EAAM,EAAK,SADE,CAAA,EAAA,UACjB,EAAI,EAAA,OACV,EAAW,KAAK,EAAE,qBAFD,wBAInB,SAAA,QAAY,GACZ,CAAA,EAAO,GAAiB,UAYhB,EAAA,UAAA,gBAAV,SAA0B,GAMxB,OALM,GAA8B,GAE9B,EAAgB,GAAU,MAAQ,EAAO,cACzC,EAAU,EAAgB,KAAK,iBAAmB,KAAK,QACvD,EAAe,KAAK,WAAW,GAC5B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAI,GAAiB,CAAC,EAAQ,GAAG,UAE/B,SAEF,EAAa,KACT,CAAC,KAAM,EAAQ,GAAG,aAAc,OAAQ,EAAa,KAE3D,MAAO,IAiCT,OAAA,eAAI,EAAA,UAAA,eAAY,KAIhB,WACE,MAAO,MAAK,mBALd,SAAiB,GACf,KAAK,cAAgB,mCAOvB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,AAAI,KAAK,aAAe,GACtB,MAAK,WAAa,EAClB,KAAK,iBAAmB,qCAI5B,EAAA,UAAA,QAAA,WACE,GAAM,GAAS,EAAA,UAAM,QAAO,KAAA,MAC5B,GAAI,EAAO,uBAAyB,GAAK,KAAK,WAAa,MACvD,KAAK,kBACP,GAAM,GAAmC,EAAA,SAAa,WACtD,KAAK,WAAW,UAChB,EAAO,sBACH,EAAmC,EAAA,SAAa,WAEtD,MAAO,IAGD,EAAA,UAAA,mBAAR,WAEE,GAAI,GAEJ,GAAI,MAAO,MAAK,MAAS,SACvB,EAAY,GAAY,KAAK,cACpB,MAAM,QAAQ,KAAK,OAC5B,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,OAAM,sDAGpB,EAAa,KAAK,KAAkB,IAAI,SAAA,GAAQ,MAAA,IAAY,UAG5D,GAAM,GAAc,OAAO,KAAK,KAAK,MACrC,EAAY,GAGZ,OAFM,GACF,KAAK,KACgB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,MAAO,GAAO,IAAgB,SAChC,EAAU,GACN,GAAY,EAAO,QAEvB,MAAM,IAAI,OAAM,uDAItB,MAAO,IAGD,EAAA,UAAA,qBAAR,WAEE,GAAI,MAAO,MAAK,SAAY,UACxB,MAAO,MAAK,SAAY,WAC1B,MAAO,CAAC,GAAY,GAA4B,KAAK,WAChD,GAAI,MAAM,QAAQ,KAAK,SAC5B,MAAO,MAAK,QAAQ,IAChB,SAAA,GAAU,MAAA,IAAY,GAA4B,MAEtD,GAAM,GAAyD,GAC/D,OAAW,KAAO,MAAK,QACrB,EAAmB,GACf,GAAY,GAA4B,KAAK,QAAQ,KAE3D,MAAO,IAID,EAAA,UAAA,kBAAV,WACE,MAAO,CACL,KAAM,KAAK,qBACX,QAAS,KAAK,uBACd,iBAAkB,CAChB,WAAY,KAAK,UAAU,eAC3B,OAAQ,KAAK,UAAU,eAQ7B,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,EAAe,kBAAoB,KACrC,KAAM,IAAI,OAAM,gDAElB,GAAI,EAAe,cAAgB,KACjC,KAAM,IAAI,OAAM,8CAElB,GAAI,EAAe,oBAAsB,KACvC,KAAM,IAAI,OAAM,oDAGlB,GAAM,GAAW,GAAoB,EAAe,kBAE9C,EAAY,GAAY,GAE1B,EACJ,GAAI,MAAO,GAAe,MAAS,SACjC,EAAO,GAAY,EAAe,cACzB,MAAM,QAAQ,EAAe,MACtC,EAAO,EAAe,KAAK,IAAI,SAAA,GAAa,MAAA,IAAY,aAC/C,EAAe,MAAQ,MAChC,EAAO,GACP,OAAW,KAAO,GAAe,KAC/B,EAAK,GAAO,GAAY,EAAe,KAAK,IAIhD,GAAI,GACJ,GAAI,MAAM,QAAQ,EAAe,SAC/B,EAAU,EAAe,QAAQ,IAAI,SAAA,GAAU,MAAA,IAAY,aAClD,EAAe,SAAW,MACnC,EAAU,GACV,OAAW,KAAO,GAAe,QAC/B,EAAQ,GAAO,GAAY,EAAe,QAAQ,IAItD,KAAK,QAAQ,CAAC,KAAI,EAAE,QAAO,EAAE,UAAS,KAoFlC,EAAA,UAAA,KAAN,SAAW,EAAmC,+HAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,GACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,GACN,gHAKF,MAAA,CAAA,EAAM,EAAA,GAAG,cAAc,KAAK,gBAAgB,kBAD1C,GACF,EAAA,OAEE,EAAe,GACf,EAAgB,KAChB,EAAc,KAAK,OAAO,EAAW,GACrC,EAAoC,CACxC,cAAe,EACf,OAAQ,GACR,YAAa,8BAA8B,GAC3C,YAAa,MAGT,EAAmB,GAAU,KAAO,GAAQ,EAAO,iBACrD,GAAoB,KAAK,WAAa,KACxC,GAAe,eAAiB,KAAK,oBAC/B,EAAa,YAET,EAAA,GAAA,EAAA,IAAG,cAAc,CAAA,EAAM,KAAK,UAAU,eAJ9C,CAAA,EAAA,UAIE,MAAA,CAAA,EAAM,EAAA,MAAA,EAAA,CAAiB,EAAA,OAAmC,YADxD,EACF,EAAA,OADS,EAAmB,EAAA,KAAS,EAAoB,EAAA,MAE7D,GAAA,EAAmB,OAAM,KAAI,MAAA,EAAI,GACjC,EAAmB,KAAO,EAAA,GAAG,wBACzB,CAAC,EAAmB,KAAM,qBAGhC,MAAI,MAAK,qBAAuB,MAExB,GAAY,GAClB,GAAyB,KAAK,oBAAqB,KAAK,KAAM,GAC9D,EAAe,oBAAsB,KAAK,qBAG5C,EAAe,WAAa,EAAmB,KAC/C,EAAe,YAAc,EAAmB,MAChD,CAAA,EAAO,EAAa,KAAK,UAW3B,EAAA,UAAA,uBAAA,SAAuB,GACrB,GAAyB,EAAqB,KAAK,MACnD,KAAK,oBAAsB,GAc7B,EAAA,UAAA,uBAAA,WACE,MAAO,MAAK,qBAv4CP,EAAA,UAAY,QAy4CrB,GA74CiC,IA84CjC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DACS,SAAA,UAAY,aACrB,GAFgC,IAGhC,EAAA,cAAc,cAAc,IC72D5B,YA8DI,EACA,uHACF,AAAM,iBAAmB,IACvB,GAAwB,CAAC,cAAe,IAE1C,EAAwB,EAEpB,EAAgB,EAAsB,cAC1C,AAAI,EAAc,cAAmB,MAMnC,GAAgB,EAAc,cAE1B,EACF,GAAoB,GAClB,EAAQ,GAAY,EAAU,GAEhC,EAAsB,iBAAmB,KAItB,CAAA,EAAM,EAAA,GAAG,YAC1B,EAAsB,gBAAiB,EAAsB,WAC7D,EAAM,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,iBANrC,CAAA,EAAA,UAUF,IANM,EAAe,EAAA,OAKf,EAAqC,GAC3C,EAAA,EAAqB,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,IAAV,EAAM,EAAA,GACf,EAAmB,EAAO,cACtB,EAAa,EAAO,cAG1B,EAAM,YAAY,GAElB,EAAA,QAAQ,oBAEV,MAAA,CAAA,EAAO,oBAsIL,EACA,4EAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,MAAO,IAAoB,UAE7B,GADM,EAAW,EAAA,GAAG,gBAAgB,EAAiB,GACjD,EAAS,SAAW,EAKtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAiB,YAC5C,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAe,MAE7B,EAAkB,EAAS,GAE7B,MAAA,CAAA,EAAO,GAA6B,EAAiB,OAAW,oBAc9D,EAAuB,EACvB,mHAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,EAAQ,MAAQ,KAClB,KAAM,IAAI,GACN,iHAGY,MAAA,CAAA,EAAM,EAAQ,kBAA1B,EAAY,EAAA,OACd,EAAgB,EAAU,cAC9B,AAAI,EAAc,cAAmB,MACnC,GAAgB,EAAc,cAG1B,EAAS,EAAQ,QAAU,KAAO,GAAO,EAAQ,OAMjD,EACF,EAAU,YAAc,MAAQ,EAAU,aAAe,MAAQ,EAC/D,EACF,GACI,GAAoB,GACpB,EAAe,GAEjB,EAAiB,EAAU,eACjC,AAAI,GAAkB,MACpB,EAAM,mBAAmB,GAE3B,AAAI,EAAU,qBAAuB,MACnC,EAAM,uBAAuB,EAAU,sBAIrC,GAAU,YAAc,MAAxB,MAAA,CAAA,EAAA,GAEF,GAAI,EAAU,aAAe,KAC3B,KAAM,IAAI,GACN,wHAIA,GAAmC,GACrC,EAAU,WAAY,EAAU,aAD7B,EAAY,EAAA,aAAE,EAAgB,EAAA,iBAErC,EAAM,YAAY,EAAc,GAE5B,EAAM,WAAa,MAAQ,EAAiB,OAAS,EACvD,CAAA,EAAM,EAAM,UAAU,WAAW,IAD/B,CAAA,EAAA,UACF,EAAA,wBAIF,EAAA,QAAQ,GACR,EAAA,QAAQ,EAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,2BAEtC,MAAA,CAAA,EAAO,QAGT,YACI,EAAqB,GAEvB,GAAM,GAAc,EAAA,GAAG,cAAc,EAAQ,GACvC,EAA+B,GAC/B,EAAkC,GACxC,SAAM,QAAQ,SAAA,GACZ,AAAI,EAAK,QAAU,YACjB,EAAiB,KAAK,CAAC,KAAM,EAAK,KAAM,OAAQ,EAAY,EAAK,QAEjE,EAAa,EAAK,MAAQ,EAAY,EAAK,QAGxC,CAAC,aAAY,EAAE,iBAAgB,sBAwCR,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CAAC,OAAQ,GAAI,QAAS,MAAI,KAUhC,GATA,EAAO,GAAQ,GAEf,EAAK,UAAY,GACjB,EAAK,MAAQ,GAGb,EAAK,KAAQ,EAAK,MAAQ,KAAQ,EAAK,KAAO,GAAO,eAGjD,EAAK,QAAU,KACjB,OAAoB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,IAAI,YAOP,SAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAQ,EAAM,aAAa,GAAG,cAAc,GAAG,MACrD,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,GAAI,IACtB,KAAM,IAAI,GACN,kDACG,GAAM,KAAI,uBACV,GAAM,aAAa,GAAG,aAAa,GAAG,MAAK,OAyBtD,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GACF,YAAiB,IAAc,YAAiB,IAChD,EACJ,GAAI,GAEF,GADA,EAAa,EACT,EAAW,QAAQ,SAAW,EAChC,KAAM,IAAI,GACN,yHAKN,GAAI,EAAW,OAAO,SAAW,EAC/B,KAAM,IAAI,GACN,uHAOR,GAAI,KAAK,QAAQ,SAAW,GAE1B,GAAI,EAAM,aAAa,SAAW,GAEhC,GAAI,EAAM,iBAAmB,KAC3B,KAAM,IAAI,GACN,iGAIN,GAAM,GAAI,GAAM,CACd,WAAY,EAAM,gBAClB,MAAO,EAAM,MACb,KAAM,EAAM,KAAO,WAIrB,EAAM,MAAM,GAGd,GAAI,EACF,KAAK,QAAU,EAAW,QAC1B,KAAK,OAAS,EAAW,YAEzB,GAAI,EAAM,aAAa,SAAW,EAChC,KAAM,IAAI,GACN,2DACA,yDACI,EAAM,KAAI,KACd,cAAa,EAAM,aAAa,OAAM,0BACtC,gBAGN,GAAI,EAAM,aAAa,GAAG,cAAc,SAAW,EACjD,KAAM,IAAI,GACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,EAAM,aAAa,GAAG,cAAc,IACpD,KAAK,OAAS,GAAgB,KAAK,QAAQ,IAG7C,KAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,KACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,KAAK,OACnB,cAAe,KAAK,QAEpB,WAAY,GAA2B,KAAM,KAAK,OAAO,QACzD,YAAa,CAAC,MACd,YAAa,KAAK,OAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QACpC,aAAc,KAAK,QAAQ,GAAG,aAGhC,GAAM,GAAe,EAAM,MAAM,KAAK,QAAQ,IAC9C,GAAI,MAAM,QAAQ,GAChB,KAAM,IAAI,WACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,GAEhB,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,OAGvD,KAAK,OAAO,KAAK,GACjB,KAAK,MAAQ,IAQf,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,OAAO,SAAW,EACzB,KAAM,IAAI,WAAU,qCAItB,GADA,KAAK,OAAO,MACR,KAAK,OAAO,SAAW,EACzB,KAAK,QAAU,GACf,KAAK,aAAe,GACpB,KAAK,cAAgB,QAErB,GAAM,GAAiB,KAAK,OAAO,OAAS,EAC5C,KAAK,OAAO,GAAgB,cAAgB,GAC5C,KAAK,QAAU,CAAC,KAAK,OAAO,GAAgB,QAE5C,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,SAIzD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,KAAK,EAAQ,IAGjC,EAAA,UAAA,MAAA,SAAM,GAKJ,GAFA,GAAmB,GAEf,KAAK,OAAO,SAAW,GAAK,KAAK,QAAQ,SAAW,EACtD,KAAM,IAAI,WACN,4EAIN,KAAK,MAAQ,GAAI,IAAY,CAC3B,OAAQ,KAAK,OACb,QAAS,KAAK,QAAQ,GACtB,KAAM,KAAK,KAAO,WAEpB,KAAK,MAAM,UAAY,KAAK,UAG5B,KAAK,gBAAkB,KAAK,MAAM,gBAElC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,uBAAyB,KAAK,MAAM,uBACzC,KAAK,yBAA2B,KAAK,MAAM,yBAC3C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,wBAA0B,KAAK,MAAM,wBAC1C,KAAK,0BAA4B,KAAK,MAAM,0BAC5C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,WAAa,KAAK,MAAM,WAG7B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,WACE,MAAK,MAAK,OACR,KAAK,QAEA,EAAA,UAAM,YAAW,KAAA,OAiC1B,EAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAAA,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC9D,AAAK,KAAK,OACR,KAAK,QAEP,EAAA,UAAM,QAAO,KAAA,KAAC,EAAY,EAAW,IASvC,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,OAAS,MAChB,KAAK,QAEP,KAAK,MAAM,WAAW,IAmCxB,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GACF,GADE,AAAA,IAAA,QAAA,GAAA,IACE,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAO,MAAK,MAAM,SAAS,EAAG,EAAG,IAyB7B,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAA,CAAA,EAAO,KAAK,MAAM,gBAAgB,EAAS,SA8B7C,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC1B,AAAI,KAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,QAAQ,EAAG,IAU/B,EAAA,UAAA,eAAA,SAAe,GACb,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,eAAe,IAQnC,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,QACL,KAAK,MAAM,QAAQ,GACnB,KAAK,WAAa,KAAK,MAAM,UAE7B,KAAK,iBAAoB,KAAK,MAAc,iBAC5C,KAAK,KAAO,KAAK,MAAM,KACvB,KAAK,QAAU,KAAK,MAAM,QAG1B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,aAAe,KAAK,MAAM,cAIjC,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAO,OAAY,KAAK,MAAM,eAGrD,SAAc,GACZ,KAAK,MAAM,UAAY,mCAkCnB,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,IAAI,EAAG,EAAG,SAwFxB,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,WAAW,EAAS,SA0BlC,EAAA,UAAA,aAAN,SACI,EACA,sEAEF,MAAA,CAAA,EAAO,KAAK,MAAM,aAAa,EAAG,SAK7B,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GACA,EAA6C,GACjD,GAAI,YAAkB,QACpB,GAAI,CAAE,GAAO,GAAG,WAAa,OACzB,EAAO,GAAG,YAAiB,QAC7B,KAAM,IAAI,GAAW,kDAEvB,EAAc,MAEd,GAAA,KAAK,OACD,EAAO,QAAa,KACpB,WACI,MAAA,wHAER,EAAc,EAAO,OACrB,MAAO,GAAO,OACd,EAAmB,EAGrB,GAAM,GAAQ,GAAI,GAAI,GACtB,GAAI,CAAE,aAAiB,IACrB,KAAM,IAAI,IACN,yDAAyD,GAE/D,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAA0C,OAC1C,EAAQ,GACI,EAAkC,EAClC,GAClB,AAAI,GACF,EAAM,6BAA6B,IAErC,EAAM,IAAI,GAEZ,MAAO,IA+BT,OAAA,eAAI,EAAA,UAAA,eAAY,KAWhB,WACE,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,MAAO,MAAK,MAAM,kBAjBpB,SAAiB,GAGf,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,KAAK,MAAM,aAAe,mCAe5B,EAAA,UAAA,UAAA,WAME,OADM,GAAqC,GACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAiC,GACvC,EAAK,UAAe,EAAM,eAC1B,EAAK,OAAY,EAAM,YACvB,EAAO,KAAK,GAEd,MAAO,CAAC,KAAM,KAAK,KAAM,OAAM,IAvsB1B,EAAA,UAAY,aAysBrB,GA3sBgC,IA4sBhC,EAAA,cAAc,cAAc,IC3kC5B,YAqEsB,GACpB,MAAO,IAAI,IAAY,eA+DE,GACzB,MAAO,IAAI,IAAW,eA+FpB,EACA,GACF,MAAI,IAAW,MACb,GAAU,IAEL,GAAwB,EAAiB,eA0B5B,GACpB,MAAO,IAAM,eAIX,EACA,GACF,GAA4B,4BACxB,EAAgB,GC5QtB,AAwBA,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAEE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GALyC,EAAA,cAAc,cAWvD,GAAA,SAAA,GAAyB,EAAA,EAAA,GAAzB,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,GAAM,EAAG,IATF,EAAA,UAAY,MAW9B,GAbyB,IAczB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAK,EAAA,KAAS,OAF9B,EAAA,UAAY,QAI9B,GAN2B,IAO3B,EAAA,cAAc,cAAc,IAG5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAFO,EAAA,UAAY,SAI9B,GAN4B,IAO5B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,QAAY,IAFL,EAAA,UAAY,UAI9B,GAN6B,IAO7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAAjC,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAc,IAFP,EAAA,UAAY,cAI9B,GANiC,IAOjC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,SAAa,IAFN,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAW,IAFJ,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAeE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,QAAY,EAAG,IAdR,EAAA,UAAY,UAgB9B,GAlB6B,IAmB7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DAgBE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,WAAe,EAAG,IAfX,EAAA,UAAY,aAiB9B,GAnBgC,IAoBhC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,EAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAE,IAAI,IAAQ,IAAI,MATlC,EAAA,UAAY,QAW9B,GAb2B,IAc3B,EAAA,cAAc,cAAc,gBAEQ,GAClC,MAAO,GAAW,2BAIjB,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACM,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,MAChB,GAAM,GAAmC,GACzC,SAAO,UAAe,SACtB,EAAO,OAAY,GACZ,GAAsB,GAE/B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCjQjC,AAiBA,YAA0B,GACxB,GAAI,GAAQ,MAAQ,MAAO,IAAS,SAClC,KAAM,IAAI,OACN,mEACA,0BAAyB,IAOjC,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DAEA,MAAA,IAF0C,EAAA,cAAc,cAqBxD,GAAA,SAAA,GAA0B,EAAA,EAAA,GAQxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAEP,UAAiB,GAEjB,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,MAAQ,EAAK,KAAO,EACzB,EAAK,MAAQ,EAAK,KAAO,IAO3B,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,GAAyB,EAAA,MAAM,CAAC,IACpC,MAAI,GAAK,OACP,GAAiB,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAI,OAEhE,AAAI,EAAK,OACP,GACI,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,GAAS,OAEjD,EAAe,cAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,GAAM,KAAK,GAAI,GAAM,KAAK,KAI7B,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,CAAC,GAAI,EAAO,GAAiB,GAAI,EAAO,MA3ClD,EAAA,UAAY,OA6CrB,GA/C0B,IAgD1B,EAAA,cAAc,cAAc,gBAET,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,gBAGvC,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,IAOnD,GAAM,IACyC,CAChD,KAAQ,oBAGuB,GAEnC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGU,GAG7B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAuB,OACzB,OAAI,aAAsB,IACxB,EAEA,GAAuB,GClJlC,AAmCA,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAKxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,MACV,GAAK,SAAW,EAAK,YAIzB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,EAAS,GAAoB,GAC7B,GAAI,GAAS,EAAA,KAAK,GAClB,MAAI,MAAK,UAAY,MACnB,GAAS,EAAA,YAAY,EAAQ,EAAG,KAAK,WAEhC,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,SAAU,KAAK,UACnD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA5BF,EAAA,UAAY,OA8BrB,GAhC0B,IAiC1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAO7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,GAIvB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,UAAU,EAAG,KAAK,QAG3B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,YA4BrB,GA9B+B,IA+B/B,EAAA,cAAc,cAAc,IA6B5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAWzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAU/B,GAbO,EAAA,0BAAmD,QAI1D,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,gBAAkB,GACvB,EAAK,iBACD,GAAe,EAAK,kBAAoB,EAAK,2BACjD,EAAK,iBAAmB,GAAe,EAAK,kBAC5C,EAAK,gBAAkB,GAAc,EAAK,iBACtC,EAAK,YAAc,KACrB,EAAK,WAAa,aACT,MAAM,QAAQ,EAAK,YAC5B,EAAK,WAAa,EAAK,mBACd,MAAO,GAAK,YAAe,SACpC,EAAK,WAAa,CAAC,EAAK,gBAExB,MAAM,IAAI,GACN,8DACA,YAAW,EAAK,sBAIxB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAChC,GAAM,GAAoB,EAAW,MAAM,GAC3C,GAAI,KAAK,YAAc,KACrB,OAAgB,GAAA,EAAA,EAAA,KAAK,WAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAW,EAAI,GAAK,EAGxB,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAAM,KAAK,iBAEtC,GAAM,GAAiC,GACvC,GAAI,KAAK,YAAc,KACrB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,EAAK,GAAK,EAAW,GAGzB,KAAK,UAAY,CAAC,GAAI,IAAU,CAC9B,KAAM,EAAW,OACjB,KAAI,KAEN,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,SAAS,GAAoB,GACtB,EAAA,MAAM,EAAQ,KAAK,MAAM,SAGlC,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,iBAAkB,GAAqB,KAAK,kBAC5C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAoB,KAAK,iBAC1C,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAxEF,EAAA,UAAY,QA0ErB,GA5E2B,IA6E3B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAK/B,GARO,EAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGL,EAAK,OAAS,MAAQ,EAAK,QAAU,EAAK,cAC5C,KAAM,IAAI,IACN,4BAA4B,EAAK,MAAK,4CAI5C,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,IAAI,IAGb,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjCF,EAAA,UAAY,MAmCrB,GArCyB,IAsCzB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAOnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAE,IAAI,GAAK,EAAE,QAAQ,KAAK,OAAQ,aAG3C,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,kBA6BrB,GA/BqC,IAgCrC,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,aAAe,EAItB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,QAAU,GAAI,MAAoB,MACvC,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,OAG3D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,MAAK,QAAQ,EAAG,KAAK,OAG9B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,KAAM,KAAK,MAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,UA6BrB,GA/B6B,IAgC7B,EAAA,cAAc,cAAc,ICxV5B,YAwBI,EAAwB,EAAW,GACrC,GAAI,MAAO,IAAU,SACnB,MAAO,IAAa,EAAO,GAE3B,GAAI,EAAM,SAAW,EACnB,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EAAC,aACxD,eAAc,EAAM,OAAM,eAEhC,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAc,EAAM,GAC1B,GAAI,CAAC,GAAU,GACb,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EACvD,yBAAwB,KAAK,UAAU,GAAM,gBAC7C,wBAAuB,IAG/B,MAAO,eAaP,EAAqB,EAAoB,EACzC,EAAgB,GAClB,GADkB,AAAA,IAAA,QAAA,GAAA,GACd,GAAe,KACjB,MAAO,GAET,GAAM,GAAoB,EAAc,GAAa,GAAM,GAAW,GAClE,EACJ,MAAI,KAAY,OACd,EAAe,EAEf,EAAe,EAAc,EAAoB,EAE5C,KAAK,MAAO,GAAe,EAAS,GAAK,eAI9C,EAAiB,EAAoB,EACrC,GACF,GAAI,GAAW,KACb,MAAO,MAGT,GAAI,IAAY,QACd,EAAU,EAAU,EAAa,GAAI,CAAC,EAAa,EAAY,YACtD,IAAY,OACrB,EAAU,EAAU,MAEpB,MAAM,IAAI,GAAW,2BAA2B,EAAO,KAEzD,MAAO,GCpFT,YAwCI,EAAW,GAEb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,IAE3B,gBAWT,EAAW,GACb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,IAE9B,gBAsBT,EAAW,EAAgB,EAAc,EAAa,EACtD,EAAyB,GADgB,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,SAC7B,AAAA,IAAA,QAAA,GAAA,GACpB,EAAA,KAAK,WAMV,GALA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAEZ,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,+DACG,GAAE,MAAM,OAAM,cAEvB,GAAI,EAAO,MAAM,SAAW,EAC1B,KAAM,IAAI,GACN,iEACG,GAAO,MAAM,OAAM,aAE5B,GAAI,GAAQ,MAAQ,EAAK,MAAM,SAAW,EACxC,KAAM,IAAI,GACN,+DACG,GAAO,MAAM,OAAM,aAM5B,GAHA,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,KAE1B,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,GAAI,GAAY,EAAA,OACZ,EAA0B,EAAoB,EAC9C,IAAY,OAAS,OAAS,QAAS,MAAO,GAClD,MAAI,IAAQ,MACV,GAAI,GAAU,EAAG,IAEZ,gBAoDP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EAAyB,EAC5C,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,IACvD,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,gEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,iEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,MAAU,OAAO,CACnB,EAAG,EACH,OAAQ,EACR,QAAS,EACT,IAAK,IAAY,OAAS,OAAS,QACnC,UAAW,EACX,WAAY,OACZ,KAAI,EACJ,WAAU,IAEZ,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBA8BP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EACnB,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,EAAG,IAC1D,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,MAEf,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,oEACG,GAAE,KAAI,MAEf,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,OACA,EACA,EAAmC,EACnC,IAAY,OAAS,OAAS,QAAS,QAAS,GACpD,AAAI,GAAQ,MACV,GAAI,GAAU,EAAG,IAEnB,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAmHX,GAAA,IAAA,SAAA,GAAuC,EAAA,EAAA,GAwBrC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,IAAkB,KAIxB,GAVQ,EAAA,KAAsB,KAEvB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAIzD,EAAS,WAAW,GACpB,EAAK,KAAO,EACZ,GAAoC,EAAK,KAAM,QAC3C,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,EAAK,OAAS,EACtD,KAAM,IAAI,IACN,qDACI,EAAK,KAAI,6BAqBnB,GAlBA,EAAK,WAAa,GAAe,EAAK,WAAY,EAAM,cACxD,EAAK,QAAU,GACX,EAAK,SAAW,KAAO,EAAI,EAAK,QAAS,EAAM,WACnD,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,WAAa,GAAc,EAAK,YACrC,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAClD,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,aAAe,GAChB,EAAK,cAAgB,KAAO,EAAI,EAAK,aAAc,EACnD,gBACA,EAAK,OAAS,GACb,MAAM,QAAQ,EAAK,eAAiB,EAAK,aAAa,SAAW,EACpE,KAAM,IAAI,GACN,iGAEA,IAAG,KAAK,UAAU,EAAK,gBACtB,GAAI,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aAAe,CAAC,EAAK,aAAc,EAAK,sBACpC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,gEACA,8BAA6B,KAAK,UAAU,EAAK,wBAE9C,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aACD,CAAC,EAAK,aAAc,EAAK,aAAc,EAAK,sBACvC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,kEACA,8BAA6B,KAAK,UAAU,EAAK,yBAK1C,SAAA,WAAjB,SAA4B,GAI1B,GAFA,GACI,cAAgB,GAAM,2CACtB,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,oEACA,oCACI,KAAK,UAAU,EAAK,YAAW,OAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,gBAAiB,GAAqB,KAAK,iBAC3C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA/GuC,IAqHvC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAcjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAA0B,KAZ9B,SAAA,OAAwB,KAahC,EAAK,WAAW,GAChB,EAAK,QAAU,EAAK,QACpB,GAAoC,EAAK,QAAS,WAClD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,kBAAoB,GAAe,EAAK,qBAG/C,SAAA,UAAA,MAAA,SAAM,SACJ,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAE1B,GAAM,GAAW,EAAW,GAEtB,EAAc,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,UAE3D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,KAAM,KAAK,gBACnC,KAAK,gBAAiB,GAAM,KAAK,iBAGvC,KAAK,UAAY,CAAC,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,KACtE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GACE,EAAY,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACjD,EAAsB,GACU,EAAK,WAAW,gBAEtD,GAAI,GAAuB,MAAQ,EAAK,OAAS,EAC/C,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,aACtB,QAEJ,GAAI,EAAK,OAAS,EAChB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAQ,GACpD,EAAK,QAAS,EAAK,WAAY,EAAK,aAAa,YAC5C,EAAK,OAAS,EAEvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,sBACjB,EAAK,OAAS,EACvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,kBAE1B,MAAM,IAAI,IACN,yDAGN,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAIpC,MAAO,MAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAKhC,OAJM,GAAqB,GACrB,EAAS,KAAK,aAAe,eAC/B,EAAW,MAAM,EAAG,EAAW,OAAS,GACxC,EAAW,MAAM,GACZ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAM,GAAS,GACX,EAAM,GAAI,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,GACzD,MAAO,MAAK,cAAiB,SAAW,KAAK,aACL,KAAK,aAAa,IAC9D,EAAS,KAAK,GAGhB,GAAI,GAAc,CAAC,EAAW,IAC9B,MAAI,MAAK,aAAe,eACtB,GAAc,EAAY,OAAO,GACjC,EAAY,KAAK,KAAK,UAEtB,GAAY,KAAK,KAAK,SACtB,EAAc,EAAY,OAAO,IAE5B,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,kBAAmB,GAAqB,KAAK,mBAC7C,iBAAkB,GAAoB,KAAK,mBAEvC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,CAAE,YAAa,KAAS,MAAO,GAAK,SAAY,UAChD,EAAK,QAAU,EACjB,KAAM,IAAI,GACN,kEACA,YAAW,KAAK,UAAU,EAAK,YAGzC,GA1ImC,IA4InC,GAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAK,MAAO,GAAK,YAAe,UAC5B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,gCAA+B,KAAK,UAAU,EAAK,YAAW,OAnB/D,EAAA,UAAY,SAsBrB,GAxB4B,IAyB5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UACzB,CAAE,OAAM,QAAQ,EAAK,aAClB,GAAK,WAAW,SAAW,GAAK,EAAK,WAAW,SAAW,IAChE,KAAM,IAAI,GACN,mDACA,4CACI,KAAK,UAAU,EAAK,YAAW,OApBtC,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAGX,GAFA,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KAEnC,EAAK,UAAY,QAAU,EAAK,UAAY,QAC9C,KAAM,IAAI,GACN,gEACA,2CAA0C,EAAK,mBAIvD,SAAA,UAAA,MAAA,SAAM,SAGJ,GAFA,EAAa,GAAmB,GAE5B,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,mDACA,KAAK,UAAU,IAGrB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,wEAGN,GAAM,GAAW,EAAW,GACtB,EAAc,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,IAE1D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,UAAW,KAAK,kBACvC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,GAAM,KAAK,iBAIvC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MAC3D,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAM,MAAM,SAAW,EACzB,KAAM,IAAI,GACN,iEACA,8BAA6B,EAAM,MAAM,SAG/C,GAAM,GAAa,EAAM,MACnB,EAAY,EAAW,GAEzB,EACA,EACJ,AAAI,EAAK,aAAe,gBACtB,GAAQ,EACR,EAAQ,GAER,GAAQ,EACR,EAAQ,GAGV,GAAM,GAAS,EAAW,GACpB,EAAQ,EAAW,GACnB,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,QAAQ,GACvB,EAAU,EAAK,QAAQ,GAGvB,EAAY,GAAa,EAAQ,EAAS,EAAS,EAAK,SACxD,EAAW,GAAa,EAAO,EAAS,EAAS,EAAK,SAMtD,EACF,CAAC,EAAW,EAAW,EAAU,EAAK,SAE1C,AAAI,EAAK,aAAe,gBACtB,GAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,KAEzC,GAAI,GAAU,EAAA,gBACV,EAAmB,EAAK,OAAO,OAAoB,EACnD,EAAK,QAA6B,EAAK,SAC3C,MAAI,GAAK,aAAe,gBACtB,GAAU,EAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,KAG7C,AAAI,EAAK,MAAQ,MACf,GACI,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEhD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAE3B,EACA,EACA,EACJ,AAAI,KAAK,aAAe,gBACtB,GAAc,EACd,EAAa,EACb,EAAY,GAEZ,GAAc,EACd,EAAa,EACb,EAAY,GAGd,GAAM,GAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,QAAQ,GACvB,EAAU,KAAK,QAAQ,GAE7B,SAAY,GAAe,KAAK,QAChC,EAAY,GACR,GAAa,EAAY,GAAa,EAAS,EAAS,KAAK,SACjE,EAAY,GACR,GAAa,EAAY,GAAY,EAAS,EAAS,KAAK,SACzD,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,aACP,GA9IF,EAAA,UAAY,kBAgJrB,GAlJqC,IAmJrC,EAAA,cAAc,cAAc,IA0C5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAqBjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAAO,KAEnB,GAXO,EAAA,8BACL,gBACK,EAAA,8BACL,gBAEM,EAAA,gBAAiC,KACjC,EAAA,gBAAiC,KAKrC,EAAO,SAAW,KACpB,KAAM,IAAI,GACN,uFAGN,GAAI,EAAO,mBAAqB,MAAQ,EAAO,mBAAqB,MAChE,EAAO,kBAAoB,KAC7B,KAAM,IAAI,GACN,sPAKN,GAAI,EAAO,SAAW,MAAQ,EAAO,UAAY,QAC7C,EAAO,UAAY,QACrB,KAAM,IAAI,GACN,gBAAgB,EAAK,KAAI,kCACzB,qCAAoC,KAAK,UAAU,EAAO,WAGhE,SAAK,gBACD,EAAO,iBAAmB,KAAO,EAAI,EAAO,gBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,qBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,uBAGlD,SAAA,UAAA,MAAA,SAAM,SAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,KAAK,KAAO,EAClC,KAAM,IAAI,GACN,0BAA0B,KAAK,KAAI,sBAChC,MAAK,KAAO,EAAC,gCAChB,IAAG,KAAK,UAAU,KAExB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,0DACA,cAAa,KAAK,UAAU,EAAW,MAO7C,OAJM,GAAW,EAAW,GACtB,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,kBACrC,EAAuB,GACpB,EAAI,EAAG,EAAI,KAAK,KAAM,EAAE,EAC/B,EAAqB,KAAK,GAE5B,EAAqB,KAAK,EAAW,KAAK,gBAAiB,KAAK,SAEhE,GAAM,GAAY,GAClB,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,EAAW,KAAK,gBAE1C,KAAK,KAAO,KAGd,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MACvE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAE7B,GAAI,GACJ,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,IACN,oDACC,MAAI,GAAK,OAAS,GACvB,CAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAG3C,EAAS,EAAA,gBACL,EAAoB,EAAK,gBAAgB,OACzC,EAAK,gBAAgB,OACrB,EAAK,QAA6B,EAAK,QACvC,EAAK,aAAkC,SAG7C,AAAI,EAAK,SACP,GAAS,GAAU,EAAQ,EAAK,KAAK,OAAQ,EAAK,aAEpD,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAGjC,AAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAEpC,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,kBACd,MAAO,GAAO,kBACd,MAAO,GAAO,iBACd,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,qBAC7B,EAAO,oBACH,GAAoB,KAAK,qBACtB,GAzJF,EAAA,UAAY,gBA2JrB,GA7JmC,IA+JnC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,SACV,GAAA,KAAA,KAAM,EAAG,IAAK,KAFT,SAAA,UAAY,kBAIrB,GANqC,IAOrC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,GAClB,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,WACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,2BAA0B,KAAK,UAAU,EAAK,YAAW,OArB1D,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAgC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAM9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,MAAI,OAAO,GAAK,UAAa,SAC3B,EAAK,SACD,CAAC,CAAC,EAAK,SAAU,EAAK,UAAW,CAAC,EAAK,SAAU,EAAK,WACrD,AAAI,MAAO,GAAK,SAAS,IAAO,SACrC,EAAK,SAAW,CACd,CAAC,EAAK,SAAS,GAAI,EAAK,SAAS,IACjC,CAAC,EAAK,SAAS,GAAc,EAAK,SAAS,KAG7C,EAAK,SAAW,EAAK,SAEvB,EAAK,WACD,EAAK,aAAe,OAAY,eAAiB,EAAK,WAC1D,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAI,MAAK,aAAe,gBACf,CACL,EAAW,GAAI,EAAW,GAC1B,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,IAGlD,CACL,EAAW,GACX,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GAAI,EAAW,KAK5E,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAFA,EAAS,GAAoB,GAEzB,EAAK,aAAe,gBACtB,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,QAEjE,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,OAKvE,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,SAAU,KAAK,SAAU,WAAY,KAAK,YACpD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAhEF,EAAA,UAAY,aAkErB,GApEgC,IAqEhC,EAAA,cAAc,cAAc,IAwB5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAOhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KALM,SAAA,aAAe,CAAC,EAAG,GAMpC,EAAK,UAAY,CAAC,CAAC,KAAM,IACzB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,aAGtD,SAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,KAAK,aAAe,iBACtB,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,QAE9C,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAO,EAAW,MAIrD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAC1B,EAAa,EAAM,MAEzB,GAAI,EAAK,aAAe,iBACtB,EAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,IACvC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GAClC,EAAU,EAAM,sBAAsB,CAAC,EAAQ,IACrD,MAAO,GAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,SAExC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GACxC,MAAO,GAAM,sBAAsB,CAAC,EAAQ,QAKlD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,KAAM,KAAK,KAAM,WAAY,KAAK,YAC5C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlDF,EAAA,UAAY,eAoDrB,GAtDkC,IAuDlC,EAAA,cAAc,cAAc,ICrvC5B,YA6CI,EAAW,EAAyB,EACpC,EAAmB,EACnB,GAFoC,MAAA,KAAA,QAAA,GAAA,CAA6B,EAAG,IACpE,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WACV,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAChB,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,OAEf,GAAI,EAAgB,OAAS,EAC3B,KAAM,IAAI,GACN,yDACG,GAAgB,KAAI,OAE7B,SAAI,EAAA,gBACA,EAAe,EAA6B,EAC5C,IAAY,OAAS,OAAS,QAAS,OAAQ,GACnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,IAsCX,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAsB,KAHzB,SAAA,gBAAiC,KAIvC,EAAK,gBACD,EAAK,iBAAmB,KAAO,EAAI,EAAK,gBAC5C,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,4BACtC,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,qBAAuB,GAAe,EAAK,wBAGlD,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,iDACA,0BAAyB,KAAK,UAAU,GAAW,MAEzD,GAAM,GAAc,KAAK,aAAe,gBAAkB,EAAI,EAC9D,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,iEACA,4BAA2B,EAAW,GAAY,OAExD,GAAM,GAAW,EAAW,GACtB,EAA8B,CAClC,KAAK,WAAW,GAAI,KAAK,WAAW,GAAI,EAAU,KAAK,iBAGzD,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,EAAW,KAAK,iBAAkB,KAAM,KAAK,gBACtD,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GAAU,GACV,EAAQ,EAAK,gBAAgB,OAAQ,EAAK,QAC1C,EAAK,QAAS,EAAK,WAAY,MAEnC,MAAI,GAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEtD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EAAa,KAAK,aAAe,gBACnC,EAAW,GAAK,KAAK,gBACrB,EAAW,GAAK,KAAK,gBACnB,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACnD,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,MAAI,MAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAY,EAAS,GAGrC,CAAC,EAAW,GAAI,EAAS,EAAS,IAI7C,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,SAAO,gBAAqB,KAAK,gBACjC,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,sBACtB,GAjGF,EAAA,UAAY,kBAmGrB,GArGqC,IAsGrC,EAAA,cAAc,cAAc,IClN5B,YA0DI,EACA,EACA,EACA,GAKF,GAAI,MAAM,QAAQ,IAChB,GAAI,GAAgB,MAAQ,GAAa,KACvC,KAAM,IAAI,GACN,iFAGN,AAAI,GAAgB,MAClB,GAAY,EAAO,MAAM,EAAO,OAAS,EAAc,EAAO,QAC9D,EAAS,EAAO,MAAM,EAAG,EAAO,OAAS,IAE3C,AAAI,EAAO,OAAS,GAClB,GAAe,EAAO,MAAM,EAAG,EAAO,SAExC,EAAS,EAAO,GAGlB,WAAsB,GAEpB,MAAI,IAAK,MAAQ,MAAM,QAAQ,GACtB,EAEA,CAAC,GAIZ,SAAe,EAAa,GAC5B,EAAY,EAAa,GAElB,CAAC,OAAM,EAAE,aAAY,EAAE,UAAS,eA+CrC,EAA+B,EAAgB,EAC/C,EAAqB,EAAe,EAAsB,EAC1D,GADA,MAAA,KAAA,QAAA,GAAA,IAA0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,IACK,EAAA,KAAS,WACd,GAAM,GAAO,EAAO,MAAM,OAC1B,GAAI,EAAO,EACT,KAAM,IAAI,GAAW,uCAAuC,EAAI,MAKlE,GAAM,GAAO,CAAC,EAAG,GAAG,OAAO,GAAiB,EAAG,IAG/C,GAFA,EAAS,EAAA,UAAc,EAAQ,GAE3B,GAAa,KACf,KAAM,IAAI,IACN,kFAKN,AAAI,GACF,QAAQ,KACJ,qGAIN,AAAI,GAAQ,MACV,GAAO,EAAK,OAAO,QAAQ,OAAO,WAClC,AAAI,EAAK,OAAS,EAAO,GACvB,GAAO,EAAA,WAAe,EAAM,KAE9B,EAAO,EAAA,UAAc,EAAM,IAG7B,AAAI,GACF,GAAS,EAAA,QAAY,EAAQ,GAC7B,AAAI,GAAQ,MACV,GAAO,EAAA,QAAY,EAAM,KAc7B,GAAM,GAA2B,GAC7B,EACA,EAAS,EACP,EAAY,EAAO,MAAM,GACzB,EAAgB,EAAA,QAAY,GAC9B,EACJ,AAAI,GAAQ,MACV,GAAe,EAAA,QAAY,IAG7B,mBAAS,GACP,GAAM,GAAe,EAAc,GAC7B,EAAc,EAAA,KAAS,WAAM,MAAA,GAAa,EAAc,KAE9D,GAAI,GAAQ,KACV,EAAa,EAAY,GACzB,EAAS,EAAY,QAErB,GAAM,GAAgB,EAAA,KAAS,WAC7B,GAAM,GAAW,EAAa,GACxB,EAAc,EAAA,SAAa,GAAU,IAAI,GAEzC,EACF,EAAY,GAAG,IAAI,GAAU,IAAI,EAAO,GAAG,IAAI,IAC7C,EAAY,EAAO,IAAI,SAAC,EAAO,GACnC,MAAO,GAAY,GAAG,GAAG,IAAI,GAAU,IAAI,EAAM,IAAI,MAEvD,MAAO,CAAC,OAAM,EAAE,UAAS,KAE3B,EAAa,EAAc,OAC3B,EAAS,EAAc,UAGzB,AAAI,GACF,EAAe,KAAK,IAxBf,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,GA2BT,GAAI,GACJ,GAAI,GACF,GAAM,GAAO,EACb,EAAU,EAAA,MAAU,EAAgB,GAEtC,MAAO,CAAC,EAAY,EAAS,wBAyGR,EAAA,EAAA,GAqBvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACP,EACJ,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,GACN,wDAMN,GALO,AAAI,MAAM,QAAQ,EAAK,MAC5B,EAAO,GAAI,IAAgB,CAAC,MAAO,EAAK,OAExC,EAAO,EAAK,KAEV,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,qGAGN,SAAK,KAAO,EACZ,EAAK,gBACD,EAAK,iBAAmB,KAAO,GAAQ,EAAK,gBAChD,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,UAAY,EAAK,UAAY,KAAO,GAAQ,EAAK,SACtD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAQ,EAAK,OAEjD,EAAK,gBAAkB,GACvB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KACvC,EAAK,UAAY,KACjB,EAAK,QAAU,KAEf,EAAK,aAAe,KAIpB,EAAK,WAAa,KAKpB,SAAA,UAAA,UAAA,WACE,GAAI,KAAK,SAAW,MAClB,GAAM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EACtE,MAAO,IAAiB,EAAG,GAAW,IAAI,SAAA,GAAK,MAAA,YAE/C,OAAO,MAAK,SAMhB,EAAA,UAAA,UAAA,SAAU,GACR,KAAK,QAAU,GAGjB,EAAA,UAAA,mBAAA,SAAmB,GACjB,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAGb,GAAI,GAAY,KAAK,KAAK,UAC1B,AAAK,MAAM,QAAQ,IACjB,GAAY,CAAC,IAEf,GAAM,GAAY,EAAU,GACxB,EAOJ,GANA,AAAI,KAAK,gBACP,EAAc,CAAC,EAAW,GAAI,EAAW,GAAI,GAE7C,EAAc,CAAC,EAAW,GAAI,GAG5B,KAAK,aAEP,OADM,GAAsB,GACV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAW,KAAK,CAAC,EAAW,GAAI,IAElC,MAAO,CAAC,GAAa,OAAO,OAE5B,OAAO,IAIX,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAS,WACd,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAM,GAAa,EAAK,gBAAkB,EAAO,KAEjD,GAAI,EAAK,aACP,GAAM,GAAY,EAAK,OAAO,IAAI,SAAA,GAAK,MAAA,QACvC,MAAO,CAAC,GAAY,OAAO,OAE3B,OAAO,MAWb,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,SAAW,MAIlB,OAHM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EAChE,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,EAAO,KAAK,MAEd,MAAO,OAEP,OAAO,MAAK,aAIhB,SAAW,GACT,KAAK,QAAU,mCAGV,EAAA,UAAA,MAAP,SAAa,GAIX,GAAI,KAAK,cAAgB,KACvB,KAAM,IAAI,IACN,oDAGN,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAEb,GAAM,GAAoB,KAAK,SAAW,EAAW,GAAK,KACpD,EAAW,EAAW,MAAM,GAClC,KAAK,UAAU,GAAK,GAAI,IAAU,CAAC,MAAK,CAAG,EAAW,MAAI,OAAK,KAI/D,GAAM,GAAiB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAK7D,KAAK,KAAK,MAAM,GAIlB,GAAI,GAOJ,GANA,AAAI,MAAM,QAAQ,KAAK,KAAK,WAC1B,EAAY,KAAK,KAAK,UAEtB,EAAY,CAAC,KAAK,KAAK,WAGrB,KAAK,WAAa,MACpB,GAAI,CAAC,EAAA,KAAK,YACF,KAAK,UAAU,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,EAAK,MAAM,OAAS,KAC1D,GACN,KAAM,IAAI,GACN,0DACA,uCAAsC,KAAK,UAAS,MACpD,8BAA6B,KAAK,KAAK,gBAG7C,MAAK,UACD,EAAU,IAAI,SAAA,GAAO,MAAA,IAAI,IAAU,CAAC,MAAO,CAAC,KAAM,OAExD,AAAI,KAAK,UACP,KAAK,eAqBT,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAK,WACH,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAEN,GAAM,GAAY,EAAK,UAAU,GAAG,MAAM,GAC1C,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUAQN,GAAI,EAAK,SAAW,KAClB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAU,CAAC,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,qBAEzC,GAAU,KAEnB,EAAA,QAAY,EAAK,SAEjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAQ,GAAK,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,iBAMpD,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAER,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,IAAa,GAKf,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GACf,EAAM,MAAM,QAAQ,EAAK,KAAK,WAChC,EAAK,KAAK,UAAU,GACpB,EAAK,KAAK,UACR,EAAgB,CAAC,EAAW,GAClC,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAEhB,EAAK,QAAQ,GAAS,GAG1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAI5D,EAAA,UAAA,MAAA,SACI,EACA,GAEF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAGX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAC1D,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAMzB,GAAI,GAAiD,GACjD,EAA+B,GACnC,GAAI,GAAgB,MAClB,EAAO,aAAkB,EACzB,EAAmB,EAAiB,OAAO,GAC3C,KAAK,UAAY,GACjB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,UAAU,KAAK,GAAI,IAAU,CAAC,MAAO,EAAM,SAKlD,EAAkB,EAAgB,OAAO,KAAK,WAEhD,AAAI,GAAa,MACf,GAAO,UAAe,EACtB,EAAmB,EAAiB,OAAO,GAE3C,KAAK,aAAe,EAAU,QAGhC,GAAM,GAAW,EAAiB,YAAc,IAChD,GAAI,GAEF,GAAM,GACF,CAAC,GAAQ,OAAO,GACd,EAAgB,KAAK,UAAU,OAAO,GAEtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GAAS,EAAA,UAAM,MAAK,KAAA,KAAC,EAAW,GACtC,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAK/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KAIE,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC5C,EACA,GAAU,KAAO,KAAO,EAAO,aAEnC,EAAS,GAAoB,GAC7B,AAAI,GAAgB,MAClB,CAAI,EAAK,SACP,EAAe,EAAK,QAEpB,EAAe,EAAK,gBAAgB,IAIxC,GAAM,GACF,MAAM,QAAQ,EAAK,KAAK,WAAa,EAAK,KAAK,UAAU,OAAS,EACtE,GAAI,EAAa,SAAW,EAC1B,KAAM,IAAI,GACN,iBAAiB,EAAS,4BACvB,GAAa,OAAM,uBAE5B,AAAI,EAAK,QACP,QAAQ,KACJ,oEAGN,GAAM,GAAyB,CAAC,SAAQ,GAGlC,EAAO,SAAC,EAAgB,GAG5B,GAAM,GACF,EAAK,KAAK,KAAK,CAAC,GAAQ,OAAO,GAAS,GAE5C,MAAO,CAAC,EAAQ,GAAI,EAAQ,MAAM,KAK9B,EACF,GAAI,EAAM,EAAQ,EAAc,EAAK,YAAa,EAAM,KACpD,EAAK,OAAQ,EAAK,iBACpB,EAAa,EAAW,GACxB,EAAU,EAAW,GACrB,EAAS,EAAW,GAE1B,AAAI,EAAK,UACP,EAAK,YAAY,EAAQ,GAG3B,GAAM,GAAS,EAAK,gBAAkB,EAAU,EAIhD,MAAI,GAAK,YACA,CAAC,GAAQ,OAAO,GAEhB,KAKb,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAAI,GAAe,EAAA,MAAU,EAAO,OAKpC,MAHA,GAAe,EAAA,IAAQ,EAAc,CAAC,EAAG,IACzC,EAAe,GAAa,GAExB,MAAM,QAAQ,EAAK,KAAK,WACnB,EAAK,KAAK,UAAU,IACvB,SAAA,GAAO,MAAA,GAAM,EAAI,GAAO,EAAc,CAAC,EAAG,IAAQ,IAE/C,EAAK,KAAK,UAAY,EACzB,CAAC,GAAO,EAAc,CAAC,EAAG,EAAK,KAAK,aACpC,CAAC,MAKX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAK,MAAK,UAIH,KAAK,KAAK,iBAHR,oCAMX,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,MAAK,MAAK,UAGH,KAAK,KAAK,oBAFR,KAAK,KAAK,yCAKrB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,MAAQ,MACf,KAAK,KAAK,6BAA6B,IAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,gBAAiB,KAAK,gBACtB,YAAa,KAAK,YAClB,YAAa,KAAK,YAClB,SAAU,KAAK,SACf,OAAQ,KAAK,QAGf,AAAI,KAAK,cAAgB,MACvB,GAAO,aAAkB,KAAK,cAGhC,GAAM,GAAa,KAAK,KAAK,YAE7B,MAAI,MAAK,iBAAmB,EAAI,WAC9B,GAAO,KAAU,CACf,UAAa,KAAK,KAAK,eACvB,OAAU,IAKd,GAAA,GAAW,EAAe,EAAe,IAIpC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAa,EAAO,KACpB,EAAO,GAAY,EAAY,GACrC,MAAO,IAAI,GAAI,OAAO,OAAO,EAAQ,CAAC,KAAI,MArfrC,EAAA,UAAY,MAufrB,GAzfyB,IA0fzB,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAAtC,4DAQA,MAAA,IARsC,IAwFtC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAiCjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KANJ,SAAA,mBAAqB,OACrB,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAIzD,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,YAAc,KAAO,EAAK,mBAAqB,EAAK,YAC7D,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAG9B,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAEhC,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAW,EAAW,OAAS,GAAI,KAAK,OAAQ,KAC3D,KAAK,kBAAmB,KAAK,kBAAmB,GAChD,KAAK,kBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,OAAQ,KAC9C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IASf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8CAA8C,EAAO,OAAM,KAEjE,GAAI,GAAa,EAAO,GACxB,EAAS,EAAO,GAChB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,KAG/B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,KAGxC,GAAI,GACE,EAAiB,EAAK,YACtB,EAAoB,EAAK,qBAC/B,AAAI,GAAU,KACZ,EAAI,GAAM,EAAA,IAAQ,EAAQ,GAAS,EAAK,OAAO,QAE/C,EAAI,GAAM,EAAQ,EAAK,OAAO,QAEhC,AAAI,EAAK,MAAQ,MACf,GAAI,GAAU,EAAG,EAAK,KAAK,SAE7B,AAAI,GAAa,MACf,GAAa,EAAA,IAAQ,EAAY,IAEnC,GAAI,GAAS,EAAA,IAAQ,EAAG,GAAM,EAAY,EAAK,gBAAgB,SAC/D,MAAI,GAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAI1B,CAAC,EAAQ,MAIpB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,kBAGzB,MAAA,IAAA,GAAW,EAAe,IArKrB,EAAA,UAAY,gBAuKrB,GAzKmC,IA0KnC,EAAA,cAAc,cAAc,IA2F5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAG7B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAK,KAAO,GAAI,IAAc,GAC9B,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IA7BV,EAAA,UAAY,YA+BrB,GAjC+B,IAkC/B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAqC3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAbO,EAAA,mBAAqB,OACrB,EAAA,6BAAqD,cAErD,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAQrD,EAAK,WACP,KAAM,IAAI,GACN,+DAEN,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,KAAK,gBACrC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,uDACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACzD,EAAW,EAAO,GACtB,EAAS,EAAO,GAKhB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EAAY,EAAK,qBACnB,EACA,EACA,EAEJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAU,GAAM,EAAQ,EAAK,OAAO,QACxC,AAAI,EAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,SAEzC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAGzC,GAAM,GAAuB,EAAK,gBAAgB,OAC5C,EAAA,EAAA,MAAA,EAAA,CAAA,EAAA,EAAA,MAAA,EAAA,OAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAGN,EAAc,GAAM,EAAU,GAE9B,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GACT,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAE/C,GAAM,GAAa,GAAM,EAAA,IAAQ,EAAG,GAAW,GAC/C,EAAK,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,IAEvC,GAAM,GACF,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,IAAK,IAElE,MAAO,CAAC,EAAG,MAIf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,eACrB,WAAY,IAGd,MAAA,IAAA,GAAW,EAAe,IAvMrB,EAAA,UAAY,UAyMrB,GA3M6B,IA4M7B,EAAA,cAAc,cAAc,IA8B5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAQ,GACxB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,MAuCrB,GAzCyB,IA0CzB,EAAA,cAAc,cAAc,IAuC5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAsC5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAZJ,SAAA,mBAAqB,OACrB,EAAA,6BAA+B,cAC/B,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAEhC,EAAA,yBAA2B,QASlC,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,EAAK,eAE3B,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,CAAC,EAAK,MAAO,EAAK,OACnC,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,GAAI,GACJ,GAAI,KAAK,SACP,GAAI,KAAK,gBACP,GAAM,GAAmB,KAAK,gBACxB,EAAgB,KAAK,MAC3B,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GAAM,GAAK,EAAiB,MAAM,CAAC,IAC7B,EAAM,GAAI,MAAQ,MAAM,CAAC,IACzB,EAAS,EAAiB,MAAM,CAAC,EAAgB,IACvD,MAAO,IACH,GAAuB,EAAI,GAAK,IAExC,GAZgD,IAEvC,EAAA,UAAY,oBAYrB,GAAkB,KAAK,gBAEzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,EAAiB,KAAK,gBACtD,GAAM,KAAK,oBAEf,MAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,wDACG,GAAO,OAAM,MAEtB,GAAI,GAAW,EAAO,GAChB,EAAW,EAAO,GACxB,EAAS,EAAO,GAChB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EACF,EAAK,qBAKL,EACA,EACA,EACA,EACJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAI,GAAM,EAAQ,EAAK,OAAO,QAClC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAEzC,EAAI,EAAA,IAAQ,EAAG,GAAM,EAAU,EAAK,gBAAgB,SACpD,AAAI,EAAK,SACP,GAAI,GAAU,EAAG,EAAK,KAAK,SAGvB,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,KACnE,EAAI,EAAK,oBAAoB,MAAM,GAEnC,GAAM,GAAI,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,IAE3C,MAAO,CAAC,EAAG,EAAG,MAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,eAAgB,KAAK,eACrB,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,gBAGvB,MAAA,IAAA,GAAW,EAAe,IAnNrB,EAAA,UAAY,WAqNrB,GAvN8B,IAwN9B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAGxB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAS,GACzB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,OAuCrB,GAzC0B,IA0C1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAME,OADM,GAAsB,GACT,EAAA,EAAA,EAAA,KAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAU,KAAI,MAAd,EAAkB,EAAK,WAEvB,EAAU,KAAK,EAAK,WAGxB,MAAO,oCAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,EAKT,OAJI,GAAS,EAAO,MAAM,GAGpB,EAA2B,GACd,EAAA,EAAA,EAAA,EAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAa,KAAK,EAAO,OAAO,EAAG,EAAK,UAAU,SAElD,EAAa,KAAK,EAAO,OAAO,EAAG,IAGvC,EAAa,UAKb,OAFM,GAA8B,GAChC,EACK,EAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAO,EAAK,MAAM,GACxB,EAAS,EAAa,GAEtB,AAAI,IAAM,EACR,EAAa,CAAC,EAAO,IAAI,OAAO,GAEhC,EAAa,CAAC,EAAW,IAAI,OAAO,GAEtC,EAAa,EAAK,KAAK,EAAY,GACnC,EAAgB,KAAK,EAAW,MAAM,IAIxC,EAAS,GACT,OAAyB,GAAA,EAAA,EAAA,EAAgB,QAAQ,UAAxB,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAO,KAAI,MAAX,EAAe,GAEjB,MAAO,CAAC,EAAW,IAAI,OAAO,MAI3B,EAAA,UAAA,MAAP,SAAa,GACX,AAAI,GAAgB,IAGlB,GAAc,EAAuB,IAEvC,EAAa,EACb,GAAI,GACJ,KAAK,MAAM,QAAQ,SAAC,EAAM,GACxB,GAAU,WAAW,EAAK,WAGxB,EAAK,MAAM,GACX,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAY,EAAK,UAAU,GAE3B,EAAY,EAAK,UAEnB,EAAa,CAAC,EAAW,GAAI,OAGjC,KAAK,MAAQ,IAGf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAgB,SAAC,GACrB,MAAO,CACL,UAAa,EAAK,eAClB,OAAU,EAAK,cAIb,EAAc,KAAK,MAAM,IAAI,GAE7B,EAAS,CAAC,MAAS,GAEzB,MAAA,IAAA,GAAW,EAAe,IAIrB,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAElB,OADM,GAAmB,GACA,EAAA,EAAA,EAAC,EAAO,MAAR,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAM,KAAK,GAAY,EAAY,IAErC,MAAO,IAAI,GAAI,CAAC,MAAK,KAGvB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,kBAEvB,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,qBAEvB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACvB,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAiB,KAAI,MAArB,EAAyB,EAAK,kBAEhC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAQT,EAAA,UAAA,WAAA,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,SAEvB,MAAO,IAAc,IASvB,EAAA,UAAA,WAAA,SAAW,GAET,OADM,GAAyC,GAC5B,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,IAGjB,OAHS,GAAI,EAAA,GACP,EAAY,EAAK,QAAQ,OACzB,EAAe,EAAQ,OAAO,GAC3B,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,EACzC,EAAO,KAAK,CAAC,EAAK,QAAQ,GAAI,EAAa,KAG/C,GAAc,IA5KT,EAAA,UAAY,kBAgLrB,GAlLqC,IAmLrC,EAAA,cAAc,cAAc,gBAEQ,GAM3B,GAAA,GAAA,EAAA,KAAM,EAAA,EAAA,KAAM,EAAA,EAAA,SAAA,EAAA,IAAA,OAAA,GAAA,EAAkB,EAAA,EAAA,MAAA,EAAA,IAAA,OAAA,EAAA,EAE/B,EAAgB,WAAM,MAAA,IAAU,IAAQ,IAExC,EAAa,WAAM,MAAA,IAAe,EAAe,EAAM,IAG7D,GAAI,CAAC,GAAS,GAAS,EACrB,MAAO,GAAA,KAAS,IAAa,SAG/B,GAAM,GAAQ,MAAM,GAAO,KAAK,QAAW,IAAI,GAE/C,MAAO,GAAM,IAAI,SAAA,GAAK,MAAA,GAAA,KAAS,EAAE,WC/iEnC,AAiFA,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAArC,4DAyBA,MAAA,IAzBqC,IAiCrC,GAAA,SAAA,GAAwB,EAAA,EAAA,GAMtB,WAAY,GAAZ,GAAA,GAAA,KACE,GAAI,EAAK,OACP,KAAM,IAAI,IACN,sDAGN,GAAI,MAAM,QAAQ,EAAK,MACrB,KAAM,IAAI,IACN,kEAGN,SAAA,EAAA,KAAA,KAAM,IAAqB,KAE3B,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WAad,GAZA,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aAEtB,EAAK,KAAK,YAAc,MAG1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBAEtB,EAAK,KAAK,qBAAuB,MAG/B,GAAU,EAAO,UACnB,KAAM,IAAI,GAAW,6CAGvB,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KAEtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAE1C,EACF,GAAU,KAAO,KAAO,EAAO,aAEnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAI3D,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GAAkB,KAAK,yBAAyB,GAEpD,MAAK,MAAK,iBACR,GAAQ,CAAI,EAAS,IAAE,OAAK,EAAS,MAAM,KAG7C,AAAI,KAAK,aACP,GAAQ,CACH,GAAQ,OAAK,MAAM,GAAG,KAAI,CAAE,EAAW,IAAE,OAAK,EAAS,MAAM,QAG7D,GAGT,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACP,GAAA,GAAA,EAAA,KAAA,UAED,EAAa,EAAO,MAEpB,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAe,EAAA,MAAU,GAE/B,MAAI,OAAM,QAAQ,GACT,MAAM,EAAU,QAAQ,KAAK,GAG/B,CAAC,MAIZ,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAS,WACP,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAGN,GAAM,GAAa,EAAK,UAAU,GAAG,MAE/B,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAY,EAAW,GAE7B,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUASN,GAAI,EAAK,aAAe,KACtB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAU,CAAC,EAAA,MAAU,YAEnB,GAAU,KAEnB,EAAA,QAAY,EAAK,SAGjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAQ,GAAK,EAAA,MAAU,QAO9B,GAJA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAGR,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,EAKF,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GAEf,EAAgB,EAEtB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAGhB,EAAK,QAAQ,GAAS,GAI1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAIlD,EAAA,UAAA,yBAAV,SAAmC,GAC3B,GAAA,GAAA,KAAA,KAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,QAAS,EAAA,EAAA,aAGpD,EAAkB,IAAe,gBAEjC,EAAI,EAAW,EAAkB,EAAI,GACrC,EAAI,EAAW,EAAkB,EAAI,GAErC,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAClD,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAElD,EACD,EAAW,MAAM,EAAG,GAAE,OACrB,EAAkB,CAAC,EAAS,EAAM,GAAQ,CAAC,EAAM,EAAM,IAG7D,MAAO,IAhMF,EAAA,UAAY,YAkMrB,GApMwB,IAyMxB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAWlC,WAAY,GAAZ,GAAA,GAAA,KAEI,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,aAGF,SAAA,EAAA,KAAA,KAAA,GAAA,GAAU,EAAI,CAAE,MAAO,MAAS,KAEhC,EAAK,QAAU,EACf,GAAsB,EAAK,QAAS,WAEpC,EAAK,WAAa,GAAe,EAAY,EAAG,cAChD,EAAK,WAAW,QAAQ,SAAA,GAAQ,MAAA,IAAsB,EAAM,gBAE5D,EAAK,QAAU,GAAe,GAAW,EAAG,EAAG,WAC/C,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,IAAsB,EAAQ,aAE7D,EAAK,QAAU,GAAW,QAC1B,GAAiB,EAAK,SAEtB,EAAK,WAAa,GAAc,eAChC,GAAgB,EAAK,YAErB,EAAK,aAAe,GAAe,GAAgB,EAAG,EAAG,gBACzD,EAAK,aAAa,QACd,SAAA,GAAQ,MAAA,IAAsB,EAAM,oBAGnC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAEhC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAElE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAG1B,GAAM,GAAW,EAAW,GAEtB,EAAe,EAEf,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,QAAU,IAErD,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBAEvC,GAAM,GACF,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,KAAK,QAAU,IAOzD,GALA,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBAEL,KAAK,SACP,GAAI,GAAe,OAEnB,GAAI,KAAK,gBACP,GAAM,GAAO,KAAK,gBAEZ,EAAU,KAAK,QAErB,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAQ,EAAK,MAAM,CAAC,IACpB,EAAQ,EAAA,KAAS,CAAC,IAClB,EAAY,EAAK,MAAM,CAAC,EAAU,IACxC,MAAO,IAAc,CAAC,EAAO,EAAO,KAExC,GAVgD,IAEvC,EAAA,UAAY,oBAUrB,GAAkB,KAAK,gBAGzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,QAAU,GAAe,KAAM,EAC7C,KAAK,gBAAiB,GAAM,KAAK,gBAGvC,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAsB,GAA3B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8DACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,GAEjC,EAAI,EAAO,GACX,EAAW,EAAO,GAClB,EAAW,EAAO,GAElB,EAAe,EAIrB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAI9B,GAAM,GAAc,EAAK,YAEnB,EACF,SAAC,GAAe,GAAoB,IAClC,MAAI,CAAC,IAAQ,CAAC,GAAK,IACV,GAGF,EAAA,IAAQ,GAAK,IAAQ,KAG9B,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAEtC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAIvC,GAAM,GAAiB,EAAK,qBAExB,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAE1C,EAAoB,EAEpB,EAAA,EAAA,MAAA,EAAA,OAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAG5B,EAAA,EAAA,uDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAI5B,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAEvC,GAAA,GAAA,EAAA,MAAA,EAAA,gBAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAI3C,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAE5B,GAAM,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAA,IACN,EAAA,IAAQ,GAAG,GACX,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,MAC3C,GAAI,EAAA,IACN,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC3C,EAAK,WAAW,MAAM,KAE1B,MAAO,CAAC,GAAG,GAAG,OAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAA,EAAA,UAAA,UAAA,KAAA,MAAC,EAAA,EAAA,MAAY,EAAA,GAAA,EAAA,CAAA,UAEb,EAAuC,CAC3C,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,QAAS,KAAK,SAGhB,MAAA,IAAA,GAAW,EAAe,IAG5B,EAAA,UAAA,UAAA,SAAU,EAAW,EAAW,EAAY,GAC1C,GAAM,GAAM,EAAA,OACR,EAAmB,EAAmB,KAAK,QAC1C,GAAW,QACZ,KAAK,aAAe,gBAAkB,OAAS,OAC/C,KAAK,cAET,MAAI,GACK,GAAU,EAAK,EAAG,KAAK,YAGzB,GAGT,EAAA,UAAA,cAAA,SAAc,EAAW,GACvB,GAAM,GAAU,EAEhB,MAAO,GAAA,OACH,EAAmB,EAAmB,EAAS,OAC/C,KAAK,aAAe,gBAAkB,OAAS,SAzO9C,EAAA,UAAY,iBA2OrB,GA7OoC,IA+OpC,EAAA,cAAkB,cAAc,IAKhC,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,GAAI,IAAe,GAEhC,SAAA,EAAA,KAAA,KAAM,GAAA,GAAI,EAAI,CAAE,KAAI,MAAwB,OAIvC,SAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IAZV,EAAA,UAAY,aAcrB,GAhBgC,IAkBhC,EAAA,cAAkB,cAAc,ICjkBhC,AAkDA,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,KAAK,IAAI,KAAK,IAAI,EAAK,KAAM,GAAI,GAE7C,EAAK,WAAa,EAAK,WACvB,EAAK,KAAO,EAAK,KACjB,EAAK,gBAAkB,KAGf,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,KAAK,YAAc,KACrB,MAAO,MAAK,WAId,OAFM,GAAa,EAAM,MACnB,EAAoB,GACjB,EAAI,EAAG,EAAI,KAAK,WAAW,OAAQ,EAAE,EAC5C,EAAW,KACP,KAAK,WAAW,IAAM,KAAO,EAAW,GAAK,KAAK,WAAW,IAEnE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAI,EAAK,MAAQ,EAAK,KAAO,GAC/B,GAAM,GACF,EAAO,UAAe,KAAO,GAAQ,EAAO,SAC1C,EAAa,EAAK,cAAc,GAChC,EAAS,GACX,WAAM,MAAA,IAAU,EAAO,EAAK,KAAM,EAAY,EAAK,OACnD,WAAM,MAAA,IAAO,GACjB,MAAO,GAET,MAAO,MAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,KACX,WAAY,KAAK,WACjB,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAxDf,EAAA,UAAY,UA0DrB,GA5D6B,IA6D7B,EAAA,cAAc,cAAc,IA4D5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAIpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,CAAC,KAAM,MAGjB,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAM,GAAa,EAAM,MACzB,MAAO,CAAC,EAAW,GAAI,EAAG,EAAW,KAThC,EAAA,UAAY,mBAWrB,GAbsC,IActC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAmBzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAhBM,EAAA,WAA2B,KAC3B,EAAA,QAAU,GAGV,EAAA,OAAwB,KACxB,EAAA,KAAsB,KAErB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAQrD,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MACnD,EAAK,UAAY,MAGnB,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAK,gBAAkB,CAAC,EAAW,EAAK,UAG1C,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GAAc,EAAK,YACrC,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAK,SAEtB,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,gBAAkB,GAEvB,EAAK,UAAY,CAAC,CAAC,QAAS,MAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAe,EAAW,EAAW,OAAS,GACpD,AAAI,KAAK,QAAU,MACjB,MAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAc,KAAK,OAAQ,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,kBAIzC,KAAK,UAAY,CAAC,CAAC,QAAS,EAAG,KAAI,GAAA,GAAG,EAAC,IAAK,EAAY,KACxD,KAAK,MAAQ,IAGf,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,SAAY,EAAY,OAAS,GAAK,KAAK,MACpC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAM,GAAQ,GAAoB,GAC5B,EACF,GAA2B,EAAK,WAAW,gBAC3C,EAEJ,MAAI,IAAuB,KACzB,EAAS,GACL,EAAO,EAAK,OAAO,OAAQ,EAC3B,EAAK,KAAO,EAAK,KAAK,OAAS,MAEnC,GAAS,GAAM,EAAO,EAAK,OAAO,QAClC,AAAI,EAAK,MAAQ,MACf,GAAS,GAAU,EAAQ,EAAK,KAAK,SAEvC,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,KAI5B,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArHF,EAAA,UAAY,QAuHrB,GAzH2B,IA0H3B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAO,GAAQ,GACf,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,UAAY,CAAC,CAAC,QAAS,IAC5B,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,OAAkB,GAAA,EAAA,EAAA,EAAW,MAAM,GAAjB,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAO,KACT,KAAM,IAAI,GACN,4DACA,SAAQ,EAAW,MAAM,GAAE,oCAC3B,mFAIR,MAAO,CAAC,EAAW,GAAI,GAAU,EAAY,KAG/C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAK,aAAe,iBAAmB,EAAM,KAAO,GAEtD,OADM,GAAwB,CAAC,GACtB,EAAI,EAAG,EAAI,EAAM,KAAM,EAAE,EAChC,EAAY,KAAK,GAEnB,EAAY,KAAK,GACjB,EAAQ,EAAM,UAAU,GAG1B,MAAO,IAAe,MAI1B,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,GACzC,AAAI,KAAK,YAAc,MACrB,GAAO,WAAgB,KAAK,YAE9B,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA/CF,EAAA,UAAY,UAiDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAK9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,WAAa,GAAc,EAAK,cAGvC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAK,WAAW,MAAM,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,GAAoB,KAAK,aAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArBF,EAAA,UAAY,aAuBrB,GAzBgC,IA0BhC,EAAA,cAAc,cAAc,IAc5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAKhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,EAAI,EAAK,EACd,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,KAAK,EAAG,EAAW,KAG5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAS,GAAoB,GACtB,GAAS,EAAQ,EAAK,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,EAAG,KAAK,GAEJ,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,eA4BrB,GA9BkC,IA+BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,YAAc,EAAK,YAGxB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,EAAE,EAC7C,AAAI,EAAK,UAAU,EAAK,YAAY,KAClC,GAAK,YAAY,GAAK,eAKpB,SAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,GAAM,GAAK,GAAO,MAiBnB,EAAA,UAAA,oBAAR,SAA4B,EAAmB,GAK7C,OAJM,GAAW,6CACX,EAAa,EAAY,QAC3B,EAAQ,EACR,EAAU,KACL,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAM,EAAW,GACvB,GAAI,KAAK,UAAU,GACjB,GAAI,IAAY,KACd,EAAU,MAEV,MAAM,IAAI,GAAW,gDAGvB,IAAS,EAIb,GAAM,GAAe,GAAU,GAC/B,GAAI,IAAY,MACd,GAAI,IAAU,GAAK,EAAe,IAAU,EAC1C,KAAM,IAAI,GAAW,GAEvB,EAAW,GAAW,EAAe,UAC5B,IAAiB,EAC1B,KAAM,IAAI,GAAW,GAGvB,MAAO,IAGT,EAAA,UAAA,mBAAA,SAAmB,GAEjB,OADI,GAAiB,GACZ,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,GAAI,KAAK,UAAU,EAAW,KAC5B,EAAiB,GACjB,MAIJ,MAAI,GACK,EAAW,MAAM,EAAG,GAAG,OAAO,KAAK,aAEnC,EAAW,MAAM,EAAG,GAAG,OAC1B,KAAK,oBAAoB,EAAW,MAAM,GAAI,KAAK,eAI3D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAc,EAAW,MAAM,EAAG,GAAG,OACvC,EAAK,oBAAoB,EAAW,MAAM,GAAI,EAAK,cACvD,MAAO,GAAM,QAAQ,MAIzB,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlGF,EAAA,UAAY,UAoGrB,GAtG6B,IAuG7B,EAAA,cAAc,cAAc,IAY5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAM3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,OACN,mFAGN,GAAI,CAAC,MAAM,QAAQ,EAAK,MACtB,KAAM,IAAI,OACN,oEACG,GAAK,KAAI,cAIlB,GAAM,GAAwB,GAAM,EAAG,EAAK,KAAK,OAAS,GAC1D,GAAI,CAAC,EAAA,KAAK,YAAY,EAAK,KAAK,QAAQ,OAAQ,GAC9C,KAAM,IAAI,OACN,+BAAiC,KAAK,UAAU,EAAK,MACrD,8DAGN,SAAK,KAAO,EAAK,KACjB,EAAK,mBAAqB,CAAC,GAAG,OAAO,EAAK,MAC1C,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,EAAK,KAAK,OAAS,OAG5D,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,YAAK,KAAK,QAAQ,SAAC,EAAa,GAC9B,EAAY,EAAI,GAAM,EAAqB,KAEtC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,UAAU,GAAoB,GAAS,KAAK,qBAGrD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjDF,EAAA,UAAY,UAmDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,KACV,EAAK,UAAY,EAAK,WAAa,KAAO,EAAI,EAAK,UAEnD,EAAK,UAAY,IAIrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,UAAW,KAAK,WAChC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACb,MAAO,GAAA,IAAI,EAAA,SAAS,EAAO,KAAK,WAAY,IAG9C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACP,EAAW,GACX,EAAc,EAAA,IAAI,EAAA,SAAS,EAAO,EAAK,WAAY,EAAM,GACzD,EAAS,EAAM,IAAI,EAAY,OAAO,EAAM,QAClD,MAAO,MAtCJ,EAAA,UAAY,UAyCrB,GA3C6B,IA4C7B,EAAA,cAAc,cAAc,IClqB5B,AA2EA,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAgB7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GATM,EAAA,WAA4B,KAE3B,EAAA,+BACL,gBAME,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MAKrD,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,AAAI,EAAK,aAAe,KAGtB,EAAK,gBAAkB,CAAC,EAAW,MAInC,EAAK,gBACD,CAAC,GAAW,OAAO,GAAqB,EAAK,cAGrD,SAAK,SAAW,EAAK,SACrB,GAAoC,EAAK,SAAU,YACnD,EAAK,UAAY,EAAK,UACtB,GAAoC,EAAK,UAAW,aACpD,EAAK,sBAAwB,GACzB,EAAK,uBAAyB,EAAK,gCACvC,EAAK,sBAAwB,GAAe,EAAK,uBACjD,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,qBAAuB,GAAc,EAAK,sBAC/C,EAAK,SAAW,EAAK,SACrB,EAAK,gBAAkB,EAAK,SAC5B,EAAK,YAAc,EAAK,cAGnB,SAAA,UAAA,MAAP,SAAa,GACX,KAAK,WAAa,KAAK,UACnB,aAAc,CAAC,KAAK,SAAU,KAAK,WAAY,KAAK,MACpD,KAAK,sBAAuB,KAAK,sBAAuB,GACxD,KAAK,sBACT,KAAK,MAAQ,IAKL,EAAA,UAAA,6BAAV,SAAuC,KAEvC,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAK,GAAK,SAGR,GAAS,GAAoB,GACtB,EAAA,SAAS,EAAQ,EAAA,UAAU,KAH3B,QAQb,EAAA,UAAA,mBAAA,SAAmB,GAEjB,GADA,EAAa,GAAmB,GAC5B,KAAK,aAAe,KACtB,MAAW,GAAU,OAAA,CAAE,KAAK,YAG9B,GAAM,GAAmB,GAAqB,KAAK,aACnD,GAAI,EAAO,SAAW,EAAW,OAAS,EACxC,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IAG7B,OADI,GAAI,EACC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAK,EAAO,GACZ,EAAK,EAAW,EAAI,GAC1B,GAAK,GAAM,MAAU,GAAM,MAAU,IAAO,EAC1C,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IACxB,AAAI,GAAM,MACf,GAAO,GAAK,GAEd,IAGJ,MAAA,CAAQ,EAAW,IAAE,OAAK,EAAM,CAAE,KAAK,aAGzC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,AAAI,EAAM,QAAU,SAClB,GAAQ,GAAO,EAAO,UAExB,GAAM,GAAS,GAAS,EAAK,WAAW,OAAQ,EAAM,QACtD,MAAO,GAAO,QACV,GAAmB,EAAK,mBAAmB,EAAM,YAIzD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,UAAW,KAAK,UAChB,sBAAuB,GAAqB,KAAK,uBACjD,sBAAuB,GAAqB,KAAK,uBACjD,oBAAqB,GAAqB,KAAK,qBAC/C,qBAAsB,GAAoB,KAAK,sBAC/C,SAAU,KAAK,SACf,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjIF,EAAA,UAAY,YAmIrB,GArI+B,IAsI/B,EAAA,cAAc,cAAc,ICjN5B,AA+BA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAGlC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAQ,KAAG,KACjB,SAAK,gBAAkB,KAOf,SAAA,UAAA,cAAV,SAAwB,GACtB,KAAM,IAAI,KAaJ,EAAA,UAAA,gCAAR,SAAwC,EAAe,GACrD,GAAI,GAAU,MAAQ,GAAU,KAC9B,MAAO,MACF,GAAI,EAAO,OAAS,EAAO,OAChC,MAAO,MAAK,gCAAgC,EAAQ,GAC/C,GAAI,EAAO,SAAW,EAC3B,MAAO,GAGT,OADM,GAAqB,EAAO,MAAM,EAAG,EAAO,OAAS,EAAO,QACzD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,EAAO,OAAS,EAAO,OAAS,GAC3C,EAAI,EAAO,GACjB,GAAI,GAAK,MAAQ,GAAK,MAAQ,EAAI,GAAK,EAAI,EACzC,EAAY,KAAK,cACR,IAAM,EACf,EAAY,KAAK,WACR,IAAM,EACf,EAAY,KAAK,QAEjB,GAAI,IAAM,EACR,KAAM,IAAI,GACN,wDACA,KAAK,UAAU,GAAU,IAAM,KAAK,UAAU,IAEpD,EAAY,KAAK,IAGrB,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GAOJ,GALA,AAAI,MAAM,QAAQ,IAAe,CAAC,MAAM,QAAQ,EAAW,KAEzD,GAAa,CAAC,GAAmB,KAEnC,EAAa,EACT,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,mEACA,SAAQ,EAAW,OAAM,eAM/B,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAI1B,GADA,EAAa,GAAqB,GAC9B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,qDACA,6BAA4B,KAAK,UAAU,GAAW,MAK5D,OAFI,GACA,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GAC9C,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,GAAM,GAAW,EAAW,IAAI,SAAA,GAAS,MAAA,GAAM,SAC/C,AAAI,EAAW,QAAQ,QAAU,IAC7B,GAAqB,GAAU,SAAW,EAC5C,KAAK,gBAAkB,GAEvB,KAAK,gBAAkB,IAI3B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAK,iBACP,GAAM,GAA2B,GAC3B,EAAY,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,OAC5C,GAAI,EAAU,QAAQ,QAAU,IAI9B,OADM,GAAU,GAAc,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAEZ,OAFO,GAAC,EAAA,GACF,EAAQ,EAAE,KACP,EAAI,EAAG,EAAI,EAAU,EAAO,EAAE,EACrC,EAAI,GAAa,EAAG,GAEtB,EAAe,KAAK,GAEtB,MAAO,GAAK,cAAc,QAK1B,OADI,GAAa,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,KAChB,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAY,EAAO,GACnB,EAAW,EAAO,MAAM,GAAG,OAAO,CAAC,IACrC,EAAc,EAAE,QAChB,CAAC,GAAW,OAAO,GAAoB,EAAO,MAAM,MACxD,EAAc,EAAA,UAAc,EAAa,CAAC,EAAG,IAC7C,EAAc,EAAY,QAAQ,GAClC,EAAe,KAAK,GACpB,EAAa,WACJ,EAAQ,GACjB,GAAM,GAAO,GAAgB,EAAG,GAAO,OAAO,CAAC,IAC/C,EAAe,KAAK,EAAA,UAAc,EAAG,IACrC,EAAa,OAGb,GAAe,KAAK,GAGxB,GAAI,GAAI,EAAK,cAAc,GACrB,EAAQ,EAAE,KAChB,GAAI,GAGF,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAQ,EAAO,OACf,EAAY,EAAO,EAAQ,GAC3B,EACF,CAAC,GAAW,OAAO,EAAO,MAAM,EAAG,EAAO,OAAS,IACvD,EAAI,EAAA,UAAc,EAAE,QAAQ,CAAC,GAAI,IAAa,CAAC,EAAG,IACzC,QAAQ,WACR,EAAQ,GACjB,GAAM,GAAO,CAAC,EAAQ,GAAG,OAAO,GAAgB,EAAG,EAAQ,IAC3D,EAAI,EAAA,UAAc,EAAG,IAGzB,MAAO,QAGT,OAAO,GAAK,cAAc,MAKhC,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,EACb,GAAI,GACJ,AAAI,EAAW,IAAM,KACnB,EAAc,KAEd,EAAc,EAAW,GAAG,MAAM,GAEpC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAG1B,SAAa,GAAqB,GAClC,AAAI,EAAW,SAAW,EACxB,EAAc,EAAW,OAAO,GAEhC,EAAc,CAAC,MAAM,OAAO,GAEvB,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6BAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+BAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,kGAEA,KAAI,EAAO,OAAM,OAAO,EAAK,OAAM,MAEzC,GAAI,EAAK,MAAM,SAAA,GAAK,MAAA,IAAK,OACvB,MAAO,MAET,EAAO,EAAK,IAAI,SAAA,GAAK,MAAA,IAAK,KAAO,EAAI,EAAA,WAAe,EAAG,KAEvD,OADI,GAAS,EAAK,GACT,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAS,EAAA,WAAe,EAAQ,EAAK,IAEvC,MAAO,MAGb,GAlOoC,IAoOpC,GAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,MAcrB,GAhByB,IAiBzB,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAG5B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,WAcrB,GAhB8B,IAiB9B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,GAAA,IAAQ,EAAI,EAAO,OAAQ,MAX/B,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA2D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IAiE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAM/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAJJ,SAAA,aAAe,GAKtB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,MACxD,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,yEAGN,EAAa,EAGb,OADI,GAAe,GACC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,GAAS,MACX,EAAe,GACf,OAGJ,GAAI,EACF,OAIF,OADM,GAAoB,GACjB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAyB,EAAW,GAAG,QAC7C,EAAuB,OAAO,KAAK,KAAM,GAEzC,OADI,GAAS,GACO,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAA,KAAK,YAAY,EAAO,IAC1B,EAAS,GACT,OAGJ,AAAK,GACH,EAAS,KAAK,GAGlB,GAAI,EAAS,OAAS,EACpB,KAAM,IAAI,GACN,4GAEA,KAAK,UAAU,KAIb,EAAA,UAAA,cAAV,SAAwB,GAAxB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAO,IAAc,EAAQ,EAAK,SAItC,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,KAC1D,KAAM,IAAI,GACN,+DAON,OALM,GAAc,EACd,EAAc,EAAY,GAAG,QAC7B,EAAO,KAAK,KAAO,EAAI,EAAY,OAAS,KAAK,KAAO,KAAK,KAG/C,EAAA,EAAA,EAAA,EAAY,MAAM,GAAlB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAY,IAAS,MAAQ,EAAM,IAAS,MAC9C,EAAY,GAAQ,KACpB,MAEF,EAAY,IAAS,EAAM,GAE7B,MAAO,IAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6CAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+CAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,mCAAmC,EAAK,OAAM,KAC9C,8BAA6B,EAAO,OAAM,MAEhD,MAAO,GAAA,KAAS,WACd,GAAI,GAAe,GAOnB,GANA,EAAK,QAAQ,SAAA,GACX,GAAI,GAAK,MACP,EAAe,GACf,UAGA,EACF,MAAO,MAGT,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAK,IAAM,KAEb,EAAY,KAAK,EAAA,SAAa,EAAO,IAAI,OAAO,SAC3C,AAAI,EAAK,GAAG,KAAO,EAAO,GAAG,KAElC,EAAY,KAAK,EAAA,WAAe,EAAK,GAAI,KAEzC,EAAY,KAAK,EAAK,IAG1B,GAAM,GAAoB,EAAA,OAAW,EAAa,EAAK,MACvD,MAAO,GAAA,IAAQ,EAAmB,GAAI,OAI1C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,MAET,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArIF,EAAA,UAAY,cAuIrB,GAzIiC,IA0IjC,EAAA,cAAc,cAAc,IAuF5B,YAAuB,EAAc,GACnC,KAAO,EAAO,GACZ,GAAQ,EAEV,MAAO,GAGT,YAAkB,EAAW,EAAW,GACtC,GAAI,EAAE,MAAM,OAAS,GAAK,EAAE,MAAM,OAAS,EACzC,KAAM,IAAI,IACN,oEAeN,GAbA,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAC3B,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAE3B,AAAI,MAAO,IAAS,UAClB,GAAO,CAAC,EAAM,IAGZ,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,KAAM,IAAI,IACN,+DAGN,GAAM,GAAQ,EAAE,MAAM,OAChB,EAAQ,EAAE,MAAM,OACtB,AAAI,GAAQ,MAEV,GAAO,CAAC,EAAQ,EAAG,EAAQ,IAE7B,GAAM,GAAY,EAElB,MAAO,GAAA,KAAS,WACd,GAAI,GACJ,GAAI,EAAQ,GACV,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,YACpB,EAAQ,GACjB,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,QAE7B,GAAO,EAGT,GAAI,GACJ,GAAI,EAAE,MAAM,SAAW,GAAK,EAAE,MAAM,SAAW,EAC7C,AAAI,EAAU,KAAO,EAAU,GAC7B,EAAM,EAAE,IAAI,GAAG,IAAI,EAAU,IAE7B,EAAM,EAAE,UAAU,CAAC,EAAG,IAAI,IAAI,GAAG,IAAI,EAAU,SAGjD,GAAM,GAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EACzC,EAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EAC/C,EAAM,EAAE,OAAO,EAAG,EAAM,GAG1B,GAAI,EAAO,GACT,GAAI,GAAG,OACP,AAAI,EAAQ,EACV,EAAM,EAAQ,EAAQ,EAEtB,EAAM,EAAQ,EAGhB,OADM,GAAwB,GACrB,EAAI,EAAK,EAAI,EAAM,EAAM,EAAE,EAClC,EAAY,KAAK,GAEnB,EAAM,EAAI,QAAQ,GAEpB,MAAI,GAAI,MAAM,SAAW,GACvB,GAAM,EAAI,WAAW,IAEhB,IAIX,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,EAAK,KACjB,EAAK,UAAY,EAAK,WAAa,KAAO,GAAQ,EAAK,UACvD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAS,EAAW,GACpB,EAAS,EAAW,GAC1B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,GAAI,EAAO,EAAK,MAAQ,EAAO,EAAK,IAClC,KAAM,IAAI,GACN,8BACG,GAAO,EAAK,IAAG,QAAQ,EAAO,EAAK,OAIpC,EAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,qDACA,iBAAgB,EAAO,OAAM,eAGnC,GAAI,GAAK,EAAO,GACZ,EAAK,EAAO,GACZ,EACJ,MAAK,OAAM,QAAQ,KAAK,MAMtB,EAAO,KAAK,KAAK,IACN,SAAC,EAAM,GAAM,MAAA,IACT,EAAM,EAAO,GAAG,MAAM,UAPrC,EAAO,CACL,GAAc,KAAK,KAAM,EAAG,MAAM,QAClC,GAAc,KAAK,KAAM,EAAG,MAAM,SAOtC,AAAI,KAAK,WACP,GAAK,GAAY,EAAI,EAAK,IAC1B,EAAK,GAAY,EAAI,EAAK,KAErB,GAAS,EAAI,EAAI,IAGlB,EAAA,UAAA,cAAR,SAAsB,EAAe,GACnC,GAAI,GACJ,MAAK,OAAM,QAAQ,KAAK,MAQtB,EAAO,KAAK,KANZ,EAAO,CACL,GAAc,KAAK,KAAM,EAAO,QAChC,GAAc,KAAK,KAAM,EAAO,SAM7B,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAU,EAAW,GAAa,QAClC,EAAU,EAAW,GAAa,QACxC,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAG,GACjB,GAAM,GAAc,EAAO,OAAO,GAClC,MAAI,GAAY,SAAW,GACzB,EAAY,KAAK,GAEZ,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,OAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,KACb,UAAa,KAAK,WAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA7GF,EAAA,UAAY,MA+GrB,GAjHyB,IAkHzB,EAAA,cAAc,cAAc,IChjC5B,AA2BA,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAKjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,OAAS,EAAK,SAGrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,OAAQ,KAAK,QAC7B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAS,WACX,MAAA,IAAe,EAAM,MAAO,EAAG,EAAK,QAAQ,IAAI,IAC9C,EACF,GAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAC9D,MAAO,MA5BJ,EAAA,UAAY,gBA+BrB,GAjCmC,IAkCnC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAS,WACb,GAAM,GAAS,KAAK,KAAK,EAAK,KAAQ,GAAI,EAAK,OAC/C,MAAO,GAAM,IAAI,GAAe,EAAM,MAAO,EAAG,KAElD,MAAO,IAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAEnE,MAAO,MA/BJ,EAAA,UAAY,kBAkCrB,GApCqC,IAqCrC,EAAA,cAAc,cAAc,IAyC5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAMhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,KACjB,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,YAAc,GAAoB,GAAQ,OAGxD,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAa,EAAK,eAAe,GAEjC,EAAgB,WACpB,GAAM,GAAQ,GAAoB,GAE5B,EAAQ,mBACR,EAAQ,mBAER,EAAS,CAAC,EAAQ,EAEpB,EAAU,EAAA,aAAa,EAAA,cAAc,GAAa,EAAK,MAE3D,EAAU,GAAO,EAAS,WAG1B,GAAM,GAAI,KAAA,IAAE,GAAI,EAAK,MAAS,GAAI,EAAK,KAAO,KAAA,IAAA,EAAU,IAAO,KACzD,EAAI,CAAC,EAAI,EAAS,EAAK,KAGvB,EAAI,EAAM,IAAI,GAAS,IAAI,EAAQ,IAAI,IAAI,IAAI,IAErD,MAAO,GAAE,IAAI,GAAG,IAAI,IAEtB,MAAO,IACH,EAAe,WAAM,MAAA,IAAoB,IACzC,EAAO,UAAe,IAE5B,MAAO,MAxDJ,EAAA,UAAY,eA2DrB,GA7DkC,IA8DlC,EAAA,cAAc,cAAc,IChN5B,YA4CI,EAAW,EAAc,EAAkB,EAAe,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GACJ,GAAI,EAAE,OAAS,EACb,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EAEpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EACpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,OAElC,MAAM,IAAI,IACN,2DAA2D,EAAE,KAAI,QAGvE,MAAO,GAoBT,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WACH,GAAM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EACF,GAAmB,EAAG,EAAM,EAAU,EAAM,EAAO,GACvD,MAAO,CAAC,EAAQ,EAAM,KAqBjC,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKH,OAJM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EAAwB,GACX,EAAA,EAAA,EAAA,GAAiB,EAAG,EAAE,MAAtB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAc,QAAQ,KAAU,GAClC,EAAY,KAAK,GAEjB,EAAY,KAAK,EAAE,MAAM,IAG7B,GAAM,GAAgB,EAAK,QAAQ,GAC7B,EAAoB,EAAS,QAAQ,GACrC,EACF,GAAS,KAAO,KAAO,EAAM,QAAQ,GACnC,EACF,GAAQ,KAAO,KAAO,EAAK,QAAQ,GACjC,EAAS,GACX,EAAG,EAAe,EAAmB,EACrC,EAAgB,GACpB,MAAO,CAAC,EAAQ,EAAM,iBAgB7B,EAAW,EAAe,EAAc,EACxC,GACF,MADE,AAAA,KAAA,QAAA,GAAA,MACE,EAAA,KAAK,YACD,EAAc,QAAQ,OAAQ,GAAiB,EAAG,EAAE,KAAO,IAC1D,GACH,EAAG,EAAO,EAAM,EAAe,GAE5B,GACH,EAAG,EAAO,EAAM,EAAe,GAsFvC,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAqBtC,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,IAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KAC1C,EAAK,SAAW,EAAK,UAAY,KAAO,IAAO,EAAK,SACpD,EAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,sBACD,GAAe,EAAK,uBAAyB,SACjD,EAAK,0BACD,GAAe,EAAK,2BAA6B,QACrD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAc,EAAK,iBAC1C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,oBAGvC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAO,KAAK,MAAQ,EAAI,KAAK,KAAQ,KAAK,KAAO,EAAW,OAC5D,EAAM,EAAW,GACvB,GAAI,GAAO,KACT,KAAM,IAAI,GACN,QAAQ,EAAI,+FAET,MAAK,UAAU,GAAW,MAEnC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAW,OAAQ,KAAI,GAAA,GAAG,EAAC,GAAO,EAAG,MAC/D,GAAM,GAAQ,CAAC,GACf,AAAI,KAAK,OACP,MAAK,MAAQ,KAAK,UACd,QAAS,EAAO,KAAM,KAAK,iBAAkB,KAAK,iBAClD,GAAM,KAAK,kBAEjB,AAAI,KAAK,QACP,MAAK,KAAO,KAAK,UACb,OAAQ,EAAO,KAAM,KAAK,gBAAiB,KAAK,gBAAiB,GACjE,KAAK,iBAEX,KAAK,WAAa,KAAK,UACnB,cAAe,EAAO,KAAM,KAAK,sBAAuB,KAAM,IAClE,KAAK,eAAiB,KAAK,UACvB,kBAAmB,EAAO,KAAM,KAAK,0BAA2B,KAChE,IACJ,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACvD,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAO,EAAW,OAClB,EAAgB,GAAiB,EAAG,GACpC,EAAO,EAAK,MAAQ,EAAI,EAAK,KAAQ,EAAK,KAAO,EACvD,EAAc,OAAO,EAAM,GAC3B,GAAM,GAAiB,GAA2B,EAAG,GACrD,EAAe,GAAQ,EAAW,GAElC,GAAM,GAAsB,EAAc,QAC1C,EAAoB,OACpB,GAAM,GAAoB,CAAC,EAAA,KAAK,YAC5B,EAAqB,GAAiB,EAAG,GAAM,MAAM,EAAG,EAAO,IAE7D,EAAmC,WACvC,GAAI,GACF,GAAM,GACF,EAAK,WAAW,OAAO,QAAQ,GAC7B,EACF,EAAK,eAAe,OAAO,QAAQ,GACjC,EACF,EAAK,OAAS,EAAK,KAAK,OAAO,QAAQ,GAAkB,KACvD,EACF,EAAK,MAAQ,EAAK,MAAM,OAAO,QAAQ,GAAkB,KAC7D,MAAO,IACH,EAAO,EAAqB,EAC5B,EAAe,EAAgB,EAAK,aAExC,OAAO,IACH,EAAO,EAAK,WAAW,OAAQ,EAAK,eAAe,OACnD,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACrC,EAAK,OAAS,KAAO,KAAO,EAAK,MAAM,OAAQ,EAAK,UAI5D,GAAI,CAAC,EACH,MAAO,KAGH,GAAA,GAAA,GAAA,EAAA,EAAA,MAAA,OAAA,EAAA,KAAA,OAAA,EAAA,EAAA,SAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAM,EAAA,EAAA,GAIvB,EACF,SAAC,EAAyB,EAAe,GACvC,EAAA,KAAS,WACP,GAAM,GAAQ,EAAI,EACZ,EAAY,EAAS,OACrB,EAAc,EAAU,IAAI,GAAO,IAAI,GAC7C,EAAS,MAAM,EAAU,IAAI,OAU/B,EAA8B,WAClC,EAAgB,EAAK,WAAY,EAAM,EAAK,UAC5C,EAAgB,EAAK,eAAgB,EAAU,EAAK,WAEtD,WAEO,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,SAAU,KAAK,SACf,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,sBAAuB,GAAqB,KAAK,uBACjD,0BACI,GAAqB,KAAK,2BAC9B,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,eAAgB,GAAoB,KAAK,gBACzC,gBAAiB,GAAoB,KAAK,kBAEtC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArKF,EAAA,UAAY,qBAuKrB,GAzKwC,IA0KxC,EAAA,cAAc,cAAc,IAkD5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAgBtC,WAAY,GAAZ,GAAA,GAAA,KAOE,GANA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KACtC,MAAO,GAAK,MAAS,UACvB,GAAI,CAAC,OAAO,UAAU,EAAK,MACzB,KAAM,IAAI,OACN,gDAAgD,EAAK,cAElD,MAAM,QAAQ,EAAK,MAC5B,OAAmB,GAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,6CACA,iBAAgB,KAAK,UAAU,EAAK,YAI5C,MAAM,IAAI,OACN,2DACA,iBAAgB,KAAK,UAAU,EAAK,QAG1C,SAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,kBAE5C,EAAK,gBAAkB,KAGlB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAQ,EAAW,OAGzB,AAAI,MAAO,MAAK,MAAS,UACvB,MAAK,KAAO,CAAC,KAAK,OAEpB,OAAS,GAAI,EAAG,EAAI,KAAK,KAAK,OAAQ,EAAE,EACtC,AAAI,KAAK,KAAK,GAAK,GACjB,MAAK,KAAK,IAAM,GAKpB,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,GAAK,GAAQ,EACtB,KAAM,IAAI,OAAM,iBAAiB,GAGrC,GAAI,KAAK,KAAK,SAAW,GAAqB,KAAK,MAAM,OACvD,KAAM,IAAI,OAAM,4BAA4B,KAAK,MAGnD,GAAM,GAAa,KAAK,KAAK,IAAI,SAAA,GAAQ,MAAA,GAAW,KAE9C,EAAY,GAClB,AAAI,KAAK,MACP,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAE3B,KAAK,MAAQ,KAEf,AAAI,KAAK,OACP,KAAK,KAAO,KAAK,UACb,OAAQ,EAAY,UAAW,KAAK,gBACpC,KAAK,gBAAiB,GAE1B,KAAK,KAAO,KAGd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACQ,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAQ,EAAW,OAEzB,MAAO,GAAA,KAAK,WAIV,OAHM,GAAW,GACb,EAAA,EAAA,QAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,SACL,EAAiB,GAA2B,EAAG,GACnC,EAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAe,GAAO,EAAW,GAuBnC,OApBM,GAAY,SAAC,GACjB,MAAI,IAAK,MAAQ,EAAE,MAAM,SAAW,GAChC,EAAK,OAAS,CAAC,EAAQ,GAClB,EAAE,QAAQ,GAEV,GAIP,EAAQ,EAAU,EAAK,MAAM,QAC7B,EAAS,EAAU,EAAK,KAAK,QAQ3B,EAA0B,GAC1B,EAA8B,GAC3B,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,AAAK,EAAK,KAAkB,QAAQ,KAAO,GACzC,GAAc,KAAK,EAAW,IAC9B,EAAkB,KAAK,IAEvB,GAAc,KAAK,GACnB,EAAkB,KAAK,EAAW,KAGtC,SAAO,EAAK,KAAK,GACjB,EAAW,EAAS,KAAK,GACzB,EAAQ,EAAM,KAAK,GACnB,EAAS,EAAO,KAAK,GAEd,GACH,EAAO,EAAM,EAAU,EAAQ,EAAO,EAAK,YAInD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,mBAExC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjKF,EAAA,UAAY,qBAmKrB,GArKwC,IAsKxC,EAAA,cAAc,cAAc,IC9nB5B,YAoEI,EAAW,EACX,GACF,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,kEACG,GAAE,KAAI,eAMf,GAHA,AAAI,GAAW,MACb,GAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,KAErB,EAAQ,SAAW,GAAK,EAAQ,GAAG,SAAW,GAC9C,EAAQ,GAAG,SAAW,EACxB,KAAM,IAAI,GACN,+GAON,GAHA,AAAI,GAAc,MAChB,GAAa,MAEX,IAAe,gBAAkB,IAAe,gBAClD,KAAM,IAAI,GACN,wBAAwB,EAAU,mEAIxC,GAAI,GACJ,MAAI,KAAe,gBACjB,EAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,IAE/C,EAAU,CAAC,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,GAAI,CAAC,EAAG,IAG1C,EAAA,IAAQ,EAAG,KA6BtB,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAMjC,WAAY,GAAZ,GAAA,GAAA,KAUE,GATA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,WACD,EAAK,YAAc,KAAO,KAAoB,EAAK,WAGnD,EAAK,SAAW,KAClB,EAAK,QAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,YACnB,MAAO,GAAK,SAAY,SACjC,EAAK,QACD,CAAC,CAAC,EAAK,QAAS,EAAK,SAAU,CAAC,EAAK,QAAS,EAAK,eAGvD,GADA,EAAK,QAAU,EAAK,QAChB,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,6DACA,sBAAqB,EAAK,QAAQ,OAAM,YAG9C,GAAI,GAAa,OACb,EAAY,OAChB,GAAI,MAAO,GAAK,QAAQ,IAAO,SAC7B,EAAgB,CAAC,EAAK,QAAQ,GAAI,EAAK,QAAQ,IAC/C,EAAe,CAAC,EAAK,QAAQ,GAAc,EAAK,QAAQ,SAIxD,GAFA,EAAK,QAAU,EAAK,QAEhB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,gEACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAIrD,GAFA,EAAgB,EAAK,QAAQ,GAEzB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,+DACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAErD,EAAe,EAAK,QAAQ,GAE9B,EAAK,QAAU,CAAC,EAAe,GAEjC,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAEhC,GAAI,GACA,EACJ,MAAI,MAAK,aAAe,gBACtB,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,IAE5C,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,MAIlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,IACF,GAAoB,GAAS,EAAK,QAAS,EAAK,eAG1D,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAnGF,EAAA,UAAY,gBAqGrB,GAvGmC,IAwGnC,EAAA,cAAc,cAAc,IC5O5B,YA0CI,EAAW,EAA4B,EACvC,EAAuB,EACvB,GACF,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,IAEhB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAKb,EAAI,GAAsB,EAAG,GAC7B,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MAEf,EAAI,EAAA,QAAY,EAAe,EAAU,EAAS,GAIlD,EAAI,EAAA,QAEA,EAA0B,EAAU,EAAS,GAEnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBAeP,EAAa,EACb,EAAoC,EACpC,EAAyB,GAC3B,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,EAAG,IAEnB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAIb,EAAI,GAAsB,EAAa,GACvC,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MACf,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAExC,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAE1C,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAsBX,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAWtC,WAAY,GAAZ,GAAA,GAAA,KAKE,GAJA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,GAElB,EAAA,EAAA,KAAA,KAAM,IAAK,KACP,MAAO,GAAK,UAAa,SAC3B,EAAK,SAAW,CAAC,EAAK,kBAEpB,MAAM,QAAQ,EAAK,WAClB,EAAK,SAAsB,SAAW,GACvC,MAAQ,GAAK,SAAsB,IAAO,SAC5C,EAAK,SAAW,EAAK,aAErB,MAAM,IAAI,GACN,qGAEA,IAAG,KAAK,UAAU,EAAK,YAG7B,GADA,GAAsB,EAAK,SAAU,YACjC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBAEhB,MAAO,GAAK,SAAY,SAC1B,EAAK,QAAU,CAAC,EAAK,iBAEnB,MAAM,QAAQ,EAAK,UAClB,EAAK,QAAqB,SAAW,GACtC,MAAQ,GAAK,QAAqB,IAAO,SAC3C,EAAK,QAAU,EAAK,YAEpB,MAAM,IAAI,GACN,oGAEA,IAAG,KAAK,UAAU,EAAK,WAG/B,UAAsB,EAAK,QAAS,WAEpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAS,GACX,EAAW,GAAI,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IAChE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAW,KAO5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,EAAS,GAAa,GAAoB,GAAS,GACnD,GAAM,GAAS,EAAK,gBAChB,GAAoB,GAAS,CAAC,EAAK,SAAS,GAAI,GAChD,CAAC,EAAK,QAAQ,GAAI,GAAI,EAAK,QAAS,gBAExC,MAAO,GAAA,QAAY,EAAQ,CAAC,OAIhC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,SAEV,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAxFwC,IA0FxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,IAEtB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,UACrB,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,SAErC,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAKrE,MAJA,GACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,GAErC,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,KAQlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAhFwC,IAkFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,EAAG,IAEzB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,SAAU,EAAK,UACpC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,QAAS,EAAK,SAEnD,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAOrE,MANA,GAAS,GACL,EAAQ,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,EAAM,GAE7C,CAAC,EAAW,GAAI,EAAQ,EAAM,EAAM,EAAW,KAS1D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GArFwC,IAuFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,eAcrB,GAhBkC,IAiBlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,mBAcrB,GAhBsC,IAiBtC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAC5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,EAAW,KAGpC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAEd,GAb8C,IAe9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAG1C,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,KAAS,EAAO,MARpB,EAAA,UAAY,yBAWrB,GAb4C,IAc5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAGtC,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,IAAQ,EAAO,MARnB,EAAA,UAAY,qBAWrB,GAbwC,IAcxC,EAAA,cAAc,cAAc,IAiB5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAE5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GAEjB,MADA,GAAa,EACT,KAAK,aAAe,eACf,CAAC,EAAW,GAAI,EAAW,IAE3B,CAAC,EAAW,GAAI,EAAW,KAItC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAGZ,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,KAAK,YAC3B,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA7B8C,IA+B9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAA5C,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,KAAS,EAAO,CAAC,EAAG,IAEpB,EAAA,KAAS,EAAO,CAAC,EAAG,OAR1B,EAAA,UAAY,yBAYrB,GAd4C,IAe5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAAxC,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,IAAQ,EAAO,CAAC,EAAG,IAEnB,EAAA,IAAQ,EAAO,CAAC,EAAG,OARzB,EAAA,UAAY,qBAYrB,GAdwC,IAexC,EAAA,cAAc,cAAc,IC5rB5B,AA4CA,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,GAAZ,GAAA,GAQE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAKf,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAIE,MAAI,MAAK,OAAS,KACT,KAAK,MAAM,UAEX,QAIX,SAAc,GAIZ,AAAI,KAAK,OAAS,MAChB,MAAK,MAAM,UAAY,oCAI3B,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,MAAM,kDAIpB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAIpB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WAEE,MAAQ,MAAK,MAAc,0CAK7B,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,MAAM,wCAKpB,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,MAAM,cAGpB,EAAA,UAAA,WAAA,SAAW,GACT,KAAK,MAAM,WAAW,IAGxB,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAS,CACP,UAAa,KAAK,MAAM,eACxB,OAAU,KAAK,MAAM,cAGnB,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,OAAS,MAChB,KAAK,MAAM,6BAA6B,IAKrC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAc,EAAO,MACrB,EAAQ,GAAY,EAAa,GACvC,MAAO,GAAO,MACd,GAAM,GAAY,CAAC,MAAK,GACxB,cAAO,OAAO,EAAW,GAClB,GAAI,GAAI,IAEnB,GAvGsC,IAyGtC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,oEACA,gBAAe,KAAK,UAAU,KAEpC,KAAK,UAAY,CAAC,CAAC,MAAO,IAC1B,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAChE,AAAK,KAAK,MAAM,OACd,MAAK,MAAM,MAAM,GACjB,KAAK,MAAM,MAAQ,IAErB,EAAA,UAAM,MAAK,KAAA,KAAC,IAGd,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAC1D,EACF,KAAK,MAAM,mBAAmB,GAC5B,EAAY,EAAW,GAC7B,MAAO,CAAC,EAAiB,GAAI,GAAW,OAAO,EAAiB,MAAM,KAGxE,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,EAAS,GAAoB,GAI7B,GAAM,GAAwB,SAAC,EAAgB,GAK7C,GAAM,GAAS,GAAoB,EAAK,MAAM,KAAK,EAAQ,IAC3D,MAAO,CAAC,EAAQ,KAEZ,EACF,GAAI,EAAM,EAAQ,GAAI,GAAyB,KAC3C,KAAsB,GACtB,IACF,EAAI,EAAW,GAGrB,MAAO,MArDJ,EAAA,UAAY,kBA0DrB,GA5DqC,IA6DrC,EAAA,cAAc,cAAc,gBAEgB,GAC1C,GACI,GAAiC,yBAA0B,GAmBjE,GAAM,IAA2D,SAEjE,GAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAUL,EAAc,EAAK,MAAM,YACzB,EAAqC,GAC3C,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,aAAe,GAAY,GAChC,EAAY,YACR,IAAY,cAAmB,IACnC,GAAM,GAAqC,GAW3C,GAVA,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,cAAgB,GAAY,GACjC,EAAK,aAAa,KAAO,WAAa,EAAK,aAAa,KACxD,EAAK,cAAc,KAAO,YAAc,EAAK,cAAc,KAE3D,EAAK,UAAY,EAAK,YAAc,OAChC,GACA,EAAK,UACT,GAA4B,EAAK,WAC7B,EAAK,QACP,KAAM,IAAI,IACN,mEAEN,SAAK,UAAY,EAAK,MAAM,SAC5B,EAAK,gBAAkB,EAAK,MAAM,gBAClC,EAAK,YAAc,EAAK,MAAM,YAC9B,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAClB,EAAK,UAAY,EAAK,MAAM,UAC5B,EAAK,aAAe,OAGtB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GAIZ,KAAK,WAAa,EAClB,AAAI,KAAK,cAAgB,MACvB,MAAK,aAAa,UAAY,GAEhC,AAAI,KAAK,eAAiB,MACxB,MAAK,cAAc,UAAY,oCAInC,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,aAAa,aAAa,OAClC,KAAK,cAAc,eAGzB,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAa,EAAQ,OACrB,EAAiB,KAAK,MAAM,EAAa,GAC/C,KAAK,aAAa,WAAW,EAAQ,MAAM,EAAG,IAC9C,KAAK,cAAc,WAAW,EAAQ,MAAM,KAG9C,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GACA,KAAK,aAAa,mBAAmB,GACzC,AAAM,MAAM,QAAQ,IAAgB,MAAM,QAAQ,EAAY,KAC5D,GAAc,CAAC,IAEjB,EAAc,EAEd,GAAI,GACA,EACA,EAiBJ,MAhBA,AAAI,MAAK,aACP,GAAa,EAAY,MAAM,IAC/B,EAAc,EAAY,GAI5B,EAAc,EACd,AAAI,KAAK,YAAc,SACrB,GAAY,EAAY,OAAS,IAAM,EACvC,EAAe,CAAC,IACX,AAAI,KAAK,WAAa,KAC3B,EAAe,CAAC,EAAa,EAAY,SAEzC,EAAe,CAAC,GAGd,KAAK,YACH,KAAK,WAAa,KACb,EAAa,OAAO,GAAY,OAAO,EAAW,SAEpD,CAAC,GAAa,OAAO,GAAY,OAAO,EAAW,SAErD,GAA+B,IAGxC,EAAA,UAAA,MAAA,SACI,EACA,GACF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAEX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAU1D,GATA,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAEzB,AAAI,MAAM,QAAQ,IAChB,GAAgB,EAAuC,MAAM,GAC7D,EAAU,EAAuC,IAG9C,IAAgB,MAAQ,EAAa,SAAW,IACjD,GAAa,KACf,MAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,GAE7B,GAAM,GAAiD,GACjD,EAA+B,GACrC,GAAI,GAAgB,MAClB,GAAM,GAAY,EAAa,OAC/B,GAAI,EAAY,EAAI,EAClB,KAAM,IAAI,GACN,iIAIN,EAAO,aAAkB,EACzB,EAAiB,KAAI,MAArB,EAAyB,GACzB,GAAM,GAAc,EACI,IAAI,SAAA,GAAS,MAAA,IAAI,IAAU,CAAC,MAAO,EAAM,UACjE,KAAK,aAAa,UAAY,EAAW,MAAM,EAAG,EAAY,GAC9D,KAAK,cAAc,UAAY,EAAW,MAAM,EAAY,GAC5D,EAAgB,KAAI,MAApB,EAAwB,GAE1B,GAAI,GAAa,KACf,KAAM,IAAI,IACN,yEAKN,OADM,GAAmB,EAAiB,YAAc,IACnC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,YAAkB,MAAmB,EACvC,KAAM,IAAI,GACN,gHAKR,GAAI,GAEF,GAAM,GAAY,CAAC,GAAQ,OAAO,GAC5B,EAAgB,KAAK,UAAU,OAAO,GAUtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GACF,EAAA,UAAM,MAAK,KAAA,KAAC,EAA0C,GAC1D,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAI/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAe,EAAO,aAExB,EACA,EACJ,GAAI,GAAgB,KAClB,EAAI,EAAK,aAAa,KAAK,EAAQ,GACnC,EAAO,EAAK,cAAc,KAAK,EAAQ,QAEvC,GAAM,GAAe,EAAa,MAAM,EAAG,EAAa,OAAS,GAC3D,EAAgB,EAAa,MAAM,EAAa,OAAS,GAC/D,EAAI,EAAK,aAAa,KAClB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KACjD,EAAO,EAAK,cAAc,KACtB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KAGnD,GAAI,GACJ,AAAI,EAAK,aACP,CAAI,MAAM,QAAQ,IAChB,GAAS,EAAE,MAAM,GAAG,OAAQ,EAAkB,MAAM,KAGtD,EAAK,EAAe,GACpB,EAAQ,EAAkB,IAG5B,AAAI,EAAK,iBACP,GAAO,EAAA,QAAY,EAAgB,IAGrC,GAAI,GAcJ,MAbA,AAAI,GAAK,YAAc,SACrB,EAAS,GAAc,CAAC,EAAa,IAChC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,GAAI,EAAA,IAAQ,EAAa,IACrC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,WAAa,MAC3B,GAAS,CAAC,EAAa,IAIrB,EAAK,YACH,EAAK,WAAa,KACZ,EAAoB,OAAO,GAE9B,CAAC,GAAkB,OAAO,GAE5B,KAIX,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,aAAa,cAClB,KAAK,cAAc,eAGrB,EAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,GAAU,KAAK,aAAa,KAAM,WAChC,EAAK,aAAa,MAAM,KAE1B,GAAU,KAAK,cAAc,KAAM,WACjC,EAAK,cAAc,MAAM,KAE3B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,SAAY,EAAyB,GAEnC,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAI,GAcJ,GAbA,AAAI,KAAK,gBACP,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,EAAM,GAEpB,EAAa,EAGf,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,KAAM,MAEpB,EAAa,KAGb,KAAK,aACP,GAAM,GAAS,KAAK,aAAa,OAC3B,EAAsB,EAAO,IAAI,SAAA,GAAS,MAAA,QAChD,MAAI,OAAM,QAAQ,GACT,EAAW,OAAO,GAAW,OAAO,GAEpC,CAAC,GAAY,OAAO,GAAW,OAAO,OAG/C,OAAO,IAIX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,aAAa,iBAAiB,OACtC,KAAK,cAAc,mDAGzB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,aAAa,oBAAoB,OACzC,KAAK,cAAc,sDAKzB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,cAAgB,MACvB,KAAK,aAAa,6BAA6B,GAEjD,AAAI,KAAK,eAAiB,MACxB,KAAK,cAAc,6BAA6B,IAIpD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,UAAa,KAAK,WAGd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAIF,EAAA,WAAP,SACI,EACA,GACF,GAAM,GACF,GAAY,EAAO,OAGvB,GAFA,MAAO,GAAO,MAEV,EAAO,cAAmB,KAC5B,KAAM,IAAI,IACN,4FAIN,GAAM,GAAkC,EACxC,SAAU,MAAW,EACd,GAAI,GAAI,IA7VV,EAAA,UAAY,gBA+VrB,GAjWmC,IAkWnC,EAAA,cAAc,cAAc,IC7kB5B,YAkE2B,GACzB,MAAO,IAAI,IAAW,eA6BJ,GAClB,MAAO,IAAI,IAAI,eAoBI,GACnB,MAAO,IAAI,IAAK,eAuBQ,GACxB,MAAO,IAAI,IAAU,eAwBD,GACpB,MAAO,IAAI,IAAM,eAmBK,GACtB,MAAO,IAAI,IAAQ,eA2BW,GAC9B,MAAO,IAAI,IAAgB,eAyBN,GACrB,MAAO,IAAI,IAAO,eAqBG,GACrB,MAAO,IAAI,IAAO,eAsCY,GAC9B,MAAO,IAAI,IAAgB,eAqBN,GACrB,MAAO,IAAI,IAAO,eAgCY,GAC9B,MAAO,IAAI,IAAgB,eAkCF,GACzB,MAAO,IAAI,IAAW,eA2BK,GAC3B,MAAO,IAAI,IAAa,eAeM,GAC9B,MAAO,IAAI,IAAgB,eAoCF,GACzB,MAAO,IAAI,IAAW,eAoCF,GACpB,MAAO,IAAI,IAAM,eAaK,GACtB,MAAO,IAAI,IAAQ,eAmCY,GAC/B,MAAO,IAAI,IAAiB,eAsBN,GACtB,MAAO,IAAI,IAAQ,eAiBQ,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,eA+BG,GACtB,MAAO,IAAI,IAAQ,eAcK,GACxB,MAAO,IAAI,IAAU,eAyBH,GAClB,MAAO,IAAI,IAAI,eAqBO,GACtB,MAAO,IAAI,IAAQ,eAuBO,GAC1B,MAAO,IAAI,IAAY,eAqBD,GACtB,MAAO,IAAI,IAAQ,eAqBG,GACtB,MAAO,IAAI,IAAQ,eAsBI,GACvB,MAAO,IAAI,IAAS,eAyBF,GAClB,MAAO,IAAI,IAAI,eA0BkB,GACjC,MAAO,IAAI,IAAmB,eAuBG,GACjC,MAAO,IAAI,IAAmB,eA2BF,GAC5B,MAAO,IAAI,IAAc,eAgBM,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eA0BO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAwBO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAYa,GACrC,MAAO,IAAI,IAAuB,eAiBG,GACrC,MAAO,IAAI,IAAuB,eAYD,GACjC,MAAO,IAAI,IAAmB,eAiBG,GACjC,MAAO,IAAI,IAAmB,eAYH,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eA4BN,GAClB,MAAO,IAAI,IAAI,eAgDO,GACtB,MAAO,IAAI,IAAQ,eA0BA,GACnB,MAAO,IAAI,IAAK,eAgDO,GACvB,MAAO,IAAI,IAAS,eA2BI,GACxB,MAAO,IAAI,IAAU,eAgDO,GAC5B,MAAO,IAAI,IAAc,eA8BA,GACzB,MAAO,IAAI,IAAW,eAoCO,GAC7B,MAAO,IAAI,IAAe,eA+DR,GAClB,MAAO,IAAI,IAAI,eAUe,GAC9B,MAAO,IAAI,IAAgB,eAMC,GAC5B,MAAO,IAAI,IAAc,eAiDK,GAC9B,MAAO,IAAI,IAAgB,GAItB,GAAM,IAAkB,GAClB,GAAkB,GAClB,GAAY,GACZ,GAAY,eA2BK,GAC5B,MAAO,IAAI,IAAc,eA2BK,GAC9B,MAAO,IAAI,IAAgB,eAkCA,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,glCClnDU,EAAe,GAC5C,MAAO,IAAuB,EAAO,eAoBJ,EAAe,GAChD,MAAO,IAA2B,EAAO,eAuBvC,EAAe,GACjB,MAAO,IAAkC,EAAO,eAqBd,EAAe,GACjD,MAAO,IAA4B,EAAO,eAcJ,EAAe,GACrD,MAAO,IAAgC,EAAO,eAsCtB,EAAe,GACvC,MAAO,IAAkB,EAAO,eAsCX,EAAe,GACpC,MAAO,IAAe,EAAO,eAwBC,EAAe,GAC7C,MAAO,IAAuB,EAAO,eAuBL,EAAe,GAC/C,MAAO,IAAyB,EAAO,eAsBrC,EAAe,GACjB,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAqBlB,EAAe,GAC9C,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,kSC1TxC,yCCAA,YAqBqB,GACnB,MAAO,IAAI,IAAK,eAYC,GACjB,MAAO,IAAgB,eAYN,GACjB,MAAO,IAAgB,+CChDzB,mBAkBuC,EAAA,EAAA,GAAvC,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KAEE,SAAA,MAAqB,OAErB,SAAA,UAAA,SAAA,SAAS,GACP,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,OAAM,yDAElB,KAAK,MAAQ,GAEjB,GAVuC,IAsEvC,YAAc,EAAiB,GAC7B,MAAO,GAAU,EAGnB,YAAiB,EAAiB,GAChC,MAAO,GAAU,qBAOgB,EAAA,EAAA,GAcjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GAHA,AAAI,GAAQ,MACV,GAAO,IAEL,EAAK,mBACP,KAAM,IAAI,IACN,sEAGN,SAAK,QAAU,EAAK,SAAW,WAC/B,EAAK,SAAW,KAAK,IAAI,EAAK,UAAY,GAC1C,EAAK,SAAW,EAAK,UAAY,EACjC,EAAK,QAAU,EAAK,SAAW,EAC/B,EAAK,KAAO,EAAK,MAAQ,OACzB,EAAK,SAAW,EAAK,SAErB,AAAI,CAAC,OAAQ,MAAO,OAAO,QAAQ,EAAK,QAAU,IAChD,SAAQ,KACJ,uBAAuB,EAAK,KAAI,8CAEpC,EAAK,KAAO,QAGd,AAAI,EAAK,OAAS,MAChB,EAAK,YAAc,GACd,AAAI,EAAK,OAAS,OAInB,EAAK,QAAQ,QAAQ,SAAW,GAHpC,EAAK,YAAc,GAMjB,EAAK,YAAc,GAIvB,AAAI,EAAK,cAAgB,IACvB,GAAK,UAAY,MAIf,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,aAAe,EACpB,AAAI,KAAK,UAAY,KACnB,KAAK,KAAO,KAAK,SAEjB,KAAK,KAAO,KAAK,cAAgB,GAAO,SAAW,mBAIjD,EAAA,UAAA,WAAN,SAAiB,EAAe,mGAC9B,MAAA,CAAA,EAAM,GAAqB,WAE3B,MAFA,GAAA,OACM,EAAU,KAAK,gBAAgB,GACjC,GAAW,KACb,CAAA,GAGF,CAAI,KAAK,YAAY,EAAU,KAAK,SAAU,KAAK,MACjD,MAAK,KAAO,EACZ,KAAK,KAAO,GAGZ,MAAK,OACL,AAAI,KAAK,MAAQ,KAAK,UACpB,MAAK,aAAe,EACpB,KAAK,MAAM,aAAe,gBAM1B,EAAA,UAAA,WAAN,SAAiB,sEACf,MAAI,MAAK,aAAe,GAAK,KAAK,SAChC,QAAQ,IAAI,SAAS,KAAK,aAAY,8BAIlC,EAAA,UAAA,gBAAR,SAAwB,GACtB,AAAI,GAAQ,MACV,GAAO,IAET,GAAM,GAAe,EAAK,KAAK,SAC/B,MAAI,IAAgB,MAClB,QAAQ,KACJ,4BAA4B,KAAK,QAAO,sBACxC,2BAA0B,OAAO,KAAK,KAErC,GAEX,GAzGmC,gBAsJL,GAC5B,MAAO,IAAI,IAAc,MAGd,IAAY,CAAC,cAAa,i8DC9PvC,AA8BA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,WAAA,GAAA,aACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,UACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,aAAA,GAAA,eACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,IAAA,UACA,EAAA,EAAA,SAAA,IAAA,WACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,YAAA,IAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,iBAAA,KAAA,mBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,gBAAA,KAAA,oBA7BU,IAAA,IAAQ,QAyRH,IAAjB,AAAA,UAAiB,GAEf,GAAY,GAAZ,AAAA,UAAY,GAAyB,EAAA,EAAA,OAAA,GAAA,SAAc,EAAA,EAAA,GAAA,GAAA,KAAU,EAAA,EAAA,GAAA,GAAA,OAAjD,EAAA,EAAA,yBAAA,GAAA,wBAAuB,OAFpB,IAAA,IAAQ,KCtTzB,AAmBA,GAAM,IAAwC,eA2BnB,EAAc,GACvC,GAAM,GAAqB,CACzB,SAAU,EACV,SAAU,SACV,OAAQ,GACR,MAAO,GACP,eAAgB,GAGlB,GAAW,GAAQ,cAUW,GAC9B,MAAO,IAAW,eAUS,GAC3B,MAAO,IAAW,GC9EpB,WAwBI,EAAmB,EAAY,EAC/B,GACF,GAAM,GAAa,EAAK,YAAY,GACpC,GAAI,GAAc,EAAW,kBAAoB,QAC/C,GAAM,GAAQ,EAAW,gBACnB,EAAM,EAAW,gBAAkB,EACrC,OACC,EAAW,gBAAkB,OAAY,EAAQ,EACR,EAAW,cACzD,GAAI,EAAW,OAAS,SACtB,MAAO,IACH,EAAK,WAAW,EAAW,iBAAkB,EAAW,GAE9D,GAAI,EAAW,OAAS,WACtB,GAAM,GAAS,EAAK,WAAW,MAAM,EAAO,GAE5C,MAAO,GAAO,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAEvD,GAAM,GACF,GAAU,EAAK,WAAW,MAAM,GAAO,GAAI,EAAW,GACpD,EAAO,EAAO,WACpB,MAAO,GAAW,OAAS,SACvB,EAAK,GACL,EAAA,KAAK,cAAc,EAAO,MAAO,GAEvC,GAAM,GAAY,EAAK,WAAW,GAClC,MAAO,IAAa,EAAU,kBAU5B,EAAc,EACd,GACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAY,EAAQ,kBAAkB,KAAK,SAAA,GAC/C,MAAO,CAAC,CAAC,EAAW,GAAyB,EAAU,MAGzD,MAAO,KAAc,OACjB,EAAW,GAAyB,EAAU,IAAY,GAC1D,mBASF,EAAc,EACd,GACF,MAAO,GAAW,GAAyB,EAAM,EAAQ,+BAUvD,EAAmB,GACf,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEjB,MAAO,CACL,GAAyB,EAAU,GAAW,EAAQ,kBACtD,GAIJ,YAAkC,EAAc,GAC9C,MAAO,AAAE,GAAe,EAAI,IAAI,EAAc,cAGlB,GAC5B,GAAM,GAAQ,EAAK,MAAM,KACzB,GAAI,EAAM,SAAW,EACnB,MAAO,CAAC,EAAM,GAGhB,GAAM,GAAW,EAAM,GACvB,MAAO,CAAC,EAAU,OAAO,EAAM,EAAM,OAAS,iBAW5C,EAAY,EACZ,GACF,GAAI,GAAM,EAAc,MAAO,EAAM,EAAW,GAChD,GAAI,IAAQ,YAEV,EAAM,EAAc,mBAAoB,EAAM,EAAW,GAIzD,OAHM,GAEF,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,IACxB,EAAI,EAAG,EAAI,EAAG,IACrB,EAAgB,GAAG,GAAM,EAAiB,EAAI,GAC9C,EAAgB,GAAG,GAAM,EAAiB,EAAI,EAAI,GAEpD,MAAO,GAET,MAAO,eAYmB,GAC1B,MAAO,GAAO,KAAO,EAAS,EAAA,MAAM,GCrJtC,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,mCCnLxE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,UAC7D,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,YAGjE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC7dxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAElD,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,QAC1D,CAAC,OAAU,mBAAoB,KAAQ,iBAAkB,KAAQ,QACjE,CACE,OAAU,2BACV,KAAQ,yBACR,KAAQ,QAEV,CAAC,OAAU,oBAAqB,KAAQ,OAAQ,KAAQ,YAG5D,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CACR,OAAU,QACV,KAAQ,QACR,KAAQ,QACR,aAAgB,MAGpB,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAAU,CACrD,OAAU,wBACV,KAAQ,sBACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,YAG3C,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,YAE3D,CACE,SAAY,cACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,KACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,iBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,WAG9D,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,wCCnWrE,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAAa,CAC7D,OAAU,yBACV,KAAQ,sBACR,KAAQ,QAEV,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAC/C,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,OAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,WACV,KAAQ,WACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,gBAAiB,KAAQ,gBAAiB,KAAQ,QAAS,CACpE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,eACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,mBACV,KAAQ,gBACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,QAItB,CACE,SAAY,sBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,aAE9C,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,kBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,wBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,6BACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,aACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,QAAS,KAAQ,YAAa,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,yCCpVnD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,WAEtC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,SAAU,aAAgB,GAClE,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,SAAU,aAAgB,IAErE,MAAS,CACP,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAEzD,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,IAEjE,MAAS,CAAC,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAExD,CACE,SAAY,kBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,OACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAAW,CACpD,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,WAE7C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAC3C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,UAC7C,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,SACzC,CAAC,OAAU,eAAgB,KAAQ,eAAgB,KAAQ,wCC3KjE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,YAGnD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,WAEjD,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,cACV,KAAQ,YACR,KAAQ,QACR,aAAgB,IAElB,CACE,OAAU,yBACV,KAAQ,qBACR,KAAQ,UAId,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,UAC/C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,WAE5C,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCChGtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAE3D,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,yCCvC3C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,yBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CACE,SAAY,cACZ,SAAY,QACZ,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CAAC,SAAY,QAAS,SAAY,SAAU,CAC1C,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,YACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,SACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAEvC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,UAAW,CAC1D,OAAU,UACV,KAAQ,SACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,SACR,aAAgB,KAItB,CAAC,SAAY,OAAQ,SAAY,QAAS,OAAU,IAAK,CACvD,SAAY,eACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,0BACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,UACzC,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,yCC3G/C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,wBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,aAE3C,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAAW,CACxD,OAAU,sBACV,KAAQ,qBACR,KAAQ,yCCvDhB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,eACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC9ItB,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,eACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAAW,CAC3D,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCCtItB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,eACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GACrE,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,aACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,gBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCCjJtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,QACrD,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,uCC1GvD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,GAAI,KAAQ,UAAW,KAAQ,WACnD,CAAC,MAAS,GAAI,KAAQ,OAAQ,KAAQ,WAExC,MACI,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAEtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,WAClD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAG3E,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,OAAQ,aAAgB,MAGjE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,eACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,YACpC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CACP,CACE,OAAU,aACV,KAAQ,YACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,WACV,KAAQ,UACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,cACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,mBACV,KAAQ,iBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,YAEpD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGzE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,MACV,KAAQ,MACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,MAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAC/D,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,YACV,KAAQ,kBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,kBAAmB,KAAQ,YAChD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAG5C,CACE,SAAY,gBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCC1NtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,mCChDxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,OACV,KAAQ,SACR,KAAQ,QACR,aAAgB,IAElB,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAGhD,CACE,SAAY,aACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,MACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CAAC,CACR,OAAU,iBACV,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,QACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAAa,CACnD,MAAS,EACT,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,OACV,iBAAoB,eACpB,KAAQ,OACR,KAAQ,cAGZ,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,cAG7C,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,eACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,YAG5D,CACE,SAAY,cACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,iCChIb,AAyCA,GAAA,IAAA,WAWE,aACE,GAAM,GAAM,CACV,GAAY,GAAW,GAAS,GAAa,GAAU,GACvD,GAAY,GAAS,GAAO,GAAO,GAAU,GAAe,GAC5D,GAAW,GAAU,IAEjB,EAA0B,GAAG,OAAM,MAAT,GAAa,EAAI,IAAI,SAAA,GAAM,MAAA,GAAG,QAE9D,KAAK,UAAY,EAAY,OACzB,SAAC,EAAK,GACJ,SAAI,EAAO,UAAY,EAChB,GAET,IAlBN,cAAA,eAAkB,EAAA,WAAQ,KAA1B,WACE,MAAO,MAAK,WAAc,MAAK,UAAY,GAAI,wCAsBjD,EAAA,UAAA,eAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAM,KAChB,EAAuB,GACvB,EAAkB,GAClB,EAAoB,GACpB,EAAQ,EAAQ,OAA8B,SAAC,EAAK,GACxD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,GAAG,WAAW,eACrB,EAAa,KAAK,EAAI,EAAK,OACtB,AAAI,EAAK,KAAO,QACrB,EAAQ,KAAK,EAAI,EAAK,OACjB,AAAI,GAAK,OAAS,MAAQ,EAAK,MAAM,SAAW,IACrD,EAAU,KAAK,EAAI,EAAK,OAEnB,GACN,IAEC,EAAiB,GACf,EAAkB,GACpB,EAA8C,GAC9C,EAA+C,GACnD,AAAI,GAAa,MACf,GAAqB,KAAK,oBAAoB,EAAU,QACxD,EAAsB,KAAK,oBAAoB,EAAU,UAE3D,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAMlC,AAAI,OAAO,KAAK,GAAqB,SAAW,EAC9C,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,AAAI,EAAK,SAAS,SAAW,GAC3B,EAAQ,KAAK,KAIjB,OAAO,KAAK,GAAqB,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,aAAe,EAAoB,GACxC,EAAQ,KAAK,MAKnB,AAAI,OAAO,KAAK,GAAoB,OAAS,EAC3C,OAAO,KAAK,GAAoB,QAAQ,SAAA,GAC/B,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GACF,GAAK,aAAe,EAAmB,GACvC,EAAO,KAAK,MAIhB,EAAS,EAGX,GAAI,GAAY,GAChB,AAAI,EAAM,SAAW,MAAQ,EAAM,QAAQ,UAAY,MACrD,GAAY,EAAM,QAAQ,SAAS,OAAO,SAAC,EAAW,GACpD,SAAU,EAAK,UAAU,MAAQ,EAAK,YAAY,GAC3C,GACN,KAGL,GAAM,GACF,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,EAAE,UAAS,GAExE,MAAI,GAAU,OAAS,GACrB,GAAO,UAAY,GAGd,GAGD,EAAA,UAAA,oBAAR,SAA4B,GAC1B,MAAO,QAAO,KAAK,GAAW,IACzB,OAAgC,SAAC,EAAM,GACtC,SAAK,EAAQ,GAAM,MAAQ,EACpB,GACN,KAGD,EAAA,UAAA,QAAR,SAAgB,GAGd,GAAM,GACF,GAAgB,EAAK,KAAO,KAAK,UAAU,EAAK,KAAO,GAC3D,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAGd,GAAM,GAAgB,CACpB,KAAM,EAAK,KACX,GAAI,EAAK,GACT,SAAU,EAAO,SACjB,WACK,GAAK,OACL,IAAI,IAAI,SAAA,GAAS,MAAA,GAAM,WAAW,KAAO,EAAM,OAAO,GAAK,IAChE,OAAQ,GACR,SAAU,GACV,YAAa,GACb,WAAY,GACZ,SAAU,EAAK,MAGjB,MAAI,GAAO,QAAU,MACnB,GAAQ,YACJ,EAAO,OAAO,OACV,SAAC,EAAK,GACJ,SAAI,EAAM,MAAQ,CAChB,KAAM,EAAM,KACZ,gBAAiB,EAAM,MACvB,cAAe,EAAM,KAEhB,GAET,KAEV,AAAI,EAAO,OAAS,MAClB,GAAQ,WACJ,EAAO,MAAM,OAAoC,SAAC,EAAK,GACrD,GAAM,GAAO,EAAM,KACf,EAAQ,OACZ,OAAQ,EAAM,UACP,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAChB,EAAM,cAAgB,GAC3B,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,aACA,UACH,cAEA,KAAM,IAAI,OACN,2BAA2B,EAAM,KAAI,YAAY,EAAK,IAE9D,SAAI,EAAM,MAAQ,CAAC,MAAK,EAAE,KAAI,GACvB,GACN,KAEF,GAID,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAU,EAAY,QACtB,EAAuB,GACvB,EAAkB,GACpB,EAA+B,GACnC,AAAI,GAAW,MACb,GAAQ,EAAQ,OAA8B,SAAC,EAAK,GAClD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,KAAO,SACd,EAAQ,KAAK,EAAI,EAAK,OAEjB,GACN,KAEL,GAAM,GAAiB,GACjB,EAAkB,GAExB,EAAY,UAAU,SAAS,QAAQ,SAAA,GAC9B,GAAA,GAAA,GAAA,EAAA,MAAA,GACD,EAAa,CACjB,KAAM,EACN,GAAI,cACJ,OAAQ,GACR,WAAY,GACZ,SAAU,QACV,YAAa,GACb,WAAY,CAAC,MAAO,CAAC,MAAO,GAAgB,EAAI,MAAO,KAAM,UAC7D,SAAU,IAEZ,EAAK,aAAe,EAAI,KACxB,EAAO,KAAK,GACZ,EAAM,GAAY,IAGpB,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAIlC,GAAM,GAAgB,EAAY,IAElC,EAAY,UAAU,UAAU,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,cAAgB,EACrB,EAAQ,KAAK,MAIjB,GAAM,GAAY,KAAK,mBAAmB,GAC1C,MAAO,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,IAG1D,EAAA,UAAA,mBAAR,SAA2B,GAA3B,GAAA,GAAA,KAEE,MAAO,CACL,WAAY,EAAY,UAAU,KAClC,OAAQ,EAAY,UAAU,SAAS,OACnC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,GACjC,GAET,IACJ,QAAS,EAAY,UAAU,UAAU,OACrC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,EAAK,EAAY,KAClD,GAET,MAIA,EAAA,UAAA,mBAAR,SACI,EACA,GACF,GAAI,GAAO,EAAI,KACf,MAAI,IAAW,MACb,GAAO,EAAQ,IAEV,CAAC,KAAI,EAAE,MAAO,EAAI,OAE7B,iBAE6B,GAC3B,GAAM,GAAS,EAAA,MAAM,OACrB,GAAI,MAAO,GAAO,MAAS,YACzB,MAAO,GAAO,KAAK,GACd,GAAI,MAAO,SAAW,YAC3B,MAAO,IAAI,QAAO,EAAM,UAAU,WAElC,KAAM,IAAI,OACN,gGAKyB,EAAc,GAC7C,GAAM,GACF,MAAM,QAAQ,GAAK,OAAO,aAAa,MAAM,KAAM,GAAK,GAAa,GACzE,MAAO,GAAW,EAAQ,EAAM,0BAI9B,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,KACJ,GAAiB,EAAM,EAAG,GAE5B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAO,GAAQ,EAAM,EAAI,cAIvB,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,IAAS,GACvB,EACF,EAAM,GAAQ,KAAO,EAAM,EAAQ,EAAM,GAAQ,KAAO,EAAM,EAAO,EACzE,MAAQ,OAAO,IAAU,SAAY,EAAQ,SAAS,EAAO,gBAG/B,GAC9B,AAAI,MAAQ,IAAW,UAErB,GAAQ,GAAoB,IAE9B,OAAQ,OACD,IAAoB,SACvB,MAAO,cACJ,IAAoB,aACpB,IAAoB,aACpB,IAAoB,YACpB,IAAoB,SACvB,MAAO,YACJ,IAAoB,QACvB,MAAO,WACJ,IAAoB,UACvB,MAAO,cACJ,IAAoB,UACvB,MAAO,iBAIP,MAAO,mBAKT,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,EAAM,KAAK,KAEb,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,GAAgB,EAAM,MAExB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,KAC7B,EAAM,KAAK,KAAK,IAAI,SAAA,GAAK,MAAA,IAAgB,KAE3C,cAG6B,GAEpC,MAAI,GAAM,YACR,OAEE,EAAM,KAAO,KACR,EAAM,IAAI,IACb,SAAA,GACI,MAAC,OAAO,GAAI,MAAS,SAAY,EAAI,KAAO,SAAS,EAAI,KAAM,MAElE,eAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MACV,GAAsB,EAAM,OAE9B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,GACO,IAAM,KAAK,GAAK,EAAM,KAAK,EAAE,OAAS,EAAM,KAAK,EACX,EAAM,KAAK,IAClD,IACH,IAAI,SAAA,GAAK,MAAC,OAAO,IAAM,SAAY,EAAI,SAAS,EAAG,MAEnD,cAIL,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAAE,IAAI,SAAC,GACvB,MAAO,IAAiB,EAAG,KAGxB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,MAC7B,EAAM,KAAK,MAAM,IAAI,SAAC,GAC3B,MAAO,IAAsB,KAG1B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAEb,ECtkBT,AA4BA,GAAA,IAAA,WAGE,WACY,EAAoB,EACpB,GAFZ,GAAA,GAAA,KACY,KAAA,KAAA,EAAoB,KAAA,UAAA,EACpB,KAAA,QAAA,EAJI,KAAA,OAAmB,GACnB,KAAA,MAAoC,GAIlD,KAAK,OAAS,EAAK,WAAW,IAAI,SAAA,GAAQ,MAAA,GAAK,SAAS,KACxD,AAAI,EAAK,UAAY,MACnB,MAAK,MAAQ,OAAO,KAAK,EAAK,UACZ,OAAO,SAAC,EAAmC,GAC1C,SAAM,GAAO,EAAK,QAAQ,GACnB,GACN,KAQhB,SAAA,UAAA,SAAR,SAAiB,GACf,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,UAOtC,EAAA,UAAA,QAAR,SAAgB,EAAc,GAC5B,GAAM,GAAQ,KAAK,KAAK,SAAS,GACjC,GAAI,EAAM,QAAU,KAClB,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,SAE9C,GAAI,EAAM,GAAK,MAAQ,EAAM,GAAK,KAChC,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAa,KAAK,KAAK,SAAU,EAAM,GAEhD,GAAI,EAAM,OAAS,KACjB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,MAAQ,KAChB,MAAO,IAAc,KAAK,KAAK,SAAU,EAAM,GAEjD,GAAI,EAAM,MAAQ,MAChB,GAAI,EAAM,KAAK,GAAK,MAAQ,EAAM,KAAK,GAAK,KAC1C,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,OAAS,KACtB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,MAAQ,KACrB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAIlC,MAAO,IAEX,KCtGA,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,cACA,YACA,MACH,MAAO,CAAC,EAAA,IACH,EAAc,IAAK,EAAM,EAAW,GACrC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,UAAW,EAAM,EAAW,SAE3C,eACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,cACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,oBACH,MAAO,CAAC,EAAA,kBACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC7F9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACA,aACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,OAAQ,EAAM,EAAW,SACxC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,YACA,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,eAAgB,EAAM,EAAW,GAC/C,EAAc,eAAgB,EAAM,EAAW,SAEhD,QACH,MAAO,CAAC,EAAA,MACJ,GAAU,EAAK,WAAW,GAAI,EAAW,SAC1C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SACxC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SACzC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAE5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC5K9C,YAyBI,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EAAA,KAAK,OACD,GAA8B,EAAQ,GACtC,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlB,EAAc,GAC1D,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,IAAM,EAAG,KAAO,IAAM,EAAG,KAAO,EAAG,GAC/C,MAAO,GAGX,MAAO,GCzCT,AA+BA,GAAA,IAAA,WAIE,WACa,EAAuB,EAAyB,EACjD,EAAiC,EAChC,EAA+B,GAF/B,KAAA,KAAA,EAAuB,KAAA,MAAA,EAAyB,KAAA,QAAA,EACjD,KAAA,aAAA,EAAiC,KAAA,uBAAA,EAChC,KAAA,YAAA,EAA+B,KAAA,eAAA,EANpC,KAAA,QAA6B,GAC7B,KAAA,QAAU,GAMhB,KAAK,SAAW,EAAA,OAAO,GACvB,EAAA,KAAK,KAAK,UAGZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAMd,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,OAAO,MAChD,EAAO,OAAO,YAGlB,KAAK,QAAU,GACf,KAAK,QAAU,GACf,KAAK,SAAS,WAGhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAOtB,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,GAAS,KAAK,OAC7B,KAAM,IAAI,OAAM,4BAA4B,EAAK,wBAC7C,KAAK,QAGX,GAAM,GAAkB,KAAK,QAAQ,GACrC,GAAI,EAAgB,QAClB,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0BACpB,EAAK,wGAIf,MAAI,MAAK,gBACP,GAAgB,QAAU,IAG5B,EAAgB,KAAO,GAChB,EAAgB,QAMzB,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GAAS,MAAA,GAAK,KAAK,MAQxC,EAAA,UAAA,MAAA,SAAM,EAAe,GACnB,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,CAAC,KAAK,aAAe,GAAS,KAAK,QAClD,KAAM,IAAI,OAAM,2BACZ,EAAK,8CAA8C,KAAK,SAG9D,GAAM,GAAI,KAAK,QAAQ,IAAU,GAEjC,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,eACZ,KAAK,KAAI,0CAA0C,EAAK;uCAExD,EAAO,MAAK,8BAA8B,KAAK,MAAK,KAc1D,GAVA,AAAI,KAAK,SAAW,GACf,MAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,IAC7D,MAAK,aAAe,EAAO,OAG7B,GACI,KAAK,aAAc,EAAO,MAC1B,eAAe,KAAK,KAAI,0CACpB,EAAK,KAET,EAAE,KACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,uCAGf,GAAI,EAAE,QACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,0CAGf,EAAE,OAAS,EACX,EAAA,KAAK,GACL,EAAE,QAAU,GAEZ,KAAK,QAAQ,GAAS,GAMxB,EAAA,UAAA,UAAA,SAAU,EAAmB,GAA7B,GAAA,GAAA,KACE,GAAI,EAAQ,SAAW,EAAQ,OAC7B,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,sCACxB,4BACI,EAAQ,OAAM,qCACd,EAAQ,OAAM,MAGxB,EAAQ,QAAQ,SAAC,EAAG,GAAU,MAAA,GAAK,MAAM,EAAG,EAAQ,OAWtD,EAAA,UAAA,OAAA,SAAO,EAAoB,GACzB,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAK,EAMH,EAAU,EAAQ,MAAM,EAAG,KAAK,aALhC,EAAU,GACV,OAAS,GAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAMjB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAKpC,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAAO,gCAElC,EAAA,MAAM,EAAS,IAMxB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAI,KAAK,SAAW,EAClB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAIpC,OADM,GAAU,GACP,EAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAGf,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAC9B,mDACI,KAAK,aAAY,4BAA4B,EAAQ,GAAG,MAAK,KAE9D,EAAA,OAAO,EAAS,IASzB,EAAA,UAAA,QAAA,SAAQ,EAAmB,GACzB,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAGhD,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,CAAC,KAAK,aAAe,GAAY,KAAK,QACxC,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,KAAK,QAAO,KAGtE,KAAK,UAAU,EAAS,EAAA,QAAQ,EAAQ,KAS1C,EAAA,UAAA,MAAA,SAAM,EAAkB,GAAxB,GAAA,GAAA,KACE,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAEhD,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEZ,EAAW,4BAA4B,EAAO,OAGpD,GAAI,CAAC,KAAK,aAAe,EAAO,SAAW,KAAK,QAC9C,KAAM,IAAI,OACN,2DACI,KAAK,QAAO,QAAQ,EAAO,OAAM,kEAI3C,GAAM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,GAC1B,EAAA,KAAK,WACH,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,EAAK,cAE3D,MAAO,KAGT,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAQ,GAAK,EAEf,KAAK,UAAU,EAAS,IAE5B,KC3TA,AAoCA,GAAA,IAAA,WAeE,WACa,EAA4B,EAC5B,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IADxB,KAAA,QAAA,EAA4B,KAAA,aAAA,EAC5B,KAAA,aAAA,EACX,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,IAAiB,EAAO,MAC1B,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,EAAO,OAEhD,GACI,EAAc,EAAO,MAAO,+BAEhC,EAAA,KAAK,KAGT,KAAK,SAAW,EAAA,OAAO,GACvB,KAAK,eAAiB,EACtB,EAAA,KAAK,KAAK,UA5BZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAiCvB,EAAA,UAAA,KAAA,WACE,MAAO,IAAI,GACH,KAAK,QAAO,QAAG,KAAK,aAAc,KAAK,eAMjD,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,MACzC,EAAO,YAGX,KAAK,QAAQ,OAAS,EACtB,KAAK,SAAS,WAKhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAUtB,EAAA,UAAA,MAAA,SAAM,EAAwB,EAAwB,GAAtD,GAAA,GAAA,KAEE,GAFoD,AAAA,IAAA,QAAA,GAAA,IAEhD,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,IAAgB,IAAM,KAAK,QAAQ,SAAW,EAChD,KAAM,IAAI,OAAM,kCACZ,EAAW,iCACX,KAAK,QAAQ,OAAM,cAEzB,UACI,EAAc,KAAK,aAAc,+BAC9B,EAAA,KAAK,WACV,GAAM,GACF,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAA,QAAQ,EAAQ,KAC/C,MAAO,GAAA,MAAM,EAAiB,MASlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAC9B,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAG9C,GAAI,KAAK,SAAW,EAClB,KAAM,IAAI,OAAM,qCAGlB,GAAM,GAAS,KAAK,QAAQ,MAC5B,UACI,EAAO,MAAO,EAAc,+BACzB,EAAA,QAAQ,EAAQ,IAOzB,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAM9C,GAHA,GACI,EAAO,MAAO,KAAK,aAAc,+BAEjC,KAAK,iBAAmB,KAAK,OAC/B,KAAM,IAAI,OAAM,4CAElB,EAAA,KAAK,GACL,KAAK,QAAQ,KAAK,IAOpB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAO,EACT,KAAM,IAAI,OACN,0DAA0D,GAGhE,GAAI,KAAK,iBAAmB,IAAM,EAAO,KAAK,eAC5C,KAAM,IAAI,OAAM,+BACZ,EAAI,6BAA6B,KAAK,eAAc,KAE1D,KAAK,QAAQ,OAAS,GASxB,EAAA,UAAA,QAAA,SAAQ,EAAsB,EAAwB,GAEpD,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,EAAe,GAAK,EAAe,KAAK,QAAQ,OAClD,KAAM,IAAI,OAAM,4BACZ,EAAY,mBAAmB,KAAK,QAAQ,OAAM,cAGxD,GAAI,KAAK,QAAQ,IAAiB,KAChC,KAAM,IAAI,OAAM,oBAAoB,EAAY,aAGlD,UACI,KAAK,QAAQ,GAAc,MAAO,EAClC,+BAEG,KAAK,QAAQ,IAQtB,EAAA,UAAA,QAAA,SAAQ,EAAsB,GAC5B,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAG9C,GAAI,EAAe,GACf,KAAK,iBAAmB,IAAM,GAAgB,KAAK,eACrD,KAAM,IAAI,OAAM,yBACZ,EAAY,uBAAuB,KAAK,eAAc,cAG5D,GACI,KAAK,aAAc,EAAO,MAAO,+BACrC,EAAA,KAAK,GACL,KAAK,QAAQ,GAAgB,GAU/B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAwB,GAAlD,GAAA,GAAA,KAEE,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAU9C,MAPA,IACI,KAAK,aAAc,EAAc,+BAIrC,EAAU,EAAQ,MAAM,EAAG,KAAK,QAE5B,EAAQ,SAAW,EACd,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAK,QAAQ,GAAI,KAC1D,MAAO,GAAA,MAAM,EAAS,MAS1B,EAAA,UAAA,OAAA,SAAO,EAAwB,GAA/B,GAAA,GAAA,KACE,GAAI,CAAC,CAAC,GAAgB,IAAiB,KAAK,aAC1C,KAAM,IAAI,OAAM,uBACZ,KAAK,aAAY,+BAA+B,GAMtD,MAHA,IACI,KAAK,aAAc,EAAc,+BAEjC,KAAK,SAAW,EACX,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAK,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAG,KACjD,MAAO,GAAA,OAAO,EAAS,MAG7B,iBAQI,EAAgB,EAAwB,GAC1C,GAAM,GAAQ,EAAO,MACrB,GAAI,EAAO,MAAM,OAAS,EACxB,KAAM,IAAI,OACN,oDAAoD,EAAO,OAEjE,GAAI,EAAO,QAAU,EACnB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,GAEzC,GAAM,GAAc,EAAO,MAAM,MAAM,GACvC,GACI,EAAa,EAAc,+BAE/B,GAAM,GAAuB,EAAA,QAAQ,GACrC,MAAO,IAAI,IAAW,EAAY,EAAc,eAU9C,EAAwB,EAAwB,GAClD,MAAO,IAAI,IAAW,GAAI,EAAc,EAAc,eAWpD,EAAgB,EAAmB,EACnC,GACF,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,GAAe,MAAQ,IAAgB,IAAM,GAAY,EAC3D,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,EAAW,KAGrE,GAAM,GAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,GACtD,EAAU,EAAA,QAAQ,EAAQ,GAChC,SAAQ,QAAQ,SAAC,EAAO,GACtB,EAAK,QAAQ,EAAO,EAAQ,MAEvB,cAWL,EAAgB,EAAkB,GACpC,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEV,EAAW,4BAA4B,EAAO,OAmBtD,OAhBM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,EAAA,KAAK,WAC7B,GAAM,GAAU,GAChB,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,GAEtD,SAAO,UACA,IAGH,EAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,EAAO,QAE1D,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAK,QAAQ,EAAG,EAAQ,IAE1B,MAAO,GC3YT,AAiBA,GAAA,IAAA,OAUa,GAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,kUACnB,EAAA,EAAK,iBACN,KAAA,MAAA,CAAA,EAAA,OACA,cAAA,MAAA,CAAA,EAAA,OAgBA,QAAA,MAAA,CAAA,EAAA,OACA,iBAAA,MAAA,CAAA,EAAA,OAsDA,WAAA,MAAA,CAAA,EAAA,OAIA,SAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QAOA,OAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAoBA,qBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAQA,sBAAA,MAAA,CAAA,EAAA,QAUA,uBAAA,MAAA,CAAA,EAAA,QAWA,sBAAA,MAAA,CAAA,EAAA,QAQA,qBAAA,MAAA,CAAA,EAAA,QAWA,oBAAA,MAAA,CAAA,EAAA,QAMA,qBAAA,MAAA,CAAA,EAAA,QAOA,oBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAaA,sBAAA,MAAA,CAAA,EAAA,QACA,oBAAA,MAAA,CAAA,EAAA,QAcA,oBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAYA,kBAAA,MAAA,CAAA,EAAA,QAYA,uBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAUA,qBAAA,MAAA,CAAA,EAAA,QASA,oBAAA,MAAA,CAAA,EAAA,QAUA,kBAAA,MAAA,CAAA,EAAA,wBAzTG,SACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClC,CAAA,EAAM,EAAK,eAC7B,MADM,GAAY,GAAA,OACd,EAAU,GACZ,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,gBAE1C,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAKtC,SACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAO,EAAc,OAAQ,EAAM,EAAW,GAI/C,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAFxC,SACD,GAAA,OAEC,EAAS,EAAK,IAAI,SAAA,IAAU,MAAA,IAAO,KACzB,CAAA,EAAM,EAAW,GAAG,eAAhC,EAAY,GAAA,OAEhB,EAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAChD,GAAO,YAIP,EAAmB,gFAIf,UAAa,EAEV,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACzC,EAAQ,EAAQ,eAAgB,EAAQ,uBAD5C,SAAS,GAAA,OAEH,GAAY,EAAO,IAAI,SAAA,IAAU,MAAA,IAAO,KAI9C,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,YAMN,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAQ,EAAQ,eAAgB,EAAQ,uBAF1C,UACD,GAAA,OAEO,CAAA,EAAM,GAAW,GAAG,eAAhC,SAAY,GAAA,OAEZ,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,4CA1BN,GAAU,WAAE,CAAA,EAAA,wCA8BnB,MAAA,CAAA,EAAO,UAGD,SAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAChD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClD,AAAK,EAAK,MACR,GAAO,GAAY,IAGb,CAAA,EAAM,EAAK,gBAAnB,MAAA,CAAA,EAAQ,GAAA,OAAmB,GAAK,CAAC,OAAW,GAAQ,CAAC,EAAM,iBAK3D,MAFM,GAAY,EAAK,WAAW,KAC9B,SAAA,IAAQ,MAAA,IAAU,GAAM,EAAW,KAAa,SAChD,EACI,GAAO,GAAU,EAAW,EAAW,GAC7C,CAAA,EAAO,CAAC,GAAY,MAEtB,CAAA,EAAO,gBAGD,SACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,WAAW,GACnB,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,YACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,gBACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,yBAA0B,EAAM,EAAW,GAEvD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAc,GAAI,IACpB,EAAM,EAAO,EAAM,EAAc,EAAwB,EACzD,GACJ,EAAQ,eAAe,GACvB,CAAA,EAAO,CAAC,EAAY,SAAU,EAAA,OAAO,aAG/B,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAmB,EAAQ,eAAe,EAAG,IACnD,EAAiB,MAAM,EAAO,GAC9B,CAAA,EAAO,CAAC,EAAiB,mBAGnB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAkB,EAAQ,eAAe,EAAO,IACtD,CAAA,EAAO,CAAC,EAAgB,KAAK,aAGvB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAoB,EAAQ,eAAe,EAAS,IAC1D,CAAA,EAAO,CAAC,EAAkB,OAAO,EAAe,aAG1C,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAqB,EAAQ,eAAe,GAAU,IAC5D,GAAmB,QAAQ,GAAgB,IAC3C,CAAA,EAAO,CAAC,GAAmB,mBAGrB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAoB,EAAQ,eAAe,GAAS,IACpD,GACF,EAAc,QAAS,EAAM,EAAW,GAC5C,CAAA,EAAO,CAAC,GAAkB,OAAO,cAG3B,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,MAAM,GAAS,IAChC,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAkB,EAAQ,eAAe,GAAO,IACtD,CAAA,EAAO,CAAC,EAAA,OAAO,GAAgB,OAAQ,mBAGjC,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,gBACjB,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,QAAQ,EAAO,GAC1B,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAE7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAW,EAAc,cAI9C,UACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GACF,GAAQ,GAAe,GAAgB,EAAc,IACzD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,GAAQ,EAAc,GAAc,IACvD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,EAAS,IAClD,CAAA,EAAO,CAAC,GAAW,OAAO,EAAe,GAAc,aAGjD,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,MAAM,EAAc,GAAc,cAG/C,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,GAAW,GAAQ,EAAc,IACpD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAC5C,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,CAAA,EAAO,CAAC,GAAW,OAAO,GAAa,aAGjC,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,SAAS,GACpB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAc,cAGnC,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,UAAW,EAAM,EAAW,GAExC,GAAa,GAAM,GAAa,GAAS,GAC/C,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGnB,KAAM,WAAU,aAAa,EAAK,GAAE,6BCvW1C,AA2BA,YACI,EAAY,EAA4B,GACpC,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAC7B,EAAc,IAAY,iBAE1B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,yGAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,oFAIR,GAAI,EACF,KAAM,IAAI,OACN,yEAEN,GAAM,GAAS,EAAc,UAAW,EAAM,EAAW,GACnD,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGhB,MAAO,CACL,OAAM,EACN,IAAG,EACH,WAAU,EACV,UAAS,EACT,QAAO,EACP,SAAQ,EACR,eAAc,GAIX,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAQ,EAAyB,EACjC,QAED,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,gBACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAIvB,8BACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,gBAAgB,CAClC,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAGvB,0BACA,mBACH,GAAM,GAAQ,EACI,cAAe,EAAM,EACrB,GAEZ,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GACxC,MAAO,CAAC,EAAA,gBACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAO,CAAC,EAAO,GAAI,EAAO,IAAK,QAEhC,4BACA,mBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cAET,MAAO,CAAC,EAAA,gBACJ,EAAc,QAAS,EAAM,EAAW,GAExC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GAEzC,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,EACnC,EACA,CAAC,EAAU,GAAI,EAAU,GAAI,EAAU,UAExC,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,qBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,sBAAuB,EAAM,EAAW,GAEpD,EAAA,EAAA,kBAAA,EAAA,IAAA,EAAA,EAAA,GAAA,CAAA,EAAA,GAAA,EAAA,IAAA,CAAA,EAAA,GAAA,EAAA,IAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,QAIf,MAAO,CAAC,EAAQ,OAEb,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,cACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,YAAa,EAAM,EAAW,GAG1C,EAAe,EAAQ,GACvB,EAAc,EAAQ,GAGtB,EAAiB,EAAU,GAC3B,EAAgB,EAAU,GAEhC,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAc,GAAc,EAC7B,CAAC,EAAgB,GAAgB,iBAIrC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC/S9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,KAAW,EAAO,EAAO,QAE9B,YACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAM,EAAc,MAAO,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,SAAe,EAAO,EAAM,QAEjC,eACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,YAAkB,EAAQ,EAAY,QAE3C,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OAAa,EAAS,EAAO,EAAS,QAE3C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cAEJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,SAEzC,SACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,MACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,SAGzC,mBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,gBACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,GAExC,QAED,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1H9C,AAiBA,GAAA,IAAA,OAUA,YACI,EAAY,EAA4B,GAC1C,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EAAS,EAAc,SAAU,EAAM,EAAW,GAClD,EACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CACL,MAAK,EACL,OAAM,EACN,cAAa,EACb,aAAY,EACZ,eAAc,EACd,aAAY,GAIT,GAAM,IAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,mHACnB,EAAA,EAAK,iBACN,sBAAA,MAAA,CAAA,EAAA,OAgBA,sBAAA,MAAA,CAAA,EAAA,OAcA,sBAAA,MAAA,CAAA,EAAA,OACA,sBAAA,MAAA,CAAA,EAAA,OAQA,QAAA,MAAA,CAAA,EAAA,OAQA,WAAA,MAAA,CAAA,EAAA,uBA9CG,SAOF,GAAU,EAAM,EAAW,GAN7B,EAAK,EAAA,MACL,EAAM,EAAA,OACN,EAAa,EAAA,cACb,EAAY,EAAA,aACZ,EAAc,EAAA,eACd,EAAY,EAAA,aAGC,CAAA,EAAM,EAAA,MAAY,gCAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,wBAGjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAG3D,EACF,EAAc,qBAAsB,EAAM,EAAW,GAG1C,CAAA,EAAM,EAAA,MAAY,6BAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,sBAIjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAGzD,CAAA,EAAM,EAAA,MAAY,uBACtB,EAAmB,EAAoB,EAAe,EACtD,WAFJ,MAAA,CAAA,EAAA,CAAQ,EAAA,gBAKF,SAAY,EAAA,KACb,EAAc,YAAa,EAAM,EAAW,GAC7C,QACY,CAAA,EAAM,EAAA,WAAiB,WAAjC,SAAM,CAAI,EAAA,QAChB,EAAU,UACV,CAAA,EAAO,UAGP,MAAA,CAAA,EAAO,EAAA,eACH,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,6BC3G1C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAS,EAAA,KAAW,EAAG,EAAG,GAChC,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAS,EAAA,OAAa,GAC5B,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAS,EAAA,OAAa,EAAG,GAC/B,MAAO,CAAC,EAAO,OAAQ,EAAO,iBAG9B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCpDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,GAAU,EAAK,UAEnB,yBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,IAAY,OACjD,cACH,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,QACrC,eACA,mBACA,2BACH,GAAM,GAAO,EAAc,IAAK,EAAM,EAAW,GACjD,MAAO,CAAC,GAAY,QAEjB,YACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,IAAY,SACjC,WACH,GAAM,GACD,EAAc,IAAK,EAAM,EAAW,GACzC,MAAO,CAAC,GAAY,QACjB,QACH,MAAO,CAAC,EAAA,SACH,EAAc,IAAK,EAAM,EAAW,GAAoB,MACzD,cACD,SACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,GAAA,SAAe,EAAE,aACtC,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OAAa,QAClB,QACH,GAAM,GAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,QAAQ,KACJ,kGAEJ,QAAQ,IAAI,GACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,QAAQ,IAAI,MAAM,UAAU,MAAM,KAAK,EAAK,GAAG,YAC9B,MAAM,EAAG,IAE5B,MAAO,CAAC,WAGR,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,eAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,yBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,sBAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,iBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,WAAY,EAAM,EAAW,GACzC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,qBAAsB,EAAM,EAAW,GAEzD,MAAO,CAAC,EAAA,MAAY,cAChB,EAAmB,EAAmB,EACtC,EAA8B,EAC9B,YAGJ,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzE9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,eACH,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACA,WACH,MAAO,CAAC,EAAA,MACJ,EAAc,YAAa,EAAM,EAAW,GAC5C,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACA,oBACA,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,aAAc,EAAM,EAAW,GAC7C,EAAc,aAAc,EAAM,EAAW,SAG9C,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SAExC,eACG,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAE7B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,sFAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,iEAGF,GAAA,GAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEhB,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,KAAM,EACN,WAAY,EACZ,uBAAwB,aAI1B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCjF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,qBACA,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,MACH,MAAO,CAAC,EAAA,2BACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,OAAQ,EAAM,EAAW,SAExC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cACJ,EAAc,gBAAiB,EAAM,EAAW,GAEhD,EAAc,cAAe,EAAM,EAAW,GAC9C,EAAc,eAAgB,EAAM,EAAW,GAE/C,EAAc,eAAgB,EAAM,EAAW,aAInD,KAAM,WAAU,aAAa,EAAK,GAAE,yBC9E9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,EAAW,YAGf,KAAM,WAAU,aAAa,EAAK,GAAE,yBCtH9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,eACA,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACvC,EACA,EAAc,UAAW,EAAM,EAAW,GAC9C,SAAS,EAAO,MAAM,EAAG,GAClB,CAAC,EAAA,OAAa,EAAQ,QAE1B,eACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OAAa,EAAO,EAAA,KAAW,EAAS,SAAU,QAEvD,gBACA,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,QAAc,EAAO,QAE1B,SAEH,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAEhD,EAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,gBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,MAAO,EAAM,EAAW,GACpC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAE/C,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,aACJ,EAAQ,EAAO,EAAK,EAAS,EAAW,EAAS,EACjD,EAAa,QAEd,OACH,MAAO,GAAA,KAAK,WACV,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GAGxC,EAAQ,EAAQ,GAAG,MACnB,EAAgB,EAAA,QAAc,EAAQ,IAAI,MAC1C,EAAS,EAAQ,IAAI,SAAA,GACzB,GAAM,GAAY,EAAA,KAAK,YAAY,EAAO,MAAO,GACjD,GAAI,CAAC,GACD,CAAC,EAAA,KAAK,YACF,EAAA,QAAc,GAAQ,MAAO,GACnC,KAAM,IAAI,OAAM,0CAElB,MAAO,GAAY,EAAS,EAAA,QAAc,EAAQ,KAEpD,MAAO,CAAC,EAAA,MAAY,EAAQ,UAG3B,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GAC7C,MAAO,GAAA,QAAc,EAAQ,OAE1B,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,YACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,kBAAmB,EAAM,EAAW,GAGhD,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,GAAA,MAAY,EAAQ,EAAiB,OAEzC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,UAAgB,EAAS,EAAQ,QAEtC,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,SAAe,EAAG,QAEvB,iBACH,GAAM,GACF,EAAc,gBAAiB,EAAM,EAAW,GAE9C,EACF,EAAc,cAAe,EAAM,EAAW,GAE5C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,MAAO,CAAC,EAAA,cACJ,EAAS,EAAc,EACvB,EAAa,QAAU,EAAa,MAChC,EACA,EAAA,KAAW,EAAc,EAAa,iBAG9C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1K9C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBChDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAGzC,cACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAGzD,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,UAAW,EAAM,EAAW,GAE1C,EAAc,gBAAiB,EAAM,EAAW,SAGjD,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,gBACH,GAAM,GACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACrC,cAEb,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAW,QAEZ,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAG5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnG9C,YAgDI,EAAY,EACZ,GACF,GAAM,GACD,SAAC,EAAY,EAA4B,GACxC,OAAQ,EAAK,cACN,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,cACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAsB,EAAM,EAAW,SAC9C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,UACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAkB,EAAM,EAAW,SACtD,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,gBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAwB,EAAM,EAAW,SAChD,YACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,iBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAyB,EAAM,EAAW,SACjD,SACH,GAAM,GAAW,GAAgB,EAAK,IACtC,GAAI,GAAY,EAAS,eACvB,MAAO,GAAS,eACZ,GAAI,IAAc,EAAM,EAAW,IAEvC,KAAM,WAAU,aAAa,EAAK,GAAE,+BAGtC,KAAM,WACF,eAAe,EAAK,GAAE,yIAI7B,EAAM,EAAW,GACxB,MAAI,aAAiB,SACZ,EAAM,KAAK,SAAC,GAAS,MAAA,GAAG,OAAO,KAEjC,GAAG,OAAO,GCvEnB,GAAA,IAAA,WAME,WACa,EACA,EACA,EACA,GAHA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IAHA,KAAA,UAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,YAAA,EATL,KAAA,YAAc,CAAC,GAAI,EAAG,UAAW,GAAI,YAAa,GAClD,KAAA,SAAmC,CAAC,KAAK,aACzC,KAAA,OAAS,EAQf,KAAK,4BAGC,SAAA,UAAA,SAAR,SAAiB,EAAY,GAC3B,MAAO,CAAC,GAAE,EAAE,UAAS,EAAE,YAAa,IAQtC,OAAA,eAAI,EAAA,UAAA,iBAAc,KAOlB,WACE,MAAO,MAAK,cARd,SAAmB,GACjB,AAAI,KAAK,WAAa,GACpB,MAAK,SAAW,EAChB,KAAK,8DAWT,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,mBAAmB,oCAOjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGN,EAAA,UAAA,0BAAR,WAEE,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,KAAK,SAAS,OAAS,EAAG,KAC5C,GAAM,GAAW,KAAK,SAAS,MAAM,EAAG,KAAK,SAAS,OAAS,GAC/D,EAAM,KAAK,KAAK,qBAAqB,IAEvC,EAAM,KAAK,IACX,KAAK,mBAAqB,GAGpB,EAAA,UAAA,qBAAR,SAA6B,GAC3B,MAAO,GACH,EACK,IACG,SAAA,GAAW,MAAC,GAAQ,KAAO,GAAK,EAAQ,cAAgB,EACpD,GACG,EAAQ,UAAS,IAAI,EAAQ,cACvC,KAAK,KACV,IAON,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,UACP,MAAK,SACL,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,KAAK,KAAK,SAAS,KAAK,OAAQ,IAC9C,KAAK,mBAAmB,QAAQ,KAAK,qBAAqB,KAAK,aAQnE,EAAA,UAAA,UAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,EAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,OAAO,IACrB,KAAK,kBAAkB,YAEvB,MAAM,IAAI,OAAM,4CAQpB,EAAA,UAAA,cAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,GAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SACL,GAAM,GACF,OAAO,OAAO,GAAI,KAAK,SAAS,KAAK,SAAS,OAAS,IAC3D,EAAQ,aAAe,EACvB,EAAQ,GAAK,KAAK,OAClB,KAAK,SAAS,OAAO,GAAI,EAAG,GAC5B,KAAK,mBAAmB,OACpB,EAAG,EAAG,KAAK,qBAAqB,KAAK,eAEzC,MAAM,IAAI,OAAM,0DAIpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,UAAU,IAGxB,EAAA,UAAA,eAAA,SAAe,GACb,KAAK,eAAe,EAAY,IAAM,GAGxC,EAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,eAAe,IAG7B,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,cAAc,EAAW,IAAM,GAGtC,EAAA,UAAA,cAAA,SAAc,GACZ,MAAO,MAAK,cAAc,IAG5B,EAAA,UAAA,QAAA,SAAQ,GACN,OAAW,KAAO,MAAK,eACrB,KAAK,eAAe,GAAK,cAAc,GAGzC,OAAW,KAAO,MAAK,cACrB,KAAK,cAAc,GAAK,cAAc,IAG5C,KC1LA,YAyCI,EAAwB,EAAiB,EACzC,GACF,GAAM,GAAY,GAAI,KAChB,EAA0B,GAC5B,EAAoB,KACpB,EAAuB,KAIrB,EAAO,GAAI,KACX,EACF,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAEpD,EAA0B,GAC9B,AAAI,GAAa,MACf,GAAgB,EAAU,IAAI,SAAA,GAAQ,MAAA,IAAc,EAAK,MAAM,MAIjE,OADM,GAAe,EAAO,QACrB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MAWtB,GAVA,AAAI,IAAc,IAAS,GAAe,KACxC,CAAI,GAAe,MACjB,GAAc,EACd,EAAa,EAAY,SAAS,IAAI,SAAA,GAAS,MAAA,GAAM,OACnC,OAAO,SAAA,GAAQ,MAAA,GAAU,IAAI,OAGnD,EAAU,IAAI,EAAK,MAGf,EAAU,EAAK,OAAS,KAC1B,SAGF,GAAI,EAAe,QAAQ,EAAK,QAAU,GACxC,SAGF,GAAI,EAAc,QAAQ,EAAK,QAAU,GACvC,SAEF,GAAI,EAAK,OAAO,SAAW,GACzB,EAAc,KAAK,EAAK,MACxB,SAEF,EAAK,OAAO,QAAQ,SAAA,GAElB,GAAI,EAAK,IAAI,EAAM,MACjB,OAEF,EAAK,IAAI,EAAM,MACf,EAAS,KAAK,KAGlB,MAAO,CAAC,OAAM,EAAE,QAAO,EAAE,UAAS,EAAE,cAAa,EAAE,YAAW,EAAE,WAAU,eAQxE,EAAc,EACd,GACK,GAAA,GAAA,EAAA,UAAW,EAAA,EAAA,OACZ,EAAmB,GACnB,EAAa,OAAO,KAAK,GACP,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAChC,IAAI,SAAA,GAAQ,MAAA,GAAM,MAAM,KAC1C,EAAY,EAAM,UAExB,EAAW,QAAQ,SAAA,GACjB,AAAI,EAAU,IAAI,EAAM,OACtB,EAAS,KAAK,KAGlB,EAAM,QAAQ,QAAQ,SAAA,GACpB,AAAI,EAAU,IAAI,EAAO,OACvB,EAAS,KAAK,KAGlB,AAAI,GAAa,MACf,EAAU,QAAQ,SAAA,GAChB,AAAI,EAAU,IAAI,EAAK,OACrB,EAAS,KAAK,KAMpB,OAFM,GAAO,GAAI,KACX,EAAuB,GACtB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MACtB,EAAK,IAAI,EAAK,MACd,AAAK,EAAU,EAAK,OAClB,EAAa,KAAK,GAEpB,EAAK,SAAS,QAAQ,SAAA,GACpB,AAAI,CAAC,EAAK,IAAI,EAAM,OAAS,EAAU,IAAI,EAAM,OAC7C,EAAM,OAAO,MAAM,SAAA,GAAS,MAAA,GAAK,IAAI,EAAM,SAC7C,EAAS,KAAK,KAIpB,MAAO,GAGT,GAAM,IAAmB,CACvB,SAAU,QAAS,QAAS,OAAQ,gBAAiB,cACrD,iBAAkB,KAAM,SAEpB,GAAoB,CACxB,sBAAuB,sBAAuB,sBAAuB,qBAGzC,GAC5B,MAAO,IAAiB,QAAQ,EAAK,KAAO,cAGf,GAC7B,MAAO,IAAkB,QAAQ,EAAK,KAAO,ECjK/C,AAkCA,GAAA,IAAA,WAsFE,WAAoB,EAAsB,GAA1C,GAAA,GAAA,KAAoB,KAAA,MAAA,EAAsB,KAAA,OAAA,EArFlC,KAAA,YAAmC,GAAI,KACvC,KAAA,WAA8B,GAM9B,KAAA,UAAY,IACZ,KAAA,WAAqC,GACrC,KAAA,qBAA0D,GA6EhE,KAAK,SAAW,EAAM,QACtB,KAAK,QAAU,EAAM,OACrB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UAExB,AAAI,EAAM,WAAa,MACrB,OAAO,KAAK,EAAM,WAAW,QAAQ,SAAA,GACnC,EAAK,qBAAqB,GACtB,GAAI,GAAc,EAAM,UAAU,GAAO,KApFnD,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,4CAGpD,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,oBACZ,KAAK,sDAG5B,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,gBAGpD,SAAc,GACZ,GAAM,GAAY,OAAO,KAAK,GAAW,IACrC,SAAA,GAAO,MAAA,GAAU,GAAK,IAAI,SAAA,GAAU,MAAA,GAAO,OAC/C,KAAK,WAAa,GAAG,OAAM,MAAT,GAAa,GAC/B,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GACtB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,IAAI,SAAA,GACvB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,cAAgB,EAAK,wCAG5D,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,IAAI,SAAC,GACxB,GAAM,GAAO,EAAK,cAAgB,EAAK,KACvC,MAAO,GAAK,cAAoB,EAAI,IAAI,EAAK,cAAmB,qCAIpE,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAAA,GAAA,GAAA,KACE,MAAO,QAAO,KAAK,KAAK,YAAY,OAAO,SAAC,EAAK,GAC/C,SAAI,GAAO,EAAK,WAAW,GAAK,UACzB,GACN,qCA0BG,EAAA,UAAA,kBAAR,SAA0B,EAAgB,GACxC,GAAM,GAAe,EAAO,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OAC7C,EAAgB,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OACrD,MAAO,GAAa,KAAK,KAAK,WAAa,KACvC,EAAc,KAAK,KAAK,YAOtB,EAAA,UAAA,QAAR,SAAgB,EAAwB,GACtC,GAAM,GACF,GAAqB,EAAQ,EAAS,KAAK,UAAW,KAAK,YACxD,EAAA,EAAA,cAAe,EAAA,EAAA,YAAa,EAAA,EAAA,WACnC,GAAI,GAAe,KACjB,KAAM,IAAI,OACN,qCAAqC,EAAY,KAAI,gBACrD,oBAAmB,EAAY,GAAE,kBACjC,6DACA,qCAAoC,EAAU,MAGpD,GAAI,EAAc,OAAS,GACzB,GAAM,GAAW,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,OAC9B,EAAU,OAAO,KAAK,GAC5B,KAAM,IAAI,OACN,+BAA+B,EAAQ,8BACvC,KAAI,EAAO,qCAAqC,EAAa,MAGnE,MAAO,IACH,KAAK,MAAO,KAAK,UAAW,IAYlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAAhC,GAAA,GAAA,KACE,EAAS,KAAK,UAAU,GACxB,GAAM,GAAQ,OAAO,KAAK,GAAQ,OAClC,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,GAClB,GAAM,GACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC5D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAG/D,AAAI,EAAY,SAAW,GACzB,GAAc,KAAK,UAGrB,GAAM,GAAiB,KAAK,kBAAkB,EAAY,GAGtD,EAAe,KAAK,YAAY,IAAI,GACxC,AAAI,GAAgB,MAClB,GAAe,KAAK,QAAQ,EAAQ,GACpC,KAAK,YAAY,IAAI,EAAgB,IAGvC,GAAM,GAAiC,GACjC,EAA+B,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,GAAI,IAChB,EAAK,UAAW,EAAgB,EAChC,EAAK,qBACH,EAAU,GAAA,GAAwB,EAAK,WAE7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAKzB,OAFM,GAAgB,EAAK,mBAAmB,GACxC,EAA2D,GACxD,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAO,EAAa,GAC1B,GAAI,CAAC,EAAW,EAAK,OACnB,GAAM,GAAU,GAAU,EAAM,EAAY,GAC5C,GAAI,YAAmB,SACrB,KAAM,IAAI,OACN,4BAA4B,EAAK,GAAE,kEAGzC,EAAW,EAAK,MAAQ,EACxB,EAAK,uBACD,EAAK,KAAM,EAAM,EAAY,EAAS,EACtC,EAAiB,IAIzB,MAAI,GAAK,QAAU,MACjB,EAAQ,QAAQ,GAEX,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAY,QAInD,EAAA,UAAA,mBAAR,SAA2B,GACzB,GAAM,GAAM,GAAG,OAAO,MAClB,GACA,OAAO,KAAK,GACP,IAAI,SAAA,GAAO,MAAA,GAAU,KACrB,IAAI,SAAA,GAAW,MAAA,GAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,QACrD,MAAO,IAAI,KAAI,IAET,EAAA,UAAA,uBAAR,SACI,EAAkB,EAAY,EAC9B,EAA2B,EAC3B,EACA,GAGF,GAAI,EAAK,WAAa,WAAa,EAAY,QAAQ,KAAc,GACnE,OAGF,EAAU,GAAU,QAAQ,SAAA,GAC1B,AAAI,GAAU,MACZ,GAAgC,EAAO,IAClC,GAAgC,EAAO,KAAO,GAC/C,EAAK,SAAS,UAGtB,EAAK,OAAO,QAAQ,SAAA,GAGlB,GAAI,EAAM,WAAa,WACrB,GAAM,GACF,GAA6B,EAAM,KAAM,EAAW,GACxD,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,GAAU,CAAC,EAAc,IAAI,EAAO,KACtC,GAAM,GAAQ,EAAgC,EAAO,IACrD,AAAI,IAAU,EACZ,GAAO,UACP,MAAO,GAAgC,EAAO,KACzC,AAAI,GAAS,MAGlB,EAAgC,EAAO,aAkB/C,EAAA,UAAA,aAAN,SAAmB,EAAwB,sEAEzC,MAAA,CAAA,EAAO,KAAK,cAAc,EAAQ,SAiBtB,EAAA,UAAA,cAAd,SACI,EAAwB,EAAmB,EAC3C,EACA,GAF2C,MAAA,KAAA,QAAA,GAAA,IAC3C,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,uGACF,MAAK,IACH,GAAS,KAAK,UAAU,GACxB,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,IAGd,EAAU,GAAI,IAChB,KAAK,UAAW,EAAgB,EAChC,KAAK,qBAKS,CAAA,EAAM,KAAK,uBACzB,EAAQ,EAAS,EAAS,WADxB,SAAY,EAAA,OAEZ,EAAU,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAGzD,EAAY,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAW,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,GAAO,GAAM,KACxD,EACF,GAAI,KAAgB,EAAS,OAAK,EAAa,KAAK,YACxD,OAAO,KAAK,GAAW,QAAQ,SAAA,GAC7B,GAAM,GAAc,EAAU,GAC9B,EAAY,QAAQ,SAAA,GAClB,AAAI,GAAU,CAAC,EAAO,YAAc,CAAC,EAAQ,IAAI,EAAO,KACtD,EAAO,cAKb,AAAI,KAAK,QAAU,MACjB,EAAQ,QAAQ,GAGlB,CAAA,EAAO,SAGH,EAAA,UAAA,qBAAN,SACI,EAAkB,EAClB,mFACI,SAAe,EAAO,OAAO,SAAC,EAAK,EAAQ,GAC/C,SAAI,EAAK,OAAO,GAAO,MAAQ,EACxB,GACN,IAEH,CAAA,EAAO,KAAK,cACR,EAAc,KAAK,YAAa,GAAM,EAAgB,SAS9C,EAAA,UAAA,uBAAd,SACI,EAAwB,EAA2B,EACnD,0IACI,EAAQ,OAAO,KAAK,GACpB,EACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAY,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC9D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAC3D,EACF,GAAqB,EAAQ,EAAa,KAAK,WAD5C,EAAS,EAAA,UAAE,EAAa,EAAA,cAAE,EAAW,EAAA,YAAE,EAAU,EAAA,WAGlD,EACE,EAAU,OAAK,KAAK,MAAM,SAAS,IAAI,SAAA,GACzC,MAAO,CAAC,KAAI,EAAE,SAAU,EAAQ,kBAEhC,EAAU,GAAA,GAAwB,KAAK,WAC7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAEnB,EAA2D,GAC3D,EAAgB,KAAK,mBAAmB,GACxC,EAAkC,0BACjC,GAAM,OAAS,EACd,GAAW,KAAK,aAClB,EAAY,EAAO,EAAS,EAAY,EAAO,EAC/C,EAAiB,EAAiC,GACtD,CAAA,EAAM,QAAQ,IAAI,KAJG,CAAA,EAAA,UAIrB,SAAA,oBAaF,GAXA,AAAI,GAAe,MAAQ,CAAC,GAC1B,QAAQ,KACJ,mIAGA,EACF,EACK,OACG,SAAA,GAAQ,MAAA,CAAC,GAAc,IACnB,CAAC,GAAU,EAAK,KAAM,EAAY,KACzC,IAAI,SAAA,GAAQ,MAAA,GAAK,OACtB,EAAe,OAAS,EACtB,QAAiB,GACrB,AAAI,GAAe,MACjB,GACI,gEACA,4BAA2B,EAAU,MAErC,GAAI,OACN,+BAA+B,EAAc,uBAC7C,YAAW,EAAK,gDAChB,KAAI,EAAa,MAAM,IAE7B,MAAA,CAAA,EAAO,SAGD,EAAA,UAAA,aAAR,SACI,EAAoB,EAA2B,EAC/C,EAA4B,EAC5B,EAA4B,EAC5B,EACA,GAEF,OAPF,GAAA,KAMQ,EAAqC,gBAEzC,GAAM,GAAO,EAAM,MACnB,EAAQ,eAAiB,EAAK,SAC9B,GAAI,GAAW,GAUf,GANA,AAAI,EAAK,KAAK,KAAO,SACjB,EAAc,aAAc,EAAK,KAAM,EAAW,IACnD,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAIC,EAAW,QAAQ,EAAK,QAAU,IACpC,GAAM,GAAU,GAAU,EAAK,KAAM,EAAW,GAChD,AAAK,GACF,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAEH,GAAM,GAAiB,EAAQ,eAC/B,AAAI,YAAmB,SACrB,EAAS,KAAK,EAAQ,KAAK,SAAA,GACzB,SAAU,GAAY,EACtB,EAAQ,eAAiB,EACzB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,GAC1C,KAGT,GAAU,GAAY,EACtB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,QAGnD,GAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,WAxC9C,EAAM,OAAS,OA2CtB,MAAO,IAGD,EAAA,UAAA,kBAAR,SACI,EAAY,EAA2B,EACvC,EAA4B,EAC5B,GACF,EAAK,SAAS,QAAQ,SAAC,GACd,GAAA,GAAA,GAAA,EAAA,KAAA,GAAA,GACP,GAAI,EAAM,IAAa,CAAC,EAAU,IAAI,EAAU,MAC9C,OAGF,AAAI,EAAU,KAAO,QACnB,AAAI,EAAU,WAAW,KAAK,SAAA,GACxB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAExC,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,KAGpD,AAAI,EAAU,WAAW,MAAM,SAAA,GACzB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAE5C,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,QAQ1D,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,OAAO,KAAK,KAAK,WACZ,QACG,SAAA,GAAO,MAAA,GAAK,UAAU,GAAK,QAAQ,SAAA,GAAU,MAAA,GAAO,eAGtD,EAAA,UAAA,uBAAR,SAA+B,GAA/B,GAAA,GAAA,KACE,OAAO,KAAK,GAAQ,QAAQ,SAAA,GAC1B,GAAM,GAAQ,EAAO,GACd,EAAA,GAAA,GAAA,GACD,EAAO,EAAK,MAAM,MAAM,GAC9B,GAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,GAAM,GAAQ,EAAK,WAAW,MAAS,MACjC,EAAQ,EAAM,SAAW,EAAM,MAAM,QACvC,EAAM,MAAM,MACR,SAAC,EAAK,GAAU,MAAA,GAAM,KAAW,IAAM,EAAM,KAAW,IAChE,EAAA,KAAK,OACD,EACA,WAAM,MAAA,sBAAsB,EAAK,KAAI,kBACjC,iCAAgC,EAAK,eACrC,KAAI,EAAM,MAAK,OAEzB,AAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,EAAA,KAAK,OACD,EAAM,QAAU,EAAK,WAAW,MAAS,MACzC,WAAM,MAAA,sBAAsB,EAAK,KAAI,8CAE9B,GAAK,WAAW,MAAS,MAAK,aAAa,EAAM,YAK1D,EAAA,UAAA,UAAR,SAAkB,GAChB,GAAM,GAAyB,GAC/B,OAAW,KAAa,GACtB,GAAI,KAAK,YAAc,MAAQ,KAAK,WAAW,QAAU,MACrD,KAAK,WAAW,OAAO,IAAc,MACvC,GAAM,GAAS,KAAK,WAAW,OAAO,GACtC,EAAO,EAAO,MAAQ,EAAO,OAE7B,GAAO,GAAa,EAAO,GAG/B,MAAO,IAGD,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAa,OAAO,KAAK,GAAQ,OAAO,SAAA,GACrC,GAAA,GAAA,GAAA,GAAA,GACP,MAAO,GAAK,MAAM,MAAM,IAAa,OAEvC,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,OACN,gDACA,WAAU,EAAU,kCAIpB,EAAA,UAAA,WAAR,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GACjB,GAAI,EAAK,YAAc,MAAQ,EAAK,WAAW,SAAW,MACtD,EAAK,WAAW,QAAQ,IAAS,MACnC,GAAM,GAAS,EAAK,WAAW,QAAQ,GACvC,MAAO,GAAO,KAEhB,MAAO,IACN,KAGG,EAAA,UAAA,aAAR,SAAqB,GAArB,GAAA,GAAA,KACE,EAAQ,QAAQ,SAAA,GACP,GAAA,GAAA,GAAA,GAAA,GACP,GAAI,CAAC,EAAK,MAAM,MAAM,GACpB,KAAM,IAAI,OAAM,eAAe,EAAI,kCAI3C,KC3lBA,AAyBO,GAAM,IAAqB,oBACrB,GAAqB,2BAoDhC,WACY,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IADA,KAAA,SAAA,EACA,KAAA,YAAA,EAzCJ,KAAA,QAAU,MA0ChB,AAAI,GAAe,MACjB,MAAK,YAAc,IArCvB,cAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,yCAGd,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,SAAS,4CAGvB,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,6CAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,SAAS,wCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,yCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,2CAoBf,EAAA,UAAA,cAAR,WACE,GAAM,GAAO,KAAK,SAClB,GAAK,EAAsB,MAAQ,KAEjC,KAAK,QAAU,UACN,KAAK,YAAY,aAAe,KACzC,KAAK,QAAU,EAAA,GAAG,mBAAmB,EAAgB,KAAK,kBAE1D,GAAM,GAAW,EAAA,GAAG,gBAAgB,EAAgB,KAAK,aACzD,GAAI,EAAS,SAAW,EAGtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAgB,KAAK,sBAChD,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,CAAC,GAAK,MAEpB,KAAK,QAAU,EAAS,KAQtB,EAAA,UAAA,KAAN,2GAEE,GADA,KAAK,gBACD,KAAK,QAAQ,MAAQ,KACvB,KAAM,IAAI,OACN,iHAGY,MAAA,CAAA,EAAM,KAAK,QAAQ,eAA/B,SAAY,EAAA,OAElB,CAAA,EAAO,KAAK,SAAS,UASvB,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,UAAY,EACjB,GAAM,GAAQ,KAAK,UAAU,cACzB,EAAY,GAChB,AAAI,KAAK,UAAU,qBAAuB,MACxC,GACK,KAAK,UAAU,oBAA4B,WAIlD,KAAK,QAAa,EAAM,SAAS,SAAQ,IAAI,EAAM,SAAS,YAC5D,GAAM,GACF,EAAA,GAAG,cAAc,KAAK,UAAU,WAAY,KAAK,UAAU,aAK/D,GAJA,KAAK,SAAW,GAAI,IAChB,GAAgB,SAAS,eAAe,EAAO,IACnD,KAAK,SAAS,UAAY,KAAK,6BAA6B,GAExD,EAAU,kBAAoB,MAChC,GAAM,GACF,GAAgB,SAAS,eAAe,EAAU,kBACtD,KAAK,YAAc,GAAI,IAAc,GACrC,KAAK,YAAY,UAAY,KAAK,SAAS,UAC3C,KAAK,YAAY,QAAQ,GAAI,IAG/B,MAAO,IA+CH,EAAA,UAAA,KAAN,SAAW,EAAmC,4EAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,OACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,OACN,+GAIN,MAAA,CAAA,EAAO,EAAa,KAAK,KAAK,iBAyChC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,MAAO,MAAK,QAAQ,EAAQ,KAAK,cAG3B,EAAA,UAAA,gBAAR,SAAwB,GAEtB,GAAI,CAAE,aAAkB,GAAA,SAAW,CAAC,MAAM,QAAQ,GAEhD,MAAO,GAGT,GADA,EAAS,MAAM,QAAQ,GAAU,EAAS,CAAC,GACvC,EAAO,SAAW,KAAK,WAAW,OACpC,KAAM,IAAI,OACN,+BACA,wBAAuB,KAAK,WAAW,OAAM,mBAC7C,oBAAmB,EAAO,OAAM,oBAEtC,MAAO,MAAK,WAAW,OAAO,SAAC,EAAK,EAAW,GAC7C,SAAI,GAAc,EAAoB,GAC/B,GACN,KAGG,EAAA,UAAA,iBAAR,SAAyB,GACvB,SAAU,GAAW,KAAK,YACnB,AAAC,MAAM,QAAQ,GAAuB,EAAZ,CAAC,IAmBpC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,EAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GAChC,GAAM,GAAS,KAAK,SAAS,QAAQ,EAAQ,GAC7C,MAAO,GAAO,OAAS,EAAI,EAAS,EAAO,IAkBvC,EAAA,UAAA,aAAN,SACI,EACA,mGACF,SAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GACjB,CAAA,EAAM,KAAK,SAAS,aAAa,EAAQ,WAAlD,SAAS,EAAA,OACf,CAAA,EAAO,EAAO,OAAS,EAAI,EAAS,EAAO,UAGrC,EAAA,UAAA,6BAAR,SAAqC,GACnC,MAAO,QAAO,KAAK,GAAK,OAAO,SAAC,EAAyB,GACvD,SAAO,GAAO,CAAC,EAAI,IACZ,GACN,KAQL,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UAEd,AAAI,KAAK,aACP,KAAK,YAAY,WAGvB,iBAiCI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,6FACF,GAAI,GAAY,KACd,KAAM,IAAI,OACN,0GAGN,MAAI,IAAW,MACb,GAAU,IAGZ,AAAI,EAAQ,WACV,CAAK,EAA0B,MAAQ,MACrC,CAAM,EAAoB,SAAS,MACjC,GAAY,EAAsB,KAEpC,EAAW,GAAG,EAAW,GAAqB,KAG5C,EAAQ,GAAI,IAAW,EAAU,GACvC,CAAA,EAAM,EAAM,eAAZ,SAAA,OACA,CAAA,EAAO,QC7ZT,GAGM,IAAU,4GCHhB,o8DC2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAEZ,GAAQ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,YAkDwB,EAAY,GAElC,MAAO,IAAgB,EAAO,GAShC,YACI,EAAY,EACZ,EAAiC,GAEnC,GAFE,AAAA,IAAA,QAAA,GAAA,GAA0B,MAAO,AAAA,IAAA,QAAA,GAAA,GAA2B,MAE1D,GAAS,KACX,MAAO,MAET,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAI,EAAK,IAAI,GACX,MAAO,GAAK,IAAI,GAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAGL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,GAChB,OAAW,KAAK,IACd,GAAM,GAAQ,EAAM,GACd,EAAc,GAAgB,EAAO,EAAO,EAAM,GACxD,EAAe,GAAK,EAEtB,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAdzD,UAAK,IAAI,EAAO,EAAO,OAChB,EAAO,kBA2Cd,EAAe,GAAA,MAAA,KAAA,QAAA,GAAA,IACV,GAAgB,EAAQ,GAOjC,YACI,EAAe,EACf,GAAA,AAAA,IAAA,QAAA,GAAA,GAA2B,MAG7B,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAEL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,kBACL,GACT,GAAM,GAAW,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC7B,EAAc,GAAgB,EAAU,EAAO,GACrD,EAAe,GAAK,GAHtB,OAAW,KAAK,KAAL,GAKX,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAbzD,OAAO,GAAO,kBAkBQ,GACxB,MAAI,KAAM,KACD,KAIL,GAAW,EAAE,IACR,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAG,QAAS,gBAsC3B,EAAY,+GACR,EAA+B,GAAI,KAGzC,GAAgB,EAAO,EAAO,OAMZ,EAAA,MAAM,KAAK,EAAK,+BAAhB,GAAA,EAAA,OAAP,GAAG,EAAA,GACN,EAAQ,EAAK,IAAI,GACnB,YAAiB,SACC,CAAA,EAAM,GADxB,CAAA,EAAA,IAFmC,CAAA,EAAA,UAG/B,EAAc,EAAA,OACpB,EAAK,IAAI,EAAK,oBAJA,wBAWZ,SAAS,GAAgB,EAAO,EAAO,GAC7C,CAAA,EAAO,oBASkB,GACzB,MAAO,IAAO,MAAS,CAAC,YAAY,OAAO,IACtC,OAAM,QAAQ,IACb,MAAO,IAAQ,UAAY,CAAE,aAAe,IAAA,qBAYvB,GAC3B,MAAO,IAAO,MAAQ,GAAY,IAAQ,MAAM,QAAQ,IACnD,MAAO,IAAQ,UAAa,YAAe,IAAA,QAC5C,GAAA,KAAQ,aAAa,GAO3B,YAAqB,GACnB,MACI,KAAU,MACT,MAAO,IAAU,UAAY,MAAO,IAAU,WCnRrD,YAqB6B,GAC3B,MAAO,IAAQ,EAAW,IAI5B,YAAuB,GACrB,MAAI,aAAgB,IAAA,OACV,CAAC,MAAO,EAAK,QAAS,QAAS,IAC9B,GAAW,GACb,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAM,QAAS,IChClC,AAqBA,GAAA,IAAA,WAcE,WAAmB,GACjB,GADiB,KAAA,SAAA,EAVT,KAAA,MAAQ,EACR,KAAA,IAAM,EAUV,GAAY,KACd,KAAM,IAAI,YAAW,mDAEvB,GAAI,EAAW,EACb,KAAM,IAAI,YAAW,6CAEvB,KAAK,KAAO,GAAI,OAAS,GACzB,KAAK,gBAAkB,EAAI,EAMnB,SAAA,UAAA,KAAV,SAAe,GAEb,KAAO,EAAQ,GACb,GAAS,KAAK,gBAEhB,MAAO,GAAQ,KAAK,iBAGZ,EAAA,UAAA,IAAV,SAAc,GACZ,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,MAAO,MAAK,KAAK,EAAQ,KAAK,WAGtB,EAAA,UAAA,IAAV,SAAc,EAAe,GAC3B,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,KAAK,KAAK,EAAQ,KAAK,UAAY,GAMrC,EAAA,UAAA,OAAA,WACE,GAAI,GAAS,KAAK,IAAM,KAAK,MAC7B,MAAI,GAAS,GACX,GAAS,KAAK,gBAAkB,GAE3B,GAQT,EAAA,UAAA,OAAA,WACE,MAAO,MAAK,WAAa,KAAK,UAQhC,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,WAAa,GAM3B,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,IAAI,KAAK,IAAK,GACnB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,IAMlC,EAAA,UAAA,QAAA,SAAQ,GACN,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,KAAK,KAOd,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,GAChC,GAAM,GAAS,KAAK,IAAI,KAAK,KAC7B,YAAK,IAAI,KAAK,IAAK,QACZ,GAMT,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GACpC,KAAK,IAAI,KAAK,MAAO,IAMvB,EAAA,UAAA,MAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAS,KAAK,IAAI,KAAK,OAC7B,YAAK,IAAI,KAAK,MAAO,QACrB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GAC7B,GAYT,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAQ,KAAK,KAAK,KAAK,MAAQ,GAC/B,EAAS,KAAK,IAAI,GACxB,YAAK,IAAI,EAAO,KAAK,OACd,GAEX,KC/KA,AAoBA,GAAA,IAAA,SAAA,GAA0C,GAAA,EAAA,GAMxC,mBACE,GAAA,KAAA,KAAM,EAAkB,mBAAiB,KAG3C,SAAA,UAAA,OAAA,WACE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,GACH,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,KAAI,KAAA,KAAC,IAGb,EAAA,UAAA,QAAA,SAAQ,GACN,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,QAAO,KAAA,KAAC,IAMR,EAAA,UAAA,OAAR,WAOE,OANM,GAAc,KAAK,SAAW,EAC9B,EAAU,GAAI,OAAS,GACvB,EAAM,KAAK,SAIR,EAAI,EAAG,EAAI,EAAK,IACvB,EAAQ,GAAK,KAAK,IAAI,KAAK,KAAK,KAAK,MAAQ,IAG/C,KAAK,KAAO,EACZ,KAAK,SAAW,EAChB,KAAK,gBAAkB,EAAI,KAAK,SAChC,KAAK,MAAQ,EACb,KAAK,IAAM,GA7CE,EAAA,iBAAmB,GA+CpC,GAhD0C,ICpB1C,YAuCqC,GACnC,MAAO,IAAI,IAAc,eAyBvB,GAEF,MAAO,IAAI,IAAqB,eAgB9B,EACA,GACF,MAAO,IAAI,IAAgB,EAAe,eAmDxC,EACA,GAAA,MAAA,KAAA,QAAA,GAAgC,GAAgB,MAC3C,GAAI,IAAe,EAAW,GAUvC,GAAA,IAAA,WAAA,cAwBQ,SAAA,UAAA,QAAN,6GACQ,SAAc,GACZ,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,KAAK,gBAAf,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAcH,EAAA,UAAA,eAAN,+GACQ,SAAS,KAAK,SAAS,KACvB,EAAc,GACZ,CAAA,EAAM,EAAO,eAAjB,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,EAAO,gBAAjB,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAUH,EAAA,UAAA,aAAN,2GACU,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACR,CAAA,EAAM,KAAK,eAAf,SAAI,EAAA,oCAWF,EAAA,UAAA,aAAN,SAAmB,qGACT,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,8BACzB,CAAC,EAAE,MAAS,EACd,CAAA,EAAM,KAAK,QADiB,CAAA,EAAA,UAChC,SAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,oCAgBjC,EAAA,UAAA,aAAA,SAAa,GACX,MAAO,IAAI,IAA0B,KAAM,IAa7C,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,IAAI,IAAe,KAAM,IAWlC,EAAA,UAAA,IAAA,SAAO,GACL,MAAO,IAAI,IAAY,KAAM,IAW/B,EAAA,UAAA,SAAA,SAAY,GACV,MAAO,IAAI,IAAiB,KAAM,IAWpC,EAAA,UAAA,eAAA,SAAkB,GAChB,MAAO,IAAI,IAAiB,KAAM,GAAW,UAW/C,EAAA,UAAA,QAAA,SAAW,GACT,MAAO,IAAI,IAAgB,KAAM,IAQ7B,EAAA,UAAA,aAAN,SAAmB,sEACjB,MAAA,CAAA,EAAO,KAAK,IAAI,GAAG,qBAUf,EAAA,UAAA,cAAN,SAAoB,sEAClB,MAAA,CAAA,EAAO,KAAK,eAAe,GAAG,aAAa,SAAA,GAAK,MAAC,KAAM,WAqBzD,EAAA,UAAA,cAAA,SAAc,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,IACxB,GAAI,IAAsB,KAAM,EAAW,IAmCpD,EAAA,UAAA,iBAAA,SACI,EAAmB,EAEnB,GAFmB,AAAA,IAAA,QAAA,GAAA,IAEnB,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAAa,KAAK,cAAc,EAAW,GAGjD,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,MAaxC,EAAA,UAAA,YAAA,SACI,EACA,GACF,MAAO,IAAI,IACP,GAAkB,CAAC,KAAM,IAAY,IAU3C,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAShC,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAYhC,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,IAAI,IAAiB,KAAM,IAapC,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAC1B,MAAO,IAAI,IAAgB,KAAM,EAAY,IAO/C,EAAA,UAAA,OAAA,WACE,MAAO,IAAI,IAAe,OAE9B,KAWA,GAAA,SAAA,GAA+B,GAAA,EAAA,GAE7B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,MAAA,EADd,EAAA,KAAO,IAKf,SAAA,UAAA,QAAA,WACE,MAAO,YAAY,KAAK,MAAM,OAAM,UAGhC,EAAA,UAAA,KAAN,oFACE,MAAI,MAAK,MAAQ,KAAK,MAAM,OAC1B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAO,KAAK,MAAM,KAAK,MAC7B,KAAK,OACL,CAAA,EAAO,CAAC,MAAO,GAAU,GAAO,KAAM,WAE1C,GAlB+B,IAoB/B,GAAA,SAAA,GAAsC,GAAA,EAAA,GACpC,WACc,GADd,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,OAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAO,iBAGH,EAAA,UAAA,KAAN,8EACE,IACE,MAAA,CAAA,EAAO,KAAK,gBACL,GAEP,QAAE,QACE,mDAAmD,EAAE,QACnD,iBAGZ,GApBsC,IAsBtC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,8EACE,MAAA,CAAA,EAAO,KAAK,SAAS,aAEzB,GA1BgC,IA4BhC,GAAA,SAAA,GAA8B,GAAA,EAAA,GAQ5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAF3D,EAAA,MAAQ,EAIN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,iHAKS,MAAK,QAAU,KAAK,SACT,CAAA,EAAM,KAAK,SAAS,QADH,CAAA,EAAA,UAGjC,MAFM,GAAU,EAAA,OAEZ,EAAQ,KACV,CAAA,EAAO,GAET,IAAA,QAAW,EAAQ,qBAErB,MAAA,CAAA,EAAO,KAAK,SAAS,cAEzB,GAzC8B,IA2C9B,GAAA,SAAA,GAA8B,GAAA,EAAA,GAE5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAD3D,EAAA,MAAQ,IAKR,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,8EACE,MAAI,MAAK,SAAW,KAAK,SACvB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,CAAA,EAAO,KAAK,SAAS,aAEzB,GAhB8B,IAqB9B,GAAA,SAAA,GAAuC,GAAA,EAAA,GAKrC,WACc,EAAqC,EACrC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EAAqC,EAAA,UAAA,EACrC,EAAA,qBAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,qBAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,6GACQ,EAAa,0BACZ,GAAM,OAAS,KAAK,UACZ,CAAA,EAAM,KAAK,SAAS,QADC,CAAA,EAAA,UAElC,MADM,GAAO,EAAA,OACT,EAAK,KACH,KAAK,sBAAwB,EAAM,OAAS,EAC9C,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,KAE9B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAM,KAAK,EAAK,qBAElB,MAAA,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,WAEhC,GAvCuC,IAyCvC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,2GAEiB,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,MAAQ,KAAK,UAAU,EAAK,OACnC,CAAA,EAAO,GAET,IAAA,QAAW,EAAK,qCAGtB,GAlCgC,IAoChC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAC9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,WAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAc7B,IAZM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAOzD,EAAS,KAAK,UAAU,EAAK,OAC7B,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCgC,IAqChC,GAAA,SAAA,GAA2C,GAAA,EAAA,GAEzC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,QAAA,EAHd,EAAA,MAAQ,EAKN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,oBAO7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGR,EAAA,UAAA,WAAN,yJAGa,CAAA,EAAM,KAAK,SAAS,eAA3B,MAAA,CAAA,EAAO,EAAA,eAEP,kBAAI,AAAC,KAAK,QAAQ,SAChB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,+CAUrC,GA1C2C,IA4C3C,GAAA,SAAA,GAAqC,GAAA,EAAA,GACnC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,KACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAe,GAAA,YAAe,sBAAsB,EAAK,OAOhD,CAAA,EAAM,KAAK,UAAU,EAAK,gBAKzC,IALM,EAAS,EAAA,OACT,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCqC,IAgDrC,GAAA,SAAA,GAAmD,GAAA,EAAA,GAQjD,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,YAAc,GAAI,IACvB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAiBR,EAAA,UAAA,WAAN,2GAIS,MAAK,YAAY,WAAa,EAE9B,CAAA,EAAM,KAAK,QAFoB,CAAA,EAAA,UAEpC,MAAK,GAAA,aACH,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,YAG/B,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,YAAY,QAAS,KAAM,WAEnD,GAjDmD,IAkDnD,GAAA,SAAA,GAAoC,GAAA,EAAA,GAClC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,eAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,IAeT,IAbM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAMzD,EAAc,KAAK,UAAU,EAAK,OAClC,EACF,GAAA,YAAe,sBAAsB,GACzC,KAAK,YAAY,QAAQ,GAIzB,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAIN,MAAA,CAAA,EAAO,UAEX,GArCoC,IAgDpC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAStC,WACI,EACiB,GAFrB,GAAA,GAGE,EAAA,KAAA,OAAO,KADY,SAAA,iBAAA,EARb,EAAA,SAAuC,KAGvC,EAAA,SAA4B,KAOlC,EAAK,cAAgB,IAGvB,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,8CAC1B,MAAU,GAAiB,eAGvB,EAAA,UAAA,KAAN,8EACE,YAAK,SAAW,KAAK,cAAc,KAAK,UACxC,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,cAAd,SAA4B,qGAO1B,MAAA,CAAA,EAAM,gBAAN,GAAA,OACI,KAAK,UAAY,KACI,CAAA,EAAM,KAAK,cAAc,QAD9C,CAAA,EAAA,UAEF,GADM,EAAiB,EAAA,OACnB,EAAe,KAEjB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,KAAK,SAAW,EAAe,MAC/B,AAAI,KAAK,kBAAoB,MAC3B,MAAK,SAAW,KAAK,SAAS,aAAa,KAAK,oCAGjC,MAAA,CAAA,EAAM,KAAK,SAAS,eACvC,MADM,GAAa,EAAA,OACf,EAAW,KACb,MAAK,SAAW,KAChB,CAAA,EAAO,KAAK,cAAc,KAE5B,CAAA,EAAO,SAEX,GApDwC,IAsD5B,GAAZ,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,YAHU,IAAA,IAAe,KAmC3B,GAAA,IAAA,SAAA,GAAwD,GAAA,EAAA,GAItD,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAgC,GAAgB,MAFvE,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,UAAA,EACA,EAAA,aAAA,EALf,EAAA,MAAQ,EACR,EAAA,eAA6C,OAQrD,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,0CAC1B,MAAO,IAAI,EAAiB,YAGhB,EAAA,UAAA,UAAd,SAAwB,2CAWtB,WAAiB,GACf,GAAI,YAAqB,KACvB,GAAM,GAAS,EAAU,OACzB,MAAO,CACL,MAAO,EAAO,KAAK,SAAA,GACjB,WACA,AAAI,EAAE,MACJ,IAEK,EAAE,QAEX,QAAS,QAGX,OAAO,CAAC,MAAO,KAAM,QAAS,gEArBlC,MAAA,CAAA,EAAM,UAAN,SAAA,OAII,EAAe,EACf,EAAgB,EAoBF,CAAA,EAAM,GAAmB,KAAK,UAAW,WAE3D,GAFM,EAAY,EAAA,OAEd,IAAiB,EAEnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAI,EAAgB,EAClB,OAAQ,KAAK,kBACN,IAAgB,KACnB,KAAM,IAAI,OACN,+CACA,0BAAyB,KAAK,MAAK,UACpC,IAAgB,SACnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,SACxB,IAAgB,SAMzB,YAAK,QACL,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAGzB,EAAA,UAAA,KAAN,8EACE,YAAK,eAAiB,KAAK,UAAU,KAAK,gBAC1C,CAAA,EAAO,KAAK,qBAEhB,GAxEwD,IAoFxD,GAAA,SAAA,GAAyC,GAAA,EAAA,GAGvC,WACc,EAAqC,GADnD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAAqC,EAAA,WAAA,EAEjD,EAAK,OAAS,GAAI,IAAuC,KAG3D,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAOzB,EAAA,UAAA,OAAV,WACE,KAAO,CAAC,KAAK,OAAO,WAClB,GAAM,GAAI,KAAK,SAAS,OACxB,KAAK,OAAO,KAAK,KAIrB,EAAA,UAAA,KAAA,WACE,YAAK,SAIE,KAAK,OAAO,SAEvB,GA/ByC,IAuCzC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAUtC,WACc,EAAqC,EAC/C,GAFJ,GAAA,GAGE,EAAA,KAAA,KAAM,EAAU,IAAW,KAFf,SAAA,SAAA,EAAqC,EAAA,WAAA,EAH3C,EAAA,kBAAoB,GAM1B,EAAK,OAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACpD,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGN,EAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,MAAK,MAAM,KAAK,SAAW,IAG1B,EAAA,UAAA,YAAV,WACE,MAAO,MAAK,UAAU,KAAK,OAAO,WAG9B,EAAA,UAAA,WAAN,6GAEE,AAAK,KAAK,mBACR,KAAK,gCAEC,MAAK,OAAO,UAAS,CAAA,EAAA,GACrB,GAAc,KAAK,cACV,CAAA,EAAM,KAAK,OAAO,cAAc,YAC/C,GADM,EAAS,EAAA,OACX,EAAO,KACT,KAAK,kBAAoB,OAEzB,aAAK,SACL,CAAA,EAAO,sBAGX,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,WAE/B,GApDwC,ICjnCxC,kBA6DA,aAWW,KAAA,KAAe,KA+DxB,SAAA,UAAA,MAAA,SAAM,EAAmB,GAAzB,GAAA,GAAA,KAAyB,AAAA,IAAA,QAAA,GAAA,IACvB,GAAM,GAAO,KACb,GAAA,KAAQ,OACJ,EAAY,EAAG,WAAM,MAAA;QACrB,IACJ,GAAI,GACJ,MAAI,MAAK,OAAS,UAAY,KAAK,MAAQ,KAGzC,EAAO,KAAK,KACP,AAAI,EAGT,EAAO,KAAK,KAAK,KAAK,KAAO,GAI7B,EAAO,KAAK,MAAM,KAAK,KAAO,GAEzB,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OACH,iBAAiB,EAAW,EAAgB,WAChD,IAkBL,EAAA,UAAA,YAAA,SAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,UAAY,EAAQ,OAAS,SAG7C,EAAO,SACF,AAAI,KAAK,MAAQ,MAAQ,EAAQ,MAAQ,KAG9C,EAAO,KAAK,KAAO,EAAQ,KAI3B,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,EAAA,MAAA,IAAA,KAAA,SAAA,0BACK,MAAA,CAAA,EAAM,EAAK,mBAAZ,SAAA,GAAC,EAAA,QAAuB,YAAY,CAAA,EAAM,EAAQ,mBAAlD,MAAA,CAAA,EAAA,EAAA,MAAA,EAAA,CAAoC,EAAA,gBACxC,IAmBN,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,SAEhB,EAAO,SAIP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,OAAO,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAClE,IAmBC,EAAA,UAAA,aAAN,SAAmB,6FACT,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,aAAa,UAkB9C,EAAA,UAAA,IAAA,SAAkC,GAAlC,GAAA,GAAA,KACQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,IAAI,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAC/D,KAAK,OA0BV,EAAA,UAAA,SAAA,SAAuC,GAAvC,GAAA,GAAA,KAEQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,SAAS,UACvC,KAAK,OAYV,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,GAAI,GAAc,KAChB,KAAM,IAAI,YACN,6DAGN,GAAM,GAAO,KACb,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,SAAS,UAAa,KAAK,OAqBrE,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,EAAQ,EAI/B,EAAO,KAAK,KAAO,EACd,AAAI,IAAU,EAEnB,EAAO,EACF,AAAI,KAAK,MAAQ,MAAS,KAAU,QAAa,EAAQ,GAG9D,EAAO,SAGP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mDACrB,SAAmB,GACrB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,MAAA,IAAA,KAAA,SAAA,sCAAqB,CAAA,EAAM,EAAK,mBAApB,MAAA,CAAA,EAAE,GAAA,MAAO,EAAA,OAAuB,EAAA,KAAM,GAAK,WAC3D,CAAA,EAAO,GAAyB,EAAiB,KAAK,UACrD,IAoBL,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,GAAS,GAAK,KAAK,MAAQ,EAIlD,EAAO,KAAK,KAAO,EACd,AACH,KAAK,MAAQ,MACZ,MAAK,KAAO,GAAS,IAAU,QAAa,EAAQ,GAGvD,EAAO,EAGP,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IA4BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,EAAe,GAA3C,GAAA,GAAA,KAEE,GAFyC,AAAA,IAAA,QAAA,GAAA,IAErC,GAAc,MAAQ,EAAa,EACrC,KAAI,MAAK,MAAQ,KACT,GAAI,YACN,4DAEE,GAAI,YACN,+KAGA,oCAAmC,KAAK,KAAI,eAGpD,GAAM,GAAO,KACP,EAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACrD,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACvB,SAAQ,EAAO,QACnB,AAAI,GACF,IAAS,EAAO,SAEV,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,QAAQ,EAAY,EAAM,mBACxD,KAAK,OAoBV,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,KAAK,KAAO,EAGnC,EAAO,EACF,AAAI,KAAK,MAAQ,MAAQ,KAAK,MAAQ,EAG3C,EAAO,KAAK,KAGZ,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IAmBjD,EAAA,UAAA,QAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,iBAc3B,EAAA,UAAA,eAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,wBA1HjB,EAAA,gBAAkB,IA4HpC,iBAeI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,GAAA,UAAA,GAAkB,GAAA,EAAA,GAAd,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KACT,SAAA,KAAO,IAMD,SAAA,UAAA,SAAN,8EACE,MAAA,CAAA,EAAO,UAEX,GAVyB,iBAgCyB,GAApD,GAAA,GAAA,KACE,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,GAAkB,SAAQ,EAAM,oBA4CA,GAAlD,GAAA,GAAA,KAGE,GAAI,CAAC,GAAW,GACd,KAAM,IAAI,OAAM,qDAElB,GAAI,GACJ,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAO,GAAQ,KAAQ,EAAS,GAAkB,KAC5B,KAAK,IAAI,EAAO,EAAS,GAAkB,cAE1D,YAAoB,QAC7B,OAAW,KAAM,GACf,EAAO,GAAQ,KAAQ,EAAS,GAAmB,KAC7B,KAAK,IAAI,EAAO,EAAS,GAAmB,MAGtE,MAAO,IAAyB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACd,MAAA,CAAA,EAAM,GAAmB,EAAU,SAAA,GACjD,GAAI,YAAa,IACf,MAAO,CAAC,MAAO,EAAE,WAAY,QAAS,IACjC,GAAI,GAAW,GACpB,MAAO,CAAC,MAAO,KAAM,QAAS,IAE9B,KAAM,IAAI,OACN,uFAPF,SAAU,EAAA,OAWhB,CAAA,EAAO,GAAsB,EAAS,GAAgB,iBACrD,GAWL,YAAyB,GACvB,GAAI,IAAS,KACX,MAAO,MAIT,GAAM,GAAa,EAAK,GAExB,GAAI,GAAa,IAEf,GAAM,GAAQ,GAAY,GAC1B,MAAO,CAAC,MAAK,EAAE,QAAS,IAI1B,MAAO,CAAC,MAAO,KAAM,QAAS,IAOhC,YAAwD,GAEtD,GAAI,EAAO,SAAW,EAEpB,KAAM,IAAI,OAAM,wCAGlB,MAAI,GAAO,YAAc,IAAA,OAEhB,GAAA,MAAS,GAGT,GAAA,OAAU,GC9rBrB,mBA2BqC,GAAA,EAAA,GAMnC,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,IAIzB,SAAA,UAAA,SAAN,+GACwB,MAAA,CAAA,EAAM,KAAK,MAAM,mBAAjC,SAAgB,EAAA,OAChB,EAAe,EAAc,aAC7B,EAAe,EAAa,MAAM;GAAM,IAAI,SAAA,GAEhD,MAAI,GAAK,SAAS,OAChB,GAAO,EAAK,MAAM,EAAG,KAEhB,IAET,CAAA,EAAO,SAEX,GAtBqC,IC3BrC,AAyBA,GAAM,IAAa,IACb,GAAY,OAAO,OACnB,GAAc,OAAO,SACrB,GAAc,OAAO,SACrB,GAA0B,OAAO,mBACjC,GAA8B,OAAO,+BAeX,GAAA,EAAA,GAgI9B,WAA+B,EAAmB,GAAlD,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,EA9HvB,EAAA,UAAY,GACZ,EAAA,gBAA4B,KAC5B,EAAA,qBAAuB,GACvB,EAAA,cAA+C,KAC/C,EAAA,sBAAwB,GACxB,EAAA,UAAY,IACZ,EAAA,gBAAkB,GA0HxB,EAAK,KAAO,GAAI,IAAgB,GAChC,AAAK,GACH,GAAY,IAEd,EAAK,UAAY,IAAU,YAAc,IACzC,EAAK,gBAAkB,EAAU,YACjC,EAAK,cAAgB,EAAU,cAC/B,EAAK,sBAAwB,EAAU,sBACvC,AAAI,EAAU,gBACZ,IAAA,KAAK,OACD,EAAU,WAAa,KACvB,WACI,MAAA,mEACR,EAAK,gBAAkB,GACvB,EAAK,UAAY,KAEjB,EAAK,UAAY,EAAU,UAAY,EAAU,UAAY,MA9H3D,SAAA,UAAA,YAAN,2GACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEF,MAAA,CAAA,EAAO,KAAK,sBAAwB,OAAO,KAAK,KAAK,eACjB,KAAK,uBAW7B,EAAA,UAAA,eAAd,8HAC8B,MAAA,CAAA,EAAM,KAAK,8BACvC,GADM,EAAsB,EAAA,OACxB,CAAC,KAAK,iBAAmB,CAAC,EAE5B,KAAM,IAAI,OACN,6DA0BN,GAzBO,AAAI,KAAK,iBAAmB,GAEjC,GAAA,KAAK,OACD,EAAoB,SAAW,KAAK,gBAAgB,OACpD,WAAM,MAAA,uCACF,EAAK,gBAAgB,OAAO,WAC5B,kEACW,EAAoB,OAAO,WAAa,OAE7D,AAAK,KAAK,iBACR,MAAK,gBAAkB,GAGnB,EAAkC,KAAK,gBAAgB,OACzD,SAAC,EAAmC,GAClC,SAAS,GAAS,EAAS,GAAQ,GAAM,EAClC,GAET,IACE,EACF,OAAO,KAAK,GAAQ,OAAO,SAAC,GAAS,MAAC,GAAO,GAAQ,IACzD,GAAA,KAAK,OACD,EAAe,SAAW,EAC1B,WAAM,MAAA,iCAAmC,EAAe,aAExD,KAAK,eACP,IAAA,EAAA,EAAkB,EAAA,OAAO,KAAK,KAAK,eAAjB,EAAA,EAAA,OAAA,IAEhB,GAFS,EAAG,EAAA,GACN,EAAQ,KAAK,gBAAgB,QAAQ,GACvC,IAAU,GACZ,KAAM,IAAI,OACN,YAAc,EACd,uEACY,KAAK,gBAAgB,WAAa,MAIxD,YAAK,qBAAuB,aAGhB,EAAA,UAAA,oBAAd,uHACM,MAAK,UACM,CAAA,EAAM,KAAK,KAAK,YAD3B,CAAA,EAAA,UACI,SAAO,EAAA,OACQ,CAAA,EAAM,EAAK,eAChC,GADM,EAAe,EAAA,OACjB,EAAa,KACf,KAAM,IAAI,OAAM,sCAEZ,SAAoB,EAAa,MACjC,EAAU,KAAK,SAAS,EAAW,IACzC,CAAA,EAAO,UAEP,MAAA,CAAA,EAAO,YA0DL,EAAA,UAAA,SAAN,wHACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEU,MAAA,CAAA,EAAM,KAAK,KAAK,mBAAxB,SAAQ,EAAA,OACZ,AAAI,KAAK,WAGP,GAAQ,EAAM,KAAK,IAErB,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,GAAK,gBAAgB,YAG7C,EAAA,UAAA,gBAAA,SAAgB,GAKd,OAJM,GAAS,KAAK,SAAS,GACvB,EAA6C,GAC7C,EAA2C,GAExC,EAAI,EAAG,EAAI,KAAK,gBAAgB,OAAQ,KAC/C,GAAM,GAAM,KAAK,gBAAgB,GAC3B,EAAS,KAAK,cAAgB,KAAK,cAAc,GAAO,KAC9D,GAAI,KAAK,uBAAyB,CAAC,EAEjC,SAEA,GAAM,GAAQ,EAAO,GACjB,EAAc,KAClB,GAAI,IAAU,GAGZ,GAAI,GAAU,EAAO,UAAY,OAC/B,EAAc,EAAO,aAChB,GAAI,GAAW,GAAO,UAAY,EAAO,SAC9C,KAAM,IAAI,OACN,mBAAmB,EAAG,2BAA2B,GAErD,EAAc,YAIhB,GAAM,GAAa,OAAO,GAC1B,GAAI,MAAM,GAGR,AAAI,GAAU,EAAO,QAAU,OAC7B,EAAc,KAAK,WAAW,GAG9B,EAAc,UAEP,CAAC,GAAU,CAAC,EAAO,MAG5B,EAAc,MAId,QAAQ,EAAO,WACR,UACH,EAAc,EACd,UACG,QACH,EAAc,KAAK,MAAM,GACzB,UACG,OACH,EAAc,KAAK,WAAW,GAC9B,cAEA,EAAc,GAKtB,AAAC,GAAU,EAAO,QAAW,EAAO,GAAO,EACd,EAAS,GAAO,EAKjD,MAAI,QAAO,KAAK,GAAQ,SAAW,EAC1B,EAGA,CAAC,GAAI,EAAU,GAAI,IAItB,EAAA,UAAA,WAAR,SAAmB,GACjB,MAAI,KAAU,KAAO,EAAM,gBAAkB,OACpC,EAEA,GAKH,EAAA,UAAA,SAAR,SAAiB,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IAM7B,OALM,GAAmB,GACrB,EAAa,EACX,EAAa,EAAK,OACpB,EAAe,GAEV,EAAI,EAAG,EAAI,EAAY,IAC9B,OAAQ,OAED,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAa,EAAI,EACjB,EAAe,GACf,UAEG,MAAK,UAIR,GAHA,EAAa,EAAI,EAGb,KAAK,YAAc,KAAO,KAAK,gBACjC,MAEF,EAAO,KAAK,IACZ,EAAe,GACf,cAGA,EAAe,GACf,EAAa,EACb,MAEJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,IACvC,EAAe,GACf,EAAa,EAAI,EACjB,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,EAAI,IAC3C,EAAe,GACf,EAAa,EAAI,EACjB,UAEG,IACH,EAAe,GACf,cAGA,EAAe,GACf,MAEJ,UACG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,MAWN,GANA,AAAI,IAAiB,GACnB,EAAO,KAAK,EAAK,UAAU,EAAY,EAAa,IAEpD,EAAO,KAAK,EAAK,UAAU,IAGzB,GAAwB,EAAO,SAAW,KAAK,gBAAgB,OACjE,KAAM,IAAI,OAAM,wCACZ,KAAK,gBAAgB,OAAM,+BAA+B,GAEhE,MAAO,IAEX,GA1VgC,IC7ChC,AA4BA,GAAA,IAAA,SAAA,GAAwC,GAAA,EAAA,GAgBtC,WAAuC,GAAvC,GAAA,GACE,EAAA,KAAA,OAAO,KAD8B,EAAA,iBAAA,EAf/B,EAAA,SAAW,GAiBjB,EAAK,QAAU,EAAiB,SAAW,KAC3C,GAAM,GAAc,KAAK,KAAK,EAAK,SACnC,GAAI,EAAK,QAAU,GAAK,EAAc,GAAK,EAAc,IACrD,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,oDACA,gCAA+B,EAAK,UAc1C,GAXA,EAAK,UAAY,EAAiB,yBAA2B,GAC7D,EAAK,aAAe,EAAiB,aACrC,EAAK,qBACD,EAAiB,sBAAwB,EAAK,QAClD,EAAK,sBAAwB,EAAiB,sBAC9C,EAAK,sBAAwB,EAAiB,uBAAyB,EAEvE,EAAK,mBACD,IAAiB,qBAAuB,IAC5C,EAAK,gBACD,EAAiB,kBAAoB,GACrC,CAAC,EAAK,oBAAsB,CAAC,EAAK,gBACpC,KAAM,IAAI,OACN,iHAKR,SAAA,UAAA,QAAA,WACE,MAAO,cAII,EAAA,OAAb,SAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,6FAClB,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGA,SAAqB,GAAI,GAAmB,GAGlD,CAAA,EAAM,EAAmB,gBAAzB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,8IAEI,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,KAAK,uBAAyB,KAAO,GACA,KAAK,sBACjD,MAAO,aAHT,SAAK,OAAS,EAAA,qCAMR,GAAI,OACN,iDAAiD,EAAE,gBAGzD,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,2CAQlB,GALM,EAED,OAAe,cAAiB,OAAe,mBACpD,KAAK,aAAe,GAAI,GAEpB,CAAC,KAAK,aAGR,KAAK,aAAe,KAAK,aAAa,mBAC7B,KAAK,aAAa,aAAe,KAAK,aAC/C,KAAM,IAAI,OACN,8BACA,cAAa,KAAK,aAAY,MAC9B,YAAW,KAAK,aAAa,aAG7B,SAAe,KAAK,aAAa,wBAAwB,KAAK,QACpE,KAAK,SAAW,KAAK,aAAa,iBAClC,KAAK,SAAS,QAAU,KAAK,QAAU,EACvC,KAAK,SAAS,sBAAwB,KAAK,sBAC3C,EAAa,QAAQ,KAAK,UAC1B,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,CAAA,SAGI,EAAA,UAAA,KAAN,mHACE,MAAI,MAAK,SACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAMN,CAAA,EAAM,KAAK,uBAA5B,SAAiB,EAAA,OACvB,AAAI,KAAK,oBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAoB,KAAK,4BACrB,EAAU,CAAC,KAAK,UAAW,KAAK,qBAAsB,KAE5D,AAAI,KAAK,iBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAiB,KAAK,4BAClB,EAAU,CAAC,KAAK,UAAY,KAAK,QAAS,KAGhD,CAAA,EAAO,CACL,MAAO,CAAC,YAAe,EAAmB,SAAY,GACtD,KAAM,WAMJ,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAIf,EAAA,UAAA,aAAd,+FAEQ,SAAgC,GAChC,EAAgC,GAClC,EAAgB,EACpB,CAAA,EAAO,GAAI,SAAQ,SAAA,GACjB,GAAM,GAAa,YAAY,WAC7B,AAAI,EAAK,oBACP,GAAK,SAAS,sBAAsB,EAAK,UAEzC,AAAI,EAAK,SAAS,KAAO,WACvB,EAAQ,CAAC,cAAa,EAAE,cAAa,IAEvC,EAAc,KAAK,EAAK,SAAS,MAAM,EAAG,EAAK,wBAEjD,AAAI,EAAK,iBACP,GAAK,SAAS,uBAAuB,EAAK,UAC1C,EAAc,KAAK,EAAK,SAAS,UAInC,AAAI,EAAE,IAAkB,EAAK,WAC3B,eAAc,GACd,EAAQ,CAAC,cAAa,EAAE,cAAa,MAEtC,EAAK,QAAU,EAAK,aAAe,aAK1C,EAAA,UAAA,KAAA,WACE,AAAK,KAAK,UACR,MAAK,SAAW,GAChB,KAAK,SAAS,aACd,KAAK,aAAa,QAClB,AAAI,KAAK,QAAU,MAAQ,KAAK,OAAO,YAAY,OAAS,GAC1D,KAAK,OAAO,YAAY,GAAG,SAMjC,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAIlB,EAAA,UAAA,cAAA,WACE,MAAO,MAAK,cAGN,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAY,EAAM,GAAG,OACrB,EAAW,GAAI,cAAa,EAAM,OAAS,GACjD,SAAM,QAAQ,SAAC,EAAM,GAAM,MAAA,GAAS,IAAI,EAAM,EAAI,KAC3C,GAGD,EAAA,UAAA,4BAAR,SAAoC,EAAwB,GAE1D,GAAM,GAAO,GAAI,cAAa,GAAA,KAAK,cAAc,IAEjD,SAAK,IAAI,EAAU,EAAK,OAAS,EAAS,QACnC,GAAA,OAAO,EAAM,IAExB,GA7MwC,IC5BxC,AA0BA,GAAA,IAAA,SAAA,GAAoC,GAAA,EAAA,GAQlC,WACuB,EACA,GAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KACP,GAHqB,EAAA,mBAAA,EACA,EAAA,aAAA,EATf,EAAA,SAAW,GAEX,EAAA,OAAS,GASX,EAAK,eAKP,GAJA,EAAK,OAAS,GACd,EAAK,SACD,CAAC,EAAK,aAAa,aAAc,EAAK,aAAa,aACvD,EAAK,WAAa,GAAA,SAAS,CAAC,GAAI,SAC5B,EAAK,aAAa,YAEpB,GAAM,GACF,EAAK,aAAa,YAAc,EAAM,EAAK,mBAAmB,MAC5D,EAAsB,EAAK,aAAa,aAAe,EACzD,EAAK,mBAAmB,OACtB,EAAkB,GAAI,GAAsB,EAC5C,EAAmB,GAAI,GAAuB,EAC9C,EAAe,EAAiB,EAChC,EAAgB,EAAsB,EAC5C,EAAK,QAAU,GAAA,SACX,CAAC,EAAiB,EAAgB,EAAe,GACjD,CAAC,EAAG,QAER,GAAK,QAAU,GAAA,SAAS,CAAC,EAAG,EAAG,EAAG,GAAI,CAAC,EAAG,aAKhD,SAAA,UAAA,QAAA,WACE,MAAO,UAII,EAAA,OAAb,SACI,EAAuC,GAAA,MAAA,KAAA,QAAA,GAAA,6FACzC,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGN,GAAI,CAAC,GAIH,GADA,EAAqB,SAAS,cAAc,SACxC,CAAC,EAAa,aAAe,CAAC,EAAa,aAC7C,KAAM,IAAI,OACN,0GAGN,EAAmB,MAAQ,EAAa,YACxC,EAAmB,OAAS,EAAa,aAErC,SAAiB,GAAI,GAAe,EAAoB,GAG9D,CAAA,EAAM,EAAe,gBAArB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,oHACE,AAAI,KAAK,aAAa,YACpB,GAAA,KAAK,OACA,KAAK,aAAa,aAAe,QAC7B,KAAK,aAAa,aAAe,cACtC,WACI,MAAA,+BAA+B,EAAK,aAAa,WAAU,2FAKnE,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,CACL,SAAU,KAAK,aAAa,SAC5B,WAAY,KAAK,aAAa,WAC1B,KAAK,aAAa,WAClB,OACJ,MAAO,KAAK,mBAAmB,MAC/B,OAAQ,KAAK,mBAAmB,kBAPpC,SAAK,OAAS,EAAA,qCAYd,EAAE,QAAU,iDAAiD,EAAE,QACzD,SAGR,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,uCAIlB,IACE,KAAK,mBAAmB,UAAY,KAAK,aAClC,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,OAAO,IAAI,gBAAgB,KAAK,QAGhE,YAAK,mBAAmB,OAExB,KAAK,SAAW,GAEhB,CAAA,EAAO,GAAI,SAAc,SAAA,GAEvB,EAAK,mBAAmB,iBAAmB,WACzC,cAKA,EAAA,UAAA,KAAN,oFACE,GAAI,KAAK,SACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAI7B,IACE,EAAM,GAAA,QAAQ,WAAW,KAAK,0BACvB,GACP,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAU,IAEjE,GAAI,KAAK,OACP,IACE,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,mBAAmB,GAAM,KAAM,WAC5C,GACP,KAAM,IAAI,OAAM,oCAAoC,EAAE,iBAEtD,EAAI,cAGN,OAAA,CAAA,EAAO,CAAC,MAAO,EAAK,KAAM,oBAItB,EAAA,UAAA,aAAR,WAIE,MAAI,QAAK,aAAa,aAAe,KAAK,aAAa,cAClD,MAAK,mBAAmB,QAAU,KAAK,aAAa,aACpD,KAAK,mBAAmB,SAAW,KAAK,aAAa,gBAO5D,EAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,IAAA,KAAK,WACV,GAAM,GAA0B,EAAI,UAAU,WAAW,GACrD,EACJ,EAAe,GAAA,MAAM,cACjB,EAAe,EAAK,QAAS,EAAK,WAAY,EAAK,SACnD,YAEJ,GAAM,GAAQ,EAAa,MAC3B,MAAO,GAAa,QAAQ,EAAM,MAAM,OAMtC,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAI7B,EAAA,UAAA,KAAA,WACE,GAAM,GAAS,KAAK,OAAO,YAE3B,EAAO,QAAQ,SAAA,GAAS,MAAA,GAAM,SAE9B,IACE,KAAK,mBAAmB,UAAY,WAC7B,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,KAEhC,KAAK,SAAW,IAIlB,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAEpB,GApMoC,IC1BpC,AA2BA,GAAA,IAAA,WAAA,cAUA,MAAA,MCrCA,AAoBA,GAAA,IAAA,SAAA,GAA6C,GAAA,EAAA,GAA7C,4DAmBE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAI,IAAc,KAAM,IAEnC,GAtB6C,IAmC7C,GAAA,SAAA,GAA4B,GAAA,EAAA,GAG1B,WAAsB,EAAgC,GAAtD,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAkB,EAAU,KAG9C,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf4B,IAiB5B,GAAA,SAAA,GAAgC,GAAA,EAAA,GAI9B,WACc,EAA0C,GADxD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAA0C,EAAA,UAAA,EAHxD,EAAA,UAAY,KAOZ,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAAc,KAAK,UAAS,MAGzD,EAAA,UAAA,KAAN,mHACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eACxC,GADM,EAAc,EAAA,OAChB,EAAY,KACd,MAAI,MAAK,YAAc,GACrB,CAAA,EAAO,IAKT,MAAK,YAAY,KAAK,KAAK,WAC3B,KAAK,UAAY,GACjB,CAAA,EAAO,KAQT,IANM,EAAQ,EAAY,MAAM,MAAM,KAAK,WAK3C,EAAM,GAAK,KAAK,UAAY,EAAM,GAClC,EAAA,EAAmB,EAAA,EAAM,MAAM,EAAG,IAAf,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,KAAK,YAAY,KAAK,GAExB,YAAK,UAAY,EAAM,EAAM,OAAS,GAEtC,CAAA,EAAO,UAEX,GAvCgC,ICxEhC,AAsBA,GAAA,IAAA,SAAA,GAAgD,GAAA,EAAA,GAAhD,4DAUE,SAAA,UAAA,WAAA,WACE,MAAO,IAAI,IAAa,OAE5B,GAbgD,IA0BhD,GAAA,SAAA,GAA2B,GAAA,EAAA,GAGzB,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAiB,KAGnC,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf2B,IAuC3B,GAAA,SAAA,GAA+B,GAAA,EAAA,GAM7B,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAF6B,EAAA,SAAA,EAEzB,GAAA,MAAM,IAAI,cACZ,EAAK,QAAU,GAAI,aAAY,cAGxB,GAAA,GAAA,KAAA,cACP,EAAK,QAAU,GAAI,GAAc,iBAGrC,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,+GACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eAExC,MAFM,GAAc,EAAA,OAEhB,EAAY,KACd,CAAA,EAAO,IAEP,GAAQ,EAAY,MAItB,AAAI,GAAA,MAAM,IAAI,cACZ,EAAO,KAAK,QAAQ,OAAO,EAAO,CAAC,OAAQ,KAE3C,EAAO,KAAK,QAAQ,MAAM,OAAO,KAAK,EAAM,SAE9C,KAAK,YAAY,KAAK,GACtB,CAAA,EAAO,WAEX,GAtC+B,ICvF/B,AAqCA,GAAA,IAAA,SAAA,GAAuC,GAAA,EAAA,GAIrC,WACc,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,KAAA,EACA,EAAA,QAAA,EAEZ,GAAA,KAAK,OACA,YAAgB,aACZ,IAAA,MAAM,IAAI,cACL,YAAgB,OAAQ,YAAgB,MACzC,IACT,WAAM,MAAA,yEAEV,EAAK,OAAS,EAAQ,QAAU,EAEhC,EAAK,UAAY,EAAQ,WAAa,KAAO,OAG/C,SAAA,UAAA,QAAA,WACE,MAAO,cAAc,KAAK,MAGtB,EAAA,UAAA,KAAN,oHACE,MAAI,MAAK,QAAY,MAAK,eAAgB,YAClB,KAAK,KAAK,WACV,KAAK,KAAK,MAChC,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAQ,GAAI,SAAoB,SAAC,EAAS,GAC9C,GAAM,GAAM,EAAK,OAAS,EAAK,UAC/B,GAAI,EAAK,eAAgB,YAGvB,EAAQ,GAAI,YAAW,EAAK,KAAK,MAAM,EAAK,OAAQ,UAMpD,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GACnB,GAAI,GAAsC,EAAW,OAOrD,GAHA,AAAI,YAAgB,cAClB,GAAO,GAAI,YAAW,IAEpB,CAAE,aAAgB,aACpB,MAAO,GAAO,GAAI,WAAU,sCAE9B,EAAQ,IAEV,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,aAE1B,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,EAAM,QAIhC,GAAM,GAAQ,EAAK,KAAK,MAAM,EAAK,OAAQ,GAG3C,EAAW,kBAAkB,GAE/B,EAAK,OAAS,SAEA,CAAA,EAAM,WAAtB,MAAA,CAAA,EAAQ,GAAA,MAAQ,EAAA,OAAc,EAAA,KAAM,GAAK,UAE7C,GAxEuC,ICrCvC,YA6BI,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,qGAGpB,MAAK,OAAO,IAAS,SACnB,EAAY,EAEZ,GAAa,EAAgB,IAC7B,EAAc,GAA0B,IAEzB,CAAA,EAAM,GAAA,KAAK,MAAM,EAAW,iBAAvC,GAAW,EAAA,OACb,EAAS,MACY,WAAU,KAAC,CAAA,EAAM,EAAS,gBAD/C,CAAA,EAAA,UACI,SAAa,GAAA,GAAA,MAAI,WAAU,CAAA,OAAC,EAAA,UAClC,CAAA,EAAO,GAAI,IAAkB,EAAY,WAEzC,KAAM,IAAI,OAAM,EAAS,iBAK7B,GAAM,IAA4B,SAAC,GACjC,GAAM,GAAO,CACX,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,KACd,KAAM,EAAQ,KACd,YAAa,EAAQ,YACrB,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,SAAU,EAAQ,SAClB,UAAW,EAAQ,WAErB,MAAO,IC5DT,YAqB4B,GAC1B,MAAQ,OAAO,IAAW,UAAa,EAAO,OAAO,EAAG,KAAO,UCtBjE,mBA6BoC,GAAA,EAAA,GASlC,WACc,EACS,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,MAAA,EACS,EAAA,QAAA,IAIjB,SAAA,UAAA,SAAN,oFACE,MAAI,IAAY,KAAK,QAAU,GAAA,MAAM,IAAI,YAEjC,GAAK,cACX,KAAK,MAAQ,EAAG,aAAc,KAAK,MAAiB,OAAO,KAI7D,CAAA,EAAO,GAAI,IAAkB,KAAK,MAAsB,KAAK,eAEjE,GAzBoC,IC7BpC,mBA4BmC,GAAA,EAAA,GAQjC,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,IAAA,EACA,EAAA,YAAA,IAQjB,SAAA,UAAA,SAAN,8EACE,MAAI,IAAY,KAAK,KACnB,CAAA,EAAQ,GAAI,IAAe,KAAK,IAAe,KAAK,aAC/C,YAEL,CAAA,EAAO,GAAiB,KAAK,IAAK,KAAK,mBAG7C,GA1BmC,IC5BnC,YA0GI,EAAqB,GAAA,MAAA,KAAA,QAAA,GAAA,IAChB,GAAI,IAAW,GAAI,IAAc,GAAS,eA4B/C,GADJ,GAAA,GAAA,KAEQ,EAAO,GAAqB,GAClC,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,qBAgEvC,GADJ,GAAA,GAAA,KAEE,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACf,MAAA,CAAA,EAAM,YAAZ,SAAM,EAAA,OACZ,CAAA,EAAO,GAAqB,WAAM,MAAA,GAAI,6BAmCtC,EACA,sEACF,MAAA,CAAA,EAAO,GAAe,OAAO,EAAoB,oBAqClB,sEAE/B,MAAA,CAAA,EAAO,GAAmB,OAAO,QCxRnC,GAGM,IAAU,oNCHhB,oBA2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9GnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC7EnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UClFnC,oBAKA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAER,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAG,EAC9B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9FnC,oBAyBA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UChJnC,oBAIA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UCnGnC,oBAwBA,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMjB,MAAO,KAAW,UAAY,GAAO,SACxC,GAAO,QAAU,EAEjB,IACE,EAAa,WACN,SACJ,AAAK,OAAO,SAAW,YAAc,OAAO,KACjD,OAAO,WAAa,MAAO,OAK3B,GACA,QCrPF,oBAYA,GAAI,IAAO,KAKP,GAAS,KAKT,GAAS,KAQT,GAAY,KASZ,GAAU,KAOV,GAAS,KAIT,GAAK,KAET,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAO,QAAU,6vDC3DjB,YAoBI,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MACF,GAAM,8DC7BpB,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBACvC,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAM/B,YACI,EAAyB,EAAW,EACpC,GACF,GAAI,IAAe,SACjB,MAAO,GAAQ,OAAO,GACjB,GAAI,IAAe,OACxB,MAAO,GAAQ,KAAK,GACf,GAAI,IAAe,MACxB,MAAO,GAAA,IAAO,GACT,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,GAChB,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,EAAG,GAE1B,KAAM,IAAI,OACN,cAAc,EAAU,qEAeM,GAAA,EAAA,GAMlC,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KANF,SAAA,UAAY,GAGX,EAAA,SAAW,GAIjB,EAAK,KAAO,GAAI,GAAA,YAAY,EAAM,EAAA,YAGpC,SAAA,UAAA,MAAA,SAAM,EAAoC,EAAiB,GAEzD,AAAI,KAAK,UACP,MAAK,SAAW,GAChB,AAAI,EAAA,MAAM,IAAI,YACZ,EAAA,aAAa,KACT;;;gCAYR,GAAM,GAAS,GAEf,YAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,IAEzC,GAST,EAAA,UAAA,eAAA,SACI,EAAiB,EACjB,GACF,GAAM,GAAQ,KAAK,MAAM,EAAQ,EAAO,GAExC,MAAO,CAAC,OAAQ,EAAO,MAAK,EAAE,MAAK,IAIrC,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,YAIb,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,aAIf,EAAA,UAAA,KAAA,SACI,EAAgB,EAAoC,EACpD,GACF,KAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,KAGlD,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,KAAK,cAGb,EAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,SAAS,SAEvB,EAAA,UAAA,SAAA,SAAS,GACD,GAAA,GAAA,KAAA,KAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,mBAEd,GAAI,IAAU,aACZ,GAAM,GACF,KAAK,SAAS,EAAmB,KAAK,QACpC,EACF,KAAK,SAAS,EAAmB,KAAK,QAC1C,MAAO,GAAA,aAAa,uBAAuB,EAAY,GAGzD,MAAO,MAAK,KAAK,IAAI,GAAQ,QAGvB,EAAA,UAAA,WAAR,SAAmC,GACjC,GAAM,GAAO,KAAK,SAAS,EAAE,QACzB,EAAc,EAClB,GAAI,EAAE,QAAU,SACd,IAEE,EAAe,EAAsB,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,WAChE,GACA,KAAM,IAAI,OAAM,oDAGpB,MAAO,GAAA,OAAU,EAAE,MAAO,EAAE,MAAO,IAGrC,EAAA,UAAA,WAAA,SACI,EAAoC,EAAiB,GACvD,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAG7D,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,KAAK,IAAI,IACT,GAAA,GAAA,KAAA,KAAA,IAAA,GAAA,mBAEP,AAAI,GAAsB,MACxB,MAAK,YAAY,EAAmB,KAAK,QACzC,KAAK,YAAY,EAAmB,KAAK,SAG3C,KAAK,KAAK,OAAO,KAIrB,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GAEjC,EAAW,WAEX,AAAI,EAAW,SAAW,GACxB,KAAK,YAAY,KAKjB,EAAA,UAAA,KAAN,SAAW,8EACH,SAAQ,EAAA,KAAK,MACnB,IACM,EAAW,EAAA,KAAK,MAAQ,EAC9B,CAAA,EAAO,CAAC,SAAQ,SAGlB,EAAA,UAAA,OAAA,WACE,MAAO,CAEL,WAAY,GACZ,QACI,CAAC,wHAKT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,GAAiB,EAAG,gBAEpB,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAU,GAAI,GAKvB,OAFM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GACpB,EAAI,EAAG,EAAI,EAAO,KAAM,KAI/B,OAHM,GAAM,EAAO,WAAW,GAExB,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,GAAK,EAAQ,GAAK,EAAM,GAE1C,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAM,OAAM,IAGrC,MAAO,GAAO,YAGhB,EAAA,UAAA,KAAA,SAAK,GAIH,OAHM,GAAQ,KAAK,SAAS,EAAE,QACxB,EAAS,EAAA,OAAU,CAAC,EAAE,KAAM,EAAE,MAAO,EAAE,OACvC,EAAO,EAAO,OACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAK,EAAI,EAAE,KAAO,GAAK,EAAM,GAE/B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,EAAA,MAAS,EAAG,EAAO,GAAM,QAAQ,GAE5C,MAAO,IAGT,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAiB,EAAG,WAKpB,OAHM,GAAS,EAAA,OAAU,EAAE,MAAO,EAAE,OAC9B,EAAO,KAAK,WAAW,cAEpB,GACP,GAAM,GAAS,EAAO,WAAW,GAC3B,EAAQ,EAAO,QACrB,EAAK,QAAQ,SAAA,GAAM,MAAA,GAAM,GAAM,EAAE,MAAM,GAAM,EAAI,EAAM,KACvD,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAK,OAAM,KAJ3B,EAAI,EAAG,EAAI,EAAO,KAAM,MAAxB,GAOT,MAAO,GAAO,YAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,UAAiB,EAAG,OAGb,EAAA,IAAO,EAAA,OAAU,IAAK,IAG/B,EAAA,UAAA,KAAA,SAAuB,GAAvB,GAAA,GAAA,KACE,GAAiB,EAAS,QAK1B,OAHM,GAAO,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAK,SAAS,EAAE,UACxC,EAAS,EAAA,OAAU,EAAQ,GAAG,MAAO,EAAQ,GAAG,OAChD,EAAa,EAAO,OACjB,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAElC,OADM,GAAW,EAAK,GACb,EAAI,EAAG,EAAI,EAAW,OAAQ,IACrC,EAAW,IAAM,EAAS,GAG9B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAGhD,EAAI,EAAA,IAAO,EAAQ,EAAS,QAAQ,IACpC,EAAI,EAAA,IAAO,GACX,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAO,EAAG,IAGnB,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,IAAI,EAAQ,MAIjE,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAiB,CAAC,EAAG,GAAI,UAqBzB,OAnBM,GAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAU,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5C,EAAW,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7C,EAAW,EAAE,MAAM,GAEnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,8DAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAAY,EAAA,EAAA,GAGrB,EAAA,8DAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAIzB,EAAO,EAAU,EACjB,EAAS,EAAA,OAAU,CAAC,EAAU,EAAS,GAAW,EAAE,OACpD,EAAU,EAAO,OACjB,EAAY,KAAK,UAEd,EAAI,EAAG,EAAI,EAAU,IAC5B,OAAS,GAAK,EAAG,EAAK,EAAS,GAAM,EACnC,OAAS,GAAK,EAAG,EAAK,EAAU,GAAM,EACpC,OAAS,GAAK,EAAG,EAAK,EAAW,GAAM,EAMrC,OAJM,GAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAE/B,EAAI,EAAI,EAAI,EAAQ,IAC3B,OAAS,GAAI,EAAI,EAAI,EAAQ,KAG3B,OAFI,GAAM,EAED,EAAI,EAAI,EAAI,EAAQ,IAC3B,GAAO,EAAQ,EAAI,EAAS,EAAI,EAAa,EAAI,GAC7C,EAAQ,EAAI,EAAa,EAAI,EAAa,EAAI,GAEpD,EAAQ,EAAI,EAAQ,GAAI,EAAW,KAAO,EAOtD,MAAO,GAAO,YAGhB,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE/C,EAAS,KAAK,YAAY,EAAG,EAAG,EAAY,GAChD,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAIvC,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAiB,CAAC,EAAG,GAAI,YAEzB,GAAM,GAAK,SAAC,EAAW,GAAc,MAAA,MAAK,MAAM,EAAI,IAC9C,EAAc,QACpB,MAAO,MAAK,oBAAoB,EAAG,EAAG,EAAa,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MASvD,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EACD,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAO,EAAM,EAAS,GAExB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,GAAiB,EAAG,OAUpB,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAO,EACF,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAQ,EAAM,EAAS,GAEzB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,EAAG,sBAOpB,OALM,GAAM,GAIN,EAAW,EAAE,KAAO,EAAW,KAC5B,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,EAAW,WAAW,EAAI,GAGzC,OAAS,GAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAY,EAAA,OAAU,EAAG,SACzB,EAAO,EAAA,MAAS,EAAW,GAAY,OAAO,WAC9C,EAAM,EAAK,IAAI,GAAG,IAAI,GAC5B,EAAI,KAAK,GAGX,MAAO,GAAA,MAAS,IAGlB,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAIlD,GAFA,GAAiB,EAAG,UAEhB,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,oDAAoD,GAAE,KAAO,GAAC,IAC9D,iBAAgB,IAWtB,OATM,GAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAE,MAAO,GAC3B,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAW,EAAE,MAAM,EAAE,KAAO,GAC5B,EAAgB,EAClB,SAAC,EAAW,GAAc,MAAA,GAAI,EAAW,EAAI,GAC7C,SAAC,EAAW,GAAc,MAAA,GAAI,GACzB,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACrC,OAAS,GAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAM,EAAc,EAAG,GAC7B,GAAI,IAAM,EACR,EAAK,GAAO,EAAY,EAAI,EAAM,QAElC,GAAM,GAAU,EAAc,EAAG,EAAI,GACrC,EAAK,GAAO,EAAY,EAAM,GAAW,EAAK,GACtB,EAAM,GAAO,EAAK,IAIhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,UAAiB,CAAC,EAAG,GAAI,YAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,UAAiB,CAAC,EAAG,GAAI,QAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,UAAiB,CAAC,EAAG,GAAI,gBAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,UAAiB,CAAC,EAAG,GAAI,cAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAiB,CAAC,EAAW,EAAG,GAAI,UAYpC,OAVM,GAAS,KAAK,SAAS,EAAU,QACjC,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,MAAS,EAAE,MAAO,EAAA,WAAW,EAAE,MAAO,EAAE,QACjD,EAAY,KAAK,SAAS,EAAO,QACnC,EAAQ,EACN,EAAS,EAAU,OAAS,GAAK,EAAU,KAAO,GAAK,EAAE,OAAS,EACpE,EACA,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAE5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,OAAS,GAAI,EAAG,EAAI,EAAQ,IAC1B,AAAI,EAAO,KAAO,EAChB,EAAU,KAAW,EAAQ,GAE7B,EAAU,KAAW,EAAQ,GAKnC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GACJ,GAAiB,CAAC,GAAY,SAE9B,GAAM,GAAW,KAAK,SAAS,EAAU,QACzC,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAiB,EAAG,QAEpB,GAAM,GAAQ,KAAK,SAAS,EAAE,QAC9B,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAM,EAAO,EACnB,MAAK,GAAO,GAAK,EAAO,GAAO,GAAQ,GAAK,GAAQ,EAC3C,EAEC,GAAM,GAAQ,KAK5B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAM,GAAO,EAEf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAS,EAAM,GACV,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAS,GAAU,EAErB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,UAAiB,CAAC,EAAG,GAAI,qBAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAO,EAAO,EACpB,MAAO,GAAO,KAIlB,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,IAGT,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,EAAG,EAAO,IAElC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,KAAK,IAAI,EAAG,EAAO,IAAK,GAEhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MACR,SAAC,EAAQ,GAAW,MAAA,GAAS,EAAI,EAAS,EAAS,KAGhE,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAiB,CAAC,EAAI,GAAI,UAK1B,OAHM,GAAe,GAAI,cAAa,EAAE,MAClC,EAAS,KAAK,SAAS,EAAE,QACzB,EAAW,KAAK,SAAS,EAAG,QACzB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,GACjB,AAAI,GAAK,EACP,EAAa,GAAK,EAAS,GAE3B,EAAa,GAAK,EAAS,GAAM,GAAI,GAGzC,MAAO,MAAK,WAAW,EAAc,EAAE,MAAO,YAGhD,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,MAAM,EAAQ,MAInE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,OAAO,EAAO,EAAQ,GAExC,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAiB,CAAC,EAAG,GAAS,UAyB9B,OAvBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAiB,EAAS,aAAe,eAEzC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OAEnC,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAChD,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAEhD,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EACf,EAAW,EAAI,EACZ,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EACxB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,IAAK,EAAW,EAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAC7B,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,GAAK,GAC1B,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,GAAK,IAClB,GAAO,EAAM,GAAW,IAE9B,IAAY,EAAS,cAOjC,MAAO,GAAE,YAGX,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAiBpC,OAfM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAI,EAAA,OAAmB,EAAS,SAAU,EAAE,OAE5C,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAKF,OAHM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAElC,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAS,YACpC,EAAW,EAAK,EAAS,YAAc,EACpC,GAAK,EAAG,GAAK,EAAa,MACjC,GAAM,IAAK,EAAW,GAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,GAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAAS,WACtC,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,IACrB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,KAAO,GAAO,EAAM,GAAW,IAElD,IAAY,EAAS,eASrC,MAAO,GAAE,YAGX,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,kBAkC/B,OAhCM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAiB,IAAe,eAChC,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GACjD,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GAE9C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,IAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAOjC,OANM,IAAK,GAAK,EAAc,GACxB,GACF,EAAe,EAAI,EAAa,GAAK,EAAa,GAChD,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,EAAiB,IAC7C,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,GAAM,IAAW,EAAe,EAAI,EAAa,GAC7C,EAAa,GAAK,EAAiB,EACvC,EAAS,IAAY,GAK7B,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GA6BF,OA5BM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAE1B,EAAA,EAAA,UACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,GAAA,EAAA,YAEI,GAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,GAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,GAAU,EAAc,EAAI,EAAS,QAAQ,KAE1C,GAAI,EAAG,GAAI,EAAW,EAAE,GAC/B,OAAS,IAAK,EAAG,GAAK,EAAY,EAAE,GAElC,OAAS,IAAK,EAAG,GAAK,EAAS,EAAE,GAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAGzC,GAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,KACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,IAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAc,GAErB,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAQjC,OAPM,IAAK,GAAK,GAAc,GACxB,GACF,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GACxC,GAAY,EAAS,GAAc,EAAI,IACzC,EAAS,GAAe,EAAI,IAC5B,EAAS,GAAc,EAAI,IAAM,EAAQ,GAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAK3B,EAAS,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GAAK,IACpD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAiB,CAAC,EAAG,GAAK,mBAa1B,OAXM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eACzC,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,EAC3C,OAAS,GAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAG5C,OADI,GAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,AAAI,EACF,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAEnD,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAK3D,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAKpC,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAqBzC,OAnBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAC9C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAEhB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAEvB,EAAK,EAAG,EAAK,EAAa,EAAE,EAMnC,OALM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAW,GAAM,IAChD,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAW,GAAM,GACtD,EAAW,EAAK,EAEb,EAAK,EAAG,EAAK,EAAc,EAAE,EAOpC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,GAAQ,KAAK,IACf,EAAS,UACR,GAAS,SAAW,EAAS,GAAM,GAClC,GAAW,EAAK,EAAO,EAEpB,GAAK,EAAG,GAAK,EAAa,EAAE,GAOnC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,IAAM,IAC/C,GAAQ,KAAK,IACf,EAAS,SACR,GAAS,QAAU,EAAU,IAAM,GAClC,GAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,GAG3C,OAFM,IAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,IAE5C,OADI,IAAU,EACL,GAAI,EAAG,GAAI,EAAS,UAAW,EAAE,GAIxC,OAHM,IAAW,GAAI,EACf,GAAW,GAAI,EAEZ,GAAK,EAAO,GAAK,EAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAe,EAC9B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,IACjC,GAAM,IAAK,GAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAE7B,IACI,EAAQ,GAAW,IAAM,EAAS,GAAW,IAKzD,EAAS,GAAW,IAAM,GAMpC,MAAO,GAAG,YAGZ,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,gBAAgB,EAAO,EAAQ,GAEjD,MAAI,IAGF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAiB,CAAC,EAAG,GAAS,mBAc9B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WACxC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OACnC,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,EAAE,GACpC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,EAAE,GACnC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAMF,OAJM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAS,WACtC,EAAW,EACX,EAAW,EACN,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,GAE3C,OADM,GAAO,EAAM,EAAW,GACrB,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAM,EAAW,IAAM,EAAO,EAAM,EAAW,GAEjD,GAAY,EACZ,GAAY,IAQxB,MAAO,GAAE,YAGX,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,2BA0B/B,OAxBM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAQ,EAAc,EAEnB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,GAAK,EAAG,EAAK,EAAU,EAAE,EAMhC,OALM,IAAW,EAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAMjC,OALM,IAAK,GAAK,EAAc,GACxB,GAAW,EAAO,EAAI,EAAO,GAAK,EAAO,GACzC,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAO,EAAE,IAC7B,GAAM,IAAK,EAAK,EAAQ,GAClB,GAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,EAAS,EAAO,EAAI,EAAO,EAAK,EAAO,GAAK,GAAM,GAK1D,MAAO,GAAG,YAGZ,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAiB,CAAC,EAAG,GAAK,4BAc1B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WAExC,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAK5C,OAJM,GAAK,KAAK,MAAM,EAAK,GACrB,EAAK,EAAK,EAEZ,EAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,GAAW,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAIhE,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAIlC,MAAO,GAAG,YAGZ,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,UAAiB,EAAG,QACb,GAAK,KAAK,WAAW,GAAI,IAGlC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,GAAiB,CAAC,EAAG,GAAU,UAE/B,GAAM,GAAqB,EAAE,MAAM,QAC7B,EAAgB,KAAK,SAAS,EAAQ,QAC5C,EAAS,GAAQ,EAAc,OAI/B,OAHM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GAEpB,EAAI,EAAG,EAAI,EAAO,KAAM,EAAE,GACjC,GAAM,GAAS,EAAO,WAAW,GAE3B,EAAwB,EAAO,QACrC,EAAY,GAAQ,EAAc,EAAO,IAEzC,GAAM,GAAgB,EAAK,WAAW,GACtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,YAGhB,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAa,EAAE,QAAQ,GAAW,GAC7B,QAAQ,GACR,MAAM,EAAkB,IAG9B,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAiB,EAAG,UA8BpB,OA5BM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,OAAU,EAAS,SAAU,EAAE,OACxC,EAAa,EAAO,OAEpB,EAAqB,EAAS,SAAS,GAAK,EAAS,SAAS,GAChE,EAAS,SAAS,GAAK,EAAS,SAAS,GACvC,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAGhD,OAFM,GAAoB,EAAQ,EAC5B,EAAmB,EAAQ,EAAE,QAAQ,GAClC,EAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAMf,OAJM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAChD,EACF,EAAoB,EAAS,EACxB,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAKb,OAHM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,GAAkB,EAAoB,EAAO,EAC1C,GAAO,EAAG,GAAO,EAAS,SAAU,EAAE,IAG7C,OAFM,IAAa,GAAO,EAAc,EACpC,GAAU,GACP,GAAU,GACf,IAAW,EASb,OAPM,IACF,KAAK,IAAI,EAAS,QAAS,EAAuB,IAEhD,GAAkB,GAAkB,GAAO,EAC7C,GAAc,EACd,GAAW,EACX,GAAQ,EACH,GAAS,EAAW,GAAS,EACjC,IAAU,GAEb,OADM,IAAe,EAAmB,GAAS,EAAE,QAAQ,GAClD,GAAO,EAAS,GAAO,EAC3B,IAAQ,GAEX,OADM,IAAa,GAAe,GAAO,EAAE,QAAQ,GAC1C,GAAO,GAAS,GAAO,GAC3B,IAAQ,GACX,GAAM,IAAa,GAAa,GAAO,EAAE,QAAQ,GAC3C,GAAQ,EAAQ,GAAa,GAOnC,GANA,AAAK,IAAa,OAAS,GAAQ,GACjC,GAAc,GACT,AAAI,IAAa,OACtB,KAAY,GACZ,MAEE,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAM,IAAe,GAAkB,EACvC,EAAW,IACP,IAAa,MAAQ,GAAW,GAAQ,KAMtD,MAAO,GAAO,YAGhB,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGzC,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,qBAuB1B,OArBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAgB,EAAK,GAAc,EAAe,GAElD,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,KAIjB,EAAG,IACC,EAAU,EAAe,EAAO,EAAS,EAAO,EAChD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGjC,EAAA,UAAA,mBAAR,SAA2B,EAAa,GAiBtC,OAfM,GAAe,EAAA,OAAU,EAAS,SAAU,SAC5C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,KAAK,WAAW,GACpB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAIf,OAFM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAC7C,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAIb,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAO,EAAG,EAAO,EAAS,SAAU,EAAE,GAG7C,OAFM,GAAa,EAAO,EAAc,EACpC,EAAU,EACP,EAAU,GACf,GAAW,EASb,OAPM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAGlD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAS,EAAW,EAAS,EACjC,GAAU,EAEb,OADM,GAAS,EAAS,EACf,EAAO,EAAS,EAAO,EAC3B,GAAQ,EAEX,OADM,GAAO,EAAO,EACX,GAAO,EAAS,GAAO,EAC3B,IAAQ,GACX,GAAM,IAAO,GAAO,EACd,GAAQ,EAAK,IAAI,EAAO,EAAQ,EAAM,GAAM,GAClD,AAAI,IAAS,GACX,GAAW,GACX,EAAc,EAAS,EACf,EACJ,EAAO,EAAwB,IAM3C,EAAa,IAAI,EAAa,EAAO,EAAQ,EAAM,EAAM,KAMnE,MAAO,GAAa,YAGtB,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAiB,CAAC,EAAG,GAAI,qBAoBzB,OAlBM,GAAe,KAAK,mBAAmB,EAAG,GAC1C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAY,KAAK,WAAW,GAC5B,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,GAAS,EACP,EAAwB,EAC5B,EACA,EAAU,IAAI,EAAO,EAAS,EAAO,EAAO,GAC1C,GACF,EAAS,EAAwB,EACjC,EAAO,EAAuB,EAE5B,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,GAAQ,KAIzB,EAAG,IAAI,EAAS,EAAO,EAAS,EAAO,EAAO,GAMxD,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,kBAqBpB,OAnBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cACf,EAAA,KAAK,cAAc,CAAC,EAAO,EAAW,EAAU,KAE9C,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAE9C,EAAY,EACV,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GACvC,EAAI,EAAG,EAAI,EAAO,IACzB,OAAS,GAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EAAgB,KAAK,IAAI,EAAY,EAAG,KAAK,KAAK,IAClD,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAiB,EAAE,QAAQ,GAC7D,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAgB,EAAE,QAAQ,GACzD,EAAI,EAAG,EAAI,EAAU,IAU5B,OATM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EACF,KAAK,IAAI,EAAW,EAAG,KAAK,KAAK,IAC/B,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GACvD,EAAI,EAAG,EAAI,EAAa,KAI/B,GAAM,GAAU,EAAQ,EAAgB,GAClC,EAAa,EAAQ,EAAgB,GACrC,GAAW,EAAQ,EAAiB,GACpC,GAAc,EAAQ,EAAiB,GAEvC,GAAM,EAAW,IAAW,GAAW,EACvC,GAAS,EAAc,IAAc,GAAc,EACnD,GAAW,GAAO,IAAS,IAAO,EAExC,EAAO,KAAe,GAK9B,MAAO,GAAA,OAAU,EAAQ,CAAC,EAAO,EAAW,EAAU,KAGxD,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAChD,GAAiB,CAAC,EAAI,GAAI,0BA+B1B,OA7BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GAOrD,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAMhD,EAAW,KAAK,SAAS,EAAG,QAC9B,EAAS,EACJ,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAU,EAAI,EAAE,QAAQ,GACrB,EAAI,EAAG,EAAI,EAAS,IAU3B,OATM,GAAM,EAAI,EACV,EAAc,KAAK,MAAM,GACzB,EAAiB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAU,GAEpD,EAAe,EAAU,EAAc,EAAE,QAAQ,GACjD,EAAkB,EAAU,EAAiB,EAAE,QAAQ,GAEvD,EAAU,EAAM,EAChB,EAAiB,EAAM,EACpB,EAAI,EAAG,EAAI,EAAQ,IAmB1B,OAlBM,GAAM,EAAI,EACV,EAAe,KAAK,MAAM,GAC1B,EAAgB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAS,GAClD,EAAU,EAAM,EAChB,EAAiB,EAAM,EAEvB,EAAkB,EAAe,EAAe,EAAE,QAAQ,GAC1D,EAAmB,EAAe,EAAgB,EAAE,QAAQ,GAC5D,GACF,EAAkB,EAAe,EAAE,QAAQ,GACzC,GACF,EAAkB,EAAgB,EAAE,QAAQ,GAE1C,GACF,EAAiB,EACf,GAA6B,EAAiB,EAC9C,GAA6B,EAAU,EACvC,GAAsB,EAAU,EAC7B,GAAI,EAAG,GAAI,EAAO,MACzB,GAAM,IAAQ,EAAS,KACvB,EAAO,EAAkB,KACrB,GAAQ,GACZ,EAAO,EAAmB,KAAM,GAAQ,GACxC,EAAO,GAAqB,KACxB,GAAQ,GACZ,EAAO,GAAsB,KAAM,GAAQ,GAKnD,MAAO,GAAA,SAAY,EAAQ,CAAC,EAAO,EAAQ,EAAS,GAAQ,EAAE,QAGhE,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,yBAsBpB,OApBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cAAa,EAAQ,EAAY,EAAW,GAEzD,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GAE5C,EAAe,EACV,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAY,EACZ,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAc,EAAmB,EAAE,QAAQ,GACpD,EAAI,EAAG,EAAI,EAAU,IAO5B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAW,EACX,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAY,EAAmB,EAAE,QAAQ,GAClD,EAAI,EAAG,EAAI,EAAa,KAG/B,GAAM,GAAS,EAAQ,EAAY,GACnC,EAAO,KAAkB,EAKjC,MAAO,GAAA,OACH,EAAQ,CAAC,EAAO,EAAW,EAAU,GAAc,EAAE,QAG3D,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,iCAiC1B,OA/BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GACrD,EAAW,KAAK,SAAS,EAAG,QAK5B,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAGzC,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAS,IAM3B,OALM,GAAY,EAAc,EAAI,EAAE,QAAQ,GAGxC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAY,GAC7C,EAAI,EAAG,EAAI,EAAQ,IAO1B,OANM,GAAY,EAAY,EAAI,EAAE,QAAQ,GAGtC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAW,GAE5C,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHI,GAAQ,EAGH,EAAW,EAAG,EAAW,EAAW,KAC3C,GAAM,GAAM,EAAW,EAEvB,GAAI,EAAM,GAAK,GAAO,EACpB,SAGF,GAAM,GAAY,EAAc,EAAM,EAAG,QAAQ,GAC3C,GAAgB,EAAM,EACtB,GAAmB,KAAK,IAC1B,EAAU,EACV,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAC9B,GAAI,IAAM,GACR,SAEF,OAAS,IAAW,EAAG,GAAW,EAAU,MAC1C,GAAM,IAAM,GAAW,EAEvB,GAAI,GAAM,GAAK,IAAO,EACpB,SAGF,GAAM,IAAY,EAAY,GAAM,EAAG,QAAQ,GACzC,GAAgB,GAAM,EACtB,GAAmB,KAAK,IAC1B,EAAS,EACT,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAE9B,AAAI,IAAM,IACR,IAAS,EAAS,GAAY,KAIpC,EAAO,EAAY,GAAK,EAKhC,MAAO,GAAA,SAAY,EAAQ,EAAE,MAAO,EAAE,QAGxC,EAAA,UAAA,6BAAA,SACI,EAAa,EAAqB,EAAc,EAChD,GACF,GAAiB,EAAG,gCAEpB,GAAM,GAAW,EAAE,MAAM,GACnB,EAAO,EAAW,EAClB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAO,EAAE,KACT,EAAS,GAAI,cAAa,GAEhC,WAA2B,GAQzB,OAPM,GAAiB,EAAS,EAC5B,EACA,EAAS,EAAiB,KAAK,IAAI,EAAG,EAAiB,GACrD,EAAe,EAAS,EAC1B,KAAK,IAAI,EAAiB,EAAa,GAEvC,EAAM,EACH,GAAkB,EAAc,KACrC,GAAM,GAAI,EAAQ,GAClB,GAAO,EAAI,EAEb,MAAO,GAGT,OAAS,GAAS,EAAG,EAAS,EAAM,KAClC,GAAM,GAAM,EAAkB,GACxB,EAAM,EAAQ,GAAU,KAAK,IAAI,EAAO,EAAQ,EAAK,CAAC,GAC5D,EAAO,GAAU,EAGnB,MAAO,GAAA,SAAY,EAAQ,EAAE,QAG/B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAiB,EAAI,WAQrB,OAPM,GAAW,EAAG,MAAM,GACpB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAmB,KAAK,SAAS,EAAW,QAC5C,EAAoB,KAAK,SAAS,EAAY,QAC9C,EAAS,GAAI,cAAa,EAAG,MAC7B,EAAO,EAAG,KAEP,EAAS,EAAG,EAAS,EAAM,KAQlC,OAPM,GAAiB,EAAS,EAC1B,EACD,EAAS,EAAkB,KAAK,IAAI,EAAG,EAAiB,GACvD,EAAY,EAAS,EACvB,KAAK,IAAI,EAAU,EAAiB,EAAc,GAElD,EAAO,EACF,EAAI,EAAY,EAAI,EAAU,IACrC,GAAQ,KAAK,IAAI,EAAiB,GAAI,GAExC,EAAO,EAAQ,EAAO,EAEtB,OAAS,GAAI,EAAY,EAAI,EAAU,KACrC,GAAI,GAAM,GAAK,EAAQ,EAAO,EAAiB,GAC3C,EAAkB,GAAU,EAChC,AAAI,IAAW,GACb,IAAO,KAAK,IAAI,EAAM,CAAC,IAEzB,GAAO,EAAS,GAChB,EAAO,IAAM,GAGjB,MAAO,GAAA,SAAY,EAAQ,EAAG,QAGhC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAiB,EAAQ,eASzB,OAPM,GAAgB,EAAa,EAAS,EAAA,QAAW,GACjD,EAAY,EAAc,MAAM,GAChC,EAAY,EAAc,MAAM,GAChC,EAAM,EAAA,MAAkB,CAAC,EAAW,GAAa,SACjD,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAW,KAAK,SAAS,EAAc,QAEpC,EAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,GAAS,EAAI,EAGb,EAAM,GAAI,cAAa,EAAY,GACzC,EAAI,GAAK,EAAS,GAClB,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,EAAE,EACxC,EAAI,GAAS,EAAI,EAAQ,GAAK,EAAS,EAAS,GAKlD,OAFM,GAAS,GAAA,KAAgB,EAAK,YAC9B,EAAY,EAAI,EACb,EAAW,EAAG,EAAW,EAAY,EAAE,GAC9C,GAAM,GAAI,IAGV,EAAQ,EAAY,GAAY,EAAI,OAEpC,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,IACtC,GAAI,EAAI,EAAI,IACV,EAAQ,EAAY,GAAY,EAChC,QAKR,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAiB,EAAS,UAE1B,GAAM,GAAM,GAAI,cAAa,EAAQ,KAAO,GAC5C,EAAI,KAAK,GAGT,OAFM,GAAa,KAAK,SAAS,EAAQ,QAEhC,EAAQ,EAAG,EAAQ,EAAQ,KAAM,EAAE,EAC1C,AAAI,EAAW,IAAU,GAAK,EAAW,GAAS,GAChD,GAAI,EAAQ,EAAQ,EAAW,IAAU,GAG7C,MAAO,GAAA,SAAY,EAAK,CAAC,EAAQ,KAAM,GAAQ,UAGjD,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,GAAiB,EAAO,qBAExB,GAAM,GAAY,KAAK,SAAS,EAAM,QAChC,EAAa,KAAK,SAAS,EAAO,QACxC,MAAO,IACH,EAAW,EAAY,EAAe,EAAc,IAG1D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,IAAe,OACf,WAAM,MAAA,+DACF,IACR,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAgB9D,OAdM,GAAY,EAAE,MAAM,GACpB,EAAc,EAAE,MAAM,GACtB,EAAa,EAAE,MAAM,GACrB,EAAa,EAAE,MAAM,GAErB,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAU,KAAK,SAAS,EAAE,QAC1B,EACF,GAAI,cAAa,EAAY,EAAe,EAAc,GAE1D,EAAY,EACP,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAI,EAAG,EAAI,EAAc,EAAE,EAGlC,OAFM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACZ,EAAI,EAAG,EAAI,EAAa,EAAE,EAIjC,OAHM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACf,EAAW,GAAU,EAAY,GAAW,EACzC,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAM,EAAI,EACV,EACF,EAAM,EAAc,GAAM,EAAc,GAAM,EAAc,IAChE,EAAO,KAAe,EAAQ,GAKtC,MAAO,GAAA,SACH,EAAQ,CAAC,EAAW,EAAc,EAAa,KAG7C,EAAA,UAAA,oBAAR,SACI,EAAW,EAAW,EACtB,GACF,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAE,MAAO,EAAE,OAC9D,EAAS,EAAA,OAAU,EAAU,GAC7B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GACxD,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GAExD,EAAU,EAAO,OACvB,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,EACpC,EAAQ,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAK3D,QAFM,GAAO,KAAK,WAAW,GACvB,EAAO,KAAK,WAAW,cACpB,GACP,GAAM,GAAM,EAAO,WAAW,GAExB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAEzB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAE/B,EAAQ,GAAK,EAAG,EAAM,GAAS,EAAM,KAX9B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,IAA7B,GAcX,MAAO,GAAO,YAGhB,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,QAAA,aAEA,EAAA,UAAA,eAAA,WACE,MAAO,KAIT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAGtB,EAAA,UAAA,cAAA,SACI,EACA,EACA,EACA,EACA,EACA,GAmBF,OAjBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GACjC,EAAW,EAAM,MAAM,GAEtB,EAAA,EAAA,GAAY,EAAA,EAAA,GACb,EACF,EAAA,OAAU,CAAC,EAAU,EAAY,EAAW,GAAc,WAExD,EAAU,KAAK,SAAS,EAAM,QAC9B,EAAa,KAAK,SAAS,EAAS,QACpC,EAAY,KAAK,SAAS,EAAO,QAEjC,EAAW,EAAO,QAClB,EAAY,EAAO,QAKhB,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAW,EAAI,EACf,EAAK,EAAQ,GACb,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GAExB,EAAe,EAAW,GAChC,GAAI,GAAQ,EACV,SASF,OANM,GAAe,EAAa,EAC7B,GAAK,GAAO,GAAc,GAAM,GAAa,GAC9C,EACE,EACD,EAAY,EAAM,GAAK,GAAO,GAAa,GAAM,GAAY,GAAK,EAE9D,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAgB,EAAa,EAC/B,EAAM,GAAc,GAAK,EAAK,EAC9B,GAAO,GAAK,GAAO,GAAc,GAErC,GAAI,EAAO,GAAK,EAAO,EAAc,GACnC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAGzB,SAGF,GAAI,IAAW,WAKb,OAJM,GAAS,KAAK,MAAM,GACpB,EAAY,KAAK,KAAK,GACtB,GAAQ,EAAO,EAEZ,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAOF,OAJM,IAAU,KAAK,MAAM,IACrB,GAAW,KAAK,KAAK,IACrB,GAAQ,GAAO,GAEZ,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAI,GAAM,EAAI,GAAU,EAAS,GAAK,EAAS,EAAS,GACpD,EAAO,EAAS,GACd,GAAU,EAAU,GAE1B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAS,EAAS,GACjD,EAAO,EAAS,GACpB,GAAM,IAAW,EAAU,GAE3B,EAAM,EAAI,GAAU,EAAS,GAAK,EAAY,EAAS,GACnD,EAAO,EAAS,GACpB,GAAM,IAAa,EAAU,GAE7B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAY,EAAS,GACpD,EAAO,EAAS,GACpB,GAAM,IAAc,EAAU,GAExB,GAAM,GAAW,IAAW,IAAW,GACvC,GAAS,GAAc,IAAc,IAAc,GAEzD,EAAM,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC9D,EAAO,OAAO,GAAO,GAAQ,IAAS,IAAO,QAIjD,QAAS,GAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAKF,OAFM,IAAW,KAAK,MAAM,IACtB,GAAW,KAAK,MAAM,GACnB,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,IAAQ,EAAI,GAAW,EAAS,GAClC,GAAW,EAAS,GAAK,EAAO,EAAS,GACvC,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,IAAU,EAAU,OAM5C,MAAO,GAAO,YAGhB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAiB,GACvB,MAAO,MAAK,QACR,EAAe,EAAc,EAAa,EAAY,EACtD,EAAY,EAAW,EAAS,EAAc,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAE1C,GAAI,IAAc,EAChB,MAAO,GAAA,OAAU,GAAI,EAAa,EAAE,OAOtC,OAJM,GAAS,GAAI,GAAA,aAAa,CAAC,EAAW,GAAY,EAAE,OACpD,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAQ,KAAK,SAAS,EAAE,QAErB,EAAI,EAAG,EAAI,EAAW,KAG7B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,GAAgB,EAAM,EAAQ,GAC9B,EAAM,KAAK,GAEb,GAAI,EAAe,GAAK,GAAgB,EAAE,KAAO,EAC/C,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,EAAE,OAGzD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,EAAO,OAAO,EAAI,EAAY,GAAK,EAAM,EAAe,EAAY,GAGxE,MAAO,GAAO,WAAW,QAAQ,IAGnC,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAe,EAAA,OAAU,GACzB,EAAiB,GACvB,MAAO,MAAK,QACR,EAAS,EAAS,EAAO,EAAY,EAAW,EAAY,EAC5D,EAAS,EAAc,IAG7B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,EAAQ,GAAS,EAAA,KAAK,WAAW,GACjC,GAAM,GACF,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IACrD,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,OAGnD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAEhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,GAAM,GAAS,EAAA,KAAK,kBACD,EAAE,MAAO,EAAA,KAAK,cAAc,EAAE,QACjD,MAAO,MAAK,WAAW,EAAQ,EAAE,MAAO,EAAE,QAG5C,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGxC,EAAA,UAAA,QAAR,SACI,EAAiB,EAAiB,EAAoB,EACtD,EAAmB,EAAoB,EACvC,EAAmB,EACnB,GACF,GAAM,GAAe,CAAC,EAAa,EAAW,GAExC,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAc,KAAK,SAAS,EAAQ,QAE1C,GAAI,IAAe,EACjB,MAAO,GAAA,OAAU,GAAI,EAAO,EAAQ,OAGtC,GAAM,GAAS,GAAI,GAAA,aAAa,EAAc,EAAQ,OACtD,EAAO,OAAO,KAAM,KAAK,SAAS,EAAa,QAAuB,IAEtE,OAAS,GAAI,EAAG,EAAI,EAAY,KAG9B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,EAAM,KAAK,GACX,GAAgB,EAAM,EAAQ,GAGhC,GAAI,EAAe,GAAK,GAAgB,EAAa,EACnD,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,GAGvD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,AAAI,EACF,EAAO,OAAO,EAAe,EAAY,IACrC,EAAY,EAAI,EAAY,GAEhC,EAAO,OAAO,EAAe,EAAY,GAAK,EAAQ,OAAS,EAC3D,EAAY,GACZ,EAAY,EAAI,EAAY,GAItC,MAAO,GAAO,WAAW,QAAQ,IAErC,GAzjFoC,EAAA,eC5DpC,YAqB8B,GAE5B,OADM,GAAe,GAAI,cAAa,EAAK,QAClC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAa,GAAK,KAAK,IAAI,EAAK,IAElC,MAAO,GAGF,GAAM,IACT,SAAC,GACQ,GAAA,GAAA,EAAA,OAAA,EACD,EAAa,EAAK,QACpB,EAAe,GAAI,cAAa,EAAA,KAAK,cAAc,EAAE,QACzD,GAAI,EAAE,QAAU,aACd,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC7C,EAAe,GAAc,OAS7B,QAPM,GAAc,EAAW,KAAK,IAAI,EAAE,QACpC,EAAO,EAAY,mBAAmB,KACtC,EAAO,EAAY,mBAAmB,KACtC,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC/B,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAO,EAAS,GAChB,EAAO,EAAS,GACtB,EAAa,GAAK,KAAK,MAAM,EAAM,GAGvC,MAAO,GAAW,WAAW,EAAc,EAAE,MAAO,YAG7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICzDd,YAwB6C,GAE3C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAmB,GACzB,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAE3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAEhC,EACF,EAAA,KAAK,uBAAuB,EAA0B,GAEpD,EAAQ,EAAO,OACf,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAE7D,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAG1D,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAE5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,KAX7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,IAA5B,GAeX,MAAO,CAAC,EAAQ,IClEpB,YAqBwB,GAEf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,KAAM,EAAA,EAAA,KAEP,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACzC,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAEzC,EAAc,EAAQ,eAAe,EAAK,MAAO,aAEjD,EAAU,EAAQ,KAAK,IAAI,EAAY,QAK7C,SAAQ,mBAAqB,CAC3B,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,GACpD,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,IAG/C,EAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC/Cd,YAsBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IClCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAGP,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAIvC,GAAM,GAAc,EAAA,MAAS,EAAE,OACzB,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,EAAE,MAAO,CAAC,MAAO,aAEpD,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAQ,KAAM,GAAc,QAAO,IAE/D,SAAY,UACZ,EAAQ,8BAA8B,GAE/B,EAIT,GAAI,EAAE,QAAU,aACd,GAAM,GAAW,GAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,IAC5C,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAO,EAAE,MAAO,CAAC,MAAK,KAElE,SAAQ,8BAA8B,GAE/B,EAGT,GAAI,CAAC,EAAA,KAAK,gBAAgB,EAAE,MAAO,IAGjC,GAAM,GAAS,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAC7C,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAO,MAAO,MAAK,GAG3D,GAAI,IAAU,SACZ,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAe,WAAW,KAAK,GACrC,MAAO,GAAQ,eAAe,EAAE,MAAO,QAAS,GAGlD,GAAI,IAAU,QAIZ,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAO,EAAA,KAAK,aAAa,CAAC,GAAI,EAAE,OAEhC,EAAA,GAAA,SAAA,EAAA,GAAA,MAAA,KAAA,EAAA,EAAA,IAAA,EAAA,MAAA,GAAA,EAAA,EAAA,QAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAQ,eAAe,EAAa,OAAQ,GAGrD,KAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,GAG1D,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC9Fd,YAsCI,EAAc,EACd,EAAuC,GACzC,MAAI,IAAe,KACV,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAiB,CAAC,EAAG,GAAI,GAEzB,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,IAInD,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,aACzC,GAAM,GAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGjC,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAY,KAAM,GAAa,QAAS,IAE5D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,OAEP,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,iBASd,GAE5C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAyB,EACzB,GACN,GAAM,GAAc,EAAA,aAAa,2BAA2B,EAAQ,GAC9D,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAY,OACzB,EAAgB,EAAA,KAAK,eAAe,GAEpC,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GACxD,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GAExD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAEvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GACvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GAEvD,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAO,EAAI,EAAM,OACjB,EAAO,EAAI,EAAM,OAEjB,EACF,EAAG,EAAM,EAAO,GAAI,EAAM,EAAO,EAAI,GAAI,EAAM,EAAO,GACnD,EAAM,EAAO,EAAI,IAExB,EAAe,GAAK,EAAO,KAC3B,EAAe,GAAK,EAAO,SAG7B,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EACF,EAAG,EAAM,EAAS,GAAI,EAAM,EAAS,EAAI,GAAI,EAAM,EAAS,GACzD,EAAM,EAAS,EAAI,IAE1B,EAAe,GAAK,EAAS,KAC7B,EAAe,GAAK,EAAS,MAhBtB,EAAI,EAAG,EAAI,EAAe,OAAQ,MAAlC,GAmBX,MAAO,CAAC,EAAgB,EAAgB,ICvL5C,AAsBO,GAAM,IAAU,GAA8B,SAAC,EAAG,GAAM,MAAA,GAAI,IACtD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAGpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAwBsC,GAEpC,MAAO,UAAC,EAAQ,EAAO,GAGrB,OAFM,GACF,EAAA,KAAK,uBAAuB,EAA0B,EAAO,QACxD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,IChCX,YAiCI,EAAc,EAA0B,GAC1C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAQlB,OALM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,GAAS,EAAE,MACpB,EAAY,EAAA,KAAK,kBAAkB,EAAQ,GACxC,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,gBAclD,EAAc,EAA4B,GAC5C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAGlB,GAAM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAS,GAAS,EAAE,MACpB,EAAY,EAAU,EAAQ,EAAQ,GAC5C,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,IC3EtD,AAsBO,GAAM,IAAW,GAAsB,SAAC,GAAO,MAAA,MAAK,KAAK,KACnD,GAAiB,GAAwB,EAAA,KAAM,IAE/C,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,YAoBI,EAAmB,EAAoB,EACvC,GAIF,OAHM,GAAO,EAAA,KAAK,uBACd,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,GCpCT,AAqBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CACL,KAAM,EAAQ,EAAQ,EAAQ,EAC9B,KAAM,EAAQ,EAAQ,EAAQ,KAIzB,GACT,GAAiB,EAAA,SAAU,GAAc,IAEhC,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICrCd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,GAAI,KAAK,KAAK,KACxD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,YAuBI,EAAkB,EAAiB,EAAgB,EACnD,GACF,GAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,GACF,GAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAIhD,OADM,GAAU,EAAA,KAAK,uBAAuB,EAA0B,GAC7D,EAAI,EAAG,EAAI,EAAQ,EAAE,GAC5B,GAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,SAAC,EAAa,GAAM,MAAA,GAAM,EAAM,KAC/C,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEpB,MAAO,eAIL,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAAO,EAAA,EAAA,KAEd,GAAiB,EAAG,SAEd,GAAA,GAAA,EAAA,WAAA,iBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAA,WAAW,kBAAkB,EAAG,EAAQ,GAExC,GAAM,GAAO,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAClC,EAAU,GAAU,EAAM,EAAQ,EAAO,EAAE,MAAO,EAAE,OAC1D,MAAO,GAAQ,eAAe,EAAO,EAAE,MAAO,GAGzC,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IClEd,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAEpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAqBI,EAAmB,EAAkB,EAAiB,EACtD,GASF,OARM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GAEjC,EAAS,EAAA,KAAK,uBAChB,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAO,EAAE,GAK3B,OAJM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAGhC,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,EAAK,IAGvB,GAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE3B,MAAO,GC3CT,YAoBI,EAAuB,EAAc,EAAiB,GAgExD,OA1DM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAyDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GACtB,EAAI,EAAG,EAAI,EAAO,IACzB,EAAS,IAAM,EAAM,GAEvB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACxC,EAAS,IAAM,EAAM,GAevB,OAVM,GAA0C,GAG1C,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAA0B,GAC1B,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EAC/C,EAAI,EAAG,EAAI,EAAM,GAAQ,KAEhC,GAAI,GAAO,OACX,GAAI,EAEF,EAAU,EAAO,GAAG,gBAGpB,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG1C,EAAU,EAAW,KAAK,KAI5B,GAAI,EAAe,KAAa,OAC9B,EAAQ,GAAK,EAAe,QAE5B,GAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAOvB,GAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,GAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,SAAC,EAAoB,GACzC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAOxE,GAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAE7B,CACL,aAAc,EAAa,OAC3B,YAAW,EACX,QAAO,GCzJX,+MCAA,GAGM,IAAU,QCHhB,AA8BA,EAAA,gBAAgB,MAAO,WAAM,MAAA,IAAI,KAAkB,GC9BnD,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,YAoBI,EAAqB,EAAkB,EAAiB,EACxD,EACA,GAsBF,OArBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAS,EAAA,OAAO,EAAS,SAAU,GACnC,EAAa,EAAO,OAEpB,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAoB,EAAI,EACxB,EAAmB,EAAI,EAAQ,GAC5B,EAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAM1C,OALM,GAAW,EAAK,EAAe,EAC/B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,EAAkB,EAAoB,EAAK,EACxC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAQzC,OAPM,GAAW,EAAK,EAAc,EAC9B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAc,EACd,EAAW,EACX,EAAQ,EACH,EAAK,EAAO,EAAK,EAAO,GAAM,GAErC,OADM,GAAW,EAAmB,EAAK,EAAQ,GACxC,GAAK,EAAO,GAAK,EAAO,IAAM,GACrC,GAAM,IAAW,EAAW,GAAK,EAAQ,GACnC,GAAQ,EAAQ,GAAW,GACjC,AAAK,IAAa,OAAS,GAAQ,EACjC,EAAc,GACT,AAAI,IAAa,OACtB,IAAY,GACZ,KAGJ,GAAI,MAAM,GACR,MAGJ,GAAM,IAAe,EAAkB,EAAK,EAAmB,EAC/D,EAAW,IACP,IAAa,MAAQ,EAAW,EAAQ,EAKpD,MAAO,eAIL,EAAqB,EAAkB,EACvC,EAAmC,EACnC,GADmC,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,IAYF,OAXM,GAAe,EAAA,OAAO,EAAS,SAAU,SACzC,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,EAAA,OAAO,EAAQ,EAAO,GAC1B,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,GAG1C,OAFM,GAAW,EAAK,EAAe,EACjC,EAAQ,EACL,EAAQ,GACb,GAAS,EAKX,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAGzC,OAFM,GAAW,EAAK,EAAc,EAChC,EAAQ,EACL,EAAQ,GACb,GAAS,EAOX,OALM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAK,EAAO,EAAK,EAAO,GAAM,EAErC,OADM,GAAK,EAAK,EACP,EAAK,EAAO,EAAK,EAAO,GAAM,GACrC,GAAM,GAAK,EAAK,EACV,EAAQ,EAAK,IAAI,EAAG,EAAI,EAAI,GAClC,AAAI,EAAQ,GACV,GAAW,EACX,AAAI,EACF,EAAc,EACR,IAAI,EAAS,SAAW,GAAM,EAAS,QAAU,GAC3C,EAAS,WACb,EACH,GAAK,EAAS,QAAU,GAAM,EAAS,WAAa,EAEzD,EAAc,EAAK,EAAuB,GAKlD,EAAa,IAAI,EAAa,EAAG,EAAI,EAAI,IAKjD,MAAO,GCrJT,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAqBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBAyB9B,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAgB,EAAK,GAAe,GAEpC,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAGF,GAAM,GAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAGf,EAAG,IAAI,EAAU,EAAe,EAAG,EAAK,EAAK,GAKrD,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,IC1Fd,YAsBoC,GAK3B,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EAAG,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,KAAM,EAAA,EAAA,SAE/B,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGV,GAAiB,CAAC,EAAG,EAAM,EAAU,EAAO,GAAS,aAEhD,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAsBpB,OAnBM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACtC,EAAU,EAAQ,KAAK,IAAI,EAAS,QAAQ,OAC5C,EAAQ,EAAQ,EAAQ,KAAK,IAAI,EAAM,QAAQ,OAC/B,GAAI,cAAa,CAAC,IAClC,EAAU,EACZ,EAAQ,KAAK,IAAI,EAAO,QAAQ,OAChC,GAAI,cAAa,CAAC,IAChB,EAAU,GAAI,cAAa,EAAM,QAEjC,EAAgB,EAAQ,OACxB,EAAc,EAAM,OACpB,EAAgB,EAAQ,OACxB,EAAc,EAAM,OAEtB,EAAO,EACP,EAAK,EACL,EAAK,EACL,EAAK,EACA,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,EAAQ,GAAK,EAAQ,KAChB,GAAM,GAAK,EAAM,MAAS,EAAM,KAC7B,KAAK,KAAK,EAAQ,KAAQ,GAClC,AAAI,GAAQ,GACV,GAAO,GAET,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAGT,MAAO,GAAQ,eAAe,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAM,IAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,IC5Fd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,YAAa,SAAC,EAAI,GAC9D,GAAM,GAAY,EAClB,MAAI,GAAK,EAAU,aACV,EAAU,aAEZ,EAAK,EAAU,aAAe,EAAU,aAAe,IAGnD,GAA2B,CACtC,WAAY,EAAA,YACZ,YAAa,MACb,WAAY,IChCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YAsBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAED,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,EAAQ,OAAO,EAAE,QAEjB,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAEjC,GAAI,EAAM,oBAAsB,MAC9B,GAAM,GAAO,EAAM,mBAAmB,KAChC,EAAO,EAAM,mBAAmB,KAEtC,EAAK,MAAQ,EACb,EAAK,MAAQ,EAGf,MAAO,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,OAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,ICzDd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,KAED,EAAQ,EAAA,KAAK,eAAe,EAAM,EAAO,GAAG,OAAO,GACrD,EAAW,EAAA,aAAa,gBAAgB,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtE,GAAI,EAAA,KAAK,cAAc,KAAc,EACnC,MAAO,GAAQ,eAAe,EAAU,EAAO,GAAG,MAAO,IAI3D,GAAM,GAAU,EAAO,OAAO,SAAA,GAAK,MAAA,GAAA,KAAK,cAAc,EAAE,OAAS,IACjE,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAM,GAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAGlC,GAFA,EAAA,aAAa,uBAAuB,EAAQ,GAExC,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAC5D,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAE5D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAC3D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAE3D,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAO,IAEtE,SAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAUT,GAAM,GAAW,EAAQ,IAAI,SAAA,GAC3B,GAAM,GAAY,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAC7C,EAAQ,CAAC,GAAI,GACnB,MAAO,IAAQ,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAO,EAAE,MAAO,CAAC,MAAK,OAIxD,EACI,EAAA,aAAa,gBAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAE7D,GAAM,GAAU,EAAA,KAAK,uBACjB,EAAQ,GAAG,MAAoB,EAAA,KAAK,cAAc,IAEtD,GAAI,EAAS,GAAG,MAAM,KAAO,GAE3B,GAAI,GAAS,EACb,EAAS,QAAQ,SAAA,GACf,GAAM,GAAM,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACjC,EAAO,EAAA,KAAK,cAAc,EAAE,OAElC,EAAQ,IAAI,EAAK,GACjB,GAAU,SAGZ,GAAI,GAAY,EAEhB,EAAS,QAAQ,SAAA,GAKf,OAJM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAErC,EAAO,EAEF,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EAEpC,OADM,GAAS,EAAM,EAAS,GAAK,EAC1B,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EACpC,EAAQ,EAAS,GAAO,EAAM,KAIlC,GAAa,EAAE,MAAM,KAIzB,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtD,EACF,EAAQ,eAAe,EAAe,EAAO,GAAG,MAAO,GAE3D,SAAS,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KAErD,EAGF,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,ICjId,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,SAAC,GAwCX,OAxCY,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OACJ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAE,MAAM,OAEhB,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAChD,EAAa,EAAO,MAAM,OAE1B,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOI,EAAU,EAAA,KAAK,cAAc,GAC7B,EAAU,EAAS,OACnB,EAAa,EAAA,KAAK,kBAAkB,EAAE,MAAO,GAM1C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,IAAO,EAAO,EAAc,EAAQ,KACjC,GAAI,EAAG,GAAI,EAAY,EAAE,IAEhC,OADI,IAAS,OAAO,iBACX,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,GAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAS,EAAA,KAAK,WAChB,CAAC,EAAG,GAAK,GAAK,IAAI,EAAO,EAAA,KAAK,eAAe,EAAE,QAC7C,GAAc,EAAA,KAAK,WACrB,CAAC,GAAG,GAAG,IAAI,EACX,EAAA,KAAK,eAAe,EAAO,QACzB,GAAM,EAAM,IAAU,EAAW,IACvC,AAAI,GAAM,IACR,IAAS,MAMnB,GAAM,IAAc,EAAA,KAAK,WACrB,CAAC,EAAG,EAAM,EAAM,IAAI,EAAS,EAAA,KAAK,eAAe,IACrD,EAAW,IAAe,GAMlC,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAY,EAAE,OAAQ,EAAU,EAAE,OAExD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAU,MAAO,EAAE,SCrG9C,AAsBO,GAAM,IAA+C,CAC1D,WAAY,EAAA,yBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,yBAAwB,QACtC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EAAY,EAAA,KAAK,0BACD,EAAO,MAAO,EAAO,OAOlC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAO,EACP,GAAO,EACF,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAO,GACP,GAAO,MAMjB,EAAU,IAAM,IAAM,IAAM,EAAI,GAAG,GAAM,GAAM,GAMvD,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAO,MAAO,EAAO,OAEhE,MAAO,CAAC,OAAM,GAAE,MAAO,EAAO,MAAO,MAAO,EAAO,SCrHvD,AAsBO,GAAM,IAA8C,CACzD,WAAY,EAAA,wBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,wBAAuB,QACrC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EACF,EAAA,KAAK,0BAA0B,EAAE,MAAO,EAAE,OAOrC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAU,EAAO,EAAK,EAAI,EAC1B,GAAU,EAAO,EAAK,EAAI,EACrB,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAS,GACT,GAAS,MAMnB,EAAU,GAAG,IAAQ,IAAQ,IAAM,EAAI,GAAG,GAAM,GAAM,GAM9D,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAE,MAAO,EAAE,OAEtD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrH7C,AAsBO,GAAM,IACT,GAA6B,SAAC,EAAW,GAAc,MAAA,GAAI,IAClD,GAAM,GAAiB,EAAA,IAAK,IAE5B,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC7Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,IAAM,EAAI,EAAM,KAAK,IAAI,GAAM,IAEnD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC3Bd,AAqBA,GAAM,IAAI,EAAA,aAAa,MACjB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAEX,GAAgB,GACzB,EAAA,IACA,SAAC,GACC,GAAM,GAAO,KAAK,KAAK,GACjB,EAAI,KAAK,IAAI,GACb,EAAI,EAAO,GAAM,GAAI,GAC3B,MAAO,GACF,GACK,OAAK,EAAI,IAAM,EAAK,IAAM,EAAI,IAAM,EAAI,IAAM,EAC/C,KAAK,IAAI,CAAC,EAAI,MAIhB,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Cd,YAmCI,EAAmB,EACnB,GAgBF,OAfM,GAAa,EAAM,MACnB,EAAQ,EAAW,GACnB,EAAW,EAAW,GAEtB,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EAAS,EAAU,mBAAmB,KACtC,EAAS,EAAU,mBAAmB,KAGtC,EAAc,CAAC,EAAO,GACtB,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAA,KAAK,uBAAuB,UAAW,GACpD,EAAa,EAAA,KAAK,uBAAuB,UAAW,GAEjD,EAAI,EAAG,EAAI,EAAO,KAmBzB,OAjBM,GAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAE7B,EAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAG7B,EAAQ,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAG,KAAM,GAAI,QAAS,IAGtD,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,KACP,EAAM,EAAA,aAAa,uBAAuB,EAAM,GAE7C,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAI,EAAA,aAAa,oBAAoB,EAAK,GAChD,EAAW,EAAI,EAAW,GAAK,EAAE,KACjC,EAAW,EAAI,EAAW,GAAK,EAAE,KAGnC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAG3C,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,GAChD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,GAAY,QAAS,IAE1D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,cAIL,EAAmB,EACnB,GACF,GAAM,GAAY,EAAA,KAAK,cAAc,EAAM,OAErC,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAG5D,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAGlE,GAAI,GAAc,IAChB,GAAM,GACF,GAAU,EAAU,EAAU,EAAW,EAAS,GAEhD,EAAc,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAEjD,GAAI,GACF,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,EAAO,MACvD,EACF,EAAW,eAAe,EAAa,UAAW,EAAO,MAEvD,EAAuB,EAAW,eACpC,GAAI,UACJ,EAAA,KAAK,kBAAkB,EAA8B,YACnD,EACF,GAAS,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAS,IAExC,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAW,QAAS,IAEhD,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAe,QAAS,IAGpD,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OACtC,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OAE5C,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,CAAC,KAAM,EAAa,KAAM,GAGnC,MAAO,QAEP,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EACF,GAAyB,EAAM,EAAW,GAE9C,MAAO,GAAA,aAAa,uBAAuB,IAI/C,YAAuB,GACrB,MAAQ,GAAO,EAAO,KAAO,EAI/B,YACI,EAAwB,EAAwB,EAChD,EACA,GACF,GAAI,IAAS,EACX,MAAO,CAAC,KAAM,EAAU,KAAM,GAGhC,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EAAO,EAAO,EAEd,EAAc,EAAA,aAAa,qBAAqB,GAEhD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAa,EAAA,aAAa,oBAAoB,GAE9C,EAAc,EAAW,KACzB,EAAc,EAAW,KAEzB,EAAW,CAAC,EAAY,QAExB,EACF,EAAW,eAAe,EAAU,UAAW,GAC7C,EACF,EAAW,eAAe,EAAU,UAAW,GAE7C,EAAgB,GAClB,CAAC,OAAQ,CAAC,KAAM,EAAa,KAAM,GAAc,QAAS,IAGxD,EACF,GAAU,EAAc,EAAc,EAAM,EAAS,GAEnD,EAAgB,EAAa,KAC7B,EAAgB,EAAa,KAE7B,EAAa,CAAC,EAAc,QAE5B,EACF,EAAW,eAAe,EAAY,UAAW,GAC/C,EACF,EAAW,eAAe,EAAY,UAAW,GAE/C,EAAkB,GAAQ,CAC9B,OAAQ,CAAC,KAAM,EAAe,KAAM,GACpC,QAAS,IAGL,EACF,GAAU,EAAa,EAAa,EAAM,EAAS,GAEjD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAI,EAAA,aAAa,UAAU,EAAM,GACjC,EAAS,CAAC,EAAE,KAAK,QAEjB,EAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAC3D,GAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAE3D,GAAc,GAChB,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,IAAY,QAAS,IAEpD,GACF,GACI,CAAC,OAAQ,CAAC,EAAG,GAAa,EAAG,GAAiB,QAAS,IAGzD,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAErB,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAGrB,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAEV,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAGV,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAC9C,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAEpD,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IAElC,CAAC,KAAM,GAAW,KAAM,IAIjC,YACI,EAAkB,EAAc,GAGlC,OAFM,GAAM,GAAI,cAAa,EAAO,GAE3B,EAAI,EAAG,EAAI,EAAM,KAGxB,OAFI,GAAO,EACP,EAAO,EACF,EAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAI,EAAA,aAAa,SAAS,EAAI,EAAG,EAAM,GACvC,EAAO,EAAA,aAAa,oBAAoB,EAAsB,GACpE,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAC3C,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAE7C,AAAI,GACF,IAAQ,EACR,GAAQ,GAEV,EAAA,aAAa,mBAAmB,EAAK,EAAM,EAAM,GAEnD,MAAO,GCjVT,YAuBoB,GAEX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAO,GAElC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICtDd,AAsBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,MACb,WAAY,SAAC,GAUX,OAVY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GAEX,EAAS,KAAK,MAAM,EAAa,GACjC,EAAS,EAAc,EAAY,EAAY,EAEjD,EAAc,EAAU,GAE5B,GAAI,GAAU,GAAK,EAAS,GAE1B,GAAM,GAAmB,EAAS,EAC5B,EACF,EAAc,EAAY,EAAmB,EACjD,EAAc,EAAU,GAE1B,EAAO,GAAU,EAMzB,GAAM,GAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,EAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SCpErD,YAuBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAM,GAEjC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtDd,AAqBO,GAAM,IACT,GAAgB,EAAA,SAAU,SAAC,GAAO,MAAA,QAAO,SAAS,GAAM,EAAI,GAAG,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,IAAI,KAAQ,SAAW,EAAI,GAAG,QAEzD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,QAAO,MAAM,GAAM,EAAI,GAAG,QAEhD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAK,EAAI,GAAG,QAEvC,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC3Bd,AA2BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAa,EACf,EAAS,EAAE,MACT,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,EAAkB,GACnD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACvD,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC1C,GAAI,GAAgB,MAElB,OADM,GAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAO,EAAa,IAGpC,EAAQ,GAAc,EAAO,EAAQ,EAAE,MAAO,EAAc,GAC5D,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAElD,EAAS,EAGX,GAAiB,EAAG,OACpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGd,EAAa,EAAA,KAAK,cAAc,GAEhC,EAAS,GAAQ,EAAO,EAAY,EAAa,EAAE,OACnD,EAAS,EAAW,MAAM,EAAQ,EAAa,EAAE,OAEnD,EAAW,EACf,GAAI,GAEF,GAAM,GAAW,EAAA,aAAa,qBAAqB,EAAa,GAChE,EAAW,EAGb,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,SCtE9C,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAyBlC,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAY,EAAA,OACd,EAAS,SAAU,EAAE,MACrB,GAAiB,EAAS,EAAE,MAAO,EAAE,MAAO,GAAU,QACpD,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,GAAM,GAAS,EAAwB,EAAuB,EACzD,EAAU,IAAI,EAAG,EAAK,EAAK,GAC1B,GAAS,EAAK,EAAuB,EAErC,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAAQ,IAGvB,EAAG,IAAI,EAAS,EAAG,EAAK,EAAK,GAKrC,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,ICnGd,YAoBI,EAAqB,EAAkB,EACvC,EAA8B,GAChC,GAAM,GAAU,EAAA,KAAK,eAAe,GAC9B,EAAW,GAAK,EAAS,EAAQ,EAAO,EAAS,EAAU,OAC3D,EAAe,GACjB,EAAS,EAAQ,EAAO,EAAU,GAAM,GAE5C,MAAO,CAAC,EAAS,OAAQ,EAAa,QC3BxC,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAa,EACnB,GAAiB,EAAG,qBAEpB,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,CAAC,EAAG,GAAI,GACN,EAAA,GAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGT,EACF,EAAW,MAAM,EAAwB,EAAS,SAAU,EAAE,OAC5D,EACF,EAAW,MAAM,EAAuB,EAAS,SAAU,EAAE,OACjE,MAAO,CACL,CAAC,OAAQ,EAAc,MAAO,EAAS,SAAU,MAAO,EAAE,OAC1D,CAAC,OAAQ,EAAe,MAAO,EAAS,SAAU,MAAO,YC/C/D,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEnB,GAAiB,EAAO,2BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC3C7B,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEnB,GAAiB,EAAO,8BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAG,GAAM,MAAC,KAAM,EAAK,EAAI,IAC/C,GACT,GAAiB,EAAA,SAAU,GAAc,KAAsB,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC9Bd,YAuBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,SAAU,EAAA,EAAA,cAEjB,GAAiB,EAAG,OAEpB,GAAM,GAAW,EAAS,IACtB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAE,MAAM,GAAK,EAAE,KAE9C,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAE5B,EAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAQ,EAAE,MAAM,OAChB,EAAW,EAAA,KAAK,eAAe,EAAE,OAEjC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EACF,EAAA,KAAK,uBAAuB,EAAE,MAA0B,GAE5D,AAAI,IAAkB,GACpB,EAAQ,KAAK,GAGf,OAAS,GAAI,EAAG,EAAI,EAAO,KACzB,GAAM,GAAS,EAAA,KAAK,WAAW,EAAG,EAAO,GACnC,EAAY,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAM,KAC3C,EAAW,EAAA,KAAK,WAAW,EAAW,EAAY,GAExD,EAAQ,GAAY,EAAM,GAG5B,GAAM,GAAQ,EAAQ,MAAM,EAAS,EAAU,EAAE,OAEjD,MAAO,CAAC,OAAQ,EAAO,MAAO,EAAU,MAAO,EAAE,OAG5C,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICnEd,AAqBO,GAAM,IAAuB,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAI,IAE/D,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,MACb,WAAY,SAAC,GAiBX,OAjBY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAmB,IAEnB,EAAY,KAAK,IAAI,GACrB,EAAY,KAAK,IAAI,GACrB,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GACX,EAAI,EAAO,GAGb,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACrD,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACzD,EAAS,KAAK,MAAM,EAAS,GAC7B,EAAS,KAAK,MAAM,EAAS,GAE7B,GAAI,GAAc,EAUlB,GATA,AAAI,MAAO,IAAc,UACvB,CAAI,IAAY,EACd,EAAc,EAEd,EAAc,EAAU,IAKxB,GAAU,GAAK,EAAS,GAAc,GAAU,GAChD,EAAS,GAEX,GAAM,GAAmB,EAAU,GAAa,GAC1C,EAAmB,EAAS,EAC5B,GACF,EAAc,EAAmB,EAAmB,EACxD,EAAc,EAAU,IAG1B,GAAM,IAAS,EAAc,EAAY,EAAY,EACrD,EAAO,IAAU,EAMzB,GAAM,IAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,GAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SC3FrD,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAErD,GAAM,GAAO,KAAK,MAAM,GACxB,MAAI,GAAK,EAAO,GACP,KAAK,MAAM,GACT,EAAK,EAAO,GACd,KAAK,KAAK,GAEb,EAAO,IAAQ,EACV,EAEA,EAAO,IAKP,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICxCd,AAqBA,GAAM,IAAa,EAAA,aAAa,gBAC1B,GAAQ,EAAA,aAAa,WAEd,GAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,IAAM,EACD,GAAQ,EAER,GAAc,MAAK,IAAI,GAAM,KAI3B,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICnCd,AAqBO,GAAM,IACT,GAAgB,EAAA,QAAS,SAAC,GAAO,MAAA,GAAK,GAAI,KAAK,IAAI,CAAC,MAE3C,GAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,GAAK,EACA,GACE,EAAK,EACP,EAEA,IAIE,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IClCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AA0BA,GAAM,IAAU,sBACV,GAAY,KAAK,IAAI,IAAW,EAEzB,GAAqB,GAAgB,EAAA,SAAU,SAAC,GAG3D,GAAM,GAAW,EAAK,CAAC,GAIjB,EAAW,EAAK,GAEhB,EAAO,KAAK,IAAI,GAClB,EAEJ,MAAI,GACF,EAAS,EACJ,AAAI,EACT,EAAS,EAET,EAAS,KAAK,IAAI,EAAM,GAEnB,IAGI,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICtDd,YAwB0B,GAKjB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EACA,EAAA,EAAA,KAEP,GAAiB,EAAG,aAKpB,OAHM,GAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAS,GAAc,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAEvD,EAAS,EAAQ,MAAM,EAAQ,EAAU,EAAE,OACjD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,OAGrC,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,MACb,WAAY,ICpDd,YA0B+B,GAKtB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,WAAY,EAAA,EAAA,SAEnB,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAA,KAAK,cAAc,GAE1B,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAA0B,GAE1B,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,GAAY,WAAW,CACrC,OAAQ,CAAC,EAAC,GACV,QAAO,EACP,MAAO,CAAC,SAAU,EAAkB,cAAe,KAG/C,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EACF,EAAA,aAAa,oBAAoB,EAAQ,MAAO,EAAY,EAAM,IAEhE,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,MAAO,GACrC,EACF,GAAQ,CAAC,OAAQ,EAAe,QAAO,EAAE,MAAO,IAE9C,EAAmC,CAAC,EAAG,GACvC,EACe,CAAC,KAAM,GACtB,EACF,GAAU,CAAC,OAAQ,EAAiB,QAAO,EAAE,MAAO,IAElD,EAAqC,CAAC,EAAG,GACzC,EAAmC,CAAC,MAAO,GAC3C,EAAS,GACX,CAAC,OAAQ,EAAqB,QAAO,EAAE,MAAO,IAElD,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAAqC,CAChD,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,ICvFd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,EACD,EAAa,EACnB,GAAiB,EAAG,UAIpB,OAFM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAY,GAAI,cAAa,EAAO,QACjC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAQ,EAAO,GACrB,EAAU,GAAK,EAAQ,EAEzB,GAAM,GAAS,EAAW,MAAM,EAAW,EAAE,MAAO,EAAE,OACtD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrC7C,AAsBO,GAAM,IAAwB,GAA8B,SAAC,EAAG,GACrE,GAAM,GAAO,EAAI,EACjB,MAAO,GAAO,IAEH,GACT,GAAiB,EAAA,kBAAmB,IAE3B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,IChCd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,EAAI,GACvD,GAAM,GAAY,EAClB,MAAI,OAAM,GACD,IAEA,EAAK,EAAI,EAAI,EAAU,QAIrB,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICjCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,YAyBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAEpB,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,IC5Cd,AA6FA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,0yDCvKjB,AAiBA,GAAM,IAAmD,GAEnD,GAA2C,CAC/C,MAAO,GACP,UAAW,GACX,mBAAoB,GACpB,sBAAuB,GACvB,MAAO,GACP,QAAS,GACT,6BAA8B,gBAI5B,EAAsB,GACxB,GAAS,GAAgB,cAGK,GAC9B,GAAI,CAAE,KAAgB,MACpB,GAAM,GAAS,GAAyB,GACxC,GAAI,IAAW,KACb,GAAS,GAAgB,MAEzB,gBAAQ,IAAI,0CAA2C,GAChD,KAGX,GAAM,GAAK,GAAS,GACpB,MAAI,GAAG,gBACL,OAAO,IAAS,GACT,GAAgB,IAGzB,GAAG,QAAQ,EAAG,YACd,EAAG,QAAQ,EAAG,cACd,EAAG,QAAQ,EAAG,OACd,EAAG,QAAQ,EAAG,QACd,EAAG,QAAQ,EAAG,qBACd,EAAG,QAAQ,EAAG,iBACd,EAAG,OAAO,EAAG,cACb,EAAG,OAAO,EAAG,WACb,EAAG,SAAS,EAAG,MAER,GAAS,IAGlB,YAAsB,GACpB,GAAI,MAAO,kBAAoB,aAAe,IAAiB,EAC7D,MAAO,IAAI,iBAAgB,IAAK,KAC3B,GAAI,MAAO,WAAa,YAC7B,MAAO,UAAS,cAAc,UAE9B,KAAM,IAAI,OAAM,0CAIpB,YAAkC,GAChC,GAAI,IAAiB,GAAK,IAAiB,EACzC,KAAM,IAAI,OAAM,0DAElB,GAAM,GAAS,GAAa,GAM5B,MAJA,GAAO,iBAAiB,mBAAoB,SAAC,GAC3C,EAAG,iBACH,MAAO,IAAS,IACf,IACC,IAAiB,EACX,EAAO,WAAW,QAAS,KAC3B,EAAO,WAAW,qBAAsB,IAG3C,EAAO,WAAW,SAAU,ICxFrC,AAoBA,GAAY,IAAZ,AAAA,UAAY,GAgBV,EAAA,EAAA,MAAA,GAAA,QAkBA,EAAA,EAAA,aAAA,GAAA,iBAlCU,IAAA,IAAa,KAqCzB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,SAAA,GAAA,aAJU,IAAA,IAAY,KAOxB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,yBAAA,GAAA,2BACA,EAAA,EAAA,mBAAA,GAAA,qBACA,EAAA,EAAA,mBAAA,GAAA,uBALU,IAAA,IAAmB,iBAqC3B,EAAc,GAChB,MAAO,CAAC,EAAS,eAIf,EAAoB,GACtB,MAAO,GAAa,cAWW,GAC/B,GAAM,GAAO,EAAA,KAAK,cAAc,GAC1B,EAAe,KAAK,KAAK,EAAO,GACtC,MAAO,GAAA,KAAK,oBAAoB,eA6B9B,EAAc,GAChB,MAAO,CACL,KAAK,IAAI,EAAG,KAAK,KAAK,EAAU,IAAK,KAAK,IAAI,EAAG,KAAK,KAAK,EAAO,iBAKlE,EAAc,GACV,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,GAAI,EAAI,cAsBb,EAA2B,GAE7B,GAAM,GAAQ,EAEV,EACA,EACA,EACA,EACA,EAEA,EACA,EAEA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAsB,EAAM,KAC5B,EAA0B,EAAM,KAChC,EAAgC,EAAM,QACtC,EAA4B,EAAM,QAClC,EAAqB,EAAM,IAC3B,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,EAAM,WAC7B,EAAmB,EAAM,OAEzB,GAAsB,EAAG,KACzB,EAA0B,EAAG,KAC7B,EAAgC,EAAG,KACnC,EAA4B,EAAM,KAClC,EAAqB,EAAG,KACxB,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,GAA6B,KAChD,EAA0B,eAC1B,KACJ,EAAmB,EAAG,OAExB,EAAwB,EAAG,KAEpB,CACL,oBAAmB,EACnB,wBAAuB,EACvB,8BAA6B,EAC7B,0BAAyB,EACzB,mBAAkB,EAClB,sBAAqB,EACrB,0BAAyB,EACzB,mBAAkB,EAClB,qBAAoB,EACpB,iBAAgB,GCzOpB,YAsBgC,EAA2B,GACzD,GAAM,GAAc,IACpB,MAAI,GAAA,MAAM,QAAQ,UAChB,GAAgB,GAEX,EAGT,YAAyB,GACvB,GAAM,GAAQ,EAAG,WACjB,GAAI,IAAU,EAAG,SACf,KAAM,IAAI,OAAM,gBAAkB,GAAqB,EAAI,IAK/D,GAAM,IAAc,QACd,GAAc,kBAEa,GAC/B,MAAI,KAAA,MAAM,QAAQ,iCAAmC,IAAQ,GACxD,GAAc,KAAK,IAAI,IAAQ,KAAK,IAAI,GAAO,gBAOlD,EAA2B,GAC7B,OAAQ,OACD,GAAG,SACN,MAAO,eACJ,GAAG,aACN,MAAO,mBACJ,GAAG,cACN,MAAO,oBACJ,GAAG,kBACN,MAAO,wBACJ,GAAG,8BACN,MAAO,oCACJ,GAAG,cACN,MAAO,oBACJ,GAAG,mBACN,MAAO,6BAEP,MAAO,sBAAsB,eAK/B,EAA2B,GAC7B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,aAAa,IAC1B,cAAgB,EAAgB,gDAIlC,EAA2B,GAC7B,GAAM,GAA4B,GAC9B,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,gBAC7B,wCAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAc,KACrD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAc,EAAG,kBAAoB,GAC7D,cAAQ,IAAI,EAAG,iBAAiB,IAC1B,GAAI,OAAM,oCAElB,MAAO,eAIL,EAA2B,GAC7B,GAAM,GAA8B,GAChC,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,kBAC7B,0CAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAgB,KACvD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAgB,EAAG,kBAAoB,GAC/D,SACI,EAAsB,EAAG,iBAAiB,IACxC,GAAI,OAAM,sCAElB,MAAO,GAGT,GAAM,IAAkB,2BACxB,YACI,EAAsB,GACxB,GAAM,GAAwB,GAAgB,KAAK,GACnD,GAAI,GAAyB,MAC3B,QAAQ,IAAI,wCAAwC,GACpD,QAAQ,IAAI,GACZ,OAWF,OARM,GAAa,CAAC,EAAsB,GAEpC,EAAc,EAAa,MAAM;GACjC,EAAM,EAAY,OAAO,WAAW,OAAS,EAC7C,EAAuB,EAAY,IACrC,SAAC,EAAM,GACH,MAAA,GAAA,KAAK,SAAU,GAAa,GAAG,WAAY,GAAO,IACtD,EAAgB,EACX,EAAI,EAAG,EAAI,EAAqB,OAAQ,IAC/C,EAAgB,KAAK,IAAI,EAAqB,GAAG,OAAQ,GAG3D,GAAM,GAAmB,EAAqB,MAAM,EAAG,EAAa,GAC9D,EAAY,EAAqB,MAAM,EAAa,EAAG,GACvD,EAAkB,EAAqB,MAAM,GAEnD,QAAQ,IAAI,EAAiB,KAAK;IAClC,QAAQ,IAAI,EAAc,MAAM;GAAM,IACtC,QAAQ,IACJ,MAAM,EAAA,KAAK,SAAS,EAAU,GAAI,GAClC,iEACJ,QAAQ,IAAI,EAAgB,KAAK;gBAGL,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGR,EAA2B,GAErD,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,KAClC,EAAG,oBAAoB,EAAS,EAAG,eAAiB,GACtD,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,2DAKhB,EAA2B,GAE7B,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,KACtC,EAAG,oBAAoB,EAAS,EAAG,mBAAqB,GAC1D,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,iDAKhB,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,EAAM,EAAG,eACxD,cAIL,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,KAC9D,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,EAAM,EAAG,eACvD,gBAIP,MAAI,GAAA,MAAM,UAAU,mBAAqB,EAChC,EAEF,cAGqB,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGA,EAAe,GACjD,GAAM,GAAiB,EAAA,MAAM,UAAU,0BACvC,GAAK,GAAS,GAAO,GAAU,GAC7B,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IACrC,KAAM,IAAI,OAAM,0BAA4B,EAAY,gBAE1D,GAAK,EAAQ,GAAoB,EAAS,GACxC,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IAC/B,EAAM,IAAI,EAAc,IAAI,EAAc,IAChD,KAAM,IAAI,OACN,0BAA4B,EAC5B,qDAAuD,EAAM,kBAInC,GAChC,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,qBAAqB,kDAIpC,EAA2B,EAAuB,EAClD,EAAqB,EAA6B,EAClD,GACF,GAAM,GAAM,EAAG,kBAAkB,EAAS,GAC1C,MAAI,KAAQ,GAGH,GAET,IAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GACI,EACA,WAAM,MAAA,GAAG,oBACL,EAAK,EAAqB,EAAG,MAAO,GAAO,EAC3C,KACR,GAAa,EAAI,WAAM,MAAA,GAAG,wBAAwB,KAC3C,gBAIL,EAA2B,EAAuB,GACpD,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,iBAInD,EAA2B,GAC7B,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAInD,EAA2B,EAC3B,GACF,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,mBAAmB,EAAS,IACzC,YAAc,EAAc,yCAI9B,EAA2B,EAC3B,GACF,MAAO,GAAG,mBAAmB,EAAS,eAIpC,EAA2B,EAC3B,EAA8C,GAChD,GAAa,EAAI,WAAM,MAAA,IAAgB,EAAI,EAAS,KACpD,GAAa,EAAI,WAAM,MAAA,GAAG,UAAU,EAAwB,iBAGtB,GACtC,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QAC1D,GAAa,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,UACpE,GAAa,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,sBAIjE,EAA2B,EAC3B,GACF,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,iBAIpE,EAA2B,GAC7B,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,KAAM,iBAGjC,GAClC,GAAM,GAAS,EAAG,uBAAuB,EAAG,aAC5C,GAAI,IAAW,EAAG,qBAChB,KAAM,IAAI,OACN,8BAAgC,GAA2B,EAAI,gBAKnE,EAA2B,GAC7B,OAAQ,OACD,GAAG,kCACN,MAAO,wCACJ,GAAG,0CACN,MAAO,gDACJ,GAAG,kCACN,MAAO,wCACJ,GAAG,wBACN,MAAO,kCAEP,MAAO,iBAAiB,GAI9B,YACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAa,EAAI,WAAM,MAAA,OAC/C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,GAElB,MAAO,GAGT,YAA6B,EAA2B,GACtD,GAAM,GAAiB,EAAG,iCAAmC,EACvD,EAAgB,EAAc,EAAG,SACvC,GAAI,EAAgB,EAAG,UAAY,EAAgB,GACjD,GAAM,GAAmB,2BAA2B,EAAc,IAClE,KAAM,IAAI,OAAM,0BAA0B,EAAgB,kBAIlC,EAAiB,GAAA,MAAA,KAAA,QAAA,GAAA,GACpC,EAAA,KAAK,cAAc,EAAM,MAAM,EAAG,EAAM,OAAS,gBAG9B,GAC1B,GAAI,EAAM,SAAW,EACnB,KAAM,OAAM,wDAGd,MAAO,CACL,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EAAG,EAAM,EAAM,OAAS,gBAI5C,GAC3B,GAAI,GAAsC,CAAC,EAAG,EAAG,GAC3C,EAAW,EAAM,SAAW,GAAM,EAAM,SAAW,GAAK,EAAM,KAAO,EAC3E,MAAK,IACH,GACI,CAAC,GAAY,IAAM,OAAK,GAAY,KAEnC,cAIL,EAAoB,SAAA,AAAA,IAAA,QAAA,GAAA,IACtB,GAAI,GAAa,EAAA,MAAM,UAAU,0BAsBjC,GArBA,AAAI,GACF,GAAa,EAAa,EAO1B,EAAW,EAAS,IAChB,SAAC,EAAG,GAAM,MAAA,IAAK,EAAS,OAAS,EAC7B,EAAA,KAAK,kBAAkB,EAAS,IAChC,EAAS,KAIjB,AAAI,EAAS,SAAW,GACtB,GAAW,CAAC,EAAG,EAAS,MAKxB,EAAS,SAAW,GACtB,GAAM,GAAgB,EAAA,KAAK,aAAa,GACxC,EAAW,EAAc,SAG3B,GAAI,GAAO,EAAA,KAAK,cAAc,GAC9B,GAAI,EAAS,QAAU,GAAK,GAAQ,EAClC,MAAO,CAAC,EAAG,GACN,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,IAAM,EACjB,MAAO,GACF,GACH,EAAS,SAAW,GAAK,EAAS,GAAK,EAAS,IAAM,GACtD,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAI,EAAS,IACvC,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,IAAM,EAC/B,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,IACvC,GACH,EAAS,SAAW,GACpB,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,GAC3C,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAK,EAAS,GAAI,EAAS,IACrD,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,EAC7C,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,GAAK,EAAS,IAE1D,GAAI,GAOF,GAAM,GAAW,GAAY,GACzB,EAAO,EAAG,EAAO,EACrB,MAAI,GAAS,QACX,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,IAET,EAAO,EAAY,GAAO,GAAM,GAAO,GAChC,EAAA,KAAK,oBAAoB,GAAM,IAAI,SAAA,GAAK,MAAA,GAAI,IAErD,MAAO,GAAA,KAAK,oBAAoB,GAIpC,YAAgB,GACd,MAAO,GAAI,IAAM,cAOW,EAAkB,GAI9C,GAHA,EAAS,EAAO,MAAM,IACtB,EAAS,EAAO,MAAM,IAElB,EAAA,KAAK,YAAY,EAAQ,GAC3B,MAAO,GAGT,GAAI,CAAC,EAAO,QAAU,CAAC,EAAO,OAC5B,MAAO,GAGT,GAAI,EAAO,KAAO,GAAK,EAAO,KAAO,GAAK,EAAO,KAAO,GACpD,EAAO,KAAO,EAChB,MAAO,GAGT,GAAI,EAAO,SAAW,EAAO,QAC3B,GAAM,GAAa,EAAO,MAAM,IAAI,GAC9B,EAAa,EAAO,MAAM,IAAI,GACpC,GAAI,IAAe,EACjB,MAAO,GAGT,GAAI,GAAO,IAAe,GAAO,IAC5B,GAAO,KAAO,GAAK,EAAO,KAAO,GACpC,MAAO,GAGX,MAAO,GAAO,KAAO,EAAO,IAAM,GAAO,EAAO,KAAO,GAAO,EAAO,IAMvE,GAAI,IACA,eAEmC,GACrC,GAAI,IAAoB,MACtB,GAAM,GAAK,GAAgB,GAC3B,GAAmB,EAAG,aAAa,EAAG,kBAExC,MAAO,kBAIP,GAAmB,mBAGnB,GAAyB,iBAGY,GACrC,GAAI,IAA0B,MAC5B,GAAM,GAAK,GAAgB,GAC3B,GAAyB,EAAG,aAAa,EAAG,yBAG9C,MAAO,MAAK,IAAI,GAAI,gBAG4B,GAEhD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAI,GACE,EAAK,GAAgB,GAE3B,MAAI,IAAa,EAAI,oCACjB,IAAiB,EACnB,EAAoB,EACf,AAAI,GAAa,EAAI,4BAC1B,EAAoB,EAEpB,EAAoB,EAEf,cAGoB,EAA2B,GACtD,GAAM,GAAM,EAAG,aAAa,GAC5B,MAAO,IAAO,iBAGsB,GACpC,IACE,GAAM,GAAK,GAAgB,GAC3B,GAAI,GAAM,KACR,MAAO,SAEF,GACP,eAAQ,IAAI,qCAAsC,GAC3C,GAET,MAAO,eAG0C,GAEjD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,WAGL,CAAC,GAAa,EAAI,0BACpB,MAAO,GAIX,GAAM,GAAwB,GAAuC,GACrE,MAAO,eAYqC,GAC5C,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,GAET,GAAI,CAAC,GAAa,EAAI,4BACpB,MAAO,QAGT,GAAI,GAAa,EAAI,0BACnB,MAAO,IAAuC,GAGhD,GAAM,GAA0B,8BAChC,GAAI,GAAa,EAAI,IACnB,GAAM,GACF,EAAG,aAAa,GACpB,MAAO,IACH,EAAI,GAGV,MAAO,GAGT,GAAM,GAAwB,GAAuC,GACrE,MAAO,GAGT,YAAgD,GAE9C,GAAM,GAAY,GAAiB,GAE7B,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,oBAAqB,EAAO,EAAQ,EAChE,EAAU,mBAAoB,EAAU,iBAAkB,MAE9D,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,EAGT,YAEI,EAA2B,GAC7B,GAAM,GAAY,GAAiB,EAAI,GACjC,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,wBAAyB,EAAO,EAAQ,EACpE,EAAU,mBAAoB,EAAU,qBAAsB,MAElE,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,cAG2B,GAClC,GAAI,IAAiB,EACnB,MAAO,GAET,GAAM,GAAK,GAAgB,GAGrB,EAAa,EAAW,WAAa,KAC3C,MAAO,eAIL,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MAAG,GAAM,smCC7qBzB,AAqBA,GAAM,IAAM,EAAA,MASZ,GAAI,aAAa,YAAa,WAAM,MAAA,IAAI,UAAU,iBAAmB,IAGrE,GAAI,aAAa,gBAAiB,WAChC,MAAI,IAAsB,GACjB,EACE,GAAsB,GACxB,EAEF,IAIT,GAAI,aAAa,iCAAkC,WAAM,MAAA,KAEzD,GAAI,aACA,yBAA0B,WAAM,MAAA,IAAI,IAAI,mBAAqB,IAGjE,GAAI,aAAa,oBAAqB,WAAM,MAAA,KAG5C,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aAAa,aAAc,WAAM,MAAA,IAAI,QAAQ,eAGjD,GAAI,aAAa,2BAA4B,WAAM,MAAA,IAAI,QAAQ,gBAG/D,GAAI,aAAa,kBAAmB,WAAM,MAAA,IAAI,QAAQ,gBAItD,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aACA,+BAAgC,WAAM,MAAA,IAAI,QAAQ,gBAGtD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aAAa,sBAAuB,WAAM,MAAA,IAAI,QAAQ,gBAG1D,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aACA,yBACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAG/C,GAAI,aACA,+BACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAU/C,GAAI,aAAa,+CAAgD,WAC/D,GAAM,GAAe,GAAI,UAAU,iBAEnC,MAAI,KAAiB,EACZ,EAEF,GAAkC,KAO3C,GAAI,aACA,gDACA,WAAM,MAAA,IAAI,UAAU,gDAAkD,GAClE,CAAC,EAAA,YAAY,aAKrB,GAAI,aACA,+BACA,WAAM,MAAA,IAAmC,GAAI,UAAU,oBAM3D,GAAI,aAAa,+BAAgC,WAC/C,MAAO,IAAI,QAAQ,4BACf,GACA,GAAI,QAAQ,kCAOlB,GAAI,aACA,+BACA,WAAM,MAAA,IAA8B,GAAI,UAAU,oBAGtD,GAAI,aACA,0BACA,WAAM,MAAA,IAAoB,GAAI,UAAU,oBAK5C,GAAI,aAAa,4BAA6B,WAK5C,GAAM,GAAc,GAAI,QAAQ,gCAChC,MAAO,GAAc,EAAI,IAU3B,GAAI,aACA,iCACA,WACE,MAAO,IAET,SAAA,GACE,GAAI,EAAY,GAAK,IAAc,GACjC,KAAM,IAAI,OACN,+DACA,mCAAkC,EAAS,QC1LvD,AAiBO,YAAuB,GAC1B,KAAM,GAAe,GAAI,cAAa,EAAK,QAC3C,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC/B,EAAa,GAAK,KAAK,IAAI,EAAK,IAEpC,MAAO,GCtBX,AAoBO,YAAsC,GACzC,MAAO,CAAC,EAAQ,EAAQ,EAAO,EAAO,KAClC,KAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAC3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAQ,EAAO,OACf,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAC7D,GAAI,EAAe,OAAS,EAAe,SAAW,EAClD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAI5D,QAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACjC,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GACrC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GACtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAC5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,IAG5C,MAAO,CAAC,EAAQ,IClDxB,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAG,IAAM,EAAI,GCnBnE,AAoBO,YAA+B,GAClC,MAAO,CAAC,EAAQ,EAAO,KACnB,KAAM,GAAY,EAAA,KAAK,uBAAuB,EAAO,EAAO,QAC5D,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAU,GAAK,EAAG,EAAO,GAAI,GAEjC,MAAO,IC1Bf,AAmBO,KAAM,IAAW,GAAsB,AAAC,GAAO,KAAK,KAAK,ICnBhE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAiBO,YAAiB,EAAO,EAAY,EAAU,GACjD,KAAM,GAAO,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACnE,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAC/B,KAAM,GAAS,EAAI,EACnB,GAAI,GAAM,EAAM,GAChB,OAAS,GAAI,EAAG,EAAI,EAAY,EAAE,GAC9B,KAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACR,GAAM,GAGd,EAAK,GAAK,EAEd,MAAO,GC9BX,AAmBO,KAAM,IAAe,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBvF,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,EAAI,KAAK,KAAK,ICnBrE,AAkBO,YAAmB,EAAM,EAAO,EAAM,EAAO,GAChD,KAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GACrC,GAAI,GACA,KAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAElD,KAAM,GAAU,EAAA,KAAK,uBAAuB,EAAO,GACnD,OAAS,GAAI,EAAG,EAAI,EAAQ,EAAE,GAC1B,KAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,CAAC,EAAK,IAAM,EAAM,EAAM,IACvC,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEtB,MAAO,GCnCX,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBlF,AAiBO,YAAuB,EAAO,EAAQ,EAAO,EAAM,GACtD,KAAM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GACjC,EAAS,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACrE,OAAS,GAAI,EAAG,EAAI,EAAO,EAAE,GACzB,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAEhC,EAAS,GAAI,OAAM,EAAI,QAC7B,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAO,GAAK,EAAI,EAAK,IAEzB,KAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE7B,MAAO,GCjCX,AAiBO,YAAoB,EAAQ,EAAM,EAAO,GAE5C,KAAM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAwDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GAC/B,OAAS,GAAI,EAAG,EAAI,EAAO,IACvB,EAAS,IAAM,EAAM,GAEzB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACtC,EAAS,IAAM,EAAM,GAIzB,KAAM,GAAiB,GAGjB,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAAgB,GAChB,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EACxD,OAAS,GAAI,EAAG,EAAI,EAAM,GAAQ,KAE9B,GAAI,GACJ,GAAI,EAEA,EAAU,EAAO,GAAG,gBAGpB,KAAM,GAAa,GACnB,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG9C,EAAU,EAAW,KAAK,KAG9B,GAAI,EAAe,KAAa,OAC5B,EAAQ,GAAK,EAAe,QAG5B,KAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAM3B,KAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,KAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,CAAC,EAAoB,KACvC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAM9E,KAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAC7B,CACH,aAAc,EAAa,OAC3B,cACA,WC9IR,AAuBE,GAAA,IAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GCpCF,AAmBA,GAAA,IAAA,WAKE,WAAY,EAAuB,GAHnC,KAAA,YAAwB,GAItB,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,UAAU,EAAQ,SAAS,EAAQ,oBAInD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;yBAEZ,EAAS;;;MAKlC,MAAA,MClDA,AAmBA,GAAA,IAAA,WAOE,WAAY,EAAuB,GALnC,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,SAAS,EAAQ,SAAS,EAAQ,oBAIlD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;wBAEb,EAAS;;;MAKjC,MAAA,MCpDA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAqC,EACrC,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAC9B,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,KAAK,YAAc,CAAC,EAAW,GAC/B,GAAM,GAAU,IAAO,MAAS,IAAM,IAChC,EAAe,EACjB,gBACA,+CAEJ,KAAK,SAAW;;;;;kCAKc,EAAU;;;;;8BAKd,EAAU;wBAChB,EAAY;;0BAEV,EAAM;;;;;;;MAShC,MAAA,MC5DA,YAiB+B,EAAc,GAC3C,MAAO,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAG,GAAI,IAAI,gBAG/C,EAAc,GACxC,MAAI,KAAS,EACJ,CAAC,GAEH,GAAe,EAAM,eAGE,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,KAIT,OADI,GAAS,GACJ,EAAI,EAAG,EAAI,EAAM,IACxB,GAAU,EAAK,GACf,AAAI,EAAI,EAAO,GACb,IAAU,KAGd,MAAO,GCxCT,cAgCE,GAAI,GACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAU,kBACV,EAAY,KACZ,EAAY,MACZ,EAAY,KACZ,EAAY,UACZ,EAAS,cACT,EAAe,wBAMf,EAAmB;;;;;;;;;;;MAcnB,EAAmB,GACnB,EAAc;;;;;;;;;OAWd,GAAU,GACV,EAAY,YACZ,EAAY,UACZ,EAAY,UACZ,EAAY,YACZ,EAAS,eACT,EAAe,GAEf,EAAmB;;;;;;;;MASnB,EAAmB;;;;;;;;;MAUnB,EAAc;;;;;;;;OAWT,CACL,QAAO,EACP,UAAS,EACT,UAAS,EACT,UAAS,EACT,UAAS,EACT,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,iBAAgB,EAChB,YAAW,GClIf,YAyBI,EAAkB,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,SACrC,GAAM,GAAU,EAAA,KAAK,eAAe,GACpC,MAAO,GACF,IAAI,SAAC,EAAQ,GACZ,GAAM,GAAQ,OAAO,EAAO,GAAE,MAAM,EAAK,MAAM,EACzC,EAAQ,IAAM,EAAQ,OAAS,EACjC,OAAO,EAAO,EAAI,GAAE,MAAM,EAAK,MAAM,EAAO,GAAE,MAAM,EACpD,YAAY,EAAO,GAAE,MAAM,EAC/B,MAAU,GAAK,KAAK,EAAK,MAE1B,KAAK,gBA+CuB,GACjC,GAAM,GAAU,EAAA,KAAK,eAAe,GAAO,IAAI,SAAA,GAAK,MAAA,GAAE,aAEtD,MAAO;;wBAEe,EAAQ,GAAE,iBAAiB,EAAQ,GAAE;;EAKtD,GAAM,IAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EC5FpC,AAkBO,GAAA,IAAA,EAAA,aAAA,6BAkBH,EAAyB,EAAwB,EACjD,GACF,GAAM,GAA2B,GACjC,EAAW,QAAQ,SAAA,GACjB,GAAM,GAAO,EAAA,KAAK,cAAc,EAAE,UAAU,cAG5C,AAAI,EAAE,UAAU,UACd,EAAe,KACX,iBAAiB,EAAE,KAAO,GAAO,EAAI,IAAI,EAAI,IAAM,IAAE,KAEzD,GAAe,KAAK,qBAAqB,EAAE,KAAI,KAC/C,EAAe,KAAK,qBAAqB,EAAE,KAAI,QAGnD,GAAM,GAAqB,EAAe,KAAK;GAEzC,EACF,EACK,IAAI,SAAA,GAAK,MAAA,IAAwB,EAAG,EAAa,KACjD,KAAK;GACR,EAAc,EAAY,SAC1B,EAAO,KACP,EAA4B,GAA6B,GAC3D,EACA,EACA,EAAe,GAAgB,GAEnC,AAAI,EAAY,SACd,GACI,GAA+B,EAAY,aAAc,GAC7D,EAA+B,GAA8B,IAE7D,GACI,GAAyB,EAAY,aAAc,GACvD,EAA+B,GAA2B,IAG5D,AAAI,GACF,IAAgB,IAGlB,GAAM,GAAS,CACb,EAAc,EAA2B,EACzC,EAAoB,EAAuB,EAAsB,GACjE,KAAK;GACP,MAAO,GAGT,YAA8B,GAC5B,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAiB,OACrB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,WAEpB,KAAM,IAAI,OACH,EAAM,OAAM,2CAKzB,YAAoC,GAClC,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAuB,OAC3B,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,WAE1B,MAAO,IAAmB,IAIhC,YACI,EAAmB,EACnB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAM,GACV,AAAI,EACF,GAAO,GAA2B,GAElC,GAAO,GAAqB,GAG9B,GAAM,GAAU,EAAO,UAAU,aAC3B,EAAW,EAAa,aAC9B,MAAI,GAAQ,QAAU,EAAS,QAC7B,CAAI,EACF,GAAO,GAA+B,EAAQ,GAE9C,GAAO,GAAyB,EAAQ,IAGrC,EAGT,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAwB,EAAsB,OAClD,GACH,MAAO,IAAwB,EAA8B,OAC1D,GACH,MAAO,IACH,EAAsC,WAE1C,MAAO,IAAwB,EAAU,IAI/C,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAkB,EAAsB,OAC5C,GACH,MAAO,IAAkB,EAA8B,OACpD,GACH,MAAO,IACH,EAAsC,OACvC,GACH,MAAO,IACH,EAA8C,OAC/C,GACH,MAAO,IACH,EAAsD,OACvD,GACH,MAAO,IACH,EACA,WAEJ,KAAM,IAAI,OACH,EAAS,OAAM,4CAI5B,YAAsC,GACpC,MAAO;;eAEM,EAAK,UAAS;;IAK7B,YAAoC,GAClC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAuC,GACrC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAyB,GACvB,GAAM,GAAmB,EAAK,QAAO;;;;MAIjC,EAAK,UAAS;MACd,EAAK,aAAY;;;;;;;;;;;;;;;;;;;;;;;MAuBjB,EAAK,iBAAgB;MACrB,EAAK,iBAAgB;MACrB,EAAK,YAAW;;;;;;;;;;;;;;;;;;;;;;;;;MAyBhB,GAAiB;MACjB,GAAiB;MACjB,GAAiB;IAGrB,MAAO,GAGT,GAAM,IAAoB;;;;;;;;;;;;EAcpB,GAAoB;;;;;;;;EAUpB,GAAoB;;;;;;;;;EAWpB,GAAuB;;;;;;;;;;;EAa7B,cACE,MAAO;;;;IAOT,YACI,EAAiB,GACnB,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,MAAI,GAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAKjD,EAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAK9C;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;IAKlD,YACI,EAAiB,GACnB,MAAI,GAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIvC,EAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIpC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;4BACnC,EAAS,GAAE;;IAKvC,YACI,EAAiC,GACnC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAqB,KAAK,KAAK,EAAM,GAAK,GAC1C,EAAgB,EAAqB,KAAK,KAAK,EAAM,GAAK,GAEhE,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;wBAE1B,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAiC,GACnC,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EAAiB,GAWnB,OAVM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEnD,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EACF,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EAAiB,EACjB,EAAU,GACV,EAAS,UAEJ,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAkB,EAAM,EAAM,OAAS,EAAI,GAC3C,EAAU;aACD,EAAC,cAAc,EAAc;kBACxB,EAAC,MAAM,EAAc;MAC/B,EACJ,EAAS,IAAI,EAAC,KAAO,EAGvB,MAAO;UACC,EAAM,OAAM;;oCAEc,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;QAE1C,EAAO;;wBAES,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;mBAE3B,EAAM,OAAM,IAAI,EAAM;;IAKzC,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,MAAO,GAE3B,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,MAAO,GAEjC,MAAO;;kDAEyC,EAAS,GAAE;+BAC9B,EAAS,GAAE;;iCAET,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,KAAM,MAAO,GAEvC,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EAAyB,GAC3B,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,GAAI,EAAA,KAAK,YAAY,EAAO,GAC1B,MAAO;;8CAEmC,EAAe,GAAE,KACvD,EAAe,GAAE;;MAMvB,GAAM,GAAqB,KAAK,KAAK,EAAM,GAAK,GAWhD,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;;iCAE1C,EAAe,GAAE;6BACrB,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAyB,GAC3B,MAAI,GAAA,KAAK,YAAY,EAAO,GACnB;;0CAE+B,EAAS,GAAE,KAAK,EAAS,GAAE;;MAI/D,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKxC,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKrC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;wBACpB,EAAM,GAAE;4BACJ,EAAM,GAAE;;;IAMpC,YAAkC,GAChC,MAAO,SAAS,EAGlB,YAAgC,GAC9B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAO,KACb,MAAO;WACE,EAAQ;eACJ,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAA0B,GACxB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACzE,GAAI,EAAU,UAAU,UACtB,MAAO,SAAS,EAAQ,cAAc,EAAO,KAEzC,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAChB,GAAI,IAAY,GAAK,IAAY,EAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAK9B,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,KAAK,EAAM;6BAC1B,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAe,GAAE,KAAK,EAAe,GAAE;eAClC,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEzE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;UACZ,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAQ,EAAS,GACjB,EAAQ,EAAS,GAEvB,GAAI,IAAU,GAAK,IAAU,EAC3B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAI,KAAU,EACL;cACG,EAAQ;6CACuB,EAAM,cAAc,EAAK;+BACvC,EAAO;;MAIhC,IAAU,EACL;cACG,EAAQ;wCACkB,EAAM,cAAc,EAAK;+BAClC,EAAO;;MAI7B;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,aAAa,EAAM;6BAClC,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAE/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAO,KACb,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,GAC9C,MAAO;aACE,EAAQ;qDACgC,EAAO,OAAO,EAAO;;iBAEzD,EAAK,UAAS,IAAI,EAAO;;MAKxC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAe,KAAK,KAAK,EAAM,GAAK,GAE1C,MAAO;WACE,EAAQ;iCACc,EAAY,KAAK,EAAe,GAAE,KAC7D,EAAe,GAAE;eACR,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAErC,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,IAC9C,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACzB,MAAO;YACC,EAAQ;mDAC+B,EAAO,OAAO,EAAO;6BAC3C,EAAO;;IAK5B,GAAA,GAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,OACvB,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;qDAC+B,EAAM,GAAE;UACnD,GAAkB,GAAU;;MAKpC,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAS,GAAyB,GACxC,MAAI,KAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;4CACtB,EAAO;6BACtB,EAAO;;IAI9B,IAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;uCAC3B,EAAO;6BACjB,EAAO;;IAK3B;UACC,EAAQ;;wBAEM,EAAM,GAAE,YAAY,EAAM;2BACvB,EAAO,KAAK,EAAO;2BACnB,EAAO;;EAKlC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEzD,GAAI,EAAM,KAAO,GACf,GAAM,GAAgB,EAAM,MAAM,GAC5B,EAAW,CAAC,EAAG,GACf,EAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,IAAK,MAAO,OAC5B,MAAO;UACD,GAA2B,GAAa;eACnC,EAAQ;mBACJ,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAM,GAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpC,EAAgB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAO,KAAK,EAAO,KAAK,EAAa,KAAK,EAAY;eACjD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAAK,EAAM,GAC3B,EAAU,EAAM,GAEhB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,SAC9B,MAAO;UACD,GAAqB,GAAa;gBAC5B,EAAQ;mBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO;UAC1C,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAa,EAAU,UAAU,WACvC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;gBACK,EAAQ;;oDAE4B,EAAO;;4BAE/B,EAAO,OAAO,EAAO;iCAChB,EAAO;;QAKtC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;YACC,EAAQ;8CAC0B,EAAM,GAAE;;qDAED,EAAO,OAAO,EAAO;6BAC7C,EAAO;;IAKlC,GAAM,GAAS,GAAyB,GACxC,MAAO;cACK,EAAQ;;4BAEM,EAAO,YAAY,EAAO,cAAc,EAAM;+BAC3C,EAAO,KAAK,EAAO;+BACnB,EAAO;;IAKtC,YAA4B,GAe1B,OAdM,GAAQ,EAAU,UAAU,aAC5B,EAAO,EAAM,OACb,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC7C,EAAgB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC3D,EAAS,0BACT,EAAQ,OAAO,EAAa,kBAAkB,EAAY,eACrD,EAAI,EAAG,EAAI,EAAO,EAAG,IAC5B,EAAS,QAAQ,EAAC,KAAO,EACzB,GAAiB,EAAM,EAAO,EAAI,GAClC,EAAQ,IAAI,EAAC,MAAM,EAAa,MAAQ,EAE1C,GAAM,GAAO,KACb,MAAO;WACE,EAAQ,IAAI,EAAM;oBACT,EAAK;2BACE,EAAO;kCACA,EAAO;qDACY,EAAO,KAAK,EAAO;eACzD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,UACvC,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO,KAAK,EAAO;UACtD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;uBAIC,EAAO,KAAK,EAAO;;0BAEhB,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;gCAEU,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGvC,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO;oBAChC,EAAO;6BACE,EAAO,KAAK,EAAO,aAAa,EAAM;6BACtC,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,UACjD,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;iBAGL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;UAElD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;gCAGU,EAAO,KAAK,EAAO,KAAK,EAAO;;0BAErC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAKpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGxB,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAM;6BACpB,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEnE,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,SAAU,UAC3D,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;;iBAEL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAM,GAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAE3B,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;mBAGzC,EAAO;UAChB,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;0BAGlC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBACzC,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE;iBACnB,EAAM,GAAE;;;yBAGA,EAAO,OAAO,EAAO;+BACf,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;;0BAGM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAO,eAAe,EAAM;6BAC1C,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAA2B,GACzB,GAAM,GAAU,EAAU,KACpB,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAEtD,MAAI,GAAS,EACJ,UAAU,EAAO,IAEnB;0BACiB,EAAM;;iBAEf,EAAO;;;IAMxB,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAEpC,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAE7C,EAAO,GAAkB,GACzB,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,AAAI,EAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAGpC,GAAI,GAAS,sBACP,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAChD,EAAgB,IAAW,EAC3B,EAAU,EAAA,KAAK,cAAc,EAAa,cAC1C,EAAiB,IAAY,EAEnC,GAAI,IAAW,GAAK,CAAC,GAAiB,CAAC,EACrC,EAAS;;cAGA,GAAiB,CAAC,EAC3B,AAAI,IAAY,EACd,EAAS;;QAIT,EAAS;;gBAIF,EAAc,QACvB,GAAM,GAAO,EAAS,EAChB,EAAO,EAAS,EAEtB,AAAI,EAAc,QAAQ,GAAQ,IAAM,EAAc,QAAQ,GAAQ,GACpE,EAAS,8BACJ,AAAI,EAAc,QAAQ,GAAQ,GACvC,EAAS,2EAEJ,AAAI,EAAc,QAAQ,GAAQ,IACvC,GAAS,gDAIb,MAAO;WACE,EAAQ;QACX,EAAI;QACJ,EAAa;8BACS,EAAc,IAAI,EAAqB;QAC7D,EAAM;;IAKd,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAc,EAAa,SAC3B,EAAa,EAAU,UAAU,SACjC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAE1C,GAAI,CAAC,EAAU,UAAU,WAAa,IAAW,GAC7C,EAAU,UAAU,YAAc,MAClC,EAAA,KAAK,YAAY,EAAY,GAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAKpC,GAAM,GAAO,GAAkB,GACzB,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAC7C,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,MAAI,GAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAG7B;YACG,EAAQ;QACZ,EAAI;QACJ,EAAa;kBACH,EAAc,IAAI,EAAqB;;gBAKvB,GAChC,GAAI,GAAQ,EACV,MAAO,MACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QAEP,KAAM,OAAM,gBAAgB,EAAI,yBAKpC,YACI,EAAmB,GAErB,GAAM,GAA0B,KAAK,MAAM,KAAK,UAAU,IAC1D,SAAa,UAAU,aAAe,EAC/B,EAGT,YAA2B,EAAkB,GAC3C,MAAO,GAAS,IAAI,SAAA,GAAK,MAAA,GAAO,KAAI,KAAK,MCt2C3C,AAuBA,GAAA,IAAA,WAOE,WACI,EAAiB,EAAoB,EACrC,GARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,EAAA,KAAK,OACD,EAAM,OAAS,EACf,WAAM,MAAA,aACF,GAAG,OAAO,GAAG,cACb,EAAG,MAAM,IAAE,6CACnB,GAAM,GAAS,EAAM,EAAM,OAAS,GAC9B,EAAU,KAAK,KAAK,EAAS,GACnC,KAAK,YAAc,EAAM,MAAM,EAAG,IAClC,AAAI,EAAU,GACZ,KAAK,YAAY,KAAK,GAExB,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,GAAM,GAAW,KAAK,YAChB,EAAO,EAAS,OAChB,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAEjC,EACA,EACJ,GAAI,IAAY,GACd,EAAa,EAAO,EACpB,GAAM,GAAiB,GAAkB,GACzC,EAAiB;UACb,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE,QAEtB,GAAa,EACb,EAAiB;UACb,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE,IAExB,GAAM,GAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACnD,EAAY,IAAM,EAAS,EAAa,GACxC,EAAc,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IACzC,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAE/C,EAAU,IAAO,MAAS,cAAgB,WAC1C,EAAoB,EAAY,GAAK;sDACO,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM,OAE7D,EAAa;0BACG,EAAW,OAAM;uCACJ,EAAW,OAAM;uCACjB,EAAW,OAAM;qDACH,EAAW,OAAM,UAE5D,EAAgC,EAAY,GAAK;qCACtB,EAAY,OAAM;4CACX,EAAS,OAAM;iDACV,EAAS,MAAM,IAAI,OAAM;SAGtE,KAAK,SAAW;0BACM,EAAY,OAAM;iCACX,EAAS,OAAM;sCACV,EAAS,MAAM,IAAI,OAAM;;QAEvD,EAA6B;;UAE3B,EAAK;4BACa,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;4BAC5C,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;UAC9D,EAAc;yCACiB,EAAS,eAAe,EAAS;sBACpD,EAAS,eAAe,EAAS,OAAO,EAAU;;;2BAG7C,EAAU;;8BAEP,EAAU;;YAE5B,EAAiB;6BACA,EAAU;;;mBAGpB,EAAM;;;;;;;;;;;MAazB,MAAA,MCjJA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAe,GAE1C,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;0CACT,EAAa;;;;;;;;;;;;;;gCAcvB,EAAqB;oBACjC,EAAc;gDACc,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;mBACnC,EAAa;kDACkB,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;;;;;;;MAevD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAc,EAAe,GAExD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAa;;;;;;;;;;;;;;;;;gCAiBvB,EAAoB;oBAChC,EAAa;gDACe,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;;;;;;;;MAgBzD,MAAA,MC3KA,AAwBO,GAAM,IAAmB,CAC9B,KAAM,wCACN,KAAM,yCAGR,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,QAAS,QAAS,QAAS,SAK1C,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GAEnE,KAAK,SAAW;;;UAGV,EAAE;;;;;;;;;;MAYZ,MAAA,MCpDA,AAqBA,GAAM,IAAoB;;;EAKb,GAAM,gBACN,GAAM,gBACN,GAAM,gBAMN,GAAU;;;;;;;;;;EAYV,GAAM;;;;;;;;;EAYN,GAAQ,wBAER,GAAY,wBAEZ,GAAO,uBAEP,GAAa,wBAEb,GAAU,uBAEV,GAAgB,wBAEhB,GAAc,sCAEd,GAAa,sCAEb,GAAM,GAAoB;;EAG1B,GAAM,GAAoB;;EAG1B,GAAM;qBAGN,GAAU,yCAEV,GAAQ,+BAErB,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;MAUZ,MAAA,MC1GA,AAuBA,GAAM,IAAoB;;;;;EAOb,GAAU;;;;;;;;;;;;;;;;;;;;;EAuBV,GAAM;;;;;;;;;;;;;;IAef,GAAoB;;EAIX,GAAQ;;;EAKR,GAAU;;;EAKV,GAAQ;;EAIR,GAAY;;EAIZ,GAAO;;EAIP,GAAa;;EAIb,GAAU;;EAIV,GAAgB;;EAIhB,GAAc;;;;EAMd,GAAa;;;;;EAOb,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIxB,GAAA,WAQE,WACI,EAAY,EAAkB,EAC9B,GAAA,AAAA,IAAA,QAAA,GAAA,IATJ,KAAA,cAAgB,CAAC,IAAK,KAGtB,KAAA,qBAAuB,GACvB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,GAAM,GAAO,KAAK,YAAY,OAC1B,EAAyB,GAC7B,GAAI,EACF,GAAI,IAAS,GAAK,EAAA,KAAK,cAAc,KAAK,eAAiB,EACzD,EAAyB;;;;eAMzB,GAAM,GAAQ,GAAkB,GAIhC,GAHA,EAAyB;YACrB,EAAK;UAEL,IAAS,EACX,GAA0B;yCACK,KAAK,YAAY,GAAE;;;iBAKlD,GAAM,GAAW,GAAY,SAAU,GACvC,GAA0B;;iBAEnB,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;iBAExD,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;;;aASrE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;UAQF,EAAsB;;;;MAMhC,MAAA,MC9MA,AAoBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,KASf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAgBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KCzDA,AAoBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;MAiBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KC5DA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,OAAQ,QAKvB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAepB,MAAA,MCzCA,AAoBA,GAAA,IAAA,WAME,WAAY,GAJZ,KAAA,YAAwB,GAKtB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAI1C,OADM,GAAW,CAAC,YAAY,EAAQ,GAAE,+BAC/B,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAC1B,EAAS,KACL,iBAAiB,EAAQ,GAAE,KAC3B,kBAAiB,EAAC,WAAW,EAAK,QAExC,GAAM,GAAY,EAAQ,OACpB,EAAY,EAAQ,EAAQ,OAAS,GAC3C,EAAS,KAAK,sBAAsB,EAAS,WAAW,EAAS,OAEjE,KAAK,SAAW;;;;;;UAMV,EAAS,KAAK;WAAa;;MAIrC,MAAA,MCzDA,AAuBA,GAAA,IAAA,WAOE,WAAY,EAAoB,GALhC,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,GAAM,GAAQ,KAAK,YACb,EAAO,EAAM,OACb,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAC/B,EAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACzD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAW1C,OARM,GAAU,EAAS,GACnB,EAAe,EAAS,MAAM,IAC9B,EAAc,EAAS,OAEzB,EAAkB,OAAO,EAAO,MAAM,EAAQ,GAAE;;oBAEpC,EAAW,WAAW,EAAa,OAAM;WAEhD,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAK1B,GAAmB;cACX,EAAO,MAAM,EAAQ,GAAE,QAAQ,EAAO,OAAO,EAAQ,EAAI,GAAE;;kBAEvD,EAAC,IAAI,GAAgB,EAAU,EAAS,GAAM;mBAC7C,GAAgB,EAAc,EAAS,GAAM;WAG5D,GAAM,GAAY,EAAQ,OACpB,EAAQ,EAAQ,EAAQ,OAAS,GACvC,GAAmB;;gBAEP,EAAS,IAAI,GAAgB,EAAU,EAAS,GAAM;iBACrD,GAAgB,EAAc,EAAS,GAAM,MAE1D,KAAK,SAAW;uBACG,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IAAE;UAC1C,EAAe;;;;UAIf,EAAK;sCACuB,EAAM;;UAElC,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;cACrC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;;MAMtC,MAAA,MAaA,YAAyB,EAAoB,EAAiB,GAC5D,GAAM,GAAa,EAAS,QAAQ,GAC9B,EAAM,EAAS,IAAI,SAAC,EAAG,GAC3B,MAAI,KAAQ,EACA,EAAC,MAAM,EAEV,IAGX,MAAO,GAAI,OC3Hb,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAiB,EAAS,aAAe,eAE/C,KAAK,SAAW;;;;;;;;;;;;8BAYU,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;oBAIhC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAExC,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;wCAEI,EAAM,aAAa,EAAM;;;;;;;gCAOjC,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;oCAEF,EAAS,YAAW;;oBAEpC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAEjC,KAAK,SAAW;;;;;;;;;;;8BAWU,EAAS,UAAS;kCACd,EAAS,SAAQ;iCAClB,EAAW,MAAM,EAAQ;;kCAExB,EAAS,QAAO;;;;oCAId,EAAS,UAAS;mCACnB,EAAY,MAAM,EAAM;;oCAEvB,EAAS,SAAQ;;;;sCAIf,EAAS,SAAQ;qCAClB,EAAW,MAAM,EAAO;;sCAEvB,EAAS,QAAO;;;;;;;;;;;;;MAetD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAEnD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAchC,EAAW;gDACK,EAAW;;oCAEvB,EAAS,SAAQ;;;;;yBAK5B,EAAW;;kCAEF,EAAY;kDACI,EAAY;;sCAExB,EAAS,UAAS;;;;;;2BAM7B,EAAY;;oCAEH,EAAW;oDACK,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;6BAM5B,EAAW;;sCAEF,EAAS,YAAW;;;;;;;;;;MAY1D,MAAA,MCzSA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;;;;;;;wBAOI,EAAU;;;;;8BAKJ,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;;;;;;;;;MAcpD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;gCAYnB,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;;oCAGF,EAAU;8BAChB,EAAU;;;;;;;;;MAWxC,MAAA,MCzIA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAChD,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAEpC,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;;2BAGT,EAAM,aAAa,EAAM;;;;;;;gCAOpB,EAAY;qCACP,EAAc;;gCAEnB,EAAS,SAAQ;;;;kCAIf,EAAW;uCACN,EAAa;;kCAElB,EAAS,QAAO;;;;oCAId,EAAqB;;;;;;;;oBAQrC,EAAc;;;;;;;;;;;;;;;;;;;kBAmBhB,KAA4B,GAAC;;oBAE3B,EAAc;;0CAEQ,EAAqB;mCAC5B,EAAqB;;;kCAGtB,EAAqB;mCACpB,EAAqB;;;yBAG/B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;;;;;yBAK5B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;gCACrB,EAAqB;;;;;;;;;;UAU3C,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAEtD,KAAK,SAAW;oCACgB,EAAW,KAAK,EAAY,KACxD,EAAW;iCACc,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;gCAgBhC,EAAW;qCACN,EAAa;;gCAElB,EAAS,QAAO;;;;kCAId,EAAY;uCACP,EAAc;;kCAEnB,EAAS,SAAQ;;;;oCAIf,EAAW;yCACN,EAAa;;oCAElB,EAAS,QAAO;;;;sCAId,EAAqB;;;;;;;;;;;;;;;;;oBAiBvC,KAA4B,GAAC;;4CAEL,EAAqB;qCAC5B,EAAqB;2BAC/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;;;2BAG/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;qCACrB,EAAqB;;;;;;;;;MAW1D,MAAA,MC3TA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAa,EAAS,YAAc,EAAS,WAE/C,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;wBAO3B,EAAU;4BACN,EAAU;;;;;;;;;gCASN,EAAY;qCACP,EAAc;;gCAEnB,EAAQ;;;;kCAIN,EAAW;uCACN,EAAa;;kCAElB,EAAQ;;;;;;;;;;;UAWhC,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MCrHA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAR/B,KAAA,cAAgB,CAAC,IAAK,KACtB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,SAgB5B,OAdM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAEjB,EAAW,gCAEN,EAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY;wBACI,EAAC,IAAI,EAAI,EAAC;mBACf,EAAC,IAAI,EAAC;mBACN,EAAC,IAAI,EAAC,eAYrB,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAS,EAAG,EAAS,EAAc,KAC1C,GAAM,GAAI,EAAS,EAOnB,GALA,GAAY;4BACQ,EAAI,EAAc;4BAClB,EAAI,EAAa;UAGjC,IAAgB,GAClB,GAAI,EAAI,GAEN,CAAI,EAAU,IAAM,EAUlB,GAAY;;qCAEW,EAAQ,mCAC3B,EAAQ;2BACC,EAAC,IAAI,EAAC;;;;uCAIM,EAAQ;6BAClB,EAAC,IAAI,EAAC;;;2BAGR,EAAC,IAAI,EAAC;;;;qCAII,EAAQ,mCAC3B,EAAQ;;;;;uCAKa,EAAQ;;;;sBAIzB,EAAC,IAAI,EAAC,+BAA+B,EAAC,IAAI,EAAC;;sBAE3C,EAAC,IAAI,EAAC,wBAAwB,EAAC,IAAI,EAAC;;gBAK5C,GAAY;qCACW,EAAQ,uBAAuB,EAAQ;2BACjD,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;oBAGb,EAAC,IAAI,EAAC,aAAa,EAAC,IAAI,EAAC;gBAI7B,EAAI,EAAI,IAOV,GAAM,GAAkB,EAAU,IAAM,EACpC,EAAA,KAAK,kBAAkB,GACvB,EAEJ,AAAK,EAAgB,IAAM,GAAK,EAAU,IAAM,GAC3C,EAAgB,IAAM,GAAK,EAAU,IAAM,EAC9C,IAAY;oCACQ,EAAU,EAAC,MAAM,EAAe;;uCAE7B,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;kBAMvB,AAAI,EAAgB,GAClB,IAAY;;yCAEW,EAAQ;oDACG,EAAQ;+BAC7B,EAAC,IAAI,EAAC;;+BAEN,EAAC,IAAI,EAAC;;qBAKrB,GAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;mBAG5C,GAAY;oCACQ,EAAe;;uCAEZ,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;;sBAGjB,EAAC,IAAI,GAAI,GAAC,aAAa,EAAC,IAAI,GAAI,GAAC;uBAM7C,AAAI,GAAI,GACN,IAAY;mCACW,EAAQ;cAS/B,AAAI,EAAU,IAAM,EAClB,IAAY;sCACY,EAAW;iDACA,EAAQ;2BAC9B,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;6CAGY,EAAQ;2BAC1B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;;wCAEY,EAAW;mDACA,EAAQ;;;sBAGrC,EAAC,IAAI,GAAI,GAAC,kBAAkB,EAAC,IAAI,GAAI,GAAC;oBAI9C,IAAY;qCACW,EAAQ;2BAClB,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;kCAGC,EAAW;iDACI,EAAQ;2BAC9B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;oBAKhD,GAAY,KAIhB,AAAI,EAAI,GACN,IAAY;0BACI,EAAC,IAAI,EAAC,WAAW,EAAC,KAAK,EAAC;gBAClC,EAAC,IAAI,EAAC,kBAAkB,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,EAAC;YAGzD,AAAI,EAAI,EAAI,GACV,IAAY;4BACI,EAAC,IAAI,GAAI,GAAC,WAAW,EAAC,KAAK,GAAI,GAAC;kBAC1C,EAAC,IAAI,GAAI,GAAC;8BACE,EAAC,IAAI,GAAI,GAAC,eAAe,EAAC,IAAI,GAAI,GAAC,UAM7D,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY,gBAAgB,EAAC,IAAI,EAAC,QAAQ,EAAC,IAAI,EAAC,IAIpD,GAAI,GAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;UAezC,EAAQ;;;UAGR,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MC3UA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAA4B,EAC5B,GAPJ,KAAA,cAAgB,CAAC,QAAS,QAAS,UACnC,KAAA,YAAwB,GAOf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAChC,EAAA,EAAA,GACA,EAAA,EAAA,GAAY,EAAA,EAAA,GACnB,KAAK,YAAc,CAAC,EAAU,EAAY,EAAW,GACrD,GAAM,GAAW,IAAW,WAAa,EAAI,EAEvC,EAAA,CAAA,EAAA,EAAA,KAAA,EAAA,EAAA,MAAC,EAAA,EAAA,GAAkB,EAAA,EAAA,GAGnB,EAAA,EAAA,oHAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAa,EAAA,EAAA,GAW3B,EAAA,EAAA,kHAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAe/B,KAAK,SAAW;yCACqB,EAAW;wCACZ,EAAU;;;;;;;;;;;;;;;;iCAgBjB,EAAK;;;;+BAIP,EAAW;8BACZ,EAAU;;uBAEjB,EAAG;mCACS,EAAgB;4BACvB,EAAkB;;;uBAGvB,EAAG;mCACS,EAAe;4BACtB,EAAkB;;;;;aAKjC,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;MA0BrB,MAAA,MC1GA,GAAA,WAQE,WAAY,EAAiB,EAAoB,GAPjD,KAAA,cAAgB,CAAC,KAQf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAM,OACb,EAAM,EAAY,MAAQ,QAAQ,GAAU,EAAM,UAAS,IAC3D,EAAS,EAAM,EAAM,OAAS,GAChC,EAAY,GACZ,EAAY,GAIhB,AAAI,EACF,GAAY,EAAU,UAAU,GAAS,GAAM,WAC/C,EAAY,EAAU,UAAY,WAElC,GAAY,EAAU,gBAAgB,EAAW,cACjD,EAAa,EAAU,aAAe,cAGxC,KAAK,SAAW;;;UAGV,GAAkB,GAAK;oBACb,GAAc,EAAM,UAAS;sBAC3B,EAAG;;cAEX,EAAS;sBACD,EAAS;YACnB,GAAc,EAAM,UAAS;wBACjB,GAAU,EAAM,UAAS;;;;MAO/C,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,OAAS,MAChB,GAAK,MAAQ,EAAM,mBAAmB,EAAc,UAEtD,EAAM,GAAG,UAAU,EAAK,MAAO,KAGrC,KAEA,YAAmB,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,KACpB,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,KAC/B,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,OAAO,EAAI,KAE/C,KAAM,OAAM,2BAA2B,EAAI,yBAI/C,YAAuB,EAAc,GACnC,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KAEd,KAAM,OAAM,2BAA2B,EAAI,yBCjG/C,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAEjB,KAAA,aAAe,GACf,KAAA,aAAe,GAEf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAuB,EAAmB,GAP9C,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAOtB,KAAK,YAAc,EACnB,KAAK,UAAY,EACjB,KAAK,WAAa,EAClB,KAAK,SAAW;;;;gBAIJ,KAAK,uBAAsB;gBAC3B,KAAK,sBAAqB;gBAC1B,KAAK,sBAAqB;;uBAEnB,EAAS;+BACD,EAAS;uBACjB,EAAS;+BACD,EAAS;mCACL,EAAS;UAClC,KAAK,qBAAoB;;;uBAGZ,KAAK,yBAAwB;;;IAM1C,SAAA,UAAA,qBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,mBAAR,WACE,MAAI,MAAK,aAAe,OACf,KAAK,YAAY,GAEjB,KAAK,YAAY,IAIpB,EAAA,UAAA,uBAAR,WACE,MAAI,MAAK,aAAe,OACf,4BAEA,6BAGb,KC5FA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,CAAC,EAAM,GAC1B,KAAK,SAAW;;;;;;MAQpB,MAAA,MClCA,AAsBA,GAAA,IAAA,WAME,WAAY,GALZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;UAIlB,EAAK,OAAM;;MAIrB,MAAA,MCxCA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;;UAKlB,EAAK,OAAM;;MAIrB,MAAA,MC3CA,AAqBA,GAAA,IAAA,WAKE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IANJ,KAAA,cAAgB,CAAC,KAOf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAS,SACb,AAAI,GACF,GAAS,8BAGX,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;8BAUrB,EAAK;kCACD,EAAK;iDACU,EAAK,OAAO,EAAM;wBAC3C,EAAK,UAAS;;;;;;;;;;;;;;UAc5B,EAAK,OAAM,WAAW,EAAM;;MAItC,MAAA,MCtEA,AAqCA,GAAA,IAAA,WAOE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAW,GACX,EAAS,SACb,AAAI,GACF,GAAS,8BAGX,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAC1B,GAAM,GAAU,EAAM,EAAI,EAE1B,GAAY;;gCAEY,EAAG,MAAM,EAAY,GAAE;gCACvB,EAAG;kCACD,EAAG,MAAM,EAAY,GAAE;kCACvB,EAAG;;;;;;;gCAOL,EAAK;oCACD,EAAK;kDACS,EAAK,OAAO,EAAM;yBAC3C,EAAK,UAAS;;;yBAGd,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;;;UAQ5B,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;;UAWzC,EAAQ;;UAER,EAAK,OAAM,MAAM,EAAM;;MAIjC,MAAA,MC/GA,AAmBO,GAAM,IAAc,CACzB,KAAM,oCACN,KAAM,qCAGR,GAAA,WAKE,WAAY,EAAY,EAA8B,GAJtD,KAAA,cAAgB,CAAC,OAAQ,QAKvB,GAAM,GAAW,EAAW,GAC5B,KAAK,YAAc,EAEnB,GAAM,GACF,EAAU,SAAS,KAAK,GAAO,UAAU,KAAK,GAC5C,EAAoB,EAAa,EAAQ,KAAO,MAEtD,KAAK,SAAW;yCACqB,EAAyB;;;UAGxD,EAAE;;;;kDAIsC,EAAQ;;;;;;8BAM5B,EAAQ;;;;;;;;;yDASmB,EAAiB;;;;;;;;;;MAY1E,MAAA,MCxEA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAiB,GAL7B,KAAA,YAAwB,GAMtB,KAAK,cAAgB,CAAC,KACtB,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;;MASlB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,UAEhE,EAAM,GAAG,UAAU,EAAK,SAAU,KAGxC,KChDA,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,EAAuB,GALrD,KAAA,cAAgB,CAAC,IAAK,WAMpB,GAAM,GAAwB,EAAO,QACrC,EAAY,GAAQ,EACpB,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,EAAQ,GAE7C,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,EAAkB,GACzC,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,mBAAmB,EAAI,yBAErC,GAAI,IAAS,EACX,MAAO,yBAMT,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAElD,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,IAAM,EACR,EAAa,KAAK,kBAAkB,EAAc,GAAE,MAEpD,EAAa,KAAK,GAAG,EAAc,IAGvC,MAAO,GAAa,OC3CtB,GAAA,IAAA,WAIE,WACY,EAA0B,EAAmB,GAA7C,KAAA,SAAA,EAA0B,KAAA,QAAA,EAJtC,KAAA,cAAgB,CAAC,IAAK,WAKpB,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAChC,EAAe,KAAK,SAAW,EAAI,aAAe,UACxD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,KAAK,QAAO;;YAElD,EAAK;;gCAEe,KAAK,SAAQ;;sCAEP,EAAY;;;;QAMlD,MAAA,MC1CA,YAwBmC,GACjC,GAAM,GAAO,KACP,EAAwB,EAAK,QAAO;;MAEtC,EAAK,UAAS;MACd,EAAK,UAAS;MACd,EAAK,UAAS;;;;;OAMlB,MAAO,IAA8B,EAAI,eAGR,GAEjC,GAAM,GAAc,GAAI,cACpB,CAAC,GAAI,EAAG,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,GAAI,EAAG,EAAG,IAClE,MAAO,IAAoC,EAAI,eAGf,GAEhC,GAAM,GAAwB,GAAI,aAAY,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,IAC9D,MAAO,IAAmC,EAAI,GAGhD,YACI,EAA2B,EAAe,EAC1C,EAAwB,EACxB,GACF,GAA+B,EAAO,GACtC,GAAM,GAAU,GAAyB,GAEnC,EAAQ,EAAG,WACjB,UAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAO,KACxD,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAO,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAC5C,EAAa,QACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,QACzD,cAIL,GACF,MAAO,GAAc,gCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAG,mBAIvC,GACF,MAAO,GAAc,oCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAc,kCAIlD,GACF,MAAO,GAAc,kCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAG,2BAIL,GACF,MAAO,GAAc,sCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EAAQ,GAAwC,GAC3D,EAAG,KAAM,EAAG,mBAId,GACF,MAAO,GAAc,0CAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAc,kCAIhB,EAA2B,EAC3B,GACF,GAAM,GAAY,EACZ,EAAW,EAAI,EACf,EAAU,EAAI,EAAM,EAAI,EAC9B,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KAC7C,GAAM,GAAU,GACZ,EAAI,EAAS,eAAgB,EAAc,EAAG,EAAQ,GAC1D,MAAO,IACH,GACI,EAAI,EAAS,KAAM,EAAc,EAAG,EAAQ,eAIlD,EAA2B,EAAuB,EAClD,EAAgB,EAAkB,GACpC,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAEhE,GAAI,GAA2B,EAAuB,EACtD,AAAI,YAAgB,YAClB,GAAgB,GAAI,YAAW,EAAQ,EAAS,GAChD,EAAgB,EAAG,cACnB,EAAiB,EAAG,MAEpB,GAAgB,GAAI,cAAa,EAAQ,EAAS,GAClD,EAAgB,EAAG,MACnB,EAAiB,EAAc,2BAGjC,EAAc,IAAI,GAElB,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAAG,KACvD,EAAe,KAEvB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA2B,EAC3B,GAEF,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAChE,AAAK,EAAqB,eAAgB,YACxC,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAO,MAAO,EAAO,OAAQ,EAAG,EAAG,KAC9D,EAAG,cAAgB,EAAqB,QAEhD,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAG,KAAM,EAAG,cACvC,KAIV,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA6B,EAAc,EAC3C,GAEF,GAAM,GAAS,EAAI,eACnB,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,KAGrD,GAAM,GAAgB,EAChB,EAAiB,EACjB,EAAkB,EAAgB,EAAiB,EAAO,EAEhE,UACI,EACA,WAAM,MAAA,GAAI,WACN,EAAI,kBAAmB,EAAiB,EAAI,eAIpD,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAG,EAAG,EAAS,EAAM,EAAI,KAAM,EAAI,MAAO,KAExE,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,QAE9C,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAM,EAEN,EAAiB,GAAI,cAAa,GAExC,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAGJ,EAAc,EACd,EAAiB,GAAI,YACvB,GAA4C,EAAO,EAAS,IAEhE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAG,EAAG,EAAc,sBAAuB,EAAG,cACpD,KAID,GAAI,cAAa,EAAe,oBAIrC,EAA2B,EAAqB,EAAe,EAC/D,EAAc,EAAsB,EACpC,GACF,GAAM,GAAM,EAEN,EACF,GAAI,cAAa,GACb,EAAc,IAEtB,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAa,GAAI,cAAa,EAAe,EAAe,GAClE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAc,EAAc,EAAG,KAAM,EAAG,MAAO,KAEtD,owBCxTT,kBAgDE,WAAY,GANZ,KAAA,cAAmC,KACnC,KAAA,QAA6B,KACrB,KAAA,SAAW,GAqOX,KAAA,oBAAsB,GAsPtB,KAAA,YAA0B,GAtdhC,GAAM,GAAY,EAAA,MAAM,UAAU,iBAClC,AAAI,GAAM,KACR,MAAK,GAAK,EACV,GAAgB,EAAW,IAE3B,KAAK,GAAK,GAAgB,GAG5B,GAAI,GAAqB,2BACnB,EAA0B,8BAChC,GAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAM,GAAgB,oBAChB,EAAqB,yBAI3B,GAFA,KAAK,sBACD,GAA+B,KAAK,GAAI,GACxC,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,sHAKN,GADA,KAAK,0BAA4B,KAAK,GAAG,aAAa,GAClD,GAAwB,KAAK,GAAI,GACnC,KAAK,8BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,uIAIN,EAAqB,yBACjB,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,KAAK,GAAG,aAAa,WAChB,GAAwB,KAAK,GAAI,GAC1C,KAAK,8BACD,KAAK,GAAG,aAAa,OAEzB,MAAM,IAAI,OAAM,uDAIpB,KAAK,aAAe,GAA8B,KAAK,IACvD,KAAK,YAAc,GAA6B,KAAK,IACrD,KAAK,YAAc,GAA6B,KAAK,IAErD,KAAK,cACD,GAA0B,KAAK,GAAI,KAAK,2BAG9C,cAAA,eAAY,EAAA,UAAA,QAAK,KAAjB,WACE,MAAO,GAAA,MAAM,QAAQ,0CAGhB,EAAA,UAAA,QAAP,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAEF,AAAI,KAAK,SAAW,MAClB,QAAQ,KACJ,wKAIN,AAAI,KAAK,eAAiB,MACxB,QAAQ,KACJ,sMAKN,GAAM,GAAK,KAAK,GAChB,GAAwB,EAAI,WAAM,MAAA,GAAG,WACrC,GAAwB,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QACrE,GAAwB,EAAI,WAAM,MAAA,GAAG,kBAAkB,EAAK,eAC5D,GAAwB,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,QACjE,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,QACrD,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAK,eACvD,KAAK,SAAW,IAGX,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,yBAAP,SACI,EACA,GACF,KAAK,kBACL,GAAoC,KAAK,GAAI,EAAS,IAGjD,EAAA,UAAA,2BAAP,SACI,EAAuB,EAAe,EAAgB,GACxD,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,EAAO,EAAQ,EAAM,KAAK,gBAG3C,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,0BAAP,SAAiC,EAAc,GAE7C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,oBAAP,SAA2B,GAA3B,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,KAAK,gBAAkB,GACzB,IAA6C,KAAK,GAAI,KAAK,aAC3D,KAAK,cAAgB,MAEvB,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAGxD,EAAA,UAAA,gDAAP,SACI,EAAuB,EAAc,GADzC,GAAA,GAAA,KAEE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAM,EAAS,EAAK,kBAGhC,EAAA,UAAA,+BAAP,SACI,EAAqB,EAAe,EAAc,EAClD,EAAsB,GACxB,MAAO,IACH,KAAK,GAAI,EAAQ,EAAO,EAAM,EAAS,EAAc,EACrD,KAAK,gBAGJ,EAAA,UAAA,gCAAP,SAAuC,EAAqB,GAE1D,MAAO,IAA2C,KAAK,GAAI,EAAQ,IAG9D,EAAA,UAAA,wBAAP,SACI,EAAuB,EAAc,GACvC,KAAK,yBAAyB,GAC9B,GAAM,GAAS,GACX,KAAK,GAA8B,EAAM,EAAS,KAAK,eAC3D,YAAK,6BACE,GAGF,EAAA,UAAA,sBAAP,WACE,GAAM,GAAe,KAAK,YAAY,KAAK,IAC3C,MAAO,MAAK,UAAU,IAGhB,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACM,EACA,EAEJ,GAAI,EAAA,MAAM,QAAQ,4BAChB,GAAM,GAAM,EAEN,EAAO,EAAI,UAAU,EAAI,2BAA4B,GAC3D,EAAG,QAEH,EAAgB,WACd,GAAM,GAAS,EAAI,eAAe,EAAM,EAAG,GAC3C,MAAO,KAAW,EAAI,kBAClB,IAAW,EAAI,qBAGrB,EAAQ,MACH,AACH,GAAA,MAAM,UAAU,gDAAkD,EACpE,GAAQ,KAAK,aACb,KAAK,WACL,EAAgB,WAAM,MAAA,GAAK,iBACvB,EACA,EAAA,MAAM,UAAU,mDAMpB,EAAgB,WAAM,MAAA,IAGxB,MAAO,CAAC,MAAK,EAAE,cAAa,IAGvB,EAAA,UAAA,gCAAP,SACI,EAAuB,EACvB,GAFJ,GAAA,GAAA,KAGE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAc,MAK1B,EAAA,UAAA,cAAP,SAAqB,GACnB,KAAK,kBACL,GAAM,GAAK,KAAK,GACV,EACF,GAAgC,EAAI,GAClC,EAA4B,GAA8B,GAC1D,EAAwB,GAAyB,GACvD,UAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAuB,EAAI,GAC3B,AAAI,KAAK,OACP,GAA2B,EAAI,GAEjC,AAAK,KAAK,qBACR,MAAK,WAAW,GAChB,KAAK,oBAAsB,GACvB,EAAI,KAAK,QAAS,KAAK,eAEtB,GAGF,EAAA,UAAA,cAAP,SAAqB,GAArB,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,IAAY,KAAK,SACnB,MAAK,QAAU,MAEjB,AAAI,GAAW,MACb,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAI1D,EAAA,UAAA,WAAP,SAAkB,GAAlB,GAAA,GAAA,KACE,KAAK,kBACL,KAAK,QAAU,EACf,AAAK,KAAK,SAAW,MAAS,KAAK,OACjC,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,WAAW,MAGrD,EAAA,UAAA,mBAAP,SACI,EAAuB,EACvB,GAEF,MAFE,AAAA,KAAA,QAAA,GAAA,IACF,KAAK,kBACD,EACK,GACH,KAAK,GAAI,EAAS,GAEf,GACH,KAAK,GAAI,EAAS,IAInB,EAAA,UAAA,qBAAP,SAA4B,EAAuB,GAAnD,GAAA,GAAA,KAEE,YAAK,kBACE,GACH,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,kBAAkB,EAAS,MAGjD,EAAA,UAAA,0BAAP,SAAiC,EAAuB,GAEtD,YAAK,kBACE,KAAK,GAAG,mBAAmB,EAAS,IAGtC,EAAA,UAAA,sBAAP,SACI,EAAkC,EAClC,GACF,KAAK,kBACL,KAAK,mBACL,GACI,KAAK,GAAI,EAAoB,EAAiB,IAG7C,EAAA,UAAA,uBAAP,SACI,EAAmC,EAAc,GACnD,KAAK,6BAA6B,EAAqB,EAAS,IAG3D,EAAA,UAAA,6BAAP,SACI,EAAyC,EAAc,GACzD,KAAK,kBACC,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,KAAK,6BAA6B,EAA2B,EAAO,IAG/D,EAAA,UAAA,2BAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAK,iCACD,EAAa,EAAU,EAAY,IAGlC,EAAA,UAAA,iCAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAM,IAAI,OAAM,sDAGX,EAAA,UAAA,cAAP,WACE,AAAI,KAAK,SAAW,MAClB,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAA+B,KAAK,KAG/B,EAAA,UAAA,eAAP,WACE,KAAK,kBACL,KAAK,mBACL,GAAM,GAAK,KAAK,GAChB,AAAI,KAAK,OACP,KAAK,gBAEP,GACI,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,UAAW,EAAG,EAAG,eAAgB,MAG7D,EAAA,UAAA,+BAAP,WAAA,GAAA,GAAA,KACE,KAAK,kBACL,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,YAGzC,EAAA,UAAA,uBAAR,WAEE,MAAI,MAAK,6BAA+B,MACtC,MAAK,4BACD,GACI,KAAK,GACL,EAAA,MAAM,UACF,kDAAoD,EACpD,kCACA,6BAIP,KAAK,6BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGd,EAAA,UAAA,WAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EAAQ,EAAI,cAClB,SAAI,WAAW,EAAI,iBAAkB,GAC9B,EAET,GAAM,GAAM,KAAK,+BACX,EAAQ,EAAI,iBAClB,SAAI,cAAc,EAAI,iBAAkB,GACjC,GAGT,EAAA,UAAA,SAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BACjB,EAAI,SAAS,EAAI,kBACjB,OAEF,GAAM,GAAM,KAAK,+BACjB,EAAI,YAAY,EAAI,mBAGT,EAAA,UAAA,uBAAb,SAAoC,wGAClC,MAAA,CAAA,EAAM,EAAA,KAAK,YACP,WAAM,MAAA,GAAK,UAGP,EAAK,iBACD,EACA,EAAA,MAAM,UACF,2DAPhB,SAAA,OAQA,CAAA,EAAO,KAAK,aACR,EAAO,EAAA,MAAM,UAAU,wDAGrB,EAAA,UAAA,aAAR,SAAqB,EAAmB,GACtC,GAAI,IAAsB,EACxB,MAAO,MAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GAEX,EAAmB,EAAI,kBAAkB,EAAO,EAAI,cAE1D,MAAO,GAAmB,SAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,kBAErC,MAAO,GAAmB,MAItB,EAAA,UAAA,iBAAR,SAAyB,EAAmB,GAE1C,GAAI,IAAsB,EACxB,MAAO,GAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,wBACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,cAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,4BACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,WAI9B,EAAA,UAAA,UAAA,SAAU,GAAV,GAAA,GAAA,KACE,MAAO,IAAI,SAAc,SAAA,GACvB,EAAK,cAAc,WAAM,MAAA,GAAa,iBAAiB,WAAM,MAAA,UAMjE,EAAA,UAAA,UAAA,WAGE,OADM,GAAQ,GAAqB,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,YACtD,EAAI,EAAG,GAAK,EAAO,EAAE,GACrB,GAAA,GAAA,KAAA,YAAA,GAAA,UACP,IAEF,KAAK,YAAc,KAAK,YAAY,MAAM,EAAQ,IAG5C,EAAA,UAAA,cAAR,SAAsB,EAAyB,GAA/C,GAAA,GAAA,KAEE,GADA,KAAK,YAAY,KAAK,CAAC,SAAQ,EAAE,UAAS,IACtC,KAAK,YAAY,OAAS,EAE5B,OAGF,EAAA,KAAK,YAAY,WACf,SAAK,YAEE,EAAK,YAAY,SAAW,KAI/B,EAAA,UAAA,yBAAR,SAAiC,GAC/B,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,KAAK,aAC3B,AAAI,KAAK,OACP,GAA+B,KAAK,KAIhC,EAAA,UAAA,2BAAR,WACE,AAAI,KAAK,eAAiB,KACxB,IACI,KAAK,GAAI,KAAK,cAAe,KAAK,aACtC,AAAI,KAAK,OACP,GAA+B,KAAK,KAGtC,GAA6C,KAAK,GAAI,KAAK,cAIvD,EAAA,UAAA,qBAAR,SACI,EACA,GACF,KAAK,yBAAyB,GAC9B,GAAM,GAAS,IACf,YAAK,6BAEE,GAGD,EAAA,UAAA,6BAAR,SACI,EAA8C,EAC9C,GACF,KAAK,kBACL,GAAM,GAAK,KAAK,GAChB,GACI,EAAI,EAAgC,KAAK,aAC7C,AAAI,KAAK,OACP,GAA+B,GAEjC,KAAK,cAAgB,EACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAO,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAO,MAGpD,EAAA,UAAA,iCAAR,SACI,EAAW,EAAW,EAAe,GADzC,GAAA,GAAA,KAEE,KAAK,kBACL,GACI,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,QAAQ,EAAG,EAAG,EAAO,MAG1C,EAAA,UAAA,gBAAR,WACE,GAAI,KAAK,SACP,KAAM,IAAI,OAAM,4CAIZ,EAAA,UAAA,iBAAR,WACE,GAAI,KAAK,SAAW,KAClB,KAAM,IAAI,OAAM,qCAGtB,iBAaqC,GAEnC,OADI,GAAI,EACD,EAAI,EAAI,OAAQ,EAAE,GACvB,GAAM,GAAS,EAAI,KACnB,GAAI,CAAC,EACH,MAGJ,MAAO,GAAI,ECtnBb,YAgEI,EAAqB,EAAuB,EAC5C,GACF,GAAM,GAAW,EAAQ,SACnB,EAA0B,EAAO,IAAI,SAAC,EAAO,GACjD,GAAM,GAAuB,CAC3B,aAAc,EAAM,MACpB,SAAU,EAAM,UAAY,KAAO,EAAM,QAAQ,SACjD,UAAW,EAAM,UACjB,SAAU,EAAM,UAAY,GAAQ,EAAM,QAAQ,SAClD,WAAY,MAEd,MAAI,GAAM,SAAW,MAAQ,EAAM,QAAQ,OAAS,MAChD,EAAM,QAAQ,MAAM,WAAa,GACnC,GAAU,WAAa,EAAM,QAAQ,MAAM,YAEtC,CAAC,KAAM,EAAQ,cAAc,GAAI,UAAS,KAE7C,EAAe,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,YACrC,EAA0B,CAC9B,aAAc,EAAO,MACrB,SAAU,EAAO,QAAQ,SACzB,UAAW,GACX,SAAU,EAAO,QAAQ,SACzB,WAAY,MAER,EAAS,GACX,EAAY,EAAc,EAAU,EAAQ,cAE1C,EAAe,EAAM,cAAc,GAGrC,EAA+B,KAC7B,EAAS,EAAM,mBAAmB,EAAc,MAAO,IAC7D,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAS,EAAM,mBAAmB,EAAc,WAAY,KAK9D,OADM,GAA2D,GACxD,EAAI,EAAG,EAAI,EAAQ,cAAc,OAAQ,KAChD,GAAM,GAAU,EAAQ,cAAc,GAChC,EAAc,GACpB,EAAiB,GACb,EAAM,mBAAmB,EAAc,EAAS,GACpD,EAAiB,SAAS,GACtB,EAAM,mBAAmB,EAAc,SAAS,EAAW,GAGjE,MAAO,CACL,QAAO,EACP,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,aAAY,EACZ,aAAY,EACZ,OAAM,EACN,OAAM,GAIV,YACI,EAAyB,GAC3B,GAAI,EAAW,SAAW,EAAO,OAC/B,KAAM,OACF,4BAA4B,EAAW,OAAM,gBAC7C,sBAAqB,EAAO,OAAM,YAGxC,EAAW,QAAQ,SAAC,EAAG,GACrB,GAAM,GAAS,EAAE,aACX,EAAQ,EAAO,GACf,EAAS,EAAM,MAErB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAQ,GAC5B,KAAM,OACF,kDACA,6BAA4B,EAAM,QAAQ,EAAM,gBAGtD,GAAI,EAAE,WAAa,EAAM,UACvB,OAGF,GAAM,GAAY,EAAE,SACd,EAAY,EAAM,UAAY,KAAO,EAAM,QAAQ,SACzD,GAAI,CAAC,EAAA,KAAK,YAAY,EAAW,GAC/B,KAAM,OACF,6DACA,yBAAwB,EAAS,QAAQ,EAAS,8BAMxD,EAAqB,EAAqB,EAC1C,EACA,GAEF,GAAyB,EAAO,aAAc,GAC9C,GAAyB,CAAC,EAAO,cAAe,CAAC,IAEjD,GAAM,GAAS,EAAO,QAAQ,QACxB,EAAc,EAAO,QAAQ,SACnC,AAAI,EAAO,QAAQ,SACjB,EAAM,6BAA6B,EAAQ,EAAY,GAAI,EAAY,IAEvE,EAAM,uBAAuB,EAAQ,EAAY,GAAI,EAAY,IAEnE,EAAM,WAAW,EAAO,cAGxB,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,CAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,WAGtC,AAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,KAIpC,EAAO,QAAQ,SAAC,EAAO,GACrB,GAAM,GAAU,EAAO,QAAQ,cAAc,GACvC,EAAS,EAAO,iBAAiB,GACjC,EAAe,EAAO,iBAAiB,SAAS,GAEtD,GAAI,GAAU,KAEZ,OAGF,GAAI,EAAM,WAER,GAAI,EAAA,KAAK,cAAc,EAAM,OAAS,EACpC,EAAM,GAAG,UAAU,EAAQ,EAAM,cAAc,SAE/C,GAAI,GAAO,EAAM,cACjB,AAAM,YAAgB,eACpB,GAAO,GAAI,cAAa,IAE1B,EAAM,GAAG,WAAW,EAAQ,GAE9B,OAIF,AAAI,EAAM,QAAQ,OAAS,MAAQ,GAAgB,MACjD,EAAM,GAAG,UAAU,EAAc,EAAM,QAAQ,MAAM,YAGvD,EAAM,sBAAsB,EAAM,QAAQ,QAAS,EAAQ,KAG7D,AAAI,GAAe,MACjB,EAAY,EAAO,EAAO,cAE5B,EAAM,6BAIJ,EAAuB,EAAsB,GAC/C,GAAI,GAAY,GAChB,EAAO,OAAO,GAAQ,QAAQ,SAAA,GAC5B,GAAM,GAAY,EAAE,SAAW,MAAQ,EAAE,QAAQ,OAAS,MACtD,EAAE,QAAQ,MAAM,WAAa,EAC3B,EAAW,EAAE,UAAY,UAAY,EAAE,QAAQ,SACrD,GAAgB,EAAE,MAAK,IAAI,EAAQ,IAAI,IAEzC,GAAM,GAAc,EAAQ,SACxB,EAAM,EAAQ,YAAY,KAE9B,UAAO,IAAM,EAAY,IAAM,EACxB,EC5OT,AAqBA,GAAA,IAAA,WAOE,WACI,EAAuB,EACvB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAsBnB,OAnBE,GAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,cACA,EAAA,EAAA,eACA,EAAA,EAAA,WAEK,EAAA,EAAA,KAAM,EAAA,EAAA,IACP,EAAmB,EAAa,EAChC,EAAO,KACP,EAAiB,IAAe,eAChC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAEhC,EAAW,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,GAAY;gCACY,EAAG;yBACV,EAAG;;4BAEA,EAAY,GAAE,aAAa,EAAY,GAAE;0CAC3B,EAAQ,QAAQ,EAAY,MAC1D,EAAG;6BACc,EAAc,aAAa,EAAgB;;sBAElD,EAAW,GAAO;;qDAEa,EAAQ,QACjD,EAAW,OAAO,EAAI;+BACH,EAAa,2BAChC,EAAgB,QAAQ,EAAU;;wBAEtB,EAAW,GAAO;;2CAEC,EAAU;;sBAE/B,EAAc;;2BAET,GAAM,EAAI,GAAG;;;;;2BAKb,GAAM,EAAI,GAAG;;;;;;;UAWpC,KAAK,SAAW;;;;;;;;;UASV,EAAQ;;UAER,EAAK,OAAM;;MAIrB,MAAA,MC5GA,AAmBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAgB,EAAc,EAChD,GANJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMtB,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;;wBASI,EAAG,UAAU,EAAG;;oCAEJ,EAAI;;;;;0BAKd,EAAW;;;MAKrC,MAAA,MClEA,AAmBA,GAAA,IAAA,WAUE,WACI,EAAsB,EAAqB,EAAc,EACzD,GAXJ,KAAA,cAAgB,CAAC,aAAc,cAAe,MAC9C,KAAA,YAAwB,GAWtB,KAAK,YAAc,EACnB,KAAK,MAAQ,EAAW,GACxB,KAAK,YAAc,EACnB,KAAK,KAAO,EACZ,KAAK,MAAQ,EACb,KAAK,KAAO,EACZ,KAAK,SAAW;;;;;;;;8BAQU,KAAK,MAAK;oDACY,EAAW;yCACtB,KAAK,MAAK;0BACzB,EAAW;;;sCAGC,KAAK,MAAK;;;;;;;;;;;;;;;yBAevB,EAAK,oBAAoB,EAAI;;;;;;;yCAOb,EAAK;0BACpB,EAAI;;;;0CAIY,EAAI;;;;;;;;;;;;;;MAgB9C,MAAA,MC/FA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAgB,EAAc,EAChD,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;gCAQY,KAAK,YAAY,GAAE;gCACnB,KAAK,YAAY,GAAE;;;;;;;;;;;;;;;iCAelB,EAAG;;;;;;;;;;;yBAWX,EAAG,UAAU,EAAG;;;6DAGoB,EAAI;;;;;;;;;;;;;;;;;;;;;0CAqBvB,EAAW;;;MAKrD,MAAA,MC/GA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAY,EAAwB,EAAuB,EACjE,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAcnB,EAAqB;kBACnC,EAAc;gDACgB,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;kDACJ,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;gCAOvB,EAAS;;;;qCAIJ,EAAoB;;;;;;;;MAUzD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EACF,EAAuB,EAAwB,EAAuB,EAC1E,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;gCAiBhC,EAAoB;mBACjC,EAAa;gDACgB,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;kCAOvB,EAAS;;;;;;yBAMlB,EAAqB,MAAM,EAAoB;yBAC/C,EAAoB;;;;;;;;;MAW7C,MAAA,MChLA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkC,EAClC,EAAoB,EAAoB,EACxC,EAA2B,GAD3B,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IACxC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAT/B,KAAA,cAAgB,CAAC,UAAW,WAC5B,KAAA,aAAe,GACf,KAAA,aAAe,GAQb,KAAK,YAAc,EAEnB,GAAM,GAAY,EAAa,EAAO,GAAK,EAAO,GAC5C,EAAwB,KAAK,KAAK,EAAY,GAE9C,EAAU,EAAa,cAAgB,cACvC,EAAU,EAAa,cAAgB,cACvC,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAC1D,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAE5D,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;sCAEa,EAAqB;;;;8BAI7B,EAAqB;sCACb,EAAO;sCACP,EAAO;;;;uBAItB,EAAS,GAAE,MAAM,EAAS,GAAE;uBAC5B,EAAS,GAAE,MAAM,EAAS,GAAE;;;;;;;;;UASzC,EAAc;;UAEd,EAAsB;;;;MAMhC,MAAA,MChGA,AAoBA,GAAA,IAAA,WAQE,WAAY,EAAmB,EAAqB,GAPpD,KAAA,cAAgB,CAAC,SAQf,KAAK,YAAc,CAAC,EAAW,GAE/B,KAAK,SAAW;;;;;;;;;;8BAUU,GAAc,GAAC;;;;;;;;;;0BAUnB,GAAc,GAAC;;MAKvC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAM,mBAAmB,EAAc,SAExD,EAAM,GAAG,UAAU,EAAK,QAAS,KAGvC,KChEA,AAmBA,GAAA,IAAA,WAQE,WACI,EAAoB,EAAe,EAAiB,GARxD,KAAA,cAAgB,CAAC,WASf,KAAK,YAAc,CAAC,EAAY,GAEhC,KAAK,SAAW;;;;8BAIU,EAAQ,YAAY,EAAO;;;MAKzD,MAAA,MCxCA,AAqBA,GAAA,IAAA,WAOE,WACI,GAPJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEzB,GAAI,IAAS,EACX,KAAK,SAAW;;;;aAMhB,GAAM,GAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EACF,GAAwB,EAAM,EAAa,GACzC,EAAQ,GACV,EAAM,EAAY,EAAY,OAAS,GACvC,EAAY,EAAY,OAAS,GAAI,GACnC,EAAS,GAAU,EAAa,GAEtC,KAAK,SAAW;;YAEV,EAAK;;eAEF,EAAoB;;;cAGrB,EAAK;;6BAEU,EAAM;;;SAMnC,MAAA,MAEA,YAA4B,EAAc,GAGxC,OAFM,GAAS,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAG1B,OAFI,GAAW,KAAQ,EAAI,IAAM,OAAK,KAAK,KAAQ,EAAI,IAAM,OAEpD,EAAI,EAAG,EAAI,EAAM,IACxB,EAAW,EAAK,EAAK,OAAS,EAAI,GAAE,IAAM,EAG5C,EAAO,KAAK,GAGhB,MAAO,GAGT,YACI,EAAc,EAAiB,GACjC,GAAI,IAAS,EACX,MAAO,QAAQ,EAAM,GAIvB,OADI,GAAO,GACF,EAAI,EAAO,EAAG,EAAI,EAAM,IAC/B,GAAW,EAAK,GAAE,OAAO,EAAM,GAC/B,AAAI,EAAI,EAAO,GACb,IAAQ,MAIZ,MAAO,GAGT,YACI,EAAc,EAAc,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,GAGT,GAAM,GAAY,EAAK,MAAM,IAE7B,MAAO;cACK,EAAU,GAAE;cACZ,EAAU,GAAE;;;;0BAIA,EAAI;0BACJ,EAAI;IAI9B,YAAmB,EAAiB,GAClC,GAAM,GAAO,EAAM,OACb,EAAe,GAAmB,EAAM,GAC9C,MAAI,KAAS,EACJ;wBACa,EAAM,GAAE;kBAIvB,QAAQ,EAAa,GAAE;8BACF,EAAa,GAAE;8BACf,EAAa,GAAE;uCACN,EAAa,GAAE,ICrItD,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAClB,GANJ,KAAA,cAAgB,CAAC,KAOf,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KACnD,GAAM,GAAO,EAAO,OACd,EAAO,GAAkB,GAEzB,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EACF,CAAC,YAAa,YAAa,YAAa,aAAa,MAAM,EAAG,GAElE,GAAI,IAAS,GACX,KAAK,SAAW;sBACA,EAAK;oBACP,EAAG;;;;;8BAKO,EAAa;;;;;QAMrC,OAEF,KAAK,SAAW;QACZ,EAAI,YAAY,EAAI,IAAI,EAAK;QAC7B,EAAI,UAAU,EAAI,IAAI,EAAG;;;UAGvB,EAAI;;4BAEc,EAAa;;YAE7B,EAAI;2BACW,EAAc;;;MAKzC,MAAA,MCrEA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAClB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KA4BnD,OA3BM,GAAO,EAAO,OACd,EAAQ,GAAkB,GAE1B,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EAAS,GAAY,KAAM,GAC3B,EAAS,GAAY,SAAU,GAC/B,EAAY,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAC1D,EACF,IAAS,EAAI,SAAW,QAAQ,EAAO,MAAM,IAAI,OAAM,IAErD,EAAiB,CAClB,EAAK,mBAAuB,EAAO,EAAO,GAAE;YACzC,EAAM;QAEZ,IAAS,EAAI,GAAK;;SAEf,EAAO,EAAO,GAAE;YACb,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAAE,MACtD,IAAS,EAAI,GAAK,KAAK,EAAO,EAAO,GAAE;cAC/B,EAAM,OAGV,EAAc,IAAS,EACzB,0BACA,6DACA,EAAW,GACN,EAAI,EAAG,EAAI,IAAS,EAAI,EAAI,EAAG,EAAI,EAAG,IAC7C,GAAY;UACR,EAAe,GAAE;cACb,EAAW;mBACN,EAAC,aAAa,EAAa;;YAElC,EAAK;mBACE,EAAC,uBAAuB,EAAO,OAAM,MAAM,EAAS;;QAInE,GAAa,IAAS,EAAI,KAAO,KAEjC,KAAK,SAAW;cACN,EAAK,YAAY,EAAK,IAAI,EAAK;cAC/B,EAAK,UAAU,EAAK,IAAI,EAAG;;;UAG/B,EAAK;;UAEL,EAAQ;;;MAKlB,MAAA,MCrFA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MACzB,EAA0B,cAAc,EAAS,SAAQ,YAC3D,EAAS,QAAO,YAAY,EAAS,WAAU,OAC7C,EACF,SAAS,EAAS,QAAO,YAAY,EAAS,WAAU,OAExD,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;sCACgB,EAAY,KAAK,EAAW;mCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBnB,EAAqB;sBACjC,EAAc;;;kCAGF,EAAS,SAAQ;;;;oCAIf,EAAoB;wBAChC,EAAa;;;oCAGD,EAAS,QAAO;;;;;;;;;;0BAU1B,EAAS;;;mCAIzB,GAAoB,EAAsB,EACA,EACvB,QAAQ,EAAoB,SAAO;;;;;;QAO1D,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;oCACgB,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;0CACT,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;kCAkBZ,EAAmB;;;;gCAIrB,EAAqB;oBACjC,EAAc;;;gCAGF,EAAS,SAAQ;;;;kCAIf,EAAsB;uCACjB,EAAa;;;;yCAIX,EAAa;6CACT,EAAa;6CACb,EAAa;;;cAG5C,EAAa;;;gCAGK,EAAsB;gBACtC,KAA6B,GAAC;;;;;;;;cAQhC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;;;;;cAKxC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;6CACT,EAAa;;;;cAI5C,EAAa;;;oBAGP,EAAW;;MAI/B,MAAA,MAEA,GAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MAE3B,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;;oBAEF,EAAW,KAAK,EAAY,KAAK,EAAW;mCAC7B,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBhC,EAAoB;sBAChC,EAAa;;;kCAGD,EAAS,QAAO;;;;oCAId,EAAqB;wBACjC,EAAc;;;oCAGF,EAAS,SAAQ;;;;sCAIf,EAAoB;0BAChC,EAAa;;;sCAGD,EAAS,QAAO;;;;;;;;;;4BAU1B,EAAS;;;qCAI3B,GACK,EACI,cAAc,EAAS,QAAO,YAC1B,EAAS,SAAQ,YAAY,EAAS,QAAO,YAC7C,EAAS,WAAU,QACvB,UAAU,EAAS,SAAQ,YACvB,EAAS,QAAO,YAAY,EAAS,WAAU,QACxD,QAAQ,EAAqB,MAAM,EAAoB;6BACxC,EAAoB,SAAO;;;;;;;QAQlD,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;;gBAEJ,EAAW,KAAK,EAAY,KAAK,EAAW;iCAC3B,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;;kCAmBZ,EAAmB;;;;gCAIrB,EAAoB;oBAChC,EAAa;;;gCAGD,EAAS,QAAO;;;;kCAId,EAAqB;oBACnC,EAAc;;;kCAGA,EAAS,SAAQ;;;;oCAIf,EAAsB;yCACjB,EAAa;;;;+CAIP,EAAa;mDACT,EAAa;mDACb,EAAa;;;gBAGhD,EAAa;;;kCAGK,EAAsB;kBACtC,KAA6B,GAAC;;;;;;;;gBAQhC,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;;;;;gBAK5C,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;mDACT,EAAa;;;;gBAIhD,EAAa;;;sBAGP,EAAW;;;MAKjC,MAAA,MCvcA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,OAAQ,EAAA,EAAA,QACtC,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAI,GAAsB,MACtB,EAAY,GAEhB,AAAI,IAAe,OACjB,EAAsB,MACjB,AAAI,IAAe,MAExB,GAAsB,cACtB,EAAY,OACP,AAAI,IAAe,OAExB,GAAsB,eACtB,EAAY,OAGd,GAAI,GAAiB,EAAU,IAAI,EAAU,IAAI,EAAU,qEAG3D,AAAI,IAAe,MACjB,EAAc,WACT,AAAI,IAAe,OACxB,EAAc,YACT,AAAI,IAAe,MACxB,EAAc,WACT,AAAI,IAAe,OACxB,GAAc,YAGhB,GAAM,GAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEzC,EAAgB;YACZ,KAAe,OAAK;;mBAEb,KAAe,QAAM;;;;wBAIhB,EAAS;;MAIzB,EAAU,OAEd,AAAI,IAAe,MACjB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SACL,AAAI,IAAe,OACxB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SAGZ,GAAI,GAAmB,GACvB,AAAI,EAAS,EAAa,GACxB,GAAmB;oCACW,EAAM;;;SAKtC,KAAK,SAAW;0CACsB,EAAmB;;;;UAInD,EAAgB;;;;;;;;kCAQQ,EAAU;;kCAEV,EAAmB;;;;;;8BAMvB,EAAqB;;YAEvC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;YAC/B,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCvKA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAuC,GANnD,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAGnB,OADI,GAAW,GACN,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAS,eACb,AAAI,EAAI,IAAM,GACZ,IAAU,kBAEZ,AAAI,EAAI,GACN,IAAU,kBAGZ,GAAY;UACR,EAAM;UACN,GAAI,EAAI,0CAA4C,IAAE;;;;;;mBAM7C,EAAC;;UAEV,GAAI,EAAI,IAAM,IAAE;QAItB,KAAK,SAAW;QACZ,GAAuB,GAAW;QAClC,GAA+B,GAAY;;;;;;;;qBAQ9B,EAAY,GAAE;qBACd,EAAY,GAAE;;UAEzB,EAAQ;;;;MAMlB,MAAA,MAEA,YAAgC,GAC9B,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;QAED,EAAsB;;;ICnF9B,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;;sDAMY,GAAU,GAAC;;;;;;qDAMZ,GAAS,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAgC/D,MAAA,MC1IA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MA+B/D,MAAA,MC3EA,AAmBA,GAAA,IAAA,WAOE,WACI,EAA8C,EAC9C,EAAkB,GARtB,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;uCACxB,EAAQ;;;;;;;;;;;;;;;;;;;;;;gCAsBf,GAAQ,GAAC;uCACF,GAAW,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAkDnD,MAAA,MCzHA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;sBAKpB,EAAe,GAAE;sCACD,EAAe,GAAE;;;wBAG/B,EAAe,GAAE;wCACD,EAAe,GAAE;;;4BAG7B,EAAO;kBACjB,EAAY;;;;4BAIF,EAAM;kBAChB,EAAY;;;;;;;;;;;;MAc9B,MAAA,MC/HA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAI5C,EAAY,EAAe,MAAQ,MAEzC,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;8DAaD,EAAS;;;;;;MAQvE,MAAA,MCpEA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAI5C,GAFA,KAAK,YAAc,EAEf,IAAS,GACX,KAAK,SAAW;;;2BAGK,EAAO,GAAE;;QAG9B,OAEF,GAAM,GAAa,SAAC,GAClB,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,aAAa,EAAC,QAE5B,UAAU,EAAC,KAEd,EAAW,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,KAAI,KAAK,KACpD,EAAO,GAAkB,GAE/B,KAAK,SAAW;;UAEV,EAAI;yBACW,EAAQ;;MAIjC,MAAA,MC1DA,AAqBA,GAAA,IAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAE5C,KAAK,YAAc,EACnB,GAAM,GAAW,GAAY,KAAM,GAC7B,EACC,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACrD,EAAa,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACjE,EAAO,GAAkB,GAC/B,AAAI,IAAS,EACX,KAAK,SAAW;;;;uCAIiB,EAAO,GAAE;cAClC,EAAO,GAAE;eACR,EAAU;2CACkB,EAAO,GAAE;kBAClC,EAAO,GAAE;;;;QAMrB,KAAK,SAAW;;YAEV,EAAI;;uBAEO,EAAK,EAAS,SAAQ;eAC9B,EAAU;yBACA,EAAK,EAAS,SAAQ;;eAEhC,EAAO;yBACG,EAAK,EAAS,SAAQ;iBAC9B,EAAU;2BACA,EAAK,EAAS,SAAQ;;;;;MAQ7C,WAAc,GACZ,MAAO,GAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QAChD,EAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAoB,GAClB,GAAM,GAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,EAAG,KACnD,EAAW,EAAc,KAAK,KAC9B,EAAY,EAAc,MAAM,IAAI,KAAK,KAC/C,MAAO,mBAAmB,EAAQ,WAAW,EAAS,KAGxD,WAAoB,EAAW,GAC7B,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,MAAM,EAAU,GAAE,OAE9B,GAAG,EAAU,IAI5B,MAAA,MC7GA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAoB,EAAkB,EACtC,EAAqB,EAAmB,EACxC,GAPJ,KAAA,cAAgB,CAAC,UAAW,UAAW,gBAQrC,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAClC,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,QAElB,GAAM,GAAiB,cAAc,EAAa,IAE9C,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,gBAElB,GAAM,GAAiB,cAAc,EAAa,IAE5C,EAAe,EAAW,EAAI,aAAe,UACnD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,EAAO;;;YAG7C,EAAK;;;gCAGe,EAAU;;kCAER,EAAQ;kCACR,EAAc;0CACN,EAAY;;;uBAG/B,EAAc;;;;;;QAQrC,MAAA,MCvEA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,IAAK,cAOpB,GAAM,GAAa,EAAU,WACvB,EAAY,EAAU,UACtB,EAAS,EAAU,OACnB,EAAc,EAAU,YACxB,EAAU,EAAc,KAAK,KAAK,EAAS,GACjD,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAM,GAAsB,MACtB,EAAc,WAEd,EAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEvC,EAAgB;;MAIlB,EAAwB,GAC5B,AAAI,EAAS,EAAa,GACxB,GAAwB;oCACM,EAAM;;;SAMtC,GAAI,GAA4B,GAChC,AAAI,EAAS,EAAa,GACxB,GAA4B;oCACE,EAAM;;;SAMtC,KAAK,SAAW;0CACsB,EAAmB;;;UAGnD,EAAqB;;;;;UAKrB,EAAyB;;;;;;;;;YASvB,EAAW,cAAc,EAAU;wDACS,EAAW;;;;8BAIrC,EAAqB;;;;;;;;;;;;;;;;YAgBvC,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;;;;;;;;;;;;;;;;;YAiB/B,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCjKA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAe,EAAiB,GAJ5C,KAAA,cAAgB,CAAC,IAAK,IAAK,KAKzB,KAAK,YAAc,EAEnB,GAAI,GACA,EACJ,GAAI,EAAO,EACT,KAAM,OAAM,kBAAkB,EAAI,yBAGpC,GAAI,IAAS,EACX,EAAW,QACX,EAAU,aAKV,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAClD,EAAa,GACb,EAAc,GACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAY,KAAK,GAAG,EAAc,IAClC,AAAI,EAAI,GACN,EAAW,KAAK,GAAG,EAAc,IAGrC,EAAU,EAAW,OACrB,EAAW,EAAY,OAGzB,GAAM,GAAQ,GAAkB,GAEhC,KAAK,SAAW;;UAEV,EAAK;4BACa,EAAO;;2BAER,EAAQ;;2BAER,EAAQ;;;MAKnC,MAAA,MCjEA,AAqBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,UASf,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAc,qBAAqB,KAAK,KAAI,KAC5C,EAAe,GAAU,KAAK,MAEhC,EACE,EAAW,EAAS,IAAI,SAAC,EAAG,GAChC,MAAO,aAAa,GAAO,GAAE,YAAY,EAAC,cAAc,GAAO,GAAE,MAEnE,EAAO;UACD,EAAK;UACL,EAAK;UACL,EAAS,KAAK;GAAK;QAEzB,KAAK,SAAW;QACZ,EAAW;;UAET,EAAI;8BACgB,EAAY;;MAKxC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KAEM,GAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,YAAmB,GACjB,GAAI,IAAS,EACX,MAAO,YACF,GAAI,GAAQ,EACjB,MAAO,IAAO,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAA,aAAe,IAAG,KAAK,KAE7D,KAAM,OAAM,oBAAoB,EAAI,yBCpFxC,AAsBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,UACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAS,GAAY,SAAU,KAAK,MACpC,EAAY,GAAY,YAAa,KAAK,MAE1C,EACF,KAAK,OAAS,EAAI,YAAc,QAAQ,EAAU,MAAM,IAAI,OAAM,IAChE,EACF,wBAAwB,EAAU,OAAM,MAAM,EAAS,IACrD,EAAW;mBACF,EAAU;cACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;YACnB,EAAU,KAAK,KAAO,GAAE;;MAG1B,EAAW,KAAK,OAAS,EAAI,GAAK;UAClC,EAAO,KAAK,KAAO,GAAE;cACjB,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;gBACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;cACpD,EAAU,KAAK,KAAO,GAAE;uBACf,EAAU;;;MAKvB,EAAiB,KAAK,MAAQ,EAChC;cACM,EAAK,IAAI,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,SAAS,EAAC,MAAK,OAAM,KAC3D,EAAS,IAAI,SAAC,EAAG,GAAM,MAAG,GAAU,GAAE,MAAM,EAAO,GAAE,YAAY,EAAC,OAC7D,KAAK;GACd,KAAK,SAAW;0BACM,KAAK,KAAI;;UAEzB,EAAK;UACL,EAAK;UACL,EAAc;;UAEd,EAAQ;UACR,EAAQ;;;MAMhB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KCtGA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAiB,EAAmB,GAJhD,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAK,OACZ,EAAa,GAAkB,EAAK,QACpC,EAAQ,GAAkB,EAAK,QAEjC,EAAY,GAChB,GAAI,IAAS,EACX,EAAY,gCAEZ,GAAI,GAAa,EACjB,EACI,EAAK,IAAI,SAAC,EAAG,GACP,WACO,EAAK,SAAW,EACnB,oBAAoB,EAAC,aAAa,EAAC,IACnC,UAAU,GAAa,GAAC,eAAe,EAAC,aAAa,EAAC,MAE3D,KAAK,KAGhB,KAAK,SAAW;QACZ,EAAU,YAAY,EAAU,IAAI,EAAK;QACzC,EAAU,cAAc,EAAU,IAAI,EAAO;;;UAG3C,EAAK;yBACU,EAAS;;MAIlC,MAAA,MCxDA,AAuBA,GAAA,IAAA,WAUE,WAAoB,GAAA,KAAA,MAAA,EATZ,KAAA,gBAAkB,EAClB,KAAA,gBAAkB,EAClB,KAAA,mBAAqB,EACrB,KAAA,cAAgB,EAEhB,KAAA,aAAkD,GAClD,KAAA,WAAa,GACb,KAAA,aAAkD,GAI1D,SAAA,UAAA,eAAA,SACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAkC,EAAO,GAE3D,EAAW,GAAuB,EAAS,EAAiB,GAClE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAEhC,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAS,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cACpD,GAEJ,GAAI,KAAK,aAAa,GAAU,OAAS,GACvC,KAAK,kBACL,KAAK,kBACL,KAAK,eAAiB,EACtB,KAAK,MACL,GAAM,GAAa,KAAK,aAAa,GAAU,QAC/C,YAAK,aAAa,GAAU,KAAK,GAC1B,EAGT,GAAI,GACJ,MAAI,KAAoB,GAAoB,mBAC1C,EAAa,KAAK,MAAM,0BAA0B,EAAQ,GAAI,EAAQ,IACjE,AAAI,IAAoB,GAAoB,mBACjD,EACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,IAC/D,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AACH,IAAoB,GAAoB,0BAC1C,GACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,KAEtE,KAAK,aAAa,GAAU,KAAK,GAEjC,KAAK,kBACL,KAAK,oBAAsB,EAC3B,KAAK,MAEE,GAGT,EAAA,UAAA,eAAA,SACI,EAAuB,EACvB,EAA8B,GAChC,GAAI,KAAK,cAAgB,KAEvB,OAEF,GAAM,GACF,GAAkC,EAAgB,GAChD,EAAW,GAAuB,EAAO,EAAiB,GAChE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAO,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cAClD,GACE,EAAqB,EAAA,MAAM,IAAI,kCACrC,AAAI,IAAuB,IACvB,KAAK,mBAAqB,EAC5B,MAAK,MAAM,oBAAoB,GAC/B,KAAK,oBAAsB,GAE3B,MAAK,aAAa,GAAU,KAAK,GACjC,KAAK,kBACL,KAAK,eAAiB,GAGxB,KAAK,kBAEL,GAAM,GAAU,KAAK,aAAa,GAC5B,EAAW,EAAQ,QAAQ,GACjC,GAAI,EAAW,EACb,KAAM,IAAI,OACN,4EAGN,EAAQ,OAAO,EAAU,GACzB,KAAK,OAGC,EAAA,UAAA,IAAR,WACE,GAAI,CAAC,KAAK,WACR,OAEF,GAAM,GAAQ,KAAK,gBAAkB,KAAK,gBAC1C,QAAQ,IACJ,YAAgB,KAAK,gBAAe,MAAM,KAAK,gBAC/C,IAAI,EAAK,KACb,GAAM,GAAY,KAAK,cAAgB,KAAK,mBAC5C,QAAQ,IAAI,oBAAoB,KAAK,oBACrC,QAAQ,IAAI,iBAAiB,KAAK,cAAa,KAC3C,KAAK,MAAM,IAAM,GAAU,OAGjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGd,OAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,+CAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,cAAgB,KAEvB,OAEF,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,KAAK,aAAe,KACpB,KAAK,aAAe,KACpB,KAAK,gBAAkB,EACvB,KAAK,gBAAkB,EACvB,KAAK,mBAAqB,EAC1B,KAAK,cAAgB,GAEzB,KAEA,YACI,EAA2B,GAE7B,GAAM,GAAQ,EACd,GAAI,IAAmB,EAAM,KAC3B,MAAO,GACF,GAAI,IAAmB,EAAM,KAClC,MAAO,GACF,GAAI,IAAmB,EAAM,QAClC,MAAO,IACF,GAAI,IAAmB,EAAG,KAC/B,MAAO,IACF,GAAI,IAAmB,EAAM,QAClC,MAAO,GAET,KAAM,IAAI,OAAM,2BAA2B,eAIzC,EAAyB,EACzB,EAA2B,EAC3B,GAMF,GAAM,GACF,GAAiC,EAAiB,GAElD,EACJ,GAAI,GACI,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAEpB,EAAc,EAAc,OAGtB,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,EAAc,EAAQ,EAGxB,GAAM,GAAkB,GAA0B,EAAI,GACtD,MAAO,GAAc,EAGvB,YACI,EACA,GACF,OAAQ,OACD,IAAoB,mBACvB,MAAO,IAAwC,OAC5C,IAAoB,mBACvB,MAAO,IAA+C,OACnD,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,yBACvB,MAAO,IAA+C,WAEtD,KAAM,IAAI,OAAM,iCAAiC,IAIvD,YAAwC,GAEtC,MAAI,GAAA,MAAM,QAAQ,gCACZ,EACK,GAAoB,mBAEtB,GAAoB,iBAGzB,EACK,GAAoB,mBAEtB,GAAoB,iBAG7B,YACI,EAA8B,GAChC,GAAI,IAAmB,GAAa,OAClC,MAAO,IAAoB,mBACtB,GAAI,IAAmB,GAAa,QAAU,GAAkB,KACrE,MAAO,IAA+B,GACjC,GACH,IAAmB,GAAa,UAChC,IAAmB,GAAa,OAClC,MAAO,IAAoB,yBAE7B,KAAM,IAAI,OAAM,gCAAgC,GAGlD,YACI,EAAgC,EAChC,GACF,MAAU,GAAa,GAAE,IAAI,EAAa,GAAE,IAAI,EAAe,IAAI,ECxRrE,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,GAEpC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,GAErC,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,GACvB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,iBAAiB,EAAI,yBAEnC,GAAI,IAAS,EACX,MAAO,eAAe,EAAO,GAAE,IAMjC,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,UAAW,WAE7D,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAa,KAAK,QAAQ,EAAc,GAAE,KAAK,EAAO,GAAE,KAE1D,MAAO,GAAa,OC5DtB,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MAEM,GAAoB,0BAEb,GAAS,YAET,GAAM,iBAEN,GAAO,GAAoB;;EAI3B,GAAQ,GAAoB;;EAI5B,GAAM,0CAEN,GAAO;;;uBAGG,EAAA,aAAa,gBAAe;kBACjC,EAAA,aAAa,WAAU;;cAIpB,GAAA,MAAA,KAAA,QAAA,GAAA,GACZ,GAAoB;mCACM,EAAK;KAIjC,GAAM,IAAM,aAEN,GAAO,kBAEP,GAAQ,mBAER,GAAO;;;EAKP,GAAS,0BAET,GAAS,0BAET,GAAY,wCAEZ,GAAQ;;;;;;;;;;;;;;;EAiBR,GAAM,iBAEN,GAAQ,uBAER,GAAM;kBAGN,GAAQ,uBAER,GAAO,kBAEP,GAAQ,yBAER,GAAU,sCAeV,GAAW;;;;;;;;;;;;;;;;;;;;EAsBX,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;EAI3B,GAAO;;;EAKP,GAAO;;;EAKP,GAAO;;;EAKP,GAAQ,GAAoB,qCAE5B,GAAQ,GAAoB;;sCAI5B,GAAQ,GAAoB;;+CAI5B,GAAM;;;;cAIL,EAAA,aAAa,MAAK;eACjB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;;;;;;EAQrB,GAAa,kBAEb,GAAc,6BAEd,GAAS,wBAET,GAAQ,YC9NrB,AAmBO,GAAM,IAAS,YAET,GAAM;;;;;;;;;EAWN,GAAO;;;;;;;;;;EAYP,GAAQ;;;;;;;;;;EAYR,GAAM;;;;;;;;;EAWnB,GAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MCzFA,AAqBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEnB,EAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EAAe,GAAgB,EAAM,GACrC,EAAY,EAAS,MAAM,IAC3B,EAAS,GAAQ,EAAI,KAAO,QAAQ,EAAU,KAAK,KAAI,IAE7D,KAAK,SAAW;;UAEV,EAAK;kCACmB,EAAY;;4CAEF,EAAM;;MAIlD,MAAA,MC/CA,AA2BO,GAAA,IAAA,EAAA,aAAA,aACD,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAiFlB,GAAkB,KAClB,GAAkB,KA4BzB,GAAuE,eAE9C,GAC7B,MAAI,KAAgB,KAGpB,IAAa,GAAgB,IACtB,GAAa,GAGtB,YACI,EAAqC,GACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACnC,IAAe,SACjB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,OACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,MACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GAET,KAAM,IAAI,OAAM,cACZ,EAAU,oDAKhB,GAAM,IAA6B,IAK7B,GAAyB,IAC/B,cACE,MAAI,GAAA,MAAM,OAAO,QAAU,KAClB,KAED,EAAA,MAAM,OAAO,OAAO,OAAS,EAAA,MAAM,OAAO,OAAO,MACjD,OAAO,iBACX,GAAyB,KAAO,KAM/B,GAAM,IAA8B,mBAEL,GAAA,EAAA,GAkCpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GA/BM,EAAA,YAAc,GAAI,SAGlB,EAAA,gBAAkB,GAAI,SAGtB,EAAA,aAAe,GAAI,SACnB,EAAA,cAAgB,EAOhB,EAAA,aAAe,EAEf,EAAA,eAAiB,EAUjB,EAAA,kBAAoB,GACpB,EAAA,sBAAwB,GAgXxB,EAAA,eAAiB,EAq+DjB,EAAA,SAAW,GAj1Eb,CAAC,EAAA,MAAM,QAAQ,aACjB,KAAM,IAAI,OAAM,yCAGlB,GAAI,GAAS,MACX,GAAM,GAAK,GAAgB,EAAA,MAAM,UAAU,kBAC3C,EAAK,YAAc,GAAe,EAAA,MAAM,UAAU,kBAClD,EAAK,MAAQ,GAAI,IAAa,GAC9B,EAAK,OAAS,EAAG,OACjB,EAAK,oBAAsB,OAE3B,GAAK,MAAQ,EACb,EAAK,YAAc,GACnB,EAAK,oBAAsB,GAC3B,EAAK,OAAS,EAAM,GAAG,OAEzB,SAAK,eAAiB,GAAI,IAAe,EAAK,OAC9C,EAAK,mBAAqB,KAE1B,EAAK,QAAU,GAAI,GAAA,YAAY,EAAM,EAAA,YAGvC,SAAA,UAAA,WAAA,WACE,MAAO,MAAK,QAAQ,aACf,MAAK,WAAa,KAAK,WAAW,aAAe,GAClD,KAAK,gBAGX,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAK5C,GAJA,AAAI,GAAA,MAAM,QAAQ,mCACd,EAAA,MAAM,QAAQ,WAChB,KAAK,uBAAuB,GAE1B,IAAU,aAAe,GAAU,KACrC,KAAM,IAAI,OACN,yEAGN,GAAM,GAAS,GACf,YAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,IAC1D,GAIT,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,YAIV,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,aAIZ,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAK3D,GAHA,AAAI,EAAA,MAAM,QAAQ,UAChB,KAAK,uBAAuB,GAE1B,IAAU,YACZ,KAAM,IAAI,OACN,yEAGN,KAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,KAGnE,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAc,KAAK,QAAQ,IAAI,GAErC,EAAY,WAEZ,AAAI,EAAY,SAAW,GACzB,KAAK,YAAY,KAKvB,EAAA,UAAA,SAAA,SAAS,GACP,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,eAAgB,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SAKpD,GAAI,GAAS,MACX,GAAI,GAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEtC,GAAM,GACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,SAAS,EAAI,QAC/B,YAAK,8BAA8B,GAC5B,EAET,GAAI,GAAU,KACZ,MAAO,MAAK,qBAAqB,GAEnC,GAAI,IAAU,SACZ,MAAO,GAET,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GACJ,GAAI,IAAU,aACZ,GAAM,GAAa,EAAe,KAAK,WACjC,EAAa,EAAe,KAAK,WACvC,EAAS,EAAA,aAAa,uBAAuB,EAAY,OAEzD,GAAS,KAAK,qBAAqB,GAGrC,MAAI,IACF,MAAK,gBAAkB,EAAA,KAAK,MAAQ,GAE/B,KAAK,qBAAqB,EAAQ,IAGrC,EAAA,UAAA,KAAN,SAAW,6IACT,GAAI,KAAK,YAAY,IAAI,GACjB,SAAc,KAAK,YAAY,IAAI,GACzC,CAAA,EAAO,GAAI,SAAoB,SAAA,GAAW,MAAA,GAAY,KAAK,MAQ7D,GANM,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAyD,EAAO,OAAxD,EAAiD,EAAO,MAAjD,EAA0C,EAAO,MAA1C,EAAmC,EAAO,MAAnC,EAA4B,EAAO,eAAnB,EAAY,EAAO,SAKnE,GAAS,KACP,SAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEhC,EACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,KAAK,EAAI,QAC3B,KAAK,8BAA8B,GACnC,CAAA,EAAO,GAGT,GAAI,GAAU,KACZ,MAAA,CAAA,EAAO,KAAK,qBAAqB,IAGnC,GAAI,CAAC,EAAA,MAAM,QAAQ,iCACf,EAAA,MAAM,UAAU,mBAAqB,EACvC,KAAM,IAAI,OACN,sGAIF,GAAS,KAGb,AAAI,IAAU,aAAe,EAAA,MAAM,IAAI,2BAErC,GAAoB,KAAK,OAAO,GAC1B,EAAU,KAAK,QAAQ,IAAI,EAAkB,QAEnD,EAAS,GAAA,KAAK,OAAM,wBAAuB,MAAA,EAAA,CACvC,EAAQ,SAAO,OAAK,GAA0B,MAGpD,KAAK,YAAY,IAAI,EAAQ,IAEzB,IAAU,YAEZ,CAAA,EAAM,KAAK,MAAM,yBAFf,CAAA,EAAA,UAEF,EAAA,8BAKE,KAAU,YACD,CAAA,EAAM,QAAQ,IACrB,CAAC,EAAe,KAAK,OAAQ,EAAe,KAAK,UAFnD,CAAA,EAAA,UACI,SAAK,EAAA,OAEL,EAAa,EAAG,GAChB,EAAa,EAAG,GACtB,EAAO,EAAA,aAAa,uBAChB,EAA4B,gBAC3B,AAAI,GAAU,KACnB,EAAO,KAAK,qBAAqB,GAE3B,GAAO,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,MAAM,gCAAgC,EAAQ,qBAE5D,MAAI,IAAqB,MACvB,KAAK,8BAA8B,GAE/B,EAAY,KAAK,qBAAqB,EAAQ,GAE9C,EAAc,KAAK,YAAY,IAAI,GACzC,KAAK,YAAY,OAAO,GAGxB,EAAY,QAAQ,SAAA,GAAW,MAAA,GAAQ,KACvC,AAAI,KAAK,gBAAgB,IAAI,IAC3B,MAAK,gBAAgB,OAAO,GAC5B,KAAK,YAAY,GACjB,KAAK,kBAEP,CAAA,EAAO,SAGD,EAAA,UAAA,uBAAR,SAA+B,GAC7B,GAAI,GAAU,KACZ,OAEF,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,KACjC,GAAM,GAAM,EAAO,GACnB,GAAI,CAAC,GAA4B,GAC/B,KAAI,GAAA,MAAM,QAAQ,gCACV,MACF,aAAa,EAAG,iJAIhB,MAAM,aAAa,EAAG,4CAK1B,EAAA,UAAA,qBAAR,SAA6B,SACrB,EAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SACf,EAAO,EAAA,KAAK,cAAc,GAChC,GAAI,EAAA,MAAM,QAAQ,iCAChB,GAAM,GAAY,KAAK,OAAO,GACxB,EAAU,KAAK,QAAQ,IAAI,EAAU,QACrC,EAAO,GAAA,KAAK,OACA,gCAA+B,MAAA,EAAA,CAC5B,EAAQ,SAAO,OAAK,GAA0B,KACjD,SAAS,EAAG,GAE9B,YAAK,8BAA8B,GAE5B,EAGT,GAAM,GACF,EAAA,MAAM,QAAQ,eAAiB,IAAa,GAC1C,EACF,EAAyB,GAAwB,GAAS,EACxD,EAAU,EACZ,GAAI,IAAyB,GAC7B,GAAI,IAAmB,GACrB,EAAS,KAAK,gBAChB,EAAS,CAAC,CAAC,MAAO,EAAa,MAAK,EAAE,OAAM,IAAI,WAC9C,EAAU,KAAK,QAAQ,IAAI,EAAO,QAClC,EACF,KAAK,MACA,gDACG,EAAQ,QAAS,EAAQ,SAAS,GAAI,EAAQ,SAAS,IAC1D,SAAS,EAAG,GACrB,YAAK,8BAA8B,GAE5B,GAGH,EAAA,UAAA,KAAN,SAAW,qHACH,GAAkB,KAAK,aACvB,EAA+B,GAEjC,EAAgB,GACpB,AAAI,KAAK,oBAAsB,KAC7B,MAAK,mBAAqB,EAC1B,EAAgB,IAEhB,KAAK,aAAa,KAAK,GAEzB,KAAK,aAAe,EAEpB,IAGM,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,SACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OACpB,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,QACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OAE1B,KAAK,aAAe,EAEpB,AAAI,GACF,MAAK,mBAAqB,MAGtB,EAAuB,CAC3B,aAAc,KAAK,aACnB,eAAgB,KAAK,eACrB,SAAU,KACV,OAAQ,MAGN,EAAA,MAAM,UAAU,iDAAmD,EACpD,CAAA,EAAM,QAAQ,IAAI,IADjC,CAAA,EAAA,UACI,SAAW,EAAA,OAEjB,EAAI,SAAc,EAAA,KAAK,IAAI,GAC3B,EAAI,oBAAyB,WACzB,MAAA,GAAS,IAAI,SAAC,EAAG,GAAM,MAAC,CAAC,KAAM,EAA0B,GAAI,GAAI,KAC5D,IAAI,SAAA,GAAK,MAAG,GAAE,KAAI,KAAK,EAAE,KACzB,KAAK,oBAEd,EAAI,SAAc,CAChB,MAAO,8EAIX,YAAK,aAAe,EACpB,KAAK,eAAiB,EACtB,CAAA,EAAO,SAET,EAAA,UAAA,OAAA,WACE,MAAO,CACL,WAAY,GACZ,cAAe,KAAK,cACpB,uBAAwB,KAAK,eAAe,kBAC5C,kBAAmB,KAAK,eAAe,eAInC,EAAA,UAAA,WAAR,WACE,MAAI,GAAA,MAAM,UAAU,iDAAmD,EAC9D,KAAK,MAAM,aAEb,CAAC,QAAS,EAAA,KAAK,MAAO,MAAO,OAG9B,EAAA,UAAA,SAAR,SAAiB,GACf,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,MAAK,MAAM,WACJ,GAER,GAAwB,MAAQ,EAAA,KAAK,MAC/B,IAGK,EAAA,UAAA,aAAd,SAA2B,4EACzB,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,CAAA,EAAO,KAAK,MAAM,uBAAuB,IAErC,GAAa,EACnB,CAAA,EAAO,EAAW,MAAQ,EAAW,eAKvC,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,gBAAgB,IAAI,GAC3B,OAEF,GAAI,KAAK,YAAY,IAAI,IACvB,KAAK,gBAAgB,IAAI,GACzB,KAAK,iBACL,OAGF,GAAI,CAAC,KAAK,QAAQ,IAAI,GACpB,OAGF,KAAK,eAAe,GACb,GAAA,GAAA,KAAA,QAAA,IAAA,GAAA,eACP,AAAI,GAAkB,MACpB,GAAe,KAAK,UACpB,EAAe,KAAK,WAEtB,KAAK,QAAQ,OAAO,IAGd,EAAA,UAAA,eAAR,SAAuB,GACf,GAAA,GAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAE5C,EAAM,GAAS,EAAM,YAAc,EACnC,EAAW,KAAK,aAAa,IAAI,GACvC,AAAI,EAAW,EACb,KAAK,aAAa,IAAI,EAAK,EAAW,GAEtC,MAAK,aAAa,OAAO,GACzB,AAAI,GAAW,MACb,MAAK,eAAiB,KAAK,aAAa,EAAU,GAClD,KAAK,eAAe,eAAe,EAAS,EAAU,EAAO,KAGjE,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,QAAU,KAClB,EAAQ,SAAW,KACnB,EAAQ,SAAW,GACnB,EAAQ,MAAQ,MAGlB,EAAA,UAAA,WAAA,SAAW,GACT,YAAK,YAAY,GACV,KAAK,QAAQ,IAAI,GAAQ,SAOlC,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,MAAK,QAAQ,IAAI,IAGlB,EAAA,UAAA,cAAR,WACE,MAAK,GAAA,MAAM,QAAQ,qBAInB,CAAI,KAAK,YAAc,MACrB,MAAK,WAAa,EAAA,SAAS,YAAY,QAGlC,KAAK,YAPH,MAiBX,EAAA,UAAA,mBAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAa,KAAK,gBACxB,MAAI,CAAC,KAAK,uBAAyB,GAAc,MAC/C,SAAQ,KACJ,6NAKJ,KAAK,sBAAwB,IAGxB,GAAc,MACjB,EAAO,MACH,SAAA,GAAS,MAAA,GAAK,QAAQ,IAAI,EAAM,QAAQ,SAAW,MAC/C,EAAA,KAAK,cAAc,EAAM,OAAS,KAGhD,EAAA,UAAA,gBAAA,WACE,MAAO,MAAK,OAGd,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,GAAM,GAAS,KAAK,WAAW,EAAK,MAAO,aACrC,EAAa,KAAK,QAAQ,IAAI,EAAO,QAI3C,SAAW,eAAiB,CAC1B,KAAM,EAAA,SAAS,KAAK,EAAK,SACzB,KAAM,EAAA,SAAS,KAAK,EAAK,UAGpB,GAET,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAExC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAGxC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAO,EAAM,EAAE,MAChE,EAAE,OACN,MAAO,MAAK,WAAW,EAAM,EAAE,MAAO,GAGxC,GAAI,EAAA,KAAK,cAAc,KAAU,EAC/B,MAAO,GAAA,OAAO,GAAI,EAAM,EAAE,OAErB,GAAA,GAAA,KAAA,QAAA,IAAA,EAAA,QAAA,SACD,EAAc,EAAA,WAAW,iBAAiB,EAAE,MAAO,EAAO,GAChE,GAAI,GAAY,CAAC,GACf,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAmB,GACvB,GAAI,IAAa,GACf,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,GAEhD,YAAK,YAAY,EAAE,QACZ,KAAK,aAAa,EAAG,EAAO,IAG7B,EAAA,UAAA,aAAR,SAAqB,EAAW,EAAiB,GAC/C,GAAM,GAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAI,KAAK,WAAW,EAAM,EAAE,OAC5B,EAAa,KAAK,QAAQ,IAAI,EAAE,QAEtC,OAAO,OAAO,EAAY,GAC1B,EAAW,MAAQ,EACnB,EAAW,MAAQ,EAAE,MACrB,GAAI,GAAa,EAAA,WAAW,kBAAkB,EAAO,EAAE,SACvD,AAAI,EAAS,OAGX,IAAc,EAAS,MAAM,YAE/B,EAAW,MAAQ,CACjB,WAAU,EAEV,WAAY,EAAS,OAAS,EAAS,MAAM,YAAc,EAAE,QAI/D,GAAM,GAAW,KAAK,aAAa,IAAI,EAAW,MAAM,aAAe,EACvE,YAAK,aAAa,IAAI,EAAW,MAAM,WAAY,EAAW,GAEvD,GAGT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GAD1C,GAAA,GAAA,KAEQ,EAAS,KAAK,mBAChB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,aAAa,EAAG,EAAO,EAAK,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAO,GAAI,GAGpB,GAAM,GAAU,GAAI,IAAoB,EAAO,EAAS,GACxD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAqB,EAAE,MAAO,GAClC,GAAI,IAAe,EAAE,MAAO,GAChC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,GAAI,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KAChC,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KACtC,MAAO,GAAA,QAAQ,KAAK,OAAO,EAAO,GAAO,KAAK,OAAO,EAAO,IAG9D,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAEjB,GAAI,EAAQ,OAAS,EAAA,MAAM,UAAU,iCACnC,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,OAAO,EAAQ,MAAM,EAAG,GAAW,GACnD,EAAY,KAAK,OAAO,EAAQ,MAAM,GAAW,GACvD,MAAO,MAAK,OAAO,CAAC,EAAU,GAAY,GAE5C,GAAI,EAAA,MAAM,QAAQ,gCAAkC,EAAQ,GAAG,KAAO,GACpE,GAAM,GAAU,GAAI,IAAoB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACnE,MAAO,MAAK,cAAc,EAAS,GASrC,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACtD,EACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAAK,GAAI,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,OAC3D,EAAU,GAAI,IAAc,EAAU,IAAI,SAAA,GAAK,MAAA,GAAE,SACjD,EAAc,KAAK,cAAc,EAAS,GAChD,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAsB,GAAtB,GAAA,GAAA,KACQ,EAAS,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,IAAI,KACtE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAE/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAM,GAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAA,EAAA,MAAC,EAAA,EAAA,GAIP,GAAK,KAAgB,GAAK,IAAgB,IACtC,EAAY,IACd,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAE1B,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAG1B,GAAM,GAAM,IAAgB,EAAI,EAAI,EAAE,KAAK,EAAO,EAAW,GACvD,EAAO,IAAgB,EAAI,EAAI,EAC/B,EAAM,IAAgB,EAAI,EAAE,KAAK,EAAO,EAAG,GAAa,EAC9D,MAAO,MAAK,SAAS,EAAK,GAAK,IAAI,EAAM,IAG3C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,GAC5D,MAAO,MAAK,cAAwB,EAAS,CAAC,EAAG,GAAI,IAGvD,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE7C,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAA,EAAA,MAAC,EAAA,EAAA,GAED,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,EACxD,EAAS,EAAiB,GACxB,EAAuB,CAAC,EAAG,GACjC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAwB,EAAS,EAAQ,IAGvD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aACd,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OACrD,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OAErD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAExD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAE/C,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,EAGT,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,EAAE,QAG/C,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAQ,EAAM,EAAO,GACnD,GAAI,IAAW,EAAE,MAAO,EAAQ,EAAM,EAAO,GACjD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAM,GACF,GAAI,IAAe,EAAW,MAAO,EAAa,EAAM,EAAO,GACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAY,EAAa,KAG/D,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAI,EAAE,QAAU,UACd,GAAM,GAAO,KAAK,SAAS,EAAE,QACvB,EAAc,EAAK,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,KAC9C,EAAM,EAAA,OAAO,EAAE,MAAO,EAAE,MAAO,GACrC,MAAO,IAAK,EAAK,GAEnB,GAAM,GAAU,GAAI,IAAY,EAAE,MAAO,GACzC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAU,GACxC,GAAI,IAAW,EAAE,MAAO,EAAU,GACtC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAAlD,GAAA,GAAA,KACQ,EAAS,KAAK,mBAChB,CAAC,EAAG,GAAU,WAAM,MAAA,GAAK,WAAW,OAAO,EAAG,EAAS,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAQ,KAAM,GACzD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAEV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAU,EAAE,QAAQ,GAAW,GAC1B,QAAQ,GACR,MAAM,EAAkB,IAGtC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAGV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAAyB,GACzB,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,EAAE,IAAI,GAEhB,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EAAe,EAAA,aAAa,oBAC9B,EAAQ,MAAO,EAAY,EAAM,IAE/B,EAAW,EAAA,UACb,EAAQ,QAAQ,GAChB,GACJ,MAAO,GAAA,QAAQ,EAAU,IAGnB,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,KAAK,KAAK,EAAS,GAC7B,EAAa,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,QAAO,GACpD,EAAU,GAAI,IAAc,EAAY,GACxC,EAAS,KAAK,cAAwB,EAAS,CAAC,GAAI,GAE1D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,OAAO,EAAQ,EAAY,IAGjC,EAAA,UAAA,UAAR,SACI,EAAa,EACb,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACrB,AAAI,GAAgB,MAClB,GAAY,EAAa,MAAM,GAC/B,EAAS,EAAa,MAAM,IAE9B,GAAM,GAAa,EAAA,aAAa,yBAAyB,GACnD,EAAa,CACjB,WAAU,EACV,OAAM,EACN,UAAS,EACT,QAAS,KAAK,KAAK,EAAS,IAExB,EACF,GAAI,IAAiB,EAAY,EAAY,GAAgB,MAC3D,EAAS,CAAC,GAChB,AAAI,GAAgB,MAClB,EAAO,KAAK,GAEd,GAAM,GAAS,KAAK,cAAwB,EAAS,EAAQ,SAE7D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,UAAU,EAAG,EAAY,IAG/B,EAAA,UAAA,gBAAR,SACI,EAAW,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MACtC,GAAM,GAAU,GAAgB,KAAO,EAAa,MAAQ,EAAE,MACxD,EAAS,EAAQ,EAAQ,OAAS,GAClC,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,GAAI,IAChB,EAAS,EAAY,EAAY,GAAgB,MAC/C,EAAS,GAAgB,KAAO,CAAC,GAAK,CAAC,EAAG,GAC1C,EAAS,KAAK,cAAsB,EAAS,EAAQ,SAC3D,MAAI,GAAO,OAAS,EAAE,KACb,KAAK,gBAAgB,EAAG,EAAY,GAEtC,GAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,MAAO,GAAa,QAAQ,IAGtD,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAC/D,GAAI,EACF,MAAO,GAGH,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,OAAQ,GAAa,QAAQ,IAGvD,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAI,GAAO,EACL,EAAc,EAAA,aAAa,mBAAmB,CAAC,GAAO,EAAE,MAC1D,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,EAAA,UAAU,EAAG,GACzB,EAAO,EAAA,aAAa,iBAAiB,EAAG,EAAE,MAAM,IAGlD,GAAM,GACF,GAAa,gBAAgB,EAAU,MAAO,EAAM,GAClD,EAAS,EAAA,KAAK,cAAc,CAAC,EAAU,MAAM,KAC7C,EAAM,EAAU,KAAK,GAAI,GACzB,EAAc,EAAA,WAAc,EAAE,OAChC,EACA,KAAK,aACG,EAAK,qBAAsB,EAAY,EAAa,GACvD,QAAQ,GACjB,MAAI,IAAe,MACjB,GACI,EAAA,UAAU,EAAQ,EAAA,aAAa,uBAAuB,KAErD,GAGD,EAAA,UAAA,aAAR,SACI,EAAa,EAAiC,EAC9C,EAAiB,GACnB,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EACF,GAAa,8BAA8B,EAAQ,GACjD,EAAY,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,YAAW,GACvD,EAAU,GAAI,IAAiB,EAAW,GAC1C,EACF,KAAK,cAAwB,EAAS,CAAC,EAAG,GAAa,GAE3D,MAAI,GAAO,MAAM,KAAO,EACf,EAET,GAAa,EAAA,MAAM,EAAG,GAAa,KAAK,CAAC,EAAS,IAC3C,KAAK,aAAa,EAAQ,EAAW,EAAY,EAAO,KAGzD,EAAA,UAAA,gBAAR,SAAwB,EAAW,EAAc,GAE/C,GAAM,GAAO,CAAC,GAId,GAHA,EAAA,aAAa,2BACT,MAAQ,EAAW,OAAO,GAAG,cAAgB,EAAW,MAAM,GAAI,EAClE,EAAE,MACF,CAAC,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAQ,GAC7C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,UAAU,EAAK,GAAY,QAAQ,GAEjD,MAAO,MAAK,gBAAgB,EAAG,IAGjC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,GAAI,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,kDAAkD,GAAE,KAAO,GAAC,IAC5D,iBAAgB,IAMtB,OAJM,GAAO,EAAE,MAAM,GACjB,EAAS,EAGJ,EAAI,EAAG,GAAK,KAAK,KAAK,KAAK,KAAK,IAAS,EAAG,KACnD,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,GAAO,GAC5C,EAAc,EAAQ,mBAAmB,GACzC,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,GAAS,EAAO,MAAO,GAC7D,EAAW,UAIb,GAAI,GACF,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAW,GAChD,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,IACtC,EAAW,UAGb,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA2B,QAE9D,GAAM,GAAU,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAE,OACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA+B,QAElE,GAAM,GACF,GAAI,IAAgB,GAAwB,EAAE,MAAO,EAAE,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAClE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA0B,QAG7D,GAAM,GAAU,GAAI,IAAgB,GAAmB,EAAE,MAAO,EAAE,OAClE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA6B,QAGhE,GAAM,GAAU,GAAI,IAAgB,GAAsB,EAAE,MAAO,EAAE,OACrE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAAmC,QAE/C,GAAM,GACF,GAAI,IAAgB,GAA4B,EAAE,MAAO,EAAE,OAC/D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAiC,QAEpE,GAAM,GACF,GAAI,IAAgB,GAA0B,EAAE,MAAO,EAAE,OAC7D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAM,GAAU,GAAI,IAAc,EAAU,KAAM,EAAE,MAAO,EAAE,MAC7D,MAAO,MAAK,cACR,EAAS,CAAC,EAAW,EAAG,GAAI,EAAA,WAAW,EAAE,MAAO,EAAE,SAGxD,EAAA,UAAA,MAAA,SAAM,GACJ,EAAA,aAAa,KACT,yEAEJ,GAAM,GAAW,EAAU,WAC3B,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAM,GAAQ,EAAE,WAChB,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAK,GACL,EAAc,QACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAA6B,GAEzC,GAAM,GAAU,GAAI,IAAgB,EAAI,EAAE,MAAO,EAAE,OACnD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAG1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,GAErD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAG7C,EAAA,UAAA,cAAR,SAAsB,EAAe,EAAY,GAC/C,GAAM,GAAU,GAAI,IAAqB,EAAE,MAAO,GAClD,MAAO,MAAK,cAAsB,EAAS,CAAC,GAAI,IAG1C,EAAA,UAAA,eAAR,SACI,EAAe,EAAe,EAAY,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GACF,GAAI,IAAsB,EAAI,EAAE,MAAO,EAAE,MAAO,GACpD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAO7C,EAAA,UAAA,yBAAR,SAAiC,EAAW,EAAW,GAAvD,GAAA,GAAA,KACQ,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAA,4SAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAcP,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,GAMD,EAAA,UAAA,+BAAR,SACI,EAAuB,GACzB,MAAO,CACL,OAAQ,EAAY,OACpB,MAAO,EAAY,MACnB,MAAO,EAAc,QAIzB,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAIjB,GAAI,EAAQ,OAAS,EAAA,MAAM,IAAI,iCAC7B,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,KAAK,EAAQ,MAAM,EAAG,IACtC,EAAY,KAAK,KAAK,EAAQ,MAAM,IAC1C,MAAO,MAAK,KAAK,CAAC,EAAU,IAG9B,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAO,OAAO,SAAC,EAAI,GAAO,MAAA,GAAA,WAAW,EAAI,KAC1D,EAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAE5B,EAAc,EAAA,MAAM,QAAQ,cAC5B,EAAU,EACZ,GAAI,IAAkB,EAAQ,GAAG,MAAO,GACxC,GAAI,IAAY,EAAQ,GAAG,MAAO,GACtC,MAAO,MAAK,cAAiB,EAAS,EAAS,IAGjD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,GAAM,GAAc,EAAA,MAAM,QAAQ,gCAC5B,EAAU,EACZ,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OAC/C,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,MAAO,MAAK,cAAiB,EAAS,CAAC,EAAG,GAAI,IAGhD,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAY,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACnE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAe,EAAE,OAGhD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAG1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAChD,EAAI,KAAK,SAAS,EAAQ,EAAS,QAAQ,IAC3C,EAAI,KAAK,IAAI,GACb,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAI,EAAG,IAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAGtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAE3C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA2B,EAAE,MAAO,EAAM,OAC9C,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAM,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAEtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA6B,EAAG,MAAO,EAAE,OAC7C,GAAI,IAAgB,GAAsB,EAAG,MAAO,EAAE,OAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,UAG1C,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,GAAI,GACJ,AAAI,EAAA,MAAM,QAAQ,mBAChB,EAAU,GAAI,IAAkB,EAAE,OAElC,EAAU,GAAI,IAAY,EAAE,OAE9B,GAAM,GAAc,EAAQ,mBAAmB,EAAK,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,IAGhD,EAAA,UAAA,IAAA,SAAsB,GAEpB,GAAI,KAAK,mBAAmB,CAAC,KAAO,EAAE,QAAU,aAC9C,GAAM,GACF,GAAiB,KAAK,QAAQ,IAAI,EAAE,QAAQ,QAChD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAU,GAAI,IAAkB,EAAE,OAClC,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAG9D,MAAO,MAAK,cAAsB,EAAS,IAG7C,EAAA,UAAA,QAAA,SAA0B,GACxB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,GAAc,IAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,KAG9B,EAAA,UAAA,eAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAGF,GAAM,GAAS,EAAE,MACX,EAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAkB,EAAS,WAC3B,EAAc,EAAO,GAAK,EAAO,GAAK,EAAO,GAC7C,EAAmB,EAAS,YAC5B,EAAiB,EAAS,aAAe,eACzC,EAAa,GACb,EAAa,GAIb,EACD,KAAgB,GAAK,IAAqB,IAC3C,EAAkB,GAChB,EAAyB,EAAO,GAAK,IAAM,GAAK,CAAC,CAAC,EAAS,SAEjE,GAAI,GAA6B,CAAC,EAAA,MAAM,QAAQ,wBAC5C,CAAC,EAAA,MAAM,QAAQ,iCACf,CAAC,GACH,GAAM,GAAc,EAAiB,EAAO,GAAK,EAAO,GAAK,EAAO,GAC/B,EAAO,GAAK,EAAO,GAAK,EAAO,GAC9D,EAAY,EAAA,QAAQ,EAAG,CAAC,EAAG,EAAa,EAAS,aACjD,EACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAS,KAAK,iBAAiB,CACnC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAExB,MAAO,GAAA,QAAQ,EAAQ,EAAS,UAWlC,GAAM,GAAc,EAChB,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACrC,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACnC,EAAwB,CAC5B,OAAQ,EAAE,OACV,MAAO,CAAC,EAAG,EAAa,EAAS,YACjC,MAAO,EAAE,OAUL,EAAwB,EAAS,MACvC,EAAS,MAAQ,EAAS,MAAM,QAChC,EAAS,MAAM,EAAS,MAAM,OAAS,KACvC,EAAA,KAAK,OACD,GAAyB,EAAS,MAAO,EAAU,OACnD,WAAM,MAAA,kBAAkB,EAAS,MAAK,OAClC,EAAU,MAAK,gBACvB,GAAM,GACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAgB,KAAK,iBAAiB,CAC1C,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAElB,EAAuB,KAAK,QAAQ,IAAI,EAAc,QAC5D,SAAA,KAAK,OACD,EAAqB,SACrB,WAAM,MAAA,gDAEV,EAAS,MAAQ,EAGjB,EAAqB,MAAQ,EAAS,SAC/B,EAAA,SAAS,qBACL,EAAc,OAAQ,EAAS,SAAU,EAAc,QAI5D,EAAA,UAAA,iBAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAQA,GAAA,GAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,WAGI,EAAiB,IAAe,eAEhC,EAAY,EAAc,EAAe,EACzC,EAAU,EAAY,EACtB,EAAa,CAAC,EAAW,GACzB,EAAa,GACb,EAAa,GAEb,EAAY,EAAE,QAAQ,CAAC,IACvB,EAAQ,EAAO,QAAQ,CAAC,EAAG,EAAW,KAEtC,EACF,GAAI,IAAoB,EAAY,EAAU,MAAO,GACnD,EACF,KAAK,cAAwB,EAAe,CAAC,IAAY,QAAQ,CAC/D,EAAG,EAAW,GAAI,EAAW,KAG7B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAgB,GAAI,IACtB,EAAO,MAAO,CAAC,EAAG,EAAS,EAAS,aAAc,EAClD,EAAY,EAAS,EAAiB,GACpC,EAAuB,CAAC,EAAQ,GACtC,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEd,GAAM,GAAU,KAAK,cAAwB,EAAe,GAE5D,MAAI,GACK,EAAQ,QAAQ,CAAC,EAAG,EAAW,EAAU,EAAS,cAElD,EAAQ,QAAQ,CAAC,EAAG,EAAS,YAAa,EAAW,KAIhE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAEjD,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAM,MAAM,KAAO,EAC3D,MAAO,MAAK,iBACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAGjD,GAAM,GAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAS,KAC7D,EAAU,GAAI,IAChB,EAAU,EAAS,EAAiB,GAClC,EAAuB,CAAC,EAAO,GACrC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAc,EAAS,IAGrC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eAAe,EAAG,EAAQ,GAExC,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAM,KAAO,EACvD,MAAO,MAAK,iBAAiB,EAAG,EAAQ,GAE1C,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAExC,EAA0B,EAAA,MAAM,QAAQ,6BAC1C,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EAC7C,EAAkB,EACpB,GAA6B,EAAY,GACzC,KACE,EAAmB,CAAC,EAAO,GAE3B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KAC5D,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAGd,GAAI,GACJ,MAAI,GACF,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,IAGrC,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,KAGrC,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,6BACd,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EACjD,GAAU,GAAI,IAA6B,GACpC,KAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,GAAU,GAAI,IAAuB,GAC9B,KAAK,cAAc,EAAS,CAAC,EAAG,MAGzC,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAA+B,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAM,GAAU,GAAI,IAAgC,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAA,aAAa,WAAW,EAAG,EAAO,OAG3C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,KAAK,MAAM,EAAG,EAAO,GAAM,QAAQ,GAE9C,MAAO,IAGT,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GAA2B,GAAI,IAAyB,GAC9D,MAAO,MAAK,cAAc,EAA0B,CAAC,GAAK,EAAE,QAG9D,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAM,GAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,KAAK,cAAc,EAA2B,CAAC,IAC7C,EAA2B,GAAI,IAAyB,GACxD,EAAS,KAAK,cAChB,EAA0B,CAAC,EAAI,GAAqB,EAAE,OAC1D,SAAmB,UACZ,GAGT,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IACA,EAAE,MAAO,EAAW,EAAU,GAClC,GAAI,IAAsB,EAAE,MAAO,EAAW,EAAU,GAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,GAAM,GAAU,GAAI,IAA8B,EAAI,EAAG,GAEzD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,GAAI,IAChB,EAAE,MAAO,EAAW,EAAU,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GACF,GAAI,IAAoC,EAAI,EAAG,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAM,GAAQ,EAAa,EAAS,EAAA,QAAQ,GACtC,EAAY,EAAM,MAAM,GACxB,EAAc,EAAM,MAAM,GAC1B,EAAU,GAAI,IAAmB,EAAW,EAAa,GACzD,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAQ,QAAS,IAGvD,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAM,GAAU,GAAI,IAAc,EAAQ,KAAM,EAAO,EAAS,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAAI,IAAY,EAAE,MAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,GAAM,GAAU,GAAI,IAChB,EAAM,MAAO,EAAM,MAAO,EAAU,EAAQ,GAChD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAO,EAAO,GAAW,YAG/D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAE9D,GAAM,GAAY,EAAE,MAAM,GACpB,EAAe,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAE5D,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAe,IAAe,OAChC,CAAC,EAAW,EAAc,EAAa,GACvC,CAAC,EAAW,EAAa,EAAc,GAErC,EAAU,GAAI,IAAoB,EAAa,EAAW,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAG5C,EAAe,CAAC,EAAa,EAAW,GACxC,EAAiB,EAAQ,QAAQ,CAAC,EAAY,IAC9C,EAAW,EAAQ,QAAQ,CAAC,EAAY,IAE9C,GAAI,IAAe,EACjB,MAAO,GAAA,aAAa,cAAc,EAAA,OAAO,IAAK,GAEhD,GAAM,GAAe,EAAA,OAAO,GACtB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAe,KAAM,EAAS,KAAM,EAC3D,GACE,EACF,KAAK,cAAc,EAAS,CAAC,EAAU,EAAgB,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAGjC,EAAiB,GACjB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAc,KAAM,EAAa,KAAM,EAC9D,CAAC,EAAY,GAAI,GACf,EAAc,KAAK,cACrB,EAAS,CAAC,EAAc,EAAe,IAC3C,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGzB,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGjB,EAAA,UAAA,QAAR,SAAgB,EAAa,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAGxD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAU,KAAK,QAAQ,EAAM,GAAM,KAAK,EAAE,MAAM,GAAI,EAAE,MAAM,IAClE,SAAK,UACL,EAAK,UACE,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAGpC,EAAiB,EAAQ,QAAQ,CAAC,EAAW,IAC7C,EAAW,EAAE,QAAQ,CAAC,EAAE,KAAO,EAAW,IAC1C,EACF,GAAI,IAAgB,EAAW,EAAS,CAAC,EAAW,IAClD,EAAc,KAAK,cAAc,EAAS,CAAC,EAAU,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAG5C,GAFA,EAAQ,GAAS,EAAA,KAAK,WAAW,GAE7B,IAAU,UAEZ,GAAM,GAAS,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IAChE,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,WAEjD,GAAM,GAAU,GAAI,IAAY,EAAO,GACjC,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,GAAI,EAAO,KAIlD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAIhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAE,QAAU,SAAW,GAAK,EAAG,EAAE,QAG7D,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GAEpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGhD,EAAA,UAAA,eAAA,SAAe,EAAiB,EAAiB,GAE/C,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,YAAK,QAAQ,IAAI,GAAQ,MAAQ,KAC1B,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAGtB,EAAA,UAAA,WAAR,SACI,EAAiB,EAAiB,GAC7B,GAAA,GAAA,KAAA,eAAA,EAAA,EAAA,GAAA,OACP,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAGrD,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAU,GAAI,IAAc,EAAM,OACxC,MAAO,MAAK,gBAAgB,EAAS,CAAC,GAAQ,EAAM,QAG9C,EAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAU,GAAI,IAAY,EAAM,OAChC,EAA8B,GACpC,MAAO,MAAK,gBACR,EAAS,CAAC,GAAQ,EAAM,MAAO,KAC/B,IAGE,EAAA,UAAA,cAAR,SAAsB,EAAmB,GACvC,GAAM,GAAe,CACnB,GAAuB,EAAM,eAC1B,GAAuB,EAAM,QAE5B,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EAAiB,CACrB,GAAuB,WAAgB,GAAuB,IAG1D,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,KAAK,gBAChB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,QAG1D,EAAA,UAAA,OAAR,SAAe,GACb,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,MAClB,EACF,GAAwB,GACxB,EACJ,AAAI,EACF,EAAU,GAAI,IAA0B,GAExC,EAAU,GAAI,IAAoB,GAEpC,GAAM,GAAgC,GAChC,EAAM,KAAK,gBACb,EAAS,CAAC,CAAC,MAAO,EAAW,MAAK,EAAE,OAAM,IAAI,EAC9C,KAAwB,GAC5B,MAAO,CAAC,MAAK,EAAE,MAAK,EAAE,OAAQ,EAAI,SAGpC,EAAA,UAAA,gBAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAHJ,GAAA,GAAA,KAGI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,KAAK,eAAe,EAAQ,YAAa,GAClD,EAAU,KAAK,QAAQ,IAAI,EAAO,QAIxC,GAHA,AAAI,EAAQ,cACV,GAAQ,SAAW,IAEjB,EAAQ,mBAAqB,GAAuB,OACtD,GAAM,GAAa,GAA0B,EAAQ,aAKrD,EAAQ,SAAW,EAAW,IAAI,SAAA,GAAK,MAAA,GAAI,IAK7C,GAHA,AAAI,EAAQ,aAAe,MACzB,GAAQ,MAAQ,EAAQ,aAEtB,EAAA,KAAK,cAAc,EAAO,SAAW,EAGvC,SAAQ,OACJ,EAAA,KAAK,uBAAuB,EAAO,MAAoB,GACpD,EAGT,GAAM,GAA8B,GAC9B,EAA2B,EAAO,IAAI,SAAA,GAC1C,GAAI,EAAM,QAAU,YAClB,KAAM,IAAI,OACN,mIAKN,GAAI,GAAU,EAAK,QAAQ,IAAI,EAAM,QAErC,GAAI,EAAQ,SAAW,MACrB,GAAI,CAAC,EAAQ,cACT,EAAA,KAAK,cAAc,EAAM,QACrB,EAAA,MAAM,UAAU,6BAMtB,MAAO,CACL,MAAO,EAAM,MACb,QAAS,KACT,UAAW,GACX,cAAe,EAAQ,QAM3B,AAAI,EAAQ,cACV,GAAQ,SAAW,GACnB,EAAQ,MAAQ,EAAM,eAEf,CAAC,CAAC,EAAQ,WAAa,CAAC,CAAC,EAAQ,aAC1C,EAAQ,EAAQ,SAAW,EAAK,aAAa,GAClB,EAAK,WAAW,GAC3C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,gBAE/B,EAAQ,UACR,CAAC,GAAyB,EAAQ,MAAO,EAAM,QAQjD,GAAM,GAAa,EACb,EAAc,EAAM,MAE1B,EAAM,MAAQ,EAAQ,MACtB,EAAQ,EAAK,cAAc,EAAiB,GAC5C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,QAEjC,EAAW,MAAQ,EAGrB,SAAK,YAAY,EAAM,QAChB,CAAC,MAAO,EAAM,MAAO,QAAO,EAAE,UAAW,MAGlD,KAAK,YAAY,EAAO,QACxB,GAAM,GACW,CAAC,MAAO,EAAO,MAAO,QAAS,EAAS,UAAW,IAC9D,EAAM,GAAyB,EAAS,EAAY,GACpD,EAAS,KAAK,iBAAiB,EAAK,WACxC,MAAO,IACH,EAAK,MAAO,EAAS,EAAY,KAEjC,EAAoB,KAAK,cAAgB,KAC3C,EAgBJ,GAfA,AAAI,GACF,GAAQ,KAAK,cAGf,GACI,KAAK,MAAO,EAAQ,EAAY,EAAY,GAEhD,EAAc,QAAQ,SAAA,GAAQ,MAAA,GAAK,8BAA8B,KAEjE,AAAI,GACF,GAAQ,KAAK,SAAS,GACtB,KAAK,aAAa,KACd,CAAC,KAAM,EAAQ,YAAY,KAAM,MAAO,KAAK,aAAa,MAG5D,CAAC,EAAA,MAAM,QAAQ,wBAA0B,EAAQ,UACjD,IAAkC,IACpC,GAAM,GAAW,KAAK,aAAa,GACnC,YAAK,8BAA8B,GAC5B,EAET,MAAO,IAGT,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,EAAc,GAAe,EAAO,GAAG,MACvC,GAAM,GAAU,KAAK,gBACjB,EAAS,EAAQ,EAAa,EAC9B,GACJ,MAAO,GAAA,SAAS,qBACL,EAAQ,OAAQ,EAAQ,MAAO,EAAQ,QAG5C,EAAA,UAAA,iBAAR,SAAyB,EAAa,GAEpC,MAAM,KAAO,MAAK,aAChB,MAAK,YAAY,GAAO,KAEnB,KAAK,YAAY,IAG1B,EAAA,UAAA,kBAAA,WACE,MAAO,MAAK,gBAKd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAIF,GAAI,CAAC,EAAA,MAAM,QAAQ,YACjB,GAAM,GAAU,OAAO,KAAK,KAAK,aACjC,EAAQ,QAAQ,SAAA,GACd,EAAK,MAAM,cAAc,EAAK,YAAY,GAAK,cAC/C,MAAO,GAAK,YAAY,KAG5B,KAAK,eAAe,UACpB,AAAI,KAAK,QAAU,MACd,MAAQ,oBAAuB,aAC/B,KAAK,iBAAkB,mBAC1B,KAAK,OAAO,SAEZ,KAAK,OAAS,KAEhB,AAAI,KAAK,qBACP,MAAK,MAAM,QAAU,KACrB,KAAK,MAAM,WAEb,KAAK,SAAW,IAGlB,EAAA,UAAA,eAAA,WAAA,GAAA,GAAA,KACE,MAAI,MAAK,qBAAuB,MAC9B,MAAK,oBAAsB,EAAA,KAAK,WAC9B,GAAI,CAAC,EAAA,MAAM,IAAI,iCAGb,GAAM,GAAY,EAAA,MAAM,QAAQ,SAChC,EAAA,MAAM,IAAI,QAAS,IACnB,GAAM,GAAsB,EAAK,IAAI,EAAA,OAAO,OAAO,WAAW,GAG9D,GAFA,EAAA,MAAM,IAAI,QAAS,GAEf,EAAsB,EACxB,MAAO,IAGX,MAAO,OAGJ,KAAK,qBAGd,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAGlD,EAAA,UAAA,YAAR,SAAoB,SACZ,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAE7C,GAAI,GAAW,KAEb,OAEF,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GAAW,EAAQ,SAMvB,GALA,AAAI,GAAY,MACd,GAAW,GAA2C,EAAO,GAC7D,EAAQ,SAAW,GAGjB,GAAU,MACZ,GAAM,GAAY,GAAwB,GAEtC,EAAO,OACP,EAAQ,EAAS,GAAI,EAAS,EAAS,GACrC,EAAc,YAAkB,YAEtC,AAAI,EACF,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,EAAU,GAAI,IACV,EAAW,CAAC,EAAQ,GAAQ,IAEhC,EACI,GAAI,IAAoB,EAAW,CAAC,EAAQ,GAAQ,GAG1D,GAAM,GAAuB,KAAK,eAAe,CAAC,EAAQ,GAAQ,GAClE,AAAI,EACF,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEjB,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEnB,KAAK,MAAM,2BACP,KAAK,WAAW,EAAqB,QAAS,EAAO,EACrD,GAIJ,GAAM,GAAwB,GACxB,EAAsB,KAAK,gBAC7B,EAAS,CAAC,GAAuB,EAAO,KAAM,GAG5C,EAAgB,KAAK,QAAQ,IAAI,EAAoB,QAC3D,EAAQ,QAAU,EAAc,QAChC,EAAQ,SAAW,EAAc,SACjC,EAAQ,SAAW,EAAc,SACjC,EAAQ,MAAQ,EAAc,MAE9B,KAAK,8BAA8B,GACnC,KAAK,QAAQ,OAAO,EAAoB,QAGxC,EAAQ,OAAS,KACjB,AAAI,GACF,MAAK,cAAgB,EAAA,KAAK,MAAQ,QAGpC,GAAM,GAAa,KAAK,eAAe,EAAU,EAAO,EAAO,GAC/D,EAAQ,QAAU,IAId,EAAA,UAAA,qBAAR,SAA6B,EAAgB,GAE3C,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAEP,YAAK,eAAe,GAEpB,AAAI,GAAiB,MACnB,GAAQ,OAAS,GAAoB,EAAe,IAE/C,EAAQ,QAGT,EAAA,UAAA,eAAR,SACI,EAA4B,EAAuB,EACnD,GAEF,GADA,KAAK,eAAiB,KAAK,aAAa,EAAU,GAC9C,CAAC,KAAK,mBACN,KAAK,cAAgB,KAAK,mBAAqB,KAAO,MACxD,GAAM,GAAM,MAAK,cAAgB,KAAO,MAAM,QAAQ,GACtD,KAAK,kBAAoB,GACzB,QAAQ,KACJ,6BAA6B,EAAE,yCAGrC,MAAO,MAAK,eAAe,eAAe,EAAU,EAAS,IAGvD,EAAA,UAAA,aAAR,SAAqB,EAAyB,GAC5C,MAAO,GAAM,GAAK,EAAM,GAAK,EAAA,KAAK,gBAAgB,IAG5C,EAAA,UAAA,mBAAR,SACI,EAAsB,GACxB,GAAI,KAAK,mBAAmB,GAC1B,IACE,MAAO,WACA,GACP,GAAI,EAAA,MAAM,QAAQ,WAChB,KAAM,IAAI,OAAM,yBAItB,MAAO,OAEX,GAriFsC,EAAA,eAuiFtC,YACI,EAAiB,GACnB,GAAI,IAAU,WAAa,IAAU,YACnC,MAAO,GACF,GAAI,IAAU,SAAW,IAAU,QAGxC,OAFM,GAAU,IAAU,QAAW,GAAI,YAAW,EAAE,QACjB,GAAI,YAAW,EAAE,QAC7C,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,KAAK,MAAM,EAAE,IAE3B,MAAO,OAEP,MAAM,IAAI,OAAM,iBAAiB,GChwFrC,GAGM,IAAU,QCHhB,cAmCE,EAAA,MAAM,IAAI,2BAA4B,ICnCxC,AAuBA,AAAI,EAAA,YAAY,aACd,EAAA,gBAAgB,QAAS,WAAM,MAAA,IAAI,KAAoB,MAQ5C,IAAQ,CAAC,eAAc,ICzBvB,GAA0B,0BAE1B,GAA2B;;;EAK3B,GAAkC;;;;;cAWf,GAC9B,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EACD,EAAe,EACf,EAAU,GAAI,IAAe,EAAE,MAAO,GAC5C,MAAO,GAAa,gBAAgB,EAAS,CAAC,GAAI,EAAE,oBAepD,EAAmB,EACnB,EAA4C,GAG9C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAe,EACf,EAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,EAAiB,EAAE,MAAO,EAAE,MAC5B,CAAC,CAAC,GACN,GAAI,IAAgB,EAAW,EAAE,MAAO,EAAE,OACxC,EAAS,GAAS,EAAE,MACpB,EAAS,EAAa,gBAAgB,EAAS,CAAC,EAAG,GAAI,GAC7D,MAAO,IC3DX,AAsBA,GAAM,IAAQ,GAA2B;;EAInC,GAAe;;;IAIjB,GAAkC;;EAIzB,GAAkB,GAAiB,GAAO,IAE1C,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,QACb,WAAY,ICvCd,YAqBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,QACb,WAAY,ICjCd,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,WAG/C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBACvB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAyB,GAAI,IAAyB,GAC5D,MAAO,GAAQ,gBAAgB,EAAwB,CAAC,GAAK,EAAE,OAG1D,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,IC3Cd,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,YAAwB,GAOtB,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,MACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,MACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;yBAKK,EAAa;wBACd,EAAY;2DACuB,EAAe;;;MAK1E,MAAA,MC5DA,AAoBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,YACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,YACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;wBAEI,EAAa;uBACd,EAAY;;;;;;yDAMsB,EAAe;;;;MAMxE,MAAA,MChEA,AAuBO,GAAM,IAIM,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,EAAA,EAAG,EAAA,EAAA,KAAM,EAAA,EAAA,SAAU,EAAA,EAAA,OAAQ,EAAA,EAAA,MAElC,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGL,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAGpB,GAAM,GAAc,CAAC,EAAG,EAAM,GAE1B,EAAc,KAClB,AAAI,GAAU,MACZ,GAAc,EAAO,MACrB,EAAY,KAAK,IAGnB,GAAI,GAAa,KACjB,AAAI,GAAS,MACX,GAAa,EAAM,MACnB,EAAY,KAAK,IAGnB,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACJ,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACF,EACF,EAAQ,gBAAgB,EAAS,EAAa,EAAY,GAAG,OAEjE,MAAO,IAGI,GAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,QACb,WAAY,IC/Ed,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAuBA,GAAM,IAAM;;;;eAQN,GAAa;;;;;;;;;;;;;;;;;;EAoBN,GAAgB,GACzB,GAAK,GAAY,IAER,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,ICzDd,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAItB,GAAM,GAAa,EAAW,GAC9B,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;yBAKK,EAAU;;uCAEI,EAAU;;;;;;;MASjD,MAAA,MC5CA,AAuBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,MACD,EAAe,EAEf,EAAU,GAAI,IAAsB,EAAmB,OACvD,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KChCX,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;uDAMmC,EAAK,OAAO,EAAM;;wBAEjD,EAAK,UAAS;;;;;;;;;;;;;;MAgBtC,MAAA,MCrDA,AAoBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;;8BAeU,EAAK,OAAO,EAAM;4BACpB,EAAK,UAAS;;;;;;;;;;;;;;;;UAgBhC,EAAK,OAAM;;MAIrB,MAAA,MCnEA,AA0BO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,QACb,WAAY,IAGV,GAEJ,YAAoB,GAKX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACnB,EAAA,EAAA,OACE,EAAA,EAAA,YAED,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAOR,EAA6B,CAAC,EAAQ,GACtC,EAAW,CAAC,EAAQ,EAAO,GAEjC,AAAI,IAAW,IACb,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAGpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA+C,EAAG,EAAG,EAAO,GAChE,EAAS,GAAoB,QAG/B,GAAM,GAAkB,EAAQ,eAAe,EAAU,SAEzD,EAAQ,QAAQ,IAAI,EAAgB,QAAQ,MAAQ,GAAa,OACjE,EAAQ,MAAM,yBACV,EAAQ,WAAW,EAAgB,QAAS,GAChD,GAAM,GAAU,EAAA,MAAM,QAAQ,cAC1B,GAAI,IAAwB,GAC5B,GAAI,IAAkB,GACpB,EAAM,EAAQ,gBAAgB,EAAS,CAAC,GAAkB,SAChE,SAAQ,YAAY,EAAgB,QAC7B,EC/ET,AA0BA,YAA4B,GAI1B,OAFM,GAAS,GAER,EAAO,SAAW,GAAK,EAAO,EAAO,OAAS,GAAG,UAAY,IAClE,GAAM,GACF,EAAO,OAAS,EAAO,EAAO,OAAS,GAAG,QAAU,EAAQ,GAC1D,EAAa,EAAA,aAAa,yBAAyB,GACzD,EAAO,KAAK,CACV,OAAQ,EACR,WAAU,EACV,QAAS,KAAK,KAAK,EAAU,KAIjC,MAAO,eAIL,EAAe,EAAiB,EAChC,GAIF,OAHM,GAAkB,GAAmB,EAAE,OAEzC,EAAS,EACJ,EAAI,EAAG,EAAI,EAAgB,OAAQ,KACpC,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,WAAY,EAAA,EAAA,QAErB,EAAU,GAAI,IAChB,CAAC,WAAU,EAAE,OAAM,EAAE,UAAW,EAAE,MAAM,GAAI,QAAO,GAAG,GACpD,EAAiB,EACvB,EAAS,EAAQ,gBAAgB,EAAS,CAAC,GAAS,GAEpD,AAAI,EAAe,SAAW,EAAE,QAC9B,EAAQ,YAAY,EAAe,QAIvC,MAAO,GC/DT,YAwBI,EAAmB,EACnB,GACF,GAAM,GACF,CAAC,GAAY,EAAM,QAAM,OACrB,GAAY,EAAM,QACpB,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EACF,CAAC,GAAY,IAAW,OACpB,GAAY,IAEd,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,EAAQ,gBACnB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,OC3ClE,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MACD,EAAe,EAEf,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QAC5C,MAAI,GAAS,UAAY,CAAC,GAAc,EAAE,MAAO,IAC7C,CAAE,GAAS,UAAY,MAAQ,GAAc,EAAS,MAAO,IACxD,GAAc,EAAG,EAAQ,GAGlC,GAAa,OAAO,EAAE,QAEf,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,QAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICzDd,YAwBI,EAAe,EAAuB,EACtC,GACF,GAAM,GAAS,EAAA,KAAK,cAAc,GAC5B,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAY,EAAQ,EACpB,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAC,GAAG,MAAO,CAAC,MAAO,CAAC,EAAW,IAAU,QAAO,IAEhE,EAAU,GAAO,EAAe,EAAE,MAAO,MAAO,GAChD,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAU,MAAO,CAAC,MAAO,GAAW,QAAO,IAEpE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,ECvCT,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAEjC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAW,GAAkB,GAEnC,KAAK,SAAW;;QAEZ,EAAK;uBACU,EAAQ;;MAI/B,MAAA,MAEA,YAA2B,GACzB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,sBAAsB,EAAI,yBAKxC,OAHM,GACF,CAAC,UAAW,UAAW,UAAW,UAAW,UAAW,WACtD,EAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAe,EAAO,IAAM,EAAc,GAE5C,MAAO,GAAe,OCxDxB,AAqBA,GAAA,IAAA,WAQE,WAAY,EAAkB,GAP9B,KAAA,cAAgB,CAAC,KAIjB,KAAA,aAAe,GACf,KAAA,aAAe,GAIb,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAIjC,GAFA,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACpB,KAAK,KAAO,EACd,KAAM,OACF,6BAA6B,KAAK,KAAI,0BAM5C,OAJM,GAAQ,GAAkB,KAAK,MAE/B,EAAc,GAAe,KAAM,KAAK,MACxC,EAAgB,GAAI,OAAM,KAAK,MAC5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAc,EAAO,IAAM,EAAY,GAEzC,GAAM,GAAY,QAAQ,EAAc,MAAM,IAAI,OAAM,IAClD,EACF,KAAK,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAC3D,EAAO,mBAAmB,EAAc,OAAM,MAAM,EAAS,IAEnE,KAAK,SAAW;;QAEZ,EAAK;;oBAEO,EAAI;WACb,EAAU;sBACC,EAAI;;UAEhB,EAAY,KAAK,KAAO,GAAE;aACvB,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAAE;sBACjD,EAAI;aACb,EAAU;wBACC,EAAI;;;;;MAO5B,MAAA,MCvEA,YAyBI,EAAe,EAAgB,GACjC,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAuB,EAAE,MAAO,GACpC,GAAI,IAAiB,EAAE,MAAO,GAClC,MAAO,GAAQ,gBAAgB,EAAS,CAAC,GAAI,EAAE,OC7BjD,AA0BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAW,EAAA,KAAK,eAAe,EAAkB,EAAE,OACrD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACrD,EAAuB,GAAgB,KACvC,EAAqB,EAAa,mBAAmB,CAAC,IAExD,EAAW,EACf,GAAI,GACF,GAAI,GAKF,OAJM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAa,IAErC,GAAM,GACF,GAAiB,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAc,GAE7D,EAAW,EAAa,eAAe,EAAU,EAAE,OACnD,GAAM,GAAe,EAAa,QAAQ,IAAI,EAAS,QACvD,EAAa,OAAS,MAEtB,GAAW,GAAc,EAAG,EAAc,GAG5C,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAGpD,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGhB,EAAW,EACf,AAAI,GAEF,GAAW,EAAA,aAAa,qBAAqB,EAAa,IAG5D,GAAI,GACJ,GAAI,GACF,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAY,GACd,EAAQ,EAAA,KAAK,cAAc,GAAc,EAAU,EAAE,OAEzD,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAQ,EAAU,EAAa,EAAU,GAGjD,MAAI,IACF,EAAa,8BAA8B,GAGtC,IC9FX,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,EAAE,OAGjD,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAuBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAC3B,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,EAAQ,gBAAgB,EAAyB,CAAC,GAAI,EAAE,OAEtD,EAAyB,GAAI,IAAyB,GACtD,EAAS,EAAQ,gBACnB,EAAwB,CAAC,EAAI,GAAmB,EAAE,OACtD,SAAQ,8BAA8B,GAC/B,EAGF,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,ICrDd,YAuBI,EAAe,EACf,EACA,GACF,GAAI,GAAU,GAAI,IAAc,EAAU,MAAO,IAC3C,EAAa,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAEzD,EAAU,GAAI,IAAc,EAAU,MAAO,GAAM,GAAM,GACzD,GAAM,GAAc,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAC1D,MAAO,CAAC,EAAY,GC/BtB,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAe,EAErB,EAAA,KAAK,OACD,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,uDACF,EAAE,MAAM,OAAM,MACtB,GAAM,GAA8B,CAAC,EAAG,GACxC,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,GAET,EAAA,GAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAEf,MAAO,CAAC,EAAQ,KCjDpB,AAoBO,GAAM,IAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAG9B,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EAE1B,MAAO,GAAA,aAAa,wBAChB,EAAW,EAAY,EAAkB,EACzC,KC3CR,AAkBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC1C7B,AAmBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAqBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EACA,GAPJ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAOtB,GAAM,GAAc,EAAW,GACzB,EAAa,EAAW,GACxB,EAAY,KAAK,IAAI,GAAS,QAAQ,GACtC,EAAY,KAAK,IAAI,GAAS,QAAQ,GAC5C,KAAK,YAAc,EAEb,GAAA,GAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAgB,EAAQ,QAAQ,GAChC,EAAgB,EAAQ,QAAQ,GAElC,EAAc,GAClB,AAAI,MAAO,IAAc,SACvB,EAAc,uBAAuB,EAAU,QAAQ,GAAE,IAEzD,EAAc;2BACO,EAAU,KAAK,KAAI;8CAI1C,KAAK,SAAW;;;;;4CAKwB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;4CACpB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;iDACf,EAAa;iDACb,EAAa;YAClD,EAAW;uCACgB,EAAU,+BACzC,EAAW;;;;;MAOnB,MAAA,MCtEA,AAuBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAe,EAEf,EAAU,GAAI,IACf,EAAmB,MAAO,EAAS,EAAW,GAC7C,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KClCX,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAqBA,GAAM,IAAS,gBAEF,GAAmB,GAAgB,IAEnC,GAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IC5Bd,AAqBA,GAAM,IAAqB,4BAEd,GACT,GAAiB,GAAoB,IAE5B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,IC7Bd,AAqBA,GAAM,IAAM,iBAEC,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC5Bd,AAwBO,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,QACb,WAAY,SAAC,GAQX,OARY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACA,EAAA,EAAA,KACD,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAI,GACJ,GAAI,EAAa,mBAAmB,CAAC,KACnC,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QACtC,EAAS,EAAS,OAClB,EAAY,GAAa,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAE/D,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAc,EAAG,EAAM,GAE/B,MAAO,KCnDX,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAGpB,QAAQ,KACJ,YACA,8DACJ,GAAM,GAAS,EAAQ,SAAS,EAAE,QAC5B,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IChDd,AA4CA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,+RCxEjB,AAGA,GAAM,IAAU,QCHhB,GAsCa,IAAU,CACrB,YAAa,GAAA,aACb,mBAAoB,GAAA,YACpB,qBAAsB,GAAA,cACtB,YAAa,GAAA,aACb,cAAe,GAAA,eACf,iBAAkB,GAAA,kBAClB,KAAQ,yZC7CV,oBAAM,IAAK,KAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,GAAG,IAAI,EAAI,WAAY,GAChC,EAAO,GAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,GAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,GAAG,IAAI,EAAW,GAC5B,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,GAAG,IAAI,EAAU,GACtC,EAAoB,GAAG,IAAI,EAAS,GACpC,EAAc,GAAG,IAAI,EAAoB,GACzC,EAAS,GAAG,IAAI,EAAmB,GACnC,EAAO,GAAG,IAAI,EAAmB,GACjC,EAAkB,GAAG,IAAI,EAAQ,GACjC,EAAgB,GAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,IAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,IAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,GAAG,SAAS,KAAK,aAChC,KAAK,UAAY,GAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,GAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,GAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,GAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,GAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,GACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,GAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,IAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,kBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,oBAAM,IAAK,KAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,GAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,YAAc,KClDtB,kBAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,GAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,GAAQ,wBAA0B,KCvFlC,eACA,KAAM,IAAK,KACL,GAAW,KACX,GAAY,KACZ,GAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,GAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,GAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,GAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,GAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,GAAU,yBAAyB,GACtD,EAAkB,GAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,GAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,GAAS,YAAY,GAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,GAAS,WAAW,GACpC,GAAI,GAAO,GAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,GAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,GAAS,oBAAoB,EAAe,GACxD,EAAc,GAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,GAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,GAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,GAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,GAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,GAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,EAAgB,KAAM,GAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,EAAa,QAAS,EAAiB,KAAM,GAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,EAAkB,KAAK,UAAU,QAAQ,GAAG,OAAO,CAAC,EAAa,KACjE,EAAqB,EAAe,WAC1C,EAAe,UACf,KAAM,GAAc,EAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,EAAkB,KAAM,GAAsB,KAAK,aAAa,EAAa,EAAY,EAAgB,IACtH,EAAe,EAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,EAAmB,KAAM,GAAuB,KAAK,aAAa,EAAc,EAAa,GAC1G,EAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,GAAiC,GAC5C,IAAsB,EAAW,EAAkB,QACnD,GAAsB,EAAW,EAAmB,UAE/C,AAAI,EAAgC,EACzC,GAAsB,EAAW,EAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,EAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,GAAyB,KAAK,sBAAsB,EAAW,EAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,EAAoB,SAC1F,EAAY,EAAU,OAAO,GAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,GAAG,QAAQ,GACX,KAAM,GAAe,GAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,GAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,GAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,GAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,kBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,0CAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,oBAAM,IAAK,KACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QAAS,GAAQ,GAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,GAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,GAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,oBAAM,IAAK,KAEL,GAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,GAAG,QAAQ,WAAW,GAC/B,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,IAAO,KAAK,IAAO,IAAM,KAAM,IAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,GAAO,IAGhB,kBAA0B,GACxB,MAAK,IAAO,QAAQ,IAAO,OAAS,KAAM,IAAG,eAAe,EAAO,KAAK,OAAO,YACxE,GAAO,OAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,IAAI,WAC3B,UAAQ,EACD,GAET,GAAI,GACJ,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,GAAG,IAAI,EAAQ,CAAC,MAC1B,GAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,IAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,GAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,IAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OAC3B,EAAI,OAAS,KAAK,MAAM,IAAM,EAAK,IAAM,GAAK,SAAW,OACzD,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KCzDrB,oBAAM,IAAK,KAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,GAAG,KAAK,KACrB,KAAM,GAAS,GAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,IAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,GAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,GAAG,IAAI,EAAG,CAAC,QAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAY,GAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,oBAAM,IAAK,KAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,oBAAM,IAAK,KACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,eACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,oBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,kBAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,GAAQ,cAAgB,GAAQ,UAAU,OAC1C,GAAQ,QAAU,GAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,GAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,GAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,GAAQ,QAAQ,GAAa,GAAQ,QAAQ,KACnI,GAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,oBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,GAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,GAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,GAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,GAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,GAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,GAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,GAAQ,YAAc,KCnDtB,oBAAM,IAAY,KACZ,GAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,GAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,GAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,GAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,GAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,oBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,oBAAM,IAAK,KACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,GAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,GAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,GAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,GAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,GAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,GAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,IAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,GAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,IAAG,OAAS,EAAQ,GAAG,QAAQ,WAAW,GAEpE,GAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,IAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,GAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,GAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,GAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,GAAQ,kBAAoB,KCrH5B,oBAAM,IAAK,KACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,oBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,GAAQ,KAAO,GAAa,KAC5B,GAAQ,QAAU,GAAa,QAE/B,GAAQ,UAAY,GAAe,UACnC,GAAQ,oBAAsB,GAAe,oBAC7C,GAAQ,aAAe,GAAU,aACjC,GAAQ,QAAU,GAAU,QAC5B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,eAAiB,GAAK,eAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,kBAAoB,GAAK,kBACjC,GAAQ,UAAY,GAAK,YCnBzB,oBAAM,IAAK,KAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,GAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,SAAW,KCtEnB,oBAAM,IAAK,KACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,GAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,GAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,GAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,IAAG,KAAK,KACb,KAAM,GAAa,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,GAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,IAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,IAAG,KAAK,KACb,KAAM,GAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,IAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,GAAG,KAAK,IAAM,GAAG,IAAI,GAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,GAAG,KAAK,IAAM,GAAG,QAAQ,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,GAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,GAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,GAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,kBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,2BAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,KCzEtB,oBAAM,IAAK,KACL,GAAW,KACX,GAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,IAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,IAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,GAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,GAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,GAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,GAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,GAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,oBAAM,IAAK,KACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,GAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,IAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,0CAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GACT,SAAU,CACR,UAAW,sCACX,UAAW,IACX,SAAU,GACV,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCACX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,EACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GACZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,EACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,6vECnEjB,oBAAM,IAAK,KACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,IAGJ,KAAM,IAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,GAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,GAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,GAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,GAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,IAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,GAAG,eAAiB,GAAO,SAC7B,IAAI,iCAAkC,GAAO,SAC7C,KAAM,IAAG,WAAW,GAAO,SAC3B,KAAM,IAAG,SAGX,GAAI,GACJ,AAAI,GAAG,eAAiB,SACtB,GAAkC,GAAG,MAAM,IAAI,4BAC/C,GAAG,MAAM,IAAI,2BAA4B,KAI3C,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,GAAO,OACvF,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,IAAI,SAAW,CAAC,GAAO,KAAK,IAAO,IAAM,KAAM,IAAO,QAAQ,KACrG,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,OAAO,SAAW,CAAC,GAAO,QAAQ,IAAO,OAAS,KAAM,IAAO,WAAW,KACjH,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,QAAQ,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,KAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,QAAQ,cAAc,EAAO,GAAO,MAAQ,GAC/F,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAAQ,GAChG,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,GAAO,KAAK,SACd,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAChE,EAAK,KAAO,KAAK,MAAM,KAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,KACZ,KAAM,GAAW,GAAO,KAAK,IAAI,SAAW,GAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,IAAU,GACrH,EAAK,UAAY,KAAK,MAAM,KAAQ,GAEpC,EAAY,KACZ,KAAM,GAAc,GAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,IAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,KAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,GAAG,SAAS,WAId,GAAG,MAAM,IAAI,2BAA4B,GAGzC,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,QAAU,GAClB,GAAQ,SAAW,GACnB,GAAQ,GAAK,GACb,GAAQ,QAAU,GAAI",
+  "sourcesContent": ["", "", "", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n  kernelMs: number|{error: string};\n  getExtraProfileInfo?(): string;  // a field for additional timing information\n                                   // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n  disposeData(dataId: DataId): void;\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void;\n  memory(): {unreliable: boolean;};  // Backend-specific information.\n  /** Returns number of data ids currently in the storage. */\n  numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage<T> {\n  private data = new WeakMap<DataId, T>();\n  private dataIdsCount = 0;\n\n  constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n  get(dataId: DataId) {\n    if (!this.data.has(dataId)) {\n      this.dataMover.moveData(this.backend, dataId);\n    }\n    return this.data.get(dataId);\n  }\n\n  set(dataId: DataId, value: T): void {\n    this.dataIdsCount++;\n    this.data.set(dataId, value);\n  }\n\n  has(dataId: DataId): boolean {\n    return this.data.has(dataId);\n  }\n\n  delete(dataId: DataId): boolean {\n    this.dataIdsCount--;\n    return this.data.delete(dataId);\n  }\n\n  numDataIds(): number {\n    return this.dataIdsCount;\n  }\n}\n\nexport interface DataMover {\n  /**\n   * To be called by backends whenever they see a dataId that they don't own.\n   * Upon calling this method, the mover will fetch the tensor from another\n   * backend and register it with the current active backend.\n   */\n  moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo>;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n  time(f: () => void): Promise<BackendTimingInfo> {\n    return notYetImplemented('time');\n  }\n  read(dataId: object): Promise<BackendValues> {\n    return notYetImplemented('read');\n  }\n  readSync(dataId: object): BackendValues {\n    return notYetImplemented('readSync');\n  }\n  numDataIds(): number {\n    return notYetImplemented('numDataIds');\n  }\n  disposeData(dataId: object): void {\n    return notYetImplemented('disposeData');\n  }\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    return notYetImplemented('write');\n  }\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    return notYetImplemented('move');\n  }\n  memory(): {unreliable: boolean; reasons?: string[]} {\n    return notYetImplemented('memory');\n  }\n  /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n  floatPrecision(): 16|32 {\n    return notYetImplemented('floatPrecision');\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    return notYetImplemented('batchMatMul');\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          FusedBatchMatMulConfig): Tensor3D {\n    return notYetImplemented('fusedBatchMatMul');\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    return notYetImplemented('slice');\n  }\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    return notYetImplemented('stridedSlice');\n  }\n  unstack(x: Tensor, axis: number): Tensor[] {\n    return notYetImplemented('unstack');\n  }\n  reverse<T extends Tensor>(a: T, axis: number[]): T {\n    return notYetImplemented('reverse');\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    return notYetImplemented('concat');\n  }\n\n  neg<T extends Tensor>(a: T): T {\n    return notYetImplemented('neg');\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('add');\n  }\n  addN<T extends Tensor>(tensors: T[]): T {\n    return notYetImplemented('addN');\n  }\n  subtract(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('subtract');\n  }\n  multiply(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('multiply');\n  }\n  realDivide(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('realDivide');\n  }\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('floorDiv');\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('sum');\n  }\n  prod(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('prod');\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    return notYetImplemented('unsortedSegmentSum');\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMin');\n  }\n  argMax(x: Tensor, axis: number): Tensor {\n    return notYetImplemented('argMax');\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('equal');\n  }\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('notEqual');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('less');\n  }\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('lessEqual');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greater');\n  }\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('greaterEqual');\n  }\n\n  logicalNot<T extends Tensor>(a: T): T {\n    return notYetImplemented('logicalNot');\n  }\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalAnd');\n  }\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('logicalOr');\n  }\n\n  where(condition: Tensor): Tensor2D {\n    return notYetImplemented('where');\n  }\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('select');\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    return notYetImplemented('topk');\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('min');\n  }\n  minimum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('minimum');\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('mod');\n  }\n\n  max(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('max');\n  }\n  maximum(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('maximum');\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('all');\n  }\n  any(x: Tensor, axes: number[]): Tensor {\n    return notYetImplemented('any');\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    return notYetImplemented('squaredDifference');\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    return notYetImplemented('ceil');\n  }\n  floor<T extends Tensor>(x: T): T {\n    return notYetImplemented('floor');\n  }\n  round<T extends Tensor>(x: T): T {\n    return notYetImplemented('round');\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    return notYetImplemented('sign');\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    return notYetImplemented('isNaN');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    return notYetImplemented('isInf');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    return notYetImplemented('isFinite');\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    return notYetImplemented('pow');\n  }\n  exp<T extends Tensor>(x: T): T {\n    return notYetImplemented('exp');\n  }\n  expm1<T extends Tensor>(x: T): T {\n    return notYetImplemented('expm1');\n  }\n  softmax<T extends Tensor>(x: T, dim: number): T {\n    return notYetImplemented('softmax');\n  }\n  log<T extends Tensor>(x: T): T {\n    return notYetImplemented('log');\n  }\n  log1p<T extends Tensor>(x: T): T {\n    return notYetImplemented('log1p');\n  }\n  sqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('sqrt');\n  }\n  rsqrt<T extends Tensor>(x: T): T {\n    return notYetImplemented('rsqrt');\n  }\n  square<T extends Tensor>(x: T): T {\n    return notYetImplemented('square');\n  }\n  reciprocal<T extends Tensor>(x: T): T {\n    return notYetImplemented('reciprocal');\n  }\n  relu<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu');\n  }\n  relu6<T extends Tensor>(x: T): T {\n    return notYetImplemented('relu6');\n  }\n  prelu<T extends Tensor>(x: T, a: T): T {\n    return notYetImplemented('prelu');\n  }\n  elu<T extends Tensor>(x: T): T {\n    return notYetImplemented('elu');\n  }\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    return notYetImplemented('eluDer');\n  }\n  selu<T extends Tensor>(x: T): T {\n    return notYetImplemented('selu');\n  }\n  int<T extends Tensor>(x: T): T {\n    return notYetImplemented('int');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    return notYetImplemented('clip');\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    return notYetImplemented('abs');\n  }\n  complexAbs<T extends Tensor>(x: T): T {\n    return notYetImplemented('complexAbs');\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    return notYetImplemented('sigmoid');\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    return notYetImplemented('softplus');\n  }\n\n  sin<T extends Tensor>(x: T): T {\n    return notYetImplemented('sin');\n  }\n  cos<T extends Tensor>(x: T): T {\n    return notYetImplemented('cos');\n  }\n  tan<T extends Tensor>(x: T): T {\n    return notYetImplemented('tan');\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    return notYetImplemented('asin');\n  }\n  acos<T extends Tensor>(x: T): T {\n    return notYetImplemented('acos');\n  }\n  atan<T extends Tensor>(x: T): T {\n    return notYetImplemented('atan');\n  }\n  atan2<T extends Tensor>(a: T, b: T): T {\n    return notYetImplemented('atan2');\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('sinh');\n  }\n  cosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('cosh');\n  }\n  tanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('tanh');\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    return notYetImplemented('asinh');\n  }\n  acosh<T extends Tensor>(x: T): T {\n    return notYetImplemented('acosh');\n  }\n  atanh<T extends Tensor>(x: T): T {\n    return notYetImplemented('atanh');\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    return notYetImplemented('erf');\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    return notYetImplemented('step');\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedConv2d');\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2d');\n  }\n  conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('conv2dDerInput');\n  }\n  conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('conv2dDerFilter');\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          FusedConv2DConfig): Tensor4D {\n    return notYetImplemented('fusedDepthwiseConv2D');\n  }\n\n  depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2D');\n  }\n  depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerInput');\n  }\n  depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('depthwiseConv2DDerFilter');\n  }\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3d');\n  }\n  conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n      Tensor5D {\n    return notYetImplemented('conv3dDerInput');\n  }\n  conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('conv3dDerFilter');\n  }\n  maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('maxPool');\n  }\n  maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n      Tensor4D {\n    return notYetImplemented('maxPoolBackprop');\n  }\n  avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPool');\n  }\n  avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n    return notYetImplemented('avgPoolBackprop');\n  }\n  avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3d');\n  }\n  avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('avgPool3dBackprop');\n  }\n  maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3d');\n  }\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n    return notYetImplemented('maxPool3dBackprop');\n  }\n\n  reshape<T extends Tensor, R extends Rank>(x: T, shape: ShapeMap[R]):\n      Tensor<R> {\n    return notYetImplemented('reshape');\n  }\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return notYetImplemented('cast');\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    return notYetImplemented('tile');\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    return notYetImplemented('pad');\n  }\n\n  transpose<T extends Tensor>(x: T, perm: number[]): T {\n    return notYetImplemented('transpose');\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    return notYetImplemented('gather');\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    return notYetImplemented('gatherND');\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    return notYetImplemented('scatterND');\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    return notYetImplemented('batchToSpaceND');\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: number[][]): T {\n    return notYetImplemented('spaceToBatchND');\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeBilinear');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    return notYetImplemented('resizeBilinearBackprop');\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHEight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighbor');\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    return notYetImplemented('resizeNearestNeighborBackprop');\n  }\n\n  batchNorm(\n      x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n      offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n      varianceEpsilon?: number): Tensor4D {\n    return notYetImplemented('batchNorm');\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    return notYetImplemented('localResponseNormalization4D');\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n      bias: number, alpha: number, beta: number): Tensor4D {\n    return notYetImplemented('LRNGrad');\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    return notYetImplemented('multinomial');\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    return notYetImplemented('oneHot');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    return notYetImplemented('cumsum');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold?: number): Tensor1D {\n    return notYetImplemented('nonMaxSuppression');\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('fft');\n  }\n  ifft(x: Tensor2D): Tensor2D {\n    return notYetImplemented('ifft');\n  }\n  complex<T extends Tensor>(real: T, imag: T): T {\n    return notYetImplemented('complex');\n  }\n  real<T extends Tensor>(input: T): T {\n    return notYetImplemented('real');\n  }\n  imag<T extends Tensor>(input: T): T {\n    return notYetImplemented('imag');\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    return notYetImplemented('cropAndResize');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n    return notYetImplemented('depthToSpace');\n  }\n\n  // Aligns with the \"SplitV\" kernel in TensorFlow.\n  split<T extends Tensor>(value: T, sizeSplits: number[], axis: number): T[] {\n    return notYetImplemented('split');\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    return notYetImplemented('sparseToDense');\n  }\n\n  diag(x: Tensor): Tensor {\n    return notYetImplemented('diag');\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    return notYetImplemented('fill');\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('onesLike');\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return notYetImplemented('zerosLike');\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return notYetImplemented('linspace');\n  }\n\n  dispose(): void {\n    return notYetImplemented('dispose');\n  }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n  throw new Error(\n      `'${kernelName}' not yet implemented or not found in the registry. ` +\n      `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise<FlagValue>);\nexport type Flags = {\n  [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n  evaluationFn: FlagEvaluationFn;\n  setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n  private flags: Flags = {};\n  private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n  private urlFlags: Flags = {};\n\n  platformName: string;\n  platform: Platform;\n\n  // tslint:disable-next-line: no-any\n  constructor(public global: any) {\n    this.populateURLFlags();\n  }\n\n  setPlatform(platformName: string, platform: Platform) {\n    if (this.platform != null) {\n      console.warn(\n          `Platform ${this.platformName} has already been set. ` +\n          `Overwriting the platform with ${platform}.`);\n    }\n    this.platformName = platformName;\n    this.platform = platform;\n  }\n\n  registerFlag(\n      flagName: string, evaluationFn: FlagEvaluationFn,\n      setHook?: (value: FlagValue) => void) {\n    this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n    // Override the flag value from the URL. This has to happen here because the\n    // environment is initialized before flags get registered.\n    if (this.urlFlags[flagName] != null) {\n      const flagValue = this.urlFlags[flagName];\n      console.warn(\n          `Setting feature override from URL ${flagName}: ${flagValue}.`);\n      this.set(flagName, flagValue);\n    }\n  }\n\n  async getAsync(flagName: string): Promise<FlagValue> {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    this.flags[flagName] = await this.evaluateFlag(flagName);\n    return this.flags[flagName];\n  }\n\n  get(flagName: string): FlagValue {\n    if (flagName in this.flags) {\n      return this.flags[flagName];\n    }\n\n    const flagValue = this.evaluateFlag(flagName);\n    if (flagValue instanceof Promise) {\n      throw new Error(\n          `Flag ${flagName} cannot be synchronously evaluated. ` +\n          `Please use getAsync() instead.`);\n    }\n\n    this.flags[flagName] = flagValue;\n\n    return this.flags[flagName];\n  }\n\n  getNumber(flagName: string): number {\n    return this.get(flagName) as number;\n  }\n\n  getBool(flagName: string): boolean {\n    return this.get(flagName) as boolean;\n  }\n\n  getFlags(): Flags {\n    return this.flags;\n  }\n  // For backwards compatibility.\n  get features(): Flags {\n    return this.flags;\n  }\n\n  set(flagName: string, value: FlagValue): void {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot set flag ${flagName} as it has not been registered.`);\n    }\n    this.flags[flagName] = value;\n    if (this.flagRegistry[flagName].setHook != null) {\n      this.flagRegistry[flagName].setHook(value);\n    }\n  }\n\n  private evaluateFlag(flagName: string): FlagValue|Promise<FlagValue> {\n    if (this.flagRegistry[flagName] == null) {\n      throw new Error(\n          `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n    }\n    return this.flagRegistry[flagName].evaluationFn();\n  }\n\n  setFlags(flags: Flags) {\n    this.flags = Object.assign({}, flags);\n  }\n\n  reset() {\n    this.flags = {};\n    this.urlFlags = {};\n    this.populateURLFlags();\n  }\n\n  private populateURLFlags(): void {\n    if (typeof this.global === 'undefined' ||\n        typeof this.global.location === 'undefined' ||\n        typeof this.global.location.search === 'undefined') {\n      return;\n    }\n\n    const urlParams = getQueryParams(this.global.location.search);\n    if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n      const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n      keyValues.forEach(keyValue => {\n        const [key, value] = keyValue.split(':') as [string, string];\n        this.urlFlags[key] = parseValue(key, value);\n      });\n    }\n  }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n  const params = {};\n  queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n    decodeParam(params, t[0], t[1]);\n    return t.join('=');\n  });\n  return params;\n}\n\nfunction decodeParam(\n    params: {[key: string]: string}, name: string, value?: string) {\n  params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n  value = value.toLowerCase();\n  if (value === 'true' || value === 'false') {\n    return value === 'true';\n  } else if (`${+ value}` === value) {\n    return +value;\n  }\n  throw new Error(\n      `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n  return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n  ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map<string, any>};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map<string, any>} {\n  if (globalNameSpace == null) {\n    // tslint:disable-next-line:no-any\n    let ns: any;\n    if (typeof (window) !== 'undefined') {\n      ns = window;\n    } else if (typeof (global) !== 'undefined') {\n      ns = global;\n    } else if (typeof (process) !== 'undefined') {\n      ns = process;\n    } else if (typeof (self) !== 'undefined') {\n      ns = self;\n    } else {\n      throw new Error('Could not find a global object');\n    }\n    globalNameSpace = ns;\n  }\n  return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map<string, any> {\n  const ns = getGlobalNamespace();\n  if (ns._tfGlobals == null) {\n    ns._tfGlobals = new Map();\n  }\n  return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n *             the first time it is fetched.\n */\nexport function getGlobal<T>(key: string, init: () => T): T {\n  const globalMap = getGlobalMap();\n  if (globalMap.has(key)) {\n    return globalMap.get(key);\n  } else {\n    const singleton = init();\n    globalMap.set(key, singleton);\n    return globalMap.get(key);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AllAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AnyAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMaxAttrs {\n  axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ArgMinAttrs {\n  axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface AvgPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick<NamedTensorInfoMap, 'dy'|'input'>;\nexport interface AvgPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\nexport interface BatchMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BatchToSpaceNDAttrs {\n  blockShape: number[];\n  crops: number[][];\n}\n\nexport type BinaryInputs = Pick<NamedTensorInfoMap, 'a'|'b'>;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface BroadCastToAttrs {\n  shape: number[];\n  inputShape: number[];  // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n  dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n  clipValueMin: number;\n  clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick<NamedTensorInfoMap, 'real'|'imag'>;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n  axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'dy'>;\nexport interface Conv2DBackpropFilterAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'|'filter'>;\nexport interface Conv2DBackpropInputAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n  inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Conv3DAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick<NamedTensorInfoMap, 'x'|'y'>;\n\nexport interface Conv3DBackpropFilterAttrs {\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick<NamedTensorInfoMap, 'dy'>;\nexport interface Conv3DBackpropInputAttrs {\n  pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface CumsumAttrs {\n  axis: number;\n  exclusive: boolean;\n  reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n    Pick<NamedTensorInfoMap, 'image'|'boxes'|'boxInd'>;\nexport interface CropAndResizeAttrs {\n  cropSize: [number, number];\n  method: 'bilinear'|'nearest';\n  extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface DepthToSpaceAttrs {\n  blockSize: number;\n  dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface DepthwiseConv2dNativeAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n    'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'dy'>;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n    'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'dy'>;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick<NamedTensorInfoMap, 'x'|'filter'>;\nexport interface Dilation2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n    Pick<NamedTensorInfoMap, 'x'|'filter'|'dy'>;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick<NamedTensorInfoMap, 'dy'|'y'>;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n  shape: number[];\n  value: number|string;\n  dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick<NamedTensorInfoMap, 'image'>;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n    Pick<NamedTensorInfoMap, 'x'|'scale'|'offset'|'mean'|'variance'>;\nexport interface FusedBatchNormAttrs {\n  varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick<NamedTensorInfoMap, 'x'|'indices'>;\nexport interface GatherV2Attrs {\n  axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick<NamedTensorInfoMap, 'params'|'indices'>;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n  start: number;\n  stop: number;\n  num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface LogSoftmaxAttrs {\n  axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface LRNAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick<NamedTensorInfoMap, 'x'|'y'|'dy'>;\nexport interface LRNBackpropAttrs {\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxAttrs {\n  reductionIndices: number|number[];\n  keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPoolBackpropAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPool3DAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NDHWC'|'NCDHW';\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n    Pick<NamedTensorInfoMap, 'dy'|'input'|'output'>;\nexport interface MaxPool3DBackpropAttrs {\n  filterSize: [number, number, number]|number;\n  strides: [number, number, number]|number;\n  pad: 'valid'|'same'|number;\n  dilations?: [number, number, number]|number;\n  dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MaxPoolWithArgmaxAttrs {\n  filterSize: [number, number]|number;\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MeanAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface MinAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV3Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV4Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n    Pick<NamedTensorInfoMap, 'boxes'|'scores'>;\nexport interface NonMaxSuppressionV5Attrs {\n  maxOutputSize: number;\n  iouThreshold: number;\n  scoreThreshold: number;\n  softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick<NamedTensorInfoMap, 'indices'>;\nexport interface OneHotAttrs {\n  depth: number;\n  onValue: number;\n  offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface PadV2Attrs {\n  paddings: Array<[number, number]>;\n  constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick<NamedTensorInfoMap, 'x'|'alpha'>;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ProdAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n  start: number;\n  stop: number;\n  step: number;\n  dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick<NamedTensorInfoMap, 'input'>;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReshapeAttrs {\n  shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeNearestNeighborAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n    Pick<NamedTensorInfoMap, 'images'>;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick<NamedTensorInfoMap, 'images'>;\nexport interface ResizeBilinearAttrs {\n  alignCorners: boolean;\n  size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick<NamedTensorInfoMap, 'images'>;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface ReverseAttrs {\n  dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick<NamedTensorInfoMap, 'indices'|'updates'>;\nexport interface ScatterNdAttrs {\n  shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick<NamedTensorInfoMap, 'condition'|'t'|'e'>;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SliceAttrs {\n  begin: number|number[];\n  size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SumAttrs {\n  axis: number|number[];\n  keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SpaceToBatchNDAttrs {\n  blockShape: number[];\n  paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface SplitVAttrs {\n  numOrSizeSplits: number[]|number;\n  axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick<NamedTensorInfoMap, 'logits'>;\nexport interface SoftmaxAttrs {\n  dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n    Pick<NamedTensorInfoMap, 'sparseIndices'|'sparseValues'|'defaultValue'>;\nexport interface SparseToDenseAttrs {\n  outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface StridedSliceAttrs {\n  begin: number[];\n  end: number[];\n  strides: number[];\n  beginMask: number;\n  endMask: number;\n  ellipsisMask: number;\n  newAxisMask: number;\n  shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TileAttrs {\n  reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TopKAttrs {\n  k: number;\n  sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface TransposeAttrs {\n  perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick<NamedTensorInfoMap, 'x'>;\nexport interface UniqueAttrs {\n  axis: number;\n}\n\nexport type UnaryInputs = Pick<NamedTensorInfoMap, 'x'>;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick<NamedTensorInfoMap, 'value'>;\nexport interface UnpackAttrs {\n  axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n    Pick<NamedTensorInfoMap, 'x'|'segmentIds'>;\nexport interface UnsortedSegmentSumAttrs {\n  numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n  alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n  pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n      HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n  numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick<NamedTensorInfoMap, 'image'>;\nexport interface RotateWithOffsetAttrs {\n  radians: number;\n  fillValue: number|[number, number, number];\n  center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n  a: TensorInfo;\n  b: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n  transposeA: boolean;\n  transposeB: boolean;\n  activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number|ExplicitPadding;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n  x: TensorInfo;\n  filter: TensorInfo;\n  bias?: TensorInfo;\n  preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n  strides: [number, number]|number;\n  pad: 'valid'|'same'|number;\n  dataFormat: 'NHWC'|'NCHW';\n  dilations: [number, number]|number;\n  dimRoundingMode: 'floor'|'round'|'ceil';\n  activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n    getGlobal('kernelRegistry', () => new Map<string, KernelConfig>());\nconst gradRegistry =\n    getGlobal('gradRegistry', () => new Map<string, GradConfig>());\n\nexport type DataId = object;\n\ntype AttributeValue =\n    number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray<AttributeValue>;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n  inputs: NamedTensorInfoMap,\n  backend: {},\n  attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n    (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n        NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n  kernelName: string;\n  backendName: string;\n  kernelFunc: KernelFunc;\n  setupFunc?: KernelSetupFunc;\n  disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n  kernelName: string;\n  inputsToSave?: string[];\n  // When saveAllInputs is true, all inputs will be saved. Only use this flag\n  // if inputs is an array of Tensors.\n  saveAllInputs?: boolean;\n  outputsToSave?: boolean[];\n  gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n  dataId: DataId;\n  shape: number[];\n  dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n  [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n  [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n    kernelName: string, backendName: string): KernelConfig {\n  const key = makeKey(kernelName, backendName);\n  return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n  return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n  const it = kernelRegistry.entries();\n  const result: KernelConfig[] = [];\n\n  while (true) {\n    const {done, value} = it.next();\n    if (done) {\n      break;\n    }\n    const [key, config] = value;\n    const [backend, ] = key.split('_');\n    if (backend === backendName) {\n      result.push(config);\n    }\n  }\n  return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n  const {kernelName, backendName} = config;\n  const key = makeKey(kernelName, backendName);\n  if (kernelRegistry.has(key)) {\n    console.warn(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is already registered`);\n  }\n  kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n  const {kernelName} = config;\n\n  if (gradRegistry.has(kernelName)) {\n    // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n    // to debug mode.\n    if (env().getBool('DEBUG')) {\n      console.warn(`Overriding the gradient for '${kernelName}'`);\n    }\n  }\n  gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n    kernelName: string, backendName: string): void {\n  const key = makeKey(kernelName, backendName);\n  if (!kernelRegistry.has(key)) {\n    throw new Error(\n        `The kernel '${kernelName}' for backend ` +\n        `'${backendName}' is not registered`);\n  }\n  kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n  if (!gradRegistry.has(kernelName)) {\n    throw new Error(\n        `The gradient '${kernelName}' for backend is not registered`);\n  }\n  gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n    registeredBackendName: string, newBackendName: string): void {\n  const kernels = getKernelsForBackend(registeredBackendName);\n  kernels.forEach(kernelConfig => {\n    const newKernelConfig =\n        Object.assign({}, kernelConfig, {backendName: newBackendName});\n    registerKernel(newKernelConfig);\n  });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n  return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n                        Float32Array): void {\n  let counter = array.length;\n  let temp = 0;\n  let index = 0;\n  // While there are elements in the array\n  while (counter > 0) {\n    // Pick a random index\n    index = (Math.random() * counter) | 0;\n    // Decrease counter by 1\n    counter--;\n    // And swap the last element with it\n    temp = array[counter];\n    array[counter] = array[index];\n    array[index] = temp;\n  }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n  return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n  return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n  let sum = 0;\n  for (let i = 0; i < arr.length; i++) {\n    sum += arr[i];\n  }\n  return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n  const r = Math.random();\n  return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n  let result = 0;\n  for (let i = 0; i < a.length; i++) {\n    const diff = Number(a[i]) - Number(b[i]);\n    result += diff * diff;\n  }\n  return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n *     error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n  if (!expr) {\n    throw new Error(typeof msg === 'string' ? msg : msg());\n  }\n}\n\nexport function assertShapesMatch(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  assert(\n      arraysEqual(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n  assert(\n      a != null,\n      () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten<number>().\n/**\n *  Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n *  @param arr The nested array to flatten.\n *  @param result The destination array which holds the elements.\n *  @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n *      to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten<T extends number|boolean|string|Promise<number>|TypedArray>(\n    arr: T|RecursiveArray<T>, result: T[] = [], skipTypedArray = false): T[] {\n  if (result == null) {\n    result = [];\n  }\n  if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n    for (let i = 0; i < arr.length; ++i) {\n      flatten(arr[i], result, skipTypedArray);\n    }\n  } else {\n    result.push(arr as T);\n  }\n  return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n  if (shape.length === 0) {\n    // Scalar.\n    return 1;\n  }\n  let size = shape[0];\n  for (let i = 1; i < shape.length; i++) {\n    size *= shape[i];\n  }\n  return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n  return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n  if (n1 === n2) {\n    return true;\n  }\n  if (n1 == null || n2 == null) {\n    return false;\n  }\n\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function isInt(a: number): boolean {\n  return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n  // tslint:disable-next-line:no-any\n  if ((Math as any).tanh != null) {\n    // tslint:disable-next-line:no-any\n    return (Math as any).tanh(x);\n  }\n  if (x === Infinity) {\n    return 1;\n  } else if (x === -Infinity) {\n    return -1;\n  } else {\n    const e2x = Math.exp(2 * x);\n    return (e2x - 1) / (e2x + 1);\n  }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n  const width = Math.ceil(Math.sqrt(size));\n  return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n  const shuffledIndices = new Uint32Array(n);\n  for (let i = 0; i < n; ++i) {\n    shuffledIndices[i] = i;\n  }\n  shuffle(shuffledIndices);\n  return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n  if (size <= a.length) {\n    return a;\n  }\n  return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n    checkFn: () => boolean, delayFn = (counter: number) => 0,\n    maxCounter?: number): Promise<void> {\n  return new Promise<void>((resolve, reject) => {\n    let tryCount = 0;\n\n    const tryFn = () => {\n      if (checkFn()) {\n        resolve();\n        return;\n      }\n\n      tryCount++;\n\n      const nextBackoff = delayFn(tryCount);\n\n      if (maxCounter != null && tryCount >= maxCounter) {\n        reject();\n        return;\n      }\n      setTimeout(tryFn, nextBackoff);\n    };\n\n    tryFn();\n  });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n    shape: number[], size: number): number[] {\n  let shapeProd = 1;\n  let implicitIdx = -1;\n\n  for (let i = 0; i < shape.length; ++i) {\n    if (shape[i] >= 0) {\n      shapeProd *= shape[i];\n    } else if (shape[i] === -1) {\n      if (implicitIdx !== -1) {\n        throw Error(\n            `Shapes can only have 1 implicit size. ` +\n            `Found -1 at dim ${implicitIdx} and dim ${i}`);\n      }\n      implicitIdx = i;\n    } else if (shape[i] < 0) {\n      throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n    }\n  }\n\n  if (implicitIdx === -1) {\n    if (size > 0 && size !== shapeProd) {\n      throw Error(`Size(${size}) must match the product of shape ${shape}`);\n    }\n    return shape;\n  }\n\n  if (shapeProd === 0) {\n    throw Error(\n        `Cannot infer the missing size in [${shape}] when ` +\n        `there are 0 elements`);\n  }\n  if (size % shapeProd !== 0) {\n    throw Error(\n        `The implicit shape can't be a fractional number. ` +\n        `Got ${size} / ${shapeProd}`);\n  }\n\n  const newShape = shape.slice();\n  newShape[implicitIdx] = size / shapeProd;\n  return newShape;\n}\n\nexport function parseAxisParam(\n    axis: number|number[], shape: number[]): number[] {\n  const rank = shape.length;\n\n  // Normalize input\n  axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n  // Check for valid range\n  assert(\n      axis.every(ax => ax >= -rank && ax < rank),\n      () =>\n          `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n          `got axis ${axis}`);\n\n  // Check for only integers\n  assert(\n      axis.every(ax => isInt(ax)),\n      () => `All values in axis param must be integers but ` +\n          `got axis ${axis}`);\n\n  // Handle negative axis.\n  return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n    {newShape: number[], keptDims: number[]} {\n  const newShape: number[] = [];\n  const keptDims: number[] = [];\n  const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n  const axes = (axis == null || isEmptyArray) ?\n      null :\n      parseAxisParam(axis, shape).sort();\n  let j = 0;\n  for (let i = 0; i < shape.length; ++i) {\n    if (axes != null) {\n      if (axes[j] === i && shape[i] !== 1) {\n        throw new Error(\n            `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n      }\n      if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n        newShape.push(shape[i]);\n        keptDims.push(i);\n      }\n      if (axes[j] <= i) {\n        j++;\n      }\n    }\n    if (shape[i] !== 1) {\n      newShape.push(shape[i]);\n      keptDims.push(i);\n    }\n  }\n  return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType<D extends NumericDataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType<D extends DataType>(\n    dtype: D, size: number): DataTypeMap[D] {\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else if (dtype === 'string') {\n    values = new Array<'string'>(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors<D extends DataType>(\n    vals: DataTypeMap[D]|number[], dtype: D): void {\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n    }\n  }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n  return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n      dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n  if (newType === 'complex64') {\n    return false;\n  }\n  if (newType === 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n    return false;\n  }\n  if (newType === 'bool' && oldType === 'bool') {\n    return false;\n  }\n  return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n  return a instanceof Float32Array || a instanceof Int32Array ||\n      a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n  if (dtype === 'float32' || dtype === 'int32') {\n    return 4;\n  } else if (dtype === 'complex64') {\n    return 8;\n  } else if (dtype === 'bool') {\n    return 1;\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n  if (arr == null) {\n    return 0;\n  }\n  let bytes = 0;\n  arr.forEach(x => bytes += x.length);\n  return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n  return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n  return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n  return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n  if (Array.isArray(values)) {\n    return inferDtype(values[0]);\n  }\n  if (values instanceof Float32Array) {\n    return 'float32';\n  } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n    return 'int32';\n  } else if (isNumber(values)) {\n    return 'float32';\n  } else if (isString(values)) {\n    return 'string';\n  } else if (isBoolean(values)) {\n    return 'bool';\n  }\n  return 'float32';\n}\n\nexport function isFunction(f: Function) {\n  return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n  for (let i = start; i < size; ++i) {\n    if (size % i === 0) {\n      return i;\n    }\n  }\n  return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n  const rank = shape.length;\n  if (rank < 2) {\n    return [];\n  }\n\n  // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n  // strides.\n  const strides = new Array(rank - 1);\n  strides[rank - 2] = shape[rank - 1];\n  for (let i = rank - 3; i >= 0; --i) {\n    strides[i] = strides[i + 1] * shape[i + 1];\n  }\n  return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n    value: DataType, dtype: DataType): BackendValues {\n  if (dtype === 'string') {\n    return encodeString(value);\n  }\n\n  return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n  if (dtype === 'string') {\n    throw new Error('Cannot convert a string[] to a TypedArray');\n  }\n  if (Array.isArray(a)) {\n    a = flatten(a);\n  }\n\n  if (env().getBool('DEBUG')) {\n    checkConversionForErrors(a as number[], dtype);\n  }\n  if (noConversionNeeded(a, dtype)) {\n    return a as TypedArray;\n  }\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(a as number[]);\n  } else if (dtype === 'int32') {\n    return new Int32Array(a as number[]);\n  } else if (dtype === 'bool') {\n    const bool = new Uint8Array((a as number[]).length);\n    for (let i = 0; i < bool.length; ++i) {\n      if (Math.round((a as number[])[i]) !== 0) {\n        bool[i] = 1;\n      }\n    }\n    return bool;\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n  const ret = new Array();\n  if (shape.length === 1) {\n    const d = shape[0];\n    for (let i = 0; i < d; i++) {\n      ret[i] = a[offset + i];\n    }\n  } else {\n    const d = shape[0];\n    const rest = shape.slice(1);\n    const len = rest.reduce((acc, c) => acc * c);\n    for (let i = 0; i < d; i++) {\n      ret[i] = createNestedArray(offset + i * len, rest, a);\n    }\n  }\n  return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n  if (shape.length === 0) {\n    // Scalar type should return a single number.\n    return a[0];\n  }\n  const size = shape.reduce((acc, c) => acc * c);\n  if (size === 0) {\n    // A tensor with shape zero should be turned into empty list.\n    return [];\n  }\n  if (size !== a.length) {\n    throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n  }\n\n  return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n  return (a instanceof Float32Array && dtype === 'float32') ||\n      (a instanceof Int32Array && dtype === 'int32') ||\n      (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  const array = makeZerosTypedArray(size, dtype);\n  for (let i = 0; i < array.length; i++) {\n    array[i] = 1;\n  }\n  return array;\n}\n\nexport function makeZerosTypedArray<D extends DataType>(\n    size: number, dtype: D): DataTypeMap[D] {\n  if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n    return new Float32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'int32') {\n    return new Int32Array(size) as DataTypeMap[D];\n  } else if (dtype === 'bool') {\n    return new Uint8Array(size) as DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray<D extends DataType>(\n    shape: number[], dtype: D) {\n  const size = shape.reduce((prev, curr) => prev * curr, 1);\n  if (dtype == null || dtype === 'float32') {\n    return toNestedArray(shape, new Float32Array(size));\n  } else if (dtype === 'int32') {\n    return toNestedArray(shape, new Int32Array(size));\n  } else if (dtype === 'bool') {\n    return toNestedArray(shape, new Uint8Array(size));\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n  return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n  shape.forEach(dimSize => {\n    assert(\n        Number.isInteger(dimSize) && dimSize >= 0,\n        () =>\n            `Tensor must have a shape comprised of positive integers but got ` +\n            `shape [${shape}].`);\n  });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n    path: string, requestInits?: RequestInit): Promise<Response> {\n  return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n  encoding = encoding || 'utf-8';\n  return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n  encoding = encoding || 'utf-8';\n  return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n    locs: number[], rank: number, strides: number[]): number {\n  if (rank === 0) {\n    return 0;\n  } else if (rank === 1) {\n    return locs[0];\n  }\n  let index = locs[locs.length - 1];\n  for (let i = 0; i < locs.length - 1; ++i) {\n    index += strides[i] * locs[i];\n  }\n  return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n    index: number, rank: number, strides: number[]): number[] {\n  if (rank === 0) {\n    return [];\n  } else if (rank === 1) {\n    return [index];\n  }\n  const locs: number[] = new Array(rank);\n  for (let i = 0; i < locs.length - 1; ++i) {\n    locs[i] = Math.floor(index / strides[i]);\n    index -= locs[i] * strides[i];\n  }\n  locs[locs.length - 1] = index;\n  return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n  kernelName: string,\n  outputs: Tensor[],\n  inputs: NamedTensorMap,\n  timeMs: Promise<number|{error: string}>,\n  extraInfo: Promise<string>\n};\n\nexport class Profiler {\n  constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n    if (logger == null) {\n      this.logger = new Logger();\n    }\n  }\n\n  profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n      KernelProfile {\n    let outputs: Tensor[];\n    const holdResultWrapperFn = () => {\n      outputs = f();\n    };\n    const timer = this.backendTimer.time(holdResultWrapperFn);\n\n    for (let i = 0; i < outputs.length; i++) {\n      const output = outputs[i];\n      // Dangling promise here because we don't want to propagate up\n      // asynchronicity.\n      output.data().then(tensorVals => {\n        checkComputationForErrors(tensorVals, output.dtype, kernelName);\n      });\n    }\n\n    const kernelProfile = {\n      kernelName,\n      outputs,\n      inputs,\n      timeMs: timer.then(timing => timing.kernelMs),\n      extraInfo: timer.then(\n          timing => timing.getExtraProfileInfo != null ?\n              timing.getExtraProfileInfo() :\n              '')\n    };\n    return kernelProfile;\n  }\n\n  logKernelProfile(kernelProfile: KernelProfile): void {\n    const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n    outputs.forEach(result => {\n      Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n        this.logger.logKernelProfile(\n            kernelName, result, valueContainer[0], valueContainer[1], inputs,\n            valueContainer[2]);\n      });\n    });\n  }\n}\n\nexport function checkComputationForErrors<D extends DataType>(\n    vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n  if (dtype !== 'float32') {\n    // Only floating point computations will generate NaN values\n    return false;\n  }\n  for (let i = 0; i < vals.length; i++) {\n    const num = vals[i] as number;\n    if (isNaN(num) || !isFinite(num)) {\n      // Throwing custom exception so behavior is testable.\n      console.warn(`Found ${num} in the result of '${kernelName}'`);\n      return true;\n    }\n  }\n  return false;\n}\n\nexport class Logger {\n  logKernelProfile(\n      name: string, result: Tensor, vals: TypedArray,\n      timeMs: number|{error: string}, inputs: NamedTensorMap,\n      extraInfo?: string) {\n    const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n                                              timeMs['error'];\n    const paddedName = util.rightPad(name, 25);\n    const rank = result.rank;\n    const size = result.size;\n    const shape = util.rightPad(result.shape.toString(), 14);\n    let inputShapesDescription = '';\n\n    for (const name in inputs) {\n      const input = inputs[name];\n      if (input != null) {\n        // The input might be a non-tensor (e.g HTMLImageElement), in which case\n        // we claim the output shape as input shape.\n        const inputShape = input.shape || result.shape;\n        const inputRank = inputShape.length;\n        inputShapesDescription +=\n            `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n      }\n    }\n\n    console.log(\n        `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n            inputShapesDescription}\\t%c${extraInfo}`,\n        'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n        'color: green', 'color: steelblue');\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n  id: number;\n  kernelName: string;\n  outputs: Tensor[];\n  inputs: NamedTensorMap;\n  // Optional params, defined only for ops with gradient impl.\n  gradient?: (dys: Tensor[]) => NamedGradientMap;\n  saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n  [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n    tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n  // Forward pass to compute all the nodes and Tensors that are transitively a\n  // function of x.\n  const tensorsFromX: {[tensorId: number]: boolean} = {};\n  const nodesFromX: {[nodeId: number]: boolean} = {};\n  for (let i = 0; i < xs.length; i++) {\n    tensorsFromX[xs[i].id] = true;\n  }\n\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n    for (const inputName in nodeInputs) {\n      const input = nodeInputs[inputName];\n\n      let anyInputFromX = false;\n      for (let j = 0; j < xs.length; j++) {\n        if (tensorsFromX[input.id]) {\n          node.outputs.forEach(output => tensorsFromX[output.id] = true);\n          anyInputFromX = true;\n          nodesFromX[node.id] = true;\n          break;\n        }\n      }\n\n      if (anyInputFromX) {\n        break;\n      }\n    }\n  }\n\n  // Backward pass to find all of the nodes and Tensors that lead to y.\n  const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n  tensorsLeadToY[y.id] = true;\n  const nodesToY: {[nodeId: number]: boolean} = {};\n\n  for (let i = tape.length - 1; i >= 0; i--) {\n    const node = tape[i];\n    const nodeInputs = node.inputs;\n\n    // If any of the outputs lead to y, mark all of the inputs as leading to y.\n    for (let j = 0; j < node.outputs.length; j++) {\n      if (tensorsLeadToY[node.outputs[j].id]) {\n        for (const inputName in nodeInputs) {\n          tensorsLeadToY[nodeInputs[inputName].id] = true;\n          nodesToY[node.id] = true;\n        }\n        break;\n      }\n    }\n  }\n\n  // Return the paths that come from x and lead to y.\n  const filteredTape: TapeNode[] = [];\n  for (let i = 0; i < tape.length; i++) {\n    const node = tape[i];\n\n    if (nodesFromX[node.id] && nodesToY[node.id]) {\n      // Prune the inputs from the node that aren't a function of x.\n      const prunedInputs: {[inputName: string]: Tensor} = {};\n      for (const inputName in node.inputs) {\n        const nodeInput = node.inputs[inputName];\n        if (tensorsFromX[nodeInput.id]) {\n          prunedInputs[inputName] = nodeInput;\n        }\n      }\n\n      // Copy the node and overwrite inputsAndArgs to the pruned version.\n      const prunedNode = Object.assign({}, node);\n      prunedNode.inputs = prunedInputs;\n      prunedNode.outputs = node.outputs;\n\n      filteredTape.push(prunedNode);\n    }\n  }\n\n  return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n    tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n    filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n    add: (a: Tensor, b: Tensor) => Tensor) {\n  // Walk the tape backward and keep a map of Tensor to its gradient.\n  for (let i = filteredTape.length - 1; i >= 0; i--) {\n    const node = filteredTape[i];\n\n    const dys: Tensor[] = [];\n    node.outputs.forEach(o => {\n      const gradTensor = tensorAccumulatedGradientMap[o.id];\n      if (gradTensor != null) {\n        dys.push(gradTensor);\n      } else {\n        // This particular output is not in the back-propagation subgraph, so it\n        // does not affect the final output, thus we put null for its dy.\n        dys.push(null);\n      }\n    });\n\n    if (node.gradient == null) {\n      throw new Error(\n          `Cannot compute gradient: gradient function not found ` +\n          `for ${node.kernelName}.`);\n    }\n\n    // Backprop dy through this node and accumulate gradients over the inputs.\n    const inputGradients = node.gradient(dys);\n\n    for (const inputName in node.inputs) {\n      if (!(inputName in inputGradients)) {\n        throw new Error(\n            `Cannot backprop through input ${inputName}. ` +\n            `Available gradients found: ${Object.keys(inputGradients)}.`);\n      }\n\n      // Call the gradient function.\n      const dx = tidy(() => inputGradients[inputName]());\n      if (dx.dtype !== 'float32') {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n      }\n      const x = node.inputs[inputName];\n      if (!util.arraysEqual(dx.shape, x.shape)) {\n        throw new Error(\n            `Error in gradient for op ${\n                node.kernelName}. The gradient of input ` +\n            `'${inputName}' has shape '${dx.shape}', which does not match ` +\n            `the shape of the input '${x.shape}'`);\n      }\n\n      if (tensorAccumulatedGradientMap[x.id] == null) {\n        tensorAccumulatedGradientMap[x.id] = dx;\n      } else {\n        const curGradient = tensorAccumulatedGradientMap[x.id];\n        tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n        curGradient.dispose();\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    verbose: boolean) {\n  const strides = computeStrides(shape);\n  const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n  const rank = shape.length;\n  const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n  const lines = ['Tensor'];\n  if (verbose) {\n    lines.push(`  dtype: ${dtype}`);\n    lines.push(`  rank: ${rank}`);\n    lines.push(`  shape: [${shape}]`);\n    lines.push(`  values:`);\n  }\n  lines.push(valsLines.map(l => '    ' + l).join('\\n'));\n  return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[]): number[] {\n  const n = sizeFromShape(shape);\n  const numCols = strides[strides.length - 1];\n  const padPerCol = new Array(numCols).fill(0);\n  const rank = shape.length;\n  const valuesOrTuples =\n      dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n  if (rank > 1) {\n    for (let row = 0; row < n / numCols; row++) {\n      const offset = row * numCols;\n      for (let j = 0; j < numCols; j++) {\n        padPerCol[j] = Math.max(\n            padPerCol[j],\n            valToString(valuesOrTuples[offset + j], 0, dtype).length);\n      }\n    }\n  }\n  return padPerCol;\n}\n\nfunction valToString(\n    val: number|string|[number, number], pad: number, dtype: DataType) {\n  let valStr: string;\n  if (Array.isArray(val)) {\n    valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n        `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n  } else if (isString(val)) {\n    valStr = `'${val}'`;\n  } else if (dtype === 'bool') {\n    valStr = boolNumToString(val);\n  } else {\n    valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n  }\n\n  return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n  return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n    vals: TypedArray|string[], shape: number[], dtype: DataType,\n    strides: number[], padPerCol: number[], isLast = true): string[] {\n  const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n  const size = shape[0];\n  const rank = shape.length;\n  if (rank === 0) {\n    if (dtype === 'complex64') {\n      const complexTuple = createComplexTuples(vals);\n      return [valToString(complexTuple[0], 0, dtype)];\n    }\n    if (dtype === 'bool') {\n      return [boolNumToString(vals[0] as number)];\n    }\n    return [vals[0].toString()];\n  }\n\n  if (rank === 1) {\n    if (size > FORMAT_LIMIT_NUM_VALS) {\n      const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n      let firstVals = Array.from<number|string|[number, number]>(\n          vals.slice(0, firstValsSize));\n      let lastVals = Array.from<number|string|[number, number]>(vals.slice(\n          (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n          size * storagePerElement));\n      if (dtype === 'complex64') {\n        firstVals = createComplexTuples(firstVals);\n        lastVals = createComplexTuples(lastVals);\n      }\n      return [\n        '[' +\n        firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n            .join(', ') +\n        ', ..., ' +\n        lastVals\n            .map(\n                (x, i) => valToString(\n                    x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n            .join(', ') +\n        ']'\n      ];\n    }\n    const displayVals: Array<number|string|[number, number]> =\n        dtype === 'complex64' ? createComplexTuples(vals) :\n                                Array.from<number|string>(vals);\n\n    return [\n      '[' +\n      displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n          .join(', ') +\n      ']'\n    ];\n  }\n\n  // The array is rank 2 or more.\n  const subshape = shape.slice(1);\n  const substrides = strides.slice(1);\n  const stride = strides[0] * storagePerElement;\n  const lines: string[] = [];\n  if (size > FORMAT_LIMIT_NUM_VALS) {\n    for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          false /* isLast */));\n    }\n    lines.push('...');\n    for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  } else {\n    for (let i = 0; i < size; i++) {\n      const start = i * stride;\n      const end = start + stride;\n      lines.push(...subTensorToString(\n          vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n          i === size - 1 /* isLast */));\n    }\n  }\n  const sep = rank === 2 ? ',' : '';\n  lines[0] = '[' + lines[0] + sep;\n  for (let i = 1; i < lines.length - 1; i++) {\n    lines[i] = ' ' + lines[i] + sep;\n  }\n  let newLineSep = ',\\n';\n  for (let i = 2; i < rank; i++) {\n    newLineSep += '\\n';\n  }\n  lines[lines.length - 1] =\n      ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n  return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n                             TypedArray): Array<[number, number]> {\n  const complexTuples: Array<[number, number]> = [];\n  for (let i = 0; i < vals.length; i += 2) {\n    complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n  }\n  return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData<D extends DataType> {\n  dataId?: DataId;\n  values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer<R extends Rank, D extends DataType = 'float32'> {\n  size: number;\n  shape: ShapeMap[R];\n  strides: number[];\n  values: DataTypeMap[D];\n\n  constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.size = util.sizeFromShape(shape);\n\n    if (values != null) {\n      const n = values.length;\n      util.assert(\n          n === this.size,\n          () => `Length of values '${n}' does not match the size ` +\n              `inferred by the shape '${this.size}'.`);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `complex64 dtype TensorBuffers are not supported. Please create ` +\n          `a TensorBuffer for the real and imaginary parts separately and ` +\n          `call tf.complex(real, imag).`);\n    }\n    this.values = values || util.getArrayFromDType(dtype, this.size);\n    this.strides = computeStrides(shape);\n  }\n\n  /**\n   * Sets a value in the buffer at a given location.\n   *\n   * @param value The value to set.\n   * @param locs  The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  set(value: SingleValueMap[D], ...locs: number[]): void {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    util.assert(\n        locs.length === this.rank,\n        () => `The number of provided coordinates (${locs.length}) must ` +\n            `match the rank (${this.rank})`);\n\n    const index = this.locToIndex(locs);\n    this.values[index] = value as number;\n  }\n\n  /**\n   * Returns the value in the buffer at the provided location.\n   *\n   * @param locs The location indices.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  get(...locs: number[]): SingleValueMap[D] {\n    if (locs.length === 0) {\n      locs = [0];\n    }\n    let i = 0;\n    for (const loc of locs) {\n      if (loc < 0 || loc >= this.shape[i]) {\n        const msg = `Requested out of range element at ${locs}. ` +\n            `  Buffer shape=${this.shape}`;\n        throw new Error(msg);\n      }\n      i++;\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return this.values[index] as SingleValueMap[D];\n  }\n\n  locToIndex(locs: number[]): number {\n    if (this.rank === 0) {\n      return 0;\n    } else if (this.rank === 1) {\n      return locs[0];\n    }\n    let index = locs[locs.length - 1];\n    for (let i = 0; i < locs.length - 1; ++i) {\n      index += this.strides[i] * locs[i];\n    }\n    return index;\n  }\n\n  indexToLoc(index: number): number[] {\n    if (this.rank === 0) {\n      return [];\n    } else if (this.rank === 1) {\n      return [index];\n    }\n    const locs: number[] = new Array(this.shape.length);\n    for (let i = 0; i < locs.length - 1; ++i) {\n      locs[i] = Math.floor(index / this.strides[i]);\n      index -= locs[i] * this.strides[i];\n    }\n    locs[locs.length - 1] = index;\n    return locs;\n  }\n\n  get rank() {\n    return this.shape.length;\n  }\n\n  /**\n   * Creates an immutable `tf.Tensor` object from the buffer.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Creation'}\n   */\n  toTensor(): Tensor<R> {\n    return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n        Tensor<R>;\n  }\n}\n\nexport interface TensorTracker {\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: Backend): Tensor;\n  makeVariable(\n      initialValue: Tensor, trainable?: boolean, name?: string,\n      dtype?: DataType): Variable;\n  incRef(a: Tensor, backend: Backend): void;\n  disposeTensor(t: Tensor): void;\n  disposeVariable(v: Variable): void;\n  read(dataId: DataId): Promise<BackendValues>;\n  readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n  cast<T extends Tensor>(x: T, dtype: DataType): T;\n  buffer<R extends Rank, D extends DataType>(\n      shape: ShapeMap[R], dtype: D,\n      values?: DataTypeMap[D]): TensorBuffer<R, D>;\n  print<T extends Tensor>(x: T, verbose: boolean): void;\n  clone<T extends Tensor>(x: T): T;\n  // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n  trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n  opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n  deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object;  // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor<R extends Rank = Rank> {\n  /** Unique id of this tensor. */\n  readonly id: number;\n  /**\n   * Id of the bucket holding the data for this tensor. Multiple arrays can\n   * point to the same bucket (e.g. when calling array.reshape()).\n   */\n  dataId: DataId;\n  /** The shape of the tensor. */\n  readonly shape: ShapeMap[R];\n  /** Number of elements in the tensor. */\n  readonly size: number;\n  /** The data type for the array. */\n  readonly dtype: DataType;\n  /** The rank type for the array (see `Rank` enum). */\n  readonly rankType: R;\n\n  /** Whether this tensor has been globally kept. */\n  kept = false;\n  /** The id of the scope this tensor is being tracked in. */\n  scopeId: number;\n\n  /**\n   * Number of elements to skip in each dimension when indexing. See\n   * https://docs.scipy.org/doc/numpy/reference/generated/\\\n   * numpy.ndarray.strides.html\n   */\n  readonly strides: number[];\n\n  constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n    this.shape = shape.slice() as ShapeMap[R];\n    this.dtype = dtype || 'float32';\n    this.size = util.sizeFromShape(shape);\n    this.strides = computeStrides(shape);\n    this.dataId = dataId;\n    this.id = id;\n    this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n  }\n\n  get rank(): number {\n    return this.shape.length;\n  }\n\n  /**\n   * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async buffer<D extends DataType = 'float32'>(): Promise<TensorBuffer<R, D>> {\n    const vals = await this.data<D>();\n    return opHandler.buffer(this.shape, this.dtype as D, vals);\n  }\n\n  /**\n   * Returns a `tf.TensorBuffer` that holds the underlying data.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  bufferSync<D extends DataType = 'float32'>(): TensorBuffer<R, D> {\n    return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * asynchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async array(): Promise<ArrayMap[R]> {\n    const vals = await this.data();\n    return toNestedArray(this.shape, vals) as ArrayMap[R];\n  }\n\n  /**\n   * Returns the tensor data as a nested array. The transfer of data is done\n   * synchronously.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  arraySync(): ArrayMap[R] {\n    return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n  }\n\n  /**\n   * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n   * promise of `TypedArray` that resolves when the computation has finished.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  async data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]> {\n    this.throwIfDisposed();\n    const data = trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      const bytes = await data as Uint8Array[];\n      try {\n        return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as Promise<DataTypeMap[D]>;\n  }\n\n  /**\n   * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n   * UI thread until the values are ready, which can cause performance issues.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D] {\n    this.throwIfDisposed();\n    const data = trackerFn().readSync(this.dataId);\n    if (this.dtype === 'string') {\n      try {\n        return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n            DataTypeMap[D];\n      } catch {\n        throw new Error(\n            'Failed to decode the string bytes into utf-8. ' +\n            'To get the original bytes, call tensor.bytes().');\n      }\n    }\n    return data as DataTypeMap[D];\n  }\n\n  /** Returns the underlying bytes of the tensor's data. */\n  async bytes(): Promise<Uint8Array[]|Uint8Array> {\n    this.throwIfDisposed();\n    const data = await trackerFn().read(this.dataId);\n    if (this.dtype === 'string') {\n      return data as Uint8Array[];\n    } else {\n      return new Uint8Array((data as TypedArray).buffer);\n    }\n  }\n\n  /**\n   * Disposes `tf.Tensor` from memory.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  dispose(): void {\n    if (this.isDisposed) {\n      return;\n    }\n    trackerFn().disposeTensor(this);\n    this.isDisposedInternal = true;\n  }\n\n  protected isDisposedInternal = false;\n  get isDisposed(): boolean {\n    return this.isDisposedInternal;\n  }\n\n  throwIfDisposed() {\n    if (this.isDisposed) {\n      throw new Error(`Tensor is disposed.`);\n    }\n  }\n\n  /**\n   * Prints the `tf.Tensor`. See `tf.print` for details.\n   *\n   * @param verbose Whether to print verbose information about the tensor,\n   *    including dtype and size.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  print(verbose = false): void {\n    return opHandler.print(this, verbose);\n  }\n\n  /**\n   * Returns a copy of the tensor. See `tf.clone` for details.\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  clone<T extends Tensor>(this: T): T {\n    this.throwIfDisposed();\n    return opHandler.clone(this);\n  }\n\n  /**\n   * Returns a human-readable description of the tensor. Useful for logging.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  toString(verbose = false): string {\n    const vals = this.dataSync();\n    return tensorToString(vals, this.shape, this.dtype, verbose);\n  }\n\n  cast<T extends this>(dtype: DataType): T {\n    this.throwIfDisposed();\n    return opHandler.cast(this as T, dtype);\n  }\n  variable(trainable = true, name?: string, dtype?: DataType): Variable<R> {\n    this.throwIfDisposed();\n    return trackerFn().makeVariable(this, trainable, name, dtype) as\n        Variable<R>;\n  }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n  value: (instance: Tensor) => {\n    // Implementation note: we should use properties of the object that will be\n    // defined before the constructor body has finished executing (methods).\n    // This is because when this code is transpiled by babel, babel will call\n    // classCallCheck before the constructor body is run.\n    // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n    return !!instance && instance.data != null && instance.dataSync != null &&\n        instance.throwIfDisposed != null;\n  }\n});\n\nexport interface NumericTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: NumericDataType;\n  dataSync<D extends DataType = NumericDataType>(): DataTypeMap[D];\n  data<D extends DataType = NumericDataType>(): Promise<DataTypeMap[D]>;\n}\n\nexport interface StringTensor<R extends Rank = Rank> extends Tensor<R> {\n  dtype: 'string';\n  dataSync<D extends DataType = 'string'>(): DataTypeMap[D];\n  data<D extends DataType = 'string'>(): Promise<DataTypeMap[D]>;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor<Rank.R0>;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor<Rank.R1>;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor<Rank.R2>;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor<Rank.R3>;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor<Rank.R4>;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor<Rank.R5>;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor<Rank.R6>;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable<R extends Rank = Rank> extends Tensor<R> {\n  name: string;\n\n  constructor(\n      initialValue: Tensor<R>, public trainable: boolean, name: string,\n      tensorId: number) {\n    super(\n        initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n    this.name = name;\n  }\n\n  /**\n   * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n   * the same shape and dtype as the old `tf.Tensor`.\n   *\n   * @param newValue New tensor to be assigned to this variable.\n   *\n   * @doc {heading: 'Tensors', subheading: 'Classes'}\n   */\n  assign(newValue: Tensor<R>): void {\n    if (newValue.dtype !== this.dtype) {\n      throw new Error(\n          `dtype of the new value (${newValue.dtype}) and ` +\n          `previous value (${this.dtype}) must match`);\n    }\n    if (!util.arraysEqual(newValue.shape, this.shape)) {\n      throw new Error(\n          `shape of the new value (${newValue.shape}) and ` +\n          `previous value (${this.shape}) must match`);\n    }\n    trackerFn().disposeTensor(this);\n    this.dataId = newValue.dataId;\n    trackerFn().incRef(this, null /* backend */);\n  }\n\n  dispose(): void {\n    trackerFn().disposeVariable(this);\n    this.isDisposedInternal = true;\n  }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n  value: (instance: Variable) => {\n    return instance instanceof Tensor && instance.assign != null &&\n        instance.assign instanceof Function;\n  }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n  R0: number[];\n  R1: [number];\n  R2: [number, number];\n  R3: [number, number, number];\n  R4: [number, number, number, number];\n  R5: [number, number, number, number, number];\n  R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n  R0: number;\n  R1: number[];\n  R2: number[][];\n  R3: number[][][];\n  R4: number[][][][];\n  R5: number[][][][][];\n  R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n  float32: Float32Array;\n  int32: Int32Array;\n  bool: Uint8Array;\n  complex64: Float32Array;\n  string: string[];\n}\n\nexport interface SingleValueMap {\n  bool: boolean;\n  int32: number;\n  float32: number;\n  complex64: number;\n  string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n  R0 = 'R0',\n  R1 = 'R1',\n  R2 = 'R2',\n  R3 = 'R3',\n  R4 = 'R4',\n  R5 = 'R5',\n  R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray<T> =\n    T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray<T extends any> {\n  [index: number]: T|RecursiveArray<T>;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'int32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n  'float32' = 'float32',\n  'int32' = 'int32',\n  'bool' = 'bool',\n  'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n  'float32' = 'float32',\n  'int32' = 'float32',\n  'bool' = 'float32',\n  'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n  'float32' = 'complex64',\n  'int32' = 'complex64',\n  'bool' = 'complex64',\n  'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n  'float32': UpcastFloat32AndMap,\n  'int32': UpcastInt32AndMap,\n  'bool': UpcastBoolAndMap,\n  'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n  if (typeA === 'string' || typeB === 'string') {\n    if (typeA === 'string' && typeB === 'string') {\n      return 'string';\n    }\n    throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n  }\n  return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n  return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n    TypedArray|number|boolean|string|RecursiveArray<number|number[]|TypedArray>|\n    RecursiveArray<boolean>|RecursiveArray<string>|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n    string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n    boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n    boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n    TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n    string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n    TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n    string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n  width: number;\n  height: number;\n  data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch<T extends Tensor>(a: T, b: T): [T, T] {\n  if (a.dtype === b.dtype) {\n    return [a, b];\n  }\n  const dtype = upcastType(a.dtype, b.dtype);\n  return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n  assert(\n      a.dtype === b.dtype,\n      () => `The dtypes of the first(${a.dtype}) and` +\n          ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n  return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n *   `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n *   is safe to pass any object here, except that `Promise`s are not\n *   supported.\n * @returns An array of `Tensors` found within the passed object. If the\n *   argument is simply a `Tensor', a list containing that `Tensor` is\n *   returned. If the object is not a `Tensor` or does not\n *   contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n  const list: Tensor[] = [];\n  const seen = new Set<{}|void>();\n  walkTensorContainer(result, list, seen);\n  return list;\n}\n\nfunction walkTensorContainer(\n    container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n  if (container == null) {\n    return;\n  }\n  if (container instanceof Tensor) {\n    list.push(container);\n    return;\n  }\n  if (!isIterable(container)) {\n    return;\n  }\n  // Iteration over keys works also for arrays.\n  const iterable = container as TensorContainerArray;\n  for (const k in iterable) {\n    const val = iterable[k];\n    if (!seen.has(val)) {\n      seen.add(val);\n      walkTensorContainer(val, list, seen);\n    }\n  }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n  return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc<T> = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n *   value: Tensor,\n *   gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc<T extends Tensor> =\n    (...inputs: Array<Tensor|GradSaveFunc>) => {\n      value: T;\n      gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n    };\n\nexport type MemoryInfo = {\n  numTensors: number; numDataBuffers: number; numBytes: number;\n  unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n  name: string; bytesAdded: number; totalBytesSnapshot: number;\n  tensorsAdded: number;\n  totalTensorsSnapshot: number;\n  inputShapes: number[][];\n  outputShapes: number[][];\n  kernelTimeMs: number | {error: string} | Promise<number|{error: string}>;\n  extraInfo: string | Promise<string>;\n};\n\nexport type ProfileInfo = {\n  newBytes: number; newTensors: number; peakBytes: number;\n  kernels: KernelInfo[];\n  result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n  wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn<T extends TensorContainer> = () => T;\n\ninterface ScopeState {\n  track: Tensor[];\n  name: string;\n  id: number;\n}\n\nclass EngineState {\n  // Public since optimizers will use it.\n  registeredVariables: NamedVariableMap = {};\n\n  nextTapeNodeId = 0;\n  numBytes = 0;\n  numTensors = 0;\n  numStringTensors = 0;\n  numDataBuffers = 0;\n\n  activeTape: TapeNode[];\n  // Number of nested tf.grad() statements when computing higher-order\n  // gradients. E.g. `1` for first-order gradients and `2` for second-order\n  // gradients. Used to track if the tape should be removed after a backprop.\n  gradientDepth = 0;\n  // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n  // off the tape.\n  kernelDepth = 0;\n\n  // Keep Tensors that parallel the tapes.\n  activeScope: ScopeState;\n  scopeStack: ScopeState[] = [];\n  /**\n   * Keeps track of the number of data moves during a kernel execution. We\n   * maintain a stack since kernels can call other kernels, recursively.\n   */\n  numDataMovesStack: number[] = [];\n  nextScopeId = 0;\n\n  tensorInfo = new WeakMap<DataId, {\n    backend: KernelBackend,\n    bytes: number,\n    dtype: DataType,\n    shape: number[],\n    refCount: number\n  }>();\n\n  profiling = false;\n  activeProfile: ProfileInfo =\n      {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n  dispose() {\n    for (const variableName in this.registeredVariables) {\n      this.registeredVariables[variableName].dispose();\n    }\n  }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n  state: EngineState;\n  backendName: string;\n  registry: {[id: string]: KernelBackend} = {};\n  registryFactory: {\n    [id: string]: {\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority: number\n    }\n  } = {};\n\n  private profiler: Profiler;\n  private backendInstance: KernelBackend;\n  private pendingBackendInit: Promise<boolean>;\n  private pendingBackendInitId = 0;\n\n  constructor(public ENV: Environment) {\n    this.state = new EngineState();\n  }\n\n  async ready(): Promise<void> {\n    if (this.pendingBackendInit != null) {\n      return this.pendingBackendInit.then(() => {});\n    }\n    if (this.backendInstance != null) {\n      return;\n    }\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const success = await this.initializeBackend(backendName).success;\n      if (success) {\n        await this.setBackend(backendName);\n        return;\n      }\n    }\n\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  get backend(): KernelBackend {\n    if (this.pendingBackendInit != null) {\n      throw new Error(\n          `Backend '${this.backendName}' has not yet been initialized. Make ` +\n          `sure to await tf.ready() or await tf.setBackend() before calling ` +\n          `other methods`);\n    }\n    if (this.backendInstance == null) {\n      const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n      if (asyncInit) {\n        throw new Error(\n            `The highest priority backend '${name}' has not yet been ` +\n            `initialized. Make sure to await tf.ready() or ` +\n            `await tf.setBackend() before calling other methods`);\n      }\n      this.setBackend(name);\n    }\n    return this.backendInstance;\n  }\n\n  backendNames(): string[] {\n    return Object.keys(this.registryFactory);\n  }\n\n  findBackend(backendName: string): KernelBackend {\n    if (!(backendName in this.registry)) {\n      // If the backend hasn't been initialized but we have a registry entry for\n      // it, initialize it and return it.\n      if (backendName in this.registryFactory) {\n        const {asyncInit} = this.initializeBackend(backendName);\n        if (asyncInit) {\n          // Backend is not ready yet.\n          return null;\n        }\n      } else {\n        return null;\n      }\n    }\n    return this.registry[backendName];\n  }\n\n  findBackendFactory(backendName: string):\n      () => KernelBackend | Promise<KernelBackend> {\n    if (!(backendName in this.registryFactory)) {\n      return null;\n    }\n    return this.registryFactory[backendName].factory;\n  }\n\n  registerBackend(\n      backendName: string,\n      factory: () => KernelBackend | Promise<KernelBackend>,\n      priority = 1): boolean {\n    if (backendName in this.registryFactory) {\n      console.warn(\n          `${backendName} backend was already registered. ` +\n          `Reusing existing backend factory.`);\n      return false;\n    }\n    this.registryFactory[backendName] = {factory, priority};\n    return true;\n  }\n\n  async setBackend(backendName: string): Promise<boolean> {\n    if (this.registryFactory[backendName] == null) {\n      throw new Error(`Backend name '${backendName}' not found in registry`);\n    }\n    this.backendName = backendName;\n    if (this.registry[backendName] == null) {\n      this.backendInstance = null;\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      const result = asyncInit ? await success : success;\n      if (!result) {\n        return false;\n      }\n    }\n    this.backendInstance = this.registry[backendName];\n    this.setupRegisteredKernels();\n    // Reset the profiler.\n    this.profiler = new Profiler(this.backendInstance);\n\n    return true;\n  }\n\n  private setupRegisteredKernels(): void {\n    const kernels = getKernelsForBackend(this.backendName);\n    kernels.forEach(kernel => {\n      if (kernel.setupFunc != null) {\n        kernel.setupFunc(this.backendInstance);\n      }\n    });\n  }\n\n  private disposeRegisteredKernels(backendName: string): void {\n    const kernels = getKernelsForBackend(backendName);\n    kernels.forEach(kernel => {\n      if (kernel.disposeFunc != null) {\n        kernel.disposeFunc(this.registry[backendName]);\n      }\n    });\n  }\n\n  /**\n   * Initializes a backend by looking up the backend name in the factory\n   * registry and calling the factory method. Returns a boolean representing\n   * whether the initialization of the backend suceeded. Throws an error if\n   * there is no backend in the factory registry.\n   */\n  private initializeBackend(backendName: string):\n      {success: boolean|Promise<boolean>, asyncInit: boolean} {\n    const registryFactoryEntry = this.registryFactory[backendName];\n    if (registryFactoryEntry == null) {\n      throw new Error(\n          `Cannot initialize backend ${backendName}, no registration found.`);\n    }\n\n    try {\n      const backend = registryFactoryEntry.factory();\n      /* Test if the factory returns a promise.\n      Done in a more liberal way than\n      previous 'Promise.resolve(backend)===backend'\n      as we needed to account for custom Promise\n      implementations (e.g. Angular) */\n      if (backend && !(backend instanceof KernelBackend)\n          && typeof backend.then === 'function') {\n        const promiseId = ++this.pendingBackendInitId;\n        const success =\n            backend\n                .then(backendInstance => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.registry[backendName] = backendInstance;\n                  this.pendingBackendInit = null;\n                  return true;\n                })\n                .catch(err => {\n                  // Outdated promise. Another backend was set in the meantime.\n                  if (promiseId < this.pendingBackendInitId) {\n                    return false;\n                  }\n                  this.pendingBackendInit = null;\n                  console.warn(\n                      `Initialization of backend ${backendName} failed`);\n                  console.warn(err.stack || err.message);\n                  return false;\n                });\n        this.pendingBackendInit = success;\n        return {success, asyncInit: true};\n      } else {\n        this.registry[backendName] = backend as KernelBackend;\n        return {success: true, asyncInit: false};\n      }\n    } catch (err) {\n      console.warn(`Initialization of backend ${backendName} failed`);\n      console.warn(err.stack || err.message);\n      return {success: false, asyncInit: false};\n    }\n  }\n\n  removeBackend(backendName: string): void {\n    if (!(backendName in this.registryFactory)) {\n      throw new Error(`${backendName} backend not found in registry`);\n    }\n    if (this.backendName === backendName && this.pendingBackendInit != null) {\n      // There is a pending promise of the backend we want to remove. Make it\n      // obsolete.\n      this.pendingBackendInitId++;\n    }\n\n    if (backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n\n    delete this.registryFactory[backendName];\n\n    // Unset the backend if it is active.\n    if (this.backendName === backendName) {\n      this.pendingBackendInit = null;\n      this.backendName = null;\n      this.backendInstance = null;\n    }\n  }\n\n  private getSortedBackends(): string[] {\n    if (Object.keys(this.registryFactory).length === 0) {\n      throw new Error('No backend found in registry.');\n    }\n    return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n      // Highest priority comes first.\n      return this.registryFactory[b].priority -\n          this.registryFactory[a].priority;\n    });\n  }\n\n  private initializeBackendsAndReturnBest():\n      {name: string, asyncInit: boolean} {\n    const sortedBackends = this.getSortedBackends();\n\n    for (let i = 0; i < sortedBackends.length; i++) {\n      const backendName = sortedBackends[i];\n      const {success, asyncInit} = this.initializeBackend(backendName);\n      if (asyncInit || success) {\n        return {name: backendName, asyncInit};\n      }\n    }\n    throw new Error(\n        `Could not initialize any backends, all backend initializations ` +\n        `failed.`);\n  }\n\n  moveData(backend: KernelBackend, dataId: DataId) {\n    const info = this.state.tensorInfo.get(dataId);\n    const srcBackend = info.backend;\n    const values = this.readSync(dataId);\n    // Delete the tensor from the old backend and move it to the new\n    // backend.\n    srcBackend.disposeData(dataId);\n    info.backend = backend;\n    backend.move(dataId, values, info.shape, info.dtype);\n    if (this.shouldCheckForMemLeaks()) {\n      // Track the number of moves during a kernel execution to correctly\n      // detect memory leaks.\n      this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n    }\n  }\n\n  tidy<T extends TensorContainer>(nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>):\n      T {\n    let name: string = null;\n    if (fn == null) {\n      // Called with only 1 argument.\n      if (typeof nameOrFn !== 'function') {\n        throw new Error('Please provide a function to tidy()');\n      }\n      fn = nameOrFn;\n    } else {\n      // Called with 2 arguments.\n      if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n        throw new Error(\n            'When calling with two arguments, the first argument ' +\n            'to tidy() must be a string');\n      }\n      if (typeof fn !== 'function') {\n        throw new Error(\n            'When calling with two arguments, the 2nd argument ' +\n            'to tidy() must be a function');\n      }\n      name = nameOrFn as string;\n      // TODO(nsthorat,smilkov): Do operation logging and performance\n      // profiling.\n    }\n    let result: T;\n    return this.scopedRun(\n        () => this.startScope(name), () => this.endScope(result), () => {\n          result = fn();\n          if (result instanceof Promise) {\n            console.error('Cannot return a Promise inside of tidy.');\n          }\n          return result;\n        });\n  }\n\n  private scopedRun<T>(start: () => void, end: () => void, f: () => T): T {\n    start();\n    try {\n      const res = f();\n      end();\n      return res;\n    } catch (ex) {\n      end();\n      throw ex;\n    }\n  }\n\n  private static nextTensorId = 0;\n  private nextTensorId(): number {\n    return Engine.nextTensorId++;\n  }\n\n  private static nextVariableId = 0;\n  private nextVariableId(): number {\n    return Engine.nextVariableId++;\n  }\n\n  /**\n   * This method is called instead of the public-facing tensor.clone() when\n   * saving a tensor for backwards pass. It makes sure to add the clone\n   * operation to the tape regardless of being called inside a kernel\n   * execution.\n   *\n   * This method will go away once all kernels are modularized since we won't\n   * need to turn off the tape inside runKernel().\n   */\n  private clone(x: Tensor): Tensor {\n    const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n    const inputs = {x};\n    const grad = (dy: Tensor) => ({\n      x: () => {\n        const dtype = 'float32';\n        const gradInputs = {x: dy};\n        const attrs = {dtype};\n\n        return ENGINE.runKernelFunc(\n            backend => backend.cast(dy, dtype),\n            gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n            attrs as {} as NamedAttrMap);\n      }\n    });\n    const saved: Tensor[] = [];\n    this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n    return y;\n  }\n\n  /**\n   * Execute a kernel with the given name and return the output tensor.\n   *\n   * @param kernelName The name of the kernel to execute.\n   * @param inputs A map of input names to tensors.\n   * @param attrs A map of attribute names to their values. An attribute is a\n   *     primitive (non-tensor) input to the kernel.\n   * @param inputsToSave A list of tensors, inputs to save for the backprop\n   *     computation.\n   * @param outputsToSave A list of booleans, specifying which output to save\n   *     for the backprop computation. These are booleans since the output\n   * tensors are not visible to the user.\n   */\n  runKernel(\n      kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n      inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n    const forwardFunc: null = null;\n    const backwardsFunc: null = null;\n    // Call runKernel as a stop-gap until we modularize all kernels.\n    // Once we modularize all kernels, we will remove the existing\n    // `runKernelFunc`.\n    return this.runKernelFunc(\n        forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n        outputsToSave);\n  }\n\n  private shouldCheckForMemLeaks(): boolean {\n    return this.ENV.getBool('IS_TEST');\n  }\n\n  private checkKernelForMemLeak(\n      kernelName: string, numDataIdsBefore: number,\n      outInfos: TensorInfo[]): void {\n    const numDataIdsAfter = this.backend.numDataIds();\n\n    // Count the number of data ids associated with the result of the kernel.\n    let numOutputDataIds = 0;\n    outInfos.forEach(info => {\n      // Complex numbers allocate 3 data ids, one for 'real', one for\n      // 'imaginary', and one for the container that holds the former two.\n      numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n    });\n\n    // Account for the number of moves during kernel execution. A \"data move\"\n    // can happen in the middle of a kernel execution, placing a new (key,value)\n    // pair in the data storage. Since data moves have net zero effect (we\n    // always remove the data from the old backend), we have to cancel them out\n    // when detecting memory leaks.\n    const numMoves =\n        this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n    const dataIdsLeaked =\n        numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n    if (dataIdsLeaked > 0) {\n      throw new Error(\n          `Backend '${this.backendName}' has an internal memory leak ` +\n          `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n    }\n  }\n\n  /**\n   * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n   *     only for kernels that are not yet fully modularized.\n   */\n  runKernelFunc<T extends Tensor|Tensor[], I extends NamedTensorMap>(\n      forwardFunc: ForwardFunc<T>, inputs: I,\n      backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n      kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n      outputsToSave?: boolean[]): T {\n    let outputs: Tensor[];\n    let saved: Tensor[] = [];\n    const isTapeOn = this.isTapeOn();\n    if (kernelName == null) {\n      kernelName =\n          this.state.activeScope != null ? this.state.activeScope.name : '';\n    }\n\n    const startingBytecount = this.state.numBytes;\n    const startingNumTensors = this.state.numTensors;\n\n    if (this.shouldCheckForMemLeaks()) {\n      this.state.numDataMovesStack.push(0);\n    }\n\n    let kernelFunc: () => Tensor[];\n    const kernel = getKernel(kernelName, this.backendName);\n    let out: TensorInfo|TensorInfo[];\n    if (kernel != null) {\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n        const outInfos = Array.isArray(out) ? out : [out];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n        }\n        const outTensors = outInfos.map(\n            ({dataId, shape, dtype}) =>\n                this.makeTensorFromDataId(dataId, shape, dtype));\n\n        // Save the inputs and outputs.\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (isTapeOn) {\n          let tensorsToSave =\n              this.getTensorsForGradient(kernelName, inputs, outTensors);\n          if (tensorsToSave == null) {\n            // Fallback for ops that call runKernelFunc and pass in\n            // inputsToSave and outputsToSave. Currently this is the set of ops\n            // with kernel support in the WASM backend. Once those ops and\n            // respective gradients are modularised we can remove this path.\n            if (outputsToSave == null) {\n              outputsToSave = [];\n            }\n            const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n            tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n          }\n          saved = this.saveTensorsForBackwardMode(tensorsToSave);\n        }\n        return outTensors;\n      };\n    } else {\n      const saveFunc: GradSaveFunc = (tensors) => {\n        // Do not save unless we are recording to the tape. Otherwise it would\n        // cause a mem leak since we would never run backprop, which disposes\n        // the kept tensors.\n        if (!isTapeOn) {\n          return;\n        }\n        saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n      };\n\n      kernelFunc = () => {\n        const numDataIdsBefore = this.backend.numDataIds();\n        out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n        const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n        if (this.shouldCheckForMemLeaks()) {\n          this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n        }\n        return outs;\n      };\n    }\n\n    // Stop recording to a tape when running a kernel.\n    let kernelProfile: KernelProfile;\n    this.scopedRun(\n        () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n          if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n            outputs = kernelFunc();\n          } else {\n            kernelProfile = this.profiler.profileKernel(\n                kernelName, inputs, () => kernelFunc());\n            if (this.ENV.getBool('DEBUG')) {\n              this.profiler.logKernelProfile(kernelProfile);\n            }\n            outputs = kernelProfile.outputs;\n          }\n        });\n\n    if (isTapeOn) {\n      this.addTapeNode(\n          kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n    }\n\n    if (this.state.profiling) {\n      this.state.activeProfile.kernels.push({\n        name: kernelName,\n        bytesAdded: this.state.numBytes - startingBytecount,\n        totalBytesSnapshot: this.state.numBytes,\n        tensorsAdded: this.state.numTensors - startingNumTensors,\n        totalTensorsSnapshot: this.state.numTensors,\n        inputShapes: Object.keys(inputs).map(\n            key => inputs[key] != null ? inputs[key].shape : null),\n        outputShapes: outputs.map(item => item.shape),\n        kernelTimeMs: kernelProfile.timeMs,\n        extraInfo: kernelProfile.extraInfo\n      });\n    }\n    return (Array.isArray(out) ? outputs : outputs[0]) as T;\n  }\n\n  /**\n   * Saves tensors used in forward mode for use in backward mode.\n   *\n   * @param tensors the list of tensors to save.\n   */\n  private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n    const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n    return saved;\n  }\n\n  /**\n   * Returns a list of tensors to save for a given gradient calculation.\n   *\n   * Returns undefined if their is no registered gradient for this kernel in the\n   * gradient registry.\n   *\n   * @param kernelName name of kernel to look up gradient for.\n   * @param inputs a map of input tensors.\n   * @param outputs an array of output tensors from forward mode of kernel.\n   */\n  private getTensorsForGradient(\n      kernelName: string, inputs: NamedTensorMap,\n      outputs: Tensor[]): Tensor[]|null {\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      const inputsToSave: string[] = gradConfig.inputsToSave || [];\n      const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n      // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n      // specified in inputsToSave will be saved.\n      let inputTensorsToSave: Tensor[];\n      if (gradConfig.saveAllInputs) {\n        util.assert(\n            Array.isArray(inputs),\n            () => 'saveAllInputs is true, expected inputs to be an array.');\n\n        inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n      } else {\n        inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n      }\n\n      const outputTensorsToSave: Tensor[] =\n          outputs.filter((_, i) => outputsToSave[i]);\n\n      return inputTensorsToSave.concat(outputTensorsToSave);\n    }\n    // TODO(yassogba) throw exception here once all runkernelFunc calls with\n    // inputsToSave/outputsToSave are removed\n    return null;\n  }\n\n  /**\n   * Internal method used by public APIs for tensor creation. Makes a new\n   * tensor with the provided shape, dtype and values. It always\n   * creates a new data id and writes the values to the underlying backend.\n   */\n  makeTensor(\n      values: DataValues, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    if (values == null) {\n      throw new Error('Values passed to engine.makeTensor() are null');\n    }\n    dtype = dtype || 'float32';\n    backend = backend || this.backend;\n    let backendVals = values as BackendValues;\n    if (dtype === 'string' && util.isString(values[0])) {\n      backendVals = (values as string[]).map(d => util.encodeString(d));\n    }\n    const dataId = backend.write(backendVals, shape, dtype);\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n\n    // Count bytes for string tensors.\n    if (dtype === 'string') {\n      const info = this.state.tensorInfo.get(dataId);\n      const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n      this.state.numBytes += newBytes - info.bytes;\n      info.bytes = newBytes;\n    }\n    return t;\n  }\n\n  /**\n   * Internal method used by backends. Makes a new tensor\n   * that is a wrapper around an existing data id. It doesn't create\n   * a new data id, only increments the ref count used in memory tracking.\n   */\n  makeTensorFromDataId(\n      dataId: DataId, shape: number[], dtype: DataType,\n      backend?: KernelBackend): Tensor {\n    dtype = dtype || 'float32';\n    const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n    this.incRef(t, backend);\n    return t;\n  }\n\n  makeVariable(\n      initialValue: Tensor, trainable = true, name?: string,\n      dtype?: DataType): Variable {\n    name = name || this.nextVariableId().toString();\n    if (dtype != null && dtype !== initialValue.dtype) {\n      initialValue = initialValue.cast(dtype);\n    }\n    const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n    if (this.state.registeredVariables[v.name] != null) {\n      throw new Error(`Variable with name ${v.name} was already registered`);\n    }\n    this.state.registeredVariables[v.name] = v;\n    this.incRef(v, this.backend);\n    return v;\n  }\n\n  incRef(a: Tensor, backend: KernelBackend): void {\n    const refCount = this.state.tensorInfo.has(a.dataId) ?\n        this.state.tensorInfo.get(a.dataId).refCount :\n        0;\n    this.state.numTensors++;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors++;\n    }\n    if (refCount === 0) {\n      this.state.numDataBuffers++;\n\n      // Bytes for complex numbers are counted by their components. Bytes for\n      // string tensors are counted when writing values.\n      let bytes = 0;\n      if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n        bytes = a.size * util.bytesPerElement(a.dtype);\n      }\n      this.state.tensorInfo.set(a.dataId, {\n        backend: backend || this.backend,\n        dtype: a.dtype,\n        shape: a.shape,\n        bytes,\n        refCount: 0\n      });\n      this.state.numBytes += bytes;\n    }\n\n    this.state.tensorInfo.get(a.dataId).refCount++;\n\n    if (!(a instanceof Variable)) {\n      this.track(a);\n    }\n  }\n\n  disposeTensor(a: Tensor): void {\n    if (!this.state.tensorInfo.has(a.dataId)) {\n      return;\n    }\n\n    this.state.numTensors--;\n    if (a.dtype === 'string') {\n      this.state.numStringTensors--;\n    }\n    const info = this.state.tensorInfo.get(a.dataId);\n    const refCount = info.refCount;\n\n    if (refCount <= 1) {\n      // Don't count bytes for complex numbers as they are counted by their\n      // components.\n      if (a.dtype !== 'complex64') {\n        this.state.numBytes -= info.bytes;\n      }\n      this.state.numDataBuffers--;\n\n      info.backend.disposeData(a.dataId);\n      this.state.tensorInfo.delete(a.dataId);\n    } else {\n      this.state.tensorInfo.get(a.dataId).refCount--;\n    }\n    // TODO(nsthorat): Construct an error and save the stack trace for\n    // debugging when in debug mode. Creating a stack trace is too expensive\n    // to do unconditionally.\n  }\n\n  disposeVariables(): void {\n    for (const varName in this.state.registeredVariables) {\n      const v = this.state.registeredVariables[varName];\n      this.disposeVariable(v);\n    }\n  }\n\n  disposeVariable(v: Variable): void {\n    this.disposeTensor(v);\n    if (this.state.registeredVariables[v.name] != null) {\n      delete this.state.registeredVariables[v.name];\n    }\n  }\n\n  memory(): MemoryInfo {\n    const info = this.backend.memory() as MemoryInfo;\n    info.numTensors = this.state.numTensors;\n    info.numDataBuffers = this.state.numDataBuffers;\n    info.numBytes = this.state.numBytes;\n    if (this.state.numStringTensors > 0) {\n      info.unreliable = true;\n      if (info.reasons == null) {\n        info.reasons = [];\n      }\n      info.reasons.push(\n          'Memory usage by string tensors is approximate ' +\n          '(2 bytes per character)');\n    }\n    return info;\n  }\n\n  async profile(query: () => (TensorContainer | Promise<TensorContainer>)):\n      Promise<ProfileInfo> {\n    this.state.profiling = true;\n\n    const startBytes = this.state.numBytes;\n    const startNumTensors = this.state.numTensors;\n\n    this.state.activeProfile.kernels = [];\n    this.state.activeProfile.result = await query();\n\n    this.state.profiling = false;\n\n    this.state.activeProfile.peakBytes = Math.max(\n        ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n    this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n    this.state.activeProfile.newTensors =\n        this.state.numTensors - startNumTensors;\n    for (const kernel of this.state.activeProfile.kernels) {\n      kernel.kernelTimeMs = await kernel.kernelTimeMs;\n      kernel.extraInfo = await kernel.extraInfo;\n    }\n    return this.state.activeProfile;\n  }\n\n  isTapeOn(): boolean {\n    return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n  }\n\n  private addTapeNode(\n      kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n      gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n    const tapeNode: TapeNode =\n        {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n    const gradConfig = getGradient(kernelName);\n    if (gradConfig != null) {\n      gradientsFunc = gradConfig.gradFunc;\n    }\n    if (gradientsFunc != null) {\n      tapeNode.gradient = (dys: Tensor[]) => {\n        // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n        // the backprop graph to the user as null instead of zeros\n        dys = dys.map((dy, i) => {\n          if (dy == null) {\n            const output = outputs[i];\n            const vals = util.makeZerosTypedArray(output.size, output.dtype);\n            return this.makeTensor(vals, output.shape, output.dtype);\n          }\n          return dy;\n        });\n        // Grad functions of ops with single outputs expect a dy, while ops\n        // with multiple outputs expect dys (array of dy).\n        return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n      };\n    }\n    this.state.activeTape.push(tapeNode);\n  }\n\n  keep<T extends Tensor>(result: T): T {\n    result.kept = true;\n    return result;\n  }\n\n  private startTape() {\n    if (this.state.gradientDepth === 0) {\n      this.state.activeTape = [];\n    }\n    this.state.gradientDepth++;\n  }\n\n  private endTape() {\n    this.state.gradientDepth--;\n  }\n\n  /**\n   * Start a scope. Use this with endScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  startScope(name?: string) {\n    const scopeInfo: ScopeState = {\n      track: [],\n      name: 'unnamed scope',\n      id: this.state.nextScopeId++\n    };\n    if (name) {\n      scopeInfo.name = name;\n    }\n    this.state.scopeStack.push(scopeInfo);\n    this.state.activeScope = scopeInfo;\n  }\n\n  /**\n   * End a scope. Use this with startScope() to achieve the same functionality\n   * as scope() without the need for a function closure.\n   */\n  endScope(result?: TensorContainer) {\n    const tensorsToTrackInParent = getTensorsInContainer(result);\n    const tensorsToTrackInParentSet =\n        new Set(tensorsToTrackInParent.map(t => t.id));\n\n    // Dispose the arrays tracked in this scope.\n    for (let i = 0; i < this.state.activeScope.track.length; i++) {\n      const tensor = this.state.activeScope.track[i];\n      if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n        tensor.dispose();\n      }\n    }\n\n    const oldScope = this.state.scopeStack.pop();\n    this.state.activeScope = this.state.scopeStack.length === 0 ?\n        null :\n        this.state.scopeStack[this.state.scopeStack.length - 1];\n\n    // Track the current result in the parent scope.\n    tensorsToTrackInParent.forEach(tensor => {\n      // Only track the tensor if was allocated in the inner scope and is not\n      // globally kept.\n      if (!tensor.kept && tensor.scopeId === oldScope.id) {\n        this.track(tensor);\n      }\n    });\n  }\n\n  /**\n   * Returns gradients of `f` with respect to each of the `xs`. The gradients\n   * returned are of the same length as `xs`, but some might be null if `f`\n   * was not a function of that `x`. It also takes optional dy to multiply the\n   * gradient, which defaults to `1`.\n   */\n  gradients<T extends Tensor>(\n      f: () => T, xs: Tensor[], dy?: T,\n      allowNoGradients = false): {value: T, grads: Tensor[]} {\n    util.assert(\n        xs.length > 0, () => 'gradients() received an empty list of xs.');\n    if (dy != null && dy.dtype !== 'float32') {\n      throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n    }\n\n    const y = this.scopedRun(\n        () => this.startTape(), () => this.endTape(),\n        () => this.tidy('forward', f));\n\n    util.assert(\n        y instanceof Tensor,\n        () => 'The result y returned by f() must be a tensor.');\n    // Filter out the nodes that don't connect x => y.\n    const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n    if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n      throw new Error(\n          'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n          'that the f you passed encloses all operations that lead from x ' +\n          'to y.');\n    }\n\n    return this.tidy('backward', () => {\n      const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n      accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n      // Backprop gradients through the filtered nodes.\n      backpropagateGradients(\n          accumulatedGradientMap, filteredTape,\n          // Pass the tidy function to avoid circular dep with `tape.ts`.\n          f => this.tidy(f as ScopeFn<Tensor>),\n          // Pass an add function to avoide a circular dep with `tape.ts`.\n          add);\n      const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n      if (this.state.gradientDepth === 0) {\n        // This means that we are not computing higher-order gradients\n        // and can clean up the tape.\n        this.state.activeTape.forEach(node => {\n          for (const tensor of node.saved) {\n            tensor.dispose();\n          }\n        });\n        this.state.activeTape = null;\n      }\n      return {value: y, grads};\n    });\n  }\n\n  customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n      (...args: Array<Tensor|GradSaveFunc>) => T {\n    util.assert(\n        util.isFunction(f),\n        () => 'The f passed in customGrad(f) must be a function.');\n    return (...inputs: Tensor[]): T => {\n      util.assert(\n          inputs.every(t => t instanceof Tensor),\n          () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n              'tensors');\n\n      let res: {\n        value: T,\n        gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n      };\n      const inputMap: NamedTensorMap = {};\n      inputs.forEach((input, i) => {\n        inputMap[i] = input;\n      });\n      return this.runKernelFunc(\n          (_, save) => {\n            res = f(...[...inputs, save]);\n            util.assert(\n                res.value instanceof Tensor,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.value` is a tensor');\n            util.assert(\n                util.isFunction(res.gradFunc),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function.');\n            return res.value;\n          },\n          inputMap,\n          (dy: T, saved: Tensor[]) => {\n            const gradRes = res.gradFunc(dy, saved);\n            const grads: Tensor[] =\n                Array.isArray(gradRes) ? gradRes : [gradRes];\n            util.assert(\n                grads.length === inputs.length,\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'the same number of tensors as inputs passed to f(...).');\n            util.assert(\n                grads.every(t => t instanceof Tensor),\n                () => 'The function f passed in customGrad(f) must return an ' +\n                    'object where `obj.gradFunc` is a function that returns ' +\n                    'a list of only tensors.');\n            const gradMap: {[key: string]: () => Tensor} = {};\n            grads.forEach((grad, i) => {\n              gradMap[i] = () => grad;\n            });\n            return gradMap;\n          });\n    };\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.readSync(dataId);\n  }\n  read(dataId: DataId): Promise<BackendValues> {\n    // Route the read to the correct backend.\n    const info = this.state.tensorInfo.get(dataId);\n    return info.backend.read(dataId);\n  }\n\n  async time(query: () => void): Promise<TimingInfo> {\n    const start = now();\n    const timingInfo = await this.backend.time(query) as TimingInfo;\n    timingInfo.wallMs = now() - start;\n    return timingInfo;\n  }\n\n  /**\n   * Tracks a Tensor in the current scope to be automatically cleaned up\n   * when the current scope ends, and returns the value.\n   *\n   * @param result The Tensor to track in the current scope.\n   */\n  private track<T extends Tensor>(result: T): T {\n    if (this.state.activeScope != null) {\n      result.scopeId = this.state.activeScope.id;\n      this.state.activeScope.track.push(result);\n    }\n\n    return result;\n  }\n\n  get registeredVariables(): NamedVariableMap {\n    return this.state.registeredVariables;\n  }\n\n  /**\n   * Resets the engine state. Removes all backends but does not remove\n   * registered backend factories.\n   */\n  reset(): void {\n    // Make any pending promise obsolete.\n    this.pendingBackendInitId++;\n\n    this.state.dispose();\n    this.ENV.reset();\n    this.state = new EngineState();\n\n    for (const backendName in this.registry) {\n      this.disposeRegisteredKernels(backendName);\n      this.registry[backendName].dispose();\n      delete this.registry[backendName];\n    }\n    this.backendName = null;\n    this.backendInstance = null;\n    this.pendingBackendInit = null;\n  }\n}\n\nfunction ones(shape: number[]): Tensor {\n  const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n  return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n  const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n  if (ns._tfengine == null) {\n    const environment = new Environment(ns);\n    ns._tfengine = new Engine(environment);\n  }\n  setEnvironmentGlobal(ns._tfengine.ENV);\n\n  // Tell the current tensor interface that the global engine is responsible\n  // for tracking.\n  setTensorTracker(() => ns._tfengine);\n  return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n  // We duplicate Add here to avoid a circular dependency with add.ts.\n  const inputs = {a, b};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.add(a, b);\n    save([a, b]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n  return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n  if (_isNavigatorDefined()) {\n    // tslint:disable-next-line:no-any\n    const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n    // tslint:disable-next-line:max-line-length\n    return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n               .test(a) ||\n        // tslint:disable-next-line:max-line-length\n        /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n            .test(a.substr(0, 4));\n  }\n  return false;\n}\n\nexport function isBrowser(): boolean {\n  return (typeof window !== 'undefined' && window.document != null) ||\n      //@ts-ignore\n      (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n  if (debugValue) {\n    console.warn(\n        'Debugging mode is ON. The output of every math call will ' +\n        'be downloaded to CPU and checked for NaNs. ' +\n        'This significantly impacts performance.');\n  }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n    'IS_NODE',\n    () => (typeof process !== 'undefined') &&\n        (typeof process.versions !== 'undefined') &&\n        (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n    'IS_CHROME',\n    () => typeof navigator !== 'undefined' && navigator != null &&\n        navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n        /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n    'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n  let firstElem: typeof val = val;\n\n  if (isTypedArray(val)) {\n    return dtype === 'string' ? [] : [val.length];\n  }\n  if (!Array.isArray(val)) {\n    return [];  // Scalar.\n  }\n  const shape: number[] = [];\n\n  while (Array.isArray(firstElem) ||\n         isTypedArray(firstElem) && dtype !== 'string') {\n    shape.push(firstElem.length);\n    firstElem = firstElem[0];\n  }\n  if (Array.isArray(val) &&\n      env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n    deepAssertShapeConsistency(val, shape, []);\n  }\n\n  return shape;\n}\n\nfunction deepAssertShapeConsistency(\n    val: TensorLike, shape: number[], indices: number[]) {\n  indices = indices || [];\n  if (!(Array.isArray(val)) && !isTypedArray(val)) {\n    assert(\n        shape.length === 0,\n        () => `Element arr[${indices.join('][')}] is a primitive, ` +\n            `but should be an array/TypedArray of ${shape[0]} elements`);\n    return;\n  }\n  assert(\n      shape.length > 0,\n      () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n          `but is an array of ${val.length} elements`);\n  assert(\n      val.length === shape[0],\n      () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n          `elements, but has ${val.length} elements`);\n  const subShape = shape.slice(1);\n  for (let i = 0; i < val.length; ++i) {\n    deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n  }\n}\n\nfunction assertDtype(\n    expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n    functionName: string) {\n  if (expectedDtype == null) {\n    return;\n  }\n  if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n      expectedDtype === 'numeric' && actualDType === 'string') {\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must ` +\n        `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n  }\n}\n\nexport function convertToTensor<T extends Tensor>(\n    x: T|TensorLike, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T {\n  if (x instanceof Tensor) {\n    assertDtype(parseAsDtype, x.dtype, argName, functionName);\n    return x;\n  }\n  let inferredDtype = inferDtype(x);\n  // If the user expects a bool/int/float, use that info to update the\n  // inferredDtype when it is not a string.\n  if (inferredDtype !== 'string' &&\n      ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n    inferredDtype = parseAsDtype as DataType;\n  }\n  assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n  if ((x == null) ||\n      (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n       typeof x !== 'boolean' && typeof x !== 'string')) {\n    const type = x == null ? 'null' : (x as {}).constructor.name;\n    throw new Error(\n        `Argument '${argName}' passed to '${functionName}' must be a ` +\n        `Tensor or TensorLike, but got '${type}'`);\n  }\n  const inferredShape = inferShape(x, inferredDtype);\n  if (!isTypedArray(x) && !Array.isArray(x)) {\n    x = [x] as number[];\n  }\n  const skipTypedArray = true;\n  const values = inferredDtype !== 'string' ?\n      toTypedArray(x, inferredDtype as DataType) :\n      flatten(x as string[], [], skipTypedArray) as string[];\n  return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray<T extends Tensor>(\n    arg: Array<T|TensorLike>, argName: string, functionName: string,\n    parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n  if (!Array.isArray(arg)) {\n    throw new Error(\n        `Argument ${argName} passed to ${functionName} must be a ` +\n        '`Tensor[]` or `TensorLike[]`');\n  }\n  const tensors = arg as T[];\n  return tensors.map(\n      (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n      parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op<T extends Function>(f: {[name: string]: T}): T {\n  const keys = Object.keys(f);\n  if (keys.length !== 1) {\n    throw new Error(\n        `Please provide an object with a single key ` +\n        `(operation name) mapping to a function. Got an object with ` +\n        `${keys.length} keys.`);\n  }\n\n  let opName = keys[0];\n  const fn = f[opName];\n\n  // Strip the underscore from the end of the function name.\n  if (opName.endsWith('_')) {\n    opName = opName.substring(0, opName.length - 1);\n  }\n\n  // add an __op suffix to distinguish ops from kernels in tf.profile\n  opName = opName + OP_SCOPE_SUFFIX;\n\n  // tslint:disable-next-line:no-any\n  const f2 = (...args: any[]) => {\n    ENGINE.startScope(opName);\n    try {\n      const result = fn(...args);\n      if (result instanceof Promise) {\n        console.error('Cannot return a Promise inside of tidy.');\n      }\n      ENGINE.endScope(result);\n      return result;\n    } catch (ex) {\n      ENGINE.endScope(null);\n      throw ex;\n    }\n  };\n  Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n  // tslint:disable-next-line:no-any\n  return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_<T extends Tensor>(real: T|TensorLike, imag: T|TensorLike): T {\n  const $real = convertToTensor(real, 'real', 'complex');\n  const $imag = convertToTensor(imag, 'imag', 'complex');\n  util.assertShapesMatch(\n      $real.shape, $imag.shape,\n      `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n          `must match in call to tf.complex().`);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.complex($real, $imag);\n  };\n  const inputs: ComplexInputs = {real: $real, imag: $imag};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n    values: TensorLike, shape: number[], inferredShape: number[],\n    dtype?: DataType): Tensor {\n  if (dtype == null) {\n    dtype = inferDtype(values);\n  }\n  if (dtype === 'complex64') {\n    throw new Error(\n        `Cannot construct a complex64 tensor directly. ` +\n        `Please use tf.complex(real, imag).`);\n  }\n  if (!isTypedArray(values) && !Array.isArray(values) &&\n      typeof values !== 'number' && typeof values !== 'boolean' &&\n      typeof values !== 'string') {\n    throw new Error(\n        'values passed to tensor(values) must be a number/boolean/string or ' +\n        'an array of numbers/booleans/strings, or a TypedArray');\n  }\n  if (shape != null) {\n    assertNonNegativeIntegerDimensions(shape);\n\n    const providedSize = sizeFromShape(shape);\n    const inferredSize = sizeFromShape(inferredShape);\n    assert(\n        providedSize === inferredSize,\n        () =>\n            `Based on the provided shape, [${shape}], the tensor should have ` +\n            `${providedSize} values but has ${inferredSize}`);\n\n    for (let i = 0; i < inferredShape.length; ++i) {\n      const inferred = inferredShape[i];\n      const flatDimsDontMatch = i === inferredShape.length - 1 ?\n          inferred !== sizeFromShape(shape.slice(i)) :\n          true;\n      assert(\n          inferredShape[i] === shape[i] || !flatDimsDontMatch,\n          () => `Error creating a new Tensor. Inferred shape ` +\n              `(${inferredShape}) does not match the provided ` +\n              `shape (${shape}). `);\n    }\n  }\n\n  if (!isTypedArray(values) && !Array.isArray(values)) {\n    values = [values] as number[];\n  }\n\n  shape = shape || inferredShape;\n  values = dtype !== 'string' ?\n      toTypedArray(values, dtype) :\n      flatten(values as string[], [], true) as string[];\n  return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`. If the values are strings,\n *     they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor<R extends Rank>(\n    values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor<R> {\n  const inferredShape = inferShape(values, dtype);\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n  'float32': 4,\n  'float16': 2,\n  'int32': 4,\n  'uint16': 2,\n  'uint8': 1,\n  'bool': 1,\n  'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n  /**\n   * An ordered list of paths.\n   *\n   * Paths are intentionally abstract in order to be general. For example, they\n   * can be relative URL paths or relative paths on the file system.\n   */\n  paths: string[];\n\n  /**\n   * Specifications of the weights stored in the paths.\n   */\n  weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n  /**\n   * Name of the weight, e.g., 'Dense_1/bias'\n   */\n  name: string;\n\n  /**\n   * Shape of the weight.\n   */\n  shape: number[];\n\n  /**\n   * Data type of the weight.\n   */\n  dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n  /**\n   * Type of the weight.\n   *\n   * Optional.\n   *\n   * The value 'optimizer' indicates the weight belongs to an optimizer\n   * (i.e., used only during model training and not during inference).\n   */\n  group?: WeightGroup;\n\n  /**\n   * Information for dequantization of the weight.\n   */\n  quantization?: {\n    scale?: number,  // The scaling constant to multiply by.\n    min?: number,    // The (possibly nudged) minimum weight to add.\n       dtype: 'uint16'|'uint8'|'float16'  // The dtype of the quantized weights.\n  };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n  /**\n   * Whether to save only the trainable weights of the model, ignoring the\n   * non-trainable ones.\n   */\n  trainableOnly?: boolean;\n\n  /**\n   * Whether the optimizer will be saved (if exists).\n   *\n   * Default: `false`.\n   */\n  includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n  /**\n   * Information about the model artifacts saved.\n   */\n  modelArtifactsInfo: ModelArtifactsInfo;\n\n  /**\n   * HTTP responses from the server that handled the model-saving request (if\n   * any). This is applicable only to server-based saving routes.\n   */\n  responses?: Response[];\n\n  /**\n   * Error messages and related data (if any).\n   */\n  errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n  /**\n   * Timestamp for when the model is saved.\n   */\n  dateSaved: Date;\n\n  /**\n   * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n   * come in a JSON format and none of our IOHandlers support a non json\n   * format. We could conder replacing this with 'Binary' if we want to\n   * allow future handlers to save to non json formats (though they will\n   * probably want more information than 'Binary').\n   * Type of the model topology\n   *\n   * Type of the model topology\n   *\n   * Possible values:\n   *   - JSON: JSON config (human-readable, e.g., Keras JSON).\n   *   - GraphDef: TensorFlow\n   *     [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n   *     protocol buffer (binary).\n   */\n  modelTopologyType: 'JSON'|'GraphDef';\n\n  /**\n   * Size of model topology (Keras JSON or GraphDef), in bytes.\n   */\n  modelTopologyBytes?: number;\n\n  /**\n   * Size of weight specification or manifest, in bytes.\n   */\n  weightSpecsBytes?: number;\n\n  /**\n   * Size of weight value data, in bytes.\n   */\n  weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  // See\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n  /** Optimizer used for the model training. */\n  optimizer_config: {};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Loss function(s) for the model's output(s). */\n  loss: string|string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  /** Metric function(s) for the model's output(s). */\n  metrics?: string[]|{[key: string]: string};\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  weighted_metrics?: string[];\n\n  // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n  sample_weight_mode?: string;\n\n  loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers.  For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology?: {}|ArrayBuffer;\n\n  /**\n   * Serialized configuration for the model's training.\n   */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weight specifications.\n   *\n   * This corresponds to the weightsData below.\n   */\n  weightSpecs?: WeightsManifestEntry[];\n\n  /**\n   * Binary buffer for all weight values concatenated in the order specified\n   * by `weightSpecs`.\n   */\n  weightData?: ArrayBuffer;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n  /**\n   * Model topology.\n   *\n   * For Keras-style `tf.Model`s, this is a JSON object.\n   * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n   * encoding of the `GraphDef` protocol buffer.\n   */\n  modelTopology: {};\n\n  /** Model training configuration. */\n  trainingConfig?: TrainingConfig;\n\n  /**\n   * Weights manifest.\n   *\n   * The weights manifest consists of an ordered list of weight-manifest\n   * groups. Each weight-manifest group consists of a number of weight values\n   * stored in a number of paths. See the documentation of\n   * `WeightsManifestConfig` for more details.\n   */\n  weightsManifest: WeightsManifestConfig;\n\n  /**\n   * Hard-coded format name for models saved from TensorFlow.js or converted\n   * by TensorFlow.js Converter.\n   */\n  format?: string;\n\n  /**\n   * What library is responsible for originally generating this artifact.\n   *\n   * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n   */\n  generatedBy?: string;\n\n  /**\n   * What library or tool is responsible for converting the original model\n   * to this format, applicable only if the model is output by a converter.\n   *\n   * Used for debugging purposes.  E.g., 'TensorFlow.js Converter v1.0.0'.\n   *\n   * A value of `null` means the model artifacts are generated without any\n   * conversion process (e.g., saved directly from a TensorFlow.js\n   * `tf.LayersModel` instance.)\n   */\n  convertedBy?: string|null;\n\n  /**\n   * User-defined metadata about the model.\n   */\n  userDefinedMetadata?: {};\n\n  /**\n   * Initializer for the model.\n   */\n  modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise<ModelArtifacts>;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n    Promise<SaveResult>;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n  save?: SaveHandler;\n  load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n  /**\n   * List all models in the model store.\n   *\n   * @returns A dictionary mapping paths of existing models to their\n   *   model artifacts info. Model artifacts info include type of the model's\n   *   topology, byte sizes of the topology, weights, etc.\n   */\n  listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n  /**\n   * Remove a model specified by `path`.\n   *\n   * @param path\n   * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n   *   is successful).\n   * @throws Error if deletion fails, e.g., if no model exists at `path`.\n   */\n  removeModel(path: string): Promise<ModelArtifactsInfo>;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n  /**\n   * RequestInit (options) for HTTP requests.\n   *\n   * For detailed information on the supported fields, see\n   * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n   *     https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n   */\n  requestInit?: RequestInit;\n\n  /**\n   * Progress callback.\n   */\n  onProgress?: OnProgressCallback;\n\n  /**\n   * A function used to override the `window.fetch` function.\n   */\n  fetchFunc?: Function;\n\n  /**\n   * Strict loading model: whether extraneous weights or missing\n   * weights should trigger an `Error`.\n   *\n   * If `true`, require that the provided weights exactly match those\n   * required by the layers. `false` means that both extra weights\n   * and missing weights will be silently ignored.\n   *\n   * Default: `true`.\n   */\n  strict?: boolean;\n\n  /**\n   * Path prefix for weight files, by default this is calculated from the\n   * path of the model JSON file.\n   *\n   * For instance, if the path to the model JSON file is\n   * `http://localhost/foo/model.json`, then the default path prefix will be\n   * `http://localhost/foo/`. If a weight file has the path value\n   * `group1-shard1of2` in the weight manifest, then the weight file will be\n   * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n   * if you provide a `weightPathPrefix` value of\n   * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n   * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n   */\n  weightPathPrefix?: string;\n\n  /**\n   * Whether the module or model is to be loaded from TF Hub.\n   *\n   * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n   * standard model file name and the query parameters.\n   *\n   * Default: `false`.\n   */\n  fromTFHub?: boolean;\n\n  /**\n   * An async function to convert weight file name to URL. The weight file\n   * names are stored in model.json's weightsManifest.paths field. By default we\n   * consider weight files are colocated with the model.json file. For example:\n   *     model.json URL: https://www.google.com/models/1/model.json\n   *     group1-shard1of1.bin url:\n   *        https://www.google.com/models/1/group1-shard1of1.bin\n   *\n   * With this func you can convert the weight file name to any URL.\n   */\n  weightUrlConverter?: (weightFileName: string) => Promise<string>;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n  /**\n   * Is this request for a binary file (as opposed to a json file)\n   */\n  isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n *     concatenated.\n *   - An `Array` of `WeightManifestEntry`s, carrying information including\n *     tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n    tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n    Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n  // TODO(adarob, cais): Support quantization.\n  const specs: WeightsManifestEntry[] = [];\n  const dataPromises: Array<Promise<TypedArray>> = [];\n\n  const names: string[] = Array.isArray(tensors) ?\n      tensors.map(tensor => tensor.name) :\n      Object.keys(tensors);\n\n  for (let i = 0; i < names.length; ++i) {\n    const name = names[i];\n    const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n    if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n        t.dtype !== 'string' && t.dtype !== 'complex64') {\n      throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n    }\n    const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n    if (t.dtype === 'string') {\n      const utf8bytes = new Promise<TypedArray>(async resolve => {\n        const vals = await t.bytes() as Uint8Array[];\n        const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n            NUM_BYTES_STRING_LENGTH * vals.length;\n        const bytes = new Uint8Array(totalNumBytes);\n        let offset = 0;\n        for (let i = 0; i < vals.length; i++) {\n          const val = vals[i];\n          const bytesOfLength =\n              new Uint8Array(new Uint32Array([val.length]).buffer);\n          bytes.set(bytesOfLength, offset);\n          offset += NUM_BYTES_STRING_LENGTH;\n          bytes.set(val, offset);\n          offset += val.length;\n        }\n        resolve(bytes);\n      });\n      dataPromises.push(utf8bytes);\n    } else {\n      dataPromises.push(t.data());\n    }\n    if (group != null) {\n      spec.group = group;\n    }\n    specs.push(spec);\n  }\n\n  const tensorValues = await Promise.all(dataPromises);\n  return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n *   concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n *   whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n *   to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n    buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n  // TODO(adarob, cais): Support quantization.\n  const out: NamedTensorMap = {};\n  let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n  let offset = 0;\n  for (const spec of specs) {\n    const name = spec.name;\n    const dtype = spec.dtype;\n    const shape = spec.shape;\n    const size = sizeFromShape(shape);\n    let values: TypedArray|string[]|Uint8Array[];\n\n    if ('quantization' in spec) {\n      const quantization = spec.quantization;\n      if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n        if (!('min' in quantization && 'scale' in quantization)) {\n          throw new Error(\n              `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n              `doesn't have corresponding metadata min and scale.`);\n        }\n      } else if (quantization.dtype === 'float16') {\n        if (dtype !== 'float32') {\n          throw new Error(\n              `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n              `which only supports weights of type float32 not ${dtype}.`);\n        }\n      } else {\n        throw new Error(\n            `Weight ${spec.name} has unknown ` +\n            `quantization dtype ${quantization.dtype}. ` +\n            `Supported quantization dtypes are: ` +\n            `'uint8', 'uint16', and 'float16'.`);\n      }\n      const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n      const byteBuffer =\n          buffer.slice(offset, offset + size * quantizationSizeFactor);\n      const quantizedArray = (quantization.dtype === 'uint8') ?\n          new Uint8Array(byteBuffer) :\n          new Uint16Array(byteBuffer);\n      if (dtype === 'float32') {\n        if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n          values = new Float32Array(quantizedArray.length);\n          for (let i = 0; i < quantizedArray.length; i++) {\n            const v = quantizedArray[i];\n            values[i] = v * quantization.scale + quantization.min;\n          }\n        } else if (quantization.dtype === 'float16') {\n          if (float16Decode === undefined) {\n            float16Decode = getFloat16Decoder();\n          }\n          values = float16Decode(quantizedArray as Uint16Array);\n        } else {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type float32.`);\n        }\n      } else if (dtype === 'int32') {\n        if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n          throw new Error(\n              `Unsupported quantization type ${quantization.dtype} ` +\n              `for weight type int32.`);\n        }\n        values = new Int32Array(quantizedArray.length);\n        for (let i = 0; i < quantizedArray.length; i++) {\n          const v = quantizedArray[i];\n          values[i] = Math.round(v * quantization.scale + quantization.min);\n        }\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * quantizationSizeFactor;\n    } else if (dtype === 'string') {\n      const size = sizeFromShape(spec.shape);\n      values = [];\n      for (let i = 0; i < size; i++) {\n        const byteLength = new Uint32Array(\n            buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n        offset += NUM_BYTES_STRING_LENGTH;\n        const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n        (values as Uint8Array[]).push(bytes);\n        offset += byteLength;\n      }\n    } else {\n      const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n      const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n      if (dtype === 'float32') {\n        values = new Float32Array(byteBuffer);\n      } else if (dtype === 'int32') {\n        values = new Int32Array(byteBuffer);\n      } else if (dtype === 'bool') {\n        values = new Uint8Array(byteBuffer);\n      } else if (dtype === 'complex64') {\n        values = new Float32Array(byteBuffer);\n        const real = new Float32Array(values.length / 2);\n        const image = new Float32Array(values.length / 2);\n        for (let i = 0; i < real.length; i++) {\n          real[i] = values[i * 2];\n          image[i] = values[i * 2 + 1];\n        }\n        const realTensor = tensor(real, shape, 'float32');\n        const imageTensor = tensor(image, shape, 'float32');\n        out[name] = complex(realTensor, imageTensor);\n      } else {\n        throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n      }\n      offset += size * dtypeFactor;\n    }\n    if (dtype !== 'complex64') {\n      out[name] = tensor(values, shape, dtype);\n    }\n  }\n  return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n  // TODO(adarob, cais): Support quantization.\n  if (xs === null) {\n    throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n  }\n\n  let totalByteLength = 0;\n\n  // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n  // can have a different byte length from that of the `TypedArray` itself,\n  // for example, when the `TypedArray` is created from an offset in an\n  // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n  // the `TypedArray` in byte length. If an element of `xs` does not show\n  // this property, a new `TypedArray` that satisfy this property will be\n  // constructed and pushed into `normalizedXs`.\n  const normalizedXs: TypedArray[] = [];\n  xs.forEach((x: TypedArray) => {\n    totalByteLength += x.byteLength;\n    // tslint:disable:no-any\n    normalizedXs.push(\n        x.byteLength === x.buffer.byteLength ? x :\n                                               new (x.constructor as any)(x));\n    if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n          x as any instanceof Uint8Array)) {\n      throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n    }\n    // tslint:enable:no-any\n  });\n\n  const y = new Uint8Array(totalByteLength);\n  let offset = 0;\n  normalizedXs.forEach((x: TypedArray) => {\n    y.set(new Uint8Array(x.buffer), offset);\n    offset += x.byteLength;\n  });\n\n  return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n    (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n     typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n  if (useNodeBuffer) {\n    return Buffer.byteLength(str);\n  }\n  return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n  if (useNodeBuffer) {\n    return Buffer.from(buffer).toString('base64');\n  }\n  const buf = new Uint8Array(buffer);\n  let s = '';\n  for (let i = 0, l = buf.length; i < l; i++) {\n    s += String.fromCharCode(buf[i]);\n  }\n  return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n  if (useNodeBuffer) {\n    const buf = Buffer.from(str, 'base64');\n    return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n  }\n  const s = atob(str);\n  const buffer = new Uint8Array(s.length);\n  for (let i = 0; i < s.length; ++i) {\n    buffer.set([s.charCodeAt(i)], i);\n  }\n  return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n  if (buffers.length === 1) {\n    return buffers[0];\n  }\n\n  let totalByteLength = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    totalByteLength += buffer.byteLength;\n  });\n\n  const temp = new Uint8Array(totalByteLength);\n  let offset = 0;\n  buffers.forEach((buffer: ArrayBuffer) => {\n    temp.set(new Uint8Array(buffer), offset);\n    offset += buffer.byteLength;\n  });\n  return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n  const SEPARATOR = '/';\n  path = path.trim();\n  while (path.endsWith(SEPARATOR)) {\n    path = path.slice(0, path.length - 1);\n  }\n  const items = path.split(SEPARATOR);\n  return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n    ModelArtifactsInfo {\n  if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n    throw new Error('Expected JSON model topology, received ArrayBuffer.');\n  }\n\n  return {\n    dateSaved: new Date(),\n    modelTopologyType: 'JSON',\n    modelTopologyBytes: modelArtifacts.modelTopology == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n    weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n        0 :\n        stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n    weightDataBytes: modelArtifacts.weightData == null ?\n        0 :\n        modelArtifacts.weightData.byteLength,\n  };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n  const convertMantissa = (i: number): number => {\n    let m = i << 13;\n    let e = 0;\n\n    while ((m & 0x00800000) === 0) {\n      e -= 0x00800000;\n      m <<= 1;\n    }\n    m &= ~0x00800000;\n    e += 0x38800000;\n\n    return m | e;\n  };\n\n  const mantisaTable = new Uint32Array(2048);\n\n  mantisaTable[0] = 0;\n  for (let i = 1; i < 1024; i++) {\n    mantisaTable[i] = convertMantissa(i);\n  }\n  for (let i = 1024; i < 2048; i++) {\n    mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n  }\n\n  return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n  const exponentTable = new Uint32Array(64);\n\n  exponentTable[0] = 0;\n  exponentTable[31] = 0x47800000;\n  exponentTable[32] = 0x80000000;\n  exponentTable[63] = 0xc7800000;\n  for (let i = 1; i < 31; i++) {\n    exponentTable[i] = i << 23;\n  }\n  for (let i = 33; i < 63; i++) {\n    exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n  }\n\n  return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n  const offsetTable = new Uint32Array(64);\n\n  for (let i = 0; i < 64; i++) {\n    offsetTable[i] = 1024;\n  }\n  offsetTable[0] = offsetTable[32] = 0;\n\n  return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n *          the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n  // Algorithm is based off of\n  // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n  // Cache lookup tables\n  const mantisaTable = computeFloat16MantisaTable();\n  const exponentTable = computeFloat16ExponentTable();\n  const offsetTable = computeFloat16OffsetTable();\n\n  return (quantizedArray: Uint16Array) => {\n    const buffer = new ArrayBuffer(4 * quantizedArray.length);\n    const bufferUint32View = new Uint32Array(buffer);\n    for (let index = 0; index < quantizedArray.length; index++) {\n      const float16Bits = quantizedArray[index];\n      const float32Bits =\n          mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n          exponentTable[float16Bits >> 10];\n      bufferUint32View[index] = float32Bits;\n    }\n    return new Float32Array(buffer);\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n    IOHandler;\n\nexport class IORouterRegistry {\n  // Singleton instance.\n  private static instance: IORouterRegistry;\n\n  private saveRouters: IORouter[];\n  private loadRouters: IORouter[];\n\n  private constructor() {\n    this.saveRouters = [];\n    this.loadRouters = [];\n  }\n\n  private static getInstance(): IORouterRegistry {\n    if (IORouterRegistry.instance == null) {\n      IORouterRegistry.instance = new IORouterRegistry();\n    }\n    return IORouterRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerSaveRouter(saveRouter: IORouter) {\n    IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n  }\n\n  /**\n   * Register a load-handler router.\n   *\n   * @param loadRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `load` method defined or `null`.\n   */\n  static registerLoadRouter(loadRouter: IORouter) {\n    IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n  }\n\n  /**\n   * Look up IOHandler for saving, given a URL-like string.\n   *\n   * @param url\n   * @returns If only one match is found, an instance of IOHandler with the\n   * `save` method defined. If no match is found, `null`.\n   * @throws Error, if more than one match is found.\n   */\n  static getSaveHandlers(url: string|string[]): IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'save');\n  }\n\n  /**\n   * Look up IOHandler for loading, given a URL-like string.\n   *\n   * @param url\n   * @param loadOptions Optional, custom load options.\n   * @returns All valid handlers for `url`, given the currently registered\n   *   handler routers.\n   */\n  static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n      IOHandler[] {\n    return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n  }\n\n  private static getHandlers(\n      url: string|string[], handlerType: 'save'|'load',\n      loadOptions?: LoadOptions): IOHandler[] {\n    const validHandlers: IOHandler[] = [];\n    const routers = handlerType === 'load' ?\n        IORouterRegistry.getInstance().loadRouters :\n        IORouterRegistry.getInstance().saveRouters;\n    routers.forEach(router => {\n      const handler = router(url, loadOptions);\n      if (handler !== null) {\n        validHandlers.push(handler);\n      }\n    });\n    return validHandlers;\n  }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n    IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n    IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n    (url: string|string[], loadOptions?: LoadOptions) =>\n        IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n//    as the type of topology (JSON vs binary), byte size of the topology, byte\n//    size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise<void> {\n  const idbFactory = getIndexedDBFactory();\n\n  return new Promise<void>((resolve, reject) => {\n    const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n    deleteRequest.onsuccess = () => resolve();\n    deleteRequest.onerror = error => reject(error);\n  });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n  if (!env().getBool('IS_BROWSER')) {\n    // TODO(cais): Add more info about what IOHandler subtypes are available.\n    //   Maybe point to a doc page on the web and/or automatically determine\n    //   the available IOHandlers and print them in the error message.\n    throw new Error(\n        'Failed to obtain IndexedDB factory because the current environment' +\n        'is not a web browser.');\n  }\n  // tslint:disable-next-line:no-any\n  const theWindow: any = typeof window === 'undefined' ? self : window;\n  const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n      theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n      theWindow.shimIndexedDB;\n  if (factory == null) {\n    throw new Error(\n        'The current browser does not appear to support IndexedDB.');\n  }\n  return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n  const db = openRequest.result as IDBDatabase;\n  db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n  db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n  protected readonly indexedDB: IDBFactory;\n  protected readonly modelPath: string;\n\n  static readonly URL_SCHEME = 'indexeddb://';\n\n  constructor(modelPath: string) {\n    this.indexedDB = getIndexedDBFactory();\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For IndexedDB, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    // TODO(cais): Support saving GraphDef models.\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    return this.databaseAction(this.modelPath, modelArtifacts) as\n        Promise<SaveResult>;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    return this.databaseAction(this.modelPath) as Promise<ModelArtifacts>;\n  }\n\n  /**\n   * Perform database action to put model artifacts into or read model artifacts\n   * from IndexedDB object store.\n   *\n   * Whether the action is put or get depends on whether `modelArtifacts` is\n   * specified. If it is specified, the action will be put; otherwise the action\n   * will be get.\n   *\n   * @param modelPath A unique string path for the model.\n   * @param modelArtifacts If specified, it will be the model artifacts to be\n   *   stored in IndexedDB.\n   * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n   *   of `ModelArtifacts`, if the action is get.\n   */\n  private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n      Promise<ModelArtifacts|SaveResult> {\n    return new Promise<ModelArtifacts|SaveResult>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n\n        if (modelArtifacts == null) {\n          // Read model out from object store.\n          const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n          const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n          const getRequest = modelStore.get(this.modelPath);\n          getRequest.onsuccess = () => {\n            if (getRequest.result == null) {\n              db.close();\n              return reject(new Error(\n                  `Cannot find model with path '${this.modelPath}' ` +\n                  `in IndexedDB.`));\n            } else {\n              resolve(getRequest.result.modelArtifacts);\n            }\n          };\n          getRequest.onerror = error => {\n            db.close();\n            return reject(getRequest.error);\n          };\n          modelTx.oncomplete = () => db.close();\n        } else {\n          // Put model into object store.\n          const modelArtifactsInfo: ModelArtifactsInfo =\n              getModelArtifactsInfoForJSON(modelArtifacts);\n          // First, put ModelArtifactsInfo into info store.\n          const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n          let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n          const putInfoRequest =\n              infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n          let modelTx: IDBTransaction;\n          putInfoRequest.onsuccess = () => {\n            // Second, put model data into model store.\n            modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n            const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n            const putModelRequest = modelStore.put({\n              modelPath: this.modelPath,\n              modelArtifacts,\n              modelArtifactsInfo\n            });\n            putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n            putModelRequest.onerror = error => {\n              // If the put-model request fails, roll back the info entry as\n              // well.\n              infoStore = infoTx.objectStore(INFO_STORE_NAME);\n              const deleteInfoRequest = infoStore.delete(this.modelPath);\n              deleteInfoRequest.onsuccess = () => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n              deleteInfoRequest.onerror = error => {\n                db.close();\n                return reject(putModelRequest.error);\n              };\n            };\n          };\n          putInfoRequest.onerror = error => {\n            db.close();\n            return reject(putInfoRequest.error);\n          };\n          infoTx.oncomplete = () => {\n            if (modelTx == null) {\n              db.close();\n            } else {\n              modelTx.oncomplete = () => db.close();\n            }\n          };\n        }\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n      return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n *   which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n  return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n      key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n      key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n  private indexedDB: IDBFactory;\n\n  constructor() {\n    this.indexedDB = getIndexedDBFactory();\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    return new Promise<{[path: string]: ModelArtifactsInfo}>(\n        (resolve, reject) => {\n          const openRequest =\n              this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n          openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n          openRequest.onsuccess = () => {\n            const db = openRequest.result;\n            const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n            const store = tx.objectStore(INFO_STORE_NAME);\n            // tslint:disable:max-line-length\n            // Need to cast `store` as `any` here because TypeScript's DOM\n            // library does not have the `getAll()` method even though the\n            // method is supported in the latest version of most mainstream\n            // browsers:\n            // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n            // tslint:enable:max-line-length\n            // tslint:disable-next-line:no-any\n            const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n            getAllInfoRequest.onsuccess = () => {\n              const out: {[path: string]: ModelArtifactsInfo} = {};\n              for (const item of getAllInfoRequest.result) {\n                out[item.modelPath] = item.modelArtifactsInfo;\n              }\n              resolve(out);\n            };\n            getAllInfoRequest.onerror = error => {\n              db.close();\n              return reject(getAllInfoRequest.error);\n            };\n            tx.oncomplete = () => db.close();\n          };\n          openRequest.onerror = error => reject(openRequest.error);\n        });\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    return new Promise<ModelArtifactsInfo>((resolve, reject) => {\n      const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n      openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n      openRequest.onsuccess = () => {\n        const db = openRequest.result;\n        const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n        const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n        const getInfoRequest = infoStore.get(path);\n        let modelTx: IDBTransaction;\n        getInfoRequest.onsuccess = () => {\n          if (getInfoRequest.result == null) {\n            db.close();\n            return reject(new Error(\n                `Cannot find model with path '${path}' ` +\n                `in IndexedDB.`));\n          } else {\n            // First, delete the entry in the info store.\n            const deleteInfoRequest = infoStore.delete(path);\n            const deleteModelData = () => {\n              // Second, delete the entry in the model store.\n              modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n              const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n              const deleteModelRequest = modelStore.delete(path);\n              deleteModelRequest.onsuccess = () =>\n                  resolve(getInfoRequest.result.modelArtifactsInfo);\n              deleteModelRequest.onerror = error =>\n                  reject(getInfoRequest.error);\n            };\n            // Proceed with deleting model data regardless of whether deletion\n            // of info data succeeds or not.\n            deleteInfoRequest.onsuccess = deleteModelData;\n            deleteInfoRequest.onerror = error => {\n              deleteModelData();\n              db.close();\n              return reject(getInfoRequest.error);\n            };\n          }\n        };\n        getInfoRequest.onerror = error => {\n          db.close();\n          return reject(getInfoRequest.error);\n        };\n\n        infoTx.oncomplete = () => {\n          if (modelTx == null) {\n            db.close();\n          } else {\n            modelTx.oncomplete = () => db.close();\n          }\n        };\n      };\n      openRequest.onerror = error => reject(openRequest.error);\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n  if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n      typeof window.localStorage === 'undefined') {\n    throw new Error(\n        'purgeLocalStorageModels() cannot proceed because local storage is ' +\n        'unavailable in the current environment.');\n  }\n  const LS = window.localStorage;\n  const purgedModelPaths: string[] = [];\n  for (let i = 0; i < LS.length; ++i) {\n    const key = LS.key(i);\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    if (key.startsWith(prefix) && key.length > prefix.length) {\n      LS.removeItem(key);\n      const modelName = getModelPathFromKey(key);\n      if (purgedModelPaths.indexOf(modelName) === -1) {\n        purgedModelPaths.push(modelName);\n      }\n    }\n  }\n  return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n} {\n  return {\n    info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n    topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n    weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n    weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n    modelMetadata:\n        [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n  };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n  const items = key.split(PATH_SEPARATOR);\n  if (items.length < 3) {\n    throw new Error(`Invalid key format: ${key}`);\n  }\n  return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n  return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n      key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n      key;\n}\n\ndeclare type LocalStorageKeys = {\n  info: string,\n  topology: string,\n  weightSpecs: string,\n  weightData: string,\n  modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n  protected readonly LS: Storage;\n  protected readonly modelPath: string;\n  protected readonly keys: LocalStorageKeys;\n\n  static readonly URL_SCHEME = 'localstorage://';\n\n  constructor(modelPath: string) {\n    if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n        typeof window.localStorage === 'undefined') {\n      // TODO(cais): Add more info about what IOHandler subtypes are\n      // available.\n      //   Maybe point to a doc page on the web and/or automatically determine\n      //   the available IOHandlers and print them in the error message.\n      throw new Error(\n          'The current environment does not support local storage.');\n    }\n    this.LS = window.localStorage;\n\n    if (modelPath == null || !modelPath) {\n      throw new Error(\n          'For local storage, modelPath must not be null, undefined or empty.');\n    }\n    this.modelPath = modelPath;\n    this.keys = getModelKeys(this.modelPath);\n  }\n\n  /**\n   * Save model artifacts to browser local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @param modelArtifacts The model artifacts to be stored.\n   * @returns An instance of SaveResult.\n   */\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserLocalStorage.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const topology = JSON.stringify(modelArtifacts.modelTopology);\n      const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n      const modelArtifactsInfo: ModelArtifactsInfo =\n          getModelArtifactsInfoForJSON(modelArtifacts);\n\n      try {\n        this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n        this.LS.setItem(this.keys.topology, topology);\n        this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n        this.LS.setItem(\n            this.keys.weightData,\n            arrayBufferToBase64String(modelArtifacts.weightData));\n        this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n          format: modelArtifacts.format,\n          generatedBy: modelArtifacts.generatedBy,\n          convertedBy: modelArtifacts.convertedBy,\n          userDefinedMetadata: modelArtifacts.userDefinedMetadata\n        }));\n\n        return {modelArtifactsInfo};\n      } catch (err) {\n        // If saving failed, clean up all items saved so far.\n        this.LS.removeItem(this.keys.info);\n        this.LS.removeItem(this.keys.topology);\n        this.LS.removeItem(this.keys.weightSpecs);\n        this.LS.removeItem(this.keys.weightData);\n        this.LS.removeItem(this.keys.modelMetadata);\n\n        throw new Error(\n            `Failed to save model '${this.modelPath}' to local storage: ` +\n            `size quota being exceeded is a possible cause of this failure: ` +\n            `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n            `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n            `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n      }\n    }\n  }\n\n  /**\n   * Load a model from local storage.\n   *\n   * See the documentation to `browserLocalStorage` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const info =\n        JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n    if (info == null) {\n      throw new Error(\n          `In local storage, there is no model with name '${this.modelPath}'`);\n    }\n\n    if (info.modelTopologyType !== 'JSON') {\n      throw new Error(\n          'BrowserLocalStorage does not support loading non-JSON model ' +\n          'topology yet.');\n    }\n\n    const out: ModelArtifacts = {};\n\n    // Load topology.\n    const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n    if (topology == null) {\n      throw new Error(\n          `In local storage, the topology of model '${this.modelPath}' ` +\n          `is missing.`);\n    }\n    out.modelTopology = topology;\n\n    // Load weight specs.\n    const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n    if (weightSpecs == null) {\n      throw new Error(\n          `In local storage, the weight specs of model '${this.modelPath}' ` +\n          `are missing.`);\n    }\n    out.weightSpecs = weightSpecs;\n\n    // Load meta-data fields.\n    const metadataString = this.LS.getItem(this.keys.modelMetadata);\n    if (metadataString != null) {\n      const metadata = JSON.parse(metadataString) as ModelArtifacts;\n      out.format = metadata['format'];\n      out.generatedBy = metadata['generatedBy'];\n      out.convertedBy = metadata['convertedBy'];\n      out.userDefinedMetadata = metadata['userDefinedMetadata'];\n    }\n\n    // Load weight data.\n    const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n    if (weightDataBase64 == null) {\n      throw new Error(\n          `In local storage, the binary weight values of model ` +\n          `'${this.modelPath}' are missing.`);\n    }\n    out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n    return out;\n  }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n      return browserLocalStorage(\n          url.slice(BrowserLocalStorage.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n *   - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n *     model, such as date saved, type of the topology, size in bytes, etc.\n *   - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n *     style models, this is a stringized JSON.\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n *     model, can be used to decode the saved binary weight values (see\n *     item below).\n *   - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n *     weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n *   non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n *   `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n  return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n  private readonly LS: Storage;\n\n  constructor() {\n    assert(\n        env().getBool('IS_BROWSER'),\n        () => 'Current environment is not a web browser');\n    assert(\n        typeof window === 'undefined' ||\n            typeof window.localStorage !== 'undefined',\n        () => 'Current browser does not appear to support localStorage');\n    this.LS = window.localStorage;\n  }\n\n  async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n    const out: {[path: string]: ModelArtifactsInfo} = {};\n    const prefix = PATH_PREFIX + PATH_SEPARATOR;\n    const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n    for (let i = 0; i < this.LS.length; ++i) {\n      const key = this.LS.key(i);\n      if (key.startsWith(prefix) && key.endsWith(suffix)) {\n        const modelPath = getModelPathFromKey(key);\n        out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n      }\n    }\n    return out;\n  }\n\n  async removeModel(path: string): Promise<ModelArtifactsInfo> {\n    path = maybeStripScheme(path);\n    const keys = getModelKeys(path);\n    if (this.LS.getItem(keys.info) == null) {\n      throw new Error(`Cannot find model at path '${path}'`);\n    }\n    const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n    this.LS.removeItem(keys.info);\n    this.LS.removeItem(keys.topology);\n    this.LS.removeItem(keys.weightSpecs);\n    this.LS.removeItem(keys.weightData);\n    return info;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n *   string.\n * - Moving or copying model from one path to another in the same medium or from\n *   one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n  // Singleton instance.\n  private static instance: ModelStoreManagerRegistry;\n\n  private managers: {[scheme: string]: ModelStoreManager};\n\n  private constructor() {\n    this.managers = {};\n  }\n\n  private static getInstance(): ModelStoreManagerRegistry {\n    if (ModelStoreManagerRegistry.instance == null) {\n      ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n    }\n    return ModelStoreManagerRegistry.instance;\n  }\n\n  /**\n   * Register a save-handler router.\n   *\n   * @param saveRouter A function that maps a URL-like string onto an instance\n   * of `IOHandler` with the `save` method defined or `null`.\n   */\n  static registerManager(scheme: string, manager: ModelStoreManager) {\n    assert(scheme != null, () => 'scheme must not be undefined or null.');\n    if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n      scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n    }\n    assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n    const registry = ModelStoreManagerRegistry.getInstance();\n    assert(\n        registry.managers[scheme] == null,\n        () => `A model store manager is already registered for scheme '${\n            scheme}'.`);\n    registry.managers[scheme] = manager;\n  }\n\n  static getManager(scheme: string): ModelStoreManager {\n    const manager = this.getInstance().managers[scheme];\n    if (manager == null) {\n      throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n    }\n    return manager;\n  }\n\n  static getSchemes(): string[] {\n    return Object.keys(this.getInstance().managers);\n  }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n *   Scheme: e.g., 'localstorage' in the example above.\n *   Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n  if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n    throw new Error(\n        `The url string provided does not contain a scheme. ` +\n        `Supported schemes are: ` +\n        `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n  }\n  return {\n    scheme: url.split(URL_SCHEME_SUFFIX)[0],\n    path: url.split(URL_SCHEME_SUFFIX)[1],\n  };\n}\n\nasync function cloneModelInternal(\n    sourceURL: string, destURL: string,\n    deleteSource = false): Promise<ModelArtifactsInfo> {\n  assert(\n      sourceURL !== destURL,\n      () => `Old path and new path are the same: '${sourceURL}'`);\n\n  const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n  assert(\n      loadHandlers.length > 0,\n      () => `Copying failed because no load handler is found for source URL ${\n          sourceURL}.`);\n  assert(\n      loadHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `load handlers for source URL ${sourceURL}.`);\n  const loadHandler = loadHandlers[0];\n\n  const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n  assert(\n      saveHandlers.length > 0,\n      () => `Copying failed because no save handler is found for destination ` +\n          `URL ${destURL}.`);\n  assert(\n      saveHandlers.length < 2,\n      () => `Copying failed because more than one (${loadHandlers.length}) ` +\n          `save handlers for destination URL ${destURL}.`);\n  const saveHandler = saveHandlers[0];\n\n  const sourceScheme = parseURL(sourceURL).scheme;\n  const sourcePath = parseURL(sourceURL).path;\n  const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n  const modelArtifacts = await loadHandler.load();\n\n  // If moving within the same storage medium, remove the old model as soon as\n  // the loading is done. Without doing this, it is possible that the combined\n  // size of the two models will cause the cloning to fail.\n  if (deleteSource && sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  const saveResult = await saveHandler.save(modelArtifacts);\n\n  // If moving between mediums, the deletion is done after the save succeeds.\n  // This guards against the case in which saving to the destination medium\n  // fails.\n  if (deleteSource && !sameMedium) {\n    await ModelStoreManagerRegistry.getManager(sourceScheme)\n        .removeModel(sourcePath);\n  }\n\n  return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n *   'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n  const schemes = ModelStoreManagerRegistry.getSchemes();\n  const out: {[url: string]: ModelArtifactsInfo} = {};\n  for (const scheme of schemes) {\n    const schemeOut =\n        await ModelStoreManagerRegistry.getManager(scheme).listModels();\n    for (const path in schemeOut) {\n      const url = scheme + URL_SCHEME_SUFFIX + path;\n      out[url] = schemeOut[path];\n    }\n  }\n  return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n *   'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n *   is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise<ModelArtifactsInfo> {\n  const schemeAndPath = parseURL(url);\n  const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n  return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function copyModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = false;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n *    `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n *     'localstorage://demo/management/model1',\n *     'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n *   is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n *   if `oldPath` and `newPath` are identical.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Management',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nasync function moveModel(\n    sourceURL: string, destURL: string): Promise<ModelArtifactsInfo> {\n  const deleteSource = true;\n  return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n  // According to the spec, the built-in encoder can do only UTF-8 encoding.\n  // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n  private textEncoder: TextEncoder;\n\n  fetch(path: string, init?: RequestInit): Promise<Response> {\n    return fetch(path, init);\n  }\n\n  now(): number {\n    return performance.now();\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Browser's encoder only supports utf-8, but got ${encoding}`);\n    }\n    if (this.textEncoder == null) {\n      this.textEncoder = new TextEncoder();\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    return new TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_BROWSER')) {\n  env().setPlatform('browser', new PlatformBrowser());\n\n  // Register LocalStorage IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n  } catch (err) {\n  }\n\n  // Register IndexedDB IOHandler\n  try {\n    ModelStoreManagerRegistry.registerManager(\n        BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n  } catch (err) {\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n  // tslint:disable-next-line:no-require-imports\n  importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise<Response>;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n  systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n  systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n  return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n  private textEncoder: TextEncoder;\n  // tslint:disable-next-line:no-any\n  util: any;\n\n  constructor() {\n    // tslint:disable-next-line:no-require-imports\n    this.util = require('util');\n    // According to the spec, the built-in encoder can do only UTF-8 encoding.\n    // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n    this.textEncoder = new this.util.TextEncoder();\n  }\n\n  fetch(path: string, requestInits?: RequestInit): Promise<Response> {\n    if (env().global.fetch != null) {\n      return env().global.fetch(path, requestInits);\n    }\n\n    if (systemFetch == null) {\n      systemFetch = getNodeFetch.importFetch();\n    }\n    return systemFetch(path, requestInits);\n  }\n\n  now(): number {\n    const time = process.hrtime();\n    return time[0] * 1000 + time[1] / 1000000;\n  }\n\n  encode(text: string, encoding: string): Uint8Array {\n    if (encoding !== 'utf-8' && encoding !== 'utf8') {\n      throw new Error(\n          `Node built-in encoder only supports utf-8, but got ${encoding}`);\n    }\n    return this.textEncoder.encode(text);\n  }\n  decode(bytes: Uint8Array, encoding: string): string {\n    if (bytes.length === 0) {\n      return '';\n    }\n    return new this.util.TextDecoder(encoding).decode(bytes);\n  }\n}\n\nif (env().get('IS_NODE')) {\n  env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer<R extends Rank, D extends DataType = 'float32'>(\n    shape: ShapeMap[R], dtype: D = 'float32' as D,\n    values?: DataTypeMap[D]): TensorBuffer<R, D> {\n  dtype = dtype || 'float32' as D;\n  util.assertNonNegativeIntegerDimensions(shape);\n  return new TensorBuffer<R, D>(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_<T extends Tensor>(x: T|TensorLike, dtype: DataType): T {\n  const $x = convertToTensor(x, 'x', 'cast');\n\n  // Sanity checks.\n  if (!util.isValidDtype(dtype)) {\n    throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n  }\n  if (dtype === 'string' && $x.dtype !== 'string' ||\n      dtype !== 'string' && $x.dtype === 'string') {\n    throw new Error('Only strings can be casted to strings');\n  }\n\n  const inputs: CastInputs = {x: $x};\n  const attrs: CastAttrs = {dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n      null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'clone', null);\n  const forward = () =>\n      ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n  const inputs: IdentityInputs = {x: $x};\n\n  // Note this op is called tf.identity in python. Hence the kernel name used\n  // here.\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print<T extends Tensor>(x: T, verbose = false): void {\n  console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n  buffer,\n  cast,\n  clone,\n  print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer<T>(f: () => T): Promise<T> {\n  return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n  private readonly modelTopologyFileName: string;\n  private readonly weightDataFileName: string;\n  private readonly jsonAnchor: HTMLAnchorElement;\n  private readonly weightDataAnchor: HTMLAnchorElement;\n\n  static readonly URL_SCHEME = 'downloads://';\n\n  constructor(fileNamePrefix?: string) {\n    if (!env().getBool('IS_BROWSER')) {\n      // TODO(cais): Provide info on what IOHandlers are available under the\n      //   current environment.\n      throw new Error(\n          'browserDownloads() cannot proceed because the current environment ' +\n          'is not a browser.');\n    }\n\n    if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n      fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n    }\n    if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n      fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n    }\n\n    this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n    this.weightDataFileName =\n        fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (typeof (document) === 'undefined') {\n      throw new Error(\n          'Browser downloads are not supported in ' +\n          'this environment since `document` is not present');\n    }\n    const weightsURL = window.URL.createObjectURL(new Blob(\n        [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserDownloads.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    } else {\n      const weightsManifest: WeightsManifestConfig = [{\n        paths: ['./' + this.weightDataFileName],\n        weights: modelArtifacts.weightSpecs\n      }];\n      const modelTopologyAndWeightManifest: ModelJSON = {\n        modelTopology: modelArtifacts.modelTopology,\n        format: modelArtifacts.format,\n        generatedBy: modelArtifacts.generatedBy,\n        convertedBy: modelArtifacts.convertedBy,\n        weightsManifest\n      };\n      const modelTopologyAndWeightManifestURL =\n          window.URL.createObjectURL(new Blob(\n              [JSON.stringify(modelTopologyAndWeightManifest)],\n              {type: 'application/json'}));\n\n      // If anchor elements are not provided, create them without attaching them\n      // to parents, so that the downloaded file names can be controlled.\n      const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n                                                   this.jsonAnchor;\n      jsonAnchor.download = this.modelTopologyFileName;\n      jsonAnchor.href = modelTopologyAndWeightManifestURL;\n      // Trigger downloads by evoking a click event on the download anchors.\n      // When multiple downloads are started synchronously, Firefox will only\n      // save the last one.\n      await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n      if (modelArtifacts.weightData != null) {\n        const weightDataAnchor = this.weightDataAnchor == null ?\n            document.createElement('a') :\n            this.weightDataAnchor;\n        weightDataAnchor.download = this.weightDataFileName;\n        weightDataAnchor.href = weightsURL;\n        await defer(\n            () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n      }\n\n      return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n    }\n  }\n}\n\nclass BrowserFiles implements IOHandler {\n  private readonly files: File[];\n\n  constructor(files: File[]) {\n    if (files == null || files.length < 1) {\n      throw new Error(\n          `When calling browserFiles, at least 1 file is required, ` +\n          `but received ${files}`);\n    }\n    this.files = files;\n  }\n\n  async load(): Promise<ModelArtifacts> {\n    const jsonFile = this.files[0];\n    const weightFiles = this.files.slice(1);\n\n    return new Promise<ModelArtifacts>((resolve, reject) => {\n      const jsonReader = new FileReader();\n      jsonReader.onload = (event: Event) => {\n        // tslint:disable-next-line:no-any\n        const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n        const modelTopology = modelJSON.modelTopology;\n        if (modelTopology == null) {\n          reject(new Error(\n              `modelTopology field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        if (weightFiles.length === 0) {\n          resolve({modelTopology});\n        }\n\n        const weightsManifest = modelJSON.weightsManifest;\n        if (weightsManifest == null) {\n          reject(new Error(\n              `weightManifest field is missing from file ${jsonFile.name}`));\n          return;\n        }\n\n        let pathToFile: {[path: string]: File};\n        try {\n          pathToFile =\n              this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n        } catch (err) {\n          reject(err);\n          return;\n        }\n\n        const weightSpecs: WeightsManifestEntry[] = [];\n        const paths: string[] = [];\n        const perFileBuffers: ArrayBuffer[] = [];\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            paths.push(path);\n            perFileBuffers.push(null);\n          });\n          weightSpecs.push(...weightsGroup.weights);\n        });\n\n        weightsManifest.forEach(weightsGroup => {\n          weightsGroup.paths.forEach(path => {\n            const weightFileReader = new FileReader();\n            weightFileReader.onload = (event: Event) => {\n              // tslint:disable-next-line:no-any\n              const weightData = (event.target as any).result as ArrayBuffer;\n              const index = paths.indexOf(path);\n              perFileBuffers[index] = weightData;\n              if (perFileBuffers.indexOf(null) === -1) {\n                resolve({\n                  modelTopology,\n                  weightSpecs,\n                  weightData: concatenateArrayBuffers(perFileBuffers),\n                  format: modelJSON.format,\n                  generatedBy: modelJSON.generatedBy,\n                  convertedBy: modelJSON.convertedBy,\n                  userDefinedMetadata: modelJSON.userDefinedMetadata\n                });\n              }\n            };\n            weightFileReader.onerror = error =>\n                reject(`Failed to weights data from file of path '${path}'.`);\n            weightFileReader.readAsArrayBuffer(pathToFile[path]);\n          });\n        });\n      };\n      jsonReader.onerror = error => reject(\n          `Failed to read model topology and weights manifest JSON ` +\n          `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n          `Keras-style tf.Model artifacts only.`);\n      jsonReader.readAsText(jsonFile);\n    });\n  }\n\n  /**\n   * Check the compatibility between weights manifest and weight files.\n   */\n  private checkManifestAndWeightFiles(\n      manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n    const basenames: string[] = [];\n    const fileNames = files.map(file => basename(file.name));\n    const pathToFile: {[path: string]: File} = {};\n    for (const group of manifest) {\n      group.paths.forEach(path => {\n        const pathBasename = basename(path);\n        if (basenames.indexOf(pathBasename) !== -1) {\n          throw new Error(\n              `Duplicate file basename found in weights manifest: ` +\n              `'${pathBasename}'`);\n        }\n        basenames.push(pathBasename);\n        if (fileNames.indexOf(pathBasename) === -1) {\n          throw new Error(\n              `Weight file with basename '${pathBasename}' is not provided.`);\n        } else {\n          pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n        }\n      });\n    }\n\n    if (basenames.length !== files.length) {\n      throw new Error(\n          `Mismatch in the number of files in weights manifest ` +\n          `(${basenames.length}) and the number of weight files provided ` +\n          `(${files.length}).`);\n    }\n    return pathToFile;\n  }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n  if (!env().getBool('IS_BROWSER')) {\n    return null;\n  } else {\n    if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n      return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n    } else {\n      return null;\n    }\n  }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n *     {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * //   'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n *   `tf.Model`, `fileNamePrefix` should follow either of the following two\n *   formats:\n *   1. `null` or `undefined`, in which case the default file\n *      names will be used:\n *      - 'model.json' for the JSON file containing the model topology and\n *        weights manifest.\n *      - 'model.weights.bin' for the binary file containing the binary weight\n *        values.\n *   2. A single string or an Array of a single string, as the file name prefix.\n *      For example, if `'foo'` is provided, the downloaded JSON\n *      file and binary weights file will be named 'foo.json' and\n *      'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n  return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * //   elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (`<input type=\"file\" ...>`)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n *   which an `Array` of `File`s is expected (in that order):\n *   - A JSON file containing the model topology and weight manifest.\n *   - Optionally, One or more binary files containing the binary weights.\n *     These files must have names that match the paths in the `weightsManifest`\n *     contained by the aforementioned JSON file, or errors will be thrown\n *     during loading. These weights files have the same format as the ones\n *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n *     Python PIP package. If no weights files are provided, only the model\n *     topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n  return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n    promises: Array<Promise<{}|void>>, onProgress: OnProgressCallback,\n    startFraction?: number, endFraction?: number) {\n  checkPromises(promises);\n  startFraction = startFraction == null ? 0 : startFraction;\n  endFraction = endFraction == null ? 1 : endFraction;\n  checkFraction(startFraction, endFraction);\n  let resolvedPromise = 0;\n\n  const registerMonitor = (promise: Promise<{}>) => {\n    promise.then(value => {\n      const fraction = startFraction +\n          ++resolvedPromise / promises.length * (endFraction - startFraction);\n      // pass fraction as parameter to callback function.\n      onProgress(fraction);\n      return value;\n    });\n    return promise;\n  };\n\n  function checkPromises(promises: Array<Promise<{}|void>>): void {\n    assert(\n        promises != null && Array.isArray(promises) && promises.length > 0,\n        () => 'promises must be a none empty array');\n  }\n\n  function checkFraction(startFraction: number, endFraction: number): void {\n    assert(\n        startFraction >= 0 && startFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got startFraction ${startFraction}`);\n    assert(\n        endFraction >= 0 && endFraction <= 1,\n        () => `Progress fraction must be in range [0, 1], but ` +\n            `got endFraction ${endFraction}`);\n    assert(\n        endFraction >= startFraction,\n        () => `startFraction must be no more than endFraction, but ` +\n            `got startFraction ${startFraction} and endFraction ` +\n            `${endFraction}`);\n  }\n\n  return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n *   before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n *   length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n    fetchURLs: string[], loadOptions?: LoadOptions): Promise<ArrayBuffer[]> {\n  if (loadOptions == null) {\n    loadOptions = {};\n  }\n\n  const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n                                                    loadOptions.fetchFunc;\n\n  // Create the requests for all of the weights in parallel.\n  const requests = fetchURLs.map(\n      fetchURL =>\n          fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n  const fetchStartFraction = 0;\n  const fetchEndFraction = 0.5;\n\n  const responses = loadOptions.onProgress == null ?\n      await Promise.all(requests) :\n      await monitorPromisesProgress(\n          requests, loadOptions.onProgress, fetchStartFraction,\n          fetchEndFraction);\n\n  const bufferPromises = responses.map(response => response.arrayBuffer());\n\n  const bufferStartFraction = 0.5;\n  const bufferEndFraction = 1;\n\n  const buffers = loadOptions.onProgress == null ?\n      await Promise.all(bufferPromises) :\n      await monitorPromisesProgress(\n          bufferPromises, loadOptions.onProgress, bufferStartFraction,\n          bufferEndFraction);\n  return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n *     Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n    manifest: WeightsManifestConfig, filePathPrefix = '',\n    weightNames?: string[],\n    requestInit?: RequestInit): Promise<NamedTensorMap> {\n  // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n  // single weight from a group, the whole group will be fetched. At a future\n  // date, we should support fetching only the individual shards within a\n  // group that are needed to reconstruct the requested weight.\n  // TODO(cais): Use `decodeWeights` for implementation.\n\n  const fetchWeights = (fetchUrls: string[]) =>\n      loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n  const loadWeights = weightsLoaderFactory(fetchWeights);\n\n  return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n *   filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n *   fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n    fetchWeightsFunction: (fetchUrls: string[]) => Promise<ArrayBuffer[]>):\n    (manifest: WeightsManifestConfig, filePathPrefix?: string,\n     weightNames?: string[]) => Promise<NamedTensorMap> {\n  return async(\n             manifest: WeightsManifestConfig, filePathPrefix = '',\n             weightNames?: string[]): Promise<NamedTensorMap> => {\n    // Collect all the groups, weights, and their relative offsets to be\n    // fetched.\n    const groupIndicesToFetchMap = manifest.map(() => false);\n    const groupWeightsToFetch: {\n      [group: number]: Array<{\n        manifestEntry: WeightsManifestEntry; groupOffset: number;\n        sizeBytes: number;\n      }>\n    } = {};\n    const weightsFound =\n        weightNames != null ? weightNames.map(() => false) : [];\n    const allManifestWeightNames: string[] = [];\n    manifest.forEach((manifestGroupConfig, groupIndex) => {\n      let groupOffset = 0;\n      manifestGroupConfig.weights.forEach(weightsEntry => {\n        const rawDtype = ('quantization' in weightsEntry) ?\n            weightsEntry.quantization.dtype :\n            weightsEntry.dtype;\n\n        const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n            util.sizeFromShape(weightsEntry.shape);\n\n        const enqueueWeightsForFetchingFn = () => {\n          groupIndicesToFetchMap[groupIndex] = true;\n          if (groupWeightsToFetch[groupIndex] == null) {\n            groupWeightsToFetch[groupIndex] = [];\n          }\n\n          groupWeightsToFetch[groupIndex].push({\n            manifestEntry: weightsEntry,\n            groupOffset,\n            sizeBytes: weightsBytes\n          });\n        };\n\n        if (weightNames != null) {\n          weightNames.forEach((weightName, weightIndex) => {\n            if (weightName === weightsEntry.name) {\n              enqueueWeightsForFetchingFn();\n              weightsFound[weightIndex] = true;\n            }\n          });\n        } else {\n          enqueueWeightsForFetchingFn();\n        }\n\n        allManifestWeightNames.push(weightsEntry.name);\n        groupOffset += weightsBytes;\n      });\n    });\n\n    if (!weightsFound.every(found => found)) {\n      const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n      throw new Error(\n          `Could not find weights in manifest with names: ` +\n          `${weightsNotFound.join(', ')}. \\n` +\n          `Manifest JSON has weights with names: ` +\n          `${allManifestWeightNames.join(', ')}.`);\n    }\n\n    // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n    // IDs.\n    const groupIndicesToFetch =\n        groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n          if (shouldFetch) {\n            accumulator.push(i);\n          }\n          return accumulator;\n        }, []);\n\n    const fetchUrls: string[] = [];\n    groupIndicesToFetch.forEach(i => {\n      manifest[i].paths.forEach(filepath => {\n        const fetchUrl = filePathPrefix +\n            (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n        fetchUrls.push(fetchUrl);\n      });\n    });\n    const buffers = await fetchWeightsFunction(fetchUrls);\n\n    const weightsTensorMap: NamedTensorMap = {};\n    let bufferIndexOffset = 0;\n    groupIndicesToFetch.forEach(i => {\n      const numBuffers = manifest[i].paths.length;\n\n      let groupBytes = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        groupBytes += buffers[bufferIndexOffset + i].byteLength;\n      }\n\n      // Create a buffer for the whole group.\n      const groupBuffer = new ArrayBuffer(groupBytes);\n      const groupByteBuffer = new Uint8Array(groupBuffer);\n      let groupBufferOffset = 0;\n      for (let i = 0; i < numBuffers; i++) {\n        const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n        groupByteBuffer.set(buffer, groupBufferOffset);\n        groupBufferOffset += buffer.byteLength;\n      }\n\n      const weightsEntries = groupWeightsToFetch[i];\n      weightsEntries.forEach(weightsEntry => {\n        const byteBuffer = groupBuffer.slice(\n            weightsEntry.groupOffset,\n            weightsEntry.groupOffset + weightsEntry.sizeBytes);\n        const nameToTensorMap =\n            decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n        for (const name in nameToTensorMap) {\n          weightsTensorMap[name] = nameToTensorMap[name];\n        }\n      });\n\n      bufferIndexOffset += numBuffers;\n    });\n\n    return weightsTensorMap;\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n  protected readonly path: string;\n  protected readonly requestInit: RequestInit;\n\n  private readonly fetch: Function;\n  private readonly weightUrlConverter: (weightName: string) => Promise<string>;\n\n  readonly DEFAULT_METHOD = 'POST';\n\n  static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n  private readonly weightPathPrefix: string;\n  private readonly onProgress: OnProgressCallback;\n\n  constructor(path: string, loadOptions?: LoadOptions) {\n    if (loadOptions == null) {\n      loadOptions = {};\n    }\n    this.weightPathPrefix = loadOptions.weightPathPrefix;\n    this.onProgress = loadOptions.onProgress;\n    this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n    if (loadOptions.fetchFunc != null) {\n      assert(\n          typeof loadOptions.fetchFunc === 'function',\n          () => 'Must pass a function that matches the signature of ' +\n              '`fetch` (see ' +\n              'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n      this.fetch = loadOptions.fetchFunc;\n    } else {\n      this.fetch = env().platform.fetch;\n    }\n\n    assert(\n        path != null && path.length > 0,\n        () => 'URL path for http must not be null, undefined or ' +\n            'empty.');\n\n    if (Array.isArray(path)) {\n      assert(\n          path.length === 2,\n          () => 'URL paths for http must have a length of 2, ' +\n              `(actual length is ${path.length}).`);\n    }\n    this.path = path;\n\n    if (loadOptions.requestInit != null &&\n        loadOptions.requestInit.body != null) {\n      throw new Error(\n          'requestInit is expected to have no pre-existing body, but has one.');\n    }\n    this.requestInit = loadOptions.requestInit || {};\n  }\n\n  async save(modelArtifacts: ModelArtifacts): Promise<SaveResult> {\n    if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n      throw new Error(\n          'BrowserHTTPRequest.save() does not support saving model topology ' +\n          'in binary formats yet.');\n    }\n\n    const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n    init.body = new FormData();\n\n    const weightsManifest: WeightsManifestConfig = [{\n      paths: ['./model.weights.bin'],\n      weights: modelArtifacts.weightSpecs,\n    }];\n    const modelTopologyAndWeightManifest: ModelJSON = {\n      modelTopology: modelArtifacts.modelTopology,\n      format: modelArtifacts.format,\n      generatedBy: modelArtifacts.generatedBy,\n      convertedBy: modelArtifacts.convertedBy,\n      userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n      weightsManifest\n    };\n\n    init.body.append(\n        'model.json',\n        new Blob(\n            [JSON.stringify(modelTopologyAndWeightManifest)],\n            {type: JSON_TYPE}),\n        'model.json');\n\n    if (modelArtifacts.weightData != null) {\n      init.body.append(\n          'model.weights.bin',\n          new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n          'model.weights.bin');\n    }\n\n    const response = await this.fetch(this.path, init);\n\n    if (response.ok) {\n      return {\n        modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n        responses: [response],\n      };\n    } else {\n      throw new Error(\n          `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n          `${response.status}.`);\n    }\n  }\n\n  /**\n   * Load model artifacts via HTTP request(s).\n   *\n   * See the documentation to `tf.io.http` for details on the saved\n   * artifacts.\n   *\n   * @returns The loaded model artifacts (if loading succeeds).\n   */\n  async load(): Promise<ModelArtifacts> {\n    const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n    if (!modelConfigRequest.ok) {\n      throw new Error(\n          `Request to ${this.path} failed with status code ` +\n          `${modelConfigRequest.status}. Please verify this URL points to ` +\n          `the model JSON of the model to load.`);\n    }\n    let modelConfig: ModelJSON;\n    try {\n      modelConfig = await modelConfigRequest.json();\n    } catch (e) {\n      let message = `Failed to parse model JSON of response from ${this.path}.`;\n      // TODO(nsthorat): Remove this after some time when we're comfortable that\n      // .pb files are mostly gone.\n      if (this.path.endsWith('.pb')) {\n        message += ' Your path contains a .pb file extension. ' +\n            'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n            'in favor of .json models. You can re-convert your Python ' +\n            'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n            'or you can convert your.pb models with the \\'pb2json\\'' +\n            'NPM script in the tensorflow/tfjs-converter repository.';\n      } else {\n        message += ' Please make sure the server is serving valid ' +\n            'JSON for this request.';\n      }\n      throw new Error(message);\n    }\n    const modelTopology = modelConfig.modelTopology;\n    const weightsManifest = modelConfig.weightsManifest;\n    const generatedBy = modelConfig.generatedBy;\n    const convertedBy = modelConfig.convertedBy;\n    const format = modelConfig.format;\n    const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n    // We do not allow both modelTopology and weightsManifest to be missing.\n    if (modelTopology == null && weightsManifest == null) {\n      throw new Error(\n          `The JSON from HTTP path ${this.path} contains neither model ` +\n          `topology or manifest for weights.`);\n    }\n\n    let weightSpecs: WeightsManifestEntry[];\n    let weightData: ArrayBuffer;\n    if (weightsManifest != null) {\n      const results = await this.loadWeights(weightsManifest);\n      [weightSpecs, weightData] = results;\n    }\n\n    const artifacts: ModelArtifacts = {\n      modelTopology,\n      weightSpecs,\n      weightData,\n      userDefinedMetadata,\n      generatedBy,\n      convertedBy,\n      format\n    };\n\n    const initializer = modelConfig.modelInitializer;\n    if (initializer) {\n      artifacts.modelInitializer = initializer;\n    }\n\n    return artifacts;\n  }\n\n  private async loadWeights(weightsManifest: WeightsManifestConfig):\n      Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n    const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n    const [prefix, suffix] = parseUrl(weightPath);\n    const pathPrefix = this.weightPathPrefix || prefix;\n\n    const weightSpecs = [];\n    for (const entry of weightsManifest) {\n      weightSpecs.push(...entry.weights);\n    }\n\n    const fetchURLs: string[] = [];\n    const urlPromises: Array<Promise<string>> = [];\n    for (const weightsGroup of weightsManifest) {\n      for (const path of weightsGroup.paths) {\n        if (this.weightUrlConverter != null) {\n          urlPromises.push(this.weightUrlConverter(path));\n        } else {\n          fetchURLs.push(pathPrefix + path + suffix);\n        }\n      }\n    }\n\n    if (this.weightUrlConverter) {\n      fetchURLs.push(...await Promise.all(urlPromises));\n    }\n\n    const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n      requestInit: this.requestInit,\n      fetchFunc: this.fetch,\n      onProgress: this.onProgress\n    });\n    return [weightSpecs, concatenateArrayBuffers(buffers)];\n  }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n  const lastSlash = url.lastIndexOf('/');\n  const lastSearchParam = url.lastIndexOf('?');\n  const prefix = url.substring(0, lastSlash);\n  const suffix =\n      lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n  return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n  return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n    (url: string, loadOptions?: LoadOptions) => {\n      if (typeof fetch === 'undefined' &&\n          (loadOptions == null || loadOptions.fetchFunc == null)) {\n        // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n        // an environment that is not the browser or node they have to setup a\n        // global fetch polyfill.\n        return null;\n      } else {\n        let isHTTP = true;\n        if (Array.isArray(url)) {\n          isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n        } else {\n          isHTTP = isHTTPScheme(url);\n        }\n        if (isHTTP) {\n          return http(url, loadOptions);\n        }\n      }\n      return null;\n    };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n *   - A JSON file consisting of `modelTopology` and `weightsManifest`.\n *   - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n *     tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n *     'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n *   Can be an absolute HTTP path (e.g.,\n *   'http://localhost:8000/model-upload)') or a relative path (e.g.,\n *   './model-upload').\n * @param requestInit Request configurations to be used when sending\n *    HTTP request to server using `fetch`. It can contain fields such as\n *    `method`, `credentials`, `headers`, `mode`, etc. See\n *    https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n *    for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n *   following fields:\n *   - weightPathPrefix Optional, this specifies the path prefix for weight\n *     files, by default this is calculated from the path param.\n *   - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n *     the `fetch` from node-fetch can be used here.\n *   - onProgress Optional, progress callback function, fired periodically\n *     before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n *   heading: 'Models',\n *   subheading: 'Loading',\n *   namespace: 'io',\n *   ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n  return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n    path: string, loadOptions?: LoadOptions): IOHandler {\n  return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n  constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n  async load(): Promise<ModelArtifacts> {\n    return this.modelArtifacts;\n  }\n}\n\nclass PassthroughSaver implements IOHandler {\n  constructor(\n      private readonly saveHandler:\n          (artifacts: ModelArtifacts) => Promise<SaveResult>) {}\n\n  async save(modelArtifacts: ModelArtifacts) {\n    return this.saveHandler(modelArtifacts);\n  }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n *     modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n *   the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n *   names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n *   concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n    modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n    weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n  if (arguments.length === 1) {\n    const isModelArtifacts =\n        (modelArtifacts as ModelArtifacts).modelTopology != null ||\n        (modelArtifacts as ModelArtifacts).weightSpecs != null;\n    if (isModelArtifacts) {\n      return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n    } else {\n      // Legacy support: with only modelTopology.\n      // TODO(cais): Remove this deprecated API.\n      console.warn(\n          'Please call tf.io.fromMemory() with only one argument. ' +\n          'The argument should be of type ModelArtifacts. ' +\n          'The multi-argument signature of tf.io.fromMemory() has been ' +\n          'deprecated and will be removed in a future release.');\n      return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n    }\n  } else {\n    // Legacy support.\n    // TODO(cais): Remove this deprecated API.\n    console.warn(\n        'Please call tf.io.fromMemory() with only one argument. ' +\n        'The argument should be of type ModelArtifacts. ' +\n        'The multi-argument signature of tf.io.fromMemory() has been ' +\n        'deprecated and will be removed in a future release.');\n    return new PassthroughLoader({\n      modelTopology: modelArtifacts as {},\n      weightSpecs,\n      weightData,\n      trainingConfig\n    });\n  }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n *   // ... do something with the artifacts ...\n *   return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n *     `SaveResult`.\n */\nexport function withSaveHandler(\n    saveHandler: (artifacts: ModelArtifacts) =>\n        Promise<SaveResult>): IOHandler {\n  return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n  browserFiles,\n  browserHTTPRequest,\n  concatenateArrayBuffers,\n  decodeWeights,\n  encodeWeights,\n  fromMemory,\n  getLoadHandlers,\n  getModelArtifactsInfoForJSON,\n  getSaveHandlers,\n  http,\n  IOHandler,\n  isHTTPScheme,\n  LoadHandler,\n  LoadOptions,\n  loadWeights,\n  ModelArtifacts,\n  ModelArtifactsInfo,\n  ModelJSON,\n  ModelStoreManager,\n  OnProgressCallback,\n  registerLoadRouter,\n  registerSaveRouter,\n  RequestDetails,\n  SaveConfig,\n  SaveHandler,\n  SaveResult,\n  WeightGroup,\n  weightsLoaderFactory,\n  WeightsManifestConfig,\n  WeightsManifestEntry,\n  withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  const $x = convertToTensor(x, 'x', 'reshape', null);\n\n  const inputs: ReshapeInputs = {x: $x};\n  const attrs: ReshapeAttrs = {shape};\n  const forward: ForwardFunc<\n      Tensor<R>> = (backend: KernelBackend, save: GradSaveFunc) => {\n    shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n    util.assert(\n        $x.size === util.sizeFromShape(shape),\n        () => 'new shape and old shape must have the same number of elements.');\n    save([$x]);\n    return backend.reshape($x, shape);\n  };\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n      attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print();  // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike, transposeA = false,\n    transposeB = false): T {\n  let $a = convertToTensor(a, 'a', 'matMul');\n  let $b = convertToTensor(b, 'b', 'matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n          `got ranks ${$a.rank} and ${$b.rank}.`);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n                           reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n                           reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([a3D, b3D]);\n\n    return backend.batchMatMul(\n        a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n  };\n\n  const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n  const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n      attrs as {} as NamedAttrMap);\n\n  return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n *     not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n    indices: Tensor|TensorLike, depth: number, onValue = 1,\n    offValue = 0): Tensor {\n  if (depth < 2) {\n    throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n  }\n  const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n  const outShape = [...$indices.shape, depth];\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$indices]);\n    return reshape(\n        backend.oneHot(\n            reshape($indices, [$indices.size]), depth, onValue, offValue),\n        outShape);\n  };\n\n  const inputs: OneHotInputs = {indices: $indices};\n  const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print();  // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_<T extends Tensor>(x: T|TensorLike, perm?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'transpose');\n\n  if (perm == null) {\n    perm = $x.shape.map((s, i) => i).reverse();\n  }\n  util.assert(\n      $x.rank === perm.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of perm ${perm}.`);\n  perm.forEach(axis => {\n    util.assert(\n        axis >= 0 && axis < $x.rank,\n        () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n            ` but got ${perm}`);\n  });\n\n  if ($x.rank <= 1) {\n    return $x.clone();\n  }\n\n  const inputs: TransposeInputs = {x: $x};\n  const attrs: TransposeAttrs = {perm};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n      null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * //  [0, 1, 1],\n * //  [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n *   for the classes. The shape is `[numExamples]`, where\n *   `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n *   0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n *   Its value must be larger than the largest element in `labels` and\n *   `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n *   row `r` and column `c` is the number of times examples of actual class\n *   `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n    labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n    numClasses: number): Tensor2D {\n  const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n  util.assert(\n      numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n      () => `If provided, numClasses must be a positive integer, ` +\n          `but got ${numClasses}`);\n  util.assert(\n      $labels.rank === 1,\n      () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n  util.assert(\n      $predictions.rank === 1,\n      () => `Expected the rank of predictions to be 1, ` +\n          `but got ${$predictions.rank}`);\n  util.assert(\n      $labels.shape[0] === $predictions.shape[0],\n      () => `Mismatch in the number of examples: ` +\n          `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n          `Labels and predictions should have the same number of elements.`);\n  util.assert(\n      numClasses > 0 && Number.isInteger(numClasses),\n      () => `numClasses is required to be a positive integer, but got ` +\n          `${numClasses}`);\n  // TODO(cais): In the future, if oneHot supports tensors inputs for\n  //   `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n  const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n  const oneHotPredictions =\n      oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n  const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n  return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided,  it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n    values: TensorLike3D, shape?: [number, number, number],\n    dtype?: DataType): Tensor3D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 3) {\n    throw new Error('tensor3d() requires shape to have three numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor3d() requires values to be number[][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor3d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement,\n    numChannels = 3): Tensor3D {\n  // Sanity checks.\n  if (numChannels > 4) {\n    throw new Error(\n        'Cannot construct Tensor with more than 4 channels from pixels.');\n  }\n  if (pixels == null) {\n    throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n  }\n  let isPixelData = false;\n  let isImageData = false;\n  let isVideo = false;\n  let isImage = false;\n  let isCanvasLike = false;\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    isPixelData = true;\n  } else if (\n      typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n    isImageData = true;\n  } else if (\n      typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement) {\n    isVideo = true;\n  } else if (\n      typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement) {\n    isImage = true;\n    // tslint:disable-next-line: no-any\n  } else if ((pixels as any).getContext != null) {\n    isCanvasLike = true;\n  } else {\n    throw new Error(\n        'pixels passed to tf.browser.fromPixels() must be either an ' +\n        `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n        `in browser, or OffscreenCanvas, ImageData in webworker` +\n        ` or {data: Uint32Array, width: number, height: number}, ` +\n        `but was ${(pixels as {}).constructor.name}`);\n  }\n  if (isVideo) {\n    const HAVE_CURRENT_DATA_READY_STATE = 2;\n    if (isVideo &&\n        (pixels as HTMLVideoElement).readyState <\n            HAVE_CURRENT_DATA_READY_STATE) {\n      throw new Error(\n          'The video element has not loaded data yet. Please wait for ' +\n          '`loadeddata` event on the <video> element.');\n    }\n  }\n  // If the current backend has 'FromPixels' registered, it has a more\n  // efficient way of handling pixel uploads, so we call that.\n  const kernel = getKernel(FromPixels, ENGINE.backendName);\n  if (kernel != null) {\n    const inputs: FromPixelsInputs = {pixels};\n    const attrs: FromPixelsAttrs = {numChannels};\n    return ENGINE.runKernel(\n               FromPixels, inputs as {} as NamedTensorMap,\n               attrs as {} as NamedAttrMap) as Tensor3D;\n  }\n\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n  let vals: Uint8ClampedArray|Uint8Array;\n\n  if (isCanvasLike) {\n    vals =\n        // tslint:disable-next-line:no-any\n        (pixels as any).getContext('2d').getImageData(0, 0, width, height).data;\n  } else if (isImageData || isPixelData) {\n    vals = (pixels as PixelData | ImageData).data;\n  } else if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement, 0, 0, width, height);\n    vals = fromPixels2DContext.getImageData(0, 0, width, height).data;\n  }\n  let values: Int32Array;\n  if (numChannels === 4) {\n    values = new Int32Array(vals);\n  } else {\n    const numPixels = width * height;\n    values = new Int32Array(numPixels * numChannels);\n    for (let i = 0; i < numPixels; i++) {\n      for (let channel = 0; channel < numChannels; ++channel) {\n        values[i * numChannels + channel] = vals[i * 4 + channel];\n      }\n    }\n  }\n  const outShape: [number, number, number] = [height, width, numChannels];\n  return tensor3d(values, outShape, 'int32');\n}\n\n/**\n * Draws a `tf.Tensor` of pixel values to a byte array or optionally a\n * canvas.\n *\n * When the dtype of the input is 'float32', we assume values in the range\n * [0-1]. Otherwise, when input is 'int32', we assume values in the range\n * [0-255].\n *\n * Returns a promise that resolves when the canvas has been drawn to.\n *\n * @param img A rank-2 or rank-3 tensor. If rank-2, draws grayscale. If\n *     rank-3, must have depth of 1, 3 or 4. When depth of 1, draws\n * grayscale. When depth of 3, we draw with the first three components of\n * the depth dimension corresponding to r, g, b and alpha = 1. When depth of\n * 4, all four components of the depth dimension correspond to r, g, b, a.\n * @param canvas The canvas to draw to.\n *\n * @doc {heading: 'Browser', namespace: 'browser'}\n */\nexport async function toPixels(\n    img: Tensor2D|Tensor3D|TensorLike,\n    canvas?: HTMLCanvasElement): Promise<Uint8ClampedArray> {\n  let $img = convertToTensor(img, 'img', 'toPixels');\n  if (!(img instanceof Tensor)) {\n    // Assume int32 if user passed a native array.\n    const originalImgTensor = $img;\n    $img = cast(originalImgTensor, 'int32');\n    originalImgTensor.dispose();\n  }\n  if ($img.rank !== 2 && $img.rank !== 3) {\n    throw new Error(\n        `toPixels only supports rank 2 or 3 tensors, got rank ${$img.rank}.`);\n  }\n  const [height, width] = $img.shape.slice(0, 2);\n  const depth = $img.rank === 2 ? 1 : $img.shape[2];\n\n  if (depth > 4 || depth === 2) {\n    throw new Error(\n        `toPixels only supports depth of size ` +\n        `1, 3 or 4 but got ${depth}`);\n  }\n\n  if ($img.dtype !== 'float32' && $img.dtype !== 'int32') {\n    throw new Error(\n        `Unsupported type for toPixels: ${$img.dtype}.` +\n        ` Please use float32 or int32 tensors.`);\n  }\n\n  const data = await $img.data();\n  const multiplier = $img.dtype === 'float32' ? 255 : 1;\n  const bytes = new Uint8ClampedArray(width * height * 4);\n\n  for (let i = 0; i < height * width; ++i) {\n    const rgba = [0, 0, 0, 255];\n\n    for (let d = 0; d < depth; d++) {\n      const value = data[i * depth + d];\n\n      if ($img.dtype === 'float32') {\n        if (value < 0 || value > 1) {\n          throw new Error(\n              `Tensor values for a float32 Tensor must be in the ` +\n              `range [0 - 1] but encountered ${value}.`);\n        }\n      } else if ($img.dtype === 'int32') {\n        if (value < 0 || value > 255) {\n          throw new Error(\n              `Tensor values for a int32 Tensor must be in the ` +\n              `range [0 - 255] but encountered ${value}.`);\n        }\n      }\n\n      if (depth === 1) {\n        rgba[0] = value * multiplier;\n        rgba[1] = value * multiplier;\n        rgba[2] = value * multiplier;\n      } else {\n        rgba[d] = value * multiplier;\n      }\n    }\n\n    const j = i * 4;\n    bytes[j + 0] = Math.round(rgba[0]);\n    bytes[j + 1] = Math.round(rgba[1]);\n    bytes[j + 2] = Math.round(rgba[2]);\n    bytes[j + 3] = Math.round(rgba[3]);\n  }\n\n  if (canvas != null) {\n    canvas.width = width;\n    canvas.height = height;\n    const ctx = canvas.getContext('2d');\n    const imageData = new ImageData(bytes, width, height);\n    ctx.putImageData(imageData, 0, 0);\n  }\n  if ($img !== img) {\n    $img.dispose();\n  }\n  return bytes;\n}\n\nexport const fromPixels = op({fromPixels_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\nimport {computeStrides} from '../util';\n\n/**\n * Validate gather nd inputs.\n *\n * @param tensor The tensor contains the source values.\n * @param indices The tensor contains the indices to slice the source.\n *\n * @returns [resultShape, numUpdates, sliceSize, strides]\n */\nexport function prepareAndValidate(\n    tensor: Tensor, indices: Tensor): [number[], number, number, number[]] {\n  if (tensor.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the input to be rank 1 or higher,' +\n        ` but the rank was ${tensor.rank}.`);\n  }\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.gatherND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(\n        'tf.gatherND() expects the indices to be int32 type,' +\n        ` but the dtype was ${indices.dtype}.`);\n  }\n  if (indices.shape[indices.rank - 1] > tensor.rank) {\n    throw new Error(\n        'index innermost dimension length must be <= tensor rank; saw: ' +\n        `${indices.shape[indices.rank - 1]} vs. ${tensor.rank}`);\n  }\n\n  if (tensor.size === 0) {\n    throw new Error(\n        'Requested more than 0 entries, but input is empty.' +\n        ` Input shape: ${tensor.shape}.`);\n  }\n\n  const indicesShape = indices.shape;\n  const sliceRank = indicesShape[indicesShape.length - 1];\n\n  // The result shape is\n  //   indices.shape[:-1] + params.shape[indices.shape[-1]:]\n  let nResult = 1;\n  for (let i = 0; i < indicesShape.length - 1; ++i) {\n    nResult *= indicesShape[i];\n  }\n\n  const inputShape = tensor.shape;\n\n  const resultShape = indicesShape.slice();\n  resultShape.pop();\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < tensor.rank; ++i) {\n    sliceSize *= inputShape[i];\n    resultShape.push(inputShape[i]);\n  }\n\n  const strides =\n      [...computeStrides(tensor.shape).map(stride => stride / sliceSize),\n       1].slice(0, sliceRank);\n\n  return [resultShape, nResult, sliceSize, strides];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {computeStrides, sizeFromShape} from '../util';\n\n/**\n * Check whether updates.shape = indices.shape[:batchDim] +\n * shape[sliceDim:]\n *\n * @param x The input tensor.\n */\nexport function validateUpdateShape(\n    shape: number[], indices: Tensor, updates: Tensor) {\n  const sliceDim = (indices.rank > 1) ? indices.shape[indices.rank - 1] : 1;\n  const batchDim = (indices.rank > 1) ? indices.rank - 1 : 1;\n\n  const shapeError = 'Must have updates.shape = indices.shape[:batchDim] + ' +\n      `shape[sliceDim:], got updates.shape: ${updates.shape}` +\n      `, indices.shape: ${indices.shape}, shape: ${shape}` +\n      `, sliceDim: ${sliceDim}, and batchDim: ${batchDim}.`;\n\n  if (updates.rank < batchDim) {\n    throw new Error(shapeError + ` update.rank < ${batchDim}. `);\n  }\n  if (shape.length < sliceDim + (updates.rank - batchDim)) {\n    throw new Error(\n        shapeError +\n        ` Output shape length < ${sliceDim + (updates.rank - batchDim)}`);\n  }\n  if (updates.rank !== batchDim + shape.length - sliceDim) {\n    throw new Error(\n        shapeError + ` update.rank != ${batchDim + shape.length - sliceDim}`);\n  }\n  for (let d = 0; d < batchDim; ++d) {\n    if (updates.shape[d] !== indices.shape[d]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d}] (${updates.shape[d]}) != indices.shape[${d}] (${\n              indices.shape[d]}).`);\n    }\n  }\n  for (let d = 0; d < updates.rank - batchDim; ++d) {\n    if (updates.shape[d + batchDim] !== shape[d + sliceDim]) {\n      throw new Error(\n          shapeError +\n          ` updates.shape[${d + batchDim}] (${\n              updates.shape[d + batchDim]}) != shape[${d + batchDim}] (${\n              shape[d + batchDim]})`);\n    }\n  }\n}\n\nexport interface ScatterShapeInfo {\n  sliceRank: number;\n  numUpdates: number;\n  sliceSize: number;\n  strides: number[];\n  outputSize: number;\n}\n/**\n * Validate scatter nd inputs.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n */\nexport function validateInput(\n    updates: Tensor, indices: Tensor, shape: number[]) {\n  if (indices.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the indices to be rank 1 or higher,' +\n        ` but the rank was ${indices.rank}.`);\n  }\n  if (updates.rank < 1) {\n    throw new Error(\n        'tf.scatterND() expects the updates to be rank 1 or higher,' +\n        ` but the rank was ${updates.rank}.`);\n  }\n  if (indices.dtype !== 'int32') {\n    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${\n        indices.dtype}`);\n  }\n  if (shape.length < 1) {\n    throw new Error(\n        `Output rank must be greater or equal to 1, but got shape: ${shape}`);\n  }\n\n  if (shape.length === 0) {\n    if (indices.size === 0) {\n      throw new Error(`Indices specified for empty output. indices shape: ${\n          indices.shape}`);\n    }\n    if (updates.size === 0) {\n      throw new Error(`Updates specified for empty output. updates shape: ${\n          updates.shape}`);\n    }\n  }\n\n  validateUpdateShape(shape, indices, updates);\n}\n\n/**\n * Calculate the shape information for the output.\n *\n * @param update The tensor contains the update values.\n * @param indices The tensor contains the indices for the update values.\n * @param shape The shape of the output tensor.\n *\n * @returns ScatterShapeInfo\n */\nexport function calculateShapes(\n    updates: TensorInfo, indices: TensorInfo,\n    shape: number[]): ScatterShapeInfo {\n  // Calculate the number of dimensions in indices\n  const indicesRank = indices.shape.length;\n  const sliceRank = (indicesRank > 1) ? indices.shape[indicesRank - 1] : 1;\n\n  // Calculate the number of elements that make up each slice of our updated\n  // tensor. This allows us to work with flattened tensors and copy over whole\n  // slices at a time.\n  const totalNd = shape.length;\n\n  let sliceSize = 1;\n  for (let i = sliceRank; i < totalNd; ++i) {\n    sliceSize *= shape[i];\n  }\n\n  const safeSliceDim = (sliceRank < 1) ? 1 : sliceRank;\n  const numUpdates = sizeFromShape(indices.shape) / safeSliceDim;\n\n  const strides = [...computeStrides(shape.slice(0, sliceRank)), 1];\n  const outputSize = sizeFromShape(shape);\n  return {sliceRank, numUpdates, sliceSize, strides, outputSize};\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '../kernel_registry';\nimport * as util from '../util';\n\nexport function assertParamsValid(\n    input: TensorInfo, begin: number[], size: number[]): void {\n  const inputRank = input.shape.length;\n  util.assert(\n      inputRank === begin.length,\n      () => `Error in slice${inputRank}D: Length of begin ${begin} must ` +\n          `match the rank of the array (${inputRank}).`);\n  util.assert(\n      inputRank === size.length,\n      () => `Error in slice${inputRank}D: Length of size ${size} must ` +\n          `match the rank of the array (${inputRank}).`);\n\n  for (let i = 0; i < inputRank; ++i) {\n    util.assert(\n        begin[i] + size[i] <= input.shape[i],\n        () => `Error in slice${inputRank}D: begin[${i}] + size[${i}] ` +\n            `(${begin[i] + size[i]}) would overflow input.shape[${i}] (${\n                  input.shape[i]})`);\n  }\n}\n\n/** Converts a binary mask to an array of axes. Used in stridedSlice(). */\nexport function maskToAxes(mask: number): number[] {\n  const axes = [];\n  let axis = 0;\n  while (mask > 0) {\n    if (mask & 1) {\n      axes.push(axis);\n    }\n    mask /= 2;\n    axis++;\n  }\n  return axes;\n}\n\n/** Computes the output shape given the strided slice params. */\nexport function computeOutShape(\n    begin: number[], end: number[], strides: number[]): number[] {\n  const size = [];\n  for (let axis = 0; axis < begin.length; axis++) {\n    size[axis] = Math.ceil((end[axis] - begin[axis]) / strides[axis]);\n  }\n  return size;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stride value. Otherwise, insert.\nexport function stridesWithElidedDims(\n    strides: number[], ellipsisInsertionIndex: number, numElidedAxes: number,\n    inputShape: number[]): number[] {\n  const newStrides = [...strides];\n  for (let i = newStrides.length; i < inputShape.length; i++) {\n    newStrides.push(1);\n  }\n  for (let i = 0; i < numElidedAxes; i++) {\n    if (i === 0) {\n      newStrides[ellipsisInsertionIndex] = 1;\n    } else {\n      newStrides.splice(\n          ellipsisInsertionIndex, 0 /* num elements to delete */,\n          1 /* element to add */);\n      newStrides.pop();\n    }\n  }\n  return newStrides;\n}\n\nfunction unnormalizeAxis(\n    ellipsisInsertionIndex: number, numElidedAxes: number,\n    normalizedAxis: number): number {\n  if (normalizedAxis <= ellipsisInsertionIndex) {\n    return normalizedAxis;\n  }\n\n  return normalizedAxis - (numElidedAxes - 1);\n}\n\nfunction getElidedAxes(numElidedAxes: number, ellipsisInsertionIndex: number) {\n  const elidedAxes = [];\n  for (let i = 0; i < numElidedAxes; i++) {\n    elidedAxes.push(ellipsisInsertionIndex + i);\n  }\n  return elidedAxes;\n}\n\n// Normalize the start, end and strides.\nexport function getNormalizedAxes(\n    inputShape: number[], ellipsisAxes: number[], numInterpolatedAxes: number,\n    begin: number[], end: number[], strides: number[], beginMask: number,\n    endMask: number,\n    ellipsisMask: number): {begin: number[], end: number[], strides: number[]} {\n  const inputRank = inputShape.length;\n  let normalizedBegin = new Array(inputRank),\n      normalizedEnd = new Array(inputRank),\n      normalizedStrides = new Array(inputRank);\n  if (ellipsisAxes.length && numInterpolatedAxes > 0) {\n    const fullIndex = ellipsisAxes[0];\n\n    // The ellipsis applies to the masked index as well as any dimensions\n    // that are interpolated.\n    const numElidedAxes = numInterpolatedAxes + 1;\n    normalizedBegin = startIndicesWithElidedDims(\n        beginMask, fullIndex, numElidedAxes, begin, inputShape);\n    normalizedEnd = stopIndicesWithElidedDims(\n        endMask, fullIndex, numElidedAxes, end, inputShape);\n    normalizedStrides =\n        stridesWithElidedDims(strides, fullIndex, numElidedAxes, inputShape);\n  } else {\n    for (let axis = 0; axis < inputRank; axis++) {\n      normalizedBegin[axis] = startForAxis(\n          beginMask, begin, strides, inputShape, axis, ellipsisMask);\n      normalizedEnd[axis] =\n          stopForAxis(endMask, end, strides, inputShape, axis, ellipsisMask);\n      normalizedStrides[axis] = stridesForAxis(strides, axis, ellipsisMask);\n    }\n  }\n\n  return {\n    begin: normalizedBegin,\n    end: normalizedEnd,\n    strides: normalizedStrides\n  };\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current start value. Otherwise, insert.\nexport function startIndicesWithElidedDims(\n    beginMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalBegin: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = 0;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalBegin[originalAxis];\n      if (beginMask & 1 << originalAxis) {\n        originalValue = 0;\n      }\n\n      newIndices[axis] = originalValue;\n    }\n  }\n  return newIndices;\n}\n\n// Creates full selection at the elided dimensions. If the dimension matches\n// the ellipsis mask, override the current stop value. Otherwise, insert.\nexport function stopIndicesWithElidedDims(\n    endMask: number, ellipsisInsertionIndex: number, numElidedAxes: number,\n    originalEnd: number[], inputShape: number[]): number[] {\n  const newIndices = [...inputShape];\n  const elidedAxes = getElidedAxes(numElidedAxes, ellipsisInsertionIndex);\n\n  for (let axis = 0; axis < newIndices.length; axis++) {\n    if (elidedAxes.indexOf(axis) > -1) {\n      newIndices[axis] = Number.MAX_SAFE_INTEGER;\n    } else {\n      const originalAxis =\n          unnormalizeAxis(ellipsisInsertionIndex, numElidedAxes, axis);\n      let originalValue = originalEnd[originalAxis];\n      if (endMask & 1 << originalAxis) {\n        originalValue = Number.MAX_SAFE_INTEGER;\n      }\n      newIndices[axis] = originalValue;\n    }\n  }\n\n  for (let i = 0; i < newIndices.length; i++) {\n    // Handle negative indices\n    const axisSize = inputShape[i];\n    if (newIndices[i] < 0) {\n      newIndices[i] += axisSize;\n    }\n    newIndices[i] = util.clamp(0, newIndices[i], inputShape[i]);\n  }\n  return newIndices;\n}\n\nexport function stridesForAxis(\n    strides: number[], axis: number, ellipsisMask: number): number {\n  let stride = strides[axis];\n  if (ellipsisMask & (1 << axis) || stride == null) {\n    stride = 1;\n  }\n\n  return stride;\n}\n\nexport function startForAxis(\n    beginMask: number, startIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let start = startIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or the begin index is not set\n  // for the axis.\n  if (beginMask & 1 << axis || ellipsisMask & 1 << axis || start == null) {\n    if (stride > 0) {\n      // Forward iteration - use the first element. These values will get\n      // clamped below (Note: We could have set them to 0 and axis_size-1, but\n      // use lowest() and max() to maintain symmetry with StopForAxis())\n      start = Number.MIN_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the last element.\n      start = Number.MAX_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (start < 0) {\n    start += axisSize;\n  }\n\n  // Clamping\n  start = util.clamp(0, start, axisSize - 1);\n\n  return start;\n}\n\nexport function stopForAxis(\n    endMask: number, stopIndices: number[], strides: number[],\n    inputShape: number[], axis: number, ellipsisMask: number): number {\n  // Begin with the specified index\n  let stop = stopIndices[axis];\n  const stride = strides[axis] || 1;\n\n  // Check the axis bit from right of masked axes, or if the stop index is not\n  // set for this axis.\n  if (endMask & (1 << axis) || ellipsisMask & (1 << axis) || stop == null) {\n    if (stride > 0) {\n      // Forward iteration - use the last element. These values will get\n      // clamped below\n      stop = Number.MAX_SAFE_INTEGER;\n    } else {\n      // Backward iteration - use the first element.\n      stop = Number.MIN_SAFE_INTEGER;\n    }\n  }\n\n  // Handle negative indices\n  const axisSize = inputShape[axis];\n  if (stop < 0) {\n    stop += axisSize;\n  }\n\n  // Clamping\n  // Because the end index points one past the last element, we need slightly\n  // different clamping ranges depending on the direction.\n  if (stride > 0) {\n    // Forward iteration\n    stop = util.clamp(0, stop, axisSize);\n  } else {\n    // Backward iteration\n    stop = util.clamp(-1, stop, axisSize - 1);\n  }\n\n  return stop;\n}\n\n/**\n * Returns true if the slice occupies a continous set of elements in the\n * 'flat' space.\n */\nexport function isSliceContinous(\n    shape: number[], begin: number[], size: number[]) {\n  // Index of the first axis that has size > 1.\n  let firstNonOneAxis = size.length;\n  for (let i = 0; i < size.length; i++) {\n    if (size[i] > 1) {\n      firstNonOneAxis = i;\n      break;\n    }\n  }\n\n  for (let i = firstNonOneAxis + 1; i < size.length; i++) {\n    if (begin[i] > 0 || size[i] !== shape[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function computeFlatOffset(begin: number[], strides: number[]): number {\n  let flatOffset = begin.length > 0 ? begin[begin.length - 1] : 1;\n  for (let i = 0; i < begin.length - 1; i++) {\n    flatOffset += begin[i] * strides[i];\n  }\n  return flatOffset;\n}\n\nexport function parseSliceParams(\n    x: TensorInfo, begin: number|number[], size?: number|number[]) {\n  // The following logic allows for more ergonomic calls.\n  let begin_: number[];\n  const xRank = x.shape.length;\n  if (typeof begin === 'number') {\n    begin_ = [begin, ...new Array(xRank - 1).fill(0)];\n  } else if (begin.length < xRank) {\n    begin_ = begin.concat(new Array(xRank - begin.length).fill(0));\n  } else {\n    begin_ = begin.slice();\n  }\n  begin_.forEach(d => {\n    util.assert(\n        d !== -1, () => 'slice() does not support negative begin indexing.');\n  });\n  let size_: number[];\n  if (size == null) {\n    size_ = new Array(xRank).fill(-1);\n  } else if (typeof size === 'number') {\n    size_ = [size, ...new Array(xRank - 1).fill(-1)];\n  } else if (size.length < xRank) {\n    size_ = size.concat(new Array(xRank - size.length).fill(-1));\n  } else {\n    size_ = size;\n  }\n  size_ = size_.map((d, i) => {\n    if (d >= 0) {\n      return d;\n    } else {\n      util.assert(\n          d === -1,\n          () => `Negative size values should be exactly -1 but got ` +\n              `${d} for the slice() size at index ${i}.`);\n      return x.shape[i] - begin_[i];\n    }\n  });\n  return [begin_, size_];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from './util';\n\n/**\n * Types to support JSON-esque data structures internally.\n *\n * Internally ConfigDict's use camelCase keys and values where the\n * values are class names to be instantiated.  On the python side, these\n * will be snake_case.  Internally we allow Enums into the values for better\n * type safety, but these need to be converted to raw primitives (usually\n * strings) for round-tripping with python.\n *\n * toConfig returns the TS-friendly representation. model.toJSON() returns\n * the pythonic version as that's the portable format.  If you need to\n * python-ify a non-model level toConfig output, you'll need to use a\n * convertTsToPythonic from serialization_utils in -Layers.\n *\n */\nexport declare type ConfigDictValue =\n    boolean | number | string | null | ConfigDictArray | ConfigDict;\nexport declare interface ConfigDict {\n  [key: string]: ConfigDictValue;\n}\nexport declare interface ConfigDictArray extends Array<ConfigDictValue> {}\n\n/**\n * Type to represent the class-type of Serializable objects.\n *\n * Ie the class prototype with access to the constructor and any\n * static members/methods. Instance methods are not listed here.\n *\n * Source for this idea: https://stackoverflow.com/a/43607255\n */\nexport declare type SerializableConstructor<T extends Serializable> = {\n  // tslint:disable-next-line:no-any\n  new (...args: any[]): T; className: string; fromConfig: FromConfigMethod<T>;\n};\nexport declare type FromConfigMethod<T extends Serializable> =\n    (cls: SerializableConstructor<T>, config: ConfigDict) => T;\n\n/**\n * Serializable defines the serialization contract.\n *\n * TFJS requires serializable classes to return their className when asked\n * to avoid issues with minification.\n */\nexport abstract class Serializable {\n  /**\n   * Return the class name for this class to use in serialization contexts.\n   *\n   * Generally speaking this will be the same thing that constructor.name\n   * would have returned.  However, the class name needs to be robust\n   * against minification for serialization/deserialization to work properly.\n   *\n   * There's also places such as initializers.VarianceScaling, where\n   * implementation details between different languages led to different\n   * class hierarchies and a non-leaf node is used for serialization purposes.\n   */\n  getClassName(): string {\n    return (this.constructor as SerializableConstructor<Serializable>)\n        .className;\n  }\n\n  /**\n   * Return all the non-weight state needed to serialize this object.\n   */\n  abstract getConfig(): ConfigDict;\n\n  /**\n   * Creates an instance of T from a ConfigDict.\n   *\n   * This works for most descendants of serializable.  A few need to\n   * provide special handling.\n   * @param cls A Constructor for the class to instantiate.\n   * @param config The Configuration for the object.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config);\n  }\n}\n\n/**\n * Maps string keys to class constructors.\n *\n * Used during (de)serialization from the cross-language JSON format, which\n * requires the class name in the serialization format matches the class\n * names as used in Python, should it exist.\n */\nexport class SerializationMap {\n  private static instance: SerializationMap;\n  classNameMap: {\n    [className: string]:\n        [SerializableConstructor<Serializable>, FromConfigMethod<Serializable>]\n  };\n\n  private constructor() {\n    this.classNameMap = {};\n  }\n\n  /**\n   * Returns the singleton instance of the map.\n   */\n  static getMap(): SerializationMap {\n    if (SerializationMap.instance == null) {\n      SerializationMap.instance = new SerializationMap();\n    }\n    return SerializationMap.instance;\n  }\n\n  /**\n   * Registers the class as serializable.\n   */\n  static register<T extends Serializable>(cls: SerializableConstructor<T>) {\n    SerializationMap.getMap().classNameMap[cls.className] =\n        [cls, cls.fromConfig];\n  }\n}\n\n/**\n * Register a class with the serialization map of TensorFlow.js.\n *\n * This is often used for registering custom Layers, so they can be\n * serialized and deserialized.\n *\n * Example:\n *\n * ```js\n * class MyCustomLayer extends tf.layers.Layer {\n *   static className = 'MyCustomLayer';\n *\n *   constructor(config) {\n *     super(config);\n *   }\n * }\n * tf.serialization.registerClass(MyCustomLayer);\n * ```\n *\n * @param cls The class to be registered. It must have a public static member\n *   called `className` defined and the value must be a non-empty string.\n *\n * @doc {heading: 'Models', subheading: 'Serialization', ignoreCI: true}\n */\nexport function registerClass<T extends Serializable>(\n    cls: SerializableConstructor<T>) {\n  assert(\n      cls.className != null,\n      () => `Class being registered does not have the static className ` +\n          `property defined.`);\n  assert(\n      typeof cls.className === 'string',\n      () => `className is required to be a string, but got type ` +\n          typeof cls.className);\n  assert(\n      cls.className.length > 0,\n      () => `Class being registered has an empty-string as its className, ` +\n          `which is disallowed.`);\n\n  SerializationMap.register(cls);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {inferShape} from './tensor_util_env';\nimport {RecursiveArray, TensorLike, TypedArray} from './types';\nimport {arraysEqual, flatten, isString, isTypedArray} from './util';\n\nconst TEST_EPSILON_FLOAT32 = 1e-3;\nexport const TEST_EPSILON_FLOAT16 = 1e-1;\n\nexport function expectArraysClose(\n    actual: TypedArray|number|RecursiveArray<number>,\n    expected: TypedArray|number|RecursiveArray<number>, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, epsilon));\n}\n\nexport function testEpsilon() {\n  return ENGINE.backend.floatPrecision() === 32 ? TEST_EPSILON_FLOAT32 :\n                                                  TEST_EPSILON_FLOAT16;\n}\n\nfunction expectArraysPredicate(\n    actual: TensorLike, expected: TensorLike,\n    predicate: (a: {}, b: {}) => boolean) {\n  let checkClassType = true;\n  if (isTypedArray(actual) || isTypedArray(expected)) {\n    checkClassType = false;\n  }\n  if (isTypedArray(actual) && isTypedArray(expected)) {\n    checkClassType = true;\n  }\n  if (checkClassType) {\n    const aType = actual.constructor.name;\n    const bType = expected.constructor.name;\n\n    if (aType !== bType) {\n      throw new Error(\n          `Arrays are of different type. Actual: ${aType}. ` +\n          `Expected: ${bType}`);\n    }\n  }\n\n  if (Array.isArray(actual) && Array.isArray(expected)) {\n    const actualShape = inferShape(actual);\n    const expectedShape = inferShape(expected);\n    if (!arraysEqual(actualShape, expectedShape)) {\n      throw new Error(\n          `Arrays have different shapes. ` +\n          `Actual: [${actualShape}]. Expected: [${expectedShape}]`);\n    }\n  }\n\n  const actualFlat =\n      isTypedArray(actual) ? actual : flatten(actual as RecursiveArray<number>);\n  const expectedFlat = isTypedArray(expected) ?\n      expected :\n      flatten(expected as RecursiveArray<number>);\n\n  if (actualFlat.length !== expectedFlat.length) {\n    throw new Error(\n        `Arrays have different lengths actual: ${actualFlat.length} vs ` +\n        `expected: ${expectedFlat.length}.\\n` +\n        `Actual:   ${actualFlat}.\\n` +\n        `Expected: ${expectedFlat}.`);\n  }\n  for (let i = 0; i < expectedFlat.length; ++i) {\n    const a = actualFlat[i];\n    const e = expectedFlat[i];\n\n    if (!predicate(a, e)) {\n      throw new Error(\n          `Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${e}.\\n` +\n          `Actual:   ${actualFlat}.\\n` +\n          `Expected: ${expectedFlat}.`);\n    }\n  }\n}\n\nexport interface DoneFn {\n  (): void;\n  fail: (message?: Error|string) => void;\n}\n\nexport function expectPromiseToFail(fn: () => Promise<{}>, done: DoneFn): void {\n  fn().then(() => done.fail(), () => done());\n}\n\nexport function expectArraysEqual(actual: TensorLike, expected: TensorLike) {\n  const exp = typeof expected === 'string' || typeof expected === 'number' ||\n          typeof expected === 'boolean' ?\n      [expected] as number[] :\n      expected as number[];\n  if (isString(actual) || isString((actual as string[])[0]) ||\n      isString(expected) || isString((expected as string[])[0])) {\n    // tslint:disable-next-line: triple-equals\n    return expectArraysPredicate(actual, exp, (a, b) => a == b);\n  }\n  return expectArraysPredicate(\n      actual, expected, (a, b) => areClose(a as number, b as number, 0));\n}\n\nexport function expectNumbersClose(a: number, e: number, epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  if (!areClose(a, e, epsilon)) {\n    throw new Error(`Numbers differ: actual === ${a}, expected === ${e}`);\n  }\n}\n\nfunction areClose(a: number, e: number, epsilon: number): boolean {\n  if (!isFinite(a) && !isFinite(e)) {\n    return true;\n  }\n  if (isNaN(a) || isNaN(e) || Math.abs(a - e) > epsilon) {\n    return false;\n  }\n  return true;\n}\n\nexport function expectValuesInRange(\n    actual: TypedArray|number[], low: number, high: number) {\n  for (let i = 0; i < actual.length; i++) {\n    if (actual[i] < low || actual[i] > high) {\n      throw new Error(\n          `Value out of range:${actual[i]} low: ${low}, high: ${high}`);\n    }\n  }\n}\n\nexport function expectArrayBuffersEqual(\n    actual: ArrayBuffer, expected: ArrayBuffer) {\n  // Safari & Jasmine don't like comparing ArrayBuffers directly. Wrapping in\n  // a Float32Array solves this issue.\n  expect(new Float32Array(actual)).toEqual(new Float32Array(expected));\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from './backends/backend';\nimport {ENGINE, Engine, MemoryInfo, ProfileInfo, ScopeFn, TimingInfo} from './engine';\nimport {env} from './environment';\n\nimport {Platform} from './platforms/platform';\nimport {setDeprecationWarningFn, Tensor} from './tensor';\nimport {TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\n\n/**\n * Enables production mode which disables correctness checks in favor of\n * performance.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableProdMode(): void {\n  env().set('PROD', true);\n}\n\n/**\n * Enables debug mode which will log information about all executed kernels:\n * the elapsed time of the kernel execution, as well as the rank, shape, and\n * size of the output tensor.\n *\n * Debug mode will significantly slow down your application as it will\n * download the result of every operation to the CPU. This should not be used in\n * production. Debug mode does not affect the timing information of the kernel\n * execution as we do not measure download time in the kernel execution time.\n *\n * See also: `tf.profile`, `tf.memory`.\n *\n * @doc {heading: 'Environment'}\n */\nexport function enableDebugMode(): void {\n  env().set('DEBUG', true);\n}\n\n/** Globally disables deprecation warnings */\nexport function disableDeprecationWarnings(): void {\n  env().set('DEPRECATION_WARNINGS_ENABLED', false);\n  console.warn(`TensorFlow.js deprecation warnings have been disabled.`);\n}\n\n/** Warn users about deprecated functionality. */\nexport function deprecationWarn(msg: string) {\n  if (env().getBool('DEPRECATION_WARNINGS_ENABLED')) {\n    console.warn(\n        msg + ' You can disable deprecation warnings with ' +\n        'tf.disableDeprecationWarnings().');\n  }\n}\nsetDeprecationWarningFn(deprecationWarn);\n\n/**\n * Dispose all variables kept in backend engine.\n *\n * @doc {heading: 'Environment'}\n */\nexport function disposeVariables(): void {\n  ENGINE.disposeVariables();\n}\n\n/**\n * It returns the global engine that keeps track of all tensors and backends.\n *\n * @doc {heading: 'Environment'}\n */\nexport function engine(): Engine {\n  return ENGINE;\n}\n\n/**\n * Returns memory info at the current time in the program. The result is an\n * object with the following properties:\n *\n * - `numBytes`: Number of bytes allocated (undisposed) at this time.\n * - `numTensors`: Number of unique tensors allocated.\n * - `numDataBuffers`: Number of unique data buffers allocated\n *   (undisposed) at this time, which is \u2264 the number of tensors\n *   (e.g. `a.reshape(newShape)` makes a new Tensor that shares the same\n *   data buffer with `a`).\n * - `unreliable`: True if the memory usage is unreliable. See `reasons` when\n *    `unreliable` is true.\n * - `reasons`: `string[]`, reasons why the memory is unreliable, present if\n *    `unreliable` is true.\n *\n * WebGL Properties:\n * - `numBytesInGPU`: Number of bytes allocated (undisposed) in the GPU only at\n *     this time.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function memory(): MemoryInfo {\n  return ENGINE.memory();\n}\n\n/**\n * Executes the provided function `f()` and returns a promise that resolves\n * with information about the function's memory use:\n * - `newBytes`: the number of new bytes allocated\n * - `newTensors`: the number of new tensors created\n * - `peakBytes`: the peak number of bytes allocated\n * - `kernels`: an array of objects for each kernel involved that reports\n * their input and output shapes, number of bytes used, and number of new\n * tensors created.\n *\n * ```js\n * const profile = await tf.profile(() => {\n *   const x = tf.tensor1d([1, 2, 3]);\n *   let x2 = x.square();\n *   x2.dispose();\n *   x2 = x.square();\n *   x2.dispose();\n *   return x;\n * });\n *\n * console.log(`newBytes: ${profile.newBytes}`);\n * console.log(`newTensors: ${profile.newTensors}`);\n * console.log(`byte usage over all kernels: ${profile.kernels.map(k =>\n * k.totalBytesSnapshot)}`);\n * ```\n *\n *\n * @doc {heading: 'Performance', subheading: 'Profile'}\n */\nexport function profile(f: () => (TensorContainer | Promise<TensorContainer>)):\n    Promise<ProfileInfo> {\n  return ENGINE.profile(f);\n}\n\n/**\n * Executes the provided function `fn` and after it is executed, cleans up all\n * intermediate tensors allocated by `fn` except those returned by `fn`.\n * `fn` must not return a Promise (async functions not allowed). The returned\n * result can be a complex object.\n *\n * Using this method helps avoid memory leaks. In general, wrap calls to\n * operations in `tf.tidy` for automatic memory cleanup.\n *\n * NOTE: Variables do *not* get cleaned up when inside a tidy(). If you want to\n * dispose variables, please use `tf.disposeVariables` or call dispose()\n * directly on variables.\n *\n * ```js\n * // y = 2 ^ 2 + 1\n * const y = tf.tidy(() => {\n *   // a, b, and one will be cleaned up when the tidy ends.\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *   const b = a.square();\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * y.print();\n * ```\n *\n * @param nameOrFn The name of the closure, or the function to execute.\n *     If a name is provided, the 2nd argument should be the function.\n *     If debug mode is on, the timing and the memory usage of the function\n *     will be tracked and displayed on the console using the provided name.\n * @param fn The function to execute.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function tidy<T extends TensorContainer>(\n    nameOrFn: string|ScopeFn<T>, fn?: ScopeFn<T>): T {\n  return ENGINE.tidy(nameOrFn, fn);\n}\n\n/**\n * Disposes any `tf.Tensor`s found within the provided object.\n *\n * @param container an object that may be a `tf.Tensor` or may directly\n *     contain `tf.Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. If\n *     the object is not a `tf.Tensor` or does not contain `Tensors`, nothing\n *     happens. In general it is safe to pass any object here, except that\n *     `Promise`s are not supported.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function dispose(container: TensorContainer) {\n  const tensors = getTensorsInContainer(container);\n  tensors.forEach(tensor => tensor.dispose());\n}\n\n/**\n * Keeps a `tf.Tensor` generated inside a `tf.tidy` from being disposed\n * automatically.\n *\n * ```js\n * let b;\n * const y = tf.tidy(() => {\n *   const one = tf.scalar(1);\n *   const a = tf.scalar(2);\n *\n *   // b will not be cleaned up by the tidy. a and one will be cleaned up\n *   // when the tidy ends.\n *   b = tf.keep(a.square());\n *\n *   console.log('numTensors (in tidy): ' + tf.memory().numTensors);\n *\n *   // The value returned inside the tidy function will return\n *   // through the tidy, in this case to the variable y.\n *   return b.add(one);\n * });\n *\n * console.log('numTensors (outside tidy): ' + tf.memory().numTensors);\n * console.log('y:');\n * y.print();\n * console.log('b:');\n * b.print();\n * ```\n *\n * @param result The tensor to keep from being disposed.\n *\n * @doc {heading: 'Performance', subheading: 'Memory'}\n */\nexport function keep<T extends Tensor>(result: T): T {\n  return ENGINE.keep(result);\n}\n\n/**\n * Executes `f()` and returns a promise that resolves with timing\n * information.\n *\n * The result is an object with the following properties:\n *\n * - `wallMs`: Wall execution time.\n * - `kernelMs`: Kernel execution time, ignoring data transfer. If using the\n * WebGL backend and the query timer extension is not available, this will\n * return an error object.\n * - On `WebGL` The following additional properties exist:\n *   - `uploadWaitMs`: CPU blocking time on texture uploads.\n *   - `downloadWaitMs`: CPU blocking time on texture downloads (readPixels).\n *\n * ```js\n * const x = tf.randomNormal([20, 20]);\n * const time = await tf.time(() => x.matMul(x));\n *\n * console.log(`kernelMs: ${time.kernelMs}, wallTimeMs: ${time.wallMs}`);\n * ```\n *\n * @param f The function to execute and time.\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nexport function time(f: () => void): Promise<TimingInfo> {\n  return ENGINE.time(f);\n}\n\n/**\n * Sets the backend (cpu, webgl, wasm, etc) responsible for creating tensors and\n * executing operations on those tensors. Returns a promise that resolves\n * to a boolean if the backend initialization was successful.\n *\n * Note this disposes the current backend, if any, as well as any tensors\n * associated with it. A new backend is initialized, even if it is of the\n * same type as the previous one.\n *\n * @param backendName The name of the backend. Currently supports\n *     `'webgl'|'cpu'` in the browser, `'tensorflow'` under node.js\n *     (requires tfjs-node), and `'wasm'` (requires tfjs-backend-wasm).\n *\n * @doc {heading: 'Backends'}\n */\nexport function setBackend(backendName: string): Promise<boolean> {\n  return ENGINE.setBackend(backendName);\n}\n\n/**\n * Returns a promise that resolves when the currently selected backend (or the\n * highest priority one) has initialized. Await this promise when you are using\n * a backend that has async initialization.\n *\n * @doc {heading: 'Backends'}\n */\nexport function ready(): Promise<void> {\n  return ENGINE.ready();\n}\n\n/**\n * Returns the current backend name (cpu, webgl, etc). The backend is\n * responsible for creating tensors and executing operations on those tensors.\n *\n * @doc {heading: 'Backends'}\n */\nexport function getBackend(): string {\n  return ENGINE.backendName;\n}\n\n/**\n * Removes a backend and the registered factory.\n *\n * @doc {heading: 'Backends'}\n */\nexport function removeBackend(name: string): void {\n  ENGINE.removeBackend(name);\n}\n\n/**\n * Finds the backend registered under the provided name. Returns null if the\n * name is not in the registry, or the registration hasn't finished yet.\n */\nexport function findBackend(name: string): KernelBackend {\n  return ENGINE.findBackend(name);\n}\n\n/**\n * Finds the backend factory registered under the provided name. Returns a\n * function that produces a new backend when called. Returns null if the name\n * is not in the registry.\n */\nexport function findBackendFactory(name: string): () =>\n    KernelBackend | Promise<KernelBackend> {\n  return ENGINE.findBackendFactory(name);\n}\n\n/**\n * Registers a global backend. The registration should happen when importing\n * a module file (e.g. when importing `backend_webgl.ts`), and is used for\n * modular builds (e.g. custom tfjs bundle with only webgl support).\n *\n * @param factory The backend factory function. When called, it should\n * return a backend instance, or a promise of an instance.\n * @param priority The priority of the backend (higher = more important).\n *     In case multiple backends are registered, the priority is used to find\n *     the best backend. Defaults to 1.\n * @return False if there is already a registered backend under this name, true\n *     if not.\n *\n * @doc {heading: 'Backends'}\n */\nexport function registerBackend(\n    name: string, factory: () => KernelBackend | Promise<KernelBackend>,\n    priority = 1): boolean {\n  return ENGINE.registerBackend(name, factory, priority);\n}\n\n/**\n * Gets the current backend. If no backends have been initialized, this will\n * attempt to initialize the best backend. Will throw an error if the highest\n * priority backend has async initialization, in which case, you should call\n * 'await tf.ready()' before running other code.\n *\n * @doc {heading: 'Backends'}\n */\nexport function backend(): KernelBackend {\n  return ENGINE.backend;\n}\n\n/**\n * Sets the global platform.\n *\n * @param platformName The name of this platform.\n * @param platform A platform implementation.\n */\nexport function setPlatform(platformName: string, platform: Platform) {\n  env().setPlatform(platformName, platform);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Add, AddInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n *\n * ```js\n * // Broadcast add a with b.\n * const a = tf.scalar(5);\n * const b = tf.tensor1d([10, 20, 30, 40]);\n *\n * a.add(b).print();  // or tf.add(a, b)\n * ```\n * @param a The first `tf.Tensor` to add.\n * @param b The second `tf.Tensor` to add. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction add_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'add');\n  let $b = convertToTensor(b, 'b', 'add');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.add($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: AddInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Add) as T;\n}\n\nexport const add = op({add_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FloorDiv, FloorDivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n * The result is rounded with floor function.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.floorDiv(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.floorDiv(b).print();  // or tf.floorDiv(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction floorDiv_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'floorDiv');\n  let $b = convertToTensor(b, 'b', 'floorDiv');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.floorDiv($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: FloorDivInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             FloorDiv) as T;\n}\n\nexport const floorDiv = op({floorDiv_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Div, DivInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {floorDiv} from './floorDiv';\nimport {op} from './operation';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n *\n * a.div(b).print();  // or tf.div(a, b)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction div_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'int32' && $b.dtype === 'int32') {\n    return floorDiv($a, $b);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.realDivide($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: DivInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Div,\n             attrs) as T;\n}\n\nexport const div = op({div_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Multiply, MultiplyInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Multiplies two `tf.Tensor`s element-wise, A * B. Supports broadcasting.\n *\n * We also expose `tf.mulStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.tensor1d([2, 3, 4, 5]);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n *\n * ```js\n * // Broadcast mul a with b.\n * const a = tf.tensor1d([1, 2, 3, 4]);\n * const b = tf.scalar(5);\n *\n * a.mul(b).print();  // or tf.mul(a, b)\n * ```\n * @param a The first tensor to multiply.\n * @param b The second tensor to multiply. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mul_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mul');\n  let $b = convertToTensor(b, 'b', 'mul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.multiply($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MultiplyInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Multiply) as T;\n}\nexport const mul = op({mul_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Abs, AbsInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes absolute value element-wise: `abs(x)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.abs().print();  // or tf.abs(x)\n * ```\n * @param x The input `tf.Tensor`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction abs_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'abs');\n\n  const inputs: AbsInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    if ($x.dtype === 'complex64') {\n      return backend.complexAbs($x);\n    }\n\n    return backend.abs($x);\n  }, inputs as {} as NamedTensorMap, null /* grad */, Abs);\n}\n\nexport const abs = op({abs_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Acos, AcosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes acos of the input `tf.Tensor` element-wise: `acos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.acos().print();  // or tf.acos(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acos');\n  const inputs: AcosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acos);\n}\nexport const acos = op({acos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Acosh, AcoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the inverse hyperbolic cos of the input `tf.Tensor` element-wise:\n * `acosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([10, 1, 3, 5.7]);\n *\n * x.acosh().print();  // or tf.acosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction acosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'acosh');\n  const inputs: AcoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.acosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Acosh);\n}\nexport const acosh = op({acosh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AddN, AddNInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Adds a list of `tf.Tensor`s element-wise, each with the same shape and dtype.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n *\n * tf.addN([a, b, c]).print();\n * ```\n * @param tensors A list of tensors with the same shape and dtype.\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction addN_<T extends Tensor>(tensors: Array<T|TensorLike>): T {\n  util.assert(\n      Array.isArray(tensors),\n      () => 'The argument passed to tf.addN() must be a list of tensors');\n  util.assert(\n      tensors.length >= 1,\n      () => `Must pass at least one tensor to tf.addN(), but got ` +\n          `${tensors.length}`);\n\n  const $tensors =\n      tensors.map((t, i) => convertToTensor(t, `tensors${i}`, 'addN'));\n\n  const firstTensor = $tensors[0];\n  $tensors.forEach(t => {\n    if (t.dtype !== firstTensor.dtype) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same dtype');\n    }\n  });\n\n  $tensors.forEach(t => {\n    if (!util.arraysEqual(t.shape, firstTensor.shape)) {\n      throw new Error(\n          'All tensors passed to tf.addN() must have the same shape');\n    }\n  });\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.addN($tensors);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: AddNInputs = $tensors;\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, AddN) as\n      T;\n}\n\nexport const addN = op({addN_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\n/**\n * Returns true if the axis specifies the inner most dimensions of the\n * array.\n */\nexport function axesAreInnerMostDims(axes: number[], rank: number): boolean {\n  for (let i = 0; i < axes.length; ++i) {\n    if (axes[axes.length - i - 1] !== rank - 1 - i) {\n      return false;\n    }\n  }\n  return true;\n}\n\nexport function combineLocations(\n    outputLoc: number[], reduceLoc: number[], axes: number[]): number[] {\n  const rank = outputLoc.length + reduceLoc.length;\n  const loc = [];\n  let outIdx = 0;\n  let reduceIdx = 0;\n  \u00A0 for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      loc.push(outputLoc[outIdx++]);\n    } else {\n      loc.push(reduceLoc[reduceIdx++]);\n    }\n  }\n  return loc;\n}\n\nexport function computeOutAndReduceShapes(\n    aShape: number[], axes: number[]): [number[], number[]] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (axes.indexOf(dim) === -1) {\n      outShape.push(aShape[dim]);\n    }\n  }\n  const reduceShape = axes.map(dim => aShape[dim]);\n  return [outShape, reduceShape];\n}\n\nexport function expandShapeToKeepDim(\n    shape: number[], axes: number[]): number[] {\n  const reduceSubShape = axes.map(x => 1);\n  return combineLocations(shape, reduceSubShape, axes);\n}\n\nexport function assertAxesAreInnerMostDims(\n    msg: string, axes: number[], rank: number): void {\n  util.assert(\n      axesAreInnerMostDims(axes, rank),\n      () => `${msg} supports only inner-most axes for now. ` +\n          `Got axes ${axes} and rank-${rank} input.`);\n}\n\n/**\n * Returns the axes permutation to be used with `tf.transpose`, if such\n * permutation is necessary. Otherwise it returns null. This method is used by\n * operations that operate only on inner-most axes.\n */\nexport function getAxesPermutation(axes: number[], rank: number): number[]|\n    null {\n  if (axesAreInnerMostDims(axes, rank)) {\n    return null;\n  }\n  const result: number[] = [];\n  for (let i = 0; i < rank; ++i) {\n    if (axes.indexOf(i) === -1) {\n      result.push(i);\n    }\n  }\n  axes.forEach(axis => result.push(axis));\n  return result;\n}\n\n/** Returns the axes permutation that undoes the original permutation. */\nexport function getUndoAxesPermutation(axes: number[]): number[] {\n  return axes.map((axis, i) => [i, axis])\n      .sort((a, b) => a[1] - b[1])\n      .map(x => x[0]);\n}\n\nexport function getInnerMostAxes(numAxes: number, rank: number): number[] {\n  const res: number[] = [];\n  for (let i = rank - numAxes; i < rank; ++i) {\n    res.push(i);\n  }\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {All, AllAttrs, AllInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical and of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.all().print();  // or tf.all(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.all(axis).print();  // or tf.all(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction all_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'all', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.all($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AllInputs = {x: $x};\n  const attrs: AllAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, All,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const all = op({all_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Any, AnyAttrs, AnyInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the logical or of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 1, 1], 'bool');\n *\n * x.any().print();  // or tf.any(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 1, 0, 0], [2, 2], 'bool');\n *\n * const axis = 1;\n * x.any(axis).print();  // or tf.any(x, axis)\n * ```\n *\n * @param x The input tensor. Must be of dtype bool.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction any_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'any', 'bool');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const origAxes = parseAxisParam(axis, $x.shape);\n    let axes = origAxes;\n    const permutedAxes = getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = getInnerMostAxes(axes.length, $x.rank);\n    }\n    const res = backend.any($x, axes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(res.shape, origAxes);\n      return reshape(res, newShape);\n    }\n    return res as T;\n  };\n\n  const inputs: AnyInputs = {x: $x};\n  const attrs: AnyAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Any,\n             attrs as {} as NamedAttrMap) as T;\n}\n\n// tslint:disable-next-line:variable-name\nexport const any = op({any_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMax, ArgMaxAttrs, ArgMaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the maximum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMax().print();  // or tf.argMax(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMax(axis).print();  // or tf.argMax(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMax_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMax');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMax($x, axes[0]);\n  };\n\n  const inputs: ArgMaxInputs = {x: $x};\n  const attrs: ArgMaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMax,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMax = op({argMax_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ArgMin, ArgMinAttrs, ArgMinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Returns the indices of the minimum values along an `axis`.\n *\n * The result has the same shape as `input` with the dimension along `axis`\n * removed.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.argMin().print();  // or tf.argMin(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 4, 3], [2, 2]);\n *\n * const axis = 1;\n * x.argMin(axis).print();  // or tf.argMin(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension to reduce. Defaults to 0 (outer-most dimension).\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction argMin_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  let $x = convertToTensor(x, 'x', 'argMin');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if (axis == null) {\n      axis = 0;\n    }\n    let axes = util.parseAxisParam(axis, $x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n    if (permutedAxes != null) {\n      $x = transpose($x, permutedAxes);\n      axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n    }\n    return backend.argMin($x, axes[0]);\n  };\n\n  const inputs: ArgMinInputs = {x: $x};\n  const attrs: ArgMinAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, ArgMin,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const argMin = op({argMin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asin, AsinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes asin of the input `tf.Tensor` element-wise: `asin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asin().print();  // or tf.asin(x)\n * ```\n * @param x The input tensor.\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asin');\n  const inputs: AsinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asin);\n}\nexport const asin = op({asin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Asinh, AsinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic sin of the input `tf.Tensor` element-wise:\n * `asinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.asinh().print();  // or tf.asinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction asinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'asinh');\n\n  const inputs: AsinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.asinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Asinh);\n}\nexport const asinh = op({asinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atan, AtanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes atan of the input `tf.Tensor` element-wise: `atan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.atan().print();  // or tf.atan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atan');\n\n  const inputs: AtanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atan);\n}\nexport const atan = op({atan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Atan2, Atan2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes arctangent of `tf.Tensor`s a / b element-wise: `atan2(a, b)`.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1.0, 1.0, -1.0, .7]);\n * const b = tf.tensor1d([2.0, 13.0, 3.5, .21]);\n *\n * tf.atan2(a, b).print()\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atan2_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'atan2');\n  let $b = convertToTensor(b, 'b', 'atan2');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.atan2($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: Atan2Inputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Atan2) as T;\n}\n\nexport const atan2 = op({atan2_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Atanh, AtanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes inverse hyperbolic tan of the input `tf.Tensor` element-wise:\n * `atanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.atanh().print();  // or tf.atanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction atanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'atanh');\n\n  const inputs: AtanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.atanh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Atanh);\n}\nexport const atanh = op({atanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\ntype PadType = 'SAME'|'VALID'|'NUMBER'|'EXPLICIT';\n\n// For NHWC should be in the following form:\n//  [[0, 0], [pad_top,pad_bottom], [pad_left, pad_right], [0, 0]]\n// For NCHW should be in the following form:\n//  [[0, 0], [0, 0], [pad_top,pad_bottom], [pad_left, pad_right]]\n// Reference: https://www.tensorflow.org/api_docs/python/tf/nn/conv2d\nexport type ExplicitPadding =\n    [[number, number], [number, number], [number, number], [number, number]];\n\nexport type PadInfo = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  type: PadType\n};\n\nexport type PadInfo3D = {\n  top: number,\n  left: number,\n  right: number,\n  bottom: number,\n  front: number,\n  back: number,\n  type: PadType\n};\n\n/**\n * Information about the forward pass of a convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv2DInfo = {\n  batchSize: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideHeight: number,\n  strideWidth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo,\n  inShape: [number, number, number, number],\n  outShape: [number, number, number, number],\n  filterShape: [number, number, number, number]\n};\n\n/**\n *\n * @param inputShape Input tensor shape is of the following dimensions:\n *     `[batch, height, width, inChannels]`.\n * @param filterShape The filter shape is of the following dimensions:\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat The data format of the input and output data.\n *     Defaults to 'NHWC'.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.\n *     Defaults to `[1, 1]`. If `dilations` is a single number, then\n *     `dilationHeight == dilationWidth`.\n */\nexport function computeDilation2DInfo(\n    inputShape: [number, number, number, number],\n    filterShape: [number, number, number], strides: number|[number, number],\n    pad: 'same'|'valid'|number, dataFormat: 'NHWC' = 'NHWC',\n    dilations: number|[number, number]) {\n  // `computerConv2DInfo` require filterShape to be in the dimension of:\n  // `[filterHeight, filterWidth, depth, outDepth]`, dilation2d doesn't have\n  // outDepth, it should have the same depth as the input.\n  // Input shape: [batch, height, width, inChannels]\n  const inputChannels = inputShape[3];\n  const $filterShape =\n      [...filterShape, inputChannels] as [number, number, number, number];\n  const $dataFormat = convertConv2DDataFormat(dataFormat);\n\n  return computeConv2DInfo(\n      inputShape, $filterShape, strides, dilations, pad,\n      null /* roundingMode */, null /* depthWise */, $dataFormat);\n}\n\nexport function computePool2DInfo(\n    inShape: [number, number, number, number],\n    filterSize: [number, number]|number, strides: number|[number, number],\n    dilations: number|[number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  const [filterHeight, filterWidth] = parseTupleParam(filterSize);\n\n  let filterShape: [number, number, number, number];\n  if (dataFormat === 'channelsLast') {\n    filterShape = [filterHeight, filterWidth, inShape[3], inShape[3]];\n  } else if (dataFormat === 'channelsFirst') {\n    filterShape = [filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv2DInfo(\n      inShape, filterShape, strides, dilations, pad, roundingMode, false,\n      dataFormat);\n}\n\n/**\n * Computes the information for a forward pass of a pooling3D operation.\n */\nexport function computePool3DInfo(\n    inShape: [number, number, number, number, number],\n    filterSize: number|[number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    roundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC'): Conv3DInfo {\n  const [filterDepth, filterHeight, filterWidth] = parse3TupleParam(filterSize);\n\n  let filterShape: [number, number, number, number, number];\n  let $dataFormat: 'channelsFirst'|'channelsLast';\n  if (dataFormat === 'NDHWC') {\n    $dataFormat = 'channelsLast';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[4], inShape[4]];\n  } else if (dataFormat === 'NCDHW') {\n    $dataFormat = 'channelsFirst';\n    filterShape =\n        [filterDepth, filterHeight, filterWidth, inShape[1], inShape[1]];\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  return computeConv3DInfo(\n      inShape, filterShape, strides, dilations, pad, false, $dataFormat,\n      roundingMode);\n}\n\n/**\n * Computes the information for a forward pass of a convolution/pooling\n * operation.\n */\nexport function computeConv2DInfo(\n    inShape: [number, number, number, number],\n    filterShape: [number, number, number, number],\n    strides: number|[number, number], dilations: number|[number, number],\n    pad: 'same'|'valid'|number|ExplicitPadding,\n    roundingMode?: 'floor'|'round'|'ceil', depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast'): Conv2DInfo {\n  let [batchSize, inHeight, inWidth, inChannels] = [-1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterHeight, filterWidth, , filterChannels] = filterShape;\n  const [strideHeight, strideWidth] = parseTupleParam(strides);\n  const [dilationHeight, dilationWidth] = parseTupleParam(dilations);\n\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outHeight, outWidth} = getPadAndOutInfo(\n      pad, inHeight, inWidth, strideHeight, strideWidth, effectiveFilterHeight,\n      effectiveFilterWidth, roundingMode, dataFormat);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inHeight,\n    inWidth,\n    inChannels,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideHeight,\n    strideWidth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\n/**\n * Information about the forward pass of a 3D convolution/pooling operation.\n * It includes input and output shape, strides, filter size and padding\n * information.\n */\nexport type Conv3DInfo = {\n  batchSize: number,\n  inDepth: number,\n  inHeight: number,\n  inWidth: number,\n  inChannels: number,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number,\n  outChannels: number,\n  dataFormat: 'channelsFirst'|'channelsLast',\n  strideDepth: number,\n  strideHeight: number,\n  strideWidth: number,\n  dilationDepth: number,\n  dilationHeight: number,\n  dilationWidth: number,\n  filterDepth: number,\n  filterHeight: number,\n  filterWidth: number,\n  effectiveFilterDepth: number,\n  effectiveFilterHeight: number,\n  effectiveFilterWidth: number,\n  padInfo: PadInfo3D,\n  inShape: [number, number, number, number, number],\n  outShape: [number, number, number, number, number],\n  filterShape: [number, number, number, number, number]\n};\n\n/**\n * Computes the information for a forward pass of a 3D convolution/pooling\n * operation.\n */\nexport function computeConv3DInfo(\n    inShape: [number, number, number, number, number],\n    filterShape: [number, number, number, number, number],\n    strides: number|[number, number, number],\n    dilations: number|[number, number, number], pad: 'same'|'valid'|number,\n    depthwise = false,\n    dataFormat: 'channelsFirst'|'channelsLast' = 'channelsLast',\n    roundingMode?: 'floor'|'round'|'ceil'): Conv3DInfo {\n  let [batchSize, inDepth, inHeight, inWidth, inChannels] =\n      [-1, -1, -1, -1, -1];\n  if (dataFormat === 'channelsLast') {\n    [batchSize, inDepth, inHeight, inWidth, inChannels] = inShape;\n  } else if (dataFormat === 'channelsFirst') {\n    [batchSize, inChannels, inDepth, inHeight, inWidth] = inShape;\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n\n  const [filterDepth, filterHeight, filterWidth, , filterChannels] =\n      filterShape;\n  const [strideDepth, strideHeight, strideWidth] = parse3TupleParam(strides);\n  const [dilationDepth, dilationHeight, dilationWidth] =\n      parse3TupleParam(dilations);\n\n  const effectiveFilterDepth =\n      getEffectiveFilterSize(filterDepth, dilationDepth);\n  const effectiveFilterHeight =\n      getEffectiveFilterSize(filterHeight, dilationHeight);\n  const effectiveFilterWidth =\n      getEffectiveFilterSize(filterWidth, dilationWidth);\n  const {padInfo, outDepth, outHeight, outWidth} = get3DPadAndOutInfo(\n      pad, inDepth, inHeight, inWidth, strideDepth, strideHeight, strideWidth,\n      effectiveFilterDepth, effectiveFilterHeight, effectiveFilterWidth,\n      roundingMode);\n\n  const outChannels = depthwise ? filterChannels * inChannels : filterChannels;\n\n  let outShape: [number, number, number, number, number];\n  if (dataFormat === 'channelsFirst') {\n    outShape = [batchSize, outChannels, outDepth, outHeight, outWidth];\n  } else if (dataFormat === 'channelsLast') {\n    outShape = [batchSize, outDepth, outHeight, outWidth, outChannels];\n  }\n\n  return {\n    batchSize,\n    dataFormat,\n    inDepth,\n    inHeight,\n    inWidth,\n    inChannels,\n    outDepth,\n    outHeight,\n    outWidth,\n    outChannels,\n    padInfo,\n    strideDepth,\n    strideHeight,\n    strideWidth,\n    filterDepth,\n    filterHeight,\n    filterWidth,\n    effectiveFilterDepth,\n    effectiveFilterHeight,\n    effectiveFilterWidth,\n    dilationDepth,\n    dilationHeight,\n    dilationWidth,\n    inShape,\n    outShape,\n    filterShape\n  };\n}\n\nfunction computeOutputShape2D(\n    inShape: [number, number], fieldSize: number, stride: number,\n    zeroPad?: number, roundingMode?: 'floor'|'round'|'ceil'): [number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputRows = inShape[0];\n  const inputCols = inShape[1];\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputRows, outputCols];\n}\n\nfunction computeOutputShape4D(\n    inShape: [number, number, number, number], fieldSize: number,\n    outChannels: number, stride: number, zeroPad?: number,\n    roundingMode?: 'floor'|'round'|'ceil'): [number, number, number, number] {\n  if (zeroPad == null) {\n    zeroPad = computeDefaultPad(inShape, fieldSize, stride);\n  }\n  const inputDepth = inShape[0];\n  const inputRows = inShape[1];\n  const inputCols = inShape[2];\n\n  const outputDepths = conditionalRound(\n      (inputDepth - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputDepths),\n      () => `The output # of depths (${outputDepths}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputRows = conditionalRound(\n      (inputRows - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputRows),\n      () => `The output # of rows (${outputRows}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  const outputCols = conditionalRound(\n      (inputCols - fieldSize + 2 * zeroPad) / stride + 1, roundingMode);\n  util.assert(\n      util.isInt(outputCols),\n      () => `The output # of columns (${outputCols}) must be an integer. ` +\n          `Change the stride and/or zero pad parameters`);\n\n  return [outputDepths, outputRows, outputCols, outChannels];\n}\n\nexport function computeDefaultPad(\n    inputShape: [number, number]|[number, number, number, number],\n    fieldSize: number, stride: number, dilation = 1): number {\n  const effectiveFieldSize = getEffectiveFilterSize(fieldSize, dilation);\n  return Math.floor(\n      (inputShape[0] * (stride - 1) - stride + effectiveFieldSize) / 2);\n}\n\nfunction parseTupleParam(param: number|number[]): [number, number, number] {\n  if (typeof param === 'number') {\n    return [param, param, param];\n  }\n  if (param.length === 2) {\n    return [param[0], param[1], 1];\n  }\n  return param as [number, number, number];\n}\n\nfunction parse3TupleParam(param: number|[number, number, number]):\n    [number, number, number] {\n  return typeof param === 'number' ? [param, param, param] : param;\n}\n\n/* See https://www.tensorflow.org/api_docs/python/tf/nn/atrous_conv2d\n * Atrous convolution is equivalent to standard convolution with upsampled\n * filters with effective_filter_height =\n * filter_height + (filter_height - 1) * (dilation - 1)\n * and effective_filter_width =\n * filter_width + (filter_width - 1) * (dilation - 1),\n * produced by inserting dilation - 1 zeros along consecutive elements across\n * the filters' spatial dimensions.\n * When there is a dilation, this converts a filter dimension to the\n * effective filter dimension, so it can be used in a standard convolution.\n */\nfunction getEffectiveFilterSize(filterSize: number, dilation: number) {\n  if (dilation <= 1) {\n    return filterSize;\n  }\n\n  return filterSize + (filterSize - 1) * (dilation - 1);\n}\n\nfunction getPadAndOutInfo(\n    pad: 'same'|'valid'|number|ExplicitPadding, inHeight: number,\n    inWidth: number, strideHeight: number, strideWidth: number,\n    filterHeight: number, filterWidth: number,\n    roundingMode: 'floor'|'round'|'ceil',\n    dataFormat: 'channelsFirst'|\n    'channelsLast'): {padInfo: PadInfo, outHeight: number, outWidth: number} {\n  let padInfo: PadInfo;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {top: pad, bottom: pad, left: pad, right: pad, type: padType};\n    const outShape = computeOutputShape2D(\n        [inHeight, inWidth], filterHeight, strideHeight, pad, roundingMode);\n    outHeight = outShape[0];\n    outWidth = outShape[1];\n  } else if (pad === 'same') {\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongHeight =\n        Math.max(0, (outHeight - 1) * strideHeight + filterHeight - inHeight);\n    const padAlongWidth =\n        Math.max(0, (outWidth - 1) * strideWidth + filterWidth - inWidth);\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n    padInfo = {top, bottom, left, right, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {top: 0, bottom: 0, left: 0, right: 0, type: 'VALID'};\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else if (typeof pad === 'object') {\n    const top = dataFormat === 'channelsLast' ? pad[1][0] : pad[2][0];\n    const bottom = dataFormat === 'channelsLast' ? pad[1][1] : pad[2][1];\n    const left = dataFormat === 'channelsLast' ? pad[2][0] : pad[3][0];\n    const right = dataFormat === 'channelsLast' ? pad[2][1] : pad[3][1];\n    const padType = (top === 0 && bottom === 0 && left === 0 && right === 0) ?\n        'VALID' :\n        'EXPLICIT';\n    padInfo = {top, bottom, left, right, type: padType};\n    outHeight = conditionalRound(\n        (inHeight - filterHeight + top + bottom) / strideHeight + 1,\n        roundingMode);\n    outWidth = conditionalRound(\n        (inWidth - filterWidth + left + right) / strideWidth + 1, roundingMode);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outHeight, outWidth};\n}\n\nfunction get3DPadAndOutInfo(\n    pad: 'same'|'valid'|number, inDepth: number, inHeight: number,\n    inWidth: number, strideDepth: number, strideHeight: number,\n    strideWidth: number, filterDepth: number, filterHeight: number,\n    filterWidth: number, roundingMode?: 'floor'|'round'|'ceil'): {\n  padInfo: PadInfo3D,\n  outDepth: number,\n  outHeight: number,\n  outWidth: number\n} {\n  let padInfo: PadInfo3D;\n  let outDepth: number;\n  let outHeight: number;\n  let outWidth: number;\n\n  if (typeof pad === 'number') {\n    const padType = (pad === 0) ? 'VALID' : 'NUMBER';\n    padInfo = {\n      top: pad,\n      bottom: pad,\n      left: pad,\n      right: pad,\n      front: pad,\n      back: pad,\n      type: padType\n    };\n    const outShape = computeOutputShape4D(\n        [inDepth, inHeight, inWidth, 1], filterDepth, 1, strideDepth, pad,\n        roundingMode);\n    outDepth = outShape[0];\n    outHeight = outShape[1];\n    outWidth = outShape[2];\n  } else if (pad === 'same') {\n    outDepth = Math.ceil(inDepth / strideDepth);\n    outHeight = Math.ceil(inHeight / strideHeight);\n    outWidth = Math.ceil(inWidth / strideWidth);\n    const padAlongDepth = (outDepth - 1) * strideDepth + filterDepth - inDepth;\n    const padAlongHeight =\n        (outHeight - 1) * strideHeight + filterHeight - inHeight;\n    const padAlongWidth = (outWidth - 1) * strideWidth + filterWidth - inWidth;\n    const front = Math.floor(padAlongDepth / 2);\n    const back = padAlongDepth - front;\n    const top = Math.floor(padAlongHeight / 2);\n    const bottom = padAlongHeight - top;\n    const left = Math.floor(padAlongWidth / 2);\n    const right = padAlongWidth - left;\n\n    padInfo = {top, bottom, left, right, front, back, type: 'SAME'};\n  } else if (pad === 'valid') {\n    padInfo = {\n      top: 0,\n      bottom: 0,\n      left: 0,\n      right: 0,\n      front: 0,\n      back: 0,\n      type: 'VALID'\n    };\n    outDepth = Math.ceil((inDepth - filterDepth + 1) / strideDepth);\n    outHeight = Math.ceil((inHeight - filterHeight + 1) / strideHeight);\n    outWidth = Math.ceil((inWidth - filterWidth + 1) / strideWidth);\n  } else {\n    throw Error(`Unknown padding parameter: ${pad}`);\n  }\n  return {padInfo, outDepth, outHeight, outWidth};\n}\n\n/**\n * Rounds a value depending on the rounding mode\n * @param value\n * @param roundingMode\n */\nfunction conditionalRound(\n    value: number, roundingMode?: 'floor'|'round'|'ceil') {\n  if (!roundingMode) {\n    return value;\n  }\n  switch (roundingMode) {\n    case 'round':\n      // used for Caffe Conv\n      return Math.round(value);\n    case 'ceil':\n      // used for Caffe Pool\n      return Math.ceil(value);\n    case 'floor':\n      return Math.floor(value);\n    default:\n      throw new Error(`Unknown roundingMode ${roundingMode}`);\n  }\n}\n\nexport function tupleValuesAreOne(param: number|number[]): boolean {\n  const [dimA, dimB, dimC] = parseTupleParam(param);\n  return dimA === 1 && dimB === 1 && dimC === 1;\n}\n\nexport function eitherStridesOrDilationsAreOne(\n    strides: number|number[], dilations: number|number[]): boolean {\n  return tupleValuesAreOne(strides) || tupleValuesAreOne(dilations);\n}\n\n/**\n * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to\n *    'channelsLast'|'channelsFirst'\n * @param dataFormat in 'NHWC'|'NCHW' mode\n * @return dataFormat in 'channelsLast'|'channelsFirst' mode\n * @throws unknown dataFormat\n */\nexport function convertConv2DDataFormat(dataFormat: 'NHWC'|'NCHW'):\n    'channelsLast'|'channelsFirst' {\n  if (dataFormat === 'NHWC') {\n    return 'channelsLast';\n  } else if (dataFormat === 'NCHW') {\n    return 'channelsFirst';\n  } else {\n    throw new Error(`Unknown dataFormat ${dataFormat}`);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D average pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction avgPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'avgPool', 'float32');\n  const dilations = 1;\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in avgPool: x must be rank 4 but got rank ${x4D.rank}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    save([x4D]);\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      return x4D.clone();\n    }\n\n    return backend.avgPool(x4D, convInfo);\n  };\n\n  const inputs: AvgPoolInputs = {x: x4D};\n\n  const attrs: AvgPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, $x.dtype);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool = op({avgPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {AvgPool3D, AvgPool3DAttrs, AvgPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D average pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.avgPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates:\n *     `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction avgPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'avgPool3d', 'float32');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in avgPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in avgPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in avgPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n\n    save([x5D]);\n\n    return backend.avgPool3d(x5D, convInfo);\n  };\n\n  const inputs: AvgPool3DInputs = {x: x5D};\n\n  const attrs: AvgPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  let res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, AvgPool3D,\n      attrs as {} as NamedAttrMap);\n\n  res = cast(res, x5D.dtype);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3d = op({avgPool3d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as util from '../util';\n\nexport function assertParamsConsistent(shapes: number[][], axis: number) {\n  const rank = shapes[0].length;\n  shapes.forEach((shape, i) => {\n    util.assert(\n        shape.length === rank,\n        () =>\n            `Error in concat${rank}D: rank of tensors[${i}] must be the same ` +\n            `as the rank of the rest (${rank})`);\n  });\n\n  util.assert(\n      axis >= 0 && axis < rank,\n      () => `Error in concat${rank}D: axis must be between 0 and ${rank - 1}.`);\n\n  const firstShape = shapes[0];\n  shapes.forEach((shape, i) => {\n    for (let r = 0; r < rank; r++) {\n      util.assert(\n          (r === axis) || (shape[r] === firstShape[r]),\n          () => `Error in concat${rank}D: Shape of tensors[${i}] (${shape}) ` +\n              `does not match the shape of the rest (${firstShape}) ` +\n              `along the non-concatenated axis ${i}.`);\n    }\n  });\n}\n\nexport function computeOutShape(shapes: number[][], axis: number): number[] {\n  const outputShape = shapes[0].slice();\n  for (let i = 1; i < shapes.length; i++) {\n    outputShape[axis] += shapes[i][axis];\n  }\n  return outputShape;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Concat, ConcatAttrs, ConcatInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, parseAxisParam, sizeFromShape} from '../util';\n\nimport {assertParamsConsistent, computeOutShape} from './concat_util';\nimport {op} from './operation';\nimport {tensor} from './tensor';\n\n/**\n * Concatenates a list of `tf.Tensor`s along a given axis.\n *\n * The tensors ranks and types must match, and their sizes must match in all\n * dimensions except `axis`.\n *\n * Also available are stricter rank-specific methods that assert that\n * `tensors` are of the given rank:\n *   - `tf.concat1d`\n *   - `tf.concat2d`\n *   - `tf.concat3d`\n *   - `tf.concat4d`\n *\n * Except `tf.concat1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * a.concat(b).print();  // or a.concat(b)\n * ```\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.concat([a, b, c]).print();\n * ```\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [10, 20]]);\n * const b = tf.tensor2d([[3, 4], [30, 40]]);\n * const axis = 1;\n * tf.concat([a, b], axis).print();\n * ```\n * @param tensors A list of tensors to concatenate.\n * @param axis The axis to concate along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction concat_<T extends Tensor>(tensors: Array<T|TensorLike>, axis = 0): T {\n  assert(tensors.length >= 1, () => 'Pass at least one tensor to concat');\n\n  let $tensors = convertToTensorArray(tensors, 'tensors', 'concat');\n\n  if ($tensors[0].dtype === 'complex64') {\n    $tensors.forEach(tensor => {\n      if (tensor.dtype !== 'complex64') {\n        throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${tensor.dtype}. `);\n      }\n    });\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $axis = parseAxisParam(axis, $tensors[0].shape)[0];\n    const outShape = computeOutShape($tensors.map(t => t.shape), $axis);\n    if (sizeFromShape(outShape) === 0) {\n      return tensor([], outShape) as T;\n    }\n    // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n    $tensors = $tensors.filter(t => t.size > 0);\n    if ($tensors.length === 1) {\n      return $tensors[0];\n    }\n\n    const shapes = $tensors.map(t => t.shape);\n    assertParamsConsistent(shapes, $axis);\n\n    const res = backend.concat($tensors, $axis);\n    save($tensors);\n    return res;\n  };\n\n  const inputs: ConcatInputs = $tensors;\n  const attr: ConcatAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Concat,\n             attr as {} as NamedAttrMap) as T;\n}\n\nexport const concat = op({concat_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sigmoid, SigmoidInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sigmoid element-wise, `1 / (1 + exp(-x))`\n *\n * ```js\n * const x = tf.tensor1d([0, -1, 2, -3]);\n *\n * x.sigmoid().print();  // or tf.sigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sigmoid');\n\n  const inputs: SigmoidInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sigmoid($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sigmoid);\n}\nexport const sigmoid = op({sigmoid_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Slice, SliceAttrs, SliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as slice_util from './slice_util';\n\n/**\n * Extracts a slice from a `tf.Tensor` starting at coordinates `begin`\n * and is of size `size`.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `x` is of the given rank:\n *   - `tf.slice1d`\n *   - `tf.slice2d`\n *   - `tf.slice3d`\n *   - `tf.slice4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.slice([1], [2]).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * x.slice([1, 0], [1, 2]).print();\n * ```\n * @param x The input `tf.Tensor` to slice from.\n * @param begin The coordinates to start the slice from. The length can be\n *     less than the rank of x - the rest of the axes will have implicit 0 as\n *     start. Can also be a single number, in which case it specifies the\n *     first axis.\n * @param size The size of the slice. The length can be less than the rank of\n *     x - the rest of the axes will have implicit -1. A value of -1 requests\n *     the rest of the dimensions in the axis. Can also be a single number,\n *     in which case it specifies the size of the first axis.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction slice_<R extends Rank, T extends Tensor<R>>(\n    x: T|TensorLike, begin: number|number[], size?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'slice');\n\n  if ($x.rank === 0) {\n    throw new Error('Slicing scalar is not possible');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const [begin_, size_] = slice_util.parseSliceParams($x, begin, size);\n    slice_util.assertParamsValid($x, begin_, size_);\n    save([$x]);\n    return backend.slice($x, begin_, size_);\n  };\n\n  const inputs: SliceInputs = {x: $x};\n  const attrs: SliceAttrs = {begin, size};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Slice,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const slice = op({slice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tanh, TanhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic tangent of the input `tf.Tensor` element-wise: `tanh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, 70]);\n *\n * x.tanh().print();  // or tf.tanh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tanh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tanh');\n\n  const inputs: TanhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const y = backend.tanh($x);\n    save([y]);\n    return y;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tanh);\n}\nexport const tanh = op({tanh_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {add} from './add';\nimport {concat} from './concat';\nimport {matMul} from './mat_mul';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {slice} from './slice';\nimport {tanh} from './tanh';\n\n/**\n * Computes the next state and output of a BasicLSTMCell.\n *\n * Returns `[newC, newH]`.\n *\n * Derived from tf.contrib.rnn.BasicLSTMCell.\n *\n * @param forgetBias Forget bias for the cell.\n * @param lstmKernel The weights for the cell.\n * @param lstmBias The bias for the cell.\n * @param data The input to the cell.\n * @param c Previous cell state.\n * @param h Previous cell output.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction basicLSTMCell_(\n    forgetBias: Scalar|TensorLike, lstmKernel: Tensor2D|TensorLike,\n    lstmBias: Tensor1D|TensorLike, data: Tensor2D|TensorLike,\n    c: Tensor2D|TensorLike, h: Tensor2D|TensorLike): [Tensor2D, Tensor2D] {\n  const $forgetBias =\n      convertToTensor(forgetBias, 'forgetBias', 'basicLSTMCell');\n  const $lstmKernel =\n      convertToTensor(lstmKernel, 'lstmKernel', 'basicLSTMCell');\n  const $lstmBias = convertToTensor(lstmBias, 'lstmBias', 'basicLSTMCell');\n  const $data = convertToTensor(data, 'data', 'basicLSTMCell');\n  const $c = convertToTensor(c, 'c', 'basicLSTMCell');\n  const $h = convertToTensor(h, 'h', 'basicLSTMCell');\n\n  const combined = concat([$data, $h], 1);\n  const weighted = matMul(combined, $lstmKernel);\n  const res: Tensor2D = add(weighted, $lstmBias);\n\n  // i = input_gate, j = new_input, f = forget_gate, o = output_gate\n  const batchSize = res.shape[0];\n  const sliceCols = res.shape[1] / 4;\n  const sliceSize: [number, number] = [batchSize, sliceCols];\n  const i = slice(res, [0, 0], sliceSize);\n  const j = slice(res, [0, sliceCols], sliceSize);\n  const f = slice(res, [0, sliceCols * 2], sliceSize);\n  const o = slice(res, [0, sliceCols * 3], sliceSize);\n\n  const newC: Tensor2D =\n      add(mul(sigmoid(i), tanh(j)),\n          mul($c, sigmoid(add($forgetBias, f)) as Tensor2D));\n  const newH: Tensor2D = mul(tanh(newC), sigmoid(o));\n  return [newC, newH];\n}\n\nexport const basicLSTMCell = op({basicLSTMCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchToSpaceND, BatchToSpaceNDAttrs, BatchToSpaceNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation reshapes the \"batch\" dimension 0 into `M + 1` dimensions of\n * shape `blockShape + [batch]`, interleaves these blocks back into the grid\n * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with\n * the same rank as the input. The spatial dimensions of this intermediate\n * result are then optionally cropped according to `crops` to produce the\n * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise\n * description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);\n * const blockShape = [2, 2];\n * const crops = [[0, 0], [0, 0]];\n *\n * x.batchToSpaceND(blockShape, crops).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param crops A 2-D array.  Must have shape `[M, 2]`, all values must be >= 0.\n * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input\n * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required\n * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,\n * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,\n * x.shape[N-1]]`\n *\n * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /\n * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *\n * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`\n * according to `crops` to produce the output of shape: `[batch /\n * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],\n * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -\n * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction batchToSpaceND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], crops: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'batchToSpaceND');\n  const prod = blockShape.reduce((a, b) => a * b);\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank is ${$x.rank} but should be > than blockShape.length ${\n          blockShape.length}`);\n\n  util.assert(\n      crops.length === blockShape.length,\n      () => `crops.length is ${\n          crops.length} but should be equal to blockShape.length  ${\n          blockShape.length}`);\n\n  util.assert(\n      $x.shape[0] % prod === 0,\n      () => `input tensor batch is ${\n                $x.shape[0]} but is not divisible by the product of ` +\n          `the elements of blockShape ${blockShape.join(' * ')} === ${prod}`);\n\n  const forward: ForwardFunc<T> = backend => {\n    return backend.batchToSpaceND($x, blockShape, crops);\n  };\n\n  const inputs: BatchToSpaceNDInputs = {x: $x};\n  const attrs: BatchToSpaceNDAttrs = {blockShape, crops};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      BatchToSpaceND, attrs as {} as NamedAttrMap);\n}\n\nexport const batchToSpaceND = op({batchToSpaceND_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, Tensor4D} from '../tensor';\nimport {Rank} from '../types';\nimport {reshape} from './reshape';\n\nexport function xAs4D<R extends Rank>(x: Tensor<R>) {\n  let x4D: Tensor4D;\n  if (x.rank === 0 || x.rank === 1) {\n    x4D = reshape(x, [1, 1, 1, x.size]);\n  } else if (x.rank === 2) {\n    x4D = reshape(x, [1, 1, x.shape[0], x.shape[1]]);\n  } else if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  } else {\n    x4D = x as Tensor4D;\n  }\n\n  return x4D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {xAs4D} from './batchnorm_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Batch normalization.\n *\n * As described in\n * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).\n *\n * Mean, variance, scale, and offset can be of two shapes:\n *   - The same shape as the input.\n *   - In the common case, the depth dimension is the last dimension of x, so\n *     the values would be an `tf.Tensor1D` of shape [depth].\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that parameters passed are of given rank\n *   - `tf.batchNorm2d`\n *   - `tf.batchNorm3d`\n *   - `tf.batchNorm4d`\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction batchNorm_<R extends Rank>(\n    x: Tensor<R>|TensorLike, mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor<R>|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor<R>|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n\n  util.assert(\n      $mean.rank === $variance.rank,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      $offset == null || $mean.rank === $offset.rank,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      $scale == null || $mean.rank === $scale.rank,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  const x4D: Tensor4D = xAs4D($x);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([x4D, $mean, $variance, $scale]);\n\n    return backend.batchNorm(\n        x4D, as1DOr4D($mean), as1DOr4D($variance), as1DOr4D($offset),\n        as1DOr4D($scale), varianceEpsilon);\n  };\n\n  const inputs: FusedBatchNormInputs = {\n    x: x4D,\n    scale: $scale,\n    offset: $offset,\n    mean: $mean,\n    variance: $variance\n  };\n\n  const attrs: FusedBatchNormAttrs = {varianceEpsilon};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      FusedBatchNorm, attrs as {} as NamedAttrMap);\n\n  return reshape(res, $x.shape);\n}\n\nfunction as1DOr4D(x: Tensor): Tensor4D|Tensor1D {\n  if (x == null) {\n    return null;\n  }\n  if (x.rank === 0) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [x.size]) as Tensor1D;\n  } else if (x.rank === 1) {\n    return x as Tensor1D;\n  } else if (x.rank === 2) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, 1, x.shape[0], x.shape[1]]) as Tensor4D;\n  } else if (x.rank === 3) {\n    // tslint:disable-next-line:no-unnecessary-type-assertion\n    return reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]) as Tensor4D;\n  }\n  return x as Tensor4D;\n}\n\nexport const batchNorm = op({batchNorm_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 2D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm2d_(\n    x: Tensor2D|TensorLike, mean: Tensor2D|Tensor1D|TensorLike,\n    variance: Tensor2D|Tensor1D|TensorLike,\n    offset?: Tensor2D|Tensor1D|TensorLike, scale?: Tensor2D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor2D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor2D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 2,\n      () => `Error in batchNorm2D: x must be rank 2 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 2 || $mean.rank === 1,\n      () => `Error in batchNorm2D: mean must be rank 2 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 2 || $variance.rank === 1,\n      () => `Error in batchNorm2D: variance must be rank 2 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 2 || $scale.rank === 1,\n        () => `Error in batchNorm2D: scale must be rank 2 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 2 || $offset.rank === 1,\n        () => `Error in batchNorm2D: offset must be rank 2 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm2d = op({batchNorm2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 3D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm3d_(\n    x: Tensor3D|TensorLike, mean: Tensor3D|Tensor1D|TensorLike,\n    variance: Tensor3D|Tensor1D|TensorLike,\n    offset?: Tensor3D|Tensor1D|TensorLike, scale?: Tensor3D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor3D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor3D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 3,\n      () => `Error in batchNorm3D: x must be rank 3 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 3 || $mean.rank === 1,\n      () => `Error in batchNorm3D: mean must be rank 3 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 3 || $variance.rank === 1,\n      () => `Error in batchNorm3D: variance must be rank 3 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 3 || $scale.rank === 1,\n        () => `Error in batchNorm3D: scale must be rank 3 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 3 || $offset.rank === 1,\n        () => `Error in batchNorm3D: offset must be rank 3 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm3d = op({batchNorm3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {batchNorm} from './batchnorm';\nimport {op} from './operation';\n\n/**\n * Batch normalization, strictly for 4D. For the more relaxed version, see\n * `tf.batchNorm`.\n *\n * @param x The input Tensor.\n * @param mean A mean Tensor.\n * @param variance A variance Tensor.\n * @param offset An offset Tensor.\n * @param scale A scale Tensor.\n * @param varianceEpsilon A small float number to avoid dividing by 0.\n */\nfunction batchNorm4d_(\n    x: Tensor4D|TensorLike, mean: Tensor4D|Tensor1D|TensorLike,\n    variance: Tensor4D|Tensor1D|TensorLike,\n    offset?: Tensor4D|Tensor1D|TensorLike, scale?: Tensor4D|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'batchNorm');\n  const $mean = convertToTensor(mean, 'mean', 'batchNorm');\n  const $variance = convertToTensor(variance, 'variance', 'batchNorm');\n  let $scale: Tensor4D|Tensor1D;\n  if (scale != null) {\n    $scale = convertToTensor(scale, 'scale', 'batchNorm');\n  }\n  let $offset: Tensor4D|Tensor1D;\n  if (offset != null) {\n    $offset = convertToTensor(offset, 'offset', 'batchNorm');\n  }\n  util.assert(\n      $x.rank === 4,\n      () => `Error in batchNorm4D: x must be rank 4 but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $mean.rank === 4 || $mean.rank === 1,\n      () => `Error in batchNorm4D: mean must be rank 4 or rank 1 but ` +\n          `got rank ${$mean.rank}.`);\n  util.assert(\n      $variance.rank === 4 || $variance.rank === 1,\n      () => `Error in batchNorm4D: variance must be rank 4 or rank 1 ` +\n          `but got rank ${$variance.rank}.`);\n  if ($scale != null) {\n    util.assert(\n        $scale.rank === 4 || $scale.rank === 1,\n        () => `Error in batchNorm4D: scale must be rank 4 or rank 1 ` +\n            `but got rank ${$scale.rank}.`);\n  }\n  if ($offset != null) {\n    util.assert(\n        $offset.rank === 4 || $offset.rank === 1,\n        () => `Error in batchNorm4D: offset must be rank 4 or rank 1 ` +\n            `but got rank ${$offset.rank}.`);\n  }\n  return batchNorm($x, $mean, $variance, $offset, $scale, varianceEpsilon);\n}\n\nexport const batchNorm4d = op({batchNorm4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE} from '../engine';\nimport {BroadcastTo, BroadCastToAttrs, BroadcastToInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Broadcast an array to a compatible shape NumPy-style.\n *\n * The tensor's shape is compared to the broadcast shape from end to beginning.\n * Ones are prepended to the tensor's shape until is has the same length as\n * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is\n * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then\n * the input tensor is tiled N times along that axis (using tf.tile).\n *\n * @param input The tensor that is to be broadcasted.\n * @param shape The input is to be broadcast to this shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction broadcastTo_<R extends Rank>(\n    x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor<R> {\n  let input = convertToTensor(x, 'broadcastTo', 'x');\n  const xShape = input.shape;\n\n  if (shape.some(d => !(d > 0) || d % 1 !== 0)) {\n    throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);\n  }\n\n  if (shape.length < input.rank) {\n    throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${\n        input.rank}.`);\n  }\n\n  if (shape.length > input.rank) {\n    const newShape = input.shape.slice();\n    while (newShape.length < shape.length) {\n      newShape.unshift(1);\n    }\n    input = reshape(input, newShape);\n  }\n\n  const inputShape = input.shape;\n  const reps: number[] = Array.from(shape);\n  for (let i = shape.length - 1; i >= 0; i--) {\n    if (inputShape[i] === shape[i]) {\n      reps[i] = 1;\n    } else if (input.shape[i] !== 1) {\n      throw new Error(\n          `broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);\n    }\n  }\n  const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);\n\n  if (axes.length === 0) {\n    return clone(input) as Tensor<R>;\n  }\n\n  const forward = (backend: KernelBackend) => backend.tile(input, reps);\n\n  const inputs: BroadcastToInputs = {x: input};\n  const attrs: BroadCastToAttrs = {shape, inputShape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             BroadcastTo, attrs as unknown as NamedAttrMap) as Tensor<R>;\n}\n\nexport const broadcastTo = op({broadcastTo_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Ceil, CeilInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes ceiling of input `tf.Tensor` element-wise: `ceil(x)`\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.ceil().print();  // or tf.ceil(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction ceil_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'ceil');\n\n  const inputs: CeilInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.ceil($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Ceil);\n}\nexport const ceil = op({ceil_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {ClipByValue, ClipByValueAttrs, ClipByValueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)\n * ```\n * @param x The input tensor.\n * @param clipValueMin Lower-bound of range to be clipped to.\n * @param clipValueMax Upper-bound of range to be clipped to.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction clipByValue_<T extends Tensor>(\n    x: T|TensorLike, clipValueMin: number, clipValueMax: number): T {\n  const $x = convertToTensor(x, 'x', 'clipByValue');\n  util.assert(\n      (clipValueMin <= clipValueMax),\n      () => `Error in clip: min (${clipValueMin}) must be ` +\n          `less than or equal to max (${clipValueMax}).`);\n\n  const inputs: ClipByValueInputs = {x: $x};\n  const attrs: ClipByValueAttrs = {clipValueMin, clipValueMax};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const res = backend.clip($x, clipValueMin, clipValueMax);\n        save([$x]);\n        return res;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, ClipByValue,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const clipByValue = op({clipByValue_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor1D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(3) = |r1, g1, b1|\n * B: shape(2) = |r2, g2|\n * C = tf.concat1d([A, B]) == |r1, g1, b1, r2, g2|\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @return The concatenated array.\n */\nfunction concat1d_(tensors: Array<Tensor1D|TensorLike>): Tensor1D {\n  return concat(tensors, 0 /* axis */);\n}\n\nexport const concat1d = op({concat1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of`tf.Tensor2D`s along an axis. See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *\n * B: shape(2, 3) = | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * C = tf.concat2d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 3) = | r1, g1, b1 |\n *                  | r2, g2, b2 |\n *                  | r3, g3, b3 |\n *                  | r4, g4, b4 |\n *\n * if axis = 1:\n * C = shape(2, 6) = | r1, g1, b1, r3, g3, b3 |\n *                   | r2, g2, b2, r4, g4, b4 |\n *\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concatenate along.\n * @return The concatenated array.\n */\nfunction concat2d_(\n    tensors: Array<Tensor2D|TensorLike>, axis: number): Tensor2D {\n  return concat(tensors, axis);\n}\n\nexport const concat2d = op({concat2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor3D`s along an axis.\n * See `concat` for details.\n *\n * For example, if:\n * A: shape(2, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *\n * B: shape(2, 1, 3) = | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * C = tf.concat3d([A, B], axis)\n *\n * if axis = 0:\n * C: shape(4, 1, 3) = | r1, g1, b1 |\n *                     | r2, g2, b2 |\n *                     | r3, g3, b3 |\n *                     | r4, g4, b4 |\n *\n * if axis = 1:\n * C: shape(2, 2, 3) = | r1, g1, b1, r3, g3, b3 |\n *                     | r2, g2, b2, r4, g4, b4 |\n *\n * if axis = 2:\n * C = shape(2, 1, 6) = | r1, g1, b1, r3, g3, b3 |\n *                      | r2, g2, b2, r4, g4, b4 |\n *\n * @param tensors A list of`tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat3d_(\n    tensors: Array<Tensor3D|TensorLike>, axis: number): Tensor3D {\n  return concat(tensors, axis);\n}\n\nexport const concat3d = op({concat3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\n\nimport {concat} from './concat';\nimport {op} from './operation';\n\n/**\n * Concatenates a list of `tf.Tensor4D`s along an axis.\n * See `concat` for details.\n *\n * @param tensors A list of `tf.Tensor`s to concatenate.\n * @param axis The axis to concate along.\n * @return The concatenated array.\n */\nfunction concat4d_(\n    tensors: Array<Tensor4D|TensorLike>, axis: number): Tensor4D {\n  return concat(tensors, axis);\n}\n\nexport const concat4d = op({concat4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2D, Conv2DAttrs, Conv2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 2D convolution over the input x.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2d: input must be rank 4, but got rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  util.assert(\n      inDepth === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${inDepth}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        false, $dataFormat);\n    const res = backend.conv2d(x4D, $filter, convInfo);\n\n    save([x4D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DInputs = {x: x4D, filter: $filter};\n  const attrs:\n      Conv2DAttrs = {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv2D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2d = op({conv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D, Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 1D convolution over the input x.\n *\n * @param x The input tensor, of rank 3 or rank 2, of shape\n *     `[batch, width, inChannels]`. If rank 2, batch of 1 is assumed.\n * @param filter The filter, rank 3, of shape\n *     `[filterWidth, inDepth, outDepth]`.\n * @param stride The number of entries by which the filter is moved right at\n *     each step.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from \"NWC\", \"NCW\". Defaults to \"NWC\",\n *     the data is stored in the order of [batch, in_width, in_channels]. Only\n *     \"NWC\" is currently supported.\n * @param dilation The dilation rate in which we sample input values in\n *     atrous convolution. Defaults to `1`. If it is greater than 1, then\n *     stride must be `1`.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv1d_<T extends Tensor2D|Tensor3D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike, stride: number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv1d');\n  const $filter = convertToTensor(filter, 'filter', 'conv1d');\n\n  let x3D = $x as Tensor3D;\n  let reshapedTo3D = false;\n  if ($x.rank === 2) {\n    reshapedTo3D = true;\n    x3D = reshape($x, [1, $x.shape[0], $x.shape[1]]);\n  }\n\n  util.assert(\n      x3D.rank === 3,\n      () => `Error in conv1d: input must be rank 3, but got rank ${x3D.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in conv1d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv1d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x3D.shape[2] === $filter.shape[1],\n      () => `Error in conv1d: depth of input (${x3D.shape[2]}) must match ` +\n          `input depth for filter ${$filter.shape[1]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(stride, dilation),\n      () => 'Error in conv1D: Either stride or dilation must be 1. ' +\n          `Got stride ${stride} and dilation '${dilation}'`);\n  util.assert(\n      dataFormat === 'NWC',\n      () => `Error in conv1d: got dataFormat of ${\n          dataFormat} but only NWC is currently supported.`);\n\n  const filter4D = reshape(\n      $filter, [1, $filter.shape[0], $filter.shape[1], $filter.shape[2]]);\n  const input4D = reshape(x3D, [x3D.shape[0], 1, x3D.shape[1], x3D.shape[2]]);\n  const strides: [number, number] = [1, stride];\n  const dilations: [number, number] = [1, dilation];\n\n  const conv2dDataFormat = 'NHWC';\n\n  const res = conv2d(\n      (input4D as Tensor4D), (filter4D as Tensor4D), strides, pad,\n      conv2dDataFormat, dilations, dimRoundingMode);\n\n  if (reshapedTo3D) {\n    return reshape(res, [res.shape[2], res.shape[3]]) as T;\n  }\n\n  return reshape(res, [res.shape[0], res.shape[2], res.shape[3]]) as T;\n}\n\nexport const conv1d = op({conv1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs, Conv2DBackpropInputInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 2D convolution.\n *\n * @param xShape The shape of the input: [batch, height, width, inDepth].\n * If length of 3, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 4 or rank 3 of shape\n *   `[batch, outHeight, outWidth, outDepth]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction conv2DBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape4D = xShape as [number, number, number, number];\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n    xShape4D = [1, xShape[0], xShape[1], xShape[2]];\n  }\n\n  util.assert(\n      xShape4D.length === 4,\n      () =>\n          `Error in conv2dDerInput: inShape must be length 4, but got length ` +\n          `${xShape4D.length}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerInput: dy must be rank 4, but got ` +\n          `rank ${dy4D.rank}`);\n  util.assert(\n      filter.rank === 4,\n      () => `Error in conv2dDerInput: filter must be rank 4, but got ` +\n          `rank ${filter.rank}`);\n  const inDepth = dataFormat === 'NHWC' ? xShape4D[3] : xShape4D[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filter.shape[2],\n      () => `Error in conv2dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[2]}.`);\n  util.assert(\n      outDepth === filter.shape[3],\n      () => `Error in conv2dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[3]}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerInput: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const dilations = 1;\n\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        xShape4D, filter.shape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    const res = backend.conv2dDerInput(dy4D, filter, convInfo);\n\n    save([dy4D, filter]);\n\n    return res;\n  };\n\n  const inputs: Conv2DBackpropInputInputs = {dy: dy4D, filter};\n\n  const attrs: Conv2DBackpropInputAttrs =\n      {strides, pad, dataFormat, dimRoundingMode, inputShape: xShape4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      Conv2DBackpropInput, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const conv2DBackpropInput = op({conv2DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv2DBackpropInput} from './conv2d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 2D convolution of an image, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 4 or rank 3, of shape\n *   `[batch, height, width, inDepth]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 4 or rank 3:\n *     `[batch, height, width, outDepth]`. If rank 3, batch of 1 is assumed.\n * @param strides The strides of the original convolution:\n *     `[strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n * @param dimRoundingMode The rounding mode used when computing output\n *    dimensions if pad is a number. If none is provided, it will not round\n *    and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv2dTranspose_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'conv2dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv2dTranspose');\n\n  return conv2DBackpropInput(\n      outputShape, $x, $filter, strides, pad, 'NHWC', dimRoundingMode);\n}\n\nexport const conv2dTranspose = op({conv2dTranspose_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3D, Conv3DAttrs, Conv3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {eitherStridesOrDilationsAreOne} from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes a 3D convolution over the input x.\n *\n * @param x The input tensor, of rank 5 or rank 4, of shape\n *     `[batch, depth, height, width, channels]`. If rank 4,\n * batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inChannels, outChannels]`.\n *      inChannels must match between input and filter.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat: An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationDepth, dilationHeight,\n *     dilationWidth]` in which we sample input values across the height\n *     and width dimensions in atrous convolution. Defaults to `[1, 1, 1]`.\n *     If `dilations` is a single number, then\n *     `dilationDepth == dilationHeight == dilationWidth`. If it is greater\n *     than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    strides: [number, number, number]|number, pad: 'valid'|'same',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations: [number, number, number]|number = [1, 1, 1]): T {\n  const $x = convertToTensor(x, 'x', 'conv3d');\n  const $filter = convertToTensor(filter, 'filter', 'conv3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3d: input must be rank 5, but got rank ${x5D.rank}.`);\n  util.assert(\n      $filter.rank === 5,\n      () => `Error in conv3d: filter must be rank 5, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x5D.shape[4] === $filter.shape[3],\n      () => `Error in conv3d: depth of input (${x5D.shape[4]}) must match ` +\n          `input depth for filter ${$filter.shape[3]}.`);\n  util.assert(\n      eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv3D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in conv3d: got dataFormat of ${\n          dataFormat} but only NDHWC is currently supported.`);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, $filter.shape, strides, dilations, pad);\n    const res = backend.conv3d(x5D, $filter, convInfo);\n\n    save([x5D, $filter]);\n\n    return res;\n  };\n\n  const inputs: Conv3DInputs = {x: x5D, filter: $filter};\n\n  const attrs: Conv3DAttrs = {strides, pad, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Conv3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3d = op({conv3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropInputAttrs, Conv3DBackpropInputInputs, Conv3DBackpropInputV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the input of a 3D convolution.\n *\n * @param xShape The shape of the input: [batch, depth, height, width,\n * in_channels]. If length of 4, batch of 1 is assumed.\n * @param dy The derivative of the output, of rank 5 or rank 4 of shape\n *   `[batch, outDepth, outHeight, outWidth, in_channels]`.\n * If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 5, of shape\n *     `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideDepth, strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm used:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n */\nfunction conv3DBackpropInput_<T extends Tensor4D|Tensor5D>(\n    xShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    dy: T, filter: Tensor5D, strides: [number, number, number]|number,\n    pad: 'valid'|'same'): T {\n  util.assert(\n      xShape.length === dy.rank,\n      () => `Length of inShape ` +\n          `(${xShape.length}) and rank of dy (${dy.rank}) must match`);\n\n  let xShape5D = xShape as [number, number, number, number, number];\n  let dy5D = dy as Tensor5D;\n  let reshapedTo5D = false;\n  if (dy.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n    xShape5D = [1, xShape[0], xShape[1], xShape[2], xShape[3]];\n  }\n\n  const inDepth = xShape5D[4];\n  const outDepth = dy5D.shape[4];\n  util.assert(\n      xShape5D.length === 5,\n      () =>\n          `Error in conv3dDerInput: inShape must be length 5, but got length ` +\n          `${xShape5D.length}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerInput: dy must be rank 5, but got ` +\n          `rank ${dy5D.rank}`);\n  util.assert(\n      filter.rank === 5,\n      () => `Error in conv3dDerInput: filter must be rank 5, but got ` +\n          `rank ${filter.rank}`);\n  util.assert(\n      inDepth === filter.shape[3],\n      () => `Error in conv3dDerInput: depth of input (${inDepth}) must ` +\n          `match input depth for filter ${filter.shape[3]}.`);\n  util.assert(\n      outDepth === filter.shape[4],\n      () => `Error in conv3dDerInput: depth of output (${outDepth}) must ` +\n          `match output depth for filter ${filter.shape[4]}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        xShape5D, filter.shape, strides, dilations, pad);\n\n    return backend.conv3dDerInput(dy5D, filter, convInfo);\n  };\n\n  const inputs: Conv3DBackpropInputInputs = {dy: dy5D};\n\n  const attrs: Conv3DBackpropInputAttrs = {pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, Conv3DBackpropInputV2,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n  return res as T;\n}\n\nexport const conv3DBackpropInput = op({conv3DBackpropInput_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D, Tensor5D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {conv3DBackpropInput} from './conv3d_backprop_input';\nimport {op} from './operation';\n\n/**\n * Computes the transposed 3D convolution of a volume, also known as a\n * deconvolution.\n *\n * @param x The input image, of rank 5 or rank 4, of shape\n *   `[batch, depth, height, width, inDepth]`. If rank 4, batch of 1 is assumed.\n * @param filter The filter, rank 4, of shape\n *     `[depth, filterHeight, filterWidth, outDepth, inDepth]`.\n *     `inDepth` must match `inDepth` in `x`.\n * @param outputShape Output shape, of rank 5 or rank 4:\n *     `[batch, depth, height, width, outDepth]`. If rank 3, batch of 1 is\n *    assumed.\n * @param strides The strides of the original convolution:\n *     `[strideDepth, strideHeight, strideWidth]`.\n * @param pad  The type of padding algorithm used in the non-transpose version\n *    of the op.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction conv3dTranspose_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filter: Tensor5D|TensorLike,\n    outputShape:\n        [number, number, number, number,\n         number]|[number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): T {\n  const $x = convertToTensor(x, 'x', 'conv3dTranspose');\n  const $filter = convertToTensor(filter, 'filter', 'conv3dTranspose');\n\n  return conv3DBackpropInput(outputShape, $x, $filter, strides, pad);\n}\n\nexport const conv3dTranspose = op({conv3dTranspose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cos, CosInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes cos of the input `tf.Tensor` element-wise: `cos(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.cos().print();  // or tf.cos(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cos_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cos');\n\n  const inputs: CosInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cos($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cos);\n}\nexport const cos = op({cos_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Cosh, CoshInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic cos of the input `tf.Tensor` element-wise: `cosh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.cosh().print();  // or tf.cosh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction cosh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'cosh');\n  const inputs: CoshInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.cosh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Cosh);\n}\nexport const cosh = op({cosh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Cumsum, CumsumAttrs, CumsumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {getAxesPermutation, getInnerMostAxes, getUndoAxesPermutation} from './axis_util';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the cumulative sum of a `tf.Tensor` along `axis`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4]);\n * x.cumsum().print();\n * ```\n * ```js\n * const x = tf.tensor([[1, 2], [3, 4]]);\n * x.cumsum().print();\n * ```\n *\n * @param x The input tensor to be summed.\n * @param axis The axis along which to sum. Optional. Defaults to 0.\n * @param exclusive Whether to perform exclusive cumulative sum. Optional.\n *     Defaults to false. If set to true then the sum of each tensor entry\n *     does not include its own value, but only the values previous to it\n *     along the specified axis.\n * @param reverse Whether to sum in the opposite direction. Optional.\n *     Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Scan'}\n */\nfunction cumsum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis = 0, exclusive = false, reverse = false): T {\n  const $x = convertToTensor(x, 'x', 'cumsum');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const permutation = getAxesPermutation([axis], $x.rank);\n        let permutedX = $x;\n        if (permutation != null) {\n          permutedX = transpose($x, permutation);\n        }\n        const permutedAxis = getInnerMostAxes(1, $x.rank)[0];\n        let value = backend.cumsum(permutedX, permutedAxis, exclusive, reverse);\n        save([$x]);\n\n        if (permutation != null) {\n          const reversePermutation = getUndoAxesPermutation(permutation);\n          value = transpose(value, reversePermutation);\n        }\n        return value;\n      };\n\n  const inputs: CumsumInputs = {x: $x};\n  const attrs: CumsumAttrs = {axis, exclusive, reverse};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Cumsum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const cumsum = op({cumsum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthToSpace, DepthToSpaceAttrs, DepthToSpaceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Rearranges data from depth into blocks of spatial data. More specifically,\n * this op outputs a copy of the input tensor where values from the `depth`\n * dimension are moved in spatial blocks to the `height` and `width` dimensions.\n * The attr `blockSize` indicates the input block size and how the data is\n * moved.\n *\n *  - Chunks of data of size `blockSize * blockSize` from depth are rearranged\n * into non-overlapping blocks of size `blockSize x blockSize`\n *\n *  - The width the output tensor is `inputWidth * blockSize`, whereas the\n * height is `inputHeight * blockSize`\n *\n *  - The Y, X coordinates within each block of the output image are determined\n * by the high order component of the input channel index\n *\n *  - The depth of the input tensor must be divisible by `blockSize *\n * blockSize`\n *\n * The `dataFormat` attr specifies the layout of the input and output tensors\n * with the following options: \"NHWC\": [ `batch, height, width, channels` ]\n * \"NCHW\": [ `batch, channels, height, width` ]\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 1, 1, 4]);\n * const blockSize = 2;\n * const dataFormat = \"NHWC\";\n *\n * tf.depthToSpace(x, blockSize, dataFormat).print();\n * ```\n *\n * @param x The input tensor of rank 4\n * @param blockSIze  An `int` that is `>= 2`. The size of the spatial block\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\"\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction depthToSpace_(\n    x: Tensor4D|TensorLike4D, blockSize: number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'depthToSpace') as Tensor4D;\n\n  const inputHeight = (dataFormat === 'NHWC') ? $x.shape[1] : $x.shape[2];\n  const inputWidth = (dataFormat === 'NHWC') ? $x.shape[2] : $x.shape[3];\n  const inputDepth = (dataFormat === 'NHWC') ? $x.shape[3] : $x.shape[1];\n\n  util.assert(\n      inputHeight * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputHeight} and ${blockSize}  for depthToSpace with input shape\n    ${$x.shape}`);\n\n  util.assert(\n      inputWidth * blockSize >= 0,\n      () => `Negative dimension size caused by overflow when multiplying\n    ${inputWidth} and ${blockSize} for depthToSpace with input shape\n        ${$x.shape}`);\n\n  util.assert(\n      (inputDepth % (blockSize * blockSize) === 0),\n      () => `Dimension size must be evenly divisible by ${\n          blockSize * blockSize} but is ${\n          inputDepth} for depthToSpace with input shape ${$x.shape}`);\n\n  const forward: ForwardFunc<Tensor4D> = backend =>\n      backend.depthToSpace($x, blockSize, dataFormat);\n\n  const inputs: DepthToSpaceInputs = {x: $x};\n  const attrs: DepthToSpaceAttrs = {blockSize, dataFormat};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      DepthToSpace, attrs as {} as NamedAttrMap);\n}\n\nexport const depthToSpace = op({depthToSpace_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs, DepthwiseConv2dNativeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Depthwise 2D convolution.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction depthwiseConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor4D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dilations: [number, number]|number = [1, 1],\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in depthwiseConv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in depthwiseConv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1];\n    }\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in depthwiseConv2d: Either strides or dilations must be ' +\n            `1. Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n        true /* depthwise */);\n    const res = backend.depthwiseConv2D(x4D, $filter, convInfo);\n    save([x4D, $filter]);\n    return res;\n  };\n\n  const inputs: DepthwiseConv2dNativeInputs = {x: x4D, filter: $filter};\n  const attrs: DepthwiseConv2dNativeAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      DepthwiseConv2dNative, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2d = op({depthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Diag, DiagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a diagonal tensor with a given diagonal values.\n *\n * Given a diagonal, this operation returns a tensor with the diagonal and\n * everything else padded with zeros.\n *\n * Assume the input has dimensions `[D1,..., Dk]`, then the output is a tensor\n * of rank 2k with dimensions `[D1,..., Dk, D1,..., Dk]`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * tf.diag(x).print()\n * ```\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4, 5, 6, 6, 8], [4, 2])\n *\n * tf.diag(x).print()\n * ```\n * @param x The input tensor.\n */\nfunction diag_(x: Tensor): Tensor {\n  const $x = convertToTensor(x, 'x', 'diag');\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const flat = reshape($x, [$x.size]);\n    const result = backend.diag(flat);\n    const outShape = [...x.shape, ...x.shape];\n\n    return reshape(result, outShape);\n  };\n\n  const inputs: DiagInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Diag);\n}\n\nexport const diag = op({diag_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DAttrs, Dilation2DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the grayscale dilation over the input `x`.\n *\n * @param x The input tensor, rank 3 or rank 4 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filter The filter tensor, rank 3, of shape\n *     `[filterHeight, filterWidth, depth]`.\n * @param strides The strides of the sliding window for each dimension of the\n *     input tensor: `[strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat Specify the data format of the input and output data.\n *      Defaults to 'NHWC'. Only 'NHWC' is currently supported. With the\n *      default format \"NHWC\", the data is stored in the order of: [batch,\n *      height, width, channels].\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     for atrous morphological dilation. Defaults to `[1, 1]`. If `dilations`\n *     is a single number, then `dilationHeight == dilationWidth`. If it is\n *     greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction dilation2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filter: Tensor3D|TensorLike,\n    strides: [number, number]|number, pad: 'valid'|'same',\n    dilations: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'dilation2d');\n  const $filter = convertToTensor(filter, 'filter', 'dilation2d');\n\n  util.assert(\n      $x.rank === 3 || $x.rank === 4,\n      () => `Error in dilation2d: input must be rank 3 or 4, but got rank ` +\n          `${$x.rank}.`);\n  util.assert(\n      $filter.rank === 3,\n      () => `Error in dilation2d: filter must be rank 3, but got rank ` +\n          `${$filter.rank}.`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in dilation2d: Only NHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n    reshapedTo4D = true;\n  }\n\n  const inputs: Dilation2DInputs = {x: x4D, filter: $filter};\n  const attrs: Dilation2DAttrs = {strides, pad, dilations};\n\n  const res = ENGINE.runKernel(\n                  Dilation2D, inputs as {} as NamedTensorMap,\n                  attrs as {} as NamedAttrMap) as T;\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res;\n}\n\nexport const dilation2d = op({dilation2d_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Returns the dimensions in the input shape that are broadcasted to\n * produce the provided output shape.\n *\n * The returned dimensions are 0-indexed and sorted. An example:\n * inShape = [4, 1, 3]\n * outShape = [5, 4, 3, 3]\n * result = [1]. Dimension 1 (2nd dimension of input) gets broadcasted 1 => 3.\n */\nexport function getBroadcastDims(\n    inShape: number[], outShape: number[]): number[] {\n  const inRank = inShape.length;\n  const dims: number[] = [];\n  for (let i = 0; i < inRank; i++) {\n    const dim = inRank - 1 - i;\n    const a = inShape[dim] || 1;\n    const b = outShape[outShape.length - 1 - i] || 1;\n    if (b > 1 && a === 1) {\n      dims.unshift(dim);\n    }\n  }\n  return dims;\n}\n\n/**\n * Returns the axes in the output space that should be reduced to produce\n * the input space.\n */\nexport function getReductionAxes(\n    inShape: number[], outShape: number[]): number[] {\n  const result: number[] = [];\n  for (let i = 0; i < outShape.length; i++) {\n    const inDim = inShape[inShape.length - i - 1];\n    const outAxis = outShape.length - i - 1;\n    const outDim = outShape[outAxis];\n    if (inDim == null || (inDim === 1 && outDim > 1)) {\n      result.unshift(outAxis);\n    }\n  }\n  return result;\n}\n\nexport function assertAndGetBroadcastShape(\n    shapeA: number[], shapeB: number[]): number[] {\n  const result: number[] = [];\n  const l = Math.max(shapeA.length, shapeB.length);\n\n  for (let i = 0; i < l; i++) {\n    let a = shapeA[shapeA.length - i - 1];\n    if (a == null) {\n      a = 1;\n    }\n    let b = shapeB[shapeB.length - i - 1];\n    if (b == null) {\n      b = 1;\n    }\n    if (a === 1) {\n      result.unshift(b);\n    } else if (b === 1) {\n      result.unshift(a);\n    } else if (a !== b) {\n      const errMsg = `Operands could not be broadcast together with shapes ` +\n          `${shapeA} and ${shapeB}.`;\n      throw Error(errMsg);\n    } else {\n      result.unshift(a);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Equal, EqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a == b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.equal(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction equal_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'equal');\n  let $b = convertToTensor(b, 'b', 'equal');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.equal($a, $b);\n\n  const inputs: EqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, Equal) as T;\n}\n\nexport const equal = op({equal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SelectV2, SelectV2Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch} from '../util';\n\nimport {broadcastTo} from './broadcast_to';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the elements, either `a` or `b` depending on the `condition`.\n *\n * If the condition is true, select from `a`, otherwise select from `b`.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const a = tf.tensor1d([1 , 2, 3]);\n * const b = tf.tensor1d([-1, -2, -3]);\n *\n * a.where(cond, b).print();\n * ```\n *\n * @param condition The input condition. Must be of dtype bool.\n * @param a If `condition` is rank 1, `a` may have a higher rank but\n *     its first dimension must match the size of `condition`.\n * @param b A tensor with the same dtype as `a` and with shape that is\n *     compatible with `a`.\n * @return A tensor with same dtype as `a` and `b`, and shape that is\n *     broadcastable from `a` and `b`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction where_<T extends Tensor>(\n    condition: Tensor|TensorLike, a: T|TensorLike, b: T|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'where');\n  const $b = convertToTensor(b, 'b', 'where');\n  const $condition = convertToTensor(condition, 'condition', 'where', 'bool');\n  // TODO: move this logic to forward function when the broadcastTo op is\n  // implemented in WASM.\n  // Find the broadcastable shape for $a and $b.\n  const broadcastShape = assertAndGetBroadcastShape($a.shape, $b.shape);\n  const $broadcastedA = broadcastTo($a, broadcastShape);\n  const $broadcastedB = broadcastTo($b, broadcastShape);\n  if ($condition.rank === 1) {\n    // If condition rank is 1, then the first dimension must match the size of\n    // condition.\n    assert(\n        $condition.shape[0] === $a.shape[0],\n        () => 'The first dimension of `a` must match the size of `condition`.');\n  }\n\n  if ($condition.rank !== 1) {\n    // A must have the same shape as condition.\n    assertShapesMatch(\n        $condition.shape, $broadcastedB.shape, 'Error in where: ');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.select($condition, $broadcastedA, $broadcastedB);\n    save([$condition]);\n    return res;\n  };\n  const inputs: SelectV2Inputs = {\n    condition: $condition,\n    t: $broadcastedA,\n    e: $broadcastedB\n  };\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* gradient */,\n             SelectV2) as T;\n}\n\nexport const where = op({where_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {ZerosLike, ZerosLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.zerosLike(x).print();\n * ```\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction zerosLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'zerosLike');\n  const inputs: ZerosLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             backend => backend.zerosLike($x), inputs as {} as NamedTensorMap,\n             null /* grad */, ZerosLike) as T;\n}\nexport const zerosLike = op({zerosLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {div} from './div';\nimport {equal} from './equal';\nimport {op} from './operation';\nimport {where} from './where';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting. Return 0\n * if denominator is 0.\n *\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 9, 16]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n * const c = tf.tensor1d([0, 0, 0, 0]);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * ```js\n * // Broadcast div a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(2);\n * const c = tf.scalar(0);\n *\n * a.divNoNan(b).print();  // or tf.divNoNan(a, b)\n * a.divNoNan(c).print();  // or tf.divNoNan(a, c)\n * ```\n *\n * @param a The first tensor as the numerator.\n * @param b The second tensor as the denominator. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction divNoNan_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  // TODO: Make this into its own kernel.\n  let $a = convertToTensor(a, 'a', 'div');\n  let $b = convertToTensor(b, 'b', 'div');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const divResult = div($a, $b);\n  const zeros = zerosLike(divResult);\n  const bEqualsZero = equal($b, zeros);\n  return where(bEqualsZero, zeros, divResult) as T;\n}\n\nexport const divNoNan = op({divNoNan_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor,} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices and/or vectors, `t1` and `t2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor2d([[1, 2], [3, 4]]);\n * const c = tf.tensor2d([[1, 2, 3], [4, 5, 6]]);\n *\n * a.dot(b).print();  // or tf.dot(a, b)\n * b.dot(a).print();\n * b.dot(c).print();\n * ```\n * @param t1 The first tensor in the dot operation.\n * @param t2 The second tensor in the dot operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction dot_(t1: Tensor|TensorLike, t2: Tensor|TensorLike): Tensor {\n  const $t1 = convertToTensor(t1, 't1', 'dot');\n  const $t2 = convertToTensor(t2, 't2', 'dot');\n\n  util.assert(\n      ($t1.rank === 1 || $t1.rank === 2) && ($t2.rank === 1 || $t2.rank === 2),\n      () => `Error in dot: inputs must all be rank 1 or 2, but got ranks ` +\n          `${$t1.rank} and ${$t2.rank}.`);\n\n  const t1Inner = ($t1.rank === 1 ? $t1.size : $t1.shape[1]);\n  const t2Inner = ($t2.rank === 1 ? $t2.size : $t2.shape[0]);\n\n  util.assert(\n      t1Inner === t2Inner,\n      () => `Error in dot: inner dimensions of inputs must match, but got ` +\n          `${t1Inner} and ${t2Inner}.`);\n\n  if ($t1.rank === 1 && $t2.rank === 1) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, []);\n  } else if ($t1.rank === 1 && $t2.rank === 2) {\n    const t12D = reshape($t1, [1, -1]);\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul(t12D, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else if ($t1.rank === 2 && $t2.rank === 1) {\n    const t22D = reshape($t2, [-1, 1]);\n    const t1t2 = matMul($t1, t22D);\n    return reshape(t1t2, [t1t2.size]);\n  } else {\n    const t22D = reshape($t2, [$t2.shape[0], $t2.shape[1]]);\n    const t1t2 = matMul($t1, t22D);\n    return t1t2;\n  }\n}\n\nexport const dot = op({dot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential linear element-wise: `x > 0 ? e ^ x - 1 : 0`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 1, -3, 2]);\n *\n * x.elu().print();  // or tf.elu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction elu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'elu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.elu($x);\n    save([y]);\n    return y;\n  };\n\n  const inputs: EluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Elu) as\n      T;\n}\n\nexport const elu = op({elu_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Erf, ErfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes gause error function of the input `tf.Tensor` element-wise:\n * `erf(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, .1, -.1, .7]);\n *\n * x.erf().print(); // or tf.erf(x);\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction erf_<T extends Tensor>(x: T|TensorLike): T {\n  let $x = convertToTensor(x, 'x', 'erf');\n  util.assert(\n      $x.dtype === 'int32' || $x.dtype === 'float32',\n      () => 'Input dtype must be `int32` or `float32`.');\n\n  if ($x.dtype === 'int32') {\n    $x = cast($x, 'float32');\n  }\n\n  const inputs: ErfInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.erf($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Erf);\n}\nexport const erf = op({erf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Exp, ExpInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.exp().print();  // or tf.exp(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction exp_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'exp');\n\n  const inputs: ExpInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.exp($x);\n    save([res]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Exp);\n}\nexport const exp = op({exp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension\n * into the tensor's shape.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const axis = 1;\n * x.expandDims(axis).print();\n * ```\n *\n * @param x The input tensor whose dimensions to be expanded.\n * @param axis The dimension index at which to insert shape of `1`. Defaults\n *     to 0 (the first dimension).\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction expandDims_<T extends Tensor>(x: Tensor|TensorLike, axis = 0): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'expandDims', parseAs);\n\n  util.assert(axis <= $x.rank, () => 'Axis must be <= rank of the tensor');\n  const newShape = $x.shape.slice();\n  if (axis < 0) {\n    // Negative value is counted from the tail of rank.\n    util.assert(\n        -($x.rank + 1) <= axis,\n        () => `Axis must be in the interval [${- ($x.rank + 1)}, ${$x.rank}]`);\n    axis = $x.rank + axis + 1;\n  }\n  newShape.splice(axis, 0, 1);\n  return reshape($x, newShape) as T;\n}\n\nexport const expandDims = op({expandDims_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Expm1, Expm1Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes exponential of the input `tf.Tensor` minus one element-wise.\n * `e ^ x - 1`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, -3]);\n *\n * x.expm1().print();  // or tf.expm1(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction expm1_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'expm1');\n\n  const inputs: Expm1Inputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.expm1($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Expm1);\n}\nexport const expm1 = op({expm1_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Tile, TileAttrs, TileInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Construct a tensor by repeating it the number of times given by reps.\n *\n * This operation creates a new tensor by replicating `input` `reps`\n * times. The output tensor's i'th dimension has `input.shape[i] *\n * reps[i]` elements, and the values of `input` are replicated\n * `reps[i]` times along the i'th dimension. For example, tiling\n * `[a, b, c, d]` by `[2]` produces `[a, b, c, d, a, b, c, d]`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n *\n * a.tile([2]).print();    // or a.tile([2])\n * ```\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.tile([1, 2]).print();  // or a.tile([1, 2])\n * ```\n * @param x The tensor to tile.\n * @param reps Determines the number of replications per dimension.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction tile_<T extends Tensor>(x: T|TensorLike, reps: number[]): T {\n  const parseAs: DataType = null;\n  const $x = convertToTensor(x, 'x', 'tile', parseAs);\n  util.assert(\n      $x.rank === reps.length,\n      () => `Error in transpose: rank of input ${$x.rank} ` +\n          `must match length of reps ${reps}.`);\n\n  const forward: ForwardFunc<T> = (backend, save) => {\n    const res = backend.tile($x, reps);\n    save([$x]);\n    return res;\n  };\n\n  const inputsToSave = [$x];\n  const inputs: TileInputs = {x: $x};\n  const attrs: TileAttrs = {reps};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, Tile,\n      attrs as unknown as NamedAttrMap, inputsToSave);\n}\n\nexport const tile = op({tile_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {DataType} from '../types';\n\nimport {buffer} from './buffer';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {tile} from './tile';\n\n/**\n * Create an identity matrix.\n *\n * @param numRows Number of rows.\n * @param numColumns Number of columns. Defaults to `numRows`.\n * @param batchShape If provided, will add the batch shape to the beginning\n *   of the shape of the returned `tf.Tensor` by repeating the identity\n *   matrix.\n * @param dtype Data type.\n * @returns Identity matrix of the specified size and data type, possibly\n *   with batch repetition if `batchShape` is specified.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction eye_(\n    numRows: number, numColumns?: number,\n    batchShape?:\n        [\n          number\n        ]|[number,\n           number]|[number, number, number]|[number, number, number, number],\n    dtype: DataType = 'float32'): Tensor2D {\n  if (numColumns == null) {\n    numColumns = numRows;\n  }\n  const buff = buffer([numRows, numColumns], dtype);\n  const n = numRows <= numColumns ? numRows : numColumns;\n  for (let i = 0; i < n; ++i) {\n    buff.set(1, i, i);\n  }\n  const out: Tensor2D = reshape(buff.toTensor(), [numRows, numColumns]);\n  if (batchShape == null) {\n    return out;\n  } else {\n    if (batchShape.length === 1) {\n      return tile(expandDims(out, 0), [batchShape[0], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 2) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(\n                 expandDims(expandDims(out, 0), 0),\n                 [batchShape[0], batchShape[1], 1, 1]) as Tensor2D;\n    } else if (batchShape.length === 3) {\n      // tslint:disable-next-line:no-unnecessary-type-assertion\n      return tile(expandDims(expandDims(expandDims(out, 0), 0), 0), [\n               batchShape[0], batchShape[1], batchShape[2], 1, 1\n             ]) as Tensor2D;\n    } else {\n      throw new Error(\n          `eye() currently supports only 1D and 2D ` +\n          // tslint:disable-next-line:no-any\n          `batchShapes, but received ${(batchShape as any).length}D.`);\n    }\n  }\n}\n\nexport const eye = op({eye_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Fill, FillAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\n/**\n * Creates a `tf.Tensor` filled with a scalar value.\n *\n * ```js\n * tf.fill([2, 2], 4).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param value The scalar value to fill the tensor with.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n * 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction fill<R extends Rank>(\n    shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n  const attrs: FillAttrs = {shape, value, dtype};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.fill(shape, value, dtype), {}, null, Fill,\n      attrs as {} as NamedAttrMap);\n}\n\nexport {fill};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Floor, FloorInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes floor of input `tf.Tensor` element-wise: `floor(x)`.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.floor().print();  // or tf.floor(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction floor_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'floor');\n\n  const inputs: FloorInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.floor($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Floor);\n}\nexport const floor = op({floor_});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inputs of size above this threshold will be parallelized by calling multiple\n * shader programs.\n */\nimport {nearestDivisor} from '../util';\n\nexport const PARALLELIZE_THRESHOLD = 30;\n\nexport interface ReduceInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  outSize: number;\n}\n\nexport function computeOptimalWindowSize(inSize: number): number {\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    return inSize;\n  }\n  return nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {nearestDivisor} from '../util';\n\nimport {PARALLELIZE_THRESHOLD} from './reduce_util';\n\nexport interface SegOpInfo {\n  windowSize: number;\n  batchSize: number;\n  inSize: number;\n  numSegments: number;\n}\n\nexport function segOpComputeOptimalWindowSize(\n    inSize: number, numSegments: number): number {\n  let done = false;\n  let res;\n\n  if (inSize <= PARALLELIZE_THRESHOLD) {\n    res = inSize;\n    done = true;\n  } else {\n    res = nearestDivisor(inSize, Math.floor(Math.sqrt(inSize)));\n  }\n\n  while (!done) {\n    if (res > numSegments || res === inSize) {\n      done = true;\n    } else {\n      res = nearestDivisor(inSize, res + 1);\n    }\n  }\n  return res;\n}\n\nexport function computeOutShape(\n    aShape: number[], axis: number, numSegments: number): number[] {\n  const outShape = [];\n  const rank = aShape.length;\n  for (let dim = 0; dim < rank; dim++) {\n    if (dim !== axis) {\n      outShape.push(aShape[dim]);\n    } else {\n      outShape.push(numSegments);\n    }\n  }\n  return outShape;\n}\n\nexport interface GatherOpShapeInfo {\n  batchSize: number;\n  sliceSize: number;\n  dimSize: number;\n  outputShape: number[];\n}\nexport function collectGatherOpShapeInfo(\n    x: Tensor, indices: Tensor, axis: number): GatherOpShapeInfo {\n  const dimSize = x.shape[axis];\n\n  const outputShape: number[] = [];\n  let batchSize = 1;\n  let sliceSize = 1;\n  for (let i = 0; i < axis; i++) {\n    outputShape.push(x.shape[i]);\n    batchSize *= x.shape[i];\n  }\n\n  for (let i = 0; i < indices.rank; i++) {\n    outputShape.push(indices.shape[i]);\n  }\n\n  for (let i = axis + 1; i < x.rank; i++) {\n    outputShape.push(x.shape[i]);\n    sliceSize *= x.shape[i];\n  }\n\n  return {batchSize, sliceSize, dimSize, outputShape};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherV2, GatherV2Attrs, GatherV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {collectGatherOpShapeInfo} from './segment_util';\n\n/**\n * Gather slices from tensor `x`'s axis `axis` according to `indices`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const indices = tf.tensor1d([1, 3, 3], 'int32');\n *\n * x.gather(indices).print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const indices = tf.tensor1d([1, 1, 0], 'int32');\n *\n * x.gather(indices).print();\n * ```\n * @param x The input tensor whose slices to be gathered.\n * @param indices The indices of the values to extract.\n * @param axis The axis over which to select values. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction gather_<T extends Tensor>(\n    x: T|TensorLike, indices: Tensor|TensorLike, axis = 0): T {\n  const $x = convertToTensor(x, 'x', 'gather');\n  const $indices = convertToTensor(indices, 'indices', 'gather', 'int32');\n\n  const inputs: GatherV2Inputs = {x: $x, indices: $indices};\n  const attrs: GatherV2Attrs = {axis};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const parsedAxis = parseAxisParam(axis, $x.shape)[0];\n    const shapeInfo = collectGatherOpShapeInfo($x, $indices, parsedAxis);\n\n    const res =\n        backend.gather($x, reshape($indices, [$indices.size]), parsedAxis);\n    save([$x, $indices]);\n\n    return reshape(res, shapeInfo.outputShape);\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, GatherV2,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const gather = op({gather_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Greater, GreaterInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a > b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greater(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greater_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greater');\n  let $b = convertToTensor(b, 'b', 'greater');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.greater($a, $b);\n\n  const inputs: GreaterInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             Greater) as T;\n}\n\nexport const greater = op({greater_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GreaterEqual, GreaterEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a >= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.greaterEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction greaterEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'greaterEqual');\n  let $b = convertToTensor(b, 'b', 'greaterEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.greaterEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: GreaterEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             GreaterEqual) as T;\n}\n\nexport const greaterEqual = op({greaterEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Imag, ImagInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n/**\n * Returns the imaginary part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the imaginary part of each element in input considered as a complex number.\n * If input is real, a tensor of all zeros is returned.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.imag(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction imag_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'imag');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.imag($input);\n  };\n\n  const inputs: ImagInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Imag) as T;\n}\n\nexport const imag = op({imag_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsFinite, IsFiniteInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are finite.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isFinite().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isFinite_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isFinite');\n\n  const inputs: IsFiniteInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isFinite($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsFinite);\n}\nexport const isFinite = op({isFinite_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsInf, IsInfInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns which elements of x are Infinity or -Infinity.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isInf().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isInf_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isInf');\n\n  const inputs: IsInfInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.isInf($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsInf);\n}\nexport const isInf = op({isInf_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {IsNan, IsNanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * RReturns which elements of x are NaN.\n *\n * ```js\n * const x = tf.tensor1d([NaN, Infinity, -Infinity, 0, 1]);\n *\n * x.isNaN().print();  // or tf.isNaN(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction isNaN_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'isNaN');\n  const inputs: IsNanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.isNaN($x), inputs as {} as NamedTensorMap,\n      null /* grad */, IsNan);\n}\nexport const isNaN = op({isNaN_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Maximum, MaximumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the max of a and b (`a > b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `tf.maximumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * ```js\n * // Broadcast maximum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.maximum(b).print();  // or tf.maximum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction maximum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'maximum');\n  let $b = convertToTensor(b, 'b', 'maximum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.maximum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: MaximumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Maximum) as T;\n}\n\nexport const maximum = op({maximum_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar} from '../tensor';\nimport {DataType} from '../types';\nimport {isTypedArray} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-0 `tf.Tensor` (scalar) with the provided value and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.scalar` as it makes the code more readable.\n *\n * ```js\n * tf.scalar(3.14).print();\n * ```\n *\n * @param value The value of the scalar.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function scalar(\n    value: number|boolean|string|Uint8Array, dtype?: DataType): Scalar {\n  if (((isTypedArray(value) && dtype !== 'string') || Array.isArray(value)) &&\n      dtype !== 'complex64') {\n    throw new Error(\n        'Error creating a new Scalar: value must be a primitive ' +\n        '(number|boolean|string)');\n  }\n  if (dtype === 'string' && isTypedArray(value) &&\n      !(value instanceof Uint8Array)) {\n    throw new Error(\n        'When making a scalar from encoded string, ' +\n        'the value must be `Uint8Array`.');\n  }\n  const shape: number[] = [];\n  const inferredShape: number[] = [];\n  return makeTensor(value, shape, inferredShape, dtype) as Scalar;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {maximum} from './maximum';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {scalar} from './scalar';\n\n/**\n * Computes leaky rectified linear element-wise.\n *\n * See\n * [http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf](\n *     http://web.stanford.edu/~awni/papers/relu_hybrid_icml2013_final.pdf)\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.leakyRelu(0.1).print();  // or tf.leakyRelu(x, 0.1)\n * ```\n * @param x The input tensor.\n * @param alpha The scaling factor for negative values, defaults to 0.2.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction leakyRelu_<T extends Tensor>(x: T|TensorLike, alpha = 0.2): T {\n  const $x = convertToTensor(x, 'x', 'leakyRelu');\n  return maximum(mul(scalar(alpha), $x), $x);\n}\n\nexport const leakyRelu = op({leakyRelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Less, LessInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a < b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.less(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction less_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'less');\n  let $b = convertToTensor(b, 'b', 'less');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = backend => backend.less($a, $b);\n\n  const inputs: LessInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Less) as\n      T;\n}\n\nexport const less = op({less_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LessEqual, LessEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a <= b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([2, 2, 2]);\n *\n * a.lessEqual(b).print();\n * ```\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction lessEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'lessEqual');\n  let $b = convertToTensor(b, 'b', 'lessEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.lessEqual($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: LessEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LessEqual) as T;\n}\n\nexport const lessEqual = op({lessEqual_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LinSpace, LinSpaceAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor1D} from '../tensor';\n\n/**\n * Return an evenly spaced sequence of numbers over the given interval.\n *\n * ```js\n * tf.linspace(0, 9, 10).print();\n * ```\n * @param start The start value of the sequence.\n * @param stop The end value of the sequence.\n * @param num The number of values to generate.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function linspace(start: number, stop: number, num: number): Tensor1D {\n  if (num <= 0) {\n    throw new Error('The number of values should be positive.');\n  }\n\n  const attrs: LinSpaceAttrs = {start, stop, num};\n  return ENGINE.runKernelFunc(\n      backend => backend.linspace(start, stop, num), {} /* inputs */,\n      null /* grad */, LinSpace, attrs as {} as NamedAttrMap);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRN, LRNAttrs, LRNInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Normalizes the activation of a local neighborhood across or within\n * channels.\n *\n * @param x The input tensor. The 4-D input tensor is treated as a 3-D array\n *     of 1D vectors (along the last dimension), and each vector is\n *     normalized independently.\n * @param depthRadius The number of adjacent channels in the 1D normalization\n *     window.\n * @param bias A constant bias term for the basis.\n * @param alpha A scale factor, usually positive.\n * @param beta An exponent.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction localResponseNormalization_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const $x = convertToTensor(x, 'x', 'localResponseNormalization');\n  util.assert(\n      $x.rank === 4 || $x.rank === 3,\n      () => `Error in localResponseNormalization: x must be rank 3 or 4 but got\n               rank ${$x.rank}.`);\n  util.assert(\n      util.isInt(depthRadius),\n      () => `Error in localResponseNormalization: depthRadius must be an ` +\n          `integer but got depthRadius ${depthRadius}.`);\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.localResponseNormalization4D(\n        x4D, depthRadius, bias, alpha, beta);\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: LRNInputs = {x: x4D};\n\n  const attrs: LRNAttrs = {depthRadius, bias, alpha, beta};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, LRN,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  } else {\n    return res as T;\n  }\n}\n\nexport const localResponseNormalization = op({localResponseNormalization_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log, LogInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` element-wise: `ln(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E]);\n *\n * x.log().print();  // or tf.log(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log');\n\n  const inputs: LogInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log);\n}\nexport const log = op({log_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Log1p, Log1pInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes natural logarithm of the input `tf.Tensor` plus one\n * element-wise: `ln(1 + x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.E - 1]);\n *\n * x.log1p().print();  // or tf.log1p(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction log1p_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'log1p');\n\n  const inputs: Log1pInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.log1p($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Log1p);\n}\nexport const log1p = op({log1p_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {CustomGradientFunc, ENGINE} from './engine';\nimport {Scalar, Tensor, Variable} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {convertToTensor, convertToTensorArray} from './tensor_util_env';\nimport {TensorLike} from './types';\nimport * as util from './util';\n\n/**\n * Provided `f(x)`, returns another function `g(x, dy?)`, which gives the\n * gradient of `f(x)` with respect to `x`.\n *\n * If `dy` is provided, the gradient of `f(x).mul(dy).sum()` with respect to\n * `x` is computed instead. `f(x)` must take a single tensor `x` and return a\n * single tensor `y`. If `f()` takes multiple inputs, use `tf.grads` instead.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.grad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * g(x).print();\n * ```\n *\n * ```js\n * // f(x) = x ^ 3\n * const f = x => x.pow(tf.scalar(3, 'int32'));\n * // f'(x) = 3x ^ 2\n * const g = tf.grad(f);\n * // f''(x) = 6x\n * const gg = tf.grad(g);\n *\n * const x = tf.tensor1d([2, 3]);\n * gg(x).print();\n * ```\n *\n * @param f The function f(x), to compute gradient for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grad(f: (x: Tensor) => Tensor): (\n    x: TensorLike|Tensor, dy?: TensorLike|Tensor) => Tensor {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grad(f) must be a function');\n  return (x: TensorLike|Tensor, dy?: TensorLike|Tensor): Tensor => {\n    // x can be of any dtype, thus null as the last argument.\n    const $x = convertToTensor(x, 'x', 'tf.grad', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grad') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f($x), [$x], $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grad(f)(x, dy) must match the shape ' +\n                'returned by f(x)');\n      }\n      checkGrads(grads);\n      return grads[0];\n    });\n  };\n}\n\n/**\n * Provided `f(x1, x2,...)`, returns another function `g([x1, x2,...], dy?)`,\n * which gives an array of gradients of `f()` with respect to each input\n * [`x1`,`x2`,...].\n *\n * If `dy` is passed when calling `g()`, the gradient of\n * `f(x1,...).mul(dy).sum()` with respect to each input is computed instead.\n * The provided `f` must take one or more tensors and return a single tensor\n * `y`. If `f()` takes a single input, we recommend using `tf.grad` instead.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df / da = b, df / db = a\n * const g = tf.grads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const [da, db] = g([a, b]);\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @param f The function `f(x1, x2,...)` to compute gradients for.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction grads(f: (...args: Tensor[]) => Tensor): (\n    args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike) => Tensor[] {\n  util.assert(\n      util.isFunction(f), () => 'The f passed in grads(f) must be a function');\n  return (args: Array<Tensor|TensorLike>, dy?: Tensor|TensorLike): Tensor[] => {\n    util.assert(\n        Array.isArray(args),\n        () => 'The args passed in grads(f)(args) must be an array ' +\n            'of `Tensor`s or `TensorLike`s');\n    // args can be of any dtype, thus null as the last argument.\n    const $args = convertToTensorArray(args, 'args', 'tf.grads', null);\n    const $dy: Tensor =\n        (dy != null) ? convertToTensor(dy, 'dy', 'tf.grads') : null;\n    return ENGINE.tidy(() => {\n      const {value, grads} = ENGINE.gradients(() => f(...$args), $args, $dy);\n      if ($dy != null) {\n        util.assertShapesMatch(\n            value.shape, $dy.shape,\n            'The shape of dy passed in grads(f)([x1,...], dy) must ' +\n                'match the shape returned by f([x1,...])');\n      }\n      checkGrads(grads);\n      return grads;\n    });\n  };\n}\n\n/**\n * Like `tf.grad`, but also returns the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grad: The gradient of `f(x)` w.r.t `x` (result of `tf.grad`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(x) = x ^ 2\n * const f = x => x.square();\n * // f'(x) = 2x\n * const g = tf.valueAndGrad(f);\n *\n * const x = tf.tensor1d([2, 3]);\n * const {value, grad} = g(x);\n *\n * console.log('value');\n * value.print();\n * console.log('grad');\n * grad.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrad<I extends Tensor, O extends Tensor>(f: (x: I) => O): (\n    x: I, dy?: O) => {\n  value: O;\n  grad: I;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrad(f) must be a function');\n  return (x: I, dy?: O) => {\n    util.assert(\n        x instanceof Tensor,\n        () => 'The x passed in valueAndGrad(f)(x) must be a tensor');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrad(f)(x, dy) must be a tensor');\n    const {grads, value} = ENGINE.gradients(() => f(x), [x], dy);\n    checkGrads(grads);\n    return {grad: grads[0] as I, value};\n  };\n}\n\n/**\n * Like `tf.grads`, but returns also the value of `f()`. Useful when `f()`\n * returns a metric you want to show.\n *\n * The result is a rich object with the following properties:\n * - grads: The gradients of `f()` w.r.t each input (result of `tf.grads`).\n * - value: The value returned by `f(x)`.\n *\n * ```js\n * // f(a, b) = a * b\n * const f = (a, b) => a.mul(b);\n * // df/da = b, df/db = a\n * const g = tf.valueAndGrads(f);\n *\n * const a = tf.tensor1d([2, 3]);\n * const b = tf.tensor1d([-2, -3]);\n * const {value, grads} = g([a, b]);\n *\n * const [da, db] = grads;\n *\n * console.log('value');\n * value.print();\n *\n * console.log('da');\n * da.print();\n * console.log('db');\n * db.print();\n * ```\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction valueAndGrads<O extends Tensor>(f: (...args: Tensor[]) => O): (\n    args: Tensor[], dy?: O) => {\n  grads: Tensor[];\n  value: O;\n} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in valueAndGrads(f) must be a function');\n  return (args: Tensor[], dy?: O) => {\n    util.assert(\n        Array.isArray(args) && args.every(arg => arg instanceof Tensor),\n        () => 'The args passed in valueAndGrads(f)(args) must be array of ' +\n            'tensors');\n    util.assert(\n        dy == null || dy instanceof Tensor,\n        () => 'The dy passed in valueAndGrads(f)(args, dy) must be a tensor');\n    const res = ENGINE.gradients(() => f(...args), args, dy);\n    if (dy != null) {\n      util.assertShapesMatch(\n          res.value.shape, dy.shape,\n          'The shape of dy passed in valueAndGrads(f)([x1,...], dy) must ' +\n              'match the shape returned by f([x1,...])');\n    }\n    checkGrads(res.grads);\n    return res;\n  };\n}\n\n/**\n * Computes and returns the gradient of f(x) with respect to the list of\n * trainable variables provided by `varList`. If no list is provided, it\n * defaults to all trainable variables.\n *\n * ```js\n * const a = tf.variable(tf.tensor1d([3, 4]));\n * const b = tf.variable(tf.tensor1d([5, 6]));\n * const x = tf.tensor1d([1, 2]);\n *\n * // f(a, b) = a * x ^ 2 + b * x\n * const f = () => a.mul(x.square()).add(b.mul(x)).sum();\n * // df/da = x ^ 2, df/db = x\n * const {value, grads} = tf.variableGrads(f);\n *\n * Object.keys(grads).forEach(varName => grads[varName].print());\n * ```\n *\n * @param f The function to execute. f() should return a scalar.\n * @param varList The list of variables to compute the gradients with respect\n *     to. Defaults to all trainable variables.\n * @returns An object with the following keys and values:\n *   - `value`: The value of the function `f`.\n *   - `grads`: A map from the names of the variables to the gradients.\n *     If the `varList` argument is provided explicitly and contains a subset of\n *     non-trainable variables, this map in the return value will contain keys\n *     that map the names of the non-trainable variables to `null`.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction variableGrads(f: () => Scalar, varList?: Variable[]):\n    {value: Scalar, grads: NamedTensorMap} {\n  util.assert(\n      util.isFunction(f),\n      () => 'The f passed in variableGrads(f) must be a function');\n  util.assert(\n      varList == null ||\n          Array.isArray(varList) && varList.every(v => v instanceof Variable),\n      () =>\n          'The varList passed in variableGrads(f, varList) must be an array ' +\n          'of variables');\n\n  const specifiedVarList = varList != null;\n  if (!specifiedVarList) {\n    // Get all of the trainable variables.\n    varList = [];\n    for (const varName in ENGINE.registeredVariables) {\n      varList.push(ENGINE.registeredVariables[varName]);\n    }\n  }\n\n  const specifiedNonTrainable: Variable[] =\n      specifiedVarList ? varList.filter(variable => !variable.trainable) : null;\n\n  // Prune non-trainable variables.\n  const originalVarCount = varList.length;\n  varList = varList.filter(variable => variable.trainable);\n  util.assert(\n      varList.length > 0,\n      () => `variableGrads() expects at least one of the input variables to ` +\n          `be trainable, but none of the ${originalVarCount} variables is ` +\n          `trainable.`);\n\n  const allowNoGradients = true;\n  const {value, grads} = ENGINE.gradients(f, varList, null, allowNoGradients);\n\n  util.assert(\n      grads.some(g => g != null),\n      () => 'Cannot find a connection between any variable and the result of ' +\n          'the loss function y=f(x). Please make sure the operations that ' +\n          'use variables are inside the function f passed to minimize().');\n  util.assert(\n      value.rank === 0,\n      () => `The f passed in variableGrads(f) must return a scalar, but it ` +\n          `returned a rank-${value.rank} tensor`);\n\n  const namedGrads: NamedTensorMap = {};\n  varList.forEach((v, i) => {\n    if (grads[i] != null) {\n      namedGrads[v.name] = grads[i];\n    }\n  });\n  if (specifiedNonTrainable != null) {\n    // If varList is explicitly provided and contains non-trainable values,\n    // add them to the returned gradients with `null` values.\n    specifiedNonTrainable.forEach(v => namedGrads[v.name] = null);\n  }\n  return {value, grads: namedGrads};\n}\n\n/**\n * Overrides the gradient computation of a function `f`.\n *\n * Takes a function\n * `f(...inputs, save) => {value: Tensor, gradFunc: (dy, saved) => Tensor[]}`\n * and returns another function `g(...inputs)` which takes the same inputs as\n * `f`. When called, `g` returns `f().value`. In backward mode, custom gradients\n * with respect to each input of `f` are computed using `f().gradFunc`.\n *\n * The `save` function passsed to `f` should be used for saving tensors needed\n * in the gradient. And the `saved` passed to the `gradFunc` is a\n * `NamedTensorMap`, which contains those saved tensor.\n *\n * ```js\n * const customOp = tf.customGrad((x, save) => {\n *   // Save x to make sure it's available later for the gradient.\n *   save([x]);\n *   // Override gradient of our custom x ^ 2 op to be dy * abs(x);\n *   return {\n *     value: x.square(),\n *     // Note `saved.x` which points to the `x` we saved earlier.\n *     gradFunc: (dy, saved) => [dy.mul(saved[0].abs())]\n *   };\n * });\n *\n * const x = tf.tensor1d([-1, -2, 3]);\n * const dx = tf.grad(x => customOp(x));\n *\n * console.log(`f(x):`);\n * customOp(x).print();\n * console.log(`f'(x):`);\n * dx(x).print();\n * ```\n *\n * @param f The function to evaluate in forward mode, which should return\n *     `{value: Tensor, gradFunc: (dy, saved) => Tensor[]}`, where `gradFunc`\n *     returns the custom gradients of `f` with respect to its inputs.\n *\n * @doc {heading: 'Training', subheading: 'Gradients'}\n */\nfunction customGrad<T extends Tensor>(f: CustomGradientFunc<T>):\n    (...args: Tensor[]) => T {\n  return ENGINE.customGrad(f);\n}\n\nfunction checkGrads(grads: Tensor[]) {\n  const numNullGradients = grads.filter(g => g == null).length;\n  if (numNullGradients > 0) {\n    throw new Error(\n        `Cannot compute gradient of y=f(x) with respect to x. Make sure that\n    the f you passed encloses all operations that lead from x to y.`);\n  }\n}\n\nexport {\n  customGrad,\n  variableGrads,\n  valueAndGrad,\n  valueAndGrads,\n  grad,\n  grads,\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Negate, NegateInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes `-1 * x` element-wise.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, -2, 0], [2, 2]);\n *\n * x.neg().print();  // or tf.neg(x)\n * ```\n *\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction neg_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'neg');\n\n  const inputs: NegateInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.neg($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Negate);\n}\nexport const neg = op({neg_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softplus, SoftplusInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes softplus of the input `tf.Tensor` element-wise: `log(exp(x) + 1)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.softplus().print();  // or tf.softplus(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction softplus_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'softplus');\n\n  const inputs: SoftplusInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.softplus($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Softplus);\n}\nexport const softplus = op({softplus_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {mul} from './mul';\nimport {neg} from './neg';\nimport {op} from './operation';\nimport {sigmoid} from './sigmoid';\nimport {softplus} from './softplus';\n\n/**\n * Computes log sigmoid of the input `tf.Tensor` element-wise:\n * `logSigmoid(x)`. For numerical stability, we use `-tf.softplus(-x)`.\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.logSigmoid().print();  // or tf.logSigmoid(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction logSigmoid_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logSigmoid');\n\n  // Use a custom gradient to maintain previous implementation.\n  // There is no LogSigmoid kernel in TF so we can't use engine.runKernel\n  // directly\n  const customOp = customGrad((x: Tensor) => {\n    // TODO(yassogba) we can remove the chained softplus call here only\n    // after backends have modualrized softplus at which point we can call\n    // engine runKernel(..., Sotfplus, ...) directly.\n    const value = neg(softplus(neg(x)));\n\n    const gradFunc = (dy: T) => {\n      const derX = mul(dy, sigmoid(neg(x)));\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\nexport const logSigmoid = op({logSigmoid_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Max, MaxAttrs, MaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the maximum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and an\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.max().print();  // or tf.max(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.max(axis).print();  // or tf.max(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction max_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'max');\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = util.parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let maxInput = $x;\n        if (permutedAxes != null) {\n          maxInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, maxInput.rank);\n        }\n\n        const y = backend.max(maxInput, axes);\n        if (permutedAxes != null) {\n          maxInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape = axis_util.expandShapeToKeepDim(\n              res.shape, util.parseAxisParam(axis, $x.shape));\n          res = reshape(res, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MaxInputs = {x: $x};\n  const attrs: MaxAttrs = {reductionIndices: axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Max,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const max = op({max_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sub, SubInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Subtracts two `tf.Tensor`s element-wise, A - B. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.tensor1d([1, 2, 3, 4]);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n *\n * ```js\n * // Broadcast subtract a with b.\n * const a = tf.tensor1d([10, 20, 30, 40]);\n * const b = tf.scalar(5);\n *\n * a.sub(b).print();  // or tf.sub(a, b)\n * ```\n * @param a The first `tf.Tensor` to subtract from.\n * @param b The second `tf.Tensor` to be subtracted. Must have the same dtype as\n * `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction sub_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'sub');\n  let $b = convertToTensor(b, 'b', 'sub');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.subtract($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SubInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sub) as\n      T;\n}\n\nexport const sub = op({sub_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Sum, SumAttrs, SumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the sum of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If axes has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.sum().print();  // or tf.sum(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.sum(axis).print();  // or tf.sum(x, axis)\n * ```\n *\n * @param x The input tensor to compute the sum over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction sum_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'sum');\n  if ($x.dtype === 'bool') {\n    $x = cast($x, 'int32');\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.sum(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n    return value;\n  };\n\n  const inputs: SumInputs = {x: $x};\n  const attrs: SumAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Sum,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const sum = op({sum_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LogSoftmax, LogSoftmaxAttrs, LogSoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log softmax.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.logSoftmax().print();  // or tf.logSoftmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param axis The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction logSoftmax_<T extends Tensor>(logits: T|TensorLike, axis = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'logSoftmax');\n\n  if (axis === -1) {\n    axis = $logits.rank - 1;\n  }\n  if (axis !== $logits.rank - 1) {\n    throw Error(\n        'Log Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and axis was ${axis}`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const keepDims = true;\n    const xMax = max(logits, axis, true);\n    const shifted = sub(logits, xMax);\n    const value =\n        sub(cast(shifted, 'float32'), log(sum(exp(shifted), axis, keepDims)));\n    save([value]);\n    return value;\n  };\n\n  const inputs: LogSoftmaxInputs = {logits: $logits};\n  const attrs: LogSoftmaxAttrs = {axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LogSoftmax, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const logSoftmax = op({logSoftmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {add} from './add';\nimport {expandShapeToKeepDim} from './axis_util';\nimport {exp} from './exp';\nimport {log} from './log';\nimport {max} from './max';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {sub} from './sub';\nimport {sum} from './sum';\n\n/**\n * Computes the log(sum(exp(elements across the reduction dimensions)).\n *\n * Reduces the input along the dimensions given in `axis`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.logSumExp().print();  // or tf.logSumExp(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.logSumExp(axis).print();  // or tf.logSumExp(a, axis)\n * ```\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. If null (the default),\n *     reduces all dimensions.\n * @param keepDims If true, retains reduced dimensions with length\n *     of 1. Defaults to false.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction logSumExp_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'logSumExp');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const xMax = max($x, axes, true /* keepDims */);\n  const a = sub($x, xMax);\n  const b = exp(a);\n  const c = sum(b, axes);\n  const d = log(c);\n  const res = add(reshape(xMax, d.shape), d);\n\n  if (keepDims) {\n    const newShape = expandShapeToKeepDim(res.shape, axes);\n    return reshape(res, newShape) as T;\n  }\n  return res as T;\n}\n\nexport const logSumExp = op({logSumExp_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {LogicalAnd, LogicalAndInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a AND b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalAnd(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalAnd_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalAnd', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalAnd', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalAndInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalAnd($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalAnd) as T;\n}\n\nexport const logicalAnd = op({logicalAnd_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalNot, LogicalNotInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `NOT x` element-wise.\n *\n * ```js\n * const a = tf.tensor1d([false, true], 'bool');\n *\n * a.logicalNot().print();\n * ```\n *\n * @param x The input tensor. Must be of dtype 'bool'.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalNot_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'logicalNot', 'bool');\n  const inputs: LogicalNotInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.logicalNot($x), inputs as {} as NamedTensorMap,\n      null /* grad */, LogicalNot);\n}\n\nexport const logicalNot = op({logicalNot_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {LogicalOr, LogicalOrInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a OR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalOr(b).print();\n * ```\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalOr_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalOr', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalOr', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const inputs: LogicalOrInputs = {a: $a, b: $b};\n  return ENGINE.runKernelFunc(\n             backend => backend.logicalOr($a, $b),\n             inputs as {} as NamedTensorMap, null /* grad */, LogicalOr) as T;\n}\nexport const logicalOr = op({logicalOr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {logicalAnd} from './logical_and';\nimport {logicalNot} from './logical_not';\nimport {logicalOr} from './logical_or';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of `a XOR b` element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([false, false, true, true], 'bool');\n * const b = tf.tensor1d([false, true, false, true], 'bool');\n *\n * a.logicalXor(b).print();\n * ```\n *\n * @param a The first input tensor. Must be of dtype bool.\n * @param b The second input tensor. Must be of dtype bool.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction logicalXor_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  const $a = convertToTensor(a, 'a', 'logicalXor', 'bool');\n  const $b = convertToTensor(b, 'b', 'logicalXor', 'bool');\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  // x ^ y = (x | y) & ~(x & y)\n  return logicalAnd(logicalOr(a, b), logicalNot(logicalAnd(a, b)));\n}\n\nexport const logicalXor = op({logicalXor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool, MaxPoolAttrs, MaxPoolInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 2D max pooling of an image.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n */\nfunction maxPool_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $x = convertToTensor(x, 'x', 'maxPool');\n  const dilations = 1;\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in maxPool: input must be rank 4 but got rank ${x4D.rank}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const convInfo = conv_util.computePool2DInfo(\n        x4D.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    let y;\n\n    if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n        util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n      y = x4D.clone();\n    } else {\n      y = backend.maxPool(x4D, convInfo);\n    }\n\n    save([x4D, y]);\n\n    return y;\n  };\n\n  const inputs: MaxPoolInputs = {x: x4D};\n\n  const attrs: MaxPoolAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const maxPool = op({maxPool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {deprecationWarn} from '../globals';\nimport {MaxPool3D, MaxPool3DAttrs, MaxPool3DInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the 3D max pooling.\n *\n * ```js\n * const x = tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]);\n * const result = tf.maxPool3d(x, 2, 1, 'valid');\n * result.print();\n * ```\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     `[batch, depth, height, width, inChannels]`.\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     If `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`.\n *     If `strides` is a single number,\n *     then `strideDepth == strideHeight == strideWidth`.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1*1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPool3d_<T extends Tensor4D|Tensor5D>(\n    x: T|TensorLike, filterSize: [number, number, number]|number = [1, 1, 1],\n    strides: [number, number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil',\n    dataFormat: 'NDHWC'|'NCDHW' = 'NDHWC',\n    dilations?: [number, number, number]|number): T {\n  if (dilations == null) {\n    dilations = [1, 1, 1];\n  } else {\n    deprecationWarn(\n        'dilations is deprecated, this field will be gone in ' +\n        'v3.0.0.');\n  }\n\n  const $x = convertToTensor(x, 'x', 'maxPool3d');\n\n  let x5D = $x as Tensor5D;\n  let reshapedTo5D = false;\n  if ($x.rank === 4) {\n    reshapedTo5D = true;\n    x5D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2], $x.shape[3]]);\n  }\n\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in maxPool3d: x must be rank 5 but got rank ${x5D.rank}.`);\n  util.assert(\n      dataFormat === 'NDHWC',\n      () => `Error in maxPool3d: Only NDHWC is currently supported, ` +\n          `but got dataFormat of ${dataFormat}`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3d: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if (dilations == null) {\n      dilations = [1, 1, 1];\n    }\n    const convInfo = conv_util.computePool3DInfo(\n        x5D.shape, filterSize, strides, dilations, pad, dimRoundingMode,\n        dataFormat);\n    const y = backend.maxPool3d(x5D, convInfo);\n    save([x5D, y]);\n    return y;\n  };\n\n  const inputs: MaxPool3DInputs = {x: x5D};\n\n  const attrs: MaxPool3DAttrs =\n      {filterSize, strides, pad, dimRoundingMode, dataFormat, dilations};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, MaxPool3D,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3d = op({maxPool3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the 2D max pooling of an image with Argmax index.\n * The indices in argmax are flattened, so that a maximum value at position `[b,\n * y, x, c]` becomes flattened index: `(y * width + x) * channels + c` if\n * include_batch_in_index is False; `((b * height + y) * width + x) * channels\n * +c` if include_batch_in_index is True.\n *\n * The indices returned are always in `[0, height) x [0, width)` before\n * flattening.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dataFormat An optional string from: \"NDHWC\", \"NCDHW\". Defaults to\n *     \"NDHWC\". Specify the data format of the input and output data. With the\n *     default format \"NDHWC\", the data is stored in the order of: [batch,\n *     depth, height, width, channels]. Only \"NDHWC\" is currently supported.\n * @param pad The type of padding algorithm.\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param includeBatchIndex Defaults to False. Whether to include batch\n *    dimension in flattened index of argmax.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction maxPoolWithArgmax_<T extends Tensor4D>(\n    x: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    includeBatchInIndex = false): NamedTensorMap {\n  const $x = convertToTensor(x, 'x', 'maxPoolWithArgmax');\n\n  const inputs: MaxPoolWithArgmaxInputs = {x: $x};\n  const attrs:\n      MaxPoolWithArgmaxAttrs = {filterSize, strides, pad, includeBatchInIndex};\n\n  const result = ENGINE.runKernel(\n                     MaxPoolWithArgmax, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {result: result[0], indexes: result[1]};\n}\n\nexport const maxPoolWithArgmax = op({maxPoolWithArgmax_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeZerosTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 0.\n *\n * ```js\n * tf.zeros([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Can\n *     be 'float32', 'int32' or 'bool'. Defaults to 'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function zeros<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = zeros(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeZerosTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {makeOnesTypedArray, sizeFromShape} from '../util';\n\nimport {complex} from './complex';\nimport {zeros} from './zeros';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1.\n *\n * ```js\n * tf.ones([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The type of an element in the resulting tensor. Defaults to\n *     'float'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function ones<R extends Rank>(\n    shape: ShapeMap[R], dtype: DataType = 'float32'): Tensor<R> {\n  if (dtype === 'complex64') {\n    const real = ones(shape, 'float32');\n    const imag = zeros(shape, 'float32');\n    return complex(real, imag);\n  }\n  const values = makeOnesTypedArray(sizeFromShape(shape), dtype);\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../gradients';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam, sizeFromShape} from '../util';\n\nimport {computeOutAndReduceShapes} from './axis_util';\nimport {cast} from './cast';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {ones} from './ones';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sum} from './sum';\n\n/**\n * Computes the mean of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces `x` along the dimensions given in `axis`. Unless `keepDims` is\n * true, the rank of the `tf.Tensor` is reduced by 1 for each entry in `axis`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axis` has no entries, all dimensions are reduced, and a `tf.Tensor` with\n * a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.mean().print();  // or tf.mean(a)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.mean(axis).print();  // or tf.mean(x, axis)\n * ```\n *\n * @param x The input tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction mean_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'mean');\n\n  const axes = parseAxisParam(axis, $x.shape);\n  const shapes = computeOutAndReduceShapes($x.shape, axes);\n  const reduceShape = shapes[1];\n  const reduceSize = sizeFromShape(reduceShape);\n\n  // Use a custom gradient to bypass 2 gradient backprops since mean is used\n  // extremely often.\n  const customOp = customGrad((x: Tensor) => {\n    const reduceSizeScalar = scalar(reduceSize);\n    // Cast if needed.\n    const xReduce = reduceSizeScalar.dtype === x.dtype ?\n        x :\n        cast(x, reduceSizeScalar.dtype);\n    const res = div(xReduce, reduceSizeScalar);\n    const value = sum(res, axis, keepDims);\n\n    const gradFunc = (dy: Tensor) => {\n      const expandedDyShape = x.shape.slice();\n      axes.forEach(axis => {\n        expandedDyShape[axis] = 1;\n      });\n      const expandedDy = reshape(dy, expandedDyShape);\n      const derX = div(mul(expandedDy, ones(x.shape, 'float32')), reduceSize);\n      return derX;\n    };\n    return {value, gradFunc};\n  });\n\n  return customOp($x) as T;\n}\n\nexport const mean = op({mean_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Min, MinAttrs, MinInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport * as axis_util from './axis_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the minimum value from the input.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the array is reduced by 1 for each entry in `axes`.\n * If `keepDims` is true, the reduced dimensions are retained with length 1.\n * If `axes` has no entries, all dimensions are reduced, and an array with a\n * single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.min().print();  // or tf.min(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.min(axis).print();  // or tf.min(x, axis)\n * ```\n *\n * @param x The input Tensor.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction min_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  const $x = convertToTensor(x, 'x', 'min');\n\n  const forward: ForwardFunc<Tensor> =\n      (backend: KernelBackend, save: GradSaveFunc) => {\n        const origAxes = parseAxisParam(axis, $x.shape);\n        let axes = origAxes;\n        const permutedAxes = axis_util.getAxesPermutation(axes, $x.rank);\n        let minInput = $x;\n        if (permutedAxes != null) {\n          minInput = transpose($x, permutedAxes);\n          axes = axis_util.getInnerMostAxes(axes.length, $x.rank);\n        }\n\n        const y = backend.min(minInput, axes);\n        if (permutedAxes != null) {\n          minInput.dispose();\n        }\n\n        let res = y;\n        if (keepDims) {\n          const expandedShape =\n              axis_util.expandShapeToKeepDim(res.shape, origAxes);\n          res = reshape(y, expandedShape) as T;\n          y.dispose();\n        }\n\n        save([$x, res]);\n        return res;\n      };\n\n  const inputs: MinInputs = {x: $x};\n  const attrs: MinAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */, Min,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const min = op({min_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Minimum, MinimumInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Returns the min of a and b (`a < b ? a : b`) element-wise.\n * Supports broadcasting.\n *\n * We also expose `minimumStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * ```js\n * // Broadcast minimum a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.minimum(b).print();  // or tf.minimum(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction minimum_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'minimum');\n  let $b = convertToTensor(b, 'b', 'minimum');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  if ($a.dtype === 'bool') {\n    $a = cast($a, 'int32');\n    $b = cast($b, 'int32');\n  }\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.minimum($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: MinimumInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Minimum) as T;\n}\n\nexport const minimum = op({minimum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Mod, ModInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns the mod of a and b element-wise.\n * `floor(x / y) * y + mod(x, y) = x`\n * Supports broadcasting.\n *\n * We also expose `tf.modStrict` which has the same signature as this op and\n * asserts that `a` and `b` are the same shape (does not broadcast).\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * ```js\n * // Broadcast a mod b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.mod(b).print();  // or tf.mod(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction mod_<T extends Tensor>(a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'mod');\n  let $b = convertToTensor(b, 'b', 'mod');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.mod($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n  const inputs: ModInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Mod) as T;\n}\n\nexport const mod = op({mod_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Computes square of `x` element-wise: `x ^ 2`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, Math.sqrt(2), -1]);\n *\n * x.square().print();  // or tf.square(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction square_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'square');\n  const attrs = {};\n  const inputsToSave = [$x];\n  const outputsToSave: boolean[] = [];\n  return ENGINE.runKernelFunc((backend, save) => {\n    save([$x]);\n    return backend.square($x);\n  }, {x: $x}, null /* grad */, 'Square', attrs, inputsToSave, outputsToSave);\n}\n\nexport const square = op({square_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim} from './axis_util';\nimport {cast} from './cast';\nimport {mean} from './mean';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {square} from './square';\nimport {sub} from './sub';\n\n/**\n * Calculates the mean and variance of `x`. The mean and variance are\n * calculated by aggregating the contents of `x` across `axes`. If `x` is\n * 1-D and `axes = [0]` this is just the mean and variance of a vector.\n *\n * @param x The input tensor.\n * @param axis The dimension(s) along with to compute mean and\n *     variance. By default it reduces all dimensions.\n * @param keepDims If true, the moments have the same dimensionality as the\n *     input.\n * @return An object with two keys: `mean` and `variance`.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction moments_(\n    x: Tensor|TensorLike, axis: number|number[] = null,\n    keepDims = false): {mean: Tensor, variance: Tensor} {\n  x = convertToTensor(x, 'x', 'moments');\n  const axes = parseAxisParam(axis, x.shape);\n  const xMean = mean(x, axes, keepDims);\n  let keepDimsShape = xMean.shape;\n  if (!keepDims) {\n    keepDimsShape = expandShapeToKeepDim(xMean.shape, axes);\n  }\n  const devSquared =\n      square(sub(cast(x, 'float32'), reshape(xMean, keepDimsShape)));\n  const variance = mean(devSquared, axes, keepDims);\n  return {mean: xMean, variance};\n}\n\nexport const moments = op({moments_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor, convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * @docalias (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D]\n */\nexport type LSTMCellFunc = {\n  (data: Tensor2D, c: Tensor2D, h: Tensor2D): [Tensor2D, Tensor2D];\n};\n\n/**\n * Computes the next states and outputs of a stack of LSTMCells.\n *\n * Each cell output is used as input to the next cell.\n *\n * Returns `[cellState, cellOutput]`.\n *\n * Derived from tf.contrib.rn.MultiRNNCell.\n *\n * @param lstmCells Array of LSTMCell functions.\n * @param data The input to the cell.\n * @param c Array of previous cell states.\n * @param h Array of previous cell outputs.\n *\n * @doc {heading: 'Operations', subheading: 'RNN'}\n */\nfunction multiRNNCell_(\n    lstmCells: LSTMCellFunc[], data: Tensor2D|TensorLike,\n    c: Array<Tensor2D|TensorLike>,\n    h: Array<Tensor2D|TensorLike>): [Tensor2D[], Tensor2D[]] {\n  const $data = convertToTensor(data, 'data', 'multiRNNCell');\n  const $c = convertToTensorArray(c, 'c', 'multiRNNCell');\n  const $h = convertToTensorArray(h, 'h', 'multiRNNCell');\n\n  let input = $data;\n  const newStates = [];\n  for (let i = 0; i < lstmCells.length; i++) {\n    const output = lstmCells[i](input, $c[i], $h[i]);\n    newStates.push(output[0]);\n    newStates.push(output[1]);\n    input = output[1];\n  }\n  const newC: Tensor2D[] = [];\n  const newH: Tensor2D[] = [];\n  for (let i = 0; i < newStates.length; i += 2) {\n    newC.push(newStates[i]);\n    newH.push(newStates[i + 1]);\n  }\n  return [newC, newH];\n}\nexport const multiRNNCell = op({multiRNNCell_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a `tf.Tensor` with values drawn from a multinomial distribution.\n *\n * ```js\n * const probs = tf.tensor([.75, .25]);\n * tf.multinomial(probs, 3).print();\n * ```\n *\n * @param logits 1D array with unnormalized log-probabilities, or\n *     2D array of shape `[batchSize, numOutcomes]`. See the `normalized`\n *     parameter.\n * @param numSamples Number of samples to draw for each row slice.\n * @param seed The seed number.\n * @param normalized Whether the provided `logits` are normalized true\n *     probabilities (sum to 1). Defaults to false.\n * @return 1D array of shape `[numSamples]`, or 2D array of shape\n *     `[batchSize, numSamples]`, depending on the rank of the input.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction multinomial_(\n    logits: Tensor1D|Tensor2D|TensorLike, numSamples: number, seed?: number,\n    normalized = false): Tensor1D|Tensor2D {\n  const $logits = convertToTensor(logits, 'logits', 'multinomial');\n  const numOutcomes = $logits.size;\n  const origRank = $logits.rank;\n  if (numOutcomes < 2) {\n    throw new Error(\n        `Error in multinomial: you need at least 2 outcomes, but got ` +\n        `${numOutcomes}.`);\n  }\n  if (origRank > 2) {\n    throw new Error(`Rank of probabilities must be 1 or 2, but is ${origRank}`);\n  }\n  seed = seed || Math.random();\n  const logits2D: Tensor2D =\n      origRank === 1 ? reshape($logits, [1, -1]) : $logits as Tensor2D;\n  const res = ENGINE.runKernelFunc(\n      backend => backend.multinomial(logits2D, normalized, numSamples, seed),\n      {logits2D});\n  // tslint:disable-next-line:no-unnecessary-type-assertion\n  return origRank === 1 ? reshape(res, [res.size]) as Tensor1D : res;\n}\n\nexport const multinomial = op({multinomial_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {NotEqual, NotEqualInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns the truth value of (a != b) element-wise. Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([0, 2, 3]);\n *\n * a.notEqual(b).print();\n * ```\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same dtype as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nfunction notEqual_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'notEqual');\n  let $b = convertToTensor(b, 'b', 'notEqual');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => backend.notEqual($a, $b);\n\n  const inputs: NotEqualInputs = {a: $a, b: $b};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             NotEqual) as T;\n}\n\nexport const notEqual = op({notEqual_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Real, RealInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Returns the real part of a complex (or real) tensor.\n *\n * Given a tensor input, this operation returns a tensor of type float that is\n * the real part of each element in input considered as a complex number.\n *\n * If the input is real, it simply makes a clone.\n *\n * ```js\n * const x = tf.complex([-2.25, 3.25], [4.75, 5.75]);\n * tf.real(x).print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction real_<T extends Tensor>(input: T|TensorLike): T {\n  const $input = convertToTensor(input, 'input', 'real');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.real($input);\n  };\n\n  const inputs: RealInputs = {input: $input};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Real) as T;\n}\n\nexport const real = op({real_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OnesLike, OnesLikeInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {complex} from './complex';\nimport {imag} from './imag';\nimport {op} from './operation';\nimport {real} from './real';\nimport {zerosLike} from './zeros_like';\n\n/**\n * Creates a `tf.Tensor` with all elements set to 1 with the same shape as the\n * given tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n * tf.onesLike(x).print();\n * ```\n * @param x A tensor.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction onesLike_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'onesLike');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    if ($x.dtype === 'complex64') {\n      const r = onesLike(real($x));\n      const i = zerosLike(imag($x));\n      return complex(r, i);\n    }\n\n    return backend.onesLike($x);\n  };\n\n  const inputs: OnesLikeInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             OnesLike) as T;\n}\n\nexport const onesLike = op({onesLike_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {matMul} from './mat_mul';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the outer product of two vectors, `v1` and `v2`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n * const b = tf.tensor1d([3, 4, 5]);\n *\n * tf.outerProduct(a, b).print();\n * ```\n * @param v1 The first vector in the outer product operation.\n * @param v2 The second vector in the outer product operation.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction outerProduct_(\n    v1: Tensor1D|TensorLike, v2: Tensor1D|TensorLike): Tensor2D {\n  const $v1 = convertToTensor(v1, 'v1', 'outerProduct');\n  const $v2 = convertToTensor(v2, 'v2', 'outerProduct');\n\n  util.assert(\n      $v1.rank === 1 && $v2.rank === 1,\n      () => `Error in outerProduct: inputs must be rank 1, but got ranks ` +\n          `${$v1.rank} and ${$v2.rank}.`);\n\n  const v12D = reshape($v1, [-1, 1]);\n  const v22D = reshape($v2, [1, -1]);\n  return matMul(v12D, v22D) as Tensor2D;\n}\n\nexport const outerProduct = op({outerProduct_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {PadV2, PadV2Attrs, PadV2Inputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Pads a `tf.Tensor` with a given value and paddings.\n *\n * This operation currently only implements the `CONSTANT` mode.\n *\n * Also available are stricter rank-specific methods with the same signature\n * as this method that assert that `paddings` is of given length.\n *   - `tf.pad1d`\n *   - `tf.pad2d`\n *   - `tf.pad3d`\n *   - `tf.pad4d`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.pad([[1, 2]]).print();\n * ```\n * @param x The tensor to pad.\n * @param paddings An array of length `R` (the rank of the tensor), where\n * each element is a length-2 tuple of ints `[padBefore, padAfter]`,\n * specifying how much to pad along each dimension of the tensor.\n * @param constantValue The pad value to use. Defaults to 0.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction pad_<T extends Tensor>(\n    x: T|TensorLike, paddings: Array<[number, number]>, constantValue = 0): T {\n  const $x = convertToTensor(x, 'x', 'pad');\n  if ($x.rank === 0) {\n    throw new Error('pad(scalar) is not defined. Pass non-scalar to pad');\n  }\n  const forward: ForwardFunc<T> = (backend, save) => {\n    save([$x]);\n    return backend.pad($x, paddings, constantValue);\n  };\n\n  const attrs: PadV2Attrs = {paddings, constantValue};\n  const inputs: PadV2Inputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      forward, inputs as unknown as NamedTensorMap, null /* grad */, PadV2,\n      attrs as unknown as NamedAttrMap);\n}\n\nexport const pad = op({pad_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor1D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor1D` with a given value and paddings. See `pad` for details.\n */\nfunction pad1d_(\n    x: Tensor1D|TensorLike, paddings: [number, number],\n    constantValue = 0): Tensor1D {\n  assert(\n      paddings.length === 2,\n      () => 'Invalid number of paddings. Must be length of 2.');\n  return pad(x, [paddings], constantValue);\n}\n\nexport const pad1d = op({pad1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor2D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor2D` with a given value and paddings. See `pad` for details.\n */\nfunction pad2d_(\n    x: Tensor2D|TensorLike, paddings: [[number, number], [number, number]],\n    constantValue = 0): Tensor2D {\n  assert(\n      paddings.length === 2 && paddings[0].length === 2 &&\n          paddings[1].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad2d = op({pad2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor3D` with a given value and paddings. See `pad` for details.\n */\nfunction pad3d_(\n    x: Tensor3D|TensorLike,\n    paddings: [[number, number], [number, number], [number, number]],\n    constantValue = 0): Tensor3D {\n  assert(\n      paddings.length === 3 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad3d = op({pad3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor4D} from '../tensor';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\nimport {op} from './operation';\nimport {pad} from './pad';\n\n/**\n * Pads a `tf.Tensor4D` with a given value and paddings. See `pad` for details.\n */\nfunction pad4d_(\n    x: Tensor4D|TensorLike,\n    paddings:\n        [\n          [number, number], [number, number], [number, number], [number, number]\n        ],\n    constantValue = 0): Tensor4D {\n  assert(\n      paddings.length === 4 && paddings[0].length === 2 &&\n          paddings[1].length === 2 && paddings[2].length === 2 &&\n          paddings[3].length === 2,\n      () => 'Invalid number of paddings. Must be length of 2 each.');\n  return pad(x, paddings, constantValue);\n}\n\nexport const pad4d = op({pad4d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * This operation divides \"spatial\" dimensions `[1, ..., M]` of the input into\n * a grid of blocks of shape `blockShape`, and interleaves these blocks with\n * the \"batch\" dimension (0) such that in the output, the spatial\n * dimensions `[1, ..., M]` correspond to the position within the grid,\n * and the batch dimension combines both the position within a spatial block\n * and the original batch position. Prior to division into blocks,\n * the spatial dimensions of the input are optionally zero padded\n * according to `paddings`. See below for a precise description.\n *\n * ```js\n * const x = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);\n * const blockShape = [2, 2];\n * const paddings = [[0, 0], [0, 0]];\n *\n * x.spaceToBatchND(blockShape, paddings).print();\n * ```\n *\n * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +\n * remainingShape`, where spatialShape has `M` dimensions.\n * @param blockShape A 1-D array. Must have shape `[M]`, all values must\n * be >= 1.\n * @param paddings A 2-D array. Must have shape `[M, 2]`, all values must be >=\n *     0. `paddings[i] = [padStart, padEnd]` specifies the amount to zero-pad\n * from input dimension `i + 1`, which corresponds to spatial dimension `i`. It\n * is required that\n * `(inputShape[i + 1] + padStart + padEnd) % blockShape[i] === 0`\n *\n * This operation is equivalent to the following steps:\n *\n * 1. Zero-pad the start and end of dimensions `[1, ..., M]` of the input\n * according to `paddings` to produce `padded` of shape paddedShape.\n *\n * 2. Reshape `padded` to `reshapedPadded` of shape:\n * `[batch] + [paddedShape[1] / blockShape[0], blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1], blockShape[M-1]] + remainingShape`\n *\n * 3. Permute dimensions of `reshapedPadded` to produce `permutedReshapedPadded`\n * of shape: `blockShape + [batch] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * 4. Reshape `permutedReshapedPadded` to flatten `blockShape` into the\n * batch dimension, producing an output tensor of shape:\n * `[batch * prod(blockShape)] + [paddedShape[1] / blockShape[0], ...,\n * paddedShape[M] / blockShape[M-1]] + remainingShape`\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction spaceToBatchND_<T extends Tensor>(\n    x: T|TensorLike, blockShape: number[], paddings: number[][]): T {\n  const $x = convertToTensor(x, 'x', 'spaceToBatchND');\n\n  util.assert(\n      $x.rank >= 1 + blockShape.length,\n      () => `input rank ${$x.rank} should be > than [blockShape] ${\n          blockShape.length}`);\n\n  util.assert(\n      paddings.length === blockShape.length,\n      () => `paddings.shape[0] ${\n          paddings.length} must be equal to [blockShape] ${blockShape.length}`);\n\n  util.assert(\n      $x.shape.reduce(\n          (a, b, i) => {\n            if (i > 0 && i <= blockShape.length) {\n              return a &&\n                  ((b + paddings[i - 1][0] + paddings[i - 1][1]) %\n                       blockShape[i - 1] ===\n                   0);\n            }\n            return a;\n          },\n          true),\n      () => `input spatial dimensions ${$x.shape.slice(1)} with paddings ${\n          paddings.toString()} must be divisible by blockShapes ${\n          blockShape.toString()}`);\n\n  const forward: ForwardFunc<T> = backend =>\n      backend.spaceToBatchND($x, blockShape, paddings);\n\n  const inputs: SpaceToBatchNDInputs = {x: $x};\n  const attrs: SpaceToBatchNDAttrs = {blockShape, paddings};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      SpaceToBatchND, attrs as {} as NamedAttrMap);\n}\n\nexport const spaceToBatchND = op({spaceToBatchND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {avgPool} from './avg_pool';\nimport {batchToSpaceND} from './batch_to_space_nd';\nimport * as conv_util from './conv_util';\nimport {maxPool} from './max_pool';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {spaceToBatchND} from './space_to_batch_nd';\n\n/**\n * Performs an N-D pooling operation\n *\n * @param input The input tensor, of rank 4 or rank 3 of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param windowShape The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param poolingType The type of pooling, either 'max' or 'avg'.\n * @param pad The type of padding algorithm:\n *    - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *    - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *    - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *         https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in dilated pooling. Defaults to `[1, 1]`. If `dilationRate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction pool_<T extends Tensor3D|Tensor4D>(\n    input: T|TensorLike, windowShape: [number, number]|number,\n    poolingType: 'avg'|'max', pad: 'valid'|'same'|number,\n    dilations?: [number, number]|number, strides?: [number, number]|number) {\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  if (strides == null) {\n    strides = 1;\n  }\n  if (pad === 0) {\n    pad = 'valid';\n  }\n\n  const $x = convertToTensor(input, 'x', 'maxPool');\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in pool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = conv_util.computePool2DInfo(\n      x4D.shape, windowShape, strides, dilations, pad);\n  const dilation: [number, number] =\n      [convInfo.dilationHeight, convInfo.dilationWidth];\n\n  // The following implementation does batchToSpace(pool(spaceToBatch(x)))\n  // whenever dilation > 1 since the TF kernels do not support dilation > 1.\n  // tslint:disable-next-line:max-line-length\n  // https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L1037\n\n  let basePadding: number[][];\n  if (pad === 'same') {\n    basePadding = withSpaceToBatchBasePaddings(\n        [convInfo.filterHeight, convInfo.filterWidth], dilation);\n  } else {\n    basePadding = [[0, 0], [0, 0]];\n  }\n\n  const isDilationOne = dilation[0] === 1 && dilation[1] === 1;\n  const [adjustedPadding, adjustedCrops] = requiredSpaceToBatchPaddings(\n      [convInfo.inHeight, convInfo.inWidth], dilation, basePadding);\n  const convertedPad = isDilationOne ? pad : 'valid';\n  const convertedX =\n      isDilationOne ? x4D : spaceToBatchND(x4D, dilation, adjustedPadding);\n\n  const forwardOp = poolingType === 'avg' ?\n      () => avgPool(convertedX, windowShape, strides, convertedPad) :\n      () => maxPool(convertedX, windowShape, strides, convertedPad);\n  const y = forwardOp();\n\n  const res = isDilationOne ? y : batchToSpaceND(y, dilation, adjustedCrops);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n\n  return res as T;\n}\n\n// Helper function to compute crops and paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/array_ops.py#L2184\nfunction requiredSpaceToBatchPaddings(\n    inputShape: [number, number], blockShape: [number, number],\n    basePadding: number[][]) {\n  const padStart = basePadding.map(b => b[0]);\n  const origPadEnd = basePadding.map(b => b[1]);\n  const fullInputShape = inputShape.concat(padStart, origPadEnd);\n  const padEndExtra = blockShape.map((b, i) => (b - fullInputShape[i] % b) % b);\n  const padEnd = origPadEnd.map((s, i) => s + padEndExtra[i]);\n  const paddings = blockShape.map((_, i) => [padStart[i], padEnd[i]]);\n  const crops = blockShape.map((_, i) => [0, padEndExtra[i]]);\n  return [paddings, crops];\n}\n\n// Helper function to compute base paddings for pool with dilation > 1.\n// tslint:disable-next-line:max-line-length\n// https://github.com/tensorflow/tensorflow/blob/50f6bb67dc98c9b74630b6047aae7a4f8a40fd02/tensorflow/python/ops/nn_ops.py#L524\nfunction withSpaceToBatchBasePaddings(\n    filterShape: [number, number], dilation: [number, number]) {\n  // Spatial dimensions of the filters and the upsampled filters in which we\n  // introduce (rate - 1) zeros between consecutive filter values.\n  const dilatedFilterShape = filterShape.map((s, i) => {\n    return s + (s - 1) * (dilation[i] - 1);\n  });\n  const padExtraShape = dilatedFilterShape.map(s => s - 1);\n\n  // When padding is odd, we pad more at end, following the same\n  // convention as conv2d.\n  const padExtraStart = padExtraShape.map(s => Math.floor(s / 2));\n  const padExtraEnd = padExtraShape.map((s, i) => s - padExtraStart[i]);\n  return padExtraShape.map((_, i) => {\n    return [padExtraStart[i], padExtraEnd[i]];\n  });\n}\n\nexport const pool = op({pool_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Pow, PowInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the power of one `tf.Tensor` to another. Supports broadcasting.\n *\n * Given a `tf.Tensor` x and a `tf.Tensor` y, this operation computes x^y for\n * corresponding elements in x and y. The result's dtype will be the upcasted\n * type of the `base` and `exp` dtypes.\n *\n * ```js\n * const a = tf.tensor([[2, 3], [4, 5]])\n * const b = tf.tensor([[1, 2], [3, 0]]).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n *\n * ```js\n * const a = tf.tensor([[1, 2], [3, 4]])\n * const b = tf.tensor(2).toInt();\n *\n * a.pow(b).print();  // or tf.pow(a, b)\n * ```\n * We also expose `powStrict` which has the same signature as this op and\n * asserts that `base` and `exp` are the same shape (does not broadcast).\n *\n * @param base The base `tf.Tensor` to pow element-wise.\n * @param exp The exponent `tf.Tensor` to pow element-wise.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction pow_<T extends Tensor>(\n    base: Tensor|TensorLike, exp: Tensor|TensorLike): T {\n  let $base = convertToTensor(base, 'base', 'pow');\n  let $exp = convertToTensor(exp, 'exp', 'pow');\n  [$base, $exp] = makeTypesMatch($base, $exp);\n\n  const inputs: PowInputs = {a: $base, b: $exp};\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const y = backend.pow($base, $exp);\n    save([$base, $exp, y]);\n    return y;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Pow) as T;\n}\n\nexport const pow = op({pow_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prelu, PreluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes leaky rectified linear element-wise with parametric alphas.\n *\n * `x < 0 ? alpha * x : f(x) = x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n * const alpha = tf.scalar(0.1);\n *\n * x.prelu(alpha).print();  // or tf.prelu(x, alpha)\n * ```\n * @param x The input tensor.\n * @param alpha Scaling factor for negative values.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction prelu_<T extends Tensor>(x: T|TensorLike, alpha: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'prelu');\n  const $alpha = convertToTensor(alpha, 'alpha', 'prelu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.prelu($x, $alpha);\n    save([$x, $alpha]);\n    return res;\n  };\n\n  const inputs: PreluInputs = {x: $x, alpha: $alpha};\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prelu) as\n      T;\n}\n\nexport const prelu = op({prelu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Prod, ProdAttrs, ProdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {expandShapeToKeepDim, getAxesPermutation, getInnerMostAxes} from './axis_util';\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {transpose} from './transpose';\n\n/**\n * Computes the product of elements across dimensions of a `tf.Tensor`.\n *\n * Reduces the input along the dimensions given in `axes`. Unless `keepDims`\n * is true, the rank of the `tf.Tensor` is reduced by 1 for each entry in\n * `axes`. If `keepDims` is true, the reduced dimensions are retained with\n * length 1. If `axes` has no entries, all dimensions are reduced, and a\n * `tf.Tensor` with a single element is returned.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3]);\n *\n * x.prod().print();  // or tf.prod(x)\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.prod(axis).print();  // or tf.prod(x, axis)\n * ```\n *\n * @param x The input tensor to compute the product over. If the dtype is `bool`\n *   it will be converted to `int32` and the output dtype will be `int32`.\n * @param axis The dimension(s) to reduce. By default it reduces\n *     all dimensions.\n * @param keepDims If true, retains reduced dimensions with size 1.\n *\n * @doc {heading: 'Operations', subheading: 'Reduction'}\n */\nfunction prod_<T extends Tensor>(\n    x: Tensor|TensorLike, axis: number|number[] = null, keepDims = false): T {\n  let $x = convertToTensor(x, 'x', 'prod');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if ($x.dtype === 'bool') {\n      $x = cast($x, 'int32');\n    }\n    const axes = parseAxisParam(axis, $x.shape);\n\n    const permutation = getAxesPermutation(axes, $x.rank);\n    let reductionAxes = axes;\n    let permutedX = $x;\n    if (permutation != null) {\n      permutedX = transpose($x, permutation);\n      reductionAxes = getInnerMostAxes(reductionAxes.length, $x.rank);\n    }\n    let value = backend.prod(permutedX, reductionAxes);\n    if (keepDims) {\n      const newShape = expandShapeToKeepDim(value.shape, axes);\n      value = reshape(value, newShape);\n    }\n\n    return value as T;\n  };\n\n  const inputs: ProdInputs = {x: $x};\n  const attrs: ProdAttrs = {axis, keepDims};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Prod,\n             attrs as {} as NamedAttrMap) as T;\n}\n\nexport const prod = op({prod_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a random number generator\n * function defined by the user.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param randFunction A random number generator function which is called\n * for each element in the output tensor.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n */\nfunction rand_<R extends Rank>(\n    shape: ShapeMap[R], randFunction: () => number,\n    dtype?: DataType): Tensor<R> {\n  const size = sizeFromShape(shape);\n  let values = null;\n  if (dtype == null || dtype === 'float32') {\n    values = new Float32Array(size);\n  } else if (dtype === 'int32') {\n    values = new Int32Array(size);\n  } else if (dtype === 'bool') {\n    values = new Uint8Array(size);\n  } else {\n    throw new Error(`Unknown data type ${dtype}`);\n  }\n  for (let i = 0; i < size; i++) {\n    values[i] = randFunction();\n  }\n  return ENGINE.makeTensor(values, shape, dtype) as Tensor<R>;\n}\n\nexport const rand = op({rand_});\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as seedrandom from 'seedrandom';\n\nimport {expectNumbersClose, testEpsilon} from '../test_util';\nimport {TypedArray} from '../types';\n\nexport interface RandomBase {\n  nextValue(): number;\n}\n\nexport interface RandomGamma {\n  nextValue(): number;\n}\n\nexport interface RandNormalDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\nexport interface RandGammaDataTypes {\n  float32: Float32Array;\n  int32: Int32Array;\n}\n\n// https://en.wikipedia.org/wiki/Marsaglia_polar_method\nexport class MPRandGauss implements RandomBase {\n  private mean: number;\n  private stdDev: number;\n  private nextVal: number;\n  private dtype?: keyof RandNormalDataTypes;\n  private truncated?: boolean;\n  private upper?: number;\n  private lower?: number;\n  private random: seedrandom.prng;\n\n  constructor(\n      mean: number, stdDeviation: number, dtype?: keyof RandNormalDataTypes,\n      truncated?: boolean, seed?: number) {\n    this.mean = mean;\n    this.stdDev = stdDeviation;\n    this.dtype = dtype;\n    this.nextVal = NaN;\n    this.truncated = truncated;\n    if (this.truncated) {\n      this.upper = this.mean + this.stdDev * 2;\n      this.lower = this.mean - this.stdDev * 2;\n    }\n    const seedValue = seed ? seed : Math.random();\n    this.random = seedrandom.alea(seedValue.toString());\n  }\n\n  /** Returns next sample from a Gaussian distribution. */\n  public nextValue(): number {\n    if (!isNaN(this.nextVal)) {\n      const value = this.nextVal;\n      this.nextVal = NaN;\n      return value;\n    }\n\n    let resultX: number, resultY: number;\n    let isValid = false;\n    while (!isValid) {\n      let v1: number, v2: number, s: number;\n      do {\n        v1 = 2 * this.random() - 1;\n        v2 = 2 * this.random() - 1;\n        s = v1 * v1 + v2 * v2;\n      } while (s >= 1 || s === 0);\n\n      const mul = Math.sqrt(-2.0 * Math.log(s) / s);\n      resultX = this.mean + this.stdDev * v1 * mul;\n      resultY = this.mean + this.stdDev * v2 * mul;\n\n      if (!this.truncated || this.isValidTruncated(resultX)) {\n        isValid = true;\n      }\n    }\n\n    if (!this.truncated || this.isValidTruncated(resultY)) {\n      this.nextVal = this.convertValue(resultY);\n    }\n    return this.convertValue(resultX);\n  }\n\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype == null || this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  /** Returns true if less than 2-standard-deviations from the mean. */\n  private isValidTruncated(value: number): boolean {\n    return value <= this.upper && value >= this.lower;\n  }\n}\n\n// Marsaglia, George, and Wai Wan Tsang. 2000. \"A Simple Method for Generating\n// Gamma Variables.\"\nexport class RandGamma implements RandomGamma {\n  private alpha: number;\n  private beta: number;\n  private d: number;\n  private c: number;\n  private dtype?: keyof RandGammaDataTypes;\n  private randu: seedrandom.prng;\n  private randn: MPRandGauss;\n\n  constructor(\n      alpha: number, beta: number, dtype: keyof RandGammaDataTypes,\n      seed?: number) {\n    this.alpha = alpha;\n    this.beta = 1 / beta;  // convert rate to scale parameter\n    this.dtype = dtype;\n\n    const seedValue = seed ? seed : Math.random();\n    this.randu = seedrandom.alea(seedValue.toString());\n    this.randn = new MPRandGauss(0, 1, dtype, false, this.randu());\n\n    if (alpha < 1) {\n      this.d = alpha + (2 / 3);\n    } else {\n      this.d = alpha - (1 / 3);\n    }\n    this.c = 1 / Math.sqrt(9 * this.d);\n  }\n\n  /** Returns next sample from a gamma distribution. */\n  public nextValue(): number {\n    let x2: number, v0: number, v1: number, x: number, u: number, v: number;\n    while (true) {\n      do {\n        x = this.randn.nextValue();\n        v = 1 + (this.c * x);\n      } while (v <= 0);\n      v *= v * v;\n      x2 = x * x;\n      v0 = 1 - (0.331 * x2 * x2);\n      v1 = (0.5 * x2) + (this.d * (1 - v + Math.log(v)));\n      u = this.randu();\n      if (u < v0 || Math.log(u) < v1) {\n        break;\n      }\n    }\n    v = (1 / this.beta) * this.d * v;\n    if (this.alpha < 1) {\n      v *= Math.pow(this.randu(), 1 / this.alpha);\n    }\n    return this.convertValue(v);\n  }\n  /** Handles proper rounding for non-floating-point numbers. */\n  private convertValue(value: number): number {\n    if (this.dtype === 'float32') {\n      return value;\n    }\n    return Math.round(value);\n  }\n}\n\nexport class UniformRandom implements RandomBase {\n  private min: number;\n  private range: number;\n  private random: seedrandom.prng;\n  private dtype?: keyof RandNormalDataTypes;\n\n  constructor(\n      min = 0, max = 1, dtype?: keyof RandNormalDataTypes,\n      seed?: string|number) {\n    this.min = min;\n    this.range = max - min;\n    this.dtype = dtype;\n    if (seed == null) {\n      seed = Math.random();\n    }\n    if (typeof seed === 'number') {\n      seed = seed.toString();\n    }\n\n    if (!this.canReturnFloat() && this.range <= 1) {\n      throw new Error(\n          `The difference between ${min} - ${max} <= 1 and dtype is not float`);\n    }\n    this.random = seedrandom.alea(seed);\n  }\n\n  /** Handles proper rounding for non floating point numbers. */\n  private canReturnFloat = () =>\n      (this.dtype == null || this.dtype === 'float32');\n\n  private convertValue(value: number): number {\n    if (this.canReturnFloat()) {\n      return value;\n    }\n    return Math.round(value);\n  }\n\n  nextValue() {\n    return this.convertValue(this.min + this.range * this.random());\n  }\n}\n\nexport function jarqueBeraNormalityTest(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test\n  const n = values.length;\n  const s = skewness(values);\n  const k = kurtosis(values);\n  const jb = n / 6 * (Math.pow(s, 2) + 0.25 * Math.pow(k - 3, 2));\n  // JB test requires 2-degress of freedom from Chi-Square @ 0.95:\n  // http://www.itl.nist.gov/div898/handbook/eda/section3/eda3674.htm\n  const CHI_SQUARE_2DEG = 5.991;\n  if (jb > CHI_SQUARE_2DEG) {\n    throw new Error(`Invalid p-value for JB: ${jb}`);\n  }\n}\n\nexport function expectArrayInMeanStdRange(\n    actual: TypedArray|number[], expectedMean: number, expectedStdDev: number,\n    epsilon?: number) {\n  if (epsilon == null) {\n    epsilon = testEpsilon();\n  }\n  const actualMean = mean(actual);\n  expectNumbersClose(actualMean, expectedMean, epsilon);\n  expectNumbersClose(\n      standardDeviation(actual, actualMean), expectedStdDev, epsilon);\n}\n\nfunction mean(values: TypedArray|number[]) {\n  let sum = 0;\n  for (let i = 0; i < values.length; i++) {\n    sum += values[i];\n  }\n  return sum / values.length;\n}\n\nfunction standardDeviation(values: TypedArray|number[], mean: number) {\n  let squareDiffSum = 0;\n  for (let i = 0; i < values.length; i++) {\n    const diff = values[i] - mean;\n    squareDiffSum += diff * diff;\n  }\n  return Math.sqrt(squareDiffSum / values.length);\n}\n\nfunction kurtosis(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Kurtosis\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum4 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum4 += Math.pow(v, 4);\n  }\n  return (1 / n) * sum4 / Math.pow((1 / n) * sum2, 2);\n}\n\nfunction skewness(values: TypedArray|number[]) {\n  // https://en.wikipedia.org/wiki/Skewness\n  const valuesMean = mean(values);\n  const n = values.length;\n  let sum2 = 0;\n  let sum3 = 0;\n  for (let i = 0; i < n; i++) {\n    const v = values[i] - valuesMean;\n    sum2 += Math.pow(v, 2);\n    sum3 += Math.pow(v, 3);\n  }\n  return (1 / n) * sum3 / Math.pow((1 / (n - 1)) * sum2, 3 / 2);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {RandGamma} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a gamma distribution.\n *\n * ```js\n * tf.randomGamma([2, 2], 1).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param alpha The shape parameter of the gamma distribution.\n * @param beta The inverse scale parameter of the gamma distribution. Defaults\n *     to 1.\n * @param dtype The data type of the output. Defaults to float32.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomGamma_<R extends Rank>(\n    shape: ShapeMap[R], alpha: number, beta = 1,\n    dtype: 'float32'|'int32' = 'float32', seed?: number): Tensor<R> {\n  if (beta == null) {\n    beta = 1;\n  }\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const rgamma = new RandGamma(alpha, beta, dtype, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = rgamma.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomGamma = op({randomGamma_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a normal distribution.\n *\n * ```js\n * tf.randomNormal([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type ${dtype}`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, false /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomNormal = op({randomNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {UniformRandom} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a uniform distribution.\n *\n * The generated values follow a uniform distribution in the range [minval,\n * maxval). The lower bound minval is included in the range, while the upper\n * bound maxval is excluded.\n *\n * ```js\n * tf.randomUniform([2, 2]).print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param minval The lower bound on the range of random values to generate.\n *   Defaults to 0.\n * @param maxval The upper bound on the range of random values to generate.\n *   Defaults to 1.\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Random'}\n */\nfunction randomUniform_<R extends Rank>(\n    shape: ShapeMap[R], minval = 0, maxval = 1, dtype: DataType = 'float32',\n    seed?: number|string): Tensor<R> {\n  const res = buffer(shape, dtype);\n  const random = new UniformRandom(minval, maxval, null, seed);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = random.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const randomUniform = op({randomUniform_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike1D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-1 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor1d` as it makes the code more readable.\n *\n * ```js\n * tf.tensor1d([1, 2, 3]).print();\n * ```\n *\n * @param values The values of the tensor. Can be array of numbers,\n *     or a `TypedArray`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor1d(values: TensorLike1D, dtype?: DataType): Tensor1D {\n  assertNonNull(values);\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 1) {\n    throw new Error('tensor1d() requires values to be a flat/TypedArray');\n  }\n  const shape: number[] = null;\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Range, RangeAttrs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {makeZerosTypedArray} from '../util';\n\nimport {tensor1d} from './tensor1d';\nimport {zeros} from './zeros';\n\n/**\n * Creates a new `tf.Tensor1D` filled with the numbers in the range provided.\n *\n * The tensor is a is half-open interval meaning it includes start, but\n * excludes stop. Decrementing ranges and negative step values are also\n * supported.sv\n *\n *\n * ```js\n * tf.range(0, 9, 2).print();\n * ```\n *\n * @param start An integer start value\n * @param stop An integer stop value\n * @param step An integer increment (will default to 1 or -1)\n * @param dtype The data type of the output tensor. Defaults to 'float32'.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function range(\n    start: number, stop: number, step = 1,\n    dtype: 'float32'|'int32' = 'float32'): Tensor1D {\n  if (step === 0) {\n    throw new Error('Cannot have a step of zero');\n  }\n\n  const forward: ForwardFunc<Tensor> = () => {\n    const sameStartStop = start === stop;\n    const increasingRangeNegativeStep = start < stop && step < 0;\n    const decreasingRangePositiveStep = stop < start && step > 1;\n\n    if (sameStartStop || increasingRangeNegativeStep ||\n        decreasingRangePositiveStep) {\n      return zeros([0], dtype);\n    }\n\n    const numElements = Math.abs(Math.ceil((stop - start) / step));\n    const values = makeZerosTypedArray(numElements, dtype);\n\n    if (stop < start && step === 1) {\n      // Auto adjust the step's sign if it hasn't been set\n      // (or was set to 1)\n      step = -1;\n    }\n\n    values[0] = start;\n    for (let i = 1; i < values.length; i++) {\n      values[i] = values[i - 1] + step;\n    }\n\n    return tensor1d(values, dtype);\n  };\n\n  const attrs: RangeAttrs = {start, stop, step, dtype};\n\n  return ENGINE.runKernelFunc(\n             forward, {} /* inputs */, null /* grad */, Range,\n             attrs as {} as NamedAttrMap) as Tensor1D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Reciprocal, ReciprocalInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of x element-wise: `1 / x`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, 2]);\n *\n * x.reciprocal().print();  // or tf.reciprocal(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction reciprocal_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'reciprocal');\n\n  const inputs: ReciprocalInputs = {x: $x};\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.reciprocal($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Reciprocal);\n}\nexport const reciprocal = op({reciprocal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu, ReluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport { cast } from './cast';\n\n/**\n * Computes rectified linear element-wise: `max(x, 0)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.relu().print();  // or tf.relu(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu($x);\n  };\n\n  const inputs: ReluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu) as\n      T;\n}\n\nexport const relu = op({relu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Relu6, Relu6Inputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\n\n/**\n * Computes rectified linear 6 element-wise: `min(max(x, 0), 6)`.\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 8]);\n *\n * x.relu6().print();  // or tf.relu6(x)\n * ```\n * @param x The input tensor. If the dtype is `bool`, the output dtype will be\n *     `int32'.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction relu6_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'relu6');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    save([$x]);\n\n    if ($x.dtype === 'bool') {\n      return cast($x, 'int32');\n    }\n\n    return backend.relu6($x);\n  };\n\n  const inputs: Relu6Inputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Relu6) as\n      T;\n}\n\nexport const relu6 = op({relu6_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reverse, ReverseAttrs, ReverseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {clone} from './clone';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Reverses a `tf.Tensor` along a specified axis.\n *\n * Also available are stricter rank-specific methods that assert that `x` is\n * of the given rank:\n *   - `tf.reverse1d`\n *   - `tf.reverse2d`\n *   - `tf.reverse3d`\n *   - `tf.reverse4d`\n *\n * Except `tf.reverse1d` (which does not have axis param), all methods have\n * same signature as this method.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.reverse().print();\n * ```\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * const axis = 1;\n * x.reverse(axis).print();\n * ```\n * @param x The input tensor to be reversed.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction reverse_<T extends Tensor>(\n    x: T|TensorLike, axis?: number|number[]): T {\n  const $x = convertToTensor(x, 'x', 'reverse');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const axes = parseAxisParam(axis, $x.shape);\n    if ($x.rank === 0) {\n      return clone($x);\n    }\n    const res = backend.reverse($x, axes);\n    return reshape(res, $x.shape);\n  };\n\n  const inputs: ReverseInputs = {x: $x};\n  const attrs: ReverseAttrs = {dims: axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* gradient */,\n             Reverse, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const reverse = op({reverse_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor1D`.\n *\n * @param x The input tensor.\n */\nfunction reverse1d_(x: Tensor1D|TensorLike): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 1,\n      () => `Error in reverse1D: x must be rank 1 but got rank ${$x.rank}.`);\n  return reverse($x, 0);\n}\n\nexport const reverse1d = op({reverse1d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor2D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse2d_(x: Tensor2D|TensorLike, axis?: number|number[]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 2,\n      () => `Error in reverse2D: x must be rank 2 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse2d = op({reverse2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor3D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse3d_(x: Tensor3D|TensorLike, axis?: number|number[]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 3,\n      () => `Error in reverse3D: x must be rank 3 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse3d = op({reverse3d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\nimport {op} from './operation';\nimport {reverse} from './reverse';\n\n/**\n * Reverses a `tf.Tensor4D` along a specified axis.\n *\n * @param x The input tensor.\n * @param axis The set of dimensions to reverse. Must be in the\n *     range [-rank(x), rank(x)). Defaults to all axes.\n */\nfunction reverse4d_(x: Tensor4D|TensorLike, axis?: number|number[]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'reverse');\n  util.assert(\n      $x.rank === 4,\n      () => `Error in reverse4D: x must be rank 4 but got rank ${$x.rank}.`);\n  return reverse($x, axis);\n}\n\nexport const reverse4d = op({reverse4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Round, RoundInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes round of input `tf.Tensor` element-wise: `round(x)`.\n * It implements banker's rounding.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3]);\n *\n * x.round().print();  // or tf.round(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction round_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'round');\n  const inputs: RoundInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n      (backend) => backend.round($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Round);\n}\n\nexport const round = op({round_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Rsqrt, RsqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes reciprocal of square root of the input `tf.Tensor` element-wise:\n * `y = 1 / sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.rsqrt().print();  // or tf.rsqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction rsqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'rsqrt');\n\n  const inputs: RsqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.rsqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Rsqrt);\n}\nexport const rsqrt = op({rsqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Selu, SeluInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes scaled exponential linear element-wise.\n *\n * `x < 0 ? scale * alpha * (exp(x) - 1) : x`\n *\n * ```js\n * const x = tf.tensor1d([-1, 2, -3, 4]);\n *\n * x.selu().print();  // or tf.selu(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction selu_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'selu');\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.selu($x);\n    save([$x]);\n    return res;\n  };\n\n  const inputs: SeluInputs = {x: $x};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, Selu) as\n      T;\n}\n\nexport const selu = op({selu_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor3D, Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {conv2d} from './conv2d';\nimport {depthwiseConv2d} from './depthwise_conv2d';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * 2-D convolution with separable filters.\n *\n * Performs a depthwise convolution that acts separately on channels followed\n * by a pointwise convolution that mixes channels. Note that this is\n * separability between dimensions [1, 2] and 3, not spatial separability\n * between dimensions 1 and 2.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/separable_conv2d)\n * for more details.\n *\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param depthwiseFilter The depthwise filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`. This is\n *     the filter used in the first step.\n * @param pointwiseFilter The pointwise filter tensor, rank 4, of shape\n *     `[1, 1, inChannels * channelMultiplier, outChannels]`. This is\n *     the filter used in the second step.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n *\n * @doc {heading: 'Operations', subheading: 'Convolution'}\n */\nfunction separableConv2d_<T extends Tensor3D|Tensor4D>(\n    x: T|TensorLike, depthwiseFilter: Tensor4D|TensorLike,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation: [number, number]|number = [1, 1],\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC'): T {\n  const $x = convertToTensor(x, 'x', 'separableConv2d');\n  const $depthwiseFilter =\n      convertToTensor(depthwiseFilter, 'depthwiseFilter', 'separableConv2d');\n  const $pointwiseFilter =\n      convertToTensor(pointwiseFilter, 'pointwiseFilter', 'separableConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n\n  if (dataFormat === 'NCHW') {\n    throw new Error(\n        'separableConv2d currently does not support dataFormat NCHW; only ' +\n        'NHWC is supported');\n  }\n\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in separableConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $depthwiseFilter.rank === 4,\n      () => `Error in separableConv2d: depthwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.rank === 4,\n      () => `Error in separableConv2d: pointwise filter must be rank 4, but ` +\n          `got rank ${$depthwiseFilter.rank}.`);\n  util.assert(\n      $pointwiseFilter.shape[0] === 1,\n      () =>\n          `Error in separableConv2d: the first dimension of pointwise filter ` +\n          ` must be 1, but got ${$pointwiseFilter.shape[0]}.`);\n  util.assert(\n      $pointwiseFilter.shape[1] === 1,\n      () => `Error in separableConv2d: the second dimension of pointwise ` +\n          `filter must be 1, but got ${$pointwiseFilter.shape[1]}.`);\n\n  const inChannels = $depthwiseFilter.shape[2];\n  const channelMultiplier = $depthwiseFilter.shape[3];\n  util.assert(\n      $pointwiseFilter.shape[2] === inChannels * channelMultiplier,\n      () =>\n          `Error in separableConv2d: the third dimension of pointwise filter ` +\n          `must be ${inChannels * channelMultiplier}, ` +\n          `but got ${$pointwiseFilter.shape[2]}.`);\n\n  const depthwise = depthwiseConv2d(\n      x4D, $depthwiseFilter, strides, pad, dataFormat, dilation);\n  const pointwiseStride = 1;\n  const res =\n      conv2d(depthwise, $pointwiseFilter, pointwiseStride, 'valid', dataFormat);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const separableConv2d = op({separableConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\n/**\n * Computes the difference between two lists of numbers.\n *\n * Given a Tensor `x` and a Tensor `y`, this operation returns a Tensor `out`\n * that represents all values that are in `x` but not in `y`. The returned\n * Tensor `out` is sorted in the same order that the numbers appear in `x`\n * (duplicates are preserved). This operation also returns a Tensor indices that\n * represents the position of each out element in `x`. In other words:\n *\n * `out[i] = x[idx[i]] for i in [0, 1, ..., out.length - 1]`\n *\n * ```js\n * const x = [1, 2, 3, 4, 5, 6];\n * const y = [1, 3, 5];\n *\n * const [out, indices] = await tf.setdiff1dAsync(x, y);\n * out.print(); // [2, 4, 6]\n * indices.print(); // [1, 3, 5]\n * ```\n *\n * @param x 1-D Tensor. Values to keep.\n * @param y 1-D Tensor. Must have the same type as x. Values to exclude in the\n *     output.\n * @returns Promise of Tensor tuple [out, indices].\n *  out: Tensor with the same type as x.\n *  indices: A Tensor of type int32.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nasync function setdiff1dAsync_(\n    x: Tensor|TensorLike, y: Tensor|TensorLike): Promise<[Tensor, Tensor]> {\n  const $x = convertToTensor(x, 'x', 'setdiff1d');\n  const $y = convertToTensor(y, 'y', 'setdiff1d');\n\n  util.assert(\n      $x.dtype === $y.dtype,\n      () => `x and y should have the same dtype, but got x (${\n          $x.dtype}) and y (${$y.dtype}).`);\n\n  util.assert(\n      $x.rank === 1, () => `x should be 1D tensor, but got x (${$x.shape}).`);\n\n  util.assert(\n      $y.rank === 1, () => `y should be 1D tensor, but got y (${$y.shape}).`);\n\n  const xVals = await $x.data();\n  const yVals = await $y.data();\n  const ySet = new Set(yVals);\n\n  let outputSize = 0;\n  for (let i = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      outputSize++;\n    }\n  }\n\n  const buffer = new TensorBuffer([outputSize], $x.dtype);\n  const indices = new TensorBuffer([outputSize], 'int32');\n  for (let i = 0, p = 0; i < xVals.length; i++) {\n    if (!ySet.has(xVals[i])) {\n      buffer.values[p] = xVals[i];\n      indices.values[p] = i;\n      p++;\n    }\n  }\n  return [buffer.toTensor(), indices.toTensor()];\n}\nexport const setdiff1dAsync = setdiff1dAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sign, SignInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Returns an element-wise indication of the sign of a number.\n *\n * ```js\n * const x = tf.tensor1d([.6, 1.1, -3.3, NaN, 0]);\n *\n * x.sign().print();  // or tf.sign(x)\n * ```\n * @param x The input Tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sign_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sign');\n  const inputs: SignInputs = {x: $x};\n  return ENGINE.runKernelFunc(\n      backend => backend.sign($x), inputs as {} as NamedTensorMap,\n      null /* grad */, Sign);\n}\nexport const sign = op({sign_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sin, SinInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes sin of the input Tensor element-wise: `sin(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.sin().print();  // or tf.sin(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sin_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sin');\n\n  const inputs: SinInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sin($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sin);\n}\nexport const sin = op({sin_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sinh, SinhInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes hyperbolic sin of the input `tf.Tensor` element-wise: `sinh(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, 1, -1, .7]);\n *\n * x.sinh().print();  // or tf.sinh(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sinh_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sinh');\n  const inputs: SinhInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sinh($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sinh);\n}\nexport const sinh = op({sinh_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 1D slice from 1D array starting at coordinates `begin` and is\n * of length `size`. See `slice` for details.\n */\nfunction slice1d_(\n    x: Tensor1D|TensorLike, begin: number, size: number): Tensor1D {\n  const $x = convertToTensor(x, 'x', 'slice1d');\n  util.assert(\n      $x.rank === 1,\n      () =>\n          `slice1d expects a rank-1 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, [begin], [size]);\n}\nexport const slice1d = op({slice1d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 2D slice from a 2D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice2d_(\n    x: Tensor2D|TensorLike, begin: [number, number],\n    size: [number, number]): Tensor2D {\n  const $x = convertToTensor(x, 'x', 'slice2d');\n  util.assert(\n      $x.rank === 2,\n      () =>\n          `slice2d expects a rank-2 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice2d = op({slice2d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 3D slice from a 3D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice3d_(\n    x: Tensor3D|TensorLike, begin: [number, number, number],\n    size: [number, number, number]): Tensor3D {\n  const $x = convertToTensor(x, 'x', 'slice3d');\n  util.assert(\n      $x.rank === 3,\n      () =>\n          `slice3d expects a rank-3 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice3d = op({slice3d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {slice} from './slice';\n\n/**\n * Extracts a 4D slice from a 4D array starting at coordinates `begin` and\n * is of size `size`. See `slice` for details.\n */\nfunction slice4d_(\n    x: Tensor4D|TensorLike, begin: [number, number, number, number],\n    size: [number, number, number, number]): Tensor4D {\n  const $x = convertToTensor(x, 'x', 'slice4d');\n  util.assert(\n      $x.rank === 4,\n      () =>\n          `slice4d expects a rank-4 tensor, but got a rank-${$x.rank} tensor`);\n  return slice($x, begin, size);\n}\nexport const slice4d = op({slice4d_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Softmax, SoftmaxAttrs, SoftmaxInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes the softmax normalized vector given the logits.\n *\n * ```js\n * const a = tf.tensor1d([1, 2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * ```js\n * const a = tf.tensor2d([2, 4, 6, 1, 2, 3], [2, 3]);\n *\n * a.softmax().print();  // or tf.softmax(a)\n * ```\n *\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction softmax_<T extends Tensor>(logits: T|TensorLike, dim = -1): T {\n  const $logits = convertToTensor(logits, 'logits', 'softmax', 'float32');\n\n  if (dim === -1) {\n    dim = $logits.rank - 1;\n  }\n  if (dim !== $logits.rank - 1) {\n    throw Error(\n        'Softmax along a non-last dimension is not yet supported. ' +\n        `Logits was rank ${$logits.rank} and dim was ${dim}`);\n  }\n\n  const inputs: SoftmaxInputs = {logits: $logits};\n  const attrs: SoftmaxAttrs = {dim};\n\n  return ENGINE.runKernelFunc(\n      (backend, save) => {\n        const y = backend.softmax($logits, dim);\n        save([y]);\n        return y;\n      },\n      inputs as {} as NamedTensorMap, null /* grad */, Softmax,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const softmax = op({softmax_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FFT, FFTInputs} from '../../kernel_names';\nimport {Tensor} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\n\n/**\n * Fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the inner-most\n * dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.fft().print();  // tf.spectral.fft(x).print();\n * ```\n * @param input The complex input to compute an fft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction fft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.fft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: FFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D = input.as2D(batch, innerDimensionSize);\n    const result = backend.fft(input2D);\n    return result.reshape(input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, FFT);\n}\n\nexport const fft = op({fft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {IFFT, IFFTInputs} from '../../kernel_names';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {assert} from '../../util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Inverse fast Fourier transform.\n *\n * Computes the inverse 1-dimensional discrete Fourier transform over the\n * inner-most dimension of input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([1, 2, 3]);\n * const x = tf.complex(real, imag);\n *\n * x.ifft().print();  // tf.spectral.ifft(x).print();\n * ```\n * @param input The complex input to compute an ifft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction ifft_(input: Tensor): Tensor {\n  assert(\n      input.dtype === 'complex64',\n      () => `The dtype for tf.spectral.ifft() must be complex64 ` +\n          `but got ${input.dtype}.`);\n\n  const inputs: IFFTInputs = {input};\n\n  return ENGINE.runKernelFunc(backend => {\n    // Collapse all outer dimensions to a single batch dimension.\n    const innerDimensionSize = input.shape[input.shape.length - 1];\n    const batch = input.size / innerDimensionSize;\n\n    const input2D: Tensor2D = reshape(input, [batch, innerDimensionSize]);\n    const result = backend.ifft(input2D);\n    return reshape(result, input.shape);\n  }, inputs as {} as NamedTensorMap, null /* gradient */, IFFT);\n}\n\nexport const ifft = op({ifft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {reverse} from '../reverse';\nimport {scalar} from '../scalar';\nimport {slice} from '../slice';\n\nimport {ifft} from './ifft';\n\n/**\n * Inversed real value input fast Fourier transform.\n *\n * Computes the 1-dimensional inversed discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n * const imag = tf.tensor1d([0, 0, 0]);\n * const x = tf.complex(real, imag);\n *\n * x.irfft().print();\n * ```\n * @param input The real value input to compute an irfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction irfft_(input: Tensor): Tensor {\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n  let ret: Tensor;\n  if (innerDimensionSize <= 2) {\n    const complexInput = reshape(input, [batch, innerDimensionSize]);\n    ret = ifft(complexInput);\n  } else {\n    // The length of unique components of the DFT of a real-valued signal\n    // is 2 * (input_len - 1)\n    const outputShape = [batch, 2 * (innerDimensionSize - 1)];\n    const realInput = reshape(real(input), [batch, innerDimensionSize]);\n    const imagInput = reshape(imag(input), [batch, innerDimensionSize]);\n\n    const realConjugate =\n        reverse(slice(realInput, [0, 1], [batch, innerDimensionSize - 2]), 1);\n    const imagConjugate: Tensor2D = mul(\n        reverse(slice(imagInput, [0, 1], [batch, innerDimensionSize - 2]), 1),\n        scalar(-1));\n\n    const r = concat([realInput, realConjugate], 1);\n    const i = concat([imagInput, imagConjugate], 1);\n    const complexInput =\n        reshape(complex(r, i), [outputShape[0], outputShape[1]]);\n    ret = ifft(complexInput);\n  }\n  ret = real(ret);\n  // reshape the result if the input is 3D tensor.\n  if (input.rank === 3 && input.shape[0] !== 0) {\n    const temp = ret;\n    const batch = input.shape[0];\n    ret = reshape(ret, [batch, ret.shape[0] / batch, ret.shape[1]]);\n    temp.dispose();\n  }\n  return ret;\n}\n\nexport const irfft = op({irfft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {TensorInfo} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {assert} from '../util';\n\n/**\n * Prepare the split size array. When the input is a number, the axis is evenly\n * divided among the split size. When the input contains the negative value, the\n * rest of the axis is allocated toward that.\n */\nexport function prepareSplitSize(\n    x: Tensor|TensorInfo, numOrSizeSplits: number[]|number,\n    axis = 0): number[] {\n  let splitSizes = [];\n  if (typeof (numOrSizeSplits) === 'number') {\n    assert(\n        x.shape[axis] % numOrSizeSplits === 0,\n        () => 'Number of splits must evenly divide the axis.');\n    splitSizes =\n        new Array(numOrSizeSplits).fill(x.shape[axis] / numOrSizeSplits);\n  } else {\n    const numOfNegs = numOrSizeSplits.reduce((count, value) => {\n      if (value === -1) {\n        count += 1;\n      }\n      return count;\n    }, 0);\n    assert(\n        numOfNegs <= 1,\n        () => 'There should be only one negative value in split array.');\n    const negIndex = numOrSizeSplits.indexOf(-1);\n    // Allow the number of split array to be -1, which indicates the rest\n    // of dimension is allocated to that split.\n    if (negIndex !== -1) {\n      const total = numOrSizeSplits.reduce((a, b) => b > 0 ? a + b : a);\n      numOrSizeSplits[negIndex] = x.shape[axis] - total;\n    }\n    assert(\n        x.shape[axis] === numOrSizeSplits.reduce((a, b) => a + b),\n        () => 'The sum of sizes must match the size of the axis dimension.');\n    splitSizes = numOrSizeSplits;\n  }\n\n  return splitSizes;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SplitV, SplitVAttrs, SplitVInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {op} from './operation';\nimport {prepareSplitSize} from './split_util';\n\n/**\n * Splits a `tf.Tensor` into sub tensors.\n *\n * If `numOrSizeSplits` is a number, splits `x` along dimension `axis`\n * into `numOrSizeSplits` smaller tensors.\n * Requires that `numOrSizeSplits` evenly divides `x.shape[axis]`.\n *\n * If `numOrSizeSplits` is a number array, splits `x` into\n * `numOrSizeSplits.length` pieces. The shape of the `i`-th piece has the\n * same size as `x` except along dimension `axis` where the size is\n * `numOrSizeSplits[i]`.\n *\n * ```js\n * const x = tf.tensor2d([1, 2, 3, 4, 5, 6, 7, 8], [2, 4]);\n * const [a, b] = tf.split(x, 2, 1);\n * a.print();\n * b.print();\n *\n * const [c, d, e] = tf.split(x, [1, 2, 1], 1);\n * c.print();\n * d.print();\n * e.print();\n * ```\n *\n * @param x The input tensor to split.\n * @param numOrSizeSplits Either an integer indicating the number of\n * splits along the axis or an array of integers containing the sizes of\n * each output tensor along the axis. If a number then it must evenly divide\n * `x.shape[axis]`; otherwise the sum of sizes must match `x.shape[axis]`.\n * Can contain one -1 indicating that dimension is to be inferred.\n * @param axis The dimension along which to split. Defaults to 0 (the first\n * dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction split_<T extends Tensor>(\n    x: Tensor|TensorLike, numOrSizeSplits: number[]|number, axis = 0): T[] {\n  const $x = convertToTensor(x, 'x', 'split');\n\n  const forward: ForwardFunc<Tensor> = (backend, _) => {\n    const $axis = parseAxisParam(axis, $x.shape)[0];\n    const splitSizes = prepareSplitSize($x, numOrSizeSplits, $axis);\n    return backend.split($x, splitSizes, $axis) as {} as T;\n  };\n\n  const inputs: SplitVInputs = {x: $x};\n  const attr: SplitVAttrs = {numOrSizeSplits, axis};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */, SplitV,\n             attr as {} as NamedAttrMap) as {} as T[];\n}\n\nexport const split = op({split_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {assert} from '../../util';\nimport {complex} from '../complex';\nimport {concat} from '../concat';\nimport {imag} from '../imag';\nimport {op} from '../operation';\nimport {real} from '../real';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {split} from '../split';\nimport {zeros} from '../zeros';\nimport {zerosLike} from '../zeros_like';\n\nimport {fft} from './fft';\n\n/**\n * Real value input fast Fourier transform.\n *\n * Computes the 1-dimensional discrete Fourier transform over the\n * inner-most dimension of the real input.\n *\n * ```js\n * const real = tf.tensor1d([1, 2, 3]);\n *\n * real.rfft().print();\n * ```\n * @param input The real value input to compute an rfft over.\n *\n * @doc {heading: 'Operations', subheading: 'Spectral', namespace: 'spectral'}\n */\nfunction rfft_(input: Tensor, fftLength?: number): Tensor {\n  assert(\n      input.dtype === 'float32',\n      () => `The dtype for rfft() must be real value but got ${input.dtype}`);\n\n  let innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = input.size / innerDimensionSize;\n\n  let adjustedInput: Tensor;\n  if (fftLength != null && fftLength < innerDimensionSize) {\n    // Need to crop\n    const begin = input.shape.map(v => 0);\n    const size = input.shape.map(v => v);\n    size[input.shape.length - 1] = fftLength;\n    adjustedInput = slice(input, begin, size);\n    innerDimensionSize = fftLength;\n  } else if (fftLength != null && fftLength > innerDimensionSize) {\n    // Need to pad with zeros\n    const zerosShape = input.shape.map(v => v);\n    zerosShape[input.shape.length - 1] = fftLength - innerDimensionSize;\n    adjustedInput = concat([input, zeros(zerosShape)], input.shape.length - 1);\n    innerDimensionSize = fftLength;\n  } else {\n    adjustedInput = input;\n  }\n\n  // Complement the input with zero imaginary numbers.\n  const zerosInput = zerosLike(adjustedInput);\n  const complexInput =\n      reshape(complex(adjustedInput, zerosInput), [batch, innerDimensionSize]);\n\n  const ret = fft(complexInput);\n\n  // Exclude complex conjugations. These conjugations are put symmetrically.\n  const half = Math.floor(innerDimensionSize / 2) + 1;\n  const realValues = real(ret);\n  const imagValues = imag(ret);\n  const realComplexConjugate = split(\n      realValues, [half, innerDimensionSize - half],\n      realValues.shape.length - 1);\n  const imagComplexConjugate = split(\n      imagValues, [half, innerDimensionSize - half],\n      imagValues.shape.length - 1);\n\n  const outputShape = adjustedInput.shape.slice();\n  outputShape[adjustedInput.shape.length - 1] = half;\n\n  return reshape(\n      complex(realComplexConjugate[0], imagComplexConjugate[0]), outputShape);\n}\n\nexport const rfft = op({rfft_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Sqrt, SqrtInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes square root of the input `tf.Tensor` element-wise: `y = sqrt(x)`\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 4, -1]);\n *\n * x.sqrt().print();  // or tf.sqrt(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction sqrt_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'sqrt');\n\n  const inputs: SqrtInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.sqrt($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Sqrt);\n}\nexport const sqrt = op({sqrt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {SquaredDifference, SquaredDifferenceInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {assertAndGetBroadcastShape} from './broadcast_util';\nimport {op} from './operation';\n\n/**\n * Returns (a - b) * (a - b) element-wise.\n * Supports broadcasting.\n *\n * ```js\n * const a = tf.tensor1d([1, 4, 3, 16]);\n * const b = tf.tensor1d([1, 2, 9, 4]);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * ```js\n * // Broadcast squared difference  a with b.\n * const a = tf.tensor1d([2, 4, 6, 8]);\n * const b = tf.scalar(5);\n *\n * a.squaredDifference(b).print();  // or tf.squaredDifference(a, b)\n * ```\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n *\n * @doc {heading: 'Operations', subheading: 'Arithmetic'}\n */\nfunction squaredDifference_<T extends Tensor>(\n    a: Tensor|TensorLike, b: Tensor|TensorLike): T {\n  let $a = convertToTensor(a, 'a', 'squaredDifference');\n  let $b = convertToTensor(b, 'b', 'squaredDifference');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  assertAndGetBroadcastShape($a.shape, $b.shape);\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.squaredDifference($a, $b);\n    save([$a, $b]);\n    return res;\n  };\n\n  const inputs: SquaredDifferenceInputs = {a: $a, b: $b};\n  const attrs = {};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as unknown as NamedTensorMap, null /* grad */,\n             SquaredDifference, attrs) as T;\n}\n\nexport const squaredDifference = op({squaredDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {squeezeShape} from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Removes dimensions of size 1 from the shape of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor([1, 2, 3, 4], [1, 1, 4]);\n * x.squeeze().print();\n * ```\n *\n * @param x The input tensor to be squeezed.\n * @param axis An optional list of numbers. If specified, only\n *     squeezes the dimensions listed. The dimension index starts at 0. It\n * is an error to squeeze a dimension that is not 1.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction squeeze_<T extends Tensor>(x: Tensor|TensorLike, axis?: number[]): T {\n  const $x = convertToTensor(x, 'x', 'squeeze');\n  return reshape($x, squeezeShape($x.shape, axis).newShape) as T;\n}\n\nexport const squeeze = op({squeeze_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensorArray} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {concat} from './concat';\nimport {expandDims} from './expand_dims';\nimport {op} from './operation';\n\n/**\n * Stacks a list of rank-`R` `tf.Tensor`s into one rank-`(R+1)` `tf.Tensor`.\n *\n * ```js\n * const a = tf.tensor1d([1, 2]);\n * const b = tf.tensor1d([3, 4]);\n * const c = tf.tensor1d([5, 6]);\n * tf.stack([a, b, c]).print();\n * ```\n *\n * @param tensors A list of tensor objects with the same shape and dtype.\n * @param axis The axis to stack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction stack_<T extends Tensor>(\n    tensors: Array<T|TensorLike>, axis = 0): Tensor {\n  const $tensors = convertToTensorArray(tensors, 'tensors', 'stack');\n\n  util.assert(\n      $tensors.length >= 1, () => 'Pass at least one tensor to tf.stack');\n\n  if ($tensors.length === 1) {\n    return expandDims($tensors[0], axis);\n  }\n\n  const rank = $tensors[0].rank;\n  const shape = $tensors[0].shape;\n  const dtype = $tensors[0].dtype;\n\n  util.assert(axis <= rank, () => 'Axis must be <= rank of the tensor');\n\n  $tensors.forEach(t => {\n    util.assertShapesMatch(\n        shape, t.shape,\n        'All tensors passed to stack must have matching shapes');\n    util.assert(\n        dtype === t.dtype,\n        () => 'All tensors passed to stack must have matching dtypes');\n  });\n\n  const expandedTensors = $tensors.map(t => expandDims(t, axis));\n  // Stack exists in the TensorFlow C++ API\n  // (https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/stack) but not\n  // in\n  // https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/core/ops/ops.pbtxt.\n  // Therefore we are treating it like a high-level op rather than\n  // creating a dedicated stack kernel.\n  return concat(expandedTensors, axis);\n}\n\nexport const stack = op({stack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Step, StepAttrs, StepInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes step of the input `tf.Tensor` element-wise: `x > 0 ? 1 : alpha * x`\n *\n * ```js\n * const x = tf.tensor1d([0, 2, -1, -3]);\n *\n * x.step(.5).print();  // or tf.step(x, .5)\n * ```\n * @param x The input tensor.\n * @param alpha The gradient when input is negative.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction step_<T extends Tensor>(x: T|TensorLike, alpha = 0.0): T {\n  const $x = convertToTensor(x, 'x', 'step');\n\n  const inputs: StepInputs = {x: $x};\n  const attrs: StepAttrs = {alpha};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.step($x, alpha), inputs as {} as NamedTensorMap,\n      null /* grad */, Step, attrs as {} as NamedAttrMap);\n}\nexport const step = op({step_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {StridedSlice, StridedSliceAttrs, StridedSliceInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\nimport {slice} from './slice';\nimport {computeOutShape, getNormalizedAxes, maskToAxes} from './slice_util';\n\n/**\n * Extracts a strided slice of a tensor.\n *\n * Roughly speaking, this op extracts a slice of size (end-begin)/stride from\n * the given input tensor (x). Starting at the location specified by begin the\n * slice continues by adding stride to the index until all dimensions are not\n * less than end. Note that a stride can be negative, which causes a reverse\n * slice.\n *\n * ```js\n * const t = tf.tensor3d([1, 1, 1 ,2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6],\n *    [3, 2, 3]);\n * t.stridedSlice([1, 0, 0], [2, 1, 3], [1, 1, 1]).print()  // [[[3, 3, 3]]]\n * t.stridedSlice([1, 0, 0], [2, 2, 3], [1, 1, 1]).print()  // [[[3, 3, 3],\n *                                                     // [4, 4, 4]]]\n * t.stridedSlice([1, -1, 0], [2, -3, 3], [1, -1, 1]).print() // [[[4, 4, 4],\n *                                                     // [3, 3, 3]]]\n * ```\n *\n * @param x The tensor to stride slice.\n * @param begin The coordinates to start the slice from.\n * @param end: The coordinates to end the slice at.\n * @param strides: The size of the slice.\n * @param beginMask: If the ith bit of beginMask is set, begin[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param endMask: If the ith bit of endMask is set, end[i] is ignored\n *      and the fullest possible range in that dimension is used instead.\n * @param shrinkAxisMask: a bitmask where bit i implies that\n * the ith specification should shrink the dimensionality. begin and end must\n * imply a slice of size 1 in the dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction stridedSlice_(\n    x: Tensor|TensorLike, begin: number[], end: number[], strides?: number[],\n    beginMask = 0, endMask = 0, ellipsisMask = 0, newAxisMask = 0,\n    shrinkAxisMask = 0): Tensor {\n  let $x = convertToTensor(x, 'x', 'stridedSlice');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    if (strides == null) {\n      strides = new Array(begin.length);\n    }\n\n    const ellipsisAxes = maskToAxes(ellipsisMask);\n    if (ellipsisAxes.length > 1) {\n      throw new Error('Multiple ellipses in slice is not allowed.');\n    }\n\n    if (ellipsisMask !== 0 && newAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and newAxisMask is not yet supported.');\n    }\n\n    if (ellipsisMask !== 0 && shrinkAxisMask !== 0) {\n      throw new Error(\n          'Using both ellipsisMask and shrinkAxisMask is not yet supported.');\n    }\n\n    const numInterpolatedAxes = $x.rank - begin.length;\n\n    // Expand the dims of x based on the newAxisMask.\n    const expandAxes = maskToAxes(newAxisMask);\n    const newShape = $x.shape.slice();\n    expandAxes.forEach(axis => {\n      begin[axis] = 0;\n      end[axis] = 1;\n      newShape.splice(axis, 0, 1);\n    });\n    $x = reshape($x, newShape);\n\n    const {\n      begin: normalizedBegin,\n      end: normalizedEnd,\n      strides: normalizedStrides\n    } =\n        getNormalizedAxes(\n            $x.shape, ellipsisAxes, numInterpolatedAxes, begin, end, strides,\n            beginMask, endMask, ellipsisMask);\n    begin = normalizedBegin;\n    end = normalizedEnd;\n    strides = normalizedStrides;\n\n    const shrinkAxes = maskToAxes(shrinkAxisMask);\n    // Adjust the ends based on the shrink mask.\n    shrinkAxes.forEach(axis => {\n      end[axis] = begin[axis] + 1;\n      strides[axis] = 1;\n    });\n\n    // Figure out the output shape.\n    const size = computeOutShape(begin, end, strides);\n    // Remove the axes based on shrinkMask.\n    const outShape = size.filter((_, axis) => shrinkAxes.indexOf(axis) === -1);\n\n    const nonStrided = strides.every(v => v === 1);\n    if (nonStrided) {\n      return reshape(slice($x, begin, size), outShape);\n    }\n\n    const res = backend.stridedSlice($x, begin, end, strides);\n    return reshape(res, outShape);\n  };\n\n  const inputs: StridedSliceInputs = {x: $x};\n  const attrs: StridedSliceAttrs = {\n    begin,\n    end,\n    strides,\n    beginMask,\n    endMask,\n    ellipsisMask,\n    newAxisMask,\n    shrinkAxisMask\n  };\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, StridedSlice,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const stridedSlice = op({stridedSlice_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tan, TanInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Computes tan of the input `tf.Tensor` element-wise, `tan(x)`\n *\n * ```js\n * const x = tf.tensor1d([0, Math.PI / 2, Math.PI * 3 / 4]);\n *\n * x.tan().print();  // or tf.tan(x)\n * ```\n * @param x The input tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Basic math'}\n */\nfunction tan_<T extends Tensor>(x: T|TensorLike): T {\n  const $x = convertToTensor(x, 'x', 'tan');\n\n  const inputs: TanInputs = {x: $x};\n\n  return ENGINE.runKernelFunc((backend, save) => {\n    const res = backend.tan($x);\n    save([$x]);\n    return res;\n  }, inputs as {} as NamedTensorMap, null /* grad */, Tan);\n}\nexport const tan = op({tan_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor2D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike2D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-2 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor2d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor2d([[1, 2], [3, 4]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n *     `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor2d(\n    values: TensorLike2D, shape?: [number, number],\n    dtype?: DataType): Tensor2D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 2) {\n    throw new Error('tensor2d() requires shape to have two numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 2 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor2d() requires values to be number[][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor2d() requires shape to be provided when `values` ' +\n        'are a flat/TypedArray');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike4D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-4 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor4d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor4d([[[[1], [2]], [[3], [4]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor4d(\n    values: TensorLike4D, shape?: [number, number, number, number],\n    dtype?: DataType): Tensor4D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 4) {\n    throw new Error('tensor4d() requires shape to have four numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 4 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor4d() requires values to be number[][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor4d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor4D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor5D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike5D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-5 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor5d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor5d([[[[[1], [2]], [[3], [4]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor5d([1, 2, 3, 4, 5, 6, 7, 8], [1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor5d(\n    values: TensorLike5D, shape?: [number, number, number, number, number],\n    dtype?: DataType): Tensor5D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 5) {\n    throw new Error('tensor5d() requires shape to have five numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 5 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor5d() requires values to be ' +\n        'number[][][][][] or flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor5d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor5D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor6D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike6D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-6 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor6d` as it makes the code more readable.\n *\n *  ```js\n * // Pass a nested array.\n * tf.tensor6d([[[[[[1],[2]],[[3],[4]]],[[[5],[6]],[[7],[8]]]]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor6d([1, 2, 3, 4, 5, 6, 7, 8], [1, 1, 2, 2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n *     or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. Optional. If not provided,\n *   it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor6d(\n    values: TensorLike6D,\n    shape?: [number, number, number, number, number, number],\n    dtype?: DataType): Tensor6D {\n  assertNonNull(values);\n  if (shape != null && shape.length !== 6) {\n    throw new Error('tensor6d() requires shape to have six numbers');\n  }\n  const inferredShape = inferShape(values, dtype);\n  if (inferredShape.length !== 6 && inferredShape.length !== 1) {\n    throw new Error(\n        'tensor6d() requires values to be number[][][][][][] or ' +\n        'flat/TypedArray');\n  }\n  if (inferredShape.length === 1 && shape == null) {\n    throw new Error(\n        'tensor6d() requires shape to be provided when `values` ' +\n        'are a flat array');\n  }\n  shape = shape ||\n      inferredShape as [number, number, number, number, number, number];\n  return makeTensor(values, shape, inferredShape, dtype) as Tensor6D;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {TopK, TopKAttrs, TopKInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {NumericTensor, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Finds the values and indices of the `k` largest entries along the last\n * dimension.\n *\n * If the input is a vector (rank=1), finds the k largest entries in the vector\n * and outputs their values and indices as vectors. Thus values[j] is the j-th\n * largest entry in input, and its index is indices[j].\n * For higher rank inputs, computes the top k entries along the last dimension.\n *\n * If two elements are equal, the lower-index element appears first.\n *\n * ```js\n * const a = tf.tensor2d([[1, 5], [4, 3]]);\n * const {values, indices} = tf.topk(a);\n * values.print();\n * indices.print();\n * ```\n * @param x 1-D or higher `tf.Tensor` with last dimension being at least `k`.\n * @param k Number of top elements to look for along the last dimension.\n * @param sorted If true, the resulting `k` elements will be sorted by the\n *     values in descending order.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction topk_<T extends Tensor>(\n    x: T|TensorLike, k = 1, sorted = true): {values: T, indices: T} {\n  const $x = convertToTensor(x, 'x', 'topk');\n  if ($x.rank === 0) {\n    throw new Error('topk() expects the input to be of rank 1 or higher');\n  }\n  const lastDim = $x.shape[$x.shape.length - 1];\n  if (k > lastDim) {\n    throw new Error(\n        `'k' passed to topk() must be <= the last dimension (${lastDim}) ` +\n        `but got ${k}`);\n  }\n\n  const inputs: TopKInputs = {x: $x};\n  const attrs: TopKAttrs = {k, sorted};\n\n  const [values, indices] = ENGINE.runKernelFunc(\n      b => b.topk($x as NumericTensor, k, sorted),\n      inputs as {} as NamedTensorMap, null /* grad */, TopK,\n      attrs as {} as NamedAttrMap);\n\n  return {values, indices} as {values: T, indices: T};\n}\n\nexport const topk = op({topk_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {buffer} from './buffer';\nimport {op} from './operation';\nimport {MPRandGauss} from './rand_util';\n\n/**\n * Creates a `tf.Tensor` with values sampled from a truncated normal\n * distribution.\n *\n * ```js\n * tf.truncatedNormal([2, 2]).print();\n * ```\n *\n * The generated values follow a normal distribution with specified mean and\n * standard deviation, except that values whose magnitude is more than 2\n * standard deviations from the mean are dropped and re-picked.\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param mean The mean of the normal distribution.\n * @param stdDev The standard deviation of the normal distribution.\n * @param dtype The data type of the output tensor.\n * @param seed The seed for the random number generator.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction truncatedNormal_<R extends Rank>(\n    shape: ShapeMap[R], mean = 0, stdDev = 1, dtype?: 'float32'|'int32',\n    seed?: number): Tensor<R> {\n  if (dtype != null && (dtype as DataType) === 'bool') {\n    throw new Error(`Unsupported data type $ { dtype }`);\n  }\n  const randGauss =\n      new MPRandGauss(mean, stdDev, dtype, true /* truncated */, seed);\n  const res = buffer(shape, dtype);\n  for (let i = 0; i < res.values.length; i++) {\n    res.values[i] = randGauss.nextValue();\n  }\n  return res.toTensor();\n}\n\nexport const truncatedNormal = op({truncatedNormal_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Unique, UniqueAttrs, UniqueInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Finds unique elements along an axis of a tensor.\n *\n * It returns a tensor `values` containing all of the unique elements along the\n * `axis` of the given tensor `x` in the same order that they occur along the\n * `axis` in `x`; `x` does not need to be sorted. It also returns a tensor\n * `indices` the same size as the number of the elements in `x` along the `axis`\n * dimension. It contains the index in the unique output `values`.\n *\n * ```js\n * // A 1-D tensor\n * const a = tf.tensor1d([1, 1, 2, 4, 4, 4, 7, 8, 8]);\n * const {values, indices} = tf.unique(a);\n * values.print();   // [1, 2, 4, 7, 8,]\n * indices.print();  // [0, 0, 1, 2, 2, 2, 3, 4, 4]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=0\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 0)\n * values.print();   // [[1, 0, 0],\n *                   //  [2, 0, 0]]\n * indices.print();  // [0, 0, 1]\n * ```\n *\n * ```js\n * // A 2-D tensor with axis=1\n * //\n * // 'a' is: [[1, 0, 0],\n * //          [1, 0, 0],\n * //          [2, 0, 0]]\n * const a = tf.tensor2d([[1, 0, 0], [1, 0, 0], [2, 0, 0]]);\n * const {values, indices} = tf.unique(a, 1)\n * values.print();   // [[1, 0],\n *                   //  [1, 0],\n *                   //  [2, 0]]\n * indices.print();  // [0, 1, 1]\n * ```\n * @param x A tensor (int32, string, bool).\n * @param axis The axis of the tensor to find the unique elements.\n * @returns [uniqueElements, indices] (see above for details)\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nfunction unique_<T extends Tensor>(\n    x: T|TensorLike, axis = 0): {values: T, indices: Tensor1D} {\n  // x can be of any dtype, thus null as the last argument.\n  const $x = convertToTensor(x, 'x', 'unique', null);\n  assert($x.rank > 0, () => 'The input tensor must be at least 1D');\n\n  const inputs: UniqueInputs = {x: $x};\n  const attrs: UniqueAttrs = {axis};\n  const [values, indices] = ENGINE.runKernel(\n                                Unique, inputs as {} as NamedTensorMap,\n                                attrs as {} as NamedAttrMap) as [T, Tensor1D];\n  return {values, indices};\n}\n\nexport const unique = op({unique_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {UnsortedSegmentSum, UnsortedSegmentSumAttrs, UnsortedSegmentSumInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, isInt} from '../util';\n\nimport {op} from './operation';\n\n/**\n * Computes the sum along segments of a `tf.Tensor`.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * const segmentIds = tf.tensor1d([1, 2, 0, 1], 'int32');\n * const numSegments = 3;\n *\n * x.unsortedSegmentSum(segmentIds, numSegments).print()\n * //or tf.unsortedSegmentSum(x, segmentIds, numSegments)\n * ```\n * @param x The `tf.Tensor` that will be summed along its segments.\n * @param segmentIds A `tf.Tensor1D` whose rank is equal to the rank of `x`'s\n * dimension along the `axis`.  Maps each element of `x` to a segment.\n * @param numSegments The number of distinct `segmentIds`.\n *\n * @doc {heading: 'Operations', subheading: 'Segment'}\n */\nfunction unsortedSegmentSum_<T extends Tensor>(\n    x: T|TensorLike, segmentIds: Tensor1D|TensorLike, numSegments: number): T {\n  const $x = convertToTensor(x, 'x', 'unsortedSegmentSum');\n  const $segmentIds =\n      convertToTensor(segmentIds, 'segmentIds', 'unsortedSegmentSum', 'int32');\n  assert(isInt(numSegments), () => 'numSegments must be of dtype int');\n\n  const inputs: UnsortedSegmentSumInputs = {x: $x, segmentIds: $segmentIds};\n  const attrs: UnsortedSegmentSumAttrs = {numSegments};\n\n  const forward: ForwardFunc<Tensor> = (backend, save) => {\n    const res = backend.unsortedSegmentSum($x, $segmentIds, numSegments);\n    save([$segmentIds]);\n    return res;\n  };\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             UnsortedSegmentSum, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const unsortedSegmentSum = op({unsortedSegmentSum_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Unpack, UnpackAttrs, UnpackInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Unstacks a `tf.Tensor` of rank-`R` into a list of rank-`(R-1)` `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * tf.unstack(a).forEach(tensor => tensor.print());\n * ```\n *\n * @param x A tensor object.\n * @param axis The axis to unstack along. Defaults to 0 (the first dim).\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nfunction unstack_(x: Tensor|TensorLike, axis = 0): Tensor[] {\n  const $x = convertToTensor(x, 'x', 'unstack');\n  util.assert(\n      axis >= -$x.shape.length && axis < $x.shape.length,\n      () =>\n          `Axis = ${axis} is not in [-${$x.shape.length}, ${$x.shape.length})`);\n  if (axis < 0) {\n    axis += $x.shape.length;\n  }\n  const inputs: UnpackInputs = {value: $x};\n  const attrs: UnpackAttrs = {axis};\n  const forward: ForwardFunc<Tensor[]> = (backend: KernelBackend) =>\n      backend.unstack($x, axis);\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, Unpack,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const unstack = op({unstack_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor, Variable} from '../tensor';\nimport {DataType, Rank} from '../types';\n\n/**\n * Creates a new variable with the provided initial value.\n * ```js\n * const x = tf.variable(tf.tensor([1, 2, 3]));\n * x.assign(tf.tensor([4, 5, 6]));\n *\n * x.print();\n * ```\n *\n * @param initialValue Initial value for the tensor.\n * @param trainable If true, optimizers are allowed to update it.\n * @param name Name of the variable. Defaults to a unique id.\n * @param dtype If set, initialValue will be converted to the given type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function variable<R extends Rank>(\n    initialValue: Tensor<R>, trainable = true, name?: string,\n    dtype?: DataType): Variable<R> {\n  return ENGINE.makeVariable(initialValue, trainable, name, dtype) as\n      Variable<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the Where kernel shared between cpu and webgl */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor2D} from '../tensor';\nimport {TypedArray} from '../types';\n\nexport function whereImpl(condShape: number[], condVals: TypedArray): Tensor2D {\n  const indices = [];\n  for (let i = 0; i < condVals.length; i++) {\n    if (condVals[i]) {\n      indices.push(i);\n    }\n  }\n\n  const inBuffer = buffer(condShape, 'int32');\n\n  const out = buffer([indices.length, condShape.length], 'int32');\n  for (let i = 0; i < indices.length; i++) {\n    const loc = inBuffer.indexToLoc(indices[i]);\n    const offset = i * condShape.length;\n    out.values.set(loc, offset);\n  }\n  return out.toTensor() as Tensor2D;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {whereImpl} from '../backends/where_impl';\nimport {Tensor, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\n/**\n * Returns the coordinates of true elements of condition.\n *\n * The coordinates are returned in a 2-D tensor where the first dimension (rows)\n * represents the number of true elements, and the second dimension (columns)\n * represents the coordinates of the true elements. Keep in mind, the shape of\n * the output tensor can vary depending on how many true values there are in\n * input. Indices are output in row-major order. The resulting tensor has the\n * shape `[numTrueElems, condition.rank]`.\n *\n * This is analogous to calling the python `tf.where(cond)` without an x or y.\n *\n * ```js\n * const cond = tf.tensor1d([false, false, true], 'bool');\n * const result = await tf.whereAsync(cond);\n * result.print();\n * ```\n *\n * @doc {heading: 'Operations', subheading: 'Logical'}\n */\nasync function whereAsync_(condition: Tensor|TensorLike): Promise<Tensor2D> {\n  const $condition =\n      convertToTensor(condition, 'condition', 'whereAsync', 'bool');\n  const vals = await $condition.data();\n  const res = whereImpl($condition.shape, vals);\n  if (condition !== $condition) {\n    $condition.dispose();\n  }\n  return res;\n}\n\nexport const whereAsync = whereAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {gather} from './gather';\nimport {reshape} from './reshape';\nimport {squeeze} from './squeeze';\nimport {whereAsync} from './where_async';\n\n/**\n * Apply boolean mask to tensor.\n *\n * ```js\n * const tensor = tf.tensor2d([1, 2, 3, 4, 5, 6], [3, 2]);\n * const mask = tf.tensor1d([1, 0, 1], 'bool');\n * const result = await tf.booleanMaskAsync(tensor, mask);\n * result.print();\n * ```\n *\n * @param tensor N-D tensor.\n * @param mask K-D boolean tensor, K <= N and K must be known statically.\n * @param axis A 0-D int Tensor representing the axis in tensor to mask from.\n *     By default, axis is 0 which will mask from the first dimension.\n *     Otherwise K + axis <= N.\n *\n * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}\n */\nasync function booleanMaskAsync_(\n    tensor: Tensor|TensorLike, mask: Tensor|TensorLike,\n    axis?: number): Promise<Tensor> {\n  const $tensor = convertToTensor(tensor, 'tensor', 'boolMask');\n  const $mask = convertToTensor(mask, 'mask', 'boolMask', 'bool');\n\n  const axisFrom = axis == null ? 0 : axis;\n  const maskDim = $mask.rank;\n  const tensorShape = $tensor.shape;\n\n  util.assert(maskDim > 0, () => 'mask cannot be scalar');\n  util.assertShapesMatch(\n      tensorShape.slice(axisFrom, axisFrom + maskDim), $mask.shape,\n      `mask's shape must match the first K dimensions of tensor's shape,`);\n\n  let leadingSize = 1;\n  for (let i = axisFrom; i < axisFrom + maskDim; i++) {\n    leadingSize *= tensorShape[i];\n  }\n  const targetTensorShape =\n      tensorShape.slice(0, axisFrom)\n          .concat([leadingSize], tensorShape.slice(axisFrom + maskDim));\n  const reshapedTensor = reshape($tensor, targetTensorShape);\n  const reshapedMask = reshape($mask, [-1]);\n  const positivePositions = await whereAsync(reshapedMask);\n  const indices = squeeze(positivePositions, [1]);\n\n  const res = gather(reshapedTensor, indices, axisFrom);\n\n  // Ensure no memory leak.\n  if (tensor !== $tensor) {\n    $tensor.dispose();\n  }\n  if (mask !== $mask) {\n    $mask.dispose();\n  }\n  indices.dispose();\n  reshapedTensor.dispose();\n  reshapedMask.dispose();\n  positivePositions.dispose();\n\n  return res;\n}\n\nexport const booleanMaskAsync = booleanMaskAsync_;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assertShapesMatch} from '../util';\n\nimport {equal} from './equal';\nimport {greater} from './greater';\nimport {greaterEqual} from './greater_equal';\nimport {less} from './less';\nimport {lessEqual} from './less_equal';\nimport {notEqual} from './not_equal';\nimport {op} from './operation';\n\n/**\n * @deprecated\n * Strict version of `tf.notEqual` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction notEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'notEqualStrict');\n  const $b = convertToTensor(b, 'b', 'notEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in notEqualStrict: ');\n  return notEqual($a, $b);\n}\n\n/**\n * @deprecated\n * Strict version of `tf.less` that forces `a` and `b` to be of the same\n * shape.\n *\n * @param a The first input tensor.\n * @param b The second input tensor. Must have the same shape and dtype as\n *     `a`.\n */\nfunction lessStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessStrict');\n  const $b = convertToTensor(b, 'b', 'lessStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessStrict: ');\n  return less($a, $b);\n}\n\nfunction equalStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'equalStrict');\n  const $b = convertToTensor(b, 'b', 'equalStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in equalStrict: ');\n  return equal($a, $b);\n}\n\nfunction lessEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'lessEqualStrict');\n  const $b = convertToTensor(b, 'b', 'lessEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in lessEqualStrict: ');\n  return lessEqual($a, $b);\n}\n\nfunction greaterStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterStrict');\n  const $b = convertToTensor(b, 'b', 'greaterStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterStrict: ');\n  return greater($a, $b);\n}\n\nfunction greaterEqualStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'greaterEqualStrict');\n  const $b = convertToTensor(b, 'b', 'greaterEqualStrict');\n  assertShapesMatch($a.shape, $b.shape, 'Error in greaterEqualStrict: ');\n  return greaterEqual($a, $b);\n}\n\nexport const equalStrict = op({equalStrict_});\nexport const greaterEqualStrict = op({greaterEqualStrict_});\nexport const greaterStrict = op({greaterStrict_});\nexport const lessEqualStrict = op({lessEqualStrict_});\nexport const lessStrict = op({lessStrict_});\nexport const notEqualStrict = op({notEqualStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {deprecationWarn} from '../globals';\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {maximum} from './maximum';\nimport {minimum} from './minimum';\nimport {mod} from './mod';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {squaredDifference} from './squared_difference';\nimport {sub} from './sub';\n\n/**\n * @deprecated\n * Adds two `tf.Tensor`s element-wise, A + B.\n *\n * Inputs must be the same shape. For broadcasting support, use add() instead.\n *\n * @param a The first Tensor to add element-wise.\n * @param b The second Tensor to add element-wise.\n */\nfunction addStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'addStrict');\n  const $b = convertToTensor(b, 'b', 'addStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in addStrict: ');\n  return add($a, $b);\n}\n\n/**\n * @deprecated\n * Subtracts two `tf.Tensor`s element-wise, A - B. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.sub` instead.\n *\n * @param a The first Tensor to subtract element-wise.\n * @param b The second Tensor to subtract element-wise.\n */\nfunction subStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'subStrict');\n  const $b = convertToTensor(b, 'b', 'subStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in subStrict: ');\n  return sub($a, $b);\n}\n\n/**\n * @deprecated\n * Computes the power of one `tf.Tensor` to another. Inputs must\n * be the same shape.\n *\n * For broadcasting support, use `tf.pow` instead.\n *\n * @param base The base tensor to pow element-wise.\n * @param exp The exponent tensor to pow element-wise.\n */\nfunction powStrict_<T extends Tensor>(base: T, exp: Tensor): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  util.assertShapesMatch(base.shape, exp.shape, 'Error in powStrict: ');\n  return pow(base, exp);\n}\n\n/**\n * @deprecated\n * Multiplies two `tf.Tensor`s element-wise, A * B.\n *\n * Inputs must be the same shape. For broadcasting support, use `tf.mul`.\n *\n * @param a The first tensor to multiply.\n * @param b The first tensor to multiply. Must have the same\n *    dtype as `a`.\n */\nfunction mulStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'mul');\n  const $b = convertToTensor(b, 'b', 'mul');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in multiplyStrict: ');\n  return mul($a, $b);\n}\n\n/**\n * @deprecated\n * Divides two `tf.Tensor`s element-wise, A / B. Inputs must\n * be the same shape.\n *\n * @param a The first tensor as the numerator for element-wise division.\n * @param b The second tensor as the denominator for element-wise division.\n */\nfunction divStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'div');\n  const $b = convertToTensor(b, 'b', 'div');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in divideStrict: ');\n  return div($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the mod of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use mod().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction modStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'modStrict');\n  const $b = convertToTensor(b, 'b', 'modStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in modStrict: ');\n  return mod($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the min of a and b (`a < b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use minimum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction minimumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'minimumStrict');\n  const $b = convertToTensor(b, 'b', 'minimumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in minimumStrict: ');\n  return minimum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns the max of a and b (`a > b ? a : b`) element-wise. Inputs must\n * be the same shape. For broadcasting support, use maximum().\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same dtype as `a`.\n */\nfunction maximumStrict_<T extends Tensor>(a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n\n  const $a = convertToTensor(a, 'a', 'maximumStrict');\n  const $b = convertToTensor(b, 'b', 'maximumStrict');\n  util.assertShapesMatch($a.shape, $b.shape, 'Error in maximumStrict: ');\n  return maximum($a, $b);\n}\n\n/**\n * @deprecated\n * Returns (a - b) * (a - b) element-wise.\n *\n * Inputs must be the same shape. For broadcasting support, use\n * `tf.squaredDifference` instead.\n *\n * @param a The first tensor.\n * @param b The second tensor. Must have the same type as `a`.\n */\nfunction squaredDifferenceStrict_<T extends Tensor>(\n    a: T|TensorLike, b: T|TensorLike): T {\n  deprecationWarn(\n      'strict variants of ops have been deprecated ' +\n      'and will be removed in future');\n  const $a = convertToTensor(a, 'a', 'squaredDifferenceStrict');\n  const $b = convertToTensor(b, 'b', 'squaredDifferenceStrict');\n  util.assertShapesMatch(\n      $a.shape, $b.shape, 'Error in squaredDifferenceStrict: ');\n  return squaredDifference($a, $b);\n}\n\nexport const addStrict = op({addStrict_});\nexport const divStrict = op({divStrict_});\nexport const maximumStrict = op({maximumStrict_});\nexport const minimumStrict = op({minimumStrict_});\nexport const modStrict = op({modStrict_});\nexport const mulStrict = op({mulStrict_});\nexport const powStrict = op({powStrict_});\nexport const squaredDifferenceStrict = op({squaredDifferenceStrict_});\nexport const subStrict = op({subStrict_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {parseAxisParam} from '../util';\n\nimport {abs} from './abs';\nimport * as axis_util from './axis_util';\nimport {max} from './max';\nimport {min} from './min';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {reshape} from './reshape';\nimport {scalar} from './scalar';\nimport {sqrt} from './sqrt';\nimport {square} from './square';\nimport {sum} from './sum';\n\n/**\n * Computes the norm of scalar, vectors, and matrices.\n * This function can compute several different vector norms (the 1-norm, the\n * Euclidean or 2-norm, the inf-norm, and in general the p-norm for p > 0)\n * and matrix norms (Frobenius, 1-norm, and inf-norm).\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n *\n * x.norm().print();  // or tf.norm(x)\n * ```\n *\n * @param x The input array.\n * @param ord Optional. Order of the norm. Supported norm types are\n * following:\n *\n *  | ord        | norm for matrices         | norm for vectors\n *  |------------|---------------------------|---------------------\n *  |'euclidean' |Frobenius norm             |2-norm\n *  |'fro'       |Frobenius norm\t           |\n *  |Infinity    |max(sum(abs(x), axis=1))   |max(abs(x))\n *  |-Infinity   |min(sum(abs(x), axis=1))   |min(abs(x))\n *  |1           |max(sum(abs(x), axis=0))   |sum(abs(x))\n *  |2           |                           |sum(abs(x)^2)^1/2*\n *\n * @param axis Optional. If axis is null (the default), the input is\n * considered a vector and a single vector norm is computed over the entire\n * set of values in the Tensor, i.e. norm(x, ord) is equivalent\n * to norm(x.reshape([-1]), ord). If axis is a integer, the input\n * is considered a batch of vectors, and axis determines the axis in x\n * over which to compute vector norms. If axis is a 2-tuple of integer it is\n * considered a batch of matrices and axis determines the axes in NDArray\n * over which to compute a matrix norm.\n * @param keepDims Optional. If true, the norm have the same dimensionality\n * as the input.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction norm_(\n    x: Tensor|TensorLike, ord: number|'euclidean'|'fro' = 'euclidean',\n    axis: number|number[] = null, keepDims = false): Tensor {\n  x = convertToTensor(x, 'x', 'norm');\n\n  const norm = normImpl(x, ord, axis);\n  let keepDimsShape = norm.shape;\n  if (keepDims) {\n    const axes = parseAxisParam(axis, x.shape);\n    keepDimsShape = axis_util.expandShapeToKeepDim(norm.shape, axes);\n  }\n  return reshape(norm, keepDimsShape);\n}\n\nfunction normImpl(\n    x: Tensor, p: number|string, axis: number|number[] = null): Tensor {\n  if (x.rank === 0) {\n    return abs(x);\n  }\n\n  // consider vector when no axis is specified\n  if (x.rank !== 1 && axis === null) {\n    return normImpl(reshape(x, [-1]), p, axis);\n  }\n\n  // vector\n  if (x.rank === 1 || typeof axis === 'number' ||\n      Array.isArray(axis) && axis.length === 1) {\n    if (p === 1) {\n      return sum(abs(x), axis);\n    }\n    if (p === Infinity) {\n      return max(abs(x), axis);\n    }\n    if (p === -Infinity) {\n      return min(abs(x), axis);\n    }\n    if (p === 'euclidean' || p === 2) {\n      // norm(x, 2) = sum(abs(xi) ^ 2) ^ 1/2\n      return sqrt(sum(pow(abs(x), scalar(2, 'int32')), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  // matrix (assumption axis[0] < axis[1])\n  if (Array.isArray(axis) && axis.length === 2) {\n    if (p === 1) {\n      return max(sum(abs(x), axis[0]), axis[1] - 1);\n    }\n    if (p === Infinity) {\n      return max(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === -Infinity) {\n      return min(sum(abs(x), axis[1]), axis[0]);\n    }\n    if (p === 'fro' || p === 'euclidean') {\n      // norm(x) = sqrt(sum(pow(x, 2)))\n      return sqrt(sum(square(x), axis));\n    }\n\n    throw new Error(`Error in norm: invalid ord value: ${p}`);\n  }\n\n  throw new Error(`Error in norm: invalid axis: ${axis}`);\n}\n\nexport const norm = op({norm_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor} from '../tensor';\nimport {assertTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {pow} from './pow';\nimport {scalar} from './scalar';\nimport {sub} from './sub';\n\n/**\n * Compute the moving average of a variable.\n *\n * Without zeroDebias, the moving average operation is defined by:\n *   `v += delta`\n * where\n *   `delta = (1 - decay) * (x - v)`\n *\n * With zeroDebias (default), the `delta` term is scaled to debias the\n * effect of the (assumed) zero-initialization of `v`.\n *   `delta /= (1 - decay ^ step)`\n *\n * For more details on the zero-debiasing algorithm, see:\n *   https://arxiv.org/abs/1412.6980\n *\n * Note that this function is completely stateless and does not keep track of\n * step count. The step count needs to be maintained by the caller and passed\n * in as `step`.\n *\n * @param v The current moving average value.\n * @param x New input value, must have the same shape and dtype as `v`.\n * @param decay The decay factor. Typical values are 0.95 and 0.99.\n * @param step Step count.\n * @param zeroDebias: Whether zeroDebias is to be performed (default: `true`).\n * @returns The new moving average value.\n *\n * @doc {heading: 'Operations', subheading: 'Moving Average'}\n */\nfunction movingAverage_<T extends Tensor>(\n    v: T|TensorLike, x: T|TensorLike, decay: number|Scalar,\n    step?: number|Scalar, zeroDebias = true): T {\n  const $v = convertToTensor(v, 'v', 'movingAverage');\n  const $x = convertToTensor(x, 'x', 'movingAverage');\n  const $decay = convertToTensor(decay, 'decay', 'movingAverage');\n\n  assertTypesMatch($v, $x);\n  util.assert(\n      util.arraysEqual($v.shape, $x.shape), () => 'Shape mismatch in v and x');\n\n  const one = scalar(1);\n  const oneMinusDecay = sub(one, $decay);\n\n  let update = mul(sub($x, $v), oneMinusDecay);\n  if (zeroDebias) {\n    util.assert(\n        step != null, () => 'When using zeroDebias: true, step is required.');\n    const $step = convertToTensor(step, 'step', 'movingAverage');\n    update = div(update, sub(one, pow($decay, $step)));\n  }\n  return add($v, update);\n}\n\nexport const movingAverage = op({movingAverage_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ScatterNd, ScatterNdAttrs, ScatterNdInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\nimport * as scatter_nd_util from './scatter_nd_util';\n\n/**\n * Creates a new tensor by applying sparse updates to individual\n * values or slices within a zero tensor of the given shape tensor according to\n * indices. This operator is the inverse of the `tf.gatherND` operator which\n * extracts values or slices from a given tensor.\n *\n * ```js\n * const indices = tf.tensor2d([4, 3, 1, 7], [4, 1], 'int32');\n * const updates = tf.tensor1d([9, 10, 11, 12]);\n * const shape = [8];\n * tf.scatterND(indices, updates, shape).print() //[0, 11, 0, 10, 9, 0, 0, 12]\n * ```\n *\n * @param indices The tensor contains the indices into the output tensor.\n * @param updates The tensor contains the value for the indices.\n * @param shape: The shape of the output tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction scatterND_<R extends Rank>(\n    indices: Tensor|TensorLike, updates: Tensor|TensorLike,\n    shape: ShapeMap[R]): Tensor<R> {\n  const $indices = convertToTensor(indices, 'indices', 'scatterND', 'int32');\n  const $updates = convertToTensor(updates, 'updates', 'scatterND');\n  scatter_nd_util.validateInput($updates, $indices, shape);\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.scatterND($indices, $updates, shape);\n  };\n\n  const inputs: ScatterNdInputs = {indices: $indices, updates: $updates};\n  const attrs: ScatterNdAttrs = {shape};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             ScatterNd, attrs as {} as NamedAttrMap) as Tensor<R>;\n}\n\nexport const scatterND = op({scatterND_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../tensor';\n\n/**\n * Validate sparseToDense inputs.\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape number[]. Shape of the dense output tensor.\n * @param validateIndices boolean. indice validation is not supported, error\n * will be thrown if it is set.\n */\nexport function validateInput(\n    sparseIndices: Tensor, sparseValues: Tensor, outputShape: number[],\n    defaultValues: Tensor) {\n  if (sparseIndices.dtype !== 'int32') {\n    throw new Error(\n        'tf.sparseToDense() expects the indices to be int32 type,' +\n        ` but the dtype was ${sparseIndices.dtype}.`);\n  }\n  if (sparseIndices.rank > 2) {\n    throw new Error(\n        'sparseIndices should be a scalar, vector, or matrix,' +\n        ` but got shape ${sparseIndices.shape}.`);\n  }\n\n  const numElems = sparseIndices.rank > 0 ? sparseIndices.shape[0] : 1;\n  const numDims = sparseIndices.rank > 1 ? sparseIndices.shape[1] : 1;\n\n  if (outputShape.length !== numDims) {\n    throw new Error(\n        'outputShape has incorrect number of elements:,' +\n        ` ${outputShape.length}, should be: ${numDims}.`);\n  }\n\n  const numValues = sparseValues.size;\n  if (!(sparseValues.rank === 0 ||\n        sparseValues.rank === 1 && numValues === numElems)) {\n    throw new Error(\n        'sparseValues has incorrect shape ' +\n        `${sparseValues.shape}, should be [] or [${numElems}]`);\n  }\n\n  if (sparseValues.dtype !== defaultValues.dtype) {\n    throw new Error('sparseValues.dtype must match defaultValues.dtype');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {SparseToDense, SparseToDenseAttrs, SparseToDenseInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport * as sparse_to_dense from '../ops/sparse_to_dense_util';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ScalarLike, ShapeMap, TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Converts a sparse representation into a dense tensor.\n *\n * Builds an array dense with shape outputShape such that:\n *\n * // If sparseIndices is scalar\n * dense[i] = (i == sparseIndices ? sparseValues : defaultValue)\n *\n * // If sparseIndices is a vector, then for each i\n * dense[sparseIndices[i]] = sparseValues[i]\n *\n * // If sparseIndices is an n by d matrix, then for each i in [0, n)\n * dense[sparseIndices[i][0], ..., sparseIndices[i][d-1]] = sparseValues[i]\n * All other values in dense are set to defaultValue. If sparseValues is a\n * scalar, all sparse indices are set to this single value.\n *\n * If indices are repeated the final value is summed over all values for those\n * indices.\n *\n * ```js\n * const indices = tf.tensor1d([4, 5, 6, 1, 2, 3], 'int32');\n * const values = tf.tensor1d([10, 11, 12, 13, 14, 15], 'float32');\n * const shape = [8];\n * tf.sparseToDense(indices, values, shape).print();\n * ```\n *\n * @param sparseIndices A 0-D, 1-D, or 2-D Tensor of type int32.\n * sparseIndices[i] contains the complete index where sparseValues[i] will be\n * placed.\n * @param sparseValues A 0-D or 1-D Tensor. Values\n * corresponding to each row of sparseIndices, or a scalar value to be used for\n * all sparse indices.\n * @param outputShape Shape of the dense output tensor. the type is inferred.\n * @param defaultValue Scalar. Value to set for indices not specified in\n * sparseIndices. Defaults to zero.\n *\n * @doc {heading: 'Operations', subheading: 'Normalization'}\n */\nfunction sparseToDense_<R extends Rank>(\n    sparseIndices: Tensor|TensorLike, sparseValues: Tensor|TensorLike,\n    outputShape: ShapeMap[R], defaultValue: Scalar|ScalarLike = 0): Tensor<R> {\n  const $sparseIndices =\n      convertToTensor(sparseIndices, 'sparseIndices', 'sparseToDense', 'int32');\n  const $sparseValues =\n      convertToTensor(sparseValues, 'sparseValues', 'sparseToDense');\n  const $defaultValue = convertToTensor(\n      defaultValue, 'defaultValue', 'sparseToDense', $sparseValues.dtype);\n\n  sparse_to_dense.validateInput(\n      $sparseIndices, $sparseValues, outputShape, $defaultValue);\n\n  const inputs: SparseToDenseInputs = {\n    sparseIndices: $sparseIndices,\n    sparseValues: $sparseValues,\n    defaultValue: $defaultValue\n  };\n\n  const attrs: SparseToDenseAttrs = {outputShape};\n\n  return ENGINE.runKernelFunc(\n      backend => backend.sparseToDense(\n          $sparseIndices, $sparseValues, outputShape, $defaultValue),\n      inputs as {} as NamedTensorMap, null /* grad */, SparseToDense,\n      attrs as {} as NamedAttrMap);\n}\n\nexport const sparseToDense = op({sparseToDense_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {GatherNd, GatherNdInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {op} from './operation';\n\n/**\n * Gather slices from input tensor into a Tensor with shape specified by\n * `indices`.\n *\n * `indices` is an K-dimensional integer tensor, best thought of as a\n * (K-1)-dimensional tensor of indices into input, where each element defines a\n * slice of input:\n * output[\\\\(i_0, ..., i_{K-2}\\\\)] = input[indices[\\\\(i_0, ..., i_{K-2}\\\\)]]\n *\n * Whereas in `tf.gather`, `indices` defines slices into the first dimension of\n * input, in `tf.gatherND`, `indices` defines slices into the first N dimensions\n * of input, where N = indices.shape[-1].\n *\n * The last dimension of indices can be at most the rank of input:\n * indices.shape[-1] <= input.rank\n *\n * The last dimension of `indices` corresponds to elements\n * (if indices.shape[-1] == input.rank) or slices\n * (if indices.shape[-1] < input.rank) along dimension indices.shape[-1] of\n * input.\n * The output tensor has shape\n * indices.shape[:-1] + input.shape[indices.shape[-1]:]\n *\n * Note that on CPU, if an out of bound index is found, an error is returned. On\n * GPU, if an out of bound index is found, a 0 is stored in the corresponding\n * output value.\n *\n * ```js\n * const indices = tf.tensor2d([0, 1, 1, 0], [2,2], 'int32');\n * const input = tf.tensor2d([9, 10, 11, 12], [2, 2]);\n * tf.gatherND(input, indices).print() // [10, 11]\n * ```\n *\n * @param x The tensor from which to gather values.\n * @param indices Index tensor, must be of type int32.\n *\n * @doc {heading: 'Operations', subheading: 'Slicing and Joining'}\n */\nfunction gatherND_(x: Tensor|TensorLike, indices: Tensor|TensorLike): Tensor {\n  const $indices = convertToTensor(indices, 'indices', 'gatherND', 'int32');\n  const $x = convertToTensor(x, 'x', 'gatherND');\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    return backend.gatherND($x, $indices);\n  };\n\n  const inputs: GatherNdInputs = {params: $x, indices: $indices};\n\n  return ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */, GatherNd);\n}\n\nexport const gatherND = op({gatherND_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\n/**\n * Normalize noise shape based on provided tensor and noise shape.\n *\n * @param x Tensor.\n * @param noiseShape The shape for the randomly generated keep/drop flags, as\n *   an array of numbers. Optional.\n * @returns Normalized noise shape.\n */\nexport function getNoiseShape(x: Tensor, noiseShape?: number[]): number[] {\n  if (noiseShape == null) {\n    return x.shape.slice();\n  }\n  if (util.arraysEqual(x.shape, noiseShape)) {\n    return noiseShape;\n  }\n  if (x.shape.length === noiseShape.length) {\n    const newDimension: number[] = [];\n    for (let i = 0; i < x.shape.length; i++) {\n      if (noiseShape[i] == null && x.shape[i] != null) {\n        newDimension.push(x.shape[i]);\n      } else {\n        newDimension.push(noiseShape[i]);\n      }\n    }\n    return newDimension;\n  }\n\n  return noiseShape;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {add} from './add';\nimport {div} from './div';\nimport {getNoiseShape} from './dropout_util';\nimport {floor} from './floor';\nimport {mul} from './mul';\nimport {op} from './operation';\nimport {randomUniform} from './random_uniform';\n\n/**\n * Computes dropout.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 2, 1]);\n * const rate = 0.75;\n * const output = tf.dropout(x, rate);\n * output.print();\n * ```\n *\n * @param x A floating point Tensor or TensorLike.\n * @param rate A float in the range [0, 1). The probability that each element\n *   of x is discarded.\n * @param noiseShape An array of numbers of type int32, representing the\n * shape for randomly generated keep/drop flags. If the noiseShape has null\n * value, it will be automatically replaced with the x's relative dimension\n * size. Optional.\n * @param seed Used to create random seeds. Optional.\n * @returns A Tensor of the same shape of x.\n *\n * @doc {heading: 'Operations', subheading: 'Dropout'}\n */\nfunction dropout_(\n    x: Tensor|TensorLike, rate: number, noiseShape?: number[],\n    seed?: number|string): Tensor {\n  const $x = convertToTensor(x, 'x', 'dropout');\n\n  util.assert(\n      $x.dtype === 'float32',\n      () => `x has to be a floating point tensor since it's going to be ` +\n          `scaled, but got a ${$x.dtype} tensor instead.`);\n  util.assert(\n      rate >= 0 && rate < 1,\n      () => `rate must be a float in the range [0, 1), but got ${rate}.`);\n\n  if (rate === 0) {\n    return x instanceof Tensor ? $x.clone() : $x;\n  }\n\n  const $noiseShape = getNoiseShape($x, noiseShape);\n  const keepProb = 1 - rate;\n  const multiplier = div(\n      floor(add(randomUniform($noiseShape, 0, 1, 'float32', seed), keepProb)),\n      keepProb);\n\n  return mul($x, multiplier);\n}\n\nexport const dropout = op({dropout_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../tensor';\nimport {tensor1d} from './tensor1d';\n\nexport function enclosingPowerOfTwo(value: number) {\n  // Return 2**N for integer N such that 2**N >= value.\n  return Math.floor(Math.pow(2, Math.ceil(Math.log(value) / Math.log(2.0))));\n}\n\nexport function cosineWindow(\n    windowLength: number, a: number, b: number): Tensor1D {\n  const even = 1 - windowLength % 2;\n  const newValues = new Float32Array(windowLength);\n  for (let i = 0; i < windowLength; ++i) {\n    const cosArg = (2.0 * Math.PI * i) / (windowLength + even - 1);\n    newValues[i] = a - b * Math.cos(cosArg);\n  }\n  return tensor1d(newValues, 'float32');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {assert, assertShapesMatch, getTypedArrayFromDType} from '../util';\nimport {tensor} from './tensor';\n\n/**\n * Returns whether the targets are in the top K predictions.\n *\n * ```js\n * const predictions = tf.tensor2d([[20, 10, 40, 30], [30, 50, -20, 10]]);\n * const targets = tf.tensor1d([2, 0]);\n * const precision = await tf.inTopKAsync(predictions, targets);\n * precision.print();\n * ```\n * @param predictions 2-D or higher `tf.Tensor` with last dimension being\n *     at least `k`.\n * @param targets 1-D or higher `tf.Tensor`.\n * @param k Optional Number of top elements to look at for computing precision,\n *     default to 1.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nasync function inTopKAsync_<T extends Tensor, U extends Tensor>(\n    predictions: T|TensorLike, targets: U|TensorLike, k = 1): Promise<U> {\n  const $predictions = convertToTensor(predictions, 'predictions', 'inTopK');\n  const $targets = convertToTensor(targets, 'targets', 'inTopK');\n\n  assert(\n      $predictions.rank > 1,\n      () => 'inTopK() expects the predictions to be of rank 2 or higher, ' +\n          `but got ${$predictions.rank}`);\n  assert(\n      $predictions.rank - 1 === $targets.rank,\n      () => `predictions rank should be 1 larger than ` +\n          `targets rank, but got predictions rank ` +\n          `${$predictions.rank} and targets rank ${$targets.rank}`);\n  assertShapesMatch(\n      $predictions.shape.slice(0, $predictions.shape.length - 1),\n      $targets.shape,\n      `predictions's shape should be align with the targets' shape, ` +\n          'except the last dimension.');\n  const lastDim = $predictions.shape[$predictions.shape.length - 1];\n  assert(\n      k > 0 && k <= lastDim,\n      () => `'k' passed to inTopK() must be > 0 && <= the predictions last ` +\n          `dimension (${lastDim}), but got ${k}`);\n\n  const predictionsVals = await $predictions.data();\n  const targetsVals = await $targets.data();\n\n  // Reshape predictionsVals into a 2d tensor [batch, lastDim]\n  // and look up topK along lastDim.\n  const [batch, size] = [predictionsVals.length / lastDim, lastDim];\n  const precision = getTypedArrayFromDType('bool', batch);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = predictionsVals.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    precision[b] = 0;\n    for (let i = 0; i < k; i++) {\n      if (valAndInd[i].index === targetsVals[b]) {\n        precision[b] = 1;\n        break;\n      }\n    }\n  }\n\n  if (predictions !== $predictions) {\n    $predictions.dispose();\n  }\n  if (targets !== $targets) {\n    $targets.dispose();\n  }\n\n  // Output precision has the same shape as targets.\n  return tensor(precision, $targets.shape, 'bool') as U;\n}\n\nexport const inTopKAsync = inTopKAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv2DBackpropFilter, Conv2DBackpropFilterAttrs, Conv2DBackpropFilterInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 2D convolution.\n *\n * @param x The input tensor, of rank 4 or rank 3 of shape\n *     [batch, height, width, inChannels]. If rank 3, batch of 1 is assumed.\n * @param dy The dy image, of rank 4 or rank 3, of shape\n *     [batch, height, width, outDepth]. If rank 3, batch of 1 is assumed.\n * @param filterShape The shape of the filter, length 4,\n *     [filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels].\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction conv2DBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    strides: [number, number]|number,\n    pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n    dataFormat: 'NHWC'|'NCHW' = 'NHWC',\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in conv2dDerFilter: input must be rank 4, but got shape ` +\n          `${x4D.shape}.`);\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in conv2dDerFilter: dy must be rank 4, but got shape ` +\n          `${dy4D.shape}.`);\n  util.assert(\n      filterShape.length === 4,\n      () => `Error in conv2dDerFilter: filterShape must be length 4, but got ` +\n          `${filterShape}.`);\n  const inDepth = dataFormat === 'NHWC' ? x4D.shape[3] : x4D.shape[1];\n  const outDepth = dataFormat === 'NHWC' ? dy4D.shape[3] : dy4D.shape[1];\n  util.assert(\n      inDepth === filterShape[2],\n      () => `Error in conv2dDerFilter: depth of input ${inDepth}) must ` +\n          `match input depth in filter (${filterShape[2]}.`);\n  util.assert(\n      outDepth === filterShape[3],\n      () => `Error in conv2dDerFilter: depth of dy (${outDepth}) must ` +\n          `match output depth for filter (${filterShape[3]}).`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in conv2dDerFilter: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n    const $dataFormat = conv_util.convertConv2DDataFormat(dataFormat);\n    const convInfo = conv_util.computeConv2DInfo(\n        x4D.shape, filterShape, strides, dilations, pad, dimRoundingMode, false,\n        $dataFormat);\n\n    return backend.conv2dDerFilter(x4D, dy4D, convInfo);\n  };\n\n  const inputs: Conv2DBackpropFilterInputs = {x: x4D, dy: dy4D};\n  const attrs:\n      Conv2DBackpropFilterAttrs = {strides, pad, dataFormat, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv2DBackpropFilter, attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const conv2DBackpropFilter = op({conv2DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\nimport * as broadcast_util from './broadcast_util';\nimport {elu} from './elu';\nimport {Activation} from './fused_types';\nimport {mul} from './mul';\nimport {prelu} from './prelu';\nimport {relu} from './relu';\nimport {relu6} from './relu6';\nimport {reshape} from './reshape';\nimport {step} from './step';\nimport {sum} from './sum';\n\n// Returns gradient for fused activation.\nexport function getFusedDyActivation(\n    dy: Tensor, y: Tensor, activation: Activation): Tensor {\n  if (activation == null || activation === 'linear') {\n    return dy;\n  }\n  if (activation === 'relu') {\n    return mul(dy, step(y));\n  }\n  throw new Error(\n      `Cannot compute gradient for fused activation ${activation}.`);\n}\n\n// Returns gradient for fused bias.\nexport function getFusedBiasGradient(\n    bias: Tensor, dyActivation: Tensor): Tensor {\n  let res = dyActivation;\n  const reduceAxes =\n      broadcast_util.getReductionAxes(bias.shape, dyActivation.shape);\n  if (reduceAxes.length > 0) {\n    res = sum(res, reduceAxes);\n  }\n  return reshape(res, bias.shape);\n}\n\nexport function applyActivation(\n    x: Tensor, activation: Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return x;\n  } else if (activation === 'relu') {\n    return relu(x);\n  } else if (activation === 'elu') {\n    return elu(x);\n  } else if (activation === 'relu6') {\n    return relu6(x);\n  } else if (activation === 'prelu') {\n    return prelu(x, preluActivationWeights);\n  }\n  throw new Error(`Unknown fused activation ${activation}.`);\n}\n\n// Whether we should call fused ops.\nexport const shouldFuse = (gradientDepth: number, activation: Activation) => {\n  const gradientMode = gradientDepth > 0;\n  return !gradientMode || activation === 'linear';\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedConv2D, FusedConv2DAttrs, FusedConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {conv2d as unfusedConv2d} from '../conv2d';\nimport {conv2DBackpropFilter} from '../conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../conv2d_backprop_input';\nimport * as conv_util from '../conv_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes a 2D convolution over the input x, optionally fused with adding a\n * bias and applying an activation.\n *\n * ```js\n * const inputDepth = 2;\n * const inShape = [2, 2, 2, inputDepth];\n * const outputDepth = 2;\n * const fSize = 1;\n * const pad = 0;\n * const strides = 1;\n *\n * const x = tf.tensor4d( [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,\n * 16], inShape);\n * const w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth,\n * outputDepth]);\n *\n * tf.fused.conv2d({ x, filter: w, strides, pad, dataFormat: 'NHWC',\n * dilations: [1, 1], bias: tf.scalar(5), activation: 'relu' }).print();\n * ```\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter, rank 4, of shape\n *     `[filterHeight, filterWidth, inDepth, outDepth]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid` output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dataFormat An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`) to be\n *     applied\n *      after biasAdd.\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number|conv_util.ExplicitPadding,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  activation = activation || 'linear';\n\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'conv2d');\n  const $filter = convertToTensor(filter, 'filter', 'conv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused conv2d: input must be rank 4, but got rank ` +\n          `${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused conv2d: filter must be rank 4, but got rank ` +\n          `${$filter.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused conv2d: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in conv2d: depth of input (${x4D.shape[3]}) must match ` +\n          `input depth for filter ${$filter.shape[2]}.`);\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in conv2D: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n  util.assert(\n      dataFormat === 'NHWC',\n      () => `Error in conv2d: got dataFormat of ${\n          dataFormat} but only NHWC is currently supported.`);\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused conv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    const [$filter, x4D, y, $bias] =\n        saved as [Tensor4D, Tensor4D, Tensor4D, Tensor];\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused conv2D: ' +\n            `dilation rates greater than 1 ` +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const xDer =\n        conv2DBackpropInput(x4D.shape, dyActivation, $filter, strides, pad);\n    const filterDer =\n        conv2DBackpropFilter(x4D, dyActivation, $filter.shape, strides, pad);\n    const der: Tensor[] = [xDer, filterDer];\n\n    if ($bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      der.push(biasDer);\n    }\n    return der;\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedConv2d({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n\n  const attrs: FusedConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const conv2d = op({fusedConv2d_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropFilterInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropFilter_<T extends Tensor3D|Tensor4D>(\n    x: T, dy: T, filterShape: [number, number, number, number],\n    convInfo: conv_util.Conv2DInfo): Tensor4D {\n  let x4D = x as Tensor4D;\n  if (x.rank === 3) {\n    x4D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2]]);\n  }\n  let dy4D = dy as Tensor4D;\n  if (dy4D.rank === 3) {\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerFilter(x4D, dy4D, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropFilterInputs = {x: x4D, dy: dy4D};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             DepthwiseConv2dNativeBackpropFilter) as Tensor4D;\n}\n\nexport const depthwiseConv2dNativeBackpropFilter =\n    op({depthwiseConv2dNativeBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {DepthwiseConv2dNativeBackpropInput, DepthwiseConv2dNativeBackpropInputInputs} from '../kernel_names';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\nfunction depthwiseConv2dNativeBackpropInput_<T extends Tensor3D|Tensor4D>(\n    xShape: [number, number, number, number]|[number, number, number], dy: T,\n    filter: Tensor4D, convInfo: conv_util.Conv2DInfo): T {\n  let dy4D = dy as Tensor4D;\n  let reshapedTo4D = false;\n  if (dy.rank === 3) {\n    reshapedTo4D = true;\n    dy4D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2]]);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.depthwiseConv2DDerInput(dy4D, filter, convInfo);\n\n  const inputs: DepthwiseConv2dNativeBackpropInputInputs = {dy: dy4D};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null,\n      DepthwiseConv2dNativeBackpropInput);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const depthwiseConv2dNativeBackpropInput =\n    op({depthwiseConv2dNativeBackpropInput_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {FusedDepthwiseConv2D, FusedDepthwiseConv2DAttrs, FusedDepthwiseConv2DInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport * as conv_util from '../conv_util';\nimport {depthwiseConv2d as unfusedDepthwiseConv2d} from '../depthwise_conv2d';\nimport {depthwiseConv2dNativeBackpropFilter} from '../depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../depthwise_conv2d_native_backprop_input';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes depthwise 2D convolution, optionally fused with adding a\n * bias and applying an activation.\n *\n * Given a 4D `input` array and a `filter` array of shape\n * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing\n * `inChannels` convolutional filters of depth 1, this op applies a\n * different filter to each input channel (expanding from 1 channel to\n * `channelMultiplier` channels for each), then concatenates the results\n * together. The output has `inChannels * channelMultiplier` channels.\n *\n * See\n * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](\n *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)\n * for more details.\n *\n * @param obj An object with the following properties:\n * @param x The input tensor, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is\n * assumed.\n * @param filter The filter tensor, rank 4, of shape\n *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.\n * @param strides The strides of the convolution: `[strideHeight,\n * strideWidth]`. If strides is a single number, then `strideHeight ==\n * strideWidth`.\n * @param pad The type of padding algorithm.\n *   - `same` and stride 1: output will be of same size as input,\n *       regardless of filter size.\n *   - `valid`: output will be smaller than input if filter is larger\n *       than 1x1.\n *   - For more info, see this guide:\n *     [https://www.tensorflow.org/api_guides/python/nn#Convolution](\n *          https://www.tensorflow.org/api_guides/python/nn#Convolution)\n * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`\n *     in which we sample input values across the height and width dimensions\n *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single\n *     number, then `dilationHeight == dilationWidth`. If it is greater than\n *     1, then all values of `strides` must be 1.\n * @param dataFormat: An optional string from: \"NHWC\", \"NCHW\". Defaults to\n *     \"NHWC\". Specify the data format of the input and output data. With the\n *     default format \"NHWC\", the data is stored in the order of: [batch,\n *     height, width, channels]. Only \"NHWC\" is currently supported.\n * @param dimRoundingMode The rounding mode used when computing output\n *     dimensions if pad is a number. If none is provided, it will not round\n *     and error if the output is of fractional size.\n * @param bias Tensor to be added to the result.\n * @param activation Name of activation kernel (defaults to `linear`).\n * @param preluActivationWeights Tensor of prelu weights to be applied as part\n *     of a `prelu` activation, typically the same shape as `x`.\n */\nfunction fusedDepthwiseConv2d_<T extends Tensor3D|Tensor4D>({\n  x,\n  filter,\n  strides,\n  pad,\n  dataFormat = 'NHWC',\n  dilations = [1, 1],\n  dimRoundingMode,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  x: T|TensorLike,\n  filter: Tensor4D|TensorLike,\n  strides: [number, number]|number,\n  pad: 'valid'|'same'|number,\n  dataFormat?: 'NHWC'|'NCHW',\n  dilations?: [number, number]|number,\n  dimRoundingMode?: 'floor'|'round'|'ceil',\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedDepthwiseConv2d(\n        x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  const $x = convertToTensor(x, 'x', 'depthwiseConv2d');\n  const $filter = convertToTensor(filter, 'filter', 'depthwiseConv2d');\n\n  let x4D = $x as Tensor4D;\n  let reshapedTo4D = false;\n  if ($x.rank === 3) {\n    reshapedTo4D = true;\n    x4D = reshape($x, [1, $x.shape[0], $x.shape[1], $x.shape[2]]);\n  }\n  util.assert(\n      x4D.rank === 4,\n      () => `Error in fused depthwiseConv2d: input must be rank 4, but got ` +\n          `rank ${x4D.rank}.`);\n  util.assert(\n      $filter.rank === 4,\n      () => `Error in fused depthwiseConv2d: filter must be rank 4, ` +\n          `but got rank ${$filter.rank}.`);\n  util.assert(\n      x4D.shape[3] === $filter.shape[2],\n      () => `Error in fused depthwiseConv2d: number of input channels ` +\n          `(${x4D.shape[3]}) must match the inChannels dimension in ` +\n          `filter ${$filter.shape[2]}.`);\n  if (dilations == null) {\n    dilations = [1, 1];\n  }\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () =>\n          'Error in fused depthwiseConv2d: Either strides or dilations must ' +\n          `be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in fused depthwiseConv2d: pad must be an integer when ` +\n            `using dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const convInfo = conv_util.computeConv2DInfo(\n      x4D.shape, $filter.shape, strides, dilations, pad, dimRoundingMode,\n      true /* depthwise */);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused conv2d');\n    [$bias] = makeTypesMatch($bias, $x);\n\n    broadcast_util.assertAndGetBroadcastShape(convInfo.outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused depthwiseConv2d');\n  }\n\n  const grad = (dy: Tensor4D, saved: Tensor[]) => {\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of fused depthwiseConv2d: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${dilations}'`);\n    const [$filter, x4D, y, bias] = saved;\n\n    const dyActivation = getFusedDyActivation(dy, y, activation) as Tensor4D;\n\n    const xDer = depthwiseConv2dNativeBackpropInput(\n        (x4D as Tensor4D).shape, dyActivation, $filter as Tensor4D, convInfo);\n    const filterDer = depthwiseConv2dNativeBackpropFilter(\n        x4D as Tensor4D, dyActivation, ($filter as Tensor4D).shape, convInfo);\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [xDer, filterDer, biasDer];\n    }\n    return [xDer, filterDer];\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const res = backend.fusedDepthwiseConv2D({\n      input: x4D,\n      filter: $filter,\n      convInfo,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return res;\n  };\n\n  const inputs: FusedDepthwiseConv2DInputs = {\n    x: x4D,\n    filter: $filter,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: FusedDepthwiseConv2DAttrs =\n      {strides, pad, dataFormat, dilations, dimRoundingMode, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((x4D: Tensor4D, filter: Tensor4D, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n    return customOp(x4D, $filter) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (x4D: Tensor4D, filter: Tensor4D, bias: Tensor, save: GradSaveFunc) => {\n          let res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              FusedDepthwiseConv2D, attrs as {} as NamedAttrMap);\n\n          save([filter, x4D, res, bias]);\n\n          if (reshapedTo4D) {\n            res = reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n          }\n\n          return {value: res, gradFunc: grad};\n        });\n\n    return customOpWithBias(x4D, $filter, $bias) as T;\n  }\n}\nexport const depthwiseConv2d = op({fusedDepthwiseConv2d_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {customGrad} from '../../gradients';\nimport {_FusedMatMul, _FusedMatMulAttrs, _FusedMatMulInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor3D} from '../../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../../tensor_types';\nimport {makeTypesMatch} from '../../tensor_util';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {add} from '../add';\nimport * as broadcast_util from '../broadcast_util';\nimport {Activation} from '../fused_types';\nimport {applyActivation, getFusedBiasGradient, getFusedDyActivation, shouldFuse} from '../fused_util';\nimport {matMul as unfusedMatMul} from '../mat_mul';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Computes the dot product of two matrices with optional activation and bias.\n *\n * ```js\n * const a = tf.tensor2d([-1, -2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const bias = tf.tensor2d([1, 2], [1, 2]);\n *\n * tf.fused.matMul({a, b, bias, activation: 'relu'}).print();\n * ```\n *\n * @param obj An object with the following properties:\n * - `a` First matrix in dot product operation.\n * - `b` Second matrix in dot product operation.\n * - `transposeA` If true, `a` is transposed before multiplication.\n * - `transposeB` If true, `b` is transposed before multiplication.\n * - `bias` Matrix to be added to the result.\n * - `activation` Name of activation kernel (defaults to `linear`).\n * - `preluActivationWeights` Tensor of prelu weights.\n */\nfunction fusedMatMul_<T extends Tensor>({\n  a,\n  b,\n  transposeA = false,\n  transposeB = false,\n  bias,\n  activation = 'linear',\n  preluActivationWeights\n}: {\n  a: T|TensorLike,\n  b: T|TensorLike,\n  transposeA?: boolean,\n  transposeB?: boolean,\n  bias?: Tensor|TensorLike,\n  activation?: Activation,\n  preluActivationWeights?: Tensor\n}): T {\n  if (shouldFuse(ENGINE.state.gradientDepth, activation) === false) {\n    let result = unfusedMatMul(a, b, transposeA, transposeB);\n    if (bias != null) {\n      result = add(result, bias);\n    }\n\n    return applyActivation(result, activation, preluActivationWeights) as T;\n  }\n\n  let $a = convertToTensor(a, 'a', 'fused matMul');\n  let $b = convertToTensor(b, 'b', 'fused matMul');\n  [$a, $b] = makeTypesMatch($a, $b);\n\n  const innerShapeA =\n      transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n  const innerShapeB =\n      transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n  const outerShapeA =\n      transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n  const outerShapeB =\n      transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n  const outerDimsA = $a.shape.slice(0, -2);\n  const outerDimsB = $b.shape.slice(0, -2);\n  const batchDimA = util.sizeFromShape(outerDimsA);\n  const batchDimB = util.sizeFromShape(outerDimsB);\n\n  util.assert(\n      $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n      () =>\n          `Error in fused matMul: inputs must have the same rank of at least ` +\n          `2, got ranks ${$a.rank} and ${$b.rank}.`);\n\n  util.assert(\n      util.arraysEqual(outerDimsA, outerDimsB),\n      () => `Error in fused matMul: outer dimensions (${outerDimsA}) and (` +\n          `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} must match.`);\n\n  util.assert(\n      innerShapeA === innerShapeB,\n      () => `Error in fused matMul: inner shapes (${innerShapeA}) and (` +\n          `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n          `${$b.shape} and transposeA=${transposeA}` +\n          ` and transposeB=${transposeB} must match.`);\n\n  const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n  const a3D: Tensor3D = transposeA ?\n      reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n      reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n  const b3D: Tensor3D = transposeB ?\n      reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n      reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n  let $bias: Tensor;\n  if (bias != null) {\n    $bias = convertToTensor(bias, 'bias', 'fused matMul');\n    [$bias] = makeTypesMatch($bias, $a);\n\n    broadcast_util.assertAndGetBroadcastShape(outShape, $bias.shape);\n  }\n\n  let $preluActivationWeights: Tensor;\n  if (preluActivationWeights != null) {\n    $preluActivationWeights = convertToTensor(\n        preluActivationWeights, 'prelu weights', 'fused matMul');\n  }\n\n  const grad = (dy: Tensor3D, saved: Tensor[]) => {\n    const [a3D, b3D, y, $bias] = saved;\n    // we reshape dy because the result of the forward is not\n    // necessarily going to be a 3d tensor due to a reshape done at the end of\n    // the customOp.\n    const dyActivation =\n        getFusedDyActivation(reshape(dy, y.shape), y, activation);\n    let aDer: Tensor;\n    let bDer: Tensor;\n\n    if (!transposeA && !transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, true, false);\n    } else if (!transposeA && transposeB) {\n      aDer = unfusedMatMul(dyActivation, b3D, false, false);\n      bDer = unfusedMatMul(dyActivation, a3D, true, false);\n    } else if (transposeA && !transposeB) {\n      aDer = unfusedMatMul(b3D, dyActivation, false, true);\n      bDer = unfusedMatMul(a3D, dyActivation, false, false);\n    } else {\n      aDer = unfusedMatMul(b3D, dyActivation, true, true);\n      bDer = unfusedMatMul(dyActivation, a3D, true, true);\n    }\n\n    if (bias != null) {\n      const biasDer = getFusedBiasGradient($bias, dyActivation);\n      return [aDer, bDer, biasDer];\n    } else {\n      return [aDer, bDer];\n    }\n  };\n\n  const forward: ForwardFunc<Tensor> = (backend) => {\n    const y = backend.fusedBatchMatMul({\n      a: a3D,\n      b: b3D,\n      transposeA,\n      transposeB,\n      bias: $bias,\n      activation,\n      preluActivationWeights: $preluActivationWeights\n    });\n    return y;\n  };\n\n  const inputs: _FusedMatMulInputs = {\n    a: a3D,\n    b: b3D,\n    bias: $bias,\n    preluActivationWeights: $preluActivationWeights\n  };\n  const attrs: _FusedMatMulAttrs = {transposeA, transposeB, activation};\n\n  // Depending on the the params passed in we will have different number of\n  // inputs and thus a a different number of elements in the gradient.\n  if (bias == null) {\n    const customOp =\n        customGrad((a3D: Tensor3D, b3D: Tensor3D, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n    return customOp(a3D, b3D) as T;\n  } else {\n    const customOpWithBias = customGrad(\n        (a3D: Tensor3D, b3D: Tensor3D, $bias: Tensor, save: GradSaveFunc) => {\n          const res = ENGINE.runKernelFunc(\n              forward, inputs as {} as NamedTensorMap, null /* grad */,\n              _FusedMatMul, attrs as {} as NamedAttrMap);\n\n          save([a3D, b3D, res, $bias]);\n\n          return {value: reshape(res, outShape), gradFunc: grad};\n        });\n\n    return customOpWithBias(a3D, b3D, $bias) as T;\n  }\n}\n\nexport const matMul = op({fusedMatMul_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {conv2d} from './fused/conv2d';\nimport {depthwiseConv2d} from './fused/depthwise_conv2d';\nimport {matMul} from './fused/mat_mul';\nimport {Activation} from './fused_types';\n\nexport {Activation, conv2d, depthwiseConv2d, matMul};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a hamming window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hammingWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hammingWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.54, 0.46);\n}\nexport const hammingWindow = op({hammingWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D} from '../../tensor';\nimport {op} from '../operation';\nimport {cosineWindow} from '../signal_ops_util';\n\n/**\n * Generate a Hann window.\n *\n * See: https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows\n *\n * ```js\n * tf.signal.hannWindow(10).print();\n * ```\n * @param The length of window\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction hannWindow_(windowLength: number): Tensor1D {\n  return cosineWindow(windowLength, 0.5, 0.5);\n}\n\nexport const hannWindow = op({hannWindow_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {fill} from '../fill';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {tensor2d} from '../tensor2d';\n\n/**\n * Expands input into frames of frameLength.\n * Slides a window size with frameStep.\n *\n * ```js\n * tf.signal.frame([1, 2, 3], 2, 1).print();\n * ```\n * @param signal The input tensor to be expanded\n * @param frameLength Length of each frame\n * @param frameStep The frame hop size in samples.\n * @param padEnd Whether to pad the end of signal with padValue.\n * @param padValue An number to use where the input signal does\n *     not exist when padEnd is True.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction frame_(\n    signal: Tensor1D, frameLength: number, frameStep: number, padEnd = false,\n    padValue = 0): Tensor {\n  let start = 0;\n  const output: Tensor[] = [];\n  while (start + frameLength <= signal.size) {\n    output.push(slice(signal, start, frameLength));\n    start += frameStep;\n  }\n\n  if (padEnd) {\n    while (start < signal.size) {\n      const padLen = (start + frameLength) - signal.size;\n      const pad = concat([\n        slice(signal, start, frameLength - padLen), fill([padLen], padValue)\n      ]);\n      output.push(pad);\n      start += frameStep;\n    }\n  }\n\n  if (output.length === 0) {\n    return tensor2d([], [0, frameLength]);\n  }\n\n  return reshape(concat(output), [output.length, frameLength]);\n}\nexport const frame = op({frame_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {concat} from '../concat';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {enclosingPowerOfTwo} from '../signal_ops_util';\nimport {slice} from '../slice';\nimport {rfft} from '../spectral/rfft';\n\nimport {frame} from './frame';\nimport {hannWindow} from './hann_window';\n\n/**\n * Computes the Short-time Fourier Transform of signals\n * See: https://en.wikipedia.org/wiki/Short-time_Fourier_transform\n *\n * ```js\n * const input = tf.tensor1d([1, 1, 1, 1, 1])\n * tf.signal.stft(input, 3, 1).print();\n * ```\n * @param signal 1-dimensional real value tensor.\n * @param frameLength The window length of samples.\n * @param frameStep The number of samples to step.\n * @param fftLength The size of the FFT to apply.\n * @param windowFn A callable that takes a window length and returns 1-d tensor.\n *\n * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}\n */\nfunction stft_(\n    signal: Tensor1D, frameLength: number, frameStep: number,\n    fftLength?: number,\n    windowFn: (length: number) => Tensor1D = hannWindow): Tensor {\n  if (fftLength == null) {\n    fftLength = enclosingPowerOfTwo(frameLength);\n  }\n  const framedSignal = frame(signal, frameLength, frameStep);\n  const windowedSignal = mul(framedSignal, windowFn(frameLength));\n  const output: Tensor[] = [];\n  for (let i = 0; i < framedSignal.shape[0]; i++) {\n    output.push(\n        rfft(slice(windowedSignal, [i, 0], [1, frameLength]), fftLength));\n  }\n  return concat(output);\n}\nexport const stft = op({stft_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {CropAndResize, CropAndResizeAttrs, CropAndResizeInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor1D, Tensor2D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Extracts crops from the input image tensor and resizes them using bilinear\n * sampling or nearest neighbor sampling (possibly with aspect ratio change)\n * to a common output size specified by crop_size.\n *\n * @param image 4d tensor of shape `[batch,imageHeight,imageWidth, depth]`,\n *     where imageHeight and imageWidth must be positive, specifying the\n *     batch of images from which to take crops\n * @param boxes 2d float32 tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the normalized\n *     coordinates of the box in the boxInd[i]'th image in the batch\n * @param boxInd 1d int32 tensor of shape `[numBoxes]` with values in range\n *     `[0, batch)` that specifies the image that the `i`-th box refers to.\n * @param cropSize 1d int32 tensor of 2 elements `[cropHeigh, cropWidth]`\n *     specifying the size to which all crops are resized to.\n * @param method Optional string from `'bilinear' | 'nearest'`,\n *     defaults to bilinear, which specifies the sampling method for resizing\n * @param extrapolationValue A threshold for deciding when to remove boxes based\n *     on score. Defaults to 0.\n * @return A 4D tensor of the shape `[numBoxes,cropHeight,cropWidth,depth]`\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction cropAndResize_(\n    image: Tensor4D|TensorLike,\n    boxes: Tensor2D|TensorLike,\n    boxInd: Tensor1D|TensorLike,\n    cropSize: [number, number],\n    method?: 'bilinear'|'nearest',\n    extrapolationValue?: number,\n    ): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'cropAndResize');\n  const $boxes = convertToTensor(boxes, 'boxes', 'cropAndResize', 'float32');\n  const $boxInd = convertToTensor(boxInd, 'boxInd', 'cropAndResize', 'int32');\n  method = method || 'bilinear';\n  extrapolationValue = extrapolationValue || 0;\n\n  const numBoxes = $boxes.shape[0];\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in cropAndResize: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n  util.assert(\n      $boxes.rank === 2 && $boxes.shape[1] === 4,\n      () => `Error in cropAndResize: boxes must be have size [${numBoxes},4] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      $boxInd.rank === 1 && $boxInd.shape[0] === numBoxes,\n      () => `Error in cropAndResize: boxInd must be have size [${numBoxes}] ` +\n          `but had shape ${$boxes.shape}.`);\n  util.assert(\n      cropSize.length === 2,\n      () => `Error in cropAndResize: cropSize must be of length 2, but got ` +\n          `length ${cropSize.length}.`);\n  util.assert(\n      cropSize[0] >= 1 && cropSize[1] >= 1,\n      () => `cropSize must be atleast [1,1], but was ${cropSize}`);\n  util.assert(\n      method === 'bilinear' || method === 'nearest',\n      () => `method must be bilinear or nearest, but was ${method}`);\n\n  const forward: ForwardFunc<Tensor4D> = (backend) => backend.cropAndResize(\n      $image, $boxes, $boxInd, cropSize, method, extrapolationValue);\n\n  const inputs:\n      CropAndResizeInputs = {image: $image, boxes: $boxes, boxInd: $boxInd};\n  const attrs: CropAndResizeAttrs = {method, extrapolationValue, cropSize};\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */, CropAndResize,\n      attrs as {} as NamedAttrMap);\n  return res;\n}\n\nexport const cropAndResize = op({cropAndResize_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {FlipLeftRight, FlipLeftRightInputs} from '../../kernel_names';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\nimport {op} from '../operation';\n\n/**\n * Flips the image left to right. Currently available in the CPU, WebGL, and\n * WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n */\n/** @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'} */\nfunction flipLeftRight_(image: Tensor4D|TensorLike): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'flipLeftRight', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in flipLeftRight: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: FlipLeftRightInputs = {image: $image};\n  const res =\n      ENGINE.runKernel(FlipLeftRight, inputs as {} as NamedTensorMap, {});\n  return res as Tensor4D;\n}\n\nexport const flipLeftRight = op({flipLeftRight_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\n\n/**\n * Rotates the input image tensor counter-clockwise with an optional offset\n * center of rotation. Currently available in the CPU, WebGL, and WASM backends.\n *\n * @param image 4d tensor of shape `[batch, imageHeight, imageWidth, depth]`.\n * @param radians The amount of rotation.\n * @param fillValue The value to fill in the empty space leftover\n *     after rotation. Can be either a single grayscale value (0-255), or an\n *     array of three numbers `[red, green, blue]` specifying the red, green,\n *     and blue channels. Defaults to `0` (black).\n * @param center The center of rotation. Can be either a single value (0-1), or\n *     an array of two numbers `[centerX, centerY]`. Defaults to `0.5` (rotates\n *     the image around its center).\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction rotateWithOffset_(\n    image: Tensor4D|TensorLike, radians: number,\n    fillValue: number|[number, number, number] = 0,\n    center: number|[number, number] = 0.5): Tensor4D {\n  const $image = convertToTensor(image, 'image', 'rotateWithOffset', 'float32');\n\n  util.assert(\n      $image.rank === 4,\n      () => 'Error in rotateWithOffset: image must be rank 4,' +\n          `but got rank ${$image.rank}.`);\n\n  const inputs: RotateWithOffsetInputs = {image: $image};\n  const attrs: RotateWithOffsetAttrs = {radians, fillValue, center};\n  const res = ENGINE.runKernel(\n      RotateWithOffset, inputs as {} as NamedTensorMap,\n      attrs as {} as NamedAttrMap);\n  return res as Tensor4D;\n}\n\nexport const rotateWithOffset = op({rotateWithOffset_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport * as util from '../util';\n\nfunction nonMaxSuppSanityCheck(\n    boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma?: number): {\n  maxOutputSize: number,\n  iouThreshold: number,\n  scoreThreshold: number,\n  softNmsSigma: number\n} {\n  if (iouThreshold == null) {\n    iouThreshold = 0.5;\n  }\n  if (scoreThreshold == null) {\n    scoreThreshold = Number.NEGATIVE_INFINITY;\n  }\n  if (softNmsSigma == null) {\n    softNmsSigma = 0.0;\n  }\n\n  const numBoxes = boxes.shape[0];\n  maxOutputSize = Math.min(maxOutputSize, numBoxes);\n\n  util.assert(\n      0 <= iouThreshold && iouThreshold <= 1,\n      () => `iouThreshold must be in [0, 1], but was '${iouThreshold}'`);\n  util.assert(\n      boxes.rank === 2,\n      () => `boxes must be a 2D tensor, but was of rank '${boxes.rank}'`);\n  util.assert(\n      boxes.shape[1] === 4,\n      () =>\n          `boxes must have 4 columns, but 2nd dimension was ${boxes.shape[1]}`);\n  util.assert(scores.rank === 1, () => 'scores must be a 1D tensor');\n  util.assert(\n      scores.shape[0] === numBoxes,\n      () => `scores has incompatible shape with boxes. Expected ${numBoxes}, ` +\n          `but was ${scores.shape[0]}`);\n  util.assert(\n      0 <= softNmsSigma && softNmsSigma <= 1,\n      () => `softNmsSigma must be in [0, 1], but was '${softNmsSigma}'`);\n  return {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n}\n\nexport {nonMaxSuppSanityCheck};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV3} from '../../kernel_names';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\nfunction nonMaxSuppression_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Tensor1D {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const attrs = {maxOutputSize, iouThreshold, scoreThreshold};\n  return ENGINE.runKernelFunc(\n      b => b.nonMaxSuppression(\n          $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold),\n      {boxes: $boxes, scores: $scores}, null /* grad */, NonMaxSuppressionV3,\n      attrs);\n}\n\nexport const nonMaxSuppression = op({nonMaxSuppression_});\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Inserts a value into a sorted array. This method allows duplicate, meaning it\n * allows inserting duplicate value, in which case, the element will be inserted\n * at the lowest index of the value.\n * @param arr The array to modify.\n * @param element The element to insert.\n * @param comparator Optional. If no comparator is specified, elements are\n * compared using array_util.defaultComparator, which is suitable for Strings\n * and Numbers in ascending arrays. If the array contains multiple instances of\n * the target value, the left-most instance will be returned. To provide a\n * comparator, it should take 2 arguments to compare and return a negative,\n * zero, or a positive number.\n */\nexport function binaryInsert<T>(\n    arr: T[], element: T, comparator?: (a: T, b: T) => number) {\n  const index = binarySearch(arr, element, comparator);\n  const insertionPoint = index < 0 ? -(index + 1) : index;\n  arr.splice(insertionPoint, 0, element);\n}\n\n/**\n * Searches the array for the target using binary search, returns the index\n * of the found element, or position to insert if element not found. If no\n * comparator is specified, elements are compared using array_\n * util.defaultComparator, which is suitable for Strings and Numbers in\n * ascending arrays. If the array contains multiple instances of the target\n * value, the left-most instance will be returned.\n * @param arr The array to be searched in.\n * @param target The target to be searched for.\n * @param comparator Should take 2 arguments to compare and return a negative,\n *    zero, or a positive number.\n * @return Lowest index of the target value if found, otherwise the insertion\n *    point where the target should be inserted, in the form of\n *    (-insertionPoint - 1).\n */\nexport function binarySearch<T>(\n    arr: T[], target: T, comparator?: (a: T, b: T) => number) {\n  return binarySearch_(arr, target, comparator || defaultComparator);\n}\n\n/**\n * Compares its two arguments for order.\n * @param a The first element to be compared.\n * @param b The second element to be compared.\n * @return A negative number, zero, or a positive number as the first\n *     argument is less than, equal to, or greater than the second.\n */\nfunction defaultComparator<T>(a: T, b: T): number {\n  return a > b ? 1 : a < b ? -1 : 0;\n}\n\nfunction binarySearch_<T>(\n    arr: T[], target: T, comparator: (a: T, b: T) => number) {\n  let left = 0;\n  let right = arr.length;\n  let middle = 0;\n  let found = false;\n  while (left < right) {\n    middle = left + ((right - left) >>> 1);\n    const compareResult = comparator(target, arr[middle]);\n    if (compareResult > 0) {\n      left = middle + 1;\n    } else {\n      right = middle;\n      // If compareResult is 0, the value is found. We record it is found,\n      // and then keep looking because there may be duplicate.\n      found = !compareResult;\n    }\n  }\n\n  return found ? left : -left - 1;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Implementation of the NonMaxSuppression kernel shared between webgl and cpu.\n */\n\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {Tensor1D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\n\nimport {binaryInsert} from './array_util';\n\ninterface Candidate {\n  score: number;\n  boxIndex: number;\n  suppressBeginIndex: number;\n}\n\nexport function nonMaxSuppressionV3Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number): Tensor1D {\n  return nonMaxSuppressionImpl_(\n             boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n             0 /* softNmsSigma */)\n             .selectedIndices as Tensor1D;\n}\n\nexport function nonMaxSuppressionV4Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    padToMaxOutputSize: boolean): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold,\n      0 /* softNmsSigma */, false /* returnScoresTensor */,\n      padToMaxOutputSize /* padToMaxOutputSize */, true\n      /* returnValidOutputs */);\n}\n\nexport function nonMaxSuppressionV5Impl(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number,\n    softNmsSigma: number): NamedTensorMap {\n  return nonMaxSuppressionImpl_(\n      boxes, scores, maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma,\n      true /* returnScoresTensor */);\n}\n\nfunction nonMaxSuppressionImpl_(\n    boxes: TypedArray, scores: TypedArray, maxOutputSize: number,\n    iouThreshold: number, scoreThreshold: number, softNmsSigma: number,\n    returnScoresTensor = false, padToMaxOutputSize = false,\n    returnValidOutputs = false): NamedTensorMap {\n  // The list is sorted in ascending order, so that we can always pop the\n  // candidate with the largest score in O(1) time.\n  const candidates = [];\n\n  for (let i = 0; i < scores.length; i++) {\n    if (scores[i] > scoreThreshold) {\n      candidates.push({score: scores[i], boxIndex: i, suppressBeginIndex: 0});\n    }\n  }\n\n  candidates.sort(ascendingComparator);\n\n  // If softNmsSigma is 0, the outcome of this algorithm is exactly same as\n  // before.\n  const scale = softNmsSigma > 0 ? (-0.5 / softNmsSigma) : 0.0;\n\n  const selectedIndices: number[] = [];\n  const selectedScores: number[] = [];\n\n  while (selectedIndices.length < maxOutputSize && candidates.length > 0) {\n    const candidate = candidates.pop();\n    const {score: originalScore, boxIndex, suppressBeginIndex} = candidate;\n\n    if (originalScore < scoreThreshold) {\n      break;\n    }\n\n    // Overlapping boxes are likely to have similar scores, therefore we\n    // iterate through the previously selected boxes backwards in order to\n    // see if candidate's score should be suppressed. We use\n    // suppressBeginIndex to track and ensure a candidate can be suppressed\n    // by a selected box no more than once. Also, if the overlap exceeds\n    // iouThreshold, we simply ignore the candidate.\n    let ignoreCandidate = false;\n    for (let j = selectedIndices.length - 1; j >= suppressBeginIndex; --j) {\n      const iou = intersectionOverUnion(boxes, boxIndex, selectedIndices[j]);\n\n      if (iou >= iouThreshold) {\n        ignoreCandidate = true;\n        break;\n      }\n\n      candidate.score =\n          candidate.score * suppressWeight(iouThreshold, scale, iou);\n\n      if (candidate.score <= scoreThreshold) {\n        break;\n      }\n    }\n\n    // At this point, if `candidate.score` has not dropped below\n    // `scoreThreshold`, then we know that we went through all of the\n    // previous selections and can safely update `suppressBeginIndex` to the\n    // end of the selected array. Then we can re-insert the candidate with\n    // the updated score and suppressBeginIndex back in the candidate list.\n    // If on the other hand, `candidate.score` has dropped below the score\n    // threshold, we will not add it back to the candidates list.\n    candidate.suppressBeginIndex = selectedIndices.length;\n\n    if (!ignoreCandidate) {\n      // Candidate has passed all the tests, and is not suppressed, so\n      // select the candidate.\n      if (candidate.score === originalScore) {\n        selectedIndices.push(boxIndex);\n        selectedScores.push(candidate.score);\n      } else if (candidate.score > scoreThreshold) {\n        // Candidate's score is suppressed but is still high enough to be\n        // considered, so add back to the candidates list.\n        binaryInsert(candidates, candidate, ascendingComparator);\n      }\n    }\n  }\n\n  // NonMaxSuppressionV4 feature: padding output to maxOutputSize.\n  const validOutputs = selectedIndices.length;\n  const elemsToPad = maxOutputSize - validOutputs;\n\n  if (padToMaxOutputSize && elemsToPad > 0) {\n    selectedIndices.push(...new Array(elemsToPad).fill(0));\n    selectedScores.push(...new Array(elemsToPad).fill(0.0));\n  }\n\n  const result:\n      NamedTensorMap = {selectedIndices: tensor1d(selectedIndices, 'int32')};\n\n  if (returnScoresTensor) {\n    result['selectedScores'] = tensor1d(selectedScores, 'float32');\n  }\n\n  if (returnValidOutputs) {\n    result['validOutputs'] = scalar(validOutputs, 'int32');\n  }\n\n  return result;\n}\n\nfunction intersectionOverUnion(boxes: TypedArray, i: number, j: number) {\n  const iCoord = boxes.subarray(i * 4, i * 4 + 4);\n  const jCoord = boxes.subarray(j * 4, j * 4 + 4);\n  const yminI = Math.min(iCoord[0], iCoord[2]);\n  const xminI = Math.min(iCoord[1], iCoord[3]);\n  const ymaxI = Math.max(iCoord[0], iCoord[2]);\n  const xmaxI = Math.max(iCoord[1], iCoord[3]);\n  const yminJ = Math.min(jCoord[0], jCoord[2]);\n  const xminJ = Math.min(jCoord[1], jCoord[3]);\n  const ymaxJ = Math.max(jCoord[0], jCoord[2]);\n  const xmaxJ = Math.max(jCoord[1], jCoord[3]);\n  const areaI = (ymaxI - yminI) * (xmaxI - xminI);\n  const areaJ = (ymaxJ - yminJ) * (xmaxJ - xminJ);\n  if (areaI <= 0 || areaJ <= 0) {\n    return 0.0;\n  }\n  const intersectionYmin = Math.max(yminI, yminJ);\n  const intersectionXmin = Math.max(xminI, xminJ);\n  const intersectionYmax = Math.min(ymaxI, ymaxJ);\n  const intersectionXmax = Math.min(xmaxI, xmaxJ);\n  const intersectionArea = Math.max(intersectionYmax - intersectionYmin, 0.0) *\n      Math.max(intersectionXmax - intersectionXmin, 0.0);\n  return intersectionArea / (areaI + areaJ - intersectionArea);\n}\n\n// A Gaussian penalty function, this method always returns values in [0, 1].\n// The weight is a function of similarity, the more overlap two boxes are, the\n// smaller the weight is, meaning highly overlapping boxe will be significantly\n// penalized. On the other hand, a non-overlapping box will not be penalized.\nfunction suppressWeight(iouThreshold: number, scale: number, iou: number) {\n  const weight = Math.exp(scale * iou * iou);\n  return iou <= iouThreshold ? weight : 0.0;\n}\n\nfunction ascendingComparator(c1: Candidate, c2: Candidate) {\n  // For objects with same scores, we make the object with the larger index go\n  // first. In an array that pops from the end, this means that the object with\n  // the smaller index will be popped first. This ensures the same output as\n  // the TensorFlow python version.\n  return (c1.score - c2.score) ||\n      ((c1.score === c2.score) && (c2.boxIndex - c1.boxIndex));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {nonMaxSuppressionV3Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This is the async version of `nonMaxSuppression`\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @return A 1D tensor with the selected box indices.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY): Promise<Tensor1D> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const inputs = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold);\n  maxOutputSize = inputs.maxOutputSize;\n  iouThreshold = inputs.iouThreshold;\n  scoreThreshold = inputs.scoreThreshold;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV3Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionAsync = nonMaxSuppressionAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionWithScore_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const inputs: NonMaxSuppressionV5Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV5Attrs =\n      {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma};\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV5, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], selectedScores: result[1]};\n}\n\nexport const nonMaxSuppressionWithScore = op({nonMaxSuppressionWithScore_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV5Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union).\n *\n * This op also supports a Soft-NMS mode (c.f.\n * Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score\n * of other overlapping boxes, therefore favoring different regions of the image\n * with high scores. To enable this Soft-NMS mode, set the `softNmsSigma`\n * parameter to be larger than 0.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param softNmsSigma A float representing the sigma parameter for Soft NMS.\n *     When sigma is 0, it falls back to nonMaxSuppression.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - selectedScores: A 1D tensor with the corresponding scores for each\n *       selected box.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionWithScoreAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    softNmsSigma = 0.0): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n  maxOutputSize = params.maxOutputSize;\n  iouThreshold = params.iouThreshold;\n  scoreThreshold = params.scoreThreshold;\n  softNmsSigma = params.softNmsSigma;\n\n  const boxesAndScores = await Promise.all([$boxes.data(), $scores.data()]);\n  const boxesVals = boxesAndScores[0];\n  const scoresVals = boxesAndScores[1];\n\n  // We call a cpu based impl directly with the typedarray data  here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV5Impl(\n      boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n      softNmsSigma);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionWithScoreAsync = nonMaxSuppressionWithScoreAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor, Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\nimport {op} from '../operation';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction nonMaxSuppressionPadded_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): NamedTensorMap {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppression');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppression');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const inputs: NonMaxSuppressionV4Inputs = {boxes: $boxes, scores: $scores};\n  const attrs: NonMaxSuppressionV4Attrs = {\n    maxOutputSize: $maxOutputSize,\n    iouThreshold: $iouThreshold,\n    scoreThreshold: $scoreThreshold,\n    padToMaxOutputSize\n  };\n\n  const result = ENGINE.runKernel(\n                     NonMaxSuppressionV4, inputs as {} as NamedTensorMap,\n                     attrs as {} as NamedAttrMap) as Tensor[];\n\n  return {selectedIndices: result[0], validOutputs: result[1]};\n}\n\nexport const nonMaxSuppressionPadded = op({nonMaxSuppressionPadded_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {nonMaxSuppressionV4Impl} from '../../backends/non_max_suppression_impl';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {nonMaxSuppSanityCheck} from '../nonmax_util';\n\n/**\n * Asynchronously performs non maximum suppression of bounding boxes based on\n * iou (intersection over union), with an option to pad results.\n *\n * @param boxes a 2d tensor of shape `[numBoxes, 4]`. Each entry is\n *     `[y1, x1, y2, x2]`, where `(y1, x1)` and `(y2, x2)` are the corners of\n *     the bounding box.\n * @param scores a 1d tensor providing the box scores of shape `[numBoxes]`.\n * @param maxOutputSize The maximum number of boxes to be selected.\n * @param iouThreshold A float representing the threshold for deciding whether\n *     boxes overlap too much with respect to IOU. Must be between [0, 1].\n *     Defaults to 0.5 (50% box overlap).\n * @param scoreThreshold A threshold for deciding when to remove boxes based\n *     on score. Defaults to -inf, which means any score is accepted.\n * @param padToMaxOutputSize Defalts to false. If true, size of output\n *     `selectedIndices` is padded to maxOutputSize.\n * @return A map with the following properties:\n *     - selectedIndices: A 1D tensor with the selected box indices.\n *     - validOutputs: A scalar denoting how many elements in `selectedIndices`\n *       are valid. Valid elements occur first, then padding.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nasync function nonMaxSuppressionPaddedAsync_(\n    boxes: Tensor2D|TensorLike, scores: Tensor1D|TensorLike,\n    maxOutputSize: number, iouThreshold = 0.5,\n    scoreThreshold = Number.NEGATIVE_INFINITY,\n    padToMaxOutputSize = false): Promise<NamedTensorMap> {\n  const $boxes = convertToTensor(boxes, 'boxes', 'nonMaxSuppressionAsync');\n  const $scores = convertToTensor(scores, 'scores', 'nonMaxSuppressionAsync');\n\n  const params = nonMaxSuppSanityCheck(\n      $boxes, $scores, maxOutputSize, iouThreshold, scoreThreshold,\n      null /* softNmsSigma */);\n  const $maxOutputSize = params.maxOutputSize;\n  const $iouThreshold = params.iouThreshold;\n  const $scoreThreshold = params.scoreThreshold;\n\n  const [boxesVals, scoresVals] =\n      await Promise.all([$boxes.data(), $scores.data()]);\n\n  // We call a cpu based impl directly with the typedarray data here rather\n  // than a kernel because all kernels are synchronous (and thus cannot await\n  // .data()).\n  const res = nonMaxSuppressionV4Impl(\n      boxesVals, scoresVals, $maxOutputSize, $iouThreshold, $scoreThreshold,\n      padToMaxOutputSize);\n\n  if ($boxes !== boxes) {\n    $boxes.dispose();\n  }\n  if ($scores !== scores) {\n    $scores.dispose();\n  }\n  return res;\n}\n\nexport const nonMaxSuppressionPaddedAsync = nonMaxSuppressionPaddedAsync_;\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * Bilinear resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeBilinear_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeBilinear');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeBilinear: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () => `Error in resizeBilinear: new shape must 2D, but got shape ` +\n          `${size}.`);\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n\n  const [newHeight, newWidth] = size;\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeBilinear(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const inputs: ResizeBilinearInputs = {images: batchImages};\n  const attrs: ResizeBilinearAttrs = {alignCorners, size};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeBilinear, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeBilinear = op({resizeBilinear_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborInputs} from '../../kernel_names';\nimport {NamedAttrMap} from '../../kernel_registry';\nimport {Tensor3D, Tensor4D} from '../../tensor';\nimport {NamedTensorMap} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport * as util from '../../util';\n\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\n\n/**\n * NearestNeighbor resize a batch of 3D images to a new shape.\n *\n * @param images The images, of rank 4 or rank 3, of shape\n *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.\n * @param size The new shape `[newHeight, newWidth]` to resize the\n *     images to. Each channel is resized individually.\n * @param alignCorners Defaults to False. If true, rescale\n *     input by `(new_height - 1) / (height - 1)`, which exactly aligns the 4\n *     corners of images and resized images. If false, rescale by\n *     `new_height / height`. Treat similarly the width dimension.\n *\n * @doc {heading: 'Operations', subheading: 'Images', namespace: 'image'}\n */\nfunction resizeNearestNeighbor_<T extends Tensor3D|Tensor4D>(\n    images: T|TensorLike, size: [number, number], alignCorners = false): T {\n  const $images = convertToTensor(images, 'images', 'resizeNearestNeighbor');\n\n  util.assert(\n      $images.rank === 3 || $images.rank === 4,\n      () => `Error in resizeNearestNeighbor: x must be rank 3 or 4, but got ` +\n          `rank ${$images.rank}.`);\n  util.assert(\n      size.length === 2,\n      () =>\n          `Error in resizeNearestNeighbor: new shape must 2D, but got shape ` +\n          `${size}.`);\n  util.assert(\n      $images.dtype === 'float32' || $images.dtype === 'int32',\n      () => '`images` must have `int32` or `float32` as dtype');\n\n  let batchImages = $images as Tensor4D;\n  let reshapedTo4D = false;\n  if ($images.rank === 3) {\n    reshapedTo4D = true;\n    batchImages = reshape(\n        $images, [1, $images.shape[0], $images.shape[1], $images.shape[2]]);\n  }\n  const [newHeight, newWidth] = size;\n\n  const inputs: ResizeNearestNeighborInputs = {images: batchImages};\n  const attrs: ResizeNearestNeighborAttrs = {alignCorners, size};\n\n  const forward: ForwardFunc<Tensor4D> = (backend, save) => {\n    save([batchImages]);\n    return backend.resizeNearestNeighbor(\n        batchImages, newHeight, newWidth, alignCorners);\n  };\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* gradient */,\n      ResizeNearestNeighbor, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const resizeNearestNeighbor = op({resizeNearestNeighbor_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assert} from '../../util';\n\nimport {greaterEqual} from '../greater_equal';\nimport {lessEqual} from '../less_equal';\nimport {logicalAnd} from '../logical_and';\nimport {op} from '../operation';\nimport {range} from '../range';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\nimport {zeros} from '../zeros';\n\n/**\n * Copy a tensor setting everything outside a central band in each innermost\n * matrix to zero.\n *\n * The band part is computed as follows: Assume input has `k` dimensions\n * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where\n * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`.\n * The indicator function\n * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))`\n * `&& (num_upper < 0 || (n-m) <= num_upper)`\n *\n * ```js\n * const x = tf.tensor2d([[ 0,  1,  2, 3],\n *                        [-1,  0,  1, 2],\n *                        [-2, -1,  0, 1],\n *                        [-3, -2, -1, 0]]);\n * let y = tf.linalg.bandPart(x, 1, -1);\n * y.print(); // [[ 0,  1,  2, 3],\n *            //  [-1,  0,  1, 2],\n *            //  [ 0, -1,  0, 1],\n *            //  [ 0, 0 , -1, 0]]\n * let z = tf.linalg.bandPart(x, 2, 1);\n * z.print(); // [[ 0,  1,  0, 0],\n *            //  [-1,  0,  1, 0],\n *            //  [-2, -1,  0, 1],\n *            //  [ 0, -2, -1, 0]]\n * ```\n *\n * @param x Rank `k` tensor\n * @param numLower Number of subdiagonals to keep.\n *   If negative, keep entire lower triangle.\n * @param numUpper Number of subdiagonals to keep.\n *   If negative, keep entire upper triangle.\n * @returns Rank `k` tensor of the same shape as input.\n *   The extracted banded tensor.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction bandPart_<T extends Tensor>(\n    a: T|TensorLike, numLower: number, numUpper: number): T {\n  assert(\n      numLower % 1 === 0,\n      () => `bandPart(): numLower must be an integer, got ${numLower}.`);\n  assert(\n      numUpper % 1 === 0,\n      () => `bandPart(): numUpper must be an integer, got ${numUpper}.`);\n\n  const $a = convertToTensor(a, 'a', 'bandPart');\n\n  assert(\n      $a.rank >= 2,\n      () => `bandPart(): Rank must be at least 2, got ${$a.rank}.`);\n\n  const shape = $a.shape;\n  const [M, N] = $a.shape.slice(-2);\n\n  if (!(numLower <= M)) {\n    throw new Error(\n        `bandPart(): numLower (${numLower})` +\n        ` must not be greater than the number of rows (${M}).`);\n  }\n  if (!(numUpper <= N)) {\n    throw new Error(\n        `bandPart(): numUpper (${numUpper})` +\n        ` must not be greater than the number of columns (${N}).`);\n  }\n\n  if (numLower < 0) {\n    numLower = M;\n  }\n  if (numUpper < 0) {\n    numUpper = N;\n  }\n\n  const i = reshape(range(0, M, 1, 'int32'), [-1, 1]);\n  const j = range(0, N, 1, 'int32');\n  const ij = sub(i, j);\n\n  const inBand = logicalAnd(\n      lessEqual(ij, scalar(+numLower, 'int32')),\n      greaterEqual(ij, scalar(-numUpper, 'int32')));\n\n  const zero = zeros([M, N], $a.dtype);\n\n  return reshape(\n             stack(unstack(reshape($a, [-1, M, N]))\n                       .map(mat => where(inBand, mat, zero))),\n             shape) as T;\n}\n\nexport const bandPart = op({bandPart_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../../engine';\nimport {Tensor1D, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {div} from '../div';\nimport {mul} from '../mul';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {split} from '../split';\nimport {squeeze} from '../squeeze';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\n/**\n * Gram-Schmidt orthogonalization.\n *\n * ```js\n * const x = tf.tensor2d([[1, 2], [3, 4]]);\n * let y = tf.linalg.gramSchmidt(x);\n * y.print();\n * console.log('Othogonalized:');\n * y.dot(y.transpose()).print();  // should be nearly the identity matrix.\n * console.log('First row direction maintained:');\n * const data = await y.array();\n * console.log(data[0][1] / data[0][0]);  // should be nearly 2.\n * ```\n *\n * @param xs The vectors to be orthogonalized, in one of the two following\n *   formats:\n *   - An Array of `tf.Tensor1D`.\n *   - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows\n *     of `xs`.\n *   In each case, all the vectors must have the same length and the length\n *   must be greater than or equal to the number of vectors.\n * @returns The orthogonalized and normalized vectors or matrix.\n *   Orthogonalization means that the vectors or the rows of the matrix\n *   are orthogonal (zero inner products). Normalization means that each\n *   vector or each row of the matrix has an L2 norm that equals `1`.\n *\n * @doc {heading:'Operations', subheading:'Linear Algebra', namespace:'linalg'}\n */\nfunction gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D {\n  let inputIsTensor2D: boolean;\n  if (Array.isArray(xs)) {\n    inputIsTensor2D = false;\n    assert(\n        xs != null && xs.length > 0,\n        () => 'Gram-Schmidt process: input must not be null, undefined, or ' +\n            'empty');\n    const dim = xs[0].shape[0];\n    for (let i = 1; i < xs.length; ++i) {\n      assert(\n          xs[i].shape[0] === dim,\n          () =>\n              'Gram-Schmidt: Non-unique lengths found in the input vectors: ' +\n              `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`);\n    }\n  } else {\n    inputIsTensor2D = true;\n    xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0]));\n  }\n\n  assert(\n      xs.length <= xs[0].shape[0],\n      () => `Gram-Schmidt: Number of vectors (${\n                (xs as Tensor1D[]).length}) exceeds ` +\n          `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`);\n\n  const ys: Tensor1D[] = [];\n  const xs1d = xs;\n  for (let i = 0; i < xs.length; ++i) {\n    ys.push(ENGINE.tidy(() => {\n      let x = xs1d[i];\n      if (i > 0) {\n        for (let j = 0; j < i; ++j) {\n          const proj = mul(sum(mul(ys[j], x)), ys[j]);\n          x = sub(x, proj);\n        }\n      }\n      return div(x, norm(x, 'euclidean'));\n    }));\n  }\n\n  if (inputIsTensor2D) {\n    return stack(ys, 0) as Tensor2D;\n  } else {\n    return ys;\n  }\n}\n\nexport const gramSchmidt = op({gramSchmidt_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../../engine';\nimport {dispose} from '../../globals';\nimport {Tensor, Tensor2D} from '../../tensor';\nimport {assert} from '../../util';\n\nimport {clone} from '../clone';\nimport {concat} from '../concat';\nimport {div} from '../div';\nimport {eye} from '../eye';\nimport {greater} from '../greater';\nimport {matMul} from '../mat_mul';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {norm} from '../norm';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {slice} from '../slice';\nimport {stack} from '../stack';\nimport {sub} from '../sub';\nimport {tensor2d} from '../tensor2d';\nimport {transpose} from '../transpose';\nimport {unstack} from '../unstack';\nimport {where} from '../where';\n\n/**\n * Compute QR decomposition of m-by-n matrix using Householder transformation.\n *\n * Implementation based on\n *   [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf]\n * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf)\n *\n * ```js\n * const a = tf.tensor2d([[1, 2], [3, 4]]);\n * let [q, r] = tf.linalg.qr(a);\n * console.log('Q');\n * q.print();\n * console.log('R');\n * r.print();\n * console.log('Orthogonalized');\n * q.dot(q.transpose()).print()  // should be nearly the identity matrix.\n * console.log('Reconstructed');\n * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]];\n * ```\n *\n * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose\n *   it has the shape `[..., M, N]`.\n * @param fullMatrices An optional boolean parameter. Defaults to `false`.\n *   If `true`, compute full-sized `Q`. If `false` (the default),\n *   compute only the leading N columns of `Q` and `R`.\n * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix,\n *   i.e., its columns all have unit norm and are mutually orthogonal.\n *   If `M >= N`,\n *     If `fullMatrices` is `false` (default),\n *       - `Q` has a shape of `[..., M, N]`,\n *       - `R` has a shape of `[..., N, N]`.\n *     If `fullMatrices` is `true` (default),\n *       - `Q` has a shape of `[..., M, M]`,\n *       - `R` has a shape of `[..., M, N]`.\n *   If `M < N`,\n *     - `Q` has a shape of `[..., M, M]`,\n *     - `R` has a shape of `[..., M, N]`.\n * @throws If the rank of `x` is less than 2.\n *\n * @doc {heading:'Operations',\n *       subheading:'Linear Algebra',\n *       namespace:'linalg'}\n */\nfunction qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] {\n  assert(\n      x.rank >= 2,\n      () => `qr() requires input tensor to have a rank >= 2, but got rank ${\n          x.rank}`);\n\n  if (x.rank === 2) {\n    return qr2d(x as Tensor2D, fullMatrices);\n  } else {\n    // Rank > 2.\n    // TODO(cais): Below we split the input into individual 2D tensors,\n    //   perform QR decomposition on them and then stack the results back\n    //   together. We should explore whether this can be parallelized.\n    const outerDimsProd = x.shape.slice(0, x.shape.length - 2)\n                              .reduce((value, prev) => value * prev);\n    const x2ds = unstack(\n        reshape(\n            x,\n            [\n              outerDimsProd, x.shape[x.shape.length - 2],\n              x.shape[x.shape.length - 1]\n            ]),\n        0);\n    const q2ds: Tensor2D[] = [];\n    const r2ds: Tensor2D[] = [];\n    x2ds.forEach(x2d => {\n      const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices);\n      q2ds.push(q2d);\n      r2ds.push(r2d);\n    });\n    const q = reshape(stack(q2ds, 0), x.shape);\n    const r = reshape(stack(r2ds, 0), x.shape);\n    return [q, r];\n  }\n}\n\nfunction qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] {\n  return ENGINE.tidy(() => {\n    assert(\n        x.shape.length === 2,\n        () => `qr2d() requires a 2D Tensor, but got a ${\n            x.shape.length}D Tensor.`);\n\n    const m = x.shape[0];\n    const n = x.shape[1];\n\n    let q = eye(m);    // Orthogonal transform so far.\n    let r = clone(x);  // Transformed matrix so far.\n\n    const one2D = tensor2d([[1]], [1, 1]);\n    let w: Tensor2D = clone(one2D);\n\n    const iters = m >= n ? n : m;\n    for (let j = 0; j < iters; ++j) {\n      // This tidy within the for-loop ensures we clean up temporary\n      // tensors as soon as they are no longer needed.\n      const rTemp = r;\n      const wTemp = w;\n      const qTemp = q;\n      [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => {\n        // Find H = I - tau * w * w', to put zeros below R(j, j).\n        const rjEnd1 = slice(r, [j, j], [m - j, 1]);\n        const normX = norm(rjEnd1);\n        const rjj = slice(r, [j, j], [1, 1]);\n\n        // The sign() function returns 0 on 0, which causes division by zero.\n        const s = where(greater(rjj, 0), tensor2d([[-1]]), tensor2d([[1]]));\n\n        const u1 = sub(rjj, mul(s, normX));\n        const wPre = div(rjEnd1, u1);\n        if (wPre.shape[0] === 1) {\n          w = clone(one2D);\n        } else {\n          w = concat(\n              [\n                one2D,\n                slice(wPre, [1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as\n                    Tensor2D\n              ],\n              0);\n        }\n        const tau = neg(div(matMul(s, u1), normX)) as Tensor2D;\n\n        // -- R := HR, Q := QH.\n        const rjEndAll = slice(r, [j, 0], [m - j, n]);\n        const tauTimesW: Tensor2D = mul(tau, w);\n        const wT: Tensor2D = transpose(w);\n        if (j === 0) {\n          r = sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n        } else {\n          const rTimesTau: Tensor2D =\n              sub(rjEndAll, matMul(tauTimesW, matMul(wT, rjEndAll)));\n          r = concat([slice(r, [0, 0], [j, n]), rTimesTau], 0);\n        }\n        const tawTimesWT: Tensor2D = transpose(tauTimesW);\n        const qAllJEnd = slice(q, [0, j], [m, q.shape[1] - j]);\n        if (j === 0) {\n          q = sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n        } else {\n          const qTimesTau: Tensor2D =\n              sub(qAllJEnd, matMul(matMul(qAllJEnd, w), tawTimesWT));\n          q = concat([slice(q, [0, 0], [m, j]), qTimesTau], 1);\n        }\n        return [w, r, q];\n      });\n      dispose([rTemp, wTemp, qTemp]);\n    }\n\n    if (!fullMatrices && m > n) {\n      q = slice(q, [0, 0], [m, n]);\n      r = slice(r, [0, 0], [n, n]);\n    }\n\n    return [q, r];\n  }) as [Tensor2D, Tensor2D];\n}\n\nexport const qr = op({qr_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport enum Reduction {\n  NONE,\n  MEAN,\n  SUM,\n  SUM_BY_NONZERO_WEIGHTS\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\n\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {Reduction} from '../loss_ops_utils';\nimport {mean} from '../mean';\nimport {mul} from '../mul';\nimport {notEqual} from '../not_equal';\nimport {ones} from '../ones';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sum} from '../sum';\n\n/**\n * Computes the weighted loss between two tensors.\n *\n * @param losses Tensor of shape `[batch_size, d1, ... dN]`.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `losses`, and must be broadcastable to `losses` (i.e., all\n *    dimensions must be either `1`, or the same as the corresponding\n *    `losses` dimension).\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction computeWeightedLoss_<T extends Tensor, O extends Tensor>(\n    losses: T|TensorLike, weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $losses = convertToTensor(losses, 'losses', 'computeWeightedLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'computeWeightedLoss');\n  }\n\n  const weightedLoss = ($weights == null) ? $losses : mul($losses, $weights);\n\n  if (reduction === Reduction.NONE) {\n    return weightedLoss as O;\n  }\n  if (reduction === Reduction.SUM) {\n    return sum(weightedLoss);\n  }\n  if (reduction === Reduction.MEAN) {\n    if ($weights == null) {\n      return mean(weightedLoss);\n    } else {\n      const broadcastFactor = $losses.size / $weights.size;\n      const result = div(sum(weightedLoss), sum($weights));\n      return broadcastFactor > 1 ? div(result, scalar(broadcastFactor)) :\n                                   result as O;\n    }\n  }\n  if (reduction === Reduction.SUM_BY_NONZERO_WEIGHTS) {\n    if ($weights == null) {\n      return div(sum(weightedLoss), scalar($losses.size));\n    } else {\n      const broadcastedWeights = mul($weights, ones($losses.shape));\n\n      const numNonZeros =\n          cast(sum(notEqual(broadcastedWeights, scalar(0))), 'float32');\n      return div(sum(weightedLoss), numNonZeros);\n    }\n  }\n\n  throw Error(`Unknown reduction: ${reduction}`);\n}\nexport const computeWeightedLoss = op({computeWeightedLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the absolute difference loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction absoluteDifference_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'absoluteDifference');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'absoluteDifference');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'absoluteDifference');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in absoluteDifference: ');\n\n  const losses = abs(sub($labels, $predictions));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const absoluteDifference = op({absoluteDifference_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the cosine distance loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param axis The dimension along which the cosine distance is computed.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction cosineDistance_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike, axis: number,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'cosineDistance');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'cosineDistance');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'cosineDistance');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in cosineDistance: ');\n\n  const one = scalar(1);\n  const losses = sub(one, sum(mul($labels, $predictions), axis, true));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const cosineDistance = op({cosineDistance_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the Hinge loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction hingeLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $labels = convertToTensor(labels, 'labels', 'hingeLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'hingeLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'hingeLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in hingeLoss: ');\n\n  const one = scalar(1);\n  // Convert binary labels to (-1, 1)\n  $labels = sub(mul(scalar(2), $labels), one);\n  const losses = relu(sub(one, mul($labels, $predictions)));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const hingeLoss = op({hingeLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {Reduction} from '../loss_ops_utils';\nimport {minimum} from '../minimum';\nimport {mul} from '../mul';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {square} from '../square';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the huber loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param delta Point where huber loss changes from quadratic to linear.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`.\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction huberLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, delta = 1.0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'huberLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'huberLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'huberLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in huberLoss: ');\n\n  const deltaScalar = scalar(delta);\n  const error = abs(sub($predictions, $labels));\n  const quadratic = minimum(error, deltaScalar);\n  const linear = sub(error, quadratic);\n\n  const losses =\n      add(mul(scalar(0.5), square(quadratic)), mul(deltaScalar, linear));\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const huberLoss = op({huberLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {log} from '../log';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the log loss between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param epsilon A small increment to avoid taking log of zero\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction logLoss_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike, epsilon = 1e-7,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'logLoss');\n  const $predictions = convertToTensor(predictions, 'predictions', 'logLoss');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'logLoss');\n  }\n  assertShapesMatch($labels.shape, $predictions.shape, 'Error in logLoss: ');\n\n  const one = scalar(1);\n  const epsilonScalar = scalar(epsilon);\n\n  const l1 = neg(mul($labels, log(add($predictions, epsilonScalar))));\n  const l2 =\n      mul(sub(one, $labels), log(add(sub(one, $predictions), epsilonScalar)));\n  const losses = sub(l1, l2);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const logLoss = op({logLoss_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {Reduction} from '../loss_ops_utils';\nimport {op} from '../operation';\nimport {squaredDifference} from '../squared_difference';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes the mean squared error between two tensors.\n *\n * @param labels The ground truth output tensor, same dimensions as\n *    'predictions'.\n * @param predictions The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc {heading: 'Training', subheading: 'Losses', namespace: 'losses'}\n */\nfunction meanSquaredError_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, predictions: T|TensorLike,\n    weights?: Tensor|TensorLike,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  const $labels = convertToTensor(labels, 'labels', 'meanSquaredError');\n  const $predictions =\n      convertToTensor(predictions, 'predictions', 'meanSquaredError');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'meanSquaredError');\n  }\n  assertShapesMatch(\n      $labels.shape, $predictions.shape, 'Error in meanSquaredError: ');\n\n  const losses = squaredDifference($labels, $predictions);\n  return computeWeightedLoss(losses, $weights, reduction);\n}\nexport const meanSquaredError = op({meanSquaredError_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../../tensor';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {abs} from '../abs';\nimport {add} from '../add';\nimport {exp} from '../exp';\nimport {log1p} from '../log1p';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {relu} from '../relu';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\nfunction sigmoidCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T|TensorLike, logits: T|TensorLike): O {\n  const $labels =\n      convertToTensor(labels, 'labels', 'sigmoidCrossEntropyWithLogits');\n  const $logits =\n      convertToTensor(logits, 'logits', 'sigmoidCrossEntropyWithLogits');\n  assertShapesMatch(\n      $labels.shape, $logits.shape, 'Error in sigmoidCrossEntropyWithLogits: ');\n\n  /**\n   * Implementation Details:\n   *\n   * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n   *     z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n   *   = (1 - z) * x + log(1 + exp(-x))\n   *   = x - x * z + log(1 + exp(-x))\n   *\n   *   For x < 0, to avoid overflow in exp(-x), we reformulate the above\n   *     x - x * z + log(1 + exp(-x))\n   *   = log(exp(x)) - x * z + log(1 + exp(-x))\n   *   = - x * z + log(1 + exp(x))\n   *\n   * Hence, to ensure stability and avoid overflow, the implementation uses\n   * this equivalent formulation:\n   *     max(x, 0) - x * z + log(1 + exp(-abs(x)))\n   */\n  const maxOutput = relu($logits);\n  const outputXTarget = mul($logits, $labels);\n  const sigmoidOutput = log1p(exp(neg(abs($logits))));\n\n  return add(sub(maxOutput, outputXTarget), sigmoidOutput);\n}\n\n/**\n * Computes the sigmoid cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newMulticlassLabels = multiclassLabels * (1 - labelSmoothing)\n *                         + 0.5 * labelSmoothing\n *\n * @param multiClassLabels The ground truth output tensor of shape\n * [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or the same rank as\n *    `labels`, and must be broadcastable to `labels` (i.e., all dimensions\n *    must be either `1`, or the same as the corresponding `losses`\n *    dimension).\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction sigmoidCrossEntropy_<T extends Tensor, O extends Tensor>(\n    multiClassLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $multiClassLabels = convertToTensor(\n      multiClassLabels, 'multiClassLabels', 'sigmoidCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'sigmoidCrossEntropy');\n  let $weights: Tensor = null;\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'sigmoidCrossEntropy');\n  }\n  assertShapesMatch(\n      $multiClassLabels.shape, $logits.shape, 'Error in sigmoidCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const half = scalar(0.5);\n\n    $multiClassLabels =\n        add(mul($multiClassLabels, sub(one, labelSmoothingScalar)),\n            mul(half, labelSmoothingScalar));\n  }\n  const losses = sigmoidCrossEntropyWithLogits_($multiClassLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const sigmoidCrossEntropy = op({sigmoidCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {customGrad} from '../../gradients';\nimport {Tensor} from '../../tensor';\nimport {GradSaveFunc} from '../../tensor_types';\nimport {convertToTensor} from '../../tensor_util_env';\nimport {TensorLike} from '../../types';\nimport {assertShapesMatch} from '../../util';\nimport {add} from '../add';\nimport {expandShapeToKeepDim} from '../axis_util';\nimport {cast} from '../cast';\nimport {div} from '../div';\nimport {exp} from '../exp';\nimport {logSumExp} from '../log_sum_exp';\nimport {Reduction} from '../loss_ops_utils';\nimport {mul} from '../mul';\nimport {neg} from '../neg';\nimport {op} from '../operation';\nimport {reshape} from '../reshape';\nimport {scalar} from '../scalar';\nimport {sub} from '../sub';\nimport {sum} from '../sum';\n\nimport {computeWeightedLoss} from './compute_weighted_loss';\n\n/**\n * Computes softmax cross entropy between logits and labels.\n *\n * Measures the probability error in discrete classification tasks in which\n * the classes are mutually exclusive (each entry is in exactly one class).\n * For example, each CIFAR-10 image is labeled with one and only one label: an\n * image can be a dog or a truck, but not both.\n *\n * `NOTE`: While the classes are mutually exclusive, their probabilities need\n * not be. All that is required is that each row of labels is a valid\n * probability distribution. If they are not, the computation of the gradient\n * will be incorrect.\n *\n * `WARNING`: This op expects unscaled logits, since it performs a softmax on\n * logits internally for efficiency. Do not call this op with the output of\n * softmax, as it will produce incorrect results.\n *\n * logits and labels must have the same shape, e.g. [batch_size, num_classes]\n * and the same dtype.\n * @param labels The labels array.\n * @param logits The logits array.\n * @param dim The dimension softmax would be performed on. Defaults to `-1`\n *     which indicates the last dimension.\n */\nfunction softmaxCrossEntropyWithLogits_<T extends Tensor, O extends Tensor>(\n    labels: T, logits: T, dim = -1): O {\n  if (dim === -1) {\n    dim = logits.rank - 1;\n  }\n\n  if (dim !== logits.rank - 1) {\n    throw Error(\n        `Softmax cross entropy along a non-last dimension is not yet ` +\n        `supported. Labels / logits was rank ${logits.rank} ` +\n        `and dim was ${dim}`);\n  }\n  // Use a custom gradient for numerical stability.\n  const customOp =\n      customGrad((labels: Tensor, logits: Tensor, save: GradSaveFunc) => {\n        // Reference:\n        //   1. http://cs231n.github.io/linear-classify/#softmax\n        //   2. https://blog.feedly.com/tricks-of-the-trade-logsumexp/\n        const keepDims = true;\n        const lse = logSumExp(logits, [dim], keepDims);\n        const logResult = sub(cast(logits, 'float32'), lse);\n        save([labels, logResult]);\n\n        const costVector = neg(mul(logResult, labels));\n        const value: O = sum(costVector, [dim]);\n\n        const gradFunc = (dy: O, saved: Tensor[]) => {\n          const [labels, logResult] = saved;\n          const dyShape = expandShapeToKeepDim(dy.shape, [dim]);\n          return [\n            mul(reshape(dy, dyShape),\n                sub(cast(labels, 'float32'), exp(logResult))),\n            mul(reshape(dy, dyShape),\n                sub(exp(logResult), cast(labels, 'float32'))),\n          ];\n        };\n        return {value, gradFunc};\n      });\n\n  return customOp(labels, logits);\n}\n\n/**\n * Computes the softmax cross entropy loss between two tensors.\n *\n * If labelSmoothing is nonzero, smooth the labels towards 1/2:\n *\n *   newOnehotLabels = onehotLabels * (1 - labelSmoothing)\n *                         + labelSmoothing / numClasses\n *\n * @param onehotLabels One hot encoded labels\n *    [batch_size, num_classes], same dimensions as 'predictions'.\n * @param logits The predicted outputs.\n * @param weights Tensor whose rank is either 0, or 1, and must be\n *    broadcastable to `loss`  of shape [batch_size]\n * @param labelSmoothing If greater than 0, then smooth the labels.\n * @param reduction Type of reduction to apply to loss. Should be of type\n *    `Reduction`\n *\n * @doc { heading: 'Training', subheading: 'Losses', namespace: 'losses' }\n */\nfunction softmaxCrossEntropy_<T extends Tensor, O extends Tensor>(\n    onehotLabels: T|TensorLike, logits: T|TensorLike,\n    weights?: Tensor|TensorLike, labelSmoothing = 0,\n    reduction = Reduction.SUM_BY_NONZERO_WEIGHTS): O {\n  let $onehotLabels =\n      convertToTensor(onehotLabels, 'onehotLabels', 'softmaxCrossEntropy');\n  const $logits = convertToTensor(logits, 'logits', 'softmaxCrossEntropy');\n  let $weights: Tensor = null;\n\n  if (weights != null) {\n    $weights = convertToTensor(weights, 'weights', 'softmaxCrossEntropy');\n  }\n\n  assertShapesMatch(\n      $onehotLabels.shape, $logits.shape, 'Error in softmaxCrossEntropy: ');\n\n  if (labelSmoothing > 0) {\n    const labelSmoothingScalar = scalar(labelSmoothing);\n    const one = scalar(1);\n    const numClasses = scalar($onehotLabels.shape[1]);\n\n    $onehotLabels =\n        add(mul($onehotLabels, sub(one, labelSmoothingScalar)),\n            div(labelSmoothingScalar, numClasses));\n  }\n\n  const losses = softmaxCrossEntropyWithLogits_($onehotLabels, $logits);\n\n  return computeWeightedLoss(losses, $weights, reduction);\n}\n\nexport const softmaxCrossEntropy = op({softmaxCrossEntropy_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Modularized ops.\nexport {abs} from './abs';\nexport {acos} from './acos';\nexport {acosh} from './acosh';\nexport {add} from './add';\nexport {addN} from './add_n';\nexport {all} from './all';\nexport {any} from './any';\nexport {argMax} from './arg_max';\nexport {argMin} from './arg_min';\nexport {asin} from './asin';\nexport {asinh} from './asinh';\nexport {atan} from './atan';\nexport {atan2} from './atan2';\nexport {atanh} from './atanh';\nexport {avgPool} from './avg_pool';\nexport {avgPool3d} from './avg_pool_3d';\nexport {basicLSTMCell} from './basic_lstm_cell';\nexport {batchToSpaceND} from './batch_to_space_nd';\nexport {batchNorm} from './batchnorm';\nexport {batchNorm2d} from './batchnorm2d';\nexport {batchNorm3d} from './batchnorm3d';\nexport {batchNorm4d} from './batchnorm4d';\nexport {broadcastTo} from './broadcast_to';\nexport {buffer} from './buffer';\nexport {cast} from './cast';\nexport {ceil} from './ceil';\nexport {clipByValue} from './clip_by_value';\nexport {clone} from './clone';\nexport {complex} from './complex';\nexport {concat} from './concat';\nexport {concat1d} from './concat_1d';\nexport {concat2d} from './concat_2d';\nexport {concat3d} from './concat_3d';\nexport {concat4d} from './concat_4d';\nexport {conv1d} from './conv1d';\nexport {conv2d} from './conv2d';\nexport {conv2dTranspose} from './conv2d_transpose';\nexport {conv3d} from './conv3d';\nexport {conv3dTranspose} from './conv3d_transpose';\nexport {cos} from './cos';\nexport {cosh} from './cosh';\nexport {cumsum} from './cumsum';\nexport {depthToSpace} from './depth_to_space';\nexport {depthwiseConv2d} from './depthwise_conv2d';\nexport {diag} from './diag';\nexport {dilation2d} from './dilation2d';\nexport {div} from './div';\nexport {divNoNan} from './div_no_nan';\nexport {dot} from './dot';\nexport {elu} from './elu';\nexport {equal} from './equal';\nexport {erf} from './erf';\nexport {exp} from './exp';\nexport {expandDims} from './expand_dims';\nexport {expm1} from './expm1';\nexport {eye} from './eye';\nexport {fill} from './fill';\nexport {floor} from './floor';\nexport {floorDiv} from './floorDiv';\nexport {gather} from './gather';\nexport {greater} from './greater';\nexport {greaterEqual} from './greater_equal';\nexport {imag} from './imag';\nexport {isFinite} from './is_finite';\nexport {isInf} from './is_inf';\nexport {isNaN} from './is_nan';\nexport {leakyRelu} from './leaky_relu';\nexport {less} from './less';\nexport {lessEqual} from './less_equal';\nexport {linspace} from './linspace';\nexport {localResponseNormalization} from './local_response_normalization';\nexport {log} from './log';\nexport {log1p} from './log1p';\nexport {logSigmoid} from './log_sigmoid';\nexport {logSoftmax} from './log_softmax';\nexport {logSumExp} from './log_sum_exp';\nexport {logicalAnd} from './logical_and';\nexport {logicalNot} from './logical_not';\nexport {logicalOr} from './logical_or';\nexport {logicalXor} from './logical_xor';\nexport {matMul} from './mat_mul';\nexport {max} from './max';\nexport {maxPool} from './max_pool';\nexport {maxPool3d} from './max_pool_3d';\nexport {maxPoolWithArgmax} from './max_pool_with_argmax';\nexport {maximum} from './maximum';\nexport {mean} from './mean';\nexport {min} from './min';\nexport {minimum} from './minimum';\nexport {mod} from './mod';\nexport {moments} from './moments';\nexport {mul} from './mul';\nexport {LSTMCellFunc, multiRNNCell} from './multi_rnn_cell';\nexport {multinomial} from './multinomial';\nexport {neg} from './neg';\nexport {notEqual} from './not_equal';\nexport {oneHot} from './one_hot';\nexport {ones} from './ones';\nexport {onesLike} from './ones_like';\nexport {outerProduct} from './outer_product';\nexport {pad} from './pad';\nexport {pad1d} from './pad1d';\nexport {pad2d} from './pad2d';\nexport {pad3d} from './pad3d';\nexport {pad4d} from './pad4d';\nexport {pool} from './pool';\nexport {pow} from './pow';\nexport {prelu} from './prelu';\nexport {print} from './print';\nexport {prod} from './prod';\nexport {rand} from './rand';\nexport {randomGamma} from './random_gamma';\nexport {randomNormal} from './random_normal';\nexport {randomUniform} from './random_uniform';\nexport {range} from './range';\nexport {real} from './real';\nexport {reciprocal} from './reciprocal';\nexport {relu} from './relu';\nexport {relu6} from './relu6';\nexport {reshape} from './reshape';\nexport {reverse} from './reverse';\nexport {reverse1d} from './reverse_1d';\nexport {reverse2d} from './reverse_2d';\nexport {reverse3d} from './reverse_3d';\nexport {reverse4d} from './reverse_4d';\nexport {round} from './round';\nexport {rsqrt} from './rsqrt';\nexport {scalar} from './scalar';\nexport {selu} from './selu';\nexport {separableConv2d} from './separable_conv2d';\nexport {setdiff1dAsync} from './setdiff1d_async';\nexport {sigmoid} from './sigmoid';\nexport {sign} from './sign';\nexport {sin} from './sin';\nexport {sinh} from './sinh';\nexport {slice} from './slice';\nexport {slice1d} from './slice1d';\nexport {slice2d} from './slice2d';\nexport {slice3d} from './slice3d';\nexport {slice4d} from './slice4d';\nexport {softmax} from './softmax';\nexport {softplus} from './softplus';\nexport {spaceToBatchND} from './space_to_batch_nd';\nexport {fft} from './spectral/fft';\nexport {ifft} from './spectral/ifft';\nexport {irfft} from './spectral/irfft';\nexport {rfft} from './spectral/rfft';\nexport {split} from './split';\nexport {sqrt} from './sqrt';\nexport {square} from './square';\nexport {squaredDifference} from './squared_difference';\nexport {squeeze} from './squeeze';\nexport {stack} from './stack';\nexport {step} from './step';\nexport {stridedSlice} from './strided_slice';\nexport {sub} from './sub';\nexport {sum} from './sum';\nexport {tan} from './tan';\nexport {tanh} from './tanh';\nexport {tensor} from './tensor';\nexport {tensor1d} from './tensor1d';\nexport {tensor2d} from './tensor2d';\nexport {tensor3d} from './tensor3d';\nexport {tensor4d} from './tensor4d';\nexport {tensor5d} from './tensor5d';\nexport {tensor6d} from './tensor6d';\nexport {tile} from './tile';\nexport {topk} from './topk';\nexport {truncatedNormal} from './truncated_normal';\nexport {unique} from './unique';\nexport {unsortedSegmentSum} from './unsorted_segment_sum';\nexport {unstack} from './unstack';\nexport {variable} from './variable';\nexport {where} from './where';\nexport {whereAsync} from './where_async';\nexport {zeros} from './zeros';\nexport {zerosLike} from './zeros_like';\n\nexport * from './boolean_mask';\nexport * from './compare';\nexport * from './binary_ops';\nexport * from './transpose';\nexport * from './norm';\nexport * from './moving_average';\nexport * from './scatter_nd';\nexport * from './sparse_to_dense';\nexport * from './gather_nd';\nexport * from './dropout';\nexport * from './signal_ops_util';\nexport * from './in_top_k';\n\nexport {op, OP_SCOPE_SUFFIX} from './operation';\n\nimport {rfft} from './spectral/rfft';\nimport {fft} from './spectral/fft';\nimport {ifft} from './spectral/ifft';\nimport {irfft} from './spectral/irfft';\nconst spectral = {\n  fft,\n  ifft,\n  rfft,\n  irfft\n};\n\nimport * as fused from './fused_ops';\n\nimport {hammingWindow} from './signal/hamming_window';\nimport {hannWindow} from './signal/hann_window';\nimport {frame} from './signal/frame';\nimport {stft} from './signal/stft';\nconst signal = {\n  hammingWindow,\n  hannWindow,\n  frame,\n  stft,\n};\n\n// Image Ops namespace\nimport {cropAndResize} from './image/crop_and_resize';\nimport {flipLeftRight} from './image/flip_left_right';\nimport {rotateWithOffset} from './image/rotate_with_offset';\nimport {nonMaxSuppression} from './image/non_max_suppression';\nimport {nonMaxSuppressionAsync} from './image/non_max_suppression_async';\nimport {nonMaxSuppressionWithScore} from './image/non_max_suppression_with_score';\nimport {nonMaxSuppressionWithScoreAsync} from './image/non_max_suppression_with_score_async';\nimport {nonMaxSuppressionPadded} from './image/non_max_suppression_padded';\nimport {nonMaxSuppressionPaddedAsync} from './image/non_max_suppression_padded_async';\nimport {resizeBilinear} from './image/resize_bilinear';\nimport {resizeNearestNeighbor} from './image/resize_nearest_neighbor';\nconst image = {\n  flipLeftRight,\n  resizeNearestNeighbor,\n  resizeBilinear,\n  rotateWithOffset,\n  cropAndResize,\n  nonMaxSuppression,\n  nonMaxSuppressionAsync,\n  nonMaxSuppressionWithScore,\n  nonMaxSuppressionWithScoreAsync,\n  nonMaxSuppressionPadded,\n  nonMaxSuppressionPaddedAsync\n};\n\n// linalg namespace\nimport {bandPart} from './linalg/band_part';\nimport {gramSchmidt} from './linalg/gram_schmidt';\nimport {qr} from './linalg/qr';\nconst linalg = {\n  bandPart,\n  gramSchmidt,\n  qr\n};\n\n// losses namespace;\nimport {absoluteDifference} from './losses/absolute_difference';\nimport {computeWeightedLoss} from './losses/compute_weighted_loss';\nimport {cosineDistance} from './losses/cosine_distance';\nimport {hingeLoss} from './losses/hinge_loss';\nimport {huberLoss} from './losses/huber_loss';\nimport {logLoss} from './losses/log_loss';\nimport {meanSquaredError} from './losses/mean_squared_error';\nimport {sigmoidCrossEntropy} from './losses/sigmoid_cross_entropy';\nimport {softmaxCrossEntropy} from './losses/softmax_cross_entropy';\nconst losses = {\n  absoluteDifference,\n  computeWeightedLoss,\n  cosineDistance,\n  hingeLoss,\n  huberLoss,\n  logLoss,\n  meanSquaredError,\n  sigmoidCrossEntropy,\n  softmaxCrossEntropy\n};\n\n// Second level exports.\nexport {image, linalg, losses, spectral, fused, signal};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {dispose} from '../globals';\nimport {variableGrads} from '../gradients';\nimport {scalar} from '../ops/ops';\nimport {Serializable} from '../serialization';\nimport {Scalar, Variable} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\n/**\n * A variable that belongs to an optimizer.\n *\n * The `originalName` field is required for keeping track of the canonical\n * name of the variable, which is usually the name of the model weight that\n * the variable is related to plus a suffix, e.g., 'dense1/kernel/momentum'.\n * The name of the `Variable` object itself cannot be used directly due to\n * possible deduplication: Every `Variable` must have a unique name but more\n * than one optimizer objects of the same type may be created for the same model\n * or the same `Variable`.\n */\nexport interface OptimizerVariable {\n  originalName: string;\n  variable: Variable;\n}\n\n/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */\nexport abstract class Optimizer extends Serializable {\n  protected iterations_: number;\n\n  /**\n   * Executes `f()` and minimizes the scalar output of `f()` by computing\n   * gradients of y with respect to the list of trainable variables provided by\n   * `varList`. If no list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to minimize.\n   * @param returnCost Whether to return the scalar cost value produced by\n   * executing `f()`.\n   * @param varList An optional list of variables to update. If specified, only\n   * the trainable variables in varList will be updated by minimize. Defaults to\n   * all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  minimize(f: () => Scalar, returnCost = false, varList?: Variable[]): Scalar\n      |null {\n    const {value, grads} = this.computeGradients(f, varList);\n\n    if (varList != null) {\n      const gradArray: NamedTensor[] =\n          varList.map(v => ({name: v.name, tensor: grads[v.name]}));\n      this.applyGradients(gradArray);\n    } else {\n      this.applyGradients(grads);\n    }\n\n    // Dispose gradients.\n    dispose(grads);\n\n    if (returnCost) {\n      return value;\n    } else {\n      value.dispose();\n      return null;\n    }\n  }\n\n  /**\n   * The number of iterations that this optimizer instance has been invoked for.\n   */\n  get iterations(): number {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return this.iterations_;\n  }\n\n  protected incrementIterations() {\n    this.iterations_ = this.iterations + 1;\n  }\n\n  /**\n   * Executes f() and computes the gradient of the scalar output of f() with\n   * respect to the list of trainable variables provided by `varList`. If no\n   * list is provided, it defaults to all trainable variables.\n   *\n   * @param f The function to execute and whose output to use for computing\n   * gradients with respect to variables.\n   * @param varList An optional list of variables to compute gradients with\n   * respect to. If specified, only the trainable variables in varList will have\n   * gradients computed with respect to. Defaults to all trainable variables.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  computeGradients(f: () => Scalar, varList?: Variable[]):\n      {value: Scalar, grads: NamedTensorMap} {\n    return variableGrads(f, varList);\n  }\n\n  /**\n   * Updates variables by using the computed gradients.\n   *\n   * @param variableGradients A mapping of variable name to its gradient value.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers'}\n   */\n  abstract applyGradients(variableGradients: NamedTensorMap|\n                          NamedTensor[]): void;\n\n  /**\n   * Dispose the variables (if any) owned by this optimizer instance.\n   */\n  dispose(): void {\n    if (this.iterations_ != null) {\n      dispose(this.iterations_);\n    }\n  }\n\n  async saveIterations(): Promise<NamedTensor> {\n    if (this.iterations_ == null) {\n      this.iterations_ = 0;\n    }\n    return {\n      name: 'iter',  // Named for Python compatibility.\n      // TODO(cais): Use 'int64' type when available.\n      tensor: scalar(this.iterations_, 'int32')\n    };\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for this optimizer yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error(\n        `setWeights() is not implemented for this optimizer class ` +\n        `${this.getClassName()}`);\n  }\n\n  /**\n   * Extract the first element of the weight values and set it\n   * as the iterations counter variable of this instance of optimizer.\n   *\n   * @param weightValues\n   * @returns Weight values with the first element consumed and excluded.\n   */\n  protected async extractIterations(weightValues: NamedTensor[]):\n      Promise<NamedTensor[]> {\n    this.iterations_ = (await weightValues[0].tensor.data())[0];\n    return weightValues.slice(1);\n  }\n}\n\nObject.defineProperty(Optimizer, Symbol.hasInstance, {\n  value: (instance: Optimizer) => {\n    return instance.minimize != null && instance.computeGradients != null &&\n        instance.applyGradients != null;\n  }\n});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/ops';\nimport {square} from '../ops/square';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdadeltaOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adadelta';  // Name matters for Python compatibility.\n  private accumulatedGrads: OptimizerVariable[] = [];\n  private accumulatedUpdates: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected rho: number,\n      protected epsilon: number = null) {\n    super();\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedGrads[i] == null) {\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accum_grad`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedUpdates[i] == null) {\n        this.accumulatedUpdates[i] = {\n          originalName: `${name}/accum_var`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n      const accumulatedUpdate = this.accumulatedUpdates[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad =\n            add(mul(accumulatedGrad, this.rho),\n                mul(square(gradient), 1 - this.rho));\n\n        const updates =\n            mul(div(sqrt(add(accumulatedUpdate, this.epsilon)),\n                    sqrt(add(accumulatedGrad, this.epsilon))),\n                gradient);\n\n        const newAccumulatedUpdate =\n            add(mul(accumulatedUpdate, this.rho),\n                mul(square(updates), 1 - this.rho));\n\n        accumulatedGrad.assign(newAccumulatedGrad);\n        accumulatedUpdate.assign(newAccumulatedUpdate);\n\n        const newValue = add(mul(updates, -this.learningRate), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedUpdates != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n      dispose(this.accumulatedUpdates.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedGrads, ...this.accumulatedUpdates];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedGrads =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedUpdates =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'rho': this.rho,\n      'epsilon': this.epsilon\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['rho'], config['epsilon']);\n  }\n}\nregisterClass(AdadeltaOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {fill} from '../ops/fill';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class AdagradOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adagrad';  // Note: Name matters for Python compatibility.\n\n  private accumulatedGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private initialAccumulatorValue = 0.1) {\n    super();\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulatedGrads[i] == null) {\n        const trainable = false;\n        this.accumulatedGrads[i] = {\n          originalName: `${name}/accumulator`,\n          variable: tidy(\n              () => fill(value.shape, this.initialAccumulatorValue)\n                        .variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedGrad = this.accumulatedGrads[i].variable;\n\n      tidy(() => {\n        const newAccumulatedGrad = add(accumulatedGrad, square(gradient));\n        accumulatedGrad.assign(newAccumulatedGrad);\n\n        const newValue = add(\n            mul(div(gradient,\n                    sqrt(add(newAccumulatedGrad, ENGINE.backend.epsilon()))),\n                -this.learningRate),\n            value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedGrads != null) {\n      dispose(this.accumulatedGrads.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulatedGrads.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulatedGrads = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'initialAccumulatorValue': this.initialAccumulatorValue,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate'], config['initialAccumulatorValue']);\n  }\n}\nregisterClass(AdagradOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adam';  // Note: Name matters for Python compatibility.\n  private accBeta1: Variable;\n  private accBeta2: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedSecondMoment: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null) {\n    super();\n    tidy(() => {\n      // accB* will be updated by batch.\n      this.accBeta1 = scalar(beta1).variable();\n      this.accBeta2 = scalar(beta2).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const oneMinusAccBeta2 = sub(1, this.accBeta2);\n\n      varNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n        if (this.accumulatedSecondMoment[i] == null) {\n          this.accumulatedSecondMoment[i] = {\n            originalName: `${name}/v`,\n            variable: tidy(() => zerosLike(value).variable(trainable))\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const secondMoment = this.accumulatedSecondMoment[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n        const newSecondMoment =\n            add(mul(secondMoment, this.beta2),\n                mul(square(gradient), 1 - this.beta2));\n\n        const biasCorrectedFirstMoment = div(newFirstMoment, oneMinusAccBeta1);\n        const biasCorrectedSecondMoment =\n            div(newSecondMoment, oneMinusAccBeta2);\n\n        firstMoment.assign(newFirstMoment);\n        secondMoment.assign(newSecondMoment);\n\n        const newValue =\n            add(mul(div(biasCorrectedFirstMoment,\n                        add(sqrt(biasCorrectedSecondMoment), this.epsilon)),\n                    -this.learningRate),\n                value);\n        value.assign(newValue);\n      });\n\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n      this.accBeta2.assign(mul(this.accBeta2, this.beta2));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.accBeta2.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedSecondMoment != null) {\n      dispose(this.accumulatedSecondMoment.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedFirstMoment, ...this.accumulatedSecondMoment];\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    tidy(() => {\n      this.accBeta1.assign(pow(this.beta1, this.iterations_ + 1));\n      this.accBeta2.assign(pow(this.beta2, this.iterations_ + 1));\n    });\n\n    const variableCount = weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedFirstMoment =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedSecondMoment =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon']);\n  }\n}\nregisterClass(AdamOptimizer);\n", "\uFEFF/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {abs} from '../ops/abs';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {maximum} from '../ops/maximum';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Variable} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\nexport class AdamaxOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'Adamax';  // Note: Name matters for Python compatbility.\n  private accBeta1: Variable;\n  private iteration: Variable;\n\n  private accumulatedFirstMoment: OptimizerVariable[] = [];\n  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected beta1: number,\n      protected beta2: number, protected epsilon: number = null,\n      protected decay = 0.0) {\n    super();\n\n    tidy(() => {\n      this.iteration = scalar(0).variable();\n      this.accBeta1 = scalar(beta1).variable();\n    });\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    tidy(() => {\n      const oneMinusAccBeta1 = sub(1, this.accBeta1);\n      const lr =\n          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));\n\n      variableNames.forEach((name, i) => {\n        const value = ENGINE.registeredVariables[name];\n        const trainable = false;\n        if (this.accumulatedFirstMoment[i] == null) {\n          this.accumulatedFirstMoment[i] = {\n            originalName: `${name}/m`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n        if (this.accumulatedWeightedInfNorm[i] == null) {\n          this.accumulatedWeightedInfNorm[i] = {\n            originalName: `${name}/v`,\n            variable: zerosLike(value).variable(trainable)\n          };\n        }\n\n        const gradient = Array.isArray(variableGradients) ?\n            variableGradients[i].tensor :\n            variableGradients[name];\n        if (gradient == null) {\n          return;\n        }\n\n        const firstMoment = this.accumulatedFirstMoment[i].variable;\n        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;\n\n        const newFirstMoment =\n            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));\n\n        const ut0 = mul(weightedInfNorm, this.beta2);\n        const ut1 = abs(gradient);\n\n        const newWeightedInfNorm = maximum(ut0, ut1);\n\n        firstMoment.assign(newFirstMoment);\n        weightedInfNorm.assign(newWeightedInfNorm);\n\n        const newValue =\n            add(mul(div(lr, oneMinusAccBeta1),\n                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),\n                value);\n\n        value.assign(newValue);\n      });\n\n      this.iteration.assign(add(this.iteration, 1));\n      this.accBeta1.assign(mul(this.accBeta1, this.beta1));\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.accBeta1.dispose();\n    this.iteration.dispose();\n\n    if (this.accumulatedFirstMoment != null) {\n      dispose(this.accumulatedFirstMoment.map(v => v.variable));\n    }\n    if (this.accumulatedWeightedInfNorm != null) {\n      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    throw new Error('getWeights() is not implemented for Adamax yet.');\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    throw new Error('setWeights() is not implemented for Adamax yet.');\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'beta1': this.beta1,\n      'beta2': this.beta2,\n      'epsilon': this.epsilon,\n      'decay': this.decay\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['beta1'], config['beta2'],\n        config['epsilon'], config['decay']);\n  }\n}\nregisterClass(AdamaxOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {keep, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar} from '../tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer} from './optimizer';\n\n/** @doclink Optimizer */\nexport class SGDOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'SGD';  // Note: Name matters for Python compatibility.\n  protected c: Scalar;\n\n  constructor(protected learningRate: number) {\n    super();\n    this.setLearningRate(learningRate);\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const varNames = Array.isArray(variableGradients) ?\n        variableGradients.map(v => v.name) :\n        Object.keys(variableGradients);\n    varNames.forEach((name, i) => {\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n      const value = ENGINE.registeredVariables[name];\n      tidy(() => {\n        const newValue = add(mul(this.c, gradient), value);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  /**\n   * Sets the learning rate of the optimizer.\n   */\n  setLearningRate(learningRate: number) {\n    this.learningRate = learningRate;\n    if (this.c != null) {\n      this.c.dispose();\n    }\n    this.c = keep(scalar(-learningRate));\n  }\n\n  dispose() {\n    this.c.dispose();\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    return [await this.saveIterations()];\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    if (weightValues.length !== 0) {\n      throw new Error('SGD optimizer does not have settable weights.');\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {'learningRate': this.learningRate};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(config['learningRate']);\n  }\n}\nregisterClass(SGDOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {Scalar, Tensor} from '../tensor';\nimport {NamedTensor, NamedVariableMap} from '../tensor_types';\n\nimport {OptimizerVariable} from './optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\n/** @doclink Optimizer */\nexport class MomentumOptimizer extends SGDOptimizer {\n  /** @nocollapse */\n  static className = 'Momentum';  // Name matters for Python compatibility.\n  private m: Scalar;\n  private accumulations: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, private momentum: number,\n      private useNesterov = false) {\n    super(learningRate);\n    this.m = scalar(this.momentum);\n  }\n\n  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      if (this.accumulations[i] == null) {\n        const trainable = false;\n        this.accumulations[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const accumulation = this.accumulations[i].variable;\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      tidy(() => {\n        let newValue: Tensor;\n        const newAccumulation = add(mul(this.m, accumulation), gradient);\n        if (this.useNesterov) {\n          newValue = add(\n              mul(this.c, add(gradient, mul(newAccumulation, this.m))), value);\n        } else {\n          newValue = add(mul(this.c, newAccumulation), value);\n        }\n        accumulation.assign(newAccumulation);\n        value.assign(newValue);\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    this.m.dispose();\n    if (this.accumulations != null) {\n      dispose(this.accumulations.map(v => v.variable));\n    }\n  }\n\n  /**\n   * Sets the momentum of the optimizer.\n   *\n   * @param momentum\n   */\n  setMomentum(momentum: number) {\n    this.momentum = momentum;\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    return [await this.saveIterations()].concat(this.accumulations.map(\n        v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const trainable = false;\n    this.accumulations = weightValues.map(\n        v => ({originalName: v.name, variable: v.tensor.variable(trainable)}));\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'momentum': this.momentum,\n      'useNesterov': this.useNesterov\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['momentum'], config['useNesterov']);\n  }\n}\nregisterClass(MomentumOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {dispose, tidy} from '../globals';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {zerosLike} from '../ops/zeros_like';\nimport {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\n\nimport {Optimizer, OptimizerVariable} from './optimizer';\n\n/** @doclink Optimizer */\nexport class RMSPropOptimizer extends Optimizer {\n  /** @nocollapse */\n  static className = 'RMSProp';  // Note: Name matters for Python compatibility.\n  private centered: boolean;\n\n  private accumulatedMeanSquares: OptimizerVariable[] = [];\n  private accumulatedMoments: OptimizerVariable[] = [];\n  private accumulatedMeanGrads: OptimizerVariable[] = [];\n\n  constructor(\n      protected learningRate: number, protected decay = 0.9,\n      protected momentum = 0.0, protected epsilon: number = null,\n      centered = false) {\n    super();\n\n    this.centered = centered;\n\n    if (epsilon == null) {\n      this.epsilon = ENGINE.backend.epsilon();\n    }\n    if (learningRate == null) {\n      throw new Error(`learningRate for RMSPropOptimizer must be defined.`);\n    }\n  }\n\n  applyGradients(variableGradients: NamedTensorMap|NamedTensor[]) {\n    const variableNames = Array.isArray(variableGradients) ?\n        variableGradients.map(item => item.name) :\n        Object.keys(variableGradients);\n\n    variableNames.forEach((name, i) => {\n      const value = ENGINE.registeredVariables[name];\n      const trainable = false;\n      if (this.accumulatedMeanSquares[i] == null) {\n        this.accumulatedMeanSquares[i] = {\n          originalName: `${name}/rms`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMoments[i] == null) {\n        this.accumulatedMoments[i] = {\n          originalName: `${name}/momentum`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n      if (this.accumulatedMeanGrads[i] == null && this.centered) {\n        this.accumulatedMeanGrads[i] = {\n          originalName: `${name}/mg`,\n          variable: tidy(() => zerosLike(value).variable(trainable))\n        };\n      }\n\n      const gradient = Array.isArray(variableGradients) ?\n          variableGradients[i].tensor :\n          variableGradients[name];\n      if (gradient == null) {\n        return;\n      }\n\n      const accumulatedMeanSquare = this.accumulatedMeanSquares[i].variable;\n      const accumulatedMoments = this.accumulatedMoments[i].variable;\n      tidy(() => {\n        const newAccumulatedMeanSquare =\n            add(mul(accumulatedMeanSquare, this.decay),\n                mul(square(gradient), 1 - this.decay));\n\n        if (this.centered) {\n          const accumulatedMeanGrad = this.accumulatedMeanGrads[i].variable;\n          // Centered gradient\n          const newAccumulatedMeanGrad =\n              add(mul(accumulatedMeanGrad, this.decay),\n                  mul(gradient, 1 - this.decay));\n\n          const gradContribution =\n              div(mul(gradient, this.learningRate),\n                  sqrt(\n                      sub(newAccumulatedMeanSquare,\n                          add(square(newAccumulatedMeanGrad), this.epsilon))));\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum), gradContribution);\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMeanGrad.assign(newAccumulatedMeanGrad);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        } else {\n          // Plain gradient\n          const newAccumulatedMeanSquare =\n              add(mul(accumulatedMeanSquare, this.decay),\n                  mul(square(gradient), 1 - this.decay));\n\n          const newAccumulatedMoments =\n              add(mul(accumulatedMoments, this.momentum),\n                  div(mul(gradient, this.learningRate),\n                      sqrt(add(newAccumulatedMeanSquare, this.epsilon))));\n\n          accumulatedMeanSquare.assign(newAccumulatedMeanSquare);\n          accumulatedMoments.assign(newAccumulatedMoments);\n\n          const newValue = sub(value, newAccumulatedMoments);\n          value.assign(newValue);\n        }\n      });\n    });\n    this.incrementIterations();\n  }\n\n  dispose(): void {\n    if (this.accumulatedMeanSquares != null) {\n      dispose(this.accumulatedMeanSquares.map(v => v.variable));\n    }\n    if (this.accumulatedMeanGrads != null && this.centered) {\n      dispose(this.accumulatedMeanGrads.map(v => v.variable));\n    }\n    if (this.accumulatedMoments != null) {\n      dispose(this.accumulatedMoments.map(v => v.variable));\n    }\n  }\n\n  async getWeights(): Promise<NamedTensor[]> {\n    // Order matters for Python compatibility.\n    const variables: OptimizerVariable[] =\n        [...this.accumulatedMeanSquares, ...this.accumulatedMoments];\n    if (this.centered) {\n      variables.push(...this.accumulatedMeanGrads);\n    }\n    return [await this.saveIterations()].concat(\n        variables.map(v => ({name: v.originalName, tensor: v.variable})));\n  }\n\n  async setWeights(weightValues: NamedTensor[]): Promise<void> {\n    weightValues = await this.extractIterations(weightValues);\n    const variableCount =\n        this.centered ? weightValues.length / 3 : weightValues.length / 2;\n    const trainable = false;\n    this.accumulatedMeanSquares =\n        weightValues.slice(0, variableCount).map(v => ({\n                                                   originalName: v.name,\n                                                   variable: v.tensor.variable(\n                                                       trainable)\n                                                 }));\n    this.accumulatedMoments =\n        weightValues.slice(variableCount, variableCount * 2)\n            .map(v => ({\n                   originalName: v.name,\n                   variable: v.tensor.variable(trainable)\n                 }));\n    if (this.centered) {\n      this.accumulatedMeanGrads =\n          weightValues.slice(variableCount * 2, variableCount * 3)\n              .map(v => ({\n                     originalName: v.name,\n                     variable: v.tensor.variable(trainable)\n                   }));\n    }\n  }\n\n  getConfig(): ConfigDict {\n    return {\n      'learningRate': this.learningRate,\n      'decay': this.decay,\n      'momentum': this.momentum,\n      'epsilon': this.epsilon,\n      'centered': this.centered\n    };\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends Serializable>(\n      cls: SerializableConstructor<T>, config: ConfigDict): T {\n    return new cls(\n        config['learningRate'], config['decay'], config['momentum'],\n        config['epsilon'], config['centered']);\n  }\n}\nregisterClass(RMSPropOptimizer);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AdadeltaOptimizer} from './adadelta_optimizer';\nimport {AdagradOptimizer} from './adagrad_optimizer';\nimport {AdamOptimizer} from './adam_optimizer';\nimport {AdamaxOptimizer} from './adamax_optimizer';\nimport {MomentumOptimizer} from './momentum_optimizer';\nimport {RMSPropOptimizer} from './rmsprop_optimizer';\nimport {SGDOptimizer} from './sgd_optimizer';\n\nexport class OptimizerConstructors {\n  /**\n   * Constructs a `tf.SGDOptimizer` that uses stochastic gradient descent.\n   *\n   * ```js\n   * // Fit a quadratic function by learning the coefficients a, b, c.\n   * const xs = tf.tensor1d([0, 1, 2, 3]);\n   * const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);\n   *\n   * const a = tf.scalar(Math.random()).variable();\n   * const b = tf.scalar(Math.random()).variable();\n   * const c = tf.scalar(Math.random()).variable();\n   *\n   * // y = a * x^2 + b * x + c.\n   * const f = x => a.mul(x.square()).add(b.mul(x)).add(c);\n   * const loss = (pred, label) => pred.sub(label).square().mean();\n   *\n   * const learningRate = 0.01;\n   * const optimizer = tf.train.sgd(learningRate);\n   *\n   * // Train the model.\n   * for (let i = 0; i < 10; i++) {\n   *   optimizer.minimize(() => loss(f(xs), ys));\n   * }\n   *\n   * // Make predictions.\n   * console.log(\n   *     `a: ${a.dataSync()}, b: ${b.dataSync()}, c: ${c.dataSync()}`);\n   * const preds = f(xs).dataSync();\n   * preds.forEach((pred, i) => {\n   *   console.log(`x: ${i}, pred: ${pred}`);\n   * });\n   * ```\n   *\n   * @param learningRate The learning rate to use for the SGD algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static sgd(learningRate: number): SGDOptimizer {\n    return new SGDOptimizer(learningRate);\n  }\n\n  /**\n   * Constructs a `tf.MomentumOptimizer` that uses momentum gradient\n   * descent.\n   *\n   * See\n   * [http://proceedings.mlr.press/v28/sutskever13.pdf](\n   * http://proceedings.mlr.press/v28/sutskever13.pdf)\n   *\n   * @param learningRate The learning rate to use for the Momentum gradient\n   * descent algorithm.\n   * @param momentum The momentum to use for the momentum gradient descent\n   * algorithm.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static momentum(learningRate: number, momentum: number, useNesterov = false):\n      MomentumOptimizer {\n    return new MomentumOptimizer(learningRate, momentum, useNesterov);\n  }\n\n  /**\n   * Constructs a `tf.RMSPropOptimizer` that uses RMSProp gradient\n   * descent. This implementation uses plain momentum and is not centered\n   * version of RMSProp.\n   *\n   * See\n   * [http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf](\n   * http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)\n   *\n   * @param learningRate The learning rate to use for the RMSProp gradient\n   * descent algorithm.\n   * @param decay The discounting factor for the history/coming gradient.\n   * @param momentum The momentum to use for the RMSProp gradient descent\n   * algorithm.\n   * @param epsilon Small value to avoid zero denominator.\n   * @param centered If true, gradients are normalized by the estimated\n   * variance of the gradient.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static rmsprop(\n      learningRate: number, decay = .9, momentum = 0.0, epsilon: number = null,\n      centered = false): RMSPropOptimizer {\n    return new RMSPropOptimizer(\n        learningRate, decay, momentum, epsilon, centered);\n  }\n\n  /**\n   * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adam gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adam(\n      learningRate = 0.001, beta1 = 0.9, beta2 = 0.999,\n      epsilon: number = null): AdamOptimizer {\n    return new AdamOptimizer(learningRate, beta1, beta2, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.\n   * See [https://arxiv.org/abs/1212.5701](https://arxiv.org/abs/1212.5701)\n   *\n   * @param learningRate The learning rate to use for the Adadelta gradient\n   * descent algorithm.\n   * @param rho The learning rate decay over each update.\n   * @param epsilon A constant epsilon used to better condition the grad\n   * update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adadelta(learningRate = .001, rho = .95, epsilon: number = null):\n      AdadeltaOptimizer {\n    return new AdadeltaOptimizer(learningRate, rho, epsilon);\n  }\n\n  /**\n   * Constructs a `tf.AdamaxOptimizer` that uses the Adamax algorithm.\n   * See [https://arxiv.org/abs/1412.6980](https://arxiv.org/abs/1412.6980)\n   *\n   * @param learningRate The learning rate to use for the Adamax gradient\n   * descent algorithm.\n   * @param beta1 The exponential decay rate for the 1st moment estimates.\n   * @param beta2 The exponential decay rate for the 2nd moment estimates.\n   * @param epsilon A small constant for numerical stability.\n   * @param decay The learning rate decay over each update.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adamax(\n      learningRate = 0.002, beta1 = 0.9, beta2 = 0.999, epsilon: number = null,\n      decay = 0.0): AdamaxOptimizer {\n    return new AdamaxOptimizer(learningRate, beta1, beta2, epsilon, decay);\n  }\n\n  /**\n   * Constructs a `tf.AdagradOptimizer` that uses the Adagrad algorithm.\n   * See\n   * [http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf](\n   * http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\n   * or\n   * [http://ruder.io/optimizing-gradient-descent/index.html#adagrad](\n   * http://ruder.io/optimizing-gradient-descent/index.html#adagrad)\n   *\n   * @param learningRate The learning rate to use for the Adagrad gradient\n   * descent algorithm.\n   * @param initialAccumulatorValue Starting value for the accumulators, must be\n   * positive.\n   *\n   * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}\n   */\n  static adagrad(learningRate: number, initialAccumulatorValue = 0.1):\n      AdagradOptimizer {\n    return new AdagradOptimizer(learningRate, initialAccumulatorValue);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// So typings can propagate.\nimport {AdadeltaOptimizer} from './optimizers/adadelta_optimizer';\nimport {AdagradOptimizer} from './optimizers/adagrad_optimizer';\nimport {AdamOptimizer} from './optimizers/adam_optimizer';\nimport {AdamaxOptimizer} from './optimizers/adamax_optimizer';\nimport {MomentumOptimizer} from './optimizers/momentum_optimizer';\nimport {OptimizerConstructors} from './optimizers/optimizer_constructors';\nimport {RMSPropOptimizer} from './optimizers/rmsprop_optimizer';\nimport {SGDOptimizer} from './optimizers/sgd_optimizer';\n\n// tslint:disable-next-line:no-unused-expression\n[MomentumOptimizer, SGDOptimizer, AdadeltaOptimizer, AdagradOptimizer,\n RMSPropOptimizer, AdamaxOptimizer, AdamOptimizer];\n\nexport const train = {\n  sgd: OptimizerConstructors.sgd,\n  momentum: OptimizerConstructors.momentum,\n  adadelta: OptimizerConstructors.adadelta,\n  adagrad: OptimizerConstructors.adagrad,\n  rmsprop: OptimizerConstructors.rmsprop,\n  adamax: OptimizerConstructors.adamax,\n  adam: OptimizerConstructors.adam\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst delayCallback: Function = (() => {\n  if (typeof requestAnimationFrame !== 'undefined') {\n    return requestAnimationFrame;\n  } else if (typeof setImmediate !== 'undefined') {\n    return setImmediate;\n  }\n  return (f: Function) => f();  // no delays\n})();\n\n/**\n * Returns a promise that resolve when a requestAnimationFrame has completed.\n *\n * On Node.js this uses setImmediate instead of requestAnimationFrame.\n *\n * This is simply a sugar method so that users can do the following:\n * `await tf.nextFrame();`\n *\n * @doc {heading: 'Performance', subheading: 'Timing'}\n */\nfunction nextFrame(): Promise<void> {\n  return new Promise<void>(resolve => delayCallback(() => resolve()));\n}\n\nexport {nextFrame};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Returns the image center in pixels.\nexport function getImageCenter(\n    center: number|[number, number], imageHeight: number,\n    imageWidth: number): [number, number] {\n  const centerX =\n      imageWidth * (typeof center === 'number' ? center : center[0]);\n  const centerY =\n      imageHeight * (typeof center === 'number' ? center : center[1]);\n  return [centerX, centerY];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Gets the new shape of the input Tensor after it's been reshaped\n * to:\n * [blockShape[0], ..., blockShape[M-1], batch / prod(blockShape),\n * inputShape[1], ..., inputShape[N-1]]\n *\n * See step 1: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshaped(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  let reshaped: number[] = [];\n  if (batchToSpace) {\n    reshaped = reshaped.concat(blockShape.slice(0));\n    reshaped.push(inputShape[0] / prod);\n    reshaped = reshaped.concat(inputShape.slice(1));\n  } else {\n    reshaped = reshaped.concat(inputShape[0]);\n    const spatialLength = blockShape.length;\n    for (let i = 0; i < spatialLength; ++i) {\n      reshaped =\n          reshaped.concat([inputShape[i + 1] / blockShape[i], blockShape[i]]);\n    }\n    reshaped = reshaped.concat(inputShape.slice(spatialLength + 1));\n  }\n  return reshaped;\n}\n\n/**\n * Gets the permutation that will transpose the dimensions of the\n * reshaped tensor to shape:\n *\n * [batch / prod(block_shape),inputShape[1], blockShape[0], ...,\n * inputShape[M], blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * see step 2: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getPermuted(\n    reshapedRank: number, blockShapeRank: number,\n    batchToSpace = true): number[] {\n  const permuted = [];\n  if (batchToSpace) {\n    permuted.push(blockShapeRank);\n    for (let i = blockShapeRank + 1; i < reshapedRank; ++i) {\n      if (i <= 2 * blockShapeRank) {\n        permuted.push(i);\n        permuted.push(i - (blockShapeRank + 1));\n      } else {\n        permuted.push(i);\n      }\n    }\n  } else {\n    const permutedBeforeBatch = [];\n    const permutedAfterBatch = [];\n    for (let i = 1; i < reshapedRank; ++i) {\n      if (i >= blockShapeRank * 2 + 1 || i % 2 === 1) {\n        permutedAfterBatch.push(i);\n      } else {\n        permutedBeforeBatch.push(i);\n      }\n    }\n    permuted.push(...permutedBeforeBatch);\n    permuted.push(0);\n    permuted.push(...permutedAfterBatch);\n  }\n  return permuted;\n}\n\n/**\n * Gets the shape of the reshaped and permuted input Tensor before any cropping\n * is applied.  The new shape will be:\n *\n * [batch / prod(blockShape),inputShape[1] * blockShape[0], ...,\n * inputShape[M] * blockShape[M-1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 3: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getReshapedPermuted(\n    inputShape: number[], blockShape: number[], prod: number,\n    batchToSpace = true): number[] {\n  const reshapedPermuted = [];\n\n  if (batchToSpace) {\n    reshapedPermuted.push(inputShape[0] / prod);\n  } else {\n    reshapedPermuted.push(inputShape[0] * prod);\n  }\n\n  for (let i = 1; i < inputShape.length; ++i) {\n    if (i <= blockShape.length) {\n      if (batchToSpace) {\n        reshapedPermuted.push(blockShape[i - 1] * inputShape[i]);\n      } else {\n        reshapedPermuted.push(inputShape[i] / blockShape[i - 1]);\n      }\n    } else {\n      reshapedPermuted.push(inputShape[i]);\n    }\n  }\n\n  return reshapedPermuted;\n}\n\n/**\n * Converts the crops argument into the beginning coordinates of a slice\n * operation.\n */\nexport function getSliceBeginCoords(\n    crops: number[][], blockShape: number): number[] {\n  const sliceBeginCoords = [0];\n  for (let i = 0; i < blockShape; ++i) {\n    sliceBeginCoords.push(crops[i][0]);\n  }\n  return sliceBeginCoords;\n}\n\n/**\n * Converts the crops argument into the size of a slice operation.  When\n * combined with getSliceBeginCoords this function allows the reshaped and\n * permuted Tensor to be cropped to its final output shape of:\n *\n * inputShape[1] * blockShape[0] - crops[0,0] - crops[0,1], ...,\n * inputShape[M] * blockShape[M-1] -crops[M-1,0] -\n * crops[M-1,1],inputShape[M+1], ..., inputShape[N-1]]\n *\n * See step 4: https://www.tensorflow.org/api_docs/python/tf/batch_to_space_nd\n */\nexport function getSliceSize(\n    uncroppedShape: number[], crops: number[][], blockShape: number): number[] {\n  const sliceSize = uncroppedShape.slice(0, 1);\n  for (let i = 0; i < blockShape; ++i) {\n    sliceSize.push(uncroppedShape[i + 1] - crops[i][0] - crops[i][1]);\n  }\n\n  return sliceSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const SELU_SCALEALPHA = 1.7580993408473768599402175208123;\nexport const SELU_SCALE = 1.0507009873554804934193349852946;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const ERF_P = 0.3275911;\nexport const ERF_A1 = 0.254829592;\nexport const ERF_A2 = -0.284496736;\nexport const ERF_A3 = 1.421413741;\nexport const ERF_A4 = -1.453152027;\nexport const ERF_A5 = 1.061405429;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\n\nexport function warn(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.warn(...msg);\n  }\n}\n\nexport function log(...msg: Array<{}>): void {\n  if (!env().getBool('IS_TEST')) {\n    console.log(...msg);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TypedArray} from '../types';\n/**\n * Merges real and imaginary Float32Arrays into a single complex Float32Array.\n *\n * The memory layout is interleaved as follows:\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n * complex: [r0, i0, r1, i1, r2, i2]\n *\n * This is the inverse of splitRealAndImagArrays.\n *\n * @param real The real values of the complex tensor values.\n * @param imag The imag values of the complex tensor values.\n * @returns A complex tensor as a Float32Array with merged values.\n */\nexport function mergeRealAndImagArrays(\n    real: Float32Array, imag: Float32Array): Float32Array {\n  if (real.length !== imag.length) {\n    throw new Error(\n        `Cannot merge real and imag arrays of different lengths. real:` +\n        `${real.length}, imag: ${imag.length}.`);\n  }\n  const result = new Float32Array(real.length * 2);\n  for (let i = 0; i < result.length; i += 2) {\n    result[i] = real[i / 2];\n    result[i + 1] = imag[i / 2];\n  }\n  return result;\n}\n\n/**\n * Splits a complex Float32Array into real and imag parts.\n *\n * The memory layout is interleaved as follows:\n * complex: [r0, i0, r1, i1, r2, i2]\n * real: [r0, r1, r2]\n * imag: [i0, i1, i2]\n *\n * This is the inverse of mergeRealAndImagArrays.\n *\n * @param complex The complex tensor values.\n * @returns An object with real and imag Float32Array components of the complex\n *     tensor.\n */\nexport function splitRealAndImagArrays(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(complex.length / 2);\n  const imag = new Float32Array(complex.length / 2);\n  for (let i = 0; i < complex.length; i += 2) {\n    real[i / 2] = complex[i];\n    imag[i / 2] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts even indexed complex values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithEvenIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.ceil(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 0; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Extracts odd indexed comple values in the given array.\n * @param complex The complex tensor values\n */\nexport function complexWithOddIndex(complex: Float32Array):\n    {real: Float32Array, imag: Float32Array} {\n  const len = Math.floor(complex.length / 4);\n  const real = new Float32Array(len);\n  const imag = new Float32Array(len);\n  for (let i = 2; i < complex.length; i += 4) {\n    real[Math.floor(i / 4)] = complex[i];\n    imag[Math.floor(i / 4)] = complex[i + 1];\n  }\n  return {real, imag};\n}\n\n/**\n * Get the map representing a complex value in the given array.\n * @param complex The complex tensor values.\n * @param index An index of the target complex value.\n */\nexport function getComplexWithIndex(\n    complex: Float32Array, index: number): {real: number, imag: number} {\n  const real = complex[index * 2];\n  const imag = complex[index * 2 + 1];\n  return {real, imag};\n}\n\n/**\n * Insert a given complex value into the TypedArray.\n * @param data The array in which the complex value is inserted.\n * @param c The complex value to be inserted.\n * @param index An index of the target complex value.\n */\nexport function assignToTypedArray(\n    data: TypedArray, real: number, imag: number, index: number) {\n  data[index * 2] = real;\n  data[index * 2 + 1] = imag;\n}\n\n/**\n * Make the list of exponent terms used by FFT.\n */\nexport function exponents(\n    n: number, inverse: boolean): {real: Float32Array, imag: Float32Array} {\n  const real = new Float32Array(n / 2);\n  const imag = new Float32Array(n / 2);\n  for (let i = 0; i < Math.ceil(n / 2); i++) {\n    const x = (inverse ? 2 : -2) * Math.PI * (i / n);\n    real[i] = Math.cos(x);\n    imag[i] = Math.sin(x);\n  }\n  return {real, imag};\n}\n\n/**\n * Make the exponent term used by FFT.\n */\nexport function exponent(\n    k: number, n: number, inverse: boolean): {real: number, imag: number} {\n  const x = (inverse ? 2 : -2) * Math.PI * (k / n);\n  const real = Math.cos(x);\n  const imag = Math.sin(x);\n  return {real, imag};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {cast} from '../ops/cast';\nimport {scalar} from '../ops/scalar';\nimport {tensor1d} from '../ops/tensor1d';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\nimport {Rank} from '../types';\nimport {DataType, ShapeMap} from '../types';\nimport {hasEncodingLoss, makeZerosTypedArray} from '../util';\n\nimport {KernelBackend} from './backend';\n\n// Utilities needed by backend consumers of tf-core.\nexport * from '../ops/axis_util';\nexport * from '../ops/broadcast_util';\nexport * from '../ops/concat_util';\nexport * from '../ops/conv_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/fused_types';\nexport * from '../ops/reduce_util';\n\nimport * as slice_util from '../ops/slice_util';\nexport {slice_util};\n\nexport {BackendValues, TypedArray, upcastType, PixelData} from '../types';\nexport {MemoryInfo, TimingInfo} from '../engine';\nexport * from '../ops/rotate_util';\nexport * from '../ops/array_ops_util';\nexport * from '../ops/gather_nd_util';\nexport * from '../ops/scatter_nd_util';\nexport * from '../ops/selu_util';\nexport * from '../ops/fused_util';\nexport * from '../ops/erf_util';\nexport * from '../log';\nexport * from '../backends/complex_util';\nexport * from '../ops/split_util';\n\nimport * as segment_util from '../ops/segment_util';\nexport {segment_util};\n\nexport function castTensor<T extends Tensor>(\n    x: T, dtype: DataType, backend: KernelBackend): T {\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return x.clone();\n    }\n    const zerosTensor = zeros(x.shape);\n    const floatX = cast(x, 'float32');\n    const result = backend.complex(floatX, zerosTensor);\n    zerosTensor.dispose();\n    floatX.dispose();\n    return result as T;\n  }\n\n  if (!hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    return ENGINE.makeTensorFromDataId(x.dataId, x.shape, dtype) as T;\n  }\n  if (x.dtype === 'complex64') {\n    const real = backend.real(x);\n    const result = cast(real, dtype);\n    real.dispose();\n    return result;\n  }\n  if (dtype === 'int32') {\n    return backend.int(x);\n  } else if (dtype === 'bool') {\n    const zero = scalar(0, x.dtype);\n    const result = backend.notEqual(x, zero) as T;\n    zero.dispose();\n    return result;\n  } else {\n    throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n  }\n}\n\nexport function reshapeTensor<T extends Tensor, R extends Rank>(\n    x: T, shape: ShapeMap[R]): Tensor<R> {\n  return ENGINE.makeTensorFromDataId(x.dataId, shape, x.dtype) as Tensor<R>;\n}\n\nexport function linspaceImpl(start: number, stop: number, num: number) {\n  const step = (stop - start) / (num - 1);\n\n  const values = makeZerosTypedArray(num, 'float32');\n  values[0] = start;\n  for (let i = 1; i < values.length; i++) {\n    values[i] = values[i - 1] + step;\n  }\n\n  return tensor1d(values, 'float32');\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\n// TODO(annxingyuan): Use this helper in WASM Split kernel once intermediate\n// kernels have been modularized in WebGL and CPU\n// https://github.com/tensorflow/tfjs/issues/2822.\n/** Shared implementation of the split kernel across WebGL and CPU. */\nexport function split<T extends Tensor>(\n    x: T, sizeSplits: number[], axis: number): T[] {\n  const begin = new Array(x.rank).fill(0);\n  const size = x.shape.slice();\n  return sizeSplits.map(s => {\n    const sliceSize = [...size];\n    sliceSize[axis] = s;\n    const sliceT = slice(x, begin, sliceSize);\n    begin[axis] += s;\n    return sliceT;\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * An implementation of the tile kernel shared between webgl and cpu for string\n * tensors only.\n */\n\nimport {buffer} from '../ops/buffer';\nimport {Tensor, TensorBuffer} from '../tensor';\nimport {DataType, Rank} from '../types';\n\nexport function tile<R extends Rank>(\n    xBuf: TensorBuffer<R, DataType>, reps: number[]): Tensor<R> {\n  const newShape: number[] = new Array(xBuf.rank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = xBuf.shape[i] * reps[i];\n  }\n  const result = buffer(newShape, xBuf.dtype);\n  for (let i = 0; i < result.values.length; ++i) {\n    const newLoc = result.indexToLoc(i);\n\n    const originalLoc: number[] = new Array(xBuf.rank);\n    for (let j = 0; j < originalLoc.length; j++) {\n      originalLoc[j] = newLoc[j] % xBuf.shape[j];\n    }\n\n    const originalIndex = xBuf.locToIndex(originalLoc);\n\n    result.values[i] = xBuf.values[originalIndex];\n  }\n  return result.toTensor() as Tensor<R>;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** An implementation of the TopK kernel shared between webgl and cpu. */\n\nimport {tensor} from '../ops/tensor';\nimport {Tensor} from '../tensor';\nimport {NumericDataType, TypedArray} from '../types';\nimport {getTypedArrayFromDType} from '../util';\n\nexport function topkImpl<T extends Tensor>(\n    x: TypedArray, xShape: number[], xDtype: NumericDataType, k: number,\n    sorted: boolean): [T, T] {\n  // Reshape into a 2d tensor [batch, lastDim] and compute topk along lastDim.\n  const lastDim = xShape[xShape.length - 1];\n  const [batch, size] = [x.length / lastDim, lastDim];\n  const allTopKVals = getTypedArrayFromDType(xDtype, batch * k);\n  const allTopKIndices = getTypedArrayFromDType('int32', batch * k);\n\n  for (let b = 0; b < batch; b++) {\n    const offset = b * size;\n    const vals = x.subarray(offset, offset + size);\n    const valAndInd: Array<{value: number, index: number}> = [];\n    for (let i = 0; i < vals.length; i++) {\n      valAndInd.push({value: vals[i], index: i});\n    }\n    valAndInd.sort((a, b) => b.value - a.value);\n\n    const outOffset = b * k;\n    const topKVals = allTopKVals.subarray(outOffset, outOffset + k);\n    const topKIndices = allTopKIndices.subarray(outOffset, outOffset + k);\n    for (let i = 0; i < k; i++) {\n      topKVals[i] = valAndInd[i].value;\n      topKIndices[i] = valAndInd[i].index;\n    }\n  }\n  // Reshape back to the original input shape, except that the last\n  // dimension is k.\n  const outputShape = xShape.slice();\n  outputShape[outputShape.length - 1] = k;\n  return [\n    tensor(allTopKVals, outputShape, xDtype) as T,\n    tensor(allTopKIndices, outputShape, 'int32') as T\n  ];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport {nonMaxSuppressionV3Impl, nonMaxSuppressionV4Impl, nonMaxSuppressionV5Impl} from './non_max_suppression_impl';\nexport {split} from './split_shared';\nexport {tile} from './tile_impl';\nexport {topkImpl} from './topk_impl';\nexport {whereImpl} from './where_impl';\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const absGradConfig: GradConfig = {\n  kernelName: Abs,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, step(cast(x, 'float32'), -1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acosGradConfig: GradConfig = {\n  kernelName: Acos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = square(cast(x, 'float32'));\n        const b = sqrt(sub(scalar(1), a));\n        return neg(div(dy, b));\n      }\n\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const acoshGradConfig: GradConfig = {\n  kernelName: Acosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(sub(square(cast(x, 'float32')), 1));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Add} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const addGradConfig: GradConfig = {\n  kernelName: Add,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AddN} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const addNGradConfig: GradConfig = {\n  kernelName: AddN,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const ders: {[key: string]: () => Tensor} = {};\n    saved.forEach((_, i) => {\n      ders[i] = () => dy.clone();\n    });\n    return ders;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMax} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMaxGradConfig: GradConfig = {\n  kernelName: ArgMax,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ArgMin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const argMinGradConfig: GradConfig = {\n  kernelName: ArgMin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => zerosLike(x)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const asinGradConfig: GradConfig = {\n  kernelName: Asin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, sqrt(sub(scalar(1), square(cast(x, 'float32')))))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {scalar} from '../ops/scalar';\nimport {sqrt} from '../ops/sqrt';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const asinhGradConfig: GradConfig = {\n  kernelName: Asinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {\n      x: () => {\n        const a = sqrt(add(scalar(1), square(cast(x, 'float32'))));\n        return div(dy, a);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const atan2GradConfig: GradConfig = {\n  kernelName: Atan2,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const d = add(square(a), square(b));\n      let res = mul(dy, div(b, d));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      const d = add(square(a), square(b));\n      let res = neg(mul(dy, div(a, d)));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, b.shape);\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const atanGradConfig: GradConfig = {\n  kernelName: Atan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, add(square(cast(x, 'float32')), 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {scalar} from '../ops/scalar';\nimport {Tensor} from '../tensor';\n\nexport const atanhGradConfig: GradConfig = {\n  kernelName: Atanh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, sub(scalar(1), square(cast(x, 'float32'))))};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPool3DBackprop, AvgPool3DBackpropAttrs, AvgPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d avg pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0. The dilation\n *     rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction avgPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in avgPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in avgPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.avgPool3dBackprop(dy5D, input5D, convInfo);\n  };\n\n  const inputs: AvgPool3DBackpropInputs = {dy: dy5D, input: input5D};\n\n  const attrs: AvgPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      AvgPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const avgPool3dBackprop = op({avgPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool3D, AvgPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPool3dBackprop} from '../ops/avg_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const avgPool3DGradConfig: GradConfig = {\n  kernelName: AvgPool3D,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as AvgPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => avgPool3dBackprop(\n          dy as Tensor5D, x, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of an 2D avg pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The input image, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction avgPoolBackprop_<T extends Tensor3D|Tensor4D>(\n    dy: T|TensorLike, input: T|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number): T {\n  const $dy = convertToTensor(dy, 'dy', 'avgPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'avgPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy (${\n          $dy.rank})`);\n\n  let input4D = $input as Tensor4D;\n  let dy4D = $dy as Tensor4D;\n  let reshapedTo4D = false;\n\n  if ($input.rank === 3) {\n    reshapedTo4D = true;\n    input4D =\n        reshape($input, [1, $input.shape[0], $input.shape[1], $input.shape[2]]);\n    dy4D = reshape($dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2]]);\n  }\n\n  util.assert(\n      dy4D.rank === 4,\n      () => `Error in avgPoolBackprop: dy must be rank 4 but got rank ` +\n          `${dy4D.rank}.`);\n  util.assert(\n      input4D.rank === 4,\n      () => `Error in avgPoolBackprop: input must be rank 4 but got rank ` +\n          `${input4D.rank}.`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        input4D.shape, filterSize, strides, 1 /* dilations */, pad);\n\n    return backend.avgPoolBackprop(dy4D, input4D, convInfo);\n  };\n\n  const inputs: AvgPoolBackpropInputs = {dy: dy4D, input: input4D};\n\n  const attrs: AvgPoolBackpropAttrs = {filterSize, strides, pad};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null, AvgPoolBackprop,\n      attrs as {} as NamedAttrMap);\n\n  if (reshapedTo4D) {\n    return reshape(res, [res.shape[1], res.shape[2], res.shape[3]]) as T;\n  }\n  return res as T;\n}\n\nexport const avgPoolBackprop = op({avgPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {AvgPool, AvgPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {avgPoolBackprop} from '../ops/avg_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const avgPoolGradConfig: GradConfig = {\n  kernelName: AvgPool,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved as [Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as AvgPoolAttrs;\n    return {\n      x: () => avgPoolBackprop(dy as Tensor4D, x, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchMatMul, BatchMatMulAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {matMul} from '../ops/mat_mul';\nimport {Tensor, Tensor3D} from '../tensor';\n\nexport const batchMatMulGradConfig: GradConfig = {\n  kernelName: BatchMatMul,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor3D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [a, b] = saved as Tensor3D[];\n\n    const {transposeA, transposeB} = attrs as {} as BatchMatMulAttrs;\n\n    if (!transposeA && !transposeB) {\n      return {\n        a: () => matMul(dy, b, false, true),\n        b: () => matMul(a, dy, true, false)\n      };\n    } else if (!transposeA && transposeB) {\n      return {\n        a: () => matMul(dy, b, false, false),\n        b: () => matMul(dy, a, true, false)\n      };\n    } else if (transposeA && !transposeB) {\n      return {\n        a: () => matMul(b, dy, false, true),\n        b: () => matMul(a, dy, false, false)\n      };\n    } else {\n      return {\n        a: () => matMul(b, dy, true, true),\n        b: () => matMul(dy, a, true, true)\n      };\n    }\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BatchToSpaceND, BatchToSpaceNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {spaceToBatchND} from '../ops/space_to_batch_nd';\nimport {Tensor} from '../tensor';\n\nexport const batchToSpaceNDGradConfig: GradConfig = {\n  kernelName: BatchToSpaceND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, crops} = attrs as {} as BatchToSpaceNDAttrs;\n    return {x: () => spaceToBatchND(dy, blockShape, crops)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BroadcastTo, BroadCastToAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const broadcastToGradConfig: GradConfig = {\n  kernelName: BroadcastTo,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const broadCastToAttrs: BroadCastToAttrs =\n        attrs as unknown as BroadCastToAttrs;\n\n    const inputShape = broadCastToAttrs.inputShape;\n    const outputShape = broadCastToAttrs.shape;\n\n    const reps: number[] = Array.from(outputShape);\n    for (let i = inputShape.length - 1; i >= 0; i--) {\n      if (inputShape[i] === outputShape[i]) {\n        reps[i] = 1;\n      } else if (inputShape[i] !== 1) {\n        throw new Error(`broadcastTo(): [${\n            inputShape}] cannot be broadcast to [${outputShape}].`);\n      }\n    }\n    const axes: number[] = [];\n    for (let i = 0; i < reps.length; i++) {\n      if (reps[i] > 1) {\n        axes.push(i);\n      }\n    }\n\n    return {x: () => sum(dy, axes, true /* keepDims */)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cast} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\n\nexport const castGradConfig: GradConfig = {\n  kernelName: Cast,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => dy.clone()};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const ceilGradConfig: GradConfig = {\n  kernelName: Ceil,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {lessEqual} from '../ops/less_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const clipByValueGradConfig: GradConfig = {\n  kernelName: ClipByValue,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {clipValueMin, clipValueMax} = attrs as {} as ClipByValueAttrs;\n    return {\n      x: () => where(\n          logicalAnd(greaterEqual(x, clipValueMin), lessEqual(x, clipValueMax)),\n          dy, zerosLike(dy)),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Concat, ConcatAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {split} from '../ops/split';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const concatGradConfig: GradConfig = {\n  kernelName: Concat,\n  saveAllInputs: true,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const shapes = saved.map(t => t.shape);\n    const {axis} = attrs as {} as ConcatAttrs;\n    const $axis = parseAxisParam(axis, saved[0].shape)[0];\n    const sizeSplits = shapes.map(s => s[$axis]);\n    const derTensors = split(dy, sizeSplits, $axis);\n    return derTensors.map(t => () => t) as {};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2D, Conv2DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {conv2DBackpropInput} from '../ops/conv2d_backprop_input';\nimport * as conv_util from '../ops/conv_util';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv2DGradConfig: GradConfig = {\n  kernelName: Conv2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x4D, $filter] = saved as [Tensor4D, Tensor4D];\n    const {dilations, strides, pad, dataFormat} = attrs as {} as Conv2DAttrs;\n\n    util.assert(\n        conv_util.tupleValuesAreOne(dilations),\n        () => 'Error in gradient of conv2D: dilation rates greater than 1 ' +\n            `are not yet supported in gradients. Got dilations '${dilations}'`);\n\n    return {\n      x: () =>\n          conv2DBackpropInput(x4D.shape, dy, $filter, strides, pad, dataFormat),\n      filter: () =>\n          conv2DBackpropFilter(x4D, dy, $filter.shape, strides, pad, dataFormat)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv2DBackpropInput, Conv2DBackpropInputAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv2d} from '../ops/conv2d';\nimport {conv2DBackpropFilter} from '../ops/conv2d_backprop_filter';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const conv2DBackpropInputGradConfig: GradConfig = {\n  kernelName: Conv2DBackpropInput,\n  inputsToSave: ['dy', 'filter'],\n  gradFunc: (ddx: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [dy, filter] = saved as [Tensor4D, Tensor4D];\n\n    const {strides, pad, dataFormat, dimRoundingMode} =\n        attrs as {} as Conv2DBackpropInputAttrs;\n\n    return {\n      dy: () => conv2d(\n          ddx, filter, strides, pad, dataFormat, 1 /* dilations */,\n          dimRoundingMode),\n      filter: () => conv2DBackpropFilter(\n          ddx, dy, filter.shape, strides, pad, dataFormat, dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Conv3DBackpropFilterAttrs, Conv3DBackpropFilterInputs, Conv3DBackpropFilterV2} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the derivative of the filter of a 3D convolution.\n *\n * @param x The input tensor, of rank 5 or rank 4 of shape\n *     [batch, depth, height, width, inChannels]. If rank 4, batch of 1 is\n *     assumed.\n * @param dy The dy image, of rank 5 or rank 4, of shape\n *     [batch, depth, height, width, outDepth]. If rank 4, batch of 1 is\n *     assumed.\n * @param filterShape The shape of the filter, length 5,\n *     [filterDepth, filterHeight, filterWidth, inDepth, outDepth].\n * @param strides The strides of the convolution: [strideDepth, strideHeight,\n * strideWidth].\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n */\nfunction conv3DBackpropFilter_<T extends Tensor4D|Tensor5D>(\n    x: T, dy: T, filterShape: [number, number, number, number, number],\n    strides: [number, number, number]|number, pad: 'valid'|'same'): Tensor5D {\n  let x5D = x as Tensor5D;\n  if (x.rank === 4) {\n    x5D = reshape(x, [1, x.shape[0], x.shape[1], x.shape[2], x.shape[3]]);\n  }\n  let dy5D = dy as Tensor5D;\n  if (dy5D.rank === 4) {\n    dy5D = reshape(dy, [1, dy.shape[0], dy.shape[1], dy.shape[2], dy.shape[3]]);\n  }\n  util.assert(\n      x5D.rank === 5,\n      () => `Error in conv3dDerFilter: input must be rank 5, but got shape ` +\n          `${x5D.shape}.`);\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in conv3dDerFilter: dy must be rank 5, but got shape ` +\n          `${dy5D.shape}.`);\n  util.assert(\n      filterShape.length === 5,\n      () => `Error in conv3dDerFilter: filterShape must be length 5, but got ` +\n          `${filterShape}.`);\n  util.assert(\n      x5D.shape[4] === filterShape[3],\n      () => `Error in conv3dDerFilter: depth of input ${x5D.shape[4]}) must ` +\n          `match input depth in filter (${filterShape[3]}.`);\n  util.assert(\n      dy5D.shape[4] === filterShape[4],\n      () => `Error in conv3dDerFilter: depth of dy (${dy5D.shape[4]}) must ` +\n          `match output depth for filter (${filterShape[4]}).`);\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const dilations = 1;\n\n    const convInfo = conv_util.computeConv3DInfo(\n        x5D.shape, filterShape, strides, dilations, pad);\n\n    return backend.conv3dDerFilter(x5D, dy5D, convInfo);\n  };\n\n  const inputs: Conv3DBackpropFilterInputs = {x: x5D, y: dy5D};\n\n  const attrs: Conv3DBackpropFilterAttrs = {strides, pad};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null,\n             Conv3DBackpropFilterV2, attrs as {} as NamedAttrMap) as Tensor5D;\n}\n\nexport const conv3DBackpropFilter = op({conv3DBackpropFilter_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Conv3D, Conv3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {conv3DBackpropFilter} from '../ops/conv3d_backprop_filter';\nimport {conv3DBackpropInput} from '../ops/conv3d_backprop_input';\nimport {tupleValuesAreOne} from '../ops/conv_util';\nimport {Tensor, Tensor5D} from '../tensor';\nimport * as util from '../util';\n\nexport const conv3DGradConfig: GradConfig = {\n  kernelName: Conv3D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor5D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad} = attrs as {} as Conv3DAttrs;\n    util.assert(\n        tupleValuesAreOne(dilations),\n        () =>\n            'Error in gradient of conv3D: dilation rates greater than 1 are ' +\n            `not yet supported in gradients. Got dilations '${dilations}'`);\n\n    const [x5D, $filter] = saved;\n\n    return {\n      x: () => conv3DBackpropInput(\n          (x5D as Tensor5D).shape, dy, $filter as Tensor5D, strides, pad),\n      filter: () => conv3DBackpropFilter(\n          x5D as Tensor5D, dy, ($filter as Tensor5D).shape, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {sin} from '../ops/sin';\nimport {Tensor} from '../tensor';\n\nexport const cosGradConfig: GradConfig = {\n  kernelName: Cos,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(neg(sin(cast(x, 'float32'))), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {sinh} from '../ops/sinh';\nimport {Tensor} from '../tensor';\n\nexport const coshGradConfig: GradConfig = {\n  kernelName: Cosh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(sinh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cumsum, CumsumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getAxesPermutation} from '../ops/axis_util';\nimport {cumsum} from '../ops/cumsum';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const cumsumGradConfig: GradConfig = {\n  kernelName: Cumsum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {axis, exclusive, reverse}: CumsumAttrs = attrs as {} as CumsumAttrs;\n\n    return {\n      x: () => {\n        const permutation = getAxesPermutation([axis], x.rank);\n\n        let out = cumsum(dy, axis, exclusive, !reverse);\n\n        if (permutation != null) {\n          out = transpose(out, permutation);\n        }\n\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {DepthwiseConv2dNative, DepthwiseConv2dNativeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as conv_util from '../ops/conv_util';\nimport {depthwiseConv2dNativeBackpropFilter} from '../ops/depthwise_conv2d_native_backprop_filter';\nimport {depthwiseConv2dNativeBackpropInput} from '../ops/depthwise_conv2d_native_backprop_input';\nimport {Tensor, Tensor4D} from '../tensor';\nimport * as util from '../util';\n\nexport const depthwiseConv2dNativeGradConfig: GradConfig = {\n  kernelName: DepthwiseConv2dNative,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dilations, strides, pad, dimRoundingMode} =\n        attrs as {} as DepthwiseConv2dNativeAttrs;\n\n    const $dilations = dilations == null ? [1, 1] : dilations;\n\n    util.assert(\n        conv_util.tupleValuesAreOne($dilations),\n        () => 'Error in gradient of depthwiseConv2dNative: dilation rates ' +\n            `greater than 1 are not yet supported. Got dilations ` +\n            `'${$dilations}'`);\n\n    const [x, filter] = saved as [Tensor4D, Tensor4D];\n\n    util.assert(\n        x.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: input must be ` +\n            `rank 4, but got rank ${x.rank}.`);\n    util.assert(\n        filter.rank === 4,\n        () => `Error in gradient of depthwiseConv2dNative: filter must be ` +\n            `rank 4, but got rank ${filter.rank}.`);\n    util.assert(\n        x.shape[3] === filter.shape[2],\n        () => `Error in gradient of depthwiseConv2d: number of input ` +\n            `channels (${x.shape[3]}) must match the inChannels dimension ` +\n            `in filter ${filter.shape[2]}.`);\n\n    util.assert(\n        conv_util.eitherStridesOrDilationsAreOne(strides, $dilations),\n        () => 'Error in gradient of depthwiseConv2d: Either strides or ' +\n            `dilations must be  1. Got strides ${strides} and dilations ` +\n            `'${$dilations}'.`);\n\n    if (dimRoundingMode != null) {\n      util.assert(\n          util.isInt(pad as number),\n          () =>\n              `Error in depthwiseConv2d: pad must be an integer when using, ` +\n              `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n    }\n\n    const convInfo = conv_util.computeConv2DInfo(\n        x.shape, filter.shape, strides, $dilations as number | [number, number],\n        pad, dimRoundingMode, true /* depthwise */);\n\n    return {\n      x: () =>\n          depthwiseConv2dNativeBackpropInput(x.shape, dy, filter, convInfo),\n      filter: () =>\n          depthwiseConv2dNativeBackpropFilter(x, dy, filter.shape, convInfo),\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Dilation2D, Dilation2DBackpropFilter, Dilation2DBackpropFilterInputs, Dilation2DBackpropInput, Dilation2DBackpropInputInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const dilation2dGradConfig: GradConfig = {\n  kernelName: Dilation2D,\n  inputsToSave: ['x', 'filter'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, filter] = saved as [Tensor4D, Tensor3D];\n\n    const inputInputs: Dilation2DBackpropInputInputs = {x, filter, dy};\n    const filterInputs: Dilation2DBackpropFilterInputs = {x, filter, dy};\n\n    return {\n      x: () => ENGINE.runKernel(\n                   Dilation2DBackpropInput, inputInputs as {} as NamedTensorMap,\n                   attrs) as Tensor,\n      filter: () => ENGINE.runKernel(\n                        Dilation2DBackpropFilter,\n                        filterInputs as {} as NamedTensorMap, attrs) as Tensor\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const divGradConfig: GradConfig = {\n  kernelName: Div,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Elu, EluGrad, EluGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const eluGradConfig: GradConfig = {\n  kernelName: Elu,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      return backend.eluDer(dy, y);\n    };\n\n    const inputs: EluGradInputs = {dy, y};\n\n    return {\n      x: () => ENGINE.runKernelFunc(\n          backPropKernelFunc, inputs as {} as NamedTensorMap, null /* grad */,\n          EluGrad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Erf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const erfGradConfig: GradConfig = {\n  kernelName: Erf,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const a = mul(exp(neg(square(x))), 2 / Math.sqrt(Math.PI));\n    return {x: () => mul(dy, a)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expGradConfig: GradConfig = {\n  kernelName: Exp,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n    return {x: () => mul(dy, y)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const expm1GradConfig: GradConfig = {\n  kernelName: Expm1,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, exp(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const floorGradConfig: GradConfig = {\n  kernelName: Floor,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FloorDiv} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {square} from '../ops/square';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const floorDivGradConfig: GradConfig = {\n  kernelName: FloorDiv,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = div(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      let res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = reshape(sum(res, reduceAxes), b.shape);\n      }\n      const tmp = square(b);\n      return neg(div(res, cast(tmp, 'float32')));\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {FusedBatchNorm, FusedBatchNormAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {rsqrt} from '../ops/rsqrt';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {tile} from '../ops/tile';\nimport {Tensor} from '../tensor';\nimport {Rank, ShapeMap} from '../types';\n\nexport const fusedBatchNormGradConfig: GradConfig = {\n  kernelName: FusedBatchNorm,\n  inputsToSave: ['x', 'mean', 'variance', 'scale'],\n  gradFunc: <R extends Rank>(\n      dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {varianceEpsilon} = attrs as {} as FusedBatchNormAttrs;\n    const [x, mean, variance, scale] = saved;\n\n    const scaleValue = scale == null ? scalar(1) : scale;\n    const reductionAxes = getReductionAxes(mean.shape, x.shape);\n    const tileShape: number[] = [];\n    if (mean.rank === 1) {\n      for (let i = 0; i < x.shape.length - 1; ++i) {\n        tileShape.push(x.shape[i]);\n      }\n      tileShape.push(1);\n    }\n\n    const xMinusMean = sub(x, mean);\n    const dyTimesScaleValue = mul(dy, scaleValue);\n    const oneOverSqrtVariance = rsqrt(add(variance, scalar(varianceEpsilon)));\n    const minusHalfRCube = mul(\n        mul(mul(oneOverSqrtVariance, oneOverSqrtVariance), oneOverSqrtVariance),\n        scalar(-0.5));\n\n    const derX = () => {\n      if (mean.rank === 1) {\n        return reshape(\n            mul(mul(dy,\n                    tile(\n                        reshape(oneOverSqrtVariance, [1, 1, 1, mean.shape[0]]),\n                        tileShape)),\n                scaleValue),\n            x.shape);\n      } else {\n        return reshape(mul(mul(dy, oneOverSqrtVariance), scaleValue), x.shape);\n      }\n    };\n    const derMean = () => {\n      let meanDer =\n          mul(mul(oneOverSqrtVariance, scalar(-1)), dyTimesScaleValue);\n      if (mean.rank === 1) {\n        meanDer = sum(meanDer, reductionAxes);\n      }\n      return reshape(meanDer, mean.shape as ShapeMap[R]);\n    };\n    const derVariance = () => {\n      let varianceDer = mul(mul(minusHalfRCube, xMinusMean), dyTimesScaleValue);\n\n      if (mean.rank === 1) {\n        varianceDer = sum(varianceDer, reductionAxes);\n      }\n      return reshape(varianceDer, mean.shape as ShapeMap[R]);\n    };\n    const derScale = () => {\n      const xMinusMean2TimesRsqrt = mul(xMinusMean, oneOverSqrtVariance);\n\n      let scaleDer = mul(dy, xMinusMean2TimesRsqrt);\n      if (mean.rank === 1) {\n        scaleDer = sum(scaleDer, reductionAxes);\n      }\n      return reshape(scaleDer, mean.shape as ShapeMap[R]);\n    };\n    const derOffset = () => {\n      let offsetDer = dy;\n      if (mean.rank === 1) {\n        offsetDer = sum(offsetDer, reductionAxes);\n      }\n      return reshape(offsetDer, mean.shape as ShapeMap[R]);\n    };\n\n    return {\n      x: derX,\n      mean: derMean,\n      variance: derVariance,\n      scale: derScale,\n      offset: derOffset\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GatherV2, GatherV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {getUndoAxesPermutation} from '../ops/axis_util';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {unsortedSegmentSum} from '../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const gatherGradConfig: GradConfig = {\n  kernelName: GatherV2,\n  inputsToSave: ['x', 'indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, indices] = saved;\n    const {axis} = attrs as {} as GatherV2Attrs;\n\n    const parsedAxis = parseAxisParam(axis, x.shape)[0];\n\n    const derX = () => {\n      const paramsShape = x.shape;\n      const indicesSize = indices.size;\n\n      const outerShape = paramsShape.slice(0, parsedAxis);\n      const outerDims = outerShape.length;\n      const innerShape = paramsShape.slice(axis, paramsShape.length).slice(1);\n      const innerDims = innerShape.length;\n\n      const outerAxesIndices = arrayRange(0, outerDims);\n      const innerAxesIndices =\n          arrayRange(outerDims + 1, outerDims + 1 + innerDims);\n\n      const valuesShape = arrayConcat([outerShape, [indicesSize], innerShape]);\n\n      const values = reshape(dy, valuesShape);\n      const reshapedIndices = reshape(indices, [indicesSize]);\n\n      const transposeDims =\n          arrayConcat([[outerDims], outerAxesIndices, innerAxesIndices]);\n      const valuesTranspose = transpose(values, transposeDims);\n      let paramsGrad = unsortedSegmentSum(\n          valuesTranspose, reshapedIndices as Tensor1D, x.shape[parsedAxis]);\n\n      const invertTransposeDims = getUndoAxesPermutation(transposeDims);\n      paramsGrad = transpose(paramsGrad, invertTransposeDims);\n\n      return paramsGrad;\n    };\n    return {x: derX, indices: () => indices};\n  }\n};\n\nfunction arrayRange(start: number, stop: number): number[] {\n  const result = [];\n  for (let i = start; i < stop; ++i) {\n    result.push(i);\n  }\n  return result;\n}\n\nfunction arrayConcat(arrays: number[][]): number[] {\n  const result = [];\n  for (let i = 0; i < arrays.length; ++i) {\n    for (let j = 0; j < arrays[i].length; ++j) {\n      result.push(arrays[i][j]);\n    }\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GreaterEqual} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const greaterEqualGradConfig: GradConfig = {\n  kernelName: GreaterEqual,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    return {a: () => zerosLike(a), b: () => zerosLike(b)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {Tensor} from '../tensor';\n\nexport const identityGradConfig: GradConfig = {\n  kernelName: Identity,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => cast(dy, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isFiniteGradConfig: GradConfig = {\n  kernelName: IsFinite,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isInfGradConfig: GradConfig = {\n  kernelName: IsInf,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const isNanGradConfig: GradConfig = {\n  kernelName: IsNan,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log1p} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const log1pGradConfig: GradConfig = {\n  kernelName: Log1p,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, add(x, 1))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Log} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {Tensor} from '../tensor';\n\nexport const logGradConfig: GradConfig = {\n  kernelName: Log,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, cast(x, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {LogSoftmax, LogSoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {exp} from '../ops/exp';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const logSoftmaxGradConfig: GradConfig = {\n  kernelName: LogSoftmax,\n  inputsToSave: [],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [value] = saved;\n    const {axis} = attrs as {} as LogSoftmaxAttrs;\n    return {\n      logits: () => {\n        const keepDims = true;\n        const softmax = exp(value);\n        return sub(dy, mul(sum(dy, axis, keepDims), softmax));\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {LRNBackprop, LRNBackpropAttrs, LRNBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nimport {op} from './operation';\n\nfunction localResponseNormalizationBackprop_<T extends Tensor4D>(\n    x: T, y: T, dy: T, depthRadius = 5, bias = 1, alpha = 1, beta = 0.5): T {\n  const forward: ForwardFunc<Tensor> = backend =>\n      backend.LRNGrad(dy, x, y, depthRadius, bias, alpha, beta);\n\n  const inputs: LRNBackpropInputs = {x, y, dy};\n\n  const attrs: LRNBackpropAttrs = {depthRadius, bias, alpha, beta};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null /* grad */,\n             LRNBackprop, attrs as {} as NamedAttrMap) as T;\n}\n\nexport const localResponseNormalizationBackprop =\n    op({localResponseNormalizationBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {LRN, LRNAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {localResponseNormalizationBackprop} from '../ops/local_response_normalization_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const lrnGradConfig: GradConfig = {\n  kernelName: LRN,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {depthRadius, bias, alpha, beta} = attrs as {} as LRNAttrs;\n\n    return {\n      x: () => localResponseNormalizationBackprop(\n          x, y, dy, depthRadius, bias, alpha, beta)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as axis_util from '../ops/axis_util';\nimport {cast} from '../ops/cast';\nimport {equal} from '../ops/equal';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\n/**\n * Gradient helper function for the min and max operations.\n */\nexport function gradForMinAndMax<T extends Tensor>(\n    dy: T, y: T, xOrig: Tensor, origAxes: number[], permutedAxes: number[]) {\n  if (y.rank < xOrig.rank) {\n    y = reshape(y, axis_util.expandShapeToKeepDim(y.shape, origAxes)) as T;\n  }\n  if (dy.rank < xOrig.rank) {\n    dy = reshape(dy, axis_util.expandShapeToKeepDim(dy.shape, origAxes)) as T;\n  }\n  return {\n    x: () => {\n      const dx = mul(dy, cast(equal(xOrig, y), dy.dtype));\n      return permutedAxes == null ? dx : transpose(dx, permutedAxes);\n    }\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const maxGradConfig: GradConfig = {\n  kernelName: Max,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const maxAttrs: MaxAttrs = attrs as {} as MaxAttrs;\n    const {reductionIndices} = maxAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const maxGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = maxGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Maximum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {less} from '../ops/less';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const maximumGradConfig: GradConfig = {\n  kernelName: Maximum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(greaterEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(less(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPool3DBackprop, MaxPool3DBackpropAttrs, MaxPool3DBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D, Tensor5D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the backprop of a 3d max pool.\n *\n * @param dy The dy error, of rank 5 of shape\n *     [batchSize, depth, height, width, channels].\n * assumed.\n * @param input The original input image, of rank 5 or rank 4 of shape\n *     [batchSize, depth, height, width, channels].\n * @param output The original output image, of rank 5 of shape\n *     [batchSize, outDepth, outHeight, outWidth, channels].\n * @param filterSize The filter size:\n *     `[filterDepth, filterHeight, filterWidth]`.\n *     `filterSize` is a single number,\n *     then `filterDepth == filterHeight == filterWidth`.\n * @param strides The strides of the pooling:\n *     `[strideDepth, strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param dilations Deprecated, this field will be gone in v3.0.0.\n *     The dilation rates: `[dilationDepth, dilationHeight, dilationWidth]`\n *     in which we sample input values across the depth, height and width\n *     dimensions in dilated pooling.\n *     Defaults to `[1, 1, 1]`. If `dilations` is a single number,\n *     then `dilationDepth == dilationHeight == dilationWidth`.\n *     If it is greater than 1, then all values of `strides` must be 1.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPool3dBackprop_<T extends Tensor4D|Tensor5D>(\n    dy: T|TensorLike, input: T|TensorLike, output: T|TensorLike,\n    filterSize: [number, number, number]|number,\n    strides: [number, number, number]|number,\n    dilations: [number, number, number]|number = [1, 1, 1],\n    pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  const $dy = convertToTensor(dy, 'dy', 'maxPool3dBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPool3dBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPool3dBackprop');\n\n  let dy5D = $dy as Tensor5D;\n  let input5D = $input as Tensor5D;\n  let output5D = $output as Tensor5D;\n  let reshapedTo5D = false;\n\n  if ($input.rank === 4) {\n    reshapedTo5D = true;\n    dy5D = reshape(\n        $dy, [1, $dy.shape[0], $dy.shape[1], $dy.shape[2], $dy.shape[3]]);\n    input5D = reshape($input, [\n      1, $input.shape[0], $input.shape[1], $input.shape[2], $input.shape[3]\n    ]);\n    output5D = reshape($output, [\n      1, $output.shape[0], $output.shape[1], $output.shape[2], $output.shape[3]\n    ]);\n  }\n\n  util.assert(\n      dy5D.rank === 5,\n      () => `Error in maxPool3dBackprop: dy must be rank 5 but got rank ` +\n          `${dy5D.rank}.`);\n  util.assert(\n      input5D.rank === 5,\n      () => `Error in maxPool3dBackprop: input must be rank 5 but got rank ` +\n          `${input5D.rank}.`);\n  util.assert(\n      output5D.rank === 5,\n      () => `Error in maxPool3dBackprop: output must be rank 5 but got rank ` +\n          `${output5D.rank}.`);\n\n  util.assert(\n      conv_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool3dBackprop: Either strides or dilations ' +\n          `must be 1. Got strides ${strides} and dilations '${dilations}'`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPool3dBackprop: pad must be an integer when ` +\n            `using, dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool3DInfo(\n        input5D.shape, filterSize, strides, dilations, pad, dimRoundingMode);\n\n    return backend.maxPool3dBackprop(dy5D, input5D, output5D, convInfo);\n  };\n\n  const inputs:\n      MaxPool3DBackpropInputs = {dy: dy5D, input: input5D, output: output5D};\n\n  const attrs: MaxPool3DBackpropAttrs =\n      {filterSize, strides, dilations, pad, dimRoundingMode};\n\n  const res = ENGINE.runKernelFunc(\n      forward, inputs as {} as NamedTensorMap, null /* grad */,\n      MaxPool3DBackprop, attrs as {} as NamedAttrMap);\n\n  if (reshapedTo5D) {\n    return reshape(\n               res, [res.shape[1], res.shape[2], res.shape[3], res.shape[4]]) as\n        T;\n  }\n\n  return res as T;\n}\n\nexport const maxPool3dBackprop = op({maxPool3dBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool3D, MaxPool3DAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPool3dBackprop} from '../ops/max_pool_3d_backprop';\nimport {Tensor, Tensor5D} from '../tensor';\n\nexport const maxPool3DGradConfig: GradConfig = {\n  kernelName: MaxPool3D,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor5D, Tensor5D];\n    const {filterSize, strides, dilations, pad, dimRoundingMode} =\n        attrs as {} as MaxPool3DAttrs;\n\n    const $dilations =\n        dilations == null ? [1, 1, 1] as [number, number, number] : dilations;\n\n    return {\n      x: () => maxPool3dBackprop(\n          dy as Tensor5D, x, y, filterSize, strides, $dilations, pad,\n          dimRoundingMode)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport * as conv_util from './conv_util';\nimport {op} from './operation';\n\n/**\n * Computes the backprop of a 2D max pool.\n *\n * @param dy The dy error, of rank 4 or rank 3 of shape\n *     [batchSize, height, width, channels]. If rank 3, batch of 1 is\n * assumed.\n * @param input The original input image, of rank 4, of shape\n *     [batchSize, height, width, channels].\n * @param output The original output image, of rank 4, of shape\n *     [batchSize, outHeight, outWidth, channels].\n * @param filterSize The filter size: `[filterHeight, filterWidth]`. If\n *     `filterSize` is a single number, then `filterHeight == filterWidth`.\n * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If\n *     `strides` is a single number, then `strideHeight == strideWidth`.\n * @param pad A string from: 'same', 'valid'. The type of padding algorithm\n *     used in the forward prop of the op.\n * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. The\n *     rounding mode used when computing output dimensions if pad is a\n *     number. If none is provided, it will not round and error if the output\n *     is of fractional size.\n */\nfunction maxPoolBackprop_(\n    dy: Tensor4D|TensorLike, input: Tensor4D|TensorLike,\n    output: Tensor4D|TensorLike, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): Tensor4D {\n  const $dy = convertToTensor(dy, 'dy', 'maxPoolBackprop');\n  const $input = convertToTensor(input, 'input', 'maxPoolBackprop');\n  const $output = convertToTensor(output, 'output', 'maxPoolBackprop');\n\n  util.assert(\n      $input.rank === $dy.rank,\n      () => `Rank of input (${$input.rank}) does not match rank of dy ` +\n          `(${$dy.rank})`);\n\n  util.assert(\n      $dy.rank === 4,\n      () => `Error in maxPoolBackprop: dy must be rank 4 but got rank ` +\n          `${$dy.rank}.`);\n  util.assert(\n      $input.rank === 4,\n      () => `Error in maxPoolBackprop: input must be rank 4 but got rank ` +\n          `${$input.rank}.`);\n  if (dimRoundingMode != null) {\n    util.assert(\n        util.isInt(pad as number),\n        () => `Error in maxPoolBackprop: pad must be an integer when using, ` +\n            `dimRoundingMode ${dimRoundingMode} but got pad ${pad}.`);\n  }\n\n  const forward: ForwardFunc<Tensor> = backend => {\n    const convInfo = conv_util.computePool2DInfo(\n        $input.shape, filterSize, strides, 1 /* dilations */, pad,\n        dimRoundingMode);\n\n    return backend.maxPoolBackprop($dy, $input, $output, convInfo);\n  };\n\n  const inputs:\n      MaxPoolBackpropInputs = {dy: $dy, input: $input, output: $output};\n\n  const attrs:\n      MaxPoolBackpropAttrs = {filterSize, strides, pad, dimRoundingMode};\n\n  return ENGINE.runKernelFunc(\n             forward, inputs as {} as NamedTensorMap, null, MaxPoolBackprop,\n             attrs as {} as NamedAttrMap) as Tensor4D;\n}\n\nexport const maxPoolBackprop = op({maxPoolBackprop_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPool, MaxPoolAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {maxPoolBackprop} from '../ops/max_pool_backprop';\nimport {Tensor, Tensor4D} from '../tensor';\n\nexport const maxPoolGradConfig: GradConfig = {\n  kernelName: MaxPool,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x, y] = saved as [Tensor4D, Tensor4D];\n    const {filterSize, strides, pad} = attrs as {} as MaxPoolAttrs;\n\n    return {\n      x: () => maxPoolBackprop(dy as Tensor4D, x, y, filterSize, strides, pad)\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Min, MinAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\nimport * as util from '../util';\n\nimport {gradForMinAndMax} from './min_max_grad_util';\n\nexport const minGradConfig: GradConfig = {\n  kernelName: Min,\n  inputsToSave: ['x'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const minAttrs: MinAttrs = attrs as {} as MinAttrs;\n    const {axis} = minAttrs;\n    const [x, y] = saved;\n    const origAxes = util.parseAxisParam(axis, x.shape);\n    const permutedAxes = axis_util.getAxesPermutation(origAxes, x.rank);\n    const minGrad = gradForMinAndMax(dy, y, x, origAxes, permutedAxes);\n    return {\n      x: () => {\n        let out = minGrad['x']();\n        if (permutedAxes != null) {\n          out = transpose(out);\n        }\n        return out;\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Minimum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const minimumGradConfig: GradConfig = {\n  kernelName: Minimum,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const derA = () => mul(dy, cast(lessEqual(a, b), 'float32'));\n    const derB = () => mul(dy, cast(greater(a, b), 'float32'));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Mod} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {div} from '../ops/div';\nimport {floor} from '../ops/floor';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const modGradConfig: GradConfig = {\n  kernelName: Mod,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(dy, reduceAxes), a.shape);\n      }\n      return dy;\n    };\n    const derB = () => {\n      const res = mul(dy, neg(floor(div(a, b))));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Multiply} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {assertAndGetBroadcastShape, getReductionAxes} from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const multiplyGradConfig: GradConfig = {\n  kernelName: Multiply,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape = assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      const res = mul(dy, cast(b, 'float32'));\n      const reduceAxes = getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), a.shape);\n      }\n      return res;\n    };\n    const derB = () => {\n      const res = mul(dy, cast(a, 'float32'));\n      const reduceAxes = getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        return reshape(sum(res, reduceAxes), b.shape);\n      }\n      return res;\n    };\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Negate} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {neg} from '../ops/neg';\nimport {Tensor} from '../tensor';\n\nexport const negateGradConfig: GradConfig = {\n  kernelName: Negate,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => neg(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OneHot} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zeros} from '../ops/zeros';\nimport {Tensor} from '../tensor';\n\nexport const oneHotGradConfig: GradConfig = {\n  kernelName: OneHot,\n  inputsToSave: ['indices'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const indices = saved[0];\n    return {indices: () => zeros(indices.shape, 'float32')};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OnesLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const onesLikeGradConfig: GradConfig = {\n  kernelName: OnesLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PadV2, PadV2Attrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {slice} from '../ops/slice';\nimport {Tensor} from '../tensor';\n\nexport const padV2GradConfig: GradConfig = {\n  kernelName: PadV2,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    // Pad introduces values around the original tensor, so the gradient\n    // slices the original shape out of the gradient.\n    const x = saved[0];\n    const {paddings} = attrs as unknown as PadV2Attrs;\n    const begin = paddings.map(p => p[0]);\n    return {x: () => slice(dy, begin, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Pow} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {cast} from '../ops/cast';\nimport {greater} from '../ops/greater';\nimport {log} from '../ops/log';\nimport {mul} from '../ops/mul';\nimport {pow} from '../ops/pow';\nimport {reshape} from '../ops/reshape';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const powGradConfig: GradConfig = {\n  kernelName: Pow,\n  inputsToSave: ['a', 'b'],\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b, y] = saved;\n    const base = a;\n    const exp = b;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(base.shape, exp.shape);\n\n    const derBase = () => {\n      const expFloat = cast(exp, 'float32');\n      let res = mul(dy, mul(expFloat, pow(base, sub(expFloat, scalar(1)))));\n      const reduceAxes = broadcast_util.getReductionAxes(base.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, base.shape);\n    };\n    const derExp = () => {\n      const condition = greater(base, 0);\n      const logBase = where(condition, log(base), zerosLike(base));\n      let res = mul(dy, mul(y, logBase));\n      const reduceAxes = broadcast_util.getReductionAxes(exp.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, exp.shape);\n    };\n    return {a: derBase, b: derExp};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Prelu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {getReductionAxes} from '../ops/broadcast_util';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const preluGradConfig: GradConfig = {\n  kernelName: Prelu,\n  inputsToSave: ['x', 'alpha'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x, alpha] = saved;\n    const mask = greater(x, 0);\n\n    return {\n      x: () => where(mask, dy, mul(dy, alpha)),\n      alpha: () => {\n        let res = where(mask, zerosLike(dy), mul(dy, x));\n        const reduceAxes = getReductionAxes(alpha.shape, dy.shape);\n        if (reduceAxes.length > 0) {\n          res = sum(res, reduceAxes);\n        }\n        return reshape(res, alpha.shape);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Reciprocal} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {neg} from '../ops/neg';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const reciprocalGradConfig: GradConfig = {\n  kernelName: Reciprocal,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => div(dy, neg(square(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu6} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {lessEqual} from '../ops/less_equal';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const relu6GradConfig: GradConfig = {\n  kernelName: Relu6,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    const mask = mul(lessEqual(x, 6), step(x));\n\n    return {x: () => mul(dy, cast(mask, 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Relu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {step} from '../ops/step';\nimport {Tensor} from '../tensor';\n\nexport const reluGradConfig: GradConfig = {\n  kernelName: Relu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, cast(step(x), 'float32'))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reshape} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\n\nexport const reshapeGradConfig: GradConfig = {\n  kernelName: Reshape,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => reshape(dy, x.shape)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearGrad, ResizeBilinearGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeBilinearGradConfig: GradConfig = {\n  kernelName: ResizeBilinear,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeBilinearAttrs;\n      return backend.resizeBilinearBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeBilinearGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeBilinearGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {ResizeNearestNeighbor, ResizeNearestNeighborAttrs, ResizeNearestNeighborGrad, ResizeNearestNeighborGradInputs} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor4D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\n\nexport const resizeNearestNeighborGradConfig: GradConfig = {\n  kernelName: ResizeNearestNeighbor,\n  inputsToSave: ['images'],\n  gradFunc: (dy: Tensor4D, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [images] = saved;\n\n    const backPropKernelFunc: ForwardFunc<Tensor> = (backend) => {\n      const {alignCorners} = attrs as {} as ResizeNearestNeighborAttrs;\n      return backend.resizeNearestNeighborBackprop(\n          dy, images as Tensor4D, alignCorners);\n    };\n\n    const inputs: ResizeNearestNeighborGradInputs = {images};\n    const imagesDer = () => ENGINE.runKernelFunc(\n        backPropKernelFunc, inputs as {} as NamedTensorMap, null /* gradient */,\n        ResizeNearestNeighborGrad, attrs);\n\n    return {images: imagesDer};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Reverse, ReverseAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {reverse} from '../ops/reverse';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const reverseGradConfig: GradConfig = {\n  kernelName: Reverse,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {dims} = attrs as {} as ReverseAttrs;\n    const axes = parseAxisParam(dims, dy.shape);\n    return {x: () => reverse(dy, axes)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Round} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const roundGradConfig: GradConfig = {\n  kernelName: Round,\n  gradFunc: (dy: Tensor) => {\n    // TODO(nsthorat): Let gradients be null for cases where we want to stop\n    // backpropgation.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rsqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {neg} from '../ops/neg';\nimport {pow} from '../ops/pow';\nimport {Tensor} from '../tensor';\n\nexport const rsqrtGradConfig: GradConfig = {\n  kernelName: Rsqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => neg(div(dy, mul(pow(x, 1.5), 2)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SelectV2} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {logicalNot} from '../ops/logical_not';\nimport {mul} from '../ops/mul';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const selectV2PoolGradConfig: GradConfig = {\n  kernelName: SelectV2,\n  inputsToSave: ['condition'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [condition] = saved;\n    return {\n      // TODO(julianoks): Return null for condition gradient\n      // when backprop supports it.\n      condition: () => cast(zerosLike(condition), 'float32'),\n      t: () => mul(dy, cast(condition, dy.dtype)),\n      e: () => mul(dy, cast(logicalNot(condition), dy.dtype))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Selu} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {exp} from '../ops/exp';\nimport {greater} from '../ops/greater';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {SELU_SCALE, SELU_SCALEALPHA} from '../ops/selu_util';\nimport {where} from '../ops/where';\nimport {Tensor} from '../tensor';\n\nexport const seluGradConfig: GradConfig = {\n  kernelName: Selu,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {\n      x: () => {\n        const mask = greater(x, scalar(0));\n\n        const scaleAlpha = scalar(SELU_SCALEALPHA);\n        const scale = scalar(SELU_SCALE);\n\n        const greaterThanZeroDer = mul(dy, scale);\n        const lessEqualZeroDer =\n            mul(mul(dy, scaleAlpha), exp(cast(x, 'float32')));\n\n        return where(mask, greaterThanZeroDer, lessEqualZeroDer);\n      }\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sigmoid} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const sigmoidGradConfig: GradConfig = {\n  kernelName: Sigmoid,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(dy, mul(y, sub(scalar(1), y)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sign} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const signGradConfig: GradConfig = {\n  kernelName: Sign,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sin} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cos} from '../ops/cos';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinGradConfig: GradConfig = {\n  kernelName: Sin,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(cos(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sinh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {cosh} from '../ops/cosh';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const sinhGradConfig: GradConfig = {\n  kernelName: Sinh,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(cosh(cast(x, 'float32')), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Slice, SliceAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {pad} from '../ops/pad';\nimport {parseSliceParams} from '../ops/slice_util';\nimport {Tensor} from '../tensor';\n\nexport const sliceGradConfig: GradConfig = {\n  kernelName: Slice,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {begin, size} = attrs as {} as SliceAttrs;\n\n    const inputShape = x.shape;\n    const [begin_, size_] = parseSliceParams(x, begin, size);\n\n    // Create an Nx2 padding where the first column represents how many\n    // zeros are prepended (at start) for each dimension, and the second\n    // column indicates how many zeros are appended (at end).\n\n    // The number of zeros to append is the shape of the input\n    // elementwise-subtracted by both the begin vector and sizes vector.\n    const paddings: Array<[number, number]> = [];\n    for (let i = 0; i < dy.rank; i++) {\n      paddings.push([begin_[i], inputShape[i] - begin_[i] - size_[i]]);\n    }\n    return {x: () => pad(dy, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softmax, SoftmaxAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sub} from '../ops/sub';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const softmaxGradConfig: GradConfig = {\n  kernelName: Softmax,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [y] = saved;\n    const {dim} = attrs as {} as SoftmaxAttrs;\n    const keepDims = true;\n\n    const dyTimesY = mul(dy, y);\n    return {\n      logits: () => sub(dyTimesY, mul(sum(dyTimesY, [dim], keepDims), y))\n    };\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Softplus} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {sigmoid} from '../ops/sigmoid';\nimport {Tensor} from '../tensor';\n\nexport const softplusGradConfig: GradConfig = {\n  kernelName: Softplus,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => mul(dy, sigmoid(x))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SpaceToBatchND, SpaceToBatchNDAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {batchToSpaceND} from '../ops/batch_to_space_nd';\nimport {Tensor} from '../tensor';\n\nexport const spaceToBatchNDGradConfig: GradConfig = {\n  kernelName: SpaceToBatchND,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const {blockShape, paddings} = attrs as {} as SpaceToBatchNDAttrs;\n    return {x: () => batchToSpaceND(dy, blockShape, paddings)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {SplitV, SplitVAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {concat} from '../ops/concat';\nimport {Tensor} from '../tensor';\n\nexport const splitVGradConfig: GradConfig = {\n  kernelName: SplitV,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const {axis} = attrs as {} as SplitVAttrs;\n\n    return {x: () => concat(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sqrt} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {div} from '../ops/div';\nimport {mul} from '../ops/mul';\nimport {sqrt} from '../ops/sqrt';\nimport {Tensor} from '../tensor';\n\nexport const sqrtGradConfig: GradConfig = {\n  kernelName: Sqrt,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, mul(sqrt(cast(x, 'float32')), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cast} from '../ops/cast';\nimport {mul} from '../ops/mul';\nimport {Tensor} from '../tensor';\n\nexport const squareGradConfig: GradConfig = {\n  kernelName: Square,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n    return {x: () => mul(dy, mul(cast(x, 'float32'), 2))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {SquaredDifference} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const squaredDifferenceGradConfig: GradConfig = {\n  kernelName: SquaredDifference,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const two = scalar(2);\n    const derA = () => mul(dy, mul(two, sub(a, b)));\n    const derB = () => mul(dy, mul(two, sub(b, a)));\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Step} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const stepGradConfig: GradConfig = {\n  kernelName: Step,\n  gradFunc: (dy: Tensor) => {\n    // TODO(manrajgrover): Return null for gradients when backprop supports\n    // it.\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Sub} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport * as broadcast_util from '../ops/broadcast_util';\nimport {neg} from '../ops/neg';\nimport {reshape} from '../ops/reshape';\nimport {sum} from '../ops/sum';\nimport {Tensor} from '../tensor';\n\nexport const subGradConfig: GradConfig = {\n  kernelName: Sub,\n  inputsToSave: ['a', 'b'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [a, b] = saved;\n    const outShape =\n        broadcast_util.assertAndGetBroadcastShape(a.shape, b.shape);\n\n    const derA = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(a.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(res, a.shape);\n    };\n    const derB = () => {\n      let res = dy;\n      const reduceAxes = broadcast_util.getReductionAxes(b.shape, outShape);\n      if (reduceAxes.length > 0) {\n        res = sum(res, reduceAxes);\n      }\n      return reshape(neg(res), b.shape);\n    };\n\n    return {a: derA, b: derB};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Sum, SumAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {ones} from '../ops/ones';\nimport {reshape} from '../ops/reshape';\nimport {Tensor} from '../tensor';\nimport {parseAxisParam} from '../util';\n\nexport const sumGradConfig: GradConfig = {\n  kernelName: Sum,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const expandedDyShape = x.shape.slice();\n    const {axis} = attrs as {} as SumAttrs;\n\n    const axes = parseAxisParam(axis, x.shape);\n    axes.forEach(axis => {\n      expandedDyShape[axis] = 1;\n    });\n    const expandedDy = reshape(dy, expandedDyShape);\n    const derX = mul(expandedDy, ones(x.shape, 'float32'));\n\n    return {x: () => derX};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tan} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {cos} from '../ops/cos';\nimport {div} from '../ops/div';\nimport {square} from '../ops/square';\nimport {Tensor} from '../tensor';\n\nexport const tanGradConfig: GradConfig = {\n  kernelName: Tan,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [x] = saved;\n\n    return {x: () => div(dy, square(cos(x)))};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tanh} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {mul} from '../ops/mul';\nimport {scalar} from '../ops/scalar';\nimport {square} from '../ops/square';\nimport {sub} from '../ops/sub';\nimport {Tensor} from '../tensor';\n\nexport const tanhGradConfig: GradConfig = {\n  kernelName: Tanh,\n  outputsToSave: [true],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [y] = saved;\n\n    return {x: () => mul(sub(scalar(1), square(y)), dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tile, TileAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {add} from '../ops/add';\nimport {slice} from '../ops/slice';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const tileGradConfig: GradConfig = {\n  kernelName: Tile,\n  inputsToSave: ['x'],\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const [x] = saved;\n    const {reps} = attrs as unknown as TileAttrs;\n\n    const derX = () => {\n      let xGrad = zerosLike(x);\n      // TODO(cais): Maybe reduce memory footprint by avoiding repeated\n      // slicing.\n      if (x.rank === 1) {\n        for (let i = 0; i < reps[0]; ++i) {\n          xGrad = add(xGrad, slice(dy, [i * x.shape[0]], [x.shape[0]]));\n        }\n      } else if (x.rank === 2) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            xGrad = add(xGrad, slice(dy, [i * x.shape[0], j * x.shape[1]], [\n                          x.shape[0], x.shape[1]\n                        ]));\n          }\n        }\n      } else if (x.rank === 3) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              xGrad =\n                  add(xGrad,\n                      slice(\n                          dy, [i * x.shape[0], j * x.shape[1], k * x.shape[2]],\n                          [x.shape[0], x.shape[1], x.shape[2]]));\n            }\n          }\n        }\n      } else if (x.rank === 4) {\n        for (let i = 0; i < reps[0]; ++i) {\n          for (let j = 0; j < reps[1]; ++j) {\n            for (let k = 0; k < reps[2]; ++k) {\n              for (let l = 0; l < reps[3]; ++l) {\n                xGrad =\n                    add(xGrad,\n                        slice(\n                            dy,\n                            [\n                              i * x.shape[0], j * x.shape[1], k * x.shape[2],\n                              l * x.shape[3]\n                            ],\n                            [x.shape[0], x.shape[1], x.shape[2], x.shape[3]]));\n              }\n            }\n          }\n        }\n      } else {\n        throw new Error(\n            `Gradient for tile operation is not implemented for rank-` +\n            `${x.rank} tensors yet.`);\n      }\n      return xGrad;\n    };\n    return {x: derX};\n  },\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Transpose, TransposeAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport * as axis_util from '../ops/axis_util';\nimport {transpose} from '../ops/transpose';\nimport {Tensor} from '../tensor';\n\nexport const transposeGradConfig: GradConfig = {\n  kernelName: Transpose,\n  gradFunc: (dy: Tensor, saved: Tensor[], attrs: NamedAttrMap) => {\n    const transposeAttrs: TransposeAttrs = attrs as {} as TransposeAttrs;\n    const {perm} = transposeAttrs;\n    const undoPerm = axis_util.getUndoAxesPermutation(perm);\n    return {x: () => transpose(dy, undoPerm)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Unpack, UnpackAttrs} from '../kernel_names';\nimport {GradConfig, NamedAttrMap} from '../kernel_registry';\nimport {stack} from '../ops/stack';\nimport {Tensor} from '../tensor';\n\nexport const unpackGradConfig: GradConfig = {\n  kernelName: Unpack,\n  gradFunc: (dy: Tensor[], saved: Tensor[], attrs: NamedAttrMap) => {\n    const unpackAttrs: UnpackAttrs = attrs as {} as UnpackAttrs;\n    const {axis} = unpackAttrs;\n    return {value: () => stack(dy, axis)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {UnsortedSegmentSum} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {expandDims} from '../ops/expand_dims';\nimport {gather} from '../ops/gather';\nimport {greaterEqual} from '../ops/greater_equal';\nimport {logicalAnd} from '../ops/logical_and';\nimport {maximum} from '../ops/maximum';\nimport {ones} from '../ops/ones';\nimport {scalar} from '../ops/scalar';\nimport {where} from '../ops/where';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor, Tensor1D} from '../tensor';\n\nexport const unsortedSegmentSumGradConfig: GradConfig = {\n  kernelName: UnsortedSegmentSum,\n  inputsToSave: ['segmentIds'],\n  gradFunc: (dy: Tensor, saved: Tensor[]) => {\n    const [segmentIds] = saved;\n\n    const derX = () => {\n      return gatherDropNegatives(dy, segmentIds as Tensor1D);\n    };\n    return {x: derX};\n  }\n};\n\nfunction gatherDropNegatives<T extends Tensor>(x: T, indices: Tensor1D) {\n  // Helper function for unsorted segment ops. Gathers params for\n  // positive segment ids and gathers 0 for inputs with negative segment id.\n  // Mirrors _GatherDropNegatives from tensorflow/python/ops/math_grad.py\n  const zeroClippedIndices = maximum(indices, zerosLike(indices));\n  const gathered = gather(x, zeroClippedIndices as Tensor1D);\n  let isPositive = greaterEqual(indices, scalar(0, 'int32'));\n  const numIters = gathered.rank - isPositive.rank;\n  for (let i = 0; i < numIters; ++i) {\n    isPositive = expandDims(isPositive, i + 1);\n  }\n  isPositive = logicalAnd(isPositive, ones(gathered.shape, 'bool'));\n  const zeroSlice = zerosLike(gathered);\n  return where(isPositive, gathered, zeroSlice);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ZerosLike} from '../kernel_names';\nimport {GradConfig} from '../kernel_registry';\nimport {zerosLike} from '../ops/zeros_like';\nimport {Tensor} from '../tensor';\n\nexport const zerosLikeGradConfig: GradConfig = {\n  kernelName: ZerosLike,\n  gradFunc: (dy: Tensor) => {\n    return {x: () => zerosLike(dy)};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {absGradConfig} from './gradients/Abs_grad';\nimport {acosGradConfig} from './gradients/Acos_grad';\nimport {acoshGradConfig} from './gradients/Acosh_grad';\nimport {addGradConfig} from './gradients/Add_grad';\nimport {addNGradConfig} from './gradients/AddN_grad';\nimport {argMaxGradConfig} from './gradients/ArgMax_grad';\nimport {argMinGradConfig} from './gradients/ArgMin_grad';\nimport {asinGradConfig} from './gradients/Asin_grad';\nimport {asinhGradConfig} from './gradients/Asinh_grad';\nimport {atan2GradConfig} from './gradients/Atan2_grad';\nimport {atanGradConfig} from './gradients/Atan_grad';\nimport {atanhGradConfig} from './gradients/Atanh_grad';\nimport {avgPool3DGradConfig} from './gradients/AvgPool3D_grad';\nimport {avgPoolGradConfig} from './gradients/AvgPool_grad';\nimport {batchMatMulGradConfig} from './gradients/BatchMatMul_grad';\nimport {batchToSpaceNDGradConfig} from './gradients/BatchToSpaceND_grad';\nimport {broadcastToGradConfig} from './gradients/BroadcastTo_grad';\nimport {castGradConfig} from './gradients/Cast_grad';\nimport {ceilGradConfig} from './gradients/Ceil_grad';\nimport {clipByValueGradConfig} from './gradients/ClipByValue_grad';\nimport {concatGradConfig} from './gradients/Concat_grad';\nimport {conv2DGradConfig} from './gradients/Conv2D_grad';\nimport {conv2DBackpropInputGradConfig} from './gradients/Conv2DBackpropInput_grad';\nimport {conv3DGradConfig} from './gradients/Conv3D_grad';\nimport {cosGradConfig} from './gradients/Cos_grad';\nimport {coshGradConfig} from './gradients/Cosh_grad';\nimport {cumsumGradConfig} from './gradients/Cumsum_grad';\nimport {depthwiseConv2dNativeGradConfig} from './gradients/DepthwiseConv2dNative_grad';\nimport {dilation2dGradConfig} from './gradients/Dilation2D_grad';\nimport {divGradConfig} from './gradients/Div_grad';\nimport {eluGradConfig} from './gradients/Elu_grad';\nimport {erfGradConfig} from './gradients/Erf_grad';\nimport {expGradConfig} from './gradients/Exp_grad';\nimport {expm1GradConfig} from './gradients/Expm1_grad';\nimport {floorGradConfig} from './gradients/Floor_grad';\nimport {floorDivGradConfig} from './gradients/FloorDiv_grad';\nimport {fusedBatchNormGradConfig} from './gradients/FusedBatchNorm_grad';\nimport {gatherGradConfig} from './gradients/GatherV2_grad';\nimport {greaterEqualGradConfig} from './gradients/GreaterEqual_grad';\nimport {identityGradConfig} from './gradients/Identity_grad';\nimport {isFiniteGradConfig} from './gradients/IsFinite_grad';\nimport {isInfGradConfig} from './gradients/IsInf_grad';\nimport {isNanGradConfig} from './gradients/IsNan_grad';\nimport {log1pGradConfig} from './gradients/Log1p_grad';\nimport {logGradConfig} from './gradients/Log_grad';\nimport {logSoftmaxGradConfig} from './gradients/LogSoftmax_grad';\nimport {lrnGradConfig} from './gradients/LRN_grad';\nimport {maxGradConfig} from './gradients/Max_grad';\nimport {maximumGradConfig} from './gradients/Maximum_grad';\nimport {maxPool3DGradConfig} from './gradients/MaxPool3D_grad';\nimport {maxPoolGradConfig} from './gradients/MaxPool_grad';\nimport {minGradConfig} from './gradients/Min_grad';\nimport {minimumGradConfig} from './gradients/Minimum_grad';\nimport {modGradConfig} from './gradients/Mod_grad';\nimport {multiplyGradConfig} from './gradients/Multiply_grad';\nimport {negateGradConfig} from './gradients/Negate_grad';\nimport {oneHotGradConfig} from './gradients/OneHot_grad';\nimport {onesLikeGradConfig} from './gradients/OnesLike_grad';\nimport {padV2GradConfig} from './gradients/PadV2_grad';\nimport {powGradConfig} from './gradients/Pow_grad';\nimport {preluGradConfig} from './gradients/Prelu_grad';\nimport {reciprocalGradConfig} from './gradients/Reciprocal_grad';\nimport {relu6GradConfig} from './gradients/Relu6_grad';\nimport {reluGradConfig} from './gradients/Relu_grad';\nimport {reshapeGradConfig} from './gradients/Reshape_grad';\nimport {resizeBilinearGradConfig} from './gradients/ResizeBilinear_grad';\nimport {resizeNearestNeighborGradConfig} from './gradients/ResizeNearestNeighbor_grad';\nimport {reverseGradConfig} from './gradients/Reverse_grad';\nimport {roundGradConfig} from './gradients/Round_grad';\nimport {rsqrtGradConfig} from './gradients/Rsqrt_grad';\nimport {selectV2PoolGradConfig} from './gradients/SelectV2_grad';\nimport {seluGradConfig} from './gradients/Selu_grad';\nimport {sigmoidGradConfig} from './gradients/Sigmoid_grad';\nimport {signGradConfig} from './gradients/Sign_grad';\nimport {sinGradConfig} from './gradients/Sin_grad';\nimport {sinhGradConfig} from './gradients/Sinh_grad';\nimport {sliceGradConfig} from './gradients/Slice_grad';\nimport {softmaxGradConfig} from './gradients/Softmax_grad';\nimport {softplusGradConfig} from './gradients/Softplus_grad';\nimport {spaceToBatchNDGradConfig} from './gradients/SpaceToBatchND_grad';\nimport {splitVGradConfig} from './gradients/SplitV_grad';\nimport {sqrtGradConfig} from './gradients/Sqrt_grad';\nimport {squareGradConfig} from './gradients/Square_grad';\nimport {squaredDifferenceGradConfig} from './gradients/SquaredDifference_grad';\nimport {stepGradConfig} from './gradients/Step_grad';\nimport {subGradConfig} from './gradients/Sub_grad';\nimport {sumGradConfig} from './gradients/Sum_grad';\nimport {tanGradConfig} from './gradients/Tan_grad';\nimport {tanhGradConfig} from './gradients/Tanh_grad';\nimport {tileGradConfig} from './gradients/Tile_grad';\nimport {transposeGradConfig} from './gradients/Transpose_grad';\nimport {unpackGradConfig} from './gradients/Unpack_grad';\nimport {unsortedSegmentSumGradConfig} from './gradients/UnsortedSegmentSum_grad';\nimport {zerosLikeGradConfig} from './gradients/ZerosLike_grad';\nimport {GradConfig} from './kernel_registry';\nimport {registerGradient} from './kernel_registry';\n\n// Export all kernel configs here so that the package can auto register them\nconst gradConfigs: GradConfig[] = [\n  absGradConfig,\n  acosGradConfig,\n  acoshGradConfig,\n  addGradConfig,\n  addNGradConfig,\n  argMaxGradConfig,\n  argMinGradConfig,\n  asinGradConfig,\n  asinhGradConfig,\n  atan2GradConfig,\n  atanGradConfig,\n  atanhGradConfig,\n  avgPool3DGradConfig,\n  avgPoolGradConfig,\n  batchMatMulGradConfig,\n  batchToSpaceNDGradConfig,\n  broadcastToGradConfig,\n  castGradConfig,\n  ceilGradConfig,\n  clipByValueGradConfig,\n  concatGradConfig,\n  conv2DBackpropInputGradConfig,\n  conv2DGradConfig,\n  conv3DGradConfig,\n  cosGradConfig,\n  coshGradConfig,\n  cumsumGradConfig,\n  depthwiseConv2dNativeGradConfig,\n  dilation2dGradConfig,\n  divGradConfig,\n  eluGradConfig,\n  erfGradConfig,\n  expGradConfig,\n  expm1GradConfig,\n  floorDivGradConfig,\n  floorGradConfig,\n  fusedBatchNormGradConfig,\n  gatherGradConfig,\n  greaterEqualGradConfig,\n  identityGradConfig,\n  isFiniteGradConfig,\n  isInfGradConfig,\n  isNanGradConfig,\n  log1pGradConfig,\n  logGradConfig,\n  logSoftmaxGradConfig,\n  lrnGradConfig,\n  maxGradConfig,\n  maxGradConfig,\n  maximumGradConfig,\n  maxPool3DGradConfig,\n  maxPoolGradConfig,\n  minGradConfig,\n  minimumGradConfig,\n  modGradConfig,\n  multiplyGradConfig,\n  negateGradConfig,\n  oneHotGradConfig,\n  onesLikeGradConfig,\n  padV2GradConfig,\n  padV2GradConfig,\n  powGradConfig,\n  preluGradConfig,\n  reciprocalGradConfig,\n  relu6GradConfig,\n  reluGradConfig,\n  reshapeGradConfig,\n  resizeBilinearGradConfig,\n  resizeNearestNeighborGradConfig,\n  reverseGradConfig,\n  roundGradConfig,\n  rsqrtGradConfig,\n  selectV2PoolGradConfig,\n  seluGradConfig,\n  sigmoidGradConfig,\n  signGradConfig,\n  sinGradConfig,\n  sinhGradConfig,\n  sliceGradConfig,\n  softmaxGradConfig,\n  softplusGradConfig,\n  spaceToBatchNDGradConfig,\n  spaceToBatchNDGradConfig,\n  splitVGradConfig,\n  splitVGradConfig,\n  sqrtGradConfig,\n  squaredDifferenceGradConfig,\n  squareGradConfig,\n  stepGradConfig,\n  subGradConfig,\n  sumGradConfig,\n  tanGradConfig,\n  tanhGradConfig,\n  tileGradConfig,\n  transposeGradConfig,\n  unpackGradConfig,\n  unsortedSegmentSumGradConfig,\n  zerosLikeGradConfig\n];\n\nfor (const gradientConfig of gradConfigs) {\n  registerGradient(gradientConfig);\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {abs} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    abs<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.abs = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return abs(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acos = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {acosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    acosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.acosh = function<T extends Tensor>(this: T) {\n  this.throwIfDisposed();\n  return acosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {addStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    addStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.addStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return addStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {add} from '../../ops/add';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    add<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.add = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return add(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {all} from '../../ops/all';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    all<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.all = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return all(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {any} from '../../ops/any';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    any<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.any = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return any(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMax} from '../../ops/arg_max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMax<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMax = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return argMax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {argMin} from '../../ops/arg_min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    argMin<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.argMin = function<T extends Tensor>(axis: number): T {\n  this.throwIfDisposed();\n  return argMin(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Scalar, Tensor} from '../../tensor';\nimport {Rank} from '../../types';\nimport {assert} from '../../util';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asScalar<T extends Tensor>(): Scalar;\n  }\n}\n\n/** Converts a size-1 `tf.Tensor` to a `tf.Scalar`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asScalar = function<T extends Tensor>(this: T): Scalar {\n  this.throwIfDisposed();\n  assert(this.size === 1, () => 'The array must have only 1 element.');\n  return reshape(this, []);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asType<T extends Tensor>(this: T, dtype: DataType): T;\n  }\n}\n\n/**\n * Casts a `tf.Tensor` to a specified dtype.\n *\n * @param dtype Data-type to cast the tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.asType = function<T extends Tensor>(\n    this: T, dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast<T>(this, dtype);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as1D<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Converts a `tf.Tensor` to a `tf.Tensor1D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as1D = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as2D<T extends Tensor>(rows: number, columns: number): Tensor2D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor2D`.\n *\n * @param rows Number of rows in `tf.Tensor2D`.\n * @param columns Number of columns in `tf.Tensor2D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as2D = function<T extends Tensor>(\n    rows: number, columns: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as3D<T extends Tensor>(rows: number, columns: number, depth: number):\n        Tensor3D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor3D`.\n *\n * @param rows Number of rows in `tf.Tensor3D`.\n * @param columns Number of columns in `tf.Tensor3D`.\n * @param depth Depth of `tf.Tensor3D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as3D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as4D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number): Tensor4D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor4D`.\n *\n * @param rows Number of rows in `tf.Tensor4D`.\n * @param columns Number of columns in `tf.Tensor4D`.\n * @param depth Depth of `tf.Tensor4D`.\n * @param depth2 4th dimension of `tf.Tensor4D`.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as4D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    as5D<T extends Tensor>(\n        rows: number, columns: number, depth: number, depth2: number,\n        depth3: number): Tensor5D;\n  }\n}\n\n/**\n * Converts a `tf.Tensor` to a `tf.Tensor5D`.\n *\n * @param rows Number of rows in `tf.Tensor5D`.\n * @param columns Number of columns in `tf.Tensor5D`.\n * @param depth Depth of `tf.Tensor5D`.\n * @param depth2 4th dimension of `tf.Tensor5D`.\n * @param depth3 5th dimension of 'tf.Tensor5D'\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.as5D = function<T extends Tensor>(\n    rows: number, columns: number, depth: number, depth2: number,\n    depth3: number): T {\n  this.throwIfDisposed();\n  return reshape(this, [rows, columns, depth, depth2, depth3]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {asinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    asinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.asinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return asinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {atan2} from '../../ops/atan2';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atan2<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.atan2 = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return atan2(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {atanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    atanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.atanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return atanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {avgPool} from '../../ops/avg_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    avgPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.avgPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return avgPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchToSpaceND} from '../../ops/batch_to_space_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchToSpaceND<R extends Rank>(blockShape: number[], crops: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.batchToSpaceND = function<R extends Rank>(\n    blockShape: number[], crops: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return batchToSpaceND(this, blockShape, crops);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {batchNorm} from '../../ops/batchnorm';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    batchNorm<T extends Tensor>(\n        mean: Tensor<R>|Tensor1D|TensorLike,\n        variance: Tensor<R>|Tensor1D|TensorLike,\n        offset?: Tensor<R>|Tensor1D|TensorLike,\n        scale?: Tensor<R>|Tensor1D|TensorLike,\n        varianceEpsilon?: number): Tensor<R>;\n  }\n}\n\nTensor.prototype.batchNorm = function<R extends Rank>(\n    mean: Tensor<R>|Tensor1D|TensorLike,\n    variance: Tensor<R>|Tensor1D|TensorLike,\n    offset?: Tensor<R>|Tensor1D|TensorLike,\n    scale?: Tensor<R>|Tensor1D|TensorLike,\n    varianceEpsilon?: number): Tensor<R> {\n  this.throwIfDisposed();\n  return batchNorm(this, mean, variance, offset, scale, varianceEpsilon);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {broadcastTo} from '../../ops/broadcast_to';\nimport {Tensor} from '../../tensor';\nimport {Rank, ShapeMap} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    broadcastTo<R extends Rank>(shape: ShapeMap[R]): Tensor<R>;\n  }\n}\n\nTensor.prototype.broadcastTo = function<R extends Rank>(shape: ShapeMap[R]):\n    Tensor<R> {\n  this.throwIfDisposed();\n  return broadcastTo(this, shape);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {DataType, Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cast<T extends Tensor>(dtype: DataType): T;\n  }\n}\n\nTensor.prototype.cast = function<T extends Tensor>(dtype: DataType): T {\n  this.throwIfDisposed();\n  return cast(this, dtype) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ceil} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ceil<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.ceil = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return ceil(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {clipByValue} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    clipByValue<T extends Tensor>(min: number, max: number): Tensor;\n  }\n}\n\nTensor.prototype.clipByValue = function<T extends Tensor>(\n    min: number, max: number): T {\n  this.throwIfDisposed();\n  return clipByValue(this, min, max) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {concat} from '../../ops/concat';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    concat<T extends Tensor>(tensors: T|Array<T|TensorLike>, axis?: number): T;\n  }\n}\n\nTensor.prototype.concat = function<T extends Tensor>(\n    x: T|Array<T|TensorLike>, axis?: number): T {\n  this.throwIfDisposed();\n  if (x instanceof Tensor) {\n    x = [x];\n  }\n  return concat([this, ...x], axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv1d} from '../../ops/conv1d';\nimport {ExplicitPadding} from '../../ops/conv_util';\nimport {Tensor, Tensor2D, Tensor3D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv1d<T extends Tensor2D|Tensor3D>(\n        filter: Tensor3D|TensorLike3D, stride: number,\n        pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n        dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv1d = function<T extends Tensor2D|Tensor3D>(\n    filter: Tensor3D|TensorLike3D, stride: number,\n    pad: 'valid'|'same'|number|ExplicitPadding, dataFormat?: 'NWC'|'NCW',\n    dilation?: number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv1d(\n             this, filter, stride, pad, dataFormat, dilation,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2dTranspose} from '../../ops/conv2d_transpose';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2dTranspose<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D,\n        outputShape: [number, number, number, number]|[number, number, number],\n        strides: [number, number]|number, pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2dTranspose = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D,\n    outputShape: [number, number, number, number]|[number, number, number],\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2dTranspose(\n             this, filter, outputShape, strides, pad, dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {conv2d} from '../../ops/conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    conv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.conv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return conv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cos} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cos<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cos = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cos(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cosh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cosh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.cosh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cosh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {cumsum} from '../../ops/cumsum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    cumsum<R extends Rank>(\n        axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R>;\n  }\n}\n\nTensor.prototype.cumsum = function<R extends Rank>(\n    axis?: number, exclusive?: boolean, reverse?: boolean): Tensor<R> {\n  this.throwIfDisposed();\n  return cumsum(this, axis, exclusive, reverse);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthToSpace} from '../../ops/depth_to_space';\nimport {Tensor, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthToSpace<T extends Tensor4D>(\n        blockSize: number, dataFormat: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.depthToSpace = function<T extends Tensor4D>(\n    blockSize: number, dataFormat: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return depthToSpace(this, blockSize, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {deprecationWarn} from '../../globals';\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2D<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\n/**\n * @deprecated Use `depthwiseConv2d` instead.\n */\nTensor.prototype.depthwiseConv2D = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  deprecationWarn('depthwiseConv2D is deprecated, use depthwiseConv2d instead');\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {depthwiseConv2d} from '../../ops/depthwise_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    depthwiseConv2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n        pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n        dilations?: [number, number]|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.depthwiseConv2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor4D|TensorLike4D, strides: [number, number]|number,\n    pad: 'valid'|'same'|number, dataFormat?: 'NHWC'|'NCHW',\n    dilations?: [number, number]|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return depthwiseConv2d(\n             this, filter, strides, pad, dataFormat, dilations,\n             dimRoundingMode) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dilation2d} from '../../ops/dilation2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike3D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dilation2d<T extends Tensor3D|Tensor4D>(\n        filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n        pad: 'valid'|'same', dilations?: [number, number]|number,\n        dataFormat?: 'NHWC'): T;\n  }\n}\n\nTensor.prototype.dilation2d = function<T extends Tensor3D|Tensor4D>(\n    filter: Tensor3D|TensorLike3D, strides: [number, number]|number,\n    pad: 'valid'|'same', dilations?: [number, number]|number,\n    dataFormat?: 'NHWC'): T {\n  this.throwIfDisposed();\n  return dilation2d(this, filter, strides, pad, dilations, dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {divNoNan} from '../../ops/div_no_nan';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divNoNan<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.divNoNan = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return divNoNan(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {divStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    divStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.divStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return divStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {div} from '../../ops/div';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    div<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.div = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return div(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {dot} from '../../ops/dot';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    dot<T extends Tensor>(b: Tensor|TensorLike): Tensor;\n  }\n}\n\nTensor.prototype.dot = function<T extends Tensor>(b: T|TensorLike): Tensor {\n  this.throwIfDisposed();\n  return dot(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {elu} from '../../ops/elu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    elu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.elu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return elu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {equalStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equalStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.equalStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return equalStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {equal} from '../../ops/equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    equal<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.equal = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return equal(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {erf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    erf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.erf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return erf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {exp} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    exp<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.exp = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return exp(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {expandDims} from '../../ops/expand_dims';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expandDims<T extends Tensor>(axis?: number): T;\n  }\n}\n\nTensor.prototype.expandDims = function<T extends Tensor>(axis?: number): T {\n  this.throwIfDisposed();\n  return expandDims(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {expm1} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    expm1<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.expm1 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return expm1(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {fft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    fft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.fft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return fft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    flatten<T extends Tensor>(): Tensor1D;\n  }\n}\n\n/** Flatten a Tensor to a 1D array.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.flatten = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return reshape(this, [this.size]) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {floor} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floor<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.floor = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return floor(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {floorDiv} from '../../ops/floorDiv';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    floorDiv<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.floorDiv = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return floorDiv(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {gather} from '../../ops/gather';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    gather<T extends Tensor>(\n        this: T, indices: Tensor|TensorLike, axis?: number): T;\n  }\n}\n\nTensor.prototype.gather = function<T extends Tensor>(\n    this: T, indices: Tensor|TensorLike, axis?: number): T {\n  this.throwIfDisposed();\n  return gather(this, indices, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return greaterEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greaterEqual} from '../../ops/greater_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greaterEqual = function<T extends Tensor>(b: Tensor|\n                                                           TensorLike): T {\n  this.throwIfDisposed();\n  return greaterEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {greaterStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greaterStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.greaterStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return greaterStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {greater} from '../../ops/greater';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    greater<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.greater = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return greater(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {ifft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    ifft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.ifft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return ifft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {irfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    irfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.irfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return irfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isFinite} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isFinite<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isFinite = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isFinite(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isInf} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isInf<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isInf = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isInf(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {isNaN} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    isNaN<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.isNaN = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return isNaN(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {leakyRelu} from '../../ops/leaky_relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    leakyRelu<T extends Tensor>(alpha: number): T;\n  }\n}\n\nTensor.prototype.leakyRelu = function<T extends Tensor>(\n    this: T, alpha: number): T {\n  this.throwIfDisposed();\n  return leakyRelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.lessEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {lessEqual} from '../../ops/less_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessEqual = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return lessEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {lessStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    lessStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.lessStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike) {\n  this.throwIfDisposed();\n  return lessStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {less} from '../../ops/less';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    less<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.less = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return less(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {localResponseNormalization} from '../../ops/local_response_normalization';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    localResponseNormalization<T extends Tensor>(\n        depthRadius?: number, bias?: number, alpha?: number, beta?: number): T;\n  }\n}\n\nTensor.prototype.localResponseNormalization = function<T extends Tensor>(\n    depthRadius?: number, bias?: number, alpha?: number, beta?: number): T {\n  this.throwIfDisposed();\n  return localResponseNormalization(this, depthRadius, bias, alpha, beta) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.logSigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return logSigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {logSoftmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSoftmax<T extends Tensor>(this: T, axis?: number): T;\n  }\n}\n\nTensor.prototype.logSoftmax = function<T extends Tensor>(\n    this: T, axis?: number): T {\n  this.throwIfDisposed();\n  return logSoftmax(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logSumExp} from '../../ops/log_sum_exp';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logSumExp<T extends Tensor>(\n        this: T, axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.logSumExp = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return logSumExp(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {log1p} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    log1p<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.log1p = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return log1p(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalAnd} from '../../ops/logical_and';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalAnd<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalAnd = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalAnd(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalNot} from '../../ops/logical_not';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalNot<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.logicalNot = function<T extends Tensor>(): T {\n  this.throwIfDisposed();\n  return logicalNot(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalOr} from '../../ops/logical_or';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalOr<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalOr = function<T extends Tensor>(b: Tensor|\n                                                        TensorLike): T {\n  this.throwIfDisposed();\n  return logicalOr(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {logicalXor} from '../../ops/logical_xor';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    logicalXor<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.logicalXor = function<T extends Tensor>(b: Tensor|\n                                                         TensorLike): T {\n  this.throwIfDisposed();\n  return logicalXor(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {matMul} from '../../ops/mat_mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    matMul<T extends Tensor>(\n        b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T;\n  }\n}\n\nTensor.prototype.matMul = function<T extends Tensor>(\n    this: T, b: T|TensorLike, transposeA?: boolean, transposeB?: boolean): T {\n  this.throwIfDisposed();\n  return matMul(this, b, transposeA, transposeB);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maxPool} from '../../ops/max_pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maxPool<T extends Tensor3D|Tensor4D>(\n        filterSize: [number, number]|number, strides: [number, number]|number,\n        pad: 'valid'|'same'|number,\n        dimRoundingMode?: 'floor'|'round'|'ceil'): T;\n  }\n}\n\nTensor.prototype.maxPool = function<T extends Tensor3D|Tensor4D>(\n    this: T, filterSize: [number, number]|number,\n    strides: [number, number]|number, pad: 'valid'|'same'|number,\n    dimRoundingMode?: 'floor'|'round'|'ceil'): T {\n  this.throwIfDisposed();\n  return maxPool(this, filterSize, strides, pad, dimRoundingMode);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {max} from '../../ops/max';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    max<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.max = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return max(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {maximumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.maximumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return maximumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {maximum} from '../../ops/maximum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    maximum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.maximum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return maximum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mean} from '../../ops/mean';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mean<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.mean = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return mean(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {min} from '../../ops/min';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    min<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.min = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return min(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {minimumStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimumStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.minimumStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return minimumStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {minimum} from '../../ops/minimum';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    minimum<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.minimum = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return minimum(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {modStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    modStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.modStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return modStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mod} from '../../ops/mod';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mod<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mod = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mod(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {mulStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mulStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.mulStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return mulStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {mul} from '../../ops/mul';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    mul<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.mul = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return mul(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {neg} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    neg<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.neg = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return neg(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {norm} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    norm<T extends Tensor>(\n        ord?: number|'euclidean'|'fro', axis?: number|number[],\n        keepDims?: boolean): Tensor;\n  }\n}\n\nTensor.prototype.norm = function<T extends Tensor>(\n    ord?: number|'euclidean'|'fro', axis?: number|number[],\n    keepDims?: boolean) {\n  this.throwIfDisposed();\n  return norm(this, ord, axis, keepDims) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {notEqualStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqualStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.notEqualStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return notEqualStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {notEqual} from '../../ops/not_equal';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    notEqual<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.notEqual = function<T extends Tensor>(b: Tensor|\n                                                       TensorLike): T {\n  this.throwIfDisposed();\n  return notEqual(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {oneHot} from '../../ops/one_hot';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    oneHot(depth: number, onValue: number, offValue: number): Tensor;\n  }\n}\n\nTensor.prototype.oneHot = function(\n    depth: number, onValue = 1, offValue = 0): Tensor {\n  this.throwIfDisposed();\n  return oneHot(this, depth, onValue, offValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {onesLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    onesLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.onesLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return onesLike(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {pad} from '../../ops/pad';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pad<T extends Tensor>(\n        paddings: Array<[number, number]>, constantValue?: number): T;\n  }\n}\n\nTensor.prototype.pad = function<T extends Tensor>(\n    this: T, paddings: Array<[number, number]>, constantValue: number): T {\n  this.throwIfDisposed();\n  return pad(this, paddings, constantValue);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pool} from '../../ops/pool';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pool<T extends Tensor3D|Tensor4D>(\n        windowShape: [number, number]|number, poolingType: 'avg'|'max',\n        padding: 'valid'|'same'|number, diationRate?: [number, number]|number,\n        strides?: [number, number]|number): T;\n  }\n}\n\nTensor.prototype.pool = function<T extends Tensor3D|Tensor4D>(\n    this: T, windowShape: [number, number]|number, poolingType: 'max'|'avg',\n    padding: 'valid'|'same'|number, dilationRate?: [number, number]|number,\n    strides?: [number, number]|number): T {\n  this.throwIfDisposed();\n  return pool(this, windowShape, poolingType, padding, dilationRate, strides);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {powStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    powStrict<T extends Tensor>(exp: Tensor): Tensor;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.powStrict = function<T extends Tensor>(exp: Tensor): T {\n  this.throwIfDisposed();\n  return powStrict(this, exp) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {pow} from '../../ops/pow';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    pow<T extends Tensor>(exp: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.pow = function<T extends Tensor>(exp: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return pow(this, exp);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prelu} from '../../ops/prelu';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prelu<T extends Tensor>(alpha: T|TensorLike): T;\n  }\n}\n\nTensor.prototype.prelu = function<T extends Tensor>(\n    this: T, alpha: T|TensorLike): T {\n  this.throwIfDisposed();\n  return prelu(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {prod} from '../../ops/prod';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    prod<T extends Tensor>(this: T, axis?: number|number[], keepDims?: boolean):\n        T;\n  }\n}\n\nTensor.prototype.prod = function<T extends Tensor>(\n    this: T, axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return prod(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {reciprocal} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reciprocal<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.reciprocal = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return reciprocal(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu} from '../../ops/relu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {relu6} from '../../ops/relu6';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    relu6<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.relu6 = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return relu6(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshapeAs<T extends Tensor>(x: T): T;\n  }\n}\n\n/**\n * Reshapes the tensor into the shape of the provided tensor.\n *\n * @param x The tensor of required shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.reshapeAs = function<T extends Tensor>(x: T): T {\n  this.throwIfDisposed();\n  return reshape(this, x.shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reshape} from '../../ops/reshape';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reshape<T extends Tensor>(shape: number[]): T;\n  }\n}\n\nTensor.prototype.reshape = function<T extends Tensor>(shape: number[]): T {\n  this.throwIfDisposed();\n  return reshape(this, shape) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeBilinear} from '../../ops/image/resize_bilinear';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeBilinear<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeBilinear = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeBilinear(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {resizeNearestNeighbor} from '../../ops/image/resize_nearest_neighbor';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    resizeNearestNeighbor<T extends Tensor3D|Tensor4D>(\n        newShape2D: [number, number], alignCorners?: boolean): T;\n  }\n}\n\nTensor.prototype.resizeNearestNeighbor = function<T extends Tensor3D|Tensor4D>(\n    this: T, newShape2D: [number, number], alignCorners?: boolean): T {\n  this.throwIfDisposed();\n  return resizeNearestNeighbor(this, newShape2D, alignCorners);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {reverse} from '../../ops/reverse';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    reverse<T extends Tensor>(this: T, axis?: number|number[]): T;\n  }\n}\n\nTensor.prototype.reverse = function<T extends Tensor>(\n    this: T, axis?: number|number[]): T {\n  this.throwIfDisposed();\n  return reverse(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rfft} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rfft<T extends Tensor>(this: Tensor): Tensor;\n  }\n}\n\nTensor.prototype.rfft = function<T extends Tensor>(this: Tensor): T {\n  this.throwIfDisposed();\n  return rfft(this) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {round} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    round<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.round = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return round(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {rsqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    rsqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.rsqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return rsqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {selu} from '../../ops/selu';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    selu<T extends Tensor>(): T;\n  }\n}\n\nTensor.prototype.selu = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return selu(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {separableConv2d} from '../../ops/separable_conv2d';\nimport {Tensor, Tensor3D, Tensor4D} from '../../tensor';\nimport {Rank, TensorLike, TensorLike4D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    separableConv2d<T extends Tensor3D|Tensor4D>(\n        depthwiseFilter: Tensor4D|TensorLike4D,\n        pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n        pad: 'valid'|'same', dilation?: [number, number]|number,\n        dataFormat?: 'NHWC'|'NCHW'): T;\n  }\n}\n\nTensor.prototype.separableConv2d = function<T extends Tensor3D|Tensor4D>(\n    depthwiseFilter: Tensor4D|TensorLike4D,\n    pointwiseFilter: Tensor4D|TensorLike, strides: [number, number]|number,\n    pad: 'valid'|'same', dilation?: [number, number]|number,\n    dataFormat?: 'NHWC'|'NCHW'): T {\n  this.throwIfDisposed();\n  return separableConv2d(\n             this, depthwiseFilter, pointwiseFilter, strides, pad, dilation,\n             dataFormat) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sigmoid} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sigmoid<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sigmoid = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sigmoid(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sign} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sign<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sign = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sign(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sin} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sin<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sin = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sin(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sinh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sinh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sinh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sinh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {slice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    slice<T extends Tensor>(\n        this: T, begin: number|number[], size?: number|number[]): T;\n  }\n}\n\nTensor.prototype.slice = function<T extends Tensor>(\n    this: T, begin: number|number[], size?: number|number[]): T {\n  this.throwIfDisposed();\n  return slice(this, begin, size);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softmax} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softmax<T extends Tensor>(this: T, dim?: number): T;\n  }\n}\n\nTensor.prototype.softmax = function<T extends Tensor>(this: T, dim: number): T {\n  this.throwIfDisposed();\n  return softmax(this, dim);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {softplus} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    softplus<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.softplus = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return softplus(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {spaceToBatchND} from '../../ops/space_to_batch_nd';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    spaceToBatchND<R extends Rank>(blockShape: number[], paddings: number[][]):\n        Tensor<R>;\n  }\n}\n\nTensor.prototype.spaceToBatchND = function<R extends Rank>(\n    blockShape: number[], paddings: number[][]): Tensor<R> {\n  this.throwIfDisposed();\n  return spaceToBatchND(this, blockShape, paddings);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {split} from '../../ops/split';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    split<T extends Tensor>(numOrSizeSplits: number[]|number, axis?: number):\n        T[];\n  }\n}\n\nTensor.prototype.split = function<T extends Tensor>(\n    numOrSizeSplits: number[]|number, axis?: number): T[] {\n  this.throwIfDisposed();\n  return split(this, numOrSizeSplits, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {sqrt} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sqrt<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.sqrt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return sqrt(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {square} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    square<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.square = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return square(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {squaredDifference} from '../../ops/squared_difference';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifference<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.squaredDifference = function<T extends Tensor>(b: Tensor|\n                                                                TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifference(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {squaredDifferenceStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squaredDifferenceStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.squaredDifferenceStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return squaredDifferenceStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {squeeze} from '../../ops/squeeze';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    squeeze<T extends Tensor>(axis?: number[]): T;\n  }\n}\n\nTensor.prototype.squeeze = function<T extends Tensor>(axis?: number[]): T {\n  this.throwIfDisposed();\n  return squeeze(this, axis);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {stack} from '../../ops/stack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stack<T extends Tensor>(x: Tensor|Tensor[], axis?: number): T;\n  }\n}\n\nTensor.prototype.stack = function<T extends Tensor>(\n    x: Tensor|Tensor[], axis?: number): T {\n  this.throwIfDisposed();\n  const tensorsToBeStacked = x instanceof Tensor ? [this, x] : [this, ...x];\n  return stack(tensorsToBeStacked, axis) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {step} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    step<T extends Tensor>(this: T, alpha?: number): T;\n  }\n}\n\nTensor.prototype.step = function<T extends Tensor>(this: T, alpha?: number) {\n  this.throwIfDisposed();\n  return step(this, alpha);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {stridedSlice} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    stridedSlice<T extends Tensor>(\n        this: Tensor, begin: number[], end: number[], strides: number[],\n        beginMask?: number, endMask?: number, ellipsisMask?: number,\n        newAxisMask?: number, shrinkAxisMask?: number): Tensor;\n  }\n}\n\nTensor.prototype.stridedSlice = function<T extends Tensor>(\n    this: Tensor, begin: number[], end: number[], strides: number[],\n    beginMask?: number, endMask?: number, ellipsisMask?: number,\n    newAxisMask?: number, shrinkAxisMask?: number): T {\n  this.throwIfDisposed();\n  return stridedSlice(\n             this, begin, end, strides, beginMask, endMask, ellipsisMask,\n             newAxisMask, shrinkAxisMask) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {subStrict} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    subStrict<T extends Tensor>(this: T, x: T|TensorLike): T;\n  }\n}\n\n/**\n * @deprecated strict variants of ops have been deprecated\n */\nTensor.prototype.subStrict = function<T extends Tensor>(\n    this: T, x: T|TensorLike): T {\n  this.throwIfDisposed();\n  return subStrict(this, x);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sub} from '../../ops/sub';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sub<T extends Tensor>(b: Tensor|TensorLike): T;\n  }\n}\n\nTensor.prototype.sub = function<T extends Tensor>(b: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return sub(this, b);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {sum} from '../../ops/sum';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    sum<T extends Tensor>(axis?: number|number[], keepDims?: boolean): T;\n  }\n}\n\nTensor.prototype.sum = function<T extends Tensor>(\n    axis?: number|number[], keepDims?: boolean): T {\n  this.throwIfDisposed();\n  return sum(this, axis, keepDims);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tan} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tan<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tan = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tan(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {tanh} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tanh<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.tanh = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return tanh(this);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tile} from '../../ops/tile';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    tile<T extends Tensor>(b: number[]): T;\n  }\n}\n\nTensor.prototype.tile = function<T extends Tensor>(reps: number[]): T {\n  this.throwIfDisposed();\n  return tile(this, reps) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toBool<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `bool`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toBool = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'bool');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toFloat<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `float32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toFloat = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'float32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {cast} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    toInt<T extends Tensor>(this: T): T;\n  }\n}\n\n/** Casts the array to type `int32`\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nTensor.prototype.toInt = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return cast<T>(this, 'int32');\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {topk} from '../../ops/topk';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    topk<T extends Tensor>(this: T, k?: number, sorted?: boolean):\n        {values: T, indices: T};\n  }\n}\n\nTensor.prototype.topk = function<T extends Tensor>(\n    this: T, k?: number, sorted?: boolean): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return topk(this, k, sorted) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {transpose} from '../../ops/transpose';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    transpose<T extends Tensor>(perm?: number[]): T;\n  }\n}\n\nTensor.prototype.transpose = function<T extends Tensor>(\n    this: T, perm?: number[]): T {\n  this.throwIfDisposed();\n  return transpose(this, perm);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unique} from '../../ops/unique';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unique<T extends Tensor>(this: T, axis?: number): {values: T, indices: T};\n  }\n}\n\nTensor.prototype.unique = function<T extends Tensor>(\n    this: T, axis?: number): {values: T, indices: T} {\n  this.throwIfDisposed();\n  return unique(this, axis) as {values: T, indices: T};\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {unsortedSegmentSum} from '../../ops/unsorted_segment_sum';\nimport {Tensor, Tensor1D} from '../../tensor';\nimport {Rank, TensorLike1D} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unsortedSegmentSum<T extends Tensor>(\n        this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T;\n  }\n}\n\nTensor.prototype.unsortedSegmentSum = function<T extends Tensor>(\n    this: T, segmentIds: Tensor1D|TensorLike1D, numSegments: number): T {\n  this.throwIfDisposed();\n  return unsortedSegmentSum(this, segmentIds, numSegments);\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {unstack} from '../../ops/unstack';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    unstack<T extends Tensor>(axis?: number): T[];\n  }\n}\n\nTensor.prototype.unstack = function<T extends Tensor>(axis?: number): T[] {\n  this.throwIfDisposed();\n  return unstack(this, axis) as T[];\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {where} from '../../ops/where';\nimport {Tensor} from '../../tensor';\nimport {Rank, TensorLike} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    where<T extends Tensor>(condition: Tensor|TensorLike, x: Tensor|TensorLike):\n        T;\n  }\n}\n\nTensor.prototype.where = function<T extends Tensor>(\n    condition: Tensor|TensorLike, x: Tensor|TensorLike): T {\n  this.throwIfDisposed();\n  return where(condition, this, x) as T;\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// TODO update import path once op is modularized.\nimport {zerosLike} from '../../ops/ops';\nimport {Tensor} from '../../tensor';\nimport {Rank} from '../../types';\n\ndeclare module '../../tensor' {\n  interface Tensor<R extends Rank = Rank> {\n    zerosLike<T extends Tensor>(this: T): T;\n  }\n}\n\nTensor.prototype.zerosLike = function<T extends Tensor>(this: T): T {\n  this.throwIfDisposed();\n  return zerosLike(this);\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {backend} from '@tensorflow/tfjs-core';\nimport {DataFormat} from '../keras_format/common';\n\nlet _epsilon: number;\n\n/**\n * Returns the value of the fuzz factor used in numeric expressions.\n */\nexport function epsilon() {\n  if (_epsilon == null) {\n    _epsilon = backend().epsilon();\n  }\n  return _epsilon;\n}\n\n/**\n * Sets the value of the fuzz factor used in numeric expressions.\n * @param e New value of epsilon.\n */\nexport function setEpsilon(e: number) {\n  _epsilon = e;\n}\n\n/**\n * Returns the default image data format convention.\n */\nexport function imageDataFormat(): DataFormat {\n  return 'channelsLast';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Explicit error types.\n *\n * See the following link for more information about why the code includes\n * calls to setPrototypeOf:\n *\n * https://github.com/Microsoft/TypeScript-wiki/blob/master/Breaking-Changes.md#extending-built-ins-like-error-array-and-map-may-no-longer-work\n */\n// tslint:enable\n\n/**\n * Equivalent of Python's AttributeError.\n */\nexport class AttributeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AttributeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's RuntimeError.\n */\nexport class RuntimeError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, RuntimeError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's ValueError.\n */\nexport class ValueError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, ValueError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's NotImplementedError.\n */\nexport class NotImplementedError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, NotImplementedError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's AssertionError.\n */\nexport class AssertionError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, AssertionError.prototype);\n  }\n}\n\n/**\n * Equivalent of Python's IndexError.\n */\nexport class IndexError extends Error {\n  constructor(message?: string) {\n    super(message);\n    // Set the prototype explicitly.\n    Object.setPrototypeOf(this, IndexError.prototype);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {DataType, fused, serialization, util} from '@tensorflow/tfjs-core';\n\nimport {AssertionError, ValueError} from '../errors';\n\n// tslint:enable\n\n/**\n * If `value` is an Array, equivalent to Python's `value * numValues`.\n * If `value` is not an Array, equivalent to Python's `[value] * numValues`\n */\n// tslint:disable-next-line:no-any\nexport function pyListRepeat(value: any, numValues: number): any[] {\n  if (Array.isArray(value)) {\n    // tslint:disable-next-line:no-any\n    let newArray: any[] = [];\n    for (let i = 0; i < numValues; i++) {\n      newArray = newArray.concat(value);\n    }\n    return newArray;\n  } else {\n    const newArray = new Array(numValues);\n    newArray.fill(value);\n    return newArray;\n  }\n}\n\nexport function assert(val: boolean, message?: string): void {\n  if (!val) {\n    throw new AssertionError(message);\n  }\n}\n\n/**\n * Count the number of elements of the `array` that are equal to `reference`.\n */\nexport function count<T>(array: T[], refernce: T) {\n  let counter = 0;\n  for (const item of array) {\n    if (item === refernce) {\n      counter++;\n    }\n  }\n  return counter;\n}\n\n/**\n * If an array is of length 1, just return the first element. Otherwise, return\n * the full array.\n * @param tensors\n */\nexport function singletonOrArray<T>(xs: T[]): T|T[] {\n  if (xs.length === 1) {\n    return xs[0];\n  }\n  return xs;\n}\n\n/**\n * Normalizes a list/tensor into a list.\n *\n * If a tensor is passed, we return\n * a list of size 1 containing the tensor.\n *\n * @param x target object to be normalized.\n */\n// tslint:disable-next-line:no-any\nexport function toList(x: any): any[] {\n  if (Array.isArray(x)) {\n    return x;\n  }\n  return [x];\n}\n\n/**\n * Generate a UID for a list\n */\n// tslint:disable-next-line:no-any\nexport function objectListUid(objs: any|any[]): string {\n  const objectList = toList(objs);\n  let retVal = '';\n  for (const obj of objectList) {\n    if (obj.id == null) {\n      throw new ValueError(\n          `Object ${obj} passed to objectListUid without an id`);\n    }\n    if (retVal !== '') {\n      retVal = retVal + ', ';\n    }\n    retVal = `${retVal}${Math.abs(obj.id)}`;\n  }\n  return retVal;\n}\n/**\n * Converts string to snake-case.\n * @param name\n */\nexport function toSnakeCase(name: string): string {\n  const intermediate = name.replace(/(.)([A-Z][a-z0-9]+)/g, '$1_$2');\n  const insecure =\n      intermediate.replace(/([a-z])([A-Z])/g, '$1_$2').toLowerCase();\n  /*\n   If the class is private the name starts with \"_\" which is not secure\n   for creating scopes. We prefix the name with \"private\" in this case.\n   */\n  if (insecure[0] !== '_') {\n    return insecure;\n  }\n  return 'private' + insecure;\n}\n\nexport function toCamelCase(identifier: string): string {\n  // quick return for empty string or single character strings\n  if (identifier.length <= 1) {\n    return identifier;\n  }\n  // Check for the underscore indicating snake_case\n  if (identifier.indexOf('_') === -1) {\n    return identifier;\n  }\n  return identifier.replace(/[_]+(\\w|$)/g, (m, p1) => p1.toUpperCase());\n}\n\n// tslint:disable-next-line:no-any\nlet _GLOBAL_CUSTOM_OBJECTS = {} as {[objName: string]: any};\n\nexport function serializeKerasObject(instance: serialization.Serializable):\n    serialization.ConfigDictValue {\n  if (instance === null || instance === undefined) {\n    return null;\n  }\n  const dict: serialization.ConfigDictValue = {};\n  dict['className'] = instance.getClassName();\n  dict['config'] = instance.getConfig();\n  return dict;\n}\n\n/**\n * Replace ndarray-style scalar objects in serialization objects with numbers.\n *\n * Background: In some versions of tf.keras, certain scalar values in the HDF5\n * model save file can be serialized as: `{'type': 'ndarray', 'value': num}`,\n * where in `num` is a plain number. This method converts such serialization\n * to a `number`.\n *\n * @param config The keras-format serialization object to be processed\n *   (in place).\n */\nfunction convertNDArrayScalarsInConfig(config: serialization.ConfigDictValue):\n    void {\n  if (config == null || typeof config !== 'object') {\n    return;\n  } else if (Array.isArray(config)) {\n    config.forEach(configItem => convertNDArrayScalarsInConfig(configItem));\n  } else {\n    const fields = Object.keys(config);\n    for (const field of fields) {\n      const value = config[field];\n      if (value != null && typeof value === 'object') {\n        if (!Array.isArray(value) && value['type'] === 'ndarray' &&\n            typeof value['value'] === 'number') {\n          config[field] = value['value'];\n        } else {\n          convertNDArrayScalarsInConfig(value as serialization.ConfigDict);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Deserialize a saved Keras Object\n * @param identifier either a string ID or a saved Keras dictionary\n * @param moduleObjects a list of Python class names to object constructors\n * @param customObjects a list of Python class names to object constructors\n * @param printableModuleName debug text for the object being reconstituted\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns a TensorFlow.js Layers object\n */\n// tslint:disable:no-any\nexport function deserializeKerasObject(\n    identifier: string|serialization.ConfigDict,\n    moduleObjects = {} as {[objName: string]: any},\n    customObjects = {} as {[objName: string]: any},\n    printableModuleName = 'object', fastWeightInit = false): any {\n  // tslint:enable\n  if (typeof identifier === 'string') {\n    const functionName = identifier;\n    let fn;\n    if (functionName in customObjects) {\n      fn = customObjects[functionName];\n    } else if (functionName in _GLOBAL_CUSTOM_OBJECTS) {\n      fn = _GLOBAL_CUSTOM_OBJECTS[functionName];\n    } else {\n      fn = moduleObjects[functionName];\n      if (fn == null) {\n        throw new ValueError(\n            `Unknown ${printableModuleName}: ${identifier}. ` +\n            `This may be due to one of the following reasons:\\n` +\n            `1. The ${printableModuleName} is defined in Python, in which ` +\n            `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n            `code.\\n` +\n            `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n            `but is not registered properly with ` +\n            `tf.serialization.registerClass().`);\n        // TODO(cais): Add link to tutorial page on custom layers.\n      }\n    }\n    return fn;\n  } else {\n    // In this case we are dealing with a Keras config dictionary.\n    const config = identifier;\n    if (config['className'] == null || config['config'] == null) {\n      throw new ValueError(\n          `${printableModuleName}: Improper config format: ` +\n          `${JSON.stringify(config)}.\\n` +\n          `'className' and 'config' must set.`);\n    }\n    const className = config['className'] as string;\n    let cls, fromConfig;\n    if (className in customObjects) {\n      [cls, fromConfig] = customObjects[className];\n    } else if (className in _GLOBAL_CUSTOM_OBJECTS) {\n      [cls, fromConfig] = _GLOBAL_CUSTOM_OBJECTS['className'];\n    } else if (className in moduleObjects) {\n      [cls, fromConfig] = moduleObjects[className];\n    }\n    if (cls == null) {\n      throw new ValueError(\n          `Unknown ${printableModuleName}: ${className}. ` +\n          `This may be due to one of the following reasons:\\n` +\n          `1. The ${printableModuleName} is defined in Python, in which ` +\n          `case it needs to be ported to TensorFlow.js or your JavaScript ` +\n          `code.\\n` +\n          `2. The custom ${printableModuleName} is defined in JavaScript, ` +\n          `but is not registered properly with ` +\n          `tf.serialization.registerClass().`);\n      // TODO(cais): Add link to tutorial page on custom layers.\n    }\n    if (fromConfig != null) {\n      // Porting notes: Instead of checking to see whether fromConfig accepts\n      // customObjects, we create a customObjects dictionary and tack it on to\n      // config['config'] as config['config'].customObjects. Objects can use it,\n      // if they want.\n\n      // tslint:disable-next-line:no-any\n      const customObjectsCombined = {} as {[objName: string]: any};\n      for (const key of Object.keys(_GLOBAL_CUSTOM_OBJECTS)) {\n        customObjectsCombined[key] = _GLOBAL_CUSTOM_OBJECTS[key];\n      }\n      for (const key of Object.keys(customObjects)) {\n        customObjectsCombined[key] = customObjects[key];\n      }\n      // Add the customObjects to config\n      const nestedConfig = config['config'] as serialization.ConfigDict;\n      nestedConfig['customObjects'] = customObjectsCombined;\n\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      convertNDArrayScalarsInConfig(config['config']);\n      const returnObj =\n          fromConfig(cls, config['config'], customObjects, fastWeightInit);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n\n      return returnObj;\n    } else {\n      // Then `cls` may be a function returning a class.\n      // In this case by convention `config` holds\n      // the kwargs of the function.\n      const backupCustomObjects = {..._GLOBAL_CUSTOM_OBJECTS};\n      for (const key of Object.keys(customObjects)) {\n        _GLOBAL_CUSTOM_OBJECTS[key] = customObjects[key];\n      }\n      // In python this is **config['config'], for tfjs-layers we require\n      // classes that use this fall-through construction method to take\n      // a config interface that mimics the expansion of named parameters.\n      const returnObj = new cls(config['config']);\n      _GLOBAL_CUSTOM_OBJECTS = {...backupCustomObjects};\n      return returnObj;\n    }\n  }\n}\n\n/**\n * Compares two numbers for sorting.\n * @param a\n * @param b\n */\nexport function numberCompare(a: number, b: number) {\n  return (a < b) ? -1 : ((a > b) ? 1 : 0);\n}\n\n/**\n * Comparison of two numbers for reverse sorting.\n * @param a\n * @param b\n */\nexport function reverseNumberCompare(a: number, b: number) {\n  return -1 * numberCompare(a, b);\n}\n\n/**\n * Convert a string into the corresponding DType.\n * @param dtype\n * @returns An instance of DType.\n */\nexport function stringToDType(dtype: string): DataType {\n  switch (dtype) {\n    case 'float32':\n      return 'float32';\n    default:\n      throw new ValueError(`Invalid dtype: ${dtype}`);\n  }\n}\n\n/**\n * Test the element-by-element equality of two Arrays of strings.\n * @param xs First array of strings.\n * @param ys Second array of strings.\n * @returns Wether the two arrays are all equal, element by element.\n */\nexport function stringsEqual(xs: string[], ys: string[]): boolean {\n  if (xs == null || ys == null) {\n    return xs === ys;\n  }\n  if (xs.length !== ys.length) {\n    return false;\n  }\n  for (let i = 0; i < xs.length; ++i) {\n    if (xs[i] !== ys[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Get the unique elements of an array.\n * @param xs Array.\n * @returns An Array consisting of the unique elements in `xs`.\n */\nexport function unique<T>(xs: T[]): T[] {\n  if (xs == null) {\n    return xs;\n  }\n  const out: T[] = [];\n  // TODO(cais): Maybe improve performance by sorting.\n  for (const x of xs) {\n    if (out.indexOf(x) === -1) {\n      out.push(x);\n    }\n  }\n  return out;\n}\n\n/**\n * Determine if an Object is empty (i.e., does not have own properties).\n * @param obj Object\n * @returns Whether the Object is empty.\n * @throws ValueError: If object is `null` or `undefined`.\n */\nexport function isObjectEmpty(obj: {}): boolean {\n  if (obj == null) {\n    throw new ValueError(`Invalid value in obj: ${JSON.stringify(obj)}`);\n  }\n  for (const key in obj) {\n    if (obj.hasOwnProperty(key)) {\n      return false;\n    }\n  }\n  return true;\n}\n\n/**\n * Helper function used to build type union/enum run-time checkers.\n * @param values The list of allowed values.\n * @param label A string name for the type\n * @param value The value to test.\n * @throws ValueError: If the value is not in values nor `undefined`/`null`.\n */\nexport function checkStringTypeUnionValue(\n    values: string[], label: string, value: string): void {\n  if (value == null) {\n    return;\n  }\n  if (values.indexOf(value) < 0) {\n    throw new ValueError(`${value} is not a valid ${label}.  Valid values are ${\n        values} or null/undefined.`);\n  }\n}\n\n/**\n * Helper function for verifying the types of inputs.\n *\n * Ensures that the elements of `x` are all of type `expectedType`.\n * Also verifies that the length of `x` is within bounds.\n *\n * @param x Object to test.\n * @param expectedType The string expected type of all of the elements in the\n * Array.\n * @param minLength Return false if x.length is less than this.\n * @param maxLength Return false if x.length is greater than this.\n * @returns true if and only if `x` is an `Array<expectedType>` with\n * length >= `minLength` and <= `maxLength`.\n */\n// tslint:disable:no-any\nexport function checkArrayTypeAndLength(\n    x: any, expectedType: string, minLength = 0,\n    maxLength = Infinity): boolean {\n  assert(minLength >= 0);\n  assert(maxLength >= minLength);\n  return (\n      Array.isArray(x) && x.length >= minLength && x.length <= maxLength &&\n      x.every(e => typeof e === expectedType));\n}\n// tslint:enable:no-any\n\n/**\n * Assert that a value or an array of value are positive integer.\n *\n * @param value The value being asserted on. May be a single number or an array\n *   of numbers.\n * @param name Name of the value, used to make the error message.\n */\nexport function assertPositiveInteger(value: number|number[], name: string) {\n  if (Array.isArray(value)) {\n    util.assert(\n        value.length > 0, () => `${name} is unexpectedly an empty array.`);\n    value.forEach(\n        (v, i) => assertPositiveInteger(v, `element ${i + 1} of ${name}`));\n  } else {\n    util.assert(\n        Number.isInteger(value) && value > 0,\n        () => `Expected ${name} to be a positive integer, but got ` +\n            `${formatAsFriendlyString(value)}.`);\n  }\n}\n\n/**\n * Format a value into a display-friendly, human-readable fashion.\n *\n * - `null` is formatted as `'null'`\n * - Strings are formated with flanking pair of quotes.\n * - Arrays are formatted with flanking pair of square brackets.\n *\n * @param value The value to display.\n * @return Formatted string.\n */\n// tslint:disable-next-line:no-any\nexport function formatAsFriendlyString(value: any): string {\n  if (value === null) {\n    return 'null';\n  } else if (Array.isArray(value)) {\n    return '[' + value.map(v => formatAsFriendlyString(v)).join(',') + ']';\n  } else if (typeof value === 'string') {\n    return `\"${value}\"`;\n  } else {\n    return `${value}`;\n  }\n}\n\n/**\n * Returns a function `f2` (decorator) which wraps the original function\n * `f`. `f2` guarantees that `f` can be called at most once\n * every `waitMs` ms. If `f2` is called more often, it will return\n * the last returned result of `f`.\n *\n * @param f The original function `f` to wrap.\n * @param waitMs The time between two consecutive calls to `f` in ms.\n */\nexport function debounce<T>(\n    f: (...args: Array<{}>) => T, waitMs: number): (...args: Array<{}>) => T {\n  let lastTime = util.now();\n  let lastResult: T;\n  const f2 = (...args: Array<{}>) => {\n    const now = util.now();\n    if (now - lastTime < waitMs) {\n      return lastResult;\n    }\n    lastTime = now;\n    lastResult = f(...args);\n    return lastResult;\n  };\n  return f2;\n}\n\n/**\n * Returns the fusable activation given a layers identifier.\n *\n * @param activationName The layers identifier string.\n * @return The name of the fusable activation.\n */\nexport function mapActivationToFusedKernel(activationName: string):\n    fused.Activation {\n  if (activationName === 'relu') {\n    return 'relu';\n  }\n  if (activationName === 'linear') {\n    return 'linear';\n  }\n  if (activationName === 'elu') {\n    return 'elu';\n  }\n  return null;\n}\n\ntype PossibleValues = Array<Array<boolean|string|number>>;\n\n/**\n * Returns the cartesian product of sets of values.\n * This works the same as itertools.product in Python.\n *\n * Example:\n *\n * filters = [128, 256, 512]\n * paddings = ['same', 'valid']\n *\n * product = [ [128, 'same'], [128, 'valid'], [256, 'same'], [256, 'valid'],\n * [512, 'same'], [512, 'valid']]\n *\n * @param arrayOfValues List/array of values.\n * @return The cartesian product.\n */\nexport function getCartesianProductOfValues(...arrayOfValues: PossibleValues):\n    PossibleValues {\n  assert(arrayOfValues.length > 0, 'arrayOfValues is empty');\n\n  for (const values of arrayOfValues) {\n    assert(Array.isArray(values), 'one of the values is not an array');\n    assert(values.length > 0, 'one of the values is empty');\n  }\n\n  return arrayOfValues.reduce((products, values) => {\n    if (products.length === 0) {\n      return values.map(value => [value]);\n    }\n\n    return values\n        .map(value => {\n          return products.map((prevValue) => [...prevValue, value]);\n        })\n        .reduce((flattenedProduct, unflattenedProduct) => {\n          return flattenedProduct.concat(unflattenedProduct);\n        }, []);\n  }, [] as PossibleValues);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/contraints.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport {epsilon} from './backend/common';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\n/**\n * Helper function used by many of the Constraints to find the L2Norms.\n */\nfunction calcL2Norms(w: Tensor, axis: number): Tensor {\n  return tidy(() => tfc.sqrt(tfc.sum(tfc.mul(w, w), axis, true)));\n}\n\n/**\n * Base class for functions that impose constraints on weight values\n *\n * @doc {\n *   heading: 'Constraints',\n *   subheading: 'Classes',\n *   namespace: 'constraints'\n * }\n */\nexport abstract class Constraint extends serialization.Serializable {\n  /* Porting note: was __call__, apply chosen to match other similar choices */\n  abstract apply(w: Tensor): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport interface MaxNormArgs {\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   *\n   *  For instance, in a `Dense` layer the weight matrix\n   *  has shape `[inputDim, outputDim]`,\n   *  set `axis` to `0` to constrain each weight vector\n   *  of length `[inputDim,]`.\n   *  In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   *  the weight tensor has shape\n   *  `[rows, cols, inputDepth, outputDepth]`,\n   *  set `axis` to `[0, 1, 2]`\n   *  to constrain the weights of each filter tensor of size\n   *  `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class MaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MaxNorm';\n  private maxValue: number;\n  private axis: number;\n  private readonly defaultMaxValue = 2;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MaxNormArgs) {\n    super();\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.clipByValue(norms, 0, this.maxValue);\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {maxValue: this.maxValue, axis: this.axis};\n  }\n}\nserialization.registerClass(MaxNorm);\n\nexport interface UnitNormArgs {\n  /**\n   * Axis along which to calculate norms.\n   *\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * [rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n}\n\nexport class UnitNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'UnitNorm';\n  private axis: number;\n  private readonly defaultAxis = 0;\n  constructor(args: UnitNormArgs) {\n    super();\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(\n        () => tfc.div(w, tfc.add(epsilon(), calcL2Norms(w, this.axis))));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {axis: this.axis};\n  }\n}\nserialization.registerClass(UnitNorm);\n\nexport class NonNeg extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'NonNeg';\n\n  apply(w: Tensor): Tensor {\n    return tfc.relu(w);\n  }\n}\nserialization.registerClass(NonNeg);\n\nexport interface MinMaxNormArgs {\n  /**\n   * Minimum norm for incoming weights\n   */\n  minValue?: number;\n  /**\n   * Maximum norm for incoming weights\n   */\n  maxValue?: number;\n  /**\n   * Axis along which to calculate norms.\n   * For instance, in a `Dense` layer the weight matrix\n   * has shape `[inputDim, outputDim]`,\n   * set `axis` to `0` to constrain each weight vector\n   * of length `[inputDim,]`.\n   * In a `Conv2D` layer with `dataFormat=\"channels_last\"`,\n   * the weight tensor has shape\n   * `[rows, cols, inputDepth, outputDepth]`,\n   * set `axis` to `[0, 1, 2]`\n   * to constrain the weights of each filter tensor of size\n   * `[rows, cols, inputDepth]`.\n   */\n  axis?: number;\n  /**\n   * Rate for enforcing the constraint: weights will be rescaled to yield:\n   * `(1 - rate) * norm + rate * norm.clip(minValue, maxValue)`.\n   * Effectively, this means that rate=1.0 stands for strict\n   * enforcement of the constraint, while rate<1.0 means that\n   * weights will be rescaled at each step to slowly move\n   * towards a value inside the desired interval.\n   */\n  rate?: number;\n}\n\nexport class MinMaxNorm extends Constraint {\n  /** @nocollapse */\n  static readonly className = 'MinMaxNorm';\n  private minValue: number;\n  private maxValue: number;\n  private rate: number;\n  private axis: number;\n  private readonly defaultMinValue = 0.0;\n  private readonly defaultMaxValue = 1.0;\n  private readonly defaultRate = 1.0;\n  private readonly defaultAxis = 0;\n\n  constructor(args: MinMaxNormArgs) {\n    super();\n    this.minValue =\n        args.minValue != null ? args.minValue : this.defaultMinValue;\n    this.maxValue =\n        args.maxValue != null ? args.maxValue : this.defaultMaxValue;\n    this.rate = args.rate != null ? args.rate : this.defaultRate;\n    this.axis = args.axis != null ? args.axis : this.defaultAxis;\n  }\n\n  apply(w: Tensor): Tensor {\n    return tidy(() => {\n      const norms = calcL2Norms(w, this.axis);\n      const desired = tfc.add(\n          tfc.mul(\n              this.rate, tfc.clipByValue(norms, this.minValue, this.maxValue)),\n          tfc.mul(1.0 - this.rate, norms));\n      return tfc.mul(w, tfc.div(desired, tfc.add(epsilon(), norms)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      minValue: this.minValue,\n      maxValue: this.maxValue,\n      rate: this.rate,\n      axis: this.axis\n    };\n  }\n}\nserialization.registerClass(MinMaxNorm);\n\n/** @docinline */\nexport type ConstraintIdentifier =\n    'maxNorm'|'minMaxNorm'|'nonNeg'|'unitNorm'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in ConstraintIdentifier]: string} = {\n      'maxNorm': 'MaxNorm',\n      'minMaxNorm': 'MinMaxNorm',\n      'nonNeg': 'NonNeg',\n      'unitNorm': 'UnitNorm'\n    };\n\nexport function serializeConstraint(constraint: Constraint):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeConstraint(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Constraint {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'constraint');\n}\n\nexport function getConstraint(identifier: ConstraintIdentifier|\n                              serialization.ConfigDict|Constraint): Constraint {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        CONSTRAINT_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeConstraint(config);\n  } else if (identifier instanceof Constraint) {\n    return identifier;\n  } else {\n    return deserializeConstraint(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constraint, MaxNorm, MaxNormArgs, MinMaxNorm, MinMaxNormArgs, NonNeg, UnitNorm, UnitNormArgs} from './constraints';\n\n/**\n * MaxNorm weight constraint.\n *\n * Constrains the weights incident to each hidden unit\n * to have a norm less than or equal to a desired value.\n *\n * References\n *       - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting\n * Srivastava, Hinton, et al.\n * 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Constraints',namespace: 'constraints'}\n */\nexport function maxNorm(args: MaxNormArgs): Constraint {\n  return new MaxNorm(args);\n}\n\n/**\n * Constrains the weights incident to each hidden unit to have unit norm.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function unitNorm(args: UnitNormArgs): Constraint {\n  return new UnitNorm(args);\n}\n\n/**\n * Constains the weight to be non-negative.\n *\n * @doc {heading: 'Constraints', namespace: 'constraints'}\n */\nexport function nonNeg(): Constraint {\n  return new NonNeg();\n}\n\n/** @doc {heading: 'Constraints', namespace: 'constraints'} */\nexport function minMaxNorm(config: MinMaxNormArgs): Constraint {\n  return new MinMaxNorm(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// TODO(huan): add layer-specific input shape types (see: https://github.com/tensorflow/tfjs-layers/pull/492)\n/** @docalias (null | number)[] */\nexport type Shape = Array<null | number>;\n\n// The tfjs-core version of DataType must stay synced with this.\nexport type DataType = 'float32'|'int32'|'bool'|'complex64'|'string';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type DataFormat = 'channelsFirst'|'channelsLast';\nexport const VALID_DATA_FORMAT_VALUES = ['channelsFirst', 'channelsLast'];\n\n// These constants have a snake vs. camel distinction.\nexport type DataFormatSerialization = 'channels_first'|'channels_last';\n\n/** @docinline */\nexport type PaddingMode = 'valid'|'same'|'causal';\nexport const VALID_PADDING_MODE_VALUES = ['valid', 'same', 'causal'];\n\n/** @docinline */\nexport type PoolMode = 'max'|'avg';\nexport const VALID_POOL_MODE_VALUES = ['max', 'avg'];\n\n/** @docinline */\nexport type BidirectionalMergeMode = 'sum'|'mul'|'concat'|'ave';\nexport const VALID_BIDIRECTIONAL_MERGE_MODES = ['sum', 'mul', 'concat', 'ave'];\n\n/** @docinline */\nexport type SampleWeightMode = 'temporal';\nexport const VALID_SAMPLE_WEIGHT_MODES = ['temporal'];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Common functions for TensorFlow.js Layers.\n */\nimport {VALID_DATA_FORMAT_VALUES, VALID_PADDING_MODE_VALUES, VALID_POOL_MODE_VALUES} from './keras_format/common';\nimport {checkStringTypeUnionValue} from './utils/generic_utils';\n\n// A map from the requested scoped name of a Tensor to the number of Tensors\n// wanting that name so far.  This allows enforcing name uniqueness by appending\n// an incrementing index, e.g. scope/name, scope/name_1, scope/name_2, etc.\nconst nameMap: Map<string, number> = new Map<string, number>();\n\nexport function checkDataFormat(value?: string): void {\n  checkStringTypeUnionValue(VALID_DATA_FORMAT_VALUES, 'DataFormat', value);\n}\n\nexport function checkPaddingMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_PADDING_MODE_VALUES, 'PaddingMode', value);\n}\n\nexport function checkPoolMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_POOL_MODE_VALUES, 'PoolMode', value);\n}\n\nconst _nameScopeStack: string[] = [];\nconst _nameScopeDivider = '/';\n\n/**\n * Enter namescope, which can be nested.\n */\nexport function nameScope<T>(name: string, fn: () => T): T {\n  _nameScopeStack.push(name);\n  try {\n    const val: T = fn();\n    _nameScopeStack.pop();\n    return val;\n  } catch (e) {\n    _nameScopeStack.pop();\n    throw e;\n  }\n}\n\n/**\n * Get the current namescope as a flat, concatenated string.\n */\nfunction currentNameScopePrefix(): string {\n  if (_nameScopeStack.length === 0) {\n    return '';\n  } else {\n    return _nameScopeStack.join(_nameScopeDivider) + _nameScopeDivider;\n  }\n}\n\n/**\n * Get the name a Tensor (or Variable) would have if not uniqueified.\n * @param tensorName\n * @return Scoped name string.\n */\nexport function getScopedTensorName(tensorName: string): string {\n  if (!isValidTensorName(tensorName)) {\n    throw new Error('Not a valid tensor name: \\'' + tensorName + '\\'');\n  }\n  return currentNameScopePrefix() + tensorName;\n}\n\n/**\n * Get unique names for Tensors and Variables.\n * @param scopedName The fully-qualified name of the Tensor, i.e. as produced by\n *  `getScopedTensorName()`.\n * @return A unique version of the given fully scoped name.\n *   If this is the first time that the scoped name is seen in this session,\n *   then the given `scopedName` is returned unaltered.  If the same name is\n *   seen again (producing a collision), an incrementing suffix is added to the\n *   end of the name, so it takes the form 'scope/name_1', 'scope/name_2', etc.\n */\nexport function getUniqueTensorName(scopedName: string): string {\n  if (!isValidTensorName(scopedName)) {\n    throw new Error('Not a valid tensor name: \\'' + scopedName + '\\'');\n  }\n  if (!nameMap.has(scopedName)) {\n    nameMap.set(scopedName, 0);\n  }\n  const index = nameMap.get(scopedName);\n  nameMap.set(scopedName, nameMap.get(scopedName) + 1);\n\n  if (index > 0) {\n    const result = `${scopedName}_${index}`;\n    // Mark the composed name as used in case someone wants\n    // to call getUniqueTensorName(\"name_1\").\n    nameMap.set(result, 1);\n    return result;\n  } else {\n    return scopedName;\n  }\n}\n\nconst tensorNameRegex = new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\\._\\/]*$/);\n\n/**\n * Determine whether a string is a valid tensor name.\n * @param name\n * @returns A Boolean indicating whether `name` is a valid tensor name.\n */\nexport function isValidTensorName(name: string): boolean {\n  return !!name.match(tensorNameRegex);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Math utility functions.\n *\n * This file contains some frequently used math function that operates on\n * number[] or Float32Array and return a number. Many of these functions are\n * not-so-thick wrappers around TF.js Core functions. But they offer the\n * convenience of\n * 1) not having to convert the inputs into Tensors,\n * 2) not having to convert the returned Tensors to numbers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar, Tensor1D, tensor1d} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\n\nexport type ArrayTypes = Uint8Array|Int32Array|Float32Array;\n\n/**\n * Determine if a number is an integer.\n */\nexport function isInteger(x: number): boolean {\n  return x === parseInt(x.toString(), 10);\n}\n\n/**\n * Calculate the product of an array of numbers.\n * @param array The array to calculate the product over.\n * @param begin Beginning index, inclusive.\n * @param end Ending index, exclusive.\n * @return The product.\n */\nexport function arrayProd(\n    array: number[]|ArrayTypes, begin?: number, end?: number): number {\n  if (begin == null) {\n    begin = 0;\n  }\n  if (end == null) {\n    end = array.length;\n  }\n\n  let prod = 1;\n  for (let i = begin; i < end; ++i) {\n    prod *= array[i];\n  }\n  return prod;\n}\n\n/**\n * A helper function transforms the two input types to an instance of Tensor1D,\n * so the return value can be fed directly into various TF.js Core functions.\n * @param array\n */\nfunction toArray1D(array: number[]|Float32Array): Tensor1D {\n  array = Array.isArray(array) ? new Float32Array(array) : array;\n  return tensor1d(array);\n}\n\n/**\n * Compute minimum value.\n * @param array\n * @return minimum value.\n */\nexport function min(array: number[]|Float32Array): number {\n  return tfc.min(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute maximum value.\n * @param array\n * @return maximum value\n */\nexport function max(array: number[]|Float32Array): number {\n  return tfc.max(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute sum of array.\n * @param array\n * @return The sum.\n */\nexport function sum(array: number[]|Float32Array): number {\n  return tfc.sum(toArray1D(array)).dataSync()[0];\n}\n\n/**\n * Compute mean of array.\n * @param array\n * @return The mean.\n */\nexport function mean(array: number[]|Float32Array): number {\n  return sum(array) / array.length;\n}\n\n/**\n * Compute variance of array.\n * @param array\n * @return The variance.\n */\nexport function variance(array: number[]|Float32Array): number {\n  const demeaned = tfc.sub(toArray1D(array), scalar(mean(array)));\n  const sumSquare = tfc.sum(tfc.mul(demeaned, demeaned)).dataSync()[0];\n  return sumSquare / array.length;\n}\n\n/**\n * Compute median of array.\n * @param array\n * @return The median value.\n */\nexport function median(array: number[]|Float32Array): number {\n  const arraySorted = array.slice().sort((a, b) => a - b);\n  const lowIdx = Math.floor((arraySorted.length - 1) / 2);\n  const highIdx = Math.ceil((arraySorted.length - 1) / 2);\n  if (lowIdx === highIdx) {\n    return arraySorted[lowIdx];\n  }\n  return (arraySorted[lowIdx] + arraySorted[highIdx]) / 2;\n}\n\n/**\n * Generate an array of integers in [begin, end).\n * @param begin Beginning integer, inclusive.\n * @param end Ending integer, exclusive.\n * @returns Range array.\n * @throws ValueError, iff `end` < `begin`.\n */\nexport function range(begin: number, end: number): number[] {\n  if (end < begin) {\n    throw new ValueError(`end (${end}) < begin (${begin}) is forbidden.`);\n  }\n  const out: number[] = [];\n  for (let i = begin; i < end; ++i) {\n    out.push(i);\n  }\n  return out;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * deeplearn.js backend.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {onesLike as coreOnesLike, scalar, Tensor, Tensor1D, tensor1d, Tensor2D, Tensor3D, Tensor4D, Tensor5D, tidy, where, zerosLike as coreZerosLike} from '@tensorflow/tfjs-core';\nimport {checkDataFormat} from '../common';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {HasShape} from '../types';\nimport * as math_utils from '../utils/math_utils';\n\nimport {imageDataFormat} from './common';\n\n// tslint:enable\n\n/* Setting and getting backend from deeplearn.js. */\n\n// Default deeplearn.js backend is WebGL (GPU).\nlet backend: 'cpu'|'webgl' = 'webgl';\n\nexport function setBackend(requestedBackend: 'cpu'|'webgl') {\n  tfc.setBackend(requestedBackend);\n  backend = requestedBackend;\n}\n\nexport function getBackend(): 'cpu'|'webgl' {\n  return backend;\n}\n\n/**\n * Indicates whether the backend is operating symbolically.\n *\n * This function will be used to determine how to interpret user code. If\n * it returns true, calls to the backend construct a symbolic graph; if\n * it returns false, calls to the backend execute immediately.\n */\nexport function isBackendSymbolic(): boolean {\n  return false;\n}\n\n/**\n * Get the number of elements in a Tensor.\n * @param x The Tensor.\n * @return Number of elements in `x`.\n */\nexport function countParams(x: HasShape): number {\n  const shape = x.shape;\n  if (shape.length > 0) {\n    return shape.reduce((a: number, b: number) => a * b);\n  } else {\n    // Scalar.\n    return 1;\n  }\n}\n\n/**\n * Casts a tensor to a different dtype and returns it.\n * @param x Input tensor.\n * @param dtype String: 'float32'|'int32'|'bool'.\n * @returns Tensor of the specified `dtype`.\n */\nexport function cast(x: Tensor, dtype: tfc.DataType): Tensor {\n  return x.asType(dtype);\n}\n\n/**\n * Adds a 1-sized dimension at index \"axis\".\n * @param x Input tensor.\n * @param axis Position where to add the new axis.\n * @returns Result of the dimension expansion.\n */\nexport function expandDims(x: Tensor, axis = -1): Tensor {\n  const outShape = x.shape.slice();\n  if (axis < 0) {\n    axis = outShape.length + axis + 1;\n  }\n  outShape.splice(axis, 0, 1);\n  return x.reshape(outShape);\n}\n\n/**\n * Repeats a 2D tensor.\n *\n * If `x` has shape `[samples, dim]` and `n` is 2, for example, the output\n * will have shape `[samples, 2, dim]`.\n *\n * @param x Input tensor.\n * @param n Integer, number of times to repeat.\n * @returns The result of the repeat operation.\n * @throws ValueError: If input tensor is not 2D.\n */\nexport function repeat(x: Tensor, n: number): Tensor {\n  return tidy(() => {\n    if (x.shape.length !== 2) {\n      throw new ValueError(\n          `repeat() expects a rank-2 tensor, but received a ` +\n          `rank-${x.shape.length} tensor.`);\n    }\n    const y = expandDims(x, 1);\n    return tile(y, [1, n, 1]);\n  });\n}\n\n/**\n * Flatten a Tensor into 1D.\n * @param x Input tensor.\n * @return The result of the flattening `x`.\n */\nexport function flatten(x: Tensor): Tensor {\n  const newShape = [math_utils.arrayProd(x.shape)];\n  return x.reshape(newShape);\n}\n\n/**\n * Turn a nD tensor into a 2D tensor with same 0th dimension.\n * In other words, it flattens each data samples of a batch.\n *\n * @param x The tensor to flatten. The rank of this tensor is required to be 2\n *   or higher.\n * @return The result of the flattening.\n */\nexport function batchFlatten(x: Tensor): Tensor {\n  if (x.rank <= 1) {\n    throw new ValueError(\n        `batchFlatten requires a minimum rank of 2. Got rank: ${x.rank}.`);\n  }\n  const newShape = [x.shape[0], math_utils.arrayProd(x.shape, 1)];\n  return x.reshape(newShape);\n}\n\n/**\n * Do slicing along the first axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the first axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongFirstAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [start, 0], [size, array.shape[1]]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [start, 0, 0],\n            [size, array.shape[1], array.shape[2]]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [start, 0, 0, 0],\n            [size, array.shape[1], array.shape[2], array.shape[3]]);\n      case 5:\n        return tfc.slice(array as Tensor5D, [start, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4]\n        ]);\n      case 6:\n        return tfc.slice(array, [start, 0, 0, 0, 0, 0], [\n          size, array.shape[1], array.shape[2], array.shape[3], array.shape[4],\n          array.shape[5]\n        ]);\n      default:\n        throw new ValueError(\n            `sliceAlongFirstAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the last axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size size of the slice along the last axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongLastAxis(\n    array: Tensor, start: number, size: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        return tfc.slice2d(\n            array as Tensor2D, [0, start], [array.shape[0], size]);\n      case 3:\n        return tfc.slice3d(\n            array as Tensor3D, [0, 0, start],\n            [array.shape[0], array.shape[1], size]);\n      case 4:\n        return tfc.slice4d(\n            array as Tensor4D, [0, 0, 0, start],\n            [array.shape[0], array.shape[1], array.shape[2], size]);\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Do slicing along the sepcified axis.\n * @param array input `tf.Tensor`.\n * @param start starting index, inclusive.\n * @param size of the slice along the chosen axis.\n * @param choose an axis.\n * @returns result of the slicing.\n * @throws ValueError: If `array` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function sliceAlongAxis(\n    array: Tensor, start: number, size: number, axis: number): Tensor {\n  return tidy(() => {\n    switch (array.rank) {\n      case 1:\n        return tfc.slice1d(array as Tensor1D, start, size);\n      case 2:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 3:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice3d(\n                array as Tensor3D, [0, start, 0],\n                [array.shape[0], size, array.shape[2]]);\n          case 3:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      case 4:\n        switch (axis) {\n          case 1:\n            return sliceAlongFirstAxis(array, start, size);\n          case 2:\n            return tfc.slice4d(\n                array as Tensor4D, [0, start, 0, 0],\n                [array.shape[0], size, array.shape[2], array.shape[3]]);\n          case 3:\n            return tfc.slice4d(\n                array as Tensor4D, [0, 0, start, 0],\n                [array.shape[0], array.shape[1], size, array.shape[3]]);\n          case 4:\n            return sliceAlongLastAxis(array, start, size);\n          default:\n            throw new ValueError(\n                `The axis is not within the rank of the tensor ` +\n                `${axis}`);\n        }\n      default:\n        throw new ValueError(\n            `sliceAlongLastAxis() received an unsupported tensor rank: ` +\n            `${array.rank}`);\n    }\n  });\n}\n\n/**\n * Concatenates a list of tensors alongside the specified axis.\n * @param tensors `Array` of tensors to concatenate.\n * @param axis Concatenation axis.\n * @returns The result of the concatenation.\n */\nexport function concatenate(tensors: Tensor[], axis = -1): Tensor {\n  let rank: number;\n  if (axis < 0) {\n    rank = tensors[0].rank;\n    if (rank !== 0) {\n      axis = rank;\n    } else {\n      axis = 0;\n    }\n  }\n  if (axis === tensors[0].rank) {\n    // Porting Note: This is necessary because tfc.concat() requires axis to be\n    //   in the interval [-rank, rank).\n    axis = -1;\n  }\n  // Porting Note: Sparse concat is not supported yet.\n  return tfc.concat(tensors, axis);\n}\n\n/**\n * Concatenate two arrays along the first dimension.\n * @param a The 1st `tf.Tensor` to concatenate.\n * @param b The 2nd `tf.Tensor` to concatenate.\n * @returns Result of the concatenation.\n * @throws ValueError: If `a` is of an unsupported subtype of `tf.Tensor`.\n */\nexport function concatAlongFirstAxis(a: Tensor, b: Tensor): Tensor {\n  switch (a.rank) {\n    case 1:\n      return tfc.concat1d([a as Tensor1D, b as Tensor1D]);\n    case 2:\n      return tfc.concat2d([a as Tensor2D, b as Tensor2D], 0);\n    case 3:\n      return tfc.concat3d([a as Tensor3D, b as Tensor3D], 0);\n    case 4:\n      return tfc.concat4d([a as Tensor4D, b as Tensor4D], 0);\n    default:\n      throw new ValueError(\n          `concatAlongFirstAxis() received an unsupported ` +\n          `tensor rank: ${a.rank}`);\n  }\n}\n\n/**\n * Creates a tensor by tiling `x` by `n`.\n * @param x A tensor.\n * @param n An Array of integers or a single integer. If an Array, the length\n *   must be the same as the number of dimensions in `x`. If a single integer,\n *   it will be treated as an Array of length 1.\n */\nexport function tile(x: Tensor, n: number|number[]): Tensor {\n  if (!Array.isArray(n)) {\n    n = [n];\n  }\n  if (x.rank !== n.length) {\n    throw new ValueError(\n        `The length of input n (${n.length}) does not match ` +\n        `the number of dimensions in input x (${x.rank})`);\n  }\n  return tfc.tile(x, n);\n}\n\n/* Creation of random tensors. */\n\n/**\n * Get a tensor with normal distribution of values.\n *\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @return The normal tensor.\n */\nexport function randomNormal(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: 'float32'|'int32',\n    seed?: number): Tensor {\n  return tfc.randomNormal(shape, mean, stddev, dtype, seed);\n}\n\n/* Linear Algebra */\n\n/**\n * Multiply two tensors and returns the result as a tensor.\n *\n * For 2D tensors, this is equivalent to matrix multiplication (matMul).\n * For tensors of higher ranks, it follows the Theano behavior,\n * (e.g. `(2, 3) * (4, 3, 5) -> (2, 4, 5)`).  From the Theano documentation:\n *\n * For N dimensions it is a sum product over the last axis of x and the\n * second-to-last of y:\n *\n * @param a A tensor of at least rank 2.\n * @param b A tensor of at least rank 2.\n * @param activation (optional) A string identifying the activation\n *   function.\n * @return Result of the dot operation.\n */\nexport function dot(\n    a: Tensor, b: Tensor, activation?: tfc.fused.Activation,\n    bias?: Tensor): Tensor {\n  if ((a.rank < 2) || (b.rank < 2)) {\n    throw new NotImplementedError(\n        `dot requires both inputs to be rank >= 2` +\n        ` but got x shape = ${a.shape} and y shape = ${b.shape}`);\n  }\n  if (b.rank >= 3) {\n    const xLastDim = a.shape.slice(-1)[0];\n    const ySecondLastDim = b.shape.slice(-2)[0];\n    if (xLastDim !== ySecondLastDim) {\n      throw new NotImplementedError(\n          `If rank y >= 3, then the second last dim` +\n          ` of y must equal the last dim of x but got x shape = ${\n              a.shape} and ` +\n          ` y shape = ${b.shape}`);\n    }\n  }\n  // Handle basic 2D x 2D case.\n  if ((a.rank === 2) && (b.rank === 2)) {\n    const transposeA = false;\n    const transposeB = false;\n    // tfc.fused.matMul only fuses certain activation functions. Unsupported\n    // activation functions are treated as 'linear' activations, which is\n    // equivalent to a no-op.\n    return tfc.fused.matMul({\n      a,\n      b: b as Tensor2D,\n      transposeA,\n      transposeB,\n      bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n      activation\n    });\n  } else {\n    // Reshape x into the analogous 2D Tensor.\n    const aFirstDims = a.shape.slice();  // Holds all but the last dim of x.\n    const aLastDim = aFirstDims.pop();\n    a = a.reshape([-1, aLastDim]);\n\n    // Reshape y into the analogous 2D Tensor, and keep track of the\n    // required dimensions to reproduce the output shape.\n    const bShape = b.shape.slice();\n    const bLastDim = bShape.pop();\n    const ySecondLastDim = bShape.pop();\n    const yOtherDims = [...bShape, bLastDim];\n    // permutation should be like [r-2, 0, 1, 2, ... r-4, r-3, r-1]\n    // where r is the rank of y.\n    const perm = Array.from({length: b.rank}, (_, i) => {\n      if (i === 0) {\n        return b.rank - 2;\n      } else if (i <= b.rank - 2) {\n        return i - 1;\n      }\n      return i;\n    });\n    b = b.transpose(perm).reshape([ySecondLastDim, -1]);\n\n    // Multiply x and y as 2D Tensors, and then reshape back to original.\n    const outputShape = [...aFirstDims, ...yOtherDims];\n    const transposeA = false;\n    const transposeB = false;\n    return tfc.fused\n        .matMul({\n          a,\n          b,\n          transposeA,\n          transposeB,\n          bias: bias ? reshapeBias(a.rank, bias, imageDataFormat()) : null,\n          activation\n        })\n        .reshape(outputShape);\n  }\n}\n\n/**\n * Compute the sign Tensor of an input Tensor.\n *\n * Elements of the input `tf.Tensor` that are === 0 are mapped to 0.\n * Elements of the input `tf.Tensor` that are > 0 are mapped to 1.\n * Elements of the input `tf.Tensor` that are < 0 are mapped to -1.\n *\n * @param x Input `tf.Tensor`.\n * @return The sign `tf.Tensor`.\n */\nexport function sign(x: Tensor): Tensor {\n  // TODO(cais): Move to the core.\n  return tidy(() => {\n    const zerosLikeX = coreZerosLike(x);\n    const onesLikeX = coreOnesLike(x);\n    return where(\n        tfc.equal(x, zerosLikeX), zerosLikeX,\n        where(\n            tfc.greater(x, coreZerosLike(x)), onesLikeX,\n            tfc.mul(-1, onesLikeX)));\n  });\n}\n\n/**\n * Computes the one-hot representation of an integer tensor.\n * @param indices nD integer tensor of shape\n *   `(batch_size, dim1, dim2, ... dim(n-1))`\n * @param numClasses Integer, number of classes to consider.\n * @returns (n + 1)D one hot representation of the input\n *   with shape `(batch_size, dim1, dim2, ... dim(n-1), num_classes)`\n */\nexport function oneHot(indices: Tensor, numClasses: number): Tensor {\n  return tidy(() => {\n    if (indices.rank !== 1) {\n      throw new Error(\n          'Only 1D one-hot tensors are supported in the ' +\n          'deeplearn backend, at present.');\n    }\n    indices = indices.toInt();\n    return tfc.oneHot(indices as Tensor1D, numClasses).toFloat();\n  });\n}\n\n/* Elementary math functions. */\n\n/**\n * Retrieves the elements of indices `indices` in the tensor `reference`.\n * @param reference A tensor.\n * @param indices An integer tensor of indices or an `Array` of integers.\n * @param axis Axis along which to perform the gather operation.\n * @returns The result of the gathering as a tensor.\n */\nexport function gather(\n    reference: Tensor, indices: number[]|Tensor1D, axis?: number): Tensor {\n  return tidy(() => {\n    if (Array.isArray(indices)) {\n      indices = tensor1d(indices, 'int32');\n    } else {\n      indices = indices.toInt();\n    }\n    return tfc.gather(reference, indices, axis);\n  });\n}\n\n/**\n * Element-wise square.\n * @param x Input tensor.\n * @return element-wise x^2\n */\nexport function square(x: Tensor): Tensor {\n  return tfc.mul(x, x);\n}\n\n/**\n * Element-wise exponentiation.\n *\n * Porting Note: In PyKeras, `a` (the exponent) is a Python integer, which\n *   takes advatnage of the backend's (e.g., TensorFlow's) automatic\n * conversion to tensor. Here we allow `a` to be either a number or a tensor.\n *\n * @param x The base tensor.\n * @param a The exponent, tensor or number. If a number, it is rounded to the\n *   nearest integer and converted to a tensor.\n * @returns A tensor of the same shape as `x`.\n */\nexport function pow(x: Tensor, a: Tensor|number): Tensor {\n  return tidy(() => {\n    if (typeof (a) === 'number') {\n      a = scalar(Math.round(a), 'int32');\n    }\n    if (a.dtype !== 'int32') {\n      throw new NotImplementedError(\n          `Non-int32 dtype (${a.dtype}) is not supported by pow() yet`);\n    }\n    return tfc.pow(x, a);\n  });\n}\n\n/**\n * Reshapes bias tensor according to rank of x.\n */\nfunction reshapeBias(xRank: number, bias: Tensor, dataFormat: string) {\n  const biasShape = bias.shape;\n\n  if (bias.rank !== 1 && bias.rank !== xRank) {\n    throw new ValueError(\n        `Unexpected bias dimensions: ${bias.rank}` +\n        `; expected it to be 1 or ${xRank}`);\n  }\n\n  if (xRank === 5) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1, 1]);\n      } else {\n        return bias.reshape(\n            [1, biasShape[3], biasShape[0], biasShape[1], biasShape[2]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 4) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1, 1]);\n      } else {\n        return bias.reshape([1, biasShape[2], biasShape[0], biasShape[1]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank === 3) {\n    if (dataFormat === 'channelsFirst') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, biasShape[0], 1]);\n      } else {\n        return bias.reshape([1, biasShape[1], biasShape[0]]);\n      }\n    } else if (dataFormat === 'channelsLast') {\n      if (biasShape.length === 1) {\n        return bias.reshape([1, 1, biasShape[0]]);\n      } else {\n        return bias.reshape([1].concat(biasShape));\n      }\n    }\n  } else if (xRank < 3) {\n    return bias;\n  }\n  throw new ValueError(`Unsupported input rank by biasAdd: ${bias.rank}`);\n}\n\n/* Neural-network operations. */\n\n/**\n * Add a bias to a tensor.\n *\n * @param x The tensor to add the bias to.\n * @param bias The bias to add to `x`. Must be 1D or the same rank as `x`.\n * @return Result of the bias adding.\n * @throws ValueError: If the rank of `bias` is incorrect.\n */\nexport function biasAdd(\n    x: Tensor, bias: Tensor, dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n\n    return x.add(reshapeBias(x.rank, bias, dataFormat));\n  });\n}\n\n/**\n * Exponential linear unit (ELU).\n * @param x A tensor or variable to compute the activation function for.\n * @param alpha: A scalar, a scaling factor for the negative section.\n * @return Output of the ELU operation.\n */\nexport function elu(x: Tensor, alpha = 1): Tensor {\n  // TODO(cais): Add support for alpha values other than 1.\n  if (alpha !== 1) {\n    throw new NotImplementedError(\n        `Support for alpha values other than 1 (${alpha}) is not implemented ` +\n        `yet.`);\n  }\n  return tfc.elu(x);\n}\n\n/**\n * Softsign of a tensor.\n *\n * Defined as x / (abs(x) + 1), element-wise.\n *\n * @param x: Input.\n * @returns Output.\n */\nexport function softsign(x: Tensor): Tensor {\n  return tidy(() => tfc.div(x, tfc.abs(x).add(1)));\n}\n\n/**\n * Sets entries in `x` to zero at random, while scaling the entire tensor.\n *\n * @param x input tensor.\n * @param level fraction of the entries in the tensor that will be set to 0.\n * @param noiseShape shape of randomly generated keep/drop flags, must be\n *   broadcastable to the shape of `x`. Optional.\n * @param seed random seed to ensure determinism. Optional.\n * @returns Result of the dropout operation.\n */\nexport function dropout(\n    x: Tensor, level: number, noiseShape?: number[], seed?: number): Tensor {\n  return tidy(() => tfc.dropout(x, level, noiseShape, seed));\n}\n\n/**\n * Element-wise, segment-wise linear approximation of sigmoid.\n *\n * Returns `0.` if `x < -2.5`, `1.` if `x > 2.5`.\n * In `-2.5 <= x <= 2.5`, returns `0.2 * x + 0.5`.\n *\n * @param x Input tensor.\n * @returns Output tensor.\n */\nexport function hardSigmoid(x: Tensor): Tensor {\n  return tidy(() => {\n    const y = tfc.add(.5, tfc.mul(.2, x));\n    return tfc.clipByValue(y, 0, 1);\n  });\n}\n\n/**\n * Invoke `x` in the training phase, and `alt` otherwise.\n *\n * Porting Note: We do not create placeholder tensors for the `training`\n * boolean flag here, because there is no such thing in the TF.js imperative\n * backend.\n *\n * @param x The function to invoke iff `training` is `true`.\n * @param alt The function to invoke iff `training` is `false`.\n * @param training Boolean flag for whether training phase is active.\n * @returns The return value of `x()` if `training` is `true`, or the return\n *   value of `alt()` if `training` is `false`.\n */\nexport function inTrainPhase<T>(x: () => T, alt: () => T, training = false): T {\n  return training ? x() : alt();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {BaseSerialization} from './types';\n\n// TODO(soergel): Move the CamelCase versions back out of keras_format\n// e.g. to src/common.ts.  Maybe even duplicate *all* of these to be pedantic?\n/** @docinline */\nexport type FanMode = 'fanIn'|'fanOut'|'fanAvg';\nexport const VALID_FAN_MODE_VALUES = ['fanIn', 'fanOut', 'fanAvg'];\n\n// These constants have a snake vs. camel distinction.\nexport type FanModeSerialization = 'fan_in'|'fan_out'|'fan_avg';\n\n/** @docinline */\nexport type Distribution = 'normal'|'uniform'|'truncatedNormal';\nexport const VALID_DISTRIBUTION_VALUES =\n    ['normal', 'uniform', 'truncatedNormal'];\n// These constants have a snake vs. camel distinction.\nexport type DistributionSerialization = 'normal'|'uniform'|'truncated_normal';\n\nexport type ZerosSerialization = BaseSerialization<'Zeros', {}>;\n\nexport type OnesSerialization = BaseSerialization<'Ones', {}>;\n\nexport type ConstantConfig = {\n  value: number;\n};\n\nexport type ConstantSerialization =\n    BaseSerialization<'Constant', ConstantConfig>;\n\nexport type RandomNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type RandomNormalSerialization =\n    BaseSerialization<'RandomNormal', RandomNormalConfig>;\n\nexport type RandomUniformConfig = {\n  minval?: number;\n  maxval?: number;\n  seed?: number;\n};\n\nexport type RandomUniformSerialization =\n    BaseSerialization<'RandomUniform', RandomUniformConfig>;\n\nexport type TruncatedNormalConfig = {\n  mean?: number;\n  stddev?: number;\n  seed?: number;\n};\n\nexport type TruncatedNormalSerialization =\n    BaseSerialization<'TruncatedNormal', TruncatedNormalConfig>;\n\nexport type VarianceScalingConfig = {\n  scale?: number;\n\n  mode?: FanModeSerialization;\n  distribution?: DistributionSerialization;\n  seed?: number;\n};\n\nexport type VarianceScalingSerialization =\n    BaseSerialization<'VarianceScaling', VarianceScalingConfig>;\n\nexport type OrthogonalConfig = {\n  seed?: number;\n  gain?: number;\n};\n\nexport type OrthogonalSerialization =\n    BaseSerialization<'Orthogonal', OrthogonalConfig>;\n\nexport type IdentityConfig = {\n  gain?: number;\n};\n\nexport type IdentitySerialization =\n    BaseSerialization<'Identity', IdentityConfig>;\n\n// Update initializerClassNames below in concert with this.\nexport type InitializerSerialization = ZerosSerialization|OnesSerialization|\n    ConstantSerialization|RandomUniformSerialization|RandomNormalSerialization|\n    TruncatedNormalSerialization|IdentitySerialization|\n    VarianceScalingSerialization|OrthogonalSerialization;\n\nexport type InitializerClassName = InitializerSerialization['class_name'];\n\n// We can't easily extract a string[] from the string union type, but we can\n// recapitulate the list, enforcing at compile time that the values are valid\n// and that we have the right number of them.\n\n/**\n * A string array of valid Initializer class names.\n *\n * This is guaranteed to match the `InitializerClassName` union type.\n */\nexport const initializerClassNames: InitializerClassName[] = [\n  'Zeros', 'Ones', 'Constant', 'RandomNormal', 'RandomUniform',\n  'TruncatedNormal', 'VarianceScaling', 'Orthogonal', 'Identity'\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, eye, linalg, mul, ones, randomUniform, scalar, serialization, Tensor, Tensor2D, tidy, truncatedNormal, zeros} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {checkDataFormat} from './common';\nimport {NotImplementedError, ValueError} from './errors';\nimport {DataFormat, Shape} from './keras_format/common';\nimport {Distribution, FanMode, VALID_DISTRIBUTION_VALUES, VALID_FAN_MODE_VALUES} from './keras_format/initializer_config';\nimport {checkStringTypeUnionValue, deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\nimport {arrayProd} from './utils/math_utils';\n\nexport function checkFanMode(value?: string): void {\n  checkStringTypeUnionValue(VALID_FAN_MODE_VALUES, 'FanMode', value);\n}\n\nexport function checkDistribution(value?: string): void {\n  checkStringTypeUnionValue(VALID_DISTRIBUTION_VALUES, 'Distribution', value);\n}\n\n/**\n * Initializer base class.\n *\n * @doc {\n *   heading: 'Initializers', subheading: 'Classes', namespace: 'initializers'}\n */\nexport abstract class Initializer extends serialization.Serializable {\n  public fromConfigUsesCustomObjects(): boolean {\n    return false;\n  }\n  /**\n   * Generate an initial value.\n   * @param shape\n   * @param dtype\n   * @return The init value.\n   */\n  abstract apply(shape: Shape, dtype?: DataType): Tensor;\n\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\nexport class Zeros extends Initializer {\n  /** @nocollapse */\n  static className = 'Zeros';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return zeros(shape, dtype);\n  }\n}\nserialization.registerClass(Zeros);\n\nexport class Ones extends Initializer {\n  /** @nocollapse */\n  static className = 'Ones';\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return ones(shape, dtype);\n  }\n}\nserialization.registerClass(Ones);\n\nexport interface ConstantArgs {\n  /** The value for each element in the variable. */\n  value: number;\n}\n\nexport class Constant extends Initializer {\n  /** @nocollapse */\n  static className = 'Constant';\n  private value: number;\n  constructor(args: ConstantArgs) {\n    super();\n    if (typeof args !== 'object') {\n      throw new ValueError(\n          `Expected argument of type ConstantConfig but got ${args}`);\n    }\n    if (args.value === undefined) {\n      throw new ValueError(`config must have value set but got ${args}`);\n    }\n    this.value = args.value;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => mul(scalar(this.value), ones(shape, dtype)));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      value: this.value,\n    };\n  }\n}\nserialization.registerClass(Constant);\n\nexport interface RandomUniformArgs {\n  /** Lower bound of the range of random values to generate. */\n  minval?: number;\n  /** Upper bound of the range of random values to generate. */\n  maxval?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomUniform extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomUniform';\n  readonly DEFAULT_MINVAL = -0.05;\n  readonly DEFAULT_MAXVAL = 0.05;\n  private minval: number;\n  private maxval: number;\n  private seed: number;\n\n  constructor(args: RandomUniformArgs) {\n    super();\n    this.minval = args.minval || this.DEFAULT_MINVAL;\n    this.maxval = args.maxval || this.DEFAULT_MAXVAL;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return randomUniform(shape, this.minval, this.maxval, dtype);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {minval: this.minval, maxval: this.maxval, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomUniform);\n\nexport interface RandomNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class RandomNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'RandomNormal';\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: RandomNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `randomNormal does not support dType ${dtype}.`);\n    }\n\n    return K.randomNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(RandomNormal);\n\nexport interface TruncatedNormalArgs {\n  /** Mean of the random values to generate. */\n  mean?: number;\n  /** Standard deviation of the random values to generate. */\n  stddev?: number;\n  /** Used to seed the random generator. */\n  seed?: number;\n}\n\nexport class TruncatedNormal extends Initializer {\n  /** @nocollapse */\n  static className = 'TruncatedNormal';\n\n  readonly DEFAULT_MEAN = 0.;\n  readonly DEFAULT_STDDEV = 0.05;\n  private mean: number;\n  private stddev: number;\n  private seed: number;\n\n  constructor(args: TruncatedNormalArgs) {\n    super();\n    this.mean = args.mean || this.DEFAULT_MEAN;\n    this.stddev = args.stddev || this.DEFAULT_STDDEV;\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    dtype = dtype || 'float32';\n    if (dtype !== 'float32' && dtype !== 'int32') {\n      throw new NotImplementedError(\n          `truncatedNormal does not support dType ${dtype}.`);\n    }\n    return truncatedNormal(shape, this.mean, this.stddev, dtype, this.seed);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {mean: this.mean, stddev: this.stddev, seed: this.seed};\n  }\n}\nserialization.registerClass(TruncatedNormal);\n\nexport interface IdentityArgs {\n  /**\n   * Multiplicative factor to apply to the identity matrix.\n   */\n  gain?: number;\n}\n\nexport class Identity extends Initializer {\n  /** @nocollapse */\n  static className = 'Identity';\n  private gain: number;\n  constructor(args: IdentityArgs) {\n    super();\n    this.gain = args.gain != null ? args.gain : 1.0;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length !== 2 || shape[0] !== shape[1]) {\n        throw new ValueError(\n            'Identity matrix initializer can only be used for' +\n            ' 2D square matrices.');\n      } else {\n        return mul(this.gain, eye(shape[0]));\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {gain: this.gain};\n  }\n}\nserialization.registerClass(Identity);\n\n/**\n * Computes the number of input and output units for a weight shape.\n * @param shape Shape of weight.\n * @param dataFormat data format to use for convolution kernels.\n *   Note that all kernels in Keras are standardized on the\n *   CHANNEL_LAST ordering (even when inputs are set to CHANNEL_FIRST).\n * @return An length-2 array: fanIn, fanOut.\n */\nfunction computeFans(\n    shape: Shape, dataFormat: DataFormat = 'channelsLast'): number[] {\n  let fanIn: number;\n  let fanOut: number;\n  checkDataFormat(dataFormat);\n  if (shape.length === 2) {\n    fanIn = shape[0];\n    fanOut = shape[1];\n  } else if ([3, 4, 5].indexOf(shape.length) !== -1) {\n    if (dataFormat === 'channelsFirst') {\n      const receptiveFieldSize = arrayProd(shape, 2);\n      fanIn = shape[1] * receptiveFieldSize;\n      fanOut = shape[0] * receptiveFieldSize;\n    } else if (dataFormat === 'channelsLast') {\n      const receptiveFieldSize = arrayProd(shape, 0, shape.length - 2);\n      fanIn = shape[shape.length - 2] * receptiveFieldSize;\n      fanOut = shape[shape.length - 1] * receptiveFieldSize;\n    }\n  } else {\n    const shapeProd = arrayProd(shape);\n    fanIn = Math.sqrt(shapeProd);\n    fanOut = Math.sqrt(shapeProd);\n  }\n\n  return [fanIn, fanOut];\n}\n\nexport interface VarianceScalingArgs {\n  /** Scaling factor (positive float). */\n  scale?: number;\n\n  /** Fanning mode for inputs and outputs. */\n  mode?: FanMode;\n\n  /** Probabilistic distribution of the values. */\n  distribution?: Distribution;\n\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class VarianceScaling extends Initializer {\n  /** @nocollapse */\n  static className = 'VarianceScaling';\n  private scale: number;\n  private mode: FanMode;\n  private distribution: Distribution;\n  private seed: number;\n\n  /**\n   * Constructor of VarianceScaling.\n   * @throws ValueError for invalid value in scale.\n   */\n  constructor(args: VarianceScalingArgs) {\n    super();\n    if (args.scale < 0.0) {\n      throw new ValueError(\n          `scale must be a positive float. Got: ${args.scale}`);\n    }\n    this.scale = args.scale == null ? 1.0 : args.scale;\n    this.mode = args.mode == null ? 'fanIn' : args.mode;\n    checkFanMode(this.mode);\n    this.distribution =\n        args.distribution == null ? 'normal' : args.distribution;\n    checkDistribution(this.distribution);\n    this.seed = args.seed;\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    const fans = computeFans(shape);\n    const fanIn = fans[0];\n    const fanOut = fans[1];\n    let scale = this.scale;\n    if (this.mode === 'fanIn') {\n      scale /= Math.max(1, fanIn);\n    } else if (this.mode === 'fanOut') {\n      scale /= Math.max(1, fanOut);\n    } else {\n      scale /= Math.max(1, (fanIn + fanOut) / 2);\n    }\n\n    if (this.distribution === 'normal') {\n      const stddev = Math.sqrt(scale);\n      dtype = dtype || 'float32';\n      if (dtype !== 'float32' && dtype !== 'int32') {\n        throw new NotImplementedError(\n            `${this.getClassName()} does not support dType ${dtype}.`);\n      }\n      return truncatedNormal(shape, 0, stddev, dtype, this.seed);\n    } else {\n      const limit = Math.sqrt(3 * scale);\n      return randomUniform(shape, -limit, limit, dtype);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      scale: this.scale,\n      mode: this.mode,\n      distribution: this.distribution,\n      seed: this.seed\n    };\n  }\n}\nserialization.registerClass(VarianceScaling);\n\nexport interface SeedOnlyInitializerArgs {\n  /** Random number generator seed. */\n  seed?: number;\n}\n\nexport class GlorotUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotUniform';\n\n  /**\n   * Constructor of GlorotUniform\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotUniform);\n\nexport class GlorotNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'GlorotNormal';\n\n  /**\n   * Constructor of GlorotNormal.\n   * @param scale\n   * @param mode\n   * @param distribution\n   * @param seed\n   */\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanAvg',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, GlorotNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(GlorotNormal);\n\nexport class HeNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeNormal);\n\nexport class HeUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'HeUniform';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 2.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, HeUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(HeUniform);\n\nexport class LeCunNormal extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'normal',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunNormal is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunNormal);\n\nexport class LeCunUniform extends VarianceScaling {\n  /** @nocollapse */\n  static className = 'LeCunNormal';\n\n  constructor(args?: SeedOnlyInitializerArgs) {\n    super({\n      scale: 1.0,\n      mode: 'fanIn',\n      distribution: 'uniform',\n      seed: args == null ? null : args.seed\n    });\n  }\n\n  getClassName(): string {\n    // In Python Keras, LeCunUniform is not a class, but a helper method\n    // that creates a VarianceScaling object. Use 'VarianceScaling' as\n    // class name to be compatible with that.\n    return VarianceScaling.className;\n  }\n}\nserialization.registerClass(LeCunUniform);\n\nexport interface OrthogonalArgs extends SeedOnlyInitializerArgs {\n  /**\n   * Multiplicative factor to apply to the orthogonal matrix. Defaults to 1.\n   */\n  gain?: number;\n}\n\nexport class Orthogonal extends Initializer {\n  /** @nocollapse */\n  static className = 'Orthogonal';\n  readonly DEFAULT_GAIN = 1;\n  protected readonly gain: number;\n  protected readonly seed: number;\n\n  constructor(args?: OrthogonalArgs) {\n    super();\n    this.gain = args.gain == null ? this.DEFAULT_GAIN : args.gain;\n    this.seed = args.seed;\n\n    if (this.seed != null) {\n      throw new NotImplementedError(\n          'Random seed is not implemented for Orthogonal Initializer yet.');\n    }\n  }\n\n  apply(shape: Shape, dtype?: DataType): Tensor {\n    return tidy(() => {\n      if (shape.length < 2) {\n        throw new NotImplementedError('Shape must be at least 2D.');\n      }\n      if (shape[0] * shape[1] > 2000) {\n        console.warn(\n            `Orthogonal initializer is being called on a matrix with more ` +\n            `than 2000 (${shape[0] * shape[1]}) elements: ` +\n            `Slowness may result.`);\n      }\n\n      // TODO(cais): Add seed support.\n      const normalizedShape =\n          shape[0] > shape[1] ? [shape[1], shape[0]] : shape;\n      const a = K.randomNormal(normalizedShape, 0, 1, 'float32') as Tensor2D;\n      let q = linalg.gramSchmidt(a) as Tensor2D;\n      if (shape[0] > shape[1]) {\n        q = q.transpose();\n      }\n      return mul(this.gain, q);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      gain: this.gain,\n      seed: this.seed,\n    };\n  }\n}\nserialization.registerClass(Orthogonal);\n\n/** @docinline */\nexport type InitializerIdentifier =\n    'constant'|'glorotNormal'|'glorotUniform'|'heNormal'|'heUniform'|'identity'|\n    'leCunNormal'|'leCunUniform'|'ones'|'orthogonal'|'randomNormal'|\n    'randomUniform'|'truncatedNormal'|'varianceScaling'|'zeros'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding registry\n// symbols.\nexport const INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in InitializerIdentifier]: string} = {\n      'constant': 'Constant',\n      'glorotNormal': 'GlorotNormal',\n      'glorotUniform': 'GlorotUniform',\n      'heNormal': 'HeNormal',\n      'heUniform': 'HeUniform',\n      'identity': 'Identity',\n      'leCunNormal': 'LeCunNormal',\n      'leCunUniform': 'LeCunUniform',\n      'ones': 'Ones',\n      'orthogonal': 'Orthogonal',\n      'randomNormal': 'RandomNormal',\n      'randomUniform': 'RandomUniform',\n      'truncatedNormal': 'TruncatedNormal',\n      'varianceScaling': 'VarianceScaling',\n      'zeros': 'Zeros'\n    };\n\nfunction deserializeInitializer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Initializer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'initializer');\n}\n\nexport function serializeInitializer(initializer: Initializer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(initializer);\n}\n\nexport function getInitializer(identifier: InitializerIdentifier|Initializer|\n                               serialization.ConfigDict): Initializer {\n  if (typeof identifier === 'string') {\n    const className = identifier in INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        INITIALIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    /* We have four 'helper' classes for common initializers that\n    all get serialized as 'VarianceScaling' and shouldn't go through\n    the deserializeInitializer pathway. */\n    if (className === 'GlorotNormal') {\n      return new GlorotNormal();\n    } else if (className === 'GlorotUniform') {\n      return new GlorotUniform();\n    } else if (className === 'HeNormal') {\n      return new HeNormal();\n    } else if (className === 'HeUniform') {\n      return new HeUniform();\n    } else if (className === 'LeCunNormal') {\n      return new LeCunNormal();\n    } else if (className === 'LeCunUniform') {\n      return new LeCunUniform();\n    } else {\n      const config: serialization.ConfigDict = {};\n      config['className'] = className;\n      config['config'] = {};\n      return deserializeInitializer(config);\n    }\n  } else if (identifier instanceof Initializer) {\n    return identifier;\n  } else {\n    return deserializeInitializer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n// tslint:disable-next-line:max-line-length\nimport {Constant, ConstantArgs, GlorotNormal, GlorotUniform, HeNormal, HeUniform, Identity, IdentityArgs, Initializer, LeCunNormal, LeCunUniform, Ones, Orthogonal, OrthogonalArgs, RandomNormal, RandomNormalArgs, RandomUniform, RandomUniformArgs, SeedOnlyInitializerArgs, TruncatedNormal, TruncatedNormalArgs, VarianceScaling, VarianceScalingArgs, Zeros} from './initializers';\n\n/**\n * Initializer that generates tensors initialized to 0.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function zeros(): Zeros {\n  return new Zeros();\n}\n\n/**\n * Initializer that generates tensors initialized to 1.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function ones(): Initializer {\n  return new Ones();\n}\n\n/**\n * Initializer that generates values initialized to some constant.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function constant(args: ConstantArgs): Initializer {\n  return new Constant(args);\n}\n\n/**\n * Initializer that generates random values initialized to a uniform\n * distribution.\n *\n * Values will be distributed uniformly between the configured minval and\n * maxval.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomUniform(args: RandomUniformArgs): Initializer {\n  return new RandomUniform(args);\n}\n\n/**\n * Initializer that generates random values initialized to a normal\n * distribution.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function randomNormal(args: RandomNormalArgs): Initializer {\n  return new RandomNormal(args);\n}\n\n/**\n * Initializer that generates random values initialized to a truncated normal.\n * distribution.\n *\n * These values are similar to values from a `RandomNormal` except that values\n * more than two standard deviations from the mean are discarded and re-drawn.\n * This is the recommended initializer for neural network weights and filters.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function truncatedNormal(args: TruncatedNormalArgs): Initializer {\n  return new TruncatedNormal(args);\n}\n\n/**\n * Initializer that generates the identity matrix.\n * Only use for square 2D matrices.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function identity(args: IdentityArgs): Initializer {\n  return new Identity(args);\n}\n\n/**\n * Initializer capable of adapting its scale to the shape of weights.\n * With distribution=NORMAL, samples are drawn from a truncated normal\n * distribution centered on zero, with `stddev = sqrt(scale / n)` where n is:\n *   - number of input units in the weight tensor, if mode = FAN_IN.\n *   - number of output units, if mode = FAN_OUT.\n *   - average of the numbers of input and output units, if mode = FAN_AVG.\n * With distribution=UNIFORM,\n * samples are drawn from a uniform distribution\n * within [-limit, limit], with `limit = sqrt(3 * scale / n)`.\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function varianceScaling(config: VarianceScalingArgs): Initializer {\n  return new VarianceScaling(config);\n}\n\n/**\n * Glorot uniform initializer, also called Xavier uniform initializer.\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotUniform(args);\n}\n\n/**\n * Glorot normal initializer, also called Xavier normal initializer.\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / (fan_in + fan_out))`\n * where `fan_in` is the number of input units in the weight tensor\n * and `fan_out` is the number of output units in the weight tensor.\n *\n * Reference:\n *   Glorot & Bengio, AISTATS 2010\n *       http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function glorotNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new GlorotNormal(args);\n}\n\n/**\n * He normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(2 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function heNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeNormal(args);\n}\n\n/**\n * He uniform initializer.\n *\n * It draws samples from a uniform distribution within [-limit, limit]\n * where `limit` is `sqrt(6 / fan_in)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * Reference:\n *     He et al., http://arxiv.org/abs/1502.01852\n *\n * @doc {heading: 'Initializers',namespace: 'initializers'}\n */\nexport function heUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new HeUniform(args);\n}\n\n/**\n * LeCun normal initializer.\n *\n * It draws samples from a truncated normal distribution centered on 0\n * with `stddev = sqrt(1 / fanIn)`\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * References:\n *   [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *   [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunNormal(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunNormal(args);\n}\n\n/**\n * LeCun uniform initializer.\n *\n * It draws samples from a uniform distribution in the interval\n * `[-limit, limit]` with `limit = sqrt(3 / fanIn)`,\n * where `fanIn` is the number of input units in the weight tensor.\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function leCunUniform(args: SeedOnlyInitializerArgs): Initializer {\n  return new LeCunUniform(args);\n}\n\n/**\n * Initializer that generates a random orthogonal matrix.\n *\n * Reference:\n * [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)\n *\n * @doc {heading: 'Initializers', namespace: 'initializers'}\n */\nexport function orthogonal(args: OrthogonalArgs): Initializer {\n  return new Orthogonal(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Utilities related to persistent state in the backend.\n */\n\n/**\n * An ID to track `tf.SymbolicTensor`s and derived classes.\n * Required in different places in engine/topology.ts to identify unique\n * tensors.\n */\nlet _nextUniqueTensorId = 0;\n\nexport function getNextUniqueTensorId(): number {\n  return _nextUniqueTensorId++;\n}\n\nconst _uidPrefixes: {[prefix: string]: number} = {};\n\n/**\n * Provides a unique UID given a string prefix.\n *\n * @param prefix\n */\nexport function getUid(prefix = ''): string {\n  if (!(prefix in _uidPrefixes)) {\n    _uidPrefixes[prefix] = 0;\n  }\n  _uidPrefixes[prefix] += 1;\n  return prefix + _uidPrefixes[prefix].toString();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: utils/generic_utils.py */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\n// tslint:enable\n\n/**\n * Determine whether the input is an Array of Shapes.\n */\nexport function isArrayOfShapes(x: Shape|Shape[]): boolean {\n  return Array.isArray(x) && Array.isArray(x[0]);\n}\n\n/**\n * Special case of normalizing shapes to lists.\n *\n * @param x A shape or list of shapes to normalize into a list of Shapes.\n * @return A list of Shapes.\n */\nexport function normalizeShapeList(x: Shape|Shape[]): Shape[] {\n  if (x.length === 0) {\n    return [];\n  }\n  if (!Array.isArray(x[0])) {\n    return [x] as Shape[];\n  }\n  return x as Shape[];\n}\n\n/**\n * Helper function to obtain exactly one Tensor.\n * @param xs: A single `tf.Tensor` or an `Array` of `tf.Tensor`s.\n * @return A single `tf.Tensor`. If `xs` is an `Array`, return the first one.\n * @throws ValueError: If `xs` is an `Array` and its length is not 1.\n */\nexport function getExactlyOneTensor(xs: Tensor|Tensor[]): Tensor {\n  let x: Tensor;\n  if (Array.isArray(xs)) {\n    if (xs.length !== 1) {\n      throw new ValueError(`Expected Tensor length to be 1; got ${xs.length}`);\n    }\n    x = xs[0];\n  } else {\n    x = xs;\n  }\n  return x;\n}\n\n/**\n * Helper function to obtain exactly on instance of Shape.\n *\n * @param shapes Input single `Shape` or Array of `Shape`s.\n * @returns If input is a single `Shape`, return it unchanged. If the input is\n *   an `Array` containing exactly one instance of `Shape`, return the instance.\n *   Otherwise, throw a `ValueError`.\n * @throws ValueError: If input is an `Array` of `Shape`s, and its length is not\n *   1.\n */\nexport function getExactlyOneShape(shapes: Shape|Shape[]): Shape {\n  if (Array.isArray(shapes) && Array.isArray(shapes[0])) {\n    if (shapes.length === 1) {\n      shapes = shapes as Shape[];\n      return shapes[0];\n    } else {\n      throw new ValueError(`Expected exactly 1 Shape; got ${shapes.length}`);\n    }\n  } else {\n    return shapes as Shape;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {LayerVariable} from '../variables';\n\n/**\n * Count the elements in an Array of LayerVariables.\n *\n * @param weights: The LayerVariables of which the constituent numbers are to\n *   be counted.\n * @returns A count of the elements in all the LayerVariables\n */\nexport function countParamsInWeights(weights: LayerVariable[]): number {\n  let count = 0;\n  for (const weight of weights) {\n    if (weight.shape.length === 0) {\n      count += 1;\n    } else {\n      count += weight.shape.reduce((a, b) => a * b);\n    }\n  }\n  return count;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, Tensor, variableGrads} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId} from './backend/state';\nimport {getScopedTensorName, getUniqueTensorName} from './common';\nimport {Constraint} from './constraints';\nimport {NotImplementedError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {HasShape} from './types';\n\nconst DEFAULT_VARIABLE_NAME_PREFIX = 'Variable';\n\n/**\n * A `tf.layers.LayerVariable` is similar to a `tf.Tensor` in that it has a\n * dtype and shape, but its value is mutable.  The value is itself represented\n * as a`tf.Tensor`, and can be read with the `read()` method and updated with\n * the `write()` method.\n */\nexport class LayerVariable {\n  readonly dtype: DataType;\n  readonly shape: Shape;\n\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName: string;\n  private trainable_: boolean;\n\n  protected readonly val: tfc.Variable;\n  readonly constraint: Constraint;\n  /**\n   * Construct Variable from a `tf.Tensor`.\n   *\n   * If not explicitly named, the Variable will be given a name with the\n   * prefix 'Variable'. Variable names are unique. In the case of name\n   * collision, suffixies '_<num>' will be added to the name.\n   *\n   * @param val Initial value of the Variable.\n   * @param name Name of the variable. If `null` or `undefined` is provided, it\n   *   will default a name with the prefix 'Variable'.\n   * @param constraint Optional, projection function to be applied to the\n   * variable after optimize updates\n   * @throws ValueError if `name` is `null` or `undefined`.\n   */\n  constructor(\n      val: Tensor, dtype: DataType = 'float32',\n      name = DEFAULT_VARIABLE_NAME_PREFIX, trainable = true,\n      constraint: Constraint = null) {\n    this.dtype = dtype == null ? 'float32' : dtype;\n    this.shape = val.shape;\n    this.id = getNextUniqueTensorId();\n\n    name = name == null ? DEFAULT_VARIABLE_NAME_PREFIX : name;\n    this.originalName = getScopedTensorName(name);\n    this.name = getUniqueTensorName(this.originalName);\n\n    this.trainable_ = trainable;\n    this.constraint = constraint;\n\n    this.val = tfc.variable(val, this.trainable_, this.name, this.dtype);\n  }\n\n  /**\n   * Get a snapshot of the Variable's value.\n   *\n   * The returned value is a snapshot of the Variable's value at the time of\n   * the invocation. Future mutations in the value of the tensor will only\n   * be reflected by future calls to this method.\n   */\n  read(): Tensor {\n    this.assertNotDisposed();\n    return this.val;\n  }\n\n  /**\n   * Update the value of the Variable.\n   *\n   * @param newVal: The new value to update to. Must be consistent with the\n   *   dtype and shape of the Variable.\n   * @return This Variable.\n   */\n  write(newVal: Tensor) {\n    // TODO(cais): Once  TF.js Core supports Tensor.dtype, check dtype match.\n    this.assertNotDisposed();\n    checkShapesMatch(this.val, newVal);\n    // Skip updating if this is the exact same tensor.\n    if (this.val.id !== newVal.id) {\n      this.val.assign(newVal);\n      if (this.constraint != null) {\n        this.val.assign(this.constraint.apply(this.val));\n      }\n    }\n    return this;\n  }\n\n  /**\n   * Dispose this LayersVariable instance from memory.\n   */\n  dispose(): void {\n    this.assertNotDisposed();\n    this.val.dispose();\n  }\n\n  protected assertNotDisposed(): void {\n    if (this.val.isDisposed) {\n      throw new Error(`LayersVariable ${this.name} is already disposed.`);\n    }\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.trainable_ = trainable;\n    this.val.trainable = trainable;\n  }\n}\n\nfunction checkShapesMatch(x: HasShape, y: HasShape): void {\n  if (x.shape.toString() !== y.shape.toString()) {\n    throw new Error(\n        'Shape mismatch: ' + JSON.stringify(x.shape) + ' vs. ' +\n        JSON.stringify(y.shape));\n  }\n}\n\n/**\n * Create a Variable.\n * @param x The initial value of the `Variable`.\n * @param dtype optional, the type of the variable.\n * @param name optional, the name of the variable, default provided by\n * Variable.\n * @param constraint optional, a constraint to be applied after every update.\n * @return The newly instantiated `Variable`.\n */\nexport function variable(\n    x: Tensor, dtype?: DataType, name?: string,\n    constraint?: Constraint): LayerVariable {\n  return new LayerVariable(x, dtype, name, true, constraint);\n}\n\n/**\n * Instantiates an all-zeros Variable and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-zero Variable.\n */\nexport function zerosVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  return new LayerVariable(tfc.zeros(shape), dtype, name);\n}\n\n/**\n * Instantiates an all-zeros tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function zerosLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.zerosLike(x), dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor and returns it.\n *\n * @param shape Shape of the tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return An all-ones Variable.\n */\nexport function onesVariable(\n    shape: Shape, dtype?: DataType, name?: string): LayerVariable {\n  // TODO(cais): Implement logic for dtype.\n  const allocated = tfc.ones(shape);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiates an all-ones tensor of the same shape as another tensor.\n *\n * @param x The other tensor.\n * @param dtype DType of the tensor.\n * @param name Name of the tensor.\n * @return A newly instantiated Variable.\n */\nexport function onesLike(\n    x: Tensor, dtype?: DataType, name?: string): LayerVariable {\n  const allocated = tfc.onesLike(x);\n  return new LayerVariable(allocated, dtype, name);\n}\n\n/**\n * Instantiate an identity matrix and returns it, as a Variable\n *\n * @param size Number of rows/columns.\n * @param dtype Data type of returned Variable.\n * @param name Name of returned Variable.\n * @return A Variable, an identity matrix.\n */\nexport function eyeVariable(\n    size: number, dtype?: DataType, name?: string): LayerVariable {\n  return new LayerVariable(tfc.eye(size), dtype, name);\n}\n\n/**\n * Get a Variable with uniform distribution of values.\n * @param shape Shape of the tensor.\n * @param minval Lower bound of the uniform distribution.\n * @param maxval Upper bound of the uniform distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The uniform-random Variable.\n */\nexport function randomUniformVariable(\n    shape: Shape, minval: number, maxval: number, dtype?: DataType,\n    seed?: number, name = 'randomUniform'): LayerVariable {\n  return new LayerVariable(\n      tfc.randomUniform(shape, minval, maxval, dtype), dtype, name);\n}\n\n/**\n * Get a Variable with truncated-normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function truncatedNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'truncatedNormal'): LayerVariable {\n  // TODO(cais): Implement logic for dtype and seed once they are supported\n  // by deeplearn.js.\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormal does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.truncatedNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n/**\n * Get a Variable with normal distribution of values.\n * @param shape Shape of the tensor.\n * @param mean mean value of the normal distribution.\n * @param stddev standard deviation of the normal distribution.\n * @param dtype\n * @param seed\n * @param name Optional name.\n * @return The truncated-normal-random Variable.\n */\nexport function randomNormalVariable(\n    shape: Shape, mean = 0.0, stddev = 1.0, dtype?: DataType, seed?: number,\n    name = 'randomNormal'): LayerVariable {\n  dtype = dtype || 'float32';\n  if (dtype !== 'float32' && dtype !== 'int32') {\n    throw new NotImplementedError(\n        `randomNormalVariable does not support dType ${dtype}.`);\n  }\n  return new LayerVariable(\n      tfc.randomNormal(shape, mean, stddev, dtype, seed), dtype, name);\n}\n\n/**\n * Update the value of a Variable.\n * @param x The Variable to be updated.\n * @param xNew The new value to update to.\n * @return The Variable updated.\n */\nexport function update(x: LayerVariable, xNew: Tensor): LayerVariable {\n  return x.write(xNew);\n}\n\n/**\n * Update the value of a Variable by adding an increment.\n * @param x The Variable to be updated.\n * @param increment The incrment to add to `x`.\n * @return The Variable updated.\n */\nexport function updateAdd(x: LayerVariable, increment: Tensor): LayerVariable {\n  return x.write(tfc.add(x.read(), increment));\n}\n\n/**\n * Update the value of a Variable by subtracting a decrement.\n * @param x The Variable to be updated.\n * @param decrement The decrement to subtract from `x`.\n * @return The Variable updated.\n */\nexport function updateSub(x: LayerVariable, decrement: Tensor): LayerVariable {\n  return x.write(tfc.sub(x.read(), decrement));\n}\n\n/**\n * Get the values of an array of Variables.\n *\n * @param tensors An `Array` of `Variable`s to get the values of.\n * @return The values of the inputs, as an `Array` of`tf.Tensor`s.\n */\nexport function batchGetValue(xs: LayerVariable[]): Tensor[] {\n  return xs.map(x => x.read());\n}\n\n/**\n * Update the value of multiple Variables at once.\n *\n * @param variablesAndValues An `Array`, each element is of type\n *   [Variable, Tensor]. The first item is the\n *   `Variable` of which the value is to be updated. The second item\n *   carries the new value.\n */\nexport function batchSetValue(\n    variablesAndValues: Array<[LayerVariable, Tensor]>): void {\n  variablesAndValues.forEach(variableAndValue => {\n    const variable: LayerVariable = variableAndValue[0];\n    variable.write(variableAndValue[1]);\n  });\n}\n\n/**\n * Returns the gradients of `variables` w.r.t. the return value of `lossFn`.\n * @param lossFn A function which returns a Scalar to be used as the function\n *   value (i.e., numerator) for differentiation.\n * @param variables List of variables to be used as the independent variables\n *   (i.e., denominator) for differentiation.\n * @returns An Array of gradients tensors.\n */\nexport function gradients(\n    lossFn: () => tfc.Scalar, variables: LayerVariable[]): Tensor[] {\n  // TODO(cais): The return type signature can be simplified if deeplearn makes\n  //   the corresponding type public.\n  const variableList =\n      variables.map(variable => variable.read() as tfc.Variable);\n  const valudAndGrads = variableGrads(lossFn, variableList);\n  return variables.map(variable => valudAndGrads.grads[variable.name]);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {DataType, Scalar, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {getNextUniqueTensorId, getUid} from '../backend/state';\nimport {getScopedTensorName, getUniqueTensorName, nameScope} from '../common';\nimport {Constraint} from '../constraints';\nimport {AttributeError, NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {getInitializer, Initializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs, RegularizerFn} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as types_utils from '../utils/types_utils';\nimport * as variable_utils from '../utils/variable_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\n\n// TODO(michaelterry): This is a stub until it's defined.\nexport type Op = (x: LayerVariable) => LayerVariable;\n\n/**\n * Constructor arguments for InputSpec.\n */\nexport interface InputSpecArgs {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n}\n\n/**\n * Specifies the ndim, dtype and shape of every input to a layer.\n *\n * Every layer should expose (if appropriate) an `inputSpec` attribute:\n * a list of instances of InputSpec (one per input tensor).\n *\n * A null entry in a shape is compatible with any dimension,\n * a null shape is compatible with any shape.\n */\nexport class InputSpec {\n  /** Expected datatype of the input. */\n  dtype?: DataType;\n  /** Expected shape of the input (may include null for unchecked axes). */\n  shape?: Shape;\n  /** Expected rank of the input. */\n  ndim?: number;\n  /** Maximum rank of the input. */\n  maxNDim?: number;\n  /** Minimum rank of the input. */\n  minNDim?: number;\n  /** Dictionary mapping integer axes to a specific dimension value. */\n  axes?: {[axis: number]: number};\n\n  constructor(args: InputSpecArgs) {\n    this.dtype = args.dtype;\n    this.shape = args.shape;\n    /*\n      TODO(michaelterry): Could throw error if ndim and shape are both defined\n        (then backport).\n    */\n    if (args.shape != null) {\n      this.ndim = args.shape.length;\n    } else {\n      this.ndim = args.ndim;\n    }\n    this.maxNDim = args.maxNDim;\n    this.minNDim = args.minNDim;\n    this.axes = args.axes || {};\n  }\n}\n\n/**\n * `tf.SymbolicTensor` is a placeholder for a Tensor without any concrete value.\n *\n * They are most often encountered when building a graph of `Layer`s for a\n * a `tf.LayersModel` and the input data's shape, but not values are known.\n *\n * @doc {heading: 'Models', 'subheading': 'Classes'}\n */\nexport class SymbolicTensor {\n  /* A unique ID for the tensor to be able to differentiate tensors. */\n  readonly id: number;\n  // The fully scoped name of this Variable, including a unique suffix if needed\n  readonly name: string;\n  // The originally requested fully scoped name of this Variable, not including\n  // any unique suffix.  This may be needed when restoring weights because this\n  // original name is used as a key.\n  readonly originalName?: string;\n  /**\n   * Rank/dimensionality of the tensor.\n   */\n  readonly rank: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  nodeIndex: number;\n  /**\n   * Replacement for _keras_history.\n   */\n  tensorIndex: number;\n\n  /**\n   *\n   * @param dtype\n   * @param shape\n   * @param sourceLayer The Layer that produced this symbolic tensor.\n   * @param inputs The inputs passed to sourceLayer's __call__() method.\n   * @param nodeIndex\n   * @param tensorIndex\n   * @param callArgs The keyword arguments passed to the __call__() method.\n   * @param name\n   * @param outputTensorIndex The index of this tensor in the list of outputs\n   *   returned by apply().\n   */\n  constructor(\n      readonly dtype: DataType, readonly shape: Shape,\n      public sourceLayer: Layer, readonly inputs: SymbolicTensor[],\n      readonly callArgs: Kwargs, name?: string,\n      readonly outputTensorIndex?: number) {\n    this.id = getNextUniqueTensorId();\n    if (name != null) {\n      this.originalName = getScopedTensorName(name);\n      this.name = getUniqueTensorName(this.originalName);\n    }\n    this.rank = shape.length;\n  }\n}\n\n/**\n * Constructor arguments for Node.\n */\nexport interface NodeArgs {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`.\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n}\n\n/**\n * The type of the return value of Layer.dispose() and Container.dispose().\n */\nexport interface DisposeResult {\n  /**\n   * Reference count after the dispose call.\n   */\n  refCountAfterDispose: number;\n\n  /**\n   * Number of variables dispose in this dispose call.\n   */\n  numDisposedVariables: number;\n}\n\nlet _nextNodeID = 0;\n\n/**\n * A `Node` describes the connectivity between two layers.\n *\n * Each time a layer is connected to some new input,\n * a node is added to `layer.inboundNodes`.\n *\n * Each time the output of a layer is used by another layer,\n * a node is added to `layer.outboundNodes`.\n *\n * `nodeIndices` and `tensorIndices` are basically fine-grained coordinates\n * describing the origin of the `inputTensors`, verifying the following:\n *\n * `inputTensors[i] ==\n * inboundLayers[i].inboundNodes[nodeIndices[i]].outputTensors[\n *   tensorIndices[i]]`\n *\n * A node from layer A to layer B is added to:\n *     A.outboundNodes\n *     B.inboundNodes\n */\nexport class Node {\n  /**\n   * The layer that takes `inputTensors` and turns them into `outputTensors`\n   * (the node gets created when the `call` method of the layer is called).\n   */\n  outboundLayer: Layer;\n  /**\n   * A list of layers, the same length as `inputTensors`, the layers from where\n   * `inputTensors` originate.\n   */\n  inboundLayers: Layer[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `nodeIndices[i]` is\n   * the origin node of `inputTensors[i]` (necessary since each inbound layer\n   * might have several nodes, e.g. if the layer is being shared with a\n   * different data stream).\n   */\n  nodeIndices: number[];\n  /**\n   * A list of integers, the same length as `inboundLayers`. `tensorIndices[i]`\n   * is the index of `inputTensors[i]` within the output of the inbound layer\n   * (necessary since each inbound layer might have multiple tensor outputs,\n   * with each one being independently manipulable).\n   */\n  tensorIndices: number[];\n  /** List of input tensors. */\n  inputTensors: SymbolicTensor[];\n  /** List of output tensors. */\n  outputTensors: SymbolicTensor[];\n  /** List of input masks (a mask can be a tensor, or null). */\n  inputMasks: Tensor[];\n  /** List of output masks (a mask can be a tensor, or null). */\n  outputMasks: Tensor[];\n  /** List of input shape tuples. */\n  inputShapes: Shape|Shape[];\n  /** List of output shape tuples. */\n  outputShapes: Shape|Shape[];\n\n  readonly id: number;\n\n  constructor(\n      args: NodeArgs,\n      // TODO(michaelterry): Define actual type for this.\n      public callArgs?: Kwargs) {\n    this.id = _nextNodeID++;\n    /*\n      Layer instance (NOT a list).\n      this is the layer that takes a list of input tensors\n      and turns them into a list of output tensors.\n      the current node will be added to\n      the inboundNodes of outboundLayer.\n    */\n    this.outboundLayer = args.outboundLayer;\n\n    /*\n        The following 3 properties describe where\n        the input tensors come from: which layers,\n        and for each layer, which node and which\n        tensor output of each node.\n    */\n\n    // List of layer instances.\n    this.inboundLayers = args.inboundLayers;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.nodeIndices = args.nodeIndices;\n    // List of integers, 1:1 mapping with inboundLayers.\n    this.tensorIndices = args.tensorIndices;\n\n    /*\n        Following 2 properties:\n        tensor inputs and outputs of outboundLayer.\n    */\n\n    // List of tensors. 1:1 mapping with inboundLayers.\n    this.inputTensors = args.inputTensors;\n    // List of tensors, created by outboundLayer.call().\n    this.outputTensors = args.outputTensors;\n\n    /*\n        Following 2 properties: input and output masks.\n        List of tensors, 1:1 mapping with inputTensor.\n    */\n    this.inputMasks = args.inputMasks;\n    // List of tensors, created by outboundLayer.computeMask().\n    this.outputMasks = args.outputMasks;\n\n    // Following 2 properties: input and output shapes.\n\n    // List of shape tuples, shapes of inputTensors.\n    this.inputShapes = args.inputShapes;\n    // List of shape tuples, shapes of outputTensors.\n    this.outputShapes = args.outputShapes;\n\n    // Add nodes to all layers involved.\n    for (const layer of args.inboundLayers) {\n      if (layer != null) {\n        layer.outboundNodes.push(this);\n      }\n    }\n    args.outboundLayer.inboundNodes.push(this);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const inboundNames: string[] = [];\n    for (const layer of this.inboundLayers) {\n      if (layer != null) {\n        inboundNames.push(layer.name);\n      } else {\n        inboundNames.push(null);\n      }\n    }\n    return {\n      outboundLayer: this.outboundLayer ? this.outboundLayer.name : null,\n      inboundLayers: inboundNames,\n      nodeIndices: this.nodeIndices,\n      tensorIndices: this.tensorIndices\n    };\n  }\n}\n\n/** Constructor arguments for Layer. */\nexport declare interface LayerArgs {\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  inputShape?: Shape;\n  /**\n   * If defined, will be used to create an input layer to insert before this\n   * layer. If both `inputShape` and `batchInputShape` are defined,\n   * `batchInputShape` will be used. This argument is only applicable to input\n   * layers (the first layer of a model).\n   */\n  batchInputShape?: Shape;\n  /**\n   * If `inputShape` is specified and `batchInputShape` is *not* specified,\n   * `batchSize` is used to construct the `batchInputShape`: `[batchSize,\n   * ...inputShape]`\n   */\n  batchSize?: number;\n  /**\n   * The data-type for this layer. Defaults to 'float32'.\n   * This argument is only applicable to input layers (the first layer of a\n   * model).\n   */\n  dtype?: DataType;\n  /** Name for this layer. */\n  name?: string;\n  /**\n   * Whether the weights of this layer are updatable by `fit`.\n   * Defaults to true.\n   */\n  trainable?: boolean;\n  /**\n   * Initial weight values of the layer.\n   */\n  weights?: Tensor[];\n  /** Legacy support. Do not use for new code. */\n  inputDType?: DataType;\n}\n\n// If necessary, add `output` arguments to the CallHook function.\n// This is currently used for testing only, but may be used for debugger-related\n// purposes in the future.\nexport type CallHook = (inputs: Tensor|Tensor[], kwargs: Kwargs) => void;\n\nlet _nextLayerID = 0;\n\n/**\n * A layer is a grouping of operations and weights that can be composed to\n * create a `tf.LayersModel`.\n *\n * Layers are constructed by using the functions under the\n * [tf.layers](#Layers-Basic) namespace.\n *\n * @doc {heading: 'Layers', subheading: 'Classes', namespace: 'layers'}\n */\nexport abstract class Layer extends serialization.Serializable {\n  /** Name for this layer. Must be unique within a model. */\n  name: string;\n  /**\n   * List of InputSpec class instances.\n   *\n   * Each entry describes one required input:\n   * - ndim\n   * - dtype\n   * A layer with `n` input tensors must have an `inputSpec` of length `n`.\n   */\n  inputSpec: InputSpec[];\n  supportsMasking: boolean;\n  /** Whether the layer weights will be updated during training. */\n  protected trainable_: boolean;\n  batchInputShape: Shape;\n  dtype: DataType;\n  initialWeights: Tensor[];\n\n  inboundNodes: Node[];\n  outboundNodes: Node[];\n\n  activityRegularizer: Regularizer;\n\n  protected _trainableWeights: LayerVariable[];\n  private _nonTrainableWeights: LayerVariable[];\n  private _losses: RegularizerFn[];\n  // TODO(cais): _updates is currently unused.\n  private _updates: Tensor[];\n  private _built: boolean;\n  private _callHook: CallHook = null;\n\n  private _addedWeightNames: string[] = [];\n\n  readonly id: number;\n\n  // Porting Notes: PyKeras does not have this property in this base Layer\n  //   class. Instead lets Layer subclass set it dynamically and checks the\n  //   value with `hasattr`. In tfjs-layers, we let this be a member of this\n  //   base class.\n  protected _stateful = false;\n\n  protected _refCount: number|null;\n\n  // A flag for whether fast (i.e., all-zero) weight initialization is to\n  // be used during `build()` call. This speeds up weight initialization\n  // by saving unnecessary calls to expensive initializers in cases where\n  // the initialized values will be overwritten by loaded weight values\n  // during model loading.\n  private fastWeightInitDuringBuild: boolean;\n\n  constructor(args: LayerArgs = {}) {\n    super();\n    this.id = _nextLayerID++;\n\n    this.activityRegularizer = null;\n\n    this.inputSpec = null;\n    this.supportsMasking = false;\n\n    // These properties will be set upon call of this.build()\n    this._trainableWeights = [];\n    this._nonTrainableWeights = [];\n    this._losses = [];\n    this._updates = [];\n    this._built = false;\n\n    /*\n      These lists will be filled via successive calls\n      to this.addInboundNode().\n     */\n    this.inboundNodes = [];\n    this.outboundNodes = [];\n\n    let name = args.name;\n    if (!name) {\n      const prefix = this.getClassName();\n      name = generic_utils.toSnakeCase(prefix) + '_' + getUid(prefix);\n    }\n    this.name = name;\n\n    this.trainable_ = args.trainable == null ? true : args.trainable;\n\n    if (args.inputShape != null || args.batchInputShape != null) {\n      /*\n        In this case we will later create an input layer\n        to insert before the current layer\n       */\n      let batchInputShape: Shape;\n      if (args.batchInputShape != null) {\n        batchInputShape = args.batchInputShape;\n      } else if (args.inputShape != null) {\n        let batchSize: number = null;\n        if (args.batchSize != null) {\n          batchSize = args.batchSize;\n        }\n        batchInputShape = [batchSize].concat(args.inputShape);\n      }\n      this.batchInputShape = batchInputShape;\n\n      // Set dtype.\n      let dtype = args.dtype;\n      if (dtype == null) {\n        dtype = args.inputDType;\n      }\n      if (dtype == null) {\n        dtype = 'float32';\n      }\n      this.dtype = dtype;\n    }\n\n    if (args.weights != null) {\n      this.initialWeights = args.weights;\n    } else {\n      this.initialWeights = null;\n    }\n\n    // The value of `_refCount` is initialized to null. When the layer is used\n    // in a symbolic way for the first time, it will be set to 1.\n    this._refCount = null;\n\n    this.fastWeightInitDuringBuild = false;\n  }\n\n  /**\n   * Converts a layer and its index to a unique (immutable type) name.\n   * This function is used internally with `this.containerNodes`.\n   * @param layer The layer.\n   * @param nodeIndex The layer's position (e.g. via enumerate) in a list of\n   *   nodes.\n   *\n   * @returns The unique name.\n   */\n  protected static nodeKey(layer: Layer, nodeIndex: number) {\n    return layer.name + '_ib-' + nodeIndex.toString();\n  }\n\n  /**\n   * Returns this.inboundNode at index nodeIndex.\n   *\n   * Porting note: This is a replacement for _get_node_attribute_at_index()\n   * @param nodeIndex\n   * @param attrName The name of the attribute related to request for this node.\n   */\n  private getNodeAtIndex(nodeIndex: number, attrName: string): Node {\n    if (this.inboundNodes.length === 0) {\n      throw new RuntimeError(\n          'The layer has never been called ' +\n          `and thus has no defined ${attrName}.`);\n    }\n    if (this.inboundNodes.length <= nodeIndex) {\n      throw new ValueError(\n          `Asked to get ${attrName} at node ${nodeIndex}, ` +\n          `but the layer has only ${this.inboundNodes.length} inbound nodes.`);\n    }\n    return this.inboundNodes[nodeIndex];\n  }\n\n  /**\n   * Retrieves the input tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple inputs).\n   */\n  getInputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer at a given node.\n   *\n   * @param nodeIndex Integer, index of the node from which to retrieve the\n   *   attribute. E.g. `nodeIndex=0` will correspond to the first time the layer\n   *   was called.\n   *\n   * @return A tensor (or list of tensors if the layer has multiple outputs).\n   */\n  getOutputAt(nodeIndex: number): SymbolicTensor|SymbolicTensor[] {\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(nodeIndex, 'output').outputTensors);\n  }\n\n  // Properties\n\n  /**\n   * Retrieves the input tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Input tensor or list of input tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get input(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer input\" ' +\n          'is ill-defined. ' +\n          'Use `getInputAt(nodeIndex)` instead.');\n    } else if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' is not connected, no input to return.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'input').inputTensors);\n  }\n\n  /**\n   * Retrieves the output tensor(s) of a layer.\n   *\n   * Only applicable if the layer has exactly one inbound node,\n   * i.e. if it is connected to one incoming layer.\n   *\n   * @return Output tensor or list of output tensors.\n   *\n   * @exception AttributeError if the layer is connected to more than one\n   *   incoming layers.\n   */\n  get output(): SymbolicTensor|SymbolicTensor[] {\n    if (this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has no inbound nodes.');\n    }\n    if (this.inboundNodes.length > 1) {\n      throw new AttributeError(\n          `Layer ${this.name}` +\n          ' has multiple inbound nodes, ' +\n          'hence the notion of \"layer output\" ' +\n          'is ill-defined. ' +\n          'Use `getOutputAt(nodeIndex)` instead.');\n    }\n    return generic_utils.singletonOrArray(\n        this.getNodeAtIndex(0, 'output').outputTensors);\n  }\n\n  get losses(): RegularizerFn[] {\n    return this._losses;\n  }\n\n  /**\n   * Retrieves the Layer's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Layer.loss in PyKeras.\n    //   In PyKeras, Layer.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return this.losses.map(lossFn => lossFn());\n  }\n\n  get updates(): Tensor[] {\n    return this._updates;\n  }\n\n  get built(): boolean {\n    return this._built;\n  }\n\n  set built(built: boolean) {\n    this._built = built;\n  }\n\n  get trainable(): boolean {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this._trainableWeights.forEach(w => w.trainable = trainable);\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (this.trainable_) {\n      return this._trainableWeights.filter(w => w.trainable);\n    } else {\n      return [];\n    }\n  }\n\n  set trainableWeights(weights: LayerVariable[]) {\n    this._trainableWeights = weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    if (this.trainable) {\n      return this._trainableWeights.filter(w => !w.trainable)\n          .concat(this._nonTrainableWeights);\n    } else {\n      return this._trainableWeights.concat(this._nonTrainableWeights);\n    }\n  }\n\n  set nonTrainableWeights(weights: LayerVariable[]) {\n    this._nonTrainableWeights = weights;\n  }\n\n  /**\n   * The concatenation of the lists trainableWeights and nonTrainableWeights\n   * (in this order).\n   */\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  get stateful(): boolean {\n    return this._stateful;\n  }\n\n  /**\n   * Reset the states of the layer.\n   *\n   * This method of the base Layer class is essentially a no-op.\n   * Subclasses that are stateful (e.g., stateful RNNs) should override this\n   * method.\n   */\n  resetStates(): void {\n    if (!this.stateful) {\n      throw new Error(\n          'Cannot call the resetStates() method of a non-stateful Layer ' +\n          'object.');\n    }\n  }\n\n  /**\n   * Checks compatibility between the layer and provided inputs.\n   *\n   * This checks that the tensor(s) `input`\n   * verify the input assumptions of the layer\n   * (if any). If not, exceptions are raised.\n   *\n   * @param inputs Input tensor or list of input tensors.\n   *\n   * @exception ValueError in case of mismatch between\n   *   the provided inputs and the expectations of the layer.\n   */\n  protected assertInputCompatibility(inputs: Tensor|Tensor[]|SymbolicTensor|\n                                     SymbolicTensor[]): void {\n    inputs = generic_utils.toList(inputs);\n    if (this.inputSpec == null || this.inputSpec.length === 0) {\n      return;\n    }\n    const inputSpec = generic_utils.toList(this.inputSpec);\n    if (inputs.length !== inputSpec.length) {\n      throw new ValueError(\n          `Layer ${this.name} expects ${inputSpec.length} inputs, ` +\n          `but it received ${inputs.length} input tensors. ` +\n          `Input received: ${inputs}`);\n    }\n    for (let inputIndex = 0; inputIndex < inputs.length; inputIndex++) {\n      const x = inputs[inputIndex];\n      const spec: InputSpec = inputSpec[inputIndex];\n      if (spec == null) {\n        continue;\n      }\n\n      // Check ndim.\n      const ndim = x.rank;\n      if (spec.ndim != null) {\n        if (ndim !== spec.ndim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}: ` +\n              `expected ndim=${spec.ndim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.maxNDim != null) {\n        if (ndim > spec.maxNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected max_ndim=${spec.maxNDim}, found ndim=${ndim}`);\n        }\n      }\n      if (spec.minNDim != null) {\n        if (ndim < spec.minNDim) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name}` +\n              `: expected min_ndim=${spec.minNDim}, found ndim=${ndim}.`);\n        }\n      }\n\n      // Check dtype.\n      if (spec.dtype != null) {\n        if (x.dtype !== spec.dtype) {\n          throw new ValueError(\n              `Input ${inputIndex} is incompatible with layer ${this.name} ` +\n              `: expected dtype=${spec.dtype}, found dtype=${x.dtype}.`);\n        }\n      }\n\n      // Check specific shape axes.\n      if (spec.axes) {\n        const xShape = x.shape;\n        for (const key in spec.axes) {\n          const axis = Number(key);\n          const value = spec.axes[key];\n          // Perform Python-style slicing in case axis < 0;\n          // TODO(cais): Use https://github.com/alvivi/typescript-underscore to\n          // ensure type safety through Underscore calls.\n          const xShapeAtAxis =\n              axis >= 0 ? xShape[axis] : xShape[xShape.length + axis];\n          if (value != null && [value, null].indexOf(xShapeAtAxis) === -1) {\n            throw new ValueError(\n                `Input ${inputIndex} is incompatible with layer ` +\n                `${this.name}: expected axis ${axis} of input shape to ` +\n                `have value ${value} but got shape ${xShape}.`);\n          }\n        }\n      }\n\n      // Check shape.\n      if (spec.shape != null) {\n        for (let i = 0; i < spec.shape.length; ++i) {\n          const specDim = spec.shape[i];\n          const dim = x.shape[i];\n          if (specDim != null && dim != null) {\n            if (specDim !== dim) {\n              throw new ValueError(\n                  `Input ${inputIndex} is incompatible with layer ` +\n                  `${this.name}: expected shape=${spec.shape}, ` +\n                  `found shape=${x.shape}.`);\n            }\n          }\n        }\n      }\n    }\n  }\n\n  /**\n   * This is where the layer's logic lives.\n   *\n   * @param inputs Input tensor, or list/tuple of input tensors.\n   * @param kwargs Additional keyword arguments.\n   *\n   * @return A tensor or list/tuple of tensors.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return inputs;\n  }\n\n  protected invokeCallHook(inputs: Tensor|Tensor[], kwargs: Kwargs) {\n    if (this._callHook != null) {\n      this._callHook(inputs, kwargs);\n    }\n  }\n\n  /**\n   * Set call hook.\n   * This is currently used for testing only.\n   * @param callHook\n   */\n  setCallHook(callHook: CallHook) {\n    this._callHook = callHook;\n  }\n\n  /**\n   * Clear call hook.\n   * This is currently used for testing only.\n   */\n  clearCallHook() {\n    this._callHook = null;\n  }\n\n  /**\n   * Builds or executes a `Layer's logic.\n   *\n   * When called with `tf.Tensor`(s), execute the `Layer`s computation and\n   * return Tensor(s). For example:\n   *\n   * ```js\n   * const denseLayer = tf.layers.dense({\n   *   units: 1,\n   *   kernelInitializer: 'zeros',\n   *   useBias: false\n   * });\n   *\n   * // Invoke the layer's apply() method with a `tf.Tensor` (with concrete\n   * // numeric values).\n   * const input = tf.ones([2, 2]);\n   * const output = denseLayer.apply(input);\n   *\n   * // The output's value is expected to be [[0], [0]], due to the fact that\n   * // the dense layer has a kernel initialized to all-zeros and does not have\n   * // a bias.\n   * output.print();\n   * ```\n   *\n   * When called with `tf.SymbolicTensor`(s), this will prepare the layer for\n   * future execution.  This entails internal book-keeping on shapes of\n   * expected Tensors, wiring layers together, and initializing weights.\n   *\n   * Calling `apply` with `tf.SymbolicTensor`s are typically used during the\n   * building of non-`tf.Sequential` models. For example:\n   *\n   * ```js\n   * const flattenLayer = tf.layers.flatten();\n   * const denseLayer = tf.layers.dense({units: 1});\n   *\n   * // Use tf.layers.input() to obtain a SymbolicTensor as input to apply().\n   * const input = tf.input({shape: [2, 2]});\n   * const output1 = flattenLayer.apply(input);\n   *\n   * // output1.shape is [null, 4]. The first dimension is the undetermined\n   * // batch size. The second dimension comes from flattening the [2, 2]\n   * // shape.\n   * console.log(JSON.stringify(output1.shape));\n   *\n   * // The output SymbolicTensor of the flatten layer can be used to call\n   * // the apply() of the dense layer:\n   * const output2 = denseLayer.apply(output1);\n   *\n   * // output2.shape is [null, 1]. The first dimension is the undetermined\n   * // batch size. The second dimension matches the number of units of the\n   * // dense layer.\n   * console.log(JSON.stringify(output2.shape));\n   *\n   * // The input and output and be used to construct a model that consists\n   * // of the flatten and dense layers.\n   * const model = tf.model({inputs: input, outputs: output2});\n   * ```\n   *\n   * @param inputs a `tf.Tensor` or `tf.SymbolicTensor` or an Array of them.\n   * @param kwargs Additional keyword arguments to be passed to `call()`.\n   *\n   * @return Output of the layer's `call` method.\n   *\n   * @exception ValueError error in case the layer is missing shape information\n   *   for its `build` call.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  // Porting Note: This is a replacement for __call__() in Python.\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    kwargs = kwargs || {};\n\n    this.assertNotDisposed();\n\n    // Ensure inputs are all the same type.\n    const inputsList = generic_utils.toList(inputs);\n\n    let allAreSymbolic = true;\n    for (const input of inputsList) {\n      if (!(input instanceof SymbolicTensor)) {\n        allAreSymbolic = false;\n        break;\n      }\n    }\n    let noneAreSymbolic = true;\n    for (const input of inputsList) {\n      if (input instanceof SymbolicTensor) {\n        noneAreSymbolic = false;\n        break;\n      }\n    }\n\n    if (allAreSymbolic === noneAreSymbolic) {\n      throw new ValueError(\n          'Arguments to apply() must be all ' +\n          'SymbolicTensors or all Tensors');\n    }\n\n    // TODO(michaelterry): nameScope() may not be necessary.\n    return nameScope(this.name, () => {\n      // Handle laying building (weight creating, input spec locking).\n      if (!this.built) {\n        /*\n          Throw exceptions in case the input is not compatible\n          with the inputSpec specified in the layer constructor.\n         */\n        this.assertInputCompatibility(inputs);\n\n        // Collect input shapes to build layer.\n        const inputShapes: Shape[] = [];\n        for (const xElem of generic_utils.toList(inputs)) {\n          inputShapes.push(xElem.shape);\n        }\n        this.build(generic_utils.singletonOrArray(inputShapes));\n        this.built = true;\n\n        // Load weights that were specified at layer instantiation.\n        if (this.initialWeights) {\n          this.setWeights(this.initialWeights);\n        }\n\n        if (this._refCount === null && noneAreSymbolic) {\n          // The first use of this layer is a non-symbolic call, set ref count\n          // to 1 so the Layer can be properly disposed if its dispose() method\n          // is called.\n          this._refCount = 1;\n        }\n      }\n\n      /*\n        Throw exceptions in case the input is not compatible\n        with the inputSpec set at build time.\n      */\n      this.assertInputCompatibility(inputs);\n\n      // Handle mask propagation.\n      // TODO(michaelterry): Mask propagation not currently implemented.\n\n      // Actually call the layer, collecting output(s), mask(s), and shape(s).\n      if (noneAreSymbolic) {\n        let output = this.call(inputs as Tensor | Tensor[], kwargs);\n        // TODO(michaelterry): Compute the outputMask\n\n        // If the layer returns tensors from its inputs, unmodified,\n        // we copy them to avoid loss of tensor metadata.\n        const outputList: Tensor[] = generic_utils.toList(output);\n        const outputListCopy: Tensor[] = [];\n        // TODO(michaelterry): This copying may not be necessary given our eager\n        // backend.\n        for (let x of outputList) {\n          if (inputsList.indexOf(x) !== -1) {\n            x = x.clone();\n          }\n          outputListCopy.push(x);\n        }\n        output = generic_utils.singletonOrArray(outputListCopy);\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        // TODO(michaelterry): Call addInboundNode()?\n        return output;\n      } else {\n        const inputShape = collectInputShape(inputs);\n        const outputShape = this.computeOutputShape(inputShape);\n        let output: SymbolicTensor|SymbolicTensor[];\n        const outputDType = guessOutputDType(inputs);\n        this.warnOnIncompatibleInputShape(\n            Array.isArray(inputs) ? inputShape[0] as Shape :\n                                    inputShape as Shape);\n\n        if (outputShape != null && outputShape.length > 0 &&\n            Array.isArray(outputShape[0])) {\n          // We have multiple output shapes. Create multiple output tensors.\n          output = (outputShape as Shape[])\n                       .map(\n                           (shape, index) => new SymbolicTensor(\n                               outputDType, shape, this,\n                               generic_utils.toList(inputs), kwargs, this.name,\n                               index));\n        } else {\n          output = new SymbolicTensor(\n              outputDType, outputShape as Shape, this,\n              generic_utils.toList(inputs), kwargs, this.name);\n        }\n\n        /*\n          Add an inbound node to the layer, so that it keeps track\n          of the call and of all new variables created during the call.\n          This also updates the layer history of the output tensor(s).\n          If the input tensor(s) had no previous history,\n          this does nothing.\n        */\n        this.addInboundNode(\n            inputs as SymbolicTensor | SymbolicTensor[], output, null, null,\n            inputShape, outputShape, kwargs);\n        this._refCount++;\n\n        if (this.activityRegularizer != null) {\n          throw new NotImplementedError(\n              'Layer invocation in the presence of activity ' +\n              'regularizer(s) is not supported yet.');\n        }\n\n        return output;\n      }\n    });\n  }\n\n  /**\n   * Check compatibility between input shape and this layer's batchInputShape.\n   *\n   * Print warning if any incompatibility is found.\n   *\n   * @param inputShape Input shape to be checked.\n   */\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {\n    if (this.batchInputShape == null) {\n      return;\n    } else if (inputShape.length !== this.batchInputShape.length) {\n      console.warn(\n          `The rank of the input tensor provided (shape: ` +\n          `${JSON.stringify(inputShape)}) does not match that of the ` +\n          `batchInputShape (${JSON.stringify(this.batchInputShape)}) ` +\n          `of the layer ${this.name}`);\n    } else {\n      let dimMismatch = false;\n      this.batchInputShape.forEach((dimension, i) => {\n        if (dimension != null && inputShape[i] != null &&\n            inputShape[i] !== dimension) {\n          dimMismatch = true;\n        }\n      });\n      if (dimMismatch) {\n        console.warn(\n            `The shape of the input tensor ` +\n            `(${JSON.stringify(inputShape)}) does not ` +\n            `match the expectation of layer ${this.name}: ` +\n            `${JSON.stringify(this.batchInputShape)}`);\n      }\n    }\n  }\n\n  /**\n   * Retrieves the output shape(s) of a layer.\n   *\n   * Only applicable if the layer has only one inbound node, or if all inbound\n   * nodes have the same output shape.\n   *\n   * @returns Output shape or shapes.\n   * @throws AttributeError: if the layer is connected to more than one incoming\n   *   nodes.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  get outputShape(): Shape|Shape[] {\n    if (this.inboundNodes == null || this.inboundNodes.length === 0) {\n      throw new AttributeError(\n          `The layer ${this.name} has never been called and thus has no ` +\n          `defined output shape.`);\n    }\n    const allOutputShapes: string[] = [];\n    for (const node of this.inboundNodes) {\n      const shapeString = JSON.stringify(node.outputShapes);\n      if (allOutputShapes.indexOf(shapeString) === -1) {\n        allOutputShapes.push(shapeString);\n      }\n    }\n    if (allOutputShapes.length === 1) {\n      const outputShapes = this.inboundNodes[0].outputShapes;\n      if (Array.isArray(outputShapes) && Array.isArray(outputShapes[0]) &&\n          outputShapes.length === 1) {\n        return (outputShapes as Shape[])[0];\n      } else {\n        return outputShapes;\n      }\n\n    } else {\n      throw new AttributeError(\n          `The layer ${this.name} has multiple inbound nodes with different ` +\n          `output shapes. Hence the notion of \"output shape\" is ill-defined ` +\n          `for the layer.`);\n      // TODO(cais): Implement getOutputShapeAt().\n    }\n  }\n\n  /**\n   * Counts the total number of numbers (e.g., float32, int32) in the\n   * weights.\n   *\n   * @returns An integer count.\n   * @throws RuntimeError: If the layer is not built yet (in which case its\n   *   weights are not defined yet.)\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  countParams(): number {\n    if (!this.built) {\n      throw new RuntimeError(\n          `You tried to call countParams() on ${this.name}, ` +\n          `but the layer is not built yet. Build it first by calling ` +\n          `build(batchInputShape).`);\n    }\n    return variable_utils.countParamsInWeights(this.weights);\n  }\n\n  /**\n   * Creates the layer weights.\n   *\n   * Must be implemented on all layers that have weights.\n   *\n   * Called when apply() is called to construct the weights.\n   *\n   * @param inputShape A `Shape` or array of `Shape` (unused).\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  build(inputShape: Shape|Shape[]) {\n    this.built = true;\n  }\n\n  /**\n   * Returns the current values of the weights of the layer.\n   *\n   * @param trainableOnly Whether to get the values of only trainable weights.\n   * @returns Weight values as an `Array` of `tf.Tensor`s.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getWeights(trainableOnly = false): Tensor[] {\n    return batchGetValue(trainableOnly ? this.trainableWeights : this.weights);\n  }\n\n  /**\n   * Sets the weights of the layer, from Tensors.\n   *\n   * @param weights a list of Tensors. The number of arrays and their shape\n   *   must match number of the dimensions of the weights of the layer (i.e.\n   *   it should match the output of `getWeights`).\n   *\n   * @exception ValueError If the provided weights list does not match the\n   *   layer's specifications.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  setWeights(weights: Tensor[]): void {\n    tidy(() => {\n      const params = this.weights;\n      if (params.length !== weights.length) {\n        // TODO(cais): Restore the following and use `providedWeights`, instead\n        // of `weights` in the error message, once the deeplearn.js bug is\n        // fixed: https://github.com/PAIR-code/deeplearnjs/issues/498 const\n        // providedWeights = JSON.stringify(weights).substr(0, 50);\n        throw new ValueError(\n            `You called setWeights(weights) on layer \"${this.name}\" ` +\n            `with a weight list of length ${weights.length}, ` +\n            `but the layer was expecting ${params.length} weights. ` +\n            `Provided weights: ${weights}...`);\n      }\n      if (params.length === 0) {\n        return;\n      }\n      const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n      const paramValues = batchGetValue(params);\n      for (let i = 0; i < paramValues.length; ++i) {\n        const pv = paramValues[i];\n        const p = params[i];\n        const w = weights[i];\n        if (!util.arraysEqual(pv.shape, w.shape)) {\n          throw new ValueError(\n              `Layer weight shape ${pv.shape} ` +\n              `not compatible with provided weight shape ${w.shape}`);\n        }\n        weightValueTuples.push([p, w]);\n      }\n      batchSetValue(weightValueTuples);\n    });\n  }\n\n  /**\n   * Adds a weight variable to the layer.\n   *\n   * @param name Name of the new weight variable.\n   * @param shape The shape of the weight.\n   * @param dtype The dtype of the weight.\n   * @param initializer An initializer instance.\n   * @param regularizer A regularizer instance.\n   * @param trainable Whether the weight should be trained via backprop or not\n   *   (assuming that the layer itself is also trainable).\n   * @param constraint An optional trainable.\n   * @return The created weight variable.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  protected addWeight(\n      name: string, shape: Shape, dtype?: DataType, initializer?: Initializer,\n      regularizer?: Regularizer, trainable?: boolean,\n      constraint?: Constraint): LayerVariable {\n    // Reject duplicate weight names.\n    if (this._addedWeightNames.indexOf(name) !== -1) {\n      throw new ValueError(\n          `Duplicate weight name ${name} for layer ${this.name}`);\n    }\n    this._addedWeightNames.push(name);\n\n    if (dtype == null) {\n      dtype = 'float32';\n    }\n\n    if (this.fastWeightInitDuringBuild) {\n      initializer = getInitializer('zeros');\n    }\n    const initValue = initializer.apply(shape, dtype);\n    const weight =\n        new LayerVariable(initValue, dtype, name, trainable, constraint);\n    initValue.dispose();\n    // Request backend not to dispose the weights of the model on scope() exit.\n    if (regularizer != null) {\n      this.addLoss(() => regularizer.apply(weight.read()));\n    }\n    if (trainable == null) {\n      trainable = true;\n    }\n    if (trainable) {\n      this._trainableWeights.push(weight);\n    } else {\n      this._nonTrainableWeights.push(weight);\n    }\n    return weight;\n  }\n\n  /**\n   * Set the fast-weight-initialization flag.\n   *\n   * In cases where the initialized weight values will be immediately\n   * overwritten by loaded weight values during model loading, setting\n   * the flag to `true` saves unnecessary calls to potentially expensive\n   * initializers and speeds up the loading process.\n   *\n   * @param value Target value of the flag.\n   */\n  setFastWeightInitDuringBuild(value: boolean) {\n    this.fastWeightInitDuringBuild = value;\n  }\n\n  /**\n   * Add losses to the layer.\n   *\n   * The loss may potentionally be conditional on some inputs tensors,\n   * for instance activity losses are conditional on the layer's inputs.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  addLoss(losses: RegularizerFn|RegularizerFn[]): void {\n    if (losses == null || Array.isArray(losses) && losses.length === 0) {\n      return;\n    }\n    // Update this.losses\n    losses = generic_utils.toList(losses);\n    if (this._losses !== undefined && this._losses !== null) {\n      this.losses.push(...losses);\n    }\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (!this.supportsMasking) {\n      if (mask != null) {\n        if (Array.isArray(mask)) {\n          mask.forEach(maskElement => {\n            if (maskElement != null) {\n              throw new TypeError(\n                  `Layer ${this.name} does not support masking, ` +\n                  'but was passed an inputMask.');\n            }\n          });\n        } else {\n          throw new TypeError(\n              `Layer ${this.name} does not support masking, ` +\n              'but was passed an inputMask.');\n        }\n      }\n      // masking not explicitly supported: return null as mask\n      return null;\n    }\n    // if masking is explictly supported, by default\n    // carry over the input mask\n    return mask;\n  }\n\n  /**\n   * Internal method to create an inbound node for the layer.\n   *\n   * @param inputTensors List of input tensors.\n   * @param outputTensors List of output tensors.\n   * @param inputMasks List of input masks (a mask can be a tensor, or null).\n   * @param outputMasks List of output masks (a mask can be a tensor, or null).\n   * @param inputShapes List of input shape tuples.\n   * @param outputShapes List of output shape tuples.\n   * @param kwargs Dictionary of keyword arguments that were passed to the\n   *   `call` method of the layer at the call that created the node.\n   */\n  private addInboundNode(\n      inputTensors: SymbolicTensor|SymbolicTensor[],\n      outputTensors: SymbolicTensor|SymbolicTensor[],\n      inputMasks: Tensor|Tensor[], outputMasks: Tensor|Tensor[],\n      inputShapes: Shape|Shape[], outputShapes: Shape|Shape[],\n      kwargs: {} = null): void {\n    const inputTensorList: SymbolicTensor[] =\n        generic_utils.toList(inputTensors);\n    outputTensors = generic_utils.toList(outputTensors);\n    inputMasks = generic_utils.toList(inputMasks);\n    outputMasks = generic_utils.toList(outputMasks);\n    inputShapes = types_utils.normalizeShapeList(inputShapes);\n    outputShapes = types_utils.normalizeShapeList(outputShapes);\n\n    // Collect input tensor(s) coordinates.\n    const inboundLayers: Layer[] = [];\n    const nodeIndices: number[] = [];\n    const tensorIndices: number[] = [];\n    for (const x of inputTensorList) {\n      /*\n       * TODO(michaelterry): Keras adds this value to tensors; it's not\n       * clear whether we'll use this or not.\n       */\n      inboundLayers.push(x.sourceLayer);\n      nodeIndices.push(x.nodeIndex);\n      tensorIndices.push(x.tensorIndex);\n    }\n\n    // Create node, add it to inbound nodes.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node(\n        {\n          outboundLayer: this,\n          inboundLayers,\n          nodeIndices,\n          tensorIndices,\n          inputTensors: inputTensorList,\n          outputTensors,\n          inputMasks,\n          outputMasks,\n          inputShapes,\n          outputShapes\n        },\n        kwargs);\n\n    // Update tensor history\n    for (let i = 0; i < outputTensors.length; i++) {\n      // TODO(michaelterry: _uses_learning_phase not tracked.\n      outputTensors[i].sourceLayer = this;\n      outputTensors[i].nodeIndex = this.inboundNodes.length - 1;\n      outputTensors[i].tensorIndex = i;\n    }\n  }\n\n  /**\n   * Returns the config of the layer.\n   *\n   * A layer config is a TS dictionary (serializable)\n   * containing the configuration of a layer.\n   * The same layer can be reinstantiated later\n   * (without its trained weights) from this configuration.\n   *\n   * The config of a layer does not include connectivity\n   * information, nor the layer class name.  These are handled\n   * by 'Container' (one layer of abstraction above).\n   *\n   * Porting Note: The TS dictionary follows TS naming standrds for\n   * keys, and uses tfjs-layers type-safe Enums.  Serialization methods\n   * should use a helper function to convert to the pythonic storage\n   * standard. (see serialization_utils.convertTsToPythonic)\n   *\n   * @returns TS dictionary of configuration.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  getConfig(): serialization.ConfigDict {\n    const config:\n        serialization.ConfigDict = {name: this.name, trainable: this.trainable};\n    if (this.batchInputShape != null) {\n      config['batchInputShape'] = this.batchInputShape;\n    }\n    if (this.dtype != null) {\n      config['dtype'] = this.dtype;\n    }\n    return config;\n  }\n\n  /**\n   * Dispose the weight variables that this Layer instance holds.\n   *\n   * @returns {number} Number of disposed variables.\n   */\n  protected disposeWeights(): number {\n    this.weights.forEach(weight => weight.dispose());\n    return this.weights.length;\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Layer '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose layer's weights.\n   *\n   * This method decrease the reference count of the Layer object by 1.\n   *\n   * A Layer is reference-counted. Its reference count is incremented by 1\n   * the first item its `apply()` method is called and when it becomes a part\n   * of a new `Node` (through calling the `apply()`) method on a\n   * `tf.SymbolicTensor`).\n   *\n   * If the reference count of a Layer becomes 0, all the weights will be\n   * disposed and the underlying memory (e.g., the textures allocated in WebGL)\n   * will be freed.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * weights of the Layer will *not* be disposed.\n   *\n   * After a Layer is disposed, it cannot be used in calls such as `apply()`,\n   * `getWeights()` or `setWeights()` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the Container after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the layer has already\n   *   been disposed.\n   *\n   * @doc {heading: 'Models', 'subheading': 'Classes'}\n   */\n  dispose(): DisposeResult {\n    if (!this.built) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been ` +\n          `built yet.`);\n    }\n\n    if (this._refCount === null) {\n      throw new Error(\n          `Cannot dispose Layer ${this.name} because it has not been used ` +\n          `yet.`);\n    }\n\n    this.assertNotDisposed();\n\n    let numDisposedVariables = 0;\n    if (--this._refCount === 0) {\n      numDisposedVariables = this.disposeWeights();\n    }\n\n    return {refCountAfterDispose: this._refCount, numDisposedVariables};\n  }\n}\n\n/**\n * Collects the input shape(s) of a list of `tf.Tensor`s or\n * `tf.SymbolicTensor`s.\n *\n * TODO(michaelterry): Update PyKeras docs (backport).\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return List of shape tuples (or single tuple), one tuple per input.\n */\nfunction collectInputShape(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                           Tensor[]): Shape|Shape[] {\n  inputTensors =\n      generic_utils.toList(inputTensors) as SymbolicTensor[] | Tensor[];\n  const shapes: Shape[] = [];\n  for (const x of inputTensors) {\n    shapes.push(x.shape);\n  }\n  return generic_utils.singletonOrArray(shapes);\n}\n\n/**\n * Guesses output dtype based on inputs.\n *\n * At present, just returns 'float32' for any input.\n *\n * @param inputTensors List of input tensors (or single input tensor).\n *\n * @return The guessed DType. At present, always returns 'float32'.\n */\nfunction guessOutputDType(inputTensors: SymbolicTensor|SymbolicTensor[]|Tensor|\n                          Tensor[]): DataType {\n  return 'float32';\n}\n\n/**\n * Returns the list of input tensors necessary to compute `tensor`.\n *\n * Output will always be a list of tensors (potentially with 1 element).\n *\n * @param tensor The tensor to start from.\n * @param layer Origin layer of the tensor.\n * @param nodeIndex Origin node index of the tensor.\n *\n * @return Array of input tensors.\n */\nexport function getSourceInputs(\n    tensor: SymbolicTensor, layer?: Layer,\n    nodeIndex?: number): SymbolicTensor[] {\n  if (layer == null || (nodeIndex != null && nodeIndex > 0)) {\n    layer = tensor.sourceLayer;\n    nodeIndex = tensor.nodeIndex;\n  }\n  if (layer.inboundNodes.length === 0) {\n    return [tensor];\n  } else {\n    const node = layer.inboundNodes[nodeIndex];\n    if (node.inboundLayers.length === 0) {\n      return node.inputTensors;\n    } else {\n      const sourceTensors: SymbolicTensor[] = [];\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const x = node.inputTensors[i];\n        const layer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const previousSources = getSourceInputs(x, layer, nodeIndex);\n        // Avoid input redundancy.\n        for (const x of previousSources) {\n          if (sourceTensors.indexOf(x) === -1) {\n            sourceTensors.push(x);\n          }\n        }\n      }\n      return sourceTensors;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {DataType, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\n\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/**\n * Constructor arguments for InputLayer.\n *\n * Note: You should provide only inputShape or batchInputShape (not both).\n * If only inputShape is provided, then the batchInputShape is determined by\n * the batchSize argument and the inputShape: [batchSize].concat(inputShape).\n */\nexport declare interface InputLayerArgs {\n  /** Input shape, not including the batch axis. */\n  inputShape?: Shape;\n  /** Optional input batch size (integer or null). */\n  batchSize?: number;\n  /** Batch input shape, including the batch axis. */\n  batchInputShape?: Shape;\n  /** Datatype of the input.  */\n  dtype?: DataType;\n  /**\n   * Whether the placeholder created is meant to be sparse.\n   */\n  sparse?: boolean;  // TODO(michaelterry): Not clear whether we'll need this.\n\n  /** Name of the layer. */\n  name?: string;\n}\n\nexport class InputLayer extends Layer {\n  /** @nocollapse */\n  static readonly className = 'InputLayer';\n  sparse: boolean;\n  constructor(args: InputLayerArgs) {\n    super({\n      dtype: args.dtype,\n      name: args.name != null ? args.name : getUid('input').toString()\n    });\n    // Normalize config.batchSize and config.sparse\n    if (args.batchSize == null) {\n      args.batchSize = null;\n    }\n    if (args.sparse == null) {\n      args.sparse = false;\n    }\n\n    this.trainable = false;\n    this.built = true;\n    this.sparse = args.sparse;\n\n    if (args.inputShape != null && args.batchInputShape != null) {\n      throw new ValueError(\n          'Only provide the inputShape OR ' +\n          'batchInputShape argument to inputLayer, not both at the same time.');\n    }\n    let batchInputShape = args.batchInputShape;\n    if (batchInputShape == null) {\n      if (args.inputShape == null) {\n        throw new ValueError(\n            'An InputLayer should be passed either a ' +\n            '`batchInputShape` or an `inputShape`.');\n      } else {\n        batchInputShape = [args.batchSize].concat(args.inputShape);\n      }\n    } else {\n      // TODO(michaelterry): Backport to PyKeras\n      if (args.batchSize != null) {\n        throw new ValueError(\n            'Cannot specify batchSize if batchInputShape is ' +\n            'specified when creating an InputLayer.');\n      }\n    }\n\n    const dtype = args.dtype || 'float32';\n\n    this.batchInputShape = batchInputShape;\n    this.dtype = dtype;\n    // TODO(michaelterry): Backport this to PyKeras?\n    this.inputSpec = [{shape: batchInputShape}];\n\n    const inputTensor = new SymbolicTensor(\n        this.dtype, this.batchInputShape, this, [], {}, this.name);\n    inputTensor.nodeIndex = 0;\n    inputTensor.tensorIndex = 0;\n\n    // Create an input node to add to this.outboundNode.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: [inputTensor],\n      outputTensors: [inputTensor],\n      inputMasks: [null],\n      outputMasks: [null],\n      inputShapes: [batchInputShape],\n      outputShapes: [batchInputShape]\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor {\n    throw new ValueError(\n        'Cannot pass any input to an ' +\n        `InputLayer's apply() method. InputLayer name: ${this.name}`);\n  }\n\n  dispose(): DisposeResult {\n    // dispose() for InputLayer is overridden as no-op.\n    return {refCountAfterDispose: this._refCount, numDisposedVariables: 0};\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {\n      batchInputShape: this.batchInputShape,\n      dtype: this.dtype,\n      sparse: this.sparse,\n      name: this.name\n    };\n  }\n}\nserialization.registerClass(InputLayer);\n\n/**\n * Config for the Input function.\n *\n * Note: You should provide only shape or batchShape (not both).\n * If only shape is provided, then the batchShape becomes\n * [null].concat(inputShape).\n */\nexport interface InputConfig {\n  /**\n   * A shape, not including the batch size. For instance, `shape=[32]`\n   * indicates that the expected input will be batches of 32-dimensional\n   * vectors.\n   */\n  shape?: Shape;\n  /**\n   * A shape tuple (integer), including the batch size. For instance,\n   * `batchShape=[10, 32]` indicates that the expected input will be batches of\n   * 10 32-dimensional vectors. `batchShape=[null, 32]` indicates batches of an\n   * arbitrary number of 32-dimensional vectors.\n   */\n  batchShape?: Shape;\n  /**\n   * An optional name string for the layer. Should be unique in a model (do not\n   * reuse the same name twice). It will be autogenerated if it isn't provided.\n   */\n  name?: string;\n  dtype?: DataType;\n  /**\n   * A boolean specifying whether the placeholder to be created is sparse.\n   */\n  sparse?: boolean;\n}\n\nexport function Input(config: InputConfig): SymbolicTensor {\n  if (config.batchShape == null && config.shape == null) {\n    throw new Error(\n        'Please provide to Input either a `shape`' +\n        ' or a `batchShape` argument. Note that ' +\n        '`shape` does not include the batch ' +\n        'dimension.');\n  }\n  if (config.batchShape != null && config.shape != null) {\n    // TODO(michaelterry): Backport to PyKeras.\n    throw new ValueError(\n        'Please provide either a `shape` or `batchShape` ' +\n        'argument to Input, but not both.');\n  }\n  let batchShape = config.batchShape;\n  if (config.shape != null && batchShape == null) {\n    batchShape = [null].concat(config.shape);\n  }\n\n  let dtype = config.dtype;\n  if (dtype == null) {\n    dtype = 'float32';\n  }\n\n  const inputLayer = new InputLayer({\n    batchInputShape: batchShape,\n    name: config.name,\n    dtype,\n    sparse: config.sparse\n  });\n\n  const outputs = inputLayer.inboundNodes[0].outputTensors;\n  return outputs[0];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, Scalar} from '@tensorflow/tfjs-core';\n\n/**\n * Logs in which values can be either numbers or Tensors (Scalars).\n *\n * Used internally.\n */\nexport type UnresolvedLogs = {\n  [key: string]: number|Scalar;\n};\n\n/**\n * Turn any Scalar values in a Logs object into actual number values.\n *\n * @param logs The `Logs` object to be resolved in place.\n */\nexport async function resolveScalarsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n  const keys: string[] = [];\n  const scalarsToDispose: Scalar[] = [];\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      const valueScalar = value;\n      promises.push(valueScalar.data());\n      keys.push(key);\n      scalarsToDispose.push(valueScalar);\n    }\n  }\n  if (promises.length > 0) {\n    const values = await Promise.all(promises);\n    for (let i = 0; i < values.length; ++i) {\n      logs[keys[i]] = values[i][0];\n    }\n    // Dispose the original scalar tensors.\n    dispose(scalarsToDispose);\n  }\n}\n\n/**\n * Dispose all Tensors in an UnresolvedLogs object.\n *\n * @param logs An `UnresolvedLogs` object potentially containing `tf.Tensor`s in\n *   places where the values can be `tf.Tensor` or `number`.\n */\nexport function disposeTensorsInLogs(logs: UnresolvedLogs) {\n  if (logs == null) {\n    return;\n  }\n  for (const key in logs) {\n    const value = logs[key];\n    if (typeof value !== 'number') {\n      value.dispose();\n    }\n  }\n}\n\n/**\n * Logs in which values can only be numbers.\n *\n * Used when calling client-provided custom callbacks.\n */\nexport type Logs = {\n  [key: string]: number;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {add, div, keep, mul, nextFrame, Scalar, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Container} from './engine/container';\nimport {ValueError} from './errors';\nimport {Logs, resolveScalarsInLogs, UnresolvedLogs} from './logs';\nimport * as generic_utils from './utils/generic_utils';\n\n/** Verbosity logging level when fitting a model. */\nexport enum ModelLoggingVerbosity {\n  SILENT = 0,\n  VERBOSE = 1\n}\n\n/** How often to yield to the main thread when training (in ms). */\nexport const DEFAULT_YIELD_EVERY_MS = 125;\n\nexport type Params = {\n  [key: string]: number|string|boolean|number[]|string[]|boolean[];\n};\n\nexport type YieldEveryOptions = 'auto'|'batch'|'epoch'|'never'|number;\n\n/**\n * Abstract base class used to build new callbacks.\n *\n * The `logs` dictionary that callback methods take as argument will contain\n * keys for quantities relevant to the current batch or epoch.\n *\n * Currently, the `.fit()` method of the `Sequential` model class\n * will include the following quantities in the `logs` that\n * it passes to its callbacks:\n *\n * onEpochEnd: Logs include `acc` and `loss`, and optionally include `valLoss`\n *   (if validation is enabled in `fit`), and `valAcc` (if validation and\n *   accuracy monitoring are enabled).\n * onBatchBegin: Logs include `size`, the number of samples in the current\n *   batch.\n * onBatchEnd: Logs include `loss`, and optionally `acc` (if accuracy monitoring\n *   is enabled).\n */\nexport abstract class BaseCallback {\n  // TODO(michaelterry): This type is a best guess.\n  validationData: Tensor|Tensor[] = null;\n  /**\n   * Training parameters (eg. verbosity, batch size, number of epochs...).\n   */\n  params: Params;\n\n  setParams(params: Params): void {\n    this.params = params;\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {}\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {}\n\n  async onTrainBegin(logs?: UnresolvedLogs) {}\n\n  async onTrainEnd(logs?: UnresolvedLogs) {}\n\n  // LayersModel needs to call Callback.setModel(), but cannot actually depend\n  // on Callback because that creates a cyclic dependency.  Providing this no-op\n  // method on BaseCallback breaks the cycle: this way LayersModel can depend on\n  // BaseCallback but not on Callback.  The argument is typed as `Container`\n  // (the superclass of LayersModel) to avoid recapitulating the cycle. Callback\n  // overrides this method and enforces that the argument is really a\n  // LayersModel.\n  setModel(model: Container): void {\n    // Do nothing. Use Callback instead of BaseCallback to track the model.\n  }\n}\n\n/**\n * Container abstracting a list of callbacks.\n */\nexport class CallbackList {\n  callbacks: BaseCallback[];\n  queueLength: number;\n\n  // TODO(cais): When the need arises, uncomment the following lines and\n  // implement the queue for time values.\n  // private deltaTBatch: number;\n  // private deltaTsBatchBegin: Array<number>;\n  // private deltaTsBatchEnd: Array<number>;\n\n  /**\n   * Constructor of CallbackList.\n   * @param callbacks Array of `Callback` instances.\n   * @param queueLength Queue length for keeping running statistics over\n   *   callback execution time.\n   */\n  constructor(callbacks?: BaseCallback[], queueLength = 10) {\n    // TODO(cais): Make use of queueLength when implementing the queue for time\n    // values.\n    if (callbacks == null) {\n      callbacks = [];\n    }\n    this.callbacks = callbacks;\n    this.queueLength = queueLength;\n  }\n\n  append(callback: BaseCallback): void {\n    this.callbacks.push(callback);\n  }\n\n  setParams(params: Params): void {\n    for (const callback of this.callbacks) {\n      callback.setParams(params);\n    }\n  }\n\n  setModel(model: Container): void {\n    for (const callback of this.callbacks) {\n      callback.setModel(model);\n    }\n  }\n\n  /**\n   * Called at the start of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochBegin(epoch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of an epoch.\n   * @param epoch Index of epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onEpochEnd(epoch, logs);\n    }\n  }\n\n  /**\n   * Called  right before processing a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchBegin(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the end of a batch.\n   * @param batch Index of batch within the current epoch.\n   * @param logs Dictionary of logs.\n   */\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onBatchEnd(batch, logs);\n    }\n  }\n\n  /**\n   * Called at the beginning of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainBegin(logs);\n    }\n  }\n\n  /**\n   * Called at the end of training.\n   * @param logs Dictionary of logs.\n   */\n  async onTrainEnd(logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    for (const callback of this.callbacks) {\n      await callback.onTrainEnd(logs);\n    }\n  }\n}\n\n/**\n * Callback that accumulates epoch averages of metrics.\n *\n * This callback is automatically applied to every LayersModel.\n */\nexport class BaseLogger extends BaseCallback {\n  private seen: number;\n  private totals: UnresolvedLogs;\n\n  constructor() {\n    super();\n  }\n\n  async onEpochBegin(epoch: number) {\n    this.seen = 0;\n    this.totals = {};\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const batchSize = logs['size'] == null ? 0 : logs['size'] as number;\n    this.seen += batchSize;\n    for (const key in logs) {\n      const value = logs[key];\n      if (typeof value === 'number') {\n        if (!this.totals.hasOwnProperty(key)) {\n          this.totals[key] = 0;\n        }\n        this.totals[key] = this.totals[key] as number + value * batchSize;\n      } else {\n        let oldTotalsToDispose: Scalar;\n        if (key in this.totals) {\n          oldTotalsToDispose = this.totals[key] as Scalar;\n        } else {\n          this.totals[key] = 0;\n        }\n        const total: Scalar =\n            tidy(() => add((this.totals[key]), mul(value, batchSize)));\n        this.totals[key] = total;\n        if (oldTotalsToDispose != null) {\n          oldTotalsToDispose.dispose();\n        }\n      }\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs != null) {\n      for (const key of this.params['metrics'] as string[]) {\n        if (this.totals[key] == null) {\n          continue;\n        }\n        if (typeof this.totals[key] === 'number') {\n          logs[key] = this.totals[key] as number / this.seen;\n        } else {\n          tidy(() => {\n            const log: Scalar = mul(div(1, this.seen), this.totals[key]);\n            logs[key] = log;\n            (this.totals[key] as Tensor).dispose();\n            keep(logs[key] as Scalar);\n          });\n        }\n      }\n    }\n  }\n}\n\n/**\n * Callback that records events into a `History` object. This callback is\n * automatically applied to every TF.js Layers model. The `History` object\n * gets returned by the `fit` method of models.\n */\nexport class History extends BaseCallback {\n  epoch: number[];\n  history: {[key: string]: Array<number|Tensor>};\n\n  async onTrainBegin(logs?: UnresolvedLogs) {\n    this.epoch = [];\n    this.history = {};\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs) {\n    if (logs == null) {\n      logs = {};\n    }\n    this.epoch.push(epoch);\n    for (const key in logs) {\n      if (this.history[key] == null) {\n        this.history[key] = [];\n      }\n      this.history[key].push(logs[key]);\n    }\n  }\n\n  /**\n   * Await the values of all losses and metrics.\n   */\n  async syncData() {\n    const promises: Array<Promise<Float32Array|Int32Array|Uint8Array>> = [];\n    const keys: string[] = [];\n    const indices: number[] = [];\n    for (const key in this.history) {\n      const valueArray = this.history[key];\n      for (let i = 0; i < valueArray.length; ++i) {\n        if (typeof valueArray[i] !== 'number') {\n          const valueScalar = valueArray[i] as Tensor;\n          promises.push(valueScalar.data());\n          keys.push(key);\n          indices.push(i);\n        }\n      }\n    }\n    const values = await Promise.all(promises);\n    for (let n = 0; n < values.length; ++n) {\n      const tensorToDispose = this.history[keys[n]][indices[n]] as Tensor;\n      tensorToDispose.dispose();\n      this.history[keys[n]][indices[n]] = values[n][0];\n    }\n  }\n}\n\nexport interface CustomCallbackArgs {\n  onTrainBegin?: (logs?: Logs) => void | Promise<void>;\n  onTrainEnd?: (logs?: Logs) => void | Promise<void>;\n  onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;\n  onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;\n  onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n}\n\n/**\n * Custom callback for training.\n */\nexport class CustomCallback extends BaseCallback {\n  protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;\n  protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;\n  protected readonly epochBegin:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly epochEnd:\n      (epoch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchBegin:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly batchEnd:\n      (batch: number, logs?: Logs) => void | Promise<void>;\n  protected readonly yield:\n      (epoch: number, batch: number, logs: Logs) => void | Promise<void>;\n\n  private yieldEvery: YieldEveryOptions;\n  private currentEpoch = 0;\n\n  constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions) {\n    super();\n    this.yieldEvery = yieldEvery || 'auto';\n    if (this.yieldEvery === 'auto') {\n      this.yieldEvery = DEFAULT_YIELD_EVERY_MS;\n    }\n    if (this.yieldEvery === 'never' && args.onYield != null) {\n      throw new Error(\n          'yieldEvery is `never` but you provided an `onYield` callback. ' +\n          'Either change `yieldEvery` or remove the callback');\n    }\n    if (util.isNumber(this.yieldEvery)) {\n      // Decorate `maybeWait` so it will be called at most once every\n      // `yieldEvery` ms.\n      this.maybeWait = generic_utils.debounce(\n          this.maybeWait.bind(this), this.yieldEvery as number);\n    }\n    this.trainBegin = args.onTrainBegin;\n    this.trainEnd = args.onTrainEnd;\n    this.epochBegin = args.onEpochBegin;\n    this.epochEnd = args.onEpochEnd;\n    this.batchBegin = args.onBatchBegin;\n    this.batchEnd = args.onBatchEnd;\n    this.yield = args.onYield;\n  }\n\n  async maybeWait(epoch: number, batch: number, logs: UnresolvedLogs) {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.yield != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.yield(epoch, batch, logs as Logs));\n    }\n    ps.push(nextFrame());\n    await Promise.all(ps);\n  }\n\n  async onEpochBegin(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    this.currentEpoch = epoch;\n    if (this.epochBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.epochBegin(epoch, logs as Logs);\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.epochEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.epochEnd(epoch, logs as Logs));\n    }\n    if (this.yieldEvery === 'epoch') {\n      ps.push(nextFrame());\n    }\n    await Promise.all(ps);\n  }\n\n  async onBatchBegin(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    if (this.batchBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.batchBegin(batch, logs as Logs);\n    }\n  }\n\n  async onBatchEnd(batch: number, logs?: UnresolvedLogs): Promise<void> {\n    const ps: Array<void|Promise<void>> = [];\n    if (this.batchEnd != null) {\n      await resolveScalarsInLogs(logs);\n      ps.push(this.batchEnd(batch, logs as Logs));\n    }\n    if (this.yieldEvery === 'batch') {\n      ps.push(nextFrame());\n    } else if (util.isNumber(this.yieldEvery)) {\n      ps.push(this.maybeWait(this.currentEpoch, batch, logs));\n    }\n    await Promise.all(ps);\n  }\n\n  async onTrainBegin(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainBegin != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainBegin(logs as Logs);\n    }\n  }\n\n  async onTrainEnd(logs?: UnresolvedLogs): Promise<void> {\n    if (this.trainEnd != null) {\n      await resolveScalarsInLogs(logs);\n      await this.trainEnd(logs as Logs);\n    }\n  }\n}\n\n/**\n * Standardize callbacks or configurations of them to an Array of callbacks.\n */\nexport function standardizeCallbacks(\n    callbacks: BaseCallback|BaseCallback[]|CustomCallbackArgs|\n    CustomCallbackArgs[],\n    yieldEvery: YieldEveryOptions): BaseCallback[] {\n  if (callbacks == null) {\n    callbacks = {} as BaseCallback;\n  }\n  if (callbacks instanceof BaseCallback) {\n    return [callbacks];\n  }\n  if (Array.isArray(callbacks) && callbacks[0] instanceof BaseCallback) {\n    return callbacks as BaseCallback[];\n  }\n  // Convert custom callback configs to custom callback objects.\n  const callbackConfigs =\n      generic_utils.toList(callbacks) as CustomCallbackArgs[];\n  return callbackConfigs.map(\n      callbackConfig => new CustomCallback(callbackConfig, yieldEvery));\n}\n\nexport declare type BaseCallbackConstructor = {\n  new (): BaseCallback\n};\n\n/**\n * A global registry for callback constructors to be used during\n * LayersModel.fit().\n */\nexport class CallbackConstructorRegistry {\n  private static constructors:\n      {[verbosityLevel: number]: BaseCallbackConstructor[]} = {};\n\n  /**\n   * Blocks public access to constructor.\n   */\n  private constructor() {}\n\n  /**\n   * Register a tf.LayersModel.fit() callback constructor.\n   *\n   * The registered callback constructor will be used to instantiate\n   * callbacks for every tf.LayersModel.fit() call afterwards.\n   *\n   * @param verbosityLevel Level of verbosity at which the `callbackConstructor`\n   *   is to be reigstered.\n   * @param callbackConstructor A no-arg constructor for `tf.Callback`.\n   * @throws Error, if the same callbackConstructor has been registered before,\n   *   either at the same or a different `verbosityLevel`.\n   */\n  static registerCallbackConstructor(\n      verbosityLevel: number, callbackConstructor: BaseCallbackConstructor) {\n    util.assert(\n        verbosityLevel >= 0 && Number.isInteger(verbosityLevel),\n        () => `Verbosity level is expected to be an integer >= 0, ` +\n            `but got ${verbosityLevel}`);\n    CallbackConstructorRegistry.checkForDuplicate(callbackConstructor);\n    if (CallbackConstructorRegistry.constructors[verbosityLevel] == null) {\n      CallbackConstructorRegistry.constructors[verbosityLevel] = [];\n    }\n    CallbackConstructorRegistry.constructors[verbosityLevel].push(\n        callbackConstructor);\n  }\n\n  private static checkForDuplicate(callbackConstructor:\n                                       BaseCallbackConstructor) {\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const constructors = CallbackConstructorRegistry.constructors[+levelName];\n      constructors.forEach(ctor => {\n        if (ctor === callbackConstructor) {\n          throw new ValueError('Duplicate callback constructor.');\n        }\n      });\n    }\n  }\n\n  /**\n   * Clear all registered callback constructors.\n   */\n  protected static clear() {\n    CallbackConstructorRegistry.constructors = {};\n  }\n\n  /**\n   * Create callbacks using the registered callback constructors.\n   *\n   * Given `verbosityLevel`, all constructors registered at that level or above\n   * will be called and the instantiated callbacks will be used.\n   *\n   * @param verbosityLevel: Level of verbosity.\n   */\n  static createCallbacks(verbosityLevel: number): BaseCallback[] {\n    const constructors: BaseCallbackConstructor[] = [];\n    for (const levelName in CallbackConstructorRegistry.constructors) {\n      const level = +levelName;\n      if (verbosityLevel >= level) {\n        constructors.push(...CallbackConstructorRegistry.constructors[level]);\n      }\n    }\n    return constructors.map(ctor => new ctor());\n  }\n}\n\nexport function configureCallbacks(\n    callbacks: BaseCallback[], verbose: ModelLoggingVerbosity, epochs: number,\n    initialEpoch: number, numTrainSamples: number, stepsPerEpoch: number,\n    batchSize: number, doValidation: boolean,\n    callbackMetrics: string[]): {callbackList: CallbackList, history: History} {\n  const history = new History();\n  const actualCallbacks: BaseCallback[] = [\n    new BaseLogger(), ...CallbackConstructorRegistry.createCallbacks(verbose)\n  ];\n  if (callbacks != null) {\n    actualCallbacks.push(...callbacks);\n  }\n  actualCallbacks.push(history);\n  const callbackList = new CallbackList(actualCallbacks);\n\n  // TODO(cais): Figure out when this LayersModel instance can have a\n  // dynamically\n  //   set property called 'callback_model' as in PyKeras.\n\n  callbackList.setParams({\n    epochs,\n    initialEpoch,\n    samples: numTrainSamples,\n    steps: stepsPerEpoch,\n    batchSize,\n    verbose,\n    doValidation,\n    metrics: callbackMetrics,\n  });\n  return {callbackList, history};\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source layers/__init__.py */\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {deserializeKerasObject} from '../utils/generic_utils';\n\n/**\n * Instantiate a layer from a config dictionary.\n * @param config dict of the form {class_name: str, config: dict}\n * @param customObjects dict mapping class names (or function names)\n *   of custom (non-Keras) objects to class/functions\n * @param fastWeightInit Optional flag to use fast weight initialization\n *   during deserialization. This is applicable to cases in which\n *   the initialization will be immediately overwritten by loaded weight\n *   values. Default: `false`.\n * @returns Layer instance (may be LayersModel, Sequential, Layer...)\n */\nexport function deserialize(\n    config: serialization.ConfigDict,\n    customObjects = {} as serialization.ConfigDict,\n    fastWeightInit = false): serialization.Serializable {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'layer', fastWeightInit);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: losses.py */\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\nimport * as K from './backend/tfjs_backend';\nimport {ValueError} from './errors';\nimport {LossOrMetricFn} from './types';\n\n/**\n * Normalizes a tensor wrt the L2 norm alongside the specified axis.\n * @param x\n * @param axis Axis along which to perform normalization.\n */\nexport function l2Normalize(x: Tensor, axis?: number): Tensor {\n  return tidy(() => {\n    if (x.dtype !== 'float32') {\n      x = x.asType('float32');\n    }\n    const squareSum = tfc.sum(K.square(x), axis, true);\n    const epsilonTensor = tfc.fill(squareSum.shape, epsilon());\n    const norm = tfc.sqrt(tfc.maximum(squareSum, epsilonTensor));\n    return tfc.div(x, norm);\n  });\n}\n\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(K.square(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => tfc.mean(tfc.abs(tfc.sub(yPred, yTrue)), -1));\n}\n\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const diff = tfc.sub(yTrue, yPred);\n    const clippedTrue =\n        tfc.clipByValue(tfc.abs(yTrue), epsilon(), Number.MAX_VALUE);\n    const absResult = tfc.abs(tfc.div(diff, clippedTrue));\n    return tfc.mul(100, tfc.mean(absResult, -1));\n  });\n}\n\nexport function meanSquaredLogarithmicError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), Number.MAX_VALUE);\n    const firstLog = tfc.log(tfc.add(1, clippedPred));\n\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), Number.MAX_VALUE);\n    const secondLog = tfc.log(tfc.add(1, clippedTrue));\n\n    return tfc.mean(K.square(tfc.sub(firstLog, secondLog)), -1);\n  });\n}\n\nexport function squaredHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(K.square(maxResult), -1);\n  });\n}\n\nexport function hinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const maxResult = tfc.maximum(0, tfc.sub(1, tfc.mul(yTrue, yPred)));\n    return tfc.mean(maxResult, -1);\n  });\n}\n\nexport function categoricalHinge(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const pos = tfc.sum(tfc.mul(yTrue, yPred), -1);\n    const neg = tfc.max(tfc.mul(tfc.sub(1, yTrue), yPred), -1);\n    return tfc.maximum(0, tfc.add(1, tfc.sub(neg, pos)));\n  });\n}\n\n/**\n * Logarithm of the hyperbolic cosine of the prediction error.\n *\n * `log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and\n * to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly\n * like the mean squared error, but will not be so strongly affected by the\n * occasional wildly incorrect prediction.\n */\nexport function logcosh(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const log2 = Math.log(2);\n    const predictionDiff = tfc.sub(yPred, yTrue);\n    const logcoshResult = tfc.sub(\n        tfc.add(predictionDiff, tfc.softplus(tfc.mul(-2, predictionDiff))),\n        log2);\n    return tfc.mean(logcoshResult, -1);\n  });\n}\n\nexport function categoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    if (fromLogits) {\n      output = tfc.softmax(output);\n    } else {\n      // scale preds so that the class probabilities of each sample sum to 1.\n      const outputSum = tfc.sum(output, output.shape.length - 1, true);\n      output = tfc.div(output, outputSum);\n    }\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    return tfc.neg(tfc.sum(\n        tfc.mul(target.toFloat(), tfc.log(output)), output.shape.length - 1));\n  });\n}\n\n/**\n * Categorical crossentropy with integer targets.\n *\n * @param target An integer tensor.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n */\nexport function sparseCategoricalCrossentropy(\n    target: Tensor, output: Tensor, fromLogits = false): Tensor {\n  return tidy(() => {\n    const flatTarget = tfc.floor(K.flatten(target)).toInt() as Tensor1D;\n    output = tfc.clipByValue(output, epsilon(), 1 - epsilon());\n    const outputShape = output.shape;\n    const oneHotTarget =\n        tfc.oneHot(flatTarget, outputShape[outputShape.length - 1])\n            .reshape(outputShape);\n    return categoricalCrossentropy(oneHotTarget, output, fromLogits);\n  });\n}\n\n/**\n * From TensorFlow's implementation in nn_impl.py:\n *\n * For brevity, let `x = logits`, `z = labels`.  The logistic loss is\n *      z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))\n *    = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))\n *    = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))\n *    = (1 - z) * x + log(1 + exp(-x))\n *    = x - x * z + log(1 + exp(-x))\n * For x < 0, to avoid overflow in exp(-x), we reformulate the above\n *      x - x * z + log(1 + exp(-x))\n *    = log(exp(x)) - x * z + log(1 + exp(-x))\n *    = - x * z + log(1 + exp(x))\n * Hence, to ensure stability and avoid overflow, the implementation uses this\n * equivalent formulation\n *    max(x, 0) - x * z + log(1 + exp(-abs(x)))\n *\n * @param labels The labels.\n * @param logits The logits.\n */\nexport function sigmoidCrossEntropyWithLogits(\n    labels: Tensor, logits: Tensor): Tensor {\n  if (!util.arraysEqual(labels.shape, logits.shape)) {\n    throw new ValueError(\n        `logits and labels must have the same shape, but got shapes ` +\n        `${JSON.stringify(labels.shape)} and ${JSON.stringify(logits.shape)}`);\n  }\n  return tidy(() => {\n    // The logistic loss formula from above is\n    //   x - x * z + log(1 + exp(-x))\n    // For x < 0, a more numerically stable formula is\n    //   -x * z + log(1 + exp(x))\n    // Note that these two expressions can be combined into the following:\n    //   max(x, 0) - x * z + log(1 + exp(-abs(x)))\n    const reluLogits = logits.relu();\n    const negAbsLogits = logits.abs().neg();\n    return reluLogits.sub(logits.mul(labels)).add(negAbsLogits.exp().log1p());\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    let y: Tensor;\n    y = tfc.clipByValue(yPred, epsilon(), 1 - epsilon());\n    y = tfc.log(tfc.div(y, tfc.sub(1, y)));\n    return tfc.mean(sigmoidCrossEntropyWithLogits(yTrue, y), -1);\n  });\n}\n\nexport function kullbackLeiblerDivergence(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const clippedTrue = tfc.clipByValue(yTrue, epsilon(), 1);\n    const clippedPred = tfc.clipByValue(yPred, epsilon(), 1);\n    return tfc.sum(\n        tfc.mul(yTrue, tfc.log(tfc.div(clippedTrue, clippedPred))), -1);\n  });\n}\n\nexport function poisson(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const logPred = tfc.log(tfc.add(epsilon(), yPred));\n    return tfc.mean(tfc.sub(yPred, tfc.mul(yTrue, logPred)), -1);\n  });\n}\n\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const trueNormalized = l2Normalize(yTrue, -1);\n    const predNormalized = l2Normalize(yPred, -1);\n    const trueXPred = tfc.mul(trueNormalized, predNormalized);\n    return tfc.neg(tfc.sum(trueXPred, -1));\n  });\n}\n\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const msle = meanSquaredLogarithmicError;\nexport const MSLE = meanSquaredLogarithmicError;\nexport const kld = kullbackLeiblerDivergence;\nexport const KLD = kullbackLeiblerDivergence;\nexport const cosine = cosineProximity;\n\n// TODO(michaelterry): Add deserialize() function.\n\nexport const lossesMap: {[functionName: string]: LossOrMetricFn} = {\n  meanSquaredError,\n  meanAbsoluteError,\n  meanAbsolutePercentageError,\n  meanSquaredLogarithmicError,\n  squaredHinge,\n  hinge,\n  categoricalHinge,\n  logcosh,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  binaryCrossentropy,\n  kullbackLeiblerDivergence,\n  poisson,\n  cosineProximity\n};\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function get(identifierOrFn: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifierOrFn === 'string') {\n    if (identifierOrFn in lossesMap) {\n      return lossesMap[identifierOrFn];\n    }\n    let errMsg = `Unknown loss ${identifierOrFn}`;\n    if (identifierOrFn.toLowerCase().includes('softmaxcrossentropy')) {\n      errMsg = `Unknown loss ${identifierOrFn}. ` +\n          'Use \"categoricalCrossentropy\" as the string name for ' +\n          'tf.losses.softmaxCrossEntropy';\n    }\n    throw new ValueError(errMsg);\n  } else {\n    return identifierOrFn;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Built-in metrics.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport * as K from './backend/tfjs_backend';\nimport {NotImplementedError, ValueError} from './errors';\nimport {categoricalCrossentropy as categoricalCrossentropyLoss, cosineProximity, meanAbsoluteError, meanAbsolutePercentageError, meanSquaredError, sparseCategoricalCrossentropy as sparseCategoricalCrossentropyLoss} from './losses';\nimport {binaryCrossentropy as lossBinaryCrossentropy} from './losses';\nimport {lossesMap} from './losses';\nimport {LossOrMetricFn} from './types';\nimport * as util from './utils/generic_utils';\n\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const threshold = tfc.mul(.5, tfc.onesLike(yPred));\n    const yPredThresholded = K.cast(tfc.greater(yPred, threshold), yTrue.dtype);\n    return tfc.mean(tfc.equal(yTrue, yPredThresholded), -1);\n  });\n}\n\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(\n      () => K.cast(\n          tfc.equal(tfc.argMax(yTrue, -1), tfc.argMax(yPred, -1)), 'float32'));\n}\n\nfunction truePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nfunction falseNegatives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(1), yPred.equal(0)).sum().cast('float32');\n  });\n}\n\nfunction falsePositives(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    return tfc.logicalAnd(yTrue.equal(0), yPred.equal(1)).sum().cast('float32');\n  });\n}\n\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fp = falsePositives(yTrue, yPred);\n\n    const denominator = tp.add(fp);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return tidy(() => {\n    const tp = truePositives(yTrue, yPred);\n    const fn = falseNegatives(yTrue, yPred);\n\n    const denominator = tp.add(fn);\n\n    return tfc.where(tfc.greater(denominator, 0), tp.div(denominator), 0)\n        .cast('float32');\n  });\n}\n\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return lossBinaryCrossentropy(yTrue, yPred);\n}\n\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  if (yTrue.rank === yPred.rank) {\n    yTrue = yTrue.squeeze([yTrue.rank - 1]);\n  }\n  yPred = yPred.argMax(-1);\n  if (yPred.dtype !== yTrue.dtype) {\n    yPred = yPred.asType(yTrue.dtype);\n  }\n  return tfc.equal(yTrue, yPred).asType('float32');\n}\n\nexport function topKCategoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\nexport function sparseTopKCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  throw new NotImplementedError();\n}\n\n// Aliases.\nexport const mse = meanSquaredError;\nexport const MSE = meanSquaredError;\nexport const mae = meanAbsoluteError;\nexport const MAE = meanAbsoluteError;\nexport const mape = meanAbsolutePercentageError;\nexport const MAPE = meanAbsolutePercentageError;\nexport const categoricalCrossentropy = categoricalCrossentropyLoss;\nexport const cosine = cosineProximity;\nexport const sparseCategoricalCrossentropy = sparseCategoricalCrossentropyLoss;\n\n// TODO(cais, nielsene): Add serialize().\n\nexport const metricsMap: {[functionName: string]: LossOrMetricFn} = {\n  binaryAccuracy,\n  categoricalAccuracy,\n  precision,\n  categoricalCrossentropy,\n  sparseCategoricalCrossentropy,\n  mse,\n  MSE,\n  mae,\n  MAE,\n  mape,\n  MAPE,\n  cosine\n};\n\nexport function get(identifier: string|LossOrMetricFn): LossOrMetricFn {\n  if (typeof identifier === 'string' && identifier in metricsMap) {\n    return metricsMap[identifier];\n  } else if (typeof identifier !== 'string' && identifier != null) {\n    return identifier;\n  } else {\n    throw new ValueError(`Unknown metric ${identifier}`);\n  }\n}\n\n/**\n * Get the shortcut function name.\n *\n * If the fn name is a string,\n *   directly return the string name.\n * If the function is included in metricsMap or lossesMap,\n *   return key of the map.\n *   - If the function relative to multiple keys,\n *     return the first found key as the function name.\n *   - If the function exists in both lossesMap and metricsMap,\n *     search lossesMap first.\n * If the function is not included in metricsMap or lossesMap,\n *   return the function name.\n *\n * @param fn loss function, metric function, or short cut name.\n * @returns Loss or Metric name in string.\n */\nexport function getLossOrMetricName(fn: string|LossOrMetricFn): string {\n  util.assert(fn !== null, `Unknown LossOrMetricFn ${fn}`);\n  if (typeof fn === 'string') {\n    return fn;\n  } else {\n    let fnName;\n    for (const key of Object.keys(lossesMap)) {\n      if (lossesMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    for (const key of Object.keys(metricsMap)) {\n      if (metricsMap[key] === fn) {\n        fnName = key;\n        break;\n      }\n    }\n    if (fnName !== undefined) {\n      return fnName;\n    }\n    return (fn as Function).name;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Optimizers.\n */\n\nimport {Optimizer, train} from '@tensorflow/tfjs-core';\n\nimport {epsilon} from './backend/common';\n\nimport {ValueError} from './errors';\n\n// Add (de)serialize()\n\n// Porting note: This diverges from the PyKeras implementation and may need to\n// change based on (de)serialization requirements.\nexport function getOptimizer(identifier: string): Optimizer {\n  const optimizerMap: {[optimizerName: string]: () => Optimizer} = {\n    'Adagrad': () => train.adagrad(0.01),\n    'Adadelta': () => train.adadelta(1, 0.95, epsilon()),\n    'Adam': () => train.adam(0.001, 0.9, 0.999, epsilon()),\n    'Adamax': () => train.adamax(0.002, 0.9, 0.999, epsilon(), 0),\n    'RMSProp': () => train.rmsprop(0.001, 0.9, 0, epsilon()),\n    'SGD': () => train.sgd(0.01)\n  };\n  optimizerMap['adagrad'] = optimizerMap['Adagrad'];\n  optimizerMap['adadelta'] = optimizerMap['Adadelta'];\n  optimizerMap['adam'] = optimizerMap['Adam'];\n  optimizerMap['adamax'] = optimizerMap['Adamax'];\n  optimizerMap['rmsprop'] = optimizerMap['RMSProp'];\n  optimizerMap['sgd'] = optimizerMap['SGD'];\n\n  if (identifier in optimizerMap) {\n    return optimizerMap[identifier]();\n  }\n  throw new ValueError(`Unknown Optimizer ${identifier}`);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/** Utility functions related to user-defined metadata. */\n\n// Maximum recommended serialized size for user-defined metadata.\n// Beyond this limit, a warning message will be printed during model loading and\n// saving.\nexport const MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH = 1 * 1024 * 1024;\n\n/**\n * Check validity of user-defined metadata.\n *\n * @param userDefinedMetadata\n * @param modelName Name of the model that the user-defined metadata belongs to.\n *   Used during construction of error messages.\n * @param checkSize Whether to check the size of the metadata is under\n *   recommended limit. Default: `false`. If `true`, will try stringify the\n *   JSON object and print a console warning if the serialzied size is above the\n *   limit.\n * @throws Error if `userDefinedMetadata` is not a plain JSON object.\n */\nexport function checkUserDefinedMetadata(\n    userDefinedMetadata: {}, modelName: string, checkSize = false): void {\n  if (userDefinedMetadata == null ||\n      typeof userDefinedMetadata !== 'object' ||\n      Object.getPrototypeOf(userDefinedMetadata) !== Object.prototype ||\n      !plainObjectCheck(userDefinedMetadata)) {\n    throw new Error(\n        'User-defined metadata is expected to be a JSON object, but is not.');\n  }\n\n  if (checkSize) {\n    const out = JSON.stringify(userDefinedMetadata);\n    if (out.length > MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH) {\n      console.warn(\n          `User-defined metadata of model \"${modelName}\" is too large in ` +\n          `size (length=${out.length} when serialized). It is not ` +\n          `recommended to store such large objects in user-defined metadata. ` +\n          `Please make sure its serialized length is <= ` +\n          `${MAX_USER_DEFINED_METADATA_SERIALIZED_LENGTH}.`);\n    }\n  }\n}\n\n/**\n * Check if an input is plain JSON object or any valid subfield of it.\n *\n * @param x The input to be checked.\n * @param assertObject Whether to assert `x` is a JSON object, i.e., reject\n *   cases of arrays and primitives.\n * @return Returns `true` if and only if `x` is a plain JSON object,\n *   a JSON-valid primitive including string, number, boolean and null,\n *   or an array of the said types.\n */\n// tslint:disable-next-line:no-any\nexport function plainObjectCheck(x: any): boolean {\n  if (x === null) {\n    // Note: typeof `null` is 'object', and `null` is valid in JSON.\n    return true;\n  } else if (typeof x === 'object') {\n    if (Object.getPrototypeOf(x) === Object.prototype) {\n      // `x` is a JavaScript object and its prototype is Object.\n      const keys = Object.keys(x);\n      for (const key of keys) {\n        if (typeof key !== 'string') {\n          // JSON keys must be strings.\n          return false;\n        }\n        if (!plainObjectCheck(x[key])) {  // Recursive call.\n          return false;\n        }\n      }\n      return true;\n    } else {\n      // `x` is a JavaScript object but its prototype is not Object.\n      if (Array.isArray(x)) {\n        // `x` is a JavaScript array.\n        for (const item of x) {\n          if (!plainObjectCheck(item)) {  // Recursive call.\n            return false;\n          }\n        }\n        return true;\n      } else {\n        // `x` is a JavaScript object and its prototype is not Object,\n        // and it's not an Array. I.e., it's a complex object such as\n        // `Error` and `Date`.\n        return false;\n      }\n    }\n  } else {\n    // `x` is not a JavaScript object or `null`.\n    const xType = typeof x;\n    return xType === 'string' || xType === 'number' || xType === 'boolean';\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {Container} from '../engine/container';\nimport {Layer, Node} from '../engine/topology';\nimport {countParamsInWeights} from './variable_utils';\n\n/**\n * Print the summary of a LayersModel object.\n *\n * @param model tf.LayersModel instance.\n * @param lineLength Total length of printed lines. Set this to adapt to the\n *   display to different terminal or console sizes.\n * @param positions Relative or absolute positions of log elements in each\n *   line. Each number corresponds to right-most (i.e., ending) position of a\n *   column.\n *   If not provided, defaults to `[0.45, 0.85, 1]` for sequential-like\n *   models and `[0.33, 0.55, 0.67, 1]` for non-sequential like models.\n * @param printFn Print function to use.\n *   It will be called on each line of the summary. You can provide a custom\n *   function in order to capture the string summary. Defaults to `console.log`.\n */\nexport function printSummary(\n    model: Container, lineLength?: number, positions?: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void =\n        console.log): void {\n  const sequentialLike = isModelSequentialLike(model);\n\n  // Header names for different log elements.\n  const toDisplay: string[] = ['Layer (type)', 'Output shape', 'Param #'];\n  if (sequentialLike) {\n    lineLength = lineLength || 65;\n    positions = positions || [0.45, 0.85, 1];\n  } else {\n    lineLength = lineLength || 98;\n    positions = positions || [0.33, 0.55, 0.67, 1];\n    // Header names for different log elements.\n  }\n\n  if (positions[positions.length - 1] <= 1) {\n    // `positions` is relative. Convert it to absolute positioning.\n    positions = positions.map(p => Math.floor(lineLength * p));\n  }\n\n  let relevantNodes: Node[];\n  if (!sequentialLike) {\n    toDisplay.push('Receives inputs');\n    relevantNodes = [];\n    for (const depth in model.nodesByDepth) {\n      relevantNodes.push(...model.nodesByDepth[depth]);\n    }\n  }\n\n  printFn('_'.repeat(lineLength));\n  printRow(toDisplay, positions, printFn);\n  printFn('='.repeat(lineLength));\n\n  const layers = model.layers;\n  for (let i = 0; i < layers.length; ++i) {\n    if (sequentialLike) {\n      printLayerSummary(layers[i], positions, printFn);\n    } else {\n      printLayerSummaryWithConnections(\n          layers[i], positions, relevantNodes, printFn);\n    }\n    printFn((i === layers.length - 1 ? '=' : '_').repeat(lineLength));\n  }\n\n  // tslint:disable-next-line:no-any\n  (model as any).checkTrainableWeightsConsistency();\n\n  const trainableCount = countTrainableParams(model);\n  const nonTrainableCount = countParamsInWeights(model.nonTrainableWeights);\n\n  printFn(`Total params: ${trainableCount + nonTrainableCount}`);\n  printFn(`Trainable params: ${trainableCount}`);\n  printFn(`Non-trainable params: ${nonTrainableCount}`);\n  printFn('_'.repeat(lineLength));\n}\n\nfunction countTrainableParams(model: Container): number {\n  let trainableCount: number;\n  // tslint:disable:no-any\n  if ((model as any).collectedTrainableWeights != null) {\n    trainableCount =\n        countParamsInWeights((model as any).collectedTrainableWeights);\n  } else {\n    trainableCount = countParamsInWeights(model.trainableWeights);\n  }\n  // tslint:enable:no-any\n  return trainableCount;\n}\n\nfunction isModelSequentialLike(model: Container): boolean {\n  let sequentialLike = true;\n  const nodesByDepth: Node[][] = [];\n  const nodes: Node[] = [];\n  for (const depth in model.nodesByDepth) {\n    nodesByDepth.push(model.nodesByDepth[depth]);\n  }\n  for (const depthNodes of nodesByDepth) {\n    if (depthNodes.length > 1 ||\n        depthNodes.length === 1 && depthNodes[0].inboundLayers.length > 1) {\n      sequentialLike = false;\n      break;\n    }\n    nodes.push(...depthNodes);\n  }\n  if (sequentialLike) {\n    // Search for shared layers.\n    for (const layer of model.layers) {\n      let flag = false;\n      for (const node of layer.inboundNodes) {\n        if (nodes.indexOf(node) !== -1) {\n          if (flag) {\n            sequentialLike = false;\n            break;\n          } else {\n            flag = true;\n          }\n        }\n      }\n      if (!sequentialLike) {\n        break;\n      }\n    }\n  }\n  return sequentialLike;\n}\n\nfunction printRow(\n    fields: string[], positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void = console.log) {\n  let line = '';\n  for (let i = 0; i < fields.length; ++i) {\n    if (i > 0) {\n      line = line.slice(0, line.length - 1) + ' ';\n    }\n    line += fields[i];\n    line = line.slice(0, positions[i]);\n    line += ' '.repeat(positions[i] - line.length);\n  }\n  printFn(line);\n}\n\n/**\n * Prints a summary for a single Layer, without connectivity information.\n *\n * @param layer: Layer instance to print.\n */\nfunction printLayerSummary(\n    layer: Layer, positions: number[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const name = layer.name;\n  const className = layer.getClassName();\n  const fields: string[] =\n      [`${name} (${className})`, outputShape, layer.countParams().toString()];\n  printRow(fields, positions, printFn);\n}\n\n/**\n * Prints a summary for a single Layer, with connectivity information.\n */\nfunction printLayerSummaryWithConnections(\n    layer: Layer, positions: number[], relevantNodes: Node[],\n    // tslint:disable-next-line:no-any\n    printFn: (message?: any, ...optionalParams: any[]) => void) {\n  let outputShape: string;\n  try {\n    outputShape = JSON.stringify(layer.outputShape);\n  } catch (err) {\n    outputShape = 'multiple';\n  }\n\n  const connections: string[] = [];\n  for (const node of layer.inboundNodes) {\n    if (relevantNodes != null && relevantNodes.length > 0 &&\n        relevantNodes.indexOf(node) === -1) {\n      continue;\n    }\n    for (let i = 0; i < node.inboundLayers.length; ++i) {\n      const inboundLayer = node.inboundLayers[i].name;\n      const inboundLayerIndex = node.nodeIndices[i];\n      const inboundTensorIndex = node.tensorIndices[i];\n      connections.push(\n          `${inboundLayer}[${inboundLayerIndex}][${inboundTensorIndex}]`);\n    }\n  }\n  const name = layer.name;\n  const className = layer.getClassName();\n  const firstConnection = connections.length === 0 ? '' : connections[0];\n  const fields: string[] = [\n    `${name} (${className})`, outputShape, layer.countParams().toString(),\n    firstConnection\n  ];\n\n  printRow(fields, positions, printFn);\n  for (let i = 1; i < connections.length; ++i) {\n    printRow(['', '', '', connections[i]], positions, printFn);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Porting note: This file doesn't exist in PyKeras.\n// Its purpose here is to centralize the boundary layer between\n// tfjs-layers's internal Config TS-Centric format and PyKeras's\n// serialized Python Config format.\n\nimport {serialization} from '@tensorflow/tfjs-core';\n\nimport {PyJsonValue} from '../keras_format/types';\nimport * as generic_utils from '../utils/generic_utils';\n// tslint:enable\n\n/**\n * Test whether a value in an array is the name of a LayersModel or Layer.\n * @param key The key name that the value is found under. Note that the key\n *   may not be at the level immediately above the value, if the value is in a\n *   nested array.\n * @param index Index of the value in the Array that it is found in.\n * @param value The value object.\n * @returns A boolean indicating whether value is a name.\n */\nfunction isArrayItemInputOrOutputName<T>(\n    key: string, index: number, value: T): boolean {\n  return (key === 'inboundNodes' || key === 'outputLayers' ||\n          key === 'inputLayers') &&\n      index === 0 && typeof value === 'string';\n}\n\n/**\n * Convert a Pythonic config object to TypeScript config object.\n * @param pythonicConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertPythonicToTs(\n    pythonicConfig: PyJsonValue, key?: string): serialization.ConfigDictValue {\n  if (pythonicConfig === null) {\n    return null;\n  } else if (typeof pythonicConfig === 'string') {\n    return generic_utils.toCamelCase(pythonicConfig);\n  } else if (\n      (typeof pythonicConfig === 'number') ||\n      (typeof pythonicConfig === 'boolean')) {\n    return pythonicConfig;\n  } else if (pythonicConfig instanceof Array) {\n    const tsArray = [];\n    const arrayLength = pythonicConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = pythonicConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        tsArray.push(item);\n      } else {\n        tsArray.push(convertPythonicToTs(item, key));\n      }\n    }\n    return tsArray;\n  } else {\n    const tsDict: serialization.ConfigDict = {};\n    for (const pythonicKey of Object.keys(pythonicConfig)) {\n      const pythonicValue = pythonicConfig[pythonicKey];\n      if (pythonicKey === 'name' && typeof pythonicValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // camel-case conversion.\n        tsDict[pythonicKey] = pythonicValue;\n      } else {\n        const tsKey = generic_utils.toCamelCase(pythonicKey);\n        tsDict[tsKey] = convertPythonicToTs(pythonicValue, tsKey);\n      }\n    }\n    return tsDict;\n  }\n}\n\n/**\n * Convert a TypeScript config object to Python config object.\n * @param tsConfig The config object to convert.\n * @param key Optional key name of the object being converted.\n * @returns Result of the conversion.\n */\nexport function convertTsToPythonic(\n    tsConfig: serialization.ConfigDictValue, key?: string): PyJsonValue {\n  if (tsConfig === null || tsConfig === undefined) {\n    return null;\n  } else if (typeof tsConfig === 'string') {\n    return generic_utils.toSnakeCase(tsConfig);\n  } else if (\n      (typeof tsConfig === 'number') || (typeof tsConfig === 'boolean')) {\n    return tsConfig;\n  } else if (tsConfig instanceof Array) {\n    const pyArray = [];\n    const arrayLength = tsConfig.length;\n    for (let i = 0; i < arrayLength; ++i) {\n      const item = tsConfig[i];\n      if (isArrayItemInputOrOutputName(key, i, item)) {\n        pyArray.push(item);\n      } else {\n        pyArray.push(convertTsToPythonic(item, key));\n      }\n    }\n    return pyArray;\n  } else {\n    const pyDict: serialization.ConfigDict = {};\n    for (const tsKey of Object.keys(tsConfig)) {\n      const tsValue = tsConfig[tsKey];\n      const pyKey = generic_utils.toSnakeCase(tsKey);\n      if ((tsKey === 'name' || tsKey === 'className') &&\n          typeof tsValue === 'string') {\n        // Special case the 'name' key with a string value. Name values, such as\n        // the names of LayersModel and Layer instances, should not undergo the\n        // snake-case conversion.\n        pyDict[pyKey] = tsValue;\n      } else {\n        pyDict[pyKey] = convertTsToPythonic(tsValue, tsKey);\n      }\n    }\n    return pyDict;\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Executor: Evaluates SymbolicTensor based on feeds.\n */\n\nimport {cast, dispose, memory, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {ValueError} from '../errors';\nimport {Kwargs} from '../types';\nimport {toList} from '../utils/generic_utils';\n\nimport {InputLayer} from './input_layer';\nimport {SymbolicTensor} from './topology';\n\n/**\n * Helper function to check the dtype and shape compatibility of a feed value.\n */\nfunction assertFeedCompatibility(key: SymbolicTensor, val: Tensor): Tensor {\n  // Check dtype compatibility.\n  if (key.dtype == null || key.dtype === val.dtype) {\n    //  a.  If types match, return val tensor as is.\n    return val;\n  }\n  try {\n    //  b. Attempt to convert to expected type.\n    return cast(val, key.dtype);\n  } catch (err) {\n    //  c. If conversion fails, return helpful error.\n    throw new ValueError(\n        `The dtype of the feed (${val.dtype}) can not be cast to the dtype ` +\n        `of the key '${key.name}' (${key.dtype}).`);\n  }\n}\n\n/**\n * A concrete Tensor value for a symbolic tensor as the key.\n */\nexport interface Feed {\n  key: SymbolicTensor;\n  value: Tensor;\n}\n\n/**\n * FeedDict: A mapping from unique SymbolicTensors to feed values for them.\n * A feed value is a concrete value represented as an `Tensor`.\n */\nexport class FeedDict {\n  private id2Value: {[id: number]: Tensor} = {};\n  private id2Mask: {[id: number]: Tensor} = {};\n  private name2Id: {[name: string]: number} = {};\n\n  /**\n   * Constructor, optionally does copy-construction.\n   * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case\n   *   copy-construction will be performed.\n   */\n  constructor(feeds?: Feed[]|FeedDict) {\n    if (feeds instanceof FeedDict) {\n      for (const id in feeds.id2Value) {\n        this.id2Value[id] = feeds.id2Value[id];\n        if (id in feeds.id2Mask) {\n          this.id2Mask[id] = feeds.id2Mask[id];\n        }\n      }\n    } else {\n      if (feeds == null) {\n        return;\n      }\n      for (const feed of feeds) {\n        this.add(feed.key, feed.value);\n      }\n    }\n  }\n\n  /**\n   * Add a key-value pair to the FeedDict.\n   *\n   * @param key The key of the feed.\n   * @param value The value of the tensor feed.\n   * @param mask The value of the mask feed (optional).\n   * @returns This `FeedDict`.\n   * @throws ValueError: If the key `SymbolicTensor` already exists in the\n   *   `FeedDict`.\n   */\n  add(key: SymbolicTensor, value: Tensor, mask?: Tensor): FeedDict {\n    if (this.id2Value[key.id] == null) {\n      this.id2Value[key.id] = assertFeedCompatibility(key, value);\n      this.name2Id[key.name] = key.id;\n      if (mask != null) {\n        this.id2Mask[key.id] = mask;\n      }\n    } else {\n      throw new ValueError(`Duplicate key: name=${key.name}, id=${key.id}`);\n    }\n    return this;\n  }\n\n  /**\n   * Add a Feed to the FeedDict.\n   * @param feed The new `Feed` to add.\n   * @returns This `FeedDict`.\n   */\n  addFeed(feed: Feed) {\n    this.add(feed.key, feed.value);\n  }\n\n  /**\n   * Probe whether a key already exists in the FeedDict.\n   * @param key\n   */\n  hasKey(key: SymbolicTensor): boolean {\n    return this.id2Value[key.id] != null;\n  }\n\n  /**\n   * Get all the SymbolicTensor available in this FeedDict.\n   */\n  names(): string[] {\n    return Object.keys(this.name2Id);\n  }\n\n  /**\n   * Get the feed value for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed value.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getValue(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Value[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Value[id];\n    }\n  }\n\n  /**\n   * Get the feed mask for given key.\n   * @param key The SymbolicTensor, or its name (as a string), of which the\n   *     value is sought.\n   * @returns If `key` exists, the corresponding feed mask.\n   * @throws ValueError: If `key` does not exist in this `FeedDict`.\n   */\n  getMask(key: SymbolicTensor|string): Tensor {\n    if (key instanceof SymbolicTensor) {\n      if (this.id2Value[key.id] == null) {\n        throw new ValueError(`Nonexistent key: ${key.name}`);\n      } else {\n        return this.id2Mask[key.id];\n      }\n    } else {\n      const id = this.name2Id[key];\n      if (id == null) {\n        throw new ValueError(`Feed dict has no SymbolicTensor name: ${key}`);\n      }\n      return this.id2Mask[id];\n    }\n  }\n\n  /** Dispose all mask Tensors held by this object. */\n  disposeMasks() {\n    if (this.id2Mask != null) {\n      dispose(this.id2Mask);\n    }\n  }\n}\n\n// Cache for topologically sorted SymbolicTensors for given execution\n// targets (i.e., fetches).\nconst cachedSorted: {[concatFetchNames: string]: SymbolicTensor[]} = {};\n\n// Cache for recipient count maps for given execution targets (i.e., fetches).\nconst cachedRecipientCounts:\n    {[concatFetchNames: string]: {[fetchName: string]: number}} = {};\n\n/**\n * Interface for the optional object used for probing the memory\n * usage and other statistics during execution.\n */\nexport interface ExecutionProbe {\n  /**\n   * Maximum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  maxNumTensors?: number;\n\n  /**\n   * Minimum number of tensors that exist during all steps of the\n   * execution. Tensor counts are measured at the beginning of every\n   * step.\n   */\n  minNumTensors?: number;\n}\n\n/**\n * Execute a SymbolicTensor by using concrete feed values.\n *\n * A `SymbolicTensor` object is a node in a computation graph of TF.js\n * Layers. The object is backed by a source layer and input\n * `SymbolicTensor`s to the source layer. This method evaluates\n * the `call()` method of the source layer, using concrete values of the\n * inputs obtained from either\n * * `feedDict`, if the input key exists in `feedDict`, or else,\n * * a recursive call to `execute()` itself.\n *\n * @param x: The `SymbolicTensor` to execute.\n * @param feedDict: The feed values, as base condition of the recursion.\n *   execution.\n * @param kwargs: Optional keyword arguments.\n * @param probe: A probe object (of interface `ExecutionProbe`) used for\n *   testing memory footprint of `execute` calls.\n * @returns Result of the execution.\n * @throws ValueError: If any `SymbolicTensor`s from `InputLayer`s\n *   encountered during the execution lacks a feed value in `feedDict`.\n */\nexport function execute(\n    fetches: SymbolicTensor|SymbolicTensor[], feedDict: FeedDict,\n    kwargs?: Kwargs, probe?: ExecutionProbe): Tensor|\n    Tensor[]|[Tensor | Tensor[]] {\n  const training: boolean = kwargs == null ? false : kwargs['training'];\n\n  const arrayFetches = Array.isArray(fetches);\n  const fetchArray: SymbolicTensor[] =\n      arrayFetches ? fetches as SymbolicTensor[] : [fetches as SymbolicTensor];\n\n  const outputNames = fetchArray.map(t => t.name);\n  const finalOutputs: Tensor[] = [];\n  const feedNames = feedDict.names();\n  for (const outputName of outputNames) {\n    if (feedNames.indexOf(outputName) !== -1) {\n      finalOutputs.push(feedDict.getValue(outputName));\n    } else {\n      finalOutputs.push(null);\n    }\n  }\n\n  if (probe != null) {\n    // For optional probing of memory footprint during execution.\n    probe.maxNumTensors = -Infinity;\n    probe.minNumTensors = Infinity;\n  }\n\n  // Check cache.\n  const fetchAndFeedKey =\n      outputNames.join(',') + '|' + feedDict.names().join(',');\n  let sorted: SymbolicTensor[];\n  let recipientCounts: {[fetchName: string]: number};\n  if (cachedSorted[fetchAndFeedKey] == null) {\n    // Cache doesn't contain the desired combination of fetches. Compute\n    // topological sort for the combination for the first time.\n    const out = getTopologicalSortAndRecipientCounts(fetchArray, feedDict);\n    sorted = out.sorted;\n    recipientCounts = out.recipientCounts;\n\n    // Store results in cache for future use.\n    cachedSorted[fetchAndFeedKey] = sorted;\n    cachedRecipientCounts[fetchAndFeedKey] = recipientCounts;\n  }\n  sorted = cachedSorted[fetchAndFeedKey];\n  recipientCounts = {};\n  if (!training) {\n    Object.assign(recipientCounts, cachedRecipientCounts[fetchAndFeedKey]);\n  }\n\n  const internalFeedDict = new FeedDict(feedDict);\n\n  // Start iterative execution on the topologically-sorted SymbolicTensors.\n  for (let i = 0; i < sorted.length; ++i) {\n    if (probe != null) {\n      // For optional probing of memory usage during execution.\n      const numTensors = memory().numTensors;\n      if (numTensors > probe.maxNumTensors) {\n        probe.maxNumTensors = numTensors;\n      }\n      if (numTensors < probe.minNumTensors) {\n        probe.minNumTensors = numTensors;\n      }\n    }\n\n    const symbolic = sorted[i];\n    const srcLayer = symbolic.sourceLayer;\n    if (srcLayer instanceof InputLayer) {\n      continue;\n    }\n    const inputValues: Tensor[] = [];\n    const inputMasks: Tensor[] = [];\n    const tensorsToDispose: Tensor[] = [];\n\n    let maskExists = false;\n    for (const input of symbolic.inputs) {\n      const value = internalFeedDict.getValue(input);\n      const mask = internalFeedDict.getMask(input);\n      inputValues.push(value);\n      inputMasks.push(mask);\n      if (mask != null) {\n        maskExists = true;\n      }\n      if (!training) {\n        recipientCounts[input.name]--;\n        if (recipientCounts[input.name] === 0 && !feedDict.hasKey(input) &&\n            outputNames.indexOf(input.name) === -1 && !value.isDisposed &&\n            input.sourceLayer.stateful !== true) {\n          tensorsToDispose.push(value);\n        }\n      }\n    }\n\n    if (maskExists) {\n      kwargs = kwargs || {};\n      kwargs['mask'] = inputMasks[0];\n    }\n    const outputTensors =\n        toList(srcLayer.apply(inputValues, kwargs)) as Tensor[];\n    let outputMask: Tensor|Tensor[] = null;\n    if (srcLayer.supportsMasking) {\n      outputMask = srcLayer.computeMask(inputValues, inputMasks);\n    }\n    const layerOutputs = getNodeOutputs(symbolic);\n    const outputSymbolicTensors =\n        Array.isArray(layerOutputs) ? layerOutputs : [layerOutputs];\n    for (let i = 0; i < outputSymbolicTensors.length; ++i) {\n      if (!internalFeedDict.hasKey(outputSymbolicTensors[i])) {\n        internalFeedDict.add(\n            outputSymbolicTensors[i], outputTensors[i],\n            Array.isArray(outputMask) ? outputMask[0] : outputMask);\n      }\n      const index = outputNames.indexOf(outputSymbolicTensors[i].name);\n      if (index !== -1) {\n        finalOutputs[index] = outputTensors[i];\n      }\n    }\n\n    if (!training) {\n      // Clean up Tensors that are no longer needed.\n      dispose(tensorsToDispose);\n    }\n  }\n  // NOTE(cais): Unlike intermediate tensors, we don't discard mask\n  // tensors as we go, because these tensors are sometimes passed over a\n  // series of mutliple layers, i.e., not obeying the immediate input\n  // relations in the graph. If this becomes a memory-usage concern,\n  // we can improve this in the future.\n  internalFeedDict.disposeMasks();\n\n  return arrayFetches ? finalOutputs : finalOutputs[0];\n}\n\ntype RecipientCounts = {\n  [fetchName: string]: number\n};\n\nexport type RecipientMap = {\n  [fetchName: string]: Set<string>;\n};\n\n/**\n * Sort the `SymbolicTensor`s topologically, for an array of fetches.\n *\n * This function calls getTopologicalSortAndRecipientCountsForOneFetch and\n * merges their results.\n *\n * @param fetch The array of fetches requested. Must be a non-empty array.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientCounts: Recipient counts for all SymbolicTensors in `sorted`.\n */\nfunction getTopologicalSortAndRecipientCounts(\n    fetches: SymbolicTensor[], feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientCounts: RecipientCounts} {\n  util.assert(\n      fetches != null && fetches.length > 0,\n      () => `Expected at least one fetch, got none`);\n\n  let finalSorted: SymbolicTensor[] = [];\n  let finalRecipientMap: RecipientMap = {};\n  if (fetches.length === 1) {\n    // Special-casing 1 fetch for efficiency.\n    const out =\n        getTopologicalSortAndRecipientCountsForOneFetch(fetches[0], feedDict);\n    finalSorted = out.sorted;\n    finalRecipientMap = out.recipientMap;\n  } else {\n    const visited = new Set<string>();\n    for (const fetch of fetches) {\n      const {sorted, recipientMap} =\n          getTopologicalSortAndRecipientCountsForOneFetch(fetch, feedDict);\n\n      // Merge sorted SymbolicTensor Arrays.\n      for (const symbolicTensor of sorted) {\n        if (!visited.has(symbolicTensor.name)) {\n          finalSorted.push(symbolicTensor);\n          visited.add(symbolicTensor.name);\n        }\n      }\n\n      // Merge recipient maps.\n      for (const name in recipientMap) {\n        if (finalRecipientMap[name] == null) {\n          finalRecipientMap[name] = new Set<string>();\n        }\n        recipientMap[name].forEach(\n            recipient => finalRecipientMap[name].add(recipient));\n      }\n    }\n  }\n  return {\n    sorted: finalSorted,\n    recipientCounts: recipientMap2Counts(finalRecipientMap)\n  };\n}\n\nfunction recipientMap2Counts(recipientMap: RecipientMap): RecipientCounts {\n  const recipientCounts: RecipientCounts = {};\n  for (const name in recipientMap) {\n    recipientCounts[name] = recipientMap[name].size;\n  }\n  return recipientCounts;\n}\n\n/**\n * Sort the `SymbolicTensor`s topologically, for a single fetch.\n *\n * This helper function processes the upstream SymbolicTensors of a single\n * fetch.\n *\n * @param fetch The single fetch requested.\n * @param feedDict The dictionary of fed values.\n * @returns sorted: Topologically-sorted array of SymbolicTensors.\n *   recipientMap: Recipient names for all SymbolicTensors in `sorted`.\n */\nexport function getTopologicalSortAndRecipientCountsForOneFetch(\n    fetch: SymbolicTensor, feedDict: FeedDict):\n    {sorted: SymbolicTensor[], recipientMap: RecipientMap} {\n  const visited = new Set<string>();\n  const sorted: SymbolicTensor[] = [];\n  const recipientMap: RecipientMap = {};\n\n  // Put keys of the feedDict into visited first, so they don't have to be\n  // walked. This is needed in case where there are feeds for intermediate\n  // SymbolicTensors of the graph.\n  for (const key of feedDict.names()) {\n    visited.add(key);\n  }\n\n  const stack: SymbolicTensor[] = [];\n  const marks: number[] = [];\n\n  // Initial population of stack and marks.\n  stack.push(fetch);\n\n  while (stack.length > 0) {\n    const top = stack[stack.length - 1];\n    if (visited.has(top.name)) {\n      stack.pop();\n      continue;\n    }\n    const topIsMarked = marks[marks.length - 1] === stack.length - 1;\n    if (top.inputs.length === 0 || topIsMarked) {\n      // Input SymbolicTensor or all children have been visited.\n      stack.pop();\n      sorted.push(top);\n      visited.add(top.name);\n      if (topIsMarked) {\n        marks.pop();\n      }\n    } else {\n      // A non-input SymbolicTensor whose upstream SymbolicTensors haven't\n      // been visited yet. Push them onto the stack.\n      marks.push(stack.length - 1);\n      for (const input of top.inputs) {\n        // Increment the recipient count. Note that this needs to happen\n        // regardless of whether the SymbolicTensor has been visited before.\n        if (recipientMap[input.name] == null) {\n          recipientMap[input.name] = new Set<string>();\n        }\n        recipientMap[input.name].add(top.name);\n\n        if (visited.has(input.name)) {\n          continue;  // Avoid repeated visits to the same SymbolicTensor.\n        }\n        stack.push(input);\n      }\n    }\n  }\n  return {sorted, recipientMap};\n}\n\n/**\n * Get the symbolic output tensors of the node to which a given fetch belongs.\n * @param fetch The fetched symbolic tensor.\n * @returns The Array of symbolic tensors output by the node to which `fetch`\n *   belongs.\n */\nfunction getNodeOutputs(fetch: SymbolicTensor): SymbolicTensor|\n    SymbolicTensor[] {\n  let layerOutputs: SymbolicTensor|SymbolicTensor[];\n  if (fetch.sourceLayer.inboundNodes.length === 1) {\n    layerOutputs = fetch.sourceLayer.output;\n  } else {\n    let nodeIndex: number = null;\n    for (let i = 0; i < fetch.sourceLayer.inboundNodes.length; ++i) {\n      for (const outputTensor of fetch.sourceLayer.inboundNodes[i]\n               .outputTensors) {\n        if (outputTensor.id === fetch.id) {\n          nodeIndex = i;\n          break;\n        }\n      }\n    }\n    layerOutputs = fetch.sourceLayer.getOutputAt(nodeIndex);\n  }\n  return layerOutputs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/engine/topology.py */\n\nimport {NamedTensorMap, Scalar, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {getUid} from '../backend/state';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {TensorKeyWithArgsArray} from '../keras_format/node_config';\nimport {PyJsonDict} from '../keras_format/types';\nimport {deserialize as deserializeLayer} from '../layers/serialization';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {convertTsToPythonic} from '../utils/serialization_utils';\nimport * as types_utils from '../utils/types_utils';\nimport {batchSetValue, LayerVariable} from '../variables';\nimport {version as layersVersion} from '../version';\n\nimport {execute, FeedDict} from './executor';\nimport {InputLayer} from './input_layer';\nimport {DisposeResult, Layer, Node, SymbolicTensor} from './topology';\n\n/** Constructor config for Container. */\nexport interface ContainerArgs {\n  inputs: SymbolicTensor|SymbolicTensor[];\n  outputs: SymbolicTensor|SymbolicTensor[];\n  name?: string;\n}\n\n/**\n * A Container is a directed acyclic graph of layers.\n *\n * It is the topological form of a \"model\". A LayersModel\n * is simply a Container with added training routines.\n *\n */\nexport abstract class Container extends Layer {\n  inputs: SymbolicTensor[];\n  outputs: SymbolicTensor[];\n\n  inputLayers: Layer[];\n  inputLayersNodeIndices: number[];\n  inputLayersTensorIndices: number[];\n\n  outputLayers: Layer[];\n  outputLayersNodeIndices: number[];\n  outputLayersTensorIndices: number[];\n\n  layers: Layer[];\n  layersByDepth: {[depth: string]: Layer[]};\n  nodesByDepth: {[depth: string]: Node[]};\n\n  internalContainerRefs: Container[];\n\n  containerNodes = new Set<string>();\n\n  // TODO(michaelterry): Add cache support\n  // private outputMaskCache: any;\n  // private outputTensorCache: any;\n  // private outputShapeCache: any;\n\n  inputNames: string[];\n  outputNames: string[];\n  feedInputShapes: Shape[];\n\n  protected internalInputShapes: Shape[];\n  protected internalOutputShapes: Shape[];\n  // TODO(cais): Maybe 'feed' should not in the names of these variables,\n  //   due to the fact that our backend is not symbolic.\n  protected feedInputNames: string[];\n  protected feedOutputNames: string[];\n\n  constructor(args: ContainerArgs) {\n    // No args passed to super's constructor.\n    super({});\n    this.name = args.name;\n    if (this.name == null) {\n      const prefix = this.getClassName().toLowerCase();\n      this.name = getUid(prefix);\n    }\n\n    this.supportsMasking = false;\n    this.trainable_ = true;\n\n    // TODO(michaelterry): Initialize perInputLosses/Updates here.\n\n    // Container-specific properties.\n    if (Array.isArray(args.inputs)) {\n      this.inputs = args.inputs.slice();\n    } else {\n      this.inputs = [args.inputs];\n    }\n    if (Array.isArray(args.outputs)) {\n      this.outputs = args.outputs.slice();\n    } else {\n      this.outputs = [args.outputs];\n    }\n\n    // Check for redundancy in inputs.\n    if (generic_utils.unique(this.inputs).length !== this.inputs.length) {\n      throw new ValueError(\n          'The list of inputs passed to the model is ' +\n          'redundant. All inputs should only appear once. Found: ' +\n          `${this.inputs.map(x => x.name)}`);\n    }\n\n    // Check for redundancy in outputs.\n    if (generic_utils.unique(this.outputs).length !== this.outputs.length) {\n      console.warn(\n          'The list of outputs passed to the model is redundant. ' +\n          'All outputs should only appear once. Found: ' +\n          `${this.outputs.map(x => x.name)}`);\n    }\n\n    /*\n      List of initial layers (1 to 1 mapping with this.inputs, hence the same\n      layer might appear twice)\n    */\n    this.inputLayers = [];\n    this.inputLayersNodeIndices = [];\n    this.inputLayersTensorIndices = [];\n    /*\n      List of layers (1 to 1 mapping with this.outputs, hence the same layer\n      might appear twice)\n    */\n    this.outputLayers = [];\n    this.outputLayersNodeIndices = [];\n    this.outputLayersTensorIndices = [];\n    /*\n      All layers in order of horizontal graph traversal. Entries are unique.\n      Includes input and output layers.\n    */\n    this.layers = [];\n\n    /*\n      References to container layers that were constructed internally. We need\n      these to properly dispose of tensors from nested containers.\n    */\n    this.internalContainerRefs = [];\n\n    // TODO(michaelterry): Determine if caching still needed with eager\n    // backend.\n    /*\n      This is for performance optimization when calling the Container on new\n      inputs. Every time the Container is called on a set on input tensors,\n      we compute the output tensors, output masks and output shapes in one pass,\n      then cache them here. When one of these outputs is queried later,\n      we retrieve it from there instead of recomputing it.\n    */\n    // this.outputTensorCache = {};\n    // this.outputShapeCache = {};\n\n    // Build this.outputLayers:\n    for (const x of this.outputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      this.outputLayers.push(layer);\n      this.outputLayersNodeIndices.push(nodeIndex);\n      this.outputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // TODO(michaelterry): Add output mask cache code.\n\n    // Build this.inputLayers:\n    for (const x of this.inputs) {\n      const layer = x.sourceLayer;\n      const nodeIndex = x.nodeIndex;\n      const tensorIndex = x.tensorIndex;\n      /*\n        It's supposed to be an input layer, so only one node\n        and one tensor output.\n      */\n      generic_utils.assert(nodeIndex === 0, 'input layer has >1 nodes');\n      generic_utils.assert(tensorIndex === 0, 'input layer has >1 tensors');\n      this.inputLayers.push(layer);\n      this.inputLayersNodeIndices.push(nodeIndex);\n      this.inputLayersTensorIndices.push(tensorIndex);\n    }\n\n    // Build this.inputNames and this.outputNames.\n    this.inputNames = [];\n    this.outputNames = [];\n    this.feedInputShapes = [];\n    this.feedInputNames = [];\n    this.feedOutputNames = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      // Check that layer is an InputLayer.\n      if (!(layer instanceof InputLayer)) {\n        throw new TypeError(\n            'Input layers to a LayersModel must be InputLayer objects. ' +\n            `Received inputs: ${args.inputs}. ` +\n            `Input ${i} (0-based) originates ` +\n            `from layer type ${layer.getClassName()}.`);\n      }\n      this.inputNames.push(layer.name);\n      this.feedInputShapes.push(layer.batchInputShape);\n\n      this.feedInputNames.push(layer.name);\n    }\n    for (const layer of this.outputLayers) {\n      this.outputNames.push(layer.name);\n    }\n\n    this.internalInputShapes = this.inputs.map(x => x.shape);\n    this.internalOutputShapes = this.outputs.map(x => x.shape);\n\n    /*\n      Container_nodes: set of nodes included in the graph (not all nodes\n      included in the layers are relevant to the current graph).\n    */\n    // ids of all nodes relevant to the Container:\n    const nodesDepths: {[nodeID: string]: number} = {};\n    // To recover nodes from their ID.\n    const nodeIDToNode: {[nodeID: string]: Node} = {};\n    const layersDepths: {[layerID: string]: number} = {};\n    // To layers from their ID.\n    const layerIDToLayer: {[layerID: string]: Layer} = {};\n    const layerIndices: {[layerID: string]: number} = {};\n    const nodesInDecreasingDepth: Node[] = [];\n\n    /**\n     * Builds a map of the graph of layers.\n     *\n     * This recursively updates the map `layerIndices`,\n     * the list `nodesInDecreasingDepth` and the set `containerNodes`.\n     *\n     * @param tensor Some tensor in a graph.\n     * @param finishedNodes Set of nodes whose subgraphs have been traversed\n     *         completely. Useful to prevent duplicated work.\n     * @param nodesInProgress Set of nodes that are currently active on the\n     *         recursion stack. Useful to detect cycles.\n     * @param layer Layer from which `tensor` comes from. If not provided,\n     *   will be obtained from tensor.sourceLayer.\n     * @param nodeIndex Node index from which `tensor` comes from.\n     * @param tensorIndex TensorIndex from which `tensor` comes from.\n     *\n     * @exception RuntimeError if a cycle is detected.\n     */\n    const buildMapOfGraph =\n        (tensor: SymbolicTensor, finishedNodes: Node[], nodesInProgress: Node[],\n         layer?: Layer, nodeIndex?: number, tensorIndex?: number) => {\n          if (layer == null || nodeIndex == null || tensorIndex == null) {\n            layer = tensor.sourceLayer;\n            nodeIndex = tensor.nodeIndex;\n            tensorIndex = tensor.tensorIndex;\n          }\n          const node = layer.inboundNodes[nodeIndex];\n\n          // Prevent cycles.\n          if (nodesInProgress.indexOf(node) !== -1) {\n            throw new RuntimeError(\n                `The tensor ${tensor.name} at layer \"${layer.name}\" ` +\n                'is part of a cycle.');\n          }\n\n          // Don't repeat work for shared subgraphs\n          if (finishedNodes.indexOf(node) !== -1) {\n            return;\n          }\n\n          // Update containerNodes.\n          this.containerNodes.add(Container.nodeKey(layer, nodeIndex));\n\n          // Store the traversal order for layer sorting.\n          if (!(layer.id in layerIndices)) {\n            layerIndices[layer.id] = Object.keys(layerIndices).length;\n          }\n\n          if (nodesInProgress.indexOf(node) === -1) {\n            nodesInProgress.push(node);\n          }\n\n          // Propagate to all previous tensors connected to this node.\n          const numInboundLayers = node.inboundLayers.length;\n          for (let i = 0; i < numInboundLayers; i++) {\n            const x = node.inputTensors[i];\n            const layer = node.inboundLayers[i];\n            const nodeIndex = node.nodeIndices[i];\n            const tensorIndex = node.tensorIndices[i];\n            buildMapOfGraph(\n                x, finishedNodes, nodesInProgress, layer, nodeIndex,\n                tensorIndex);\n          }\n          finishedNodes.push(node);\n          while (nodesInProgress.indexOf(node) >= 0) {\n            nodesInProgress.splice(nodesInProgress.indexOf(node), 1);\n          }\n          nodesInDecreasingDepth.push(node);\n        };\n\n    const finishedNodes: Node[] = [];\n    const nodesInProgress: Node[] = [];\n    for (const x of this.outputs) {\n      buildMapOfGraph(x, finishedNodes, nodesInProgress);\n    }\n\n    const reversedNodesInDecreasingDepth =\n        nodesInDecreasingDepth.slice().reverse();\n    for (const node of reversedNodesInDecreasingDepth) {\n      nodeIDToNode[node.id] = node;\n      // If the depth is not set, the node has no outbound nodes (depth 0).\n      if (!(node.id in nodesDepths)) {\n        nodesDepths[node.id] = 0;\n      }\n      let depth = nodesDepths[node.id];\n\n      // Update the depth of the corresponding layer\n      const previousDepth =\n          (layersDepths[node.outboundLayer.id] == null ?\n               0 :\n               layersDepths[node.outboundLayer.id]);\n\n      /*\n        If we've seen this layer before at a higher depth, we should use that\n        depth instead of the node depth.  This is necessary for shared layers\n        that have inputs at different depth levels in the graph.\n      */\n      depth = Math.max(depth, previousDepth);\n      layersDepths[node.outboundLayer.id] = depth;\n      layerIDToLayer[node.outboundLayer.id] = node.outboundLayer;\n      nodesDepths[node.id] = depth;\n\n      // Update the depth of inbound nodes.\n      for (let i = 0; i < node.inboundLayers.length; i++) {\n        const inboundLayer = node.inboundLayers[i];\n        const nodeIndex = node.nodeIndices[i];\n        const inboundNode = inboundLayer.inboundNodes[nodeIndex];\n        const previousDepth =\n            (nodesDepths[inboundNode.id] == null ? 0 :\n                                                   nodesDepths[inboundNode.id]);\n        nodesDepths[inboundNode.id] = Math.max(depth + 1, previousDepth);\n        nodeIDToNode[inboundNode.id] = inboundNode;\n      }\n    }\n\n    // Build a dict {depth: list of nodes with this depth}\n    const nodesByDepth: {[depth: string]: Node[]} = {};\n    for (const nodeID in nodesDepths) {\n      const depth = nodesDepths[nodeID];\n      if (!(depth in nodesByDepth)) {\n        nodesByDepth[depth] = [];\n      }\n      nodesByDepth[depth].push(nodeIDToNode[nodeID]);\n    }\n\n    // Build a dict {depth: list of layers with this depth}\n    const layersByDepth: {[depth: string]: Layer[]} = {};\n    for (const layerID in layersDepths) {\n      const depth = layersDepths[layerID];\n      if (!(depth in layersByDepth)) {\n        layersByDepth[depth] = [];\n      }\n      layersByDepth[depth].push(layerIDToLayer[layerID]);\n    }\n\n    // Get sorted list of layer depths.\n    let depthKeys = Object.keys(layersByDepth)\n                        .map(x => parseInt(x, 10))\n                        .sort(generic_utils.reverseNumberCompare);\n\n    // Set this.layers and this.layersByDepth.\n    this.layers = [];\n    for (const depth of depthKeys) {\n      const layersForDepth = layersByDepth[depth];\n      // Container.layers needs to have a deterministic order:\n      // here we order them by traversal order.\n      layersForDepth.sort((a, b) => {\n        const aIndex = layerIndices[a.id];\n        const bIndex = layerIndices[b.id];\n        if (aIndex < bIndex) {\n          return -1;\n        }\n        if (aIndex > bIndex) {\n          return 1;\n        }\n        return 0;\n      });\n      for (const layer of layersForDepth) {\n        if (layer instanceof Container) {\n          this.internalContainerRefs.push(layer);\n        }\n        this.layers.push(layer);\n      }\n    }\n    this.layersByDepth = layersByDepth;\n\n    // Get sorted list of node depths;\n    depthKeys = Object.keys(nodesByDepth)\n                    .map(x => parseInt(x, 10))\n                    .sort(generic_utils.reverseNumberCompare);\n\n    // Check that all tensors required are computable.\n    // computable_tensors: all tensors in the graph\n    // that can be computed from the inputs provided.\n    const computableTensors = this.inputs.slice();\n\n    // To provide a better error msg.\n    const layersWithCompleteInput: string[] = [];\n    for (const depth of depthKeys) {\n      for (const node of nodesByDepth[depth]) {\n        const layer = node.outboundLayer;\n        if (layer != null) {\n          for (const x of node.inputTensors) {\n            if (computableTensors.indexOf(x) === -1) {\n              throw new RuntimeError(\n                  `Graph disconnected: cannot obtain value for tensor ${x}` +\n                  ` at layer \"${layer.name}\". ` +\n                  'The following previous layers were accessed without ' +\n                  `issue: ${layersWithCompleteInput}`);\n            }\n          }\n          for (const x of node.outputTensors) {\n            computableTensors.push(x);\n          }\n          layersWithCompleteInput.push(layer.name);\n        }\n      }\n    }\n\n    // Set this.containerNodes and this.nodesByDepth.\n    this.nodesByDepth = nodesByDepth;\n\n    // Ensure name unicity, which will be crucial for serialization\n    // (since serialized nodes refer to layers by their name).\n    const allNames = this.layers.map(x => x.name);\n    for (const name of allNames) {\n      const numOccurrences = allNames.filter(x => x === name).length;\n      if (numOccurrences !== 1) {\n        throw new RuntimeError(\n            `The name \"${name}\" is used ${numOccurrences} times ` +\n            'in the model. All layer names should be unique. Layer names: ' +\n            JSON.stringify(allNames));\n      }\n    }\n\n    // Layer parameters.\n    // The new container starts with a single inbound node\n    // for its inputs, and no outbound nodes.\n    // Will be appended to by future calls to apply().\n    this.outboundNodes = [];\n    // Will be appended to below, and by future calls to apply().\n    this.inboundNodes = [];\n\n    // Create the node linking internal inputs to internal outputs.\n    // (This call has side effects.)\n    // tslint:disable-next-line:no-unused-expression\n    new Node({\n      outboundLayer: this,\n      inboundLayers: [],\n      nodeIndices: [],\n      tensorIndices: [],\n      inputTensors: this.inputs,\n      outputTensors: this.outputs,\n      inputMasks: this.inputs.map(x => null),\n      outputMasks: this.outputs.map(x => null),\n      inputShapes: this.inputs.map(x => x.shape),\n      outputShapes: this.outputs.map(x => x.shape)\n    });\n    this.built = true;\n    this._refCount = 1;  // The ref count of a container always start at 1.\n  }\n\n  protected assertNotDisposed() {\n    if (this._refCount === 0) {\n      throw new Error(`Container '${this.name}' is already disposed.`);\n    }\n  }\n\n  /**\n   * Attempt to dispose a LayersModel's weights.\n   *\n   * This method decrease the reference count of the LayersModel object by 1.\n   *\n   * A LayersModel is reference-counted. Its reference count is incremented by 1\n   * when it is first constructed and when it is used as a Layer of another\n   * LayersModel.\n   *\n   * If the reference count of a LayersModel becomes 0, the `dispose` method of\n   * all its constituent `Layer`s will be called.\n   *\n   * Note: If the reference count is greater than 0 after the decrement, the\n   * `dispose` method of its constituent `Layer`s will *not* be called.\n   *\n   * After a LayersModel is disposed, it cannot be used in calls such as\n   * 'predict`, `evaluate` or `fit` anymore.\n   *\n   * @returns A DisposeResult Object with the following fields:\n   *   - refCountAfterDispose: The reference count of the LayersModel after this\n   *     `dispose()` call.\n   *   - numDisposedVariables: Number of `tf.Variable`s (i.e., weights) disposed\n   *     during this `dispose()` call.\n   * @throws {Error} If the layer is not built yet, or if the LayersModel has\n   *   already been disposed.\n   */\n  dispose(): DisposeResult {\n    this.assertNotDisposed();\n    const result:\n        DisposeResult = {refCountAfterDispose: null, numDisposedVariables: 0};\n    if (--this._refCount === 0) {\n      for (const layer of this.layers) {\n        result.numDisposedVariables += layer.dispose().numDisposedVariables;\n      }\n\n      // Call dispose on each internally created container layer again to ensure\n      // their refCounts hit zero and their tensors are subsequently deleted.\n      for (const container of this.internalContainerRefs) {\n        result.numDisposedVariables += container.dispose().numDisposedVariables;\n      }\n    }\n    result.refCountAfterDispose = this._refCount;\n    return result;\n  }\n\n  get trainable() {\n    return this.trainable_;\n  }\n\n  set trainable(trainable: boolean) {\n    this.layers.forEach(layer => {\n      // tslint:disable-next-line:no-any\n      ((layer as any)._trainableWeights as LayerVariable[])\n          .forEach(w => w.trainable = trainable);\n    });\n    this.trainable_ = trainable;\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    // Porting Note: This check below is to prevent errors where the\n    //   _trainableWeights inherited from the parent class (Layer) gets\n    //   inadvertently used.\n    if (this._trainableWeights.length > 0) {\n      throw new ValueError(\n          'Container instance unexpectedly contains _trainableWeights.' +\n          'The trainable weights of a Container are a union of the ' +\n          'trainable weights of its consituent Layers. Its own ' +\n          '_trainableWeights must remain an empty Array.');\n    }\n\n    if (!this.trainable) {\n      return [];\n    }\n    let weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights = weights.concat(layer.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const layer of this.layers) {\n      weights.push(...layer.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const layer of this.layers) {\n        trainableWeights.push(...layer.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  get weights(): LayerVariable[] {\n    return this.trainableWeights.concat(this.nonTrainableWeights);\n  }\n\n  /**\n   * Loads all layer weights from a JSON object.\n   *\n   * Porting Note: HDF5 weight files cannot be directly loaded in JavaScript /\n   *   TypeScript. The utility script at `scripts/pykeras.py` offers means\n   *   to convert them into JSON strings compatible with this method.\n   * Porting Note: TensorFlow.js Layers supports only loading by name currently.\n   *\n   * @param weights A JSON mapping weight names to weight values as nested\n   *   arrays of numbers, or a `NamedTensorMap`, i.e., a JSON mapping weight\n   *   names to `tf.Tensor` objects.\n   * @param strict Require that the provided weights exactly match those\n   *   required by the container.  Default: `true`.  Passing `false` means that\n   *   extra weights and missing weights will be silently ignored.\n   */\n  loadWeights(weights: NamedTensorMap, strict = true) {\n    const nameToWeight: {[name: string]: LayerVariable} = {};\n    let totalWeightsCount = 0;\n    for (const layer of this.layers) {\n      for (const weight of layer.weights) {\n        if (nameToWeight[weight.originalName] != null) {\n          throw new ValueError(`Duplicate weight name: ${weight.originalName}`);\n        }\n        nameToWeight[weight.originalName] = weight;\n        totalWeightsCount++;\n      }\n    }\n\n    const weightValueTuples: Array<[LayerVariable, Tensor]> = [];\n    for (const name in weights) {\n      // TF 2.2.0 added cell name to the weight name in the format of\n      // layer_name/cell_name/weight_name, we need to remove\n      // the inner cell name.\n      let validatedName = name;\n      if (nameToWeight[name] == null) {\n        const tokens = name.split('/');\n        const shortenNameArray =\n            tokens.slice(0, -2).concat([tokens[tokens.length - 1]]);\n        validatedName = shortenNameArray.join('/');\n      }\n      if (nameToWeight[validatedName] != null) {\n        weightValueTuples.push([nameToWeight[validatedName], weights[name]]);\n      } else if (strict) {\n        throw new ValueError(\n            `Provided weight data has no target variable: ${name}`);\n      }\n      delete nameToWeight[validatedName];\n    }\n\n    if (strict) {\n      // Check that all weights are set.\n      const unsetNames: string[] = [];\n      for (const name in nameToWeight) {\n        unsetNames.push(name);\n      }\n      if (unsetNames.length > 0) {\n        throw new ValueError(\n            `${unsetNames.length} of ${\n                totalWeightsCount} weights are not set: ` +\n            `${unsetNames}`);\n      }\n    }\n\n    batchSetValue(weightValueTuples);\n  }\n\n  /**\n   * Util shared between different serialization methods.\n   * @returns LayersModel config with Keras version information added.\n   */\n  protected updatedConfig(): serialization.ConfigDict {\n    const theConfig = this.getConfig();\n    const modelConfig: serialization.ConfigDict = {};\n    modelConfig['className'] = this.getClassName();\n    modelConfig['config'] = theConfig;\n    modelConfig['kerasVersion'] = `tfjs-layers ${layersVersion}`;\n    // TODO(nielsene): Replace something like K.backend() once\n    // possible.\n    modelConfig['backend'] = 'TensorFlow.js';\n    return modelConfig;\n  }\n\n  /**\n   * Returns a JSON string containing the network configuration.\n   *\n   * To load a network from a JSON save file, use\n   * models.modelFromJSON(jsonString);\n   * @param extraJsonArgs Unused in tfjs-layers, maintained for PyKeras\n   * @param returnString Whether the return value should be stringified\n   *    (default: `true`).\n   * @returns a JSON string if `returnString` (default), or a JSON object if\n   *   `!returnString`.\n   */\n  // tslint:disable-next-line:no-any\n  toJSON(unused?: any, returnString = true): string|PyJsonDict {\n    const modelConfig = convertTsToPythonic(this.updatedConfig()) as PyJsonDict;\n    return returnString ? JSON.stringify(modelConfig) : modelConfig;\n  }\n\n  /**\n   * Call the model on new inputs.\n   *\n   * In this case `call` just reapplies all ops in the graph to the new inputs\n   * (e.g. build a new computational graph from the provided inputs).\n   *\n   * @param inputs A tensor or list of tensors.\n   * @param mask A mask or list of masks. A mask can be either a tensor or null\n   *   (no mask).\n   *\n   * @return A tensor if there is a single output, or a list of tensors if there\n   *   are more than one outputs.\n   */\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      const feedDict = new FeedDict();\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n      return execute(this.outputs, feedDict, kwargs) as Tensor | Tensor[];\n    });\n  }\n\n  /**\n   * Computes an output mask tensor.\n   *\n   * @param inputs Tensor or list of tensors.\n   * @param mask Tensor or list of tensors.\n   *\n   * @return null or a tensor (or list of tensors, one per output tensor of the\n   * layer).\n   */\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tidy(() => {\n      inputs = generic_utils.toList(inputs);\n      let masks: Tensor[];\n      if (mask == null) {\n        masks = generic_utils.pyListRepeat(null, inputs.length);\n      } else {\n        masks = generic_utils.toList(mask);\n      }\n      // TODO(michaelterry): Add support for mask caching.\n      return this.runInternalGraph(inputs, masks)[1];\n    });\n  }\n\n  /**\n   * Computes the output shape of the layer.\n   *\n   * Assumes that the layer will be built to match that input shape provided.\n   *\n   * @param inputShape A shape (tuple of integers) or a list of shape tuples\n   *   (one per output tensor of the layer). Shape tuples can include null for\n   *   free dimensions, instead of an integer.\n   */\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    const inputShapes = types_utils.normalizeShapeList(inputShape);\n    if (inputShapes.length !== this.inputLayers.length) {\n      throw new ValueError(\n          `Invalid inputShape argument ${inputShape}: ` +\n          `model has ${this.inputLayers.length} tensor inputs.`);\n    }\n\n    // TODO(michaelterry): Add caching\n    const layersToOutputShapes: {[shapeKey: string]: Shape} = {};\n    for (let i = 0; i < inputShapes.length; i++) {\n      const layer = this.inputLayers[i];\n      const inputShape = inputShapes[i];\n      // It's an input layer: computeOutputShape is identity,\n      // and there is only one node and one tensor output.\n      const shapeKey = layer.name + '_0_0';\n      layersToOutputShapes[shapeKey] = inputShape;\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    // Iterate over nodes, by depth level.\n    if (depthKeys.length > 1) {\n      for (const depth of depthKeys) {\n        const nodes = this.nodesByDepth[depth];\n        for (const node of nodes) {\n          // This is always a single layer, never a list.\n          const layer = node.outboundLayer;\n          if (this.inputLayers.map(x => x.id).indexOf(layer.id) !== -1) {\n            // We've already covered the input layers a few lines above.\n            continue;\n          }\n          // Potentially redundant list, same size of node.inputTensors.\n          const inputShapes: Shape[] = [];\n          for (let j = 0; j < node.inboundLayers.length; j++) {\n            const inboundLayer = node.inboundLayers[j];\n            const nodeIndex = node.nodeIndices[j];\n            const tensorIndex = node.tensorIndices[j];\n            const shapeKey = `${inboundLayer.name}_${nodeIndex}_${tensorIndex}`;\n            const inputShape = layersToOutputShapes[shapeKey];\n            inputShapes.push(inputShape);\n          }\n\n          const outputShape = layer.computeOutputShape(\n              generic_utils.singletonOrArray(inputShapes));\n\n          const outputShapes = types_utils.normalizeShapeList(outputShape);\n          const nodeIndex = layer.inboundNodes.indexOf(node);\n          for (let j = 0; j < outputShapes.length; j++) {\n            const shapeKey = `${layer.name}_${nodeIndex}_${j}`;\n            layersToOutputShapes[shapeKey] = outputShapes[j];\n          }\n        }\n      }\n    }\n\n    // Read final output shapes from layersToOutputShapes.\n    const outputShapes: Shape[] = [];\n    const outputShapeKeys: string[] = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      const shapeKey = `${layer.name}_${nodeIndex}_${tensorIndex}`;\n      outputShapeKeys.push(shapeKey);\n    }\n\n    for (let i = 0; i < outputShapeKeys.length; i++) {\n      const key = outputShapeKeys[i];\n      generic_utils.assert(key in layersToOutputShapes);\n      outputShapes.push(layersToOutputShapes[key]);\n    }\n\n    // TODO(michaelterry): Update cache\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  /**\n   * Computes output tensors for new inputs.\n   *\n   * Note:\n   *   - Expects `inputs` to be a list (potentially with 1 element).\n   *\n   * @param inputs List of tensors\n   * @param masks List of masks (tensors or null).\n   * @return Three lists: outputTensors, outputMasks, outputShapes\n   */\n  protected runInternalGraph(inputs: Tensor[], masks?: Tensor[]):\n      [Tensor[], Tensor[], Shape[]] {\n    if (masks == null) {\n      masks = generic_utils.pyListRepeat(null, inputs.length);\n    }\n\n    // Dictionary mapping reference tensors to tuples\n    // (computed tensor, compute mask)\n    // we assume a 1:1 mapping from tensor to mask\n    // TODO: raise exception when a `.computeMask()` call\n    // does not return a list the same size as `call`\n    const tensorMap: {[tensorID: string]: [Tensor, Tensor]} = {};\n    for (let i = 0; i < this.inputs.length; ++i) {\n      const x = this.inputs[i];\n      const y = inputs[i];\n      const mask = masks[i];\n      tensorMap[x.id] = [y, mask];\n    }\n\n    const depthKeys = Object.keys(this.nodesByDepth)\n                          .map(x => parseInt(x, 10))\n                          .sort(generic_utils.reverseNumberCompare);\n    for (const depth of depthKeys) {\n      const nodes = this.nodesByDepth[depth];\n      for (const node of nodes) {\n        // This is always a single layer, never a list.\n        const layer = node.outboundLayer;\n        const referenceInputTensors = node.inputTensors;\n        const referenceOutputTensors = node.outputTensors;\n\n        // If all previous input tensors are available in tensorMap,\n        // then call node.inboundLayer on them.\n        // List of tuples [input, mask]:\n        const computedData = new Array<[Tensor, Tensor]>();\n        for (const x of referenceInputTensors) {\n          if (x.id in tensorMap) {\n            computedData.push(tensorMap[x.id]);\n          }\n        }\n        if (computedData.length === referenceInputTensors.length) {\n          // TODO(michaelterry): Add K.name_scope here, if we need it.\n          let kwargs: Kwargs = {};\n          let computedTensors: Tensor[];\n          let computedMasks: Tensor[];\n          let outputTensors: Tensor[];\n          let outputMasks: Tensor[];\n          // call layer\n          if (node.callArgs != null) {\n            kwargs = node.callArgs;\n          }\n          if (computedData.length === 1) {\n            const [computedTensor, computedMask] = computedData[0];\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMask;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensor, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensor, computedMask));\n            computedTensors = [computedTensor];\n            computedMasks = [computedMask];\n          } else {\n            computedTensors = computedData.map(x => x[0]);\n            computedMasks = computedData.map(x => x[1]);\n            if (kwargs['mask'] == null) {\n              kwargs['mask'] = computedMasks;\n            }\n            outputTensors =\n                generic_utils.toList(layer.call(computedTensors, kwargs));\n            outputMasks = generic_utils.toList(\n                layer.computeMask(computedTensors, computedMasks));\n          }\n\n          if (layer.activityRegularizer) {\n            throw new NotImplementedError(\n                'LayersModel invocation with concrete Tensor value(s) in the ' +\n                'presence of activity regularizer(s) is not supported yet.');\n          }\n          // TODO(michaelterry): Add model updates and losses\n\n          // Update tensor map.\n          for (let i = 0; i < referenceOutputTensors.length; ++i) {\n            const x = referenceOutputTensors[i];\n            const y = outputTensors[i];\n            const mask = outputMasks[i];\n            tensorMap[x.id] = [y, mask];\n          }\n        }\n      }\n    }\n\n    const outputTensors: Tensor[] = [];\n    const outputMasks: Tensor[] = [];\n    const outputShapes: Shape[] = [];\n    for (const x of this.outputs) {\n      generic_utils.assert(\n          x.id in tensorMap, `Could not compute output ${x.name} : ${x.id}`);\n      const [tensor, mask] = tensorMap[x.id];\n      outputShapes.push(tensor.shape);\n      outputTensors.push(tensor);\n      outputMasks.push(mask);\n    }\n\n    // TODO(michaelterry): Add support for caches.\n    return [outputTensors, outputMasks, outputShapes];\n  }\n\n  /**\n   * Builds a map of internal node keys to node ordering.\n   * Used in serializaion a node orderings may change as unused nodes are\n   * dropped. Porting Note:  This helper method was pulled out of getConfig to\n   * improve readability.\n   * @param layers An array of Layers in the model.\n   * @returns Map of Node Keys to index order within the layer.\n   */\n  private buildNodeConversionMap(layers: Layer[]): {[nodeKey: string]: number} {\n    const nodeConversionMap: {[nodeKey: string]: number} = {};\n    let keptNodes: number;\n    for (const layer of this.layers) {\n      keptNodes = layer instanceof Container ? 1 : 0;\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        if (this.containerNodes.has(nodeKey)) {\n          // i.e. we mark it to be saved\n          nodeConversionMap[nodeKey] = keptNodes;\n          keptNodes += 1;\n        }\n      }\n    }\n    return nodeConversionMap;\n  }\n\n  /**\n   * Retrieves a layer based on either its name (unique) or index.\n   *\n   * Indices are based on order of horizontal graph traversal (bottom-up).\n   *\n   * If both `name` and `index` are specified, `index` takes precedence.\n   *\n   * @param name Name of layer.\n   * @param index Index of layer.\n   * @returns A Layer instance.\n   * @throws ValueError: In case of invalid layer name or index.\n   *\n   * @doc {\n   *    heading: 'Layers',\n   *    subheading: 'Classes',\n   *    namespace: 'layers',\n   *    subclasses: ['LayersModel']\n   * }\n   */\n  getLayer(name?: string, index?: number): Layer {\n    if (index != null) {\n      if (this.layers.length <= index) {\n        throw new ValueError(\n            `Was asked to retrieve layer at index ${index}, but model only ` +\n            `has ${this.layers.length} layer(s).`);\n      } else {\n        return this.layers[index];\n      }\n    } else {\n      if (name == null) {\n        throw new ValueError('Provide either a layer name or layer index');\n      }\n    }\n\n    for (const layer of this.layers) {\n      if (layer.name === name) {\n        return layer;\n      }\n    }\n    throw new ValueError(`No such layer: ${name}`);\n  }\n\n  /**\n   * Retrieves the Container's current loss values.\n   *\n   * Used for regularizers during training.\n   */\n  calculateLosses(): Scalar[] {\n    // Porting Node: This is an augmentation to Container.loss in PyKeras.\n    //   In PyKeras, Container.loss returns symbolic tensors. Here a concrete\n    //   Tensor (specifically Scalar) values are returned. This is due to the\n    //   imperative backend.\n    return tidy(() => {\n      const losses: Scalar[] = [];\n      for (const layer of this.layers) {\n        for (let nodeIndex = 0; nodeIndex < layer.inboundNodes.length;\n             ++nodeIndex) {\n          const nodeKey = Container.nodeKey(layer, nodeIndex);\n          if (this.containerNodes.has(nodeKey)) {\n            losses.push(...layer.calculateLosses());\n          }\n        }\n      }\n      // TODO(cais): Add any unconditional model-level losses?\n      return losses;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {name: this.name};\n\n    // Build a map from layer unique name (self._node_key)\n    // to the index of the nodes that are saved in the config.\n    // Only nodes in container_nodes are saved.\n    const nodeConversionMap: {[nodeKey: string]: number} =\n        this.buildNodeConversionMap(this.layers);\n\n    // Serialize and save the layers in layerConfigs\n    const layerConfigs = [];\n    for (const layer of this.layers) {\n      const layerClassName = layer.getClassName();\n      const layerConfig = layer.getConfig();\n      const filteredInboundNodes = [];\n      for (let originalNodeIndex = 0;\n           originalNodeIndex < layer.inboundNodes.length; originalNodeIndex++) {\n        const node = layer.inboundNodes[originalNodeIndex];\n        const nodeKey = Container.nodeKey(layer, originalNodeIndex);\n        let kwargs = {};\n        if (this.containerNodes.has(nodeKey)) {\n          // The node is relevant to the model:\n          // add to filteredInboundNodes.\n          if (node.callArgs) {\n            try {\n              JSON.stringify(node.callArgs);\n              kwargs = node.callArgs;\n            } catch (err) {\n              console.warn(\n                  `Layer ${layer.name} was passed ` +\n                  `non-serializable keyword arguments: ` +\n                  `${node.callArgs}. They will not be included ` +\n                  `in the serialized model (and thus will be ` +\n                  `missing at deserialization time).`);\n              kwargs = {};\n            }\n          }\n          if (node.inboundLayers.length > 0) {\n            const nodeData = [];\n            for (let i = 0; i < node.inboundLayers.length; i++) {\n              const inboundLayer = node.inboundLayers[i];\n              const nodeIndex = node.nodeIndices[i];\n              const tensorIndex = node.tensorIndices[i];\n              const nodeKey = Container.nodeKey(inboundLayer, nodeIndex);\n              let newNodeIndex = nodeConversionMap[nodeKey];\n              if (newNodeIndex == null) {\n                newNodeIndex = 0;\n              }\n              nodeData.push(\n                  [inboundLayer.name, newNodeIndex, tensorIndex, kwargs]);\n            }\n            filteredInboundNodes.push(nodeData);\n          }\n        }\n      }\n      const dict: serialization.ConfigDict = {};\n      dict['name'] = layer.name;\n      dict['className'] = layerClassName;\n      dict['config'] = layerConfig;\n      dict['inboundNodes'] = filteredInboundNodes;\n      layerConfigs.push(dict);\n    }\n    config['layers'] = layerConfigs;\n    // Gather info about inputs and outputs\n    const modelInputs = [];\n    for (let i = 0; i < this.inputLayers.length; i++) {\n      const layer = this.inputLayers[i];\n      const nodeIndex = this.inputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.inputLayersTensorIndices[i];\n      modelInputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['inputLayers'] = modelInputs;\n\n    const modelOutputs = [];\n    for (let i = 0; i < this.outputLayers.length; i++) {\n      const layer = this.outputLayers[i];\n      const nodeIndex = this.outputLayersNodeIndices[i];\n\n      const nodeKey = Container.nodeKey(layer, nodeIndex);\n      if (!this.containerNodes.has(nodeKey)) {\n        continue;\n      }\n      let newNodeIndex = nodeConversionMap[nodeKey];\n      if (newNodeIndex === null || newNodeIndex === undefined) {\n        newNodeIndex = 0;\n      }\n      const tensorIndex = this.outputLayersTensorIndices[i];\n      modelOutputs.push([layer.name, newNodeIndex, tensorIndex]);\n    }\n    config['outputLayers'] = modelOutputs;\n    return config;\n  }\n\n  /**\n   * Instantiates a LayersModel from its config (output of `get_config()`).\n   * @param cls the class to create\n   * @param config LayersModel config dictionary.\n   * @param customObjects An optional dictionary of custom objects.\n   * @param fastWeightInit Optional flag to use fast weight initialization\n   *   during deserialization. This is applicable to cases in which\n   *   the initialization will be immediately overwritten by loaded weight\n   *   values. Default: `false`.\n   * @returns A LayersModel instance.\n   * @throws ValueError: In case of improperly formatted config dict.\n   */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    // Layer instances created during\n    // the graph reconstruction process\n    const createdLayers: {[layerName: string]: Layer} = {};\n\n    // Dictionary mapping layer instances to\n    // node data that specifies a layer call.\n    // It acts as a queue that maintains any unprocessed\n    // layer call until it becomes possible to process it\n    // (i.e. until the input tensors to the call all exist).\n    const unprocessedNodes: {[layer: string]: TensorKeyWithArgsArray[][]} = {};\n    function addUnprocessedNode(\n        layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      if (!(layer.name in unprocessedNodes)) {\n        unprocessedNodes[layer.name] = [nodeData];\n      } else {\n        unprocessedNodes[layer.name].push(nodeData);\n      }\n    }\n\n    function processNode(layer: Layer, nodeData: TensorKeyWithArgsArray[]) {\n      const inputTensors: SymbolicTensor[] = [];\n      let kwargs;\n      for (const inputData of nodeData) {\n        const inboundLayerName = inputData[0];\n        const inboundNodeIndex = inputData[1];\n        const inboundTensorIndex = inputData[2];\n\n        kwargs = inputData[3] == null ?\n            {} :\n            inputData[3] as serialization.ConfigDict;\n        if (!(inboundLayerName in createdLayers)) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundLayer = createdLayers[inboundLayerName];\n        if (inboundLayer.inboundNodes.length <= inboundNodeIndex) {\n          addUnprocessedNode(layer, nodeData);\n          return;\n        }\n        const inboundNode = inboundLayer.inboundNodes[inboundNodeIndex];\n        inputTensors.push(inboundNode.outputTensors[inboundTensorIndex]);\n      }\n      // Call layer on its inputs, thus creating the node\n      // and building the layer if needed.\n      // Note: This has Eager vs Graph Implications.\n      if (inputTensors.length > 0) {\n        layer.apply(\n            generic_utils.singletonOrArray(inputTensors),\n            kwargs);  // was ** kwargs\n      }\n    }\n\n    /**\n     * Deserialize a layer, then call it on appropriate inputs.\n     * @param layerData: layer config dict.\n     * @throws ValueError: In case of improperly formatted `layer_data`\n     * dict.\n     */\n    function processLayer(layerData: serialization.ConfigDict|null) {\n      const layerName = layerData['name'] as string;\n      // Instantiate layer.\n      const layer =\n          deserializeLayer(\n              layerData,\n              config['customObjects'] != null ?\n                  config['customObjects'] as serialization.ConfigDict :\n                  {}) as Layer;\n      layer.setFastWeightInitDuringBuild(fastWeightInit);\n      createdLayers[layerName] = layer;\n      // Gather layer inputs.\n      const inboundNodesData =\n          layerData['inboundNodes'] as TensorKeyWithArgsArray[][];\n      inboundNodesData.forEach(nodeData => {\n        if (!(nodeData instanceof Array)) {\n          throw new ValueError(\n              `Corrupted configuration, expected array for nodeData: ${\n                  nodeData}`);\n        }\n        // We don't process nodes (i.e. make layer calls)\n        // on the fly because the inbound node may not yet exist,\n        // in case of layer shared at different topological depths\n        // (e.g.a model such as A(B(A(B(x)))))\n        addUnprocessedNode(layer, nodeData);\n      });\n    }\n\n    // First, we create all layers and enqueue nodes to be processed.\n    const name = config['name'];\n    const layersFromConfig = config['layers'] as serialization.ConfigDict[];\n    for (const layerData of layersFromConfig) {\n      processLayer(layerData);\n    }\n\n    // Then we process nodes in order of layer depth.\n    // Nodes that cannot yet be processed(if the inbound node\n    // does not yet exist) are re - enqueued, and the process\n    // is repeated until all nodes are processed.\n    while (!generic_utils.isObjectEmpty(unprocessedNodes)) {\n      for (const layerData of layersFromConfig) {\n        const layer = createdLayers[layerData['name'] as string];\n        if (layer.name in unprocessedNodes) {\n          const currentUnprocessedNodesForLayer = unprocessedNodes[layer.name];\n          delete unprocessedNodes[layer.name];\n          for (const nodeData of currentUnprocessedNodesForLayer) {\n            processNode(layer, nodeData);\n          }\n        }\n      }\n    }\n\n    const inputTensors: SymbolicTensor[] = [];\n    const outputTensors: SymbolicTensor[] = [];\n    const inputLayersFromConfig =\n        config['inputLayers'] as serialization.ConfigDict[];\n    for (const layerData of inputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      inputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    const outputLayersFromConfig =\n        config['outputLayers'] as serialization.ConfigDict[];\n    for (const layerData of outputLayersFromConfig) {\n      const layerName = layerData[0] as string;\n      const nodeIndex = layerData[1] as number;\n      const tensorIndex = layerData[2] as number;\n      generic_utils.assert(layerName in createdLayers);\n      const layer = createdLayers[layerName];\n      const layerOutputTensors = layer.inboundNodes[nodeIndex].outputTensors;\n      outputTensors.push(layerOutputTensors[tensorIndex]);\n    }\n    return new cls({inputs: inputTensors, outputs: outputTensors, name});\n  }\n\n  /**\n   * Determine whether the container is stateful.\n   *\n   * Porting Note: this is the equivalent of the stateful @property of\n   *   the Container class in PyKeras.\n   */\n  get stateful(): boolean {\n    // Porting Note: This check is to prevent inadvertent setting of the\n    //   _stateful property of the Container instance.\n    if (this._stateful) {\n      throw new ValueError(\n          'Container instance unexpectedly has _stateful = true. The ' +\n          'statefulness of a Container is determined by the Layers it ' +\n          'contains. Its _stateful property must remain the default false.');\n    }\n    for (const layer of this.layers) {\n      if (layer.stateful) {\n        return true;\n      }\n    }\n    return false;\n  }\n\n  /**\n   * Reset the state of all stateful constituent layers (if any).\n   *\n   * Examples of stateful layers include RNN layers whose `stateful` property\n   * is set as `true`.\n   */\n  resetStates() {\n    tidy(() => {\n      this.layers.forEach(layer => {\n        // tslint:disable:no-any\n        if (layer.stateful) {\n          layer.resetStates();\n        }\n        // tslint:enable:no-any\n      });\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {dispose, mul, Tensor, Tensor1D, tensor1d, tidy} from '@tensorflow/tfjs-core';\n\n/**\n * For multi-class classification problems, this object is designed to store a\n * mapping from class index to the \"weight\" of the class, where higher weighted\n * classes have larger impact on loss, accuracy, and other metrics.\n *\n * This is useful for cases in which you want the model to \"pay more attention\"\n * to examples from an under-represented class, e.g., in unbalanced datasets.\n */\nexport type ClassWeight = {\n  [classIndex: number]: number\n};\n\n/**\n * Class weighting for a model with multiple outputs.\n *\n * This object maps each output name to a class-weighting object.\n */\nexport type ClassWeightMap = {\n  [outputName: string]: ClassWeight\n};\n\nfunction standardizeSampleOrClassWeights(\n    xWeight: ClassWeight|ClassWeight[]|ClassWeightMap, outputNames: string[],\n    weightType: 'sampleWeight'|'classWeight'): ClassWeight[] {\n  const numOutputs = outputNames.length;\n  if (xWeight == null || (Array.isArray(xWeight) && xWeight.length === 0)) {\n    return outputNames.map(name => null);\n  }\n  if (numOutputs === 1) {\n    if (Array.isArray(xWeight) && xWeight.length === 1) {\n      return xWeight;\n    } else if (typeof xWeight === 'object' && outputNames[0] in xWeight) {\n      return [(xWeight as ClassWeightMap)[outputNames[0]]];\n    } else {\n      return [xWeight as ClassWeight];\n    }\n  }\n  if (Array.isArray(xWeight)) {\n    if (xWeight.length !== numOutputs) {\n      throw new Error(\n          `Provided ${weightType} is an array of ${xWeight.length} ` +\n          `element(s), but the model has ${numOutputs} outputs. ` +\n          `Make sure a set of weights is provided for each model output.`);\n    }\n    return xWeight;\n  } else if (\n      typeof xWeight === 'object' && Object.keys(xWeight).length > 0 &&\n      typeof (xWeight as ClassWeightMap)[Object.keys(xWeight)[0]] ===\n          'object') {\n    const output: ClassWeight[] = [];\n    outputNames.forEach(outputName => {\n      if (outputName in xWeight) {\n        output.push((xWeight as ClassWeightMap)[outputName]);\n      } else {\n        output.push(null);\n      }\n    });\n    return output;\n  } else {\n    throw new Error(\n        `The model has multiple (${numOutputs}) outputs, ` +\n        `so ${weightType} must be either an array with ` +\n        `${numOutputs} elements or an object with ${outputNames} keys. ` +\n        `Provided ${weightType} not understood: ${JSON.stringify(xWeight)}`);\n  }\n}\n\n/**\n * Standardize class weighting objects.\n *\n * This function takes a single class-weighting object, an array of them,\n * or a map from output name to class-weighting object. It compares it to the\n * output name(s) of the model, base on which it outputs an array of\n * class-weighting objects of which the length matches the number of outputs.\n *\n * @param classWeight Input class-weighting object(s).\n * @param outputNames All output name(s) of the model.\n * @return An array of class-weighting objects. The length of the array matches\n *   the model's number of outputs.\n */\nexport function standardizeClassWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'classWeight');\n}\n\nexport function standardizeSampleWeights(\n    classWeight: ClassWeight|ClassWeight[]|ClassWeightMap,\n    outputNames: string[]): ClassWeight[] {\n  return standardizeSampleOrClassWeights(\n      classWeight, outputNames, 'sampleWeight');\n}\n\n/**\n * Standardize by-sample and/or by-class weights for training.\n *\n * Note that this function operates on one model output at a time. For a model\n * with multiple outputs, you must call this function multiple times.\n *\n * @param y The target tensor that the by-sample and/or by-class weight is for.\n *     The values of y are assumed to encode the classes, either directly\n *     as an integer index, or as one-hot encoding.\n * @param sampleWeight By-sample weights.\n * @param classWeight By-class weights: an object mapping class indices\n *     (integers) to a weight (float) to apply to the model's loss for the\n *     samples from this class during training. This can be useful to tell the\n *     model to \"pay more attention\" to samples from an under-represented class.\n * @param sampleWeightMode The mode for the sample weights.\n * @return A Promise of weight tensor, of which the size of the first dimension\n *     matches that of `y`.\n */\nexport async function standardizeWeights(\n    y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight,\n    sampleWeightMode?: 'temporal'): Promise<Tensor> {\n  if (sampleWeight != null || sampleWeightMode != null) {\n    // TODO(cais): Once 'temporal' mode is implemented, document it in the doc\n    // string.\n    throw new Error('Support sampleWeight is not implemented yet');\n  }\n\n  if (classWeight != null) {\n    // Apply class weights per sample.\n    const yClasses: Tensor1D = tidy(() => {\n      if (y.shape.length === 1) {\n        // Assume class indices.\n        return y.clone() as Tensor1D;\n      } else if (y.shape.length === 2) {\n        if (y.shape[1] > 1) {\n          // Assume one-hot encoding of classes.\n          const axis = 1;\n          return y.argMax(axis);\n        } else if (y.shape[1] === 1) {\n          // Class index.\n          return y.reshape([y.shape[0]]);\n        } else {\n          throw new Error(\n              `Encountered unexpected last-dimension size (${y.shape[1]}) ` +\n              `during handling of class weights. The size is expected to be ` +\n              `>= 1.`);\n        }\n      } else {\n        throw new Error(\n            `Unexpected rank of target (y) tensor (${y.rank}) during ` +\n            `handling of class weights. The rank is expected to be 1 or 2.`);\n      }\n    });\n\n    const yClassIndices = Array.from(await yClasses.data());\n    dispose(yClasses);\n    const classSampleWeight: number[] = [];\n    yClassIndices.forEach(classIndex => {\n      if (classWeight[classIndex] == null) {\n        throw new Error(\n            `classWeight must contain all classes in the training data. ` +\n            `The class ${classIndex} exists in the data but not in ` +\n            `classWeight`);\n      } else {\n        classSampleWeight.push(classWeight[classIndex]);\n      }\n    });\n\n    return tensor1d(classSampleWeight, 'float32');\n  } else {\n    return null;\n  }\n}\n\n/**\n * Apply per-sample weights on the loss values from a number of samples.\n *\n * @param losses Loss tensor of shape `[batchSize]`.\n * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.\n * @returns Tensor of the same shape as`losses`.\n */\nexport function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor) {\n  return mul(losses, sampleWeights);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using TensorFlow.js datasets.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {scalar} from '@tensorflow/tfjs-core';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {TensorOrArrayOrMap} from '../types';\nimport {singletonOrArray, toList} from '../utils/generic_utils';\n\nimport {Dataset, LazyIterator} from './dataset_stub';\nimport {ClassWeight, ClassWeightMap, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Interface for configuring model training based on a dataset object.\n */\nexport interface ModelFitDatasetArgs<T> {\n  /**\n   * (Optional) Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. It should\n   * typically be equal to the number of samples of your dataset divided by\n   * the batch size, so that `fitDataset`() call can utilize the entire dataset.\n   * If it is not provided, use `done` return value in `iterator.next()` as\n   * signal to finish an epoch.\n   */\n  batchesPerEpoch?: number;\n\n  /**\n   * Integer number of times to iterate over the training dataset.\n   */\n  epochs: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be any of the following:\n   *\n   *   - An array `[xVal, yVal]`, where the two values may be `tf.Tensor`,\n   *     an array of Tensors, or a map of string to Tensor.\n   *   - Similarly, an array ` [xVal, yVal, valSampleWeights]`\n   *     (not implemented yet).\n   *   - a `Dataset` object with elements of the form `{xs: xVal, ys: yVal}`,\n   *     where `xs` and `ys` are the feature and label tensors, respectively.\n   *\n   * If `validationData` is an Array of Tensor objects, each `tf.Tensor` will be\n   * sliced into batches during validation, using the parameter\n   * `validationBatchSize` (which defaults to 32). The entirety of the\n   * `tf.Tensor` objects will be used in the validation.\n   *\n   * If `validationData` is a dataset object, and the `validationBatches`\n   * parameter is specified, the validation will use `validationBatches` batches\n   * drawn from the dataset object. If `validationBatches` parameter is not\n   * specified, the validation will stop when the dataset is exhausted.\n   *\n   * The model will not be trained on this data.\n   */\n  validationData?: [\n    TensorOrArrayOrMap, TensorOrArrayOrMap\n  ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|Dataset<T>;\n\n  /**\n   * Optional batch size for validation.\n   *\n   * Used only if `validationData` is an array of `tf.Tensor` objects, i.e., not\n   * a dataset object.\n   *\n   * If not specified, its value defaults to 32.\n   */\n  validationBatchSize?: number;\n\n  /**\n   * (Optional) Only relevant if `validationData` is specified and is a dataset\n   * object.\n   *\n   * Total number of batches of samples to draw from `validationData` for\n   * validation purpose before stopping at the end of every epoch. If not\n   * specified, `evaluateDataset` will use `iterator.next().done` as signal to\n   * stop validation.\n   */\n  validationBatches?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - a `number`: Will yield every `number` milliseconds.\n   *   - `'never'`: never yield. (But yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n}\n\nexport interface FitDatasetElement {\n  xs: TensorOrArrayOrMap;\n  ys: TensorOrArrayOrMap;\n}\n\n/**\n * Interface for configuring model evaluation based on a dataset object.\n */\nexport interface ModelEvaluateDatasetArgs {\n  /**\n   * Number of batches to draw from the dataset object before ending the\n   * evaluation.\n   */\n  batches?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n}\n\n// Default batch size used during tensor-based validation.\nconst DEFAULT_VALIDATION_BATCH_SIZE = 32;\n\n/**\n * Standardize the output of a dataset iterator for use by\n * LayersModel.fitDataset().\n *\n * @param model: A `tf.LayersModel` object.\n * @param iteratorOut The output of a dataset iterator. It is required to be\n *   an object of the form `{xs: TensorOrArrayOrMap, ys:\n * TensorOrArrayOrMap}`, where `TensorOrArrayOrMap` is a single `tf.Tensor`,\n * a `tf.Tensor[]`, or a flat map from string names to `tf.Tensor`s.\n * @returns A flat array of `tf.Tensor` objects: the input `tf.Tensor`s\n *   followed by the target `tf.Tensor`s.  When `tf.Tensor`s are provided\n *   as a map, the order in the resulting array is taken from the `inputNames`\n *   and `outputNames` of the model.\n */\nfunction standardizeDataIteratorOutput(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, iteratorOut: {}): {xs: tfc.Tensor[], ys: tfc.Tensor[]} {\n  let xs: TensorOrArrayOrMap;\n  let ys: TensorOrArrayOrMap;\n\n  const iteratorOutObj = iteratorOut as FitDatasetElement;\n  xs = iteratorOutObj['xs'];\n  ys = iteratorOutObj['ys'];\n  tfc.util.assert(\n      xs != null && ys != null,\n      () => 'A Dataset iterator for fitDataset() is expected to generate ' +\n          'objects of the form `{xs: xVal, ys: yVal}`, where the two ' +\n          'values may be `tf.Tensor`, an array of Tensors, or a map of ' +\n          'string to Tensor.  The provided Dataset instead generates ' +\n          `${iteratorOut}`);\n\n  const flattenedXs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('input', model.inputNames, xs);\n  const flattenedYs: tfc.Tensor[] =\n      flattenTensorOrArrayOrMap('output', model.outputNames, ys);\n\n  const batchSize: number = flattenedXs[0].shape[0];\n\n  tfc.util.assert(\n      flattenedXs.length === model.inputs.length,\n      () => `LayersModel has ${model.inputs.length} inputs, but the dataset ` +\n          `provides ${flattenedXs.length} inputs.  (Expected input keys: ` +\n          `${JSON.stringify(model.inputNames)})`);\n\n  tfc.util.assert(\n      flattenedYs.length === model.outputs.length,\n      () =>\n          `LayersModel has ${model.outputs.length} outputs, but the dataset ` +\n          `provides ${flattenedYs.length} outputs.  (Expected output keys: ` +\n          `${JSON.stringify(model.outputNames)})`);\n\n  for (let xIndex = 0; xIndex < flattenedXs.length; xIndex++) {\n    tfc.util.assert(\n        flattenedXs[xIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: input ` +\n            `${model.inputNames[xIndex]} has ${\n                  flattenedXs[xIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  for (let yIndex = 0; yIndex < flattenedYs.length; yIndex++) {\n    tfc.util.assert(\n        flattenedYs[yIndex].shape[0] === batchSize,\n        () => `Batch size mismatch: output ` +\n            `${model.outputNames[yIndex]} has ${\n                  flattenedYs[yIndex].shape[0]}; ` +\n            `expected  ${batchSize} based on input ${model.inputNames[0]}.`);\n  }\n\n  return {xs: flattenedXs, ys: flattenedYs};\n}\n\nfunction flattenTensorOrArrayOrMap(\n    inputOrOutput: string, names: string[], values: TensorOrArrayOrMap) {\n  if (values instanceof tfc.Tensor) {\n    return [values];\n  } else if (Array.isArray(values)) {\n    tfc.util.assert(\n        values.length === names.length,\n        () => `Received an array of ${values.length} Tensors, but expected ${\n            names.length} to match the ${inputOrOutput} keys ${names}.`);\n    return values;\n  } else {\n    const result: tfc.Tensor[] = [];\n    // Check that all the required keys are available.\n    for (const name of names) {\n      if (values[name] == null) {\n        throw new ValueError(\n            `The feature data generated by the dataset lacks the required ` +\n            `${inputOrOutput} key '${name}'.`);\n      }\n      result.push(values[name]);\n    }\n    return result;\n  }\n}\n\nfunction standardizeTensorValidationData<T>(\n    data:\n        [\n          tfc.Tensor|tfc.Tensor[], tfc.Tensor|tfc.Tensor[]\n        ]|[tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n           tfc.Tensor | tfc.Tensor[]]):\n    {xs: tfc.Tensor|tfc.Tensor[], ys: tfc.Tensor|tfc.Tensor[]} {\n  if (data.length === 3) {\n    throw new NotImplementedError(\n        'Validation with sample weights is not implemented yet.');\n  }\n  return {xs: data[0], ys: data[1]};\n}\n\nexport async function fitDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>,\n    args: ModelFitDatasetArgs<T>): Promise<History> {\n  const hasBatchesPerEpoch = args.batchesPerEpoch != null;\n  tfc.util.assert(\n      model.optimizer != null,\n      () => 'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileConfig).');\n\n  tfc.util.assert(\n      args != null,\n      () => `For fitDataset(), the 2nd argument (config) is required, ` +\n          `but it is not provided in this call.`);\n  tfc.util.assert(\n      args.epochs != null && args.epochs > 0 && Number.isInteger(args.epochs),\n      () => `For fitDataset(), config.epochs is expected to be a positive ` +\n          `integer, but got ${args.epochs}`);\n  tfc.util.assert(\n      !hasBatchesPerEpoch ||\n          (args.batchesPerEpoch > 0 && Number.isInteger(args.batchesPerEpoch)),\n      () => `For fitDataset(), config.batchesPerEpoch is expected to be a ` +\n          `positive integer if specified, but got ${args.batchesPerEpoch}`);\n  tfc.util.assert(\n      // tslint:disable-next-line:no-any\n      (args as any)['validationSplit'] == null,\n      () => '`validationSplit` is not supported by `fitDataset()`. ' +\n          'Use validationData instead.');\n\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n\n  try {\n    const doValidation = args.validationData != null;\n    let valXs: tfc.Tensor|tfc.Tensor[];\n    let valYs: tfc.Tensor|tfc.Tensor[];\n    if (doValidation) {\n      if (isDatasetObject(args.validationData)) {\n        tfc.util.assert(\n            args.validationBatches == null ||\n                (args.validationBatches > 0 &&\n                 Number.isInteger(args.validationBatches)),\n            () => `For fitDataset() with dataset-based validation, ` +\n                `config.validationBatches is expected not to be provided, ` +\n                `or to be a positive integer, ` +\n                `but got ${args.validationBatches}`);\n      } else {\n        const validationData = standardizeTensorValidationData(\n            args.validationData as\n                    [tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[]] |\n            [\n              tfc.Tensor | tfc.Tensor[], tfc.Tensor | tfc.Tensor[],\n              tfc.Tensor | tfc.Tensor[]\n            ]);\n        valXs = validationData.xs;\n        valYs = validationData.ys;\n      }\n    }\n\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let callbackMetrics: string[];\n    if (doValidation) {\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const verbose = args.verbose == null ? 1 : args.verbose;\n    const {callbackList, history} = configureCallbacks(\n        callbacks, verbose, args.epochs, null, null,\n        getStepsPerEpoch(dataset, args),\n        null,  // Batch size determined by the dataset itself.\n        doValidation, callbackMetrics);\n    callbackList.setModel(model);\n    model.history = history;\n\n    await callbackList.onTrainBegin();\n    model.stopTraining_ = false;\n    let epoch = args.initialEpoch == null ? 0 : args.initialEpoch;\n\n    let dataIterator = await dataset.iterator();\n    while (epoch < args.epochs) {\n      const epochLogs: UnresolvedLogs = {};\n      await callbackList.onEpochBegin(epoch);\n      let stepsDone = 0;\n      let batchIndex = 0;\n      if (!hasBatchesPerEpoch) {\n        dataIterator = await dataset.iterator();\n      }\n      while (hasBatchesPerEpoch ? stepsDone < args.batchesPerEpoch : true) {\n        const iteratorOut = await dataIterator.next();\n\n        // If `batchesPerEpoch` is specified, the dataset should not be\n        // exhausted until all epoches are done.\n        if (hasBatchesPerEpoch && iteratorOut.done) {\n          console.warn(\n              'You provided `batchesPerEpoch` as ' +\n              `${args.batchesPerEpoch}, ` +\n              'but your dataset iterator ran out of data after ' +\n              `${stepsDone} batches; ` +\n              'interrupting training. Make sure that your ' +\n              'dataset can generate at least `batchesPerEpoch * epochs` ' +\n              'batches (in this case, ' +\n              `${args.batchesPerEpoch * args.epochs} batches). ` +\n              'You may need to use the repeat() function when building ' +\n              'your dataset.');\n          break;\n        }\n\n        if (iteratorOut.value != null) {\n          const {xs, ys} =\n              standardizeDataIteratorOutput(model, iteratorOut.value);\n          const batchLogs: UnresolvedLogs = {};\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = xs[0].shape[0];\n\n          await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n          const sampleWeights: tfc.Tensor[] = [];\n          if (args.classWeight != null) {\n            const standardClassWeights =\n                standardizeClassWeights(args.classWeight, model.outputNames);\n            for (let i = 0; i < standardClassWeights.length; ++i) {\n              sampleWeights.push(await standardizeWeights(\n                  ys[i], null, standardClassWeights[i]));\n            }\n          }\n\n          // Train on batch.\n          const ins = xs.concat(ys).concat(sampleWeights);\n          const outs = trainFunction(ins);\n          tfc.dispose(ins);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n          }\n\n          await callbackList.onBatchEnd(batchIndex, batchLogs);\n          disposeTensorsInLogs(batchLogs);\n\n          batchIndex++;\n          stepsDone++;\n        }\n\n        if (hasBatchesPerEpoch ? stepsDone >= args.batchesPerEpoch :\n                                 iteratorOut.done) {\n          // Epoch finished. Perform validation.\n          if (doValidation) {\n            let valOuts: tfc.Scalar[];\n            if (isDatasetObject(args.validationData)) {\n              valOuts = toList(await model.evaluateDataset(\n                  args.validationData, {batches: args.validationBatches}));\n            } else {\n              valOuts = toList(model.evaluate(valXs, valYs, {\n                batchSize: args.validationBatchSize == null ?\n                    DEFAULT_VALIDATION_BATCH_SIZE :\n                    args.validationBatchSize,\n                verbose: 0\n              }));\n            }\n            for (let i = 0; i < model.metricsNames.length; ++i) {\n              epochLogs[`val_${model.metricsNames[i]}`] = valOuts[i];\n            }\n          }\n          // Call `break` to exit one epoch lopp after validation is done. If\n          // config.batchesPerEpoch is specified, an epoch while loop will\n          // stop when `stepsDone >= config.batchesPerEpoch`. When\n          // config.batchesPerEpoch is not provided, the following `break` is\n          // required to exit the while lopp after dataset is exhausted.\n          break;\n        }\n\n        if (model.stopTraining_) {\n          break;\n        }\n      }\n      await callbackList.onEpochEnd(epoch, epochLogs);\n      epoch++;\n      if (model.stopTraining_) {\n        break;\n      }\n    }\n    await callbackList.onTrainEnd();\n    await model.history.syncData();\n    return model.history;\n  } finally {\n    model.isTraining = false;\n  }\n}\n\n/** Helper function that determines number of steps (batches) per epoch. */\nfunction getStepsPerEpoch<T>(\n    dataset: Dataset<T>, args: ModelFitDatasetArgs<T>): number {\n  // Attempt to determine # of batches in an epoch.\n  let stepsPerEpoch: number = null;\n  if (args.batchesPerEpoch != null) {\n    stepsPerEpoch = args.batchesPerEpoch;\n  } else if (Number.isFinite(dataset.size)) {\n    stepsPerEpoch = dataset.size;\n  }\n  return stepsPerEpoch;\n}\n\n// Check if provided object is a Dataset object by checking its .iterator\n// element.\nfunction isDatasetObject<T>(\n    dataset:\n        [\n          TensorOrArrayOrMap, TensorOrArrayOrMap\n        ]|[TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]|\n    Dataset<T>): boolean {\n  return (typeof (dataset as Dataset<T>).iterator === 'function');\n}\n\n// Check if provided object is a LazyIterator object by checking it's .next\n// element.\nfunction isLazyIteratorObject<T>(iterator: Dataset<T>|\n                                 LazyIterator<T>): boolean {\n  return (typeof (iterator as LazyIterator<T>).next === 'function');\n}\n\nexport async function evaluateDataset<T>(\n    // Type `model` as `any` here to avoid circular dependency w/\n    // training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, dataset: Dataset<T>|LazyIterator<T>,\n    args: ModelEvaluateDatasetArgs): Promise<tfc.Scalar|tfc.Scalar[]> {\n  args = args || {};\n  const hasBatches = args.batches != null;\n  const f = model.testFunction;\n  let outs: tfc.Scalar[] = [];\n  if (args.verbose > 0) {\n    throw new NotImplementedError('Verbose mode is not implemented yet.');\n  }\n\n  tfc.util.assert(\n      !hasBatches || (args.batches > 0 && Number.isInteger(args.batches)),\n      () => 'Test loop expects `batches` to be a positive integer, but ' +\n          `received ${JSON.stringify(args.batches)}`);\n  const dataIterator = isLazyIteratorObject(dataset) ?\n      dataset as LazyIterator<T>:\n      await (dataset as Dataset<T>).iterator();\n  // Keeps track of number of examples used in this evaluation.\n  let numExamples = 0;\n  let batch = 0;\n\n  while (hasBatches ? batch < args.batches : true) {\n    const iteratorOut = await dataIterator.next();\n    outs = tfc.tidy(() => {\n      if (iteratorOut.value) {\n        // TODO(cais): Once real dataset is available, use\n        //   `map(x => standardizeDataIteratorOutput(model, x).map(f)`.\n        const {xs, ys} =\n            standardizeDataIteratorOutput(model, iteratorOut.value);\n        const xsAndYs = xs.concat(ys);\n        const batchOuts = tfc.tidy(() => f(xsAndYs));\n        tfc.dispose(xsAndYs);\n\n        if (batch === 0) {\n          for (let i = 0; i < batchOuts.length; ++i) {\n            outs.push(scalar(0));\n          }\n        }\n\n        const batchSize = xsAndYs[0].shape[0];\n        for (let i = 0; i < batchOuts.length; ++i) {\n          const batchOut = batchOuts[i];\n          const oldScalar = outs[i];\n          outs[i] =\n              tfc.tidy(() => tfc.add(outs[i], tfc.mul(batchSize, batchOut)));\n          if (batch > 0) {\n            tfc.dispose(oldScalar);\n          }\n        }\n        tfc.dispose(batchOuts);\n        numExamples += batchSize;\n\n        ++batch;\n      }\n      return outs;\n    });\n\n    if (iteratorOut.done) {\n      if (hasBatches) {\n        console.warn(\n            'Your dataset iterator ran out of data during evaluateDataset(). ' +\n            'Interrupting evalution. Make sure that your ' +\n            'dataset can generate at least `batches` ' +\n            `batches (in this case, ${args.batches} batches). ` +\n            'You may need to use the repeat() function when building ' +\n            'your dataset.');\n      }\n      break;\n    }\n  }\n\n  for (let i = 0; i < outs.length; ++i) {\n    const oldScalar = outs[i];\n    outs[i] = tfc.div(outs[i], numExamples);\n    tfc.dispose(oldScalar);\n  }\n\n  return singletonOrArray(outs);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Interfaces and methods for training models using tf.Tensor objects.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Scalar, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport {expandDims, gather, sliceAlongFirstAxis} from '../backend/tfjs_backend';\nimport {BaseCallback, configureCallbacks, CustomCallbackArgs, History, ModelLoggingVerbosity, standardizeCallbacks, YieldEveryOptions} from '../base_callbacks';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {disposeTensorsInLogs, UnresolvedLogs} from '../logs';\nimport {range} from '../utils/math_utils';\nimport {ClassWeight, ClassWeightMap} from './training_utils';\n\n/**\n * Interface configuration model training based on data as `tf.Tensor`s.\n */\nexport interface ModelFitArgs {\n  /**\n   * Number of samples per gradient update. If unspecified, it\n   * will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Integer number of times to iterate over the training data arrays.\n   */\n  epochs?: number;\n\n  /**\n   * Verbosity level.\n   *\n   * Expected to be 0, 1, or 2. Default: 1.\n   *\n   * 0 - No printed message during fit() call.\n   * 1 - In Node.js (tfjs-node), prints the progress bar, together with\n   *     real-time updates of loss and metric values and training speed.\n   *     In the browser: no action. This is the default.\n   * 2 - Not implemented yet.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * List of callbacks to be called during training.\n   * Can have one or more of the following callbacks:\n   *   - `onTrainBegin(logs)`: called when training starts.\n   *   - `onTrainEnd(logs)`: called when training ends.\n   *   - `onEpochBegin(epoch, logs)`: called at the start of every epoch.\n   *   - `onEpochEnd(epoch, logs)`: called at the end of every epoch.\n   *   - `onBatchBegin(batch, logs)`: called at the start of every batch.\n   *   - `onBatchEnd(batch, logs)`: called at the end of every batch.\n   *   - `onYield(epoch, batch, logs)`: called every `yieldEvery` milliseconds\n   *      with the current epoch, batch and logs. The logs are the same\n   *      as in `onBatchEnd()`. Note that `onYield` can skip batches or\n   *      epochs. See also docs for `yieldEvery` below.\n   */\n  callbacks?: BaseCallback[]|CustomCallbackArgs|CustomCallbackArgs[];\n\n  /**\n   * Float between 0 and 1: fraction of the training data\n   * to be used as validation data. The model will set apart this fraction of\n   * the training data, will not train on it, and will evaluate the loss and\n   * any model metrics on this data at the end of each epoch.\n   * The validation data is selected from the last samples in the `x` and `y`\n   * data provided, before shuffling.\n   */\n  validationSplit?: number;\n\n  /**\n   * Data on which to evaluate the loss and any model\n   * metrics at the end of each epoch. The model will not be trained on this\n   * data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal,\n   * valSampleWeights]. The model will not be trained on this data.\n   * `validationData` will override `validationSplit`.\n   */\n  validationData?: [\n    Tensor|Tensor[], Tensor|Tensor[]\n  ]|[Tensor | Tensor[], Tensor|Tensor[], Tensor|Tensor[]];\n\n  /**\n   * Whether to shuffle the training data before each epoch. Has\n   * no effect when `stepsPerEpoch` is not `null`.\n   */\n  shuffle?: boolean;\n\n  /**\n   * Optional object mapping class indices (integers) to\n   * a weight (float) to apply to the model's loss for the samples from this\n   * class during training. This can be useful to tell the model to \"pay more\n   * attention\" to samples from an under-represented class.\n   *\n   * If the model has multiple outputs, a class weight can be specified for\n   * each of the outputs by setting this field an array of weight object\n   * or a object that maps model output names (e.g., `model.outputNames[0]`)\n   * to weight objects.\n   */\n  classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap;\n\n  /**\n   * Optional array of the same length as x, containing\n   * weights to apply to the model's loss for each sample. In the case of\n   * temporal data, you can pass a 2D array with shape (samples,\n   * sequenceLength), to apply a different weight to every timestep of every\n   * sample. In this case you should make sure to specify\n   * sampleWeightMode=\"temporal\" in compile().\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * Epoch at which to start training (useful for resuming a previous training\n   * run). When this is used, `epochs` is the index of the \"final epoch\".\n   * The model is not trained for a number of iterations given by `epochs`,\n   * but merely until the epoch of index `epochs` is reached.\n   */\n  initialEpoch?: number;\n\n  /**\n   * Total number of steps (batches of samples) before\n   * declaring one epoch finished and starting the next epoch. When training\n   * with Input Tensors such as TensorFlow data tensors, the default `null` is\n   * equal to the number of unique samples in your dataset divided by the\n   * batch size, or 1 if that cannot be determined.\n   */\n  stepsPerEpoch?: number;\n\n  /**\n   * Only relevant if `stepsPerEpoch` is specified. Total number of steps\n   * (batches of samples) to validate before stopping.\n   */\n  validationSteps?: number;\n\n  /**\n   * Configures the frequency of yielding the main thread to other tasks.\n   *\n   * In the browser environment, yielding the main thread can improve the\n   * responsiveness of the page during training. In the Node.js environment,\n   * it can ensure tasks queued in the event loop can be handled in a timely\n   * manner.\n   *\n   * The value can be one of the following:\n   *   - `'auto'`: The yielding happens at a certain frame rate (currently set\n   *               at 125ms). This is the default.\n   *   - `'batch'`: yield every batch.\n   *   - `'epoch'`: yield every epoch.\n   *   - any `number`: yield every `number` milliseconds.\n   *   - `'never'`: never yield. (yielding can still happen through `await\n   *      nextFrame()` calls in custom callbacks.)\n   */\n  yieldEvery?: YieldEveryOptions;\n}\n\nexport function checkBatchSize(batchSize: number) {\n  tfc.util.assert(\n      batchSize > 0 && Number.isInteger(batchSize),\n      () => `batchSize is required to be a positive integer, but got ${\n          batchSize}`);\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by start and stop indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArraysByIndices()` together.\n *\n * @param arrays: the input.\n * @param start: the starting index (inclusive).\n * @param stop: the stopping index (exclusive).\n * @returns The result of the slicing. If `arrays` is an `Array` of\n *   `tf.Tensor`s, the slicing will be applied to all elements of the `Array`\n *   in the same way.\n */\nexport function sliceArrays(\n    arrays: Tensor|Tensor[], start: number, stop: number): Tensor|Tensor[] {\n  if (arrays == null) {\n    return [null];\n  } else if (Array.isArray(arrays)) {\n    return arrays.map(array => sliceAlongFirstAxis(array, start, stop - start));\n  } else {  // Tensor.\n    return sliceAlongFirstAxis(arrays, start, stop - start);\n  }\n}\n\n/**\n * Slice a Tensor or an Array of Tensors, by random-order indices.\n *\n * Porting Note: The `_slice_arrays` function in PyKeras is covered by this\n *   function and `sliceArrays()` together.\n *\n * @param arrays The input `tf.Tensor` or `Array` of `tf.Tensor`s to slice.\n *   If an `Array` of `tf.Tensor`s, all `tf.Tensor`s will be sliced in the\n *   same fashion.\n * @param indices The indices to use for slicing along the first (batch)\n *   dimension.\n * @returns Result(s) of the slicing.\n */\nexport function sliceArraysByIndices(\n    arrays: Tensor|Tensor[], indices: Tensor1D): Tensor|Tensor[] {\n  return tfc.tidy(() => {\n    if (arrays == null) {\n      return null;\n    } else if (Array.isArray(arrays)) {\n      return arrays.map(\n          array => (sliceArraysByIndices(array, indices) as Tensor));\n    } else {\n      // TODO(cais): indices should be a pre-constructed Tensor1D to avoid\n      //   tensor1d() calls.\n      return gather(\n          arrays, indices.dtype === 'int32' ? indices : indices.toInt());\n    }\n  });\n}\n\n/**\n * Returns a list of batch indices (tuples of indices).\n * @param size: Integer, total size of the data to slice into batches.\n * @param batchSize: Integer, batch size.\n * @returns An Array of [batchStart, batchEnd] tuples. batchStart is\n *   inclusive; batchEnd is exclusive. I.e., each batch consists of indices x\n *   that satisfy batchStart <= x < batchEnd.\n */\nexport function makeBatches(\n    size: number, batchSize: number): Array<[number, number]> {\n  const output: Array<[number, number]> = [];\n  let batchStart = 0;\n  let batchEnd: number = null;\n  while (batchStart < size) {\n    batchEnd = batchStart + batchSize;\n    if (batchEnd >= size) {\n      batchEnd = size;\n    }\n    output.push([batchStart, batchEnd]);\n    batchStart = batchEnd;\n  }\n  return output;\n}\n\n/**\n * Abstract fit function for `f(ins)`.\n * @param f A Function returning a list of tensors. For training, this\n *   function is expected to perform the updates to the variables.\n * @param ins List of tensors to be fed to `f`.\n * @param outLabels List of strings, display names of the outputs of `f`.\n * @param batchSize Integer batch size or `== null` if unknown. Default : 32.\n * @param epochs Number of times to iterate over the data. Default : 1.\n * @param verbose Verbosity mode: 0, 1, or 2. Default: 1.\n * @param callbacks List of callbacks to be called during training.\n * @param valF Function to call for validation.\n * @param valIns List of tensors to be fed to `valF`.\n * @param shuffle Whether to shuffle the data at the beginning of every\n * epoch. Default : true.\n * @param callbackMetrics List of strings, the display names of the metrics\n *   passed to the callbacks. They should be the concatenation of the\n *   display names of the outputs of `f` and the list of display names\n *   of the outputs of `valF`.\n * @param initialEpoch Epoch at which to start training (useful for\n *   resuming a previous training run). Default : 0.\n * @param stepsPerEpoch Total number of steps (batches on samples) before\n *   declaring one epoch finished and starting the next epoch. Ignored with\n *   the default value of `undefined` or `null`.\n * @param validationSteps Number of steps to run validation for (only if\n *   doing validation from data tensors). Not applicable for tfjs-layers.\n * @returns A `History` object.\n */\nasync function fitLoop(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, f: (data: Tensor[]) => Scalar[], ins: Tensor[],\n    outLabels?: string[], batchSize?: number, epochs?: number, verbose?: number,\n    callbacks?: BaseCallback[], valF?: (data: Tensor[]) => Scalar[],\n    valIns?: Tensor[], shuffle?: boolean|string, callbackMetrics?: string[],\n    initialEpoch?: number, stepsPerEpoch?: number,\n    validationSteps?: number): Promise<History> {\n  if (batchSize == null) {\n    batchSize = 32;\n  }\n  if (epochs == null) {\n    epochs = 1;\n  }\n  if (shuffle == null) {\n    shuffle = true;\n  }\n  if (initialEpoch == null) {\n    initialEpoch = 0;\n  }\n\n  // TODO(cais): Change const to let below when implementing validation.\n  let doValidation = false;\n  if (valF != null && valIns != null) {\n    doValidation = true;\n    // TODO(cais): verbose message.\n  }\n  if (validationSteps != null) {\n    doValidation = true;\n    if (stepsPerEpoch == null) {\n      throw new ValueError(\n          'Can only use `validationSteps` when doing step-wise training, ' +\n          'i.e., `stepsPerEpoch` must be set.');\n    }\n  }\n\n  const numTrainSamples =\n      model.checkNumSamples(ins, batchSize, stepsPerEpoch, 'steps_per_epoch');\n  let indexArray: number[];\n  if (numTrainSamples != null) {\n    indexArray = range(0, numTrainSamples);\n  }\n\n  if (verbose == null) {\n    verbose = 1;\n  }\n\n  const {callbackList, history} = configureCallbacks(\n      callbacks, verbose, epochs, initialEpoch, numTrainSamples, stepsPerEpoch,\n      batchSize, doValidation, callbackMetrics);\n  callbackList.setModel(model);\n  model.history = history;\n  await callbackList.onTrainBegin();\n  model.stopTraining_ = false;\n  // TODO(cais): Take care of callbacks.validation_data as in PyKeras.\n  // TODO(cais): Pre-convert feeds for performance as in PyKeras.\n\n  for (let epoch = initialEpoch; epoch < epochs; ++epoch) {\n    await callbackList.onEpochBegin(epoch);\n    const epochLogs: UnresolvedLogs = {};\n    if (stepsPerEpoch != null) {\n      throw new NotImplementedError(\n          'stepsPerEpoch mode is not implemented yet.');\n    } else {\n      if (shuffle === 'batch') {\n        throw new NotImplementedError('batch shuffling is not implemneted yet');\n      } else if (shuffle) {\n        util.shuffle(indexArray);\n      }\n      // Convert the potentially shuffled indices to Tensor1D, to avoid the\n      // cost of repeated creation of Array1Ds later on.\n      const epochIndexArray1D = tensor1d(indexArray);\n\n      const batches = makeBatches(numTrainSamples, batchSize);\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchLogs: UnresolvedLogs = {};\n        await callbackList.onBatchBegin(batchIndex, batchLogs);\n\n        tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds = sliceAlongFirstAxis(\n                               epochIndexArray1D, batchStart,\n                               batchEnd - batchStart) as Tensor1D;\n          batchLogs['batch'] = batchIndex;\n          batchLogs['size'] = batchEnd - batchStart;\n\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Tensor[];\n          const outs = f(insBatch);\n          for (let i = 0; i < outLabels.length; ++i) {\n            const label = outLabels[i];\n            const out = outs[i];\n            batchLogs[label] = out;\n            tfc.keep(out);\n            // TODO(cais): Use scope() to avoid ownership.\n          }\n\n          if (batchIndex === batches.length - 1) {  // Last batch.\n            if (doValidation) {\n              const valOuts = model.testLoop(valF, valIns, batchSize);\n              // Porting Notes: In tfjs-layers, valOuts is always an Array.\n              for (let i = 0; i < outLabels.length; ++i) {\n                const label = outLabels[i];\n                const out = valOuts[i];\n                tfc.keep(out);\n                // TODO(cais): Use scope() to avoid ownership.\n                epochLogs['val_' + label] = out;\n              }\n            }\n          }\n        });\n\n        await callbackList.onBatchEnd(batchIndex, batchLogs);\n        disposeTensorsInLogs(batchLogs);\n\n        if (model.stopTraining_) {\n          break;\n        }\n        // TODO(cais): return outs as list of Tensor.\n      }\n\n      epochIndexArray1D.dispose();\n    }\n    // TODO(cais): Run validation at the end of the epoch.\n    await callbackList.onEpochEnd(epoch, epochLogs);\n    if (model.stopTraining_) {\n      break;\n    }\n  }\n  await callbackList.onTrainEnd();\n\n  await model.history.syncData();\n  return model.history;\n}\n\nexport async function fitTensors(\n    // Type `model` as `any` here to avoid circular dependency w/ training.ts.\n    // tslint:disable-next-line:no-any\n    model: any, x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    args: ModelFitArgs = {}): Promise<History> {\n  if (model.isTraining) {\n    throw new Error(\n        'Cannot start training because another fit() call is ongoing.');\n  }\n  model.isTraining = true;\n  let inputs: Tensor[];\n  let targets: Tensor[];\n  let inputValX: Tensor|Tensor[];\n  let inputValY: Tensor|Tensor[];\n  let valX: Tensor|Tensor[];\n  let valY: Tensor|Tensor[];\n  let sampleWeights: Tensor[];\n  try {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // Validate user data.\n    // TODO(cais): Support sampleWeight.\n    const checkBatchAxis = false;\n    const standardizedOuts =\n        await model.standardizeUserData(\n            x, y, args.sampleWeight, args.classWeight, checkBatchAxis,\n            batchSize) as [Tensor[], Tensor[], Tensor[]];\n    inputs = standardizedOuts[0];\n    targets = standardizedOuts[1];\n    sampleWeights = standardizedOuts[2];\n\n    // Prepare validation data.\n    let doValidation = false;\n    let valIns: Tensor[];\n    if (args.validationData != null && args.validationData.length > 0) {\n      doValidation = true;\n      if (args.validationData.length === 2) {\n        // config.validationData consists of valX and valY.\n        inputValX = args.validationData[0];\n        inputValY = args.validationData[1];\n      } else if (args.validationData.length === 3) {\n        throw new NotImplementedError(\n            'validationData including sample weights is not supported yet.');\n      } else {\n        throw new ValueError(\n            `When passing validation data, it must contain 2 (valX, valY) ` +\n            `or 3 (valX, valY, valSampleWeight) items; ` +\n            `${args.validationData} is invalid.`);\n      }\n\n      const checkBatchAxis = true;\n      const valStandardized =\n          await model.standardizeUserData(\n              inputValX, inputValY, null, /** Unused sample weights. */\n              null,                       /** Unused class weights. */\n              checkBatchAxis, batchSize) as [Tensor[], Tensor[], Tensor[]];\n      valX = valStandardized[0];\n      valY = valStandardized[1];\n      valIns = valX.concat(valY);\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (\n        args.validationSplit != null && args.validationSplit > 0 &&\n        args.validationSplit < 1) {\n      doValidation = true;\n      // Porting Note: In tfjs-layers, inputs[0] is always a Tensor.\n      const splitAt =\n          Math.floor(inputs[0].shape[0] * (1 - args.validationSplit));\n      const originalBatchSize = inputs[0].shape[0];\n      valX = sliceArrays(inputs, splitAt, originalBatchSize) as Tensor[];\n      inputs = sliceArrays(inputs, 0, splitAt) as Tensor[];\n      valY = sliceArrays(targets, splitAt, originalBatchSize) as Tensor[];\n      targets = sliceArrays(targets, 0, splitAt) as Tensor[];\n      // TODO(cais): Once sampleWeights becomes available, slice it to get\n      //   valSampleWeights.\n      valIns = valX.concat(valY);\n\n      // TODO(cais): Add useLearningPhase data properly.\n    } else if (args.validationSteps != null) {\n      doValidation = true;\n      // TODO(cais): Add useLearningPhase.\n    }\n\n    const ins = inputs.concat(targets).concat(sampleWeights);\n\n    model.checkTrainableWeightsConsistency();\n\n    // TODO(cais): Handle use_learning_phase and learning_phase?\n\n    // Porting Note: Here we see a key deviation of tfjs-layers from\n    // Keras.\n    //  Due to the imperative nature of tfjs-layers' backend (tfjs-core),\n    //  we do not construct symbolic computation graphs to embody the\n    //  training process. Instead, we define a function that performs the\n    //  training action. In PyKeras, the data (inputs and targets) are fed\n    //  through graph placeholders. In tfjs-layers, the data are fed as\n    //  function arguments. Since the function are defined below in the\n    //  scope, we don't have equivalents of PyKeras's\n    //  `_make_train_funciton`.\n    const trainFunction = model.makeTrainFunction();\n    const outLabels = model.getDedupedMetricsNames() as string[];\n\n    let valFunction: (data: Tensor[]) => Scalar[];\n    let callbackMetrics: string[];\n    if (doValidation) {\n      model.makeTestFunction();\n      valFunction = model.testFunction;\n      callbackMetrics =\n          outLabels.slice().concat(outLabels.map(n => 'val_' + n));\n    } else {\n      valFunction = null;\n      valIns = [];\n      callbackMetrics = outLabels.slice();\n    }\n\n    const callbacks = standardizeCallbacks(args.callbacks, args.yieldEvery);\n    const out = await fitLoop(\n        model, trainFunction, ins, outLabels, batchSize, args.epochs,\n        args.verbose, callbacks, valFunction, valIns, args.shuffle,\n        callbackMetrics, args.initialEpoch, null, null);\n    return out;\n  } finally {\n    model.isTraining = false;\n    // Memory clean up.\n    disposeNewTensors(inputs, x);\n    disposeNewTensors(targets, y);\n    disposeNewTensors(valX as Tensor[], inputValX);\n    disposeNewTensors(valY as Tensor[], inputValY);\n    if (sampleWeights != null) {\n      tfc.dispose(sampleWeights);\n    }\n  }\n  // TODO(cais): Add value to outLabels.\n}\n\n/**\n * Ensure tensors all have a rank of at least 2.\n *\n * If a tensor has a rank of 1, it is dimension-expanded to rank 2.\n * If any tensor has a rank of 0 (i.e., is a scalar), an error will be thrown.\n */\nexport function ensureTensorsRank2OrHigher(tensors: Tensor|Tensor[]): Tensor[] {\n  const outs: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    tensors = [tensors];\n  }\n\n  // Make Tensors at least 2D.\n  for (let i = 0; i < tensors.length; ++i) {\n    const tensor = tensors[i];\n    if (tensor.rank === 1) {\n      outs.push(expandDims(tensor, 1));\n    } else if (tensor.rank === 0) {\n      throw new Error(\n          'Expected tensor to be at least 1D, but received a 0D tensor ' +\n          '(scalar).');\n    } else {\n      outs.push(tensor);\n    }\n  }\n  return outs;\n}\n\n/**\n * Compare a set of tensors with a reference (old) set, discard the ones\n * in the new set that are not present in the reference set.\n *\n * This method is used for memory clenaup during calls such as\n * LayersModel.fit().\n *\n * @param tensors New set which may contain Tensors not present in\n *   `refTensors`.\n * @param refTensors Reference Tensor set.\n */\n// TODO(cais, kangyizhang): Deduplicate with tfjs-data.\nexport function disposeNewTensors(\n    tensors: Tensor|Tensor[]|{[inputName: string]: Tensor},\n    refTensors: Tensor|Tensor[]|{[inputName: string]: Tensor}): void {\n  if (tensors == null) {\n    return;\n  }\n  const oldTensorIds: number[] = [];\n  if (refTensors instanceof Tensor) {\n    oldTensorIds.push(refTensors.id);\n  } else if (Array.isArray(refTensors)) {\n    refTensors.forEach(t => oldTensorIds.push(t.id));\n  } else if (refTensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in refTensors) {\n      const oldTensor = refTensors[name];\n      oldTensorIds.push(oldTensor.id);\n    }\n  }\n\n  const tensorsToDispose: Tensor[] = [];\n  if (tensors instanceof Tensor) {\n    if (oldTensorIds.indexOf(tensors.id) === -1) {\n      tensorsToDispose.push(tensors);\n    }\n  } else if (Array.isArray(tensors)) {\n    tensors.forEach(t => {\n      if (oldTensorIds.indexOf(t.id) === -1) {\n        tensorsToDispose.push(t);\n      }\n    });\n  } else if (tensors != null) {\n    // `oldTensors` is a map from string name to Tensor.\n    for (const name in tensors) {\n      const tensor = tensors[name];\n      if (oldTensorIds.indexOf(tensor.id) === -1) {\n        tensorsToDispose.push(tensor);\n      }\n    }\n  }\n\n  tensorsToDispose.forEach(t => {\n    if (!t.isDisposed) {\n      t.dispose();\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original Source: engine/training.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {io, ModelPredictConfig as ModelPredictArgs, NamedTensorMap, Optimizer, Scalar, scalar, serialization, Tensor, Tensor1D, tensor1d, util} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {History, ModelLoggingVerbosity} from '../base_callbacks';\nimport {nameScope} from '../common';\nimport {NotImplementedError, RuntimeError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {LossIdentifier} from '../keras_format/loss_config';\nimport {OptimizerSerialization} from '../keras_format/optimizer_config';\nimport {MetricsIdentifier, TrainingConfig} from '../keras_format/training_config';\nimport {deserialize} from '../layers/serialization';\nimport * as losses from '../losses';\nimport * as Metrics from '../metrics';\nimport * as optimizers from '../optimizers';\nimport {LossOrMetricFn, NamedTensor} from '../types';\nimport {checkUserDefinedMetadata} from '../user_defined_metadata';\nimport {count, pyListRepeat, singletonOrArray, toCamelCase, toSnakeCase, unique} from '../utils/generic_utils';\nimport {printSummary} from '../utils/layer_utils';\nimport {range} from '../utils/math_utils';\nimport {convertPythonicToTs} from '../utils/serialization_utils';\nimport {LayerVariable} from '../variables';\nimport {version} from '../version';\n\nimport {Container, ContainerArgs} from './container';\nimport {Dataset} from './dataset_stub';\nimport {execute, FeedDict} from './executor';\nimport {DisposeResult, SymbolicTensor} from './topology';\nimport {evaluateDataset, fitDataset, ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './training_dataset';\nimport {checkBatchSize, disposeNewTensors, ensureTensorsRank2OrHigher, fitTensors, makeBatches, ModelFitArgs, sliceArrays, sliceArraysByIndices} from './training_tensors';\nimport {ClassWeight, ClassWeightMap, computeWeightedLoss, standardizeClassWeights, standardizeWeights} from './training_utils';\n\n/**\n * Helper function for polymorphic input data: 1. singleton Tensor.\n */\nexport function isDataTensor(x: Tensor|Tensor[]|{[inputName: string]: Tensor}|\n                             {[inputName: string]: Tensor[]}): boolean {\n  return x instanceof Tensor;\n}\n\n/**\n * Helper function for polymorphic input data: 2. Array of Tensor.\n */\nexport function isDataArray(x: Tensor|Tensor[]|\n                            {[inputName: string]: Tensor}): boolean {\n  return Array.isArray(x);\n}\n\n/**\n * Helper function for polymorphic input data: 3. \"dict\" of Tensor.\n */\nexport function isDataDict(x: Tensor|Tensor[]|\n                           {[inputName: string]: Tensor}): boolean {\n  return !isDataTensor(x) && !isDataArray(x);\n}\n\n/**\n * Normalizes inputs and targets provided by users.\n * @param data User-provided input data (polymorphic).\n * @param names An Array of expected Tensor names.\n * @param shapes Optional Array of expected Tensor shapes.\n * @param checkBatchAxis Whether to check that the batch axis of the arrays\n *   match  the expected value found in `shapes`.\n * @param exceptionPrefix String prefix used for exception formatting.\n * @returns List of standardized input Tensors (one Tensor per model input).\n * @throws ValueError: in case of improperly formatted user data.\n */\nexport function standardizeInputData(\n    data: Tensor|Tensor[]|{[inputName: string]: Tensor}, names: string[],\n    shapes?: Shape[], checkBatchAxis = true, exceptionPrefix = ''): Tensor[] {\n  if (names == null || names.length === 0) {\n    // Check for the case where the model expected no data, but some data got\n    // sent.\n    if (data != null) {\n      let gotUnexpectedData = false;\n      if (isDataArray(data) && (data as Tensor[]).length > 0) {\n        gotUnexpectedData = true;\n      } else if (isDataDict(data)) {\n        for (const key in data) {\n          if (data.hasOwnProperty(key)) {\n            gotUnexpectedData = true;\n            break;\n          }\n        }\n      } else {\n        // `data` is a singleton Tensor in this case.\n        gotUnexpectedData = true;\n      }\n      if (gotUnexpectedData) {\n        throw new ValueError(\n            `Error when checking model ${exceptionPrefix} expected no data, ` +\n            `but got ${data}`);\n      }\n    }\n    return [];\n  }\n  if (data == null) {\n    return names.map(name => null);\n  }\n\n  let arrays: Tensor[];\n  if (isDataDict(data)) {\n    data = data as {[inputName: string]: Tensor};\n    arrays = [];\n    for (const name of names) {\n      if (data[name] == null) {\n        throw new ValueError(\n            `No data provided for \"${name}\". Need data for each key in: ` +\n            `${names}`);\n      }\n      arrays.push(data[name]);\n    }\n  } else if (isDataArray(data)) {\n    data = data as Tensor[];\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `model expected. Expected to see ${names.length} Tensor(s), but ` +\n          `instead got the following list of Tensor(s): ${data}`);\n    }\n    arrays = data;\n  } else {\n    data = data as Tensor;\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model ${exceptionPrefix} expects ${names.length} Tensor(s), ` +\n          `but only received one Tensor. Found: Tensor with shape ${\n              data.shape}`);\n    }\n    arrays = [data];\n  }\n\n  arrays = ensureTensorsRank2OrHigher(arrays);\n\n  // Check shape compatibility.\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s). but got array with ` +\n            `shape ${array.shape}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          // Skip the first (batch) axis.\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null && refDim >= 0 && dim !== refDim) {\n          throw new ValueError(\n              `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n              `to have shape [${shapes[i]}], but got array with shape ` +\n              `[${array.shape}].`);\n        }\n      }\n    }\n  }\n  return arrays;\n}\n\n/**\n * User input validation for Tensors.\n * @param inputs `Array` of `tf.Tensor`s for inputs.\n * @param targets `Array` of `tf.Tensor`s for targets.\n * @param weights Optional `Array` of `tf.Tensor`s for sample weights.\n * @throws ValueError: in case of incorrectly formatted data.\n */\nexport function checkArrayLengths(\n    inputs: Tensor[], targets: Tensor[], weights?: Tensor[]) {\n  const setX = unique(inputs.map(input => input.shape[0]));\n  setX.sort();\n  const setY = unique(targets.map(target => target.shape[0]));\n  setY.sort();\n  // TODO(cais): Check `weights` as well.\n  if (setX.length > 1) {\n    throw new ValueError(\n        `All input Tensors (x) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(inputs.map(input => input.shape))}`);\n  }\n  if (setY.length > 1) {\n    throw new ValueError(\n        `All target Tensors (y) should have the same number of samples. ` +\n        `Got array shapes: ` +\n        `${JSON.stringify(targets.map(target => target.shape))}`);\n  }\n  if (setX.length > 0 && setY.length > 0 && !util.arraysEqual(setX, setY)) {\n    throw new ValueError(\n        `Input Tensors should have the same number of samples as target ` +\n        `Tensors. Found ${setX[0]} input sample(s) and ${setY[0]} target ` +\n        `sample(s).`);\n  }\n}\n\n/**\n * Validation on the compatibility of targes and loss functions.\n *\n * This helps prevent users from using loss functions incorrectly.\n *\n * @param targets `Array` of `tf.Tensor`s of targets.\n * @param lossFns `Array` of loss functions.\n * @param outputShapes `Array` of shapes of model outputs.\n */\nfunction checkLossAndTargetCompatibility(\n    targets: Tensor[], lossFns: LossOrMetricFn[], outputShapes: Shape[]) {\n  // TODO(cais): Dedicated test coverage?\n  const keyLosses = [\n    losses.meanSquaredError, losses.binaryCrossentropy,\n    losses.categoricalCrossentropy\n  ];\n  for (let i = 0; i < targets.length; ++i) {\n    const y = targets[i];\n    const loss = lossFns[i];\n    const shape = outputShapes[i];\n    if (loss == null) {\n      continue;\n    }\n    if (loss === losses.categoricalCrossentropy) {\n      if (y.shape[y.shape.length - 1] === 1) {\n        throw new ValueError(\n            `You are passing a target array of shape ${y.shape} while using ` +\n            `a loss 'categorical_crossentropy'. 'categorical_crossentropy'` +\n            `expects targets to be binary matrices (1s and 0s) of shape ` +\n            `[samples, classes].`);\n        // TODO(cais): Example code in error message.\n      }\n    }\n    if (keyLosses.indexOf(loss) !== -1) {\n      const slicedYShape = y.shape.slice(1);\n      const slicedShape = shape.slice(1);\n      for (let j = 0; j < slicedYShape.length; ++j) {\n        const targetDim = slicedYShape[j];\n        const outDim = slicedShape[j];\n        if (outDim != null && targetDim !== outDim) {\n          throw new ValueError(\n              `A target Tensor with shape ${y.shape} was passed for an ` +\n              `output of shape ${shape}, while using a loss function that ` +\n              `expects targets to have the same shape as the output.`);\n        }\n      }\n    }\n  }\n}\n\n/**\n * Check inputs provided by the user.\n *\n * Porting Note: This corresponds to _standardize_input_data() in Python\n *   Keras. Because of the strong typing in TF.js, we do not need to convert\n *   the data. Specifically:\n *   1) in PyKeras, `data` can be `DataFrame` instances from pandas, for\n *      example. We don't need to worry about that here because there is no\n *      widely popular javascript/typesdcript equivalent of pandas (so far).\n *      If one becomes available in the future, we can add support.\n *   2) in PyKeras, inputs can be Python dict. But here we are stipulating\n * that the data is either a single `tf.Tensor` or an Array of `tf.Tensor`s. We\n * may add support for `Object` data inputs in the future when the need\n * arises.\n *\n * Instead, we perform basic checks for number of parameters and shapes.\n *\n * @param data: The input data.\n * @param names: Name for the inputs, from the model.\n * @param shapes: Expected shapes for the input data, from the model.\n * @param checkBatchAxis: Whether the size along the batch axis (i.e., the\n *   first dimension) will be checked for matching.\n * @param exceptionPrefix: Execption prefix message, used in generating error\n *   messages.\n * @throws ValueError: on incorrect number of inputs or mismatches in shapes.\n */\nfunction checkInputData(\n    data: Tensor|Tensor[], names: string[], shapes?: Shape[],\n    checkBatchAxis = true, exceptionPrefix = '') {\n  let arrays: Tensor[];\n  if (Array.isArray(data)) {\n    if (data.length !== names.length) {\n      throw new ValueError(\n          `Error when checking model ${exceptionPrefix}: the Array of ` +\n          `Tensors that you are passing to your model is not the size the ` +\n          `the model expected. Expected to see ${names.length} Tensor(s),` +\n          ` but instead got ${data.length} Tensors(s).`);\n    }\n    arrays = data;\n  } else {\n    if (names.length > 1) {\n      throw new ValueError(\n          `The model expects ${names.length} ${exceptionPrefix} Tensors, ` +\n          `but only received one Tensor. Found: array with shape ` +\n          `${JSON.stringify(data.shape)}.`);\n    }\n    arrays = [data];\n  }\n\n  if (shapes != null) {\n    for (let i = 0; i < names.length; ++i) {\n      if (shapes[i] == null) {\n        continue;\n      }\n      const array = arrays[i];\n      if (array.shape.length !== shapes[i].length) {\n        throw new ValueError(\n            `Error when checking ${exceptionPrefix}: expected ${names[i]} ` +\n            `to have ${shapes[i].length} dimension(s), but got array with ` +\n            `shape ${JSON.stringify(array.shape)}`);\n      }\n      for (let j = 0; j < shapes[i].length; ++j) {\n        if (j === 0 && !checkBatchAxis) {\n          continue;\n        }\n        const dim = array.shape[j];\n        const refDim = shapes[i][j];\n        if (refDim != null) {\n          if (refDim !== dim) {\n            throw new ValueError(\n                `Error when checking ${exceptionPrefix}: expected ` +\n                `${names[i]} to have shape ${JSON.stringify(shapes[i])} but ` +\n                `got array with shape ${JSON.stringify(array.shape)}.`);\n          }\n        }\n      }\n    }\n  }\n}\n\n/**\n * Maps metric functions to model outputs.\n * @param metrics An shortcut strings name, metric function, `Array` or dict\n *   (`Object`) of metric functions.\n * @param outputNames An `Array` of the names of model outputs.\n * @returns An `Array` (one entry per model output) of `Array` of metric\n *   functions. For instance, if the model has 2 outputs, and for the first\n *   output we want to compute `binaryAccuracy` and `binaryCrossentropy`,\n *   and just `binaryAccuracy` for the second output, the `Array` would look\n *   like:\n *     `[[binaryAccuracy, binaryCrossentropy],  [binaryAccuracy]]`\n * @throws TypeError: incompatible metrics format.\n */\nexport function collectMetrics(\n    metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n    {[outputName: string]: string | LossOrMetricFn},\n    outputNames: string[]): Array<Array<string|LossOrMetricFn>> {\n  if (metrics == null || Array.isArray(metrics) && metrics.length === 0) {\n    return outputNames.map(name => []);\n  }\n\n  let wrappedMetrics: Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  if (typeof metrics === 'string' || typeof metrics === 'function') {\n    wrappedMetrics = [metrics];\n  } else if (Array.isArray(metrics) || typeof metrics === 'object') {\n    wrappedMetrics = metrics as Array<string|LossOrMetricFn>|\n        {[outputName: string]: string} | {[outputName: string]: LossOrMetricFn};\n  } else {\n    throw new TypeError(\n        'Type of metrics argument not understood. Expected an string,' +\n        `function, Array, or Object, found: ${metrics}`);\n  }\n\n  if (Array.isArray(wrappedMetrics)) {\n    // We then apply all metrics to all outputs.\n    return outputNames.map(\n        name => wrappedMetrics as Array<string|LossOrMetricFn>);\n  } else {\n    // In this case, metrics is a dict.\n    const nestedMetrics: Array<Array<string|LossOrMetricFn>> = [];\n    for (const name of outputNames) {\n      let outputMetrics: string|LossOrMetricFn|Array<string|LossOrMetricFn> =\n          wrappedMetrics.hasOwnProperty(name) ? wrappedMetrics[name] : [];\n      if (!Array.isArray(outputMetrics)) {\n        outputMetrics = [outputMetrics];\n      }\n      nestedMetrics.push(outputMetrics);\n    }\n    return nestedMetrics;\n  }\n}\n\nexport interface ModelEvaluateArgs {\n  /**\n   * Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Verbosity mode.\n   */\n  verbose?: ModelLoggingVerbosity;\n\n  /**\n   * Tensor of weights to weight the contribution of different samples to the\n   * loss and metrics.\n   */\n  sampleWeight?: Tensor;\n\n  /**\n   * integer: total number of steps (batches of samples)\n   * before declaring the evaluation round finished. Ignored with the default\n   * value of `undefined`.\n   */\n  steps?: number;\n}\n\n/**\n * Configuration for calls to `LayersModel.compile()`.\n */\nexport interface ModelCompileArgs {\n  /**\n   * An instance of `tf.train.Optimizer` or a string name for an Optimizer.\n   */\n  optimizer: string|Optimizer;\n\n  /**\n   * Object function(s) or name(s) of object function(s).\n   * If the model has multiple outputs, you can use a different loss\n   * on each output by passing a dictionary or an Array of losses.\n   * The loss value that will be minimized by the model will then be the sum\n   * of all individual losses.\n   */\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n\n  /**\n   * List of metrics to be evaluated by the model during training and testing.\n   * Typically you will use `metrics=['accuracy']`.\n   * To specify different metrics for different outputs of a multi-output\n   * model, you could also pass a dictionary.\n   */\n  metrics?: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n\n  // TODO(cais): Add lossWeights, sampleWeightMode, weightedMetrics, and\n  //   targetTensors.\n}\n\nconst LAYERS_MODEL_FORMAT_NAME = 'layers-model';\n\n/**\n * A `tf.LayersModel` is a directed, acyclic graph of `tf.Layer`s plus methods\n * for training, evaluation, prediction and saving.\n *\n * `tf.LayersModel` is the basic unit of training, inference and evaluation in\n * TensorFlow.js. To create a `tf.LayersModel`, use `tf.LayersModel`.\n *\n * See also:\n *   `tf.Sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class LayersModel extends Container implements tfc.InferenceModel {\n  // The class name is 'Model' rather than 'LayersModel' for backwards\n  // compatibility since this class name shows up in the serialization format.\n  /** @nocollapse */\n  static className = 'Model';\n  protected optimizer_: Optimizer;\n  // Whether the model instance owns the optimizer: `true` if and only if\n  // `optimizer` is created from a string parameter during `compile()` call.\n  protected isOptimizerOwned: boolean;\n\n  loss: string|string[]|{[outputName: string]: string}|LossOrMetricFn|\n      LossOrMetricFn[]|{[outputName: string]: LossOrMetricFn};\n  lossFunctions: LossOrMetricFn[];\n\n  // TODO(cais): These private variables should probably not have the string\n  //   'feed' in their names, because we are not dealing with a symbolic\n  //   backend.\n  private feedOutputShapes: Shape[];\n  private feedLossFns: LossOrMetricFn[];\n  private collectedTrainableWeights: LayerVariable[];\n  private testFunction: (data: Tensor[]) => Scalar[];\n  history: History;\n\n  // A public property that can be set by Callbacks to order early stopping\n  // during `fit()` calls.\n  protected stopTraining_: boolean;\n  protected isTraining: boolean;\n\n  metrics: string|LossOrMetricFn|Array<string|LossOrMetricFn>|\n      {[outputName: string]: string | LossOrMetricFn};\n  metricsNames: string[];\n  // Porting Note: `metrics_tensors` in PyKeras is a symbolic tensor. But given\n  //   the imperative nature of tfjs-core, `metricsTensors` is a\n  //   TypeScript function here.\n  //   Also note that due to the imperative nature of tfjs-core, `metricsTensor`\n  //   here needs an output index to keep track of which output of the\n  //   LayersModel a metric belongs to. This is unlike `metrics_tensors` in\n  //   PyKeras, which is a `list` of symbolic tensors, each of which has\n  //   implicit \"knowledge\" of the outputs it depends on.\n  metricsTensors: Array<[LossOrMetricFn, number]>;\n\n  // User defind metadata (if any).\n  private userDefinedMetadata: {};\n\n  constructor(args: ContainerArgs) {\n    super(args);\n    this.isTraining = false;\n  }\n\n  /**\n   * Print a text summary of the model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - If the model has non-sequential-like topology, the inputs each layer\n   *   receives\n   * - The total number of trainable and non-trainable parameters of the model.\n   *\n   * ```js\n   * const input1 = tf.input({shape: [10]});\n   * const input2 = tf.input({shape: [20]});\n   * const dense1 = tf.layers.dense({units: 4}).apply(input1);\n   * const dense2 = tf.layers.dense({units: 8}).apply(input2);\n   * const concat = tf.layers.concatenate().apply([dense1, dense2]);\n   * const output =\n   *     tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat);\n   *\n   * const model = tf.model({inputs: [input1, input2], outputs: output});\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      throw new ValueError(\n          `This model has never been called, thus its weights have not been ` +\n          `created yet. So no summary can be displayed. Build the model ` +\n          `first (e.g., by calling it on some test data).`);\n    }\n    printSummary(this, lineLength, positions, printFn);\n  }\n\n  /**\n   * Configures and prepares the model for training and evaluation.  Compiling\n   * outfits the model with an optimizer, loss, and/or metrics.  Calling `fit`\n   * or `evaluate` on an un-compiled model will throw an error.\n   *\n   * @param args a `ModelCompileArgs` specifying the loss, optimizer, and\n   * metrics to be used for fitting and evaluating this model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  compile(args: ModelCompileArgs): void {\n    if (args.loss == null) {\n      args.loss = [];\n    }\n    this.loss = args.loss;\n\n    if (typeof args.optimizer === 'string') {\n      this.optimizer_ = optimizers.getOptimizer(args.optimizer);\n      this.isOptimizerOwned = true;\n    } else {\n      if (!(args.optimizer instanceof Optimizer)) {\n        throw new ValueError(\n            `User-defined optimizer must be an instance of tf.Optimizer.`);\n      }\n      this.optimizer_ = args.optimizer;\n      this.isOptimizerOwned = false;\n    }\n\n    // TODO(cais): Add lossWeights.\n    // TODO(cais): Add sampleWeightMode.\n\n    // Prepare loss functions.\n    let lossFunctions: LossOrMetricFn[] = [];\n    if (!Array.isArray(args.loss) && typeof args.loss !== 'string' &&\n        typeof args.loss !== 'function') {\n      args.loss = args.loss as {[outputName: string]: string};\n      for (const name in args.loss) {\n        if (this.outputNames.indexOf(name) === -1) {\n          throw new ValueError(\n              `Unknown entry in loss dictionary: \"${name}\". ` +\n              `Only expected the following keys: ${this.outputNames}`);\n        }\n      }\n      for (const name of this.outputNames) {\n        if (args.loss[name] == null) {\n          console.warn(\n              `Output \"${name}\" is missing from loss dictionary. We assume ` +\n              `this was done on purpose, and we will not be expecting data ` +\n              `to be passed to ${name} during training`);\n        }\n        lossFunctions.push(losses.get(args.loss[name]));\n      }\n    } else if (Array.isArray(args.loss)) {\n      if (args.loss.length !== this.outputs.length) {\n        throw new ValueError(\n            `When passing an Array as loss, it should have one entry per ` +\n            `model output. The model has ${this.outputs.length} output(s), ` +\n            `but you passed loss=${args.loss}.`);\n      }\n      const theLosses = args.loss as Array<string|LossOrMetricFn>;\n      lossFunctions = theLosses.map(l => losses.get(l));\n    } else {\n      const lossFunction = losses.get(args.loss);\n      this.outputs.forEach(_ => {\n        lossFunctions.push(lossFunction);\n      });\n    }\n\n    this.lossFunctions = lossFunctions;\n\n    this.feedOutputNames = [];\n    this.feedOutputShapes = [];\n    this.feedLossFns = [];\n    for (let i = 0; i < this.outputs.length; ++i) {\n      // TODO(cais): Logic for skipping target(s).\n      const shape = this.internalOutputShapes[i];\n      const name = this.outputNames[i];\n      this.feedOutputNames.push(name);\n      this.feedOutputShapes.push(shape);\n      this.feedLossFns.push(this.lossFunctions[i]);\n    }\n\n    // TODO(cais): Add logic for output masks.\n    // TODO(cais): Add logic for sample weights.\n    const skipTargetIndices: number[] = [];\n\n    // Prepare metrics.\n    this.metrics = args.metrics;\n    // TODO(cais): Add weightedMetrics.\n    this.metricsNames = ['loss'];\n    this.metricsTensors = [];\n\n    // Compute total loss.\n    // Porting Note: In PyKeras, metrics_tensors are symbolic tensor objects.\n    //   Here, metricsTensors are TypeScript functions. This difference is due\n    //   to the difference in symbolic/imperative property of the backends.\n    nameScope('loss', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        // TODO(cais): Add weightedLoss, sampleWeight and mask.\n        //   The following line should be weightedLoss\n        const weightedLoss = this.lossFunctions[i];\n        if (this.outputs.length > 1) {\n          this.metricsTensors.push([weightedLoss, i]);\n          this.metricsNames.push(this.outputNames[i] + '_loss');\n        }\n      }\n\n      // Porting Note: Due to the imperative nature of the backend, we calculate\n      //   the regularizer penalties in the totalLossFunction, instead of here.\n    });\n\n    const nestedMetrics = collectMetrics(args.metrics, this.outputNames);\n    // TODO(cais): Add nestedWeightedMetrics.\n\n    /**\n     * Helper function used in loop below.\n     */\n    const appendMetric =\n        (outputIndex: number, metricName: string,\n         metricTensor: LossOrMetricFn) => {\n          if (this.outputNames.length > 1) {\n            metricName = this.outputNames[outputIndex] + '_' + metricName;\n          }\n          this.metricsNames.push(metricName);\n          this.metricsTensors.push([metricTensor, outputIndex]);\n        };\n\n    nameScope('metric', () => {\n      for (let i = 0; i < this.outputs.length; ++i) {\n        if (skipTargetIndices.indexOf(i) !== -1) {\n          continue;\n        }\n        const outputMetrics = nestedMetrics[i];\n        // TODO(cais): Add weights and outputWeightedMetrics.\n\n        // TODO(cais): Add optional arg `weights` to the following function.\n        const handleMetrics = (metrics: Array<string|LossOrMetricFn>) => {\n          const metricNamePrefix = '';\n          let metricName: string;\n          let accFn: LossOrMetricFn;\n          let weightedMetricFn: LossOrMetricFn;\n          //  TODO(cais): Use 'weights_' for weighted metrics.\n\n          for (const metric of metrics) {\n            if (typeof metric === 'string' &&\n                ['accuracy', 'acc', 'crossentropy', 'ce'].indexOf(metric) !==\n                    -1) {\n              const outputShape = this.internalOutputShapes[i];\n\n              if (outputShape[outputShape.length - 1] === 1 ||\n                  this.lossFunctions[i] === losses.binaryCrossentropy) {\n                // case: binary accuracy/crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.binaryCrossentropy;\n                }\n              } else if (\n                  this.lossFunctions[i] ===\n                  losses.sparseCategoricalCrossentropy) {\n                // case: categorical accuracy / crossentropy with sparse\n                // targets.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.sparseCategoricalCrossentropy;\n                }\n              } else {\n                // case: categorical accuracy / crossentropy.\n                if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalAccuracy;\n                } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                  accFn = Metrics.categoricalCrossentropy;\n                }\n              }\n              let suffix: string;\n              if (['accuracy', 'acc'].indexOf(metric) !== -1) {\n                suffix = 'acc';\n              } else if (['crossentropy', 'ce'].indexOf(metric) !== -1) {\n                suffix = 'ce';\n              }\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = accFn;\n              metricName = metricNamePrefix + suffix;\n            } else {\n              const metricFn = Metrics.get(metric);\n              // TODO(cais): Add weighting actually.\n              weightedMetricFn = metricFn;\n              metricName =\n                  metricNamePrefix + Metrics.getLossOrMetricName(metric);\n            }\n\n            // TODO(cais): Add weighting and masking to metricResult.\n            let metricResult: LossOrMetricFn;\n            nameScope(metricName, () => {\n              metricResult = weightedMetricFn;\n            });\n            appendMetric(i, metricName, metricResult);\n          }\n        };\n\n        handleMetrics(outputMetrics);\n        // TODO(cais): Call handleMetrics with weights.\n      }\n    });\n\n    // Porting Notes: Given the imperative backend of tfjs-core,\n    //   there is no need for constructing the symbolic graph and placeholders.\n    this.collectedTrainableWeights = this.trainableWeights;\n  }\n\n  /**\n   * Check trainable weights count consistency.\n   *\n   * This will raise a warning if `this.trainableWeights` and\n   * `this.collectedTrainableWeights` are inconsistent (i.e., have different\n   * numbers of parameters).\n   * Inconsistency will typically arise when one modifies `model.trainable`\n   * without calling `model.compile()` again.\n   */\n  protected checkTrainableWeightsConsistency(): void {\n    if (this.collectedTrainableWeights == null) {\n      return;\n    }\n    if (this.trainableWeights.length !==\n        this.collectedTrainableWeights.length) {\n      console.warn(\n          'Discrepancy between trainableweights and collected trainable ' +\n          'weights. Did you set `model.trainable` without calling ' +\n          '`model.compile()` afterwards?');\n    }\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(\n   *     tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4});\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateArgs`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    const batchSize = args.batchSize == null ? 32 : args.batchSize;\n    checkBatchSize(batchSize);\n\n    // TODO(cais): Standardize `config.sampleWeights` as well.\n    // Validate user data.\n    const checkBatchAxis = true;\n    const standardizedOuts =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    try {\n      // TODO(cais): If uses `useLearningPhase`, set the corresponding element\n      // of the input to 0.\n      const ins = standardizedOuts[0].concat(standardizedOuts[1]);\n      this.makeTestFunction();\n      const f = this.testFunction;\n      const testOuts =\n          this.testLoop(f, ins, batchSize, args.verbose, args.steps);\n      return singletonOrArray(testOuts);\n    } finally {\n      disposeNewTensors(standardizedOuts[0], x);\n      disposeNewTensors(standardizedOuts[1], y);\n    }\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args?: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    this.makeTestFunction();\n    return evaluateDataset(this, dataset, args);\n  }\n\n  /**\n   * Get number of samples provided for training, evaluation or prediction.\n   *\n   * @param ins Input `tf.Tensor`.\n   * @param batchSize Integer batch size, optional.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring loop finished. Optional.\n   * @param stepsName The public API's parameter name for `steps`.\n   * @returns Number of samples provided.\n   */\n  private checkNumSamples(\n      ins: Tensor|Tensor[], batchSize?: number, steps?: number,\n      stepsName = 'steps'): number {\n    let numSamples: number;\n    if (steps != null) {\n      numSamples = null;\n      if (batchSize != null) {\n        throw new ValueError(\n            `If ${stepsName} is set, batchSize must be null or undefined.` +\n            `Got batchSize = ${batchSize}`);\n      }\n    } else if (ins != null) {\n      if (Array.isArray(ins)) {\n        numSamples = ins[0].shape[0];\n      } else {\n        numSamples = ins.shape[0];\n      }\n    } else {\n      throw new ValueError(\n          `Either the input data should have a defined shape, or ` +\n          `${stepsName} shoud be specified.`);\n    }\n    return numSamples;\n  }\n\n  /**\n   * Execute internal tensors of the model with input data feed.\n   * @param inputs Input data feed. Must match the inputs of the model.\n   * @param outputs Names of the output tensors to be fetched. Must match\n   *   names of the SymbolicTensors that belong to the graph.\n   * @returns Fetched values for `outputs`.\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs: string|string[]):\n      Tensor|Tensor[] {\n    if (Array.isArray(outputs) && outputs.length === 0) {\n      throw new ValueError(\n          '`outputs` is an empty Array, which is not allowed.');\n    }\n\n    const outputsIsArray = Array.isArray(outputs);\n    const outputNames =\n        (outputsIsArray ? outputs as string[] : [outputs as string]);\n    const outputSymbolicTensors = this.retrieveSymbolicTensors(outputNames);\n\n    // Format the input into a FeedDict.\n    const feedDict = new FeedDict();\n    if (inputs instanceof Tensor) {\n      inputs = [inputs];\n    }\n    if (Array.isArray(inputs)) {\n      if (inputs.length !== this.inputs.length) {\n        throw new ValueError(\n            `The number of inputs provided (${inputs.length}) ` +\n            `does not match the number of inputs of this model ` +\n            `(${this.inputs.length}).`);\n      }\n      for (let i = 0; i < this.inputs.length; ++i) {\n        feedDict.add(this.inputs[i], inputs[i]);\n      }\n    } else {\n      for (const input of this.inputs) {\n        const tensorValue = inputs[input.name];\n        if (tensorValue == null) {\n          throw new ValueError(\n              `No value is provided for the model's input ${input.name}`);\n        }\n        feedDict.add(input, tensorValue);\n      }\n    }\n\n    // Run execution.\n    const executeOutputs = execute(outputSymbolicTensors, feedDict) as Tensor[];\n    return outputsIsArray ? executeOutputs : executeOutputs[0];\n  }\n\n  /**\n   * Retrieve the model's internal symbolic tensors from symbolic-tensor names.\n   */\n  private retrieveSymbolicTensors(symbolicTensorNames: string[]):\n      SymbolicTensor[] {\n    const outputSymbolicTensors: SymbolicTensor[] =\n        pyListRepeat(null, symbolicTensorNames.length);\n    let outputsRemaining = symbolicTensorNames.length;\n    for (const layer of this.layers) {\n      const layerOutputs: SymbolicTensor[] =\n          Array.isArray(layer.output) ? layer.output : [layer.output];\n      const layerOutputNames = layerOutputs.map(output => output.name);\n      for (let i = 0; i < symbolicTensorNames.length; ++i) {\n        const index = layerOutputNames.indexOf(symbolicTensorNames[i]);\n        if (index !== -1) {\n          outputSymbolicTensors[i] = layerOutputs[index];\n          outputsRemaining--;\n        }\n        if (outputsRemaining === 0) {\n          break;\n        }\n      }\n      if (outputsRemaining === 0) {\n        break;\n      }\n    }\n\n    if (outputsRemaining > 0) {\n      const remainingNames: string[] = [];\n      outputSymbolicTensors.forEach((tensor, i) => {\n        if (tensor == null) {\n          remainingNames.push(symbolicTensorNames[i]);\n        }\n      });\n      throw new ValueError(\n          `Cannot find SymbolicTensors for output name(s): ` +\n          `${JSON.stringify(remainingNames)}`);\n    }\n    return outputSymbolicTensors;\n  }\n\n  /**\n   * Helper method to loop over some data in batches.\n   *\n   * Porting Note: Not using the functional approach in the Python equivalent\n   *   due to the imperative backend.\n   * Porting Note: Does not support step mode currently.\n   *\n   * @param ins: input data\n   * @param batchSize: integer batch size.\n   * @param verbose: verbosity model\n   * @returns: Predictions as `tf.Tensor` (if a single output) or an `Array` of\n   *   `tf.Tensor` (if multipe outputs).\n   */\n  private predictLoop(ins: Tensor|Tensor[], batchSize = 32, verbose = false):\n      Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins);\n      if (verbose) {\n        throw new NotImplementedError(\n            'Verbose predictLoop() is not implemented yet.');\n      }\n\n      // Sample-based predictions.\n      // Porting Note: Tensor currently does not support sliced assignments as\n      //   in numpy, e.g., x[1:3] = y. Therefore we use concatenation while\n      //   iterating over the batches.\n\n      const batches = makeBatches(numSamples, batchSize);\n      const outsBatches: Tensor[][] = this.outputs.map(output => []);\n\n      // TODO(cais): Can the scope() be pushed down inside the for loop?\n      for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n        const batchOuts = tfc.tidy(() => {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          // TODO(cais): Take care of the case of the last element is a flag for\n          //   training/test.\n          const insBatch = sliceArrays(ins, batchStart, batchEnd);\n\n          // Construct the feeds for execute();\n          const feeds = [];\n          if (Array.isArray(insBatch)) {\n            for (let i = 0; i < insBatch.length; ++i) {\n              feeds.push({key: this.inputs[i], value: insBatch[i]});\n            }\n          } else {\n            feeds.push({key: this.inputs[0], value: insBatch});\n          }\n          const feedDict = new FeedDict(feeds);\n          return execute(this.outputs, feedDict) as Tensor[];\n        });\n        batchOuts.forEach((batchOut, i) => outsBatches[i].push(batchOut));\n      }\n      return singletonOrArray(\n          outsBatches.map(batches => tfc.concat(batches, 0)));\n    });\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFlow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([8, 10]), {batchSize: 4}).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param args A `ModelPredictArgs` object containing optional fields.\n   *\n   * @return Prediction results as a `tf.Tensor`(s).\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    const xsRank2OrHigher = ensureTensorsRank2OrHigher(x);\n    checkInputData(\n        xsRank2OrHigher, this.inputNames, this.feedInputShapes, false);\n    try {\n      // TODO(cais): Take care of stateful models.\n      //   if (this.stateful) ...\n      // TODO(cais): Take care of the learning_phase boolean flag.\n      //   if (this.useLearningPhase) ...\n      const batchSize = args.batchSize == null ? 32 : args.batchSize;\n      checkBatchSize(batchSize);\n      return this.predictLoop(xsRank2OrHigher, batchSize);\n    } finally {\n      disposeNewTensors(xsRank2OrHigher, x);\n    }\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predictOnBatch(tf.ones([8, 10])).print();\n   * ```\n   * @param x: Input samples, as a Tensor (for models with exactly one\n   *   input) or an array of Tensors (for models with more than one input).\n   * @return Tensor(s) of predictions\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predictOnBatch(x: Tensor|Tensor[]): Tensor|Tensor[] {\n    checkInputData(x, this.inputNames, this.feedInputShapes, true);\n    // TODO(cais): Take care of the learning_phase boolean flag.\n    //   if (this.useLearningPhase) ...\n    const batchSize = (Array.isArray(x) ? x[0] : x).shape[0];\n    return this.predictLoop(x, batchSize);\n  }\n\n  protected standardizeUserDataXY(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor}, checkBatchAxis = true,\n      batchSize?: number): [Tensor[], Tensor[]] {\n    // TODO(cais): Add sampleWeight, classWeight\n    if (this.optimizer_ == null) {\n      throw new RuntimeError(\n          'You must compile a model before training/testing. Use ' +\n          'LayersModel.compile(modelCompileArgs).');\n    }\n    const outputShapes: Shape[] = [];\n    for (let i = 0; i < this.feedOutputShapes.length; ++i) {\n      const outputShape = this.feedOutputShapes[i];\n      const lossFn = this.feedLossFns[i];\n      if (lossFn === losses.sparseCategoricalCrossentropy) {\n        outputShapes.push(\n            outputShape.slice(0, outputShape.length - 1).concat([1]));\n      } else {\n        // Porting Note: Because of strong typing `lossFn` must be a function.\n        outputShapes.push(outputShape);\n      }\n    }\n    x = standardizeInputData(\n        x, this.feedInputNames, this.feedInputShapes, false, 'input');\n    y = standardizeInputData(\n        y, this.feedOutputNames, outputShapes, false, 'target');\n    // TODO(cais): Standardize sampleWeights & classWeights.\n    checkArrayLengths(x, y, null);\n    // TODO(cais): Check sampleWeights as well.\n    checkLossAndTargetCompatibility(y, this.feedLossFns, this.feedOutputShapes);\n    if (this.stateful && batchSize != null && batchSize > 0) {\n      if (x[0].shape[0] % batchSize !== 0) {\n        throw new ValueError(\n            `In a stateful network, you should only pass inputs with a ` +\n            `number of samples that is divisible by the batch size ` +\n            `${batchSize}. Found: ${x[0].shape[0]} sample(s).`);\n      }\n    }\n    return [x, y];\n  }\n\n  protected async standardizeUserData(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      sampleWeight?: Tensor|Tensor[]|{[outputName: string]: Tensor},\n      classWeight?: ClassWeight|ClassWeight[]|ClassWeightMap,\n      checkBatchAxis = true,\n      batchSize?: number): Promise<[Tensor[], Tensor[], Tensor[]]> {\n    const [standardXs, standardYs] =\n        this.standardizeUserDataXY(x, y, checkBatchAxis, batchSize);\n    // TODO(cais): Handle sampleWeights.\n    if (sampleWeight != null) {\n      throw new Error('sample weight is not supported yet.');\n    }\n\n    let standardSampleWeights: Tensor[] = null;\n    if (classWeight != null) {\n      const classWeights =\n          standardizeClassWeights(classWeight, this.outputNames);\n      standardSampleWeights = [];\n      for (let i = 0; i < classWeights.length; ++i) {\n        standardSampleWeights.push(\n            await standardizeWeights(standardYs[i], null, classWeights[i]));\n      }\n    }\n\n    // TODO(cais): Deal with the case of model.stateful == true.\n    return [standardXs, standardYs, standardSampleWeights];\n  }\n\n  /**\n   * Loop over some test data in batches.\n   * @param f A Function returning a list of tensors.\n   * @param ins Array of tensors to be fed to `f`.\n   * @param batchSize Integer batch size or `null` / `undefined`.\n   * @param verbose verbosity mode.\n   * @param steps Total number of steps (batches of samples) before\n   * declaring test finished. Ignored with the default value of `null` /\n   * `undefined`.\n   * @returns Array of Scalars.\n   */\n  private testLoop(\n      f: (data: Tensor[]) => Scalar[], ins: Tensor[], batchSize?: number,\n      verbose = 0, steps?: number): Scalar[] {\n    return tfc.tidy(() => {\n      const numSamples = this.checkNumSamples(ins, batchSize, steps, 'steps');\n      const outs: Scalar[] = [];\n      if (verbose > 0) {\n        throw new NotImplementedError('Verbose mode is not implemented yet.');\n      }\n      // TODO(cais): Use `indicesForConversionToDense' to prevent slow down.\n      if (steps != null) {\n        throw new NotImplementedError(\n            'steps mode in testLoop() is not implemented yet');\n      } else {\n        const batches = makeBatches(numSamples, batchSize);\n        const indexArray = tensor1d(range(0, numSamples));\n        for (let batchIndex = 0; batchIndex < batches.length; ++batchIndex) {\n          const batchStart = batches[batchIndex][0];\n          const batchEnd = batches[batchIndex][1];\n          const batchIds =\n              K.sliceAlongFirstAxis(\n                  indexArray, batchStart, batchEnd - batchStart) as Tensor1D;\n          // TODO(cais): In ins, train flag can be a number, instead of an\n          //   Tensor? Do we need to handle this in tfjs-layers?\n          const insBatch = sliceArraysByIndices(ins, batchIds) as Scalar[];\n          const batchOuts = f(insBatch);\n          if (batchIndex === 0) {\n            for (let i = 0; i < batchOuts.length; ++i) {\n              outs.push(scalar(0));\n            }\n          }\n          for (let i = 0; i < batchOuts.length; ++i) {\n            const batchOut = batchOuts[i];\n            outs[i] =\n                tfc.add(outs[i], tfc.mul(batchEnd - batchStart, batchOut));\n          }\n        }\n        for (let i = 0; i < outs.length; ++i) {\n          outs[i] = tfc.div(outs[i], numSamples);\n        }\n      }\n      return outs;\n    });\n  }\n\n  protected getDedupedMetricsNames(): string[] {\n    const outLabels = this.metricsNames;\n    // Rename duplicated metrics names (can happen with an output layer\n    // shared among multiple dataflows).\n    const dedupedOutLabels = [];\n    for (let i = 0; i < outLabels.length; ++i) {\n      const label = outLabels[i];\n      let newLabel = label;\n      if (count(outLabels, label) > 1) {\n        const dupIndex = count(outLabels.slice(0, i), label);\n        newLabel += `_${dupIndex}`;\n      }\n      dedupedOutLabels.push(newLabel);\n    }\n    return dedupedOutLabels;\n  }\n\n  /**\n   * Creates a function that performs the following actions:\n   *\n   * 1. computes the losses\n   * 2. sums them to get the total loss\n   * 3. call the optimizer computes the gradients of the LayersModel's\n   *    trainable weights w.r.t. the total loss and update the variables\n   * 4. calculates the metrics\n   * 5. returns the values of the losses and metrics.\n   */\n  protected makeTrainFunction(): (data: Tensor[]) => Scalar[] {\n    return (data: Tensor[]) => {\n      const lossValues: Scalar[] = [];\n\n      const inputs = data.slice(0, this.inputs.length);\n      const targets = data.slice(\n          this.inputs.length, this.inputs.length + this.outputs.length);\n      const sampleWeights = data.slice(\n          this.inputs.length + this.outputs.length,\n          this.inputs.length + this.outputs.length * 2);\n\n      const metricsValues: Scalar[] = [];\n\n      // Create a function that computes the total loss based on the\n      // inputs. This function is used for obtaining gradients through\n      // backprop.\n      const totalLossFunction = () => {\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs =\n            execute(this.outputs, feedDict, {'training': true}) as Tensor[];\n        // TODO(cais): Take care of the case of multiple outputs from a\n        //   single layer?\n\n        let totalLoss: Tensor;\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          let loss = lossFunction(targets[i], outputs[i]);\n          if (sampleWeights[i] != null) {\n            loss = computeWeightedLoss(loss, sampleWeights[i]);\n          }\n\n          // TODO(cais): push Scalar instead.\n          const meanLoss: Scalar = tfc.mean(loss);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          lossValues.push(meanLoss);\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n        }\n\n        // Compute the metrics.\n        // TODO(cais): These should probably be calculated outside\n        //   totalLossFunction to benefit speed?\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          let weightedMetric: Scalar;\n\n          if (this.outputs.length > 1 && i < this.outputs.length) {\n            weightedMetric = lossValues[i];\n          } else {\n            const metric = this.metricsTensors[i][0];\n            const outputIndex = this.metricsTensors[i][1];\n            weightedMetric =\n                tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          }\n\n          tfc.keep(weightedMetric);\n          // TODO(cais): Use a scope() instead, to avoid ownership.\n          metricsValues.push(weightedMetric);\n        }\n\n        totalLoss = tfc.mean(totalLoss);\n\n        // Add regularizer penalties.\n        this.calculateLosses().forEach(regularizerLoss => {\n          totalLoss = tfc.add(totalLoss, regularizerLoss);\n        });\n\n        return totalLoss as Scalar;\n      };\n\n      const variables = this.collectedTrainableWeights.map(\n          param => param.read() as tfc.Variable);\n      const returnCost = true;\n      const totalLossValue =\n          this.optimizer_.minimize(totalLossFunction, returnCost, variables);\n\n      return [totalLossValue].concat(metricsValues);\n    };\n  }\n\n  /**\n   * Create a function which, when invoked with an array of `tf.Tensor`s as a\n   * batch of inputs, returns the prespecified loss and metrics of the model\n   * under the batch of input data.\n   */\n  private makeTestFunction() {\n    this.testFunction = (data: Tensor[]) => {\n      return tfc.tidy(() => {\n        const valOutputs: Scalar[] = [];\n        let totalLoss: Scalar;\n        const inputs = data.slice(0, this.inputs.length);\n        const targets = data.slice(\n            this.inputs.length, this.inputs.length + this.outputs.length);\n        const feeds = [];\n        for (let i = 0; i < this.inputs.length; ++i) {\n          feeds.push({key: this.inputs[i], value: inputs[i]});\n        }\n        const feedDict = new FeedDict(feeds);\n        const outputs = execute(this.outputs, feedDict) as Tensor[];\n        // Compute total loss.\n        for (let i = 0; i < this.lossFunctions.length; ++i) {\n          const lossFunction = this.lossFunctions[i];\n          // TODO(cais): Add sample weighting and replace the simple\n          // averaging.\n          const loss: Scalar = tfc.mean(lossFunction(targets[i], outputs[i]));\n          if (i === 0) {\n            totalLoss = loss;\n          } else {\n            totalLoss = tfc.add(totalLoss, loss);\n          }\n          valOutputs.push(totalLoss);\n        }\n        // Compute the metrics.\n        for (let i = 0; i < this.metricsTensors.length; ++i) {\n          const metric = this.metricsTensors[i][0];\n          const outputIndex = this.metricsTensors[i][1];\n          // TODO(cais): Replace K.mean() with a proper weighting function.\n          const meanMetric =\n              tfc.mean(metric(targets[outputIndex], outputs[outputIndex]));\n          valOutputs.push(meanMetric as Scalar);\n        }\n        return valOutputs;\n      });\n    };\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a\n   * dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *     layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * for (let i = 1; i < 5 ; ++i) {\n   *   const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *       batchSize: 4,\n   *       epochs: 3\n   *   });\n   *   console.log(\"Loss after Epoch \" + i + \" : \" + h.history.loss[0]);\n   * }\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you\n   * can also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named,\n   * you can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args A `ModelFitArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input\n   * data and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    return fitTensors(this, x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below when it's possible to instantiate\n  //   actual dataset objects.\n  /**\n   * Trains the model using a dataset object.\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for training. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs.\n   *   Of the two items in the array, the first is the input feature(s) and\n   *   the second is the output target(s).\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    return fitDataset(this, dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    // TODO(cais): Support sampleWeight and classWeight.\n    // TODO(cais): Support Dataset objects.\n    const standardizeOut = await this.standardizeUserData(x, y);\n    const inputs = standardizeOut[0];\n    const targets = standardizeOut[1];\n    const trainFunction = this.makeTrainFunction();\n    const losses = trainFunction(inputs.concat(targets));\n    const lossValues: number[] = [];\n    for (const loss of losses) {\n      const v = await loss.data();\n      lossValues.push(v[0]);\n    }\n    tfc.dispose(losses);\n    return singletonOrArray(lossValues);\n  }\n\n  /**\n   * Extract weight values of the model.\n   *\n   * @param config: An instance of `io.SaveConfig`, which specifies\n   * model-saving options such as whether only trainable weights are to be\n   * saved.\n   * @returns A `NamedTensorMap` mapping original weight names (i.e.,\n   *   non-uniqueified weight names) to their values.\n   */\n  protected getNamedWeights(config?: io.SaveConfig): NamedTensor[] {\n    const namedWeights: NamedTensor[] = [];\n\n    const trainableOnly = config != null && config.trainableOnly;\n    const weights = trainableOnly ? this.trainableWeights : this.weights;\n    const weightValues = this.getWeights(trainableOnly);\n    for (let i = 0; i < weights.length; ++i) {\n      if (trainableOnly && !weights[i].trainable) {\n        // Optionally skip non-trainable weights.\n        continue;\n      }\n      namedWeights.push(\n          {name: weights[i].originalName, tensor: weightValues[i]});\n    }\n    return namedWeights;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const input = tf.input({shape: [10]});\n   * const output = tf.layers.dense({units: 1}).apply(input);\n   * const model = tf.model({inputs: [input], outputs: [output]});\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10\n   * values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    this.stopTraining_ = stop;\n  }\n\n  get stopTraining(): boolean {\n    return this.stopTraining_;\n  }\n\n  get optimizer(): Optimizer {\n    return this.optimizer_;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    if (this.optimizer_ !== optimizer) {\n      this.optimizer_ = optimizer;\n      this.isOptimizerOwned = false;\n    }\n  }\n\n  dispose(): DisposeResult {\n    const result = super.dispose();\n    if (result.refCountAfterDispose === 0 && this.optimizer != null &&\n        this.isOptimizerOwned) {\n      const numTensorsBeforeOptmizerDisposal = tfc.memory().numTensors;\n      this.optimizer_.dispose();\n      result.numDisposedVariables +=\n          numTensorsBeforeOptmizerDisposal - tfc.memory().numTensors;\n    }\n    return result;\n  }\n\n  private getLossIdentifiers(): LossIdentifier|LossIdentifier[]|\n      {[outputName: string]: LossIdentifier} {\n    let lossNames: LossIdentifier|LossIdentifier[]|\n        {[outputName: string]: LossIdentifier};\n    if (typeof this.loss === 'string') {\n      lossNames = toSnakeCase(this.loss) as LossIdentifier;\n    } else if (Array.isArray(this.loss)) {\n      for (const loss of this.loss) {\n        if (typeof loss !== 'string') {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n      lossNames = (this.loss as string[]).map(name => toSnakeCase(name)) as\n          LossIdentifier[];\n    } else {\n      const outputNames = Object.keys(this.loss);\n      lossNames = {} as {[outputName: string]: LossIdentifier};\n      const losses =\n          this.loss as {[outputName: string]: LossOrMetricFn | string};\n      for (const outputName of outputNames) {\n        if (typeof losses[outputName] === 'string') {\n          lossNames[outputName] =\n              toSnakeCase(losses[outputName] as string) as LossIdentifier;\n        } else {\n          throw new Error('Serialization of non-string loss is not supported.');\n        }\n      }\n    }\n    return lossNames;\n  }\n\n  private getMetricIdentifiers(): MetricsIdentifier[]|\n      {[key: string]: MetricsIdentifier} {\n    if (typeof this.metrics === 'string' ||\n        typeof this.metrics === 'function') {\n      return [toSnakeCase(Metrics.getLossOrMetricName(this.metrics))];\n    } else if (Array.isArray(this.metrics)) {\n      return this.metrics.map(\n          metric => toSnakeCase(Metrics.getLossOrMetricName(metric)));\n    } else {\n      const metricsIdentifiers: {[key: string]: MetricsIdentifier} = {};\n      for (const key in this.metrics) {\n        metricsIdentifiers[key] =\n            toSnakeCase(Metrics.getLossOrMetricName(this.metrics[key]));\n      }\n      return metricsIdentifiers;\n    }\n  }\n\n  protected getTrainingConfig(): TrainingConfig {\n    return {\n      loss: this.getLossIdentifiers(),\n      metrics: this.getMetricIdentifiers(),\n      optimizer_config: {\n        class_name: this.optimizer.getClassName(),\n        config: this.optimizer.getConfig()\n      } as OptimizerSerialization\n    };\n    // TODO(cais): Add weight_metrics when they are supported.\n    // TODO(cais): Add sample_weight_mode when it's supported.\n    // TODO(cais): Add loss_weights when it's supported.\n  }\n\n  loadTrainingConfig(trainingConfig: TrainingConfig) {\n    if (trainingConfig.weighted_metrics != null) {\n      throw new Error('Loading weight_metrics is not supported yet.');\n    }\n    if (trainingConfig.loss_weights != null) {\n      throw new Error('Loading loss_weights is not supported yet.');\n    }\n    if (trainingConfig.sample_weight_mode != null) {\n      throw new Error('Loading sample_weight_mode is not supported yet.');\n    }\n\n    const tsConfig = convertPythonicToTs(trainingConfig.optimizer_config) as\n        serialization.ConfigDict;\n    const optimizer = deserialize(tsConfig) as Optimizer;\n\n    let loss;\n    if (typeof trainingConfig.loss === 'string') {\n      loss = toCamelCase(trainingConfig.loss);\n    } else if (Array.isArray(trainingConfig.loss)) {\n      loss = trainingConfig.loss.map(lossEntry => toCamelCase(lossEntry));\n    } else if (trainingConfig.loss != null) {\n      loss = {} as {[outputName: string]: LossIdentifier};\n      for (const key in trainingConfig.loss) {\n        loss[key] = toCamelCase(trainingConfig.loss[key]) as LossIdentifier;\n      }\n    }\n\n    let metrics;\n    if (Array.isArray(trainingConfig.metrics)) {\n      metrics = trainingConfig.metrics.map(metric => toCamelCase(metric));\n    } else if (trainingConfig.metrics != null) {\n      metrics = {} as {[outputName: string]: MetricsIdentifier};\n      for (const key in trainingConfig.metrics) {\n        metrics[key] = toCamelCase(trainingConfig.metrics[key]);\n      }\n    }\n\n    this.compile({loss, metrics, optimizer});\n  }\n\n  /**\n   * Save the configuration and/or weights of the LayersModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 2. Saving `model`'s topology and weights to browser\n   * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n   * then load it back.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * console.log('Prediction from original model:');\n   * model.predict(tf.ones([1, 3])).print();\n   *\n   * const saveResults = await model.save('indexeddb://my-model-1');\n   *\n   * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * loadedModel.predict(tf.ones([1, 3])).print();\n   * ```\n   *\n   * Example 3. Saving `model`'s topology and weights as two files\n   * (`my-model-1.json` and `my-model-1.weights.bin`) downloaded from\n   * browser.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('downloads://my-model-1');\n   * ```\n   *\n   * Example 4. Send  `model`'s topology and weights to an HTTP server.\n   * See the documentation of `tf.io.http` for more details\n   * including specifying request parameters and implementation of the\n   * server.\n   *\n   * ```js\n   * const model = tf.sequential(\n   *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n   * const saveResults = await model.save('http://my-server/model/upload');\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new ValueError(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new ValueError(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new ValueError(\n          'LayersModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    const weightDataAndSpecs =\n        await io.encodeWeights(this.getNamedWeights(config));\n\n    const returnString = false;\n    const unusedArg: {} = null;\n    const modelConfig = this.toJSON(unusedArg, returnString);\n    const modelArtifacts: io.ModelArtifacts = {\n      modelTopology: modelConfig,\n      format: LAYERS_MODEL_FORMAT_NAME,\n      generatedBy: `TensorFlow.js tfjs-layers v${version}`,\n      convertedBy: null,\n    };\n\n    const includeOptimizer = config == null ? false : config.includeOptimizer;\n    if (includeOptimizer && this.optimizer != null) {\n      modelArtifacts.trainingConfig = this.getTrainingConfig();\n      const weightType = 'optimizer';\n      const {data: optimizerWeightData, specs: optimizerWeightSpecs} =\n          await io.encodeWeights(await this.optimizer.getWeights(), weightType);\n      weightDataAndSpecs.specs.push(...optimizerWeightSpecs);\n      weightDataAndSpecs.data = io.concatenateArrayBuffers(\n          [weightDataAndSpecs.data, optimizerWeightData]);\n    }\n\n    if (this.userDefinedMetadata != null) {\n      // Check serialized size of user-defined metadata.\n      const checkSize = true;\n      checkUserDefinedMetadata(this.userDefinedMetadata, this.name, checkSize);\n      modelArtifacts.userDefinedMetadata = this.userDefinedMetadata;\n    }\n\n    modelArtifacts.weightData = weightDataAndSpecs.data;\n    modelArtifacts.weightSpecs = weightDataAndSpecs.specs;\n    return handlerOrURL.save(modelArtifacts);\n  }\n\n  /**\n   * Set user-defined metadata.\n   *\n   * The set metadata will be serialized together with the topology\n   * and weights of the model during `save()` calls.\n   *\n   * @param setUserDefinedMetadata\n   */\n  setUserDefinedMetadata(userDefinedMetadata: {}): void {\n    checkUserDefinedMetadata(userDefinedMetadata, this.name);\n    this.userDefinedMetadata = userDefinedMetadata;\n  }\n\n  /**\n   * Get user-defined metadata.\n   *\n   * The metadata is supplied via one of the two routes:\n   *   1. By calling `setUserDefinedMetadata()`.\n   *   2. Loaded during model loading (if the model is constructed\n   *      via `tf.loadLayersModel()`.)\n   *\n   * If no user-defined metadata is available from either of the\n   * two routes, this function will return `undefined`.\n   */\n  getUserDefinedMetadata(): {} {\n    return this.userDefinedMetadata;\n  }\n}\nserialization.registerClass(LayersModel);\n\n/**\n * A `tf.Functional` is an alias to `tf.LayersModel`.\n *\n * See also:\n *   `tf.LayersModel`, `tf.Sequential`, `tf.loadLayersModel`.\n */\n/** @doc {heading: 'Models', subheading: 'Classes'} */\nexport class Functional extends LayersModel {\n  static className = 'Functional';\n}\nserialization.registerClass(Functional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source keras/models.py */\n\nimport {dispose, io, NamedTensorMap, Optimizer, Scalar, serialization, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {getUid} from './backend/state';\nimport {History} from './base_callbacks';\nimport {Dataset} from './engine/dataset_stub';\nimport {Input} from './engine/input_layer';\nimport {getSourceInputs, Layer, Node, SymbolicTensor} from './engine/topology';\nimport {LayersModel, ModelCompileArgs, ModelEvaluateArgs} from './engine/training';\nimport {ModelEvaluateDatasetArgs, ModelFitDatasetArgs} from './engine/training_dataset';\nimport {ModelFitArgs} from './engine/training_tensors';\nimport {NotImplementedError, RuntimeError, ValueError} from './errors';\nimport {Shape} from './keras_format/common';\nimport {TrainingConfig} from './keras_format/training_config';\nimport {PyJsonDict} from './keras_format/types';\nimport {deserialize} from './layers/serialization';\nimport {Kwargs, NamedTensor} from './types';\nimport * as generic_utils from './utils/generic_utils';\nimport {convertPythonicToTs} from './utils/serialization_utils';\nimport {getExactlyOneShape} from './utils/types_utils';\n\n/**\n * Parses a JSON model configuration file and returns a model instance.\n *\n * ```js\n * // This example shows how to serialize a model using `toJSON()` and\n * // deserialize it as another model using `tf.models.modelFromJSON()`.\n * // Note: this example serializes and deserializes only the topology\n * // of the model; the weights of the loaded model will be different\n * // from those of the the original model, due to random weight\n * // initialization.\n * // To load the topology and weights of a model, use `tf.loadLayersModel()`.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.repeatVector({inputShape: [2], n: 4}));\n * // Serialize `model1` as a JSON object.\n * const model1JSON = model1.toJSON(null, false);\n * model1.summary();\n *\n * const model2 = await tf.models.modelFromJSON(model1JSON);\n * model2.summary();\n * ```\n *\n *  @param modelAndWeightsConfig JSON object or string encoding a model and\n *       weights configuration. It can also be only the topology JSON of the\n *       model, in which case the weights will not be loaded.\n *  @param custom_objects Optional dictionary mapping names\n *       (strings) to custom classes or functions to be\n *       considered during deserialization.\n * @returns A TensorFlow.js Layers `tf.LayersModel` instance (uncompiled).\n */\nexport async function modelFromJSON(\n    modelAndWeightsConfig: ModelAndWeightsConfig|PyJsonDict,\n    customObjects?: serialization.ConfigDict): Promise<LayersModel> {\n  if (!('modelTopology' in modelAndWeightsConfig)) {\n    modelAndWeightsConfig = {modelTopology: modelAndWeightsConfig};\n  }\n  modelAndWeightsConfig = modelAndWeightsConfig as ModelAndWeightsConfig;\n\n  let modelTopology = modelAndWeightsConfig.modelTopology;\n  if (modelTopology['model_config'] != null) {\n    // If the model-topology JSON contains a 'model_config' field, then it is\n    // a full model JSON (e.g., from `keras.Model.save()`), which contains\n    // not only the model's architecture in its 'model_config' field, but\n    // additional information such as the model's optimizer. We use only the\n    // 'model_config' field currently.\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n  const tsConfig =\n      convertPythonicToTs(modelTopology) as serialization.ConfigDict;\n  const model = deserialize(tsConfig, customObjects) as LayersModel;\n\n  if (modelAndWeightsConfig.weightsManifest != null) {\n    // Load the weight values keyed by the original tensor names in the model\n    // file that was loaded.  These should match the keys of the weight\n    // manifest.\n    const weightValues = await io.loadWeights(\n        modelAndWeightsConfig.weightsManifest, modelAndWeightsConfig.pathPrefix,\n        model.weights.map(weight => weight.originalName));\n\n    // Map the weights to the unique tensor names generated during model loading\n    const uniqueWeightValues: NamedTensorMap = {};\n    for (const weight of model.weights) {\n      uniqueWeightValues[weight.originalName] =\n          weightValues[weight.originalName];\n    }\n\n    model.loadWeights(uniqueWeightValues);\n    // Dispose temporary weight values.\n    dispose(weightValues);\n  }\n  return model;\n}\n\n/**\n * Options for loading a saved mode in TensorFlow.js format.\n */\nexport interface ModelAndWeightsConfig {\n  /**\n   * A JSON object or JSON string containing the model config.\n   *\n   * This can be either of the following two formats:\n   *   - A model archiecture-only config,  i.e., a format consistent with the\n   *     return value of`keras.Model.to_json()`.\n   *   - A full model config, containing not only model architecture, but also\n   *     training options and state, i.e., a format consistent with the return\n   *     value of `keras.models.save_model()`.\n   */\n  modelTopology: PyJsonDict;\n\n  /**\n   * A weights manifest in TensorFlow.js format.\n   */\n  weightsManifest?: io.WeightsManifestConfig;\n\n  /**\n   * Path to prepend to the paths in `weightManifest` before fetching.\n   *\n   * The path may optionally end in a slash ('/').\n   */\n  pathPrefix?: string;\n}\n\n// TODO(nielsene): Remove after: https://github.com/tensorflow/tfjs/issues/400\nexport interface ModelPredictArgs {\n  /**\n   * Optional. Batch size (Integer). If unspecified, it will default to 32.\n   */\n  batchSize?: number;\n\n  /**\n   * Optional. Verbosity mode. Defaults to false.\n   */\n  verbose?: boolean;\n}\n\n/**\n * Load a model, including its topology and optionally weights.  See the\n * Tutorial named \"How to import a Keras Model\" for usage examples.\n *\n * Example 1: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 2. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * Example 4. Load a model from an HTTP server.\n *\n * ```js\n * const model = await\n *     tf.loadLayersModel('https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. This path will be\n *      interpreted as a relative HTTP path, to which `fetch` will be used to\n *      request the model topology and weight manifest JSON.\n *      The content of the JSON file is assumed to be a JSON object with the\n *      following fields and values:\n *      - 'modelTopology': A JSON object that can be either of:\n *        1. a model architecture JSON consistent with the format of the return\n *            value of `keras.Model.to_json()`\n *        2. a full model JSON in the format of `keras.models.save_model()`.\n *      - 'weightsManifest': A TensorFlow.js weights manifest.\n *      See the Python converter function `save_model()` for more details.\n *      It is also assumed that model weights can be accessed from relative\n *      paths described by the `paths` fields in weights manifest.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A progress callback of the form:\n *     `(fraction: number) => void`. This callback can be used to monitor the\n *     model-loading process.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n */\nexport async function loadLayersModelInternal(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (typeof pathOrIOHandler === 'string') {\n    const handlers = io.getLoadHandlers(pathOrIOHandler, options);\n    if (handlers.length === 0) {\n      // For backward compatibility: if no load handler can be found,\n      // assume it is a relative http path.\n      // TODO(cais): Reformat the args into a single `LoadOptions` once the core\n      // is refactored.\n      handlers.push(io.browserHTTPRequest(pathOrIOHandler, options));\n    } else if (handlers.length > 1) {\n      throw new ValueError(\n          `Found more than one (${handlers.length}) load handlers for ` +\n          `URL '${pathOrIOHandler}'`);\n    }\n    pathOrIOHandler = handlers[0];\n  }\n  return loadLayersModelFromIOHandler(pathOrIOHandler, undefined, options);\n}\n\n/**\n * Load a model and optionally its weights, using an IOHandler object.\n *\n * @param handler The instance of `IOHandler` to be used during the model\n *   loading.\n * @param customObjects Any optional custom objects to be used during model\n *   loading.\n * @param strict Whether the weight loading will be done in strict mode.\n *   Default: `true`.\n */\nexport async function loadLayersModelFromIOHandler(\n    handler: io.IOHandler, customObjects?: serialization.ConfigDict,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  if (handler.load == null) {\n    throw new ValueError(\n        'Cannot proceed with model loading because the IOHandler provided ' +\n        'does not have the `load` method implemented.');\n  }\n  const artifacts = await handler.load();\n  let modelTopology = artifacts.modelTopology as PyJsonDict;\n  if (modelTopology['model_config'] != null) {\n    modelTopology = modelTopology['model_config'] as PyJsonDict;\n  }\n\n  const strict = options.strict == null ? true : options.strict;\n  // If weights are provided and the weight-loading mode is strict, use\n  // fast weight initialization. This skips costly initializers such as\n  // 'orthogonal' and saves unnecessary computation in cases where\n  // the initialized weight values will immediately be overwritten by\n  // loaded weight values.\n  const fastWeightInit =\n      artifacts.weightData != null && artifacts.weightSpecs != null && strict;\n  const model =\n      deserialize(\n          convertPythonicToTs(modelTopology) as serialization.ConfigDict,\n          customObjects, fastWeightInit) as LayersModel;\n\n  const trainingConfig = artifacts.trainingConfig as TrainingConfig;\n  if (trainingConfig != null) {\n    model.loadTrainingConfig(trainingConfig);\n  }\n  if (artifacts.userDefinedMetadata != null) {\n    model.setUserDefinedMetadata(artifacts.userDefinedMetadata);\n  }\n\n  // If weightData is present, load the weights into the model.\n  if (artifacts.weightData != null) {\n    // Loading weights requires weightSpecs.\n    if (artifacts.weightSpecs == null) {\n      throw new ValueError(\n          'LayersModel artifacts contains weight data, but not weight specs. ' +\n          'Therefore loading of weights cannot proceed.');\n    }\n\n    const {modelWeights, optimizerWeights} = decodeModelAndOptimizerWeights(\n        artifacts.weightData, artifacts.weightSpecs);\n    model.loadWeights(modelWeights, strict);\n\n    if (model.optimizer != null && optimizerWeights.length > 0) {\n      await model.optimizer.setWeights(optimizerWeights);\n    }\n\n    // Dispose temporary weight values.\n    dispose(modelWeights);\n    dispose(optimizerWeights.map(w => w.tensor));\n  }\n  return model;\n}\n\nfunction decodeModelAndOptimizerWeights(\n    buffer: ArrayBuffer, specs: io.WeightsManifestEntry[]):\n    {modelWeights: NamedTensorMap, optimizerWeights: NamedTensor[]} {\n  const name2Tensor = io.decodeWeights(buffer, specs);\n  const modelWeights: NamedTensorMap = {};\n  const optimizerWeights: NamedTensor[] = [];\n  specs.forEach(spec => {\n    if (spec.group === 'optimizer') {\n      optimizerWeights.push({name: spec.name, tensor: name2Tensor[spec.name]});\n    } else {\n      modelWeights[spec.name] = name2Tensor[spec.name];\n    }\n  });\n  return {modelWeights, optimizerWeights};\n}\n\n/**\n * Configuration for a Sequential model.\n */\nexport interface SequentialArgs {\n  /** Stack of layers for the model. */\n  layers?: Layer[];\n\n  /** The name of this model. */\n  name?: string;\n}\n\n/**\n * A model with a stack of layers, feeding linearly from one to the next.\n *\n * `tf.sequential` is a factory function that creates an instance of\n * `tf.Sequential`.\n *\n * ```js\n *  // Define a model for linear regression.\n *  const model = tf.sequential();\n *  model.add(tf.layers.dense({units: 1, inputShape: [1]}));\n *\n *  // Prepare the model for training: Specify the loss and the optimizer.\n *  model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n *\n *  // Generate some synthetic data for training.\n *  const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);\n *  const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);\n *\n *  // Train the model using the data then do inference on a data point the\n *  // model hasn't seen:\n *  await model.fit(xs, ys);\n *  model.predict(tf.tensor2d([5], [1, 1])).print();\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class Sequential extends LayersModel {\n  /** @nocollapse */\n  static className = 'Sequential';\n  private model: LayersModel;\n  constructor(args?: SequentialArgs) {\n    super({inputs: [], outputs: []});\n    args = args || {};\n\n    this.trainable = true;\n    this.built = false;\n\n    // Set model name.\n    this.name = (args.name != null) ? args.name : getUid('sequential_');\n\n    // Add to the model any layers passed to the constructor.\n    if (args.layers != null) {\n      for (const layer of args.layers) {\n        this.add(layer);\n      }\n    }\n  }\n\n  // Helper function to Sequential.add  Throws if the new output shape will be\n  // invalid.\n  private checkShape(layer: Layer) {\n    const shape = layer.inboundNodes[0].outputTensors[0].shape;\n    if (shape.some(x => x < 0)) {\n      throw new ValueError(\n          'Negative dimension size caused by adding layer ' +\n          `${layer.name} with input shape [` +\n          `${layer.inboundNodes[0].inputTensors[0].shape}]`);\n    }\n  }\n\n  /**\n   * Adds a layer instance on top of the layer stack.\n   *\n   * ```js\n   *  const model = tf.sequential();\n   *  model.add(tf.layers.dense({units: 8, inputShape: [1]}));\n   *  model.add(tf.layers.dense({units: 4, activation: 'relu6'}));\n   *  model.add(tf.layers.dense({units: 1, activation: 'relu6'}));\n   *  // Note that the untrained model is random at this point.\n   *  model.predict(tf.randomNormal([10, 1])).print();\n   * ```\n   * @param layer Layer instance.\n   *\n   * @exception ValueError In case the `layer` argument does not know its\n   * input shape.\n   * @exception ValueError In case the `layer` argument has multiple output\n   *   tensors, or is already connected somewhere else (forbidden in\n   *   `Sequential` models).\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  add(layer: Layer): void {\n    const isLayerModelInstance =\n        layer instanceof Sequential || layer instanceof LayersModel;\n    let modelLayer: LayersModel;\n    if (isLayerModelInstance) {\n      modelLayer = layer as LayersModel;\n      if (modelLayer.outputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      if (modelLayer.inputs.length !== 1) {\n        throw new ValueError(\n            'All layers in a Sequential model ' +\n            'should have a single input tensor. ' +\n            'For multi-input layers, ' +\n            'use the functional API.');\n      }\n    }\n\n    if (this.outputs.length === 0) {\n      // first layer in model: check that it is an input layer\n      if (layer.inboundNodes.length === 0) {\n        // create an input layer\n        if (layer.batchInputShape == null) {\n          throw new ValueError(\n              'The first layer in a Sequential model must ' +\n              'get an `inputShape` or `batchInputShape` argument.');\n        }\n        // Instantiate the input layer.\n        const x = Input({\n          batchShape: layer.batchInputShape,\n          dtype: layer.dtype,\n          name: layer.name + '_input'\n        });\n        // This will build the current layer and create the node connecting\n        // the current layer to the input layer we just created.\n        layer.apply(x);\n      }\n\n      if (isLayerModelInstance) {\n        this.outputs = modelLayer.outputs;\n        this.inputs = modelLayer.inputs;\n      } else {\n        if (layer.inboundNodes.length !== 1) {\n          throw new ValueError(\n              'A layer added to a Sequential model must not already be ' +\n              `connected somewhere else. LayersModel received layer ${\n                  layer.name} ` +\n              `which has ${layer.inboundNodes.length} pre-existing inbound ` +\n              'connections.');\n        }\n\n        if (layer.inboundNodes[0].outputTensors.length !== 1) {\n          throw new ValueError(\n              'All layers in a Sequential model ' +\n              'should have a single output tensor. ' +\n              'For multi-output layers, ' +\n              'use the functional API.');\n        }\n        this.checkShape(layer);\n        this.outputs = [layer.inboundNodes[0].outputTensors[0]];\n        this.inputs = getSourceInputs(this.outputs[0]);\n      }\n\n      this.inboundNodes = [];\n      // We create an input node, which we will keep updated\n      // as we add more layers.\n      // (This call has side effects.)\n      // tslint:disable-next-line:no-unused-expression\n      new Node({\n        outboundLayer: this,\n        inboundLayers: [],\n        nodeIndices: [],\n        tensorIndices: [],\n        inputTensors: this.inputs,\n        outputTensors: this.outputs,\n        // no model-level masking for now\n        inputMasks: generic_utils.pyListRepeat(null, this.inputs.length),\n        outputMasks: [null],\n        inputShapes: this.inputs.map(x => x.shape),\n        outputShapes: this.outputs[0].shape\n      });\n    } else {\n      const outputTensor = layer.apply(this.outputs[0]);\n      if (Array.isArray(outputTensor)) {\n        throw new TypeError(\n            'All layers in a Sequential model ' +\n            'should have a single output tensor. ' +\n            'For multi-output layers, ' +\n            'use the functional API.');\n      }\n      this.checkShape(layer);\n      this.outputs = [outputTensor as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n\n    this.layers.push(layer);\n    this.built = false;\n  }\n\n  /**\n   * Removes the last layer in the model.\n   *\n   * @exception TypeError if there are no layers in the model.\n   */\n  pop(): void {\n    if (this.layers.length === 0) {\n      throw new TypeError('There are no layers in the model.');\n    }\n\n    this.layers.pop();\n    if (this.layers.length === 0) {\n      this.outputs = [];\n      this.inboundNodes = [];\n      this.outboundNodes = [];\n    } else {\n      const lastLayerIndex = this.layers.length - 1;\n      this.layers[lastLayerIndex].outboundNodes = [];\n      this.outputs = [this.layers[lastLayerIndex].output as SymbolicTensor];\n      // update self.inbound_nodes\n      this.inboundNodes[0].outputTensors = this.outputs;\n      this.inboundNodes[0].outputShapes = [this.outputs[0].shape];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.call(inputs, kwargs);\n  }\n\n  build(inputShape?: Shape|Shape[]) {\n    // Call `getExactlyOneShape` without using its return value,\n    // to verify that exactly one input shape is provided.\n    getExactlyOneShape(inputShape);\n\n    if (this.inputs.length === 0 || this.outputs.length === 0) {\n      throw new TypeError(\n          'Sequential model cannot be built: model is empty.' +\n          ' Add some layers first.');\n    }\n    // actually create the model\n    this.model = new LayersModel({\n      inputs: this.inputs,\n      outputs: this.outputs[0],\n      name: this.name + '_model'\n    });\n    this.model.trainable = this.trainable;\n\n    // mirror model attributes\n    this.supportsMasking = this.model.supportsMasking;\n    // TODO(michaelterry): Add caches\n    this.inputLayers = this.model.inputLayers;\n    this.inputLayersNodeIndices = this.model.inputLayersNodeIndices;\n    this.inputLayersTensorIndices = this.model.inputLayersTensorIndices;\n    this.outputLayers = this.model.outputLayers;\n    this.outputLayersNodeIndices = this.model.outputLayersNodeIndices;\n    this.outputLayersTensorIndices = this.model.outputLayersTensorIndices;\n    this.nodesByDepth = this.model.nodesByDepth;\n    this.containerNodes = this.model.containerNodes;\n    this.outputNames = this.model.outputNames;\n    this.inputNames = this.model.inputNames;\n    // TODO(michaelterry): Add feedInputNames, feedInputs, if needed.\n    // TODO(michaelterry): Add callbackModel if needed.\n    this.built = true;\n  }\n\n  countParams(): number {\n    if (!this.built) {\n      this.build();\n    }\n    return super.countParams();\n  }\n\n  /**\n   * Print a text summary of the Sequential model's layers.\n   *\n   * The summary includes\n   * - Name and type of all layers that comprise the model.\n   * - Output shape(s) of the layers\n   * - Number of weight parameters of each layer\n   * - The total number of trainable and non-trainable parameters of the\n   * model.\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(\n   *     tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'}));\n   * model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n   *\n   * model.summary();\n   * ```\n   *\n   * @param lineLength Custom line length, in number of characters.\n   * @param positions Custom widths of each of the columns, as either\n   *   fractions of `lineLength` (e.g., `[0.5, 0.75, 1]`) or absolute number\n   *   of characters (e.g., `[30, 50, 65]`). Each number corresponds to\n   *   right-most (i.e., ending) position of a column.\n   * @param printFn Custom print function. Can be used to replace the default\n   *   `console.log`. For example, you can use `x => {}` to mute the printed\n   *   messages in the console.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  summary(\n      lineLength?: number, positions?: number[],\n      printFn:\n          // tslint:disable-next-line:no-any\n      (message?: any, ...optionalParams: any[]) => void = console.log) {\n    if (!this.built) {\n      this.build();\n    }\n    super.summary(lineLength, positions, printFn);\n  }\n\n  /**\n   * Sets the weights of the model.\n   *\n   * @param weights Should be a list of Tensors with shapes and types matching\n   *   the output of `model.getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    if (this.model == null) {\n      this.build();\n    }\n    this.model.setWeights(weights);\n  }\n\n  /**\n   * Returns the loss value & metrics values for the model in test mode.\n   *\n   * Loss and metrics are specified during `compile()`, which needs to happen\n   * before calls to `evaluate()`.\n   *\n   * Computation is done in batches.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   * });\n   * result.print();\n   * ```\n   *\n   * @param x `tf.Tensor` of test data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple inputs.\n   * @param y `tf.Tensor` of target data, or an `Array` of `tf.Tensor`s if the\n   * model has multiple outputs.\n   * @param args A `ModelEvaluateConfig`, containing optional fields.\n   *\n   * @return `Scalar` test loss (if the model has a single output and no\n   *   metrics) or `Array` of `Scalar`s (if the model has multiple outputs\n   *   and/or metrics). The attribute `model.metricsNames`\n   *   will give you the display labels for the scalar outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  evaluate(\n      x: Tensor|Tensor[], y: Tensor|Tensor[],\n      args: ModelEvaluateArgs = {}): Scalar|Scalar[] {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluate(x, y, args);\n  }\n\n  // TODO(cais): Add code snippet below once real dataset objects are\n  //   available.\n  /**\n   * Evaluate model using a dataset object.\n   *\n   * Note: Unlike `evaluate()`, this method is asynchronous (`async`);\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected\n   *   to generate a dataset iterator object, the `next()` method of which\n   *   is expected to produce data batches for evaluation. The return value\n   *   of the `next()` call ought to contain a boolean `done` field and a\n   *   `value` field. The `value` field is expected to be an array of two\n   *   `tf.Tensor`s or an array of two nested `tf.Tensor` structures. The former\n   *   case is for models with exactly one input and one output (e.g..\n   *   a sequential model). The latter case is for models with multiple\n   *   inputs and/or multiple outputs. Of the two items in the array, the\n   *   first is the input feature(s) and the second is the output target(s).\n   * @param args A configuration object for the dataset-based evaluation.\n   * @returns Loss and metric values as an Array of `Scalar` objects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async evaluateDataset(dataset: Dataset<{}>, args: ModelEvaluateDatasetArgs):\n      Promise<Scalar|Scalar[]> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before being used.');\n    }\n    return this.model.evaluateDataset(dataset, args);\n  }\n\n  /**\n   * Generates output predictions for the input samples.\n   *\n   * Computation is done in batches.\n   *\n   * Note: the \"step\" mode of predict() is currently not supported.\n   *   This is because the TensorFow.js core backend is imperative only.\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.predict(tf.ones([2, 10])).print();\n   * ```\n   *\n   * @param x The input data, as a Tensor, or an `Array` of `tf.Tensor`s if\n   *   the model has multiple inputs.\n   * @param conifg A `ModelPredictConfig` object containing optional fields.\n   *\n   * @return `tf.Tensor`(s) of predictions.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and the model's expectations, or in case a stateful model receives a\n   *   number of samples that is not a multiple of the batch size.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(x: Tensor|Tensor[], args: ModelPredictArgs = {}): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predict(x, args);\n  }\n\n  /**\n   * Returns predictions for a single batch of samples.\n   *\n   * @param x: Input samples, as a Tensor, or list of Tensors (if the model\n   *   has multiple inputs).\n   * @return Tensor(s) of predictions\n   */\n  predictOnBatch(x: Tensor): Tensor|Tensor[] {\n    if (this.model == null) {\n      this.build();\n    }\n    return this.model.predictOnBatch(x);\n  }\n\n  /**\n   * See `LayersModel.compile`.\n   *\n   * @param args\n   */\n  compile(args: ModelCompileArgs): void {\n    this.build();\n    this.model.compile(args);\n    this.optimizer_ = this.model.optimizer;\n    // tslint:disable-next-line:no-any\n    this.isOptimizerOwned = (this.model as any).isOptimizerOwned;\n    this.loss = this.model.loss;\n    this.metrics = this.model.metrics;\n    // TODO(cais): Add this.lossWeights, this.sampleWeightMode,\n    //   this.weightedMetrics, this.targets.\n    this.metricsTensors = this.model.metricsTensors;\n    this.metricsNames = this.model.metricsNames;\n    // TODO(cais): Add sampleWeights.\n  }\n\n  get optimizer(): Optimizer {\n    return this.model == null ? undefined : this.model.optimizer;\n  }\n\n  set optimizer(optimizer: Optimizer) {\n    this.model.optimizer = optimizer;\n  }\n\n  /**\n   * Trains the model for a fixed number of epochs (iterations on a dataset).\n   *\n   * ```js\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [10]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {\n   *   batchSize: 4,\n   *   epochs: 3\n   * });\n   * console.log(history.history.loss[0]);\n   * ```\n   *\n   * @param x `tf.Tensor` of training data, or an array of `tf.Tensor`s if the\n   * model has multiple inputs. If all inputs in the model are named, you can\n   * also pass a dictionary mapping input names to `tf.Tensor`s.\n   * @param y `tf.Tensor` of target (label) data, or an array of `tf.Tensor`s if\n   * the model has multiple outputs. If all outputs in the model are named, you\n   *  can also pass a dictionary mapping output names to `tf.Tensor`s.\n   * @param args  A `ModelFitConfig`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @exception ValueError In case of mismatch between the provided input data\n   *   and what the model expects.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async fit(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      args: ModelFitArgs = {}): Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fit(x, y, args);\n  }\n\n  /**\n   * Trains the model using a dataset object.\n   *\n   * ```js\n   * const xArray = [\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   *   [1, 1, 1, 1, 1, 1, 1, 1, 1],\n   * ];\n   * const yArray = [1, 1, 1, 1];\n   * // Create a dataset from the JavaScript array.\n   * const xDataset = tf.data.array(xArray);\n   * const yDataset = tf.data.array(yArray);\n   * // Zip combines the `x` and `y` Datasets into a single Dataset, the\n   * // iterator of which will return an object containing of two tensors,\n   * // corresponding to `x` and `y`.  The call to `batch(4)` will bundle\n   * // four such samples into a single object, with the same keys now pointing\n   * // to tensors that hold 4 examples, organized along the batch dimension.\n   * // The call to `shuffle(4)` causes each iteration through the dataset to\n   * // happen in a different order.  The size of the shuffle window is 4.\n   * const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset})\n   *     .batch(4)\n   *     .shuffle(4);\n   * const model = tf.sequential({\n   *   layers: [tf.layers.dense({units: 1, inputShape: [9]})]\n   * });\n   * model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});\n   * const history = await model.fitDataset(xyDataset, {\n   *   epochs: 4,\n   *   callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)}\n   * });\n   * ```\n   *\n   * @param dataset A dataset object. Its `iterator()` method is expected to\n   *   generate a dataset iterator object, the `next()` method of which is\n   *   expected to produce data batches for evaluation. The return value of the\n   *   `next()` call ought to contain a boolean `done` field and a `value`\n   *   field.\n   *\n   *   The `value` field is expected to be an object of with fields\n   *   `xs` and `ys`, which point to the feature tensor and the target tensor,\n   *   respectively. This case is for models with exactly one input and one\n   *   output (e.g.. a sequential model). For example:\n   *   ```js\n   *   {value: {xs: xsTensor, ys: ysTensor}, done: false}\n   *   ```\n   *\n   *   If the model has multiple inputs, the `xs` field of `value` should\n   *   be an object mapping input names to their respective feature tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: {\n   *         input_1: xsTensor1,\n   *         input_2: xsTensor2\n   *       },\n   *       ys: ysTensor\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   *   If the model has multiple outputs, the `ys` field of `value` should\n   *   be an object mapping output names to their respective target tensors.\n   *   For example:\n   *   ```js\n   *   {\n   *     value: {\n   *       xs: xsTensor,\n   *       ys: {\n   *         output_1: ysTensor1,\n   *         output_2: ysTensor2\n   *       },\n   *     },\n   *     done: false\n   *   }\n   *   ```\n   * @param args A `ModelFitDatasetArgs`, containing optional fields.\n   *\n   * @return A `History` instance. Its `history` attribute contains all\n   *   information collected during training.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async fitDataset<T>(dataset: Dataset<T>, args: ModelFitDatasetArgs<T>):\n      Promise<History> {\n    if (!this.built) {\n      throw new RuntimeError(\n          'The model needs to be compiled before ' +\n          'being used.');\n    }\n    return this.model.fitDataset(dataset, args);\n  }\n\n  /**\n   * Runs a single gradient update on a single batch of data.\n   *\n   * This method differs from `fit()` and `fitDataset()` in the following\n   * regards:\n   *   - It operates on exactly one batch of data.\n   *   - It returns only the loss and matric values, instead of\n   *     returning the batch-by-batch loss and metric values.\n   *   - It doesn't support fine-grained options such as verbosity and\n   *     callbacks.\n   *\n   * @param x Input data. It could be one of the following:\n   *   - A `tf.Tensor`, or an Array of `tf.Tensor`s (in case the model has\n   *     multiple inputs).\n   *   - An Object mapping input names to corresponding `tf.Tensor` (if the\n   *     model has named inputs).\n   * @param y Target darta. It could be either a `tf.Tensor` a multiple\n   *   `tf.Tensor`s. It should be consistent with `x`.\n   * @returns Training loss or losses (in case the model has\n   *   multiple outputs), along with metrics (if any), as numbers.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async trainOnBatch(\n      x: Tensor|Tensor[]|{[inputName: string]: Tensor},\n      y: Tensor|Tensor[]|\n      {[inputName: string]: Tensor}): Promise<number|number[]> {\n    return this.model.trainOnBatch(x, y);\n  }\n\n  /* See parent class for JsDoc */\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict,\n      fastWeightInit = false): T {\n    let configArray: serialization.ConfigDictArray;\n    let extraModelConfig: serialization.ConfigDict = {};\n    if (config instanceof Array) {\n      if (!(config[0].className != null) ||\n          config[0]['className'] === 'Merge') {\n        throw new ValueError('Legacy serialization format not supported yet.');\n      }\n      configArray = config;\n    } else {\n      util.assert(\n          config['layers'] != null,\n          () =>\n              `When the config data for a Sequential model is not an Array, ` +\n              `it must be an Object that contains the 'layers' field.`);\n      configArray = config['layers'] as serialization.ConfigDictArray;\n      delete config['layers'];\n      extraModelConfig = config;\n    }\n\n    const model = new cls(extraModelConfig);\n    if (!(model instanceof Sequential)) {\n      throw new NotImplementedError(\n          `Sequential.fromConfig called on non-Sequential input: ${model}`);\n    }\n    for (const conf of configArray) {\n      const customObjects: serialization.ConfigDict = undefined;\n      const layer = deserialize(\n                        conf as serialization.ConfigDict, customObjects,\n                        fastWeightInit) as Layer;\n      if (fastWeightInit) {\n        layer.setFastWeightInitDuringBuild(true);\n      }\n      model.add(layer);\n    }\n    return model;\n  }\n\n  /**\n   * Setter used for force stopping of LayersModel.fit() (i.e., training).\n   *\n   * Example:\n   *\n   * ```js\n   * const model = tf.sequential();\n   * model.add(tf.layers.dense({units: 1, inputShape: [10]}));\n   * model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});\n   * const xs = tf.ones([8, 10]);\n   * const ys = tf.zeros([8, 1]);\n   *\n   * const history = await model.fit(xs, ys, {\n   *   epochs: 10,\n   *   callbacks: {\n   *     onEpochEnd: async (epoch, logs) => {\n   *       if (epoch === 2) {\n   *         model.stopTraining = true;\n   *       }\n   *     }\n   *   }\n   * });\n   *\n   * // There should be only 3 values in the loss array, instead of 10 values,\n   * // due to the stopping after 3 epochs.\n   * console.log(history.history.loss);\n   * ```\n   */\n  set stopTraining(stop: boolean) {\n    // TODO(cais): When refactoring to remove the composition pattern happens,\n    // remove this method overriding.\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot set the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    this.model.stopTraining = stop;\n  }\n\n  get stopTraining(): boolean {\n    if (this.model == null) {\n      throw new ValueError(\n          'Cannot get the stopTraining property of a sequential model before ' +\n          'it is compiled.');\n    }\n    return this.model.stopTraining;\n  }\n\n  // TODO(cais): Override get trainableWeights() here\n\n  // tslint:disable-next-line:no-any\n  getConfig(): any {\n    // NOTE(cais): We override the return type of getConfig() to `any` here,\n    //   because the `Sequential` class is a special case among `Container`\n    //   subtypes in that its getConfig() method returns an Array (not a\n    //   dict).\n    const layers: serialization.ConfigDict[] = [];\n    for (const layer of this.layers) {\n      const dict: serialization.ConfigDict = {};\n      dict['className'] = layer.getClassName();\n      dict['config'] = layer.getConfig();\n      layers.push(dict);\n    }\n    return {name: this.name, layers};\n  }\n}\nserialization.registerClass(Sequential);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Exported functions.\n */\n\nimport {io} from '@tensorflow/tfjs-core';\n\nimport {BaseCallbackConstructor, CallbackConstructorRegistry} from './base_callbacks';\nimport {ContainerArgs} from './engine/container';\nimport {Input, InputConfig,} from './engine/input_layer';\nimport {SymbolicTensor} from './engine/topology';\nimport {LayersModel} from './engine/training';\nimport {loadLayersModelInternal, Sequential, SequentialArgs} from './models';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// LayersModel and related factory methods.\n\n/**\n * A model is a data structure that consists of `Layers` and defines inputs\n * and outputs.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.model` is more generic, supporting an arbitrary graph (without\n * cycles) of layers. `tf.sequential` is less generic and supports only a linear\n * stack of layers.\n *\n * When creating a `tf.LayersModel`, specify its input(s) and output(s). Layers\n * are used to wire input(s) to output(s).\n *\n * For example, the following code snippet defines a model consisting of\n * two `dense` layers, with 10 and 4 units, respectively.\n *\n * ```js\n * // Define input, which has a size of 5 (not including batch dimension).\n * const input = tf.input({shape: [5]});\n *\n * // First dense layer uses relu activation.\n * const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});\n * // Second dense layer uses softmax activation.\n * const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});\n *\n * // Obtain the output symbolic tensor by applying the layers on the input.\n * const output = denseLayer2.apply(denseLayer1.apply(input));\n *\n * // Create the model based on the inputs.\n * const model = tf.model({inputs: input, outputs: output});\n *\n * // The model can be used for training, evaluation and prediction.\n * // For example, the following line runs prediction with the model on\n * // some fake data.\n * model.predict(tf.ones([2, 5])).print();\n * ```\n * See also:\n *   `tf.sequential`, `tf.loadLayersModel`.\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function model(args: ContainerArgs): LayersModel {\n  return new LayersModel(args);\n}\n\n/**\n * Creates a `tf.Sequential` model.  A sequential model is any model where the\n * outputs of one layer are the inputs to the next layer, i.e. the model\n * topology is a simple 'stack' of layers, with no branching or skipping.\n *\n * This means that the first layer passed to a `tf.Sequential` model should have\n * a defined input shape. What that means is that it should have received an\n * `inputShape` or `batchInputShape` argument, or for some type of layers\n * (recurrent, Dense...) an `inputDim` argument.\n *\n * The key difference between `tf.model` and `tf.sequential` is that\n * `tf.sequential` is less generic, supporting only a linear stack of layers.\n * `tf.model` is more generic and supports an arbitrary graph (without\n * cycles) of layers.\n *\n * Examples:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have an input shape defined.\n * model.add(tf.layers.dense({units: 32, inputShape: [50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output, which equals\n * // `[null, 4]`. The 1st dimension is the undetermined batch dimension; the\n * // 2nd is the output size of the model's last layer.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * It is also possible to specify a batch size (with potentially undetermined\n * batch dimension, denoted by \"null\") for the first layer using the\n * `batchInputShape` key. The following example is equivalent to the above:\n *\n * ```js\n * const model = tf.sequential();\n *\n * // First layer must have a defined input shape\n * model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));\n * // Afterwards, TF.js does automatic shape inference.\n * model.add(tf.layers.dense({units: 4}));\n *\n * // Inspect the inferred shape of the model's output.\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * You can also use an `Array` of already-constructed `Layer`s to create\n * a `tf.Sequential` model:\n *\n * ```js\n * const model = tf.sequential({\n *   layers: [tf.layers.dense({units: 32, inputShape: [50]}),\n *            tf.layers.dense({units: 4})]\n * });\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Models', subheading: 'Creation'}\n */\nexport function sequential(config?: SequentialArgs): Sequential {\n  return new Sequential(config);\n}\n\n/**\n * Load a model composed of Layer objects, including its topology and optionally\n * weights. See the Tutorial named \"How to import a Keras Model\" for usage\n * examples.\n *\n * This method is applicable to:\n *\n * 1. Models created with the `tf.layers.*`, `tf.sequential`, and\n * `tf.model` APIs of TensorFlow.js and later saved with the\n * `tf.LayersModel.save` method.\n * 2. Models converted from Keras or TensorFlow tf.keras using the\n * [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).\n *\n * This mode is *not* applicable to TensorFlow `SavedModel`s or their converted\n * forms. For those models, use `tf.loadGraphModel`.\n *\n * Example 1. Load a model from an HTTP server.\n *\n * ```js\n * const model = await tf.loadLayersModel(\n *     'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json');\n * model.summary();\n * ```\n *\n * Example 2: Save `model`'s topology and weights to browser [local\n * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('localstorage://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('localstorage://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 3. Saving `model`'s topology and weights to browser\n * [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API);\n * then load it back.\n *\n * ```js\n * const model = tf.sequential(\n *     {layers: [tf.layers.dense({units: 1, inputShape: [3]})]});\n * console.log('Prediction from original model:');\n * model.predict(tf.ones([1, 3])).print();\n *\n * const saveResults = await model.save('indexeddb://my-model-1');\n *\n * const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1');\n * console.log('Prediction from loaded model:');\n * loadedModel.predict(tf.ones([1, 3])).print();\n * ```\n *\n * Example 4. Load a model from user-selected files from HTML\n * [file input\n * elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).\n *\n * ```js\n * // Note: this code snippet will not work without the HTML elements in the\n * //   page\n * const jsonUpload = document.getElementById('json-upload');\n * const weightsUpload = document.getElementById('weights-upload');\n *\n * const model = await tf.loadLayersModel(\n *     tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));\n * ```\n *\n * @param pathOrIOHandler Can be either of the two formats\n *   1. A string path to the `ModelAndWeightsConfig` JSON describing\n *      the model in the canonical TensorFlow.js format. For file://\n *      (tfjs-node-only), http:// and https:// schemas, the path can be\n *      either absolute or relative.\n *   2. An `tf.io.IOHandler` object that loads model artifacts with its `load`\n *      method.\n * @param options Optional configuration arguments for the model loading,\n *   including:\n *   - `strict`: Require that the provided weights exactly match those required\n *     by the layers.  Default true.  Passing false means that both extra\n *     weights and missing weights will be silently ignored.\n *   - `onProgress`: A function of the signature `(fraction: number) => void',\n *     that can be used as the progress callback for the model loading.\n * @returns A `Promise` of `tf.LayersModel`, with the topology and weights\n *     loaded.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport function loadLayersModel(\n    pathOrIOHandler: string|io.IOHandler,\n    options?: io.LoadOptions): Promise<LayersModel> {\n  if (options == null) {\n    options = {};\n  }\n  return loadLayersModelInternal(pathOrIOHandler, options);\n}\n\n/**\n * Used to instantiate an input to a model as a `tf.SymbolicTensor`.\n *\n * Users should call the `input` factory function for\n * consistency with other generator functions.\n *\n * Example:\n *\n * ```js\n * // Defines a simple logistic regression model with 32 dimensional input\n * // and 3 dimensional output.\n * const x = tf.input({shape: [32]});\n * const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x);\n * const model = tf.model({inputs: x, outputs: y});\n * model.predict(tf.ones([2, 32])).print();\n * ```\n *\n * Note: `input` is only necessary when using `model`. When using\n * `sequential`, specify `inputShape` for the first layer or use `inputLayer`\n * as the first layer.\n *\n * @doc {heading: 'Models', subheading: 'Inputs'}\n */\nexport function input(config: InputConfig): SymbolicTensor {\n  return Input(config);\n}\n\nexport function registerCallbackConstructor(\n    verbosityLevel: number,\n    callbackConstructor: BaseCallbackConstructor): void {\n  CallbackConstructorRegistry.registerCallbackConstructor(\n      verbosityLevel, callbackConstructor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n// Layer activation functions\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {ActivationIdentifier} from './keras_format/activation_config';\nimport {deserializeKerasObject} from './utils/generic_utils';\n\n/**\n * Base class for Activations.\n *\n * Special note: due to cross-language compatibility reasons, the\n * static readonly className field in this family of classes must be set to\n * the initialLowerCamelCase name of the activation.\n */\nexport abstract class Activation extends serialization.Serializable {\n  abstract apply(tensor: Tensor, axis?: number): Tensor;\n  getConfig(): serialization.ConfigDict {\n    return {};\n  }\n}\n\n/**\n * Exponential linear unit (ELU).\n * Reference: https://arxiv.org/abs/1511.07289\n */\nexport class Elu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'elu';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x: Input.\n   * @param alpha: Scaling factor the negative section.\n   * @return Output of the ELU activation.\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return K.elu(x, alpha);\n  }\n}\nserialization.registerClass(Elu);\n\n/**\n * Scaled Exponential Linear Unit. (Klambauer et al., 2017).\n * Reference: Self-Normalizing Neural Networks, https://arxiv.org/abs/1706.02515\n * Notes:\n *   - To be used together with the initialization \"lecunNormal\".\n *   - To be used together with the dropout variant \"AlphaDropout\".\n */\nexport class Selu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'selu';\n  apply(x: Tensor): Tensor {\n    return tfc.selu(x);\n  }\n}\nserialization.registerClass(Selu);\n\n/**\n *  Rectified linear unit\n */\nexport class Relu extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu';\n  apply(x: Tensor): Tensor {\n    return tfc.relu(x);\n  }\n}\nserialization.registerClass(Relu);\n\n/**\n * Rectified linear unit activation maxing out at 6.0.\n */\nexport class Relu6 extends Activation {\n  /** @nocollapse */\n  static readonly className = 'relu6';\n  apply(x: Tensor): Tensor {\n    return tidy(() => tfc.minimum(6.0, tfc.relu(x)));\n  }\n}\nserialization.registerClass(Relu6);\n\n//* Linear activation (no-op) */\nexport class Linear extends Activation {\n  /** @nocollapse */\n  static readonly className = 'linear';\n  apply(x: Tensor): Tensor {\n    return x;\n  }\n}\nserialization.registerClass(Linear);\n\n/**\n * Sigmoid activation function.\n */\nexport class Sigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'sigmoid';\n  apply(x: Tensor): Tensor {\n    return tfc.sigmoid(x);\n  }\n}\nserialization.registerClass(Sigmoid);\n\n/**\n * Segment-wise linear approximation of sigmoid.\n */\nexport class HardSigmoid extends Activation {\n  /** @nocollapse */\n  static readonly className = 'hardSigmoid';\n  apply(x: Tensor): Tensor {\n    return K.hardSigmoid(x);\n  }\n}\nserialization.registerClass(HardSigmoid);\n\n/**\n * Softplus activation function.\n */\nexport class Softplus extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softplus';\n  apply(x: Tensor): Tensor {\n    return tfc.softplus(x);\n  }\n}\nserialization.registerClass(Softplus);\n\n/**\n * Softsign activation function.\n */\nexport class Softsign extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softsign';\n  apply(x: Tensor): Tensor {\n    return K.softsign(x);\n  }\n}\nserialization.registerClass(Softsign);\n\n/**\n * Hyperbolic tangent function.\n */\nexport class Tanh extends Activation {\n  /** @nocollapse */\n  static readonly className = 'tanh';\n  apply(x: Tensor): Tensor {\n    return tfc.tanh(x);\n  }\n}\nserialization.registerClass(Tanh);\n\n/**\n * Softmax activation function\n */\nexport class Softmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'softmax';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.softmax(x, axis);\n  }\n}\nserialization.registerClass(Softmax);\n\n/**\n * Log softmax activation function\n */\nexport class LogSoftmax extends Activation {\n  /** @nocollapse */\n  static readonly className = 'logSoftmax';\n  /**\n   * Calculate the activation function of log softmax:\n   * log( exp(x_i) / sum(exp(x)) )\n   *\n   * @param x Tensor.\n   * @param axis Integer, axis along which the softmax normalization is applied.\n   * Invalid if < 2, as softmax across 1 (the batch dimension) is assumed to be\n   * an error.\n   *\n   * @returns a Tensor of the same shape as x\n   *\n   * @throws ValueError: In case `dim(x) < 2`.\n   */\n  apply(x: Tensor, axis: number = (-1)): Tensor {\n    return tfc.logSoftmax(x, axis);\n  }\n}\nserialization.registerClass(LogSoftmax);\n\n/**\n * Swish activation function\n */\nexport class Swish extends Activation {\n  /** @nocollapse */\n  static readonly className = 'swish';\n  /**\n   * Calculate the activation function.\n   *\n   * @param x Tensor.\n   * @param alpha Scaling factor for the sigmoid function.\n   * @returns a Tensor of the same shape as x\n   */\n  apply(x: Tensor, alpha = 1): Tensor {\n    return tidy(() => tfc.sigmoid(x.mul(alpha)).mul(x));\n  }\n}\nserialization.registerClass(Swish);\n\nexport function serializeActivation(activation: Activation): string {\n  return activation.getClassName();\n}\n\nexport function deserializeActivation(\n   config: serialization.ConfigDict,\n   customObjects: serialization.ConfigDict = {}): Activation {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'activation');\n}\n\nexport function getActivation(identifier: ActivationIdentifier|\n                              serialization.ConfigDict|Activation): Activation {\n  if (identifier == null) {\n    const config: serialization.ConfigDict = {};\n    config['className'] = 'linear';\n    config['config'] = {};\n    return deserializeActivation(config);\n  }\n  if (typeof identifier === 'string') {\n    const config: serialization.ConfigDict = {};\n    config['className'] = identifier;\n    config['config'] = {};\n    return deserializeActivation(config);\n  } else if (identifier instanceof Activation) {\n    return identifier;\n  } else {\n    return deserializeActivation(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* original source: keras/regularizers.py */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {abs, add, Scalar, serialization, sum, Tensor, tidy, zeros} from '@tensorflow/tfjs-core';\nimport * as K from './backend/tfjs_backend';\nimport {deserializeKerasObject, serializeKerasObject} from './utils/generic_utils';\n\nfunction assertObjectArgs(args: L1Args | L2Args | L1L2Args): void {\n  if (args != null && typeof args !== 'object') {\n    throw new Error(\n        `Argument to L1L2 regularizer's constructor is expected to be an ` +\n        `object, but received: ${args}`);\n  }\n}\n\n/**\n * Regularizer base class.\n */\nexport abstract class Regularizer extends serialization.Serializable {\n  abstract apply(x: Tensor): Scalar;\n}\n\nexport interface L1L2Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1?: number;\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2?: number;\n}\n\nexport interface L1Args {\n  /** L1 regularization rate. Defaults to 0.01. */\n  l1: number;\n}\n\nexport interface L2Args {\n  /** L2 regularization rate. Defaults to 0.01. */\n  l2: number;\n}\n\nexport class L1L2 extends Regularizer {\n  /** @nocollapse */\n  static className = 'L1L2';\n\n  private readonly l1: number;\n  private readonly l2: number;\n  private readonly hasL1: boolean;\n  private readonly hasL2: boolean;\n  constructor(args?: L1L2Args) {\n    super();\n\n    assertObjectArgs(args);\n\n    this.l1 = args == null || args.l1 == null ? 0.01 : args.l1;\n    this.l2 = args == null || args.l2 == null ? 0.01 : args.l2;\n    this.hasL1 = this.l1 !== 0;\n    this.hasL2 = this.l2 !== 0;\n  }\n\n  /**\n   * Porting note: Renamed from __call__.\n   * @param x Variable of which to calculate the regularization score.\n   */\n  apply(x: Tensor): Scalar {\n    return tidy(() => {\n      let regularization: Tensor = zeros([1]);\n      if (this.hasL1) {\n        regularization = add(regularization, sum(tfc.mul(this.l1, abs(x))));\n      }\n      if (this.hasL2) {\n        regularization =\n            add(regularization, sum(tfc.mul(this.l2, K.square(x))));\n      }\n      return regularization.asScalar();\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    return {'l1': this.l1, 'l2': this.l2};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls({l1: config['l1'] as number, l2: config['l2'] as number});\n  }\n}\nserialization.registerClass(L1L2);\n\nexport function l1(args?: L1Args) {\n  assertObjectArgs(args);\n  return new L1L2({l1: args != null ? args.l1 : null, l2: 0});\n}\n\nexport function l2(args: L2Args) {\n  assertObjectArgs(args);\n  return new L1L2({l2: args != null ? args.l2 : null, l1: 0});\n}\n\n/** @docinline */\nexport type RegularizerIdentifier = 'l1l2'|string;\n\n// Maps the JavaScript-like identifier keys to the corresponding keras symbols.\nexport const REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP:\n    {[identifier in RegularizerIdentifier]: string} = {\n      'l1l2': 'L1L2'\n    };\n\nexport function serializeRegularizer(constraint: Regularizer):\n    serialization.ConfigDictValue {\n  return serializeKerasObject(constraint);\n}\n\nexport function deserializeRegularizer(\n    config: serialization.ConfigDict,\n    customObjects: serialization.ConfigDict = {}): Regularizer {\n  return deserializeKerasObject(\n      config, serialization.SerializationMap.getMap().classNameMap,\n      customObjects, 'regularizer');\n}\n\nexport function getRegularizer(identifier: RegularizerIdentifier|\n                               serialization.ConfigDict|\n                               Regularizer): Regularizer {\n  if (identifier == null) {\n    return null;\n  }\n  if (typeof identifier === 'string') {\n    const className = identifier in REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP ?\n        REGULARIZER_IDENTIFIER_REGISTRY_SYMBOL_MAP[identifier] :\n        identifier;\n    const config = {className, config: {}};\n    return deserializeRegularizer(config);\n  } else if (identifier instanceof Regularizer) {\n    return identifier;\n  } else {\n    return deserializeRegularizer(identifier);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n *  Advanced activation layers.\n */\n\nimport {clipByValue, elu, leakyRelu, prelu, relu, serialization, Tensor} from '@tensorflow/tfjs-core';\n\nimport {Softmax as softmaxActivation} from '../activations';\nimport {cast} from '../backend/tfjs_backend';\nimport {Constraint, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface ReLULayerArgs extends LayerArgs {\n  /**\n   * Float, the maximum output value.\n   */\n  maxValue?: number;\n}\n\nexport class ReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ReLU';\n  maxValue: number;\n\n  constructor(args?: ReLULayerArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maxValue = args.maxValue;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    let output = relu(inputs);\n    if (this.maxValue != null) {\n      output = clipByValue(output, 0, this.maxValue);\n    }\n    return output;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {maxValue: this.maxValue};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ReLU);\n\nexport declare interface LeakyReLULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `0.3`.\n   */\n  alpha?: number;\n}\n\nexport class LeakyReLU extends Layer {\n  /** @nocollapse */\n  static className = 'LeakyReLU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 0.3;\n\n  constructor(args?: LeakyReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return leakyRelu(x, this.alpha);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LeakyReLU);\n\nexport declare interface PReLULayerArgs extends LayerArgs {\n  /**\n   * Initializer for the learnable alpha.\n   */\n  alphaInitializer?: Initializer|InitializerIdentifier;\n\n  /**\n   * Regularizer for the learnable alpha.\n   */\n  alphaRegularizer?: Regularizer;\n\n  /**\n   * Constraint for the learnable alpha.\n   */\n  alphaConstraint?: Constraint;\n\n  /**\n   * The axes along which to share learnable parameters for the activation\n   * function. For example, if the incoming feature maps are from a 2D\n   * convolution with output shape `[numExamples, height, width, channels]`,\n   * and you wish to share parameters across space (height and width) so that\n   * each filter channels has only one set of parameters, set\n   * `shared_axes: [1, 2]`.\n   */\n  sharedAxes?: number|number[];\n}\n\nexport class PReLU extends Layer {\n  /** @nocollapse */\n  static className = 'PReLU';\n  private readonly alphaInitializer: Initializer;\n  private readonly alphaRegularizer: Regularizer;\n  private readonly alphaConstraint: Constraint;\n  private readonly sharedAxes: number[];\n  private alpha: LayerVariable;\n\n  readonly DEFAULT_ALPHA_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args?: PReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.supportsMasking = true;\n    this.alphaInitializer =\n        getInitializer(args.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER);\n    this.alphaRegularizer = getRegularizer(args.alphaRegularizer);\n    this.alphaConstraint = getConstraint(args.alphaConstraint);\n    if (args.sharedAxes == null) {\n      this.sharedAxes = null;\n    } else if (Array.isArray(args.sharedAxes)) {\n      this.sharedAxes = args.sharedAxes;\n    } else if (typeof args.sharedAxes === 'number') {\n      this.sharedAxes = [args.sharedAxes];\n    } else {\n      throw new ValueError(\n          `Expected sharedAxes to be a number or an array of numbers, ` +\n          `but got ${args.sharedAxes}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]) {\n    inputShape = getExactlyOneShape(inputShape);\n    const paramShape: Shape = inputShape.slice(1);\n    if (this.sharedAxes != null) {\n      for (const i of this.sharedAxes) {\n        paramShape[i - 1] = 1;\n      }\n    }\n    this.alpha = this.addWeight(\n        'alpha', paramShape, 'float32', this.alphaInitializer,\n        this.alphaRegularizer, true, this.alphaConstraint);\n    // Set input spec.\n    const axes: {[axis: number]: number} = {};\n    if (this.sharedAxes != null) {\n      for (let i = 1; i < inputShape.length; ++i) {\n        axes[i] = inputShape[i];\n      }\n    }\n    this.inputSpec = [new InputSpec({\n      ndim: inputShape.length,\n      axes,\n    })];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    inputs = getExactlyOneTensor(inputs);\n    return prelu(inputs, this.alpha.read());\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      alphaInitializer: serializeInitializer(this.alphaInitializer),\n      alphaRegularizer: serializeRegularizer(this.alphaRegularizer),\n      alphaConstraint: serializeConstraint(this.alphaConstraint),\n      sharedAxes: this.sharedAxes\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(PReLU);\n\nexport declare interface ELULayerArgs extends LayerArgs {\n  /**\n   * Float `>= 0`. Negative slope coefficient. Defaults to `1.0`.\n   */\n  alpha?: number;\n}\n\nexport class ELU extends Layer {\n  /** @nocollapse */\n  static className = 'ELU';\n  readonly alpha: number;\n\n  readonly DEFAULT_ALPHA = 1.0;\n\n  constructor(args?: ELULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    if (args.alpha != null && args.alpha !== this.DEFAULT_ALPHA) {\n      throw new NotImplementedError(\n          `Non-default alpha value (${args.alpha}) is not supported by the ` +\n          `ELU layer yet.`);\n    }\n\n    this.alpha = args.alpha == null ? this.DEFAULT_ALPHA : args.alpha;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return elu(x);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {alpha: this.alpha};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ELU);\n\nexport declare interface ThresholdedReLULayerArgs extends LayerArgs {\n  /**\n   * Float >= 0. Threshold location of activation.\n   */\n  theta?: number;\n}\n\nexport class ThresholdedReLU extends Layer {\n  /** @nocollapse */\n  static className = 'ThresholdedReLU';\n  readonly theta: number;\n\n  readonly DEFAULT_THETA = 1.0;\n\n  constructor(args?: ThresholdedReLULayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n\n    this.theta = args.theta == null ? this.DEFAULT_THETA : args.theta;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return x.mul(cast(x.greater(this.theta), 'float32'));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {theta: this.theta};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ThresholdedReLU);\n\nexport declare interface SoftmaxLayerArgs extends LayerArgs {\n  /**\n   * Integer, axis along which the softmax normalization is applied.\n   * Defaults to `-1` (i.e., the last axis).\n   */\n  axis?: number;\n}\n\nexport class Softmax extends Layer {\n  /** @nocollapse */\n  static className = 'Softmax';\n  readonly axis: number;\n  readonly softmax: (t: Tensor, a?: number) => Tensor;\n  readonly DEFAULT_AXIS = 1.0;\n\n  constructor(args?: SoftmaxLayerArgs) {\n    super(args == null ? {} : args);\n    if (args == null) {\n      args = {};\n    }\n    this.softmax = new softmaxActivation().apply;\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const x = getExactlyOneTensor(inputs);\n    return this.softmax(x, this.axis);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {axis: this.axis};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Softmax);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {ValueError} from '../errors';\nimport {PaddingMode} from '../keras_format/common';\n\nimport {pyListRepeat} from './generic_utils';\nimport {isInteger, max} from './math_utils';\n\n/**\n * Transforms a single number of array of numbers into an array of numbers.\n * @param value\n * @param n: The size of the tuple to be returned.\n * @param name: Name of the parameter, used for generating error messages.\n * @returns An array of numbers.\n */\nexport function normalizeArray(\n    value: number|number[], n: number, name: string): number[] {\n  if (typeof value === 'number') {\n    return pyListRepeat(value, n);\n  } else {\n    if (value.length !== n) {\n      throw new ValueError(\n          `The ${name} argument must be an integer or tuple of ${n} integers.` +\n          ` Received: ${value.length} elements.`);\n    }\n    for (let i = 0; i < n; ++i) {\n      const singleValue = value[i];\n      if (!isInteger(singleValue)) {\n        throw new ValueError(\n            `The ${name} argument must be an integer or tuple of ${n}` +\n            ` integers. Received: ${JSON.stringify(value)} including a` +\n            ` non-integer number ${singleValue}`);\n      }\n    }\n    return value;\n  }\n}\n\n/**\n * Determines output length of a convolution given input length.\n * @param inputLength\n * @param filterSize\n * @param padding\n * @param stride\n * @param dilation: dilation rate.\n */\nexport function convOutputLength(\n    inputLength: number, filterSize: number, padding: PaddingMode,\n    stride: number, dilation = 1): number {\n  if (inputLength == null) {\n    return inputLength;\n  }\n  const dilatedFilterSize = filterSize + (filterSize - 1) * (dilation - 1);\n  let outputLength: number;\n  if (padding === 'same') {\n    outputLength = inputLength;\n  } else {  // VALID\n    outputLength = inputLength - dilatedFilterSize + 1;\n  }\n  return Math.floor((outputLength + stride - 1) / stride);\n}\n\nexport function deconvLength(\n    dimSize: number, strideSize: number, kernelSize: number,\n    padding: PaddingMode): number {\n  if (dimSize == null) {\n    return null;\n  }\n\n  if (padding === 'valid') {\n    dimSize = dimSize * strideSize + max([kernelSize - strideSize, 0]);\n  } else if (padding === 'same') {\n    dimSize = dimSize * strideSize;\n  } else {\n    throw new ValueError(`Unsupport padding mode: ${padding}.`);\n  }\n  return dimSize;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {fused, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, PaddingMode, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, deconvLength, normalizeArray} from '../utils/conv_utils';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Transpose and cast the input before the conv2d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv2DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  // TODO(cais): Cast type to float32 if not.\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 1]);  // NCHW -> NHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * Transpose and cast the input before the conv3d.\n * @param x Input image tensor.\n * @param dataFormat\n */\nexport function preprocessConv3DInput(\n    x: Tensor, dataFormat: DataFormat): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    if (dataFormat === 'channelsFirst') {\n      return tfc.transpose(x, [0, 2, 3, 4, 1]);  // NCDHW -> NDHWC.\n    } else {\n      return x;\n    }\n  });\n}\n\n/**\n * 1D-convolution with bias added.\n *\n * Porting Note: This function does not exist in the Python Keras backend.\n *   It is exactly the same as `conv2d`, except the added `bias`.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.\n * @param bias Bias, rank-3, of shape `[outDepth]`.\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    // Check the ranks of x, kernel and bias.\n    if (x.shape.length !== 3) {\n      throw new ValueError(\n          `The input of a conv1dWithBias operation should be 3, but is ` +\n          `${x.shape.length} instead.`);\n    }\n    if (kernel.shape.length !== 3) {\n      throw new ValueError(\n          `The kernel for a conv1dWithBias operation should be 3, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    if (bias != null && bias.shape.length !== 1) {\n      throw new ValueError(\n          `The bias for a conv1dWithBias operation should be 1, but is ` +\n          `${kernel.shape.length} instead`);\n    }\n    // TODO(cais): Support CAUSAL padding mode.\n    if (dataFormat === 'channelsFirst') {\n      x = tfc.transpose(x, [0, 2, 1]);  // NCW -> NWC.\n    }\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    let y: Tensor = tfc.conv1d(\n        x as Tensor2D | Tensor3D, kernel as Tensor3D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias);\n    }\n    return y;\n  });\n}\n\n/**\n * 1D-convolution.\n *\n * @param x Input tensor, rank-3, of shape `[batchSize, width, inChannels]`.\n * @param kernel Kernel, rank-3, of shape `[filterWidth, inDepth, outDepth]`.s\n * @param strides\n * @param padding Padding mode.\n * @param dataFormat Data format.\n * @param dilationRate\n * @returns The result of the 1D convolution.\n * @throws ValueError, if `x`, `kernel` or `bias` is not of the correct rank.\n */\nexport function conv1d(\n    x: Tensor, kernel: Tensor, strides = 1, padding = 'valid',\n    dataFormat?: DataFormat, dilationRate = 1): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv1dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 2D pooling.\n */\nexport function conv2d(\n    x: Tensor, kernel: Tensor, strides = [1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv2dWithBiasActivation(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 2D Convolution with an added bias and optional activation.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv2d`, except the added `bias`.\n */\nexport function conv2dWithBiasActivation(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat, dilationRate?: [number, number],\n    activation: fused.Activation = null): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 3 && x.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects input to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    if (kernel.rank !== 3 && kernel.rank !== 4) {\n      throw new ValueError(\n          `conv2dWithBiasActivation expects kernel to be of rank 3 or 4, ` +\n          `but received ${x.rank}.`);\n    }\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv1dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.fused.conv2d({\n      x: y as Tensor3D | Tensor4D,\n      filter: kernel as Tensor4D,\n      strides: strides as [number, number],\n      pad: padding === 'same' ? 'same' : 'valid',\n      dilations: dilationRate,\n      dataFormat: 'NHWC',\n      bias,\n      activation\n    });\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\n/**\n * 3D Convolution.\n * @param x\n * @param kernel kernel of the convolution.\n * @param strides strides array.\n * @param padding padding mode. Default to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param dilationRate dilation rate array.\n * @returns Result of the 3D convolution.\n */\nexport function conv3d(\n    x: Tensor, kernel: Tensor, strides = [1, 1, 1], padding = 'valid',\n    dataFormat?: DataFormat, dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    return conv3dWithBias(\n        x, kernel, null, strides, padding, dataFormat, dilationRate);\n  });\n}\n\n/**\n * 3D Convolution with an added bias.\n * Note: This function does not exist in the Python Keras Backend. This function\n * is exactly the same as `conv3d`, except the added `bias`.\n */\nexport function conv3dWithBias(\n    x: Tensor, kernel: Tensor, bias: Tensor, strides = [1, 1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    if (x.rank !== 4 && x.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects input to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    if (kernel.rank !== 4 && kernel.rank !== 5) {\n      throw new ValueError(\n          `conv3dWithBias expects kernel to be of rank 4 or 5, but received ` +\n          `${x.rank}.`);\n    }\n    let y = preprocessConv3DInput(x, dataFormat);\n    if (padding === 'causal') {\n      throw new NotImplementedError(\n          'The support for CAUSAL padding mode in conv3dWithBias is not ' +\n          'implemented yet.');\n    }\n    y = tfc.conv3d(\n        y as Tensor4D | tfc.Tensor<tfc.Rank.R5>,\n        kernel as tfc.Tensor<tfc.Rank.R5>, strides as [number, number, number],\n        padding === 'same' ? 'same' : 'valid', 'NDHWC', dilationRate);\n    if (bias != null) {\n      y = K.biasAdd(y, bias as Tensor1D);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);\n    }\n    return y;\n  });\n}\n\n/**\n * Base LayerConfig for depthwise and non-depthwise convolutional layers.\n */\nexport declare interface BaseConvLayerArgs extends LayerArgs {\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number]|[number, number, number];\n\n  /**\n   * Activation function of the layer.\n   *\n   * If you don't specify the activation, none is applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector. Defaults to `true`.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the convolutional kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the convolutional kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\n/**\n * LayerConfig for non-depthwise convolutional layers.\n * Applies to non-depthwise convolution of all ranks (e.g, Conv1D, Conv2D,\n * Conv3D).\n */\nexport declare interface ConvLayerArgs extends BaseConvLayerArgs {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n}\n\n/**\n * Abstract convolution layer.\n */\nexport abstract class BaseConv extends Layer {\n  protected readonly rank: number;\n  protected readonly kernelSize: number[];\n  protected readonly strides: number[];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n  protected readonly activation: Activation;\n  protected readonly useBias: boolean;\n  protected readonly dilationRate: number[];\n\n  // Bias-related members are here because all convolution subclasses use the\n  // same configuration parmeters to control bias.  Kernel-related members\n  // are in subclass `Conv` because some subclasses use different parameters to\n  // control kernel properties, for instance, `DepthwiseConv2D` uses\n  // `depthwiseInitializer` instead of `kernelInitializer`.\n  protected readonly biasInitializer?: Initializer;\n  protected readonly biasConstraint?: Constraint;\n  protected readonly biasRegularizer?: Regularizer;\n\n  protected bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(rank: number, args: BaseConvLayerArgs) {\n    super(args as LayerArgs);\n    BaseConv.verifyArgs(args);\n    this.rank = rank;\n    generic_utils.assertPositiveInteger(this.rank, 'rank');\n    if (this.rank !== 1 && this.rank !== 2 && this.rank !== 3) {\n      throw new NotImplementedError(\n          `Convolution layer for rank other than 1, 2, or 3 (${\n              this.rank}) is ` +\n          `not implemented yet.`);\n    }\n    this.kernelSize = normalizeArray(args.kernelSize, rank, 'kernelSize');\n    this.strides = normalizeArray(\n        args.strides == null ? 1 : args.strides, rank, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.activation = getActivation(args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.dilationRate = normalizeArray(\n        args.dilationRate == null ? 1 : args.dilationRate, rank,\n        'dilationRate');\n    if (this.rank === 1 &&\n        (Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)) {\n      throw new ValueError(\n          `dilationRate must be a number or an array of a single number ` +\n          `for 1D convolution, but received ` +\n          `${JSON.stringify(this.dilationRate)}`);\n    } else if (this.rank === 2) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate = [this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 2) {\n        throw new ValueError(\n            `dilationRate must be a number or array of two numbers for 2D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    } else if (this.rank === 3) {\n      if (typeof this.dilationRate === 'number') {\n        this.dilationRate =\n            [this.dilationRate, this.dilationRate, this.dilationRate];\n      } else if (this.dilationRate.length !== 3) {\n        throw new ValueError(\n            `dilationRate must be a number or array of three numbers for 3D ` +\n            `convolution, but received ${JSON.stringify(this.dilationRate)}`);\n      }\n    }\n  }\n\n  protected static verifyArgs(args: BaseConvLayerArgs) {\n    // Check config.kernelSize type and shape.\n    generic_utils.assert(\n        'kernelSize' in args, `required key 'kernelSize' not in config`);\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 3)) {\n      throw new ValueError(\n          `BaseConv expects config.kernelSize to be number or number[] with ` +\n          `length 1, 2, or 3, but received ${\n              JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      kernelSize: this.kernelSize,\n      strides: this.strides,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\n/**\n * Abstract nD convolution layer.  Ancestor of convolution layers which reduce\n * across channels, i.e., Conv1D and Conv2D, but not DepthwiseConv2D.\n */\nexport abstract class Conv extends BaseConv {\n  protected readonly filters: number;\n\n  protected kernel: LayerVariable = null;\n\n  // Bias-related properties are stored in the superclass `BaseConv` because all\n  // convolution subclasses use the same configuration parameters to control\n  // bias. Kernel-related properties are defined here rather than in the\n  // superclass because some convolution subclasses use different names and\n  // configuration parameters for their internal kernel state.\n  protected readonly kernelInitializer?: Initializer;\n  protected readonly kernelConstraint?: Constraint;\n  protected readonly kernelRegularizer?: Regularizer;\n\n  constructor(rank: number, args: ConvLayerArgs) {\n    super(rank, args as BaseConvLayerArgs);\n    Conv.verifyArgs(args);\n    this.filters = args.filters;\n    generic_utils.assertPositiveInteger(this.filters, 'filters');\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n    const inputDim = inputShape[channelAxis];\n\n    const kernelShape = this.kernelSize.concat([inputDim, this.filters]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.inputSpec = [{ndim: this.rank + 2, axes: {[channelAxis]: inputDim}}];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs: Tensor;\n      const biasValue = this.bias == null ? null : this.bias.read();\n      const fusedActivationName = generic_utils.mapActivationToFusedKernel(\n                                            this.activation.getClassName());\n\n      if (fusedActivationName != null && this.rank === 2) {\n        outputs = conv2dWithBiasActivation(\n            inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n            this.dataFormat, this.dilationRate as [number, number],\n            fusedActivationName);\n      } else {\n        if (this.rank === 1) {\n          outputs = conv1dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides[0],\n              this.padding, this.dataFormat, this.dilationRate[0]);\n        } else if (this.rank === 2) {\n          // TODO(cais): Move up to constructor.\n          outputs = conv2dWithBiasActivation(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number]);\n        } else if (this.rank === 3) {\n          outputs = conv3dWithBias(\n              inputs, this.kernel.read(), biasValue, this.strides, this.padding,\n              this.dataFormat, this.dilationRate as [number, number, number]);\n        } else {\n          throw new NotImplementedError(\n              'convolutions greater than 3D are not implemented yet.');\n        }\n\n        if (this.activation != null) {\n          outputs = this.activation.apply(outputs);\n        }\n      }\n\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const newSpace: number[] = [];\n    const space = (this.dataFormat === 'channelsLast') ?\n        inputShape.slice(1, inputShape.length - 1) :\n        inputShape.slice(2);\n    for (let i = 0; i < space.length; ++i) {\n      const newDim = convOutputLength(\n          space[i], this.kernelSize[i], this.padding, this.strides[i],\n          typeof this.dilationRate === 'number' ? this.dilationRate :\n                                                  this.dilationRate[i]);\n      newSpace.push(newDim);\n    }\n\n    let outputShape = [inputShape[0]];\n    if (this.dataFormat === 'channelsLast') {\n      outputShape = outputShape.concat(newSpace);\n      outputShape.push(this.filters);\n    } else {\n      outputShape.push(this.filters);\n      outputShape = outputShape.concat(newSpace);\n    }\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      filters: this.filters,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // Check config.filters type, shape, and value.\n    if (!('filters' in args) || typeof args.filters !== 'number' ||\n        args.filters < 1) {\n      throw new ValueError(\n          `Convolution layer expected config.filters to be a 'number' > 0 ` +\n          `but got ${JSON.stringify(args.filters)}`);\n    }\n  }\n}\n\nexport class Conv2D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv2D';\n  constructor(args: ConvLayerArgs) {\n    super(2, args);\n    Conv2D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if ((typeof args.kernelSize !== 'number') &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 2)) {\n      throw new ValueError(\n          `Conv2D expects config.kernelSize to be number or number[] with ` +\n          `length 1 or 2, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv2D);\n\nexport class Conv3D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv3D';\n  constructor(args: ConvLayerArgs) {\n    super(3, args);\n    Conv3D.verifyArgs(args);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number') {\n      if (!(Array.isArray(args.kernelSize) &&\n            (args.kernelSize.length === 1 || args.kernelSize.length === 3))) {\n        throw new ValueError(\n            `Conv3D expects config.kernelSize to be number or` +\n            ` [number, number, number], but received ${\n                JSON.stringify(args.kernelSize)}.`);\n      }\n    }\n  }\n}\nserialization.registerClass(Conv3D);\n\nexport class Conv2DTranspose extends Conv2D {\n  /** @nocollapse */\n  static className = 'Conv2DTranspose';\n  inputSpec: InputSpec[];\n\n  constructor(args: ConvLayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n\n    if (this.padding !== 'same' && this.padding !== 'valid') {\n      throw new ValueError(\n          `Conv2DTranspose currently supports only padding modes 'same' ` +\n          `and 'valid', but received padding mode ${this.padding}`);\n    }\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    if (inputShape.length !== 4) {\n      throw new ValueError(\n          'Input should have rank 4; Received input shape: ' +\n          JSON.stringify(inputShape));\n    }\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          'The channel dimension of the inputs should be defined. ' +\n          'Found `None`.');\n    }\n    const inputDim = inputShape[channelAxis];\n    const kernelShape = this.kernelSize.concat([this.filters, inputDim]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, 'float32', this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    // Set input spec.\n    this.inputSpec =\n        [new InputSpec({ndim: 4, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs);\n      if (input.shape.length !== 4) {\n        throw new ValueError(\n            `Conv2DTranspose.call() expects input tensor to be rank-4, but ` +\n            `received a tensor of rank-${input.shape.length}`);\n      }\n\n      const inputShape = input.shape;\n      const batchSize = inputShape[0];\n\n      let hAxis: number;\n      let wAxis: number;\n      if (this.dataFormat === 'channelsFirst') {\n        hAxis = 2;\n        wAxis = 3;\n      } else {\n        hAxis = 1;\n        wAxis = 2;\n      }\n\n      const height = inputShape[hAxis];\n      const width = inputShape[wAxis];\n      const kernelH = this.kernelSize[0];\n      const kernelW = this.kernelSize[1];\n      const strideH = this.strides[0];\n      const strideW = this.strides[1];\n\n      // Infer the dynamic output shape.\n      const outHeight = deconvLength(height, strideH, kernelH, this.padding);\n      const outWidth = deconvLength(width, strideW, kernelW, this.padding);\n\n      // Porting Note: We don't branch based on `this.dataFormat` here,\n      // because\n      //   the tjfs-core function `conv2dTranspose` called below always\n      //   assumes channelsLast.\n      const outputShape: [number, number, number, number] =\n          [batchSize, outHeight, outWidth, this.filters];\n\n      if (this.dataFormat !== 'channelsLast') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n      }\n      let outputs = tfc.conv2dTranspose(\n          input as Tensor4D, this.kernel.read() as Tensor4D, outputShape,\n          this.strides as [number, number], this.padding as 'same' | 'valid');\n      if (this.dataFormat !== 'channelsLast') {\n        outputs = tfc.transpose(outputs, [0, 3, 1, 2]);\n      }\n\n      if (this.bias != null) {\n        outputs =\n            K.biasAdd(outputs, this.bias.read(), this.dataFormat) as Tensor4D;\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs) as Tensor4D;\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n\n    let channelAxis: number;\n    let heightAxis: number;\n    let widthAxis: number;\n    if (this.dataFormat === 'channelsFirst') {\n      channelAxis = 1;\n      heightAxis = 2;\n      widthAxis = 3;\n    } else {\n      channelAxis = 3;\n      heightAxis = 1;\n      widthAxis = 2;\n    }\n\n    const kernelH = this.kernelSize[0];\n    const kernelW = this.kernelSize[1];\n    const strideH = this.strides[0];\n    const strideW = this.strides[1];\n\n    outputShape[channelAxis] = this.filters;\n    outputShape[heightAxis] =\n        deconvLength(outputShape[heightAxis], strideH, kernelH, this.padding);\n    outputShape[widthAxis] =\n        deconvLength(outputShape[widthAxis], strideW, kernelW, this.padding);\n    return outputShape;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['dilationRate'];\n    return config;\n  }\n}\nserialization.registerClass(Conv2DTranspose);\n\nexport declare interface SeparableConvLayerArgs extends ConvLayerArgs {\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal\n   * to `filtersIn * depthMultiplier`. Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the pointwise kernel matrix.\n   */\n  pointwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the pointwise kernel matrix.\n   */\n  pointwiseRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the pointwise kernel matrix.\n   */\n  pointwiseConstraint?: ConstraintIdentifier|Constraint;\n}\n\nexport class SeparableConv extends Conv {\n  /** @nocollapse */\n  static className = 'SeparableConv';\n\n  readonly depthMultiplier: number;\n\n  protected readonly depthwiseInitializer?: Initializer;\n  protected readonly depthwiseRegularizer?: Regularizer;\n  protected readonly depthwiseConstraint?: Constraint;\n  protected readonly pointwiseInitializer?: Initializer;\n  protected readonly pointwiseRegularizer?: Regularizer;\n  protected readonly pointwiseConstraint?: Constraint;\n\n  readonly DEFAULT_DEPTHWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n  readonly DEFAULT_POINTWISE_INITIALIZER: InitializerIdentifier =\n      'glorotUniform';\n\n  protected depthwiseKernel: LayerVariable = null;\n  protected pointwiseKernel: LayerVariable = null;\n\n  constructor(rank: number, config?: SeparableConvLayerArgs) {\n    super(rank, config);\n\n    if (config.filters == null) {\n      throw new ValueError(\n          'The `filters` configuration field is required by SeparableConv, ' +\n          'but is unspecified.');\n    }\n    if (config.kernelInitializer != null || config.kernelRegularizer != null ||\n        config.kernelConstraint != null) {\n      throw new ValueError(\n          'Fields kernelInitializer, kernelRegularizer and kernelConstraint ' +\n          'are invalid for SeparableConv2D. Use depthwiseInitializer, ' +\n          'depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, ' +\n          'pointwiseRegularizer and pointwiseConstraint instead.');\n    }\n    if (config.padding != null && config.padding !== 'same' &&\n        config.padding !== 'valid') {\n      throw new ValueError(\n          `SeparableConv${this.rank}D supports only padding modes: ` +\n          `'same' and 'valid', but received ${JSON.stringify(config.padding)}`);\n    }\n\n    this.depthMultiplier =\n        config.depthMultiplier == null ? 1 : config.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER);\n    this.depthwiseRegularizer = getRegularizer(config.depthwiseRegularizer);\n    this.depthwiseConstraint = getConstraint(config.depthwiseConstraint);\n    this.pointwiseInitializer = getInitializer(\n        config.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER);\n    this.pointwiseRegularizer = getRegularizer(config.pointwiseRegularizer);\n    this.pointwiseConstraint = getConstraint(config.pointwiseConstraint);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < this.rank + 2) {\n      throw new ValueError(\n          `Inputs to SeparableConv${this.rank}D should have rank ` +\n          `${this.rank + 2}, but received input shape: ` +\n          `${JSON.stringify(inputShape)}`);\n    }\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          `The channel dimension of the inputs should be defined, ` +\n          `but found ${JSON.stringify(inputShape[channelAxis])}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape =\n        this.kernelSize.concat([inputDim, this.depthMultiplier]);\n    const pointwiseKernelShape = [];\n    for (let i = 0; i < this.rank; ++i) {\n      pointwiseKernelShape.push(1);\n    }\n    pointwiseKernelShape.push(inputDim * this.depthMultiplier, this.filters);\n\n    const trainable = true;\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, 'float32',\n        this.depthwiseInitializer, this.depthwiseRegularizer, trainable,\n        this.depthwiseConstraint);\n    this.pointwiseKernel = this.addWeight(\n        'pointwise_kernel', pointwiseKernelShape, 'float32',\n        this.pointwiseInitializer, this.pointwiseRegularizer, trainable,\n        this.pointwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.filters], 'float32', this.biasInitializer,\n          this.biasRegularizer, trainable, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n\n    this.inputSpec =\n        [new InputSpec({ndim: this.rank + 2, axes: {[channelAxis]: inputDim}})];\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      let output: Tensor;\n      if (this.rank === 1) {\n        throw new NotImplementedError(\n            '1D separable convolution is not implemented yet.');\n      } else if (this.rank === 2) {\n        if (this.dataFormat === 'channelsFirst') {\n          inputs = tfc.transpose(inputs, [0, 2, 3, 1]);  // NCHW -> NHWC.\n        }\n\n        output = tfc.separableConv2d(\n            inputs as Tensor4D, this.depthwiseKernel.read() as Tensor4D,\n            this.pointwiseKernel.read() as Tensor4D,\n            this.strides as [number, number], this.padding as 'same' | 'valid',\n            this.dilationRate as [number, number], 'NHWC');\n      }\n\n      if (this.useBias) {\n        output = K.biasAdd(output, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      if (this.dataFormat === 'channelsFirst') {\n        output = tfc.transpose(output, [0, 3, 1, 2]);  // NHWC -> NCHW.\n      }\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['kernelInitializer'];\n    delete config['kernelRegularizer'];\n    delete config['kernelConstraint'];\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['pointwiseInitializer'] =\n        serializeInitializer(this.pointwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['pointwiseRegularizer'] =\n        serializeRegularizer(this.pointwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseConstraint);\n    config['pointwiseConstraint'] =\n        serializeConstraint(this.pointwiseConstraint);\n    return config;\n  }\n}\n\nexport class SeparableConv2D extends SeparableConv {\n  /** @nocollapse */\n  static className = 'SeparableConv2D';\n  constructor(args?: SeparableConvLayerArgs) {\n    super(2, args);\n  }\n}\nserialization.registerClass(SeparableConv2D);\n\nexport class Conv1D extends Conv {\n  /** @nocollapse */\n  static className = 'Conv1D';\n  constructor(args: ConvLayerArgs) {\n    super(1, args);\n    Conv1D.verifyArgs(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    delete config['rank'];\n    delete config['dataFormat'];\n    return config;\n  }\n\n  protected static verifyArgs(args: ConvLayerArgs) {\n    // config.kernelSize must be a number or array of numbers.\n    if (typeof args.kernelSize !== 'number' &&\n        !generic_utils.checkArrayTypeAndLength(\n            args.kernelSize, 'number', 1, 1)) {\n      throw new ValueError(\n          `Conv1D expects config.kernelSize to be number or number[] with ` +\n          `length 1, but received ${JSON.stringify(args.kernelSize)}.`);\n    }\n  }\n}\nserialization.registerClass(Conv1D);\n\nexport declare interface Cropping2DLayerArgs extends LayerArgs {\n  /**\n   * Dimension of the cropping along the width and the height.\n   * - If integer: the same symmetric cropping\n   *  is applied to width and height.\n   * - If list of 2 integers:\n   *   interpreted as two different\n   *   symmetric cropping values for height and width:\n   *   `[symmetric_height_crop, symmetric_width_crop]`.\n   * - If a list of 2 list of 2 integers:\n   *   interpreted as\n   *   `[[top_crop, bottom_crop], [left_crop, right_crop]]`\n   */\n  cropping: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   * `channels_first` corresponds to inputs with shape\n   *   `(batch, channels, ...)`\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class Cropping2D extends Layer {\n  /** @nocollapse */\n  static className = 'Cropping2D';\n  protected readonly cropping: [[number, number], [number, number]];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Cropping2DLayerArgs) {\n    super(args);\n    if (typeof args.cropping === 'number') {\n      this.cropping =\n          [[args.cropping, args.cropping], [args.cropping, args.cropping]];\n    } else if (typeof args.cropping[0] === 'number') {\n      this.cropping = [\n        [args.cropping[0], args.cropping[0]],\n        [args.cropping[1] as number, args.cropping[1] as number]\n      ];\n    } else {\n      this.cropping = args.cropping as [[number, number], [number, number]];\n    }\n    this.dataFormat =\n        args.dataFormat === undefined ? 'channelsLast' : args.dataFormat;\n    this.inputSpec = [{ndim: 4}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      return [\n        inputShape[0], inputShape[1],\n        inputShape[2] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[3] - this.cropping[1][0] - this.cropping[1][1]\n      ];\n    } else {\n      return [\n        inputShape[0],\n        inputShape[1] - this.cropping[0][0] - this.cropping[0][1],\n        inputShape[2] - this.cropping[1][0] - this.cropping[1][1], inputShape[3]\n      ];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n\n      if (this.dataFormat === 'channelsLast') {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[1] - this.cropping[0][0] - this.cropping[0][1], 2);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[2] - this.cropping[1][1] - this.cropping[1][0], 3);\n      } else {\n        const hSliced = K.sliceAlongAxis(\n            inputs, this.cropping[0][0],\n            inputs.shape[2] - this.cropping[0][0] - this.cropping[0][1], 3);\n        return K.sliceAlongAxis(\n            hSliced, this.cropping[1][0],\n            inputs.shape[3] - this.cropping[1][1] - this.cropping[1][0], 4);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {cropping: this.cropping, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Cropping2D);\n\nexport declare interface UpSampling2DLayerArgs extends LayerArgs {\n  /**\n   * The upsampling factors for rows and columns.\n   *\n   * Defaults to `[2, 2]`.\n   */\n  size?: number[];\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `\"channelsLast\"` corresponds to inputs with shape\n   *   `[batch, ..., channels]`\n   *\n   *  `\"channelsFirst\"` corresponds to inputs with shape `[batch, channels,\n   * ...]`.\n   *\n   * Defaults to `\"channelsLast\"`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class UpSampling2D extends Layer {\n  /** @nocollapse */\n  static className = 'UpSampling2D';\n  protected readonly DEFAULT_SIZE = [2, 2];\n  protected readonly size: number[];\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: UpSampling2DLayerArgs) {\n    super(args);\n    this.inputSpec = [{ndim: 4}];\n    this.size = args.size == null ? this.DEFAULT_SIZE : args.size;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    if (this.dataFormat === 'channelsFirst') {\n      const height =\n          inputShape[2] == null ? null : this.size[0] * inputShape[2];\n      const width = inputShape[3] == null ? null : this.size[1] * inputShape[3];\n      return [inputShape[0], inputShape[1], height, width];\n    } else {\n      const height =\n          inputShape[1] == null ? null : this.size[0] * inputShape[1];\n      const width = inputShape[2] == null ? null : this.size[1] * inputShape[2];\n      return [inputShape[0], height, width, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      let input = getExactlyOneTensor(inputs) as Tensor4D;\n      const inputShape = input.shape;\n\n      if (this.dataFormat === 'channelsFirst') {\n        input = tfc.transpose(input, [0, 2, 3, 1]);\n        const height = this.size[0] * inputShape[2];\n        const width = this.size[1] * inputShape[3];\n        const resized = input.resizeNearestNeighbor([height, width]);\n        return tfc.transpose(resized, [0, 3, 1, 2]);\n      } else {\n        const height = this.size[0] * inputShape[1];\n        const width = this.size[1] * inputShape[2];\n        return input.resizeNearestNeighbor([height, width]);\n      }\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {size: this.size, dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(UpSampling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Depthwise Convolutional Layers\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor4D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {BaseConv, BaseConvLayerArgs, ConvLayerArgs, preprocessConv2DInput} from './convolutional';\n\n/**\n * 2D convolution with separable filters.\n * @param x Input tensor.\n * @param depthwiseKernel Convolution kernel for depthwise convolution.\n * @param strides Strides (Array of two integers).\n * @param padding Padding model.\n * @param dataFormat Data format.\n * @param dilationRate Array of two integers, dilation rates for the separable\n *   convolution.\n * @returns Output tensor.\n * @throws ValueError If depthwiseKernel is not a 4D array.\n */\nexport function depthwiseConv2d(\n    x: Tensor, depthwiseKernel: Tensor, strides: [number, number] = [1, 1],\n    padding = 'valid', dataFormat?: DataFormat,\n    dilationRate?: [number, number]): Tensor {\n  return tidy(() => {\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    checkDataFormat(dataFormat);\n    let y = preprocessConv2DInput(x, dataFormat);\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `Input for depthwiseConv2d is required to be 4-D, but is instead ` +\n          `${x.rank}-D`);\n    }\n    if (depthwiseKernel.rank !== 4) {\n      throw new ValueError(\n          `depthwiseKernel is required to be 4-D, but is instead ` +\n          `${depthwiseKernel.rank}-D`);\n    }\n    y = tfc.depthwiseConv2d(\n        y as Tensor4D, depthwiseKernel as Tensor4D, strides,\n        padding === 'same' ? 'same' : 'valid', 'NHWC', dilationRate);\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);\n    }\n    return y;\n  });\n}\n\nexport declare interface DepthwiseConv2DLayerArgs extends BaseConvLayerArgs {\n  /**\n   * An integer or Array of 2 integers, specifying the width and height of the\n   * 2D convolution window. Can be a single integer to specify the same value\n   * for all spatial dimensions.\n   */\n  kernelSize: number|[number, number];\n\n  /**\n   * The number of depthwise convolution output channels for each input\n   * channel.\n   * The total number of depthwise convolution output channels will be equal to\n   * `filtersIn * depthMultiplier`.\n   * Default: 1.\n   */\n  depthMultiplier?: number;\n\n  /**\n   * Initializer for the depthwise kernel matrix.\n   * Default: GlorotNormal.\n   */\n  depthwiseInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the depthwise kernel matrix.\n   */\n  depthwiseConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regulzarizer function for the depthwise kernel matrix.\n   */\n  depthwiseRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class DepthwiseConv2D extends BaseConv {\n  /** @nocollapse */\n  static className = 'DepthwiseConv2D';\n  private readonly depthMultiplier: number;\n  private readonly depthwiseInitializer: Initializer;\n  private readonly depthwiseConstraint: Constraint;\n  private readonly depthwiseRegularizer: Regularizer;\n\n  private depthwiseKernel: LayerVariable = null;\n\n  constructor(args: DepthwiseConv2DLayerArgs) {\n    super(2, args as ConvLayerArgs);\n    this.depthMultiplier =\n        args.depthMultiplier == null ? 1 : args.depthMultiplier;\n    this.depthwiseInitializer = getInitializer(\n        args.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.depthwiseConstraint = getConstraint(args.depthwiseConstraint);\n    this.depthwiseRegularizer = getRegularizer(args.depthwiseRegularizer);\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 4) {\n      throw new ValueError(\n          `Inputs to DepthwiseConv2D should have rank 4. ` +\n          `Received input shape: ${JSON.stringify(inputShape)}.`);\n    }\n    const channelAxis = this.dataFormat === 'channelsFirst' ? 1 : 3;\n    if (inputShape[channelAxis] == null || inputShape[channelAxis] < 0) {\n      throw new ValueError(\n          'The channel dimension of the inputs to DepthwiseConv2D should ' +\n          `be defined, but is not (${inputShape[channelAxis]}).`);\n    }\n    const inputDim = inputShape[channelAxis];\n    const depthwiseKernelShape: Shape = [\n      this.kernelSize[0], this.kernelSize[1], inputDim, this.depthMultiplier\n    ];\n\n    this.depthwiseKernel = this.addWeight(\n        'depthwise_kernel', depthwiseKernelShape, null,\n        this.depthwiseInitializer, this.depthwiseRegularizer, true,\n        this.depthwiseConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [inputDim * this.depthMultiplier], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      let outputs = depthwiseConv2d(\n          inputs, this.depthwiseKernel.read(), this.strides as [number, number],\n          this.padding, this.dataFormat, null);\n      // TODO(cais): Add support for dilation.\n      if (this.useBias) {\n        outputs = K.biasAdd(outputs, this.bias.read(), this.dataFormat);\n      }\n      if (this.activation != null) {\n        outputs = this.activation.apply(outputs);\n      }\n      return outputs;\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    const cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    const outFilters = this.dataFormat === 'channelsFirst' ?\n        inputShape[1] * this.depthMultiplier :\n        inputShape[3] * this.depthMultiplier;\n    const outRows = convOutputLength(\n        rows, this.kernelSize[0], this.padding, this.strides[0]);\n    const outCols = convOutputLength(\n        cols, this.kernelSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], outFilters, outRows, outCols];\n    } else {\n      // In this case, assume 'channelsLast'.\n      return [inputShape[0], outRows, outCols, outFilters];\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = super.getConfig();\n    config['depthMultiplier'] = this.depthMultiplier;\n    config['depthwiseInitializer'] =\n        serializeInitializer(this.depthwiseInitializer);\n    config['depthwiseRegularizer'] =\n        serializeRegularizer(this.depthwiseRegularizer);\n    config['depthwiseConstraint'] =\n        serializeConstraint(this.depthwiseRegularizer);\n    return config;\n  }\n}\nserialization.registerClass(DepthwiseConv2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Recurrent Neural Network Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {DataType, serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Activation, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, SymbolicTensor} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, Ones, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs, RnnStepFunction} from '../types';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor, isArrayOfShapes} from '../utils/types_utils';\nimport {batchGetValue, batchSetValue, LayerVariable} from '../variables';\nimport {deserialize} from './serialization';\n\n/**\n * Standardize `apply()` args to a single list of tensor inputs.\n *\n * When running a model loaded from file, the input tensors `initialState` and\n * `constants` are passed to `RNN.apply()` as part of `inputs` instead of the\n * dedicated kwargs fields. `inputs` consists of\n * `[inputs, initialState0, initialState1, ..., constant0, constant1]` in this\n * case.\n * This method makes sure that arguments are\n * separated and that `initialState` and `constants` are `Array`s of tensors\n * (or None).\n *\n * @param inputs Tensor or `Array` of  tensors.\n * @param initialState Tensor or `Array` of tensors or `null`/`undefined`.\n * @param constants Tensor or `Array` of tensors or `null`/`undefined`.\n * @returns An object consisting of\n *   inputs: A tensor.\n *   initialState: `Array` of tensors or `null`.\n *   constants: `Array` of tensors or `null`.\n * @throws ValueError, if `inputs` is an `Array` but either `initialState` or\n *   `constants` is provided.\n */\nexport function standardizeArgs(\n    inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    initialState: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    constants: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n    numConstants?: number): {\n  inputs: Tensor|SymbolicTensor,\n  initialState: Tensor[]|SymbolicTensor[],\n  constants: Tensor[]|SymbolicTensor[]\n} {\n  if (Array.isArray(inputs)) {\n    if (initialState != null || constants != null) {\n      throw new ValueError(\n          'When inputs is an array, neither initialState or constants ' +\n          'should be provided');\n    }\n    if (numConstants != null) {\n      constants = inputs.slice(inputs.length - numConstants, inputs.length);\n      inputs = inputs.slice(0, inputs.length - numConstants);\n    }\n    if (inputs.length > 1) {\n      initialState = inputs.slice(1, inputs.length);\n    }\n    inputs = inputs[0];\n  }\n\n  function toListOrNull(x: Tensor|Tensor[]|SymbolicTensor|\n                        SymbolicTensor[]): Tensor[]|SymbolicTensor[] {\n    if (x == null || Array.isArray(x)) {\n      return x as Tensor[] | SymbolicTensor[];\n    } else {\n      return [x] as Tensor[] | SymbolicTensor[];\n    }\n  }\n\n  initialState = toListOrNull(initialState);\n  constants = toListOrNull(constants);\n\n  return {inputs, initialState, constants};\n}\n\n/**\n * Iterates over the time dimension of a tensor.\n *\n * @param stepFunction RNN step function.\n *   Parameters:\n *     inputs: tensor with shape `[samples, ...]` (no time dimension),\n *       representing input for the batch of samples at a certain time step.\n *     states: an Array of tensors.\n *   Returns:\n *     outputs: tensor with shape `[samples, outputDim]` (no time dimension).\n *     newStates: list of tensors, same length and shapes as `states`. The first\n *       state in the list must be the output tensor at the previous timestep.\n * @param inputs Tensor of temporal data of shape `[samples, time, ...]` (at\n *   least 3D).\n * @param initialStates Tensor with shape `[samples, outputDim]` (no time\n *   dimension), containing the initial values of the states used in the step\n *   function.\n * @param goBackwards If `true`, do the iteration over the time dimension in\n *   reverse order and return the reversed sequence.\n * @param mask Binary tensor with shape `[sample, time, 1]`, with a zero for\n *   every element that is masked.\n * @param constants An Array of constant values passed at each step.\n * @param unroll Whether to unroll the RNN or to use a symbolic loop. *Not*\n *   applicable to this imperative deeplearn.js backend. Its value is ignored.\n * @param needPerStepOutputs Whether the per-step outputs are to be\n *   concatenated into a single tensor and returned (as the second return\n *   value). Default: `false`. This arg is included so that the relatively\n *   expensive concatenation of the stepwise outputs can be omitted unless\n *   the stepwise outputs need to be kept (e.g., for an LSTM layer of which\n *   `returnSequence` is `true`.)\n * @returns An Array: `[lastOutput, outputs, newStates]`.\n *   lastOutput: the lastest output of the RNN, of shape `[samples, ...]`.\n *   outputs: tensor with shape `[samples, time, ...]` where each entry\n *     `output[s, t]` is the output of the step function at time `t` for sample\n *     `s`. This return value is provided if and only if the\n *     `needPerStepOutputs` is set as `true`. If it is set as `false`, this\n *     return value will be `undefined`.\n *   newStates: Array of tensors, latest states returned by the step function,\n *      of shape `(samples, ...)`.\n * @throws ValueError If input dimension is less than 3.\n *\n * TODO(nielsene): This needs to be tidy-ed.\n */\nexport function rnn(\n    stepFunction: RnnStepFunction, inputs: Tensor, initialStates: Tensor[],\n    goBackwards = false, mask?: Tensor, constants?: Tensor[], unroll = false,\n    needPerStepOutputs = false): [Tensor, Tensor, Tensor[]] {\n  return tfc.tidy(() => {\n    const ndim = inputs.shape.length;\n    if (ndim < 3) {\n      throw new ValueError(`Input should be at least 3D, but is ${ndim}D.`);\n    }\n\n    // Transpose to time-major, i.e., from [batch, time, ...] to [time, batch,\n    // ...].\n    const axes = [1, 0].concat(math_utils.range(2, ndim));\n    inputs = tfc.transpose(inputs, axes);\n\n    if (constants != null) {\n      throw new NotImplementedError(\n          'The rnn() functoin of the deeplearn.js backend does not support ' +\n          'constants yet.');\n    }\n\n    // Porting Note: the unroll option is ignored by the imperative backend.\n    if (unroll) {\n      console.warn(\n          'Backend rnn(): the unroll = true option is not applicable to the ' +\n          'imperative deeplearn.js backend.');\n    }\n\n    if (mask != null) {\n      mask = mask.asType('bool').asType('float32');\n      if (mask.rank === ndim - 1) {\n        mask = tfc.expandDims(mask, -1);\n      }\n      mask = tfc.transpose(mask, axes);\n    }\n\n    if (goBackwards) {\n      inputs = tfc.reverse(inputs, 0);\n      if (mask != null) {\n        mask = tfc.reverse(mask, 0);\n      }\n    }\n\n    // Porting Note: PyKeras with TensorFlow backend uses a symbolic loop\n    //   (tf.while_loop). But for the imperative deeplearn.js backend, we just\n    //   use the usual TypeScript control flow to iterate over the time steps in\n    //   the inputs.\n    // Porting Note: PyKeras patches a \"_use_learning_phase\" attribute to\n    // outputs.\n    //   This is not idiomatic in TypeScript. The info regarding whether we are\n    //   in a learning (i.e., training) phase for RNN is passed in a different\n    //   way.\n\n    const perStepOutputs: Tensor[] = [];\n    let lastOutput: Tensor;\n    let states = initialStates;\n    const timeSteps = inputs.shape[0];\n    const perStepInputs = tfc.unstack(inputs);\n    let perStepMasks: Tensor[];\n    if (mask != null) {\n      perStepMasks = tfc.unstack(mask);\n    }\n\n    for (let t = 0; t < timeSteps; ++t) {\n      const currentInput = perStepInputs[t];\n      const stepOutputs = tfc.tidy(() => stepFunction(currentInput, states));\n\n      if (mask == null) {\n        lastOutput = stepOutputs[0];\n        states = stepOutputs[1];\n      } else {\n        const maskedOutputs = tfc.tidy(() => {\n          const stepMask = perStepMasks[t];\n          const negStepMask = tfc.onesLike(stepMask).sub(stepMask);\n          // TODO(cais): Would tfc.where() be better for performance?\n          const output =\n              stepOutputs[0].mul(stepMask).add(states[0].mul(negStepMask));\n          const newStates = states.map((state, i) => {\n            return stepOutputs[1][i].mul(stepMask).add(state.mul(negStepMask));\n          });\n          return {output, newStates};\n        });\n        lastOutput = maskedOutputs.output;\n        states = maskedOutputs.newStates;\n      }\n\n      if (needPerStepOutputs) {\n        perStepOutputs.push(lastOutput);\n      }\n    }\n    let outputs: Tensor;\n    if (needPerStepOutputs) {\n      const axis = 1;\n      outputs = tfc.stack(perStepOutputs, axis);\n    }\n    return [lastOutput, outputs, states] as [Tensor, Tensor, Tensor[]];\n  });\n}\n\nexport declare interface BaseRNNLayerArgs extends LayerArgs {\n  /**\n   * A RNN cell instance. A RNN cell is a class that has:\n   *   - a `call()` method, which takes `[Tensor, Tensor]` as the\n   *     first input argument. The first item is the input at time t, and\n   *     second item is the cell state at time t.\n   *     The `call()` method returns `[outputAtT, statesAtTPlus1]`.\n   *     The `call()` method of the cell can also take the argument `constants`,\n   *     see section \"Note on passing external constants\" below.\n   *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,\n   *       which are Layer subtypes, to accept two arguments. tfjs-layers does\n   *       not do such overriding. Instead we preseve the `call()` signature,\n   *       which due to its `Tensor|Tensor[]` argument and return value, is\n   *       flexible enough to handle the inputs and states.\n   *   - a `stateSize` attribute. This can be a single integer (single state)\n   *     in which case it is the size of the recurrent state (which should be\n   *     the same as the size of the cell output). This can also be an Array of\n   *     integers (one size per state). In this case, the first entry\n   *     (`stateSize[0]`) should be the same as the size of the cell output.\n   * It is also possible for `cell` to be a list of RNN cell instances, in which\n   * case the cells get stacked on after the other in the RNN, implementing an\n   * efficient stacked RNN.\n   */\n  cell?: RNNCell|RNNCell[];\n\n  /**\n   * Whether to return the last output in the output sequence, or the full\n   * sequence.\n   */\n  returnSequences?: boolean;\n\n  /**\n   * Whether to return the last state in addition to the output.\n   */\n  returnState?: boolean;\n\n  /**\n   * If `true`, process the input sequence backwards and return the reversed\n   * sequence (default: `false`).\n   */\n  goBackwards?: boolean;\n\n  /**\n   * If `true`, the last state for each sample at index i in a batch will be\n   * used as initial state of the sample of index i in the following batch\n   * (default: `false`).\n   *\n   * You can set RNN layers to be \"stateful\", which means that the states\n   * computed for the samples in one batch will be reused as initial states\n   * for the samples in the next batch. This assumes a one-to-one mapping\n   * between samples in different successive batches.\n   *\n   * To enable \"statefulness\":\n   *   - specify `stateful: true` in the layer constructor.\n   *   - specify a fixed batch size for your model, by passing\n   *     - if sequential model:\n   *       `batchInputShape: [...]` to the first layer in your model.\n   *     - else for functional model with 1 or more Input layers:\n   *       `batchShape: [...]` to all the first layers in your model.\n   *     This is the expected shape of your inputs\n   *     *including the batch size*.\n   *     It should be a tuple of integers, e.g., `[32, 10, 100]`.\n   *   - specify `shuffle: false` when calling `LayersModel.fit()`.\n   *\n   * To reset the state of your model, call `resetStates()` on either the\n   * specific layer or on the entire model.\n   */\n  stateful?: boolean;\n  // TODO(cais): Explore whether we can warn users when they fail to set\n  //   `shuffle: false` when training a model consisting of stateful RNNs\n  //   and any stateful Layers in general.\n\n  /**\n   * If `true`, the network will be unrolled, else a symbolic loop will be\n   * used. Unrolling can speed-up a RNN, although it tends to be more memory-\n   * intensive. Unrolling is only suitable for short sequences (default:\n   * `false`).\n   * Porting Note: tfjs-layers has an imperative backend. RNNs are executed with\n   *   normal TypeScript control flow. Hence this property is inapplicable and\n   *   ignored in tfjs-layers.\n   */\n  unroll?: boolean;\n\n  /**\n   * Dimensionality of the input (integer).\n   *   This option (or alternatively, the option `inputShape`) is required when\n   *   this layer is used as the first layer in a model.\n   */\n  inputDim?: number;\n\n  /**\n   * Length of the input sequences, to be specified when it is constant.\n   * This argument is required if you are going to connect `Flatten` then\n   * `Dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed). Note that if the recurrent layer is not the first layer in\n   * your model, you would need to specify the input length at the level of the\n   * first layer (e.g., via the `inputShape` option).\n   */\n  inputLength?: number;\n}\n\nexport class RNN extends Layer {\n  /** @nocollapse */\n  static className = 'RNN';\n  public readonly cell: RNNCell;\n  public readonly returnSequences: boolean;\n  public readonly returnState: boolean;\n  public readonly goBackwards: boolean;\n  public readonly unroll: boolean;\n\n  public stateSpec: InputSpec[];\n  protected states_: Tensor[];\n\n  // NOTE(cais): For stateful RNNs, the old states cannot be disposed right\n  // away when new states are set, because the old states may need to be used\n  // later for backpropagation through time (BPTT) and other purposes. So we\n  // keep them here for final disposal when the state is reset completely\n  // (i.e., through no-arg call to `resetStates()`).\n  protected keptStates: Tensor[][];\n\n  private numConstants: number;\n\n  constructor(args: RNNLayerArgs) {\n    super(args);\n    let cell: RNNCell;\n    if (args.cell == null) {\n      throw new ValueError(\n          'cell property is missing for the constructor of RNN.');\n    } else if (Array.isArray(args.cell)) {\n      cell = new StackedRNNCells({cells: args.cell});\n    } else {\n      cell = args.cell;\n    }\n    if (cell.stateSize == null) {\n      throw new ValueError(\n          'The RNN cell should have an attribute `stateSize` (tuple of ' +\n          'integers, one integer per RNN state).');\n    }\n    this.cell = cell;\n    this.returnSequences =\n        args.returnSequences == null ? false : args.returnSequences;\n    this.returnState = args.returnState == null ? false : args.returnState;\n    this.goBackwards = args.goBackwards == null ? false : args.goBackwards;\n    this._stateful = args.stateful == null ? false : args.stateful;\n    this.unroll = args.unroll == null ? false : args.unroll;\n\n    this.supportsMasking = true;\n    this.inputSpec = [new InputSpec({ndim: 3})];\n    this.stateSpec = null;\n    this.states_ = null;\n    // TODO(cais): Add constantsSpec and numConstants.\n    this.numConstants = null;\n    // TODO(cais): Look into the use of initial_state in the kwargs of the\n    //   constructor.\n\n    this.keptStates = [];\n  }\n\n  // Porting Note: This is the equivalent of `RNN.states` property getter in\n  //   PyKeras.\n  getStates(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      return math_utils.range(0, numStates).map(x => null);\n    } else {\n      return this.states_;\n    }\n  }\n\n  // Porting Note: This is the equivalent of the `RNN.states` property setter in\n  //   PyKeras.\n  setStates(states: Tensor[]): void {\n    this.states_ = states;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    // TODO(cais): Remove the casting once stacked RNN cells become supported.\n    let stateSize = this.cell.stateSize;\n    if (!Array.isArray(stateSize)) {\n      stateSize = [stateSize];\n    }\n    const outputDim = stateSize[0];\n    let outputShape: Shape|Shape[];\n    if (this.returnSequences) {\n      outputShape = [inputShape[0], inputShape[1], outputDim];\n    } else {\n      outputShape = [inputShape[0], outputDim];\n    }\n\n    if (this.returnState) {\n      const stateShape: Shape[] = [];\n      for (const dim of stateSize) {\n        stateShape.push([inputShape[0], dim]);\n      }\n      return [outputShape].concat(stateShape);\n    } else {\n      return outputShape;\n    }\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    return tfc.tidy(() => {\n      if (Array.isArray(mask)) {\n        mask = mask[0];\n      }\n      const outputMask = this.returnSequences ? mask : null;\n\n      if (this.returnState) {\n        const stateMask = this.states.map(s => null);\n        return [outputMask].concat(stateMask);\n      } else {\n        return outputMask;\n      }\n    });\n  }\n\n  /**\n   * Get the current state tensors of the RNN.\n   *\n   * If the state hasn't been set, return an array of `null`s of the correct\n   * length.\n   */\n  get states(): Tensor[] {\n    if (this.states_ == null) {\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      const output: Tensor[] = [];\n      for (let i = 0; i < numStates; ++i) {\n        output.push(null);\n      }\n      return output;\n    } else {\n      return this.states_;\n    }\n  }\n\n  set states(s: Tensor[]) {\n    this.states_ = s;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    // Note inputShape will be an Array of Shapes of initial states and\n    // constants if these are passed in apply().\n    const constantShape: Shape[] = null;\n    if (this.numConstants != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    }\n\n    if (isArrayOfShapes(inputShape)) {\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n\n    const batchSize: number = this.stateful ? inputShape[0] : null;\n    const inputDim = inputShape.slice(2);\n    this.inputSpec[0] = new InputSpec({shape: [batchSize, null, ...inputDim]});\n\n    // Allow cell (if RNNCell Layer) to build before we set or validate\n    // stateSpec.\n    const stepInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (constantShape != null) {\n      throw new NotImplementedError(\n          'Constants support is not implemented in RNN yet.');\n    } else {\n      this.cell.build(stepInputShape);\n    }\n\n    // Set or validate stateSpec.\n    let stateSize: number[];\n    if (Array.isArray(this.cell.stateSize)) {\n      stateSize = this.cell.stateSize;\n    } else {\n      stateSize = [this.cell.stateSize];\n    }\n\n    if (this.stateSpec != null) {\n      if (!util.arraysEqual(\n              this.stateSpec.map(spec => spec.shape[spec.shape.length - 1]),\n              stateSize)) {\n        throw new ValueError(\n            `An initialState was passed that is not compatible with ` +\n            `cell.stateSize. Received stateSpec=${this.stateSpec}; ` +\n            `However cell.stateSize is ${this.cell.stateSize}`);\n      }\n    } else {\n      this.stateSpec =\n          stateSize.map(dim => new InputSpec({shape: [null, dim]}));\n    }\n    if (this.stateful) {\n      this.resetStates();\n    }\n  }\n\n  /**\n   * Reset the state tensors of the RNN.\n   *\n   * If the `states` argument is `undefined` or `null`, will set the\n   * state tensor(s) of the RNN to all-zero tensors of the appropriate\n   * shape(s).\n   *\n   * If `states` is provided, will set the state tensors of the RNN to its\n   * value.\n   *\n   * @param states Optional externally-provided initial states.\n   * @param training Whether this call is done during training. For stateful\n   *   RNNs, this affects whether the old states are kept or discarded. In\n   *   particular, if `training` is `true`, the old states will be kept so\n   *   that subsequent backpropgataion through time (BPTT) may work properly.\n   *   Else, the old states will be discarded.\n   */\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n      const batchSize = this.inputSpec[0].shape[0];\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n      // Initialize state if null.\n      if (this.states_ == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_ = [tfc.zeros([batchSize, this.cell.stateSize])];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ =\n              this.cell.stateSize.map(dim => tfc.zeros([batchSize, dim]));\n        } else {\n          this.states_[0] = tfc.zeros([batchSize, this.cell.stateSize]);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training === true) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n          const dim = Array.isArray(this.cell.stateSize) ?\n              this.cell.stateSize[index] :\n              this.cell.stateSize;\n          const expectedShape = [batchSize, dim];\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n          this.states_[index] = value;\n        }\n      }\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    // TODO(cais): Figure out whether initialState is in kwargs or inputs.\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    // If any of `initial_state` or `constants` are specified and are\n    // `tf.SymbolicTensor`s, then add them to the inputs and temporarily modify\n    // the input_spec to include them.\n\n    let additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    let additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      kwargs['initialState'] = initialState;\n      additionalInputs = additionalInputs.concat(initialState);\n      this.stateSpec = [];\n      for (const state of initialState) {\n        this.stateSpec.push(new InputSpec({shape: state.shape}));\n      }\n      // TODO(cais): Use the following instead.\n      // this.stateSpec = initialState.map(state => new InputSpec({shape:\n      // state.shape}));\n      additionalSpecs = additionalSpecs.concat(this.stateSpec);\n    }\n    if (constants != null) {\n      kwargs['constants'] = constants;\n      additionalInputs = additionalInputs.concat(constants);\n      // TODO(cais): Add this.constantsSpec.\n      this.numConstants = constants.length;\n    }\n\n    const isTensor = additionalInputs[0] instanceof SymbolicTensor;\n    if (isTensor) {\n      // Compute full input spec, including state and constants.\n      const fullInput =\n          [inputs].concat(additionalInputs) as Tensor[] | SymbolicTensor[];\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call with temporarily replaced inputSpec.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output = super.apply(fullInput, kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  // tslint:disable-next-line:no-any\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    // Input shape: `[samples, time (padded with zeros), input_dim]`.\n    // Note that the .build() method of subclasses **must** define\n    // this.inputSpec and this.stateSpec owith complete input shapes.\n    return tidy(() => {\n      const mask = kwargs == null ? null : kwargs['mask'] as Tensor;\n      const training = kwargs == null ? null : kwargs['training'];\n      let initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      inputs = getExactlyOneTensor(inputs);\n      if (initialState == null) {\n        if (this.stateful) {\n          initialState = this.states_;\n        } else {\n          initialState = this.getInitialState(inputs);\n        }\n      }\n\n      const numStates =\n          Array.isArray(this.cell.stateSize) ? this.cell.stateSize.length : 1;\n      if (initialState.length !== numStates) {\n        throw new ValueError(\n            `RNN Layer has ${numStates} state(s) but was passed ` +\n            `${initialState.length} initial state(s).`);\n      }\n      if (this.unroll) {\n        console.warn(\n            'Ignoring unroll = true for RNN layer, due to imperative backend.');\n      }\n\n      const cellCallKwargs: Kwargs = {training};\n\n      // TODO(cais): Add support for constants.\n      const step = (inputs: Tensor, states: Tensor[]) => {\n        // `inputs` and `states` are concatenated to form a single `Array` of\n        // `tf.Tensor`s as the input to `cell.call()`.\n        const outputs =\n            this.cell.call([inputs].concat(states), cellCallKwargs) as Tensor[];\n        // Marshall the return value into output and new states.\n        return [outputs[0], outputs.slice(1)] as [Tensor, Tensor[]];\n      };\n\n      // TODO(cais): Add support for constants.\n\n      const rnnOutputs =\n          rnn(step, inputs, initialState, this.goBackwards, mask, null,\n              this.unroll, this.returnSequences);\n      const lastOutput = rnnOutputs[0];\n      const outputs = rnnOutputs[1];\n      const states = rnnOutputs[2];\n\n      if (this.stateful) {\n        this.resetStates(states, training);\n      }\n\n      const output = this.returnSequences ? outputs : lastOutput;\n\n      // TODO(cais): Porperty set learning phase flag.\n\n      if (this.returnState) {\n        return [output].concat(states);\n      } else {\n        return output;\n      }\n    });\n  }\n\n  getInitialState(inputs: Tensor): Tensor[] {\n    return tidy(() => {\n      // Build an all-zero tensor of shape [samples, outputDim].\n      // [Samples, timeSteps, inputDim].\n      let initialState = tfc.zeros(inputs.shape);\n      // [Samples].\n      initialState = tfc.sum(initialState, [1, 2]);\n      initialState = K.expandDims(initialState);  // [Samples, 1].\n\n      if (Array.isArray(this.cell.stateSize)) {\n        return this.cell.stateSize.map(\n            dim => dim > 1 ? K.tile(initialState, [1, dim]) : initialState);\n      } else {\n        return this.cell.stateSize > 1 ?\n            [K.tile(initialState, [1, this.cell.stateSize])] :\n            [initialState];\n      }\n    });\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    return this.cell.trainableWeights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    // Porting Note: In TypeScript, `this` is always an instance of `Layer`.\n    if (!this.trainable) {\n      return this.cell.weights;\n    }\n    return this.cell.nonTrainableWeights;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.cell != null) {\n      this.cell.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      returnSequences: this.returnSequences,\n      returnState: this.returnState,\n      goBackwards: this.goBackwards,\n      stateful: this.stateful,\n      unroll: this.unroll,\n    };\n\n    if (this.numConstants != null) {\n      config['numConstants'] = this.numConstants;\n    }\n\n    const cellConfig = this.cell.getConfig();\n\n    if (this.getClassName() === RNN.className) {\n      config['cell'] = {\n        'className': this.cell.getClassName(),\n        'config': cellConfig,\n      } as serialization.ConfigDictValue;\n    }\n\n    // this order is necessary, to prevent cell name from replacing layer name\n    return {...cellConfig, ...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cellConfig = config['cell'] as serialization.ConfigDict;\n    const cell = deserialize(cellConfig, customObjects) as RNNCell;\n    return new cls(Object.assign(config, {cell}));\n  }\n}\nserialization.registerClass(RNN);\n\n// Porting Note: This is a common parent class for RNN cells. There is no\n// equivalent of this in PyKeras. Having a common parent class forgoes the\n//  need for `has_attr(cell, ...)` checks or its TypeScript equivalent.\n/**\n * An RNNCell layer.\n *\n * @doc {heading: 'Layers', subheading: 'Classes'}\n */\nexport abstract class RNNCell extends Layer {\n  /**\n   * Size(s) of the states.\n   * For RNN cells with only a single state, this is a single integer.\n   */\n  public stateSize: number|number[];\n  public dropoutMask: Tensor|Tensor[];\n  public recurrentDropoutMask: Tensor|Tensor[];\n}\n\nexport declare interface SimpleRNNCellLayerArgs extends LayerArgs {\n  /**\n   * units: Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   * Default: hyperbolic tangent ('tanh').\n   * If you pass `null`,  'linear' activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the `kernel` weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the `recurrent_kernel` weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the `kernel` weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the `recurrentKernel` weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraintfunction applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Float number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\nexport class SimpleRNNCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'SimpleRNNCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  constructor(args: SimpleRNNCellLayerArgs) {\n    super(args);\n    this.units = args.units;\n    assertPositiveInteger(this.units, `units`);\n    this.activation = getActivation(\n        args.activation == null ? this.DEFAULT_ACTIVATION : args.activation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    // TODO(cais): Use regularizer.\n    this.kernel = this.addWeight(\n        'kernel', [inputShape[inputShape.length - 1], this.units], null,\n        this.kernelInitializer, this.kernelRegularizer, true,\n        this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    this.built = true;\n  }\n\n  // Porting Note: PyKeras' equivalent of this method takes two tensor inputs:\n  //   `inputs` and `states`. Here, the two tensors are combined into an\n  //   `Tensor[]` Array as the first input argument.\n  //   Similarly, PyKeras' equivalent of this method returns two values:\n  //    `output` and `[output]`. Here the two are combined into one length-2\n  //    `Tensor[]`, consisting of `output` repeated.\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `SimpleRNNCell expects 2 input Tensors, got ${inputs.length}.`);\n      }\n      let prevOutput = inputs[1];\n      inputs = inputs[0];\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training\n                           }) as Tensor;\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(prevOutput),\n                                      rate: this.recurrentDropout,\n                                      training\n                                    }) as Tensor;\n      }\n      let h: Tensor;\n      const dpMask: Tensor = this.dropoutMask as Tensor;\n      const recDpMask: Tensor = this.recurrentDropoutMask as Tensor;\n      if (dpMask != null) {\n        h = K.dot(tfc.mul(inputs, dpMask), this.kernel.read());\n      } else {\n        h = K.dot(inputs, this.kernel.read());\n      }\n      if (this.bias != null) {\n        h = K.biasAdd(h, this.bias.read());\n      }\n      if (recDpMask != null) {\n        prevOutput = tfc.mul(prevOutput, recDpMask);\n      }\n      let output = tfc.add(h, K.dot(prevOutput, this.recurrentKernel.read()));\n      if (this.activation != null) {\n        output = this.activation.apply(output);\n      }\n\n      // TODO(cais): Properly set learning phase on output tensor?\n      return [output, output];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(SimpleRNNCell);\n\nexport declare interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {\n  /**\n   * Positive integer, dimensionality of the output space.\n   */\n  units: number;\n\n  /**\n   * Activation function to use.\n   *\n   * Defaults to  hyperbolic tangent (`tanh`)\n   *\n   * If you pass `null`, no activation will be applied.\n   */\n  activation?: ActivationIdentifier;\n\n  /**\n   * Whether the layer uses a bias vector.\n   */\n  useBias?: boolean;\n\n  /**\n   * Initializer for the `kernel` weights matrix, used for the linear\n   * transformation of the inputs.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the `recurrentKernel` weights matrix, used for\n   * linear transformation of the recurrent state.\n   */\n  recurrentInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Regularizer function applied to the kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the recurrentKernel weights matrix.\n   */\n  recurrentRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Constraint function applied to the kernel weights matrix.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the recurrentKernel weights matrix.\n   */\n  recurrentConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint function applied to the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the inputs.\n   */\n  dropout?: number;\n\n  /**\n   * Number between 0 and 1. Fraction of the units to drop for the linear\n   * transformation of the recurrent state.\n   */\n  recurrentDropout?: number;\n}\n\n/**\n * RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the\n * `cell` property required. This interface is to be used with constructors\n * of concrete RNN layer subtypes.\n */\nexport declare interface RNNLayerArgs extends BaseRNNLayerArgs {\n  cell: RNNCell|RNNCell[];\n}\n\nexport class SimpleRNN extends RNN {\n  /** @nocollapse */\n  static className = 'SimpleRNN';\n  constructor(args: SimpleRNNLayerArgs) {\n    args.cell = new SimpleRNNCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\nserialization.registerClass(SimpleRNN);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n\n  /**\n   * GRU convention (whether to apply reset gate after or before matrix\n   * multiplication). false = \"before\", true = \"after\" (only false is\n   * supported).\n   */\n  resetAfter?: boolean;\n}\n\nexport class GRUCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'GRUCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number;\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION: ActivationIdentifier = 'hardSigmoid';\n\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: GRUCellLayerArgs) {\n    super(args);\n    if (args.resetAfter) {\n      throw new ValueError(\n          `GRUCell does not support reset_after parameter set to true.`);\n    }\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = this.units;\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 3], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 3], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    if (this.useBias) {\n      this.bias = this.addWeight(\n          'bias', [this.units * 3], null, this.biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 2) {\n        throw new ValueError(\n            `GRUCell expects 2 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      let hTMinus1 = inputs[1];  // Previous memory state.\n      inputs = inputs[0];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2, regardless of the actual value of\n      // config.implementation.\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 3\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 3\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor];\n      const recDpMask = this.recurrentDropoutMask as [Tensor, Tensor, Tensor];\n      let z: Tensor;\n      let r: Tensor;\n      let hh: Tensor;\n\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let matrixX = K.dot(inputs, this.kernel.read());\n      if (this.useBias) {\n        matrixX = K.biasAdd(matrixX, this.bias.read());\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n\n      const recurrentKernelValue = this.recurrentKernel.read();\n      const [rk1, rk2] = tfc.split(\n          recurrentKernelValue, [2 * this.units, this.units],\n          recurrentKernelValue.rank - 1);\n      const matrixInner = K.dot(hTMinus1, rk1);\n\n      const [xZ, xR, xH] = tfc.split(matrixX, 3, matrixX.rank - 1);\n      const [recurrentZ, recurrentR] =\n          tfc.split(matrixInner, 2, matrixInner.rank - 1);\n      z = this.recurrentActivation.apply(tfc.add(xZ, recurrentZ));\n      r = this.recurrentActivation.apply(tfc.add(xR, recurrentR));\n\n      const recurrentH = K.dot(tfc.mul(r, hTMinus1), rk2);\n      hh = this.activation.apply(tfc.add(xH, recurrentH));\n\n      const h =\n          tfc.add(tfc.mul(z, hTMinus1), tfc.mul(tfc.add(1, tfc.neg(z)), hh));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n      resetAfter: false\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(GRUCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface GRULayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   * smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class GRU extends RNN {\n  /** @nocollapse */\n  static className = 'GRU';\n  constructor(args: GRULayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new GRUCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(GRU);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we extend\n//   that interface instead of repeating the fields.\nexport declare interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *\n   * Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions.\n   *\n   * Mode 2 will batch them into fewer, larger operations. These modes will\n   * have different performance profiles on different hardware and\n   * for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this configuration field.\n   */\n  implementation?: number;\n}\n\nexport class LSTMCell extends RNNCell {\n  /** @nocollapse */\n  static className = 'LSTMCell';\n  readonly units: number;\n  readonly activation: Activation;\n  readonly recurrentActivation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n  readonly unitForgetBias: boolean;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n\n  readonly stateSize: number[];\n  readonly implementation: number;\n\n  readonly DEFAULT_ACTIVATION = 'tanh';\n  readonly DEFAULT_RECURRENT_ACTIVATION = 'hardSigmoid';\n  readonly DEFAULT_KERNEL_INITIALIZER = 'glorotNormal';\n  readonly DEFAULT_RECURRENT_INITIALIZER = 'orthogonal';\n\n  readonly DEFAULT_BIAS_INITIALIZER = 'zeros';\n\n  kernel: LayerVariable;\n  recurrentKernel: LayerVariable;\n  bias: LayerVariable;\n\n  constructor(args: LSTMCellLayerArgs) {\n    super(args);\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(\n        args.activation === undefined ? this.DEFAULT_ACTIVATION :\n                                        args.activation);\n    this.recurrentActivation = getActivation(\n        args.recurrentActivation === undefined ?\n            this.DEFAULT_RECURRENT_ACTIVATION :\n            args.recurrentActivation);\n    this.useBias = args.useBias == null ? true : args.useBias;\n\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.recurrentInitializer = getInitializer(\n        args.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER);\n\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.unitForgetBias = args.unitForgetBias;\n\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.recurrentRegularizer = getRegularizer(args.recurrentRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.recurrentConstraint = getConstraint(args.recurrentConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n\n    this.dropout = math_utils.min(\n        [1, math_utils.max([0, args.dropout == null ? 0 : args.dropout])]);\n    this.recurrentDropout = math_utils.min([\n      1,\n      math_utils.max(\n          [0, args.recurrentDropout == null ? 0 : args.recurrentDropout])\n    ]);\n    this.implementation = args.implementation;\n    this.stateSize = [this.units, this.units];\n    this.dropoutMask = null;\n    this.recurrentDropoutMask = null;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputDim = inputShape[inputShape.length - 1];\n    this.kernel = this.addWeight(\n        'kernel', [inputDim, this.units * 4], null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', [this.units, this.units * 4], null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n    let biasInitializer: Initializer;\n    if (this.useBias) {\n      if (this.unitForgetBias) {\n        const capturedBiasInit = this.biasInitializer;\n        const capturedUnits = this.units;\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): Tensor {\n            // TODO(cais): More informative variable names?\n            const bI = capturedBiasInit.apply([capturedUnits]);\n            const bF = (new Ones()).apply([capturedUnits]);\n            const bCAndH = capturedBiasInit.apply([capturedUnits * 2]);\n            return K.concatAlongFirstAxis(\n                K.concatAlongFirstAxis(bI, bF), bCAndH);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n      this.bias = this.addWeight(\n          'bias', [this.units * 4], null, biasInitializer, this.biasRegularizer,\n          true, this.biasConstraint);\n    } else {\n      this.bias = null;\n    }\n    // Porting Notes: Unlike the PyKeras implementation, we perform slicing\n    //   of the weights and bias in the call() method, at execution time.\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      inputs = inputs as Tensor[];\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `LSTMCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n      let hTMinus1 = inputs[1];    // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n      inputs = inputs[0];\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(inputs as Tensor),\n                             rate: this.dropout,\n                             training,\n                             count: 4\n                           }) as Tensor[];\n      }\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: 4\n                                    }) as Tensor[];\n      }\n      const dpMask = this.dropoutMask as [Tensor, Tensor, Tensor, Tensor];\n      const recDpMask =\n          this.recurrentDropoutMask as [Tensor, Tensor, Tensor, Tensor];\n\n      // Note: For superior performance, TensorFlow.js always uses\n      // implementation 2 regardless of the actual value of\n      // config.implementation.\n      let i: Tensor;\n      let f: Tensor;\n      let c: Tensor;\n      let o: Tensor;\n      if (0 < this.dropout && this.dropout < 1) {\n        inputs = tfc.mul(inputs, dpMask[0]);\n      }\n      let z = K.dot(inputs, this.kernel.read());\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1) {\n        hTMinus1 = tfc.mul(hTMinus1, recDpMask[0]);\n      }\n      z = tfc.add(z, K.dot(hTMinus1, this.recurrentKernel.read()));\n      if (this.useBias) {\n        z = K.biasAdd(z, this.bias.read());\n      }\n\n      const [z0, z1, z2, z3] = tfc.split(z, 4, z.rank - 1);\n\n      i = this.recurrentActivation.apply(z0);\n      f = this.recurrentActivation.apply(z1);\n      c = tfc.add(tfc.mul(f, cTMinus1), tfc.mul(i, this.activation.apply(z2)));\n      o = this.recurrentActivation.apply(z3);\n\n      const h = tfc.mul(o, this.activation.apply(c));\n      // TODO(cais): Add use_learning_phase flag properly.\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      recurrentActivation: serializeActivation(this.recurrentActivation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      recurrentInitializer: serializeInitializer(this.recurrentInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      unitForgetBias: this.unitForgetBias,\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      recurrentRegularizer: serializeRegularizer(this.recurrentRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      recurrentConstraint: serializeConstraint(this.recurrentConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint),\n      dropout: this.dropout,\n      recurrentDropout: this.recurrentDropout,\n      implementation: this.implementation,\n    };\n\n    return {...baseConfig, ...config};\n  }\n}\nserialization.registerClass(LSTMCell);\n\n// Porting Note: Since this is a superset of SimpleRNNLayerConfig, we inherit\n//   from that interface instead of repeating the fields here.\nexport declare interface LSTMLayerArgs extends SimpleRNNLayerArgs {\n  /**\n   * Activation function to use for the recurrent step.\n   *\n   * Defaults to hard sigmoid (`hardSigmoid`).\n   *\n   * If `null`, no activation is applied.\n   */\n  recurrentActivation?: ActivationIdentifier;\n\n  /**\n   * If `true`, add 1 to the bias of the forget gate at initialization.\n   * Setting it to `true` will also force `biasInitializer = 'zeros'`.\n   * This is recommended in\n   * [Jozefowicz et\n   * al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).\n   */\n  unitForgetBias?: boolean;\n\n  /**\n   * Implementation mode, either 1 or 2.\n   *   Mode 1 will structure its operations as a larger number of\n   *   smaller dot products and additions, whereas mode 2 will\n   *   batch them into fewer, larger operations. These modes will\n   *   have different performance profiles on different hardware and\n   *   for different applications.\n   *\n   * Note: For superior performance, TensorFlow.js always uses implementation\n   * 2, regardless of the actual value of this config field.\n   */\n  implementation?: number;\n}\n\nexport class LSTM extends RNN {\n  /** @nocollapse */\n  static className = 'LSTM';\n  constructor(args: LSTMLayerArgs) {\n    if (args.implementation === 0) {\n      console.warn(\n          '`implementation=0` has been deprecated, and now defaults to ' +\n          '`implementation=1`. Please update your layer call.');\n    }\n    args.cell = new LSTMCell(args);\n    super(args as RNNLayerArgs);\n    // TODO(cais): Add activityRegularizer.\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n        this.cell.dropoutMask = null;\n      }\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n        this.cell.recurrentDropoutMask = null;\n      }\n      const mask = kwargs == null ? null : kwargs['mask'];\n      const training = kwargs == null ? null : kwargs['training'];\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    if (config['implmentation'] === 0) {\n      config['implementation'] = 1;\n    }\n    return new cls(config);\n  }\n}\nserialization.registerClass(LSTM);\n\nexport declare interface StackedRNNCellsArgs extends LayerArgs {\n  /**\n   * A `Array` of `RNNCell` instances.\n   */\n  cells: RNNCell[];\n}\n\nexport class StackedRNNCells extends RNNCell {\n  /** @nocollapse */\n  static className = 'StackedRNNCells';\n  protected cells: RNNCell[];\n\n  constructor(args: StackedRNNCellsArgs) {\n    super(args);\n    this.cells = args.cells;\n  }\n\n  get stateSize(): number[] {\n    // States are a flat list in reverse order of the cell stack.\n    // This allows perserving the requirement `stack.statesize[0] ===\n    // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,\n    // assuming one LSTM has states `[h, c]`.\n    const stateSize: number[] = [];\n    for (const cell of this.cells.slice().reverse()) {\n      if (Array.isArray(cell.stateSize)) {\n        stateSize.push(...cell.stateSize);\n      } else {\n        stateSize.push(cell.stateSize);\n      }\n    }\n    return stateSize;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      let states = inputs.slice(1);\n\n      // Recover per-cell states.\n      const nestedStates: Tensor[][] = [];\n      for (const cell of this.cells.slice().reverse()) {\n        if (Array.isArray(cell.stateSize)) {\n          nestedStates.push(states.splice(0, cell.stateSize.length));\n        } else {\n          nestedStates.push(states.splice(0, 1));\n        }\n      }\n      nestedStates.reverse();\n\n      // Call the cells in order and store the returned states.\n      const newNestedStates: Tensor[][] = [];\n      let callInputs: Tensor[];\n      for (let i = 0; i < this.cells.length; ++i) {\n        const cell = this.cells[i];\n        states = nestedStates[i];\n        // TODO(cais): Take care of constants.\n        if (i === 0) {\n          callInputs = [inputs[0]].concat(states);\n        } else {\n          callInputs = [callInputs[0]].concat(states);\n        }\n        callInputs = cell.call(callInputs, kwargs) as Tensor[];\n        newNestedStates.push(callInputs.slice(1));\n      }\n\n      // Format the new states as a flat list in reverse cell order.\n      states = [];\n      for (const cellStates of newNestedStates.slice().reverse()) {\n        states.push(...cellStates);\n      }\n      return [callInputs[0]].concat(states);\n    });\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    if (isArrayOfShapes(inputShape)) {\n      // TODO(cais): Take care of input constants.\n      // const constantShape = inputShape.slice(1);\n      inputShape = (inputShape as Shape[])[0];\n    }\n    inputShape = inputShape as Shape;\n    let outputDim: number;\n    this.cells.forEach((cell, i) => {\n      nameScope(`RNNCell_${i}`, () => {\n        // TODO(cais): Take care of input constants.\n\n        cell.build(inputShape);\n        if (Array.isArray(cell.stateSize)) {\n          outputDim = cell.stateSize[0];\n        } else {\n          outputDim = cell.stateSize;\n        }\n        inputShape = [inputShape[0], outputDim] as Shape;\n      });\n    });\n    this.built = true;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const baseConfig = super.getConfig();\n\n    const getCellConfig = (cell: RNNCell) => {\n      return {\n        'className': cell.getClassName(),\n        'config': cell.getConfig(),\n      };\n    };\n\n    const cellConfigs = this.cells.map(getCellConfig);\n\n    const config = {'cells': cellConfigs};\n\n    return {...baseConfig, ...config};\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const cells: RNNCell[] = [];\n    for (const cellConfig of (config['cells'] as serialization.ConfigDict[])) {\n      cells.push(deserialize(cellConfig, customObjects) as RNNCell);\n    }\n    return new cls({cells});\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    if (!this.trainable) {\n      return [];\n    }\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.trainableWeights);\n    }\n    return weights;\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.nonTrainableWeights);\n    }\n    if (!this.trainable) {\n      const trainableWeights: LayerVariable[] = [];\n      for (const cell of this.cells) {\n        trainableWeights.push(...cell.trainableWeights);\n      }\n      return trainableWeights.concat(weights);\n    }\n    return weights;\n  }\n\n  /**\n   * Retrieve the weights of a the model.\n   *\n   * @returns A flat `Array` of `tf.Tensor`s.\n   */\n  getWeights(): Tensor[] {\n    const weights: LayerVariable[] = [];\n    for (const cell of this.cells) {\n      weights.push(...cell.weights);\n    }\n    return batchGetValue(weights);\n  }\n\n  /**\n   * Set the weights of the model.\n   *\n   * @param weights An `Array` of `tf.Tensor`s with shapes and types matching\n   *     the output of `getWeights()`.\n   */\n  setWeights(weights: Tensor[]): void {\n    const tuples: Array<[LayerVariable, Tensor]> = [];\n    for (const cell of this.cells) {\n      const numParams = cell.weights.length;\n      const inputWeights = weights.splice(numParams);\n      for (let i = 0; i < cell.weights.length; ++i) {\n        tuples.push([cell.weights[i], inputWeights[i]]);\n      }\n    }\n    batchSetValue(tuples);\n  }\n\n  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.\n}\nserialization.registerClass(StackedRNNCells);\n\nexport function generateDropoutMask(args: {\n  ones: () => tfc.Tensor,\n  rate: number,\n  training?: boolean,\n  count?: number,\n}): tfc.Tensor|tfc.Tensor[] {\n  const {ones, rate, training = false, count = 1} = args;\n\n  const droppedInputs = () => K.dropout(ones(), rate);\n\n  const createMask = () => K.inTrainPhase(droppedInputs, ones, training);\n\n  // just in case count is provided with null or undefined\n  if (!count || count <= 1) {\n    return tfc.keep(createMask().clone());\n  }\n\n  const masks = Array(count).fill(undefined).map(createMask);\n\n  return masks.map(m => tfc.keep(m.clone()));\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {Activation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode} from '../common';\nimport {Constraint} from '../constraints';\nimport {InputSpec} from '../engine/topology';\nimport {AttributeError, NotImplementedError, ValueError} from '../errors';\nimport {Initializer} from '../initializers';\nimport {DataFormat, DataType, PaddingMode, Shape} from '../keras_format/common';\nimport {Regularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {convOutputLength, normalizeArray} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\nimport {BaseRNNLayerArgs, generateDropoutMask, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNNCellLayerArgs} from './recurrent';\n\ndeclare interface ConvRNN2DCellArgs extends\n    Omit<SimpleRNNCellLayerArgs, 'units'> {\n  /**\n   * The dimensionality of the output space (i.e. the number of filters in the\n   * convolution).\n   */\n  filters: number;\n\n  /**\n   * The dimensions of the convolution window. If kernelSize is a number, the\n   * convolutional window will be square.\n   */\n  kernelSize: number|number[];\n\n  /**\n   * The strides of the convolution in each dimension. If strides is a number,\n   * strides in both dimensions are equal.\n   *\n   * Specifying any stride value != 1 is incompatible with specifying any\n   * `dilationRate` value != 1.\n   */\n  strides?: number|number[];\n\n  /**\n   * Padding mode.\n   */\n  padding?: PaddingMode;\n\n  /**\n   * Format of the data, which determines the ordering of the dimensions in\n   * the inputs.\n   *\n   * `channels_last` corresponds to inputs with shape\n   *   `(batch, ..., channels)`\n   *\n   *  `channels_first` corresponds to inputs with shape `(batch, channels,\n   * ...)`.\n   *\n   * Defaults to `channels_last`.\n   */\n  dataFormat?: DataFormat;\n\n  /**\n   * The dilation rate to use for the dilated convolution in each dimension.\n   * Should be an integer or array of two or three integers.\n   *\n   * Currently, specifying any `dilationRate` value != 1 is incompatible with\n   * specifying any `strides` value != 1.\n   */\n  dilationRate?: number|[number]|[number, number];\n}\n\nabstract class ConvRNN2DCell extends RNNCell {\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  readonly activation: Activation;\n  readonly useBias: boolean;\n\n  readonly kernelInitializer: Initializer;\n  readonly recurrentInitializer: Initializer;\n  readonly biasInitializer: Initializer;\n\n  readonly kernelConstraint: Constraint;\n  readonly recurrentConstraint: Constraint;\n  readonly biasConstraint: Constraint;\n\n  readonly kernelRegularizer: Regularizer;\n  readonly recurrentRegularizer: Regularizer;\n  readonly biasRegularizer: Regularizer;\n\n  readonly dropout: number;\n  readonly recurrentDropout: number;\n}\n\ndeclare interface ConvRNN2DLayerArgs extends BaseRNNLayerArgs,\n                                             ConvRNN2DCellArgs {}\n\n/**\n * Base class for convolutional-recurrent layers.\n */\nclass ConvRNN2D extends RNN {\n  /** @nocollapse */\n  static className = 'ConvRNN2D';\n\n  readonly cell: ConvRNN2DCell;\n\n  constructor(args: ConvRNN2DLayerArgs) {\n    if (args.unroll) {\n      throw new NotImplementedError(\n          'Unrolling is not possible with convolutional RNNs.');\n    }\n\n    if (Array.isArray(args.cell)) {\n      throw new NotImplementedError(\n          'It is not possible at the moment to stack convolutional cells.');\n    }\n\n    super(args as RNNLayerArgs);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tfc.tidy(() => {\n      if (this.cell.dropoutMask != null) {\n        tfc.dispose(this.cell.dropoutMask);\n\n        this.cell.dropoutMask = null;\n      }\n\n      if (this.cell.recurrentDropoutMask != null) {\n        tfc.dispose(this.cell.recurrentDropoutMask);\n\n        this.cell.recurrentDropoutMask = null;\n      }\n\n      if (kwargs && kwargs['constants']) {\n        throw new ValueError('ConvRNN2D cell does not support constants');\n      }\n\n      const mask = kwargs == null ? null : kwargs['mask'];\n\n      const training = kwargs == null ? null : kwargs['training'];\n\n      const initialState: Tensor[] =\n          kwargs == null ? null : kwargs['initialState'];\n\n      return super.call(inputs, {mask, training, initialState});\n    });\n  }\n\n  computeOutputShape(inputShape: Shape): Shape|Shape[] {\n    let outShape: Shape = this.computeSingleOutputShape(inputShape);\n\n    if (!this.returnSequences) {\n      outShape = [outShape[0], ...outShape.slice(2)];\n    }\n\n    if (this.returnState) {\n      outShape =\n          [outShape, ...Array(2).fill([inputShape[0], ...outShape.slice(-3)])];\n    }\n\n    return outShape;\n  }\n\n  getInitialState(inputs: tfc.Tensor): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      const {stateSize} = this.cell;\n\n      const inputShape = inputs.shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const initialState = tfc.zeros(stateShape);\n\n      if (Array.isArray(stateSize)) {\n        return Array(stateSize.length).fill(initialState);\n      }\n\n      return [initialState];\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[], training = false): void {\n    tfc.tidy(() => {\n      if (!this.stateful) {\n        throw new AttributeError(\n            'Cannot call resetStates() on an RNN Layer that is not stateful.');\n      }\n\n      const inputShape = this.inputSpec[0].shape;\n\n      const outputShape = this.computeSingleOutputShape(inputShape);\n\n      const stateShape = [outputShape[0], ...outputShape.slice(2)];\n\n      const batchSize = inputShape[0];\n\n      if (batchSize == null) {\n        throw new ValueError(\n            'If an RNN is stateful, it needs to know its batch size. Specify ' +\n            'the batch size of your input tensors: \\n' +\n            '- If using a Sequential model, specify the batch size by ' +\n            'passing a `batchInputShape` option to your first layer.\\n' +\n            '- If using the functional API, specify the batch size by ' +\n            'passing a `batchShape` option to your Input layer.');\n      }\n\n      // Initialize state if null.\n      if (this.getStates() == null) {\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_ = [tfc.zeros(stateShape)];\n        }\n      } else if (states == null) {\n        // Dispose old state tensors.\n        tfc.dispose(this.states_);\n\n        // For stateful RNNs, fully dispose kept old states.\n        if (this.keptStates != null) {\n          tfc.dispose(this.keptStates);\n          this.keptStates = [];\n        }\n\n        if (Array.isArray(this.cell.stateSize)) {\n          this.states_ = this.cell.stateSize.map(() => tfc.zeros(stateShape));\n        } else {\n          this.states_[0] = tfc.zeros(stateShape);\n        }\n      } else {\n        if (!Array.isArray(states)) {\n          states = [states];\n        }\n\n        if (states.length !== this.states_.length) {\n          throw new ValueError(\n              `Layer ${this.name} expects ${this.states_.length} state(s), ` +\n              `but it received ${states.length} state value(s). Input ` +\n              `received: ${states}`);\n        }\n\n        if (training) {\n          // Store old state tensors for complete disposal later, i.e., during\n          // the next no-arg call to this method. We do not dispose the old\n          // states immediately because that BPTT (among other things) require\n          // them.\n          this.keptStates.push(this.states_.slice());\n        } else {\n          tfc.dispose(this.states_);\n        }\n\n        for (let index = 0; index < this.states_.length; ++index) {\n          const value = states[index];\n\n          const expectedShape = stateShape;\n\n          if (!util.arraysEqual(value.shape, expectedShape)) {\n            throw new ValueError(\n                `State ${index} is incompatible with layer ${this.name}: ` +\n                `expected shape=${expectedShape}, received shape=${\n                    value.shape}`);\n          }\n\n          this.states_[index] = value;\n        }\n      }\n\n      this.states_ = this.states_.map(state => tfc.keep(state.clone()));\n    });\n  }\n\n  protected computeSingleOutputShape(inputShape: Shape): Shape {\n    const {dataFormat, filters, kernelSize, padding, strides, dilationRate} =\n        this.cell;\n\n    const isChannelsFirst = dataFormat === 'channelsFirst';\n\n    const h = inputShape[isChannelsFirst ? 3 : 2];\n    const w = inputShape[isChannelsFirst ? 4 : 3];\n\n    const hOut = convOutputLength(\n        h, kernelSize[0], padding, strides[0], dilationRate[0]);\n    const wOut = convOutputLength(\n        w, kernelSize[1], padding, strides[1], dilationRate[1]);\n\n    const outShape: Shape = [\n      ...inputShape.slice(0, 2),\n      ...(isChannelsFirst ? [filters, hOut, wOut] : [hOut, wOut, filters])\n    ];\n\n    return outShape;\n  }\n}\n\nexport declare interface ConvLSTM2DCellArgs extends\n    Omit<LSTMCellLayerArgs, 'units'>, ConvRNN2DCellArgs {}\n\nexport class ConvLSTM2DCell extends LSTMCell implements ConvRNN2DCell {\n  /** @nocollapse */\n  static className = 'ConvLSTM2DCell';\n\n  readonly filters: number;\n  readonly kernelSize: number[];\n  readonly strides: number[];\n  readonly padding: PaddingMode;\n  readonly dataFormat: DataFormat;\n  readonly dilationRate: number[];\n\n  constructor(args: ConvLSTM2DCellArgs) {\n    const {\n      filters,\n      kernelSize,\n      strides,\n      padding,\n      dataFormat,\n      dilationRate,\n    } = args;\n\n    super({...args, units: filters});\n\n    this.filters = filters;\n    assertPositiveInteger(this.filters, 'filters');\n\n    this.kernelSize = normalizeArray(kernelSize, 2, 'kernelSize');\n    this.kernelSize.forEach(size => assertPositiveInteger(size, 'kernelSize'));\n\n    this.strides = normalizeArray(strides || 1, 2, 'strides');\n    this.strides.forEach(stride => assertPositiveInteger(stride, 'strides'));\n\n    this.padding = padding || 'valid';\n    checkPaddingMode(this.padding);\n\n    this.dataFormat = dataFormat || 'channelsLast';\n    checkDataFormat(this.dataFormat);\n\n    this.dilationRate = normalizeArray(dilationRate || 1, 2, 'dilationRate');\n    this.dilationRate.forEach(\n        rate => assertPositiveInteger(rate, 'dilationRate'));\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n\n    const channelAxis =\n        this.dataFormat === 'channelsFirst' ? 1 : inputShape.length - 1;\n\n    if (inputShape[channelAxis] == null) {\n      throw new ValueError(\n          `The channel dimension of the input should be defined. ` +\n          `Found ${inputShape[channelAxis]}`);\n    }\n\n    const inputDim = inputShape[channelAxis];\n\n    const numOfKernels = 4;\n\n    const kernelShape =\n        this.kernelSize.concat([inputDim, this.filters * numOfKernels]);\n\n    this.kernel = this.addWeight(\n        'kernel', kernelShape, null, this.kernelInitializer,\n        this.kernelRegularizer, true, this.kernelConstraint);\n\n    const recurrentKernelShape =\n        this.kernelSize.concat([this.filters, this.filters * numOfKernels]);\n\n    this.recurrentKernel = this.addWeight(\n        'recurrent_kernel', recurrentKernelShape, null,\n        this.recurrentInitializer, this.recurrentRegularizer, true,\n        this.recurrentConstraint);\n\n    if (this.useBias) {\n      let biasInitializer: Initializer;\n\n      if (this.unitForgetBias) {\n        const init = this.biasInitializer;\n\n        const filters = this.filters;\n\n        biasInitializer = new (class CustomInit extends Initializer {\n          /** @nocollapse */\n          static className = 'CustomInit';\n\n          apply(shape: Shape, dtype?: DataType): tfc.Tensor {\n            const biasI = init.apply([filters]);\n            const biasF = tfc.ones([filters]);\n            const biasCAndO = init.apply([filters * 2]);\n            return K.concatenate([biasI, biasF, biasCAndO]);\n          }\n        })();\n      } else {\n        biasInitializer = this.biasInitializer;\n      }\n\n      this.bias = this.addWeight(\n          'bias', [this.filters * numOfKernels], null, biasInitializer,\n          this.biasRegularizer, true, this.biasConstraint);\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: tfc.Tensor[], kwargs: Kwargs): tfc.Tensor[] {\n    return tfc.tidy(() => {\n      if (inputs.length !== 3) {\n        throw new ValueError(\n            `ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ` +\n            `${inputs.length}.`);\n      }\n\n      const training = kwargs['training'] || false;\n\n      const x = inputs[0];         // Current input\n      const hTMinus1 = inputs[1];  // Previous memory state.\n      const cTMinus1 = inputs[2];  // Previous carry state.\n\n      const numOfKernels = 4;\n\n      type DropoutMasks = [tfc.Tensor, tfc.Tensor, tfc.Tensor, tfc.Tensor];\n\n      if (0 < this.dropout && this.dropout < 1 && this.dropoutMask == null) {\n        this.dropoutMask = generateDropoutMask({\n                             ones: () => tfc.onesLike(x),\n                             rate: this.dropout,\n                             training,\n                             count: numOfKernels\n                           }) as tfc.Tensor[];\n      }\n\n      const dropoutMask = this.dropoutMask as DropoutMasks;\n\n      const applyDropout =\n          (x: tfc.Tensor, mask: tfc.Tensor[], index: number) => {\n            if (!mask || !mask[index]) {\n              return x;\n            }\n\n            return tfc.mul(mask[index], x);\n          };\n\n      let xI = applyDropout(x, dropoutMask, 0);\n      let xF = applyDropout(x, dropoutMask, 1);\n      let xC = applyDropout(x, dropoutMask, 2);\n      let xO = applyDropout(x, dropoutMask, 3);\n\n      if (0 < this.recurrentDropout && this.recurrentDropout < 1 &&\n          this.recurrentDropoutMask == null) {\n        this.recurrentDropoutMask = generateDropoutMask({\n                                      ones: () => tfc.onesLike(hTMinus1),\n                                      rate: this.recurrentDropout,\n                                      training,\n                                      count: numOfKernels\n                                    }) as tfc.Tensor[];\n      }\n\n      const recDropoutMask = this.recurrentDropoutMask as DropoutMasks;\n\n      let hI = applyDropout(hTMinus1, recDropoutMask, 0);\n      let hF = applyDropout(hTMinus1, recDropoutMask, 1);\n      let hC = applyDropout(hTMinus1, recDropoutMask, 2);\n      let hO = applyDropout(hTMinus1, recDropoutMask, 3);\n\n      const kernelChannelAxis = 3;\n\n      const [kernelI, kernelF, kernelC, kernelO]: tfc.Tensor[] =\n          tfc.split(this.kernel.read(), numOfKernels, kernelChannelAxis);\n\n      const [biasI, biasF, biasC, biasO]: tfc.Tensor[] = this.useBias ?\n          tfc.split(this.bias.read(), numOfKernels) :\n          [null, null, null, null];\n\n      xI = this.inputConv(xI, kernelI, biasI, this.padding);\n      xF = this.inputConv(xF, kernelF, biasF, this.padding);\n      xC = this.inputConv(xC, kernelC, biasC, this.padding);\n      xO = this.inputConv(xO, kernelO, biasO, this.padding);\n\n      const [recKernelI, recKernelF, recKernelC, recKernelO]: tfc.Tensor[] =\n          tfc.split(\n              this.recurrentKernel.read(), numOfKernels, kernelChannelAxis);\n\n      hI = this.recurrentConv(hI, recKernelI);\n      hF = this.recurrentConv(hF, recKernelF);\n      hC = this.recurrentConv(hC, recKernelC);\n      hO = this.recurrentConv(hO, recKernelO);\n\n      const i = this.recurrentActivation.apply(tfc.add(xI, hI));\n      const f = this.recurrentActivation.apply(tfc.add(xF, hF));\n      const c = tfc.add(\n          tfc.mul(f, cTMinus1),\n          tfc.mul(i, this.activation.apply(tfc.add(xC, hC))));\n      const h = tfc.mul(\n          this.recurrentActivation.apply(tfc.add(xO, hO)),\n          this.activation.apply(c));\n\n      return [h, h, c];\n    });\n  }\n\n  getConfig(): tfc.serialization.ConfigDict {\n    const {'units': _, ...baseConfig} = super.getConfig();\n\n    const config: tfc.serialization.ConfigDict = {\n      filters: this.filters,\n      kernelSize: this.kernelSize,\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n      dilationRate: this.dilationRate,\n      strides: this.strides,\n    };\n\n    return {...baseConfig, ...config};\n  }\n\n  inputConv(x: Tensor, w: Tensor, b?: Tensor, padding?: PaddingMode) {\n    const out = tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, this.strides as [number, number],\n        (padding || 'valid') as 'same' | 'valid',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC',\n        this.dilationRate as [number, number]);\n\n    if (b) {\n      return K.biasAdd(out, b, this.dataFormat) as tfc.Tensor3D;\n    }\n\n    return out;\n  }\n\n  recurrentConv(x: Tensor, w: Tensor) {\n    const strides = 1;\n\n    return tfc.conv2d(\n        x as tfc.Tensor3D, w as tfc.Tensor4D, strides, 'same',\n        this.dataFormat === 'channelsFirst' ? 'NCHW' : 'NHWC');\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2DCell);\n\nexport declare interface ConvLSTM2DArgs extends\n    Omit<LSTMLayerArgs, 'units'|'cell'>, ConvRNN2DLayerArgs {}\n\nexport class ConvLSTM2D extends ConvRNN2D {\n  /** @nocollapse */\n  static className = 'ConvLSTM2D';\n\n  constructor(args: ConvLSTM2DArgs) {\n    const cell = new ConvLSTM2DCell(args);\n\n    super({...args, cell} as ConvRNN2DLayerArgs);\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends tfc.serialization.Serializable>(\n      cls: tfc.serialization.SerializableConstructor<T>,\n      config: tfc.serialization.ConfigDict): T {\n    return new cls(config);\n  }\n}\n\ntfc.serialization.registerClass(ConvLSTM2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Basic Layers.\n */\n\nimport {any, notEqual, serialization, Tensor, tidy, transpose, util} from '@tensorflow/tfjs-core';\n\nimport {Activation as ActivationFn, getActivation, serializeActivation} from '../activations';\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {DisposeResult, InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {ActivationIdentifier} from '../keras_format/activation_config';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {LayerConfig} from '../keras_format/topology_config';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport {assertPositiveInteger, mapActivationToFusedKernel} from '../utils/generic_utils';\nimport {arrayProd, range} from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface DropoutLayerArgs extends LayerArgs {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /**\n   * Integer array representing the shape of the binary dropout mask that will\n   * be multiplied with the input.\n   *\n   * For instance, if your inputs have shape `(batchSize, timesteps, features)`\n   * and you want the dropout mask to be the same for all timesteps, you can use\n   * `noise_shape=(batch_size, 1, features)`.\n   */\n  noiseShape?: number[];\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class Dropout extends Layer {\n  /** @nocollapse */\n  static className = 'Dropout';\n  private readonly rate: number;\n  private readonly noiseShape: number[];\n  private readonly seed: number;\n\n  constructor(args: DropoutLayerArgs) {\n    super(args);\n    this.rate = Math.max(Math.min(args.rate, 1), 0);\n    // So that the scalar doesn't get tidied up between executions.\n    this.noiseShape = args.noiseShape;\n    this.seed = args.seed;\n    this.supportsMasking = true;\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    if (this.noiseShape == null) {\n      return this.noiseShape;\n    }\n    const inputShape = input.shape;\n    const noiseShape: Shape = [];\n    for (let i = 0; i < this.noiseShape.length; ++i) {\n      noiseShape.push(\n          this.noiseShape[i] == null ? inputShape[i] : this.noiseShape[i]);\n    }\n    return noiseShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      if (0 < this.rate && this.rate < 1) {\n        const training =\n            kwargs['training'] == null ? false : kwargs['training'];\n        const noiseShape = this.getNoiseShape(input);\n        const output = K.inTrainPhase(\n            () => K.dropout(input, this.rate, noiseShape, this.seed),\n            () => input, training);\n        return output;\n      }\n      return inputs;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      rate: this.rate,\n      noiseShape: this.noiseShape,\n      seed: this.seed,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  dispose(): DisposeResult {\n    return super.dispose();\n  }\n}\nserialization.registerClass(Dropout);\n\nexport declare interface DenseLayerArgs extends LayerArgs {\n  /** Positive integer, dimensionality of the output space. */\n  units: number;\n  /**\n   * Activation function to use.\n   *\n   * If unspecified, no activation is applied.\n   */\n  activation?: ActivationIdentifier;\n  /** Whether to apply a bias. */\n  useBias?: boolean;\n  /**\n   * Initializer for the dense kernel weights matrix.\n   */\n  kernelInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Initializer for the bias vector.\n   */\n  biasInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * If specified, defines inputShape as `[inputDim]`.\n   */\n  inputDim?: number;\n\n  /**\n   * Constraint for the kernel weights.\n   */\n  kernelConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for the bias vector.\n   */\n  biasConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer function applied to the dense kernel weights matrix.\n   */\n  kernelRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the bias vector.\n   */\n  biasRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport interface SpatialDropout1DLayerConfig extends LayerConfig {\n  /** Float between 0 and 1. Fraction of the input units to drop. */\n  rate: number;\n\n  /** An integer to use as random seed. */\n  seed?: number;\n}\n\nexport class SpatialDropout1D extends Dropout {\n  /** @nocollapse */\n  static className = 'SpatialDropout1D';\n\n  constructor(args: SpatialDropout1DLayerConfig) {\n    super(args);\n    this.inputSpec = [{ndim: 3}];\n  }\n\n  protected getNoiseShape(input: Tensor): Shape {\n    const inputShape = input.shape;\n    return [inputShape[0], 1, inputShape[2]];\n  }\n}\nserialization.registerClass(SpatialDropout1D);\n\nexport class Dense extends Layer {\n  /** @nocollapse */\n  static className = 'Dense';\n  private units: number;\n  // Default activation: Linear (none).\n  private activation: ActivationFn = null;\n  private useBias = true;\n  private kernelInitializer: Initializer;\n  private biasInitializer: Initializer;\n  private kernel: LayerVariable = null;\n  private bias: LayerVariable = null;\n\n  readonly DEFAULT_KERNEL_INITIALIZER: InitializerIdentifier = 'glorotNormal';\n  readonly DEFAULT_BIAS_INITIALIZER: InitializerIdentifier = 'zeros';\n  private readonly kernelConstraint?: Constraint;\n  private readonly biasConstraint?: Constraint;\n  private readonly kernelRegularizer?: Regularizer;\n  private readonly biasRegularizer?: Regularizer;\n\n  constructor(args: DenseLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null &&\n        args.inputDim != null) {\n      // This logic is copied from Layer's constructor, since we can't\n      // do exactly what the Python constructor does for Dense().\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      this.batchInputShape = [batchSize, args.inputDim];\n    }\n\n    this.units = args.units;\n    assertPositiveInteger(this.units, 'units');\n    this.activation = getActivation(args.activation);\n    if (args.useBias != null) {\n      this.useBias = args.useBias;\n    }\n    this.kernelInitializer = getInitializer(\n        args.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER);\n    this.biasInitializer =\n        getInitializer(args.biasInitializer || this.DEFAULT_BIAS_INITIALIZER);\n    this.kernelConstraint = getConstraint(args.kernelConstraint);\n    this.biasConstraint = getConstraint(args.biasConstraint);\n    this.kernelRegularizer = getRegularizer(args.kernelRegularizer);\n    this.biasRegularizer = getRegularizer(args.biasRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.supportsMasking = true;\n\n    this.inputSpec = [{minNDim: 2}];\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const inputLastDim = inputShape[inputShape.length - 1];\n    if (this.kernel == null) {\n      this.kernel = this.addWeight(\n          'kernel', [inputLastDim, this.units], null, this.kernelInitializer,\n          this.kernelRegularizer, true, this.kernelConstraint);\n      if (this.useBias) {\n        this.bias = this.addWeight(\n            'bias', [this.units], null, this.biasInitializer,\n            this.biasRegularizer, true, this.biasConstraint);\n      }\n    }\n\n    this.inputSpec = [{minNDim: 2, axes: {[-1]: inputLastDim}}];\n    this.built = true;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    outputShape[outputShape.length - 1] = this.units;\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Dense layer accepts only a single input.\n      const input = getExactlyOneTensor(inputs);\n      const fusedActivationName =\n          mapActivationToFusedKernel(this.activation.getClassName());\n      let output: Tensor;\n\n      if (fusedActivationName != null) {\n        output = K.dot(\n            input, this.kernel.read(), fusedActivationName,\n            this.bias ? this.bias.read() : null);\n      } else {\n        output = K.dot(input, this.kernel.read());\n        if (this.bias != null) {\n          output = K.biasAdd(output, this.bias.read());\n        }\n        if (this.activation != null) {\n          output = this.activation.apply(output);\n        }\n      }\n\n      return output;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      units: this.units,\n      activation: serializeActivation(this.activation),\n      useBias: this.useBias,\n      kernelInitializer: serializeInitializer(this.kernelInitializer),\n      biasInitializer: serializeInitializer(this.biasInitializer),\n      kernelRegularizer: serializeRegularizer(this.kernelRegularizer),\n      biasRegularizer: serializeRegularizer(this.biasRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      kernelConstraint: serializeConstraint(this.kernelConstraint),\n      biasConstraint: serializeConstraint(this.biasConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dense);\n\nexport declare interface FlattenLayerArgs extends LayerArgs {\n  /** Image data format: channeLast (default) or channelFirst. */\n  dataFormat?: DataFormat;\n}\n\nexport class Flatten extends Layer {\n  private dataFormat: DataFormat;\n\n  /** @nocollapse */\n  static className = 'Flatten';\n  constructor(args?: FlattenLayerArgs) {\n    args = args || {};\n    super(args);\n    this.inputSpec = [{minNDim: 3}];\n    this.dataFormat = args.dataFormat;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    for (const dim of inputShape.slice(1)) {\n      if (dim == null) {\n        throw new ValueError(\n            `The shape of the input to \"Flatten\" is not fully defined ` +\n            `(got ${inputShape.slice(1)}). Make sure to pass a complete ` +\n            `\"input_shape\" or \"batch_input_shape\" argument to the first ` +\n            `layer in your model.`);\n      }\n    }\n    return [inputShape[0], arrayProd(inputShape, 1)];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n\n      let input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsFirst' && input.rank > 1) {\n        const permutation: number[] = [0];\n        for (let i = 2; i < input.rank; ++i) {\n          permutation.push(i);\n        }\n        permutation.push(1);\n        input = input.transpose(permutation);\n      }\n\n      return K.batchFlatten(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {};\n    if (this.dataFormat != null) {\n      config['dataFormat'] = this.dataFormat;\n    }\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Flatten);\n\nexport declare interface ActivationLayerArgs extends LayerArgs {\n  /**\n   * Name of the activation function to use.\n   */\n  activation: ActivationIdentifier;\n}\n\nexport class Activation extends Layer {\n  /** @nocollapse */\n  static className = 'Activation';\n  activation: ActivationFn;\n\n  constructor(args: ActivationLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n    this.activation = getActivation(args.activation);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      return this.activation.apply(input);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {activation: serializeActivation(this.activation)};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Activation);\n\nexport declare interface ReshapeLayerArgs extends LayerArgs {\n  /** The target shape. Does not include the batch axis. */\n  targetShape: Shape;\n}\n\nexport declare interface RepeatVectorLayerArgs extends LayerArgs {\n  /**\n   * The integer number of times to repeat the input.\n   */\n  n: number;\n}\n\nexport class RepeatVector extends Layer {\n  /** @nocollapse */\n  static className = 'RepeatVector';\n  readonly n: number;\n\n  constructor(args: RepeatVectorLayerArgs) {\n    super(args);\n    this.n = args.n;\n    this.inputSpec = [{ndim: 2}];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], this.n, inputShape[1]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = getExactlyOneTensor(inputs);\n      return K.repeat(inputs, this.n);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      n: this.n,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(RepeatVector);\n\nexport class Reshape extends Layer {\n  /** @nocollapse */\n  static className = 'Reshape';\n  private targetShape: Shape;\n\n  constructor(args: ReshapeLayerArgs) {\n    super(args);\n    this.targetShape = args.targetShape;\n\n    // Make sure that all unknown dimensions are represented as `null`.\n    for (let i = 0; i < this.targetShape.length; ++i) {\n      if (this.isUnknown(this.targetShape[i])) {\n        this.targetShape[i] = null;\n      }\n    }\n  }\n\n  private isUnknown(dim: number): boolean {\n    return dim < 0 || dim == null;\n  }\n\n  /**\n   * Finds and replaces a missing dimension in output shape.\n   *\n   * This is a near direct port of the internal Numpy function\n   * `_fix_unknown_dimension` in `numpy/core/src/multiarray/shape.c`.\n   *\n   * @param inputShape: Original shape of array begin reshape.\n   * @param outputShape: Target shape of the array, with at most a single\n   * `null` or negative number, which indicates an underdetermined dimension\n   * that should be derived from `inputShape` and the known dimensions of\n   *   `outputShape`.\n   * @returns: The output shape with `null` replaced with its computed value.\n   * @throws: ValueError: If `inputShape` and `outputShape` do not match.\n   */\n  private fixUnknownDimension(inputShape: Shape, outputShape: Shape): Shape {\n    const errorMsg = 'Total size of new array must be unchanged.';\n    const finalShape = outputShape.slice();\n    let known = 1;\n    let unknown = null;\n    for (let i = 0; i < finalShape.length; ++i) {\n      const dim = finalShape[i];\n      if (this.isUnknown(dim)) {\n        if (unknown === null) {\n          unknown = i;\n        } else {\n          throw new ValueError('Can only specifiy one unknown dimension.');\n        }\n      } else {\n        known *= dim;\n      }\n    }\n\n    const originalSize = arrayProd(inputShape);\n    if (unknown !== null) {\n      if (known === 0 || originalSize % known !== 0) {\n        throw new ValueError(errorMsg);\n      }\n      finalShape[unknown] = originalSize / known;\n    } else if (originalSize !== known) {\n      throw new ValueError(errorMsg);\n    }\n\n    return finalShape;\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    let anyUnknownDims = false;\n    for (let i = 0; i < inputShape.length; ++i) {\n      if (this.isUnknown(inputShape[i])) {\n        anyUnknownDims = true;\n        break;\n      }\n    }\n\n    if (anyUnknownDims) {\n      return inputShape.slice(0, 1).concat(this.targetShape);\n    } else {\n      return inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const outputShape = inputShape.slice(0, 1).concat(\n          this.fixUnknownDimension(inputShape.slice(1), this.targetShape));\n      return input.reshape(outputShape);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      targetShape: this.targetShape,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Reshape);\n\nexport declare interface PermuteLayerArgs extends LayerArgs {\n  /**\n   * Array of integers. Permutation pattern. Does not include the\n   * sample (batch) dimension. Index starts at 1.\n   * For instance, `[2, 1]` permutes the first and second dimensions\n   * of the input.\n   */\n  dims: number[];\n}\n\nexport class Permute extends Layer {\n  /** @nocollapse */\n  static className = 'Permute';\n  readonly dims: number[];\n  private readonly dimsIncludingBatch: number[];\n\n  constructor(args: PermuteLayerArgs) {\n    super(args);\n    if (args.dims == null) {\n      throw new Error(\n          'Required configuration field `dims` is missing during Permute ' +\n          'constructor call.');\n    }\n    if (!Array.isArray(args.dims)) {\n      throw new Error(\n          'Permute constructor requires `dims` to be an Array, but received ' +\n          `${args.dims} instead.`);\n    }\n\n    // Check the validity of the permutation indices.\n    const expectedSortedIndices = range(1, args.dims.length + 1);\n    if (!util.arraysEqual(args.dims.slice().sort(), expectedSortedIndices)) {\n      throw new Error(\n          'Invalid permutation `dims`: ' + JSON.stringify(args.dims) +\n          ' `dims` must contain consecutive integers starting from 1.');\n    }\n\n    this.dims = args.dims;\n    this.dimsIncludingBatch = [0].concat(this.dims);\n    this.inputSpec = [new InputSpec({ndim: this.dims.length + 1})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const outputShape = inputShape.slice();\n    this.dims.forEach((dim: number, i: number) => {\n      outputShape[i + 1] = (inputShape as Shape)[dim];\n    });\n    return outputShape;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return transpose(getExactlyOneTensor(inputs), this.dimsIncludingBatch);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      dims: this.dims,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Permute);\n\nexport declare interface MaskingArgs extends LayerArgs {\n  /**\n   * Masking Value. Defaults to `0.0`.\n   */\n  maskValue?: number;\n}\n\nexport class Masking extends Layer {\n  /** @nocollapse */\n  static className = 'Masking';\n  maskValue: number;\n\n  constructor(args?: MaskingArgs) {\n    super(args == null ? {} : args);\n    this.supportsMasking = true;\n    if (args != null) {\n      this.maskValue = args.maskValue == null ? 0 : args.maskValue;\n    } else {\n      this.maskValue = 0;\n    }\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    return inputShape;\n  }\n\n  getConfig() {\n    const baseConfig = super.getConfig();\n    const config = {maskValue: this.maskValue};\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    const input = getExactlyOneTensor(inputs);\n    const axis = -1;\n    return any(notEqual(input, this.maskValue), axis);\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      const input = getExactlyOneTensor(inputs);\n      const axis = -1;\n      const keepDims = true;\n      const booleanMask = any(notEqual(input, this.maskValue), axis, keepDims);\n      const output = input.mul(booleanMask.asType(input.dtype));\n      return output;\n    });\n  }\n}\nserialization.registerClass(Masking);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Embedding Layer.\n *\n * Original source: keras/constraints.py\n */\nimport {notEqual, serialization, Tensor, tidy, zerosLike} from '@tensorflow/tfjs-core';\n\nimport * as K from '../backend/tfjs_backend';\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nexport declare interface EmbeddingLayerArgs extends LayerArgs {\n  /**\n   * Integer > 0. Size of the vocabulary, i.e. maximum integer index + 1.\n   */\n  inputDim: number;\n  /**\n   * Integer >= 0. Dimension of the dense embedding.\n   */\n  outputDim: number;\n  /**\n   * Initializer for the `embeddings` matrix.\n   */\n  embeddingsInitializer?: InitializerIdentifier|Initializer;\n  /**\n   * Regularizer function applied to the `embeddings` matrix.\n   */\n  embeddingsRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Regularizer function applied to the activation.\n   */\n  activityRegularizer?: RegularizerIdentifier|Regularizer;\n  /**\n   * Constraint function applied to the `embeddings` matrix.\n   */\n  embeddingsConstraint?: ConstraintIdentifier|Constraint;\n  /**\n   * Whether the input value 0 is a special \"padding\" value that should be\n   * masked out. This is useful when using recurrent layers which may take\n   * variable length input.\n   *\n   * If this is `True` then all subsequent layers in the model need to support\n   * masking or an exception will be raised. If maskZero is set to `True`, as a\n   * consequence, index 0 cannot be used in the vocabulary (inputDim should\n   * equal size of vocabulary + 1).\n   */\n  maskZero?: boolean;\n  /**\n   * Length of input sequences, when it is constant.\n   *\n   * This argument is required if you are going to connect `flatten` then\n   * `dense` layers upstream (without it, the shape of the dense outputs cannot\n   * be computed).\n   */\n  inputLength?: number|number[];\n}\n\nexport class Embedding extends Layer {\n  /** @nocollapse */\n  static className = 'Embedding';\n  private inputDim: number;\n  private outputDim: number;\n  private embeddingsInitializer: Initializer;\n  private maskZero: boolean;\n  private inputLength: number|number[];\n\n  private embeddings: LayerVariable = null;\n\n  readonly DEFAULT_EMBEDDINGS_INITIALIZER: InitializerIdentifier =\n      'randomUniform';\n  private readonly embeddingsRegularizer?: Regularizer;\n  private readonly embeddingsConstraint?: Constraint;\n\n  constructor(args: EmbeddingLayerArgs) {\n    super(args);\n    if (args.batchInputShape == null && args.inputShape == null) {\n      // Porting Note: This logic is copied from Layer's constructor, since we\n      // can't do exactly what the Python constructor does for Embedding().\n      // Specifically, the super constructor can not be called after the\n      // mutation of the `config` argument.\n      let batchSize: number = null;\n      if (args.batchSize != null) {\n        batchSize = args.batchSize;\n      }\n      if (args.inputLength == null) {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (None, )\n        this.batchInputShape = [batchSize, null];\n      } else {\n        // Fix super-constructor to what it would have done if\n        // 'config.inputShape' were (config.inputLength, )\n        this.batchInputShape =\n            [batchSize].concat(generic_utils.toList(args.inputLength));\n      }\n    }\n    this.inputDim = args.inputDim;\n    generic_utils.assertPositiveInteger(this.inputDim, 'inputDim');\n    this.outputDim = args.outputDim;\n    generic_utils.assertPositiveInteger(this.outputDim, 'outputDim');\n    this.embeddingsInitializer = getInitializer(\n        args.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER);\n    this.embeddingsRegularizer = getRegularizer(args.embeddingsRegularizer);\n    this.activityRegularizer = getRegularizer(args.activityRegularizer);\n    this.embeddingsConstraint = getConstraint(args.embeddingsConstraint);\n    this.maskZero = args.maskZero;\n    this.supportsMasking = args.maskZero;\n    this.inputLength = args.inputLength;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    this.embeddings = this.addWeight(\n        'embeddings', [this.inputDim, this.outputDim], this.dtype,\n        this.embeddingsInitializer, this.embeddingsRegularizer, true,\n        this.embeddingsConstraint);\n    this.built = true;\n  }\n\n  // Override warnOnIncompatibleInputShape because an embedding layer allows\n  // the input to have varying ranks.\n  protected warnOnIncompatibleInputShape(inputShape: Shape) {}\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tidy(() => {\n      if (!this.maskZero) {\n        return null;\n      } else {\n        inputs = getExactlyOneTensor(inputs);\n        return notEqual(inputs, zerosLike(inputs));\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    if (this.inputLength == null) {\n      return [...inputShape, this.outputDim];\n    }\n    // inputLength can be an array if input is 3D or higher.\n    const inLens: number[] = generic_utils.toList(this.inputLength);\n    if (inLens.length !== inputShape.length - 1) {\n      throw new ValueError(\n          `\"inputLength\" is ${this.inputLength}, but received ` +\n          `input shape has shape ${inputShape}`);\n    } else {\n      let i = 0;\n      for (let k = 0; k < inLens.length; ++k) {\n        const s1 = inLens[k];\n        const s2 = inputShape[k + 1];\n        if ((s1 != null) && (s2 != null) && (s1 !== s2)) {\n          throw new ValueError(\n              `\"inputLength\" is ${this.inputLength}, but received ` +\n              `input shape has shape ${inputShape}`);\n        } else if (s1 == null) {\n          inLens[i] = s2;\n        }\n        i++;\n      }\n    }\n    return [inputShape[0], ...inLens, this.outputDim];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Embedding layer accepts only a single input.\n      let input = getExactlyOneTensor(inputs);\n      if (input.dtype !== 'int32') {\n        input = K.cast(input, 'int32');\n      }\n      const output = K.gather(this.embeddings.read(), input.as1D());\n      return output.reshape(\n          getExactlyOneShape(this.computeOutputShape(input.shape)));\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      inputDim: this.inputDim,\n      outputDim: this.outputDim,\n      embeddingsInitializer: serializeInitializer(this.embeddingsInitializer),\n      embeddingsRegularizer: serializeRegularizer(this.embeddingsRegularizer),\n      activityRegularizer: serializeRegularizer(this.activityRegularizer),\n      embeddingsConstraint: serializeConstraint(this.embeddingsConstraint),\n      maskZero: this.maskZero,\n      inputLength: this.inputLength\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Embedding);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Merge Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy, util} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {Shape} from '../keras_format/common';\nimport {l2Normalize} from '../losses';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as mathUtils from '../utils/math_utils';\nimport {getExactlyOneShape} from '../utils/types_utils';\n\n/**\n * Generic Merge layer for element-wise merge functions.\n *\n * Used to implement `Sum`, `Average`, `Concatenate`, etc.\n */\nexport abstract class Merge extends Layer {\n  protected reshapeRequired: boolean;\n\n  constructor(args?: LayerArgs) {\n    super(args || {});\n    this.supportsMasking = true;\n  }\n\n  /**\n   * Logic for merging multiple tensors, to be overridden by subclasses.\n   * @param inputs\n   */\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    throw new NotImplementedError();\n  }\n\n  /**\n   * Computes the shape of the result of an elementwise operation.\n   *\n   * @param shape1: Shape of the first tensor.\n   * @param shape2: Shape of the second tensor.\n   * @returns Expected output shape when an elementwise operation is carried\n   *   out on 2 tensors with shapes `shape1` and `shape2`.\n   * @throws ValueError: If `shape1` and `shape2` are not compatible for\n   *   element-wise operations.\n   */\n  private computeElementwiseOpOutputShape(shape1: Shape, shape2: Shape): Shape {\n    if (shape1 == null || shape2 == null) {\n      return null;\n    } else if (shape1.length < shape2.length) {\n      return this.computeElementwiseOpOutputShape(shape2, shape1);\n    } else if (shape2.length === 0) {\n      return shape1;\n    }\n    const outputShape: Shape = shape1.slice(0, shape1.length - shape2.length);\n    for (let k = 0; k < shape2.length; ++k) {\n      const i = shape1[shape1.length - shape2.length + k];\n      const j = shape2[k];\n      if (i == null || j == null || i < 0 || j < 0) {\n        outputShape.push(null);\n      } else if (i === 1) {\n        outputShape.push(j);\n      } else if (j === 1) {\n        outputShape.push(i);\n      } else {\n        if (i !== j) {\n          throw new ValueError(\n              'Operands could not be broadcast together with shapes ' +\n              JSON.stringify(shape1) + ' ' + JSON.stringify(shape2));\n        }\n        outputShape.push(i);\n      }\n    }\n    return outputShape;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.\n    if (Array.isArray(inputShape) && !Array.isArray(inputShape[0])) {\n      // Make sure that inputShape is an Array of shape.\n      inputShape = [getExactlyOneShape(inputShape)];\n    }\n    inputShape = inputShape as Shape[];\n    if (inputShape.length < 2) {\n      throw new ValueError(\n          'A merge layer should be called on an Array of at least 2 inputs.' +\n          ` Got ${inputShape.length} input(s).`);\n    }\n\n    // Make sure that there is at most one unique batch size among the input\n    // shapes.\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length > 1) {\n      throw new ValueError(\n          `Can not merge tensors with different batch sizes. ` +\n          `Got tensors with shapes: ${JSON.stringify(inputShape)}.`);\n    }\n\n    let outputShape: Shape =\n        inputShape[0] == null ? null : inputShape[0].slice(1);\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n    // If the inputs have different ranks, we have to reshape them to make them\n    // broadcastable.\n    const allRanks = inputShape.map(shape => shape.length);\n    if (inputShape.indexOf(null) === -1 &&\n        generic_utils.unique(allRanks).length === 1) {\n      this.reshapeRequired = false;\n    } else {\n      this.reshapeRequired = true;\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      inputs = inputs as Tensor[];\n      if (this.reshapeRequired) {\n        const reshapedInputs: Tensor[] = [];\n        const inputDims = inputs.map(input => input.rank);\n        if (inputDims.indexOf(null) === -1) {\n          // If ranks of all inputs are available, we simply expand each of them\n          // at axis=1 until all of them have the same rank.\n          const maxNDim = mathUtils.max(inputDims);\n          for (let x of inputs) {\n            const xNDim = x.rank;\n            for (let k = 0; k < maxNDim - xNDim; ++k) {\n              x = K.expandDims(x, 1);\n            }\n            reshapedInputs.push(x);\n          }\n          return this.mergeFunction(reshapedInputs);\n        } else {\n          // Transpose all inputs so that batch size is the last dimension.\n          // [batchSize, dim1, dim2, ...] -> [dim1, dim2, ..., batchSize]\n          let transposed = false;\n          for (const x of inputs) {\n            const xNDim = x.rank;\n            if (xNDim == null) {\n              const xShape = x.shape;\n              const batchSize = xShape[0];\n              const newShape = xShape.slice(1).concat([batchSize]);\n              let xTransposed = x.reshape(\n                  [batchSize].concat(mathUtils.arrayProd(xShape.slice(1))));\n              xTransposed = tfc.transpose(xTransposed, [1, 0]);\n              xTransposed = xTransposed.reshape(newShape);\n              reshapedInputs.push(xTransposed);\n              transposed = true;\n            } else if (xNDim > 1) {\n              const dims = mathUtils.range(1, xNDim).concat([0]);\n              reshapedInputs.push(tfc.transpose(x, dims));\n              transposed = true;\n            } else {\n              // We don't transpose inputs if they are 1D vectors or scalars.\n              reshapedInputs.push(x);\n            }\n          }\n          let y = this.mergeFunction(reshapedInputs);\n          const yNDim = y.rank;\n          if (transposed) {\n            // If inputs have been transposed, we have to transpose the output\n            // too.\n            if (yNDim == null) {\n              const yShape = y.shape;\n              const yNDim = yShape.length;\n              const batchSize = yShape[yNDim - 1];\n              const newShape =\n                  [batchSize].concat(yShape.slice(0, yShape.length - 1));\n              y = tfc.transpose(y.reshape([-1, batchSize]), [1, 0])\n                      .reshape(newShape);\n            } else if (yNDim > 1) {\n              const dims = [yNDim - 1].concat(mathUtils.range(0, yNDim - 1));\n              y = tfc.transpose(y, dims);\n            }\n          }\n          return y;\n        }\n      } else {\n        return this.mergeFunction(inputs);\n      }\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape[];\n    let outputShape: Shape;\n    if (inputShape[0] == null) {\n      outputShape = null;\n    } else {\n      outputShape = inputShape[0].slice(1);\n    }\n    for (let i = 1; i < inputShape.length; ++i) {\n      const shape = inputShape[i] == null ? null : inputShape[i].slice(1);\n      outputShape = this.computeElementwiseOpOutputShape(outputShape, shape);\n    }\n\n    let batchSizes: number[] = [];\n    for (const shape of inputShape) {\n      if (shape != null && shape[0] !== null) {\n        batchSizes.push(shape[0]);\n      }\n    }\n    batchSizes = generic_utils.unique(batchSizes);\n    if (batchSizes.length === 1) {\n      outputShape = batchSizes.concat(outputShape);\n    } else {\n      outputShape = [null].concat(outputShape);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return tfc.tidy(() => {\n      if (mask == null) {\n        return null;\n      }\n      if (!Array.isArray(mask)) {\n        throw new ValueError('`mask` should be an Array');\n      }\n      if (!Array.isArray(inputs)) {\n        throw new ValueError('`inputs` should be an Array');\n      }\n      if (mask.length !== inputs.length) {\n        throw new ValueError(\n            `The Array 'inputs' and 'mask' are expected to have the same ` +\n            `length, but have different lengths ` +\n            `(${inputs.length} vs ${mask.length})`);\n      }\n      if (mask.every(m => m == null)) {\n        return null;\n      }\n      mask = mask.map(m => m == null ? m : tfc.expandDims(m, 0));\n      let output = mask[0];\n      for (let i = 1; i < mask.length - 1; ++i) {\n        output = tfc.logicalAnd(output, mask[i]);\n      }\n      return output;\n    });\n  }\n}\n\nexport class Add extends Merge {\n  /** @nocollapse */\n  static className = 'Add';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Add);\n\n/**\n * Calculate the element-wise sum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Add` layer, by using no input argument\n *    or a single configuration argument. The resultant `Add` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const addLayer = tf.layers.add();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = addLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.add([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.add([input1, input2]).print();\n * // Gives [[11, 22], [33, 44]].\n *\n */\nexport function add(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Add({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Add(config);\n  }\n}\n\nexport class Multiply extends Merge {\n  /** @nocollapse */\n  static className = 'Multiply';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.mul(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Multiply);\n\n/**\n * Calculate the element-wise product of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Multiply` layer, by using no input argument\n *    or a single configuration argument. The resultant `Multiply` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const multiplyLayer = tf.layers.multiply();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = multiplyLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.multiply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.multiply([input1, input2]).print();\n * // Gives [[10, 40], [90, 160]].\n *\n */\nexport function multiply(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Multiply({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Multiply(config);\n  }\n}\n\nexport class Average extends Merge {\n  /** @nocollapse */\n  static className = 'Average';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0].clone();\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.add(output, inputs[i]);\n      }\n      return tfc.mul(1 / inputs.length, output);\n    });\n  }\n}\nserialization.registerClass(Average);\n\n/**\n * Calculate the element-wise arithmetic mean of inputs, which all have the same\n * shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Average` layer, by using no input argument\n *    or a single configuration argument. The resultant `Average` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const averageLayer = tf.layers.average();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = averageLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.average([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const input2 = tf.tensor2d([10, 20, 30, 40], [2, 2]);\n * tf.layers.average([input1, input2]).print();\n * // Gives [[5.5, 11], [16.5, 22]].\n *\n */\nexport function average(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Average({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Average(config);\n  }\n}\n\nexport class Maximum extends Merge {\n  /** @nocollapse */\n  static className = 'Maximum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.maximum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Maximum);\n\n/**\n * Calculate the element-wise maximum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Maximum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Maximum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const maximumLayer = tf.layers.maximum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = maximumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.maximum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.maximum([input1, input2]).print();\n * // Gives [[10, 20], [30, 40]].\n *\n */\nexport function maximum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Maximum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Maximum(config);\n  }\n}\n\nexport class Minimum extends Merge {\n  /** @nocollapse */\n  static className = 'Minimum';\n  constructor(args?: LayerArgs) {\n    super(args);\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      let output = inputs[0];\n      for (let i = 1; i < inputs.length; ++i) {\n        output = tfc.minimum(output, inputs[i]);\n      }\n      return output;\n    });\n  }\n}\nserialization.registerClass(Minimum);\n\n/**\n * Calculate the element-wise minimum of inputs, which all have the same shape.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Minimum` layer, by using no input argument\n *    or a single configuration argument. The resultant `Minimum` layer can then\n *    be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const minimumLayer = tf.layers.minimum();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = minimumLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const output = tf.layers.minimum([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([1, 20, 3, 40], [2, 2]);\n * const input2 = tf.tensor2d([10, 2, 30, 4], [2, 2]);\n * tf.layers.minimum([input1, input2]).print();\n * // Gives [[1, 2], [3, 4]].\n *\n */\nexport function minimum(config?: SymbolicTensor[]|Tensor[]|LayerArgs): Layer|\n    SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Minimum({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Minimum(config);\n  }\n}\n\nexport declare interface ConcatenateLayerArgs extends LayerArgs {\n  /**\n   * Axis along which to concatenate.\n   */\n  axis?: number;\n}\n\nexport class Concatenate extends Merge {\n  /** @nocollapse */\n  static className = 'Concatenate';\n  readonly DEFAULT_AXIS = -1;\n  private readonly axis: number;\n\n  constructor(args?: ConcatenateLayerArgs) {\n    super(args);\n    if (args == null) {\n      args = {};\n    }\n    this.axis = args.axis == null ? this.DEFAULT_AXIS : args.axis;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    // Used purely for shape validation.]\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0])) ||\n        inputShape.length === 1) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of at least 2 ' +\n          'inputs');\n    }\n    inputShape = inputShape as Shape[];\n\n    let allNoneShape = true;\n    for (const shape of inputShape) {\n      if (shape != null) {\n        allNoneShape = false;\n        break;\n      }\n    }\n    if (allNoneShape) {\n      return;\n    }\n\n    const shapeSet: Shape[] = [];\n    for (let i = 0; i < inputShape.length; ++i) {\n      const shapeWithoutConcatAxis = inputShape[i].slice();\n      shapeWithoutConcatAxis.splice(this.axis, 1);\n      let exists = false;\n      for (const shape of shapeSet) {\n        if (util.arraysEqual(shape, shapeWithoutConcatAxis)) {\n          exists = true;\n          break;\n        }\n      }\n      if (!exists) {\n        shapeSet.push(shapeWithoutConcatAxis);\n      }\n    }\n    if (shapeSet.length > 1) {\n      throw new ValueError(\n          'A `Concatenate` layer requires inputs with matching shapes ' +\n          'except for the concat axis. Got input shapes: ' +\n          JSON.stringify(inputShape));\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    return tidy(() => {\n      return K.concatenate(inputs, this.axis);\n    });\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    if (!(Array.isArray(inputShape) && Array.isArray(inputShape[0]))) {\n      throw new ValueError(\n          'A `Concatenate` layer should be called on a list of inputs.');\n    }\n    const inputShapes = inputShape as Shape[];\n    const outputShape = inputShapes[0].slice();\n    const axis = this.axis < 0 ? outputShape.length + this.axis : this.axis;\n    // Porting Note: the line above is because TypeScript doesn't support\n    //   negative indices.\n    for (const shape of inputShapes.slice(1)) {\n      if (outputShape[axis] == null || shape[axis] == null) {\n        outputShape[axis] = null;\n        break;\n      }\n      outputShape[axis] += shape[axis];\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    if (mask == null) {\n      return null;\n    }\n    if (!Array.isArray(mask)) {\n      throw new ValueError('`mask` should be an array for Concatenate');\n    }\n    if (!Array.isArray(inputs)) {\n      throw new ValueError('`inputs` should be an array for Concatenate');\n    }\n    if (mask.length !== inputs.length) {\n      throw new ValueError(\n          `Mismatch in the length of mask (${mask.length}) ` +\n          `and the legnth of inputs (${inputs.length})`);\n    }\n    return tfc.tidy(() => {\n      let allNullMasks = true;\n      mask.forEach(m => {\n        if (m != null) {\n          allNullMasks = false;\n          return;\n        }\n      });\n      if (allNullMasks) {\n        return null;\n      }\n      const outputMasks: Tensor[] = [];\n      for (let i = 0; i < inputs.length; ++i) {\n        if (mask[i] == null) {\n          // Input is unmasked. Append all 1's to masks.\n          outputMasks.push(tfc.onesLike(inputs[i]).asType('bool'));\n        } else if (mask[i].rank < inputs[i].rank) {\n          // Mask is smaller than the input, expand it.\n          outputMasks.push(tfc.expandDims(mask[i], -1));\n        } else {\n          outputMasks.push(mask[i]);\n        }\n      }\n      const concatenatedMasks = tfc.concat(outputMasks, this.axis);\n      return tfc.all(concatenatedMasks, -1, false);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axis': this.axis,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Concatenate);\n\n/**\n * Concatenate an `Array` of inputs.\n *\n * This function can be invoked in three ways.\n *\n * 1. Construct an instance of `Concatenate` layer, by using no input argument\n *    or a single configuration argument. The resultant `Concatenate` layer can\n *    then be used on `tf.SymbolicTensor`s or `tf.Tensor`s. For example:\n *\n * ```js\n * const concatLayer = tf.layers.concatenate();\n *\n * // The layer can be applied to inputs.\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = concatLayer.apply([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 7], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 2. Invoke directly on an `Array` of `tf.SymbolicTensor`s. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.SymbolicTensor`. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 3]});\n * const input2 = tf.input({shape: [2, 4]});\n * const output = tf.layers.concatenate([input1, input2]);\n * console.log(output.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension and the last dimension as the result of concatenating the\n * // last dimensions of the two inputs.\n * ```\n *\n * 3. Invoke directly on `tf.Tensor`s, i.e., concrete values. This constructs\n *    an `Layer` object internally and calls its `apply` method on the inputs,\n *    generating a new `tf.Tensor` as the result of the computation. For\n * example:\n *\n * ```js\n * const input1 = tf.tensor2d([[1, 2], [3, 4]], [2, 2]);\n * const input2 = tf.tensor2d([[10, 20], [30, 40]], [2, 2]);\n * tf.layers.concatenate([input1, input2]).print();\n * // Gives [[1, 2, 10, 20], [3, 4, 30, 40]].\n *\n */\nexport function concatenate(config?: SymbolicTensor[]|Tensor[]|\n                            ConcatenateLayerArgs): Layer|SymbolicTensor|Tensor {\n  if (Array.isArray(config)) {\n    const layer = new Concatenate({});\n    return layer.apply(config) as SymbolicTensor | Tensor;\n  } else {\n    return new Concatenate(config);\n  }\n}\n\nexport declare interface DotLayerArgs extends LayerArgs {\n  /**\n   * Axis or axes along which the dot product will be taken.\n   *\n   * Integer or an Array of integers.\n   */\n  axes: number|[number, number];\n\n  /**\n   * Whether to L2-normalize samples along the dot product axis\n   * before taking the dot product.\n   *\n   * If set to `true`, the output of the dot product isthe cosine\n   * proximity between the two samples.\n   */\n  normalize?: boolean;\n}\n\n/**\n * Interpretable potentially negative axis index.\n *\n * For example, given axis = -1, and dim = 3, this function will return 2.\n *\n * @param axis The axis index, may be a positive, zero or negative integer.\n * @param dim Total number of dimensions, a positive integer.\n * @returns A non-negative axis index equivalent to the input `axis`.\n */\nfunction interpretAxis(axis: number, dim: number): number {\n  while (axis < 0) {\n    axis += dim;\n  }\n  return axis;\n}\n\nfunction batchDot(x: Tensor, y: Tensor, axes: number|[number, number]): Tensor {\n  if (x.shape.length > 3 || y.shape.length > 3) {\n    throw new NotImplementedError(\n        'batchDot is not implemented for tensors of 4D or higher rank yet');\n  }\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of x to be >= 2, ` +\n          `but got ${x.shape.length}`);\n  tfc.util.assert(\n      x.shape.length >= 2,\n      () => `batchDot requires the rank of y to be >= 2, ` +\n          `but got ${y.shape.length}`);\n\n  if (typeof axes === 'number') {\n    axes = [axes, axes];\n  }\n\n  if (x.dtype === 'complex64' || y.dtype === 'complex64') {\n    throw new NotImplementedError(\n        'batchDot is not implemented for complex64-type Tensors yet.');\n  }\n\n  const xNDim = x.shape.length;\n  const yNDim = y.shape.length;\n  if (axes == null) {\n    // Behave like batchMatmul by default.\n    axes = [xNDim - 1, yNDim - 2];\n  }\n  const axesArray = axes as [number, number];\n\n  return tfc.tidy(() => {\n    let diff: number;\n    if (xNDim > yNDim) {\n      diff = xNDim - yNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      y = y.reshape(y.shape.concat(diffShape));\n    } else if (yNDim > xNDim) {\n      diff = yNDim - xNDim;\n      const diffShape: Shape = [];\n      for (let i = 0; i < diff; ++i) {\n        diffShape.push(1);\n      }\n      x = x.reshape(x.shape.concat(diffShape));\n    } else {\n      diff = 0;\n    }\n\n    let out: Tensor;\n    if (x.shape.length === 2 && y.shape.length === 2) {\n      if (axesArray[0] === axesArray[1]) {\n        out = x.mul(y).sum(axesArray[0]);\n      } else {\n        out = x.transpose([1, 0]).mul(y).sum(axesArray[1]);\n      }\n    } else {\n      const adjX = axesArray[0] !== x.shape.length - 1;\n      const adjY = axesArray[1] === y.shape.length - 1;\n      out = x.matMul(y, adjX, adjY);\n    }\n\n    if (diff > 0) {\n      let idx: number;\n      if (xNDim > yNDim) {\n        idx = xNDim + yNDim - 3;\n      } else {\n        idx = xNDim - 1;\n      }\n      const squeezeAxes: number[] = [];\n      for (let i = idx; i < idx + diff; ++i) {\n        squeezeAxes.push(i);\n      }\n      out = out.squeeze(squeezeAxes);\n    }\n    if (out.shape.length === 1) {\n      out = out.expandDims(1);\n    }\n    return out;\n  });\n}\n\nexport class Dot extends Merge {\n  /** @nocollapse */\n  static className = 'Dot';\n\n  private axes: number|[number, number];\n  private normalize: boolean;\n\n  constructor(args: DotLayerArgs) {\n    super(args);\n    this.axes = args.axes;\n    this.normalize = args.normalize == null ? false : args.normalize;\n    this.supportsMasking = true;\n    this.reshapeRequired = false;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = inputShape[0] as Shape;\n    const shape2 = inputShape[1] as Shape;\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    if (shape1[axes[0]] !== shape2[axes[1]]) {\n      throw new ValueError(\n          `Dimension incompatibility: ` +\n          `${shape1[axes[0]]} !== ${shape2[axes[1]]}`);\n    }\n  }\n\n  protected mergeFunction(inputs: Tensor[]): Tensor {\n    if (inputs.length !== 2) {\n      throw new ValueError(\n          'A `Dot` layer must be called on exactly 2 inputs, ' +\n          `but received ${inputs.length} input(s).`);\n    }\n\n    let x1 = inputs[0];\n    let x2 = inputs[1];\n    let axes: [number, number];\n    if (!Array.isArray(this.axes)) {\n      axes = [\n        interpretAxis(this.axes, x1.shape.length),\n        interpretAxis(this.axes, x2.shape.length)\n      ];\n    } else {\n      axes = this.axes.map(\n                 (axis, i) => interpretAxis(\n                     axis, inputs[i].shape.length)) as [number, number];\n    }\n    if (this.normalize) {\n      x1 = l2Normalize(x1, axes[0]);\n      x2 = l2Normalize(x2, axes[1]);\n    }\n    return batchDot(x1, x2, axes);\n  }\n\n  private interpretAxes(shape1: Shape, shape2: Shape): number[] {\n    let axes: number[];\n    if (!Array.isArray(this.axes)) {\n      // `this.axes` is a single integer.\n      axes = [\n        interpretAxis(this.axes, shape1.length),\n        interpretAxis(this.axes, shape2.length)\n      ];\n    } else {\n      // `this.axes` is an Array of integers.\n      axes = this.axes;\n    }\n    return axes;\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    tfc.util.assert(\n        Array.isArray(inputShape) && inputShape.length === 2 &&\n            Array.isArray(inputShape[0]) && Array.isArray(inputShape[1]),\n        () => 'A `Dot` layer should be called on a list of exactly 2 inputs.');\n    const shape1 = (inputShape[0] as Shape).slice();\n    const shape2 = (inputShape[1] as Shape).slice();\n    if (shape1.length > 3 || shape2.length > 3) {\n      throw new NotImplementedError(\n          'Dot layer does not support tensors of 4D or higher rank yet.');\n    }\n\n    const axes = this.interpretAxes(shape1, shape2);\n    shape1.splice(axes[0], 1);\n    shape2.splice(axes[1], 1);\n    shape2.splice(0, 1);\n    const outputShape = shape1.concat(shape2);\n    if (outputShape.length === 1) {\n      outputShape.push(1);\n    }\n    return outputShape;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor {\n    return null;\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'axes': this.axes,\n      'normalize': this.normalize\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(Dot);\n\n// TODO(cais): Add functional interfaces for the merge layers.\n", "/**\r\n * @license\r\n * Copyright 2018 Google LLC\r\n *\r\n * Use of this source code is governed by an MIT-style\r\n * license that can be found in the LICENSE file or at\r\n * https://opensource.org/licenses/MIT.\r\n * =============================================================================\r\n */\r\n\r\n/**\r\n * TensorFlow.js Layers: Noise Layers.\r\n */\r\n\r\nimport {greaterEqual, randomUniform, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\r\n\r\nimport * as K from '../backend/tfjs_backend';\r\nimport {Layer, LayerArgs} from '../engine/topology';\r\nimport {Shape} from '../keras_format/common';\r\nimport {Kwargs} from '../types';\r\nimport {getExactlyOneTensor} from '../utils/types_utils';\r\n\r\nexport declare interface GaussianNoiseArgs extends LayerArgs {\r\n  /** Standard Deviation.  */\r\n  stddev: number;\r\n}\r\n\r\nexport class GaussianNoise extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianNoise';\r\n  readonly stddev: number;\r\n\r\n  constructor(args: GaussianNoiseArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.stddev = args.stddev;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {stddev: this.stddev};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      const noised = () =>\r\n          K.randomNormal(input.shape, 0, this.stddev).add(input);\r\n      const output =\r\n          K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      return output;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianNoise);\r\n\r\nexport declare interface GaussianDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n}\r\n\r\nexport class GaussianDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'GaussianDropout';\r\n  readonly rate: number;\r\n\r\n  constructor(args: GaussianDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      this.invokeCallHook(inputs, kwargs);\r\n      const input = getExactlyOneTensor(inputs);\r\n      if (this.rate > 0 && this.rate < 1) {\r\n        const noised = () => {\r\n          const stddev = Math.sqrt(this.rate / (1 - this.rate));\r\n          return input.mul(K.randomNormal(input.shape, 1, stddev));\r\n        };\r\n        return K.inTrainPhase(noised, () => input, kwargs['training'] || false);\r\n      }\r\n      return input;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(GaussianDropout);\r\n\r\nexport declare interface AlphaDropoutArgs extends LayerArgs {\r\n  /** drop probability.  */\r\n  rate: number;\r\n  /**\r\n   * A 1-D `Tensor` of type `int32`, representing the\r\n   * shape for randomly generated keep/drop flags.\r\n   */\r\n  noiseShape?: Shape;\r\n}\r\n\r\n/**\r\n * Applies Alpha Dropout to the input.\r\n *\r\n * As it is a regularization layer, it is only active at training time.\r\n *\r\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\r\n * to their original values, in order to ensure the self-normalizing property\r\n * even after this dropout.\r\n * Alpha Dropout fits well to Scaled Exponential Linear Units\r\n * by randomly setting activations to the negative saturation value.\r\n *\r\n * Arguments:\r\n *   - `rate`: float, drop probability (as with `Dropout`).\r\n *     The multiplicative noise will have\r\n *     standard deviation `sqrt(rate / (1 - rate))`.\r\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\r\n *     shape for randomly generated keep/drop flags.\r\n *\r\n * Input shape:\r\n *   Arbitrary. Use the keyword argument `inputShape`\r\n *   (tuple of integers, does not include the samples axis)\r\n *   when using this layer as the first layer in a model.\r\n *\r\n * Output shape:\r\n *   Same shape as input.\r\n *\r\n * References:\r\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\r\n */\r\nexport class AlphaDropout extends Layer {\r\n  /** @nocollapse */\r\n  static className = 'AlphaDropout';\r\n  readonly rate: number;\r\n  readonly noiseShape: Shape;\r\n\r\n  constructor(args: AlphaDropoutArgs) {\r\n    super(args);\r\n    this.supportsMasking = true;\r\n    this.rate = args.rate;\r\n    this.noiseShape = args.noiseShape;\r\n  }\r\n\r\n  _getNoiseShape(inputs: Tensor|Tensor[]) {\r\n    return this.noiseShape || getExactlyOneTensor(inputs).shape;\r\n  }\r\n\r\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\r\n    return inputShape;\r\n  }\r\n\r\n  getConfig() {\r\n    const baseConfig = super.getConfig();\r\n    const config = {rate: this.rate};\r\n    Object.assign(config, baseConfig);\r\n    return config;\r\n  }\r\n\r\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\r\n    return tidy(() => {\r\n      if (this.rate < 1 && this.rate > 0) {\r\n        const noiseShape = this._getNoiseShape(inputs);\r\n\r\n        const droppedInputs = () => {\r\n          const input = getExactlyOneTensor(inputs);\r\n\r\n          const alpha = 1.6732632423543772848170429916717;\r\n          const scale = 1.0507009873554804934193349852946;\r\n\r\n          const alphaP = -alpha * scale;\r\n\r\n          let keptIdx = greaterEqual(randomUniform(noiseShape), this.rate);\r\n\r\n          keptIdx = K.cast(keptIdx, 'float32');  // get default dtype.\r\n\r\n          // Get affine transformation params.\r\n          const a = ((1 - this.rate) * (1 + this.rate * alphaP ** 2)) ** -0.5;\r\n          const b = -a * alphaP * this.rate;\r\n\r\n          // Apply mask.\r\n          const x = input.mul(keptIdx).add(keptIdx.add(-1).mul(alphaP));\r\n\r\n          return x.mul(a).add(b);\r\n        };\r\n        return K.inTrainPhase(\r\n            droppedInputs, () => getExactlyOneTensor(inputs),\r\n            kwargs['training'] || false);\r\n      }\r\n      return inputs;\r\n    });\r\n  }\r\n}\r\nserialization.registerClass(AlphaDropout);\r\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Normalization layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {moments, serialization, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {Constraint, ConstraintIdentifier, getConstraint, serializeConstraint} from '../constraints';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {getInitializer, Initializer, InitializerIdentifier, serializeInitializer} from '../initializers';\nimport {Shape} from '../keras_format/common';\nimport {getRegularizer, Regularizer, RegularizerIdentifier, serializeRegularizer} from '../regularizers';\nimport {Kwargs} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport * as math_utils from '../utils/math_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\n/**\n * Applies batch normalization on x given mean, var, beta and gamma.\n *\n * I.e. returns:\n *   `output = (x - mean) / (sqrt(var) + epsilon) * gamma + beta`\n *\n * @param x Input tensor.\n * @param mean Mean of batch.\n * @param variance Variance of batch.\n * @param beta Tensor with which to center the input.\n * @param gamma Tensor by which to scale the input.\n * @param epsilon Fuzz factor.\n * @returns The result of the batch normalization.\n */\nexport function batchNormalization(\n    x: Tensor, mean: Tensor, variance: Tensor, beta?: Tensor, gamma?: Tensor,\n    epsilon = 1e-3): Tensor {\n  let out: Tensor;\n  if (x.rank === 2) {\n    out = tfc.batchNorm2d(\n        x as Tensor2D, mean as Tensor2D | Tensor1D,\n        variance as Tensor2D | Tensor1D, beta as Tensor2D | Tensor1D,\n        gamma as Tensor2D | Tensor1D, epsilon);\n  } else if (x.rank === 3) {\n    // TODO(cais): Check rank; give proper error message.\n    out = tfc.batchNorm3d(\n        x as Tensor3D, mean as Tensor3D | Tensor1D,\n        variance as Tensor3D | Tensor1D, beta as Tensor3D | Tensor1D,\n        gamma as Tensor3D | Tensor1D, epsilon);\n  } else if (x.rank === 4) {\n    out = tfc.batchNorm4d(\n        x as Tensor4D, mean as Tensor4D | Tensor1D,\n        variance as Tensor4D | Tensor1D, beta as Tensor4D | Tensor1D,\n        gamma as Tensor4D | Tensor1D, epsilon);\n  } else {\n    throw new NotImplementedError(\n        `batchNormalization is not implemented for array of rank ${x.rank} ` +\n        `yet`);\n  }\n  return out;\n}\n\n/**\n * Non-broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction regularNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const normed =\n               batchNormalization(x, mean, variance, beta, gamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Broadcasting batch normalization for use in training (not inference).\n *\n * The input is normalized to zero mean and unit variance along the\n * `reductionAxes`, followed by scaling with `gamma` and shifted by `beta`.\n * The result of that is returned as the first element\n * of the returned `Array`. The other two elements are the mean and variance,\n * respectively.\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nfunction broadcastNormalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  return tidy(() => {\n           const meanAndVariance = tfc.moments(x, reductionAxes);\n           const mean = meanAndVariance.mean;\n           const variance = meanAndVariance.variance;\n           const targetShape: number[] = [];\n           for (const axis of math_utils.range(0, x.rank)) {\n             if (reductionAxes.indexOf(axis) !== -1) {\n               targetShape.push(1);\n             } else {\n               targetShape.push(x.shape[axis]);\n             }\n           }\n           const broadcastMean = mean.reshape(targetShape);\n           const broadcastVariance = variance.reshape(targetShape);\n           const broadcastGamma =\n               gamma == null ? null : gamma.reshape(targetShape);\n           const broadcastBeta =\n               beta == null ? null : beta.reshape(targetShape);\n           const normed = batchNormalization(\n               x, broadcastMean, broadcastVariance, broadcastBeta,\n               broadcastGamma, epsilon);\n           return [normed, mean, variance];\n         }) as [Tensor, Tensor, Tensor];\n}\n\n/**\n * Batch normalization for use in training (not inference).\n *\n * @param x Input tensor to be normalized.\n * @param gamma Tensor by which to scale the input.\n * @param beta Tensor by which to center the input.\n * @param reductionAxes Axes over which to normalize.\n * @param epsilon Fuzz factor.\n * @returns An `Array` of three `Tensors`:\n *   [normalized tensor, mean of input, variance of input].\n */\nexport function normalizeBatchInTraining(\n    x: Tensor, gamma: Tensor, beta: Tensor, reductionAxes: number[],\n    epsilon = 1e-3): [Tensor, Tensor, Tensor] {\n  if (util.arraysEqual(\n          reductionAxes.slice().sort(), math_utils.range(0, x.rank - 1))) {\n    return regularNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  } else {\n    return broadcastNormalizeBatchInTraining(\n        x, gamma, beta, reductionAxes, epsilon);\n  }\n}\n\nexport declare interface BatchNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The integer axis that should be normalized (typically the features axis).\n   * Defaults to -1.\n   *\n   * For instance, after a `Conv2D` layer with `data_format=\"channels_first\"`,\n   * set `axis=1` in `batchNormalization`.\n   */\n  axis?: number;\n\n  /**\n   * Momentum of the moving average. Defaults to 0.99.\n   */\n  momentum?: number;\n\n  /**\n   * Small float added to the variance to avoid dividing by zero. Defaults to\n   * 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Defaults to `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear (also e.g. `nn.relu`),\n   * this can be disabled since the scaling will be done by the next layer.\n   * Defaults to `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   *  Defaults to 'zeros'.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   *  Defaults to `ones`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving mean.\n   * Defaults to `zeros`\n   */\n  movingMeanInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the moving variance.\n   *  Defaults to 'Ones'.\n   */\n  movingVarianceInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Constraint for the beta weight.\n   */\n  betaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Constraint for gamma weight.\n   */\n  gammaConstraint?: ConstraintIdentifier|Constraint;\n\n  /**\n   * Regularizer for the beta weight.\n   */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /**\n   * Regularizer for the gamma weight.\n   */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class BatchNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'BatchNormalization';\n  private readonly axis: number;\n  private readonly momentum: number;\n  private readonly epsilon: number;\n  private readonly center: boolean;\n  private readonly scale: boolean;\n  private readonly betaInitializer: Initializer;\n  private readonly gammaInitializer: Initializer;\n  private readonly movingMeanInitializer: Initializer;\n  private readonly movingVarianceInitializer: Initializer;\n  private readonly betaConstraint: Constraint;\n  private readonly gammaConstraint: Constraint;\n  private readonly betaRegularizer: Regularizer;\n  private readonly gammaRegularizer: Regularizer;\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n  private movingMean: LayerVariable;\n  private movingVariance: LayerVariable;\n\n  constructor(args?: BatchNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.supportsMasking = true;\n    this.axis = args.axis == null ? -1 : args.axis;\n    this.momentum = args.momentum == null ? 0.99 : args.momentum;\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.movingMeanInitializer =\n        getInitializer(args.movingMeanInitializer || 'zeros');\n    this.movingVarianceInitializer =\n        getInitializer(args.movingVarianceInitializer || 'ones');\n    this.betaConstraint = getConstraint(args.betaConstraint);\n    this.gammaConstraint = getConstraint(args.gammaConstraint);\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const axis = this.axis >= 0 ? this.axis : (this.axis + inputShape.length);\n    const dim = inputShape[axis];\n    if (dim == null) {\n      throw new ValueError(\n          `Axis ${axis} of input tensor should have a defined dimension but ` +\n          `the layer received an input with shape ` +\n          `${JSON.stringify(inputShape)}.`);\n    }\n    this.inputSpec =\n        [new InputSpec({ndim: inputShape.length, axes: {[axis]: dim}})];\n    const shape = [dim];\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', shape, null, this.gammaInitializer, this.gammaRegularizer,\n          true, this.gammaConstraint);\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', shape, null, this.betaInitializer, this.betaRegularizer, true,\n          this.betaConstraint);\n    }\n    this.movingMean = this.addWeight(\n        'moving_mean', shape, null, this.movingMeanInitializer, null, false);\n    this.movingVariance = this.addWeight(\n        'moving_variance', shape, null, this.movingVarianceInitializer, null,\n        false);\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const training = kwargs['training'] == null ? false : kwargs['training'];\n      const input = getExactlyOneTensor(inputs);\n      const inputShape = input.shape;\n      const ndim = inputShape.length;\n      const reductionAxes = math_utils.range(0, ndim);\n      const axis = this.axis >= 0 ? this.axis : (this.axis + ndim);\n      reductionAxes.splice(axis, 1);\n      const broadcastShape = generic_utils.pyListRepeat(1, ndim);\n      broadcastShape[axis] = inputShape[axis];\n\n      const sortedReductionAxes = reductionAxes.slice();\n      sortedReductionAxes.sort();\n      const needsBroadcasting = !util.arraysEqual(\n          sortedReductionAxes, math_utils.range(0, ndim).slice(0, ndim - 1));\n\n      const normalizeInference: () => Tensor = () => {\n        if (needsBroadcasting) {\n          const broadcastMovingMean =\n              this.movingMean.read().reshape(broadcastShape);\n          const broadcastMovingVariance =\n              this.movingVariance.read().reshape(broadcastShape);\n          const broadcastBeta =\n              this.center ? this.beta.read().reshape(broadcastShape) : null;\n          const broadcastGamma =\n              this.scale ? this.gamma.read().reshape(broadcastShape) : null;\n          return batchNormalization(\n              input, broadcastMovingMean, broadcastMovingVariance,\n              broadcastBeta, broadcastGamma, this.epsilon);\n        } else {\n          return batchNormalization(\n              input, this.movingMean.read(), this.movingVariance.read(),\n              this.beta == null ? null : this.beta.read(),\n              this.gamma == null ? null : this.gamma.read(), this.epsilon);\n        }\n      };\n\n      if (!training) {\n        return normalizeInference();\n      }\n\n      const [normedTraining, mean, variance] = normalizeBatchInTraining(\n          input, this.gamma.read(), this.beta.read(), reductionAxes,\n          this.epsilon);\n\n      const doMovingAverage =\n          (variable: LayerVariable, value: Tensor, momentum: number): void => {\n            tfc.tidy(() => {\n              const decay = 1 - momentum;\n              const origValue = variable.read();\n              const updateDelta = origValue.sub(value).mul(decay);\n              variable.write(origValue.sub(updateDelta));\n            });\n          };\n\n      // Perform updates to moving mean and moving variance for training.\n      // Porting Note: In PyKeras, these updates to `movingMean` and\n      //   `movingAverage` are done as a deferred Graph, added to the `Layer`'s\n      //   `update`s using the `add_update()` method. Here we do it imperatively\n      //   and encapsulate the updates in a function that is invoked\n      //   immediately.\n      const updateMovingMeanAndVariance = () => {\n        doMovingAverage(this.movingMean, mean, this.momentum);\n        doMovingAverage(this.movingVariance, variance, this.momentum);\n      };\n      updateMovingMeanAndVariance();\n\n      return normedTraining;\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      momentum: this.momentum,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      movingMeanInitializer: serializeInitializer(this.movingMeanInitializer),\n      movingVarianceInitializer:\n          serializeInitializer(this.movingVarianceInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer),\n      betaConstraint: serializeConstraint(this.betaConstraint),\n      gammaConstraint: serializeConstraint(this.gammaConstraint)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(BatchNormalization);\n\nexport interface LayerNormalizationLayerArgs extends LayerArgs {\n  /**\n   * The axis or axes that should be normalized (typically, the feature axis.)\n   * Defaults to -1 (the last axis.)\n   */\n  axis?: number|number[];\n\n  /**\n   * A small positive float added to variance to avoid divison by zero.\n   * Defaults to 1e-3.\n   */\n  epsilon?: number;\n\n  /**\n   * If `true`, add offset of `beta` to normalized tensor.\n   * If `false`, `beta` is ignored.\n   * Default: `true`.\n   */\n  center?: boolean;\n\n  /**\n   * If `true`, multiply output by `gamma`.\n   * If `false`, `gamma` is not used.\n   * When the next layer is linear, this can be disabled since scaling will\n   * be done by the next layer.\n   * Default: `true`.\n   */\n  scale?: boolean;\n\n  /**\n   * Initializer for the beta weight.\n   * Default: `'zeros'`.\n   */\n  betaInitializer?: InitializerIdentifier|Initializer;\n\n  /**\n   * Initializer for the gamma weight.\n   * Default: `'ones'`.\n   */\n  gammaInitializer?: InitializerIdentifier|Initializer;\n\n  /** Regularizer for the beta weight. */\n  betaRegularizer?: RegularizerIdentifier|Regularizer;\n\n  /** Regularizer for the gamma weight. */\n  gammaRegularizer?: RegularizerIdentifier|Regularizer;\n}\n\nexport class LayerNormalization extends Layer {\n  /** @nocollapse */\n  static className = 'LayerNormalization';\n\n  private axis: number|number[];\n  readonly epsilon: number;\n  readonly center: boolean;\n  readonly scale: boolean;\n  readonly betaInitializer: Initializer;\n  readonly gammaInitializer: Initializer;\n  readonly betaRegularizer: Regularizer;\n  readonly gammaRegularizer: Regularizer;\n\n  private gamma: LayerVariable;\n  private beta: LayerVariable;\n\n  constructor(args?: LayerNormalizationLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.axis = args.axis == null ? -1 : args.axis;\n    if (typeof this.axis === 'number') {\n      if (!Number.isInteger(this.axis)) {\n        throw new Error(\n            `Expected axis to be an integer, but received ${this.axis}`);\n      }\n    } else if (Array.isArray(this.axis)) {\n      for (const axis of this.axis) {\n        if (!Number.isInteger(axis)) {\n          throw new Error(\n              `Expected axis to be an array of integers, ` +\n              `but received ${JSON.stringify(this.axis)}`);\n        }\n      }\n    } else {\n      throw new Error(\n          `Expected axis to be an integer or an array of integers, ` +\n          `but received ${JSON.stringify(this.axis)}`);\n    }\n\n    this.epsilon = args.epsilon == null ? 1e-3 : args.epsilon;\n    this.center = args.center == null ? true : args.center;\n    this.scale = args.scale == null ? true : args.scale;\n    this.betaInitializer = getInitializer(args.betaInitializer || 'zeros');\n    this.gammaInitializer = getInitializer(args.gammaInitializer || 'ones');\n    this.betaRegularizer = getRegularizer(args.betaRegularizer);\n    this.gammaRegularizer = getRegularizer(args.gammaRegularizer);\n\n    this.supportsMasking = true;\n  }\n\n  public build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    const nDims = inputShape.length;\n\n    // Convert axis to array and resolve negatives.\n    if (typeof this.axis === 'number') {\n      this.axis = [this.axis];\n    }\n    for (let i = 0; i < this.axis.length; ++i) {\n      if (this.axis[i] < 0) {\n        this.axis[i] += nDims;\n      }\n    }\n\n    // Further validate axes.\n    for (const axis of this.axis) {\n      if (axis < 0 || axis >= nDims) {\n        throw new Error(`Invalid axis: ${axis}`);\n      }\n    }\n    if (this.axis.length !== generic_utils.unique(this.axis).length) {\n      throw new Error(`Found duplicate axes in: ${this.axis}`);\n    }\n\n    const paramShape = this.axis.map(axis => inputShape[axis]) as number[];\n\n    const trainable = true;\n    if (this.scale) {\n      this.gamma = this.addWeight(\n          'gamma', paramShape, 'float32', this.gammaInitializer,\n          this.gammaRegularizer, trainable);\n    } else {\n      this.gamma = null;\n    }\n    if (this.center) {\n      this.beta = this.addWeight(\n          'beta', paramShape, 'float32', this.betaInitializer,\n          this.betaRegularizer, trainable);\n    } else {\n      this.beta = null;\n    }\n\n    this.built = true;\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    const input = getExactlyOneTensor(inputs);\n    const inputShape = input.shape;\n    const nDims = inputShape.length;\n\n    return tidy(() => {\n      const keepDims = true;\n      let {mean, variance} = moments(input, this.axis, keepDims);\n      const broadcastShape = generic_utils.pyListRepeat(1, nDims);\n      for (const dim of this.axis as number[]) {\n        broadcastShape[dim] = inputShape[dim];\n      }\n\n      const broadcast = (v: Tensor) => {\n        if (v != null && v.shape.length !== nDims &&\n            this.axis !== [nDims - 1]) {\n          return v.reshape(broadcastShape);\n        } else {\n          return v;\n        }\n      };\n\n      let scale = broadcast(this.gamma.read());\n      let offset = broadcast(this.beta.read());\n\n      // TODO(https://github.com/tensorflow/tfjs/issues/2120): The tiling below\n      // is a workaround for the limitation of core's batchNormalization?d don't\n      // support broadcasting in their gradients. In addition, the tiling is\n      // necessary to ensure correctness on the browser CPU backend regardless\n      // of forward or backward computation. Remove this workaround once the\n      // limitation is addressed. See .\n      const momentsTiling: number[] = [];\n      const scaleOffsetTiling: number[] = [];\n      for (let i = 0; i < nDims; ++i) {\n        if ((this.axis as number[]).indexOf(i) !== -1) {\n          momentsTiling.push(inputShape[i]);\n          scaleOffsetTiling.push(1);\n        } else {\n          momentsTiling.push(1);\n          scaleOffsetTiling.push(inputShape[i]);\n        }\n      }\n      mean = mean.tile(momentsTiling);\n      variance = variance.tile(momentsTiling);\n      scale = scale.tile(scaleOffsetTiling);\n      offset = offset.tile(scaleOffsetTiling);\n\n      return batchNormalization(\n          input, mean, variance, offset, scale, this.epsilon);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      axis: this.axis,\n      epsilon: this.epsilon,\n      center: this.center,\n      scale: this.scale,\n      betaInitializer: serializeInitializer(this.betaInitializer),\n      gammaInitializer: serializeInitializer(this.gammaInitializer),\n      betaRegularizer: serializeRegularizer(this.betaRegularizer),\n      gammaRegularizer: serializeRegularizer(this.gammaRegularizer)\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(LayerNormalization);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Padding Layers.\n */\n\n// Porting Note: In Python Keras, the padding layers are in convolutional.py,\n//   but we decided to put them in a separate file (padding.ts) for clarity.\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport {InputSpec, Layer, LayerArgs} from '../engine/topology';\nimport {ValueError} from '../errors';\nimport {DataFormat, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\n/**\n * Pads the middle dimension of a 3D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of 2 integers, how many zeros to add at the start and\n *   end of the middle dimension (i.e., dimension 1).\n * @return A padded 3D `tf.Tensor`.\n */\nexport function temporalPadding(x: Tensor, padding?: [number, number]): Tensor {\n  return tidy(() => {\n    if (x.rank !== 3) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 3-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [1, 1];\n    }\n    if (padding.length !== 2) {\n      throw new ValueError(\n          `temporalPadding expects input padding pattern to be a length-2 ` +\n          `array, but received a length-${padding.length} array.`);\n    }\n\n    const pattern: Array<[number, number]> = [[0, 0], padding, [0, 0]];\n    return tfc.pad(x, pattern);\n  });\n}\n\n/**\n * Pads the 2nd and 3rd dimensions of a 4D tensor.\n *\n * @param x Input `tf.Tensor` to be padded.\n * @param padding `Array` of two `Array`s, each of which is an `Array` of two\n *   integers. The amount of padding at the beginning and end of the 2nd and 3rd\n *   dimensions, respectively.\n * @param dataFormat 'channelsLast' (default) or 'channelsFirst'.\n * @return Padded 4D `tf.Tensor`.\n */\nexport function spatial2dPadding(\n    x: Tensor, padding?: [[number, number], [number, number]],\n    dataFormat?: DataFormat): Tensor {\n  return tidy(() => {\n    if (x.rank !== 4) {\n      throw new ValueError(\n          `temporalPadding expects input tensor to be 4-D, but received a ` +\n          `${x.rank}-D tensor.`);\n    }\n\n    if (padding == null) {\n      padding = [[1, 1], [1, 1]];\n    }\n    if (padding.length !== 2 || padding[0].length !== 2 ||\n        padding[1].length !== 2) {\n      throw new ValueError(\n          'spatial2dPadding expects `padding` to be an Array of two Arrays, ' +\n          'each of which is an Array of two integers.');\n    }\n\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (dataFormat !== 'channelsLast' && dataFormat !== 'channelsFirst') {\n      throw new ValueError(\n          `Unknown data format: ${dataFormat}. ` +\n          `Supported data formats are 'channelsLast' and 'channelsFirst.`);\n    }\n\n    let pattern: Array<[number, number]>;\n    if (dataFormat === 'channelsFirst') {\n      pattern = [[0, 0], [0, 0], padding[0], padding[1]];\n    } else {\n      pattern = [[0, 0], padding[0], padding[1], [0, 0]];\n    }\n\n    return tfc.pad(x, pattern);\n  });\n}\n\nexport declare interface ZeroPadding2DLayerArgs extends LayerArgs {\n  /**\n   * Integer, or `Array` of 2 integers, or `Array` of 2 `Array`s, each of\n   * which is an `Array` of 2 integers.\n   * - If integer, the same symmetric padding is applied to width and height.\n   * - If Array` of 2 integers, interpreted as two different symmetric values\n   *   for height and width:\n   *   `[symmetricHeightPad, symmetricWidthPad]`.\n   * - If `Array` of 2 `Array`s, interpreted as:\n   *   `[[topPad, bottomPad], [leftPad, rightPad]]`.\n   */\n  padding?: number|[number, number]|[[number, number], [number, number]];\n\n  /**\n   * One of `'channelsLast'` (default) and `'channelsFirst'`.\n   *\n   * The ordering of the dimensions in the inputs.\n   * `channelsLast` corresponds to inputs with shape\n   * `[batch, height, width, channels]` while `channelsFirst`\n   * corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\nexport class ZeroPadding2D extends Layer {\n  /** @nocollapse */\n  static className = 'ZeroPadding2D';\n  readonly dataFormat: DataFormat;\n  readonly padding: [[number, number], [number, number]];\n\n  constructor(args?: ZeroPadding2DLayerArgs) {\n    if (args == null) {\n      args = {};\n    }\n    super(args);\n\n    this.dataFormat =\n        args.dataFormat == null ? imageDataFormat() : args.dataFormat;\n    // TODO(cais): Maybe refactor the following logic surrounding `padding`\n    //   into a helper method.\n    if (args.padding == null) {\n      this.padding = [[1, 1], [1, 1]];\n    } else if (typeof args.padding === 'number') {\n      this.padding =\n          [[args.padding, args.padding], [args.padding, args.padding]];\n    } else {\n      args.padding = args.padding;\n      if (args.padding.length !== 2) {\n        throw new ValueError(\n            `ZeroPadding2D expects padding to be a length-2 array, but ` +\n            `received a length-${args.padding.length} array.`);\n      }\n\n      let heightPadding: [number, number];\n      let widthPadding: [number, number];\n      if (typeof args.padding[0] === 'number') {\n        heightPadding = [args.padding[0], args.padding[0]];\n        widthPadding = [args.padding[1] as number, args.padding[1] as number];\n      } else {\n        args.padding = args.padding as [[number, number], [number, number]];\n\n        if (args.padding[0].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects height padding to be a length-2 array, ` +\n              `but received a length-${args.padding[0].length} array.`);\n        }\n        heightPadding = args.padding[0] as [number, number];\n\n        if (args.padding[1].length !== 2) {\n          throw new ValueError(\n              `ZeroPadding2D expects width padding to be a length-2 array, ` +\n              `but received a length-${args.padding[1].length} array.`);\n        }\n        widthPadding = args.padding[1] as [number, number];\n      }\n      this.padding = [heightPadding, widthPadding];\n    }\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n\n    let rows: number;\n    let cols: number;\n    if (this.dataFormat === 'channelsFirst') {\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        rows = inputShape[2] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[3] != null && inputShape[3] >= 0) {\n        cols = inputShape[3] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      if (inputShape[1] != null && inputShape[1] >= 0) {\n        rows = inputShape[1] + this.padding[0][0] + this.padding[0][1];\n      } else {\n        rows = null;\n      }\n      if (inputShape[2] != null && inputShape[2] >= 0) {\n        cols = inputShape[2] + this.padding[1][0] + this.padding[1][1];\n      } else {\n        cols = null;\n      }\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(\n        () => spatial2dPadding(\n            getExactlyOneTensor(inputs), this.padding, this.dataFormat));\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      padding: this.padding,\n      dataFormat: this.dataFormat,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\nserialization.registerClass(ZeroPadding2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * TensorFlow.js Layers: Pooling Layers.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, Tensor3D, Tensor4D, Tensor5D, tidy} from '@tensorflow/tfjs-core';\n\nimport {imageDataFormat} from '../backend/common';\nimport * as K from '../backend/tfjs_backend';\nimport {checkDataFormat, checkPaddingMode, checkPoolMode} from '../common';\nimport {InputSpec} from '../engine/topology';\nimport {Layer, LayerArgs} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {DataFormat, PaddingMode, PoolMode, Shape} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {convOutputLength} from '../utils/conv_utils';\nimport {assertPositiveInteger} from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\n\nimport {preprocessConv2DInput, preprocessConv3DInput} from './convolutional';\n\n/**\n * 2D pooling.\n * @param x\n * @param poolSize\n * @param stridesdes strides. Defaults to [1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 2D pooling.\n */\nexport function pool2d(\n    x: Tensor, poolSize: [number, number], strides?: [number, number],\n    padding?: PaddingMode, dataFormat?: DataFormat,\n    poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // TODO(cais): Remove the preprocessing step once deeplearn.js supports\n    // dataFormat as an input argument.\n    x = preprocessConv2DInput(x, dataFormat);  // x is NHWC after preprocessing.\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      // TODO(cais): Rank check?\n      y = tfc.maxPool(x as Tensor4D, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      // TODO(cais): Check the dtype and rank of x and give clear error message\n      //   if those are incorrect.\n      y = tfc.avgPool(\n          // TODO(cais): Rank check?\n          x as Tensor3D | Tensor4D, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 3, 1, 2]);  // NHWC -> NCHW.\n    }\n    return y;\n  });\n}\n\n/**\n * 3D pooling.\n * @param x\n * @param poolSize. Default to [1, 1, 1].\n * @param strides strides. Defaults to [1, 1, 1].\n * @param padding padding. Defaults to 'valid'.\n * @param dataFormat data format. Defaults to 'channelsLast'.\n * @param poolMode Mode of pooling. Defaults to 'max'.\n * @returns Result of the 3D pooling.\n */\nexport function pool3d(\n    x: Tensor5D, poolSize: [number, number, number],\n    strides?: [number, number, number], padding?: PaddingMode,\n    dataFormat?: DataFormat, poolMode?: PoolMode): Tensor {\n  return tidy(() => {\n    checkDataFormat(dataFormat);\n    checkPoolMode(poolMode);\n    checkPaddingMode(padding);\n    if (strides == null) {\n      strides = [1, 1, 1];\n    }\n    if (padding == null) {\n      padding = 'valid';\n    }\n    if (dataFormat == null) {\n      dataFormat = imageDataFormat();\n    }\n    if (poolMode == null) {\n      poolMode = 'max';\n    }\n\n    // x is NDHWC after preprocessing.\n    x = preprocessConv3DInput(x as Tensor, dataFormat) as Tensor5D;\n    let y: Tensor;\n    const paddingString = (padding === 'same') ? 'same' : 'valid';\n    if (poolMode === 'max') {\n      y = tfc.maxPool3d(x, poolSize, strides, paddingString);\n    } else {  // 'avg'\n      y = tfc.avgPool3d(x, poolSize, strides, paddingString);\n    }\n    if (dataFormat === 'channelsFirst') {\n      y = tfc.transpose(y, [0, 4, 1, 2, 3]);  // NDHWC -> NCDHW.\n    }\n    return y;\n  });\n}\n\nexport declare interface Pooling1DLayerArgs extends LayerArgs {\n  /**\n   * Size of the window to pool over, should be an integer.\n   */\n  poolSize?: number|[number];\n  /**\n   * Period at which to sample the pooled values.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number];\n  /** How to fill in data that's not an integer multiple of poolSize. */\n  padding?: PaddingMode;\n}\n\n/**\n * Abstract class for different pooling 1D layers.\n */\nexport abstract class Pooling1D extends Layer {\n  protected readonly poolSize: [number];\n  protected readonly strides: [number];\n  protected readonly padding: PaddingMode;\n\n  /**\n   *\n   * @param args Parameters for the Pooling layer.\n   *\n   * config.poolSize defaults to 2.\n   */\n  constructor(args: Pooling1DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = 2;\n    }\n    super(args);\n    if (typeof args.poolSize === 'number') {\n      this.poolSize = [args.poolSize];\n    } else if (\n        Array.isArray(args.poolSize) &&\n        (args.poolSize as number[]).length === 1 &&\n        typeof (args.poolSize as number[])[0] === 'number') {\n      this.poolSize = args.poolSize;\n    } else {\n      throw new ValueError(\n          `poolSize for 1D convolutional layer must be a number or an ` +\n          `Array of a single number, but received ` +\n          `${JSON.stringify(args.poolSize)}`);\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else {\n      if (typeof args.strides === 'number') {\n        this.strides = [args.strides];\n      } else if (\n          Array.isArray(args.strides) &&\n          (args.strides as number[]).length === 1 &&\n          typeof (args.strides as number[])[0] === 'number') {\n        this.strides = args.strides;\n      } else {\n        throw new ValueError(\n            `strides for 1D convolutional layer must be a number or an ` +\n            `Array of a single number, but received ` +\n            `${JSON.stringify(args.strides)}`);\n      }\n    }\n    assertPositiveInteger(this.strides, 'strides');\n\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    checkPaddingMode(this.padding);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const length = convOutputLength(\n        inputShape[1], this.poolSize[0], this.padding, this.strides[0]);\n    return [inputShape[0], length, inputShape[2]];\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      // Add dummy last dimension.\n      inputs = K.expandDims(getExactlyOneTensor(inputs), 2);\n      const output = this.poolingFunction(\n          getExactlyOneTensor(inputs), [this.poolSize[0], 1],\n          [this.strides[0], 1], this.padding, 'channelsLast');\n      // Remove dummy last dimension.\n      return tfc.squeeze(output, [2]);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'MaxPooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling1D);\n\nexport class AveragePooling1D extends Pooling1D {\n  /** @nocollapse */\n  static className = 'AveragePooling1D';\n  constructor(args: Pooling1DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling1D);\n\nexport declare interface Pooling2DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [vertical, horizontal].\n   * Expects an integer or an array of 2 integers.\n   *\n   * For example, `[2, 2]` will halve the input in both spatial dimension.\n   * If only one integer is specified, the same window length\n   * will be used for both dimensions.\n   */\n  poolSize?: number|[number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 2 integers. Integer, tuple of 2 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 2D layers.\n */\nexport abstract class Pooling2D extends Layer {\n  protected readonly poolSize: [number, number];\n  protected readonly strides: [number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling2DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 2) {\n        throw new ValueError(\n            `If the strides property of a 2D pooling layer is an Array, ` +\n            `it is expected to have a length of 2, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    rows =\n        convOutputLength(rows, this.poolSize[0], this.padding, this.strides[0]);\n    cols =\n        convOutputLength(cols, this.poolSize[1], this.padding, this.strides[1]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], rows, cols];\n    } else {\n      return [inputShape[0], rows, cols, inputShape[3]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'MaxPooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling2D);\n\nexport class AveragePooling2D extends Pooling2D {\n  /** @nocollapse */\n  static className = 'AveragePooling2D';\n  constructor(args: Pooling2DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number], strides: [number, number],\n      padding: PaddingMode, dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool2d(inputs, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling2D);\n\nexport declare interface Pooling3DLayerArgs extends LayerArgs {\n  /**\n   * Factors by which to downscale in each dimension [depth, height, width].\n   * Expects an integer or an array of 3 integers.\n   *\n   * For example, `[2, 2, 2]` will halve the input in three dimensions.\n   * If only one integer is specified, the same window length\n   * will be used for all dimensions.\n   */\n  poolSize?: number|[number, number, number];\n\n  /**\n   * The size of the stride in each dimension of the pooling window. Expects\n   * an integer or an array of 3 integers. Integer, tuple of 3 integers, or\n   * None.\n   *\n   * If `null`, defaults to `poolSize`.\n   */\n  strides?: number|[number, number, number];\n\n  /** The padding type to use for the pooling layer. */\n  padding?: PaddingMode;\n  /** The data format to use for the pooling layer. */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different pooling 3D layers.\n */\nexport abstract class Pooling3D extends Layer {\n  protected readonly poolSize: [number, number, number];\n  protected readonly strides: [number, number, number];\n  protected readonly padding: PaddingMode;\n  protected readonly dataFormat: DataFormat;\n\n  constructor(args: Pooling3DLayerArgs) {\n    if (args.poolSize == null) {\n      args.poolSize = [2, 2, 2];\n    }\n    super(args);\n    this.poolSize = Array.isArray(args.poolSize) ?\n        args.poolSize :\n        [args.poolSize, args.poolSize, args.poolSize];\n    if (args.strides == null) {\n      this.strides = this.poolSize;\n    } else if (Array.isArray(args.strides)) {\n      if (args.strides.length !== 3) {\n        throw new ValueError(\n            `If the strides property of a 3D pooling layer is an Array, ` +\n            `it is expected to have a length of 3, but received length ` +\n            `${args.strides.length}.`);\n      }\n      this.strides = args.strides;\n    } else {\n      // `config.strides` is a number.\n      this.strides = [args.strides, args.strides, args.strides];\n    }\n    assertPositiveInteger(this.poolSize, 'poolSize');\n    assertPositiveInteger(this.strides, 'strides');\n    this.padding = args.padding == null ? 'valid' : args.padding;\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    checkPaddingMode(this.padding);\n\n    this.inputSpec = [new InputSpec({ndim: 5})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    let depths =\n        this.dataFormat === 'channelsFirst' ? inputShape[2] : inputShape[1];\n    let rows =\n        this.dataFormat === 'channelsFirst' ? inputShape[3] : inputShape[2];\n    let cols =\n        this.dataFormat === 'channelsFirst' ? inputShape[4] : inputShape[3];\n    depths = convOutputLength(\n        depths, this.poolSize[0], this.padding, this.strides[0]);\n    rows =\n        convOutputLength(rows, this.poolSize[1], this.padding, this.strides[1]);\n    cols =\n        convOutputLength(cols, this.poolSize[2], this.padding, this.strides[2]);\n    if (this.dataFormat === 'channelsFirst') {\n      return [inputShape[0], inputShape[1], depths, rows, cols];\n    } else {\n      return [inputShape[0], depths, rows, cols, inputShape[4]];\n    }\n  }\n\n  protected abstract poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor;\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      this.invokeCallHook(inputs, kwargs);\n      return this.poolingFunction(\n          getExactlyOneTensor(inputs), this.poolSize, this.strides,\n          this.padding, this.dataFormat);\n    });\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {\n      poolSize: this.poolSize,\n      padding: this.padding,\n      strides: this.strides,\n      dataFormat: this.dataFormat\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class MaxPooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'MaxPooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'max');\n  }\n}\nserialization.registerClass(MaxPooling3D);\n\nexport class AveragePooling3D extends Pooling3D {\n  /** @nocollapse */\n  static className = 'AveragePooling3D';\n  constructor(args: Pooling3DLayerArgs) {\n    super(args);\n  }\n\n  protected poolingFunction(\n      inputs: Tensor, poolSize: [number, number, number],\n      strides: [number, number, number], padding: PaddingMode,\n      dataFormat: DataFormat): Tensor {\n    checkDataFormat(dataFormat);\n    checkPaddingMode(padding);\n    return pool3d(\n        inputs as Tensor5D, poolSize, strides, padding, dataFormat, 'avg');\n  }\n}\nserialization.registerClass(AveragePooling3D);\n\n/**\n * Abstract class for different global pooling 1D layers.\n */\nexport abstract class GlobalPooling1D extends Layer {\n  constructor(args: LayerArgs) {\n    super(args);\n    this.inputSpec = [new InputSpec({ndim: 3})];\n  }\n\n  computeOutputShape(inputShape: Shape): Shape {\n    return [inputShape[0], inputShape[2]];\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n}\n\nexport class GlobalAveragePooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling1D';\n  constructor(args?: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.mean(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling1D);\n\nexport class GlobalMaxPooling1D extends GlobalPooling1D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling1D';\n  constructor(args: LayerArgs) {\n    super(args || {});\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      return tfc.max(input, 1);\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling1D);\n\nexport declare interface GlobalPooling2DLayerArgs extends LayerArgs {\n  /**\n   * One of `CHANNEL_LAST` (default) or `CHANNEL_FIRST`.\n   *\n   * The ordering of the dimensions in the inputs. `CHANNEL_LAST` corresponds\n   * to inputs with shape `[batch, height, width, channels[` while\n   * `CHANNEL_FIRST` corresponds to inputs with shape\n   * `[batch, channels, height, width]`.\n   */\n  dataFormat?: DataFormat;\n}\n\n/**\n * Abstract class for different global pooling 2D layers.\n */\nexport abstract class GlobalPooling2D extends Layer {\n  protected dataFormat: DataFormat;\n  constructor(args: GlobalPooling2DLayerArgs) {\n    super(args);\n    this.dataFormat =\n        args.dataFormat == null ? 'channelsLast' : args.dataFormat;\n    checkDataFormat(this.dataFormat);\n    this.inputSpec = [new InputSpec({ndim: 4})];\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = inputShape as Shape;\n    if (this.dataFormat === 'channelsLast') {\n      return [inputShape[0], inputShape[3]];\n    } else {\n      return [inputShape[0], inputShape[1]];\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    throw new NotImplementedError();\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config = {dataFormat: this.dataFormat};\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n}\n\nexport class GlobalAveragePooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalAveragePooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.mean(input, [1, 2]);\n      } else {\n        return tfc.mean(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalAveragePooling2D);\n\nexport class GlobalMaxPooling2D extends GlobalPooling2D {\n  /** @nocollapse */\n  static className = 'GlobalMaxPooling2D';\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const input = getExactlyOneTensor(inputs);\n      if (this.dataFormat === 'channelsLast') {\n        return tfc.max(input, [1, 2]);\n      } else {\n        return tfc.max(input, [2, 3]);\n      }\n    });\n  }\n}\nserialization.registerClass(GlobalMaxPooling2D);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/**\n * Layers that augment the functionality of a base layer.\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\nimport {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';\nimport * as K from '../backend/tfjs_backend';\nimport {nameScope} from '../common';\nimport {InputSpec, Layer, LayerArgs, SymbolicTensor} from '../engine/topology';\nimport {NotImplementedError, ValueError} from '../errors';\nimport {BidirectionalMergeMode, Shape, VALID_BIDIRECTIONAL_MERGE_MODES} from '../keras_format/common';\nimport {Kwargs} from '../types';\nimport {RegularizerFn, RnnStepFunction} from '../types';\nimport * as generic_utils from '../utils/generic_utils';\nimport {getExactlyOneShape, getExactlyOneTensor} from '../utils/types_utils';\nimport {LayerVariable} from '../variables';\n\nimport {rnn, RNN, standardizeArgs} from './recurrent';\nimport {deserialize} from './serialization';\n\nexport declare interface WrapperLayerArgs extends LayerArgs {\n  /**\n   * The layer to be wrapped.\n   */\n  layer: Layer;\n}\n\n/**\n * Abstract wrapper base class.\n *\n * Wrappers take another layer and augment it in various ways.\n * Do not use this class as a layer, it is only an abstract base class.\n * Two usable wrappers are the `TimeDistributed` and `Bidirectional` wrappers.\n */\nexport abstract class Wrapper extends Layer {\n  readonly layer: Layer;\n\n  constructor(args: WrapperLayerArgs) {\n    // Porting Note: In PyKeras, `self.layer` is set prior to the calling\n    //   `super()`. But we can't do that here due to TypeScript's restriction.\n    //   See: https://github.com/Microsoft/TypeScript/issues/8277\n    //   As a result, we have to add checks in `get trainable()` and\n    //   `set trainable()` below in order to prevent using `this.layer` when\n    //   its value is `undefined`. The super constructor does use the getter\n    //   and the setter of `this.layer`.\n    super(args);\n    this.layer = args.layer;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    this.built = true;\n  }\n\n  // TODO(cais): Implement activityRegularizer getter.\n\n  get trainable(): boolean {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      return this.layer.trainable;\n    } else {\n      return false;\n    }\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    if (this.layer != null) {\n      this.layer.trainable = value;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.layer.trainableWeights;\n  }\n  // TODO(cais): Implement setter for trainableWeights.\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.layer.nonTrainableWeights;\n  }\n  // TODO(cais): Implement setter for nonTrainableWeights.\n\n  get updates(): Tensor[] {\n    // tslint:disable-next-line:no-any\n    return (this.layer as any)._updates;\n  }\n\n  // TODO(cais): Implement getUpdatesFor().\n\n  get losses(): RegularizerFn[] {\n    return this.layer.losses;\n  }\n\n  // TODO(cais): Implement getLossesFor().\n\n  getWeights(): Tensor[] {\n    return this.layer.getWeights();\n  }\n\n  setWeights(weights: Tensor[]): void {\n    this.layer.setWeights(weights);\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'layer': {\n        'className': this.layer.getClassName(),\n        'config': this.layer.getConfig(),\n      }\n    };\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.layer != null) {\n      this.layer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict,\n      customObjects = {} as serialization.ConfigDict): T {\n    const layerConfig = config['layer'] as serialization.ConfigDict;\n    const layer = deserialize(layerConfig, customObjects) as Layer;\n    delete config['layer'];\n    const newConfig = {layer};\n    Object.assign(newConfig, config);\n    return new cls(newConfig);\n  }\n}\n\nexport class TimeDistributed extends Wrapper {\n  /** @nocollapse */\n  static className = 'TimeDistributed';\n  constructor(args: WrapperLayerArgs) {\n    super(args);\n    this.supportsMasking = true;\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    inputShape = getExactlyOneShape(inputShape);\n    if (inputShape.length < 3) {\n      throw new ValueError(\n          `TimeDistributed layer expects an input shape >= 3D, but received ` +\n          `input shape ${JSON.stringify(inputShape)}`);\n    }\n    this.inputSpec = [{shape: inputShape}];\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    if (!this.layer.built) {\n      this.layer.build(childInputShape);\n      this.layer.built = true;\n    }\n    super.build(inputShape);\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    inputShape = getExactlyOneShape(inputShape);\n    const childInputShape = [inputShape[0]].concat(inputShape.slice(2));\n    const childOutputShape =\n        this.layer.computeOutputShape(childInputShape) as Shape;\n    const timesteps = inputShape[1];\n    return [childOutputShape[0], timesteps].concat(childOutputShape.slice(1));\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      // TODO(cais): Add 'training' and 'useLearningPhase' to kwargs.\n      inputs = getExactlyOneTensor(inputs);\n      // Porting Note: In tfjs-layers, `inputs` are always concrete tensor\n      // values. Hence the inputs can't have an undetermined first (batch)\n      // dimension, which is why we always use the K.rnn approach here.\n      const step: RnnStepFunction = (inputs: Tensor, states: Tensor[]) => {\n        // TODO(cais): Add useLearningPhase.\n        // NOTE(cais): `layer.call` may return a length-1 array of Tensor in\n        //   some cases (e.g., `layer` is a `Sequential` instance), which is\n        //   why `getExactlyOneTensor` is used below.\n        const output = getExactlyOneTensor(this.layer.call(inputs, kwargs));\n        return [output, []];\n      };\n      const rnnOutputs =\n          rnn(step, inputs, [], false /* goBackwards */, null /* mask */,\n              null /* constants */, false /* unroll */,\n              true /* needPerStepOutputs */);\n      const y = rnnOutputs[1];\n      // TODO(cais): Add activity regularization.\n      // TODO(cais): Add useLearningPhase.\n      return y;\n    });\n  }\n\n  // TODO(cais): Implement detailed computeMask() logic.\n}\nserialization.registerClass(TimeDistributed);\n\nexport function checkBidirectionalMergeMode(value?: string): void {\n  generic_utils.checkStringTypeUnionValue(\n      VALID_BIDIRECTIONAL_MERGE_MODES, 'BidirectionalMergeMode', value);\n}\n\nexport declare interface BidirectionalLayerArgs extends WrapperLayerArgs {\n  /**\n   * The instance of an `RNN` layer to be wrapped.\n   */\n  layer: RNN;\n\n  /**\n   * Mode by which outputs of the forward and backward RNNs are\n   * combined. If `null` or `undefined`, the output will not be\n   * combined, they will be returned as an `Array`.\n   *\n   * If `undefined` (i.e., not provided), defaults to `'concat'`.\n   */\n  mergeMode?: BidirectionalMergeMode;\n}\n\nconst DEFAULT_BIDIRECTIONAL_MERGE_MODE: BidirectionalMergeMode = 'concat';\n\nexport class Bidirectional extends Wrapper {\n  /** @nocollapse */\n  static className = 'Bidirectional';\n  mergeMode: BidirectionalMergeMode;\n  private forwardLayer: RNN;\n  private backwardLayer: RNN;\n  private returnSequences: boolean;\n  private returnState: boolean;\n  private numConstants?: number;\n  private _trainable: boolean;\n\n  constructor(args: BidirectionalLayerArgs) {\n    super(args);\n\n    // Note: When creating `this.forwardLayer`, the original Layer object\n    //   (`config.layer`) ought to be cloned. This is why we call\n    //   `getConfig()` followed by `deserialize()`. Without this cloning,\n    //   the layer names saved during serialization will incorrectly contain\n    //   the 'forward_' prefix. In Python Keras, this is done using\n    //   `copy.copy` (shallow copy), which does not have a simple equivalent\n    //   in JavaScript. JavaScript's `Object.assign()` does not copy\n    //   methods.\n    const layerConfig = args.layer.getConfig();\n    const forwDict: serialization.ConfigDict = {};\n    forwDict['className'] = args.layer.getClassName();\n    forwDict['config'] = layerConfig;\n    this.forwardLayer = deserialize(forwDict) as RNN;\n    layerConfig['goBackwards'] =\n        layerConfig['goBackwards'] === true ? false : true;\n    const backDict: serialization.ConfigDict = {};\n    backDict['className'] = args.layer.getClassName();\n    backDict['config'] = layerConfig;\n    this.backwardLayer = deserialize(backDict) as RNN;\n    this.forwardLayer.name = 'forward_' + this.forwardLayer.name;\n    this.backwardLayer.name = 'backward_' + this.backwardLayer.name;\n\n    this.mergeMode = args.mergeMode === undefined ?\n        DEFAULT_BIDIRECTIONAL_MERGE_MODE :\n        args.mergeMode;\n    checkBidirectionalMergeMode(this.mergeMode);\n    if (args.weights) {\n      throw new NotImplementedError(\n          'weights support is not implemented for Bidirectional layer yet.');\n    }\n    this._stateful = args.layer.stateful;\n    this.returnSequences = args.layer.returnSequences;\n    this.returnState = args.layer.returnState;\n    this.supportsMasking = true;\n    this._trainable = true;\n    this.inputSpec = args.layer.inputSpec;\n    this.numConstants = null;\n  }\n\n  get trainable(): boolean {\n    return this._trainable;\n  }\n\n  set trainable(value: boolean) {\n    // Porting Note: the check of `this.layer` here is necessary due to the\n    //   way the `constructor` of this class is written (see Porting Note\n    //   above).\n    this._trainable = value;\n    if (this.forwardLayer != null) {\n      this.forwardLayer.trainable = value;\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.trainable = value;\n    }\n  }\n\n  getWeights(): Tensor[] {\n    return this.forwardLayer.getWeights().concat(\n        this.backwardLayer.getWeights());\n  }\n\n  setWeights(weights: Tensor[]): void {\n    const numWeights = weights.length;\n    const numeightsOver2 = Math.floor(numWeights / 2);\n    this.forwardLayer.setWeights(weights.slice(0, numeightsOver2));\n    this.backwardLayer.setWeights(weights.slice(numeightsOver2));\n  }\n\n  computeOutputShape(inputShape: Shape|Shape[]): Shape|Shape[] {\n    let layerShapes: Shape|Shape[] =\n        this.forwardLayer.computeOutputShape(inputShape);\n    if (!(Array.isArray(layerShapes) && Array.isArray(layerShapes[0]))) {\n      layerShapes = [layerShapes as Shape];\n    }\n    layerShapes = layerShapes as Shape[];\n\n    let outputShape: Shape;\n    let outputShapes: Shape[];\n    let stateShape: Shape[];\n    if (this.returnState) {\n      stateShape = layerShapes.slice(1);\n      outputShape = layerShapes[0];\n    } else {\n      outputShape = layerShapes[0];\n    }\n    outputShape = outputShape;\n    if (this.mergeMode === 'concat') {\n      outputShape[outputShape.length - 1] *= 2;\n      outputShapes = [outputShape];\n    } else if (this.mergeMode == null) {\n      outputShapes = [outputShape, outputShape.slice()];\n    } else {\n      outputShapes = [outputShape];\n    }\n\n    if (this.returnState) {\n      if (this.mergeMode == null) {\n        return outputShapes.concat(stateShape).concat(stateShape.slice());\n      }\n      return [outputShape].concat(stateShape).concat(stateShape.slice());\n    }\n    return generic_utils.singletonOrArray(outputShapes);\n  }\n\n  apply(\n      inputs: Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[],\n      kwargs?: Kwargs): Tensor|Tensor[]|SymbolicTensor|SymbolicTensor[] {\n    let initialState: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['initialState'];\n    let constants: Tensor[]|SymbolicTensor[] =\n        kwargs == null ? null : kwargs['constants'];\n    if (kwargs == null) {\n      kwargs = {};\n    }\n    const standardized =\n        standardizeArgs(inputs, initialState, constants, this.numConstants);\n    inputs = standardized.inputs;\n    initialState = standardized.initialState;\n    constants = standardized.constants;\n\n    if (Array.isArray(inputs)) {\n      initialState = (inputs as Tensor[] | SymbolicTensor[]).slice(1);\n      inputs = (inputs as Tensor[] | SymbolicTensor[])[0];\n    }\n\n    if ((initialState == null || initialState.length === 0) &&\n        constants == null) {\n      return super.apply(inputs, kwargs);\n    }\n    const additionalInputs: Array<Tensor|SymbolicTensor> = [];\n    const additionalSpecs: InputSpec[] = [];\n    if (initialState != null) {\n      const numStates = initialState.length;\n      if (numStates % 2 > 0) {\n        throw new ValueError(\n            'When passing `initialState` to a Bidrectional RNN, ' +\n            'the state should be an Array containing the states of ' +\n            'the underlying RNNs.');\n      }\n      kwargs['initialState'] = initialState;\n      additionalInputs.push(...initialState);\n      const stateSpecs = (initialState as Array<Tensor|SymbolicTensor>)\n                             .map(state => new InputSpec({shape: state.shape}));\n      this.forwardLayer.stateSpec = stateSpecs.slice(0, numStates / 2);\n      this.backwardLayer.stateSpec = stateSpecs.slice(numStates / 2);\n      additionalSpecs.push(...stateSpecs);\n    }\n    if (constants != null) {\n      throw new NotImplementedError(\n          'Support for constants in Bidirectional layers is not ' +\n          'implemented yet.');\n    }\n\n    const isSymbolicTensor = additionalInputs[0] instanceof SymbolicTensor;\n    for (const tensor of additionalInputs) {\n      if (tensor instanceof SymbolicTensor !== isSymbolicTensor) {\n        throw new ValueError(\n            'The initial state of a Bidirectional layer cannot be ' +\n            'specified as a mix of symbolic and non-symbolic tensors');\n      }\n    }\n\n    if (isSymbolicTensor) {\n      // Compute the full input and specs, including the states.\n      const fullInput = [inputs].concat(additionalInputs);\n      const fullInputSpec = this.inputSpec.concat(additionalSpecs);\n      // Perform the call temporarily and replace inputSpec.\n      // Note: with initial states symbolic calls and non-symbolic calls to\n      // this method differ in how the initial states are passed. For\n      // symbolic calls, the initial states are passed in the first arg, as\n      // an Array of SymbolicTensors; for non-symbolic calls, they are\n      // passed in the second arg as a part of the kwargs. Hence the need to\n      // temporarily modify inputSpec here.\n      // TODO(cais): Make refactoring so that this hacky code below is no\n      // longer needed.\n      const originalInputSpec = this.inputSpec;\n      this.inputSpec = fullInputSpec;\n      const output =\n          super.apply(fullInput as Tensor[] | SymbolicTensor[], kwargs);\n      this.inputSpec = originalInputSpec;\n      return output;\n    } else {\n      return super.apply(inputs, kwargs);\n    }\n  }\n\n  call(inputs: Tensor|Tensor[], kwargs: Kwargs): Tensor|Tensor[] {\n    return tidy(() => {\n      const initialState = kwargs['initialState'];\n\n      let y: Tensor|Tensor[];\n      let yRev: Tensor|Tensor[];\n      if (initialState == null) {\n        y = this.forwardLayer.call(inputs, kwargs);\n        yRev = this.backwardLayer.call(inputs, kwargs);\n      } else {\n        const forwardState = initialState.slice(0, initialState.length / 2);\n        const backwardState = initialState.slice(initialState.length / 2);\n        y = this.forwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: forwardState}));\n        yRev = this.backwardLayer.call(\n            inputs, Object.assign(kwargs, {initialState: backwardState}));\n      }\n\n      let states: Tensor[];\n      if (this.returnState) {\n        if (Array.isArray(y)) {\n          states = y.slice(1).concat((yRev as Tensor[]).slice(1));\n        } else {\n        }\n        y = (y as Tensor[])[0];\n        yRev = (yRev as Tensor[])[0];\n      }\n\n      if (this.returnSequences) {\n        yRev = tfc.reverse(yRev as Tensor, 1);\n      }\n\n      let output: Tensor|Tensor[];\n      if (this.mergeMode === 'concat') {\n        output = K.concatenate([y as Tensor, yRev as Tensor]);\n      } else if (this.mergeMode === 'sum') {\n        output = tfc.add(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode === 'ave') {\n        output = tfc.mul(.5, tfc.add(y as Tensor, yRev as Tensor));\n      } else if (this.mergeMode === 'mul') {\n        output = tfc.mul(y as Tensor, yRev as Tensor);\n      } else if (this.mergeMode == null) {\n        output = [y as Tensor, yRev as Tensor];\n      }\n\n      // TODO(cais): Properly set learning phase.\n      if (this.returnState) {\n        if (this.mergeMode == null) {\n          return (output as Tensor[]).concat(states);\n        }\n        return [output as Tensor].concat(states);\n      }\n      return output;\n    });\n  }\n\n  resetStates(states?: Tensor|Tensor[]): void {\n    this.forwardLayer.resetStates();\n    this.backwardLayer.resetStates();\n  }\n\n  build(inputShape: Shape|Shape[]): void {\n    nameScope(this.forwardLayer.name, () => {\n      this.forwardLayer.build(inputShape);\n    });\n    nameScope(this.backwardLayer.name, () => {\n      this.backwardLayer.build(inputShape);\n    });\n    this.built = true;\n  }\n\n  computeMask(inputs: Tensor|Tensor[], mask?: Tensor|Tensor[]): Tensor\n      |Tensor[] {\n    if (Array.isArray(mask)) {\n      mask = mask[0];\n    }\n    let outputMask: Tensor|Tensor[];\n    if (this.returnSequences) {\n      if (this.mergeMode == null) {\n        outputMask = [mask, mask];\n      } else {\n        outputMask = mask;\n      }\n    } else {\n      if (this.mergeMode == null) {\n        outputMask = [null, null];\n      } else {\n        outputMask = null;\n      }\n    }\n    if (this.returnState) {\n      const states = this.forwardLayer.states;\n      const stateMask: Tensor[] = states.map(state => null);\n      if (Array.isArray(outputMask)) {\n        return outputMask.concat(stateMask).concat(stateMask);\n      } else {\n        return [outputMask].concat(stateMask).concat(stateMask);\n      }\n    } else {\n      return outputMask;\n    }\n  }\n\n  get trainableWeights(): LayerVariable[] {\n    return this.forwardLayer.trainableWeights.concat(\n        this.backwardLayer.trainableWeights);\n  }\n\n  get nonTrainableWeights(): LayerVariable[] {\n    return this.forwardLayer.nonTrainableWeights.concat(\n        this.backwardLayer.nonTrainableWeights);\n  }\n\n  // TODO(cais): Implement constraints().\n\n  setFastWeightInitDuringBuild(value: boolean) {\n    super.setFastWeightInitDuringBuild(value);\n    if (this.forwardLayer != null) {\n      this.forwardLayer.setFastWeightInitDuringBuild(value);\n    }\n    if (this.backwardLayer != null) {\n      this.backwardLayer.setFastWeightInitDuringBuild(value);\n    }\n  }\n\n  getConfig(): serialization.ConfigDict {\n    const config: serialization.ConfigDict = {\n      'mergeMode': this.mergeMode,\n    };\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    const baseConfig = super.getConfig();\n    Object.assign(config, baseConfig);\n    return config;\n  }\n\n  /** @nocollapse */\n  static fromConfig<T extends serialization.Serializable>(\n      cls: serialization.SerializableConstructor<T>,\n      config: serialization.ConfigDict): T {\n    const rnnLayer =\n        deserialize(config['layer'] as serialization.ConfigDict) as RNN;\n    delete config['layer'];\n    // TODO(cais): Add logic for `numConstants` once the property is added.\n    if (config['numConstants'] != null) {\n      throw new NotImplementedError(\n          `Deserialization of a Bidirectional layer with numConstants ` +\n          `present is not supported yet.`);\n    }\n    // tslint:disable-next-line:no-any\n    const newConfig: {[key: string]: any} = config;\n    newConfig['layer'] = rnnLayer;\n    return new cls(newConfig);\n  }\n}\nserialization.registerClass(Bidirectional);\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nimport {InputLayer, InputLayerArgs} from './engine/input_layer';\nimport {Layer, LayerArgs} from './engine/topology';\nimport {input} from './exports';\nimport {ELU, ELULayerArgs, LeakyReLU, LeakyReLULayerArgs, PReLU, PReLULayerArgs, ReLU, ReLULayerArgs, Softmax, SoftmaxLayerArgs, ThresholdedReLU, ThresholdedReLULayerArgs} from './layers/advanced_activations';\nimport {Conv1D, Conv2D, Conv2DTranspose, Conv3D, ConvLayerArgs, Cropping2D, Cropping2DLayerArgs, SeparableConv2D, SeparableConvLayerArgs, UpSampling2D, UpSampling2DLayerArgs} from './layers/convolutional';\nimport {DepthwiseConv2D, DepthwiseConv2DLayerArgs} from './layers/convolutional_depthwise';\nimport {ConvLSTM2D, ConvLSTM2DArgs, ConvLSTM2DCell, ConvLSTM2DCellArgs} from './layers/convolutional_recurrent';\nimport {Activation, ActivationLayerArgs, Dense, DenseLayerArgs, Dropout, DropoutLayerArgs, Flatten, FlattenLayerArgs, Masking, MaskingArgs, Permute, PermuteLayerArgs, RepeatVector, RepeatVectorLayerArgs, Reshape, ReshapeLayerArgs, SpatialDropout1D, SpatialDropout1DLayerConfig} from './layers/core';\nimport {Embedding, EmbeddingLayerArgs} from './layers/embeddings';\nimport {Add, Average, Concatenate, ConcatenateLayerArgs, Dot, DotLayerArgs, Maximum, Minimum, Multiply} from './layers/merge';\nimport {AlphaDropout, AlphaDropoutArgs, GaussianDropout, GaussianDropoutArgs, GaussianNoise, GaussianNoiseArgs} from './layers/noise';\nimport {BatchNormalization, BatchNormalizationLayerArgs, LayerNormalization, LayerNormalizationLayerArgs} from './layers/normalization';\nimport {ZeroPadding2D, ZeroPadding2DLayerArgs} from './layers/padding';\nimport {AveragePooling1D, AveragePooling2D, AveragePooling3D, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalPooling2DLayerArgs, MaxPooling1D, MaxPooling2D, MaxPooling3D, Pooling1DLayerArgs, Pooling2DLayerArgs, Pooling3DLayerArgs} from './layers/pooling';\nimport {GRU, GRUCell, GRUCellLayerArgs, GRULayerArgs, LSTM, LSTMCell, LSTMCellLayerArgs, LSTMLayerArgs, RNN, RNNCell, RNNLayerArgs, SimpleRNN, SimpleRNNCell, SimpleRNNCellLayerArgs, SimpleRNNLayerArgs, StackedRNNCells, StackedRNNCellsArgs} from './layers/recurrent';\nimport {Bidirectional, BidirectionalLayerArgs, TimeDistributed, WrapperLayerArgs} from './layers/wrappers';\n\n// TODO(cais): Add doc string to all the public static functions in this\n//   class; include exectuable JavaScript code snippets where applicable\n//   (b/74074458).\n\n// Input Layer.\n/**\n * An input layer is an entry point into a `tf.LayersModel`.\n *\n * `InputLayer` is generated automatically for `tf.Sequential`` models by\n * specifying the `inputshape` or `batchInputShape` for the first layer.  It\n * should not be specified explicitly. However, it can be useful sometimes,\n * e.g., when constructing a sequential model from a subset of another\n * sequential model's layers. Like the code snippet below shows.\n *\n * ```js\n * // Define a model which simply adds two inputs.\n * const model1 = tf.sequential();\n * model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));\n * model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));\n * model1.summary();\n * model1.predict(tf.zeros([1, 4])).print();\n *\n * // Construct another model, reusing the second layer of `model1` while\n * // not using the first layer of `model1`. Note that you cannot add the second\n * // layer of `model` directly as the first layer of the new sequential model,\n * // because doing so will lead to an error related to the fact that the layer\n * // is not an input layer. Instead, you need to create an `inputLayer` and add\n * // it to the new sequential model before adding the reused layer.\n * const model2 = tf.sequential();\n * // Use an inputShape that matches the input shape of `model1`'s second\n * // layer.\n * model2.add(tf.layers.inputLayer({inputShape: [3]}));\n * model2.add(model1.layers[1]);\n * model2.summary();\n * model2.predict(tf.zeros([1, 3])).print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}\n */\nexport function inputLayer(args: InputLayerArgs): Layer {\n  return new InputLayer(args);\n}\n\n// Advanced Activation Layers.\n\n/**\n * Exponetial Linear Unit (ELU).\n *\n * It follows:\n * `f(x) =  alpha * (exp(x) - 1.) for x < 0`,\n * `f(x) = x for x >= 0`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Fast and Accurate Deep Network Learning by Exponential Linear Units\n * (ELUs)](https://arxiv.org/abs/1511.07289v1)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function elu(args?: ELULayerArgs): Layer {\n  return new ELU(args);\n}\n\n/**\n * Rectified Linear Unit activation function.\n *\n * Input shape:\n *   Arbitrary. Use the config field `inputShape` (Array of integers, does\n *   not include the sample axis) when using this layer as the first layer\n *   in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function reLU(args?: ReLULayerArgs): Layer {\n  return new ReLU(args);\n}\n\n/**\n * Leaky version of a rectified linear unit.\n *\n * It allows a small gradient when the unit is not active:\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function leakyReLU(args?: LeakyReLULayerArgs): Layer {\n  return new LeakyReLU(args);\n}\n\n/**\n * Parameterized version of a leaky rectified linear unit.\n *\n * It follows\n * `f(x) = alpha * x for x < 0.`\n * `f(x) = x for x >= 0.`\n * wherein `alpha` is a trainable weight.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function prelu(args?: PReLULayerArgs): Layer {\n  return new PReLU(args);\n}\n\n/**\n * Softmax activation layer.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function softmax(args?: SoftmaxLayerArgs): Layer {\n  return new Softmax(args);\n}\n\n/**\n * Thresholded Rectified Linear Unit.\n *\n * It follows:\n * `f(x) = x for x > theta`,\n * `f(x) = 0 otherwise`.\n *\n * Input shape:\n *   Arbitrary. Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n * Output shape:\n *   Same shape as the input.\n *\n * References:\n *   - [Zero-Bias Autoencoders and the Benefits of Co-Adapting\n * Features](http://arxiv.org/abs/1402.3337)\n *\n * @doc {\n *   heading: 'Layers',\n *   subheading: 'Advanced Activation',\n *   namespace: 'layers'\n * }\n */\nexport function thresholdedReLU(args?: ThresholdedReLULayerArgs): Layer {\n  return new ThresholdedReLU(args);\n}\n\n// Convolutional Layers.\n\n/**\n * 1D convolution layer (e.g., temporal convolution).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input over a single spatial (or temporal) dimension\n * to produce a tensor of outputs.\n *\n * If `use_bias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model, provide an\n * `inputShape` argument `Array` or `null`.\n *\n * For example, `inputShape` would be:\n * - `[10, 128]` for sequences of 10 vectors of 128-dimensional vectors\n * - `[null, 128]` for variable-length sequences of 128-dimensional vectors.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional',  namespace: 'layers'}\n */\nexport function conv1d(args: ConvLayerArgs): Layer {\n  return new Conv1D(args);\n}\n\n/**\n * 2D convolution layer (e.g. spatial convolution over images).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 3]` for 128x128 RGB pictures\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2d(args: ConvLayerArgs): Layer {\n  return new Conv2D(args);\n}\n\n/**\n * Transposed convolutional layer (sometimes called Deconvolution).\n *\n * The need for transposed convolutions generally arises\n * from the desire to use a transformation going in the opposite direction of\n * a normal convolution, i.e., from something that has the shape of the output\n * of some convolution to something that has the shape of its input while\n * maintaining a connectivity pattern that is compatible with said\n * convolution.\n *\n * When using this layer as the first layer in a model, provide the\n * configuration `inputShape` (`Array` of integers, does not include the\n * sample axis), e.g., `inputShape: [128, 128, 3]` for 128x128 RGB pictures in\n * `dataFormat: 'channelsLast'`.\n *\n * Input shape:\n *   4D tensor with shape:\n *   `[batch, channels, rows, cols]` if `dataFormat` is `'channelsFirst'`.\n *   or 4D tensor with shape\n *   `[batch, rows, cols, channels]` if `dataFormat` is `'channelsLast`.\n *\n * Output shape:\n *   4D tensor with shape:\n *   `[batch, filters, newRows, newCols]` if `dataFormat` is\n * `'channelsFirst'`. or 4D tensor with shape:\n *   `[batch, newRows, newCols, filters]` if `dataFormat` is `'channelsLast'`.\n *\n * References:\n *   - [A guide to convolution arithmetic for deep\n * learning](https://arxiv.org/abs/1603.07285v1)\n *   - [Deconvolutional\n * Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv2dTranspose(args: ConvLayerArgs): Layer {\n  return new Conv2DTranspose(args);\n}\n\n/**\n * 3D convolution layer (e.g. spatial convolution over volumes).\n *\n * This layer creates a convolution kernel that is convolved\n * with the layer input to produce a tensor of outputs.\n *\n * If `useBias` is True, a bias vector is created and added to the outputs.\n *\n * If `activation` is not `null`, it is applied to the outputs as well.\n *\n * When using this layer as the first layer in a model,\n * provide the keyword argument `inputShape`\n * (Array of integers, does not include the sample axis),\n * e.g. `inputShape=[128, 128, 128, 1]` for 128x128x128 grayscale volumes\n * in `dataFormat='channelsLast'`.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function conv3d(args: ConvLayerArgs): Layer {\n  return new Conv3D(args);\n}\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Separable convolution consists of first performing\n * a depthwise spatial convolution\n * (which acts on each input channel separately)\n * followed by a pointwise convolution which mixes together the resulting\n * output channels. The `depthMultiplier` argument controls how many\n * output channels are generated per input channel in the depthwise step.\n *\n * Intuitively, separable convolutions can be understood as\n * a way to factorize a convolution kernel into two smaller kernels,\n * or as an extreme version of an Inception block.\n *\n * Input shape:\n *   4D tensor with shape:\n *     `[batch, channels, rows, cols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, rows, cols, channels]` if data_format='channelsLast'.\n *\n * Output shape:\n *   4D tensor with shape:\n *     `[batch, filters, newRows, newCols]` if data_format='channelsFirst'\n *   or 4D tensor with shape:\n *     `[batch, newRows, newCols, filters]` if data_format='channelsLast'.\n *     `rows` and `cols` values might have changed due to padding.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function separableConv2d(args: SeparableConvLayerArgs): Layer {\n  return new SeparableConv2D(args);\n}\n\n/**\n * Cropping layer for 2D input (e.g., image).\n *\n * This layer can crop an input\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, croppedRows, croppedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, croppedRows, croppedCols]`.\n *\n * Examples\n * ```js\n *\n * const model = tf.sequential();\n * model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]],\n *                                inputShape: [128, 128, 3]}));\n * //now output shape is [batch, 124, 124, 3]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function cropping2D(args: Cropping2DLayerArgs): Layer {\n  return new Cropping2D(args);\n}\n\n/**\n * Upsampling layer for 2D inputs.\n *\n * Repeats the rows and columns of the data\n * by size[0] and size[1] respectively.\n *\n *\n * Input shape:\n *    4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *         `[batch, rows, cols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *        `[batch, channels, rows, cols]`\n *\n * Output shape:\n *     4D tensor with shape:\n *     - If `dataFormat` is `\"channelsLast\"`:\n *        `[batch, upsampledRows, upsampledCols, channels]`\n *     - If `dataFormat` is `\"channelsFirst\"`:\n *         `[batch, channels, upsampledRows, upsampledCols]`\n *\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function upSampling2d(args: UpSampling2DLayerArgs): Layer {\n  return new UpSampling2D(args);\n}\n\n// Convolutional(depthwise) Layers.\n\n/**\n * Depthwise separable 2D convolution.\n *\n * Depthwise Separable convolutions consists in performing just the first step\n * in a depthwise spatial convolution (which acts on each input channel\n * separately). The `depthMultplier` argument controls how many output channels\n * are generated per input channel in the depthwise step.\n *\n * @doc {heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}\n */\nexport function depthwiseConv2d(args: DepthwiseConv2DLayerArgs): Layer {\n  return new DepthwiseConv2D(args);\n}\n\n// Basic Layers.\n\n/**\n * Applies an activation function to an output.\n *\n * This layer applies element-wise activation function.  Other layers, notably\n * `dense` can also apply activation functions.  Use this isolated activation\n * function to extract the values before and after the\n * activation. For instance:\n *\n * ```js\n * const input = tf.input({shape: [5]});\n * const denseLayer = tf.layers.dense({units: 1});\n * const activationLayer = tf.layers.activation({activation: 'relu6'});\n *\n * // Obtain the output symbolic tensors by applying the layers in order.\n * const denseOutput = denseLayer.apply(input);\n * const activationOutput = activationLayer.apply(denseOutput);\n *\n * // Create the model based on the inputs.\n * const model = tf.model({\n *     inputs: input,\n *     outputs: [denseOutput, activationOutput]\n * });\n *\n * // Collect both outputs and print separately.\n * const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));\n * denseOut.print();\n * activationOut.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function activation(args: ActivationLayerArgs): Layer {\n  return new Activation(args);\n}\n\n/**\n * Creates a dense (fully connected) layer.\n *\n * This layer implements the operation:\n *   `output = activation(dot(input, kernel) + bias)`\n *\n * `activation` is the element-wise activation function\n *   passed as the `activation` argument.\n *\n * `kernel` is a weights matrix created by the layer.\n *\n * `bias` is a bias vector created by the layer (only applicable if `useBias`\n * is `true`).\n *\n * **Input shape:**\n *\n *   nD `tf.Tensor` with shape: `(batchSize, ..., inputDim)`.\n *\n *   The most common situation would be\n *   a 2D input with shape `(batchSize, inputDim)`.\n *\n * **Output shape:**\n *\n *   nD tensor with shape: `(batchSize, ..., units)`.\n *\n *   For instance, for a 2D input with shape `(batchSize, inputDim)`,\n *   the output would have shape `(batchSize, units)`.\n *\n * Note: if the input to the layer has a rank greater than 2, then it is\n * flattened prior to the initial dot product with the kernel.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dense(args: DenseLayerArgs): Layer {\n  return new Dense(args);\n}\n\n/**\n * Applies\n * [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to\n * the input.\n *\n * Dropout consists in randomly setting a fraction `rate` of input units to 0 at\n * each update during training time, which helps prevent overfitting.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function dropout(args: DropoutLayerArgs): Layer {\n  return new Dropout(args);\n}\n\n/**\n * Spatial 1D version of Dropout.\n *\n * This Layer type performs the same function as the Dropout layer, but it drops\n * entire 1D feature maps instead of individual elements. For example, if an\n * input example consists of 3 timesteps and the feature map for each timestep\n * has a size of 4, a `spatialDropout1d` layer may zero out the feature maps\n * of the 1st timesteps and 2nd timesteps completely while sparing all feature\n * elements of the 3rd timestep.\n *\n * If adjacent frames (timesteps) are strongly correlated (as is normally the\n * case in early convolution layers), regular dropout will not regularize the\n * activation and will otherwise just result in merely an effective learning\n * rate decrease. In this case, `spatialDropout1d` will help promote\n * independence among feature maps and should be used instead.\n *\n * **Arguments:**\n *   rate: A floating-point number >=0 and <=1. Fraction of the input elements\n *     to drop.\n *\n * **Input shape:**\n *   3D tensor with shape `(samples, timesteps, channels)`.\n *\n * **Output shape:**\n *   Same as the input shape.\n *\n * References:\n *   - [Efficient Object Localization Using Convolutional\n *      Networks](https://arxiv.org/abs/1411.4280)\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function spatialDropout1d(args: SpatialDropout1DLayerConfig): Layer {\n  return new SpatialDropout1D(args);\n}\n\n/**\n * Flattens the input. Does not affect the batch size.\n *\n * A `Flatten` layer flattens each batch in its inputs to 1D (making the output\n * 2D).\n *\n * For example:\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const flattenLayer = tf.layers.flatten();\n * // Inspect the inferred output shape of the flatten layer, which\n * // equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the\n * // flattening. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(flattenLayer.apply(input).shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function flatten(args?: FlattenLayerArgs): Layer {\n  return new Flatten(args);\n}\n\n/**\n * Repeats the input n times in a new dimension.\n *\n * ```js\n *  const model = tf.sequential();\n *  model.add(tf.layers.repeatVector({n: 4, inputShape: [2]}));\n *  const x = tf.tensor2d([[10, 20]]);\n *  // Use the model to do inference on a data point the model hasn't see\n *  model.predict(x).print();\n *  // output shape is now [batch, 2, 4]\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function repeatVector(args: RepeatVectorLayerArgs): Layer {\n  return new RepeatVector(args);\n}\n\n/**\n * Reshapes an input to a certain shape.\n *\n * ```js\n * const input = tf.input({shape: [4, 3]});\n * const reshapeLayer = tf.layers.reshape({targetShape: [2, 6]});\n * // Inspect the inferred output shape of the Reshape layer, which\n * // equals `[null, 2, 6]`. (The 1st dimension is the undermined batch size.)\n * console.log(JSON.stringify(reshapeLayer.apply(input).shape));\n * ```\n *\n * Input shape:\n *   Arbitrary, although all dimensions in the input shape must be fixed.\n *   Use the configuration `inputShape` when using this layer as the\n *   first layer in a model.\n *\n *\n * Output shape:\n *   [batchSize, targetShape[0], targetShape[1], ...,\n *    targetShape[targetShape.length - 1]].\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function reshape(args: ReshapeLayerArgs): Layer {\n  return new Reshape(args);\n}\n\n/**\n * Permutes the dimensions of the input according to a given pattern.\n *\n * Useful for, e.g., connecting RNNs and convnets together.\n *\n * Example:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.permute({\n *   dims: [2, 1],\n *   inputShape: [10, 64]\n * }));\n * console.log(model.outputShape);\n * // Now model's output shape is [null, 64, 10], where null is the\n * // unpermuted sample (batch) dimension.\n * ```\n *\n * Input shape:\n *   Arbitrary. Use the configuration field `inputShape` when using this\n *   layer as the first layer in a model.\n *\n * Output shape:\n *   Same rank as the input shape, but with the dimensions re-ordered (i.e.,\n *   permuted) according to the `dims` configuration of this layer.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function permute(args: PermuteLayerArgs): Layer {\n  return new Permute(args);\n}\n\n/**\n * Maps positive integers (indices) into dense vectors of fixed size.\n * eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]\n *\n * **Input shape:** 2D tensor with shape: `[batchSize, sequenceLength]`.\n *\n * **Output shape:** 3D tensor with shape: `[batchSize, sequenceLength,\n * outputDim]`.\n *\n * @doc {heading: 'Layers', subheading: 'Basic', namespace: 'layers'}\n */\nexport function embedding(args: EmbeddingLayerArgs): Layer {\n  return new Embedding(args);\n}\n\n// Merge Layers.\n\n/**\n * Layer that performs element-wise addition on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). The inputs are specified as an\n * `Array` when the `apply` method of the `Add` layer instance is called. For\n * example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const addLayer = tf.layers.add();\n * const sum = addLayer.apply([input1, input2]);\n * console.log(JSON.stringify(sum.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function add(args?: LayerArgs): Layer {\n  return new Add(args);\n}\n\n/**\n * Layer that performs element-wise averaging on an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape, and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const averageLayer = tf.layers.average();\n * const average = averageLayer.apply([input1, input2]);\n * console.log(JSON.stringify(average.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function average(args?: LayerArgs): Layer {\n  return new Average(args);\n}\n\n/**\n * Layer that concatenates an `Array` of inputs.\n *\n * It takes a list of tensors, all of the same shape except for the\n * concatenation axis, and returns a single tensor, the concatenation\n * of all inputs. For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 3]});\n * const concatLayer = tf.layers.concatenate();\n * const output = concatLayer.apply([input1, input2]);\n * console.log(JSON.stringify(output.shape));\n * // You get [null, 2, 5], with the first dimension as the undetermined batch\n * // dimension. The last dimension (5) is the result of concatenating the\n * // last dimensions of the inputs (2 and 3).\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function concatenate(args?: ConcatenateLayerArgs): Layer {\n  return new Concatenate(args);\n}\n\n/**\n * Layer that computes the element-wise maximum an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const maxLayer = tf.layers.maximum();\n * const max = maxLayer.apply([input1, input2]);\n * console.log(JSON.stringify(max.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function maximum(args?: LayerArgs): Layer {\n  return new Maximum(args);\n}\n\n/**\n * Layer that computes the element-wise minimum of an `Array` of inputs.\n *\n * It takes as input a list of tensors, all of the same shape and returns a\n * single tensor (also of the same shape). For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const minLayer = tf.layers.minimum();\n * const min = minLayer.apply([input1, input2]);\n * console.log(JSON.stringify(min.shape));\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function minimum(args?: LayerArgs): Layer {\n  return new Minimum(args);\n}\n\n/**\n * Layer that multiplies (element-wise) an `Array` of inputs.\n *\n * It takes as input an Array of tensors, all of the same\n * shape, and returns a single tensor (also of the same shape).\n * For example:\n *\n * ```js\n * const input1 = tf.input({shape: [2, 2]});\n * const input2 = tf.input({shape: [2, 2]});\n * const input3 = tf.input({shape: [2, 2]});\n * const multiplyLayer = tf.layers.multiply();\n * const product = multiplyLayer.apply([input1, input2, input3]);\n * console.log(product.shape);\n * // You get [null, 2, 2], with the first dimension as the undetermined batch\n * // dimension.\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function multiply(args?: LayerArgs): Layer {\n  return new Multiply(args);\n}\n\n/**\n * Layer that computes a dot product between samples in two tensors.\n *\n * E.g., if applied to a list of two tensors `a` and `b` both of shape\n * `[batchSize, n]`, the output will be a tensor of shape `[batchSize, 1]`,\n * where each entry at index `[i, 0]` will be the dot product between\n * `a[i, :]` and `b[i, :]`.\n *\n * Example:\n *\n * ```js\n * const dotLayer = tf.layers.dot({axes: -1});\n * const x1 = tf.tensor2d([[10, 20], [30, 40]]);\n * const x2 = tf.tensor2d([[-1, -2], [-3, -4]]);\n *\n * // Invoke the layer's apply() method in eager (imperative) mode.\n * const y = dotLayer.apply([x1, x2]);\n * y.print();\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Merge', namespace: 'layers'}\n */\nexport function dot(args: DotLayerArgs): Layer {\n  return new Dot(args);\n}\n\n// Normalization Layers.\n\n/**\n * Batch normalization layer (Ioffe and Szegedy, 2014).\n *\n * Normalize the activations of the previous layer at each batch,\n * i.e. applies a transformation that maintains the mean activation\n * close to 0 and the activation standard deviation close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape` (Array of integers, does\n *   not include the sample axis) when calling the constructor of this class,\n *   if this layer is used as a first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Batch Normalization: Accelerating Deep Network Training by Reducing\n * Internal Covariate Shift](https://arxiv.org/abs/1502.03167)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function batchNormalization(args?: BatchNormalizationLayerArgs): Layer {\n  return new BatchNormalization(args);\n}\n\n/**\n * Layer-normalization layer (Ba et al., 2016).\n *\n * Normalizes the activations of the previous layer for each given example in a\n * batch independently, instead of across a batch like in `batchNormalization`.\n * In other words, this layer applies a transformation that maintanis the mean\n * activation within each example close to0 and activation variance close to 1.\n *\n * Input shape:\n *   Arbitrary. Use the argument `inputShape` when using this layer as the first\n *   layer in a model.\n *\n * Output shape:\n *   Same as input.\n *\n * References:\n *   - [Layer Normalization](https://arxiv.org/abs/1607.06450)\n *\n * @doc {heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}\n */\nexport function layerNormalization(args?: LayerNormalizationLayerArgs): Layer {\n  return new LayerNormalization(args);\n}\n\n// Padding Layers.\n\n/**\n * Zero-padding layer for 2D input (e.g., image).\n *\n * This layer can add rows and columns of zeros\n * at the top, bottom, left and right side of an image tensor.\n *\n * Input shape:\n *   4D tensor with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, rows, cols, channels]`\n *   - If `data_format` is `\"channels_first\"`:\n *     `[batch, channels, rows, cols]`.\n *\n * Output shape:\n *   4D with shape:\n *   - If `dataFormat` is `\"channelsLast\"`:\n *     `[batch, paddedRows, paddedCols, channels]`\n *    - If `dataFormat` is `\"channelsFirst\"`:\n *     `[batch, channels, paddedRows, paddedCols]`.\n *\n * @doc {heading: 'Layers', subheading: 'Padding', namespace: 'layers'}\n */\nexport function zeroPadding2d(args?: ZeroPadding2DLayerArgs): Layer {\n  return new ZeroPadding2D(args);\n}\n\n// Pooling Layers.\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape: `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * `tf.avgPool1d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling1d(args: Pooling1DLayerArgs): Layer {\n  return new AveragePooling1D(args);\n}\nexport function avgPool1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling1d(args: Pooling1DLayerArgs): Layer {\n  return averagePooling1d(args);\n}\n\n/**\n * Average pooling operation for spatial data.\n *\n * Input shape:\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, rows, cols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *  - If `dataFormat === CHANNEL_LAST`:\n *      4D tensor with shape:\n *      `[batchSize, pooleRows, pooledCols, channels]`\n *  - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *      `[batchSize, channels, pooleRows, pooledCols]`\n *\n * `tf.avgPool2d` is an alias.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling2d(args: Pooling2DLayerArgs): Layer {\n  return new AveragePooling2D(args);\n}\nexport function avgPool2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling2d(args: Pooling2DLayerArgs): Layer {\n  return averagePooling2d(args);\n}\n\n/**\n * Average pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      4D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function averagePooling3d(args: Pooling3DLayerArgs): Layer {\n  return new AveragePooling3D(args);\n}\nexport function avgPool3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n// For backwards compatibility.\n// See https://github.com/tensorflow/tfjs/issues/152\nexport function avgPooling3d(args: Pooling3DLayerArgs): Layer {\n  return averagePooling3d(args);\n}\n\n/**\n * Global average pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling1d(args?: LayerArgs): Layer {\n  return new GlobalAveragePooling1D(args);\n}\n\n/**\n * Global average pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalAveragePooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalAveragePooling2D(args);\n}\n\n/**\n * Global max pooling operation for temporal data.\n *\n * Input Shape: 3D tensor with shape: `[batchSize, steps, features]`.\n *\n * Output Shape:2D tensor with shape: `[batchSize, features]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling1d(args?: LayerArgs): Layer {\n  return new GlobalMaxPooling1D(args);\n}\n\n/**\n * Global max pooling operation for spatial data.\n *\n * Input shape:\n *   - If `dataFormat` is `CHANNEL_LAST`:\n *       4D tensor with shape: `[batchSize, rows, cols, channels]`.\n *   - If `dataFormat` is `CHANNEL_FIRST`:\n *       4D tensor with shape: `[batchSize, channels, rows, cols]`.\n *\n * Output shape:\n *   2D tensor with shape: `[batchSize, channels]`.\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function globalMaxPooling2d(args: GlobalPooling2DLayerArgs): Layer {\n  return new GlobalMaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for temporal data.\n *\n * Input shape:  `[batchSize, inLength, channels]`\n *\n * Output shape: `[batchSize, pooledLength, channels]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling1d(args: Pooling1DLayerArgs): Layer {\n  return new MaxPooling1D(args);\n}\n\n/**\n * Max pooling operation for spatial data.\n *\n * Input shape\n *   - If `dataFormat === CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, rows, cols, channels]`\n *   - If `dataFormat === CHANNEL_FIRST`:\n *      4D tensor with shape:\n *       `[batchSize, channels, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=CHANNEL_LAST`:\n *       4D tensor with shape:\n *       `[batchSize, pooleRows, pooledCols, channels]`\n *   - If `dataFormat=CHANNEL_FIRST`:\n *       4D tensor with shape:\n *       `[batchSize, channels, pooleRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling2d(args: Pooling2DLayerArgs): Layer {\n  return new MaxPooling2D(args);\n}\n\n/**\n * Max pooling operation for 3D data.\n *\n * Input shape\n *   - If `dataFormat === channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, depths, rows, cols, channels]`\n *   - If `dataFormat === channelsFirst`:\n *      5D tensor with shape:\n *       `[batchSize, channels, depths, rows, cols]`\n *\n * Output shape\n *   - If `dataFormat=channelsLast`:\n *       5D tensor with shape:\n *       `[batchSize, pooledDepths, pooledRows, pooledCols, channels]`\n *   - If `dataFormat=channelsFirst`:\n *       5D tensor with shape:\n *       `[batchSize, channels, pooledDepths, pooledRows, pooledCols]`\n *\n * @doc {heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}\n */\nexport function maxPooling3d(args: Pooling3DLayerArgs): Layer {\n  return new MaxPooling3D(args);\n}\n\n// Recurrent Layers.\n\n/**\n * Gated Recurrent Unit - Cho et al. 2014.\n *\n * This is an `RNN` layer consisting of one `GRUCell`. However, unlike\n * the underlying `GRUCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.gru({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `GRUCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gru(args: GRULayerArgs): Layer {\n  return new GRU(args);\n}\n\n/**\n * Cell class for `GRU`.\n *\n * `GRUCell` is distinct from the `RNN` subclass `GRU` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `GRU` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.gruCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `GRUCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.gruCell({units: 4}),\n *   tf.layers.gruCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `gruCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `GRUCell`, use the\n * `tf.layers.gru`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function gruCell(args: GRUCellLayerArgs): RNNCell {\n  return new GRUCell(args);\n}\n\n/**\n * Long-Short Term Memory layer - Hochreiter 1997.\n *\n * This is an `RNN` layer consisting of one `LSTMCell`. However, unlike\n * the underlying `LSTMCell`, the `apply` method of `LSTM` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const lstm = tf.layers.lstm({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = lstm.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `LSTMCell`'s number of units.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstm(args: LSTMLayerArgs): Layer {\n  return new LSTM(args);\n}\n\n/**\n * Cell class for `LSTM`.\n *\n * `LSTMCell` is distinct from the `RNN` subclass `LSTM` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `LSTM` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.lstmCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `LSTMCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.lstmCell({units: 4}),\n *   tf.layers.lstmCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `lstmCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `LSTMCell`, use the\n * `tf.layers.lstm`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function lstmCell(args: LSTMCellLayerArgs): RNNCell {\n  return new LSTMCell(args);\n}\n\n/**\n * Fully-connected RNN where the output is to be fed back to input.\n *\n * This is an `RNN` layer consisting of one `SimpleRNNCell`. However, unlike\n * the underlying `SimpleRNNCell`, the `apply` method of `SimpleRNN` operates\n * on a sequence of inputs. The shape of the input (not including the first,\n * batch dimension) needs to be at least 2-D, with the first dimension being\n * time steps. For example:\n *\n * ```js\n * const rnn = tf.layers.simpleRNN({units: 8, returnSequences: true});\n *\n * // Create an input with 10 time steps.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the `SimpleRNNCell`'s number of units.\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNN(args: SimpleRNNLayerArgs): Layer {\n  return new SimpleRNN(args);\n}\n\n/**\n * Cell class for `SimpleRNN`.\n *\n * `SimpleRNNCell` is distinct from the `RNN` subclass `SimpleRNN` in that its\n * `apply` method takes the input data of only a single time step and returns\n * the cell's output at the time step, while `SimpleRNN` takes the input data\n * over a number of time steps. For example:\n *\n * ```js\n * const cell = tf.layers.simpleRNNCell({units: 2});\n * const input = tf.input({shape: [10]});\n * const output = cell.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10]: This is the cell's output at a single time step. The 1st\n * // dimension is the unknown batch size.\n * ```\n *\n * Instance(s) of `SimpleRNNCell` can be used to construct `RNN` layers. The\n * most typical use of this workflow is to combine a number of cells into a\n * stacked RNN cell (i.e., `StackedRNNCell` internally) and use it to create an\n * RNN. For example:\n *\n * ```js\n * const cells = [\n *   tf.layers.simpleRNNCell({units: 4}),\n *   tf.layers.simpleRNNCell({units: 8}),\n * ];\n * const rnn = tf.layers.rnn({cell: cells, returnSequences: true});\n *\n * // Create an input with 10 time steps and a length-20 vector at each step.\n * const input = tf.input({shape: [10, 20]});\n * const output = rnn.apply(input);\n *\n * console.log(JSON.stringify(output.shape));\n * // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the\n * // same as the sequence length of `input`, due to `returnSequences`: `true`;\n * // 3rd dimension is the last `SimpleRNNCell`'s number of units.\n * ```\n *\n * To create an `RNN` consisting of only *one* `SimpleRNNCell`, use the\n * `tf.layers.simpleRNN`.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function simpleRNNCell(args: SimpleRNNCellLayerArgs): RNNCell {\n  return new SimpleRNNCell(args);\n}\n\n/**\n * Convolutional LSTM layer - Xingjian Shi 2015.\n *\n * This is an `ConvRNN2D` layer consisting of one `ConvLSTM2DCell`. However,\n * unlike the underlying `ConvLSTM2DCell`, the `apply` method of `ConvLSTM2D`\n * operates on a sequence of inputs. The shape of the input (not including the\n * first, batch dimension) needs to be 4-D, with the first dimension being time\n * steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const batchSize = 4;\n * const sequenceLength = 2;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [batchSize, sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const layer = tf.layers.convLstm2d({filters, kernelSize});\n *\n * const output = layer.apply(input);\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2d(args: ConvLSTM2DArgs): ConvLSTM2D {\n  return new ConvLSTM2D(args);\n}\n\n/**\n * Cell class for `ConvLSTM2D`.\n *\n * `ConvLSTM2DCell` is distinct from the `ConvRNN2D` subclass `ConvLSTM2D` in\n * that its `call` method takes the input data of only a single time step and\n * returns the cell's output at the time step, while `ConvLSTM2D` takes the\n * input data over a number of time steps. For example:\n *\n * ```js\n * const filters = 3;\n * const kernelSize = 3;\n *\n * const sequenceLength = 1;\n * const size = 5;\n * const channels = 3;\n *\n * const inputShape = [sequenceLength, size, size, channels];\n * const input = tf.ones(inputShape);\n *\n * const cell = tf.layers.convLstm2dCell({filters, kernelSize});\n *\n * cell.build(input.shape);\n *\n * const outputSize = size - kernelSize + 1;\n * const outShape = [sequenceLength, outputSize, outputSize, filters];\n *\n * const initialH = tf.zeros(outShape);\n * const initialC = tf.zeros(outShape);\n *\n * const [o, h, c] = cell.call([input, initialH, initialC], {});\n * ```\n */\n/** @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'} */\nexport function convLstm2dCell(args: ConvLSTM2DCellArgs): ConvLSTM2DCell {\n  return new ConvLSTM2DCell(args);\n}\n\n/**\n * Base class for recurrent layers.\n *\n * Input shape:\n *   3D tensor with shape `[batchSize, timeSteps, inputDim]`.\n *\n * Output shape:\n *   - if `returnState`, an Array of tensors (i.e., `tf.Tensor`s). The first\n *     tensor is the output. The remaining tensors are the states at the\n *     last time step, each with shape `[batchSize, units]`.\n *   - if `returnSequences`, the output will have shape\n *     `[batchSize, timeSteps, units]`.\n *   - else, the output will have shape `[batchSize, units]`.\n *\n * Masking:\n *   This layer supports masking for input data with a variable number\n *   of timesteps. To introduce masks to your data,\n *   use an embedding layer with the `mask_zero` parameter\n *   set to `True`.\n *\n * Notes on using statefulness in RNNs:\n *   You can set RNN layers to be 'stateful', which means that the states\n *   computed for the samples in one batch will be reused as initial states\n *   for the samples in the next batch. This assumes a one-to-one mapping\n *   between samples in different successive batches.\n *\n *   To enable statefulness:\n *     - specify `stateful: true` in the layer constructor.\n *     - specify a fixed batch size for your model, by passing\n *       if sequential model:\n *         `batchInputShape=[...]` to the first layer in your model.\n *       else for functional model with 1 or more Input layers:\n *         `batchShape=[...]` to all the first layers in your model.\n *       This is the expected shape of your inputs *including the batch size*.\n *       It should be a tuple of integers, e.g. `(32, 10, 100)`.\n *     - specify `shuffle=False` when calling fit().\n *\n *   To reset the states of your model, call `.resetStates()` on either\n *   a specific layer, or on your entire model.\n *\n * Note on specifying the initial state of RNNs\n *   You can specify the initial state of RNN layers symbolically by\n *   calling them with the option `initialState`. The value of\n *   `initialState` should be a tensor or list of tensors representing\n *   the initial state of the RNN layer.\n *\n *   You can specify the initial state of RNN layers numerically by\n *   calling `resetStates` with the keyword argument `states`. The value of\n *   `states` should be a numpy array or list of numpy arrays representing\n *   the initial state of the RNN layer.\n *\n * Note on passing external constants to RNNs\n *   You can pass \"external\" constants to the cell using the `constants`\n *   keyword argument of `RNN.call` method. This requires that the `cell.call`\n *   method accepts the same keyword argument `constants`. Such constants\n *   can be used to conditon the cell transformation on additional static inputs\n *   (not changing over time), a.k.a an attention mechanism.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function rnn(args: RNNLayerArgs): Layer {\n  return new RNN(args);\n}\n\n/**\n * Wrapper allowing a stack of RNN cells to behave as a single cell.\n *\n * Used to implement efficient stacked RNNs.\n *\n * @doc {heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}\n */\nexport function stackedRNNCells(args: StackedRNNCellsArgs): RNNCell {\n  return new StackedRNNCells(args);\n}\n\n// Wrapper Layers.\n\n/** @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'} */\nexport function bidirectional(args: BidirectionalLayerArgs): Bidirectional {\n  return new Bidirectional(args);\n}\n\n/**\n * This wrapper applies a layer to every temporal slice of an input.\n *\n * The input should be at least 3D,  and the dimension of the index `1` will be\n * considered to be the temporal dimension.\n *\n * Consider a batch of 32 samples, where each sample is a sequence of 10 vectors\n * of 16 dimensions. The batch input shape of the layer is then `[32,  10,\n * 16]`, and the `inputShape`, not including the sample dimension, is\n * `[10, 16]`.\n *\n * You can then use `TimeDistributed` to apply a `Dense` layer to each of the 10\n * timesteps, independently:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.dense({units: 8}),\n *   inputShape: [10, 16],\n * }));\n *\n * // Now model.outputShape = [null, 10, 8].\n * // The output will then have shape `[32, 10, 8]`.\n *\n * // In subsequent layers, there is no need for `inputShape`:\n * model.add(tf.layers.timeDistributed({layer: tf.layers.dense({units: 32})}));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * // Now model.outputShape = [null, 10, 32].\n * ```\n *\n * The output will then have shape `[32, 10, 32]`.\n *\n * `TimeDistributed` can be used with arbitrary layers, not just `Dense`, for\n * instance a `Conv2D` layer.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.timeDistributed({\n *   layer: tf.layers.conv2d({filters: 64, kernelSize: [3, 3]}),\n *   inputShape: [10, 299, 299, 3],\n * }));\n * console.log(JSON.stringify(model.outputs[0].shape));\n * ```\n *\n * @doc {heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}\n */\nexport function timeDistributed(args: WrapperLayerArgs): Layer {\n  return new TimeDistributed(args);\n}\n\n// Aliases for pooling.\nexport const globalMaxPool1d = globalMaxPooling1d;\nexport const globalMaxPool2d = globalMaxPooling2d;\nexport const maxPool1d = maxPooling1d;\nexport const maxPool2d = maxPooling2d;\n\nexport {Layer, RNN, RNNCell, input /* alias for tf.input */};\n\n/**\n * Apply additive zero-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * This is useful to mitigate overfitting\n * (you could see it as a form of random data augmentation).\n * Gaussian Noise (GS) is a natural choice as corruption process\n * for real valued inputs.\n *\n * # Arguments\n *     stddev: float, standard deviation of the noise distribution.\n *\n * # Input shape\n *         Arbitrary. Use the keyword argument `input_shape`\n *         (tuple of integers, does not include the samples axis)\n *         when using this layer as the first layer in a model.\n *\n * # Output shape\n *         Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianNoise(args: GaussianNoiseArgs): GaussianNoise {\n  return new GaussianNoise(args);\n}\n\n/**\n * Apply multiplicative 1-centered Gaussian noise.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting](\n *      http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function gaussianDropout(args: GaussianDropoutArgs): GaussianDropout {\n  return new GaussianDropout(args);\n}\n\n/**\n * Applies Alpha Dropout to the input.\n *\n * As it is a regularization layer, it is only active at training time.\n *\n * Alpha Dropout is a `Dropout` that keeps mean and variance of inputs\n * to their original values, in order to ensure the self-normalizing property\n * even after this dropout.\n * Alpha Dropout fits well to Scaled Exponential Linear Units\n * by randomly setting activations to the negative saturation value.\n *\n * Arguments:\n *   - `rate`: float, drop probability (as with `Dropout`).\n *     The multiplicative noise will have\n *     standard deviation `sqrt(rate / (1 - rate))`.\n *   - `noise_shape`: A 1-D `Tensor` of type `int32`, representing the\n *     shape for randomly generated keep/drop flags.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * References:\n *   - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)\n *\n * @doc {heading: 'Layers', subheading: 'Noise', namespace: 'layers'}\n */\nexport function alphaDropout(args: AlphaDropoutArgs): AlphaDropout {\n  return new AlphaDropout(args);\n}\n\n/**\n * Masks a sequence by using a mask value to skip timesteps.\n *\n * If all features for a given sample timestep are equal to `mask_value`,\n * then the sample timestep will be masked (skipped) in all downstream layers\n * (as long as they support masking).\n *\n * If any downstream layer does not support masking yet receives such\n * an input mask, an exception will be raised.\n *\n * Arguments:\n *   - `maskValue`: Either None or mask value to skip.\n *\n * Input shape:\n *   Arbitrary. Use the keyword argument `inputShape`\n *   (tuple of integers, does not include the samples axis)\n *   when using this layer as the first layer in a model.\n *\n * Output shape:\n *   Same shape as input.\n *\n * @doc {heading: 'Layers', subheading: 'Mask', namespace: 'layers'}\n */\nexport function masking(args?: MaskingArgs): Layer {\n  return new Masking(args);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport * as losses from './losses';\nimport * as metrics from './metrics';\n\n/**\n * Binary accuracy metric function.\n *\n * `yTrue` and `yPred` can have 0-1 values. Example:\n * ```js\n * const x = tf.tensor2d([[1, 1, 1, 1], [0, 0, 0, 0]], [2, 4]);\n * const y = tf.tensor2d([[1, 0, 1, 0], [0, 0, 0, 1]], [2, 4]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * `yTrue` and `yPred` can also have floating-number values between 0 and 1, in\n * which case the values will be thresholded at 0.5 to yield 0-1 values (i.e.,\n * a value >= 0.5 and <= 1.0 is interpreted as 1.\n * )\n * Example:\n * ```js\n * const x = tf.tensor1d([1, 1, 1, 1, 0, 0, 0, 0]);\n * const y = tf.tensor1d([0.2, 0.4, 0.6, 0.8, 0.2, 0.3, 0.4, 0.7]);\n * const accuracy = tf.metrics.binaryAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryAccuracy(yTrue, yPred);\n}\n\n/**\n * Binary crossentropy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0], [1], [1], [1]]);\n * const y = tf.tensor2d([[0], [0], [0.5], [1]]);\n * const crossentropy = tf.metrics.binaryCrossentropy(x, y);\n * crossentropy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth.\n * @param yPred Binary Tensor of prediction, probabilities for the `1` case.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function binaryCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.binaryCrossentropy(yTrue, yPred);\n}\n\n/**\n * Sparse categorical accuracy metric function.\n *\n * Example:\n * ```js\n *\n * const yTrue = tf.tensor1d([1, 1, 2, 2, 0]);\n * const yPred = tf.tensor2d(\n *      [[0, 1, 0], [1, 0, 0], [0, 0.4, 0.6], [0, 0.6, 0.4], [0.7, 0.3, 0]]);\n * const crossentropy = tf.metrics.sparseCategoricalAccuracy(yTrue, yPred);\n * crossentropy.print();\n * ```\n *\n * @param yTrue True labels: indices.\n * @param yPred Predicted probabilities or logits.\n * @returns Accuracy tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function sparseCategoricalAccuracy(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.sparseCategoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical accuracy metric function.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d([[0, 0, 0, 1], [0, 0, 0, 1]]);\n * const y = tf.tensor2d([[0.1, 0.8, 0.05, 0.05], [0.1, 0.05, 0.05, 0.8]]);\n * const accuracy = tf.metrics.categoricalAccuracy(x, y);\n * accuracy.print();\n * ```\n *\n * @param yTrue Binary Tensor of truth: one-hot encoding of categories.\n * @param yPred Binary Tensor of prediction: probabilities or logits for the\n *   same categories as in `yTrue`.\n * @return Accuracy Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalAccuracy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalAccuracy(yTrue, yPred);\n}\n\n/**\n * Categorical crossentropy between an output tensor and a target tensor.\n *\n * @param target A tensor of the same shape as `output`.\n * @param output A tensor resulting from a softmax (unless `fromLogits` is\n *  `true`, in which case `output` is expected to be the logits).\n * @param fromLogits Boolean, whether `output` is the result of a softmax, or is\n *   a tensor of logits.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function categoricalCrossentropy(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.categoricalCrossentropy(yTrue, yPred);\n}\n\n/**\n * Computes the precision of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const precision = tf.metrics.precision(x, y);\n * precision.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Precision Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function precision(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.precision(yTrue, yPred);\n}\n\n/**\n * Computes the recall of the predictions with respect to the labels.\n *\n * Example:\n * ```js\n * const x = tf.tensor2d(\n *    [\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [1, 0, 0, 0],\n *      [0, 0, 1, 0]\n *    ]\n * );\n *\n * const y = tf.tensor2d(\n *    [\n *      [0, 0, 1, 0],\n *      [0, 1, 0, 0],\n *      [0, 0, 0, 1],\n *      [0, 1, 0, 0],\n *      [0, 1, 0, 0]\n *    ]\n * );\n *\n * const recall = tf.metrics.recall(x, y);\n * recall.print();\n * ```\n *\n * @param yTrue The ground truth values. Expected to be contain only 0-1 values.\n * @param yPred The predicted values. Expected to be contain only 0-1 values.\n * @return Recall Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function recall(yTrue: Tensor, yPred: Tensor): Tensor {\n  return metrics.recall(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Cosine proximity.\n *\n * Mathematically, cosine proximity is defined as:\n *   `-sum(l2Normalize(yTrue) * l2Normalize(yPred))`,\n * wherein `l2Normalize()` normalizes the L2 norm of the input to 1 and `*`\n * represents element-wise multiplication.\n *\n * ```js\n * const yTrue = tf.tensor2d([[1, 0], [1, 0]]);\n * const yPred = tf.tensor2d([[1 / Math.sqrt(2), 1 / Math.sqrt(2)], [0, 1]]);\n * const proximity = tf.metrics.cosineProximity(yTrue, yPred);\n * proximity.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Cosine proximity Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function cosineProximity(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.cosineProximity(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute error.\n *\n * Mathematically, mean absolute error is defined as:\n *   `mean(abs(yPred - yTrue))`,\n * wherein the `mean` is applied over feature dimensions.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [0, 0], [2, 3]]);\n * const yPred = tf.tensor2d([[0, 1], [0, 1], [-2, -3]]);\n * const mse = tf.metrics.meanAbsoluteError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsoluteError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsoluteError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean absolute percentage error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [10, 20]]);\n * const yPred = tf.tensor2d([[0, 1], [11, 24]]);\n * const mse = tf.metrics.meanAbsolutePercentageError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MAPE`, `tf.metrics.mape`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean absolute percentage error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanAbsolutePercentageError(\n    yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function MAPE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\nexport function mape(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanAbsolutePercentageError(yTrue, yPred);\n}\n\n/**\n * Loss or metric function: Mean squared error.\n *\n * ```js\n * const yTrue = tf.tensor2d([[0, 1], [3, 4]]);\n * const yPred = tf.tensor2d([[0, 1], [-3, -4]]);\n * const mse = tf.metrics.meanSquaredError(yTrue, yPred);\n * mse.print();\n * ```\n *\n * Aliases: `tf.metrics.MSE`, `tf.metrics.mse`.\n *\n * @param yTrue Truth Tensor.\n * @param yPred Prediction Tensor.\n * @return Mean squared error Tensor.\n *\n * @doc {heading: 'Metrics', namespace: 'metrics'}\n */\nexport function meanSquaredError(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function MSE(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n\nexport function mse(yTrue: Tensor, yPred: Tensor): Tensor {\n  return losses.meanSquaredError(yTrue, yPred);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\nexport {modelFromJSON} from './models';\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\nimport * as regularizers from './regularizers';\n// tslint:disable-next-line:max-line-length\nimport {L1Args, L1L2, L1L2Args, L2Args, Regularizer} from './regularizers';\n\n/**\n * Regularizer for L1 and L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x)) + sum(l2 * x^2)\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1l2(config?: L1L2Args): Regularizer {\n  return new L1L2(config);\n}\n\n/**\n * Regularizer for L1 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l1 * abs(x))\n * @param args l1 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l1(config?: L1Args): Regularizer {\n  return regularizers.l1(config);\n}\n\n/**\n * Regularizer for L2 regularization.\n *\n * Adds a term to the loss to penalize large weights:\n * loss += sum(l2 * x^2)\n * @param args l2 config.\n *\n * @doc {heading: 'Regularizers', namespace: 'regularizers'}\n */\nexport function l2(config?: L2Args): Regularizer {\n  return regularizers.l2(config);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC\n *\n * Use of this source code is governed by an MIT-style\n * license that can be found in the LICENSE file or at\n * https://opensource.org/licenses/MIT.\n * =============================================================================\n */\n\n/* Original source: keras/callbacks.py */\n\nimport {BaseCallback} from './base_callbacks';\nimport {Container} from './engine/container';\nimport {LayersModel} from './engine/training';\nimport {NotImplementedError} from './errors';\nimport {Logs, resolveScalarsInLogs} from './logs';\n\nexport abstract class Callback extends BaseCallback {\n  /** Instance of `keras.models.Model`. Reference of the model being trained. */\n  model: LayersModel = null;\n\n  setModel(model: Container): void {\n    if (!(model instanceof LayersModel)) {\n      throw new Error('model must be a LayersModel, not some other Container');\n    }\n    this.model = model;\n  }\n}\n\nexport interface EarlyStoppingCallbackArgs {\n  /**\n   * Quantity to be monitored.\n   *\n   * Defaults to 'val_loss'.\n   */\n  monitor?: string;\n\n  /**\n   * Minimum change in the monitored quantity to qualify as improvement,\n   * i.e., an absolute change of less than `minDelta` will count as no\n   * improvement.\n   *\n   * Defaults to 0.\n   */\n  minDelta?: number;\n\n  /**\n   * Number of epochs with no improvement after which training will be stopped.\n   *\n   * Defaults to 0.\n   */\n  patience?: number;\n\n  /** Verbosity mode. */\n  verbose?: number;\n\n  /**\n   * Mode: one of 'min', 'max', and 'auto'.\n   * - In 'min' mode, training will be stopped when the quantity monitored has\n   *   stopped decreasing.\n   * - In 'max' mode, training will be stopped when the quantity monitored has\n   *   stopped increasing.\n   * - In 'auto' mode, the direction is inferred automatically from the name of\n   *   the monitored quantity.\n   *\n   * Defaults to 'auto'.\n   */\n  mode?: 'auto'|'min'|'max';\n\n  /**\n   * Baseline value of the monitored quantity.\n   *\n   * If specified, training will be stopped if the model doesn't show\n   * improvement over the baseline.\n   */\n  baseline?: number;\n\n  /**\n   * Whether to restore model weights from the epoch with the best value\n   * of the monitored quantity. If `False`, the model weights obtained at the\n   * at the last step of training are used.\n   *\n   * **`True` is not supported yet.**\n   */\n  restoreBestWeights?: boolean;\n}\n\nfunction less(currVal: number, prevVal: number) {\n  return currVal < prevVal;\n}\n\nfunction greater(currVal: number, prevVal: number) {\n  return currVal > prevVal;\n}\n\n/**\n * A Callback that stops training when a monitored quantity has stopped\n * improving.\n */\nexport class EarlyStopping extends Callback {\n  protected readonly monitor: string;\n  protected readonly minDelta: number;\n  protected readonly patience: number;\n  protected readonly baseline: number;\n  protected readonly verbose: number;\n  protected readonly mode: 'auto'|'min'|'max';\n\n  protected monitorFunc: (currVal: number, prevVal: number) => boolean;\n\n  private wait: number;\n  private stoppedEpoch: number;\n  private best: number;\n\n  constructor(args?: EarlyStoppingCallbackArgs) {\n    super();\n    if (args == null) {\n      args = {};\n    }\n    if (args.restoreBestWeights) {\n      throw new NotImplementedError(\n          'restoreBestWeights = True is not implemented in EarlyStopping yet.');\n    }\n\n    this.monitor = args.monitor || 'val_loss';\n    this.minDelta = Math.abs(args.minDelta || 0);\n    this.patience = args.patience || 0;\n    this.verbose = args.verbose || 0;\n    this.mode = args.mode || 'auto';\n    this.baseline = args.baseline;\n\n    if (['auto', 'min', 'max'].indexOf(this.mode) === -1) {\n      console.warn(\n          `EarlyStopping mode '${this.mode}' is invalid. ` +\n          `Falling back to mode 'auto'.`);\n      this.mode = 'auto';\n    }\n\n    if (this.mode === 'min') {\n      this.monitorFunc = less;\n    } else if (this.mode === 'max') {\n      this.monitorFunc = greater;\n    } else {\n      // For mode === 'auto'.\n      if (this.monitor.indexOf('acc') !== -1) {\n        this.monitorFunc = greater;\n      } else {\n        this.monitorFunc = less;\n      }\n    }\n\n    if (this.monitorFunc === less) {\n      this.minDelta *= -1;\n    }\n  }\n\n  async onTrainBegin(logs?: Logs) {\n    this.wait = 0;\n    this.stoppedEpoch = 0;\n    if (this.baseline != null) {\n      this.best = this.baseline;\n    } else {\n      this.best = this.monitorFunc === less ? Infinity : -Infinity;\n    }\n  }\n\n  async onEpochEnd(epoch: number, logs?: Logs) {\n    await resolveScalarsInLogs(logs);\n    const current = this.getMonitorValue(logs);\n    if (current == null) {\n      return;\n    }\n\n    if (this.monitorFunc(current - this.minDelta, this.best)) {\n      this.best = current;\n      this.wait = 0;\n      // TODO(cais): Logic for restoreBestWeights.\n    } else {\n      this.wait++;\n      if (this.wait >= this.patience) {\n        this.stoppedEpoch = epoch;\n        this.model.stopTraining = true;\n      }\n      // TODO(cais): Logic for restoreBestWeights.\n    }\n  }\n\n  async onTrainEnd(logs?: Logs) {\n    if (this.stoppedEpoch > 0 && this.verbose) {\n      console.log(`Epoch ${this.stoppedEpoch}: early stopping.`);\n    }\n  }\n\n  private getMonitorValue(logs: Logs) {\n    if (logs == null) {\n      logs = {};\n    }\n    const monitorValue = logs[this.monitor];\n    if (monitorValue == null) {\n      console.warn(\n          `Metric for EarlyStopping ${this.monitor} is not available. ` +\n          `Available metrics are: ${Object.keys(logs)}`);\n    }\n    return monitorValue;\n  }\n}\n\n/**\n * Factory function for a Callback that stops training when a monitored\n * quantity has stopped improving.\n *\n * Early stopping is a type of regularization, and protects model against\n * overfitting.\n *\n * The following example based on fake data illustrates how this callback\n * can be used during `tf.LayersModel.fit()`:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense({\n *   units: 3,\n *   activation: 'softmax',\n *   kernelInitializer: 'ones',\n *   inputShape: [2]\n * }));\n * const xs = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n * const ys = tf.tensor2d([[1, 0, 0], [0, 1, 0]], [2, 3]);\n * const xsVal = tf.tensor2d([4, 3, 2, 1], [2, 2]);\n * const ysVal = tf.tensor2d([[0, 0, 1], [0, 1, 0]], [2, 3]);\n * model.compile(\n *     {loss: 'categoricalCrossentropy', optimizer: 'sgd', metrics: ['acc']});\n *\n * // Without the EarlyStopping callback, the val_acc value would be:\n * //   0.5, 0.5, 0.5, 0.5, ...\n * // With val_acc being monitored, training should stop after the 2nd epoch.\n * const history = await model.fit(xs, ys, {\n *   epochs: 10,\n *   validationData: [xsVal, ysVal],\n *   callbacks: tf.callbacks.earlyStopping({monitor: 'val_acc'})\n * });\n *\n * // Expect to see a length-2 array.\n * console.log(history.history.val_acc);\n * ```\n *\n * @doc {\n *   heading: 'Callbacks',\n *   namespace: 'callbacks'\n * }\n */\nexport function earlyStopping(args?: EarlyStoppingCallbackArgs) {\n  return new EarlyStopping(args);\n}\n\nexport const callbacks = {earlyStopping};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/* tslint:disable */\n\n/** Properties of an Any. */\nexport declare interface IAny {\n  /** Any typeUrl */\n  typeUrl?: (string|null);\n\n  /** Any value */\n  value?: (Uint8Array|null);\n}\n\n/** DataType enum. */\nexport enum DataType {\n  'DT_INVALID' = 0,\n  'DT_FLOAT' = 1,\n  'DT_DOUBLE' = 2,\n  'DT_INT32' = 3,\n  'DT_UINT8' = 4,\n  'DT_INT16' = 5,\n  'DT_INT8' = 6,\n  'DT_STRING' = 7,\n  'DT_COMPLEX64' = 8,\n  'DT_INT64' = 9,\n  'DT_BOOL' = 10,\n  'DT_QINT8' = 11,\n  'DT_QUINT8' = 12,\n  'DT_QINT32' = 13,\n  'DT_BFLOAT16' = 14,\n  'DT_FLOAT_REF' = 101,\n  'DT_DOUBLE_REF' = 102,\n  'DT_INT32_REF' = 103,\n  'DT_UINT8_REF' = 104,\n  'DT_INT16_REF' = 105,\n  'DT_INT8_REF' = 106,\n  'DT_STRING_REF' = 107,\n  'DT_COMPLEX64_REF' = 108,\n  'DT_INT64_REF' = 109,\n  'DT_BOOL_REF' = 110,\n  'DT_QINT8_REF' = 111,\n  'DT_QUINT8_REF' = 112,\n  'DT_QINT32_REF' = 113,\n  'DT_BFLOAT16_REF' = 114\n}\n\n/** Properties of a TensorShape. */\nexport declare interface ITensorShape {\n  /** TensorShape dim */\n  dim?: (TensorShape.IDim[]|null);\n\n  /** TensorShape unknownRank */\n  unknownRank?: (boolean|null);\n}\n\nexport namespace TensorShape {\n  /** Properties of a Dim. */\n  export declare interface IDim {\n    /** Dim size */\n    size?: (number|string|null);\n\n    /** Dim name */\n    name?: (string|null);\n  }\n}\n\n/** Properties of a Tensor. */\nexport declare interface ITensor {\n  /** Tensor dtype */\n  dtype?: (DataType|null);\n\n  /** Tensor tensorShape */\n  tensorShape?: (ITensorShape|null);\n\n  /** Tensor versionNumber */\n  versionNumber?: (number|null);\n\n  /** Tensor tensorContent */\n  tensorContent?: (Uint8Array|null);\n\n  /** Tensor floatVal */\n  floatVal?: (number[]|null);\n\n  /** Tensor doubleVal */\n  doubleVal?: (number[]|null);\n\n  /** Tensor intVal */\n  intVal?: (number[]|null);\n\n  /** Tensor stringVal */\n  stringVal?: (Uint8Array[]|null);\n\n  /** Tensor scomplexVal */\n  scomplexVal?: (number[]|null);\n\n  /** Tensor int64Val */\n  int64Val?: ((number | string)[]|null);\n\n  /** Tensor boolVal */\n  boolVal?: (boolean[]|null);\n\n  /** Tensor uint32Val */\n  uint32Val?: (number[]|null);\n\n  /** Tensor uint64Val */\n  uint64Val?: ((number | string)[]|null);\n}\n\n/** Properties of an AttrValue. */\nexport declare interface IAttrValue {\n  /** AttrValue list */\n  list?: (AttrValue.IListValue|null);\n\n  /** AttrValue s */\n  s?: (string|null);\n\n  /** AttrValue i */\n  i?: (number|string|null);\n\n  /** AttrValue f */\n  f?: (number|null);\n\n  /** AttrValue b */\n  b?: (boolean|null);\n\n  /** AttrValue type */\n  type?: (DataType|null);\n\n  /** AttrValue shape */\n  shape?: (ITensorShape|null);\n\n  /** AttrValue tensor */\n  tensor?: (ITensor|null);\n\n  /** AttrValue placeholder */\n  placeholder?: (string|null);\n\n  /** AttrValue func */\n  func?: (INameAttrList|null);\n}\n\nexport namespace AttrValue {\n  /** Properties of a ListValue. */\n  export declare interface IListValue {\n    /** ListValue s */\n    s?: (string[]|null);\n\n    /** ListValue i */\n    i?: ((number | string)[]|null);\n\n    /** ListValue f */\n    f?: (number[]|null);\n\n    /** ListValue b */\n    b?: (boolean[]|null);\n\n    /** ListValue type */\n    type?: (DataType[]|null);\n\n    /** ListValue shape */\n    shape?: (ITensorShape[]|null);\n\n    /** ListValue tensor */\n    tensor?: (ITensor[]|null);\n\n    /** ListValue func */\n    func?: (INameAttrList[]|null);\n  }\n}\n\n/** Properties of a NameAttrList. */\nexport declare interface INameAttrList {\n  /** NameAttrList name */\n  name?: (string|null);\n\n  /** NameAttrList attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a NodeDef. */\nexport declare interface INodeDef {\n  /** NodeDef name */\n  name?: (string|null);\n\n  /** NodeDef op */\n  op?: (string|null);\n\n  /** NodeDef input */\n  input?: (string[]|null);\n\n  /** NodeDef device */\n  device?: (string|null);\n\n  /** NodeDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n}\n\n/** Properties of a VersionDef. */\nexport declare interface IVersionDef {\n  /** VersionDef producer */\n  producer?: (number|null);\n\n  /** VersionDef minConsumer */\n  minConsumer?: (number|null);\n\n  /** VersionDef badConsumers */\n  badConsumers?: (number[]|null);\n}\n\n/** Properties of a GraphDef. */\nexport declare interface IGraphDef {\n  /** GraphDef node */\n  node?: (INodeDef[]|null);\n\n  /** GraphDef versions */\n  versions?: (IVersionDef|null);\n\n  /** GraphDef library */\n  library?: (IFunctionDefLibrary|null);\n}\n\n/** Properties of a CollectionDef. */\nexport declare interface ICollectionDef {\n  /** CollectionDef nodeList */\n  nodeList?: (CollectionDef.INodeList|null);\n\n  /** CollectionDef bytesList */\n  bytesList?: (CollectionDef.IBytesList|null);\n\n  /** CollectionDef int64List */\n  int64List?: (CollectionDef.IInt64List|null);\n\n  /** CollectionDef floatList */\n  floatList?: (CollectionDef.IFloatList|null);\n\n  /** CollectionDef anyList */\n  anyList?: (CollectionDef.IAnyList|null);\n}\n\nexport namespace CollectionDef {\n  /** Properties of a NodeList. */\n  export declare interface INodeList {\n    /** NodeList value */\n    value?: (string[]|null);\n  }\n\n  /** Properties of a BytesList. */\n  export declare interface IBytesList {\n    /** BytesList value */\n    value?: (Uint8Array[]|null);\n  }\n\n  /** Properties of an Int64List. */\n  export declare interface IInt64List {\n    /** Int64List value */\n    value?: ((number | string)[]|null);\n  }\n\n  /** Properties of a FloatList. */\n  export declare interface IFloatList {\n    /** FloatList value */\n    value?: (number[]|null);\n  }\n\n  /** Properties of an AnyList. */\n  export declare interface IAnyList {\n    /** AnyList value */\n    value?: (IAny[]|null);\n  }\n}\n\n/** Properties of a SaverDef. */\nexport declare interface ISaverDef {\n  /** SaverDef filenameTensorName */\n  filenameTensorName?: (string|null);\n\n  /** SaverDef saveTensorName */\n  saveTensorName?: (string|null);\n\n  /** SaverDef restoreOpName */\n  restoreOpName?: (string|null);\n\n  /** SaverDef maxToKeep */\n  maxToKeep?: (number|null);\n\n  /** SaverDef sharded */\n  sharded?: (boolean|null);\n\n  /** SaverDef keepCheckpointEveryNHours */\n  keepCheckpointEveryNHours?: (number|null);\n\n  /** SaverDef version */\n  version?: (SaverDef.CheckpointFormatVersion|null);\n}\n\nexport namespace SaverDef {\n  /** CheckpointFormatVersion enum. */\n  export enum CheckpointFormatVersion {'LEGACY' = 0, 'V1' = 1, 'V2' = 2}\n}\n\n/** Properties of a TensorInfo. */\nexport declare interface ITensorInfo {\n  /** TensorInfo name */\n  name?: (string|null);\n\n  /** TensorInfo cooSparse */\n  cooSparse?: (TensorInfo.ICooSparse|null);\n\n  /** TensorInfo dtype */\n  dtype?: (DataType|null);\n\n  /** TensorInfo tensorShape */\n  tensorShape?: (ITensorShape|null);\n}\n\nexport namespace TensorInfo {\n  /** Properties of a CooSparse. */\n  export declare interface ICooSparse {\n    /** CooSparse valuesTensorName */\n    valuesTensorName?: (string|null);\n\n    /** CooSparse indicesTensorName */\n    indicesTensorName?: (string|null);\n\n    /** CooSparse denseShapeTensorName */\n    denseShapeTensorName?: (string|null);\n  }\n}\n\n/** Properties of a SignatureDef. */\nexport declare interface ISignatureDef {\n  /** SignatureDef inputs */\n  inputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef outputs */\n  outputs?: ({[k: string]: ITensorInfo}|null);\n\n  /** SignatureDef methodName */\n  methodName?: (string|null);\n}\n\n/** Properties of an AssetFileDef. */\nexport declare interface IAssetFileDef {\n  /** AssetFileDef tensorInfo */\n  tensorInfo?: (ITensorInfo|null);\n\n  /** AssetFileDef filename */\n  filename?: (string|null);\n}\n\n/** Properties of an OpDef. */\nexport declare interface IOpDef {\n  /** OpDef name */\n  name?: (string|null);\n\n  /** OpDef inputArg */\n  inputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef outputArg */\n  outputArg?: (OpDef.IArgDef[]|null);\n\n  /** OpDef attr */\n  attr?: (OpDef.IAttrDef[]|null);\n\n  /** OpDef deprecation */\n  deprecation?: (OpDef.IOpDeprecation|null);\n\n  /** OpDef summary */\n  summary?: (string|null);\n\n  /** OpDef description */\n  description?: (string|null);\n\n  /** OpDef isCommutative */\n  isCommutative?: (boolean|null);\n\n  /** OpDef isAggregate */\n  isAggregate?: (boolean|null);\n\n  /** OpDef isStateful */\n  isStateful?: (boolean|null);\n\n  /** OpDef allowsUninitializedInput */\n  allowsUninitializedInput?: (boolean|null);\n}\n\nexport namespace OpDef {\n  /** Properties of an ArgDef. */\n  export declare interface IArgDef {\n    /** ArgDef name */\n    name?: (string|null);\n\n    /** ArgDef description */\n    description?: (string|null);\n\n    /** ArgDef type */\n    type?: (DataType|null);\n\n    /** ArgDef typeAttr */\n    typeAttr?: (string|null);\n\n    /** ArgDef numberAttr */\n    numberAttr?: (string|null);\n\n    /** ArgDef typeListAttr */\n    typeListAttr?: (string|null);\n\n    /** ArgDef isRef */\n    isRef?: (boolean|null);\n  }\n\n  /** Properties of an AttrDef. */\n  export declare interface IAttrDef {\n    /** AttrDef name */\n    name?: (string|null);\n\n    /** AttrDef type */\n    type?: (string|null);\n\n    /** AttrDef defaultValue */\n    defaultValue?: (IAttrValue|null);\n\n    /** AttrDef description */\n    description?: (string|null);\n\n    /** AttrDef hasMinimum */\n    hasMinimum?: (boolean|null);\n\n    /** AttrDef minimum */\n    minimum?: (number|string|null);\n\n    /** AttrDef allowedValues */\n    allowedValues?: (IAttrValue|null);\n  }\n\n  /** Properties of an OpDeprecation. */\n  export declare interface IOpDeprecation {\n    /** OpDeprecation version */\n    version?: (number|null);\n\n    /** OpDeprecation explanation */\n    explanation?: (string|null);\n  }\n}\n\n/** Properties of an OpList. */\nexport declare interface IOpList {\n  /** OpList op */\n  op?: (IOpDef[]|null);\n}\n\n/** Properties of a MetaGraphDef. */\nexport declare interface IMetaGraphDef {\n  /** MetaGraphDef metaInfoDef */\n  metaInfoDef?: (MetaGraphDef.IMetaInfoDef|null);\n\n  /** MetaGraphDef graphDef */\n  graphDef?: (IGraphDef|null);\n\n  /** MetaGraphDef saverDef */\n  saverDef?: (ISaverDef|null);\n\n  /** MetaGraphDef collectionDef */\n  collectionDef?: ({[k: string]: ICollectionDef}|null);\n\n  /** MetaGraphDef signatureDef */\n  signatureDef?: ({[k: string]: ISignatureDef}|null);\n\n  /** MetaGraphDef assetFileDef */\n  assetFileDef?: (IAssetFileDef[]|null);\n}\n\nexport namespace MetaGraphDef {\n  /** Properties of a MetaInfoDef. */\n  export declare interface IMetaInfoDef {\n    /** MetaInfoDef metaGraphVersion */\n    metaGraphVersion?: (string|null);\n\n    /** MetaInfoDef strippedOpList */\n    strippedOpList?: (IOpList|null);\n\n    /** MetaInfoDef anyInfo */\n    anyInfo?: (IAny|null);\n\n    /** MetaInfoDef tags */\n    tags?: (string[]|null);\n\n    /** MetaInfoDef tensorflowVersion */\n    tensorflowVersion?: (string|null);\n\n    /** MetaInfoDef tensorflowGitVersion */\n    tensorflowGitVersion?: (string|null);\n  }\n}\n\n/** Properties of a SavedModel. */\nexport declare interface ISavedModel {\n  /** SavedModel savedModelSchemaVersion */\n  savedModelSchemaVersion?: (number|string|null);\n\n  /** SavedModel metaGraphs */\n  metaGraphs?: (IMetaGraphDef[]|null);\n}\n\n/** Properties of a FunctionDefLibrary. */\nexport declare interface IFunctionDefLibrary {\n  /** FunctionDefLibrary function */\n  'function'?: (IFunctionDef[]|null);\n\n  /** FunctionDefLibrary gradient */\n  gradient?: (IGradientDef[]|null);\n}\n\n/** Properties of a FunctionDef. */\nexport declare interface IFunctionDef {\n  /** FunctionDef signature */\n  signature?: (IOpDef|null);\n\n  /** FunctionDef attr */\n  attr?: ({[k: string]: IAttrValue}|null);\n\n  /** FunctionDef nodeDef */\n  nodeDef?: (INodeDef[]|null);\n\n  /** FunctionDef ret */\n  ret?: ({[k: string]: string}|null);\n}\n\n/** Properties of a GradientDef. */\nexport declare interface IGradientDef {\n  /** GradientDef functionName */\n  functionName?: (string|null);\n\n  /** GradientDef gradientFunc */\n  gradientFunc?: (string|null);\n}\n", "\n/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpExecutor, OpMapper} from '../types';\n\nconst CUSTOM_OPS: {[key: string]: OpMapper} = {};\n\n/**\n * Register an Op for graph model executor. This allow you to register\n * TensorFlow custom op or override existing op.\n *\n * Here is an example of registering a new MatMul Op.\n * ```js\n * const customMatmul = (node) =>\n *    tf.matMul(\n *        node.inputs[0], node.inputs[1],\n *        node.attrs['transpose_a'], node.attrs['transpose_b']);\n *\n * tf.registerOp('MatMul', customMatmul);\n * ```\n * The inputs and attrs of the node object is based on the TensorFlow op\n * registry.\n *\n * @param name The Tensorflow Op name.\n * @param opFunc An op function which is called with the current graph node\n * during execution and needs to return a tensor or a list of tensors. The node\n * has the following attributes:\n *    - attr: A map from attribute name to its value\n *    - inputs: A list of input tensors\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function registerOp(name: string, opFunc: OpExecutor) {\n  const opMapper: OpMapper = {\n    tfOpName: name,\n    category: 'custom',\n    inputs: [],\n    attrs: [],\n    customExecutor: opFunc\n  };\n\n  CUSTOM_OPS[name] = opMapper;\n}\n\n/**\n * Retrieve the OpMapper object for the registered op.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function getRegisteredOp(name: string): OpMapper {\n  return CUSTOM_OPS[name];\n}\n\n/**\n * Deregister the Op for graph model executor.\n *\n * @param name The Tensorflow Op name.\n *\n * @doc {heading: 'Models', subheading: 'Op Registry'}\n */\nexport function deregisterOp(name: string) {\n  delete CUSTOM_OPS[name];\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {clone, Tensor, util} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {Node, ValueType} from '../types';\n\nexport function getParamValue(\n    paramName: string, node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  const inputParam = node.inputParams[paramName];\n  if (inputParam && inputParam.inputIndexStart !== undefined) {\n    const start = inputParam.inputIndexStart;\n    const end = inputParam.inputIndexEnd === 0 ?\n        undefined :\n        (inputParam.inputIndexEnd === undefined ? start + 1 :\n                                                  inputParam.inputIndexEnd);\n    if (inputParam.type === 'tensor') {\n      return getTensor(\n          node.inputNames[inputParam.inputIndexStart], tensorMap, context);\n    }\n    if (inputParam.type === 'tensors') {\n      const inputs = node.inputNames.slice(start, end);\n\n      return inputs.map(name => getTensor(name, tensorMap, context));\n    }\n    const tensor =\n        getTensor(node.inputNames.slice(start)[0], tensorMap, context);\n    const data = tensor.dataSync();\n    return inputParam.type === 'number' ?\n        data[0] :\n        util.toNestedArray(tensor.shape, data);\n  }\n  const attrParam = node.attrParams[paramName];\n  return attrParam && attrParam.value;\n}\n\n/**\n * Retrieve the tensor based on input name by extracting the node name and\n * output index information.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensor(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor {\n  const [nodeName, index] = parseNodeName(name);\n  const contextId = context.currentContextIds.find(contextId => {\n    return !!tensorsMap[getNodeNameWithContextId(nodeName, contextId)];\n  });\n\n  return contextId !== undefined ?\n      tensorsMap[getNodeNameWithContextId(nodeName, contextId)][index] :\n      undefined;\n}\n\n/**\n * Retrieve the tensors based on input name for current context.\n * @param name Node input name\n * @param tensorsMap Tensors map keyed by the node\n */\nexport function getTensorsForCurrentContenxt(\n    name: string, tensorsMap: NamedTensorsMap,\n    context: ExecutionContext): Tensor[] {\n  return tensorsMap[getNodeNameWithContextId(name, context.currentContextId)];\n}\n\n/**\n * Returns the node name and index from the Node input name.\n * @param inputName The input name of the node, in format of\n * node_name:output_index, i.e. MatMul:0, if the output_index is not set, it is\n * default to 0.\n */\nexport function getNodeNameAndIndex(\n    inputName: string, context?: ExecutionContext): [string, number] {\n  const [nodeName, index] = parseNodeName(inputName);\n\n  return [\n    getNodeNameWithContextId(nodeName, context && context.currentContextId),\n    index\n  ];\n}\n\nfunction getNodeNameWithContextId(name: string, contextId?: string): string {\n  return !!contextId ? `${name}-${contextId}` : name;\n}\n\nexport function parseNodeName(name: string): [string, number] {\n  const parts = name.split(':');\n  if (parts.length === 1) {\n    return [name, 0];\n  }\n\n  const nodeName = parts[0];\n  return [nodeName, Number(parts[parts.length - 1])];\n}\n\nexport function split(arr: number[], size: number) {\n  const res = [];\n  for (let i = 0; i < arr.length; i += size) {\n    res.push(arr.slice(i, i + size));\n  }\n  return res;\n}\nexport function getPadding(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): ValueType {\n  let pad = getParamValue('pad', node, tensorMap, context);\n  if (pad === 'explicit') {\n    // This is 1d array, we need to convert it to 2d array\n    pad = getParamValue('explicitPaddings', node, tensorMap, context);\n    const explicitPadding: [\n      [number, number], [number, number], [number, number], [number, number]\n    ] = [[0, 0], [0, 0], [0, 0], [0, 0]];\n    for (let i = 0; i < 4; i++) {\n      explicitPadding[i][0] = (pad as number[])[i * 2];\n      explicitPadding[i][1] = (pad as number[])[i * 2 + 1];\n    }\n    return explicitPadding;\n  }\n  return pad;\n}\n\n/**\n *  Reuse the tensor if it is marked as keep, otherwise clone the tensor to\n *  avoid disposal. This is important for TensorArray and TensorList ops, since\n *  internally they use a tensor as the id for TensorArray and TensorList, and\n * to simplify lookup, they also use Tensor.id as the key to the internal map.\n * These id tensors have been marked as kept in the backend, we need avoid clone\n * them in order to create new Tensor.id.\n * @param tensor\n */\nexport function cloneTensor(tensor: Tensor): Tensor {\n  return tensor.kept ? tensor : clone(tensor);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Add',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddV2',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AddN',\n    'category': 'arithmetic',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'BiasAdd',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sub',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'RealDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Div',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'DivNoNan',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorDiv',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mul',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Maximum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Minimum',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Pow',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SquaredDifference',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Mod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'FloorMod',\n    'category': 'arithmetic',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Abs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atan2',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ceil',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ClipByValue',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'clip_value_min', 'name': 'clipValueMin', 'type': 'number'},\n      {'tfName': 'clip_value_max', 'name': 'clipValueMax', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Complex',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'real', 'type': 'tensor'},\n      {'start': 1, 'name': 'imag', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ComplexAbs',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cos',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Cosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Elu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Exp',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Floor',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Imag',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Neg',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Real',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'Tout',\n        'name': 'outputType',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Prelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'alpha', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Relu6',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'clipValueMin',\n        'name': 'clipValueMin',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'clipValueMax',\n        'name': 'clipValueMax',\n        'type': 'number',\n        'defaultValue': 6\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Selu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sigmoid',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sin',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Rsqrt',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Square',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tan',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Tanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Sign',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Round',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Expm1',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Log1p',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Reciprocal',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Softplus',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Asinh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Acosh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Atanh',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Erf',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axes', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'keep_dims',\n        'name': 'keepDims',\n        'type': 'bool',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LeakyRelu',\n    'category': 'basic_math',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 0.2\n      },\n      {\n        'tfName': 'T',\n        'name': 'dtype',\n        'type': 'dtype',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'LoopCond',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'pred', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Switch',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'data', 'type': 'tensor'},\n      {'start': 1, 'name': 'pred', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'Merge',\n    'category': 'control',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Enter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'frame_name', 'name': 'frameName', 'type': 'string'},\n      {'tfName': 'is_constant', 'name': 'isConstant', 'type': 'bool'}\n    ]\n  },\n  {\n    'tfOpName': 'Exit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NextIteration',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'size', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'dynamic_size', 'name': 'dynamicSize', 'type': 'bool'},\n      {'tfName': 'clear_after_read', 'name': 'clearAfterRead', 'type': 'bool'},\n      {\n        'tfName': 'identical_element_shapes',\n        'name': 'identicalElementShapes',\n        'type': 'bool'\n      },\n      {'tfName': 'tensor_array_name', 'name': 'name', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayWriteV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayReadV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{\n      'tfName': 'dtype',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  },\n  {\n    'tfOpName': 'TensorArrayGatherV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayScatterV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArrayConcatV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}, {\n        'tfName': 'element_shape_except0',\n        'name': 'elementShapeExcept0',\n        'type': 'shape',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'TensorArraySplitV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n      {'start': 3, 'name': 'flowIn', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorArraySizeV3',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'},\n      {'start': 1, 'name': 'flowIn', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorArrayCloseV3',\n    'category': 'control',\n    'inputs': [{'start': 0, 'name': 'tensorArrayId', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'StatelessIf',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'If',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'cond', 'type': 'tensor'},\n      {'start': 1, 'end': 0, 'name': 'args', 'type': 'tensors'}\n    ],\n    'attrs': [\n      {'tfName': 'then_branch', 'name': 'thenBranch', 'type': 'func'},\n      {'tfName': 'else_branch', 'name': 'elseBranch', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'StatelessWhile',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'While',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'cond', 'name': 'cond', 'type': 'func'},\n      {'tfName': 'body', 'name': 'body', 'type': 'func'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListScatter',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListScatterV2',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 3, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGather',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'number[]'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListGetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListSetItem',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'index', 'type': 'number'},\n      {'start': 2, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListReserve',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 1, 'name': 'numElements', 'type': 'number'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListFromTensor',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'}\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListStack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs': [\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'},\n      {'tfName': 'num_elements', 'name': 'numElements', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListSplit',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n      {'start': 2, 'name': 'lengths', 'type': 'number[]'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListConcat',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'element_shape', 'name': 'elementShape', 'type': 'shape'},\n      {'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'TensorListPopBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'elementShape', 'type': 'shape'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TensorListPushBack',\n    'category': 'control',\n    'inputs': [\n      {'start': 0, 'name': 'tensorListId', 'type': 'tensor'},\n      {'start': 1, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs':\n        [{'tfName': 'element_dtype', 'name': 'elementDType', 'type': 'dtype'}]\n  },\n];\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'AvgPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPoolWithArgmax',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'}, {\n        'tfName': 'include_batch_in_index',\n        'name': 'includeBatchInIndex',\n        'type': 'bool'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'AvgPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MaxPool3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {'tfName': 'ksize', 'name': 'kernelSize', 'type': 'number[]'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Conv1D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'stride', 'name': 'stride', 'type': 'number'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NWC'\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'dilation',\n        'name': 'dilation',\n        'type': 'number',\n        'defaultValue': 1\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {'tfName': 'useCudnnOnGpu', 'name': 'useCudnnOnGpu', 'type': 'bool'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': '_FusedConv2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'use_cudnn_on_gpu',\n        'name': 'useCudnnOnGpu',\n        'type': 'bool',\n        'defaultValue': true\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n    ]\n  },\n  {\n    'tfOpName': 'Conv2DBackpropInput',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 2, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 0, 'name': 'outputShape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'},\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2d',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'input', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'explicit_paddings',\n        'name': 'explicitPaddings',\n        'type': 'number[]',\n        'defaultValue': []\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'FusedDepthwiseConv2dNative',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true},\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {\n        'tfName': 'dilations',\n        'name': 'dilations',\n        'type': 'number[]',\n        'defaultValue': [1, 1, 1, 1]\n      },\n      {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Conv3D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}, {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'defaultValue': 'NHWC'\n      },\n      {'tfName': 'dilations', 'name': 'dilations', 'type': 'number[]'}\n    ],\n  },\n  {\n    'tfOpName': 'Dilation2D',\n    'category': 'convolution',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'filter', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'strides', 'name': 'strides', 'type': 'number[]'},\n      {'tfName': 'rates', 'name': 'dilations', 'type': 'number[]'},\n      {'tfName': 'padding', 'name': 'pad', 'type': 'string'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Fill',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n      {'start': 1, 'name': 'value', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'LinSpace',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'num', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'OneHot',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'depth', 'type': 'number'},\n      {'start': 2, 'name': 'onValue', 'type': 'number', 'defaultValue': 1},\n      {'start': 3, 'name': 'offValue', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [\n      {\n        'tfName': 'axis',\n        'name': 'axis',\n        'type': 'number',\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Ones',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'OnesLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'RandomUniform',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'minval',\n        'name': 'minval',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'maxval',\n        'name': 'maxval',\n        'type': 'number',\n        'defaultValue': 1\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Range',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'start', 'type': 'number'},\n      {'start': 1, 'name': 'stop', 'type': 'number'},\n      {'start': 2, 'name': 'step', 'type': 'number', 'defaultValue': 0},\n    ],\n    'attrs': [{'tfName': 'Tidx', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'TruncatedNormal',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'means',\n        'name': 'mean',\n        'type': 'number',\n        'defaultValue': 0.0\n      },\n      {\n        'tfName': 'stddev',\n        'name': 'stdDev',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'}, {\n        'tfName': 'seed2',\n        'name': 'seed2',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      },\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'T', 'name': 'T', 'type': 'number', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Zeros',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'ZerosLike',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{'tfName': 'T', 'name': 'dtype', 'type': 'dtype'}]\n  },\n  {\n    'tfOpName': 'Multinomial',\n    'category': 'creation',\n    'inputs': [\n      {'start': 0, 'name': 'logits', 'type': 'tensor'},\n      {'start': 1, 'name': 'numSamples', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'seed', 'name': 'seed', 'type': 'number'},\n      {'tfName': 'seed2', 'name': 'seed2', 'type': 'number'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype'},\n      {'tfName': 'output_dtype', 'name': 'output_dtype', 'type': 'dtype'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'NonMaxSuppressionV2',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV3',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV4',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'}\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}, {\n        'tfName': 'T_threshold',\n        'name': 'threshold',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {\n        'tfName': 'pad_to_max_output_size',\n        'name': 'padToMaxOutputSize',\n        'type': 'bool'\n      }\n    ]\n  },\n  {\n    'tfOpName': 'NonMaxSuppressionV5',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 1, 'name': 'scores', 'type': 'tensor'},\n      {'start': 2, 'name': 'maxOutputSize', 'type': 'number'},\n      {'start': 3, 'name': 'iouThreshold', 'type': 'number'},\n      {'start': 4, 'name': 'scoreThreshold', 'type': 'number'},\n      {'start': 5, 'name': 'softNmsSigma', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Where',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ListDiff',\n    'category': 'dynamic',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'y', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'TopKV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'k', 'type': 'number'},\n    ],\n    'attrs': [{'tfName': 'sorted', 'name': 'sorted', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Unique',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n  },\n  {\n    'tfOpName': 'UniqueV2',\n    'category': 'evaluation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n  },\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'PlaceholderWithDefault',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'default', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'Placeholder',\n    'category': 'graph',\n    'attrs': [\n      {'tfName': 'shape', 'name': 'shape', 'type': 'shape'},\n      {'tfName': 'dtype', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {'tfOpName': 'Const', 'category': 'graph'}, {\n    'tfOpName': 'Identity',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IdentityN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Snapshot',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Rank',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Size',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'Shape',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'ShapeN',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'end': 0, 'name': 'x', 'type': 'tensors'}]\n  },\n  {\n    'tfOpName': 'Print',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'data', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'message', 'name': 'message', 'type': 'string'}, {\n        'tfName': 'first_n',\n        'name': 'firstN',\n        'type': 'number',\n        'notSupported': true\n      },\n      {\n        'tfName': 'summarize',\n        'name': 'summarize',\n        'type': 'number',\n        'defaultValue': 3\n      }\n    ]\n  },\n  {'tfOpName': 'NoOp', 'category': 'graph', 'inputs': []}, {\n    'tfOpName': 'StopGradient',\n    'category': 'graph',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'FakeQuantWithMinMaxVars',\n    'category': 'graph',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'min', 'name': 'min', 'type': 'number'},\n      {'tfName': 'max', 'name': 'max', 'type': 'number'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ResizeBilinear',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ResizeNearestNeighbor',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'images', 'type': 'tensor'},\n      {'start': 1, 'name': 'size', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'align_corners', 'name': 'alignCorners', 'type': 'bool'},\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'CropAndResize',\n    'category': 'image',\n    'inputs': [\n      {'start': 0, 'name': 'image', 'type': 'tensor'},\n      {'start': 1, 'name': 'boxes', 'type': 'tensor'},\n      {'start': 2, 'name': 'boxInd', 'type': 'tensor'},\n      {'start': 3, 'name': 'cropSize', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {'tfName': 'method', 'name': 'method', 'type': 'string'}, {\n        'tfName': 'extrapolation_value',\n        'name': 'extrapolationValue',\n        'type': 'number'\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Equal',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'NotEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Greater',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'GreaterEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Less',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LessEqual',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalAnd',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalNot',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'LogicalOr',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Select',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'SelectV2',\n    'category': 'logical',\n    'inputs': [\n      {'start': 0, 'name': 'condition', 'type': 'tensor'},\n      {'start': 1, 'name': 'a', 'type': 'tensor'},\n      {'start': 2, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {OpMapper} from '../types';\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': '_FusedMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n      {'start': 2, end: 0, 'name': 'args', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'num_args', 'name': 'numArgs', 'type': 'number'}, {\n        'tfName': 'fused_ops',\n        'name': 'fusedOps',\n        'type': 'string[]',\n        'defaultValue': []\n      },\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.0001\n      },\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'MatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'transpose_a',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'transpose_b',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMul',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'BatchMatMulV2',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'a', 'type': 'tensor'},\n      {'start': 1, 'name': 'b', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'adj_x',\n        'name': 'transposeA',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {\n        'tfName': 'adj_y',\n        'name': 'transposeB',\n        'type': 'bool',\n        'defaultValue': false\n      },\n      {'tfName': 'T', 'name': 'dtype', 'type': 'dtype', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'Transpose',\n    'category': 'matrices',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'perm', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'T',\n      'name': 'dtype',\n      'type': 'dtype',\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FusedBatchNorm',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV2',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'FusedBatchNormV3',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'scale', 'type': 'tensor'},\n      {'start': 2, 'name': 'offset', 'type': 'tensor'},\n      {'start': 3, 'name': 'mean', 'type': 'tensor'},\n      {'start': 4, 'name': 'variance', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'epsilon',\n        'name': 'epsilon',\n        'type': 'number',\n        'defaultValue': 0.001\n      },\n      {\n        'tfName': 'data_format',\n        'name': 'dataFormat',\n        'type': 'string',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'LRN',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'depth_radius',\n        'name': 'radius',\n        'type': 'number',\n        'defaultValue': 5\n      },\n      {'tfName': 'bias', 'name': 'bias', 'type': 'number', 'defaultValue': 1.0},\n      {\n        'tfName': 'alpha',\n        'name': 'alpha',\n        'type': 'number',\n        'defaultValue': 1.0\n      },\n      {\n        'tfName': 'beta',\n        'name': 'beta',\n        'type': 'number',\n        'defaultValue': 0.5\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Softmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'LogSoftmax',\n    'category': 'normalization',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'normalization',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': true,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Max',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Mean',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Min',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Sum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'All',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Any',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'ArgMax',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'ArgMin',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Prod',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'},\n    ],\n    'attrs': [{'tfName': 'keep_dims', 'name': 'keepDims', 'type': 'bool'}]\n  },\n  {\n    'tfOpName': 'Cumsum',\n    'category': 'reduction',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'},\n    ],\n    'attrs': [\n      {'tfName': 'exclusive', 'name': 'exclusive', 'type': 'bool'},\n      {'tfName': 'reverse', 'name': 'reverse', 'type': 'bool'}\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'ConcatV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': -1, 'name': 'tensors', 'type': 'tensors'},\n      {'start': -1, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs':\n        [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n  },\n  {\n    'tfOpName': 'Concat',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 1, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n      {'start': 0, 'name': 'axis', 'type': 'number'}\n    ],\n    'attrs': [{'tfName': 'N', 'name': 'n', 'type': 'number', 'defaultValue': 2}]\n\n  },\n  {\n    'tfOpName': 'GatherV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Gather',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'validate_indices',\n        'name': 'validateIndices',\n        'type': 'bool',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reverse',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'dims', 'type': 'bool', 'notSupported': true}\n    ]\n  },\n  {\n    'tfOpName': 'ReverseV2',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Slice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'size', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'StridedSlice',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'begin', 'type': 'number[]'},\n      {'start': 2, 'name': 'end', 'type': 'number[]'},\n      {'start': 3, 'name': 'strides', 'type': 'number[]'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'begin_mask',\n        'name': 'beginMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'end_mask',\n        'name': 'endMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'new_axis_mask',\n        'name': 'newAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'ellipsis_mask',\n        'name': 'ellipsisMask',\n        'type': 'number',\n        'defaultValue': 0\n      },\n      {\n        'tfName': 'shrink_axis_mask',\n        'name': 'shrinkAxisMask',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Pack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'end': 0, 'name': 'tensors', 'type': 'tensors'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'Unpack',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'tensor', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'axis', 'name': 'axis', 'type': 'number', 'defaultValue': 0}, {\n        'tfName': 'num',\n        'name': 'num',\n        'type': 'number',\n        'defaultValue': 0,\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Tile',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'reps', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Split',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'axis', 'type': 'number', 'defaultValue': 0},\n      {'start': 1, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'num_split',\n      'name': 'numOrSizeSplits',\n      'type': 'number',\n      'defaultValue': 1\n    }]\n  },\n  {\n    'tfOpName': 'SplitV',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'numOrSizeSplits', 'type': 'number[]'},\n      {'start': 2, 'name': 'axis', 'type': 'number', 'defaultValue': 0}\n    ]\n  },\n  {\n    'tfOpName': 'ScatterNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'indices', 'type': 'tensor'},\n      {'start': 1, 'name': 'values', 'type': 'tensor'},\n      {'start': 2, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'GatherNd',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'indices', 'type': 'tensor'}\n    ]\n  },\n  {\n    'tfOpName': 'SparseToDense',\n    'category': 'slice_join',\n    'inputs': [\n      {'start': 0, 'name': 'sparseIndices', 'type': 'tensor'},\n      {'start': 1, 'name': 'outputShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'sparseValues', 'type': 'tensor'},\n      {'start': 3, 'name': 'defaultValue', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'validate_indices',\n      'name': 'validateIndices',\n      'type': 'bool',\n      'defaultValue': false,\n      'notSupported': true\n    }]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'FFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'IFFT',\n    'category': 'spectral',\n    'inputs': [{'start': 0, 'name': 'x', 'type': 'tensor'}]\n  },\n  {\n    'tfOpName': 'RFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  },\n  {\n    'tfOpName': 'IRFFT',\n    'category': 'spectral',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'}, {\n        'start': 1,\n        'name': 'fft_length',\n        'type': 'number',\n        'notSupported': true\n      }\n    ]\n  }\n];\n", "import {OpMapper} from '../types';\n\n/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport const json: OpMapper[] = [\n  {\n    'tfOpName': 'Cast',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {\n        'tfName': 'SrcT',\n        'name': 'sdtype',\n        'type': 'dtype',\n        'notSupported': true\n      },\n      {'tfName': 'DstT', 'name': 'dtype', 'type': 'dtype'}\n    ]\n  },\n  {\n    'tfOpName': 'ExpandDims',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'axis', 'type': 'number'}\n    ]\n  },\n  {\n    'tfOpName': 'Pad',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'},\n    ],\n    'attrs': [{\n      'tfName': 'constant_value',\n      'name': 'constantValue',\n      'type': 'number',\n      'defaultValue': 0\n    }]\n  },\n  {\n    'tfOpName': 'PadV2',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'padding', 'type': 'number[]'}, {\n        'start': 2,\n        'name': 'constantValue',\n        'type': 'number',\n        'defaultValue': 0\n      }\n    ]\n  },\n  {\n    'tfOpName': 'Reshape',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'Squeeze',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [{\n      'tfName': 'axis',\n      'tfDeprecatedName': 'squeeze_dims',\n      'name': 'axis',\n      'type': 'number[]'\n    }]\n  },\n  {\n    'tfOpName': 'SpaceToBatchND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'paddings', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'BatchToSpaceND',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'blockShape', 'type': 'number[]'},\n      {'start': 2, 'name': 'crops', 'type': 'number[]'}\n    ]\n  },\n  {\n    'tfOpName': 'DepthToSpace',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n    ],\n    'attrs': [\n      {'tfName': 'block_size', 'name': 'blockSize', 'type': 'number'},\n      {'tfName': 'data_format', 'name': 'dataFormat', 'type': 'string'}\n    ]\n  },\n  {\n    'tfOpName': 'BroadcastTo',\n    'category': 'transformation',\n    'inputs': [\n      {'start': 0, 'name': 'x', 'type': 'tensor'},\n      {'start': 1, 'name': 'shape', 'type': 'number[]'},\n    ],\n    'attrs': []\n  }\n];\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, env} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\n\nimport {getRegisteredOp} from './custom_op/register';\nimport {getNodeNameAndIndex} from './executors/utils';\nimport * as arithmetic from './op_list/arithmetic';\nimport * as basicMath from './op_list/basic_math';\nimport * as control from './op_list/control';\nimport * as convolution from './op_list/convolution';\nimport * as creation from './op_list/creation';\nimport * as dynamic from './op_list/dynamic';\nimport * as evaluation from './op_list/evaluation';\nimport * as graph from './op_list/graph';\nimport * as image from './op_list/image';\nimport * as logical from './op_list/logical';\nimport * as matrices from './op_list/matrices';\nimport * as normalization from './op_list/normalization';\nimport * as reduction from './op_list/reduction';\nimport * as sliceJoin from './op_list/slice_join';\nimport * as spectral from './op_list/spectral';\nimport * as transformation from './op_list/transformation';\nimport {Graph, InputParamValue, Node, OpMapper, ParamValue} from './types';\n\nexport class OperationMapper {\n  private static _instance: OperationMapper;\n\n  private opMappers: {[key: string]: OpMapper};\n\n  // Singleton instance for the mapper\n  public static get Instance() {\n    return this._instance || (this._instance = new this());\n  }\n\n  // Loads the op mapping from the JSON file.\n  private constructor() {\n    const ops = [\n      arithmetic, basicMath, control, convolution, creation, dynamic,\n      evaluation, logical, image, graph, matrices, normalization, reduction,\n      sliceJoin, spectral, transformation\n    ];\n    const mappersJson: OpMapper[] = [].concat(...ops.map(op => op.json));\n\n    this.opMappers = mappersJson.reduce<{[key: string]: OpMapper}>(\n        (map, mapper: OpMapper) => {\n          map[mapper.tfOpName] = mapper;\n          return map;\n        },\n        {});\n  }\n\n  // Converts the model inference graph from Tensorflow GraphDef to local\n  // representation for TensorFlow.js API\n  transformGraph(\n      graph: tensorflow.IGraphDef,\n      signature: tensorflow.ISignatureDef = {}): Graph {\n    const tfNodes = graph.node;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    const initNodes: Node[] = [];\n    const nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n      map[node.name] = this.mapNode(node);\n      if (node.op.startsWith('Placeholder')) {\n        placeholders.push(map[node.name]);\n      } else if (node.op === 'Const') {\n        weights.push(map[node.name]);\n      } else if (node.input == null || node.input.length === 0) {\n        initNodes.push(map[node.name]);\n      }\n      return map;\n    }, {});\n\n    let inputs: Node[] = [];\n    const outputs: Node[] = [];\n    let inputNodeNameToKey: {[key: string]: string} = {};\n    let outputNodeNameToKey: {[key: string]: string} = {};\n    if (signature != null) {\n      inputNodeNameToKey = this.mapSignatureEntries(signature.inputs);\n      outputNodeNameToKey = this.mapSignatureEntries(signature.outputs);\n    }\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    // if signature has not outputs set, add any node that does not have\n    // outputs.\n    if (Object.keys(outputNodeNameToKey).length === 0) {\n      allNodes.forEach(key => {\n        const node = nodes[key];\n        if (node.children.length === 0) {\n          outputs.push(node);\n        }\n      });\n    } else {\n      Object.keys(outputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node != null) {\n          node.signatureKey = outputNodeNameToKey[name];\n          outputs.push(node);\n        }\n      });\n    }\n\n    if (Object.keys(inputNodeNameToKey).length > 0) {\n      Object.keys(inputNodeNameToKey).forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        const node = nodes[nodeName];\n        if (node) {\n          node.signatureKey = inputNodeNameToKey[name];\n          inputs.push(node);\n        }\n      });\n    } else {\n      inputs = placeholders;\n    }\n\n    let functions = {};\n    if (graph.library != null && graph.library.function != null) {\n      functions = graph.library.function.reduce((functions, func) => {\n        functions[func.signature.name] = this.mapFunction(func);\n        return functions;\n      }, {} as {[key: string]: Graph});\n    }\n\n    const result: Graph =\n        {nodes, inputs, outputs, weights, placeholders, signature, functions};\n\n    if (initNodes.length > 0) {\n      result.initNodes = initNodes;\n    }\n\n    return result;\n  }\n\n  private mapSignatureEntries(entries: {[k: string]: tensorflow.ITensorInfo}) {\n    return Object.keys(entries || {})\n        .reduce<{[key: string]: string}>((prev, curr) => {\n          prev[entries[curr].name] = curr;\n          return prev;\n        }, {});\n  }\n\n  private mapNode(node: tensorflow.INodeDef): Node {\n    // Unsupported ops will cause an error at run-time (not parse time), since\n    // they may not be used by the actual execution subgraph.\n    const mapper =\n        getRegisteredOp(node.op) || this.opMappers[node.op] || {} as OpMapper;\n    if (node.attr == null) {\n      node.attr = {};\n    }\n\n    const newNode: Node = {\n      name: node.name,\n      op: node.op,\n      category: mapper.category,\n      inputNames:\n          (node.input ||\n           []).map(input => input.startsWith('^') ? input.substr(1) : input),\n      inputs: [],\n      children: [],\n      inputParams: {},\n      attrParams: {},\n      rawAttrs: node.attr\n    };\n\n    if (mapper.inputs != null) {\n      newNode.inputParams =\n          mapper.inputs.reduce<{[key: string]: InputParamValue}>(\n              (map, param) => {\n                map[param.name] = {\n                  type: param.type,\n                  inputIndexStart: param.start,\n                  inputIndexEnd: param.end\n                };\n                return map;\n              },\n              {});\n    }\n    if (mapper.attrs != null) {\n      newNode.attrParams =\n          mapper.attrs.reduce<{[key: string]: ParamValue}>((map, param) => {\n            const type = param.type;\n            let value = undefined;\n            switch (param.type) {\n              case 'string':\n                value = getStringParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'string[]':\n                value = getStringArrayParam(\n                    node.attr, param.tfName, param.defaultValue as string[]);\n\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getStringArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string[]);\n                }\n                break;\n              case 'number':\n                value = getNumberParam(\n                    node.attr, param.tfName,\n                    (param.defaultValue || 0) as number);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumberParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number);\n                }\n                break;\n              case 'number[]':\n                value = getNumericArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getNumericArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'bool':\n                value = getBoolParam(\n                    node.attr, param.tfName, param.defaultValue as boolean);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean);\n                }\n                break;\n              case 'bool[]':\n                value = getBoolArrayParam(\n                    node.attr, param.tfName, param.defaultValue as boolean[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getBoolArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as boolean[]);\n                }\n                break;\n              case 'shape':\n                value = getTensorShapeParam(\n                    node.attr, param.tfName, param.defaultValue as number[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[]);\n                }\n                break;\n              case 'shape[]':\n                value = getTensorShapeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as number[][]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getTensorShapeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as number[][]);\n                }\n                break;\n              case 'dtype':\n                value = getDtypeParam(\n                    node.attr, param.tfName, param.defaultValue as DataType);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType);\n                }\n                break;\n              case 'dtype[]':\n                value = getDtypeArrayParam(\n                    node.attr, param.tfName, param.defaultValue as DataType[]);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getDtypeArrayParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as DataType[]);\n                }\n                break;\n              case 'func':\n                value = getFuncParam(\n                    node.attr, param.tfName, param.defaultValue as string);\n                if (value === undefined && !!param.tfDeprecatedName) {\n                  value = getFuncParam(\n                      node.attr, param.tfDeprecatedName,\n                      param.defaultValue as string);\n                }\n                break;\n              case 'tensor':\n              case 'tensors':\n                break;\n              default:\n                throw new Error(\n                    `Unsupported param type: ${param.type} for op: ${node.op}`);\n            }\n            map[param.name] = {value, type};\n            return map;\n          }, {});\n    }\n    return newNode;\n  }\n\n  // map the TFunctionDef to TFJS graph object\n  private mapFunction(functionDef: tensorflow.IFunctionDef): Graph {\n    const tfNodes = functionDef.nodeDef;\n    const placeholders: Node[] = [];\n    const weights: Node[] = [];\n    let nodes: {[key: string]: Node} = {};\n    if (tfNodes != null) {\n      nodes = tfNodes.reduce<{[key: string]: Node}>((map, node) => {\n        map[node.name] = this.mapNode(node);\n        if (node.op === 'Const') {\n          weights.push(map[node.name]);\n        }\n        return map;\n      }, {});\n    }\n    const inputs: Node[] = [];\n    const outputs: Node[] = [];\n\n    functionDef.signature.inputArg.forEach(arg => {\n      const [nodeName, ] = getNodeNameAndIndex(arg.name);\n      const node: Node = {\n        name: nodeName,\n        op: 'Placeholder',\n        inputs: [],\n        inputNames: [],\n        category: 'graph',\n        inputParams: {},\n        attrParams: {dtype: {value: parseDtypeParam(arg.type), type: 'dtype'}},\n        children: []\n      };\n      node.signatureKey = arg.name;\n      inputs.push(node);\n      nodes[nodeName] = node;\n    });\n\n    const allNodes = Object.keys(nodes);\n    allNodes.forEach(key => {\n      const node = nodes[key];\n      node.inputNames.forEach(name => {\n        const [nodeName, ] = getNodeNameAndIndex(name);\n        node.inputs.push(nodes[nodeName]);\n        nodes[nodeName].children.push(node);\n      });\n    });\n\n    const returnNodeMap = functionDef.ret;\n\n    functionDef.signature.outputArg.forEach(output => {\n      const [nodeName, index] = getNodeNameAndIndex(returnNodeMap[output.name]);\n      const node = nodes[nodeName];\n      if (node != null) {\n        node.defaultOutput = index;\n        outputs.push(node);\n      }\n    });\n\n    const signature = this.mapArgsToSignature(functionDef);\n    return {nodes, inputs, outputs, weights, placeholders, signature};\n  }\n\n  private mapArgsToSignature(functionDef: tensorflow.IFunctionDef):\n      tensorflow.ISignatureDef {\n    return {\n      methodName: functionDef.signature.name,\n      inputs: functionDef.signature.inputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n      outputs: functionDef.signature.outputArg.reduce(\n          (map, arg) => {\n            map[arg.name] = this.mapArgToTensorInfo(arg, functionDef.ret);\n            return map;\n          },\n          {} as {[key: string]: tensorflow.ITensorInfo}),\n    };\n  }\n\n  private mapArgToTensorInfo(\n      arg: tensorflow.OpDef.IArgDef,\n      nameMap?: {[key: string]: string}): tensorflow.ITensorInfo {\n    let name = arg.name;\n    if (nameMap != null) {\n      name = nameMap[name];\n    }\n    return {name, dtype: arg.type};\n  }\n}\n\nexport function decodeBase64(text: string): string {\n  const global = env().global;\n  if (typeof global.atob !== 'undefined') {\n    return global.atob(text);\n  } else if (typeof Buffer !== 'undefined') {\n    return new Buffer(text, 'base64').toString();\n  } else {\n    throw new Error(\n        'Unable to decode base64 in this environment. ' +\n        'Missing built-in atob() or Buffer()');\n  }\n}\n\nexport function parseStringParam(s: []|string, keepCase: boolean): string {\n  const value =\n      Array.isArray(s) ? String.fromCharCode.apply(null, s) : decodeBase64(s);\n  return keepCase ? value : value.toLowerCase();\n}\n\nexport function getStringParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string,\n    keepCase = false): string {\n  const param = attrs[name];\n  if (param != null) {\n    return parseStringParam(param.s, keepCase);\n  }\n  return def;\n}\n\nexport function getBoolParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean): boolean {\n  const param = attrs[name];\n  return param ? param.b : def;\n}\n\nexport function getNumberParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number): number {\n  const param = attrs[name] || {};\n  const value =\n      param['i'] != null ? param['i'] : (param['f'] != null ? param['f'] : def);\n  return (typeof value === 'number') ? value : parseInt(value, 10);\n}\n\nexport function parseDtypeParam(value: string|tensorflow.DataType): DataType {\n  if (typeof (value) === 'string') {\n    // tslint:disable-next-line:no-any\n    value = tensorflow.DataType[value as any];\n  }\n  switch (value) {\n    case tensorflow.DataType.DT_FLOAT:\n      return 'float32';\n    case tensorflow.DataType.DT_INT32:\n    case tensorflow.DataType.DT_INT64:\n    case tensorflow.DataType.DT_INT8:\n    case tensorflow.DataType.DT_UINT8:\n      return 'int32';\n    case tensorflow.DataType.DT_BOOL:\n      return 'bool';\n    case tensorflow.DataType.DT_DOUBLE:\n      return 'float32';\n    case tensorflow.DataType.DT_STRING:\n      return 'string';\n    default:\n      // Unknown dtype error will happen at runtime (instead of parse time),\n      // since these nodes might not be used by the actual subgraph execution.\n      return null;\n  }\n}\n\nexport function getFuncParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: string): string {\n  const param = attrs[name];\n  if (param && param.func) {\n    return param.func.name;\n  }\n  return def;\n}\n\nexport function getDtypeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType): DataType {\n  const param = attrs[name];\n  if (param && param.type) {\n    return parseDtypeParam(param.type);\n  }\n  return def;\n}\n\nexport function getDtypeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: DataType[]): DataType[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.type) {\n    return param.list.type.map(v => parseDtypeParam(v));\n  }\n  return def;\n}\n\nexport function parseTensorShapeParam(shape: tensorflow.ITensorShape): number[]|\n    undefined {\n  if (shape.unknownRank) {\n    return undefined;\n  }\n  if (shape.dim != null) {\n    return shape.dim.map(\n        dim =>\n            (typeof dim.size === 'number') ? dim.size : parseInt(dim.size, 10));\n  }\n  return [];\n}\n\nexport function getTensorShapeParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def?: number[]): number[]|undefined {\n  const param = attrs[name];\n  if (param && param.shape) {\n    return parseTensorShapeParam(param.shape);\n  }\n  return def;\n}\n\nexport function getNumericArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[]): number[] {\n  const param = attrs[name];\n  if (param) {\n    return ((param.list.f && param.list.f.length ? param.list.f :\n                                                   param.list.i) ||\n            [])\n        .map(v => (typeof v === 'number') ? v : parseInt(v, 10));\n  }\n  return def;\n}\n\nexport function getStringArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string, def: string[],\n    keepCase = false): string[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.s) {\n    return param.list.s.map((v) => {\n      return parseStringParam(v, keepCase);\n    });\n  }\n  return def;\n}\n\nexport function getTensorShapeArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: number[][]): number[][] {\n  const param = attrs[name];\n  if (param && param.list && param.list.shape) {\n    return param.list.shape.map((v) => {\n      return parseTensorShapeParam(v);\n    });\n  }\n  return def;\n}\n\nexport function getBoolArrayParam(\n    attrs: {[key: string]: tensorflow.IAttrValue}, name: string,\n    def: boolean[]): boolean[] {\n  const param = attrs[name];\n  if (param && param.list && param.list.b) {\n    return param.list.b;\n  }\n  return def;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {getTensor} from '../executors/utils';\nimport {getBoolArrayParam, getBoolParam, getDtypeArrayParam, getDtypeParam, getNumberParam, getNumericArrayParam, getStringArrayParam, getStringParam, getTensorShapeArrayParam, getTensorShapeParam} from '../operation_mapper';\nimport {GraphNode, Node, ValueType} from '../types';\n\n/**\n * Helper class for lookup inputs and params for nodes in the model graph.\n */\nexport class NodeValueImpl implements GraphNode {\n  public readonly inputs: Tensor[] = [];\n  public readonly attrs: {[key: string]: ValueType} = {};\n  constructor(\n      private node: Node, private tensorMap: NamedTensorsMap,\n      private context: ExecutionContext) {\n    this.inputs = node.inputNames.map(name => this.getInput(name));\n    if (node.rawAttrs != null) {\n      this.attrs = Object.keys(node.rawAttrs)\n                       .reduce((attrs: {[key: string]: ValueType}, key) => {\n                         attrs[key] = this.getAttr(key);\n                         return attrs;\n                       }, {});\n    }\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getInput(name: string): Tensor {\n    return getTensor(name, this.tensorMap, this.context);\n  }\n\n  /**\n   * Return the value of the attribute or input param.\n   * @param name String: name of attribute or input param.\n   */\n  private getAttr(name: string, defaultValue?: ValueType): ValueType {\n    const value = this.node.rawAttrs[name];\n    if (value.tensor != null) {\n      return getTensor(name, this.tensorMap, this.context);\n    }\n    if (value.i != null || value.f != null) {\n      return getNumberParam(this.node.rawAttrs, name, defaultValue as number);\n    }\n    if (value.s != null) {\n      return getStringParam(this.node.rawAttrs, name, defaultValue as string);\n    }\n    if (value.b != null) {\n      return getBoolParam(this.node.rawAttrs, name, defaultValue as boolean);\n    }\n    if (value.shape != null) {\n      return getTensorShapeParam(\n          this.node.rawAttrs, name, defaultValue as number[]);\n    }\n    if (value.type != null) {\n      return getDtypeParam(this.node.rawAttrs, name, defaultValue as DataType);\n    }\n    if (value.list != null) {\n      if (value.list.i != null || value.list.f != null) {\n        return getNumericArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[]);\n      }\n      if (value.list.s != null) {\n        return getStringArrayParam(\n            this.node.rawAttrs, name, defaultValue as string[]);\n      }\n      if (value.list.shape != null) {\n        return getTensorShapeArrayParam(\n            this.node.rawAttrs, name, defaultValue as number[][]);\n      }\n      if (value.list.b != null) {\n        return getBoolArrayParam(\n            this.node.rawAttrs, name, defaultValue as boolean[]);\n      }\n      if (value.list.type != null) {\n        return getDtypeArrayParam(\n            this.node.rawAttrs, name, defaultValue as DataType[]);\n      }\n    }\n\n    return defaultValue;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BiasAdd':\n        case 'AddV2':\n        case 'Add': {\n          return [tfOps.add(\n              (getParamValue('a', node, tensorMap, context) as Tensor),\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'AddN': {\n          return [tfOps.addN((\n              getParamValue('tensors', node, tensorMap, context) as Tensor[]))];\n        }\n        case 'FloorMod':\n        case 'Mod':\n          return [tfOps.mod(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'Mul':\n          return [tfOps.mul(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        case 'RealDiv':\n        case 'Div': {\n          return [tfOps.div(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'DivNoNan': {\n          return [tfOps.divNoNan(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'FloorDiv': {\n          return [tfOps.floorDiv(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sub': {\n          return [tfOps.sub(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Minimum': {\n          return [tfOps.minimum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Maximum': {\n          return [tfOps.maximum(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Pow': {\n          return [tfOps.pow(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'SquaredDifference': {\n          return [tfOps.squaredDifference(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'arithmetic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Abs':\n        case 'ComplexAbs':\n          return [tfOps.abs(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acos':\n          return [tfOps.acos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Acosh':\n          return [tfOps.acosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asin':\n          return [tfOps.asin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Asinh':\n          return [tfOps.asinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan':\n          return [tfOps.atan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Atan2':\n          return [tfOps.atan2(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('y', node, tensorMap, context) as Tensor)];\n        case 'Atanh':\n          return [tfOps.atanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Ceil':\n          return [tfOps.ceil(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Complex':\n          return [tfOps.complex(\n              getParamValue('real', node, tensorMap, context) as Tensor,\n              getParamValue('imag', node, tensorMap, context) as Tensor)];\n        case 'Cos':\n          return [tfOps.cos(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Cosh':\n          return [tfOps.cosh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Elu':\n          return [tfOps.elu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Erf':\n          return [tfOps.erf(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Exp':\n          return [tfOps.exp(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Expm1': {\n          return [tfOps.expm1(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Floor':\n          return [tfOps.floor(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log':\n          return [tfOps.log(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Log1p': {\n          return [tfOps.log1p(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Imag':\n          return [tfOps.imag(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n\n        case 'Neg':\n          return [tfOps.neg(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Reciprocal': {\n          return [tfOps.reciprocal(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Real':\n          return [tfOps.real(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu':\n          return [tfOps.relu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Round': {\n          return [tfOps.round(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Selu':\n          return [tfOps.selu(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sigmoid':\n          return [tfOps.sigmoid(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sin':\n          return [tfOps.sin(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Sign': {\n          return [tfOps.sign(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sinh': {\n          return [tfOps.sinh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Softplus': {\n          return [tfOps.softplus(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Sqrt': {\n          return [tfOps.sqrt(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Square': {\n          return [tfOps.square(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tanh': {\n          return [tfOps.tanh(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'Tan':\n          return [tfOps.tan(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        case 'Relu6':\n        case 'ClipByValue':\n          return [tfOps.clipByValue(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('clipValueMin', node, tensorMap, context) as number,\n              getParamValue('clipValueMax', node, tensorMap, context) as\n                  number)];\n        case 'Rsqrt':\n          return [tfOps.rsqrt(\n              getTensor(node.inputNames[0], tensorMap, context))];\n        case 'Prod':\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('axes', node, tensorMap, context) as number[])];\n        case 'LeakyRelu':\n          return [tfOps.leakyRelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as number)];\n        case 'Prelu':\n          return [tfOps.prelu(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('alpha', node, tensorMap, context) as Tensor)];\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'basic_math';\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n/**\n * This differs from util.assertShapesMatch in that it allows values of\n * negative one, an undefined size of a dimensinon, in a shape to match\n * anything.\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function assertShapesMatchAllowUndefinedSize(\n    shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n  util.assert(\n      shapesEqualAllowUndefinedSize(shapeA, shapeB),\n      () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function shapesEqualAllowUndefinedSize(n1: number[], n2: number[]) {\n  if (n1.length !== n2.length) {\n    return false;\n  }\n  for (let i = 0; i < n1.length; i++) {\n    if (n1[i] !== -1 && n2[i] !== -1 && n1[i] !== n2[i]) {\n      return false;\n    }\n  }\n  return true;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\nexport interface TensorWithState {\n  tensor?: Tensor;\n  written?: boolean;\n  read?: boolean;\n  cleared?: boolean;\n}\n/**\n * The TensorArray object keeps an array of Tensors.  It\n * allows reading from the array and writing to the array.\n */\nexport class TensorArray {\n  private tensors: TensorWithState[] = [];\n  private closed_ = false;\n  readonly idTensor: Tensor;\n  constructor(\n      readonly name: string, readonly dtype: DataType, private maxSize: number,\n      private elementShape: number[], readonly identicalElementShapes: boolean,\n      readonly dynamicSize: boolean, readonly clearAfterRead: boolean) {\n    this.idTensor = scalar(0);\n    keep(this.idTensor);\n  }\n\n  get id() {\n    return this.idTensor.id;\n  }\n\n  get closed() {\n    return this.closed_;\n  }\n\n  /**\n   * Dispose the tensors and idTensor and mark the TensoryArray as closed.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.tensor.id)) {\n        tensor.tensor.dispose();\n      }\n    });\n    this.tensors = [];\n    this.closed_ = true;\n    this.idTensor.dispose();\n  }\n\n  size(): number {\n    return this.tensors.length;\n  }\n\n  /**\n   * Read the value at location index in the TensorArray.\n   * @param index Number the index to read from.\n   */\n  read(index: number): Tensor {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || index >= this.size()) {\n      throw new Error(`Tried to read from index ${index}, but array size is: ${\n          this.size()}`);\n    }\n\n    const tensorWithState = this.tensors[index];\n    if (tensorWithState.cleared) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not read index ${\n              index} twice because it was cleared after a previous read ` +\n          `(perhaps try setting clear_after_read = false?).`);\n    }\n\n    if (this.clearAfterRead) {\n      tensorWithState.cleared = true;\n    }\n\n    tensorWithState.read = true;\n    return tensorWithState.tensor;\n  }\n\n  /**\n   * Helper method to read multiple tensors from the specified indices.\n   */\n  readMany(indices: number[]): Tensor[] {\n    return indices.map(index => this.read(index));\n  }\n\n  /**\n   * Write value into the index of the TensorArray.\n   * @param index number the index to write to.\n   * @param tensor\n   */\n  write(index: number, tensor: Tensor) {\n    if (this.closed_) {\n      throw new Error(`TensorArray ${this.name} has already been closed.`);\n    }\n\n    if (index < 0 || !this.dynamicSize && index >= this.maxSize) {\n      throw new Error(`Tried to write to index ${\n          index}, but array is not resizeable and size is: ${this.maxSize}`);\n    }\n\n    const t = this.tensors[index] || {};\n\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray ${\n          this.name}: Could not write to TensorArray index ${index},\n          because the value dtype is ${\n          tensor.dtype}, but TensorArray dtype is ${this.dtype}.`);\n    }\n\n    // Set the shape for the first time write to unknow shape tensor array\n    if (this.size() === 0 &&\n        (this.elementShape == null || this.elementShape.length === 0)) {\n      this.elementShape = tensor.shape;\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape,\n        `TensorArray ${this.name}: Could not write to TensorArray index ${\n            index}.`);\n\n    if (t.read) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been read.`);\n    }\n\n    if (t.written) {\n      throw new Error(\n          `TensorArray ${this.name}: Could not write to TensorArray index ${\n              index}, because it has already been written.`);\n    }\n\n    t.tensor = tensor;\n    keep(tensor);\n    t.written = true;\n\n    this.tensors[index] = t;\n  }\n\n  /**\n   * Helper method to write multiple tensors to the specified indices.\n   */\n  writeMany(indices: number[], tensors: Tensor[]) {\n    if (indices.length !== tensors.length) {\n      throw new Error(\n          `TensorArray ${this.name}: could not write multiple tensors,` +\n          `because the index size: ${\n              indices.length} is not the same as tensors size: ${\n              tensors.length}.`);\n    }\n\n    indices.forEach((i, index) => this.write(i, tensors[index]));\n  }\n\n  /**\n   * Return selected values in the TensorArray as a packed Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param [indices] number[] Optional. Taking values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size(). If not specified returns\n   *    all tensors in the original order.\n   * @param [dtype]\n   */\n  gather(indices?: number[], dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but gather requested dtype ${dtype}`);\n    }\n\n    if (!indices) {\n      indices = [];\n      for (let i = 0; i < this.size(); i++) {\n        indices.push(i);\n      }\n    } else {\n      indices = indices.slice(0, this.size());\n    }\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    // Read all the PersistentTensors into a vector to keep track of\n    // their memory.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape, 'TensorArray shape mismatch: ');\n\n    return stack(tensors, 0);\n  }\n\n  /**\n   * Return the values in the TensorArray as a concatenated Tensor.\n   */\n  concat(dtype?: DataType): Tensor {\n    if (!!dtype && dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but concat requested dtype ${dtype}`);\n    }\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    const indices = [];\n    for (let i = 0; i < this.size(); i++) {\n      indices.push(i);\n    }\n    // Collect all the tensors from the tensors array.\n    const tensors = this.readMany(indices);\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensors[0].shape,\n        `TensorArray shape mismatch: tensor array shape (${\n            this.elementShape}) vs first tensor shape (${tensors[0].shape})`);\n\n    return concat(tensors, 0);\n  }\n\n  /**\n   * Scatter the values of a Tensor in specific indices of a TensorArray.\n   * @param indices nummber[] values in [0, max_value). If the\n   *    TensorArray is not dynamic, max_value=size().\n   * @param tensor Tensor input tensor.\n   */\n  scatter(indices: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n\n    if (indices.length !== tensor.shape[0]) {\n      throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n          indices.length} vs. ${tensor.shape[0]}`);\n    }\n\n    const maxIndex = Math.max(...indices);\n\n    if (!this.dynamicSize && maxIndex >= this.maxSize) {\n      throw new Error(\n          `Max index must be < array size (${maxIndex}  vs. ${this.maxSize})`);\n    }\n\n    this.writeMany(indices, unstack(tensor, 0));\n  }\n\n  /**\n   * Split the values of a Tensor into the TensorArray.\n   * @param length number[] with the lengths to use when splitting value along\n   *    its first dimension.\n   * @param tensor Tensor, the tensor to split.\n   */\n  split(length: number[], tensor: Tensor) {\n    if (tensor.dtype !== this.dtype) {\n      throw new Error(`TensorArray dtype is ${\n          this.dtype} but tensor has dtype ${tensor.dtype}`);\n    }\n    let totalLength = 0;\n    const cumulativeLengths = length.map(len => {\n      totalLength += len;\n      return totalLength;\n    });\n\n    if (totalLength !== tensor.shape[0]) {\n      throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n    }\n\n    if (!this.dynamicSize && length.length !== this.maxSize) {\n      throw new Error(\n          `TensorArray's size is not equal to the size of lengths (${\n              this.maxSize} vs. ${length.length}), ` +\n          'and the TensorArray is not marked as dynamically resizeable');\n    }\n\n    const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n    const tensors: Tensor[] = [];\n    tidy(() => {\n      tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n      for (let i = 0; i < length.length; ++i) {\n        const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n        const indices = [0, previousLength, 0];\n        const sizes = [1, length[i], elementPerRow];\n        tensors[i] = reshape(slice(tensor, indices, sizes), this.elementShape);\n      }\n      return tensors;\n    });\n    const indices = [];\n    for (let i = 0; i < length.length; i++) {\n      indices[i] = i;\n    }\n    this.writeMany(indices, tensors);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {concat, DataType, keep, reshape, scalar, slice, stack, Tensor, tensor, tidy, unstack} from '@tensorflow/tfjs-core';\n\nimport {assertShapesMatchAllowUndefinedSize} from './tensor_utils';\n\n/**\n * TensorList stores a container of `tf.Tensor` objects, which are accessible\n * via tensors field.\n *\n * In order to get a copy of the underlying list, use the copy method:\n * ```\n *    TensorList b = a.copy();\n *    b.tensors().pushBack(t);  // This does not modify a.tensors().\n * ```\n *\n * Note that this is not a deep copy: the memory locations of the underlying\n * tensors will still point to the same locations of the corresponding tensors\n * in the original.\n */\n\nexport class TensorList {\n  readonly idTensor: Tensor;\n  maxNumElements: number;\n\n  get id() {\n    return this.idTensor.id;\n  }\n  /**\n   *\n   * @param tensors list of tensors\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param maxNumElements The maximum allowed size of `tensors`. Defaults to -1\n   *   meaning that the size of `tensors` is unbounded.\n   */\n  constructor(\n      readonly tensors: Tensor[], readonly elementShape: number[],\n      readonly elementDtype: DataType, maxNumElements = -1) {\n    if (tensors != null) {\n      tensors.forEach(tensor => {\n        if (elementDtype !== tensor.dtype) {\n          throw new Error(`Invalid data types; op elements ${\n              elementDtype}, but list elements ${tensor.dtype}`);\n        }\n        assertShapesMatchAllowUndefinedSize(\n            elementShape, tensor.shape, 'TensorList shape mismatch: ');\n\n        keep(tensor);\n      });\n    }\n    this.idTensor = scalar(0);\n    this.maxNumElements = maxNumElements;\n    keep(this.idTensor);\n  }\n\n  /**\n   * Get a new TensorList containing a copy of the underlying tensor container.\n   */\n  copy(): TensorList {\n    return new TensorList(\n        [...this.tensors], this.elementShape, this.elementDtype);\n  }\n\n  /**\n   * Dispose the tensors and idTensor and clear the tensor list.\n   */\n  clearAndClose(keepIds?: Set<number>) {\n    this.tensors.forEach(tensor => {\n      if (keepIds == null || !keepIds.has(tensor.id)) {\n        tensor.dispose();\n      }\n    });\n    this.tensors.length = 0;\n    this.idTensor.dispose();\n  }\n  /**\n   * The size of the tensors in the tensor list.\n   */\n  size() {\n    return this.tensors.length;\n  }\n\n  /**\n   * Return a tensor that stacks a list of rank-R tf.Tensors into one rank-(R+1)\n   * tf.Tensor.\n   * @param elementShape shape of each tensor\n   * @param elementDtype data type of each tensor\n   * @param numElements the number of elements to stack\n   */\n  stack(elementShape: number[], elementDtype: DataType, numElements = -1):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (numElements !== -1 && this.tensors.length !== numElements) {\n      throw new Error(`Operation expected a list with ${\n          numElements} elements but got a list with ${\n          this.tensors.length} elements.`);\n    }\n    assertShapesMatchAllowUndefinedSize(\n        elementShape, this.elementShape, 'TensorList shape mismatch: ');\n    return tidy(() => {\n      const reshapedTensors =\n          this.tensors.map(tensor => reshape(tensor, elementShape));\n      return stack(reshapedTensors, 0);\n    });\n  }\n\n  /**\n   * Pop a tensor from the end of the list.\n   * @param elementShape shape of the tensor\n   * @param elementDtype data type of the tensor\n   */\n  popBack(elementShape: number[], elementDtype: DataType): Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (this.size() === 0) {\n      throw new Error('Trying to pop from an empty list.');\n    }\n\n    const tensor = this.tensors.pop();\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, elementShape, 'TensorList shape mismatch: ');\n    return reshape(tensor, elementShape);\n  }\n\n  /**\n   * Push a tensor to the end of the list.\n   * @param tensor Tensor to be pushed.\n   */\n  pushBack(tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        tensor.shape, this.elementShape, 'TensorList shape mismatch: ');\n\n    if (this.maxNumElements === this.size()) {\n      throw new Error(`Trying to push element into a full list.`);\n    }\n    keep(tensor);\n    this.tensors.push(tensor);\n  }\n\n  /**\n   * Update the size of the list.\n   * @param size the new size of the list.\n   */\n  resize(size: number) {\n    if (size < 0) {\n      throw new Error(\n          `TensorListResize expects size to be non-negative. Got: ${size}`);\n    }\n\n    if (this.maxNumElements !== -1 && size > this.maxNumElements) {\n      throw new Error(`TensorListResize input size ${\n          size} is greater maxNumElement ${this.maxNumElements}.`);\n    }\n    this.tensors.length = size;\n  }\n\n  /**\n   * Retrieve the element at the provided index\n   * @param elementShape shape of the tensor\n   * @param elementDtype dtype of the tensor\n   * @param elementIndex index of the tensor\n   */\n  getItem(elementIndex: number, elementShape: number[], elementDtype: DataType):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n    if (elementIndex < 0 || elementIndex > this.tensors.length) {\n      throw new Error(`Trying to access element ${\n          elementIndex} in a list with ${this.tensors.length} elements.`);\n    }\n\n    if (this.tensors[elementIndex] == null) {\n      throw new Error(`element at index ${elementIndex} is null.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.tensors[elementIndex].shape, elementShape,\n        'TensorList shape mismatch: ');\n\n    return this.tensors[elementIndex];\n  }\n\n  /**\n   * Set the tensor at the index\n   * @param elementIndex index of the tensor\n   * @param tensor the tensor to be inserted into the list\n   */\n  setItem(elementIndex: number, tensor: Tensor) {\n    if (tensor.dtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          tensor.dtype}, but list elements ${this.elementDtype}`);\n    }\n\n    if (elementIndex < 0 ||\n        this.maxNumElements !== -1 && elementIndex >= this.maxNumElements) {\n      throw new Error(`Trying to set element ${\n          elementIndex} in a list with max ${this.maxNumElements} elements.`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, tensor.shape, 'TensorList shape mismatch: ');\n    keep(tensor);\n    this.tensors[elementIndex] = tensor;\n  }\n\n  /**\n   * Return selected values in the TensorList as a stacked Tensor. All of\n   * selected values must have been written and their shapes must all match.\n   * @param indices indices of tensors to gather\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  gather(indices: number[], elementDtype: DataType, elementShape: number[]):\n      Tensor {\n    if (elementDtype !== this.elementDtype) {\n      throw new Error(`Invalid data types; op elements ${\n          elementDtype}, but list elements ${this.elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    // When indices is greater than the size of the list, indices beyond the\n    // size of the list are ignored.\n    indices = indices.slice(0, this.size());\n\n    if (indices.length === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = indices.map(i => reshape(this.tensors[i], elementShape));\n      return stack(tensors, 0);\n    });\n  }\n\n  /**\n   * Return the values in the TensorList as a concatenated Tensor.\n   * @param elementDtype output tensor dtype\n   * @param elementShape output tensor element shape\n   */\n  concat(elementDtype: DataType, elementShape: number[]): Tensor {\n    if (!!elementDtype && elementDtype !== this.elementDtype) {\n      throw new Error(`TensorList dtype is ${\n          this.elementDtype} but concat requested dtype ${elementDtype}`);\n    }\n\n    assertShapesMatchAllowUndefinedSize(\n        this.elementShape, elementShape, 'TensorList shape mismatch: ');\n\n    if (this.size() === 0) {\n      return tensor([], [0].concat(this.elementShape));\n    }\n\n    return tidy(() => {\n      const tensors = this.tensors.map(t => reshape(t, elementShape));\n      return concat(tensors, 0);\n    });\n  }\n}\n\n/**\n * Creates a TensorList which, when stacked, has the value of tensor.\n * @param tensor from tensor\n * @param elementShape output tensor element shape\n */\nexport function fromTensor(\n    tensor: Tensor, elementShape: number[], elementDtype: DataType) {\n  const dtype = tensor.dtype;\n  if (tensor.shape.length < 1) {\n    throw new Error(\n        `Tensor must be at least a vector, but saw shape: ${tensor.shape}`);\n  }\n  if (tensor.dtype !== elementDtype) {\n    throw new Error(`Invalid data types; op elements ${\n        tensor.dtype}, but list elements ${elementDtype}`);\n  }\n  const outputShape = tensor.shape.slice(1);\n  assertShapesMatchAllowUndefinedSize(\n      outputShape, elementShape, 'TensorList shape mismatch: ');\n\n  const tensorList: Tensor[] = unstack(tensor);\n  return new TensorList(tensorList, elementShape, dtype);\n}\n\n/**\n * Return a TensorList of the given size with empty elements.\n * @param elementShape the shape of the future elements of the list\n * @param elementDtype the desired type of elements in the list\n * @param numElements the number of elements to reserve\n */\nexport function reserve(\n    elementShape: number[], elementDtype: DataType, numElements: number) {\n  return new TensorList([], elementShape, elementDtype, numElements);\n}\n\n/**\n * Put tensors at specific indices of a stacked tensor into a TensorList.\n * @param indices list of indices on how to scatter the tensor.\n * @param tensor input tensor.\n * @param elementShape the shape of the future elements of the list\n * @param numElements the number of elements to scatter\n */\nexport function scatter(\n    tensor: Tensor, indices: number[], elementShape: number[],\n    numElements?: number): TensorList {\n  if (indices.length !== tensor.shape[0]) {\n    throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${\n        indices.length} vs. ${tensor.shape[0]}`);\n  }\n\n  const maxIndex = Math.max(...indices);\n\n  if (numElements != null && numElements !== -1 && maxIndex >= numElements) {\n    throw new Error(\n        `Max index must be < array size (${maxIndex}  vs. ${numElements})`);\n  }\n\n  const list = new TensorList([], elementShape, tensor.dtype, numElements);\n  const tensors = unstack(tensor, 0);\n  indices.forEach((value, index) => {\n    list.setItem(value, tensors[index]);\n  });\n  return list;\n}\n\n/**\n * Split the values of a Tensor into a TensorList.\n * @param length the lengths to use when splitting value along\n *    its first dimension.\n * @param tensor the tensor to split.\n * @param elementShape the shape of the future elements of the list\n */\nexport function split(\n    tensor: Tensor, length: number[], elementShape: number[]) {\n  let totalLength = 0;\n  const cumulativeLengths = length.map(len => {\n    totalLength += len;\n    return totalLength;\n  });\n\n  if (totalLength !== tensor.shape[0]) {\n    throw new Error(`Expected sum of lengths to be equal to\n          tensor.shape[0], but sum of lengths is\n        ${totalLength}, and tensor's shape is: ${tensor.shape}`);\n  }\n\n  const elementPerRow = totalLength === 0 ? 0 : tensor.size / totalLength;\n  const tensors: Tensor[] = tidy(() => {\n    const tensors = [];\n    tensor = reshape(tensor, [1, totalLength, elementPerRow]);\n    for (let i = 0; i < length.length; ++i) {\n      const previousLength = (i === 0) ? 0 : cumulativeLengths[i - 1];\n      const indices = [0, previousLength, 0];\n      const sizes = [1, length[i], elementPerRow];\n      tensors[i] = reshape(slice(tensor, indices, sizes), elementShape);\n    }\n    tensor.dispose();\n    return tensors;\n  });\n\n  const list = new TensorList([], elementShape, tensor.dtype, length.length);\n\n  for (let i = 0; i < tensors.length; i++) {\n    list.setItem(i, tensors[i]);\n  }\n  return list;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, scalar, Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {TensorArray} from '../../executor/tensor_array';\nimport {fromTensor, reserve, scatter, split} from '../../executor/tensor_list';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'If':\n    case 'StatelessIf': {\n      const thenFunc =\n          getParamValue('thenBranch', node, tensorMap, context) as string;\n      const elseFunc =\n          getParamValue('elseBranch', node, tensorMap, context) as string;\n      const cond = getParamValue('cond', node, tensorMap, context) as Tensor;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n      const condValue = await cond.data();\n      if (condValue[0]) {\n        return context.functionMap[thenFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      } else {\n        return context.functionMap[elseFunc].executeFunctionAsync(\n            args, context.tensorArrayMap, context.tensorListMap);\n      }\n    }\n    case 'While':\n    case 'StatelessWhile': {\n      const bodyFunc =\n          getParamValue('body', node, tensorMap, context) as string;\n      const condFunc =\n          getParamValue('cond', node, tensorMap, context) as string;\n      const args = getParamValue('args', node, tensorMap, context) as Tensor[];\n\n      // Calculate the condition of the loop\n      const condResult =\n          (await context.functionMap[condFunc].executeFunctionAsync(\n              args, context.tensorArrayMap, context.tensorListMap));\n      const argIds = args.map(tensor => tensor.id);\n      let condValue = await condResult[0].data();\n      // Dispose the intermediate tensors for condition function\n      condResult.forEach(tensor => {\n        if (!tensor.kept && argIds.indexOf(tensor.id) === -1) {\n          tensor.dispose();\n        }\n      });\n\n      let result: Tensor[] = args;\n\n      while (condValue[0]) {\n        // Record the previous result for intermediate tensor tracking\n        const origResult = result;\n        // Execution the body of the loop\n        result = await context.functionMap[bodyFunc].executeFunctionAsync(\n            result, context.tensorArrayMap, context.tensorListMap);\n        const resultIds = result.map(tensor => tensor.id);\n\n        // Dispose the intermediate tensor for body function that is not global\n        // kept, not input/output of the body function\n        origResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n\n        // Recalcuate the condition of the loop using the latest results.\n        const condResult =\n            (await context.functionMap[condFunc].executeFunctionAsync(\n                result, context.tensorArrayMap, context.tensorListMap));\n        condValue = await condResult[0].data();\n        // Dispose the intermediate tensors for condition function\n        condResult.forEach(tensor => {\n          if (!tensor.kept && argIds.indexOf(tensor.id) === -1 &&\n              resultIds.indexOf(tensor.id) === -1) {\n            tensor.dispose();\n          }\n        });\n      }\n      return result;\n    }\n    case 'LoopCond': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      return [cloneTensor(pred)];\n    }\n    case 'Switch': {\n      const pred = getParamValue('pred', node, tensorMap, context) as Tensor;\n      let data = getParamValue('data', node, tensorMap, context) as Tensor;\n      if (!data.kept) {\n        data = cloneTensor(data);\n      }\n      // Outputs nodes :0 => false, :1 => true\n      return (await pred.data())[0] ? [undefined, data] : [data, undefined];\n    }\n    case 'Merge': {\n      const inputName = node.inputNames.find(\n          name => getTensor(name, tensorMap, context) !== undefined);\n      if (inputName) {\n        const data = getTensor(inputName, tensorMap, context);\n        return [cloneTensor(data)];\n      }\n      return undefined;\n    }\n    case 'Enter': {\n      const frameId =\n          getParamValue('frameName', node, tensorMap, context) as string;\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.enterFrame(frameId);\n      return [cloneTensor(data)];\n    }\n    case 'Exit': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.exitFrame();\n      return [cloneTensor(data)];\n    }\n    case 'NextIteration': {\n      const data = getParamValue('tensor', node, tensorMap, context) as Tensor;\n      context.nextIteration();\n      return [cloneTensor(data)];\n    }\n    case 'TensorArrayV3': {\n      const size = getParamValue('size', node, tensorMap, context) as number;\n      const dtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const dynamicSize =\n          getParamValue('dynamicSize', node, tensorMap, context) as boolean;\n      const clearAfterRead =\n          getParamValue('clearAfterRead', node, tensorMap, context) as boolean;\n      const identicalElementShapes =\n          getParamValue('identicalElementShapes', node, tensorMap, context) as\n          boolean;\n      const name = getParamValue('name', node, tensorMap, context) as string;\n      const tensorArray = new TensorArray(\n          name, dtype, size, elementShape, identicalElementShapes, dynamicSize,\n          clearAfterRead);\n      context.addTensorArray(tensorArray);\n      return [tensorArray.idTensor, scalar(1.0)];\n    }\n    case 'TensorArrayWriteV3': {\n      const id =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const writeTensorArray = context.getTensorArray(id.id);\n      writeTensorArray.write(index, writeTensor);\n      return [writeTensorArray.idTensor];\n    }\n    case 'TensorArrayReadV3': {\n      const readId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const readTensorArray = context.getTensorArray(readId.id);\n      return [readTensorArray.read(readIndex)];\n    }\n    case 'TensorArrayGatherV3': {\n      const gatherId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const gatherDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const gatherTensorArray = context.getTensorArray(gatherId.id);\n      return [gatherTensorArray.gather(gatherIndices, gatherDtype)];\n    }\n    case 'TensorArrayScatterV3': {\n      const scatterId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const scatterTensorArray = context.getTensorArray(scatterId.id);\n      scatterTensorArray.scatter(scatterIndices, scatterTensor);\n      return [scatterTensorArray.idTensor];\n    }\n    case 'TensorArrayConcatV3': {\n      const concatId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const concatTensorArray = context.getTensorArray(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      return [concatTensorArray.concat(concatDtype)];\n    }\n    case 'TensorArraySplitV3': {\n      const splitId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n      const splitTensorArray = context.getTensorArray(splitId.id);\n      splitTensorArray.split(lengths, splitTensor);\n      return [splitTensorArray.idTensor];\n    }\n    case 'TensorArraySizeV3': {\n      const sizeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const sizeTensorArray = context.getTensorArray(sizeId.id);\n      return [scalar(sizeTensorArray.size(), 'int32')];\n    }\n    case 'TensorArrayCloseV3': {\n      const closeId =\n          getParamValue('tensorArrayId', node, tensorMap, context) as Tensor;\n      const closeTensorArray = context.getTensorArray(closeId.id);\n      closeTensorArray.clearAndClose();\n      return [closeTensorArray.idTensor];\n    }\n    case 'TensorListSetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const index = getParamValue('index', node, tensorMap, context) as number;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.setItem(index, writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGetItem': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const readIndex =\n          getParamValue('index', node, tensorMap, context) as number;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.getItem(readIndex, elementShape, elementDType)];\n    }\n    case 'TensorListScatterV2':\n    case 'TensorListScatter': {\n      const scatterIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const scatterTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList =\n          scatter(scatterTensor, scatterIndices, elementShape, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListReserve': {\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = reserve(elementShape, elementDtype, numElements);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListGather': {\n      const gatherId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const gatherIndices =\n          getParamValue('indices', node, tensorMap, context) as number[];\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(gatherId.id);\n      return [tensorList.gather(gatherIndices, elementDtype, elementShape)];\n    }\n    case 'TensorListStack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const numElements =\n          getParamValue('numElements', node, tensorMap, context) as number;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.stack(elementShape, elementDtype, numElements)];\n    }\n    case 'TensorListFromTensor': {\n      const tensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDtype =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = fromTensor(tensor, elementShape, elementDtype);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListConcat': {\n      const concatId =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(concatId.id);\n      const concatDtype =\n          getParamValue('dtype', node, tensorMap, context) as DataType;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      return [tensorList.concat(concatDtype, elementShape)];\n    }\n    case 'TensorListPushBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const writeTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const tensorList = context.getTensorList(idTensor.id);\n      tensorList.pushBack(writeTensor);\n      return [tensorList.idTensor];\n    }\n    case 'TensorListPopBack': {\n      const idTensor =\n          getParamValue('tensorListId', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const elementDType =\n          getParamValue('elementDType', node, tensorMap, context) as DataType;\n      const tensorList = context.getTensorList(idTensor.id);\n      return [tensorList.popBack(elementShape, elementDType)];\n    }\n    case 'TensorListSplit': {\n      const splitTensor =\n          getParamValue('tensor', node, tensorMap, context) as Tensor;\n      const elementShape =\n          getParamValue('elementShape', node, tensorMap, context) as number[];\n      const lengths =\n          getParamValue('lengths', node, tensorMap, context) as number[];\n\n      const tensorList = split(splitTensor, lengths, elementShape);\n      context.addTensorList(tensorList);\n      return [tensorList.idTensor];\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'control';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Rank, Tensor, Tensor3D, Tensor4D, Tensor5D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getPadding, getParamValue} from './utils';\n\nfunction fusedConvAndDepthWiseParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const [extraOp, activationFunc] =\n      (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n  const isBiasAdd = extraOp === 'biasadd';\n  const isPrelu = activationFunc === 'prelu';\n  const isBatchNorm = extraOp === 'fusedbatchnorm';\n\n  const numArgs =\n      (getParamValue('numArgs', node, tensorMap, context) as number);\n  if (isBiasAdd) {\n    if (isPrelu && numArgs !== 2) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu ' +\n          'must have two extra arguments: bias and alpha.');\n    }\n    if (!isPrelu && numArgs !== 1) {\n      throw new Error(\n          'FusedConv2d and DepthwiseConv2d with BiasAdd must have ' +\n          'one extra argument: bias.');\n    }\n  }\n  if (isBatchNorm) {\n    throw new Error(\n        'FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.');\n  }\n  const stride = getParamValue('strides', node, tensorMap, context) as number[];\n  const pad = getPadding(node, tensorMap, context);\n  const dataFormat =\n      (getParamValue('dataFormat', node, tensorMap, context) as string)\n          .toUpperCase();\n  const dilations =\n      getParamValue('dilations', node, tensorMap, context) as number[];\n  const [biasArg, preluArg] =\n      getParamValue('args', node, tensorMap, context) as Tensor[];\n\n  return {\n    stride,\n    pad,\n    dataFormat,\n    dilations,\n    biasArg,\n    preluArg,\n    activationFunc\n  };\n}\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Conv1D': {\n          const stride =\n              getParamValue('stride', node, tensorMap, context) as number;\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilation =\n              getParamValue('dilation', node, tensorMap, context) as number;\n          return [tfOps.conv1d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              stride, pad as 'valid' | 'same', dataFormat as 'NWC' | 'NCW',\n              dilation)];\n        }\n        case 'Conv2D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case '_FusedConv2D': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.conv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n\n        case 'FusedDepthwiseConv2dNative': {\n          const {\n            stride,\n            pad,\n            dataFormat,\n            dilations,\n            biasArg,\n            preluArg,\n            activationFunc\n          } = fusedConvAndDepthWiseParams(node, tensorMap, context);\n\n          return [tfOps.fused.depthwiseConv2d({\n            x: getParamValue('x', node, tensorMap, context) as Tensor3D |\n                Tensor4D,\n            filter: getParamValue('filter', node, tensorMap, context) as\n                Tensor4D,\n            strides: [stride[1], stride[2]],\n            pad: pad as 'valid' | 'same',\n            dataFormat: dataFormat as 'NHWC' | 'NCHW',\n            dilations: [dilations[1], dilations[2]],\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n        }\n        case 'Conv2DBackpropInput':\n        case 'Conv2dTranspose': {\n          const shape = getParamValue(\n                            'outputShape', node, tensorMap,\n                            context) as [number, number, number] |\n              [number, number, number, number];\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          return [tfOps.conv2dTranspose(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              shape, [stride[1], stride[2]], pad as 'valid' | 'same')];\n        }\n        case 'DepthwiseConv2dNative':\n        case 'DepthwiseConv2d': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getPadding(node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n\n          return [tfOps.depthwiseConv2d(\n              getParamValue('input', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor4D,\n              [stride[1], stride[2]], pad as 'valid' | 'same',\n              dataFormat as 'NHWC' | 'NCHW', [dilations[1], dilations[2]])];\n        }\n        case 'Conv3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as string)\n                  .toUpperCase();\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n          return [tfOps.conv3d(\n              getParamValue('x', node, tensorMap, context) as Tensor4D |\n                  Tensor<Rank.R5>,\n              getParamValue('filter', node, tensorMap, context) as\n                  Tensor<Rank.R5>,\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same',\n              dataFormat as 'NDHWC' | 'NCDHW',\n              [dilations[1], dilations[2], dilations[3]])];\n        }\n        case 'AvgPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPool': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same')];\n        }\n        case 'MaxPoolWithArgmax': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n          const includeBatchInIndex =\n              getParamValue('includeBatchInIndex', node, tensorMap, context) as\n              boolean;\n          const {result, indexes} = tfOps.maxPoolWithArgmax(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              [kernelSize[1], kernelSize[2]], [stride[1], stride[2]],\n              pad as 'valid' | 'same', includeBatchInIndex);\n          return [result, indexes];\n        }\n        case 'AvgPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.avgPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'MaxPool3D': {\n          const stride =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const kernelSize =\n              getParamValue('kernelSize', node, tensorMap, context) as number[];\n\n          return [tfOps.maxPool3d(\n              getParamValue('x', node, tensorMap, context) as Tensor5D,\n              [kernelSize[1], kernelSize[2], kernelSize[3]],\n              [stride[1], stride[2], stride[3]], pad as 'valid' | 'same')];\n        }\n\n        case 'Dilation2D': {\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const pad = getParamValue('pad', node, tensorMap, context);\n          const dilations =\n              getParamValue('dilations', node, tensorMap, context) as number[];\n\n          // strides: [1, stride_height, stride_width, 1].\n          const strideHeight = strides[1];\n          const strideWidth = strides[2];\n\n          // dilations: [1, dilation_height, dilation_width, 1].\n          const dilationHeight = dilations[1];\n          const dilationWidth = dilations[2];\n\n          return [tfOps.dilation2d(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('filter', node, tensorMap, context) as Tensor3D,\n              [strideHeight, strideWidth], pad as 'valid' | 'same',\n              [dilationHeight, dilationWidth], 'NHWC' /* dataFormat */)];\n        }\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'convolution';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, Tensor, Tensor1D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Fill': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const dtype =\n              getParamValue('dtype', node, tensorMap, context) as DataType;\n          const value =\n              getParamValue('value', node, tensorMap, context) as number;\n          return [tfOps.fill(shape, value, dtype)];\n        }\n        case 'LinSpace': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const num = getParamValue('num', node, tensorMap, context) as number;\n          return [tfOps.linspace(start, stop, num)];\n        }\n        case 'Multinomial': {\n          const logits =\n              getParamValue('logits', node, tensorMap, context) as Tensor1D;\n          const numSamples =\n              getParamValue('numSamples', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.multinomial(logits, numSamples, seed)];\n        }\n        case 'OneHot': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          const depth =\n              getParamValue('depth', node, tensorMap, context) as number;\n          const onValue =\n              getParamValue('onValue', node, tensorMap, context) as number;\n          const offValue =\n              getParamValue('offValue', node, tensorMap, context) as number;\n          return [tfOps.oneHot(indices, depth, onValue, offValue)];\n        }\n        case 'Ones': {\n          return [tfOps.ones(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'OnesLike': {\n          return [tfOps.onesLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'RandomUniform': {\n          return [tfOps.randomUniform(\n              // tslint:disable-next-line:no-any\n              getParamValue('shape', node, tensorMap, context) as any,\n              getParamValue('minval', node, tensorMap, context) as number,\n              getParamValue('maxval', node, tensorMap, context) as number,\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'Range': {\n          const start =\n              getParamValue('start', node, tensorMap, context) as number;\n          const stop =\n              getParamValue('stop', node, tensorMap, context) as number;\n          const step =\n              getParamValue('step', node, tensorMap, context) as number;\n          return [tfOps.range(\n              start, stop, step,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32')];\n        }\n        case 'TruncatedNormal': {\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          const mean =\n              getParamValue('mean', node, tensorMap, context) as number;\n          const stdDev =\n              getParamValue('stdDev', node, tensorMap, context) as number;\n          const seed =\n              getParamValue('seed', node, tensorMap, context) as number;\n          return [tfOps.truncatedNormal(\n              shape, mean, stdDev,\n              getParamValue('dtype', node, tensorMap, context) as 'float32' |\n                  'int32',\n              seed)];\n        }\n        case 'Zeros': {\n          return [tfOps.zeros(\n              getParamValue('shape', node, tensorMap, context) as number[],\n              getParamValue('dtype', node, tensorMap, context) as DataType)];\n        }\n        case 'ZerosLike': {\n          return [tfOps.zerosLike(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'creation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpAsyncExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nfunction nmsParams(\n    node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) {\n  const boxes = getParamValue('boxes', node, tensorMap, context) as Tensor;\n  const scores = getParamValue('scores', node, tensorMap, context) as Tensor;\n  const maxOutputSize =\n      getParamValue('maxOutputSize', node, tensorMap, context) as number;\n  const iouThreshold =\n      getParamValue('iouThreshold', node, tensorMap, context) as number;\n  const scoreThreshold =\n      getParamValue('scoreThreshold', node, tensorMap, context) as number;\n  const softNmsSigma =\n      getParamValue('softNmsSigma', node, tensorMap, context) as number;\n\n  return {\n    boxes,\n    scores,\n    maxOutputSize,\n    iouThreshold,\n    scoreThreshold,\n    softNmsSigma\n  };\n}\n\nexport const executeOp: InternalOpAsyncExecutor = async(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): Promise<Tensor[]> => {\n  switch (node.op) {\n    case 'NonMaxSuppressionV5': {\n      const {\n        boxes,\n        scores,\n        maxOutputSize,\n        iouThreshold,\n        scoreThreshold,\n        softNmsSigma\n      } = nmsParams(node, tensorMap, context);\n\n      const result = await tfOps.image.nonMaxSuppressionWithScoreAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, softNmsSigma);\n\n      return [result.selectedIndices, result.selectedScores];\n    }\n    case 'NonMaxSuppressionV4': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      const padToMaxOutputSize =\n          getParamValue('padToMaxOutputSize', node, tensorMap, context) as\n          boolean;\n\n      const result = await tfOps.image.nonMaxSuppressionPaddedAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold, padToMaxOutputSize);\n\n      return [result.selectedIndices, result.validOutputs];\n    }\n    case 'NonMaxSuppressionV3':\n    case 'NonMaxSuppressionV2': {\n      const {boxes, scores, maxOutputSize, iouThreshold, scoreThreshold} =\n          nmsParams(node, tensorMap, context);\n\n      return [await tfOps.image.nonMaxSuppressionAsync(\n          boxes as Tensor2D, scores as Tensor1D, maxOutputSize, iouThreshold,\n          scoreThreshold)];\n    }\n    case 'Where': {\n      const condition = tfOps.cast(\n          (getParamValue('condition', node, tensorMap, context) as Tensor),\n          'bool');\n      const result = [await tfOps.whereAsync(condition)];\n      condition.dispose();\n      return result;\n    }\n    case 'ListDiff': {\n      return tfOps.setdiff1dAsync(\n          getParamValue('x', node, tensorMap, context) as Tensor,\n          getParamValue('y', node, tensorMap, context) as Tensor);\n    }\n    default:\n      throw TypeError(`Node type ${node.op} is not implemented`);\n  }\n};\n\nexport const CATEGORY = 'dynamic';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'TopKV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const k = getParamValue('k', node, tensorMap, context) as number;\n              const sorted =\n                  getParamValue('sorted', node, tensorMap, context) as boolean;\n              const result = tfOps.topk(x, k, sorted);\n              return [result.values, result.indices];\n            }\n            case 'Unique': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const result = tfOps.unique(x);\n              return [result.values, result.indices];\n            }\n            case 'UniqueV2': {\n              const x = getParamValue('x', node, tensorMap, context) as Tensor;\n              const axis =\n                  getParamValue('axis', node, tensorMap, context) as number;\n              const result = tfOps.unique(x, axis);\n              return [result.values, result.indices];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'evaluation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {cloneTensor, getParamValue, getTensor} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Const': {\n          return tensorMap[node.name];\n        }\n        case 'PlaceholderWithDefault':\n          const def =\n              getParamValue('default', node, tensorMap, context) as Tensor;\n          return [getTensor(node.name, tensorMap, context) || def];\n        case 'Placeholder':\n          return [getTensor(node.name, tensorMap, context)];\n        case 'Identity':\n        case 'StopGradient':\n        case 'FakeQuantWithMinMaxVars': {  // This op is currently ignored.\n          const data = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [cloneTensor(data)];\n        }\n        case 'IdentityN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => cloneTensor(t));\n        case 'Snapshot':\n          const snapshot =\n              (getParamValue('x', node, tensorMap, context) as Tensor);\n          return [cloneTensor(snapshot)];\n        case 'Shape':\n          return [tfOps.tensor1d(\n              (getParamValue('x', node, tensorMap, context) as Tensor).shape,\n              'int32')];\n        case 'ShapeN':\n          return (getParamValue('x', node, tensorMap, context) as Tensor[])\n              .map((t: Tensor) => tfOps.tensor1d(t.shape));\n        case 'Size':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).size,\n              'int32')];\n        case 'Rank':\n          return [tfOps.scalar(\n              (getParamValue('x', node, tensorMap, context) as Tensor).rank,\n              'int32')];\n        case 'NoOp':\n          return [tfOps.scalar(1)];\n        case 'Print':\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const data =\n              getParamValue('data', node, tensorMap, context) as Tensor[];\n          const message =\n              getParamValue('message', node, tensorMap, context) as string;\n          const summarize =\n              getParamValue('summarize', node, tensorMap, context) as number;\n          console.warn(\n              'The graph has a tf.print() operation,' +\n              'usually used for debugging, which slows down performance.');\n          console.log(message);\n          for (let i = 0; i < data.length; i++) {\n            console.log(Array.prototype.slice.call(data[i].dataSync())\n                            .slice(0, summarize));\n          }\n          return [input];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'graph';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ResizeBilinear': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeBilinear(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'ResizeNearestNeighbor': {\n          const images =\n              getParamValue('images', node, tensorMap, context) as Tensor;\n          const size =\n              getParamValue('size', node, tensorMap, context) as number[];\n          const alignCorners =\n              getParamValue('alignCorners', node, tensorMap, context) as\n              boolean;\n          return [tfOps.image.resizeNearestNeighbor(\n              images as Tensor3D | Tensor4D, [size[0], size[1]], alignCorners)];\n        }\n        case 'CropAndResize': {\n          const image =\n              getParamValue('image', node, tensorMap, context) as Tensor;\n          const boxes =\n              getParamValue('boxes', node, tensorMap, context) as Tensor;\n          const boxInd =\n              getParamValue('boxInd', node, tensorMap, context) as Tensor;\n          const cropSize =\n              getParamValue('cropSize', node, tensorMap, context) as number[];\n          const method =\n              getParamValue('method', node, tensorMap, context) as string;\n          const extrapolationValue =\n              getParamValue('extrapolationValue', node, tensorMap, context) as\n              number;\n          return [tfOps.image.cropAndResize(\n              image as Tensor4D, boxes as Tensor2D, boxInd as Tensor1D,\n              cropSize as [number, number], method as 'bilinear' | 'nearest',\n              extrapolationValue)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'image';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Equal': {\n          return [tfOps.equal(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'NotEqual': {\n          return [tfOps.notEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Greater': {\n          return [tfOps.greater(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'GreaterEqual': {\n          return [tfOps.greaterEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Less': {\n          return [tfOps.less(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LessEqual': {\n          return [tfOps.lessEqual(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalAnd': {\n          return [tfOps.logicalAnd(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalNot': {\n          return [tfOps.logicalNot(\n              getParamValue('a', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogicalOr': {\n          return [tfOps.logicalOr(\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        case 'Select':\n        case 'SelectV2': {\n          return [tfOps.where(\n              getParamValue('condition', node, tensorMap, context) as Tensor,\n              getParamValue('a', node, tensorMap, context) as Tensor,\n              getParamValue('b', node, tensorMap, context) as Tensor)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'logical';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor2D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'BatchMatMul':\n        case 'BatchMatMulV2':\n        case 'MatMul':\n          return [tfOps.matMul(\n              getParamValue('a', node, tensorMap, context) as Tensor2D,\n              getParamValue('b', node, tensorMap, context) as Tensor2D,\n              getParamValue('transposeA', node, tensorMap, context) as boolean,\n              getParamValue('transposeB', node, tensorMap, context) as\n                  boolean)];\n\n        case 'Transpose':\n          return [tfOps.transpose(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('perm', node, tensorMap, context) as number[])];\n\n        case '_FusedMatMul':\n          const [extraOp, activationFunc] =\n              (getParamValue('fusedOps', node, tensorMap, context) as string[]);\n\n          const isBiasAdd = extraOp === 'biasadd';\n          const isPrelu = activationFunc === 'prelu';\n\n          const numArgs =\n              (getParamValue('numArgs', node, tensorMap, context) as number);\n          if (isBiasAdd) {\n            if (isPrelu && numArgs !== 2) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd and Prelu must have two ' +\n                  'extra arguments: bias and alpha.');\n            }\n            if (!isPrelu && numArgs !== 1) {\n              throw new Error(\n                  'Fused MatMul with BiasAdd must have one extra argument: bias.');\n            }\n          }\n          const [biasArg, preluArg] =\n              getParamValue('args', node, tensorMap, context) as Tensor[];\n          return [tfOps.fused.matMul({\n            a: getParamValue('a', node, tensorMap, context) as Tensor2D,\n            b: getParamValue('b', node, tensorMap, context) as Tensor2D,\n            transposeA: getParamValue('transposeA', node, tensorMap, context) as\n                boolean,\n            transposeB: getParamValue('transposeB', node, tensorMap, context) as\n                boolean,\n            bias: biasArg,\n            activation: activationFunc as tfOps.fused.Activation,\n            preluActivationWeights: preluArg\n          })];\n\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'matrices';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor3D, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'FusedBatchNorm':\n        case 'FusedBatchNormV2': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'FusedBatchNormV3': {\n          return [tfOps.batchNorm(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('mean', node, tensorMap, context) as Tensor,\n              getParamValue('variance', node, tensorMap, context) as Tensor,\n              getParamValue('offset', node, tensorMap, context) as Tensor,\n              getParamValue('scale', node, tensorMap, context) as Tensor,\n              getParamValue('epsilon', node, tensorMap, context) as number)];\n        }\n        case 'LRN': {\n          return [tfOps.localResponseNormalization(\n              getParamValue('x', node, tensorMap, context) as Tensor3D |\n                  Tensor4D,\n              getParamValue('radius', node, tensorMap, context) as number,\n              getParamValue('bias', node, tensorMap, context) as number,\n              getParamValue('alpha', node, tensorMap, context) as number,\n              getParamValue('beta', node, tensorMap, context) as number)];\n        }\n        case 'Softmax': {\n          return [tfOps.softmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'LogSoftmax': {\n          return [tfOps.logSoftmax(\n              getParamValue('x', node, tensorMap, context) as Tensor)];\n        }\n        case 'SparseToDense': {\n          return [tfOps.sparseToDense(\n              getParamValue('sparseIndices', node, tensorMap, context) as\n                  Tensor,\n              getParamValue('outputShape', node, tensorMap, context) as Tensor,\n              getParamValue('sparseValues', node, tensorMap, context) as\n                  number[],\n              getParamValue('defaultValue', node, tensorMap, context) as\n                  Scalar)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'normalization';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Max': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.max(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Mean': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.mean(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Min': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.min(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Sum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.sum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'All': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.all(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Any': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.any(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'ArgMax': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMax(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'ArgMin': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.argMin(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Prod': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const keepDims =\n              getParamValue('keepDims', node, tensorMap, context) as boolean;\n          return [tfOps.prod(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              keepDims)];\n        }\n        case 'Cumsum': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const exclusive =\n              getParamValue('exclusive', node, tensorMap, context) as boolean;\n          const reverse =\n              getParamValue('reverse', node, tensorMap, context) as boolean;\n          return [tfOps.cumsum(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis,\n              exclusive, reverse)];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'reduction';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Scalar, Tensor, Tensor1D, tidy, util} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'ConcatV2':\n        case 'Concat': {\n          const n = getParamValue('n', node, tensorMap, context) as number;\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          let inputs =\n              getParamValue('tensors', node, tensorMap, context) as Tensor[];\n          inputs = inputs.slice(0, n);\n          return [tfOps.concat(inputs, axis)];\n        }\n        case 'GatherV2':\n        case 'Gather': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor1D;\n          return [tfOps.gather(input, tfOps.cast(indices, 'int32'), axis)];\n        }\n        case 'ReverseV2':\n        case 'Reverse': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          const input = getParamValue('x', node, tensorMap, context) as Tensor;\n          return [tfOps.reverse(input, axis)];\n        }\n        case 'Slice': {\n          // tslint:disable-next-line:no-any\n          const begin = getParamValue('begin', node, tensorMap, context) as any;\n          // tslint:disable-next-line:no-any\n          const size = getParamValue('size', node, tensorMap, context) as any;\n          return [tfOps.slice(\n              getParamValue('x', node, tensorMap, context) as Tensor, begin,\n              size)];\n        }\n        case 'StridedSlice': {\n          const begin =\n              getParamValue('begin', node, tensorMap, context) as number[];\n          const end =\n              getParamValue('end', node, tensorMap, context) as number[];\n          const strides =\n              getParamValue('strides', node, tensorMap, context) as number[];\n          const beginMask =\n              getParamValue('beginMask', node, tensorMap, context) as number;\n          const endMask =\n              getParamValue('endMask', node, tensorMap, context) as number;\n          const ellipsisMask =\n              getParamValue('ellipsisMask', node, tensorMap, context) as number;\n          const newAxisMask =\n              getParamValue('newAxisMask', node, tensorMap, context) as number;\n          const shrinkAxisMask =\n              getParamValue('shrinkAxisMask', node, tensorMap, context) as\n              number;\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return [tfOps.stridedSlice(\n              tensor, begin, end, strides, beginMask, endMask, ellipsisMask,\n              newAxisMask, shrinkAxisMask)];\n        }\n        case 'Pack': {\n          return tidy(() => {\n            const axis =\n                getParamValue('axis', node, tensorMap, context) as number;\n            const tensors =\n                getParamValue('tensors', node, tensorMap, context) as Tensor[];\n            // Reshape the tensors to the first tensor's shape if they don't\n            // match.\n            const shape = tensors[0].shape;\n            const squeezedShape = tfOps.squeeze(tensors[0]).shape;\n            const mapped = tensors.map(tensor => {\n              const sameShape = util.arraysEqual(tensor.shape, shape);\n              if (!sameShape &&\n                  !util.arraysEqual(\n                      tfOps.squeeze(tensor).shape, squeezedShape)) {\n                throw new Error('the input tensors shape does not match');\n              }\n              return sameShape ? tensor : tfOps.reshape(tensor, shape);\n            });\n            return [tfOps.stack(mapped, axis)];\n          });\n        }\n        case 'Unpack': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const tensor =\n              getParamValue('tensor', node, tensorMap, context) as Tensor;\n          return tfOps.unstack(tensor, axis);\n        }\n        case 'Tile': {\n          const reps =\n              getParamValue('reps', node, tensorMap, context) as number[];\n          return [tfOps.tile(\n              getParamValue('x', node, tensorMap, context) as Tensor, reps)];\n        }\n        case 'Split':\n        case 'SplitV': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          const numOrSizeSplits =\n              getParamValue('numOrSizeSplits', node, tensorMap, context) as\n                  number |\n              number[];\n          const tensor = getParamValue('x', node, tensorMap, context) as Tensor;\n\n          return tfOps.split(tensor, numOrSizeSplits, axis);\n        }\n        case 'ScatterNd': {\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          const values =\n              getParamValue('values', node, tensorMap, context) as Tensor;\n          const shape =\n              getParamValue('shape', node, tensorMap, context) as number[];\n          return [tfOps.scatterND(indices, values, shape)];\n        }\n        case 'GatherNd': {\n          const x = getParamValue('x', node, tensorMap, context) as Tensor;\n          const indices =\n              getParamValue('indices', node, tensorMap, context) as Tensor;\n          return [tfOps.gatherND(x, indices)];\n        }\n        case 'SparseToDense': {\n          const indices =\n              getParamValue('sparseIndices', node, tensorMap, context) as\n              Tensor;\n          const shape =\n              getParamValue('outputShape', node, tensorMap, context) as\n              number[];\n          const sparseValues =\n              getParamValue('sparseValues', node, tensorMap, context) as Tensor;\n          const defaultValue =\n              getParamValue('defaultValue', node, tensorMap, context) as Scalar;\n          return [tfOps.sparseToDense(\n              indices, sparseValues, shape,\n              sparseValues.dtype === defaultValue.dtype ?\n                  defaultValue :\n                  tfOps.cast(defaultValue, sparseValues.dtype))];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'slice_join';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext):\n        Tensor[] => {\n          switch (node.op) {\n            case 'FFT': {\n              return [tfOps.fft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IFFT': {\n              return [tfOps.ifft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'RFFT': {\n              return [tfOps.rfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            case 'IRFFT': {\n              return [tfOps.irfft(\n                  getParamValue('x', node, tensorMap, context) as Tensor)];\n            }\n            default:\n              throw TypeError(`Node type ${node.op} is not implemented`);\n          }\n        };\n\nexport const CATEGORY = 'spectral';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor, Tensor4D} from '@tensorflow/tfjs-core';\n// tslint:disable-next-line: no-imports-from-dist\nimport * as tfOps from '@tensorflow/tfjs-core/dist/ops/ops_for_converter';\n\nimport {NamedTensorsMap} from '../../data/types';\nimport {ExecutionContext} from '../../executor/execution_context';\nimport {InternalOpExecutor, Node} from '../types';\n\nimport {getParamValue} from './utils';\n\nexport const executeOp: InternalOpExecutor =\n    (node: Node, tensorMap: NamedTensorsMap,\n     context: ExecutionContext): Tensor[] => {\n      switch (node.op) {\n        case 'Cast': {\n          return [tfOps.cast(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('dtype', node, tensorMap, context) as 'int32' |\n                  'float32' | 'bool')];\n        }\n        case 'ExpandDims': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number;\n          return [tfOps.expandDims(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n        case 'Squeeze': {\n          const axis =\n              getParamValue('axis', node, tensorMap, context) as number[];\n          return [tfOps.squeeze(\n              getParamValue('x', node, tensorMap, context) as Tensor, axis)];\n        }\n\n        case 'Reshape': {\n          return [tfOps.reshape(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        case 'PadV2':\n        case 'Pad': {\n          return [tfOps.pad(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('padding', node, tensorMap, context) as\n                  Array<[number, number]>,\n              getParamValue('constantValue', node, tensorMap, context) as\n                  number)];\n        }\n        case 'SpaceToBatchND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const paddings =\n              getParamValue('paddings', node, tensorMap, context) as number[][];\n          return [tfOps.spaceToBatchND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, paddings)];\n        }\n        case 'BatchToSpaceND': {\n          const blockShape =\n              getParamValue('blockShape', node, tensorMap, context) as number[];\n          const crops =\n              getParamValue('crops', node, tensorMap, context) as number[][];\n          return [tfOps.batchToSpaceND(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              blockShape, crops)];\n        }\n        case 'DepthToSpace': {\n          const blockSize =\n              getParamValue('blockSize', node, tensorMap, context) as number;\n          const dataFormat =\n              (getParamValue('dataFormat', node, tensorMap, context) as\n               string).toUpperCase() as 'NHWC' |\n              'NCHW';\n          return [tfOps.depthToSpace(\n              getParamValue('x', node, tensorMap, context) as Tensor4D,\n              blockSize, dataFormat)];\n        }\n        case 'BroadcastTo': {\n          return [tfOps.broadcastTo(\n              getParamValue('x', node, tensorMap, context) as Tensor,\n              getParamValue('shape', node, tensorMap, context) as number[])];\n        }\n        default:\n          throw TypeError(`Node type ${node.op} is not implemented`);\n      }\n    };\n\nexport const CATEGORY = 'transformation';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tfc from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {ExecutionContext} from '../executor/execution_context';\n\nimport {NodeValueImpl} from './custom_op/node_value_impl';\nimport {getRegisteredOp} from './custom_op/register';\nimport * as arithmetic from './executors/arithmetic_executor';\nimport * as basicMath from './executors/basic_math_executor';\nimport * as control from './executors/control_executor';\nimport * as convolution from './executors/convolution_executor';\nimport * as creation from './executors/creation_executor';\nimport * as dynamic from './executors/dynamic_executor';\nimport * as evaluation from './executors/evaluation_executor';\nimport * as graph from './executors/graph_executor';\nimport * as image from './executors/image_executor';\nimport * as logical from './executors/logical_executor';\nimport * as matrices from './executors/matrices_executor';\nimport * as normalization from './executors/normalization_executor';\nimport * as reduction from './executors/reduction_executor';\nimport * as sliceJoin from './executors/slice_join_executor';\nimport * as spectral from './executors/spectral_executor';\nimport * as transformation from './executors/transformation_executor';\nimport {Node} from './types';\n\n/**\n * Executes the op defined by the node object.\n * @param node\n * @param tensorMap contains tensors for executed nodes and weights\n */\nexport function executeOp(\n    node: Node, tensorMap: NamedTensorsMap,\n    context: ExecutionContext): tfc.Tensor[]|Promise<tfc.Tensor[]> {\n  const value =\n      ((node: Node, tensorMap: NamedTensorsMap, context: ExecutionContext) => {\n        switch (node.category) {\n          case 'arithmetic':\n            return tfc.tidy(\n                () => arithmetic.executeOp(node, tensorMap, context));\n          case 'basic_math':\n            return tfc.tidy(\n                () => basicMath.executeOp(node, tensorMap, context));\n          case 'control':\n            return control.executeOp(node, tensorMap, context);\n          case 'convolution':\n            return tfc.tidy(\n                () => convolution.executeOp(node, tensorMap, context));\n          case 'creation':\n            return tfc.tidy(() => creation.executeOp(node, tensorMap, context));\n          case 'dynamic':\n            return dynamic.executeOp(node, tensorMap, context);\n          case 'evaluation':\n            return tfc.tidy(\n                () => evaluation.executeOp(node, tensorMap, context));\n          case 'image':\n            return tfc.tidy(() => image.executeOp(node, tensorMap, context));\n          case 'graph':\n            return tfc.tidy(() => graph.executeOp(node, tensorMap, context));\n          case 'logical':\n            return tfc.tidy(() => logical.executeOp(node, tensorMap, context));\n          case 'matrices':\n            return tfc.tidy(() => matrices.executeOp(node, tensorMap, context));\n          case 'normalization':\n            return tfc.tidy(\n                () => normalization.executeOp(node, tensorMap, context));\n          case 'reduction':\n            return tfc.tidy(\n                () => reduction.executeOp(node, tensorMap, context));\n          case 'slice_join':\n            return tfc.tidy(\n                () => sliceJoin.executeOp(node, tensorMap, context));\n          case 'spectral':\n            return tfc.tidy(() => spectral.executeOp(node, tensorMap, context));\n          case 'transformation':\n            return tfc.tidy(\n                () => transformation.executeOp(node, tensorMap, context));\n          case 'custom':\n            const opMapper = getRegisteredOp(node.op);\n            if (opMapper && opMapper.customExecutor) {\n              return opMapper.customExecutor(\n                  new NodeValueImpl(node, tensorMap, context));\n            } else {\n              throw TypeError(`Custom op ${node.op} is not registered.`);\n            }\n          default:\n            throw TypeError(\n                `Unknown op '${node.op}'. File an issue at ` +\n                `https://github.com/tensorflow/tfjs/issues so we can add it` +\n                `, or register a custom execution with tf.registerOp()`);\n        }\n      })(node, tensorMap, context);\n  if (value instanceof Promise) {\n    return value.then((data) => [].concat(data));\n  }\n  return [].concat(value);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {Tensor} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap, TensorArrayMap, TensorListMap} from '../data/types';\n\nimport {TensorArray} from './tensor_array';\nimport {TensorList} from './tensor_list';\nimport {FunctionExecutor} from './types';\n\nexport interface ExecutionContextInfo {\n  id: number;           // the unique id of the context info\n  frameName: string;    // The frame name of the loop, this comes from\n                        // the TensorFlow NodeDef.\n  iterationId: number;  // The iteration id of the loop\n}\n\n/**\n * ExecutionContext captures the runtime environment of the node. It keeps\n * track of the current frame and iteration for the control flow ops.\n *\n * For example, typical Dynamic RNN model may contain loops, for which\n * TensorFlow will generate graphs with Enter/Exit nodes to control the\n * current execution frame, and NextIteration Nodes for iteration id increment.\n * For model with branch logic, TensorFLow will generate Switch/Merge ops.\n */\nexport class ExecutionContext {\n  private rootContext = {id: 0, frameName: '', iterationId: 0};\n  private contexts: ExecutionContextInfo[] = [this.rootContext];\n  private lastId = 0;\n  private _currentContextIds: string[];\n\n  constructor(\n      readonly weightMap: NamedTensorsMap = {},\n      readonly tensorArrayMap: TensorArrayMap = {},\n      readonly tensorListMap: TensorListMap = {},\n      readonly functionMap: {[key: string]: FunctionExecutor} = {}) {\n    this.generateCurrentContextIds();\n  }\n\n  private newFrame(id: number, frameName: string) {\n    return {id, frameName, iterationId: 0};\n  }\n\n  /**\n   * Set the current context\n   * @param contexts: ExecutionContextInfo[] the current path of execution\n   * frames\n   */\n  set currentContext(contexts: ExecutionContextInfo[]) {\n    if (this.contexts !== contexts) {\n      this.contexts = contexts;\n      this.generateCurrentContextIds();\n    }\n  }\n\n  get currentContext(): ExecutionContextInfo[] {\n    return this.contexts;\n  }\n\n  /**\n   * Returns the current context in string format.\n   */\n  get currentContextId(): string {\n    return this._currentContextIds[0];\n  }\n\n  /**\n   * Returns the current context and all parent contexts in string format.\n   * This allow access to the nodes in the current and parent frames.\n   */\n  get currentContextIds(): string[] {\n    return this._currentContextIds;\n  }\n\n  private generateCurrentContextIds() {\n    const names = [];\n    for (let i = 0; i < this.contexts.length - 1; i++) {\n      const contexts = this.contexts.slice(0, this.contexts.length - i);\n      names.push(this.contextIdforContexts(contexts));\n    }\n    names.push('');\n    this._currentContextIds = names;\n  }\n\n  private contextIdforContexts(contexts: ExecutionContextInfo[]) {\n    return contexts ?\n        contexts\n            .map(\n                context => (context.id === 0 && context.iterationId === 0) ?\n                    '' :\n                    `${context.frameName}-${context.iterationId}`)\n            .join('/') :\n        '';\n  }\n\n  /**\n   * Enter a new frame, a new context is pushed on the current context list.\n   * @param frameId new frame id\n   */\n  enterFrame(frameId: string) {\n    if (this.contexts) {\n      this.lastId++;\n      this.contexts = this.contexts.slice();\n      this.contexts.push(this.newFrame(this.lastId, frameId));\n      this._currentContextIds.unshift(this.contextIdforContexts(this.contexts));\n    }\n  }\n\n  /**\n   * Exit the current frame, the last context is removed from the current\n   * context list.\n   */\n  exitFrame() {\n    if (this.contexts && this.contexts.length > 1) {\n      this.contexts = this.contexts.slice();\n      this.contexts.splice(-1);\n      this.currentContextIds.shift();\n    } else {\n      throw new Error('Cannot exit frame, the context is empty');\n    }\n  }\n\n  /**\n   * Enter the next iteration of a loop, the iteration id of last context is\n   * increased.\n   */\n  nextIteration() {\n    if (this.contexts && this.contexts.length > 0) {\n      this.contexts = this.contexts.slice();\n      this.lastId++;\n      const context =\n          Object.assign({}, this.contexts[this.contexts.length - 1]);\n      context.iterationId += 1;\n      context.id = this.lastId;\n      this.contexts.splice(-1, 1, context);\n      this._currentContextIds.splice(\n          0, 1, this.contextIdforContexts(this.contexts));\n    } else {\n      throw new Error('Cannot increase frame iteration, the context is empty');\n    }\n  }\n\n  getWeight(name: string): Tensor[] {\n    return this.weightMap[name];\n  }\n\n  addTensorArray(tensorArray: TensorArray) {\n    this.tensorArrayMap[tensorArray.id] = tensorArray;\n  }\n\n  getTensorArray(id: number): TensorArray {\n    return this.tensorArrayMap[id];\n  }\n\n  addTensorList(tensorList: TensorList) {\n    this.tensorListMap[tensorList.id] = tensorList;\n  }\n\n  getTensorList(id: number): TensorList {\n    return this.tensorListMap[id];\n  }\n\n  dispose(keepIds: Set<number>) {\n    for (const key in this.tensorArrayMap) {\n      this.tensorArrayMap[key].clearAndClose(keepIds);\n    }\n\n    for (const key in this.tensorListMap) {\n      this.tensorListMap[key].clearAndClose(keepIds);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NamedTensorMap} from '@tensorflow/tfjs-core';\n\nimport {NamedTensorsMap} from '../data/types';\nimport {parseNodeName} from '../operations/executors/utils';\nimport {Graph, Node} from '../operations/types';\n\nexport interface ExecutionInfo {\n  inputs: NamedTensorMap;\n  outputs: Node[];\n  usedNodes: Set<string>;\n  missingInputs: string[];\n  dynamicNode: Node;\n  syncInputs: string[];\n}\n\n/**\n * Given graph inputs and desired outputs, find the minimal set of nodes\n * to execute in order to compute the outputs. In addition return other useful\n * info such:\n * - Missing inputs needed to compute the output.\n * - Whether the subgraph contains dynamic ops (control flow, dynamic shape).\n * - Alternative inputs in order to avoid async (dynamic op) execution.\n */\nexport function getExecutionSubgraph(\n    inputs: NamedTensorMap, outputs: Node[], weightMap: NamedTensorsMap,\n    initNodes?: Node[]): ExecutionInfo {\n  const usedNodes = new Set<string>();\n  const missingInputs: string[] = [];\n  let dynamicNode: Node = null;\n  let syncInputs: string[] = null;\n\n  // Start with the outputs, going backwards and find all the nodes that are\n  // needed to compute those outputs.\n  const seen = new Set<string>();\n  const inputNodeNames =\n      Object.keys(inputs).map(name => parseNodeName(name)[0]);\n\n  let initNodeNames: string[] = [];\n  if (initNodes != null) {\n    initNodeNames = initNodes.map(node => parseNodeName(node.name)[0]);\n  }\n\n  const frontier = [...outputs];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    if (isControlFlow(node) || isDynamicShape(node)) {\n      if (dynamicNode == null) {\n        dynamicNode = node;\n        syncInputs = dynamicNode.children.map(child => child.name)\n                         .filter(name => usedNodes.has(name));\n      }\n    }\n    usedNodes.add(node.name);\n\n    // Weights are dead end since we already have their values.\n    if (weightMap[node.name] != null) {\n      continue;\n    }\n    // This node is a dead end since it's one of the user-provided inputs.\n    if (inputNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    // This node is a dead end since it doesn't have any inputs.\n    if (initNodeNames.indexOf(node.name) !== -1) {\n      continue;\n    }\n    if (node.inputs.length === 0) {\n      missingInputs.push(node.name);\n      continue;\n    }\n    node.inputs.forEach(input => {\n      // Don't add to the frontier if it is already there.\n      if (seen.has(input.name)) {\n        return;\n      }\n      seen.add(input.name);\n      frontier.push(input);\n    });\n  }\n  return {inputs, outputs, usedNodes, missingInputs, dynamicNode, syncInputs};\n}\n\n/**\n * Given the execution info, return a list of nodes in topological order that\n * need to be executed to compute the output.\n */\nexport function getNodesInTopologicalOrder(\n    graph: Graph, weightMap: NamedTensorsMap,\n    executionInfo: ExecutionInfo): Node[] {\n  const {usedNodes, inputs} = executionInfo;\n  const frontier: Node[] = [];\n  const inputNodes = Object.keys(inputs)\n                         .map(name => parseNodeName(name)[0])\n                         .map(name => graph.nodes[name]);\n  const initNodes = graph.initNodes;\n\n  inputNodes.forEach(input => {\n    if (usedNodes.has(input.name)) {\n      frontier.push(input);\n    }\n  });\n  graph.weights.forEach(weight => {\n    if (usedNodes.has(weight.name)) {\n      frontier.push(weight);\n    }\n  });\n  if (initNodes != null) {\n    initNodes.forEach(node => {\n      if (usedNodes.has(node.name)) {\n        frontier.push(node);\n      }\n    });\n  }\n  const seen = new Set<string>();\n  const orderedNodes: Node[] = [];\n  while (frontier.length > 0) {\n    const node = frontier.pop();\n    seen.add(node.name);\n    if (!weightMap[node.name]) {\n      orderedNodes.push(node);\n    }\n    node.children.forEach(child => {\n      if (!seen.has(child.name) && usedNodes.has(child.name) &&\n          child.inputs.every(input => seen.has(input.name))) {\n        frontier.push(child);\n      }\n    });\n  }\n  return orderedNodes;\n}\n\nconst CONTROL_FLOW_OPS = [\n  'Switch', 'Merge', 'Enter', 'Exit', 'NextIteration', 'StatelessIf',\n  'StatelessWhile', 'if', 'While'\n];\nconst DYNAMIC_SHAPE_OPS = [\n  'NonMaxSuppressionV2', 'NonMaxSuppressionV3', 'NonMaxSuppressionV5', 'Where'\n];\n\nexport function isControlFlow(node: Node) {\n  return CONTROL_FLOW_OPS.indexOf(node.op) >= 0;\n}\n\nexport function isDynamicShape(node: Node) {\n  return DYNAMIC_SHAPE_OPS.indexOf(node.op) >= 0;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NamedTensorMap, Tensor, tidy, util} from '@tensorflow/tfjs-core';\n\nimport {ISignatureDef} from '../data/compiled_api';\nimport {NamedTensorsMap, TensorArrayMap, TensorInfo, TensorListMap} from '../data/types';\nimport {getNodeNameAndIndex, getParamValue, getTensor, getTensorsForCurrentContenxt, parseNodeName} from '../operations/executors/utils';\nimport {executeOp} from '../operations/operation_executor';\nimport {Graph, Node} from '../operations/types';\n\nimport {ExecutionContext, ExecutionContextInfo} from './execution_context';\nimport {getExecutionSubgraph, getNodesInTopologicalOrder, isControlFlow} from './model_analysis';\nimport {FunctionExecutor} from './types';\n\ninterface NodeWithContexts {\n  contexts: ExecutionContextInfo[];\n  node: Node;\n}\n\nexport class GraphExecutor implements FunctionExecutor {\n  private compiledMap: Map<string, Node[]> = new Map();\n  private _weightMap: NamedTensorsMap = {};\n  private _weightIds: number[];\n  private _signature: ISignatureDef;\n  private _inputs: Node[];\n  private _outputs: Node[];\n  private _initNodes: Node[];\n  private SEPERATOR = ',';\n  private _functions: {[key: string]: Graph} = {};\n  private _functionExecutorMap: {[key: string]: FunctionExecutor} = {};\n\n  get weightIds(): number[] {\n    return this.parent ? this.parent.weightIds : this._weightIds;\n  }\n\n  get functionExecutorMap(): {[key: string]: FunctionExecutor} {\n    return this.parent ? this.parent.functionExecutorMap :\n                         this._functionExecutorMap;\n  }\n\n  get weightMap(): NamedTensorsMap {\n    return this.parent ? this.parent.weightMap : this._weightMap;\n  }\n\n  set weightMap(weightMap: NamedTensorsMap) {\n    const weightIds = Object.keys(weightMap).map(\n        key => weightMap[key].map(tensor => tensor.id));\n    this._weightIds = [].concat(...weightIds);\n    this._weightMap = weightMap;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this._inputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get outputs(): TensorInfo[] {\n    return this._outputs.map(node => {\n      return {\n        name: node.name,\n        shape: node.attrParams['shape'] ?\n            node.attrParams['shape'].value as number[] :\n            undefined,\n        dtype: node.attrParams['dtype'] ?\n            node.attrParams['dtype'].value as DataType :\n            undefined\n      };\n    });\n  }\n\n  get inputNodes(): string[] {\n    return this._inputs.map(node => node.signatureKey || node.name);\n  }\n\n  get outputNodes(): string[] {\n    return this._outputs.map((node) => {\n      const name = node.signatureKey || node.name;\n      return node.defaultOutput ? (`${name}:${node.defaultOutput}`) : name;\n    });\n  }\n\n  get functions(): {[key: string]: ISignatureDef} {\n    return Object.keys(this._functions).reduce((map, key) => {\n      map[key] = this._functions[key].signature;\n      return map;\n    }, {} as {[key: string]: ISignatureDef});\n  }\n\n  /**\n   *\n   * @param graph Graph the model or function graph to be executed.\n   * @param parent When building function exector you need to set the parent\n   * executor. Since the weights and function executor maps are set at parant\n   * level, that function executor can access the function maps and weight maps\n   * through the parent.\n   */\n  constructor(private graph: Graph, private parent?: GraphExecutor) {\n    this._outputs = graph.outputs;\n    this._inputs = graph.inputs;\n    this._initNodes = graph.initNodes;\n    this._signature = graph.signature;\n    this._functions = graph.functions;\n    // create sub-graph executors\n    if (graph.functions != null) {\n      Object.keys(graph.functions).forEach(name => {\n        this._functionExecutorMap[name] =\n            new GraphExecutor(graph.functions[name], this);\n      });\n    }\n  }\n\n  private getCompilationKey(inputs: Node[], outputs: Node[]): string {\n    const sortedInputs = inputs.map(node => node.name).sort();\n    const sortedOutputs = outputs.map(node => node.name).sort();\n    return sortedInputs.join(this.SEPERATOR) + '--' +\n        sortedOutputs.join(this.SEPERATOR);\n  }\n\n  /**\n   * Compiles the inference graph and returns the minimal set of nodes that are\n   * required for execution, in the correct execution order.\n   */\n  private compile(inputs: NamedTensorMap, outputs: Node[]): Node[] {\n    const executionInfo =\n        getExecutionSubgraph(inputs, outputs, this.weightMap, this._initNodes);\n    const {missingInputs, dynamicNode, syncInputs} = executionInfo;\n    if (dynamicNode != null) {\n      throw new Error(\n          `This execution contains the node '${dynamicNode.name}', which has ` +\n          `the dynamic op '${dynamicNode.op}'. Please use ` +\n          `model.executeAsync() instead. Alternatively, to avoid the ` +\n          `dynamic ops, specify the inputs [${syncInputs}]`);\n    }\n\n    if (missingInputs.length > 0) {\n      const outNames = outputs.map(n => n.name);\n      const inNames = Object.keys(inputs);\n      throw new Error(\n          `Cannot compute the outputs [${outNames}] from the provided inputs ` +\n          `[${inNames}]. Missing the following inputs: [${missingInputs}]`);\n    }\n\n    return getNodesInTopologicalOrder(\n        this.graph, this.weightMap, executionInfo);\n  }\n\n  /**\n   * Executes the inference for given input tensors.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  execute(inputs: NamedTensorMap, outputs: string[]): Tensor[] {\n    inputs = this.mapInputs(inputs);\n    const names = Object.keys(inputs).sort();\n    this.checkInputs(inputs);\n    this.checkInputShapeAndType(inputs);\n    outputs = this.mapOutputs(outputs);\n    this.checkOutputs(outputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputs.map(name => parseNodeName(name)[0]);\n    let outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n\n    // If no outputs are specified, then use the default outputs of the model.\n    if (outputNodes.length === 0) {\n      outputNodes = this._outputs;\n    }\n\n    const compilationKey = this.getCompilationKey(inputNodes, outputNodes);\n\n    // Do nothing if the compiled graph cache contains the input.\n    let orderedNodes = this.compiledMap.get(compilationKey);\n    if (orderedNodes == null) {\n      orderedNodes = this.compile(inputs, outputNodes);\n      this.compiledMap.set(compilationKey, orderedNodes);\n    }\n\n    const tensorArrayMap: TensorArrayMap = {};\n    const tensorListMap: TensorListMap = {};\n    return tidy(() => {\n      const context = new ExecutionContext(\n          this.weightMap, tensorArrayMap, tensorListMap,\n          this.functionExecutorMap);\n      const tensorsMap: NamedTensorsMap = {...this.weightMap};\n\n      Object.keys(inputs).forEach(name => {\n        const [nodeName, index] = parseNodeName(name);\n        const tensors: Tensor[] = [];\n        tensors[index] = inputs[name];\n        tensorsMap[nodeName] = tensors;\n      });\n\n      const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n      const intermediateTensorConsumerCount: {[key: number]: number} = {};\n      for (let i = 0; i < orderedNodes.length; i++) {\n        const node = orderedNodes[i];\n        if (!tensorsMap[node.name]) {\n          const tensors = executeOp(node, tensorsMap, context) as Tensor[];\n          if (tensors instanceof Promise) {\n            throw new Error(\n                `The execution of the op '${node.op}' returned a promise. ` +\n                `Please use model.executeAsync() instead.`);\n          }\n          tensorsMap[node.name] = tensors;\n          this.checkTensorForDisposal(\n              node.name, node, tensorsMap, context, tensorsToKeep,\n              outputNodeNames, intermediateTensorConsumerCount);\n        }\n      }\n      // dispose the context for the root executor\n      if (this.parent == null) {\n        context.dispose(tensorsToKeep);\n      }\n      return outputs.map(name => getTensor(name, tensorsMap, context));\n    });\n  }\n\n  private getFrozenTensorIds(tensorMap: NamedTensorsMap): Set<number> {\n    const ids = [].concat.apply(\n        [],\n        Object.keys(tensorMap)\n            .map(key => tensorMap[key])\n            .map(tensors => tensors.map(tensor => tensor.id)));\n    return new Set(ids);\n  }\n  private checkTensorForDisposal(\n      nodeName: string, node: Node, tensorMap: NamedTensorsMap,\n      context: ExecutionContext, tensorsToKeep: Set<number>,\n      outputNames: string[],\n      intermediateTensorConsumerCount: {[key: string]: number}) {\n    // Skip output nodes and any control flow nodes, since its dependency is\n    // tricky to track correctly.\n    if (node.category === 'control' || outputNames.indexOf(nodeName) !== -1) {\n      return;\n    }\n\n    tensorMap[nodeName].forEach(tensor => {\n      if (tensor != null) {\n        intermediateTensorConsumerCount[tensor.id] =\n            (intermediateTensorConsumerCount[tensor.id] || 0) +\n            node.children.length;\n      }\n    });\n    node.inputs.forEach(input => {\n      // Skip any control flow nodes, since its dependency is tricky to track\n      // correctly.\n      if (input.category !== 'control') {\n        const tensors =\n            getTensorsForCurrentContenxt(input.name, tensorMap, context);\n        if (tensors != null) {\n          tensors.forEach(tensor => {\n            if (tensor && !tensorsToKeep.has(tensor.id)) {\n              const count = intermediateTensorConsumerCount[tensor.id];\n              if (count === 1) {\n                tensor.dispose();\n                delete intermediateTensorConsumerCount[tensor.id];\n              } else if (count != null) {\n                // only intermediate nodes has count set, inputs and weights are\n                // not.\n                intermediateTensorConsumerCount[tensor.id]--;\n              }\n            }\n          });\n        }\n      }\n    });\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   */\n  async executeAsync(inputs: NamedTensorMap, outputs: string[]):\n      Promise<Tensor[]> {\n    return this._executeAsync(inputs, outputs);\n  }\n\n  /**\n   * Executes the inference for given input tensors in Async fashion.\n   * @param inputs Tensor map for the model inputs, keyed by the input node\n   * names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   * @param isFunctionExecution Flag for executing a function.\n   * @param tensorArrayMap Optional, global TensorArray map by id. Used for\n   * function execution.\n   * @param tensorArrayMap Optinal global TensorList map by id. Used for\n   * function execution.\n   */\n  private async _executeAsync(\n      inputs: NamedTensorMap, outputs: string[], isFunctionExecution = false,\n      tensorArrayMap: TensorArrayMap = {},\n      tensorListMap: TensorListMap = {}): Promise<Tensor[]> {\n    if (!isFunctionExecution) {\n      inputs = this.mapInputs(inputs);\n      this.checkInputs(inputs);\n      this.checkInputShapeAndType(inputs);\n      outputs = this.mapOutputs(outputs);\n      this.checkOutputs(outputs);\n    }\n\n    const context = new ExecutionContext(\n        this.weightMap, tensorArrayMap, tensorListMap,\n        this.functionExecutorMap);\n\n    // Graph with control flow op requires runtime evaluation of the execution\n    // order, while without control flow the execution order is pre-determined\n    // in the compile method.\n    const tensorMap = await this.executeWithControlFlow(\n        inputs, context, outputs, isFunctionExecution);\n    const results = outputs.map(name => getTensor(name, tensorMap, context));\n\n    // dispose all the intermediate tensors\n    const outputIds = results.map(t => t.id);\n    const inputIds = Object.keys(inputs).map(name => inputs[name].id);\n    const keepIds =\n        new Set<number>([...outputIds, ...inputIds, ...this.weightIds]);\n    Object.keys(tensorMap).forEach(key => {\n      const tensorArray = tensorMap[key];\n      tensorArray.forEach(tensor => {\n        if (tensor && !tensor.isDisposed && !keepIds.has(tensor.id)) {\n          tensor.dispose();\n        }\n      });\n    });\n    // dispose the context for the root executor\n    if (this.parent == null) {\n      context.dispose(keepIds);\n    }\n\n    return results;\n  }\n\n  async executeFunctionAsync(\n      inputs: Tensor[], tensorArrayMap: TensorArrayMap,\n      tensorListMap: TensorListMap): Promise<Tensor[]> {\n    const mappedInputs = inputs.reduce((map, tensor, index) => {\n      map[this.inputs[index].name] = tensor;\n      return map;\n    }, {} as NamedTensorMap);\n\n    return this._executeAsync(\n        mappedInputs, this.outputNodes, true, tensorArrayMap, tensorListMap);\n  }\n  /**\n   * When there are control flow nodes in the graph, the graph execution use\n   * ExecutionContext to keep track of the frames and loop iterators.\n   * @param inputs placeholder tensors for the graph.\n   * @param context the execution context object for current execution.\n   * @param isFunctionExecution Flag for executing a function.\n   */\n  private async executeWithControlFlow(\n      inputs: NamedTensorMap, context: ExecutionContext, outputNames: string[],\n      isFunctionExecution: boolean): Promise<NamedTensorsMap> {\n    const names = Object.keys(inputs);\n    const inputNodes =\n        names.map(name => this.graph.nodes[parseNodeName(name)[0]]);\n    const outputNodeNames = outputNames.map(name => parseNodeName(name)[0]);\n    const outputNodes = outputNodeNames.map(name => this.graph.nodes[name]);\n    const {usedNodes, missingInputs, dynamicNode, syncInputs} =\n        getExecutionSubgraph(inputs, outputNodes, this.weightMap);\n\n    const stack: NodeWithContexts[] =\n        [...inputNodes, ...this.graph.weights].map(node => {\n          return {node, contexts: context.currentContext};\n        });\n    const tensorsMap: NamedTensorsMap = {...this.weightMap};\n    Object.keys(inputs).forEach(name => {\n      const [nodeName, index] = parseNodeName(name);\n      const tensors: Tensor[] = [];\n      tensors[index] = inputs[name];\n      tensorsMap[nodeName] = tensors;\n    });\n    const intermediateTensorConsumerCount: {[key: number]: number} = {};\n    const tensorsToKeep = this.getFrozenTensorIds(tensorsMap);\n    const added: {[key: string]: boolean} = {};\n    while (stack.length > 0) {\n      const promises = this.processStack(\n          inputNodes, stack, context, tensorsMap, added, tensorsToKeep,\n          outputNodeNames, intermediateTensorConsumerCount, usedNodes);\n      await Promise.all(promises);\n    }\n    if (dynamicNode == null && !isFunctionExecution) {\n      console.warn(\n          `This model execution did not contain any nodes with control flow ` +\n          `or dynamic output shapes. You can use model.execute() instead.`);\n    }\n    const missingOutputs =\n        outputNodes\n            .filter(\n                node => !isControlFlow(node) &&\n                    !getTensor(node.name, tensorsMap, context))\n            .map(node => node.name);\n    if (missingOutputs.length > 0) {\n      let alternativeMsg = '';\n      if (dynamicNode != null) {\n        alternativeMsg =\n            `Alternatively, to avoid the dynamic ops, use model.execute() ` +\n            `and specify the inputs [${syncInputs}]`;\n      }\n      throw new Error(\n          `Cannot compute the outputs [${missingOutputs}] from the provided ` +\n          `inputs [${names}]. Consider providing the following inputs: ` +\n          `[${missingInputs}]. ${alternativeMsg}`);\n    }\n    return tensorsMap;\n  }\n\n  private processStack(\n      inputNodes: Node[], stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      tensorsToKeep: Set<number>, outputNames: string[],\n      intermediateTensorConsumerCount: {[key: number]: number},\n      usedNodes: Set<string>) {\n    const promises: Array<Promise<Tensor[]>> = [];\n    while (stack.length > 0) {\n      const item = stack.pop();\n      context.currentContext = item.contexts;\n      let nodeName = '';\n      // The tensor of the Enter op with isConstant set should be set\n      // in the parent scope, so it will be available as constant for the\n      // whole loop.\n      if (item.node.op === 'Enter' &&\n          getParamValue('isConstant', item.node, tensorMap, context)) {\n        [nodeName] = getNodeNameAndIndex(item.node.name, context);\n      }\n\n      // only process nodes that are not provided as input nodes.\n      if (inputNodes.indexOf(item.node) === -1) {\n        const tensors = executeOp(item.node, tensorMap, context);\n        if (!nodeName) {\n          [nodeName] = getNodeNameAndIndex(item.node.name, context);\n        }\n        const currentContext = context.currentContext;\n        if (tensors instanceof Promise) {\n          promises.push(tensors.then(t => {\n            tensorMap[nodeName] = t;\n            context.currentContext = currentContext;\n            this.checkTensorForDisposal(\n                nodeName, item.node, tensorMap, context, tensorsToKeep,\n                outputNames, intermediateTensorConsumerCount);\n            this.processChildNodes(\n                item.node, stack, context, tensorMap, added, usedNodes);\n            return t;\n          }));\n        } else {\n          tensorMap[nodeName] = tensors;\n          this.checkTensorForDisposal(\n              nodeName, item.node, tensorMap, context, tensorsToKeep,\n              outputNames, intermediateTensorConsumerCount);\n          this.processChildNodes(\n              item.node, stack, context, tensorMap, added, usedNodes);\n        }\n      } else {\n        this.processChildNodes(\n            item.node, stack, context, tensorMap, added, usedNodes);\n      }\n    }\n    return promises;\n  }\n\n  private processChildNodes(\n      node: Node, stack: NodeWithContexts[], context: ExecutionContext,\n      tensorMap: NamedTensorsMap, added: {[key: string]: boolean},\n      usedNodes: Set<string>) {\n    node.children.forEach((childNode) => {\n      const [nodeName, ] = getNodeNameAndIndex(childNode.name, context);\n      if (added[nodeName] || !usedNodes.has(childNode.name)) {\n        return;\n      }\n      // Merge op can be pushed if any of its inputs has value.\n      if (childNode.op === 'Merge') {\n        if (childNode.inputNames.some(name => {\n              return !!getTensor(name, tensorMap, context);\n            })) {\n          added[nodeName] = true;\n          stack.push({contexts: context.currentContext, node: childNode});\n        }\n      } else  // Otherwise all inputs must to have value.\n          if (childNode.inputNames.every(name => {\n                return !!getTensor(name, tensorMap, context);\n              })) {\n        added[nodeName] = true;\n        stack.push({contexts: context.currentContext, node: childNode});\n      }\n    });\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   */\n  dispose() {\n    Object.keys(this.weightMap)\n        .forEach(\n            key => this.weightMap[key].forEach(tensor => tensor.dispose()));\n  }\n\n  private checkInputShapeAndType(inputs: NamedTensorMap) {\n    Object.keys(inputs).forEach(name => {\n      const input = inputs[name];\n      const [nodeName, ] = parseNodeName(name);\n      const node = this.graph.nodes[nodeName];\n      if (node.attrParams['shape'] && node.attrParams['shape'].value) {\n        const shape = node.attrParams['shape'].value as number[];\n        const match = shape.length === input.shape.length &&\n            input.shape.every(\n                (dim, index) => shape[index] === -1 || shape[index] === dim);\n        util.assert(\n            match,\n            () => `The shape of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be [${shape}], but was ` +\n                `[${input.shape}]`);\n      }\n      if (node.attrParams['dtype'] && node.attrParams['dtype'].value) {\n        util.assert(\n            input.dtype === node.attrParams['dtype'].value as string,\n            () => `The dtype of dict['${node.name}'] provided in ` +\n                `model.execute(dict) must be ` +\n                `${node.attrParams['dtype'].value}, but was ${input.dtype}`);\n      }\n    });\n  }\n\n  private mapInputs(inputs: NamedTensorMap) {\n    const result: NamedTensorMap = {};\n    for (const inputName in inputs) {\n      if (this._signature != null && this._signature.inputs != null &&\n          this._signature.inputs[inputName] != null) {\n        const tensor = this._signature.inputs[inputName];\n        result[tensor.name] = inputs[inputName];\n      } else {\n        result[inputName] = inputs[inputName];\n      }\n    }\n    return result;\n  }\n\n  private checkInputs(inputs: NamedTensorMap) {\n    const notInGraph = Object.keys(inputs).filter(name => {\n      const [nodeName] = parseNodeName(name);\n      return this.graph.nodes[nodeName] == null;\n    });\n    if (notInGraph.length > 0) {\n      throw new Error(\n          `The dict provided in model.execute(dict) has ` +\n          `keys: [${notInGraph}] that are not part of graph`);\n    }\n  }\n\n  private mapOutputs(outputs: string[]) {\n    return outputs.map(name => {\n      if (this._signature != null && this._signature.outputs != null &&\n          this._signature.outputs[name] != null) {\n        const tensor = this._signature.outputs[name];\n        return tensor.name;\n      }\n      return name;\n    }, {});\n  }\n\n  private checkOutputs(outputs: string[]): void {\n    outputs.forEach(name => {\n      const [normalizedName] = parseNodeName(name);\n      if (!this.graph.nodes[normalizedName]) {\n        throw new Error(`The output '${name}' is not found in the graph`);\n      }\n    });\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {InferenceModel, io, ModelPredictConfig, NamedTensorMap, Tensor} from '@tensorflow/tfjs-core';\n\nimport * as tensorflow from '../data/compiled_api';\nimport {NamedTensorsMap, TensorInfo} from '../data/types';\nimport {OperationMapper} from '../operations/operation_mapper';\n\nimport {GraphExecutor} from './graph_executor';\n\nexport const TFHUB_SEARCH_PARAM = '?tfjs-format=file';\nexport const DEFAULT_MODEL_NAME = 'model.json';\n/**\n * A `tf.GraphModel` is a directed, acyclic graph built from a\n * SavedModel GraphDef and allows inference execution.\n *\n * A `tf.GraphModel` can only be created by loading from a model converted from\n * a [TensorFlow SavedModel](https://www.tensorflow.org/guide/saved_model) using\n * the command line converter tool and loaded via `tf.loadGraphModel`.\n *\n * @doc {heading: 'Models', subheading: 'Classes'}\n */\nexport class GraphModel implements InferenceModel {\n  private executor: GraphExecutor;\n  private version = 'n/a';\n  private handler: io.IOHandler;\n  private artifacts: io.ModelArtifacts;\n  private initializer: GraphExecutor;\n\n  // Returns the version information for the tensorflow model GraphDef.\n  get modelVersion(): string {\n    return this.version;\n  }\n\n  get inputNodes(): string[] {\n    return this.executor.inputNodes;\n  }\n\n  get outputNodes(): string[] {\n    return this.executor.outputNodes;\n  }\n\n  get inputs(): TensorInfo[] {\n    return this.executor.inputs;\n  }\n\n  get outputs(): TensorInfo[] {\n    return this.executor.outputs;\n  }\n\n  get weights(): NamedTensorsMap {\n    return this.executor.weightMap;\n  }\n\n  /**\n   * @param modelUrl url for the model, or an `io.IOHandler`.\n   * @param weightManifestUrl url for the weight file generated by\n   * scripts/convert.py script.\n   * @param requestOption options for Request, which allows to send credentials\n   * and custom headers.\n   * @param onProgress Optional, progress callback function, fired periodically\n   * before the load is completed.\n   */\n  constructor(\n      private modelUrl: string|io.IOHandler,\n      private loadOptions: io.LoadOptions = {}) {\n    if (loadOptions == null) {\n      this.loadOptions = {};\n    }\n  }\n\n  private findIOHandler() {\n    const path = this.modelUrl;\n    if ((path as io.IOHandler).load != null) {\n      // Path is an IO Handler.\n      this.handler = path as io.IOHandler;\n    } else if (this.loadOptions.requestInit != null) {\n      this.handler = io.browserHTTPRequest(path as string, this.loadOptions);\n    } else {\n      const handlers = io.getLoadHandlers(path as string, this.loadOptions);\n      if (handlers.length === 0) {\n        // For backward compatibility: if no load handler can be found,\n        // assume it is a relative http path.\n        handlers.push(io.browserHTTPRequest(path as string, this.loadOptions));\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) load handlers for ` +\n            `URL '${[path]}'`);\n      }\n      this.handler = handlers[0];\n    }\n  }\n\n  /**\n   * Loads the model and weight files, construct the in memory weight map and\n   * compile the inference graph.\n   */\n  async load(): Promise<boolean> {\n    this.findIOHandler();\n    if (this.handler.load == null) {\n      throw new Error(\n          'Cannot proceed with model loading because the IOHandler provided ' +\n          'does not have the `load` method implemented.');\n    }\n    const artifacts = await this.handler.load();\n\n    return this.loadSync(artifacts);\n  }\n\n  /**\n   * Synchronously construct the in memory weight map and\n   * compile the inference graph. Also initialize hashtable if any.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  loadSync(artifacts: io.ModelArtifacts) {\n    this.artifacts = artifacts;\n    const graph = this.artifacts.modelTopology as tensorflow.IGraphDef;\n    let signature = {};\n    if (this.artifacts.userDefinedMetadata != null) {\n      signature =  // tslint:disable-next-line:no-any\n          (this.artifacts.userDefinedMetadata as any).signature as\n          tensorflow.ISignatureDef;\n    }\n\n    this.version = `${graph.versions.producer}.${graph.versions.minConsumer}`;\n    const weightMap =\n        io.decodeWeights(this.artifacts.weightData, this.artifacts.weightSpecs);\n    this.executor = new GraphExecutor(\n        OperationMapper.Instance.transformGraph(graph, signature));\n    this.executor.weightMap = this.convertTensorMapToTensorsMap(weightMap);\n\n    if (artifacts.modelInitializer != null) {\n      const initializer =\n          OperationMapper.Instance.transformGraph(artifacts.modelInitializer);\n      this.initializer = new GraphExecutor(initializer);\n      this.initializer.weightMap = this.executor.weightMap;\n      this.initializer.execute({}, []);\n    }\n\n    return true;\n  }\n\n  /**\n   * Save the configuration and/or weights of the GraphModel.\n   *\n   * An `IOHandler` is an object that has a `save` method of the proper\n   * signature defined. The `save` method manages the storing or\n   * transmission of serialized data (\"artifacts\") that represent the\n   * model's topology and weights onto or via a specific medium, such as\n   * file downloads, local storage, IndexedDB in the web browser and HTTP\n   * requests to a server. TensorFlow.js provides `IOHandler`\n   * implementations for a number of frequently used saving mediums, such as\n   * `tf.io.browserDownloads` and `tf.io.browserLocalStorage`. See `tf.io`\n   * for more details.\n   *\n   * This method also allows you to refer to certain types of `IOHandler`s\n   * as URL-like string shortcuts, such as 'localstorage://' and\n   * 'indexeddb://'.\n   *\n   * Example 1: Save `model`'s topology and weights to browser [local\n   * storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage);\n   * then load it back.\n   *\n   * ```js\n   * const modelUrl =\n   *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n   * const model = await tf.loadGraphModel(modelUrl);\n   * const zeros = tf.zeros([1, 224, 224, 3]);\n   * model.predict(zeros).print();\n   *\n   * const saveResults = await model.save('localstorage://my-model-1');\n   *\n   * const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');\n   * console.log('Prediction from loaded model:');\n   * model.predict(zeros).print();\n   * ```\n   *\n   * @param handlerOrURL An instance of `IOHandler` or a URL-like,\n   * scheme-based string shortcut for `IOHandler`.\n   * @param config Options for saving the model.\n   * @returns A `Promise` of `SaveResult`, which summarizes the result of\n   * the saving, such as byte sizes of the saved artifacts for the model's\n   *   topology and weight values.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes', ignoreCI: true}\n   */\n  async save(handlerOrURL: io.IOHandler|string, config?: io.SaveConfig):\n      Promise<io.SaveResult> {\n    if (typeof handlerOrURL === 'string') {\n      const handlers = io.getSaveHandlers(handlerOrURL);\n      if (handlers.length === 0) {\n        throw new Error(\n            `Cannot find any save handlers for URL '${handlerOrURL}'`);\n      } else if (handlers.length > 1) {\n        throw new Error(\n            `Found more than one (${handlers.length}) save handlers for ` +\n            `URL '${handlerOrURL}'`);\n      }\n      handlerOrURL = handlers[0];\n    }\n    if (handlerOrURL.save == null) {\n      throw new Error(\n          'GraphModel.save() cannot proceed because the IOHandler ' +\n          'provided does not have the `save` attribute defined.');\n    }\n\n    return handlerOrURL.save(this.artifacts);\n  }\n\n  /**\n   * Execute the inference for the input tensors.\n   *\n   * @param input The input tensors, when there is single input for the model,\n   * inputs param should be a `tf.Tensor`. For models with mutliple inputs,\n   * inputs params should be in either `tf.Tensor`[] if the input order is\n   * fixed, or otherwise NamedTensorMap format.\n   *\n   * For model with multiple inputs, we recommend you use NamedTensorMap as the\n   * input type, if you use `tf.Tensor`[], the order of the array needs to\n   * follow the\n   * order of inputNodes array. @see {@link GraphModel.inputNodes}\n   *\n   * You can also feed any intermediate nodes using the NamedTensorMap as the\n   * input type. For example, given the graph\n   *    InputNode => Intermediate => OutputNode,\n   * you can execute the subgraph Intermediate => OutputNode by calling\n   *    model.execute('IntermediateNode' : tf.tensor(...));\n   *\n   * This is useful for models that uses tf.dynamic_rnn, where the intermediate\n   * state needs to be fed manually.\n   *\n   * For batch inference execution, the tensors for each input need to be\n   * concatenated together. For example with mobilenet, the required input shape\n   * is [1, 244, 244, 3], which represents the [batch, height, width, channel].\n   * If we are provide a batched data of 100 images, the input tensor should be\n   * in the shape of [100, 244, 244, 3].\n   *\n   * @param config Prediction configuration for specifying the batch size and\n   * output node names. Currently the batch size option is ignored for graph\n   * model.\n   *\n   * @returns Inference result tensors. The output would be single `tf.Tensor`\n   * if model has single output node, otherwise Tensor[] or NamedTensorMap[]\n   * will be returned for model with multiple outputs.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  predict(inputs: Tensor|Tensor[]|NamedTensorMap, config?: ModelPredictConfig):\n      Tensor|Tensor[]|NamedTensorMap {\n    return this.execute(inputs, this.outputNodes);\n  }\n\n  private normalizeInputs(inputs: Tensor|Tensor[]|\n                          NamedTensorMap): NamedTensorMap {\n    if (!(inputs instanceof Tensor) && !Array.isArray(inputs)) {\n      // The input is already a NamedTensorMap.\n      return inputs;\n    }\n    inputs = Array.isArray(inputs) ? inputs : [inputs];\n    if (inputs.length !== this.inputNodes.length) {\n      throw new Error(\n          'Input tensor count mismatch,' +\n          `the graph model has ${this.inputNodes.length} placeholders, ` +\n          `while there are ${inputs.length} input tensors.`);\n    }\n    return this.inputNodes.reduce((map, inputName, i) => {\n      map[inputName] = (inputs as Tensor[])[i];\n      return map;\n    }, {} as NamedTensorMap);\n  }\n\n  private normalizeOutputs(outputs: string|string[]): string[] {\n    outputs = outputs || this.outputNodes;\n    return !Array.isArray(outputs) ? [outputs] : outputs;\n  }\n\n  /**\n   * Executes inference for the model for given input tensors.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no\n   * outputs are specified, the default outputs of the model would be used.\n   * You can inspect intermediate nodes of the model by adding them to the\n   * outputs array.\n   *\n   * @returns A single tensor if provided with a single output or no outputs\n   * are provided and there is only one default output, otherwise return a\n   * tensor array. The order of the tensor array is the same as the outputs\n   * if provided, otherwise the order of outputNodes attribute of the model.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  execute(inputs: Tensor|Tensor[]|NamedTensorMap, outputs?: string|string[]):\n      Tensor|Tensor[] {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = this.executor.execute(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n  /**\n   * Executes inference for the model for given input tensors in async\n   * fashion, use this method when your model contains control flow ops.\n   * @param inputs tensor, tensor array or tensor map of the inputs for the\n   * model, keyed by the input node names.\n   * @param outputs output node name from the Tensorflow model, if no outputs\n   * are specified, the default outputs of the model would be used. You can\n   * inspect intermediate nodes of the model by adding them to the outputs\n   * array.\n   *\n   * @returns A Promise of single tensor if provided with a single output or\n   * no outputs are provided and there is only one default output, otherwise\n   * return a tensor map.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  async executeAsync(\n      inputs: Tensor|Tensor[]|NamedTensorMap,\n      outputs?: string|string[]): Promise<Tensor|Tensor[]> {\n    inputs = this.normalizeInputs(inputs);\n    outputs = this.normalizeOutputs(outputs);\n    const result = await this.executor.executeAsync(inputs, outputs);\n    return result.length > 1 ? result : result[0];\n  }\n\n  private convertTensorMapToTensorsMap(map: NamedTensorMap): NamedTensorsMap {\n    return Object.keys(map).reduce((newMap: NamedTensorsMap, key) => {\n      newMap[key] = [map[key]];\n      return newMap;\n    }, {});\n  }\n\n  /**\n   * Releases the memory used by the weight tensors.\n   *\n   * @doc {heading: 'Models', subheading: 'Classes'}\n   */\n  dispose() {\n    this.executor.dispose();\n\n    if (this.initializer) {\n      this.initializer.dispose();\n    }\n  }\n}\n\n/**\n * Load a graph model given a URL to the model definition.\n *\n * Example of loading MobileNetV2 from a URL and making a prediction with a\n * zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';\n * const model = await tf.loadGraphModel(modelUrl);\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n *\n * Example of loading MobileNetV2 from a TF Hub URL and making a prediction with\n * a zeros input:\n *\n * ```js\n * const modelUrl =\n *    'https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/2';\n * const model = await tf.loadGraphModel(modelUrl, {fromTFHub: true});\n * const zeros = tf.zeros([1, 224, 224, 3]);\n * model.predict(zeros).print();\n * ```\n * @param modelUrl The url or an `io.IOHandler` that loads the model.\n * @param options Options for the HTTP request, which allows to send credentials\n *    and custom headers.\n *\n * @doc {heading: 'Models', subheading: 'Loading'}\n */\nexport async function loadGraphModel(\n    modelUrl: string|io.IOHandler,\n    options: io.LoadOptions = {}): Promise<GraphModel> {\n  if (modelUrl == null) {\n    throw new Error(\n        'modelUrl in loadGraphModel() cannot be null. Please provide a url ' +\n        'or an IOHandler that loads the model');\n  }\n  if (options == null) {\n    options = {};\n  }\n\n  if (options.fromTFHub) {\n    if ((modelUrl as io.IOHandler).load == null) {\n      if (!(modelUrl as string).endsWith('/')) {\n        modelUrl = (modelUrl as string) + '/';\n      }\n      modelUrl = `${modelUrl}${DEFAULT_MODEL_NAME}${TFHUB_SEARCH_PARAM}`;\n    }\n  }\n  const model = new GraphModel(modelUrl, options);\n  await model.load();\n  return model;\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\n\n// tslint:disable:no-any\n\n/**\n * A return value for a mapping function that can be applied via deepMap.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapResult = {\n  value: any,\n  recurse: boolean\n};\n\n/**\n * Apply a mapping function to a nested structure in a recursive manner.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapResult`.  The `DeepMapResult` either provides a\n *   replacement value for that node (i.e., replacing the subtree), or indicates\n *   that the node should be processed recursively.\n */\nexport function deepMap(input: any, mapFn: (x: any) => DeepMapResult): any|\n    any[] {\n  return deepMapInternal(input, mapFn);\n}\n\n/**\n * @param seen: A Map of known object mappings (i.e., memoized results of\n *   `mapFn()`)\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepMapInternal(\n    input: any, mapFn: (x: any) => DeepMapResult,\n    seen: Map<any, any> = new Map(), containedIn: Set<{}> = new Set()): any|\n    any[] {\n  if (input == null) {\n    return null;\n  }\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  if (seen.has(input)) {\n    return seen.get(input);\n  }\n  const result = mapFn(input);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep map function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    seen.set(input, result.value);\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const child = input[k];\n      const childResult = deepMapInternal(child, mapFn, seen, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// TODO(soergel, kangyizhang) Reconsider naming of deepZip() to avoid confusion\n// with zip()\n\n/**\n * Zip nested structures together in a recursive manner.\n *\n * This has the effect of transposing or pivoting data, e.g. converting it from\n * a row-major representation to a column-major representation.\n *\n * For example, `deepZip([{a: 1, b: 2}, {a: 3, b: 4}])` returns\n * `{a: [1, 3], b: [2, 4]}`.\n *\n * The inputs should all have the same nested structure (i.e., of arrays and\n * dicts).  The result is a single object with the same nested structure, where\n * the leaves are arrays collecting the values of the inputs at that location\n * (or, optionally, the result of a custom function applied to those arrays).\n *\n * @param inputs: An array of the objects to zip together.\n * @param zipFn: (optional) A function that expects an array of elements at a\n *   single node of the object tree, and returns a `DeepMapResult`.  The\n *   `DeepMapResult` either provides a result value for that node (i.e.,\n *   representing the subtree), or indicates that the node should be processed\n *   recursively.  The default zipFn recurses as far as possible and places\n *   arrays at the leaves.\n */\nexport function deepZip(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult = zipToList): any|any[] {\n  return deepZipInternal(inputs, zipFn);\n}\n\n/**\n * @param containedIn: An set containing objects on the reference path currently\n *   being processed (used to detect cycles).\n */\nfunction deepZipInternal(\n    inputs: any[], zipFn: (xs: any[]) => DeepMapResult,\n    containedIn: Set<{}> = new Set()): any|any[] {\n  // The recursion follows the structure of input 0; it's assumed that all the\n  // other inputs have the same structure.\n  const input = inputs[0];\n  if (containedIn.has(input)) {\n    throw new Error('Circular references are not supported.');\n  }\n  const result = zipFn(inputs);\n\n  if (result.recurse && result.value !== null) {\n    throw new Error(\n        'A deep zip function may not return both a value and recurse=true.');\n  }\n\n  if (!result.recurse) {\n    return result.value;\n  } else if (isIterable(input)) {\n    // tslint:disable-next-line:no-any\n    const mappedIterable: any|any[] = Array.isArray(input) ? [] : {};\n    containedIn.add(input);\n    for (const k in input) {\n      const children = inputs.map(x => x[k]);\n      const childResult = deepZipInternal(children, zipFn, containedIn);\n      mappedIterable[k] = childResult;\n    }\n    containedIn.delete(input);\n    return mappedIterable;\n  } else {\n    throw new Error(`Can't recurse into non-iterable type: ${input}`);\n  }\n}\n\n// tslint:disable-next-line:no-any\nexport function zipToList(x: any[]): DeepMapResult {\n  if (x === null) {\n    return null;\n  }\n  // TODO(soergel): validate array type?\n\n  if (isIterable(x[0])) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: x, recurse: false};\n  }\n}\n\n/**\n * A return value for an async map function for use with deepMapAndAwaitAll.\n *\n * If recurse is true, the value should be empty, and iteration will continue\n * into the object or array.\n */\nexport type DeepMapAsyncResult = {\n  value: Promise<any>,\n  recurse: boolean\n};\n\n/**\n * Apply an async mapping function to a nested structure in a recursive manner.\n *\n * This first creates a nested structure of Promises, and then awaits all of\n * those, resulting in a single Promise for a resolved nested structure.\n *\n * The result of the mapping is an object with the same nested structure (i.e.,\n * of arrays and dicts) as the input, except that some subtrees are replaced,\n * according to the results of the mapping function.\n *\n * Mappings are memoized.  Thus, if the nested structure contains the same\n * object in multiple positions, the output will contain the same mapped object\n * in those positions.  Cycles are not supported, however.\n *\n * @param input: The object to which to apply the mapping function.\n * @param mapFn: A function that expects a single node of the object tree, and\n *   returns a `DeepMapAsyncResult`.  The `DeepMapAsyncResult` either provides\n *   a `Promise` for a replacement value for that node (i.e., replacing the\n *   subtree), or indicates that the node should be processed recursively.  Note\n *   that the decision whether or not to recurse must be made immediately; only\n *   the mapped value may be promised.\n */\nexport async function deepMapAndAwaitAll(\n    input: any, mapFn: (x: any) => DeepMapAsyncResult): Promise<any|any[]> {\n  const seen: Map<any, Promise<any>> = new Map();\n\n  // First do a normal deepMap, collecting Promises in 'seen' as a side effect.\n  deepMapInternal(input, mapFn, seen);\n\n  // Replace the Promises in 'seen' in place.\n  // Note TypeScript provides no async map iteration, and regular map iteration\n  // is broken too, so sadly we have to do Array.from() to make it work.\n  // (There's no advantage to Promise.all(), and that would be tricky anyway.)\n  for (const key of Array.from(seen.keys())) {\n    const value = seen.get(key);\n    if (value instanceof Promise) {\n      const mappedValue = await value;\n      seen.set(key, mappedValue);\n    }\n  }\n\n  // Normal deepMap again, this time filling in the resolved values.\n  // It's unfortunate that we have to do two passes.\n  // TODO(soergel): test performance and think harder about a fast solution.\n  const result = deepMapInternal(input, mapFn, seen);\n  return result;\n}\n\n/**\n * Determine whether the argument is iterable.\n *\n * @returns true if the argument is an array or any non-Tensor object.\n */\n// tslint:disable-next-line:no-any\nexport function isIterable(obj: any): boolean {\n  return obj != null && (!ArrayBuffer.isView(obj)) &&\n      (Array.isArray(obj) ||\n       (typeof obj === 'object' && !(obj instanceof tf.Tensor)));\n}\n\n/**\n * Determine whether the argument can be converted to Tensor.\n *\n * Tensors, primitives, arrays, and TypedArrays all qualify; anything else does\n * not.\n *\n * @returns true if the argument can be converted to Tensor.\n */\n// tslint:disable-next-line:no-any\nexport function canTensorify(obj: any): boolean {\n  return obj == null || isPrimitive(obj) || Array.isArray(obj) ||\n      (typeof obj === 'object' && (obj instanceof tf.Tensor)) ||\n      tf.util.isTypedArray(obj);\n}\n\n/**\n * Returns true if the given `value` is a primitive type. Otherwise returns\n * false. This is equivalant to node util.isPrimitive\n */\nfunction isPrimitive(value: any): boolean {\n  return (\n      value === null ||\n      (typeof value !== 'object' && typeof value !== 'function'));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {deepMap, DeepMapResult, isIterable} from './deep_map';\n\nexport function deepClone<T>(container: T): T {\n  return deepMap(container, cloneIfTensor);\n}\n\n// tslint:disable-next-line: no-any\nfunction cloneIfTensor(item: any): DeepMapResult {\n  if (item instanceof tf.Tensor) {\n    return ({value: item.clone(), recurse: false});\n  } else if (isIterable(item)) {\n    return {value: null, recurse: true};\n  } else {\n    return {value: item, recurse: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n/**\n * A ring buffer, providing O(1) FIFO, LIFO, and related operations.\n */\nexport class RingBuffer<T> {\n  // Note we store the indices in the range 0 <= index < 2*capacity.\n  // This allows us to distinguish the full from the empty case.\n  // See https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/\n  protected begin = 0;  // inclusive\n  protected end = 0;    // exclusive\n  protected doubledCapacity: number;\n\n  protected data: T[];\n\n  /**\n   * Constructs a `RingBuffer`.\n   * @param capacity The number of items that the buffer can accomodate.\n   */\n  constructor(public capacity: number) {\n    if (capacity == null) {\n      throw new RangeError('Can\\'t create a ring buffer of unknown capacity.');\n    }\n    if (capacity < 1) {\n      throw new RangeError('Can\\'t create ring buffer of capacity < 1.');\n    }\n    this.data = new Array<T>(capacity);\n    this.doubledCapacity = 2 * capacity;\n  }\n\n  /**\n   * Map any index into the range 0 <= index < 2*capacity.\n   */\n  protected wrap(index: number) {\n    // don't trust % on negative numbers\n    while (index < 0) {\n      index += this.doubledCapacity;\n    }\n    return index % this.doubledCapacity;\n  }\n\n  protected get(index: number) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t get item at a negative index.');\n    }\n    return this.data[index % this.capacity];\n  }\n\n  protected set(index: number, value: T) {\n    if (index < 0) {\n      throw new RangeError('Can\\'t set item at a negative index.');\n    }\n    this.data[index % this.capacity] = value;\n  }\n\n  /**\n   * Returns the current number of items in the buffer.\n   */\n  length(): number {\n    let length = this.end - this.begin;\n    if (length < 0) {\n      length = this.doubledCapacity + length;\n    }\n    return length;\n  }\n\n  /**\n   * Reports whether the buffer is full.\n   * @returns true if the number of items in the buffer equals its capacity, and\n   *   false otherwise.\n   */\n  isFull() {\n    return this.length() === this.capacity;\n  }\n\n  /**\n   * Reports whether the buffer is empty.\n   * @returns true if the number of items in the buffer equals zero, and\n   *   false otherwise.\n   */\n  isEmpty() {\n    return this.length() === 0;\n  }\n\n  /**\n   * Adds an item to the end of the buffer.\n   */\n  push(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.set(this.end, value);\n    this.end = this.wrap(this.end + 1);\n  }\n\n  /**\n   * Adds many items to the end of the buffer, in order.\n   */\n  pushAll(values: T[]) {\n    for (const value of values) {\n      this.push(value);\n    }\n  }\n\n  /**\n   * Removes and returns the last item in the buffer.\n   */\n  pop(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    this.end = this.wrap(this.end - 1);\n    const result = this.get(this.end);\n    this.set(this.end, undefined);\n    return result;\n  }\n\n  /**\n   * Adds an item to the beginning of the buffer.\n   */\n  unshift(value: T) {\n    if (this.isFull()) {\n      throw new RangeError('Ring buffer is full.');\n    }\n    this.begin = this.wrap(this.begin - 1);\n    this.set(this.begin, value);\n  }\n\n  /**\n   * Removes and returns the first item in the buffer.\n   */\n  shift(): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const result = this.get(this.begin);\n    this.set(this.begin, undefined);\n    this.begin = this.wrap(this.begin + 1);\n    return result;\n  }\n\n  /**\n   * Removes and returns a specific item in the buffer, and moves the last item\n   * to the vacated slot.  This is useful for implementing a shuffling stream.\n   * Note that this operation necessarily scrambles the original order.\n   *\n   * @param relativeIndex: the index of the item to remove, relative to the\n   *   first item in the buffer (e.g., hiding the ring nature of the underlying\n   *   storage).\n   */\n  shuffleExcise(relativeIndex: number): T {\n    if (this.isEmpty()) {\n      throw new RangeError('Ring buffer is empty.');\n    }\n    const index = this.wrap(this.begin + relativeIndex);\n    const result = this.get(index);\n    this.set(index, this.pop());\n    return result;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {RingBuffer} from './ring_buffer';\n\nexport class GrowingRingBuffer<T> extends RingBuffer<T> {\n  private static INITIAL_CAPACITY = 32;\n\n  /**\n   * Constructs a `GrowingRingBuffer`.\n   */\n  constructor() {\n    super(GrowingRingBuffer.INITIAL_CAPACITY);\n  }\n\n  isFull() {\n    return false;\n  }\n\n  push(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.push(value);\n  }\n\n  unshift(value: T) {\n    if (super.isFull()) {\n      this.expand();\n    }\n    super.unshift(value);\n  }\n\n  /**\n   * Doubles the capacity of the buffer.\n   */\n  private expand() {\n    const newCapacity = this.capacity * 2;\n    const newData = new Array<T>(newCapacity);\n    const len = this.length();\n\n    // Rotate the buffer to start at index 0 again, since we can't just\n    // allocate more space at the end.\n    for (let i = 0; i < len; i++) {\n      newData[i] = this.get(this.wrap(this.begin + i));\n    }\n\n    this.data = newData;\n    this.capacity = newCapacity;\n    this.doubledCapacity = 2 * this.capacity;\n    this.begin = 0;\n    this.end = len;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {Container} from '../types';\nimport {deepClone} from '../util/deep_clone';\nimport {deepMapAndAwaitAll, DeepMapAsyncResult, DeepMapResult, deepZip, zipToList} from '../util/deep_map';\nimport {GrowingRingBuffer} from '../util/growing_ring_buffer';\nimport {RingBuffer} from '../util/ring_buffer';\n\n/**\n * A nested structure of LazyIterators, used as the input to zip().\n */\nexport type IteratorContainer = Container<LazyIterator<tf.TensorContainer>>;\n\n// Here we implement a simple asynchronous iterator.\n// This lets us avoid using either third-party stream libraries or\n// recent TypeScript language support requiring polyfills.\n\n/**\n * Create a `LazyIterator` from an array of items.\n */\nexport function iteratorFromItems<T>(items: T[]): LazyIterator<T> {\n  return new ArrayIterator(items);\n}\n\n/**\n * Create a `LazyIterator` of incrementing integers.\n */\nexport function iteratorFromIncrementing(start: number): LazyIterator<number> {\n  let i = start;\n  return iteratorFromFunction(() => ({value: i++, done: false}));\n}\n\n/**\n * Create a `LazyIterator` from a function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * await iter.forEachAsync(e => console.log(e));\n * ```\n *\n * @param func A function that produces data on each call.\n */\nexport function iteratorFromFunction<T>(\n    func: () =>\n        IteratorResult<T>| Promise<IteratorResult<T>>): LazyIterator<T> {\n  return new FunctionCallIterator(func);\n}\n\n/**\n * Create a `LazyIterator` by concatenating underlying streams, which are\n * themselves provided as a stream.\n *\n * This can also be thought of as a \"stream flatten\" operation.\n *\n * @param baseIterators A stream of streams to be concatenated.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenated<T>(\n    baseIterators: LazyIterator<LazyIterator<T>>,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return new ChainedIterator(baseIterators, baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by concatenating streams produced by calling a\n * stream-generating function a given number of times.\n *\n * Since a `LazyIterator` is read-once, it cannot be repeated, but this\n * function can be used to achieve a similar effect:\n *\n *   LazyIterator.ofConcatenatedFunction(() => new MyIterator(), 6);\n *\n * @param iteratorFunc: A function that produces a new stream on each call.\n * @param count: The number of times to call the function.\n * @param baseErrorHandler An optional function that can intercept `Error`s\n *   raised during a `next()` call on the base stream.  This function can decide\n *   whether the error should be propagated, whether the error should be\n *   ignored, or whether the base stream should be terminated.\n */\nexport function iteratorFromConcatenatedFunction<T>(\n    iteratorFunc: () => IteratorResult<LazyIterator<T>>, count: number,\n    baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n  return iteratorFromConcatenated(\n      iteratorFromFunction(iteratorFunc).take(count), baseErrorHandler);\n}\n\n/**\n * Create a `LazyIterator` by zipping together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nexport function iteratorFromZipped<O extends tf.TensorContainer>(\n    iterators: IteratorContainer,\n    mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL): LazyIterator<O> {\n  return new ZipIterator<O>(iterators, mismatchMode);\n}\n\n/**\n * An asynchronous iterator, providing lazy access to a potentially\n * unbounded stream of elements.\n *\n * Iterator can be obtained from a dataset:\n * `const iter = await dataset.iterator();`\n */\nexport abstract class LazyIterator<T> {\n  // This class implements AsyncIterator<T>, but we have not yet set the\n  // TypeScript --downlevelIteration flag to enable that.\n\n  abstract summary(): string;\n\n  /**\n   * Returns a `Promise` for the next element in the stream.\n   *\n   * When an item can be provided successfully, the return value is\n   * `{value:T, done:false}`.\n   *\n   * Calling next() on a closed stream returns `{value:null, done:true}`.\n   */\n  abstract async next(): Promise<IteratorResult<T>>;\n\n  /**\n   * Collect all remaining elements of a bounded stream into an array.\n   * Obviously this will succeed only for small streams that fit in memory.\n   * Useful for testing.\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArray(): Promise<T[]> {\n    const result: T[] = [];\n    let x = await this.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await this.next();\n    }\n    return result;\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of stream elements, which will resolve\n   *   when the stream is exhausted.\n   */\n  async toArrayForTest(): Promise<T[]> {\n    const stream = this.prefetch(100);\n    const result: T[] = [];\n    let x = await stream.next();\n    while (!x.done) {\n      result.push(x.value);\n      x = await stream.next();\n    }\n    return result;\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveFully(): Promise<void> {\n    let x = await this.next();\n    while (!x.done) {\n      x = await this.next();\n    }\n  }\n\n  /**\n   * Draw items from the stream until it is exhausted, or a predicate fails.\n   *\n   * This can be useful when the stream has side effects but no output.  In\n   * that case, calling this function guarantees that the stream will be\n   * fully processed.\n   */\n  async resolveWhile(predicate: (r: T) => boolean): Promise<void> {\n    let x = await this.next();\n    let shouldContinue = predicate(x.value);\n    while ((!x.done) && shouldContinue) {\n      x = await this.next();\n      shouldContinue = predicate(x.value);\n    }\n  }\n\n  /**\n   * Handles errors thrown on this stream using a provided handler function.\n   *\n   * @param handler A function that handles any `Error` thrown during a `next()`\n   *   call and returns true if the stream should continue (dropping the failed\n   *   call) or false if the stream should quietly terminate.  If the handler\n   *   itself throws (or rethrows) an `Error`, that will be propagated.\n   *\n   * @returns A `LazyIterator` of elements passed through from upstream,\n   *   possibly filtering or terminating on upstream `next()` calls that\n   *   throw an `Error`.\n   */\n  handleErrors(handler: (error: Error) => boolean): LazyIterator<T> {\n    return new ErrorHandlingLazyIterator(this, handler);\n  }\n\n  // TODO(soergel): Implement reduce() etc.\n\n  /**\n   * Filters this stream according to `predicate`.\n   *\n   * @param predicate A function mapping a stream element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `LazyIterator` of elements for which the predicate was true.\n   */\n  filter(predicate: (value: T) => boolean): LazyIterator<T> {\n    return new FilterIterator(this, predicate);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  map<O>(transform: (value: T) => O): LazyIterator<O> {\n    return new MapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through an async 1-to-1 transform.\n   *\n   * @param transform A function mapping a stream element to a `Promise` for a\n   *   transformed stream element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  mapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform);\n  }\n\n  /**\n   * Maps this stream through a 1-to-1 transform, forcing serial execution.\n   *\n   * @param transform A function mapping a stream element to a transformed\n   *   element.\n   *\n   * @returns A `LazyIterator` of transformed elements.\n   */\n  serialMapAsync<O>(transform: (value: T) => Promise<O>): LazyIterator<O> {\n    return new AsyncMapIterator(this, transform).serial();\n  }\n\n  /**\n   * Maps this stream through a 1-to-many transform.\n   *\n   * @param transform A function mapping a stream element to an array of\n   *   transformed elements.\n   *\n   * @returns A `DataStream` of transformed elements.\n   */\n  flatmap<O>(transform: (value: T) => O[]): LazyIterator<O> {\n    return new FlatmapIterator(this, transform);\n  }\n\n  /**\n   * Apply a function to every element of the stream.\n   *\n   * @param f A function to apply to each stream element.\n   */\n  async forEachAsync(f: (value: T) => void): Promise<void> {\n    return this.map(f).resolveFully();\n  }\n\n  /**\n   * Apply a function to every element of the stream, forcing serial execution.\n   *\n   * @param f A function to apply to each stream element.  Should return 'true'\n   *   to indicate that the stream should continue, or 'false' to cause it to\n   *   terminate.\n   */\n  async serialForEach(f: (value: T) => Promise<boolean>): Promise<void> {\n    return this.serialMapAsync(f).resolveWhile(x => (x === true));\n  }\n\n  /**\n   * Groups elements into batches, represented as arrays of elements.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"Row-major\" means that the resulting batch is simply a collection of\n   * rows: `[row1, row2, row3, ...]`.  This is contrast to the column-major\n   * form, which is needed for vectorized computation.\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `LazyIterator` of batches of elements, represented as arrays\n   *   of the original element type.\n   */\n  rowMajorBatch(batchSize: number, smallLastBatch = true): LazyIterator<T[]> {\n    return new RowMajorBatchIterator(this, batchSize, smallLastBatch);\n  }\n\n  /**\n   * Groups elements into batches, represented in column-major form.\n   *\n   * We can think of the elements of this iterator as 'rows' (even if they are\n   * nested structures).  By the same token, consecutive values for a given\n   * key within the elements form a 'column'.  This matches the usual sense of\n   * 'row' and 'column' when processing tabular data (e.g., parsing a CSV).\n   *\n   * Thus, \"column-major\" means that the resulting batch is a (potentially\n   * nested) structure representing the columns.  Each column entry, then,\n   * contains a collection of the values found in that column for a range of\n   * input elements.  This representation allows for vectorized computation, in\n   * contrast to the row-major form.\n   *\n   * The inputs should all have the same nested structure (i.e., of arrays and\n   * dicts).  The result is a single object with the same nested structure,\n   * where the leaves are arrays collecting the values of the inputs at that\n   * location (or, optionally, the result of a custom function applied to those\n   * arrays).\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @param zipFn: (optional) A function that expects an array of elements at a\n   *   single node of the object tree, and returns a `DeepMapResult`.  The\n   *   `DeepMapResult` either provides a result value for that node (i.e.,\n   *   representing the subtree), or indicates that the node should be processed\n   *   recursively.  The default zipFn recurses as far as possible and places\n   *   arrays at the leaves.\n   * @returns A `LazyIterator` of batches of elements, represented as an object\n   *   with collections at the leaves.\n   */\n  columnMajorBatch(\n      batchSize: number, smallLastBatch = true,\n      // tslint:disable-next-line:no-any\n      zipFn: (xs: any[]) => DeepMapResult = zipToList):\n      LazyIterator<tf.TensorContainer> {\n    // First collect the desired number of input elements as a row-major batch.\n    const rowBatches = this.rowMajorBatch(batchSize, smallLastBatch);\n    // Now 'rotate' or 'pivot' the data, collecting all values from each column\n    // in the batch (i.e., for each key within the elements) into an array.\n    return rowBatches.map(x => deepZip(x, zipFn));\n  }\n\n  /**\n   * Concatenate this `LazyIterator` with another.\n   *\n   * @param iterator A `LazyIterator` to be concatenated onto this one.\n   * @param baseErrorHandler An optional function that can intercept `Error`s\n   *   raised during a `next()` call on the base stream.  This function can\n   *   decide whether the error should be propagated, whether the error should\n   *   be ignored, or whether the base stream should be terminated.\n   * @returns A `LazyIterator`.\n   */\n  concatenate(\n      iterator: LazyIterator<T>,\n      baseErrorHandler?: (e: Error) => boolean): LazyIterator<T> {\n    return new ChainedIterator(\n        iteratorFromItems([this, iterator]), baseErrorHandler);\n  }\n\n  /**\n   * Limits this stream to return at most `count` items.\n   *\n   * @param count The maximum number of items to provide from the stream. If\n   * a negative or undefined value is given, the entire stream is returned\n   *   unaltered.\n   */\n  take(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new TakeIterator(this, count);\n  }\n\n  /**\n   * Skips the first `count` items in this stream.\n   *\n   * @param count The number of items to skip.  If a negative or undefined\n   * value is given, the entire stream is returned unaltered.\n   */\n  skip(count: number): LazyIterator<T> {\n    if (count < 0 || count == null) {\n      return this;\n    }\n    return new SkipIterator(this, count);\n  }\n\n  /**\n   * Prefetch the first `bufferSize` items in this stream.\n   *\n   * Note this prefetches Promises, but makes no guarantees about when those\n   * Promises resolve.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   */\n  prefetch(bufferSize: number): LazyIterator<T> {\n    return new PrefetchIterator(this, bufferSize);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  /**\n   * Randomly shuffles the elements of this stream.\n   *\n   * @param bufferSize: An integer specifying the number of elements from\n   * this stream from which the new stream will sample.\n   * @param seed: (Optional.) An integer specifying the random seed that\n   * will be used to create the distribution.\n   */\n  shuffle(windowSize: number, seed?: string): LazyIterator<T> {\n    return new ShuffleIterator(this, windowSize, seed);\n  }\n\n  /**\n   * Force an iterator to execute serially: each next() call will await the\n   * prior one, so that they cannot execute concurrently.\n   */\n  serial(): LazyIterator<T> {\n    return new SerialIterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on LazyIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// Iterators that just extend LazyIterator directly\n// ============================================================================\n\nclass ArrayIterator<T> extends LazyIterator<T> {\n  private trav = 0;\n  constructor(protected items: T[]) {\n    super();\n  }\n\n  summary() {\n    return `Array of ${this.items.length} items`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.trav >= this.items.length) {\n      return {value: null, done: true};\n    }\n    const item = this.items[this.trav];\n    this.trav++;\n    return {value: deepClone(item), done: false};\n  }\n}\n\nclass FunctionCallIterator<T> extends LazyIterator<T> {\n  constructor(\n      protected nextFn: () => IteratorResult<T>| Promise<IteratorResult<T>>) {\n    super();\n  }\n\n  summary() {\n    return `Function call`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    try {\n      return this.nextFn();\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message =\n          `Error thrown while iterating through a dataset: ${e.message}`;\n      throw e;\n    }\n  }\n}\n\nclass SerialIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(protected upstream: LazyIterator<T>) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Serial`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    return this.upstream.next();\n  }\n}\n\nclass SkipIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  count = 0;\n\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Skip`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider tradeoffs of reading in parallel, eg.\n    // collecting next() promises in an Array and then waiting for\n    // Promise.all() of those. Benefit: pseudo-parallel execution.  Drawback:\n    // maybe delayed GC.\n    while (this.count++ < this.maxCount) {\n      const skipped = await this.upstream.next();\n      // short-circuit if upstream is already empty\n      if (skipped.done) {\n        return skipped;\n      }\n      tf.dispose(skipped.value as {});\n    }\n    return this.upstream.next();\n  }\n}\n\nclass TakeIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(protected upstream: LazyIterator<T>, protected maxCount: number) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Take`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    if (this.count++ >= this.maxCount) {\n      return {value: null, done: true};\n    }\n    return this.upstream.next();\n  }\n}\n\n// Note this batch just groups items into row-wise element arrays.\n// Rotating these to a column-wise representation happens only at the dataset\n// level.\nclass RowMajorBatchIterator<T> extends LazyIterator<T[]> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T[]>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected batchSize: number,\n      protected enableSmallLastBatch = true) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> RowMajorBatch`;\n  }\n\n  async next(): Promise<IteratorResult<T[]>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T[]>> {\n    const batch: T[] = [];\n    while (batch.length < this.batchSize) {\n      const item = await this.upstream.next();\n      if (item.done) {\n        if (this.enableSmallLastBatch && batch.length > 0) {\n          return {value: batch, done: false};\n        }\n        return {value: null, done: true};\n      }\n      batch.push(item.value);\n    }\n    return {value: batch, done: false};\n  }\n}\n\nclass FilterIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected predicate: (value: T) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Filter`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      const item = await this.upstream.next();\n      if (item.done || this.predicate(item.value)) {\n        return item;\n      }\n      tf.dispose(item.value as {});\n    }\n  }\n}\n\nclass MapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Map`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\nclass ErrorHandlingLazyIterator<T> extends LazyIterator<T> {\n  count = 0;\n  constructor(\n      protected upstream: LazyIterator<T>,\n      protected handler: (error: Error) => boolean) {\n    super();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> handleErrors`;\n  }\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    while (true) {\n      try {\n        return await this.upstream.next();\n      } catch (e) {\n        if (!this.handler(e)) {\n          return {value: null, done: true};\n        }\n        // If the handler returns true, loop and fetch the next upstream item.\n\n        // If the upstream iterator throws an endless stream of errors, and if\n        // the handler says to ignore them, then we loop forever here.  That is\n        // the correct behavior-- it's up to the handler to decide when to stop.\n      }\n    }\n  }\n}\n\nclass AsyncMapIterator<I, O> extends LazyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => Promise<O>) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> AsyncMap`;\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return {value: null, done: true};\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // That's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying\n    // any intermediate Tensors.  Here we are concerned only about the\n    // inputs.\n    const mapped = await this.transform(item.value);\n    const outputTensors = tf.tensor_util.getTensorsInContainer(mapped as {});\n\n    // TODO(soergel) faster intersection\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n    return {value: mapped, done: false};\n  }\n}\n\n// Iterators that maintain a queue of pending items\n// ============================================================================\n\n/**\n * A base class for transforming streams that operate by maintaining an\n * output queue of elements that are ready to return via next().  This is\n * commonly required when the transformation is 1-to-many:  A call to next()\n * may trigger a call to the underlying stream, which will produce many\n * mapped elements of this stream-- of which we need to return only one, so\n * we have to queue the rest.\n */\nexport abstract class OneToManyIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  protected outputQueue: RingBuffer<T>;\n\n  constructor() {\n    super();\n    this.outputQueue = new GrowingRingBuffer<T>();\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  /**\n   * Read one or more chunks from upstream and process them, possibly\n   * reading or writing a carryover, and adding processed items to the\n   * output queue.  Note it's possible that no items are added to the queue\n   * on a given pump() call, even if the upstream stream is not closed\n   * (e.g., because items are filtered).\n   *\n   * @return `true` if any action was taken, i.e. fetching items from the\n   *   upstream source OR adding items to the output queue.  `false` if the\n   *   upstream source is exhausted AND nothing was added to the queue\n   * (i.e., any remaining carryover).\n   */\n  protected abstract async pump(): Promise<boolean>;\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // Fetch so that the queue contains at least one item if possible.\n    // If the upstream source is exhausted, AND there are no items left in\n    // the output queue, then this stream is also exhausted.\n    while (this.outputQueue.length() === 0) {\n      // TODO(soergel): consider parallel reads.\n      if (!await this.pump()) {\n        return {value: null, done: true};\n      }\n    }\n    return {value: this.outputQueue.shift(), done: false};\n  }\n}\nclass FlatmapIterator<I, O> extends OneToManyIterator<O> {\n  constructor(\n      protected upstream: LazyIterator<I>,\n      protected transform: (value: I) => O[]) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Flatmap`;\n  }\n\n  async pump(): Promise<boolean> {\n    const item = await this.upstream.next();\n    if (item.done) {\n      return false;\n    }\n    const inputTensors = tf.tensor_util.getTensorsInContainer(item.value as {});\n    // Careful: the transform may mutate the item in place.\n    // that's why we have to remember the input Tensors above, and then\n    // below dispose only those that were not passed through to the output.\n    // Note too that the transform function is responsible for tidying any\n    // intermediate Tensors.  Here we are concerned only about the inputs.\n    const mappedArray = this.transform(item.value);\n    const outputTensors =\n        tf.tensor_util.getTensorsInContainer(mappedArray as {});\n    this.outputQueue.pushAll(mappedArray);\n\n    // TODO(soergel) faster intersection, and deduplicate outputTensors\n    // TODO(soergel) move to tf.disposeExcept(in, out)?\n    for (const t of inputTensors) {\n      if (!tf.tensor_util.isTensorInList(t, outputTensors)) {\n        t.dispose();\n      }\n    }\n\n    return true;\n  }\n}\n\n/**\n * Provides a `LazyIterator` that concatenates a stream of underlying\n * streams.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n */\nexport class ChainedIterator<T> extends LazyIterator<T> {\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>> = null;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private iterator: LazyIterator<T> = null;\n  private moreIterators: LazyIterator<LazyIterator<T>>;\n\n  constructor(\n      iterators: LazyIterator<LazyIterator<T>>,\n      private readonly baseErrorHandler?: (e: Error) => boolean) {\n    super();\n    this.moreIterators = iterators;\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of chained summaries';\n    return `${upstreamSummaries} -> Chained`;\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    this.lastRead = this.readFromChain(this.lastRead);\n    return this.lastRead;\n  }\n\n  private async readFromChain(lastRead: Promise<IteratorResult<T>>):\n      Promise<IteratorResult<T>> {\n    // Must await on the previous read since the previous read may have advanced\n    // the stream of streams, from which we need to read.\n    // This is unfortunate since we can't parallelize reads. Which means\n    // prefetching of chained streams is a no-op.\n    // One solution is to prefetch immediately upstream of this.\n    await lastRead;\n    if (this.iterator == null) {\n      const iteratorResult = await this.moreIterators.next();\n      if (iteratorResult.done) {\n        // No more streams to stream from.\n        return {value: null, done: true};\n      }\n      this.iterator = iteratorResult.value;\n      if (this.baseErrorHandler != null) {\n        this.iterator = this.iterator.handleErrors(this.baseErrorHandler);\n      }\n    }\n    const itemResult = await this.iterator.next();\n    if (itemResult.done) {\n      this.iterator = null;\n      return this.readFromChain(lastRead);\n    }\n    return itemResult;\n  }\n}\n\nexport enum ZipMismatchMode {\n  FAIL,      // require zipped streams to have the same length\n  SHORTEST,  // terminate zip when the first stream is exhausted\n  LONGEST    // use nulls for exhausted streams; use up the longest stream.\n}\n\n/**\n * Provides a `LazyIterator` that zips together an array, dict, or nested\n * structure of `LazyIterator`s (and perhaps additional constants).\n *\n * The underlying streams must provide elements in a consistent order such\n * that they correspond.\n *\n * Typically, the underlying streams should have the same number of\n * elements. If they do not, the behavior is determined by the\n * `mismatchMode` argument.\n *\n * The nested structure of the `iterators` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Doing this in a concurrency-safe way requires some trickery.  In\n * particular, we want this stream to return the elements from the\n * underlying streams in the correct order according to when next() was\n * called, even if the resulting Promises resolve in a different order.\n *\n * @param iterators: An array or object containing LazyIterators at the\n * leaves.\n * @param mismatchMode: Determines what to do when one underlying iterator\n * is exhausted before the others.  `ZipMismatchMode.FAIL` (the default)\n * causes an error to be thrown in this case.  `ZipMismatchMode.SHORTEST`\n * causes the zipped iterator to terminate with the furst underlying\n * streams, so elements remaining on the longer streams are ignored.\n * `ZipMismatchMode.LONGEST` causes the zipped stream to continue, filling\n * in nulls for the exhausted streams, until all streams are exhausted.\n */\nclass ZipIterator<O extends tf.TensorContainer> extends LazyIterator<O> {\n  private count = 0;\n  private currentPromise: Promise<IteratorResult<O>> = null;\n\n  constructor(\n      protected readonly iterators: IteratorContainer,\n      protected readonly mismatchMode: ZipMismatchMode = ZipMismatchMode.FAIL) {\n    super();\n  }\n\n  summary() {\n    const upstreamSummaries = 'TODO: fill in upstream of zip summaries';\n    return `{${upstreamSummaries}} -> Zip`;\n  }\n\n  private async nextState(afterState: Promise<IteratorResult<O>>):\n      Promise<IteratorResult<O>> {\n    // This chaining ensures that the underlying next() are not even called\n    // before the previous ones have resolved.\n    await afterState;\n\n    // Collect underlying iterator \"done\" signals as a side effect in\n    // getNext()\n    let numIterators = 0;\n    let iteratorsDone = 0;\n\n    function getNext(container: IteratorContainer): DeepMapAsyncResult {\n      if (container instanceof LazyIterator) {\n        const result = container.next();\n        return {\n          value: result.then(x => {\n            numIterators++;\n            if (x.done) {\n              iteratorsDone++;\n            }\n            return x.value;\n          }),\n          recurse: false\n        };\n      } else {\n        return {value: null, recurse: true};\n      }\n    }\n\n    const mapped: O = await deepMapAndAwaitAll(this.iterators, getNext);\n\n    if (numIterators === iteratorsDone) {\n      // The streams have all ended.\n      return {value: null, done: true};\n    }\n    if (iteratorsDone > 0) {\n      switch (this.mismatchMode) {\n        case ZipMismatchMode.FAIL:\n          throw new Error(\n              'Zipped streams should have the same length. ' +\n              `Mismatched at element ${this.count}.`);\n        case ZipMismatchMode.SHORTEST:\n          return {value: null, done: true};\n        case ZipMismatchMode.LONGEST:\n        default:\n          // Continue.  The exhausted streams already produced value: null.\n      }\n    }\n\n    this.count++;\n    return {value: mapped, done: false};\n  }\n\n  async next(): Promise<IteratorResult<O>> {\n    this.currentPromise = this.nextState(this.currentPromise);\n    return this.currentPromise;\n  }\n}\n\n// Iterators that maintain a ring buffer of pending promises\n// ============================================================================\n\n/**\n * A stream that prefetches a given number of items from an upstream source,\n * returning them in FIFO order.\n *\n * Note this prefetches Promises, but makes no guarantees about when those\n * Promises resolve.\n */\nexport class PrefetchIterator<T> extends LazyIterator<T> {\n  protected buffer: RingBuffer<Promise<IteratorResult<T>>>;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected bufferSize: number) {\n    super();\n    this.buffer = new RingBuffer<Promise<IteratorResult<T>>>(bufferSize);\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Prefetch`;\n  }\n\n  /**\n   * Refill the prefetch buffer.  Returns only after the buffer is full, or\n   * the upstream source is exhausted.\n   */\n  protected refill() {\n    while (!this.buffer.isFull()) {\n      const v = this.upstream.next();\n      this.buffer.push(v);\n    }\n  }\n\n  next(): Promise<IteratorResult<T>> {\n    this.refill();\n    // This shift will never throw an error because the buffer is always\n    // full after a refill. If the stream is exhausted, the buffer will be\n    // full of Promises that will resolve to the end-of-stream signal.\n    return this.buffer.shift();\n  }\n}\n\n/**\n * A stream that performs a sliding-window random shuffle on an upstream\n * source. This is like a `PrefetchIterator` except that the items are\n * returned in randomized order.  Mixing naturally improves as the buffer\n * size increases.\n */\nexport class ShuffleIterator<T> extends PrefetchIterator<T> {\n  private readonly random: seedrandom.prng;\n\n  // Strict Promise execution order:\n  // a next() call may not even begin until the previous one completes.\n  private lastRead: Promise<IteratorResult<T>>;\n\n  // Local state that should not be clobbered by out-of-order execution.\n  private upstreamExhausted = false;\n\n  constructor(\n      protected upstream: LazyIterator<T>, protected windowSize: number,\n      seed?: string) {\n    super(upstream, windowSize);\n    this.random = seedrandom.alea(seed || tf.util.now().toString());\n    this.lastRead = Promise.resolve({value: null, done: false});\n  }\n\n  async next(): Promise<IteratorResult<T>> {\n    // This sets this.lastRead to a new Promise right away, as opposed to\n    // saying `await this.lastRead; this.lastRead = this.serialNext();` which\n    // would not work because this.nextRead would be updated only after the\n    // promise resolves.\n    this.lastRead = this.lastRead.then(() => this.serialNext());\n    return this.lastRead;\n  }\n\n  private randomInt(max: number) {\n    return Math.floor(this.random() * max);\n  }\n\n  protected chooseIndex(): number {\n    return this.randomInt(this.buffer.length());\n  }\n\n  async serialNext(): Promise<IteratorResult<T>> {\n    // TODO(soergel): consider performance\n    if (!this.upstreamExhausted) {\n      this.refill();\n    }\n    while (!this.buffer.isEmpty()) {\n      const chosenIndex = this.chooseIndex();\n      const result = await this.buffer.shuffleExcise(chosenIndex);\n      if (result.done) {\n        this.upstreamExhausted = true;\n      } else {\n        this.refill();\n        return result;\n      }\n    }\n    return {value: null, done: true};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {TensorContainer, TensorLike} from '@tensorflow/tfjs-core';\nimport * as seedrandom from 'seedrandom';\n\nimport {iteratorFromConcatenated, iteratorFromFunction, iteratorFromItems, iteratorFromZipped, LazyIterator, ZipMismatchMode} from './iterators/lazy_iterator';\nimport {Container} from './types';\nimport {canTensorify, deepMapAndAwaitAll, DeepMapResult, isIterable} from './util/deep_map';\n\n/**\n * A nested structure of Datasets, used as the input to zip().\n */\nexport type DatasetContainer = Container<Dataset<TensorContainer>>;\n\n// TODO(soergel): consider vectorized operations within the pipeline.\n\n/**\n * Represents a potentially large list of independent data elements (typically\n * 'samples' or 'examples').\n *\n * A 'data example' may be a primitive, an array, a map from string keys to\n * values, or any nested structure of these.\n *\n * A `Dataset` represents an ordered collection of elements, together with a\n * chain of transformations to be performed on those elements. Each\n * transformation is a method of `Dataset` that returns another `Dataset`, so\n * these may be chained, e.g.\n * `const processedDataset = rawDataset.filter(...).map(...).batch(...)`.\n *\n * Data loading and transformation is done in a lazy, streaming fashion.  The\n * dataset may be iterated over multiple times; each iteration starts the data\n * loading anew and recapitulates the transformations.\n *\n * A `Dataset` is typically processed as a stream of unbatched examples --i.e.,\n * its transformations are applied one example at a time. Batching produces a\n * new `Dataset` where each element is a batch. Batching should usually come\n * last in a pipeline, because data transformations are easier to express on a\n * per-example basis than on a per-batch basis.\n *\n * The following code examples are calling `await dataset.forEachAsync(...)` to\n * iterate once over the entire dataset in order to print out the data.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport abstract class Dataset<T extends tf.TensorContainer> {\n  /*\n   * Provide a new stream of elements.  Note this will also start new streams\n   * from any underlying `Dataset`s.\n   *\n   * CAUTION: Any Tensors contained within the elements returned from\n   * this stream *must* be manually disposed to avoid a GPU memory leak.\n   * The tf.tidy() approach cannot be used in an asynchronous context.\n   */\n  abstract async iterator(): Promise<LazyIterator<T>>;\n\n  readonly size: number = null;\n\n  // TODO(soergel): Make Datasets report whether repeated iterator() calls\n  // produce the same result (e.g., reading from a file) or different results\n  // (e.g., from the webcam).  Currently we don't make this distinction but it\n  // could be important for the user to know.\n  // abstract isDeterministic(): boolean;\n\n  /**\n   * Groups elements into batches.\n   *\n   * It is assumed that each of the incoming dataset elements has the same\n   * structure-- i.e. the same set of keys at each location in an object\n   * hierarchy.  For each key, the resulting `Dataset` provides a batched\n   * element collecting all of the incoming values for that key.\n   *\n   *  * Incoming primitives are grouped into a 1-D Tensor.\n   *  * Incoming Tensors are grouped into a new Tensor where the 0'th axis is\n   *    the batch dimension.\n   *  * Incoming arrays are converted to Tensor and then batched.\n   *  * A nested array is interpreted as an n-D Tensor, so the batched result\n   *    has n+1 dimensions.\n   *  * An array that cannot be converted to Tensor produces an error.\n   *\n   * If an array should not be batched as a unit, it should first be converted\n   * to an object with integer keys.\n   *\n   * Here are a few examples:\n   *\n   * Batch a dataset of numbers:\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8]).batch(4);\n   * await a.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of arrays:\n   * ```js\n   * const b = tf.data.array([[1], [2], [3], [4], [5], [6], [7], [8]]).batch(4);\n   * await b.forEachAsync(e => e.print());\n   * ```\n   *\n   * Batch a dataset of objects:\n   * ```js\n   * const c = tf.data.array([{a: 1, b: 11}, {a: 2, b: 12}, {a: 3, b: 13},\n   *   {a: 4, b: 14}, {a: 5, b: 15}, {a: 6, b: 16}, {a: 7, b: 17},\n   *   {a: 8, b: 18}]).batch(4);\n   * await c.forEachAsync(e => {\n   *   console.log('{');\n   *   for(var key in e) {\n   *     console.log(key+':');\n   *     e[key].print();\n   *   }\n   *   console.log('}');\n   * })\n   * ```\n   *\n   * @param batchSize The number of elements desired per batch.\n   * @param smallLastBatch Whether to emit the final batch when it has fewer\n   *   than batchSize elements. Default true.\n   * @returns A `Dataset`, from which a stream of batches can be obtained.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  batch(batchSize: number, smallLastBatch = true): Dataset<tf.TensorContainer> {\n    const base = this;\n    tf.util.assert(\n        batchSize > 0, () => `batchSize needs to be positive, but it is\n      ${batchSize}`);\n    let size;\n    if (this.size === Infinity || this.size == null) {\n      // If the size of this dataset is infinity or null, the new size keeps the\n      // same.\n      size = this.size;\n    } else if (smallLastBatch) {\n      // If the size of this dataset is known and include small last batch, the\n      // new size is full batch count plus last batch.\n      size = Math.ceil(this.size / batchSize);\n    } else {\n      // If the size of this dataset is known and not include small last batch,\n      // the new size is full batch count.\n      size = Math.floor(this.size / batchSize);\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator())\n          .columnMajorBatch(batchSize, smallLastBatch, deepBatchConcat);\n    }, size);\n  }\n\n  /**\n   * Concatenates this `Dataset` with another.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * const b = tf.data.array([4, 5, 6]);\n   * const c = a.concatenate(b);\n   * await c.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param dataset A `Dataset` to be concatenated onto this one.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  concatenate(dataset: Dataset<T>): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity || dataset.size === Infinity) {\n      // If the size of any of these two dataset is infinity, new size is\n      // infinity.\n      size = Infinity;\n    } else if (this.size != null && dataset.size != null) {\n      // If the size of both datasets are known and not infinity, new size is\n      // sum the size of these two datasets.\n      size = this.size + dataset.size;\n    } else {\n      // If neither of these two datasets has infinite size and any of these two\n      // datasets' size is null, the new size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () =>\n            (await base.iterator()).concatenate(await dataset.iterator()),\n        size);\n  }\n\n  /**\n   * Filters this dataset according to `predicate`.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])\n   *   .filter(x => x%2 === 0);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param predicate A function mapping a dataset element to a boolean or a\n   * `Promise` for one.\n   *\n   * @returns A `Dataset` of elements for which the predicate was true.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  filter(predicate: (value: T) => boolean): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size === Infinity) {\n      // If the size of this dataset is infinity, new size is infinity\n      size = Infinity;\n    } else {\n      // If this dataset has limited elements, new size is null because it might\n      // exhausted randomly.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).filter(x => tf.tidy(() => predicate(x)));\n    }, size);\n  }\n\n  /**\n   * Apply a function to every element of the dataset.\n   *\n   * After the function is applied to a dataset element, any Tensors contained\n   * within that element are disposed.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param f A function to apply to each dataset element.\n   * @returns A `Promise` that resolves after all elements have been processed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async forEachAsync(f: (input: T) => void): Promise<void> {\n    return (await this.iterator()).forEachAsync(f);\n  }\n\n  /**\n   * Maps this dataset through a 1-to-1 transform.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).map(x => x*x);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a transformed\n   *   dataset element.\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  map<O extends tf.TensorContainer>(transform: (value: T) => O): Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).map(x => tf.tidy(() => transform(x)));\n    }, this.size);\n  }\n\n  /**\n   * Maps this dataset through an async 1-to-1 transform.\n   *\n   * ```js\n   * const a =\n   *  tf.data.array([1, 2, 3]).mapAsync(x => new Promise(function(resolve){\n   *    setTimeout(() => {\n   *      resolve(x * x);\n   *    }, Math.random()*1000 + 500);\n   *  }));\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @param transform A function mapping a dataset element to a `Promise` for a\n   *   transformed dataset element.  This transform is responsible for disposing\n   *   any intermediate `Tensor`s, i.e. by wrapping its computation in\n   *   `tf.tidy()`; that cannot be automated here (as it is in the synchronous\n   *   `map()` case).\n   *\n   * @returns A `Dataset` of transformed elements.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  mapAsync<O extends tf.TensorContainer>(transform: (value: T) => Promise<O>):\n      Dataset<O> {\n    const base = this;\n    return datasetFromIteratorFn(async () => {\n      return (await base.iterator()).mapAsync(transform);\n    }, this.size);\n  }\n\n  /**\n   *  Creates a `Dataset` that prefetches elements from this dataset.\n   *\n   * @param bufferSize: An integer specifying the number of elements to be\n   *   prefetched.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  prefetch(bufferSize: number): Dataset<T> {\n    if (bufferSize == null) {\n      throw new RangeError(\n          '`Dataset.prefetch()` requires bufferSize to be specified.');\n    }\n\n    const base = this;\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).prefetch(bufferSize), this.size);\n  }\n\n  /**\n   * Repeats this dataset `count` times.\n   *\n   * NOTE: If this dataset is a function of global state (e.g. a random number\n   * generator), then different repetitions may produce different elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3]).repeat(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: (Optional) An integer, representing the number of times\n   *   the dataset should be repeated. The default behavior (if `count` is\n   *   `undefined` or negative) is for the dataset be repeated indefinitely.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  repeat(count?: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count > 0) {\n      // If this dataset has size and count is positive, new size is current\n      // size multiply count. This also covers the case that current size is\n      // infinity.\n      size = this.size * count;\n    } else if (count === 0) {\n      // If count is 0, new size is 0.\n      size = 0;\n    } else if (this.size != null && (count === undefined || count < 0)) {\n      // If this dataset has size and count is undefined or negative, the\n      // dataset will be repeated indefinitely and new size is infinity.\n      size = Infinity;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(async () => {\n      const iteratorIterator = iteratorFromFunction(\n          async () => ({value: await base.iterator(), done: false}));\n      return iteratorFromConcatenated(iteratorIterator.take(count));\n    }, size);\n  }\n\n  /**\n   * Creates a `Dataset` that skips `count` initial elements from this dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).skip(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be skipped\n   *   to form the new dataset.  If `count` is greater than the size of this\n   *   dataset, the new dataset will contain no elements.  If `count`\n   *   is `undefined` or negative, skips the entire dataset.\n   *\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  skip(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && count >= 0 && this.size >= count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is current size minus skipped size.This also covers the case that\n      // current size is infinity.\n      size = this.size - count;\n    } else if (\n        this.size != null &&\n        (this.size < count || count === undefined || count < 0)) {\n      // If the size of this dataset is smaller than count, or count is\n      // undefined or negative, skips the entire dataset and the new size is 0.\n      size = 0;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).skip(count), size);\n  }\n\n  // TODO(soergel): deep sharded shuffle, where supported\n\n  static readonly MAX_BUFFER_SIZE = 10000;\n\n  /**\n   * Pseudorandomly shuffles the elements of this dataset. This is done in a\n   * streaming manner, by sampling from a given number of prefetched elements.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).shuffle(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param bufferSize: An integer specifying the number of elements from this\n   *   dataset from which the new dataset will sample.\n   * @param seed: (Optional) An integer specifying the random seed that will\n   *   be used to create the distribution.\n   * @param reshuffleEachIteration: (Optional) A boolean, which if true\n   *   indicates that the dataset should be pseudorandomly reshuffled each time\n   *   it is iterated over. If false, elements will be returned in the same\n   *   shuffled order on each iteration. (Defaults to `true`.)\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  shuffle(bufferSize: number, seed?: string, reshuffleEachIteration = true):\n      Dataset<T> {\n    if (bufferSize == null || bufferSize < 0) {\n      if (this.size == null) {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.');\n      } else {\n        throw new RangeError(\n            '`Dataset.shuffle()` requires bufferSize to be specified.  ' +\n            'If your data fits in main memory (for regular JS objects), ' +\n            'and/or GPU memory (for `tf.Tensor`s), consider setting ' +\n            `bufferSize to the dataset size (${this.size} elements)`);\n      }\n    }\n    const base = this;\n    const random = seedrandom.alea(seed || tf.util.now().toString());\n    return datasetFromIteratorFn(async () => {\n      let seed2 = random.int32();\n      if (reshuffleEachIteration) {\n        seed2 += random.int32();\n      }\n      return (await base.iterator()).shuffle(bufferSize, seed2.toString());\n    }, this.size);\n  }\n\n  /**\n   * Creates a `Dataset` with at most `count` initial elements from this\n   * dataset.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]).take(3);\n   * await a.forEachAsync(e => console.log(e));\n   * ```\n   *\n   * @param count: The number of elements of this dataset that should be taken\n   *   to form the new dataset.  If `count` is `undefined` or negative, or if\n   *   `count` is greater than the size of this dataset, the new dataset will\n   *   contain all elements of this dataset.\n   * @returns A `Dataset`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  take(count: number): Dataset<T> {\n    const base = this;\n    let size;\n    if (this.size != null && this.size > count) {\n      // If the size of this dataset is greater than count, the new dataset's\n      // size is count.\n      size = count;\n    } else if (this.size != null && this.size <= count) {\n      // If the size of this dataset is equal or smaller than count, the new\n      // dataset's size is the size of this dataset.\n      size = this.size;\n    } else {\n      // If the size of this dataset is null, the new dataset's size is null.\n      size = null;\n    }\n    return datasetFromIteratorFn(\n        async () => (await base.iterator()).take(count), size);\n  }\n\n  /**\n   * Collect all elements of this dataset into an array.\n   *\n   * Obviously this will succeed only for small datasets that fit in memory.\n   * Useful for testing and generally should be avoided if possible.\n   *\n   * ```js\n   * const a = tf.data.array([1, 2, 3, 4, 5, 6]);\n   * console.log(await a.toArray());\n   * ```\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async toArray() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArray();\n  }\n\n  /**\n   * Collect all elements of this dataset into an array with prefetching 100\n   * elements. This is useful for testing, because the prefetch changes the\n   * order in which the Promises are resolved along the processing pipeline.\n   * This may help expose bugs where results are dependent on the order of\n   * Promise resolution rather than on the logical order of the stream (i.e.,\n   * due to hidden mutable state).\n   *\n   * @returns A Promise for an array of elements, which will resolve\n   *   when a new stream has been obtained and fully consumed.\n   */\n  async toArrayForTest() {\n    if (this.size === Infinity) {\n      throw new Error('Can not convert infinite data stream to array.');\n    }\n    return (await this.iterator()).toArrayForTest();\n  }\n}\n\n/**\n * Create a `Dataset` defined by a provided iterator() function.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const iter = tf.data.iteratorFromFunction(func);\n * const ds = tf.data.datasetFromIteratorFn(iter);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n */\nexport function datasetFromIteratorFn<T extends tf.TensorContainer>(\n    iteratorFn: () => Promise<LazyIterator<T>>,\n    size: number = null): Dataset<T> {\n  return new class extends Dataset<T> {\n    size = size;\n\n    /*\n     * Provide a new stream of elements.  Note this will also start new streams\n     * from any underlying `Dataset`s.\n     */\n    async iterator(): Promise<LazyIterator<T>> {\n      return iteratorFn();\n    }\n  }\n  ();\n}\n\n/**\n * Create a `Dataset` from an array of elements.\n *\n * Create a Dataset from an array of objects:\n * ```js\n * const a = tf.data.array([{'item': 1}, {'item': 2}, {'item': 3}]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n *\n * Create a Dataset from an array of numbers:\n * ```js\n * const a = tf.data.array([4, 5, 6]);\n * await a.forEachAsync(e => console.log(e));\n * ```\n * @param items An array of elements that will be parsed as items in a dataset.\n *\n * @doc {heading: 'Data', subheading: 'Creation', namespace: 'data'}\n */\nexport function array<T extends tf.TensorContainer>(items: T[]): Dataset<T> {\n  return datasetFromIteratorFn(\n      async () => iteratorFromItems(items), items.length);\n}\n\n/**\n * Create a `Dataset` by zipping together an array, dict, or nested\n * structure of `Dataset`s (and perhaps additional constants).\n * The underlying datasets must provide elements in a consistent order such that\n * they correspond.\n *\n * The number of elements in the resulting dataset is the same as the size of\n * the smallest dataset in datasets.\n *\n * The nested structure of the `datasets` argument determines the\n * structure of elements in the resulting iterator.\n *\n * Note this means that, given an array of two datasets that produce dict\n * elements, the result is a dataset that produces elements that are arrays\n * of two dicts:\n *\n * Zip an array of datasets:\n * ```js\n * console.log('Zip two datasets of objects:');\n * const ds1 = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const ds2 = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const ds3 = tf.data.zip([ds1, ds2]);\n * await ds3.forEachAsync(e => console.log(JSON.stringify(e)));\n *\n * // If the goal is to merge the dicts in order to produce elements like\n * // {a: ..., b: ...}, this requires a second step such as:\n * console.log('Merge the objects:');\n * const ds4 = ds3.map(x => {return {a: x[0].a, b: x[1].b}});\n * await ds4.forEachAsync(e => console.log(e));\n * ```\n *\n * Zip a dict of datasets:\n * ```js\n * const a = tf.data.array([{a: 1}, {a: 2}, {a: 3}]);\n * const b = tf.data.array([{b: 4}, {b: 5}, {b: 6}]);\n * const c = tf.data.zip({c: a, d: b});\n * await c.forEachAsync(e => console.log(JSON.stringify(e)));\n * ```\n *\n * @doc {heading: 'Data', subheading: 'Operations', namespace: 'data'}\n */\nexport function zip<O extends tf.TensorContainer>(datasets: DatasetContainer):\n    Dataset<O> {\n  // manually type-check the argument for JS users\n  if (!isIterable(datasets)) {\n    throw new Error('The argument to zip() must be an object or array.');\n  }\n  let size;\n  if (Array.isArray(datasets)) {\n    for (let i = 0; i < datasets.length; i++) {\n      size = size == null ? (datasets[i] as Dataset<O>).size :\n                            Math.min(size, (datasets[i] as Dataset<O>).size);\n    }\n  } else if (datasets instanceof Object) {\n    for (const ds in datasets) {\n      size = size == null ? (datasets[ds] as Dataset<O>).size :\n                            Math.min(size, (datasets[ds] as Dataset<O>).size);\n    }\n  }\n  return datasetFromIteratorFn<O>(async () => {\n    const streams = await deepMapAndAwaitAll(datasets, d => {\n      if (d instanceof Dataset) {\n        return {value: d.iterator(), recurse: false};\n      } else if (isIterable(d)) {\n        return {value: null, recurse: true};\n      } else {\n        throw new Error(\n            'Leaves of the structure passed to zip() must be Datasets, ' +\n            'not primitives.');\n      }\n    });\n    return iteratorFromZipped<O>(streams, ZipMismatchMode.SHORTEST);\n  }, size);\n}\n\n/**\n * A zip function for use with deepZip, passed via the columnMajorBatch call.\n *\n * Accepts an array of identically-structured nested elements and either batches\n * them (if they are primitives, numeric arrays, or Tensors) or requests\n * recursion (if not).\n */\n// tslint:disable-next-line:no-any\nfunction deepBatchConcat(rows: any[]): DeepMapResult {\n  if (rows === null) {\n    return null;\n  }\n\n  // use the first item to decide whether to recurse or batch here.\n  const exampleRow = rows[0];\n\n  if (canTensorify(exampleRow)) {\n    // rows is an array of primitives, Tensors, or arrays.  Batch them.\n    const value = batchConcat(rows);\n    return {value, recurse: false};\n  }\n\n  // the example row is an object, so recurse into it.\n  return {value: null, recurse: true};\n}\n\n/**\n * Assembles a list of same-shaped numbers, number arrays, or Tensors\n * into a single new Tensor where axis 0 is the batch dimension.\n */\nfunction batchConcat<T extends(TensorLike | tf.Tensor)>(arrays: T[]):\n    tf.Tensor {\n  if (arrays.length === 0) {\n    // We can't return an empty Tensor because we don't know the element shape.\n    throw new Error('Can\\'t make a batch of zero elements.');\n  }\n\n  if (arrays[0] instanceof tf.Tensor) {\n    // Input is an array of Tensors\n    return tf.stack(arrays as tf.Tensor[]);\n  } else {\n    // Input is a possibly-nested array of numbers.\n    return tf.tensor(arrays as TensorLike);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\n\n/**\n * Represents a potentially large collection of text lines.\n *\n * The results are not batched.\n */\nexport class TextLineDataset extends Dataset<string> {\n  /**\n   * Create a `TextLineDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   */\n  constructor(protected readonly input: DataSource) {\n    super();\n  }\n\n  async iterator(): Promise<LazyIterator<string>> {\n    const inputIterator = await this.input.iterator();\n    const utf8Iterator = inputIterator.decodeUTF8();\n    const lineIterator = utf8Iterator.split('\\n').map(line => {\n      // Windows/DOS format text file has extra line breaker at the end of line.\n      if (line.endsWith('\\r')) {\n        line = line.slice(0, -1);\n      }\n      return line;\n    });\n    return lineIterator;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {Dataset} from '../dataset';\nimport {DataSource} from '../datasource';\nimport {LazyIterator} from '../iterators/lazy_iterator';\nimport {ColumnConfig, CSVConfig} from '../types';\nimport {TextLineDataset} from './text_line_dataset';\n\nconst CODE_QUOTE = '\"';\nconst STATE_OUT = Symbol('out');\nconst STATE_FIELD = Symbol('field');\nconst STATE_QUOTE = Symbol('quote');\nconst STATE_QUOTE_AFTER_QUOTE = Symbol('quoteafterquote');\nconst STATE_WITHIN_QUOTE_IN_QUOTE = Symbol('quoteinquote');\n\n/**\n * Represents a potentially large collection of delimited text records.\n *\n * The produced `TensorContainer`s each contain one key-value pair for\n * every column of the table.  When a field is empty in the incoming data, the\n * resulting value is `undefined`, or throw error if it is required.  Values\n * that can be parsed as numbers are emitted as type `number`, other values\n * are parsed as `string`.\n *\n * The results are not batched.\n *\n * @doc {heading: 'Data', subheading: 'Classes', namespace: 'data'}\n */\nexport class CSVDataset extends Dataset<TensorContainer> {\n  base: TextLineDataset;\n  private hasHeader = true;\n  private fullColumnNames: string[] = null;\n  private columnNamesValidated = false;\n  private columnConfigs: {[key: string]: ColumnConfig} = null;\n  private configuredColumnsOnly = false;\n  private delimiter = ',';\n  private delimWhitespace = false;\n\n  /**\n   * Returns column names of the csv dataset. If `configuredColumnsOnly` is\n   * true, return column names in `columnConfigs`. If `configuredColumnsOnly` is\n   * false and `columnNames` is provided, `columnNames`. If\n   * `configuredColumnsOnly` is false and `columnNames` is not provided, return\n   * all column names parsed from the csv file. For example usage please go to\n   * `tf.data.csv`.\n   *\n   * @doc {heading: 'Data', subheading: 'Classes'}\n   */\n  async columnNames() {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    return this.configuredColumnsOnly ? Object.keys(this.columnConfigs) :\n                                        this.fullColumnNames;\n  }\n\n  /* 1) If `columnNames` is provided as string[], use this string[] as output\n   * keys in corresponding order. The length must match the number of inferred\n   * columns if `hasHeader` is true .\n   * 2) If `columnNames` is not provided, parse header line as `columnNames` if\n   * hasHeader is true. If `hasHeader` is false, throw an error.\n   * 3) If `columnConfigs` is provided, all the keys in `columnConfigs` must\n   * exist in parsed `columnNames`.\n   */\n  private async setColumnNames() {\n    const columnNamesFromFile = await this.maybeReadHeaderLine();\n    if (!this.fullColumnNames && !columnNamesFromFile) {\n      // Throw an error if columnNames is not provided and no header line.\n      throw new Error(\n          'Column names must be provided if there is no header line.');\n    } else if (this.fullColumnNames && columnNamesFromFile) {\n      // Check provided columnNames match header line.\n      util.assert(\n          columnNamesFromFile.length === this.fullColumnNames.length,\n          () => 'The length of provided columnNames (' +\n              this.fullColumnNames.length.toString() +\n              ') does not match the length of the header line read from ' +\n              'file (' + columnNamesFromFile.length.toString() + ').');\n    }\n    if (!this.fullColumnNames) {\n      this.fullColumnNames = columnNamesFromFile;\n    }\n    // Check if there are duplicate column names.\n    const counts: {[key: string]: number} = this.fullColumnNames.reduce(\n        (countAcc: {[key: string]: number}, name) => {\n          countAcc[name] = (countAcc[name] + 1) || 1;\n          return countAcc;\n        },\n        {});\n    const duplicateNames =\n        Object.keys(counts).filter((name) => (counts[name] > 1));\n    util.assert(\n        duplicateNames.length === 0,\n        () => 'Duplicate column names found: ' + duplicateNames.toString());\n    // Check if keys in columnConfigs match columnNames.\n    if (this.columnConfigs) {\n      for (const key of Object.keys(this.columnConfigs)) {\n        const index = this.fullColumnNames.indexOf(key);\n        if (index === -1) {\n          throw new Error(\n              'The key \"' + key +\n              '\" provided in columnConfigs does not match any of the column ' +\n              'names (' + this.fullColumnNames.toString() + ').');\n        }\n      }\n    }\n    this.columnNamesValidated = true;\n  }\n\n  private async maybeReadHeaderLine() {\n    if (this.hasHeader) {\n      const iter = await this.base.iterator();\n      const firstElement = await iter.next();\n      if (firstElement.done) {\n        throw new Error('No data was found for CSV parsing.');\n      }\n      const firstLine: string = firstElement.value;\n      const headers = this.parseRow(firstLine, false);\n      return headers;\n    } else {\n      return null;\n    }\n  }\n\n  /**\n   * Create a `CSVDataset`.\n   *\n   * @param input A `DataSource` providing a chunked, UTF8-encoded byte stream.\n   * @param csvConfig (Optional) A CSVConfig object that contains configurations\n   *     of reading and decoding from CSV file(s).\n   *\n   *     hasHeader: (Optional) A boolean value that indicates whether the first\n   *     row of provided CSV file is a header line with column names, and should\n   *     not be included in the data. Defaults to `true`.\n   *\n   *     columnNames: (Optional) A list of strings that corresponds to\n   *     the CSV column names, in order. If provided, it ignores the column\n   *     names inferred from the header row. If not provided, infers the column\n   *     names from the first row of the records. If hasHeader is false and\n   *     columnNames is not provided, this method throws an error.\n   *\n   *     columnConfigs: (Optional) A dictionary whose key is column names, value\n   *     is an object stating if this column is required, column's data type,\n   *     default value, and if this column is label. If provided, keys must\n   *     correspond to names provided in columnNames or inferred from the file\n   *     header lines. If isLabel is true any column, returns an array of two\n   *     items: the first item is a dict of features key/value pairs, the second\n   *     item is a dict of labels key/value pairs. If no feature is marked as\n   *     label, returns a dict of features only.\n   *\n   *     configuredColumnsOnly (Optional) If true, only columns provided in\n   *     columnConfigs will be parsed and provided during iteration.\n   *\n   *     delimiter (Optional) The string used to parse each line of the input\n   *     file. Defaults to `,`.\n   */\n  constructor(protected readonly input: DataSource, csvConfig?: CSVConfig) {\n    super();\n    this.base = new TextLineDataset(input);\n    if (!csvConfig) {\n      csvConfig = {};\n    }\n    this.hasHeader = csvConfig.hasHeader === false ? false : true;\n    this.fullColumnNames = csvConfig.columnNames;\n    this.columnConfigs = csvConfig.columnConfigs;\n    this.configuredColumnsOnly = csvConfig.configuredColumnsOnly;\n    if (csvConfig.delimWhitespace) {\n      util.assert(\n          csvConfig.delimiter == null,\n          () =>\n              'Delimiter should not be provided when delimWhitespace is true.');\n      this.delimWhitespace = true;\n      this.delimiter = ' ';\n    } else {\n      this.delimiter = csvConfig.delimiter ? csvConfig.delimiter : ',';\n    }\n  }\n\n  async iterator(): Promise<LazyIterator<TensorContainer>> {\n    if (!this.columnNamesValidated) {\n      await this.setColumnNames();\n    }\n    let lines = await this.base.iterator();\n    if (this.hasHeader) {\n      // We previously read the first line to get the columnNames.\n      // Now that we're providing data, skip it.\n      lines = lines.skip(1);\n    }\n    return lines.map(x => this.makeDataElement(x));\n  }\n\n  makeDataElement(line: string): TensorContainer {\n    const values = this.parseRow(line);\n    const features: {[key: string]: TensorContainer} = {};\n    const labels: {[key: string]: TensorContainer} = {};\n\n    for (let i = 0; i < this.fullColumnNames.length; i++) {\n      const key = this.fullColumnNames[i];\n      const config = this.columnConfigs ? this.columnConfigs[key] : null;\n      if (this.configuredColumnsOnly && !config) {\n        // This column is not selected.\n        continue;\n      } else {\n        const value = values[i];\n        let parsedValue = null;\n        if (value === '') {\n          // If default value is provided, use it. If default value is not\n          // provided, set as undefined.\n          if (config && config.default !== undefined) {\n            parsedValue = config.default;\n          } else if (config && (config.required || config.isLabel)) {\n            throw new Error(\n                `Required column ${key} is empty in this line: ${line}`);\n          } else {\n            parsedValue = undefined;\n          }\n        } else {\n          // A value is present, so parse it based on type\n          const valueAsNum = Number(value);\n          if (isNaN(valueAsNum)) {\n            // The value is a string and this column is declared as boolean\n            // in config, parse it as boolean.\n            if (config && config.dtype === 'bool') {\n              parsedValue = this.getBoolean(value);\n            } else {\n              // Set value as string\n              parsedValue = value;\n            }\n          } else if (!config || !config.dtype) {\n            // If this value is a number and no type config is provided, return\n            // it as number.\n            parsedValue = valueAsNum;\n          } else {\n            // If this value is a number and data type is provided, parse it\n            // according to provided data type.\n            switch (config.dtype) {\n              case 'float32':\n                parsedValue = valueAsNum;\n                break;\n              case 'int32':\n                parsedValue = Math.floor(valueAsNum);\n                break;\n              case 'bool':\n                parsedValue = this.getBoolean(value);\n                break;\n              default:\n                parsedValue = valueAsNum;\n            }\n          }\n        }\n        // Check if this column is label.\n        (config && config.isLabel) ? labels[key] = parsedValue :\n                                     features[key] = parsedValue;\n      }\n    }\n    // If label exists, return an object of features and labels as {xs:features,\n    // ys:labels}, otherwise return features only.\n    if (Object.keys(labels).length === 0) {\n      return features;\n\n    } else {\n      return {xs: features, ys: labels};\n    }\n  }\n\n  private getBoolean(value: string): number {\n    if (value === '1' || value.toLowerCase() === 'true') {\n      return 1;\n    } else {\n      return 0;\n    }\n  }\n\n  // adapted from https://beta.observablehq.com/@mbostock/streaming-csv\n  private parseRow(line: string, validateElementCount = true): string[] {\n    const result: string[] = [];\n    let readOffset = 0;\n    const readLength = line.length;\n    let currentState = STATE_OUT;\n    // Goes through the line to parse quote.\n    for (let i = 0; i < readLength; i++) {\n      switch (currentState) {\n        // Before enter a new field\n        case STATE_OUT:\n          switch (line.charAt(i)) {\n            // Enter a quoted field\n            case CODE_QUOTE:\n              readOffset = i + 1;\n              currentState = STATE_QUOTE;\n              break;\n            // Read an empty field\n            case this.delimiter:\n              readOffset = i + 1;\n              // If delimiter is white space and configured to collapse\n              // multiple white spaces, ignore this white space.\n              if (this.delimiter === ' ' && this.delimWhitespace) {\n                break;\n              }\n              result.push('');\n              currentState = STATE_OUT;\n              break;\n            // Enter an unquoted field\n            default:\n              currentState = STATE_FIELD;\n              readOffset = i;\n              break;\n          }\n          break;\n        // In an unquoted field\n        case STATE_FIELD:\n          switch (line.charAt(i)) {\n            // Exit an unquoted field, add it to result\n            case this.delimiter:\n              result.push(line.substring(readOffset, i));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            default:\n          }\n          break;\n        // In a quoted field\n        case STATE_QUOTE:\n          switch (line.charAt(i)) {\n            // Read a quote after a quote\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE_AFTER_QUOTE;\n              break;\n            default:\n          }\n          break;\n        // This state means it's right after a second quote in a field\n        case STATE_QUOTE_AFTER_QUOTE:\n          switch (line.charAt(i)) {\n            // Finished a quoted field\n            case this.delimiter:\n              result.push(line.substring(readOffset, i - 1));\n              currentState = STATE_OUT;\n              readOffset = i + 1;\n              break;\n            // Finished a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            // In a quoted part in a quoted field\n            default:\n              currentState = STATE_WITHIN_QUOTE_IN_QUOTE;\n              break;\n          }\n          break;\n        case STATE_WITHIN_QUOTE_IN_QUOTE:\n          switch (line.charAt(i)) {\n            // Exit a quoted part in a quoted field\n            case CODE_QUOTE:\n              currentState = STATE_QUOTE;\n              break;\n            default:\n          }\n          break;\n        default:\n      }\n    }\n    // Adds last item based on if it is quoted.\n    if (currentState === STATE_QUOTE_AFTER_QUOTE) {\n      result.push(line.substring(readOffset, readLength - 1));\n    } else {\n      result.push(line.substring(readOffset));\n    }\n    // Check if each row has the same number of elements as column names.\n    if (validateElementCount && result.length !== this.fullColumnNames.length) {\n      throw new Error(`Invalid row in csv file. Should have ${\n          this.fullColumnNames.length} elements in a row, but got ${result}`);\n    }\n    return result;\n  }\n}\n\n// TODO(soergel): add more basic datasets for parity with tf.data\n// tf.data.FixedLengthRecordDataset()\n// tf.data.TFRecordDataset()\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env, Tensor, tensor, Tensor2D, Tensor3D, TensorContainer, util} from '@tensorflow/tfjs-core';\nimport {MicrophoneConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of tensors from microphone audio stream. The tensors are\n * representing audio data as frequency-domain spectrogram generated with\n * browser's native FFT. Tensors representing time-domain waveform is available\n * based on configuration. Only works in browser environment.\n */\nexport class MicrophoneIterator extends LazyIterator<TensorContainer> {\n  private isClosed = false;\n  private stream: MediaStream;\n  private readonly fftSize: number;\n  private readonly columnTruncateLength: number;\n  private freqData: Float32Array;\n  private timeData: Float32Array;\n  private readonly numFrames: number;\n  private analyser: AnalyserNode;\n  private audioContext: AudioContext;\n  private sampleRateHz: number;\n  private readonly audioTrackConstraints: MediaTrackConstraints;\n  private readonly smoothingTimeConstant: number;\n  private readonly includeSpectrogram: boolean;\n  private readonly includeWaveform: boolean;\n\n  private constructor(protected readonly microphoneConfig: MicrophoneConfig) {\n    super();\n    this.fftSize = microphoneConfig.fftSize || 1024;\n    const fftSizeLog2 = Math.log2(this.fftSize);\n    if (this.fftSize < 0 || fftSizeLog2 < 4 || fftSizeLog2 > 14 ||\n        !Number.isInteger(fftSizeLog2)) {\n      throw new Error(\n          `Invalid fftSize: it must be a power of 2 between ` +\n          `2 to 4 and 2 to 14, but got ${this.fftSize}`);\n    }\n\n    this.numFrames = microphoneConfig.numFramesPerSpectrogram || 43;\n    this.sampleRateHz = microphoneConfig.sampleRateHz;\n    this.columnTruncateLength =\n        microphoneConfig.columnTruncateLength || this.fftSize;\n    this.audioTrackConstraints = microphoneConfig.audioTrackConstraints;\n    this.smoothingTimeConstant = microphoneConfig.smoothingTimeConstant || 0;\n\n    this.includeSpectrogram =\n        microphoneConfig.includeSpectrogram === false ? false : true;\n    this.includeWaveform =\n        microphoneConfig.includeWaveform === true ? true : false;\n    if (!this.includeSpectrogram && !this.includeWaveform) {\n      throw new Error(\n          'Both includeSpectrogram and includeWaveform are false. ' +\n          'At least one type of data should be returned.');\n    }\n  }\n\n  summary() {\n    return `microphone`;\n  }\n\n  // Construct a MicrophoneIterator and start the audio stream.\n  static async create(microphoneConfig: MicrophoneConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'microphone API is only supported in browser environment.');\n    }\n\n    const microphoneIterator = new MicrophoneIterator(microphoneConfig);\n\n    // Call async function start() to initialize the audio stream.\n    await microphoneIterator.start();\n\n    return microphoneIterator;\n  }\n\n  // Start the audio stream and FFT.\n  async start(): Promise<void> {\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        audio: this.audioTrackConstraints == null ? true :\n                                                    this.audioTrackConstraints,\n        video: false\n      });\n    } catch (e) {\n      throw new Error(\n          `Error thrown while initializing video stream: ${e.message}`);\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain audio from microphone.');\n    }\n\n    const ctxConstructor =\n        // tslint:disable-next-line:no-any\n        (window as any).AudioContext || (window as any).webkitAudioContext;\n    this.audioContext = new ctxConstructor();\n\n    if (!this.sampleRateHz) {\n      // If sample rate is not provided, use the available sample rate on\n      // device.\n      this.sampleRateHz = this.audioContext.sampleRate;\n    } else if (this.audioContext.sampleRate !== this.sampleRateHz) {\n      throw new Error(\n          `Mismatch in sampling rate: ` +\n          `Expected: ${this.sampleRateHz}; ` +\n          `Actual: ${this.audioContext.sampleRate}`);\n    }\n\n    const streamSource = this.audioContext.createMediaStreamSource(this.stream);\n    this.analyser = this.audioContext.createAnalyser();\n    this.analyser.fftSize = this.fftSize * 2;\n    this.analyser.smoothingTimeConstant = this.smoothingTimeConstant;\n    streamSource.connect(this.analyser);\n    this.freqData = new Float32Array(this.fftSize);\n    this.timeData = new Float32Array(this.fftSize);\n    return;\n  }\n\n  async next(): Promise<IteratorResult<TensorContainer>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let spectrogramTensor: Tensor;\n    let waveformTensor: Tensor;\n\n    const audioDataQueue = await this.getAudioData();\n    if (this.includeSpectrogram) {\n      const freqData = this.flattenQueue(audioDataQueue.freqDataQueue);\n      spectrogramTensor = this.getTensorFromAudioDataArray(\n          freqData, [this.numFrames, this.columnTruncateLength, 1]);\n    }\n    if (this.includeWaveform) {\n      const timeData = this.flattenQueue(audioDataQueue.timeDataQueue);\n      waveformTensor = this.getTensorFromAudioDataArray(\n          timeData, [this.numFrames * this.fftSize, 1]);\n    }\n\n    return {\n      value: {'spectrogram': spectrogramTensor, 'waveform': waveformTensor},\n      done: false\n    };\n  }\n\n  // Capture one result from the audio stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<{spectrogram: Tensor3D, waveform: Tensor2D}> {\n    return (await this.next()).value as\n        {spectrogram: Tensor3D, waveform: Tensor2D};\n  }\n\n  private async getAudioData():\n      Promise<{freqDataQueue: Float32Array[], timeDataQueue: Float32Array[]}> {\n    const freqDataQueue: Float32Array[] = [];\n    const timeDataQueue: Float32Array[] = [];\n    let currentFrames = 0;\n    return new Promise(resolve => {\n      const intervalID = setInterval(() => {\n        if (this.includeSpectrogram) {\n          this.analyser.getFloatFrequencyData(this.freqData);\n          // If the audio stream is initializing, return empty queue.\n          if (this.freqData[0] === -Infinity) {\n            resolve({freqDataQueue, timeDataQueue});\n          }\n          freqDataQueue.push(this.freqData.slice(0, this.columnTruncateLength));\n        }\n        if (this.includeWaveform) {\n          this.analyser.getFloatTimeDomainData(this.timeData);\n          timeDataQueue.push(this.timeData.slice());\n        }\n\n        // Clean interval and return when all frames have been collected\n        if (++currentFrames === this.numFrames) {\n          clearInterval(intervalID);\n          resolve({freqDataQueue, timeDataQueue});\n        }\n      }, this.fftSize / this.sampleRateHz * 1e3);\n    });\n  }\n\n  // Stop the audio stream and pause the iterator.\n  stop(): void {\n    if (!this.isClosed) {\n      this.isClosed = true;\n      this.analyser.disconnect();\n      this.audioContext.close();\n      if (this.stream != null && this.stream.getTracks().length > 0) {\n        this.stream.getTracks()[0].stop();\n      }\n    }\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor[]> {\n    throw new Error('Can not convert infinite audio stream to array.');\n  }\n\n  // Return audio sampling rate in Hz\n  getSampleRate(): number {\n    return this.sampleRateHz;\n  }\n\n  private flattenQueue(queue: Float32Array[]): Float32Array {\n    const frameSize = queue[0].length;\n    const freqData = new Float32Array(queue.length * frameSize);\n    queue.forEach((data, i) => freqData.set(data, i * frameSize));\n    return freqData;\n  }\n\n  private getTensorFromAudioDataArray(freqData: Float32Array, shape: number[]):\n      Tensor {\n    const vals = new Float32Array(util.sizeFromShape(shape));\n    // If the data is less than the output shape, the rest is padded with zeros.\n    vals.set(freqData, vals.length - freqData.length);\n    return tensor(vals, shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {browser, env, image, tensor1d, Tensor1D, tensor2d, Tensor2D, Tensor3D, Tensor4D, tidy, util} from '@tensorflow/tfjs-core';\nimport {WebcamConfig} from '../types';\nimport {LazyIterator} from './lazy_iterator';\n\n/**\n * Provide a stream of image tensors from webcam video stream. Only works in\n * browser environment.\n */\nexport class WebcamIterator extends LazyIterator<Tensor3D> {\n  private isClosed = true;\n  private stream: MediaStream;\n  private resize = false;\n  private cropSize: [number, number];\n  private cropBox: Tensor2D;\n  private cropBoxInd: Tensor1D;\n\n  private constructor(\n      protected readonly webcamVideoElement: HTMLVideoElement,\n      protected readonly webcamConfig: WebcamConfig) {\n    super();\n    if (this.needToResize()) {\n      this.resize = true;\n      this.cropSize =\n          [this.webcamConfig.resizeHeight, this.webcamConfig.resizeWidth];\n      this.cropBoxInd = tensor1d([0], 'int32');\n      if (this.webcamConfig.centerCrop) {\n        // Calculate the box based on resizing shape.\n        const widthCroppingRatio =\n            this.webcamConfig.resizeWidth * 1.0 / this.webcamVideoElement.width;\n        const heightCroppingRatio = this.webcamConfig.resizeHeight * 1.0 /\n            this.webcamVideoElement.height;\n        const widthCropStart = (1 - widthCroppingRatio) / 2;\n        const heightCropStart = (1 - heightCroppingRatio) / 2;\n        const widthCropEnd = widthCropStart + widthCroppingRatio;\n        const heightCropEnd = heightCroppingRatio + heightCropStart;\n        this.cropBox = tensor2d(\n            [heightCropStart, widthCropStart, heightCropEnd, widthCropEnd],\n            [1, 4]);\n      } else {\n        this.cropBox = tensor2d([0, 0, 1, 1], [1, 4]);\n      }\n    }\n  }\n\n  summary() {\n    return `webcam`;\n  }\n\n  // Construct a WebcamIterator and start it's video stream.\n  static async create(\n      webcamVideoElement?: HTMLVideoElement, webcamConfig: WebcamConfig = {}) {\n    if (env().get('IS_NODE')) {\n      throw new Error(\n          'tf.data.webcam is only supported in browser environment.');\n    }\n\n    if (!webcamVideoElement) {\n      // If webcam video element is not provided, create a hidden video element\n      // with provided width and height.\n      webcamVideoElement = document.createElement('video');\n      if (!webcamConfig.resizeWidth || !webcamConfig.resizeHeight) {\n        throw new Error(\n            'Please provide webcam video element, or resizeWidth and ' +\n            'resizeHeight to create a hidden video element.');\n      }\n      webcamVideoElement.width = webcamConfig.resizeWidth;\n      webcamVideoElement.height = webcamConfig.resizeHeight;\n    }\n    const webcamIterator = new WebcamIterator(webcamVideoElement, webcamConfig);\n\n    // Call async function to initialize the video stream.\n    await webcamIterator.start();\n\n    return webcamIterator;\n  }\n\n  // Async function to start video stream.\n  async start(): Promise<void> {\n    if (this.webcamConfig.facingMode) {\n      util.assert(\n          (this.webcamConfig.facingMode === 'user') ||\n              (this.webcamConfig.facingMode === 'environment'),\n          () =>\n              `Invalid webcam facing mode: ${this.webcamConfig.facingMode}. ` +\n              `Please provide 'user' or 'environment'`);\n    }\n\n    try {\n      this.stream = await navigator.mediaDevices.getUserMedia({\n        video: {\n          deviceId: this.webcamConfig.deviceId,\n          facingMode: this.webcamConfig.facingMode ?\n              this.webcamConfig.facingMode :\n              'user',\n          width: this.webcamVideoElement.width,\n          height: this.webcamVideoElement.height\n        }\n      });\n    } catch (e) {\n      // Modify the error message but leave the stack trace intact\n      e.message = `Error thrown while initializing video stream: ${e.message}`;\n      throw e;\n    }\n\n    if (!this.stream) {\n      throw new Error('Could not obtain video from webcam.');\n    }\n\n    // Older browsers may not have srcObject\n    try {\n      this.webcamVideoElement.srcObject = this.stream;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = window.URL.createObjectURL(this.stream);\n    }\n    // Start the webcam video stream\n    this.webcamVideoElement.play();\n\n    this.isClosed = false;\n\n    return new Promise<void>(resolve => {\n      // Add event listener to make sure the webcam has been fully initialized.\n      this.webcamVideoElement.onloadedmetadata = () => {\n        resolve();\n      };\n    });\n  }\n\n  async next(): Promise<IteratorResult<Tensor3D>> {\n    if (this.isClosed) {\n      return {value: null, done: true};\n    }\n\n    let img;\n    try {\n      img = browser.fromPixels(this.webcamVideoElement);\n    } catch (e) {\n      throw new Error(\n          `Error thrown converting video to pixels: ${JSON.stringify(e)}`);\n    }\n    if (this.resize) {\n      try {\n        return {value: this.cropAndResizeFrame(img), done: false};\n      } catch (e) {\n        throw new Error(`Error thrown cropping the video: ${e.message}`);\n      } finally {\n        img.dispose();\n      }\n    } else {\n      return {value: img, done: false};\n    }\n  }\n\n  private needToResize() {\n    // If resizeWidth and resizeHeight are provided, and different from the\n    // width and height of original HTMLVideoElement, then resizing and cropping\n    // is required.\n    if (this.webcamConfig.resizeWidth && this.webcamConfig.resizeHeight &&\n        (this.webcamVideoElement.width !== this.webcamConfig.resizeWidth ||\n         this.webcamVideoElement.height !== this.webcamConfig.resizeHeight)) {\n      return true;\n    }\n    return false;\n  }\n\n  // Cropping and resizing each frame based on config\n  cropAndResizeFrame(img: Tensor3D): Tensor3D {\n    return tidy(() => {\n      const expandedImage: Tensor4D = img.toFloat().expandDims(0);\n      let resizedImage;\n      resizedImage = image.cropAndResize(\n          expandedImage, this.cropBox, this.cropBoxInd, this.cropSize,\n          'bilinear');\n      // Extract image from batch cropping.\n      const shape = resizedImage.shape;\n      return resizedImage.reshape(shape.slice(1) as [number, number, number]);\n    });\n  }\n\n  // Capture one frame from the video stream, and extract the value from\n  // iterator.next() result.\n  async capture(): Promise<Tensor3D> {\n    return (await this.next()).value;\n  }\n\n  // Stop the video stream and pause webcam iterator.\n  stop(): void {\n    const tracks = this.stream.getTracks();\n\n    tracks.forEach(track => track.stop());\n\n    try {\n      this.webcamVideoElement.srcObject = null;\n    } catch (error) {\n      console.log(error);\n      this.webcamVideoElement.src = null;\n    }\n    this.isClosed = true;\n  }\n\n  // Override toArray() function to prevent collecting.\n  toArray(): Promise<Tensor3D[]> {\n    throw new Error('Can not convert infinite video stream to array.');\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {ByteChunkIterator} from './iterators/byte_chunk_iterator';\n\n/**\n * Represents a data source readable as a stream of binary data chunks.\n *\n * Because `Dataset`s can be read repeatedly (via `Dataset.iterator()`), this\n * provides a means to repeatedly create streams from the underlying data\n * sources.\n */\nexport abstract class DataSource {\n  /**\n   * Obtain a new stream of binary data chunks.\n   *\n   * Starts the new stream from the beginning of the data source, even if other\n   * streams have been obtained previously.\n   */\n  abstract async iterator(): Promise<ByteChunkIterator>;\n\n  // TODO(soergel): consider chainable Dataset construction here\n}\n\n// TODO(soergel): consider convenience factory functions here\n// in combination with chainable source->dataset above, e.g.:\n// tf.data.url(...).asCsvDataset().shuffle().batch()\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\n\nexport abstract class StringIterator extends LazyIterator<string> {\n  /**\n   * Splits a string stream on a given separator.\n   *\n   * It is assumed that the incoming chunk boundaries have no semantic meaning,\n   * so conceptually the incoming stream is treated simply as the concatenation\n   * of its elements.\n   *\n   * The outgoing stream provides chunks corresponding to the results of the\n   * standard string split() operation (even if such a chunk spanned incoming\n   * chunks).  The separators are not included.\n   *\n   * A typical usage is to split a text file (represented as a stream with\n   * arbitrary chunk boundaries) into lines.\n   *\n   * @param upstream A readable stream of strings that can be treated as\n   *   concatenated.\n   * @param separator A character to split on.\n   */\n  split(separator: string): StringIterator {\n    return new SplitIterator(this, separator);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on StringIterator.  Unfortunately they can't be placed in separate files, due\n// to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class SplitIterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass SplitIterator extends StringIterator {\n  private impl: SplitIteratorImpl;\n\n  constructor(protected upstream: LazyIterator<string>, separator: string) {\n    super();\n    this.impl = new SplitIteratorImpl(upstream, separator);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\nclass SplitIteratorImpl extends OneToManyIterator<string> {\n  // A partial string at the end of an upstream chunk\n  carryover = '';\n\n  constructor(\n      protected upstream: LazyIterator<string>, protected separator: string) {\n    super();\n  }\n\n  summary() {\n    return `${this.upstream.summary()} -> Split('${this.separator}')`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    if (chunkResult.done) {\n      if (this.carryover === '') {\n        return false;\n      }\n\n      // Pretend that the pump succeeded in order to emit the small last batch.\n      // The next pump() call will actually fail.\n      this.outputQueue.push(this.carryover);\n      this.carryover = '';\n      return true;\n    }\n    const lines = chunkResult.value.split(this.separator);\n    // Note the behavior: \" ab \".split(' ') === ['', 'ab', '']\n    // Thus the carryover may be '' if the separator falls on a chunk\n    // boundary; this produces the correct result.\n\n    lines[0] = this.carryover + lines[0];\n    for (const line of lines.slice(0, -1)) {\n      this.outputQueue.push(line);\n    }\n    this.carryover = lines[lines.length - 1];\n\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {LazyIterator, OneToManyIterator} from './lazy_iterator';\nimport {StringIterator} from './string_iterator';\n\nexport abstract class ByteChunkIterator extends LazyIterator<Uint8Array> {\n  /**\n   * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n   *\n   * The byte arrays producetd from the ByteChunkIterator on which this is\n   * called will be interpreted as concatenated.  No assumptions are made about\n   * the boundaries of the incoming chunks, so a multi-byte UTF8 encoding of a\n   * character may span the boundary between chunks.  This naturally happens,\n   * for instance, when reading fixed-size byte arrays from a file.\n   */\n  decodeUTF8(): StringIterator {\n    return new Utf8Iterator(this);\n  }\n}\n\n// ============================================================================\n// The following private classes serve to implement the chainable methods\n// on ByteChunkIterator.  Unfortunately they can't be placed in separate files,\n// due to resulting trouble with circular imports.\n// ============================================================================\n\n// We wanted multiple inheritance, e.g.\n//   class Utf8Iterator extends QueueIterator<string>, StringIterator\n// but the TypeScript mixin approach is a bit hacky, so we take this adapter\n// approach instead.\n\nclass Utf8Iterator extends StringIterator {\n  private impl: Utf8IteratorImpl;\n\n  constructor(protected upstream: LazyIterator<Uint8Array>) {\n    super();\n    this.impl = new Utf8IteratorImpl(upstream);\n  }\n\n  summary() {\n    return this.impl.summary();\n  }\n\n  async next() {\n    return this.impl.next();\n  }\n}\n\n/**\n * Decode a stream of UTF8-encoded byte arrays to a stream of strings.\n *\n * This is tricky because the incoming byte array boundaries may disrupt a\n * multi-byte UTF8 character. Thus any incomplete character data at the end of\n * a chunk must be carried over and prepended to the next chunk before\n * decoding. Luckily with native decoder, TextDecoder in browser and\n * string_decoder in node, byte array boundaries are handled automatically.\n *\n * In the context of an input pipeline for machine learning, UTF8 decoding is\n * needed to parse text files containing training examples or prediction\n * requests (e.g., formatted as CSV or JSON). We cannot use the built-in\n * decoding provided by FileReader.readAsText() because here we are in a\n * streaming context, which FileReader does not support.\n *\n * @param upstream A `LazyIterator` of `Uint8Arrays` containing UTF8-encoded\n *   text, which should be interpreted as concatenated.  No assumptions are\n *   made about the boundaries of the incoming chunks, so a multi-byte UTF8\n *   encoding of a character may span the boundary between chunks.  This\n *   naturally happens, for instance, when reading fixed-size byte arrays from a\n *   file.\n */\nclass Utf8IteratorImpl extends OneToManyIterator<string> {\n  // `decoder` as `any` here to dynamically assign value based on the\n  // environment.\n  // tslint:disable-next-line:no-any\n  decoder: any;\n\n  constructor(protected readonly upstream: LazyIterator<Uint8Array>) {\n    super();\n    if (env().get('IS_BROWSER')) {\n      this.decoder = new TextDecoder('utf-8');\n    } else {\n      // tslint:disable-next-line:no-require-imports\n      const {StringDecoder} = require('string_decoder');\n      this.decoder = new StringDecoder('utf8');\n    }\n  }\n  summary() {\n    return `${this.upstream.summary()} -> Utf8`;\n  }\n\n  async pump(): Promise<boolean> {\n    const chunkResult = await this.upstream.next();\n    let chunk;\n    if (chunkResult.done) {\n      return false;\n    } else {\n      chunk = chunkResult.value;\n    }\n\n    let text: string;\n    if (env().get('IS_BROWSER')) {\n      text = this.decoder.decode(chunk, {stream: true});\n    } else {\n      text = this.decoder.write(Buffer.from(chunk.buffer));\n    }\n    this.outputQueue.push(text);\n    return true;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// inspired by https://github.com/maxogden/filereader-stream\nimport {env, util} from '@tensorflow/tfjs-core';\nimport {FileElement} from '../types';\nimport {ByteChunkIterator} from './byte_chunk_iterator';\n\nexport interface FileChunkIteratorOptions {\n  /** The byte offset at which to begin reading the File or Blob. Default 0. */\n  offset?: number;\n  /** The number of bytes to read at a time. Default 1MB. */\n  chunkSize?: number;\n}\n\n/**\n * Provide a stream of chunks from a File, Blob, or Uint8Array.\n * @param file The source File, Blob or Uint8Array.\n * @param options Optional settings controlling file reading.\n * @returns a lazy Iterator of Uint8Arrays containing sequential chunks of the\n *   input File, Blob or Uint8Array.\n */\nexport class FileChunkIterator extends ByteChunkIterator {\n  offset: number;\n  chunkSize: number;\n\n  constructor(\n      protected file: FileElement,\n      protected options: FileChunkIteratorOptions = {}) {\n    super();\n    util.assert(\n        (file instanceof Uint8Array) ||\n            (env().get('IS_BROWSER') ?\n                 (file instanceof File || file instanceof Blob) :\n                 false),\n        () => 'FileChunkIterator only supports File, Blob and Uint8Array ' +\n            'right now.');\n    this.offset = options.offset || 0;\n    // default 1MB chunk has tolerable perf on large files\n    this.chunkSize = options.chunkSize || 1024 * 1024;\n  }\n\n  summary() {\n    return `FileChunks ${this.file}`;\n  }\n\n  async next(): Promise<IteratorResult<Uint8Array>> {\n    if (this.offset >= ((this.file instanceof Uint8Array) ?\n                            this.file.byteLength :\n                            this.file.size)) {\n      return {value: null, done: true};\n    }\n    const chunk = new Promise<Uint8Array>((resolve, reject) => {\n      const end = this.offset + this.chunkSize;\n      if (this.file instanceof Uint8Array) {\n        // Note if end > this.uint8Array.byteLength, we just get a small last\n        // chunk.\n        resolve(new Uint8Array(this.file.slice(this.offset, end)));\n      } else {\n        // This branch assumes that this.file type is File or Blob, which\n        // means it is in the browser environment.\n\n        // TODO(soergel): is this a performance issue?\n        const fileReader = new FileReader();\n        fileReader.onload = (event) => {\n          let data: string|ArrayBuffer|Uint8Array = fileReader.result;\n          // Not sure we can trust the return type of\n          // FileReader.readAsArrayBuffer See e.g.\n          // https://github.com/node-file-api/FileReader/issues/2\n          if (data instanceof ArrayBuffer) {\n            data = new Uint8Array(data);\n          }\n          if (!(data instanceof Uint8Array)) {\n            return reject(new TypeError('FileReader returned unknown type.'));\n          }\n          resolve(data);\n        };\n        fileReader.onabort = (event) => {\n          return reject(new Error('Aborted'));\n        };\n        fileReader.onerror = (event) => {\n          return reject(new Error(event.type));\n        };\n        // TODO(soergel): better handle onabort, onerror\n        // Note if end > this.file.size, we just get a small last chunk.\n        const slice = this.file.slice(this.offset, end);\n        // We can't use readAsText here (even if we know the file is text)\n        // because the slice boundary may fall within a multi-byte character.\n        fileReader.readAsArrayBuffer(slice);\n      }\n      this.offset = end;\n    });\n    return {value: (await chunk), done: false};\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\nimport {FileChunkIterator, FileChunkIteratorOptions} from './file_chunk_iterator';\n\n/**\n * Provide a stream of chunks from a URL.\n *\n * Note this class first downloads the entire file into memory before providing\n * the first element from the stream.  This is because the Fetch API does not\n * yet reliably provide a reader stream for the response body.\n */\nexport async function urlChunkIterator(\n    url: RequestInfo, options: FileChunkIteratorOptions = {}) {\n  let urlString;\n  let requestInit;\n  if ((typeof url) === 'string') {\n    urlString = url as string;\n  } else {\n    urlString = (url as Request).url;\n    requestInit = getRequestInitFromRequest(url as Request);\n  }\n  const response = await util.fetch(urlString, requestInit);\n  if (response.ok) {\n    const uint8Array = new Uint8Array(await response.arrayBuffer());\n    return new FileChunkIterator(uint8Array, options);\n  } else {\n    throw new Error(response.statusText);\n  }\n}\n\n// Generate RequestInit from Request to match tf.util.fetch signature.\nconst getRequestInitFromRequest = (request: Request) => {\n  const init = {\n    method: request.method,\n    headers: request.headers,\n    body: request.body,\n    mode: request.mode,\n    credentials: request.credentials,\n    cache: request.cache,\n    redirect: request.redirect,\n    referrer: request.referrer,\n    integrity: request.integrity,\n  };\n  return init;\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\n// Skip tslint any type check cause this method is aiming to check type of\n// input.\n// tslint:disable-next-line:no-any\nexport function isLocalPath(source: any): boolean {\n  return (typeof source === 'string') && source.substr(0, 7) === 'file://';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIterator, FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {FileElement} from '../types';\nimport {isLocalPath} from '../util/source_util';\n\n/**\n * Represents a file, blob, or Uint8Array readable as a stream of binary data\n * chunks.\n */\nexport class FileDataSource extends DataSource {\n  /**\n   * Create a `FileDataSource`.\n   *\n   * @param input Local file path, or `File`/`Blob`/`Uint8Array` object to\n   *     read. Local file only works in node environment.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected input: FileElement|string,\n      protected readonly options: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.input) && env().get('IS_NODE')) {\n      // tslint:disable-next-line:no-require-imports\n      const fs = require('fs');\n      this.input = fs.readFileSync((this.input as string).substr(7));\n    }\n    // TODO(kangyizhang): Add LocalFileChunkIterator to split local streaming\n    // with file in browser.\n    return new FileChunkIterator(this.input as FileElement, this.options);\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {DataSource} from '../datasource';\nimport {ByteChunkIterator} from '../iterators/byte_chunk_iterator';\nimport {FileChunkIteratorOptions} from '../iterators/file_chunk_iterator';\nimport {urlChunkIterator} from '../iterators/url_chunk_iterator';\nimport {isLocalPath} from '../util/source_util';\nimport {FileDataSource} from './file_data_source';\n\n/*\n * Represents a URL readable as a stream of binary data chunks.\n */\nexport class URLDataSource extends DataSource {\n  /**\n   * Create a `URLDataSource`.\n   *\n   * @param url A source URL string, or a `Request` object.\n   * @param options Options passed to the underlying `FileChunkIterator`s,\n   *   such as {chunksize: 1024}.\n   */\n  constructor(\n      protected readonly url: RequestInfo,\n      protected readonly fileOptions: FileChunkIteratorOptions = {}) {\n    super();\n  }\n\n  // TODO(soergel): provide appropriate caching options.  Currently this\n  // will download the URL anew for each call to iterator().  Since we have\n  // to treat the downloaded file as a blob/buffer anyway, we may as well retain\n  // it-- but that raises GC issues.  Also we may want a persistent disk cache.\n  async iterator(): Promise<ByteChunkIterator> {\n    if (isLocalPath(this.url)) {\n      return (new FileDataSource(this.url as string, this.fileOptions))\n          .iterator();\n    } else {\n      return urlChunkIterator(this.url, this.fileOptions);\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n *\n * =============================================================================\n */\n\nimport {TensorContainer} from '@tensorflow/tfjs-core';\nimport {Dataset, datasetFromIteratorFn} from './dataset';\nimport {CSVDataset} from './datasets/csv_dataset';\nimport {iteratorFromFunction} from './iterators/lazy_iterator';\nimport {MicrophoneIterator} from './iterators/microphone_iterator';\nimport {WebcamIterator} from './iterators/webcam_iterator';\nimport {URLDataSource} from './sources/url_data_source';\nimport {CSVConfig, MicrophoneConfig, WebcamConfig} from './types';\n\n/**\n * Create a `CSVDataset` by reading and decoding CSV file(s) from provided URL\n * or local path if it's in Node environment.\n *\n * Note: If isLabel in columnConfigs is `true` for at least one column, the\n * element in returned `CSVDataset` will be an object of\n * `{xs:features, ys:labels}`: xs is a dict of features key/value pairs, ys\n * is a dict of labels key/value pairs. If no column is marked as label,\n * returns a dict of features only.\n *\n * ```js\n * const csvUrl =\n * 'https://storage.googleapis.com/tfjs-examples/multivariate-linear-regression/data/boston-housing-train.csv';\n *\n * async function run() {\n *   // We want to predict the column \"medv\", which represents a median value of\n *   // a home (in $1000s), so we mark it as a label.\n *   const csvDataset = tf.data.csv(\n *     csvUrl, {\n *       columnConfigs: {\n *         medv: {\n *           isLabel: true\n *         }\n *       }\n *     });\n *\n *   // Number of features is the number of column names minus one for the label\n *   // column.\n *   const numOfFeatures = (await csvDataset.columnNames()).length - 1;\n *\n *   // Prepare the Dataset for training.\n *   const flattenedDataset =\n *     csvDataset\n *     .map(({xs, ys}) =>\n *       {\n *         // Convert xs(features) and ys(labels) from object form (keyed by\n *         // column name) to array form.\n *         return {xs:Object.values(xs), ys:Object.values(ys)};\n *       })\n *     .batch(10);\n *\n *   // Define the model.\n *   const model = tf.sequential();\n *   model.add(tf.layers.dense({\n *     inputShape: [numOfFeatures],\n *     units: 1\n *   }));\n *   model.compile({\n *     optimizer: tf.train.sgd(0.000001),\n *     loss: 'meanSquaredError'\n *   });\n *\n *   // Fit the model using the prepared Dataset\n *   return model.fitDataset(flattenedDataset, {\n *     epochs: 10,\n *     callbacks: {\n *       onEpochEnd: async (epoch, logs) => {\n *         console.log(epoch + ':' + logs.loss);\n *       }\n *     }\n *   });\n * }\n *\n * await run();\n * ```\n *\n * @param source URL or local path to get CSV file. If it's a local path, it\n * must have prefix `file://` and it only works in node environment.\n * @param csvConfig (Optional) A CSVConfig object that contains configurations\n *     of reading and decoding from CSV file(s).\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function csv(\n    source: RequestInfo, csvConfig: CSVConfig = {}): CSVDataset {\n  return new CSVDataset(new URLDataSource(source), csvConfig);\n}\n\n/**\n * Create a `Dataset` that produces each element by calling a provided function.\n *\n * Note that repeated iterations over this `Dataset` may produce different\n * results, because the function will be called anew for each element of each\n * iteration.\n *\n * Also, beware that the sequence of calls to this function may be out of order\n * in time with respect to the logical order of the Dataset. This is due to the\n * asynchronous lazy nature of stream processing, and depends on downstream\n * transformations (e.g. .shuffle()). If the provided function is pure, this is\n * no problem, but if it is a closure over a mutable state (e.g., a traversal\n * pointer), then the order of the produced elements may be scrambled.\n *\n * ```js\n * let i = -1;\n * const func = () =>\n *    ++i < 5 ? {value: i, done: false} : {value: null, done: true};\n * const ds = tf.data.func(func);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param f A function that produces one data element on each call.\n */\nexport function func<T extends TensorContainer>(\n    f: () => IteratorResult<T>| Promise<IteratorResult<T>>): Dataset<T> {\n  const iter = iteratorFromFunction(f);\n  return datasetFromIteratorFn(async () => iter);\n}\n\n/**\n * Create a `Dataset` that produces each element from provided JavaScript\n * generator, which is a function*\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions),\n * or a function that returns an\n * iterator\n * (https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Iterators_and_Generators#Generator_functions).\n *\n * The returned iterator should have `.next()` function that returns element in\n * format of `{value: TensorContainer, done:boolean}`.\n *\n * Example of creating a dataset from an iterator factory:\n * ```js\n * function makeIterator() {\n *   const numElements = 10;\n *   let index = 0;\n *\n *   const iterator = {\n *     next: () => {\n *       let result;\n *       if (index < numElements) {\n *         result = {value: index, done: false};\n *         index++;\n *         return result;\n *       }\n *       return {value: index, done: true};\n *     }\n *   };\n *   return iterator;\n * }\n * const ds = tf.data.generator(makeIterator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * Example of creating a dataset from a generator:\n * ```js\n * function* dataGenerator() {\n *   const numElements = 10;\n *   let index = 0;\n *   while (index < numElements) {\n *     const x = index;\n *     index++;\n *     yield x;\n *   }\n * }\n *\n * const ds = tf.data.generator(dataGenerator);\n * await ds.forEachAsync(e => console.log(e));\n * ```\n *\n * @param generator A Javascript generator function that returns a JavaScript\n *     iterator.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   configParamIndices: [1]\n *  }\n */\nexport function generator<T extends TensorContainer>(\n    generator: () => Iterator<T>| Promise<Iterator<T>>): Dataset<T> {\n  return datasetFromIteratorFn(async () => {\n    const gen = await generator();\n    return iteratorFromFunction(() => gen.next());\n  });\n}\n\n/**\n * Create an iterator that generate `Tensor`s from webcam video stream. This API\n * only works in Browser environment when the device has webcam.\n *\n * Note: this code snippet only works when the device has a webcam. It will\n * request permission to open the webcam when running.\n * ```js\n * const videoElement = document.createElement('video');\n * videoElement.width = 100;\n * videoElement.height = 100;\n * const cam = await tf.data.webcam(videoElement);\n * const img = await cam.capture();\n * img.print();\n * cam.stop();\n * ```\n *\n * @param webcamVideoElement A `HTMLVideoElement` used to play video from\n *     webcam. If this element is not provided, a hidden `HTMLVideoElement` will\n *     be created. In that case, `resizeWidth` and `resizeHeight` must be\n *     provided to set the generated tensor shape.\n * @param webcamConfig A `WebcamConfig` object that contains configurations of\n *     reading and manipulating data from webcam video stream.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function webcam(\n    webcamVideoElement?: HTMLVideoElement,\n    webcamConfig?: WebcamConfig): Promise<WebcamIterator> {\n  return WebcamIterator.create(webcamVideoElement, webcamConfig);\n}\n\n/**\n * Create an iterator that generate frequency-domain spectrogram `Tensor`s from\n * microphone audio stream with browser's native FFT. This API only works in\n * browser environment when the device has microphone.\n *\n * Note: this code snippet only works when the device has a microphone. It will\n * request permission to open the microphone when running.\n * ```js\n * const mic = await tf.data.microphone({\n *   fftSize: 1024,\n *   columnTruncateLength: 232,\n *   numFramesPerSpectrogram: 43,\n *   sampleRateHz:44100,\n *   includeSpectrogram: true,\n *   includeWaveform: true\n * });\n * const audioData = await mic.capture();\n * const spectrogramTensor = audioData.spectrogram;\n * spectrogramTensor.print();\n * const waveformTensor = audioData.waveform;\n * waveformTensor.print();\n * mic.stop();\n * ```\n *\n * @param microphoneConfig A `MicrophoneConfig` object that contains\n *     configurations of reading audio data from microphone.\n *\n * @doc {\n *   heading: 'Data',\n *   subheading: 'Creation',\n *   namespace: 'data',\n *   ignoreCI: true\n *  }\n */\nexport async function microphone(microphoneConfig?: MicrophoneConfig):\n    Promise<MicrophoneIterator> {\n  return MicrophoneIterator.create(microphoneConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "// A port of an algorithm by Johannes Baag\u00F8e <baagoe@baagoe.com>, 2010\n// http://baagoe.com/en/RandomMusings/javascript/\n// https://github.com/nquinlan/better-random-numbers-for-javascript-mirror\n// Original work is under MIT license -\n\n// Copyright (C) 2010 by Johannes Baag\u00F8e <baagoe@baagoe.org>\n//\n// Permission is hereby granted, free of charge, to any person obtaining a copy\n// of this software and associated documentation files (the \"Software\"), to deal\n// in the Software without restriction, including without limitation the rights\n// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n// copies of the Software, and to permit persons to whom the Software is\n// furnished to do so, subject to the following conditions:\n// \n// The above copyright notice and this permission notice shall be included in\n// all copies or substantial portions of the Software.\n// \n// THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n// THE SOFTWARE.\n\n\n\n(function(global, module, define) {\n\nfunction Alea(seed) {\n  var me = this, mash = Mash();\n\n  me.next = function() {\n    var t = 2091639 * me.s0 + me.c * 2.3283064365386963e-10; // 2^-32\n    me.s0 = me.s1;\n    me.s1 = me.s2;\n    return me.s2 = t - (me.c = t | 0);\n  };\n\n  // Apply the seeding algorithm from Baagoe.\n  me.c = 1;\n  me.s0 = mash(' ');\n  me.s1 = mash(' ');\n  me.s2 = mash(' ');\n  me.s0 -= mash(seed);\n  if (me.s0 < 0) { me.s0 += 1; }\n  me.s1 -= mash(seed);\n  if (me.s1 < 0) { me.s1 += 1; }\n  me.s2 -= mash(seed);\n  if (me.s2 < 0) { me.s2 += 1; }\n  mash = null;\n}\n\nfunction copy(f, t) {\n  t.c = f.c;\n  t.s0 = f.s0;\n  t.s1 = f.s1;\n  t.s2 = f.s2;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new Alea(seed),\n      state = opts && opts.state,\n      prng = xg.next;\n  prng.int32 = function() { return (xg.next() * 0x100000000) | 0; }\n  prng.double = function() {\n    return prng() + (prng() * 0x200000 | 0) * 1.1102230246251565e-16; // 2^-53\n  };\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nfunction Mash() {\n  var n = 0xefc8249d;\n\n  var mash = function(data) {\n    data = data.toString();\n    for (var i = 0; i < data.length; i++) {\n      n += data.charCodeAt(i);\n      var h = 0.02519603282416938 * n;\n      n = h >>> 0;\n      h -= n;\n      h *= n;\n      n = h >>> 0;\n      h -= n;\n      n += h * 0x100000000; // 2^32\n    }\n    return (n >>> 0) * 2.3283064365386963e-10; // 2^-32\n  };\n\n  return mash;\n}\n\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.alea = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xor128\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n\n  // Set up generator function.\n  me.next = function() {\n    var t = me.x ^ (me.x << 11);\n    me.x = me.y;\n    me.y = me.z;\n    me.z = me.w;\n    return me.w ^= (me.w >>> 19) ^ t ^ (t >>> 8);\n  };\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor128 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorwow\" prng algorithm by\n// George Marsaglia.  See http://www.jstatsoft.org/v08/i14/paper\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var t = (me.x ^ (me.x >>> 2));\n    me.x = me.y; me.y = me.z; me.z = me.w; me.w = me.v;\n    return (me.d = (me.d + 362437 | 0)) +\n       (me.v = (me.v ^ (me.v << 4)) ^ (t ^ (t << 1))) | 0;\n  };\n\n  me.x = 0;\n  me.y = 0;\n  me.z = 0;\n  me.w = 0;\n  me.v = 0;\n\n  if (seed === (seed | 0)) {\n    // Integer seed.\n    me.x = seed;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 64; k++) {\n    me.x ^= strseed.charCodeAt(k) | 0;\n    if (k == strseed.length) {\n      me.d = me.x << 10 ^ me.x >>> 4;\n    }\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.x = f.x;\n  t.y = f.y;\n  t.z = f.z;\n  t.w = f.w;\n  t.v = f.v;\n  t.d = f.d;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorwow = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "// A Javascript implementaion of the \"xorshift7\" algorithm by\n// Fran\u00E7ois Panneton and Pierre L'ecuyer:\n// \"On the Xorgshift Random Number Generators\"\n// http://saluc.engr.uconn.edu/refs/crypto/rng/panneton05onthexorshift.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    // Update xor generator.\n    var X = me.x, i = me.i, t, v, w;\n    t = X[i]; t ^= (t >>> 7); v = t ^ (t << 24);\n    t = X[(i + 1) & 7]; v ^= t ^ (t >>> 10);\n    t = X[(i + 3) & 7]; v ^= t ^ (t >>> 3);\n    t = X[(i + 4) & 7]; v ^= t ^ (t << 7);\n    t = X[(i + 7) & 7]; t = t ^ (t << 13); v ^= t ^ (t << 9);\n    X[i] = v;\n    me.i = (i + 1) & 7;\n    return v;\n  };\n\n  function init(me, seed) {\n    var j, w, X = [];\n\n    if (seed === (seed | 0)) {\n      // Seed state array using a 32-bit integer.\n      w = X[0] = seed;\n    } else {\n      // Seed state using a string.\n      seed = '' + seed;\n      for (j = 0; j < seed.length; ++j) {\n        X[j & 7] = (X[j & 7] << 15) ^\n            (seed.charCodeAt(j) + X[(j + 1) & 7] << 13);\n      }\n    }\n    // Enforce an array length of 8, not all zeroes.\n    while (X.length < 8) X.push(0);\n    for (j = 0; j < 8 && X[j] === 0; ++j);\n    if (j == 8) w = X[7] = -1; else w = X[j];\n\n    me.x = X;\n    me.i = 0;\n\n    // Discard an initial 256 values.\n    for (j = 256; j > 0; --j) {\n      me.next();\n    }\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.x = f.x.slice();\n  t.i = f.i;\n  return t;\n}\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.x) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xorshift7 = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n", "// A Javascript implementaion of Richard Brent's Xorgens xor4096 algorithm.\n//\n// This fast non-cryptographic random number generator is designed for\n// use in Monte-Carlo algorithms. It combines a long-period xorshift\n// generator with a Weyl generator, and it passes all common batteries\n// of stasticial tests for randomness while consuming only a few nanoseconds\n// for each prng generated.  For background on the generator, see Brent's\n// paper: \"Some long-period random number generators using shifts and xors.\"\n// http://arxiv.org/pdf/1004.3115v1.pdf\n//\n// Usage:\n//\n// var xor4096 = require('xor4096');\n// random = xor4096(1);                        // Seed with int32 or string.\n// assert.equal(random(), 0.1520436450538547); // (0, 1) range, 53 bits.\n// assert.equal(random.int32(), 1806534897);   // signed int32, 32 bits.\n//\n// For nonzero numeric keys, this impelementation provides a sequence\n// identical to that by Brent's xorgens 3 implementaion in C.  This\n// implementation also provides for initalizing the generator with\n// string seeds, or for saving and restoring the state of the generator.\n//\n// On Chrome, this prng benchmarks about 2.1 times slower than\n// Javascript's built-in Math.random().\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this;\n\n  // Set up generator function.\n  me.next = function() {\n    var w = me.w,\n        X = me.X, i = me.i, t, v;\n    // Update Weyl generator.\n    me.w = w = (w + 0x61c88647) | 0;\n    // Update xor generator.\n    v = X[(i + 34) & 127];\n    t = X[i = ((i + 1) & 127)];\n    v ^= v << 13;\n    t ^= t << 17;\n    v ^= v >>> 15;\n    t ^= t >>> 12;\n    // Update Xor generator array state.\n    v = X[i] = v ^ t;\n    me.i = i;\n    // Result is the combination.\n    return (v + (w ^ (w >>> 16))) | 0;\n  };\n\n  function init(me, seed) {\n    var t, v, i, j, w, X = [], limit = 128;\n    if (seed === (seed | 0)) {\n      // Numeric seeds initialize v, which is used to generates X.\n      v = seed;\n      seed = null;\n    } else {\n      // String seeds are mixed into v and X one character at a time.\n      seed = seed + '\\0';\n      v = 0;\n      limit = Math.max(limit, seed.length);\n    }\n    // Initialize circular array and weyl value.\n    for (i = 0, j = -32; j < limit; ++j) {\n      // Put the unicode characters into the array, and shuffle them.\n      if (seed) v ^= seed.charCodeAt((j + 32) % seed.length);\n      // After 32 shuffles, take v as the starting w value.\n      if (j === 0) w = v;\n      v ^= v << 10;\n      v ^= v >>> 15;\n      v ^= v << 4;\n      v ^= v >>> 13;\n      if (j >= 0) {\n        w = (w + 0x61c88647) | 0;     // Weyl.\n        t = (X[j & 127] ^= (v + w));  // Combine xor and weyl to init array.\n        i = (0 == t) ? i + 1 : 0;     // Count zeroes.\n      }\n    }\n    // We have detected all zeroes; make the key nonzero.\n    if (i >= 128) {\n      X[(seed && seed.length || 0) & 127] = -1;\n    }\n    // Run the generator 512 times to further mix the state before using it.\n    // Factoring this as a function slows the main generator, so it is just\n    // unrolled here.  The weyl generator is not advanced while warming up.\n    i = 127;\n    for (j = 4 * 128; j > 0; --j) {\n      v = X[(i + 34) & 127];\n      t = X[i = ((i + 1) & 127)];\n      v ^= v << 13;\n      t ^= t << 17;\n      v ^= v >>> 15;\n      t ^= t >>> 12;\n      X[i] = v ^ t;\n    }\n    // Storing state as object members is faster than using closure variables.\n    me.w = w;\n    me.X = X;\n    me.i = i;\n  }\n\n  init(me, seed);\n}\n\nfunction copy(f, t) {\n  t.i = f.i;\n  t.w = f.w;\n  t.X = f.X.slice();\n  return t;\n};\n\nfunction impl(seed, opts) {\n  if (seed == null) seed = +(new Date);\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (state.X) copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.xor4096 = impl;\n}\n\n})(\n  this,                                     // window object or global\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n", "// A Javascript implementaion of the \"Tyche-i\" prng algorithm by\n// Samuel Neves and Filipe Araujo.\n// See https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n\n(function(global, module, define) {\n\nfunction XorGen(seed) {\n  var me = this, strseed = '';\n\n  // Set up generator function.\n  me.next = function() {\n    var b = me.b, c = me.c, d = me.d, a = me.a;\n    b = (b << 25) ^ (b >>> 7) ^ c;\n    c = (c - d) | 0;\n    d = (d << 24) ^ (d >>> 8) ^ a;\n    a = (a - b) | 0;\n    me.b = b = (b << 20) ^ (b >>> 12) ^ c;\n    me.c = c = (c - d) | 0;\n    me.d = (d << 16) ^ (c >>> 16) ^ a;\n    return me.a = (a - b) | 0;\n  };\n\n  /* The following is non-inverted tyche, which has better internal\n   * bit diffusion, but which is about 25% slower than tyche-i in JS.\n  me.next = function() {\n    var a = me.a, b = me.b, c = me.c, d = me.d;\n    a = (me.a + me.b | 0) >>> 0;\n    d = me.d ^ a; d = d << 16 ^ d >>> 16;\n    c = me.c + d | 0;\n    b = me.b ^ c; b = b << 12 ^ d >>> 20;\n    me.a = a = a + b | 0;\n    d = d ^ a; me.d = d = d << 8 ^ d >>> 24;\n    me.c = c = c + d | 0;\n    b = b ^ c;\n    return me.b = (b << 7 ^ b >>> 25);\n  }\n  */\n\n  me.a = 0;\n  me.b = 0;\n  me.c = 2654435769 | 0;\n  me.d = 1367130551;\n\n  if (seed === Math.floor(seed)) {\n    // Integer seed.\n    me.a = (seed / 0x100000000) | 0;\n    me.b = seed | 0;\n  } else {\n    // String seed.\n    strseed += seed;\n  }\n\n  // Mix in string seed, then discard an initial batch of 64 values.\n  for (var k = 0; k < strseed.length + 20; k++) {\n    me.b ^= strseed.charCodeAt(k) | 0;\n    me.next();\n  }\n}\n\nfunction copy(f, t) {\n  t.a = f.a;\n  t.b = f.b;\n  t.c = f.c;\n  t.d = f.d;\n  return t;\n};\n\nfunction impl(seed, opts) {\n  var xg = new XorGen(seed),\n      state = opts && opts.state,\n      prng = function() { return (xg.next() >>> 0) / 0x100000000; };\n  prng.double = function() {\n    do {\n      var top = xg.next() >>> 11,\n          bot = (xg.next() >>> 0) / 0x100000000,\n          result = (top + bot) / (1 << 21);\n    } while (result === 0);\n    return result;\n  };\n  prng.int32 = xg.next;\n  prng.quick = prng;\n  if (state) {\n    if (typeof(state) == 'object') copy(state, xg);\n    prng.state = function() { return copy(xg, {}); }\n  }\n  return prng;\n}\n\nif (module && module.exports) {\n  module.exports = impl;\n} else if (define && define.amd) {\n  define(function() { return impl; });\n} else {\n  this.tychei = impl;\n}\n\n})(\n  this,\n  (typeof module) == 'object' && module,    // present in node.js\n  (typeof define) == 'function' && define   // present with an AMD loader\n);\n\n\n", "/*\nCopyright 2014 David Bau.\n\nPermission is hereby granted, free of charge, to any person obtaining\na copy of this software and associated documentation files (the\n\"Software\"), to deal in the Software without restriction, including\nwithout limitation the rights to use, copy, modify, merge, publish,\ndistribute, sublicense, and/or sell copies of the Software, and to\npermit persons to whom the Software is furnished to do so, subject to\nthe following conditions:\n\nThe above copyright notice and this permission notice shall be\nincluded in all copies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND,\nEXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF\nMERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.\nIN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY\nCLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,\nTORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE\nSOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n\n*/\n\n(function (pool, math) {\n//\n// The following constants are related to IEEE 754 limits.\n//\nvar global = this,\n    width = 256,        // each RC4 output is 0 <= x < 256\n    chunks = 6,         // at least six RC4 outputs for each double\n    digits = 52,        // there are 52 significant digits in a double\n    rngname = 'random', // rngname: name for Math.random and Math.seedrandom\n    startdenom = math.pow(width, chunks),\n    significance = math.pow(2, digits),\n    overflow = significance * 2,\n    mask = width - 1,\n    nodecrypto;         // node.js crypto module, initialized at the bottom.\n\n//\n// seedrandom()\n// This is the seedrandom function described above.\n//\nfunction seedrandom(seed, options, callback) {\n  var key = [];\n  options = (options == true) ? { entropy: true } : (options || {});\n\n  // Flatten the seed string or build one from local entropy if needed.\n  var shortseed = mixkey(flatten(\n    options.entropy ? [seed, tostring(pool)] :\n    (seed == null) ? autoseed() : seed, 3), key);\n\n  // Use the seed to initialize an ARC4 generator.\n  var arc4 = new ARC4(key);\n\n  // This function returns a random double in [0, 1) that contains\n  // randomness in every bit of the mantissa of the IEEE 754 value.\n  var prng = function() {\n    var n = arc4.g(chunks),             // Start with a numerator n < 2 ^ 48\n        d = startdenom,                 //   and denominator d = 2 ^ 48.\n        x = 0;                          //   and no 'extra last byte'.\n    while (n < significance) {          // Fill up all significant digits by\n      n = (n + x) * width;              //   shifting numerator and\n      d *= width;                       //   denominator and generating a\n      x = arc4.g(1);                    //   new least-significant-byte.\n    }\n    while (n >= overflow) {             // To avoid rounding up, before adding\n      n /= 2;                           //   last byte, shift everything\n      d /= 2;                           //   right using integer math until\n      x >>>= 1;                         //   we have exactly the desired bits.\n    }\n    return (n + x) / d;                 // Form the number within [0, 1).\n  };\n\n  prng.int32 = function() { return arc4.g(4) | 0; }\n  prng.quick = function() { return arc4.g(4) / 0x100000000; }\n  prng.double = prng;\n\n  // Mix the randomness into accumulated entropy.\n  mixkey(tostring(arc4.S), pool);\n\n  // Calling convention: what to return as a function of prng, seed, is_math.\n  return (options.pass || callback ||\n      function(prng, seed, is_math_call, state) {\n        if (state) {\n          // Load the arc4 state from the given state if it has an S array.\n          if (state.S) { copy(state, arc4); }\n          // Only provide the .state method if requested via options.state.\n          prng.state = function() { return copy(arc4, {}); }\n        }\n\n        // If called as a method of Math (Math.seedrandom()), mutate\n        // Math.random because that is how seedrandom.js has worked since v1.0.\n        if (is_math_call) { math[rngname] = prng; return seed; }\n\n        // Otherwise, it is a newer calling convention, so return the\n        // prng directly.\n        else return prng;\n      })(\n  prng,\n  shortseed,\n  'global' in options ? options.global : (this == math),\n  options.state);\n}\nmath['seed' + rngname] = seedrandom;\n\n//\n// ARC4\n//\n// An ARC4 implementation.  The constructor takes a key in the form of\n// an array of at most (width) integers that should be 0 <= x < (width).\n//\n// The g(count) method returns a pseudorandom integer that concatenates\n// the next (count) outputs from ARC4.  Its return value is a number x\n// that is in the range 0 <= x < (width ^ count).\n//\nfunction ARC4(key) {\n  var t, keylen = key.length,\n      me = this, i = 0, j = me.i = me.j = 0, s = me.S = [];\n\n  // The empty key [] is treated as [0].\n  if (!keylen) { key = [keylen++]; }\n\n  // Set up S using the standard key scheduling algorithm.\n  while (i < width) {\n    s[i] = i++;\n  }\n  for (i = 0; i < width; i++) {\n    s[i] = s[j = mask & (j + key[i % keylen] + (t = s[i]))];\n    s[j] = t;\n  }\n\n  // The \"g\" method returns the next (count) outputs as one number.\n  (me.g = function(count) {\n    // Using instance members instead of closure state nearly doubles speed.\n    var t, r = 0,\n        i = me.i, j = me.j, s = me.S;\n    while (count--) {\n      t = s[i = mask & (i + 1)];\n      r = r * width + s[mask & ((s[i] = s[j = mask & (j + t)]) + (s[j] = t))];\n    }\n    me.i = i; me.j = j;\n    return r;\n    // For robust unpredictability, the function call below automatically\n    // discards an initial batch of values.  This is called RC4-drop[256].\n    // See http://google.com/search?q=rsa+fluhrer+response&btnI\n  })(width);\n}\n\n//\n// copy()\n// Copies internal state of ARC4 to or from a plain object.\n//\nfunction copy(f, t) {\n  t.i = f.i;\n  t.j = f.j;\n  t.S = f.S.slice();\n  return t;\n};\n\n//\n// flatten()\n// Converts an object tree to nested arrays of strings.\n//\nfunction flatten(obj, depth) {\n  var result = [], typ = (typeof obj), prop;\n  if (depth && typ == 'object') {\n    for (prop in obj) {\n      try { result.push(flatten(obj[prop], depth - 1)); } catch (e) {}\n    }\n  }\n  return (result.length ? result : typ == 'string' ? obj : obj + '\\0');\n}\n\n//\n// mixkey()\n// Mixes a string seed into a key that is an array of integers, and\n// returns a shortened string seed that is equivalent to the result key.\n//\nfunction mixkey(seed, key) {\n  var stringseed = seed + '', smear, j = 0;\n  while (j < stringseed.length) {\n    key[mask & j] =\n      mask & ((smear ^= key[mask & j] * 19) + stringseed.charCodeAt(j++));\n  }\n  return tostring(key);\n}\n\n//\n// autoseed()\n// Returns an object for autoseeding, using window.crypto and Node crypto\n// module if available.\n//\nfunction autoseed() {\n  try {\n    var out;\n    if (nodecrypto && (out = nodecrypto.randomBytes)) {\n      // The use of 'out' to remember randomBytes makes tight minified code.\n      out = out(width);\n    } else {\n      out = new Uint8Array(width);\n      (global.crypto || global.msCrypto).getRandomValues(out);\n    }\n    return tostring(out);\n  } catch (e) {\n    var browser = global.navigator,\n        plugins = browser && browser.plugins;\n    return [+new Date, global, plugins, global.screen, tostring(pool)];\n  }\n}\n\n//\n// tostring()\n// Converts an array of charcodes to a string\n//\nfunction tostring(a) {\n  return String.fromCharCode.apply(0, a);\n}\n\n//\n// When seedrandom.js is loaded, we immediately mix a few bits\n// from the built-in RNG into the entropy pool.  Because we do\n// not want to interfere with deterministic PRNG state later,\n// seedrandom will not call math.random on its own again after\n// initialization.\n//\nmixkey(math.random(), pool);\n\n//\n// Nodejs and AMD support: export the implementation as a module using\n// either convention.\n//\nif ((typeof module) == 'object' && module.exports) {\n  module.exports = seedrandom;\n  // When in node.js, try using crypto package for autoseeding.\n  try {\n    nodecrypto = require('crypto');\n  } catch (ex) {}\n} else if ((typeof define) == 'function' && define.amd) {\n  define(function() { return seedrandom; });\n}\n\n// End anonymous scope, and pass initial values.\n})(\n  [],     // pool: entropy pool starts empty\n  Math    // math: package containing random, pow, and seedrandom\n);\n", "// A library of seedable RNGs implemented in Javascript.\n//\n// Usage:\n//\n// var seedrandom = require('seedrandom');\n// var random = seedrandom(1); // or any seed.\n// var x = random();       // 0 <= x < 1.  Every bit is random.\n// var x = random.quick(); // 0 <= x < 1.  32 bits of randomness.\n\n// alea, a 53-bit multiply-with-carry generator by Johannes Baag\u00F8e.\n// Period: ~2^116\n// Reported to pass all BigCrush tests.\nvar alea = require('./lib/alea');\n\n// xor128, a pure xor-shift generator by George Marsaglia.\n// Period: 2^128-1.\n// Reported to fail: MatrixRank and LinearComp.\nvar xor128 = require('./lib/xor128');\n\n// xorwow, George Marsaglia's 160-bit xor-shift combined plus weyl.\n// Period: 2^192-2^32\n// Reported to fail: CollisionOver, SimpPoker, and LinearComp.\nvar xorwow = require('./lib/xorwow');\n\n// xorshift7, by Fran\u00E7ois Panneton and Pierre L'ecuyer, takes\n// a different approach: it adds robustness by allowing more shifts\n// than Marsaglia's original three.  It is a 7-shift generator\n// with 256 bits, that passes BigCrush with no systmatic failures.\n// Period 2^256-1.\n// No systematic BigCrush failures reported.\nvar xorshift7 = require('./lib/xorshift7');\n\n// xor4096, by Richard Brent, is a 4096-bit xor-shift with a\n// very long period that also adds a Weyl generator. It also passes\n// BigCrush with no systematic failures.  Its long period may\n// be useful if you have many generators and need to avoid\n// collisions.\n// Period: 2^4128-2^32.\n// No systematic BigCrush failures reported.\nvar xor4096 = require('./lib/xor4096');\n\n// Tyche-i, by Samuel Neves and Filipe Araujo, is a bit-shifting random\n// number generator derived from ChaCha, a modern stream cipher.\n// https://eden.dei.uc.pt/~sneves/pubs/2011-snfa2.pdf\n// Period: ~2^127\n// No systematic BigCrush failures reported.\nvar tychei = require('./lib/tychei');\n\n// The original ARC4-based prng included in this library.\n// Period: ~2^1600\nvar sr = require('./seedrandom');\n\nsr.alea = alea;\nsr.xor128 = xor128;\nsr.xorwow = xorwow;\nsr.xorshift7 = xorshift7;\nsr.xor4096 = xor4096;\nsr.tychei = tychei;\n\nmodule.exports = sr;\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${\n              opName} does not support complex64 tensors in the CPU backend.`);\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {backend_util, BackendTimingInfo, DataStorage, DataType, DataValues, engine, env, kernel_impls, KernelBackend, max, NumericDataType, Rank, Scalar, ShapeMap, slice_util, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, TensorInfo, TypedArray, upcastType, util} from '@tensorflow/tfjs-core';\n\nconst nonMaxSuppressionV3Impl = kernel_impls.nonMaxSuppressionV3Impl;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\nimport * as seedrandom from 'seedrandom';\nimport {assertNotComplex} from './cpu_util';\n\ninterface DataId {}\n\nfunction mapActivation(\n    backend: MathBackendCPU, x: Tensor, activation: backend_util.Activation,\n    preluActivationWeights?: Tensor): Tensor {\n  if (activation === 'linear') {\n    return backend.linear(x);\n  } else if (activation === 'relu') {\n    return backend.relu(x);\n  } else if (activation === 'elu') {\n    return tf.elu(x);\n  } else if (activation === 'relu6') {\n    return backend.relu6(x);\n  } else if (activation === 'prelu') {\n    return backend.prelu(x, preluActivationWeights);\n  }\n  throw new Error(\n      `Activation ${activation} has not been implemented for the CPU backend.`);\n}\n\nexport interface TensorData<D extends DataType> {\n  values?: backend_util.BackendValues;\n  dtype: D;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensorInfos field.\n  complexTensorInfos?: {real: TensorInfo, imag: TensorInfo};\n  // refCount keeps track of how many tensors reference it. Used for memory\n  // management.\n  refCount: number;\n}\n\nexport class MathBackendCPU extends KernelBackend {\n  public blockSize = 48;\n\n  data: DataStorage<TensorData<DataType>>;\n  private firstUse = true;\n\n  constructor() {\n    super();\n    this.data = new DataStorage(this, engine());\n  }\n\n  write(values: backend_util.BackendValues, shape: number[], dtype: DataType):\n      DataId {\n    if (this.firstUse) {\n      this.firstUse = false;\n      if (env().get('IS_NODE')) {\n        backend_util.warn(\n            '\\n============================\\n' +\n            'Hi there \uD83D\uDC4B. Looks like you are running TensorFlow.js in ' +\n            'Node.js. To speed things up dramatically, install our node ' +\n            'backend, which binds to TensorFlow C++, by running ' +\n            'npm i @tensorflow/tfjs-node, ' +\n            'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +\n            'Then call require(\\'@tensorflow/tfjs-node\\'); (-gpu ' +\n            'suffix for CUDA) at the start of your program. ' +\n            'Visit https://github.com/tensorflow/tfjs-node for more details.' +\n            '\\n============================');\n      }\n    }\n    const dataId = {};\n\n    this.data.set(dataId, {values, dtype, refCount: 1});\n\n    return dataId;\n  }\n\n  /**\n   * Create a data bucket in cpu backend.\n   * @param shape Shape of the `TensorInfo`.\n   * @param dtype DType of the `TensorInfo`.\n   * @param values The value of the `TensorInfo` stored as a flattened array.\n   */\n  makeTensorInfo(\n      shape: number[], dtype: DataType,\n      values?: backend_util.BackendValues): TensorInfo {\n    const outId = this.write(values, shape, dtype);\n\n    return {dataId: outId, shape, dtype};\n  }\n\n  /** Increase refCount of a `TensorData`. */\n  incRef(dataId: DataId): void {\n    const tensorData = this.data.get(dataId);\n    tensorData.refCount++;\n  }\n\n  /** Decrease refCount of a `TensorData`. */\n  decRef(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n      tensorData.refCount--;\n    }\n  }\n\n  move(\n      dataId: DataId, values: backend_util.BackendValues, shape: number[],\n      dtype: DataType): void {\n    this.data.set(dataId, {values, dtype, refCount: 1});\n  }\n\n  numDataIds(): number {\n    return this.data.numDataIds();\n  }\n\n  async read(dataId: DataId): Promise<backend_util.BackendValues> {\n    return this.readSync(dataId);\n  }\n  readSync(dataId: DataId): backend_util.BackendValues {\n    const {dtype, complexTensorInfos} = this.data.get(dataId);\n\n    if (dtype === 'complex64') {\n      const realValues =\n          this.readSync(complexTensorInfos.real.dataId) as Float32Array;\n      const imagValues =\n          this.readSync(complexTensorInfos.imag.dataId) as Float32Array;\n      return backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    }\n\n    return this.data.get(dataId).values;\n  }\n\n  private bufferSync<R extends Rank>(t: Tensor<R>): TensorBuffer<R> {\n    const data = this.readSync(t.dataId);\n    let decodedData = data as DataValues;\n    if (t.dtype === 'string') {\n      try {\n        // Decode the bytes into string.\n        decodedData = (data as Uint8Array[]).map(d => util.decodeString(d));\n      } catch {\n        throw new Error('Failed to decode encoded string bytes into utf-8');\n      }\n    }\n    return tf.buffer(t.shape, t.dtype, decodedData) as TensorBuffer<R>;\n  }\n\n  makeOutput<T extends Tensor>(\n      values: backend_util.BackendValues, shape: number[], dtype: DataType): T {\n    const dataId = this.write(values, shape, dtype);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  disposeData(dataId: DataId): void {\n    if (this.data.has(dataId)) {\n      const {complexTensorInfos} = this.data.get(dataId);\n\n      if (complexTensorInfos != null) {\n        this.disposeData(complexTensorInfos.real.dataId);\n        this.disposeData(complexTensorInfos.imag.dataId);\n      }\n\n      this.data.delete(dataId);\n    }\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.data.has(dataId)) {\n      const tensorData = this.data.get(dataId);\n\n      tensorData.refCount--;\n\n      if (tensorData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  async time(f: () => void): Promise<BackendTimingInfo> {\n    const start = util.now();\n    f();\n    const kernelMs = util.now() - start;\n    return {kernelMs};\n  }\n\n  memory() {\n    return {\n      // Unreliable due to automatic gc. The numbers above are cumulative.\n      unreliable: true,\n      reasons:\n          ['The reported memory is an upper bound. Due to automatic garbage ' +\n           'collection, the true allocated memory may be less.']\n    };\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    assertNotComplex(x, 'stridedSlice');\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tf.tensor([], outShape) as T;\n    }\n\n    const buffer = tf.buffer(outShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n    for (let i = 0; i < buffer.size; i++) {\n      const loc = buffer.indexToLoc(i);\n\n      const newLoc: number[] = new Array(loc.length);\n      for (let j = 0; j < newLoc.length; j++) {\n        newLoc[j] = loc[j] * strides[j] + begin[j];\n      }\n      buffer.set(xBuf.get(...newLoc), ...loc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  diag(x: Tensor): Tensor {\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const buffer = tf.buffer([x.size, x.size], x.dtype);\n    const vals = buffer.values;\n    for (let i = 0; i < xVals.length; i++) {\n      vals[i * x.size + i] = xVals[i];\n    }\n    return buffer.toTensor();\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = tf.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    assertNotComplex(x, 'reverse');\n\n    const buffer = tf.buffer(x.shape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < buffer.size; i++) {\n      const outLoc = buffer.indexToLoc(i);\n      const inLoc = outLoc.slice();\n      axis.forEach(ax => inLoc[ax] = x.shape[ax] - 1 - inLoc[ax]);\n      buffer.set(xBuf.get(...inLoc), ...outLoc);\n    }\n\n    return buffer.toTensor() as T;\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'neg');\n\n    // TODO(lina128): Use mul directly once neg is modularized.\n    return tf.mul(tf.scalar(-1), x);\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    assertNotComplex(tensors, 'addN');\n\n    const vals = tensors.map(t => this.readSync(t.dataId) as TypedArray);\n    const result = tf.buffer(tensors[0].shape, tensors[0].dtype as 'float32');\n    const resultVals = result.values;\n    for (let i = 0; i < tensors.length; i++) {\n      const currVals = vals[i];\n      for (let j = 0; j < resultVals.length; j++) {\n        resultVals[j] += currVals[j];\n      }\n    }\n    return result.toTensor() as T;\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n\n    // TODO(lina128): Use sub directly once softmax is modularized.\n    const a = tf.sub(logits, maxLogit.reshape(expandedShape));\n    const b = tf.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax\n    // kernel modularization.\n    return tf.div(b, sumExp);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    assertNotComplex([a, b], 'pow');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.pow(aValue, bValue)) as\n        T;\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    assertNotComplex([a, b], 'matMul');\n\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const leftDim = transposeA ? a.shape[2] : a.shape[1];\n    const rightDim = transposeB ? b.shape[1] : b.shape[2];\n    const batchDim = a.shape[0];\n\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const [aBatch, aOuterStep, aInnerStep] = transposeA ?\n        [a.strides[0], 1, a.strides[1]] :\n        [a.strides[0], a.strides[1], 1];\n    const [bInnerStep, bOuterStep, bBatch] = transposeB ?\n        [1, b.strides[1], b.strides[0]] :\n        [b.strides[1], 1, b.strides[0]];\n\n    const size = leftDim * rightDim;\n    const result = tf.buffer([batchDim, leftDim, rightDim], a.dtype);\n    const resVals = result.values as TypedArray;\n    const blockSize = this.blockSize;\n\n    for (let b = 0; b < batchDim; b++) {\n      for (let i0 = 0; i0 < leftDim; i0 += blockSize) {\n        for (let j0 = 0; j0 < rightDim; j0 += blockSize) {\n          for (let k0 = 0; k0 < sharedDim; k0 += blockSize) {\n            // for when blockSize doesn't evenly divide the input\n            const iBlock = Math.min(i0 + blockSize, leftDim);\n            const jBlock = Math.min(j0 + blockSize, rightDim);\n            const kBlock = Math.min(k0 + blockSize, sharedDim);\n\n            for (let i = i0; i < iBlock; i++) {\n              for (let j = j0; j < jBlock; j++) {\n                let sum = 0.0;\n\n                for (let k = k0; k < kBlock; k++) {\n                  sum += aValues[b * aBatch + i * aOuterStep + k * aInnerStep] *\n                      bValues[k * bInnerStep + j * bOuterStep + b * bBatch];\n                }\n                resVals[b * size + (i * rightDim + j)] += sum;\n              }\n            }\n          }\n        }\n      }\n    }\n    return result.toTensor() as Tensor3D;\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    let result = this.batchMatMul(a, b, transposeA, transposeB);\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedBatchMatMul is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor3D;\n    }\n\n    return result;\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'floorDiv');\n\n    const op = (a: number, b: number) => Math.floor(a / b);\n    const outputDtype = 'int32';\n    return this.broadcastedBinaryOp(a, b, outputDtype, op);\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let sum = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        sum += aVals[offset + j];\n      }\n      vals[i] = sum;\n    }\n    return result;\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'sum');\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(outShape, resultDtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let prod = 1;\n      for (let j = 0; j < reduceSize; ++j) {\n        prod *= aVals[offset + j];\n      }\n      vals[i] = prod;\n    }\n    return result;\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    assertNotComplex(x, 'unsortedSegmentSum');\n\n    const res = [];\n\n    // Reshape the segment id's so that they can be broadcast with\n    // x. The new shape should be [segmentIds.shape, 1, ..., 1]\n    const numIters = x.rank - segmentIds.rank;\n    for (let i = 0; i < numIters; ++i) {\n      segmentIds = segmentIds.expandDims(i + 1);\n    }\n\n    for (let i = 0; i < numSegments; ++i) {\n      const segmentId = tf.scalar(i, 'int32');\n      const mask = tf.equal(segmentId, segmentIds).asType('float32');\n      const sum = mask.mul(x).sum(0);\n      res.push(sum);\n    }\n\n    return tf.stack(res);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMin');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMin', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      let minIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n          minIndex = j;\n        }\n      }\n      vals[i] = minIndex;\n    }\n    return result;\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    assertNotComplex(x, 'argMax');\n\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims('argMax', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, 'int32');\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let max = aVals[offset];\n      let maxIndex = 0;\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value > max) {\n          max = value;\n          maxIndex = j;\n        }\n      }\n      vals[i] = maxIndex;\n    }\n    return result;\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    assertNotComplex(x, 'cumsum');\n\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `backend.cumsum in CPU expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const resultDtype = upcastType(x.dtype, 'int32');\n    const result = tf.zeros(x.shape, resultDtype);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    const finalDim = x.shape[x.rank - 1];\n    const indexAdjuster = reverse ?\n        (i: number, j: number) => i + finalDim - j - 1 :\n        (i: number, j: number) => i + j;\n    for (let i = 0; i < aVals.length; i += finalDim) {\n      for (let j = 0; j < finalDim; j++) {\n        const idx = indexAdjuster(i, j);\n        if (j === 0) {\n          vals[idx] = exclusive ? 0 : aVals[idx];\n        } else {\n          const prevIdx = indexAdjuster(i, j - 1);\n          vals[idx] = exclusive ? aVals[prevIdx] + vals[prevIdx] :\n                                  aVals[idx] + vals[prevIdx];\n        }\n      }\n    }\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'equal');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal === bVal) ? 1 : 0;\n    });\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'notEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal !== bVal) ? 1 : 0;\n    });\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'less');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal < bVal) ? 1 : 0;\n    });\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'lessEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal <= bVal) ? 1 : 0;\n    });\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greater');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal > bVal) ? 1 : 0;\n    });\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'greaterEqual');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return (aVal >= bVal) ? 1 : 0;\n    });\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalAnd');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal && bVal;\n    });\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'logicalOr');\n\n    return this.broadcastedBinaryOp(a, b, 'bool', (aVal, bVal) => {\n      return aVal || bVal;\n    });\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([condition, a, b], 'select');\n\n    const values = this.readSync(condition.dataId) as TypedArray;\n    const aValues = this.readSync(a.dataId) as TypedArray;\n    const bValues = this.readSync(b.dataId) as TypedArray;\n    const result = tf.zeros(a.shape, upcastType(a.dtype, b.dtype));\n    const newValues = this.readSync(result.dataId) as TypedArray;\n    let index = 0;\n    const offset = condition.rank === 0 || condition.rank > 1 || a.rank === 1 ?\n        1 :\n        util.sizeFromShape(a.shape.slice(1));\n\n    for (let i = 0; i < values.length; i++) {\n      for (let j = 0; j < offset; j++) {\n        if (values[i] === 1) {\n          newValues[index++] = aValues[i];\n        } else {\n          newValues[index++] = bValues[i];\n        }\n      }\n    }\n\n    return result;\n  }\n\n  where(condition: Tensor): Tensor2D {\n    assertNotComplex([condition], 'where');\n\n    const condVals = this.readSync(condition.dataId) as TypedArray;\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    assertNotComplex(x, 'topk');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'min');\n\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let min = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        if (value < min) {\n          min = value;\n        }\n      }\n      vals[i] = min;\n    }\n    return result;\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'minimum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.min(aVal, bVal));\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'mod');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const rem = aVal % bVal;\n      if ((aVal < 0 && bVal < 0) || (aVal >= 0 && bVal >= 0)) {\n        return rem;\n      } else {\n        return (rem + bVal) % bVal;\n      }\n    });\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'maximum');\n\n    return this.broadcastedBinaryOp(\n        a, b, a.dtype, (aVal, bVal) => Math.max(aVal, bVal));\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'all');\n\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let all = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        all = all && value;\n      }\n      vals[i] = all;\n    }\n    return result;\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    assertNotComplex(x, 'any');\n\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const result = tf.zeros(outShape, x.dtype);\n    const reduceSize = util.sizeFromShape(reduceShape);\n    const vals = this.readSync(result.dataId) as TypedArray;\n\n    const aVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < vals.length; ++i) {\n      const offset = i * reduceSize;\n      let anyVal = aVals[offset];\n      for (let j = 0; j < reduceSize; ++j) {\n        const value = aVals[offset + j];\n        anyVal = anyVal || value;\n      }\n      vals[i] = anyVal;\n    }\n    return result;\n  }\n\n  squaredDifference(a: Tensor, b: Tensor): Tensor {\n    assertNotComplex([a, b], 'squaredDifference');\n\n    return this.broadcastedBinaryOp(a, b, a.dtype, (aVal, bVal) => {\n      const diff = aVal - bVal;\n      return diff * diff;\n    });\n  }\n\n  linear<T extends Tensor>(x: T): T {\n    return x;\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.max(0, inVals[i]);\n    }\n    return res as T;\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    assertNotComplex(x, 'relu');\n\n    const res = tf.zeros(x.shape, x.dtype);\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const inVals = this.readSync(x.dataId) as TypedArray;\n    for (let i = 0; i < inVals.length; ++i) {\n      resVals[i] = Math.min(Math.max(0, inVals[i]), 6);\n    }\n    return res as T;\n  }\n\n  prelu<T extends Tensor>(x: T, a: T): T {\n    assertNotComplex([x, a], 'prelu');\n\n    return this.broadcastedBinaryOp(\n               x, a, x.dtype,\n               (xValue, aValue) => xValue < 0 ? aValue * xValue : xValue) as T;\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    assertNotComplex([dy, y], 'eluDer');\n\n    const resultValues = new Float32Array(y.size);\n    const values = this.readSync(y.dataId) as TypedArray;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    for (let i = 0; i < values.length; ++i) {\n      const v = values[i];\n      if (v >= 1) {\n        resultValues[i] = dyValues[i];\n      } else {\n        resultValues[i] = dyValues[i] * (v + 1);\n      }\n    }\n    return this.makeOutput(resultValues, y.shape, 'float32');\n  }\n\n  atan2<T extends Tensor>(a: T, b: T): T {\n    assertNotComplex([a, b], 'atan2');\n\n    return this.broadcastedBinaryOp(\n               a, b, a.dtype, (aValue, bValue) => Math.atan2(aValue, bValue)) as\n        T;\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.conv2d(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedConv2d is modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, filter], 'conv2d');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n\n    const xBatchStride = x.strides[0];\n    const xRowStride = isChannelsLast ? x.strides[1] : x.strides[2];\n    const xColStride = isChannelsLast ? x.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : x.strides[1];\n    const yBatchStride = y.strides[0];\n    const yRowStride = isChannelsLast ? y.strides[1] : y.strides[2];\n    const yColStride = isChannelsLast ? y.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : y.strides[1];\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * xBatchStride;\n      const yOffset1 = b * yBatchStride;\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * yRowStride;\n        const xRCorner = yR * convInfo.strideHeight - padTop;\n        for (let wR = 0; wR < filterHeight; wR++) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * xRowStride;\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * yColStride;\n            const xCCorner = yC * convInfo.strideWidth - padLeft;\n            for (let wC = 0; wC < filterWidth; wC++) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * xColStride;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1 * xChannelStride];\n                for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                  yVals[yOffset3 + d2 * yChannelStride] +=\n                      xVal * wVals[wOffset3 + d2];\n                }\n                wOffset3 += convInfo.outChannels;\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor() as Tensor4D;\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padFront = convInfo.padInfo.front;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const y = tf.buffer<Rank.R5>(convInfo.outShape, x.dtype as 'float32');\n\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yF = 0; yF < convInfo.outDepth; ++yF) {\n        const yOffset2 = yOffset1 + yF * y.strides[1];\n        const xFCorner = yF * convInfo.strideDepth - padFront;\n        for (let wF = 0; wF < filterDepth; wF++) {\n          const xF = xFCorner + wF * dilationDepth;\n          if (xF < 0 || xF >= convInfo.inDepth) {\n            continue;\n          }\n          const wOffset1 = wF * filter.strides[0];\n          const xOffset2 = xOffset1 + xF * x.strides[1];\n\n          for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n            const yOffset3 = yOffset2 + yR * y.strides[2];\n            const xRCorner = yR * convInfo.strideHeight - padTop;\n            for (let wR = 0; wR < filterHeight; wR++) {\n              const xR = xRCorner + wR * dilationHeight;\n              if (xR < 0 || xR >= convInfo.inHeight) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wR * filter.strides[1];\n              const xOffset3 = xOffset2 + xR * x.strides[2];\n              for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n                const yOffset4 = yOffset3 + yC * convInfo.outChannels;\n                const xCCorner = yC * convInfo.strideWidth - padLeft;\n                for (let wC = 0; wC < filterWidth; wC++) {\n                  const xC = xCCorner + wC * dilationWidth;\n                  if (xC < 0 || xC >= convInfo.inWidth) {\n                    continue;\n                  }\n                  const wOffset3 = wOffset2 + wC * filter.strides[2];\n                  const xOffset4 = xOffset3 + xC * convInfo.inChannels;\n                  let wOffset4 = wOffset3;\n                  for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                    const xVal = xVals[xOffset4 + d1];\n                    for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n                      yVals[yOffset4 + d2] += xVal * wVals[wOffset4 + d2];\n                    }\n                    wOffset4 += convInfo.outChannels;\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    }\n    return y.toTensor();\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'conv2dDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth,\n      dataFormat\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const xBatchStride = dx.strides[0];\n    const xRowStride = isChannelsLast ? dx.strides[1] : dx.strides[2];\n    const xColStride = isChannelsLast ? dx.strides[2] : 1;\n    const xChannelStride = isChannelsLast ? 1 : dx.strides[1];\n    const yBatchStride = dy.strides[0];\n    const yRowStride = isChannelsLast ? dy.strides[1] : dy.strides[2];\n    const yColStride = isChannelsLast ? dy.strides[2] : 1;\n    const yChannelStride = isChannelsLast ? 1 : dy.strides[1];\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset =\n                    yBatchStride * b + yRowStride * yR + yColStride * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let d2 = 0; d2 < outChannels; ++d2) {\n                  const pixel = dyValues[dyOffset + yChannelStride * d2];\n                  const weight = fltValues[fltOffset + d2];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            const dxOffset = xBatchStride * b + xRowStride * xR +\n                xColStride * xC + xChannelStride * d1;\n            dxValues[dxOffset] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const dx = tf.buffer<Rank.R5>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2, dxS3] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2, fltS3] = filter.strides;\n    const {\n      batchSize,\n      filterDepth,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inDepth,\n      inHeight,\n      inWidth,\n      outChannels,\n      outDepth,\n      outHeight,\n      outWidth,\n      strideDepth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const frontPad = filterDepth - 1 - convInfo.padInfo.front;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        // Frames of depth\n        for (let xF = 0; xF < inDepth; ++xF) {\n          const xFCorner = xF - frontPad;\n          const xFMin = Math.max(0, Math.ceil(xFCorner / strideDepth));\n          const yFMax =\n              Math.min(outDepth, (filterDepth + xFCorner) / strideDepth);\n\n          // Rows as per standard 2d matrix notation\n          for (let xR = 0; xR < inHeight; ++xR) {\n            const xRCorner = xR - topPad;\n            const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n            const yRMax =\n                Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n            // Columns as per standard 2d matrix notation\n            for (let xC = 0; xC < inWidth; ++xC) {\n              const xCCorner = xC - leftPad;\n              const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n              const yCMax =\n                  Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n              let dotProd = 0;\n              for (let yF = xFMin; yF < yFMax; ++yF) {\n                const wF = yF * strideDepth - xFCorner;\n\n                for (let yR = xRMin; yR < yRMax; ++yR) {\n                  const wR = yR * strideHeight - xRCorner;\n\n                  for (let yC = xCMin; yC < yCMax; ++yC) {\n                    const wC = yC * strideWidth - xCCorner;\n                    const dyOffset =\n                        dyS0 * b + dyS1 * yF + dyS2 * yR + dyS3 * yC;\n                    const fltOffset = fltS0 * (filterDepth - 1 - wF) +\n                        fltS1 * (filterHeight - 1 - wR) +\n                        fltS2 * (filterWidth - 1 - wC) + fltS3 * d1;\n\n                    for (let d2 = 0; d2 < outChannels; ++d2) {\n                      const pixel = dyValues[dyOffset + d2];\n                      const weight = fltValues[fltOffset + d2];\n                      dotProd += pixel * weight;\n                    }\n                  }\n                }\n              }\n              dxValues[dxS0 * b + dxS1 * xF + dxS2 * xR + dxS3 * xC + d1] =\n                  dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    assertNotComplex([x, dy], 'conv2dDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n          for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n            // Need to convolve.\n            let dotProd = 0;\n            for (let b = 0; b < convInfo.batchSize; ++b) {\n              for (let yR = yRMin; yR < yRMax; ++yR) {\n                const xR = wR + yR * strideHeight - topPad;\n                for (let yC = yCMin; yC < yCMax; ++yC) {\n                  const xC = wC + yC * strideWidth - leftPad;\n                  if (isChannelsLast) {\n                    dotProd +=\n                        xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n                  } else {\n                    dotProd +=\n                        xBuf.get(b, d1, xR, xC) * dyBuf.get(b, d2, yR, yC);\n                  }\n                }\n              }\n            }\n            dW.set(dotProd, wR, wC, d1, d2);\n          }\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const dw = tf.buffer<Rank.R5>(convInfo.filterShape, 'float32');\n    const dwValues = dw.values;\n    const [dwS0, dwS1, dwS2, dwS3] = dw.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2, dyS3] = dy.strides;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const [xS0, xS1, xS2, xS3] = x.strides;\n\n    const frontPad = convInfo.padInfo.front;\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n\n    for (let wF = 0; wF < filterDepth; ++wF) {\n      const yFMin = Math.max(0, Math.ceil((frontPad - wF) / strideDepth));\n      const yFMax = Math.min(\n          convInfo.outDepth, (convInfo.inDepth + frontPad - wF) / strideDepth);\n      const wOffset1 = wF * dwS0;\n\n      for (let wR = 0; wR < filterHeight; ++wR) {\n        const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n        const yRMax = Math.min(\n            convInfo.outHeight,\n            (convInfo.inHeight + topPad - wR) / strideHeight);\n        const wOffset2 = wR * dwS1 + wOffset1;\n\n        for (let wC = 0; wC < filterWidth; ++wC) {\n          const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n          const yCMax = Math.min(\n              convInfo.outWidth,\n              (convInfo.inWidth + leftPad - wC) / strideWidth);\n          const wOffset3 = wC * dwS2 + wOffset2;\n\n          for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n            const wOffset4 = d1 * dwS3 + wOffset3;\n\n            for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n              let dotProd = 0;\n              for (let b = 0; b < convInfo.batchSize; ++b) {\n                const xOffset1 = b * xS0;\n                const yOffset1 = b * dyS0;\n\n                for (let yF = yFMin; yF < yFMax; ++yF) {\n                  const xF = wF + yF * strideDepth - frontPad;\n                  const xOffset2 = xF * xS1 + xOffset1;\n                  const yOffset2 = yF * dyS1 + yOffset1;\n\n                  for (let yR = yRMin; yR < yRMax; ++yR) {\n                    const xR = wR + yR * strideHeight - topPad;\n                    const xOffset3 = xR * xS2 + xOffset2;\n                    const yOffset3 = yR * dyS2 + yOffset2;\n\n                    for (let yC = yCMin; yC < yCMax; ++yC) {\n                      const xC = wC + yC * strideWidth - leftPad;\n                      const xOffset4 = xC * xS3 + xOffset3;\n                      const yOffset4 = yC * dyS3 + yOffset3;\n\n                      dotProd +=\n                          xValues[xOffset4 + d1] * dyValues[yOffset4 + d2];\n                    }\n                  }\n                }\n              }\n              dwValues[wOffset4 + d2] = dotProd;\n            }\n          }\n        }\n      }\n    }\n    return dw.toTensor();\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    let result = this.depthwiseConv2D(input, filter, convInfo);\n\n    if (bias) {\n      // TODO(lina128): Use add directly once fusedDepthwiseConv2D is\n      // modularized.\n      result = tf.add(result, bias);\n    }\n    if (activation) {\n      result =\n          mapActivation(this, result, activation, preluActivationWeights) as\n          Tensor4D;\n    }\n    return result;\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, filter], 'depthwiseConv2D');\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const padLeft = convInfo.padInfo.left;\n    const padTop = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n    const y = tf.buffer(convInfo.outShape, x.dtype as 'float32');\n    const xVals = this.readSync(x.dataId) as TypedArray;\n    const wVals = this.readSync(filter.dataId) as TypedArray;\n    const yVals = y.values;\n\n    for (let b = 0; b < convInfo.batchSize; ++b) {\n      const xOffset1 = b * x.strides[0];\n      const yOffset1 = b * y.strides[0];\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const yOffset2 = yOffset1 + yR * y.strides[1];\n        const xRCorner = yR * convInfo.strideHeight - padLeft;\n        for (let wR = 0; wR < filterHeight; ++wR) {\n          const xR = xRCorner + wR * dilationHeight;\n          if (xR < 0 || xR >= convInfo.inHeight) {\n            continue;\n          }\n          const wOffset1 = wR * filter.strides[0];\n          const xOffset2 = xOffset1 + xR * x.strides[1];\n          for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n            const yOffset3 = yOffset2 + yC * y.strides[2];\n            const xCCorner = yC * convInfo.strideWidth - padTop;\n            for (let wC = 0; wC < filterWidth; ++wC) {\n              const xC = xCCorner + wC * dilationWidth;\n              if (xC < 0 || xC >= convInfo.inWidth) {\n                continue;\n              }\n              const wOffset2 = wOffset1 + wC * filter.strides[1];\n              const xOffset3 = xOffset2 + xC * convInfo.inChannels;\n              let yOffset4 = yOffset3;\n              let wOffset3 = wOffset2;\n              for (let d1 = 0; d1 < convInfo.inChannels; ++d1) {\n                const xVal = xVals[xOffset3 + d1];\n                for (let q = 0; q < chMul; ++q) {\n                  yVals[yOffset4 + q] += xVal * wVals[wOffset3 + q];\n                }\n                yOffset4 += chMul;\n                wOffset3 += chMul;\n              }\n            }\n          }\n        }\n      }\n    }\n\n    return y.toTensor() as Tensor4D;\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([dy, filter], 'depthwiseConv2DDerInput');\n\n    const dx = tf.buffer<Rank.R4>(convInfo.inShape, 'float32');\n    const dxValues = dx.values;\n    const [dxS0, dxS1, dxS2] = dx.strides;\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const [dyS0, dyS1, dyS2] = dy.strides;\n    const fltValues = this.readSync(filter.dataId) as TypedArray;\n    const [fltS0, fltS1, fltS2] = filter.strides;\n    const {\n      batchSize,\n      filterHeight,\n      filterWidth,\n      inChannels,\n      inHeight,\n      inWidth,\n      outChannels,\n      outHeight,\n      outWidth,\n      strideHeight,\n      strideWidth\n    } = convInfo;\n    const topPad = filterHeight - 1 - convInfo.padInfo.top;\n    const leftPad = filterWidth - 1 - convInfo.padInfo.left;\n    const chMul = outChannels / inChannels;\n\n    for (let b = 0; b < batchSize; ++b) {\n      for (let d1 = 0; d1 < inChannels; ++d1) {\n        for (let xR = 0; xR < inHeight; ++xR) {\n          const xRCorner = xR - topPad;\n          const xRMin = Math.max(0, Math.ceil(xRCorner / strideHeight));\n          const yRMax =\n              Math.min(outHeight, (filterHeight + xRCorner) / strideHeight);\n\n          for (let xC = 0; xC < inWidth; ++xC) {\n            const xCCorner = xC - leftPad;\n            const xCMin = Math.max(0, Math.ceil(xCCorner / strideWidth));\n            const yCMax =\n                Math.min(outWidth, (filterWidth + xCCorner) / strideWidth);\n\n            let dotProd = 0;\n            for (let yR = xRMin; yR < yRMax; ++yR) {\n              const wR = yR * strideHeight - xRCorner;\n\n              for (let yC = xCMin; yC < yCMax; ++yC) {\n                const wC = yC * strideWidth - xCCorner;\n                const dyOffset = dyS0 * b + dyS1 * yR + dyS2 * yC;\n                const fltOffset = fltS0 * (filterHeight - 1 - wR) +\n                    fltS1 * (filterWidth - 1 - wC) + fltS2 * d1;\n\n                for (let dm = 0; dm < chMul; ++dm) {\n                  const d2 = d1 * chMul + dm;\n                  const pixel = dyValues[dyOffset + d2];\n                  const weight = fltValues[fltOffset + dm];\n                  dotProd += pixel * weight;\n                }\n              }\n            }\n            dxValues[dxS0 * b + dxS1 * xR + dxS2 * xC + d1] = dotProd;\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    assertNotComplex([x, dy], 'depthwiseConv2DDerFilter');\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dW = tf.buffer<Rank.R4>(convInfo.filterShape, 'float32');\n\n    const leftPad = convInfo.padInfo.left;\n    const topPad = convInfo.padInfo.top;\n    const chMul = convInfo.outChannels / convInfo.inChannels;\n\n    const xBuf = this.bufferSync(x);\n    const dyBuf = this.bufferSync(dy);\n    for (let wR = 0; wR < filterHeight; ++wR) {\n      const yRMin = Math.max(0, Math.ceil((topPad - wR) / strideHeight));\n      const yRMax = Math.min(\n          convInfo.outHeight, (convInfo.inHeight + topPad - wR) / strideHeight);\n\n      for (let wC = 0; wC < filterWidth; ++wC) {\n        const yCMin = Math.max(0, Math.ceil((leftPad - wC) / strideWidth));\n        const yCMax = Math.min(\n            convInfo.outWidth, (convInfo.inWidth + leftPad - wC) / strideWidth);\n\n        for (let d2 = 0; d2 < convInfo.outChannels; ++d2) {\n          const d1 = Math.trunc(d2 / chMul);\n          const dm = d2 % chMul;\n\n          let dotProd = 0;\n          for (let b = 0; b < convInfo.batchSize; ++b) {\n            for (let yR = yRMin; yR < yRMax; ++yR) {\n              const xR = wR + yR * strideHeight - topPad;\n              for (let yC = yCMin; yC < yCMax; ++yC) {\n                const xC = wC + yC * strideWidth - leftPad;\n                dotProd += xBuf.get(b, xR, xC, d1) * dyBuf.get(b, yR, yC, d2);\n              }\n            }\n          }\n          dW.set(dotProd, wR, wC, d1, dm);\n        }\n      }\n    }\n    return dW.toTensor();\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    assertNotComplex(x, 'tile');\n    return tile(this.bufferSync(x), reps) as T;\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    assertNotComplex([x, indices], 'gather');\n\n    const newShape: number[] = x.shape.slice();\n    const indicesValues = this.readSync(indices.dataId) as TypedArray;\n    newShape[axis] = indicesValues.length;\n    const result = tf.buffer(newShape, x.dtype);\n    const xBuf = this.bufferSync(x);\n\n    for (let i = 0; i < result.size; ++i) {\n      const newLoc = result.indexToLoc(i);\n\n      const originalLoc: number[] = newLoc.slice();\n      originalLoc[axis] = indicesValues[newLoc[axis]];\n\n      const originalIndex = xBuf.locToIndex(originalLoc);\n      result.values[i] = xBuf.values[originalIndex];\n    }\n    return result.toTensor() as T;\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    assertNotComplex([x], 'batchToSpaceND');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return tf.transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  private pool3d(\n      x: Tensor5D, convInfo: backend_util.Conv3DInfo,\n      poolType: 'max'|'avg'): Tensor5D {\n    assertNotComplex(x, 'pool3d');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const initialValue =\n        (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                              Number.POSITIVE_INFINITY);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = tf.buffer(convInfo.outShape, x.dtype);\n    const outputVals = output.values;\n\n    const outputBatchStrides = convInfo.outShape[1] * convInfo.outShape[2] *\n        convInfo.outShape[3] * convInfo.outShape[4];\n    const outputDepthStrides =\n        convInfo.outShape[2] * convInfo.outShape[3] * convInfo.outShape[4];\n    const outputRowStrides = convInfo.outShape[3] * convInfo.outShape[4];\n    const outputColStrides = convInfo.outShape[4];\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      const outputBatchOffset = batch * outputBatchStrides;\n      const inputBatchOffset = batch * x.strides[0];\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          const outputDepthOffset =\n              outputBatchOffset + yDepth * outputDepthStrides;\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            const outputRowOffset = outputDepthOffset + yRow * outputRowStrides;\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n              // Shader code begins\n              const outputColOffset = outputRowOffset + yCol * outputColStrides;\n              let minMaxValue = initialValue;\n              let avgValue = 0;\n              let count = 0;\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const xDepthOffset = inputBatchOffset + xDepth * x.strides[1];\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const xRowOffset = xDepthOffset + xRow * x.strides[2];\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const xColOffset = xRowOffset + xCol * x.strides[3];\n                    const pixel = xValues[xColOffset + channel];\n                    if ((poolType === 'max' && pixel > minMaxValue)) {\n                      minMaxValue = pixel;\n                    } else if (poolType === 'avg') {\n                      avgValue += pixel;\n                      count++;\n                    }\n                    if (isNaN(minMaxValue)) {\n                      break;\n                    }\n                  }\n                  if (isNaN(minMaxValue)) {\n                    break;\n                  }\n                }\n                if (isNaN(minMaxValue)) {\n                  break;\n                }\n              }\n              const outputOffset = outputColOffset + channel;\n              outputVals[outputOffset] =\n                  poolType === 'avg' ? avgValue / count : minMaxValue;\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor5D;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'avgPool3d');\n\n    return this.pool3d(x, convInfo, 'avg').toFloat();\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([dy, x], 'avgPool3dBackprop');\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins.\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel;\n                  }\n                }\n              }\n              dx.set(\n                  dotProd * avgMultiplier, batch, dxDepth, dxRow, dxCol,\n                  channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex(x, 'maxPool3d');\n\n    return this.pool3d(x, convInfo, 'max').toFloat();\n  }\n\n  private maxPool3dPositions(x: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const maxPositions = tf.buffer(convInfo.outShape, 'int32');\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    const xBuf = this.bufferSync(x);\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let yDepth = 0; yDepth < convInfo.outDepth; ++yDepth) {\n          const xDepthCorner = yDepth * strideDepth - padFront;\n          let xDepthMin = xDepthCorner;\n          while (xDepthMin < 0) {\n            xDepthMin += dilationDepth;\n          }\n          const xDepthMax =\n              Math.min(convInfo.inDepth, effectiveFilterDepth + xDepthCorner);\n          for (let yRow = 0; yRow < convInfo.outHeight; ++yRow) {\n            const xRowCorner = yRow * strideHeight - padTop;\n            let xRowMin = xRowCorner;\n            while (xRowMin < 0) {\n              xRowMin += dilationHeight;\n            }\n            const xRowMax =\n                Math.min(convInfo.inHeight, effectiveFilterHeight + xRowCorner);\n            for (let yCol = 0; yCol < convInfo.outWidth; ++yCol) {\n              const xColCorner = yCol * strideWidth - padLeft;\n              let xColMin = xColCorner;\n              while (xColMin < 0) {\n                xColMin += dilationWidth;\n              }\n              const xColMax =\n                  Math.min(convInfo.inWidth, effectiveFilterWidth + xColCorner);\n\n              // Shader code begins\n              let maxValue = Number.NEGATIVE_INFINITY;\n              let maxPosition = -1;\n\n              for (let xDepth = xDepthMin; xDepth < xDepthMax;\n                   xDepth += dilationDepth) {\n                const wDepth = xDepth - xDepthCorner;\n                for (let xRow = xRowMin; xRow < xRowMax;\n                     xRow += dilationHeight) {\n                  const wRow = xRow - xRowCorner;\n                  for (let xCol = xColMin; xCol < xColMax;\n                       xCol += dilationWidth) {\n                    const wCol = xCol - xColCorner;\n                    const pixel = xBuf.get(batch, xDepth, xRow, xCol, channel);\n                    if (pixel >= maxValue) {\n                      maxValue = pixel;\n                      maxPosition = wDepth * effectiveFilterHeight *\n                              effectiveFilterWidth +\n                          wRow * effectiveFilterHeight + wCol;\n                    }\n                  }\n                }\n              }\n\n              maxPositions.set(maxPosition, batch, yDepth, yRow, yCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return maxPositions.toTensor() as Tensor5D;\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    assertNotComplex([x, y], 'maxPool3dBackprop');\n\n    const maxPositions = this.maxPool3dPositions(x, convInfo);\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const dx = tf.buffer<Rank.R5>(x.shape, 'float32');\n\n    const maxPosBuf = this.bufferSync(maxPositions);\n    const dyBuf = this.bufferSync(dy);\n\n    for (let batch = 0; batch < convInfo.batchSize; ++batch) {\n      for (let channel = 0; channel < convInfo.inChannels; ++channel) {\n        for (let dxDepth = 0; dxDepth < convInfo.inDepth; ++dxDepth) {\n          for (let dxRow = 0; dxRow < convInfo.inHeight; ++dxRow) {\n            for (let dxCol = 0; dxCol < convInfo.inWidth; ++dxCol) {\n              // Shader code begins\n              const dyDepthCorner = dxDepth - padFront;\n              const dyRowCorner = dxRow - padTop;\n              const dyColCorner = dxCol - padLeft;\n              let dotProd = 0;\n              for (let wDepth = 0; wDepth < effectiveFilterDepth;\n                   wDepth += dilationDepth) {\n                const dyDepth = (dyDepthCorner + wDepth) / strideDepth;\n                if (dyDepth < 0 || dyDepth >= convInfo.outDepth ||\n                    Math.floor(dyDepth) !== dyDepth) {\n                  continue;\n                }\n                for (let wRow = 0; wRow < effectiveFilterHeight;\n                     wRow += dilationHeight) {\n                  const dyRow = (dyRowCorner + wRow) / strideHeight;\n                  if (dyRow < 0 || dyRow >= convInfo.outHeight ||\n                      Math.floor(dyRow) !== dyRow) {\n                    continue;\n                  }\n                  for (let wCol = 0; wCol < effectiveFilterWidth;\n                       wCol += dilationWidth) {\n                    const dyCol = (dyColCorner + wCol) / strideWidth;\n                    if (dyCol < 0 || dyCol >= convInfo.outWidth ||\n                        Math.floor(dyCol) !== dyCol) {\n                      continue;\n                    }\n\n                    const maxPos = effectiveFilterDepth *\n                            effectiveFilterHeight * effectiveFilterWidth -\n                        1 -\n                        maxPosBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    const curPos =\n                        wDepth * effectiveFilterHeight * effectiveFilterWidth +\n                        wRow * effectiveFilterWidth + wCol;\n\n                    const mask = maxPos === curPos ? 1 : 0;\n                    if (mask === 0) {\n                      continue;\n                    }\n\n                    const pixel =\n                        dyBuf.get(batch, dyDepth, dyRow, dyCol, channel);\n                    dotProd += pixel * mask;\n                  }\n                }\n              }\n              dx.set(dotProd, batch, dxDepth, dxRow, dxCol, channel);\n            }\n          }\n        }\n      }\n    }\n    return dx.toTensor();\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeBilinear');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result = new Float32Array(\n        util.sizeFromShape([batch, newHeight, newWidth, numChannels]));\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n    let outputIdx = 0;\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n    for (let b = 0; b < batch; b++) {\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceRowFloor = Math.floor(sourceFracRow);\n        const rowFrac = sourceFracRow - sourceRowFloor;\n        const sourceRowCeil = Math.min(oldHeight - 1, Math.ceil(sourceFracRow));\n        const topRowOffset = b * x.strides[0] + sourceRowFloor * x.strides[1];\n        const botRowOffset = b * x.strides[0] + sourceRowCeil * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceColFloor = Math.floor(sourceFracCol);\n          const colFrac = sourceFracCol - sourceColFloor;\n          const sourceColCeil =\n              Math.min(oldWidth - 1, Math.ceil(sourceFracCol));\n          const topLeftOffest = topRowOffset + sourceColFloor * x.strides[2];\n          const botLeftOffset = botRowOffset + sourceColFloor * x.strides[2];\n          const topRightOffset = topRowOffset + sourceColCeil * x.strides[2];\n          const botRightOffest = botRowOffset + sourceColCeil * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n\n            // Compute the fractional index of the source.\n            const topLeft = xValues[topLeftOffest + d];\n            const bottomLeft = xValues[botLeftOffset + d];\n            const topRight = xValues[topRightOffset + d];\n            const bottomRight = xValues[botRightOffest + d];\n\n            const top = topLeft + (topRight - topLeft) * colFrac;\n            const bottom = bottomLeft + (bottomRight - bottomLeft) * colFrac;\n            const newValue = top + (bottom - top) * rowFrac;\n\n            result[outputIdx++] = newValue;\n          }\n        }\n      }\n    }\n    return tf.tensor(result, [batch, newHeight, newWidth, numChannels]);\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeBilinearBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass and add the\n    // corresponding coefficient from dy to the gradient (with some\n    // interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/3039375c86a5bbc9610c7725dcaa95d635f87ba2/tensorflow/core/kernels/resize_bilinear_op.cc#L275\n\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    let offset = 0;\n    for (let b = 0; b < batch; b++) {\n      const bOffset = b * x.strides[0];\n      for (let r = 0; r < yHeight; r++) {\n        const dxR = r * heightScale;\n        const topDxRIndex = Math.floor(dxR);\n        const bottomDxRIndex = Math.min(Math.ceil(dxR), xHeight - 1);\n\n        const topDxROffset = bOffset + topDxRIndex * x.strides[1];\n        const bottomDxROffset = bOffset + bottomDxRIndex * x.strides[1];\n\n        const dxRLerp = dxR - topDxRIndex;\n        const inverseDxRLerp = 1.0 - dxRLerp;\n        for (let c = 0; c < yWidth; c++) {\n          const dxC = c * widthScale;\n          const leftDxCIndex = Math.floor(dxC);\n          const rightDxCIndex = Math.min(Math.ceil(dxC), xWidth - 1);\n          const dxCLerp = dxC - leftDxCIndex;\n          const inverseDxCLerp = 1.0 - dxCLerp;\n\n          const topLeftRCOffset = topDxROffset + leftDxCIndex * x.strides[2];\n          const topRightRCOffset = topDxROffset + rightDxCIndex * x.strides[2];\n          const bottomLeftRCOffset =\n              bottomDxROffset + leftDxCIndex * x.strides[2];\n          const bottomRightRCOffset =\n              bottomDxROffset + rightDxCIndex * x.strides[2];\n\n          const inverseDxRLerpTimesInverseDxCLerp =\n              inverseDxRLerp * inverseDxCLerp;\n          const inverseDxRLerpTimesDxCLerp = inverseDxRLerp * dxCLerp;\n          const dxRLerpTimesInverseDxCLerp = dxRLerp * inverseDxCLerp;\n          const dxRLerpTimesDxCLerp = dxRLerp * dxCLerp;\n          for (let d = 0; d < depth; d++) {\n            const dyVal = dyValues[offset++];\n            output[topLeftRCOffset + d] +=\n                dyVal * inverseDxRLerpTimesInverseDxCLerp;\n            output[topRightRCOffset + d] += dyVal * inverseDxRLerpTimesDxCLerp;\n            output[bottomLeftRCOffset + d] +=\n                dyVal * dxRLerpTimesInverseDxCLerp;\n            output[bottomRightRCOffset + d] += dyVal * dxRLerpTimesDxCLerp;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, [batch, xWidth, xHeight, depth], x.dtype);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    assertNotComplex(x, 'resizeNearestNeighbor');\n\n    const [batch, oldHeight, oldWidth, numChannels] = x.shape;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const output = new Float32Array(batch * newHeight * newWidth * numChannels);\n\n    const effectiveInputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutputSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    const effectiveRowSizeRatio =\n        effectiveInputSize[0] / effectiveOutputSize[0];\n    const effectiveColSizeRatio =\n        effectiveInputSize[1] / effectiveOutputSize[1];\n\n    let outputOffset = 0;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < newHeight; r++) {\n        const sourceFracRow = effectiveRowSizeRatio * r;\n        const sourceNearestRow = Math.min(\n            oldHeight - 1,\n            alignCorners ? Math.round(sourceFracRow) :\n                           Math.floor(sourceFracRow));\n        const rowOffset = batchOffset + sourceNearestRow * x.strides[1];\n        for (let c = 0; c < newWidth; c++) {\n          const sourceFracCol = effectiveColSizeRatio * c;\n          const sourceNearestCol = Math.min(\n              oldWidth - 1,\n              alignCorners ? Math.round(sourceFracCol) :\n                             Math.floor(sourceFracCol));\n          const colOffset = rowOffset + sourceNearestCol * x.strides[2];\n          for (let d = 0; d < numChannels; d++) {\n            // Begin shader.\n            // Compute the fractional index of the source.\n            const newVal = xValues[colOffset + d];\n            output[outputOffset++] = newVal;\n          }\n        }\n      }\n    }\n    return tf.tensor(\n        output, [batch, newHeight, newWidth, numChannels], x.dtype);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    assertNotComplex([dy, x], 'resizeNearestNeighborBackprop');\n\n    const [batch, xHeight, xWidth, depth] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    const output = new Float32Array(batch * xHeight * xWidth * depth);\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n\n    // In the backwards pass, we want to find the pixels that were generated\n    // for each pixel in the input image the forward pass\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    // Loop over the output space.\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = b * x.strides[0];\n      for (let r = 0; r < xHeight; r++) {\n        const rowOffset = batchOffset + r * x.strides[1];\n\n        // Compute bounds for where in dy we will look\n        const startRLerp = Math.floor(r * invHeightScale);\n        const startDyR = Math.floor(startRLerp - (winHeight / 2));\n        for (let c = 0; c < xWidth; c++) {\n          const colOffset = rowOffset + c * x.strides[2];\n\n          // Compute bounds for where in dy we will look\n          const startCLerp = Math.floor(c * invWidthScale);\n          const startDyC = Math.floor(startCLerp - (winWidth / 2));\n\n          for (let d = 0; d < depth; d++) {\n            let accum = 0;\n            // loop over dy\n\n            for (let dyRIndex = 0; dyRIndex < winHeight; dyRIndex++) {\n              const dyR = dyRIndex + startDyR;\n              // Guard against the window exceeding the bounds of dy\n              if (dyR < 0 || dyR >= yHeight) {\n                continue;\n              }\n\n              const dyROffset = batchOffset + dyR * dy.strides[1];\n              const sourceFracRow = dyR * heightScale;\n              const sourceNearestRow = Math.min(\n                  xHeight - 1,\n                  alignCorners ? Math.round(sourceFracRow) :\n                                 Math.floor(sourceFracRow));\n              if (r !== sourceNearestRow) {\n                continue;\n              }\n              for (let dyCIndex = 0; dyCIndex < winWidth; dyCIndex++) {\n                const dyC = dyCIndex + startDyC;\n                // Guard against the window exceeding the bounds of dy\n                if (dyC < 0 || dyC >= yWidth) {\n                  continue;\n                }\n\n                const dyCOffset = dyROffset + dyC * dy.strides[2];\n                const sourceFracCol = dyC * widthScale;\n                const sourceNearestCol = Math.min(\n                    xWidth - 1,\n                    alignCorners ? Math.round(sourceFracCol) :\n                                   Math.floor(sourceFracCol));\n\n                if (c === sourceNearestCol) {\n                  accum += dyValues[dyCOffset + d];\n                }\n              }\n            }\n            output[colOffset + d] = accum;\n          }\n        }\n      }\n    }\n    return tf.tensor4d(output, x.shape, x.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(x, 'localResponseNormalization4D');\n\n    const channels = x.shape[3];\n    const maxD = channels - 1;\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const size = x.size;\n    const result = new Float32Array(size);\n\n    function sumAcrossChannels(offset: number) {\n      const currentChannel = offset % channels;\n      let beginSumOffset =\n          offset - currentChannel + Math.max(0, currentChannel - depthRadius);\n      const endSumOffset = offset - currentChannel +\n          Math.min(currentChannel + depthRadius, maxD);\n\n      let sum = 0.0;\n      for (; beginSumOffset <= endSumOffset; beginSumOffset++) {\n        const z = xValues[beginSumOffset];\n        sum += z * z;\n      }\n      return sum;\n    }\n\n    for (let offset = 0; offset < size; offset++) {\n      const sum = sumAcrossChannels(offset);\n      const val = xValues[offset] * Math.pow(bias + alpha * sum, -beta);\n      result[offset] = val;\n    }\n\n    return tf.tensor4d(result, x.shape);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    assertNotComplex(dy, 'LRNGrad');\n    const channels = dy.shape[3];\n    const dyValues = this.readSync(dy.dataId) as TypedArray;\n    const inputImageValues = this.readSync(inputImage.dataId) as TypedArray;\n    const outputImageValues = this.readSync(outputImage.dataId) as TypedArray;\n    const result = new Float32Array(dy.size);\n    const size = dy.size;\n\n    for (let offset = 0; offset < size; offset++) {\n      const currentChannel = offset % channels;\n      const depthBegin =\n          (offset - currentChannel) + Math.max(0, currentChannel - depthRadius);\n      const depthEnd = (offset - currentChannel) +\n          Math.min(channels, currentChannel + depthRadius + 1);\n\n      let norm = 0;\n      for (let k = depthBegin; k < depthEnd; k++) {\n        norm += Math.pow(inputImageValues[k], 2);\n      }\n      norm = alpha * norm + bias;\n\n      for (let k = depthBegin; k < depthEnd; k++) {\n        let dyi = -2 * alpha * beta * inputImageValues[k] *\n            outputImageValues[offset] / norm;\n        if (offset === k) {\n          dyi += Math.pow(norm, -beta);\n        }\n        dyi *= dyValues[offset];\n        result[k] += dyi;\n      }\n    }\n    return tf.tensor4d(result, dy.shape);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    assertNotComplex(logits, 'multinomial');\n\n    const probabilities = normalized ? logits : tf.softmax(logits);\n    const batchSize = probabilities.shape[0];\n    const numEvents = probabilities.shape[1];\n    const res = tf.zeros<Rank.R2>([batchSize, numSamples], 'int32');\n    const resVals = this.readSync(res.dataId) as TypedArray;\n    const probVals = this.readSync(probabilities.dataId) as TypedArray;\n\n    for (let b = 0; b < batchSize; ++b) {\n      const offset = b * numEvents;\n      // The cdf won't include the last event. It will be implicit if no other\n      // event happened.\n      const cdf = new Float32Array(numEvents - 1);\n      cdf[0] = probVals[offset];\n      for (let event = 1; event < cdf.length; ++event) {\n        cdf[event] = cdf[event - 1] + probVals[offset + event];\n      }\n\n      const random = seedrandom.alea(seed.toString());\n      const outOffset = b * numSamples;\n      for (let sampleId = 0; sampleId < numSamples; ++sampleId) {\n        const r = random();\n\n        // Assume last event happened by default.\n        resVals[outOffset + sampleId] = cdf.length;\n\n        for (let event = 0; event < cdf.length; event++) {\n          if (r < cdf[event]) {\n            resVals[outOffset + sampleId] = event;\n            break;\n          }\n        }\n      }\n    }\n    return res;\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    assertNotComplex(indices, 'oneHot');\n\n    const res = new Float32Array(indices.size * depth);\n    res.fill(offValue);\n    const indicesVal = this.readSync(indices.dataId) as TypedArray;\n\n    for (let event = 0; event < indices.size; ++event) {\n      if (indicesVal[event] >= 0 && indicesVal[event] < depth) {\n        res[event * depth + indicesVal[event]] = onValue;\n      }\n    }\n    return tf.tensor2d(res, [indices.size, depth], 'int32');\n  }\n\n  nonMaxSuppression(\n      boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n      iouThreshold: number, scoreThreshold: number): Tensor1D {\n    assertNotComplex(boxes, 'nonMaxSuppression');\n\n    const boxesVals = this.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = this.readSync(scores.dataId) as TypedArray;\n    return nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold);\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        dataFormat === 'NHWC',\n        () => `Only NHWC dataFormat supported on CPU for depthToSpace. Got ${\n            dataFormat}`);\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = x.shape[1];\n    const inputWidth = x.shape[2];\n    const inputDepth = x.shape[3];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const xValues = this.readSync(x.dataId) as TypedArray;\n    const result =\n        new Float32Array(batchSize * outputHeight * outputWidth * outputDepth);\n\n    let outputIdx = 0;\n    for (let b = 0; b < batchSize; ++b) {\n      for (let h = 0; h < outputHeight; ++h) {\n        const inH = Math.floor(h / blockSize);\n        const offsetH = (h % blockSize);\n        for (let w = 0; w < outputWidth; ++w) {\n          const inW = Math.floor(w / blockSize);\n          const offsetW = (w % blockSize);\n          const offsetD = (offsetH * blockSize + offsetW) * outputDepth;\n          for (let d = 0; d < outputDepth; ++d) {\n            const inD = d + offsetD;\n            const inputIdx =\n                inD + inputDepth * (inW + inputWidth * (inH + inputHeight * b));\n            result[outputIdx++] = xValues[inputIdx];\n          }\n        }\n      }\n    }\n    return tf.tensor4d(\n        result, [batchSize, outputHeight, outputWidth, outputDepth]);\n  }\n\n  private broadcastedBinaryOp(\n      a: Tensor, b: Tensor, dtype: DataType,\n      op: (a: number, b: number) => number): Tensor {\n    const newShape = backend_util.assertAndGetBroadcastShape(a.shape, b.shape);\n    const result = tf.buffer(newShape, dtype);\n    const aVals = this.readSync(a.dataId) as TypedArray;\n    const bVals = this.readSync(b.dataId) as TypedArray;\n    const aBroadcastDims = backend_util.getBroadcastDims(a.shape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(b.shape, newShape);\n\n    const resVals = result.values;\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resVals.length; ++i) {\n        resVals[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      const aBuf = this.bufferSync(a);\n      const bBuf = this.bufferSync(b);\n      for (let i = 0; i < resVals.length; ++i) {\n        const loc = result.indexToLoc(i);\n\n        const aLoc = loc.slice(-a.rank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = aBuf.locToIndex(aLoc);\n\n        const bLoc = loc.slice(-b.rank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = bBuf.locToIndex(bLoc);\n\n        resVals[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n    return result.toTensor();\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  dispose() {}\n\n  floatPrecision(): 16|32 {\n    return 32;\n  }\n\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return super.epsilon();\n  }\n\n  cropAndResize(\n      images: Tensor4D,\n      boxes: Tensor2D,\n      boxIndex: Tensor1D,\n      cropSize: [number, number],\n      method: string,\n      extrapolationValue: number,\n  ) {\n    const [batch, imageHeight, imageWidth, numChannels] = images.shape;\n    const numBoxes = boxes.shape[0];\n\n    const [cropHeight, cropWidth] = cropSize;\n    const output =\n        tf.buffer([numBoxes, cropHeight, cropWidth, numChannels], 'float32');\n\n    const boxVals = this.readSync(boxes.dataId) as TypedArray;\n    const boxIndVals = this.readSync(boxIndex.dataId) as TypedArray;\n    const imageVals = this.readSync(images.dataId) as TypedArray;\n\n    const inStride = images.strides;   // to calculate flat indexes into image\n    const outStride = output.strides;  // to calculate flat indexes into output\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op.cc\n    for (let b = 0; b < numBoxes; b++) {\n      const startInd = b * 4;\n      const y1 = boxVals[startInd];\n      const x1 = boxVals[startInd + 1];\n      const y2 = boxVals[startInd + 2];\n      const x2 = boxVals[startInd + 3];\n\n      const bInd: number = boxIndVals[b];\n      if (bInd >= batch) {\n        continue;\n      }\n\n      const heightScale = (cropHeight > 1) ?\n          (y2 - y1) * (imageHeight - 1) / (cropHeight - 1) :\n          0;\n      const widthScale =\n          (cropWidth > 1) ? (x2 - x1) * (imageWidth - 1) / (cropWidth - 1) : 0;\n\n      for (let y = 0; y < cropHeight; y++) {\n        const yInd: number = (cropHeight > 1) ?\n            y1 * (imageHeight - 1) + y * (heightScale) :\n            0.5 * (y1 + y2) * (imageHeight - 1);\n\n        if (yInd < 0 || yInd > imageHeight - 1) {\n          for (let x = 0; x < cropWidth; x++) {\n            for (let c = 0; c < numChannels; c++) {\n              const ind =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = extrapolationValue;\n            }\n          }\n          continue;\n        }\n\n        if (method === 'bilinear') {\n          const topInd = Math.floor(yInd);\n          const bottomInd = Math.ceil(yInd);\n          const yLerp = yInd - topInd;\n\n          for (let x = 0; x < cropWidth; x++) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const leftInd = Math.floor(xInd);\n            const rightInd = Math.ceil(xInd);\n            const xLerp = xInd - leftInd;\n\n            for (let c = 0; c < numChannels; c++) {\n              let ind = c + leftInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + topInd * inStride[1] +\n                  bInd * inStride[0];\n              const topRight = imageVals[ind];\n\n              ind = c + leftInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomLeft = imageVals[ind];\n\n              ind = c + rightInd * inStride[2] + bottomInd * inStride[1] +\n                  bInd * inStride[0];\n              const bottomRight = imageVals[ind];\n\n              const top = topLeft + (topRight - topLeft) * xLerp;\n              const bottom = bottomLeft + (bottomRight - bottomLeft) * xLerp;\n\n              ind = c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[ind] = top + ((bottom - top) * yLerp);\n            }\n          }\n        } else {  // method == \"nearest\"\n          for (let x = 0; x < cropWidth; ++x) {\n            const xInd = (cropWidth > 1) ?\n                x1 * (imageWidth - 1) + x * widthScale :\n                0.5 * (x1 + x2) * (imageWidth - 1);\n\n            if (xInd < 0 || xInd > imageWidth - 1) {\n              for (let c = 0; c < numChannels; c++) {\n                const ind =\n                    c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n                output.values[ind] = extrapolationValue;\n              }\n              continue;\n            }\n\n            const closestX = Math.round(xInd);\n            const closestY = Math.round(yInd);\n            for (let c = 0; c < numChannels; c++) {\n              const inInd = c + closestX * inStride[2] +\n                  closestY * inStride[1] + bInd * inStride[0];\n              const outInd =\n                  c + x * outStride[2] + y * outStride[1] + b * outStride[0];\n              output.values[outInd] = imageVals[inInd];\n            }\n          }\n        }\n      }\n    }\n    return output.toTensor() as Tensor4D;\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n    const sumDupeIndices = false;\n    return this.scatter(\n        sparseIndices, sparseValues, outputShape, outputSize, sliceSize,\n        numUpdates, sliceRank, strides, defaultValue, sumDupeIndices);\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n    if (numSlices === 0) {\n      return tf.tensor([], resultShape, x.dtype);\n    }\n\n    const buffer = new TensorBuffer([numSlices, sliceSize], x.dtype);\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const xData = this.readSync(x.dataId) as TypedArray;\n\n    for (let i = 0; i < numSlices; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        flattenIndex += dim * strides[j];\n        index.push(dim);\n      }\n      if (flattenIndex < 0 || flattenIndex >= x.size / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${x.shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        buffer.values[i * sliceSize + k] = xData[flattenIndex * sliceSize + k];\n      }\n    }\n    return buffer.toTensor().reshape(resultShape);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n    const defaultValue = tf.scalar(0);\n    const sumDupeIndices = true;\n    return this.scatter(\n        indices, updates, shape, outputSize, sliceSize, numUpdates, sliceRank,\n        strides, defaultValue, sumDupeIndices);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n    const values =\n        util.getArrayFromDType(dtype, util.sizeFromShape(shape)) as TypedArray;\n    values.fill(value as number);\n    return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported for string tensors');\n    } else {\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    const values = util.getArrayFromDType(\n                       x.dtype, util.sizeFromShape(x.shape)) as TypedArray;\n    return this.makeOutput(values, x.shape, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  private scatter<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R], outputSize: number,\n      sliceSize: number, numUpdates: number, sliceRank: number,\n      strides: number[], defaultValue: Scalar,\n      sumDupeIndices: boolean): Tensor<R> {\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n\n    const indicesData = this.readSync(indices.dataId) as TypedArray;\n    const updatesData = this.readSync(updates.dataId) as TypedArray;\n\n    if (outputSize === 0) {\n      return tf.tensor([], shape, updates.dtype);\n    }\n\n    const buffer = new TensorBuffer(flattenShape, updates.dtype as 'float32');\n    buffer.values.fill((this.readSync(defaultValue.dataId) as TypedArray)[0]);\n\n    for (let i = 0; i < numUpdates; i++) {\n      const index = [];\n      let flattenIndex = 0;\n      for (let j = 0; j < sliceRank; j++) {\n        const dim = indicesData[i * sliceRank + j];\n        index.push(dim);\n        flattenIndex += dim * strides[j];\n      }\n\n      if (flattenIndex < 0 || flattenIndex >= outputSize / sliceSize) {\n        throw new Error(\n            `Invalid indices: ${index} does not index into ${shape}`);\n      }\n\n      for (let k = 0; k < sliceSize; k++) {\n        if (sumDupeIndices) {\n          buffer.values[flattenIndex * sliceSize + k] +=\n              updatesData[i * sliceSize + k];\n        } else {\n          buffer.values[flattenIndex * sliceSize + k] = updates.rank === 0 ?\n              updatesData[0] :\n              updatesData[i * sliceSize + k];\n        }\n      }\n    }\n    return buffer.toTensor().reshape(shape);\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Abs, AbsInputs, KernelConfig, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function simpleAbsImpl(vals: TypedArray): Float32Array {\n  const resultValues = new Float32Array(vals.length);\n  for (let i = 0; i < vals.length; ++i) {\n    resultValues[i] = Math.abs(vals[i]);\n  }\n  return resultValues;\n}\n\nexport const absKernelFunc =\n    (args: {inputs: AbsInputs, backend: MathBackendCPU}) => {\n      const {x} = args.inputs;\n      const cpuBackend = args.backend;\n      let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n      if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n        resultValues = simpleAbsImpl(values);\n      } else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals =\n            cpuBackend.data.get(real.dataId).values as Float32Array;\n        const imagVals =\n            cpuBackend.data.get(imag.dataId).values as Float32Array;\n        for (let i = 0; i < realVals.length; i++) {\n          const real = realVals[i];\n          const imag = imagVals[i];\n          resultValues[i] = Math.hypot(real, imag);\n        }\n      }\n      return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n    };\n\nexport const absConfig: KernelConfig = {\n  kernelName: Abs,\n  backendName: 'cpu',\n  kernelFunc: absKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleBinaryKernelImpl, SimpleBinaryOperation} from './binary_types';\n\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op: SimpleBinaryOperation):\n    SimpleBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aVals: TypedArray,\n          bVals: TypedArray, dtype: DataType): [TypedArray, number[]] => {\n    const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    const resultRank = newShape.length;\n    const resultStrides = util.computeStrides(newShape);\n    const resultSize = util.sizeFromShape(newShape);\n\n    const result =\n        util.getTypedArrayFromDType(dtype as NumericDataType, resultSize);\n\n    const aRank = aShape.length;\n    const bRank = bShape.length;\n\n    const aStrides = util.computeStrides(aShape);\n    const bStrides = util.computeStrides(bShape);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < result.length; ++i) {\n        result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n      }\n    } else {\n      for (let i = 0; i < result.length; ++i) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        result[i] = op(aVals[aIndex], bVals[bIndex]);\n      }\n    }\n\n    return [result, newShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Complex, ComplexInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function complex(args: {inputs: ComplexInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {real, imag} = inputs;\n\n  const realVals = backend.data.get(real.dataId).values as TypedArray;\n  const imagVals = backend.data.get(imag.dataId).values as TypedArray;\n\n  const complexInfo = backend.makeTensorInfo(real.shape, 'complex64');\n\n  const complex = backend.data.get(complexInfo.dataId);\n\n  // The complex tensor owns the underlying real and imag tensorInfos, only the\n  // complex tensor tracks refCount, when complexData is disposed the\n  // underlying tensorData will be disposed.\n  complex.complexTensorInfos = {\n    real: backend.makeTensorInfo(real.shape, 'float32', realVals),\n    imag: backend.makeTensorInfo(imag.shape, 'float32', imagVals)\n  };\n\n  return complexInfo;\n}\n\nexport const complexConfig: KernelConfig = {\n  kernelName: Complex,\n  backendName: 'cpu',\n  kernelFunc: complex as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendCPU}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'cpu',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Real, RealInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function real(args: {inputs: RealInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const real = backend.data.get(input.dataId).complexTensorInfos.real;\n  const realVal = backend.data.get(real.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the real value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(real.shape, real.dtype, realVal);\n}\n\nexport const realConfig: KernelConfig = {\n  kernelName: Real,\n  backendName: 'cpu',\n  kernelFunc: real as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport * as tf from '@tensorflow/tfjs-core';\nimport {Cast, CastAttrs, CastInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\n\nimport {complex} from './Complex';\nimport {identity} from './Identity';\nimport {real} from './Real';\n\nexport function cast(\n    args: {inputs: CastInputs, backend: MathBackendCPU, attrs: CastAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {dtype} = attrs;\n\n  // Casting to complex64.\n  if (dtype === 'complex64') {\n    if (x.dtype === 'complex64') {\n      return identity({inputs: {x}, backend});\n    }\n\n    // TODO(lina128): Import kernel function once zeros is modularized.\n    const zerosTensor = tf.zeros(x.shape);\n    const floatX = cast({inputs: {x}, backend, attrs: {dtype: 'float32'}});\n\n    const result =\n        complex({inputs: {real: floatX, imag: zerosTensor}, backend});\n\n    zerosTensor.dispose();\n    backend.disposeIntermediateTensorInfo(floatX);\n\n    return result;\n  }\n\n  // Casting from complex64\n  if (x.dtype === 'complex64') {\n    const realPart = real({inputs: {input: x}, backend});\n    const result = cast({inputs: {x: realPart}, backend, attrs: {dtype}});\n\n    backend.disposeIntermediateTensorInfo(realPart);\n\n    return result;\n  }\n\n  if (!util.hasEncodingLoss(x.dtype, dtype)) {\n    // We don't change the underlying data, since we cast to higher\n    // precision.\n    const result = identity({inputs: {x}, backend});\n    return {dataId: result.dataId, shape: result.shape, dtype};\n  }\n\n  if (dtype === 'int32') {\n    const values = backend.data.get(x.dataId).values as TypedArray;\n    const resultValues = Int32Array.from(values);\n    return backend.makeTensorInfo(x.shape, 'int32', resultValues);\n  }\n\n  if (dtype === 'bool') {\n    // This is essentially the result of notEqual(x, 0). We avoid using\n    // kernel notEqual to avoid circular dependency, i.e. binary_utils ->\n    // cast -> notEqual -> binary_utils.\n    const xVals = backend.data.get(x.dataId).values as TypedArray;\n    const zero = util.toTypedArray([0], x.dtype);\n\n    const [resultData, resultShape] = createSimpleBinaryKernelImpl(\n        (a, b) => (a !== b) ? 1 : 0)(x.shape, [], xVals, zero, 'bool');\n\n    return backend.makeTensorInfo(resultShape, 'bool', resultData);\n  }\n\n  throw new Error(`Error in Cast: failed to cast ${x.dtype} to ${dtype}`);\n}\n\nexport const castConfig: KernelConfig = {\n  kernelName: Cast,\n  backendName: 'cpu',\n  kernelFunc: cast as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, BinaryInputs, DataType, KernelFunc, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {cast} from '../kernels/Cast';\nimport {complex} from '../kernels/Complex';\n\nimport {ComplexBinaryKernelImpl, ComplexBinaryOperation, SimpleBinaryKernelImpl} from './binary_types';\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param name Kernel name.\n * @param binaryKernelImpl A `SimpleBinaryKernelImpl` for the kernel.\n * @param binaryKernelComplexImpl Optional. If exists, represents a\n *     `ComplexBinaryKernelImpl` for the kernel, will be used when input dtype\n *     is `complex64`.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    name: string, simpleImpl: SimpleBinaryKernelImpl,\n    complexImpl?: ComplexBinaryKernelImpl, dtype?: DataType): KernelFunc {\n  if (complexImpl == null) {\n    return ({inputs, backend}) => {\n      const {a, b} = inputs as BinaryInputs;\n      const cpuBackend = backend as MathBackendCPU;\n\n      assertNotComplex([a, b], name);\n\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    };\n  }\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    if (a.dtype === 'complex64' || b.dtype === 'complex64') {\n      const $aComplex = cast(\n          {inputs: {x: a}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $aComplexVals = cpuBackend.data.get($aComplex.dataId);\n\n      const aReal = $aComplexVals.complexTensorInfos.real;\n      const aImag = $aComplexVals.complexTensorInfos.imag;\n\n      const aRealVals =\n          cpuBackend.data.get(aReal.dataId).values as Float32Array;\n      const aImagVals =\n          cpuBackend.data.get(aImag.dataId).values as Float32Array;\n\n      const $bComplex = cast(\n          {inputs: {x: b}, backend: cpuBackend, attrs: {dtype: 'complex64'}});\n\n      const $bComplexVals = cpuBackend.data.get($bComplex.dataId);\n\n      const bReal = $bComplexVals.complexTensorInfos.real;\n      const bImag = $bComplexVals.complexTensorInfos.imag;\n\n      const bRealVals =\n          cpuBackend.data.get(bReal.dataId).values as Float32Array;\n      const bImagVals =\n          cpuBackend.data.get(bImag.dataId).values as Float32Array;\n\n      const [resultRealData, resultImagData, resultShape] = complexImpl(\n          a.shape, b.shape, aRealVals, aImagVals, bRealVals, bImagVals);\n\n      const resultReal =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultRealData);\n\n      const resultImag =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', resultImagData);\n\n      const result = complex(\n          {inputs: {real: resultReal, imag: resultImag}, backend: cpuBackend});\n\n      cpuBackend.disposeIntermediateTensorInfo($aComplex);\n      cpuBackend.disposeIntermediateTensorInfo($bComplex);\n      cpuBackend.disposeIntermediateTensorInfo(resultReal);\n      cpuBackend.disposeIntermediateTensorInfo(resultImag);\n\n      return result;\n    } else {\n      const aVals = cpuBackend.data.get(a.dataId).values as TypedArray;\n      const bVals = cpuBackend.data.get(b.dataId).values as TypedArray;\n\n      const $dtype = dtype || a.dtype;\n\n      const [resultData, resultShape] =\n          simpleImpl(a.shape, b.shape, aVals, bVals, $dtype);\n\n      return cpuBackend.makeTensorInfo(resultShape, $dtype, resultData);\n    }\n  };\n}\n\n/**\n * Template that creates the complex type implementation for binary ops.\n * Supports broadcast.\n */\nexport function createComplexBinaryKernelImpl(op: ComplexBinaryOperation):\n    ComplexBinaryKernelImpl {\n  return (aShape: number[], bShape: number[], aRealVals: Float32Array,\n          aImagVals: Float32Array, bRealVals: Float32Array,\n          bImagVals: Float32Array): [TypedArray, TypedArray, number[]] => {\n    const resultShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const resultSize = util.sizeFromShape(resultShape);\n    const resultRank = resultShape.length;\n    const resultStrides = util.computeStrides(resultShape);\n\n    const resultRealVals = util.getTypedArrayFromDType('float32', resultSize);\n    const resultImagVals = util.getTypedArrayFromDType('float32', resultSize);\n\n    const aBroadcastDims = backend_util.getBroadcastDims(aShape, resultShape);\n    const bBroadcastDims = backend_util.getBroadcastDims(bShape, resultShape);\n\n    const aVals = backend_util.mergeRealAndImagArrays(aRealVals, aImagVals);\n    const bVals = backend_util.mergeRealAndImagArrays(bRealVals, bImagVals);\n\n    const aRank = aShape.length;\n    const aStrides = util.computeStrides(aShape);\n\n    const bRank = bShape.length;\n    const bStrides = util.computeStrides(bShape);\n\n    if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const aIdx = i % aVals.length;\n        const bIdx = i % bVals.length;\n\n        const result =\n            op(aVals[aIdx * 2], aVals[aIdx * 2 + 1], bVals[bIdx * 2],\n               bVals[bIdx * 2 + 1]);\n\n        resultRealVals[i] = result.real;\n        resultImagVals[i] = result.imag;\n      }\n    } else {\n      for (let i = 0; i < resultRealVals.length; i++) {\n        const loc = util.indexToLoc(i, resultRank, resultStrides);\n\n        const aLoc = loc.slice(-aRank);\n        aBroadcastDims.forEach(d => aLoc[d] = 0);\n        const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n\n        const bLoc = loc.slice(-bRank);\n        bBroadcastDims.forEach(d => bLoc[d] = 0);\n        const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n\n        const opResult =\n            op(aVals[aIndex * 2], aVals[aIndex * 2 + 1], bVals[bIndex * 2],\n               bVals[bIndex * 2 + 1]);\n\n        resultRealVals[i] = opResult.real;\n        resultImagVals[i] = opResult.imag;\n      }\n    }\n    return [resultRealVals, resultImagVals, resultShape];\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Add, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal + bReal, imag: aImag + bImag};\n    }));\n\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\n\nexport const addConfig: KernelConfig = {\n  kernelName: Add,\n  backendName: 'cpu',\n  kernelFunc: add\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NumericDataType, util} from '@tensorflow/tfjs-core';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op: SimpleUnaryOperation):\n    SimpleUnaryImpl {\n  return (values, dtype, attrs) => {\n    const newValues =\n        util.getTypedArrayFromDType(dtype as NumericDataType, values.length);\n    for (let i = 0; i < values.length; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return newValues;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelFunc, TypedArray, UnaryInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {SimpleUnaryImpl, SimpleUnaryOperation} from './unary_types';\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param name Kernel name.\n * @param op A `SimpleUnaryOperation` for the kernel.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFunc(\n    name: string, op: SimpleUnaryOperation, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xSize = util.sizeFromShape(x.shape);\n    const $dtype = dtype || x.dtype;\n    const newValues = util.getArrayFromDType($dtype, xSize);\n    for (let i = 0; i < xSize; ++i) {\n      newValues[i] = op(values[i], attrs);\n    }\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for unary ops from the given\n * `SimpleUnaryImpl`..\n * @param name Kernel name.\n * @param unaryImpl A `SimpleUnaryImpl` that implements the op.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the input. This is mainly used in certain\n *     kernels that return bool type, such as isFinite, isInf, etc.\n */\nexport function unaryKernelFuncFromImpl(\n    name: string, unaryImpl: SimpleUnaryImpl, dtype?: DataType): KernelFunc {\n  return ({inputs, attrs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    assertNotComplex(x, name);\n    if (x.dtype === 'string' || dtype === 'string') {\n      throw new Error('unaryKernelFunc does not support string input/output');\n    }\n\n    const cpuBackend = backend as MathBackendCPU;\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const $dtype = dtype || x.dtype;\n    const newValues = unaryImpl(values, $dtype, attrs);\n    return cpuBackend.makeTensorInfo(x.shape, $dtype, newValues);\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Ceil, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\n\nexport const ceilConfig: KernelConfig = {\n  kernelName: Ceil,\n  backendName: 'cpu',\n  kernelFunc: ceilKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Exp, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\n\nexport const expConfig: KernelConfig = {\n  kernelName: Exp,\n  backendName: 'cpu',\n  kernelFunc: expKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Expm1, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\n\nexport const expm1Config: KernelConfig = {\n  kernelName: Expm1,\n  backendName: 'cpu',\n  kernelFunc: expm1KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Floor, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\n\nexport const floorConfig: KernelConfig = {\n  kernelName: Floor,\n  backendName: 'cpu',\n  kernelFunc: floorKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\n\nexport const logConfig: KernelConfig = {\n  kernelName: Log,\n  backendName: 'cpu',\n  kernelFunc: logKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function maxImpl(\n    aVals: TypedArray, reduceSize: number, outShape: number[],\n    dtype: DataType): TypedArray {\n  const vals = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(outShape));\n\n  for (let i = 0; i < vals.length; ++i) {\n    const offset = i * reduceSize;\n    let max = aVals[offset];\n    for (let j = 0; j < reduceSize; ++j) {\n      const value = aVals[offset + j];\n      if (value > max) {\n        max = value;\n      }\n    }\n    vals[i] = max;\n  }\n  return vals;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Multiply} from '@tensorflow/tfjs-core';\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const multiplyImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n      };\n    }));\n\nexport const multiply =\n    binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\n\nexport const multiplyConfig: KernelConfig = {\n  kernelName: Multiply,\n  backendName: 'cpu',\n  kernelFunc: multiply\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Rsqrt} from '@tensorflow/tfjs-core';\n\nimport {createSimpleUnaryImpl} from '../utils/unary_impl';\nimport {unaryKernelFuncFromImpl} from '../utils/unary_utils';\n\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\n\nexport const rsqrtConfig: KernelConfig = {\n  kernelName: Rsqrt,\n  backendName: 'cpu',\n  kernelFunc: rsqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, KernelConfig, KernelFunc, NumericDataType, Slice, slice_util, SliceAttrs, SliceInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function sliceImpl(\n    vals: TypedArray, begin: number[], size: number[], shape: number[],\n    dtype: DataType): TypedArray {\n  const isContinous = slice_util.isSliceContinous(shape, begin, size);\n  const length = util.sizeFromShape(size);\n  const xStrides = util.computeStrides(shape);\n\n  if (isContinous) {\n    const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n    return vals.subarray(flatOffset, flatOffset + length);\n  }\n\n  const outVals = util.getTypedArrayFromDType(dtype as NumericDataType, length);\n  for (let i = 0; i < length; ++i) {\n    const rank = size.length;\n    const strides = util.computeStrides(size);\n    const loc = util.indexToLoc(i, rank, strides);\n    const xLoc = loc.map((idx: number, j) => idx + begin[j]);\n    const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n    outVals[i] = vals[xIndex];\n  }\n  return outVals;\n}\n\nexport function slice(\n    args: {inputs: SliceInputs, backend: MathBackendCPU, attrs: SliceAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {begin, size} = attrs;\n\n  assertNotComplex(x, 'slice');\n\n  const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n  slice_util.assertParamsValid(x, $begin, $size);\n\n  const vals = backend.data.get(x.dataId).values as TypedArray;\n  const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n  return backend.makeTensorInfo($size, x.dtype, outVals);\n}\n\nexport const sliceConfig: KernelConfig = {\n  kernelName: Slice,\n  backendName: 'cpu',\n  kernelFunc: slice as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sub} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc, createComplexBinaryKernelImpl} from '../utils/kernel_utils';\n\nexport const subImpl =\n    createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl =\n    createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n      return {real: aReal - bReal, imag: aImag - bImag};\n    }));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\n\nexport const subConfig: KernelConfig = {\n  kernelName: Sub,\n  backendName: 'cpu',\n  kernelFunc: sub\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {util} from '@tensorflow/tfjs-core';\n\nexport function transposeImpl(\n    xVals: TypedArray, xShape: number[], dtype: DataType, perm: number[],\n    newShape: number[]): TypedArray {\n  const xRank = xShape.length;\n  const xSize = util.sizeFromShape(xShape);\n  const xStrides = util.computeStrides(xShape);\n  const newStrides = util.computeStrides(newShape);\n\n  const result = util.getTypedArrayFromDType(\n      dtype as NumericDataType, util.sizeFromShape(newShape));\n\n  for (let i = 0; i < xSize; ++i) {\n    const loc = util.indexToLoc(i, xRank, xStrides);\n\n    // Permute location.\n    const newLoc: number[] = new Array(loc.length);\n    for (let i = 0; i < newLoc.length; i++) {\n      newLoc[i] = loc[perm[i]];\n    }\n\n    const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n    result[newIndex] = xVals[i];\n  }\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendValues, DataType, TensorBuffer, TypedArray, util} from '@tensorflow/tfjs-core';\n\nexport function uniqueImpl(\n    values: BackendValues, axis: number, shape: number[], dtype: DataType): {\n  outputValues: BackendValues,\n  outputShape: number[],\n  indices: BackendValues\n} {\n  // Normalize and validate axis.\n  const $axis = util.parseAxisParam(axis, shape)[0];\n\n  // Calculate the new shape that is suitable for extracting data along the\n  // given axis.\n  //\n  // The rank is 3.\n  // The size of the 1st dimension is the size of all the axes < the given axis.\n  // The size of the 2nd dimension is the same as the size of the given axis.\n  // The size of the 3rd dimension is the size of all the axes > the given axis.\n  //\n  // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n  // newShape would be: [2*3, 5, 4].\n  //\n  // Note that this is not the final output shape. This will be the shape for an\n  // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n  // values along the given axis. To demonstrate how it works, consider the\n  // following example:\n  //\n  // Input: a 3D tensor, with shape [1, 2, 3]\n  // [\n  //   [\n  //      [1,2,3],\n  //      [4,5,6]\n  //   ]\n  // ]\n  // Axis: 2 (the last axis).\n  // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n  //\n  // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n  // 1*2. The re-shaped data would look like:\n  //\n  // [\n  //   [\n  //     [1], [2], [3]\n  //   ],\n  //   [\n  //     [4], [5], [6]\n  //   ]\n  // ]\n  //\n  // Then, we can construct a 3-level nested loop by the following dimension\n  // order to extract the values along the axis (dimension1):\n  // i: dimension1       // 0,1,2 (newShape[1])\n  //   m: dimension0     // 0,1   (newShape[0])\n  //     n: dimension2   // 0     (newShape[2])\n  //\n  //                       m, i, n\n  //                      ---------\n  // Iteration 0: data at [0, 0, 0] => \"1\"\n  // Iteration 1: data at [1, 0, 0] => \"4\"\n  // We got [1,4].\n  // Iteration 2: data at [0, 1, 0] => \"2\"\n  // Iteration 3: data at [1, 1, 0] => \"5\"\n  // We got [2,5].\n  // Iteration 4: data at [0, 2, 0] => \"3\"\n  // Iteration 5: data at [1, 2, 0] => \"6\"\n  // We got [3,6].\n  const newShape = [1, shape[0], 1];\n  for (let i = 0; i < $axis; i++) {\n    newShape[0] *= shape[i];\n  }\n  newShape[1] = shape[$axis];\n  for (let i = $axis + 1; i < shape.length; i++) {\n    newShape[2] *= shape[i];\n  }\n\n  // A map from unique elements (their string representations) to their values\n  // in \"indices\" (below).\n  const uniqueElements: {[key: string]: number} = {};\n  // The indices of each unique element in the original tensor along the given\n  // axis. It is 1D and has the same size as the given axis.\n  const indices = new Int32Array(shape[$axis]);\n  // Create a buffer so we can easily extract value at a given location.\n  const inputBuffer = new TensorBuffer(newShape, dtype, values as TypedArray);\n  // The indices along the given axis that have unique elements. This is a\n  // de-duped version of \"indices\" above.\n  const uniqueIndices: number[] = [];\n  const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n  for (let i = 0; i < shape[$axis]; i++) {\n    // Extract values along the axis.\n    let element: string;\n    if (is1DTensor) {\n      // Fast path for 1D tensor input.\n      element = values[i].toString();\n    } else {\n      const axisValues = [];\n      for (let m = 0; m < newShape[0]; m++) {\n        for (let n = 0; n < newShape[2]; n++) {\n          axisValues.push(inputBuffer.get(m, i, n));\n        }\n      }\n      element = axisValues.join(',');\n    }\n\n    // Dedup and update various indices.\n    if (uniqueElements[element] !== undefined) {\n      indices[i] = uniqueElements[element];\n    } else {\n      const uniqueIndex = Object.keys(uniqueElements).length;\n      uniqueElements[element] = uniqueIndex;\n      indices[i] = uniqueIndex;\n      uniqueIndices.push(i);\n    }\n  }\n\n  // Now we know where each of the unique elements are located along the axis\n  // (uniqueIndices). Extract them from input buffer and store them in the\n  // output buffer.\n  const outputTmpShape = newShape.slice();\n  outputTmpShape[1] = Object.keys(uniqueElements).length;\n  const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n  uniqueIndices.forEach((uniqueElementIndex, i) => {\n    for (let m = 0; m < newShape[0]; m++) {\n      for (let n = 0; n < newShape[2]; n++) {\n        outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n      }\n    }\n  });\n\n  // The output shape can be calculated from the input shape with the size of\n  // the given axis replaced by the number of unique elements along that axis.\n  const outputShape = shape.slice();\n  outputShape[$axis] = outputTmpShape[1];\n\n  return {\n    outputValues: outputBuffer.values as BackendValues,\n    outputShape,\n    indices,\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Shared kernel impls for use in other backends.\nexport {simpleAbsImpl} from './kernels/Abs';\nexport {addImpl} from './kernels/Add';\nexport {ceilImpl} from './kernels/Ceil';\nexport {expImpl} from './kernels/Exp';\nexport {expm1Impl} from './kernels/Expm1';\nexport {floorImpl} from './kernels/Floor';\nexport {logImpl} from './kernels/Log';\nexport {maxImpl} from './kernels/Max_impl';\nexport {multiplyImpl} from './kernels/Multiply';\nexport {rsqrtImpl} from './kernels/Rsqrt';\nexport {sliceImpl} from './kernels/Slice';\nexport {subImpl} from './kernels/Sub';\nexport {transposeImpl} from './kernels/Transpose_impl';\nexport {uniqueImpl} from './kernels/Unique_impl';\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/*\n * base.ts contains all the exports from tfjs-backend-cpu\n * without auto-kernel registration\n */\nimport {registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from './backend_cpu';\nimport * as shared from './shared';\n\nexport {MathBackendCPU} from './backend_cpu';\nexport {version as version_cpu} from './version';\nexport {shared};\n\n// Side effects for default initialization of MathBackendCPU\nregisterBackend('cpu', () => new MathBackendCPU(), 1 /* priority */);\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acosKernelFunc = unaryKernelFunc(Acos, (xi) => Math.acos(xi));\n\nexport const acosConfig: KernelConfig = {\n  kernelName: Acos,\n  backendName: 'cpu',\n  kernelFunc: acosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Acosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const acoshKernelFunc = unaryKernelFunc(Acosh, (xi) => Math.acosh(xi));\n\nexport const acoshConfig: KernelConfig = {\n  kernelName: Acosh,\n  backendName: 'cpu',\n  kernelFunc: acoshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asin, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinKernelFunc = unaryKernelFunc(Asin, (xi) => Math.asin(xi));\n\nexport const asinConfig: KernelConfig = {\n  kernelName: Asin,\n  backendName: 'cpu',\n  kernelFunc: asinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Asinh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const asinhKernelFunc = unaryKernelFunc(Asinh, (xi) => Math.asinh(xi));\n\nexport const asinhConfig: KernelConfig = {\n  kernelName: Asinh,\n  backendName: 'cpu',\n  kernelFunc: asinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanKernelFunc = unaryKernelFunc(Atan, (xi) => Math.atan(xi));\n\nexport const atanConfig: KernelConfig = {\n  kernelName: Atan,\n  backendName: 'cpu',\n  kernelFunc: atanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atanh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const atanhKernelFunc = unaryKernelFunc(Atanh, (xi) => Math.atanh(xi));\n\nexport const atanhConfig: KernelConfig = {\n  kernelName: Atanh,\n  backendName: 'cpu',\n  kernelFunc: atanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, buffer, DataType, Rank, TensorBuffer, TypedArray} from '@tensorflow/tfjs-core';\n\nexport function pool(\n    xValues: TypedArray, xShape: number[], dtype: DataType, strides: number[],\n    convInfo: backend_util.Conv2DInfo,\n    poolType: 'max'|'avg'): TensorBuffer<Rank, DataType> {\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const initialValue =\n      (poolType === 'max' ? Number.NEGATIVE_INFINITY :\n                            Number.POSITIVE_INFINITY);\n\n  const output = buffer(convInfo.outShape, dtype);\n  const outputVals = output.values;\n\n  const outputBatchStrides =\n      convInfo.outShape[1] * convInfo.outShape[2] * convInfo.outShape[3];\n  const outputRowStrides = convInfo.outShape[2] * convInfo.outShape[3];\n  const outputColStrides = convInfo.outShape[3];\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    const outputBatchOffset = b * outputBatchStrides;\n    const inputBatchOffset = b * strides[0];\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        const outputRowOffset = outputBatchOffset + yR * outputRowStrides;\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          const xCMin = Math.max(0, xCCorner);\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let minMaxValue = initialValue;\n          let avgValue = 0;\n          let count = 0;\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const xROffset = inputBatchOffset + xR * strides[1];\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const xCOffset = xROffset + xC * strides[2];\n              const pixel = xValues[xCOffset + d];\n              if ((poolType === 'max' && pixel > minMaxValue)) {\n                minMaxValue = pixel;\n              } else if (poolType === 'avg') {\n                avgValue += pixel;\n                count++;\n              }\n            }\n            if (isNaN(minMaxValue)) {\n              break;\n            }\n          }\n          const outputOffset = outputRowOffset + yC * outputColStrides + d;\n          outputVals[outputOffset] =\n              poolType === 'avg' ? avgValue / count : minMaxValue;\n        }\n      }\n    }\n  }\n  return output;\n}\n\nexport function maxPoolPositions(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    convInfo: backend_util.Conv2DInfo, flattenPositions = false,\n    includeBatchInIndex = false): TensorBuffer<Rank, 'int32'> {\n  const maxPositions = buffer(convInfo.outShape, 'int32');\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padTop = convInfo.padInfo.top;\n  const padLeft = convInfo.padInfo.left;\n\n  const xBuf = buffer(xShape, dtype, xValues);\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let yR = 0; yR < convInfo.outHeight; ++yR) {\n        const xRCorner = yR * strideHeight - padTop;\n        let xRMin = xRCorner;\n        while (xRMin < 0) {\n          xRMin += dilationHeight;\n        }\n        // const xRMin = Math.max(0, xRCorner);\n        const xRMax =\n            Math.min(convInfo.inHeight, effectiveFilterHeight + xRCorner);\n        for (let yC = 0; yC < convInfo.outWidth; ++yC) {\n          const xCCorner = yC * strideWidth - padLeft;\n          let xCMin = xCCorner;\n          while (xCMin < 0) {\n            xCMin += dilationWidth;\n          }\n          const xCMax =\n              Math.min(convInfo.inWidth, effectiveFilterWidth + xCCorner);\n          let maxValue = Number.NEGATIVE_INFINITY;\n          let maxPosition = -1;\n\n          for (let xR = xRMin; xR < xRMax; xR += dilationHeight) {\n            const wR = xR - xRCorner;\n            for (let xC = xCMin; xC < xCMax; xC += dilationWidth) {\n              const wC = xC - xCCorner;\n              const pixel = xBuf.get(b, xR, xC, d);\n              if (pixel > maxValue) {\n                maxValue = pixel as number;\n                if (flattenPositions) {\n                  maxPosition = includeBatchInIndex ?\n                      ((b * convInfo.inHeight + xR) * convInfo.inWidth + xC) *\n                              convInfo.inChannels +\n                          d :\n                      (xR * convInfo.inWidth + xC) * convInfo.inChannels + d;\n                } else {\n                  maxPosition = wR * effectiveFilterWidth + wC;\n                }\n              }\n            }\n          }\n          maxPositions.set(maxPosition, b, yR, yC, d);\n        }\n      }\n    }\n  }\n  return maxPositions;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function avgPool(\n    args:\n        {inputs: AvgPoolInputs, backend: MathBackendCPU, attrs: AvgPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'avg');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'cpu',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, buffer, KernelConfig, KernelFunc, Rank, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const filterHeight = convInfo.filterHeight;\n  const filterWidth = convInfo.filterWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel;\n            }\n          }\n          dx.set(dotProd * avgMultiplier, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function batchNormKernelFunc(args: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendCPU,\n  attrs: FusedBatchNormAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x, scale, offset, mean, variance} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  assertNotComplex([x, mean, variance, scale, offset], 'batchNorm');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const mVals = backend.data.get(mean.dataId).values as TypedArray;\n  const varVals = backend.data.get(variance.dataId).values as TypedArray;\n  const sVals = scale ? backend.data.get(scale.dataId).values as TypedArray :\n                        new Float32Array([1]);\n  const offVals = offset ?\n      backend.data.get(offset.dataId).values as TypedArray :\n      new Float32Array([0]);\n  const outVals = new Float32Array(xVals.length);\n\n  const offValsLength = offVals.length;\n  const sValsLength = sVals.length;\n  const varValsLength = varVals.length;\n  const mValsLength = mVals.length;\n\n  let offi = 0;\n  let mi = 0;\n  let si = 0;\n  let vi = 0;\n  for (let i = 0; i < xVals.length; ++i) {\n    outVals[i] = offVals[offi++] +\n        (xVals[i] - mVals[mi++]) * sVals[si++] /\n            Math.sqrt(varVals[vi++] + varianceEpsilon);\n    if (offi >= offValsLength) {\n      offi = 0;\n    }\n    if (mi >= mValsLength) {\n      mi = 0;\n    }\n    if (si >= sValsLength) {\n      si = 0;\n    }\n    if (vi >= varValsLength) {\n      vi = 0;\n    }\n  }\n  return backend.makeTensorInfo(x.shape, x.dtype, outVals);\n}\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'cpu',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ClipByValue, ClipByValueAttrs, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const clipKernelFunc = unaryKernelFunc(ClipByValue, (xi, attrs) => {\n  const clipAttrs = attrs as {} as ClipByValueAttrs;\n  if (xi > clipAttrs.clipValueMax) {\n    return clipAttrs.clipValueMax;\n  }\n  return xi < clipAttrs.clipValueMin ? clipAttrs.clipValueMin : xi;\n});\n\nexport const clipConfig: KernelConfig = {\n  kernelName: ClipByValue,\n  backendName: 'cpu',\n  kernelFunc: clipKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Imag, ImagInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function imag(args: {inputs: ImagInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const imag = backend.data.get(input.dataId).complexTensorInfos.imag;\n  const imagVal = backend.data.get(imag.dataId).values;\n\n  // When complex tensor is disposed, its underlying parts will be disposed too.\n  // Make new tensor out of the imag value of the complex. This makes sure the\n  // value is still accessible even if complex tensor is disposed.\n  return backend.makeTensorInfo(imag.shape, imag.dtype, imagVal);\n}\n\nexport const imagConfig: KernelConfig = {\n  kernelName: Imag,\n  backendName: 'cpu',\n  kernelFunc: imag as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport function reshape(\n    args:\n        {inputs: ReshapeInputs, backend: MathBackendCPU, attrs: ReshapeAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  backend.incRef(x.dataId);\n\n  const xData = backend.data.get(x.dataId);\n\n  if (xData.complexTensorInfos != null) {\n    const real = xData.complexTensorInfos.real;\n    const imag = xData.complexTensorInfos.imag;\n\n    real.shape = $shape;\n    imag.shape = $shape;\n  }\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'cpu',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Concat, ConcatAttrs, ConcatInputs, KernelConfig, KernelFunc, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nimport {complex} from './Complex';\nimport {imag} from './Imag';\nimport {real} from './Real';\nimport {reshape} from './Reshape';\n\nexport function concat(\n    args: {inputs: ConcatInputs, backend: MathBackendCPU, attrs: ConcatAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {axis} = attrs;\n\n  const $axis = util.parseAxisParam(axis, inputs[0].shape)[0];\n  let outShape = backend_util.computeOutShape(inputs.map(t => t.shape), $axis);\n\n  if (util.sizeFromShape(outShape) === 0) {\n    return backend.makeTensorInfo(outShape, inputs[0].dtype, []);\n  }\n\n  // Keep only non-empty tensors (ignore tensors with 0 in their shape).\n  const $inputs = inputs.filter(t => util.sizeFromShape(t.shape) > 0);\n  if ($inputs.length === 1) {\n    return $inputs[0];\n  }\n\n  const shapes = $inputs.map(t => t.shape);\n  backend_util.assertParamsConsistent(shapes, $axis);\n\n  if ($inputs[0].dtype === 'complex64') {\n    const reals = $inputs.map((t) => real({inputs: {input: t}, backend}));\n    const imags = $inputs.map((t) => imag({inputs: {input: t}, backend}));\n\n    const realConcated = concat({inputs: reals, backend, attrs: {axis}});\n    const imagConcated = concat({inputs: imags, backend, attrs: {axis}});\n\n    const result =\n        complex({inputs: {real: realConcated, imag: imagConcated}, backend});\n\n    reals.forEach(r => backend.disposeIntermediateTensorInfo(r));\n    imags.forEach(i => backend.disposeIntermediateTensorInfo(i));\n    backend.disposeIntermediateTensorInfo(realConcated);\n    backend.disposeIntermediateTensorInfo(imagConcated);\n\n    return result;\n  }\n\n  // Any concat of n-dimensional tensors across any axis can be reduced to\n  // a concatenation of two-dimensional tensors across the axis 1 by first\n  // partitioning the axes of the original tensors into those less than the\n  // axis to be concatenated and the rest. Then reshape the tensors\n  // into a two-dimensional tensor by collapsing these two sets of axes and\n  // concatenate the resulting matrices across the axis 1, finally reshaping\n  // the result to have the proper shape.\n  const inputs2D = $inputs.map(t => {\n    const innerSize = util.sizeFromShape(t.shape.slice($axis));\n    const shape = [-1, innerSize];\n    return reshape({inputs: {x: t}, backend, attrs: {shape}});\n  });\n\n  // Concats 2d tensors along axis=1.\n  outShape =\n      backend_util.computeOutShape(inputs2D.map(t => t.shape), 1 /* axis */);\n\n  const outVals = util.getTypedArrayFromDType(\n      $inputs[0].dtype as 'float32', util.sizeFromShape(outShape));\n\n  if (inputs2D[0].shape[0] === 1) {\n    // Use built-in TypedArray.set() method for speed.\n    let offset = 0;\n    inputs2D.forEach(t => {\n      const val = backend.data.get(t.dataId).values as TypedArray;\n      const size = util.sizeFromShape(t.shape);\n\n      outVals.set(val, offset);\n      offset += size;\n    });\n  } else {\n    let colOffset = 0;\n\n    inputs2D.forEach(t => {\n      const tVals = backend.data.get(t.dataId).values as TypedArray;\n\n      let tIdx = 0;\n\n      for (let row = 0; row < t.shape[0]; ++row) {\n        const resIdx = row * outShape[1] + colOffset;\n        for (let col = 0; col < t.shape[1]; ++col) {\n          outVals[resIdx + col] = tVals[tIdx++];\n        }\n      }\n\n      colOffset += t.shape[1];\n    });\n  }\n\n  const finalOutShape =\n      backend_util.computeOutShape($inputs.map(t => t.shape), $axis);\n\n  const outInfo =\n      backend.makeTensorInfo(finalOutShape, inputs[0].dtype, outVals);\n\n  inputs2D.forEach(t => backend.disposeIntermediateTensorInfo(t));\n\n  return outInfo;\n}\n\nexport const concatConfig: KernelConfig = {\n  kernelName: Concat,\n  backendName: 'cpu',\n  kernelFunc: concat as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const cosKernelFunc = unaryKernelFunc(Cos, (xi) => Math.cos(xi));\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'cpu',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cosh, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const coshKernelFunc = unaryKernelFunc(Cosh, (xi) => Math.cosh(xi));\n\nexport const coshConfig: KernelConfig = {\n  kernelName: Cosh,\n  backendName: 'cpu',\n  kernelFunc: coshKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2D, Dilation2DAttrs, Dilation2DInputs, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dConfig: KernelConfig = {\n  kernelName: Dilation2D,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter} = inputs as Dilation2DInputs;\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const xRank = x.shape.length;\n\n    const filterVals = cpuBackend.data.get(filter.dataId).values as TypedArray;\n    const filterRank = filter.shape.length;\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    const outSize = util.sizeFromShape(outShape);\n    const outRank = outShape.length;\n    const outputVals = util.getArrayFromDType(x.dtype, outSize);\n\n    // Upsampling the input by fill in `dilation size - 1` values between each\n    // input value.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const xIndex = util.locToIndex(\n                        [b, hIn, wIn, d], xRank, util.computeStrides(x.shape));\n                    const filterIndex = util.locToIndex(\n                        [h, w, d], filterRank,\n                        util.computeStrides(filter.shape));\n                    const val = xVals[xIndex] + filterVals[filterIndex];\n                    if (val > curVal) {\n                      curVal = val;\n                    }\n                  }\n                }\n              }\n            }\n            const outputIndex = util.locToIndex(\n                [b, hOut, wOut, d], outRank, util.computeStrides(outShape));\n            outputVals[outputIndex] = curVal;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(outputVals, x.dtype), outShape, x.dtype);\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropFilter, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropFilterConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropFilter,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropFilter}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed filter gradients has the same dimensions as the filter:\n    // [filterHeight, filterWidth, depth]\n    const gradients = util.makeZerosNestedTypedArray(\n                          filter.shape, filter.dtype) as number[][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hMax = 0;\n            let wMax = 0;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hMax = h;\n                      wMax = w;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[hMax][wMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), filter.shape, filter.dtype);\n\n    return {dataId, shape: filter.shape, dtype: filter.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Dilation2DAttrs, Dilation2DBackpropInput, Tensor3D, Tensor4D, TypedArray, util} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const dilation2dBackpropInputConfig: KernelConfig = {\n  kernelName: Dilation2DBackpropInput,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {x, filter, dy} =\n        inputs as {x: Tensor4D, filter: Tensor3D, dy: Tensor4D};\n    const {strides, pad, dilations} = attrs as {} as Dilation2DAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const $x =\n        util.toNestedArray(\n            x.shape, cpuBackend.data.get(x.dataId).values as TypedArray) as\n        number[][][][];\n\n    const $filter = util.toNestedArray(\n                        filter.shape,\n                        cpuBackend.data.get(filter.dataId).values as\n                            TypedArray) as number[][][];\n\n    const {\n      batchSize,\n      inHeight,\n      inWidth,\n      inChannels,\n      outHeight,\n      outWidth,\n      padInfo,\n      strideHeight,\n      strideWidth,\n      filterHeight,\n      filterWidth,\n      dilationHeight,\n      dilationWidth,\n      outShape\n    } =\n        backend_util.computeDilation2DInfo(\n            x.shape as [number, number, number, number],\n            filter.shape as [number, number, number], strides, pad,\n            'NHWC' /* dataFormat */, dilations);\n\n    util.assert(\n        dy.rank === outShape.length,\n        () => `Error in ${Dilation2DBackpropInput}, dy ` +\n            `must have the same rank as output ${outShape.length}, but got ` +\n            `${dy.rank}`);\n\n    const $dy =\n        util.toNestedArray(\n            outShape, cpuBackend.data.get(dy.dataId).values as TypedArray) as\n        number[][][][];\n\n    // The computed gradients has the same dimensions as the input:\n    // [batch, inputHeight, inputCols, inChannel]\n    const gradients =\n        util.makeZerosNestedTypedArray(x.shape, x.dtype) as number[][][][];\n\n    // In the case of multiple argmax branches, we only back-propagate along the\n    // last branch, i.e., the one with largest value of `h * filter_cols + w`,\n    // similarly to the max-pooling backward routines.\n    // This implementation follows the TF c++ implementation:\n    // https://github.com/tensorflow/tensorflow/blob/d9a3a849edc198e90172bc58eb293de457f9d986/tensorflow/core/kernels/dilation_ops.cc\n    for (let b = 0; b < batchSize; ++b) {\n      for (let hOut = 0; hOut < outHeight; ++hOut) {\n        const hBeg = hOut * strideHeight - padInfo.top;\n        for (let wOut = 0; wOut < outWidth; ++wOut) {\n          const wBeg = wOut * strideWidth - padInfo.left;\n          for (let d = 0; d < inChannels; ++d) {\n            let curVal = Number.MIN_SAFE_INTEGER;\n            let hInMax = (hBeg < 0) ? 0 : hBeg;\n            let wInMax = (wBeg < 0) ? 0 : wBeg;\n            for (let h = 0; h < filterHeight; ++h) {\n              const hIn = hBeg + h * dilationHeight;\n              if (hIn >= 0 && hIn < inHeight) {\n                for (let w = 0; w < filterWidth; ++w) {\n                  const wIn = wBeg + w * dilationWidth;\n                  if (wIn >= 0 && wIn < inWidth) {\n                    const val = $x[b][hIn][wIn][d] + $filter[h][w][d];\n                    if (val > curVal) {\n                      curVal = val;\n                      hInMax = hIn;\n                      wInMax = wIn;\n                    }\n                  }\n                }\n              }\n            }\n            gradients[b][hInMax][wInMax][d] += $dy[b][hOut][wOut][d];\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(\n        util.toTypedArray(gradients, x.dtype), x.shape, x.dtype);\n\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const divImpl =\n    createSimpleBinaryKernelImpl((a: number, b: number) => a / b);\nexport const div = binaryKernelFunc(Div, divImpl);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'cpu',\n  kernelFunc: div\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Elu, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const eluKernelFunc =\n    unaryKernelFunc(Elu, (xi) => xi >= 0 ? xi : (Math.exp(xi) - 1));\n\nexport const eluConfig: KernelConfig = {\n  kernelName: Elu,\n  backendName: 'cpu',\n  kernelFunc: eluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Erf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst p = backend_util.ERF_P;\nconst a1 = backend_util.ERF_A1;\nconst a2 = backend_util.ERF_A2;\nconst a3 = backend_util.ERF_A3;\nconst a4 = backend_util.ERF_A4;\nconst a5 = backend_util.ERF_A5;\n\nexport const erfKernelFunc = unaryKernelFunc(\n    Erf,\n    (xi) => {\n      const sign = Math.sign(xi);\n      const v = Math.abs(xi);\n      const t = 1.0 / (1.0 + p * v);\n      return sign *\n          (1.0 -\n           (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t *\n               Math.exp(-v * v));\n    },\n);\n\nexport const erfConfig: KernelConfig = {\n  kernelName: Erf,\n  backendName: 'cpu',\n  kernelFunc: erfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, Tensor, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {add} from '../kernels/Add';\nimport {complex} from '../kernels/Complex';\nimport {concat} from '../kernels/Concat';\nimport {divConfig} from '../kernels/Div';\nimport {identity} from '../kernels/Identity';\nimport {imag} from '../kernels/Imag';\nimport {multiply} from '../kernels/Multiply';\nimport {real} from '../kernels/Real';\nimport {slice} from '../kernels/Slice';\nimport {sub} from '../kernels/Sub';\n\n/**\n * Calculate FFT of inner most elements of batch tensor.\n */\nexport function fftBatch(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): TensorInfo {\n  const inputShape = input.shape;\n  const batch = inputShape[0];\n  const innerDim = inputShape[1];\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const real2D = inputVals.complexTensorInfos.real;\n  const imag2D = inputVals.complexTensorInfos.imag;\n\n  // Collects real and imaginary values separately.\n  const resultShape = [batch, innerDim];\n  const resultSize = util.sizeFromShape(resultShape);\n  const resultReal = util.getTypedArrayFromDType('float32', resultSize);\n  const resultImag = util.getTypedArrayFromDType('float32', resultSize);\n\n  for (let b = 0; b < batch; b++) {\n    // TODO: Support slice ops for complex type.\n    const r = slice({\n      inputs: {x: real2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n    const i = slice({\n      inputs: {x: imag2D},\n      backend: cpuBackend,\n      attrs: {begin: [b, 0], size: [1, innerDim]}\n    });\n\n    const input = complex({inputs: {real: r, imag: i}, backend: cpuBackend});\n\n    // Run FFT by batch element.\n    const {real, imag} = fftImpl(input, inverse, cpuBackend);\n    const res = backend_util.mergeRealAndImagArrays(real, imag);\n\n    for (let d = 0; d < innerDim; d++) {\n      const c = backend_util.getComplexWithIndex(res, d);\n      resultReal[b * innerDim + d] = c.real;\n      resultImag[b * innerDim + d] = c.imag;\n    }\n\n    cpuBackend.disposeIntermediateTensorInfo(r);\n    cpuBackend.disposeIntermediateTensorInfo(i);\n    cpuBackend.disposeIntermediateTensorInfo(input);\n  }\n\n  const $realInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultReal);\n  const $imagInfo: TensorInfo =\n      cpuBackend.makeTensorInfo(resultShape, 'float32', resultImag);\n\n  const result = complex(\n      {inputs: {real: $realInfo, imag: $imagInfo}, backend: cpuBackend});\n\n  cpuBackend.disposeIntermediateTensorInfo($realInfo);\n  cpuBackend.disposeIntermediateTensorInfo($imagInfo);\n\n  return result;\n}\n\nexport function fftImpl(\n    input: TensorInfo, inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  const inputSize = util.sizeFromShape(input.shape);\n\n  const inputVals = cpuBackend.data.get(input.dataId);\n\n  const realVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.real.dataId).values as\n      Float32Array;\n\n  const imagVals =\n      cpuBackend.data.get(inputVals.complexTensorInfos.imag.dataId).values as\n      Float32Array;\n\n  if (isExponentOf2(inputSize)) {\n    const result =\n        fftRadix2(realVals, imagVals, inputSize, inverse, cpuBackend);\n\n    const resultShape = [input.shape[0], input.shape[1]];\n\n    if (inverse) {\n      const realInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.real);\n      const imagInfo: TensorInfo =\n          cpuBackend.makeTensorInfo(resultShape, 'float32', result.imag);\n\n      const sizeInfo: TensorInfo = cpuBackend.makeTensorInfo(\n          [], 'float32',\n          util.createScalarValue(inputSize as {} as 'float32', 'float32'));\n      const sizeInfoCopy =\n          identity({inputs: {x: sizeInfo}, backend: cpuBackend});\n\n      const divRealInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: realInfo, b: sizeInfo}, backend: cpuBackend}) as\n          TensorInfo;\n      const divImagInfo =\n          divConfig.kernelFunc(\n              {inputs: {a: imagInfo, b: sizeInfoCopy}, backend: cpuBackend}) as\n          TensorInfo;\n\n      const divRealVals =\n          cpuBackend.data.get(divRealInfo.dataId).values as Float32Array;\n      const divImagVals =\n          cpuBackend.data.get(divImagInfo.dataId).values as Float32Array;\n\n      cpuBackend.disposeIntermediateTensorInfo(realInfo);\n      cpuBackend.disposeIntermediateTensorInfo(imagInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfo);\n      cpuBackend.disposeIntermediateTensorInfo(sizeInfoCopy);\n      cpuBackend.disposeIntermediateTensorInfo(divRealInfo);\n      cpuBackend.disposeIntermediateTensorInfo(divImagInfo);\n\n      return {real: divRealVals, imag: divImagVals};\n    }\n\n    return result;\n  } else {\n    const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n    const rawOutput =\n        fourierTransformByMatmul(data, inputSize, inverse) as Float32Array;\n\n    return backend_util.splitRealAndImagArrays(rawOutput);\n  }\n}\n\nfunction isExponentOf2(size: number): boolean {\n  return (size & size - 1) === 0;\n}\n\n// FFT using Cooley-Tukey algorithm on radix 2 dimensional input.\nfunction fftRadix2(\n    realVals: Float32Array, imagVals: Float32Array, size: number,\n    inverse: boolean,\n    cpuBackend: MathBackendCPU): {real: Float32Array, imag: Float32Array} {\n  if (size === 1) {\n    return {real: realVals, imag: imagVals};\n  }\n\n  const data = backend_util.mergeRealAndImagArrays(realVals, imagVals);\n\n  const half = size / 2;\n\n  const evenComplex = backend_util.complexWithEvenIndex(data);\n\n  const evenRealVals = evenComplex.real;\n  const evenImagVals = evenComplex.imag;\n\n  const evenShape = [evenRealVals.length];\n\n  const evenRealInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenRealVals);\n  const evenImagInfo =\n      cpuBackend.makeTensorInfo(evenShape, 'float32', evenImagVals);\n\n  const evenTensorInfo = complex(\n      {inputs: {real: evenRealInfo, imag: evenImagInfo}, backend: cpuBackend});\n\n  const oddComplex = backend_util.complexWithOddIndex(data);\n\n  const oddRealVals = oddComplex.real;\n  const oddImagVals = oddComplex.imag;\n\n  const oddShape = [oddRealVals.length];\n\n  const oddRealInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddRealVals);\n  const oddImagInfo =\n      cpuBackend.makeTensorInfo(oddShape, 'float32', oddImagVals);\n\n  const oddTensorInfo = complex(\n      {inputs: {real: oddRealInfo, imag: oddImagInfo}, backend: cpuBackend});\n\n  // Recursive call for half part of original input.\n  const $evenComplex =\n      fftRadix2(evenRealVals, evenImagVals, half, inverse, cpuBackend);\n\n  const $evenRealVals = $evenComplex.real;\n  const $evenImagVals = $evenComplex.imag;\n\n  const $evenShape = [$evenRealVals.length];\n\n  const $evenRealInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenRealVals);\n  const $evenImagInfo =\n      cpuBackend.makeTensorInfo($evenShape, 'float32', $evenImagVals);\n\n  const $evenTensorInfo = complex({\n    inputs: {real: $evenRealInfo, imag: $evenImagInfo},\n    backend: cpuBackend\n  });\n\n  const $oddComplex =\n      fftRadix2(oddRealVals, oddImagVals, half, inverse, cpuBackend);\n\n  const $oddRealVals = $oddComplex.real;\n  const $oddImagVals = $oddComplex.imag;\n\n  const $oddShape = [$oddRealVals.length];\n\n  const $oddRealInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddRealVals);\n  const $oddImagInfo =\n      cpuBackend.makeTensorInfo($oddShape, 'float32', $oddImagVals);\n\n  const $oddTensorInfo = complex(\n      {inputs: {real: $oddRealInfo, imag: $oddImagInfo}, backend: cpuBackend});\n\n  const e = backend_util.exponents(size, inverse);\n  const eShape = [e.real.length];\n\n  const eRealInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.real);\n  const eImagInfo = cpuBackend.makeTensorInfo(eShape, 'float32', e.imag);\n\n  const complexInfo = complex(\n      {inputs: {real: eRealInfo, imag: eImagInfo}, backend: cpuBackend});\n\n  const exponentInfo =\n      multiply(\n          {inputs: {a: complexInfo, b: $oddTensorInfo}, backend: cpuBackend}) as\n      TensorInfo;\n\n  const addPart = add({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n  const subPart = sub({\n                    inputs: {a: $evenTensorInfo, b: exponentInfo},\n                    backend: cpuBackend\n                  }) as TensorInfo;\n\n  const addPartReal = real({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartReal = real({inputs: {input: subPart}, backend: cpuBackend});\n\n  const addPartImag = imag({inputs: {input: addPart}, backend: cpuBackend});\n  const subPartImag = imag({inputs: {input: subPart}, backend: cpuBackend});\n\n  const $real = concat({\n    inputs: [addPartReal as Tensor, subPartReal as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n  const $imag = concat({\n    inputs: [addPartImag as Tensor, subPartImag as Tensor],\n    backend: cpuBackend,\n    attrs: {axis: 0}\n  });\n\n  const $realVals = cpuBackend.data.get($real.dataId).values as Float32Array;\n  const $imagVals = cpuBackend.data.get($imag.dataId).values as Float32Array;\n\n  cpuBackend.disposeIntermediateTensorInfo(evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($evenTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo($oddTensorInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eRealInfo);\n  cpuBackend.disposeIntermediateTensorInfo(eImagInfo);\n  cpuBackend.disposeIntermediateTensorInfo(complexInfo);\n  cpuBackend.disposeIntermediateTensorInfo(exponentInfo);\n  cpuBackend.disposeIntermediateTensorInfo(addPart);\n  cpuBackend.disposeIntermediateTensorInfo(subPart);\n  cpuBackend.disposeIntermediateTensorInfo(addPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(addPartImag);\n  cpuBackend.disposeIntermediateTensorInfo(subPartReal);\n  cpuBackend.disposeIntermediateTensorInfo(subPartImag);\n  cpuBackend.disposeIntermediateTensorInfo($real);\n  cpuBackend.disposeIntermediateTensorInfo($imag);\n\n  return {real: $realVals, imag: $imagVals};\n}\n\n// Calculate fourier transform by multplying sinusoid matrix.\nfunction fourierTransformByMatmul(\n    data: TypedArray, size: number, inverse: boolean): TypedArray {\n  const ret = new Float32Array(size * 2);\n  // TODO: Use matmul instead once it supports complex64 type.\n  for (let r = 0; r < size; r++) {\n    let real = 0.0;\n    let imag = 0.0;\n    for (let c = 0; c < size; c++) {\n      const e = backend_util.exponent(r * c, size, inverse);\n      const term = backend_util.getComplexWithIndex(data as Float32Array, c);\n      real += term.real * e.real - term.imag * e.imag;\n      imag += term.real * e.imag + term.imag * e.real;\n    }\n    if (inverse) {\n      real /= size;\n      imag /= size;\n    }\n    backend_util.assignToTypedArray(ret, real, imag, r);\n  }\n  return ret;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {FFT, FFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function fft(args: {inputs: FFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, false, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const fftConfig: KernelConfig = {\n  kernelName: FFT,\n  backendName: 'cpu',\n  kernelFunc: fft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n\n            const coordX = Math.round(imageWidth - x);\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n\n            let outputValue = imageVals[outIdx];\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth) {\n              // set the output to the image value at the coordinate position.\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n            output[outIdx] = outputValue;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IFFT, IFFTInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {fftBatch} from '../utils/fft_utils';\nimport {reshape} from './Reshape';\n\nexport function ifft(args: {inputs: IFFTInputs, backend: MathBackendCPU}):\n    TensorInfo {\n  const {inputs, backend} = args;\n  const {input} = inputs;\n\n  const inputSize = util.sizeFromShape(input.shape);\n\n  // Collapse all outer dimensions to a single batch dimension.\n  const innerDimensionSize = input.shape[input.shape.length - 1];\n  const batch = inputSize / innerDimensionSize;\n\n  const input2D = reshape({\n    inputs: {x: input},\n    backend,\n    attrs: {shape: [batch, innerDimensionSize]}\n  });\n\n  const result = fftBatch(input2D, true, backend);\n\n  const resultReshaped =\n      reshape({inputs: {x: result}, backend, attrs: {shape: input.shape}});\n\n  backend.disposeIntermediateTensorInfo(input2D);\n  backend.disposeIntermediateTensorInfo(result);\n\n  return resultReshaped;\n}\n\nexport const ifftConfig: KernelConfig = {\n  kernelName: IFFT,\n  backendName: 'cpu',\n  kernelFunc: ifft as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsFinite, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isFiniteKernelFunc =\n    unaryKernelFunc(IsFinite, (xi) => Number.isFinite(xi) ? 1 : 0, 'bool');\n\nexport const isFiniteConfig: KernelConfig = {\n  kernelName: IsFinite,\n  backendName: 'cpu',\n  kernelFunc: isFiniteKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsInf, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isInfKernelFunc =\n    unaryKernelFunc(IsInf, (xi) => Math.abs(xi) === Infinity ? 1 : 0, 'bool');\n\nexport const isInfConfig: KernelConfig = {\n  kernelName: IsInf,\n  backendName: 'cpu',\n  kernelFunc: isInfKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IsNan, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const isNaNKernelFunc =\n    unaryKernelFunc(IsNan, (xi) => Number.isNaN(xi) ? 1 : 0, 'bool');\n\nexport const isNaNConfig: KernelConfig = {\n  kernelName: IsNan,\n  backendName: 'cpu',\n  kernelFunc: isNaNKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Log1p} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const log1pKernelFunc = unaryKernelFunc(Log1p, (xi) => Math.log1p(xi));\n\nexport const log1pConfig: KernelConfig = {\n  kernelName: Log1p,\n  backendName: 'cpu',\n  kernelFunc: log1pKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, LogicalNot} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const logicalNotKernelFunc =\n    unaryKernelFunc(LogicalNot, (xi) => xi ? 0 : 1, 'bool');\n\nexport const logicalNotConfig: KernelConfig = {\n  kernelName: LogicalNot,\n  backendName: 'cpu',\n  kernelFunc: logicalNotKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig} from '@tensorflow/tfjs-core';\nimport {TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    let xShape = x.shape;\n    const xRank = xShape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, xShape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    let xVals = cpuBackend.data.get(x.dataId).values as TypedArray;\n    if (permutedAxes != null) {\n      const newShape: number[] = new Array(xRank);\n      for (let i = 0; i < newShape.length; i++) {\n        newShape[i] = xShape[permutedAxes[i]];\n      }\n\n      xVals = transposeImpl(xVals, xShape, x.dtype, permutedAxes, newShape);\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n\n      xShape = newShape;\n    }\n\n    assertNotComplex(x, 'max');\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(xShape, axes);\n\n    const reduceSize = util.sizeFromShape(reduceShape);\n\n    const result = maxImpl(xVals, reduceSize, maxOutShape, x.dtype);\n    const dataId = cpuBackend.write(result, maxOutShape, x.dtype);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // reshape\n      const newShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n      outShape = newShape;\n    }\n\n    return {dataId, shape: outShape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {pool} from '../utils/pool_utils';\nimport {identity} from './Identity';\n\nexport function maxPool(\n    args:\n        {inputs: MaxPoolInputs, backend: MathBackendCPU, attrs: MaxPoolAttrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  let res: TensorInfo;\n\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    res = identity({inputs: {x}, backend});\n  } else {\n    const xValues = backend.data.get(x.dataId).values as TypedArray;\n    const strides = util.computeStrides(x.shape);\n    const buffer = pool(xValues, x.shape, x.dtype, strides, convInfo, 'max');\n    res = backend.makeTensorInfo(\n        convInfo.outShape, x.dtype, buffer.values as TypedArray);\n  }\n  return res;\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'cpu',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, buffer, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, Rank, TensorInfo, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\nimport {maxPoolPositions} from '../utils/pool_utils';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendCPU,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const xValues = backend.data.get(x.dataId).values as TypedArray;\n  const maxPosBuf = buffer(\n      convInfo.outShape, x.dtype,\n      maxPoolPositions(xValues, x.shape, x.dtype, convInfo).values);\n  const strideHeight = convInfo.strideHeight;\n  const strideWidth = convInfo.strideWidth;\n  const dilationHeight = convInfo.dilationHeight;\n  const dilationWidth = convInfo.dilationWidth;\n  const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n  const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n  const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n  const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n  const dx =\n      buffer<Rank.R4>(x.shape as [number, number, number, number], 'float32');\n\n  const dyData = backend.data.get(dy.dataId).values as Float32Array;\n  const dyBuf = buffer<Rank.R4>(\n      dy.shape as [number, number, number, number], 'float32', dyData);\n\n  for (let b = 0; b < convInfo.batchSize; ++b) {\n    for (let d = 0; d < convInfo.inChannels; ++d) {\n      for (let dxR = 0; dxR < convInfo.inHeight; ++dxR) {\n        for (let dxC = 0; dxC < convInfo.inWidth; ++dxC) {\n          // Shader code begins.\n          const dyRCorner = dxR - padTop;\n          const dyCCorner = dxC - padLeft;\n          let dotProd = 0;\n          for (let wR = 0; wR < effectiveFilterHeight; wR += dilationHeight) {\n            const dyR = (dyRCorner + wR) / strideHeight;\n            if (dyR < 0 || dyR >= convInfo.outHeight ||\n                Math.floor(dyR) !== dyR) {\n              continue;\n            }\n            for (let wC = 0; wC < effectiveFilterWidth; wC += dilationWidth) {\n              const dyC = (dyCCorner + wC) / strideWidth;\n              if (dyC < 0 || dyC >= convInfo.outWidth ||\n                  Math.floor(dyC) !== dyC) {\n                continue;\n              }\n              const maxPos = effectiveFilterHeight * effectiveFilterWidth - 1 -\n                  (maxPosBuf.get(b, dyR, dyC, d) as number);\n              const curPos = wR * effectiveFilterWidth + wC;\n\n              const mask = maxPos === curPos ? 1 : 0;\n              if (mask === 0) {\n                continue;\n              }\n\n              const pixel = dyBuf.get(b, dyR, dyC, d);\n              dotProd += pixel * mask;\n            }\n          }\n          dx.set(dotProd, b, dxR, dxC, d);\n        }\n      }\n    }\n  }\n  return backend.makeTensorInfo(dx.shape, dx.dtype, dx.values);\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'cpu',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, DataType, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {maxPoolPositions, pool} from '../utils/pool_utils';\nexport function maxPoolWithArgmaxImpl(\n    xValues: TypedArray, xShape: number[], dtype: DataType,\n    includeBatchInIndex: boolean, convInfo: backend_util.Conv2DInfo) {\n  const strides = util.computeStrides(xShape);\n  const maxPools = pool(xValues, xShape, dtype, strides, convInfo, 'max');\n  const maxPositions = maxPoolPositions(\n      xValues, xShape, dtype, convInfo, true, includeBatchInIndex);\n\n  return [maxPools.values, maxPositions.values];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'MaxPoolWithArgmax');\n\n    const values = cpuBackend.data.get(x.dataId).values as TypedArray;\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        [1, 1], pad);\n    const [pooled, indexes] = maxPoolWithArgmaxImpl(\n        values, x.shape, x.dtype, includeBatchInIndex, convInfo);\n\n    const pooledDataId =\n        cpuBackend.write(pooled as Float32Array, convInfo.outShape, x.dtype);\n    const indexesDataId =\n        cpuBackend.write(indexes as Int32Array, convInfo.outShape, x.dtype);\n    return [\n      {dataId: pooledDataId, shape: convInfo.outShape, dtype: x.dtype},\n      {dataId: indexesDataId, shape: convInfo.outShape, dtype: 'int32'}\n    ];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionPadded');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {kernel_impls} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const cpuBackend = backend as MathBackendCPU;\n\n    assertNotComplex(boxes, 'NonMaxSuppressionWithScore');\n\n    const boxesVals = cpuBackend.data.get(boxes.dataId).values as TypedArray;\n    const scoresVals = cpuBackend.data.get(scores.dataId).values as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NotEqual} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const notEqualImpl =\n    createSimpleBinaryKernelImpl(((a, b) => (a !== b) ? 1 : 0));\nexport const notEqual =\n    binaryKernelFunc(NotEqual, notEqualImpl, null /* complexOp */, 'bool');\n\nexport const notEqualConfig: KernelConfig = {\n  kernelName: NotEqual,\n  backendName: 'cpu',\n  kernelFunc: notEqual\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, NumericDataType, PadV2, PadV2Attrs, PadV2Inputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport function padV2(\n    args: {inputs: PadV2Inputs, backend: MathBackendCPU, attrs: PadV2Attrs}):\n    TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {paddings, constantValue} = attrs;\n\n  assertNotComplex(x, 'pad');\n\n  const outShape = paddings.map(\n      (p, i) => p[0] /* beforePad */ + x.shape[i] + p[1] /* afterPad */);\n\n  const start = paddings.map(p => p[0]);\n\n  const xVals = backend.data.get(x.dataId).values as TypedArray;\n  const xSize = util.sizeFromShape(x.shape);\n  const xRank = x.shape.length;\n  const xStrides = util.computeStrides(x.shape);\n\n  const resultSize = util.sizeFromShape(outShape);\n  const resultRank = outShape.length;\n  const resultStrides = util.computeStrides(outShape);\n  const resVals =\n      util.getTypedArrayFromDType(x.dtype as NumericDataType, resultSize);\n\n  if (constantValue !== 0) {\n    resVals.fill(constantValue);\n  }\n\n  for (let i = 0; i < xSize; i++) {\n    const coords = util.indexToLoc(i, xRank, xStrides);\n    const outCoords = coords.map((c, i) => c + start[i]);\n    const outIndex = util.locToIndex(outCoords, resultRank, resultStrides);\n\n    resVals[outIndex] = xVals[i];\n  }\n\n  const outId = backend.write(resVals, outShape, x.dtype);\n\n  return {dataId: outId, shape: outShape, dtype: x.dtype};\n}\n\nexport const padV2Config: KernelConfig = {\n  kernelName: PadV2,\n  backendName: 'cpu',\n  kernelFunc: padV2 as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Reciprocal} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const reciprocalKernelFunc = unaryKernelFunc(Reciprocal, (xi) => 1 / xi);\n\nexport const reciprocalConfig: KernelConfig = {\n  kernelName: Reciprocal,\n  backendName: 'cpu',\n  kernelFunc: reciprocalKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, NumericDataType, TypedArray} from '@tensorflow/tfjs-core';\nimport {backend_util, RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const cpuBackend = backend as MathBackendCPU;\n\n    const output = util.getTypedArrayFromDType(\n        image.dtype as NumericDataType, util.sizeFromShape(image.shape));\n    const [batch, imageHeight, imageWidth, numChannels] = image.shape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const fullOpacityValue = 255;\n\n    const sinFactor = Math.sin(radians);\n    const cosFactor = Math.cos(radians);\n    const imageVals = cpuBackend.data.get(image.dataId).values as TypedArray;\n\n    for (let batchIdx = 0; batchIdx < batch; batchIdx++) {\n      const batchOffset = batchIdx * imageWidth * imageHeight * numChannels;\n\n      for (let row = 0; row < imageHeight; row++) {\n        const rowOffset = row * (imageWidth * numChannels);\n\n        for (let col = 0; col < imageWidth; col++) {\n          const colOffset = col * numChannels;\n\n          for (let channel = 0; channel < numChannels; channel++) {\n            const coords = [batch, row, col, channel];\n\n            const x = coords[2];\n            const y = coords[1];\n\n            // coordX/coordY are the result of rotating and translating x/y.\n            let coordX = (x - centerX) * cosFactor - (y - centerY) * sinFactor;\n            let coordY = (x - centerX) * sinFactor + (y - centerY) * cosFactor;\n            coordX = Math.round(coordX + centerX);\n            coordY = Math.round(coordY + centerY);\n\n            let outputValue = fillValue;\n            if (typeof fillValue !== 'number') {\n              if (channel === 3) {\n                outputValue = fullOpacityValue;\n              } else {\n                outputValue = fillValue[channel];\n              }\n            }\n\n            // If the coordinate position falls within the image boundaries...\n            if (coordX >= 0 && coordX < imageWidth && coordY >= 0 &&\n                coordY < imageHeight) {\n              // set the output to the image value at the coordinate position.\n              const rotatedRowOffset = coordY * (imageWidth * numChannels);\n              const rotatedColOffset = coordX * numChannels;\n              const imageIdx =\n                  batchOffset + rotatedRowOffset + rotatedColOffset + channel;\n              outputValue = imageVals[imageIdx];\n            }\n\n            const outIdx = batchOffset + rowOffset + colOffset + channel;\n            output[outIdx] = outputValue as number;\n          }\n        }\n      }\n    }\n\n    const dataId = cpuBackend.write(output, image.shape, image.dtype);\n    return {dataId, shape: image.shape, dtype: image.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Round} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const roundKernelFunc = unaryKernelFunc(Round, (xi) => {\n  // The algorithm is based on banker's rounding.\n  const base = Math.floor(xi);\n  if (xi - base < 0.5) {\n    return Math.floor(xi);\n  } else if (xi - base > 0.5) {\n    return Math.ceil(xi);\n  } else {\n    if (base % 2.0 === 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n});\n\nexport const roundConfig: KernelConfig = {\n  kernelName: Round,\n  backendName: 'cpu',\n  kernelFunc: roundKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, Selu} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nconst scaleAlpha = backend_util.SELU_SCALEALPHA;\nconst scale = backend_util.SELU_SCALE;\n\nexport const seluKernelFunc = unaryKernelFunc(Selu, (xi) => {\n  if (xi >= 0) {\n    return scale * xi;\n  } else {\n    return scaleAlpha * (Math.exp(xi) - 1);\n  }\n});\n\nexport const seluConfig: KernelConfig = {\n  kernelName: Selu,\n  backendName: 'cpu',\n  kernelFunc: seluKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sigmoid} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sigmoidKernelFunc =\n    unaryKernelFunc(Sigmoid, (xi) => 1 / (1 + Math.exp(-xi)));\n\nexport const sigmoidConfig: KernelConfig = {\n  kernelName: Sigmoid,\n  backendName: 'cpu',\n  kernelFunc: sigmoidKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sign} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const signKernelFunc = unaryKernelFunc(Sign, (xi) => {\n  if (xi < 0) {\n    return -1;\n  } else if (xi > 0) {\n    return 1;\n  } else {\n    return 0;\n  }\n});\n\nexport const signConfig: KernelConfig = {\n  kernelName: Sign,\n  backendName: 'cpu',\n  kernelFunc: signKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinKernelFunc = unaryKernelFunc(Sin, (xi) => Math.sin(xi));\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'cpu',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sinh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sinhKernelFunc = unaryKernelFunc(Sinh, (xi) => Math.sinh(xi));\n\nexport const sinhConfig: KernelConfig = {\n  kernelName: Sinh,\n  backendName: 'cpu',\n  kernelFunc: sinhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Softplus} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\n// mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n\n// epsilon is the difference between 1.0 and the next representable float.\n// For a single precision 32 bit float this should be 2^-23, see:\n// https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\nconst epsilon = 1.1920928955078125e-7;\nconst threshold = Math.log(epsilon) + 2.0;\n\nexport const softplusKernelFunc = unaryKernelFunc(Softplus, (xi) => {\n  // Value above which exp(x) may overflow, but softplus(x) == x\n  // is within machine epsilon.\n  const tooLarge = xi > -threshold;\n\n  // Value below which exp(x) may underflow, but softplus(x) == exp(x)\n  // is within machine epsilon.\n  const tooSmall = xi < threshold;\n\n  const expX = Math.exp(xi);\n  let result;\n\n  if (tooSmall) {\n    result = expX;\n  } else if (tooLarge) {\n    result = xi;\n  } else {\n    result = Math.log(1.0 + expX);\n  }\n  return result;\n});\n\nexport const softplusConfig: KernelConfig = {\n  kernelName: Softplus,\n  backendName: 'cpu',\n  kernelFunc: softplusKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {transposeImpl} from './Transpose_impl';\n\nexport function transpose(args: {\n  inputs: TransposeInputs,\n  attrs: TransposeAttrs,\n  backend: MathBackendCPU\n}): TensorInfo {\n  const {inputs, attrs, backend} = args;\n  const {x} = inputs;\n  const {perm} = attrs;\n\n  assertNotComplex(x, 'transpose');\n\n  const xRank = x.shape.length;\n\n  const newShape: number[] = new Array(xRank);\n  for (let i = 0; i < newShape.length; i++) {\n    newShape[i] = x.shape[perm[i]];\n  }\n\n  const values = backend.data.get(x.dataId).values as TypedArray;\n  const result = transposeImpl(values, x.shape, x.dtype, perm, newShape);\n\n  const dataId = backend.write(result, newShape, x.dtype);\n  return {dataId, shape: newShape, dtype: x.dtype};\n}\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'cpu',\n  kernelFunc: transpose as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, KernelConfig, KernelFunc, ReshapeAttrs, ReshapeInputs, SpaceToBatchND, SpaceToBatchNDAttrs, SpaceToBatchNDInputs, TensorInfo, TransposeAttrs, TransposeInputs, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {padV2Config} from './PadV2';\nimport {reshape} from './Reshape';\nimport {transpose} from './Transpose';\n\nexport function spaceToBatchND(args: {\n  inputs: SpaceToBatchNDInputs,\n  backend: MathBackendCPU,\n  attrs: SpaceToBatchNDAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {blockShape, paddings} = attrs;\n\n  assertNotComplex([x], 'spaceToBatchND');\n\n  const prod = util.sizeFromShape(blockShape);\n\n  const completePaddings: Array<[number, number]> = [[0, 0]];\n  completePaddings.push(...(paddings as Array<[number, number]>));\n\n  for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n    completePaddings.push([0, 0]);\n  }\n\n  const paddedX = padV2Config.kernelFunc({\n    inputs: {x},\n    backend,\n    attrs: {paddings: completePaddings, constantValue: 0}\n  }) as TensorInfo;\n\n  const reshapedPaddedShape =\n      backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n  const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n      reshapedPaddedShape.length, blockShape.length, false);\n\n  const flattenShape =\n      backend_util.getReshapedPermuted(paddedX.shape, blockShape, prod, false);\n\n  const reshapeInputs: ReshapeInputs = {x: paddedX};\n  const reshapeAttrs: ReshapeAttrs = {shape: reshapedPaddedShape};\n  const paddedXReshaped =\n      reshape({inputs: reshapeInputs, backend, attrs: reshapeAttrs});\n\n  const transposeInputs: TransposeInputs = {x: paddedXReshaped};\n  const transposeAttrs:\n      TransposeAttrs = {perm: permutedReshapedPaddedPermutation};\n  const paddedXT =\n      transpose({inputs: transposeInputs, backend, attrs: transposeAttrs});\n\n  const resultReshapeInputs: ReshapeInputs = {x: paddedXT};\n  const resultReshapeAttrs: ReshapeAttrs = {shape: flattenShape};\n  const result = reshape(\n      {inputs: resultReshapeInputs, backend, attrs: resultReshapeAttrs});\n\n  backend.disposeIntermediateTensorInfo(paddedX);\n  backend.disposeIntermediateTensorInfo(paddedXReshaped);\n  backend.disposeIntermediateTensorInfo(paddedXT);\n\n  return result;\n}\n\nexport const spaceToBatchNDConfig: KernelConfig = {\n  kernelName: SpaceToBatchND,\n  backendName: 'cpu',\n  kernelFunc: spaceToBatchND as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sqrt} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const sqrtKernelFunc = unaryKernelFunc(Sqrt, (xi) => Math.sqrt(xi));\n\nexport const sqrtConfig: KernelConfig = {\n  kernelName: Sqrt,\n  backendName: 'cpu',\n  kernelFunc: sqrtKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Square, SquareInputs} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'cpu',\n  kernelFunc: ({inputs, backend}) => {\n    const {x} = inputs as SquareInputs;\n    const cpuBackend = backend as MathBackendCPU;\n    assertNotComplex(x, 'square');\n\n    const values = cpuBackend.data.get(x.dataId).values as Float32Array;\n    const newValues = new Float32Array(values.length);\n    for (let i = 0; i < values.length; ++i) {\n      const value = values[i];\n      newValues[i] = value * value;\n    }\n    const dataId = cpuBackend.write(newValues, x.shape, x.dtype);\n    return {dataId, shape: x.shape, dtype: x.dtype};\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {createSimpleBinaryKernelImpl} from '../utils/binary_impl';\nimport {binaryKernelFunc} from '../utils/kernel_utils';\n\nexport const squaredDifferenceImpl = createSimpleBinaryKernelImpl(((a, b) => {\n  const diff = a - b;\n  return diff * diff;\n}));\nexport const squaredDifference =\n    binaryKernelFunc(SquaredDifference, squaredDifferenceImpl);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'cpu',\n  kernelFunc: squaredDifference\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Step, StepAttrs} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const stepKernelFunc = unaryKernelFunc(Step, (xi, attrs) => {\n  const stepAttrs = attrs as {} as StepAttrs;\n  if (isNaN(xi)) {\n    return NaN;\n  } else {\n    return xi > 0 ? 1 : stepAttrs.alpha;\n  }\n});\n\nexport const stepConfig: KernelConfig = {\n  kernelName: Step,\n  backendName: 'cpu',\n  kernelFunc: stepKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanKernelFunc = unaryKernelFunc(Tan, (xi) => Math.tan(xi));\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'cpu',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tanh} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../utils/unary_utils';\n\nexport const tanhKernelFunc = unaryKernelFunc(Tanh, (xi) => Math.tanh(xi));\n\nexport const tanhConfig: KernelConfig = {\n  kernelName: Tanh,\n  backendName: 'cpu',\n  kernelFunc: tanhKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendCPU} from '../backend_cpu';\nimport {assertNotComplex} from '../cpu_util';\n\nimport {uniqueImpl} from './Unique_impl';\n\nexport function unique(\n    args: {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendCPU}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  const values = backend.data.get(x.dataId).values;\n  const {outputValues, outputShape, indices} =\n      uniqueImpl(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'cpu',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// We explicitly import the modular kernels so they get registered in the\n// global registry when we compile the library. A modular build would replace\n// the contents of this file and import only the kernels that are needed.\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {absConfig} from './kernels/Abs';\nimport {acosConfig} from './kernels/Acos';\nimport {acoshConfig} from './kernels/Acosh';\nimport {addConfig} from './kernels/Add';\nimport {asinConfig} from './kernels/Asin';\nimport {asinhConfig} from './kernels/Asinh';\nimport {atanConfig} from './kernels/Atan';\nimport {atanhConfig} from './kernels/Atanh';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {castConfig} from './kernels/Cast';\nimport {ceilConfig} from './kernels/Ceil';\nimport {clipConfig} from './kernels/Clip';\nimport {complexConfig} from './kernels/Complex';\nimport {concatConfig} from './kernels/Concat';\nimport {cosConfig} from './kernels/Cos';\nimport {coshConfig} from './kernels/Cosh';\nimport {dilation2dConfig} from './kernels/Dilation2D';\nimport {dilation2dBackpropFilterConfig} from './kernels/Dilation2DBackpropFilter';\nimport {dilation2dBackpropInputConfig} from './kernels/Dilation2DBackpropInput';\nimport {divConfig} from './kernels/Div';\nimport {eluConfig} from './kernels/Elu';\nimport {erfConfig} from './kernels/Erf';\nimport {expConfig} from './kernels/Exp';\nimport {expm1Config} from './kernels/Expm1';\nimport {fftConfig} from './kernels/FFT';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {floorConfig} from './kernels/Floor';\nimport {identityConfig} from './kernels/Identity';\nimport {ifftConfig} from './kernels/IFFT';\nimport {imagConfig} from './kernels/Imag';\nimport {isFiniteConfig} from './kernels/IsFinite';\nimport {isInfConfig} from './kernels/IsInf';\nimport {isNaNConfig} from './kernels/IsNaN';\nimport {logConfig} from './kernels/Log';\nimport {log1pConfig} from './kernels/Log1p';\nimport {logicalNotConfig} from './kernels/LogicalNot';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {multiplyConfig} from './kernels/Multiply';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {notEqualConfig} from './kernels/NotEqual';\nimport {padV2Config} from './kernels/PadV2';\nimport {realConfig} from './kernels/Real';\nimport {reciprocalConfig} from './kernels/Reciprocal';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {roundConfig} from './kernels/Round';\nimport {rsqrtConfig} from './kernels/Rsqrt';\nimport {seluConfig} from './kernels/Selu';\nimport {sigmoidConfig} from './kernels/Sigmoid';\nimport {signConfig} from './kernels/Sign';\nimport {sinConfig} from './kernels/Sin';\nimport {sinhConfig} from './kernels/Sinh';\nimport {sliceConfig} from './kernels/Slice';\nimport {softplusConfig} from './kernels/Softplus';\nimport {spaceToBatchNDConfig} from './kernels/SpaceToBatchND';\nimport {sqrtConfig} from './kernels/Sqrt';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {stepConfig} from './kernels/Step';\nimport {subConfig} from './kernels/Sub';\nimport {tanConfig} from './kernels/Tan';\nimport {tanhConfig} from './kernels/Tanh';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  absConfig,\n  acosConfig,\n  acoshConfig,\n  addConfig,\n  asinConfig,\n  asinhConfig,\n  atanConfig,\n  atanhConfig,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  castConfig,\n  ceilConfig,\n  clipConfig,\n  complexConfig,\n  concatConfig,\n  cosConfig,\n  coshConfig,\n  dilation2dConfig,\n  dilation2dBackpropInputConfig,\n  dilation2dBackpropFilterConfig,\n  divConfig,\n  eluConfig,\n  erfConfig,\n  expConfig,\n  expm1Config,\n  fftConfig,\n  flipLeftRightConfig,\n  floorConfig,\n  identityConfig,\n  ifftConfig,\n  imagConfig,\n  isFiniteConfig,\n  isInfConfig,\n  isNaNConfig,\n  logConfig,\n  log1pConfig,\n  logicalNotConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  maxConfig,\n  multiplyConfig,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  notEqualConfig,\n  padV2Config,\n  realConfig,\n  reciprocalConfig,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  roundConfig,\n  rsqrtConfig,\n  seluConfig,\n  sigmoidConfig,\n  signConfig,\n  sinConfig,\n  sinhConfig,\n  sliceConfig,\n  softplusConfig,\n  spaceToBatchNDConfig,\n  sqrtConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  stepConfig,\n  subConfig,\n  tanConfig,\n  tanhConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nconst contexts: {[key: string]: WebGLRenderingContext} = {};\n\nconst WEBGL_ATTRIBUTES: WebGLContextAttributes = {\n  alpha: false,\n  antialias: false,\n  premultipliedAlpha: false,\n  preserveDrawingBuffer: false,\n  depth: false,\n  stencil: false,\n  failIfMajorPerformanceCaveat: true\n};\n\nexport function setWebGLContext(\n    webGLVersion: number, gl: WebGLRenderingContext) {\n  contexts[webGLVersion] = gl;\n}\n\nexport function getWebGLContext(webGLVersion: number): WebGLRenderingContext {\n  if (!(webGLVersion in contexts)) {\n    const newCtx = getWebGLRenderingContext(webGLVersion);\n    if (newCtx !== null) {\n      contexts[webGLVersion] = newCtx;\n    } else {\n      console.log('Could not get context for WebGL version', webGLVersion);\n      return null;\n    }\n  }\n  const gl = contexts[webGLVersion];\n  if (gl.isContextLost()) {\n    delete contexts[webGLVersion];\n    return getWebGLContext(webGLVersion);\n  }\n\n  gl.disable(gl.DEPTH_TEST);\n  gl.disable(gl.STENCIL_TEST);\n  gl.disable(gl.BLEND);\n  gl.disable(gl.DITHER);\n  gl.disable(gl.POLYGON_OFFSET_FILL);\n  gl.disable(gl.SAMPLE_COVERAGE);\n  gl.enable(gl.SCISSOR_TEST);\n  gl.enable(gl.CULL_FACE);\n  gl.cullFace(gl.BACK);\n\n  return contexts[webGLVersion];\n}\n\nfunction createCanvas(webGLVersion: number) {\n  if (typeof OffscreenCanvas !== 'undefined' && webGLVersion === 2) {\n    return new OffscreenCanvas(300, 150);\n  } else if (typeof document !== 'undefined') {\n    return document.createElement('canvas');\n  } else {\n    throw new Error('Cannot create a canvas in this context');\n  }\n}\n\nfunction getWebGLRenderingContext(webGLVersion: number): WebGLRenderingContext {\n  if (webGLVersion !== 1 && webGLVersion !== 2) {\n    throw new Error('Cannot get WebGL rendering context, WebGL is disabled.');\n  }\n  const canvas = createCanvas(webGLVersion);\n\n  canvas.addEventListener('webglcontextlost', (ev: Event) => {\n    ev.preventDefault();\n    delete contexts[webGLVersion];\n  }, false);\n  if (webGLVersion === 1) {\n    return (canvas.getContext('webgl', WEBGL_ATTRIBUTES) ||\n            canvas.getContext('experimental-webgl', WEBGL_ATTRIBUTES)) as\n        WebGLRenderingContext;\n  }\n  return canvas.getContext('webgl2', WEBGL_ATTRIBUTES) as WebGLRenderingContext;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\nimport {backend_util, DataId, DataType, Tensor, util} from '@tensorflow/tfjs-core';\n\nexport enum PackingScheme {\n  /**\n   * All values in a single texel are densely packed without any constraints.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 4]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   010|011   020|021\n   * -------   -------   -------\n   * 002|003   012|013   022|023\n   *\n   * 100|101   110|111   120|121\n   * -------   -------   -------\n   * 102|103   112|113   122|123\n   *\n   */\n  DENSE,\n\n  /**\n   * Single texels contain only values from the same batch, and from adjacent\n   * rows and columns.\n   *\n   * This is how the shader encodes a tensor with shape = [2, 3, 5]\n   * (indices are [batch, row, col]).\n   *\n   * 000|001   002|003   004|xxx   020|021   022|023   024|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   * 100|101   102|103   104|xxx   120|121   122|123   124|xxx\n   * -------   -------   -------   -------   -------   -------\n   * 110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n   *\n   */\n  SHARED_BATCH\n}\n\nexport enum TextureUsage {\n  RENDER,\n  UPLOAD,\n  PIXELS,\n  DOWNLOAD\n}\n\nexport enum PhysicalTextureType {\n  UNPACKED_FLOAT16,\n  UNPACKED_FLOAT32,\n  PACKED_4X1_UNSIGNED_BYTE,\n  PACKED_2X2_FLOAT32,\n  PACKED_2X2_FLOAT16\n}\n\nexport interface TextureData {\n  // Required.\n  shape: number[];\n  dtype: DataType;\n\n  // Optional.\n  values?: backend_util.BackendValues;\n  texture?: WebGLTexture;\n  // For complex numbers, the real and imaginary parts are stored as their own\n  // individual tensors, with a parent joining the two with the\n  // complexTensors field. When this is defined, texture will be null.\n  complexTensors?: {real: Tensor, imag: Tensor};\n  /** [rows, columns] shape of the texture. */\n  texShape?: [number, number];\n  usage?: TextureUsage;\n  isPacked?: boolean;\n\n  refCount: number;\n\n  // Available when the tensor has been sliced.\n  slice?: {\n    // Offset in the 'flat index' space.\n    flatOffset: number;\n    // Used for counting how many sliced tensors point to the same texture.\n    origDataId: DataId;\n  };\n}\n\nexport function getUnpackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns, rows];\n}\n\nexport function getUnpackedArraySizeFromMatrixSize(\n    matrixSize: number, channelsPerTexture: number): number {\n  return matrixSize * channelsPerTexture;\n}\n\nexport function getColorMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [columns * 4, rows];\n}\n\n/**\n * Get shape for densely packed RGBA texture.\n */\nexport function getDenseTexShape(shape: number[]): [number, number] {\n  const size = util.sizeFromShape(shape);\n  const texelsNeeded = Math.ceil(size / 4);\n  return util.sizeToSquarishShape(texelsNeeded);\n}\n\nexport function getMatrixSizeFromUnpackedArraySize(\n    unpackedSize: number, channelsPerTexture: number): number {\n  if (unpackedSize % channelsPerTexture !== 0) {\n    throw new Error(\n        `unpackedSize (${unpackedSize}) must be a multiple of ` +\n        `${channelsPerTexture}`);\n  }\n  return unpackedSize / channelsPerTexture;\n}\n\nexport function decodeMatrixFromUnpackedColorRGBAArray(\n    unpackedArray: Float32Array, matrix: Float32Array, channels: number) {\n  const requiredSize = unpackedArray.length * channels / 4;\n  if (matrix.length < requiredSize) {\n    throw new Error(\n        `matrix length (${matrix.length}) must be >= ${requiredSize}`);\n  }\n  let dst = 0;\n  for (let src = 0; src < unpackedArray.length; src += 4) {\n    for (let c = 0; c < channels; c++) {\n      matrix[dst++] = unpackedArray[src + c];\n    }\n  }\n}\n\nexport function getPackedMatrixTextureShapeWidthHeight(\n    rows: number, columns: number): [number, number] {\n  return [\n    Math.max(1, Math.ceil(columns / 2)), Math.max(1, Math.ceil(rows / 2))\n  ];\n}\n\nexport function getPackedRGBAArraySizeFromMatrixShape(\n    rows: number, columns: number): number {\n  const [w, h] = getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return w * h * 4;\n}\n\nexport interface TextureConfig {\n  internalFormatFloat: number;\n  textureFormatFloat: number;\n  internalFormatPackedHalfFloat: number;\n  internalFormatHalfFloat: number;\n  internalFormatPackedFloat: number;\n\n  // The format to use during a gl.readPixels call.\n  downloadTextureFormat: number;\n  // How many channels need to be unpacked after a gl.readPixels call.\n  downloadUnpackNumChannels: number;\n\n  defaultNumChannels: number;\n  textureTypeHalfFloat: number;\n  textureTypeFloat: number;\n}\n\nexport function getTextureConfig(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension?: any): TextureConfig {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n\n  let internalFormatFloat: number;\n  let internalFormatHalfFloat: number;\n  let internalFormatPackedHalfFloat: number;\n  let internalFormatPackedFloat: number;\n  let textureFormatFloat: number;\n\n  let downloadTextureFormat: number;\n  let downloadUnpackNumChannels: number;\n\n  let defaultNumChannels: number;\n  let textureTypeHalfFloat: number;\n  let textureTypeFloat: number;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    internalFormatFloat = glany.R32F;\n    internalFormatHalfFloat = glany.R16F;\n    internalFormatPackedHalfFloat = glany.RGBA16F;\n    internalFormatPackedFloat = glany.RGBA32F;\n    textureFormatFloat = glany.RED;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 1;\n    textureTypeHalfFloat = glany.HALF_FLOAT;\n    textureTypeFloat = glany.FLOAT;\n  } else {\n    internalFormatFloat = gl.RGBA;\n    internalFormatHalfFloat = gl.RGBA;\n    internalFormatPackedHalfFloat = gl.RGBA;\n    internalFormatPackedFloat = glany.RGBA;\n    textureFormatFloat = gl.RGBA;\n    downloadUnpackNumChannels = 4;\n    defaultNumChannels = 4;\n    textureTypeHalfFloat = textureHalfFloatExtension != null ?\n        textureHalfFloatExtension.HALF_FLOAT_OES :\n        null;\n    textureTypeFloat = gl.FLOAT;\n  }\n  downloadTextureFormat = gl.RGBA;\n\n  return {\n    internalFormatFloat,\n    internalFormatHalfFloat,\n    internalFormatPackedHalfFloat,\n    internalFormatPackedFloat,\n    textureFormatFloat,\n    downloadTextureFormat,\n    downloadUnpackNumChannels,\n    defaultNumChannels,\n    textureTypeHalfFloat,\n    textureTypeFloat\n  };\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext} from './canvas_util';\nimport {getTextureConfig} from './tex_util';\n\nexport function callAndCheck<T>(gl: WebGLRenderingContext, func: () => T): T {\n  const returnValue = func();\n  if (env().getBool('DEBUG')) {\n    checkWebGLError(gl);\n  }\n  return returnValue;\n}\n\nfunction checkWebGLError(gl: WebGLRenderingContext) {\n  const error = gl.getError();\n  if (error !== gl.NO_ERROR) {\n    throw new Error('WebGL Error: ' + getWebGLErrorMessage(gl, error));\n  }\n}\n\n// https://en.wikipedia.org/wiki/Half-precision_floating-point_format\nconst MIN_FLOAT16 = 5.96e-8;\nconst MAX_FLOAT16 = 65504;\n\nexport function canBeRepresented(num: number): boolean {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED') || num === 0 ||\n      (MIN_FLOAT16 < Math.abs(num) && Math.abs(num) < MAX_FLOAT16)) {\n    return true;\n  }\n  return false;\n}\n\nexport function getWebGLErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.NO_ERROR:\n      return 'NO_ERROR';\n    case gl.INVALID_ENUM:\n      return 'INVALID_ENUM';\n    case gl.INVALID_VALUE:\n      return 'INVALID_VALUE';\n    case gl.INVALID_OPERATION:\n      return 'INVALID_OPERATION';\n    case gl.INVALID_FRAMEBUFFER_OPERATION:\n      return 'INVALID_FRAMEBUFFER_OPERATION';\n    case gl.OUT_OF_MEMORY:\n      return 'OUT_OF_MEMORY';\n    case gl.CONTEXT_LOST_WEBGL:\n      return 'CONTEXT_LOST_WEBGL';\n    default:\n      return `Unknown error code ${status}`;\n  }\n}\n\nexport function getExtensionOrThrow(\n    gl: WebGLRenderingContext, extensionName: string): {} {\n  return throwIfNull<{}>(\n      gl, () => gl.getExtension(extensionName),\n      'Extension \"' + extensionName + '\" not supported on this browser.');\n}\n\nexport function createVertexShader(\n    gl: WebGLRenderingContext, vertexShaderSource: string): WebGLShader {\n  const vertexShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.VERTEX_SHADER),\n      'Unable to create vertex WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(vertexShader, vertexShaderSource));\n  callAndCheck(gl, () => gl.compileShader(vertexShader));\n  if (gl.getShaderParameter(vertexShader, gl.COMPILE_STATUS) === false) {\n    console.log(gl.getShaderInfoLog(vertexShader));\n    throw new Error('Failed to compile vertex shader.');\n  }\n  return vertexShader;\n}\n\nexport function createFragmentShader(\n    gl: WebGLRenderingContext, fragmentShaderSource: string): WebGLShader {\n  const fragmentShader: WebGLShader = throwIfNull<WebGLShader>(\n      gl, () => gl.createShader(gl.FRAGMENT_SHADER),\n      'Unable to create fragment WebGLShader.');\n  callAndCheck(gl, () => gl.shaderSource(fragmentShader, fragmentShaderSource));\n  callAndCheck(gl, () => gl.compileShader(fragmentShader));\n  if (gl.getShaderParameter(fragmentShader, gl.COMPILE_STATUS) === false) {\n    logShaderSourceAndInfoLog(\n        fragmentShaderSource, gl.getShaderInfoLog(fragmentShader));\n    throw new Error('Failed to compile fragment shader.');\n  }\n  return fragmentShader;\n}\n\nconst lineNumberRegex = /ERROR: [0-9]+:([0-9]+):/g;\nfunction logShaderSourceAndInfoLog(\n    shaderSource: string, shaderInfoLog: string) {\n  const lineNumberRegexResult = lineNumberRegex.exec(shaderInfoLog);\n  if (lineNumberRegexResult == null) {\n    console.log(`Couldn't parse line number in error: ${shaderInfoLog}`);\n    console.log(shaderSource);\n    return;\n  }\n\n  const lineNumber = +lineNumberRegexResult[1];\n\n  const shaderLines = shaderSource.split('\\n');\n  const pad = shaderLines.length.toString().length + 2;\n  const linesWithLineNumbers = shaderLines.map(\n      (line, lineNumber) =>\n          util.rightPad((lineNumber + 1).toString(), pad) + line);\n  let maxLineLength = 0;\n  for (let i = 0; i < linesWithLineNumbers.length; i++) {\n    maxLineLength = Math.max(linesWithLineNumbers[i].length, maxLineLength);\n  }\n\n  const beforeErrorLines = linesWithLineNumbers.slice(0, lineNumber - 1);\n  const errorLine = linesWithLineNumbers.slice(lineNumber - 1, lineNumber);\n  const afterErrorLines = linesWithLineNumbers.slice(lineNumber);\n\n  console.log(beforeErrorLines.join('\\n'));\n  console.log(shaderInfoLog.split('\\n')[0]);\n  console.log(\n      `%c ${util.rightPad(errorLine[0], maxLineLength)}`,\n      'border:1px solid red; background-color:#e3d2d2; color:#a61717');\n  console.log(afterErrorLines.join('\\n'));\n}\n\nexport function createProgram(gl: WebGLRenderingContext): WebGLProgram {\n  return throwIfNull<WebGLProgram>(\n      gl, () => gl.createProgram(), 'Unable to create WebGLProgram.');\n}\n\nexport function linkProgram(gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.linkProgram(program));\n  if (gl.getProgramParameter(program, gl.LINK_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Failed to link vertex and fragment shaders.');\n  }\n}\n\nexport function validateProgram(\n    gl: WebGLRenderingContext, program: WebGLProgram) {\n  callAndCheck(gl, () => gl.validateProgram(program));\n  if (gl.getProgramParameter(program, gl.VALIDATE_STATUS) === false) {\n    console.log(gl.getProgramInfoLog(program));\n    throw new Error('Shader program validation failed.');\n  }\n}\n\nexport function createStaticVertexBuffer(\n    gl: WebGLRenderingContext, data: Float32Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(gl, () => gl.bufferData(gl.ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function createStaticIndexBuffer(\n    gl: WebGLRenderingContext, data: Uint16Array): WebGLBuffer {\n  const buffer: WebGLBuffer = throwIfNull<WebGLBuffer>(\n      gl, () => gl.createBuffer(), 'Unable to create WebGLBuffer');\n  callAndCheck(gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl, () => gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, data, gl.STATIC_DRAW));\n  return buffer;\n}\n\nexport function getNumChannels(): number {\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    return 1;\n  }\n  return 4;\n}\n\nexport function createTexture(gl: WebGLRenderingContext): WebGLTexture {\n  return throwIfNull<WebGLTexture>(\n      gl, () => gl.createTexture(), 'Unable to create WebGLTexture.');\n}\n\nexport function validateTextureSize(width: number, height: number) {\n  const maxTextureSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if ((width <= 0) || (height <= 0)) {\n    const requested = `[${width}x${height}]`;\n    throw new Error('Requested texture size ' + requested + ' is invalid.');\n  }\n  if ((width > maxTextureSize) || (height > maxTextureSize)) {\n    const requested = `[${width}x${height}]`;\n    const max = `[${maxTextureSize}x${maxTextureSize}]`;\n    throw new Error(\n        'Requested texture size ' + requested +\n        ' greater than WebGL maximum on this browser / GPU ' + max + '.');\n  }\n}\n\nexport function createFramebuffer(gl: WebGLRenderingContext): WebGLFramebuffer {\n  return throwIfNull<WebGLFramebuffer>(\n      gl, () => gl.createFramebuffer(), 'Unable to create WebGLFramebuffer.');\n}\n\nexport function bindVertexBufferToProgramAttribute(\n    gl: WebGLRenderingContext, program: WebGLProgram, attribute: string,\n    buffer: WebGLBuffer, arrayEntriesPerItem: number, itemStrideInBytes: number,\n    itemOffsetInBytes: number): boolean {\n  const loc = gl.getAttribLocation(program, attribute);\n  if (loc === -1) {\n    // The GPU compiler decided to strip out this attribute because it's unused,\n    // thus no need to bind.\n    return false;\n  }\n  callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, buffer));\n  callAndCheck(\n      gl,\n      () => gl.vertexAttribPointer(\n          loc, arrayEntriesPerItem, gl.FLOAT, false, itemStrideInBytes,\n          itemOffsetInBytes));\n  callAndCheck(gl, () => gl.enableVertexAttribArray(loc));\n  return true;\n}\n\nexport function bindTextureUnit(\n    gl: WebGLRenderingContext, texture: WebGLTexture, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n}\n\nexport function unbindTextureUnit(\n    gl: WebGLRenderingContext, textureUnit: number) {\n  validateTextureUnit(gl, textureUnit);\n  callAndCheck(gl, () => gl.activeTexture(gl.TEXTURE0 + textureUnit));\n  callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function getProgramUniformLocationOrThrow(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return throwIfNull<WebGLUniformLocation>(\n      gl, () => gl.getUniformLocation(program, uniformName),\n      'uniform \"' + uniformName + '\" not present in program.');\n}\n\nexport function getProgramUniformLocation(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    uniformName: string): WebGLUniformLocation {\n  return gl.getUniformLocation(program, uniformName);\n}\n\nexport function bindTextureToProgramUniformSampler(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    uniformSamplerLocation: WebGLUniformLocation, textureUnit: number) {\n  callAndCheck(gl, () => bindTextureUnit(gl, texture, textureUnit));\n  callAndCheck(gl, () => gl.uniform1i(uniformSamplerLocation, textureUnit));\n}\n\nexport function bindCanvasToFramebuffer(gl: WebGLRenderingContext) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n  callAndCheck(gl, () => gl.viewport(0, 0, gl.canvas.width, gl.canvas.height));\n  callAndCheck(gl, () => gl.scissor(0, 0, gl.canvas.width, gl.canvas.height));\n}\n\nexport function bindColorTextureToFramebuffer(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0));\n}\n\nexport function unbindColorTextureFromFramebuffer(\n    gl: WebGLRenderingContext, framebuffer: WebGLFramebuffer) {\n  callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, framebuffer));\n  callAndCheck(\n      gl,\n      () => gl.framebufferTexture2D(\n          gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, null, 0));\n}\n\nexport function validateFramebuffer(gl: WebGLRenderingContext) {\n  const status = gl.checkFramebufferStatus(gl.FRAMEBUFFER);\n  if (status !== gl.FRAMEBUFFER_COMPLETE) {\n    throw new Error(\n        'Error binding framebuffer: ' + getFramebufferErrorMessage(gl, status));\n  }\n}\n\nexport function getFramebufferErrorMessage(\n    gl: WebGLRenderingContext, status: number): string {\n  switch (status) {\n    case gl.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:\n      return 'FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT';\n    case gl.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:\n      return 'FRAMEBUFFER_INCOMPLETE_DIMENSIONS';\n    case gl.FRAMEBUFFER_UNSUPPORTED:\n      return 'FRAMEBUFFER_UNSUPPORTED';\n    default:\n      return `unknown error ${status}`;\n  }\n}\n\nfunction throwIfNull<T>(\n    gl: WebGLRenderingContext, returnTOrNull: () => T | null,\n    failureMessage: string): T {\n  const tOrNull: T|null = callAndCheck(gl, () => returnTOrNull());\n  if (tOrNull == null) {\n    throw new Error(failureMessage);\n  }\n  return tOrNull;\n}\n\nfunction validateTextureUnit(gl: WebGLRenderingContext, textureUnit: number) {\n  const maxTextureUnit = gl.MAX_COMBINED_TEXTURE_IMAGE_UNITS - 1;\n  const glTextureUnit = textureUnit + gl.TEXTURE0;\n  if (glTextureUnit < gl.TEXTURE0 || glTextureUnit > maxTextureUnit) {\n    const textureUnitRange = `[gl.TEXTURE0, gl.TEXTURE${maxTextureUnit}]`;\n    throw new Error(`textureUnit must be in ${textureUnitRange}.`);\n  }\n}\n\nexport function getBatchDim(shape: number[], dimsToSkip = 2): number {\n  return util.sizeFromShape(shape.slice(0, shape.length - dimsToSkip));\n}\n\nexport function getRowsCols(shape: number[]): [number, number] {\n  if (shape.length === 0) {\n    throw Error('Cannot get rows and columns of an empty shape array.');\n  }\n\n  return [\n    shape.length > 1 ? shape[shape.length - 2] : 1, shape[shape.length - 1]\n  ];\n}\n\nexport function getShapeAs3D(shape: number[]): [number, number, number] {\n  let shapeAs3D: [number, number, number] = [1, 1, 1];\n  const isScalar = shape.length === 0 || (shape.length === 1 && shape[0] === 1);\n  if (!isScalar) {\n    shapeAs3D =\n        [getBatchDim(shape), ...getRowsCols(shape)] as [number, number, number];\n  }\n  return shapeAs3D;\n}\n\nexport function getTextureShapeFromLogicalShape(\n    logShape: number[], isPacked = false): [number, number] {\n  let maxTexSize = env().getNumber('WEBGL_MAX_TEXTURE_SIZE');\n  if (isPacked) {\n    maxTexSize = maxTexSize * 2;\n\n    // This logic ensures we accurately count the number of packed texels needed\n    // to accommodate the tensor. We can only pack values in the same texel if\n    // they are from adjacent pairs of rows/cols within the same batch. So if a\n    // tensor has 3 rows, we pretend it has 4 rows in order to account for the\n    // fact that the texels containing the third row are half empty.\n    logShape = logShape.map(\n        (d, i) => i >= logShape.length - 2 ?\n            util.nearestLargerEven(logShape[i]) :\n            logShape[i]);\n\n    // Packed texture height is at least 2 (the channel height of a single\n    // texel).\n    if (logShape.length === 1) {\n      logShape = [2, logShape[0]];\n    }\n  }\n\n  // If logical shape is 2, we don't squeeze, since we want to match physical.\n  if (logShape.length !== 2) {\n    const squeezeResult = util.squeezeShape(logShape);\n    logShape = squeezeResult.newShape;\n  }\n\n  let size = util.sizeFromShape(logShape);\n  if (logShape.length <= 1 && size <= maxTexSize) {\n    return [1, size];\n  } else if (\n      logShape.length === 2 && logShape[0] <= maxTexSize &&\n      logShape[1] <= maxTexSize) {\n    return logShape as [number, number];\n  } else if (\n      logShape.length === 3 && logShape[0] * logShape[1] <= maxTexSize &&\n      logShape[2] <= maxTexSize) {\n    return [logShape[0] * logShape[1], logShape[2]];\n  } else if (\n      logShape.length === 3 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2]];\n  } else if (\n      logShape.length === 4 &&\n      logShape[0] * logShape[1] * logShape[2] <= maxTexSize &&\n      logShape[3] <= maxTexSize) {\n    return [logShape[0] * logShape[1] * logShape[2], logShape[3]];\n  } else if (\n      logShape.length === 4 && logShape[0] <= maxTexSize &&\n      logShape[1] * logShape[2] * logShape[3] <= maxTexSize) {\n    return [logShape[0], logShape[1] * logShape[2] * logShape[3]];\n  } else {\n    if (isPacked) {\n      // For packed textures size equals the number of channels required to\n      // accommodate the texture data. However in order to squarify such that\n      // inner dimensions stay even, we rewrite size to equal the number of\n      // texels. Then in the return statement we rehydrate the squarified\n      // dimensions to channel units.\n\n      const batchDim = getBatchDim(logShape);\n      let rows = 2, cols = 2;\n      if (logShape.length) {\n        [rows, cols] = getRowsCols(logShape);\n      }\n      size = batchDim * (rows / 2) * (cols / 2);\n      return util.sizeToSquarishShape(size).map(d => d * 2) as [number, number];\n    }\n    return util.sizeToSquarishShape(size);\n  }\n}\n\nfunction isEven(n: number): boolean {\n  return n % 2 === 0;\n}\n\n/**\n * This determines whether reshaping a packed texture requires rearranging\n * the data within the texture, assuming 2x2 packing.\n */\nexport function isReshapeFree(shape1: number[], shape2: number[]): boolean {\n  shape1 = shape1.slice(-2);\n  shape2 = shape2.slice(-2);\n\n  if (util.arraysEqual(shape1, shape2)) {\n    return true;\n  }\n\n  if (!shape1.length || !shape2.length) {  // One of the shapes is a scalar.\n    return true;\n  }\n\n  if (shape1[0] === 0 || shape1[1] === 0 || shape2[0] === 0 ||\n      shape2[1] === 0) {\n    return true;\n  }\n\n  if (shape1.length !== shape2.length) {  // One of the shapes is a vector.\n    const shape1Cols = shape1.slice(-1)[0];\n    const shape2Cols = shape2.slice(-1)[0];\n    if (shape1Cols === shape2Cols) {\n      return true;\n    }\n\n    if (isEven(shape1Cols) && isEven(shape2Cols) &&\n        (shape1[0] === 1 || shape2[0] === 1)) {\n      return true;\n    }\n  }\n  return shape1[1] === shape2[1] && isEven(shape1[0]) && isEven(shape2[0]);\n}\n\n// We cache webgl params because the environment gets reset between\n// unit tests and we don't want to constantly query the WebGLContext for\n// MAX_TEXTURE_SIZE.\nlet MAX_TEXTURE_SIZE: number;\nlet MAX_TEXTURES_IN_SHADER: number;\n\nexport function getWebGLMaxTextureSize(webGLVersion: number): number {\n  if (MAX_TEXTURE_SIZE == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURE_SIZE = gl.getParameter(gl.MAX_TEXTURE_SIZE);\n  }\n  return MAX_TEXTURE_SIZE;\n}\n\nexport function resetMaxTextureSize() {\n  MAX_TEXTURE_SIZE = null;\n}\nexport function resetMaxTexturesInShader() {\n  MAX_TEXTURES_IN_SHADER = null;\n}\n\nexport function getMaxTexturesInShader(webGLVersion: number): number {\n  if (MAX_TEXTURES_IN_SHADER == null) {\n    const gl = getWebGLContext(webGLVersion);\n    MAX_TEXTURES_IN_SHADER = gl.getParameter(gl.MAX_TEXTURE_IMAGE_UNITS);\n  }\n  // We cap at 16 to avoid spurious runtime \"memory exhausted\" error.\n  return Math.min(16, MAX_TEXTURES_IN_SHADER);\n}\n\nexport function getWebGLDisjointQueryTimerVersion(webGLVersion: number):\n    number {\n  if (webGLVersion === 0) {\n    return 0;\n  }\n\n  let queryTimerVersion: number;\n  const gl = getWebGLContext(webGLVersion);\n\n  if (hasExtension(gl, 'EXT_disjoint_timer_query_webgl2') &&\n      webGLVersion === 2) {\n    queryTimerVersion = 2;\n  } else if (hasExtension(gl, 'EXT_disjoint_timer_query')) {\n    queryTimerVersion = 1;\n  } else {\n    queryTimerVersion = 0;\n  }\n  return queryTimerVersion;\n}\n\nexport function hasExtension(gl: WebGLRenderingContext, extensionName: string) {\n  const ext = gl.getExtension(extensionName);\n  return ext != null;\n}\n\nexport function isWebGLVersionEnabled(webGLVersion: 1|2) {\n  try {\n    const gl = getWebGLContext(webGLVersion);\n    if (gl != null) {\n      return true;\n    }\n  } catch (e) {\n    console.log('Error when getting WebGL context: ', e);\n    return false;\n  }\n  return false;\n}\n\nexport function isCapableOfRenderingToFloatTexture(webGLVersion: number):\n    boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n  } else {\n    if (!hasExtension(gl, 'EXT_color_buffer_float')) {\n      return false;\n    }\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\n/**\n * Check if we can download values from a float/half-float texture.\n *\n * Note that for performance reasons we use binding a texture to a framebuffer\n * as a proxy for ability to download float values later using readPixels. The\n * texture params of this texture will not match those in readPixels exactly\n * but if we are unable to bind some kind of float texture to the frameBuffer\n * then we definitely will not be able to read float values from it.\n */\nexport function isDownloadFloatTextureEnabled(webGLVersion: number): boolean {\n  if (webGLVersion === 0) {\n    return false;\n  }\n\n  const gl = getWebGLContext(webGLVersion);\n\n  if (webGLVersion === 1) {\n    if (!hasExtension(gl, 'OES_texture_float')) {\n      return false;\n    }\n    if (!hasExtension(gl, 'WEBGL_color_buffer_float')) {\n      return false;\n    }\n  } else {\n    if (hasExtension(gl, 'EXT_color_buffer_float')) {\n      return createFloatTextureAndBindToFramebuffer(gl);\n    }\n\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (hasExtension(gl, COLOR_BUFFER_HALF_FLOAT)) {\n      const textureHalfFloatExtension =\n          gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      return createHalfFloatTextureAndBindToFramebuffer(\n          gl, textureHalfFloatExtension);\n    }\n\n    return false;\n  }\n\n  const isFrameBufferComplete = createFloatTextureAndBindToFramebuffer(gl);\n  return isFrameBufferComplete;\n}\n\nfunction createFloatTextureAndBindToFramebuffer(gl: WebGLRenderingContext):\n    boolean {\n  const texConfig = getTextureConfig(gl);\n\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nfunction createHalfFloatTextureAndBindToFramebuffer(\n    // tslint:disable-next-line:no-any\n    gl: WebGLRenderingContext, textureHalfFloatExtension: any): boolean {\n  const texConfig = getTextureConfig(gl, textureHalfFloatExtension);\n  const texture = gl.createTexture();\n  gl.bindTexture(gl.TEXTURE_2D, texture);\n\n  const width = 1;\n  const height = 1;\n  gl.texImage2D(\n      gl.TEXTURE_2D, 0, texConfig.internalFormatHalfFloat, width, height, 0,\n      texConfig.textureFormatFloat, texConfig.textureTypeHalfFloat, null);\n\n  const frameBuffer = gl.createFramebuffer();\n  gl.bindFramebuffer(gl.FRAMEBUFFER, frameBuffer);\n  gl.framebufferTexture2D(\n      gl.FRAMEBUFFER, gl.COLOR_ATTACHMENT0, gl.TEXTURE_2D, texture, 0);\n\n  const isFrameBufferComplete =\n      gl.checkFramebufferStatus(gl.FRAMEBUFFER) === gl.FRAMEBUFFER_COMPLETE;\n\n  gl.bindTexture(gl.TEXTURE_2D, null);\n  gl.bindFramebuffer(gl.FRAMEBUFFER, null);\n  gl.deleteTexture(texture);\n  gl.deleteFramebuffer(frameBuffer);\n\n  return isFrameBufferComplete;\n}\n\nexport function isWebGLFenceEnabled(webGLVersion: number) {\n  if (webGLVersion !== 2) {\n    return false;\n  }\n  const gl = getWebGLContext(webGLVersion);\n\n  // tslint:disable-next-line:no-any\n  const isEnabled = (gl as any).fenceSync != null;\n  return isEnabled;\n}\n\nexport function assertNotComplex(\n    tensor: TensorInfo|TensorInfo[], opName: string): void {\n  if (!Array.isArray(tensor)) {\n    tensor = [tensor];\n  }\n  tensor.forEach(t => {\n    if (t != null) {\n      util.assert(\n          t.dtype !== 'complex64',\n          () => `${opName} does not support complex64 tensors ` +\n              'in the WebGL backend.');\n    }\n  });\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {device_util, env} from '@tensorflow/tfjs-core';\n\nimport {getMaxTexturesInShader, getWebGLDisjointQueryTimerVersion, getWebGLMaxTextureSize, isCapableOfRenderingToFloatTexture, isDownloadFloatTextureEnabled, isWebGLFenceEnabled, isWebGLVersionEnabled} from './webgl_util';\n\nconst ENV = env();\n\n/**\n * This file contains WebGL-specific flag registrations.\n */\n\n/**\n * True if WebGL is supported.\n */\nENV.registerFlag('HAS_WEBGL', () => ENV.getNumber('WEBGL_VERSION') > 0);\n\n/** 0: No WebGL, 1: WebGL 1.0, 2: WebGL 2.0. */\nENV.registerFlag('WEBGL_VERSION', () => {\n  if (isWebGLVersionEnabled(2)) {\n    return 2;\n  } else if (isWebGLVersionEnabled(1)) {\n    return 1;\n  }\n  return 0;\n});\n\n/** Whether to check for numerical representation problems. */\nENV.registerFlag('WEBGL_CHECK_NUMERICAL_PROBLEMS', () => false);\n\nENV.registerFlag(\n    'WEBGL_BUFFER_SUPPORTED', () => ENV.get('WEBGL_VERSION') === 2);\n\n/** Whether the WebGL backend will sometimes forward ops to the CPU. */\nENV.registerFlag('WEBGL_CPU_FORWARD', () => true);\n\n/** Whether the WebGL backend will always use f16 textures for rendering. */\nENV.registerFlag('WEBGL_FORCE_F16_TEXTURES', () => false);\n\n/** Whether to turn all packing related flags on. */\nENV.registerFlag('WEBGL_PACK', () => ENV.getBool('HAS_WEBGL'));\n\n/** Whether we will pack the batchnormalization op. */\nENV.registerFlag('WEBGL_PACK_NORMALIZATION', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the clip op. */\nENV.registerFlag('WEBGL_PACK_CLIP', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack the depthwise conv op. */\n// TODO: https://github.com/tensorflow/tfjs/issues/1679\nENV.registerFlag('WEBGL_PACK_DEPTHWISECONV', () => false);\n\n/** Whether we will pack binary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_BINARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack unary ops. */\nENV.registerFlag(\n    'WEBGL_PACK_UNARY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack array ops. */\nENV.registerFlag(\n    'WEBGL_PACK_ARRAY_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack image ops. */\nENV.registerFlag(\n    'WEBGL_PACK_IMAGE_OPERATIONS', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will pack reduce ops. */\nENV.registerFlag('WEBGL_PACK_REDUCE', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether packed WebGL kernels lazily unpack their outputs. */\nENV.registerFlag('WEBGL_LAZILY_UNPACK', () => ENV.getBool('WEBGL_PACK'));\n\n/** Whether we will use the im2col algorithm to speed up convolutions. */\nENV.registerFlag('WEBGL_CONV_IM2COL', () => ENV.getBool('WEBGL_PACK'));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURE_SIZE',\n    () => getWebGLMaxTextureSize(ENV.getNumber('WEBGL_VERSION')));\n\n/** The maximum texture dimension. */\nENV.registerFlag(\n    'WEBGL_MAX_TEXTURES_IN_SHADER',\n    () => getMaxTexturesInShader(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * The disjoint_query_timer extension version.\n * 0: disabled, 1: EXT_disjoint_timer_query, 2:\n * EXT_disjoint_timer_query_webgl2.\n * In Firefox with WebGL 2.0,\n * EXT_disjoint_timer_query_webgl2 is not available, so we must use the\n * WebGL 1.0 extension.\n */\nENV.registerFlag('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION', () => {\n  const webGLVersion = ENV.getNumber('WEBGL_VERSION');\n\n  if (webGLVersion === 0) {\n    return 0;\n  }\n  return getWebGLDisjointQueryTimerVersion(webGLVersion);\n});\n\n/**\n * Whether the timer object from the disjoint_query_timer extension gives\n * timing information that is reliable.\n */\nENV.registerFlag(\n    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE',\n    () => ENV.getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0 &&\n        !device_util.isMobile());\n\n/**\n * Whether the device is physically capable of rendering to float32 textures.\n */\nENV.registerFlag(\n    'WEBGL_RENDER_FLOAT32_CAPABLE',\n    () => isCapableOfRenderingToFloatTexture(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Whether rendering to float32 textures is enabled. If disabled, renders to\n * float16 textures.\n */\nENV.registerFlag('WEBGL_RENDER_FLOAT32_ENABLED', () => {\n  return ENV.getBool('WEBGL_FORCE_F16_TEXTURES') ?\n      false :\n      ENV.getBool('WEBGL_RENDER_FLOAT32_CAPABLE');\n});\n\n/**\n * Whether downloading float textures is enabled (16 or 32 bit). If disabled,\n * uses IEEE 754 encoding of the float32 values to 4 uint8 when downloading.\n */\nENV.registerFlag(\n    'WEBGL_DOWNLOAD_FLOAT_ENABLED',\n    () => isDownloadFloatTextureEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/** Whether the fence API is available. */\nENV.registerFlag(\n    'WEBGL_FENCE_API_ENABLED',\n    () => isWebGLFenceEnabled(ENV.getNumber('WEBGL_VERSION')));\n\n/**\n * Tensors with size <= than this will be uploaded as uniforms, not textures.\n */\nENV.registerFlag('WEBGL_SIZE_UPLOAD_UNIFORM', () => {\n  // Use uniform uploads only when 32bit floats are supported. In\n  // 16bit\n  // environments there are problems with comparing a 16bit texture value\n  // with a 32bit uniform value.\n  const useUniforms = ENV.getBool('WEBGL_RENDER_FLOAT32_ENABLED');\n  return useUniforms ? 4 : 0;\n});\n\n/**\n * If the total number of bytes allocated on the GPU is greater than this\n * number, we will aggressively delete textures upon disposal with\n * gl.deleteMatrixTexture, rather than making them available for reuse.\n *\n * Default value -1 indicates that we will never aggressively delete textures.\n */\nENV.registerFlag(\n    'WEBGL_DELETE_TEXTURE_THRESHOLD',\n    () => {\n      return -1;\n    },\n    threshold => {\n      if (threshold < 0 && threshold !== -1) {\n        throw new Error(\n            `WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never ` +\n            `delete) or at least 0, but got ${threshold}.`);\n      }\n    });\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Abs, util } from '@tensorflow/tfjs-core';\nexport function simpleAbsImpl(vals) {\n    const resultValues = new Float32Array(vals.length);\n    for (let i = 0; i < vals.length; ++i) {\n        resultValues[i] = Math.abs(vals[i]);\n    }\n    return resultValues;\n}\nexport const absKernelFunc = (args) => {\n    const { x } = args.inputs;\n    const cpuBackend = args.backend;\n    let resultValues = new Float32Array(util.sizeFromShape(x.shape));\n    if (x.dtype !== 'complex64') {\n        const values = cpuBackend.data.get(x.dataId).values;\n        resultValues = simpleAbsImpl(values);\n    }\n    else {\n        const complexVals = cpuBackend.data.get(x.dataId);\n        const real = complexVals.complexTensorInfos.real;\n        const imag = complexVals.complexTensorInfos.imag;\n        const realVals = cpuBackend.data.get(real.dataId).values;\n        const imagVals = cpuBackend.data.get(imag.dataId).values;\n        for (let i = 0; i < realVals.length; i++) {\n            const real = realVals[i];\n            const imag = imagVals[i];\n            resultValues[i] = Math.hypot(real, imag);\n        }\n    }\n    return cpuBackend.makeOutput(resultValues, x.shape, 'float32');\n};\nexport const absConfig = {\n    kernelName: Abs,\n    backendName: 'cpu',\n    kernelFunc: absKernelFunc,\n};\n//# sourceMappingURL=Abs.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { backend_util, util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for binary ops. Supports broadcast.\n */\nexport function createSimpleBinaryKernelImpl(op) {\n    return (aShape, bShape, aVals, bVals, dtype) => {\n        const newShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n        const resultRank = newShape.length;\n        const resultStrides = util.computeStrides(newShape);\n        const resultSize = util.sizeFromShape(newShape);\n        const result = util.getTypedArrayFromDType(dtype, resultSize);\n        const aRank = aShape.length;\n        const bRank = bShape.length;\n        const aStrides = util.computeStrides(aShape);\n        const bStrides = util.computeStrides(bShape);\n        const aBroadcastDims = backend_util.getBroadcastDims(aShape, newShape);\n        const bBroadcastDims = backend_util.getBroadcastDims(bShape, newShape);\n        if (aBroadcastDims.length + bBroadcastDims.length === 0) {\n            for (let i = 0; i < result.length; ++i) {\n                result[i] = op(aVals[i % aVals.length], bVals[i % bVals.length]);\n            }\n        }\n        else {\n            for (let i = 0; i < result.length; ++i) {\n                const loc = util.indexToLoc(i, resultRank, resultStrides);\n                const aLoc = loc.slice(-aRank);\n                aBroadcastDims.forEach(d => aLoc[d] = 0);\n                const aIndex = util.locToIndex(aLoc, aRank, aStrides);\n                const bLoc = loc.slice(-bRank);\n                bBroadcastDims.forEach(d => bLoc[d] = 0);\n                const bIndex = util.locToIndex(bLoc, bRank, bStrides);\n                result[i] = op(aVals[aIndex], bVals[bIndex]);\n            }\n        }\n        return [result, newShape];\n    };\n}\n//# sourceMappingURL=binary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Add } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const addImpl = createSimpleBinaryKernelImpl(((a, b) => a + b));\nexport const addComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal + bReal, imag: aImag + bImag };\n}));\nexport const add = binaryKernelFunc(Add, addImpl, addComplexImpl);\nexport const addConfig = {\n    kernelName: Add,\n    backendName: 'cpu',\n    kernelFunc: add\n};\n//# sourceMappingURL=Add.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\n/**\n * Template that creates implementation for unary op.\n */\nexport function createSimpleUnaryImpl(op) {\n    return (values, dtype, attrs) => {\n        const newValues = util.getTypedArrayFromDType(dtype, values.length);\n        for (let i = 0; i < values.length; ++i) {\n            newValues[i] = op(values[i], attrs);\n        }\n        return newValues;\n    };\n}\n//# sourceMappingURL=unary_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Ceil } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const ceilImpl = createSimpleUnaryImpl((xi) => Math.ceil(xi));\nexport const ceilKernelFunc = unaryKernelFuncFromImpl(Ceil, ceilImpl);\nexport const ceilConfig = {\n    kernelName: Ceil,\n    backendName: 'cpu',\n    kernelFunc: ceilKernelFunc,\n};\n//# sourceMappingURL=Ceil.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Exp } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expImpl = createSimpleUnaryImpl((xi) => Math.exp(xi));\nexport const expKernelFunc = unaryKernelFuncFromImpl(Exp, expImpl);\nexport const expConfig = {\n    kernelName: Exp,\n    backendName: 'cpu',\n    kernelFunc: expKernelFunc,\n};\n//# sourceMappingURL=Exp.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Expm1 } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const expm1Impl = createSimpleUnaryImpl((xi) => Math.expm1(xi));\nexport const expm1KernelFunc = unaryKernelFuncFromImpl(Expm1, expm1Impl);\nexport const expm1Config = {\n    kernelName: Expm1,\n    backendName: 'cpu',\n    kernelFunc: expm1KernelFunc,\n};\n//# sourceMappingURL=Expm1.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Floor } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const floorImpl = createSimpleUnaryImpl((xi) => Math.floor(xi));\nexport const floorKernelFunc = unaryKernelFuncFromImpl(Floor, floorImpl);\nexport const floorConfig = {\n    kernelName: Floor,\n    backendName: 'cpu',\n    kernelFunc: floorKernelFunc,\n};\n//# sourceMappingURL=Floor.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Log } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const logImpl = createSimpleUnaryImpl((xi) => Math.log(xi));\nexport const logKernelFunc = unaryKernelFuncFromImpl(Log, logImpl);\nexport const logConfig = {\n    kernelName: Log,\n    backendName: 'cpu',\n    kernelFunc: logKernelFunc,\n};\n//# sourceMappingURL=Log.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function maxImpl(aVals, reduceSize, outShape, dtype) {\n    const vals = util.getTypedArrayFromDType(dtype, util.sizeFromShape(outShape));\n    for (let i = 0; i < vals.length; ++i) {\n        const offset = i * reduceSize;\n        let max = aVals[offset];\n        for (let j = 0; j < reduceSize; ++j) {\n            const value = aVals[offset + j];\n            if (value > max) {\n                max = value;\n            }\n        }\n        vals[i] = max;\n    }\n    return vals;\n}\n//# sourceMappingURL=Max_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Multiply } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const multiplyImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue * bValue));\nexport const multiplyComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return {\n        real: aReal * bReal - aImag * bImag,\n        imag: aReal * bImag + aImag * bReal\n    };\n}));\nexport const multiply = binaryKernelFunc(Multiply, multiplyImpl, multiplyComplexImpl);\nexport const multiplyConfig = {\n    kernelName: Multiply,\n    backendName: 'cpu',\n    kernelFunc: multiply\n};\n//# sourceMappingURL=Multiply.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Rsqrt } from '@tensorflow/tfjs-core';\nimport { createSimpleUnaryImpl } from '../utils/unary_impl';\nimport { unaryKernelFuncFromImpl } from '../utils/unary_utils';\nexport const rsqrtImpl = createSimpleUnaryImpl((xi) => 1 / Math.sqrt(xi));\nexport const rsqrtKernelFunc = unaryKernelFuncFromImpl(Rsqrt, rsqrtImpl);\nexport const rsqrtConfig = {\n    kernelName: Rsqrt,\n    backendName: 'cpu',\n    kernelFunc: rsqrtKernelFunc,\n};\n//# sourceMappingURL=Rsqrt.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Slice, slice_util, util } from '@tensorflow/tfjs-core';\nimport { assertNotComplex } from '../cpu_util';\nexport function sliceImpl(vals, begin, size, shape, dtype) {\n    const isContinous = slice_util.isSliceContinous(shape, begin, size);\n    const length = util.sizeFromShape(size);\n    const xStrides = util.computeStrides(shape);\n    if (isContinous) {\n        const flatOffset = slice_util.computeFlatOffset(begin, xStrides);\n        return vals.subarray(flatOffset, flatOffset + length);\n    }\n    const outVals = util.getTypedArrayFromDType(dtype, length);\n    for (let i = 0; i < length; ++i) {\n        const rank = size.length;\n        const strides = util.computeStrides(size);\n        const loc = util.indexToLoc(i, rank, strides);\n        const xLoc = loc.map((idx, j) => idx + begin[j]);\n        const xIndex = util.locToIndex(xLoc, shape.length, xStrides);\n        outVals[i] = vals[xIndex];\n    }\n    return outVals;\n}\nexport function slice(args) {\n    const { inputs, backend, attrs } = args;\n    const { x } = inputs;\n    const { begin, size } = attrs;\n    assertNotComplex(x, 'slice');\n    const [$begin, $size] = slice_util.parseSliceParams(x, begin, size);\n    slice_util.assertParamsValid(x, $begin, $size);\n    const vals = backend.data.get(x.dataId).values;\n    const outVals = sliceImpl(vals, $begin, $size, x.shape, x.dtype);\n    return backend.makeTensorInfo($size, x.dtype, outVals);\n}\nexport const sliceConfig = {\n    kernelName: Slice,\n    backendName: 'cpu',\n    kernelFunc: slice\n};\n//# sourceMappingURL=Slice.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { Sub } from '@tensorflow/tfjs-core';\nimport { createSimpleBinaryKernelImpl } from '../utils/binary_impl';\nimport { binaryKernelFunc, createComplexBinaryKernelImpl } from '../utils/kernel_utils';\nexport const subImpl = createSimpleBinaryKernelImpl(((aValue, bValue) => aValue - bValue));\nexport const subComplexImpl = createComplexBinaryKernelImpl(((aReal, aImag, bReal, bImag) => {\n    return { real: aReal - bReal, imag: aImag - bImag };\n}));\nexport const sub = binaryKernelFunc(Sub, subImpl, subComplexImpl);\nexport const subConfig = {\n    kernelName: Sub,\n    backendName: 'cpu',\n    kernelFunc: sub\n};\n//# sourceMappingURL=Sub.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { util } from '@tensorflow/tfjs-core';\nexport function transposeImpl(xVals, xShape, dtype, perm, newShape) {\n    const xRank = xShape.length;\n    const xSize = util.sizeFromShape(xShape);\n    const xStrides = util.computeStrides(xShape);\n    const newStrides = util.computeStrides(newShape);\n    const result = util.getTypedArrayFromDType(dtype, util.sizeFromShape(newShape));\n    for (let i = 0; i < xSize; ++i) {\n        const loc = util.indexToLoc(i, xRank, xStrides);\n        // Permute location.\n        const newLoc = new Array(loc.length);\n        for (let i = 0; i < newLoc.length; i++) {\n            newLoc[i] = loc[perm[i]];\n        }\n        const newIndex = util.locToIndex(newLoc, xRank, newStrides);\n        result[newIndex] = xVals[i];\n    }\n    return result;\n}\n//# sourceMappingURL=Transpose_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport { TensorBuffer, util } from '@tensorflow/tfjs-core';\nexport function uniqueImpl(values, axis, shape, dtype) {\n    // Normalize and validate axis.\n    const $axis = util.parseAxisParam(axis, shape)[0];\n    // Calculate the new shape that is suitable for extracting data along the\n    // given axis.\n    //\n    // The rank is 3.\n    // The size of the 1st dimension is the size of all the axes < the given axis.\n    // The size of the 2nd dimension is the same as the size of the given axis.\n    // The size of the 3rd dimension is the size of all the axes > the given axis.\n    //\n    // For example, for a 4D tensor with shape=[2, 3, 5, 4] and axis=2, the\n    // newShape would be: [2*3, 5, 4].\n    //\n    // Note that this is not the final output shape. This will be the shape for an\n    // intermediate TensorBuffer (see inputBuffer below) to allow us to extract\n    // values along the given axis. To demonstrate how it works, consider the\n    // following example:\n    //\n    // Input: a 3D tensor, with shape [1, 2, 3]\n    // [\n    //   [\n    //      [1,2,3],\n    //      [4,5,6]\n    //   ]\n    // ]\n    // Axis: 2 (the last axis).\n    // Along axis 2, we expect to extract 3 tensors: [1,4], [2,5], [3,6].\n    //\n    // For this example, newShape would be: [2, 3, 1], where 2 is calculated from\n    // 1*2. The re-shaped data would look like:\n    //\n    // [\n    //   [\n    //     [1], [2], [3]\n    //   ],\n    //   [\n    //     [4], [5], [6]\n    //   ]\n    // ]\n    //\n    // Then, we can construct a 3-level nested loop by the following dimension\n    // order to extract the values along the axis (dimension1):\n    // i: dimension1       // 0,1,2 (newShape[1])\n    //   m: dimension0     // 0,1   (newShape[0])\n    //     n: dimension2   // 0     (newShape[2])\n    //\n    //                       m, i, n\n    //                      ---------\n    // Iteration 0: data at [0, 0, 0] => \"1\"\n    // Iteration 1: data at [1, 0, 0] => \"4\"\n    // We got [1,4].\n    // Iteration 2: data at [0, 1, 0] => \"2\"\n    // Iteration 3: data at [1, 1, 0] => \"5\"\n    // We got [2,5].\n    // Iteration 4: data at [0, 2, 0] => \"3\"\n    // Iteration 5: data at [1, 2, 0] => \"6\"\n    // We got [3,6].\n    const newShape = [1, shape[0], 1];\n    for (let i = 0; i < $axis; i++) {\n        newShape[0] *= shape[i];\n    }\n    newShape[1] = shape[$axis];\n    for (let i = $axis + 1; i < shape.length; i++) {\n        newShape[2] *= shape[i];\n    }\n    // A map from unique elements (their string representations) to their values\n    // in \"indices\" (below).\n    const uniqueElements = {};\n    // The indices of each unique element in the original tensor along the given\n    // axis. It is 1D and has the same size as the given axis.\n    const indices = new Int32Array(shape[$axis]);\n    // Create a buffer so we can easily extract value at a given location.\n    const inputBuffer = new TensorBuffer(newShape, dtype, values);\n    // The indices along the given axis that have unique elements. This is a\n    // de-duped version of \"indices\" above.\n    const uniqueIndices = [];\n    const is1DTensor = newShape[0] === 1 && newShape[2] === 1;\n    for (let i = 0; i < shape[$axis]; i++) {\n        // Extract values along the axis.\n        let element;\n        if (is1DTensor) {\n            // Fast path for 1D tensor input.\n            element = values[i].toString();\n        }\n        else {\n            const axisValues = [];\n            for (let m = 0; m < newShape[0]; m++) {\n                for (let n = 0; n < newShape[2]; n++) {\n                    axisValues.push(inputBuffer.get(m, i, n));\n                }\n            }\n            element = axisValues.join(',');\n        }\n        // Dedup and update various indices.\n        if (uniqueElements[element] !== undefined) {\n            indices[i] = uniqueElements[element];\n        }\n        else {\n            const uniqueIndex = Object.keys(uniqueElements).length;\n            uniqueElements[element] = uniqueIndex;\n            indices[i] = uniqueIndex;\n            uniqueIndices.push(i);\n        }\n    }\n    // Now we know where each of the unique elements are located along the axis\n    // (uniqueIndices). Extract them from input buffer and store them in the\n    // output buffer.\n    const outputTmpShape = newShape.slice();\n    outputTmpShape[1] = Object.keys(uniqueElements).length;\n    const outputBuffer = new TensorBuffer(outputTmpShape, dtype);\n    uniqueIndices.forEach((uniqueElementIndex, i) => {\n        for (let m = 0; m < newShape[0]; m++) {\n            for (let n = 0; n < newShape[2]; n++) {\n                outputBuffer.set(inputBuffer.get(m, uniqueElementIndex, n), m, i, n);\n            }\n        }\n    });\n    // The output shape can be calculated from the input shape with the size of\n    // the given axis replaced by the number of unique elements along that axis.\n    const outputShape = shape.slice();\n    outputShape[$axis] = outputTmpShape[1];\n    return {\n        outputValues: outputBuffer.values,\n        outputShape,\n        indices,\n    };\n}\n//# sourceMappingURL=Unique_impl.js.map", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import shared functionality from tfjs-backend-cpu without triggering\n// side effects.\n// tslint:disable-next-line: no-imports-from-dist\nimport * as shared from '@tensorflow/tfjs-backend-cpu/dist/shared';\n\nconst {\n  simpleAbsImpl: simpleAbsImplCPU,\n  addImpl: addImplCPU,\n  ceilImpl: ceilImplCPU,\n  expImpl: expImplCPU,\n  expm1Impl: expm1ImplCPU,\n  floorImpl: floorImplCPU,\n  logImpl: logImplCPU,\n  maxImpl: maxImplCPU,\n  multiplyImpl: multiplyImplCPU,\n  rsqrtImpl: rsqrtImplCPU,\n  sliceImpl: sliceImplCPU,\n  subImpl: subImplCPU,\n  transposeImpl: transposeImplCPU,\n  uniqueImpl: uniqueImplCPU,\n} = shared;\n\nexport {\n  simpleAbsImplCPU,\n  addImplCPU,\n  ceilImplCPU,\n  expImplCPU,\n  expm1ImplCPU,\n  logImplCPU,\n  multiplyImplCPU,\n  sliceImplCPU,\n  subImplCPU,\n  floorImplCPU,\n  maxImplCPU,\n  rsqrtImplCPU,\n  transposeImplCPU,\n  uniqueImplCPU,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`float v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        float result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AddNPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(outputShape: number[], shapes: number[][]) {\n    this.outputShape = outputShape;\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const snippets: string[] = [];\n    // Get target elements from every input tensor.\n    this.variableNames.forEach(variable => {\n      snippets.push(`vec4 v${variable} = get${variable}AtOutCoords();`);\n    });\n\n    // Calculate the sum of all elements.\n    const operation = this.variableNames\n                          .map(variable => {\n                            return `v${variable}`;\n                          })\n                          .join(' + ');\n\n    this.userCode = `\n      void main() {\n        ${snippets.join('\\n        ')}\n\n        vec4 result = ${operation};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ArgMinMaxProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo, op: 'max'|'min',\n      firstPass: boolean) {\n    const {windowSize, batchSize, outSize} = reduceInfo;\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    this.outputShape = [batchSize, outSize];\n    const compOp = (op === 'max') ? '>' : '<';\n    const indexSnippet = firstPass ?\n        'inOffset + i;' :\n        'round(getBestIndicesA(batch, inOffset + i));';\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        int bestIndex = inOffset;\n        float bestValue = getA(batch, bestIndex);\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          int inIdx = ${indexSnippet};\n          float candidate = getA(batch, inIdx);\n          if (candidate ${compOp} bestValue) {\n            bestValue = candidate;\n            bestIndex = inIdx;\n          }\n        }\n        setOutput(float(bestIndex));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport function getVecChannels(name: string, rank: number): string[] {\n  return ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank).map(d => `${name}.${d}`);\n}\n\nexport function getChannels(name: string, rank: number): string[] {\n  if (rank === 1) {\n    return [name];\n  }\n  return getVecChannels(name, rank);\n}\n\nexport function getSourceCoords(rank: number, dims: string[]): string {\n  if (rank === 1) {\n    return 'rc';\n  }\n\n  let coords = '';\n  for (let i = 0; i < rank; i++) {\n    coords += dims[i];\n    if (i < rank - 1) {\n      coords += ',';\n    }\n  }\n  return coords;\n}", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '@tensorflow/tfjs-core';\n\nexport type GLSL = {\n  version: string,\n  attribute: string,\n  varyingVs: string,\n  varyingFs: string,\n  texture2D: string,\n  output: string,\n  defineOutput: string,\n  defineSpecialNaN: string,\n  defineSpecialInf: string,\n  defineRound: string\n};\n\nexport function getGlslDifferences(): GLSL {\n  let version: string;\n  let attribute: string;\n  let varyingVs: string;\n  let varyingFs: string;\n  let texture2D: string;\n  let output: string;\n  let defineOutput: string;\n  let defineSpecialNaN: string;\n  let defineSpecialInf: string;\n  let defineRound: string;\n\n  if (env().getNumber('WEBGL_VERSION') === 2) {\n    version = '#version 300 es';\n    attribute = 'in';\n    varyingVs = 'out';\n    varyingFs = 'in';\n    texture2D = 'texture';\n    output = 'outputColor';\n    defineOutput = 'out vec4 outputColor;';\n\n    // Use custom isnan definition to work across differences between\n    // implementations on various platforms. While this should happen in ANGLE\n    // we still see differences between android and windows (on chrome) when\n    // using isnan directly.\n    defineSpecialNaN = `\n      bool isnan_custom(float val) {\n        return (val > 0.0 || val < 0.0) ? false : val != 0.0;\n      }\n\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan_custom(val.x),\n          isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));\n      }\n\n      #define isnan(value) isnan_custom(value)\n    `;\n    // In webgl 2 we do not need to specify a custom isinf so there is no\n    // need for a special INFINITY constant.\n    defineSpecialInf = ``;\n    defineRound = `\n      #define round(value) newRound(value)\n      int newRound(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 newRound(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  } else {\n    version = '';\n    attribute = 'attribute';\n    varyingVs = 'varying';\n    varyingFs = 'varying';\n    texture2D = 'texture2D';\n    output = 'gl_FragColor';\n    defineOutput = '';\n    // WebGL1 has no built in isnan so we define one here.\n    defineSpecialNaN = `\n      #define isnan(value) isnan_custom(value)\n      bool isnan_custom(float val) {\n        return (val > 0. || val < 1. || val == 0.) ? false : true;\n      }\n      bvec4 isnan_custom(vec4 val) {\n        return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));\n      }\n    `;\n    defineSpecialInf = `\n      uniform float INFINITY;\n\n      bool isinf(float val) {\n        return abs(val) == INFINITY;\n      }\n      bvec4 isinf(vec4 val) {\n        return equal(abs(val), vec4(INFINITY));\n      }\n    `;\n    defineRound = `\n      int round(float value) {\n        return int(floor(value + 0.5));\n      }\n\n      ivec4 round(vec4 value) {\n        return ivec4(floor(value + vec4(0.5)));\n      }\n    `;\n  }\n\n  return {\n    version,\n    attribute,\n    varyingVs,\n    varyingFs,\n    texture2D,\n    output,\n    defineOutput,\n    defineSpecialNaN,\n    defineSpecialInf,\n    defineRound\n  };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\n/**\n * Produces GLSL code that derives logical coordinates from a flat\n * index. The code performs integer division with each stride and decrements\n * the index until the index equals the final dimension coordinate.\n */\nexport function getLogicalCoordinatesFromFlatIndex(\n    coords: string[], shape: number[], index = 'index'): string {\n  const strides = util.computeStrides(shape);\n  return strides\n      .map((stride, i) => {\n        const line1 = `int ${coords[i]} = ${index} / ${stride}`;\n        const line2 = i === strides.length - 1 ?\n            `int ${coords[i + 1]} = ${index} - ${coords[i]} * ${stride}` :\n            `index -= ${coords[i]} * ${stride}`;\n        return `${line1}; ${line2};`;\n      })\n      .join('');\n}\n\nfunction buildVec(x: string[]): string {\n  if (x.length === 1) {\n    return `${x[0]}`;\n  }\n  return `vec${x.length}(${x.join(',')})`;\n}\n\n/**\n * Produces GLSL code that computes the dot product of the input x and y\n * vectors. Handles splitting inputs into increments of vec4s when necessary.\n */\nexport function dotify(x: string[], y: string[]): string {\n  if (x.length !== y.length) {\n    throw new Error(\n        `Vectors to be dotted must be of the same length -` +\n        `got ${x.length} and ${y.length}`);\n  }\n\n  const slices: string[] = [];\n  const nearestVec4 = Math.floor(x.length / 4);\n  const nearestVec4Remainder = x.length % 4;\n\n  for (let i = 0; i < nearestVec4; i++) {\n    const xSlice = x.slice(i * 4, i * 4 + 4);\n    const ySlice = y.slice(i * 4, i * 4 + 4);\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  if (nearestVec4Remainder !== 0) {\n    let xSlice = x.slice(nearestVec4 * 4);\n    let ySlice = y.slice(nearestVec4 * 4);\n    if (xSlice.length === 1) {\n      xSlice = xSlice.map(d => `float(${d})`);\n      ySlice = ySlice.map(d => `float(${d})`);\n    }\n    slices.push(`${buildVec(xSlice)}, ${buildVec(ySlice)}`);\n  }\n\n  return slices.map((d, i) => `dot(${d})`).join('+');\n}\n\n/**\n * Produces GLSL that computes the flat index from 3D coordinates.\n */\nexport function getFlatIndexFrom3D(shape: [number, number, number]): string {\n  const strides = util.computeStrides(shape).map(d => d.toString());\n\n  return `\n  int getFlatIndex(ivec3 coords) {\n    return coords.x * ${strides[0]} + coords.y * ${strides[1]} + coords.z;\n  }\n`;\n}\n\nexport const ENCODE_FLOAT_SNIPPET = `\n  const float FLOAT_MAX = 1.70141184e38;\n  const float FLOAT_MIN = 1.17549435e-38;\n\n  lowp vec4 encode_float(highp float v) {\n    if (isnan(v)) {\n      return vec4(255, 255, 255, 255);\n    }\n\n    highp float av = abs(v);\n\n    if(av < FLOAT_MIN) {\n      return vec4(0.0, 0.0, 0.0, 0.0);\n    } else if(v > FLOAT_MAX) {\n      return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;\n    } else if(v < -FLOAT_MAX) {\n      return vec4(0.0, 0.0,  128.0, 255.0) / 255.0;\n    }\n\n    highp vec4 c = vec4(0,0,0,0);\n\n    highp float e = floor(log2(av));\n    highp float m = exp2(fract(log2(av))) - 1.0;\n\n    c[2] = floor(128.0 * m);\n    m -= c[2] / 128.0;\n    c[1] = floor(32768.0 * m);\n    m -= c[1] / 32768.0;\n    c[0] = floor(8388608.0 * m);\n\n    highp float ebias = e + 127.0;\n    c[3] = floor(ebias / 2.0);\n    ebias -= c[3] * 2.0;\n    c[2] += floor(ebias) * 128.0;\n\n    c[3] += 128.0 * step(0.0, -v);\n\n    return c / 255.0;\n  }\n`;\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\nconst {getBroadcastDims} = backend_util;\nimport {getGlslDifferences, GLSL} from './glsl_version';\nimport * as shader_util from './shader_compiler_util';\n\nexport type ShapeInfo = {\n  logicalShape: number[],\n  texShape: [number, number],\n  isUniform: boolean,\n  isPacked: boolean,\n  flatOffset: number\n};\n\nexport type InputInfo = {\n  name: string,\n  shapeInfo: ShapeInfo\n};\n\nexport function makeShader(\n    inputsInfo: InputInfo[], outputShape: ShapeInfo, userCode: string,\n    usesPackedTextures: boolean): string {\n  const prefixSnippets: string[] = [];\n  inputsInfo.forEach(x => {\n    const size = util.sizeFromShape(x.shapeInfo.logicalShape);\n\n    // Snippet when we decided to upload the values as uniform.\n    if (x.shapeInfo.isUniform) {\n      prefixSnippets.push(\n          `uniform float ${x.name}${size > 1 ? `[${size}]` : ''};`);\n    } else {\n      prefixSnippets.push(`uniform sampler2D ${x.name};`);\n      prefixSnippets.push(`uniform int offset${x.name};`);\n    }\n  });\n  const inputPrefixSnippet = prefixSnippets.join('\\n');\n\n  const inputSamplingSnippet =\n      inputsInfo\n          .map(x => getInputSamplingSnippet(x, outputShape, usesPackedTextures))\n          .join('\\n');\n  const outTexShape = outputShape.texShape;\n  const glsl = getGlslDifferences();\n  const floatTextureSampleSnippet = getFloatTextureSampleSnippet(glsl);\n  let outputSamplingSnippet: string;\n  let floatTextureSetOutputSnippet: string;\n  let shaderPrefix = getShaderPrefix(glsl);\n\n  if (outputShape.isPacked) {\n    outputSamplingSnippet =\n        getPackedOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRGBASnippet(glsl);\n  } else {\n    outputSamplingSnippet =\n        getOutputSamplingSnippet(outputShape.logicalShape, outTexShape);\n    floatTextureSetOutputSnippet = getFloatTextureSetRSnippet(glsl);\n  }\n\n  if (usesPackedTextures) {\n    shaderPrefix += SHADER_PACKED_PREFIX;\n  }\n\n  const source = [\n    shaderPrefix, floatTextureSampleSnippet, floatTextureSetOutputSnippet,\n    inputPrefixSnippet, outputSamplingSnippet, inputSamplingSnippet, userCode\n  ].join('\\n');\n  return source;\n}\n\nfunction getSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getSamplerScalar(inInfo);\n    case 1:\n      return getSampler1D(inInfo);\n    case 2:\n      return getSampler2D(inInfo);\n    case 3:\n      return getSampler3D(inInfo);\n    case 4:\n      return getSampler4D(inInfo);\n    case 5:\n      return getSampler5D(inInfo);\n    case 6:\n      return getSampler6D(inInfo);\n    default:\n      throw new Error(\n          `${shape.length}-D input sampling` +\n          ` is not yet supported`);\n  }\n}\n\nfunction getPackedSamplerFromInInfo(inInfo: InputInfo): string {\n  const shape = inInfo.shapeInfo.logicalShape;\n  switch (shape.length) {\n    case 0:\n      return getPackedSamplerScalar(inInfo);\n    case 1:\n      return getPackedSampler1D(inInfo);\n    case 2:\n      return getPackedSampler2D(inInfo);\n    case 3:\n      return getPackedSampler3D(inInfo);\n    default:\n      return getPackedSamplerND(inInfo);\n  }\n}\n\nfunction getInputSamplingSnippet(\n    inInfo: InputInfo, outShapeInfo: ShapeInfo,\n    usesPackedTextures = false): string {\n  let res = '';\n  if (usesPackedTextures) {\n    res += getPackedSamplerFromInInfo(inInfo);\n  } else {\n    res += getSamplerFromInInfo(inInfo);\n  }\n\n  const inShape = inInfo.shapeInfo.logicalShape;\n  const outShape = outShapeInfo.logicalShape;\n  if (inShape.length <= outShape.length) {\n    if (usesPackedTextures) {\n      res += getPackedSamplerAtOutputCoords(inInfo, outShapeInfo);\n    } else {\n      res += getSamplerAtOutputCoords(inInfo, outShapeInfo);\n    }\n  }\n  return res;\n}\n\nfunction getPackedOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutputPacked1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutputPacked2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutputPacked3DCoords(\n          outShape as [number, number, number], outTexShape);\n    default:\n      return getOutputPackedNDCoords(outShape, outTexShape);\n  }\n}\n\nfunction getOutputSamplingSnippet(\n    outShape: number[], outTexShape: [number, number]): string {\n  switch (outShape.length) {\n    case 0:\n      return getOutputScalarCoords();\n    case 1:\n      return getOutput1DCoords(outShape as [number], outTexShape);\n    case 2:\n      return getOutput2DCoords(outShape as [number, number], outTexShape);\n    case 3:\n      return getOutput3DCoords(\n          outShape as [number, number, number], outTexShape);\n    case 4:\n      return getOutput4DCoords(\n          outShape as [number, number, number, number], outTexShape);\n    case 5:\n      return getOutput5DCoords(\n          outShape as [number, number, number, number, number], outTexShape);\n    case 6:\n      return getOutput6DCoords(\n          outShape as [number, number, number, number, number, number],\n          outTexShape);\n    default:\n      throw new Error(\n          `${outShape.length}-D output sampling is not yet supported`);\n  }\n}\n\nfunction getFloatTextureSampleSnippet(glsl: GLSL): string {\n  return `\n    float sampleTexture(sampler2D textureSampler, vec2 uv) {\n      return ${glsl.texture2D}(textureSampler, uv).r;\n    }\n  `;\n}\n\nfunction getFloatTextureSetRSnippet(glsl: GLSL): string {\n  return `\n    void setOutput(float val) {\n      ${glsl.output} = vec4(val, 0, 0, 0);\n    }\n  `;\n}\n\nfunction getFloatTextureSetRGBASnippet(glsl: GLSL): string {\n  return `\n    void setOutput(vec4 val) {\n      ${glsl.output} = val;\n    }\n  `;\n}\n\nfunction getShaderPrefix(glsl: GLSL): string {\n  const SHADER_PREFIX = `${glsl.version}\n    precision highp float;\n    precision highp int;\n    precision highp sampler2D;\n    ${glsl.varyingFs} vec2 resultUV;\n    ${glsl.defineOutput}\n    const vec2 halfCR = vec2(0.5, 0.5);\n\n    struct ivec5\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n    };\n\n    struct ivec6\n    {\n      int x;\n      int y;\n      int z;\n      int w;\n      int u;\n      int v;\n    };\n\n    uniform float NAN;\n    ${glsl.defineSpecialNaN}\n    ${glsl.defineSpecialInf}\n    ${glsl.defineRound}\n\n    int imod(int x, int y) {\n      return x - y * (x / y);\n    }\n\n    int idiv(int a, int b, float sign) {\n      int res = a / b;\n      int mod = imod(a, b);\n      if (sign < 0. && mod != 0) {\n        res -= 1;\n      }\n      return res;\n    }\n\n    //Based on the work of Dave Hoskins\n    //https://www.shadertoy.com/view/4djSRW\n    #define HASHSCALE1 443.8975\n    float random(float seed){\n      vec2 p = resultUV * seed;\n      vec3 p3  = fract(vec3(p.xyx) * HASHSCALE1);\n      p3 += dot(p3, p3.yzx + 19.19);\n      return fract((p3.x + p3.y) * p3.z);\n    }\n\n    ${SAMPLE_1D_SNIPPET}\n    ${SAMPLE_2D_SNIPPET}\n    ${SAMPLE_3D_SNIPPET}\n  `;\n\n  return SHADER_PREFIX;\n}\n\nconst SAMPLE_1D_SNIPPET = `\nvec2 uvFromFlat(int texNumR, int texNumC, int index) {\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\nvec2 packedUVfrom1D(int texNumR, int texNumC, int index) {\n  int texelIndex = index / 2;\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_2D_SNIPPET = `\nvec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,\n  int texNumC, int row, int col) {\n  int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = texelIndex / texNumC;\n  int texC = texelIndex - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SAMPLE_3D_SNIPPET = `\nvec2 packedUVfrom3D(int texNumR, int texNumC,\n    int texelsInBatch, int texelsInLogicalRow, int b,\n    int row, int col) {\n  int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);\n  int texR = index / texNumC;\n  int texC = index - texR * texNumC;\n  return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);\n}\n`;\n\nconst SHADER_PACKED_PREFIX = `\n  float getChannel(vec4 frag, vec2 innerDims) {\n    vec2 modCoord = mod(innerDims, 2.);\n    return modCoord.x == 0. ?\n      (modCoord.y == 0. ? frag.r : frag.g) :\n      (modCoord.y == 0. ? frag.b : frag.a);\n  }\n  float getChannel(vec4 frag, int dim) {\n    float modCoord = mod(float(dim), 2.);\n    return modCoord == 0. ? frag.r : frag.g;\n  }\n`;\n\nfunction getOutputScalarCoords() {\n  return `\n    int getOutputCoords() {\n      return 0;\n    }\n  `;\n}\n\nfunction getOutputPacked1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (packedTexShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.x * ${packedTexShape[1]}.0);\n      }\n    `;\n  }\n\n  if (packedTexShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return 2 * int(resultUV.y * ${packedTexShape[0]}.0);\n      }\n    `;\n  }\n\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      return 2 * (resTexRC.x * ${packedTexShape[1]} + resTexRC.y);\n    }\n  `;\n}\n\nfunction getOutput1DCoords(\n    shape: [number], texShape: [number, number]): string {\n  if (texShape[0] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.x * ${texShape[1]}.0);\n      }\n    `;\n  }\n  if (texShape[1] === 1) {\n    return `\n      int getOutputCoords() {\n        return int(resultUV.y * ${texShape[0]}.0);\n      }\n    `;\n  }\n  return `\n    int getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      return resTexRC.x * ${texShape[1]} + resTexRC.y;\n    }\n  `;\n}\n\nfunction getOutputPacked3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texelsInLogicalRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = texelsInLogicalRow * Math.ceil(shape[1] / 2);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec3(b, r, c);\n    }\n  `;\n}\n\nfunction getOutput3DCoords(\n    shape: [number, number, number], texShape: [number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n\nfunction getOutputPackedNDCoords(\n    shape: number[], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  const texelsInLogicalRow = Math.ceil(shape[shape.length - 1] / 2);\n  const texelsInBatch =\n      texelsInLogicalRow * Math.ceil(shape[shape.length - 2] / 2);\n  let texelsInBatchN = texelsInBatch;\n  let batches = ``;\n  let coords = 'b, r, c';\n\n  for (let b = 2; b < shape.length - 1; b++) {\n    texelsInBatchN *= shape[shape.length - b - 1];\n    batches = `\n      int b${b} = index / ${texelsInBatchN};\n      index -= b${b} * ${texelsInBatchN};\n    ` + batches;\n    coords = `b${b}, ` + coords;\n  }\n\n  return `\n    ivec${shape.length} getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n\n      ${batches}\n\n      int b = index / ${texelsInBatch};\n      index -= b * ${texelsInBatch};\n\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec${shape.length}(${coords});\n    }\n  `;\n}\n\nfunction getOutput4DCoords(\n    shape: [number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2'], shape);\n\n  return `\n    ivec4 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      ${coordsFromIndexSnippet}\n      return ivec4(r, c, d, d2);\n    }\n  `;\n}\n\nfunction getOutput5DCoords(\n    shape: [number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3'], shape);\n\n  return `\n    ivec5 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx * vec2(${texShape[0]},\n                             ${texShape[1]}));\n\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec5 outShape = ivec5(r, c, d, d2, d3);\n      return outShape;\n    }\n  `;\n}\n\nfunction getOutput6DCoords(\n    shape: [number, number, number, number, number, number],\n    texShape: [number, number]): string {\n  const coordsFromIndexSnippet = shader_util.getLogicalCoordinatesFromFlatIndex(\n      ['r', 'c', 'd', 'd2', 'd3', 'd4'], shape);\n\n  return `\n    ivec6 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n        vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n\n      ${coordsFromIndexSnippet}\n\n      ivec6 result = ivec6(r, c, d, d2, d3, d4);\n      return result;\n    }\n  `;\n}\n\nfunction getOutputPacked2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return 2 * ivec2(resultUV.yx * vec2(${packedTexShape[0]}, ${\n        packedTexShape[1]}));\n      }\n    `;\n  }\n\n  // texels needed to accommodate a logical row\n  const texelsInLogicalRow = Math.ceil(shape[1] / 2);\n\n  /**\n   * getOutputCoords\n   *\n   * resTexRC: The rows and columns of the texels. If you move over one\n   * texel to the right in the packed texture, you are moving over one column\n   * (not two).\n   *\n   * index: The texel index\n   */\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${packedTexShape[0]}, ${packedTexShape[1]}));\n\n      int index = resTexRC.x * ${packedTexShape[1]} + resTexRC.y;\n      int r = 2 * (index / ${texelsInLogicalRow});\n      int c = imod(index, ${texelsInLogicalRow}) * 2;\n\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getOutput2DCoords(\n    shape: [number, number], texShape: [number, number]): string {\n  if (util.arraysEqual(shape, texShape)) {\n    return `\n      ivec2 getOutputCoords() {\n        return ivec2(resultUV.yx * vec2(${texShape[0]}, ${texShape[1]}));\n      }\n    `;\n  }\n  if (shape[1] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(index, 0);\n      }\n    `;\n  }\n  if (shape[0] === 1) {\n    return `\n      ivec2 getOutputCoords() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n                               vec2(${texShape[0]}, ${texShape[1]}));\n        int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n        return ivec2(0, index);\n      }\n    `;\n  }\n  return `\n    ivec2 getOutputCoords() {\n      ivec2 resTexRC = ivec2(resultUV.yx *\n                             vec2(${texShape[0]}, ${texShape[1]}));\n      int index = resTexRC.x * ${texShape[1]} + resTexRC.y;\n      int r = index / ${shape[1]};\n      int c = index - r * ${shape[1]};\n      return ivec2(r, c);\n    }\n  `;\n}\n\nfunction getFlatOffsetUniformName(texName: string): string {\n  return `offset${texName}`;\n}\n\nfunction getPackedSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}() {\n      return ${glsl.texture2D}(${texName}, halfCR);\n    }\n  `;\n}\n\nfunction getSamplerScalar(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  if (inputInfo.shapeInfo.isUniform) {\n    return `float ${funcName}() {return ${texName};}`;\n  }\n  const [texNumR, texNumC] = inputInfo.shapeInfo.texShape;\n  if (texNumR === 1 && texNumC === 1) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n\n  const [tNumR, tNumC] = inputInfo.shapeInfo.texShape;\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}() {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int index) {\n      vec2 uv = packedUVfrom1D(\n        ${packedTexShape[0]}, ${packedTexShape[1]}, index);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler1D(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int index) {\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const tNumR = texShape[0];\n  const tNumC = texShape[1];\n\n  if (tNumC === 1 && tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        return sampleTexture(${texName}, halfCR);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  if (tNumC === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2(0.5, (float(index + ${offset}) + 0.5) / ${tNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (tNumR === 1) {\n    return `\n      float ${funcName}(int index) {\n        vec2 uv = vec2((float(index + ${offset}) + 0.5) / ${tNumC}.0, 0.5);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  return `\n    float ${funcName}(int index) {\n      vec2 uv = uvFromFlat(${tNumR}, ${tNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getPackedSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const glsl = getGlslDifferences();\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    return `\n      vec4 ${funcName}(int row, int col) {\n        vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n\n        return ${glsl.texture2D}(${texName}, uv);\n      }\n    `;\n  }\n\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const valuesPerRow = Math.ceil(shape[1] / 2);\n\n  return `\n    vec4 ${funcName}(int row, int col) {\n      vec2 uv = packedUVfrom2D(${valuesPerRow}, ${packedTexShape[0]}, ${\n      packedTexShape[1]}, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler2D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n\n  if (texShape != null && util.arraysEqual(shape, texShape)) {\n    const texNumR = texShape[0];\n    const texNumC = texShape[1];\n    return `\n    float ${funcName}(int row, int col) {\n      vec2 uv = (vec2(col, row) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col) {\n        int index = round(dot(vec2(row, col), vec2(${shape[1]}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const offset = getFlatOffsetUniformName(texName);\n  if (texNumC === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2(0.5, (index + 0.5) / ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n  if (texNumR === 1) {\n    // index is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col) {\n      float index = dot(vec3(row, col, ${offset}), vec3(${shape[1]}, 1, 1));\n      vec2 uv = vec2((index + 0.5) / ${texNumC}.0, 0.5);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  return `\n  float ${funcName}(int row, int col) {\n    // Explicitly use integer operations as dot() only works on floats.\n    int index = row * ${shape[1]} + col + ${offset};\n    vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n    return sampleTexture(${texName}, uv);\n  }\n`;\n}\n\nfunction getPackedSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n\n  if (shape[0] === 1) {\n    const squeezedShape = shape.slice(1);\n    const keptDims = [1, 2];\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['b', 'row', 'col'];\n    return `\n        ${getPackedSamplerFromInInfo(newInputInfo)}\n        vec4 ${funcName}(int b, int row, int col) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[2] / 2);\n  const texelsInBatch = valuesPerRow * Math.ceil(shape[1] / 2);\n  const glsl = getGlslDifferences();\n\n  return `\n    vec4 ${funcName}(int b, int row, int col) {\n      vec2 uv = packedUVfrom3D(\n        ${texNumR}, ${texNumC}, ${texelsInBatch}, ${valuesPerRow}, b, row, col);\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler3D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride0 = shape[1] * shape[2];\n  const stride1 = shape[2];\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  const squeezedShape = newShape;\n  if (squeezedShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, squeezedShape);\n    const params = ['row', 'col', 'depth'];\n    return `\n        ${getSamplerFromInInfo(newInputInfo)}\n        float ${funcName}(int row, int col, int depth) {\n          return ${funcName}(${getSqueezedParams(params, keptDims)});\n        }\n      `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth) {\n        int index = round(dot(vec3(row, col, depth),\n                          vec3(${stride0}, ${stride1}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n        float ${funcName}(int row, int col, int depth) {\n          float texR = float(row);\n          float texC = dot(vec2(col, depth), vec2(${stride1}, 1));\n          vec2 uv = (vec2(texC, texR) + halfCR) /\n                     vec2(${texNumC}.0, ${texNumR}.0);\n          return sampleTexture(${texName}, uv);\n        }\n      `;\n  }\n\n  if (texNumC === stride1 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n    float ${funcName}(int row, int col, int depth) {\n      float texR = dot(vec2(row, col), vec2(${shape[1]}, 1));\n      float texC = float(depth);\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}.0, ${texNumR}.0);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n      float ${funcName}(int row, int col, int depth) {\n        // Explicitly use integer operations as dot() only works on floats.\n        int index = row * ${stride0} + col * ${stride1} + depth + ${offset};\n        vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n        return sampleTexture(${texName}, uv);\n      }\n  `;\n}\n\nfunction getPackedSamplerND(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const rank = shape.length;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const texShape = inputInfo.shapeInfo.texShape;\n  const packedTexShape =\n      [Math.ceil(texShape[0] / 2), Math.ceil(texShape[1] / 2)];\n  const texNumR = packedTexShape[0];\n  const texNumC = packedTexShape[1];\n\n  const valuesPerRow = Math.ceil(shape[rank - 1] / 2);\n  let texelsInBatch = valuesPerRow * Math.ceil(shape[rank - 2] / 2);\n  let params = `int b, int row, int col`;\n  let index = `b * ${texelsInBatch} + (row / 2) * ${valuesPerRow} + (col / 2)`;\n  for (let b = 2; b < rank - 1; b++) {\n    params = `int b${b}, ` + params;\n    texelsInBatch *= shape[rank - b - 1];\n    index = `b${b} * ${texelsInBatch} + ` + index;\n  }\n  const glsl = getGlslDifferences();\n  return `\n    vec4 ${funcName}(${params}) {\n      int index = ${index};\n      int texR = index / ${texNumC};\n      int texC = index - texR * ${texNumC};\n      vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${texNumC}, ${texNumR});\n      return ${glsl.texture2D}(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler4D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride2 = shape[3];\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        int index = round(dot(vec4(row, col, depth, depth2),\n                          vec4(${stride0}, ${stride1}, ${stride2}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = float(row);\n        float texC =\n            dot(vec3(col, depth, depth2),\n                vec3(${stride1}, ${stride2}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride2 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2) {\n        float texR = dot(vec3(row, col, depth),\n                         vec3(${shape[1] * shape[2]}, ${shape[2]}, 1));\n        float texC = float(depth2);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} +\n          depth * ${stride2} + depth2;\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index + ${offset});\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler5D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n  const stride3 = shape[4];\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float index = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          depth3;\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n                         vec4(${stride1}, ${stride2}, ${stride3}, 1));\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  if (texNumC === stride3 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n        float texR = dot(\n          vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3]},\n               ${shape[2] * shape[3]}, ${shape[3]}, 1));\n        int texC = depth3;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth, int depth2, int depth3) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getSampler6D(inputInfo: InputInfo): string {\n  const shape = inputInfo.shapeInfo.logicalShape;\n  const texName = inputInfo.name;\n  const funcName = 'get' + texName.charAt(0).toUpperCase() + texName.slice(1);\n\n  const {newShape, keptDims} = util.squeezeShape(shape);\n  if (newShape.length < shape.length) {\n    const newInputInfo = squeezeInputInfo(inputInfo, newShape);\n    const params = ['row', 'col', 'depth', 'depth2', 'depth3', 'depth4'];\n    return `\n      ${getSamplerFromInInfo(newInputInfo)}\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        return ${funcName}(${getSqueezedParams(params, keptDims)});\n      }\n    `;\n  }\n\n  const stride4 = shape[5];\n  const stride3 = shape[4] * stride4;\n  const stride2 = shape[3] * stride3;\n  const stride1 = shape[2] * stride2;\n  const stride0 = shape[1] * stride1;\n\n  if (inputInfo.shapeInfo.isUniform) {\n    // Uniform arrays will be less than 65505 (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n        int index = round(dot(\n          vec4(row, col, depth, depth2),\n          vec4(${stride0}, ${stride1}, ${stride2}, ${stride3})) +\n          dot(\n            vec2(depth3, depth4),\n            vec2(${stride4}, 1)));\n        ${getUniformSampler(inputInfo)}\n      }\n    `;\n  }\n\n  const flatOffset = inputInfo.shapeInfo.flatOffset;\n  const texShape = inputInfo.shapeInfo.texShape;\n  const texNumR = texShape[0];\n  const texNumC = texShape[1];\n  if (texNumC === stride0 && flatOffset == null) {\n    // texC is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        int texR = row;\n        float texC = dot(vec4(col, depth, depth2, depth3),\n          vec4(${stride1}, ${stride2}, ${stride3}, ${stride4})) +\n               float(depth4);\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                   vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  if (texNumC === stride4 && flatOffset == null) {\n    // texR is used directly as physical (no risk of float16 overflow).\n    return `\n      float ${funcName}(int row, int col, int depth,\n                    int depth2, int depth3, int depth4) {\n        float texR = dot(vec4(row, col, depth, depth2),\n          vec4(${shape[1] * shape[2] * shape[3] * shape[4]},\n               ${shape[2] * shape[3] * shape[4]},\n               ${shape[3] * shape[4]},\n               ${shape[4]})) + float(depth3);\n        int texC = depth4;\n        vec2 uv = (vec2(texC, texR) + halfCR) /\n                  vec2(${texNumC}.0, ${texNumR}.0);\n        return sampleTexture(${texName}, uv);\n      }\n    `;\n  }\n  const offset = getFlatOffsetUniformName(texName);\n  return `\n    float ${funcName}(int row, int col, int depth,\n                  int depth2, int depth3, int depth4) {\n      // Explicitly use integer operations as dot() only works on floats.\n      int index = row * ${stride0} + col * ${stride1} + depth * ${stride2} +\n          depth2 * ${stride3} + depth3 * ${stride4} + depth4 + ${offset};\n      vec2 uv = uvFromFlat(${texNumR}, ${texNumC}, index);\n      return sampleTexture(${texName}, uv);\n    }\n  `;\n}\n\nfunction getUniformSampler(inputInfo: InputInfo): string {\n  const texName = inputInfo.name;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n\n  if (inSize < 2) {\n    return `return ${texName};`;\n  }\n  return `\n    for (int i = 0; i < ${inSize}; i++) {\n      if (i == index) {\n        return ${texName}[i];\n      }\n    }\n  `;\n}\n\nfunction getPackedSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n\n  const type = getCoordsDataType(outRank);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  let output = `return outputValue;`;\n  const inSize = util.sizeFromShape(inputInfo.shapeInfo.logicalShape);\n  const isInputScalar = inSize === 1;\n  const outSize = util.sizeFromShape(outShapeInfo.logicalShape);\n  const isOutputScalar = outSize === 1;\n\n  if (inRank === 1 && !isInputScalar && !isOutputScalar) {\n    output = `\n      return vec4(outputValue.xy, outputValue.xy);\n    `;\n  } else if (isInputScalar && !isOutputScalar) {\n    if (outRank === 1) {\n      output = `\n        return vec4(outputValue.x, outputValue.x, 0., 0.);\n      `;\n    } else {\n      output = `\n        return vec4(outputValue.x);\n      `;\n    }\n  } else if (broadcastDims.length) {\n    const rows = inRank - 2;\n    const cols = inRank - 1;\n\n    if (broadcastDims.indexOf(rows) > -1 && broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.x);`;\n    } else if (broadcastDims.indexOf(rows) > -1) {\n      output = `return vec4(outputValue.x, outputValue.y, ` +\n          `outputValue.x, outputValue.y);`;\n    } else if (broadcastDims.indexOf(cols) > -1) {\n      output = `return vec4(outputValue.xx, outputValue.zz);`;\n    }\n  }\n\n  return `\n    vec4 ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      vec4 outputValue = get${texFuncSnippet}(${unpackedCoordsSnippet});\n      ${output}\n    }\n  `;\n}\n\nfunction getSamplerAtOutputCoords(\n    inputInfo: InputInfo, outShapeInfo: ShapeInfo) {\n  const texName = inputInfo.name;\n  const texFuncSnippet = texName.charAt(0).toUpperCase() + texName.slice(1);\n  const funcName = 'get' + texFuncSnippet + 'AtOutCoords';\n  const outTexShape = outShapeInfo.texShape;\n  const inTexShape = inputInfo.shapeInfo.texShape;\n  const inRank = inputInfo.shapeInfo.logicalShape.length;\n  const outRank = outShapeInfo.logicalShape.length;\n\n  if (!inputInfo.shapeInfo.isUniform && inRank === outRank &&\n      inputInfo.shapeInfo.flatOffset == null &&\n      util.arraysEqual(inTexShape, outTexShape)) {\n    return `\n      float ${funcName}() {\n        return sampleTexture(${texName}, resultUV);\n      }\n    `;\n  }\n\n  const type = getCoordsDataType(outRank);\n  const broadcastDims = getBroadcastDims(\n      inputInfo.shapeInfo.logicalShape, outShapeInfo.logicalShape);\n  const rankDiff = outRank - inRank;\n  let coordsSnippet: string;\n  const fields = ['x', 'y', 'z', 'w', 'u', 'v'];\n\n  if (inRank === 0) {\n    coordsSnippet = '';\n  } else if (outRank < 2 && broadcastDims.length >= 1) {\n    coordsSnippet = 'coords = 0;';\n  } else {\n    coordsSnippet =\n        broadcastDims.map(d => `coords.${fields[d + rankDiff]} = 0;`)\n            .join('\\n');\n  }\n  let unpackedCoordsSnippet = '';\n  if (outRank < 2 && inRank > 0) {\n    unpackedCoordsSnippet = 'coords';\n  } else {\n    unpackedCoordsSnippet = inputInfo.shapeInfo.logicalShape\n                                .map((s, i) => `coords.${fields[i + rankDiff]}`)\n                                .join(', ');\n  }\n\n  return `\n    float ${funcName}() {\n      ${type} coords = getOutputCoords();\n      ${coordsSnippet}\n      return get${texFuncSnippet}(${unpackedCoordsSnippet});\n    }\n  `;\n}\n\nexport function getCoordsDataType(rank: number): string {\n  if (rank <= 1) {\n    return 'int';\n  } else if (rank === 2) {\n    return 'ivec2';\n  } else if (rank === 3) {\n    return 'ivec3';\n  } else if (rank === 4) {\n    return 'ivec4';\n  } else if (rank === 5) {\n    return 'ivec5';\n  } else if (rank === 6) {\n    return 'ivec6';\n  } else {\n    throw Error(`GPU for rank ${rank} is not yet supported`);\n  }\n}\n\n/** Returns a new input info (a copy) that has a squeezed logical shape. */\nfunction squeezeInputInfo(\n    inInfo: InputInfo, squeezedShape: number[]): InputInfo {\n  // Deep copy.\n  const newInputInfo: InputInfo = JSON.parse(JSON.stringify(inInfo));\n  newInputInfo.shapeInfo.logicalShape = squeezedShape;\n  return newInputInfo;\n}\n\nfunction getSqueezedParams(params: string[], keptDims: number[]): string {\n  return keptDims.map(d => params[d]).join(', ');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ArgMinMaxPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      shape: number[], windowSize: number, op: 'max'|'min',\n      firstPass: boolean) {\n    util.assert(\n        shape.length > 2,\n        () => `Packed arg${\n            op.charAt(0).toUpperCase() +\n            op.slice(1)} supports only inputs with rank above 2.`);\n    const inSize = shape[shape.length - 1];\n    const outSize = Math.ceil(inSize / windowSize);\n    this.outputShape = shape.slice(0, -1);\n    if (outSize > 1) {\n      this.outputShape.push(outSize);\n    }\n    if (!firstPass) {\n      this.variableNames.push('bestIndicesA');\n    }\n    const outShape = this.outputShape;\n    const rank = outShape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n\n    let sourceLocSetup;\n    let sourceRank;\n    if (outSize === 1) {\n      sourceRank = rank + 1;\n      const sourceLocDType = getCoordsDataType(sourceRank);\n      sourceLocSetup = `\n        ${sourceLocDType} sourceLocR = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 1]};\n        ${sourceLocDType} sourceLocG = ${sourceLocDType}(${coords.join()}, 0);\n        ++${coords[rank - 2]};\n        ${sourceLocDType} sourceLocA = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 1]};\n        ${sourceLocDType} sourceLocB = ${sourceLocDType}(${coords.join()}, 0);\n        --${coords[rank - 2]};`;\n    } else {\n      sourceRank = rank;\n      sourceLocSetup = `\n        ${dtype} sourceLocR = coords;\n        ++${coords[rank - 1]};\n        ${dtype} sourceLocG = coords;\n        ++${coords[rank - 2]};\n        ${dtype} sourceLocA = coords;\n        --${coords[rank - 1]};\n        ${dtype} sourceLocB = coords;\n        --${coords[rank - 2]};`;\n    }\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, sourceRank);\n    const inChannel = '.' + channels[sourceRank - 1];  // e.g. \".b\" for rank 3.\n    const intChannels = channels.map(x => 'int ' + x);\n    const srcRCoords =\n        getChannels('sourceLocR', sourceRank - 1).concat('inIdx.r');\n    const srcGCoords =\n        getChannels('sourceLocG', sourceRank - 1).concat('inIdx.g');\n    const srcBCoords =\n        getChannels('sourceLocB', sourceRank - 1).concat('inIdx.b');\n    const srcACoords =\n        getChannels('sourceLocA', sourceRank - 1).concat('inIdx.a');\n\n    const compOp = (op === 'max') ? 'greaterThan' : 'lessThan';\n    const fetchCandidateIdx = firstPass ? '' : `\n          inIdx = round(vec4(getBestIndicesAChannel(${srcRCoords.join()}),\n                             getBestIndicesAChannel(${srcGCoords.join()}),\n                             getBestIndicesAChannel(${srcBCoords.join()}),\n                             getBestIndicesAChannel(${srcACoords.join()})));`;\n\n    const fetchValue = `vec4(\n            getAChannel(${srcRCoords.join()}),\n            hasNextCol ? getAChannel(${srcGCoords.join()}) : 0.,\n            hasNextRow ? getAChannel(${srcBCoords.join()}) : 0.,\n            hasNextRow && hasNextCol ? getAChannel(${srcACoords.join()}) : 0.)`;\n\n    const getBestIndicesAChannelSnippet = firstPass ? '' : `\n      float getBestIndicesAChannel(${intChannels.join()}) {\n        return getChannel(getBestIndicesA(${channels.join()}),\n                                          vec2(${channels.slice(-2).join()}));\n      }`;\n\n    this.userCode = `\n      float getAChannel(${intChannels.join()}) {\n        return getChannel(getA(${channels.join()}),\n                               vec2(${channels.slice(-2).join()}));\n      }\n      ${getBestIndicesAChannelSnippet}\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        bool hasNextCol = ${coords[rank - 1]} < ${outShape[rank - 1] - 1};\n        bool hasNextRow = ${coords[rank - 2]} < ${outShape[rank - 2] - 1};\n        ${sourceLocSetup}\n        ivec4 srcIdx = ivec4(sourceLocR${inChannel}, sourceLocG${inChannel},\n          sourceLocB${inChannel}, sourceLocA${inChannel}) * ${windowSize};\n        ivec4 inIdx = srcIdx;\n        vec4 bestIndex = vec4(inIdx);\n        vec4 bestValue = ${fetchValue};\n\n        for (int i = 0; i < ${windowSize}; i++) {\n          inIdx = srcIdx;\n          ${fetchCandidateIdx}\n          vec4 candidate = ${fetchValue};\n          bvec4 nan = isnan(candidate);\n          bvec4 replace = bvec4(\n            vec4(${compOp}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));\n\n          bestValue = vec4(replace.x  ? candidate.x : bestValue.x,\n                           replace.y  ? candidate.y : bestValue.y,\n                           replace.z  ? candidate.z : bestValue.z,\n                           replace.w  ? candidate.w : bestValue.w);\n          bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));\n          srcIdx++;\n        }\n        setOutput(bestIndex);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class AvgPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth};\n            wC+= ${dilationWidth}) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n\n            dotProd += dyValue * avgMultiplier;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class AvgPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const avgMultiplier = 1 / (filterDepth * filterHeight * filterWidth);\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float avgMultiplier = float(${avgMultiplier});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n\n              dotProd += dyValue * avgMultiplier;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\n// (Ar + Ai)(Br + Bi) =\n// ArBr + ArBi + AiBr + AiBi = ArBr - AB + ArBi + AiBr\n// Yr = ArBr - AB\n// Yi = ArBi + AiBr\nexport const COMPLEX_MULTIPLY = {\n  REAL: 'return areal * breal - aimag * bimag;',\n  IMAG: 'return areal * bimag + aimag * breal;'\n};\n\nexport class BinaryOpComplexProgram implements GPGPUProgram {\n  variableNames = ['AReal', 'AImag', 'BReal', 'BImag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n\n    this.userCode = `\n      float binaryOpComplex(\n          float areal, float aimag, float breal, float bimag) {\n        ${op}\n      }\n\n      void main() {\n        float areal = getARealAtOutCoords();\n        float aimag = getAImagAtOutCoords();\n        float breal = getBRealAtOutCoords();\n        float bimag = getBImagAtOutCoords();\n        setOutput(binaryOpComplex(areal, aimag, breal, bimag));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nconst CHECK_NAN_SNIPPET = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const ADD = 'return a + b;';\nexport const SUB = 'return a - b;';\nexport const MUL = 'return a * b;';\n\n// We use native integer division to deal with floating point imprecision. Since\n// we implement floor division and glsl implements truncated division, we\n// correct for this by subtracting 1 from result when the result is negative and\n// there is a remainder.\nexport const INT_DIV = `\n  float s = sign(a) * sign(b);\n  int ia = round(a);\n  int ib = round(b);\n  if (ib != 0) {\n    // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n    return float(idiv(ia, ib, s));\n  } else {\n    return NAN;\n  }\n`;\n\nexport const POW = `\nif(a < 0.0 && floor(b) < b){\n  return NAN;\n}\nif (b == 0.0) {\n  return 1.0;\n}\nreturn (round(mod(b, 2.0)) != 1) ?\n    pow(abs(a), b) : sign(a) * pow(abs(a), b);\n`;\nexport const SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const EQUAL = `return float(a == b);`;\n\nexport const NOT_EQUAL = `return float(a != b);`;\n\nexport const LESS = `return float(a < b);`;\n\nexport const LESS_EQUAL = `return float(a <= b);`;\n\nexport const GREATER = `return float(a > b);`;\n\nexport const GREATER_EQUAL = `return float(a >= b);`;\n\nexport const LOGICAL_AND = `return float(a >= 1.0 && b >= 1.0);`;\n\nexport const LOGICAL_OR = `return float(a >= 1.0 || b >= 1.0);`;\n\nexport const MAX = CHECK_NAN_SNIPPET + `\n  return max(a, b);\n`;\nexport const MIN = CHECK_NAN_SNIPPET + `\n  return min(a, b);\n`;\nexport const MOD = `if (b == 0.0) return NAN;\n  return mod(a, b);`;\n\nexport const ELU_DER = `return (b >= 1.0) ? a : a * (b + 1.0);`;\n\nexport const PRELU = `return (a < 0.) ? b * a : a;`;\n\nexport class BinaryOpProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, aShape: number[], bShape: number[]) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    this.userCode = `\n      float binaryOperation(float a, float b) {\n        ${op}\n      }\n\n      void main() {\n        float a = getAAtOutCoords();\n        float b = getBAtOutCoords();\n        setOutput(binaryOperation(a, b));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nconst CHECK_NAN_SNIPPET = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\nexport const INT_DIV = `\n  ivec4 ia = round(a);\n  ivec4 ib = round(b);\n  bvec4 cond = notEqual(ib, ivec4(0));\n  ivec4 result = ivec4(0);\n  vec4 s = sign(a) * sign(b);\n\n  // Windows (D3D) wants guaranteed non-zero int division at compile-time.\n  if (cond[0]) {\n    result[0] = idiv(ia[0], ib[0], s[0]);\n  }\n  if (cond[1]) {\n    result[1] = idiv(ia[1], ib[1], s[1]);\n  }\n  if (cond[2]) {\n    result[2] = idiv(ia[2], ib[2], s[2]);\n  }\n  if (cond[3]) {\n    result[3] = idiv(ia[3], ib[3], s[3]);\n  }\n  return vec4(result);\n`;\n\nexport const POW = `\n  // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.\n  vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));\n  vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);\n  vec4 result = multiplier * pow(abs(a), b);\n\n  // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS\n  bvec4 isExpZero = equal(b, vec4(0.0));\n  result.r = isExpZero.r ? 1.0 : result.r;\n  result.g = isExpZero.g ? 1.0 : result.g;\n  result.b = isExpZero.b ? 1.0 : result.b;\n  result.a = isExpZero.a ? 1.0 : result.a;\n\n  vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const PRELU = `\n  vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));\n  return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);\n`;\n\nexport const ELU_DER = `\n  vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));\n  return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));\n`;\n\nexport const EQUAL = `\n  return vec4(equal(a, b));\n`;\n\nexport const NOT_EQUAL = `\n  return vec4(notEqual(a, b));\n`;\n\nexport const LESS = `\n  return vec4(lessThan(a, b));\n`;\n\nexport const LESS_EQUAL = `\n  return vec4(lessThanEqual(a, b));\n`;\n\nexport const GREATER = `\n  return vec4(greaterThan(a, b));\n`;\n\nexport const GREATER_EQUAL = `\n  return vec4(greaterThanEqual(a, b));\n`;\n\nexport const LOGICAL_AND = `\n  return vec4(\n    vec4(greaterThanEqual(a, vec4(1.0))) *\n    vec4(greaterThanEqual(b, vec4(1.0))));\n`;\n\nexport const LOGICAL_OR = `\n  return min(\n    vec4(greaterThanEqual(a, vec4(1.0))) +\n    vec4(greaterThanEqual(b, vec4(1.0))),\n    vec4(1.0));\n`;\n\nexport const MAX = `\n  vec4 result = vec4(max(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MIN = `\n  vec4 result = vec4(min(a, b));\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport const MOD = `\n  vec4 result = mod(a, b);\n  vec4 isNaN = vec4(equal(b, vec4(0.0)));\n  ` +\n    CHECK_NAN_SNIPPET + `\n  return result;\n`;\n\nexport class BinaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A', 'B'];\n  outputShape: number[];\n  userCode: string;\n  supportsBroadcasting = true;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      op: string, aShape: number[], bShape: number[],\n      checkOutOfBounds = false) {\n    this.outputShape = backend_util.assertAndGetBroadcastShape(aShape, bShape);\n    const rank = this.outputShape.length;\n    let checkOutOfBoundsString = '';\n    if (checkOutOfBounds) {\n      if (rank === 0 || util.sizeFromShape(this.outputShape) === 1) {\n        checkOutOfBoundsString = `\n          result.y = 0.;\n          result.z = 0.;\n          result.w = 0.;\n        `;\n      } else {\n        const dtype = getCoordsDataType(rank);\n        checkOutOfBoundsString = `\n          ${dtype} coords = getOutputCoords();\n        `;\n        if (rank === 1) {\n          checkOutOfBoundsString += `\n            result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;\n            result.z = 0.;\n            result.w = 0.;\n          `;\n        } else {\n          const channels = getChannels('coords', rank);\n          checkOutOfBoundsString += `\n            bool nextRowOutOfBounds =\n              (${channels[rank - 2]} + 1) >= ${this.outputShape[rank - 2]};\n            bool nextColOutOfBounds =\n              (${channels[rank - 1]} + 1) >= ${this.outputShape[rank - 1]};\n            result.y = nextColOutOfBounds ? 0. : result.y;\n            result.z = nextRowOutOfBounds ? 0. : result.z;\n            result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;\n          `;\n        }\n      }\n    }\n\n    this.userCode = `\n      vec4 binaryOperation(vec4 a, vec4 b) {\n        ${op}\n      }\n\n      void main() {\n        vec4 a = getAAtOutCoords();\n        vec4 b = getBAtOutCoords();\n\n        vec4 result = binaryOperation(a, b);\n        ${checkOutOfBoundsString}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        float value = getAAtOutCoords();\n        if (isnan(value)) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, minVal, maxVal));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ClipPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  userCode: string;\n  outputShape: number[];\n\n  // Caching uniform locations for speed.\n  minLoc: WebGLUniformLocation;\n  maxLoc: WebGLUniformLocation;\n\n  constructor(aShape: number[]) {\n    this.outputShape = aShape;\n    this.userCode = `\n      uniform float minVal;\n      uniform float maxVal;\n\n      void main() {\n        vec4 value = getAAtOutCoords();\n\n        if (any(isnan(value))) {\n          setOutput(value);\n          return;\n        }\n\n        setOutput(clamp(value, vec4(minVal), vec4(maxVal)));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(min: number, max: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.minLoc == null) {\n        this.minLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'minVal');\n        this.maxLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'maxVal');\n      }\n      gpgpu.gl.uniform1f(this.minLoc, min);\n      gpgpu.gl.uniform1f(this.maxLoc, max);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ComplexAbsProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(shape: number[]) {\n    this.outputShape = shape;\n    this.userCode = `\n      void main() {\n        float re = abs(getRealAtOutCoords());\n        float im = abs(getImagAtOutCoords());\n        float mx = max(re, im);\n\n        // sadly the length function in glsl is not underflow-safe\n        // (at least not on Intel GPUs). So the safe solution is\n        // to ensure underflow-safety in all cases.\n        setOutput(\n          mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))\n        );\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ConcatProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  // Concats 2d tensors along axis=1. See comments in MathBackendWebGL.concat().\n  constructor(shapes: Array<[number, number]>) {\n    this.outputShape = backend_util.computeOutShape(shapes, 1 /* axis */);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][1];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][1];\n    }\n\n    const snippets = [`if (yC < ${offsets[0]}) setOutput(getT0(yR, yC));`];\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      snippets.push(\n          `else if (yC < ${offsets[i]}) ` +\n          `setOutput(getT${i}(yR, yC-${shift}));`);\n    }\n    const lastIndex = offsets.length;\n    const lastShift = offsets[offsets.length - 1];\n    snippets.push(`else setOutput(getT${lastIndex}(yR, yC-${lastShift}));`);\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int yR = coords.x;\n        int yC = coords.y;\n\n        ${snippets.join('\\n        ')}\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ConcatPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(shapes: number[][], axis: number) {\n    this.outputShape = backend_util.computeOutShape(shapes, axis);\n    const shape = this.outputShape;\n    const rank = shape.length;\n    const dtype = getCoordsDataType(rank);\n    const coords = getChannels('coords', rank);\n    const channels = ['x', 'y', 'z', 'w', 'u', 'v'].slice(0, rank);\n    this.variableNames = shapes.map((_, i) => `T${i}`);\n\n    const offsets: number[] = new Array(shapes.length - 1);\n    offsets[0] = shapes[0][axis];\n    for (let i = 1; i < offsets.length; i++) {\n      offsets[i] = offsets[i - 1] + shapes[i][axis];\n    }\n\n    const channel = channels[axis];\n    const lastChannels = channels.slice(-2);\n    const allChannels = channels.join();\n\n    let getValueSnippet = `if (${channel} < ${offsets[0]}) {\n        return getChannel(\n            getT0(${allChannels}), vec2(${lastChannels.join()}));\n        }`;\n    for (let i = 1; i < offsets.length; i++) {\n      const shift = offsets[i - 1];\n      // Note: the >= comparison below may seem unnecessary given the check\n      // above but is needed to workaround branch execution issues on some\n      // devices. It makes all the conditions exclusive without relying on\n      // execution order.\n      getValueSnippet += `\n        if (${channel} < ${offsets[i]}  && ${channel} >= ${offsets[i - 1]}) {\n          return getChannel(\n            getT${i}(${shiftedChannels(channels, channel, shift)}),\n            vec2(${shiftedChannels(lastChannels, channel, shift)}));\n        }`;\n    }\n    const lastIndex = offsets.length;\n    const shift = offsets[offsets.length - 1];\n    getValueSnippet += `\n        return getChannel(\n          getT${lastIndex}(${shiftedChannels(channels, channel, shift)}),\n          vec2(${shiftedChannels(lastChannels, channel, shift)}));`;\n\n    this.userCode = `\n      float getValue(${channels.map(x => 'int ' + x)}) {\n        ${getValueSnippet}\n      }\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        vec4 result = vec4(getValue(${coords}), 0., 0., 0.);\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} + 1;\n        if (${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.g = getValue(${coords});\n        }\n\n        ${coords[rank - 2]} = ${coords[rank - 2]} + 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]}) {\n          result.a = getValue(${coords});\n        }\n\n        ${coords[rank - 1]} = ${coords[rank - 1]} - 1;\n        if (${coords[rank - 2]} < ${shape[rank - 2]} &&\n            ${coords[rank - 1]} < ${shape[rank - 1]}) {\n          result.b = getValue(${coords});\n        }\n        setOutput(result);\n      }\n    `;\n  }\n}\n\n/**\n * Return an expression for coordinates into a vector where a given channel\n * will be offset by [shift].\n *\n * @param channels the channels to consider\n * @param channel the channel we want shifted\n * @param shift  the amount to subtract from the channel.\n *\n * @returns a string of the form 'x, y-[shift], z' where any one channel can\n * have the shift applied.\n */\nfunction shiftedChannels(channels: string[], channel: string, shift: number) {\n  const channelIdx = channels.indexOf(channel);\n  const res = channels.map((c, idx) => {\n    if (idx === channelIdx) {\n      return `${c} - ${shift}`;\n    } else {\n      return c;\n    }\n  });\n  return res.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int d2 = coords.w;\n\n        // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              if (${isChannelsLast}) {\n                float dyValue = getDy(b, yR, yC, d2);\n                float xValue = getX(b, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              } else {\n                float dyValue = getDy(b, d2, yR, yC);\n                float xValue = getX(b, d1, xR, xC);\n                dotProd += (xValue * dyValue);\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[${channelDim}];\n\n        ivec2 dyCorner = ivec2(coords[${rowDim}], coords[${colDim}]) - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n\n              if (${isChannelsLast}) {\n                float xValue = getDy(batch, idyR, idyC, d2);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              } else {\n                float xValue = getDy(batch, d2, idyR, idyC);\n                float wValue = getW(wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n\n    this.userCode = `\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int wF = coords.x;\n        int wR = coords.y;\n        int wC = coords.z;\n        int d1 = coords.w;\n        int d2 = coords.u;\n\n        float dotProd = 0.0;\n\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yF = 0; yF < ${convInfo.outDepth}; yF++) {\n            int xF = wF + yF * ${strideDepth} - ${padFront};\n\n            if (xF < 0 || xF >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n              int xR = wR + yR * ${strideHeight} - ${padTop};\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n                int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float dyValue = getDy(b, yF, yR, yC, d2);\n                float xValue = getX(b, xF, xR, xC, d1);\n                dotProd += (xValue * dyValue);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class Conv3DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padFront = filterDepth - 1 - convInfo.padInfo.front;\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d1 = coords.u;\n\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyFCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          float dyF = float(dyFCorner + wF) / ${strideDepth}.0;\n\n          if (dyF < 0.0 || dyF >= ${convInfo.outDepth}.0 || fract(dyF) > 0.0) {\n            continue;\n          }\n          int idyF = int(dyF);\n\n          int wFPerm = ${filterDepth} - 1 - wF;\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n              fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            int wRPerm = ${filterHeight} - 1 - wR;\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              int wCPerm = ${filterWidth} - 1 - wC;\n\n              for (int d2 = 0; d2 < ${convInfo.outChannels}; d2++) {\n                float xValue = getDy(batch, idyF, idyR, idyC, d2);\n                float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);\n                dotProd += xValue * wValue;\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DDerFilterProgram implements GPGPUProgram {\n  variableNames = ['x', 'dy'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.filterShape;\n\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int wR = coords.x;\n        int wC = coords.y;\n        int d1 = coords.z;\n        int dm = coords.w;\n        int d2 = d1 * ${channelMul} + dm;\n\n        float dotProd = 0.0;\n\n        // TO DO: Vec4 over the batch size\n        for (int b = 0; b < ${convInfo.batchSize}; b++) {\n          for (int yR = 0; yR < ${convInfo.outHeight}; yR++) {\n            int xR = wR + yR * ${strideHeight} - ${padTop};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int yC = 0; yC < ${convInfo.outWidth}; yC++) {\n              int xC = wC + yC * ${strideWidth} - ${padLeft};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float dyValue = getDy(b, yR, yC, d2);\n              float xValue = getX(b, xR, xC, d1);\n              dotProd += (xValue * dyValue);\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class DepthwiseConv2DDerInputProgram implements GPGPUProgram {\n  variableNames = ['dy', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n\n    const padTop = filterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = filterWidth - 1 - convInfo.padInfo.left;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d1 = coords[3];\n        ivec2 dyCorner = coords.yz - pads;\n        int dyRCorner = dyCorner.x;\n        int dyCCorner = dyCorner.y;\n\n        float dotProd = 0.0;\n\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          int wRPerm = ${filterHeight} - 1 - wR;\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            int wCPerm = ${filterWidth} - 1 - wC;\n\n            // TO DO: Vec4 over the channelMul\n            for (int dm = 0; dm < ${channelMul}; dm++) {\n              int d2 = d1 * ${channelMul} + dm;\n              float xValue = getDy(batch, idyR, idyC, d2);\n              float wValue = getW(wRPerm, wCPerm, d1, dm);\n              dotProd += xValue * wValue;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Conv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivationWeights = false) {\n    this.outputShape = convInfo.outShape;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n\n    const rowDim = isChannelsLast ? 1 : 2;\n    const colDim = isChannelsLast ? 2 : 3;\n    const channelDim = isChannelsLast ? 3 : 1;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivationWeights) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivationWeights) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d2 = coords[${channelDim}];\n\n        ivec2 xRCCorner =\n            ivec2(coords[${rowDim}], coords[${colDim}]) * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${convInfo.inWidth}) {\n              continue;\n            }\n\n            for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n              vec4 wValues = vec4(\n                getW(wR, wC, d1, d2),\n                getW(wR, wC, d1 + 1, d2),\n                getW(wR, wC, d1 + 2, d2),\n                getW(wR, wC, d1 + 3, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec4 xValues = vec4(\n                  getX(batch, xR, xC, d1),\n                  getX(batch, xR, xC, d1 + 1),\n                  getX(batch, xR, xC, d1 + 2),\n                  getX(batch, xR, xC, d1 + 3)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec4 xValues = vec4(\n                  getX(batch, d1, xR, xC),\n                  getX(batch, d1 + 1, xR, xC),\n                  getX(batch, d1 + 2, xR, xC),\n                  getX(batch, d1 + 3, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n\n            if (${inputDepthVec4Remainder === 1}) {\n\n              if (${isChannelsLast}) {\n                dotProd +=\n                    getX(batch, xR, xC, ${inputDepthNearestVec4}) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              } else {\n                dotProd +=\n                    getX(batch, ${inputDepthNearestVec4}, xR, xC) *\n                    getW(wR, wC, ${inputDepthNearestVec4}, d2);\n              }\n\n            } else if (${inputDepthVec4Remainder === 2}) {\n              vec2 wValues = vec2(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec2 xValues = vec2(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec2 xValues = vec2(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            } else if (${inputDepthVec4Remainder === 3}) {\n              vec3 wValues = vec3(\n                getW(wR, wC, ${inputDepthNearestVec4}, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                getW(wR, wC, ${inputDepthNearestVec4} + 2, d2)\n              );\n\n              if (${isChannelsLast}) {\n                vec3 xValues = vec3(\n                  getX(batch, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else {\n                vec3 xValues = vec3(\n                  getX(batch, ${inputDepthNearestVec4}, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 1, xR, xC),\n                  getX(batch, ${inputDepthNearestVec4} + 2, xR, xC)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n\n            }\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nexport class Conv3DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.outShape;\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterDepth = convInfo.filterDepth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n\n    const inputDepthNearestVec4 = Math.floor(convInfo.inChannels / 4) * 4;\n    const inputDepthVec4Remainder = convInfo.inChannels % 4;\n\n    this.userCode = `\n      const ivec3 strides = ivec3(${strideDepth}, ${strideHeight}, ${\n        strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int d2 = coords.u;\n\n        ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xFCorner = xFRCCorner.x;\n        int xRCorner = xFRCCorner.y;\n        int xCCorner = xFRCCorner.z;\n\n        // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get\n        // y(yF, yR, yC, d2). ? = to be determined. : = across all\n        // values in that axis.\n        float dotProd = 0.0;\n        for (int wF = 0; wF < ${filterDepth}; wF++) {\n          int xF = xFCorner + wF * ${dilationDepth};\n\n          if (xF < 0 || xF >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${filterHeight}; wR++) {\n            int xR = xRCorner + wR * ${dilationHeight};\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidth}; wC++) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              for (int d1 = 0; d1 < ${inputDepthNearestVec4}; d1 += 4) {\n                vec4 xValues = vec4(\n                  getX(batch, xF, xR, xC, d1),\n                  getX(batch, xF, xR, xC, d1 + 1),\n                  getX(batch, xF, xR, xC, d1 + 2),\n                  getX(batch, xF, xR, xC, d1 + 3)\n                );\n                vec4 wValues = vec4(\n                  getW(wF, wR, wC, d1, d2),\n                  getW(wF, wR, wC, d1 + 1, d2),\n                  getW(wF, wR, wC, d1 + 2, d2),\n                  getW(wF, wR, wC, d1 + 3, d2)\n                );\n\n                dotProd += dot(xValues, wValues);\n              }\n\n              if (${inputDepthVec4Remainder === 1}) {\n                dotProd +=\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}) *\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2);\n              } else if (${inputDepthVec4Remainder === 2}) {\n                vec2 xValues = vec2(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1)\n                );\n                vec2 wValues = vec2(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              } else if (${inputDepthVec4Remainder === 3}) {\n                vec3 xValues = vec3(\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4}),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 1),\n                  getX(batch, xF, xR, xC, ${inputDepthNearestVec4} + 2)\n                );\n                vec3 wValues = vec3(\n                  getW(wF, wR, wC, ${inputDepthNearestVec4}, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 1, d2),\n                  getW(wF, wR, wC, ${inputDepthNearestVec4} + 2, d2)\n                );\n                dotProd += dot(xValues, wValues);\n              }\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConv2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const channelMul = convInfo.outChannels / convInfo.inChannels;\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `float activation(float a) {\n          float b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `\n          float activation(float x) {\n            ${activation}\n          }\n        `;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2 / ${channelMul};\n        int q = d2 - d1 * ${channelMul};\n\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.\n        for (int wR = 0; wR < ${filterHeight}; wR++) {\n          int xR = xRCorner + wR * ${dilationHeight};\n\n          if (xR < 0 || xR >= ${xNumRows}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidth}; wC++) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            if (xC < 0 || xC >= ${xNumCols}) {\n              continue;\n            }\n\n            float xVal = getX(batch, xR, xC, d1);\n            float wVal = getW(wR, wC, d1, q);\n            dotProd += xVal * wVal;\n          }\n        }\n\n        float result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthwiseConvPacked2DProgram implements GPGPUProgram {\n  variableNames = ['x', 'W'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = convInfo.outShape;\n\n    const xNumRows = convInfo.inHeight;\n    const xNumCols = convInfo.inWidth;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const filterHeight = convInfo.filterHeight;\n    const filterWidth = convInfo.filterWidth;\n    const texelsAcross = filterWidth;\n\n    let mainLoop = `int xR; int xC; int xCOffset;`;\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `\n          vec4 xTexelR${r}C${c * 2} = vec4(0.);\n          vec4 wR${r}C${c} = vec4(0.);\n          vec4 xR${r}C${c} = vec4(0.);`;\n      }\n    }\n\n    /**\n     * This vectorized implementation works by gathering the values needed for\n     * each output channel's dot product into vec4's and then multiplying them\n     * all together (this happens in the final double for-loop below). Most of\n     * the main loop consists of constructing these vec4's with the minimum\n     * number of texture2D calls, which means making use of all four returned\n     * values from a texture2D call at once.\n     */\n    for (let r = 0; r < filterHeight; r++) {\n      for (let texelC = 0; texelC < texelsAcross; texelC++) {\n        const c = texelC * 2;\n\n        mainLoop += `\n          xR = xRCorner + ${r * dilationHeight};\n          xC = xCCorner + ${c * dilationWidth};\n        `;\n\n        if (strideWidth === 1) {\n          if (c < filterWidth) {\n            // If padding is odd, the outer texels have to be composed.\n            if (padLeft % 2 === 1) {\n              // TODO: Ensure vec4 previous does not result in redundant sample,\n              // and avoid setting xTexelRC's that exceed the boundary in the\n              // first place rather than resetting them to vec4(0)).\n\n              // To compute xCOffset:\n              // - If padding is odd, we must add 1 to ensure we ask for an\n              // even-numbered row.\n              // - We subtract 2 to access the previous texel.\n\n              mainLoop += `\n                xCOffset = xC + 1;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    xTexelR${r}C${c}.zw = vec2(0.);\n                  }\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + 1 - 2;\n                if(xR >= 0 && xR < ${xNumRows} && xCOffset >= 0 && xCOffset < ${\n                  xNumCols}) {\n                  vec4 previous = getX(batch, xR, xCOffset, d1);\n\n                  // Need to manually clear unused channels in case\n                  // we're reading from recycled texture.\n                  if(xCOffset + 1 >= ${xNumCols}) {\n                    previous.zw = vec2(0.);\n                  }\n\n                  xR${r}C${c} = vec4(previous.zw, xTexelR${r}C${c}.xy);\n                } else {\n                  xR${r}C${c} = vec4(0, 0, xTexelR${r}C${c}.xy);\n                }\n              `;\n            } else {\n              // Padding is even, so xRC corresponds to a single texel.\n              mainLoop += `\n                if(xR >= 0 && xR < ${xNumRows} && xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xR${r}C${c} = xTexelR${r}C${c};\n              `;\n            }\n\n            if (c + 1 < filterWidth) {\n              // If dilation is even, the second entry should match the first\n              // (either both are composed or both are single samples). But if\n              // dilation is odd, then the second entry should be the opposite\n              // of the first (if the first is composed, the second is a single\n              // sample, and vice versa.)\n\n              const nextTexelOffset = padLeft % 2 === 0 ?\n                  util.nearestLargerEven(dilationWidth) :\n                  dilationWidth;\n\n              if ((dilationWidth % 2 === 0 && padLeft % 2 === 1) ||\n                  (dilationWidth % 2 !== 0 && padLeft % 2 !== 1)) {\n                mainLoop += `\n                  xCOffset = xC + ${padLeft % 2} + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n                `;\n\n                // If dilation > 1 then the xRC's will not be able to share any\n                // values, so each xRC will require two unique calls to getX.\n                if (dilationWidth > 1) {\n                  mainLoop += `\n                    xCOffset -= 2;\n                    if(xR >= 0 && xR < ${xNumRows} &&\n                      xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                      xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                    } else {\n                      xTexelR${r}C${c} = vec4(0.);\n                    }\n                  `;\n                }\n\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.xy);\n                `;\n              } else {\n                mainLoop += `\n                  xCOffset = xC + ${nextTexelOffset};\n\n                  if(xR >= 0 && xR < ${xNumRows} &&\n                    xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                  }\n\n                  xR${r}C${c + 1} = xTexelR${r}C${c + 2};\n                `;\n              }\n            }\n          }\n        } else {  // stride > 1\n          if (c < filterWidth) {\n            mainLoop += `\n              if(xR >= 0 && xR < ${xNumRows}) {\n            `;\n\n            // Depending on whether padLeft is even or odd, we want either the\n            // xy or zw channels from X texels for xR${r}C${c}. If padLeft is\n            // even, xR${r}C${c + 1} is simply the zw channels of texels we've\n            // already sampled. But if padLeft is odd, xR${r}C{$c + 1}.zw will\n            // need to come from the xy channels of a new texel, hence the `vec4\n            // final` initialized below.\n            if (padLeft % 2 === 1) {\n              mainLoop += `\n                xCOffset = xC + 1 - ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                if(xC + 1 >= 0 && xC + 1 < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xC + 1, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  vec4 final = vec4(0.);\n                  xCOffset = xC + 1 + ${strideWidth};\n                  if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                    final = getX(batch, xR, xCOffset, d1);\n                  }\n                  xR${r}C${c + 1} = vec4(xTexelR${r}C${c + 2}.xy, final.xy);\n                `;\n              }\n            } else {\n              mainLoop += `\n                if(xC >= 0 && xC < ${xNumCols}) {\n                  xTexelR${r}C${c} = getX(batch, xR, xC, d1);\n                } else {\n                  xTexelR${r}C${c} = vec4(0.);\n                }\n\n                xCOffset = xC + ${strideWidth};\n                if(xCOffset >= 0 && xCOffset < ${xNumCols}) {\n                  xTexelR${r}C${c + 2} = getX(batch, xR, xCOffset, d1);\n                } else {\n                  xTexelR${r}C${c + 2} = vec4(0.);\n                }\n\n                xR${r}C${c} = vec4(\n                  xTexelR${r}C${c}.xy, xTexelR${r}C${c + 2}.xy);\n              `;\n\n              if (c + 1 < filterWidth) {\n                mainLoop += `\n                  xR${r}C${c + 1} = vec4(\n                    xTexelR${r}C${c}.zw, xTexelR${r}C${c + 2}.zw);\n                `;\n              }\n            }\n\n            mainLoop += `}`;\n          }\n        }\n\n        if (c < filterWidth) {\n          mainLoop += `\n            vec4 wTexelR${r}C${c} = getW(${r}, ${c}, d1, q);\n            wR${r}C${c} = vec4(wTexelR${r}C${c}.xz, wTexelR${r}C${c}.xz);\n          `;\n\n          if (c + 1 < filterWidth) {\n            mainLoop += `\n              vec4 wTexelR${r}C${c + 1} = getW(${r}, ${c + 1}, d1, q);\n              wR${r}C${c + 1} =\n                vec4(wTexelR${r}C${c + 1}.xz, wTexelR${r}C${c + 1}.xz);`;\n          }\n        }\n      }\n    }\n\n    for (let r = 0; r < filterHeight; r++) {\n      for (let c = 0; c < filterWidth; c++) {\n        mainLoop += `dotProd += xR${r}C${c} * wR${r}C${c};`;\n      }\n    }\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n\n        ivec4 coords = getOutputCoords();\n        int batch = coords.x;\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int d2 = coords.w;\n        int d1 = d2;\n        int q = 0;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        vec4 dotProd = vec4(0.);\n\n        ${mainLoop}\n\n        vec4 result = dotProd;\n        ${addBiasSnippet}\n        ${applyActivationSnippet}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class CropAndResizeProgram implements GPGPUProgram {\n  variableNames = ['Image', 'Boxes', 'BoxInd'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], boxShape: [number, number],\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number) {\n    const [batch, imageHeight, imageWidth, depth] = imageShape;\n    const [numBoxes, ] = boxShape;\n    const [cropHeight, cropWidth] = cropSize;\n    this.outputShape = [numBoxes, cropHeight, cropWidth, depth];\n    const methodId = method === 'bilinear' ? 1 : 0;\n\n    const [inputHeightFloat, inputWidthFloat] =\n        [`${imageHeight - 1}.0`, `${imageWidth - 1}.0`];\n\n    const [heightRatio, heightScale, inY] = cropHeight > 1 ?\n        [\n          `${(imageHeight - 1) / (cropHeight - 1)}`,\n          '(y2-y1) * height_ratio',\n          `y1*${inputHeightFloat} + float(y)*(height_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (y1+y2) * ${inputHeightFloat}`,\n        ];\n    const [widthRatio, widthScale, inX] = cropWidth > 1 ?\n        [\n          `${(imageWidth - 1) / (cropWidth - 1)}`,\n          '(x2-x1) * width_ratio',\n          `x1*${inputWidthFloat} + float(x)*(width_scale)`,\n        ] :\n        [\n          '0.0',\n          '0.0',\n          `0.5 * (x1+x2) * ${inputWidthFloat}`,\n        ];\n\n    // Reference implementation\n    // tslint:disable-next-line:max-line-length\n    // https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc\n    this.userCode = `\n      const float height_ratio = float(${heightRatio});\n      const float width_ratio = float(${widthRatio});\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int y = coords[1];\n        int x = coords[2];\n        int d = coords[3];\n\n        // get box vals\n        float y1 = getBoxes(b,0);\n        float x1 = getBoxes(b,1);\n        float y2 = getBoxes(b,2);\n        float x2 = getBoxes(b,3);\n\n        // get image in batch index\n        int bInd = round(getBoxInd(b));\n        if(bInd < 0 || bInd >= ${batch}) {\n          return;\n        }\n\n        float height_scale = ${heightScale};\n        float width_scale = ${widthScale};\n\n        float in_y = ${inY};\n        if( in_y < 0.0 || in_y > ${inputHeightFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n        float in_x = ${inX};\n        if( in_x < 0.0 || in_x > ${inputWidthFloat} ) {\n          setOutput(float(${extrapolationValue}));\n          return;\n        }\n\n        vec2 sourceFracIndexCR = vec2(in_x,in_y);\n        if(${methodId} == 1) {\n          // Compute the four integer indices.\n          ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);\n          ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));\n\n          float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);\n          float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);\n          float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);\n          float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);\n\n          vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);\n\n          float top = topLeft + (topRight - topLeft) * fracCR.x;\n          float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;\n          float newValue = top + (bottom - top) * fracCR.y;\n          setOutput(newValue);\n        } else {\n          // Compute the coordinators of nearest neighbor point.\n          ivec2 sourceNearestCR = ivec2(floor(\n            sourceFracIndexCR + vec2(0.5,0.5)));\n          float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);\n          setOutput(newValue);\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class CumSumProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  index: WebGLUniformLocation;\n\n  constructor(shape: number[], exclusive: boolean, reverse: boolean) {\n    this.outputShape = shape;\n    const rank = shape.length;\n    const val = exclusive ? '0.0' : `getX(${getCoords(rank, 'coords')})`;\n    const length = shape[shape.length - 1];\n    let condition = '';\n    let idxString = '';\n    // When exclusive is set, the cumsum op becomes roll op that copies the\n    // value from the previous index based on the direction specified by the\n    // reverse flag.\n    if (exclusive) {\n      condition = reverse ? `end != ${length - 1}` : 'end != 0';\n      idxString = reverse ? 'end + 1' : 'end - 1';\n    } else {\n      condition = reverse ? `end + pow2 < ${length}` : 'end >= pow2';\n      idxString = (reverse ? 'end + pow2' : 'end - pow2');\n    }\n\n    this.userCode = `\n      uniform float index;\n      void main() {\n        ${getCoordsDataType(rank)} coords = getOutputCoords();\n        int end = ${getFinalCoord(rank, 'coords')};\n        float val = ${val};\n        int pow2 = int(pow(2.0, index));\n        if (${condition}) {\n          int idx = ${idxString};\n          ${getFinalCoord(rank, 'coords')} = idx;\n          val += getX(${getCoords(rank, 'coords')});\n        }\n        setOutput(val);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(index: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.index == null) {\n        this.index = gpgpu.getUniformLocation(webGLProgram, 'index');\n      }\n      gpgpu.gl.uniform1f(this.index, index);\n    };\n  }\n}\n\nfunction getCoords(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.x, ${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.x, ${name}.y, ${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.x, ${name}.y, ${name}.z, ${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n\nfunction getFinalCoord(rank: number, name: string): string {\n  if (rank === 1) {\n    return `${name}`;\n  } else if (rank === 2) {\n    return `${name}.y`;\n  } else if (rank === 3) {\n    return `${name}.z`;\n  } else if (rank === 4) {\n    return `${name}.w`;\n  } else {\n    throw Error(`Cumulative sum for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: [number, number, number];\n  packedInputs = false;\n  packedOutput = true;\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getA(rc.x, rc.y, rc.z);\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\nimport {getDenseTexShape, PackingScheme} from './tex_util';\n\nexport class DecodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: [number, number, number];\n  outPackingScheme = PackingScheme.DENSE;\n\n  constructor(outputShape: [number, number, number]) {\n    const texShape = getDenseTexShape(outputShape);\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n\n    this.userCode = `\n      ivec3 outCoordsFromFlatIndex(int index) {\n        ${\n        shader_util.getLogicalCoordinatesFromFlatIndex(\n            ['r', 'c', 'd'], outputShape)}\n        return ivec3(r, c, d);\n      }\n\n      void main() {\n        ivec2 resTexRC = ivec2(resultUV.yx *\n          vec2(${texShape[0]}, ${texShape[1]}));\n        int index = 4 * (resTexRC.x * ${texShape[1]} + resTexRC.y);\n\n        vec4 result = vec4(0.);\n\n        for (int i=0; i<4; i++) {\n          int flatIndex = index + i;\n          ivec3 rc = outCoordsFromFlatIndex(flatIndex);\n          result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DepthToSpaceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  blockSize: number;\n  dataFormat: string;\n\n  constructor(\n      outputShape: number[], blockSize: number, dataFormat: 'NHWC'|'NCHW') {\n    this.outputShape = outputShape;\n    this.blockSize = blockSize;\n    this.dataFormat = dataFormat;\n    this.userCode = `\n    void main() {\n      ivec4 coords = getOutputCoords();\n      int b = coords[0];\n      int h = ${this.getHeightCoordString()};\n      int w = ${this.getWidthCoordString()};\n      int d = ${this.getDepthCoordString()};\n\n      int in_h = h / ${blockSize};\n      int offset_h = imod(h, ${blockSize});\n      int in_w = w / ${blockSize};\n      int offset_w = imod(w, ${blockSize});\n      int offset_d = (offset_h * ${blockSize} + offset_w) *\n        ${this.getOutputDepthSize()};\n      int in_d = d + offset_d;\n\n      float result = ${this.getInputSamplingString()};\n      setOutput(result);\n    }\n  `;\n  }\n\n  private getHeightCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[1]`;\n    } else {\n      return `coords[2]`;\n    }\n  }\n\n  private getWidthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[2]`;\n    } else {\n      return `coords[3]`;\n    }\n  }\n\n  private getDepthCoordString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `coords[3]`;\n    } else {\n      return `coords[1]`;\n    }\n  }\n\n  private getOutputDepthSize(): number {\n    if (this.dataFormat === 'NHWC') {\n      return this.outputShape[3];\n    } else {\n      return this.outputShape[1];\n    }\n  }\n\n  private getInputSamplingString(): string {\n    if (this.dataFormat === 'NHWC') {\n      return `getX(b, in_h, in_w, in_d)`;\n    } else {\n      return `getX(b, in_d, in_h, in_w)`;\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class DiagProgram implements GPGPUProgram {\n  variableNames = ['X'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(size: number) {\n    this.outputShape = [size, size];\n    this.userCode = `\n      void main() {\n          ivec2 coords = getOutputCoords();\n          float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;\n          setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        float x = getAAtOutCoords();\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {ENCODE_FLOAT_SNIPPET} from './shader_compiler_util';\nimport {TextureUsage} from './tex_util';\n\nexport class EncodeFloatPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = false;\n  outTexUsage = TextureUsage.DOWNLOAD;\n\n  constructor(outputShape: [number, number, number]) {\n    const glsl = getGlslDifferences();\n    this.outputShape = outputShape;\n    this.userCode = `\n      ${ENCODE_FLOAT_SNIPPET}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n        float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));\n        ${glsl.output} = encode_float(x);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class EncodeMatrixProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let output = `result`;\n    if (inputIsUnsignedByte) {\n      output = `floor(result * 255. + 0.5)`;\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        int flatIndex = getFlatIndex(coords);\n        int offset = imod(flatIndex, 4);\n\n        flatIndex = idiv(flatIndex, 4, 1.);\n\n        int r = flatIndex / ${width};\n        int c = imod(flatIndex, ${width});\n        vec2 uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n        vec4 values = ${glsl.texture2D}(A, uv);\n\n        float result;\n\n        if(offset == 0) {\n          result = values[0];\n        } else if(offset == 1) {\n          result = values[1];\n        } else if(offset == 2) {\n          result = values[2];\n        } else {\n          result = values[3];\n        }\n\n        ${glsl.output} = vec4(${output}, 0., 0., 0.);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\n/*\nThis is how the shader encodes a tensor with shape = [2, 3, 5]\n(indices are [batch, row, col]).\n\n000|001   002|003   004|xxx   020|021   022|023   024|xxx\n-------   -------   -------   -------   -------   -------\n010|011   012|013   014|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\n100|101   102|103   104|xxx   120|121   122|123   124|xxx\n-------   -------   -------   -------   -------   -------\n110|111   112|113   114|xxx   xxx|xxx   xxx|xxx   xxx|xxx\n\nSingle texels contain only values from the same batch, and from adjacent rows\nand columns.\n */\n\nexport class EncodeMatrixPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape: [number, number, number], texShape: [number, number],\n      inputIsUnsignedByte = false) {\n    const glsl = getGlslDifferences();\n    const [height, width] = texShape;\n    this.outputShape = outputShape;\n\n    let mainLoop = '';\n    let output = 'result';\n    if (inputIsUnsignedByte) {\n      output = 'floor(result * 255. + 0.5)';\n    }\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        const channel = row * 2 + col;\n\n        mainLoop += `\n          localCoords = coords;\n          if(localCoords[2] + ${col} < ${outputShape[2]}) {\n            localCoords[2] += ${col};\n            if(localCoords[1] + ${row} < ${outputShape[1]}) {\n              localCoords[1] += ${row};\n\n              flatIndex = getFlatIndex(localCoords);\n              offset = imod(flatIndex, 4);\n\n              flatIndex = idiv(flatIndex, 4, 1.);\n\n              r = flatIndex / ${width};\n              c = imod(flatIndex, ${width});\n              uv = (vec2(c, r) + halfCR) / vec2(${width}.0, ${height}.0);\n              values = ${glsl.texture2D}(A, uv);\n\n              if(offset == 0) {\n                result[${channel}] = values[0];\n              } else if(offset == 1) {\n                result[${channel}] = values[1];\n              } else if(offset == 2) {\n                result[${channel}] = values[2];\n              } else {\n                result[${channel}] = values[3];\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 coords = getOutputCoords();\n\n        vec4 result = vec4(0.);\n        int flatIndex, r, c, offset;\n        ivec3 localCoords;\n        vec2 uv;\n        vec4 values;\n\n        ${mainLoop}\n\n        ${glsl.output} = ${output};\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const COMPLEX_FFT = {\n  REAL: 'return real * expR - imag * expI;',\n  IMAG: 'return real * expI + imag * expR;'\n};\n\nexport class FFTProgram implements GPGPUProgram {\n  variableNames = ['real', 'imag'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(op: string, inputShape: [number, number], inverse: boolean) {\n    const innerDim = inputShape[1];\n    this.outputShape = inputShape;\n\n    const exponentMultiplierSnippet =\n        inverse ? `2.0 * ${Math.PI}` : `-2.0 * ${Math.PI}`;\n    const resultDenominator = inverse ? `${innerDim}.0` : '1.0';\n\n    this.userCode = `\n      const float exponentMultiplier = ${exponentMultiplierSnippet};\n\n      float unaryOpComplex(float real, float expR, float imag, float expI) {\n        ${op}\n      }\n\n      float mulMatDFT(int batch, int index) {\n        float indexRatio = float(index) / float(${innerDim});\n        float exponentMultiplierTimesIndexRatio =\n            exponentMultiplier * indexRatio;\n\n        float result = 0.0;\n\n        for (int i = 0; i < ${innerDim}; i++) {\n          // x = (-2|2 * PI / N) * index * i;\n          float x = exponentMultiplierTimesIndexRatio * float(i);\n          float expR = cos(x);\n          float expI = sin(x);\n          float real = getReal(batch, i);\n          float imag = getImag(batch, i);\n\n          result +=\n              unaryOpComplex(real, expR, imag, expI) / ${resultDenominator};\n        }\n\n        return result;\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        setOutput(mulMatDFT(coords[0], coords[1]));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FillProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  valueLoc: WebGLUniformLocation;\n\n  constructor(shape: number[], value: number) {\n    this.variableNames = ['x'];\n    this.outputShape = shape;\n\n    this.userCode = `\n      uniform float value;\n      void main() {\n        // Input can be obtained from uniform value.\n        setOutput(value);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(value: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.valueLoc == null) {\n        this.valueLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'value');\n      }\n      gpgpu.gl.uniform1f(this.valueLoc, value);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherProgram implements GPGPUProgram {\n  variableNames = ['A', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], indicesLength: number, axis: number) {\n    const outputShape: number[] = aShape.slice();\n    outputShape[axis] = indicesLength;\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape, axis);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[], axis: number): string {\n  const rank = aShape.length;\n  if (rank > 4) {\n    throw Error(`Gather for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `int(getIndices(resRC))`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    if (i === axis) {\n      sourceCoords.push(`int(getIndices(${currentCoords[i]}))`);\n    } else {\n      sourceCoords.push(`${currentCoords[i]}`);\n    }\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class GatherNDProgram implements GPGPUProgram {\n  variableNames = ['x', 'indices'];\n  outputShape: number[];\n  userCode: string;\n  constructor(\n      private sliceDim: number, private strides: number[], shape: number[]) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    const strideString = this.sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${this.strides});\n         void main() {\n          ${dtype} coords = getOutputCoords();\n          int flattenIndex = 0;\n          for (int j = 0; j < ${this.sliceDim}; j++) {\n            int index = round(getIndices(coords[0], j));\n            flattenIndex += index * ${strideString};\n          }\n          setOutput(getX(flattenIndex, coords[1]));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {PixelData, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {getGlslDifferences} from './glsl_version';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport * as webgl_util from './webgl_util';\n\nexport function createVertexShader(gl: WebGLRenderingContext): WebGLShader {\n  const glsl = getGlslDifferences();\n  const vertexShaderSource = `${glsl.version}\n    precision highp float;\n    ${glsl.attribute} vec3 clipSpacePos;\n    ${glsl.attribute} vec2 uv;\n    ${glsl.varyingVs} vec2 resultUV;\n\n    void main() {\n      gl_Position = vec4(clipSpacePos, 1);\n      resultUV = uv;\n    }`;\n  return webgl_util.createVertexShader(gl, vertexShaderSource);\n}\n\nexport function createVertexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // [x y z u v] * [upper-left, lower-left, upper-right, lower-right]\n  const vertexArray = new Float32Array(\n      [-1, 1, 0, 0, 1, -1, -1, 0, 0, 0, 1, 1, 0, 1, 1, 1, -1, 0, 1, 0]);\n  return webgl_util.createStaticVertexBuffer(gl, vertexArray);\n}\n\nexport function createIndexBuffer(gl: WebGLRenderingContext): WebGLBuffer {\n  // OpenGL (and WebGL) have \"CCW == front\" winding\n  const triangleVertexIndices = new Uint16Array([0, 1, 2, 2, 1, 3]);\n  return webgl_util.createStaticIndexBuffer(gl, triangleVertexIndices);\n}\n\nfunction createAndConfigureTexture(\n    gl: WebGLRenderingContext, width: number, height: number,\n    internalFormat: number, textureFormat: number,\n    textureType: number): WebGLTexture {\n  webgl_util.validateTextureSize(width, height);\n  const texture = webgl_util.createTexture(gl);\n\n  const tex2d = gl.TEXTURE_2D;\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(tex2d, texture));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MIN_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl, () => gl.texParameteri(tex2d, gl.TEXTURE_MAG_FILTER, gl.NEAREST));\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          tex2d, 0, internalFormat, width, height, 0, textureFormat,\n          textureType, null));\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n  return texture;\n}\n\nexport function getInternalFormatForFloat32MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatFloat;\n}\n\nexport function createFloat32MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat32MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16MatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatHalfFloat;\n}\n\nexport function createFloat16MatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16MatrixTexture(textureConfig),\n      textureConfig.textureFormatFloat, textureConfig.textureTypeHalfFloat);\n}\n\nexport function getInternalFormatForUnsignedBytesMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.downloadTextureFormat;\n}\n\nexport function createUnsignedBytesMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForUnsignedBytesMatrixTexture(textureConfig), gl.RGBA,\n      gl.UNSIGNED_BYTE);\n}\n\nexport function getInternalFormatForPackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedFloat;\n}\n\nexport function createPackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height, getInternalFormatForPackedMatrixTexture(textureConfig),\n      gl.RGBA, gl.FLOAT);\n}\n\nexport function getInternalFormatForFloat16PackedMatrixTexture(\n    textureConfig: TextureConfig) {\n  return textureConfig.internalFormatPackedHalfFloat;\n}\n\nexport function createFloat16PackedMatrixTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLTexture {\n  const [width, height] =\n      tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n  return createAndConfigureTexture(\n      gl, width, height,\n      getInternalFormatForFloat16PackedMatrixTexture(textureConfig), gl.RGBA,\n      textureConfig.textureTypeHalfFloat);\n}\n\nexport function bindVertexProgramAttributeStreams(\n    gl: WebGLRenderingContext, program: WebGLProgram,\n    vertexBuffer: WebGLBuffer): boolean {\n  const posOffset = 0;               // x is the first buffer element\n  const uvOffset = 3 * 4;            // uv comes after [x y z]\n  const stride = (3 * 4) + (2 * 4);  // xyz + uv, each entry is 4-byte float.\n  webgl_util.callAndCheck(\n      gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, vertexBuffer));\n  const success = webgl_util.bindVertexBufferToProgramAttribute(\n      gl, program, 'clipSpacePos', vertexBuffer, 3, stride, posOffset);\n  return success &&\n      webgl_util.bindVertexBufferToProgramAttribute(\n          gl, program, 'uv', vertexBuffer, 2, stride, uvOffset);\n}\n\nexport function uploadDenseMatrixToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture, width: number,\n    height: number, data: TypedArray, textureConfig: TextureConfig) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n\n  let dataForUpload: TypedArray, texelDataType: number, internalFormat: number;\n  if (data instanceof Uint8Array) {\n    dataForUpload = new Uint8Array(width * height * 4);\n    texelDataType = gl.UNSIGNED_BYTE;\n    internalFormat = gl.RGBA;\n  } else {\n    dataForUpload = new Float32Array(width * height * 4);\n    texelDataType = gl.FLOAT;\n    internalFormat = textureConfig.internalFormatPackedFloat;\n  }\n\n  dataForUpload.set(data);\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.texImage2D(\n          gl.TEXTURE_2D, 0, internalFormat, width, height, 0, gl.RGBA,\n          texelDataType, dataForUpload));\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function uploadPixelDataToTexture(\n    gl: WebGLRenderingContext, texture: WebGLTexture,\n    pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n    HTMLVideoElement) {\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, texture));\n  if ((pixels as PixelData).data instanceof Uint8Array) {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, pixels.width, pixels.height, 0, gl.RGBA,\n            gl.UNSIGNED_BYTE, (pixels as PixelData).data));\n  } else {\n    webgl_util.callAndCheck(\n        gl,\n        () => gl.texImage2D(\n            gl.TEXTURE_2D, 0, gl.RGBA, gl.RGBA, gl.UNSIGNED_BYTE,\n            pixels as ImageData | HTMLImageElement | HTMLCanvasElement |\n                HTMLVideoElement));\n  }\n\n  webgl_util.callAndCheck(gl, () => gl.bindTexture(gl.TEXTURE_2D, null));\n}\n\nexport function createBufferFromOutputTexture(\n    gl2: WebGL2RenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig): WebGLBuffer {\n  // Create and bind the buffer.\n  const buffer = gl2.createBuffer();\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer));\n\n  // Initialize the buffer to the size of the texture in bytes.\n  const bytesPerFloat = 4;\n  const valuesPerTexel = 4;\n  const bufferSizeBytes = bytesPerFloat * valuesPerTexel * rows * columns;\n\n  webgl_util.callAndCheck(\n      gl2,\n      () => gl2.bufferData(\n          gl2.PIXEL_PACK_BUFFER, bufferSizeBytes, gl2.STREAM_READ));\n\n  // Enqueue a command on the GPU command queue to copy of texture into the\n  // buffer.\n  webgl_util.callAndCheck(\n      gl2, () => gl2.readPixels(0, 0, columns, rows, gl2.RGBA, gl2.FLOAT, 0));\n\n  webgl_util.callAndCheck(\n      gl2, () => gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null));\n\n  return buffer;\n}\n\nexport function downloadFloat32MatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer,\n    size: number): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget = new Float32Array(size);\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadByteEncodedFloatMatrixFromOutputTexture(\n    gl: WebGLRenderingContext, rows: number, columns: number,\n    textureConfig: TextureConfig) {\n  const [w, h] =\n      tex_util.getUnpackedMatrixTextureShapeWidthHeight(rows, columns);\n\n  const numChannels = 4;\n  const downloadTarget = new Uint8Array(\n      tex_util.getUnpackedArraySizeFromMatrixSize(rows * columns, numChannels));\n\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, w, h, textureConfig.downloadTextureFormat, gl.UNSIGNED_BYTE,\n          downloadTarget));\n\n  // By wrapping the buffer in a Float32Array, we use native browser IEEE 754\n  // decoding of the 4 bytes that back each 32 bit float.\n  return new Float32Array(downloadTarget.buffer);\n}\n\nexport function downloadPackedMatrixFromBuffer(\n    gl: WebGLRenderingContext, buffer: WebGLBuffer, batch: number, rows: number,\n    cols: number, physicalRows: number, physicalCols: number,\n    textureConfig: TextureConfig): Float32Array {\n  const gl2 = gl as WebGL2RenderingContext;\n\n  const downloadTarget =\n      new Float32Array(tex_util.getPackedRGBAArraySizeFromMatrixShape(\n          physicalRows, physicalCols));\n\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, buffer);\n  gl2.getBufferSubData(gl2.PIXEL_PACK_BUFFER, 0, downloadTarget);\n  gl2.bindBuffer(gl2.PIXEL_PACK_BUFFER, null);\n\n  return downloadTarget;\n}\n\nexport function downloadMatrixFromPackedOutputTexture(\n    gl: WebGLRenderingContext, physicalRows: number,\n    physicalCols: number): Float32Array {\n  const packedRGBA = new Float32Array(physicalRows * physicalCols * 4);\n  webgl_util.callAndCheck(\n      gl,\n      () => gl.readPixels(\n          0, 0, physicalCols, physicalRows, gl.RGBA, gl.FLOAT, packedRGBA));\n\n  return packedRGBA;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, PixelData, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {getWebGLContext, setWebGLContext} from './canvas_util';\nimport * as gpgpu_util from './gpgpu_util';\nimport * as tex_util from './tex_util';\nimport {TextureConfig} from './tex_util';\nimport {WebGL1DisjointQueryTimerExtension, WebGL2DisjointQueryTimerExtension} from './webgl_types';\nimport * as webgl_util from './webgl_util';\n\nexport interface FenceContext {\n  query: WebGLQuery|WebGLSync;\n  isFencePassed(): boolean;\n}\n\nexport class GPGPUContext {\n  gl: WebGLRenderingContext;\n  textureFloatExtension: {};\n  textureHalfFloatExtension: {};\n  colorBufferFloatExtension: {};\n  colorBufferHalfFloatExtension: {};\n  disjointQueryTimerExtension: WebGL2DisjointQueryTimerExtension|\n      WebGL1DisjointQueryTimerExtension;\n  vertexBuffer: WebGLBuffer;\n  indexBuffer: WebGLBuffer;\n  framebuffer: WebGLFramebuffer;\n  outputTexture: WebGLTexture|null = null;\n  program: WebGLProgram|null = null;\n  private disposed = false;\n  private disjoint: boolean;\n  textureConfig: TextureConfig;\n\n  constructor(gl?: WebGLRenderingContext) {\n    const glVersion = env().getNumber('WEBGL_VERSION');\n    if (gl != null) {\n      this.gl = gl;\n      setWebGLContext(glVersion, gl);\n    } else {\n      this.gl = getWebGLContext(glVersion);\n    }\n    // WebGL 2.0 enables texture floats without an extension.\n    let COLOR_BUFFER_FLOAT = 'WEBGL_color_buffer_float';\n    const COLOR_BUFFER_HALF_FLOAT = 'EXT_color_buffer_half_float';\n    if (env().getNumber('WEBGL_VERSION') === 1) {\n      const TEXTURE_FLOAT = 'OES_texture_float';\n      const TEXTURE_HALF_FLOAT = 'OES_texture_half_float';\n\n      this.textureFloatExtension =\n          webgl_util.getExtensionOrThrow(this.gl, TEXTURE_FLOAT);\n      if (webgl_util.hasExtension(this.gl, TEXTURE_HALF_FLOAT)) {\n        this.textureHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, TEXTURE_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support half float textures, yet the ' +\n            'environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n\n      this.colorBufferFloatExtension = this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            webgl_util.getExtensionOrThrow(this.gl, COLOR_BUFFER_HALF_FLOAT);\n      } else if (env().get('WEBGL_FORCE_F16_TEXTURES')) {\n        throw new Error(\n            'GL context does not support color renderable half floats, yet ' +\n            'the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.');\n      }\n    } else {\n      COLOR_BUFFER_FLOAT = 'EXT_color_buffer_float';\n      if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_FLOAT)) {\n        this.colorBufferFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_FLOAT);\n      } else if (webgl_util.hasExtension(this.gl, COLOR_BUFFER_HALF_FLOAT)) {\n        this.colorBufferHalfFloatExtension =\n            this.gl.getExtension(COLOR_BUFFER_HALF_FLOAT);\n      } else {\n        throw new Error('GL context does not support color renderable floats');\n      }\n    }\n\n    this.vertexBuffer = gpgpu_util.createVertexBuffer(this.gl);\n    this.indexBuffer = gpgpu_util.createIndexBuffer(this.gl);\n    this.framebuffer = webgl_util.createFramebuffer(this.gl);\n\n    this.textureConfig =\n        tex_util.getTextureConfig(this.gl, this.textureHalfFloatExtension);\n  }\n\n  private get debug(): boolean {\n    return env().getBool('DEBUG');\n  }\n\n  public dispose() {\n    if (this.disposed) {\n      return;\n    }\n    if (this.program != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound WebGLProgram.' +\n          ' This is probably a resource leak, delete the program with ' +\n          'GPGPUContext.deleteProgram before disposing.');\n    }\n    if (this.outputTexture != null) {\n      console.warn(\n          'Disposing a GPGPUContext that still has a bound output matrix ' +\n          'texture.  This is probably a resource leak, delete the output ' +\n          'matrix texture with GPGPUContext.deleteMatrixTexture before ' +\n          'disposing.');\n    }\n    const gl = this.gl;\n    webgl_util.callAndCheck(gl, () => gl.finish());\n    webgl_util.callAndCheck(gl, () => gl.bindFramebuffer(gl.FRAMEBUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteFramebuffer(this.framebuffer));\n    webgl_util.callAndCheck(gl, () => gl.bindBuffer(gl.ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(\n        gl, () => gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, null));\n    webgl_util.callAndCheck(gl, () => gl.deleteBuffer(this.indexBuffer));\n    this.disposed = true;\n  }\n\n  public createFloat32MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat32MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createFloat16MatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16MatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createUnsignedBytesMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createUnsignedBytesMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public uploadPixelDataToTexture(\n      texture: WebGLTexture,\n      pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadPixelDataToTexture(this.gl, texture, pixels);\n  }\n\n  public uploadDenseMatrixToTexture(\n      texture: WebGLTexture, width: number, height: number, data: TypedArray) {\n    this.throwIfDisposed();\n    gpgpu_util.uploadDenseMatrixToTexture(\n        this.gl, texture, width, height, data, this.textureConfig);\n  }\n\n  public createFloat16PackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createFloat16PackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public createPackedMatrixTexture(rows: number, columns: number):\n      WebGLTexture {\n    this.throwIfDisposed();\n    return gpgpu_util.createPackedMatrixTexture(\n        this.gl, rows, columns, this.textureConfig);\n  }\n\n  public deleteMatrixTexture(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    if (this.outputTexture === texture) {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n      this.outputTexture = null;\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.deleteTexture(texture));\n  }\n\n  public downloadByteEncodedFloatMatrixFromOutputTexture(\n      texture: WebGLTexture, rows: number, columns: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadByteEncodedFloatMatrixFromOutputTexture(\n            this.gl, rows, columns, this.textureConfig));\n  }\n\n  public downloadPackedMatrixFromBuffer(\n      buffer: WebGLBuffer, batch: number, rows: number, columns: number,\n      physicalRows: number, physicalCols: number): Float32Array {\n    return gpgpu_util.downloadPackedMatrixFromBuffer(\n        this.gl, buffer, batch, rows, columns, physicalRows, physicalCols,\n        this.textureConfig);\n  }\n\n  public downloadFloat32MatrixFromBuffer(buffer: WebGLBuffer, size: number):\n      Float32Array {\n    return gpgpu_util.downloadFloat32MatrixFromBuffer(this.gl, buffer, size);\n  }\n\n  public createBufferFromTexture(\n      texture: WebGLTexture, rows: number, columns: number): WebGLBuffer {\n    this.bindTextureToFrameBuffer(texture);\n    const result = gpgpu_util.createBufferFromOutputTexture(\n        this.gl as WebGL2RenderingContext, rows, columns, this.textureConfig);\n    this.unbindTextureToFrameBuffer();\n    return result;\n  }\n\n  public createAndWaitForFence(): Promise<void> {\n    const fenceContext = this.createFence(this.gl);\n    return this.pollFence(fenceContext);\n  }\n\n  private createFence(gl: WebGLRenderingContext): FenceContext {\n    let query: WebGLQuery|WebGLSync;\n    let isFencePassed: () => boolean;\n\n    if (env().getBool('WEBGL_FENCE_API_ENABLED')) {\n      const gl2 = gl as WebGL2RenderingContext;\n\n      const sync = gl2.fenceSync(gl2.SYNC_GPU_COMMANDS_COMPLETE, 0);\n      gl.flush();\n\n      isFencePassed = () => {\n        const status = gl2.clientWaitSync(sync, 0, 0);\n        return status === gl2.ALREADY_SIGNALED ||\n            status === gl2.CONDITION_SATISFIED;\n      };\n\n      query = sync;\n    } else if (\n        env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') > 0) {\n      query = this.beginQuery();\n      this.endQuery();\n      isFencePassed = () => this.isQueryAvailable(\n          query,\n          env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n    } else {\n      // If we have no way to fence, return true immediately. This will fire in\n      // WebGL 1.0 when there is no disjoint query timer. In this case, because\n      // the fence passes immediately, we'll immediately ask for a download of\n      // the texture, which will cause the UI thread to hang.\n      isFencePassed = () => true;\n    }\n\n    return {query, isFencePassed};\n  }\n\n  public downloadMatrixFromPackedTexture(\n      texture: WebGLTexture, physicalRows: number,\n      physicalCols: number): Float32Array {\n    return this.downloadMatrixDriver(\n        texture,\n        () => gpgpu_util.downloadMatrixFromPackedOutputTexture(\n            this.gl, physicalRows, physicalCols));\n  }\n\n  private vertexAttrsAreBound = false;\n\n  public createProgram(fragmentShaderSource: string): WebGLProgram {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    const fragmentShader: WebGLShader =\n        webgl_util.createFragmentShader(gl, fragmentShaderSource);\n    const vertexShader: WebGLShader = gpgpu_util.createVertexShader(gl);\n    const program: WebGLProgram = webgl_util.createProgram(gl);\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, vertexShader));\n    webgl_util.callAndCheck(gl, () => gl.attachShader(program, fragmentShader));\n    webgl_util.linkProgram(gl, program);\n    if (this.debug) {\n      webgl_util.validateProgram(gl, program);\n    }\n    if (!this.vertexAttrsAreBound) {\n      this.setProgram(program);\n      this.vertexAttrsAreBound = gpgpu_util.bindVertexProgramAttributeStreams(\n          gl, this.program, this.vertexBuffer);\n    }\n    return program;\n  }\n\n  public deleteProgram(program: WebGLProgram) {\n    this.throwIfDisposed();\n    if (program === this.program) {\n      this.program = null;\n    }\n    if (program != null) {\n      webgl_util.callAndCheck(this.gl, () => this.gl.deleteProgram(program));\n    }\n  }\n\n  public setProgram(program: WebGLProgram|null) {\n    this.throwIfDisposed();\n    this.program = program;\n    if ((this.program != null) && this.debug) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.callAndCheck(this.gl, () => this.gl.useProgram(program));\n  }\n\n  public getUniformLocation(\n      program: WebGLProgram, uniformName: string,\n      shouldThrow = true): WebGLUniformLocation {\n    this.throwIfDisposed();\n    if (shouldThrow) {\n      return webgl_util.getProgramUniformLocationOrThrow(\n          this.gl, program, uniformName);\n    } else {\n      return webgl_util.getProgramUniformLocation(\n          this.gl, program, uniformName);\n    }\n  }\n\n  public getAttributeLocation(program: WebGLProgram, attribute: string):\n      number {\n    this.throwIfDisposed();\n    return webgl_util.callAndCheck(\n        this.gl, () => this.gl.getAttribLocation(program, attribute));\n  }\n\n  public getUniformLocationNoThrow(program: WebGLProgram, uniformName: string):\n      WebGLUniformLocation {\n    this.throwIfDisposed();\n    return this.gl.getUniformLocation(program, uniformName);\n  }\n\n  public setInputMatrixTexture(\n      inputMatrixTexture: WebGLTexture, uniformLocation: WebGLUniformLocation,\n      textureUnit: number) {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    webgl_util.bindTextureToProgramUniformSampler(\n        this.gl, inputMatrixTexture, uniformLocation, textureUnit);\n  }\n\n  public setOutputMatrixTexture(\n      outputMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.setOutputMatrixTextureDriver(outputMatrixTexture, columns, rows);\n  }\n\n  public setOutputPackedMatrixTexture(\n      outputPackedMatrixTexture: WebGLTexture, rows: number, columns: number) {\n    this.throwIfDisposed();\n    const [width, height] =\n        tex_util.getPackedMatrixTextureShapeWidthHeight(rows, columns);\n    this.setOutputMatrixTextureDriver(outputPackedMatrixTexture, width, height);\n  }\n\n  public setOutputMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    this.setOutputMatrixWriteRegionDriver(\n        startColumn, startRow, numColumns, numRows);\n  }\n\n  public setOutputPackedMatrixWriteRegion(\n      startRow: number, numRows: number, startColumn: number,\n      numColumns: number) {\n    throw new Error('setOutputPackedMatrixWriteRegion not implemented.');\n  }\n\n  public debugValidate() {\n    if (this.program != null) {\n      webgl_util.validateProgram(this.gl, this.program);\n    }\n    webgl_util.validateFramebuffer(this.gl);\n  }\n\n  public executeProgram() {\n    this.throwIfDisposed();\n    this.throwIfNoProgram();\n    const gl = this.gl;\n    if (this.debug) {\n      this.debugValidate();\n    }\n    webgl_util.callAndCheck(\n        gl, () => gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_SHORT, 0));\n  }\n\n  public blockUntilAllProgramsCompleted() {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(this.gl, () => this.gl.finish());\n  }\n\n  private getQueryTimerExtension(): WebGL1DisjointQueryTimerExtension\n      |WebGL2DisjointQueryTimerExtension {\n    if (this.disjointQueryTimerExtension == null) {\n      this.disjointQueryTimerExtension =\n          webgl_util.getExtensionOrThrow(\n              this.gl,\n              env().getNumber(\n                  'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2 ?\n                  'EXT_disjoint_timer_query_webgl2' :\n                  'EXT_disjoint_timer_query') as\n              WebGL1DisjointQueryTimerExtension |\n          WebGL2DisjointQueryTimerExtension;\n    }\n    return this.disjointQueryTimerExtension;\n  }\n\n  private getQueryTimerExtensionWebGL2(): WebGL2DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension();\n  }\n\n  private getQueryTimerExtensionWebGL1(): WebGL1DisjointQueryTimerExtension {\n    return this.getQueryTimerExtension() as WebGL1DisjointQueryTimerExtension;\n  }\n\n  beginQuery(): WebGLQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const query = gl2.createQuery();\n      gl2.beginQuery(ext.TIME_ELAPSED_EXT, query);\n      return query;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    const query = ext.createQueryEXT() as WebGLQuery;\n    ext.beginQueryEXT(ext.TIME_ELAPSED_EXT, query);\n    return query;\n  }\n\n  endQuery() {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION') === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n      gl2.endQuery(ext.TIME_ELAPSED_EXT);\n      return;\n    }\n    const ext = this.getQueryTimerExtensionWebGL1();\n    ext.endQueryEXT(ext.TIME_ELAPSED_EXT);\n  }\n\n  public async waitForQueryAndGetTime(query: WebGLQuery): Promise<number> {\n    await util.repeatedTry(\n        () => this.disposed ||  // while testing contexts are created / disposed\n                                // in rapid succession, so without this check we\n                                // may poll for the query timer indefinitely\n            this.isQueryAvailable(\n                query,\n                env().getNumber(\n                    'WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION')));\n    return this.getQueryTime(\n        query, env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION'));\n  }\n\n  private getQueryTime(query: WebGLQuery, queryTimerVersion: number): number {\n    if (queryTimerVersion === 0) {\n      return null;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n\n      const timeElapsedNanos = gl2.getQueryParameter(query, gl2.QUERY_RESULT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const timeElapsedNanos =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_EXT);\n      // Return milliseconds.\n      return timeElapsedNanos / 1000000;\n    }\n  }\n\n  private isQueryAvailable(query: WebGLQuery, queryTimerVersion: number):\n      boolean {\n    if (queryTimerVersion === 0) {\n      return true;\n    }\n\n    if (queryTimerVersion === 2) {\n      const gl2 = this.gl as WebGL2RenderingContext;\n      const ext = this.getQueryTimerExtensionWebGL2();\n\n      const available =\n          gl2.getQueryParameter(query, gl2.QUERY_RESULT_AVAILABLE);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    } else {\n      const ext = this.getQueryTimerExtensionWebGL1();\n\n      const available =\n          ext.getQueryObjectEXT(query, ext.QUERY_RESULT_AVAILABLE_EXT);\n      if (this.disjoint == null) {\n        this.disjoint = this.gl.getParameter(ext.GPU_DISJOINT_EXT);\n      }\n\n      return available && !this.disjoint;\n    }\n  }\n\n  pollFence(fenceContext: FenceContext) {\n    return new Promise<void>(resolve => {\n      this.addItemToPoll(() => fenceContext.isFencePassed(), () => resolve());\n    });\n  }\n\n  private itemsToPoll: PollItem[] = [];\n\n  pollItems(): void {\n    // Find the last query that has finished.\n    const index = linearSearchLastTrue(this.itemsToPoll.map(x => x.isDoneFn));\n    for (let i = 0; i <= index; ++i) {\n      const {resolveFn} = this.itemsToPoll[i];\n      resolveFn();\n    }\n    this.itemsToPoll = this.itemsToPoll.slice(index + 1);\n  }\n\n  private addItemToPoll(isDoneFn: () => boolean, resolveFn: () => void) {\n    this.itemsToPoll.push({isDoneFn, resolveFn});\n    if (this.itemsToPoll.length > 1) {\n      // We already have a running loop that polls.\n      return;\n    }\n    // Start a new loop that polls.\n    util.repeatedTry(() => {\n      this.pollItems();\n      // End the loop if no more items to poll.\n      return this.itemsToPoll.length === 0;\n    });\n  }\n\n  private bindTextureToFrameBuffer(texture: WebGLTexture) {\n    this.throwIfDisposed();\n    webgl_util.bindColorTextureToFramebuffer(\n        this.gl, texture, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(this.gl);\n    }\n  }\n\n  private unbindTextureToFrameBuffer() {\n    if (this.outputTexture != null) {\n      webgl_util.bindColorTextureToFramebuffer(\n          this.gl, this.outputTexture, this.framebuffer);\n      if (this.debug) {\n        webgl_util.validateFramebuffer(this.gl);\n      }\n    } else {\n      webgl_util.unbindColorTextureFromFramebuffer(this.gl, this.framebuffer);\n    }\n  }\n\n  private downloadMatrixDriver(\n      texture: WebGLTexture,\n      downloadAndDecode: () => Float32Array): Float32Array {\n    this.bindTextureToFrameBuffer(texture);\n    const result = downloadAndDecode();\n    this.unbindTextureToFrameBuffer();\n\n    return result;\n  }\n\n  private setOutputMatrixTextureDriver(\n      outputMatrixTextureMaybePacked: WebGLTexture, width: number,\n      height: number) {\n    this.throwIfDisposed();\n    const gl = this.gl;\n    webgl_util.bindColorTextureToFramebuffer(\n        gl, outputMatrixTextureMaybePacked, this.framebuffer);\n    if (this.debug) {\n      webgl_util.validateFramebuffer(gl);\n    }\n    this.outputTexture = outputMatrixTextureMaybePacked;\n    webgl_util.callAndCheck(gl, () => gl.viewport(0, 0, width, height));\n    webgl_util.callAndCheck(gl, () => gl.scissor(0, 0, width, height));\n  }\n\n  private setOutputMatrixWriteRegionDriver(\n      x: number, y: number, width: number, height: number) {\n    this.throwIfDisposed();\n    webgl_util.callAndCheck(\n        this.gl, () => this.gl.scissor(x, y, width, height));\n  }\n\n  private throwIfDisposed() {\n    if (this.disposed) {\n      throw new Error('Attempted to use disposed GPGPUContext.');\n    }\n  }\n\n  private throwIfNoProgram() {\n    if (this.program == null) {\n      throw new Error('No GPU program is currently set.');\n    }\n  }\n}\n\ntype PollItem = {\n  isDoneFn: () => boolean,\n  resolveFn: () => void\n};\n\n/**\n * Finds the index of the last true element using linear search.\n * Note: We can't do binary search because Chrome expects us to explicitly\n * test all fences before download:\n * https://github.com/tensorflow/tfjs/issues/1145\n */\nexport function linearSearchLastTrue(arr: Array<() => boolean>): number {\n  let i = 0;\n  for (; i < arr.length; ++i) {\n    const isDone = arr[i]();\n    if (!isDone) {\n      break;\n    }\n  }\n  return i - 1;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, Tensor, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport * as shader_compiler from './shader_compiler';\nimport {InputInfo, ShapeInfo} from './shader_compiler';\nimport {PackingScheme, TextureData, TextureUsage} from './tex_util';\n\nexport interface GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  /** If true, this program expects packed input textures. Defaults to false. */\n  packedInputs?: boolean;\n  /** If true, this program produces a packed texture. Defaults to false. */\n  packedOutput?: boolean;\n  /**\n   * Affects what type of texture we allocate for the output. Defaults to\n   * `TextureUsage.RENDER`.\n   */\n  outTexUsage?: TextureUsage;\n  /**\n   * The type of scheme to use when packing texels for the output values.\n   * See `PackingScheme` for details. Defaults to `PackingScheme.SHARED_BATCH`.\n   */\n  outPackingScheme?: PackingScheme;\n}\n\nexport interface GPGPUBinary {\n  webGLProgram: WebGLProgram;\n  program: GPGPUProgram;\n  uniformLocations: {[name: string]: WebGLUniformLocation};\n  source: string;\n  inShapeInfos: ShapeInfo[];\n  outShapeInfo: ShapeInfo;\n  infLoc: WebGLUniformLocation;\n  nanLoc: WebGLUniformLocation;\n}\n\nexport interface TensorData {\n  shape: number[];\n  texData: TextureData;\n  isUniform: boolean;\n  // Available when we decide to upload as uniform instead of texture.\n  uniformValues?: TypedArray;\n}\n\nexport function compileProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, program: GPGPUProgram, inputs: TensorData[],\n    output: TensorData): GPGPUBinary {\n  const userCode = program.userCode;\n  const inputInfos: InputInfo[] = inputs.map((input, i) => {\n    const shapeInfo: ShapeInfo = {\n      logicalShape: input.shape,\n      texShape: input.isUniform ? null : input.texData.texShape,\n      isUniform: input.isUniform,\n      isPacked: input.isUniform ? false : input.texData.isPacked,\n      flatOffset: null\n    };\n    if (input.texData != null && input.texData.slice != null &&\n        input.texData.slice.flatOffset > 0) {\n      shapeInfo.flatOffset = input.texData.slice.flatOffset;\n    }\n    return {name: program.variableNames[i], shapeInfo};\n  });\n  const inShapeInfos = inputInfos.map(x => x.shapeInfo);\n  const outShapeInfo: ShapeInfo = {\n    logicalShape: output.shape,\n    texShape: output.texData.texShape,\n    isUniform: false,\n    isPacked: output.texData.isPacked,\n    flatOffset: null\n  };\n  const source = shader_compiler.makeShader(\n      inputInfos, outShapeInfo, userCode, program.packedInputs);\n\n  const webGLProgram = gpgpu.createProgram(source);\n\n  // Add special uniforms (NAN, INFINITY)\n  let infLoc: WebGLUniformLocation = null;\n  const nanLoc = gpgpu.getUniformLocation(webGLProgram, 'NAN', false);\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    infLoc = gpgpu.getUniformLocation(webGLProgram, 'INFINITY', false);\n  }\n\n  // Add user-defined uniforms\n  const uniformLocations: {[name: string]: WebGLUniformLocation} = {};\n  for (let i = 0; i < program.variableNames.length; i++) {\n    const varName = program.variableNames[i];\n    const shouldThrow = false;\n    uniformLocations[varName] =\n        gpgpu.getUniformLocation(webGLProgram, varName, shouldThrow);\n    uniformLocations[`offset${varName}`] =\n        gpgpu.getUniformLocation(webGLProgram, `offset${varName}`, shouldThrow);\n  }\n\n  return {\n    program,\n    source,\n    webGLProgram,\n    uniformLocations,\n    inShapeInfos,\n    outShapeInfo,\n    infLoc,\n    nanLoc,\n  };\n}\n\nfunction validateBinaryAndProgram(\n    shapeInfos: ShapeInfo[], inputs: TensorData[]) {\n  if (shapeInfos.length !== inputs.length) {\n    throw Error(\n        `Binary was compiled with ${shapeInfos.length} inputs, but ` +\n        `was executed with ${inputs.length} inputs`);\n  }\n\n  shapeInfos.forEach((s, i) => {\n    const shapeA = s.logicalShape;\n    const input = inputs[i];\n    const shapeB = input.shape;\n\n    if (!util.arraysEqual(shapeA, shapeB)) {\n      throw Error(\n          `Binary was compiled with different shapes than ` +\n          `the current args. Shapes ${shapeA} and ${shapeB} must match`);\n    }\n    // The input is uploaded as uniform.\n    if (s.isUniform && input.isUniform) {\n      return;\n    }\n\n    const texShapeA = s.texShape;\n    const texShapeB = input.isUniform ? null : input.texData.texShape;\n    if (!util.arraysEqual(texShapeA, texShapeB)) {\n      throw Error(\n          `Binary was compiled with different texture shapes than the` +\n          ` current args. Shape ${texShapeA} and ${texShapeB} must match`);\n    }\n  });\n}\n\nexport function runProgram<T extends Tensor, K extends Tensor>(\n    gpgpu: GPGPUContext, binary: GPGPUBinary, inputs: TensorData[],\n    output: TensorData,\n    customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) =>\n        void): void {\n  validateBinaryAndProgram(binary.inShapeInfos, inputs);\n  validateBinaryAndProgram([binary.outShapeInfo], [output]);\n\n  const outTex = output.texData.texture;\n  const outTexShape = output.texData.texShape;\n  if (output.texData.isPacked) {\n    gpgpu.setOutputPackedMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  } else {\n    gpgpu.setOutputMatrixTexture(outTex, outTexShape[0], outTexShape[1]);\n  }\n  gpgpu.setProgram(binary.webGLProgram);\n\n  // Set special uniforms (NAN, INFINITY)\n  if (env().getNumber('WEBGL_VERSION') === 1) {\n    if (binary.infLoc !== null) {\n      gpgpu.gl.uniform1f(binary.infLoc, Infinity);\n    }\n  }\n  if (binary.nanLoc !== null) {\n    gpgpu.gl.uniform1f(binary.nanLoc, NaN);\n  }\n\n  // Set user-defined inputs\n  inputs.forEach((input, i) => {\n    const varName = binary.program.variableNames[i];\n    const varLoc = binary.uniformLocations[varName];\n    const varOffsetLoc = binary.uniformLocations[`offset${varName}`];\n\n    if (varLoc == null) {\n      // The compiler inferred that this variable is not used in this shader.\n      return;\n    }\n\n    if (input.isUniform) {\n      // Upload the values of the tensor as uniform.\n      if (util.sizeFromShape(input.shape) < 2) {\n        gpgpu.gl.uniform1f(varLoc, input.uniformValues[0]);\n      } else {\n        let vals = input.uniformValues;\n        if (!(vals instanceof Float32Array)) {\n          vals = new Float32Array(vals);\n        }\n        gpgpu.gl.uniform1fv(varLoc, vals);\n      }\n      return;\n    }\n\n    // If the input was sliced, upload the flat offset index.\n    if (input.texData.slice != null && varOffsetLoc != null) {\n      gpgpu.gl.uniform1i(varOffsetLoc, input.texData.slice.flatOffset);\n    }\n\n    gpgpu.setInputMatrixTexture(input.texData.texture, varLoc, i);\n  });\n\n  if (customSetup != null) {\n    customSetup(gpgpu, binary.webGLProgram);\n  }\n  gpgpu.executeProgram();\n}\n\nexport function makeShaderKey(\n    program: GPGPUProgram, inputs: TensorData[], output: TensorData): string {\n  let keyInputs = '';\n  inputs.concat(output).forEach(x => {\n    const hasOffset = x.texData != null && x.texData.slice != null &&\n        x.texData.slice.flatOffset > 0;\n    const texShape = x.isUniform ? 'uniform' : x.texData.texShape;\n    keyInputs += `${x.shape}_${texShape}_${hasOffset}`;\n  });\n  const keyUserCode = program.userCode;\n  let key = program.constructor.name;\n  // Fast string concat. See https://jsperf.com/string-concatenation/14.\n  key += '_' + keyInputs + '_' + keyUserCode;\n  return key;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {getGlslDifferences} from './glsl_version';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Im2ColPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      outputShape: number[], inputShape: number[],\n      convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = outputShape;\n\n    const {\n      filterWidth,\n      inChannels,\n      strideWidth,\n      strideHeight,\n      padInfo,\n      outWidth,\n      dilationWidth,\n      dilationHeight,\n      dataFormat\n    } = convInfo;\n    const {left, top} = padInfo;\n    const itemsPerBlockRow = inChannels * filterWidth;\n    const glsl = getGlslDifferences();\n    const isChannelsLast = dataFormat === 'channelsLast';\n    const rowDim = isChannelsLast ? 0 : 1;\n    const colDim = isChannelsLast ? 1 : 2;\n\n    let unrolled = ``;\n\n    for (let row = 0; row <= 1; row++) {\n      for (let col = 0; col <= 1; col++) {\n        unrolled += `\n          blockIndex = rc.y + ${col};\n          pos = rc.x + ${row};\n\n          if(blockIndex < ${outputShape[1]} && pos < ${outputShape[0]}) {\n            offsetY = int(blockIndex / (${outWidth})) * ${strideHeight} - ${\n            top};\n            d0 = offsetY + ${dilationHeight} * (pos / ${itemsPerBlockRow});\n\n            if(d0 < ${inputShape[rowDim]} && d0 >= 0) {\n\n              offsetX = int(mod(float(blockIndex), ${outWidth}.) * ${\n            strideWidth}. - ${left}.);\n              d1 = offsetX + ${dilationWidth} * (int(mod(float(pos), ${\n            itemsPerBlockRow}.) / ${inChannels}.));\n\n              if(d1 < ${inputShape[colDim]} && d1 >= 0) {\n\n                ch = int(mod(float(pos), ${inChannels}.));\n\n                if (${isChannelsLast}) {\n                  innerDims = vec2(d1, ch);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(d0, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                } else {\n                  innerDims = vec2(d0, d1);\n                  result[${row * 2 + col}] = getChannel(\n                    getA(ch, int(innerDims.x),\n                    int(innerDims.y)), innerDims);\n                }\n              }\n            }\n          }\n        `;\n      }\n    }\n\n    this.userCode = `\n      void main() {\n        ivec2 rc = getOutputCoords();\n\n        vec4 result = vec4(0);\n\n        int blockIndex, pos, offsetY, d0, offsetX, d1, ch;\n        vec2 innerDims;\n\n        ${unrolled}\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n        int d = coords[3];\n        float x = getX(b, r, c, d);\n        float sum = 0.0;\n        for (int j = -${rad}; j <= ${rad}; j++) {\n          int idx = d + j;\n          if (idx >= 0 && idx <=  ${maxD}) {\n            float z = getX(b, r, c, idx);\n            sum += z * z;\n          }\n        }\n        float val = x * ${powOperator};\n        setOutput(val);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNGradProgram implements GPGPUProgram {\n  variableNames = ['inputImage', 'outputImage', 'dy'];\n  outputShape: number[] = [];\n  userCode: string;\n  depthRadius: number;\n  bias: number;\n  alpha: number;\n  beta: number;\n  depth: number;\n\n  constructor(\n      inputShape: number[], depthRadius: number, bias: number, alpha: number,\n      beta: number) {\n    this.outputShape = inputShape;\n    this.depth = inputShape[3];\n    this.depthRadius = depthRadius;\n    this.bias = bias;\n    this.alpha = alpha;\n    this.beta = beta;\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int r = coords[1];\n        int c = coords[2];\n\n        float result = 0.0;\n        for (int d = 0; d < ${this.depth}; ++d) {\n          int depthBegin = int(max(0.0, float(d - ${depthRadius})));\n          int depthEnd = int(min(float(${this.depth}),\n              float(d + ${depthRadius} + 1)));\n\n          const int MIN_DEPTH_BEGIN = 0;\n          const int MAX_DEPTH_END = ${this.depth};\n\n          float norm = 0.0;\n          for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd) {\n              norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);\n            }\n            else {\n              break;\n            }\n          }\n\n          norm = float(${alpha}) * norm + float(${bias});\n\n          for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){\n            if (k < depthBegin){\n              continue;\n            }\n            else if (k >= depthBegin && k < depthEnd){\n              float dyi = -2.0 * float(${alpha})\n                * float(${beta})\n                * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)\n                / norm;\n              if (k == d) {\n                dyi += pow(norm, -1.0 * ${beta});\n              }\n              if (k == coords[3]) {\n                dyi *= getDy(b, r, c, d);\n                result += dyi;\n              }\n            }\n            else {\n              break;\n            }\n          }\n      }\n      setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class LRNPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[] = [];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], radius: number, bias: number, alpha: number,\n      beta: number) {\n    const rad = radius;\n    const maxD = xShape[3] - 1;\n    this.outputShape = xShape;\n\n    // optimize pow(bias + alpha * sum, -beta)\n    // src: https://github.com/tensorflow/tensorflow/..\n    // blob/26033a1644a9c4a5fbe3170ab2e864b6a4ccd4ca/..\n    // tensorflow/core/kernels/mkl_lrn_op.cc#L320\n    let powOperator;\n    const basis = `float(${bias}) + float(${alpha}) * sum`;\n    if (beta === 0.5) {\n      powOperator = `inversesqrt(${basis})`;\n    } else if (beta === 1.0) {\n      powOperator = `1.0/(${basis})`;\n    } else {\n      powOperator = `exp(log(${basis}) * float(-${beta}));`;\n    }\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords.x;\n        int r = coords.y;\n        int c = coords.z;\n        int d = coords.w;\n\n        bool hasNextCol = d < ${this.outputShape[3]};\n        bool hasNextRow = c < ${this.outputShape[2]};\n\n        vec4 sum = vec4(0.);\n        vec4 xFragAtOutputCoords = getX(b, r, c, d);\n\n        vec4 xAtOutputCoords = vec4(\n          getChannel(xFragAtOutputCoords, vec2(c, d)),\n          hasNextCol ?\n            getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,\n          hasNextRow ?\n            getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0\n        );\n\n        int firstChannel = d - ${rad};\n        vec2 cache = vec2(0.);\n        if(firstChannel >= 0){\n          vec4 firstChannelFrag = getX(b, r, c, firstChannel);\n          cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));\n            if(hasNextRow){\n              cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));\n            }\n        }\n\n        ivec2 depth = ivec2(d, d + 1);\n        for (int j = - ${rad}; j <= ${rad}; j++) {\n          ivec2 idx = depth + j;\n          bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));\n          bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${maxD}));\n\n          bool depthInRange = aboveLowerBound.x && belowUpperBound.x;\n          bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;\n\n          if(depthInRange || depthPlusOneInRange){\n            vec4 z = vec4(0.);\n            vec4 xFragAtCurrentDepth;\n            z.xz = cache.xy;\n            if(depthPlusOneInRange && hasNextCol){\n              xFragAtCurrentDepth = idx.y != d ?\n                getX(b, r, c, idx.y) : xFragAtOutputCoords;\n              z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));\n              if(hasNextRow){\n                z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));\n              }\n            }\n            cache.xy = z.yw;\n            sum += z * z;\n          }\n        }\n        vec4 result = xAtOutputCoords * ${powOperator};\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MaxPool2DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv2DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex = effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n\n        ivec2 dyRCCorner = coords.yz - pads;\n        int dyRCorner = dyRCCorner.x;\n        int dyCCorner = dyRCCorner.y;\n\n        // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n          wR += ${dilationHeight}) {\n          float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n          if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 || fract(dyR) > 0.0) {\n            continue;\n          }\n          int idyR = int(dyR);\n\n          for (int wC = 0; wC < ${effectiveFilterWidth}; wC++) {\n            float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n            if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                fract(dyC) > 0.0) {\n              continue;\n            }\n            int idyC = int(dyC);\n\n            float dyValue = getDy(b, idyR, idyC, d);\n            int maxPosValue = ${lastIndex} - int(getMaxPos(b, idyR, idyC, d));\n\n            // Get the current value, check it against the value from the\n            // position matrix.\n            int curPosValue = wR * ${effectiveFilterWidth} + wC;\n            float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n            dotProd += dyValue * mask;\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n\nexport class MaxPool3DBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy', 'maxPos'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(convInfo: backend_util.Conv3DInfo) {\n    this.outputShape = convInfo.inShape;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = effectiveFilterDepth - 1 - convInfo.padInfo.front;\n    const padTop = effectiveFilterHeight - 1 - convInfo.padInfo.top;\n    const padLeft = effectiveFilterWidth - 1 - convInfo.padInfo.left;\n\n    const lastIndex =\n        effectiveFilterDepth * effectiveFilterHeight * effectiveFilterWidth - 1;\n    this.userCode = `\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;\n        int dyDCorner = dyCorner.x;\n        int dyRCorner = dyCorner.y;\n        int dyCCorner = dyCorner.z;\n\n        // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get\n        // dx(xD, xR, xC, ch).\n        // ? = to be determined. : = across all values in that axis.\n        float dotProd = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n           wD += ${dilationDepth}) {\n          float dyD = float(dyDCorner + wD) / ${strideDepth}.0;\n\n          if (dyD < 0.0 || dyD >= ${convInfo.outDepth}.0 || fract(dyD) > 0.0) {\n            continue;\n          }\n          int idyD = int(dyD);\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            float dyR = float(dyRCorner + wR) / ${strideHeight}.0;\n\n            if (dyR < 0.0 || dyR >= ${convInfo.outHeight}.0 ||\n                fract(dyR) > 0.0) {\n              continue;\n            }\n            int idyR = int(dyR);\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              float dyC = float(dyCCorner + wC) / ${strideWidth}.0;\n\n              if (dyC < 0.0 || dyC >= ${convInfo.outWidth}.0 ||\n                  fract(dyC) > 0.0) {\n                continue;\n              }\n              int idyC = int(dyC);\n\n              float dyValue = getDy(batch, idyD, idyR, idyC, ch);\n              int maxPosValue = ${lastIndex} -\n                  int(getMaxPos(batch, idyD, idyR, idyC, ch));\n\n              // Get the current value, check it against the value from the\n              // position matrix.\n              int curPosValue =\n                  wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                  wR * ${effectiveFilterWidth} + wC;\n              float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);\n\n              dotProd += dyValue * mask;\n            }\n          }\n        }\n        setOutput(dotProd);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MatMulPackedProgram implements GPGPUProgram {\n  variableNames = ['matrixA', 'matrixB'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      aShape: [number, number, number], outputShape: [number, number, number],\n      transposeA = false, transposeB = false, addBias = false,\n      activation: string = null, hasPreluActivation = false) {\n    this.outputShape = outputShape;\n\n    const sharedDim = transposeA ? aShape[1] : aShape[2];\n    const sharedDimensionPacked = Math.ceil(sharedDim / 2);\n\n    const aSample = transposeA ? 'i * 2, rc.y' : 'rc.y, i * 2';\n    const bSample = transposeB ? 'rc.z, i * 2' : 'i * 2, rc.z';\n    const aSwizzle = transposeA ? ['a.xxyy', 'a.zzww'] : ['a.xxzz', 'a.yyww'];\n    const bSwizzle = transposeB ? ['b.xzxz', 'b.ywyw'] : ['b.xyxy', 'b.zwzw'];\n\n    let activationSnippet = '', applyActivationSnippet = '';\n    if (activation) {\n      if (hasPreluActivation) {\n        activationSnippet = `vec4 activation(vec4 a) {\n          vec4 b = getPreluActivationWeightsAtOutCoords();\n          ${activation}\n        }`;\n      } else {\n        activationSnippet = `vec4 activation(vec4 x) {\n          ${activation}\n        }`;\n      }\n\n      applyActivationSnippet = `result = activation(result);`;\n    }\n\n    const addBiasSnippet = addBias ? 'result += getBiasAtOutCoords();' : '';\n    if (addBias) {\n      this.variableNames.push('bias');\n    }\n\n    if (hasPreluActivation) {\n      this.variableNames.push('preluActivationWeights');\n    }\n\n    this.userCode = `\n      ${activationSnippet}\n\n      const float sharedDimension = ${sharedDimensionPacked}.0;\n\n      vec4 dot2x2ARowBCol(ivec3 rc) {\n        vec4 result = vec4(0);\n        for (int i = 0; i < ${sharedDimensionPacked}; i++) {\n          vec4 a = getMatrixA(rc.x, ${aSample});\n          vec4 b = getMatrixB(rc.x, ${bSample});\n\n          // These swizzled products need to be separately added.\n          // See: https://github.com/tensorflow/tfjs/issues/1735\n          result += (${aSwizzle[0]} * ${bSwizzle[0]});\n          result += (${aSwizzle[1]} * ${bSwizzle[1]});\n        }\n        return result;\n      }\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n        vec4 result = dot2x2ARowBCol(rc);\n\n        ${addBiasSnippet}\n\n        ${applyActivationSnippet}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class MultinomialProgram implements GPGPUProgram {\n  variableNames = ['probs'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(batchSize: number, numOutcomes: number, numSamples: number) {\n    this.outputShape = [batchSize, numSamples];\n\n    this.userCode = `\n      uniform float seed;\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n\n        float r = random(seed);\n        float cdf = 0.0;\n\n        for (int i = 0; i < ${numOutcomes - 1}; i++) {\n          cdf += getProbs(batch, i);\n\n          if (r < cdf) {\n            setOutput(float(i));\n            return;\n          }\n        }\n\n        // If no other event happened, last event happened.\n        setOutput(float(${numOutcomes - 1}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(seed: number) {\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.seedLoc == null) {\n        this.seedLoc = gpgpu.getUniformLocation(webGLProgram, 'seed');\n      }\n      gpgpu.gl.uniform1f(this.seedLoc, seed);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class OneHotProgram implements GPGPUProgram {\n  variableNames = ['indices'];\n  outputShape: number[];\n  userCode: string;\n\n  // Caching uniform location for speed.\n  seedLoc: WebGLUniformLocation;\n\n  constructor(\n      numIndices: number, depth: number, onValue: number, offValue: number) {\n    this.outputShape = [numIndices, depth];\n\n    this.userCode = `\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int index = round(getIndices(coords.x));\n        setOutput(mix(float(${offValue}), float(${onValue}),\n                      float(index == coords.y)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(\n      outputShape:\n          number[]) {  // TODO(https://github.com/tensorflow/tfjs/issues/893):\n                       // Only input / output 3D tensors.\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    if (rank === 0) {\n      this.userCode = `\n        void main() {\n          setOutput(vec4(getA(), 0., 0., 0.));\n        }\n      `;\n    } else {\n      const channels = getChannels('rc', rank);\n      const dtype = getCoordsDataType(rank);\n      const outOfBoundsCondition =\n          getOutOfBoundsCondition(rank, outputShape, channels);\n      const setup = getSetup(\n          rank, outputShape[outputShape.length - 1],\n          outputShape[outputShape.length - 2], channels);\n      const output = getOutput(outputShape, channels);\n\n      this.userCode = `\n        void main() {\n          ${dtype} rc = getOutputCoords();\n\n          if(${outOfBoundsCondition}) {\n            setOutput(vec4(0));\n          } else {\n            ${setup}\n\n            setOutput(vec4(${output}));\n          }\n        }\n      `;\n    }\n  }\n}\n\nfunction getSourceCoordsArr(rank: number, dims: string[]): string[] {\n  const coords = [];\n\n  for (let row = 0; row <= 1; row++) {\n    for (let col = 0; col <= 1; col++) {\n      let coord = `${row === 0 ? 'r' : 'rp1'}, ${col === 0 ? 'c' : 'cp1'}`;\n\n      for (let d = 2; d < rank; d++) {\n        coord = `${dims[dims.length - 1 - d]},` + coord;\n      }\n\n      coords.push(coord);\n    }\n  }\n  return coords;\n}\n\nfunction getOutOfBoundsCondition(\n    rank: number, shape: number[], dims: string[]): string {\n  if (rank === 1) {\n    return `rc > ${shape[0]}`;\n  }\n\n  let cond = '';\n  for (let i = rank - 2; i < rank; i++) {\n    cond += `${dims[i]} >= ${shape[i]}`;\n    if (i < rank - 1) {\n      cond += '||';\n    }\n  }\n\n  return cond;\n}\n\nfunction getSetup(\n    rank: number, cols: number, rows: number, dims: string[]): string {\n  if (rank === 1) {\n    return '';\n  }\n\n  const innerDims = dims.slice(-2);\n\n  return `\n    int r = ${innerDims[0]};\n    int c = ${innerDims[1]};\n    int rp1 = r + 1;\n    int cp1 = c + 1;\n\n    bool cEdge = cp1 >= ${cols};\n    bool rEdge = rp1 >= ${rows};\n  `;\n}\n\nfunction getOutput(shape: number[], dims: string[]): string {\n  const rank = shape.length;\n  const sourceCoords = getSourceCoordsArr(rank, dims);\n  if (rank === 1) {\n    return `getA(rc),\n            rc + 1 >= ${shape[0]} ? 0. : getA(rc + 1),\n            0, 0`;\n  }\n\n  return `getA(${sourceCoords[0]}),\n          cEdge ? 0. : getA(${sourceCoords[1]}),\n          rEdge ? 0. : getA(${sourceCoords[2]}),\n          rEdge || cEdge ? 0. : getA(${sourceCoords[3]})`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const type = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const unpackedCoords =\n        ['coords[0]', 'coords[1]', 'coords[2]', 'coords[3]'].slice(0, rank);\n\n    if (rank === 1) {\n      this.userCode = `\n        int start = ${start};\n        int end = ${end};\n\n        void main() {\n          int outC = getOutputCoords();\n          if (outC < start || outC >= end) {\n            setOutput(float(${constantValue}));\n          } else {\n            setOutput(getX(outC - start));\n          }\n        }\n      `;\n      return;\n    }\n    this.userCode = `\n      ${type} start = ${type}(${start});\n      ${type} end = ${type}(${end});\n\n      void main() {\n        ${type} outC = getOutputCoords();\n        if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {\n          setOutput(float(${constantValue}));\n        } else {\n          ${type} coords = outC - start;\n          setOutput(getX(${unpackedCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class PadPackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      xShape: number[], paddings: Array<[number, number]>,\n      constantValue: number) {\n    this.outputShape = paddings.map(\n        (p, i) => p[0] /* beforePad */ + xShape[i] + p[1] /* afterPad */);\n    const rank = xShape.length;\n    const dtype = getCoordsDataType(rank);\n\n    const start = paddings.map(p => p[0]).join(',');\n    const end = paddings.map((p, i) => p[0] + xShape[i]).join(',');\n    const coords = getChannels('rc', rank);\n    const source = getChannels('source', rank);\n    const cLimit = `${coords[rank - 1]} < ${this.outputShape[rank - 1]}`;\n    const innerDims =\n        rank === 1 ? 'source' : `vec2(${source.slice(-2).join()})`;\n\n    const componentSetup = [\n      `${dtype} rc = outputLoc;`, `${coords[rank - 1]} += 1;\n       if(${cLimit}) {\n      `,\n      rank === 1 ? '' : `}\n       rc = outputLoc;\n       ${coords[rank - 2]} += 1;\n       if(${coords[rank - 2]} < ${this.outputShape[rank - 2]}) {`,\n      rank === 1 ? '' : `  ${coords[rank - 1]} += 1;\n         if(${cLimit}) {`\n    ];\n\n    const paddingArea = rank === 1 ?\n        'rc < start || rc >= end' :\n        'any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))';\n    let mainLoop = '';\n    for (let i = 0, j = rank === 1 ? 2 : 4; i < j; i++) {\n      mainLoop += `\n        ${componentSetup[i]}\n        if (${paddingArea}) {\n          result[${i}] = float(${constantValue});\n        } else {\n          ${dtype} source = rc - start;\n          result[${i}] = getChannel(getX(${source.join()}), ${innerDims});\n        }\n      `;\n    }\n    mainLoop += (rank === 1 ? `} ` : `}}`);\n\n    this.userCode = `\n      const ${dtype} start = ${dtype}(${start});\n      const ${dtype} end = ${dtype}(${end});\n\n      void main() {\n        ${dtype} outputLoc = getOutputCoords();\n        vec4 result = vec4(0.);\n        ${mainLoop}\n        setOutput(result);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class Pool2DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv2DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n    const batchFlattenPositionStr = `((batch  * ${convInfo.inHeight} + xR) * ${\n        convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n    const flattenPositionStr =\n        `(xR * ${convInfo.inWidth} + xC) * ${convInfo.inChannels} + d`;\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n        const ivec2 pads = ivec2(${padTop}, ${padLeft});\n\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int batch = coords[0];\n          int d = coords[3];\n\n          ivec2 xRCCorner = coords.yz * strides - pads;\n          int xRCorner = xRCCorner.x;\n          int xCCorner = xRCCorner.y;\n\n          // max/min x(?, ?, d) to get y(yR, yC, d).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n          float avgValue = 0.0;\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n              wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${effectiveFilterWidth};\n                wC += ${dilationWidth}) {\n              int xC = xCCorner + wC;\n\n              if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                continue;\n              }\n\n              float value = getX(batch, xR, xC, d);\n\n              // If a min / max value has already been found, use it. If not,\n              // use the current value.\n              float currMinMaxValue = mix(\n                  value, minMaxValue, minMaxValueFound);\n              if (value ${compareOp} currMinMaxValue) {\n                minMaxValue = value;\n                minMaxValueFound = 1.0;\n                minMaxPosition = ${\n          flattenPositions ? (includeBatchInIndex ? batchFlattenPositionStr :\n                                                    flattenPositionStr) :\n                             `wR * ${effectiveFilterWidth} + wC`};\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec2 strides = ivec2(${strideHeight}, ${strideWidth});\n      const ivec2 pads = ivec2(${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xR, int xC, int d) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xR, xC, d);\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int batch = coords[0];\n        int d = coords[3];\n\n        ivec2 xRCCorner = coords.yz * strides - pads;\n        int xRCorner = xRCCorner.x;\n        int xCCorner = xRCCorner.y;\n\n        // max/min x(?, ?, d) to get y(yR, yC, d).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n          int xR = xRCorner + wR;\n\n          if (xR < 0 || xR >= ${convInfo.inHeight}) {\n            continue;\n          }\n\n          for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n            int xC = xCCorner + wC * ${dilationWidth};\n\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              getValue(batch, xR, xC + 3 * ${dilationWidth}, d)\n            );\n\n            ${updateSnippet}\n          }\n\n          int xC = xCCorner + ${filterWidthNearestVec4};\n          if (${filterWidthVec4Remainder === 1}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              initializationValue,\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 2}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              initializationValue,\n              initializationValue\n            );\n\n            ${updateSnippet}\n          } else if (${filterWidthVec4Remainder === 3}) {\n            vec4 values = vec4(\n              getValue(batch, xR, xC, d),\n              getValue(batch, xR, xC + ${dilationWidth}, d),\n              getValue(batch, xR, xC + 2 * ${dilationWidth}, d),\n              initializationValue\n            );\n\n            ${updateSnippet}\n          }\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n\nexport class Pool3DProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      convInfo: backend_util.Conv3DInfo, poolType: 'max'|'avg',\n      computePositions: boolean, flattenPositions = false,\n      includeBatchInIndex = false) {\n    if (poolType === 'avg' && computePositions) {\n      throw new Error('Cannot compute positions for average pool.');\n    }\n\n    const filterWidth = convInfo.filterWidth;\n    const strideDepth = convInfo.strideDepth;\n    const strideHeight = convInfo.strideHeight;\n    const strideWidth = convInfo.strideWidth;\n    const dilationDepth = convInfo.dilationDepth;\n    const dilationHeight = convInfo.dilationHeight;\n    const dilationWidth = convInfo.dilationWidth;\n    const effectiveFilterDepth = convInfo.effectiveFilterDepth;\n    const effectiveFilterHeight = convInfo.effectiveFilterHeight;\n    const effectiveFilterWidth = convInfo.effectiveFilterWidth;\n\n    const padFront = convInfo.padInfo.front;\n    const padTop = convInfo.padInfo.top;\n    const padLeft = convInfo.padInfo.left;\n    this.outputShape = convInfo.outShape;\n\n    const isAvgPool = poolType === 'avg';\n\n    let initializationValue = '0.0';\n    if (!isAvgPool) {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n    }\n\n    if (computePositions) {\n      const compareOp = '>=';\n\n      this.userCode = `\n        const ivec3 strides =\n            ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n        const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n\n        void main() {\n          ivec5 coords = getOutputCoords();\n          int batch = coords.x;\n          int ch = coords.u;\n\n          ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n          int xDCorner = xCorner.x;\n          int xRCorner = xCorner.y;\n          int xCCorner = xCorner.z;\n\n          // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).\n          // ? = to be determined\n          float minMaxValue = 0.0;\n          float minMaxValueFound = 0.0;\n          int minMaxPosition = 0;\n\n          for (int wD = 0; wD < ${effectiveFilterDepth};\n              wD += ${dilationDepth}) {\n            int xD = xDCorner + wD;\n\n            if (xD < 0 || xD >= ${convInfo.inDepth}) {\n              continue;\n            }\n\n            for (int wR = 0; wR < ${effectiveFilterHeight};\n                wR += ${dilationHeight}) {\n              int xR = xRCorner + wR;\n\n              if (xR < 0 || xR >= ${convInfo.inHeight}) {\n                continue;\n              }\n\n              for (int wC = 0; wC < ${effectiveFilterWidth};\n                  wC += ${dilationWidth}) {\n                int xC = xCCorner + wC;\n\n                if (xC < 0 || xC >= ${convInfo.inWidth}) {\n                  continue;\n                }\n\n                float value = getX(batch, xD, xR, xC, ch);\n\n                // If a min / max value has already been found, use it. If not,\n                // use the current value.\n                float currMinMaxValue = mix(\n                    value, minMaxValue, minMaxValueFound);\n                if (value ${compareOp} currMinMaxValue) {\n                  minMaxValue = value;\n                  minMaxValueFound = 1.0;\n                  minMaxPosition = ${\n          flattenPositions ?\n              (includeBatchInIndex ?\n                   `(((batch * ${convInfo.inDepth} + xD) * ${\n                       convInfo.inHeight} + xR) * ${convInfo.inWidth} + xC) * ${\n                       convInfo.inChannels} + ch` :\n                   `((xD * ${convInfo.inHeight} + xR) * ${\n                       convInfo.inWidth} + xC) * ${convInfo.inChannels} + ch`) :\n              `wD * ${effectiveFilterHeight} * ${effectiveFilterWidth} +\n                      wR * ${effectiveFilterWidth} + wC`};\n                }\n              }\n            }\n          }\n          setOutput(float(minMaxPosition));\n        }\n      `;\n      return;\n    }\n\n    const compareOp = 'max';\n\n    let returnValue = `${poolType}(${poolType}(${poolType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n    if (poolType === 'avg') {\n      returnValue = `avgValue / count`;\n    }\n\n    const filterWidthNearestVec4 = Math.floor(filterWidth / 4) * 4;\n    const filterWidthVec4Remainder = filterWidth % 4;\n\n    const updateSnippet = `\n      if (${isAvgPool}) {\n        avgValue += dot(values, ones);\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    this.userCode = `\n      const ivec3 strides =\n        ivec3(${strideDepth}, ${strideHeight}, ${strideWidth});\n      const ivec3 pads = ivec3(${padFront}, ${padTop}, ${padLeft});\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float count = 0.0;\n\n      float getValue(int batch, int xD, int xR, int xC, int ch) {\n        if (xC < 0 || xC >= ${convInfo.inWidth}) {\n          return initializationValue;\n        }\n        count += 1.0;\n        return getX(batch, xD, xR, xC, ch);\n      }\n\n      void main() {\n        ivec5 coords = getOutputCoords();\n        int batch = coords.x;\n        int ch = coords.u;\n\n        ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;\n        int xDCorner = xCorner.x;\n        int xRCorner = xCorner.y;\n        int xCCorner = xCorner.z;\n\n        // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).\n        // ? = to be determined\n        vec4 minMaxValue = vec4(${initializationValue});\n        float avgValue = 0.0;\n        count = 0.0;\n\n        for (int wD = 0; wD < ${effectiveFilterDepth};\n            wD += ${dilationDepth}) {\n          int xD = xDCorner + wD;\n\n          if (xD < 0 || xD >= ${convInfo.inDepth}) {\n            continue;\n          }\n\n          for (int wR = 0; wR < ${effectiveFilterHeight};\n            wR += ${dilationHeight}) {\n            int xR = xRCorner + wR;\n\n            if (xR < 0 || xR >= ${convInfo.inHeight}) {\n              continue;\n            }\n\n            for (int wC = 0; wC < ${filterWidthNearestVec4}; wC += 4) {\n              int xC = xCCorner + wC * ${dilationWidth};\n\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 3 * ${dilationWidth}, ch)\n              );\n\n              ${updateSnippet}\n            }\n\n            int xC = xCCorner + ${filterWidthNearestVec4};\n            if (${filterWidthVec4Remainder === 1}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                initializationValue,\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 2}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                initializationValue,\n                initializationValue\n              );\n\n              ${updateSnippet}\n            } else if (${filterWidthVec4Remainder === 3}) {\n              vec4 values = vec4(\n                getValue(batch, xD, xR, xC, ch),\n                getValue(batch, xD, xR, xC + ${dilationWidth}, ch),\n                getValue(batch, xD, xR, xC + 2 * ${dilationWidth}, ch),\n                initializationValue\n              );\n\n              ${updateSnippet}\n            }\n          }\n          setOutput(${returnValue});\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ReduceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      reduceInfo: backend_util.ReduceInfo,\n      reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod') {\n    const {windowSize, batchSize, inSize, outSize} = reduceInfo;\n    this.outputShape = [batchSize, outSize];\n\n    let initializationValue = '0.0';\n    let compareOp = ``;\n\n    if (reduceType === 'prod') {\n      initializationValue = '1.0';\n    } else if (reduceType === 'min') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '1.0 / 1e-20';\n      compareOp = `min`;\n    } else if (reduceType === 'max') {\n      // WebGL on Firefox Linux can't compile 1/0 so we do 1/eps.\n      initializationValue = '-1.0 / 1e-20';\n      compareOp = `max`;\n    }\n\n    let returnValue = `${reduceType}(${reduceType}(${reduceType}(` +\n        'minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])';\n\n    if (reduceType === 'sum') {\n      returnValue = `sumValue`;\n    } else if (reduceType === 'prod') {\n      returnValue = `prodValue`;\n    } else if (reduceType === 'all') {\n      returnValue = `allValue`;\n    } else if (reduceType === 'any') {\n      returnValue = `anyValue`;\n    }\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    let updateSnippet = `\n      if (${reduceType === 'sum'}) {\n        sumValue += dot(values, ones);\n      } else if (${reduceType === 'prod'}) {\n        vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);\n        prodValue *= tmp[0] * tmp[1];\n      } else {\n        minMaxValue = ${compareOp}(values, minMaxValue);\n      }\n    `;\n\n    let vecType = `vec4`;\n\n    if (reduceType === 'all') {\n      initializationValue = '1.0';\n      updateSnippet = `\n        bool reducedAllValue = all(values);\n        float floatedReducedAllValue = float(reducedAllValue);\n        allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    } else if (reduceType === 'any') {\n      initializationValue = '0.0';\n      updateSnippet = `\n        bool reducedAnyValue = any(values);\n        float floatedReducedAnyValue = float(reducedAnyValue);\n        anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);\n      `;\n      vecType = `bvec4`;\n    }\n\n    let checkOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n      const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);\n\n      float getValue(int batch, int inIdx) {\n        ${checkOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = outIdx * ${windowSize};\n\n        vec4 minMaxValue = vec4(${initializationValue});\n        float prodValue = 1.0;\n        float sumValue = 0.0;\n        float allValue = 1.0;\n        float anyValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          ${vecType} values = ${vecType}(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport * as shader_util from './shader_compiler_util';\n\nexport class ReshapePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: [number, number, number], inputShape: [\n    number, number, number\n  ]) {\n    this.outputShape = outputShape;\n\n    let mainLoop = ``;\n    for (let i = 0; i < 4; i++) {\n      let thisRC = `thisRC = rc;`;\n      if (i % 2 === 1) {\n        thisRC += `thisRC.z += 1;`;\n      }\n      if (i > 1) {\n        thisRC += `thisRC.y += 1;`;\n      }\n\n      mainLoop += `\n        ${thisRC}\n        ${i > 0 ? `if(thisRC.y < rows && thisRC.z < cols){` : ''}\n          int flatIndex = getFlatIndex(thisRC);\n\n          ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);\n          vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));\n\n          result[${i}] =\n            getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);\n        ${i > 0 ? '}' : ''}\n      `;\n    }\n\n    this.userCode = `\n      ${getReshapedInputCoords(inputShape)}\n      ${shader_util.getFlatIndexFrom3D(outputShape)}\n\n      void main() {\n        ivec3 rc = getOutputCoords();\n\n        vec4 result = vec4(0.);\n\n        ivec3 thisRC;\n        int rows = ${outputShape[1]};\n        int cols = ${outputShape[2]};\n\n        ${mainLoop}\n\n        setOutput(result);\n      }\n    `;\n  }\n}\n\nfunction getReshapedInputCoords(shape: [number, number, number]): string {\n  const coordsFromIndexSnippet =\n      shader_util.getLogicalCoordinatesFromFlatIndex(['r', 'c', 'd'], shape);\n\n  return `\n    ivec3 inputCoordsFromReshapedOutCoords(int index) {\n      ${coordsFromIndexSnippet}\n      return ivec3(r, c, d);\n    }\n  `;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(startRLerp - float(winHeight / 2));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(startCLerp - float(winWidth / 2));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float dxR = float(dyR) * heightScale;\n            int topDxRIndex = int(floor(dxR));\n            int bottomDxRIndex = int(min(ceil(dxR), ${xHeight - 1}.0));\n            float dxRLerp = dxR - float(topDxRIndex);\n            float inverseDxRLerp = 1.0 - dxRLerp;\n\n            float dxC = float(dyC) * widthScale;\n            int leftDxCIndex = int(floor(dxC));\n            int rightDxCIndex = int(min(ceil(dxC), ${xWidth - 1}.0));\n            float dxCLerp = dxC - float(leftDxCIndex);\n            float inverseDxCLerp = 1.0 - dxCLerp;\n\n            if (r == topDxRIndex && c == leftDxCIndex) {\n              // topLeft\n              accumulator +=\n                getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;\n            }\n\n            if (r == topDxRIndex && c == rightDxCIndex) {\n              // topRight\n              accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == leftDxCIndex) {\n              // bottomLeft\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;\n            }\n\n            if (r == bottomDxRIndex && c == rightDxCIndex) {\n              // bottomRight\n              accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec2 sourceFloorRC = ivec2(sourceFracIndexRC);\n        ivec2 sourceCeilRC = ivec2(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);\n        float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);\n        float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);\n        float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);\n\n        vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);\n\n        float top = topLeft + (topRight - topLeft) * fracRC.y;\n        float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;\n        float newValue = top + (bottom - top) * fracRC.x;\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeBilinearPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    this.userCode = `\n      const vec3 effectiveInputOverOutputRatioRC = vec3(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec3 inputShapeRC = vec3(${oldHeight}.0, ${oldWidth}.0,\n                                     ${oldWidth}.0);\n\n      float getAValue(int b, int r, int c, int d) {\n        return getChannel(getA(b, r, c, d), vec2(c, d));\n      }\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        // Calculate values for next column in yRC.z.\n        ivec3 yRC = coords.yzz + ivec3(0, 0, 1);\n\n        // Fractional source index.\n        vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the four integer indices.\n        ivec3 sourceFloorRC = ivec3(sourceFracIndexRC);\n        ivec3 sourceCeilRC = ivec3(\n          min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));\n\n        // Should we calculate next column and row elements in 2x2 packed cell.\n        bool hasNextCol = d < ${depth - 1};\n        bool hasNextRow = coords.z < ${newWidth - 1};\n\n        // In parallel, construct four corners for all four components in\n        // packed 2x2 cell.\n        vec4 topLeft = vec4(\n          getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 bottomLeft = vec4(\n          getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);\n\n        vec4 topRight = vec4(\n          getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec4 bottomRight = vec4(\n          getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),\n          hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)\n                     : 0.0,\n          hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)\n                     : 0.0,\n          (hasNextRow && hasNextCol) ?\n            getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);\n\n        vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);\n\n        vec4 top = mix(topLeft, topRight, fracRC.yyzz);\n        vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);\n        vec4 newValue = mix(top, bottom, fracRC.x);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor4D} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeigborBackpropProgram implements GPGPUProgram {\n  variableNames = ['dy'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(dy: Tensor4D, x: Tensor4D, alignCorners: boolean) {\n    this.outputShape = x.shape;\n    const [, xHeight, xWidth, ] = x.shape;\n    const [, yHeight, yWidth] = dy.shape;\n\n    // In the backwards pass, we want to find the pixels that were generated for\n    // each pixel in the input image the forward pass and add the corresponding\n    // coefficient from dy to the gradient (with some interpolation).\n\n    const effectiveXSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? xHeight - 1 : xHeight,\n      (alignCorners && yWidth > 1) ? xWidth - 1 : xWidth\n    ];\n\n    const effectiveYSize: [number, number] = [\n      (alignCorners && yHeight > 1) ? yHeight - 1 : yHeight,\n      (alignCorners && yWidth > 1) ? yWidth - 1 : yWidth\n    ];\n\n    const heightScale = effectiveXSize[0] / effectiveYSize[0];\n    const widthScale = effectiveXSize[1] / effectiveYSize[1];\n\n    const invHeightScale = 1 / heightScale;\n    const invWidthScale = 1 / widthScale;\n\n    // This defines the size of the window of values around a particular\n    // index in dy that we want to search for contributions to dx.\n    const winHeight = (Math.ceil(invHeightScale) * 2) + 2;\n    const winWidth = (Math.ceil(invWidthScale) * 2) + 2;\n\n    this.userCode = `\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        int r = coords[1];\n        int c = coords[2];\n\n        float accumulator = 0.0;\n\n        const float heightScale = float(${heightScale});\n        const float widthScale = float(${widthScale});\n\n        const float invHeightScale = float(${invHeightScale});\n        const float invWidthScale = float(${invWidthScale});\n\n        const int winHeight = int(${winHeight});\n        const int winWidth = int(${winWidth});\n\n        // Compute bounds for where in dy we will look\n        float startRLerp = floor(float(r) * invHeightScale);\n        int startDyR = int(floor(startRLerp - float(winHeight / 2)));\n\n        float startCLerp = floor(float(c) * invWidthScale);\n        int startDyC = int(floor(startCLerp - float(winWidth / 2)));\n\n        // Loop over dy\n        for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {\n          int dyR = dyROffset + startDyR;\n\n          // Guard against the window exceeding the bounds of dy\n          if (dyR < 0 || dyR >= ${yHeight}) {\n            continue;\n          }\n\n          for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {\n            int dyC = dyCOffset + startDyC;\n\n            // Guard against the window exceeding the bounds of dy\n            if (dyC < 0 || dyC >= ${yWidth}) {\n              continue;\n            }\n\n            float sourceFracRow =\n              float(${effectiveXSize[0]}) *\n                (float(dyR) / float(${effectiveYSize[0]}));\n\n            float sourceFracCol =\n                float(${effectiveXSize[1]}) *\n                  (float(dyC) / float(${effectiveYSize[1]}));\n\n            int sourceNearestRow = int(min(\n                float(int(${xHeight}) - 1),\n                ${alignCorners} ? float(round(sourceFracRow)) :\n                                  float(floor(sourceFracRow))));\n\n            int sourceNearestCol = int(min(\n                float(int(${xWidth}) - 1),\n                ${alignCorners} ? float(round(sourceFracCol)) :\n                                  float(floor(sourceFracCol))));\n\n            if (r == sourceNearestRow && c == sourceNearestCol) {\n              accumulator += getDy(b, dyR, dyC, d);\n            }\n          }\n        }\n        // End loop over dy\n\n        setOutput(accumulator);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class ResizeNearestNeighborProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      inputShape: [number, number, number, number], newHeight: number,\n      newWidth: number, alignCorners: boolean) {\n    const [batch, oldHeight, oldWidth, depth] = inputShape;\n    this.outputShape = [batch, newHeight, newWidth, depth];\n\n    const effectiveInSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? oldHeight - 1 : oldHeight,\n      (alignCorners && newWidth > 1) ? oldWidth - 1 : oldWidth\n    ];\n\n    const effectiveOutSize: [number, number] = [\n      (alignCorners && newHeight > 1) ? newHeight - 1 : newHeight,\n      (alignCorners && newWidth > 1) ? newWidth - 1 : newWidth\n    ];\n\n    // When align corners is false, we rounds the value with floor.\n    const roundBase = alignCorners ? '0.5' : '0.0';\n\n    this.userCode = `\n      const vec2 effectiveInputOverOutputRatioRC = vec2(\n          ${effectiveInSize[0] / effectiveOutSize[0]},\n          ${effectiveInSize[1] / effectiveOutSize[1]});\n      const vec2 inputShapeRC = vec2(${oldHeight}.0, ${oldWidth}.0);\n\n      void main() {\n        ivec4 coords = getOutputCoords();\n        int b = coords[0];\n        int d = coords[3];\n        ivec2 yRC = coords.yz;\n\n        // Fractional source index.\n        vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC;\n\n        // Compute the coordinators of nearest neighbor point.\n        ivec2 sourceNearestRC = ivec2(\n          min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${roundBase})));\n\n        float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);\n\n        setOutput(newValue);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReverseProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n\n    if (rank === 1) {\n      this.userCode = `\n        void main() {\n          int coord = getOutputCoords();\n          setOutput(getX(${xShape[0]} - coord - 1));\n        }\n      `;\n      return;\n    }\n    const getInCoord = (i: number) => {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - coords[${i}] - 1`;\n      }\n      return `coords[${i}]`;\n    };\n    const inCoords = xShape.map((_, i) => getInCoord(i)).join(',');\n    const type = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${type} coords = getOutputCoords();\n        setOutput(getX(${inCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ReversePackedProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(xShape: number[], axis: number[]) {\n    const rank = xShape.length;\n    if (rank > 4) {\n      throw new Error(\n          `WebGL backend: Reverse of rank-${rank} tensor is not yet supported`);\n    }\n    this.outputShape = xShape;\n    const channels = getChannels('rc', rank);\n    const nextColumn =\n        `${channels[rank - 1]} + 1 < ${this.outputShape[rank - 1]}`;\n    const nextRow = `${channels[rank - 2]} + 1 < ${this.outputShape[rank - 2]}`;\n    const type = getCoordsDataType(rank);\n    if (rank === 1) {\n      this.userCode = `\n        void main(){\n          int rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = getChannel(getX(${xShape[0]} - rc - 1),\n            ${xShape[0]} - rc - 1);\n          if(${nextColumn}){\n              result.g = getChannel(getX(${xShape[0]} - (rc  + 1) - 1),\n                ${xShape[0]} - (rc  + 1) - 1);\n          }\n          setOutput(result);\n        }\n      `;\n    } else {\n      this.userCode = `\n        void main() {\n          ${type} rc = getOutputCoords();\n          vec4 result = vec4(0.);\n          result.r = ${getR(channels.slice())};\n          if(${nextColumn}){\n            result.g = ${getG(channels.slice())};\n          }\n          if(${nextRow}) {\n            result.b = ${getB(channels.slice())};\n            if(${nextColumn}) {\n              result.a = ${getA(channels.slice())};\n            }\n          }\n          setOutput(result);\n        }\n    `;\n    }\n\n    function getR(channels: string[]): string {\n      return getChannel(channels);\n    }\n\n    function getG(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getB(channels: string[]): string {\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getA(channels: string[]): string {\n      channels[rank - 1] = '(' + channels[rank - 1] + ` + 1)`;\n      channels[rank - 2] = '(' + channels[rank - 2] + ` + 1)`;\n      return getChannel(channels);\n    }\n\n    function getChannel(channels: string[]): string {\n      const inCoordsArray = xShape.map((_, i) => getInCoord(i, channels));\n      const inCoords = inCoordsArray.join(',');\n      const innerDims = inCoordsArray.slice(-2).join(',');\n      return `getChannel(getX(${inCoords}), vec2(${innerDims}))`;\n    }\n\n    function getInCoord(i: number, channels1: string[]): string {\n      if (axis.indexOf(i) !== -1 && xShape[i] !== 1) {\n        return `${xShape[i]} - ${channels1[i]} - 1`;\n      } else {\n        return `${channels1[i]}`;\n      }\n    }\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class ScatterProgram implements GPGPUProgram {\n  variableNames = ['updates', 'indices', 'defaultValue'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      updateSize: number, sliceDim: number, indicesRank: number,\n      updatesRank: number, strides: number[], shape: number[],\n      summingDupeIndex = true) {\n    this.outputShape = shape;\n    const stridesType = getCoordsDataType(strides.length);\n    const dtype = getCoordsDataType(shape.length);\n    let indicesString = '';\n    if (indicesRank === 1) {\n      indicesString = 'i';\n    } else if (indicesRank === 2) {\n      indicesString = 'i, j';\n    }\n    const indicesSnippet = `getIndices(${indicesString})`;\n\n    let updatesString = '';\n    if (updatesRank === 1) {\n      updatesString = 'i';\n    } else if (updatesRank === 2) {\n      updatesString = 'i, coords[1]';\n    }\n    const updatesSnippet = `getUpdates(${updatesString})`;\n\n    const strideString = sliceDim > 1 ? 'strides[j]' : 'strides';\n    this.userCode = `\n        ${stridesType} strides = ${stridesType}(${strides});\n\n        void main() {\n          ${dtype} coords = getOutputCoords();\n          float sum = 0.0;\n          bool found = false;\n          for (int i = 0; i < ${updateSize}; i++) {\n            int flattenedIndex = 0;\n            for (int j = 0; j < ${sliceDim}; j++) {\n              int index = round(${indicesSnippet});\n              flattenedIndex += index * ${strideString};\n            }\n            if (flattenedIndex == coords[0]) {\n              sum += ${updatesSnippet};\n              found = true;\n            }\n          }\n          setOutput(mix(getDefaultValue(), sum, float(found)));\n        }\n      `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class SegmentOpProgram implements GPGPUProgram {\n  variableNames = ['x', 'segmentIds'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(\n      segOpInfo: backend_util.segment_util.SegOpInfo,\n      segOpType: 'unsortedSegmentSum') {\n    const windowSize = segOpInfo.windowSize;\n    const batchSize = segOpInfo.batchSize;\n    const inSize = segOpInfo.inSize;\n    const numSegments = segOpInfo.numSegments;\n    const outSize = numSegments * Math.ceil(inSize / windowSize);\n    this.outputShape = [batchSize, outSize];\n\n    const initializationValue = '0.0';\n    const returnValue = `sumValue`;\n\n    const windowSizeNearestVec4 = Math.floor(windowSize / 4) * 4;\n    const windowSizeVec4Remainder = windowSize % 4;\n\n    const updateSnippet = `\n        sumValue += dot(values, segFilter);\n    `;\n\n    let checkValueOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkValueOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return initializationValue;\n        }\n      `;\n    }\n\n    let checkSegmentIdOutOfBounds = '';\n    if (inSize % windowSize > 0) {\n      checkSegmentIdOutOfBounds = `\n        if (inIdx < 0 || inIdx >= ${inSize}) {\n          return -1.0;\n        }\n      `;\n    }\n\n    this.userCode = `\n      const float initializationValue = ${initializationValue};\n\n      float getValue(int batch, int inIdx) {\n        ${checkValueOutOfBounds}\n        return getX(batch, inIdx);\n      }\n\n      float getSegmentIdAtIndex(int inIdx) {\n        ${checkSegmentIdOutOfBounds}\n        return getSegmentIds(inIdx);\n      }\n\n      void main() {\n        ivec2 coords = getOutputCoords();\n        int batch = coords[0];\n        int outIdx = coords[1];\n        int inOffset = int(floor(float(outIdx) / float(\n          ${numSegments})) * float(${windowSize}));\n        int currentSeg = int(mod(float(outIdx), float(${numSegments})));\n\n        float sumValue = 0.0;\n\n        for (int i = 0; i < ${windowSizeNearestVec4}; i += 4) {\n          int inIdx = inOffset + i;\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            getValue(batch, inIdx + 3)\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0\n          );\n\n          ${updateSnippet}\n        }\n\n        int inIdx = inOffset + ${windowSizeNearestVec4};\n        if (${windowSizeVec4Remainder === 1}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            initializationValue,\n            initializationValue,\n            initializationValue\n          );\n\n          int inIdxSeg = int(getSegmentIdAtIndex(inIdx));\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            0,\n            0,\n            0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 2}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            initializationValue,\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n              0,\n              0\n          );\n\n          ${updateSnippet}\n        } else if (${windowSizeVec4Remainder === 3}) {\n          vec4 values = vec4(\n            getValue(batch, inIdx),\n            getValue(batch, inIdx + 1),\n            getValue(batch, inIdx + 2),\n            initializationValue\n          );\n\n          vec4 segFilter = vec4(\n            int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,\n            int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,\n            0\n          );\n\n          ${updateSnippet}\n        }\n        setOutput(${returnValue});\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SelectProgram implements GPGPUProgram {\n  variableNames = ['c', 'a', 'b'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(cRank: number, shape: number[], rank: number) {\n    this.outputShape = shape;\n\n    let cCoords;\n    let abCoords;\n    if (rank > 4) {\n      throw Error(`Where for rank ${rank} is not yet supported`);\n    }\n\n    if (rank === 1) {\n      abCoords = `resRC`;\n      cCoords = `resRC`;\n    } else {\n      const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w'];\n      const cCoordVars = [];\n      const abCoordVars = [];\n      for (let i = 0; i < shape.length; i++) {\n        abCoordVars.push(`${currentCoords[i]}`);\n        if (i < cRank) {\n          cCoordVars.push(`${currentCoords[i]}`);\n        }\n      }\n      cCoords = cCoordVars.join();\n      abCoords = abCoordVars.join();\n    }\n\n    const dtype = getCoordsDataType(rank);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        float cVal = getC(${cCoords});\n        if (cVal >= 1.0) {\n          setOutput(getA(${abCoords}));\n        } else {\n          setOutput(getB(${abCoords}));\n        }\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SliceProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const uniformPart = `uniform int start[${this.rank}];`;\n    const sourceCoords = getCoords(this.rank);\n\n    let body: string;\n    const coordSum = destSize.map((_, i) => {\n      return `sourceLoc.${coords[i]} = start[${i}] + coords.${coords[i]};`;\n    });\n    body = `\n        ${dtype} sourceLoc;\n        ${dtype} coords = getOutputCoords();\n        ${coordSum.join('\\n')}\n      `;\n    this.userCode = `\n      ${uniformPart}\n      void main() {\n        ${body}\n        setOutput(getSource(${sourceCoords}));\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n\nconst coords = ['x', 'y', 'z', 'w', 'u', 'v'];\n\nfunction getCoords(rank: number): string {\n  if (rank === 1) {\n    return 'sourceLoc';\n  } else if (rank <= 6) {\n    return coords.slice(0, rank).map(x => 'sourceLoc.' + x).join(',');\n  } else {\n    throw Error(`Slicing for rank ${rank} is not yet supported`);\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class SlicePackedProgram implements GPGPUProgram {\n  variableNames = ['source'];\n  packedInputs = true;\n  packedOutput = true;\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  // Caching uniform location for speed.\n  startLoc: WebGLUniformLocation;\n\n  constructor(destSize: number[]) {\n    this.outputShape = destSize;\n    this.rank = destSize.length;\n\n    const dtype = getCoordsDataType(this.rank);\n    const coords = getChannels('coords', this.rank);\n    const sourceLoc = getChannels('sourceLoc', this.rank);\n\n    const innerDims =\n        this.rank === 1 ? 'sourceLoc' : `vec2(${sourceLoc.slice(-2).join()})`;\n    const getChannel =\n        `getChannel(getSource(${sourceLoc.join()}), ${innerDims})`;\n    const upperRow = `\n      result.x = ${getChannel};\n      if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n        ++${sourceLoc[this.rank - 1]};\n        result.y = ${getChannel};\n        --${sourceLoc[this.rank - 1]};\n      }\n    `;\n    const lowerRow = this.rank === 1 ? '' : `\n      --${coords[this.rank - 1]};\n      if (++${coords[this.rank - 2]} < ${destSize[this.rank - 2]}) {\n        ++${sourceLoc[this.rank - 2]};\n        result.z = ${getChannel};\n        if (++${coords[this.rank - 1]} < ${destSize[this.rank - 1]}) {\n          ++${sourceLoc[this.rank - 1]};\n          result.w = ${getChannel};\n        }\n      }\n    `;\n\n    const sourceLocSetup = this.rank <= 4 ?\n        `sourceLoc = coords +\n            ${dtype}(${destSize.map((_, i) => `start[${i}]`).join()});` :\n        destSize.map((_, i) => `${sourceLoc[i]} = ${coords[i]} + start[${i}];`)\n            .join('\\n');\n    this.userCode = `\n      uniform int start[${this.rank}];\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        ${dtype} sourceLoc;\n        ${sourceLocSetup}\n        vec4 result = vec4(0.);\n        ${upperRow}\n        ${lowerRow}\n        setOutput(result);\n      }\n    `;\n  }\n\n  getCustomSetupFunc(start: number[]) {\n    if (start.length !== this.rank) {\n      throw Error(\n          `The rank (${this.rank}) of the program must match the ` +\n          `length of start (${start.length})`);\n    }\n    return (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => {\n      if (this.startLoc == null) {\n        this.startLoc = gpgpu.getUniformLocationNoThrow(webGLProgram, 'start');\n        if (this.startLoc == null) {\n          // This means the compiler has optimized and realized it doesn't need\n          // the uniform.\n          return;\n        }\n      }\n      gpgpu.gl.uniform1iv(this.startLoc, start);\n    };\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class StridedSliceProgram implements GPGPUProgram {\n  variableNames = ['x'];\n  outputShape: number[];\n  userCode: string;\n\n  constructor(begin: number[], strides: number[], size: number[]) {\n    this.outputShape = size;\n    const rank = size.length;\n    const inputDtype = getCoordsDataType(size.length);\n    const dtype = getCoordsDataType(size.length);\n\n    let newCoords = '';\n    if (rank === 1) {\n      newCoords = 'coords * strides + begin';\n    } else {\n      let outputAxis = 0;\n      newCoords =\n          size.map((_, i) => {\n                outputAxis++;\n                return size.length === 1 ?\n                    `coords * strides[${i}] + begin[${i}]` :\n                    `coords[${outputAxis - 1}] * strides[${i}] + begin[${i}]`;\n              })\n              .join(',');\n    }\n\n    this.userCode = `\n      ${inputDtype} begin = ${inputDtype}(${begin});\n      ${inputDtype} strides = ${inputDtype}(${strides});\n\n      void main() {\n        ${dtype} coords = getOutputCoords();\n        setOutput(getX(${newCoords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport {GPGPUContext} from './gpgpu_context';\nimport {getInternalFormatForFloat16MatrixTexture, getInternalFormatForFloat16PackedMatrixTexture, getInternalFormatForFloat32MatrixTexture, getInternalFormatForPackedMatrixTexture, getInternalFormatForUnsignedBytesMatrixTexture} from './gpgpu_util';\nimport {getPackedMatrixTextureShapeWidthHeight, getUnpackedMatrixTextureShapeWidthHeight, PhysicalTextureType, TextureConfig, TextureUsage} from './tex_util';\n\nexport class TextureManager {\n  private numUsedTextures = 0;\n  private numFreeTextures = 0;\n  private _numBytesAllocated = 0;\n  private _numBytesFree = 0;  // How many bytes that have been allocated\n                              // are available for reuse.\n  private freeTextures: {[shape: string]: WebGLTexture[]} = {};\n  private logEnabled = false;\n  private usedTextures: {[shape: string]: WebGLTexture[]} = {};\n\n  constructor(private gpgpu: GPGPUContext) {}\n\n  acquireTexture(\n      shapeRC: [number, number], usage: TextureUsage,\n      isPacked: boolean): WebGLTexture {\n    const physicalTexType = getPhysicalFromLogicalTextureType(usage, isPacked);\n\n    const shapeKey = getKeyFromTextureShape(shapeRC, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n    if (!(shapeKey in this.usedTextures)) {\n      this.usedTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shapeRC, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n\n    if (this.freeTextures[shapeKey].length > 0) {\n      this.numFreeTextures--;\n      this.numUsedTextures++;\n      this._numBytesFree -= texBytes;\n      this.log();\n      const newTexture = this.freeTextures[shapeKey].shift();\n      this.usedTextures[shapeKey].push(newTexture);\n      return newTexture;\n    }\n\n    let newTexture: WebGLTexture;\n    if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT32) {\n      newTexture = this.gpgpu.createPackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.PACKED_2X2_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16PackedMatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT32) {\n      newTexture =\n          this.gpgpu.createFloat32MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (physicalTexType === PhysicalTextureType.UNPACKED_FLOAT16) {\n      newTexture =\n          this.gpgpu.createFloat16MatrixTexture(shapeRC[0], shapeRC[1]);\n    } else if (\n        physicalTexType === PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE) {\n      newTexture =\n          this.gpgpu.createUnsignedBytesMatrixTexture(shapeRC[0], shapeRC[1]);\n    }\n    this.usedTextures[shapeKey].push(newTexture);\n\n    this.numUsedTextures++;\n    this._numBytesAllocated += texBytes;\n    this.log();\n\n    return newTexture;\n  }\n\n  releaseTexture(\n      texture: WebGLTexture, shape: [number, number],\n      logicalTexType: TextureUsage, isPacked: boolean): void {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    const physicalTexType =\n        getPhysicalFromLogicalTextureType(logicalTexType, isPacked);\n    const shapeKey = getKeyFromTextureShape(shape, physicalTexType, isPacked);\n    if (!(shapeKey in this.freeTextures)) {\n      this.freeTextures[shapeKey] = [];\n    }\n\n    const texBytes = computeBytes(\n        shape, physicalTexType, this.gpgpu.gl, this.gpgpu.textureConfig,\n        isPacked);\n    const deleteTexThreshold = env().get('WEBGL_DELETE_TEXTURE_THRESHOLD');\n    if (deleteTexThreshold !== -1 &&\n        this._numBytesAllocated > deleteTexThreshold) {\n      this.gpgpu.deleteMatrixTexture(texture);\n      this._numBytesAllocated -= texBytes;\n    } else {\n      this.freeTextures[shapeKey].push(texture);\n      this.numFreeTextures++;\n      this._numBytesFree += texBytes;\n    }\n\n    this.numUsedTextures--;\n\n    const texList = this.usedTextures[shapeKey];\n    const texIndex = texList.indexOf(texture);\n    if (texIndex < 0) {\n      throw new Error(\n          'Cannot release a texture that was never provided by this ' +\n          'texture manager');\n    }\n    texList.splice(texIndex, 1);\n    this.log();\n  }\n\n  private log() {\n    if (!this.logEnabled) {\n      return;\n    }\n    const total = this.numFreeTextures + this.numUsedTextures;\n    console.log(\n        'Free/Used', `${this.numFreeTextures} / ${this.numUsedTextures}`,\n        `(${total})`);\n    const freeRatio = this._numBytesFree / this._numBytesAllocated;\n    console.log(`Bytes allocated: ${this._numBytesAllocated}`);\n    console.log(`Bytes unused: ${this._numBytesFree} (${\n        Math.round(100 * freeRatio)}%)`);\n  }\n\n  get numBytesAllocated(): number {\n    return this._numBytesAllocated;\n  }\n\n  get numBytesFree(): number {\n    return this._numBytesFree;\n  }\n\n  getNumUsedTextures(): number {\n    return this.numUsedTextures;\n  }\n\n  getNumFreeTextures(): number {\n    return this.numFreeTextures;\n  }\n\n  dispose() {\n    if (this.freeTextures == null) {\n      // Already disposed.\n      return;\n    }\n    for (const texShape in this.freeTextures) {\n      this.freeTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    for (const texShape in this.usedTextures) {\n      this.usedTextures[texShape].forEach(tex => {\n        this.gpgpu.deleteMatrixTexture(tex);\n      });\n    }\n    this.freeTextures = null;\n    this.usedTextures = null;\n    this.numUsedTextures = 0;\n    this.numFreeTextures = 0;\n    this._numBytesAllocated = 0;\n    this._numBytesFree = 0;\n  }\n}\n\nfunction numBytesForInternalFormat(\n    gl: WebGLRenderingContext, internalFormat: number): number {\n  // tslint:disable-next-line:no-any\n  const glany = gl as any;\n  if (internalFormat === glany.R32F) {\n    return 4;\n  } else if (internalFormat === glany.R16F) {\n    return 2;\n  } else if (internalFormat === glany.RGBA32F) {\n    return 16;\n  } else if (internalFormat === gl.RGBA) {\n    return 16;\n  } else if (internalFormat === glany.RGBA16F) {\n    return 8;\n  }\n  throw new Error(`Unknown internal format ${internalFormat}`);\n}\n\nexport function computeBytes(\n    shape: [number, number], physicalTexType: PhysicalTextureType,\n    gl: WebGLRenderingContext, textureConfig: TextureConfig,\n    isPacked: boolean): number {\n  // It is not possible to infer packed status from the texture type because\n  // depending on the textureConfig, different  texture types may resolve to the\n  // same internal format (e.g. in WebGL1, the internal format for\n  // UNPACKED_FLOAT16 textures is gl.RGBA). Therefore we pass in `isPacked`\n  // explicitly.\n  const internalFormat =\n      internalFormatForPhysicalTexType(physicalTexType, textureConfig);\n\n  let numElements: number;\n  if (isPacked) {\n    const [packedWidth, packedHeight] =\n        getPackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = packedWidth * packedHeight;\n\n  } else {\n    const [width, height] =\n        getUnpackedMatrixTextureShapeWidthHeight(shape[0], shape[1]);\n    numElements = width * height;\n  }\n\n  const bytesPerElement = numBytesForInternalFormat(gl, internalFormat);\n  return numElements * bytesPerElement;\n}\n\nfunction internalFormatForPhysicalTexType(\n    physicalTexType: PhysicalTextureType,\n    textureConfig: TextureConfig): number {\n  switch (physicalTexType) {\n    case PhysicalTextureType.PACKED_2X2_FLOAT32:\n      return getInternalFormatForPackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_2X2_FLOAT16:\n      return getInternalFormatForFloat16PackedMatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT32:\n      return getInternalFormatForFloat32MatrixTexture(textureConfig);\n    case PhysicalTextureType.UNPACKED_FLOAT16:\n      return getInternalFormatForFloat16MatrixTexture(textureConfig);\n    case PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE:\n      return getInternalFormatForUnsignedBytesMatrixTexture(textureConfig);\n    default:\n      throw new Error(`Unknown physical texture type ${physicalTexType}`);\n  }\n}\n\nfunction getPhysicalTextureForRendering(isPacked: boolean):\n    PhysicalTextureType {\n  if (env().getBool('WEBGL_RENDER_FLOAT32_ENABLED')) {\n    if (isPacked) {\n      return PhysicalTextureType.PACKED_2X2_FLOAT32;\n    }\n    return PhysicalTextureType.UNPACKED_FLOAT32;\n  }\n\n  if (isPacked) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT16;\n  }\n  return PhysicalTextureType.UNPACKED_FLOAT16;\n}\n\nfunction getPhysicalFromLogicalTextureType(\n    logicalTexType: TextureUsage, isPacked: boolean): PhysicalTextureType {\n  if (logicalTexType === TextureUsage.UPLOAD) {\n    return PhysicalTextureType.PACKED_2X2_FLOAT32;\n  } else if (logicalTexType === TextureUsage.RENDER || logicalTexType == null) {\n    return getPhysicalTextureForRendering(isPacked);\n  } else if (\n      logicalTexType === TextureUsage.DOWNLOAD ||\n      logicalTexType === TextureUsage.PIXELS) {\n    return PhysicalTextureType.PACKED_4X1_UNSIGNED_BYTE;\n  }\n  throw new Error(`Unknown logical texture type ${logicalTexType}`);\n}\n\nfunction getKeyFromTextureShape(\n    shapeRowsCol: [number, number], physicalTexType: PhysicalTextureType,\n    isPacked: boolean): string {\n  return `${shapeRowsCol[0]}_${shapeRowsCol[1]}_${physicalTexType}_${isPacked}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TileProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], reps: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[i] * reps[i];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const sourceCoords = getSourceCoords(aShape);\n\n    this.userCode = `\n      void main() {\n        ${dtype} resRC = getOutputCoords();\n        setOutput(getA(${sourceCoords}));\n      }\n    `;\n  }\n}\n\nfunction getSourceCoords(aShape: number[]): string {\n  const rank = aShape.length;\n  if (rank > 5) {\n    throw Error(`Tile for rank ${rank} is not yet supported`);\n  }\n  if (rank === 1) {\n    return `imod(resRC, ${aShape[0]})`;\n  }\n\n  const currentCoords = ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u'];\n\n  const sourceCoords = [];\n  for (let i = 0; i < aShape.length; i++) {\n    sourceCoords.push(`imod(${currentCoords[i]}, ${aShape[i]})`);\n  }\n  return sourceCoords.join();\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class UnaryOpProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      float unaryOperation(float x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        float x = getAAtOutCoords();\n        float y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n\nconst CHECK_NAN_SNIPPET = `if (isnan(x)) return x;`;\n\nexport const LINEAR = `return x;`;\n\nexport const ABS = `return abs(x);`;\n\nexport const RELU = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : x;\n`;\n\nexport const RELU6 = CHECK_NAN_SNIPPET + `\n  return (x < 0.0) ? 0.0 : min(6.0, x);\n`;\n\nexport const ELU = `return (x >= 0.0) ? x : (exp(x) - 1.0);`;\n\nexport const SELU = `\n  // Stable and Attracting Fixed Point (0, 1) for Normalized Weights.\n  // see: https://arxiv.org/abs/1706.02515\n  float scaleAlpha = ${backend_util.SELU_SCALEALPHA};\n  float scale = ${backend_util.SELU_SCALE};\n  return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);\n`;\n\nexport function STEP(alpha = 0.0) {\n  return CHECK_NAN_SNIPPET + `\n    return x > 0.0 ? 1.0 : float(${alpha});\n  `;\n}\n\nexport const NEG = `return -x;`;\n\nexport const CEIL = `return ceil(x);`;\n\nexport const FLOOR = `return floor(x);`;\n\nexport const SIGN = `\n  if (isnan(x)) { return 0.0; }\n  return sign(x);\n`;\n\nexport const IS_NAN = `return float(isnan(x));`;\n\nexport const IS_INF = `return float(isinf(x));`;\n\nexport const IS_FINITE = `return float(!isnan(x) && !isinf(x));`;\n\nexport const ROUND = `\n  // OpenGL ES does not support round function.\n  // The algorithm is based on banker's rounding.\n  float base = floor(x);\n  if ((x - base) < 0.5) {\n    return floor(x);\n  } else if ((x - base) > 0.5) {\n    return ceil(x);\n  } else {\n    if (mod(base, 2.0) == 0.0) {\n      return base;\n    } else {\n      return base + 1.0;\n    }\n  }\n`;\n\nexport const EXP = `return exp(x);`;\n\nexport const EXPM1 = `return exp(x) - 1.0;`;\n\nexport const LOG = `if (x < 0.0) return NAN;\n  return log(x);`;\n\nexport const LOG1P = `return log(1.0 + x);`;\n\nexport const SQRT = `return sqrt(x);`;\n\nexport const RSQRT = `return inversesqrt(x);`;\n\nexport const SIGMOID = `return 1.0 / (1.0 + exp(-1.0 * x));`;\n\n/**\n * mirrors the implementation of tf.nn.softplus: https://goo.gl/vkcvwX\n *\n * epsilon is the difference between 1.0 and the next representable\n * float. For a single precision 32 bit float this should be 2^-23, see:\n * https://math.byu.edu/~schow/work/IEEEFloatingPoint.htm\n *\n * too_large = (x > -threshold) is value above which exp(x) may overflow\n * but softplus(x) == x is within machine epsilon\n *\n * too_small = (x < threshold) is value below which exp(x) may underflow,\n * but softplus(x) == exp(x) is within machine epsilon.\n */\nexport const SOFTPLUS = `\n  float epsilon = 1.1920928955078125e-7;\n  float threshold = log(epsilon) + 2.0;\n\n  bool too_large = x > -threshold;\n  bool too_small = x < threshold;\n\n  float result;\n  float exp_x = exp(x);\n\n  if (too_large){\n    result = x;\n  }\n  else if (too_small){\n    result = exp_x;\n  }\n  else{\n    result = log(exp_x + 1.0);\n  }\n  return result;\n`;\n\nexport const ASIN = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return asin(x);\n`;\n\nexport const ACOS = CHECK_NAN_SNIPPET + `\n  if (abs(x) > 1.) {\n    return NAN;\n  }\n  return acos(x);\n`;\n\nexport const ATAN = CHECK_NAN_SNIPPET + `\n  return atan(x);\n`;\n\nexport const SINH = `\n  float e2x = exp(x);\n  return (e2x - 1.0 / e2x) / 2.0;\n`;\n\nexport const COSH = `\n  float e2x = exp(-x);\n  return (e2x + 1.0 / e2x) / 2.0;\n`;\n\nexport const TANH = `\n  float e2x = exp(-2.0 * abs(x));\n  return sign(x) * (1.0 - e2x) / (1.0 + e2x);\n`;\n\nexport const ASINH = CHECK_NAN_SNIPPET + `return log(x + sqrt(x * x + 1.0));`;\n\nexport const ACOSH = CHECK_NAN_SNIPPET + `\n  if (x < 1.0) return NAN;\n  return log(x + sqrt(x * x - 1.0));`;\n\nexport const ATANH = CHECK_NAN_SNIPPET + `\n  if ((x < -1.0) || (x > 1.0)) return NAN;\n  return (log(1.0 + x) - log(1.0 - x)) / 2.0;`;\n\nexport const ERF = `\n  // Error function is calculated approximately with elementary function.\n  // See \"Handbook of Mathematical Functions with Formulas,\n  // Graphs, and Mathematical Tables\", Abramowitz and Stegun.\n  float p = ${backend_util.ERF_P};\n  float a1 = ${backend_util.ERF_A1};\n  float a2 = ${backend_util.ERF_A2};\n  float a3 = ${backend_util.ERF_A3};\n  float a4 = ${backend_util.ERF_A4};\n  float a5 = ${backend_util.ERF_A5};\n\n  float sign = sign(x);\n  x = abs(x);\n  float t = 1.0 / (1.0 + p * x);\n  return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));\n`;\n\nexport const RECIPROCAL = `return 1.0 / x;`;\n\nexport const LOGICAL_NOT = `return float(!(x >= 1.0));`;\n\nexport const TO_INT = `return float(int(x));`;\n\nexport const CLONE = 'return x;';\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport const LINEAR = `return x;`;\n\nexport const LOG = `\n  vec4 result = log(x);\n  vec4 isNaN = vec4(lessThan(x, vec4(0.0)));\n  result.r = isNaN.r == 1.0 ? NAN : result.r;\n  result.g = isNaN.g == 1.0 ? NAN : result.g;\n  result.b = isNaN.b == 1.0 ? NAN : result.b;\n  result.a = isNaN.a == 1.0 ? NAN : result.a;\n\n  return result;\n`;\n\nexport const RELU = `\n  vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const RELU6 = `\n  vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));\n  bvec4 isNaN = isnan(x);\n\n  result.r = isNaN.r ? x.r : result.r;\n  result.g = isNaN.g ? x.g : result.g;\n  result.b = isNaN.b ? x.b : result.b;\n  result.a = isNaN.a ? x.a : result.a;\n\n  return result;\n`;\n\nexport const ELU = `\n  vec4 result;\n\n  result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);\n  result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);\n  result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);\n  result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);\n\n  return result;\n`;\n\nexport class UnaryOpPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], opSnippet: string) {\n    this.outputShape = aShape;\n    this.userCode = `\n      vec4 unaryOperation(vec4 x) {\n        ${opSnippet}\n      }\n\n      void main() {\n        vec4 x = getAAtOutCoords();\n        vec4 y = unaryOperation(x);\n\n        setOutput(y);\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getChannels, getSourceCoords} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class UnpackProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  packedInputs = true;\n  packedOutput = false;\n  outputShape: number[];\n  userCode: string;\n\n  constructor(outputShape: number[]) {\n    this.outputShape = outputShape;\n    const rank = outputShape.length;\n\n    const channels = getChannels('rc', rank);\n    const dtype = getCoordsDataType(rank);\n    const sourceCoords = getSourceCoords(rank, channels);\n    const innerDims = channels.slice(-2);\n    const coords = rank <= 1 ? 'rc' : `vec2(${innerDims.join(',')})`;\n\n    this.userCode = `\n      void main() {\n        ${dtype} rc = getOutputCoords();\n        vec4 packedInput = getA(${sourceCoords});\n\n        setOutput(getChannel(packedInput, ${coords}));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Import webgl flags.\nimport './flags_webgl';\n\nimport * as tf from '@tensorflow/tfjs-core';\nimport {complex, DataId, div, engine, env, imag, max, MemoryInfo, range, real, RecursiveArray, reshape, scalar, softmax, tensor, tidy, TimingInfo, transpose} from '@tensorflow/tfjs-core';\nimport {backend_util, buffer, kernel_impls, slice_util, util} from '@tensorflow/tfjs-core';\nimport {DataStorage, DataType, KernelBackend, NumericDataType, Rank, Scalar, ShapeMap, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorInfo, TypedArray, upcastType} from '@tensorflow/tfjs-core';\n\nimport {addImplCPU, ceilImplCPU, expImplCPU, expm1ImplCPU, floorImplCPU, logImplCPU, multiplyImplCPU, rsqrtImplCPU, simpleAbsImplCPU, sliceImplCPU, subImplCPU} from './kernel_utils/shared';\n\nconst {segment_util} = backend_util;\nconst split = kernel_impls.split;\nconst tile = kernel_impls.tile;\nconst topkImpl = kernel_impls.topkImpl;\nconst whereImpl = kernel_impls.whereImpl;\n\nimport {AddNProgram} from './addn_gpu';\nimport {AddNPackedProgram} from './addn_packed_gpu';\nimport {ArgMinMaxProgram} from './argminmax_gpu';\nimport {ArgMinMaxPackedProgram} from './argminmax_packed_gpu';\nimport {AvgPool3DBackpropProgram} from './avg_pool_backprop_gpu';\nimport * as binaryop_complex_gpu from './binaryop_complex_gpu';\nimport {BinaryOpComplexProgram} from './binaryop_complex_gpu';\nimport * as binaryop_gpu from './binaryop_gpu';\nimport {BinaryOpProgram} from './binaryop_gpu';\nimport * as binaryop_packed_gpu from './binaryop_packed_gpu';\nimport {BinaryOpPackedProgram} from './binaryop_packed_gpu';\nimport {getWebGLContext} from './canvas_util';\nimport {ClipProgram} from './clip_gpu';\nimport {ClipPackedProgram} from './clip_packed_gpu';\nimport {ComplexAbsProgram} from './complex_abs_gpu';\nimport {ConcatProgram} from './concat_gpu';\nimport {ConcatPackedProgram} from './concat_packed_gpu';\nimport {Conv2DDerFilterProgram, Conv2DDerInputProgram, Conv3DDerFilterProgram, Conv3DDerInputProgram} from './conv_backprop_gpu';\nimport {DepthwiseConv2DDerFilterProgram, DepthwiseConv2DDerInputProgram} from './conv_backprop_gpu_depthwise';\nimport {Conv2DProgram, Conv3DProgram} from './conv_gpu';\nimport {DepthwiseConv2DProgram} from './conv_gpu_depthwise';\nimport {DepthwiseConvPacked2DProgram} from './conv_packed_gpu_depthwise';\nimport {CropAndResizeProgram} from './crop_and_resize_gpu';\nimport {CumSumProgram} from './cumsum_gpu';\nimport {DecodeMatrixProgram} from './decode_matrix_gpu';\nimport {DecodeMatrixPackedProgram} from './decode_matrix_packed_gpu';\nimport {DepthToSpaceProgram} from './depth_to_space_gpu';\nimport {DiagProgram} from './diag_gpu';\nimport {EncodeFloatProgram} from './encode_float_gpu';\nimport {EncodeFloatPackedProgram} from './encode_float_packed_gpu';\nimport {EncodeMatrixProgram} from './encode_matrix_gpu';\nimport {EncodeMatrixPackedProgram} from './encode_matrix_packed_gpu';\nimport * as fft_gpu from './fft_gpu';\nimport {FFTProgram} from './fft_gpu';\nimport {FillProgram} from './fill_gpu';\nimport {GatherProgram} from './gather_gpu';\nimport {GatherNDProgram} from './gather_nd_gpu';\nimport {GPGPUContext} from './gpgpu_context';\nimport * as gpgpu_math from './gpgpu_math';\nimport {GPGPUBinary, GPGPUProgram, TensorData} from './gpgpu_math';\nimport {Im2ColPackedProgram} from './im2col_packed_gpu';\nimport {LRNProgram} from './lrn_gpu';\nimport {LRNGradProgram} from './lrn_grad_gpu';\nimport {LRNPackedProgram} from './lrn_packed_gpu';\nimport {MaxPool3DBackpropProgram} from './max_pool_backprop_gpu';\nimport {MatMulPackedProgram} from './mulmat_packed_gpu';\nimport {MultinomialProgram} from './multinomial_gpu';\nimport {OneHotProgram} from './onehot_gpu';\nimport {PackProgram} from './pack_gpu';\nimport {PadProgram} from './pad_gpu';\nimport {PadPackedProgram} from './pad_packed_gpu';\nimport {Pool3DProgram} from './pool_gpu';\nimport {ReduceProgram} from './reduce_gpu';\nimport {ReshapePackedProgram} from './reshape_packed_gpu';\nimport {ResizeBilinearBackpropProgram} from './resize_bilinear_backprop_gpu';\nimport {ResizeBilinearProgram} from './resize_bilinear_gpu';\nimport {ResizeBilinearPackedProgram} from './resize_bilinear_packed_gpu';\nimport {ResizeNearestNeigborBackpropProgram} from './resize_nearest_neighbor_backprop_gpu';\nimport {ResizeNearestNeighborProgram} from './resize_nearest_neighbor_gpu';\nimport {ReverseProgram} from './reverse_gpu';\nimport {ReversePackedProgram} from './reverse_packed_gpu';\nimport {ScatterProgram} from './scatter_gpu';\nimport {SegmentOpProgram} from './segment_gpu';\nimport {SelectProgram} from './select_gpu';\nimport {SliceProgram} from './slice_gpu';\nimport {SlicePackedProgram} from './slice_packed_gpu';\nimport {StridedSliceProgram} from './strided_slice_gpu';\nimport * as tex_util from './tex_util';\nimport {TextureData, TextureUsage} from './tex_util';\nimport {TextureManager} from './texture_manager';\nimport {TileProgram} from './tile_gpu';\nimport * as unary_op from './unaryop_gpu';\nimport {UnaryOpProgram} from './unaryop_gpu';\nimport * as unary_packed_op from './unaryop_packed_gpu';\nimport {UnaryOpPackedProgram} from './unaryop_packed_gpu';\nimport {UnpackProgram} from './unpack_gpu';\nimport * as webgl_util from './webgl_util';\nimport {BackendValues} from '@tensorflow/tfjs-core';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\ntype KernelInfo = {\n  name: string; query: Promise<number>;\n};\n\nexport type TimerNode = RecursiveArray<KernelInfo>|KernelInfo;\nexport interface CPUTimerQuery {\n  startMs: number;\n  endMs?: number;\n}\n\nexport interface WebGLMemoryInfo extends MemoryInfo {\n  numBytesInGPU: number;\n  // Tracks the total number of bytes allocated on the GPU, accounting for the\n  // physical texture type.\n  numBytesInGPUAllocated: number;\n  // Tracks byte size of textures that were created and then made available for\n  // reuse (disposed).\n  numBytesInGPUFree: number;\n  unreliable: boolean;\n}\n\nexport interface WebGLTimingInfo extends TimingInfo {\n  uploadWaitMs: number;\n  downloadWaitMs: number;\n}\n\nconst binaryCaches: {[webGLVersion: string]: {[key: string]: GPGPUBinary}} = {};\n\nexport function getBinaryCache(webGLVersion: number) {\n  if (webGLVersion in binaryCaches) {\n    return binaryCaches[webGLVersion];\n  }\n  binaryCaches[webGLVersion] = {};\n  return binaryCaches[webGLVersion];\n}\n\nfunction mapActivationToShaderProgram(\n    activation: backend_util.Activation, packed = false): string {\n  if (activation === 'linear') {\n    if (packed) {\n      return unary_packed_op.LINEAR;\n    }\n    return unary_op.LINEAR;\n  } else if (activation === 'relu') {\n    if (packed) {\n      return unary_packed_op.RELU;\n    }\n    return unary_op.RELU;\n  } else if (activation === 'elu') {\n    if (packed) {\n      return unary_packed_op.ELU;\n    }\n    return unary_op.ELU;\n  } else if (activation === 'relu6') {\n    if (packed) {\n      return unary_packed_op.RELU6;\n    }\n    return unary_op.RELU6;\n  } else if (activation === 'prelu') {\n    if (packed) {\n      return binaryop_packed_gpu.PRELU;\n    }\n    return binaryop_gpu.PRELU;\n  }\n  throw new Error(`Activation ${\n      activation} has not been implemented for the WebGL backend.`);\n}\n\n// Empirically determined constant used to determine size threshold for handing\n// off execution to the CPU.\nconst CPU_HANDOFF_SIZE_THRESHOLD = 128;\n\n// Empirically determined constant used to decide the number of MB on GPU\n// before we warn about high memory use. The MB are this constant * screen area\n// * dpi / 1024 / 1024.\nconst BEFORE_PAGING_CONSTANT = 600;\nfunction numMBBeforeWarning(): number {\n  if (env().global.screen == null) {\n    return 1024;  // 1 GB.\n  }\n  return (env().global.screen.height * env().global.screen.width *\n          window.devicePixelRatio) *\n      BEFORE_PAGING_CONSTANT / 1024 / 1024;\n}\n\n// Empirically determined minimal shared dimension in matmul before we forward\n// to a.mul(b).sum() in order to take advantage of GPU parallelism. See\n// https://github.com/tensorflow/tfjs-core/pull/1379 for benchmarks.\nexport const MATMUL_SHARED_DIM_THRESHOLD = 1000;\n\nexport class MathBackendWebGL extends KernelBackend {\n  texData: DataStorage<TextureData>;\n  gpgpu: GPGPUContext;\n\n  // Maps data ids that have a pending read operation, to list of subscribers.\n  private pendingRead = new WeakMap<DataId, Array<(arr: TypedArray) => void>>();\n  // List of data ids that are scheduled for disposal, but are waiting on a\n  // pending read operation.\n  private pendingDisposal = new WeakSet<DataId>();\n  // Used to count the number of 'shallow' sliced tensors that point to the\n  // same data id.\n  private dataRefCount = new WeakMap<DataId, number>();\n  private numBytesInGPU = 0;\n\n  private canvas: HTMLCanvasElement|OffscreenCanvas;\n\n  private programTimersStack: TimerNode[];\n  private activeTimers: TimerNode[];\n  // Accumulated time spent (including blocking) in uploading data to webgl.\n  private uploadWaitMs = 0;\n  // Accumulated time spent (including blocking in downloading data from webgl.\n  private downloadWaitMs = 0;\n  private cpuBackend: KernelBackend;\n\n  // Number of bits of precision of this backend.\n  private floatPrecisionValue: 32|16;\n\n  private textureManager: TextureManager;\n  private binaryCache: {[key: string]: GPGPUBinary};\n  private gpgpuCreatedLocally: boolean;\n  private numMBBeforeWarning: number;\n  private warnedAboutMemory = false;\n  private warnedAboutCPUBackend = false;\n\n  constructor(gpgpu?: GPGPUContext) {\n    super();\n    if (!env().getBool('HAS_WEBGL')) {\n      throw new Error('WebGL is not supported on this device');\n    }\n\n    if (gpgpu == null) {\n      const gl = getWebGLContext(env().getNumber('WEBGL_VERSION'));\n      this.binaryCache = getBinaryCache(env().getNumber('WEBGL_VERSION'));\n      this.gpgpu = new GPGPUContext(gl);\n      this.canvas = gl.canvas;\n      this.gpgpuCreatedLocally = true;\n    } else {\n      this.gpgpu = gpgpu;\n      this.binaryCache = {};\n      this.gpgpuCreatedLocally = false;\n      this.canvas = gpgpu.gl.canvas;\n    }\n    this.textureManager = new TextureManager(this.gpgpu);\n    this.numMBBeforeWarning = numMBBeforeWarning();\n\n    this.texData = new DataStorage(this, engine());\n  }\n\n  numDataIds() {\n    return this.texData.numDataIds() +\n        (this.cpuBackend ? this.cpuBackend.numDataIds() : 0) -\n        this.pendingDeletes;\n  }\n\n  write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n    if (env().getBool('WEBGL_CHECK_NUMERICAL_PROBLEMS') ||\n        env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64' && values != null) {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    const dataId = {};\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n    return dataId;\n  }\n\n  /** Increase refCount of a `TextureData`. */\n  incRef(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    texData.refCount++;\n  }\n\n  /** Decrease refCount of a `TextureData`. */\n  decRef(dataId: DataId): void {\n    if (this.texData.has(dataId)) {\n      const texData = this.texData.get(dataId);\n      texData.refCount--;\n    }\n  }\n\n  move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n      void {\n    if (env().getBool('DEBUG')) {\n      this.checkNumericalProblems(values);\n    }\n    if (dtype === 'complex64') {\n      throw new Error(\n          `Cannot write to a complex64 dtype. ` +\n          `Please use tf.complex(real, imag).`);\n    }\n    this.texData.set(\n        dataId,\n        {shape, dtype, values, usage: TextureUsage.UPLOAD, refCount: 1});\n  }\n\n  disposeIntermediateTensorInfo(tensorInfo: TensorInfo): void {\n    const dataId = tensorInfo.dataId;\n\n    if (this.texData.has(dataId)) {\n      const textureData = this.texData.get(dataId);\n\n      textureData.refCount--;\n\n      if (textureData.refCount < 1) {\n        this.disposeData(dataId);\n      }\n    }\n  }\n\n  readSync(dataId: DataId): BackendValues {\n    const texData = this.texData.get(dataId);\n    const {values, dtype, complexTensors, slice, shape, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.readSync(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n    if (dtype === 'string') {\n      return values;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let result: Float32Array;\n    if (dtype === 'complex64') {\n      const realValues = complexTensors.real.dataSync() as Float32Array;\n      const imagValues = complexTensors.imag.dataSync() as Float32Array;\n      result = backend_util.mergeRealAndImagArrays(realValues, imagValues);\n    } else {\n      result = this.getValuesFromTexture(dataId);\n    }\n\n    if (shouldTimeProgram) {\n      this.downloadWaitMs += util.now() - start;\n    }\n    return this.convertAndCacheOnCPU(dataId, result);\n  }\n\n  async read(dataId: DataId): Promise<BackendValues> {\n    if (this.pendingRead.has(dataId)) {\n      const subscribers = this.pendingRead.get(dataId);\n      return new Promise<TypedArray>(resolve => subscribers.push(resolve));\n    }\n    const texData = this.texData.get(dataId);\n    const {values, shape, slice, dtype, complexTensors, isPacked} = texData;\n\n    // The presence of `slice` indicates this tensor is a shallow slice of a\n    // different tensor, and is using that original tensor's texture. Run\n    // `clone` in order to copy that texture and read from it.\n    if (slice != null) {\n      let program;\n      if (isPacked) {\n        program = new UnaryOpPackedProgram(shape, unary_op.CLONE);\n      } else {\n        program = new UnaryOpProgram(shape, unary_op.CLONE);\n      }\n      const res =\n          this.runWebGLProgram(program, [{dataId, shape, dtype}], dtype);\n      const data = this.read(res.dataId);\n      this.disposeIntermediateTensorInfo(res);\n      return data;\n    }\n\n    if (values != null) {\n      return this.convertAndCacheOnCPU(dataId);\n    }\n\n    if (!env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED') &&\n        env().getNumber('WEBGL_VERSION') === 2) {\n      throw new Error(\n          `tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and ` +\n          `WEBGL_VERSION=2 not yet supported.`);\n    }\n\n    let buffer = null;\n    let tmpDownloadTarget: TensorInfo;\n\n    if (dtype !== 'complex64' && env().get('WEBGL_BUFFER_SUPPORTED')) {\n      // Possibly copy the texture into a buffer before inserting a fence.\n      tmpDownloadTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpDownloadTarget.dataId);\n\n      buffer = this.gpgpu.createBufferFromTexture(\n          tmpData.texture, ...tex_util.getDenseTexShape(shape));\n    }\n\n    this.pendingRead.set(dataId, []);\n\n    if (dtype !== 'complex64') {\n      // Create a fence and wait for it to resolve.\n      await this.gpgpu.createAndWaitForFence();\n    }\n\n    // Download the values from the GPU.\n    let vals: Float32Array;\n    if (dtype === 'complex64') {\n      const ps = await Promise.all(\n          [complexTensors.real.data(), complexTensors.imag.data()]);\n      const realValues = ps[0];\n      const imagValues = ps[1];\n      vals = backend_util.mergeRealAndImagArrays(\n          realValues as Float32Array, imagValues as Float32Array);\n    } else if (buffer == null) {\n      vals = this.getValuesFromTexture(dataId);\n    } else {\n      const size = util.sizeFromShape(shape);\n      vals = this.gpgpu.downloadFloat32MatrixFromBuffer(buffer, size);\n    }\n    if (tmpDownloadTarget != null) {\n      this.disposeIntermediateTensorInfo(tmpDownloadTarget);\n    }\n    const dTypeVals = this.convertAndCacheOnCPU(dataId, vals);\n\n    const subscribers = this.pendingRead.get(dataId);\n    this.pendingRead.delete(dataId);\n\n    // Notify all pending reads.\n    subscribers.forEach(resolve => resolve(dTypeVals));\n    if (this.pendingDisposal.has(dataId)) {\n      this.pendingDisposal.delete(dataId);\n      this.disposeData(dataId);\n      this.pendingDeletes--;\n    }\n    return dTypeVals;\n  }\n\n  private checkNumericalProblems(values: BackendValues): void {\n    if (values == null) {\n      return;\n    }\n    for (let i = 0; i < values.length; i++) {\n      const num = values[i] as number;\n      if (!webgl_util.canBeRepresented(num)) {\n        if (env().getBool('WEBGL_RENDER_FLOAT32_CAPABLE')) {\n          throw Error(\n              `The value ${num} cannot be represented with your ` +\n              `current settings. Consider enabling float32 rendering: ` +\n              `'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`);\n        }\n        throw Error(`The value ${num} cannot be represented on this device.`);\n      }\n    }\n  }\n\n  private getValuesFromTexture(dataId: DataId): Float32Array {\n    const {shape, dtype, isPacked} = this.texData.get(dataId);\n    const size = util.sizeFromShape(shape);\n    if (env().getBool('WEBGL_DOWNLOAD_FLOAT_ENABLED')) {\n      const tmpTarget = this.decode(dataId);\n      const tmpData = this.texData.get(tmpTarget.dataId);\n      const vals = this.gpgpu\n                       .downloadMatrixFromPackedTexture(\n                           tmpData.texture, ...tex_util.getDenseTexShape(shape))\n                       .subarray(0, size);\n\n      this.disposeIntermediateTensorInfo(tmpTarget);\n\n      return vals;\n    }\n\n    const shouldUsePackedProgram =\n        env().getBool('WEBGL_PACK') && isPacked === true;\n    const outputShape =\n        shouldUsePackedProgram ? webgl_util.getShapeAs3D(shape) : shape;\n    const program = shouldUsePackedProgram ?\n        new EncodeFloatPackedProgram(outputShape as [number, number, number]) :\n        new EncodeFloatProgram(outputShape);\n    const output = this.runWebGLProgram(\n        program, [{shape: outputShape, dtype, dataId}], 'float32');\n    const tmpData = this.texData.get(output.dataId);\n    const vals =\n        this.gpgpu\n            .downloadByteEncodedFloatMatrixFromOutputTexture(\n                tmpData.texture, tmpData.texShape[0], tmpData.texShape[1])\n            .subarray(0, size);\n    this.disposeIntermediateTensorInfo(output);\n\n    return vals;\n  }\n\n  async time(f: () => void): Promise<WebGLTimingInfo> {\n    const oldActiveTimers = this.activeTimers;\n    const newActiveTimers: TimerNode[] = [];\n\n    let outerMostTime = false;\n    if (this.programTimersStack == null) {\n      this.programTimersStack = newActiveTimers;\n      outerMostTime = true;\n    } else {\n      this.activeTimers.push(newActiveTimers);\n    }\n    this.activeTimers = newActiveTimers;\n\n    f();\n\n    // needing to split these up because util.flatten only accepts certain types\n    const flattenedActiveTimerQueries =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.query))\n            .filter(d => d != null);\n    const flattenedActiveTimerNames =\n        util.flatten(this.activeTimers.map((d: KernelInfo) => d.name))\n            .filter(d => d != null);\n\n    this.activeTimers = oldActiveTimers;\n\n    if (outerMostTime) {\n      this.programTimersStack = null;\n    }\n\n    const res: WebGLTimingInfo = {\n      uploadWaitMs: this.uploadWaitMs,\n      downloadWaitMs: this.downloadWaitMs,\n      kernelMs: null,\n      wallMs: null  // will be filled by the engine\n    };\n\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      const kernelMs = await Promise.all(flattenedActiveTimerQueries);\n\n      res['kernelMs'] = util.sum(kernelMs);\n      res['getExtraProfileInfo'] = () =>\n          kernelMs.map((d, i) => ({name: flattenedActiveTimerNames[i], ms: d}))\n              .map(d => `${d.name}: ${d.ms}`)\n              .join(', ');\n    } else {\n      res['kernelMs'] = {\n        error: 'WebGL query timers are not supported in this environment.'\n      };\n    }\n\n    this.uploadWaitMs = 0;\n    this.downloadWaitMs = 0;\n    return res;\n  }\n  memory(): WebGLMemoryInfo {\n    return {\n      unreliable: false,\n      numBytesInGPU: this.numBytesInGPU,\n      numBytesInGPUAllocated: this.textureManager.numBytesAllocated,\n      numBytesInGPUFree: this.textureManager.numBytesFree\n    } as WebGLMemoryInfo;\n  }\n\n  private startTimer(): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.beginQuery();\n    }\n    return {startMs: util.now(), endMs: null};\n  }\n\n  private endTimer(query: WebGLQuery|CPUTimerQuery): WebGLQuery|CPUTimerQuery {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      this.gpgpu.endQuery();\n      return query;\n    }\n    (query as CPUTimerQuery).endMs = util.now();\n    return query;\n  }\n\n  private async getQueryTime(query: WebGLQuery|CPUTimerQuery): Promise<number> {\n    if (env().getNumber('WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE') > 0) {\n      return this.gpgpu.waitForQueryAndGetTime(query as WebGLQuery);\n    }\n    const timerQuery = query as CPUTimerQuery;\n    return timerQuery.endMs - timerQuery.startMs;\n  }\n\n  private pendingDeletes = 0;\n\n  disposeData(dataId: DataId): void {\n    if (this.pendingDisposal.has(dataId)) {\n      return;\n    }\n    if (this.pendingRead.has(dataId)) {\n      this.pendingDisposal.add(dataId);\n      this.pendingDeletes++;\n      return;\n    }\n    // No-op if already disposed.\n    if (!this.texData.has(dataId)) {\n      return;\n    }\n\n    this.releaseGPUData(dataId);\n    const {complexTensors} = this.texData.get(dataId);\n    if (complexTensors != null) {\n      complexTensors.real.dispose();\n      complexTensors.imag.dispose();\n    }\n    this.texData.delete(dataId);\n  }\n\n  private releaseGPUData(dataId: DataId): void {\n    const {texture, dtype, texShape, usage, isPacked, slice} =\n        this.texData.get(dataId);\n    const key = slice && slice.origDataId || dataId;\n    const refCount = this.dataRefCount.get(key);\n    if (refCount > 1) {\n      this.dataRefCount.set(key, refCount - 1);\n    } else {\n      this.dataRefCount.delete(key);\n      if (texture != null) {\n        this.numBytesInGPU -= this.computeBytes(texShape, dtype);\n        this.textureManager.releaseTexture(texture, texShape, usage, isPacked);\n      }\n    }\n    const texData = this.texData.get(dataId);\n    texData.texture = null;\n    texData.texShape = null;\n    texData.isPacked = false;\n    texData.slice = null;\n  }\n\n  getTexture(dataId: DataId): WebGLTexture {\n    this.uploadToGPU(dataId);\n    return this.texData.get(dataId).texture;\n  }\n\n  /**\n   * Returns internal information for the specific data bucket. Used in unit\n   * tests.\n   */\n  getDataInfo(dataId: DataId): TextureData {\n    return this.texData.get(dataId);\n  }\n\n  private getCPUBackend(): KernelBackend|null {\n    if (!env().getBool('WEBGL_CPU_FORWARD')) {\n      return null;\n    }\n\n    if (this.cpuBackend == null) {\n      this.cpuBackend = engine().findBackend('cpu');\n    }\n\n    return this.cpuBackend;\n  }\n\n  /*\n  Tests whether all the inputs to an op are small and on the CPU. This heuristic\n  determines when it would be faster to execute a kernel on the CPU. WebGL\n  kernels opt into running this check and forwarding when appropriate.\n  TODO(https://github.com/tensorflow/tfjs/issues/872): Develop a more\n  sustainable strategy for optimizing backend execution of ops.\n   */\n  shouldExecuteOnCPU(\n      inputs: TensorInfo[],\n      sizeThreshold = CPU_HANDOFF_SIZE_THRESHOLD): boolean {\n    const cpuBackend = this.getCPUBackend();\n    if (!this.warnedAboutCPUBackend && cpuBackend == null) {\n      console.warn(\n          'Your application contains ops that are small enough to be ' +\n          'executed on the CPU backend, however the CPU backend cannot ' +\n          'be found. Consider importing the CPU backend ' +\n          '(@tensorflow/tfjs-backend-cpu) for better performance.');\n\n      this.warnedAboutCPUBackend = true;\n    }\n\n    return cpuBackend != null &&\n        inputs.every(\n            input => this.texData.get(input.dataId).texture == null &&\n                util.sizeFromShape(input.shape) < sizeThreshold);\n  }\n\n  getGPGPUContext(): GPGPUContext {\n    return this.gpgpu;\n  }\n\n  complex<T extends Tensor>(real: T, imag: T): T {\n    const result = this.makeOutput(real.shape, 'complex64');\n    const resultData = this.texData.get(result.dataId);\n    // The backend owns the reference to the underlying real and imaginary\n    // clones. These will explicitly get disposed when the complex tensor is\n    // disposed.\n    resultData.complexTensors = {\n      real: engine().keep(real.clone()),\n      imag: engine().keep(imag.clone())\n    };\n\n    return result as T;\n  }\n  real<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.real.clone() as T;\n  }\n  imag<T extends Tensor>(input: T): T {\n    const resultData = this.texData.get(input.dataId);\n    return resultData.complexTensors.imag.clone() as T;\n  }\n\n  slice<T extends Tensor>(x: T, begin: number[], size: number[]): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = sliceImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, begin, size, x.shape,\n          x.dtype);\n      return this.makeOutput(size, x.dtype, outValues);\n    }\n    // Short-circuit computation if the slice is zero-sized.\n    if (util.sizeFromShape(size) === 0) {\n      return tensor([], size, x.dtype) as T;\n    }\n    const {isPacked} = this.texData.get(x.dataId);\n    const isContinous = slice_util.isSliceContinous(x.shape, begin, size);\n    if (isPacked || !isContinous) {\n      const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n          new SlicePackedProgram(size) :\n          new SliceProgram(size);\n      const customSetup = program.getCustomSetupFunc(begin);\n      return this.compileAndRun(program, [x], null, customSetup);\n    }\n    this.uploadToGPU(x.dataId);\n    return this.shallowSlice(x, begin, size) as T;\n  }\n\n  private shallowSlice(x: Tensor, begin: number[], size: number[]): Tensor {\n    const xTexData = this.texData.get(x.dataId);\n    const t = this.makeOutput(size, x.dtype);\n    const newTexData = this.texData.get(t.dataId);\n    // Copy texture data from the original tensor.\n    Object.assign(newTexData, xTexData);\n    newTexData.shape = size;\n    newTexData.dtype = x.dtype;\n    let flatOffset = slice_util.computeFlatOffset(begin, x.strides);\n    if (xTexData.slice) {\n      // We are slicing an already sliced tensor, so we have to accumulate\n      // the offset.\n      flatOffset += xTexData.slice.flatOffset;\n    }\n    newTexData.slice = {\n      flatOffset,\n      // Point to the original dataId, which is used to do ref counting.\n      origDataId: xTexData.slice && xTexData.slice.origDataId || x.dataId\n    };\n\n    // Increase the ref count for that data bucket.\n    const refCount = this.dataRefCount.get(newTexData.slice.origDataId) || 1;\n    this.dataRefCount.set(newTexData.slice.origDataId, refCount + 1);\n\n    return t;\n  }\n\n  stridedSlice<T extends Tensor>(\n      x: T, begin: number[], end: number[], strides: number[]): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x], () => this.cpuBackend.stridedSlice(x, begin, end, strides));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const outShape = slice_util.computeOutShape(begin, end, strides);\n\n    if (outShape.some(axis => axis === 0)) {\n      return tensor([], outShape) as T;\n    }\n\n    const program = new StridedSliceProgram(begin, strides, outShape);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reverse<T extends Tensor>(x: T, axis: number[]): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new ReversePackedProgram(x.shape, axis) :\n        new ReverseProgram(x.shape, axis);\n    return this.compileAndRun(program, [x]);\n  }\n\n  concat(tensors: Tensor[], axis: number): Tensor {\n    if (tensors[0].dtype === 'complex64') {\n      const reals = tensors.map((t) => real(t));\n      const imags = tensors.map((t) => imag(t));\n      return complex(this.concat(reals, axis), this.concat(imags, axis));\n    }\n\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n    if (tensors.length > env().getNumber('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.concat(tensors.slice(0, midIndex), axis);\n      const rightSide = this.concat(tensors.slice(midIndex), axis);\n      return this.concat([leftSide, rightSide], axis);\n    }\n    if (env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') && tensors[0].rank > 1) {\n      const program = new ConcatPackedProgram(tensors.map(t => t.shape), axis);\n      return this.compileAndRun(program, tensors);\n    }\n    // Any concat of n-dimensional tensors across any axis can be reduced to\n    // a concatenation of two-dimensional tensors across the axis 1 by first\n    // partitioning the axes of the original tensors into those less than the\n    // axis to be concatenated and the rest. Then reshape the tensors\n    // into a two-dimensional tensor by collapsing these two sets of axes and\n    // concatenate the resulting matrices across the axis 1, finally reshaping\n    // the result to have the proper shape.\n    const outShape =\n        backend_util.computeOutShape(tensors.map(t => t.shape), axis);\n    const tensors2D =\n        tensors.map(t => t.as2D(-1, util.sizeFromShape(t.shape.slice(axis))));\n    const program = new ConcatProgram(tensors2D.map(t => t.shape));\n    const res: Tensor = this.compileAndRun(program, tensors2D);\n    return res.reshape(outShape);\n  }\n\n  neg<T extends Tensor>(x: T): T {\n    const cpuRes = this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.neg(x));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.NEG, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.NEG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  batchMatMul(\n      a: Tensor3D, b: Tensor3D, transposeA: boolean,\n      transposeB: boolean): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const sharedDim = transposeA ? a.shape[1] : a.shape[2];\n    const [batch, , ] = a.shape;\n\n    // Since the matrices are vectors, it is faster to call mul().sum()\n    // because sum() is O(sqrt(N)) due to divide-and-conquer.\n    if ((outerShapeA === 1 || outerShapeB === 1) &&\n        sharedDim > MATMUL_SHARED_DIM_THRESHOLD) {\n      if (transposeA) {\n        a = transpose(a, [0, 2, 1]);\n      }\n      if (transposeB) {\n        b = transpose(b, [0, 2, 1]);\n      }\n\n      const a3D = outerShapeB === 1 ? a : a.as3D(batch, sharedDim, 1);\n      const axis = outerShapeB === 1 ? 2 : 1;\n      const b3D = outerShapeB === 1 ? b.as3D(batch, 1, sharedDim) : b;\n      return this.multiply(a3D, b3D).sum(axis, true /* keepDims */);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB);\n    return this.compileAndRun<Tensor3D>(program, [a, b], dtype);\n  }\n\n  fusedBatchMatMul(\n      {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n          backend_util.FusedBatchMatMulConfig): Tensor3D {\n    const outerShapeA = transposeA ? a.shape[2] : a.shape[1];\n    const outerShapeB = transposeB ? b.shape[1] : b.shape[2];\n    const [batch, , ] = a.shape;\n\n    const dtype = upcastType(a.dtype, b.dtype);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const program = new MatMulPackedProgram(\n        a.shape, [batch, outerShapeA, outerShapeB], transposeA, transposeB,\n        hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [a, b];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun<Tensor3D>(program, inputs, dtype);\n  }\n\n  multiply(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64') {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n\n      const realProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.REAL, a.shape, b.shape);\n      const imagProgram = new BinaryOpComplexProgram(\n          binaryop_complex_gpu.COMPLEX_MULTIPLY.IMAG, a.shape, b.shape);\n\n      const inputs = [\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(a, aData.complexTensors.imag),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.real),\n        this.makeComplexComponentTensorInfo(b, bData.complexTensors.imag)\n      ];\n      const real = this.compileAndRun<Tensor>(realProgram, inputs);\n      const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n\n      const complex = this.complex(real, imag);\n      real.dispose();\n      imag.dispose();\n      return complex;\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = multiplyImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.MUL, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.MUL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], a.dtype);\n  }\n\n  localResponseNormalization4D(\n      x: Tensor4D, radius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n        new LRNPackedProgram(x.shape, radius, bias, alpha, beta) :\n        new LRNProgram(x.shape, radius, bias, alpha, beta);\n    return this.compileAndRun(program, [x]);\n  }\n\n  LRNGrad(\n      dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D,\n      depthRadius: number, bias: number, alpha: number,\n      beta: number): Tensor4D {\n    const program =\n        new LRNGradProgram(inputImage.shape, depthRadius, bias, alpha, beta);\n    return this.compileAndRun(program, [inputImage, outputImage, dy]);\n  }\n\n  tile<T extends Tensor>(x: T, reps: number[]): T {\n    if (x.dtype === 'string') {\n      const data = this.readSync(x.dataId) as Uint8Array[];\n      const decodedData = data.map(d => util.decodeString(d));\n      const buf = buffer(x.shape, x.dtype, decodedData);\n      return tile(buf, reps) as T;\n    }\n    const program = new TileProgram(x.shape, reps);\n    return this.compileAndRun(program, [x]);\n  }\n\n  pad<T extends Tensor>(\n      x: T, paddings: Array<[number, number]>, constantValue: number): T {\n    const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n        new PadPackedProgram(x.shape, paddings, constantValue) :\n        new PadProgram(x.shape, paddings, constantValue);\n    return this.compileAndRun(program, [x]);\n  }\n\n  gather<T extends Tensor>(x: T, indices: Tensor1D, axis: number): T {\n    const cpuRes = this.tryRunOnCpuOrThrow(\n        [x, indices], () => this.cpuBackend.gather(x, indices, axis));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = new GatherProgram(x.shape, indices.size, axis);\n    return this.compileAndRun(program, [x, indices]);\n  }\n\n  batchToSpaceND<T extends Tensor>(\n      x: T, blockShape: number[], crops: number[][]): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'batchToSpaceND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const reshaped = backend_util.getReshaped(x.shape, blockShape, prod);\n    const permuted =\n        backend_util.getPermuted(reshaped.length, blockShape.length);\n    const reshapedPermuted =\n        backend_util.getReshapedPermuted(x.shape, blockShape, prod);\n    const sliceBeginCoords =\n        backend_util.getSliceBeginCoords(crops, blockShape.length);\n    const sliceSize =\n        backend_util.getSliceSize(reshapedPermuted, crops, blockShape.length);\n\n    return transpose(x.reshape(reshaped), permuted)\n               .reshape(reshapedPermuted)\n               .slice(sliceBeginCoords, sliceSize) as T;\n  }\n\n  spaceToBatchND<T extends Tensor>(\n      x: T, blockShape: number[], paddings: Array<[number, number]>): T {\n    util.assert(\n        x.rank <= 4,\n        () => 'spaceToBatchND for rank > 4 with a WebGL backend not ' +\n            'implemented yet');\n\n    const prod = blockShape.reduce((a, b) => a * b);\n\n    const completePaddings: Array<[number, number]> = [[0, 0]];\n    completePaddings.push(...paddings);\n    for (let i = 1 + blockShape.length; i < x.shape.length; ++i) {\n      completePaddings.push([0, 0]);\n    }\n\n    const paddedX = x.pad(completePaddings);\n\n    const reshapedPaddedShape =\n        backend_util.getReshaped(paddedX.shape, blockShape, prod, false);\n\n    const permutedReshapedPaddedPermutation = backend_util.getPermuted(\n        reshapedPaddedShape.length, blockShape.length, false);\n\n    const flattenShape = backend_util.getReshapedPermuted(\n        paddedX.shape, blockShape, prod, false);\n\n    const paddedXT = transpose(\n        paddedX.reshape(reshapedPaddedShape),\n        permutedReshapedPaddedPermutation);\n    return reshape(paddedXT, flattenShape) as T;\n  }\n\n  private reduce(\n      x: Tensor2D, reduceType: 'all'|'any'|'max'|'min'|'sum'|'prod',\n      dtype: DataType): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const outSize = Math.ceil(inSize / windowSize);\n    const reduceInfo = {windowSize, inSize, batchSize, outSize};\n    const program = new ReduceProgram(reduceInfo, reduceType);\n    const output = this.compileAndRun<Tensor2D>(program, [x], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.reduce(output, reduceType, dtype);\n  }\n\n  private argReduce(\n      x: Tensor2D, reduceType: 'max'|'min',\n      bestIndicesA: Tensor2D = null): Tensor2D {\n    let batchSize = x.shape[0];\n    let inSize = x.shape[1];\n    if (bestIndicesA != null) {\n      batchSize = bestIndicesA.shape[0];\n      inSize = bestIndicesA.shape[1];\n    }\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const reduceInfo = {\n      windowSize,\n      inSize,\n      batchSize,\n      outSize: Math.ceil(inSize / windowSize)\n    };\n    const program =\n        new ArgMinMaxProgram(reduceInfo, reduceType, bestIndicesA == null);\n    const inputs = [x];\n    if (bestIndicesA != null) {\n      inputs.push(bestIndicesA);\n    }\n    const output = this.compileAndRun<Tensor2D>(program, inputs, 'int32');\n    // No need to run another GPGPU program.\n    if (output.shape[1] === 1) {\n      return output;\n    }\n    return this.argReduce(x, reduceType, output);\n  }\n\n  private argReducePacked(\n      x: Tensor, reduceType: 'max'|'min', bestIndicesA: Tensor = null): Tensor {\n    const inShape = bestIndicesA != null ? bestIndicesA.shape : x.shape;\n    const inSize = inShape[inShape.length - 1];\n    const windowSize = backend_util.computeOptimalWindowSize(inSize);\n    const program = new ArgMinMaxPackedProgram(\n        inShape, windowSize, reduceType, bestIndicesA == null);\n    const inputs = bestIndicesA == null ? [x] : [x, bestIndicesA];\n    const output = this.compileAndRun<Tensor>(program, inputs, 'int32');\n    if (output.rank === x.rank) {\n      return this.argReducePacked(x, reduceType, output);\n    }\n    return output;\n  }\n\n  sum(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('sum', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'sum', outputDType).reshape(outShape);\n  }\n\n  prod(x: Tensor, axes: number[]): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([x], () => this.cpuBackend.prod(x, axes));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    return this.reduce(a2D, 'prod', outputDType).reshape(outShape);\n  }\n\n  unsortedSegmentSum<T extends Tensor>(\n      x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n    let axis = 0;\n    const permutation = backend_util.getAxesPermutation([axis], x.rank);\n    let permutedX = x;\n    if (permutation != null) {\n      permutedX = transpose(x, permutation);\n      axis = backend_util.getInnerMostAxes(1, x.rank)[0];\n    }\n\n    const outShape =\n        segment_util.computeOutShape(permutedX.shape, axis, numSegments);\n    const inSize = util.sizeFromShape([permutedX.shape[axis]]);\n    const a2D = permutedX.as2D(-1, inSize);\n    const outputDType = tf.sumOutType(x.dtype);\n    let result =\n        this.segOpCompute(\n                a2D, 'unsortedSegmentSum', segmentIds, outputDType, numSegments)\n            .reshape(outShape);\n    if (permutation != null) {\n      result =\n          transpose(result, backend_util.getUndoAxesPermutation(permutation));\n    }\n    return result;\n  }\n\n  private segOpCompute(\n      x: Tensor2D, segOpType: 'unsortedSegmentSum', segmentIds: Tensor1D,\n      dtype: DataType, numSegments: number): Tensor2D {\n    const batchSize = x.shape[0];\n    const inSize = x.shape[1];\n    const windowSize =\n        segment_util.segOpComputeOptimalWindowSize(inSize, numSegments);\n    const segOpInfo = {windowSize, inSize, batchSize, numSegments};\n    const program = new SegmentOpProgram(segOpInfo, segOpType);\n    const output =\n        this.compileAndRun<Tensor2D>(program, [x, segmentIds], dtype);\n    // No need to run another GPGPU program.\n    if (output.shape[1] === numSegments) {\n      return output;\n    }\n    segmentIds = range(0, numSegments).tile([inSize / windowSize]);\n    return this.segOpCompute(output, segOpType, segmentIds, dtype, numSegments);\n  }\n\n  private argMinMaxReduce(x: Tensor, axis: number, reduceType: 'min'|'max'):\n      Tensor {\n    const axes = [axis];\n    backend_util.assertAxesAreInnerMostDims(\n        'arg' + reduceType.charAt(0).toUpperCase() + reduceType.slice(1), axes,\n        x.rank);\n    if (!env().getBool('WEBGL_PACK_REDUCE') || x.rank <= 2) {\n      const [outShape, reduceShape] =\n          backend_util.computeOutAndReduceShapes(x.shape, axes);\n      const inSize = util.sizeFromShape(reduceShape);\n      const a2D = x.as2D(-1, inSize);\n      return this.argReduce(a2D, reduceType).reshape(outShape);\n    }\n    return this.argReducePacked(x, reduceType);\n  }\n\n  argMin(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'min');\n  }\n\n  argMax(x: Tensor, axis: number): Tensor {\n    return this.argMinMaxReduce(x, axis, 'max');\n  }\n\n  cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n      Tensor {\n    if (axis !== x.rank - 1) {\n      throw new Error(\n          `WebGL cumsum shader expects an inner-most axis=${x.rank - 1} ` +\n          `but got axis=${axis}`);\n    }\n    const size = x.shape[axis];\n    let result = x;\n    // Use cumsum parallel algorithm, ref:\n    // https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-39-parallel-prefix-sum-scan-cuda\n    for (let i = 0; i <= Math.ceil(Math.log2(size)) - 1; i++) {\n      const program = new CumSumProgram(x.shape, false, reverse);\n      const customSetup = program.getCustomSetupFunc(i);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result], result.dtype, customSetup);\n      prevResult.dispose();\n    }\n    // For exclusive cumsum, shift the end result in the direction of sum and\n    // add 0 to the front index.\n    if (exclusive) {\n      const program = new CumSumProgram(x.shape, exclusive, reverse);\n      const prevResult = result;\n      result = this.compileAndRun(program, [result]);\n      prevResult.dispose();\n    }\n\n    return result;\n  }\n\n  equal(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.EQUAL, 'bool');\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  notEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.NOT_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.NOT_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  less(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.less(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.LESS, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  lessEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LESS_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LESS_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greater(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.greater(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.GREATER, 'bool');\n    }\n\n    const program = new BinaryOpProgram(binaryop_gpu.GREATER, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  greaterEqual(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.GREATER_EQUAL, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.GREATER_EQUAL, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalNot<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOGICAL_NOT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  logicalAnd(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_AND, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_AND, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  logicalOr(a: Tensor, b: Tensor): Tensor {\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_packed_gpu.LOGICAL_OR, 'bool');\n    }\n    const program =\n        new BinaryOpProgram(binaryop_gpu.LOGICAL_OR, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b], 'bool');\n  }\n\n  select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n    const program = new SelectProgram(condition.rank, a.shape, a.rank);\n    return this.compileAndRun(\n        program, [condition, a, b], upcastType(a.dtype, b.dtype));\n  }\n\n  where(condition: Tensor): Tensor2D {\n    backend_util.warn(\n        'tf.where() in webgl locks the UI thread. ' +\n        'Call tf.whereAsync() instead');\n    const condVals = condition.dataSync();\n    return whereImpl(condition.shape, condVals);\n  }\n\n  topk<T extends Tensor>(x: T, k: number, sorted: boolean): [T, T] {\n    const xVals = x.dataSync();\n    return topkImpl(xVals, x.shape, x.dtype as NumericDataType, k, sorted);\n  }\n\n  min(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('min', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'min', a2D.dtype).reshape(outShape);\n  }\n\n  minimum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.minimum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MIN, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MIN, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  mod(a: Tensor, b: Tensor): Tensor {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MOD, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MOD, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  maximum(a: Tensor, b: Tensor): Tensor {\n    const cpuRes =\n        this.tryRunOnCpuOrThrow([a, b], () => this.cpuBackend.maximum(a, b));\n    if (cpuRes) {\n      return cpuRes;\n    }\n\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.MAX, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.MAX, a.shape, b.shape);\n    return this.compileAndRun(program, [a, b]);\n  }\n\n  all(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('all', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'all', a2D.dtype).reshape(outShape);\n  }\n\n  any(x: Tensor, axes: number[]): Tensor {\n    backend_util.assertAxesAreInnerMostDims('any', axes, x.rank);\n    const [outShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(x.shape, axes);\n    const inSize = util.sizeFromShape(reduceShape);\n    const a2D = x.as2D(-1, inSize);\n    return this.reduce(a2D, 'any', a2D.dtype).reshape(outShape);\n  }\n\n  floorDiv(a: Tensor, b: Tensor): Tensor {\n    const op = binaryop_gpu.INT_DIV;\n    const outputDtype = 'int32';\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(\n          a, b, binaryop_packed_gpu.INT_DIV, outputDtype);\n    }\n    const program = new BinaryOpProgram(op, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], outputDtype);\n  }\n\n  add(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.ADD);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = addImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.ADD, dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.ADD, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  private packedUnaryOp(x: TensorInfo, op: string, dtype: DataType) {\n    const program = new UnaryOpPackedProgram(x.shape, op);\n    return this.compileAndRun<Tensor>(program, [x], dtype);\n  }\n\n  private packedBinaryOp(\n      a: TensorInfo, b: TensorInfo, op: string, dtype: DataType,\n      checkOutOfBounds = false) {\n    const program =\n        new BinaryOpPackedProgram(op, a.shape, b.shape, checkOutOfBounds);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  /**\n   * Computes a complex binary operation that can be decomposed into a simple\n   * binary operation on both the real and imagary parts.\n   */\n  private complexSeparableBinaryOp(a: Tensor, b: Tensor, op: string): Tensor {\n    const aData = this.texData.get(a.dataId);\n    const bData = this.texData.get(b.dataId);\n\n    const [real, imag] = [\n      [aData.complexTensors.real, bData.complexTensors.real],\n      [aData.complexTensors.imag, bData.complexTensors.imag]\n    ].map(complexParts => {\n      const [aPart, bPart] = complexParts;\n\n      const aHandle = this.makeComplexComponentTensorInfo(a, aPart);\n      const bHandle = this.makeComplexComponentTensorInfo(b, bPart);\n\n      const program = new BinaryOpProgram(op, a.shape, b.shape);\n      return this.compileAndRun<Tensor>(\n          program, [aHandle, bHandle], upcastType(aPart.dtype, bPart.dtype));\n    });\n\n    const complex = this.complex(real, imag);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  // Returns a TensorInfo with the complex shape and the dataId of the\n  // underlying part. We need to do this because a reshaped complex tensor is\n  // not reflected in its parts.\n  private makeComplexComponentTensorInfo(\n      complexTensor: Tensor, complexPart: Tensor): TensorInfo {\n    return {\n      dataId: complexPart.dataId,\n      dtype: complexPart.dtype,\n      shape: complexTensor.shape\n    };\n  }\n\n  addN<T extends Tensor>(tensors: T[]): T {\n    if (tensors.length === 1) {\n      return tensors[0];\n    }\n\n    // Limit the number of uploaded textures for optimization.\n    if (tensors.length > env().get('WEBGL_MAX_TEXTURES_IN_SHADER')) {\n      const midIndex = Math.floor(tensors.length / 2);\n      const leftSide = this.addN(tensors.slice(0, midIndex));\n      const rightSide = this.addN(tensors.slice(midIndex));\n      return this.addN([leftSide, rightSide]);\n    }\n\n    const dtype =\n        tensors.map(t => t.dtype).reduce((d1, d2) => upcastType(d1, d2));\n    const shapes = tensors.map(t => t.shape);\n    // We can make sure shapes are identical in op level.\n    const usePackedOp = env().getBool('WEBGL_PACK');\n    const program = usePackedOp ?\n        new AddNPackedProgram(tensors[0].shape, shapes) :\n        new AddNProgram(tensors[0].shape, shapes);\n    return this.compileAndRun<T>(program, tensors, dtype);\n  }\n\n  subtract(a: Tensor, b: Tensor): Tensor {\n    if (a.dtype === 'complex64' && b.dtype === 'complex64') {\n      return this.complexSeparableBinaryOp(a, b, binaryop_gpu.SUB);\n    }\n\n    const dtype = upcastType(a.dtype, b.dtype);\n    if (this.shouldExecuteOnCPU([a, b])) {\n      const aData = this.texData.get(a.dataId);\n      const bData = this.texData.get(b.dataId);\n      const [outValues, outShape] = subImplCPU(\n          a.shape, b.shape, aData.values as TypedArray,\n          bData.values as TypedArray, dtype);\n      return this.makeOutput(outShape, dtype, outValues);\n    }\n    if (env().getBool('WEBGL_PACK_BINARY_OPERATIONS')) {\n      return this.packedBinaryOp(a, b, binaryop_gpu.SUB, a.dtype);\n    }\n    const program = new BinaryOpProgram(binaryop_gpu.SUB, a.shape, b.shape);\n    return this.compileAndRun<Tensor>(program, [a, b], dtype);\n  }\n\n  pow<T extends Tensor>(a: T, b: Tensor): T {\n    const usePackedOp = env().getBool('WEBGL_PACK_BINARY_OPERATIONS');\n    const program = usePackedOp ?\n        new BinaryOpPackedProgram(binaryop_packed_gpu.POW, a.shape, b.shape) :\n        new BinaryOpProgram(binaryop_gpu.POW, a.shape, b.shape);\n    const dtype = upcastType(a.dtype, b.dtype);\n    return this.compileAndRun<T>(program, [a, b], dtype);\n  }\n\n  ceil<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          ceilImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.CEIL, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.CEIL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  floor<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = floorImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.FLOOR, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.FLOOR);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sign<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  isNaN<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_NAN);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isInf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_INF);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n  isFinite<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.IS_FINITE);\n    return this.compileAndRun(program, [x], 'bool');\n  }\n\n  round<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ROUND);\n    return this.compileAndRun(program, [x]);\n  }\n\n  exp<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          expImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXP, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXP);\n    return this.compileAndRun(program, [x]);\n  }\n\n  expm1<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = expm1ImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.EXPM1, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.EXPM1);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softmax<T extends Tensor>(logits: T, dim: number): T {\n    const axes = util.parseAxisParam([dim], logits.shape);\n    // TODO(annxingyuan): Call maxImpl rather than op as part of softmax kernel\n    // modularization.\n    const maxLogit = max(logits, axes);\n    const expandedShape =\n        backend_util.expandShapeToKeepDim(maxLogit.shape, axes);\n    const a = this.subtract(logits, maxLogit.reshape(expandedShape));\n    const b = this.exp(a);\n    const sumExp = this.sum(b, axes).reshape(expandedShape);\n\n    // TODO(annxingyuan): Call divImpl rather than op as part of softmax kernel\n    // modularization.\n    return div(b, sumExp);\n  }\n\n  log<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues =\n          logImplCPU(this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.LOG, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG);\n    return this.compileAndRun(program, [x]);\n  }\n\n  log1p<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.LOG1P);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sqrt<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  rsqrt<T extends Tensor>(x: T): T {\n    if (this.shouldExecuteOnCPU([x])) {\n      const outValues = rsqrtImplCPU(\n          this.texData.get(x.dataId).values as TypedArray, x.dtype);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.RSQRT);\n    return this.compileAndRun(program, [x]);\n  }\n\n  reciprocal<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.RECIPROCAL);\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  relu6<T extends Tensor>(x: T): T {\n    let program: UnaryOpProgram|UnaryOpPackedProgram;\n    if (env().getBool('WEBGL_PACK')) {\n      program = new UnaryOpPackedProgram(x.shape, unary_packed_op.RELU6);\n    } else {\n      program = new UnaryOpProgram(x.shape, unary_op.RELU6);\n    }\n    return this.compileAndRun(program, [x]);\n  }\n\n  prelu<T extends Tensor>(x: T, alpha: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.PRELU, x.shape, alpha.shape) :\n        new BinaryOpProgram(binaryop_gpu.PRELU, x.shape, alpha.shape);\n    return this.compileAndRun(program, [x, alpha]);\n  }\n\n  elu<T extends Tensor>(x: T): T {\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_packed_op.ELU, x.dtype) as T;\n    }\n    const program = new UnaryOpProgram(x.shape, unary_op.ELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  eluDer<T extends Tensor>(dy: T, y: T): T {\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            binaryop_packed_gpu.ELU_DER, dy.shape, y.shape) :\n        new BinaryOpProgram(binaryop_gpu.ELU_DER, dy.shape, y.shape);\n    return this.compileAndRun(program, [dy, y]);\n  }\n\n  selu<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SELU);\n    return this.compileAndRun(program, [x]);\n  }\n\n  int<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TO_INT);\n    return this.compileAndRun(program, [x], 'int32');\n  }\n\n  clip<T extends Tensor>(x: T, min: number, max: number): T {\n    let program;\n    if (env().getBool('WEBGL_PACK_CLIP')) {\n      program = new ClipPackedProgram(x.shape);\n    } else {\n      program = new ClipProgram(x.shape);\n    }\n    const customSetup = program.getCustomSetupFunc(min, max);\n    return this.compileAndRun(program, [x], null, customSetup);\n  }\n\n  abs<T extends Tensor>(x: T): T {\n    // TODO: handle cases when x is complex.\n    if (this.shouldExecuteOnCPU([x]) && x.dtype !== 'complex64') {\n      const outValues =\n          simpleAbsImplCPU(this.texData.get(x.dataId).values as TypedArray);\n      return this.makeOutput(x.shape, x.dtype, outValues);\n    }\n\n    if (env().getBool('WEBGL_PACK_UNARY_OPERATIONS')) {\n      return this.packedUnaryOp(x, unary_op.ABS, x.dtype) as T;\n    }\n\n    const program = new UnaryOpProgram(x.shape, unary_op.ABS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  complexAbs<T extends Tensor>(x: T): T {\n    const xData = this.texData.get(x.dataId);\n\n    const program = new ComplexAbsProgram(x.shape);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    return this.compileAndRun<Tensor>(program, inputs) as T;\n  }\n\n  sigmoid<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SIGMOID);\n    return this.compileAndRun(program, [x]);\n  }\n\n  softplus<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SOFTPLUS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asin<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASIN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acos<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOS);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atan<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATAN);\n    return this.compileAndRun(program, [x]);\n  }\n\n  sinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.SINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.COSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  tanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.TANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  asinh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ASINH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  acosh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ACOSH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  atanh<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ATANH);\n    return this.compileAndRun(program, [x]);\n  }\n\n  erf<T extends Tensor>(x: T): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.ERF);\n    return this.compileAndRun(program, [x]);\n  }\n\n  step<T extends Tensor>(x: T, alpha: number): T {\n    const program = new UnaryOpProgram(x.shape, unary_op.STEP(alpha));\n    return this.compileAndRun(program, [x]);\n  }\n\n  private conv2dByMatMul(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Reshapes conv2D input to 2D tensors, uses matMul and then reshape the\n    // result from 2D to 4D.\n    const xShape = x.shape;\n    const xTexData = this.texData.get(x.dataId);\n    const sharedMatMulDim = convInfo.inChannels;\n    const outerShapeX = xShape[0] * xShape[1] * xShape[2];\n    const outerShapeFilter = convInfo.outChannels;\n    const isChannelsLast = convInfo.dataFormat === 'channelsLast';\n    const transposeA = false;\n    const transposeB = false;\n\n    // TODO: Once reduction ops are packed, batchMatMul will always be packed\n    // and we can remove this condition.\n    const batchMatMulWillBeUnpacked =\n        (outerShapeX === 1 || outerShapeFilter === 1) &&\n        sharedMatMulDim > MATMUL_SHARED_DIM_THRESHOLD;\n    const reshapeWillBeExpensive = xShape[2] % 2 !== 0 && !!xTexData.isPacked;\n\n    if (batchMatMulWillBeUnpacked || !env().getBool('WEBGL_LAZILY_UNPACK') ||\n        !env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ||\n        !reshapeWillBeExpensive) {\n      const targetShape = isChannelsLast ? xShape[0] * xShape[1] * xShape[2] :\n                                           xShape[0] * xShape[2] * xShape[3];\n      const xReshaped = reshape(x, [1, targetShape, convInfo.inChannels]);\n      const filterReshaped =\n          reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n      const result = this.fusedBatchMatMul({\n        a: xReshaped as Tensor3D,\n        b: filterReshaped as Tensor3D,\n        transposeA,\n        transposeB,\n        bias,\n        activation,\n        preluActivationWeights\n      });\n      return reshape(result, convInfo.outShape);\n    }\n\n    // Following optimization is specific to packed |x| with odd row count\n    // (For example, in channelLast mode, 'row count' refers to x.shape[2]):\n    // we avoid expensive packed 2x2 reshape by padding row count to next,\n    // even number. When x.shape[2] is odd, the result of packed batchMatMul is\n    // the same (has the same texture layout and and values in the texture) as\n    // it is for even x.shape[2] + 1. We make the odd-rows tensor to look like\n    // even-rows tensor before the operation and, after the batchMatMul,\n    // fix the even-rows result to have odd number of rows.\n    const targetShape = isChannelsLast ?\n        xShape[0] * xShape[1] * (xShape[2] + 1) :\n        xShape[0] * xShape[2] * (xShape[3] + 1);\n    const xReshaped: TensorInfo = {\n      dataId: x.dataId,\n      shape: [1, targetShape, convInfo.inChannels],\n      dtype: x.dtype\n    };\n    // xTexData.shape gets referenced from GPGPUBinary.inShapeInfos.\n    // Decrementing row count, after batchMatMul->...->compileProgram leads to\n    // invalid row count within the reference in GPGPUBinary.inShapeInfos.\n    // Alternative fix would be to provide a copy to GPGPUBinary.inShapeInfos\n    // in compileProgram method, but that would affect compilation of all\n    // programs - instead, provide a copy here, with even row count, before\n    // calling batchMatMul->...->compileProgram and after that, the original\n    // xTexData.shape is restored.\n    const originalXTexDataShape = xTexData.shape;\n    xTexData.shape = xTexData.shape.slice();\n    xTexData.shape[xTexData.shape.length - 2]++;\n    util.assert(\n        webgl_util.isReshapeFree(xTexData.shape, xReshaped.shape),\n        () => `packed reshape ${xTexData.shape} to ${\n            xReshaped.shape} isn't free`);\n    const filterReshaped =\n        reshape(filter, [1, convInfo.inChannels, convInfo.outChannels]);\n\n    const pointwiseConv = this.fusedBatchMatMul({\n      a: xReshaped as Tensor3D,\n      b: filterReshaped as Tensor3D,\n      transposeA,\n      transposeB,\n      bias,\n      activation,\n      preluActivationWeights\n    });\n    const pointwiseConvTexData = this.texData.get(pointwiseConv.dataId);\n    util.assert(\n        pointwiseConvTexData.isPacked,\n        () => 'batchMatMul result is expected to be packed');\n    // Restore the input shape to original.\n    xTexData.shape = originalXTexDataShape;\n    // Set the output shape - there is no need for expensive reshape as data\n    // layout is already correct.\n    pointwiseConvTexData.shape = convInfo.outShape;\n    return engine().makeTensorFromDataId(\n               pointwiseConv.dataId, convInfo.outShape, pointwiseConv.dtype) as\n        Tensor4D;\n  }\n\n  private conv2dWithIm2Row(\n      x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo,\n      bias?: Tensor, activation?: backend_util.Activation,\n      preluActivationWeights?: Tensor): Tensor4D {\n    // Rearranges conv2d input so each block to be convolved over forms the\n    // column of a new matrix with shape [filterWidth * filterHeight *\n    // inChannels, outHeight * outWidth]. The filter is also rearranged so each\n    // output channel forms a row of a new matrix with shape [outChannels,\n    // filterWidth * filterHeight * inChannels]. The convolution is then\n    // computed by multiplying these matrices and reshaping the result.\n    const {\n      filterWidth,\n      filterHeight,\n      inChannels,\n      outWidth,\n      outHeight,\n      dataFormat\n    } = convInfo;\n\n    const isChannelsLast = dataFormat === 'channelsLast';\n\n    const sharedDim = filterWidth * filterHeight * inChannels;\n    const numCols = outHeight * outWidth;\n    const x2ColShape = [sharedDim, numCols];\n    const transposeA = true;\n    const transposeB = false;\n\n    const xSqueezed = x.squeeze([0]);\n    const w2Row = filter.reshape([1, sharedDim, -1]);\n\n    const im2ColProgram =\n        new Im2ColPackedProgram(x2ColShape, xSqueezed.shape, convInfo);\n    const im2Col: Tensor3D =\n        this.compileAndRun<Tensor2D>(im2ColProgram, [xSqueezed]).reshape([\n          1, x2ColShape[0], x2ColShape[1]\n        ]);\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, true) : null;\n    const matmulProgram = new MatMulPackedProgram(\n        im2Col.shape, [1, numCols, convInfo.outChannels], transposeA,\n        transposeB, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [im2Col, w2Row];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    const product = this.compileAndRun<Tensor4D>(matmulProgram, inputs);\n\n    if (isChannelsLast) {\n      return product.reshape([1, outHeight, outWidth, convInfo.outChannels]);\n    } else {\n      return product.reshape([1, convInfo.outChannels, outHeight, outWidth]);\n    }\n  }\n\n  fusedConv2d(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && input.shape[0] === 1) {\n      return this.conv2dWithIm2Row(\n          input, filter, convInfo, bias, activation, preluActivationWeights);\n    }\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    const fusedActivation =\n        activation ? mapActivationToShaderProgram(activation, false) : null;\n    const program = new Conv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    const inputs: TensorInfo[] = [input, filter];\n    if (bias) {\n      inputs.push(bias);\n    }\n    if (preluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n    return this.compileAndRun(program, inputs);\n  }\n\n  conv2d(x: Tensor4D, filter: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    if (convInfo.filterHeight === 1 && convInfo.filterWidth === 1 &&\n        convInfo.dilationHeight === 1 && convInfo.dilationWidth === 1 &&\n        convInfo.strideHeight === 1 && convInfo.strideWidth === 1 &&\n        (convInfo.padInfo.type === 'SAME' ||\n         convInfo.padInfo.type === 'VALID')) {\n      return this.conv2dByMatMul(x, filter, convInfo);\n    }\n    if (env().getBool('WEBGL_CONV_IM2COL') && x.shape[0] === 1) {\n      return this.conv2dWithIm2Row(x, filter, convInfo);\n    }\n    const program = new Conv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv2dDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new Conv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv2dDerFilter(x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo):\n      Tensor4D {\n    const program = new Conv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  fusedDepthwiseConv2D(\n      {input, filter, convInfo, bias, activation, preluActivationWeights}:\n          backend_util.FusedConv2DConfig): Tensor4D {\n    const shouldPackDepthwiseConv = env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1;\n    const fusedActivation = activation ?\n        mapActivationToShaderProgram(activation, shouldPackDepthwiseConv) :\n        null;\n    const inputs: Tensor[] = [input, filter];\n\n    const hasBias = bias != null;\n    const hasPreluActivationWeights = preluActivationWeights != null;\n    if (hasBias) {\n      inputs.push(bias);\n    }\n    if (hasPreluActivationWeights) {\n      inputs.push(preluActivationWeights);\n    }\n\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (shouldPackDepthwiseConv) {\n      program = new DepthwiseConvPacked2DProgram(\n          convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n      return this.compileAndRun(program, inputs);\n    }\n\n    program = new DepthwiseConv2DProgram(\n        convInfo, hasBias, fusedActivation, hasPreluActivationWeights);\n    return this.compileAndRun(program, inputs);\n  }\n\n  depthwiseConv2D(\n      x: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    let program: DepthwiseConv2DProgram|DepthwiseConvPacked2DProgram;\n    if (env().getBool('WEBGL_PACK_DEPTHWISECONV') &&\n        convInfo.strideWidth <= 2 &&\n        convInfo.outChannels / convInfo.inChannels === 1) {\n      program = new DepthwiseConvPacked2DProgram(convInfo);\n      return this.compileAndRun(program, [x, filter]);\n    }\n\n    program = new DepthwiseConv2DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  depthwiseConv2DDerInput(\n      dy: Tensor4D, filter: Tensor4D,\n      convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  depthwiseConv2DDerFilter(\n      x: Tensor4D, dy: Tensor4D, convInfo: backend_util.Conv2DInfo): Tensor4D {\n    const program = new DepthwiseConv2DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  conv3d(x: Tensor5D, filter: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DProgram(convInfo);\n    return this.compileAndRun(program, [x, filter]);\n  }\n\n  conv3dDerInput(\n      dy: Tensor5D, filter: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Conv3DDerInputProgram(convInfo);\n    return this.compileAndRun(program, [dy, filter]);\n  }\n\n  conv3dDerFilter(x: Tensor5D, dy: Tensor5D, convInfo: backend_util.Conv3DInfo):\n      Tensor5D {\n    const program = new Conv3DDerFilterProgram(convInfo);\n    return this.compileAndRun(program, [x, dy]);\n  }\n\n  cast<T extends Tensor>(x: T, dtype: DataType): T {\n    return backend_util.castTensor(x, dtype, this);\n  }\n\n  unstack(x: Tensor, axis: number): Tensor[] {\n    const num = x.shape[axis];\n    const outShape: number[] = new Array(x.rank - 1);\n    let outIndex = 0;\n    for (let i = 0; i < x.rank; i++) {\n      if (i !== axis) {\n        outShape[outIndex++] = x.shape[i];\n      }\n    }\n\n    const begin = new Array(x.rank).fill(0);\n    const size = x.shape.slice();\n    size[axis] = 1;\n    const res = new Array(num);\n    for (let i = 0; i < res.length; i++) {\n      begin[axis] = i;\n      res[i] = this.slice(x, begin, size).reshape(outShape);\n    }\n    return res;\n  }\n\n  avgPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'avg', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  avgPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const avgPool3dBackpropProgram = new AvgPool3DBackpropProgram(convInfo);\n    return this.compileAndRun(avgPool3dBackpropProgram, [dy], x.dtype);\n  }\n\n  maxPool3d(x: Tensor5D, convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const program = new Pool3DProgram(convInfo, 'max', false);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  maxPool3dBackprop(\n      dy: Tensor5D, x: Tensor5D, y: Tensor5D,\n      convInfo: backend_util.Conv3DInfo): Tensor5D {\n    const getPositions = true;\n    const maxPool3dPositionsProgram =\n        new Pool3DProgram(convInfo, 'max', getPositions);\n    const maxPool3dPositions: Tensor5D =\n        this.compileAndRun(maxPool3dPositionsProgram, [x]);\n    const maxPool3dBackPropProgram = new MaxPool3DBackpropProgram(convInfo);\n    const result = this.compileAndRun(\n        maxPool3dBackPropProgram, [dy, maxPool3dPositions], x.dtype);\n    maxPool3dPositions.dispose();\n    return result as Tensor5D;\n  }\n\n  resizeBilinear(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = env().getBool('WEBGL_PACK_IMAGE_OPERATIONS') ?\n        new ResizeBilinearPackedProgram(\n            x.shape, newHeight, newWidth, alignCorners) :\n        new ResizeBilinearProgram(x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x], 'float32');\n  }\n\n  resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n      Tensor4D {\n    const program = new ResizeBilinearBackpropProgram(dy, x, alignCorners);\n\n    return this.compileAndRun(program, [dy]);\n  }\n\n  resizeNearestNeighbor(\n      x: Tensor4D, newHeight: number, newWidth: number,\n      alignCorners: boolean): Tensor4D {\n    const program = new ResizeNearestNeighborProgram(\n        x.shape, newHeight, newWidth, alignCorners);\n    return this.compileAndRun(program, [x]);\n  }\n\n  resizeNearestNeighborBackprop(\n      dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n    const program =\n        new ResizeNearestNeigborBackpropProgram(dy, x, alignCorners);\n    return this.compileAndRun(program, [dy]);\n  }\n\n  multinomial(\n      logits: Tensor2D, normalized: boolean, numSamples: number,\n      seed: number): Tensor2D {\n    const probs = normalized ? logits : softmax(logits);\n    const batchSize = probs.shape[0];\n    const numOutcomes = probs.shape[1];\n    const program = new MultinomialProgram(batchSize, numOutcomes, numSamples);\n    const customSetup = program.getCustomSetupFunc(seed);\n    return this.compileAndRun(program, [probs], 'int32', customSetup);\n  }\n\n  oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n      Tensor2D {\n    const program = new OneHotProgram(indices.size, depth, onValue, offValue);\n    return this.compileAndRun(program, [indices]);\n  }\n\n  diag(x: Tensor): Tensor {\n    const program = new DiagProgram(x.size);\n    return this.compileAndRun(program, [x]);\n  }\n\n  cropAndResize(\n      image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n      cropSize: [number, number], method: 'bilinear'|'nearest',\n      extrapolationValue: number): Tensor4D {\n    const program = new CropAndResizeProgram(\n        image.shape, boxes.shape, cropSize, method, extrapolationValue);\n    return this.compileAndRun(program, [image, boxes, boxIndex], 'float32');\n  }\n\n  depthToSpace(x: Tensor4D, blockSize: number, dataFormat: 'NHWC'|'NCHW'):\n      Tensor4D {\n    util.assert(\n        blockSize > 1,\n        () =>\n            `blockSize should be > 1 for depthToSpace, but was: ${blockSize}`);\n\n    const batchSize = x.shape[0];\n    const inputHeight = (dataFormat === 'NHWC') ? x.shape[1] : x.shape[2];\n    const inputWidth = (dataFormat === 'NHWC') ? x.shape[2] : x.shape[3];\n    const inputDepth = (dataFormat === 'NHWC') ? x.shape[3] : x.shape[1];\n\n    const outputHeight = inputHeight * blockSize;\n    const outputWidth = inputWidth * blockSize;\n    const outputDepth = inputDepth / (blockSize * blockSize);\n\n    const outputShape = (dataFormat === 'NHWC') ?\n        [batchSize, outputHeight, outputWidth, outputDepth] :\n        [batchSize, outputDepth, outputHeight, outputWidth];\n\n    const program = new DepthToSpaceProgram(outputShape, blockSize, dataFormat);\n    return this.compileAndRun(program, [x]);\n  }\n\n  split<T extends Tensor>(x: T, sizeSplits: number[], axis: number): T[] {\n    return split(x, sizeSplits, axis);\n  }\n\n  scatterND<R extends Rank>(\n      indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor<R> {\n    const {sliceRank, numUpdates, sliceSize, strides, outputSize} =\n        backend_util.calculateShapes(updates, indices, shape);\n\n    const flattenShape = [outputSize / sliceSize, sliceSize];\n    const flattenIndices = indices.reshape([numUpdates, sliceRank]);\n    const flattenX = updates.reshape([numUpdates, sliceSize]);\n\n    if (outputSize === 0) {\n      return backend_util.reshapeTensor(tensor([]), shape);\n    }\n    const defaultValue = scalar(0);\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, flattenIndices.rank, flattenX.rank, strides,\n        flattenShape);\n    const res: Tensor =\n        this.compileAndRun(program, [flattenX, flattenIndices, defaultValue]);\n    return res.reshape(shape);\n  }\n\n  sparseToDense<R extends Rank>(\n      sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n      defaultValue: Scalar): Tensor<R> {\n    const {sliceRank, numUpdates, strides, outputSize} =\n        backend_util.calculateShapes(sparseValues, sparseIndices, outputShape);\n\n    const sumDupeIndices = false;\n    const program = new ScatterProgram(\n        numUpdates, sliceRank, sparseIndices.rank, sparseValues.rank, strides,\n        [outputSize, 1], sumDupeIndices);\n    const res: Tensor = this.compileAndRun(\n        program, [sparseValues, sparseIndices, defaultValue]);\n    return res.reshape(outputShape);\n  }\n\n  fft(x: Tensor2D): Tensor2D {\n    const inverse = false;\n    return this.fftImpl(x, inverse);\n  }\n\n  ifft(x: Tensor2D): Tensor2D {\n    const inverse = true;\n    return this.fftImpl(x, inverse);\n  }\n\n  private fftImpl(x: Tensor2D, inverse: boolean): Tensor2D {\n    const xData = this.texData.get(x.dataId);\n\n    const realProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.REAL, x.shape, inverse);\n    const imagProgram =\n        new FFTProgram(fft_gpu.COMPLEX_FFT.IMAG, x.shape, inverse);\n    const inputs = [\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.real),\n      this.makeComplexComponentTensorInfo(x, xData.complexTensors.imag),\n    ];\n\n    const real = this.compileAndRun<Tensor>(realProgram, inputs);\n    const imag = this.compileAndRun<Tensor>(imagProgram, inputs);\n    const complex = this.complex(real, imag).as2D(x.shape[0], x.shape[1]);\n    real.dispose();\n    imag.dispose();\n    return complex;\n  }\n\n  gatherND(x: Tensor, indices: Tensor): Tensor {\n    const indicesShape = indices.shape;\n    const sliceRank = indicesShape[indicesShape.length - 1];\n\n    const [resultShape, numSlices, sliceSize, strides] =\n        backend_util.prepareAndValidate(x, indices);\n\n    const flattenIndices = indices.reshape([numSlices, sliceRank]);\n    const flattenX = x.reshape([x.size / sliceSize, sliceSize]);\n    const program =\n        new GatherNDProgram(sliceRank, strides, [numSlices, sliceSize]);\n    const res: Tensor = this.compileAndRun(program, [flattenX, flattenIndices]);\n    return res.reshape(resultShape);\n  }\n\n  fill<R extends Rank>(\n      shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor<R> {\n    dtype = dtype || util.inferDtype(value);\n\n    if (dtype === 'string') {\n      // String type should be handled in CPU memory.\n      const values = util.getArrayFromDType(dtype, util.sizeFromShape(shape));\n      values.fill(value as string);\n      return engine().makeTensor(values, shape, dtype, this) as Tensor<R>;\n    } else {\n      const program = new FillProgram(shape, value as number);\n      const customSetup = program.getCustomSetupFunc(value as number);\n      return this.compileAndRun(program, [], dtype, customSetup);\n    }\n  }\n\n  onesLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    if (x.dtype === 'string') {\n      throw new Error('onesLike is not supported under string dtype');\n    } else {\n      // TODO(cais, smilkov): Add WebGL shader for onesLike:\n      //   https://github.com/tensorflow/tfjs/issues/1293\n      return this.fill(x.shape, 1, x.dtype);\n    }\n  }\n\n  zerosLike<R extends Rank>(x: Tensor<R>): Tensor<R> {\n    return this.fill(x.shape, x.dtype === 'string' ? '' : 0, x.dtype);\n  }\n\n  linspace(start: number, stop: number, num: number): Tensor1D {\n    // TODO: Use CPU implementation due to the precision problem in Safari.\n    return backend_util.linspaceImpl(start, stop, num);\n  }\n\n  makeTensorInfo(shape: number[], dtype: DataType, values?: BackendValues):\n      TensorInfo {\n    const dataId = this.write(values, shape, dtype);\n    this.texData.get(dataId).usage = null;\n    return {dataId, shape, dtype};\n  }\n\n  private makeOutput<T extends Tensor>(\n      shape: number[], dtype: DataType, values?: BackendValues): T {\n    const {dataId} = this.makeTensorInfo(shape, dtype, values);\n    return engine().makeTensorFromDataId(dataId, shape, dtype, this) as T;\n  }\n\n  private unpackTensor(input: TensorInfo): TensorInfo {\n    const program = new UnpackProgram(input.shape);\n    return this.runWebGLProgram(program, [input], input.dtype);\n  }\n\n  private packTensor(input: TensorInfo): TensorInfo {\n    const program = new PackProgram(input.shape);\n    const preventEagerUnpackingOutput = true;\n    return this.runWebGLProgram(\n        program, [input], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOutput);\n  }\n\n  private packedReshape(input: TensorInfo, afterShape: number[]): TensorInfo {\n    const input3DShape = [\n      webgl_util.getBatchDim(input.shape),\n      ...webgl_util.getRowsCols(input.shape)\n    ] as [number, number, number];\n    const input3D: TensorInfo = {\n      dtype: input.dtype,\n      shape: input3DShape,\n      dataId: input.dataId\n    };\n    const afterShapeAs3D = [\n      webgl_util.getBatchDim(afterShape), ...webgl_util.getRowsCols(afterShape)\n    ] as [number, number, number];\n\n    const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n    const preventEagerUnpackingOfOutput = true;\n    const output = this.runWebGLProgram(\n        program, [input3D], input.dtype, null /* customSetup */,\n        preventEagerUnpackingOfOutput);\n    return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n  }\n\n  private decode(dataId: DataId): TensorInfo {\n    const texData = this.texData.get(dataId);\n    const {isPacked, shape, dtype} = texData;\n    const shapeAs3D =\n        webgl_util.getShapeAs3D(shape) as [number, number, number];\n    let program;\n    if (isPacked) {\n      program = new DecodeMatrixPackedProgram(shapeAs3D);\n    } else {\n      program = new DecodeMatrixProgram(shapeAs3D);\n    }\n    const preventEagerUnpackingOfOutput = true;\n    const out = this.runWebGLProgram(\n        program, [{shape: shapeAs3D, dtype, dataId}], dtype,\n        null /* customSetup */, preventEagerUnpackingOfOutput);\n    return {dtype, shape, dataId: out.dataId};\n  }\n\n  runWebGLProgram(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): TensorInfo {\n    const output = this.makeTensorInfo(program.outputShape, outputDtype);\n    const outData = this.texData.get(output.dataId);\n    if (program.packedOutput) {\n      outData.isPacked = true;\n    }\n    if (program.outPackingScheme === tex_util.PackingScheme.DENSE) {\n      const texelShape = tex_util.getDenseTexShape(program.outputShape);\n      // For a densely packed output, we explicitly set texShape\n      // so it doesn't get assigned later according to our typical packing\n      // scheme wherein a single texel can only contain values from adjacent\n      // rows/cols.\n      outData.texShape = texelShape.map(d => d * 2) as [number, number];\n    }\n    if (program.outTexUsage != null) {\n      outData.usage = program.outTexUsage;\n    }\n    if (util.sizeFromShape(output.shape) === 0) {\n      // Short-circuit the computation since the result is empty (has 0 in its\n      // shape).\n      outData.values =\n          util.getTypedArrayFromDType(output.dtype as 'float32', 0);\n      return output;\n    }\n\n    const dataToDispose: TensorInfo[] = [];\n    const inputsData: TensorData[] = inputs.map(input => {\n      if (input.dtype === 'complex64') {\n        throw new Error(\n            `GPGPUProgram does not support complex64 input. For complex64 ` +\n            `dtypes, please separate the program into real and imaginary ` +\n            `parts.`);\n      }\n\n      let texData = this.texData.get(input.dataId);\n\n      if (texData.texture == null) {\n        if (!program.packedInputs &&\n            util.sizeFromShape(input.shape) <=\n                env().getNumber('WEBGL_SIZE_UPLOAD_UNIFORM')) {\n          // Upload small tensors that live on the CPU as uniforms, not as\n          // textures. Do this only when the environment supports 32bit floats\n          // due to problems when comparing 16bit floats with 32bit floats.\n          // TODO(https://github.com/tensorflow/tfjs/issues/821): Make it\n          // possible for packed shaders to sample from uniforms.\n          return {\n            shape: input.shape,\n            texData: null,\n            isUniform: true,\n            uniformValues: texData.values as TypedArray\n          };\n        }\n\n        // This ensures that if a packed program's inputs have not yet been\n        // uploaded to the GPU, they get uploaded as packed right off the bat.\n        if (program.packedInputs) {\n          texData.isPacked = true;\n          texData.shape = input.shape;\n        }\n      } else if (!!texData.isPacked !== !!program.packedInputs) {\n        input = texData.isPacked ? this.unpackTensor(input) :\n                                   this.packTensor(input);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n      } else if (\n          texData.isPacked &&\n          !webgl_util.isReshapeFree(texData.shape, input.shape)) {\n        // This is a special case where a texture exists for a tensor\n        // but the shapes are incompatible (due to packing constraints) because\n        // the tensor did not have a chance to go through the packed reshape\n        // shader. This only happens when we reshape the *same* tensor to form\n        // *distinct* inputs to an op, e.g. dotting a vector with itself. This\n        // case will disappear once packed uploading is the default.\n\n        const savedInput = input;\n        const targetShape = input.shape;\n\n        input.shape = texData.shape;\n        input = this.packedReshape(input as Tensor, targetShape);\n        dataToDispose.push(input);\n        texData = this.texData.get(input.dataId);\n\n        savedInput.shape = targetShape;\n      }\n\n      this.uploadToGPU(input.dataId);\n      return {shape: input.shape, texData, isUniform: false};\n    });\n\n    this.uploadToGPU(output.dataId);\n    const outputData:\n        TensorData = {shape: output.shape, texData: outData, isUniform: false};\n    const key = gpgpu_math.makeShaderKey(program, inputsData, outputData);\n    const binary = this.getAndSaveBinary(key, () => {\n      return gpgpu_math.compileProgram(\n          this.gpgpu, program, inputsData, outputData);\n    });\n    const shouldTimeProgram = this.activeTimers != null;\n    let query: WebGLQuery|CPUTimerQuery;\n    if (shouldTimeProgram) {\n      query = this.startTimer();\n    }\n\n    gpgpu_math.runProgram(\n        this.gpgpu, binary, inputsData, outputData, customSetup);\n\n    dataToDispose.forEach(info => this.disposeIntermediateTensorInfo(info));\n\n    if (shouldTimeProgram) {\n      query = this.endTimer(query);\n      this.activeTimers.push(\n          {name: program.constructor.name, query: this.getQueryTime(query)});\n    }\n\n    if (!env().getBool('WEBGL_LAZILY_UNPACK') && outData.isPacked &&\n        preventEagerUnpackingOfOutput === false) {\n      const unpacked = this.unpackTensor(output);\n      this.disposeIntermediateTensorInfo(output);\n      return unpacked;\n    }\n    return output;\n  }\n\n  compileAndRun<K extends TensorInfo>(\n      program: GPGPUProgram, inputs: TensorInfo[], outputDtype?: DataType,\n      customSetup?: (gpgpu: GPGPUContext, webGLProgram: WebGLProgram) => void,\n      preventEagerUnpackingOfOutput = false): K {\n    outputDtype = outputDtype || inputs[0].dtype;\n    const outInfo = this.runWebGLProgram(\n        program, inputs, outputDtype, customSetup,\n        preventEagerUnpackingOfOutput);\n    return engine().makeTensorFromDataId(\n               outInfo.dataId, outInfo.shape, outInfo.dtype) as {} as K;\n  }\n\n  private getAndSaveBinary(key: string, getBinary: () => GPGPUBinary):\n      GPGPUBinary {\n    if (!(key in this.binaryCache)) {\n      this.binaryCache[key] = getBinary();\n    }\n    return this.binaryCache[key];\n  }\n\n  getTextureManager(): TextureManager {\n    return this.textureManager;\n  }\n\n  private disposed = false;\n\n  dispose() {\n    if (this.disposed) {\n      return;\n    }\n    // Avoid disposing the compiled webgl programs during unit testing because\n    // it slows down test execution.\n    if (!env().getBool('IS_TEST')) {\n      const allKeys = Object.keys(this.binaryCache);\n      allKeys.forEach(key => {\n        this.gpgpu.deleteProgram(this.binaryCache[key].webGLProgram);\n        delete this.binaryCache[key];\n      });\n    }\n    this.textureManager.dispose();\n    if (this.canvas != null &&\n        (typeof (HTMLCanvasElement) !== 'undefined' &&\n         this.canvas instanceof HTMLCanvasElement)) {\n      this.canvas.remove();\n    } else {\n      this.canvas = null;\n    }\n    if (this.gpgpuCreatedLocally) {\n      this.gpgpu.program = null;\n      this.gpgpu.dispose();\n    }\n    this.disposed = true;\n  }\n\n  floatPrecision(): 16|32 {\n    if (this.floatPrecisionValue == null) {\n      this.floatPrecisionValue = tidy(() => {\n        if (!env().get('WEBGL_RENDER_FLOAT32_ENABLED')) {\n          // Momentarily switching DEBUG flag to false so we don't throw an\n          // error trying to upload a small value.\n          const debugFlag = env().getBool('DEBUG');\n          env().set('DEBUG', false);\n          const underflowCheckValue = this.abs(scalar(1e-8)).dataSync()[0];\n          env().set('DEBUG', debugFlag);\n\n          if (underflowCheckValue > 0) {\n            return 32;\n          }\n        }\n        return 16;\n      });\n    }\n    return this.floatPrecisionValue;\n  }\n  /** Returns the smallest representable number.  */\n  epsilon(): number {\n    return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n  }\n\n  private uploadToGPU(dataId: DataId): void {\n    const texData = this.texData.get(dataId);\n    const {shape, dtype, values, texture, usage, isPacked} = texData;\n\n    if (texture != null) {\n      // Array is already on GPU. No-op.\n      return;\n    }\n    const shouldTimeProgram = this.activeTimers != null;\n    let start: number;\n    if (shouldTimeProgram) {\n      start = util.now();\n    }\n\n    let texShape = texData.texShape;\n    if (texShape == null) {\n      texShape = webgl_util.getTextureShapeFromLogicalShape(shape, isPacked);\n      texData.texShape = texShape;\n    }\n\n    if (values != null) {\n      const shapeAs3D = webgl_util.getShapeAs3D(shape);\n\n      let program;\n      let width = texShape[1], height = texShape[0];\n      const isByteArray = values instanceof Uint8Array;\n\n      if (isPacked) {\n        [width, height] = tex_util.getPackedMatrixTextureShapeWidthHeight(\n            texShape[0], texShape[1]);\n        program = new EncodeMatrixPackedProgram(\n            shapeAs3D, [height, width], isByteArray);\n      } else {\n        program =\n            new EncodeMatrixProgram(shapeAs3D, [height, width], isByteArray);\n      }\n\n      const tempDenseInputHandle = this.makeTensorInfo([height, width], dtype);\n      if (isByteArray) {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.PIXELS;\n      } else {\n        this.texData.get(tempDenseInputHandle.dataId).usage =\n            TextureUsage.UPLOAD;\n      }\n      this.gpgpu.uploadDenseMatrixToTexture(\n          this.getTexture(tempDenseInputHandle.dataId), width, height,\n          values as TypedArray);\n\n      // We want the output to remain packed regardless of the value of\n      // WEBGL_PACK.\n      const preventEagerUnpacking = true;\n      const encodedOutputTarget = this.runWebGLProgram(\n          program, [tempDenseInputHandle], dtype, null, preventEagerUnpacking);\n\n      // Have the original texture assume the identity of the encoded output.\n      const outputTexData = this.texData.get(encodedOutputTarget.dataId);\n      texData.texture = outputTexData.texture;\n      texData.texShape = outputTexData.texShape;\n      texData.isPacked = outputTexData.isPacked;\n      texData.usage = outputTexData.usage;\n\n      this.disposeIntermediateTensorInfo(tempDenseInputHandle);\n      this.texData.delete(encodedOutputTarget.dataId);\n\n      // Once uploaded, don't store the values on cpu.\n      texData.values = null;\n      if (shouldTimeProgram) {\n        this.uploadWaitMs += util.now() - start;\n      }\n    } else {\n      const newTexture = this.acquireTexture(texShape, usage, dtype, isPacked);\n      texData.texture = newTexture;\n    }\n  }\n\n  private convertAndCacheOnCPU(dataId: DataId, float32Values?: Float32Array):\n      TypedArray {\n    const texData = this.texData.get(dataId);\n    const {dtype} = texData;\n\n    this.releaseGPUData(dataId);\n\n    if (float32Values != null) {\n      texData.values = float32ToTypedArray(float32Values, dtype as 'float32');\n    }\n    return texData.values as TypedArray;\n  }\n\n  private acquireTexture(\n      texShape: [number, number], texType: TextureUsage, dtype: DataType,\n      isPacked: boolean): WebGLTexture {\n    this.numBytesInGPU += this.computeBytes(texShape, dtype);\n    if (!this.warnedAboutMemory &&\n        this.numBytesInGPU > this.numMBBeforeWarning * 1024 * 1024) {\n      const mb = (this.numBytesInGPU / 1024 / 1024).toFixed(2);\n      this.warnedAboutMemory = true;\n      console.warn(\n          `High memory usage in GPU: ${mb} MB, ` +\n          `most likely due to a memory leak`);\n    }\n    return this.textureManager.acquireTexture(texShape, texType, isPacked);\n  }\n\n  private computeBytes(shape: [number, number], dtype: DataType) {\n    return shape[0] * shape[1] * util.bytesPerElement(dtype);\n  }\n\n  private tryRunOnCpuOrThrow<T extends Tensor>(\n      inputs: TensorInfo[], fn: () => T): T|null {\n    if (this.shouldExecuteOnCPU(inputs)) {\n      try {\n        return fn();\n      } catch (e) {\n        if (env().getBool('IS_TEST')) {\n          throw new Error('CPU forwarding failed');\n        }\n      }\n    }\n    return null;\n  }\n}\n\nfunction float32ToTypedArray<D extends NumericDataType>(\n    a: Float32Array, dtype: D): tf.DataTypeMap[D] {\n  if (dtype === 'float32' || dtype === 'complex64') {\n    return a as tf.DataTypeMap[D];\n  } else if (dtype === 'int32' || dtype === 'bool') {\n    const result = (dtype === 'int32') ? new Int32Array(a.length) :\n                                         new Uint8Array(a.length);\n    for (let i = 0; i < result.length; ++i) {\n      result[i] = Math.round(a[i]);\n    }\n    return result as tf.DataTypeMap[D];\n  } else {\n    throw new Error(`Unknown dtype ${dtype}`);\n  }\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '@tensorflow/tfjs-core';\n\nimport * as gpgpu_util from './gpgpu_util';\nimport * as webgl_util from './webgl_util';\n\nexport {MathBackendWebGL, WebGLMemoryInfo, WebGLTimingInfo} from './backend_webgl';\nexport {setWebGLContext} from './canvas_util';\nexport {GPGPUContext} from './gpgpu_context';\nexport {GPGPUProgram} from './gpgpu_math';\n// WebGL specific utils.\nexport {gpgpu_util, webgl_util};\n\n/**\n * Enforce use of half precision textures if available on the platform.\n *\n * @doc {heading: 'Environment', namespace: 'webgl'}\n */\nexport function forceHalfFloat(): void {\n  env().set('WEBGL_FORCE_F16_TEXTURES', true);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// base.ts is the webgl backend without auto kernel registration.\n\nimport {device_util, registerBackend} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from './backend_webgl';\nexport {version as version_webgl} from './version';\n\nif (device_util.isBrowser()) {\n  registerBackend('webgl', () => new MathBackendWebGL(), 2 /* priority */);\n}\n\n// Export webgl utilities\nexport * from './webgl';\n\n// Export forceHalfFlost under webgl namespace for the union bundle.\nimport {forceHalfFloat} from './webgl';\nexport const webgl = {forceHalfFloat};\n", "import {BinaryInputs, DataType, env, KernelFunc, UnaryInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BinaryOpProgram} from '../binaryop_gpu';\nimport {BinaryOpPackedProgram} from '../binaryop_packed_gpu';\nimport {UnaryOpProgram} from '../unaryop_gpu';\n\nexport const CHECK_NAN_SNIPPET_UNARY = `if (isnan(x)) return x;`;\n\nexport const CHECK_NAN_SNIPPET_BINARY = `\n  if (isnan(a)) return a;\n  if (isnan(b)) return b;\n`;\n\nexport const CHECK_NAN_SNIPPET_BINARY_PACKED = `\n  result.r = isNaN.r > 0. ? NAN : result.r;\n  result.g = isNaN.g > 0. ? NAN : result.g;\n  result.b = isNaN.b > 0. ? NAN : result.b;\n  result.a = isNaN.a > 0. ? NAN : result.a;\n`;\n\n/**\n * Template that creates a `KernelFunc` for unary ops.\n * @param opSnippets Op snippet to create `UnaryOpProgram`.\n */\nexport function unaryKernelFunc(opSnippet: string): KernelFunc {\n  return ({inputs, backend}) => {\n    const {x} = inputs as UnaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = new UnaryOpProgram(x.shape, opSnippet);\n    return webglBackend.runWebGLProgram(program, [x], x.dtype);\n  };\n}\n\n/**\n * Template that creates a `KernelFunc` for binary ops.\n * @param opSnippet Op snippet to create `BinaryOpProgram`.\n * @param packedOpSnippet Op snippet to create `BinaryOpPackedProgram`.\n * @param checkOutOfBoundsForPackedProgram Whether to set checkOutOfBounds=true\n *     when creating BinaryOpPackedProgram.\n * @param dtype Optional. If set, the result has this dtype. Otherwise, the\n *     result has the same dtype as the first input. This is mainly used in\n *     comparison kernels, such as Equal, Less, Greater, etc.\n */\nexport function binaryKernelFunc(\n    opSnippet: string, packedOpSnippet: string,\n    checkOutOfBoundsForPackedProgram?: boolean, dtype?: DataType): KernelFunc {\n  // TODO(jingjin): handle complex64.\n\n  return ({inputs, backend}) => {\n    const {a, b} = inputs as BinaryInputs;\n    const webglBackend = backend as MathBackendWebGL;\n    const program = env().getBool('WEBGL_PACK_BINARY_OPERATIONS') ?\n        new BinaryOpPackedProgram(\n            packedOpSnippet, a.shape, b.shape,\n            !!checkOutOfBoundsForPackedProgram) :\n        new BinaryOpProgram(opSnippet, a.shape, b.shape);\n    const $dtype = dtype || a.dtype;\n    const output = webglBackend.runWebGLProgram(program, [a, b], $dtype);\n    return output;\n  };\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Atan2} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc, CHECK_NAN_SNIPPET_BINARY, CHECK_NAN_SNIPPET_BINARY_PACKED} from '../kernel_utils/kernel_funcs_utils';\n\nconst ATAN2 = CHECK_NAN_SNIPPET_BINARY + `\n  return atan(a, b);\n`;\n\nconst ATAN2_PACKED = `\n  vec4 result = atan(a, b);\n  vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));\n  ` +\n    CHECK_NAN_SNIPPET_BINARY_PACKED + `\n  return result;\n`;\n\nexport const atan2KernelFunc = binaryKernelFunc(ATAN2, ATAN2_PACKED);\n\nexport const atan2Config: KernelConfig = {\n  kernelName: Atan2,\n  backendName: 'webgl',\n  kernelFunc: atan2KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Identity, IdentityInputs, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport function identity(\n    args: {inputs: IdentityInputs, backend: MathBackendWebGL}): TensorInfo {\n  const {inputs, backend} = args;\n  const {x} = inputs;\n\n  backend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: x.shape, dtype: x.dtype};\n}\n\nexport const identityConfig: KernelConfig = {\n  kernelName: Identity,\n  backendName: 'webgl',\n  kernelFunc: identity as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPool, AvgPoolAttrs, AvgPoolInputs, backend_util, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function avgPool(args: {\n  inputs: AvgPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'avgPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in avgPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const avgPoolProgram = new Pool2DProgram(convInfo, 'avg', false);\n  return backend.runWebGLProgram(avgPoolProgram, [x], 'float32');\n}\n\nexport const avgPoolConfig: KernelConfig = {\n  kernelName: AvgPool,\n  backendName: 'webgl',\n  kernelFunc: avgPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {AvgPoolBackprop, AvgPoolBackpropAttrs, AvgPoolBackpropInputs, backend_util, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {AvgPool2DBackpropProgram} from '../avg_pool_backprop_gpu';\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function avgPoolBackprop(args: {\n  inputs: AvgPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: AvgPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input} = inputs;\n  const x = input;\n  assertNotComplex([dy, input], 'avgPoolBackprop');\n  const {filterSize, strides, pad} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad);\n  const avgPoolBackpropProgram = new AvgPool2DBackpropProgram(convInfo);\n  return backend.runWebGLProgram(avgPoolBackpropProgram, [dy], x.dtype);\n}\n\nexport const avgPoolBackpropConfig: KernelConfig = {\n  kernelName: AvgPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: avgPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = '0.0';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = '1.0';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        float x = getXAtOutCoords();\n        float mean = getMeanAtOutCoords();\n        float variance = getVarianceAtOutCoords();\n        float offset = ${offsetSnippet};\n        float scale = ${scaleSnippet};\n        float inv = scale * inversesqrt(variance + float(${varianceEpsilon}));\n        setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class BatchNormPackedProgram implements GPGPUProgram {\n  variableNames: string[];\n  outputShape: number[];\n  userCode: string;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(\n      xShape: number[], meanShape: number[], varianceShape: number[],\n      offsetShape: number[]|null, scaleShape: number[]|null,\n      varianceEpsilon: number) {\n    this.variableNames = ['x', 'mean', 'variance'];\n    backend_util.assertAndGetBroadcastShape(xShape, meanShape);\n    backend_util.assertAndGetBroadcastShape(xShape, varianceShape);\n\n    let offsetSnippet = 'vec4(0.0)';\n    if (offsetShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, offsetShape);\n      this.variableNames.push('offset');\n      offsetSnippet = 'getOffsetAtOutCoords()';\n    }\n\n    let scaleSnippet = 'vec4(1.0)';\n    if (scaleShape != null) {\n      backend_util.assertAndGetBroadcastShape(xShape, scaleShape);\n      this.variableNames.push('scale');\n      scaleSnippet = 'getScaleAtOutCoords()';\n    }\n\n    this.outputShape = xShape;\n    this.userCode = `\n      void main() {\n        vec4 offset = ${offsetSnippet};\n        vec4 scale = ${scaleSnippet};\n\n        vec4 x = getXAtOutCoords();\n        vec4 mean = getMeanAtOutCoords();\n        vec4 variance = getVarianceAtOutCoords();\n\n        vec4 inv = scale * inversesqrt(variance + vec4(${varianceEpsilon}));\n\n        setOutput((x - mean) * inv + offset);\n      }\n    `;\n  }\n}\n", "\n/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, FusedBatchNorm, FusedBatchNormAttrs, FusedBatchNormInputs, KernelConfig, KernelFunc, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {BatchNormProgram} from '../batchnorm_gpu';\nimport {BatchNormPackedProgram} from '../batchnorm_packed_gpu';\n\nexport const batchNormKernelFunc: (params: {\n  inputs: FusedBatchNormInputs,\n  backend: MathBackendWebGL,\n  attrs: FusedBatchNormAttrs\n}) => TensorInfo = ({inputs, backend, attrs}) => {\n  const {x, mean, variance, offset, scale} = inputs;\n\n  util.assert(\n      mean.shape.length === variance.shape.length,\n      () => 'Batch normalization gradient requires mean and variance to have ' +\n          'equal ranks.');\n  util.assert(\n      offset == null || mean.shape.length === offset.shape.length,\n      () => 'Batch normalization gradient requires mean and offset to have ' +\n          'equal ranks.');\n  util.assert(\n      scale == null || mean.shape.length === scale.shape.length,\n      () => 'Batch normalization gradient requires mean and scale to have ' +\n          'equal ranks.');\n\n  let {varianceEpsilon} = attrs;\n  if (varianceEpsilon == null) {\n    varianceEpsilon = 0.001;\n  }\n\n  const finalInputs = [x, mean, variance];\n\n  let offsetShape = null;\n  if (offset != null) {\n    offsetShape = offset.shape;\n    finalInputs.push(offset);\n  }\n\n  let scaleShape = null;\n  if (scale != null) {\n    scaleShape = scale.shape;\n    finalInputs.push(scale);\n  }\n\n  const program = env().getBool('WEBGL_PACK_NORMALIZATION') ?\n      new BatchNormPackedProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon) :\n      new BatchNormProgram(\n          x.shape, mean.shape, variance.shape, offsetShape, scaleShape,\n          varianceEpsilon);\n  const output =\n      backend.runWebGLProgram(program, finalInputs, finalInputs[0].dtype);\n\n  return output;\n};\n\nexport const batchNormConfig: KernelConfig = {\n  kernelName: FusedBatchNorm,\n  backendName: 'webgl',\n  kernelFunc: batchNormKernelFunc as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Cos, KernelConfig} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst COS = CHECK_NAN_SNIPPET_UNARY + `\n  return cos(x);\n`;\n\nexport const cosKernelFunc = unaryKernelFunc(COS);\n\nexport const cosConfig: KernelConfig = {\n  kernelName: Cos,\n  backendName: 'webgl',\n  kernelFunc: cosKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Div} from '@tensorflow/tfjs-core';\nimport {KernelConfig} from '@tensorflow/tfjs-core';\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\n// Without the equality check div produces 0.9999 for a = b, which when\n// floored can cause errors.\nconst DIV = `\nif (a == b) {\n  return 1.0;\n};\nreturn a / b;`;\n\n// We do the same as in ./binaryop_gpu, with vec4 and ivec4.\n// On Linux, the vectorized implementation produces NaNs when a and b are 0.\nconst DIV_PACKED = `\n  // vec4 one = vec4(equal(a, b));\n  // return one + (vec4(1.0) - one) * a / b;\n  vec4 result = a / b;\n  if(a.x == b.x) {\n    result.x = 1.;\n  }\n  if(a.y == b.y) {\n    result.y = 1.;\n  }\n  if(a.z == b.z) {\n    result.z = 1.;\n  }\n  if(a.w == b.w) {\n    result.w = 1.;\n  }\n\n  return result;\n`;\n\nexport const divKernelFunc = binaryKernelFunc(\n    DIV, DIV_PACKED, true /* checkOutOfBoundsForPackedProgram */);\n\nexport const divConfig: KernelConfig = {\n  kernelName: Div,\n  backendName: 'webgl',\n  kernelFunc: divKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class FlipLeftRightProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(imageShape: [number, number, number, number]) {\n    const imageWidth = imageShape[2];\n    this.outputShape = imageShape;\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n\n          int coordX = ${imageWidth} - x;\n          float outputValue;\n          if(coordX >= 0 && coordX < ${imageWidth}) {\n            outputValue = getImage(coords[0], coords[1], coordX, coords[3]);\n          } else {\n            outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {FlipLeftRight, FlipLeftRightInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {FlipLeftRightProgram} from '../flip_left_right_gpu';\n\nexport const flipLeftRightConfig: KernelConfig = {\n  kernelName: FlipLeftRight,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend}) => {\n    const {image} = inputs as FlipLeftRightInputs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new FlipLeftRightProgram((image as Tensor4D).shape);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n        vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${width}.0, ${height}.0);\n\n        vec4 values = ${glsl.texture2D}(A, uv);\n        float value;\n        if (depth == 0) {\n          value = values.r;\n        } else if (depth == 1) {\n          value = values.g;\n        } else if (depth == 2) {\n          value = values.b;\n        } else if (depth == 3) {\n          value = values.a;\n        }\n\n        setOutput(floor(value * 255.0 + 0.5));\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {getGlslDifferences} from '../../glsl_version';\nimport {GPGPUProgram} from '../../gpgpu_math';\n\nexport class FromPixelsPackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  userCode: string;\n  outputShape: number[];\n  packedInputs = false;\n  packedOutput = true;\n\n  constructor(outputShape: number[]) {\n    const glsl = getGlslDifferences();\n    const [height, width, ] = outputShape;\n    this.outputShape = outputShape;\n    this.userCode = `\n      void main() {\n        ivec3 coords = getOutputCoords();\n        int texR = coords[0];\n        int texC = coords[1];\n        int depth = coords[2];\n\n        vec4 result = vec4(0.);\n\n        for(int row=0; row<=1; row++) {\n          for(int col=0; col<=1; col++) {\n            texC = coords[1] + row;\n            depth = coords[2] + col;\n\n            vec2 uv = (vec2(texC, texR) + halfCR) /\n                       vec2(${width}.0, ${height}.0);\n            vec4 values = ${glsl.texture2D}(A, uv);\n            float value;\n            if (depth == 0) {\n              value = values.r;\n            } else if (depth == 1) {\n              value = values.g;\n            } else if (depth == 2) {\n              value = values.b;\n            } else if (depth == 3) {\n              value = values.a;\n            }\n\n            result[row * 2 + col] = floor(value * 255.0 + 0.5);\n          }\n        }\n\n        ${glsl.output} = result;\n      }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, KernelConfig, KernelFunc, TensorInfo} from '@tensorflow/tfjs-core';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {TextureUsage} from '../tex_util';\n\nimport {FromPixelsProgram} from './FromPixels_utils/from_pixels_gpu';\nimport {FromPixelsPackedProgram} from './FromPixels_utils/from_pixels_packed_gpu';\n\nexport const fromPixelsConfig: KernelConfig = {\n  kernelName: FromPixels,\n  backendName: 'webgl',\n  kernelFunc: fromPixels as {} as KernelFunc,\n};\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\nfunction fromPixels(args: {\n  inputs: FromPixelsInputs,\n  backend: MathBackendWebGL,\n  attrs: FromPixelsAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  let {pixels} = inputs;\n  const {numChannels} = attrs;\n\n  const isVideo = typeof (HTMLVideoElement) !== 'undefined' &&\n      pixels instanceof HTMLVideoElement;\n  const isImage = typeof (HTMLImageElement) !== 'undefined' &&\n      pixels instanceof HTMLImageElement;\n  const [width, height] = isVideo ?\n      [\n        (pixels as HTMLVideoElement).videoWidth,\n        (pixels as HTMLVideoElement).videoHeight\n      ] :\n      [pixels.width, pixels.height];\n\n  const texShape: [number, number] = [height, width];\n  const outShape = [height, width, numChannels];\n\n  if (isImage || isVideo) {\n    if (fromPixels2DContext == null) {\n      fromPixels2DContext = document.createElement('canvas').getContext('2d');\n    }\n\n    fromPixels2DContext.canvas.width = width;\n    fromPixels2DContext.canvas.height = height;\n    fromPixels2DContext.drawImage(\n        pixels as HTMLVideoElement | HTMLImageElement, 0, 0, width, height);\n    pixels = fromPixels2DContext.canvas;\n  }\n\n  const tempPixelHandle = backend.makeTensorInfo(texShape, 'int32');\n  // This is a byte texture with pixels.\n  backend.texData.get(tempPixelHandle.dataId).usage = TextureUsage.PIXELS;\n  backend.gpgpu.uploadPixelDataToTexture(\n      backend.getTexture(tempPixelHandle.dataId), pixels as ImageData);\n  const program = env().getBool('WEBGL_PACK') ?\n      new FromPixelsPackedProgram(outShape) :\n      new FromPixelsProgram(outShape);\n  const res = backend.runWebGLProgram(program, [tempPixelHandle], 'int32');\n  backend.disposeData(tempPixelHandle.dataId);\n  return res;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, DataType, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReduceProgram} from '../reduce_gpu';\n\ntype ReduceTypes = 'all'|'any'|'max'|'min'|'sum'|'prod';\n\n// Returns an array of configuration objects that describe each stage of the\n// reduction.\nfunction getReductionStages(inShape: number[]):\n    Array<{inSize: number, windowSize: number, outSize: number}> {\n  const stages = [];\n\n  while (stages.length === 0 || stages[stages.length - 1].outSize !== 1) {\n    const outSize: number =\n        stages.length ? stages[stages.length - 1].outSize : inShape[1];\n    const windowSize = backend_util.computeOptimalWindowSize(outSize);\n    stages.push({\n      inSize: outSize,\n      windowSize,\n      outSize: Math.ceil(outSize / windowSize)\n    });\n  }\n\n  return stages;\n}\n\nexport function reduce(\n    x: TensorInfo, dtype: DataType, reductionType: ReduceTypes,\n    backend: MathBackendWebGL): TensorInfo {\n  const reductionStages = getReductionStages(x.shape);\n\n  let result = x;\n  for (let i = 0; i < reductionStages.length; i++) {\n    const {inSize, windowSize, outSize} = reductionStages[i];\n\n    const program = new ReduceProgram(\n        {windowSize, inSize, batchSize: x.shape[0], outSize}, reductionType);\n    const previousResult = result;\n    result = backend.runWebGLProgram(program, [result], dtype);\n\n    if (previousResult.dataId !== x.dataId) {\n      backend.disposeData(previousResult.dataId);\n    }\n  }\n\n  return result;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {ReshapePackedProgram} from '../reshape_packed_gpu';\nimport {getBatchDim, getRowsCols} from '../webgl_util';\n\nexport function packedReshape(\n    input: TensorInfo, afterShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const input3DShape =\n      [getBatchDim(input.shape),\n       ...getRowsCols(input.shape)] as [number, number, number];\n  const input3D: TensorInfo = {\n    dtype: input.dtype,\n    shape: input3DShape,\n    dataId: input.dataId\n  };\n  const afterShapeAs3D =\n      [getBatchDim(afterShape),\n       ...getRowsCols(afterShape)] as [number, number, number];\n\n  const program = new ReshapePackedProgram(afterShapeAs3D, input3DShape);\n  const preventEagerUnpackingOfOutput = true;\n  const output = backend.runWebGLProgram(\n      program, [input3D], input.dtype, null /* customSetup */,\n      preventEagerUnpackingOfOutput);\n  return {dataId: output.dataId, shape: afterShape, dtype: output.dtype};\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, Reshape, ReshapeAttrs, ReshapeInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {packedReshape} from '../kernel_utils/reshape';\nimport {isReshapeFree} from '../webgl_util';\n\nexport function reshape(args: {\n  inputs: ReshapeInputs,\n  backend: MathBackendWebGL,\n  attrs: ReshapeAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  const {shape} = attrs;\n  const webglBackend = backend;\n\n  const xSize = util.sizeFromShape(x.shape);\n  const $shape = util.inferFromImplicitShape(shape, xSize);\n  const $xSize = util.sizeFromShape($shape);\n\n  util.assert(\n      xSize === $xSize,\n      () => `The new shape (${$shape}) has ${$xSize} elements and the old ` +\n          `shape (${x.shape}) has ${xSize} elements. The new shape and old ` +\n          `shape must have the same number of elements.`);\n\n  const xTexData = webglBackend.texData.get(x.dataId);\n  if (xTexData.isPacked && !isReshapeFree(x.shape, $shape) &&\n      !(xTexData.texture !== null && isReshapeFree(xTexData.shape, $shape))) {\n    return packedReshape(x, $shape, webglBackend);\n  }\n\n  webglBackend.incRef(x.dataId);\n\n  return {dataId: x.dataId, shape: $shape, dtype: x.dtype};\n}\n\nexport const reshapeConfig: KernelConfig = {\n  kernelName: Reshape,\n  backendName: 'webgl',\n  kernelFunc: reshape as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {reduce} from '../kernel_utils/reduce';\nimport {reshape} from '../kernels/Reshape';\n\nexport function maxImpl(\n    x: TensorInfo, reduceShape: number[], outShape: number[],\n    backend: MathBackendWebGL): TensorInfo {\n  const inSize = util.sizeFromShape(reduceShape);\n  const xSize = util.sizeFromShape(x.shape);\n  const batchSize = xSize / inSize;\n  const reshapedInput =\n      reshape({inputs: {x}, attrs: {shape: [batchSize, inSize]}, backend});\n\n  const reduced = reduce(reshapedInput, x.dtype, 'max', backend);\n  const reshapedOutput =\n      reshape({inputs: {x: reduced}, attrs: {shape: outShape}, backend});\n\n  backend.disposeIntermediateTensorInfo(reshapedInput);\n  backend.disposeIntermediateTensorInfo(reduced);\n\n  return reshapedOutput;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposeProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    const dtype = getCoordsDataType(this.rank);\n    const switched = getSwitchedCoords(newDim);\n\n    this.userCode = `\n    void main() {\n      ${dtype} resRC = getOutputCoords();\n      setOutput(getA(${switched}));\n    }\n    `;\n  }\n}\n\nfunction getSwitchedCoords(newDim: number[]): string {\n  const rank = newDim.length;\n  if (rank > 6) {\n    throw Error(`Transpose for rank ${rank} is not yet supported`);\n  }\n  const originalOrder =\n      ['resRC.x', 'resRC.y', 'resRC.z', 'resRC.w', 'resRC.u', 'resRC.v'];\n  const switchedCoords = new Array(rank);\n  for (let i = 0; i < newDim.length; i++) {\n    switchedCoords[newDim[i]] = originalOrder[i];\n  }\n  return switchedCoords.join();\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {GPGPUProgram} from './gpgpu_math';\nimport {getVecChannels} from './packing_util';\nimport {getCoordsDataType} from './shader_compiler';\n\nexport class TransposePackedProgram implements GPGPUProgram {\n  variableNames = ['A'];\n  outputShape: number[];\n  userCode: string;\n  rank: number;\n  packedInputs = true;\n  packedOutput = true;\n\n  constructor(aShape: number[], newDim: number[]) {\n    const outputShape: number[] = new Array(aShape.length);\n    for (let i = 0; i < outputShape.length; i++) {\n      outputShape[i] = aShape[newDim[i]];\n    }\n    this.outputShape = outputShape;\n    this.rank = outputShape.length;\n    if (this.rank > 6) {\n      throw Error(\n          `Packed transpose for rank ${this.rank} is not yet supported.`);\n    }\n    const dtype = getCoordsDataType(this.rank);\n\n    const outputOrder = getVecChannels('rc', this.rank);\n    const switchedOrder = new Array(this.rank);\n    for (let i = 0; i < newDim.length; i++) {\n      switchedOrder[newDim[i]] = outputOrder[i];\n    }\n    const innerDims = `vec2(${switchedOrder.slice(-2).join()})`;\n    const nextColumn =\n        `++${outputOrder[this.rank - 1]} < ${outputShape[this.rank - 1]}`;\n    const getc = `getChannel(getA(${switchedOrder.join()}), ${innerDims})`;\n\n    this.userCode = `\n    void main() {\n      ${dtype} rc = getOutputCoords();\n      vec4 result = vec4(0.);\n      result[0] = ${getc};\n      if(${nextColumn}) {\n        result[1] = ${getc};\n      }\n      --${outputOrder[this.rank - 1]};\n      if(++${outputOrder[this.rank - 2]} < ${outputShape[this.rank - 2]}) {\n        result[2] = ${getc};\n        if(${nextColumn}) {\n          result[3] = ${getc};\n        }\n      }\n      setOutput(result);\n    }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {transposeImplCPU} from '../kernel_utils/shared';\nimport {TransposeProgram} from '../transpose_gpu';\nimport {TransposePackedProgram} from '../transpose_packed_gpu';\n\nexport function transposeImpl(\n    x: TensorInfo, perm: number[], backend: MathBackendWebGL): TensorInfo {\n  const program = env().getBool('WEBGL_PACK_ARRAY_OPERATIONS') ?\n      new TransposePackedProgram(x.shape, perm) :\n      new TransposeProgram(x.shape, perm);\n  return backend.runWebGLProgram(program, [x], x.dtype);\n}\n\nexport {transposeImplCPU};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Max, MaxAttrs, MaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, TypedArray, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {maxImplCPU} from '../kernel_utils/shared';\n\nimport {maxImpl} from './Max_impl';\nimport {transposeImpl, transposeImplCPU} from './Transpose_impl';\n\nexport const maxConfig: KernelConfig = {\n  kernelName: Max,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxInputs;\n    const {reductionIndices, keepDims} = attrs as {} as MaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const origAxes = util.parseAxisParam(reductionIndices, x.shape);\n    let axes = origAxes;\n    const permutedAxes = backend_util.getAxesPermutation(axes, xRank);\n    const maxInputIsTransposed = permutedAxes != null;\n    const shouldExecuteOnCPU = webglBackend.shouldExecuteOnCPU([x]);\n\n    let maxInput = x;\n    if (maxInputIsTransposed) {\n      if (shouldExecuteOnCPU) {\n        const xTexData = webglBackend.texData.get(maxInput.dataId);\n        const values = xTexData.values as TypedArray;\n\n        const newShape: number[] = new Array(xRank);\n        for (let i = 0; i < newShape.length; i++) {\n          newShape[i] = x.shape[permutedAxes[i]];\n        }\n        const maxInputValues =\n            transposeImplCPU(values, x.shape, x.dtype, permutedAxes, newShape);\n\n        maxInput = webglBackend.makeTensorInfo(newShape, x.dtype);\n        const maxInputData = webglBackend.texData.get(maxInput.dataId);\n        maxInputData.values = maxInputValues;\n      } else {\n        maxInput = transposeImpl(x, permutedAxes, webglBackend);\n      }\n\n      axes = backend_util.getInnerMostAxes(axes.length, xRank);\n    }\n\n    backend_util.assertAxesAreInnerMostDims('max', axes, xRank);\n    const [maxOutShape, reduceShape] =\n        backend_util.computeOutAndReduceShapes(maxInput.shape, axes);\n\n    let outShape = maxOutShape;\n    if (keepDims) {\n      // rather than reshape at the end, set the target shape here.\n      outShape = backend_util.expandShapeToKeepDim(maxOutShape, origAxes);\n    }\n\n    let out;\n    if (shouldExecuteOnCPU) {\n      const xTexData = webglBackend.texData.get(maxInput.dataId);\n      const values = xTexData.values as TypedArray;\n\n      const outValues = maxImplCPU(\n          values, util.sizeFromShape(reduceShape), outShape, x.dtype);\n\n      out = webglBackend.makeTensorInfo(outShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = maxImpl(maxInput, reduceShape, outShape, webglBackend);\n    }\n\n    if (maxInputIsTransposed) {\n      webglBackend.disposeIntermediateTensorInfo(maxInput);\n    }\n\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPool, MaxPoolAttrs, MaxPoolInputs, TensorInfo, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\nimport {identity} from './Identity';\n\nexport function maxPool(args: {\n  inputs: MaxPoolInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {x} = inputs;\n  assertNotComplex(x, 'maxPool');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n  const dilations = 1;\n\n  util.assert(\n      backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n      () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n          `Got strides ${strides} and dilations '${dilations}'`);\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      dilations, pad, dimRoundingMode);\n  if (convInfo.filterWidth === 1 && convInfo.filterHeight === 1 &&\n      util.arraysEqual(convInfo.inShape, convInfo.outShape)) {\n    return identity({inputs: {x}, backend});\n  }\n  const maxPoolProgram = new Pool2DProgram(convInfo, 'max', false);\n  return backend.runWebGLProgram(maxPoolProgram, [x], x.dtype);\n}\n\nexport const maxPoolConfig: KernelConfig = {\n  kernelName: MaxPool,\n  backendName: 'webgl',\n  kernelFunc: maxPool as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {backend_util, KernelConfig, KernelFunc, MaxPoolBackprop, MaxPoolBackpropAttrs, MaxPoolBackpropInputs, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {MaxPool2DBackpropProgram} from '../max_pool_backprop_gpu';\nimport {Pool2DProgram} from '../pool_gpu';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function maxPoolBackprop(args: {\n  inputs: MaxPoolBackpropInputs,\n  backend: MathBackendWebGL,\n  attrs: MaxPoolBackpropAttrs\n}): TensorInfo {\n  const {inputs, backend, attrs} = args;\n  const {dy, input, output} = inputs;\n  const x = input;\n  assertNotComplex([input, output], 'maxPoolBackprop');\n  const {filterSize, strides, pad, dimRoundingMode} = attrs;\n\n  const convInfo = backend_util.computePool2DInfo(\n      x.shape as [number, number, number, number], filterSize, strides,\n      1 /* dilations */, pad, dimRoundingMode);\n  const getPositions = true;\n  const maxPoolPositionsProgram =\n      new Pool2DProgram(convInfo, 'max', getPositions);\n  const maxPoolPositions: TensorInfo =\n      backend.runWebGLProgram(maxPoolPositionsProgram, [x], x.dtype);\n\n  const maxPoolBackPropProgram = new MaxPool2DBackpropProgram(convInfo);\n  const result = backend.runWebGLProgram(\n      maxPoolBackPropProgram, [dy, maxPoolPositions], x.dtype);\n  backend.disposeIntermediateTensorInfo(maxPoolPositions);\n  return result;\n}\n\nexport const maxPoolBackpropConfig: KernelConfig = {\n  kernelName: MaxPoolBackprop,\n  backendName: 'webgl',\n  kernelFunc: maxPoolBackprop as {} as KernelFunc\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, TensorInfo} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {Pool2DProgram} from '../pool_gpu';\n\nexport function maxPoolWithArgmaxImpl(\n    x: TensorInfo, includeBatchInIndex: boolean,\n    convInfo: backend_util.Conv2DInfo,\n    backend: MathBackendWebGL): TensorInfo[] {\n  let program = new Pool2DProgram(convInfo, 'max', false);\n  const poolOutput = backend.runWebGLProgram(program, [x], 'float32');\n\n  program = new Pool2DProgram(convInfo, 'max', true, true, includeBatchInIndex);\n  const indexOutput = backend.runWebGLProgram(program, [x], 'float32');\n  return [poolOutput, indexOutput];\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {MaxPoolWithArgmax, MaxPoolWithArgmaxAttrs, MaxPoolWithArgmaxInputs} from '@tensorflow/tfjs-core';\nimport {backend_util, KernelConfig, util} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {maxPoolWithArgmaxImpl} from './MaxPoolWithArgmax_impl';\n\nexport const maxPoolWithArgmaxConfig: KernelConfig = {\n  kernelName: MaxPoolWithArgmax,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as MaxPoolWithArgmaxInputs;\n    const {filterSize, strides, pad, includeBatchInIndex} =\n        attrs as {} as MaxPoolWithArgmaxAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    util.assert(\n        x.shape.length === 4,\n        () => `Error in maxPool: input must be rank 4 but got rank ${\n            x.shape.length}.`);\n    const dilations: [number, number] = [1, 1];\n    util.assert(\n        backend_util.eitherStridesOrDilationsAreOne(strides, dilations),\n        () => 'Error in maxPool: Either strides or dilations must be 1. ' +\n            `Got strides ${strides} and dilations '${dilations}'`);\n\n    const convInfo = backend_util.computePool2DInfo(\n        x.shape as [number, number, number, number], filterSize, strides,\n        dilations, pad);\n\n    const [result, indexes] =\n        maxPoolWithArgmaxImpl(x, includeBatchInIndex, convInfo, webglBackend);\n    return [result, indexes];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV3, NonMaxSuppressionV3Attrs, NonMaxSuppressionV3Inputs, TypedArray} from '@tensorflow/tfjs-core';\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV3Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV3,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV3Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold} =\n        attrs as unknown as NonMaxSuppressionV3Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n\n    return kernel_impls.nonMaxSuppressionV3Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal);\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, NonMaxSuppressionV4, NonMaxSuppressionV4Attrs, NonMaxSuppressionV4Inputs, TypedArray} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV4Impl = kernel_impls.nonMaxSuppressionV4Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV4Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV4,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV4Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, padToMaxOutputSize} =\n        attrs as unknown as NonMaxSuppressionV4Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const {selectedIndices, validOutputs} = nonMaxSuppressionV4Impl(\n        boxesVals, scoresVals, maxOutputSize, iouThreshold, scoreThreshold,\n        padToMaxOutputSize);\n\n    return [selectedIndices, validOutputs];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util, kernel_impls, KernelConfig, TypedArray} from '@tensorflow/tfjs-core';\nimport {NonMaxSuppressionV5, NonMaxSuppressionV5Attrs, NonMaxSuppressionV5Inputs} from '@tensorflow/tfjs-core';\nconst nonMaxSuppressionV5Impl = kernel_impls.nonMaxSuppressionV5Impl;\nimport {MathBackendWebGL} from '../backend_webgl';\n\nexport const nonMaxSuppressionV5Config: KernelConfig = {\n  kernelName: NonMaxSuppressionV5,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, backend, attrs}) => {\n    backend_util.warn(\n        'tf.nonMaxSuppression() in webgl locks the UI thread. ' +\n        'Call tf.nonMaxSuppressionAsync() instead');\n\n    const {boxes, scores} = inputs as NonMaxSuppressionV5Inputs;\n    const {maxOutputSize, iouThreshold, scoreThreshold, softNmsSigma} =\n        attrs as unknown as NonMaxSuppressionV5Attrs;\n\n    const gpuBackend = backend as MathBackendWebGL;\n\n    const boxesVals = gpuBackend.readSync(boxes.dataId) as TypedArray;\n    const scoresVals = gpuBackend.readSync(scores.dataId) as TypedArray;\n\n    const maxOutputSizeVal = maxOutputSize;\n    const iouThresholdVal = iouThreshold;\n    const scoreThresholdVal = scoreThreshold;\n    const softNmsSigmaVal = softNmsSigma;\n\n    const {selectedIndices, selectedScores} = nonMaxSuppressionV5Impl(\n        boxesVals, scoresVals, maxOutputSizeVal, iouThresholdVal,\n        scoreThresholdVal, softNmsSigmaVal);\n\n    return [selectedIndices, selectedScores];\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {backend_util} from '@tensorflow/tfjs-core';\n\nimport {GPGPUProgram} from './gpgpu_math';\n\nexport class RotateProgram implements GPGPUProgram {\n  variableNames = ['Image'];\n  outputShape: number[] = [];\n  userCode: string;\n\n  constructor(\n      imageShape: [number, number, number, number], radians: number,\n      fillValue: number|[number, number, number],\n      center: number|[number, number]) {\n    const imageHeight = imageShape[1];\n    const imageWidth = imageShape[2];\n    const sinFactor = Math.sin(radians).toFixed(3);\n    const cosFactor = Math.cos(radians).toFixed(3);\n    this.outputShape = imageShape;\n\n    const [centerX, centerY] =\n        backend_util.getImageCenter(center, imageHeight, imageWidth);\n    const centerXString = centerX.toFixed(3);\n    const centerYString = centerY.toFixed(3);\n\n    let fillSnippet = '';\n    if (typeof fillValue === 'number') {\n      fillSnippet = `float outputValue = ${fillValue.toFixed(2)};`;\n    } else {\n      fillSnippet = `\n        vec3 fill = vec3(${fillValue.join(',')});\n        float outputValue = fill[coords[3]];`;\n    }\n\n    this.userCode = `\n        void main() {\n          ivec4 coords = getOutputCoords();\n          int x = coords[2];\n          int y = coords[1];\n          float coordXFloat = (float(x) - ${centerXString}) * ${\n        cosFactor} - (float(y) - ${centerYString}) * ${sinFactor};\n          float coordYFloat = (float(x) - ${centerXString}) * ${\n        sinFactor} + (float(y) - ${centerYString}) * ${cosFactor};\n          int coordX = int(round(coordXFloat + ${centerXString}));\n          int coordY = int(round(coordYFloat + ${centerYString}));\n          ${fillSnippet}\n          if(coordX >= 0 && coordX < ${imageWidth} && coordY >= 0 && coordY < ${\n        imageHeight}) {\n            outputValue = getImage(coords[0], coordY, coordX, coords[3]);\n          }\n          setOutput(outputValue);\n        }\n    `;\n  }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tensor4D} from '@tensorflow/tfjs-core';\nimport {RotateWithOffset, RotateWithOffsetAttrs, RotateWithOffsetInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {RotateProgram} from '../rotate_gpu';\n\nexport const rotateWithOffsetConfig: KernelConfig = {\n  kernelName: RotateWithOffset,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {image} = inputs as RotateWithOffsetInputs;\n    const {radians, fillValue, center} = attrs as {} as RotateWithOffsetAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const program = new RotateProgram(\n        (image as Tensor4D).shape, radians, fillValue, center);\n    const output = webglBackend.runWebGLProgram(program, [image], image.dtype);\n    return output;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Sin} from '@tensorflow/tfjs-core';\n\nimport {CHECK_NAN_SNIPPET_UNARY, unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SIN = CHECK_NAN_SNIPPET_UNARY + `\n  return sin(x);\n`;\n\nexport const sinKernelFunc = unaryKernelFunc(SIN);\n\nexport const sinConfig: KernelConfig = {\n  kernelName: Sin,\n  backendName: 'webgl',\n  kernelFunc: sinKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Square} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARE = `return x * x;`;\n\nexport const squareKernelFunc = unaryKernelFunc(SQUARE);\n\nexport const squareConfig: KernelConfig = {\n  kernelName: Square,\n  backendName: 'webgl',\n  kernelFunc: squareKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, SquaredDifference} from '@tensorflow/tfjs-core';\n\nimport {binaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst SQUARED_DIFFERENCE = 'return (a - b) * (a - b);';\n\nexport const squaredDifferenceKernelFunc =\n    binaryKernelFunc(SQUARED_DIFFERENCE, SQUARED_DIFFERENCE);\n\nexport const squaredDifferenceConfig: KernelConfig = {\n  kernelName: SquaredDifference,\n  backendName: 'webgl',\n  kernelFunc: squaredDifferenceKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, Tan} from '@tensorflow/tfjs-core';\n\nimport {unaryKernelFunc} from '../kernel_utils/kernel_funcs_utils';\n\nconst TAN = `return tan(x);`;\n\nexport const tanKernelFunc = unaryKernelFunc(TAN);\n\nexport const tanConfig: KernelConfig = {\n  kernelName: Tan,\n  backendName: 'webgl',\n  kernelFunc: tanKernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, TensorInfo, Transpose, TransposeAttrs, TransposeInputs, TypedArray} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\n\nimport {transposeImpl} from './Transpose_impl';\nimport {transposeImplCPU as cpuTranspose} from './Transpose_impl';\n\nexport const transposeConfig: KernelConfig = {\n  kernelName: Transpose,\n  backendName: 'webgl',\n  kernelFunc: ({inputs, attrs, backend}) => {\n    const {x} = inputs as TransposeInputs;\n    const {perm} = attrs as {} as TransposeAttrs;\n    const webglBackend = backend as MathBackendWebGL;\n\n    const xRank = x.shape.length;\n\n    const newShape: number[] = new Array(xRank);\n    for (let i = 0; i < newShape.length; i++) {\n      newShape[i] = x.shape[perm[i]];\n    }\n\n    let out: TensorInfo;\n    if (webglBackend.shouldExecuteOnCPU([x])) {\n      const xTexData = webglBackend.texData.get(x.dataId);\n      const values = xTexData.values as TypedArray;\n      const outValues = cpuTranspose(values, x.shape, x.dtype, perm, newShape);\n\n      out = webglBackend.makeTensorInfo(newShape, x.dtype);\n      const outData = webglBackend.texData.get(out.dataId);\n      outData.values = outValues;\n    } else {\n      out = transposeImpl(x, perm, webglBackend);\n    }\n    return out;\n  }\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the License);\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an AS IS BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelConfig, KernelFunc, TensorInfo, Unique, UniqueAttrs, UniqueInputs} from '@tensorflow/tfjs-core';\n\nimport {MathBackendWebGL} from '../backend_webgl';\nimport {uniqueImplCPU} from '../kernel_utils/shared';\nimport {assertNotComplex} from '../webgl_util';\n\nexport function unique(\n    args:\n        {inputs: UniqueInputs, attrs: UniqueAttrs, backend: MathBackendWebGL}):\n    TensorInfo[] {\n  const {inputs, attrs, backend} = args;\n  const {axis} = attrs;\n  const {x} = inputs;\n  assertNotComplex(x, 'unique');\n\n  // For now, always forward calculation to the CPU backend.\n  console.warn(\n      'WARNING: ',\n      'UI might be locked temporarily as data is being downloaded');\n  const values = backend.readSync(x.dataId);\n  const {outputValues, outputShape, indices} =\n      uniqueImplCPU(values, axis, x.shape, x.dtype);\n  return [\n    backend.makeTensorInfo(outputShape, x.dtype, outputValues),\n    backend.makeTensorInfo([indices.length], 'int32', indices),\n  ];\n}\n\nexport const uniqueConfig: KernelConfig = {\n  kernelName: Unique,\n  backendName: 'webgl',\n  kernelFunc: unique as {} as KernelFunc,\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {KernelConfig, registerKernel} from '@tensorflow/tfjs-core';\n\nimport {atan2Config} from './kernels/Atan2';\nimport {avgPoolConfig} from './kernels/AvgPool';\nimport {avgPoolBackpropConfig} from './kernels/AvgPoolBackprop';\nimport {batchNormConfig} from './kernels/BatchNorm';\nimport {cosConfig} from './kernels/Cos';\nimport {divConfig} from './kernels/Div';\nimport {flipLeftRightConfig} from './kernels/FlipLeftRight';\nimport {fromPixelsConfig} from './kernels/FromPixels';\nimport {identityConfig} from './kernels/Identity';\nimport {maxConfig} from './kernels/Max';\nimport {maxPoolConfig} from './kernels/MaxPool';\nimport {maxPoolBackpropConfig} from './kernels/MaxPoolBackprop';\nimport {maxPoolWithArgmaxConfig} from './kernels/MaxPoolWithArgmax';\nimport {nonMaxSuppressionV3Config} from './kernels/NonMaxSuppressionV3';\nimport {nonMaxSuppressionV4Config} from './kernels/NonMaxSuppressionV4';\nimport {nonMaxSuppressionV5Config} from './kernels/NonMaxSuppressionV5';\nimport {reshapeConfig} from './kernels/Reshape';\nimport {rotateWithOffsetConfig} from './kernels/RotateWithOffset';\nimport {sinConfig} from './kernels/Sin';\nimport {squareConfig} from './kernels/Square';\nimport {squaredDifferenceConfig} from './kernels/SquaredDifference';\nimport {tanConfig} from './kernels/Tan';\nimport {transposeConfig} from './kernels/Transpose';\nimport {uniqueConfig} from './kernels/Unique';\n\n// List all kernel configs here\nconst kernelConfigs: KernelConfig[] = [\n  atan2Config,\n  avgPoolConfig,\n  avgPoolBackpropConfig,\n  batchNormConfig,\n  cosConfig,\n  divConfig,\n  flipLeftRightConfig,\n  fromPixelsConfig,\n  identityConfig,\n  maxConfig,\n  maxPoolConfig,\n  maxPoolBackpropConfig,\n  maxPoolWithArgmaxConfig,\n  nonMaxSuppressionV3Config,\n  nonMaxSuppressionV4Config,\n  nonMaxSuppressionV5Config,\n  reshapeConfig,\n  rotateWithOffsetConfig,\n  sinConfig,\n  squareConfig,\n  squaredDifferenceConfig,\n  tanConfig,\n  transposeConfig,\n  uniqueConfig,\n];\n\nfor (const kernelConfig of kernelConfigs) {\n  registerKernel(kernelConfig);\n}\n", "/** @license See the LICENSE file. */\n\n// This code is auto-generated, do not modify this file!\nconst version = '2.6.0';\nexport {version};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nexport * from '@tensorflow/tfjs-core';\nexport * from '@tensorflow/tfjs-layers';\nexport * from '@tensorflow/tfjs-converter';\n\n// Export data api as tf.data\nimport * as data from '@tensorflow/tfjs-data';\nexport {data};\n\n// Import and register backends.\nimport '@tensorflow/tfjs-backend-cpu';\nimport '@tensorflow/tfjs-backend-webgl';\n\n// Import versions of all sub-packages.\nimport {version_core} from '@tensorflow/tfjs-core';\nimport {version_cpu} from '@tensorflow/tfjs-backend-cpu';\nimport {version_webgl} from '@tensorflow/tfjs-backend-webgl';\nimport {version_data} from '@tensorflow/tfjs-data';\nimport {version_layers} from '@tensorflow/tfjs-layers';\nimport {version_converter} from '@tensorflow/tfjs-converter';\nimport {version as version_union} from './version';\n\nexport const version = {\n  'tfjs-core': version_core,\n  'tfjs-backend-cpu': version_cpu,\n  'tfjs-backend-webgl': version_webgl,\n  'tfjs-data': version_data,\n  'tfjs-layers': version_layers,\n  'tfjs-converter': version_converter,\n  'tfjs': version_union\n};\n", "const tf = require('@tensorflow/tfjs');\n\nconst NUM_LANDMARKS = 6;\n\nfunction generateAnchors(inputSize) {\n  const spec = { strides: [inputSize / 16, inputSize / 8], anchors: [2, 6] };\n  const anchors = [];\n  for (let i = 0; i < spec.strides.length; i++) {\n    const stride = spec.strides[i];\n    const gridRows = Math.floor((inputSize + stride - 1) / stride);\n    const gridCols = Math.floor((inputSize + stride - 1) / stride);\n    const anchorsNum = spec.anchors[i];\n    for (let gridY = 0; gridY < gridRows; gridY++) {\n      const anchorY = stride * (gridY + 0.5);\n      for (let gridX = 0; gridX < gridCols; gridX++) {\n        const anchorX = stride * (gridX + 0.5);\n        for (let n = 0; n < anchorsNum; n++) {\n          anchors.push([anchorX, anchorY]);\n        }\n      }\n    }\n  }\n  return anchors;\n}\n\nconst disposeBox = (box) => {\n  box.startEndTensor.dispose();\n  box.startPoint.dispose();\n  box.endPoint.dispose();\n};\n\nconst createBox = (startEndTensor) => ({\n  startEndTensor,\n  startPoint: tf.slice(startEndTensor, [0, 0], [-1, 2]),\n  endPoint: tf.slice(startEndTensor, [0, 2], [-1, 2]),\n});\n\nconst scaleBox = (box, factors) => {\n  const starts = tf.mul(box.startPoint, factors);\n  const ends = tf.mul(box.endPoint, factors);\n  const newCoordinates = tf.concat2d([starts, ends], 1);\n  return createBox(newCoordinates);\n};\n\nfunction decodeBounds(boxOutputs, anchors, inputSize) {\n  const boxStarts = tf.slice(boxOutputs, [0, 1], [-1, 2]);\n  const centers = tf.add(boxStarts, anchors);\n  const boxSizes = tf.slice(boxOutputs, [0, 3], [-1, 2]);\n  const boxSizesNormalized = tf.div(boxSizes, inputSize);\n  const centersNormalized = tf.div(centers, inputSize);\n  const halfBoxSize = tf.div(boxSizesNormalized, 2);\n  const starts = tf.sub(centersNormalized, halfBoxSize);\n  const ends = tf.add(centersNormalized, halfBoxSize);\n  const startNormalized = tf.mul(starts, inputSize);\n  const endNormalized = tf.mul(ends, inputSize);\n  const concatAxis = 1;\n  return tf.concat2d([startNormalized, endNormalized], concatAxis);\n}\n\nfunction scaleBoxFromPrediction(face, scaleFactor) {\n  return tf.tidy(() => {\n    const box = face['box'] ? face['box'] : face;\n    return scaleBox(box, scaleFactor).startEndTensor.squeeze();\n  });\n}\n\nclass BlazeFaceModel {\n  constructor(model, config) {\n    this.blazeFaceModel = model;\n    this.width = config.detector.inputSize;\n    this.height = config.detector.inputSize;\n    this.maxFaces = config.detector.maxFaces;\n    this.anchorsData = generateAnchors(config.detector.inputSize);\n    this.anchors = tf.tensor2d(this.anchorsData);\n    this.inputSize = tf.tensor1d([this.width, this.height]);\n    this.iouThreshold = config.detector.iouThreshold;\n    this.scaleFaces = 0.8;\n    this.scoreThreshold = config.detector.scoreThreshold;\n  }\n\n  async getBoundingBoxes(inputImage) {\n    // sanity check on input\n    if ((!inputImage) || (inputImage.isDisposedInternal) || (inputImage.shape.length !== 4) || (inputImage.shape[1] < 1) || (inputImage.shape[2] < 1)) return null;\n    const [detectedOutputs, boxes, scores] = tf.tidy(() => {\n      const resizedImage = inputImage.resizeBilinear([this.width, this.height]);\n      const normalizedImage = tf.mul(tf.sub(resizedImage.div(255), 0.5), 2);\n      const batchedPrediction = this.blazeFaceModel.predict(normalizedImage);\n      let prediction;\n      // are we using tfhub or pinto converted model?\n      if (Array.isArray(batchedPrediction)) {\n        const sorted = batchedPrediction.sort((a, b) => a.size - b.size);\n        const concat384 = tf.concat([sorted[0], sorted[2]], 2); // dim: 384, 1 + 16\n        const concat512 = tf.concat([sorted[1], sorted[3]], 2); // dim: 512, 1 + 16\n        const concat = tf.concat([concat512, concat384], 1);\n        prediction = concat.squeeze(0);\n      } else {\n        prediction = batchedPrediction.squeeze(); // when using tfhub model\n      }\n      const decodedBounds = decodeBounds(prediction, this.anchors, this.inputSize);\n      const logits = tf.slice(prediction, [0, 0], [-1, 1]);\n      const scoresOut = tf.sigmoid(logits).squeeze();\n      return [prediction, decodedBounds, scoresOut];\n    });\n\n    const boxIndicesTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxFaces, this.iouThreshold, this.scoreThreshold);\n    const boxIndices = await boxIndicesTensor.array();\n    boxIndicesTensor.dispose();\n    let boundingBoxes = boxIndices.map((boxIndex) => tf.slice(boxes, [boxIndex, 0], [1, -1]));\n    boundingBoxes = await Promise.all(boundingBoxes.map(async (boundingBox) => {\n      const vals = await boundingBox.array();\n      boundingBox.dispose();\n      return vals;\n    }));\n    const annotatedBoxes = [];\n    for (let i = 0; i < boundingBoxes.length; i++) {\n      const boundingBox = boundingBoxes[i];\n      const annotatedBox = tf.tidy(() => {\n        const box = createBox(boundingBox);\n        const boxIndex = boxIndices[i];\n        const anchor = this.anchorsData[boxIndex];\n        const landmarks = tf\n          .slice(detectedOutputs, [boxIndex, NUM_LANDMARKS - 1], [1, -1])\n          .squeeze()\n          .reshape([NUM_LANDMARKS, -1]);\n        const probability = tf.slice(scores, [boxIndex], [1]);\n        return { box, landmarks, probability, anchor };\n      });\n      annotatedBoxes.push(annotatedBox);\n    }\n    boxes.dispose();\n    scores.dispose();\n    detectedOutputs.dispose();\n    return {\n      boxes: annotatedBoxes,\n      scaleFactor: [inputImage.shape[2] / this.width, inputImage.shape[1] / this.height],\n    };\n  }\n\n  async estimateFaces(input) {\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const { boxes, scaleFactor } = await this.getBoundingBoxes(image);\n    image.dispose();\n    return Promise.all(boxes.map(async (face) => {\n      const scaledBox = scaleBoxFromPrediction(face, scaleFactor);\n      const [landmarkData, boxData, probabilityData] = await Promise.all([face.landmarks, scaledBox, face.probability].map(async (d) => d.array()));\n      const anchor = face.anchor;\n      const [scaleFactorX, scaleFactorY] = scaleFactor;\n      const scaledLandmarks = landmarkData\n        .map((landmark) => ([\n          (landmark[0] + anchor[0]) * scaleFactorX,\n          (landmark[1] + anchor[1]) * scaleFactorY,\n        ]));\n      const normalizedFace = {\n        topLeft: boxData.slice(0, 2),\n        bottomRight: boxData.slice(2),\n        landmarks: scaledLandmarks,\n        probability: probabilityData,\n      };\n      disposeBox(face.box);\n      face.landmarks.dispose();\n      face.probability.dispose();\n      scaledBox.dispose();\n      return normalizedFace;\n    }));\n  }\n}\n\nasync function load(config) {\n  const blazeface = await tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') });\n  const model = new BlazeFaceModel(blazeface, config);\n  return model;\n}\n\nexports.load = load;\nexports.BlazeFaceModel = BlazeFaceModel;\nexports.disposeBox = disposeBox;\n", "exports.MESH_ANNOTATIONS = {\n  silhouette: [\n    10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288,\n    397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136,\n    172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109,\n  ],\n  lipsUpperOuter: [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291],\n  lipsLowerOuter: [146, 91, 181, 84, 17, 314, 405, 321, 375, 291],\n  lipsUpperInner: [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308],\n  lipsLowerInner: [78, 95, 88, 178, 87, 14, 317, 402, 318, 324, 308],\n  rightEyeUpper0: [246, 161, 160, 159, 158, 157, 173],\n  rightEyeLower0: [33, 7, 163, 144, 145, 153, 154, 155, 133],\n  rightEyeUpper1: [247, 30, 29, 27, 28, 56, 190],\n  rightEyeLower1: [130, 25, 110, 24, 23, 22, 26, 112, 243],\n  rightEyeUpper2: [113, 225, 224, 223, 222, 221, 189],\n  rightEyeLower2: [226, 31, 228, 229, 230, 231, 232, 233, 244],\n  rightEyeLower3: [143, 111, 117, 118, 119, 120, 121, 128, 245],\n  rightEyebrowUpper: [156, 70, 63, 105, 66, 107, 55, 193],\n  rightEyebrowLower: [35, 124, 46, 53, 52, 65],\n  rightEyeIris: [473, 474, 475, 476, 477],\n  leftEyeUpper0: [466, 388, 387, 386, 385, 384, 398],\n  leftEyeLower0: [263, 249, 390, 373, 374, 380, 381, 382, 362],\n  leftEyeUpper1: [467, 260, 259, 257, 258, 286, 414],\n  leftEyeLower1: [359, 255, 339, 254, 253, 252, 256, 341, 463],\n  leftEyeUpper2: [342, 445, 444, 443, 442, 441, 413],\n  leftEyeLower2: [446, 261, 448, 449, 450, 451, 452, 453, 464],\n  leftEyeLower3: [372, 340, 346, 347, 348, 349, 350, 357, 465],\n  leftEyebrowUpper: [383, 300, 293, 334, 296, 336, 285, 417],\n  leftEyebrowLower: [265, 353, 276, 283, 282, 295],\n  leftEyeIris: [468, 469, 470, 471, 472],\n  midwayBetweenEyes: [168],\n  noseTip: [1],\n  noseBottom: [2],\n  noseRightCorner: [98],\n  noseLeftCorner: [327],\n  rightCheek: [205],\n  leftCheek: [425],\n};\nexports.MESH_TO_IRIS_INDICES_MAP = [ // A mapping from facemesh model keypoints to iris model keypoints.\n  { key: 'EyeUpper0', indices: [9, 10, 11, 12, 13, 14, 15] },\n  { key: 'EyeUpper1', indices: [25, 26, 27, 28, 29, 30, 31] },\n  { key: 'EyeUpper2', indices: [41, 42, 43, 44, 45, 46, 47] },\n  { key: 'EyeLower0', indices: [0, 1, 2, 3, 4, 5, 6, 7, 8] },\n  { key: 'EyeLower1', indices: [16, 17, 18, 19, 20, 21, 22, 23, 24] },\n  { key: 'EyeLower2', indices: [32, 33, 34, 35, 36, 37, 38, 39, 40] },\n  { key: 'EyeLower3', indices: [54, 55, 56, 57, 58, 59, 60, 61, 62] },\n  { key: 'EyebrowUpper', indices: [63, 64, 65, 66, 67, 68, 69, 70] },\n  { key: 'EyebrowLower', indices: [48, 49, 50, 51, 52, 53] },\n];\n", "const tf = require('@tensorflow/tfjs');\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  return { startPoint, endPoint };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.enlargeBox = enlargeBox;\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, landmarks: box.landmarks };\n}\nexports.squarifyBox = squarifyBox;\n", "exports.IDENTITY_MATRIX = [[1, 0, 0], [0, 1, 0], [0, 0, 1]];\n/**\n * Normalizes the provided angle to the range -pi to pi.\n * @param angle The angle in radians to be normalized.\n */\nfunction normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n/**\n * Computes the angle of rotation between two anchor points.\n * @param point1 First anchor point\n * @param point2 Second anchor point\n */\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\nfunction radToDegrees(rad) {\n  return rad * 180 / Math.PI;\n}\nexports.radToDegrees = radToDegrees;\nfunction buildTranslationMatrix(x, y) {\n  return [[1, 0, x], [0, 1, y], [0, 0, 1]];\n}\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\nfunction xyDistanceBetweenPoints(a, b) {\n  return Math.sqrt(((a[0] - b[0]) ** 2) + ((a[1] - b[1]) ** 2));\n}\nexports.xyDistanceBetweenPoints = xyDistanceBetweenPoints;\n", "/* eslint-disable class-methods-use-this */\nconst tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nconst LANDMARKS_COUNT = 468;\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.25;\nconst MESH_MOUTH_INDEX = 13;\nconst MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [MESH_MOUTH_INDEX, keypoints.MESH_ANNOTATIONS['midwayBetweenEyes'][0]];\nconst BLAZEFACE_MOUTH_INDEX = 3;\nconst BLAZEFACE_NOSE_INDEX = 2;\nconst BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES = [BLAZEFACE_MOUTH_INDEX, BLAZEFACE_NOSE_INDEX];\nconst LEFT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['leftEyeLower0'];\nconst LEFT_EYE_BOUNDS = [LEFT_EYE_OUTLINE[0], LEFT_EYE_OUTLINE[LEFT_EYE_OUTLINE.length - 1]];\nconst RIGHT_EYE_OUTLINE = keypoints.MESH_ANNOTATIONS['rightEyeLower0'];\nconst RIGHT_EYE_BOUNDS = [RIGHT_EYE_OUTLINE[0], RIGHT_EYE_OUTLINE[RIGHT_EYE_OUTLINE.length - 1]];\nconst IRIS_UPPER_CENTER_INDEX = 3;\nconst IRIS_LOWER_CENTER_INDEX = 4;\nconst IRIS_IRIS_INDEX = 71;\nconst IRIS_NUM_COORDINATES = 76;\n\n// Replace the raw coordinates returned by facemesh with refined iris model coordinates. Update the z coordinate to be an average of the original and the new. This produces the best visual effect.\nfunction replaceRawCoordinates(rawCoords, newCoords, prefix, keys) {\n  for (let i = 0; i < keypoints.MESH_TO_IRIS_INDICES_MAP.length; i++) {\n    const { key, indices } = keypoints.MESH_TO_IRIS_INDICES_MAP[i];\n    const originalIndices = keypoints.MESH_ANNOTATIONS[`${prefix}${key}`];\n    const shouldReplaceAllKeys = keys == null;\n    if (shouldReplaceAllKeys || keys.includes(key)) {\n      for (let j = 0; j < indices.length; j++) {\n        const index = indices[j];\n        rawCoords[originalIndices[j]] = [\n          newCoords[index][0], newCoords[index][1],\n          (newCoords[index][2] + rawCoords[originalIndices[j]][2]) / 2,\n        ];\n      }\n    }\n  }\n}\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass Pipeline {\n  constructor(boundingBoxDetector, meshDetector, irisModel, config) {\n    // An array of facial bounding boxes.\n    this.regionsOfInterest = [];\n    this.runsWithoutFaceDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.irisModel = irisModel;\n    this.meshWidth = config.mesh.inputSize;\n    this.meshHeight = config.mesh.inputSize;\n    this.irisSize = config.iris.inputSize;\n    this.irisEnlarge = config.iris.enlargeFactor;\n  }\n\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize({ startPoint: box.startPoint, endPoint: box.endPoint });\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => ([\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]));\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => ([...util.rotatePoint(coord, coordsRotationMatrix), coord[2]]));\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint }), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => ([\n      coord[0] + originalBoxCenter[0],\n      coord[1] + originalBoxCenter[1], coord[2],\n    ]));\n  }\n\n  getLeftToRightEyeDepthDifference(rawCoords) {\n    const leftEyeZ = rawCoords[LEFT_EYE_BOUNDS[0]][2];\n    const rightEyeZ = rawCoords[RIGHT_EYE_BOUNDS[0]][2];\n    return leftEyeZ - rightEyeZ;\n  }\n\n  // Returns a box describing a cropped region around the eye fit for passing to the iris model.\n  getEyeBox(rawCoords, face, eyeInnerCornerIndex, eyeOuterCornerIndex, flip = false) {\n    const box = bounding.squarifyBox(bounding.enlargeBox(this.calculateLandmarksBoundingBox([rawCoords[eyeInnerCornerIndex], rawCoords[eyeOuterCornerIndex]]), this.irisEnlarge));\n    const boxSize = bounding.getBoxSize(box);\n    let crop = tf.image.cropAndResize(face, [[\n      box.startPoint[1] / this.meshHeight,\n      box.startPoint[0] / this.meshWidth, box.endPoint[1] / this.meshHeight,\n      box.endPoint[0] / this.meshWidth,\n    ]], [0], [this.irisSize, this.irisSize]);\n    if (flip) {\n      crop = tf.image.flipLeftRight(crop);\n    }\n    return { box, boxSize, crop };\n  }\n\n  // Given a cropped image of an eye, returns the coordinates of the contours surrounding the eye and the iris.\n  getEyeCoords(eyeData, eyeBox, eyeBoxSize, flip = false) {\n    const eyeRawCoords = [];\n    for (let i = 0; i < IRIS_NUM_COORDINATES; i++) {\n      const x = eyeData[i * 3];\n      const y = eyeData[i * 3 + 1];\n      const z = eyeData[i * 3 + 2];\n      eyeRawCoords.push([\n        (flip\n          ? (1 - (x / this.irisSize))\n          : (x / this.irisSize)) * eyeBoxSize[0] + eyeBox.startPoint[0],\n        (y / this.irisSize) * eyeBoxSize[1] + eyeBox.startPoint[1], z,\n      ]);\n    }\n    return { rawCoords: eyeRawCoords, iris: eyeRawCoords.slice(IRIS_IRIS_INDEX) };\n  }\n\n  // The z-coordinates returned for the iris are unreliable, so we take the z values from the surrounding keypoints.\n  getAdjustedIrisCoords(rawCoords, irisCoords, direction) {\n    const upperCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeUpper0`][IRIS_UPPER_CENTER_INDEX]][2];\n    const lowerCenterZ = rawCoords[keypoints.MESH_ANNOTATIONS[`${direction}EyeLower0`][IRIS_LOWER_CENTER_INDEX]][2];\n    const averageZ = (upperCenterZ + lowerCenterZ) / 2;\n    // Iris indices: 0: center | 1: right | 2: above | 3: left | 4: below\n    return irisCoords.map((coord, i) => {\n      let z = averageZ;\n      if (i === 2) {\n        z = upperCenterZ;\n      } else if (i === 4) {\n        z = lowerCenterZ;\n      }\n      return [coord[0], coord[1], z];\n    });\n  }\n\n  async predict(input, config) {\n    this.skipFrames = config.detector.skipFrames;\n    this.maxFaces = config.detector.maxFaces;\n    if (this.shouldUpdateRegionsOfInterest()) {\n      const { boxes, scaleFactor } = await this.boundingBoxDetector.getBoundingBoxes(input);\n      if (boxes.length === 0) {\n        this.regionsOfInterest = [];\n        return null;\n      }\n      const scaledBoxes = boxes.map((prediction) => {\n        const predictionBox = {\n          startPoint: prediction.box.startPoint.squeeze().arraySync(),\n          endPoint: prediction.box.endPoint.squeeze().arraySync(),\n        };\n        prediction.box.startPoint.dispose();\n        prediction.box.endPoint.dispose();\n        const scaledBox = bounding.scaleBoxCoordinates(predictionBox, scaleFactor);\n        const enlargedBox = bounding.enlargeBox(scaledBox);\n        const landmarks = prediction.landmarks.arraySync();\n        prediction.landmarks.dispose();\n        prediction.probability.dispose();\n        return { ...enlargedBox, landmarks };\n      });\n      this.updateRegionsOfInterest(scaledBoxes);\n      this.runsWithoutFaceDetector = 0;\n    } else {\n      this.runsWithoutFaceDetector++;\n    }\n    const results = tf.tidy(() => this.regionsOfInterest.map((box, i) => {\n      let angle = 0;\n      // The facial bounding box landmarks could come either from blazeface (if we are using a fresh box), or from the mesh model (if we are reusing an old box).\n      const boxLandmarksFromMeshModel = box.landmarks.length >= LANDMARKS_COUNT;\n      let [indexOfMouth, indexOfForehead] = MESH_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      if (boxLandmarksFromMeshModel === false) {\n        [indexOfMouth, indexOfForehead] = BLAZEFACE_KEYPOINTS_LINE_OF_SYMMETRY_INDICES;\n      }\n      angle = util.computeRotation(box.landmarks[indexOfMouth], box.landmarks[indexOfForehead]);\n      const faceCenter = bounding.getBoxCenter({ startPoint: box.startPoint, endPoint: box.endPoint });\n      const faceCenterNormalized = [faceCenter[0] / input.shape[2], faceCenter[1] / input.shape[1]];\n      let rotatedImage = input;\n      let rotationMatrix = util.IDENTITY_MATRIX;\n      if (angle !== 0) {\n        rotatedImage = tf.image.rotateWithOffset(input, angle, 0, faceCenterNormalized);\n        rotationMatrix = util.buildRotationMatrix(-angle, faceCenter);\n      }\n      const boxCPU = { startPoint: box.startPoint, endPoint: box.endPoint };\n      const face = bounding.cutBoxFromImageAndResize(boxCPU, rotatedImage, [this.meshHeight, this.meshWidth]).div(255);\n      // The first returned tensor represents facial contours, which are included in the coordinates.\n      const [, flag, coords] = this.meshDetector.predict(face);\n      const coordsReshaped = tf.reshape(coords, [-1, 3]);\n      let rawCoords = coordsReshaped.arraySync();\n      if (config.iris.enabled) {\n        const { box: leftEyeBox, boxSize: leftEyeBoxSize, crop: leftEyeCrop } = this.getEyeBox(rawCoords, face, LEFT_EYE_BOUNDS[0], LEFT_EYE_BOUNDS[1], true);\n        const { box: rightEyeBox, boxSize: rightEyeBoxSize, crop: rightEyeCrop } = this.getEyeBox(rawCoords, face, RIGHT_EYE_BOUNDS[0], RIGHT_EYE_BOUNDS[1]);\n        const eyePredictions = (this.irisModel.predict(tf.concat([leftEyeCrop, rightEyeCrop])));\n        const eyePredictionsData = eyePredictions.dataSync();\n        eyePredictions.dispose();\n        const leftEyeData = eyePredictionsData.slice(0, IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: leftEyeRawCoords, iris: leftIrisRawCoords } = this.getEyeCoords(leftEyeData, leftEyeBox, leftEyeBoxSize, true);\n        const rightEyeData = eyePredictionsData.slice(IRIS_NUM_COORDINATES * 3);\n        const { rawCoords: rightEyeRawCoords, iris: rightIrisRawCoords } = this.getEyeCoords(rightEyeData, rightEyeBox, rightEyeBoxSize);\n        const leftToRightEyeDepthDifference = this.getLeftToRightEyeDepthDifference(rawCoords);\n        if (Math.abs(leftToRightEyeDepthDifference) < 30) { // User is looking straight ahead.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left');\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right');\n          // If the user is looking to the left or to the right, the iris coordinates tend to diverge too much from the mesh coordinates for them to be merged. So we only update a single contour line above and below the eye.\n        } else if (leftToRightEyeDepthDifference < 1) { // User is looking towards the right.\n          replaceRawCoordinates(rawCoords, leftEyeRawCoords, 'left', ['EyeUpper0', 'EyeLower0']);\n        } else { // User is looking towards the left.\n          replaceRawCoordinates(rawCoords, rightEyeRawCoords, 'right', ['EyeUpper0', 'EyeLower0']);\n        }\n        const adjustedLeftIrisCoords = this.getAdjustedIrisCoords(rawCoords, leftIrisRawCoords, 'left');\n        const adjustedRightIrisCoords = this.getAdjustedIrisCoords(rawCoords, rightIrisRawCoords, 'right');\n        rawCoords = rawCoords.concat(adjustedLeftIrisCoords).concat(adjustedRightIrisCoords);\n      }\n      const transformedCoordsData = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      tf.dispose(rawCoords);\n      const landmarksBox = bounding.enlargeBox(this.calculateLandmarksBoundingBox(transformedCoordsData));\n      if (config.mesh.enabled) {\n        const transformedCoords = tf.tensor2d(transformedCoordsData);\n        this.regionsOfInterest[i] = { ...landmarksBox, landmarks: transformedCoords.arraySync() };\n        const prediction = {\n          coords: transformedCoords,\n          box: landmarksBox,\n          confidence: flag.squeeze(),\n          image: face,\n        };\n        return prediction;\n      }\n      const prediction = {\n        coords: null,\n        box: landmarksBox,\n        confidence: flag.squeeze(),\n        image: face,\n      };\n      return prediction;\n    }));\n    return results;\n  }\n\n  // Updates regions of interest if the intersection over union between the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(boxes) {\n    for (let i = 0; i < boxes.length; i++) {\n      const box = boxes[i];\n      const previousBox = this.regionsOfInterest[i];\n      let iou = 0;\n      if (previousBox && previousBox.startPoint) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      if (iou < UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD) {\n        this.regionsOfInterest[i] = box;\n      }\n    }\n    this.regionsOfInterest = this.regionsOfInterest.slice(0, boxes.length);\n  }\n\n  clearRegionOfInterest(index) {\n    if (this.regionsOfInterest[index] != null) {\n      this.regionsOfInterest = [\n        ...this.regionsOfInterest.slice(0, index),\n        ...this.regionsOfInterest.slice(index + 1),\n      ];\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    const roisCount = this.regionsOfInterest.length;\n    const noROIs = roisCount === 0;\n    if (this.maxFaces === 1 || noROIs) {\n      return noROIs;\n    }\n    return roisCount !== this.maxFaces && this.runsWithoutFaceDetector >= this.skipFrames;\n  }\n\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n}\nexports.Pipeline = Pipeline;\n", "exports.UV_COORDS = [\n  [0.499976992607117, 0.652534008026123],\n  [0.500025987625122, 0.547487020492554],\n  [0.499974012374878, 0.602371990680695],\n  [0.482113003730774, 0.471979022026062],\n  [0.500150978565216, 0.527155995368958],\n  [0.499909996986389, 0.498252987861633],\n  [0.499523013830185, 0.40106201171875],\n  [0.289712011814117, 0.380764007568359],\n  [0.499954998493195, 0.312398016452789],\n  [0.499987006187439, 0.269918978214264],\n  [0.500023007392883, 0.107050001621246],\n  [0.500023007392883, 0.666234016418457],\n  [0.5000159740448, 0.679224014282227],\n  [0.500023007392883, 0.692348003387451],\n  [0.499976992607117, 0.695277988910675],\n  [0.499976992607117, 0.70593398809433],\n  [0.499976992607117, 0.719385027885437],\n  [0.499976992607117, 0.737019002437592],\n  [0.499967992305756, 0.781370997428894],\n  [0.499816000461578, 0.562981009483337],\n  [0.473773002624512, 0.573909997940063],\n  [0.104906998574734, 0.254140973091125],\n  [0.365929991006851, 0.409575998783112],\n  [0.338757991790771, 0.41302502155304],\n  [0.311120003461838, 0.409460008144379],\n  [0.274657994508743, 0.389131009578705],\n  [0.393361985683441, 0.403706014156342],\n  [0.345234006643295, 0.344011008739471],\n  [0.370094001293182, 0.346076011657715],\n  [0.319321990013123, 0.347265005111694],\n  [0.297903001308441, 0.353591024875641],\n  [0.24779200553894, 0.410809993743896],\n  [0.396889001131058, 0.842755019664764],\n  [0.280097991228104, 0.375599980354309],\n  [0.106310002505779, 0.399955987930298],\n  [0.2099249958992, 0.391353011131287],\n  [0.355807989835739, 0.534406006336212],\n  [0.471751004457474, 0.65040397644043],\n  [0.474155008792877, 0.680191993713379],\n  [0.439785003662109, 0.657229006290436],\n  [0.414617002010345, 0.66654098033905],\n  [0.450374007225037, 0.680860996246338],\n  [0.428770989179611, 0.682690978050232],\n  [0.374971002340317, 0.727805018424988],\n  [0.486716985702515, 0.547628998756409],\n  [0.485300987958908, 0.527395009994507],\n  [0.257764995098114, 0.314490020275116],\n  [0.401223003864288, 0.455172002315521],\n  [0.429818987846375, 0.548614978790283],\n  [0.421351999044418, 0.533740997314453],\n  [0.276895999908447, 0.532056987285614],\n  [0.483370006084442, 0.499586999416351],\n  [0.33721199631691, 0.282882988452911],\n  [0.296391993761063, 0.293242990970612],\n  [0.169294998049736, 0.193813979625702],\n  [0.447580009698868, 0.302609980106354],\n  [0.392390012741089, 0.353887975215912],\n  [0.354490011930466, 0.696784019470215],\n  [0.067304998636246, 0.730105042457581],\n  [0.442739009857178, 0.572826027870178],\n  [0.457098007202148, 0.584792017936707],\n  [0.381974011659622, 0.694710969924927],\n  [0.392388999462128, 0.694203019142151],\n  [0.277076005935669, 0.271932005882263],\n  [0.422551989555359, 0.563233017921448],\n  [0.385919004678726, 0.281364023685455],\n  [0.383103013038635, 0.255840003490448],\n  [0.331431001424789, 0.119714021682739],\n  [0.229923993349075, 0.232002973556519],\n  [0.364500999450684, 0.189113974571228],\n  [0.229622006416321, 0.299540996551514],\n  [0.173287004232407, 0.278747975826263],\n  [0.472878992557526, 0.666198015213013],\n  [0.446828007698059, 0.668527007102966],\n  [0.422762006521225, 0.673889994621277],\n  [0.445307999849319, 0.580065965652466],\n  [0.388103008270264, 0.693961024284363],\n  [0.403039008378983, 0.706539988517761],\n  [0.403629004955292, 0.693953037261963],\n  [0.460041999816895, 0.557139039039612],\n  [0.431158006191254, 0.692366003990173],\n  [0.452181994915009, 0.692366003990173],\n  [0.475387006998062, 0.692366003990173],\n  [0.465828001499176, 0.779190003871918],\n  [0.472328990697861, 0.736225962638855],\n  [0.473087012767792, 0.717857003211975],\n  [0.473122000694275, 0.704625964164734],\n  [0.473033010959625, 0.695277988910675],\n  [0.427942007780075, 0.695277988910675],\n  [0.426479011774063, 0.703539967536926],\n  [0.423162013292313, 0.711845993995667],\n  [0.4183090031147, 0.720062971115112],\n  [0.390094995498657, 0.639572978019714],\n  [0.013953999616206, 0.560034036636353],\n  [0.499913990497589, 0.58014702796936],\n  [0.413199990987778, 0.69539999961853],\n  [0.409626007080078, 0.701822996139526],\n  [0.468080013990402, 0.601534962654114],\n  [0.422728985548019, 0.585985004901886],\n  [0.463079988956451, 0.593783974647522],\n  [0.37211999297142, 0.47341400384903],\n  [0.334562003612518, 0.496073007583618],\n  [0.411671012639999, 0.546965003013611],\n  [0.242175996303558, 0.14767599105835],\n  [0.290776997804642, 0.201445996761322],\n  [0.327338010072708, 0.256527006626129],\n  [0.399509996175766, 0.748921036720276],\n  [0.441727995872498, 0.261676013469696],\n  [0.429764986038208, 0.187834024429321],\n  [0.412198007106781, 0.108901023864746],\n  [0.288955003023148, 0.398952007293701],\n  [0.218936994671822, 0.435410976409912],\n  [0.41278201341629, 0.398970007896423],\n  [0.257135003805161, 0.355440020561218],\n  [0.427684992551804, 0.437960982322693],\n  [0.448339998722076, 0.536936044692993],\n  [0.178560003638268, 0.45755398273468],\n  [0.247308000922203, 0.457193970680237],\n  [0.286267012357712, 0.467674970626831],\n  [0.332827985286713, 0.460712015628815],\n  [0.368755996227264, 0.447206974029541],\n  [0.398963987827301, 0.432654976844788],\n  [0.476410001516342, 0.405806005001068],\n  [0.189241006970406, 0.523923993110657],\n  [0.228962004184723, 0.348950982093811],\n  [0.490725994110107, 0.562400996685028],\n  [0.404670000076294, 0.485132992267609],\n  [0.019469000399113, 0.401564002037048],\n  [0.426243007183075, 0.420431017875671],\n  [0.396993011236191, 0.548797011375427],\n  [0.266469985246658, 0.376977026462555],\n  [0.439121007919312, 0.51895797252655],\n  [0.032313998788595, 0.644356966018677],\n  [0.419054001569748, 0.387154996395111],\n  [0.462783008813858, 0.505746960639954],\n  [0.238978996872902, 0.779744982719421],\n  [0.198220998048782, 0.831938028335571],\n  [0.107550002634525, 0.540755033493042],\n  [0.183610007166862, 0.740257024765015],\n  [0.134409993886948, 0.333683013916016],\n  [0.385764002799988, 0.883153975009918],\n  [0.490967005491257, 0.579378008842468],\n  [0.382384985685349, 0.508572995662689],\n  [0.174399003386497, 0.397670984268188],\n  [0.318785011768341, 0.39623498916626],\n  [0.343364000320435, 0.400596976280212],\n  [0.396100014448166, 0.710216999053955],\n  [0.187885001301765, 0.588537991046906],\n  [0.430987000465393, 0.944064974784851],\n  [0.318993002176285, 0.898285031318665],\n  [0.266247987747192, 0.869701027870178],\n  [0.500023007392883, 0.190576016902924],\n  [0.499976992607117, 0.954452991485596],\n  [0.366169989109039, 0.398822009563446],\n  [0.393207013607025, 0.39553701877594],\n  [0.410373002290726, 0.391080021858215],\n  [0.194993004202843, 0.342101991176605],\n  [0.388664990663528, 0.362284004688263],\n  [0.365961998701096, 0.355970978736877],\n  [0.343364000320435, 0.355356991291046],\n  [0.318785011768341, 0.35834002494812],\n  [0.301414996385574, 0.363156020641327],\n  [0.058132998645306, 0.319076001644135],\n  [0.301414996385574, 0.387449026107788],\n  [0.499987989664078, 0.618434011936188],\n  [0.415838003158569, 0.624195992946625],\n  [0.445681989192963, 0.566076993942261],\n  [0.465844005346298, 0.620640993118286],\n  [0.49992299079895, 0.351523995399475],\n  [0.288718998432159, 0.819945991039276],\n  [0.335278987884521, 0.852819979190826],\n  [0.440512001514435, 0.902418971061707],\n  [0.128294005990028, 0.791940987110138],\n  [0.408771991729736, 0.373893976211548],\n  [0.455606997013092, 0.451801002025604],\n  [0.499877005815506, 0.908990025520325],\n  [0.375436991453171, 0.924192011356354],\n  [0.11421000212431, 0.615022003650665],\n  [0.448662012815475, 0.695277988910675],\n  [0.4480200111866, 0.704632043838501],\n  [0.447111994028091, 0.715808033943176],\n  [0.444831997156143, 0.730794012546539],\n  [0.430011987686157, 0.766808986663818],\n  [0.406787008047104, 0.685672998428345],\n  [0.400738000869751, 0.681069016456604],\n  [0.392399996519089, 0.677703022956848],\n  [0.367855995893478, 0.663918972015381],\n  [0.247923001646996, 0.601333022117615],\n  [0.452769994735718, 0.420849978923798],\n  [0.43639200925827, 0.359887003898621],\n  [0.416164010763168, 0.368713974952698],\n  [0.413385987281799, 0.692366003990173],\n  [0.228018000721931, 0.683571994304657],\n  [0.468268007040024, 0.352671027183533],\n  [0.411361992359161, 0.804327011108398],\n  [0.499989002943039, 0.469825029373169],\n  [0.479153990745544, 0.442654013633728],\n  [0.499974012374878, 0.439637005329132],\n  [0.432112008333206, 0.493588984012604],\n  [0.499886006116867, 0.866917014122009],\n  [0.49991300702095, 0.821729004383087],\n  [0.456548988819122, 0.819200992584229],\n  [0.344549000263214, 0.745438992977142],\n  [0.37890899181366, 0.574010014533997],\n  [0.374292999505997, 0.780184984207153],\n  [0.319687992334366, 0.570737957954407],\n  [0.357154995203018, 0.604269981384277],\n  [0.295284003019333, 0.621580958366394],\n  [0.447750002145767, 0.862477004528046],\n  [0.410986006259918, 0.508723020553589],\n  [0.31395098567009, 0.775308012962341],\n  [0.354128003120422, 0.812552988529205],\n  [0.324548006057739, 0.703992962837219],\n  [0.189096003770828, 0.646299958229065],\n  [0.279776990413666, 0.71465802192688],\n  [0.1338230073452, 0.682700991630554],\n  [0.336768001317978, 0.644733011722565],\n  [0.429883986711502, 0.466521978378296],\n  [0.455527991056442, 0.548622965812683],\n  [0.437114000320435, 0.558896005153656],\n  [0.467287987470627, 0.529924988746643],\n  [0.414712011814117, 0.335219979286194],\n  [0.37704598903656, 0.322777986526489],\n  [0.344107985496521, 0.320150971412659],\n  [0.312875986099243, 0.32233202457428],\n  [0.283526003360748, 0.333190023899078],\n  [0.241245999932289, 0.382785975933075],\n  [0.102986000478268, 0.468762993812561],\n  [0.267612010240555, 0.424560010433197],\n  [0.297879010438919, 0.433175981044769],\n  [0.333433985710144, 0.433878004550934],\n  [0.366427004337311, 0.426115989685059],\n  [0.396012008190155, 0.416696012020111],\n  [0.420121014118195, 0.41022801399231],\n  [0.007561000064015, 0.480777025222778],\n  [0.432949006557465, 0.569517970085144],\n  [0.458638995885849, 0.479089021682739],\n  [0.473466008901596, 0.545744001865387],\n  [0.476087987422943, 0.563830018043518],\n  [0.468472003936768, 0.555056989192963],\n  [0.433990985155106, 0.582361996173859],\n  [0.483518004417419, 0.562983989715576],\n  [0.482482999563217, 0.57784903049469],\n  [0.42645001411438, 0.389798998832703],\n  [0.438998997211456, 0.39649498462677],\n  [0.450067013502121, 0.400434017181396],\n  [0.289712011814117, 0.368252992630005],\n  [0.276670008897781, 0.363372981548309],\n  [0.517862021923065, 0.471948027610779],\n  [0.710287988185883, 0.380764007568359],\n  [0.526226997375488, 0.573909997940063],\n  [0.895093023777008, 0.254140973091125],\n  [0.634069979190826, 0.409575998783112],\n  [0.661242008209229, 0.41302502155304],\n  [0.688880026340485, 0.409460008144379],\n  [0.725341975688934, 0.389131009578705],\n  [0.606630027294159, 0.40370500087738],\n  [0.654766023159027, 0.344011008739471],\n  [0.629905998706818, 0.346076011657715],\n  [0.680678009986877, 0.347265005111694],\n  [0.702096998691559, 0.353591024875641],\n  [0.75221198797226, 0.410804986953735],\n  [0.602918028831482, 0.842862963676453],\n  [0.719901978969574, 0.375599980354309],\n  [0.893692970275879, 0.399959981441498],\n  [0.790081977844238, 0.391354024410248],\n  [0.643998026847839, 0.534487962722778],\n  [0.528249025344849, 0.65040397644043],\n  [0.525849997997284, 0.680191040039062],\n  [0.560214996337891, 0.657229006290436],\n  [0.585384011268616, 0.66654098033905],\n  [0.549625992774963, 0.680860996246338],\n  [0.57122802734375, 0.682691991329193],\n  [0.624852001667023, 0.72809898853302],\n  [0.513050019741058, 0.547281980514526],\n  [0.51509702205658, 0.527251958847046],\n  [0.742246985435486, 0.314507007598877],\n  [0.598631024360657, 0.454979002475739],\n  [0.570338010787964, 0.548575043678284],\n  [0.578631997108459, 0.533622980117798],\n  [0.723087012767792, 0.532054007053375],\n  [0.516445994377136, 0.499638974666595],\n  [0.662801027297974, 0.282917976379395],\n  [0.70362401008606, 0.293271005153656],\n  [0.830704987049103, 0.193813979625702],\n  [0.552385985851288, 0.302568018436432],\n  [0.607609987258911, 0.353887975215912],\n  [0.645429015159607, 0.696707010269165],\n  [0.932694971561432, 0.730105042457581],\n  [0.557260990142822, 0.572826027870178],\n  [0.542901992797852, 0.584792017936707],\n  [0.6180260181427, 0.694710969924927],\n  [0.607590973377228, 0.694203019142151],\n  [0.722943007946014, 0.271963000297546],\n  [0.577413976192474, 0.563166975975037],\n  [0.614082992076874, 0.281386971473694],\n  [0.616907000541687, 0.255886018276215],\n  [0.668509006500244, 0.119913995265961],\n  [0.770092010498047, 0.232020974159241],\n  [0.635536015033722, 0.189248979091644],\n  [0.77039098739624, 0.299556016921997],\n  [0.826722025871277, 0.278755009174347],\n  [0.527121007442474, 0.666198015213013],\n  [0.553171992301941, 0.668527007102966],\n  [0.577238023281097, 0.673889994621277],\n  [0.554691970348358, 0.580065965652466],\n  [0.611896991729736, 0.693961024284363],\n  [0.59696102142334, 0.706539988517761],\n  [0.596370995044708, 0.693953037261963],\n  [0.539958000183105, 0.557139039039612],\n  [0.568841993808746, 0.692366003990173],\n  [0.547818005084991, 0.692366003990173],\n  [0.52461302280426, 0.692366003990173],\n  [0.534089982509613, 0.779141008853912],\n  [0.527670979499817, 0.736225962638855],\n  [0.526912987232208, 0.717857003211975],\n  [0.526877999305725, 0.704625964164734],\n  [0.526966989040375, 0.695277988910675],\n  [0.572058022022247, 0.695277988910675],\n  [0.573521018028259, 0.703539967536926],\n  [0.57683801651001, 0.711845993995667],\n  [0.581691026687622, 0.720062971115112],\n  [0.609944999217987, 0.639909982681274],\n  [0.986046016216278, 0.560034036636353],\n  [0.5867999792099, 0.69539999961853],\n  [0.590372025966644, 0.701822996139526],\n  [0.531915009021759, 0.601536989212036],\n  [0.577268004417419, 0.585934996604919],\n  [0.536915004253387, 0.593786001205444],\n  [0.627542972564697, 0.473352015018463],\n  [0.665585994720459, 0.495950996875763],\n  [0.588353991508484, 0.546862006187439],\n  [0.757824003696442, 0.14767599105835],\n  [0.709249973297119, 0.201507985591888],\n  [0.672684013843536, 0.256581008434296],\n  [0.600408971309662, 0.74900496006012],\n  [0.55826598405838, 0.261672019958496],\n  [0.570303976535797, 0.187870979309082],\n  [0.588165998458862, 0.109044015407562],\n  [0.711045026779175, 0.398952007293701],\n  [0.781069993972778, 0.435405015945435],\n  [0.587247014045715, 0.398931980133057],\n  [0.742869973182678, 0.355445981025696],\n  [0.572156012058258, 0.437651991844177],\n  [0.55186802148819, 0.536570012569427],\n  [0.821442008018494, 0.457556009292603],\n  [0.752701997756958, 0.457181990146637],\n  [0.71375697851181, 0.467626988887787],\n  [0.66711300611496, 0.460672974586487],\n  [0.631101012229919, 0.447153985500336],\n  [0.6008620262146, 0.432473003864288],\n  [0.523481011390686, 0.405627012252808],\n  [0.810747981071472, 0.523926019668579],\n  [0.771045982837677, 0.348959028720856],\n  [0.509127020835876, 0.562718033790588],\n  [0.595292985439301, 0.485023975372314],\n  [0.980530977249146, 0.401564002037048],\n  [0.573499977588654, 0.420000016689301],\n  [0.602994978427887, 0.548687994480133],\n  [0.733529984951019, 0.376977026462555],\n  [0.560611009597778, 0.519016981124878],\n  [0.967685997486115, 0.644356966018677],\n  [0.580985009670258, 0.387160003185272],\n  [0.537728011608124, 0.505385041236877],\n  [0.760966002941132, 0.779752969741821],\n  [0.801778972148895, 0.831938028335571],\n  [0.892440974712372, 0.54076099395752],\n  [0.816350996494293, 0.740260004997253],\n  [0.865594983100891, 0.333687007427216],\n  [0.614073991775513, 0.883246004581451],\n  [0.508952975273132, 0.579437971115112],\n  [0.617941975593567, 0.508316040039062],\n  [0.825608015060425, 0.397674977779388],\n  [0.681214988231659, 0.39623498916626],\n  [0.656635999679565, 0.400596976280212],\n  [0.603900015354156, 0.710216999053955],\n  [0.81208598613739, 0.588539004325867],\n  [0.56801301240921, 0.944564998149872],\n  [0.681007981300354, 0.898285031318665],\n  [0.733752012252808, 0.869701027870178],\n  [0.633830010890961, 0.398822009563446],\n  [0.606792986392975, 0.39553701877594],\n  [0.589659988880157, 0.391062021255493],\n  [0.805015981197357, 0.342108011245728],\n  [0.611334979534149, 0.362284004688263],\n  [0.634037971496582, 0.355970978736877],\n  [0.656635999679565, 0.355356991291046],\n  [0.681214988231659, 0.35834002494812],\n  [0.698584973812103, 0.363156020641327],\n  [0.941866993904114, 0.319076001644135],\n  [0.698584973812103, 0.387449026107788],\n  [0.584177017211914, 0.624107003211975],\n  [0.554318010807037, 0.566076993942261],\n  [0.534153997898102, 0.62064003944397],\n  [0.711217999458313, 0.819975018501282],\n  [0.664629995822906, 0.852871000766754],\n  [0.559099972248077, 0.902631998062134],\n  [0.871706008911133, 0.791940987110138],\n  [0.591234028339386, 0.373893976211548],\n  [0.544341027736664, 0.451583981513977],\n  [0.624562978744507, 0.924192011356354],\n  [0.88577002286911, 0.615028977394104],\n  [0.551338016986847, 0.695277988910675],\n  [0.551980018615723, 0.704632043838501],\n  [0.552887976169586, 0.715808033943176],\n  [0.555167973041534, 0.730794012546539],\n  [0.569944024085999, 0.767035007476807],\n  [0.593203008174896, 0.685675978660583],\n  [0.599261999130249, 0.681069016456604],\n  [0.607599973678589, 0.677703022956848],\n  [0.631937980651855, 0.663500010967255],\n  [0.752032995223999, 0.601315021514893],\n  [0.547226011753082, 0.420395016670227],\n  [0.563543975353241, 0.359827995300293],\n  [0.583841025829315, 0.368713974952698],\n  [0.586614012718201, 0.692366003990173],\n  [0.771915018558502, 0.683578014373779],\n  [0.531597018241882, 0.352482974529266],\n  [0.588370978832245, 0.804440975189209],\n  [0.52079701423645, 0.442565023899078],\n  [0.567984998226166, 0.493479013442993],\n  [0.543282985687256, 0.819254994392395],\n  [0.655317008495331, 0.745514988899231],\n  [0.621008992195129, 0.574018001556396],\n  [0.625559985637665, 0.78031200170517],\n  [0.680198013782501, 0.570719003677368],\n  [0.64276397228241, 0.604337990283966],\n  [0.704662978649139, 0.621529996395111],\n  [0.552012026309967, 0.862591981887817],\n  [0.589071989059448, 0.508637011051178],\n  [0.685944974422455, 0.775357007980347],\n  [0.645735025405884, 0.812640011310577],\n  [0.675342977046967, 0.703978002071381],\n  [0.810858011245728, 0.646304965019226],\n  [0.72012197971344, 0.714666962623596],\n  [0.866151988506317, 0.682704985141754],\n  [0.663187026977539, 0.644596993923187],\n  [0.570082008838654, 0.466325998306274],\n  [0.544561982154846, 0.548375964164734],\n  [0.562758982181549, 0.558784961700439],\n  [0.531987011432648, 0.530140042304993],\n  [0.585271000862122, 0.335177004337311],\n  [0.622952997684479, 0.32277899980545],\n  [0.655896008014679, 0.320163011550903],\n  [0.687132000923157, 0.322345972061157],\n  [0.716481983661652, 0.333200991153717],\n  [0.758756995201111, 0.382786989212036],\n  [0.897013008594513, 0.468769013881683],\n  [0.732392013072968, 0.424547016620636],\n  [0.70211398601532, 0.433162987232208],\n  [0.66652500629425, 0.433866024017334],\n  [0.633504986763, 0.426087975502014],\n  [0.603875994682312, 0.416586995124817],\n  [0.579657971858978, 0.409945011138916],\n  [0.992439985275269, 0.480777025222778],\n  [0.567192018032074, 0.569419980049133],\n  [0.54136598110199, 0.478899002075195],\n  [0.526564002037048, 0.546118021011353],\n  [0.523913025856018, 0.563830018043518],\n  [0.531529009342194, 0.555056989192963],\n  [0.566035985946655, 0.582329034805298],\n  [0.51631098985672, 0.563053965568542],\n  [0.5174720287323, 0.577877044677734],\n  [0.573594987392426, 0.389806985855103],\n  [0.560697972774506, 0.395331978797913],\n  [0.549755990505219, 0.399751007556915],\n  [0.710287988185883, 0.368252992630005],\n  [0.723330020904541, 0.363372981548309],\n];\n", "export default [\n  127, 34, 139, 11, 0, 37, 232, 231, 120, 72, 37, 39, 128, 121, 47, 232, 121,\n  128, 104, 69, 67, 175, 171, 148, 157, 154, 155, 118, 50, 101, 73, 39, 40, 9,\n  151, 108, 48, 115, 131, 194, 204, 211, 74, 40, 185, 80, 42, 183, 40, 92,\n  186, 230, 229, 118, 202, 212, 214, 83, 18, 17, 76, 61, 146, 160, 29, 30, 56,\n  157, 173, 106, 204, 194, 135, 214, 192, 203, 165, 98, 21, 71, 68, 51, 45, 4,\n  144, 24, 23, 77, 146, 91, 205, 50, 187, 201, 200, 18, 91, 106, 182, 90, 91,\n  181, 85, 84, 17, 206, 203, 36, 148, 171, 140, 92, 40, 39, 193, 189, 244,\n  159, 158, 28, 247, 246, 161, 236, 3, 196, 54, 68, 104, 193, 168, 8, 117,\n  228, 31, 189, 193, 55, 98, 97, 99, 126, 47, 100, 166, 79, 218, 155, 154, 26,\n  209, 49, 131, 135, 136, 150, 47, 126, 217, 223, 52, 53, 45, 51, 134, 211,\n  170, 140, 67, 69, 108, 43, 106, 91, 230, 119, 120, 226, 130, 247, 63, 53,\n  52, 238, 20, 242, 46, 70, 156, 78, 62, 96, 46, 53, 63, 143, 34, 227, 173,\n  155, 133, 123, 117, 111, 44, 125, 19, 236, 134, 51, 216, 206, 205, 154, 153,\n  22, 39, 37, 167, 200, 201, 208, 36, 142, 100, 57, 212, 202, 20, 60, 99, 28,\n  158, 157, 35, 226, 113, 160, 159, 27, 204, 202, 210, 113, 225, 46, 43, 202,\n  204, 62, 76, 77, 137, 123, 116, 41, 38, 72, 203, 129, 142, 64, 98, 240, 49,\n  102, 64, 41, 73, 74, 212, 216, 207, 42, 74, 184, 169, 170, 211, 170, 149,\n  176, 105, 66, 69, 122, 6, 168, 123, 147, 187, 96, 77, 90, 65, 55, 107, 89,\n  90, 180, 101, 100, 120, 63, 105, 104, 93, 137, 227, 15, 86, 85, 129, 102,\n  49, 14, 87, 86, 55, 8, 9, 100, 47, 121, 145, 23, 22, 88, 89, 179, 6, 122,\n  196, 88, 95, 96, 138, 172, 136, 215, 58, 172, 115, 48, 219, 42, 80, 81, 195,\n  3, 51, 43, 146, 61, 171, 175, 199, 81, 82, 38, 53, 46, 225, 144, 163, 110,\n  246, 33, 7, 52, 65, 66, 229, 228, 117, 34, 127, 234, 107, 108, 69, 109, 108,\n  151, 48, 64, 235, 62, 78, 191, 129, 209, 126, 111, 35, 143, 163, 161, 246,\n  117, 123, 50, 222, 65, 52, 19, 125, 141, 221, 55, 65, 3, 195, 197, 25, 7,\n  33, 220, 237, 44, 70, 71, 139, 122, 193, 245, 247, 130, 33, 71, 21, 162,\n  153, 158, 159, 170, 169, 150, 188, 174, 196, 216, 186, 92, 144, 160, 161, 2,\n  97, 167, 141, 125, 241, 164, 167, 37, 72, 38, 12, 145, 159, 160, 38, 82, 13,\n  63, 68, 71, 226, 35, 111, 158, 153, 154, 101, 50, 205, 206, 92, 165, 209,\n  198, 217, 165, 167, 97, 220, 115, 218, 133, 112, 243, 239, 238, 241, 214,\n  135, 169, 190, 173, 133, 171, 208, 32, 125, 44, 237, 86, 87, 178, 85, 86,\n  179, 84, 85, 180, 83, 84, 181, 201, 83, 182, 137, 93, 132, 76, 62, 183, 61,\n  76, 184, 57, 61, 185, 212, 57, 186, 214, 207, 187, 34, 143, 156, 79, 239,\n  237, 123, 137, 177, 44, 1, 4, 201, 194, 32, 64, 102, 129, 213, 215, 138, 59,\n  166, 219, 242, 99, 97, 2, 94, 141, 75, 59, 235, 24, 110, 228, 25, 130, 226,\n  23, 24, 229, 22, 23, 230, 26, 22, 231, 112, 26, 232, 189, 190, 243, 221, 56,\n  190, 28, 56, 221, 27, 28, 222, 29, 27, 223, 30, 29, 224, 247, 30, 225, 238,\n  79, 20, 166, 59, 75, 60, 75, 240, 147, 177, 215, 20, 79, 166, 187, 147, 213,\n  112, 233, 244, 233, 128, 245, 128, 114, 188, 114, 217, 174, 131, 115, 220,\n  217, 198, 236, 198, 131, 134, 177, 132, 58, 143, 35, 124, 110, 163, 7, 228,\n  110, 25, 356, 389, 368, 11, 302, 267, 452, 350, 349, 302, 303, 269, 357,\n  343, 277, 452, 453, 357, 333, 332, 297, 175, 152, 377, 384, 398, 382, 347,\n  348, 330, 303, 304, 270, 9, 336, 337, 278, 279, 360, 418, 262, 431, 304,\n  408, 409, 310, 415, 407, 270, 409, 410, 450, 348, 347, 422, 430, 434, 313,\n  314, 17, 306, 307, 375, 387, 388, 260, 286, 414, 398, 335, 406, 418, 364,\n  367, 416, 423, 358, 327, 251, 284, 298, 281, 5, 4, 373, 374, 253, 307, 320,\n  321, 425, 427, 411, 421, 313, 18, 321, 405, 406, 320, 404, 405, 315, 16, 17,\n  426, 425, 266, 377, 400, 369, 322, 391, 269, 417, 465, 464, 386, 257, 258,\n  466, 260, 388, 456, 399, 419, 284, 332, 333, 417, 285, 8, 346, 340, 261,\n  413, 441, 285, 327, 460, 328, 355, 371, 329, 392, 439, 438, 382, 341, 256,\n  429, 420, 360, 364, 394, 379, 277, 343, 437, 443, 444, 283, 275, 440, 363,\n  431, 262, 369, 297, 338, 337, 273, 375, 321, 450, 451, 349, 446, 342, 467,\n  293, 334, 282, 458, 461, 462, 276, 353, 383, 308, 324, 325, 276, 300, 293,\n  372, 345, 447, 382, 398, 362, 352, 345, 340, 274, 1, 19, 456, 248, 281, 436,\n  427, 425, 381, 256, 252, 269, 391, 393, 200, 199, 428, 266, 330, 329, 287,\n  273, 422, 250, 462, 328, 258, 286, 384, 265, 353, 342, 387, 259, 257, 424,\n  431, 430, 342, 353, 276, 273, 335, 424, 292, 325, 307, 366, 447, 345, 271,\n  303, 302, 423, 266, 371, 294, 455, 460, 279, 278, 294, 271, 272, 304, 432,\n  434, 427, 272, 407, 408, 394, 430, 431, 395, 369, 400, 334, 333, 299, 351,\n  417, 168, 352, 280, 411, 325, 319, 320, 295, 296, 336, 319, 403, 404, 330,\n  348, 349, 293, 298, 333, 323, 454, 447, 15, 16, 315, 358, 429, 279, 14, 15,\n  316, 285, 336, 9, 329, 349, 350, 374, 380, 252, 318, 402, 403, 6, 197, 419,\n  318, 319, 325, 367, 364, 365, 435, 367, 397, 344, 438, 439, 272, 271, 311,\n  195, 5, 281, 273, 287, 291, 396, 428, 199, 311, 271, 268, 283, 444, 445,\n  373, 254, 339, 263, 466, 249, 282, 334, 296, 449, 347, 346, 264, 447, 454,\n  336, 296, 299, 338, 10, 151, 278, 439, 455, 292, 407, 415, 358, 371, 355,\n  340, 345, 372, 390, 249, 466, 346, 347, 280, 442, 443, 282, 19, 94, 370,\n  441, 442, 295, 248, 419, 197, 263, 255, 359, 440, 275, 274, 300, 383, 368,\n  351, 412, 465, 263, 467, 466, 301, 368, 389, 380, 374, 386, 395, 378, 379,\n  412, 351, 419, 436, 426, 322, 373, 390, 388, 2, 164, 393, 370, 462, 461,\n  164, 0, 267, 302, 11, 12, 374, 373, 387, 268, 12, 13, 293, 300, 301, 446,\n  261, 340, 385, 384, 381, 330, 266, 425, 426, 423, 391, 429, 355, 437, 391,\n  327, 326, 440, 457, 438, 341, 382, 362, 459, 457, 461, 434, 430, 394, 414,\n  463, 362, 396, 369, 262, 354, 461, 457, 316, 403, 402, 315, 404, 403, 314,\n  405, 404, 313, 406, 405, 421, 418, 406, 366, 401, 361, 306, 408, 407, 291,\n  409, 408, 287, 410, 409, 432, 436, 410, 434, 416, 411, 264, 368, 383, 309,\n  438, 457, 352, 376, 401, 274, 275, 4, 421, 428, 262, 294, 327, 358, 433,\n  416, 367, 289, 455, 439, 462, 370, 326, 2, 326, 370, 305, 460, 455, 254,\n  449, 448, 255, 261, 446, 253, 450, 449, 252, 451, 450, 256, 452, 451, 341,\n  453, 452, 413, 464, 463, 441, 413, 414, 258, 442, 441, 257, 443, 442, 259,\n  444, 443, 260, 445, 444, 467, 342, 445, 459, 458, 250, 289, 392, 290, 290,\n  328, 460, 376, 433, 435, 250, 290, 392, 411, 416, 433, 341, 463, 464, 453,\n  464, 465, 357, 465, 412, 343, 412, 399, 360, 363, 440, 437, 399, 456, 420,\n  456, 363, 401, 435, 288, 372, 383, 353, 339, 255, 249, 448, 261, 255, 133,\n  243, 190, 133, 155, 112, 33, 246, 247, 33, 130, 25, 398, 384, 286, 362, 398,\n  414, 362, 463, 341, 263, 359, 467, 263, 249, 255, 466, 467, 260, 75, 60,\n  166, 238, 239, 79, 162, 127, 139, 72, 11, 37, 121, 232, 120, 73, 72, 39,\n  114, 128, 47, 233, 232, 128, 103, 104, 67, 152, 175, 148, 173, 157, 155,\n  119, 118, 101, 74, 73, 40, 107, 9, 108, 49, 48, 131, 32, 194, 211, 184, 74,\n  185, 191, 80, 183, 185, 40, 186, 119, 230, 118, 210, 202, 214, 84, 83, 17,\n  77, 76, 146, 161, 160, 30, 190, 56, 173, 182, 106, 194, 138, 135, 192, 129,\n  203, 98, 54, 21, 68, 5, 51, 4, 145, 144, 23, 90, 77, 91, 207, 205, 187, 83,\n  201, 18, 181, 91, 182, 180, 90, 181, 16, 85, 17, 205, 206, 36, 176, 148,\n  140, 165, 92, 39, 245, 193, 244, 27, 159, 28, 30, 247, 161, 174, 236, 196,\n  103, 54, 104, 55, 193, 8, 111, 117, 31, 221, 189, 55, 240, 98, 99, 142, 126,\n  100, 219, 166, 218, 112, 155, 26, 198, 209, 131, 169, 135, 150, 114, 47,\n  217, 224, 223, 53, 220, 45, 134, 32, 211, 140, 109, 67, 108, 146, 43, 91,\n  231, 230, 120, 113, 226, 247, 105, 63, 52, 241, 238, 242, 124, 46, 156, 95,\n  78, 96, 70, 46, 63, 116, 143, 227, 116, 123, 111, 1, 44, 19, 3, 236, 51,\n  207, 216, 205, 26, 154, 22, 165, 39, 167, 199, 200, 208, 101, 36, 100, 43,\n  57, 202, 242, 20, 99, 56, 28, 157, 124, 35, 113, 29, 160, 27, 211, 204, 210,\n  124, 113, 46, 106, 43, 204, 96, 62, 77, 227, 137, 116, 73, 41, 72, 36, 203,\n  142, 235, 64, 240, 48, 49, 64, 42, 41, 74, 214, 212, 207, 183, 42, 184, 210,\n  169, 211, 140, 170, 176, 104, 105, 69, 193, 122, 168, 50, 123, 187, 89, 96,\n  90, 66, 65, 107, 179, 89, 180, 119, 101, 120, 68, 63, 104, 234, 93, 227, 16,\n  15, 85, 209, 129, 49, 15, 14, 86, 107, 55, 9, 120, 100, 121, 153, 145, 22,\n  178, 88, 179, 197, 6, 196, 89, 88, 96, 135, 138, 136, 138, 215, 172, 218,\n  115, 219, 41, 42, 81, 5, 195, 51, 57, 43, 61, 208, 171, 199, 41, 81, 38,\n  224, 53, 225, 24, 144, 110, 105, 52, 66, 118, 229, 117, 227, 34, 234, 66,\n  107, 69, 10, 109, 151, 219, 48, 235, 183, 62, 191, 142, 129, 126, 116, 111,\n  143, 7, 163, 246, 118, 117, 50, 223, 222, 52, 94, 19, 141, 222, 221, 65,\n  196, 3, 197, 45, 220, 44, 156, 70, 139, 188, 122, 245, 139, 71, 162, 145,\n  153, 159, 149, 170, 150, 122, 188, 196, 206, 216, 92, 163, 144, 161, 164, 2,\n  167, 242, 141, 241, 0, 164, 37, 11, 72, 12, 144, 145, 160, 12, 38, 13, 70,\n  63, 71, 31, 226, 111, 157, 158, 154, 36, 101, 205, 203, 206, 165, 126, 209,\n  217, 98, 165, 97, 237, 220, 218, 237, 239, 241, 210, 214, 169, 140, 171, 32,\n  241, 125, 237, 179, 86, 178, 180, 85, 179, 181, 84, 180, 182, 83, 181, 194,\n  201, 182, 177, 137, 132, 184, 76, 183, 185, 61, 184, 186, 57, 185, 216, 212,\n  186, 192, 214, 187, 139, 34, 156, 218, 79, 237, 147, 123, 177, 45, 44, 4,\n  208, 201, 32, 98, 64, 129, 192, 213, 138, 235, 59, 219, 141, 242, 97, 97, 2,\n  141, 240, 75, 235, 229, 24, 228, 31, 25, 226, 230, 23, 229, 231, 22, 230,\n  232, 26, 231, 233, 112, 232, 244, 189, 243, 189, 221, 190, 222, 28, 221,\n  223, 27, 222, 224, 29, 223, 225, 30, 224, 113, 247, 225, 99, 60, 240, 213,\n  147, 215, 60, 20, 166, 192, 187, 213, 243, 112, 244, 244, 233, 245, 245,\n  128, 188, 188, 114, 174, 134, 131, 220, 174, 217, 236, 236, 198, 134, 215,\n  177, 58, 156, 143, 124, 25, 110, 7, 31, 228, 25, 264, 356, 368, 0, 11, 267,\n  451, 452, 349, 267, 302, 269, 350, 357, 277, 350, 452, 357, 299, 333, 297,\n  396, 175, 377, 381, 384, 382, 280, 347, 330, 269, 303, 270, 151, 9, 337,\n  344, 278, 360, 424, 418, 431, 270, 304, 409, 272, 310, 407, 322, 270, 410,\n  449, 450, 347, 432, 422, 434, 18, 313, 17, 291, 306, 375, 259, 387, 260,\n  424, 335, 418, 434, 364, 416, 391, 423, 327, 301, 251, 298, 275, 281, 4,\n  254, 373, 253, 375, 307, 321, 280, 425, 411, 200, 421, 18, 335, 321, 406,\n  321, 320, 405, 314, 315, 17, 423, 426, 266, 396, 377, 369, 270, 322, 269,\n  413, 417, 464, 385, 386, 258, 248, 456, 419, 298, 284, 333, 168, 417, 8,\n  448, 346, 261, 417, 413, 285, 326, 327, 328, 277, 355, 329, 309, 392, 438,\n  381, 382, 256, 279, 429, 360, 365, 364, 379, 355, 277, 437, 282, 443, 283,\n  281, 275, 363, 395, 431, 369, 299, 297, 337, 335, 273, 321, 348, 450, 349,\n  359, 446, 467, 283, 293, 282, 250, 458, 462, 300, 276, 383, 292, 308, 325,\n  283, 276, 293, 264, 372, 447, 346, 352, 340, 354, 274, 19, 363, 456, 281,\n  426, 436, 425, 380, 381, 252, 267, 269, 393, 421, 200, 428, 371, 266, 329,\n  432, 287, 422, 290, 250, 328, 385, 258, 384, 446, 265, 342, 386, 387, 257,\n  422, 424, 430, 445, 342, 276, 422, 273, 424, 306, 292, 307, 352, 366, 345,\n  268, 271, 302, 358, 423, 371, 327, 294, 460, 331, 279, 294, 303, 271, 304,\n  436, 432, 427, 304, 272, 408, 395, 394, 431, 378, 395, 400, 296, 334, 299,\n  6, 351, 168, 376, 352, 411, 307, 325, 320, 285, 295, 336, 320, 319, 404,\n  329, 330, 349, 334, 293, 333, 366, 323, 447, 316, 15, 315, 331, 358, 279,\n  317, 14, 316, 8, 285, 9, 277, 329, 350, 253, 374, 252, 319, 318, 403, 351,\n  6, 419, 324, 318, 325, 397, 367, 365, 288, 435, 397, 278, 344, 439, 310,\n  272, 311, 248, 195, 281, 375, 273, 291, 175, 396, 199, 312, 311, 268, 276,\n  283, 445, 390, 373, 339, 295, 282, 296, 448, 449, 346, 356, 264, 454, 337,\n  336, 299, 337, 338, 151, 294, 278, 455, 308, 292, 415, 429, 358, 355, 265,\n  340, 372, 388, 390, 466, 352, 346, 280, 295, 442, 282, 354, 19, 370, 285,\n  441, 295, 195, 248, 197, 457, 440, 274, 301, 300, 368, 417, 351, 465, 251,\n  301, 389, 385, 380, 386, 394, 395, 379, 399, 412, 419, 410, 436, 322, 387,\n  373, 388, 326, 2, 393, 354, 370, 461, 393, 164, 267, 268, 302, 12, 386, 374,\n  387, 312, 268, 13, 298, 293, 301, 265, 446, 340, 380, 385, 381, 280, 330,\n  425, 322, 426, 391, 420, 429, 437, 393, 391, 326, 344, 440, 438, 458, 459,\n  461, 364, 434, 394, 428, 396, 262, 274, 354, 457, 317, 316, 402, 316, 315,\n  403, 315, 314, 404, 314, 313, 405, 313, 421, 406, 323, 366, 361, 292, 306,\n  407, 306, 291, 408, 291, 287, 409, 287, 432, 410, 427, 434, 411, 372, 264,\n  383, 459, 309, 457, 366, 352, 401, 1, 274, 4, 418, 421, 262, 331, 294, 358,\n  435, 433, 367, 392, 289, 439, 328, 462, 326, 94, 2, 370, 289, 305, 455, 339,\n  254, 448, 359, 255, 446, 254, 253, 449, 253, 252, 450, 252, 256, 451, 256,\n  341, 452, 414, 413, 463, 286, 441, 414, 286, 258, 441, 258, 257, 442, 257,\n  259, 443, 259, 260, 444, 260, 467, 445, 309, 459, 250, 305, 289, 290, 305,\n  290, 460, 401, 376, 435, 309, 250, 392, 376, 411, 433, 453, 341, 464, 357,\n  453, 465, 343, 357, 412, 437, 343, 399, 344, 360, 440, 420, 437, 456, 360,\n  420, 363, 361, 401, 288, 265, 372, 353, 390, 339, 249, 339, 448, 255];\n", "const tf = require('@tensorflow/tfjs');\nconst blazeface = require('./blazeface');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\nconst uv_coords = require('./uvcoords');\nconst triangulation = require('./triangulation').default;\n\nclass MediaPipeFaceMesh {\n  constructor(blazeFace, blazeMeshModel, irisModel, config) {\n    this.pipeline = new pipe.Pipeline(blazeFace, blazeMeshModel, irisModel, config);\n    if (config) this.config = config;\n  }\n\n  async estimateFaces(input, config) {\n    if (config) this.config = config;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) input = tf.browser.fromPixels(input);\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.predict(image, config);\n    tf.dispose(image);\n    const results = [];\n    for (const prediction of (predictions || [])) {\n      const confidence = prediction.confidence.arraySync();\n      if (confidence >= this.config.detector.minConfidence) {\n        const mesh = prediction.coords ? prediction.coords.arraySync() : null;\n        const annotations = {};\n        if (mesh && mesh.length > 0) {\n          for (const key in keypoints.MESH_ANNOTATIONS) {\n            if (this.config.iris.enabled || key.includes('Iris') === false) {\n              annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => mesh[index]);\n            }\n          }\n        }\n        results.push({\n          confidence: confidence || 0,\n          box: prediction.box ? [prediction.box.startPoint[0], prediction.box.startPoint[1], prediction.box.endPoint[0] - prediction.box.startPoint[0], prediction.box.endPoint[1] - prediction.box.startPoint[1]] : 0,\n          mesh,\n          annotations,\n          image: prediction.image ? tf.clone(prediction.image) : null,\n        });\n      }\n      prediction.confidence.dispose();\n      prediction.image.dispose();\n    }\n    return results;\n  }\n}\n\nasync function load(config) {\n  const models = await Promise.all([\n    blazeface.load(config),\n    tf.loadGraphModel(config.mesh.modelPath, { fromTFHub: config.mesh.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.iris.modelPath, { fromTFHub: config.iris.modelPath.includes('tfhub.dev') }),\n  ]);\n  const faceMesh = new MediaPipeFaceMesh(models[0], models[1], models[2], config);\n  return faceMesh;\n}\n\nexports.load = load;\nexports.MediaPipeFaceMesh = MediaPipeFaceMesh;\nexports.uv_coords = uv_coords;\nexports.triangulation = triangulation;\n", "const tf = require('@tensorflow/tfjs');\n\nconst models = {};\nlet last = { age: 0, gender: '' };\nlet frame = 0;\n\nasync function getImage(image, size) {\n  const buffer = tf.browser.fromPixels(image);\n  const resize = tf.image.resizeBilinear(buffer, [size, size]);\n  const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n  return expand;\n}\n\nasync function loadAge(config) {\n  if (!models.age) models.age = await tf.loadGraphModel(config.face.age.modelPath);\n  return models.age;\n}\n\nasync function loadGender(config) {\n  if (!models.gender) models.gender = await tf.loadGraphModel(config.face.gender.modelPath);\n  return models.gender;\n}\n\nasync function predict(image, config) {\n  if (frame > config.face.age.skipFrames) {\n    frame = 0;\n  } else {\n    frame += 1;\n  }\n  if (frame === 0) return last;\n  let enhance;\n  if (image instanceof tf.Tensor) {\n    const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);\n    enhance = tf.mul(resize, [255.0]);\n    tf.dispose(resize);\n  } else {\n    enhance = await getImage(image, config.face.age.inputSize);\n  }\n  const obj = {};\n  if (config.face.age.enabled) {\n    const ageT = await models.age.predict(enhance);\n    const data = await ageT.data();\n    obj.age = Math.trunc(10 * data[0]) / 10;\n    tf.dispose(ageT);\n  }\n  if (config.face.gender.enabled) {\n    const genderT = await models.gender.predict(enhance);\n    const data = await genderT.data();\n    const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;\n    if (confidence > config.face.gender.minConfidence) {\n      obj.gender = data[0] <= 0.5 ? 'female' : 'male';\n      obj.confidence = confidence;\n    }\n    tf.dispose(genderT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.loadAge = loadAge;\nexports.loadGender = loadGender;\n", "const tf = require('@tensorflow/tfjs');\n\nconst annotations = ['angry', 'discust', 'fear', 'happy', 'sad', 'surpise', 'neutral'];\nconst models = {};\nlet last = [];\nlet frame = 0;\nconst multiplier = 1.5;\n\nfunction getImage(image, size) {\n  const tensor = tf.tidy(() => {\n    const buffer = tf.browser.fromPixels(image, 1);\n    const resize = tf.image.resizeBilinear(buffer, [size, size]);\n    const expand = tf.cast(tf.expandDims(resize, 0), 'float32');\n    return expand;\n  });\n  return tensor;\n}\n\nasync function load(config) {\n  if (!models.emotion) models.emotion = await tf.loadGraphModel(config.face.emotion.modelPath);\n  return models.emotion;\n}\n\nasync function predict(image, config) {\n  frame += 1;\n  if (frame >= config.face.emotion.skipFrames) {\n    frame = 0;\n    return last;\n  }\n  const enhance = tf.tidy(() => {\n    if (image instanceof tf.Tensor) {\n      const resize = tf.image.resizeBilinear(image, [config.face.emotion.inputSize, config.face.emotion.inputSize], false);\n      const [r, g, b] = tf.split(resize, 3, 3);\n      if (config.face.emotion.useGrayscale) {\n        // weighted rgb to grayscale: https://www.mathworks.com/help/matlab/ref/rgb2gray.html\n        const r1 = tf.mul(r, [0.2989]);\n        const g1 = tf.mul(g, [0.5870]);\n        const b1 = tf.mul(b, [0.1140]);\n        const grayscale = tf.addN([r1, g1, b1]);\n        return grayscale;\n      }\n      return g;\n    }\n    return getImage(image, config.face.emotion.inputSize);\n  });\n  const obj = [];\n  if (config.face.emotion.enabled) {\n    const emotionT = await models.emotion.predict(enhance);\n    const data = await emotionT.data();\n    for (let i = 0; i < data.length; i++) {\n      if (multiplier * data[i] > config.face.emotion.minConfidence) obj.push({ score: Math.min(0.99, Math.trunc(100 * multiplier * data[i]) / 100), emotion: annotations[i] });\n    }\n    obj.sort((a, b) => b.score - a.score);\n    tf.dispose(emotionT);\n  }\n  tf.dispose(enhance);\n  last = obj;\n  return obj;\n}\n\nexports.predict = predict;\nexports.load = load;\n", "const tf = require('@tensorflow/tfjs');\n\nclass BaseModel {\n  constructor(model, outputStride) {\n    this.model = model;\n    this.outputStride = outputStride;\n    const inputShape = this.model.inputs[0].shape;\n    tf.util.assert((inputShape[1] === -1) && (inputShape[2] === -1), () => `Input shape [${inputShape[1]}, ${inputShape[2]}] must both be equal to or -1`);\n  }\n\n  /**\n     * Predicts intermediate Tensor representations.\n     *\n     * @param input The input RGB image of the base model.\n     * A Tensor of shape: [`inputResolution`, `inputResolution`, 3].\n     *\n     * @return A dictionary of base model's intermediate predictions.\n     * The returned dictionary should contains the following elements:\n     * heatmapScores: A Tensor3D that represents the heatmapScores.\n     * offsets: A Tensor3D that represents the offsets.\n     * displacementFwd: A Tensor3D that represents the forward displacement.\n     * displacementBwd: A Tensor3D that represents the backward displacement.\n     */\n  predict(input) {\n    return tf.tidy(() => {\n      const asFloat = this.preprocessInput(input.toFloat());\n      const asBatch = asFloat.expandDims(0);\n      const results = this.model.predict(asBatch);\n      const results3d = results.map((y) => y.squeeze([0]));\n      const namedResults = this.nameOutputResults(results3d);\n      return {\n        heatmapScores: namedResults.heatmap.sigmoid(),\n        offsets: namedResults.offsets,\n        displacementFwd: namedResults.displacementFwd,\n        displacementBwd: namedResults.displacementBwd,\n      };\n    });\n  }\n\n  /**\n     * Releases the CPU and GPU memory allocated by the model.\n     */\n  dispose() {\n    this.model.dispose();\n  }\n}\nexports.BaseModel = BaseModel;\n", "const tf = require('@tensorflow/tfjs');\nconst modelBase = require('./modelBase');\n\nclass MobileNet extends modelBase.BaseModel {\n  // eslint-disable-next-line class-methods-use-this\n  preprocessInput(input) {\n    // Normalize the pixels [0, 255] to be between [-1, 1].\n    return tf.tidy(() => tf.div(input, 127.5).sub(1.0));\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  nameOutputResults(results) {\n    const [offsets, heatmap, displacementFwd, displacementBwd] = results;\n    return { offsets, heatmap, displacementFwd, displacementBwd };\n  }\n}\nexports.MobileNet = MobileNet;\n", "// algorithm based on Coursera Lecture from Algorithms, Part 1: https://www.coursera.org/learn/algorithms-part1/lecture/ZjoSM/heapsort\nfunction half(k) {\n  return Math.floor(k / 2);\n}\nclass MaxHeap {\n  constructor(maxSize, getElementValue) {\n    this.priorityQueue = new Array(maxSize);\n    this.numberOfElements = -1;\n    this.getElementValue = getElementValue;\n  }\n\n  enqueue(x) {\n    this.priorityQueue[++this.numberOfElements] = x;\n    this.swim(this.numberOfElements);\n  }\n\n  dequeue() {\n    const max = this.priorityQueue[0];\n    this.exchange(0, this.numberOfElements--);\n    this.sink(0);\n    this.priorityQueue[this.numberOfElements + 1] = null;\n    return max;\n  }\n\n  empty() {\n    return this.numberOfElements === -1;\n  }\n\n  size() {\n    return this.numberOfElements + 1;\n  }\n\n  all() {\n    return this.priorityQueue.slice(0, this.numberOfElements + 1);\n  }\n\n  max() {\n    return this.priorityQueue[0];\n  }\n\n  swim(k) {\n    while (k > 0 && this.less(half(k), k)) {\n      this.exchange(k, half(k));\n      k = half(k);\n    }\n  }\n\n  sink(k) {\n    while (2 * k <= this.numberOfElements) {\n      let j = 2 * k;\n      if (j < this.numberOfElements && this.less(j, j + 1)) j++;\n      if (!this.less(k, j)) break;\n      this.exchange(k, j);\n      k = j;\n    }\n  }\n\n  getValueAt(i) {\n    return this.getElementValue(this.priorityQueue[i]);\n  }\n\n  less(i, j) {\n    return this.getValueAt(i) < this.getValueAt(j);\n  }\n\n  exchange(i, j) {\n    const t = this.priorityQueue[i];\n    this.priorityQueue[i] = this.priorityQueue[j];\n    this.priorityQueue[j] = t;\n  }\n}\nexports.MaxHeap = MaxHeap;\n", "const heapSort = require('./heapSort');\n\nfunction scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores) {\n  const [height, width] = scores.shape;\n  let localMaximum = true;\n  const yStart = Math.max(heatmapY - localMaximumRadius, 0);\n  const yEnd = Math.min(heatmapY + localMaximumRadius + 1, height);\n  for (let yCurrent = yStart; yCurrent < yEnd; ++yCurrent) {\n    const xStart = Math.max(heatmapX - localMaximumRadius, 0);\n    const xEnd = Math.min(heatmapX + localMaximumRadius + 1, width);\n    for (let xCurrent = xStart; xCurrent < xEnd; ++xCurrent) {\n      if (scores.get(yCurrent, xCurrent, keypointId) > score) {\n        localMaximum = false;\n        break;\n      }\n    }\n    if (!localMaximum) {\n      break;\n    }\n  }\n  return localMaximum;\n}\n/**\n * Builds a priority queue with part candidate positions for a specific image in\n * the batch. For this we find all local maxima in the score maps with score\n * values above a threshold. We create a single priority queue across all parts.\n */\nfunction buildPartWithScoreQueue(scoreThreshold, localMaximumRadius, scores) {\n  const [height, width, numKeypoints] = scores.shape;\n  const queue = new heapSort.MaxHeap(height * width * numKeypoints, ({ score }) => score);\n  for (let heatmapY = 0; heatmapY < height; ++heatmapY) {\n    for (let heatmapX = 0; heatmapX < width; ++heatmapX) {\n      for (let keypointId = 0; keypointId < numKeypoints; ++keypointId) {\n        const score = scores.get(heatmapY, heatmapX, keypointId);\n        // Only consider parts with score greater or equal to threshold as root candidates.\n        if (score < scoreThreshold) continue;\n        // Only consider keypoints whose score is maximum in a local window.\n        if (scoreIsMaximumInLocalWindow(keypointId, score, heatmapY, heatmapX, localMaximumRadius, scores)) {\n          queue.enqueue({ score, part: { heatmapY, heatmapX, id: keypointId } });\n        }\n      }\n    }\n  }\n  return queue;\n}\nexports.buildPartWithScoreQueue = buildPartWithScoreQueue;\n", "exports.partNames = [\n  'nose', 'leftEye', 'rightEye', 'leftEar', 'rightEar', 'leftShoulder',\n  'rightShoulder', 'leftElbow', 'rightElbow', 'leftWrist', 'rightWrist',\n  'leftHip', 'rightHip', 'leftKnee', 'rightKnee', 'leftAnkle', 'rightAnkle',\n];\nexports.NUM_KEYPOINTS = exports.partNames.length;\nexports.partIds = exports.partNames.reduce((result, jointName, i) => {\n  result[jointName] = i;\n  return result;\n}, {});\nconst connectedPartNames = [\n  ['leftHip', 'leftShoulder'], ['leftElbow', 'leftShoulder'],\n  ['leftElbow', 'leftWrist'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['rightHip', 'rightShoulder'],\n  ['rightElbow', 'rightShoulder'], ['rightElbow', 'rightWrist'],\n  ['rightHip', 'rightKnee'], ['rightKnee', 'rightAnkle'],\n  ['leftShoulder', 'rightShoulder'], ['leftHip', 'rightHip'],\n];\n/*\n * Define the skeleton. This defines the parent->child relationships of our\n * tree. Arbitrarily this defines the nose as the root of the tree, however\n * since we will infer the displacement for both parent->child and\n * child->parent, we can define the tree root as any node.\n */\nexports.poseChain = [\n  ['nose', 'leftEye'], ['leftEye', 'leftEar'], ['nose', 'rightEye'],\n  ['rightEye', 'rightEar'], ['nose', 'leftShoulder'],\n  ['leftShoulder', 'leftElbow'], ['leftElbow', 'leftWrist'],\n  ['leftShoulder', 'leftHip'], ['leftHip', 'leftKnee'],\n  ['leftKnee', 'leftAnkle'], ['nose', 'rightShoulder'],\n  ['rightShoulder', 'rightElbow'], ['rightElbow', 'rightWrist'],\n  ['rightShoulder', 'rightHip'], ['rightHip', 'rightKnee'],\n  ['rightKnee', 'rightAnkle'],\n];\nexports.connectedPartIndices = connectedPartNames.map(([jointNameA, jointNameB]) => ([exports.partIds[jointNameA], exports.partIds[jointNameB]]));\nexports.partChannels = [\n  'left_face',\n  'right_face',\n  'right_upper_leg_front',\n  'right_lower_leg_back',\n  'right_upper_leg_back',\n  'left_lower_leg_front',\n  'left_upper_leg_front',\n  'left_upper_leg_back',\n  'left_lower_leg_back',\n  'right_feet',\n  'right_lower_leg_front',\n  'left_feet',\n  'torso_front',\n  'torso_back',\n  'right_upper_arm_front',\n  'right_upper_arm_back',\n  'right_lower_arm_back',\n  'left_lower_arm_front',\n  'left_upper_arm_front',\n  'left_upper_arm_back',\n  'left_lower_arm_back',\n  'right_hand',\n  'right_lower_arm_front',\n  'left_hand',\n];\n", "const kpt = require('./keypoints');\n\nfunction getOffsetPoint(y, x, keypoint, offsets) {\n  return {\n    y: offsets.get(y, x, keypoint),\n    x: offsets.get(y, x, keypoint + kpt.NUM_KEYPOINTS),\n  };\n}\nexports.getOffsetPoint = getOffsetPoint;\n\nfunction getImageCoords(part, outputStride, offsets) {\n  const { heatmapY, heatmapX, id: keypoint } = part;\n  const { y, x } = getOffsetPoint(heatmapY, heatmapX, keypoint, offsets);\n  return {\n    x: part.heatmapX * outputStride + x,\n    y: part.heatmapY * outputStride + y,\n  };\n}\nexports.getImageCoords = getImageCoords;\n\nfunction fillArray(element, size) {\n  const result = new Array(size);\n  for (let i = 0; i < size; i++) {\n    result[i] = element;\n  }\n  return result;\n}\nexports.fillArray = fillArray;\n\nfunction clamp(a, min, max) {\n  if (a < min) return min;\n  if (a > max) return max;\n  return a;\n}\nexports.clamp = clamp;\n\nfunction squaredDistance(y1, x1, y2, x2) {\n  const dy = y2 - y1;\n  const dx = x2 - x1;\n  return dy * dy + dx * dx;\n}\nexports.squaredDistance = squaredDistance;\n\nfunction addVectors(a, b) {\n  return { x: a.x + b.x, y: a.y + b.y };\n}\nexports.addVectors = addVectors;\n\nfunction clampVector(a, min, max) {\n  return { y: clamp(a.y, min, max), x: clamp(a.x, min, max) };\n}\nexports.clampVector = clampVector;\n", "const keypoints = require('./keypoints');\nconst vectors = require('./vectors');\n\nconst parentChildrenTuples = keypoints.poseChain.map(([parentJoinName, childJoinName]) => ([keypoints.partIds[parentJoinName], keypoints.partIds[childJoinName]]));\nconst parentToChildEdges = parentChildrenTuples.map(([, childJointId]) => childJointId);\nconst childToParentEdges = parentChildrenTuples.map(([parentJointId]) => parentJointId);\nfunction getDisplacement(edgeId, point, displacements) {\n  const numEdges = displacements.shape[2] / 2;\n  return {\n    y: displacements.get(point.y, point.x, edgeId),\n    x: displacements.get(point.y, point.x, numEdges + edgeId),\n  };\n}\nfunction getStridedIndexNearPoint(point, outputStride, height, width) {\n  return {\n    y: vectors.clamp(Math.round(point.y / outputStride), 0, height - 1),\n    x: vectors.clamp(Math.round(point.x / outputStride), 0, width - 1),\n  };\n}\n/**\n * We get a new keypoint along the `edgeId` for the pose instance, assuming\n * that the position of the `idSource` part is already known. For this, we\n * follow the displacement vector from the source to target part (stored in\n * the `i`-t channel of the displacement tensor). The displaced keypoint\n * vector is refined using the offset vector by `offsetRefineStep` times.\n */\nfunction traverseToTargetKeypoint(edgeId, sourceKeypoint, targetKeypointId, scoresBuffer, offsets, outputStride, displacements, offsetRefineStep = 2) {\n  const [height, width] = scoresBuffer.shape;\n  // Nearest neighbor interpolation for the source->target displacements.\n  const sourceKeypointIndices = getStridedIndexNearPoint(sourceKeypoint.position, outputStride, height, width);\n  const displacement = getDisplacement(edgeId, sourceKeypointIndices, displacements);\n  const displacedPoint = vectors.addVectors(sourceKeypoint.position, displacement);\n  let targetKeypoint = displacedPoint;\n  for (let i = 0; i < offsetRefineStep; i++) {\n    const targetKeypointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n    const offsetPoint = vectors.getOffsetPoint(targetKeypointIndices.y, targetKeypointIndices.x, targetKeypointId, offsets);\n    targetKeypoint = vectors.addVectors({\n      x: targetKeypointIndices.x * outputStride,\n      y: targetKeypointIndices.y * outputStride,\n    }, { x: offsetPoint.x, y: offsetPoint.y });\n  }\n  const targetKeyPointIndices = getStridedIndexNearPoint(targetKeypoint, outputStride, height, width);\n  const score = scoresBuffer.get(targetKeyPointIndices.y, targetKeyPointIndices.x, targetKeypointId);\n  return { position: targetKeypoint, part: keypoints.partNames[targetKeypointId], score };\n}\n/**\n * Follows the displacement fields to decode the full pose of the object\n * instance given the position of a part that acts as root.\n *\n * @return An array of decoded keypoints and their scores for a single pose\n */\nfunction decodePose(root, scores, offsets, outputStride, displacementsFwd, displacementsBwd) {\n  const numParts = scores.shape[2];\n  const numEdges = parentToChildEdges.length;\n  const instanceKeypoints = new Array(numParts);\n  // Start a new detection instance at the position of the root.\n  const { part: rootPart, score: rootScore } = root;\n  const rootPoint = vectors.getImageCoords(rootPart, outputStride, offsets);\n  instanceKeypoints[rootPart.id] = {\n    score: rootScore,\n    part: keypoints.partNames[rootPart.id],\n    position: rootPoint,\n  };\n  // Decode the part positions upwards in the tree, following the backward\n  // displacements.\n  for (let edge = numEdges - 1; edge >= 0; --edge) {\n    const sourceKeypointId = parentToChildEdges[edge];\n    const targetKeypointId = childToParentEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsBwd);\n    }\n  }\n  // Decode the part positions downwards in the tree, following the forward\n  // displacements.\n  for (let edge = 0; edge < numEdges; ++edge) {\n    const sourceKeypointId = childToParentEdges[edge];\n    const targetKeypointId = parentToChildEdges[edge];\n    if (instanceKeypoints[sourceKeypointId] && !instanceKeypoints[targetKeypointId]) {\n      instanceKeypoints[targetKeypointId] = traverseToTargetKeypoint(edge, instanceKeypoints[sourceKeypointId], targetKeypointId, scores, offsets, outputStride, displacementsFwd);\n    }\n  }\n  return instanceKeypoints;\n}\nexports.decodePose = decodePose;\n", "const buildParts = require('./buildParts');\nconst decodePose = require('./decodePose');\nconst vectors = require('./vectors');\n\nfunction withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, { x, y }, keypointId) {\n  return poses.some(({ keypoints }) => {\n    const correspondingKeypoint = keypoints[keypointId].position;\n    return vectors.squaredDistance(y, x, correspondingKeypoint.y, correspondingKeypoint.x) <= squaredNmsRadius;\n  });\n}\n/* Score the newly proposed object instance without taking into account\n * the scores of the parts that overlap with any previously detected\n * instance.\n */\nfunction getInstanceScore(existingPoses, squaredNmsRadius, instanceKeypoints) {\n  const notOverlappedKeypointScores = instanceKeypoints.reduce((result, { position, score }, keypointId) => {\n    if (!withinNmsRadiusOfCorrespondingPoint(existingPoses, squaredNmsRadius, position, keypointId)) {\n      result += score;\n    }\n    return result;\n  }, 0.0);\n  return notOverlappedKeypointScores / instanceKeypoints.length;\n}\n// A point (y, x) is considered as root part candidate if its score is a\n// maximum in a window |y - y'| <= kLocalMaximumRadius, |x - x'| <=\n// kLocalMaximumRadius.\nconst kLocalMaximumRadius = 1;\n/**\n * Detects multiple poses and finds their parts from part scores and\n * displacement vectors. It returns up to `maxDetections` object instance\n * detections in decreasing root score order. It works as follows: We first\n * create a priority queue with local part score maxima above\n * `scoreThreshold`, considering all parts at the same time. Then we\n * iteratively pull the top  element of the queue (in decreasing score order)\n * and treat it as a root candidate for a new object instance. To avoid\n * duplicate detections, we reject the root candidate if it is within a disk\n * of `nmsRadius` pixels from the corresponding part of a previously detected\n * instance, which is a form of part-based non-maximum suppression (NMS). If\n * the root candidate passes the NMS check, we start a new object instance\n * detection, treating the corresponding part as root and finding the\n * positions of the remaining parts by following the displacement vectors\n * along the tree-structured part graph. We assign to the newly detected\n * instance a score equal to the sum of scores of its parts which have not\n * been claimed by a previous instance (i.e., those at least `nmsRadius`\n * pixels away from the corresponding part of all previously detected\n * instances), divided by the total number of parts `numParts`.\n *\n * @param heatmapScores 3-D tensor with shape `[height, width, numParts]`.\n * The value of heatmapScores[y, x, k]` is the score of placing the `k`-th\n * object part at position `(y, x)`.\n *\n * @param offsets 3-D tensor with shape `[height, width, numParts * 2]`.\n * The value of [offsets[y, x, k], offsets[y, x, k + numParts]]` is the\n * short range offset vector of the `k`-th  object part at heatmap\n * position `(y, x)`.\n *\n * @param displacementsFwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the forward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param displacementsBwd 3-D tensor of shape\n * `[height, width, 2 * num_edges]`, where `num_edges = num_parts - 1` is the\n * number of edges (parent-child pairs) in the tree. It contains the backward\n * displacements between consecutive part from the root towards the leaves.\n *\n * @param outputStride The output stride that was used when feed-forwarding\n * through the PoseNet model.  Must be 32, 16, or 8.\n *\n * @param maxPoseDetections Maximum number of returned instance detections per\n * image.\n *\n * @param scoreThreshold Only return instance detections that have root part\n * score greater or equal to this value. Defaults to 0.5.\n *\n * @param nmsRadius Non-maximum suppression part distance. It needs to be\n * strictly positive. Two parts suppress each other if they are less than\n * `nmsRadius` pixels away. Defaults to 20.\n *\n * @return An array of poses and their scores, each containing keypoints and\n * the corresponding keypoint scores.\n */\nfunction decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, maxPoseDetections, scoreThreshold = 0.5, nmsRadius = 20) {\n  const poses = [];\n  const queue = buildParts.buildPartWithScoreQueue(scoreThreshold, kLocalMaximumRadius, scoresBuffer);\n  const squaredNmsRadius = nmsRadius * nmsRadius;\n  // Generate at most maxDetections object instances per image in\n  // decreasing root part score order.\n  while (poses.length < maxPoseDetections && !queue.empty()) {\n    // The top element in the queue is the next root candidate.\n    const root = queue.dequeue();\n    // Part-based non-maximum suppression: We reject a root candidate if it\n    // is within a disk of `nmsRadius` pixels from the corresponding part of\n    // a previously detected instance.\n    const rootImageCoords = vectors.getImageCoords(root.part, outputStride, offsetsBuffer);\n    if (withinNmsRadiusOfCorrespondingPoint(poses, squaredNmsRadius, rootImageCoords, root.part.id)) continue;\n    // Start a new detection instance at the position of the root.\n    const keypoints = decodePose.decodePose(root, scoresBuffer, offsetsBuffer, outputStride, displacementsFwdBuffer, displacementsBwdBuffer);\n    const score = getInstanceScore(poses, squaredNmsRadius, keypoints);\n    poses.push({ keypoints, score });\n  }\n  return poses;\n}\nexports.decodeMultiplePoses = decodeMultiplePoses;\n", "const tf = require('@tensorflow/tfjs');\nconst kpt = require('./keypoints');\n\nfunction eitherPointDoesntMeetConfidence(a, b, minConfidence) {\n  return (a < minConfidence || b < minConfidence);\n}\n\nfunction getAdjacentKeyPoints(keypoints, minConfidence) {\n  return kpt.connectedPartIndices.reduce((result, [leftJoint, rightJoint]) => {\n    if (eitherPointDoesntMeetConfidence(keypoints[leftJoint].score, keypoints[rightJoint].score, minConfidence)) {\n      return result;\n    }\n    result.push([keypoints[leftJoint], keypoints[rightJoint]]);\n    return result;\n  }, []);\n}\nexports.getAdjacentKeyPoints = getAdjacentKeyPoints;\n\nconst { NEGATIVE_INFINITY, POSITIVE_INFINITY } = Number;\nfunction getBoundingBox(keypoints) {\n  return keypoints.reduce(({ maxX, maxY, minX, minY }, { position: { x, y } }) => ({\n    maxX: Math.max(maxX, x),\n    maxY: Math.max(maxY, y),\n    minX: Math.min(minX, x),\n    minY: Math.min(minY, y),\n  }), {\n    maxX: NEGATIVE_INFINITY,\n    maxY: NEGATIVE_INFINITY,\n    minX: POSITIVE_INFINITY,\n    minY: POSITIVE_INFINITY,\n  });\n}\nexports.getBoundingBox = getBoundingBox;\nfunction getBoundingBoxPoints(keypoints) {\n  const { minX, minY, maxX, maxY } = getBoundingBox(keypoints);\n  return [{ x: minX, y: minY }, { x: maxX, y: minY }, { x: maxX, y: maxY }, { x: minX, y: maxY }];\n}\nexports.getBoundingBoxPoints = getBoundingBoxPoints;\nasync function toTensorBuffers3D(tensors) {\n  return Promise.all(tensors.map((tensor) => tensor.buffer()));\n}\nexports.toTensorBuffers3D = toTensorBuffers3D;\n\nfunction scalePose(pose, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  return {\n    score: pose.score,\n    keypoints: pose.keypoints.map(({ score, part, position }) => ({\n      score,\n      part,\n      position: {\n        x: position.x * scaleX + offsetX,\n        y: position.y * scaleY + offsetY,\n      },\n    })),\n  };\n}\nexports.scalePose = scalePose;\n\nfunction scalePoses(poses, scaleY, scaleX, offsetY = 0, offsetX = 0) {\n  if (scaleX === 1 && scaleY === 1 && offsetY === 0 && offsetX === 0) {\n    return poses;\n  }\n  return poses.map((pose) => scalePose(pose, scaleY, scaleX, offsetY, offsetX));\n}\nexports.scalePoses = scalePoses;\n\nfunction getInputTensorDimensions(input) {\n  return input instanceof tf.Tensor ? [input.shape[0], input.shape[1]] : [input.height, input.width];\n}\nexports.getInputTensorDimensions = getInputTensorDimensions;\n\nfunction toInputTensor(input) {\n  return input instanceof tf.Tensor ? input : tf.browser.fromPixels(input);\n}\nexports.toInputTensor = toInputTensor;\n\nfunction toResizedInputTensor(input, resizeHeight, resizeWidth) {\n  return tf.tidy(() => {\n    const imageTensor = toInputTensor(input);\n    return imageTensor.resizeBilinear([resizeHeight, resizeWidth]);\n  });\n}\nexports.toResizedInputTensor = toResizedInputTensor;\n\nfunction padAndResizeTo(input, [targetH, targetW]) {\n  const [height, width] = getInputTensorDimensions(input);\n  const targetAspect = targetW / targetH;\n  const aspect = width / height;\n  let [padT, padB, padL, padR] = [0, 0, 0, 0];\n  if (aspect < targetAspect) {\n    // pads the width\n    padT = 0;\n    padB = 0;\n    padL = Math.round(0.5 * (targetAspect * height - width));\n    padR = Math.round(0.5 * (targetAspect * height - width));\n  } else {\n    // pads the height\n    padT = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padB = Math.round(0.5 * ((1.0 / targetAspect) * width - height));\n    padL = 0;\n    padR = 0;\n  }\n  const resized = tf.tidy(() => {\n    let imageTensor = toInputTensor(input);\n    imageTensor = tf.pad3d(imageTensor, [[padT, padB], [padL, padR], [0, 0]]);\n    return imageTensor.resizeBilinear([targetH, targetW]);\n  });\n  return { resized, padding: { top: padT, left: padL, right: padR, bottom: padB } };\n}\nexports.padAndResizeTo = padAndResizeTo;\n\nfunction scaleAndFlipPoses(poses, [height, width], [inputResolutionHeight, inputResolutionWidth], padding) {\n  const scaleY = (height + padding.top + padding.bottom) / (inputResolutionHeight);\n  const scaleX = (width + padding.left + padding.right) / (inputResolutionWidth);\n  const scaledPoses = scalePoses(poses, scaleY, scaleX, -padding.top, -padding.left);\n  return scaledPoses;\n}\nexports.scaleAndFlipPoses = scaleAndFlipPoses;\n", "const tf = require('@tensorflow/tfjs');\nconst modelMobileNet = require('./modelMobileNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst util = require('./util');\n\nclass PoseNet {\n  constructor(net) {\n    this.baseModel = net;\n  }\n\n  /**\n     * Infer through PoseNet, and estimates multiple poses using the outputs.\n     * This does standard ImageNet pre-processing before inferring through the\n     * model. The image should pixels should have values [0-255]. It detects\n     * multiple poses and finds their parts from part scores and displacement\n     * vectors using a fast greedy decoding algorithm.  It returns up to\n     * `config.maxDetections` object instance detections in decreasing root\n     * score order.\n     *\n     * @param input\n     * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement) The input\n     * image to feed through the network.\n     *\n     * @param config MultiPoseEstimationConfig object that contains parameters\n     * for the PoseNet inference using multiple pose estimation.\n     *\n     * @return An array of poses and their scores, each containing keypoints and\n     * the corresponding keypoint scores.  The positions of the keypoints are\n     * in the same scale as the original image\n     */\n  async estimatePoses(input, config) {\n    const outputStride = config.outputStride;\n    // const inputResolution = config.inputResolution;\n    const [height, width] = util.getInputTensorDimensions(input);\n    const { resized, padding } = util.padAndResizeTo(input, [config.inputResolution, config.inputResolution]);\n    const { heatmapScores, offsets, displacementFwd, displacementBwd } = this.baseModel.predict(resized);\n    const allTensorBuffers = await util.toTensorBuffers3D([heatmapScores, offsets, displacementFwd, displacementBwd]);\n    const scoresBuffer = allTensorBuffers[0];\n    const offsetsBuffer = allTensorBuffers[1];\n    const displacementsFwdBuffer = allTensorBuffers[2];\n    const displacementsBwdBuffer = allTensorBuffers[3];\n    const poses = await decodeMultiple.decodeMultiplePoses(scoresBuffer, offsetsBuffer, displacementsFwdBuffer, displacementsBwdBuffer, outputStride, config.maxDetections, config.scoreThreshold, config.nmsRadius);\n    const resultPoses = util.scaleAndFlipPoses(poses, [height, width], [config.inputResolution, config.inputResolution], padding);\n    heatmapScores.dispose();\n    offsets.dispose();\n    displacementFwd.dispose();\n    displacementBwd.dispose();\n    resized.dispose();\n    return resultPoses;\n  }\n\n  dispose() {\n    this.baseModel.dispose();\n  }\n}\nexports.PoseNet = PoseNet;\nasync function loadMobileNet(config) {\n  const graphModel = await tf.loadGraphModel(config.modelPath);\n  const mobilenet = new modelMobileNet.MobileNet(graphModel, config.outputStride);\n  return new PoseNet(mobilenet);\n}\n/**\n * Loads the PoseNet model instance from a checkpoint, with the MobileNet architecture. The model to be loaded is configurable using the\n * config dictionary ModelConfig. Please find more details in the documentation of the ModelConfig.\n *\n * @param config ModelConfig dictionary that contains parameters for\n * the PoseNet loading process. Please find more details of each parameters\n * in the documentation of the ModelConfig interface. The predefined\n * `MOBILENET_V1_CONFIG` and `RESNET_CONFIG` can also be used as references\n * for defining your customized config.\n */\nasync function load(config) {\n  return loadMobileNet(config);\n}\nexports.load = load;\n", "const modelMobileNet = require('./modelMobileNet');\nconst modelPoseNet = require('./modelPoseNet');\nconst decodeMultiple = require('./decodeMultiple');\nconst keypoints = require('./keypoints');\nconst util = require('./util');\n\nexports.load = modelPoseNet.load;\nexports.PoseNet = modelPoseNet.PoseNet;\n\nexports.MobileNet = modelMobileNet.MobileNet;\nexports.decodeMultiplePoses = decodeMultiple.decodeMultiplePoses;\nexports.partChannels = keypoints.partChannels;\nexports.partIds = keypoints.partIds;\nexports.partNames = keypoints.partNames;\nexports.poseChain = keypoints.poseChain;\nexports.getAdjacentKeyPoints = util.getAdjacentKeyPoints;\nexports.getBoundingBox = util.getBoundingBox;\nexports.getBoundingBoxPoints = util.getBoundingBoxPoints;\nexports.scaleAndFlipPoses = util.scaleAndFlipPoses;\nexports.scalePose = util.scalePose;\n", "const tf = require('@tensorflow/tfjs');\n\nfunction getBoxSize(box) {\n  return [\n    Math.abs(box.endPoint[0] - box.startPoint[0]),\n    Math.abs(box.endPoint[1] - box.startPoint[1]),\n  ];\n}\nexports.getBoxSize = getBoxSize;\n\nfunction getBoxCenter(box) {\n  return [\n    box.startPoint[0] + (box.endPoint[0] - box.startPoint[0]) / 2,\n    box.startPoint[1] + (box.endPoint[1] - box.startPoint[1]) / 2,\n  ];\n}\nexports.getBoxCenter = getBoxCenter;\n\nfunction cutBoxFromImageAndResize(box, image, cropSize) {\n  const h = image.shape[1];\n  const w = image.shape[2];\n  const boxes = [[\n    box.startPoint[1] / h, box.startPoint[0] / w, box.endPoint[1] / h,\n    box.endPoint[0] / w,\n  ]];\n  return tf.image.cropAndResize(image, boxes, [0], cropSize);\n}\nexports.cutBoxFromImageAndResize = cutBoxFromImageAndResize;\n\nfunction scaleBoxCoordinates(box, factor) {\n  const startPoint = [box.startPoint[0] * factor[0], box.startPoint[1] * factor[1]];\n  const endPoint = [box.endPoint[0] * factor[0], box.endPoint[1] * factor[1]];\n  const palmLandmarks = box.palmLandmarks.map((coord) => {\n    const scaledCoord = [coord[0] * factor[0], coord[1] * factor[1]];\n    return scaledCoord;\n  });\n  return { startPoint, endPoint, palmLandmarks };\n}\nexports.scaleBoxCoordinates = scaleBoxCoordinates;\n\nfunction enlargeBox(box, factor = 1.5) {\n  const center = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const newHalfSize = [factor * size[0] / 2, factor * size[1] / 2];\n  const startPoint = [center[0] - newHalfSize[0], center[1] - newHalfSize[1]];\n  const endPoint = [center[0] + newHalfSize[0], center[1] + newHalfSize[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.enlargeBox = enlargeBox;\n\nfunction squarifyBox(box) {\n  const centers = getBoxCenter(box);\n  const size = getBoxSize(box);\n  const maxEdge = Math.max(...size);\n  const halfSize = maxEdge / 2;\n  const startPoint = [centers[0] - halfSize, centers[1] - halfSize];\n  const endPoint = [centers[0] + halfSize, centers[1] + halfSize];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.squarifyBox = squarifyBox;\n\nfunction shiftBox(box, shiftFactor) {\n  const boxSize = [\n    box.endPoint[0] - box.startPoint[0], box.endPoint[1] - box.startPoint[1],\n  ];\n  const shiftVector = [boxSize[0] * shiftFactor[0], boxSize[1] * shiftFactor[1]];\n  const startPoint = [box.startPoint[0] + shiftVector[0], box.startPoint[1] + shiftVector[1]];\n  const endPoint = [box.endPoint[0] + shiftVector[0], box.endPoint[1] + shiftVector[1]];\n  return { startPoint, endPoint, palmLandmarks: box.palmLandmarks };\n}\nexports.shiftBox = shiftBox;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\n\nclass HandDetector {\n  constructor(model, anchors, config) {\n    this.model = model;\n    this.width = config.inputSize;\n    this.height = config.inputSize;\n    this.anchors = anchors.map((anchor) => [anchor.x_center, anchor.y_center]);\n    this.anchorsTensor = tf.tensor2d(this.anchors);\n    this.inputSizeTensor = tf.tensor1d([config.inputSize, config.inputSize]);\n    this.doubleInputSizeTensor = tf.tensor1d([config.inputSize * 2, config.inputSize * 2]);\n  }\n\n  normalizeBoxes(boxes) {\n    return tf.tidy(() => {\n      const boxOffsets = tf.slice(boxes, [0, 0], [-1, 2]);\n      const boxSizes = tf.slice(boxes, [0, 2], [-1, 2]);\n      const boxCenterPoints = tf.add(tf.div(boxOffsets, this.inputSizeTensor), this.anchorsTensor);\n      const halfBoxSizes = tf.div(boxSizes, this.doubleInputSizeTensor);\n      const startPoints = tf.mul(tf.sub(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      const endPoints = tf.mul(tf.add(boxCenterPoints, halfBoxSizes), this.inputSizeTensor);\n      return tf.concat2d([startPoints, endPoints], 1);\n    });\n  }\n\n  normalizeLandmarks(rawPalmLandmarks, index) {\n    return tf.tidy(() => {\n      const landmarks = tf.add(tf.div(rawPalmLandmarks.reshape([-1, 7, 2]), this.inputSizeTensor), this.anchors[index]);\n      return tf.mul(landmarks, this.inputSizeTensor);\n    });\n  }\n\n  async getBoundingBoxes(input) {\n    const normalizedInput = tf.tidy(() => tf.mul(tf.sub(input, 0.5), 2));\n    const batchedPrediction = this.model.predict(normalizedInput);\n    const prediction = batchedPrediction.squeeze();\n    // Regression score for each anchor point.\n    const scores = tf.tidy(() => tf.sigmoid(tf.slice(prediction, [0, 0], [-1, 1])).squeeze());\n    // Bounding box for each anchor point.\n    const rawBoxes = tf.slice(prediction, [0, 1], [-1, 4]);\n    const boxes = this.normalizeBoxes(rawBoxes);\n    const boxesWithHandsTensor = await tf.image.nonMaxSuppressionAsync(boxes, scores, this.maxHands, this.iouThreshold, this.scoreThreshold);\n    const boxesWithHands = await boxesWithHandsTensor.array();\n    const toDispose = [\n      normalizedInput, batchedPrediction, boxesWithHandsTensor, prediction,\n      boxes, rawBoxes, scores,\n    ];\n    if (boxesWithHands.length === 0) {\n      toDispose.forEach((tensor) => tensor.dispose());\n      return null;\n    }\n    const detectedHands = tf.tidy(() => {\n      const detectedBoxes = [];\n      for (const i in boxesWithHands) {\n        const boxIndex = boxesWithHands[i];\n        const matchingBox = tf.slice(boxes, [boxIndex, 0], [1, -1]);\n        const rawPalmLandmarks = tf.slice(prediction, [boxIndex, 5], [1, 14]);\n        const palmLandmarks = tf.tidy(() => this.normalizeLandmarks(rawPalmLandmarks, boxIndex).reshape([-1, 2]));\n        detectedBoxes.push({ boxes: matchingBox, palmLandmarks });\n      }\n      return detectedBoxes;\n    });\n    return detectedHands;\n  }\n\n  /**\n     * Returns a Box identifying the bounding box of a hand within the image.\n     * Returns null if there is no hand in the image.\n     *\n     * @param input The image to classify.\n     */\n  async estimateHandBounds(input, config) {\n    const inputHeight = input.shape[1];\n    const inputWidth = input.shape[2];\n    this.iouThreshold = config.iouThreshold;\n    this.scoreThreshold = config.scoreThreshold;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => input.resizeBilinear([this.width, this.height]).div(255));\n    const predictions = await this.getBoundingBoxes(image);\n    image.dispose();\n    if (!predictions || (predictions.length === 0)) return null;\n    const hands = [];\n    for (const i in predictions) {\n      const prediction = predictions[i];\n      const boundingBoxes = await prediction.boxes.array();\n      const startPoint = boundingBoxes[0].slice(0, 2);\n      const endPoint = boundingBoxes[0].slice(2, 4);\n      const palmLandmarks = await prediction.palmLandmarks.array();\n      prediction.boxes.dispose();\n      prediction.palmLandmarks.dispose();\n      hands.push(bounding.scaleBoxCoordinates({ startPoint, endPoint, palmLandmarks }, [inputWidth / this.width, inputHeight / this.height]));\n    }\n    return hands;\n  }\n}\nexports.HandDetector = HandDetector;\n", "exports.MESH_ANNOTATIONS = {\n  thumb: [1, 2, 3, 4],\n  indexFinger: [5, 6, 7, 8],\n  middleFinger: [9, 10, 11, 12],\n  ringFinger: [13, 14, 15, 16],\n  pinky: [17, 18, 19, 20],\n  palmBase: [0],\n};\n", "function normalizeRadians(angle) {\n  return angle - 2 * Math.PI * Math.floor((angle + Math.PI) / (2 * Math.PI));\n}\nexports.normalizeRadians = normalizeRadians;\n\nfunction computeRotation(point1, point2) {\n  const radians = Math.PI / 2 - Math.atan2(-(point2[1] - point1[1]), point2[0] - point1[0]);\n  return normalizeRadians(radians);\n}\nexports.computeRotation = computeRotation;\n\nconst buildTranslationMatrix = (x, y) => ([[1, 0, x], [0, 1, y], [0, 0, 1]]);\nfunction dot(v1, v2) {\n  let product = 0;\n  for (let i = 0; i < v1.length; i++) {\n    product += v1[i] * v2[i];\n  }\n  return product;\n}\nexports.dot = dot;\n\nfunction getColumnFrom2DArr(arr, columnIndex) {\n  const column = [];\n  for (let i = 0; i < arr.length; i++) {\n    column.push(arr[i][columnIndex]);\n  }\n  return column;\n}\nexports.getColumnFrom2DArr = getColumnFrom2DArr;\n\nfunction multiplyTransformMatrices(mat1, mat2) {\n  const product = [];\n  const size = mat1.length;\n  for (let row = 0; row < size; row++) {\n    product.push([]);\n    for (let col = 0; col < size; col++) {\n      product[row].push(dot(mat1[row], getColumnFrom2DArr(mat2, col)));\n    }\n  }\n  return product;\n}\nfunction buildRotationMatrix(rotation, center) {\n  const cosA = Math.cos(rotation);\n  const sinA = Math.sin(rotation);\n  const rotationMatrix = [[cosA, -sinA, 0], [sinA, cosA, 0], [0, 0, 1]];\n  const translationMatrix = buildTranslationMatrix(center[0], center[1]);\n  const translationTimesRotation = multiplyTransformMatrices(translationMatrix, rotationMatrix);\n  const negativeTranslationMatrix = buildTranslationMatrix(-center[0], -center[1]);\n  return multiplyTransformMatrices(translationTimesRotation, negativeTranslationMatrix);\n}\nexports.buildRotationMatrix = buildRotationMatrix;\n\nfunction invertTransformMatrix(matrix) {\n  const rotationComponent = [[matrix[0][0], matrix[1][0]], [matrix[0][1], matrix[1][1]]];\n  const translationComponent = [matrix[0][2], matrix[1][2]];\n  const invertedTranslation = [\n    -dot(rotationComponent[0], translationComponent),\n    -dot(rotationComponent[1], translationComponent),\n  ];\n  return [\n    rotationComponent[0].concat(invertedTranslation[0]),\n    rotationComponent[1].concat(invertedTranslation[1]),\n    [0, 0, 1],\n  ];\n}\nexports.invertTransformMatrix = invertTransformMatrix;\n\nfunction rotatePoint(homogeneousCoordinate, rotationMatrix) {\n  return [\n    dot(homogeneousCoordinate, rotationMatrix[0]),\n    dot(homogeneousCoordinate, rotationMatrix[1]),\n  ];\n}\nexports.rotatePoint = rotatePoint;\n", "const tf = require('@tensorflow/tfjs');\nconst bounding = require('./box');\nconst util = require('./util');\n\nconst UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD = 0.8;\nconst PALM_BOX_SHIFT_VECTOR = [0, -0.4];\nconst HAND_BOX_SHIFT_VECTOR = [0, -0.1];\nconst HAND_BOX_ENLARGE_FACTOR = 1.65;\nconst PALM_LANDMARK_IDS = [0, 5, 9, 13, 17, 1, 2];\nconst PALM_LANDMARKS_INDEX_OF_PALM_BASE = 0;\nconst PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE = 2;\n\n// The Pipeline coordinates between the bounding box and skeleton models.\nclass HandPipeline {\n  constructor(boundingBoxDetector, meshDetector, config) {\n    this.regionsOfInterest = [];\n    this.runsWithoutHandDetector = 0;\n    this.boundingBoxDetector = boundingBoxDetector;\n    this.meshDetector = meshDetector;\n    this.meshWidth = config.inputSize;\n    this.meshHeight = config.inputSize;\n    this.enlargeFactor = config.enlargeFactor;\n  }\n\n  // Get the bounding box surrounding the hand, given palm landmarks.\n  getBoxForPalmLandmarks(palmLandmarks, rotationMatrix) {\n    const rotatedPalmLandmarks = palmLandmarks.map((coord) => {\n      const homogeneousCoordinate = [...coord, 1];\n      return util.rotatePoint(homogeneousCoordinate, rotationMatrix);\n    });\n    const boxAroundPalm = this.calculateLandmarksBoundingBox(rotatedPalmLandmarks);\n    // boxAroundPalm only surrounds the palm - therefore we shift it\n    // upwards so it will capture fingers once enlarged + squarified.\n    return bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boxAroundPalm, PALM_BOX_SHIFT_VECTOR)), this.enlargeFactor);\n  }\n\n  // Get the bounding box surrounding the hand, given all hand landmarks.\n  getBoxForHandLandmarks(landmarks) {\n    // The MediaPipe hand mesh model is trained on hands with empty space\n    // around them, so we still need to shift / enlarge boxAroundHand even\n    // though it surrounds the entire hand.\n    const boundingBox = this.calculateLandmarksBoundingBox(landmarks);\n    const boxAroundHand = bounding.enlargeBox(bounding.squarifyBox(bounding.shiftBox(boundingBox, HAND_BOX_SHIFT_VECTOR)), HAND_BOX_ENLARGE_FACTOR);\n    const palmLandmarks = [];\n    for (let i = 0; i < PALM_LANDMARK_IDS.length; i++) {\n      palmLandmarks.push(landmarks[PALM_LANDMARK_IDS[i]].slice(0, 2));\n    }\n    boxAroundHand.palmLandmarks = palmLandmarks;\n    return boxAroundHand;\n  }\n\n  // Scale, rotate, and translate raw keypoints from the model so they map to\n  // the input coordinates.\n  transformRawCoords(rawCoords, box, angle, rotationMatrix) {\n    const boxSize = bounding.getBoxSize(box);\n    const scaleFactor = [boxSize[0] / this.meshWidth, boxSize[1] / this.meshHeight];\n    const coordsScaled = rawCoords.map((coord) => [\n      scaleFactor[0] * (coord[0] - this.meshWidth / 2),\n      scaleFactor[1] * (coord[1] - this.meshHeight / 2), coord[2],\n    ]);\n    const coordsRotationMatrix = util.buildRotationMatrix(angle, [0, 0]);\n    const coordsRotated = coordsScaled.map((coord) => {\n      const rotated = util.rotatePoint(coord, coordsRotationMatrix);\n      return [...rotated, coord[2]];\n    });\n    const inverseRotationMatrix = util.invertTransformMatrix(rotationMatrix);\n    const boxCenter = [...bounding.getBoxCenter(box), 1];\n    const originalBoxCenter = [\n      util.dot(boxCenter, inverseRotationMatrix[0]),\n      util.dot(boxCenter, inverseRotationMatrix[1]),\n    ];\n    return coordsRotated.map((coord) => [\n      coord[0] + originalBoxCenter[0], coord[1] + originalBoxCenter[1],\n      coord[2],\n    ]);\n  }\n\n  async estimateHands(image, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const useFreshBox = this.shouldUpdateRegionsOfInterest();\n    if (useFreshBox === true) {\n      const boundingBoxPredictions = await this.boundingBoxDetector.estimateHandBounds(image, config);\n      this.regionsOfInterest = [];\n      for (const i in boundingBoxPredictions) {\n        this.updateRegionsOfInterest(boundingBoxPredictions[i], true /* force update */, i);\n      }\n      this.runsWithoutHandDetector = 0;\n    } else {\n      this.runsWithoutHandDetector++;\n    }\n    // Rotate input so the hand is vertically oriented.\n    const hands = [];\n    if (!this.regionsOfInterest) return hands;\n    for (const i in this.regionsOfInterest) {\n      const currentBox = this.regionsOfInterest[i][0];\n      if (!currentBox) return hands;\n      const angle = util.computeRotation(currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_PALM_BASE], currentBox.palmLandmarks[PALM_LANDMARKS_INDEX_OF_MIDDLE_FINGER_BASE]);\n      const palmCenter = bounding.getBoxCenter(currentBox);\n      const palmCenterNormalized = [palmCenter[0] / image.shape[2], palmCenter[1] / image.shape[1]];\n      const rotatedImage = tf.image.rotateWithOffset(image, angle, 0, palmCenterNormalized);\n      const rotationMatrix = util.buildRotationMatrix(-angle, palmCenter);\n      const box = useFreshBox ? this.getBoxForPalmLandmarks(currentBox.palmLandmarks, rotationMatrix) : currentBox;\n      const croppedInput = bounding.cutBoxFromImageAndResize(box, rotatedImage, [this.meshWidth, this.meshHeight]);\n      const handImage = croppedInput.div(255);\n      croppedInput.dispose();\n      rotatedImage.dispose();\n      const prediction = this.meshDetector.predict(handImage);\n      const [flag, keypoints] = prediction;\n      handImage.dispose();\n      const flagValue = flag.dataSync()[0];\n      flag.dispose();\n      if (flagValue < config.minConfidence) {\n        keypoints.dispose();\n        this.regionsOfInterest[i] = [];\n        return hands;\n      }\n      const keypointsReshaped = tf.reshape(keypoints, [-1, 3]);\n      const rawCoords = await keypointsReshaped.array();\n      keypoints.dispose();\n      keypointsReshaped.dispose();\n      const coords = this.transformRawCoords(rawCoords, box, angle, rotationMatrix);\n      const nextBoundingBox = this.getBoxForHandLandmarks(coords);\n      this.updateRegionsOfInterest(nextBoundingBox, false /* force replace */, i);\n      const result = {\n        landmarks: coords,\n        confidence: flagValue,\n        box: {\n          topLeft: nextBoundingBox.startPoint,\n          bottomRight: nextBoundingBox.endPoint,\n        },\n      };\n      hands.push(result);\n    }\n    return hands;\n  }\n\n  // eslint-disable-next-line class-methods-use-this\n  calculateLandmarksBoundingBox(landmarks) {\n    const xs = landmarks.map((d) => d[0]);\n    const ys = landmarks.map((d) => d[1]);\n    const startPoint = [Math.min(...xs), Math.min(...ys)];\n    const endPoint = [Math.max(...xs), Math.max(...ys)];\n    return { startPoint, endPoint };\n  }\n\n  // Updates regions of interest if the intersection over union between\n  // the incoming and previous regions falls below a threshold.\n  updateRegionsOfInterest(box, forceUpdate, index) {\n    if (forceUpdate) {\n      this.regionsOfInterest[index] = [box];\n    } else {\n      const previousBox = this.regionsOfInterest[index][0];\n      let iou = 0;\n      if (previousBox != null && previousBox.startPoint != null) {\n        const [boxStartX, boxStartY] = box.startPoint;\n        const [boxEndX, boxEndY] = box.endPoint;\n        const [previousBoxStartX, previousBoxStartY] = previousBox.startPoint;\n        const [previousBoxEndX, previousBoxEndY] = previousBox.endPoint;\n        const xStartMax = Math.max(boxStartX, previousBoxStartX);\n        const yStartMax = Math.max(boxStartY, previousBoxStartY);\n        const xEndMin = Math.min(boxEndX, previousBoxEndX);\n        const yEndMin = Math.min(boxEndY, previousBoxEndY);\n        const intersection = (xEndMin - xStartMax) * (yEndMin - yStartMax);\n        const boxArea = (boxEndX - boxStartX) * (boxEndY - boxStartY);\n        const previousBoxArea = (previousBoxEndX - previousBoxStartX) * (previousBoxEndY - boxStartY);\n        iou = intersection / (boxArea + previousBoxArea - intersection);\n      }\n      this.regionsOfInterest[index][0] = iou > UPDATE_REGION_OF_INTEREST_IOU_THRESHOLD ? previousBox : box;\n    }\n  }\n\n  shouldUpdateRegionsOfInterest() {\n    return !this.regionsOfInterest || (this.regionsOfInterest.length === 0) || (this.runsWithoutHandDetector >= this.maxContinuousChecks);\n  }\n}\nexports.HandPipeline = HandPipeline;\n", "const tf = require('@tensorflow/tfjs');\nconst hand = require('./handdetector');\nconst keypoints = require('./keypoints');\nconst pipe = require('./pipeline');\n\nclass HandPose {\n  constructor(pipeline) {\n    this.pipeline = pipeline;\n  }\n\n  async estimateHands(input, config) {\n    this.maxContinuousChecks = config.skipFrames;\n    this.detectionConfidence = config.minConfidence;\n    this.maxHands = config.maxHands;\n    const image = tf.tidy(() => {\n      if (!(input instanceof tf.Tensor)) {\n        input = tf.browser.fromPixels(input);\n      }\n      return input.toFloat().expandDims(0);\n    });\n    const predictions = await this.pipeline.estimateHands(image, config);\n    image.dispose();\n    const hands = [];\n    if (!predictions) return hands;\n    for (const prediction of predictions) {\n      if (!prediction) return [];\n      const annotations = {};\n      for (const key of Object.keys(keypoints.MESH_ANNOTATIONS)) {\n        annotations[key] = keypoints.MESH_ANNOTATIONS[key].map((index) => prediction.landmarks[index]);\n      }\n      hands.push({\n        confidence: prediction.confidence || 0,\n        box: prediction.box ? [prediction.box.topLeft[0], prediction.box.topLeft[1], prediction.box.bottomRight[0] - prediction.box.topLeft[0], prediction.box.bottomRight[1] - prediction.box.topLeft[1]] : 0,\n        landmarks: prediction.landmarks,\n        annotations,\n      });\n    }\n    return hands;\n  }\n}\nexports.HandPose = HandPose;\n\nasync function loadAnchors(url) {\n  if (tf.env().features.IS_NODE) {\n    // eslint-disable-next-line global-require\n    const fs = require('fs');\n    const data = await fs.readFileSync(url.replace('file://', ''));\n    return JSON.parse(data);\n  }\n  return tf.util.fetch(url).then((d) => d.json());\n}\n\nasync function load(config) {\n  const [anchors, handDetectorModel, handPoseModel] = await Promise.all([\n    loadAnchors(config.detector.anchors),\n    tf.loadGraphModel(config.detector.modelPath, { fromTFHub: config.detector.modelPath.includes('tfhub.dev') }),\n    tf.loadGraphModel(config.skeleton.modelPath, { fromTFHub: config.skeleton.modelPath.includes('tfhub.dev') }),\n  ]);\n  const detector = new hand.HandDetector(handDetectorModel, anchors, config);\n  const pipeline = new pipe.HandPipeline(detector, handPoseModel, config);\n  const handpose = new HandPose(pipeline);\n  return handpose;\n}\nexports.load = load;\n", "/* eslint-disable indent */\n/* eslint-disable no-multi-spaces */\n\nexport default {\n  backend: 'webgl',          // select tfjs backend to use\n  console: true,             // enable debugging output to console\n  face: {\n    enabled: true,           // controls if specified modul is enabled\n                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion\n                             // (note: module is not loaded until it is required)\n    detector: {\n      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.\n                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.\n      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'\n      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance\n      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression\n      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    },\n    mesh: {\n      enabled: true,\n      modelPath: '../models/facemesh/model.json',\n      inputSize: 192,        // fixed value\n    },\n    iris: {\n      enabled: true,\n      modelPath: '../models/iris/model.json',\n      enlargeFactor: 2.3,    // empiric tuning\n      inputSize: 64,         // fixed value\n    },\n    age: {\n      enabled: true,\n      modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'\n                                                         // which determines training set for model\n      inputSize: 64,         // fixed value\n      skipFrames: 10,        // how many frames to go without re-running the detector\n    },\n    gender: {\n      enabled: true,\n      minConfidence: 0.8,    // threshold for discarding a prediction\n      modelPath: '../models/ssrnet-gender/imdb/model.json',\n    },\n    emotion: {\n      enabled: true,\n      inputSize: 64,         // fixed value\n      minConfidence: 0.5,    // threshold for discarding a prediction\n      skipFrames: 10,        // how many frames to go without re-running the detector\n      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel\n      modelPath: '../models/emotion/model.json',\n    },\n  },\n  body: {\n    enabled: true,\n    modelPath: '../models/posenet/model.json',\n    inputResolution: 257,    // fixed value\n    outputStride: 16,        // fixed value\n    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression\n  },\n  hand: {\n    enabled: true,\n    inputSize: 256,          // fixed value\n    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector\n                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis\n                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)\n    minConfidence: 0.5,      // threshold for discarding a prediction\n    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression\n    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression\n    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace\n    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance\n    detector: {\n      anchors: '../models/handdetect/anchors.json',\n      modelPath: '../models/handdetect/model.json',\n    },\n    skeleton: {\n      modelPath: '../models/handskeleton/model.json',\n    },\n  },\n};\n", "const tf = require('@tensorflow/tfjs');\nconst facemesh = require('./facemesh/facemesh.js');\nconst ssrnet = require('./ssrnet/ssrnet.js');\nconst emotion = require('./emotion/emotion.js');\nconst posenet = require('./posenet/posenet.js');\nconst handpose = require('./handpose/handpose.js');\nconst defaults = require('./config.js').default;\nconst app = require('../package.json');\n\nlet config;\n\n// object that contains all initialized models\nconst models = {\n  facemesh: null,\n  posenet: null,\n  handpose: null,\n  iris: null,\n  age: null,\n  gender: null,\n  emotion: null,\n};\nconst now = () => {\n  if (typeof performance !== 'undefined') return performance.now();\n  return parseInt(Number(process.hrtime.bigint()) / 1000 / 1000);\n};\n\nconst log = (...msg) => {\n  // eslint-disable-next-line no-console\n  if (config.console) console.log(...msg);\n};\n\n// helper function that performs deep merge of multiple objects so it allows full inheriance with overrides\nfunction mergeDeep(...objects) {\n  const isObject = (obj) => obj && typeof obj === 'object';\n  return objects.reduce((prev, obj) => {\n    Object.keys(obj || {}).forEach((key) => {\n      const pVal = prev[key];\n      const oVal = obj[key];\n      if (Array.isArray(pVal) && Array.isArray(oVal)) {\n        prev[key] = pVal.concat(...oVal);\n      } else if (isObject(pVal) && isObject(oVal)) {\n        prev[key] = mergeDeep(pVal, oVal);\n      } else {\n        prev[key] = oVal;\n      }\n    });\n    return prev;\n  }, {});\n}\n\nfunction sanity(input) {\n  if (!input) return 'input is not defined';\n  const width = input.naturalWidth || input.videoWidth || input.width || (input.shape && (input.shape[1] > 0));\n  if (!width || (width === 0)) return 'input is empty';\n  if (input.readyState && (input.readyState <= 2)) return 'input is not ready';\n  try {\n    tf.getBackend();\n  } catch {\n    return 'backend not loaded';\n  }\n  return null;\n}\n\nasync function detect(input, userConfig) {\n  config = mergeDeep(defaults, userConfig);\n\n  // sanity checks\n  const error = sanity(input);\n  if (error) {\n    log(error, input);\n    return { error };\n  }\n\n  // eslint-disable-next-line no-async-promise-executor\n  return new Promise(async (resolve) => {\n    // check number of loaded models\n    const loadedModels = Object.values(models).filter((a) => a).length;\n    if (loadedModels === 0) log('Human library starting');\n\n    // configure backend\n    if (tf.getBackend() !== config.backend) {\n      log('Human library setting backend:', config.backend);\n      await tf.setBackend(config.backend);\n      await tf.ready();\n    }\n    // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders\n    // let savedWebglPackDepthwiseConvFlag;\n    // if (tf.getBackend() === 'webgl') {\n    //   savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');\n    //  tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);\n    // }\n\n    // load models if enabled\n    if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);\n    if (config.body.enabled && !models.posenet) models.posenet = await posenet.load(config.body);\n    if (config.hand.enabled && !models.handpose) models.handpose = await handpose.load(config.hand);\n    if (config.face.enabled && config.face.age.enabled && !models.age) models.age = await ssrnet.loadAge(config);\n    if (config.face.enabled && config.face.gender.enabled && !models.gender) models.gender = await ssrnet.loadGender(config);\n    if (config.face.enabled && config.face.emotion.enabled && !models.emotion) models.emotion = await emotion.load(config);\n\n    const perf = {};\n    let timeStamp;\n\n    // run posenet\n    timeStamp = now();\n    tf.engine().startScope();\n    const poseRes = config.body.enabled ? await models.posenet.estimatePoses(input, config.body) : [];\n    tf.engine().endScope();\n    perf.body = Math.trunc(now() - timeStamp);\n\n    // run handpose\n    timeStamp = now();\n    tf.engine().startScope();\n    const handRes = config.hand.enabled ? await models.handpose.estimateHands(input, config.hand) : [];\n    tf.engine().endScope();\n    perf.hand = Math.trunc(now() - timeStamp);\n\n    // run facemesh, includes blazeface and iris\n    const faceRes = [];\n    if (config.face.enabled) {\n      timeStamp = now();\n      tf.engine().startScope();\n      const faces = await models.facemesh.estimateFaces(input, config.face);\n      perf.face = Math.trunc(now() - timeStamp);\n      for (const face of faces) {\n        // is something went wrong, skip the face\n        if (!face.image || face.image.isDisposedInternal) {\n          log('face object is disposed:', face.image);\n          continue;\n        }\n        // run ssr-net age & gender, inherits face from blazeface\n        timeStamp = now();\n        const ssrData = (config.face.age.enabled || config.face.gender.enabled) ? await ssrnet.predict(face.image, config) : {};\n        perf.agegender = Math.trunc(now() - timeStamp);\n        // run emotion, inherits face from blazeface\n        timeStamp = now();\n        const emotionData = config.face.emotion.enabled ? await emotion.predict(face.image, config) : {};\n        perf.emotion = Math.trunc(now() - timeStamp);\n        face.image.dispose();\n        // calculate iris distance\n        // iris: array[ bottom, left, top, right, center ]\n        const iris = (face.annotations.leftEyeIris && face.annotations.rightEyeIris)\n          ? Math.max(face.annotations.leftEyeIris[3][0] - face.annotations.leftEyeIris[1][0], face.annotations.rightEyeIris[3][0] - face.annotations.rightEyeIris[1][0])\n          : 0;\n        faceRes.push({\n          confidence: face.confidence,\n          box: face.box,\n          mesh: face.mesh,\n          annotations: face.annotations,\n          age: ssrData.age,\n          gender: ssrData.gender,\n          agConfidence: ssrData.confidence,\n          emotion: emotionData,\n          iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,\n        });\n      }\n      tf.engine().endScope();\n    }\n\n    // set depthwiseconv to original value\n    // tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);\n\n    // combine and return results\n    perf.total = Object.values(perf).reduce((a, b) => a + b);\n    resolve({ face: faceRes, body: poseRes, hand: handRes, performance: perf });\n  });\n}\n\nexports.detect = detect;\nexports.defaults = defaults;\nexports.config = config;\nexports.models = models;\nexports.facemesh = facemesh;\nexports.ssrnet = ssrnet;\nexports.posenet = posenet;\nexports.handpose = handpose;\nexports.tf = tf;\nexports.version = app.version;\n"],
+  "mappings": "oNAAA,kBCAA,kBCAA,0sDCAA,AAsBO,GAAM,IAAkB,KAClB,GAAkB,mBA0B7B,WAAoB,EAAgC,GAAhC,KAAA,QAAA,EAAgC,KAAA,UAAA,EAH5C,KAAA,KAAO,GAAI,SACX,KAAA,aAAe,EAIvB,SAAA,UAAA,IAAA,SAAI,GACF,MAAK,MAAK,KAAK,IAAI,IACjB,KAAK,UAAU,SAAS,KAAK,QAAS,GAEjC,KAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,IAAA,SAAI,EAAgB,GAClB,KAAK,eACL,KAAK,KAAK,IAAI,EAAQ,IAGxB,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,MAAK,KAAK,IAAI,IAGvB,EAAA,UAAA,OAAA,SAAO,GACL,YAAK,eACE,KAAK,KAAK,OAAO,IAG1B,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,cAEhB,mBAqBA,cACE,SAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,WACE,MAAO,GAAkB,eAE3B,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,GAAkB,gBAE3B,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAC5C,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAE3D,MAAO,GAAkB,SAE3B,EAAA,UAAA,OAAA,WACE,MAAO,GAAkB,WAG3B,EAAA,UAAA,eAAA,WACE,MAAO,GAAkB,mBAG3B,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAG1D,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAEnD,MAAO,GAAkB,qBAG3B,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,MAAO,GAAkB,UAE3B,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,MAAO,GAAkB,iBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,GAAkB,WAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAG3B,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,uBAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAE3B,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,GAAkB,WAG3B,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,MAAO,GAAkB,SAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAE3B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,MAAO,GAAkB,iBAG3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,MAAO,GAAkB,eAE3B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,MAAO,GAAkB,cAG3B,EAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,MAAO,GAAkB,WAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,MAAO,GAAkB,YAG3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,MAAO,GAAkB,QAG3B,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,MAAO,GAAkB,sBAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,MAAO,GAAkB,YAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,GAAkB,WAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,MAAO,GAAkB,WAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,WAAA,SAA6B,GAC3B,MAAO,GAAkB,eAG3B,EAAA,UAAA,QAAA,SAA0B,GACxB,MAAO,GAAkB,YAG3B,EAAA,UAAA,SAAA,SAA2B,GACzB,MAAO,GAAkB,aAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAE3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,MAAO,GAAkB,UAG3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAE3B,EAAA,UAAA,MAAA,SAAwB,GACtB,MAAO,GAAkB,UAG3B,EAAA,UAAA,IAAA,SAAsB,GACpB,MAAO,GAAkB,QAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAG3B,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,MAAO,GAAkB,yBAG3B,EAAA,UAAA,gBAAA,SAAgB,EAAiB,EAAkB,GAEjD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,wBAAA,SAAwB,EAAc,EAAkB,GAEtD,MAAO,GAAkB,4BAE3B,EAAA,UAAA,yBAAA,SAAyB,EAAa,EAAc,GAElD,MAAO,GAAkB,6BAE3B,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GACpC,MAAO,GAAkB,WAE3B,EAAA,UAAA,eAAA,SAAe,EAAc,EAAkB,GAE7C,MAAO,GAAkB,mBAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,EAAa,GAEtD,MAAO,GAAkB,oBAE3B,EAAA,UAAA,QAAA,SAAQ,EAAa,GACnB,MAAO,GAAkB,YAE3B,EAAA,UAAA,gBAAA,SAAgB,EAAc,EAAa,GACzC,MAAO,GAAkB,oBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SAAkB,EAAc,EAAa,GAC3C,MAAO,GAAkB,sBAE3B,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,MAAO,GAAkB,cAE3B,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAAa,GAC1C,MAAO,GAAkB,sBAG3B,EAAA,UAAA,QAAA,SAA0C,EAAM,GAE9C,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAkB,SAG3B,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,MAAO,GAAkB,QAG3B,EAAA,UAAA,UAAA,SAA4B,EAAM,GAChC,MAAO,GAAkB,cAG3B,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,MAAO,GAAkB,WAG3B,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GACpC,MAAO,GAAkB,cAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,MAAO,GAAkB,mBAG3B,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,mBAG3B,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,MAAO,GAAkB,2BAG3B,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,MAAO,GAAkB,0BAG3B,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,MAAO,GAAkB,kCAG3B,EAAA,UAAA,UAAA,SACI,EAAa,EAAyB,EACtC,EAA4B,EAC5B,GACF,MAAO,GAAkB,cAG3B,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,MAAO,GAAkB,iCAG3B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EAAuB,EAC3D,EAAc,EAAe,GAC/B,MAAO,GAAkB,YAG3B,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,MAAO,GAAkB,gBAG3B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,MAAO,GAAkB,WAG3B,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,MAAO,GAAkB,WAG3B,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,MAAO,GAAkB,sBAG3B,EAAA,UAAA,IAAA,SAAI,GACF,MAAO,GAAkB,QAE3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAE3B,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,MAAO,GAAkB,YAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAE3B,EAAA,UAAA,KAAA,SAAuB,GACrB,MAAO,GAAkB,SAG3B,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAC3C,MAAO,GAAkB,iBAI3B,EAAA,UAAA,MAAA,SAAwB,EAAU,EAAsB,GACtD,MAAO,GAAkB,UAG3B,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACF,MAAO,GAAkB,kBAG3B,EAAA,UAAA,KAAA,SAAK,GACH,MAAO,GAAkB,SAG3B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,MAAO,GAAkB,SAG3B,EAAA,UAAA,SAAA,SAAyB,GACvB,MAAO,GAAkB,aAG3B,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,GAAkB,cAG3B,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAkB,aAG3B,EAAA,UAAA,QAAA,WACE,MAAO,GAAkB,YAE7B,KAEA,WAA2B,GACzB,KAAM,IAAI,OACN,IAAI,EAAU,4HCzpBpB,AAoBA,GAAM,IAA4B,0BA6BhC,WAAmB,GAAA,KAAA,OAAA,EATX,KAAA,MAAe,GACf,KAAA,aAAwD,GAExD,KAAA,SAAkB,GAOxB,KAAK,mBAGP,SAAA,UAAA,YAAA,SAAY,EAAsB,GAChC,AAAI,KAAK,UAAY,MACnB,QAAQ,KACJ,YAAY,KAAK,aAAY,0BAC7B,kCAAiC,EAAQ,MAE/C,KAAK,aAAe,EACpB,KAAK,SAAW,GAGlB,EAAA,UAAA,aAAA,SACI,EAAkB,EAClB,GAKF,GAJA,KAAK,aAAa,GAAY,CAAC,aAAY,EAAE,QAAO,GAIhD,KAAK,SAAS,IAAa,MAC7B,GAAM,GAAY,KAAK,SAAS,GAChC,QAAQ,KACJ,qCAAqC,EAAQ,KAAK,EAAS,KAC/D,KAAK,IAAI,EAAU,KAIjB,EAAA,UAAA,SAAN,SAAe,qGACb,MAAI,KAAY,MAAK,MACnB,CAAA,EAAO,KAAK,MAAM,IAGpB,GAAA,KAAK,MAAM,EAAA,EAAY,CAAA,EAAM,KAAK,aAAa,YAA/C,SAAA,GAAuB,EAAA,OACvB,CAAA,EAAO,KAAK,MAAM,UAGpB,EAAA,UAAA,IAAA,SAAI,GACF,GAAI,IAAY,MAAK,MACnB,MAAO,MAAK,MAAM,GAGpB,GAAM,GAAY,KAAK,aAAa,GACpC,GAAI,YAAqB,SACvB,KAAM,IAAI,OACN,QAAQ,EAAQ,sEAItB,YAAK,MAAM,GAAY,EAEhB,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,QAAA,SAAQ,GACN,MAAO,MAAK,IAAI,IAGlB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,OAGd,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,uCAGd,EAAA,UAAA,IAAA,SAAI,EAAkB,GACpB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,mBAAmB,EAAQ,mCAEjC,KAAK,MAAM,GAAY,EACvB,AAAI,KAAK,aAAa,GAAU,SAAW,MACzC,KAAK,aAAa,GAAU,QAAQ,IAIhC,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAI,KAAK,aAAa,IAAa,KACjC,KAAM,IAAI,OACN,yBAAyB,EAAQ,oCAEvC,MAAO,MAAK,aAAa,GAAU,gBAGrC,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,MAAQ,OAAO,OAAO,GAAI,IAGjC,EAAA,UAAA,MAAA,WACE,KAAK,MAAQ,GACb,KAAK,SAAW,GAChB,KAAK,oBAGC,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,GAAI,MAAO,MAAK,QAAW,aACvB,MAAO,MAAK,OAAO,UAAa,aAChC,MAAO,MAAK,OAAO,SAAS,QAAW,YACzC,OAGF,GAAM,GAAY,GAAe,KAAK,OAAO,SAAS,QACtD,GAAI,KAA6B,IAC/B,GAAM,GAAY,EAAU,IAA2B,MAAM,KAC7D,EAAU,QAAQ,SAAA,GACV,GAAA,GAAA,EAAA,MAAA,KAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,SAAS,GAAO,GAAW,EAAK,OAI7C,iBAE+B,GAC7B,GAAM,GAAS,GACf,SAAY,QAAQ,8BAA+B,SAAC,UAAG,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GACrD,UAAY,EAAQ,EAAE,GAAI,EAAE,IACrB,EAAE,KAAK,OAET,EAGT,YACI,EAAiC,EAAc,GACjD,EAAO,mBAAmB,IAAS,mBAAmB,GAAS,IAGjE,YAAoB,EAAkB,GAEpC,GADA,EAAQ,EAAM,cACV,IAAU,QAAU,IAAU,QAChC,MAAO,KAAU,OACZ,GAAI,GAAG,EAAE,IAAY,EAC1B,MAAO,CAAC,EAEV,KAAM,IAAI,OACN,oCAAoC,EAAK,aAAa,EAAQ,mBAYlE,MAAO,GAAA,IAGE,EAAA,IAAmB,iBACO,GACnC,EAAA,IAAM,EC9MR,AAqBA,GAAI,kBAGF,GAAI,IAAmB,MAErB,GAAI,GAAE,OACN,GAAI,MAAQ,SAAY,YACtB,EAAK,eACI,MAAQ,SAAY,YAC7B,EAAK,eACI,MAAQ,UAAa,YAC9B,EAAK,gBACI,MAAQ,OAAU,YAC3B,EAAK,SAEL,MAAM,IAAI,OAAM,kCAElB,GAAkB,EAEpB,MAAO,IAIT,cACE,GAAM,GAAK,KACX,MAAI,GAAG,YAAc,MACnB,GAAG,WAAa,GAAI,MAEf,EAAG,uBAUiB,EAAa,GACxC,GAAM,GAAY,KAClB,GAAI,EAAU,IAAI,GAChB,MAAO,GAAU,IAAI,GAErB,GAAM,GAAY,IAClB,SAAU,IAAI,EAAK,GACZ,EAAU,IAAI,MCzCZ,IAAM,MAGN,GAAO,OAGP,GAAQ,QAGR,GAAM,MAGN,GAAO,OAGP,GAAM,MAON,GAAM,MAON,GAAS,SAMT,GAAS,SAMT,GAAO,OAGP,GAAQ,QAGR,GAAO,OAGP,GAAQ,QAGR,GAAQ,QAGR,GAAU,UASV,GAAkB,kBAQlB,GAAY,YAWZ,GAAoB,oBAUpB,GAAc,cAOd,GAAiB,iBASjB,GAAc,cAOd,GAAO,OAMP,GAAO,OAGP,GAAc,cAOd,GAAU,UAGV,GAAS,SAMT,GAAS,SAUT,GAAuB,uBASvB,GAAsB,sBAUtB,GAAS,SAST,GAAyB,yBAQzB,GAAwB,wBAMxB,GAAM,MAGN,GAAO,OAGP,GAAS,SAQT,GAAgB,gBAShB,GAAe,eAOf,GAAwB,wBAWxB,GACT,sCAIS,GACT,qCAIS,GAAO,OAGP,GAAa,aAQb,GAA0B,0BAI1B,GAA2B,2BAI3B,GAAM,MAGN,GAAM,MAGN,GAAU,UAGV,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAQ,QAGR,GAAM,MAGN,GAAO,OAOP,GAAgB,gBAGhB,GAAQ,QAGR,GAAW,WAGX,GAAiB,iBAOjB,GAAW,WAMX,GAAW,WAGX,GAAU,UAGV,GAAe,eAGf,GAAW,WAGX,GAAO,OAGP,GAAO,OAGP,GAAW,WAGX,GAAQ,QAGR,GAAQ,QAGR,GAAO,OAGP,GAAY,YAGZ,GAAW,WAMX,GAAM,MAGN,GAAQ,QAGR,GAAa,aAGb,GAAa,aAGb,GAAY,YAGZ,GAAa,aAMb,GAAM,MASN,GAAc,cASd,GAAM,MAON,GAAU,UAGV,GAAU,UASV,GAAkB,kBAUlB,GAAY,YAWZ,GAAoB,oBAWpB,GAAoB,oBASpB,GAAO,OAOP,GAAM,MAON,GAAU,UAGV,GAAM,MAGN,GAAW,WAGX,GAAS,SAGT,GAAW,WAGX,GAAsB,sBAStB,GAAsB,sBAUtB,GAAsB,sBAUtB,GAAW,WAGX,GAAS,SAQT,GAAQ,QAOR,GAAO,OAGP,GAAM,MAGN,GAAQ,QAGR,GAAO,OAOP,GAAQ,QAQR,GAAO,OAGP,GAAa,aAGb,GAAO,OAGP,GAAU,UAMV,GAAwB,wBAOxB,GAA4B,4BAI5B,GAAiB,iBAOjB,GAAqB,qBAGrB,GAAQ,QAGR,GAAU,UAMV,GAAQ,QAGR,GAAQ,QAGR,GAAY,YAMZ,GAAW,WAGX,GAAO,OAGP,GAAQ,QAMR,GAAM,MAGN,GAAO,OAGP,GAAO,OAGP,GAAU,UAGV,GAAW,WAGX,GAAO,OAGP,GAAM,MAON,GAAiB,iBAOjB,GAAS,SAOT,GAAU,UAMV,GAAoB,oBAGpB,GAAS,SAGT,GAAM,MAGN,GAAgB,gBAOhB,GAAe,eAaf,GAAM,MAGN,GAAO,OAGP,GAAO,OAMP,GAAO,OAOP,GAAY,YAMZ,GAAS,SAQT,GAAS,SAMT,GAAqB,qBAOrB,GAAY,YAMZ,GAAO,OAMP,GAAa,aASb,GAAmB,mBAQnB,GAAe,eAef,GAAc,cAgBd,GAAuB,uBC7zBpC,AAuBA,GAAM,IACF,GAAU,iBAAkB,WAAM,MAAA,IAAI,OACpC,GACF,GAAU,eAAgB,WAAM,MAAA,IAAI,mBAqEpC,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,MAAO,IAAe,IAAI,eAOA,GAC1B,MAAO,IAAa,IAAI,eAGW,GAInC,OAHM,GAAK,GAAe,UACpB,EAAyB,MAGvB,GAAA,GAAA,EAAA,OAAC,EAAA,EAAA,KAAM,EAAA,EAAA,MACb,GAAI,EACF,MAEK,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GACL,EAAA,EAAA,MAAA,KAAA,GACP,AAAI,IAAY,GACd,EAAO,KAAK,GAGhB,MAAO,eAcsB,GACtB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,YACb,EAAM,GAAQ,EAAY,GAChC,AAAI,GAAe,IAAI,IACrB,QAAQ,KACJ,eAAe,EAAU,iBACzB,KAAI,EAAW,4BAErB,GAAe,IAAI,EAAK,eAWO,GACxB,GAAA,GAAA,EAAA,WAEP,AAAI,GAAa,IAAI,IAGnB,CAAI,KAAM,QAAQ,UAChB,QAAQ,KAAK,gCAAgC,EAAU,MAG3D,GAAa,IAAI,EAAY,eAW3B,EAAoB,GACtB,GAAM,GAAM,GAAQ,EAAY,GAChC,GAAI,CAAC,GAAe,IAAI,GACtB,KAAM,IAAI,OACN,eAAe,EAAU,iBACzB,KAAI,EAAW,wBAErB,GAAe,OAAO,eAIW,GACjC,GAAI,CAAC,GAAa,IAAI,GACpB,KAAM,IAAI,OACN,iBAAiB,EAAU,mCAEjC,GAAa,OAAO,eAUlB,EAA+B,GACjC,GAAM,GAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,GAAM,GACF,OAAO,OAAO,GAAI,EAAc,CAAC,YAAa,IAClD,GAAe,KAInB,YAAiB,EAAoB,GACnC,MAAU,GAAW,IAAI,ECrN3B,YAkCwB,GAMtB,OAJI,GAAU,EAAM,OAChB,EAAO,EACP,EAAQ,EAEL,EAAU,GAEf,EAAS,KAAK,SAAW,EAAW,EAEpC,IAEA,EAAO,EAAM,GACb,EAAM,GAAW,EAAM,GACvB,EAAM,GAAS,cAKG,EAAa,EAAW,GAC5C,MAAO,MAAK,IAAI,EAAK,KAAK,IAAI,EAAG,gBAGD,GAChC,MAAO,GAAM,IAAM,EAAI,EAAM,EAAM,cAGjB,GAElB,OADI,GAAM,EACD,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,GAAO,EAAI,GAEb,MAAO,eAUmB,EAAW,GACrC,GAAM,GAAI,KAAK,SACf,MAAQ,GAAI,EAAM,GAAI,GAAK,cAID,EAAe,GAEzC,OADI,GAAS,EACJ,EAAI,EAAG,EAAI,EAAE,OAAQ,KAC5B,GAAM,GAAO,OAAO,EAAE,IAAM,OAAO,EAAE,IACrC,GAAU,EAAO,EAEnB,MAAO,cAkBc,EAAe,GACpC,GAAI,CAAC,EACH,KAAM,IAAI,OAAM,MAAO,IAAQ,SAAW,EAAM,iBAKhD,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EACI,GAAY,EAAQ,GACpB,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlC,GAC5B,EACI,GAAK,KACL,WAAM,MAAA,8EAwBR,EAA0B,EAAkB,GAI9C,GAJ4B,AAAA,IAAA,QAAA,GAAA,IAAkB,AAAA,IAAA,QAAA,GAAA,IAC9C,AAAI,GAAU,MACZ,GAAS,IAEP,MAAM,QAAQ,IAAQ,GAAa,IAAQ,CAAC,EAC9C,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAAQ,EAAI,GAAI,EAAQ,OAG1B,GAAO,KAAK,GAEd,MAAO,eAcqB,GAC5B,GAAI,EAAM,SAAW,EAEnB,MAAO,GAGT,OADI,GAAO,EAAM,GACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,GAAQ,EAAM,GAEhB,MAAO,eAGqB,GAC5B,MAAO,GAAM,SAAW,cAGE,EAAgB,GAC1C,GAAI,IAAO,EACT,MAAO,GAET,GAAI,GAAM,MAAQ,GAAM,KACtB,MAAO,GAGT,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,EAAG,GACf,MAAO,GAGX,MAAO,eAGa,GACpB,MAAO,GAAI,IAAM,cAGE,GAEnB,GAAK,KAAa,MAAQ,KAExB,MAAQ,MAAa,KAAK,GAE5B,GAAI,IAAM,SACR,MAAO,GACF,GAAI,IAAM,UACf,MAAO,GAEP,GAAM,GAAM,KAAK,IAAI,EAAI,GACzB,MAAQ,GAAM,GAAM,GAAM,eAIM,GAClC,GAAM,GAAQ,KAAK,KAAK,KAAK,KAAK,IAClC,MAAO,CAAC,EAAO,KAAK,KAAK,EAAO,gBAeI,GAEpC,OADM,GAAkB,GAAI,aAAY,GAC/B,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAgB,GAAK,EAEvB,UAAQ,GACD,cAGgB,EAAW,GAClC,MAAI,IAAQ,EAAE,OACL,EAEF,EAAI,IAAI,OAAO,EAAO,EAAE,oBAI7B,EAAwB,EACxB,GADwB,MAAA,KAAA,QAAA,GAAA,SAAW,GAAoB,MAAA,KAElD,GAAI,SAAc,SAAC,EAAS,GACjC,GAAI,GAAW,EAET,EAAQ,WACZ,GAAI,KACF,IACA,OAGF,IAEA,GAAM,GAAc,EAAQ,GAE5B,GAAI,GAAc,MAAQ,GAAY,GACpC,IACA,OAEF,WAAW,EAAO,IAGpB,kBAcA,EAAiB,GAInB,OAHI,GAAY,EACZ,EAAc,GAET,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,GAAI,EAAM,IAAM,EACd,GAAa,EAAM,WACV,EAAM,KAAO,IACtB,GAAI,IAAgB,GAClB,KAAM,OACF,yCACA,oBAAmB,EAAW,YAAY,IAEhD,EAAc,UACL,EAAM,GAAK,EACpB,KAAM,OAAM,gCAAgC,EAAM,GAAE,WAAW,GAInE,GAAI,IAAgB,IAClB,GAAI,EAAO,GAAK,IAAS,EACvB,KAAM,OAAM,QAAQ,EAAI,qCAAqC,GAE/D,MAAO,GAGT,GAAI,IAAc,EAChB,KAAM,OACF,qCAAqC,EAAK,+BAGhD,GAAI,EAAO,IAAc,EACvB,KAAM,OACF,oDACA,QAAO,EAAI,MAAM,IAGvB,GAAM,GAAW,EAAM,QACvB,SAAS,GAAe,EAAO,EACxB,cAIL,EAAuB,GACzB,GAAM,GAAO,EAAM,OAGnB,SAAO,GAAQ,KAAO,EAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAK,GAAG,OAAO,GAGzD,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,CAAC,GAAQ,EAAK,IACrC,WACI,MAAA,+CAA+C,EAAI,KAAK,EAAI,SAC5D,aAAY,KAGpB,EACI,EAAK,MAAM,SAAA,GAAM,MAAA,IAAM,KACvB,WAAM,MAAA,iDACF,aAAY,KAGb,EAAK,IAAI,SAAA,GAAK,MAAA,GAAI,EAAI,EAAO,EAAI,gBAIb,EAAiB,GAS5C,OAPM,GAAqB,GACrB,EAAqB,GACrB,EAAe,GAAQ,MAAQ,MAAM,QAAQ,IAAS,EAAK,SAAW,EACtE,EAAQ,GAAQ,MAAQ,EAC1B,KACA,GAAe,EAAM,GAAO,OAC5B,EAAI,EACC,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,GAAQ,MACV,GAAI,EAAK,KAAO,GAAK,EAAM,KAAO,EAChC,KAAM,IAAI,OACN,sBAAsB,EAAC,mBAAmB,EAAM,GAAE,cAExD,AAAK,GAAK,IAAM,MAAQ,EAAK,GAAK,IAAM,EAAM,KAAO,GACnD,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAEhB,AAAI,EAAK,IAAM,GACb,IAGJ,AAAI,EAAM,KAAO,GACf,GAAS,KAAK,EAAM,IACpB,EAAS,KAAK,IAGlB,MAAO,CAAC,SAAQ,EAAE,SAAQ,eAIxB,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAAU,GACZ,GAAI,GAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,SACnB,EAAS,GAAI,OAAgB,OAE7B,MAAM,IAAI,OAAM,qBAAqB,GAEvC,MAAO,eAIL,EAA+B,GACjC,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAC1B,KAAM,OAAM,oBAAoB,EAAK,4BAA4B,EAAG,kBAM7C,GAC3B,MAAO,KAAU,QAAU,IAAU,aAAe,IAAU,WAC1D,IAAU,SAAW,IAAU,qBAOL,EAAmB,GACjD,MAAI,KAAY,aAGZ,KAAY,WAAa,IAAY,aAGrC,IAAY,SAAW,IAAY,WAAa,IAAY,aALvD,GAQL,MAAY,QAAU,IAAY,oBAMX,GAC3B,MAAO,aAAa,eAAgB,YAAa,aAC7C,YAAa,wBAGa,GAC9B,GAAI,IAAU,WAAa,IAAU,QACnC,MAAO,GACF,GAAI,IAAU,YACnB,MAAO,GACF,GAAI,IAAU,OACnB,MAAO,GAEP,KAAM,IAAI,OAAM,iBAAiB,eAUA,GACnC,GAAI,GAAO,KACT,MAAO,GAET,GAAI,GAAQ,EACZ,SAAI,QAAQ,SAAA,GAAK,MAAA,IAAS,EAAE,SACrB,cAIgB,GACvB,MAAO,OAAO,IAAU,UAAY,YAAiB,oBAG7B,GACxB,MAAO,OAAO,IAAU,sBAGD,GACvB,MAAO,OAAO,IAAU,qBAGC,GACzB,MAAI,OAAM,QAAQ,GACT,GAAW,EAAO,IAEvB,YAAkB,cACb,UACE,YAAkB,aAAc,YAAkB,YACpD,QACE,GAAS,GACX,UACE,GAAS,GACX,SACE,GAAU,GACZ,OAEF,sBAGkB,GACzB,MAAO,CAAC,CAAE,IAAK,EAAE,aAAe,EAAE,MAAQ,EAAE,mBAGf,EAAc,GAC3C,OAAS,GAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,GAAI,EAAO,IAAM,EACf,MAAO,GAGX,MAAO,eAGsB,GAC7B,GAAM,GAAO,EAAM,OACnB,GAAI,EAAO,EACT,MAAO,GAKT,GAAM,GAAU,GAAI,OAAM,EAAO,GACjC,EAAQ,EAAO,GAAK,EAAM,EAAO,GACjC,OAAS,GAAI,EAAO,EAAG,GAAK,EAAG,EAAE,EAC/B,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAM,EAAI,GAE1C,MAAO,eAOL,EAAiB,GACnB,MAAI,KAAU,SACL,GAAa,GAGf,GAAa,CAAC,GAAQ,eAGF,EAAe,GAC1C,GAAI,IAAU,SACZ,KAAM,IAAI,OAAM,6CASlB,GAPA,AAAI,MAAM,QAAQ,IAChB,GAAI,GAAQ,IAGd,AAAI,KAAM,QAAQ,UAChB,GAAyB,EAAe,GAEtC,GAAmB,EAAG,GACxB,MAAO,GAET,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,QAEnB,OADM,GAAO,GAAI,YAAY,EAAe,QACnC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,AAAI,KAAK,MAAO,EAAe,MAAQ,GACrC,GAAK,GAAK,GAGd,MAAO,OAEP,MAAM,IAAI,OAAM,qBAAqB,GAIzC,YAA2B,EAAgB,EAAiB,GAC1D,GAAM,GAAM,GAAI,OAChB,GAAI,EAAM,SAAW,EAEnB,OADM,GAAI,EAAM,GACP,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,EAAE,EAAS,OAMtB,QAHM,GAAI,EAAM,GACV,EAAO,EAAM,MAAM,GACnB,EAAM,EAAK,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IACjC,EAAI,EAAG,EAAI,EAAG,IACrB,EAAI,GAAK,GAAkB,EAAS,EAAI,EAAK,EAAM,GAGvD,MAAO,eAIqB,EAAiB,GAC7C,GAAI,EAAM,SAAW,EAEnB,MAAO,GAAE,GAEX,GAAM,GAAO,EAAM,OAAO,SAAC,EAAK,GAAM,MAAA,GAAM,IAC5C,GAAI,IAAS,EAEX,MAAO,GAET,GAAI,IAAS,EAAE,OACb,KAAM,IAAI,OAAM,IAAI,EAAK,mCAAmC,EAAE,OAAM,KAGtE,MAAO,IAAkB,EAAG,EAAO,GAGrC,YAA4B,EAAe,GACzC,MAAQ,aAAa,eAAgB,IAAU,WAC1C,YAAa,aAAc,IAAU,SACrC,YAAa,aAAc,IAAU,mBAIxC,EAAc,GAEhB,OADM,GAAQ,GAAoB,EAAM,GAC/B,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAM,GAAK,EAEb,MAAO,eAIL,EAAc,GAChB,GAAI,GAAS,MAAQ,IAAU,WAAa,IAAU,YACpD,MAAO,IAAI,cAAa,GACnB,GAAI,IAAU,QACnB,MAAO,IAAI,YAAW,GACjB,GAAI,IAAU,OACnB,MAAO,IAAI,YAAW,GAEtB,KAAM,IAAI,OAAM,qBAAqB,eAUrC,EAAiB,GACnB,GAAM,GAAO,EAAM,OAAO,SAAC,EAAM,GAAS,MAAA,GAAO,GAAM,GACvD,GAAI,GAAS,MAAQ,IAAU,UAC7B,MAAO,IAAc,EAAO,GAAI,cAAa,IACxC,GAAI,IAAU,QACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IACtC,GAAI,IAAU,OACnB,MAAO,IAAc,EAAO,GAAI,YAAW,IAE3C,KAAM,IAAI,OAAM,qBAAqB,iBAgBvC,MAAO,MAAM,SAAS,kBAG2B,GACjD,EAAM,QAAQ,SAAA,GACZ,EACI,OAAO,UAAU,IAAY,GAAW,EACxC,WACI,MAAA,mEACA,WAAU,EAAK,sBAqBvB,EAAc,GAChB,MAAO,MAAM,SAAS,MAAM,EAAM,eAWP,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,SACtC,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAG,eAWL,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,SAC9C,EAAW,GAAY,QAChB,KAAM,SAAS,OAAO,EAAO,eAYlC,EAAgB,EAAc,GAChC,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,EAAQ,GAAK,EAAK,GAE7B,MAAO,eAYL,EAAe,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GACF,GAAI,IAAS,EAClB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,EAAQ,IACrC,GAAS,EAAK,GAAK,EAAQ,GAE7B,SAAK,EAAK,OAAS,GAAK,EACjB,22BC3xBT,AA+BA,GAAA,IAAA,WACE,WAAoB,EAAoC,GAApC,KAAA,aAAA,EAAoC,KAAA,OAAA,EACtD,AAAI,GAAU,MACZ,MAAK,OAAS,GAAI,KAItB,SAAA,UAAA,cAAA,SAAc,EAAoB,EAAwB,GAQxD,OANI,GACE,EAAsB,WAC1B,EAAU,KAEN,EAAQ,KAAK,aAAa,KAAK,cAE5B,GACP,GAAM,GAAS,EAAQ,GAGvB,EAAO,OAAO,KAAK,SAAA,GACjB,GAA0B,EAAY,EAAO,MAAO,MAL/C,EAAI,EAAG,EAAI,EAAQ,OAAQ,MAA3B,GAST,GAAM,GAAgB,CACpB,WAAU,EACV,QAAO,EACP,OAAM,EACN,OAAQ,EAAM,KAAK,SAAA,GAAU,MAAA,GAAO,WACpC,UAAW,EAAM,KACb,SAAA,GAAU,MAAA,GAAO,qBAAuB,KACpC,EAAO,sBACP,MAEV,MAAO,IAGT,EAAA,UAAA,iBAAA,SAAiB,GAAjB,GAAA,GAAA,KACS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,OAAQ,EAAA,EAAA,OAAQ,EAAA,EAAA,UAE5C,EAAQ,QAAQ,SAAA,GACd,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,IAAY,KAAK,SAAA,GACnD,EAAK,OAAO,iBACR,EAAY,EAAQ,EAAe,GAAI,EAAe,GAAI,EAC1D,EAAe,SAI3B,iBAGI,EAAsB,EAAU,GAClC,GAAI,IAAU,UAEZ,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAM,EAAK,GACjB,GAAI,MAAM,IAAQ,CAAC,SAAS,GAE1B,eAAQ,KAAK,SAAS,EAAG,sBAAsB,EAAU,KAClD,GAGX,MAAO,GAGT,GAAA,IAAA,WAAA,cACE,SAAA,UAAA,iBAAA,SACI,EAAc,EAAgB,EAC9B,EAAgC,EAChC,GACF,GAAM,GAAO,MAAO,IAAW,SAAW,GAAiB,EAAM,KAAM,GAC7B,EAAO,MAC3C,EAAa,GAAc,EAAM,IACjC,EAAO,EAAO,KACd,EAAO,EAAO,KACd,EAAQ,GAAc,EAAO,MAAM,WAAY,IACjD,EAAyB,GAE7B,OAAW,KAAQ,IACjB,GAAM,GAAQ,EAAO,GACrB,GAAI,GAAS,MAGX,GAAM,GAAa,EAAM,OAAS,EAAO,MACnC,EAAY,EAAW,OAC7B,GACO,EAAI,KAAK,EAAS,KAAK,GAAY,EAAI,EAAa,IAAE,KAIjE,QAAQ,IACJ,KAAK,EAAU,MAAO,EAAI,MAAO,EAAI,KAAK,EAAK,MAAO,EAAI,MACtD,EAAsB,MAAO,EACjC,mBAAoB,YAAa,aAAc,gBAC/C,eAAgB,qBAExB,KCjIA,YA4CI,EAAkB,EAAc,GAKlC,OAFM,GAA8C,GAC9C,EAA0C,GACvC,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,EAAa,EAAG,GAAG,IAAM,GAG3B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GACZ,EAAa,EAAK,OACxB,OAAW,KAAa,IAItB,OAHM,GAAQ,EAAW,GAErB,EAAgB,GACX,EAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAa,EAAM,KACrB,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAa,EAAO,IAAM,KACzD,EAAgB,GAChB,EAAW,EAAK,IAAM,GACtB,MAIJ,GAAI,EACF,OAMN,GAAM,GAAgD,GACtD,EAAe,EAAE,IAAM,GAGvB,OAFM,GAAwC,GAErC,EAAI,EAAK,OAAS,EAAG,GAAK,EAAG,IAKpC,OAJM,GAAO,EAAK,GACZ,EAAa,EAAK,OAGf,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,IACvC,GAAI,EAAe,EAAK,QAAQ,GAAG,KACjC,OAAW,KAAa,GACtB,EAAe,EAAW,GAAW,IAAM,GAC3C,EAAS,EAAK,IAAM,GAEtB,MAON,OADM,GAA2B,GACxB,EAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAM,GAAO,EAAK,GAElB,GAAI,EAAW,EAAK,KAAO,EAAS,EAAK,KAEvC,GAAM,GAA8C,GACpD,OAAW,KAAa,GAAK,QAC3B,GAAM,GAAY,EAAK,OAAO,GAC9B,AAAI,EAAa,EAAU,KACzB,GAAa,GAAa,GAK9B,GAAM,GAAa,OAAO,OAAO,GAAI,GACrC,EAAW,OAAS,EACpB,EAAW,QAAU,EAAK,QAE1B,EAAa,KAAK,IAItB,MAAO,eAWL,EACA,EAA0B,EAC1B,GAEF,mBAAS,GACP,GAAM,GAAO,EAAa,GAEpB,EAAgB,GAYtB,GAXA,EAAK,QAAQ,QAAQ,SAAA,GACnB,GAAM,GAAa,EAA6B,EAAE,IAClD,AAAI,GAAc,KAChB,EAAI,KAAK,GAIT,EAAI,KAAK,QAIT,EAAK,UAAY,KACnB,KAAM,IAAI,OACN,wDACA,QAAO,EAAK,WAAU,MAI5B,GAAM,GAAiB,EAAK,SAAS,cAE1B,GACT,GAAI,CAAE,KAAa,IACjB,KAAM,IAAI,OACN,iCAAiC,EAAS,KAC1C,+BAA8B,OAAO,KAAK,GAAe,MAI/D,GAAM,GAAK,EAAK,WAAM,MAAA,GAAe,OACrC,GAAI,EAAG,QAAU,UACf,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BAChB,GAAS,wCAAwC,EAAG,MAAK,MAElE,GAAM,GAAI,EAAK,OAAO,GACtB,GAAI,CAAC,GAAiB,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,OACN,4BACI,EAAK,WAAU,2BACnB,KAAI,EAAS,gBAAgB,EAAG,MAAK,4BACrC,4BAA2B,EAAE,MAAK,MAGxC,GAAI,EAA6B,EAAE,KAAO,KACxC,EAA6B,EAAE,IAAM,OAErC,GAAM,GAAc,EAA6B,EAAE,IACnD,EAA6B,EAAE,IAAM,EAAI,EAAa,GACtD,EAAY,YA7BhB,OAAW,KAAa,GAAK,SAAlB,IAxBJ,EAAI,EAAa,OAAS,EAAG,GAAK,EAAG,MAArC,GCvIX,AAqBA,GAAM,IAAwB,GAExB,GAA6B,EAE7B,GAAwB,cAG1B,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAU,GAAe,GACzB,EAAY,GAAwB,EAAM,EAAO,EAAO,GACxD,EAAO,EAAM,OACb,EAAY,GAAkB,EAAM,EAAO,EAAO,EAAS,GAC3D,EAAQ,CAAC,UACf,MAAI,IACF,GAAM,KAAK,YAAY,GACvB,EAAM,KAAK,WAAW,GACtB,EAAM,KAAK,aAAa,EAAK,KAC7B,EAAM,KAAK,cAEb,EAAM,KAAK,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,IAAG,KAAK;IACxC,EAAM,KAAK;GAGpB,YACI,EAA2B,EAAiB,EAC5C,GACF,GAAM,GAAI,GAAc,GAClB,EAAU,EAAQ,EAAQ,OAAS,GACnC,EAAY,GAAI,OAAM,GAAS,KAAK,GACpC,EAAO,EAAM,OACb,EACF,IAAU,YAAc,GAAoB,GAAQ,EAExD,GAAI,EAAO,EACT,OAAS,GAAM,EAAG,EAAM,EAAI,EAAS,IAEnC,OADM,GAAS,EAAM,EACZ,EAAI,EAAG,EAAI,EAAS,IAC3B,EAAU,GAAK,KAAK,IAChB,EAAU,GACV,GAAY,EAAe,EAAS,GAAI,EAAG,GAAO,QAI5D,MAAO,GAGT,YACI,EAAqC,EAAa,GACpD,GAAI,GACJ,MAAI,OAAM,QAAQ,GAChB,EAAY,WAAW,EAAI,GAAG,QAAQ,KAAuB,MACtD,YAAW,EAAI,GAAG,QAAQ,KAAuB,KACnD,AAAI,GAAS,GAClB,EAAS,IAAI,EAAG,IACX,AAAI,IAAU,OACnB,EAAS,GAAgB,GAEzB,EAAS,WAAW,EAAI,QAAQ,KAAwB,WAGnD,GAAS,EAAQ,GAG1B,YAAyB,GACvB,MAAO,KAAM,EAAI,QAAU,OAG7B,YACI,EAA2B,EAAiB,EAC5C,EAAmB,EAAqB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAM,GAAoB,IAAU,YAAc,EAAI,EAEhD,EAAO,EAAM,GACb,EAAO,EAAM,OACnB,GAAI,IAAS,GACX,GAAI,IAAU,aACZ,GAAM,GAAe,GAAoB,GACzC,MAAO,CAAC,GAAY,EAAa,GAAI,EAAG,IAE1C,MAAI,KAAU,OACL,CAAC,GAAgB,EAAK,KAExB,CAAC,EAAK,GAAG,YAGlB,GAAI,IAAS,GACX,GAAI,EAAO,IACT,GAAM,GAAgB,GAA6B,EAE/C,EAAY,MAAM,KAClB,EAAK,MAAM,EAAG,IACd,EAAW,MAAM,KAAqC,EAAK,MAC1D,GAAO,IAA8B,EACtC,EAAO,IACX,MAAI,KAAU,aACZ,GAAY,GAAoB,GAChC,EAAW,GAAoB,IAE1B,CACL,IACA,EAAU,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAChD,KAAK,MACV,UACA,EACK,IACG,SAAC,EAAG,GAAM,MAAA,IACN,EAAG,EAAU,EAAO,GAA6B,GAAI,KAC5D,KAAK,MACV,KAGJ,GAAM,GACF,IAAU,YAAc,GAAoB,GACpB,MAAM,KAAoB,GAEtD,MAAO,CACL,IACA,EAAY,IAAI,SAAC,EAAG,GAAM,MAAA,IAAY,EAAG,EAAU,GAAI,KAClD,KAAK,MACV,KAKJ,GAAM,GAAW,EAAM,MAAM,GACvB,EAAa,EAAQ,MAAM,GAC3B,EAAS,EAAQ,GAAK,EACtB,EAAkB,GACxB,GAAI,EAAO,IACT,OAAS,GAAI,EAAG,EAAI,GAA4B,KAC9C,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,KAEN,EAAM,KAAK,OACX,OAAS,GAAI,EAAO,GAA4B,EAAI,EAAM,KACxD,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,SAGnB,QAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAQ,EAAI,EACZ,EAAM,EAAQ,EACpB,EAAM,KAAI,MAAV,EAAc,GACV,EAAK,MAAM,EAAO,GAAM,EAAU,EAAO,EAAY,EACrD,IAAM,EAAO,IAGrB,GAAM,GAAM,IAAS,EAAI,IAAM,GAC/B,EAAM,GAAK,IAAM,EAAM,GAAK,EAC5B,OAAS,GAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,EAAM,GAAK,IAAM,EAAM,GAAK,EAG9B,OADI,GAAa;EACR,EAAI,EAAG,EAAI,EAAM,IACxB,GAAc;EAEhB,SAAM,EAAM,OAAS,GACjB,IAAM,EAAM,EAAM,OAAS,GAAK,IAAO,GAAS,GAAK,GAClD,EAGT,YAA6B,GAG3B,OADM,GAAyC,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,GAAK,EACpC,EAAc,KAAK,CAAC,EAAK,GAAI,EAAK,EAAI,KAExC,MAAO,GCnMT,kBA6CE,WAAY,EAA2B,EAAU,GAAjD,GAAA,GAAA,KAIE,GAJqC,KAAA,MAAA,EACrC,KAAK,MAAQ,EAAM,QACnB,KAAK,KAAO,GAAmB,GAE3B,GAAU,MACZ,GAAM,GAAI,EAAO,OACjB,EACI,IAAM,KAAK,KACX,WAAM,MAAA,qBAAqB,EAAC,6BACxB,2BAA0B,EAAK,KAAI,QAE7C,GAAI,IAAU,YACZ,KAAM,IAAI,OACN,8JAIN,KAAK,OAAS,GAAU,GAAuB,EAAO,KAAK,MAC3D,KAAK,QAAU,GAAe,GAWhC,SAAA,UAAA,IAAA,SAAI,UAAJ,GAAA,KAA8B,EAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,EAAA,GAAA,UAAA,GAC5B,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAEV,EACI,EAAK,SAAW,KAAK,KACrB,WAAM,MAAA,uCAAuC,EAAK,OAAM,UACpD,oBAAmB,EAAK,KAAI,OAEpC,GAAM,GAAQ,KAAK,WAAW,GAC9B,KAAK,OAAO,GAAS,GAUvB,EAAA,UAAA,IAAA,kBAAI,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACF,AAAI,EAAK,SAAW,GAClB,GAAO,CAAC,IAGV,OADI,GAAI,EACU,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,EAAM,GAAK,GAAO,KAAK,MAAM,IAC/B,GAAM,GAAM,qCAAqC,EAAI,KACjD,mBAAkB,KAAK,OAC3B,KAAM,IAAI,OAAM,GAElB,IAGF,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,MAAK,OAAO,IAGrB,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,GAAK,GAGd,OADI,GAAQ,EAAK,EAAK,OAAS,GACtB,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,GAAS,KAAK,QAAQ,GAAK,EAAK,GAElC,MAAO,IAGT,EAAA,UAAA,WAAA,SAAW,GACT,GAAI,KAAK,OAAS,EAChB,MAAO,GACF,GAAI,KAAK,OAAS,EACvB,MAAO,CAAC,GAGV,OADM,GAAiB,GAAI,OAAM,KAAK,MAAM,QACnC,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAK,GAAK,KAAK,MAAM,EAAQ,KAAK,QAAQ,IAC1C,GAAS,EAAK,GAAK,KAAK,QAAQ,GAElC,SAAK,EAAK,OAAS,GAAK,EACjB,GAGT,OAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAY,WAAW,KAAK,OAAQ,KAAK,MAAO,KAAK,QAGhE,KA8BI,GAAiC,KAEjC,GAAuB,iBAaM,GAC/B,GAAY,cAQe,GAC3B,GAAY,mBA4DZ,WAAY,EAAoB,EAAiB,EAAgB,GAXjE,KAAA,KAAO,GAoIG,KAAA,mBAAqB,GAxH7B,KAAK,MAAQ,EAAM,QACnB,KAAK,MAAQ,GAAS,UACtB,KAAK,KAAO,GAAmB,GAC/B,KAAK,QAAU,GAAe,GAC9B,KAAK,OAAS,EACd,KAAK,GAAK,EACV,KAAK,SAAY,KAAK,KAAO,EAAI,KAAK,KAAK,WAAa,SAG1D,cAAA,eAAI,EAAA,UAAA,OAAI,KAAR,WACE,MAAO,MAAK,MAAM,wCAQd,EAAA,UAAA,OAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAU,OAAO,KAAK,MAAO,KAAK,MAAY,UAOvD,EAAA,UAAA,WAAA,WACE,MAAO,IAAU,OAAO,KAAK,MAAO,KAAK,MAAY,KAAK,aAStD,EAAA,UAAA,MAAN,2GACe,MAAA,CAAA,EAAM,KAAK,eAAlB,SAAO,EAAA,OACb,CAAA,EAAO,GAAc,KAAK,MAAO,UASnC,EAAA,UAAA,UAAA,WACE,MAAO,IAAc,KAAK,MAAO,KAAK,aASlC,EAAA,UAAA,KAAN,mHACE,MAAK,kBACC,EAAO,KAAY,KAAK,KAAK,QAC/B,KAAK,QAAU,SACH,CAAA,EAAM,GADlB,CAAA,EAAA,UACI,EAAQ,EAAA,OACd,IACE,MAAA,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,IAAkB,YACxC,GACA,KAAM,IAAI,OACN,kHAIR,MAAA,CAAA,EAAO,SAST,EAAA,UAAA,SAAA,WACE,KAAK,kBACL,GAAM,GAAO,KAAY,SAAS,KAAK,QACvC,GAAI,KAAK,QAAU,SACjB,IACE,MAAQ,GAAsB,IAAI,SAAA,GAAK,MAAA,IAAkB,WAEzD,GACA,KAAM,IAAI,OACN,iGAIR,MAAO,IAIH,EAAA,UAAA,MAAN,2GACE,YAAK,kBACQ,CAAA,EAAM,KAAY,KAAK,KAAK,gBACzC,MADM,GAAO,EAAA,OACT,KAAK,QAAU,SACjB,CAAA,EAAO,GAEP,CAAA,EAAO,GAAI,YAAY,EAAoB,eAS/C,EAAA,UAAA,QAAA,WACE,GAAI,KAAK,WACP,OAEF,KAAY,cAAc,MAC1B,KAAK,mBAAqB,IAI5B,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,oDAGd,EAAA,UAAA,gBAAA,WACE,GAAI,KAAK,WACP,KAAM,IAAI,OAAM,wBAYpB,EAAA,UAAA,MAAA,SAAM,GAAA,MAAA,KAAA,QAAA,GAAA,IACG,GAAU,MAAM,KAAM,IAO/B,EAAA,UAAA,MAAA,WACE,YAAK,kBACE,GAAU,MAAM,OAQzB,EAAA,UAAA,SAAA,SAAS,GAAA,AAAA,IAAA,QAAA,GAAA,IACP,GAAM,GAAO,KAAK,WAClB,MAAO,IAAe,EAAM,KAAK,MAAO,KAAK,MAAO,IAGtD,EAAA,UAAA,KAAA,SAAqB,GACnB,YAAK,kBACE,GAAU,KAAK,KAAW,IAEnC,EAAA,UAAA,SAAA,SAAS,EAAkB,EAAe,GAAjC,MAAA,KAAA,QAAA,GAAA,IACP,KAAK,kBACE,KAAY,aAAa,KAAM,EAAW,EAAM,IAG3D,KACA,OAAO,eAAe,EAAQ,OAAO,YAAa,CAChD,MAAO,SAAC,GAMN,MAAO,CAAC,CAAC,GAAY,EAAS,MAAQ,MAAQ,EAAS,UAAY,MAC/D,EAAS,iBAAmB,2BAoCiB,GAAA,EAAA,GAGnD,WACI,EAAgC,EAAoB,EACpD,GAFJ,GAAA,GAGE,EAAA,KAAA,KACI,EAAa,MAAO,EAAa,MAAO,EAAa,OAAQ,IAAS,KAHxC,SAAA,UAAA,EAIlC,EAAK,KAAO,IAWd,SAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAS,QAAU,KAAK,MAC1B,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,GAAI,CAAC,GAAiB,EAAS,MAAO,KAAK,OACzC,KAAM,IAAI,OACN,2BAA2B,EAAS,MAAK,SACzC,oBAAmB,KAAK,MAAK,iBAEnC,KAAY,cAAc,MAC1B,KAAK,OAAS,EAAS,OACvB,KAAY,OAAO,KAAM,OAG3B,EAAA,UAAA,QAAA,WACE,KAAY,gBAAgB,MAC5B,KAAK,mBAAqB,IAE9B,GAvCqD,GAyCrD,OAAO,eAAe,GAAU,OAAO,YAAa,CAClD,MAAO,SAAC,GACN,MAAO,aAAoB,IAAU,EAAS,QAAU,MACpD,EAAS,iBAAkB,aCnhBnC,AAgEA,AAAA,UAAY,GACV,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,KACA,EAAA,GAAA,OAPU,EAAA,MAAA,GAAA,KAAI,KAqBhB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,QACA,EAAA,UAAA,cAJG,IAAA,IAAiB,KAOtB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,QACA,EAAA,KAAA,OACA,EAAA,UAAA,cAJG,IAAA,IAAgB,KAOrB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,UACA,EAAA,MAAA,UACA,EAAA,KAAA,UACA,EAAA,UAAA,cAJG,IAAA,IAAmB,KAOxB,GAAK,IAAL,AAAA,UAAK,GACH,EAAA,QAAA,YACA,EAAA,MAAA,YACA,EAAA,KAAA,YACA,EAAA,UAAA,cAJG,IAAA,IAAqB,KAO1B,GAAM,IAAgB,CACpB,QAAW,GACX,MAAS,GACT,KAAQ,GACR,UAAa,gBAGY,EAAiB,GAC1C,GAAI,IAAU,UAAY,IAAU,UAClC,GAAI,IAAU,UAAY,IAAU,SAClC,MAAO,SAET,KAAM,IAAI,OAAM,kBAAkB,EAAK,SAAS,GAElD,MAAO,IAAc,GAAO,eAIH,GACzB,MAAO,IAAW,EAAM,SCpI1B,YAsBiD,EAAM,GACrD,GAAI,EAAE,QAAU,EAAE,MAChB,MAAO,CAAC,EAAG,GAEb,GAAM,GAAQ,GAAW,EAAE,MAAO,EAAE,OACpC,MAAO,CAAC,EAAE,KAAK,GAAQ,EAAE,KAAK,gBAGC,EAAW,GAC1C,EACI,EAAE,QAAU,EAAE,MACd,WAAM,MAAA,2BAA2B,EAAE,MAAK,QACpC,YAAW,EAAE,MAAK,oCAGG,EAAgB,GAC7C,MAAO,GAAW,KAAK,SAAA,GAAK,MAAA,GAAE,KAAO,EAAO,iBAeR,GACpC,GAAM,GAAiB,GACjB,EAAO,GAAI,KACjB,UAAoB,EAAQ,EAAM,GAC3B,EAGT,YACI,EAA4B,EAAgB,GAC9C,GAAI,GAAa,KACf,OAEF,GAAI,YAAqB,IACvB,EAAK,KAAK,GACV,OAEF,GAAI,CAAC,GAAW,GACd,OAGF,GAAM,GAAW,EACjB,OAAW,KAAK,IACd,GAAM,GAAM,EAAS,GACrB,AAAK,EAAK,IAAI,IACZ,GAAK,IAAI,GACT,GAAoB,EAAK,EAAM,KAMrC,YAAoB,GAClB,MAAO,OAAM,QAAQ,IAAQ,MAAO,IAAQ,kHCrF9C,AAmFA,GAAA,IAAA,WAAA,aAEE,KAAA,oBAAwC,GAExC,KAAA,eAAiB,EACjB,KAAA,SAAW,EACX,KAAA,WAAa,EACb,KAAA,iBAAmB,EACnB,KAAA,eAAiB,EAMjB,KAAA,cAAgB,EAGhB,KAAA,YAAc,EAId,KAAA,WAA2B,GAK3B,KAAA,kBAA8B,GAC9B,KAAA,YAAc,EAEd,KAAA,WAAa,GAAI,SAQjB,KAAA,UAAY,GACZ,KAAA,cACI,CAAC,SAAU,EAAG,WAAY,EAAG,UAAW,EAAG,QAAS,GAAI,OAAQ,MAEpE,SAAA,UAAA,QAAA,WACE,OAAW,KAAgB,MAAK,oBAC9B,KAAK,oBAAoB,GAAc,WAG7C,KAEA,GAAA,WAgBE,WAAmB,GAAA,KAAA,IAAA,EAbnB,KAAA,SAA0C,GAC1C,KAAA,gBAKI,GAKI,KAAA,qBAAuB,EAG7B,KAAK,MAAQ,GAAI,IAGb,SAAA,UAAA,MAAN,iHACE,GAAI,KAAK,oBAAsB,KAC7B,MAAA,CAAA,EAAO,KAAK,mBAAmB,KAAK,eAEtC,GAAI,KAAK,iBAAmB,KAC1B,MAAA,CAAA,GAEI,EAAiB,KAAK,oBAEnB,EAAI,yBAAG,GAAI,EAAe,OAC3B,GAAc,EAAe,GACnB,CAAA,EAAM,KAAK,kBAAkB,GAAa,UAFnB,CAAA,EAAA,gBAEjC,GAAU,EAAA,OACZ,EACF,CAAA,EAAM,KAAK,WAAW,IADpB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,UALuC,wBAS3C,KAAM,IAAI,OACN,gFAIN,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,GAAI,KAAK,oBAAsB,KAC7B,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,uHAIlC,GAAI,KAAK,iBAAmB,MACpB,GAAA,GAAA,KAAA,kCAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UACb,GAAI,EACF,KAAM,IAAI,OACN,iCAAiC,EAAI,uHAI3C,KAAK,WAAW,GAElB,MAAO,MAAK,iDAGd,EAAA,UAAA,aAAA,WACE,MAAO,QAAO,KAAK,KAAK,kBAG1B,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,CAAE,KAAe,MAAK,UAGxB,GAAI,IAAe,MAAK,iBACf,GAAA,GAAA,KAAA,kBAAA,GAAA,UACP,GAAI,EAEF,MAAO,UAGT,OAAO,MAGX,MAAO,MAAK,SAAS,IAGvB,EAAA,UAAA,mBAAA,SAAmB,GAEjB,MAAM,KAAe,MAAK,gBAGnB,KAAK,gBAAgB,GAAa,QAFhC,MAKX,EAAA,UAAA,gBAAA,SACI,EACA,EACA,GACF,MADE,AAAA,KAAA,QAAA,GAAA,GACE,IAAe,MAAK,gBACtB,SAAQ,KACD,EAAW,sEAEX,IAET,MAAK,gBAAgB,GAAe,CAAC,QAAO,EAAE,SAAQ,GAC/C,KAGH,EAAA,UAAA,WAAN,SAAiB,2GACf,GAAI,KAAK,gBAAgB,IAAgB,KACvC,KAAM,IAAI,OAAM,iBAAiB,EAAW,iCAE9C,MAAK,YAAc,EACf,KAAK,SAAS,IAAgB,KAChC,MAAK,gBAAkB,KACjB,EAAuB,KAAK,kBAAkB,GAA7C,EAAO,EAAA,QAAE,EAAS,EAAA,UACV,EAAY,CAAA,EAAM,GAAlB,CAAA,EAAA,IAHb,CAAA,EAAA,UAGyB,SAAA,EAAA,oBAAgB,EAAA,mBAC3C,GADM,EAAM,EACR,CAAC,EACH,MAAA,CAAA,EAAO,qBAGX,YAAK,gBAAkB,KAAK,SAAS,GACrC,KAAK,yBAEL,KAAK,SAAW,GAAI,IAAS,KAAK,iBAElC,CAAA,EAAO,UAGD,EAAA,UAAA,uBAAR,WAAA,GAAA,GAAA,KACQ,EAAU,GAAqB,KAAK,aAC1C,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,WAAa,MACtB,EAAO,UAAU,EAAK,oBAKpB,EAAA,UAAA,yBAAR,SAAiC,GAAjC,GAAA,GAAA,KACQ,EAAU,GAAqB,GACrC,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAO,aAAe,MACxB,EAAO,YAAY,EAAK,SAAS,OAW/B,EAAA,UAAA,kBAAR,SAA0B,GAA1B,GAAA,GAAA,KAEQ,EAAuB,KAAK,gBAAgB,GAClD,GAAI,GAAwB,KAC1B,KAAM,IAAI,OACN,6BAA6B,EAAW,4BAG9C,IACE,GAAM,GAAU,EAAqB,UAMrC,GAAI,GAAW,CAAE,aAAmB,MAC7B,MAAO,GAAQ,MAAS,YAC7B,GAAM,GAAY,EAAE,KAAK,qBACnB,EACF,EACK,KAAK,SAAA,GAEJ,MAAI,GAAY,EAAK,qBACZ,GAET,GAAK,SAAS,GAAe,EAC7B,EAAK,mBAAqB,KACnB,MAER,MAAM,SAAA,GAEL,MAAI,GAAY,EAAK,sBAGrB,GAAK,mBAAqB,KAC1B,QAAQ,KACJ,6BAA6B,EAAW,WAC5C,QAAQ,KAAK,EAAI,OAAS,EAAI,UACvB,KAEjB,YAAK,mBAAqB,EACnB,CAAC,QAAO,EAAE,UAAW,QAE5B,aAAK,SAAS,GAAe,EACtB,CAAC,QAAS,GAAM,UAAW,UAE7B,GACP,eAAQ,KAAK,6BAA6B,EAAW,WACrD,QAAQ,KAAK,EAAI,OAAS,EAAI,SACvB,CAAC,QAAS,GAAO,UAAW,MAIvC,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAE,KAAe,MAAK,iBACxB,KAAM,IAAI,OAAS,EAAW,kCAEhC,AAAI,KAAK,cAAgB,GAAe,KAAK,oBAAsB,MAGjE,KAAK,uBAGP,AAAI,IAAe,MAAK,UACtB,MAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,IAGvB,MAAO,MAAK,gBAAgB,GAG5B,AAAI,KAAK,cAAgB,GACvB,MAAK,mBAAqB,KAC1B,KAAK,YAAc,KACnB,KAAK,gBAAkB,OAInB,EAAA,UAAA,kBAAR,WAAA,GAAA,GAAA,KACE,GAAI,OAAO,KAAK,KAAK,iBAAiB,SAAW,EAC/C,KAAM,IAAI,OAAM,iCAElB,MAAO,QAAO,KAAK,KAAK,iBAAiB,KAAK,SAAC,EAAW,GAExD,MAAO,GAAK,gBAAgB,GAAG,SAC3B,EAAK,gBAAgB,GAAG,YAIxB,EAAA,UAAA,gCAAR,WAIE,OAFM,GAAiB,KAAK,oBAEnB,EAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAc,EAAe,GAC7B,EAAA,KAAA,kBAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAChB,GAAI,GAAa,EACf,MAAO,CAAC,KAAM,EAAa,UAAS,GAGxC,KAAM,IAAI,OACN,2EAIN,EAAA,UAAA,SAAA,SAAS,EAAwB,GAC/B,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAa,EAAK,QAClB,EAAS,KAAK,SAAS,GAG7B,EAAW,YAAY,GACvB,EAAK,QAAU,EACf,EAAQ,KAAK,EAAQ,EAAQ,EAAK,MAAO,EAAK,OAC9C,AAAI,KAAK,0BAGP,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,MAIvE,EAAA,UAAA,KAAA,SAAgC,EAA6B,GAA7D,GAAA,GAAA,KAEM,EAAe,KACnB,GAAI,GAAM,MAER,GAAI,MAAO,IAAa,WACtB,KAAM,IAAI,OAAM,uCAElB,EAAK,OAGL,GAAI,MAAO,IAAa,UAAY,CAAE,aAAoB,SACxD,KAAM,IAAI,OACN,kFAGN,GAAI,MAAO,IAAO,WAChB,KAAM,IAAI,OACN,kFAGN,EAAO,EAIT,GAAI,GACJ,MAAO,MAAK,UACR,WAAM,MAAA,GAAK,WAAW,IAAO,WAAM,MAAA,GAAK,SAAS,IAAS,WACxD,SAAS,IACT,AAAI,YAAkB,UACpB,QAAQ,MAAM,2CAET,KAIP,EAAA,UAAA,UAAR,SAAqB,EAAmB,EAAiB,GACvD,IACA,IACE,GAAM,GAAM,IACZ,WACO,QACA,GACP,UACM,IAKF,EAAA,UAAA,aAAR,WACE,MAAO,GAAO,gBAIR,EAAA,UAAA,eAAR,WACE,MAAO,GAAO,kBAYR,EAAA,UAAA,MAAR,SAAc,GACZ,GAAM,GAAI,KAAK,qBAAqB,EAAE,OAAQ,EAAE,MAAO,EAAE,OACnD,EAAS,CAAC,EAAC,GACX,EAAO,SAAC,GAAe,MAAC,CAC5B,EAAG,WACD,GAAM,GAAQ,UACR,EAAa,CAAC,EAAG,GACjB,EAAQ,CAAC,MAAK,GAEpB,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAC5B,EAAoC,KAAiB,GACrD,MAGF,EAAkB,GACxB,YAAK,YAAY,KAAK,MAAM,YAAY,KAAM,EAAQ,CAAC,GAAI,EAAM,EAAO,IACjE,GAgBT,EAAA,UAAA,UAAA,SACI,EAAoB,EAAwB,EAC5C,EAAyB,GAC3B,GAAM,GAAoB,KACpB,EAAsB,KAI5B,MAAO,MAAK,cACR,EAAa,EAAQ,EAAe,EAAY,EAAO,EACvD,IAGE,EAAA,UAAA,uBAAR,WACE,MAAO,MAAK,IAAI,QAAQ,YAGlB,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAkB,KAAK,QAAQ,aAGjC,EAAmB,EACvB,EAAS,QAAQ,SAAA,GAGf,GAAqB,EAAK,QAAU,YAAc,EAAI,IAQxD,GAAM,GACF,KAAK,MAAM,kBAAkB,KAAK,MAAM,kBAAkB,OAAS,GACjE,EACF,EAAkB,EAAmB,EAAmB,EAC5D,GAAI,EAAgB,EAClB,KAAM,IAAI,OACN,YAAY,KAAK,YAAW,iCAC5B,KAAI,EAAa,6BAA6B,EAAU,OAQhE,EAAA,UAAA,cAAA,SACI,EAA6B,EAC7B,EACA,EAAqB,EAAsB,EAC3C,GAJJ,GAAA,GAAA,KAKM,EACA,EAAkB,GAChB,EAAW,KAAK,WACtB,AAAI,GAAc,MAChB,GACI,KAAK,MAAM,aAAe,KAAO,KAAK,MAAM,YAAY,KAAO,IAGrE,GAAM,GAAoB,KAAK,MAAM,SAC/B,EAAqB,KAAK,MAAM,WAEtC,AAAI,KAAK,0BACP,KAAK,MAAM,kBAAkB,KAAK,GAGpC,GAAI,GACE,EAAS,GAAU,EAAY,KAAK,aACtC,EACJ,GAAI,GAAU,KACZ,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAO,WAAW,CAAC,OAAM,EAAE,MAAK,EAAE,QAAS,EAAK,UACtD,GAAM,GAAW,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC7C,AAAI,EAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAE3D,GAAM,GAAa,EAAS,IACxB,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,MACb,MAAA,GAAK,qBAAqB,EAAQ,EAAO,KAMjD,GAAI,GACF,GAAI,GACA,EAAK,sBAAsB,EAAY,EAAQ,GACnD,GAAI,GAAiB,MAKnB,AAAI,GAAiB,MACnB,GAAgB,IAElB,GAAM,GAAa,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAC7D,EAAiB,IAAgB,IAAI,QAAQ,OAAO,GAEtD,EAAQ,EAAK,2BAA2B,GAE1C,MAAO,SAGT,GAAM,GAAyB,SAAC,GAI9B,GAAI,CAAC,EACH,OAEF,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,OAGrD,EAAa,WACX,GAAM,GAAmB,EAAK,QAAQ,aACtC,EAAM,EAAK,KAAK,WAAM,MAAA,GAAY,EAAK,QAAS,KAChD,GAAM,GAAQ,MAAM,QAAQ,GAAO,EAAM,CAAC,GAC1C,MAAI,GAAK,0BACP,EAAK,sBAAsB,EAAY,EAAkB,GAEpD,GAKX,GAAI,GACJ,YAAK,UACD,WAAM,MAAA,GAAK,MAAM,eAAe,WAAM,MAAA,GAAK,MAAM,eAAe,WAC9D,AAAI,CAAC,EAAK,IAAI,QAAQ,UAAY,CAAC,EAAK,MAAM,UAC5C,EAAU,IAEV,GAAgB,EAAK,SAAS,cAC1B,EAAY,EAAQ,WAAM,MAAA,OAC9B,AAAI,EAAK,IAAI,QAAQ,UACnB,EAAK,SAAS,iBAAiB,GAEjC,EAAU,EAAc,WAIhC,AAAI,GACF,KAAK,YACD,EAAY,EAAQ,EAAS,EAAe,EAAO,GAGzD,AAAI,KAAK,MAAM,WACb,KAAK,MAAM,cAAc,QAAQ,KAAK,CACpC,KAAM,EACN,WAAY,KAAK,MAAM,SAAW,EAClC,mBAAoB,KAAK,MAAM,SAC/B,aAAc,KAAK,MAAM,WAAa,EACtC,qBAAsB,KAAK,MAAM,WACjC,YAAa,OAAO,KAAK,GAAQ,IAC7B,SAAA,GAAO,MAAA,GAAO,IAAQ,KAAO,EAAO,GAAK,MAAQ,OACrD,aAAc,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,QACvC,aAAc,EAAc,OAC5B,UAAW,EAAc,YAGrB,MAAM,QAAQ,GAAO,EAAU,EAAQ,IAQzC,EAAA,UAAA,2BAAR,SAAmC,GAAnC,GAAA,GAAA,KACQ,EAAQ,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAK,KAAK,EAAK,MAAM,MACzD,MAAO,IAaD,EAAA,UAAA,sBAAR,SACI,EAAoB,EACpB,GACF,GAAM,GAAa,GAAY,GAC/B,GAAI,GAAc,MAChB,GAAM,GAAyB,EAAW,cAAgB,GACpD,EAA2B,EAAW,eAAiB,GAIzD,EAAkB,OACtB,AAAI,EAAW,cACb,GACI,MAAM,QAAQ,GACd,WAAM,MAAA,2DAEV,EAAqB,OAAO,KAAK,GAAQ,IAAI,SAAC,GAAQ,MAAA,GAAO,MAE7D,EAAqB,EAAa,IAAI,SAAC,GAAc,MAAA,GAAO,KAG9D,GAAM,GACF,EAAQ,OAAO,SAAC,EAAG,GAAM,MAAA,GAAc,KAE3C,MAAO,GAAmB,OAAO,GAInC,MAAO,OAQT,EAAA,UAAA,WAAA,SACI,EAAoB,EAAiB,EACrC,GACF,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,iDAElB,EAAQ,GAAS,UACjB,EAAU,GAAW,KAAK,QAC1B,GAAI,GAAc,EAClB,AAAI,IAAU,UAAY,GAAc,EAAO,KAC7C,GAAe,EAAoB,IAAI,SAAA,GAAK,MAAA,IAAkB,MAEhE,GAAM,GAAS,EAAQ,MAAM,EAAa,EAAO,GAC3C,EAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAIhD,GAHA,KAAK,OAAO,EAAG,GAGX,IAAU,UACZ,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACjC,EAAW,GAAqB,GACtC,KAAK,MAAM,UAAY,EAAW,EAAK,MACvC,EAAK,MAAQ,EAEf,MAAO,IAQT,EAAA,UAAA,qBAAA,SACI,EAAgB,EAAiB,EACjC,GACF,EAAQ,GAAS,UACjB,GAAM,GAAI,GAAI,GAAO,EAAO,EAAO,EAAQ,KAAK,gBAChD,YAAK,OAAO,EAAG,GACR,GAGT,EAAA,UAAA,aAAA,SACI,EAAsB,EAAkB,EACxC,GADsB,AAAA,IAAA,QAAA,GAAA,IAExB,EAAO,GAAQ,KAAK,iBAAiB,WACrC,AAAI,GAAS,MAAQ,IAAU,EAAa,OAC1C,GAAe,EAAa,KAAK,IAEnC,GAAM,GAAI,GAAI,IAAS,EAAc,EAAW,EAAM,KAAK,gBAC3D,GAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,KAC5C,KAAM,IAAI,OAAM,sBAAsB,EAAE,KAAI,2BAE9C,YAAK,MAAM,oBAAoB,EAAE,MAAQ,EACzC,KAAK,OAAO,EAAG,KAAK,SACb,GAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAM,GAAW,KAAK,MAAM,WAAW,IAAI,EAAE,QACzC,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,SACpC,EAKJ,GAJA,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAET,IAAa,GACf,KAAK,MAAM,iBAIX,GAAI,GAAQ,EACZ,AAAI,EAAE,QAAU,aAAe,EAAE,QAAU,UACzC,GAAQ,EAAE,KAAO,GAAqB,EAAE,QAE1C,KAAK,MAAM,WAAW,IAAI,EAAE,OAAQ,CAClC,QAAS,GAAW,KAAK,QACzB,MAAO,EAAE,MACT,MAAO,EAAE,MACT,MAAK,EACL,SAAU,IAEZ,KAAK,MAAM,UAAY,EAGzB,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,WAEpC,AAAM,YAAa,KACjB,KAAK,MAAM,IAIf,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,CAAC,KAAK,MAAM,WAAW,IAAI,EAAE,QAC/B,OAGF,KAAK,MAAM,aACX,AAAI,EAAE,QAAU,UACd,KAAK,MAAM,mBAEb,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,EAAE,QACnC,EAAW,EAAK,SAEtB,AAAI,GAAY,EAGd,CAAI,EAAE,QAAU,aACd,MAAK,MAAM,UAAY,EAAK,OAE9B,KAAK,MAAM,iBAEX,EAAK,QAAQ,YAAY,EAAE,QAC3B,KAAK,MAAM,WAAW,OAAO,EAAE,SAE/B,KAAK,MAAM,WAAW,IAAI,EAAE,QAAQ,YAOxC,EAAA,UAAA,iBAAA,WACE,OAAW,KAAW,MAAK,MAAM,qBAC/B,GAAM,GAAI,KAAK,MAAM,oBAAoB,GACzC,KAAK,gBAAgB,KAIzB,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,cAAc,GACnB,AAAI,KAAK,MAAM,oBAAoB,EAAE,OAAS,MAC5C,MAAO,MAAK,MAAM,oBAAoB,EAAE,OAI5C,EAAA,UAAA,OAAA,WACE,GAAM,GAAO,KAAK,QAAQ,SAC1B,SAAK,WAAa,KAAK,MAAM,WAC7B,EAAK,eAAiB,KAAK,MAAM,eACjC,EAAK,SAAW,KAAK,MAAM,SAC3B,AAAI,KAAK,MAAM,iBAAmB,GAChC,GAAK,WAAa,GAClB,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,IAEjB,EAAK,QAAQ,KACT,0EAGC,GAGH,EAAA,UAAA,QAAN,SAAc,iHAEZ,YAAK,MAAM,UAAY,GAEjB,EAAa,KAAK,MAAM,SACxB,EAAkB,KAAK,MAAM,WAEnC,KAAK,MAAM,cAAc,QAAU,GACnC,EAAA,KAAK,MAAM,cAAuB,CAAA,EAAM,YAAxC,EAAyB,OAAS,EAAA,OAElC,KAAK,MAAM,UAAY,GAEvB,KAAK,MAAM,cAAc,UAAY,KAAK,IAAG,MAAR,KAC9B,KAAK,MAAM,cAAc,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,sBACnD,KAAK,MAAM,cAAc,SAAW,KAAK,MAAM,SAAW,EAC1D,KAAK,MAAM,cAAc,WACrB,KAAK,MAAM,WAAa,MACP,EAAA,KAAK,MAAM,cAAc,+BAAzB,GAAA,EAAA,OAAV,GAAM,EAAA,GACf,EAAA,EAAsB,CAAA,EAAM,EAAO,eADgB,CAAA,EAAA,UACnD,SAAO,aAAe,EAAA,OACtB,EAAA,EAAmB,CAAA,EAAM,EAAO,kBAAhC,EAAO,UAAY,EAAA,wBAFA,wBAIrB,MAAA,CAAA,EAAO,KAAK,MAAM,qBAGpB,EAAA,UAAA,SAAA,WACE,MAAO,MAAK,MAAM,cAAgB,GAAK,KAAK,MAAM,cAAgB,GAG5D,EAAA,UAAA,YAAR,SACI,EAAoB,EAAwB,EAC5C,EAAyB,EAAiB,GAF9C,GAAA,GAAA,KAGQ,EACF,CAAC,GAAI,KAAK,MAAM,iBAAkB,WAAU,EAAE,OAAM,EAAE,QAAO,EAAE,MAAK,GAElE,EAAa,GAAY,GAC/B,AAAI,GAAc,MAChB,GAAgB,EAAW,UAE7B,AAAI,GAAiB,MACnB,GAAS,SAAW,SAAC,GAGnB,SAAM,EAAI,IAAI,SAAC,EAAI,GACjB,GAAI,GAAM,MACR,GAAM,GAAS,EAAQ,GACjB,EAAO,GAAyB,EAAO,KAAM,EAAO,OAC1D,MAAO,GAAK,WAAW,EAAM,EAAO,MAAO,EAAO,OAEpD,MAAO,KAIF,EAAc,EAAI,OAAS,EAAI,EAAM,EAAI,GAAI,EAAO,KAG/D,KAAK,MAAM,WAAW,KAAK,IAG7B,EAAA,UAAA,KAAA,SAAuB,GACrB,SAAO,KAAO,GACP,GAGD,EAAA,UAAA,UAAR,WACE,AAAI,KAAK,MAAM,gBAAkB,GAC/B,MAAK,MAAM,WAAa,IAE1B,KAAK,MAAM,iBAGL,EAAA,UAAA,QAAR,WACE,KAAK,MAAM,iBAOb,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAwB,CAC5B,MAAO,GACP,KAAM,gBACN,GAAI,KAAK,MAAM,eAEjB,AAAI,GACF,GAAU,KAAO,GAEnB,KAAK,MAAM,WAAW,KAAK,GAC3B,KAAK,MAAM,YAAc,GAO3B,EAAA,UAAA,SAAA,SAAS,GAMP,OANF,GAAA,KACQ,EAAyB,GAAsB,GAC/C,EACF,GAAI,KAAI,EAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,MAGrC,EAAI,EAAG,EAAI,KAAK,MAAM,YAAY,MAAM,OAAQ,KACvD,GAAM,GAAS,KAAK,MAAM,YAAY,MAAM,GAC5C,AAAI,CAAC,EAAO,MAAQ,CAAC,EAA0B,IAAI,EAAO,KACxD,EAAO,UAIX,GAAM,GAAW,KAAK,MAAM,WAAW,MACvC,KAAK,MAAM,YAAc,KAAK,MAAM,WAAW,SAAW,EACtD,KACA,KAAK,MAAM,WAAW,KAAK,MAAM,WAAW,OAAS,GAGzD,EAAuB,QAAQ,SAAA,GAG7B,AAAI,CAAC,EAAO,MAAQ,EAAO,UAAY,EAAS,IAC9C,EAAK,MAAM,MAWjB,EAAA,UAAA,UAAA,SACI,EAAY,EAAc,EAC1B,GAFJ,GAAA,GAAA,KAKE,GAHE,AAAA,IAAA,QAAA,GAAA,IACF,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,8CACrB,GAAM,MAAQ,EAAG,QAAU,UAC7B,KAAM,IAAI,OAAM,0CAA0C,EAAG,MAAK,KAGpE,GAAM,GAAI,KAAK,UACX,WAAM,MAAA,GAAK,aAAa,WAAM,MAAA,GAAK,WACnC,WAAM,MAAA,GAAK,KAAK,UAAW,KAE/B,EACI,YAAa,GACb,WAAM,MAAA,mDAEV,GAAM,GAAe,GAAqB,KAAK,MAAM,WAAY,EAAI,GACrE,GAAI,CAAC,GAAoB,EAAa,SAAW,GAAK,EAAG,OAAS,EAChE,KAAM,IAAI,OACN,uIAKN,MAAO,MAAK,KAAK,WAAY,WAC3B,GAAM,GAAuD,GAC7D,EAAuB,EAAE,IAAO,GAAc,GAAK,EAAE,OAGrD,GACI,EAAwB,EAExB,SAAA,GAAK,MAAA,GAAK,KAAK,IAEf,IACJ,GAAM,GAAQ,EAAG,IAAI,SAAA,GAAK,MAAA,GAAuB,EAAE,MAEnD,MAAI,GAAK,MAAM,gBAAkB,GAG/B,GAAK,MAAM,WAAW,QAAQ,SAAA,GAC5B,OAAqB,GAAA,EAAA,EAAA,EAAK,MAAL,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,EAAO,aAGX,EAAK,MAAM,WAAa,MAEnB,CAAC,MAAO,EAAG,MAAK,MAI3B,EAAA,UAAA,WAAA,SAA6B,GAA7B,GAAA,GAAA,KAEE,SACI,GAAgB,GAChB,WAAM,MAAA,sDACH,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACN,EACI,EAAO,MAAM,SAAA,GAAK,MAAA,aAAa,KAC/B,WAAM,MAAA,qEAGV,GAAI,GAIE,EAA2B,GACjC,SAAO,QAAQ,SAAC,EAAO,GACrB,EAAS,GAAK,IAET,EAAK,cACR,SAAC,EAAG,GACF,SAAM,EAAC,MAAA,OAAQ,EAAM,OAAA,CAAE,KACvB,EACI,EAAI,gBAAiB,GACrB,WAAM,MAAA,+FAEV,EACI,GAAgB,EAAI,UACpB,WAAM,MAAA,qGAEH,EAAI,OAEb,EACA,SAAC,EAAO,GACN,GAAM,GAAU,EAAI,SAAS,EAAI,GAC3B,EACF,MAAM,QAAQ,GAAW,EAAU,CAAC,GACxC,EACI,EAAM,SAAW,EAAO,OACxB,WAAM,MAAA,wKAGV,EACI,EAAM,MAAM,SAAA,GAAK,MAAA,aAAa,KAC9B,WAAM,MAAA,yIAGV,GAAM,GAAyC,GAC/C,SAAM,QAAQ,SAAC,EAAM,GACnB,EAAQ,GAAK,WAAM,MAAA,MAEd,MAKjB,EAAA,UAAA,SAAA,SAAS,GAEP,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,SAAS,IAE/B,EAAA,UAAA,KAAA,SAAK,GAEH,GAAM,GAAO,KAAK,MAAM,WAAW,IAAI,GACvC,MAAO,GAAK,QAAQ,KAAK,IAGrB,EAAA,UAAA,KAAN,SAAW,qGACH,SAAQ,KACK,CAAA,EAAM,KAAK,QAAQ,KAAK,WAArC,SAAa,EAAA,OACnB,EAAW,OAAS,KAAQ,EAC5B,CAAA,EAAO,SASD,EAAA,UAAA,MAAR,SAAgC,GAC9B,MAAI,MAAK,MAAM,aAAe,MAC5B,GAAO,QAAU,KAAK,MAAM,YAAY,GACxC,KAAK,MAAM,YAAY,MAAM,KAAK,IAG7B,GAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAOpB,EAAA,UAAA,MAAA,WAEE,KAAK,uBAEL,KAAK,MAAM,UACX,KAAK,IAAI,QACT,KAAK,MAAQ,GAAI,IAEjB,OAAW,KAAe,MAAK,SAC7B,KAAK,yBAAyB,GAC9B,KAAK,SAAS,GAAa,UAC3B,MAAO,MAAK,SAAS,GAEvB,KAAK,YAAc,KACnB,KAAK,gBAAkB,KACvB,KAAK,mBAAqB,MA/sBb,EAAA,aAAe,EAKf,EAAA,eAAiB,EA4sBlC,KAEA,YAAc,GACZ,GAAM,GAAS,GAAmB,GAAc,GAAQ,WACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,yBAIxC,GAAM,GAAK,KACX,GAAI,EAAG,WAAa,MAClB,GAAM,GAAc,GAAI,IAAY,GACpC,EAAG,UAAY,GAAI,IAAO,GAE5B,UAAqB,EAAG,UAAU,KAIlC,GAAiB,WAAM,MAAA,GAAG,YACnB,EAAG,UAGL,GAAM,GAAS,iBAQF,EAAW,GAE7B,GAAM,GAAS,CAAC,EAAC,EAAE,EAAC,GACpB,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,EAAG,GAC3B,SAAK,CAAC,EAAG,IACF,GACN,EAAgC,KAAqB,ICzrC1D,AAkBA,cACE,MAAO,OAAO,YAAc,aAAe,WAAa,mBAIxD,GAAI,MAEF,GAAM,GAAI,UAAU,WAAa,UAAU,QAAW,OAAe,MAErE,MAAO,2TACK,KAAK,IAEb,0kDACK,KAAK,EAAE,OAAO,EAAG,IAE5B,MAAO,iBAIP,MAAQ,OAAO,SAAW,aAAe,OAAO,UAAY,MAEvD,MAAO,oBAAsB,6DCvCpC,AAqBA,GAAM,IAAM,KAOZ,GAAI,aAAa,QAAS,WAAM,MAAA,IAAO,SAAA,GACrC,AAAI,GACF,QAAQ,KACJ,iJAOR,GAAI,aAAa,aAAc,WAAM,MAAA,QAGrC,GAAI,aACA,UACA,WAAM,MAAC,OAAO,UAAY,aACrB,MAAO,SAAQ,UAAa,aAC5B,MAAO,SAAQ,SAAS,MAAS,cAG1C,GAAI,aACA,YACA,WAAM,MAAA,OAAO,YAAc,aAAe,WAAa,MACnD,UAAU,WAAa,MAAQ,SAAS,KAAK,UAAU,YACvD,aAAa,KAAK,UAAU,UAMpC,GAAI,aAAa,OAAQ,WAAM,MAAA,KAM/B,GAAI,aACA,qCAAsC,WAAM,MAAA,IAAI,QAAQ,WAG5D,GAAI,aAAa,+BAAgC,WAAM,MAAA,KAGvD,GAAI,aAAa,UAAW,WAAM,MAAA,KCvElC,YAuB2B,EAAiB,GAC1C,GAAI,GAAwB,EAE5B,GAAI,GAAa,GACf,MAAO,KAAU,SAAW,GAAK,CAAC,EAAI,QAExC,GAAI,CAAC,MAAM,QAAQ,GACjB,MAAO,GAIT,OAFM,GAAkB,GAEjB,MAAM,QAAQ,IACd,GAAa,IAAc,IAAU,UAC1C,EAAM,KAAK,EAAU,QACrB,EAAY,EAAU,GAExB,MAAI,OAAM,QAAQ,IACd,KAAM,QAAQ,uCAChB,GAA2B,EAAK,EAAO,IAGlC,EAGT,YACI,EAAiB,EAAiB,GAEpC,GADA,EAAU,GAAW,GACjB,CAAE,MAAM,QAAQ,IAAS,CAAC,GAAa,IACzC,EACI,EAAM,SAAW,EACjB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,qBACnC,yCAAwC,EAAM,GAAE,eACxD,OAEF,EACI,EAAM,OAAS,EACf,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,4BACnC,uBAAsB,EAAI,OAAM,eACxC,EACI,EAAI,SAAW,EAAM,GACrB,WAAM,MAAA,eAAe,EAAQ,KAAK,MAAK,iBAAiB,EAAM,GAAE,IAC5D,sBAAqB,EAAI,OAAM,eAEvC,OADM,GAAW,EAAM,MAAM,GACpB,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAE,EAChC,GAA2B,EAAI,GAAI,EAAU,EAAQ,OAAO,IAIhE,YACI,EAAmC,EAAuB,EAC1D,GACF,GAAI,GAAiB,KACnB,OAEF,GAAI,IAAkB,WAAa,IAAkB,GACjD,IAAkB,WAAa,IAAgB,SACjD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,UAChD,OAAM,EAAa,oBAAoB,EAAW,uBAKtD,EAAiB,EAAiB,EAClC,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,YAAa,GACf,UAAY,EAAc,EAAE,MAAO,EAAS,GACrC,EAET,GAAI,GAAgB,GAAW,GAS/B,GANA,AAAI,IAAkB,UAClB,CAAC,OAAQ,QAAS,WAAW,QAAQ,IAAiB,GACxD,GAAgB,GAElB,GAAY,EAAc,EAAe,EAAS,GAE7C,GAAK,MACL,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IAAM,MAAO,IAAM,UACtD,MAAO,IAAM,WAAa,MAAO,IAAM,UAC1C,GAAM,GAAO,GAAK,KAAO,OAAU,EAAS,YAAY,KACxD,KAAM,IAAI,OACN,aAAa,EAAO,gBAAgB,EAAY,eAChD,mCAAkC,EAAI,MAE5C,GAAM,GAAgB,GAAW,EAAG,GACpC,AAAI,CAAC,GAAa,IAAM,CAAC,MAAM,QAAQ,IACrC,GAAI,CAAC,IAEP,GAAM,GAAiB,GACjB,EAAS,IAAkB,SAC7B,GAAa,EAAG,GAChB,GAAQ,EAAe,GAAI,GAC/B,MAAO,GAAO,WAAW,EAAQ,EAAe,eAI9C,EAA0B,EAAiB,EAC3C,GACF,GADE,AAAA,IAAA,QAAA,GAAA,WACE,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,OACN,YAAY,EAAO,cAAc,EAAY,2CAGnD,GAAM,GAAU,EAChB,MAAO,GAAQ,IACX,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAM,EAAO,IAAI,EAAC,IAAK,IACjD,GCnIN,GAkBa,IAAkB,kBAOQ,GACrC,GAAM,GAAO,OAAO,KAAK,GACzB,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,OACN,yGAEG,GAAK,OAAM,WAGpB,GAAI,GAAS,EAAK,GACZ,EAAK,EAAE,GAGb,AAAI,EAAO,SAAS,MAClB,GAAS,EAAO,UAAU,EAAG,EAAO,OAAS,IAI/C,EAAS,EAAS,GAGlB,GAAM,GAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,EAAO,WAAW,GAClB,IACE,GAAM,GAAS,EAAE,MAAA,OAAI,GACrB,MAAI,aAAkB,UACpB,QAAQ,MAAM,2CAEhB,EAAO,SAAS,GACT,QACA,GACP,QAAO,SAAS,MACV,IAGV,cAAO,eAAe,EAAI,OAAQ,CAAC,MAAO,EAAQ,aAAc,KAGzD,EC/DT,AA8CA,YAAoC,EAAoB,GACtD,GAAM,GAAQ,EAAgB,EAAM,OAAQ,WACtC,EAAQ,EAAgB,EAAM,OAAQ,WAC5C,GACI,EAAM,MAAO,EAAM,MACnB,yBAAyB,EAAM,MAAK,QAAQ,EAAM,MAAK,yCAG3D,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,QAAQ,EAAO,IAE1B,EAAwB,CAAC,KAAM,EAAO,KAAM,GAClD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC/DnC,YAyBI,EAAoB,EAAiB,EACrC,GAIF,GAHA,AAAI,GAAS,MACX,GAAQ,GAAW,IAEjB,IAAU,YACZ,KAAM,IAAI,OACN,oFAGN,GAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IACxC,MAAO,IAAW,UAAY,MAAO,IAAW,WAChD,MAAO,IAAW,SACpB,KAAM,IAAI,OACN,4HAGN,GAAI,GAAS,MACX,GAAmC,GAEnC,GAAM,GAAe,GAAc,GAC7B,EAAe,GAAc,GACnC,EACI,IAAiB,EACjB,WACI,MAAA,iCAAiC,EAAK,6BACnC,GAAY,mBAAmB,KAE1C,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,EAAE,GAC1C,GAAM,GAAW,EAAc,GACzB,EAAoB,IAAM,EAAc,OAAS,EACnD,IAAa,GAAc,EAAM,MAAM,IACvC,GACJ,EACI,EAAc,KAAO,EAAM,IAAM,CAAC,EAClC,WAAM,MAAA,+CACF,KAAI,EAAa,kCACjB,WAAU,EAAK,UAI3B,MAAI,CAAC,GAAa,IAAW,CAAC,MAAM,QAAQ,IAC1C,GAAS,CAAC,IAGZ,EAAQ,GAAS,EACjB,EAAS,IAAU,SACf,GAAa,EAAQ,GACrB,GAAQ,EAAoB,GAAI,IAC7B,EAAO,WAAW,EAAsB,EAAO,GC1ExD,YAqDI,EAAoB,EAAqB,GAC3C,GAAM,GAAgB,GAAW,EAAQ,GACzC,MAAO,IAAW,EAAQ,EAAO,EAAe,GCvDlD,AAsBO,GAAM,IAAkD,CAC7D,QAAW,EACX,QAAW,EACX,MAAS,EACT,OAAU,EACV,MAAS,EACT,KAAQ,EACR,UAAa,GC7Bf,AA2BA,GAAM,IAA0B,cAoB5B,EAAuC,oHAUzC,IAPM,EAAgC,GAChC,EAA2C,GAE3C,EAAkB,MAAM,QAAQ,GAClC,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,OAC7B,OAAO,KAAK,cAEP,GACP,GAAM,GAAO,EAAM,GACb,EAAI,MAAM,QAAQ,GAAW,EAAQ,GAAG,OAAS,EAAQ,GAC/D,GAAI,EAAE,QAAU,WAAa,EAAE,QAAU,SAAW,EAAE,QAAU,QAC5D,EAAE,QAAU,UAAY,EAAE,QAAU,YACtC,KAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,EAAE,OAE9D,GAAM,GAA6B,CAAC,KAAI,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,OACnE,GAAI,EAAE,QAAU,UACd,GAAM,GAAY,GAAI,SAAoB,SAAM,GAAO,MAAA,IAAA,EAAA,OAAA,OAAA,+EACxC,MAAA,CAAA,EAAM,EAAE,gBAKrB,IALM,EAAO,EAAA,OACP,EAAgB,EAAK,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAE,QAAQ,GACtD,GAA0B,EAAK,OAC7B,EAAQ,GAAI,YAAW,GACzB,EAAS,EACJ,EAAI,EAAG,EAAI,EAAK,OAAQ,IACzB,EAAM,EAAK,GACX,EACF,GAAI,YAAW,GAAI,aAAY,CAAC,EAAI,SAAS,QACjD,EAAM,IAAI,EAAe,GACzB,GAAU,GACV,EAAM,IAAI,EAAK,GACf,GAAU,EAAI,OAEhB,SAAQ,cAEV,EAAa,KAAK,OAElB,GAAa,KAAK,EAAE,QAEtB,AAAI,GAAS,MACX,GAAK,MAAQ,GAEf,EAAM,KAAK,IAjCJ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,IAA3B,GAoCY,MAAA,CAAA,EAAM,QAAQ,IAAI,WAAjC,SAAe,EAAA,OACrB,CAAA,EAAO,CAAC,KAAM,GAAuB,GAAe,MAAK,qBAmBvD,EAAqB,GAKvB,OAHM,GAAsB,GACxB,EACA,EAAS,EACM,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAO,EAAK,KACZ,EAAQ,EAAK,MACb,EAAQ,EAAK,MACb,EAAO,GAAc,GACvB,EAAM,OAEV,GAAI,gBAAkB,IACpB,GAAM,GAAe,EAAK,aAC1B,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,GAAI,CAAE,QAAS,IAAgB,SAAW,IACxC,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,+DAGxD,EAAa,QAAU,WAChC,GAAI,IAAU,UACZ,KAAM,IAAI,OACN,UAAU,EAAK,KAAI,sBAAsB,EAAa,MAAK,IAC3D,oDAAmD,EAAK,UAG9D,MAAM,IAAI,OACN,UAAU,EAAK,KAAI,gBACnB,uBAAsB,EAAa,MAAK,MACxC,wEAGN,GAAM,GAAyB,GAAqB,EAAa,OAC3D,EACF,EAAO,MAAM,EAAQ,EAAS,EAAO,GACnC,EAAkB,EAAa,QAAU,QAC3C,GAAI,YAAW,GACf,GAAI,aAAY,GACpB,GAAI,IAAU,UACZ,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,UAC3D,EAAS,GAAI,cAAa,EAAe,QACzC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,EAAI,EAAa,MAAQ,EAAa,aAE3C,EAAa,QAAU,UAChC,AAAI,IAAkB,QACpB,GAAgB,MAElB,EAAS,EAAc,OAEvB,MAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,qCAGhD,IAAU,SACnB,GAAI,EAAa,QAAU,SAAW,EAAa,QAAU,SAC3D,KAAM,IAAI,OACN,iCAAiC,EAAa,MAAK,2BAGzD,EAAS,GAAI,YAAW,EAAe,QACvC,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAI,EAAe,GACzB,EAAO,GAAK,KAAK,MAAM,EAAI,EAAa,MAAQ,EAAa,UAG/D,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,UACR,IAAU,UACnB,GAAM,GAAO,GAAc,EAAK,OAChC,EAAS,GACT,OAAS,GAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAa,GAAI,aACnB,EAAO,MAAM,EAAQ,EAAS,KAA0B,GAC5D,GAAU,GACV,GAAM,GAAQ,GAAI,YAAW,EAAO,MAAM,EAAQ,EAAS,IAC1D,EAAwB,KAAK,GAC9B,GAAU,QAGZ,GAAM,GAAc,GAAqB,GACnC,EAAa,EAAO,MAAM,EAAQ,EAAS,EAAO,GAExD,GAAI,IAAU,UACZ,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,WACf,IAAU,aACnB,EAAS,GAAI,cAAa,GAG1B,OAFM,GAAO,GAAI,cAAa,EAAO,OAAS,GACxC,EAAQ,GAAI,cAAa,EAAO,OAAS,GACtC,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAK,GAAK,EAAO,EAAI,GACrB,EAAM,GAAK,EAAO,EAAI,EAAI,GAE5B,GAAM,GAAa,GAAO,EAAM,EAAO,WACjC,EAAc,GAAO,EAAO,EAAO,WACzC,EAAI,GAAQ,GAAQ,EAAY,OAEhC,MAAM,IAAI,OAAM,gCAAgC,EAAI,MAAM,GAE5D,GAAU,EAAO,EAEnB,AAAI,IAAU,aACZ,GAAI,GAAQ,GAAO,EAAQ,EAAO,IAGtC,MAAO,eAM8B,GAErC,GAAI,IAAO,KACT,KAAM,IAAI,OAAM,wBAAwB,KAAK,UAAU,IAGzD,GAAI,GAAkB,EAShB,EAA6B,GACnC,EAAG,QAAQ,SAAC,GAMV,GALA,GAAmB,EAAE,WAErB,EAAa,KACT,EAAE,aAAe,EAAE,OAAO,WAAa,EACA,GAAK,GAAE,YAAoB,IAClE,CAAE,aAAoB,eAAgB,YAAoB,aACxD,YAAoB,aACxB,KAAM,IAAI,OAAM,mCAAmC,EAAE,YAAY,QAKrE,GAAM,GAAI,GAAI,YAAW,GACrB,EAAS,EACb,SAAa,QAAQ,SAAC,GACpB,EAAE,IAAI,GAAI,YAAW,EAAE,QAAS,GAChC,GAAU,EAAE,aAGP,EAAE,OAIX,GAAM,IAAgB,MAAO,SAAW,aACnC,OAAO,OAAS,aAAe,MAAO,OAAS,aAC/C,MAAO,OAAS,yBAWY,GAC/B,MAAI,IACK,OAAO,WAAW,GAEpB,GAAI,MAAK,CAAC,IAAM,iBASiB,GACxC,GAAI,GACF,MAAO,QAAO,KAAK,GAAQ,SAAS,UAItC,OAFM,GAAM,GAAI,YAAW,GACvB,EAAI,GACC,EAAI,EAAG,EAAI,EAAI,OAAQ,EAAI,EAAG,IACrC,GAAK,OAAO,aAAa,EAAI,IAE/B,MAAO,MAAK,eAS4B,GACxC,GAAI,IACF,GAAM,GAAM,OAAO,KAAK,EAAK,UAC7B,MAAO,GAAI,OAAO,MAAM,EAAI,WAAY,EAAI,WAAa,EAAI,YAI/D,OAFM,GAAI,KAAK,GACT,EAAS,GAAI,YAAW,EAAE,QACvB,EAAI,EAAG,EAAI,EAAE,OAAQ,EAAE,EAC9B,EAAO,IAAI,CAAC,EAAE,WAAW,IAAK,GAEhC,MAAO,GAAO,mBASwB,GACtC,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAI,GAAkB,EACtB,EAAQ,QAAQ,SAAC,GACf,GAAmB,EAAO,aAG5B,GAAM,GAAO,GAAI,YAAW,GACxB,EAAS,EACb,SAAQ,QAAQ,SAAC,GACf,EAAK,IAAI,GAAI,YAAW,GAAS,GACjC,GAAU,EAAO,aAEZ,EAAK,mBAUW,GACvB,GAAM,GAAY,IAElB,IADA,EAAO,EAAK,OACL,EAAK,SAAS,IACnB,EAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAErC,GAAM,GAAQ,EAAK,MAAM,GACzB,MAAO,GAAM,EAAM,OAAS,eAQe,GAE3C,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OAAM,uDAGlB,MAAO,CACL,UAAW,GAAI,MACf,kBAAmB,OACnB,mBAAoB,EAAe,eAAiB,KAChD,EACA,GAAiB,KAAK,UAAU,EAAe,gBACnD,iBAAkB,EAAe,aAAe,KAC5C,EACA,GAAiB,KAAK,UAAU,EAAe,cACnD,gBAAiB,EAAe,YAAc,KAC1C,EACA,EAAe,WAAW,YAUlC,cACE,GAAM,GAAkB,SAAC,GAIvB,OAHI,GAAI,GAAK,GACT,EAAI,EAEA,GAAI,WAAgB,GAC1B,GAAK,QACL,IAAM,EAER,UAAK,CAAC,QACN,GAAK,UAEE,EAAI,GAGP,EAAe,GAAI,aAAY,MAErC,EAAa,GAAK,EAClB,OAAS,GAAI,EAAG,EAAI,KAAM,IACxB,EAAa,GAAK,EAAgB,GAEpC,OAAS,GAAI,KAAM,EAAI,KAAM,IAC3B,EAAa,GAAK,UAAe,GAAI,MAAS,IAGhD,MAAO,GAST,cACE,GAAM,GAAgB,GAAI,aAAY,IAEtC,EAAc,GAAK,EACnB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,EAAc,IAAM,WACpB,OAAS,GAAI,EAAG,EAAI,GAAI,IACtB,EAAc,GAAK,GAAK,GAE1B,OAAS,GAAI,GAAI,EAAI,GAAI,IACvB,EAAc,GAAK,WAAe,GAAI,IAAO,IAG/C,MAAO,GAST,cAGE,OAFM,GAAc,GAAI,aAAY,IAE3B,EAAI,EAAG,EAAI,GAAI,IACtB,EAAY,GAAK,KAEnB,SAAY,GAAK,EAAY,IAAM,EAE5B,gBAeP,GAAM,GAAe,KACf,EAAgB,KAChB,EAAc,KAEpB,MAAO,UAAC,GAGN,OAFM,GAAS,GAAI,aAAY,EAAI,EAAe,QAC5C,EAAmB,GAAI,aAAY,GAChC,EAAQ,EAAG,EAAQ,EAAe,OAAQ,KACjD,GAAM,GAAc,EAAe,GAC7B,EACF,EAAa,EAAY,GAAe,IAAO,GAAc,OAC7D,EAAc,GAAe,IACjC,EAAiB,GAAS,EAE5B,MAAO,IAAI,cAAa,IChf5B,AAsBA,GAAA,IAAA,WAOE,aACE,KAAK,YAAc,GACnB,KAAK,YAAc,GAGN,SAAA,YAAf,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UASnB,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAS3C,EAAA,mBAAP,SAA0B,GACxB,EAAiB,cAAc,YAAY,KAAK,IAW3C,EAAA,gBAAP,SAAuB,GACrB,MAAO,GAAiB,YAAY,EAAK,SAWpC,EAAA,gBAAP,SAAuB,EAAsB,GAE3C,MAAO,GAAiB,YAAY,EAAK,OAAQ,IAGpC,EAAA,YAAf,SACI,EAAsB,EACtB,GACF,GAAM,GAA6B,GAC7B,EAAU,IAAgB,OAC5B,EAAiB,cAAc,YAC/B,EAAiB,cAAc,YACnC,SAAQ,QAAQ,SAAA,GACd,GAAM,GAAU,EAAO,EAAK,GAC5B,AAAI,IAAY,MACd,EAAc,KAAK,KAGhB,GAEX,KAEa,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAqB,SAAC,GAC/B,MAAA,IAAiB,mBAAmB,IAC3B,GAAkB,SAAC,GAC5B,MAAA,IAAiB,gBAAgB,IACxB,GACT,SAAC,EAAsB,GACnB,MAAA,IAAiB,gBAAgB,EAAK,IC/G9C,AAyBA,GAAM,IAAgB,eAChB,GAAmB,EAKnB,GAAmB,eAInB,GAAkB,mBAexB,cACE,GAAI,CAAC,KAAM,QAAQ,cAIjB,KAAM,IAAI,OACN,2FAIN,GAAM,GAAiB,MAAO,SAAW,YAAc,KAAO,OACxD,EAAU,EAAU,WAAa,EAAU,cAC7C,EAAU,iBAAmB,EAAU,aACvC,EAAU,cACd,GAAI,GAAW,KACb,KAAM,IAAI,OACN,6DAEN,MAAO,GAGT,YAAuB,GACrB,GAAM,GAAK,EAAY,OACvB,EAAG,kBAAkB,GAAkB,CAAC,QAAS,cACjD,EAAG,kBAAkB,GAAiB,CAAC,QAAS,cAQlD,GAAA,IAAA,WAME,WAAY,GAGV,GAFA,KAAK,UAAY,KAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,kEAEN,KAAK,UAAY,EAGb,SAAA,UAAA,KAAN,SAAW,sEAET,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAIN,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,UAAW,SAIvC,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,eAAe,KAAK,iBAiB1B,EAAA,UAAA,eAAR,SAAuB,EAAmB,GAA1C,GAAA,GAAA,KAEE,MAAO,IAAI,SAAmC,SAAC,EAAS,GACtD,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OAEvB,GAAI,GAAkB,MAEpB,GAAM,GAAU,EAAG,YAAY,GAAkB,YAC3C,EAAa,EAAQ,YAAY,IACjC,EAAa,EAAW,IAAI,EAAK,WACvC,EAAW,UAAY,WACrB,GAAI,EAAW,QAAU,KACvB,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAK,UAAS,oBAGlD,EAAQ,EAAW,OAAO,iBAG9B,EAAW,QAAU,SAAA,GACnB,SAAG,QACI,EAAO,EAAW,QAE3B,EAAQ,WAAa,WAAM,MAAA,GAAG,cAG9B,GAAM,GACF,GAA6B,GAE3B,EAAS,EAAG,YAAY,GAAiB,aAC3C,EAAY,EAAO,YAAY,IAC7B,EACF,EAAU,IAAI,CAAC,UAAW,EAAK,UAAW,mBAAkB,IAC5D,EACJ,EAAe,UAAY,WAEzB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAkB,EAAW,IAAI,CACrC,UAAW,EAAK,UAChB,eAAc,EACd,mBAAkB,IAEpB,EAAgB,UAAY,WAAM,MAAA,GAAQ,CAAC,mBAAkB,KAC7D,EAAgB,QAAU,SAAA,GAGxB,EAAY,EAAO,YAAY,IAC/B,GAAM,GAAoB,EAAU,OAAO,EAAK,WAChD,EAAkB,UAAY,WAC5B,SAAG,QACI,EAAO,EAAgB,QAEhC,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAgB,UAIpC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAE/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,YAKtC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,WAvHtC,EAAA,WAAa,eA0H/B,KAEa,GAA4B,SAAC,GACxC,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAmBH,GAC/B,MAAO,IAAI,IAAiB,GAG9B,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAiB,YACnC,EAAI,MAAM,GAAiB,WAAW,QACtC,EAGN,GAAA,IAAA,WAGE,aACE,KAAK,UAAY,KAGb,SAAA,UAAA,WAAN,yFACE,MAAA,CAAA,EAAO,GAAI,SACP,SAAC,EAAS,GACR,GAAM,GACF,EAAK,UAAU,KAAK,GAAe,IACvC,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAK,EAAG,YAAY,GAAiB,YACrC,EAAQ,EAAG,YAAY,IASvB,EAAqB,EAAc,SACzC,EAAkB,UAAY,WAE5B,OADM,GAA4C,GAC/B,EAAA,EAAA,EAAA,EAAkB,OAAlB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAI,EAAK,WAAa,EAAK,mBAE7B,EAAQ,IAEV,EAAkB,QAAU,SAAA,GAC1B,SAAG,QACI,EAAO,EAAkB,QAElC,EAAG,WAAa,WAAM,MAAA,GAAG,UAE3B,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAIpD,EAAA,UAAA,YAAN,SAAkB,iFAChB,SAAO,GAAiB,GACxB,CAAA,EAAO,GAAI,SAA4B,SAAC,EAAS,GAC/C,GAAM,GAAc,EAAK,UAAU,KAAK,GAAe,IACvD,EAAY,gBAAkB,WAAM,MAAA,IAAc,IAElD,EAAY,UAAY,WACtB,GAAM,GAAK,EAAY,OACjB,EAAS,EAAG,YAAY,GAAiB,aACzC,EAAY,EAAO,YAAY,IAE/B,EAAiB,EAAU,IAAI,GACjC,EACJ,EAAe,UAAY,WACzB,GAAI,EAAe,QAAU,KAC3B,SAAG,QACI,EAAO,GAAI,OACd,gCAAgC,EAAI,oBAIxC,GAAM,GAAoB,EAAU,OAAO,GACrC,EAAkB,WAEtB,EAAU,EAAG,YAAY,GAAkB,aAC3C,GAAM,GAAa,EAAQ,YAAY,IACjC,EAAqB,EAAW,OAAO,GAC7C,EAAmB,UAAY,WAC3B,MAAA,GAAQ,EAAe,OAAO,qBAClC,EAAmB,QAAU,SAAA,GACzB,MAAA,GAAO,EAAe,SAI5B,EAAkB,UAAY,EAC9B,EAAkB,QAAU,SAAA,GAC1B,WACA,EAAG,QACI,EAAO,EAAe,SAInC,EAAe,QAAU,SAAA,GACvB,SAAG,QACI,EAAO,EAAe,QAG/B,EAAO,WAAa,WAClB,AAAI,GAAW,KACb,EAAG,QAEH,EAAQ,WAAa,WAAM,MAAA,GAAG,WAIpC,EAAY,QAAU,SAAA,GAAS,MAAA,GAAO,EAAY,gBAGxD,KCjWA,AAyBA,GAAM,IAAiB,IACjB,GAAc,sBACd,GAAc,OACd,GAAwB,iBACxB,GAAsB,eACtB,GAAqB,cACrB,GAAwB,iBA8B9B,YAAsB,GAOpB,MAAO,CACL,KAAM,CAAC,GAAa,EAAM,IAAa,KAAK,IAC5C,SAAU,CAAC,GAAa,EAAM,IAAuB,KAAK,IAC1D,YAAa,CAAC,GAAa,EAAM,IAAqB,KAAK,IAC3D,WAAY,CAAC,GAAa,EAAM,IAAoB,KAAK,IACzD,cACI,CAAC,GAAa,EAAM,IAAuB,KAAK,KAWxD,YAA6B,GAC3B,GAAM,GAAQ,EAAI,MAAM,IACxB,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OAAM,uBAAuB,GAEzC,MAAO,GAAM,MAAM,EAAG,EAAM,OAAS,GAAG,KAAK,IAG/C,YAA0B,GACxB,MAAO,GAAI,WAAW,GAAoB,YACtC,EAAI,MAAM,GAAoB,WAAW,QACzC,EAgBN,GAAA,IAAA,WAOE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,eAAiB,MAAO,SAAW,aAClD,MAAO,QAAO,cAAiB,YAKjC,KAAM,IAAI,OACN,2DAIN,GAFA,KAAK,GAAK,OAAO,aAEb,GAAa,MAAQ,CAAC,EACxB,KAAM,IAAI,OACN,sEAEN,KAAK,UAAY,EACjB,KAAK,KAAO,GAAa,KAAK,WAY1B,SAAA,UAAA,KAAN,SAAW,gFACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,4FAGE,EAAW,KAAK,UAAU,EAAe,eACzC,EAAc,KAAK,UAAU,EAAe,aAE5C,EACF,GAA6B,GAEjC,IACE,YAAK,GAAG,QAAQ,KAAK,KAAK,KAAM,KAAK,UAAU,IAC/C,KAAK,GAAG,QAAQ,KAAK,KAAK,SAAU,GACpC,KAAK,GAAG,QAAQ,KAAK,KAAK,YAAa,GACvC,KAAK,GAAG,QACJ,KAAK,KAAK,WACV,GAA0B,EAAe,aAC7C,KAAK,GAAG,QAAQ,KAAK,KAAK,cAAe,KAAK,UAAU,CACtD,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,uBAGtC,CAAA,EAAO,CAAC,mBAAkB,UACnB,GAEP,WAAK,GAAG,WAAW,KAAK,KAAK,MAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,UAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,aAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,YAC7B,KAAK,GAAG,WAAW,KAAK,KAAK,eAEvB,GAAI,OACN,yBAAyB,KAAK,UAAS,sFAEvC,uBAAsB,EAAmB,mBAAkB,MAC3D,qBAAoB,EAAmB,iBAAgB,MACvD,oBAAmB,EAAmB,gBAAe,qBAazD,EAAA,UAAA,KAAN,gGAGE,GAFM,EACF,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,OACrC,GAAQ,KACV,KAAM,IAAI,OACN,kDAAkD,KAAK,UAAS,KAGtE,GAAI,EAAK,oBAAsB,OAC7B,KAAM,IAAI,OACN,6EAQN,GAJM,EAAsB,GAGtB,EAAW,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,WAClD,GAAY,KACd,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAS,iBAOhE,GAJA,EAAI,cAAgB,EAGd,EAAc,KAAK,MAAM,KAAK,GAAG,QAAQ,KAAK,KAAK,cACrD,GAAe,KACjB,KAAM,IAAI,OACN,gDAAgD,KAAK,UAAS,kBAiBpE,GAdA,EAAI,YAAc,EAGZ,EAAiB,KAAK,GAAG,QAAQ,KAAK,KAAK,eACjD,AAAI,GAAkB,MACd,GAAW,KAAK,MAAM,GAC5B,EAAI,OAAS,EAAS,OACtB,EAAI,YAAc,EAAS,YAC3B,EAAI,YAAc,EAAS,YAC3B,EAAI,oBAAsB,EAAS,qBAI/B,EAAmB,KAAK,GAAG,QAAQ,KAAK,KAAK,YAC/C,GAAoB,KACtB,KAAM,IAAI,OACN,uDACA,KAAI,KAAK,UAAS,mBAExB,SAAI,WAAa,GAA0B,GAE3C,CAAA,EAAO,QAzIO,EAAA,WAAa,kBA2I/B,KAEa,GAA+B,SAAC,GAC3C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAoB,aACrD,GACH,EAAI,MAAM,GAAoB,WAAW,SAJxC,MAUX,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBA0BA,GAClC,MAAO,IAAI,IAAoB,GAGjC,GAAA,IAAA,WAGE,aACE,EACI,KAAM,QAAQ,cACd,WAAM,MAAA,6CACV,EACI,MAAO,SAAW,aACd,MAAO,QAAO,cAAiB,YACnC,WAAM,MAAA,4DACV,KAAK,GAAK,OAAO,aAGb,SAAA,UAAA,WAAN,8FAIE,IAHM,EAA4C,GAC5C,EAAS,GAAc,GACvB,EAAS,GAAiB,GACvB,EAAI,EAAG,EAAI,KAAK,GAAG,OAAQ,EAAE,EAC9B,EAAM,KAAK,GAAG,IAAI,GACxB,AAAI,EAAI,WAAW,IAAW,EAAI,SAAS,IACnC,GAAY,GAAoB,GACtC,EAAI,GAAa,KAAK,MAAM,KAAK,GAAG,QAAQ,KAGhD,MAAA,CAAA,EAAO,QAGH,EAAA,UAAA,YAAN,SAAkB,8EAGhB,GAFA,EAAO,GAAiB,GAClB,EAAO,GAAa,GACtB,KAAK,GAAG,QAAQ,EAAK,OAAS,KAChC,KAAM,IAAI,OAAM,8BAA8B,EAAI,KAE9C,SAAO,KAAK,MAAM,KAAK,GAAG,QAAQ,EAAK,OAE7C,KAAK,GAAG,WAAW,EAAK,MACxB,KAAK,GAAG,WAAW,EAAK,UACxB,KAAK,GAAG,WAAW,EAAK,aACxB,KAAK,GAAG,WAAW,EAAK,YACxB,CAAA,EAAO,QAEX,KCvVA,AAiCA,GAAM,IAAoB,MAE1B,GAAA,WAME,aACE,KAAK,SAAW,GAGH,SAAA,YAAf,WACE,MAAI,GAA0B,UAAY,MACxC,GAA0B,SAAW,GAAI,IAEpC,EAA0B,UAS5B,EAAA,gBAAP,SAAuB,EAAgB,GACrC,EAAO,GAAU,KAAM,WAAM,MAAA,0CAC7B,AAAI,EAAO,SAAS,KAClB,GAAS,EAAO,MAAM,EAAG,EAAO,QAAQ,MAE1C,EAAO,EAAO,OAAS,EAAG,WAAM,MAAA,wCAChC,GAAM,GAAW,EAA0B,cAC3C,EACI,EAAS,SAAS,IAAW,KAC7B,WAAM,MAAA,2DACF,EAAM,OACd,EAAS,SAAS,GAAU,GAGvB,EAAA,WAAP,SAAkB,GAChB,GAAM,GAAU,KAAK,cAAc,SAAS,GAC5C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,yCAAyC,EAAM,KAEjE,MAAO,IAGF,EAAA,WAAP,WACE,MAAO,QAAO,KAAK,KAAK,cAAc,WAE1C,KAUA,YAAkB,GAChB,GAAI,EAAI,QAAQ,MAAuB,GACrC,KAAM,IAAI,OACN,6EAEA,IAAG,GAA0B,aAAa,KAAK,OAErD,MAAO,CACL,OAAQ,EAAI,MAAM,IAAmB,GACrC,KAAM,EAAI,MAAM,IAAmB,IAIvC,YACI,EAAmB,EACnB,GAAA,MAAA,KAAA,QAAA,GAAA,6GACF,SACI,IAAc,EACd,WAAM,MAAA,wCAAwC,EAAS,MAErD,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,kEACF,EAAS,MACjB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,iCAAgC,EAAS,OAC3C,EAAc,EAAa,GAE3B,EAAe,GAAiB,gBAAgB,GACtD,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,mEACF,QAAO,EAAO,OACtB,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,yCAAyC,EAAa,OAAM,KAC9D,sCAAqC,EAAO,OAC9C,EAAc,EAAa,GAE3B,EAAe,GAAS,GAAW,OACnC,EAAa,GAAS,GAAW,KACjC,EAAa,IAAiB,GAAS,GAAW,OAEjC,CAAA,EAAM,EAAY,qBAAnC,GAAiB,EAAA,OAKnB,GAAgB,EAClB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIiB,MAAA,CAAA,EAAM,EAAY,KAAK,iBAApC,GAAa,EAAA,OAKf,GAAgB,CAAC,EACnB,CAAA,EAAM,GAA0B,WAAW,GACtC,YAAY,IAFf,CAAA,EAAA,UACF,EAAA,wBAIF,MAAA,CAAA,EAAO,EAAW,yBAsCpB,4HACQ,EAAU,GAA0B,aACpC,EAA2C,OAC5B,EAAA,yBAAA,GAAA,EAAA,OAAV,GAAM,EAAA,GAEX,CAAA,EAAM,GAA0B,WAAW,GAAQ,eAF7B,CAAA,EAAA,UACpB,EACF,EAAA,OACJ,IAAW,IAAQ,GACX,EAAM,EAAS,GAAoB,EACzC,EAAI,GAAO,EAAU,oBALJ,wBAQrB,MAAA,CAAA,EAAO,QAoCT,YAA2B,8EACnB,SAAgB,GAAS,GACzB,EAAU,GAA0B,WAAW,EAAc,QACnE,CAAA,EAAO,EAAQ,YAAY,EAAc,WAkD3C,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QAiDhD,YACI,EAAmB,4EACf,SAAe,GACrB,CAAA,EAAO,GAAmB,EAAW,EAAS,QChWhD,AA0BA,GAAA,IAAA,WAAA,cAKE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,OAAM,EAAM,IAGrB,EAAA,UAAA,IAAA,WACE,MAAO,aAAY,OAGrB,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,kDAAkD,GAExD,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAAI,cAElB,KAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAO,IAAI,aAAY,GAAU,OAAO,IAE5C,KAEA,GAAI,KAAM,IAAI,eACZ,KAAM,YAAY,UAAW,GAAI,KAGjC,IACE,GAA0B,gBACtB,GAAoB,WAAY,GAAI,WACjC,IAIT,IACE,GAA0B,gBACtB,GAAiB,WAAY,GAAI,WAC9B,KCpEX,AAqBO,GAAM,IAAe,CAE1B,YAAa,WAAM,MAAA,QAIjB,GAaJ,GAAA,WAKE,aAEE,KAAK,KAAO,KAGZ,KAAK,YAAc,GAAI,MAAK,KAAK,YAGnC,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAI,MAAM,OAAO,OAAS,KACjB,KAAM,OAAO,MAAM,EAAM,GAGlC,CAAI,IAAe,MACjB,IAAc,GAAa,eAEtB,GAAY,EAAM,KAG3B,EAAA,UAAA,IAAA,WACE,GAAM,GAAO,QAAQ,SACrB,MAAO,GAAK,GAAK,IAAO,EAAK,GAAK,KAGpC,EAAA,UAAA,OAAA,SAAO,EAAc,GACnB,GAAI,IAAa,SAAW,IAAa,OACvC,KAAM,IAAI,OACN,sDAAsD,GAE5D,MAAO,MAAK,YAAY,OAAO,IAEjC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,MAAI,GAAM,SAAW,EACZ,GAEF,GAAI,MAAK,KAAK,YAAY,GAAU,OAAO,IAEtD,KAEA,AAAI,KAAM,IAAI,YACZ,KAAM,YAAY,OAAQ,GAAI,KCrFhC,YAgDI,EAAoB,EACpB,GADoB,MAAA,KAAA,QAAA,GAAW,WAEjC,EAAQ,GAAS,UACjB,GAAwC,GACjC,GAAI,IAAmB,EAAO,EAAO,GCpD9C,AAuCA,YAAiC,EAAiB,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAGnC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,mCAAmC,GAErD,GAAI,IAAU,UAAY,EAAG,QAAU,UACnC,IAAU,UAAY,EAAG,QAAU,SACrC,KAAM,IAAI,OAAM,yCAGlB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAGhB,IAAO,EAAG,CAAC,MAAK,KC3D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,QAAS,MACtC,EAAU,WACZ,MAAA,GAAO,qBAAqB,EAAG,OAAQ,EAAG,MAAO,EAAG,QAElD,EAAyB,CAAC,EAAG,GAInC,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAQ,EAAG,CAAC,OAAM,KCrD/B,YAgCwC,EAAM,GAAA,AAAA,IAAA,QAAA,GAAA,IAC5C,QAAQ,IAAI,EAAE,SAAS,ICjCzB,AAqBA,KAcA,GAAM,IAAuB,CAC3B,OAAM,GACN,KAAI,GACJ,MAAK,GACL,MAAK,IAEP,GAAa,ICzCb,AA6BA,GAAM,IAA2B,QAC3B,GAA8B,QAC9B,GAAqC,eAE3C,YAAkB,GAChB,MAAO,IAAI,SAAQ,SAAA,GAAW,MAAA,YAAW,KAAU,KAAK,GAG1D,GAAA,IAAA,WAQE,WAAY,GACV,GAAI,CAAC,KAAM,QAAQ,cAGjB,KAAM,IAAI,OACN,uFAIN,AAAI,EAAe,WAAW,EAAiB,aAC7C,GAAiB,EAAe,MAAM,EAAiB,WAAW,SAEpE,AAAI,IAAkB,MAAQ,EAAe,SAAW,IACtD,GAAiB,IAGnB,KAAK,sBAAwB,EAAiB,GAC9C,KAAK,mBACD,EAAiB,GAGjB,SAAA,UAAA,KAAN,SAAW,6GACT,GAAI,MAAQ,WAAc,YACxB,KAAM,IAAI,OACN,8FAGA,EAAa,OAAO,IAAI,gBAAgB,GAAI,MAC9C,CAAC,EAAe,YAAa,CAAC,KAAM,+BAEpC,GAAe,wBAAyB,cAAxC,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,OACN,gGAGE,SAAyC,CAAC,CAC9C,MAAO,CAAC,KAAO,KAAK,oBACpB,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,gBAAe,GAEX,EACF,OAAO,IAAI,gBAAgB,GAAI,MAC3B,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,sBAIT,EAAa,KAAK,YAAc,KAAO,SAAS,cAAc,KACvB,KAAK,WAClD,EAAW,SAAW,KAAK,sBAC3B,EAAW,KAAO,EAIlB,CAAA,EAAM,GAAM,WAAM,MAAA,GAAW,cAAc,GAAI,YAAW,0BAA1D,GAAA,OAEI,EAAe,YAAc,KACzB,GAAmB,KAAK,kBAAoB,KAC9C,SAAS,cAAc,KACvB,KAAK,iBACT,EAAiB,SAAW,KAAK,mBACjC,EAAiB,KAAO,EACxB,CAAA,EAAM,GACF,WAAM,MAAA,GAAiB,cAAc,GAAI,YAAW,cAPtD,CAAA,EAAA,UAMF,EAAA,wBAIF,MAAA,CAAA,EAAO,CAAC,mBAAoB,GAA6B,WA1E7C,EAAA,WAAa,eA6E/B,KAEA,GAAA,WAGE,WAAY,GACV,GAAI,GAAS,MAAQ,EAAM,OAAS,EAClC,KAAM,IAAI,OACN,2DACA,iBAAgB,IAEtB,KAAK,MAAQ,EAGT,SAAA,UAAA,KAAN,6FACQ,SAAW,KAAK,MAAM,GACtB,EAAc,KAAK,MAAM,MAAM,GAErC,CAAA,EAAO,GAAI,SAAwB,SAAC,EAAS,GAC3C,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GAEnB,GAAM,GAAY,KAAK,MAAO,EAAM,OAAe,QAC7C,EAAgB,EAAU,cAChC,GAAI,GAAiB,MACnB,EAAO,GAAI,OACP,4CAA4C,EAAS,OACzD,OAGF,AAAI,EAAY,SAAW,GACzB,EAAQ,CAAC,cAAa,IAGxB,GAAM,GAAkB,EAAU,gBAClC,GAAI,GAAmB,MACrB,EAAO,GAAI,OACP,6CAA6C,EAAS,OAC1D,OAGF,GAAI,GACJ,IACE,EACI,EAAK,4BAA4B,EAAiB,SAC/C,GACP,EAAO,GACP,OAGF,GAAM,GAAsC,GACtC,EAAkB,GAClB,EAAgC,GACtC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,EAAM,KAAK,GACX,EAAe,KAAK,QAEtB,EAAY,KAAI,MAAhB,EAAoB,EAAa,WAGnC,EAAgB,QAAQ,SAAA,GACtB,EAAa,MAAM,QAAQ,SAAA,GACzB,GAAM,GAAmB,GAAI,YAC7B,EAAiB,OAAS,SAAC,GAEzB,GAAM,GAAc,EAAM,OAAe,OACnC,EAAQ,EAAM,QAAQ,GAC5B,EAAe,GAAS,EACxB,AAAI,EAAe,QAAQ,QAAU,IACnC,EAAQ,CACN,cAAa,EACb,YAAW,EACX,WAAY,GAAwB,GACpC,OAAQ,EAAU,OAClB,YAAa,EAAU,YACvB,YAAa,EAAU,YACvB,oBAAqB,EAAU,uBAIrC,EAAiB,QAAU,SAAA,GACvB,MAAA,GAAO,6CAA6C,EAAI,OAC5D,EAAiB,kBAAkB,EAAW,SAIpD,EAAW,QAAU,SAAA,GAAS,MAAA,GAC1B,2DACA,eAAc,EAAS,KAAI,qCAC3B,yCACJ,EAAW,WAAW,WAOlB,EAAA,UAAA,4BAAR,SACI,EAAiC,GAInC,OAHM,GAAsB,GACtB,EAAY,EAAM,IAAI,SAAA,GAAQ,MAAA,IAAS,EAAK,QAC5C,EAAqC,GACvB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAM,MAAM,QAAQ,SAAA,GAClB,GAAM,GAAe,GAAS,GAC9B,GAAI,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,sDACA,KAAI,EAAY,MAGtB,GADA,EAAU,KAAK,GACX,EAAU,QAAQ,KAAkB,GACtC,KAAM,IAAI,OACN,8BAA8B,EAAY,sBAE9C,EAAW,GAAQ,EAAM,EAAU,QAAQ,MAKjD,GAAI,EAAU,SAAW,EAAM,OAC7B,KAAM,IAAI,OACN,uDACA,KAAI,EAAU,OAAM,8CACpB,KAAI,EAAM,OAAM,OAEtB,MAAO,IAEX,KAEa,GAAmC,SAAC,GAC/C,MAAK,MAAM,QAAQ,eAGb,EAAC,MAAM,QAAQ,IAAQ,EAAI,WAAW,GAAiB,aAClD,GAAiB,EAAI,MAAM,GAAiB,WAAW,SAHzD,MASX,GAAiB,mBAAmB,gBAyCH,GAAA,MAAA,KAAA,QAAA,GAAA,SACxB,GAAI,IAAiB,eA2CD,GAC3B,MAAO,IAAI,IAAa,GC5V1B,YA8BI,EAAmC,EACnC,EAAwB,GAC1B,EAAc,GACd,EAAgB,GAAwB,EACxC,EAAc,GAAsB,EACpC,EAAc,EAAe,GAC7B,GAAI,GAAkB,EAEhB,EAAkB,SAAC,GACvB,SAAQ,KAAK,SAAA,GACX,GAAM,GAAW,EACb,GAAE,EAAkB,EAAS,OAAU,GAAc,GAEzD,SAAW,GACJ,IAEF,GAGT,WAAuB,GACrB,EACI,GAAY,MAAQ,MAAM,QAAQ,IAAa,EAAS,OAAS,EACjE,WAAM,MAAA,wCAGZ,WAAuB,EAAuB,GAC5C,EACI,GAAiB,GAAK,GAAiB,EACvC,WAAM,MAAA,kDACF,sBAAqB,KAC7B,EACI,GAAe,GAAK,GAAe,EACnC,WAAM,MAAA,kDACF,oBAAmB,KAC3B,EACI,GAAe,EACf,WAAM,MAAA,uDACF,sBAAqB,EAAa,qBAClC,IAAG,KAGb,MAAO,SAAQ,IAAI,EAAS,IAAI,ICvElC,YAqCI,EAAqB,6HACvB,AAAI,IAAe,MACjB,GAAc,IAGV,EAAY,EAAY,WAAa,KAAO,KAAM,SAAS,MACf,EAAY,UAGxD,EAAW,EAAU,IACvB,SAAA,GACI,MAAA,GAAU,EAAU,EAAY,YAAa,CAAC,SAAU,OAE1D,EAAqB,EACrB,EAAmB,GAEP,EAAY,YAAc,KACxC,CAAA,EAAM,QAAQ,IAAI,IADJ,CAAA,EAAA,UACd,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAU,EAAY,WAAY,EAClC,WAFJ,EAAA,EAAA,8BAFE,GAAS,EAMT,EAAiB,EAAU,IAAI,SAAA,GAAY,MAAA,GAAS,gBAEpD,EAAsB,GACtB,EAAoB,EAEV,EAAY,YAAc,KACtC,CAAA,EAAM,QAAQ,IAAI,IADN,CAAA,EAAA,UACZ,SAAA,EAAA,oBACA,MAAA,CAAA,EAAM,GACF,EAAgB,EAAY,WAAY,EACxC,WAFJ,EAAA,EAAA,wBAFE,SAAO,EAKb,CAAA,EAAO,oBAaL,EAAiC,EACjC,EACA,GAFiC,MAAA,KAAA,QAAA,GAAA,wEAS7B,SAAe,SAAC,GAClB,MAAA,IAAyB,EAAW,CAAC,YAAW,KAC9C,EAAc,GAAqB,GAEzC,CAAA,EAAO,EAAY,EAAU,EAAgB,oBA4B3C,GADJ,GAAA,GAAA,KAIE,MAAO,UACI,EAAiC,EACjC,GADiC,MAAA,KAAA,QAAA,GAAA,4GAqD1C,GAjDM,EAAyB,EAAS,IAAI,WAAM,MAAA,KAC5C,EAKF,GACE,EACF,GAAe,KAAO,EAAY,IAAI,WAAM,MAAA,KAAS,GACnD,EAAmC,GACzC,EAAS,QAAQ,SAAC,EAAqB,GACrC,GAAI,GAAc,EAClB,EAAoB,QAAQ,QAAQ,SAAA,GAClC,GAAM,GAAY,gBAAkB,GAChC,EAAa,aAAa,MAC1B,EAAa,MAEX,EAAe,GAAqB,GACtC,GAAmB,EAAa,OAE9B,EAA8B,WAClC,EAAuB,GAAc,GACrC,AAAI,EAAoB,IAAe,MACrC,GAAoB,GAAc,IAGpC,EAAoB,GAAY,KAAK,CACnC,cAAe,EACf,YAAW,EACX,UAAW,KAIf,AAAI,GAAe,KACjB,EAAY,QAAQ,SAAC,EAAY,GAC/B,AAAI,IAAe,EAAa,MAC9B,KACA,EAAa,GAAe,MAIhC,IAGF,EAAuB,KAAK,EAAa,MACzC,GAAe,MAIf,CAAC,EAAa,MAAM,SAAA,GAAS,MAAA,KACzB,QAAkB,EAAY,OAAO,SAAC,EAAG,GAAM,MAAA,CAAC,EAAa,KAC7D,GAAI,OACN,kDACG,GAAgB,KAAK,MAAK;GAC7B,yCACG,GAAuB,KAAK,MAAK,MAKpC,SACF,EAAuB,OAAO,SAAC,EAAa,EAAa,GACvD,MAAI,IACF,EAAY,KAAK,GAEZ,GACN,IAED,EAAsB,GAC5B,EAAoB,QAAQ,SAAA,GAC1B,EAAS,GAAG,MAAM,QAAQ,SAAA,GACxB,GAAM,GAAW,EACZ,CAAC,EAAe,SAAS,KAAa,GAAN,KAAY,EACjD,EAAU,KAAK,OAGH,CAAA,EAAM,EAAqB,WAArC,SAAU,EAAA,OAEV,EAAmC,GACrC,EAAoB,EACxB,EAAoB,QAAQ,SAAA,GAI1B,OAHM,GAAa,EAAS,GAAG,MAAM,OAEjC,EAAa,EACR,EAAI,EAAG,EAAI,EAAY,IAC9B,GAAc,EAAQ,EAAoB,GAAG,WAO/C,OAHM,GAAc,GAAI,aAAY,GAC9B,EAAkB,GAAI,YAAW,GACnC,EAAoB,EACf,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAS,GAAI,YAAW,EAAQ,EAAoB,IAC1D,EAAgB,IAAI,EAAQ,GAC5B,GAAqB,EAAO,WAG9B,GAAM,GAAiB,EAAoB,GAC3C,EAAe,QAAQ,SAAA,GACrB,GAAM,GAAa,EAAY,MAC3B,EAAa,YACb,EAAa,YAAc,EAAa,WACtC,EACF,GAAc,EAAY,CAAC,EAAa,gBAC5C,OAAW,KAAQ,GACjB,EAAiB,GAAQ,EAAgB,KAI7C,GAAqB,IAGvB,CAAA,EAAO,SCpPX,AA+BA,GAAM,IAAyB,2BACzB,GAAY,mBAClB,GAAA,WAcE,WAAY,EAAc,GAgCxB,GAvCO,KAAA,eAAiB,OAQxB,AAAI,GAAe,MACjB,GAAc,IAEhB,KAAK,iBAAmB,EAAY,iBACpC,KAAK,WAAa,EAAY,WAC9B,KAAK,mBAAqB,EAAY,mBAEtC,AAAI,EAAY,WAAa,KAC3B,GACI,MAAO,GAAY,WAAc,WACjC,WAAM,MAAA,gIAGV,KAAK,MAAQ,EAAY,WAEzB,KAAK,MAAQ,KAAM,SAAS,MAG9B,EACI,GAAQ,MAAQ,EAAK,OAAS,EAC9B,WAAM,MAAA,4DAGV,AAAI,MAAM,QAAQ,IAChB,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,+CACF,sBAAqB,EAAK,OAAM,QAE1C,KAAK,KAAO,EAER,EAAY,aAAe,MAC3B,EAAY,YAAY,MAAQ,KAClC,KAAM,IAAI,OACN,sEAEN,KAAK,YAAc,EAAY,aAAe,GAG1C,SAAA,UAAA,KAAN,SAAW,yGACT,GAAI,EAAe,wBAAyB,aAC1C,KAAM,IAAI,OACN,2FAIA,SAAO,OAAO,OAAO,CAAC,OAAQ,KAAK,gBAAiB,KAAK,aAC/D,EAAK,KAAO,GAAI,UAEV,EAAyC,CAAC,CAC9C,MAAO,CAAC,uBACR,QAAS,EAAe,cAEpB,EAA4C,CAChD,cAAe,EAAe,cAC9B,OAAQ,EAAe,OACvB,YAAa,EAAe,YAC5B,YAAa,EAAe,YAC5B,oBAAqB,EAAe,oBACpC,gBAAe,GAGjB,EAAK,KAAK,OACN,aACA,GAAI,MACA,CAAC,KAAK,UAAU,IAChB,CAAC,KAAM,KACX,cAEJ,AAAI,EAAe,YAAc,MAC/B,EAAK,KAAK,OACN,oBACA,GAAI,MAAK,CAAC,EAAe,YAAa,CAAC,KAAM,KAC7C,qBAGW,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,WAE7C,GAFM,EAAW,EAAA,OAEb,EAAS,GACX,MAAA,CAAA,EAAO,CACL,mBAAoB,GAA6B,GACjD,UAAW,CAAC,KAGd,KAAM,IAAI,OACN,gEACG,GAAS,OAAM,YAYpB,EAAA,UAAA,KAAN,uIAC6B,MAAA,CAAA,EAAM,KAAK,MAAM,KAAK,KAAM,KAAK,qBAE5D,GAFM,EAAqB,EAAA,OAEvB,CAAC,EAAmB,GACtB,KAAM,IAAI,OACN,cAAc,KAAK,KAAI,4BACpB,GAAmB,OAAM,uCAC5B,sFAIU,CAAA,EAAM,EAAmB,eAAvC,SAAc,EAAA,qCAEV,EAAU,+CAA+C,KAAK,KAAI,IAGtE,AAAI,KAAK,KAAK,SAAS,OACrB,GAAW,+UAOX,GAAW,uEAGP,GAAI,OAAM,UAUlB,GARM,EAAgB,EAAY,cAC5B,EAAkB,EAAY,gBAC9B,EAAc,EAAY,YAC1B,EAAc,EAAY,YAC1B,EAAS,EAAY,OACrB,EAAsB,EAAY,oBAGpC,GAAiB,MAAQ,GAAmB,KAC9C,KAAM,IAAI,OACN,2BAA2B,KAAK,KAAI,mEAMtC,IAAmB,KACL,CAAA,EAAM,KAAK,YAAY,IADrC,CAAA,EAAA,UACI,EAAU,EAAA,OACf,EAAA,EAAA,GAAa,EAAA,EAAA,oBAGV,SAA4B,CAChC,cAAa,EACb,YAAW,EACX,WAAU,EACV,oBAAmB,EACnB,YAAW,EACX,YAAW,EACX,OAAM,GAGF,EAAc,EAAY,iBAChC,AAAI,GACF,GAAU,iBAAmB,GAG/B,CAAA,EAAO,SAGK,EAAA,UAAA,YAAd,SAA0B,2IAOxB,IALM,EAAa,MAAM,QAAQ,KAAK,MAAQ,KAAK,KAAK,GAAK,KAAK,KAC5D,EAAmB,GAAS,GAA3B,EAAM,EAAA,GAAE,EAAM,EAAA,GACf,EAAa,KAAK,kBAAoB,EAEtC,EAAc,GACpB,EAAA,EAAoB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAT,EAAK,EAAA,GACd,EAAY,KAAI,MAAhB,EAAoB,EAAM,SAK5B,IAFM,EAAsB,GACtB,EAAsC,GAC5C,EAAA,EAA2B,EAAA,EAAA,EAAA,EAAA,OAAA,IACzB,IADS,EAAY,EAAA,GACrB,EAAA,EAAmB,EAAA,EAAa,MAAb,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,AAAI,KAAK,oBAAsB,KAC7B,EAAY,KAAK,KAAK,mBAAmB,IAEzC,EAAU,KAAK,EAAa,EAAO,SAKrC,MAAK,sBACP,GAAA,EAAU,MAAI,SAAd,GAAkB,CAAA,EAAM,QAAQ,IAAI,KADlC,CAAA,EAAA,UACF,EAAA,MAAA,EAAA,EAAA,OAAA,CAAkB,EAAA,2BAGJ,MAAA,CAAA,EAAM,GAAyB,EAAW,CACxD,YAAa,KAAK,YAClB,UAAW,KAAK,MAChB,WAAY,KAAK,qBAHb,SAAU,EAAA,OAKhB,CAAA,EAAO,CAAC,EAAa,GAAwB,WA5M/B,EAAA,iBAAmB,eA8MrC,iBAayB,GACvB,GAAM,GAAY,EAAI,YAAY,KAC5B,EAAkB,EAAI,YAAY,KAClC,EAAS,EAAI,UAAU,EAAG,GAC1B,EACF,EAAkB,EAAY,EAAI,UAAU,GAAmB,GACnE,MAAO,CAAC,EAAS,IAAK,eAGK,GAC3B,MAAO,GAAI,MAAM,GAAY,mBAAqB,KAG7C,GAAM,IACT,SAAC,EAAa,GACZ,GAAI,MAAO,QAAU,aAChB,IAAe,MAAQ,EAAY,WAAa,MAInD,MAAO,MAEP,GAAI,GAAS,GAMb,MALA,AAAI,OAAM,QAAQ,GAChB,EAAS,EAAI,MAAM,SAAA,GAAW,MAAA,IAAa,KAE3C,EAAS,GAAa,GAEpB,EACK,GAAK,EAAK,GAGd,MAEb,GAAiB,mBAAmB,IACpC,GAAiB,mBAAmB,gBAwEf,EAAc,GACjC,MAAO,IAAI,IAAY,EAAM,eAS3B,EAAc,GAChB,MAAO,IAAK,EAAM,GC3XpB,AAuBA,GAAA,IAAA,WACE,WAA6B,GAAA,KAAA,eAAA,EAEvB,SAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,qBAEhB,KAEA,GAAA,WACE,WACqB,GAAA,KAAA,YAAA,EAGf,SAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,YAAY,SAE5B,iBAwBI,EAAmC,EACnC,EAA0B,GAC5B,GAAI,UAAU,SAAW,GACvB,GAAM,GACD,EAAkC,eAAiB,MACnD,EAAkC,aAAe,KACtD,MAAI,GACK,GAAI,IAAkB,GAI7B,SAAQ,KACJ,yNAIG,GAAI,IAAkB,CAAC,cAAe,SAK/C,gBAAQ,KACJ,yNAIG,GAAI,IAAkB,CAC3B,cAAe,EACf,YAAW,EACX,WAAU,EACV,eAAc,gBAqBhB,GAEF,MAAO,IAAI,IAAiB,GCpH9B,oYCAA,AAuDA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MAExC,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,MAAK,GAC5B,EACW,SAAC,EAAwB,GACxC,SAAQ,GAA4B,EAAO,EAAG,MAC9C,EACI,EAAG,OAAS,GAAmB,GAC/B,WAAM,MAAA,mEACV,EAAK,CAAC,IACC,EAAQ,QAAQ,EAAI,IAE7B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAEO,GAAU,EAAG,CAAC,SAAQ,KC1EnC,AA6CA,YACI,EAAiB,EAAiB,EAClC,SADkC,AAAA,IAAA,QAAA,GAAA,IAClC,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAK,EAAgB,EAAG,IAAK,UACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WAAM,MAAA,kEACF,cAAa,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE3C,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,sCAAsC,EAAU,UAC/C,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,kCAAkC,EAAW,UAC5C,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACxD,EAAM,EAAa,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAExD,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,IAEJ,EAAQ,YACX,EAAiB,EAAiB,EAAY,IAG9C,EAA4B,CAAC,EAAG,EAAK,EAAG,GAExC,EAA0B,CAAC,WAAU,EAAE,WAAU,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAO,GAAQ,EAAK,MAGT,IAAS,EAAG,CAAC,QAAO,KC9GjC,AA+CA,YACI,EAA4B,EAAe,EAC3C,GACF,GAF6C,AAAA,IAAA,QAAA,GAAA,GAC3C,AAAA,IAAA,QAAA,GAAA,GACE,EAAQ,EACV,KAAM,IAAI,OAAM,iDAAiD,GAEnE,GAAM,GAAW,EAAgB,EAAS,UAAW,SAAU,SACzD,EAAe,EAAS,MAAK,OAAA,CAAE,IAE/B,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,IACC,EACH,EAAQ,OACJ,EAAQ,EAAU,CAAC,EAAS,OAAQ,EAAO,EAAS,GACxD,IAGA,EAAuB,CAAC,QAAS,GACjC,EAAqB,CAAC,MAAK,EAAE,QAAO,EAAE,SAAQ,GAEpD,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAS,EAAG,CAAC,QAAO,KCxEjC,AA+CA,YAAsC,EAAiB,GACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAgBnC,GAdA,AAAI,GAAQ,MACV,GAAO,EAAG,MAAM,IAAI,SAAC,EAAG,GAAM,MAAA,KAAG,WAEnC,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OACzC,EAAK,QAAQ,SAAA,GACX,EACI,GAAQ,GAAK,EAAO,EAAG,KACvB,WAAM,MAAA,+CAA+C,GAAG,KAAO,GAC3D,aAAY,OAGlB,EAAG,MAAQ,EACb,MAAO,GAAG,QAGZ,GAAM,GAA0B,CAAC,EAAG,GAC9B,EAAwB,CAAC,KAAI,GAEnC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IAAO,EACxC,KAAqB,GAAW,MAGzB,IAAY,EAAG,CAAC,WAAU,KC5EvC,YA0DI,EAA6B,EAC7B,GACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,mBAC5C,EACF,EAAgB,EAAa,cAAe,mBAEhD,EACI,GAAc,MAAQ,EAAa,GAAK,OAAO,UAAU,GACzD,WAAM,MAAA,uDACF,YAAW,KACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,gDAAgD,EAAQ,OAClE,EACI,EAAa,OAAS,EACtB,WAAM,MAAA,6CACF,YAAW,EAAa,QAChC,EACI,EAAQ,MAAM,KAAO,EAAa,MAAM,GACxC,WAAM,MAAA,uCACC,GAAQ,MAAM,GAAE,QAAQ,EAAa,MAAM,GAAE,MAChD,oEACR,EACI,EAAa,GAAK,OAAO,UAAU,GACnC,WAAM,MAAA,4DACF,IAAG,KAIX,GAAM,GAAe,GAAO,GAAK,EAAS,SAAU,GAC9C,EACF,GAAO,GAAK,EAAc,SAAU,GAClC,EAA0B,GAAU,GAC1C,MAAO,IAAK,GAAO,EAAe,GAAoB,SAGjD,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KC9FnD,2CCAA,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,mDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,oEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,AA6BA,GAAI,IAyBJ,YACI,EAEA,GAEF,GAFE,AAAA,IAAA,QAAA,GAAA,GAEE,EAAc,EAChB,KAAM,IAAI,OACN,kEAEN,GAAI,GAAU,KACZ,KAAM,IAAI,OAAM,4DAElB,GAAI,GAAc,GACd,EAAc,GACd,EAAU,GACV,EAAU,GACV,EAAe,GACnB,GAAK,EAAqB,eAAgB,YACxC,EAAc,WAEZ,MAAQ,YAAe,aAAe,YAAkB,WAC1D,EAAc,WAEZ,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAER,MAAQ,mBAAsB,aAC9B,YAAkB,kBACpB,EAAU,WAEA,EAAe,YAAc,KACvC,EAAe,OAEf,MAAM,IAAI,OACN,6OAIA,YAAY,EAAc,YAAY,OAE5C,GAAI,GACF,GAAM,GAAgC,EACtC,GAAI,GACC,EAA4B,WACzB,EACN,KAAM,IAAI,OACN,yGAMR,GAAM,GAAS,GAAU,GAAY,EAAO,aAC5C,GAAI,GAAU,MACZ,GAAM,GAA2B,CAAC,OAAM,GAClC,EAAyB,CAAC,YAAW,GAC3C,MAAO,GAAO,UACH,GAAY,EACZ,GAGP,GAAA,GAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAMV,EAEJ,AAAI,EACF,EAEK,EAAe,WAAW,MAAM,aAAa,EAAG,EAAG,EAAO,GAAQ,KAClE,AAAI,GAAe,EACxB,EAAQ,EAAiC,KACpC,AAAI,IAAW,IACpB,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAEpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA4B,EAAG,EAAG,EAAO,GAC7C,EAAO,GAAoB,aAAa,EAAG,EAAG,EAAO,GAAQ,MAE/D,GAAI,GACJ,GAAI,IAAgB,EAClB,EAAS,GAAI,YAAW,QAExB,GAAM,GAAY,EAAQ,EAC1B,EAAS,GAAI,YAAW,EAAY,GACpC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAU,EAAG,EAAU,EAAa,EAAE,EAC7C,EAAO,EAAI,EAAc,GAAW,EAAK,EAAI,EAAI,GAIvD,GAAM,GAAqC,CAAC,EAAQ,EAAO,GAC3D,MAAO,IAAS,EAAQ,EAAU,qBAuBhC,EACA,iIAQF,GAPI,EAAO,EAAgB,EAAK,MAAO,YACvC,AAAM,YAAe,IAEb,GAAoB,EAC1B,EAAO,GAAK,EAAmB,SAC/B,EAAkB,WAEhB,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,OACN,wDAAwD,EAAK,KAAI,KAKvE,GAHM,EAAkB,EAAK,MAAM,MAAM,EAAG,GAArC,EAAM,EAAA,GAAE,EAAK,EAAA,GACd,EAAQ,EAAK,OAAS,EAAI,EAAI,EAAK,MAAM,GAE3C,EAAQ,GAAK,IAAU,EACzB,KAAM,IAAI,OACN,wCACA,sBAAqB,IAG3B,GAAI,EAAK,QAAU,WAAa,EAAK,QAAU,QAC7C,KAAM,IAAI,OACN,kCAAkC,EAAK,MAAK,0CAIrC,MAAA,CAAA,EAAM,EAAK,eAIxB,IAJM,EAAO,EAAA,OACP,EAAa,EAAK,QAAU,UAAY,IAAM,EAC9C,EAAQ,GAAI,mBAAkB,EAAQ,EAAS,GAE5C,EAAI,EAAG,EAAI,EAAS,EAAO,EAAE,GAGpC,IAFM,EAAO,CAAC,EAAG,EAAG,EAAG,KAEd,EAAI,EAAG,EAAI,EAAO,KAGzB,GAFM,EAAQ,EAAK,EAAI,EAAQ,GAE3B,EAAK,QAAU,WACjB,GAAI,EAAQ,GAAK,EAAQ,EACvB,KAAM,IAAI,OACN,qDACA,kCAAiC,EAAK,cAEnC,EAAK,QAAU,SACpB,GAAQ,GAAK,EAAQ,KACvB,KAAM,IAAI,OACN,mDACA,oCAAmC,EAAK,MAIhD,AAAI,IAAU,EACZ,GAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,EAClB,EAAK,GAAK,EAAQ,GAElB,EAAK,GAAK,EAAQ,EAIhB,EAAI,EAAI,EACd,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAC/B,EAAM,EAAI,GAAK,KAAK,MAAM,EAAK,IAGjC,MAAI,IAAU,MACZ,GAAO,MAAQ,EACf,EAAO,OAAS,EACV,EAAM,EAAO,WAAW,MACxB,EAAY,GAAI,WAAU,EAAO,EAAO,GAC9C,EAAI,aAAa,EAAW,EAAG,IAEjC,AAAI,IAAS,GACX,EAAK,UAEP,CAAA,EAAO,QAGF,GAAM,IAAa,EAAG,CAAC,YAAW,+DCrOrC,EAAgB,GAClB,GAAI,EAAO,KAAO,EAChB,KAAM,IAAI,OACN,0DACA,sBAAqB,EAAO,KAAI,MAEtC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,4DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OACN,sDACA,uBAAsB,EAAQ,MAAK,MAEzC,GAAI,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAAO,KAC3C,KAAM,IAAI,OACN,iEACG,GAAQ,MAAM,EAAQ,KAAO,GAAE,QAAQ,EAAO,OAGvD,GAAI,EAAO,OAAS,EAClB,KAAM,IAAI,OACN,qDACA,kBAAiB,EAAO,MAAK,MASnC,OANM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAIjD,EAAU,EACL,EAAI,EAAG,EAAI,EAAa,OAAS,EAAG,EAAE,EAC7C,GAAW,EAAa,GAG1B,GAAM,GAAa,EAAO,MAEpB,EAAc,EAAa,QACjC,EAAY,MAGZ,OADI,GAAY,EACP,EAAI,EAAW,EAAI,EAAO,KAAM,EAAE,EACzC,GAAa,EAAW,GACxB,EAAY,KAAK,EAAW,IAG9B,GAAM,GACE,GAAe,EAAO,OAAO,IAAI,SAAA,GAAU,MAAA,GAAS,IAAU,OAAA,CACjE,IAAG,MAAM,EAAG,GAEjB,MAAO,CAAC,EAAa,EAAS,EAAW,6DCtDvC,EAAiB,EAAiB,GACpC,GAAM,GAAY,EAAQ,KAAO,EAAK,EAAQ,MAAM,EAAQ,KAAO,GAAK,EAClE,EAAY,EAAQ,KAAO,EAAK,EAAQ,KAAO,EAAI,EAEnD,EAAa,wDACf,yCAAwC,EAAQ,OAChD,qBAAoB,EAAQ,MAAK,YAAY,GAC7C,gBAAe,EAAQ,mBAAmB,EAAQ,KAEtD,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OAAM,EAAa,mBAAkB,EAAQ,OAEzD,GAAI,EAAM,OAAS,EAAY,GAAQ,KAAO,GAC5C,KAAM,IAAI,OACN,EACA,2BAA0B,GAAY,GAAQ,KAAO,MAE3D,GAAI,EAAQ,OAAS,EAAW,EAAM,OAAS,EAC7C,KAAM,IAAI,OACN,EAAa,oBAAmB,GAAW,EAAM,OAAS,KAEhE,OAAS,GAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,GAAI,EAAQ,MAAM,KAAO,EAAQ,MAAM,GACrC,KAAM,IAAI,OACN,EACA,mBAAkB,EAAC,MAAM,EAAQ,MAAM,GAAE,sBAAsB,EAAC,MAC5D,EAAQ,MAAM,GAAE,OAG5B,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAO,EAAU,EAAE,EAC7C,GAAI,EAAQ,MAAM,EAAI,KAAc,EAAM,EAAI,GAC5C,KAAM,IAAI,OACN,EACA,mBAAkB,GAAI,GAAQ,MAC1B,EAAQ,MAAM,EAAI,GAAS,cAAc,GAAI,GAAQ,MACrD,EAAM,EAAI,GAAS,kBAoB7B,EAAiB,EAAiB,GACpC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,KAAO,EACjB,KAAM,IAAI,OACN,6DACA,sBAAqB,EAAQ,KAAI,MAEvC,GAAI,EAAQ,QAAU,QACpB,KAAM,IAAI,OAAM,0DACZ,EAAQ,OAEd,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,OACN,6DAA6D,GAGnE,GAAI,EAAM,SAAW,GACnB,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAEd,GAAI,EAAQ,OAAS,EACnB,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAIhB,GAAoB,EAAO,EAAS,eAalC,EAAqB,EACrB,GAWF,OATM,GAAc,EAAQ,MAAM,OAC5B,EAAa,EAAc,EAAK,EAAQ,MAAM,EAAc,GAAK,EAKjE,EAAU,EAAM,OAElB,EAAY,EACP,EAAI,EAAW,EAAI,EAAS,EAAE,EACrC,GAAa,EAAM,GAGrB,GAAM,GAAgB,EAAY,EAAK,EAAI,EACrC,EAAa,GAAc,EAAQ,OAAS,EAE5C,EAAc,GAAe,EAAM,MAAM,EAAG,IAAW,OAAA,CAAE,IACzD,EAAa,GAAc,GACjC,MAAO,CAAC,UAAS,EAAE,WAAU,EAAE,UAAS,EAAE,QAAO,EAAE,WAAU,sFCnJ/D,YAqBI,EAAmB,EAAiB,GACtC,GAAM,GAAY,EAAM,MAAM,OAC9B,EACI,IAAc,EAAM,OACpB,WAAM,MAAA,iBAAiB,EAAS,sBAAsB,EAAK,SACvD,iCAAgC,EAAS,QACjD,EACI,IAAc,EAAK,OACnB,WAAM,MAAA,iBAAiB,EAAS,qBAAqB,EAAI,SACrD,iCAAgC,EAAS,QAEjD,mBAAS,GACP,EACI,EAAM,GAAK,EAAK,IAAM,EAAM,MAAM,GAClC,WAAM,MAAA,iBAAiB,EAAS,YAAY,EAAC,YAAY,EAAC,KACtD,KAAI,GAAM,GAAK,EAAK,IAAE,gCAAgC,EAAC,MACjD,EAAM,MAAM,GAAE,QALrB,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,eAUgB,GAGzB,OAFM,GAAO,GACT,EAAO,EACJ,EAAO,GACZ,AAAI,EAAO,GACT,EAAK,KAAK,GAEZ,GAAQ,EACR,IAEF,MAAO,eAKL,EAAiB,EAAe,GAElC,OADM,GAAO,GACJ,EAAO,EAAG,EAAO,EAAM,OAAQ,IACtC,EAAK,GAAQ,KAAK,KAAM,GAAI,GAAQ,EAAM,IAAS,EAAQ,IAE7D,MAAO,eAML,EAAmB,EAAgC,EACnD,GAEF,OADM,GAAiB,EAAO,QACrB,EAAI,EAAW,OAAQ,EAAI,EAAW,OAAQ,IACrD,EAAW,KAAK,GAElB,OAAS,GAAI,EAAG,EAAI,EAAe,IACjC,AAAI,IAAM,EACR,EAAW,GAA0B,EAErC,GAAW,OACP,EAAwB,EACxB,GACJ,EAAW,OAGf,MAAO,GAGT,YACI,EAAgC,EAChC,GACF,MAAI,IAAkB,EACb,EAGF,EAAkB,GAAgB,GAG3C,YAAuB,EAAuB,GAE5C,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAe,IACjC,EAAW,KAAK,EAAyB,GAE3C,MAAO,eAKL,EAAsB,EAAwB,EAC9C,EAAiB,EAAe,EAAmB,EACnD,EACA,GACF,GAAM,GAAY,EAAW,OACzB,EAAkB,GAAI,OAAM,GAC5B,EAAgB,GAAI,OAAM,GAC1B,EAAoB,GAAI,OAAM,GAClC,GAAI,EAAa,QAAU,EAAsB,GAC/C,GAAM,GAAY,EAAa,GAIzB,EAAgB,EAAsB,EAC5C,EAAkB,GACd,EAAW,EAAW,EAAe,EAAO,GAChD,EAAgB,GACZ,EAAS,EAAW,EAAe,EAAK,GAC5C,EACI,GAAsB,EAAS,EAAW,EAAe,OAE7D,QAAS,GAAO,EAAG,EAAO,EAAW,IACnC,EAAgB,GAAQ,GACpB,EAAW,EAAO,EAAS,EAAY,EAAM,GACjD,EAAc,GACV,GAAY,EAAS,EAAK,EAAS,EAAY,EAAM,GACzD,EAAkB,GAAQ,GAAe,EAAS,EAAM,GAI5D,MAAO,CACL,MAAO,EACP,IAAK,EACL,QAAS,eAOT,EAAmB,EAAgC,EACnD,EAAyB,GAI3B,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAEnB,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAc,GAClC,AAAI,EAAY,GAAK,GACnB,GAAgB,GAGlB,EAAW,GAAQ,EAGvB,MAAO,eAML,EAAiB,EAAgC,EACjD,EAAuB,GAIzB,OAHM,GAAiB,EAAU,QAC3B,EAAa,GAAc,EAAe,GAEvC,EAAO,EAAG,EAAO,EAAW,OAAQ,IAC3C,GAAI,EAAW,QAAQ,GAAQ,GAC7B,EAAW,GAAQ,OAAO,sBAE1B,GAAM,GACF,GAAgB,EAAwB,EAAe,GACvD,EAAgB,EAAY,GAChC,AAAI,EAAU,GAAK,GACjB,GAAgB,OAAO,kBAEzB,EAAW,GAAQ,EAIvB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,KAErC,GAAM,GAAW,EAAW,GAC5B,AAAI,EAAW,GAAK,GAClB,GAAW,IAAM,GAEnB,EAAW,GAAK,GAAW,EAAG,EAAW,GAAI,EAAW,IAE1D,MAAO,eAIL,EAAmB,EAAc,GACnC,GAAI,GAAS,EAAQ,GACrB,MAAI,GAAgB,GAAK,GAAS,GAAU,OAC1C,GAAS,GAGJ,cAIL,EAAmB,EAAwB,EAC3C,EAAsB,EAAc,GAEtC,GAAI,GAAQ,EAAa,GACnB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAY,GAAK,GAAQ,EAAe,GAAK,GAAQ,GAAS,OAChE,CAAI,EAAS,EAIX,EAAQ,OAAO,iBAGf,EAAQ,OAAO,kBAKnB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAQ,GACV,IAAS,GAIX,EAAQ,GAAW,EAAG,EAAO,EAAW,GAEjC,cAIL,EAAiB,EAAuB,EACxC,EAAsB,EAAc,GAEtC,GAAI,GAAO,EAAY,GACjB,EAAS,EAAQ,IAAS,EAIhC,AAAI,GAAW,GAAK,GAAS,EAAgB,GAAK,GAAS,GAAQ,OACjE,CAAI,EAAS,EAGX,EAAO,OAAO,iBAGd,EAAO,OAAO,kBAKlB,GAAM,GAAW,EAAW,GAC5B,MAAI,GAAO,GACT,IAAQ,GAMV,AAAI,EAAS,EAEX,EAAO,GAAW,EAAG,EAAM,GAG3B,EAAO,GAAW,GAAI,EAAM,EAAW,GAGlC,cAQL,EAAiB,EAAiB,GAGpC,OADI,GAAkB,EAAK,OAClB,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,GAAI,EAAK,GAAK,GACZ,EAAkB,EAClB,MAIJ,OAAS,GAAI,EAAkB,EAAG,EAAI,EAAK,OAAQ,IACjD,GAAI,EAAM,GAAK,GAAK,EAAK,KAAO,EAAM,GACpC,MAAO,GAGX,MAAO,eAGyB,EAAiB,GAEjD,OADI,GAAa,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EACrD,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAc,EAAM,GAAK,EAAQ,GAEnC,MAAO,eAIL,EAAe,EAAwB,GAEzC,GAAI,GACE,EAAQ,EAAE,MAAM,OACtB,AAAI,MAAO,IAAU,SACnB,EAAM,CAAI,GAAK,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,IACzC,AAAI,EAAM,OAAS,EACxB,EAAS,EAAM,OAAO,GAAI,OAAM,EAAQ,EAAM,QAAQ,KAAK,IAE3D,EAAS,EAAM,QAEjB,EAAO,QAAQ,SAAA,GACb,EACI,IAAM,GAAI,WAAM,MAAA,wDAEtB,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAAI,OAAM,GAAO,KAAK,IACzB,AAAI,MAAO,IAAS,SACzB,EAAK,CAAI,GAAI,OAAK,GAAI,OAAM,EAAQ,GAAG,KAAK,KACvC,AAAI,EAAK,OAAS,EACvB,EAAQ,EAAK,OAAO,GAAI,OAAM,EAAQ,EAAK,QAAQ,KAAK,KAExD,EAAQ,EAEV,EAAQ,EAAM,IAAI,SAAC,EAAG,GACpB,MAAI,IAAK,EACA,EAEP,GACI,IAAM,GACN,WAAM,MAAA,qDACC,GAAC,kCAAkC,EAAC,OACxC,EAAE,MAAM,GAAK,EAAO,MAGxB,CAAC,EAAQ,wSCjWlB,AA8DA,GAAA,IAAA,WAAA,cAYE,SAAA,UAAA,aAAA,WACE,MAAQ,MAAK,YACR,WAiBA,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,IAEnB,KASA,GAAA,WAOE,aACE,KAAK,aAAe,GAMf,SAAA,OAAP,WACE,MAAI,GAAiB,UAAY,MAC/B,GAAiB,SAAW,GAAI,IAE3B,EAAiB,UAMnB,EAAA,SAAP,SAAwC,GACtC,EAAiB,SAAS,aAAa,EAAI,WACvC,CAAC,EAAK,EAAI,aAElB,iBA2BI,GACF,EACI,EAAI,WAAa,KACjB,WAAM,MAAA,gFAEV,EACI,MAAO,GAAI,WAAc,SACzB,WAAM,MAAA,sDACF,MAAO,GAAI,YACnB,EACI,EAAI,UAAU,OAAS,EACvB,WAAM,MAAA,sFAGV,GAAiB,SAAS,gFC/K5B,AAsBA,GAAM,IAAuB,KAChB,GAAuB,eAGhC,EACA,EAAoD,GACtD,MAAI,IAAW,MACb,GAAU,MAEL,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,mBAInE,MAAO,GAAO,QAAQ,mBAAqB,GAAK,GACA,GAGlD,YACI,EAAoB,EACpB,GACF,GAAI,GAAiB,GAOrB,GANA,AAAI,IAAa,IAAW,GAAa,KACvC,GAAiB,IAEnB,AAAI,GAAa,IAAW,GAAa,IACvC,GAAiB,IAEf,GACF,GAAM,GAAQ,EAAO,YAAY,KAC3B,EAAQ,EAAS,YAAY,KAEnC,GAAI,IAAU,EACZ,KAAM,IAAI,OACN,yCAAyC,EAAK,KAC9C,cAAa,IAIrB,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,IACzC,GAAM,GAAc,GAAW,GACzB,EAAgB,GAAW,GACjC,GAAI,CAAC,GAAY,EAAa,GAC5B,KAAM,IAAI,OACN,iCACA,aAAY,EAAW,iBAAiB,EAAa,MAI7D,GAAM,GACF,GAAa,GAAU,EAAS,GAAQ,GACtC,EAAe,GAAa,GAC9B,EACA,GAAQ,GAEZ,GAAI,EAAW,SAAW,EAAa,OACrC,KAAM,IAAI,OACN,yCAAyC,EAAW,OAAM,OAC1D,cAAa,EAAa,OAAM;GAChC,cAAa,EAAU;GACvB,cAAa,EAAY,MAE/B,OAAS,GAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAI,EAAW,GACf,EAAI,EAAa,GAEvB,GAAI,CAAC,EAAU,EAAG,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAC,OAAO,EAAC,cAAc,EAAC,OAAO,EAAC;EACzD,cAAa,EAAU;GACvB,cAAa,EAAY,mBAUC,EAAuB,GACzD,IAAK,KAAK,WAAM,MAAA,GAAK,QAAQ,WAAM,MAAA,mBAGH,EAAoB,GACpD,GAAM,GAAM,MAAO,IAAa,UAAY,MAAO,IAAa,UACxD,MAAO,IAAa,UACxB,CAAC,GACD,EACJ,MAAI,IAAS,IAAW,GAAU,EAAoB,KAClD,GAAS,IAAa,GAAU,EAAsB,IAEjD,GAAsB,EAAQ,EAAK,SAAC,EAAG,GAAM,MAAA,IAAK,IAEpD,GACH,EAAQ,EAAU,SAAC,EAAG,GAAM,MAAA,IAAS,EAAa,EAAa,iBAGlC,EAAW,EAAW,GAIvD,GAHA,AAAI,GAAW,MACb,GAAU,MAER,CAAC,GAAS,EAAG,EAAG,GAClB,KAAM,IAAI,OAAM,8BAA8B,EAAC,kBAAkB,GAIrE,YAAkB,EAAW,EAAW,GACtC,MAAI,CAAC,SAAS,IAAM,CAAC,SAAS,GACrB,GAEL,QAAM,IAAM,MAAM,IAAM,KAAK,IAAI,EAAI,GAAK,eAO5C,EAA6B,EAAa,GAC5C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,GAAI,EAAO,GAAK,GAAO,EAAO,GAAK,EACjC,KAAM,IAAI,OACN,sBAAsB,EAAO,GAAE,SAAS,EAAG,WAAW,eAM5D,EAAqB,GAGvB,OAAO,GAAI,cAAa,IAAS,QAAQ,GAAI,cAAa,4MCzJ5D,GAGM,IAAU,QCHhB,cAiCE,KAAM,IAAI,OAAQ,kBAkBlB,KAAM,IAAI,QAAS,kBAKnB,KAAM,IAAI,+BAAgC,IAC1C,QAAQ,KAAK,sEAIiB,GAC9B,AAAI,KAAM,QAAQ,iCAChB,QAAQ,KACJ,EAAM,6FAYZ,EAAO,iCASP,MAAO,iBAyBP,MAAO,GAAO,qBAgCQ,GAEtB,MAAO,GAAO,QAAQ,eA4CpB,EAA6B,GAC/B,MAAO,GAAO,KAAK,EAAU,eAcP,GACtB,GAAM,GAAU,GAAsB,GACtC,EAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,wBAmCI,GACrC,MAAO,GAAO,KAAK,eA4BA,GACnB,MAAO,GAAO,KAAK,eAkBM,GACzB,MAAO,GAAO,WAAW,iBAWzB,MAAO,GAAO,sBAUd,MAAO,GAAO,wBAQc,GAC5B,EAAO,cAAc,eAOK,GAC1B,MAAO,GAAO,YAAY,eAQO,GAEjC,MAAO,GAAO,mBAAmB,eAmB/B,EAAc,EACd,GAAA,MAAA,KAAA,QAAA,GAAA,GACK,EAAO,gBAAgB,EAAM,EAAS,iBAY7C,MAAO,GAAO,oBASY,EAAsB,GAChD,KAAM,YAAY,EAAc,GC5XlC,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCnE3B,AAqDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCvErC,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OAGjC,GAFA,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAED,EAAG,QAAU,SAAW,EAAG,QAAU,QACvC,MAAO,IAAS,EAAI,GAGtB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,WAAW,EAAI,GACnC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAC/B,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KC3E3B,AAoDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAEH,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAEA,GAAM,EAAG,CAAC,KAAI,KCpE3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GAEpC,MADA,GAAK,CAAC,IACF,EAAG,QAAU,YACR,EAAQ,WAAW,GAGrB,EAAQ,IAAI,IAClB,EAAgC,KAAiB,OAGzC,IAAM,EAAG,CAAC,KAAI,KCrD3B,AAoCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAuCA,YAAiC,GAC/B,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,+DACV,EACI,EAAQ,QAAU,EAClB,WAAM,MAAA,uDACF,IAAG,EAAQ,UAEnB,GAAM,GACF,EAAQ,IAAI,SAAC,EAAG,GAAM,MAAA,GAAgB,EAAG,UAAU,EAAK,UAEtD,EAAc,EAAS,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAI,EAAE,QAAU,EAAY,MAC1B,KAAM,IAAI,OACN,8DAIR,EAAS,QAAQ,SAAA,GACf,GAAI,CAAC,GAAiB,EAAE,MAAO,EAAY,OACzC,KAAM,IAAI,OACN,8DAIR,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,GACE,GAGH,EAAqB,EAE3B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC/E7B,YAuBqC,EAAgB,GACnD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,GAAI,EAAK,EAAK,OAAS,EAAI,KAAO,EAAO,EAAI,EAC3C,MAAO,GAGX,MAAO,eAIL,EAAqB,EAAqB,GAK1C,OAJI,GAAO,EAAU,OAAS,EAAU,OACpC,EAAM,GACR,EAAS,EACT,EAAY,EACL,EAAM,EAAG,EAAM,EAAM,IAC9B,AAAI,EAAK,QAAQ,KAAS,GACxB,EAAI,KAAK,EAAU,MAEnB,EAAI,KAAK,EAAU,MAGvB,MAAO,eAIL,EAAkB,GAGpB,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,EAAK,QAAQ,KAAS,IACxB,EAAS,KAAK,EAAO,IAGzB,GAAM,GAAc,EAAK,IAAI,SAAA,GAAO,MAAA,GAAO,KAC3C,MAAO,CAAC,EAAU,eAIhB,EAAiB,GACnB,GAAM,GAAiB,EAAK,IAAI,SAAA,GAAK,MAAA,KACrC,MAAO,IAAiB,EAAO,EAAgB,eAI7C,EAAa,EAAgB,GAC/B,EACI,GAAqB,EAAM,GAC3B,WAAM,MAAG,GAAG,2CACR,aAAY,EAAI,aAAa,EAAI,yBAQR,EAAgB,GAEjD,GAAI,GAAqB,EAAM,GAC7B,MAAO,MAGT,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,AAAI,EAAK,QAAQ,KAAO,IACtB,EAAO,KAAK,GAGhB,SAAK,QAAQ,SAAA,GAAQ,MAAA,GAAO,KAAK,KAC1B,cAI8B,GACrC,MAAO,GAAK,IAAI,SAAC,EAAM,GAAM,MAAA,CAAC,EAAG,KAC5B,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAE,KACxB,IAAI,SAAA,GAAK,MAAA,GAAE,iBAGe,EAAiB,GAEhD,OADM,GAAgB,GACb,EAAI,EAAO,EAAS,EAAI,EAAM,EAAE,EACvC,EAAI,KAAK,GAEX,MAAO,GC3GT,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KCxF3B,AA4DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,MAAO,QAElC,EAA+B,SAAC,GACpC,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAAmB,EAAM,EAAG,MACjD,AAAI,GAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAAiB,EAAK,OAAQ,EAAG,OAE1C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAIA,IAAM,EAAG,CAAC,KAAI,KCzF3B,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC7EjC,AAsDA,YAAmC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACvD,GAAI,GAAK,EAAgB,EAAG,IAAK,UAE3B,EAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IAEN,AAAI,GAAQ,MACV,GAAO,GAET,GAAI,GAAO,GAAoB,EAAM,EAAG,OAClC,EAAe,GAA6B,EAAM,EAAG,MAC3D,MAAI,IAAgB,MAClB,GAAK,GAAU,EAAI,GACnB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAE7C,EAAQ,OAAO,EAAI,EAAK,KAG3B,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KChFjC,AAqCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KC/C7B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA2CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAEH,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAQ,EAAG,CAAC,OAAM,KC7D/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,YAoGI,EACA,EAAuC,EACvC,EAA4B,EAC5B,GAD4B,AAAA,IAAA,QAAA,GAAA,QAM9B,GAAM,GAAgB,EAAW,GAC3B,EACE,EAAW,OAAA,CAAE,IACf,EAAc,GAAwB,GAE5C,MAAO,IACH,EAAY,EAAc,EAAS,EAAW,EAC9C,KAAyB,KAAsB,eAIjD,EACA,EAAqC,EACrC,EAAoC,EACpC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,gBACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GAEjB,EACJ,GAAI,IAAe,eACjB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACrD,IAAe,gBACxB,EAAc,CAAC,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAE9D,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,EAAc,GAC7D,eAOF,EACA,EACA,EACA,EAA4C,EAC5C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,SACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAE9B,EACA,EACJ,GAAI,IAAe,QACjB,EAAc,eACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,YACxD,IAAe,QACxB,EAAc,gBACd,EACI,CAAC,EAAa,EAAc,EAAa,EAAQ,GAAI,EAAQ,QAEjE,MAAM,IAAI,OAAM,sBAAsB,GAGxC,MAAO,IACH,EAAS,EAAa,EAAS,EAAW,EAAK,GAAO,EACtD,eAQF,EACA,EACA,EAAkC,EAClC,EACA,EAAuC,EACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACvC,AAAA,IAAA,QAAA,GAAA,gBACE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GACnC,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WACtB,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAU,EAAA,EAAA,OAElC,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAC9B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAc,EAAA,EAAA,GACf,EAAA,GAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAEjB,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,SAIrB,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAW,GAC1C,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAW,EAAU,IAGvC,CACL,UAAS,EACT,WAAU,EACV,SAAQ,EACR,QAAO,EACP,WAAU,EACV,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,aAAY,EACZ,YAAW,EACX,aAAY,EACZ,YAAW,EACX,sBAAqB,EACrB,qBAAoB,EACpB,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,eA2CX,EACA,EACA,EACA,EAA4C,EAC5C,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,gBAEE,GAAA,GAAA,CAAA,GAAA,GAAA,GAAA,GAAA,IAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,GAE5C,GAAI,IAAe,eAChB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,GAAS,EAAA,EAAA,WAC/B,IAAe,gBACvB,EAAA,EAAA,GAAW,EAAA,EAAA,GAAY,EAAA,EAAA,GAAS,EAAA,EAAA,GAAU,EAAA,EAAA,OAE3C,MAAM,IAAI,OAAM,sBAAsB,GAGjC,GAAA,GAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAAe,EAAA,EAAA,GAE3C,EAAA,GAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAc,EAAA,EAAA,GAC5B,EAAA,GAAA,GAAC,EAAA,EAAA,GAAe,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGhC,EACF,GAAuB,EAAa,GAClC,EACF,GAAuB,EAAc,GACnC,EACF,GAAuB,EAAa,GAClC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,SAAU,EAAA,EAAA,UAAW,EAAA,EAAA,SAK/B,EAAc,EAAY,EAAiB,EAAa,EAE1D,EACJ,MAAI,KAAe,gBACjB,EAAW,CAAC,EAAW,EAAa,EAAU,EAAW,GACpD,AAAI,IAAe,gBACxB,GAAW,CAAC,EAAW,EAAU,EAAW,EAAU,IAGjD,CACL,UAAS,EACT,WAAU,EACV,QAAO,EACP,SAAQ,EACR,QAAO,EACP,WAAU,EACV,SAAQ,EACR,UAAS,EACT,SAAQ,EACR,YAAW,EACX,QAAO,EACP,YAAW,EACX,aAAY,EACZ,YAAW,EACX,YAAW,EACX,aAAY,EACZ,YAAW,EACX,qBAAoB,EACpB,sBAAqB,EACrB,qBAAoB,EACpB,cAAa,EACb,eAAc,EACd,cAAa,EACb,QAAO,EACP,SAAQ,EACR,YAAW,GAIf,YACI,EAA2B,EAAmB,EAC9C,EAAkB,GACpB,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAY,GAGtB,YACI,EAA2C,EAC3C,EAAqB,EAAgB,EACrC,GACF,AAAI,GAAW,MACb,GAAU,GAAkB,EAAS,EAAW,IAElD,GAAM,GAAa,EAAQ,GACrB,EAAY,EAAQ,GACpB,EAAY,EAAQ,GAEpB,EAAe,GAChB,GAAa,EAAY,EAAI,GAAW,EAAS,EAAG,GACzD,EACI,GAAW,GACX,WAAM,MAAA,2BAA2B,EAAY,uEAGjD,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,EACI,GAAW,GACX,WAAM,MAAA,yBAAyB,EAAU,uEAG7C,GAAM,GAAa,GACd,GAAY,EAAY,EAAI,GAAW,EAAS,EAAG,GACxD,SACI,GAAW,GACX,WAAM,MAAA,4BAA4B,EAAU,uEAGzC,CAAC,EAAc,EAAY,EAAY,eAI5C,EACA,EAAmB,EAAgB,GAAA,AAAA,IAAA,QAAA,GAAA,GACrC,GAAM,GAAqB,GAAuB,EAAW,GAC7D,MAAO,MAAK,MACP,GAAW,GAAM,GAAS,GAAK,EAAS,GAAsB,GAGrE,YAAyB,GACvB,MAAI,OAAO,IAAU,SACZ,CAAC,EAAO,EAAO,GAEpB,EAAM,SAAW,EACZ,CAAC,EAAM,GAAI,EAAM,GAAI,GAEvB,EAGT,YAA0B,GAExB,MAAO,OAAO,IAAU,SAAW,CAAC,EAAO,EAAO,GAAS,EAc7D,YAAgC,EAAoB,GAClD,MAAI,IAAY,EACP,EAGF,EAAc,GAAa,GAAM,GAAW,GAGrD,YACI,EAA4C,EAC5C,EAAiB,EAAsB,EACvC,EAAsB,EACtB,EACA,GAEF,GAAI,GACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CAAC,IAAK,EAAK,OAAQ,EAAK,KAAM,EAAK,MAAO,EAAK,KAAM,GAC/D,GAAM,GAAW,GACb,CAAC,EAAU,GAAU,EAAc,EAAc,EAAK,GAC1D,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GACF,KAAK,IAAI,EAAI,GAAY,GAAK,EAAe,EAAe,GAC1D,EACF,KAAK,IAAI,EAAI,GAAW,GAAK,EAAc,EAAc,GACvD,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAC9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,gBAClC,IAAQ,QACjB,EAAU,CAAC,IAAK,EAAG,OAAQ,EAAG,KAAM,EAAG,MAAO,EAAG,KAAM,SACvD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,WAC1C,MAAO,IAAQ,UACxB,GAAM,GAAM,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GACzD,EAAS,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC5D,EAAO,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC1D,EAAQ,IAAe,eAAiB,EAAI,GAAG,GAAK,EAAI,GAAG,GAC3D,EAAW,IAAQ,GAAK,IAAW,GAAK,IAAS,GAAK,IAAU,EAClE,QACA,WACJ,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,KAAM,GAC3C,EAAY,GACP,GAAW,EAAe,EAAM,GAAU,EAAe,EAC1D,GACJ,EAAW,GACN,GAAU,EAAc,EAAO,GAAS,EAAc,EAAG,OAE9D,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,UAAS,EAAE,SAAQ,GAGtC,YACI,EAA4B,EAAiB,EAC7C,EAAiB,EAAqB,EACtC,EAAqB,EAAqB,EAC1C,EAAqB,GAMvB,GAAI,GACA,EACA,EACA,EAEJ,GAAI,MAAO,IAAQ,UACjB,GAAM,GAAW,IAAQ,EAAK,QAAU,SACxC,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,GAER,GAAM,GAAW,GACb,CAAC,EAAS,EAAU,EAAS,GAAI,EAAa,EAAG,EAAa,EAC9D,GACJ,EAAW,EAAS,GACpB,EAAY,EAAS,GACrB,EAAW,EAAS,WACX,IAAQ,QACjB,EAAW,KAAK,KAAK,EAAU,GAC/B,EAAY,KAAK,KAAK,EAAW,GACjC,EAAW,KAAK,KAAK,EAAU,GAC/B,GAAM,GAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EACD,GAAY,GAAK,EAAe,EAAe,EAC9C,EAAiB,GAAW,GAAK,EAAc,EAAc,EAC7D,EAAQ,KAAK,MAAM,EAAgB,GACnC,EAAO,EAAgB,EACvB,EAAM,KAAK,MAAM,EAAiB,GAClC,EAAS,EAAiB,EAC1B,EAAO,KAAK,MAAM,EAAgB,GAClC,EAAQ,EAAgB,EAE9B,EAAU,CAAC,IAAG,EAAE,OAAM,EAAE,KAAI,EAAE,MAAK,EAAE,MAAK,EAAE,KAAI,EAAE,KAAM,gBAC/C,IAAQ,QACjB,EAAU,CACR,IAAK,EACL,OAAQ,EACR,KAAM,EACN,MAAO,EACP,MAAO,EACP,KAAM,EACN,KAAM,SAER,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,GACnD,EAAY,KAAK,KAAM,GAAW,EAAe,GAAK,GACtD,EAAW,KAAK,KAAM,GAAU,EAAc,GAAK,OAEnD,MAAM,OAAM,8BAA8B,GAE5C,MAAO,CAAC,QAAO,EAAE,SAAQ,EAAE,UAAS,EAAE,SAAQ,GAQhD,YACI,EAAe,GACjB,GAAI,CAAC,EACH,MAAO,GAET,OAAQ,OACD,QAEH,MAAO,MAAK,MAAM,OACf,OAEH,MAAO,MAAK,KAAK,OACd,QACH,MAAO,MAAK,MAAM,WAElB,KAAM,IAAI,OAAM,wBAAwB,gBAIZ,GAC1B,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,MAAO,KAAS,GAAK,IAAS,GAAK,IAAS,cAI1C,EAA0B,GAC5B,MAAO,IAAkB,IAAY,GAAkB,eAUjB,GAEtC,GAAI,IAAe,OACjB,MAAO,eACF,GAAI,IAAe,OACxB,MAAO,gBAEP,KAAM,IAAI,OAAM,sBAAsB,GC5mB1C,AAoDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,WACxC,EAAY,EAElB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,mDAAmD,EAAI,KAAI,MAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAIJ,MAFA,GAAK,CAAC,IAEF,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UACvC,EAAI,QAGN,EAAQ,QAAQ,EAAK,IAGxB,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAElE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAG,OAEf,EACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KClHnC,AA6EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,WAE5C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAGrB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GAEJ,SAAK,CAAC,IAEC,EAAQ,UAAU,EAAK,IAG1B,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAEjE,EAAM,EAAO,cACb,EAAS,EAAgC,KAAiB,GAC1D,GAIJ,MAFA,GAAM,GAAK,EAAK,EAAI,OAEhB,EACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KCzJvC,YAmBuC,EAAoB,GACzD,GAAM,GAAO,EAAO,GAAG,OACvB,EAAO,QAAQ,SAAC,EAAO,GACrB,EACI,EAAM,SAAW,EACjB,WACI,MAAA,kBAAkB,EAAI,sBAAsB,EAAC,sBAC7C,6BAA4B,EAAI,SAG1C,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,kBAAkB,EAAI,iCAAiC,GAAO,GAAC,MAEzE,GAAM,GAAa,EAAO,GAC1B,EAAO,QAAQ,SAAC,EAAO,GACrB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EACK,IAAM,GAAU,EAAM,KAAO,EAAW,GACzC,WAAM,MAAA,kBAAkB,EAAI,uBAAuB,EAAC,MAAM,EAAK,KAC3D,0CAAyC,EAAU,MACnD,oCAAmC,EAAC,qBAKlB,EAAoB,GAElD,OADM,GAAc,EAAO,GAAG,QACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAY,IAAS,EAAO,GAAG,GAEjC,MAAO,GClDT,AAqEA,YAAmC,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/D,EAAO,EAAQ,QAAU,EAAG,WAAM,MAAA,uCAElC,GAAI,GAAW,GAAqB,EAAS,UAAW,UAExD,AAAI,EAAS,GAAG,QAAU,aACxB,EAAS,QAAQ,SAAA,GACf,GAAI,EAAO,QAAU,YACnB,KAAM,IAAI,OAAM;uBACD,EAAO,MAAK,QAKjC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAS,GAAG,OAAO,GAChD,EAAW,GAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAC7D,GAAI,GAAc,KAAc,EAC9B,MAAO,IAAO,GAAI,GAIpB,GADA,EAAW,EAAS,OAAO,SAAA,GAAK,MAAA,GAAE,KAAO,IACrC,EAAS,SAAW,EACtB,MAAO,GAAS,GAGlB,GAAM,GAAS,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QACnC,GAAuB,EAAQ,GAE/B,GAAM,GAAM,EAAQ,OAAO,EAAU,GACrC,SAAK,GACE,GAGH,EAAuB,EACvB,EAAoB,CAAC,KAAI,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC/GjC,AAsCA,YAAoC,GAClC,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAAwB,CAAC,EAAG,GAElC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,QAAQ,GAC5B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAU,EAAG,CAAC,SAAQ,KCjDnC,AA8DA,YACI,EAAiB,EAAwB,GAC3C,GAAM,GAAK,EAAgB,EAAG,IAAK,SAEnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,kCAGlB,GAAM,GAA+B,SAAC,EAAS,GACvC,GAAA,GAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,UAA6B,EAAI,EAAQ,GACzC,EAAK,CAAC,IACC,EAAQ,MAAM,EAAI,EAAQ,IAG7B,EAAsB,CAAC,EAAG,GAC1B,EAAoB,CAAC,MAAK,EAAE,KAAI,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCrF/B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAI,EAAQ,KAAK,GACvB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AA8CA,YACI,EAA+B,EAC/B,EAA+B,EAC/B,EAAwB,GAC1B,GAAM,GACF,EAAgB,EAAY,aAAc,iBACxC,EACF,EAAgB,EAAY,aAAc,iBACxC,EAAY,EAAgB,EAAU,WAAY,iBAClD,EAAQ,EAAgB,EAAM,OAAQ,iBACtC,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAE7B,EAAW,GAAO,CAAC,EAAO,GAAK,GAC/B,EAAW,GAAO,EAAU,GAC5B,EAAgB,GAAI,EAAU,GAG9B,EAAY,EAAI,MAAM,GACtB,EAAY,EAAI,MAAM,GAAK,EAC3B,EAA8B,CAAC,EAAW,GAC1C,EAAI,GAAM,EAAK,CAAC,EAAG,GAAI,GACvB,EAAI,GAAM,EAAK,CAAC,EAAG,GAAY,GAC/B,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GACnC,EAAI,GAAM,EAAK,CAAC,EAAG,EAAY,GAAI,GAEnC,EACF,GAAI,EAAI,GAAQ,GAAI,GAAK,IACrB,EAAI,EAAI,GAAQ,GAAI,EAAa,MACnC,EAAiB,EAAI,GAAK,GAAO,GAAQ,IAC/C,MAAO,CAAC,EAAM,MAGH,IAAgB,EAAG,CAAC,eAAc,KC/E/C,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAE7C,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,iBAAiB,EAAG,KAAI,2CAC1B,EAAW,SAEnB,EACI,EAAM,SAAW,EAAW,OAC5B,WAAM,MAAA,mBACF,EAAM,OAAM,8CACZ,EAAW,SAEnB,EACI,EAAG,MAAM,GAAK,IAAS,EACvB,WAAM,MAAA,yBACI,EAAG,MAAM,GAAE,2CACjB,+BAA8B,EAAW,KAAK,OAAM,QAAQ,KAEpE,GAAM,GAA0B,SAAA,GAC9B,MAAO,GAAQ,eAAe,EAAI,EAAY,IAG1C,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,MAAK,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,iBC1FX,GACpC,GAAI,GACJ,MAAI,GAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAG,EAAE,OACxB,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACvC,AAAI,EAAE,OAAS,EACpB,EAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAErD,EAAM,EAGD,EChCT,AAwDA,YACI,EAAyB,EACzB,EACA,EACA,EACA,GACF,AAAI,GAAmB,MACrB,GAAkB,MAEpB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,AAAI,GAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAG9C,EACI,EAAM,OAAS,EAAU,KACzB,WAAM,MAAA,iFAEV,EACI,GAAW,MAAQ,EAAM,OAAS,EAAQ,KAC1C,WAAM,MAAA,+EAEV,EACI,GAAU,MAAQ,EAAM,OAAS,EAAO,KACxC,WAAM,MAAA,8EAGV,GAAM,GAAgB,GAAM,GAEtB,EAA+B,SAAC,EAAS,GAC7C,SAAK,CAAC,EAAK,EAAO,EAAW,IAEtB,EAAQ,UACX,EAAK,GAAS,GAAQ,GAAS,GAAY,GAAS,GACpD,GAAS,GAAS,IAGlB,EAA+B,CACnC,EAAG,EACH,MAAO,EACP,OAAQ,EACR,KAAM,EACN,SAAU,GAGN,EAA6B,CAAC,gBAAe,GAE7C,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAO,GAAQ,EAAK,EAAG,OAGzB,YAAkB,GAChB,MAAI,IAAK,KACA,KAEL,EAAE,OAAS,EAEN,EAAQ,EAAG,CAAC,EAAE,OACZ,EAAE,OAAS,EACb,EACE,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,KACpC,EAAE,OAAS,EAEb,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,KAEjD,KAGI,IAAY,EAAG,CAAC,WAAU,KCrGvC,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAG/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC5C3C,YACI,EAAwB,EACxB,EACA,EAAuC,EACvC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAQ,EAAgB,EAAM,OAAQ,aACtC,EAAY,EAAgB,EAAU,WAAY,aACpD,EACJ,AAAI,GAAS,MACX,GAAS,EAAgB,EAAO,QAAS,cAE3C,GAAI,GACJ,MAAI,IAAU,MACZ,GAAU,EAAgB,EAAQ,SAAU,cAE9C,EACI,EAAG,OAAS,EACZ,WAAM,MAAA,uDACC,GAAG,KAAI,OAClB,EACI,EAAM,OAAS,GAAK,EAAM,OAAS,EACnC,WAAM,MAAA,2DACF,aAAY,EAAM,KAAI,OAC9B,EACI,EAAU,OAAS,GAAK,EAAU,OAAS,EAC3C,WAAM,MAAA,2DACF,iBAAgB,EAAU,KAAI,OACtC,AAAI,GAAU,MACZ,EACI,EAAO,OAAS,GAAK,EAAO,OAAS,EACrC,WAAM,MAAA,wDACF,iBAAgB,EAAO,KAAI,OAErC,AAAI,GAAW,MACb,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,yDACF,iBAAgB,EAAQ,KAAI,OAE/B,GAAU,EAAI,EAAO,EAAW,EAAS,EAAQ,MAG7C,IAAc,EAAG,CAAC,aAAY,KC9E3C,AA4CA,YACI,EAAsB,GACxB,GAAI,GAAQ,EAAgB,EAAG,cAAe,KACxC,EAAS,EAAM,MAErB,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,CAAE,GAAI,IAAM,EAAI,IAAM,IACxC,KAAM,IAAI,OAAM,2CAA2C,EAAK,MAGlE,GAAI,EAAM,OAAS,EAAM,KACvB,KAAM,IAAI,OAAM,+BAA+B,EAAM,OAAM,iBACvD,EAAM,KAAI,KAGhB,GAAI,EAAM,OAAS,EAAM,MAEvB,OADM,GAAW,EAAM,MAAM,QACtB,EAAS,OAAS,EAAM,QAC7B,EAAS,QAAQ,GAEnB,EAAQ,EAAQ,EAAO,GAKzB,OAFM,GAAa,EAAM,MACnB,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAM,OAAS,EAAG,GAAK,EAAG,IACrC,GAAI,EAAW,KAAO,EAAM,GAC1B,EAAK,GAAK,UACD,EAAM,MAAM,KAAO,EAC5B,KAAM,IAAI,OACN,mBAAmB,EAAM,6BAA6B,EAAK,MAGnE,GAAM,GAAO,EAAK,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,KAAI,OAAO,SAAA,GAAK,MAAA,IAAK,IAEjE,GAAI,EAAK,SAAW,EAClB,MAAO,IAAM,GAGf,GAAM,GAAU,SAAC,GAA2B,MAAA,GAAQ,KAAK,EAAO,IAE1D,EAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,MAAK,EAAE,WAAU,GAElD,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAa,MAGb,IAAc,EAAG,CAAC,aAAY,KC5F3C,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC9C7B,AAyCA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,eACnC,EACK,GAAgB,EACjB,WAAM,MAAA,uBAAuB,EAAY,aACrC,+BAA8B,EAAY,QAElD,GAAM,GAA4B,CAAC,EAAG,GAChC,EAA0B,CAAC,aAAY,EAAE,aAAY,GAE3D,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAM,EAAQ,KAAK,EAAI,EAAc,GAC3C,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAc,EAAG,CAAC,aAAY,KC7B3C,YAAmB,GACjB,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCYrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCDrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KC5BrC,YACI,EAAqC,GACvC,MAAO,IAAO,EAAS,MAGZ,IAAW,EAAG,CAAC,UAAS,KCnCrC,AA8DA,YACI,EAAiB,EACjB,EACA,EACA,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GACjE,EACI,IAAY,EAAQ,MAAM,GAC1B,WAAM,MAAA,oCAAoC,EAAO,gBAC7C,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,GAAO,GACL,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GACxC,EACY,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAS,EAAG,CAAC,QAAO,KC3EjC,YACI,EAAiB,EAA6B,EAC9C,EACA,EAAiC,EACjC,GADA,AAAA,IAAA,QAAA,GAAA,OAAiC,AAAA,IAAA,QAAA,GAAA,GAEnC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,MAG9C,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,uDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAQ,GACjD,WAAM,MAAA,yDACF,eAAc,EAAM,kBAAkB,EAAQ,OACtD,EACI,IAAe,MACf,WAAM,MAAA,sCACF,EAAU,0CAElB,GAAM,GAAW,EACb,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,KAC7D,EAAU,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAA4B,CAAC,EAAG,GAChC,EAA8B,CAAC,EAAG,GAElC,EAAmB,OAEnB,EAAM,GACP,EAAuB,EAAuB,EAAS,EACxD,EAAkB,EAAW,GAEjC,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAGxC,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,QAGhD,IAAS,EAAG,CAAC,QAAO,KCpHjC,AAoDA,YACI,EAAmE,EACnE,EAAkB,EAClB,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KAC1D,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,KAG9C,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,GAAM,GAAU,IAAe,OAAS,EAAS,GAAK,EAAS,GACzD,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAY,EAEZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEE,EAAM,EAAQ,eAAe,EAAM,EAAQ,GAEjD,SAAK,CAAC,EAAM,IAEL,GAGH,EAAoC,CAAC,GAAI,EAAM,OAAM,GAErD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,EAAE,WAAY,GAEtD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAqB,GAEzB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxF3D,YACI,EAAiB,EACjB,EACA,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IACH,EAAa,EAAI,EAAS,EAAS,EAAK,OAAQ,MAGzC,IAAkB,EAAG,CAAC,iBAAgB,KCxDnD,AA8DA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,GADA,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAExE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,wDACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAA+B,EAAS,GACxC,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,QACf,WAAM,MAAA,sCACF,EAAU,4CAElB,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,GAC5C,EAAM,EAAQ,OAAO,EAAK,EAAS,GAEzC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAuB,CAAC,EAAG,EAAK,OAAQ,GAExC,EAAqB,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,GAEzD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,KAGI,IAAS,EAAG,CAAC,QAAO,KC3HjC,AA6CA,YACI,EAGA,EAAO,EAAkB,EACzB,GACF,EACI,EAAO,SAAW,EAAG,KACrB,WAAM,MAAA,qBACF,KAAI,EAAO,OAAM,qBAAqB,EAAG,KAAI,kBAErD,GAAI,GAAW,EACX,EAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACvE,EAAW,CAAC,EAAG,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,EAAO,KAGzD,GAAM,GAAU,EAAS,GACnB,EAAW,EAAK,MAAM,GAC5B,EACI,EAAS,SAAW,EACpB,WACI,MAAA,qEACG,GAAS,OAAM,OAC1B,EACI,EAAK,OAAS,EACd,WAAM,MAAA,uDACF,SAAQ,EAAK,QACrB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,2DACF,SAAQ,EAAO,QACvB,EACI,IAAY,EAAO,MAAM,GACzB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAO,MAAM,GAAE,OACvD,EACI,IAAa,EAAO,MAAM,GAC1B,WAAM,MAAA,6CAA6C,EAAQ,UACvD,kCAAiC,EAAO,MAAM,GAAE,OAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAU,EAAO,MAAO,EAAS,EAAW,GAEhD,MAAO,GAAQ,eAAe,EAAM,EAAQ,IAGxC,EAAoC,CAAC,GAAI,GAEzC,EAAkC,CAAC,IAAG,GAEtC,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGhE,EAGF,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxE3D,YACI,EAAiB,EACjB,EAGA,EAA0C,GAC5C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAElD,MAAO,IAAoB,EAAa,EAAI,EAAS,EAAS,MAGnD,IAAkB,EAAG,CAAC,iBAAgB,KCtDnD,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AAqDA,YACI,EAAsB,EAAU,EAAmB,GAA7B,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,IAAmB,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,UAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAG,MAC9C,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,IAE5B,GAAM,GAAe,GAAiB,EAAG,EAAG,MAAM,GAC9C,EAAQ,EAAQ,OAAO,EAAW,EAAc,EAAW,GAG/D,GAFA,EAAK,CAAC,IAEF,GAAe,MACjB,GAAM,GAAqB,GAAuB,GAClD,EAAQ,GAAU,EAAO,GAE3B,MAAO,IAGP,EAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,EAAE,UAAS,EAAE,QAAO,GAEpD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KCnFjC,AAiEA,YACI,EAA0B,EAC1B,GAAA,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,gBAE7B,EAAe,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC/D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAC9D,EAAc,IAAe,OAAU,EAAG,MAAM,GAAK,EAAG,MAAM,GAEpE,EACI,EAAc,GAAa,EAC3B,WAAM,MAAA;MACN,EAAW,QAAQ,EAAS;MAC5B,EAAG,QAEP,EACI,EAAa,GAAa,EAC1B,WAAM,MAAA;MACN,EAAU,QAAQ,EAAS;UACvB,EAAG,QAEX,EACK,EAAc,GAAY,KAAe,EAC1C,WAAM,MAAA,8CACF,EAAY,EAAS,WACrB,EAAU,sCAAsC,EAAG,QAE3D,GAAM,GAAiC,SAAA,GACnC,MAAA,GAAQ,aAAa,EAAI,EAAW,IAElC,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAAC,UAAS,EAAE,WAAU,GAEvD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAc,MAGP,IAAe,EAAG,CAAC,cAAa,KCvG7C,AA2EA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,EACA,GAFA,AAAA,IAAA,QAAA,GAAA,QACA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IAE3C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,sDACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAElC,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAGlB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,iEACF,mBAAkB,EAAO,mBAAmB,EAAS,OAE7D,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IACE,EAAM,EAAQ,gBAAgB,EAAK,EAAS,GAClD,SAAK,CAAC,EAAK,IACJ,GAGH,EAAsC,CAAC,EAAG,EAAK,OAAQ,GACvD,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KC/InD,AA+CA,YAAe,GACb,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAA,GACnC,GAAM,GAAO,EAAQ,EAAI,CAAC,EAAG,OACvB,EAAS,EAAQ,KAAK,GACtB,EAAe,EAAE,MAAK,OAAK,EAAE,OAEnC,MAAO,GAAQ,EAAQ,IAGnB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,OAGnD,IAAO,EAAG,CAAC,MAAK,KChE7B,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,EACA,GADA,AAAA,IAAA,QAAA,GAAA,CAAsC,EAAG,IACzC,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAU,EAAgB,EAAQ,SAAU,cAElD,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA,gEACC,GAAG,KAAI,OAClB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,4DACC,GAAQ,KAAI,OACvB,EACI,IAAe,OACf,WAAM,MAAA,0DACF,0BAAyB,KAEjC,GAAI,GAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,KACzD,EAAe,IAGjB,GAAM,GAA2B,CAAC,EAAG,EAAK,OAAQ,GAC5C,EAAyB,CAAC,QAAO,EAAE,IAAG,EAAE,UAAS,GAEjD,EAAM,EAAO,UACH,GAAY,EACZ,GAEhB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,KAGI,IAAa,EAAG,CAAC,YAAW,KCvGzC,YA2BI,EAAmB,GAGrB,OAFM,GAAS,EAAQ,OACjB,EAAiB,GACd,EAAI,EAAG,EAAI,EAAQ,KAC1B,GAAM,GAAM,EAAS,EAAI,EACnB,EAAI,EAAQ,IAAQ,EACpB,EAAI,EAAS,EAAS,OAAS,EAAI,IAAM,EAC/C,AAAI,EAAI,GAAK,IAAM,GACjB,EAAK,QAAQ,GAGjB,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAQ,EAAQ,EAAQ,OAAS,EAAI,GACrC,EAAU,EAAS,OAAS,EAAI,EAChC,EAAS,EAAS,GACxB,AAAI,IAAS,MAAS,IAAU,GAAK,EAAS,IAC5C,EAAO,QAAQ,GAGnB,MAAO,eAIL,EAAkB,GAIpB,OAHM,GAAmB,GACnB,EAAI,KAAK,IAAI,EAAO,OAAQ,EAAO,QAEhC,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GACnC,AAAI,GAAK,MACP,GAAI,GAEN,GAAI,GAAI,EAAO,EAAO,OAAS,EAAI,GAInC,GAHA,AAAI,GAAK,MACP,GAAI,GAEF,IAAM,EACR,EAAO,QAAQ,WACN,IAAM,EACf,EAAO,QAAQ,WACN,IAAM,GACf,GAAM,GAAS,wDACR,GAAM,QAAQ,EAAM,KAC3B,KAAM,OAAM,OAEZ,GAAO,QAAQ,GAGnB,MAAO,GCrFT,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,MAAM,EAAI,IAE5D,EAAsB,CAAC,EAAG,EAAI,EAAG,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,OAG/C,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AAoDA,YACI,EAA8B,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAK,EAAgB,EAAG,IAAK,SAC7B,EAAa,EAAgB,EAAW,YAAa,QAAS,QAI9D,EAAiB,GAA2B,EAAG,MAAO,EAAG,OACzD,EAAgB,GAAY,EAAI,GAChC,EAAgB,GAAY,EAAI,GACtC,AAAI,EAAW,OAAS,GAGtB,EACI,EAAW,MAAM,KAAO,EAAG,MAAM,GACjC,WAAM,MAAA,mEAGZ,AAAI,EAAW,OAAS,GAEtB,GACI,EAAW,MAAO,EAAc,MAAO,oBAG7C,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,OAAO,EAAY,EAAe,GACtD,SAAK,CAAC,IACC,GAEH,EAAyB,CAC7B,UAAW,EACX,EAAG,EACH,EAAG,GAEL,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,OAGA,IAAQ,EAAG,CAAC,OAAM,KC5F/B,AAuCA,YAAsC,GACpC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAA0B,CAAC,EAAG,GACpC,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,IAAK,EAClC,KAAiB,OAEjB,IAAY,EAAG,CAAC,WAAU,KC9CvC,AA0DA,YACI,EAAsB,SAEpB,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAAY,GAAI,EAAI,GACpB,EAAQ,GAAU,GAClB,EAAc,GAAM,EAAI,GAC9B,MAAO,IAAM,EAAa,EAAO,MAGtB,IAAW,EAAG,CAAC,UAAS,KCvErC,AA2CA,YAAc,EAAuB,GACnC,GAAM,GAAM,EAAgB,EAAI,KAAM,OAChC,EAAM,EAAgB,EAAI,KAAM,OAEtC,EACK,GAAI,OAAS,GAAK,EAAI,OAAS,IAAO,GAAI,OAAS,GAAK,EAAI,OAAS,GACtE,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GACjD,EAAW,EAAI,OAAS,EAAI,EAAI,KAAO,EAAI,MAAM,GAOvD,GALA,EACI,IAAY,EACZ,WAAM,MAAA,gEACC,GAAO,QAAQ,EAAO,OAE7B,EAAI,OAAS,GAAK,EAAI,OAAS,GACjC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,YACZ,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAG,KACxB,EAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAM,GAC1B,MAAO,GAAQ,EAAM,CAAC,EAAK,eAClB,EAAI,OAAS,GAAK,EAAI,OAAS,GACxC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,GAAO,EAAK,GACzB,MAAO,GAAQ,EAAM,CAAC,EAAK,YAE3B,GAAM,GAAO,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,KAC7C,EAAO,GAAO,EAAK,GACzB,MAAO,OAIE,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,GACtB,SAAK,CAAC,IACC,GAGH,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCtD3B,AAyCA,YAAgC,GAC9B,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,EACI,EAAG,QAAU,SAAW,EAAG,QAAU,UACrC,WAAM,MAAA,8CAEV,AAAI,EAAG,QAAU,SACf,GAAK,GAAK,EAAI,YAGhB,GAAM,GAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KC1D3B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAyCA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GAC3D,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,aAAc,GAEjD,EAAY,GAAQ,EAAG,KAAM,WAAM,MAAA,uCACnC,GAAM,GAAW,EAAG,MAAM,QAC1B,MAAI,GAAO,GAET,GACI,CAAE,GAAG,KAAO,IAAM,EAClB,WAAM,MAAA,iCAAiC,CAAG,GAAG,KAAO,GAAE,KAAK,EAAG,KAAI,MACtE,EAAO,EAAG,KAAO,EAAO,GAE1B,EAAS,OAAO,EAAM,EAAG,GAClB,EAAQ,EAAI,MAGR,IAAa,EAAG,CAAC,YAAW,KC1DzC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AAqDA,YAAiC,EAAiB,GAChD,GAAM,GAAoB,KACpB,EAAK,EAAgB,EAAG,IAAK,OAAQ,GAC3C,EACI,EAAG,OAAS,EAAK,OACjB,WAAM,MAAA,qCAAqC,EAAG,KAAI,IAC9C,8BAA6B,EAAI,OAEzC,GAAM,GAA0B,SAAC,EAAS,GACxC,GAAM,GAAM,EAAQ,KAAK,EAAI,GAC7B,SAAK,CAAC,IACC,GAGH,EAAe,CAAC,GAChB,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,GAE9B,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,EAAkC,MAG3B,IAAO,EAAG,CAAC,MAAK,KC5E7B,AAwCA,YACI,EAAiB,EACjB,EAKA,GAAA,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAc,MAChB,GAAa,GAIf,OAFM,GAAO,GAAO,CAAC,EAAS,GAAa,GACrC,EAAI,GAAW,EAAa,EAAU,EACnC,EAAI,EAAG,EAAI,EAAG,EAAE,EACvB,EAAK,IAAI,EAAG,EAAG,GAEjB,GAAM,GAAgB,EAAQ,EAAK,WAAY,CAAC,EAAS,IACzD,GAAI,GAAc,KAChB,MAAO,GAEP,GAAI,EAAW,SAAW,EACxB,MAAO,IAAK,GAAW,EAAK,GAAI,CAAC,EAAW,GAAI,EAAG,IAC9C,GAAI,EAAW,SAAW,EAE/B,MAAO,IACI,GAAW,GAAW,EAAK,GAAI,GAC/B,CAAC,EAAW,GAAI,EAAW,GAAI,EAAG,IACxC,GAAI,EAAW,SAAW,EAE/B,MAAO,IAAK,GAAW,GAAW,GAAW,EAAK,GAAI,GAAI,GAAI,CACrD,EAAW,GAAI,EAAW,GAAI,EAAW,GAAI,EAAG,IAGzD,KAAM,IAAI,OACN,2CAEA,8BAA8B,EAAmB,OAAM,UAKpD,IAAM,EAAG,CAAC,KAAI,KCjF3B,AAqCA,YACI,EAAoB,EAAsB,GAC5C,GAAM,GAAmB,CAAC,MAAK,EAAE,MAAK,EAAE,MAAK,GAE7C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAO,EAAO,IAAQ,GAAI,KAAM,GACxD,GC3CN,AAqCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC7C/B,AAuBO,GAAM,IAAwB,eASI,GACvC,MAAI,IAAU,GACL,EAEF,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KCpCrD,YA8BI,EAAgB,GAClB,GAAI,GAAO,GACP,EASJ,IAPA,AAAI,GAAU,GACZ,GAAM,EACN,EAAO,IAEP,EAAM,GAAe,EAAQ,KAAK,MAAM,KAAK,KAAK,KAG7C,CAAC,GACN,AAAI,EAAM,GAAe,IAAQ,EAC/B,EAAO,GAEP,EAAM,GAAe,EAAQ,EAAM,GAGvC,MAAO,eAIL,EAAkB,EAAc,GAGlC,OAFM,GAAW,GACX,EAAO,EAAO,OACX,EAAM,EAAG,EAAM,EAAM,IAC5B,AAAI,IAAQ,EACV,EAAS,KAAK,EAAO,IAErB,EAAS,KAAK,GAGlB,MAAO,eAUL,EAAW,EAAiB,GAM9B,OALM,GAAU,EAAE,MAAM,GAElB,EAAwB,GAC1B,EAAY,EACZ,EAAY,EACP,EAAI,EAAG,EAAI,EAAM,IACxB,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAQ,KAAM,IAChC,EAAY,KAAK,EAAQ,MAAM,IAGjC,OAAS,GAAI,EAAO,EAAG,EAAI,EAAE,KAAM,IACjC,EAAY,KAAK,EAAE,MAAM,IACzB,GAAa,EAAE,MAAM,GAGvB,MAAO,CAAC,UAAS,EAAE,UAAS,EAAE,QAAO,EAAE,YAAW,2GC5FpD,AAoDA,YACI,EAAiB,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC/C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAW,EAAgB,EAAS,UAAW,SAAU,SAEzD,EAAyB,CAAC,EAAG,EAAI,QAAS,GAC1C,EAAuB,CAAC,KAAI,GAE5B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAa,GAAe,EAAM,EAAG,OAAO,GAC5C,EAAY,GAAyB,EAAI,EAAU,GAEnD,EACF,EAAQ,OAAO,EAAI,EAAQ,EAAU,CAAC,EAAS,OAAQ,GAC3D,SAAK,CAAC,EAAI,IAEH,EAAQ,EAAK,EAAU,cAGhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAS,EAAG,CAAC,QAAO,KC5EjC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,QAAQ,EAAI,IAE9D,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KC3DnC,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,aAAa,EAAI,GACrC,SAAK,CAAC,EAAI,IACH,GAGH,EAA6B,CAAC,EAAG,EAAI,EAAG,GAE9C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAe,EAAG,CAAC,cAAa,KC/D7C,AAsCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCnD7B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GAEnC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,SAAS,IAAK,EACnC,KAAiB,OAEV,IAAW,EAAG,CAAC,UAAS,KC/CrC,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC/C/B,AAsCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,MAAM,IAAK,EAC9B,KAAiB,OAEV,IAAQ,EAAG,CAAC,OAAM,KC9C/B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAEhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAGH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,YAsCI,EAAyC,GAC3C,GAAM,IAAa,IAAU,IAAU,UAAa,MAAM,QAAQ,KAC9D,IAAU,YACZ,KAAM,IAAI,OACN,kFAGN,GAAI,IAAU,UAAY,GAAa,IACnC,CAAE,aAAiB,aACrB,KAAM,IAAI,OACN,6EAGN,GAAM,GAAkB,GAClB,EAA0B,GAChC,MAAO,IAAW,EAAO,EAAO,EAAe,GCrDjD,AA2CA,YAAsC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,IACrD,GAAM,GAAK,EAAgB,EAAG,IAAK,aACnC,MAAO,IAAQ,EAAI,GAAO,GAAQ,GAAK,MAG5B,IAAY,EAAG,CAAC,WAAU,KChDvC,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,QAC7B,EAAK,EAAgB,EAAG,IAAK,QACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAE3D,EAAqB,CAAC,EAAG,EAAI,EAAG,GAEtC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC1D7B,AA0CA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,UAAU,EAAI,GAClC,SAAK,CAAC,EAAI,IACH,GAGH,EAA0B,CAAC,EAAG,EAAI,EAAG,GAE3C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAY,EAAG,CAAC,WAAU,KC/DvC,YAkCyB,EAAe,EAAc,GACpD,GAAI,GAAO,EACT,KAAM,IAAI,OAAM,4CAGlB,GAAM,GAAuB,CAAC,MAAK,EAAE,KAAI,EAAE,IAAG,GAC9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,SAAS,EAAO,EAAM,IAAM,GAC/C,KAAiB,GAAU,GC1CjC,AA4CA,YACI,EAAiB,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IACzD,GAAM,GAAK,EAAgB,EAAG,IAAK,8BACnC,EACI,EAAG,OAAS,GAAK,EAAG,OAAS,EAC7B,WAAM,MAAA;sBACU,EAAG,KAAI,MAC3B,EACI,GAAW,GACX,WAAM,MAAA,+DACF,gCAA+B,EAAW,OAElD,GAAI,GAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,6BACd,EAAK,EAAa,EAAM,EAAO,GAEnC,SAAK,CAAC,EAAK,IAEJ,GAGH,EAAoB,CAAC,EAAG,GAExB,EAAkB,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAEjD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEpD,KAIE,IAA6B,EAAG,CAAC,4BAA2B,KCvFzE,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAC9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KChD3B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KCjD/B,AA0DA,YAAc,GAEZ,SACI,GAAgB,GAAI,WAAM,MAAA,+CACvB,SAAC,EAAsB,GAE5B,GAAM,GAAK,EAAgB,EAAG,IAAK,UAAW,MACxC,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,WAAa,KAC1D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,kFAGN,GAAW,GACJ,EAAM,MAkCnB,YAAe,GAEb,SACI,GAAgB,GAAI,WAAM,MAAA,gDACvB,SAAC,EAAgC,GACtC,EACI,MAAM,QAAQ,GACd,WAAM,MAAA,qFAGV,GAAM,GAAQ,GAAqB,EAAM,OAAQ,WAAY,MACvD,EACD,GAAM,KAAQ,EAAgB,EAAI,KAAM,YAAc,KAC3D,MAAO,GAAO,KAAK,WACX,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,MAAA,OAAA,IAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,MAAI,IAAO,MACT,GACI,EAAM,MAAO,EAAI,MACjB,iGAGN,GAAW,GACJ,KA8Bb,YAA0D,GAKxD,SACI,GAAgB,GAChB,WAAM,MAAA,uDACH,SAAC,EAAM,GACZ,EACI,YAAa,GACb,WAAM,MAAA,wDACV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,6DACJ,GAAA,GAAA,EAAA,UAAA,WAAA,MAAA,GAAA,IAAA,CAAA,GAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MACd,UAAW,GACJ,CAAC,KAAM,EAAM,GAAS,MAAK,IAmCtC,YAAyC,GAKvC,SACI,GAAgB,GAChB,WAAM,MAAA,wDACH,SAAC,EAAgB,GACtB,EACI,MAAM,QAAQ,IAAS,EAAK,MAAM,SAAA,GAAO,MAAA,aAAe,KACxD,WAAM,MAAA,uEAEV,EACI,GAAM,MAAQ,YAAc,GAC5B,WAAM,MAAA,iEACV,GAAM,GAAM,EAAO,UAAU,WAAM,MAAA,GAAC,MAAA,OAAI,IAAO,EAAM,GACrD,MAAI,IAAM,MACR,GACI,EAAI,MAAM,MAAO,EAAG,MACpB,yGAGN,GAAW,EAAI,OACR,GAkCX,YAAuB,EAAiB,GAEtC,EACI,GAAgB,GAChB,WAAM,MAAA,wDACV,EACI,GAAW,MACP,MAAM,QAAQ,IAAY,EAAQ,MAAM,SAAA,GAAK,MAAA,aAAa,MAC9D,WACI,MAAA,kFAGR,GAAM,GAAmB,GAAW,KACpC,GAAI,CAAC,GAEH,EAAU,GACV,OAAW,KAAW,GAAO,oBAC3B,EAAQ,KAAK,EAAO,oBAAoB,IAI5C,GAAM,GACF,EAAmB,EAAQ,OAAO,SAAA,GAAY,MAAA,CAAC,EAAS,YAAa,KAGnE,EAAmB,EAAQ,OACjC,EAAU,EAAQ,OAAO,SAAA,GAAY,MAAA,GAAS,YAC9C,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,kEACF,kCAAiC,EAAgB,kBACjD,eAER,GAAM,GAAmB,GACnB,EAAA,EAAA,UAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,EACI,EAAM,KAAK,SAAA,GAAK,MAAA,IAAK,OACrB,WAAM,MAAA,iMAGV,EACI,EAAM,OAAS,EACf,WAAM,MAAA,iEACF,oBAAmB,EAAM,KAAI,aAErC,GAAM,GAA6B,GACnC,SAAQ,QAAQ,SAAC,EAAG,GAClB,AAAI,EAAM,IAAM,MACd,GAAW,EAAE,MAAQ,EAAM,MAG/B,AAAI,GAAyB,MAG3B,EAAsB,QAAQ,SAAA,GAAK,MAAA,GAAW,EAAE,MAAQ,OAEnD,CAAC,MAAK,EAAE,MAAO,GA2CxB,YAAsC,GAEpC,MAAO,GAAO,WAAW,GAG3B,YAAoB,GAClB,GAAM,GAAmB,EAAM,OAAO,SAAA,GAAK,MAAA,IAAK,OAAM,OACtD,GAAI,EAAmB,EACrB,KAAM,IAAI,OACN;sEC7XR,AAuCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAuB,CAAC,EAAG,GACjC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,IAAI,IAAK,EAC5B,KAAiB,OAEV,IAAM,EAAG,CAAC,KAAI,KC/C3B,AAsCA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,SAAS,GAC7B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAW,EAAG,CAAC,UAAS,KChDrC,AAyCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAK7B,EAAW,GAAW,SAAC,GAI3B,GAAM,GAAQ,GAAI,GAAS,GAAI,KAEzB,EAAW,SAAC,GAChB,GAAM,GAAO,EAAI,EAAI,GAAQ,GAAI,KACjC,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAEL,IAAa,EAAG,CAAC,YAAW,KC9DzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAoB,EAAM,EAAG,OAC1C,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAS,OAG1D,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GAAgB,GAClB,EAAI,MAAO,GAAoB,EAAM,EAAG,QAC5C,EAAM,EAAQ,EAAK,GACnB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,iBAAkB,EAAM,SAAQ,GAEzD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAiDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,SAAS,EAAI,GACjC,SAAK,CAAC,EAAI,IACH,GAGH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAM,EAAG,CAAC,KAAI,KCnE3B,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,OACjC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAGhB,GAAM,GAA+B,SAAC,EAAS,GAC7C,EAAK,CAAC,IACN,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,IAAI,EAAW,GACnC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAEzB,MAAO,IAGH,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAM,EAAG,CAAC,KAAI,KC/F3B,AAsDA,YAAuC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAAU,EAAgB,EAAQ,SAAU,cAKlD,GAHA,AAAI,IAAS,IACX,GAAO,EAAQ,KAAO,GAEpB,IAAS,EAAQ,KAAO,EAC1B,KAAM,OACF,gEACA,oBAAmB,EAAQ,KAAI,iBAAiB,IAGtD,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACX,EAAO,GAAI,EAAQ,EAAM,IACzB,EAAU,GAAI,EAAQ,GACtB,EACF,GAAI,GAAK,EAAS,WAAY,GAAI,GAAI,GAAI,GAAU,EAAM,KAC9D,SAAK,CAAC,IACC,GAGH,EAA2B,CAAC,OAAQ,GACpC,EAAyB,CAAC,KAAI,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAY,MAGZ,IAAa,EAAG,CAAC,YAAW,KCpFzC,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAO,GAAI,EAAI,EAAM,IACrB,EAAI,GAAI,EAAI,GACZ,EAAI,GAAI,GACR,EAAI,GAAI,EAAG,GACX,EAAI,GAAI,GACR,EAAM,GAAI,EAAQ,EAAM,EAAE,OAAQ,GAExC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAI,MAAO,GACjD,MAAO,GAAQ,EAAK,GAEtB,MAAO,MAGI,IAAY,EAAG,CAAC,WAAU,KChFvC,AAyCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA2B,CAAC,EAAG,EAAI,EAAG,GAE5C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,WAAW,EAAI,IAClC,EAAgC,KAAiB,OAGjD,IAAa,EAAG,CAAC,YAAW,KCtDzC,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,WAAW,IAAK,EACnC,KAAiB,OAGV,IAAa,EAAG,CAAC,YAAW,KC9CzC,AAwCA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,YAAa,QAC1C,EAAK,EAAgB,EAAG,IAAK,YAAa,QAChD,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA0B,CAAC,EAAG,EAAI,EAAG,GAC3C,MAAO,GAAO,cACH,SAAA,GAAW,MAAA,GAAQ,UAAU,EAAI,IACjC,EAAgC,KAAiB,OAEjD,IAAY,EAAG,CAAC,WAAU,KCnDvC,AA0CA,YACI,EAAsB,GACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,aAAc,QAC3C,EAAK,EAAgB,EAAG,IAAK,aAAc,QACjD,UAA2B,EAAG,MAAO,EAAG,OAGjC,GAAW,GAAU,EAAG,GAAI,GAAW,GAAW,EAAG,QAGjD,IAAa,EAAG,CAAC,YAAW,KCpDzC,AAwDA,YACI,EAAiB,EACjB,EAAkC,EAClC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAC7B,EAAY,EAEd,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,uDAAuD,EAAI,KAAI,MACzE,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,wDACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAmB,EACnD,GAEA,EAEJ,MAAI,GAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,GAAiB,EAAS,QAAS,EAAS,UAC9C,EAAI,EAAI,QAER,EAAI,EAAQ,QAAQ,EAAK,GAG3B,EAAK,CAAC,EAAK,IAEJ,GAGH,EAAwB,CAAC,EAAG,GAE5B,EAAsB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAEhE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAU,EAAG,CAAC,SAAQ,KCrHnC,AA2EA,YACI,EAAiB,EACjB,EAA0C,EAC1C,EACA,EACA,GAJiB,AAAA,IAAA,QAAA,GAAA,CAA+C,EAAG,EAAG,IAGtE,AAAA,IAAA,QAAA,GAAA,SAEF,AAAI,GAAa,KACf,EAAY,CAAC,EAAG,EAAG,GAEnB,GACI,+DAIN,GAAM,GAAK,EAAgB,EAAG,IAAK,aAE/B,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGxE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,qDAAqD,EAAI,KAAI,MACvE,EACI,IAAe,QACf,WAAM,MAAA,0DACF,0BAAyB,KACjC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,8DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,0DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAC,EAAS,GAC7C,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,EAAG,IAErB,GAAM,GAAW,GACb,EAAI,MAAO,EAAY,EAAS,EAAW,EAAK,EAChD,GACE,EAAI,EAAQ,UAAU,EAAK,GACjC,SAAK,CAAC,EAAK,IACJ,GAGH,EAA0B,CAAC,EAAG,GAE9B,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,EAAE,WAAU,EAAE,UAAS,GAE/D,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GAEJ,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,KAGI,IAAY,EAAG,CAAC,WAAU,KClJvC,AA4DA,YACI,EAAiB,EACjB,EAAkC,EAClC,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAK,EAAgB,EAAG,IAAK,qBAE7B,EAAkC,CAAC,EAAG,GACtC,EACuB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,oBAAmB,GAErE,EAAS,EAAO,UACH,GAAmB,EACnB,GAEnB,MAAO,CAAC,OAAQ,EAAO,GAAI,QAAS,EAAO,OAGhC,IAAoB,EAAG,CAAC,mBAAkB,KC7EvD,YAsCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAM,EAAO,WACpB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAoB,GAAc,GAAQ,GACzD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC7C1C,YAuCI,EAAoB,GACtB,GADsB,AAAA,IAAA,QAAA,GAAA,WAClB,IAAU,aACZ,GAAM,GAAO,GAAK,EAAO,WACnB,EAAO,GAAM,EAAO,WAC1B,MAAO,IAAQ,EAAM,GAEvB,GAAM,GAAS,GAAmB,GAAc,GAAQ,GACxD,MAAO,GAAO,WAAW,EAAQ,EAAO,GC9C1C,AA6DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAO,GAAe,EAAM,EAAG,OAC/B,EAAS,GAA0B,EAAG,MAAO,GAC7C,EAAc,EAAO,GACrB,EAAa,GAAc,GAI3B,EAAW,GAAW,SAAC,GAC3B,GAAM,GAAmB,GAAO,GAE1B,EAAU,EAAiB,QAAU,EAAE,MACzC,EACA,GAAK,EAAG,EAAiB,OACvB,EAAM,GAAI,EAAS,GACnB,EAAQ,GAAI,EAAK,EAAM,GAEvB,EAAW,SAAC,GAChB,GAAM,GAAkB,EAAE,MAAM,QAChC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,GAAI,EAAI,EAAY,GAAK,EAAE,MAAO,YAAa,GAC5D,MAAO,IAET,MAAO,CAAC,MAAK,EAAE,SAAQ,KAGzB,MAAO,GAAS,MAGL,IAAO,EAAG,CAAC,MAAK,KCpC7B,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EACF,SAAC,EAAwB,GACvB,GAAM,GAAW,GAAe,EAAM,EAAG,OACrC,EAAO,EACL,EAAe,GAA6B,EAAM,EAAG,MACvD,EAAW,EACf,AAAI,GAAgB,MAClB,GAAW,GAAU,EAAI,GACzB,EAAO,GAA2B,EAAK,OAAQ,EAAG,OAGpD,GAAM,GAAI,EAAQ,IAAI,EAAU,GAChC,AAAI,GAAgB,MAClB,EAAS,UAGX,GAAI,GAAM,EACV,GAAI,GACF,GAAM,GACF,GAA+B,EAAI,MAAO,GAC9C,EAAM,EAAQ,EAAG,GACjB,EAAE,UAGJ,SAAK,CAAC,EAAI,IACH,GAGP,EAAoB,CAAC,EAAG,GACxB,EAAkB,CAAC,KAAI,EAAE,SAAQ,GAEvC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAqB,GAC9D,MAGA,IAAM,EAAG,CAAC,KAAI,KCpG3B,AAwDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,WAC7B,EAAK,EAAgB,EAAG,IAAK,WACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,SACd,EAAK,GAAK,EAAI,UAGhB,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,SAAK,CAAC,EAAI,IACH,GAEH,EAAwB,CAAC,EAAG,EAAI,EAAG,GAEzC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAU,EAAG,CAAC,SAAQ,KCjFnC,AAuDA,YAAgC,EAAsB,SAChD,EAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,IAAI,EAAI,GAC5B,SAAK,CAAC,EAAI,IACH,GAEH,EAAoB,CAAC,EAAG,EAAI,EAAG,GAErC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AAmCA,YAAmC,GACjC,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAQ,GACR,EAAe,CAAC,GAChB,EAA2B,GACjC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,SAAK,CAAC,IACC,EAAQ,OAAO,IACrB,CAAC,EAAG,GAAK,KAAiB,SAAU,EAAO,EAAc,MAGjD,IAAS,EAAG,CAAC,QAAO,KC9CjC,AA4CA,YACI,EAAsB,EACtB,GADsB,AAAA,IAAA,QAAA,GAAA,MACtB,AAAA,IAAA,QAAA,GAAA,IACF,EAAI,EAAgB,EAAG,IAAK,WAC5B,GAAM,GAAO,GAAe,EAAM,EAAE,OAC9B,EAAQ,GAAK,EAAG,EAAM,GACxB,EAAgB,EAAM,MAC1B,AAAK,GACH,GAAgB,GAAqB,EAAM,MAAO,IAEpD,GAAM,GACF,GAAO,GAAI,GAAK,EAAG,WAAY,EAAQ,EAAO,KAC5C,EAAW,GAAK,EAAY,EAAM,GACxC,MAAO,CAAC,KAAM,EAAO,SAAQ,MAGlB,IAAU,EAAG,CAAC,SAAQ,KChBnC,YACI,EAA2B,EAC3B,EACA,GAOF,OANM,GAAQ,EAAgB,EAAM,OAAQ,gBACtC,EAAK,GAAqB,EAAG,IAAK,gBAClC,EAAK,GAAqB,EAAG,IAAK,gBAEpC,EAAQ,EACN,EAAY,GACT,EAAI,EAAG,EAAI,EAAU,OAAQ,KACpC,GAAM,GAAS,EAAU,GAAG,EAAO,EAAG,GAAI,EAAG,IAC7C,EAAU,KAAK,EAAO,IACtB,EAAU,KAAK,EAAO,IACtB,EAAQ,EAAO,GAIjB,OAFM,GAAmB,GACnB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,GAAK,EACzC,EAAK,KAAK,EAAU,IACpB,EAAK,KAAK,EAAU,EAAI,IAE1B,MAAO,CAAC,EAAM,MAEH,IAAe,EAAG,CAAC,cAAa,KCpE7C,AA4CA,YACI,EAAsC,EAAoB,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAgB,EAAQ,SAAU,eAC5C,EAAc,EAAQ,KACtB,EAAW,EAAQ,KACzB,GAAI,EAAc,EAChB,KAAM,IAAI,OACN,+DACG,GAAW,MAEpB,GAAI,EAAW,EACb,KAAM,IAAI,OAAM,gDAAgD,GAElE,EAAO,GAAQ,KAAK,SACpB,GAAM,GACF,IAAa,EAAI,EAAQ,EAAS,CAAC,EAAG,KAAO,EAC3C,EAAM,EAAO,cACf,SAAA,GAAW,MAAA,GAAQ,YAAY,EAAU,EAAY,EAAY,IACjE,CAAC,SAAQ,IAEb,MAAO,KAAa,EAAI,EAAQ,EAAK,CAAC,EAAI,OAAqB,KAGpD,IAAc,EAAG,CAAC,aAAY,KCpE3C,AAyCA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,YAC7B,EAAK,EAAgB,EAAG,IAAK,YACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,GAAY,MAAA,GAAQ,SAAS,EAAI,IAEjE,EAAyB,CAAC,EAAG,EAAI,EAAG,GAE1C,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KC1DrC,AAwCA,YAAiC,GAC/B,GAAM,GAAS,EAAgB,EAAO,QAAS,QAEzC,EAA+B,SAAC,GACpC,MAAO,GAAQ,KAAK,IAGhB,EAAqB,CAAC,MAAO,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAO,EAAG,CAAC,MAAK,KCrD7B,AA0CA,YAAqC,GACnC,GAAM,GAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAI,EAAG,QAAU,aACf,GAAM,GAAI,GAAS,GAAK,IAClB,EAAI,GAAU,GAAK,IACzB,MAAO,IAAQ,EAAG,GAGpB,MAAO,GAAQ,SAAS,IAGpB,EAAyB,CAAC,EAAG,GACnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAW,EAAG,CAAC,UAAS,KCtBrC,YACI,EAAyB,GAC3B,GAAM,GAAM,EAAgB,EAAI,KAAM,gBAChC,EAAM,EAAgB,EAAI,KAAM,gBAEtC,EACI,EAAI,OAAS,GAAK,EAAI,OAAS,EAC/B,WAAM,MAAA,+DACC,GAAI,KAAI,QAAQ,EAAI,KAAI,OAEnC,GAAM,GAAO,EAAQ,EAAK,CAAC,GAAI,IACzB,EAAO,EAAQ,EAAK,CAAC,EAAG,KAC9B,MAAO,IAAO,EAAM,MAGT,IAAe,EAAG,CAAC,cAAa,KCtD7C,AAmDA,YACI,EAAiB,EAAmC,GAAA,AAAA,IAAA,QAAA,GAAA,GACtD,GAAM,GAAK,EAAgB,EAAG,IAAK,OACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAA0B,SAAC,EAAS,GACxC,SAAK,CAAC,IACC,EAAQ,IAAI,EAAI,EAAU,IAG7B,EAAoB,CAAC,SAAQ,EAAE,cAAa,GAC5C,EAAsB,CAAC,EAAG,GAChC,MAAO,GAAO,cACV,EAAS,EAAqC,KAAiB,GAC/D,MAGO,IAAM,EAAG,CAAC,KAAI,KC5C3B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,qDACH,GAAI,EAAG,CAAC,GAAW,MAGf,IAAQ,EAAG,CAAC,OAAM,KCT/B,YACI,EAAwB,EACxB,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCV/B,YACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,EACvD,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCX/B,YACI,EACA,EAIA,GAAA,MAAA,KAAA,QAAA,GAAA,GACF,EACI,EAAS,SAAW,GAAK,EAAS,GAAG,SAAW,GAC5C,EAAS,GAAG,SAAW,GAAK,EAAS,GAAG,SAAW,GACnD,EAAS,GAAG,SAAW,EAC3B,WAAM,MAAA,0DACH,GAAI,EAAG,EAAU,MAGb,IAAQ,EAAG,CAAC,OAAM,KCxC/B,AA4EA,YACI,EAAiB,EAAsB,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAEnC,EACI,EAAG,MAAQ,EAAI,EAAW,OAC1B,WAAM,MAAA,cAAc,EAAG,KAAI,kCACvB,EAAW,SAEnB,EACI,EAAS,SAAW,EAAW,OAC/B,WAAM,MAAA,qBACF,EAAS,OAAM,kCAAkC,EAAW,SAEpE,EACI,EAAG,MAAM,OACL,SAAC,EAAG,EAAG,GACL,MAAI,GAAI,GAAK,GAAK,EAAW,OACpB,GACD,GAAI,EAAS,EAAI,GAAG,GAAK,EAAS,EAAI,GAAG,IACtC,EAAW,EAAI,KACnB,EAEA,GAET,IACJ,WAAM,MAAA,4BAA4B,EAAG,MAAM,MAAM,GAAE,kBAC/C,EAAS,WAAU,qCACnB,EAAW,aAEnB,GAAM,GAA0B,SAAA,GAC5B,MAAA,GAAQ,eAAe,EAAI,EAAY,IAErC,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,WAAU,EAAE,SAAQ,GAExD,MAAO,GAAO,cACV,EAAS,EAAgC,KACzC,GAAgB,MAGT,IAAiB,EAAG,CAAC,gBAAe,KCrHjD,AAwDA,YACI,EAAqB,EACrB,EAA0B,EAC1B,EAAqC,GACvC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,AAAI,GAAW,MACb,GAAU,GAEZ,AAAI,IAAQ,GACV,GAAM,SAGR,GAAM,GAAK,EAAgB,EAAO,IAAK,WACnC,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG3D,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,yDACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAC1C,EACF,CAAC,EAAS,eAAgB,EAAS,eAOnC,EACJ,AAAI,IAAQ,OACV,EAAc,GACV,CAAC,EAAS,aAAc,EAAS,aAAc,GAEnD,EAAc,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,IAG7B,GAAM,GAAgB,EAAS,KAAO,GAAK,EAAS,KAAO,EACrD,EAAA,GAAA,CAAA,EAAA,SAAA,EAAA,SAAA,EAAA,GAAC,EAAA,EAAA,GAAiB,EAAA,EAAA,GAElB,EAAe,EAAgB,EAAM,QACrC,EACF,EAAgB,EAAM,GAAe,EAAK,EAAU,GAElD,EAAY,IAAgB,MAC9B,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAChD,WAAM,MAAA,IAAQ,EAAY,EAAa,EAAS,IAC9C,EAAI,IAEJ,EAAM,EAAgB,EAAI,GAAe,EAAG,EAAU,GAE5D,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAGtD,EAMT,YACI,EAA8B,EAC9B,GACF,GAAM,GAAW,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,EAAa,EAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KACpC,EAAiB,EAAW,OAAO,EAAU,GAC7C,EAAc,EAAW,IAAI,SAAC,EAAG,GAAM,MAAC,GAAI,EAAe,GAAK,GAAK,IACrE,EAAS,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAY,KAClD,EAAW,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAS,GAAI,EAAO,MACzD,EAAQ,EAAW,IAAI,SAAC,EAAG,GAAM,MAAA,CAAC,EAAG,EAAY,MACvD,MAAO,CAAC,EAAU,GAMpB,YACI,EAA+B,GAGjC,GAAM,GAAqB,EAAY,IAAI,SAAC,EAAG,GAC7C,MAAO,GAAK,GAAI,GAAM,GAAS,GAAK,KAEhC,EAAgB,EAAmB,IAAI,SAAA,GAAK,MAAA,GAAI,IAIhD,EAAgB,EAAc,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAI,KACtD,EAAc,EAAc,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAc,KAClE,MAAO,GAAc,IAAI,SAAC,EAAG,GAC3B,MAAO,CAAC,EAAc,GAAI,EAAY,SAI7B,IAAO,EAAG,CAAC,MAAK,KChK7B,AAsDA,YACI,EAAyB,SACvB,EAAQ,EAAgB,EAAM,OAAQ,OACtC,EAAO,EAAgB,EAAK,MAAO,OACvC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,GAAM,GAAoB,CAAC,EAAG,EAAO,EAAG,GAClC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAI,EAAQ,IAAI,EAAO,GAC7B,SAAK,CAAC,EAAO,EAAM,IACZ,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,OAGA,IAAM,EAAG,CAAC,KAAI,KCxE3B,AA0CA,YAAkC,EAAiB,GACjD,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAS,EAAgB,EAAO,QAAS,SAEzC,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,MAAM,EAAI,GAC9B,SAAK,CAAC,EAAI,IACH,GAGH,EAAsB,CAAC,EAAG,EAAI,MAAO,GAC3C,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC1D/B,AA8DA,YACI,EAAsB,EAA8B,GAA9B,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IACtD,GAAI,GAAK,EAAgB,EAAG,IAAK,QAE3B,EAA+B,SAAC,GACpC,AAAI,EAAG,QAAU,QACf,GAAK,GAAK,EAAI,UAEhB,GAAM,GAAO,GAAe,EAAM,EAAG,OAE/B,EAAc,GAAmB,EAAM,EAAG,MAC5C,EAAgB,EAChB,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,GAAU,EAAI,GAC1B,EAAgB,GAAiB,EAAc,OAAQ,EAAG,OAE5D,GAAI,GAAQ,EAAQ,KAAK,EAAW,GACpC,GAAI,GACF,GAAM,GAAW,GAAqB,EAAM,MAAO,GACnD,EAAQ,EAAQ,EAAO,GAGzB,MAAO,IAGH,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,KAAI,EAAE,SAAQ,GAExC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAO,EAAG,CAAC,MAAK,KChG7B,AAiCA,YACI,EAAoB,EACpB,GACF,GAAM,GAAO,GAAc,GACvB,EAAS,KACb,GAAI,GAAS,MAAQ,IAAU,UAC7B,EAAS,GAAI,cAAa,WACjB,IAAU,QACnB,EAAS,GAAI,YAAW,WACf,IAAU,OACnB,EAAS,GAAI,YAAW,OAExB,MAAM,IAAI,OAAM,qBAAqB,GAEvC,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,IAEd,MAAO,GAAO,WAAW,EAAQ,EAAO,MAG7B,IAAO,EAAG,CAAC,MAAK,6OC1B7B,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,cAEJ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,AAyCA,GAAA,IAAA,WAUE,WACI,EAAc,EAAsB,EACpC,EAAqB,GACvB,KAAK,KAAO,EACZ,KAAK,OAAS,EACd,KAAK,MAAQ,EACb,KAAK,QAAU,IACf,KAAK,UAAY,EACjB,AAAI,KAAK,WACP,MAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,EACvC,KAAK,MAAQ,KAAK,KAAO,KAAK,OAAS,GAEzC,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,OAAS,GAAgB,EAAU,YAInC,SAAA,UAAA,UAAP,WACE,GAAI,CAAC,MAAM,KAAK,UACd,GAAM,GAAQ,KAAK,QACnB,YAAK,QAAU,IACR,EAKT,OAFI,GAAiB,EACjB,EAAU,GACP,CAAC,IACN,GAAI,GAAE,OAAU,EAAE,OAAU,EAAC,OAC7B,EACE,GAAK,EAAI,KAAK,SAAW,EACzB,EAAK,EAAI,KAAK,SAAW,EACzB,EAAI,EAAK,EAAK,EAAK,QACZ,GAAK,GAAK,IAAM,GAEzB,GAAM,GAAM,KAAK,KAAK,GAAO,KAAK,IAAI,GAAK,GAC3C,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EACzC,EAAU,KAAK,KAAO,KAAK,OAAS,EAAK,EAEzC,AAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,GAAU,IAId,MAAI,EAAC,KAAK,WAAa,KAAK,iBAAiB,KAC3C,MAAK,QAAU,KAAK,aAAa,IAE5B,KAAK,aAAa,IAInB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,OAAS,MAAQ,KAAK,QAAU,UAChC,EAEF,KAAK,MAAM,IAIZ,EAAA,UAAA,iBAAR,SAAyB,GACvB,MAAO,IAAS,KAAK,OAAS,GAAS,KAAK,OAEhD,KAIA,GAAA,WASE,WACI,EAAe,EAAc,EAC7B,GACF,KAAK,MAAQ,EACb,KAAK,KAAO,EAAI,EAChB,KAAK,MAAQ,EAEb,GAAM,GAAY,GAAc,KAAK,SACrC,KAAK,MAAQ,GAAgB,EAAU,YACvC,KAAK,MAAQ,GAAI,IAAY,EAAG,EAAG,EAAO,GAAO,KAAK,SAEtD,AAAI,EAAQ,EACV,KAAK,EAAI,EAAS,EAAI,EAEtB,KAAK,EAAI,EAAS,EAAI,EAExB,KAAK,EAAI,EAAI,KAAK,KAAK,EAAI,KAAK,GAI3B,SAAA,UAAA,UAAP,WAEE,OADI,GAAY,EAAY,EAAY,EAAW,EAAW,KAE5D,EACE,GAAI,KAAK,MAAM,YACf,EAAI,EAAK,KAAK,EAAI,QACX,GAAK,GAMd,GALA,GAAK,EAAI,EACT,EAAK,EAAI,EACT,EAAK,EAAK,KAAQ,EAAK,EACvB,EAAM,GAAM,EAAO,KAAK,EAAK,GAAI,EAAI,KAAK,IAAI,IAC9C,EAAI,KAAK,QACL,EAAI,GAAM,KAAK,IAAI,GAAK,EAC1B,MAGJ,SAAK,EAAI,KAAK,KAAQ,KAAK,EAAI,EAC/B,AAAI,KAAK,MAAQ,GACf,IAAK,KAAK,IAAI,KAAK,QAAS,EAAI,KAAK,QAEhC,KAAK,aAAa,IAGnB,EAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,QAAU,UACV,EAEF,KAAK,MAAM,IAEtB,KAEA,GAAA,WAME,WACI,EAAS,EAAS,EAClB,GAFJ,GAAA,GAAA,KAaE,GAZE,AAAA,IAAA,QAAA,GAAA,GAAS,AAAA,IAAA,QAAA,GAAA,GAoBL,KAAA,eAAiB,WACrB,MAAC,GAAK,OAAS,MAAQ,EAAK,QAAU,WAnBxC,KAAK,IAAM,EACX,KAAK,MAAQ,EAAM,EACnB,KAAK,MAAQ,EACb,AAAI,GAAQ,MACV,GAAO,KAAK,UAEd,AAAI,MAAO,IAAS,UAClB,GAAO,EAAK,YAGV,CAAC,KAAK,kBAAoB,KAAK,OAAS,EAC1C,KAAM,IAAI,OACN,0BAA0B,EAAG,MAAM,EAAG,gCAE5C,KAAK,OAAS,GAAgB,GAOxB,SAAA,UAAA,aAAR,SAAqB,GACnB,MAAI,MAAK,iBACA,EAEF,KAAK,MAAM,IAGpB,EAAA,UAAA,UAAA,WACE,MAAO,MAAK,aAAa,KAAK,IAAM,KAAK,MAAQ,KAAK,WAE1D,KCxNA,AAwCA,YACI,EAAoB,EAAe,EACnC,EAAsC,GAOxC,GARqC,AAAA,IAAA,QAAA,GAAA,GACnC,AAAA,IAAA,QAAA,GAAA,WACF,AAAI,GAAQ,MACV,GAAO,GAET,AAAI,GAAS,MACX,GAAQ,WAEN,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,OAAM,yBAAyB,GAI3C,OAFM,GAAS,GAAI,IAAU,EAAO,EAAM,EAAO,GAC3C,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAc,EAAG,CAAC,aAAY,KC5D3C,AAuCA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,yBAAyB,GAK3C,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAuB,GAC1D,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAe,EAAG,CAAC,cAAa,KCtD7C,AA4CA,YACI,EAAoB,EAAY,EAAY,EAC5C,GADoB,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,WAI9C,OAFM,GAAM,GAAO,EAAO,GACpB,EAAS,GAAI,IAAc,EAAQ,EAAQ,KAAM,GAC9C,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAO,YAEzB,MAAO,GAAI,cAGA,IAAgB,EAAG,CAAC,eAAc,KCvD/C,YAwCyB,EAAsB,GAC7C,GAAc,GACd,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,EAC3B,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAkB,KACxB,MAAO,IAAW,EAAQ,EAAO,EAAe,GC/ClD,YA8CI,EAAe,EAAc,EAC7B,GACF,GAF+B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAA,WACE,IAAS,EACX,KAAM,IAAI,OAAM,8BAGlB,GAAM,GAA+B,WACnC,GAAM,GAAgB,IAAU,EAC1B,EAA8B,EAAQ,GAAQ,EAAO,EACrD,EAA8B,EAAO,GAAS,EAAO,EAE3D,GAAI,GAAiB,GACjB,EACF,MAAO,IAAM,CAAC,GAAI,GAGpB,GAAM,GAAc,KAAK,IAAI,KAAK,KAAM,GAAO,GAAS,IAClD,EAAS,GAAoB,EAAa,GAEhD,AAAI,EAAO,GAAS,IAAS,GAG3B,GAAO,IAGT,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,IAGpB,EAAoB,CAAC,MAAK,EAAE,KAAI,EAAE,KAAI,EAAE,MAAK,GAEnD,MAAO,GAAO,cACH,EAAS,GAAiB,KAAiB,GAC3C,GCnFb,AAsCA,YAAuC,GACrC,GAAM,GAAK,EAAgB,EAAG,IAAK,cAE7B,EAA2B,CAAC,EAAG,GACrC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,WAAW,GAC/B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAa,EAAG,CAAC,YAAW,KChDzC,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,KAAK,IAGhB,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAwCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAG7C,MAFA,GAAK,CAAC,IAEF,EAAG,QAAU,OACR,GAAK,EAAI,SAGX,EAAQ,MAAM,IAGjB,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAQ,EAAG,CAAC,OAAM,KC5D/B,AA6DA,YACI,EAAiB,GACnB,GAAM,GAAK,EAAgB,EAAG,IAAK,WAE7B,EAA+B,SAAC,GACpC,GAAM,GAAO,GAAe,EAAM,EAAG,OACrC,GAAI,EAAG,OAAS,EACd,MAAO,IAAM,GAEf,GAAM,GAAM,EAAQ,QAAQ,EAAI,GAChC,MAAO,GAAQ,EAAK,EAAG,QAGnB,EAAwB,CAAC,EAAG,GAC5B,EAAsB,CAAC,KAAM,GAEnC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAS,MAGT,IAAU,EAAG,CAAC,SAAQ,KClFnC,AA6BA,YAAoB,GAClB,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCrCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AA+BA,YAAoB,EAAwB,GAC1C,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WAAM,MAAA,qDAAqD,EAAG,KAAI,MAC/D,GAAQ,EAAI,MAGR,IAAY,EAAG,CAAC,WAAU,KCvCvC,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAC7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cACV,SAAC,GAAY,MAAA,GAAQ,MAAM,IAAK,EAChC,KAAiB,OAGV,IAAQ,EAAG,CAAC,OAAM,KChD/B,AAuCA,YAAkC,GAChC,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAAsB,CAAC,EAAG,GAEhC,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,MAAM,GAC1B,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAQ,EAAG,CAAC,OAAM,KClD/B,AAwCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GAGH,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,OAI1D,IAAO,EAAG,CAAC,MAAK,KCe7B,YACI,EAAiB,EACjB,EAAsC,EACtC,EAAqB,EACrB,GADqB,AAAA,IAAA,QAAA,GAAA,CAAqC,EAAG,IAC7D,AAAA,IAAA,QAAA,GAAA,QACF,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EACF,EAAgB,EAAiB,kBAAmB,mBAClD,EACF,EAAgB,EAAiB,kBAAmB,mBAEpD,EAAM,EACN,EAAe,GAMnB,GALA,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAGvD,IAAe,OACjB,KAAM,IAAI,OACN,sFAIN,EACI,EAAI,OAAS,EACb,WAAM,MAAA,2DACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,OAAS,EAC1B,WAAM,MAAA,kEACF,aAAY,EAAiB,KAAI,OACzC,EACI,EAAiB,MAAM,KAAO,EAC9B,WACI,MAAA,qEACA,wBAAuB,EAAiB,MAAM,GAAE,OACxD,EACI,EAAiB,MAAM,KAAO,EAC9B,WAAM,MAAA,+DACF,8BAA6B,EAAiB,MAAM,GAAE,OAE9D,GAAM,GAAa,EAAiB,MAAM,GACpC,EAAoB,EAAiB,MAAM,GACjD,EACI,EAAiB,MAAM,KAAO,EAAa,EAC3C,WACI,MAAA,qEACA,YAAW,EAAa,EAAiB,MACzC,YAAW,EAAiB,MAAM,GAAE,OAE5C,GAAM,GAAY,GACd,EAAK,EAAkB,EAAS,EAAK,EAAY,GAC/C,EAAkB,EAClB,EACF,GAAO,EAAW,EAAkB,EAAiB,QAAS,GAElE,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,KAGI,IAAkB,EAAG,CAAC,iBAAgB,KCxFnD,YACI,EAAsB,uHAClB,SAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aAEnC,EACI,EAAG,QAAU,EAAG,MAChB,WAAM,MAAA,kDACF,EAAG,MAAK,YAAY,EAAG,MAAK,OAEpC,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAEtE,EACI,EAAG,OAAS,EAAG,WAAM,MAAA,qCAAqC,EAAG,MAAK,OAExD,CAAA,EAAM,EAAG,eAAjB,SAAQ,EAAA,OACA,CAAA,EAAM,EAAG,eAIvB,IAJM,EAAQ,EAAA,OACR,EAAO,GAAI,KAAI,GAEjB,EAAa,EACR,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,AAAK,EAAK,IAAI,EAAM,KAClB,IAMJ,IAFM,EAAS,GAAI,IAAa,CAAC,GAAa,EAAG,OAC3C,EAAU,GAAI,IAAa,CAAC,GAAa,SACtC,EAAI,EAAG,EAAI,EAAG,EAAI,EAAM,OAAQ,IACvC,AAAK,EAAK,IAAI,EAAM,KAClB,GAAO,OAAO,GAAK,EAAM,GACzB,EAAQ,OAAO,GAAK,EACpB,KAGJ,MAAA,CAAA,EAAO,CAAC,EAAO,WAAY,EAAQ,qBAExB,IAAiB,GCxF9B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAC/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,IAAK,EAC7B,KAAiB,OAEV,IAAO,EAAG,CAAC,MAAK,KC7C7B,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAC7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KChD7B,AA6BA,YACI,EAAwB,EAAe,GACzC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,CAAC,GAAQ,CAAC,OAEhB,IAAU,EAAG,CAAC,SAAQ,KCtCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AA6BA,YACI,EAAwB,EACxB,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,SACI,EAAG,OAAS,EACZ,WACI,MAAA,mDAAmD,EAAG,KAAI,YAC3D,GAAM,EAAI,EAAO,MAEb,IAAU,EAAG,CAAC,SAAQ,KCvCnC,AAgDA,YAAoC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxD,GAAM,GAAU,EAAgB,EAAQ,SAAU,UAAW,WAK7D,GAHA,AAAI,IAAQ,IACV,GAAM,EAAQ,KAAO,GAEnB,IAAQ,EAAQ,KAAO,EACzB,KAAM,OACF,4DACA,oBAAmB,EAAQ,KAAI,gBAAgB,IAGrD,GAAM,GAAwB,CAAC,OAAQ,GACjC,EAAsB,CAAC,IAAG,GAEhC,MAAO,GAAO,cACV,SAAC,EAAS,GACR,GAAM,GAAI,EAAQ,QAAQ,EAAS,GACnC,SAAK,CAAC,IACC,GAET,EAAgC,KAAiB,GACjD,MAGO,IAAU,EAAG,CAAC,SAAQ,KCzEnC,AAyCA,YAAc,GACZ,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,qDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAoB,CAAC,MAAK,GAEhC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAU,EAAM,KAAK,EAAO,GAC5B,EAAS,EAAQ,IAAI,GAC3B,MAAO,GAAO,QAAQ,EAAM,QAC3B,EAAgC,KAAqB,OAG7C,IAAM,EAAG,CAAC,KAAI,KC5D3B,AA0CA,YAAe,GACb,EACI,EAAM,QAAU,YAChB,WAAM,MAAA,sDACF,YAAW,EAAM,MAAK,OAE9B,GAAM,GAAqB,CAAC,MAAK,GAEjC,MAAO,GAAO,cAAc,SAAA,GAE1B,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EAErB,EAAoB,EAAQ,EAAO,CAAC,EAAO,IAC3C,EAAS,EAAQ,KAAK,GAC5B,MAAO,GAAQ,EAAQ,EAAM,QAC5B,EAAgC,KAAqB,OAG7C,IAAO,EAAG,CAAC,MAAK,KC7D7B,AAgDA,YAAgB,GACd,GAAM,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAM,KAAO,EACvB,EACJ,GAAI,GAAsB,GACxB,GAAM,GAAe,EAAQ,EAAO,CAAC,EAAO,IAC5C,EAAM,GAAK,QAIX,GAAM,GAAc,CAAC,EAAO,EAAK,GAAqB,IAChD,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IACzC,EAAY,EAAQ,GAAK,GAAQ,CAAC,EAAO,IAEzC,EACF,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACjE,EAA0B,EAC5B,GAAQ,GAAM,EAAW,CAAC,EAAG,GAAI,CAAC,EAAO,EAAqB,IAAK,GACnE,GAAO,KAEL,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EAAI,GAAO,CAAC,EAAW,GAAgB,GACvC,EACF,EAAQ,GAAQ,EAAG,GAAI,CAAC,EAAY,GAAI,EAAY,KACxD,EAAM,GAAK,GAIb,GAFA,EAAM,GAAK,GAEP,EAAM,OAAS,GAAK,EAAM,MAAM,KAAO,GACzC,GAAM,GAAO,EACP,EAAQ,EAAM,MAAM,GAC1B,EAAM,EAAQ,EAAK,CAAC,EAAO,EAAI,MAAM,GAAK,EAAO,EAAI,MAAM,KAC3D,EAAK,UAEP,MAAO,MAGI,IAAQ,EAAG,CAAC,OAAM,iBC3D3B,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAa,GACjB,GAAI,MAAQ,IAAqB,SAC/B,EACI,EAAE,MAAM,GAAQ,IAAoB,EACpC,WAAM,MAAA,kDACV,EACI,GAAI,OAAM,GAAiB,KAAK,EAAE,MAAM,GAAQ,QAEpD,GAAM,GAAY,EAAgB,OAAO,SAAC,EAAO,GAC/C,MAAI,KAAU,IACZ,IAAS,GAEJ,GACN,GACH,EACI,GAAa,EACb,WAAM,MAAA,4DACV,GAAM,GAAW,EAAgB,QAAQ,IAGzC,GAAI,IAAa,IACf,GAAM,GAAQ,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,EAAI,EAAI,EAAI,IAC/D,EAAgB,GAAY,EAAE,MAAM,GAAQ,EAE9C,EACI,EAAE,MAAM,KAAU,EAAgB,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IACvD,WAAM,MAAA,gEACV,EAAa,EAGf,MAAO,GC1DT,AA+DA,YACI,EAAsB,EAAkC,GAAA,AAAA,IAAA,QAAA,GAAA,GAC1D,GAAM,GAAK,EAAgB,EAAG,IAAK,SAE7B,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAQ,GAAe,EAAM,EAAG,OAAO,GACvC,EAAa,GAAiB,EAAI,EAAiB,GACzD,MAAO,GAAQ,MAAM,EAAI,EAAY,IAGjC,EAAuB,CAAC,EAAG,GAC3B,EAAoB,CAAC,gBAAe,EAAE,KAAI,GAEhD,MAAO,GAAO,cACH,EAAS,EAAgC,KAAiB,GAC1D,MAGA,IAAQ,EAAG,CAAC,OAAM,KCjF/B,AA+CA,YAAe,EAAe,GAC5B,EACI,EAAM,QAAU,UAChB,WAAM,MAAA,mDAAmD,EAAM,QAEnE,GAAI,GAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACpD,EAAQ,EAAM,KAAO,EAEvB,EACJ,GAAI,GAAa,MAAQ,EAAY,GAEnC,GAAM,GAAQ,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAC7B,EAAO,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KAClC,EAAK,EAAM,MAAM,OAAS,GAAK,EAC/B,EAAgB,GAAM,EAAO,EAAO,GACpC,EAAqB,UACZ,GAAa,MAAQ,EAAY,GAE1C,GAAM,GAAa,EAAM,MAAM,IAAI,SAAA,GAAK,MAAA,KACxC,EAAW,EAAM,MAAM,OAAS,GAAK,EAAY,EACjD,EAAgB,GAAO,CAAC,EAAO,GAAM,IAAc,EAAM,MAAM,OAAS,GACxE,EAAqB,MAErB,GAAgB,EAIlB,GAAM,GAAa,GAAU,GACvB,EACF,EAAQ,GAAQ,EAAe,GAAa,CAAC,EAAO,IAElD,EAAM,GAAI,GAGV,EAAO,KAAK,MAAM,EAAqB,GAAK,EAC5C,EAAa,GAAK,GAClB,EAAa,GAAK,GAClB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GACxB,EAAuB,GACzB,EAAY,CAAC,EAAM,EAAqB,GACxC,EAAW,MAAM,OAAS,GAExB,EAAc,EAAc,MAAM,QACxC,SAAY,EAAc,MAAM,OAAS,GAAK,EAEvC,EACH,GAAQ,EAAqB,GAAI,EAAqB,IAAK,MAGpD,IAAO,EAAG,CAAC,MAAK,KClG7B,AAsCA,YAAiC,GAC/B,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GAE/B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,KAAK,GACzB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAO,EAAG,CAAC,MAAK,KCjD7B,AAoDA,YACI,EAAsB,SACpB,EAAK,EAAgB,EAAG,IAAK,qBAC7B,EAAK,EAAgB,EAAG,IAAK,qBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAA2B,EAAG,MAAO,EAAG,OAExC,GAAM,GAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,kBAAkB,EAAI,GAC1C,SAAK,CAAC,EAAI,IACH,GAGH,EAAkC,CAAC,EAAG,EAAI,EAAG,GAC7C,EAAQ,GAEd,MAAO,GAAO,cACH,EAAS,EAAqC,KAC9C,GAAmB,MAGnB,IAAoB,EAAG,CAAC,mBAAkB,KC1EvD,AAwCA,YAAoC,EAAsB,GACxD,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,MAAO,GAAQ,EAAI,GAAa,EAAG,MAAO,GAAM,aAGrC,IAAU,EAAG,CAAC,SAAQ,KC7CnC,AAyCA,YACI,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,GAChC,GAAM,GAAW,GAAqB,EAAS,UAAW,SAK1D,GAHA,EACI,EAAS,QAAU,EAAG,WAAM,MAAA,yCAE5B,EAAS,SAAW,EACtB,MAAO,IAAW,EAAS,GAAI,GAGjC,GAAM,GAAO,EAAS,GAAG,KACnB,EAAQ,EAAS,GAAG,MACpB,EAAQ,EAAS,GAAG,MAE1B,EAAY,GAAQ,EAAM,WAAM,MAAA,uCAEhC,EAAS,QAAQ,SAAA,GACf,GACI,EAAO,EAAE,MACT,yDACJ,EACI,IAAU,EAAE,MACZ,WAAM,MAAA,4DAGZ,GAAM,GAAkB,EAAS,IAAI,SAAA,GAAK,MAAA,IAAW,EAAG,KAOxD,MAAO,IAAO,EAAiB,MAGpB,IAAQ,EAAG,CAAC,OAAM,KC7E/B,AAwCA,YAAiC,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAChD,GAAM,GAAK,EAAgB,EAAG,IAAK,QAE7B,EAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,MAAK,GAE/B,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,KAAK,EAAI,IAAQ,EACpC,KAAiB,GAAM,MAEhB,IAAO,EAAG,CAAC,MAAK,KClD7B,AA+DA,YACI,EAAsB,EAAiB,EAAe,EACtD,EAAe,EAAa,EAAkB,EAC9C,GADA,AAAA,IAAA,QAAA,GAAA,GAAe,AAAA,IAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAAkB,AAAA,IAAA,QAAA,GAAA,GAC9C,AAAA,IAAA,QAAA,GAAA,GACF,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAE3B,EAA+B,SAAC,GACpC,AAAI,GAAW,MACb,GAAU,GAAI,OAAM,EAAM,SAG5B,GAAM,GAAe,GAAW,GAChC,GAAI,EAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAI,IAAiB,GAAK,IAAgB,EACxC,KAAM,IAAI,OACN,iEAGN,GAAI,IAAiB,GAAK,IAAmB,EAC3C,KAAM,IAAI,OACN,oEAGN,GAAM,GAAsB,EAAG,KAAO,EAAM,OAGtC,EAAa,GAAW,GACxB,EAAW,EAAG,MAAM,QAC1B,EAAW,QAAQ,SAAA,GACjB,EAAM,GAAQ,EACd,EAAI,GAAQ,EACZ,EAAS,OAAO,EAAM,EAAG,KAE3B,EAAK,EAAQ,EAAI,GAEX,GAAA,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GACJ,EAAA,EAAA,MACA,EAAA,EAAA,IACA,EAAA,EAAA,QAKF,EAAQ,EACR,EAAM,EACN,EAAU,EAEV,GAAM,GAAa,GAAW,GAE9B,EAAW,QAAQ,SAAA,GACjB,EAAI,GAAQ,EAAM,GAAQ,EAC1B,EAAQ,GAAQ,IAIlB,GAAM,GAAO,GAAgB,EAAO,EAAK,GAEnC,EAAW,EAAK,OAAO,SAAC,EAAG,GAAS,MAAA,GAAW,QAAQ,KAAU,KAEjE,EAAa,EAAQ,MAAM,SAAA,GAAK,MAAA,KAAM,IAC5C,GAAI,EACF,MAAO,GAAQ,GAAM,EAAI,EAAO,GAAO,GAGzC,GAAM,GAAM,EAAQ,aAAa,EAAI,EAAO,EAAK,GACjD,MAAO,GAAQ,EAAK,IAGhB,EAA6B,CAAC,EAAG,GACjC,EAA2B,CAC/B,MAAK,EACL,IAAG,EACH,QAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,YAAW,EACX,eAAc,GAGhB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAe,EAAG,CAAC,cAAa,KCvJ7C,AAsCA,YAAgC,GAC9B,GAAM,GAAK,EAAgB,EAAG,IAAK,OAE7B,EAAoB,CAAC,EAAG,GAE9B,MAAO,GAAO,cAAc,SAAC,EAAS,GACpC,GAAM,GAAM,EAAQ,IAAI,GACxB,SAAK,CAAC,IACC,GACN,EAAgC,KAAiB,OAEzC,IAAM,EAAG,CAAC,KAAI,KCjD3B,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,kEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,gFAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,sEAEN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GChElD,YAgDI,EAAsB,EACtB,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,kDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,wEAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,MAAO,IAAW,EAAQ,EAAO,EAAe,GCjElD,YAgDI,EACA,EACA,GAEF,GADA,GAAc,GACV,GAAS,MAAQ,EAAM,SAAW,EACpC,KAAM,IAAI,OAAM,iDAElB,GAAM,GAAgB,GAAW,EAAQ,GACzC,GAAI,EAAc,SAAW,GAAK,EAAc,SAAW,EACzD,KAAM,IAAI,OACN,0EAGN,GAAI,EAAc,SAAW,GAAK,GAAS,KACzC,KAAM,IAAI,OACN,2EAGN,SAAQ,GACJ,EACG,GAAW,EAAQ,EAAO,EAAe,GCpElD,AAmDA,YACI,EAAiB,EAAO,GAAP,AAAA,IAAA,QAAA,GAAA,GAAO,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAK,EAAgB,EAAG,IAAK,QACnC,GAAI,EAAG,OAAS,EACd,KAAM,IAAI,OAAM,sDAElB,GAAM,GAAU,EAAG,MAAM,EAAG,MAAM,OAAS,GAC3C,GAAI,EAAI,EACN,KAAM,IAAI,OACN,uDAAuD,EAAO,KAC9D,YAAW,IAGjB,GAAM,GAAqB,CAAC,EAAG,GACzB,EAAmB,CAAC,EAAC,EAAE,OAAM,GAE7B,EAAA,EAAA,cAAA,SAAA,GAAA,MAAA,GAAA,KAAA,EAAA,EAAA,IAAA,EAAA,KAAA,GAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAKf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAO,EAAG,CAAC,MAAK,KC3E7B,AA4CA,YACI,EAAoB,EAAU,EAAY,EAC1C,GACF,GAFsB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAE5B,GAAS,MAAS,IAAuB,OAC3C,KAAM,IAAI,OAAM,qCAKlB,OAHM,GACF,GAAI,IAAY,EAAM,EAAQ,EAAO,GAAsB,GACzD,EAAM,GAAO,EAAO,GACjB,EAAI,EAAG,EAAI,EAAI,OAAO,OAAQ,IACrC,EAAI,OAAO,GAAK,EAAU,YAE5B,MAAO,GAAI,cAGA,IAAkB,EAAG,CAAC,iBAAgB,KC3DnD,AA6EA,YACI,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,GAEnB,GAAM,GAAK,EAAgB,EAAG,IAAK,SAAU,MAC7C,EAAO,EAAG,KAAO,EAAG,WAAM,MAAA,yCAE1B,GAAM,GAAuB,CAAC,EAAG,GAC3B,EAAqB,CAAC,KAAI,GAC1B,EAAA,EAAA,UAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGf,MAAO,CAAC,OAAM,EAAE,QAAO,MAGZ,IAAS,EAAG,CAAC,QAAO,KC3FjC,AA8CA,YACI,EAAiB,EAAiC,GACpD,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EACF,EAAgB,EAAY,aAAc,qBAAsB,SACpE,EAAO,GAAM,GAAc,WAAM,MAAA,qCAEjC,GAAM,GAAmC,CAAC,EAAG,EAAI,WAAY,GACvD,EAAiC,CAAC,YAAW,GAE7C,EAA+B,SAAC,EAAS,GAC7C,GAAM,GAAM,EAAQ,mBAAmB,EAAI,EAAa,GACxD,SAAK,CAAC,IACC,GAGT,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAoB,MAGpB,IAAqB,EAAG,CAAC,oBAAmB,KCnEzD,AA2CA,YAAkB,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,GACtC,GAAM,GAAK,EAAgB,EAAG,IAAK,WACnC,EACI,GAAQ,CAAC,EAAG,MAAM,QAAU,EAAO,EAAG,MAAM,OAC5C,WACI,MAAA,UAAU,EAAI,gBAAgB,EAAG,MAAM,OAAM,KAAK,EAAG,MAAM,OAAM,MACzE,AAAI,EAAO,GACT,IAAQ,EAAG,MAAM,QAEnB,GAAM,GAAuB,CAAC,MAAO,GAC/B,EAAqB,CAAC,KAAI,GAC1B,EAAiC,SAAC,GACpC,MAAA,GAAQ,QAAQ,EAAI,IACxB,MAAO,GAAO,cACV,EAAS,EAAgC,KAAiB,GAC1D,MAGO,IAAU,EAAG,CAAC,SAAQ,KC7DnC,YAsCI,EAAyB,EAAkB,EAC3C,GADyB,MAAA,KAAA,QAAA,GAAA,IAEpB,EAAO,aAAa,EAAc,EAAW,EAAM,GCxC5D,YAuB0B,EAAqB,GAE7C,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,AAAI,EAAS,IACX,EAAQ,KAAK,GAOjB,OAHM,GAAW,GAAO,EAAW,SAE7B,EAAM,GAAO,CAAC,EAAQ,OAAQ,EAAU,QAAS,SAC9C,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAM,EAAS,WAAW,EAAQ,IAClC,EAAS,EAAI,EAAU,OAC7B,EAAI,OAAO,IAAI,EAAK,GAEtB,MAAO,GAAI,WCEb,YAA2B,uGACnB,SACF,EAAgB,EAAW,YAAa,aAAc,QAC7C,CAAA,EAAM,EAAW,eAAxB,SAAO,EAAA,OACP,EAAM,GAAU,EAAW,MAAO,GACxC,AAAI,IAAc,GAChB,EAAW,UAEb,CAAA,EAAO,WAGI,IAAa,GCpD1B,AA6CA,YACI,EAA2B,EAC3B,2HAcF,IAbM,EAAU,EAAgB,EAAQ,SAAU,YAC5C,EAAQ,EAAgB,EAAM,OAAQ,WAAY,QAElD,EAAW,GAAe,EAC1B,EAAU,EAAM,KAChB,EAAc,EAAQ,MAE5B,EAAY,EAAU,EAAG,WAAM,MAAA,0BAC/B,GACI,EAAY,MAAM,EAAU,EAAW,GAAU,EAAM,MACvD,qEAEA,EAAc,EACT,EAAI,EAAU,EAAI,EAAW,EAAS,IAC7C,GAAe,EAAY,GAEvB,SACF,EAAY,MAAM,EAAG,GAChB,OAAO,CAAC,GAAc,EAAY,MAAM,EAAW,IACtD,EAAiB,EAAQ,EAAS,GAClC,EAAe,EAAQ,EAAO,CAAC,KACX,CAAA,EAAM,GAAW,WAArC,SAAoB,EAAA,OACpB,EAAU,GAAQ,EAAmB,CAAC,IAEtC,EAAM,GAAO,EAAgB,EAAS,GAG5C,AAAI,IAAW,GACb,EAAQ,UAEV,AAAI,IAAS,GACX,EAAM,UAER,EAAQ,UACR,EAAe,UACf,EAAa,UACb,EAAkB,UAElB,CAAA,EAAO,WAGI,IAAmB,GCzFhC,AAuCA,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,kBAC7B,EAAK,EAAgB,EAAG,IAAK,kBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,6BAC/B,GAAS,EAAI,GAYtB,YAAuC,EAAiB,GACtD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,cAC7B,EAAK,EAAgB,EAAG,IAAK,cACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,yBAC/B,GAAK,EAAI,GAGlB,YAAwC,EAAiB,GACvD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,eAC7B,EAAK,EAAgB,EAAG,IAAK,eACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,0BAC/B,GAAM,EAAI,GAGnB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAK,EAAgB,EAAG,IAAK,mBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,8BAC/B,GAAU,EAAI,GAGvB,YAA0C,EAAiB,GACzD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,4BAC/B,GAAQ,EAAI,GAGrB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,sBAC7B,EAAK,EAAgB,EAAG,IAAK,sBACnC,UAAkB,EAAG,MAAO,EAAG,MAAO,iCAC/B,GAAa,EAAI,MAGb,IAAc,EAAG,CAAC,aAAY,KAC9B,GAAqB,EAAG,CAAC,oBAAmB,KAC5C,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAkB,EAAG,CAAC,iBAAgB,KACtC,GAAa,EAAG,CAAC,YAAW,KAC5B,GAAiB,EAAG,CAAC,gBAAe,KCpHjD,AA2CA,YAAsC,EAAiB,GACrD,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAajB,YAAsC,EAAS,GAC7C,UACI,6EAGJ,GAAuB,EAAK,MAAO,EAAI,MAAO,wBACvC,GAAI,EAAM,GAanB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,6BACpC,EAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,OAC7B,EAAK,EAAgB,EAAG,IAAK,OACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,2BACpC,GAAI,EAAI,GAWjB,YAAsC,EAAiB,GACrD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,aAC7B,EAAK,EAAgB,EAAG,IAAK,aACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,wBACpC,GAAI,EAAI,GAWjB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAWrB,YAA0C,EAAiB,GACzD,GACI,6EAGJ,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBACnC,UAAuB,EAAG,MAAO,EAAG,MAAO,4BACpC,GAAQ,EAAI,GAarB,YACI,EAAiB,GACnB,GACI,6EAEJ,GAAM,GAAK,EAAgB,EAAG,IAAK,2BAC7B,EAAK,EAAgB,EAAG,IAAK,2BACnC,UACI,EAAG,MAAO,EAAG,MAAO,sCACjB,GAAkB,EAAI,MAGlB,IAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAgB,EAAG,CAAC,eAAc,KAClC,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAAY,EAAG,CAAC,WAAU,KAC1B,GAA0B,EAAG,CAAC,yBAAwB,KACtD,GAAY,EAAG,CAAC,WAAU,KC5NvC,AAwEA,YACI,EAAsB,EACtB,EAA8B,GADR,AAAA,IAAA,QAAA,GAAA,aACtB,AAAA,IAAA,QAAA,GAAA,MAA8B,AAAA,IAAA,QAAA,GAAA,IAChC,EAAI,EAAgB,EAAG,IAAK,QAE5B,GAAM,GAAO,GAAS,EAAG,EAAK,GAC1B,EAAgB,EAAK,MACzB,GAAI,GACF,GAAM,GAAO,GAAe,EAAM,EAAE,OACpC,EAAgB,GAA+B,EAAK,MAAO,GAE7D,MAAO,GAAQ,EAAM,GAGvB,YACI,EAAW,EAAkB,GAC/B,GAD+B,AAAA,IAAA,QAAA,GAAA,MAC3B,EAAE,OAAS,EACb,MAAO,IAAI,GAIb,GAAI,EAAE,OAAS,GAAK,IAAS,KAC3B,MAAO,IAAS,EAAQ,EAAG,CAAC,KAAM,EAAG,GAIvC,GAAI,EAAE,OAAS,GAAK,MAAO,IAAS,UAChC,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAErB,GAAI,IAAM,aAAe,IAAM,EAE7B,MAAO,IAAK,GAAI,GAAI,GAAI,GAAI,GAAO,EAAG,UAAW,IAGnD,KAAM,IAAI,OAAM,qCAAqC,GAIvD,GAAI,MAAM,QAAQ,IAAS,EAAK,SAAW,GACzC,GAAI,IAAM,EACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,GAAK,GAE7C,GAAI,IAAM,SACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,UACR,MAAO,IAAI,GAAI,GAAI,GAAI,EAAK,IAAK,EAAK,IAExC,GAAI,IAAM,OAAS,IAAM,YAEvB,MAAO,IAAK,GAAI,GAAO,GAAI,IAG7B,KAAM,IAAI,OAAM,qCAAqC,GAGvD,KAAM,IAAI,OAAM,gCAAgC,MAGrC,IAAO,EAAG,CAAC,MAAK,KC3I7B,AA2DA,YACI,EAAiB,EAAiB,EAClC,EAAsB,GAAA,AAAA,IAAA,QAAA,GAAA,IACxB,GAAM,GAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAK,EAAgB,EAAG,IAAK,iBAC7B,EAAS,EAAgB,EAAO,QAAS,iBAE/C,GAAiB,EAAI,GACrB,EACI,GAAiB,EAAG,MAAO,EAAG,OAAQ,WAAM,MAAA,8BAEhD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAI,EAAK,GAE3B,EAAS,EAAI,GAAI,EAAI,GAAK,GAC9B,GAAI,GACF,EACI,GAAQ,KAAM,WAAM,MAAA,mDACxB,GAAM,GAAQ,EAAgB,EAAM,OAAQ,iBAC5C,EAAS,GAAI,EAAQ,GAAI,EAAK,GAAI,EAAQ,KAE5C,MAAO,IAAI,EAAI,MAGJ,IAAgB,EAAG,CAAC,eAAc,KCnF/C,AA+CA,YACI,EAA4B,EAC5B,GACF,GAAM,GAAW,EAAgB,EAAS,UAAW,YAAa,SAC5D,EAAW,EAAgB,EAAS,UAAW,aACrD,GAA8B,EAAU,EAAU,GAElD,GAAM,GAA+B,SAAC,GACpC,MAAO,GAAQ,UAAU,EAAU,EAAU,IAGzC,EAA0B,CAAC,QAAS,EAAU,QAAS,GACvD,EAAwB,CAAC,MAAK,GAEpC,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAW,MAGX,IAAY,EAAG,CAAC,WAAU,iBClCnC,EAAuB,EAAsB,EAC7C,GACF,GAAI,EAAc,QAAU,QAC1B,KAAM,IAAI,OACN,2DACA,uBAAsB,EAAc,MAAK,MAE/C,GAAI,EAAc,KAAO,EACvB,KAAM,IAAI,OACN,uDACA,mBAAkB,EAAc,MAAK,MAG3C,GAAM,GAAW,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAC7D,EAAU,EAAc,KAAO,EAAI,EAAc,MAAM,GAAK,EAElE,GAAI,EAAY,SAAW,EACzB,KAAM,IAAI,OACN,iDACA,KAAI,EAAY,OAAM,gBAAgB,EAAO,MAGnD,GAAM,GAAY,EAAa,KAC/B,GAAI,CAAE,GAAa,OAAS,GACtB,EAAa,OAAS,GAAK,IAAc,GAC7C,KAAM,IAAI,OACN,oCACG,GAAa,MAAK,sBAAsB,EAAQ,MAGzD,GAAI,EAAa,QAAU,EAAc,MACvC,KAAM,IAAI,OAAM,qDC/DpB,AAkEA,YACI,EAAkC,EAClC,EAA0B,GAAA,AAAA,IAAA,QAAA,GAAA,GAC5B,GAAM,GACF,EAAgB,EAAe,gBAAiB,gBAAiB,SAC/D,EACF,EAAgB,EAAc,eAAgB,iBAC5C,EAAgB,EAClB,EAAc,eAAgB,gBAAiB,EAAc,OAEjE,GACI,EAAgB,EAAe,EAAa,GAEhD,GAAM,GAA8B,CAClC,cAAe,EACf,aAAc,EACd,aAAc,GAGV,EAA4B,CAAC,YAAW,GAE9C,MAAO,GAAO,cACV,SAAA,GAAW,MAAA,GAAQ,cACf,EAAgB,EAAe,EAAa,IAChD,EAAgC,KAAiB,GACjD,MAGO,IAAgB,EAAG,CAAC,eAAc,KC9F/C,AA8DA,YAAmB,EAAsB,GACvC,GAAM,GAAW,EAAgB,EAAS,UAAW,WAAY,SAC3D,EAAK,EAAgB,EAAG,IAAK,YAE7B,EAA+B,SAAC,GACpC,MAAO,GAAQ,SAAS,EAAI,IAGxB,EAAyB,CAAC,OAAQ,EAAI,QAAS,GAErD,MAAO,GAAO,cACV,EAAS,EAAgC,KAAqB,OAGvD,IAAW,EAAG,CAAC,UAAS,KC5ErC,YA4B8B,EAAW,GACvC,GAAI,GAAc,KAChB,MAAO,GAAE,MAAM,QAEjB,GAAI,GAAiB,EAAE,MAAO,GAC5B,MAAO,GAET,GAAI,EAAE,MAAM,SAAW,EAAW,QAEhC,OADM,GAAyB,GACtB,EAAI,EAAG,EAAI,EAAE,MAAM,OAAQ,IAClC,AAAI,EAAW,IAAM,MAAQ,EAAE,MAAM,IAAM,KACzC,EAAa,KAAK,EAAE,MAAM,IAE1B,EAAa,KAAK,EAAW,IAGjC,MAAO,GAGT,MAAO,GC/CT,AAoDA,YACI,EAAsB,EAAc,EACpC,GACF,GAAM,GAAK,EAAgB,EAAG,IAAK,WAUnC,GARA,EACI,EAAG,QAAU,UACb,WAAM,MAAA,8DACF,sBAAqB,EAAG,MAAK,sBACrC,EACI,GAAQ,GAAK,EAAO,EACpB,WAAM,MAAA,qDAAqD,EAAI,MAE/D,IAAS,EACX,MAAO,aAAa,GAAS,EAAG,QAAU,EAG5C,GAAM,GAAc,GAAc,EAAI,GAChC,EAAW,EAAI,EACf,EAAa,GACf,GAAM,GAAI,GAAc,EAAa,EAAG,EAAG,UAAW,GAAO,IAC7D,GAEJ,MAAO,GAAI,EAAI,MAGJ,IAAU,EAAG,CAAC,SAAQ,KC9EnC,YAoBoC,GAElC,MAAO,MAAK,MAAM,KAAK,IAAI,EAAG,KAAK,KAAK,KAAK,IAAI,GAAS,KAAK,IAAI,kBAIjE,EAAsB,EAAW,GAGnC,OAFM,GAAO,EAAI,EAAe,EAC1B,EAAY,GAAI,cAAa,GAC1B,EAAI,EAAG,EAAI,EAAc,EAAE,GAClC,GAAM,GAAU,EAAM,KAAK,GAAK,EAAM,GAAe,EAAO,GAC5D,EAAU,GAAK,EAAI,EAAI,KAAK,IAAI,GAElC,MAAO,IAAS,EAAW,WCjC7B,AAwCA,YACI,EAA2B,EAAuB,GAAA,MAAA,KAAA,QAAA,GAAA,wHAC9C,SAAe,EAAgB,EAAa,cAAe,UAC3D,EAAW,EAAgB,EAAS,UAAW,UAErD,EACI,EAAa,KAAO,EACpB,WAAM,MAAA,+DACF,YAAW,EAAa,QAChC,EACI,EAAa,KAAO,IAAM,EAAS,KACnC,WAAM,MAAA,mFAEC,GAAa,KAAI,qBAAqB,EAAS,QAC1D,GACI,EAAa,MAAM,MAAM,EAAG,EAAa,MAAM,OAAS,GACxD,EAAS,MACT,2FAEE,EAAU,EAAa,MAAM,EAAa,MAAM,OAAS,GAC/D,EACI,EAAI,GAAK,GAAK,EACd,WAAM,MAAA,iEACF,eAAc,EAAO,cAAc,KAEnB,CAAA,EAAM,EAAa,eAArC,SAAkB,EAAA,OACJ,CAAA,EAAM,EAAS,eAOnC,IAPM,EAAc,EAAA,OAId,EAAgB,CAAC,EAAgB,OAAS,EAAS,GAAlD,EAAK,EAAA,GAAE,EAAI,EAAA,GACZ,EAAY,GAAuB,OAAQ,GAExC,EAAI,EAAG,EAAI,EAAO,KAIzB,IAHM,EAAS,EAAI,EACb,EAAO,EAAgB,SAAS,EAAQ,EAAS,GACjD,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAKzC,IAHA,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAErC,EAAU,GAAK,EACN,EAAI,EAAG,EAAI,EAAG,IACrB,GAAI,EAAU,GAAG,QAAU,EAAY,IACrC,EAAU,GAAK,EACf,OAKN,MAAI,KAAgB,GAClB,EAAa,UAEf,AAAI,IAAY,GACd,EAAS,UAIX,CAAA,EAAO,GAAO,EAAW,EAAS,MAAO,iBAG9B,IAAc,GCtG3B,AAiDA,YACI,EAAM,EAAO,EACb,EACA,EACA,EACA,GADA,AAAA,IAAA,QAAA,GAAA,QAEF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE5D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,GAAM,GAAU,IAAe,OAAS,EAAI,MAAM,GAAK,EAAI,MAAM,GAC3D,EAAW,IAAe,OAAS,EAAK,MAAM,GAAK,EAAK,MAAM,GACpE,EACI,IAAY,EAAY,GACxB,WAAM,MAAA,4CAA4C,EAAO,UACrD,iCAAgC,EAAY,GAAE,OACtD,EACI,IAAa,EAAY,GACzB,WAAM,MAAA,0CAA0C,EAAQ,UACpD,mCAAkC,EAAY,GAAE,QACxD,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EACZ,EAAc,GAAkC,GAChD,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,EAAK,EAAiB,GAClE,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,GAAI,GAClD,EAC0B,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,gBAAe,GAE1E,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAsB,GAG5B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC/G7D,YAgCI,EAAY,EAAW,GACzB,GAAI,GAAc,MAAQ,IAAe,SACvC,MAAO,GAET,GAAI,IAAe,OACjB,MAAO,GAAI,EAAI,GAAK,IAEtB,KAAM,IAAI,OACN,gDAAgD,EAAU,iBAK5D,EAAc,GAChB,GAAI,GAAM,EACJ,EACF,GAAgC,EAAK,MAAO,EAAa,OAC7D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,mBAIvB,EAAW,EACX,GACF,GAAI,IAAe,SACjB,MAAO,GACF,GAAI,IAAe,OACxB,MAAO,IAAK,GACP,GAAI,IAAe,MACxB,MAAO,IAAI,GACN,GAAI,IAAe,QACxB,MAAO,IAAM,GACR,GAAI,IAAe,QACxB,MAAO,IAAM,EAAG,GAElB,KAAM,IAAI,OAAM,4BAA4B,EAAU,KAIjD,GAAM,IAAa,SAAC,EAAuB,GAChD,GAAM,GAAe,EAAgB,EACrC,MAAO,CAAC,GAAgB,IAAe,UC3EzC,AA8FA,YAAmD,MACjD,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAeA,GAFA,EAAa,GAAc,SAEvB,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,UAC7B,EAAU,EAAgB,EAAQ,SAAU,UAE9C,EAAM,EACN,EAAe,GAEnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,6DACC,GAAI,KAAI,OACnB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,8DACC,GAAQ,KAAI,OACvB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,6DACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,oCAAoC,EAAI,MAAM,GAAE,gBAClD,2BAA0B,EAAQ,MAAM,GAAE,OAClD,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,gBAAe,EAAO,mBAAmB,EAAS,OAC1D,EACI,IAAe,OACf,WAAM,MAAA,sCACF,EAAU,2CAElB,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,GAEnD,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACpB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAGlB,EAAe,GAAqB,EAAI,EAAG,GAEjD,EACI,GAA4B,GAC5B,WAAM,MAAA,oEAEF,uDAAsD,EAAS,OAEvE,GAAM,GACF,GAAoB,EAAI,MAAO,EAAc,EAAS,EAAS,GAC7D,EACF,GAAqB,EAAK,EAAc,EAAQ,MAAO,EAAS,GAC9D,GAAgB,CAAC,EAAM,GAE7B,GAAI,GAAS,MACX,GAAM,IAAU,GAAqB,EAAO,GAC5C,GAAI,KAAK,IAEX,MAAO,KAGH,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,YAAY,CAC9B,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA4B,CAChC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAGpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAa,GAEjB,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAS,EAAG,CAAC,aAAY,KC5QtC,AAyBA,YACI,EAAM,EAAO,EACb,GACF,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEvD,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,yBAAyB,EAAK,EAAM,IAE1C,EAAoD,CAAC,EAAG,EAAK,GAAI,GAEvE,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,IAGN,GAAM,IACT,EAAG,CAAC,qCAAoC,KChD5C,AAyBA,YACI,EAAmE,EACnE,EAAkB,GACpB,GAAI,GAAO,EACP,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAG5D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,wBAAwB,EAAM,EAAQ,IAE5C,EAAmD,CAAC,GAAI,GAExD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,IAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IACT,EAAG,CAAC,oCAAmC,KCnD3C,AAyFA,YAA4D,MAC1D,GAAA,EAAA,EACA,EAAA,EAAA,OACA,EAAA,EAAA,QACA,EAAA,EAAA,IACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,OAAA,EACA,EAAA,EAAA,UAAA,EAAA,IAAA,OAAA,CAAA,EAAA,GAAA,EACA,EAAA,EAAA,gBACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAaA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GACT,EAAG,EAAQ,EAAS,EAAK,EAAY,EAAW,GACpD,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAM,GAAK,EAAgB,EAAG,IAAK,mBAC7B,EAAU,EAAgB,EAAQ,SAAU,mBAE9C,EAAM,EACN,EAAe,GACnB,AAAI,EAAG,OAAS,GACd,GAAe,GACf,EAAM,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAE3D,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACF,SAAQ,EAAI,KAAI,OACxB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,0DACF,iBAAgB,EAAQ,KAAI,OACpC,EACI,EAAI,MAAM,KAAO,EAAQ,MAAM,GAC/B,WAAM,MAAA,4DACF,KAAI,EAAI,MAAM,GAAE,6CAChB,WAAU,EAAQ,MAAM,GAAE,OAClC,AAAI,GAAa,MACf,GAAY,CAAC,EAAG,IAElB,EACI,GAAyC,EAAS,GAClD,WACI,MAAA,oEACA,sBAAqB,EAAO,mBAAmB,EAAS,OAEhE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,+DACF,0BAAyB,EAAe,gBAAgB,EAAG,OAGrE,GAAM,GAAW,GACb,EAAI,MAAO,EAAQ,MAAO,EAAS,EAAW,EAAK,EACnD,IAEA,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAS,SAAU,EAAM,QAGrE,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,0BAG/C,GAAM,GAAO,SAAC,EAAc,GAC1B,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAS,OACd,GAAA,GAAA,EAAA,GAAS,EAAA,EAAA,GAAK,EAAA,EAAA,GAAG,EAAA,EAAA,GAElB,EAAe,GAAqB,EAAI,EAAG,GAE3C,EAAO,GACR,EAAiB,MAAO,EAAc,EAAqB,GAC1D,EAAY,GACd,EAAiB,EAAe,EAAqB,MAAO,GAEhE,GAAI,GAAQ,MACV,GAAM,GAAU,GAAqB,EAAO,GAC5C,MAAO,CAAC,EAAM,EAAW,GAE3B,MAAO,CAAC,EAAM,IAGV,EAA+B,SAAC,GACpC,GAAM,GAAM,EAAQ,qBAAqB,CACvC,MAAO,EACP,OAAQ,EACR,SAAQ,EACR,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAAqC,CACzC,EAAG,EACH,OAAQ,EACR,KAAM,EACN,uBAAwB,GAEpB,EACF,CAAC,QAAO,EAAE,IAAG,EAAE,WAAU,EAAE,UAAS,EAAE,gBAAe,EAAE,WAAU,GAIrE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAkB,GAC3C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,IAEnB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAEpC,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAkB,EAAc,GAC9C,GAAI,GAAM,EAAO,cACb,EAAS,EAAgC,KACzC,GAAsB,GAE1B,SAAK,CAAC,EAAQ,EAAK,EAAK,IAExB,AAAI,GACF,GAAM,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGrD,CAAC,MAAO,EAAK,SAAU,KAGpC,MAAO,GAAiB,EAAK,EAAS,IAGnC,GAAM,IAAkB,EAAG,CAAC,sBAAqB,KClQxD,AAwDA,YAAwC,SACtC,EAAA,EAAA,EACA,EAAA,EAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,GAAA,EACA,EAAA,EAAA,KACA,EAAA,EAAA,WAAA,EAAA,IAAA,OAAA,SAAA,EACA,EAAA,EAAA,uBAUA,GAAI,GAAW,EAAO,MAAM,cAAe,KAAgB,IACzD,GAAI,GAAS,GAAc,EAAG,EAAG,EAAY,GAC7C,MAAI,IAAQ,MACV,GAAS,GAAI,EAAQ,IAGhB,GAAgB,EAAQ,EAAY,GAG7C,GAAI,GAAK,EAAgB,EAAG,IAAK,gBAC7B,EAAK,EAAgB,EAAG,IAAK,gBACjC,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,GAAM,GACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GACtD,EACF,EAAa,EAAG,MAAM,EAAG,KAAO,GAAK,EAAG,MAAM,EAAG,KAAO,GAEtD,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAa,EAAG,MAAM,MAAM,EAAG,IAC/B,EAAY,GAAmB,GAC/B,EAAY,GAAmB,GAErC,EACI,EAAG,MAAQ,GAAK,EAAG,MAAQ,GAAK,EAAG,OAAS,EAAG,KAC/C,WACI,MAAA,qEACA,iBAAgB,EAAG,KAAI,QAAQ,EAAG,KAAI,OAE9C,EACI,GAAiB,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAU,UACrD,GAAU,4BAA4B,EAAG,MAAK,SAC9C,GAAG,MAAK,kBAEnB,EACI,IAAgB,EAChB,WAAM,MAAA,wCAAwC,EAAW,UAClD,GAAW,4BAA4B,EAAG,MAAK,SAC/C,GAAG,MAAK,mBAAmB,GAC9B,oBAAmB,EAAU,kBAErC,GAAM,GAAW,EAAG,MAAM,MAAM,EAAG,IAAI,OAAO,CAAC,EAAa,IAEtD,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IACnC,EAAgB,EAClB,EAAQ,EAAI,CAAC,EAAW,EAAa,IACrC,EAAQ,EAAI,CAAC,EAAW,EAAa,IAErC,EACJ,AAAI,GAAQ,MACV,GAAQ,EAAgB,EAAM,OAAQ,gBACrC,EAAA,GAAA,EAAA,GAAA,GAED,GAA0C,EAAU,EAAM,QAG5D,GAAI,GACJ,AAAI,GAA0B,MAC5B,GAA0B,EACtB,EAAwB,gBAAiB,iBAG/C,GAAM,GAAO,SAAC,EAAc,GACnB,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAAK,GAAA,EAAA,GAAG,GAAA,EAAA,GAId,GACF,GAAqB,EAAQ,EAAI,GAAE,OAAQ,GAAG,GAC9C,GACA,GAgBJ,GAdA,AAAI,CAAC,GAAc,CAAC,EAClB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAM,KACzC,AAAI,CAAC,GAAc,EACxB,IAAO,GAAc,GAAc,EAAK,GAAO,IAC/C,GAAO,GAAc,GAAc,EAAK,GAAM,KACzC,AAAI,GAAc,CAAC,EACxB,IAAO,GAAc,EAAK,GAAc,GAAO,IAC/C,GAAO,GAAc,EAAK,GAAc,GAAO,KAE/C,IAAO,GAAc,EAAK,GAAc,GAAM,IAC9C,GAAO,GAAc,GAAc,EAAK,GAAM,KAG5C,GAAQ,MACV,GAAM,IAAU,GAAqB,GAAO,IAC5C,MAAO,CAAC,GAAM,GAAM,QAEpB,OAAO,CAAC,GAAM,KAIZ,EAA+B,SAAC,GACpC,GAAM,GAAI,EAAQ,iBAAiB,CACjC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EACN,WAAU,EACV,uBAAwB,IAE1B,MAAO,IAGH,EAA6B,CACjC,EAAG,EACH,EAAG,EACH,KAAM,EACN,uBAAwB,GAEpB,EAA2B,CAAC,WAAU,EAAE,WAAU,EAAE,WAAU,GAIpE,GAAI,GAAQ,MACV,GAAM,GACF,GAAW,SAAC,EAAe,EAAe,GACxC,GAAM,GAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,IAET,CAAC,MAAO,EAAQ,EAAK,GAAW,SAAU,KAEvD,MAAO,GAAS,EAAK,QAErB,GAAM,GAAmB,GACrB,SAAC,EAAe,EAAe,EAAe,GAC5C,GAAM,IAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAc,GAElB,SAAK,CAAC,EAAK,EAAK,GAAK,IAEd,CAAC,MAAO,EAAQ,GAAK,GAAW,SAAU,KAGvD,MAAO,GAAiB,EAAK,EAAK,IAI/B,GAAM,IAAS,EAAG,CAAC,aAAY,KClOtC,+DCAA,AAiCA,YAAwB,GACtB,MAAO,IAAa,EAAc,IAAM,KAEnC,GAAM,IAAgB,EAAG,CAAC,eAAc,KCpC/C,AAiCA,YAAqB,GACnB,MAAO,IAAa,EAAc,GAAK,IAGlC,GAAM,IAAa,EAAG,CAAC,YAAW,KCrCzC,AAyCA,YACI,EAAkB,EAAqB,EAAmB,EAC1D,GAD0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,GAGF,OAFI,GAAQ,EACN,EAAmB,GAClB,EAAQ,GAAe,EAAO,MACnC,EAAO,KAAK,GAAM,EAAQ,EAAO,IACjC,GAAS,EAGX,GAAI,EACF,KAAO,EAAQ,EAAO,OACpB,GAAM,GAAU,EAAQ,EAAe,EAAO,KACxC,EAAM,GAAO,CACjB,GAAM,EAAQ,EAAO,EAAc,GAAS,GAAK,CAAC,GAAS,KAE7D,EAAO,KAAK,GACZ,GAAS,EAIb,MAAI,GAAO,SAAW,EACb,GAAS,GAAI,CAAC,EAAG,IAGnB,EAAQ,GAAO,GAAS,CAAC,EAAO,OAAQ,IAE1C,GAAM,IAAQ,EAAG,CAAC,OAAM,KCpE/B,AA4CA,YACI,EAAkB,EAAqB,EACvC,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,AAAI,GAAa,MACf,GAAY,GAAoB,IAKlC,OAHM,GAAe,GAAM,EAAQ,EAAa,GAC1C,EAAiB,EAAI,EAAc,EAAS,IAC5C,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAa,MAAM,GAAI,IACzC,EAAO,KACH,GAAK,GAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,EAAG,IAAe,IAE5D,MAAO,IAAO,GAET,GAAM,IAAO,EAAG,CAAC,MAAK,KC5D7B,AAmDA,YACI,EACA,EACA,EACA,EACA,EACA,GAEF,GAAM,GAAS,EAAgB,EAAO,QAAS,iBACzC,EAAS,EAAgB,EAAO,QAAS,gBAAiB,WAC1D,EAAU,EAAgB,EAAQ,SAAU,gBAAiB,SACnE,EAAS,GAAU,WACnB,EAAqB,GAAsB,EAE3C,GAAM,GAAW,EAAO,MAAM,GAE9B,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OACnC,EACI,EAAO,OAAS,GAAK,EAAO,MAAM,KAAO,EACzC,WAAM,MAAA,oDAAoD,EAAQ,OAC9D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAQ,OAAS,GAAK,EAAQ,MAAM,KAAO,EAC3C,WAAM,MAAA,qDAAqD,EAAQ,KAC/D,kBAAiB,EAAO,MAAK,OACrC,EACI,EAAS,SAAW,EACpB,WAAM,MAAA,iEACF,WAAU,EAAS,OAAM,OACjC,EACI,EAAS,IAAM,GAAK,EAAS,IAAM,EACnC,WAAM,MAAA,2CAA2C,IACrD,EACI,IAAW,YAAc,IAAW,UACpC,WAAM,MAAA,+CAA+C,IAEzD,GAAM,GAAiC,SAAC,GAAY,MAAA,GAAQ,cACxD,EAAQ,EAAQ,EAAS,EAAU,EAAQ,IAEzC,EACoB,CAAC,MAAO,EAAQ,MAAO,EAAQ,OAAQ,GAC3D,EAA4B,CAAC,OAAM,EAAE,mBAAkB,EAAE,SAAQ,GACjE,EAAM,EAAO,cACf,EAAS,EAAgC,KAAiB,GAC1D,GACJ,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KCtG/C,AAiCA,YAAwB,GACtB,GAAM,GAAS,EAAgB,EAAO,QAAS,gBAAiB,WAEhE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,gDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAA8B,CAAC,MAAO,GACtC,EACF,EAAO,UAAU,GAAe,EAAgC,IACpE,MAAO,GAGF,GAAM,IAAgB,EAAG,CAAC,eAAc,KC/C/C,AA4CA,YACI,EAA4B,EAC5B,EACA,GADA,AAAA,IAAA,QAAA,GAAA,GACA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,mBAAoB,WAEnE,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,mDACF,iBAAgB,EAAO,KAAI,OAEnC,GAAM,GAAiC,CAAC,MAAO,GACzC,EAA+B,CAAC,QAAO,EAAE,UAAS,EAAE,OAAM,GAC1D,EAAM,EAAO,UACf,GAAkB,EAClB,GACJ,MAAO,GAGF,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC/DrD,AAoBA,YACI,EAAiB,EAAkB,EACnC,EAAsB,EAAwB,GAMhD,AAAI,GAAgB,MAClB,GAAe,IAEjB,AAAI,GAAkB,MACpB,GAAiB,OAAO,mBAE1B,AAAI,GAAgB,MAClB,GAAe,GAGjB,GAAM,GAAW,EAAM,MAAM,GAC7B,SAAgB,KAAK,IAAI,EAAe,GAExC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAClE,EACI,EAAM,OAAS,EACf,WAAM,MAAA,+CAA+C,EAAM,KAAI,MACnE,EACI,EAAM,MAAM,KAAO,EACnB,WACI,MAAA,oDAAoD,EAAM,MAAM,KACxE,EAAY,EAAO,OAAS,EAAG,WAAM,MAAA,+BACrC,EACI,EAAO,MAAM,KAAO,EACpB,WAAM,MAAA,sDAAsD,EAAQ,KAChE,YAAW,EAAO,MAAM,MAChC,EACI,GAAK,GAAgB,GAAgB,EACrC,WAAM,MAAA,4CAA4C,EAAY,MAC3D,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GC3DnE,AA0BA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBAC1B,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAExB,GAAM,GAAQ,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,GAC1D,MAAO,GAAO,cACV,SAAA,GAAK,MAAA,GAAE,kBACH,EAAQ,EAAS,EAAe,EAAc,IAClD,CAAC,MAAO,EAAQ,OAAQ,GAAU,KAAiB,GACnD,GAGC,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC/CvD,YA+BI,EAAU,EAAY,GACxB,GAAM,GAAQ,GAAa,EAAK,EAAS,GACnC,EAAiB,EAAQ,EAAI,CAAE,GAAQ,GAAK,EAClD,EAAI,OAAO,EAAgB,EAAG,eAmB5B,EAAU,EAAW,GACvB,MAAO,IAAc,EAAK,EAAQ,GAAc,IAUlD,YAA8B,EAAM,GAClC,MAAO,GAAI,EAAI,EAAI,EAAI,EAAI,GAAK,EAGlC,YACI,EAAU,EAAW,GAKvB,OAJI,GAAO,EACP,EAAQ,EAAI,OACZ,EAAS,EACT,EAAQ,GACL,EAAO,IACZ,EAAS,EAAS,GAAQ,IAAU,GACpC,GAAM,GAAgB,EAAW,EAAQ,EAAI,IAC7C,AAAI,EAAgB,EAClB,EAAO,EAAS,EAEhB,GAAQ,EAGR,EAAQ,CAAC,GAIb,MAAO,GAAQ,EAAO,CAAC,EAAO,ECvFhC,YAoCI,EAAmB,EAAoB,EACvC,EAAsB,GACxB,MAAO,IACI,EAAO,EAAQ,EAAe,EAAc,EAC5C,GACC,4BAIV,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAC5C,EAAsB,GACtB,EAA6C,gBAK/C,EAAmB,EAAoB,EACvC,EAAsB,EACtB,GACF,MAAO,IACH,EAAO,EAAQ,EAAe,EAAc,EAAgB,EAC5D,IAGN,YACI,EAAmB,EAAoB,EACvC,EAAsB,EAAwB,EAC9C,EAA4B,EAC5B,GADA,AAAA,IAAA,QAAA,GAAA,IAA4B,AAAA,IAAA,QAAA,GAAA,IAC5B,AAAA,IAAA,QAAA,GAAA,IAKF,OAFM,GAAa,GAEV,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,EAAO,GAAK,GACd,EAAW,KAAK,CAAC,MAAO,EAAO,GAAI,SAAU,EAAG,mBAAoB,IAIxE,EAAW,KAAK,IAShB,OALM,GAAQ,EAAe,EAAK,IAAO,EAAgB,EAEnD,EAA4B,GAC5B,EAA2B,GAE1B,EAAgB,OAAS,GAAiB,EAAW,OAAS,IACnE,GAAM,GAAY,EAAW,MACtB,EAAA,EAAA,MAAsB,EAAA,EAAA,SAAU,EAAA,EAAA,mBAEvC,GAAI,EAAgB,EAClB,MAUF,OADI,GAAkB,GACb,EAAI,EAAgB,OAAS,EAAG,GAAK,EAAoB,EAAE,GAClE,GAAM,GAAM,GAAsB,EAAO,EAAU,EAAgB,IAEnE,GAAI,GAAO,GACT,EAAkB,GAClB,MAMF,GAHA,EAAU,MACN,EAAU,MAAQ,GAAe,EAAc,EAAO,GAEtD,EAAU,OAAS,EACrB,MAWJ,EAAU,mBAAqB,EAAgB,OAE/C,AAAK,GAGH,CAAI,EAAU,QAAU,EACtB,GAAgB,KAAK,GACrB,EAAe,KAAK,EAAU,QACzB,AAAI,EAAU,MAAQ,GAG3B,GAAa,EAAY,EAAW,KAM1C,GAAM,GAAe,EAAgB,OAC/B,EAAa,EAAgB,EAEnC,AAAI,GAAsB,EAAa,GACrC,GAAgB,KAAI,MAApB,EAAwB,GAAI,OAAM,GAAY,KAAK,IACnD,EAAe,KAAI,MAAnB,EAAuB,GAAI,OAAM,GAAY,KAAK,KAGpD,GAAM,GACe,CAAC,gBAAiB,GAAS,EAAiB,UAEjE,MAAI,IACF,GAAO,eAAoB,GAAS,EAAgB,YAGtD,AAAI,GACF,GAAO,aAAkB,GAAO,EAAc,UAGzC,EAGT,YAA+B,EAAmB,EAAW,GAC3D,GAAM,GAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAS,EAAM,SAAS,EAAI,EAAG,EAAI,EAAI,GACvC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAQ,KAAK,IAAI,EAAO,GAAI,EAAO,IACnC,EAAS,GAAQ,GAAU,GAAQ,GACnC,EAAS,GAAQ,GAAU,GAAQ,GACzC,GAAI,GAAS,GAAK,GAAS,EACzB,MAAO,GAET,GAAM,GAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAO,GACnC,EAAmB,KAAK,IAAI,EAAmB,EAAkB,GACnE,KAAK,IAAI,EAAmB,EAAkB,GAClD,MAAO,GAAoB,GAAQ,EAAQ,GAO7C,YAAwB,EAAsB,EAAe,GAC3D,GAAM,GAAS,KAAK,IAAI,EAAQ,EAAM,GACtC,MAAO,IAAO,EAAe,EAAS,EAGxC,YAA6B,EAAe,GAK1C,MAAQ,GAAG,MAAQ,EAAG,OAChB,EAAG,QAAU,EAAG,OAAW,EAAG,SAAW,EAAG,SC7MpD,AA2CA,YACI,EAA4B,EAC5B,EAAuB,EACvB,GADuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,wHACpB,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,GAClD,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eAED,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,GACxD,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAyB,GC1EtC,AAyDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,GACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEtB,GAAM,GAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EACF,CAAC,cAAa,EAAE,aAAY,EAAE,eAAc,EAAE,aAAY,GAExD,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,eAAgB,EAAO,IAGtD,GAAM,IAA6B,EAAG,CAAC,4BAA2B,KChCzE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,wGACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,GACJ,EAAgB,EAAO,cACvB,EAAe,EAAO,aACtB,EAAiB,EAAO,eACxB,EAAe,EAAO,aAEC,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBAA3D,SAAiB,EAAA,OACjB,EAAY,EAAe,GAC3B,EAAa,EAAe,GAK5B,EAAM,GACR,EAAW,EAAY,EAAe,EAAc,EACpD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAAkC,GCxF/C,AAmDA,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,AAAA,IAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAEzB,EAAoC,CAAC,MAAO,EAAQ,OAAQ,GAC5D,EAAkC,CACtC,cAAe,EACf,aAAc,EACd,eAAgB,EAChB,mBAAkB,GAGd,EAAS,EAAO,UACH,GAAqB,EACrB,GAEnB,MAAO,CAAC,gBAAiB,EAAO,GAAI,aAAc,EAAO,IAGpD,GAAM,IAA0B,EAAG,CAAC,yBAAwB,KCnCnE,YACI,EAA4B,EAC5B,EAAuB,EACvB,EACA,GAFuB,MAAA,KAAA,QAAA,GAAA,IACvB,AAAA,IAAA,QAAA,GAAiB,OAAO,mBACxB,AAAA,IAAA,QAAA,GAAA,+GACI,SAAS,EAAgB,EAAO,QAAS,0BACzC,EAAU,EAAgB,EAAQ,SAAU,0BAE5C,EAAS,GACX,EAAQ,EAAS,EAAe,EAAc,EAC9C,MACE,EAAiB,EAAO,cACxB,EAAgB,EAAO,aACvB,EAAkB,EAAO,eAG3B,CAAA,EAAM,QAAQ,IAAI,CAAC,EAAO,OAAQ,EAAQ,iBADxC,SACF,EAAA,OADG,EAAS,EAAA,GAAE,EAAU,EAAA,GAMtB,EAAM,GACR,EAAW,EAAY,EAAgB,EAAe,EACtD,GAEJ,AAAI,IAAW,GACb,EAAO,UAET,AAAI,IAAY,GACd,EAAQ,UAEV,CAAA,EAAO,QAGF,GAAM,IAA+B,GChF5C,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,2DACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WAAM,MAAA,6DACC,GAAI,OAEf,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAG9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GACZ,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,eACX,EAAa,EAAW,EAAU,IAGlC,EAA+B,CAAC,OAAQ,GACxC,EAA6B,CAAC,aAAY,EAAE,KAAI,GAEhD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAgB,GAEpB,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCpFjD,AA2CA,YACI,EAAsB,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IAChD,GAAM,GAAU,EAAgB,EAAQ,SAAU,yBAElD,EACI,EAAQ,OAAS,GAAK,EAAQ,OAAS,EACvC,WAAM,MAAA,kEACF,SAAQ,EAAQ,KAAI,OAC5B,EACI,EAAK,SAAW,EAChB,WACI,MAAA,oEACG,GAAI,OACf,EACI,EAAQ,QAAU,WAAa,EAAQ,QAAU,QACjD,WAAM,MAAA,qDAEV,GAAI,GAAc,EACd,EAAe,GACnB,AAAI,EAAQ,OAAS,GACnB,GAAe,GACf,EAAc,EACV,EAAS,CAAC,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAE9D,GAAA,GAAA,EAAA,GAAW,EAAA,EAAA,GAEZ,EAAsC,CAAC,OAAQ,GAC/C,EAAoC,CAAC,aAAY,EAAE,KAAI,GAEvD,EAAiC,SAAC,EAAS,GAC/C,SAAK,CAAC,IACC,EAAQ,sBACX,EAAa,EAAW,EAAU,IAGlC,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAuB,GAE3B,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAwB,EAAG,CAAC,uBAAsB,KCxF/D,AAyEA,YACI,EAAiB,EAAkB,GACrC,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAClE,EACI,EAAW,IAAM,EACjB,WAAM,MAAA,gDAAgD,EAAQ,MAElE,GAAM,GAAK,EAAgB,EAAG,IAAK,YAEnC,EACI,EAAG,MAAQ,EACX,WAAM,MAAA,4CAA4C,EAAG,KAAI,MAE7D,GAAM,GAAQ,EAAG,MACX,EAAA,EAAA,MAAA,MAAA,IAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,kDAAiD,EAAC,OAExD,GAAI,CAAE,IAAY,GAChB,KAAM,IAAI,OACN,yBAAyB,EAAQ,IACjC,qDAAoD,EAAC,OAG3D,AAAI,EAAW,GACb,GAAW,GAEb,AAAI,EAAW,GACb,GAAW,GAGb,GAAM,GAAI,EAAQ,GAAM,EAAG,EAAG,EAAG,SAAU,CAAC,GAAI,IAC1C,EAAI,GAAM,EAAG,EAAG,EAAG,SACnB,EAAK,GAAI,EAAG,GAEZ,EAAS,GACX,GAAU,EAAI,GAAO,CAAC,EAAU,UAChC,GAAa,EAAI,GAAO,CAAC,EAAU,WAEjC,EAAO,GAAM,CAAC,EAAG,GAAI,EAAG,OAE9B,MAAO,GACI,GAAM,GAAQ,EAAQ,EAAI,CAAC,GAAI,EAAG,KACvB,IAAI,SAAA,GAAO,MAAA,IAAM,EAAQ,EAAK,MACzC,GAGN,GAAM,IAAW,EAAG,CAAC,UAAS,KC7HrC,AA2DA,YAAsB,GACpB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,EAAkB,GAClB,EACI,GAAM,MAAQ,EAAG,OAAS,EAC1B,WAAM,MAAA,sEAGV,OADM,GAAM,EAAG,GAAG,MAAM,cACf,GACP,EACI,EAAG,GAAG,MAAM,KAAO,EACnB,WACI,MAAA,gEACA,KAAK,EAAkB,GAAG,MAAM,GAAE,QAAQ,EAAG,QAL9C,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,OAQT,GAAkB,GAClB,EAAK,GAAM,EAAI,EAAG,MAAM,GAAI,GAAG,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,CAAC,MAGtD,EACI,EAAG,QAAU,EAAG,GAAG,MAAM,GACzB,WAAM,MAAA,oCACK,EAAkB,OAAM,aAC/B,0BAA0B,EAAkB,GAAG,MAAM,GAAE,QAI/D,OAFM,GAAiB,GACjB,EAAO,aACJ,GACP,EAAG,KAAK,EAAO,KAAK,WAClB,GAAI,GAAI,EAAK,GACb,GAAI,EAAI,EACN,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAO,EAAI,GAAI,EAAI,EAAG,GAAI,IAAK,EAAG,IACxC,EAAI,GAAI,EAAG,GAGf,MAAO,IAAI,EAAG,GAAK,EAAG,kBATjB,EAAI,EAAG,EAAI,EAAG,OAAQ,EAAE,IAAxB,GAaT,MAAI,GACK,GAAM,EAAI,GAEV,EAIJ,GAAM,IAAc,EAAG,CAAC,aAAY,KC5G3C,AAmFA,YAAa,EAAW,GAMtB,GANsB,AAAA,IAAA,QAAA,GAAA,IACtB,EACI,EAAE,MAAQ,EACV,WAAM,MAAA,gEACF,EAAE,OAEN,EAAE,OAAS,EACb,MAAO,IAAK,EAAe,GAM3B,GAAM,GAAgB,EAAE,MAAM,MAAM,EAAG,EAAE,MAAM,OAAS,GAC7B,OAAO,SAAC,EAAO,GAAS,MAAA,GAAQ,IACrD,EAAO,GACT,EACI,EACA,CACE,EAAe,EAAE,MAAM,EAAE,MAAM,OAAS,GACxC,EAAE,MAAM,EAAE,MAAM,OAAS,KAE/B,GACE,EAAmB,GACnB,EAAmB,GACzB,EAAK,QAAQ,SAAA,GACL,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACZ,EAAK,KAAK,GACV,EAAK,KAAK,KAEZ,GAAM,GAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OAC9B,EAAI,EAAQ,GAAM,EAAM,GAAI,EAAE,OACpC,MAAO,CAAC,EAAG,GAIf,YAAc,EAAa,GAAA,MAAA,KAAA,QAAA,GAAA,IAClB,EAAO,KAAK,WACjB,EACI,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,0CACF,EAAE,MAAM,OAAM,cAYtB,OAVM,GAAI,EAAE,MAAM,GACZ,EAAI,EAAE,MAAM,GAEd,EAAI,GAAI,GACR,EAAI,GAAM,GAER,EAAQ,GAAS,CAAC,CAAC,IAAK,CAAC,EAAG,IAC9B,EAAc,GAAM,GAElB,EAAQ,GAAK,EAAI,EAAI,aAClB,SAGD,EAAQ,EACR,EAAQ,EACR,EAAQ,EACd,EAAA,EAAA,KAAA,wfAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GA8CP,GAAQ,CAAC,EAAO,EAAO,KApDhB,EAAI,EAAG,EAAI,EAAO,EAAE,IAApB,GAuDT,MAAI,CAAC,GAAgB,EAAI,GACvB,GAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,IACzB,EAAI,GAAM,EAAG,CAAC,EAAG,GAAI,CAAC,EAAG,KAGpB,CAAC,EAAG,KAIR,GAAM,IAAK,EAAG,CAAC,IAAG,KCxMzB,AAiBA,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,IAAA,GAAA,MACA,EAAA,EAAA,uBAAA,GAAA,2BAJU,EAAA,WAAA,GAAA,UAAS,KCyBrB,YACI,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GAAM,GAAgB,GAAY,KAAQ,EAAU,EAAI,EAAS,GAEjE,GAAI,IAAc,EAAA,UAAU,KAC1B,MAAO,GAET,GAAI,IAAc,EAAA,UAAU,IAC1B,MAAO,IAAI,GAEb,GAAI,IAAc,EAAA,UAAU,MAC1B,GAAI,GAAY,KACd,MAAO,IAAK,GAEZ,GAAM,GAAkB,EAAQ,KAAO,EAAS,KAC1C,EAAS,GAAI,GAAI,GAAe,GAAI,IAC1C,MAAO,GAAkB,EAAI,GAAI,EAAQ,GAAO,IACnB,EAGjC,GAAI,IAAc,EAAA,UAAU,wBAC1B,GAAI,GAAY,KACd,MAAO,IAAI,GAAI,GAAe,GAAO,EAAQ,OAE7C,GAAM,GAAqB,EAAI,EAAU,GAAK,EAAQ,QAEhD,EACF,GAAK,GAAI,GAAS,EAAoB,GAAO,KAAM,WACvD,MAAO,IAAI,GAAI,GAAe,GAIlC,KAAM,OAAM,sBAAsB,GAE7B,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCnF3D,AA2CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,sBAC5C,EACF,EAAgB,EAAa,cAAe,sBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,uBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,iCAEvC,GAAM,GAAS,GAAI,GAAI,EAAS,IAChC,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAqB,EAAG,CAAC,oBAAmB,KChBzD,YACI,EAAsB,EAA2B,EACjD,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,kBAC5C,EACF,EAAgB,EAAa,cAAe,kBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,mBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,6BAEvC,GAAM,GAAM,GAAO,GACb,EAAS,GAAI,EAAK,GAAI,EAAI,EAAS,GAAe,EAAM,KAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAiB,EAAG,CAAC,gBAAe,KCnBjD,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAU,EAAgB,EAAQ,SAAU,aAC1C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAM,GAAO,GAEnB,EAAU,GAAI,EAAI,GAAO,GAAI,GAAU,GACvC,GAAM,GAAS,GAAK,GAAI,EAAK,EAAI,EAAS,KAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KC9DvC,AAiDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,aAC5C,EAAe,EAAgB,EAAa,cAAe,aAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,cAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,wBAErD,GAAM,GAAc,GAAO,GACrB,EAAQ,GAAI,GAAI,EAAc,IAC9B,EAAY,GAAQ,EAAO,GAC3B,EAAS,GAAI,EAAO,GAEpB,EACF,GAAI,EAAI,GAAO,IAAM,GAAO,IAAa,EAAI,EAAa,IAC9D,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAY,EAAG,CAAC,WAAU,KCtEvC,AAgDA,YACI,EAAsB,EACtB,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,MAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,WAC5C,EAAe,EAAgB,EAAa,cAAe,WAC7D,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,YAEjD,GAAkB,EAAQ,MAAO,EAAa,MAAO,sBAErD,GAAM,GAAM,GAAO,GACb,EAAgB,GAAO,GAEvB,EAAK,GAAI,EAAI,EAAS,GAAI,GAAI,EAAc,MAC5C,EACF,EAAI,GAAI,EAAK,GAAU,GAAI,GAAI,GAAI,EAAK,GAAe,KACrD,EAAS,GAAI,EAAI,GACvB,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAU,EAAG,CAAC,SAAQ,KCrEnC,AA0CA,YACI,EAAsB,EACtB,EACA,GAAA,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAM,GAAU,EAAgB,EAAQ,SAAU,oBAC5C,EACF,EAAgB,EAAa,cAAe,oBAC5C,EAAmB,KACvB,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,qBAEjD,GACI,EAAQ,MAAO,EAAa,MAAO,+BAEvC,GAAM,GAAS,GAAkB,EAAS,GAC1C,MAAO,IAAoB,EAAQ,EAAU,GAExC,GAAM,IAAmB,EAAG,CAAC,kBAAiB,KC3DrD,AAmCA,YACI,EAAsB,GACxB,GAAM,GACF,EAAgB,EAAQ,SAAU,iCAChC,EACF,EAAgB,EAAQ,SAAU,iCACtC,GACI,EAAQ,MAAO,EAAQ,MAAO,4CAsBlC,GAAM,GAAY,GAAK,GACjB,EAAgB,EAAI,EAAS,GAC7B,EAAgB,GAAM,GAAI,GAAI,GAAI,MAExC,MAAO,IAAI,GAAI,EAAW,GAAgB,GAwB5C,YACI,EAAgC,EAChC,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GAAoB,EACpB,EAAkB,mBAAoB,uBACpC,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KAOvB,GANA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAEjD,GACI,EAAkB,MAAO,EAAQ,MAAO,kCAExC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAO,GAAO,IAEpB,EACI,GAAI,EAAI,EAAmB,GAAI,EAAK,IAChC,EAAI,EAAM,IAEpB,GAAM,GAAS,GAA+B,EAAmB,GAEjE,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KCxH3D,AA+DA,YACI,EAAW,EAAW,GAKxB,GALwB,AAAA,IAAA,QAAA,GAAA,IACxB,AAAI,IAAQ,IACV,GAAM,EAAO,KAAO,GAGlB,IAAQ,EAAO,KAAO,EACxB,KAAM,OACF,+DACA,wCAAuC,EAAO,KAAI,KAClD,gBAAe,IAGrB,GAAM,GACF,GAAW,SAAC,EAAgB,EAAgB,GAI1C,GAAM,GAAW,GACX,EAAM,GAAU,EAAQ,CAAC,GAAM,GAC/B,EAAY,GAAI,GAAK,EAAQ,WAAY,GAC/C,EAAK,CAAC,EAAQ,IAEd,GAAM,GAAa,GAAI,EAAI,EAAW,IAChC,EAAW,GAAI,EAAY,CAAC,IAE5B,EAAW,SAAC,EAAO,GAChB,GAAA,GAAA,EAAA,GAAQ,EAAA,EAAA,GACT,EAAU,GAAqB,EAAG,MAAO,CAAC,IAChD,MAAO,CACL,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAK,EAAQ,WAAY,GAAI,KACrC,EAAI,EAAQ,EAAI,GACZ,GAAI,GAAI,GAAY,GAAK,EAAQ,eAGzC,MAAO,CAAC,MAAK,EAAE,SAAQ,KAG7B,MAAO,GAAS,EAAQ,GAsB1B,YACI,EAA4B,EAC5B,EAA6B,EAC7B,GAD6B,AAAA,IAAA,QAAA,GAAA,GAC7B,AAAA,IAAA,QAAA,GAAY,EAAA,UAAU,wBACxB,GAAI,GACA,EAAgB,EAAc,eAAgB,uBAC5C,EAAU,EAAgB,EAAQ,SAAU,uBAC9C,EAAmB,KASvB,GAPA,AAAI,GAAW,MACb,GAAW,EAAgB,EAAS,UAAW,wBAGjD,GACI,EAAc,MAAO,EAAQ,MAAO,kCAEpC,EAAiB,GACnB,GAAM,GAAuB,GAAO,GAC9B,EAAM,GAAO,GACb,EAAa,GAAO,EAAc,MAAM,IAE9C,EACI,GAAI,EAAI,EAAe,GAAI,EAAK,IAC5B,GAAI,EAAsB,IAGpC,GAAM,GAAS,GAA+B,EAAe,GAE7D,MAAO,IAAoB,EAAQ,EAAU,GAGxC,GAAM,IAAsB,EAAG,CAAC,qBAAoB,KC3J3D,GAuNM,IAAW,CACf,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,MAAK,IASD,GAAS,CACb,cAAa,GACb,WAAU,GACV,MAAK,GACL,KAAI,IAeA,GAAQ,CACZ,cAAa,GACb,sBAAqB,GACrB,eAAc,GACd,iBAAgB,GAChB,cAAa,GACb,kBAAiB,GACjB,uBAAsB,GACtB,2BAA0B,GAC1B,gCAA+B,GAC/B,wBAAuB,GACvB,6BAA4B,IAOxB,GAAS,CACb,SAAQ,GACR,YAAW,GACX,GAAE,IAaE,GAAS,CACb,mBAAkB,GAClB,oBAAmB,GACnB,eAAc,GACd,UAAS,GACT,UAAS,GACT,QAAO,GACP,iBAAgB,GAChB,oBAAmB,GACnB,oBAAmB,IClSrB,mBAyCwC,GAAA,EAAA,GAAxC,4DAiBE,SAAA,UAAA,SAAA,SAAS,EAAiB,EAAoB,GAApB,AAAA,IAAA,QAAA,GAAA,IAElB,GAAA,GAAA,KAAA,iBAAA,EAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAEd,GAAI,GAAW,MACb,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,KAAM,OAAQ,EAAM,EAAE,SACrD,KAAK,eAAe,OAEpB,MAAK,eAAe,GAMtB,MAFA,IAAQ,GAEJ,EACK,EAEP,GAAM,UACC,OAOX,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAEd,KAAK,6CAGJ,EAAA,UAAA,oBAAV,WACE,KAAK,YAAc,KAAK,WAAa,GAgBvC,EAAA,UAAA,iBAAA,SAAiB,EAAiB,GAEhC,MAAO,IAAc,EAAG,IAgB1B,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,aAAe,MACtB,GAAQ,KAAK,cAIX,EAAA,UAAA,eAAN,8EACE,MAAI,MAAK,aAAe,MACtB,MAAK,YAAc,GAErB,CAAA,EAAO,CACL,KAAM,OAEN,OAAQ,GAAO,KAAK,YAAa,gBAI/B,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,gEAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OACN,4DACA,IAAG,KAAK,sBAUE,EAAA,UAAA,kBAAhB,SAAkC,mGAEhC,SAAA,KAAoB,CAAA,EAAM,EAAa,GAAG,OAAO,eAAjD,SAAK,YAAe,EAAA,OAAqC,GACzD,CAAA,EAAO,EAAa,MAAM,UAE9B,GA5HwC,IA8HxC,OAAO,eAAe,GAAW,OAAO,YAAa,CACnD,MAAO,SAAC,GACN,MAAO,GAAS,UAAY,MAAQ,EAAS,kBAAoB,MAC7D,EAAS,gBAAkB,QC1KnC,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAAgC,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,MAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,IAAA,EAChC,EAAA,QAAA,EALN,EAAA,iBAAwC,GACxC,EAAA,mBAA0C,GAOhD,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,cACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,aACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAC3C,EAAoB,EAAK,mBAAmB,GAAG,SAErD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAiB,EAAK,KAC1B,EAAI,GAAO,GAAW,EAAI,EAAK,MAEjC,EACF,EAAI,GAAI,GAAK,GAAI,EAAmB,EAAK,UACjC,GAAK,GAAI,EAAiB,EAAK,WACnC,GAEF,EACF,GAAI,EAAI,EAAmB,EAAK,KAC5B,EAAI,GAAO,GAAU,EAAI,EAAK,MAEtC,EAAgB,OAAO,GACvB,EAAkB,OAAO,GAEzB,GAAM,GAAW,GAAI,EAAI,EAAS,CAAC,EAAK,cAAe,GACvD,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,oBAAsB,MAC7B,IAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,YACzC,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,cAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,iBAAgB,OAAK,KAAK,oBAC/B,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,iBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,IAAO,KAAK,IACZ,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,IAAQ,EAAO,UAjHxD,EAAA,UAAY,WAmHrB,GArHuC,IAsHvC,GAAc,ICrJd,mBA+BsC,GAAA,EAAA,GAMpC,WACc,EAA8B,GAAA,AAAA,IAAA,QAAA,GAAA,IAD5C,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,aAAA,EAA8B,EAAA,wBAAA,EAHpC,EAAA,iBAAwC,KAOhD,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,iBAAiB,IAAM,MAC9B,GAAM,GAAY,GAClB,EAAK,iBAAiB,GAAK,CACzB,aAAiB,EAAI,eACrB,SAAU,GACN,WAAM,MAAA,IAAK,EAAM,MAAO,EAAK,yBAClB,SAAS,MAI5B,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAkB,EAAK,iBAAiB,GAAG,SAEjD,GAAK,WACH,GAAM,GAAqB,GAAI,EAAiB,GAAO,IACvD,EAAgB,OAAO,GAEvB,GAAM,GAAW,GACb,EAAI,GAAI,EACA,GAAK,GAAI,EAAoB,EAAO,QAAQ,aAChD,CAAC,EAAK,cACV,GACJ,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,kBAAoB,MAC3B,GAAQ,KAAK,iBAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIvC,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,iBAAiB,IAC9D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,iBAAmB,EAAa,IACjC,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,wBAA2B,KAAK,0BAK7B,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,aAAiB,EAAO,0BA/EzC,EAAA,UAAY,UAiFrB,GAnFsC,IAoFtC,GAAc,ICnHd,mBAkCmC,GAAA,EAAA,GASjC,WACc,EAAgC,EAChC,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MAFvC,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EAL/B,EAAA,uBAA8C,GAC9C,EAAA,wBAA+C,GAMrD,GAAK,WAEH,EAAK,SAAW,GAAO,GAAO,WAC9B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EAAmB,GAAI,EAAG,EAAK,UAErC,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,wBAAwB,IAAM,MACrC,GAAK,wBAAwB,GAAK,CAChC,aAAiB,EAAI,KACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAe,EAAK,wBAAwB,GAAG,SAE/C,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QACvD,EACF,GAAI,EAAI,EAAc,EAAK,OACvB,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EAA2B,GAAI,EAAgB,GAC/C,EACF,GAAI,EAAiB,GAEzB,EAAY,OAAO,GACnB,EAAa,OAAO,GAEpB,GAAM,GACF,GAAI,EAAI,GAAI,EACA,GAAI,GAAK,GAA4B,EAAK,UAC9C,CAAC,EAAK,cACV,GACR,EAAM,OAAO,KAGf,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,QAC7C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,SAAS,UAEd,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,yBAA2B,MAClC,GAAQ,KAAK,wBAAwB,IAAI,SAAA,GAAK,MAAA,GAAE,aAI9C,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,yBACrC,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,4GACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACf,GAAK,WACH,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,IACxD,EAAK,SAAS,OAAO,GAAI,EAAK,MAAO,EAAK,YAAc,MAGpD,EAAgB,EAAa,OAAS,EACtC,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,wBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,gBAI7C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,UAKb,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,UA5IN,EAAA,UAAY,OA8IrB,GAhJmC,IAiJnC,GAAc,ICnLb,mBAiCoC,GAAA,EAAA,GASnC,WACc,EAAgC,EAChC,EAAyB,EACzB,GADyB,AAAA,IAAA,QAAA,GAAA,MACzB,AAAA,IAAA,QAAA,GAAA,GAHd,GAAA,GAIE,EAAA,KAAA,OAAO,KAHK,SAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,MAAA,EAAyB,EAAA,QAAA,EACzB,EAAA,MAAA,EANN,EAAA,uBAA8C,GAC9C,EAAA,2BAAkD,GAQxD,GAAK,WACH,EAAK,UAAY,GAAO,GAAG,WAC3B,EAAK,SAAW,GAAO,GAAO,aAGhC,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,aAIlC,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,GAAK,WACH,GAAM,GAAmB,GAAI,EAAG,EAAK,UAC/B,EACF,GAAI,CAAC,EAAK,aAAc,GAAI,EAAI,EAAK,UAAW,EAAK,OAAQ,IAEjE,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAGxC,AAAI,EAAK,2BAA2B,IAAM,MACxC,GAAK,2BAA2B,GAAK,CACnC,aAAiB,EAAI,KACrB,SAAU,GAAU,GAAO,SAAS,KAIxC,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAc,EAAK,uBAAuB,GAAG,SAC7C,EAAkB,EAAK,2BAA2B,GAAG,SAErD,EACF,GAAI,EAAI,EAAa,EAAK,OAAQ,EAAI,EAAU,EAAI,EAAK,QAEvD,EAAM,EAAI,EAAiB,EAAK,OAChC,EAAM,GAAI,GAEV,EAAqB,GAAQ,EAAK,GAExC,EAAY,OAAO,GACnB,EAAgB,OAAO,GAEvB,GAAM,GACF,GAAI,EAAI,GAAI,EAAI,GACR,GAAI,EAAgB,GAAI,EAAoB,EAAK,WACrD,GAER,EAAM,OAAO,KAGf,EAAK,UAAU,OAAO,GAAI,EAAK,UAAW,IAC1C,EAAK,SAAS,OAAO,EAAI,EAAK,SAAU,EAAK,UAE/C,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UACd,KAAK,UAAU,UAEf,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,4BAA8B,MACrC,GAAQ,KAAK,2BAA2B,IAAI,SAAA,GAAK,MAAA,GAAE,aAIjD,EAAA,UAAA,WAAN,8EACE,KAAM,IAAI,OAAM,wDAGZ,EAAA,UAAA,WAAN,SAAiB,sEACf,KAAM,IAAI,OAAM,wDAGlB,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,MAAS,KAAK,MACd,QAAW,KAAK,QAChB,MAAS,KAAK,QAKX,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,MAChD,EAAO,QAAY,EAAO,QAvHzB,EAAA,UAAY,SAyHrB,GA3HqC,IA4HrC,GAAc,IC7Jd,mBA6BkC,GAAA,EAAA,GAKhC,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,aAAA,EAEpB,EAAK,gBAAgB,KAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAW,MAAM,QAAQ,GAC3B,EAAkB,IAAI,SAAA,GAAK,MAAA,GAAE,OAC7B,OAAO,KAAK,GAChB,EAAS,QAAQ,SAAC,EAAM,GACtB,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAEF,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAK,WACH,GAAM,GAAW,GAAI,EAAI,EAAK,EAAG,GAAW,GAC5C,EAAM,OAAO,OAGjB,KAAK,uBAMP,EAAA,UAAA,gBAAA,SAAgB,GACd,KAAK,aAAe,EACpB,AAAI,KAAK,GAAK,MACZ,KAAK,EAAE,UAET,KAAK,EAAI,GAAK,GAAO,CAAC,KAGxB,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,WAGH,EAAA,UAAA,WAAN,qGACU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAA,CAAQ,EAAA,eAGJ,EAAA,UAAA,WAAN,SAAiB,6FACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAC5C,GADA,EAAe,EAAA,OACX,EAAa,SAAW,EAC1B,KAAM,IAAI,OAAM,iEAIpB,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,aAAgB,KAAK,eAIxB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GAAI,EAAO,eA7DjB,EAAA,UAAY,MA+DrB,GAjEkC,IAkElC,GAAc,IC/Fd,mBA+BuC,GAAA,EAAA,GAMrC,WACc,EAA8B,EAChC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFZ,GAAA,GAGE,EAAA,KAAA,KAAM,IAAa,KAFP,SAAA,aAAA,EAA8B,EAAA,SAAA,EAChC,EAAA,YAAA,EAJJ,EAAA,cAAqC,GAM3C,EAAK,EAAI,GAAO,EAAK,YAGvB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACzC,GAAI,EAAK,cAAc,IAAM,MAC3B,GAAM,GAAY,GAClB,EAAK,cAAc,GAAK,CACtB,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,MAInD,GAAM,GAAe,EAAK,cAAc,GAAG,SACrC,EAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAK,WACH,GAAI,GACE,EAAkB,GAAI,EAAI,EAAK,EAAG,GAAe,GACvD,AAAI,EAAK,YACP,EAAW,GACP,EAAI,EAAK,EAAG,GAAI,EAAU,EAAI,EAAiB,EAAK,KAAM,GAE9D,EAAW,GAAI,EAAI,EAAK,EAAG,GAAkB,GAE/C,EAAa,OAAO,GACpB,EAAM,OAAO,OAGjB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,KAAK,EAAE,UACP,AAAI,KAAK,eAAiB,MACxB,GAAQ,KAAK,cAAc,IAAI,SAAA,GAAK,MAAA,GAAE,aAS1C,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,SAAW,GAGZ,EAAA,UAAA,WAAN,qGAEU,MAAA,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OAAO,KAAK,cAAc,IAC3D,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGvC,EAAA,UAAA,WAAN,SAAiB,mGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EAAY,GAClB,KAAK,cAAgB,EAAa,IAC9B,SAAA,GAAK,MAAC,CAAC,aAAc,EAAE,KAAM,SAAU,EAAE,OAAO,SAAS,gBAG/D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,SAAY,KAAK,SACjB,YAAe,KAAK,cAKjB,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,SAAa,EAAO,cA3FlD,EAAA,UAAY,WA6FrB,GA/FuC,IAgGvC,GAAc,IC/Hd,mBAgCsC,GAAA,EAAA,GASpC,WACc,EAAgC,EAChC,EAA0B,EACpC,GAF0C,AAAA,IAAA,QAAA,GAAA,IAChC,AAAA,IAAA,QAAA,GAAA,GAA0B,AAAA,IAAA,QAAA,GAAA,MACpC,AAAA,IAAA,QAAA,GAAA,IAHJ,GAAA,GAIE,EAAA,KAAA,OAAO,KAOP,GAVY,EAAA,aAAA,EAAgC,EAAA,MAAA,EAChC,EAAA,SAAA,EAA0B,EAAA,QAAA,EANhC,EAAA,uBAA8C,GAC9C,EAAA,mBAA0C,GAC1C,EAAA,qBAA4C,GAQlD,EAAK,SAAW,EAEhB,AAAI,GAAW,MACb,GAAK,QAAU,EAAO,QAAQ,WAE5B,GAAgB,KAClB,KAAM,IAAI,OAAM,+DAIpB,SAAA,UAAA,eAAA,SAAe,GAAf,GAAA,GAAA,KACQ,EAAgB,MAAM,QAAQ,GAChC,EAAkB,IAAI,SAAA,GAAQ,MAAA,GAAK,OACnC,OAAO,KAAK,GAEhB,EAAc,QAAQ,SAAC,EAAM,GAC3B,GAAM,GAAQ,EAAO,oBAAoB,GACnC,EAAY,GAClB,AAAI,EAAK,uBAAuB,IAAM,MACpC,GAAK,uBAAuB,GAAK,CAC/B,aAAiB,EAAI,OACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,mBAAmB,IAAM,MAChC,GAAK,mBAAmB,GAAK,CAC3B,aAAiB,EAAI,YACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAGnD,AAAI,EAAK,qBAAqB,IAAM,MAAQ,EAAK,UAC/C,GAAK,qBAAqB,GAAK,CAC7B,aAAiB,EAAI,MACrB,SAAU,GAAK,WAAM,MAAA,IAAU,GAAO,SAAS,OAInD,GAAM,GAAW,MAAM,QAAQ,GAC3B,EAAkB,GAAG,OACrB,EAAkB,GACtB,GAAI,GAAY,KACd,OAGF,GAAM,GAAwB,EAAK,uBAAuB,GAAG,SACvD,EAAqB,EAAK,mBAAmB,GAAG,SACtD,GAAK,WACH,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEvC,GAAI,EAAK,UACP,GAAM,GAAsB,EAAK,qBAAqB,GAAG,SAEnD,EACF,GAAI,EAAI,EAAqB,EAAK,OAC9B,EAAI,EAAU,EAAI,EAAK,QAEzB,EACF,GAAI,EAAI,EAAU,EAAK,cACnB,GACI,GAAI,EACA,GAAI,GAAO,GAAyB,EAAK,YACnD,EACF,GAAI,EAAI,EAAoB,EAAK,UAAW,GAEhD,EAAsB,OAAO,GAC7B,EAAoB,OAAO,GAC3B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAGb,GAAM,GACF,GAAI,EAAI,EAAuB,EAAK,OAChC,EAAI,GAAO,GAAW,EAAI,EAAK,QAEjC,EACF,GAAI,EAAI,EAAoB,EAAK,UAC7B,GAAI,EAAI,EAAU,EAAK,cACnB,GAAK,GAAI,EAA0B,EAAK,YAEpD,EAAsB,OAAO,GAC7B,EAAmB,OAAO,GAE1B,GAAM,GAAW,GAAI,EAAO,GAC5B,EAAM,OAAO,QAInB,KAAK,uBAGP,EAAA,UAAA,QAAA,WACE,AAAI,KAAK,wBAA0B,MACjC,GAAQ,KAAK,uBAAuB,IAAI,SAAA,GAAK,MAAA,GAAE,YAEjD,AAAI,KAAK,sBAAwB,MAAQ,KAAK,UAC5C,GAAQ,KAAK,qBAAqB,IAAI,SAAA,GAAK,MAAA,GAAE,YAE/C,AAAI,KAAK,oBAAsB,MAC7B,GAAQ,KAAK,mBAAmB,IAAI,SAAA,GAAK,MAAA,GAAE,aAIzC,EAAA,UAAA,WAAN,2GAEQ,SACE,KAAK,uBAAsB,OAAK,KAAK,oBAC7C,AAAI,KAAK,UACP,EAAU,KAAI,MAAd,EAAkB,KAAK,sBAEjB,CAAA,EAAM,KAAK,yBAAnB,MAAA,CAAA,EAAO,CAAC,EAAA,QAA6B,OACjC,EAAU,IAAI,SAAA,GAAK,MAAC,CAAC,KAAM,EAAE,aAAc,OAAQ,EAAE,oBAGrD,EAAA,UAAA,WAAN,SAAiB,qGACA,MAAA,CAAA,EAAM,KAAK,kBAAkB,WAA5C,SAAe,EAAA,OACT,EACF,KAAK,SAAW,EAAa,OAAS,EAAI,EAAa,OAAS,EAC9D,EAAY,GAClB,KAAK,uBACD,EAAa,MAAM,EAAG,GAAe,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SACf,MAEnD,KAAK,mBACD,EAAa,MAAM,EAAe,EAAgB,GAC7C,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,MAE3C,AAAI,KAAK,UACP,MAAK,qBACD,EAAa,MAAM,EAAgB,EAAG,EAAgB,GACjD,IAAI,SAAA,GAAK,MAAC,CACJ,aAAc,EAAE,KAChB,SAAU,EAAE,OAAO,SAAS,iBAK/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,aAAgB,KAAK,aACrB,MAAS,KAAK,MACd,SAAY,KAAK,SACjB,QAAW,KAAK,QAChB,SAAY,KAAK,WAKd,EAAA,WAAP,SACI,EAAiC,GACnC,MAAO,IAAI,GACP,EAAO,aAAiB,EAAO,MAAU,EAAO,SAChD,EAAO,QAAY,EAAO,WA5KzB,EAAA,UAAY,UA8KrB,GAhLsC,IAiLtC,GAAc,ICjNd,AAyBA,GAAA,IAAA,WAAA,cAsCS,SAAA,IAAP,SAAW,GACT,MAAO,IAAI,IAAa,IAkBnB,EAAA,SAAP,SAAgB,EAAsB,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE/C,GAAI,IAAkB,EAAc,EAAU,IAuBhD,EAAA,QAAP,SACI,EAAsB,EAAY,EAAgB,EAClD,GADsB,MAAA,KAAA,QAAA,GAAA,IAAY,AAAA,IAAA,QAAA,GAAA,GAAgB,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,IACK,GAAI,IACP,EAAc,EAAO,EAAU,EAAS,IAevC,EAAA,KAAP,SACI,EAAsB,EAAa,EACnC,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MACnC,AAAA,IAAA,QAAA,GAAA,MACK,GAAI,IAAc,EAAc,EAAO,EAAO,IAehD,EAAA,SAAP,SAAgB,EAAqB,EAAW,GAAhC,MAAA,KAAA,QAAA,GAAA,MAAqB,AAAA,IAAA,QAAA,GAAA,KAAW,AAAA,IAAA,QAAA,GAAA,MAEvC,GAAI,IAAkB,EAAc,EAAK,IAgB3C,EAAA,OAAP,SACI,EAAsB,EAAa,EAAe,EAClD,GADA,MAAA,KAAA,QAAA,GAAA,MAAsB,AAAA,IAAA,QAAA,GAAA,IAAa,AAAA,IAAA,QAAA,GAAA,MAAe,AAAA,IAAA,QAAA,GAAA,MAClD,AAAA,IAAA,QAAA,GAAA,GACK,GAAI,IAAgB,EAAc,EAAO,EAAO,EAAS,IAmB3D,EAAA,QAAP,SAAe,EAAsB,GAAA,MAAA,KAAA,QAAA,GAAA,IAE5B,GAAI,IAAiB,EAAc,IAE9C,KC5LA,GA+Ba,IAAQ,CACnB,IAAK,GAAsB,IAC3B,SAAU,GAAsB,SAChC,SAAU,GAAsB,SAChC,QAAS,GAAsB,QAC/B,QAAS,GAAsB,QAC/B,OAAQ,GAAsB,OAC9B,KAAM,GAAsB,MCtC9B,AAiBA,GAAM,IAA2B,WAC/B,MAAI,OAAO,wBAA0B,YAC5B,sBACE,MAAO,eAAiB,YAC1B,aAEF,SAAC,GAAgB,MAAA,SAa1B,cACE,MAAO,IAAI,SAAc,SAAA,GAAW,MAAA,IAAc,WAAM,MAAA,SCrC1D,YAmBI,EAAiC,EACjC,GACF,GAAM,GACF,EAAc,OAAO,IAAW,SAAW,EAAS,EAAO,IACzD,EACF,EAAe,OAAO,IAAW,SAAW,EAAS,EAAO,IAChE,MAAO,CAAC,EAAS,GCzBnB,YA0BI,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAqB,GACzB,GAAI,EACF,EAAW,EAAS,OAAO,EAAW,MAAM,IAC5C,EAAS,KAAK,EAAW,GAAK,GAC9B,EAAW,EAAS,OAAO,EAAW,MAAM,SAE5C,EAAW,EAAS,OAAO,EAAW,IAEtC,OADM,GAAgB,EAAW,OACxB,EAAI,EAAG,EAAI,EAAe,EAAE,EACnC,EACI,EAAS,OAAO,CAAC,EAAW,EAAI,GAAK,EAAW,GAAI,EAAW,KAErE,EAAW,EAAS,OAAO,EAAW,MAAM,EAAgB,IAE9D,MAAO,eAaL,EAAsB,EACtB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAW,GACjB,GAAI,GACF,EAAS,KAAK,GACd,OAAS,GAAI,EAAiB,EAAG,EAAI,EAAc,EAAE,EACnD,AAAI,GAAK,EAAI,EACX,GAAS,KAAK,GACd,EAAS,KAAK,EAAK,GAAiB,KAEpC,EAAS,KAAK,QAMlB,OAFM,GAAsB,GACtB,EAAqB,GAClB,EAAI,EAAG,EAAI,EAAc,EAAE,EAClC,AAAI,GAAK,EAAiB,EAAI,GAAK,EAAI,IAAM,EAC3C,EAAmB,KAAK,GAExB,EAAoB,KAAK,GAG7B,EAAS,KAAI,MAAb,EAAiB,GACjB,EAAS,KAAK,GACd,EAAS,KAAI,MAAb,EAAiB,GAEnB,MAAO,eAaL,EAAsB,EAAsB,EAC5C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAmB,GAEzB,AAAI,EACF,EAAiB,KAAK,EAAW,GAAK,GAEtC,EAAiB,KAAK,EAAW,GAAK,GAGxC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,GAAK,EAAW,OAClB,AAAI,EACF,EAAiB,KAAK,EAAW,EAAI,GAAK,EAAW,IAErD,EAAiB,KAAK,EAAW,GAAK,EAAW,EAAI,IAGvD,EAAiB,KAAK,EAAW,IAIrC,MAAO,eAQL,EAAmB,GAErB,OADM,GAAmB,CAAC,GACjB,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAiB,KAAK,EAAM,GAAG,IAEjC,MAAO,eAeL,EAA0B,EAAmB,GAE/C,OADM,GAAY,EAAe,MAAM,EAAG,GACjC,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,EAAU,KAAK,EAAe,EAAI,GAAK,EAAM,GAAG,GAAK,EAAM,GAAG,IAGhE,MAAO,GCvJT,AAiBO,GAAM,IAAkB,mBAClB,GAAa,mBClB1B,AAiBO,GAAM,IAAQ,SACR,GAAS,WACT,GAAS,YACT,GAAS,YACT,GAAS,aACT,GAAS,YCtBtB,qBAmBqB,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACnB,AAAK,KAAM,QAAQ,YACjB,QAAQ,KAAI,MAAZ,QAAgB,wBAIA,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GAClB,AAAK,KAAM,QAAQ,YACjB,QAAQ,IAAG,MAAX,QAAe,GC3BnB,YAiCI,EAAoB,GACtB,GAAI,EAAK,SAAW,EAAK,OACvB,KAAM,IAAI,OACN,gEACG,GAAK,OAAM,WAAW,EAAK,OAAM,MAG1C,OADM,GAAS,GAAI,cAAa,EAAK,OAAS,GACrC,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EACtC,EAAO,GAAK,EAAK,EAAI,GACrB,EAAO,EAAI,GAAK,EAAK,EAAI,GAE3B,MAAO,eAiB8B,GAIrC,OAFM,GAAO,GAAI,cAAa,EAAQ,OAAS,GACzC,EAAO,GAAI,cAAa,EAAQ,OAAS,GACtC,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,EAAI,GAAK,EAAQ,GACtB,EAAK,EAAI,GAAK,EAAQ,EAAI,GAE5B,MAAO,CAAC,KAAI,EAAE,KAAI,eAOiB,GAKnC,OAHM,GAAM,KAAK,KAAK,EAAQ,OAAS,GACjC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAOgB,GAKlC,OAHM,GAAM,KAAK,MAAM,EAAQ,OAAS,GAClC,EAAO,GAAI,cAAa,GACxB,EAAO,GAAI,cAAa,GACrB,EAAI,EAAG,EAAI,EAAQ,OAAQ,GAAK,EACvC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,GAClC,EAAK,KAAK,MAAM,EAAI,IAAM,EAAQ,EAAI,GAExC,MAAO,CAAC,KAAI,EAAE,KAAI,eAShB,EAAuB,GACzB,GAAM,GAAO,EAAQ,EAAQ,GACvB,EAAO,EAAQ,EAAQ,EAAI,GACjC,MAAO,CAAC,KAAI,EAAE,KAAI,eAUhB,EAAkB,EAAc,EAAc,GAChD,EAAK,EAAQ,GAAK,EAClB,EAAK,EAAQ,EAAI,GAAK,cAOpB,EAAW,GAGb,OAFM,GAAO,GAAI,cAAa,EAAI,GAC5B,EAAO,GAAI,cAAa,EAAI,GACzB,EAAI,EAAG,EAAI,KAAK,KAAK,EAAI,GAAI,KACpC,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GAC9C,EAAK,GAAK,KAAK,IAAI,GACnB,EAAK,GAAK,KAAK,IAAI,GAErB,MAAO,CAAC,KAAI,EAAE,KAAI,eAOhB,EAAW,EAAW,GACxB,GAAM,GAAK,GAAU,EAAI,IAAM,KAAK,GAAM,GAAI,GACxC,EAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GACtB,MAAO,CAAC,KAAI,EAAE,KAAI,GCvJpB,YAyDI,EAAM,EAAiB,GACzB,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,GAAE,QAEX,GAAM,GAAc,GAAM,EAAE,OACtB,EAAS,GAAK,EAAG,WACjB,EAAS,EAAQ,QAAQ,EAAQ,GACvC,SAAY,UACZ,EAAO,UACA,EAGT,GAAI,CAAC,GAAgB,EAAE,MAAO,GAG5B,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAE,MAAO,GAExD,GAAI,EAAE,QAAU,aACd,GAAM,GAAO,EAAQ,KAAK,GACpB,EAAS,GAAK,EAAM,GAC1B,SAAK,UACE,EAET,GAAI,IAAU,QACZ,MAAO,GAAQ,IAAI,GACd,GAAI,IAAU,QACnB,GAAM,GAAO,GAAO,EAAG,EAAE,OACnB,EAAS,EAAQ,SAAS,EAAG,GACnC,SAAK,UACE,MAEP,MAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,eAK/D,EAAM,GACR,MAAO,GAAO,qBAAqB,EAAE,OAAQ,EAAO,EAAE,mBAG3B,EAAe,EAAc,GACxD,GAAM,GAAQ,GAAO,GAAU,GAAM,GAE/B,EAAS,GAAoB,EAAK,WACxC,EAAO,GAAK,EACZ,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAO,EAAI,GAAK,EAG9B,MAAO,IAAS,EAAQ,qvCC3G1B,YAyBI,EAAM,EAAsB,GAC9B,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,MAAO,GAAW,IAAI,SAAA,GACpB,GAAM,GAAgB,EAAI,QAC1B,EAAU,GAAQ,EAClB,GAAM,GAAS,GAAM,EAAG,EAAO,GAC/B,SAAM,IAAS,EACR,ICjCX,YA2BI,EAAiC,GAEnC,OADM,GAAqB,GAAI,OAAM,EAAK,MACjC,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAK,MAAM,GAAK,EAAK,GAGrC,OADM,GAAS,GAAO,EAAU,EAAK,OAC5B,EAAI,EAAG,EAAI,EAAO,OAAO,OAAQ,EAAE,GAI1C,OAHM,GAAS,EAAO,WAAW,GAE3B,EAAwB,GAAI,OAAM,EAAK,MACpC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,MAAM,GAG1C,GAAM,GAAgB,EAAK,WAAW,GAEtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,WC7ChB,YAyBI,EAAe,EAAkB,EAAyB,EAC1D,GAOF,OALM,GAAU,EAAO,EAAO,OAAS,GACjC,EAAA,CAAA,EAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAc,GAAuB,EAAQ,EAAQ,GACrD,EAAiB,GAAuB,QAAS,EAAQ,GAEtD,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHM,GAAS,EAAI,EACb,EAAO,EAAE,SAAS,EAAQ,EAAS,GACnC,EAAmD,GAChD,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,EAAU,KAAK,CAAC,MAAO,EAAK,GAAI,MAAO,IAEzC,EAAU,KAAK,SAAC,EAAG,GAAM,MAAA,GAAE,MAAQ,EAAE,QAKrC,OAHM,GAAY,EAAI,EAChB,EAAW,EAAY,SAAS,EAAW,EAAY,GACvD,EAAc,EAAe,SAAS,EAAW,EAAY,GAC1D,EAAI,EAAG,EAAI,EAAG,IACrB,EAAS,GAAK,EAAU,GAAG,MAC3B,EAAY,GAAK,EAAU,GAAG,MAKlC,GAAM,GAAc,EAAO,QAC3B,SAAY,EAAY,OAAS,GAAK,EAC/B,CACL,GAAO,EAAa,EAAa,GACjC,GAAO,EAAgB,EAAa,UCxDxC,mJCAA,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,WAAY,SC7BtD,AA4BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAO,GAAK,EAAG,YACnB,EAAI,GAAK,GAAI,GAAO,GAAI,IAC9B,MAAO,IAAI,GAAI,EAAI,QCtC3B,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAK,EAAG,YAAa,IAC/C,MAAO,IAAI,EAAI,OCnCvB,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAGxB,MAAO,CAAC,EAAS,EAAG,KChDxB,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAsC,GAC5C,SAAM,QAAQ,SAAC,EAAG,GAChB,EAAK,GAAK,WAAM,MAAA,GAAG,WAEd,IC7BX,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,mBChChE,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CACL,EAAG,WACD,GAAM,GAAI,GAAK,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,cAC7C,MAAO,IAAI,EAAI,OCpCvB,AA6BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,EAAI,EAAI,GAAI,EAAG,IACnB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAM,GAAI,GAAI,GAAO,GAAI,GAAO,IAC5B,EAAM,GAAI,EAAI,EAAI,GAAI,EAAG,KACvB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAExB,MAAO,CAAC,EAAS,EAAG,KCtDxB,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAK,EAAG,YAAa,QC/B7D,AA0BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,GAAI,GAAO,GAAK,EAAG,kBC/B3D,AA2DA,YACI,EAAkB,EAClB,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBAE3C,EAAO,EACP,EAAU,EACV,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,MAIvE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OAEvB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OAErE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,IAG5C,EAAkC,CAAC,GAAI,EAAM,MAAO,GAEpD,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KChIvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAY,EAAS,EAAY,EACpD,OCpCV,AA8CA,YACI,EAAkB,EAAqB,EACvC,EAAkC,GACpC,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBAE/C,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,gCAC/B,EAAI,KAAI,MAEhB,GAAI,GAAU,EACV,EAAO,EACP,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EACI,EAAQ,EAAQ,CAAC,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KACvE,EAAO,EAAQ,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,MAGhE,EACI,EAAK,OAAS,EACd,WAAM,MAAA,4DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,+DACC,GAAQ,KAAI,OAEvB,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAmB,GAE3D,MAAO,GAAQ,gBAAgB,EAAM,EAAS,IAG1C,EAAgC,CAAC,GAAI,EAAM,MAAO,GAElD,EAA8B,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,GAEvD,EAAM,EAAO,cACf,EAAS,EAAgC,KAAM,GAC/C,GAEJ,MAAI,GACK,EAAQ,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAEtD,EAGF,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAC5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAY,EAAS,OC7BvE,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,WAEnB,MAAI,CAAC,GAAc,CAAC,EACX,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,MAEtB,CAAC,GAAc,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,MAEtB,GAAc,CAAC,EACjB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,KAC9B,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAO,MAGzB,CACL,EAAG,WAAM,MAAA,IAAO,EAAG,EAAI,GAAM,KAC7B,EAAG,WAAM,MAAA,IAAO,EAAI,EAAG,GAAM,QChDrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,MACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAsBO,GAAM,IAAoC,CAC/C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAQtC,OAPM,GACF,EAEE,EAAa,EAAiB,WAC9B,EAAc,EAAiB,MAE/B,EAAiB,MAAM,KAAK,GACzB,EAAI,EAAW,OAAS,EAAG,GAAK,EAAG,IAC1C,GAAI,EAAW,KAAO,EAAY,GAChC,EAAK,GAAK,UACD,EAAW,KAAO,EAC3B,KAAM,IAAI,OAAM,mBACZ,EAAU,6BAA6B,EAAW,MAI1D,OADM,GAAiB,GACd,EAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,EAAK,GAAK,GACZ,EAAK,KAAK,GAId,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAM,QC/CnC,AAqBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,GAAG,YCxBxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAET,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AA0BO,GAAM,IAAoC,CAC/C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,aAAc,EAAA,EAAA,aACrB,MAAO,CACL,EAAG,WAAM,MAAA,IACL,GAAW,GAAa,EAAG,GAAe,GAAU,EAAG,IACvD,EAAI,GAAU,QCnCxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,cAAe,GACf,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAS,EAAM,IAAI,SAAA,GAAK,MAAA,GAAE,QACzB,EAAA,EAAA,KACD,EAAQ,GAAe,EAAM,EAAM,GAAG,OAAO,GAC7C,EAAa,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAa,GAAM,EAAI,EAAY,GACzC,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,YAAM,MAAA,QC/BrC,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GACN,EAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAEhC,SACI,GAA4B,GAC5B,WAAM,MAAA,8DACF,uDAAsD,EAAS,OAEhE,CACL,EAAG,WACC,MAAA,IAAoB,EAAI,MAAO,EAAI,EAAS,EAAS,EAAK,IAC9D,OAAQ,WACJ,MAAA,IAAqB,EAAK,EAAI,EAAQ,MAAO,EAAS,EAAK,OCxCrE,AAsBO,GAAM,IAA4C,CACvD,WAAY,GACZ,aAAc,CAAC,KAAM,UACrB,SAAU,SAAC,EAAe,EAAiB,GACnC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,WAAY,EAAA,EAAA,gBAGjC,MAAO,CACL,GAAI,WAAM,MAAA,IACN,EAAK,EAAQ,EAAS,EAAK,EAAY,EACvC,IACJ,OAAQ,WAAM,MAAA,IACV,EAAK,EAAI,EAAO,MAAO,EAAS,EAAK,EAAY,OCpC3D,AA2CA,YACI,EAAM,EAAO,EACb,EAA0C,GAC5C,GAAI,GAAM,EACV,AAAI,EAAE,OAAS,GACb,GAAM,EAAQ,EAAG,CAAC,EAAG,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,MAEnE,GAAI,GAAO,EACX,AAAI,EAAK,OAAS,GAChB,GAAO,EAAQ,EAAI,CAAC,EAAG,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,GAAI,EAAG,MAAM,MAEzE,EACI,EAAI,OAAS,EACb,WAAM,MAAA,iEACC,GAAI,MAAK,OACpB,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,MAAK,OACrB,EACI,EAAY,SAAW,EACvB,WAAM,MAAA,mEACC,GAAW,OACtB,EACI,EAAI,MAAM,KAAO,EAAY,GAC7B,WAAM,MAAA,4CAA4C,EAAI,MAAM,GAAE,UAC1D,iCAAgC,EAAY,GAAE,OACtD,EACI,EAAK,MAAM,KAAO,EAAY,GAC9B,WAAM,MAAA,0CAA0C,EAAK,MAAM,GAAE,UACzD,mCAAkC,EAAY,GAAE,QAExD,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAY,EAEZ,EAAW,GACb,EAAI,MAAO,EAAa,EAAS,EAAW,GAEhD,MAAO,GAAQ,gBAAgB,EAAK,EAAM,IAGtC,EAAqC,CAAC,EAAG,EAAK,EAAG,GAEjD,EAAmC,CAAC,QAAO,EAAE,IAAG,GAEtD,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAwB,GAG9B,GAAM,IAAuB,EAAG,CAAC,sBAAqB,KC7F7D,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAC3B,EACI,GAAkB,GAClB,WACI,MAAA,kEACA,mDAAkD,EAAS,OAE5D,GAAA,GAAA,EAAA,GAAK,EAAA,EAAA,GAEZ,MAAO,CACL,EAAG,WAAM,MAAA,IACJ,EAAiB,MAAO,EAAI,EAAqB,EAAS,IAC/D,OAAQ,WAAM,MAAA,IACV,EAAiB,EAAK,EAAqB,MAAO,EAAS,OCzCrE,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAI,GAAK,EAAG,aAAc,OC/BvD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAwBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,KAAM,EAAA,EAAA,UAAW,EAAA,EAAA,QAExB,MAAO,CACL,EAAG,WACD,GAAM,GAAc,GAAmB,CAAC,GAAO,EAAE,MAE7C,EAAM,GAAO,EAAI,EAAM,EAAW,CAAC,GAEvC,MAAI,IAAe,MACjB,GAAM,GAAU,EAAK,IAGhB,MCzCf,AAwBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAG1B,EAAa,GAAoB,CAAC,EAAG,GAE3C,EACI,GAA4B,GAC5B,WAAM,MAAA,kHAEF,KAAI,EAAU,OAEhB,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEV,EACI,EAAE,OAAS,EACX,WAAM,MAAA,6DACF,yBAAwB,EAAE,KAAI,OACtC,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,8DACF,yBAAwB,EAAO,KAAI,OAC3C,EACI,EAAE,MAAM,KAAO,EAAO,MAAM,GAC5B,WAAM,MAAA,yDACF,cAAa,EAAE,MAAM,GAAE,0CACvB,cAAa,EAAO,MAAM,GAAE,OAEpC,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,sCAAqC,EAAO,mBAC5C,KAAI,EAAU,QAEtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WACI,MAAA,gEACA,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAAW,GACb,EAAE,MAAO,EAAO,MAAO,EAAS,EAChC,EAAK,EAAiB,IAE1B,MAAO,CACL,EAAG,WACC,MAAA,IAAmC,EAAE,MAAO,EAAI,EAAQ,IAC5D,OAAQ,WACJ,MAAA,IAAoC,EAAG,EAAI,EAAO,MAAO,OC7EnE,AAuBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,IAAK,UACpB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAEJ,EAA6C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAC3D,EAA+C,CAAC,EAAC,EAAE,OAAM,EAAE,GAAE,GAEnE,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,UACH,GAAyB,EACzB,IACb,OAAQ,WAAM,MAAA,GAAO,UACH,GACA,EAAsC,OCtC9D,AA6BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OACnD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAgC,EAAE,MAAO,GAC5D,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GAC/C,MAAO,GAAQ,OAAO,EAAI,IAGtB,EAAwB,CAAC,GAAE,EAAE,EAAC,GAEpC,MAAO,CACL,EAAG,WAAM,MAAA,GAAO,cACZ,EAAoB,EAAgC,KACpD,QCrCV,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAI,EAAI,GAAI,GAAI,GAAO,KAAM,EAAI,KAAK,KAAK,KAAK,KACtD,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC/B7B,AAsBO,GAAM,IAA4B,CACvC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,OC3B7B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAI,QC5BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AA6BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,GAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAI,GAAM,EAAI,EAAI,GAAK,EAAG,YACpB,EAAa,GAAiB,EAAE,MAAO,GAC7C,AAAI,EAAW,OAAS,GACtB,GAAM,EAAQ,GAAI,EAAK,GAAa,EAAE,QAExC,GAAM,GAAM,GAAO,GACnB,MAAO,IAAI,GAAI,EAAK,GAAK,EAAK,cAEhC,MAAO,CAAC,EAAS,EAAG,KCrDxB,AA8BO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,IAAK,OAAQ,WAAY,SACxC,SAAU,SACN,EAAY,EAAiB,GACxB,GAAA,GAAA,EAAA,gBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GAAM,EAAA,EAAA,GAAU,EAAA,EAAA,GAEpB,EAAa,GAAgB,GAAO,GACpC,EAAgB,GAAiB,EAAK,MAAO,EAAE,OAC/C,EAAsB,GAC5B,GAAI,EAAK,OAAS,GAChB,OAAS,GAAI,EAAG,EAAI,EAAE,MAAM,OAAS,EAAG,EAAE,EACxC,EAAU,KAAK,EAAE,MAAM,IAEzB,EAAU,KAAK,GAGjB,GAAM,GAAa,GAAI,EAAG,GACpB,EAAoB,EAAI,EAAI,GAC5B,EAAsB,GAAM,GAAI,EAAU,GAAO,KACjD,EAAiB,EACnB,EAAI,EAAI,EAAqB,GAAsB,GACnD,GAAO,MAEL,EAAO,WACX,MAAI,GAAK,OAAS,EACT,EACH,EAAI,EAAI,EACA,GACI,EAAQ,EAAqB,CAAC,EAAG,EAAG,EAAG,EAAK,MAAM,KAClD,IACR,GACJ,EAAE,OAEC,EAAQ,EAAI,EAAI,EAAI,GAAsB,GAAa,EAAE,QAG9D,EAAU,WACd,GAAI,GACA,EAAI,EAAI,EAAqB,GAAO,KAAM,GAC9C,MAAI,GAAK,OAAS,GAChB,GAAU,GAAI,EAAS,IAElB,EAAQ,EAAS,EAAK,QAEzB,EAAc,WAClB,GAAI,GAAc,EAAI,EAAI,EAAgB,GAAa,GAEvD,MAAI,GAAK,OAAS,GAChB,GAAc,GAAI,EAAa,IAE1B,EAAQ,EAAa,EAAK,QAE7B,EAAW,WACf,GAAM,GAAwB,EAAI,EAAY,GAE1C,EAAW,EAAI,EAAI,GACvB,MAAI,GAAK,OAAS,GAChB,GAAW,GAAI,EAAU,IAEpB,EAAQ,EAAU,EAAK,QAE1B,EAAY,WAChB,GAAI,GAAY,EAChB,MAAI,GAAK,OAAS,GAChB,GAAY,GAAI,EAAW,IAEtB,EAAQ,EAAW,EAAK,QAGjC,MAAO,CACL,EAAG,EACH,KAAM,EACN,SAAU,EACV,MAAO,EACP,OAAQ,KC1Gd,AA0BO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,IAAK,WACpB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACH,EAAA,EAAA,KAED,EAAa,GAAe,EAAM,EAAE,OAAO,GAE3C,EAAO,WACX,GAAM,GAAc,EAAE,MAChB,EAAc,EAAQ,KAEtB,EAAa,EAAY,MAAM,EAAG,GAClC,EAAY,EAAW,OACvB,EAAa,EAAY,MAAM,EAAM,EAAY,QAAQ,MAAM,GAC/D,EAAY,EAAW,OAEvB,EAAmB,GAAW,EAAG,GACjC,EACF,GAAW,EAAY,EAAG,EAAY,EAAI,GAExC,EAAc,GAAY,CAAC,EAAY,CAAC,GAAc,IAEtD,EAAS,EAAQ,EAAI,GACrB,EAAkB,EAAQ,EAAS,CAAC,IAEpC,EACF,GAAY,CAAC,CAAC,GAAY,EAAkB,IAC1C,EAAkB,GAAU,EAAQ,GACtC,EAAa,GACb,EAAiB,EAA6B,EAAE,MAAM,IAEpD,EAAsB,GAAuB,GACnD,SAAa,GAAU,EAAY,GAE5B,GAET,MAAO,CAAC,EAAG,EAAM,QAAS,WAAM,MAAA,OAIpC,YAAoB,EAAe,GAEjC,OADM,GAAS,GACN,EAAI,EAAO,EAAI,EAAM,EAAE,EAC9B,EAAO,KAAK,GAEd,MAAO,GAGT,YAAqB,GAEnB,OADM,GAAS,GACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,EACtC,EAAO,KAAK,EAAO,GAAG,IAG1B,MAAO,GCnFT,AAqBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,IAAI,EAAG,WAAM,MAAA,IAAU,OC1BtD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAK,EAAI,eCzB9B,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC1B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC5B/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,EAAG,QC5BpC,AAuBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAK,EAAG,gBC5BrC,AAyBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,GACd,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KACP,MAAO,CACL,OAAQ,WACN,GAAM,GAAW,GACX,EAAU,GAAI,GACpB,MAAO,IAAI,EAAI,EAAI,GAAI,EAAI,EAAM,GAAW,QCpCpD,AAyBA,YACI,EAAM,EAAM,EAAO,EAAiB,EAAU,EAAW,GAAtC,AAAA,IAAA,QAAA,GAAA,GAAiB,AAAA,IAAA,QAAA,GAAA,GAAU,AAAA,IAAA,QAAA,GAAA,GAAW,AAAA,IAAA,QAAA,GAAA,IAC3D,GAAM,GAA+B,SAAA,GACjC,MAAA,GAAQ,QAAQ,EAAI,EAAG,EAAG,EAAa,EAAM,EAAO,IAElD,EAA4B,CAAC,EAAC,EAAE,EAAC,EAAE,GAAE,GAErC,EAA0B,CAAC,YAAW,EAAE,KAAI,EAAE,MAAK,EAAE,KAAI,GAE/D,MAAO,GAAO,cACH,EAAS,EAAgC,KACzC,GAAa,GAGnB,GAAM,IACT,EAAG,CAAC,oCAAmC,KCxC3C,AAqBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAc,EAAiB,GAClC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,YAAa,EAAA,EAAA,KAAM,EAAA,EAAA,MAAO,EAAA,EAAA,KAEjC,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAG,EAAG,EAAI,EAAa,EAAM,EAAO,OC/B9C,YA6BI,EAAO,EAAM,EAAe,EAAoB,GAClD,MAAI,GAAE,KAAO,EAAM,MACjB,GAAI,EAAQ,EAAG,GAA+B,EAAE,MAAO,KAEzD,AAAI,EAAG,KAAO,EAAM,MAClB,GAAK,EAAQ,EAAI,GAA+B,EAAG,MAAO,KAErD,CACL,EAAG,WACD,GAAM,GAAK,EAAI,EAAI,GAAK,GAAM,EAAO,GAAI,EAAG,QAC5C,MAAO,IAAgB,KAAO,EAAK,GAAU,EAAI,KCvCvD,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,iBACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAkB,EAAE,OACnD,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAa,EAAG,GAAI,aAC9C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,EAAG,GAAI,aAC5C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA6DA,YACI,EAAkB,EAAqB,EACvC,EACA,EACA,EACA,EAA4B,GAD5B,AAAA,IAAA,QAAA,GAAA,CAA8C,EAAG,EAAG,IAEtD,GAAM,GAAM,EAAgB,EAAI,KAAM,qBAChC,EAAS,EAAgB,EAAO,QAAS,qBACzC,EAAU,EAAgB,EAAQ,SAAU,qBAE9C,EAAO,EACP,EAAU,EACV,EAAW,EACX,EAAe,GAEnB,AAAI,EAAO,OAAS,GAClB,GAAe,GACf,EAAO,EACH,EAAK,CAAC,EAAG,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KACjE,EAAU,EAAQ,EAAQ,CACxB,EAAG,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,GAAI,EAAO,MAAM,KAErE,EAAW,EAAQ,EAAS,CAC1B,EAAG,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,GAAI,EAAQ,MAAM,MAI3E,EACI,EAAK,OAAS,EACd,WAAM,MAAA,8DACC,GAAK,KAAI,OACpB,EACI,EAAQ,OAAS,EACjB,WAAM,MAAA,iEACC,GAAQ,KAAI,OACvB,EACI,EAAS,OAAS,EAClB,WAAM,MAAA,kEACC,GAAS,KAAI,OAExB,EACI,GAAyC,EAAS,GAClD,WAAM,MAAA,2DACF,2BAA0B,EAAO,mBAAmB,EAAS,OACrE,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,2DACF,2BAA0B,EAAe,gBAAgB,EAAG,OAGtE,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAQ,MAAO,EAAY,EAAS,EAAW,EAAK,GAExD,MAAO,GAAQ,kBAAkB,EAAM,EAAS,EAAU,IAGtD,EACwB,CAAC,GAAI,EAAM,MAAO,EAAS,OAAQ,GAE3D,EACF,CAAC,WAAU,EAAE,QAAO,EAAE,UAAS,EAAE,IAAG,EAAE,gBAAe,GAEnD,EAAM,EAAO,cACf,EAAS,EAAgC,KACzC,GAAmB,GAEvB,MAAI,GACK,EACI,EAAK,CAAC,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,GAAI,EAAI,MAAM,KAIhE,EAGF,GAAM,IAAoB,EAAG,CAAC,mBAAkB,KC1IvD,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,IAAK,EAAA,EAAA,gBAGtC,EACF,GAAoB,CAAC,EAAG,EAAG,GAE/B,MAAO,CACL,EAAG,WAAM,MAAA,IACL,EAAgB,EAAG,EAAG,EAAY,EAAS,EAAY,EACvD,OCrCV,AAkDA,YACI,EAAyB,EACzB,EAA6B,EAC7B,EAAkC,EAClC,GACF,GAAM,GAAM,EAAgB,EAAI,KAAM,mBAChC,EAAS,EAAgB,EAAO,QAAS,mBACzC,EAAU,EAAgB,EAAQ,SAAU,mBAElD,EACI,EAAO,OAAS,EAAI,KACpB,WAAM,MAAA,kBAAkB,EAAO,KAAI,+BAC/B,KAAI,EAAI,KAAI,OAEpB,EACI,EAAI,OAAS,EACb,WAAM,MAAA,4DACC,GAAI,KAAI,OACnB,EACI,EAAO,OAAS,EAChB,WAAM,MAAA,+DACC,GAAO,KAAI,OACtB,AAAI,GAAmB,MACrB,EACI,GAAW,GACX,WAAM,MAAA,gEACF,oBAAmB,EAAe,gBAAgB,EAAG,OAG/D,GAAM,GAA+B,SAAA,GACnC,GAAM,GAAW,GACb,EAAO,MAAO,EAAY,EAAS,EAAmB,EACtD,GAEJ,MAAO,GAAQ,gBAAgB,EAAK,EAAQ,EAAS,IAGjD,EACsB,CAAC,GAAI,EAAK,MAAO,EAAQ,OAAQ,GAEvD,EACqB,CAAC,WAAU,EAAE,QAAO,EAAE,IAAG,EAAE,gBAAe,GAErE,MAAO,GAAO,cACH,EAAS,EAAgC,KAAM,GAC/C,GAGN,GAAM,IAAkB,EAAG,CAAC,iBAAgB,KClGnD,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAE5B,MAAO,CACL,EAAG,WAAM,MAAA,IAAgB,EAAgB,EAAG,EAAG,EAAY,EAAS,OC/B1E,AA0BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAqB,EACpB,EAAA,EAAA,KACA,EAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAAoB,EAAM,EAAE,OACvC,EAAe,GAA6B,EAAU,EAAE,MACxD,EAAU,GAAiB,EAAI,EAAG,EAAG,EAAU,GACrD,MAAO,CACL,EAAG,WACD,GAAI,GAAM,EAAQ,IAClB,MAAI,IAAgB,MAClB,GAAM,GAAU,IAEX,MC3Cf,AAyBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAU,EAAG,GAAI,aAC3C,EAAO,WAAM,MAAA,GAAI,EAAI,GAAK,GAAQ,EAAG,GAAI,aAC/C,MAAO,CAAC,EAAG,EAAM,EAAG,KChCxB,AA4BO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAI,GAAa,EAAE,OAEjC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAI,GAAM,GAAI,EAAG,MAC/B,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KClDxB,AA0BO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAW,GAA2B,EAAE,MAAO,EAAE,OAEjD,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAEH,EAAO,WACX,GAAM,GAAM,EAAI,EAAI,GAAK,EAAG,YACtB,EAAa,GAAiB,EAAE,MAAO,GAC7C,MAAI,GAAW,OAAS,EACf,EAAQ,GAAI,EAAK,GAAa,EAAE,OAElC,GAET,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,OCzBzB,AAsBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,WACf,SAAU,SAAC,EAAY,GACrB,GAAM,GAAU,EAAM,GACtB,MAAO,CAAC,QAAS,WAAM,MAAA,IAAM,EAAQ,MAAO,eC3BhD,AAsBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAGtC,GAAM,GAAI,EAAM,GACT,EAAA,EAAA,SACD,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAClC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAM,EAAI,EAAO,EAAE,WC/BxC,AAgCO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GAAG,EAAA,EAAA,GACP,EAAO,EACP,EAAM,EACN,EACF,GAA0C,EAAK,MAAO,EAAI,OAExD,EAAU,WACd,GAAM,GAAW,GAAK,EAAK,WACvB,EAAM,EAAI,EAAI,EAAI,EAAU,GAAI,EAAM,GAAI,EAAU,GAAO,OACzD,EAAa,GAAgC,EAAK,MAAO,GAC/D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAK,QAErB,EAAS,WACb,GAAM,GAAY,GAAQ,EAAM,GAC1B,EAAU,GAAM,EAAW,GAAI,GAAO,GAAU,IAClD,EAAM,EAAI,EAAI,EAAI,EAAG,IACnB,EAAa,GAAgC,EAAI,MAAO,GAC9D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAI,QAE1B,MAAO,CAAC,EAAG,EAAS,EAAG,KC9D3B,AA2BO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,IAAK,SACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAO,GAAQ,EAAG,GAExB,MAAO,CACL,EAAG,WAAM,MAAA,IAAM,EAAM,EAAI,EAAI,EAAI,KACjC,MAAO,WACL,GAAI,GAAM,GAAM,EAAM,GAAU,GAAK,EAAI,EAAI,IACvC,EAAa,GAAiB,EAAM,MAAO,EAAG,OACpD,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAM,WC1ClC,AAwBO,GAAM,IAAmC,CAC9C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAI,GAAO,SC7BxC,AAwBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACD,EAAO,EAAI,GAAU,EAAG,GAAI,GAAK,IAEvC,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAM,gBC/BxC,AAuBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAK,GAAI,gBC5B3C,AAqBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAQ,EAAI,EAAE,WC1BnC,AAuBO,GAAM,IAAuC,CAClD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,uBACX,EAAI,EAAoB,IAGxB,EAAmC,CAAC,OAAM,GAC1C,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAAoB,IAExB,MAAO,CAAC,OAAQ,KCxCpB,AAuBO,GAAM,IAA8C,CACzD,WAAY,GACZ,aAAc,CAAC,UACf,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,GAED,EAA0C,SAAC,GACxC,GAAA,GAAA,EAAA,aACP,MAAO,GAAQ,8BACX,EAAI,EAAoB,IAGxB,EAA0C,CAAC,OAAM,GACjD,EAAY,WAAM,MAAA,GAAO,cAC3B,EAAoB,EAAgC,KACpD,GAA2B,IAE/B,MAAO,CAAC,OAAQ,KCxCpB,AAsBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,KACD,EAAO,GAAe,EAAM,EAAG,OACrC,MAAO,CAAC,EAAG,WAAM,MAAA,IAAQ,EAAI,OC3BjC,AAsBO,GAAM,IAA8B,CACzC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAyBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,GAAI,EAAI,EAAI,GAAI,EAAG,KAAM,SC9BlD,AAyBO,GAAM,IAAqC,CAChD,WAAY,GACZ,aAAc,CAAC,aACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAGL,UAAW,WAAM,MAAA,IAAK,GAAU,GAAY,YAC5C,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,EAAW,EAAG,SACpC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAK,GAAW,GAAY,EAAG,YCnCtD,AA2BO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CACL,EAAG,WACD,GAAM,GAAO,GAAQ,EAAG,GAAO,IAEzB,EAAa,GAAO,IACpB,EAAQ,GAAO,IAEf,EAAqB,EAAI,EAAI,GAC7B,EACF,EAAI,EAAI,EAAI,GAAa,GAAI,GAAK,EAAG,aAEzC,MAAO,IAAM,EAAM,EAAoB,OC3C/C,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,EAAG,GAAI,GAAO,GAAI,SC9BnD,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAK,EAAG,YAAa,OC7BlD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAK,GAAK,EAAG,YAAa,OC9BnD,AAuBO,GAAM,IAA8B,CACzC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GActC,OAbO,GAAA,EAAA,GACD,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,KAER,EAAa,EAAE,MACf,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAQT,EAAoC,GACjC,EAAI,EAAG,EAAI,EAAG,KAAM,IAC3B,EAAS,KAAK,CAAC,EAAO,GAAI,EAAW,GAAK,EAAO,GAAK,EAAM,KAE9D,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,OC3C7B,AAwBO,GAAM,IAAgC,CAC3C,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,IACD,EAAW,GAEX,EAAW,EAAI,EAAI,GACzB,MAAO,CACL,OAAQ,WAAM,MAAA,IAAI,EAAU,EAAI,GAAI,EAAU,CAAC,GAAM,GAAW,QClCtE,AAuBO,GAAM,IAAiC,CAC5C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,GAAQ,QC7BrC,AAsBO,GAAM,IAAuC,CAClD,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GAChC,GAAA,GAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,SACnB,MAAO,CAAC,EAAG,WAAM,MAAA,IAAe,EAAI,EAAY,OC1BpD,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACjC,GAAA,GAAA,EAAA,KAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAO,EAAI,OC1BhC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,EAAI,GAAK,GAAK,EAAG,YAAa,QC/B3D,AAuBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GACP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,EAAI,EAAI,GAAK,EAAG,WAAY,QC5BrD,AAwBO,GAAM,IAA0C,CACrD,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EAAM,GAAO,GACb,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MACrC,EAAO,WAAM,MAAA,GAAI,EAAI,EAAI,EAAK,GAAI,EAAG,MAC3C,MAAO,CAAC,EAAS,EAAG,KChCxB,AAsBO,GAAM,IAA6B,CACxC,WAAY,GACZ,SAAU,SAAC,GAGT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OC3B/B,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,IAAK,KACpB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAAG,EAAA,EAAA,GACJ,EACF,GAA0C,EAAE,MAAO,EAAE,OAEnD,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,EAAK,EAAE,QAElB,EAAO,WACX,GAAI,GAAM,EACJ,EAAa,GAAgC,EAAE,MAAO,GAC5D,MAAI,GAAW,OAAS,GACtB,GAAM,GAAI,EAAK,IAEV,EAAQ,GAAI,GAAM,EAAE,QAG7B,MAAO,CAAC,EAAS,EAAG,KCjDxB,AAyBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACD,EAAkB,EAAE,MAAM,QACzB,EAAA,EAAA,KAED,EAAO,GAAe,EAAM,EAAE,OACpC,EAAK,QAAQ,SAAA,GACX,EAAgB,GAAQ,IAE1B,GAAM,GAAa,EAAQ,EAAI,GACzB,EAAO,EAAI,EAAY,GAAK,EAAE,MAAO,YAE3C,MAAO,CAAC,EAAG,WAAM,MAAA,OCxCrB,AAwBO,GAAM,IAA4B,CACvC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,IAAI,EAAI,GAAO,GAAI,SC9BxC,AAyBO,GAAM,IAA6B,CACxC,WAAY,GACZ,cAAe,CAAC,IAChB,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAEP,MAAO,CAAC,EAAG,WAAM,MAAA,GAAI,GAAI,GAAO,GAAI,GAAO,IAAK,OC/BpD,AAwBO,GAAM,IAA6B,CACxC,WAAY,GACZ,aAAc,CAAC,KACf,SAAU,SAAC,EAAY,EAAiB,GAC/B,GAAA,GAAA,EAAA,GACA,EAAA,EAAA,KAED,EAAO,WACX,GAAI,GAAQ,GAAU,GAGtB,GAAI,EAAE,OAAS,EACb,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,IAAK,CAAC,EAAE,MAAM,cAEjD,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EAAQ,GAAI,EAAO,GAAM,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IAAK,CACjD,EAAE,MAAM,GAAI,EAAE,MAAM,cAI7B,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EAAI,CAAC,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,IACjD,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,cAI1C,EAAE,OAAS,EACpB,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,OAAS,GAAI,EAAG,EAAI,EAAK,GAAI,EAAE,EAC7B,EACI,GAAI,EACA,GACI,EACA,CACE,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAAI,EAAI,EAAE,MAAM,GAC5C,EAAI,EAAE,MAAM,IAEd,CAAC,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,GAAI,EAAE,MAAM,UAMjE,MAAM,IAAI,OACN,2DACG,GAAE,KAAI,kBAEf,MAAO,IAET,MAAO,CAAC,EAAG,KCpFf,AAuBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,EAAY,EAAiB,GACtC,GAAM,GAAiC,EAChC,EAAA,EAAA,KACD,EAAW,GAAiC,GAClD,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,EAAI,OC7BnC,AAqBO,GAAM,IAA+B,CAC1C,WAAY,GACZ,SAAU,SAAC,EAAc,EAAiB,GACxC,GAAM,GAA2B,EAC1B,EAAA,EAAA,KACP,MAAO,CAAC,MAAO,WAAM,MAAA,IAAM,EAAI,OC1BnC,AA8BO,GAAM,IAA2C,CACtD,WAAY,GACZ,aAAc,CAAC,cACf,SAAU,SAAC,EAAY,GACd,GAAA,GAAA,EAAA,GAED,EAAO,WACX,MAAO,IAAoB,EAAI,IAEjC,MAAO,CAAC,EAAG,KAIf,YAA+C,EAAM,GAQnD,OAJM,GAAqB,GAAQ,EAAS,GAAU,IAChD,EAAW,GAAO,EAAG,GACvB,EAAa,GAAa,EAAS,GAAO,EAAG,UAC3C,EAAW,EAAS,KAAO,EAAW,KACnC,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,GAAW,EAAY,EAAI,GAE1C,EAAa,GAAW,EAAY,GAAK,EAAS,MAAO,SACzD,GAAM,GAAY,GAAU,GAC5B,MAAO,IAAM,EAAY,EAAU,GCxDrC,AAsBO,GAAM,IAAkC,CAC7C,WAAY,GACZ,SAAU,SAAC,GACT,MAAO,CAAC,EAAG,WAAM,MAAA,IAAU,OCzB/B,AAkHA,GAAM,IAA4B,CAChC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA6B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAxB,GAAM,IAAc,GAAA,IACvB,GAAiB,ICxNnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA2BA,EAAO,UAAU,IAAM,SACV,EAAwB,GACnC,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA0BA,EAAO,UAAU,OAAS,SAA2B,GACnD,YAAK,kBACE,GAAO,KAAM,IC5BtB,AA+BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACL,EAAO,KAAK,OAAS,EAAG,WAAM,MAAA,wCACvB,EAAQ,KAAM,KClCvB,AAmCA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAQ,KAAM,ICtCvB,AA8BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AAkCA,EAAO,UAAU,KAAO,SACpB,EAAc,GAChB,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,KCrC9B,AAoCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,GACjC,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,KCvCvC,AAqCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,GAChD,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,KCxC9C,AAwCA,EAAO,UAAU,KAAO,SACpB,EAAc,EAAiB,EAAe,EAC9C,GACF,YAAK,kBACE,EAAQ,KAAM,CAAC,EAAM,EAAS,EAAO,EAAQ,KC5CtD,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA+BA,EAAO,UAAU,UAAY,SACzB,EACA,EACA,EACA,EACA,GACF,YAAK,kBACE,GAAU,KAAM,EAAM,EAAU,EAAQ,EAAO,ICtCxD,AA2BA,EAAO,UAAU,YAAc,SAAyB,GAEtD,YAAK,kBACE,GAAY,KAAM,IC9B3B,AA4BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,YAAc,SAC3B,EAAa,GACf,YAAK,kBACE,GAAY,KAAM,EAAK,IC/BhC,AA0BA,EAAO,UAAU,OAAS,SACtB,EAA0B,GAC5B,YAAK,kBACL,AAAI,YAAa,IACf,GAAI,CAAC,IAEA,GAAM,CAAE,MAAI,OAAK,GAAI,IChC9B,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4C,EAC5C,EAAmB,GACrB,YAAK,kBACE,GACI,KAAM,EAAQ,EAAQ,EAAK,EAAY,EACvC,ICrCb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EACA,EAAkC,EAClC,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAa,EAAS,EAAK,ICrCtD,AA8BA,EAAO,UAAU,OAAS,SACtB,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAqB,GACtC,YAAK,kBACE,GAAO,KAAM,EAAM,EAAW,IC9BvC,AA2BA,EAAO,UAAU,aAAe,SAC5B,EAAmB,GACrB,YAAK,kBACE,GAAa,KAAM,EAAW,IC9BvC,AAkCA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,UAAgB,8DAChB,KAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,IC3Cb,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EAA+B,EAC/B,EAA4B,EAC5B,EACA,GACF,YAAK,kBACE,GACI,KAAM,EAAQ,EAAS,EAAK,EAAY,EACxC,ICtCb,AA6BA,EAAO,UAAU,WAAa,SAC1B,EAA+B,EAC/B,EAAqB,EACrB,GACF,YAAK,kBACE,GAAW,KAAM,EAAQ,EAAS,EAAK,EAAW,IClC3D,AA2BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC9BxB,AA4BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC/BzB,AA2BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC7BnB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC5Bb,AA+BA,EAAO,UAAU,YAAc,SAClB,GACX,YAAK,kBACE,GAAY,KAAM,IClC3B,AA0BA,EAAO,UAAU,MAAQ,SAA2B,GAClD,YAAK,kBACE,GAAM,KAAM,IC5BrB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA0BA,EAAO,UAAU,WAAa,SAA2B,GACvD,YAAK,kBACE,GAAW,KAAM,IC5B1B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,EAAQ,KAAM,CAAC,KAAK,QChC7B,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA4BA,EAAO,UAAU,OAAS,SACb,EAA4B,GACvC,YAAK,kBACE,GAAO,KAAM,EAAS,IC/B/B,AA+BA,EAAO,UAAU,mBAAqB,SACzB,GACX,YAAK,kBACE,GAAmB,KAAM,IClClC,AA0BA,EAAO,UAAU,aAAe,SAA2B,GAEzD,YAAK,kBACE,GAAa,KAAM,IC7B5B,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA+BA,EAAO,UAAU,gBAAkB,SACtB,GACX,YAAK,kBACE,GAAgB,KAAM,IClC/B,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA0BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC5BpB,AA2BA,EAAO,UAAU,2BAA6B,SAC1C,EAAsB,EAAe,EAAgB,GACvD,YAAK,kBACE,GAA2B,KAAM,EAAa,EAAM,EAAO,IC9BpE,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA4BA,EAAO,UAAU,WAAa,SACjB,GACX,YAAK,kBACE,GAAW,KAAM,IC/B1B,AA2BA,EAAO,UAAU,UAAY,SAChB,EAAwB,GACnC,YAAK,kBACE,GAAU,KAAM,EAAM,IC9B/B,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA0BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC5BpB,AA0BA,EAAO,UAAU,UAAY,SAA2B,GAEtD,YAAK,kBACE,GAAU,KAAM,IC7BzB,AA0BA,EAAO,UAAU,WAAa,SAA2B,GAEvD,YAAK,kBACE,GAAW,KAAM,IC7B1B,AA2BA,EAAO,UAAU,OAAS,SACb,EAAiB,EAAsB,GAClD,YAAK,kBACE,GAAO,KAAM,EAAG,EAAY,IC9BrC,AA6BA,EAAO,UAAU,QAAU,SACd,EACT,EAAkC,EAClC,GACF,YAAK,kBACE,GAAQ,KAAM,EAAY,EAAS,EAAK,IClCjD,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,KAAO,SACpB,EAAwB,GAC1B,YAAK,kBACE,GAAK,KAAM,EAAM,IC7B1B,AA2BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC9BzB,AA+BA,EAAO,UAAU,cAAgB,SACpB,GACX,YAAK,kBACE,GAAc,KAAM,IClC7B,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,EAAI,KAAM,IC5BnB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA8BA,EAAO,UAAU,KAAO,SACpB,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAK,EAAM,IClC/B,AA+BA,EAAO,UAAU,eAAiB,SACrB,GACX,YAAK,kBACE,GAAe,KAAM,IClC9B,AA0BA,EAAO,UAAU,SAAW,SAA2B,GAErD,YAAK,kBACE,GAAS,KAAM,IC7BxB,AA2BA,EAAO,UAAU,OAAS,SACtB,EAAe,EAAa,GAAb,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,GAC9B,KAAK,kBACE,GAAO,KAAM,EAAO,EAAS,IC9BtC,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA4BA,EAAO,UAAU,IAAM,SACV,EAAmC,GAC9C,YAAK,kBACE,GAAI,KAAM,EAAU,IC/B7B,AA6BA,EAAO,UAAU,KAAO,SACX,EAAsC,EAC/C,EAAgC,EAChC,GACF,YAAK,kBACE,GAAK,KAAM,EAAa,EAAa,EAAS,EAAc,IClCrE,AA+BA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,GAAU,KAAM,ICjCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,MAAQ,SACZ,GACX,YAAK,kBACE,GAAM,KAAM,IC7BrB,AA2BA,EAAO,UAAU,KAAO,SACX,EAAwB,GACnC,YAAK,kBACE,GAAK,KAAM,EAAM,IC9B1B,AA4BA,EAAO,UAAU,WAAa,WAC5B,YAAK,kBACE,GAAW,OC9BpB,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA0BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC5Bf,AAkCA,EAAO,UAAU,UAAY,SAA2B,GACtD,YAAK,kBACE,EAAQ,KAAM,EAAE,QCpCzB,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,EAAQ,KAAM,IC5BvB,AA2BA,EAAO,UAAU,eAAiB,SACrB,EAA8B,GACzC,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,sBAAwB,SAC5B,EAA8B,GACzC,YAAK,kBACE,GAAsB,KAAM,EAAY,IC9BjD,AA0BA,EAAO,UAAU,QAAU,SACd,GACX,YAAK,kBACE,GAAQ,KAAM,IC7BvB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA4BA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAM,OC9Bf,AA0BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC5Bd,AA8BA,EAAO,UAAU,gBAAkB,SAC/B,EACA,EAAsC,EACtC,EAAqB,EACrB,GACF,YAAK,kBACE,GACI,KAAM,EAAiB,EAAiB,EAAS,EAAK,EACtD,ICtCb,AA4BA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,OC9BjB,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA6BA,EAAO,UAAU,MAAQ,SACZ,EAAwB,GACnC,YAAK,kBACE,GAAM,KAAM,EAAO,IChC5B,AA4BA,EAAO,UAAU,QAAU,SAAoC,GAC7D,YAAK,kBACE,GAAQ,KAAM,IC9BvB,AA4BA,EAAO,UAAU,SAAW,WAC1B,YAAK,kBACE,GAAS,OC9BlB,AA2BA,EAAO,UAAU,eAAiB,SAC9B,EAAsB,GACxB,YAAK,kBACE,GAAe,KAAM,EAAY,IC9B1C,AA2BA,EAAO,UAAU,MAAQ,SACrB,EAAkC,GACpC,YAAK,kBACE,GAAM,KAAM,EAAiB,IC9BtC,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA4BA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAO,OC9BhB,AA2BA,EAAO,UAAU,kBAAoB,SAA2B,GAE9D,YAAK,kBACE,GAAkB,KAAM,IC9BjC,AA+BA,EAAO,UAAU,wBAA0B,SAC9B,GACX,YAAK,kBACE,GAAwB,KAAM,IClCvC,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA0BA,EAAO,UAAU,MAAQ,SACrB,EAAoB,GACtB,KAAK,kBACL,GAAM,GAAqB,YAAa,GAAS,CAAC,KAAM,GAAE,CAAI,MAAI,OAAK,GACvE,MAAO,IAAM,EAAoB,IC9BnC,AA4BA,EAAO,UAAU,KAAO,SAAoC,GAC1D,YAAK,kBACE,GAAK,KAAM,IC9BpB,AA+BA,EAAO,UAAU,aAAe,SACd,EAAiB,EAAe,EAC9C,EAAoB,EAAkB,EACtC,EAAsB,GACxB,YAAK,kBACE,GACI,KAAM,EAAO,EAAK,EAAS,EAAW,EAAS,EAC/C,EAAa,ICtC1B,AA+BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IClCzB,AA0BA,EAAO,UAAU,IAAM,SAA2B,GAChD,YAAK,kBACE,GAAI,KAAM,IC5BnB,AA0BA,EAAO,UAAU,IAAM,SACnB,EAAwB,GAC1B,YAAK,kBACE,GAAI,KAAM,EAAM,IC7BzB,AA4BA,EAAO,UAAU,IAAM,WACrB,YAAK,kBACE,GAAI,OC9Bb,AA4BA,EAAO,UAAU,KAAO,WACtB,YAAK,kBACE,GAAK,OC9Bd,AA2BA,EAAO,UAAU,KAAO,SAA2B,GACjD,YAAK,kBACE,GAAK,KAAM,IC7BpB,AAgCA,EAAO,UAAU,OAAS,WACxB,YAAK,kBACE,GAAQ,KAAM,SClCvB,AAgCA,EAAO,UAAU,QAAU,WACzB,YAAK,kBACE,GAAQ,KAAM,YClCvB,AAgCA,EAAO,UAAU,MAAQ,WACvB,YAAK,kBACE,GAAQ,KAAM,UClCvB,AA4BA,EAAO,UAAU,KAAO,SACX,EAAY,GACvB,YAAK,kBACE,GAAK,KAAM,EAAG,IC/BvB,AA2BA,EAAO,UAAU,UAAY,SAChB,GACX,YAAK,kBACE,GAAU,KAAM,IC9BzB,AA2BA,EAAO,UAAU,OAAS,SACb,GACX,YAAK,kBACE,GAAO,KAAM,IC9BtB,AA4BA,EAAO,UAAU,mBAAqB,SACzB,EAAmC,GAC9C,YAAK,kBACE,GAAmB,KAAM,EAAY,IC/B9C,AA0BA,EAAO,UAAU,QAAU,SAA2B,GACpD,YAAK,kBACE,GAAQ,KAAM,IC5BvB,AA4BA,EAAO,UAAU,MAAQ,SACrB,EAA8B,GAChC,YAAK,kBACE,GAAM,EAAW,KAAM,IC/BhC,AA4BA,EAAO,UAAU,UAAY,WAC3B,YAAK,kBACE,GAAU,4vQC9BnB,AAaA,GAAI,kBAMF,MAAI,KAAY,MACd,IAAW,EAAA,UAAU,WAEhB,iBAeP,MAAO,eCrCT,AAuBA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAChC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAa,aAE7C,MAAA,IANkC,OAWlC,EAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OAWhC,GAAA,SAAA,GAAyC,EAAA,EAAA,GACvC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAoB,aAEpD,MAAA,IANyC,OAWzC,GAAA,SAAA,GAAoC,EAAA,EAAA,GAClC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAe,aAE/C,MAAA,IANoC,OAWpC,GAAA,SAAA,GAAgC,EAAA,EAAA,GAC9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAQ,KAEd,cAAO,eAAe,EAAM,EAAW,aAE3C,MAAA,IANgC,OC9EhC,YAuB6B,EAAY,GACvC,GAAI,MAAM,QAAQ,IAGhB,OADI,GAAkB,GACb,EAAI,EAAG,EAAI,EAAW,IAC7B,EAAW,EAAS,OAAO,GAE7B,MAAO,QAEP,GAAM,GAAW,GAAI,OAAM,GAC3B,SAAS,KAAK,GACP,eAIY,EAAc,GACnC,GAAI,CAAC,EACH,KAAM,IAAI,IAAe,eAOJ,EAAY,GAEnC,OADI,GAAU,EACK,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,IAAS,GACX,IAGJ,MAAO,eAQ2B,GAClC,MAAI,GAAG,SAAW,EACT,EAAG,GAEL,cAYc,GACrB,MAAI,OAAM,QAAQ,GACT,EAEF,CAAC,eA0BkB,GAC1B,GAAM,GAAe,EAAK,QAAQ,uBAAwB,SACpD,EACF,EAAa,QAAQ,kBAAmB,SAAS,cAKrD,MAAI,GAAS,KAAO,IACX,EAEF,UAAY,cAGO,GAE1B,MAAI,GAAW,QAAU,GAIrB,EAAW,QAAQ,OAAS,GAHvB,EAMF,EAAW,QAAQ,cAAe,SAAC,EAAG,GAAO,MAAA,GAAG,gBAIzD,GAAI,IAAyB,eAEQ,GAEnC,GAAI,GAAa,KACf,MAAO,MAET,GAAM,GAAsC,GAC5C,SAAK,UAAe,EAAS,eAC7B,EAAK,OAAY,EAAS,YACnB,EAcT,YAAuC,GAErC,GAAI,GAAU,MAAQ,MAAO,IAAW,SACtC,OACK,GAAI,MAAM,QAAQ,GACvB,EAAO,QAAQ,SAAA,GAAc,MAAA,IAA8B,SAG3D,QADM,GAAS,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAO,GACrB,AAAI,GAAS,MAAQ,MAAO,IAAU,UACpC,CAAI,CAAC,MAAM,QAAQ,IAAU,EAAM,OAAY,WAC3C,MAAO,GAAM,OAAa,SAC5B,EAAO,GAAS,EAAM,MAEtB,GAA8B,iBAqBpC,EACA,EACA,EACA,EAAgC,aAElC,GAJE,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,UAAgC,AAAA,IAAA,QAAA,GAAA,IAE9B,MAAO,IAAe,UACxB,GAAM,GAAe,EACjB,EAAE,OACN,GAAI,IAAgB,GAClB,EAAK,EAAc,WACV,IAAgB,IACzB,EAAK,GAAuB,WAE5B,EAAK,EAAc,GACf,GAAM,KACR,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAU;EAE7C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAKR,MAAO,QAGP,GAAM,GAAS,EACf,GAAI,EAAO,WAAgB,MAAQ,EAAO,QAAa,KACrD,KAAM,IAAI,GACH,EAAmB,6BACnB,MAAK,UAAU,GAAO;GACzB,sCAEN,GAAM,GAAY,EAAO,UACrB,EAAG,OAAE,EAAU,OAQnB,GAPA,AAAI,IAAa,GACf,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,GAAA,UAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IACD,AAAI,IAAa,IACtB,GAAA,EAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,IAEJ,GAAO,KACT,KAAM,IAAI,GACN,WAAW,EAAmB,KAAK,EAAS;EAE5C,WAAU,EAAmB,oCAC7B;EAEA,kBAAiB,EAAmB,+BACpC,yEAIN,GAAI,GAAc,MAQhB,OADM,GAAwB,GACZ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,GAAuB,GAEtD,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAsB,GAAO,EAAc,GAG7C,GAAM,GAAe,EAAO,OAC5B,EAAa,cAAmB,EAGhC,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAE9C,GAA8B,EAAO,QACrC,GAAM,GACF,EAAW,EAAK,EAAO,OAAW,EAAe,GACrD,UAAsB,GAAA,GAAO,GAEtB,OAMP,OADM,GAAmB,GAAA,GAAO,IACd,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAuB,GAAO,EAAc,GAK9C,GAAM,GAAY,GAAI,GAAI,EAAO,QACjC,UAAsB,GAAA,GAAO,GACtB,gBAUiB,EAAW,GACvC,MAAQ,GAAI,EAAK,GAAO,EAAI,EAAK,EAAI,cAQF,EAAW,GAC9C,MAAO,GAAK,GAAc,EAAG,eA2CL,GACxB,GAAI,GAAM,KACR,MAAO,GAIT,OAFM,GAAW,GAED,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAI,QAAQ,KAAO,IACrB,EAAI,KAAK,GAGb,MAAO,eASqB,GAC5B,GAAI,GAAO,KACT,KAAM,IAAI,GAAW,yBAAyB,KAAK,UAAU,IAE/D,OAAW,KAAO,GAChB,GAAI,EAAI,eAAe,GACrB,MAAO,GAGX,MAAO,eAWL,EAAkB,EAAe,GACnC,GAAI,GAAS,KACX,OAEF,GAAI,EAAO,QAAQ,GAAS,EAC1B,KAAM,IAAI,GAAc,EAAK,mBAAmB,EAAK,uBACjD,EAAM,mCAoBV,EAAQ,EAAsB,EAC9B,GAD8B,MAAA,KAAA,QAAA,GAAA,GAC9B,AAAA,IAAA,QAAA,GAAA,UACF,GAAO,GAAa,GACpB,GAAO,GAAa,GAEhB,MAAM,QAAQ,IAAM,EAAE,QAAU,GAAa,EAAE,QAAU,GACzD,EAAE,MAAM,SAAA,GAAK,MAAA,OAAO,KAAM,gBAWM,EAAwB,GAC5D,AAAI,MAAM,QAAQ,GAChB,GAAA,KAAK,OACD,EAAM,OAAS,EAAG,WAAM,MAAG,GAAI,qCACnC,EAAM,QACF,SAAC,EAAG,GAAM,MAAA,IAAsB,EAAG,WAAW,GAAI,GAAC,OAAO,MAE9D,EAAA,KAAK,OACD,OAAO,UAAU,IAAU,EAAQ,EACnC,WAAM,MAAA,YAAY,EAAI,sCACf,IAAuB,GAAM,mBAeL,GACrC,MAAI,KAAU,KACL,OACE,MAAM,QAAQ,GAChB,IAAM,EAAM,IAAI,SAAA,GAAK,MAAA,IAAuB,KAAI,KAAK,KAAO,IAC1D,MAAO,IAAU,SACnB,IAAI,EAAK,IAET,GAAG,cAcV,EAA8B,GAChC,GAAI,GAAW,EAAA,KAAK,MAChB,EACE,EAAK,kBAAC,GAAA,GAAA,EAAA,EAAA,EAAA,UAAA,OAAA,IAAA,EAAA,GAAA,UAAA,GACV,GAAM,GAAM,EAAA,KAAK,MACjB,MAAI,GAAM,EAAW,GAGrB,GAAW,EACX,EAAa,EAAC,MAAA,OAAI,IACX,GAET,MAAO,eASkC,GAEzC,MAAI,KAAmB,OACd,OAEL,IAAmB,SACd,SAEL,IAAmB,MACd,MAEF,KCvgBT,AAoBA,YAAqB,EAAW,GAC9B,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAI,EAAM,OAY1D,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAGE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GANyC,EAAA,cAAc,cA8BvD,GAAA,SAAA,GAA6B,EAAA,EAAA,GAQ3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAJQ,SAAA,gBAAkB,EAClB,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,YAAgB,EAAO,EAAG,EAAK,UAC/C,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,SAAU,KAAK,SAAU,KAAM,KAAK,OAtB9B,EAAA,UAAY,UAwB9B,GA1B6B,IA2B7B,EAAA,cAAc,cAAc,IAoB5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAK5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAFQ,SAAA,YAAc,EAG7B,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,KAAW,GAAY,EAAG,EAAK,WAG9D,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OAdL,EAAA,UAAY,WAgB9B,GAlB8B,IAmB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAIE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAHF,EAAA,UAAY,SAK9B,GAP4B,IAQ5B,EAAA,cAAc,cAAc,IAoC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAY9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KANQ,SAAA,gBAAkB,EAClB,EAAA,gBAAkB,EAClB,EAAA,YAAc,EACd,EAAA,YAAc,EAI7B,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,SACD,EAAK,UAAY,KAAO,EAAK,SAAW,EAAK,gBACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,YACjD,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,EAAK,cAGnD,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAY,EAAG,EAAK,MAC5B,EAAU,EAAA,IACZ,EAAA,IACI,EAAK,KAAM,EAAA,YAAgB,EAAO,EAAK,SAAU,EAAK,WAC1D,EAAA,IAAQ,EAAM,EAAK,KAAM,IAC7B,MAAO,GAAA,IAAQ,EAAG,EAAA,IAAQ,EAAS,EAAA,IAAQ,KAAW,QAI1D,EAAA,UAAA,UAAA,WACE,MAAO,CACL,SAAU,KAAK,SACf,SAAU,KAAK,SACf,KAAM,KAAK,KACX,KAAM,KAAK,OApCC,EAAA,UAAY,aAuC9B,GAzCgC,IA0ChC,EAAA,cAAc,cAAc,IAQrB,GAAM,IACwC,CAC/C,QAAW,UACX,WAAc,aACd,OAAU,SACV,SAAY,wBAGkB,GAElC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA0C,GAC1C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCpQjC,YAyBwB,GACtB,MAAO,IAAI,IAAQ,eAQI,GACvB,MAAO,IAAI,IAAS,iBASpB,MAAO,IAAI,gBAIc,GACzB,MAAO,IAAI,IAAW,0ECjDxB,AAqBO,GAAM,IAA2B,CAAC,gBAAiB,gBAO7C,GAA4B,CAAC,QAAS,OAAQ,UAI9C,GAAyB,CAAC,MAAO,OAIjC,GAAkC,CAAC,MAAO,MAAO,SAAU,OCpCxE,AAmBA,GAAM,IAA+B,GAAI,iBAET,GAC9B,GAA0B,GAA0B,aAAc,eAGnC,GAC/B,GAA0B,GAA2B,cAAe,eAGxC,GAC5B,GAA0B,GAAwB,WAAY,GAGhE,GAAM,IAA4B,GAC5B,GAAoB,gBAKG,EAAc,GACzC,GAAgB,KAAK,GACrB,IACE,GAAM,GAAS,IACf,UAAgB,MACT,QACA,GACP,SAAgB,MACV,GAOV,cACE,MAAI,IAAgB,SAAW,EACtB,GAEA,GAAgB,KAAK,IAAqB,eASjB,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,MAAO,MAA2B,cAaA,GAClC,GAAI,CAAC,GAAkB,GACrB,KAAM,IAAI,OAAM,6BAAgC,EAAa,KAE/D,AAAK,GAAQ,IAAI,IACf,GAAQ,IAAI,EAAY,GAE1B,GAAM,GAAQ,GAAQ,IAAI,GAG1B,GAFA,GAAQ,IAAI,EAAY,GAAQ,IAAI,GAAc,GAE9C,EAAQ,GACV,GAAM,GAAY,EAAU,IAAI,EAGhC,UAAQ,IAAI,EAAQ,GACb,MAEP,OAAO,GAIX,GAAM,IAAkB,GAAI,QAAO,+CAOD,GAChC,MAAO,CAAC,CAAC,EAAK,MAAM,ICjHtB,YA8B0B,GACxB,MAAO,KAAM,SAAS,EAAE,WAAY,gBAWlC,EAA4B,EAAgB,GAC9C,AAAI,GAAS,MACX,GAAQ,GAEV,AAAI,GAAO,MACT,GAAM,EAAM,QAId,OADI,GAAO,EACF,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,GAAQ,EAAM,GAEhB,MAAO,GAQT,YAAmB,GACjB,SAAQ,MAAM,QAAQ,GAAS,GAAI,cAAa,GAAS,EAClD,EAAA,SAAS,eAQE,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAQ1B,GAClB,MAAO,GAAA,IAAQ,GAAU,IAAQ,WAAW,eAsDxB,EAAe,GACnC,GAAI,EAAM,EACR,KAAM,IAAI,GAAW,QAAQ,EAAG,cAAc,EAAK,mBAGrD,OADM,GAAgB,GACb,EAAI,EAAO,EAAI,EAAK,EAAE,EAC7B,EAAI,KAAK,GAEX,MAAO,GChJT,YAwEqB,EAAW,GAC9B,MAAO,GAAE,OAAO,eASS,EAAW,GAAA,AAAA,IAAA,QAAA,GAAA,IACpC,GAAM,GAAW,EAAE,MAAM,QACzB,MAAI,GAAO,GACT,GAAO,EAAS,OAAS,EAAO,GAElC,EAAS,OAAO,EAAM,EAAG,GAClB,EAAE,QAAQ,eAcI,EAAW,GAChC,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,oDACA,SAAQ,EAAE,MAAM,OAAM,aAE5B,GAAM,GAAI,GAAW,EAAG,GACxB,MAAO,IAAK,EAAG,CAAC,EAAG,EAAG,kBASF,GACtB,GAAM,GAAW,CAAC,GAAqB,EAAE,QACzC,MAAO,GAAE,QAAQ,eAWU,GAC3B,GAAI,EAAE,MAAQ,EACZ,KAAM,IAAI,GACN,wDAAwD,EAAE,KAAI,KAEpE,GAAM,GAAW,CAAC,EAAE,MAAM,GAAI,GAAqB,EAAE,MAAO,IAC5D,MAAO,GAAE,QAAQ,eAYf,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,GAAI,CAAC,EAAM,EAAM,MAAM,SACnD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,GAC9B,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAO,EAAG,EAAG,GACjC,CAAC,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,MAAU,EAAmB,CAAC,EAAO,EAAG,EAAG,EAAG,GAAI,CACvD,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,SAEjE,GACH,MAAO,GAAA,MAAU,EAAO,CAAC,EAAO,EAAG,EAAG,EAAG,EAAG,GAAI,CAC9C,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAClE,EAAM,MAAM,aAGd,KAAM,IAAI,GACN,8DACA,IAAG,EAAM,sBAcjB,EAAe,EAAe,GAChC,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,GAAQ,CAAC,EAAM,MAAM,GAAI,QACjD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,GAC1B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,QAClC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAG,GAC7B,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,YAErD,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAejB,EAAe,EAAe,EAAc,GAC9C,MAAO,GAAA,KAAK,WACV,OAAQ,EAAM,UACP,GACH,MAAO,GAAA,QAAY,EAAmB,EAAO,OAC1C,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,GAC9B,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpC,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,QAER,GACH,OAAQ,OACD,GACH,MAAO,IAAoB,EAAO,EAAO,OACtC,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAO,EAAG,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,GAAI,EAAM,MAAM,SACpD,GACH,MAAO,GAAA,QACH,EAAmB,CAAC,EAAG,EAAG,EAAO,GACjC,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,GAAI,EAAM,EAAM,MAAM,SACpD,GACH,MAAO,IAAmB,EAAO,EAAO,WAExC,KAAM,IAAI,GACN,iDACA,IAAG,YAGX,KAAM,IAAI,GACN,6DACA,IAAG,EAAM,sBAWO,EAAmB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC7C,GAAI,GACJ,MAAI,GAAO,GACT,GAAO,EAAQ,GAAG,KAClB,AAAI,IAAS,EACX,EAAO,EAEP,EAAO,GAGX,AAAI,IAAS,EAAQ,GAAG,MAGtB,GAAO,IAGF,EAAA,OAAW,EAAS,eAUQ,EAAW,GAC9C,OAAQ,EAAE,UACH,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,QACjC,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,OACjD,GACH,MAAO,GAAA,SAAa,CAAC,EAAe,GAAgB,WAEpD,KAAM,IAAI,GACN,kDACA,iBAAgB,EAAE,oBAWP,EAAW,GAI9B,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAI,CAAC,IAEH,EAAE,OAAS,EAAE,OACf,KAAM,IAAI,GACN,0BAA0B,EAAE,OAAM,oBAClC,yCAAwC,EAAE,KAAI,MAEpD,MAAO,GAAA,KAAS,EAAG,eAgBjB,EAAc,EAAY,EAAc,EACxC,GADc,MAAA,KAAA,QAAA,GAAA,GAAY,AAAA,IAAA,QAAA,GAAA,GAErB,EAAA,aAAiB,EAAO,EAAM,EAAQ,EAAO,eAsBlD,EAAW,EAAW,EACtB,GACF,GAAK,EAAE,KAAO,GAAO,EAAE,KAAO,EAC5B,KAAM,IAAI,IACN,2CACA,uBAAsB,EAAE,MAAK,kBAAkB,EAAE,QAEvD,GAAI,EAAE,MAAQ,GACZ,GAAM,GAAW,EAAE,MAAM,MAAM,IAAI,GAC7B,EAAiB,EAAE,MAAM,MAAM,IAAI,GACzC,GAAI,IAAa,EACf,KAAM,IAAI,IACN,2CACA,yDACI,EAAE,MAAK,SACX,eAAc,EAAE,QAIxB,GAAK,EAAE,OAAS,GAAO,EAAE,OAAS,GAChC,GAAM,GAAa,GACb,EAAa,GAInB,MAAO,GAAA,MAAU,OAAO,CACtB,EAAC,EACD,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,SAIZ,GAAM,GAAa,EAAE,MAAM,QACrB,EAAW,EAAW,MAC5B,EAAI,EAAE,QAAQ,CAAC,GAAI,IAInB,GAAM,GAAS,EAAE,MAAM,QACjB,EAAW,EAAO,MAClB,EAAiB,EAAO,MACxB,EAAiB,EAAM,OAAA,CAAE,IAGzB,EAAO,MAAM,KAAK,CAAC,OAAQ,EAAE,MAAO,SAAC,EAAG,GAC5C,MAAI,KAAM,EACD,EAAE,KAAO,EACP,GAAK,EAAE,KAAO,EAChB,EAAI,EAEN,IAET,EAAI,EAAE,UAAU,GAAM,QAAQ,CAAC,EAAgB,KAG/C,GAAM,GAAkB,EAAU,OAAK,GACjC,EAAa,GACb,EAAa,GACnB,MAAO,GAAA,MACF,OAAO,CACN,EAAC,EACD,EAAC,EACD,WAAU,EACV,WAAU,EACV,KAAM,EAAO,GAAY,EAAE,KAAM,EAAM,MAAqB,KAC5D,WAAU,IAEX,QAAQ,gBAyDb,EAAmB,EAA4B,GACjD,MAAO,GAAA,KAAK,WACV,MAAI,OAAM,QAAQ,GAChB,EAAU,EAAA,SAAS,EAAS,SAE5B,EAAU,EAAQ,QAEb,EAAA,OAAW,EAAW,EAAS,iBASnB,GACrB,MAAO,GAAA,IAAQ,EAAG,GA+BpB,YAAqB,EAAe,EAAc,GAChD,GAAM,GAAY,EAAK,MAEvB,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,EACnC,KAAM,IAAI,GACN,+BAA+B,EAAK,KACpC,6BAA4B,IAGlC,GAAI,IAAU,GACZ,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,EAAG,IAErC,EAAK,QACR,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,GAAI,EAAU,KAEzD,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAG,EAAU,KAEpC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAG,IAElC,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,GAAI,EAAU,KAE3D,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAG,EAAU,KAEjC,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,IAAU,GACnB,GAAI,IAAe,gBACjB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,IAE/B,EAAK,QAAQ,CAAC,EAAG,EAAU,GAAI,EAAU,KAE7C,GAAI,IAAe,eACxB,MAAI,GAAU,SAAW,EAChB,EAAK,QAAQ,CAAC,EAAG,EAAG,EAAU,KAE9B,EAAK,QAAQ,CAAC,GAAG,OAAO,YAG1B,EAAQ,EACjB,MAAO,GAET,KAAM,IAAI,GAAW,sCAAsC,EAAK,kBAc9D,EAAW,EAAc,GAC3B,MAAO,GAAA,KAAK,WACV,MAAI,IAAc,MAChB,GAAa,MAEf,GAAgB,GAET,EAAE,IAAI,GAAY,EAAE,KAAM,EAAM,kBAUvB,EAAW,GAE7B,GAF6B,AAAA,IAAA,QAAA,GAAA,GAEzB,IAAU,EACZ,KAAM,IAAI,IACN,0CAA0C,EAAK,6BAGrD,MAAO,GAAA,IAAQ,eAWQ,GACvB,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAQ,EAAG,EAAA,IAAQ,GAAG,IAAI,kBAc1C,EAAW,EAAe,EAAuB,GACnD,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAG,EAAO,EAAY,iBAY1B,GAC1B,MAAO,GAAA,KAAK,WACV,GAAM,GAAI,EAAA,IAAQ,GAAI,EAAA,IAAQ,GAAI,IAClC,MAAO,GAAA,YAAgB,EAAG,EAAG,iBAiBD,EAAY,EAAc,GAAA,MAAA,KAAA,QAAA,GAAA,IACjD,EAAW,IAAM,IC7sB1B,AAgBO,GAAM,IAAwB,CAAC,QAAS,SAAU,UAO5C,GACT,CAAC,SAAU,UAAW,mBCxB1B,YAoB6B,GAC3B,GAA0B,GAAuB,UAAW,eAG5B,GAChC,GAA0B,GAA2B,eAAgB,GASvE,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DACS,SAAA,UAAA,4BAAP,WACE,MAAO,IAUT,EAAA,UAAA,UAAA,WACE,MAAO,IAEX,GAf0C,EAAA,cAAc,cAiBxD,GAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,MAAM,EAAO,IAHf,EAAA,UAAY,QAKrB,GAP2B,IAQ3B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAIE,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,KAAK,EAAO,IAHd,EAAA,UAAY,OAKrB,GAP0B,IAQ1B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,GACN,oDAAoD,GAE1D,GAAI,EAAK,QAAU,OACjB,KAAM,IAAI,GAAW,sCAAsC,GAE7D,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,IAAI,EAAA,OAAO,EAAK,OAAQ,EAAA,KAAK,EAAO,OAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,QApBT,EAAA,UAAY,WAuBrB,GAzB8B,IA0B9B,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GASjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,eAAiB,KACjB,EAAA,eAAiB,IAOxB,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,MAAO,GAAA,cAAc,EAAO,KAAK,OAAQ,KAAK,OAAQ,IAGxD,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,OAAQ,KAAK,OAAQ,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAnBxD,EAAA,UAAY,gBAqBrB,GAvBmC,IAwBnC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAShC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,uCAAuC,EAAK,KAGlD,MAAO,IAAe,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGnE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,eA2BrB,GA7BkC,IA8BlC,EAAA,cAAc,cAAc,IAW5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAPA,SAAA,aAAe,EACf,EAAA,eAAiB,IAOxB,EAAK,KAAO,EAAK,MAAQ,EAAK,aAC9B,EAAK,OAAS,EAAK,QAAU,EAAK,eAClC,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACN,0CAA0C,EAAK,KAErD,MAAO,GAAA,gBAAgB,EAAO,KAAK,KAAM,KAAK,OAAQ,EAAO,KAAK,OAGpE,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,KAAM,OAAQ,KAAK,OAAQ,KAAM,KAAK,OAzBpD,EAAA,UAAY,kBA2BrB,GA7BqC,IA8BrC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,KAAO,IAG9C,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,SAAW,GAAK,EAAM,KAAO,EAAM,GAC3C,KAAM,IAAI,GACN,wEAGJ,MAAO,GAAA,IAAI,EAAK,KAAM,EAAA,IAAI,EAAM,QAKtC,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,KAAM,KAAK,OApBd,EAAA,UAAY,WAsBrB,GAxB8B,IAyB9B,EAAA,cAAc,cAAc,IAU5B,YACI,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,gBAChB,GAAI,GACA,EAEJ,GADA,GAAgB,GACZ,EAAM,SAAW,EACnB,EAAQ,EAAM,GACd,EAAS,EAAM,WACN,CAAC,EAAG,EAAG,GAAG,QAAQ,EAAM,UAAY,IAC7C,GAAI,IAAe,iBACjB,GAAM,GAAqB,GAAU,EAAO,GAC5C,EAAQ,EAAM,GAAK,EACnB,EAAS,EAAM,GAAK,UACX,IAAe,gBACxB,GAAM,GAAqB,GAAU,EAAO,EAAG,EAAM,OAAS,GAC9D,EAAQ,EAAM,EAAM,OAAS,GAAK,EAClC,EAAS,EAAM,EAAM,OAAS,GAAK,QAGrC,GAAM,GAAY,GAAU,GAC5B,EAAQ,KAAK,KAAK,GAClB,EAAS,KAAK,KAAK,GAGrB,MAAO,CAAC,EAAO,GAiBjB,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAYnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAAI,EAAK,MAAQ,EACf,KAAM,IAAI,GACN,wCAAwC,EAAK,OAEnD,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAM,EAAK,MAC7C,EAAK,KAAO,EAAK,MAAQ,KAAO,QAAU,EAAK,KAC/C,GAAa,EAAK,MAClB,EAAK,aACD,EAAK,cAAgB,KAAO,SAAW,EAAK,aAChD,GAAkB,EAAK,cACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAO,GAAY,GACnB,EAAQ,EAAK,GACb,EAAS,EAAK,GAChB,EAAQ,KAAK,MASjB,GARA,AAAI,KAAK,OAAS,QAChB,GAAS,KAAK,IAAI,EAAG,GAChB,AAAI,KAAK,OAAS,SACvB,GAAS,KAAK,IAAI,EAAG,GAErB,GAAS,KAAK,IAAI,EAAI,GAAQ,GAAU,GAGtC,KAAK,eAAiB,UACxB,GAAM,GAAS,KAAK,KAAK,GAEzB,GADA,EAAQ,GAAS,UACb,IAAU,WAAa,IAAU,QACnC,KAAM,IAAI,IACH,KAAK,eAAc,2BAA2B,EAAK,KAE5D,MAAO,GAAA,gBAAgB,EAAO,EAAG,EAAQ,EAAO,KAAK,WAErD,GAAM,GAAQ,KAAK,KAAK,EAAI,GAC5B,MAAO,GAAA,cAAc,EAAO,CAAC,EAAO,EAAO,KAI/C,EAAA,UAAA,UAAA,WACE,MAAO,CACL,MAAO,KAAK,MACZ,KAAM,KAAK,KACX,aAAc,KAAK,aACnB,KAAM,KAAK,OAzDR,EAAA,UAAY,kBA4DrB,GA9DqC,IA+DrC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,gBAwBrB,GA1BmC,IA2BnC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAWhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,SACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAtBlB,EAAA,UAAY,eAwBrB,GA1BkC,IA2BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAI5B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,WAiBrB,GAnB8B,IAoB9B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAI7B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,YAiBrB,GAnB+B,IAoB/B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAI/B,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,SACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBiC,IAoBjC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAIhC,WAAY,SACV,GAAA,KAAA,KAAM,CACJ,MAAO,EACP,KAAM,QACN,aAAc,UACd,KAAM,GAAQ,KAAO,KAAO,EAAK,QACjC,KAGJ,SAAA,UAAA,aAAA,WAIE,MAAO,IAAgB,WAflB,EAAA,UAAY,cAiBrB,GAnBkC,IAoBlC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAO9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GATO,EAAA,aAAe,EAMtB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,KAAO,EAAK,KAEb,EAAK,MAAQ,KACf,KAAM,IAAI,IACN,2EAIR,SAAA,UAAA,MAAA,SAAM,EAAc,GAApB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,IAAoB,8BAEhC,AAAI,EAAM,GAAK,EAAM,GAAK,KACxB,QAAQ,KACJ,gEACA,eAAc,EAAM,GAAK,EAAM,GAAE,gBACjC,wBAIN,GAAM,GACF,EAAM,GAAK,EAAM,GAAK,CAAC,EAAM,GAAI,EAAM,IAAM,EAC3C,EAAI,GAAe,EAAiB,EAAG,EAAG,WAC5C,EAAI,EAAA,OAAO,YAAY,GAC3B,MAAI,GAAM,GAAK,EAAM,IACnB,GAAI,EAAE,aAED,EAAA,IAAI,EAAK,KAAM,MAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CACL,KAAM,KAAK,KACX,KAAM,KAAK,OA3CR,EAAA,UAAY,aA8CrB,GAhDgC,IAiDhC,EAAA,cAAc,cAAc,IAUrB,GAAM,IACyC,CAChD,SAAY,WACZ,aAAgB,eAChB,cAAiB,gBACjB,SAAY,WACZ,UAAa,YACb,SAAY,WACZ,YAAe,cACf,aAAgB,eAChB,KAAQ,OACR,WAAc,aACd,aAAgB,eAChB,cAAiB,gBACjB,gBAAmB,kBACnB,gBAAmB,kBACnB,MAAS,SAGf,YACI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGgB,GAEnC,MAAO,IAAqB,eAGC,GAE7B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EAIJ,GAAI,IAAc,eAChB,MAAO,IAAI,IACN,GAAI,IAAc,gBACvB,MAAO,IAAI,IACN,GAAI,IAAc,WACvB,MAAO,IAAI,IACN,GAAI,IAAc,YACvB,MAAO,IAAI,IACN,GAAI,IAAc,cACvB,MAAO,IAAI,IACN,GAAI,IAAc,eACvB,MAAO,IAAI,IAEX,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAuB,OAE3B,OAAI,aAAsB,IACxB,EAEA,GAAuB,GCxoBlC,cAkBE,MAAO,IAAI,kBASX,MAAO,IAAI,gBAQY,GACvB,MAAO,IAAI,IAAS,eAYQ,GAC5B,MAAO,IAAI,IAAc,eASE,GAC3B,MAAO,IAAI,IAAa,eAaM,GAC9B,MAAO,IAAI,IAAgB,eASJ,GACvB,MAAO,IAAI,IAAS,eAgBU,GAC9B,MAAO,IAAI,IAAgB,eAgBC,GAC5B,MAAO,IAAI,IAAc,eAgBE,GAC3B,MAAO,IAAI,IAAa,eAeD,GACvB,MAAO,IAAI,IAAS,eAeI,GACxB,MAAO,IAAI,IAAU,eAgBK,GAC1B,MAAO,IAAI,IAAY,eAYI,GAC3B,MAAO,IAAI,IAAa,eAWC,GACzB,MAAO,IAAI,IAAW,kPChNxB,AAmBA,GAAI,IAAsB,gBAGxB,MAAO,MAGT,GAAM,IAA2C,eAO1B,GAAA,MAAA,KAAA,QAAA,GAAA,IACrB,AAAM,IAAU,KACd,IAAa,GAAU,GAEzB,GAAa,IAAW,EACjB,EAAS,GAAa,GAAQ,WCrCvC,YAoBgC,GAC9B,MAAO,OAAM,QAAQ,IAAM,MAAM,QAAQ,EAAE,gBASV,GACjC,MAAI,GAAE,SAAW,EACR,GAEJ,MAAM,QAAQ,EAAE,IAGd,EAFE,CAAC,eAWwB,GAClC,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAG,SAAW,EAChB,KAAM,IAAI,GAAW,uCAAuC,EAAG,QAEjE,EAAI,EAAG,OAEP,GAAI,EAEN,MAAO,eAa0B,GACjC,GAAI,MAAM,QAAQ,IAAW,MAAM,QAAQ,EAAO,KAChD,GAAI,EAAO,SAAW,EACpB,SAAS,EACF,EAAO,GAEd,KAAM,IAAI,GAAW,iCAAiC,EAAO,YAG/D,OAAO,GC9EX,YAmBqC,GAEnC,OADI,GAAQ,EACS,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,AAAI,EAAO,MAAM,SAAW,EAC1B,GAAS,EAET,GAAS,EAAO,MAAM,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAG/C,MAAO,GC5BT,AAoBA,GAAM,IAA+B,yBAqCnC,WACI,EAAa,EACb,EAAqC,EACrC,GAFa,AAAA,IAAA,QAAA,GAAA,WACb,AAAA,IAAA,QAAA,GAAA,IAAqC,AAAA,IAAA,QAAA,GAAA,IACrC,AAAA,IAAA,QAAA,GAAA,MACF,KAAK,MAAQ,GAAgB,UAC7B,KAAK,MAAQ,EAAI,MACjB,KAAK,GAAK,KAEV,EAAO,GAAe,GACtB,KAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,cAErC,KAAK,WAAa,EAClB,KAAK,WAAa,EAElB,KAAK,IAAM,EAAA,SAAa,EAAK,KAAK,WAAY,KAAK,KAAM,KAAK,OAUhE,SAAA,UAAA,KAAA,WACE,YAAK,oBACE,KAAK,KAUd,EAAA,UAAA,MAAA,SAAM,GAEJ,YAAK,oBACL,GAAiB,KAAK,IAAK,GAE3B,AAAI,KAAK,IAAI,KAAO,EAAO,IACzB,MAAK,IAAI,OAAO,GAChB,AAAI,KAAK,YAAc,MACrB,KAAK,IAAI,OAAO,KAAK,WAAW,MAAM,KAAK,OAGxC,MAMT,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,KAAK,IAAI,WAGD,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,IAAI,WACX,KAAM,IAAI,OAAM,kBAAkB,KAAK,KAAI,0BAI/C,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,WAAa,EAClB,KAAK,IAAI,UAAY,mCAEzB,KAEA,YAA0B,EAAa,GACrC,GAAI,EAAE,MAAM,aAAe,EAAE,MAAM,WACjC,KAAM,IAAI,OACN,mBAAqB,KAAK,UAAU,EAAE,OAAS,QAC/C,KAAK,UAAU,EAAE,oBA0LK,GAC5B,MAAO,GAAG,IAAI,SAAA,GAAK,MAAA,GAAE,qBAYnB,GACF,EAAmB,QAAQ,SAAA,GACzB,GAAM,GAA0B,EAAiB,GACjD,EAAS,MAAM,EAAiB,MClVpC,kBAuEE,WAAY,GACV,KAAK,MAAQ,EAAK,MAClB,KAAK,MAAQ,EAAK,MAKlB,AAAI,EAAK,OAAS,KAChB,KAAK,KAAO,EAAK,MAAM,OAEvB,KAAK,KAAO,EAAK,KAEnB,KAAK,QAAU,EAAK,QACpB,KAAK,QAAU,EAAK,QACpB,KAAK,KAAO,EAAK,MAAQ,GAE7B,MAAA,oBA6CE,WACa,EAA0B,EAC5B,EAA6B,EAC3B,EAAkB,EAClB,GAHA,KAAA,MAAA,EAA0B,KAAA,MAAA,EAC5B,KAAA,YAAA,EAA6B,KAAA,OAAA,EAC3B,KAAA,SAAA,EACA,KAAA,kBAAA,EACX,KAAK,GAAK,KACV,AAAI,GAAQ,MACV,MAAK,aAAe,GAAoB,GACxC,KAAK,KAAO,GAAoB,KAAK,eAEvC,KAAK,KAAO,EAAM,OAEtB,MAAA,MA2DI,GAAc,EAsBlB,GAAA,WAwCE,WACI,EAEO,GAAA,KAAA,SAAA,EACT,KAAK,GAAK,KAQV,KAAK,cAAgB,EAAK,cAU1B,KAAK,cAAgB,EAAK,cAE1B,KAAK,YAAc,EAAK,YAExB,KAAK,cAAgB,EAAK,cAQ1B,KAAK,aAAe,EAAK,aAEzB,KAAK,cAAgB,EAAK,cAM1B,KAAK,WAAa,EAAK,WAEvB,KAAK,YAAc,EAAK,YAKxB,KAAK,YAAc,EAAK,YAExB,KAAK,aAAe,EAAK,aAGzB,OAAoB,GAAA,EAAA,EAAA,EAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MACX,EAAM,cAAc,KAAK,MAG7B,EAAK,cAAc,aAAa,KAAK,MAGvC,SAAA,UAAA,UAAA,WAEE,OADM,GAAyB,GACX,EAAA,EAAA,EAAA,KAAK,cAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,KACX,EAAa,KAAK,EAAM,MAExB,EAAa,KAAK,MAGtB,MAAO,CACL,cAAe,KAAK,cAAgB,KAAK,cAAc,KAAO,KAC9D,cAAe,EACf,YAAa,KAAK,YAClB,cAAe,KAAK,gBAG1B,KAkDI,GAAe,EAWnB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAmDlC,WAAY,GAAA,AAAA,IAAA,QAAA,GAAA,IAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAtBD,EAAA,UAAsB,KAEtB,EAAA,kBAA8B,GAQ5B,EAAA,UAAY,GAapB,EAAK,GAAK,KAEV,EAAK,oBAAsB,KAE3B,EAAK,UAAY,KACjB,EAAK,gBAAkB,GAGvB,EAAK,kBAAoB,GACzB,EAAK,qBAAuB,GAC5B,EAAK,QAAU,GACf,EAAK,SAAW,GAChB,EAAK,OAAS,GAMd,EAAK,aAAe,GACpB,EAAK,cAAgB,GAErB,GAAI,GAAO,EAAK,KAChB,GAAI,CAAC,GACH,GAAM,GAAS,EAAK,eACpB,EAAO,GAA0B,GAAU,IAAM,GAAO,GAM1D,GAJA,EAAK,KAAO,EAEZ,EAAK,WAAa,EAAK,WAAa,KAAO,GAAO,EAAK,UAEnD,EAAK,YAAc,MAAQ,EAAK,iBAAmB,MAKrD,GAAI,GAAe,OACnB,GAAI,EAAK,iBAAmB,KAC1B,EAAkB,EAAK,wBACd,EAAK,YAAc,MAC5B,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAkB,CAAC,GAAW,OAAO,EAAK,YAE5C,EAAK,gBAAkB,EAGvB,GAAI,GAAQ,EAAK,MACjB,AAAI,GAAS,MACX,GAAQ,EAAK,YAEf,AAAI,GAAS,MACX,GAAQ,WAEV,EAAK,MAAQ,EAGf,MAAI,GAAK,SAAW,KAClB,EAAK,eAAiB,EAAK,QAE3B,EAAK,eAAiB,KAKxB,EAAK,UAAY,KAEjB,EAAK,0BAA4B,KAYlB,SAAA,QAAjB,SAAyB,EAAc,GACrC,MAAO,GAAM,KAAO,OAAS,EAAU,YAUjC,EAAA,UAAA,eAAR,SAAuB,EAAmB,GACxC,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,mCACA,4BAA2B,EAAQ,MAEzC,GAAI,KAAK,aAAa,QAAU,EAC9B,KAAM,IAAI,GACN,gBAAgB,EAAQ,YAAY,EAAS,KAC7C,2BAA0B,KAAK,aAAa,OAAM,oBAExD,MAAO,MAAK,aAAa,IAY3B,EAAA,UAAA,WAAA,SAAW,GACT,MAAO,IACH,KAAK,eAAe,EAAW,SAAS,eAY9C,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,IACH,KAAK,eAAe,EAAW,UAAU,gBAgB/C,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,uHAIC,GAAI,KAAK,aAAa,SAAW,EACtC,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0CAEN,MAAO,IACH,KAAK,eAAe,EAAG,SAAS,+CActC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,aAAa,SAAW,EAC/B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,0BAEN,GAAI,KAAK,aAAa,OAAS,EAC7B,KAAM,IAAI,IACN,SAAS,KAAK,KACd,yHAKN,MAAO,IACH,KAAK,eAAe,EAAG,UAAU,gDAGvC,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAQd,EAAA,UAAA,gBAAA,WAKE,MAAO,MAAK,OAAO,IAAI,SAAA,GAAU,MAAA,QAGnC,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,0CAGd,OAAA,eAAI,EAAA,UAAA,QAAK,KAAT,WACE,MAAO,MAAK,YAGd,SAAU,GACR,KAAK,OAAS,mCAGhB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,kBAAkB,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,IAClD,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAI,MAAK,WACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,GAAE,YAErC,QAIX,SAAqB,GACnB,KAAK,kBAAoB,mCAG3B,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAI,MAAK,UACA,KAAK,kBAAkB,OAAO,SAAA,GAAK,MAAA,CAAC,EAAE,YACxC,OAAO,KAAK,sBAEV,KAAK,kBAAkB,OAAO,KAAK,2BAI9C,SAAwB,GACtB,KAAK,qBAAuB,mCAO9B,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAG3C,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WACE,MAAO,MAAK,2CAUd,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,SACR,KAAM,IAAI,OACN,yEAiBE,EAAA,UAAA,yBAAV,SAAmC,GAGjC,GADA,EAAS,GAAqB,GAC1B,KAAK,WAAa,MAAQ,KAAK,UAAU,SAAW,EACtD,OAEF,GAAM,GAAY,GAAqB,KAAK,WAC5C,GAAI,EAAO,SAAW,EAAU,OAC9B,KAAM,IAAI,GACN,SAAS,KAAK,KAAI,YAAY,EAAU,OAAM,YAC9C,oBAAmB,EAAO,OAAM,oBAChC,oBAAmB,IAEzB,OAAS,GAAa,EAAG,EAAa,EAAO,OAAQ,KACnD,GAAM,GAAI,EAAO,GACX,EAAkB,EAAU,GAClC,GAAI,GAAQ,KACV,SAIF,GAAM,GAAO,EAAE,KACf,GAAI,EAAK,MAAQ,MACX,IAAS,EAAK,KAChB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,KAC3D,kBAAiB,EAAK,KAAI,gBAAgB,IAGlD,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,IAG3D,GAAI,EAAK,SAAW,MACd,EAAO,EAAK,QACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KACvD,wBAAuB,EAAK,QAAO,gBAAgB,EAAI,MAK/D,GAAI,EAAK,OAAS,MACZ,EAAE,QAAU,EAAK,MACnB,KAAM,IAAI,GACN,SAAS,EAAU,+BAA+B,KAAK,KAAI,IAC3D,qBAAoB,EAAK,MAAK,iBAAiB,EAAE,MAAK,MAK9D,GAAI,EAAK,MACP,GAAM,GAAS,EAAE,MACjB,OAAW,KAAO,GAAK,MACrB,GAAM,GAAO,OAAO,GACd,EAAQ,EAAK,KAAK,GAIlB,EACF,GAAQ,EAAI,EAAO,GAAQ,EAAO,EAAO,OAAS,GACtD,GAAI,GAAS,MAAQ,CAAC,EAAO,MAAM,QAAQ,KAAkB,GAC3D,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,mBAAmB,EAAI,uBACnC,eAAc,EAAK,kBAAkB,EAAM,OAMrD,GAAI,EAAK,OAAS,KAChB,OAAS,GAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAU,EAAK,MAAM,GACrB,EAAM,EAAE,MAAM,GACpB,GAAI,GAAW,MAAQ,GAAO,MACxB,IAAY,EACd,KAAM,IAAI,GACN,SAAS,EAAU,+BAChB,MAAK,KAAI,oBAAoB,EAAK,MAAK,MAC1C,gBAAe,EAAE,MAAK,SAgBtC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,IAGC,EAAA,UAAA,eAAV,SAAyB,EAAyB,GAChD,AAAI,KAAK,WAAa,MACpB,KAAK,UAAU,EAAQ,IAS3B,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,UAAY,GAOnB,EAAA,UAAA,cAAA,WACE,KAAK,UAAY,MAwEnB,EAAA,UAAA,MAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAGE,EAAS,GAAU,GAEnB,KAAK,oBAML,OAHM,GAAa,GAAqB,GAEpC,EAAiB,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,CAAE,aAAiB,MACrB,EAAiB,GACjB,OAIJ,OADI,GAAkB,GACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,YAAiB,KACnB,EAAkB,GAClB,OAIJ,GAAI,IAAmB,EACrB,KAAM,IAAI,GACN,mEAKN,MAAO,IAAU,KAAK,KAAM,WAE1B,GAAI,CAAC,EAAK,OAKR,EAAK,yBAAyB,GAI9B,OADM,GAAuB,GACT,EAAA,EAAA,EAAA,GAAqB,GAArB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,KAAK,EAAM,OAEzB,EAAK,MAAM,GAA+B,IAC1C,EAAK,MAAQ,GAGb,AAAI,EAAK,gBACP,EAAK,WAAW,EAAK,gBAGvB,AAAI,EAAK,YAAc,MAAQ,GAI7B,GAAK,UAAY,GAcrB,GANA,EAAK,yBAAyB,GAM1B,GAUF,OATI,GAAS,EAAK,KAAK,EAA6B,GAK9C,EAAuB,GAAqB,GAC5C,EAA2B,GAGnB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAT,GAAI,GAAC,EAAA,GACR,AAAI,EAAW,QAAQ,KAAO,IAC5B,GAAI,EAAE,SAER,EAAe,KAAK,GAItB,GAFA,EAAS,GAA+B,GAEpC,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAKN,MAAO,QAEP,GAAM,GAAa,GAAkB,GAC/B,EAAc,EAAK,mBAAmB,GACxC,EAAM,OACJ,EAAc,GAAiB,GAgCrC,GA/BA,EAAK,6BACD,MAAM,QAAQ,GAAU,EAAW,GACX,GAE5B,AAAI,GAAe,MAAQ,EAAY,OAAS,GAC5C,MAAM,QAAQ,EAAY,IAE5B,EAAU,EACI,IACG,SAAC,EAAO,GAAU,MAAA,IAAI,IAClB,EAAa,EAAO,EACpB,GAAqB,GAAS,EAAQ,EAAK,KAC3C,KAErB,EAAS,GAAI,IACT,EAAa,EAAsB,EACnC,GAAqB,GAAS,EAAQ,EAAK,MAUjD,EAAK,eACD,EAA6C,EAAQ,KAAM,KAC3D,EAAY,EAAa,GAC7B,EAAK,YAED,EAAK,qBAAuB,KAC9B,KAAM,IAAI,IACN,qFAIN,MAAO,OAYH,EAAA,UAAA,6BAAV,SAAuC,GACrC,GAAI,KAAK,iBAAmB,KAC1B,OACK,GAAI,EAAW,SAAW,KAAK,gBAAgB,OACpD,QAAQ,KACJ,iDACG,MAAK,UAAU,GAAW,iCAC7B,qBAAoB,KAAK,UAAU,KAAK,iBAAgB,MACxD,iBAAgB,KAAK,YAEzB,GAAI,GAAc,GAClB,KAAK,gBAAgB,QAAQ,SAAC,EAAW,GACvC,AAAI,GAAa,MAAQ,EAAW,IAAM,MACtC,EAAW,KAAO,GACpB,GAAc,MAGlB,AAAI,GACF,QAAQ,KACJ,iCACA,KAAI,KAAK,UAAU,GAAW,eAC9B,mCAAkC,KAAK,KAAI,MAC3C,IAAG,KAAK,UAAU,KAAK,qBAiBjC,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,GAAI,KAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,EAC5D,KAAM,IAAI,IACN,aAAa,KAAK,KAAI,gEAI5B,OADM,GAA4B,GACf,EAAA,EAAA,EAAA,KAAK,aAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAAc,KAAK,UAAU,EAAK,cACxC,AAAI,EAAgB,QAAQ,KAAiB,IAC3C,EAAgB,KAAK,GAGzB,GAAI,EAAgB,SAAW,GAC7B,GAAM,GAAe,KAAK,aAAa,GAAG,aAC1C,MAAI,OAAM,QAAQ,IAAiB,MAAM,QAAQ,EAAa,KAC1D,EAAa,SAAW,EAClB,EAAyB,GAE1B,MAIT,MAAM,IAAI,IACN,aAAa,KAAK,KAAI,+JAiB9B,EAAA,UAAA,YAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,sCAAsC,KAAK,KAAI,uFAIrD,MAAO,IAAoC,KAAK,UAclD,EAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAWf,EAAA,UAAA,WAAA,SAAW,GAAA,MAAA,KAAA,QAAA,GAAA,IACF,GAAc,EAAgB,KAAK,iBAAmB,KAAK,UAepE,EAAA,UAAA,WAAA,SAAW,GAAX,GAAA,GAAA,KACE,EAAA,KAAK,WACH,GAAM,GAAS,EAAK,QACpB,GAAI,EAAO,SAAW,EAAQ,OAK5B,KAAM,IAAI,GACN,4CAA4C,EAAK,KAAI,KACrD,iCAAgC,EAAQ,OAAM,MAC9C,gCAA+B,EAAO,OAAM,cAC5C,sBAAqB,EAAO,QAElC,GAAI,EAAO,SAAW,EACpB,OAIF,OAFM,GAAoD,GACpD,EAAc,GAAc,GACzB,EAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,GACxC,GAAM,GAAK,EAAY,GACjB,EAAI,EAAO,GACX,EAAI,EAAQ,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAG,MAAO,EAAE,OAChC,KAAM,IAAI,GACN,sBAAsB,EAAG,MAAK,IAC9B,8CAA6C,EAAE,QAErD,EAAkB,KAAK,CAAC,EAAG,IAE7B,GAAc,MAmBR,EAAA,UAAA,UAAV,SACI,EAAc,EAAc,EAAkB,EAC9C,EAA2B,EAC3B,GAEF,GAAI,KAAK,kBAAkB,QAAQ,KAAU,GAC3C,KAAM,IAAI,GACN,yBAAyB,EAAI,cAAc,KAAK,MAEtD,KAAK,kBAAkB,KAAK,GAE5B,AAAI,GAAS,MACX,GAAQ,WAGV,AAAI,KAAK,2BACP,GAAc,GAAe,UAE/B,GAAM,GAAY,EAAY,MAAM,EAAO,GACrC,EACF,GAAI,IAAc,EAAW,EAAO,EAAM,EAAW,GACzD,SAAU,UAEV,AAAI,GAAe,MACjB,KAAK,QAAQ,WAAM,MAAA,GAAY,MAAM,EAAO,UAE9C,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,EACF,KAAK,kBAAkB,KAAK,GAE5B,KAAK,qBAAqB,KAAK,GAE1B,GAaT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,KAAK,0BAA4B,GAWnC,EAAA,UAAA,QAAA,SAAQ,SACN,GAAI,GAAU,MAAQ,MAAM,QAAQ,IAAW,EAAO,SAAW,EAC/D,OAGF,EAAS,GAAqB,GAC9B,AAAI,KAAK,UAAY,QAAa,KAAK,UAAY,MACjD,GAAA,KAAK,QAAO,KAAI,MAAA,EAAI,IAexB,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAYT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,GAAI,CAAC,KAAK,iBACR,GAAI,GAAQ,KACV,GAAI,MAAM,QAAQ,GAChB,EAAK,QAAQ,SAAA,GACX,GAAI,GAAe,KACjB,KAAM,IAAI,WACN,SAAS,EAAK,KAAI,iEAK1B,MAAM,IAAI,WACN,SAAS,KAAK,KAAI,2DAK1B,MAAO,MAIT,MAAO,IAeD,EAAA,UAAA,eAAR,SACI,EACA,EACA,EAA6B,EAC7B,EAA4B,EAC5B,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAM,GACF,GAAqB,GACzB,EAAgB,GAAqB,GACrC,EAAa,GAAqB,GAClC,EAAc,GAAqB,GACnC,EAAc,GAA+B,GAC7C,EAAe,GAA+B,GAM9C,OAHM,GAAyB,GACzB,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GAKV,EAAc,KAAK,EAAE,aACrB,EAAY,KAAK,EAAE,WACnB,EAAc,KAAK,EAAE,aAMvB,GAAI,IACA,CACE,cAAe,KACf,cAAa,EACb,YAAW,EACX,cAAa,EACb,aAAc,EACd,cAAa,EACb,WAAU,EACV,YAAW,EACX,YAAW,EACX,aAAY,GAEd,GAGJ,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,IAExC,EAAc,GAAG,YAAc,KAC/B,EAAc,GAAG,UAAY,KAAK,aAAa,OAAS,EACxD,EAAc,GAAG,YAAc,GAyBnC,EAAA,UAAA,UAAA,WACE,GAAM,GACyB,CAAC,KAAM,KAAK,KAAM,UAAW,KAAK,WACjE,MAAI,MAAK,iBAAmB,MAC1B,GAAO,gBAAqB,KAAK,iBAEnC,AAAI,KAAK,OAAS,MAChB,GAAO,MAAW,KAAK,OAElB,GAQC,EAAA,UAAA,eAAV,WACE,YAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,GAAO,YAC/B,KAAK,QAAQ,QAGZ,EAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,UAAU,KAAK,KAAI,2BAkCvC,EAAA,UAAA,QAAA,WACE,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,uCAIvC,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,OACN,wBAAwB,KAAK,KAAI,sCAIvC,KAAK,oBAEL,GAAI,GAAuB,EAC3B,MAAI,EAAE,KAAK,YAAc,GACvB,GAAuB,KAAK,kBAGvB,CAAC,qBAAsB,KAAK,UAAW,qBAAoB,IAEtE,GAxoCoC,EAAA,cAAc,cAopClD,YAA2B,GAEzB,EACI,GAAqB,GAEzB,OADM,GAAkB,GACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAO,KAAK,EAAE,OAEhB,MAAO,IAA+B,GAYxC,YAA0B,GAExB,MAAO,sBAeL,EAAwB,EACxB,GAKF,GAJA,AAAI,IAAS,MAAS,GAAa,MAAQ,EAAY,IACrD,GAAQ,EAAO,YACf,EAAY,EAAO,WAEjB,EAAM,aAAa,SAAW,EAChC,MAAO,CAAC,GAER,GAAM,GAAO,EAAM,aAAa,GAChC,GAAI,EAAK,cAAc,SAAW,EAChC,MAAO,GAAK,aAGZ,OADM,GAAkC,GAC/B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,IAM7C,OALM,GAAI,EAAK,aAAa,GACtB,EAAQ,EAAK,cAAc,GAC3B,EAAY,EAAK,YAAY,GAC7B,EAAkB,GAAgB,EAAG,EAAO,GAElC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAc,QAAQ,KAAO,IAC/B,EAAc,KAAK,GAIzB,MAAO,GCvmDb,AA4CA,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CACJ,MAAO,EAAK,MACZ,KAAM,EAAK,MAAQ,KAAO,EAAK,KAAO,GAAO,SAAS,cACtD,KAaF,GAXA,AAAI,EAAK,WAAa,MACpB,GAAK,UAAY,MAEnB,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,IAGhB,EAAK,UAAY,GACjB,EAAK,MAAQ,GACb,EAAK,OAAS,EAAK,OAEf,EAAK,YAAc,MAAQ,EAAK,iBAAmB,KACrD,KAAM,IAAI,GACN,qGAGN,GAAI,GAAkB,EAAK,gBAC3B,GAAI,GAAmB,MACrB,GAAI,EAAK,YAAc,KACrB,KAAM,IAAI,GACN,iFAGJ,EAAkB,CAAC,EAAK,WAAW,OAAO,EAAK,oBAI7C,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,yFAKR,GAAM,GAAQ,EAAK,OAAS,UAE5B,EAAK,gBAAkB,EACvB,EAAK,MAAQ,EAEb,EAAK,UAAY,CAAC,CAAC,MAAO,IAE1B,GAAM,GAAc,GAAI,IACpB,EAAK,MAAO,EAAK,gBAAiB,EAAM,GAAI,GAAI,EAAK,MACzD,SAAY,UAAY,EACxB,EAAY,YAAc,EAK1B,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,CAAC,GACf,cAAe,CAAC,GAChB,WAAY,CAAC,MACb,YAAa,CAAC,MACd,YAAa,CAAC,GACd,aAAc,CAAC,OAInB,SAAA,UAAA,MAAA,SACI,EACA,GACF,KAAM,IAAI,GACN,+BACA,kDAAiD,KAAK,QAG5D,EAAA,UAAA,QAAA,WAEE,MAAO,CAAC,qBAAsB,KAAK,UAAW,qBAAsB,IAGtE,EAAA,UAAA,UAAA,WACE,MAAO,CACL,gBAAiB,KAAK,gBACtB,MAAO,KAAK,MACZ,OAAQ,KAAK,OACb,KAAM,KAAK,OAzFC,EAAA,UAAY,aA4F9B,GA9FgC,IA+FhC,EAAA,cAAc,cAAc,gBAmCN,GACpB,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAC/C,KAAM,IAAI,OACN,gIAKN,GAAI,EAAO,YAAc,MAAQ,EAAO,OAAS,KAE/C,KAAM,IAAI,GACN,oFAGN,GAAI,GAAa,EAAO,WACxB,AAAI,EAAO,OAAS,MAAQ,GAAc,MACxC,GAAa,CAAC,MAAM,OAAO,EAAO,QAGpC,GAAI,GAAQ,EAAO,MACnB,AAAI,GAAS,MACX,GAAQ,WAGV,GAAM,GAAa,GAAI,IAAW,CAChC,gBAAiB,EACjB,KAAM,EAAO,KACb,MAAK,EACL,OAAQ,EAAO,SAGX,EAAU,EAAW,aAAa,GAAG,cAC3C,MAAO,GAAQ,GC9MjB,YA0B2C,iHACzC,GAAI,GAAQ,KACV,MAAA,CAAA,GAEI,EAA+D,GAC/D,EAAiB,GACjB,EAA6B,GACnC,IAAW,IAAO,GACV,EAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACb,GAAc,EACpB,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAiB,KAAK,UAGtB,GAAS,OAAS,EACL,CAAA,EAAM,QAAQ,IAAI,IAD/B,CAAA,EAAA,UAEF,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAK,EAAK,IAAM,EAAO,GAAG,GAG5B,EAAA,QAAQ,+CAUyB,GACnC,GAAI,GAAQ,KACV,OAEF,OAAW,KAAO,IAChB,GAAM,GAAQ,EAAK,GACnB,AAAI,MAAO,IAAU,UACnB,EAAM,WCjEZ,AAoBA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,QAAA,GAAA,YAFU,IAAA,IAAqB,KAM1B,GAAM,IAAyB,IA0BtC,GAAA,WAAA,aAEE,KAAA,eAAkC,KAMlC,SAAA,UAAA,UAAA,SAAU,GACR,KAAK,OAAS,GAGV,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,EAAe,qFAE5B,EAAA,UAAA,WAAN,SAAiB,EAAe,qFAE1B,EAAA,UAAA,aAAN,SAAmB,qFAEb,EAAA,UAAA,WAAN,SAAiB,qFASjB,EAAA,UAAA,SAAA,SAAS,KAGX,mBAqBE,WAAY,EAA4B,GAAA,AAAA,IAAA,QAAA,GAAA,IAGtC,AAAI,GAAa,MACf,GAAY,IAEd,KAAK,UAAY,EACjB,KAAK,YAAc,EAGrB,SAAA,UAAA,OAAA,SAAO,GACL,KAAK,UAAU,KAAK,IAGtB,EAAA,UAAA,UAAA,SAAU,GACR,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,UAAU,KAIvB,EAAA,UAAA,SAAA,SAAS,GACP,OAAuB,GAAA,EAAA,EAAA,KAAK,UAAL,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAS,SAAS,KAShB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,aAAN,SAAmB,EAAe,uGAChC,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,EAAO,KADA,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAUnB,EAAA,UAAA,WAAN,SAAiB,EAAe,uGAC9B,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,EAAO,KADE,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,aAAN,SAAmB,uGACjB,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,aAAa,KADO,CAAA,EAAA,UACnC,EAAA,wBADqB,wCASnB,EAAA,UAAA,WAAN,SAAiB,uGACf,AAAI,GAAQ,MACV,GAAO,QAEc,EAAA,KAAK,iCAAL,GAAA,EAAA,OAAZ,GAAQ,EAAA,GACjB,CAAA,EAAM,EAAS,WAAW,KADS,CAAA,EAAA,UACnC,EAAA,wBADqB,wCAI3B,KAOA,GAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,mBACE,GAAA,KAAA,OAAO,KAGH,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,OAAS,YAGV,EAAA,UAAA,WAAN,SAAiB,EAAe,yFAC9B,AAAI,GAAQ,MACV,GAAO,IAEH,EAAY,EAAK,MAAW,KAAO,EAAI,EAAK,KAClD,KAAK,MAAQ,aACF,GACT,GAAM,GAAQ,EAAK,GACnB,GAAI,MAAO,IAAU,SACnB,AAAK,EAAK,OAAO,eAAe,IAC9B,GAAK,OAAO,GAAO,GAErB,EAAK,OAAO,GAAO,EAAK,OAAO,GAAiB,EAAQ,OAExD,GAAI,GAAkB,OACtB,AAAI,IAAO,GAAK,OACd,EAAqB,EAAK,OAAO,GAEjC,EAAK,OAAO,GAAO,EAErB,GAAM,GACF,EAAA,KAAK,WAAM,MAAA,GAAA,IAAK,EAAK,OAAO,GAAO,EAAA,IAAI,EAAO,MAClD,EAAK,OAAO,GAAO,EACnB,AAAI,GAAsB,MACxB,EAAmB,mBAlBzB,IAAW,IAAO,KAAP,kBAwBP,EAAA,UAAA,WAAN,SAAiB,EAAe,2FAC9B,GAAI,GAAQ,KACV,eAAW,GACT,GAAI,EAAK,OAAO,IAAQ,sBAGxB,AAAI,MAAO,GAAK,OAAO,IAAS,SAC9B,EAAK,GAAO,EAAK,OAAO,GAAiB,EAAK,KAE9C,EAAA,KAAK,WACH,GAAM,GAAc,EAAA,IAAI,EAAA,IAAI,EAAG,EAAK,MAAO,EAAK,OAAO,IACvD,EAAK,GAAO,EACX,EAAK,OAAO,GAAgB,UAC7B,EAAA,KAAK,EAAK,cAXhB,EAAA,EAAkB,EAAA,KAAK,OAAO,QAAZ,EAAA,EAAA,OAAA,IAAP,EAAG,EAAA,KAAH,kBAiBjB,GA9DgC,mBAqEH,EAAA,EAAA,GAA7B,4DAIQ,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,MAAQ,GACb,KAAK,QAAU,YAGX,EAAA,UAAA,WAAN,SAAiB,EAAe,4EAC9B,AAAI,GAAQ,MACV,GAAO,IAET,KAAK,MAAM,KAAK,GAChB,IAAW,IAAO,GAChB,AAAI,KAAK,QAAQ,IAAQ,MACvB,MAAK,QAAQ,GAAO,IAEtB,KAAK,QAAQ,GAAK,KAAK,EAAK,mBAO1B,EAAA,UAAA,SAAN,6HACQ,EAA+D,GAC/D,EAAiB,GACjB,EAAoB,GAC1B,IAAW,IAAO,MAAK,QAErB,IADM,EAAa,KAAK,QAAQ,GACvB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,AAAI,MAAO,GAAW,IAAO,UACrB,GAAc,EAAW,GAC/B,EAAS,KAAK,EAAY,QAC1B,EAAK,KAAK,GACV,EAAQ,KAAK,IAIJ,MAAA,CAAA,EAAM,QAAQ,IAAI,WACjC,IADM,EAAS,EAAA,OACN,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EAC7B,EAAkB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IACtD,EAAgB,UAChB,KAAK,QAAQ,EAAK,IAAI,EAAQ,IAAM,EAAO,GAAG,mBAGpD,GA/C6B,mBA8DO,EAAA,EAAA,GAiBlC,WAAY,EAA0B,GAAtC,GAAA,GACE,EAAA,KAAA,OAAO,KAKP,GARM,EAAA,aAAe,EAIrB,EAAK,WAAa,GAAc,OAChC,AAAI,EAAK,aAAe,QACtB,GAAK,WAAa,IAEhB,EAAK,aAAe,SAAW,EAAK,SAAW,KACjD,KAAM,IAAI,OACN,mHAGN,MAAI,GAAA,KAAK,SAAS,EAAK,aAGrB,GAAK,UAAY,GACb,EAAK,UAAU,KAAK,GAAO,EAAK,aAEtC,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,WAAa,EAAK,aACvB,EAAK,SAAW,EAAK,WACrB,EAAK,MAAQ,EAAK,UAGd,SAAA,UAAA,UAAN,SAAgB,EAAe,EAAe,yGACtC,GAAgC,GAClC,KAAK,OAAS,KAChB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,MAAM,EAAO,EAAO,qBAEnC,SAAG,KAAK,EAAA,aACR,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAChC,MAAK,aAAe,EAChB,KAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,SACtB,EAAG,KAAK,EAAA,aAEV,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,EAAe,mGAC5B,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,EAAO,WAA7B,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,EAAe,yGACxB,GAAgC,GAClC,KAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,EAAA,OACA,EAAG,KAAK,KAAK,SAAS,EAAO,qBAE/B,MAAI,MAAK,aAAe,QACtB,EAAG,KAAK,EAAA,aACH,AAAI,EAAA,KAAK,SAAS,KAAK,aAC5B,EAAG,KAAK,KAAK,UAAU,KAAK,aAAc,EAAO,IAEnD,CAAA,EAAM,QAAQ,IAAI,WAAlB,SAAA,iBAGI,EAAA,UAAA,aAAN,SAAmB,mGACb,MAAK,YAAc,KACrB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,WAAW,WAAtB,EAAA,wCAIE,EAAA,UAAA,WAAN,SAAiB,mGACX,MAAK,UAAY,KACnB,CAAA,EAAM,GAAqB,IADzB,CAAA,EAAA,UACF,SAAA,OACA,CAAA,EAAM,KAAK,SAAS,WAApB,EAAA,wCAGN,GA3GoC,gBAiHhC,EAEA,GAIF,GAHA,AAAI,GAAa,MACf,GAAY,IAEV,YAAqB,IACvB,MAAO,CAAC,GAEV,GAAI,MAAM,QAAQ,IAAc,EAAU,YAAc,IACtD,MAAO,GAGT,GAAM,GACF,GAAqB,GACzB,MAAO,GAAgB,IACnB,SAAA,GAAkB,MAAA,IAAI,IAAe,EAAgB,KAW3D,GAAA,IAAA,WAOE,cAcO,SAAA,4BAAP,SACI,EAAwB,GAC1B,EAAA,KAAK,OACD,GAAkB,GAAK,OAAO,UAAU,GACxC,WAAM,MAAA,sDACF,YAAW,KACnB,EAA4B,kBAAkB,GAC9C,AAAI,EAA4B,aAAa,IAAmB,MAC9D,GAA4B,aAAa,GAAkB,IAE7D,EAA4B,aAAa,GAAgB,KACrD,IAGS,EAAA,kBAAf,SAAiC,GAE/B,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAe,EAA4B,aAAa,CAAC,GAC/D,EAAa,QAAQ,SAAA,GACnB,GAAI,IAAS,EACX,KAAM,IAAI,GAAW,uCASZ,EAAA,MAAjB,WACE,EAA4B,aAAe,IAWtC,EAAA,gBAAP,SAAuB,GACrB,GAAM,GAA0C,GAChD,OAAW,KAAa,GAA4B,cAClD,GAAM,GAAQ,CAAC,EACf,AAAI,GAAkB,GACpB,EAAa,KAAI,MAAjB,EAAqB,EAA4B,aAAa,IAGlE,MAAO,GAAa,IAAI,SAAA,GAAQ,MAAA,IAAI,MArEvB,EAAA,aAC6C,GAsE9D,iBAGI,EAA2B,EAAgC,EAC3D,EAAsB,EAAyB,EAC/C,EAAmB,EACnB,GACF,GAAM,GAAU,GAAI,IACd,EAAe,CACnB,GAAI,YAAiB,GAA4B,gBAAgB,IAEnE,AAAI,GAAa,MACf,EAAgB,KAAI,MAApB,EAAwB,GAE1B,EAAgB,KAAK,GACrB,GAAM,GAAe,GAAI,IAAa,GAMtC,SAAa,UAAU,CACrB,OAAM,EACN,aAAY,EACZ,QAAS,EACT,MAAO,EACP,UAAS,EACT,QAAO,EACP,aAAY,EACZ,QAAS,IAEJ,CAAC,aAAY,EAAE,QAAO,GCnlB/B,YA2BI,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAgB,IAChB,AAAA,IAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,QAAS,GChC9B,YAwB4B,EAAW,GACrC,MAAO,GAAA,KAAK,WACV,AAAI,EAAE,QAAU,WACd,GAAI,EAAE,OAAO,YAEf,GAAM,GAAY,EAAA,IAAQ,GAAS,GAAI,EAAM,IACvC,EAAgB,EAAA,KAAS,EAAU,MAAO,MAC1C,EAAO,EAAA,KAAS,EAAA,QAAY,EAAW,IAC7C,MAAO,GAAA,IAAQ,EAAG,iBAIW,EAAe,GAC9C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAO,IAAS,kBAG5B,EAAe,GAC/C,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,KAAS,EAAA,IAAQ,EAAA,IAAQ,EAAO,IAAS,kBAIzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,EAAA,IAAQ,EAAO,GACtB,EACF,EAAA,YAAgB,EAAA,IAAQ,GAAQ,KAAW,OAAO,WAChD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAM,IACxC,MAAO,GAAA,IAAQ,IAAK,EAAA,KAAS,EAAW,mBAKxC,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAE9B,EAAc,EAAA,YAAgB,EAAO,KAAW,OAAO,WACvD,EAAY,EAAA,IAAQ,EAAA,IAAQ,EAAG,IAErC,MAAO,GAAA,KAAS,GAAS,EAAA,IAAQ,EAAU,IAAa,kBAI/B,EAAe,GAC1C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,GAAS,GAAY,kBAInB,EAAe,GACnC,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAO,KAC3D,MAAO,GAAA,KAAS,EAAW,kBAIE,EAAe,GAC9C,MAAO,GAAA,KAAK,WACV,GAAM,GAAM,EAAA,IAAQ,EAAA,IAAQ,EAAO,GAAQ,IACrC,EAAM,EAAA,IAAQ,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAQ,GAAQ,IACvD,MAAO,GAAA,QAAY,EAAG,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAK,mBAY1B,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,KAAK,IAAI,GAChB,EAAiB,EAAA,IAAQ,EAAO,GAChC,EAAgB,EAAA,IAClB,EAAA,IAAQ,EAAgB,EAAA,SAAa,EAAA,IAAQ,GAAI,KACjD,GACJ,MAAO,GAAA,KAAS,EAAe,kBAK/B,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAI,EACF,EAAS,EAAA,QAAY,QAGrB,GAAM,GAAY,EAAA,IAAQ,EAAQ,EAAO,MAAM,OAAS,EAAG,IAC3D,EAAS,EAAA,IAAQ,EAAQ,GAE3B,SAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MACzC,EAAA,IAAQ,EAAA,IACX,EAAA,IAAQ,EAAO,UAAW,EAAA,IAAQ,IAAU,EAAO,MAAM,OAAS,kBActE,EAAgB,EAAgB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC3B,EAAA,KAAK,WACV,GAAM,GAAa,EAAA,MAAU,GAAU,IAAS,QAChD,EAAS,EAAA,YAAgB,EAAQ,KAAW,EAAI,MAChD,GAAM,GAAc,EAAO,MACrB,EACF,EAAA,OAAW,EAAY,EAAY,EAAY,OAAS,IACnD,QAAQ,GACjB,MAAO,IAAwB,EAAc,EAAQ,iBA0BrD,EAAgB,GAClB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAO,MAAO,EAAO,OACzC,KAAM,IAAI,GACN,8DACG,MAAK,UAAU,EAAO,OAAM,QAAQ,KAAK,UAAU,EAAO,SAEnE,MAAO,GAAA,KAAK,WAOV,GAAM,GAAa,EAAO,OACpB,EAAe,EAAO,MAAM,MAClC,MAAO,GAAW,IAAI,EAAO,IAAI,IAAS,IAAI,EAAa,MAAM,uBAIlC,EAAe,GAChD,MAAO,GAAA,KAAK,WACV,GAAI,GACJ,SAAI,EAAA,YAAgB,EAAO,KAAW,EAAI,MAC1C,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,EAAG,KAC3B,EAAA,KAAS,GAA8B,EAAO,GAAI,kBAKzD,EAAe,GACjB,MAAO,GAAA,KAAK,WACV,GAAM,GAAc,EAAA,YAAgB,EAAO,KAAW,GAChD,EAAc,EAAA,YAAgB,EAAO,KAAW,GACtD,MAAO,GAAA,IACH,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAA,IAAQ,EAAa,KAAgB,kBAI5C,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,EAAA,IAAQ,EAAA,IAAQ,KAAW,IAC3C,MAAO,GAAA,KAAS,EAAA,IAAQ,EAAO,EAAA,IAAQ,EAAO,IAAW,kBAI7B,EAAe,GAC7C,MAAO,GAAA,KAAK,WACV,GAAM,GAAiB,GAAY,EAAO,IACpC,EAAiB,GAAY,EAAO,IACpC,EAAY,EAAA,IAAQ,EAAgB,GAC1C,MAAO,GAAA,IAAQ,EAAA,IAAQ,EAAW,OAkB/B,GAAM,IAAsD,CACjE,iBAAgB,GAChB,kBAAiB,GACjB,4BAA2B,GAC3B,4BAA2B,GAC3B,aAAY,GACZ,MAAK,GACL,iBAAgB,GAChB,QAAO,GACP,wBAAuB,GACvB,8BAA6B,GAC7B,mBAAkB,GAClB,0BAAyB,GACzB,QAAO,GACP,gBAAe,gBAKG,GAClB,GAAI,MAAO,IAAmB,UAC5B,GAAI,IAAkB,IACpB,MAAO,IAAU,GAEnB,GAAI,GAAS,gBAAgB,EAC7B,KAAI,GAAe,cAAc,SAAS,wBACxC,GAAS,gBAAgB,EAAc,wFAInC,GAAI,GAAW,OAErB,OAAO,GC7QX,YAyB+B,EAAe,GAC5C,MAAO,GAAA,KAAK,WACV,GAAM,GAAY,EAAA,IAAQ,GAAI,EAAA,SAAa,IACrC,EAAmB,GAAO,EAAA,QAAY,EAAO,GAAY,EAAM,OACrE,MAAO,GAAA,KAAS,EAAA,MAAU,EAAO,GAAmB,kBAIpB,EAAe,GACjD,MAAO,GAAA,KACH,WAAM,MAAA,IACF,EAAA,MAAU,EAAA,OAAW,EAAO,IAAK,EAAA,OAAW,EAAO,KAAM,aAGnE,YAAuB,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,aAIrE,YAAwB,EAAe,GACrC,MAAO,GAAA,KAAK,WACV,MAAO,GAAA,WAAe,EAAM,MAAM,GAAI,EAAM,MAAM,IAAI,MAAM,KAAK,yBAI3C,EAAe,GACvC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIS,EAAe,GACpC,MAAO,GAAA,KAAK,WACV,GAAM,GAAK,GAAc,EAAO,GAC1B,EAAK,GAAe,EAAO,GAE3B,EAAc,EAAG,IAAI,GAE3B,MAAO,GAAA,MAAU,EAAA,QAAY,EAAa,GAAI,EAAG,IAAI,GAAc,GAC9D,KAAK,yBAIqB,EAAe,GAChD,MAAO,IAAuB,EAAO,eAInC,EAAe,GACjB,MAAI,GAAM,OAAS,EAAM,MACvB,GAAQ,EAAM,QAAQ,CAAC,EAAM,KAAO,KAEtC,EAAQ,EAAM,OAAO,IACrB,AAAI,EAAM,QAAU,EAAM,OACxB,GAAQ,EAAM,OAAO,EAAM,QAEtB,EAAA,MAAU,EAAO,GAAO,OAAO,WAajC,GAAM,IAAM,GACN,GAAM,GACN,GAAM,GACN,GAAM,GACN,GAAO,GACP,GAAO,GACP,GAA0B,GAC1B,GAAS,GACT,GAAgC,GAIhC,GAAuD,CAClE,eAAc,GACd,oBAAmB,GACnB,UAAS,GACT,wBAAuB,GACvB,8BAA6B,GAC7B,IAAG,GACH,IAAG,GACH,IAAG,GACH,IAAG,GACH,KAAI,GACJ,KAAI,GACJ,OAAM,gBAGY,GAClB,GAAI,MAAO,IAAe,UAAY,IAAc,IAClD,MAAO,IAAW,GACb,GAAI,MAAO,IAAe,UAAY,GAAc,KACzD,MAAO,GAEP,KAAM,IAAI,GAAW,kBAAkB,eAqBP,GAElC,GADA,GAAY,IAAO,KAAM,0BAA0B,GAC/C,MAAO,IAAO,SAChB,MAAO,GAGP,OADI,GAAM,OACQ,EAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAU,KAAS,GACrB,EAAS,EACT,OAGJ,GAAI,IAAW,OACb,MAAO,GAET,OAAkB,GAAA,EAAA,EAAA,OAAO,KAAK,IAAZ,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAW,KAAS,GACtB,EAAS,EACT,OAGJ,MAAI,KAAW,OACN,EAED,EAAgB,KCzL5B,YAwB6B,GAC3B,GAAM,GAA2D,CAC/D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,MAC/B,SAAY,WAAM,MAAA,GAAA,MAAM,SAAS,EAAG,IAAM,OAC1C,KAAQ,WAAM,MAAA,GAAA,MAAM,KAAK,KAAO,GAAK,KAAO,OAC5C,OAAU,WAAM,MAAA,GAAA,MAAM,OAAO,KAAO,GAAK,KAAO,KAAW,IAC3D,QAAW,WAAM,MAAA,GAAA,MAAM,QAAQ,KAAO,GAAK,EAAG,OAC9C,IAAO,WAAM,MAAA,GAAA,MAAM,IAAI,OASzB,GAPA,EAAa,QAAa,EAAa,QACvC,EAAa,SAAc,EAAa,SACxC,EAAa,KAAU,EAAa,KACpC,EAAa,OAAY,EAAa,OACtC,EAAa,QAAa,EAAa,QACvC,EAAa,IAAS,EAAa,IAE/B,IAAc,GAChB,MAAO,GAAa,KAEtB,KAAM,IAAI,GAAW,qBAAqB,GC3C5C,AAeO,GAAM,IAA8C,EAAI,KAAO,iBAelE,EAAyB,EAAmB,GAC9C,GAD8C,AAAA,IAAA,QAAA,GAAA,IAC1C,GAAuB,MACvB,MAAO,IAAwB,UAC/B,OAAO,eAAe,KAAyB,OAAO,WACtD,CAAC,GAAiB,GACpB,KAAM,IAAI,OACN,sEAGN,GAAI,GACF,GAAM,GAAM,KAAK,UAAU,GAC3B,AAAI,EAAI,OAAS,IACf,QAAQ,KACJ,mCAAmC,EAAS,qBAC5C,iBAAgB,EAAI,OAAM,iCAC1B,kHAEG,IAA2C,mBAgBvB,GAC/B,GAAI,IAAM,KAER,MAAO,GACF,GAAI,MAAO,IAAM,SACtB,GAAI,OAAO,eAAe,KAAO,OAAO,WAGtC,OADM,GAAO,OAAO,KAAK,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,MAAO,IAAQ,SAEjB,MAAO,GAET,GAAI,CAAC,GAAiB,EAAE,IACtB,MAAO,GAGX,MAAO,WAGH,MAAM,QAAQ,IAEhB,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,GAAiB,GACpB,MAAO,GAGX,MAAO,OAKP,OAAO,QAKX,GAAM,GAAQ,MAAO,GACrB,MAAO,KAAU,UAAY,IAAU,UAAY,IAAU,WCrGjE,YA8BI,EAAkB,EAAqB,EAEvC,GAAA,AAAA,IAAA,QAAA,GACI,QAAQ,KACd,GAAM,GAAiB,GAAsB,GAGvC,EAAsB,CAAC,eAAgB,eAAgB,WAC7D,AAAI,EACF,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAEtC,GAAa,GAAc,GAC3B,EAAY,GAAa,CAAC,IAAM,IAAM,IAAM,IAI9C,AAAI,EAAU,EAAU,OAAS,IAAM,GAErC,GAAY,EAAU,IAAI,SAAA,GAAK,MAAA,MAAK,MAAM,EAAa,MAGzD,GAAI,GACJ,GAAI,CAAC,GACH,EAAU,KAAK,mBACf,EAAgB,GAChB,OAAW,KAAS,GAAM,aACxB,EAAc,KAAI,MAAlB,EAAsB,EAAM,aAAa,IAI7C,EAAQ,IAAI,OAAO,IACnB,GAAS,EAAW,EAAW,GAC/B,EAAQ,IAAI,OAAO,IAGnB,OADM,GAAS,EAAM,OACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EACF,GAAkB,EAAO,GAAI,EAAW,GAExC,GACI,EAAO,GAAI,EAAW,EAAe,GAE3C,EAAS,KAAM,EAAO,OAAS,EAAI,IAAM,KAAK,OAAO,IAItD,EAAc,mCAEf,GAAM,GAAiB,GAAqB,GACtC,EAAoB,GAAqB,EAAM,qBAErD,EAAQ,iBAAiB,GAAiB,IAC1C,EAAQ,qBAAqB,GAC7B,EAAQ,yBAAyB,GACjC,EAAQ,IAAI,OAAO,IAGrB,YAA8B,GAC5B,GAAI,GAEJ,MAAK,GAAc,2BAA6B,KAC9C,EACI,GAAsB,EAAc,2BAExC,EAAiB,GAAqB,EAAM,kBAGvC,EAGT,YAA+B,GAC7B,GAAI,GAAiB,GACf,EAAyB,GACzB,EAAgB,GACtB,OAAW,KAAS,GAAM,aACxB,EAAa,KAAK,EAAM,aAAa,IAEvC,OAAyB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,EAAW,OAAS,GACpB,EAAW,SAAW,GAAK,EAAW,GAAG,cAAc,OAAS,GAClE,EAAiB,GACjB,MAEF,EAAM,KAAI,MAAV,EAAc,GAEhB,GAAI,EAEF,OAAoB,GAAA,EAAA,EAAA,EAAM,OAAN,EAAA,EAAA,OAAA,KAElB,OAFS,GAAK,EAAA,GACV,EAAO,GACQ,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAM,QAAQ,KAAU,GAC1B,GAAI,GACF,EAAiB,GACjB,UAEA,GAAO,GAIb,GAAI,CAAC,EACH,MAIN,MAAO,GAGT,YACI,EAAkB,EAElB,GAAA,AAAA,IAAA,QAAA,GAA6D,QAAQ,KAEvE,OADI,GAAO,GACF,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAI,GACN,GAAO,EAAK,MAAM,EAAG,EAAK,OAAS,GAAK,KAE1C,GAAQ,EAAO,GACf,EAAO,EAAK,MAAM,EAAG,EAAU,IAC/B,GAAQ,IAAI,OAAO,EAAU,GAAK,EAAK,QAEzC,EAAQ,GAQV,YACI,EAAc,EAEd,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAGhB,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EACF,CAAI,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,YAChE,GAAS,EAAQ,EAAW,GAM9B,YACI,EAAc,EAAqB,EAEnC,GACF,GAAI,GACJ,IACE,EAAc,KAAK,UAAU,EAAM,mBAC5B,GACP,EAAc,WAIhB,OADM,GAAwB,GACX,EAAA,EAAA,EAAA,EAAM,aAAN,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,GAAiB,MAAQ,EAAc,OAAS,GAChD,EAAc,QAAQ,KAAU,GAClC,SAEF,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAAG,KACrC,EAAoB,EAAK,YAAY,GACrC,EAAqB,EAAK,cAAc,GAC9C,EAAY,KACL,EAAY,IAAI,EAAiB,KAAK,EAAkB,MAGnE,GAAM,GAAO,EAAM,KACb,EAAY,EAAM,eAClB,EAAkB,EAAY,SAAW,EAAI,GAAK,EAAY,GAC9D,EAAmB,CACpB,EAAI,KAAK,EAAS,IAAK,EAAa,EAAM,cAAc,WAC3D,GAGF,GAAS,EAAQ,EAAW,GAC5B,OAAS,GAAI,EAAG,EAAI,EAAY,OAAQ,EAAE,EACxC,GAAS,CAAC,GAAI,GAAI,GAAI,EAAY,IAAK,EAAW,GCvNtD,AA8BA,YACI,EAAa,EAAe,GAC9B,MAAQ,KAAQ,gBAAkB,IAAQ,gBAClC,IAAQ,gBACZ,IAAU,GAAK,MAAO,IAAU,qBAUlC,EAA6B,GAC/B,GAAI,IAAmB,KACrB,MAAO,MACF,GAAI,MAAO,IAAmB,SACnC,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAmB,UAC1B,MAAO,IAAmB,UAC7B,MAAO,GACF,GAAI,YAA0B,QAGnC,OAFM,GAAU,GACV,EAAc,EAAe,OAC1B,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAe,GAC5B,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACf,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAArB,GAAM,GAAW,EAAA,GACd,EAAgB,EAAe,GACrC,GAAI,IAAgB,QAAU,MAAO,IAAkB,SAIrD,EAAO,GAAe,OAEtB,GAAM,GAAQ,GAA0B,GACxC,EAAO,GAAS,GAAoB,EAAe,IAGvD,MAAO,gBAWP,EAAyC,GAC3C,GAAI,GAAa,KACf,MAAO,MACF,GAAI,MAAO,IAAa,SAC7B,MAAO,IAA0B,GAC5B,GACF,MAAO,IAAa,UAAc,MAAO,IAAa,UACzD,MAAO,GACF,GAAI,YAAoB,QAG7B,OAFM,GAAU,GACV,EAAc,EAAS,OACpB,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAO,EAAS,GACtB,AAAI,GAA6B,EAAK,EAAG,GACvC,EAAQ,KAAK,GAEb,EAAQ,KAAK,GAAoB,EAAM,IAG3C,MAAO,QAGP,OADM,GAAmC,GACrB,EAAA,EAAA,EAAA,OAAO,KAAK,GAAZ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAU,EAAS,GACnB,EAAQ,GAA0B,GACxC,AAAK,KAAU,QAAU,IAAU,cAC/B,MAAO,IAAY,SAIrB,EAAO,GAAS,EAEhB,EAAO,GAAS,GAAoB,EAAS,GAGjD,MAAO,IC7HX,GAGM,IAAU,QCHhB,AA0BA,YAAiC,EAAqB,GAEpD,GAAI,EAAI,OAAS,MAAQ,EAAI,QAAU,EAAI,MAEzC,MAAO,GAET,IAEE,MAAO,GAAA,KAAK,EAAK,EAAI,aACd,GAEP,KAAM,IAAI,GACN,0BAA0B,EAAI,MAAK,kCACnC,gBAAe,EAAI,KAAI,MAAM,EAAI,MAAK,QAgB9C,GAAA,IAAA,WAUE,WAAY,GACV,GAVM,KAAA,SAAmC,GACnC,KAAA,QAAkC,GAClC,KAAA,QAAoC,GAQtC,YAAiB,GACnB,OAAW,KAAM,GAAM,SACrB,KAAK,SAAS,GAAM,EAAM,SAAS,GACnC,AAAI,IAAM,GAAM,SACd,MAAK,QAAQ,GAAM,EAAM,QAAQ,SAIrC,GAAI,GAAS,KACX,OAEF,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,KAAK,IAAI,EAAK,IAAK,EAAK,SAe9B,SAAA,UAAA,IAAA,SAAI,EAAqB,EAAe,GACtC,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAK,SAAS,EAAI,IAAM,GAAwB,EAAK,GACrD,KAAK,QAAQ,EAAI,MAAQ,EAAI,GAC7B,AAAI,GAAQ,MACV,MAAK,QAAQ,EAAI,IAAM,OAGzB,MAAM,IAAI,GAAW,uBAAuB,EAAI,KAAI,QAAQ,EAAI,IAElE,MAAO,OAQT,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,IAAI,EAAK,IAAK,EAAK,QAO1B,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,MAAK,SAAS,EAAI,KAAO,MAMlC,EAAA,UAAA,MAAA,WACE,MAAO,QAAO,KAAK,KAAK,UAU1B,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,SAAS,EAAI,SAG3B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,SAAS,KAWzB,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,YAAe,KACjB,GAAI,KAAK,SAAS,EAAI,KAAO,KAC3B,KAAM,IAAI,GAAW,oBAAoB,EAAI,MAE7C,MAAO,MAAK,QAAQ,EAAI,SAG1B,GAAM,GAAK,KAAK,QAAQ,GACxB,GAAI,GAAM,KACR,KAAM,IAAI,GAAW,yCAAyC,GAEhE,MAAO,MAAK,QAAQ,KAKxB,EAAA,UAAA,aAAA,WACE,AAAI,KAAK,SAAW,MAClB,EAAA,QAAQ,KAAK,UAGnB,KAIM,GAA+D,GAG/D,GAC4D,eA4C9D,EAA0C,EAC1C,EAAiB,GAWnB,OATM,GAAoB,GAAU,KAAO,GAAQ,EAAO,SAEpD,EAAe,MAAM,QAAQ,GAC7B,EACF,EAAe,EAA8B,CAAC,GAE5C,EAAc,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,OACpC,EAAyB,GACzB,EAAY,EAAS,QACF,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,AAAI,EAAU,QAAQ,KAAgB,GACpC,EAAa,KAAK,EAAS,SAAS,IAEpC,EAAa,KAAK,MAItB,AAAI,GAAS,MAEX,GAAM,cAAgB,UACtB,EAAM,cAAgB,UAIxB,GAAM,GACF,EAAY,KAAK,KAAO,IAAM,EAAS,QAAQ,KAAK,KACpD,EACA,EACJ,GAAI,GAAa,IAAoB,MAGnC,GAAM,GAAM,GAAqC,EAAY,GAC7D,EAAS,EAAI,OACb,EAAkB,EAAI,gBAGtB,GAAa,GAAmB,EAChC,GAAsB,GAAmB,EAE3C,EAAS,GAAa,GACtB,EAAkB,GAClB,AAAK,GACH,OAAO,OAAO,EAAiB,GAAsB,IAMvD,OAHM,GAAmB,GAAI,IAAS,GAG7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAI,GAAS,MAEX,GAAM,GAAa,EAAA,SAAS,WAC5B,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAExB,AAAI,EAAa,EAAM,eACrB,GAAM,cAAgB,GAI1B,GAAM,GAAW,EAAO,GAClB,EAAW,EAAS,YAC1B,GAAI,YAAoB,IACtB,SAOF,OALM,GAAwB,GACxB,EAAuB,GACvB,EAA6B,GAE/B,EAAa,GACG,EAAA,EAAA,EAAA,EAAS,OAAT,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAQ,EAAiB,SAAS,GAClC,EAAO,EAAiB,QAAQ,GACtC,EAAY,KAAK,GACjB,EAAW,KAAK,GAChB,AAAI,GAAQ,MACV,GAAa,IAEf,AAAK,GACH,GAAgB,EAAM,QACtB,AAAI,EAAgB,EAAM,QAAU,GAAK,CAAC,EAAS,OAAO,IACtD,EAAY,QAAQ,EAAM,QAAU,IAAM,CAAC,EAAM,YACjD,EAAM,YAAY,WAAa,IACjC,EAAiB,KAAK,IAK5B,AAAI,GACF,GAAS,GAAU,GACnB,EAAO,KAAU,EAAW,IAE9B,GAAM,GACF,GAAO,EAAS,MAAM,EAAa,IACnC,EAA8B,KAClC,AAAI,EAAS,iBACX,GAAa,EAAS,YAAY,EAAa,IAKjD,OAHM,GAAe,GAAe,GAC9B,EACF,MAAM,QAAQ,GAAgB,EAAe,CAAC,GACzC,EAAI,EAAG,EAAI,EAAsB,OAAQ,EAAE,GAClD,AAAK,EAAiB,OAAO,EAAsB,KACjD,EAAiB,IACb,EAAsB,GAAI,EAAc,GACxC,MAAM,QAAQ,GAAc,EAAW,GAAK,GAElD,GAAM,GAAQ,EAAY,QAAQ,EAAsB,GAAG,MAC3D,AAAI,IAAU,IACZ,GAAa,GAAS,EAAc,IAIxC,AAAK,GAEH,EAAA,QAAQ,GAQZ,SAAiB,eAEV,EAAe,EAAe,EAAa,GAsBpD,YACI,EAA2B,GAE7B,EAAA,KAAK,OACD,GAAW,MAAQ,EAAQ,OAAS,EACpC,WAAM,MAAA,0CAEV,GAAI,GAAgC,GAChC,EAAkC,GACtC,GAAI,EAAQ,SAAW,GAErB,GAAM,GACF,GAAgD,EAAQ,GAAI,GAChE,EAAc,EAAI,OAClB,EAAoB,EAAI,iBAGxB,QADM,GAAU,GAAI,KACA,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAKlB,OALS,GAAK,EAAA,GACR,EAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,aAIc,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAxB,GAAM,GAAc,EAAA,GACvB,AAAK,EAAQ,IAAI,EAAe,OAC9B,GAAY,KAAK,GACjB,EAAQ,IAAI,EAAe,sBAKpB,GACT,AAAI,EAAkB,IAAS,MAC7B,GAAkB,GAAQ,GAAI,MAEhC,EAAa,GAAM,QACf,SAAA,GAAa,MAAA,GAAkB,GAAM,IAAI,MAL/C,OAAW,KAAQ,KAAR,GASf,MAAO,CACL,OAAQ,EACR,gBAAiB,GAAoB,IAIzC,YAA6B,GAC3B,GAAM,GAAmC,GACzC,OAAW,KAAQ,GACjB,EAAgB,GAAQ,EAAa,GAAM,KAE7C,MAAO,eAeL,EAAuB,GASzB,OAPM,GAAU,GAAI,KACd,EAA2B,GAC3B,EAA6B,GAKjB,EAAA,EAAA,EAAA,EAAS,QAAT,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAQ,IAAI,GAGd,GAAM,GAA0B,GAC1B,EAAkB,GAKxB,IAFA,EAAM,KAAK,GAEJ,EAAM,OAAS,IACpB,GAAM,GAAM,EAAM,EAAM,OAAS,GACjC,GAAI,EAAQ,IAAI,EAAI,OAClB,EAAM,MACN,SAEF,GAAM,GAAc,EAAM,EAAM,OAAS,KAAO,EAAM,OAAS,EAC/D,GAAI,EAAI,OAAO,SAAW,GAAK,EAE7B,EAAM,MACN,EAAO,KAAK,GACZ,EAAQ,IAAI,EAAI,MAChB,AAAI,GACF,EAAM,WAKR,EAAM,KAAK,EAAM,OAAS,GAC1B,OAAoB,GAAA,EAAA,EAAA,EAAI,OAAJ,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GAQd,GALA,AAAI,EAAa,EAAM,OAAS,MAC9B,GAAa,EAAM,MAAQ,GAAI,MAEjC,EAAa,EAAM,MAAM,IAAI,EAAI,MAE7B,EAAQ,IAAI,EAAM,MACpB,SAEF,EAAM,KAAK,KAIjB,MAAO,CAAC,OAAM,EAAE,aAAY,GAS9B,YAAwB,GAEtB,GAAI,GACJ,GAAI,EAAM,YAAY,aAAa,SAAW,EAC5C,EAAe,EAAM,YAAY,YAGjC,OADI,GAAoB,KACf,EAAI,EAAG,EAAI,EAAM,YAAY,aAAa,OAAQ,EAAE,EAC3D,OAA2B,GAAA,EAAA,EAAA,EAAM,YAAY,aAAa,GAChD,cADiB,EAAA,EAAA,OAAA,KAAtB,GAAM,GAAY,EAAA,GAErB,GAAI,EAAa,KAAO,EAAM,IAC5B,EAAY,EACZ,OAIN,EAAe,EAAM,YAAY,YAAY,GAE/C,MAAO,GCjhBT,AA6CA,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAoCtC,WAAY,GAAZ,GAAA,GAEE,EAAA,KAAA,KAAM,KAAG,KAET,GAtBF,EAAA,eAAiB,GAAI,KAqBnB,EAAK,KAAO,EAAK,KACb,EAAK,MAAQ,MACf,GAAM,GAAS,EAAK,eAAe,cACnC,EAAK,KAAO,GAAO,GAqBrB,GAlBA,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAKlB,AAAI,MAAM,QAAQ,EAAK,QACrB,EAAK,OAAS,EAAK,OAAO,QAE1B,EAAK,OAAS,CAAC,EAAK,QAEtB,AAAI,MAAM,QAAQ,EAAK,SACrB,EAAK,QAAU,EAAK,QAAQ,QAE5B,EAAK,QAAU,CAAC,EAAK,SAInB,GAAqB,EAAK,QAAQ,SAAW,EAAK,OAAO,OAC3D,KAAM,IAAI,GACN,mGAEA,IAAG,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,SAIhC,AAAI,GAAqB,EAAK,SAAS,SAAW,EAAK,QAAQ,QAC7D,QAAQ,KACJ,qGAEA,IAAG,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,SAOjC,EAAK,YAAc,GACnB,EAAK,uBAAyB,GAC9B,EAAK,yBAA2B,GAKhC,EAAK,aAAe,GACpB,EAAK,wBAA0B,GAC/B,EAAK,0BAA4B,GAKjC,EAAK,OAAS,GAMd,EAAK,sBAAwB,GAe7B,OAAgB,GAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YACtB,EAAK,aAAa,KAAK,GACvB,EAAK,wBAAwB,KAAK,GAClC,EAAK,0BAA0B,KAAK,GAMtC,OAAgB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,YACV,EAAY,EAAE,UACd,EAAc,EAAE,YAKtB,GAAqB,IAAc,EAAG,4BACtC,GAAqB,IAAgB,EAAG,8BACxC,EAAK,YAAY,KAAK,GACtB,EAAK,uBAAuB,KAAK,GACjC,EAAK,yBAAyB,KAAK,GAIrC,EAAK,WAAa,GAClB,EAAK,YAAc,GACnB,EAAK,gBAAkB,GACvB,EAAK,eAAiB,GACtB,EAAK,gBAAkB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,EAAK,YAAY,GAE/B,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,WACN,6DACA,qBAAoB,EAAK,OAAM,MAC/B,UAAS,EAAC,0BACV,oBAAmB,EAAM,eAAc,MAE7C,EAAK,WAAW,KAAK,EAAM,MAC3B,EAAK,gBAAgB,KAAK,EAAM,iBAEhC,EAAK,eAAe,KAAK,EAAM,MAEjC,OAAoB,GAAA,EAAA,EAAA,EAAK,aAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,YAAY,KAAK,EAAM,MAG9B,EAAK,oBAAsB,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QAClD,EAAK,qBAAuB,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,QAwFpD,OAjFM,GAA0C,GAE1C,EAAyC,GACzC,EAA4C,GAE5C,EAA6C,GAC7C,EAA4C,GAC5C,EAAiC,GAoBjC,EACF,SAAC,GAAwB,GAAuB,GAC/C,GAAe,GAAoB,IAClC,AAAI,KAAS,MAAQ,IAAa,MAAQ,IAAe,OACvD,IAAQ,GAAO,YACf,GAAY,GAAO,UACnB,GAAc,GAAO,aAEvB,GAAM,IAAO,GAAM,aAAa,IAGhC,GAAI,GAAgB,QAAQ,MAAU,GACpC,KAAM,IAAI,IACN,cAAc,GAAO,KAAI,cAAc,GAAM,KAAI,yBAKvD,GAAI,GAAc,QAAQ,MAAU,GAClC,OAIF,EAAK,eAAe,IAAI,EAAU,QAAQ,GAAO,KAGjD,AAAM,GAAM,KAAM,IAChB,GAAa,GAAM,IAAM,OAAO,KAAK,GAAc,QAGrD,AAAI,GAAgB,QAAQ,MAAU,IACpC,GAAgB,KAAK,IAKvB,OADM,IAAmB,GAAK,cAAc,OACnC,GAAI,EAAG,GAAI,GAAkB,MACpC,GAAM,IAAI,GAAK,aAAa,IACtB,GAAQ,GAAK,cAAc,IAC3B,GAAY,GAAK,YAAY,IAC7B,GAAc,GAAK,cAAc,IACvC,EACI,GAAG,GAAe,GAAiB,GAAO,GAC1C,IAGN,IADA,GAAc,KAAK,IACZ,GAAgB,QAAQ,KAAS,GACtC,GAAgB,OAAO,GAAgB,QAAQ,IAAO,GAExD,EAAuB,KAAK,KAG5B,EAAwB,GACxB,EAA0B,GAChB,EAAA,EAAA,EAAA,EAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAgB,EAAG,EAAe,GAKpC,OAFM,GACF,EAAuB,QAAQ,UAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAa,EAAK,IAAM,EAExB,AAAM,EAAK,KAAM,IACf,GAAY,EAAK,IAAM,GAEzB,GAAI,GAAQ,EAAY,EAAK,IAGvB,EACD,EAAa,EAAK,cAAc,KAAO,KACnC,EACA,EAAa,EAAK,cAAc,IAOzC,EAAQ,KAAK,IAAI,EAAO,GACxB,EAAa,EAAK,cAAc,IAAM,EACtC,EAAe,EAAK,cAAc,IAAM,EAAK,cAC7C,EAAY,EAAK,IAAM,EAGvB,OAAS,GAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAa,aAAa,GACxC,EACD,EAAY,EAAY,KAAO,KAAO,EACA,EAAY,EAAY,IACnE,EAAY,EAAY,IAAM,KAAK,IAAI,EAAQ,EAAG,GAClD,EAAa,EAAY,IAAM,GAKnC,GAAM,GAA0C,GAChD,OAAW,KAAU,IACnB,GAAM,GAAQ,EAAY,GAC1B,AAAM,IAAS,IACb,GAAa,GAAS,IAExB,EAAa,GAAO,KAAK,EAAa,IAIxC,GAAM,IAA4C,GAClD,OAAW,MAAW,IACpB,GAAM,GAAQ,EAAa,IAC3B,AAAM,IAAS,KACb,IAAc,GAAS,IAEzB,GAAc,GAAO,KAAK,EAAe,KAI3C,GAAI,IAAY,OAAO,KAAK,IACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAG1B,EAAK,OAAS,GACd,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACR,GAAiB,GAAc,GAGrC,GAAe,KAAK,SAAC,GAAG,IACtB,GAAM,IAAS,EAAa,GAAE,IACxB,GAAS,EAAa,GAAE,IAC9B,MAAI,IAAS,GACJ,GAEL,GAAS,GACJ,EAEF,IAET,OAAoB,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAf,GAAM,GAAK,GAAA,IACd,AAAI,YAAiB,IACnB,EAAK,sBAAsB,KAAK,GAElC,EAAK,OAAO,KAAK,IAGrB,EAAK,cAAgB,GAGrB,GAAY,OAAO,KAAK,GACP,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IAStB,OAJM,IAAoB,EAAK,OAAO,QAGhC,GAAoC,GACtB,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,KAClB,OADS,GAAK,GAAA,IACK,GAAA,EAAA,GAAA,EAAa,GAAb,GAAA,GAAA,OAAA,MAAd,GAAM,GAAI,GAAA,IACP,EAAQ,EAAK,cACnB,GAAI,GAAS,MACX,OAAgB,IAAA,EAAA,GAAA,EAAK,aAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAI,GAAkB,QAAQ,KAAO,GACnC,KAAM,IAAI,IACN,sDAAsD,EACtD,eAAc,EAAM,KAAI,OACxB,uDACA,WAAU,KAGlB,OAAgB,IAAA,EAAA,GAAA,EAAK,cAAL,GAAA,GAAA,OAAA,MAAX,GAAM,GAAC,GAAA,IACV,GAAkB,KAAK,GAEzB,GAAwB,KAAK,EAAM,OAMzC,EAAK,aAAe,EAKpB,OADM,IAAW,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,mBAC7B,IACT,GAAM,IAAiB,GAAS,OAAO,SAAA,IAAK,MAAA,MAAM,KAAM,OACxD,GAAI,KAAmB,EACrB,KAAM,IAAI,IACN,aAAa,GAAI,aAAa,GAAc,uEAE5C,KAAK,UAAU,MANJ,GAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAd,GAAM,IAAI,GAAA,OAAJ,IAcX,SAAK,cAAgB,GAErB,EAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,EACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,EAAK,OACnB,cAAe,EAAK,QACpB,WAAY,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,QACjC,YAAa,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,QACnC,YAAa,EAAK,OAAO,IAAI,SAAA,IAAK,MAAA,IAAE,QACpC,aAAc,EAAK,QAAQ,IAAI,SAAA,IAAK,MAAA,IAAE,UAExC,EAAK,MAAQ,GACb,EAAK,UAAY,IAGT,SAAA,UAAA,kBAAV,WACE,GAAI,KAAK,YAAc,EACrB,KAAM,IAAI,OAAM,cAAc,KAAK,KAAI,2BA8B3C,EAAA,UAAA,QAAA,WACE,KAAK,oBACL,GAAM,GACc,CAAC,qBAAsB,KAAM,qBAAsB,GACvE,GAAI,EAAE,KAAK,YAAc,GACvB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAO,sBAAwB,EAAM,UAAU,qBAKjD,OAAwB,GAAA,EAAA,EAAA,KAAK,sBAAL,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAO,sBAAwB,EAAU,UAAU,sBAGvD,SAAO,qBAAuB,KAAK,UAC5B,GAGT,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,KAAK,OAAO,QAAQ,SAAA,GAEhB,EAAc,kBACX,QAAQ,SAAA,GAAK,MAAA,GAAE,UAAY,MAElC,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WAIE,GAAI,KAAK,kBAAkB,OAAS,EAClC,KAAM,IAAI,GACN,wNAMN,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADI,GAA2B,GACX,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAU,EAAQ,OAAO,EAAM,kBAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACb,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAQ,KAAI,MAAZ,EAAgB,EAAM,qBAExB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACtB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAiB,KAAI,MAArB,EAAyB,EAAM,kBAEjC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,iBAAiB,OAAO,KAAK,sDAkB3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,IAGnC,OAFM,GAAgD,GAClD,EAAoB,EACJ,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACO,EAAA,EAAA,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,EAAa,EAAO,eAAiB,KACvC,KAAM,IAAI,GAAW,0BAA0B,EAAO,cAExD,EAAa,EAAO,cAAgB,EACpC,IAIJ,GAAM,GAAoD,GAC1D,OAAW,KAAQ,IAIjB,GAAI,GAAgB,EACpB,GAAI,EAAa,IAAS,MACxB,GAAM,GAAS,EAAK,MAAM,KACpB,EACF,EAAO,MAAM,EAAG,IAAI,OAAO,CAAC,EAAO,EAAO,OAAS,KACvD,EAAgB,EAAiB,KAAK,KAExC,GAAI,EAAa,IAAkB,KACjC,EAAkB,KAAK,CAAC,EAAa,GAAgB,EAAQ,aACpD,EACT,KAAM,IAAI,GACN,gDAAgD,GAEtD,MAAO,GAAa,GAGtB,GAAI,GAEF,GAAM,GAAuB,GAC7B,OAAW,KAAQ,GACjB,EAAW,KAAK,GAElB,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,GACH,EAAW,OAAM,OAChB,EAAiB,yBACrB,IAAG,IAIX,GAAc,IAON,EAAA,UAAA,cAAV,WACE,GAAM,GAAY,KAAK,YACjB,EAAwC,GAC9C,SAAY,UAAe,KAAK,eAChC,EAAY,OAAY,EACxB,EAAY,aAAkB,eAAe,GAG7C,EAAY,QAAa,gBAClB,GAeT,EAAA,UAAA,OAAA,SAAO,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IACnB,GAAM,GAAc,GAAoB,KAAK,iBAC7C,MAAO,GAAe,KAAK,UAAU,GAAe,GAgBtD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAE9B,OADM,GAAW,GAAI,IACZ,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,EAAK,OAAO,GAAI,EAAO,IAEtC,MAAO,IAAQ,EAAK,QAAS,EAAU,MAa3C,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAK,WACV,EAAS,GAAqB,GAC9B,GAAI,GACJ,MAAI,IAAQ,KACV,EAAQ,GAA2B,KAAM,EAAO,QAEhD,EAAQ,GAAqB,GAGxB,EAAK,iBAAiB,EAAQ,GAAO,MAahD,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAM,GAAc,GAA+B,GACnD,GAAI,EAAY,SAAW,KAAK,YAAY,OAC1C,KAAM,IAAI,GACN,+BAA+B,EAAU,KACzC,cAAa,KAAK,YAAY,OAAM,oBAK1C,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,EAAY,OAAQ,KACtC,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAa,EAAY,GAGzB,EAAW,EAAM,KAAO,OAC9B,EAAqB,GAAY,EAGnC,GAAM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,GAAK,MAAA,UAAS,EAAG,MACrB,KAAK,IAE5B,GAAI,EAAU,OAAS,EACrB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GAEP,EAAQ,EAAK,cACnB,GAAI,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,QAAQ,EAAM,MAAQ,GAExD,SAIF,OADM,GAAuB,GACpB,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAc,EAAa,KAAI,IAAI,EAAS,IAAI,EAChD,EAAa,EAAqB,GACxC,EAAY,KAAK,GAQnB,OALM,GAAc,EAAM,mBACtB,GAA+B,IAE7B,EAAe,GAA+B,GAC9C,EAAY,EAAM,aAAa,QAAQ,GACpC,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAqB,GAAY,EAAa,IAStD,OAFM,GAAwB,GACxB,EAA4B,GACzB,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GACzC,EAAc,KAAK,0BAA0B,GAC7C,EAAc,EAAM,KAAI,IAAI,EAAS,IAAI,EAC/C,EAAgB,KAAK,GAGvB,OAAS,GAAI,EAAG,EAAI,EAAgB,OAAQ,KAC1C,GAAM,GAAM,EAAgB,GAC5B,GAAqB,IAAO,IAC5B,EAAa,KAAK,EAAqB,IAIzC,MAAO,IAA+B,IAa9B,EAAA,UAAA,iBAAV,SAA2B,EAAkB,GAE3C,AAAI,GAAS,MACX,GAAQ,GAA2B,KAAM,EAAO,SASlD,OADM,GAAoD,GACjD,EAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,GACxC,GAAM,GAAI,KAAK,OAAO,GAChB,EAAI,EAAO,GACX,EAAO,EAAM,GACnB,EAAU,EAAE,IAAM,CAAC,EAAG,GAMxB,OAHM,GAAY,OAAO,KAAK,KAAK,cACZ,IAAI,SAAA,IAAK,MAAA,UAAS,GAAG,MACrB,KAAK,IACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,IAElB,OAFS,GAAK,EAAA,GACR,EAAQ,KAAK,aAAa,GACb,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAUjB,OAVS,GAAI,EAAA,GAEP,EAAQ,EAAK,cACb,EAAwB,EAAK,aAC7B,EAAyB,EAAK,cAK9B,EAAe,GAAI,OACT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,AAAI,EAAE,KAAM,IACV,EAAa,KAAK,EAAU,EAAE,KAGlC,GAAI,EAAa,SAAW,EAAsB,QAEhD,GAAI,GAAiB,GACjB,EAAe,OACf,EAAa,OACb,EAAa,OACb,EAAW,OAKf,GAHA,AAAI,EAAK,UAAY,MACnB,GAAS,EAAK,UAEZ,EAAa,SAAW,GACpB,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GACvB,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAgB,IACpD,EAAc,GACV,EAAM,YAAY,EAAgB,IACtC,EAAkB,CAAC,GACnB,EAAgB,CAAC,OAEjB,GAAkB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KAC1C,EAAgB,EAAa,IAAI,SAAA,IAAK,MAAA,IAAE,KACxC,AAAI,EAAO,MAAW,MACpB,GAAO,KAAU,GAEnB,EACI,GAAqB,EAAM,KAAK,EAAiB,IACrD,EAAc,GACV,EAAM,YAAY,EAAiB,IAGzC,GAAI,EAAM,oBACR,KAAM,IAAI,IACN,yHAMN,OAAS,GAAI,EAAG,EAAI,EAAuB,OAAQ,EAAE,GACnD,GAAM,GAAI,EAAuB,GAC3B,EAAI,EAAc,GAClB,EAAO,EAAY,GACzB,EAAU,EAAE,IAAM,CAAC,EAAG,KAS9B,OAHM,GAA0B,GAC1B,EAAwB,GACxB,EAAwB,GACd,EAAA,EAAA,EAAA,KAAK,QAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,GACI,EAAE,KAAM,GAAW,4BAA4B,EAAE,KAAI,MAAM,EAAE,IAC3D,GAAA,GAAA,EAAA,EAAA,IAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAa,KAAK,EAAO,OACzB,EAAc,KAAK,GACnB,EAAY,KAAK,GAInB,MAAO,CAAC,EAAe,EAAa,IAW9B,EAAA,UAAA,uBAAR,SAA+B,GAG7B,OAFM,GAAiD,GACnD,EACgB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAY,YAAiB,GAAY,EAAI,EAC7C,OAAS,GAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,KAAK,eAAe,IAAI,IAE1B,GAAkB,GAAW,EAC7B,GAAa,IAInB,MAAO,IAsBT,EAAA,UAAA,SAAA,SAAS,EAAe,GACtB,GAAI,GAAS,MACX,GAAI,KAAK,OAAO,QAAU,EACxB,KAAM,IAAI,GACN,wCAAwC,EAAK,oBAC7C,QAAO,KAAK,OAAO,OAAM,eAE7B,MAAO,MAAK,OAAO,WAGjB,GAAQ,KACV,KAAM,IAAI,GAAW,8CAIzB,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,OAAS,EACjB,MAAO,GAGX,KAAM,IAAI,GAAW,kBAAkB,IAQzC,EAAA,UAAA,gBAAA,WAAA,GAAA,GAAA,KAKE,MAAO,GAAA,KAAK,WAEV,OADM,GAAmB,GACL,EAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,IAClB,OADS,GAAK,EAAA,GACL,EAAY,EAAG,EAAY,EAAM,aAAa,OAClD,EAAE,GACL,GAAM,GAAU,EAAU,QAAQ,EAAO,GACzC,AAAI,EAAK,eAAe,IAAI,IAC1B,EAAO,KAAI,MAAX,EAAe,EAAM,mBAK3B,MAAO,MAIX,EAAA,UAAA,UAAA,WAWE,OAVM,GAAmC,CAAC,KAAM,KAAK,MAK/C,EACF,KAAK,uBAAuB,KAAK,QAG/B,EAAe,GACD,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EAAiB,EAAM,eACvB,EAAc,EAAM,YACpB,EAAuB,GACpB,EAAoB,EACxB,EAAoB,EAAM,aAAa,OAAQ,KAClD,GAAM,GAAO,EAAM,aAAa,GAC1B,EAAU,EAAU,QAAQ,EAAO,GACrC,EAAS,GACb,GAAI,KAAK,eAAe,IAAI,IAG1B,GAAI,EAAK,SACP,IACE,KAAK,UAAU,EAAK,UACpB,EAAS,EAAK,eACP,GACP,QAAQ,KACJ,SAAS,EAAM,KAAI,mDAEhB,GAAK,SAAQ,gCAChB,+EAEJ,EAAS,GAGb,GAAI,EAAK,cAAc,OAAS,GAE9B,OADM,GAAW,GACR,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,KAC7C,GAAM,GAAe,EAAK,cAAc,GAClC,EAAY,EAAK,YAAY,GAC7B,EAAc,EAAK,cAAc,GACjC,EAAU,EAAU,QAAQ,EAAc,GAC5C,EAAe,EAAkB,GACrC,AAAI,GAAgB,MAClB,GAAe,GAEjB,EAAS,KACL,CAAC,EAAa,KAAM,EAAc,EAAa,IAErD,EAAqB,KAAK,KAIhC,GAAM,GAAiC,GACvC,EAAK,KAAU,EAAM,KACrB,EAAK,UAAe,EACpB,EAAK,OAAY,EACjB,EAAK,aAAkB,EACvB,EAAa,KAAK,GAEpB,EAAO,OAAY,EAGnB,OADM,GAAc,GACX,EAAI,EAAG,EAAI,KAAK,YAAY,OAAQ,KAC3C,GAAM,GAAQ,KAAK,YAAY,GACzB,EAAY,KAAK,uBAAuB,GAExC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,yBAAyB,GAClD,EAAY,KAAK,CAAC,EAAM,KAAM,EAAc,IAE9C,EAAO,YAAiB,EAGxB,OADM,GAAe,GACZ,EAAI,EAAG,EAAI,KAAK,aAAa,OAAQ,KAC5C,GAAM,GAAQ,KAAK,aAAa,GAC1B,EAAY,KAAK,wBAAwB,GAEzC,EAAU,EAAU,QAAQ,EAAO,GACzC,GAAI,CAAC,KAAK,eAAe,IAAI,GAC3B,SAEF,GAAI,GAAe,EAAkB,GACrC,AAAI,GAAiB,MACnB,GAAe,GAEjB,GAAM,GAAc,KAAK,0BAA0B,GACnD,EAAa,KAAK,CAAC,EAAM,KAAM,EAAc,IAE/C,SAAO,aAAkB,EAClB,GAgBF,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAA8C,GAO9C,EAAkE,GACxE,WACI,EAAc,GAChB,AAAM,EAAM,OAAQ,GAGlB,EAAiB,EAAM,MAAM,KAAK,GAFlC,EAAiB,EAAM,MAAQ,CAAC,GAMpC,WAAqB,EAAc,GAGjC,OAFM,GAAiC,GACnC,GACoB,GAAA,EAAA,GAAA,EAAA,GAAA,GAAA,OAAA,MAAnB,GAAM,IAAS,GAAA,IACZ,GAAmB,GAAU,GAC7B,GAAmB,GAAU,GAC7B,GAAqB,GAAU,GAKrC,GAHA,GAAS,GAAU,IAAM,KACrB,GACA,GAAU,GACV,CAAE,MAAoB,KACxB,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAe,EAAc,IACnC,GAAI,GAAa,aAAa,QAAU,IACtC,EAAmB,EAAO,GAC1B,OAEF,GAAM,IAAc,GAAa,aAAa,IAC9C,EAAa,KAAK,GAAY,cAAc,KAK9C,AAAI,EAAa,OAAS,GACxB,EAAM,MACF,GAA+B,GAC/B,IAUR,WAAsB,GACpB,GAAM,GAAY,EAAU,KAEtB,EACF,GACI,EACA,EAAO,eAAoB,KACvB,EAAO,cACP,IACZ,EAAM,6BAA6B,GACnC,EAAc,GAAa,EAE3B,GAAM,IACF,EAAU,aACd,GAAiB,QAAQ,SAAA,IACvB,GAAI,CAAE,cAAoB,QACxB,KAAM,IAAI,GACN,yDACI,IAMV,EAAmB,EAAO,MAO9B,OAFM,GAAO,EAAO,KACd,EAAmB,EAAO,OACR,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GAClB,EAAa,GAOf,KAAO,CAAC,GAA4B,IAClC,OAAwB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAQ,EAAc,EAAU,MACtC,GAAI,EAAM,OAAQ,IAChB,GAAM,GAAkC,EAAiB,EAAM,MAC/D,MAAO,GAAiB,EAAM,MAC9B,OAAuB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAlB,GAAM,GAAQ,EAAA,GACjB,EAAY,EAAO,KAU3B,OAJM,GAAiC,GACjC,EAAkC,GAClC,EACF,EAAO,YACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAa,KAAK,EAAmB,IAIvC,OAFM,GACF,EAAO,aACa,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAnB,GAAM,GAAS,EAAA,GACZ,EAAY,EAAU,GACtB,EAAY,EAAU,GACtB,EAAc,EAAU,GAC9B,GAAqB,IAAa,IAClC,GAAM,GAAQ,EAAc,GACtB,EAAqB,EAAM,aAAa,GAAW,cACzD,EAAc,KAAK,EAAmB,IAExC,MAAO,IAAI,GAAI,CAAC,OAAQ,EAAc,QAAS,EAAe,KAAI,KASpE,OAAA,eAAI,EAAA,UAAA,WAAQ,KAAZ,WAGE,GAAI,KAAK,UACP,KAAM,IAAI,GACN,wLAIN,OAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAM,SACR,MAAO,GAGX,MAAO,oCAST,EAAA,UAAA,YAAA,WAAA,GAAA,GAAA,KACE,EAAA,KAAK,WACH,EAAK,OAAO,QAAQ,SAAA,GAElB,AAAI,EAAM,UACR,EAAM,mBAMhB,GA3vCwC,IC7CxC,AAiCA,YACI,EAAmD,EACnD,GACF,GAAM,GAAa,EAAY,OAC/B,GAAI,GAAW,MAAS,MAAM,QAAQ,IAAY,EAAQ,SAAW,EACnE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,QAEjC,GAAI,IAAe,EACjB,MAAI,OAAM,QAAQ,IAAY,EAAQ,SAAW,EACxC,EACE,MAAO,IAAY,UAAY,EAAY,IAAM,GACnD,CAAE,EAA2B,EAAY,KAEzC,CAAC,GAGZ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAQ,SAAW,EACrB,KAAM,IAAI,OACN,YAAY,EAAU,mBAAmB,EAAQ,OAAM,IACvD,kCAAiC,EAAU,cAC3C,iEAEN,MAAO,WAEL,MAAO,IAAY,UAAY,OAAO,KAAK,GAAS,OAAS,GAC7D,MAAQ,GAA2B,OAAO,KAAK,GAAS,KACpD,UACN,GAAM,GAAwB,GAC9B,SAAY,QAAQ,SAAA,GAClB,AAAI,IAAc,GAChB,EAAO,KAAM,EAA2B,IAExC,EAAO,KAAK,QAGT,MAEP,MAAM,IAAI,OACN,2BAA2B,EAAU,cACrC,OAAM,EAAU,kCACb,GAAU,+BAA+B,EAAW,WACvD,aAAY,EAAU,oBAAoB,KAAK,UAAU,iBAkB7D,EACA,GACF,MAAO,IACH,EAAa,EAAa,2BA6B5B,EAAW,EAAuB,EAClC,2GACF,GAAI,GAAgB,MAAQ,GAAoB,KAG9C,KAAM,IAAI,OAAM,qDAGd,IAAe,KAEX,GAAqB,EAAA,KAAK,WAC9B,GAAI,EAAE,MAAM,SAAW,EAErB,MAAO,GAAE,QACJ,GAAI,EAAE,MAAM,SAAW,EAC5B,GAAI,EAAE,MAAM,GAAK,GAEf,GAAM,GAAO,EACb,MAAO,GAAE,OAAO,QACX,GAAI,EAAE,MAAM,KAAO,EAExB,MAAO,GAAE,QAAQ,CAAC,EAAE,MAAM,KAE1B,KAAM,IAAI,OACN,+CAA+C,EAAE,MAAM,GAAE,4EAK/D,MAAM,IAAI,OACN,yCAAyC,EAAE,KAAI,4EAKjC,EAAA,GAAA,OAAM,KAAK,CAAA,EAAM,EAAS,SA3B9C,CAAA,EAAA,UA2BI,SAAgB,EAAA,MAAA,EAAA,CAAW,EAAA,SACjC,EAAA,QAAQ,GACF,EAA8B,GACpC,EAAc,QAAQ,SAAA,GACpB,GAAI,EAAY,IAAe,KAC7B,KAAM,IAAI,OACN,8DACA,cAAa,EAAU,mCACvB,eAEJ,EAAkB,KAAK,EAAY,MAIvC,CAAA,EAAO,EAAA,SAAS,EAAmB,mBAEnC,MAAA,CAAA,EAAO,uBAWyB,EAAgB,GAClD,MAAO,GAAA,IAAI,EAAQ,GC5LrB,AAyLA,GAAM,IAAgC,GAgBtC,YAII,EAAY,GACd,GAAI,GACA,EAEE,EAAiB,EACvB,EAAK,EAAe,GACpB,EAAK,EAAe,GACpB,EAAA,KAAS,OACL,GAAM,MAAQ,GAAM,KACpB,WAAM,MAAA,+OAIF,IAAG,KAEX,GAAM,GACF,GAA0B,QAAS,EAAM,WAAY,GACnD,EACF,GAA0B,SAAU,EAAM,YAAa,GAErD,EAAoB,EAAY,GAAG,MAAM,GAE/C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,OAAO,OACpC,WAAM,MAAA,mBAAmB,EAAM,OAAO,OAAM,4BACxC,aAAY,EAAY,OAAM,oCAC3B,MAAK,UAAU,EAAM,YAAW,OAE3C,EAAA,KAAS,OACL,EAAY,SAAW,EAAM,QAAQ,OACrC,WACI,MAAA,mBAAmB,EAAM,QAAQ,OAAM,6BACvC,aAAY,EAAY,OAAM,sCAC3B,MAAK,UAAU,EAAM,aAAY,OAE5C,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,8BACC,GAAM,WAAW,GAAO,QACrB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,mBAAS,GACP,EAAA,KAAS,OACL,EAAY,GAAQ,MAAM,KAAO,EACjC,WAAM,MAAA,+BACC,GAAM,YAAY,GAAO,QACtB,EAAY,GAAQ,MAAM,GAAE,MAClC,cAAa,EAAS,mBAAmB,EAAM,WAAW,GAAE,QAN7D,EAAS,EAAG,EAAS,EAAY,OAAQ,MAAzC,GAST,MAAO,CAAC,GAAI,EAAa,GAAI,GAG/B,YACI,EAAuB,EAAiB,GAC1C,GAAI,YAAkB,GAAA,OACpB,MAAO,CAAC,GACH,GAAI,MAAM,QAAQ,GACvB,SAAA,KAAS,OACL,EAAO,SAAW,EAAM,OACxB,WAAM,MAAA,wBAAwB,EAAO,OAAM,0BACvC,EAAM,OAAM,iBAAiB,EAAa,SAAS,EAAK,MACzD,EAIP,OAFM,GAAuB,GAEV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,IAAS,KAClB,KAAM,IAAI,GACN,gEACG,GAAa,SAAS,EAAI,OAEnC,EAAO,KAAK,EAAO,IAErB,MAAO,GAIX,YACI,GAMF,GAAI,EAAK,SAAW,EAClB,KAAM,IAAI,IACN,0DAEN,MAAO,CAAC,GAAI,EAAK,GAAI,GAAI,EAAK,gBAO5B,EAAY,EACZ,yKA0BF,GAzBM,EAAqB,EAAK,iBAAmB,KACnD,EAAA,KAAS,OACL,EAAM,WAAa,KACnB,WAAM,MAAA,mGAGV,EAAA,KAAS,OACL,GAAQ,KACR,WAAM,MAAA,kGAEV,EAAA,KAAS,OACL,EAAK,QAAU,MAAQ,EAAK,OAAS,GAAK,OAAO,UAAU,EAAK,QAChE,WAAM,MAAA,gEACF,qBAAoB,EAAK,UACjC,EAAA,KAAS,OACL,CAAC,GACI,EAAK,gBAAkB,GAAK,OAAO,UAAU,EAAK,iBACvD,WAAM,MAAA,gEACF,2CAA0C,EAAK,mBACvD,EAAA,KAAS,OAEJ,EAAa,iBAAsB,KACpC,WAAM,MAAA,sFAGN,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,mDAGX,EAAe,EAAK,gBAAkB,KACxC,EAAK,OACL,EAAK,OACT,AAAI,GACF,CAAI,GAAgB,EAAK,gBACvB,EAAA,KAAS,OACL,EAAK,mBAAqB,MACrB,EAAK,kBAAoB,GACzB,OAAO,UAAU,EAAK,mBAC3B,WAAM,MAAA,yIAGF,YAAW,EAAK,qBAElB,GAAiB,GACnB,EAAK,gBAMT,EAAQ,EAAe,GACvB,EAAQ,EAAe,KAIrB,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAe,OACnB,AAAI,EACF,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,IAAK,MAAA,OAAS,MAEzD,EAAkB,EAAU,QAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YACtD,EAAU,EAAK,SAAW,KAAO,EAAI,EAAK,QAC1C,EAA0B,GAC5B,EAAW,EAAS,EAAK,OAAQ,KAAM,KACvC,GAAiB,EAAS,GAC1B,KACA,EAAc,GAJX,EAAY,EAAA,aAAE,EAAA,EAAA,QAKrB,EAAa,SAAS,GACtB,EAAM,QAAU,EAEhB,CAAA,EAAM,EAAa,uBAAnB,SAAA,OACA,EAAM,cAAgB,GAClB,EAAQ,EAAK,cAAgB,KAAO,EAAI,EAAK,aAE9B,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BACZ,GAAQ,EAAK,OACZ,GAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,KAFR,CAAA,EAAA,iBAExB,GAAA,OACI,EAAY,EACZ,EAAa,EACZ,EAAD,CAAA,EAAA,GACa,CAAA,EAAM,EAAQ,mBAA7B,EAAe,EAAA,8BAEV,GAAqB,EAAY,EAAK,gBAAkB,IACzC,CAAA,EAAM,EAAa,QAD0B,CAAA,EAAA,WAKjE,MAJM,GAAc,EAAA,OAIhB,GAAsB,EAAY,KACpC,SAAQ,KACJ,qCACG,GAAK,gBAAe,MACvB,mDACG,GAAS,cACZ,8HAGG,GAAK,gBAAkB,EAAK,OAAM,eACrC,yEAEJ,CAAA,EAAA,KAGE,EAAY,OAAS,KACjB,GACF,GAA8B,EAAO,EAAY,OAD9C,EAAE,EAAA,GAAE,EAAE,EAAA,GAEP,EAA4B,GAClC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAG,GAAG,MAAM,GAEhC,CAAA,EAAM,EAAa,aAAa,EAAY,KAP1C,CAAA,EAAA,cAOF,EAAA,OAEM,EAA8B,IAChC,GAAK,aAAe,MAApB,MAAA,CAAA,EAAA,IACI,EACF,GAAwB,EAAK,YAAa,EAAM,aAC3C,EAAI,2BAAG,GAAI,EAAqB,OACvC,GAAA,GAAA,GAAc,KAAK,CAAA,EAAM,GACrB,EAAG,GAAI,KAAM,EAAqB,MAFO,CAAA,EAAA,YAC7C,EAAA,MAAA,EAAA,CAAmB,EAAA,4BAD4B,QAAE,iBAUrD,IAHM,EAAM,EAAG,OAAO,GAAI,OAAO,GAC3B,EAAO,EAAc,GAC3B,EAAA,QAAY,GACH,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EAChC,EAAQ,EAAU,GAClB,EAAM,EAAK,GACjB,EAAU,GAAS,EACnB,EAAA,KAAS,GAGX,MAAA,CAAA,EAAM,EAAa,WAAW,EAAY,YAA1C,EAAA,OACA,GAAqB,GAErB,IACA,6BAGE,GAAqB,GAAa,EAAK,gBAClB,EAAY,MAE/B,EACE,GAAO,OACP,GAAgB,EAAK,gBACb,GAAA,GAAO,CAAA,EAAM,EAAM,gBACzB,EAAK,eAAgB,CAAC,QAAS,EAAK,sBAFtC,CAAA,EAAA,KAFF,CAAA,EAAA,IAHF,CAAA,EAAA,YAME,SAAU,EAAA,MAAA,OAAA,CAAO,EAAA,wBAGjB,EAAU,GAAO,EAAM,SAAS,EAAO,EAAO,CAC5C,UAAW,EAAK,qBAAuB,KACnC,GACA,EAAK,oBACT,QAAS,wBAGb,IAAS,EAAI,EAAG,EAAI,EAAM,aAAa,OAAQ,EAAE,EAC/C,EAAU,OAAO,EAAM,aAAa,IAAQ,EAAQ,sBAQxD,MAAA,CAAA,EAAA,YAGF,MAAI,GAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,YAErC,MAFA,GAAA,OACA,IACI,EAAM,cACR,CAAA,EAAA,kBAGJ,MAAA,CAAA,EAAM,EAAa,sBAAnB,SAAA,OACA,CAAA,EAAM,EAAM,QAAQ,oBAApB,SAAA,OACA,CAAA,EAAO,EAAM,iBAEb,SAAM,WAAa,8BAKvB,YACI,EAAqB,GAEvB,GAAI,GAAwB,KAC5B,MAAI,GAAK,iBAAmB,KAC1B,EAAgB,EAAK,gBAChB,AAAI,OAAO,SAAS,EAAQ,OACjC,GAAgB,EAAQ,MAEnB,EAKT,YACI,GAKF,MAAQ,OAAQ,GAAuB,UAAa,WAKtD,YAAiC,GAE/B,MAAQ,OAAQ,GAA6B,MAAS,uBAOpD,EAAY,EACZ,uHAKF,GAJA,EAAO,GAAQ,GACT,EAAa,EAAK,SAAW,KAC7B,EAAI,EAAM,aACZ,EAAqB,GACrB,EAAK,QAAU,EACjB,KAAM,IAAI,IAAoB,8CAGhC,GAAA,KAAS,OACL,CAAC,GAAe,EAAK,QAAU,GAAK,OAAO,UAAU,EAAK,SAC1D,WAAM,MAAA,6DACF,aAAY,KAAK,UAAU,EAAK,YACnB,GAAqB,GACtC,GAAA,SADiB,CAAA,EAAA,UAEjB,MAAA,CAAA,EAAO,EAAuB,mBAA9B,EAAA,EAAA,wBAFE,EAAY,EAId,EAAc,EACd,EAAQ,uEAGU,MAAA,CAAA,EAAM,EAAa,eAmCvC,MAnCM,GAAc,EAAA,OACpB,EAAO,EAAA,KAAS,WACd,GAAI,EAAY,OAGR,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAEL,EAAU,EAAG,OAAO,GACpB,EAAY,EAAA,KAAS,WAAM,MAAA,GAAE,KAGnC,GAFA,EAAA,QAAY,GAER,IAAU,EACZ,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAKrB,OADM,GAAY,EAAQ,GAAG,MAAM,cAC1B,GACP,GAAM,GAAW,EAAU,GACrB,EAAY,EAAK,GACvB,EAAK,GACD,EAAA,KAAS,WAAM,MAAA,GAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,MACvD,AAAI,EAAQ,GACV,EAAA,QAAY,IANP,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,IAA/B,GAST,EAAA,QAAY,GACZ,GAAe,EAEf,EAAE,EAEJ,MAAO,KAGL,EAAY,KACd,CAAI,GACF,QAAQ,KACJ,uJAGA,2BAA0B,EAAK,QAAO,eACtC,qHA3CH,GAAa,EAAQ,EAAK,QAAU,YAAI,CAAA,EAAA,2DAkD/C,IAAS,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC3B,EAAY,EAAK,GACvB,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAC3B,EAAA,QAAY,GAGd,MAAA,CAAA,EAAO,GAAiB,SCrmB1B,YAiK+B,GAC7B,EAAA,KAAS,OACL,EAAY,GAAK,OAAO,UAAU,GAClC,WAAM,MAAA,2DACF,gBAiBN,EAAyB,EAAe,GAC1C,MAAI,IAAU,KACL,CAAC,MACC,MAAM,QAAQ,GAChB,EAAO,IAAI,SAAA,GAAS,MAAA,IAAoB,EAAO,EAAO,EAAO,KAE7D,GAAoB,EAAQ,EAAO,EAAO,eAkBjD,EAAyB,GAC3B,MAAO,GAAA,KAAS,WACd,MAAI,IAAU,KACL,KACE,MAAM,QAAQ,GAChB,EAAO,IACV,SAAA,GAAS,MAAC,IAAqB,EAAO,KAInC,GACH,EAAQ,EAAQ,QAAU,QAAU,EAAU,EAAQ,uBAc5D,EAAc,GAIhB,OAHM,GAAkC,GACpC,EAAa,EACb,EAAmB,KAChB,EAAa,GAClB,EAAW,EAAa,EACxB,AAAI,GAAY,GACd,GAAW,GAEb,EAAO,KAAK,CAAC,EAAY,IACzB,EAAa,EAEf,MAAO,GA8BT,YAGI,EAAY,EAAiC,EAC7C,EAAsB,EAAoB,EAAiB,EAC3D,EAA4B,EAC5B,EAAmB,EAA0B,EAC7C,EAAuB,EACvB,mHAoBF,GAnBA,AAAI,GAAa,MACf,GAAY,IAEd,AAAI,GAAU,MACZ,GAAS,GAEX,AAAI,GAAW,MACb,GAAU,IAEZ,AAAI,GAAgB,MAClB,GAAe,GAIb,EAAe,GACnB,AAAI,GAAQ,MAAQ,GAAU,MAC5B,GAAe,IAGb,GAAmB,MACrB,GAAe,GACX,GAAiB,MACnB,KAAM,IAAI,GACN,oGAKF,SACF,EAAM,gBAAgB,EAAK,EAAW,EAAe,mBAEzD,AAAI,GAAmB,MACrB,GAAa,GAAM,EAAG,IAGxB,AAAI,GAAW,MACb,GAAU,GAGN,EAA0B,GAC5B,EAAW,EAAS,EAAQ,EAAc,EAAiB,EAC3D,EAAW,EAAc,GAFtB,EAAY,EAAA,aAAE,EAAO,EAAA,QAG5B,EAAa,SAAS,GACtB,EAAM,QAAU,EAChB,CAAA,EAAM,EAAa,uBAAnB,EAAA,OACA,EAAM,cAAgB,cAIb,qEACP,MAAA,CAAA,EAAM,EAAa,aAAa,cAAhC,EAAA,OACM,EAA4B,IAC9B,IAAiB,MAAjB,MAAA,CAAA,EAAA,GACF,KAAM,IAAI,IACN,qDAEJ,GAAI,IAAY,QACd,KAAM,IAAI,IAAoB,0CACzB,AAAI,GACT,EAAA,KAAK,QAAQ,GAIT,EAAoB,EAAA,SAAS,GAE7B,EAAU,GAAY,EAAiB,cACpC,2DACD,SAA4B,GAClC,CAAA,EAAM,EAAa,aAAa,EAAY,WAA5C,SAAA,OAEA,EAAA,KAAS,WACP,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,GAAW,GACI,EAAmB,EACnB,EAAW,GAChC,EAAU,MAAW,EACrB,EAAU,KAAU,EAAW,EAM/B,OAFM,IAAW,GAAqB,EAAK,IACrC,GAAO,EAAE,IACN,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAK,IACjB,EAAU,IAAS,GACnB,EAAA,KAAS,IAIX,GAAI,IAAe,EAAQ,OAAS,GAC9B,EAGF,OAFM,IAAU,EAAM,SAAS,EAAM,EAAQ,GAEpC,GAAI,EAAG,GAAI,EAAU,OAAQ,EAAE,IACtC,GAAM,IAAQ,EAAU,IAClB,GAAM,GAAQ,IACpB,EAAA,KAAS,IAET,EAAU,OAAS,IAAS,MAMpC,CAAA,EAAM,EAAa,WAAW,EAAY,WAG1C,MAHA,GAAA,OACA,GAAqB,GAEjB,EAAM,kCA3CH,EAAa,yBAAG,GAAa,EAAQ,YAArC,IAA2C,CAAA,EAAA,iEAAE,QAAE,eAiDxD,EAAkB,2BAGpB,MAAA,CAAA,EAAM,EAAa,WAAW,EAAO,WACrC,MADA,GAAA,OACI,EAAM,kCAtEH,EAAQ,yBAAc,GAAQ,OAA9B,IAAoC,CAAA,EAAA,iEAAE,QAAE,eA0EjD,MAAA,CAAA,EAAM,EAAa,qBAAnB,SAAA,OAEA,CAAA,EAAM,EAAM,QAAQ,mBAApB,SAAA,OACA,CAAA,EAAO,EAAM,0BAMX,EAAY,EACZ,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,yIACF,GAAI,EAAM,WACR,KAAM,IAAI,OACN,gEAEN,EAAM,WAAa,iDASX,EAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIT,EAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAG,EAAG,EAAK,aAAc,EAAK,YAAa,EAC3C,cAHF,EACF,EAAA,OAGJ,EAAS,EAAiB,GAC1B,EAAU,EAAiB,GAC3B,EAAgB,EAAiB,GAG7B,EAAe,GACf,EAAM,QACN,GAAK,gBAAkB,MAAQ,EAAK,eAAe,OAAS,GAA5D,MAAA,CAAA,EAAA,GAEF,GADA,EAAe,GACX,EAAK,eAAe,SAAW,EAEjC,EAAY,EAAK,eAAe,GAChC,EAAY,EAAK,eAAe,OAC3B,MAAI,GAAK,eAAe,SAAW,EAClC,GAAI,IACN,iEAEE,GAAI,GACN,0GAEG,GAAK,eAAc,iBAGtB,SAAiB,GAEnB,CAAA,EAAM,EAAM,oBACR,EAAW,EAAW,KACtB,KACA,EAAgB,WAJlB,SACF,EAAA,OAIJ,EAAO,EAAgB,GACvB,EAAO,EAAgB,GACvB,EAAS,EAAK,OAAO,gBAEhB,AACH,EAAK,iBAAmB,MAAQ,EAAK,gBAAkB,GACvD,EAAK,gBAAkB,EACzB,GAAe,GAET,EACF,KAAK,MAAM,EAAO,GAAG,MAAM,GAAM,GAAI,EAAK,kBACxC,EAAoB,EAAO,GAAG,MAAM,GAC1C,EAAO,GAAY,EAAQ,EAAS,GACpC,EAAS,GAAY,EAAQ,EAAG,GAChC,EAAO,GAAY,EAAS,EAAS,GACrC,EAAU,GAAY,EAAS,EAAG,GAGlC,EAAS,EAAK,OAAO,IAGhB,AAAI,EAAK,iBAAmB,MACjC,GAAe,qBAIX,SAAM,EAAO,OAAO,GAAS,OAAO,GAE1C,EAAM,mCAcA,EAAgB,EAAM,oBACtB,EAAY,EAAM,yBAEpB,EAAW,OACX,EAAe,OACnB,AAAI,EACF,GAAM,mBACN,EAAc,EAAM,aACpB,EACI,EAAU,QAAQ,OAAO,EAAU,IAAI,SAAA,GAAK,MAAA,OAAS,MAEzD,GAAc,KACd,EAAS,GACT,EAAkB,EAAU,SAGxB,EAAY,GAAqB,EAAK,UAAW,EAAK,YAChD,CAAA,EAAM,GACd,EAAO,EAAe,EAAK,EAAW,EAAW,EAAK,OACtD,EAAK,QAAS,EAAW,EAAa,EAAQ,EAAK,QACnD,EAAiB,EAAK,aAAc,KAAM,cAHxC,SAAM,EAAA,OAIZ,CAAA,EAAO,UAEP,SAAM,WAAa,GAEnB,GAAkB,EAAQ,GAC1B,GAAkB,EAAS,GAC3B,GAAkB,EAAkB,GACpC,GAAkB,EAAkB,GACpC,AAAI,GAAiB,MACnB,EAAA,QAAY,yCAYyB,GACzC,GAAM,GAAiB,GACvB,AAAI,YAAmB,GAAA,QACrB,GAAU,CAAC,IAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAS,EAAQ,GACvB,GAAI,EAAO,OAAS,EAClB,EAAK,KAAK,GAAW,EAAQ,SACxB,GAAI,EAAO,OAAS,EACzB,KAAM,IAAI,OACN,yEAGJ,EAAK,KAAK,IAGd,MAAO,eAgBL,EACA,GACF,GAAI,GAAW,KACb,OAEF,GAAM,GAAyB,GAC/B,GAAI,YAAsB,GAAA,OACxB,EAAa,KAAK,EAAW,YACpB,MAAM,QAAQ,GACvB,EAAW,QAAQ,SAAA,GAAK,MAAA,GAAa,KAAK,EAAE,cACnC,GAAc,KAEvB,OAAW,KAAQ,IACjB,GAAM,GAAY,EAAW,GAC7B,EAAa,KAAK,EAAU,IAIhC,GAAM,GAA6B,GACnC,GAAI,YAAmB,GAAA,OACrB,AAAI,EAAa,QAAQ,EAAQ,MAAQ,IACvC,EAAiB,KAAK,WAEf,MAAM,QAAQ,GACvB,EAAQ,QAAQ,SAAA,GACd,AAAI,EAAa,QAAQ,EAAE,MAAQ,IACjC,EAAiB,KAAK,aAGjB,GAAW,KAEpB,OAAW,KAAQ,IACjB,GAAM,GAAS,EAAQ,GACvB,AAAI,EAAa,QAAQ,EAAO,MAAQ,IACtC,EAAiB,KAAK,GAK5B,EAAiB,QAAQ,SAAA,GACvB,AAAK,EAAE,YACL,EAAE,YCrnBR,YA+C6B,GAE3B,MAAO,aAAa,GAAA,mBAMM,GAE1B,MAAO,OAAM,QAAQ,eAMI,GAEzB,MAAO,CAAC,GAAa,IAAM,CAAC,GAAY,eAetC,EAAqD,EACrD,EAAkB,EAAuB,GAC3C,GADoB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACvC,GAAS,MAAQ,EAAM,SAAW,GAGpC,GAAI,GAAQ,MACV,GAAI,GAAoB,GACxB,GAAI,GAAY,IAAU,EAAkB,OAAS,EACnD,EAAoB,WACX,GAAW,IACpB,OAAW,KAAO,GAChB,GAAI,EAAK,eAAe,IACtB,EAAoB,GACpB,WAKJ,GAAoB,GAEtB,GAAI,EACF,KAAM,IAAI,GACN,6BAA6B,EAAe,sBAC5C,YAAW,IAGnB,MAAO,GAET,GAAI,GAAQ,KACV,MAAO,GAAM,IAAI,SAAA,GAAQ,MAAA,QAG3B,GAAI,GACJ,GAAI,GAAW,IACb,EAAO,EACP,EAAS,GACT,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAK,IAAS,KAChB,KAAM,IAAI,GACN,yBAAyB,EAAI,iCAC7B,IAAG,IAET,EAAO,KAAK,EAAK,aAEV,GAAY,IAErB,GADA,EAAO,EACH,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,oCAAmC,EAAM,OAAM,oBAC/C,iDAAgD,IAEtD,EAAS,OAGT,GADA,EAAO,EACH,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,aAAa,EAAe,YAAY,EAAM,OAAM,eACpD,2DACI,EAAK,QAEf,EAAS,CAAC,GAMZ,GAHA,EAAS,GAA2B,GAGhC,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,EAAM,QAErB,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EAEd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MAAQ,GAAU,GAAK,IAAQ,EAC3C,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,mBAAkB,EAAO,GAAE,gCAC3B,KAAI,EAAM,MAAK,QAK3B,MAAO,eAWL,EAAkB,EAAmB,GACvC,GAAM,GAAO,GAAO,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,MAAM,MACpD,EAAK,OACL,GAAM,GAAO,GAAO,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,MAAM,MAGvD,GAFA,EAAK,OAED,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,mFAEA,IAAG,KAAK,UAAU,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,WAElD,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,GACN,oFAEA,IAAG,KAAK,UAAU,EAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,WAErD,GAAI,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,CAAC,EAAA,KAAK,YAAY,EAAM,GAChE,KAAM,IAAI,GACN,kEACA,mBAAkB,EAAK,GAAE,wBAAwB,EAAK,GAAE,YACxD,cAaR,YACI,EAAmB,EAA2B,GAMhD,OAJM,GAAY,CAChB,GAAyB,GACzB,IAEO,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAM,GAAI,EAAQ,GACZ,EAAO,EAAQ,GACf,EAAQ,EAAa,GAC3B,GAAI,GAAQ,KACV,SAEF,GAAI,IAAS,IACP,EAAE,MAAM,EAAE,MAAM,OAAS,KAAO,EAClC,KAAM,IAAI,GACN,2CAA2C,EAAE,MAAK,4JAO1D,GAAI,EAAU,QAAQ,KAAU,GAG9B,OAFM,GAAe,EAAE,MAAM,MAAM,GAC7B,EAAc,EAAM,MAAM,GACvB,EAAI,EAAG,EAAI,EAAa,OAAQ,EAAE,GACzC,GAAM,GAAY,EAAa,GACzB,EAAS,EAAY,GAC3B,GAAI,GAAU,MAAQ,IAAc,EAClC,KAAM,IAAI,GACN,8BAA8B,EAAE,MAAK,sBACrC,oBAAmB,EAAK,uCACxB,2DAiCd,YACI,EAAuB,EAAiB,EACxC,EAAuB,GAAvB,AAAA,IAAA,QAAA,GAAA,IAAuB,AAAA,IAAA,QAAA,GAAA,IACzB,GAAI,GACJ,GAAI,MAAM,QAAQ,IAChB,GAAI,EAAK,SAAW,EAAM,OACxB,KAAM,IAAI,GACN,6BAA6B,EAAe,iFAE5C,wCAAuC,EAAM,OAAM,eACnD,qBAAoB,EAAK,OAAM,iBAErC,EAAS,OAET,GAAI,EAAM,OAAS,EACjB,KAAM,IAAI,GACN,qBAAqB,EAAM,OAAM,IAAI,EAAe,mEAEjD,MAAK,UAAU,EAAK,OAAM,MAEnC,EAAS,CAAC,GAGZ,GAAI,GAAU,KACZ,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAI,EAAO,IAAM,KACf,SAEF,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAM,MAAM,SAAW,EAAO,GAAG,OACnC,KAAM,IAAI,GACN,uBAAuB,EAAe,cAAc,EAAM,GAAE,IAC5D,YAAW,EAAO,GAAG,OAAM,sCAC3B,UAAS,KAAK,UAAU,EAAM,SAEpC,OAAS,GAAI,EAAG,EAAI,EAAO,GAAG,OAAQ,EAAE,GACtC,GAAI,IAAM,GAAK,CAAC,EACd,SAEF,GAAM,GAAM,EAAM,MAAM,GAClB,EAAS,EAAO,GAAG,GACzB,GAAI,GAAU,MACR,IAAW,EACb,KAAM,IAAI,GACN,uBAAuB,EAAe,cACnC,GAAM,GAAE,kBAAkB,KAAK,UAAU,EAAO,IAAG,SACtD,yBAAwB,KAAK,UAAU,EAAM,OAAM,oBAsB/D,EAEA,GACF,GAAI,GAAW,MAAQ,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAClE,MAAO,GAAY,IAAI,SAAA,GAAQ,MAAA,KAGjC,GAAI,GAEJ,GAAI,MAAO,IAAY,UAAY,MAAO,IAAY,WACpD,EAAiB,CAAC,WACT,MAAM,QAAQ,IAAY,MAAO,IAAY,SACtD,EAAiB,MAGjB,MAAM,IAAI,WACN,+DACA,uCAAsC,IAG5C,GAAI,MAAM,QAAQ,GAEhB,MAAO,GAAY,IACf,SAAA,GAAQ,MAAA,KAIZ,OADM,GAAqD,GACxC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACT,EACA,EAAe,eAAe,GAAQ,EAAe,GAAQ,GACjE,AAAK,MAAM,QAAQ,IACjB,GAAgB,CAAC,IAEnB,EAAc,KAAK,GAErB,MAAO,GA6DX,GAAM,IAA2B,8BAcA,EAAA,EAAA,GA4C/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WAAa,KAsCpB,SAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAGF,GAHE,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC1D,CAAC,KAAK,MACR,KAAM,IAAI,GACN,gLAIN,GAAa,KAAM,EAAY,EAAW,IAa5C,EAAA,UAAA,QAAA,SAAQ,GAAR,GAAA,GAAA,KAME,GALA,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAEd,KAAK,KAAO,EAAK,KAEb,MAAO,GAAK,WAAc,SAC5B,KAAK,WAAa,GAAwB,EAAK,WAC/C,KAAK,iBAAmB,QAExB,GAAI,CAAE,GAAK,oBAAqB,GAAA,WAC9B,KAAM,IAAI,GACN,+DAEN,KAAK,WAAa,EAAK,UACvB,KAAK,iBAAmB,GAO1B,GAAI,GAAkC,GACtC,GAAI,CAAC,MAAM,QAAQ,EAAK,OAAS,MAAO,GAAK,MAAS,UAClD,MAAO,GAAK,MAAS,YACvB,EAAK,KAAO,EAAK,KACjB,OAAW,KAAQ,GAAK,KACtB,GAAI,KAAK,YAAY,QAAQ,KAAU,GACrC,KAAM,IAAI,GACN,sCAAsC,EAAI,MAC1C,sCAAqC,KAAK,cAGlD,OAAmB,GAAA,EAAA,EAAA,KAAK,YAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAK,KAAK,IAAS,MACrB,QAAQ,KACJ,WAAW,EAAI,4GAEf,oBAAmB,EAAI,qBAE7B,EAAc,KAAK,GAAW,EAAK,KAAK,cAEjC,MAAM,QAAQ,EAAK,OAC5B,GAAI,EAAK,KAAK,SAAW,KAAK,QAAQ,OACpC,KAAM,IAAI,GACN,+DACA,gCAA+B,KAAK,QAAQ,OAAM,gBAClD,wBAAuB,EAAK,KAAI,MAEtC,GAAM,GAAY,EAAK,KACvB,EAAgB,EAAU,IAAI,SAAA,GAAK,MAAA,IAAW,UAE9C,GAAM,GAAe,GAAW,EAAK,MACrC,KAAK,QAAQ,QAAQ,SAAA,GACnB,EAAc,KAAK,KAIvB,KAAK,cAAgB,EAErB,KAAK,gBAAkB,GACvB,KAAK,iBAAmB,GACxB,KAAK,YAAc,GACnB,OAAS,GAAI,EAAG,EAAI,KAAK,QAAQ,OAAQ,EAAE,GAEzC,GAAM,GAAQ,KAAK,qBAAqB,GAClC,EAAO,KAAK,YAAY,GAC9B,KAAK,gBAAgB,KAAK,GAC1B,KAAK,iBAAiB,KAAK,GAC3B,KAAK,YAAY,KAAK,KAAK,cAAc,IAK3C,GAAM,GAA8B,GAGpC,KAAK,QAAU,EAAK,QAEpB,KAAK,aAAe,CAAC,QACrB,KAAK,eAAiB,GAMtB,GAAU,OAAQ,WAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,GACzC,GAAI,EAAkB,QAAQ,KAAO,GACnC,SAIF,GAAM,GAAe,EAAK,cAAc,GACxC,AAAI,EAAK,QAAQ,OAAS,GACxB,GAAK,eAAe,KAAK,CAAC,EAAc,IACxC,EAAK,aAAa,KAAK,EAAK,YAAY,GAAK,aAQnD,GAAM,GAAgB,GAAe,EAAK,QAAS,KAAK,aAMlD,EACF,SAAC,EAAqB,EACrB,GACC,AAAI,EAAK,YAAY,OAAS,GAC5B,GAAa,EAAK,YAAY,GAAe,IAAM,GAErD,EAAK,aAAa,KAAK,GACvB,EAAK,eAAe,KAAK,CAAC,EAAc,KAG9C,GAAU,SAAU,WAClB,mBAAS,GACP,GAAI,EAAkB,QAAQ,KAAO,oBAGrC,GAAM,GAAgB,EAAc,GAI9B,EAAgB,SAAC,GAOrB,OANM,GAAmB,GACrB,EACA,EACA,aAGO,GACT,GAAI,MAAO,IAAW,UAClB,CAAC,WAAY,MAAO,eAAgB,MAAM,QAAQ,KAC9C,IACN,GAAM,GAAc,EAAK,qBAAqB,GAE9C,AAAI,EAAY,EAAY,OAAS,KAAO,GACxC,EAAK,cAAc,KAAO,GAE5B,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAEL,AACH,EAAK,cAAc,KACnB,GAGF,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAIV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAQ,GACH,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAQ,IAGZ,GAAI,GAAM,OACV,AAAI,CAAC,WAAY,OAAO,QAAQ,KAAY,GAC1C,EAAS,MACJ,AAAI,CAAC,eAAgB,MAAM,QAAQ,KAAY,IACpD,GAAS,MAGX,EAAmB,EACnB,EAAa,EAAmB,OAEhC,GAAM,GAAW,GAAY,GAE7B,EAAmB,EACnB,EACI,EAAmB,GAA4B,GAIrD,GAAI,GACJ,GAAU,EAAY,WACpB,EAAe,IAEjB,EAAa,EAAG,EAAY,IAtDT,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,KAAN,KA0Db,EAAc,IAzEP,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,IAAlC,KAgFX,KAAK,0BAA4B,KAAK,kBAY9B,EAAA,UAAA,iCAAV,WACE,GAAI,KAAK,2BAA6B,KACpC,OAEF,AAAI,KAAK,iBAAiB,SACtB,KAAK,0BAA0B,QACjC,QAAQ,KACJ,sJAqCR,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,GAAe,GAIf,GAAM,GAAiB,GACjB,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GACrD,IAGE,GAAM,GAAM,EAAiB,GAAG,OAAO,EAAiB,IACxD,KAAK,mBACL,GAAM,GAAI,KAAK,aACT,EACF,KAAK,SAAS,EAAG,EAAK,EAAW,EAAK,QAAS,EAAK,OACxD,MAAO,IAAiB,WAExB,GAAkB,EAAiB,GAAI,GACvC,GAAkB,EAAiB,GAAI,KA0BrC,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,YAAK,mBACL,CAAA,EAAO,GAAgB,KAAM,EAAS,SAahC,EAAA,UAAA,gBAAR,SACI,EAAsB,EAAoB,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,SACF,GAAI,GACJ,GAAI,GAAS,MAEX,GADA,EAAa,KACT,GAAa,KACf,KAAM,IAAI,GACN,MAAM,EAAS,gDACf,oBAAmB,YAEhB,GAAO,KAChB,AAAI,MAAM,QAAQ,GAChB,EAAa,EAAI,GAAG,MAAM,GAE1B,EAAa,EAAI,MAAM,OAGzB,MAAM,IAAI,GACN,yDACG,GAAS,yBAElB,MAAO,IAUT,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,GAAI,MAAM,QAAQ,IAAY,EAAQ,SAAW,EAC/C,KAAM,IAAI,GACN,sDAGN,GAAM,GAAiB,MAAM,QAAQ,GAC/B,EACD,EAAiB,EAAsB,CAAC,GACvC,EAAwB,KAAK,wBAAwB,GAGrD,EAAW,GAAI,IAIrB,GAHA,AAAI,YAAkB,GAAA,QACpB,GAAS,CAAC,IAER,MAAM,QAAQ,IAChB,GAAI,EAAO,SAAW,KAAK,OAAO,OAChC,KAAM,IAAI,GACN,kCAAkC,EAAO,OAAM,uDAE/C,KAAI,KAAK,OAAO,OAAM,OAE5B,OAAS,GAAI,EAAG,EAAI,KAAK,OAAO,OAAQ,EAAE,EACxC,EAAS,IAAI,KAAK,OAAO,GAAI,EAAO,QAGtC,QAAoB,GAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAc,EAAO,EAAM,MACjC,GAAI,GAAe,KACjB,KAAM,IAAI,GACN,8CAA8C,EAAM,MAE1D,EAAS,IAAI,EAAO,GAKxB,GAAM,GAAiB,GAAQ,EAAuB,GACtD,MAAO,GAAiB,EAAiB,EAAe,IAMlD,EAAA,UAAA,wBAAR,SAAgC,GAK9B,OAHM,GACF,GAAa,KAAM,EAAoB,QACvC,EAAmB,EAAoB,OACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAIlB,OAJS,GAAK,EAAA,GACR,EACF,MAAM,QAAQ,EAAM,QAAU,EAAM,OAAS,CAAC,EAAM,QAClD,EAAmB,EAAa,IAAI,SAAA,GAAU,MAAA,GAAO,OAClD,EAAI,EAAG,EAAI,EAAoB,OAAQ,EAAE,GAChD,GAAM,GAAQ,EAAiB,QAAQ,EAAoB,IAK3D,GAJA,AAAI,IAAU,IACZ,GAAsB,GAAK,EAAa,GACxC,KAEE,IAAqB,EACvB,MAGJ,GAAI,IAAqB,EACvB,MAIJ,GAAI,EAAmB,GACrB,GAAM,GAA2B,GACjC,QAAsB,QAAQ,SAAC,EAAQ,GACrC,AAAI,GAAU,MACZ,EAAe,KAAK,EAAoB,MAGtC,GAAI,GACN,mDACA,IAAG,KAAK,UAAU,KAExB,MAAO,IAgBD,EAAA,UAAA,YAAR,SAAoB,EAAsB,EAAgB,GAA1D,GAAA,GAAA,KAA0C,MAAA,KAAA,QAAA,GAAA,IAAgB,AAAA,IAAA,QAAA,GAAA,IAEjD,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,GACxC,GAAI,EACF,KAAM,IAAI,IACN,iDAYN,OAJM,GAAU,GAAY,EAAY,GAClC,EAA0B,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,gBAGlD,GACP,GAAM,GAAY,EAAA,KAAS,WACzB,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAG/B,EAAW,GAAY,EAAK,EAAY,GAGxC,EAAQ,GACd,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,EAAE,EACrC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAS,SAGnD,GAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,IAE1C,GAAM,GAAW,GAAI,IAAS,GAC9B,MAAO,IAAQ,EAAK,QAAS,KAE/B,EAAU,QAAQ,SAAC,EAAU,GAAM,MAAA,GAAY,GAAG,KAAK,MApBhD,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,IAA/C,GAsBT,MAAO,IACH,EAAY,IAAI,SAAA,GAAW,MAAA,GAAA,OAAW,EAAS,SA+BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,AAAA,IAAA,QAAA,GAAA,IAC1B,GAAM,GAAkB,GAA2B,GACnD,GACI,EAAiB,KAAK,WAAY,KAAK,gBAAiB,IAC5D,IAKE,GAAM,GAAY,EAAK,WAAa,KAAO,GAAK,EAAK,UACrD,UAAe,GACR,KAAK,YAAY,EAAiB,WAEzC,GAAkB,EAAiB,KAmBvC,EAAA,UAAA,eAAA,SAAe,GACb,GAAe,EAAG,KAAK,WAAY,KAAK,gBAAiB,IAGzD,GAAM,GAAa,OAAM,QAAQ,GAAK,EAAE,GAAK,GAAG,MAAM,GACtD,MAAO,MAAK,YAAY,EAAG,IAGnB,EAAA,UAAA,sBAAV,SACI,EACA,EAAkD,EAClD,GAEF,GAAI,KAAK,YAAc,KACrB,KAAM,IAAI,IACN,gGAIN,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,KAAK,iBAAiB,OAAQ,EAAE,GAClD,GAAM,GAAc,KAAK,iBAAiB,GACpC,EAAS,KAAK,YAAY,GAChC,AAAI,IAAW,GACb,EAAa,KACT,EAAY,MAAM,EAAG,EAAY,OAAS,GAAG,OAAO,CAAC,KAGzD,EAAa,KAAK,GAWtB,GARA,EAAI,GACA,EAAG,KAAK,eAAgB,KAAK,gBAAiB,GAAO,SACzD,EAAI,GACA,EAAG,KAAK,gBAAiB,EAAc,GAAO,UAElD,GAAkB,EAAG,GAErB,GAAgC,EAAG,KAAK,YAAa,KAAK,kBACtD,KAAK,UAAY,GAAa,MAAQ,EAAY,GAChD,EAAE,GAAG,MAAM,GAAK,IAAc,EAChC,KAAM,IAAI,GACN,mHAEG,GAAS,YAAY,EAAE,GAAG,MAAM,GAAE,gBAG7C,MAAO,CAAC,EAAG,IAGG,EAAA,UAAA,oBAAhB,SACI,EACA,EACA,EACA,EACA,EACA,GADA,MAAA,KAAA,QAAA,GAAA,2GAKF,GAHM,EACF,KAAK,sBAAsB,EAAG,EAAG,EAAgB,GAD9C,EAAU,EAAA,GAAE,EAAU,EAAA,GAGzB,GAAgB,KAClB,KAAM,IAAI,OAAM,0CAGd,EAAkC,MAClC,IAAe,MAAf,MAAA,CAAA,EAAA,GACI,EACF,GAAwB,EAAa,KAAK,aAC9C,EAAwB,GACf,EAAI,yBAAG,GAAI,EAAa,OAC/B,GAAA,GAAA,GAAsB,KAClB,CAAA,EAAM,GAAmB,EAAW,GAAI,KAAM,EAAa,MAF1B,CAAA,EAAA,UACrC,EAAA,MAAA,EAAA,CACI,EAAA,0BAFmC,QAAE,eAO7C,MAAA,CAAA,EAAO,CAAC,EAAY,EAAY,UAc1B,EAAA,UAAA,SAAR,SACI,EAAiC,EAAe,EAChD,EAAa,GAFjB,GAAA,GAAA,KAEI,MAAA,KAAA,QAAA,GAAA,GACK,EAAA,KAAS,WACd,GAAM,GAAa,EAAK,gBAAgB,EAAK,EAAW,EAAO,SACzD,EAAiB,GACvB,GAAI,EAAU,EACZ,KAAM,IAAI,IAAoB,wCAGhC,GAAI,GAAS,KACX,KAAM,IAAI,IACN,mDAIJ,OAFM,GAAU,GAAY,EAAY,GAClC,EAAa,EAAA,SAAS,GAAM,EAAG,IAC5B,EAAa,EAAG,EAAa,EAAQ,OAAQ,EAAE,GACtD,GAAM,GAAa,EAAQ,GAAY,GACjC,EAAW,EAAQ,GAAY,GAC/B,EACF,GACI,EAAY,EAAY,EAAW,GAGrC,EAAW,GAAqB,EAAK,GACrC,EAAY,EAAE,GACpB,GAAI,IAAe,EACjB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,EACtC,EAAK,KAAK,EAAA,OAAO,IAGrB,OAAS,GAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAW,EAAU,GAC3B,EAAK,GACD,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAQ,EAAW,EAAY,KAGxD,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAK,GAAK,EAAA,IAAQ,EAAK,GAAI,GAG/B,MAAO,MAID,EAAA,UAAA,uBAAV,WAKE,OAJM,GAAY,KAAK,aAGjB,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAU,OAAQ,EAAE,GACtC,GAAM,GAAQ,EAAU,GACpB,EAAW,EACf,GAAI,GAAM,EAAW,GAAS,GAC5B,GAAM,GAAW,GAAM,EAAU,MAAM,EAAG,GAAI,GAC9C,GAAY,IAAI,EAElB,EAAiB,KAAK,GAExB,MAAO,IAaC,EAAA,UAAA,kBAAV,WAAA,GAAA,GAAA,KACE,MAAO,UAAC,GACN,GAAM,GAAuB,GAEvB,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAgB,EAAK,MACvB,EAAK,OAAO,OAAS,EAAK,QAAQ,OAClC,EAAK,OAAO,OAAS,EAAK,QAAQ,OAAS,GAEzC,EAA0B,GAK1B,EAAoB,WAExB,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KASjD,OAPM,GAAW,GAAI,IAAS,GACxB,EACF,GAAQ,EAAK,QAAS,EAAU,CAAC,SAAY,KAI7C,EACK,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GACpC,EAAO,EAAa,EAAQ,GAAI,EAAQ,IAC5C,AAAI,EAAc,IAAM,MACtB,GAAO,GAAoB,EAAM,EAAc,KAIjD,GAAM,GAAmB,EAAA,KAAS,GAElC,EAAW,KAAK,GAChB,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAOnC,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAI,GAAc,OAElB,GAAI,EAAK,QAAQ,OAAS,GAAK,EAAI,EAAK,QAAQ,OAC9C,EAAiB,EAAW,QAE5B,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAC3C,EACI,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAGpD,EAAA,KAAS,GAET,EAAc,KAAK,GAGrB,SAAY,EAAA,KAAS,GAGrB,EAAK,kBAAkB,QAAQ,SAAA,GAC7B,EAAY,EAAA,IAAQ,EAAW,KAG1B,GAGH,EAAY,EAAK,0BAA0B,IAC7C,SAAA,GAAS,MAAA,GAAM,SACb,EAAa,GACb,EACF,EAAK,WAAW,SAAS,EAAmB,EAAY,GAE5D,MAAO,CAAC,GAAgB,OAAO,KAS3B,EAAA,UAAA,iBAAR,WAAA,GAAA,GAAA,KACE,KAAK,aAAe,SAAC,GACnB,MAAO,GAAA,KAAS,WAOd,OANM,GAAuB,GACzB,EACE,EAAS,EAAK,MAAM,EAAG,EAAK,OAAO,QACnC,EAAU,EAAK,MACjB,EAAK,OAAO,OAAQ,EAAK,OAAO,OAAS,EAAK,QAAQ,QACpD,EAAQ,GACL,EAAI,EAAG,EAAI,EAAK,OAAO,OAAQ,EAAE,EACxC,EAAM,KAAK,CAAC,IAAK,EAAK,OAAO,GAAI,MAAO,EAAO,KAKjD,OAHM,GAAW,GAAI,IAAS,GACxB,EAAU,GAAQ,EAAK,QAAS,GAE7B,EAAI,EAAG,EAAI,EAAK,cAAc,OAAQ,EAAE,GAC/C,GAAM,GAAe,EAAK,cAAc,GAGlC,EAAe,EAAA,KAAS,EAAa,EAAQ,GAAI,EAAQ,KAC/D,AAAI,IAAM,EACR,EAAY,EAEZ,EAAY,EAAA,IAAQ,EAAW,GAEjC,EAAW,KAAK,GAGlB,OAAS,GAAI,EAAG,EAAI,EAAK,eAAe,OAAQ,EAAE,GAChD,GAAM,GAAS,EAAK,eAAe,GAAG,GAChC,EAAc,EAAK,eAAe,GAAG,GAErC,EACF,EAAA,KAAS,EAAO,EAAQ,GAAc,EAAQ,KAClD,EAAW,KAAK,GAElB,MAAO,OAuCP,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,MAAA,CAAA,EAAO,GAAW,KAAM,EAAG,EAAG,SA0B1B,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,MAAA,CAAA,EAAO,GAAW,KAAM,EAAS,SA0B7B,EAAA,UAAA,aAAN,SACI,EACA,qHAIqB,MAAA,CAAA,EAAM,KAAK,oBAAoB,EAAG,WAAnD,EAAiB,EAAA,OACjB,EAAS,EAAe,GACxB,EAAU,EAAe,GACzB,EAAgB,KAAK,oBACrB,EAAS,EAAc,EAAO,OAAO,IACrC,EAAuB,OACV,EAAA,yBAAA,GAAA,EAAA,OAAR,GAAI,EAAA,GACH,CAAA,EAAM,EAAK,SADE,CAAA,EAAA,UACjB,EAAI,EAAA,OACV,EAAW,KAAK,EAAE,qBAFD,wBAInB,SAAA,QAAY,GACZ,CAAA,EAAO,GAAiB,UAYhB,EAAA,UAAA,gBAAV,SAA0B,GAMxB,OALM,GAA8B,GAE9B,EAAgB,GAAU,MAAQ,EAAO,cACzC,EAAU,EAAgB,KAAK,iBAAmB,KAAK,QACvD,EAAe,KAAK,WAAW,GAC5B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,GACpC,GAAI,GAAiB,CAAC,EAAQ,GAAG,UAE/B,SAEF,EAAa,KACT,CAAC,KAAM,EAAQ,GAAG,aAAc,OAAQ,EAAa,KAE3D,MAAO,IAiCT,OAAA,eAAI,EAAA,UAAA,eAAY,KAIhB,WACE,MAAO,MAAK,mBALd,SAAiB,GACf,KAAK,cAAgB,mCAOvB,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GACZ,AAAI,KAAK,aAAe,GACtB,MAAK,WAAa,EAClB,KAAK,iBAAmB,qCAI5B,EAAA,UAAA,QAAA,WACE,GAAM,GAAS,EAAA,UAAM,QAAO,KAAA,MAC5B,GAAI,EAAO,uBAAyB,GAAK,KAAK,WAAa,MACvD,KAAK,kBACP,GAAM,GAAmC,EAAA,SAAa,WACtD,KAAK,WAAW,UAChB,EAAO,sBACH,EAAmC,EAAA,SAAa,WAEtD,MAAO,IAGD,EAAA,UAAA,mBAAR,WAEE,GAAI,GAEJ,GAAI,MAAO,MAAK,MAAS,SACvB,EAAY,GAAY,KAAK,cACpB,MAAM,QAAQ,KAAK,OAC5B,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,MAAO,IAAS,SAClB,KAAM,IAAI,OAAM,sDAGpB,EAAa,KAAK,KAAkB,IAAI,SAAA,GAAQ,MAAA,IAAY,UAG5D,GAAM,GAAc,OAAO,KAAK,KAAK,MACrC,EAAY,GAGZ,OAFM,GACF,KAAK,KACgB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,GAAI,MAAO,GAAO,IAAgB,SAChC,EAAU,GACN,GAAY,EAAO,QAEvB,MAAM,IAAI,OAAM,uDAItB,MAAO,IAGD,EAAA,UAAA,qBAAR,WAEE,GAAI,MAAO,MAAK,SAAY,UACxB,MAAO,MAAK,SAAY,WAC1B,MAAO,CAAC,GAAY,GAA4B,KAAK,WAChD,GAAI,MAAM,QAAQ,KAAK,SAC5B,MAAO,MAAK,QAAQ,IAChB,SAAA,GAAU,MAAA,IAAY,GAA4B,MAEtD,GAAM,GAAyD,GAC/D,OAAW,KAAO,MAAK,QACrB,EAAmB,GACf,GAAY,GAA4B,KAAK,QAAQ,KAE3D,MAAO,IAID,EAAA,UAAA,kBAAV,WACE,MAAO,CACL,KAAM,KAAK,qBACX,QAAS,KAAK,uBACd,iBAAkB,CAChB,WAAY,KAAK,UAAU,eAC3B,OAAQ,KAAK,UAAU,eAQ7B,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,EAAe,kBAAoB,KACrC,KAAM,IAAI,OAAM,gDAElB,GAAI,EAAe,cAAgB,KACjC,KAAM,IAAI,OAAM,8CAElB,GAAI,EAAe,oBAAsB,KACvC,KAAM,IAAI,OAAM,oDAGlB,GAAM,GAAW,GAAoB,EAAe,kBAE9C,EAAY,GAAY,GAE1B,EACJ,GAAI,MAAO,GAAe,MAAS,SACjC,EAAO,GAAY,EAAe,cACzB,MAAM,QAAQ,EAAe,MACtC,EAAO,EAAe,KAAK,IAAI,SAAA,GAAa,MAAA,IAAY,aAC/C,EAAe,MAAQ,MAChC,EAAO,GACP,OAAW,KAAO,GAAe,KAC/B,EAAK,GAAO,GAAY,EAAe,KAAK,IAIhD,GAAI,GACJ,GAAI,MAAM,QAAQ,EAAe,SAC/B,EAAU,EAAe,QAAQ,IAAI,SAAA,GAAU,MAAA,IAAY,aAClD,EAAe,SAAW,MACnC,EAAU,GACV,OAAW,KAAO,GAAe,QAC/B,EAAQ,GAAO,GAAY,EAAe,QAAQ,IAItD,KAAK,QAAQ,CAAC,KAAI,EAAE,QAAO,EAAE,UAAS,KAoFlC,EAAA,UAAA,KAAN,SAAW,EAAmC,+HAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,GACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,GACN,gHAKF,MAAA,CAAA,EAAM,EAAA,GAAG,cAAc,KAAK,gBAAgB,kBAD1C,GACF,EAAA,OAEE,EAAe,GACf,EAAgB,KAChB,EAAc,KAAK,OAAO,EAAW,GACrC,EAAoC,CACxC,cAAe,EACf,OAAQ,GACR,YAAa,8BAA8B,GAC3C,YAAa,MAGT,EAAmB,GAAU,KAAO,GAAQ,EAAO,iBACrD,GAAoB,KAAK,WAAa,KACxC,GAAe,eAAiB,KAAK,oBAC/B,EAAa,YAET,EAAA,GAAA,EAAA,IAAG,cAAc,CAAA,EAAM,KAAK,UAAU,eAJ9C,CAAA,EAAA,UAIE,MAAA,CAAA,EAAM,EAAA,MAAA,EAAA,CAAiB,EAAA,OAAmC,YADxD,EACF,EAAA,OADS,EAAmB,EAAA,KAAS,EAAoB,EAAA,MAE7D,GAAA,EAAmB,OAAM,KAAI,MAAA,EAAI,GACjC,EAAmB,KAAO,EAAA,GAAG,wBACzB,CAAC,EAAmB,KAAM,qBAGhC,MAAI,MAAK,qBAAuB,MAExB,GAAY,GAClB,GAAyB,KAAK,oBAAqB,KAAK,KAAM,GAC9D,EAAe,oBAAsB,KAAK,qBAG5C,EAAe,WAAa,EAAmB,KAC/C,EAAe,YAAc,EAAmB,MAChD,CAAA,EAAO,EAAa,KAAK,UAW3B,EAAA,UAAA,uBAAA,SAAuB,GACrB,GAAyB,EAAqB,KAAK,MACnD,KAAK,oBAAsB,GAc7B,EAAA,UAAA,uBAAA,WACE,MAAO,MAAK,qBAv4CP,EAAA,UAAY,QAy4CrB,GA74CiC,IA84CjC,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DACS,SAAA,UAAY,aACrB,GAFgC,IAGhC,EAAA,cAAc,cAAc,IC72D5B,YA8DI,EACA,uHACF,AAAM,iBAAmB,IACvB,GAAwB,CAAC,cAAe,IAE1C,EAAwB,EAEpB,EAAgB,EAAsB,cAC1C,AAAI,EAAc,cAAmB,MAMnC,GAAgB,EAAc,cAE1B,EACF,GAAoB,GAClB,EAAQ,GAAY,EAAU,GAEhC,EAAsB,iBAAmB,KAItB,CAAA,EAAM,EAAA,GAAG,YAC1B,EAAsB,gBAAiB,EAAsB,WAC7D,EAAM,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,iBANrC,CAAA,EAAA,UAUF,IANM,EAAe,EAAA,OAKf,EAAqC,GAC3C,EAAA,EAAqB,EAAA,EAAM,QAAN,EAAA,EAAA,OAAA,IAAV,EAAM,EAAA,GACf,EAAmB,EAAO,cACtB,EAAa,EAAO,cAG1B,EAAM,YAAY,GAElB,EAAA,QAAQ,oBAEV,MAAA,CAAA,EAAO,oBAsIL,EACA,4EAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,MAAO,IAAoB,UAE7B,GADM,EAAW,EAAA,GAAG,gBAAgB,EAAiB,GACjD,EAAS,SAAW,EAKtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAiB,YAC5C,EAAS,OAAS,EAC3B,KAAM,IAAI,GACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAe,MAE7B,EAAkB,EAAS,GAE7B,MAAA,CAAA,EAAO,GAA6B,EAAiB,OAAW,oBAc9D,EAAuB,EACvB,mHAIF,GAHA,AAAI,GAAW,MACb,GAAU,IAER,EAAQ,MAAQ,KAClB,KAAM,IAAI,GACN,iHAGY,MAAA,CAAA,EAAM,EAAQ,kBAA1B,EAAY,EAAA,OACd,EAAgB,EAAU,cAC9B,AAAI,EAAc,cAAmB,MACnC,GAAgB,EAAc,cAG1B,EAAS,EAAQ,QAAU,KAAO,GAAO,EAAQ,OAMjD,EACF,EAAU,YAAc,MAAQ,EAAU,aAAe,MAAQ,EAC/D,EACF,GACI,GAAoB,GACpB,EAAe,GAEjB,EAAiB,EAAU,eACjC,AAAI,GAAkB,MACpB,EAAM,mBAAmB,GAE3B,AAAI,EAAU,qBAAuB,MACnC,EAAM,uBAAuB,EAAU,sBAIrC,GAAU,YAAc,MAAxB,MAAA,CAAA,EAAA,GAEF,GAAI,EAAU,aAAe,KAC3B,KAAM,IAAI,GACN,wHAIA,GAAmC,GACrC,EAAU,WAAY,EAAU,aAD7B,EAAY,EAAA,aAAE,EAAgB,EAAA,iBAErC,EAAM,YAAY,EAAc,GAE5B,EAAM,WAAa,MAAQ,EAAiB,OAAS,EACvD,CAAA,EAAM,EAAM,UAAU,WAAW,IAD/B,CAAA,EAAA,UACF,EAAA,wBAIF,EAAA,QAAQ,GACR,EAAA,QAAQ,EAAiB,IAAI,SAAA,GAAK,MAAA,GAAE,2BAEtC,MAAA,CAAA,EAAO,QAGT,YACI,EAAqB,GAEvB,GAAM,GAAc,EAAA,GAAG,cAAc,EAAQ,GACvC,EAA+B,GAC/B,EAAkC,GACxC,SAAM,QAAQ,SAAA,GACZ,AAAI,EAAK,QAAU,YACjB,EAAiB,KAAK,CAAC,KAAM,EAAK,KAAM,OAAQ,EAAY,EAAK,QAEjE,EAAa,EAAK,MAAQ,EAAY,EAAK,QAGxC,CAAC,aAAY,EAAE,iBAAgB,sBAwCR,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,CAAC,OAAQ,GAAI,QAAS,MAAI,KAUhC,GATA,EAAO,GAAQ,GAEf,EAAK,UAAY,GACjB,EAAK,MAAQ,GAGb,EAAK,KAAQ,EAAK,MAAQ,KAAQ,EAAK,KAAO,GAAO,eAGjD,EAAK,QAAU,KACjB,OAAoB,GAAA,EAAA,EAAA,EAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,EAAK,IAAI,YAOP,SAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAQ,EAAM,aAAa,GAAG,cAAc,GAAG,MACrD,GAAI,EAAM,KAAK,SAAA,GAAK,MAAA,GAAI,IACtB,KAAM,IAAI,GACN,kDACG,GAAM,KAAI,uBACV,GAAM,aAAa,GAAG,aAAa,GAAG,MAAK,OAyBtD,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GACF,YAAiB,IAAc,YAAiB,IAChD,EACJ,GAAI,GAEF,GADA,EAAa,EACT,EAAW,QAAQ,SAAW,EAChC,KAAM,IAAI,GACN,yHAKN,GAAI,EAAW,OAAO,SAAW,EAC/B,KAAM,IAAI,GACN,uHAOR,GAAI,KAAK,QAAQ,SAAW,GAE1B,GAAI,EAAM,aAAa,SAAW,GAEhC,GAAI,EAAM,iBAAmB,KAC3B,KAAM,IAAI,GACN,iGAIN,GAAM,GAAI,GAAM,CACd,WAAY,EAAM,gBAClB,MAAO,EAAM,MACb,KAAM,EAAM,KAAO,WAIrB,EAAM,MAAM,GAGd,GAAI,EACF,KAAK,QAAU,EAAW,QAC1B,KAAK,OAAS,EAAW,YAEzB,GAAI,EAAM,aAAa,SAAW,EAChC,KAAM,IAAI,GACN,2DACA,yDACI,EAAM,KAAI,KACd,cAAa,EAAM,aAAa,OAAM,0BACtC,gBAGN,GAAI,EAAM,aAAa,GAAG,cAAc,SAAW,EACjD,KAAM,IAAI,GACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,EAAM,aAAa,GAAG,cAAc,IACpD,KAAK,OAAS,GAAgB,KAAK,QAAQ,IAG7C,KAAK,aAAe,GAKpB,GAAI,IAAK,CACP,cAAe,KACf,cAAe,GACf,YAAa,GACb,cAAe,GACf,aAAc,KAAK,OACnB,cAAe,KAAK,QAEpB,WAAY,GAA2B,KAAM,KAAK,OAAO,QACzD,YAAa,CAAC,MACd,YAAa,KAAK,OAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QACpC,aAAc,KAAK,QAAQ,GAAG,aAGhC,GAAM,GAAe,EAAM,MAAM,KAAK,QAAQ,IAC9C,GAAI,MAAM,QAAQ,GAChB,KAAM,IAAI,WACN,yHAKN,KAAK,WAAW,GAChB,KAAK,QAAU,CAAC,GAEhB,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,OAGvD,KAAK,OAAO,KAAK,GACjB,KAAK,MAAQ,IAQf,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,OAAO,SAAW,EACzB,KAAM,IAAI,WAAU,qCAItB,GADA,KAAK,OAAO,MACR,KAAK,OAAO,SAAW,EACzB,KAAK,QAAU,GACf,KAAK,aAAe,GACpB,KAAK,cAAgB,QAErB,GAAM,GAAiB,KAAK,OAAO,OAAS,EAC5C,KAAK,OAAO,GAAgB,cAAgB,GAC5C,KAAK,QAAU,CAAC,KAAK,OAAO,GAAgB,QAE5C,KAAK,aAAa,GAAG,cAAgB,KAAK,QAC1C,KAAK,aAAa,GAAG,aAAe,CAAC,KAAK,QAAQ,GAAG,SAIzD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,KAAK,EAAQ,IAGjC,EAAA,UAAA,MAAA,SAAM,GAKJ,GAFA,GAAmB,GAEf,KAAK,OAAO,SAAW,GAAK,KAAK,QAAQ,SAAW,EACtD,KAAM,IAAI,WACN,4EAIN,KAAK,MAAQ,GAAI,IAAY,CAC3B,OAAQ,KAAK,OACb,QAAS,KAAK,QAAQ,GACtB,KAAM,KAAK,KAAO,WAEpB,KAAK,MAAM,UAAY,KAAK,UAG5B,KAAK,gBAAkB,KAAK,MAAM,gBAElC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,uBAAyB,KAAK,MAAM,uBACzC,KAAK,yBAA2B,KAAK,MAAM,yBAC3C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,wBAA0B,KAAK,MAAM,wBAC1C,KAAK,0BAA4B,KAAK,MAAM,0BAC5C,KAAK,aAAe,KAAK,MAAM,aAC/B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,YAAc,KAAK,MAAM,YAC9B,KAAK,WAAa,KAAK,MAAM,WAG7B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,WACE,MAAK,MAAK,OACR,KAAK,QAEA,EAAA,UAAM,YAAW,KAAA,OAiC1B,EAAA,UAAA,QAAA,SACI,EAAqB,EACrB,GAAA,AAAA,IAAA,QAAA,GAEoD,QAAQ,KAC9D,AAAK,KAAK,OACR,KAAK,QAEP,EAAA,UAAM,QAAO,KAAA,KAAC,EAAY,EAAW,IASvC,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,OAAS,MAChB,KAAK,QAEP,KAAK,MAAM,WAAW,IAmCxB,EAAA,UAAA,SAAA,SACI,EAAoB,EACpB,GACF,GADE,AAAA,IAAA,QAAA,GAAA,IACE,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAO,MAAK,MAAM,SAAS,EAAG,EAAG,IAyB7B,EAAA,UAAA,gBAAN,SAAsB,EAAsB,sEAE1C,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAEN,MAAA,CAAA,EAAO,KAAK,MAAM,gBAAgB,EAAS,SA8B7C,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,IAC1B,AAAI,KAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,QAAQ,EAAG,IAU/B,EAAA,UAAA,eAAA,SAAe,GACb,MAAI,MAAK,OAAS,MAChB,KAAK,QAEA,KAAK,MAAM,eAAe,IAQnC,EAAA,UAAA,QAAA,SAAQ,GACN,KAAK,QACL,KAAK,MAAM,QAAQ,GACnB,KAAK,WAAa,KAAK,MAAM,UAE7B,KAAK,iBAAoB,KAAK,MAAc,iBAC5C,KAAK,KAAO,KAAK,MAAM,KACvB,KAAK,QAAU,KAAK,MAAM,QAG1B,KAAK,eAAiB,KAAK,MAAM,eACjC,KAAK,aAAe,KAAK,MAAM,cAIjC,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAO,OAAY,KAAK,MAAM,eAGrD,SAAc,GACZ,KAAK,MAAM,UAAY,mCAkCnB,EAAA,UAAA,IAAN,SACI,EACA,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,gEACF,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,IAAI,EAAG,EAAG,SAwFxB,EAAA,UAAA,WAAN,SAAoB,EAAqB,sEAEvC,GAAI,CAAC,KAAK,MACR,KAAM,IAAI,IACN,qDAGN,MAAA,CAAA,EAAO,KAAK,MAAM,WAAW,EAAS,SA0BlC,EAAA,UAAA,aAAN,SACI,EACA,sEAEF,MAAA,CAAA,EAAO,KAAK,MAAM,aAAa,EAAG,SAK7B,EAAA,WAAP,SACI,EACA,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GACA,EAA6C,GACjD,GAAI,YAAkB,QACpB,GAAI,CAAE,GAAO,GAAG,WAAa,OACzB,EAAO,GAAG,YAAiB,QAC7B,KAAM,IAAI,GAAW,kDAEvB,EAAc,MAEd,GAAA,KAAK,OACD,EAAO,QAAa,KACpB,WACI,MAAA,wHAER,EAAc,EAAO,OACrB,MAAO,GAAO,OACd,EAAmB,EAGrB,GAAM,GAAQ,GAAI,GAAI,GACtB,GAAI,CAAE,aAAiB,IACrB,KAAM,IAAI,IACN,yDAAyD,GAE/D,OAAmB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACP,EAA0C,OAC1C,EAAQ,GACI,EAAkC,EAClC,GAClB,AAAI,GACF,EAAM,6BAA6B,IAErC,EAAM,IAAI,GAEZ,MAAO,IA+BT,OAAA,eAAI,EAAA,UAAA,eAAY,KAWhB,WACE,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,MAAO,MAAK,MAAM,kBAjBpB,SAAiB,GAGf,GAAI,KAAK,OAAS,KAChB,KAAM,IAAI,GACN,qFAGN,KAAK,MAAM,aAAe,mCAe5B,EAAA,UAAA,UAAA,WAME,OADM,GAAqC,GACvB,EAAA,EAAA,EAAA,KAAK,OAAL,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACR,EAAiC,GACvC,EAAK,UAAe,EAAM,eAC1B,EAAK,OAAY,EAAM,YACvB,EAAO,KAAK,GAEd,MAAO,CAAC,KAAM,KAAK,KAAM,OAAM,IAvsB1B,EAAA,UAAY,aAysBrB,GA3sBgC,IA4sBhC,EAAA,cAAc,cAAc,IC3kC5B,YAqEsB,GACpB,MAAO,IAAI,IAAY,eA+DE,GACzB,MAAO,IAAI,IAAW,eA+FpB,EACA,GACF,MAAI,IAAW,MACb,GAAU,IAEL,GAAwB,EAAiB,eA0B5B,GACpB,MAAO,IAAM,eAIX,EACA,GACF,GAA4B,4BACxB,EAAgB,GC5QtB,AAwBA,GAAA,IAAA,SAAA,GAAyC,EAAA,EAAA,GAAzC,4DAEE,SAAA,UAAA,UAAA,WACE,MAAO,IAEX,GALyC,EAAA,cAAc,cAWvD,GAAA,SAAA,GAAyB,EAAA,EAAA,GAAzB,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,GAAM,EAAG,IATF,EAAA,UAAY,MAW9B,GAbyB,IAczB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAK,EAAA,KAAS,OAF9B,EAAA,UAAY,QAI9B,GAN2B,IAO3B,EAAA,cAAc,cAAc,IAG5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAA5B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAFO,EAAA,UAAY,SAI9B,GAN4B,IAO5B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,QAAY,IAFL,EAAA,UAAY,UAI9B,GAN6B,IAO7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAAjC,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAc,IAFP,EAAA,UAAY,cAI9B,GANiC,IAOjC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,SAAa,IAFN,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAA9B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAW,IAFJ,EAAA,UAAY,WAI9B,GAN8B,IAO9B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAA1B,4DAGE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,GAAA,KAAS,IAFF,EAAA,UAAY,OAI9B,GAN0B,IAO1B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAA7B,4DAeE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,QAAY,EAAG,IAdR,EAAA,UAAY,UAgB9B,GAlB6B,IAmB7B,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAAhC,4DAgBE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAgB,IACxB,EAAA,WAAe,EAAG,IAfX,EAAA,UAAY,aAiB9B,GAnBgC,IAoBhC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAA3B,4DAUE,SAAA,UAAA,MAAA,SAAM,EAAW,GAAA,MAAA,KAAA,QAAA,GAAA,GACR,EAAA,KAAK,WAAM,MAAA,GAAA,QAAY,EAAE,IAAI,IAAQ,IAAI,MATlC,EAAA,UAAY,QAW9B,GAb2B,IAc3B,EAAA,cAAc,cAAc,gBAEQ,GAClC,MAAO,GAAW,2BAIjB,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACM,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,0BAGS,GAE5B,GAAI,GAAc,MAChB,GAAM,GAAmC,GACzC,SAAO,UAAe,SACtB,EAAO,OAAY,GACZ,GAAsB,GAE/B,GAAI,MAAO,IAAe,UACxB,GAAM,GAAmC,GACzC,SAAO,UAAe,EACtB,EAAO,OAAY,GACZ,GAAsB,OACxB,OAAI,aAAsB,IACxB,EAEA,GAAsB,GCjQjC,AAiBA,YAA0B,GACxB,GAAI,GAAQ,MAAQ,MAAO,IAAS,SAClC,KAAM,IAAI,OACN,mEACA,0BAAyB,IAOjC,GAAA,IAAA,SAAA,GAA0C,EAAA,EAAA,GAA1C,4DAEA,MAAA,IAF0C,EAAA,cAAc,cAqBxD,GAAA,SAAA,GAA0B,EAAA,EAAA,GAQxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAEP,UAAiB,GAEjB,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,GAAK,GAAQ,MAAQ,EAAK,IAAM,KAAO,IAAO,EAAK,GACxD,EAAK,MAAQ,EAAK,KAAO,EACzB,EAAK,MAAQ,EAAK,KAAO,IAO3B,SAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,GAAyB,EAAA,MAAM,CAAC,IACpC,MAAI,GAAK,OACP,GAAiB,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,EAAA,IAAI,OAEhE,AAAI,EAAK,OACP,GACI,EAAA,IAAI,EAAgB,EAAA,IAAI,EAAA,IAAQ,EAAK,GAAI,GAAS,OAEjD,EAAe,cAI1B,EAAA,UAAA,UAAA,WACE,MAAO,CAAC,GAAM,KAAK,GAAI,GAAM,KAAK,KAI7B,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,CAAC,GAAI,EAAO,GAAiB,GAAI,EAAO,MA3ClD,EAAA,UAAY,OA6CrB,GA/C0B,IAgD1B,EAAA,cAAc,cAAc,gBAET,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,gBAGvC,GACjB,UAAiB,GACV,GAAI,IAAK,CAAC,GAAI,GAAQ,KAAO,EAAK,GAAK,KAAM,GAAI,IAOnD,GAAM,IACyC,CAChD,KAAQ,oBAGuB,GAEnC,MAAO,IAAqB,eAI1B,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,IACK,GACH,EAAQ,EAAA,cAAc,iBAAiB,SAAS,aAChD,EAAe,2BAGU,GAG7B,GAAI,GAAc,KAChB,MAAO,MAET,GAAI,MAAO,IAAe,UACxB,GAAM,GAAY,IAAc,IAC5B,GAA2C,GAC3C,EACE,EAAS,CAAC,UAAS,EAAE,OAAQ,IACnC,MAAO,IAAuB,OACzB,OAAI,aAAsB,IACxB,EAEA,GAAuB,GClJlC,AAmCA,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAKxB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,MACV,GAAK,SAAW,EAAK,YAIzB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,EAAS,GAAoB,GAC7B,GAAI,GAAS,EAAA,KAAK,GAClB,MAAI,MAAK,UAAY,MACnB,GAAS,EAAA,YAAY,EAAQ,EAAG,KAAK,WAEhC,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,SAAU,KAAK,UACnD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA5BF,EAAA,UAAY,OA8BrB,GAhC0B,IAiC1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAO7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,GAIvB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,UAAU,EAAG,KAAK,QAG3B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,YA4BrB,GA9B+B,IA+B/B,EAAA,cAAc,cAAc,IA6B5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAWzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAU/B,GAbO,EAAA,0BAAmD,QAI1D,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,gBAAkB,GACvB,EAAK,iBACD,GAAe,EAAK,kBAAoB,EAAK,2BACjD,EAAK,iBAAmB,GAAe,EAAK,kBAC5C,EAAK,gBAAkB,GAAc,EAAK,iBACtC,EAAK,YAAc,KACrB,EAAK,WAAa,aACT,MAAM,QAAQ,EAAK,YAC5B,EAAK,WAAa,EAAK,mBACd,MAAO,GAAK,YAAe,SACpC,EAAK,WAAa,CAAC,EAAK,gBAExB,MAAM,IAAI,GACN,8DACA,YAAW,EAAK,sBAIxB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAChC,GAAM,GAAoB,EAAW,MAAM,GAC3C,GAAI,KAAK,YAAc,KACrB,OAAgB,GAAA,EAAA,EAAA,KAAK,WAAL,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACV,EAAW,EAAI,GAAK,EAGxB,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAAM,KAAK,iBAEtC,GAAM,GAAiC,GACvC,GAAI,KAAK,YAAc,KACrB,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,EAAK,GAAK,EAAW,GAGzB,KAAK,UAAY,CAAC,GAAI,IAAU,CAC9B,KAAM,EAAW,OACjB,KAAI,KAEN,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,SAAS,GAAoB,GACtB,EAAA,MAAM,EAAQ,KAAK,MAAM,SAGlC,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,iBAAkB,GAAqB,KAAK,kBAC5C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAoB,KAAK,iBAC1C,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAxEF,EAAA,UAAY,QA0ErB,GA5E2B,IA6E3B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAK/B,GARO,EAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGL,EAAK,OAAS,MAAQ,EAAK,QAAU,EAAK,cAC5C,KAAM,IAAI,IACN,4BAA4B,EAAK,MAAK,4CAI5C,SAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAA,IAAI,IAGb,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjCF,EAAA,UAAY,MAmCrB,GArCyB,IAsCzB,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAOnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,cAAgB,EAIvB,AAAI,GAAQ,MACV,GAAO,IAGT,EAAK,MAAQ,EAAK,OAAS,KAAO,EAAK,cAAgB,EAAK,QAG9D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,GAAE,IAAI,GAAK,EAAE,QAAQ,KAAK,OAAQ,aAG3C,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,MAAO,KAAK,OAChD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,kBA6BrB,GA/BqC,IAgCrC,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAHxB,SAAA,aAAe,EAItB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,QAAU,GAAI,MAAoB,MACvC,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,OAG3D,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,GAAM,GAAI,GAAoB,GAC9B,MAAO,MAAK,QAAQ,EAAG,KAAK,OAG9B,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CAAC,KAAM,KAAK,MAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA3BF,EAAA,UAAY,UA6BrB,GA/B6B,IAgC7B,EAAA,cAAc,cAAc,ICxV5B,YAwBI,EAAwB,EAAW,GACrC,GAAI,MAAO,IAAU,SACnB,MAAO,IAAa,EAAO,GAE3B,GAAI,EAAM,SAAW,EACnB,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EAAC,aACxD,eAAc,EAAM,OAAM,eAEhC,OAAS,GAAI,EAAG,EAAI,EAAG,EAAE,GACvB,GAAM,GAAc,EAAM,GAC1B,GAAI,CAAC,GAAU,GACb,KAAM,IAAI,GACN,OAAO,EAAI,4CAA4C,EACvD,yBAAwB,KAAK,UAAU,GAAM,gBAC7C,wBAAuB,IAG/B,MAAO,eAaP,EAAqB,EAAoB,EACzC,EAAgB,GAClB,GADkB,AAAA,IAAA,QAAA,GAAA,GACd,GAAe,KACjB,MAAO,GAET,GAAM,GAAoB,EAAc,GAAa,GAAM,GAAW,GAClE,EACJ,MAAI,KAAY,OACd,EAAe,EAEf,EAAe,EAAc,EAAoB,EAE5C,KAAK,MAAO,GAAe,EAAS,GAAK,eAI9C,EAAiB,EAAoB,EACrC,GACF,GAAI,GAAW,KACb,MAAO,MAGT,GAAI,IAAY,QACd,EAAU,EAAU,EAAa,GAAI,CAAC,EAAa,EAAY,YACtD,IAAY,OACrB,EAAU,EAAU,MAEpB,MAAM,IAAI,GAAW,2BAA2B,EAAO,KAEzD,MAAO,GCpFT,YAwCI,EAAW,GAEb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,IAE3B,gBAWT,EAAW,GACb,MAAO,GAAA,KAAK,WAEV,MADA,IAAgB,GACZ,IAAe,gBACV,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,IAE9B,gBAsBT,EAAW,EAAgB,EAAc,EAAa,EACtD,EAAyB,GADgB,MAAA,KAAA,QAAA,GAAA,GAAa,AAAA,IAAA,QAAA,GAAA,SAC7B,AAAA,IAAA,QAAA,GAAA,GACpB,EAAA,KAAK,WAMV,GALA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAEZ,EAAE,MAAM,SAAW,EACrB,KAAM,IAAI,GACN,+DACG,GAAE,MAAM,OAAM,cAEvB,GAAI,EAAO,MAAM,SAAW,EAC1B,KAAM,IAAI,GACN,iEACG,GAAO,MAAM,OAAM,aAE5B,GAAI,GAAQ,MAAQ,EAAK,MAAM,SAAW,EACxC,KAAM,IAAI,GACN,+DACG,GAAO,MAAM,OAAM,aAM5B,GAHA,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,KAE1B,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,GAAI,GAAY,EAAA,OACZ,EAA0B,EAAoB,EAC9C,IAAY,OAAS,OAAS,QAAS,MAAO,GAClD,MAAI,IAAQ,MACV,GAAI,GAAU,EAAG,IAEZ,gBAoDP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EAAyB,EAC5C,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,IACvD,AAAA,IAAA,QAAA,GAAA,SACA,AAAA,IAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,gEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,iEACA,iBAAgB,EAAE,KAAI,MAE5B,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,MAAU,OAAO,CACnB,EAAG,EACH,OAAQ,EACR,QAAS,EACT,IAAK,IAAY,OAAS,OAAS,QACnC,UAAW,EACX,WAAY,OACZ,KAAI,EACJ,WAAU,IAEZ,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBA8BP,EAAW,EAAgB,EAAc,EACzC,EAAmB,EACnB,GAFyC,MAAA,KAAA,QAAA,GAAA,CAAW,EAAG,EAAG,IAC1D,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WAKV,GAJA,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GACZ,EAAE,OAAS,GAAK,EAAE,OAAS,EAC7B,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,MAEf,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,GACN,oEACG,GAAE,KAAI,MAEf,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,IAAY,SACd,KAAM,IAAI,IACN,iFAGN,SAAI,EAAA,OACA,EACA,EAAmC,EACnC,IAAY,OAAS,OAAS,QAAS,QAAS,GACpD,AAAI,GAAQ,MACV,GAAI,GAAU,EAAG,IAEnB,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAmHX,GAAA,IAAA,SAAA,GAAuC,EAAA,EAAA,GAwBrC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,IAAkB,KAIxB,GAVQ,EAAA,KAAsB,KAEvB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAIzD,EAAS,WAAW,GACpB,EAAK,KAAO,EACZ,GAAoC,EAAK,KAAM,QAC3C,EAAK,OAAS,GAAK,EAAK,OAAS,GAAK,EAAK,OAAS,EACtD,KAAM,IAAI,IACN,qDACI,EAAK,KAAI,6BAqBnB,GAlBA,EAAK,WAAa,GAAe,EAAK,WAAY,EAAM,cACxD,EAAK,QAAU,GACX,EAAK,SAAW,KAAO,EAAI,EAAK,QAAS,EAAM,WACnD,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,WAAa,GAAc,EAAK,YACrC,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAClD,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,aAAe,GAChB,EAAK,cAAgB,KAAO,EAAI,EAAK,aAAc,EACnD,gBACA,EAAK,OAAS,GACb,MAAM,QAAQ,EAAK,eAAiB,EAAK,aAAa,SAAW,EACpE,KAAM,IAAI,GACN,iGAEA,IAAG,KAAK,UAAU,EAAK,gBACtB,GAAI,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aAAe,CAAC,EAAK,aAAc,EAAK,sBACpC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,gEACA,8BAA6B,KAAK,UAAU,EAAK,wBAE9C,EAAK,OAAS,GACvB,GAAI,MAAO,GAAK,cAAiB,SAC/B,EAAK,aACD,CAAC,EAAK,aAAc,EAAK,aAAc,EAAK,sBACvC,EAAK,aAAa,SAAW,EACtC,KAAM,IAAI,GACN,kEACA,8BAA6B,KAAK,UAAU,EAAK,yBAK1C,SAAA,WAAjB,SAA4B,GAI1B,GAFA,GACI,cAAgB,GAAM,2CACtB,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,oEACA,oCACI,KAAK,UAAU,EAAK,YAAW,OAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,gBAAiB,GAAqB,KAAK,iBAC3C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA/GuC,IAqHvC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAcjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAA0B,KAZ9B,SAAA,OAAwB,KAahC,EAAK,WAAW,GAChB,EAAK,QAAU,EAAK,QACpB,GAAoC,EAAK,QAAS,WAClD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,kBAAoB,GAAe,EAAK,qBAG/C,SAAA,UAAA,MAAA,SAAM,SACJ,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAE1B,GAAM,GAAW,EAAW,GAEtB,EAAc,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,UAE3D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,KAAM,KAAK,gBACnC,KAAK,gBAAiB,GAAM,KAAK,iBAGvC,KAAK,UAAY,CAAC,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,KACtE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GACE,EAAY,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACjD,EAAsB,GACU,EAAK,WAAW,gBAEtD,GAAI,GAAuB,MAAQ,EAAK,OAAS,EAC/C,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,aACtB,QAEJ,GAAI,EAAK,OAAS,EAChB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAQ,GACpD,EAAK,QAAS,EAAK,WAAY,EAAK,aAAa,YAC5C,EAAK,OAAS,EAEvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,sBACjB,EAAK,OAAS,EACvB,EAAU,GACN,EAAQ,EAAK,OAAO,OAAQ,EAAW,EAAK,QAAS,EAAK,QAC1D,EAAK,WAAY,EAAK,kBAE1B,MAAM,IAAI,IACN,yDAGN,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAIpC,MAAO,MAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAKhC,OAJM,GAAqB,GACrB,EAAS,KAAK,aAAe,eAC/B,EAAW,MAAM,EAAG,EAAW,OAAS,GACxC,EAAW,MAAM,GACZ,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,GAClC,GAAM,GAAS,GACX,EAAM,GAAI,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,GACzD,MAAO,MAAK,cAAiB,SAAW,KAAK,aACL,KAAK,aAAa,IAC9D,EAAS,KAAK,GAGhB,GAAI,GAAc,CAAC,EAAW,IAC9B,MAAI,MAAK,aAAe,eACtB,GAAc,EAAY,OAAO,GACjC,EAAY,KAAK,KAAK,UAEtB,GAAY,KAAK,KAAK,SACtB,EAAc,EAAY,OAAO,IAE5B,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,kBAAmB,GAAqB,KAAK,mBAC7C,iBAAkB,GAAoB,KAAK,mBAEvC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,CAAE,YAAa,KAAS,MAAO,GAAK,SAAY,UAChD,EAAK,QAAU,EACjB,KAAM,IAAI,GACN,kEACA,YAAW,KAAK,UAAU,EAAK,YAGzC,GA1ImC,IA4InC,GAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAK,MAAO,GAAK,YAAe,UAC5B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,gCAA+B,KAAK,UAAU,EAAK,YAAW,OAnB/D,EAAA,UAAY,SAsBrB,GAxB4B,IAyB5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,KAGpB,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UACzB,CAAE,OAAM,QAAQ,EAAK,aAClB,GAAK,WAAW,SAAW,GAAK,EAAK,WAAW,SAAW,IAChE,KAAM,IAAI,GACN,mDACA,4CACI,KAAK,UAAU,EAAK,YAAW,OApBtC,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAGX,GAFA,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KAEnC,EAAK,UAAY,QAAU,EAAK,UAAY,QAC9C,KAAM,IAAI,GACN,gEACA,2CAA0C,EAAK,mBAIvD,SAAA,UAAA,MAAA,SAAM,SAGJ,GAFA,EAAa,GAAmB,GAE5B,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,mDACA,KAAK,UAAU,IAGrB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,wEAGN,GAAM,GAAW,EAAW,GACtB,EAAc,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,IAE1D,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,UAAW,KAAK,kBACvC,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,GAAM,KAAK,iBAIvC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MAC3D,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAM,MAAM,SAAW,EACzB,KAAM,IAAI,GACN,iEACA,8BAA6B,EAAM,MAAM,SAG/C,GAAM,GAAa,EAAM,MACnB,EAAY,EAAW,GAEzB,EACA,EACJ,AAAI,EAAK,aAAe,gBACtB,GAAQ,EACR,EAAQ,GAER,GAAQ,EACR,EAAQ,GAGV,GAAM,GAAS,EAAW,GACpB,EAAQ,EAAW,GACnB,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,WAAW,GAC1B,EAAU,EAAK,QAAQ,GACvB,EAAU,EAAK,QAAQ,GAGvB,EAAY,GAAa,EAAQ,EAAS,EAAS,EAAK,SACxD,EAAW,GAAa,EAAO,EAAS,EAAS,EAAK,SAMtD,EACF,CAAC,EAAW,EAAW,EAAU,EAAK,SAE1C,AAAI,EAAK,aAAe,gBACtB,GAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,KAEzC,GAAI,GAAU,EAAA,gBACV,EAAmB,EAAK,OAAO,OAAoB,EACnD,EAAK,QAA6B,EAAK,SAC3C,MAAI,GAAK,aAAe,gBACtB,GAAU,EAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,KAG7C,AAAI,EAAK,MAAQ,MACf,GACI,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEhD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAE3B,EACA,EACA,EACJ,AAAI,KAAK,aAAe,gBACtB,GAAc,EACd,EAAa,EACb,EAAY,GAEZ,GAAc,EACd,EAAa,EACb,EAAY,GAGd,GAAM,GAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,WAAW,GAC1B,EAAU,KAAK,QAAQ,GACvB,EAAU,KAAK,QAAQ,GAE7B,SAAY,GAAe,KAAK,QAChC,EAAY,GACR,GAAa,EAAY,GAAa,EAAS,EAAS,KAAK,SACjE,EAAY,GACR,GAAa,EAAY,GAAY,EAAS,EAAS,KAAK,SACzD,GAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,aACP,GA9IF,EAAA,UAAY,kBAgJrB,GAlJqC,IAmJrC,EAAA,cAAc,cAAc,IA0C5B,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAqBjC,WAAY,EAAc,GAA1B,GAAA,GACE,EAAA,KAAA,KAAM,EAAM,IAAO,KAEnB,GAXO,EAAA,8BACL,gBACK,EAAA,8BACL,gBAEM,EAAA,gBAAiC,KACjC,EAAA,gBAAiC,KAKrC,EAAO,SAAW,KACpB,KAAM,IAAI,GACN,uFAGN,GAAI,EAAO,mBAAqB,MAAQ,EAAO,mBAAqB,MAChE,EAAO,kBAAoB,KAC7B,KAAM,IAAI,GACN,sPAKN,GAAI,EAAO,SAAW,MAAQ,EAAO,UAAY,QAC7C,EAAO,UAAY,QACrB,KAAM,IAAI,GACN,gBAAgB,EAAK,KAAI,kCACzB,qCAAoC,KAAK,UAAU,EAAO,WAGhE,SAAK,gBACD,EAAO,iBAAmB,KAAO,EAAI,EAAO,gBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,qBAChD,EAAK,qBAAuB,GACxB,EAAO,sBAAwB,EAAK,+BACxC,EAAK,qBAAuB,GAAe,EAAO,sBAClD,EAAK,oBAAsB,GAAc,EAAO,uBAGlD,SAAA,UAAA,MAAA,SAAM,SAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,KAAK,KAAO,EAClC,KAAM,IAAI,GACN,0BAA0B,KAAK,KAAI,sBAChC,MAAK,KAAO,EAAC,gCAChB,IAAG,KAAK,UAAU,KAExB,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAClE,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,0DACA,cAAa,KAAK,UAAU,EAAW,MAO7C,OAJM,GAAW,EAAW,GACtB,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,kBACrC,EAAuB,GACpB,EAAI,EAAG,EAAI,KAAK,KAAM,EAAE,EAC/B,EAAqB,KAAK,GAE5B,EAAqB,KAAK,EAAW,KAAK,gBAAiB,KAAK,SAEhE,GAAM,GAAY,GAClB,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,UAC1C,KAAK,qBAAsB,KAAK,qBAAsB,EACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,SAAU,UAAW,KAAK,gBACxC,KAAK,gBAAiB,EAAW,KAAK,gBAE1C,KAAK,KAAO,KAGd,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,KAAK,KAAO,EAAG,KAAI,GAAA,GAAG,EAAC,GAAc,EAAQ,MACvE,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAE7B,GAAI,GACJ,GAAI,EAAK,OAAS,EAChB,KAAM,IAAI,IACN,oDACC,MAAI,GAAK,OAAS,GACvB,CAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAG3C,EAAS,EAAA,gBACL,EAAoB,EAAK,gBAAgB,OACzC,EAAK,gBAAgB,OACrB,EAAK,QAA6B,EAAK,QACvC,EAAK,aAAkC,SAG7C,AAAI,EAAK,SACP,GAAS,GAAU,EAAQ,EAAK,KAAK,OAAQ,EAAK,aAEpD,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAGjC,AAAI,EAAK,aAAe,iBACtB,GAAS,EAAA,UAAc,EAAQ,CAAC,EAAG,EAAG,EAAG,KAEpC,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,kBACd,MAAO,GAAO,kBACd,MAAO,GAAO,iBACd,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,qBAC7B,EAAO,oBACH,GAAoB,KAAK,qBACtB,GAzJF,EAAA,UAAY,gBA2JrB,GA7JmC,IA+JnC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,SACV,GAAA,KAAA,KAAM,EAAG,IAAK,KAFT,SAAA,UAAY,kBAIrB,GANqC,IAOrC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA4B,EAAA,EAAA,GAG1B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAK,KACd,SAAO,WAAW,GAClB,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,aAAO,GAAO,KACd,MAAO,GAAO,WACP,GAGQ,EAAA,WAAjB,SAA4B,GAE1B,GAAI,MAAO,GAAK,YAAe,UAC3B,CAAC,GACG,EAAK,WAAY,SAAU,EAAG,GACpC,KAAM,IAAI,GACN,kEACA,2BAA0B,KAAK,UAAU,EAAK,YAAW,OArB1D,EAAA,UAAY,SAwBrB,GA1B4B,IA2B5B,EAAA,cAAc,cAAc,IAgC5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAM9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,MAAI,OAAO,GAAK,UAAa,SAC3B,EAAK,SACD,CAAC,CAAC,EAAK,SAAU,EAAK,UAAW,CAAC,EAAK,SAAU,EAAK,WACrD,AAAI,MAAO,GAAK,SAAS,IAAO,SACrC,EAAK,SAAW,CACd,CAAC,EAAK,SAAS,GAAI,EAAK,SAAS,IACjC,CAAC,EAAK,SAAS,GAAc,EAAK,SAAS,KAG7C,EAAK,SAAW,EAAK,SAEvB,EAAK,WACD,EAAK,aAAe,OAAY,eAAiB,EAAK,WAC1D,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAI,MAAK,aAAe,gBACf,CACL,EAAW,GAAI,EAAW,GAC1B,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,IAGlD,CACL,EAAW,GACX,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GACvD,EAAW,GAAK,KAAK,SAAS,GAAG,GAAK,KAAK,SAAS,GAAG,GAAI,EAAW,KAK5E,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAFA,EAAS,GAAoB,GAEzB,EAAK,aAAe,gBACtB,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,QAEjE,GAAM,GAAU,GACZ,EAAQ,EAAK,SAAS,GAAG,GACzB,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,GACjE,MAAO,IACH,EAAS,EAAK,SAAS,GAAG,GAC1B,EAAO,MAAM,GAAK,EAAK,SAAS,GAAG,GAAK,EAAK,SAAS,GAAG,GAAI,OAKvE,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,SAAU,KAAK,SAAU,WAAY,KAAK,YACpD,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAhEF,EAAA,UAAY,aAkErB,GApEgC,IAqEhC,EAAA,cAAc,cAAc,IAwB5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAOhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KALM,SAAA,aAAe,CAAC,EAAG,GAMpC,EAAK,UAAY,CAAC,CAAC,KAAM,IACzB,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,aAGtD,SAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,KAAK,aAAe,iBACtB,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,QAE9C,GAAM,GACF,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvD,EAAQ,EAAW,IAAM,KAAO,KAAO,KAAK,KAAK,GAAK,EAAW,GACvE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAO,EAAW,MAIrD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,GAAoB,GAC1B,EAAa,EAAM,MAEzB,GAAI,EAAK,aAAe,iBACtB,EAAQ,EAAA,UAAc,EAAO,CAAC,EAAG,EAAG,EAAG,IACvC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GAClC,EAAU,EAAM,sBAAsB,CAAC,EAAQ,IACrD,MAAO,GAAA,UAAc,EAAS,CAAC,EAAG,EAAG,EAAG,SAExC,GAAM,GAAS,EAAK,KAAK,GAAK,EAAW,GACnC,EAAQ,EAAK,KAAK,GAAK,EAAW,GACxC,MAAO,GAAM,sBAAsB,CAAC,EAAQ,QAKlD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,KAAM,KAAK,KAAM,WAAY,KAAK,YAC5C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlDF,EAAA,UAAY,eAoDrB,GAtDkC,IAuDlC,EAAA,cAAc,cAAc,ICrvC5B,YA6CI,EAAW,EAAyB,EACpC,EAAmB,EACnB,GAFoC,MAAA,KAAA,QAAA,GAAA,CAA6B,EAAG,IACpE,AAAA,IAAA,QAAA,GAAA,SAEK,EAAA,KAAK,WACV,AAAI,GAAc,MAChB,GAAa,MAEf,GAAgB,GAChB,GAAI,GAAI,GAAsB,EAAG,GACjC,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,mEACG,GAAE,KAAI,OAEf,GAAI,EAAgB,OAAS,EAC3B,KAAM,IAAI,GACN,yDACG,GAAgB,KAAI,OAE7B,SAAI,EAAA,gBACA,EAAe,EAA6B,EAC5C,IAAY,OAAS,OAAS,QAAS,OAAQ,GACnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,IAsCX,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAUnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,EAAG,IAAsB,KAHzB,SAAA,gBAAiC,KAIvC,EAAK,gBACD,EAAK,iBAAmB,KAAO,EAAI,EAAK,gBAC5C,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,4BACtC,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,qBAAuB,GAAe,EAAK,wBAGlD,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,iDACA,0BAAyB,KAAK,UAAU,GAAW,MAEzD,GAAM,GAAc,KAAK,aAAe,gBAAkB,EAAI,EAC9D,GAAI,EAAW,IAAgB,MAAQ,EAAW,GAAe,EAC/D,KAAM,IAAI,GACN,iEACA,4BAA2B,EAAW,GAAY,OAExD,GAAM,GAAW,EAAW,GACtB,EAA8B,CAClC,KAAK,WAAW,GAAI,KAAK,WAAW,GAAI,EAAU,KAAK,iBAGzD,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,EAAW,KAAK,iBAAkB,KAAM,KAAK,gBACtD,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,GAAoB,GAC7B,GAAI,GAAU,GACV,EAAQ,EAAK,gBAAgB,OAAQ,EAAK,QAC1C,EAAK,QAAS,EAAK,WAAY,MAEnC,MAAI,GAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,OAAQ,EAAK,aAEtD,AAAI,EAAK,YAAc,MACrB,GAAU,EAAK,WAAW,MAAM,IAE3B,KAIX,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EACF,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAC/D,EAAa,KAAK,aAAe,gBACnC,EAAW,GAAK,KAAK,gBACrB,EAAW,GAAK,KAAK,gBACnB,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACnD,EAAU,GACZ,EAAM,KAAK,WAAW,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,MAAI,MAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAY,EAAS,GAGrC,CAAC,EAAW,GAAI,EAAS,EAAS,IAI7C,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,EAAA,UAAM,UAAS,KAAA,MAC9B,SAAO,gBAAqB,KAAK,gBACjC,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,qBACH,GAAqB,KAAK,sBAC9B,EAAO,oBACH,GAAoB,KAAK,sBACtB,GAjGF,EAAA,UAAY,kBAmGrB,GArGqC,IAsGrC,EAAA,cAAc,cAAc,IClN5B,YA0DI,EACA,EACA,EACA,GAKF,GAAI,MAAM,QAAQ,IAChB,GAAI,GAAgB,MAAQ,GAAa,KACvC,KAAM,IAAI,GACN,iFAGN,AAAI,GAAgB,MAClB,GAAY,EAAO,MAAM,EAAO,OAAS,EAAc,EAAO,QAC9D,EAAS,EAAO,MAAM,EAAG,EAAO,OAAS,IAE3C,AAAI,EAAO,OAAS,GAClB,GAAe,EAAO,MAAM,EAAG,EAAO,SAExC,EAAS,EAAO,GAGlB,WAAsB,GAEpB,MAAI,IAAK,MAAQ,MAAM,QAAQ,GACtB,EAEA,CAAC,GAIZ,SAAe,EAAa,GAC5B,EAAY,EAAa,GAElB,CAAC,OAAM,EAAE,aAAY,EAAE,UAAS,eA+CrC,EAA+B,EAAgB,EAC/C,EAAqB,EAAe,EAAsB,EAC1D,GADA,MAAA,KAAA,QAAA,GAAA,IAA0D,AAAA,IAAA,QAAA,GAAA,IAC1D,AAAA,IAAA,QAAA,GAAA,IACK,EAAA,KAAS,WACd,GAAM,GAAO,EAAO,MAAM,OAC1B,GAAI,EAAO,EACT,KAAM,IAAI,GAAW,uCAAuC,EAAI,MAKlE,GAAM,GAAO,CAAC,EAAG,GAAG,OAAO,GAAiB,EAAG,IAG/C,GAFA,EAAS,EAAA,UAAc,EAAQ,GAE3B,GAAa,KACf,KAAM,IAAI,IACN,kFAKN,AAAI,GACF,QAAQ,KACJ,qGAIN,AAAI,GAAQ,MACV,GAAO,EAAK,OAAO,QAAQ,OAAO,WAClC,AAAI,EAAK,OAAS,EAAO,GACvB,GAAO,EAAA,WAAe,EAAM,KAE9B,EAAO,EAAA,UAAc,EAAM,IAG7B,AAAI,GACF,GAAS,EAAA,QAAY,EAAQ,GAC7B,AAAI,GAAQ,MACV,GAAO,EAAA,QAAY,EAAM,KAc7B,GAAM,GAA2B,GAC7B,EACA,EAAS,EACP,EAAY,EAAO,MAAM,GACzB,EAAgB,EAAA,QAAY,GAC9B,EACJ,AAAI,GAAQ,MACV,GAAe,EAAA,QAAY,IAG7B,mBAAS,GACP,GAAM,GAAe,EAAc,GAC7B,EAAc,EAAA,KAAS,WAAM,MAAA,GAAa,EAAc,KAE9D,GAAI,GAAQ,KACV,EAAa,EAAY,GACzB,EAAS,EAAY,QAErB,GAAM,GAAgB,EAAA,KAAS,WAC7B,GAAM,GAAW,EAAa,GACxB,EAAc,EAAA,SAAa,GAAU,IAAI,GAEzC,EACF,EAAY,GAAG,IAAI,GAAU,IAAI,EAAO,GAAG,IAAI,IAC7C,EAAY,EAAO,IAAI,SAAC,EAAO,GACnC,MAAO,GAAY,GAAG,GAAG,IAAI,GAAU,IAAI,EAAM,IAAI,MAEvD,MAAO,CAAC,OAAM,EAAE,UAAS,KAE3B,EAAa,EAAc,OAC3B,EAAS,EAAc,UAGzB,AAAI,GACF,EAAe,KAAK,IAxBf,EAAI,EAAG,EAAI,EAAW,EAAE,IAAxB,GA2BT,GAAI,GACJ,GAAI,GACF,GAAM,GAAO,EACb,EAAU,EAAA,MAAU,EAAgB,GAEtC,MAAO,CAAC,EAAY,EAAS,wBAyGR,EAAA,EAAA,GAqBvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACP,EACJ,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,GACN,wDAMN,GALO,AAAI,MAAM,QAAQ,EAAK,MAC5B,EAAO,GAAI,IAAgB,CAAC,MAAO,EAAK,OAExC,EAAO,EAAK,KAEV,EAAK,WAAa,KACpB,KAAM,IAAI,GACN,qGAGN,SAAK,KAAO,EACZ,EAAK,gBACD,EAAK,iBAAmB,KAAO,GAAQ,EAAK,gBAChD,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,YAAc,EAAK,aAAe,KAAO,GAAQ,EAAK,YAC3D,EAAK,UAAY,EAAK,UAAY,KAAO,GAAQ,EAAK,SACtD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAQ,EAAK,OAEjD,EAAK,gBAAkB,GACvB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,KACvC,EAAK,UAAY,KACjB,EAAK,QAAU,KAEf,EAAK,aAAe,KAIpB,EAAK,WAAa,KAKpB,SAAA,UAAA,UAAA,WACE,GAAI,KAAK,SAAW,MAClB,GAAM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EACtE,MAAO,IAAiB,EAAG,GAAW,IAAI,SAAA,GAAK,MAAA,YAE/C,OAAO,MAAK,SAMhB,EAAA,UAAA,UAAA,SAAU,GACR,KAAK,QAAU,GAGjB,EAAA,UAAA,mBAAA,SAAmB,GACjB,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAGb,GAAI,GAAY,KAAK,KAAK,UAC1B,AAAK,MAAM,QAAQ,IACjB,GAAY,CAAC,IAEf,GAAM,GAAY,EAAU,GACxB,EAOJ,GANA,AAAI,KAAK,gBACP,EAAc,CAAC,EAAW,GAAI,EAAW,GAAI,GAE7C,EAAc,CAAC,EAAW,GAAI,GAG5B,KAAK,aAEP,OADM,GAAsB,GACV,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAW,KAAK,CAAC,EAAW,GAAI,IAElC,MAAO,CAAC,GAAa,OAAO,OAE5B,OAAO,IAIX,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KAEE,MAAO,GAAA,KAAS,WACd,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAM,GAAa,EAAK,gBAAkB,EAAO,KAEjD,GAAI,EAAK,aACP,GAAM,GAAY,EAAK,OAAO,IAAI,SAAA,GAAK,MAAA,QACvC,MAAO,CAAC,GAAY,OAAO,OAE3B,OAAO,MAWb,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,GAAI,KAAK,SAAW,MAIlB,OAHM,GACF,MAAM,QAAQ,KAAK,KAAK,WAAa,KAAK,KAAK,UAAU,OAAS,EAChE,EAAmB,GAChB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,EAAO,KAAK,MAEd,MAAO,OAEP,OAAO,MAAK,aAIhB,SAAW,GACT,KAAK,QAAU,mCAGV,EAAA,UAAA,MAAP,SAAa,GAIX,GAAI,KAAK,cAAgB,KACvB,KAAM,IAAI,IACN,oDAGN,AAAI,GAAgB,IAClB,GAAc,EAAuB,IAEvC,EAAa,EAEb,GAAM,GAAoB,KAAK,SAAW,EAAW,GAAK,KACpD,EAAW,EAAW,MAAM,GAClC,KAAK,UAAU,GAAK,GAAI,IAAU,CAAC,MAAK,CAAG,EAAW,MAAI,OAAK,KAI/D,GAAM,GAAiB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAK7D,KAAK,KAAK,MAAM,GAIlB,GAAI,GAOJ,GANA,AAAI,MAAM,QAAQ,KAAK,KAAK,WAC1B,EAAY,KAAK,KAAK,UAEtB,EAAY,CAAC,KAAK,KAAK,WAGrB,KAAK,WAAa,MACpB,GAAI,CAAC,EAAA,KAAK,YACF,KAAK,UAAU,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,EAAK,MAAM,OAAS,KAC1D,GACN,KAAM,IAAI,GACN,0DACA,uCAAsC,KAAK,UAAS,MACpD,8BAA6B,KAAK,KAAK,gBAG7C,MAAK,UACD,EAAU,IAAI,SAAA,GAAO,MAAA,IAAI,IAAU,CAAC,MAAO,CAAC,KAAM,OAExD,AAAI,KAAK,UACP,KAAK,eAqBT,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAK,WACH,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAEN,GAAM,GAAY,EAAK,UAAU,GAAG,MAAM,GAC1C,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUAQN,GAAI,EAAK,SAAW,KAClB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAU,CAAC,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,qBAEzC,GAAU,KAEnB,EAAA,QAAY,EAAK,SAEjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QACD,EAAK,KAAK,UAAU,IAAI,SAAA,GAAO,MAAA,GAAA,MAAU,CAAC,EAAW,MAEzD,EAAK,QAAQ,GAAK,EAAA,MAAU,CAAC,EAAW,EAAK,KAAK,iBAMpD,GAHA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAER,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,IAAa,GAKf,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GACf,EAAM,MAAM,QAAQ,EAAK,KAAK,WAChC,EAAK,KAAK,UAAU,GACpB,EAAK,KAAK,UACR,EAAgB,CAAC,EAAW,GAClC,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAEhB,EAAK,QAAQ,GAAS,GAG1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAI5D,EAAA,UAAA,MAAA,SACI,EACA,GAEF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAGX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAC1D,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAMzB,GAAI,GAAiD,GACjD,EAA+B,GACnC,GAAI,GAAgB,MAClB,EAAO,aAAkB,EACzB,EAAmB,EAAiB,OAAO,GAC3C,KAAK,UAAY,GACjB,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,UAAU,KAAK,GAAI,IAAU,CAAC,MAAO,EAAM,SAKlD,EAAkB,EAAgB,OAAO,KAAK,WAEhD,AAAI,GAAa,MACf,GAAO,UAAe,EACtB,EAAmB,EAAiB,OAAO,GAE3C,KAAK,aAAe,EAAU,QAGhC,GAAM,GAAW,EAAiB,YAAc,IAChD,GAAI,GAEF,GAAM,GACF,CAAC,GAAQ,OAAO,GACd,EAAgB,KAAK,UAAU,OAAO,GAEtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GAAS,EAAA,UAAM,MAAK,KAAA,KAAC,EAAW,GACtC,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAK/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KAIE,MAAO,GAAA,KAAK,WACV,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC5C,EACA,GAAU,KAAO,KAAO,EAAO,aAEnC,EAAS,GAAoB,GAC7B,AAAI,GAAgB,MAClB,CAAI,EAAK,SACP,EAAe,EAAK,QAEpB,EAAe,EAAK,gBAAgB,IAIxC,GAAM,GACF,MAAM,QAAQ,EAAK,KAAK,WAAa,EAAK,KAAK,UAAU,OAAS,EACtE,GAAI,EAAa,SAAW,EAC1B,KAAM,IAAI,GACN,iBAAiB,EAAS,4BACvB,GAAa,OAAM,uBAE5B,AAAI,EAAK,QACP,QAAQ,KACJ,oEAGN,GAAM,GAAyB,CAAC,SAAQ,GAGlC,EAAO,SAAC,EAAgB,GAG5B,GAAM,GACF,EAAK,KAAK,KAAK,CAAC,GAAQ,OAAO,GAAS,GAE5C,MAAO,CAAC,EAAQ,GAAI,EAAQ,MAAM,KAK9B,EACF,GAAI,EAAM,EAAQ,EAAc,EAAK,YAAa,EAAM,KACpD,EAAK,OAAQ,EAAK,iBACpB,EAAa,EAAW,GACxB,EAAU,EAAW,GACrB,EAAS,EAAW,GAE1B,AAAI,EAAK,UACP,EAAK,YAAY,EAAQ,GAG3B,GAAM,GAAS,EAAK,gBAAkB,EAAU,EAIhD,MAAI,GAAK,YACA,CAAC,GAAQ,OAAO,GAEhB,KAKb,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAGV,GAAI,GAAe,EAAA,MAAU,EAAO,OAKpC,MAHA,GAAe,EAAA,IAAQ,EAAc,CAAC,EAAG,IACzC,EAAe,GAAa,GAExB,MAAM,QAAQ,EAAK,KAAK,WACnB,EAAK,KAAK,UAAU,IACvB,SAAA,GAAO,MAAA,GAAM,EAAI,GAAO,EAAc,CAAC,EAAG,IAAQ,IAE/C,EAAK,KAAK,UAAY,EACzB,CAAC,GAAO,EAAc,CAAC,EAAG,EAAK,KAAK,aACpC,CAAC,MAKX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAK,MAAK,UAIH,KAAK,KAAK,iBAHR,oCAMX,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,MAAK,MAAK,UAGH,KAAK,KAAK,oBAFR,KAAK,KAAK,yCAKrB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,MAAQ,MACf,KAAK,KAAK,6BAA6B,IAI3C,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,gBAAiB,KAAK,gBACtB,YAAa,KAAK,YAClB,YAAa,KAAK,YAClB,SAAU,KAAK,SACf,OAAQ,KAAK,QAGf,AAAI,KAAK,cAAgB,MACvB,GAAO,aAAkB,KAAK,cAGhC,GAAM,GAAa,KAAK,KAAK,YAE7B,MAAI,MAAK,iBAAmB,EAAI,WAC9B,GAAO,KAAU,CACf,UAAa,KAAK,KAAK,eACvB,OAAU,IAKd,GAAA,GAAW,EAAe,EAAe,IAIpC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAa,EAAO,KACpB,EAAO,GAAY,EAAY,GACrC,MAAO,IAAI,GAAI,OAAO,OAAO,EAAQ,CAAC,KAAI,MArfrC,EAAA,UAAY,MAufrB,GAzfyB,IA0fzB,EAAA,cAAc,cAAc,IAU5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAAtC,4DAQA,MAAA,IARsC,IAwFtC,GAAA,SAAA,GAAmC,EAAA,EAAA,GAiCjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KANJ,SAAA,mBAAqB,OACrB,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAIzD,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,YAAc,KAAO,EAAK,mBAAqB,EAAK,YAC7D,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAG9B,SAAA,UAAA,MAAA,SAAM,GACJ,EAAa,GAAmB,GAEhC,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAW,EAAW,OAAS,GAAI,KAAK,OAAQ,KAC3D,KAAK,kBAAmB,KAAK,kBAAmB,GAChD,KAAK,kBACT,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,OAAQ,KAC9C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAEd,KAAK,MAAQ,IASf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8CAA8C,EAAO,OAAM,KAEjE,GAAI,GAAa,EAAO,GACxB,EAAS,EAAO,GAChB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,KAG/B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,KAGxC,GAAI,GACE,EAAiB,EAAK,YACtB,EAAoB,EAAK,qBAC/B,AAAI,GAAU,KACZ,EAAI,GAAM,EAAA,IAAQ,EAAQ,GAAS,EAAK,OAAO,QAE/C,EAAI,GAAM,EAAQ,EAAK,OAAO,QAEhC,AAAI,EAAK,MAAQ,MACf,GAAI,GAAU,EAAG,EAAK,KAAK,SAE7B,AAAI,GAAa,MACf,GAAa,EAAA,IAAQ,EAAY,IAEnC,GAAI,GAAS,EAAA,IAAQ,EAAG,GAAM,EAAY,EAAK,gBAAgB,SAC/D,MAAI,GAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,IAI1B,CAAC,EAAQ,MAIpB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,kBAGzB,MAAA,IAAA,GAAW,EAAe,IArKrB,EAAA,UAAY,gBAuKrB,GAzKmC,IA0KnC,EAAA,cAAc,cAAc,IA2F5B,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAG7B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAK,KAAO,GAAI,IAAc,GAC9B,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IA7BV,EAAA,UAAY,YA+BrB,GAjC+B,IAkC/B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAqC3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAbO,EAAA,mBAAqB,OACrB,EAAA,6BAAqD,cAErD,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAChC,EAAA,yBAAkD,QAQrD,EAAK,WACP,KAAM,IAAI,GACN,+DAEN,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAEhD,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,EAAK,MACtB,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,AAAI,KAAK,QACP,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,KAAK,gBACrC,KAAK,gBAAiB,GAAM,KAAK,gBAErC,KAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,uDACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACzD,EAAW,EAAO,GACtB,EAAS,EAAO,GAKhB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EAAY,EAAK,qBACnB,EACA,EACA,EAEJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAU,GAAM,EAAQ,EAAK,OAAO,QACxC,AAAI,EAAK,SACP,GAAU,GAAU,EAAS,EAAK,KAAK,SAEzC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAGzC,GAAM,GAAuB,EAAK,gBAAgB,OAC5C,EAAA,EAAA,MAAA,EAAA,CAAA,EAAA,EAAA,MAAA,EAAA,OAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAGN,EAAc,GAAM,EAAU,GAE9B,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GACT,EAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,EAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAE/C,GAAM,GAAa,GAAM,EAAA,IAAQ,EAAG,GAAW,GAC/C,EAAK,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,IAEvC,GAAM,GACF,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAA,IAAQ,EAAG,EAAA,IAAQ,IAAK,IAElE,MAAO,CAAC,EAAG,MAIf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,eACrB,WAAY,IAGd,MAAA,IAAA,GAAW,EAAe,IAvMrB,EAAA,UAAY,UAyMrB,GA3M6B,IA4M7B,EAAA,cAAc,cAAc,IA8B5B,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAQ,GACxB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,MAuCrB,GAzCyB,IA0CzB,EAAA,cAAc,cAAc,IAuC5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAsC5B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAZJ,SAAA,mBAAqB,OACrB,EAAA,6BAA+B,cAC/B,EAAA,2BAA6B,eAC7B,EAAA,8BAAgC,aAEhC,EAAA,yBAA2B,QASlC,EAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GACd,EAAK,aAAe,OAAY,EAAK,mBACL,EAAK,YACzC,EAAK,oBAAsB,GACvB,EAAK,sBAAwB,OACzB,EAAK,6BACL,EAAK,qBACb,EAAK,QAAU,EAAK,SAAW,KAAO,GAAO,EAAK,QAElD,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,qBAAuB,GACxB,EAAK,sBAAwB,EAAK,+BAEtC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,eAAiB,EAAK,eAE3B,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,qBAAuB,GAAe,EAAK,sBAChD,EAAK,gBAAkB,GAAe,EAAK,iBAE3C,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,oBAAsB,GAAc,EAAK,qBAC9C,EAAK,eAAiB,GAAc,EAAK,gBAEzC,EAAK,QAAU,GACX,CAAC,EAAG,GAAe,CAAC,EAAG,EAAK,SAAW,KAAO,EAAI,EAAK,YAC3D,EAAK,iBAAmB,GAAe,CACrC,EACA,GACI,CAAC,EAAG,EAAK,kBAAoB,KAAO,EAAI,EAAK,qBAEnD,EAAK,eAAiB,EAAK,eAC3B,EAAK,UAAY,CAAC,EAAK,MAAO,EAAK,OACnC,EAAK,YAAc,KACnB,EAAK,qBAAuB,OAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAW,EAAW,EAAW,OAAS,GAChD,KAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAU,KAAK,MAAQ,GAAI,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,CAAC,KAAK,MAAO,KAAK,MAAQ,GAAI,KAClD,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBACT,GAAI,GACJ,GAAI,KAAK,SACP,GAAI,KAAK,gBACP,GAAM,GAAmB,KAAK,gBACxB,EAAgB,KAAK,MAC3B,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAElB,GAAM,GAAK,EAAiB,MAAM,CAAC,IAC7B,EAAM,GAAI,MAAQ,MAAM,CAAC,IACzB,EAAS,EAAiB,MAAM,CAAC,EAAgB,IACvD,MAAO,IACH,GAAuB,EAAI,GAAK,IAExC,GAZgD,IAEvC,EAAA,UAAY,oBAYrB,GAAkB,KAAK,gBAEzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,MAAQ,GAAI,KAAM,EAAiB,KAAK,gBACtD,GAAM,KAAK,oBAEf,MAAK,KAAO,KAId,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SAE7D,GADA,EAAS,EACL,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,wDACG,GAAO,OAAM,MAEtB,GAAI,GAAW,EAAO,GAChB,EAAW,EAAO,GACxB,EAAS,EAAO,GAChB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAG9B,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAGvC,GAAM,GAAS,EAAK,YACd,EACF,EAAK,qBAKL,EACA,EACA,EACA,EACJ,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GACrC,GAAS,EAAA,IAAQ,EAAQ,EAAO,KAElC,GAAI,GAAI,GAAM,EAAQ,EAAK,OAAO,QAClC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACvD,GAAW,EAAA,IAAQ,EAAU,EAAU,KAEzC,EAAI,EAAA,IAAQ,EAAG,GAAM,EAAU,EAAK,gBAAgB,SACpD,AAAI,EAAK,SACP,GAAI,GAAU,EAAG,EAAK,KAAK,SAGvB,GAAA,GAAA,EAAA,MAAA,EAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAAI,EAAA,EAAA,GAEnB,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAK,oBAAoB,MAAM,GACnC,EAAI,EAAA,IAAQ,EAAA,IAAQ,EAAG,GAAW,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,KACnE,EAAI,EAAK,oBAAoB,MAAM,GAEnC,GAAM,GAAI,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,IAE3C,MAAO,CAAC,EAAG,EAAG,MAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,oBAAqB,GAAoB,KAAK,qBAC9C,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,eAAgB,KAAK,eACrB,kBAAmB,GAAqB,KAAK,mBAC7C,qBAAsB,GAAqB,KAAK,sBAChD,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,oBAAqB,GAAoB,KAAK,qBAC9C,eAAgB,GAAoB,KAAK,gBACzC,QAAS,KAAK,QACd,iBAAkB,KAAK,iBACvB,eAAgB,KAAK,gBAGvB,MAAA,IAAA,GAAW,EAAe,IAnNrB,EAAA,UAAY,WAqNrB,GAvN8B,IAwN9B,EAAA,cAAc,cAAc,IAqC5B,GAAA,IAAA,SAAA,GAA0B,EAAA,EAAA,GAGxB,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,GAAK,iBAAmB,GAC1B,QAAQ,KACJ,kHAGN,EAAK,KAAO,GAAI,IAAS,GACzB,EAAA,EAAA,KAAA,KAAM,IAAqB,OAI7B,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aACtB,EAAK,KAAK,YAAc,MAE1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBACtB,EAAK,KAAK,qBAAuB,MAEnC,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KACtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAC1C,EACF,GAAU,KAAO,KAAO,EAAO,aACnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAKpD,EAAA,WAAP,SACI,EACA,GACF,MAAI,GAAO,gBAAqB,GAC9B,GAAO,eAAoB,GAEtB,GAAI,GAAI,IArCV,EAAA,UAAY,OAuCrB,GAzC0B,IA0C1B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAME,OADM,GAAsB,GACT,EAAA,EAAA,EAAA,KAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAU,KAAI,MAAd,EAAkB,EAAK,WAEvB,EAAU,KAAK,EAAK,WAGxB,MAAO,oCAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAS,EAKT,OAJI,GAAS,EAAO,MAAM,GAGpB,EAA2B,GACd,EAAA,EAAA,EAAA,EAAK,MAAM,QAAQ,UAAnB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAa,KAAK,EAAO,OAAO,EAAG,EAAK,UAAU,SAElD,EAAa,KAAK,EAAO,OAAO,EAAG,IAGvC,EAAa,UAKb,OAFM,GAA8B,GAChC,EACK,EAAI,EAAG,EAAI,EAAK,MAAM,OAAQ,EAAE,GACvC,GAAM,GAAO,EAAK,MAAM,GACxB,EAAS,EAAa,GAEtB,AAAI,IAAM,EACR,EAAa,CAAC,EAAO,IAAI,OAAO,GAEhC,EAAa,CAAC,EAAW,IAAI,OAAO,GAEtC,EAAa,EAAK,KAAK,EAAY,GACnC,EAAgB,KAAK,EAAW,MAAM,IAIxC,EAAS,GACT,OAAyB,GAAA,EAAA,EAAA,EAAgB,QAAQ,UAAxB,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAO,KAAI,MAAX,EAAe,GAEjB,MAAO,CAAC,EAAW,IAAI,OAAO,MAI3B,EAAA,UAAA,MAAP,SAAa,GACX,AAAI,GAAgB,IAGlB,GAAc,EAAuB,IAEvC,EAAa,EACb,GAAI,GACJ,KAAK,MAAM,QAAQ,SAAC,EAAM,GACxB,GAAU,WAAW,EAAK,WAGxB,EAAK,MAAM,GACX,AAAI,MAAM,QAAQ,EAAK,WACrB,EAAY,EAAK,UAAU,GAE3B,EAAY,EAAK,UAEnB,EAAa,CAAC,EAAW,GAAI,OAGjC,KAAK,MAAQ,IAGf,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAE5B,EAAgB,SAAC,GACrB,MAAO,CACL,UAAa,EAAK,eAClB,OAAU,EAAK,cAIb,EAAc,KAAK,MAAM,IAAI,GAE7B,EAAS,CAAC,MAAS,GAEzB,MAAA,IAAA,GAAW,EAAe,IAIrB,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAElB,OADM,GAAmB,GACA,EAAA,EAAA,EAAC,EAAO,MAAR,EAAA,EAAA,OAAA,KAApB,GAAM,GAAU,EAAA,GACnB,EAAM,KAAK,GAAY,EAAY,IAErC,MAAO,IAAI,GAAI,CAAC,MAAK,KAGvB,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,GAAI,CAAC,KAAK,UACR,MAAO,GAGT,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,kBAEvB,MAAO,oCAGT,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,qBAEvB,GAAI,CAAC,KAAK,WAER,OADM,GAAoC,GACvB,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAiB,KAAI,MAArB,EAAyB,EAAK,kBAEhC,MAAO,GAAiB,OAAO,GAEjC,MAAO,oCAQT,EAAA,UAAA,WAAA,WAEE,OADM,GAA2B,GACd,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,EAAQ,KAAI,MAAZ,EAAgB,EAAK,SAEvB,MAAO,IAAc,IASvB,EAAA,UAAA,WAAA,SAAW,GAET,OADM,GAAyC,GAC5B,EAAA,EAAA,EAAA,KAAK,MAAL,EAAA,EAAA,OAAA,IAGjB,OAHS,GAAI,EAAA,GACP,EAAY,EAAK,QAAQ,OACzB,EAAe,EAAQ,OAAO,GAC3B,EAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,EAAE,EACzC,EAAO,KAAK,CAAC,EAAK,QAAQ,GAAI,EAAa,KAG/C,GAAc,IA5KT,EAAA,UAAY,kBAgLrB,GAlLqC,IAmLrC,EAAA,cAAc,cAAc,gBAEQ,GAM3B,GAAA,GAAA,EAAA,KAAM,EAAA,EAAA,KAAM,EAAA,EAAA,SAAA,EAAA,IAAA,OAAA,GAAA,EAAkB,EAAA,EAAA,MAAA,EAAA,IAAA,OAAA,EAAA,EAE/B,EAAgB,WAAM,MAAA,IAAU,IAAQ,IAExC,EAAa,WAAM,MAAA,IAAe,EAAe,EAAM,IAG7D,GAAI,CAAC,GAAS,GAAS,EACrB,MAAO,GAAA,KAAS,IAAa,SAG/B,GAAM,GAAQ,MAAM,GAAO,KAAK,QAAW,IAAI,GAE/C,MAAO,GAAM,IAAI,SAAA,GAAK,MAAA,GAAA,KAAS,EAAE,WC/iEnC,AAiFA,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAArC,4DAyBA,MAAA,IAzBqC,IAiCrC,GAAA,SAAA,GAAwB,EAAA,EAAA,GAMtB,WAAY,GAAZ,GAAA,GAAA,KACE,GAAI,EAAK,OACP,KAAM,IAAI,IACN,sDAGN,GAAI,MAAM,QAAQ,EAAK,MACrB,KAAM,IAAI,IACN,kEAGN,SAAA,EAAA,KAAA,KAAM,IAAqB,KAE3B,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WAad,GAZA,AAAI,EAAK,KAAK,aAAe,MAC3B,GAAA,QAAY,EAAK,KAAK,aAEtB,EAAK,KAAK,YAAc,MAG1B,AAAI,EAAK,KAAK,sBAAwB,MACpC,GAAA,QAAY,EAAK,KAAK,sBAEtB,EAAK,KAAK,qBAAuB,MAG/B,GAAU,EAAO,UACnB,KAAM,IAAI,GAAW,6CAGvB,GAAM,GAAO,GAAU,KAAO,KAAO,EAAO,KAEtC,EAAW,GAAU,KAAO,KAAO,EAAO,SAE1C,EACF,GAAU,KAAO,KAAO,EAAO,aAEnC,MAAO,GAAA,UAAM,KAAI,KAAA,EAAC,EAAQ,CAAC,KAAI,EAAE,SAAQ,EAAE,aAAY,OAI3D,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GAAkB,KAAK,yBAAyB,GAEpD,MAAK,MAAK,iBACR,GAAQ,CAAI,EAAS,IAAE,OAAK,EAAS,MAAM,KAG7C,AAAI,KAAK,aACP,GAAQ,CACH,GAAQ,OAAK,MAAM,GAAG,KAAI,CAAE,EAAW,IAAE,OAAK,EAAS,MAAM,QAG7D,GAGT,EAAA,UAAA,gBAAA,SAAgB,GAAhB,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACP,GAAA,GAAA,EAAA,KAAA,UAED,EAAa,EAAO,MAEpB,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAe,EAAA,MAAU,GAE/B,MAAI,OAAM,QAAQ,GACT,MAAM,EAAU,QAAQ,KAAK,GAG/B,CAAC,MAIZ,EAAA,UAAA,YAAA,SAAY,EAA0B,GAAtC,GAAA,GAAA,KAAsC,AAAA,IAAA,QAAA,GAAA,IACpC,EAAA,KAAS,WACP,GAAI,CAAC,EAAK,SACR,KAAM,IAAI,IACN,mEAGN,GAAM,GAAa,EAAK,UAAU,GAAG,MAE/B,EAAc,EAAK,yBAAyB,GAE5C,EAAU,CAAI,EAAY,IAAE,OAAK,EAAY,MAAM,IAEnD,EAAY,EAAW,GAE7B,GAAI,GAAa,KACf,KAAM,IAAI,GACN,yUASN,GAAI,EAAK,aAAe,KACtB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAU,CAAC,EAAA,MAAU,YAEnB,GAAU,KAEnB,EAAA,QAAY,EAAK,SAGjB,AAAI,EAAK,YAAc,MACrB,GAAA,QAAY,EAAK,YACjB,EAAK,WAAa,IAGpB,AAAI,MAAM,QAAQ,EAAK,KAAK,WAC1B,EAAK,QAAU,EAAK,KAAK,UAAU,IAAI,WAAM,MAAA,GAAA,MAAU,KAEvD,EAAK,QAAQ,GAAK,EAAA,MAAU,QAO9B,GAJA,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAGR,EAAO,SAAW,EAAK,QAAQ,OACjC,KAAM,IAAI,GACN,SAAS,EAAK,KAAI,YAAY,EAAK,QAAQ,OAAM,cACjD,oBAAmB,EAAO,OAAM,2BAChC,cAAa,IAGnB,AAAI,EAKF,EAAK,WAAW,KAAK,EAAK,QAAQ,SAElC,EAAA,QAAY,EAAK,SAGnB,OAAS,GAAQ,EAAG,EAAQ,EAAK,QAAQ,OAAQ,EAAE,GACjD,GAAM,GAAQ,EAAO,GAEf,EAAgB,EAEtB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAM,MAAO,GACjC,KAAM,IAAI,GACN,SAAS,EAAK,+BAA+B,EAAK,KAAI,KACtD,mBAAkB,EAAa,oBAC3B,EAAM,QAGhB,EAAK,QAAQ,GAAS,GAI1B,EAAK,QAAU,EAAK,QAAQ,IAAI,SAAA,GAAS,MAAA,GAAA,KAAS,EAAM,cAIlD,EAAA,UAAA,yBAAV,SAAmC,GAC3B,GAAA,GAAA,KAAA,KAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,QAAS,EAAA,EAAA,aAGpD,EAAkB,IAAe,gBAEjC,EAAI,EAAW,EAAkB,EAAI,GACrC,EAAI,EAAW,EAAkB,EAAI,GAErC,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAClD,EAAO,GACT,EAAG,EAAW,GAAI,EAAS,EAAQ,GAAI,EAAa,IAElD,EACD,EAAW,MAAM,EAAG,GAAE,OACrB,EAAkB,CAAC,EAAS,EAAM,GAAQ,CAAC,EAAM,EAAM,IAG7D,MAAO,IAhMF,EAAA,UAAY,YAkMrB,GApMwB,IAyMxB,GAAA,SAAA,GAAoC,EAAA,EAAA,GAWlC,WAAY,GAAZ,GAAA,GAAA,KAEI,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,aAGF,SAAA,EAAA,KAAA,KAAA,GAAA,GAAU,EAAI,CAAE,MAAO,MAAS,KAEhC,EAAK,QAAU,EACf,GAAsB,EAAK,QAAS,WAEpC,EAAK,WAAa,GAAe,EAAY,EAAG,cAChD,EAAK,WAAW,QAAQ,SAAA,GAAQ,MAAA,IAAsB,EAAM,gBAE5D,EAAK,QAAU,GAAe,GAAW,EAAG,EAAG,WAC/C,EAAK,QAAQ,QAAQ,SAAA,GAAU,MAAA,IAAsB,EAAQ,aAE7D,EAAK,QAAU,GAAW,QAC1B,GAAiB,EAAK,SAEtB,EAAK,WAAa,GAAc,eAChC,GAAgB,EAAK,YAErB,EAAK,aAAe,GAAe,GAAgB,EAAG,EAAG,gBACzD,EAAK,aAAa,QACd,SAAA,GAAQ,MAAA,IAAsB,EAAM,oBAGnC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAEhC,GAAM,GACF,KAAK,aAAe,gBAAkB,EAAI,EAAW,OAAS,EAElE,GAAI,EAAW,IAAgB,KAC7B,KAAM,IAAI,GACN,yDACA,UAAS,EAAW,KAG1B,GAAM,GAAW,EAAW,GAEtB,EAAe,EAEf,EACF,KAAK,WAAW,OAAO,CAAC,EAAU,KAAK,QAAU,IAErD,KAAK,OAAS,KAAK,UACf,SAAU,EAAa,KAAM,KAAK,kBAClC,KAAK,kBAAmB,GAAM,KAAK,kBAEvC,GAAM,GACF,KAAK,WAAW,OAAO,CAAC,KAAK,QAAS,KAAK,QAAU,IAOzD,GALA,KAAK,gBAAkB,KAAK,UACxB,mBAAoB,EAAsB,KAC1C,KAAK,qBAAsB,KAAK,qBAAsB,GACtD,KAAK,qBAEL,KAAK,SACP,GAAI,GAAe,OAEnB,GAAI,KAAK,gBACP,GAAM,GAAO,KAAK,gBAEZ,EAAU,KAAK,QAErB,EAAkB,GAAA,GAAA,SAAA,GAA8B,EAAA,EAAA,GAAzB,4DAIrB,SAAA,UAAA,MAAA,SAAM,EAAc,GAClB,GAAM,GAAQ,EAAK,MAAM,CAAC,IACpB,EAAQ,EAAA,KAAS,CAAC,IAClB,EAAY,EAAK,MAAM,CAAC,EAAU,IACxC,MAAO,IAAc,CAAC,EAAO,EAAO,KAExC,GAVgD,IAEvC,EAAA,UAAY,oBAUrB,GAAkB,KAAK,gBAGzB,KAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,QAAU,GAAe,KAAM,EAC7C,KAAK,gBAAiB,GAAM,KAAK,gBAGvC,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAsB,GAA3B,GAAA,GAAA,KACE,MAAO,GAAA,KAAS,WACd,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,8DACG,GAAO,OAAM,MAGtB,GAAM,GAAW,EAAO,UAAe,GAEjC,EAAI,EAAO,GACX,EAAW,EAAO,GAClB,EAAW,EAAO,GAElB,EAAe,EAIrB,AAAI,EAAI,EAAK,SAAW,EAAK,QAAU,GAAK,EAAK,aAAe,MAC9D,GAAK,YAAc,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,QACX,SAAQ,EACR,MAAO,KAI9B,GAAM,GAAc,EAAK,YAEnB,EACF,SAAC,GAAe,GAAoB,IAClC,MAAI,CAAC,IAAQ,CAAC,GAAK,IACV,GAGF,EAAA,IAAQ,GAAK,IAAQ,KAG9B,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAClC,EAAK,EAAa,EAAG,EAAa,GAEtC,AAAI,EAAI,EAAK,kBAAoB,EAAK,iBAAmB,GACrD,EAAK,sBAAwB,MAC/B,GAAK,qBAAuB,GAAoB,CAClB,KAAM,WAAM,MAAA,GAAA,SAAa,IACzB,KAAM,EAAK,iBACX,SAAQ,EACR,MAAO,KAIvC,GAAM,GAAiB,EAAK,qBAExB,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAC5C,EAAK,EAAa,EAAU,EAAgB,GAE1C,EAAoB,EAEpB,EAAA,EAAA,MAAA,EAAA,OAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAAS,EAAA,EAAA,GAG5B,EAAA,EAAA,uDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAI5B,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAC7C,EAAK,EAAK,UAAU,EAAI,EAAS,EAAO,EAAK,SAEvC,GAAA,GAAA,EAAA,MAAA,EAAA,gBAAA,OAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAI3C,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAC5B,EAAK,EAAK,cAAc,EAAI,GAE5B,GAAM,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC/C,GAAI,EAAA,IACN,EAAA,IAAQ,GAAG,GACX,EAAA,IAAQ,EAAG,EAAK,WAAW,MAAM,EAAA,IAAQ,EAAI,MAC3C,GAAI,EAAA,IACN,EAAK,oBAAoB,MAAM,EAAA,IAAQ,EAAI,IAC3C,EAAK,WAAW,MAAM,KAE1B,MAAO,CAAC,GAAG,GAAG,OAIlB,EAAA,UAAA,UAAA,WACE,GAAM,GAAA,EAAA,UAAA,UAAA,KAAA,MAAC,EAAA,EAAA,MAAY,EAAA,GAAA,EAAA,CAAA,UAEb,EAAuC,CAC3C,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,QAAS,KAAK,QACd,WAAY,KAAK,WACjB,aAAc,KAAK,aACnB,QAAS,KAAK,SAGhB,MAAA,IAAA,GAAW,EAAe,IAG5B,EAAA,UAAA,UAAA,SAAU,EAAW,EAAW,EAAY,GAC1C,GAAM,GAAM,EAAA,OACR,EAAmB,EAAmB,KAAK,QAC1C,GAAW,QACZ,KAAK,aAAe,gBAAkB,OAAS,OAC/C,KAAK,cAET,MAAI,GACK,GAAU,EAAK,EAAG,KAAK,YAGzB,GAGT,EAAA,UAAA,cAAA,SAAc,EAAW,GACvB,GAAM,GAAU,EAEhB,MAAO,GAAA,OACH,EAAmB,EAAmB,EAAS,OAC/C,KAAK,aAAe,gBAAkB,OAAS,SAzO9C,EAAA,UAAY,iBA2OrB,GA7OoC,IA+OpC,EAAA,cAAkB,cAAc,IAKhC,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAI9B,WAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,GAAI,IAAe,GAEhC,SAAA,EAAA,KAAA,KAAM,GAAA,GAAI,EAAI,CAAE,KAAI,MAAwB,OAIvC,SAAA,WAAP,SACI,EACA,GACF,MAAO,IAAI,GAAI,IAZV,EAAA,UAAY,aAcrB,GAhBgC,IAkBhC,EAAA,cAAkB,cAAc,ICjkBhC,AAkDA,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAO3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,KAAK,IAAI,KAAK,IAAI,EAAK,KAAM,GAAI,GAE7C,EAAK,WAAa,EAAK,WACvB,EAAK,KAAO,EAAK,KACjB,EAAK,gBAAkB,KAGf,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,KAAK,YAAc,KACrB,MAAO,MAAK,WAId,OAFM,GAAa,EAAM,MACnB,EAAoB,GACjB,EAAI,EAAG,EAAI,KAAK,WAAW,OAAQ,EAAE,EAC5C,EAAW,KACP,KAAK,WAAW,IAAM,KAAO,EAAW,GAAK,KAAK,WAAW,IAEnE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAI,EAAK,MAAQ,EAAK,KAAO,GAC/B,GAAM,GACF,EAAO,UAAe,KAAO,GAAQ,EAAO,SAC1C,EAAa,EAAK,cAAc,GAChC,EAAS,GACX,WAAM,MAAA,IAAU,EAAO,EAAK,KAAM,EAAY,EAAK,OACnD,WAAM,MAAA,IAAO,GACjB,MAAO,GAET,MAAO,MAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,KACX,WAAY,KAAK,WACjB,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAxDf,EAAA,UAAY,UA0DrB,GA5D6B,IA6D7B,EAAA,cAAc,cAAc,IA4D5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAIpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,CAAC,KAAM,MAGjB,SAAA,UAAA,cAAV,SAAwB,GACtB,GAAM,GAAa,EAAM,MACzB,MAAO,CAAC,EAAW,GAAI,EAAG,EAAW,KAThC,EAAA,UAAY,mBAWrB,GAbsC,IActC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA2B,EAAA,EAAA,GAmBzB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAhBM,EAAA,WAA2B,KAC3B,EAAA,QAAU,GAGV,EAAA,OAAwB,KACxB,EAAA,KAAsB,KAErB,EAAA,2BAAoD,eACpD,EAAA,yBAAkD,QAQrD,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MACnD,EAAK,UAAY,MAGnB,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,EAAK,gBAAkB,CAAC,EAAW,EAAK,UAG1C,SAAK,MAAQ,EAAK,MAClB,GAAsB,EAAK,MAAO,SAClC,EAAK,WAAa,GAAc,EAAK,YACrC,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAK,SAEtB,EAAK,kBAAoB,GACrB,EAAK,mBAAqB,EAAK,4BACnC,EAAK,gBACD,GAAe,EAAK,iBAAmB,EAAK,0BAChD,EAAK,iBAAmB,GAAc,EAAK,kBAC3C,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,kBAAoB,GAAe,EAAK,mBAC7C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,gBAAkB,GAEvB,EAAK,UAAY,CAAC,CAAC,QAAS,MAGvB,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAe,EAAW,EAAW,OAAS,GACpD,AAAI,KAAK,QAAU,MACjB,MAAK,OAAS,KAAK,UACf,SAAU,CAAC,EAAc,KAAK,OAAQ,KAAM,KAAK,kBACjD,KAAK,kBAAmB,GAAM,KAAK,kBACvC,AAAI,KAAK,SACP,MAAK,KAAO,KAAK,UACb,OAAQ,CAAC,KAAK,OAAQ,KAAM,KAAK,gBACjC,KAAK,gBAAiB,GAAM,KAAK,kBAIzC,KAAK,UAAY,CAAC,CAAC,QAAS,EAAG,KAAI,GAAA,GAAG,EAAC,IAAK,EAAY,KACxD,KAAK,MAAQ,IAGf,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,SAAY,EAAY,OAAS,GAAK,KAAK,MACpC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAM,GAAQ,GAAoB,GAC5B,EACF,GAA2B,EAAK,WAAW,gBAC3C,EAEJ,MAAI,IAAuB,KACzB,EAAS,GACL,EAAO,EAAK,OAAO,OAAQ,EAC3B,EAAK,KAAO,EAAK,KAAK,OAAS,MAEnC,GAAS,GAAM,EAAO,EAAK,OAAO,QAClC,AAAI,EAAK,MAAQ,MACf,GAAS,GAAU,EAAQ,EAAK,KAAK,SAEvC,AAAI,EAAK,YAAc,MACrB,GAAS,EAAK,WAAW,MAAM,KAI5B,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAO,KAAK,MACZ,WAAY,GAAoB,KAAK,YACrC,QAAS,KAAK,QACd,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,kBAAmB,GAAqB,KAAK,mBAC7C,gBAAiB,GAAqB,KAAK,iBAC3C,oBAAqB,GAAqB,KAAK,qBAC/C,iBAAkB,GAAoB,KAAK,kBAC3C,eAAgB,GAAoB,KAAK,iBAErC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArHF,EAAA,UAAY,QAuHrB,GAzH2B,IA0H3B,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GAAA,KACE,SAAO,GAAQ,GACf,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,UAAY,CAAC,CAAC,QAAS,IAC5B,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,OAAkB,GAAA,EAAA,EAAA,EAAW,MAAM,GAAjB,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,GAAI,GAAO,KACT,KAAM,IAAI,GACN,4DACA,SAAQ,EAAW,MAAM,GAAE,oCAC3B,mFAIR,MAAO,CAAC,EAAW,GAAI,GAAU,EAAY,KAG/C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,GAAI,EAAK,aAAe,iBAAmB,EAAM,KAAO,GAEtD,OADM,GAAwB,CAAC,GACtB,EAAI,EAAG,EAAI,EAAM,KAAM,EAAE,EAChC,EAAY,KAAK,GAEnB,EAAY,KAAK,GACjB,EAAQ,EAAM,UAAU,GAG1B,MAAO,IAAe,MAI1B,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,GACzC,AAAI,KAAK,YAAc,MACrB,GAAO,WAAgB,KAAK,YAE9B,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA/CF,EAAA,UAAY,UAiDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAAgC,EAAA,EAAA,GAK9B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,WAAa,GAAc,EAAK,cAGvC,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAK,WAAW,MAAM,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,GAAoB,KAAK,aAC/C,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArBF,EAAA,UAAY,aAuBrB,GAzBgC,IA0BhC,EAAA,cAAc,cAAc,IAc5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAKhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,EAAI,EAAK,EACd,EAAK,UAAY,CAAC,CAAC,KAAM,MAG3B,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,KAAK,EAAG,EAAW,KAG5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAS,GAAoB,GACtB,GAAS,EAAQ,EAAK,MAIjC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,EAAG,KAAK,GAEJ,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA1BF,EAAA,UAAY,eA4BrB,GA9BkC,IA+BlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,YAAc,EAAK,YAGxB,OAAS,GAAI,EAAG,EAAI,EAAK,YAAY,OAAQ,EAAE,EAC7C,AAAI,EAAK,UAAU,EAAK,YAAY,KAClC,GAAK,YAAY,GAAK,eAKpB,SAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,GAAM,GAAK,GAAO,MAiBnB,EAAA,UAAA,oBAAR,SAA4B,EAAmB,GAK7C,OAJM,GAAW,6CACX,EAAa,EAAY,QAC3B,EAAQ,EACR,EAAU,KACL,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAM,EAAW,GACvB,GAAI,KAAK,UAAU,GACjB,GAAI,IAAY,KACd,EAAU,MAEV,MAAM,IAAI,GAAW,gDAGvB,IAAS,EAIb,GAAM,GAAe,GAAU,GAC/B,GAAI,IAAY,MACd,GAAI,IAAU,GAAK,EAAe,IAAU,EAC1C,KAAM,IAAI,GAAW,GAEvB,EAAW,GAAW,EAAe,UAC5B,IAAiB,EAC1B,KAAM,IAAI,GAAW,GAGvB,MAAO,IAGT,EAAA,UAAA,mBAAA,SAAmB,GAEjB,OADI,GAAiB,GACZ,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,EACvC,GAAI,KAAK,UAAU,EAAW,KAC5B,EAAiB,GACjB,MAIJ,MAAI,GACK,EAAW,MAAM,EAAG,GAAG,OAAO,KAAK,aAEnC,EAAW,MAAM,EAAG,GAAG,OAC1B,KAAK,oBAAoB,EAAW,MAAM,GAAI,KAAK,eAI3D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAc,EAAW,MAAM,EAAG,GAAG,OACvC,EAAK,oBAAoB,EAAW,MAAM,GAAI,EAAK,cACvD,MAAO,GAAM,QAAQ,MAIzB,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAlGF,EAAA,UAAY,UAoGrB,GAtG6B,IAuG7B,EAAA,cAAc,cAAc,IAY5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAM3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GAAI,EAAK,MAAQ,KACf,KAAM,IAAI,OACN,mFAGN,GAAI,CAAC,MAAM,QAAQ,EAAK,MACtB,KAAM,IAAI,OACN,oEACG,GAAK,KAAI,cAIlB,GAAM,GAAwB,GAAM,EAAG,EAAK,KAAK,OAAS,GAC1D,GAAI,CAAC,EAAA,KAAK,YAAY,EAAK,KAAK,QAAQ,OAAQ,GAC9C,KAAM,IAAI,OACN,+BAAiC,KAAK,UAAU,EAAK,MACrD,8DAGN,SAAK,KAAO,EAAK,KACjB,EAAK,mBAAqB,CAAC,GAAG,OAAO,EAAK,MAC1C,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,EAAK,KAAK,OAAS,OAG5D,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAc,EAAW,QAC/B,YAAK,KAAK,QAAQ,SAAC,EAAa,GAC9B,EAAY,EAAI,GAAM,EAAqB,KAEtC,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,UAAU,GAAoB,GAAS,KAAK,qBAGrD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,KAAM,KAAK,MAEP,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjDF,EAAA,UAAY,UAmDrB,GArD6B,IAsD7B,EAAA,cAAc,cAAc,IAS5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAK3B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAe,KAAU,KAC/B,SAAK,gBAAkB,GACvB,AAAI,GAAQ,KACV,EAAK,UAAY,EAAK,WAAa,KAAO,EAAI,EAAK,UAEnD,EAAK,UAAY,IAIrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,UAAW,KAAK,WAChC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACb,MAAO,GAAA,IAAI,EAAA,SAAS,EAAO,KAAK,WAAY,IAG9C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAO,GACP,EAAW,GACX,EAAc,EAAA,IAAI,EAAA,SAAS,EAAO,EAAK,WAAY,EAAM,GACzD,EAAS,EAAM,IAAI,EAAY,OAAO,EAAM,QAClD,MAAO,MAtCJ,EAAA,UAAY,UAyCrB,GA3C6B,IA4C7B,EAAA,cAAc,cAAc,IClqB5B,AA2EA,GAAA,IAAA,SAAA,GAA+B,EAAA,EAAA,GAgB7B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,GATM,EAAA,WAA4B,KAE3B,EAAA,+BACL,gBAME,EAAK,iBAAmB,MAAQ,EAAK,YAAc,MAKrD,GAAI,GAAoB,KACxB,AAAI,EAAK,WAAa,MACpB,GAAY,EAAK,WAEnB,AAAI,EAAK,aAAe,KAGtB,EAAK,gBAAkB,CAAC,EAAW,MAInC,EAAK,gBACD,CAAC,GAAW,OAAO,GAAqB,EAAK,cAGrD,SAAK,SAAW,EAAK,SACrB,GAAoC,EAAK,SAAU,YACnD,EAAK,UAAY,EAAK,UACtB,GAAoC,EAAK,UAAW,aACpD,EAAK,sBAAwB,GACzB,EAAK,uBAAyB,EAAK,gCACvC,EAAK,sBAAwB,GAAe,EAAK,uBACjD,EAAK,oBAAsB,GAAe,EAAK,qBAC/C,EAAK,qBAAuB,GAAc,EAAK,sBAC/C,EAAK,SAAW,EAAK,SACrB,EAAK,gBAAkB,EAAK,SAC5B,EAAK,YAAc,EAAK,cAGnB,SAAA,UAAA,MAAP,SAAa,GACX,KAAK,WAAa,KAAK,UACnB,aAAc,CAAC,KAAK,SAAU,KAAK,WAAY,KAAK,MACpD,KAAK,sBAAuB,KAAK,sBAAuB,GACxD,KAAK,sBACT,KAAK,MAAQ,IAKL,EAAA,UAAA,6BAAV,SAAuC,KAEvC,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAK,GAAK,SAGR,GAAS,GAAoB,GACtB,EAAA,SAAS,EAAQ,EAAA,UAAU,KAH3B,QAQb,EAAA,UAAA,mBAAA,SAAmB,GAEjB,GADA,EAAa,GAAmB,GAC5B,KAAK,aAAe,KACtB,MAAW,GAAU,OAAA,CAAE,KAAK,YAG9B,GAAM,GAAmB,GAAqB,KAAK,aACnD,GAAI,EAAO,SAAW,EAAW,OAAS,EACxC,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IAG7B,OADI,GAAI,EACC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAK,EAAO,GACZ,EAAK,EAAW,EAAI,GAC1B,GAAK,GAAM,MAAU,GAAM,MAAU,IAAO,EAC1C,KAAM,IAAI,GACN,oBAAoB,KAAK,YAAW,kBACpC,0BAAyB,IACxB,AAAI,GAAM,MACf,GAAO,GAAK,GAEd,IAGJ,MAAA,CAAQ,EAAW,IAAE,OAAK,EAAM,CAAE,KAAK,aAGzC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,GAAI,GAAQ,GAAoB,GAChC,AAAI,EAAM,QAAU,SAClB,GAAQ,GAAO,EAAO,UAExB,GAAM,GAAS,GAAS,EAAK,WAAW,OAAQ,EAAM,QACtD,MAAO,GAAO,QACV,GAAmB,EAAK,mBAAmB,EAAM,YAIzD,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,UAAW,KAAK,UAChB,sBAAuB,GAAqB,KAAK,uBACjD,sBAAuB,GAAqB,KAAK,uBACjD,oBAAqB,GAAqB,KAAK,qBAC/C,qBAAsB,GAAoB,KAAK,sBAC/C,SAAU,KAAK,SACf,YAAa,KAAK,aAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjIF,EAAA,UAAY,YAmIrB,GArI+B,IAsI/B,EAAA,cAAc,cAAc,ICjN5B,AA+BA,GAAA,IAAA,SAAA,GAAoC,EAAA,EAAA,GAGlC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,GAAQ,KAAG,KACjB,SAAK,gBAAkB,KAOf,SAAA,UAAA,cAAV,SAAwB,GACtB,KAAM,IAAI,KAaJ,EAAA,UAAA,gCAAR,SAAwC,EAAe,GACrD,GAAI,GAAU,MAAQ,GAAU,KAC9B,MAAO,MACF,GAAI,EAAO,OAAS,EAAO,OAChC,MAAO,MAAK,gCAAgC,EAAQ,GAC/C,GAAI,EAAO,SAAW,EAC3B,MAAO,GAGT,OADM,GAAqB,EAAO,MAAM,EAAG,EAAO,OAAS,EAAO,QACzD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,EAAO,OAAS,EAAO,OAAS,GAC3C,EAAI,EAAO,GACjB,GAAI,GAAK,MAAQ,GAAK,MAAQ,EAAI,GAAK,EAAI,EACzC,EAAY,KAAK,cACR,IAAM,EACf,EAAY,KAAK,WACR,IAAM,EACf,EAAY,KAAK,QAEjB,GAAI,IAAM,EACR,KAAM,IAAI,GACN,wDACA,KAAK,UAAU,GAAU,IAAM,KAAK,UAAU,IAEpD,EAAY,KAAK,IAGrB,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GAOJ,GALA,AAAI,MAAM,QAAQ,IAAe,CAAC,MAAM,QAAQ,EAAW,KAEzD,GAAa,CAAC,GAAmB,KAEnC,EAAa,EACT,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,mEACA,SAAQ,EAAW,OAAM,eAM/B,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAI1B,GADA,EAAa,GAAqB,GAC9B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,qDACA,6BAA4B,KAAK,UAAU,GAAW,MAK5D,OAFI,GACA,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GAC9C,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,GAAM,GAAW,EAAW,IAAI,SAAA,GAAS,MAAA,GAAM,SAC/C,AAAI,EAAW,QAAQ,QAAU,IAC7B,GAAqB,GAAU,SAAW,EAC5C,KAAK,gBAAkB,GAEvB,KAAK,gBAAkB,IAI3B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,GADA,EAAS,EACL,EAAK,iBACP,GAAM,GAA2B,GAC3B,EAAY,EAAO,IAAI,SAAA,GAAS,MAAA,GAAM,OAC5C,GAAI,EAAU,QAAQ,QAAU,IAI9B,OADM,GAAU,GAAc,GAChB,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAEZ,OAFO,GAAC,EAAA,GACF,EAAQ,EAAE,KACP,EAAI,EAAG,EAAI,EAAU,EAAO,EAAE,EACrC,EAAI,GAAa,EAAG,GAEtB,EAAe,KAAK,GAEtB,MAAO,GAAK,cAAc,QAK1B,OADI,GAAa,GACD,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAX,GAAM,GAAC,EAAA,GACJ,EAAQ,EAAE,KAChB,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAY,EAAO,GACnB,EAAW,EAAO,MAAM,GAAG,OAAO,CAAC,IACrC,EAAc,EAAE,QAChB,CAAC,GAAW,OAAO,GAAoB,EAAO,MAAM,MACxD,EAAc,EAAA,UAAc,EAAa,CAAC,EAAG,IAC7C,EAAc,EAAY,QAAQ,GAClC,EAAe,KAAK,GACpB,EAAa,WACJ,EAAQ,GACjB,GAAM,GAAO,GAAgB,EAAG,GAAO,OAAO,CAAC,IAC/C,EAAe,KAAK,EAAA,UAAc,EAAG,IACrC,EAAa,OAGb,GAAe,KAAK,GAGxB,GAAI,GAAI,EAAK,cAAc,GACrB,EAAQ,EAAE,KAChB,GAAI,GAGF,GAAI,GAAS,MACX,GAAM,GAAS,EAAE,MACX,EAAQ,EAAO,OACf,EAAY,EAAO,EAAQ,GAC3B,EACF,CAAC,GAAW,OAAO,EAAO,MAAM,EAAG,EAAO,OAAS,IACvD,EAAI,EAAA,UAAc,EAAE,QAAQ,CAAC,GAAI,IAAa,CAAC,EAAG,IACzC,QAAQ,WACR,EAAQ,GACjB,GAAM,GAAO,CAAC,EAAQ,GAAG,OAAO,GAAgB,EAAG,EAAQ,IAC3D,EAAI,EAAA,UAAc,EAAG,IAGzB,MAAO,QAGT,OAAO,GAAK,cAAc,MAKhC,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,EACb,GAAI,GACJ,AAAI,EAAW,IAAM,KACnB,EAAc,KAEd,EAAc,EAAW,GAAG,MAAM,GAEpC,OAAS,GAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAQ,EAAW,IAAM,KAAO,KAAO,EAAW,GAAG,MAAM,GACjE,EAAc,KAAK,gCAAgC,EAAa,GAIlE,OADI,GAAuB,GACP,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,AAAI,GAAS,MAAQ,EAAM,KAAO,MAChC,EAAW,KAAK,EAAM,IAG1B,SAAa,GAAqB,GAClC,AAAI,EAAW,SAAW,EACxB,EAAc,EAAW,OAAO,GAEhC,EAAc,CAAC,MAAM,OAAO,GAEvB,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,GAAA,KAAS,WACd,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6BAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+BAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,kGAEA,KAAI,EAAO,OAAM,OAAO,EAAK,OAAM,MAEzC,GAAI,EAAK,MAAM,SAAA,GAAK,MAAA,IAAK,OACvB,MAAO,MAET,EAAO,EAAK,IAAI,SAAA,GAAK,MAAA,IAAK,KAAO,EAAI,EAAA,WAAe,EAAG,KAEvD,OADI,GAAS,EAAK,GACT,EAAI,EAAG,EAAI,EAAK,OAAS,EAAG,EAAE,EACrC,EAAS,EAAA,WAAe,EAAQ,EAAK,IAEvC,MAAO,MAGb,GAlOoC,IAoOpC,GAAA,SAAA,GAAyB,EAAA,EAAA,GAGvB,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,MAcrB,GAhByB,IAiBzB,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA8B,EAAA,EAAA,GAG5B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,MAXJ,EAAA,UAAY,WAcrB,GAhB8B,IAiB9B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GAAG,QACd,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,IAAQ,EAAQ,EAAO,IAElC,MAAO,GAAA,IAAQ,EAAI,EAAO,OAAQ,MAX/B,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA2D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IA0D5B,GAAA,IAAA,SAAA,GAA6B,EAAA,EAAA,GAG3B,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,cAAV,SAAwB,GACtB,MAAO,GAAA,KAAK,WAEV,OADI,GAAS,EAAO,GACX,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAS,EAAA,QAAY,EAAQ,EAAO,IAEtC,MAAO,MAXJ,EAAA,UAAY,UAcrB,GAhB6B,IAiB7B,EAAA,cAAc,cAAc,IAiE5B,GAAA,IAAA,SAAA,GAAiC,EAAA,EAAA,GAM/B,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAJJ,SAAA,aAAe,GAKtB,AAAI,GAAQ,MACV,GAAO,IAET,EAAK,KAAO,EAAK,MAAQ,KAAO,EAAK,aAAe,EAAK,KACzD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,MACxD,EAAW,SAAW,EACxB,KAAM,IAAI,GACN,yEAGN,EAAa,EAGb,OADI,GAAe,GACC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,GAAS,MACX,EAAe,GACf,OAGJ,GAAI,EACF,OAIF,OADM,GAAoB,GACjB,EAAI,EAAG,EAAI,EAAW,OAAQ,EAAE,GACvC,GAAM,GAAyB,EAAW,GAAG,QAC7C,EAAuB,OAAO,KAAK,KAAM,GAEzC,OADI,GAAS,GACO,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAA,KAAK,YAAY,EAAO,IAC1B,EAAS,GACT,OAGJ,AAAK,GACH,EAAS,KAAK,GAGlB,GAAI,EAAS,OAAS,EACpB,KAAM,IAAI,GACN,4GAEA,KAAK,UAAU,KAIb,EAAA,UAAA,cAAV,SAAwB,GAAxB,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,MAAO,IAAc,EAAQ,EAAK,SAItC,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,CAAE,OAAM,QAAQ,IAAe,MAAM,QAAQ,EAAW,KAC1D,KAAM,IAAI,GACN,+DAON,OALM,GAAc,EACd,EAAc,EAAY,GAAG,QAC7B,EAAO,KAAK,KAAO,EAAI,EAAY,OAAS,KAAK,KAAO,KAAK,KAG/C,EAAA,EAAA,EAAA,EAAY,MAAM,GAAlB,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,GAAI,EAAY,IAAS,MAAQ,EAAM,IAAS,MAC9C,EAAY,GAAQ,KACpB,MAEF,EAAY,IAAS,EAAM,GAE7B,MAAO,IAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GAArC,GAAA,GAAA,KACE,GAAI,GAAQ,KACV,MAAO,MAET,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,6CAEvB,GAAI,CAAC,MAAM,QAAQ,GACjB,KAAM,IAAI,GAAW,+CAEvB,GAAI,EAAK,SAAW,EAAO,OACzB,KAAM,IAAI,GACN,mCAAmC,EAAK,OAAM,KAC9C,8BAA6B,EAAO,OAAM,MAEhD,MAAO,GAAA,KAAS,WACd,GAAI,GAAe,GAOnB,GANA,EAAK,QAAQ,SAAA,GACX,GAAI,GAAK,MACP,EAAe,GACf,UAGA,EACF,MAAO,MAGT,OADM,GAAwB,GACrB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,AAAI,EAAK,IAAM,KAEb,EAAY,KAAK,EAAA,SAAa,EAAO,IAAI,OAAO,SAC3C,AAAI,EAAK,GAAG,KAAO,EAAO,GAAG,KAElC,EAAY,KAAK,EAAA,WAAe,EAAK,GAAI,KAEzC,EAAY,KAAK,EAAK,IAG1B,GAAM,GAAoB,EAAA,OAAW,EAAa,EAAK,MACvD,MAAO,GAAA,IAAQ,EAAmB,GAAI,OAI1C,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,MAET,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArIF,EAAA,UAAY,cAuIrB,GAzIiC,IA0IjC,EAAA,cAAc,cAAc,IAuF5B,YAAuB,EAAc,GACnC,KAAO,EAAO,GACZ,GAAQ,EAEV,MAAO,GAGT,YAAkB,EAAW,EAAW,GACtC,GAAI,EAAE,MAAM,OAAS,GAAK,EAAE,MAAM,OAAS,EACzC,KAAM,IAAI,IACN,oEAeN,GAbA,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAC3B,EAAA,KAAS,OACL,EAAE,MAAM,QAAU,EAClB,WAAM,MAAA,+CACF,YAAW,EAAE,MAAM,UAE3B,AAAI,MAAO,IAAS,UAClB,GAAO,CAAC,EAAM,IAGZ,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,KAAM,IAAI,IACN,+DAGN,GAAM,GAAQ,EAAE,MAAM,OAChB,EAAQ,EAAE,MAAM,OACtB,AAAI,GAAQ,MAEV,GAAO,CAAC,EAAQ,EAAG,EAAQ,IAE7B,GAAM,GAAY,EAElB,MAAO,GAAA,KAAS,WACd,GAAI,GACJ,GAAI,EAAQ,GACV,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,YACpB,EAAQ,GACjB,EAAO,EAAQ,EAEf,OADM,GAAmB,GAChB,EAAI,EAAG,EAAI,EAAM,EAAE,EAC1B,EAAU,KAAK,GAEjB,EAAI,EAAE,QAAQ,EAAE,MAAM,OAAO,QAE7B,GAAO,EAGT,GAAI,GACJ,GAAI,EAAE,MAAM,SAAW,GAAK,EAAE,MAAM,SAAW,EAC7C,AAAI,EAAU,KAAO,EAAU,GAC7B,EAAM,EAAE,IAAI,GAAG,IAAI,EAAU,IAE7B,EAAM,EAAE,UAAU,CAAC,EAAG,IAAI,IAAI,GAAG,IAAI,EAAU,SAGjD,GAAM,GAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EACzC,EAAO,EAAU,KAAO,EAAE,MAAM,OAAS,EAC/C,EAAM,EAAE,OAAO,EAAG,EAAM,GAG1B,GAAI,EAAO,GACT,GAAI,GAAG,OACP,AAAI,EAAQ,EACV,EAAM,EAAQ,EAAQ,EAEtB,EAAM,EAAQ,EAGhB,OADM,GAAwB,GACrB,EAAI,EAAK,EAAI,EAAM,EAAM,EAAE,EAClC,EAAY,KAAK,GAEnB,EAAM,EAAI,QAAQ,GAEpB,MAAI,GAAI,MAAM,SAAW,GACvB,GAAM,EAAI,WAAW,IAEhB,IAIX,GAAA,IAAA,SAAA,GAAyB,EAAA,EAAA,GAOvB,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,KAAO,EAAK,KACjB,EAAK,UAAY,EAAK,WAAa,KAAO,GAAQ,EAAK,UACvD,EAAK,gBAAkB,GACvB,EAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GACJ,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAS,EAAW,GACpB,EAAS,EAAW,GAC1B,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,GAAI,EAAO,EAAK,MAAQ,EAAO,EAAK,IAClC,KAAM,IAAI,GACN,8BACG,GAAO,EAAK,IAAG,QAAQ,EAAO,EAAK,OAIpC,EAAA,UAAA,cAAV,SAAwB,GACtB,GAAI,EAAO,SAAW,EACpB,KAAM,IAAI,GACN,qDACA,iBAAgB,EAAO,OAAM,eAGnC,GAAI,GAAK,EAAO,GACZ,EAAK,EAAO,GACZ,EACJ,MAAK,OAAM,QAAQ,KAAK,MAMtB,EAAO,KAAK,KAAK,IACN,SAAC,EAAM,GAAM,MAAA,IACT,EAAM,EAAO,GAAG,MAAM,UAPrC,EAAO,CACL,GAAc,KAAK,KAAM,EAAG,MAAM,QAClC,GAAc,KAAK,KAAM,EAAG,MAAM,SAOtC,AAAI,KAAK,WACP,GAAK,GAAY,EAAI,EAAK,IAC1B,EAAK,GAAY,EAAI,EAAK,KAErB,GAAS,EAAI,EAAI,IAGlB,EAAA,UAAA,cAAR,SAAsB,EAAe,GACnC,GAAI,GACJ,MAAK,OAAM,QAAQ,KAAK,MAQtB,EAAO,KAAK,KANZ,EAAO,CACL,GAAc,KAAK,KAAM,EAAO,QAChC,GAAc,KAAK,KAAM,EAAO,SAM7B,GAGT,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAA,KAAS,OACL,MAAM,QAAQ,IAAe,EAAW,SAAW,GAC/C,MAAM,QAAQ,EAAW,KAAO,MAAM,QAAQ,EAAW,IAC7D,WAAM,MAAA,kEACV,GAAM,GAAU,EAAW,GAAa,QAClC,EAAU,EAAW,GAAa,QACxC,GAAI,EAAO,OAAS,GAAK,EAAO,OAAS,EACvC,KAAM,IAAI,IACN,gEAGN,GAAM,GAAO,KAAK,cAAc,EAAQ,GACxC,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAK,GAAI,GACvB,EAAO,OAAO,EAAG,GACjB,GAAM,GAAc,EAAO,OAAO,GAClC,MAAI,GAAY,SAAW,GACzB,EAAY,KAAK,GAEZ,GAGT,EAAA,UAAA,YAAA,SAAY,EAAyB,GACnC,MAAO,OAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAQ,KAAK,KACb,UAAa,KAAK,WAEd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GA7GF,EAAA,UAAY,MA+GrB,GAjHyB,IAkHzB,EAAA,cAAc,cAAc,IChjC5B,AA2BA,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAKjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,OAAS,EAAK,SAGrB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,OAAQ,KAAK,QAC7B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAC5B,EAAS,WACX,MAAA,IAAe,EAAM,MAAO,EAAG,EAAK,QAAQ,IAAI,IAC9C,EACF,GAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAC9D,MAAO,MA5BJ,EAAA,UAAY,gBA+BrB,GAjCmC,IAkCnC,EAAA,cAAc,cAAc,IAO5B,GAAA,IAAA,SAAA,GAAqC,EAAA,EAAA,GAKnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,OAGnB,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAC5B,GAAM,GAAQ,GAAoB,GAClC,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAS,WACb,GAAM,GAAS,KAAK,KAAK,EAAK,KAAQ,GAAI,EAAK,OAC/C,MAAO,GAAM,IAAI,GAAe,EAAM,MAAO,EAAG,KAElD,MAAO,IAAe,EAAQ,WAAM,MAAA,IAAO,EAAO,UAAe,IAEnE,MAAO,MA/BJ,EAAA,UAAY,kBAkCrB,GApCqC,IAqCrC,EAAA,cAAc,cAAc,IAyC5B,GAAA,IAAA,SAAA,GAAkC,EAAA,EAAA,GAMhC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,KACjB,EAAK,WAAa,EAAK,aAGzB,SAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,YAAc,GAAoB,GAAQ,OAGxD,EAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,IAGT,EAAA,UAAA,UAAA,WACE,GAAM,GAAa,EAAA,UAAM,UAAS,KAAA,MAC5B,EAAS,CAAC,KAAM,KAAK,MAC3B,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAI,EAAK,KAAO,GAAK,EAAK,KAAO,GAC/B,GAAM,GAAa,EAAK,eAAe,GAEjC,EAAgB,WACpB,GAAM,GAAQ,GAAoB,GAE5B,EAAQ,mBACR,EAAQ,mBAER,EAAS,CAAC,EAAQ,EAEpB,EAAU,EAAA,aAAa,EAAA,cAAc,GAAa,EAAK,MAE3D,EAAU,GAAO,EAAS,WAG1B,GAAM,GAAI,KAAA,IAAE,GAAI,EAAK,MAAS,GAAI,EAAK,KAAO,KAAA,IAAA,EAAU,IAAO,KACzD,EAAI,CAAC,EAAI,EAAS,EAAK,KAGvB,EAAI,EAAM,IAAI,GAAS,IAAI,EAAQ,IAAI,IAAI,IAAI,IAErD,MAAO,GAAE,IAAI,GAAG,IAAI,IAEtB,MAAO,IACH,EAAe,WAAM,MAAA,IAAoB,IACzC,EAAO,UAAe,IAE5B,MAAO,MAxDJ,EAAA,UAAY,eA2DrB,GA7DkC,IA8DlC,EAAA,cAAc,cAAc,IChN5B,YA4CI,EAAW,EAAc,EAAkB,EAAe,EAC1D,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GACJ,GAAI,EAAE,OAAS,EACb,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EAEpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,WACzB,EAAE,OAAS,EACpB,EAAM,EAAA,YACF,EAAe,EACf,EAAiC,EACjC,EAA8B,OAElC,MAAM,IAAI,IACN,2DAA2D,EAAE,KAAI,QAGvE,MAAO,GAoBT,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WACH,GAAM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EACF,GAAmB,EAAG,EAAM,EAAU,EAAM,EAAO,GACvD,MAAO,CAAC,EAAQ,EAAM,KAqBjC,YACI,EAAW,EAAe,EAAc,EACxC,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,EAAA,KAAK,WAKH,OAJM,GAAkB,EAAA,QAAY,EAAG,GACjC,EAAO,EAAgB,KACvB,EAAW,EAAgB,SAC3B,EAAwB,GACX,EAAA,EAAA,EAAA,GAAiB,EAAG,EAAE,MAAtB,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,AAAI,EAAc,QAAQ,KAAU,GAClC,EAAY,KAAK,GAEjB,EAAY,KAAK,EAAE,MAAM,IAG7B,GAAM,GAAgB,EAAK,QAAQ,GAC7B,EAAoB,EAAS,QAAQ,GACrC,EACF,GAAS,KAAO,KAAO,EAAM,QAAQ,GACnC,EACF,GAAQ,KAAO,KAAO,EAAK,QAAQ,GACjC,EAAS,GACX,EAAG,EAAe,EAAmB,EACrC,EAAgB,GACpB,MAAO,CAAC,EAAQ,EAAM,iBAgB7B,EAAW,EAAe,EAAc,EACxC,GACF,MADE,AAAA,KAAA,QAAA,GAAA,MACE,EAAA,KAAK,YACD,EAAc,QAAQ,OAAQ,GAAiB,EAAG,EAAE,KAAO,IAC1D,GACH,EAAG,EAAO,EAAM,EAAe,GAE5B,GACH,EAAG,EAAO,EAAM,EAAe,GAsFvC,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAqBtC,WAAY,GAAZ,GAAA,GAAA,KACE,MAAI,IAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,gBAAkB,GACvB,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KAC1C,EAAK,SAAW,EAAK,UAAY,KAAO,IAAO,EAAK,SACpD,EAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,sBACD,GAAe,EAAK,uBAAyB,SACjD,EAAK,0BACD,GAAe,EAAK,2BAA6B,QACrD,EAAK,eAAiB,GAAc,EAAK,gBACzC,EAAK,gBAAkB,GAAc,EAAK,iBAC1C,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,oBAGvC,SAAA,UAAA,MAAP,SAAa,SACX,EAAa,GAAmB,GAChC,GAAM,GAAO,KAAK,MAAQ,EAAI,KAAK,KAAQ,KAAK,KAAO,EAAW,OAC5D,EAAM,EAAW,GACvB,GAAI,GAAO,KACT,KAAM,IAAI,GACN,QAAQ,EAAI,+FAET,MAAK,UAAU,GAAW,MAEnC,KAAK,UACD,CAAC,GAAI,IAAU,CAAC,KAAM,EAAW,OAAQ,KAAI,GAAA,GAAG,EAAC,GAAO,EAAG,MAC/D,GAAM,GAAQ,CAAC,GACf,AAAI,KAAK,OACP,MAAK,MAAQ,KAAK,UACd,QAAS,EAAO,KAAM,KAAK,iBAAkB,KAAK,iBAClD,GAAM,KAAK,kBAEjB,AAAI,KAAK,QACP,MAAK,KAAO,KAAK,UACb,OAAQ,EAAO,KAAM,KAAK,gBAAiB,KAAK,gBAAiB,GACjE,KAAK,iBAEX,KAAK,WAAa,KAAK,UACnB,cAAe,EAAO,KAAM,KAAK,sBAAuB,KAAM,IAClE,KAAK,eAAiB,KAAK,UACvB,kBAAmB,EAAO,KAAM,KAAK,0BAA2B,KAChE,IACJ,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAW,EAAO,UAAe,KAAO,GAAQ,EAAO,SACvD,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAO,EAAW,OAClB,EAAgB,GAAiB,EAAG,GACpC,EAAO,EAAK,MAAQ,EAAI,EAAK,KAAQ,EAAK,KAAO,EACvD,EAAc,OAAO,EAAM,GAC3B,GAAM,GAAiB,GAA2B,EAAG,GACrD,EAAe,GAAQ,EAAW,GAElC,GAAM,GAAsB,EAAc,QAC1C,EAAoB,OACpB,GAAM,GAAoB,CAAC,EAAA,KAAK,YAC5B,EAAqB,GAAiB,EAAG,GAAM,MAAM,EAAG,EAAO,IAE7D,EAAmC,WACvC,GAAI,GACF,GAAM,GACF,EAAK,WAAW,OAAO,QAAQ,GAC7B,EACF,EAAK,eAAe,OAAO,QAAQ,GACjC,EACF,EAAK,OAAS,EAAK,KAAK,OAAO,QAAQ,GAAkB,KACvD,EACF,EAAK,MAAQ,EAAK,MAAM,OAAO,QAAQ,GAAkB,KAC7D,MAAO,IACH,EAAO,EAAqB,EAC5B,EAAe,EAAgB,EAAK,aAExC,OAAO,IACH,EAAO,EAAK,WAAW,OAAQ,EAAK,eAAe,OACnD,EAAK,MAAQ,KAAO,KAAO,EAAK,KAAK,OACrC,EAAK,OAAS,KAAO,KAAO,EAAK,MAAM,OAAQ,EAAK,UAI5D,GAAI,CAAC,EACH,MAAO,KAGH,GAAA,GAAA,GAAA,EAAA,EAAA,MAAA,OAAA,EAAA,KAAA,OAAA,EAAA,EAAA,SAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAM,EAAA,EAAA,GAIvB,EACF,SAAC,EAAyB,EAAe,GACvC,EAAA,KAAS,WACP,GAAM,GAAQ,EAAI,EACZ,EAAY,EAAS,OACrB,EAAc,EAAU,IAAI,GAAO,IAAI,GAC7C,EAAS,MAAM,EAAU,IAAI,OAU/B,EAA8B,WAClC,EAAgB,EAAK,WAAY,EAAM,EAAK,UAC5C,EAAgB,EAAK,eAAgB,EAAU,EAAK,WAEtD,WAEO,KAIX,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,SAAU,KAAK,SACf,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,sBAAuB,GAAqB,KAAK,uBACjD,0BACI,GAAqB,KAAK,2BAC9B,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,eAAgB,GAAoB,KAAK,gBACzC,gBAAiB,GAAoB,KAAK,kBAEtC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GArKF,EAAA,UAAY,qBAuKrB,GAzKwC,IA0KxC,EAAA,cAAc,cAAc,IAkD5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAgBtC,WAAY,GAAZ,GAAA,GAAA,KAOE,GANA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,KAAO,EAAK,MAAQ,KAAO,GAAK,EAAK,KACtC,MAAO,GAAK,MAAS,UACvB,GAAI,CAAC,OAAO,UAAU,EAAK,MACzB,KAAM,IAAI,OACN,gDAAgD,EAAK,cAElD,MAAM,QAAQ,EAAK,MAC5B,OAAmB,GAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,6CACA,iBAAgB,KAAK,UAAU,EAAK,YAI5C,MAAM,IAAI,OACN,2DACA,iBAAgB,KAAK,UAAU,EAAK,QAG1C,SAAK,QAAU,EAAK,SAAW,KAAO,KAAO,EAAK,QAClD,EAAK,OAAS,EAAK,QAAU,KAAO,GAAO,EAAK,OAChD,EAAK,MAAQ,EAAK,OAAS,KAAO,GAAO,EAAK,MAC9C,EAAK,gBAAkB,GAAe,EAAK,iBAAmB,SAC9D,EAAK,iBAAmB,GAAe,EAAK,kBAAoB,QAChE,EAAK,gBAAkB,GAAe,EAAK,iBAC3C,EAAK,iBAAmB,GAAe,EAAK,kBAE5C,EAAK,gBAAkB,KAGlB,SAAA,UAAA,MAAP,SAAa,GACX,EAAa,GAAmB,GAChC,GAAM,GAAQ,EAAW,OAGzB,AAAI,MAAO,MAAK,MAAS,UACvB,MAAK,KAAO,CAAC,KAAK,OAEpB,OAAS,GAAI,EAAG,EAAI,KAAK,KAAK,OAAQ,EAAE,EACtC,AAAI,KAAK,KAAK,GAAK,GACjB,MAAK,KAAK,IAAM,GAKpB,OAAmB,GAAA,EAAA,EAAA,KAAK,KAAL,EAAA,EAAA,OAAA,KAAd,GAAM,GAAI,EAAA,GACb,GAAI,EAAO,GAAK,GAAQ,EACtB,KAAM,IAAI,OAAM,iBAAiB,GAGrC,GAAI,KAAK,KAAK,SAAW,GAAqB,KAAK,MAAM,OACvD,KAAM,IAAI,OAAM,4BAA4B,KAAK,MAGnD,GAAM,GAAa,KAAK,KAAK,IAAI,SAAA,GAAQ,MAAA,GAAW,KAE9C,EAAY,GAClB,AAAI,KAAK,MACP,KAAK,MAAQ,KAAK,UACd,QAAS,EAAY,UAAW,KAAK,iBACrC,KAAK,iBAAkB,GAE3B,KAAK,MAAQ,KAEf,AAAI,KAAK,OACP,KAAK,KAAO,KAAK,UACb,OAAQ,EAAY,UAAW,KAAK,gBACpC,KAAK,gBAAiB,GAE1B,KAAK,KAAO,KAGd,KAAK,MAAQ,IAGf,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACQ,EAAQ,GAAoB,GAC5B,EAAa,EAAM,MACnB,EAAQ,EAAW,OAEzB,MAAO,GAAA,KAAK,WAIV,OAHM,GAAW,GACb,EAAA,EAAA,QAAA,EAAA,EAAA,KAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,SACL,EAAiB,GAA2B,EAAG,GACnC,EAAA,EAAA,EAAA,EAAK,KAAL,EAAA,EAAA,OAAA,KAAb,GAAM,GAAG,EAAA,GACZ,EAAe,GAAO,EAAW,GAuBnC,OApBM,GAAY,SAAC,GACjB,MAAI,IAAK,MAAQ,EAAE,MAAM,SAAW,GAChC,EAAK,OAAS,CAAC,EAAQ,GAClB,EAAE,QAAQ,GAEV,GAIP,EAAQ,EAAU,EAAK,MAAM,QAC7B,EAAS,EAAU,EAAK,KAAK,QAQ3B,EAA0B,GAC1B,EAA8B,GAC3B,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,AAAK,EAAK,KAAkB,QAAQ,KAAO,GACzC,GAAc,KAAK,EAAW,IAC9B,EAAkB,KAAK,IAEvB,GAAc,KAAK,GACnB,EAAkB,KAAK,EAAW,KAGtC,SAAO,EAAK,KAAK,GACjB,EAAW,EAAS,KAAK,GACzB,EAAQ,EAAM,KAAK,GACnB,EAAS,EAAO,KAAK,GAEd,GACH,EAAO,EAAM,EAAU,EAAQ,EAAO,EAAK,YAInD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,KAAM,KAAK,KACX,QAAS,KAAK,QACd,OAAQ,KAAK,OACb,MAAO,KAAK,MACZ,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,kBAC5C,gBAAiB,GAAqB,KAAK,iBAC3C,iBAAkB,GAAqB,KAAK,mBAExC,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAjKF,EAAA,UAAY,qBAmKrB,GArKwC,IAsKxC,EAAA,cAAc,cAAc,IC9nB5B,YAoEI,EAAW,EACX,GACF,MAAO,GAAA,KAAK,WACV,GAAI,EAAE,OAAS,EACb,KAAM,IAAI,GACN,kEACG,GAAE,KAAI,eAMf,GAHA,AAAI,GAAW,MACb,GAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,KAErB,EAAQ,SAAW,GAAK,EAAQ,GAAG,SAAW,GAC9C,EAAQ,GAAG,SAAW,EACxB,KAAM,IAAI,GACN,+GAON,GAHA,AAAI,GAAc,MAChB,GAAa,MAEX,IAAe,gBAAkB,IAAe,gBAClD,KAAM,IAAI,GACN,wBAAwB,EAAU,mEAIxC,GAAI,GACJ,MAAI,KAAe,gBACjB,EAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,IAE/C,EAAU,CAAC,CAAC,EAAG,GAAI,EAAQ,GAAI,EAAQ,GAAI,CAAC,EAAG,IAG1C,EAAA,IAAQ,EAAG,KA6BtB,GAAA,IAAA,SAAA,GAAmC,EAAA,EAAA,GAMjC,WAAY,GAAZ,GAAA,GAAA,KAUE,GATA,AAAI,GAAQ,MACV,GAAO,IAET,EAAA,EAAA,KAAA,KAAM,IAAK,KAEX,EAAK,WACD,EAAK,YAAc,KAAO,KAAoB,EAAK,WAGnD,EAAK,SAAW,KAClB,EAAK,QAAU,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,YACnB,MAAO,GAAK,SAAY,SACjC,EAAK,QACD,CAAC,CAAC,EAAK,QAAS,EAAK,SAAU,CAAC,EAAK,QAAS,EAAK,eAGvD,GADA,EAAK,QAAU,EAAK,QAChB,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,6DACA,sBAAqB,EAAK,QAAQ,OAAM,YAG9C,GAAI,GAAa,OACb,EAAY,OAChB,GAAI,MAAO,GAAK,QAAQ,IAAO,SAC7B,EAAgB,CAAC,EAAK,QAAQ,GAAI,EAAK,QAAQ,IAC/C,EAAe,CAAC,EAAK,QAAQ,GAAc,EAAK,QAAQ,SAIxD,GAFA,EAAK,QAAU,EAAK,QAEhB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,gEACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAIrD,GAFA,EAAgB,EAAK,QAAQ,GAEzB,EAAK,QAAQ,GAAG,SAAW,EAC7B,KAAM,IAAI,GACN,+DACA,0BAAyB,EAAK,QAAQ,GAAG,OAAM,YAErD,EAAe,EAAK,QAAQ,GAE9B,EAAK,QAAU,CAAC,EAAe,GAEjC,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAEhC,GAAI,GACA,EACJ,MAAI,MAAK,aAAe,gBACtB,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,IAE5C,CAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAET,AAAI,EAAW,IAAM,MAAQ,EAAW,IAAM,EAC5C,EAAO,EAAW,GAAK,KAAK,QAAQ,GAAG,GAAK,KAAK,QAAQ,GAAG,GAE5D,EAAO,KAEF,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,MAIlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KACH,WAAM,MAAA,IACF,GAAoB,GAAS,EAAK,QAAS,EAAK,eAG1D,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAnGF,EAAA,UAAY,gBAqGrB,GAvGmC,IAwGnC,EAAA,cAAc,cAAc,IC5O5B,YA0CI,EAAW,EAA4B,EACvC,EAAuB,EACvB,GACF,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,IAEhB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAKb,EAAI,GAAsB,EAAG,GAC7B,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MAEf,EAAI,EAAA,QAAY,EAAe,EAAU,EAAS,GAIlD,EAAI,EAAA,QAEA,EAA0B,EAAU,EAAS,GAEnD,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,KAE1B,gBAeP,EAAa,EACb,EAAoC,EACpC,EAAyB,GAC3B,MAAO,GAAA,KAAK,WACV,GAAgB,GAChB,GAAc,GACd,GAAiB,GACjB,AAAI,GAAW,MACb,GAAU,CAAC,EAAG,EAAG,IAEnB,AAAI,GAAW,MACb,GAAU,SAEZ,AAAI,GAAc,MAChB,GAAa,MAEf,AAAI,GAAY,MACd,GAAW,OAIb,EAAI,GAAsB,EAAa,GACvC,GAAI,GACE,EAAiB,IAAY,OAAU,OAAS,QACtD,MAAI,KAAa,MACf,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAExC,EAAI,EAAA,UAAc,EAAG,EAAU,EAAS,GAE1C,AAAI,IAAe,iBACjB,GAAI,EAAA,UAAc,EAAG,CAAC,EAAG,EAAG,EAAG,EAAG,KAE7B,IAsBX,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAWtC,WAAY,GAAZ,GAAA,GAAA,KAKE,GAJA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,GAElB,EAAA,EAAA,KAAA,KAAM,IAAK,KACP,MAAO,GAAK,UAAa,SAC3B,EAAK,SAAW,CAAC,EAAK,kBAEpB,MAAM,QAAQ,EAAK,WAClB,EAAK,SAAsB,SAAW,GACvC,MAAQ,GAAK,SAAsB,IAAO,SAC5C,EAAK,SAAW,EAAK,aAErB,MAAM,IAAI,GACN,qGAEA,IAAG,KAAK,UAAU,EAAK,YAG7B,GADA,GAAsB,EAAK,SAAU,YACjC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBAEhB,MAAO,GAAK,SAAY,SAC1B,EAAK,QAAU,CAAC,EAAK,iBAEnB,MAAM,QAAQ,EAAK,UAClB,EAAK,QAAqB,SAAW,GACtC,MAAQ,GAAK,QAAqB,IAAO,SAC3C,EAAK,QAAU,EAAK,YAEpB,MAAM,IAAI,GACN,oGAEA,IAAG,KAAK,UAAU,EAAK,WAG/B,UAAsB,EAAK,QAAS,WAEpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,GAAiB,EAAK,SACtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAS,GACX,EAAW,GAAI,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IAChE,MAAO,CAAC,EAAW,GAAI,EAAQ,EAAW,KAO5C,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,EAAK,eAAe,EAAQ,GAE5B,EAAS,GAAa,GAAoB,GAAS,GACnD,GAAM,GAAS,EAAK,gBAChB,GAAoB,GAAS,CAAC,EAAK,SAAS,GAAI,GAChD,CAAC,EAAK,QAAQ,GAAI,GAAI,EAAK,QAAS,gBAExC,MAAO,GAAA,QAAY,EAAQ,CAAC,OAIhC,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,SAEV,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAxFwC,IA0FxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,IAEtB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,UACrB,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,SAErC,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAKrE,MAJA,GACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAM,GAErC,CAAC,EAAW,GAAI,EAAM,EAAM,EAAW,KAQlD,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GAhFwC,IAkFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,eAYrB,GAdkC,IAelC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAA4B,EAC5C,EAAsB,GACxB,UAAgB,GAChB,GAAiB,GACV,GAAO,EAAQ,EAAU,EAAS,EAAS,EAAY,QAVzD,EAAA,UAAY,mBAYrB,GAdsC,IAetC,EAAA,cAAc,cAAc,IA+B5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAMtC,WAAY,GAAZ,GAAA,GAAA,KAQE,GAPA,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,CAAC,EAAG,EAAG,IAEzB,EAAA,EAAA,KAAA,KAAM,IAAK,KACX,EAAK,SAAW,MAAM,QAAQ,EAAK,UAC/B,EAAK,SACL,CAAC,EAAK,SAAU,EAAK,SAAU,EAAK,UACpC,EAAK,SAAW,KAClB,EAAK,QAAU,EAAK,iBACX,MAAM,QAAQ,EAAK,UAC5B,GAAI,EAAK,QAAQ,SAAW,EAC1B,KAAM,IAAI,GACN,wHAEG,GAAK,QAAQ,OAAM,MAE5B,EAAK,QAAU,EAAK,YAGpB,GAAK,QAAU,CAAC,EAAK,QAAS,EAAK,QAAS,EAAK,SAEnD,UAAsB,EAAK,SAAU,YACrC,GAAsB,EAAK,QAAS,WACpC,EAAK,QAAU,EAAK,SAAW,KAAO,QAAU,EAAK,QACrD,EAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,GAAiB,EAAK,SAEtB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAI,GACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GACjE,EACA,KAAK,aAAe,gBAAkB,EAAW,GAAK,EAAW,GAOrE,MANA,GAAS,GACL,EAAQ,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACzD,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACxE,EACI,GAAiB,EAAM,KAAK,SAAS,GAAI,KAAK,QAAS,KAAK,QAAQ,IACpE,KAAK,aAAe,gBACf,CAAC,EAAW,GAAI,EAAW,GAAI,EAAQ,EAAM,GAE7C,CAAC,EAAW,GAAI,EAAQ,EAAM,EAAM,EAAW,KAS1D,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,SAAK,eAAe,EAAQ,GACrB,EAAK,gBACR,GAAoB,GAAS,EAAK,SAAU,EAAK,QACjD,EAAK,QAAS,EAAK,eAI3B,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CACb,SAAU,KAAK,SACf,QAAS,KAAK,QACd,QAAS,KAAK,QACd,WAAY,KAAK,YAEb,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GArFwC,IAuFxC,GAAA,SAAA,GAAkC,EAAA,EAAA,GAGhC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,eAcrB,GAhBkC,IAiBlC,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,SACV,GAAA,KAAA,KAAM,IAAK,KAGH,SAAA,UAAA,gBAAV,SACI,EAAgB,EAChB,EAAmC,EACnC,GACF,UAAgB,GAChB,GAAiB,GACV,GACH,EAAoB,EAAU,EAAS,EAAS,EAAY,QAZ3D,EAAA,UAAY,mBAcrB,GAhBsC,IAiBtC,EAAA,cAAc,cAAc,IAK5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAC5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GACjB,MAAO,CAAC,EAAW,GAAI,EAAW,KAGpC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAEd,GAb8C,IAe9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAG1C,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,KAAS,EAAO,MARpB,EAAA,UAAY,yBAWrB,GAb4C,IAc5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAGtC,WAAY,SACV,GAAA,KAAA,KAAM,GAAQ,KAAG,KAGnB,SAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAO,GAAA,IAAQ,EAAO,MARnB,EAAA,UAAY,qBAWrB,GAbwC,IAcxC,EAAA,cAAc,cAAc,IAiB5B,GAAA,IAAA,SAAA,GAA8C,EAAA,EAAA,GAE5C,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,WACD,EAAK,YAAc,KAAO,eAAiB,EAAK,WACpD,GAAgB,EAAK,YACrB,EAAK,UAAY,CAAC,GAAI,IAAU,CAAC,KAAM,OAGzC,SAAA,UAAA,mBAAA,SAAmB,GAEjB,MADA,GAAa,EACT,KAAK,aAAe,eACf,CAAC,EAAW,GAAI,EAAW,IAE3B,CAAC,EAAW,GAAI,EAAW,KAItC,EAAA,UAAA,KAAA,SAAK,EAAyB,GAC5B,KAAM,IAAI,KAGZ,EAAA,UAAA,UAAA,WACE,GAAM,GAAS,CAAC,WAAY,KAAK,YAC3B,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAEX,GA7B8C,IA+B9C,GAAA,SAAA,GAA4C,EAAA,EAAA,GAA5C,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,KAAS,EAAO,CAAC,EAAG,IAEpB,EAAA,KAAS,EAAO,CAAC,EAAG,OAR1B,EAAA,UAAY,yBAYrB,GAd4C,IAe5C,EAAA,cAAc,cAAc,IAE5B,GAAA,IAAA,SAAA,GAAwC,EAAA,EAAA,GAAxC,4DAIE,SAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAQ,GAAoB,GAClC,MAAI,GAAK,aAAe,eACf,EAAA,IAAQ,EAAO,CAAC,EAAG,IAEnB,EAAA,IAAQ,EAAO,CAAC,EAAG,OARzB,EAAA,UAAY,qBAYrB,GAdwC,IAexC,EAAA,cAAc,cAAc,IC5rB5B,AA4CA,GAAA,IAAA,SAAA,GAAsC,EAAA,EAAA,GAGpC,WAAY,GAAZ,GAAA,GAQE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,MAAQ,EAAK,QAGpB,SAAA,UAAA,MAAA,SAAM,GACJ,KAAK,MAAQ,IAKf,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAIE,MAAI,MAAK,OAAS,KACT,KAAK,MAAM,UAEX,QAIX,SAAc,GAIZ,AAAI,KAAK,OAAS,MAChB,MAAK,MAAM,UAAY,oCAI3B,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,MAAM,kDAIpB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,MAAM,qDAIpB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WAEE,MAAQ,MAAK,MAAc,0CAK7B,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,MAAM,wCAKpB,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,MAAM,cAGpB,EAAA,UAAA,WAAA,SAAW,GACT,KAAK,MAAM,WAAW,IAGxB,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,MAAS,CACP,UAAa,KAAK,MAAM,eACxB,OAAU,KAAK,MAAM,cAGnB,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAGT,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,OAAS,MAChB,KAAK,MAAM,6BAA6B,IAKrC,EAAA,WAAP,SACI,EACA,EACA,GAAA,AAAA,IAAA,QAAA,GAAgB,IAClB,GAAM,GAAc,EAAO,MACrB,EAAQ,GAAY,EAAa,GACvC,MAAO,GAAO,MACd,GAAM,GAAY,CAAC,MAAK,GACxB,cAAO,OAAO,EAAW,GAClB,GAAI,GAAI,IAEnB,GAvGsC,IAyGtC,GAAA,SAAA,GAAqC,EAAA,EAAA,GAGnC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KACX,SAAK,gBAAkB,KAGzB,SAAA,UAAA,MAAA,SAAM,GAEJ,GADA,EAAa,GAAmB,GAC5B,EAAW,OAAS,EACtB,KAAM,IAAI,GACN,oEACA,gBAAe,KAAK,UAAU,KAEpC,KAAK,UAAY,CAAC,CAAC,MAAO,IAC1B,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAChE,AAAK,KAAK,MAAM,OACd,MAAK,MAAM,MAAM,GACjB,KAAK,MAAM,MAAQ,IAErB,EAAA,UAAM,MAAK,KAAA,KAAC,IAGd,EAAA,UAAA,mBAAA,SAAmB,GACjB,EAAa,GAAmB,GAChC,GAAM,GAAkB,CAAC,EAAW,IAAI,OAAO,EAAW,MAAM,IAC1D,EACF,KAAK,MAAM,mBAAmB,GAC5B,EAAY,EAAW,GAC7B,MAAO,CAAC,EAAiB,GAAI,GAAW,OAAO,EAAiB,MAAM,KAGxE,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WAEV,EAAS,GAAoB,GAI7B,GAAM,GAAwB,SAAC,EAAgB,GAK7C,GAAM,GAAS,GAAoB,EAAK,MAAM,KAAK,EAAQ,IAC3D,MAAO,CAAC,EAAQ,KAEZ,EACF,GAAI,EAAM,EAAQ,GAAI,GAAyB,KAC3C,KAAsB,GACtB,IACF,EAAI,EAAW,GAGrB,MAAO,MArDJ,EAAA,UAAY,kBA0DrB,GA5DqC,IA6DrC,EAAA,cAAc,cAAc,gBAEgB,GAC1C,GACI,GAAiC,yBAA0B,GAmBjE,GAAM,IAA2D,SAEjE,GAAA,SAAA,GAAmC,EAAA,EAAA,GAWjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,KAAM,IAAK,KAUL,EAAc,EAAK,MAAM,YACzB,EAAqC,GAC3C,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,aAAe,GAAY,GAChC,EAAY,YACR,IAAY,cAAmB,IACnC,GAAM,GAAqC,GAW3C,GAVA,EAAS,UAAe,EAAK,MAAM,eACnC,EAAS,OAAY,EACrB,EAAK,cAAgB,GAAY,GACjC,EAAK,aAAa,KAAO,WAAa,EAAK,aAAa,KACxD,EAAK,cAAc,KAAO,YAAc,EAAK,cAAc,KAE3D,EAAK,UAAY,EAAK,YAAc,OAChC,GACA,EAAK,UACT,GAA4B,EAAK,WAC7B,EAAK,QACP,KAAM,IAAI,IACN,mEAEN,SAAK,UAAY,EAAK,MAAM,SAC5B,EAAK,gBAAkB,EAAK,MAAM,gBAClC,EAAK,YAAc,EAAK,MAAM,YAC9B,EAAK,gBAAkB,GACvB,EAAK,WAAa,GAClB,EAAK,UAAY,EAAK,MAAM,UAC5B,EAAK,aAAe,OAGtB,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,gBAGd,SAAc,GAIZ,KAAK,WAAa,EAClB,AAAI,KAAK,cAAgB,MACvB,MAAK,aAAa,UAAY,GAEhC,AAAI,KAAK,eAAiB,MACxB,MAAK,cAAc,UAAY,oCAInC,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,aAAa,aAAa,OAClC,KAAK,cAAc,eAGzB,EAAA,UAAA,WAAA,SAAW,GACT,GAAM,GAAa,EAAQ,OACrB,EAAiB,KAAK,MAAM,EAAa,GAC/C,KAAK,aAAa,WAAW,EAAQ,MAAM,EAAG,IAC9C,KAAK,cAAc,WAAW,EAAQ,MAAM,KAG9C,EAAA,UAAA,mBAAA,SAAmB,GACjB,GAAI,GACA,KAAK,aAAa,mBAAmB,GACzC,AAAM,MAAM,QAAQ,IAAgB,MAAM,QAAQ,EAAY,KAC5D,GAAc,CAAC,IAEjB,EAAc,EAEd,GAAI,GACA,EACA,EAiBJ,MAhBA,AAAI,MAAK,aACP,GAAa,EAAY,MAAM,IAC/B,EAAc,EAAY,GAI5B,EAAc,EACd,AAAI,KAAK,YAAc,SACrB,GAAY,EAAY,OAAS,IAAM,EACvC,EAAe,CAAC,IACX,AAAI,KAAK,WAAa,KAC3B,EAAe,CAAC,EAAa,EAAY,SAEzC,EAAe,CAAC,GAGd,KAAK,YACH,KAAK,WAAa,KACb,EAAa,OAAO,GAAY,OAAO,EAAW,SAEpD,CAAC,GAAa,OAAO,GAAY,OAAO,EAAW,SAErD,GAA+B,IAGxC,EAAA,UAAA,MAAA,SACI,EACA,GACF,GAAI,GACA,GAAU,KAAO,KAAO,EAAO,aAC/B,EACA,GAAU,KAAO,KAAO,EAAO,UACnC,AAAI,GAAU,MACZ,GAAS,IAEX,GAAM,GACF,GAAgB,EAAQ,EAAc,EAAW,KAAK,cAU1D,GATA,EAAS,EAAa,OACtB,EAAe,EAAa,aAC5B,EAAY,EAAa,UAEzB,AAAI,MAAM,QAAQ,IAChB,GAAgB,EAAuC,MAAM,GAC7D,EAAU,EAAuC,IAG9C,IAAgB,MAAQ,EAAa,SAAW,IACjD,GAAa,KACf,MAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,GAE7B,GAAM,GAAiD,GACjD,EAA+B,GACrC,GAAI,GAAgB,MAClB,GAAM,GAAY,EAAa,OAC/B,GAAI,EAAY,EAAI,EAClB,KAAM,IAAI,GACN,iIAIN,EAAO,aAAkB,EACzB,EAAiB,KAAI,MAArB,EAAyB,GACzB,GAAM,GAAc,EACI,IAAI,SAAA,GAAS,MAAA,IAAI,IAAU,CAAC,MAAO,EAAM,UACjE,KAAK,aAAa,UAAY,EAAW,MAAM,EAAG,EAAY,GAC9D,KAAK,cAAc,UAAY,EAAW,MAAM,EAAY,GAC5D,EAAgB,KAAI,MAApB,EAAwB,GAE1B,GAAI,GAAa,KACf,KAAM,IAAI,IACN,yEAKN,OADM,GAAmB,EAAiB,YAAc,IACnC,EAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAhB,GAAM,GAAM,EAAA,GACf,GAAI,YAAkB,MAAmB,EACvC,KAAM,IAAI,GACN,gHAKR,GAAI,GAEF,GAAM,GAAY,CAAC,GAAQ,OAAO,GAC5B,EAAgB,KAAK,UAAU,OAAO,GAUtC,EAAoB,KAAK,UAC/B,KAAK,UAAY,EACjB,GAAM,GACF,EAAA,UAAM,MAAK,KAAA,KAAC,EAA0C,GAC1D,YAAK,UAAY,EACV,MAEP,OAAO,GAAA,UAAM,MAAK,KAAA,KAAC,EAAQ,IAI/B,EAAA,UAAA,KAAA,SAAK,EAAyB,GAA9B,GAAA,GAAA,KACE,MAAO,GAAA,KAAK,WACV,GAAM,GAAe,EAAO,aAExB,EACA,EACJ,GAAI,GAAgB,KAClB,EAAI,EAAK,aAAa,KAAK,EAAQ,GACnC,EAAO,EAAK,cAAc,KAAK,EAAQ,QAEvC,GAAM,GAAe,EAAa,MAAM,EAAG,EAAa,OAAS,GAC3D,EAAgB,EAAa,MAAM,EAAa,OAAS,GAC/D,EAAI,EAAK,aAAa,KAClB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KACjD,EAAO,EAAK,cAAc,KACtB,EAAQ,OAAO,OAAO,EAAQ,CAAC,aAAc,KAGnD,GAAI,GACJ,AAAI,EAAK,aACP,CAAI,MAAM,QAAQ,IAChB,GAAS,EAAE,MAAM,GAAG,OAAQ,EAAkB,MAAM,KAGtD,EAAK,EAAe,GACpB,EAAQ,EAAkB,IAG5B,AAAI,EAAK,iBACP,GAAO,EAAA,QAAY,EAAgB,IAGrC,GAAI,GAcJ,MAbA,AAAI,GAAK,YAAc,SACrB,EAAS,GAAc,CAAC,EAAa,IAChC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,GAAI,EAAA,IAAQ,EAAa,IACrC,AAAI,EAAK,YAAc,MAC5B,EAAS,EAAA,IAAQ,EAAa,GACzB,AAAI,EAAK,WAAa,MAC3B,GAAS,CAAC,EAAa,IAIrB,EAAK,YACH,EAAK,WAAa,KACZ,EAAoB,OAAO,GAE9B,CAAC,GAAkB,OAAO,GAE5B,KAIX,EAAA,UAAA,YAAA,SAAY,GACV,KAAK,aAAa,cAClB,KAAK,cAAc,eAGrB,EAAA,UAAA,MAAA,SAAM,GAAN,GAAA,GAAA,KACE,GAAU,KAAK,aAAa,KAAM,WAChC,EAAK,aAAa,MAAM,KAE1B,GAAU,KAAK,cAAc,KAAM,WACjC,EAAK,cAAc,MAAM,KAE3B,KAAK,MAAQ,IAGf,EAAA,UAAA,YAAA,SAAY,EAAyB,GAEnC,AAAI,MAAM,QAAQ,IAChB,GAAO,EAAK,IAEd,GAAI,GAcJ,GAbA,AAAI,KAAK,gBACP,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,EAAM,GAEpB,EAAa,EAGf,AAAI,KAAK,WAAa,KACpB,EAAa,CAAC,KAAM,MAEpB,EAAa,KAGb,KAAK,aACP,GAAM,GAAS,KAAK,aAAa,OAC3B,EAAsB,EAAO,IAAI,SAAA,GAAS,MAAA,QAChD,MAAI,OAAM,QAAQ,GACT,EAAW,OAAO,GAAW,OAAO,GAEpC,CAAC,GAAY,OAAO,GAAW,OAAO,OAG/C,OAAO,IAIX,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,aAAa,iBAAiB,OACtC,KAAK,cAAc,mDAGzB,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,aAAa,oBAAoB,OACzC,KAAK,cAAc,sDAKzB,EAAA,UAAA,6BAAA,SAA6B,GAC3B,EAAA,UAAM,6BAA4B,KAAA,KAAC,GACnC,AAAI,KAAK,cAAgB,MACvB,KAAK,aAAa,6BAA6B,GAEjD,AAAI,KAAK,eAAiB,MACxB,KAAK,cAAc,6BAA6B,IAIpD,EAAA,UAAA,UAAA,WACE,GAAM,GAAmC,CACvC,UAAa,KAAK,WAGd,EAAa,EAAA,UAAM,UAAS,KAAA,MAClC,cAAO,OAAO,EAAQ,GACf,GAIF,EAAA,WAAP,SACI,EACA,GACF,GAAM,GACF,GAAY,EAAO,OAGvB,GAFA,MAAO,GAAO,MAEV,EAAO,cAAmB,KAC5B,KAAM,IAAI,IACN,4FAIN,GAAM,GAAkC,EACxC,SAAU,MAAW,EACd,GAAI,GAAI,IA7VV,EAAA,UAAY,gBA+VrB,GAjWmC,IAkWnC,EAAA,cAAc,cAAc,IC7kB5B,YAkE2B,GACzB,MAAO,IAAI,IAAW,eA6BJ,GAClB,MAAO,IAAI,IAAI,eAoBI,GACnB,MAAO,IAAI,IAAK,eAuBQ,GACxB,MAAO,IAAI,IAAU,eAwBD,GACpB,MAAO,IAAI,IAAM,eAmBK,GACtB,MAAO,IAAI,IAAQ,eA2BW,GAC9B,MAAO,IAAI,IAAgB,eAyBN,GACrB,MAAO,IAAI,IAAO,eAqBG,GACrB,MAAO,IAAI,IAAO,eAsCY,GAC9B,MAAO,IAAI,IAAgB,eAqBN,GACrB,MAAO,IAAI,IAAO,eAgCY,GAC9B,MAAO,IAAI,IAAgB,eAkCF,GACzB,MAAO,IAAI,IAAW,eA2BK,GAC3B,MAAO,IAAI,IAAa,eAeM,GAC9B,MAAO,IAAI,IAAgB,eAoCF,GACzB,MAAO,IAAI,IAAW,eAoCF,GACpB,MAAO,IAAI,IAAM,eAaK,GACtB,MAAO,IAAI,IAAQ,eAmCY,GAC/B,MAAO,IAAI,IAAiB,eAsBN,GACtB,MAAO,IAAI,IAAQ,eAiBQ,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,eA+BG,GACtB,MAAO,IAAI,IAAQ,eAcK,GACxB,MAAO,IAAI,IAAU,eAyBH,GAClB,MAAO,IAAI,IAAI,eAqBO,GACtB,MAAO,IAAI,IAAQ,eAuBO,GAC1B,MAAO,IAAI,IAAY,eAqBD,GACtB,MAAO,IAAI,IAAQ,eAqBG,GACtB,MAAO,IAAI,IAAQ,eAsBI,GACvB,MAAO,IAAI,IAAS,eAyBF,GAClB,MAAO,IAAI,IAAI,eA0BkB,GACjC,MAAO,IAAI,IAAmB,eAuBG,GACjC,MAAO,IAAI,IAAmB,eA2BF,GAC5B,MAAO,IAAI,IAAc,eAgBM,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eA0BO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAwBO,GAC/B,MAAO,IAAI,IAAiB,eAEJ,GACxB,MAAO,IAAiB,eAIG,GAC3B,MAAO,IAAiB,eAYa,GACrC,MAAO,IAAI,IAAuB,eAiBG,GACrC,MAAO,IAAI,IAAuB,eAYD,GACjC,MAAO,IAAI,IAAmB,eAiBG,GACjC,MAAO,IAAI,IAAmB,eAYH,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eAwBG,GAC3B,MAAO,IAAI,IAAa,eA4BN,GAClB,MAAO,IAAI,IAAI,eAgDO,GACtB,MAAO,IAAI,IAAQ,eA0BA,GACnB,MAAO,IAAI,IAAK,eAgDO,GACvB,MAAO,IAAI,IAAS,eA2BI,GACxB,MAAO,IAAI,IAAU,eAgDO,GAC5B,MAAO,IAAI,IAAc,eA8BA,GACzB,MAAO,IAAI,IAAW,eAoCO,GAC7B,MAAO,IAAI,IAAe,eA+DR,GAClB,MAAO,IAAI,IAAI,eAUe,GAC9B,MAAO,IAAI,IAAgB,eAMC,GAC5B,MAAO,IAAI,IAAc,eAiDK,GAC9B,MAAO,IAAI,IAAgB,GAItB,GAAM,IAAkB,GAClB,GAAkB,GAClB,GAAY,GACZ,GAAY,eA2BK,GAC5B,MAAO,IAAI,IAAc,eA2BK,GAC9B,MAAO,IAAI,IAAgB,eAkCA,GAC3B,MAAO,IAAI,IAAa,eA0BF,GACtB,MAAO,IAAI,IAAQ,glCClnDU,EAAe,GAC5C,MAAO,IAAuB,EAAO,eAoBJ,EAAe,GAChD,MAAO,IAA2B,EAAO,eAuBvC,EAAe,GACjB,MAAO,IAAkC,EAAO,eAqBd,EAAe,GACjD,MAAO,IAA4B,EAAO,eAcJ,EAAe,GACrD,MAAO,IAAgC,EAAO,eAsCtB,EAAe,GACvC,MAAO,IAAkB,EAAO,eAsCX,EAAe,GACpC,MAAO,IAAe,EAAO,eAwBC,EAAe,GAC7C,MAAO,IAAuB,EAAO,eAuBL,EAAe,GAC/C,MAAO,IAAyB,EAAO,eAsBrC,EAAe,GACjB,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAG9B,EAAe,GAClC,MAAO,IAAmC,EAAO,eAqBlB,EAAe,GAC9C,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,eAGpB,EAAe,GACjC,MAAO,IAAwB,EAAO,kSC1TxC,yCCAA,YAqBqB,GACnB,MAAO,IAAI,IAAK,eAYC,GACjB,MAAO,IAAgB,eAYN,GACjB,MAAO,IAAgB,+CChDzB,mBAkBuC,EAAA,EAAA,GAAvC,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KAEE,SAAA,MAAqB,OAErB,SAAA,UAAA,SAAA,SAAS,GACP,GAAI,CAAE,aAAiB,KACrB,KAAM,IAAI,OAAM,yDAElB,KAAK,MAAQ,GAEjB,GAVuC,IAsEvC,YAAc,EAAiB,GAC7B,MAAO,GAAU,EAGnB,YAAiB,EAAiB,GAChC,MAAO,GAAU,qBAOgB,EAAA,EAAA,GAcjC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KAIP,GAHA,AAAI,GAAQ,MACV,GAAO,IAEL,EAAK,mBACP,KAAM,IAAI,IACN,sEAGN,SAAK,QAAU,EAAK,SAAW,WAC/B,EAAK,SAAW,KAAK,IAAI,EAAK,UAAY,GAC1C,EAAK,SAAW,EAAK,UAAY,EACjC,EAAK,QAAU,EAAK,SAAW,EAC/B,EAAK,KAAO,EAAK,MAAQ,OACzB,EAAK,SAAW,EAAK,SAErB,AAAI,CAAC,OAAQ,MAAO,OAAO,QAAQ,EAAK,QAAU,IAChD,SAAQ,KACJ,uBAAuB,EAAK,KAAI,8CAEpC,EAAK,KAAO,QAGd,AAAI,EAAK,OAAS,MAChB,EAAK,YAAc,GACd,AAAI,EAAK,OAAS,OAInB,EAAK,QAAQ,QAAQ,SAAW,GAHpC,EAAK,YAAc,GAMjB,EAAK,YAAc,GAIvB,AAAI,EAAK,cAAgB,IACvB,GAAK,UAAY,MAIf,SAAA,UAAA,aAAN,SAAmB,sEACjB,YAAK,KAAO,EACZ,KAAK,aAAe,EACpB,AAAI,KAAK,UAAY,KACnB,KAAK,KAAO,KAAK,SAEjB,KAAK,KAAO,KAAK,cAAgB,GAAO,SAAW,mBAIjD,EAAA,UAAA,WAAN,SAAiB,EAAe,mGAC9B,MAAA,CAAA,EAAM,GAAqB,WAE3B,MAFA,GAAA,OACM,EAAU,KAAK,gBAAgB,GACjC,GAAW,KACb,CAAA,GAGF,CAAI,KAAK,YAAY,EAAU,KAAK,SAAU,KAAK,MACjD,MAAK,KAAO,EACZ,KAAK,KAAO,GAGZ,MAAK,OACL,AAAI,KAAK,MAAQ,KAAK,UACpB,MAAK,aAAe,EACpB,KAAK,MAAM,aAAe,gBAM1B,EAAA,UAAA,WAAN,SAAiB,sEACf,MAAI,MAAK,aAAe,GAAK,KAAK,SAChC,QAAQ,IAAI,SAAS,KAAK,aAAY,8BAIlC,EAAA,UAAA,gBAAR,SAAwB,GACtB,AAAI,GAAQ,MACV,GAAO,IAET,GAAM,GAAe,EAAK,KAAK,SAC/B,MAAI,IAAgB,MAClB,QAAQ,KACJ,4BAA4B,KAAK,QAAO,sBACxC,2BAA0B,OAAO,KAAK,KAErC,GAEX,GAzGmC,gBAsJL,GAC5B,MAAO,IAAI,IAAc,MAGd,IAAY,CAAC,cAAa,i8DC9PvC,AA8BA,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,WAAA,GAAA,aACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,UACA,EAAA,EAAA,UAAA,GAAA,YACA,EAAA,EAAA,aAAA,GAAA,eACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,IAAA,UACA,EAAA,EAAA,SAAA,IAAA,WACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,UAAA,IAAA,YACA,EAAA,EAAA,YAAA,IAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,iBAAA,KAAA,mBACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,YAAA,KAAA,cACA,EAAA,EAAA,aAAA,KAAA,eACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,cAAA,KAAA,gBACA,EAAA,EAAA,gBAAA,KAAA,oBA7BU,IAAA,IAAQ,QAyRH,IAAjB,AAAA,UAAiB,GAEf,GAAY,GAAZ,AAAA,UAAY,GAAyB,EAAA,EAAA,OAAA,GAAA,SAAc,EAAA,EAAA,GAAA,GAAA,KAAU,EAAA,EAAA,GAAA,GAAA,OAAjD,EAAA,EAAA,yBAAA,GAAA,wBAAuB,OAFpB,IAAA,IAAQ,KCtTzB,AAmBA,GAAM,IAAwC,eA2BnB,EAAc,GACvC,GAAM,GAAqB,CACzB,SAAU,EACV,SAAU,SACV,OAAQ,GACR,MAAO,GACP,eAAgB,GAGlB,GAAW,GAAQ,cAUW,GAC9B,MAAO,IAAW,eAUS,GAC3B,MAAO,IAAW,GC9EpB,WAwBI,EAAmB,EAAY,EAC/B,GACF,GAAM,GAAa,EAAK,YAAY,GACpC,GAAI,GAAc,EAAW,kBAAoB,QAC/C,GAAM,GAAQ,EAAW,gBACnB,EAAM,EAAW,gBAAkB,EACrC,OACC,EAAW,gBAAkB,OAAY,EAAQ,EACR,EAAW,cACzD,GAAI,EAAW,OAAS,SACtB,MAAO,IACH,EAAK,WAAW,EAAW,iBAAkB,EAAW,GAE9D,GAAI,EAAW,OAAS,WACtB,GAAM,GAAS,EAAK,WAAW,MAAM,EAAO,GAE5C,MAAO,GAAO,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAEvD,GAAM,GACF,GAAU,EAAK,WAAW,MAAM,GAAO,GAAI,EAAW,GACpD,EAAO,EAAO,WACpB,MAAO,GAAW,OAAS,SACvB,EAAK,GACL,EAAA,KAAK,cAAc,EAAO,MAAO,GAEvC,GAAM,GAAY,EAAK,WAAW,GAClC,MAAO,IAAa,EAAU,kBAU5B,EAAc,EACd,GACI,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAY,EAAQ,kBAAkB,KAAK,SAAA,GAC/C,MAAO,CAAC,CAAC,EAAW,GAAyB,EAAU,MAGzD,MAAO,KAAc,OACjB,EAAW,GAAyB,EAAU,IAAY,GAC1D,mBASF,EAAc,EACd,GACF,MAAO,GAAW,GAAyB,EAAM,EAAQ,+BAUvD,EAAmB,GACf,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEjB,MAAO,CACL,GAAyB,EAAU,GAAW,EAAQ,kBACtD,GAIJ,YAAkC,EAAc,GAC9C,MAAO,AAAE,GAAe,EAAI,IAAI,EAAc,cAGlB,GAC5B,GAAM,GAAQ,EAAK,MAAM,KACzB,GAAI,EAAM,SAAW,EACnB,MAAO,CAAC,EAAM,GAGhB,GAAM,GAAW,EAAM,GACvB,MAAO,CAAC,EAAU,OAAO,EAAM,EAAM,OAAS,iBAW5C,EAAY,EACZ,GACF,GAAI,GAAM,EAAc,MAAO,EAAM,EAAW,GAChD,GAAI,IAAQ,YAEV,EAAM,EAAc,mBAAoB,EAAM,EAAW,GAIzD,OAHM,GAEF,CAAC,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,GAAI,CAAC,EAAG,IACxB,EAAI,EAAG,EAAI,EAAG,IACrB,EAAgB,GAAG,GAAM,EAAiB,EAAI,GAC9C,EAAgB,GAAG,GAAM,EAAiB,EAAI,EAAI,GAEpD,MAAO,GAET,MAAO,eAYmB,GAC1B,MAAO,GAAO,KAAO,EAAS,EAAA,MAAM,GCrJtC,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,mCCnLxE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,UAC7D,CAAC,OAAU,iBAAkB,KAAQ,eAAgB,KAAQ,YAGjE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,OACV,KAAQ,aACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,eACV,KAAQ,eACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,MACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC7dxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAElD,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,aAE/D,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,QAC1D,CAAC,OAAU,mBAAoB,KAAQ,iBAAkB,KAAQ,QACjE,CACE,OAAU,2BACV,KAAQ,yBACR,KAAQ,QAEV,CAAC,OAAU,oBAAqB,KAAQ,OAAQ,KAAQ,YAG5D,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CACR,OAAU,QACV,KAAQ,QACR,KAAQ,QACR,aAAgB,MAGpB,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAAU,CACrD,OAAU,wBACV,KAAQ,sBACR,KAAQ,QACR,aAAgB,MAItB,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,YAG3C,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,YAE3D,CACE,SAAY,cACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,KACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,QACxD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,UAG5D,CACE,SAAY,iBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,OAAQ,KAAQ,YAEjD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,QAC3C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAG/C,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YACxC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,WAE9C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,uBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,eAAgB,KAAQ,cAAe,KAAQ,WAG9D,CACE,SAAY,kBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,SAC7C,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,mBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,SAC5D,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAGhE,CACE,SAAY,oBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAE/C,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,WAEnE,CACE,SAAY,qBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MACI,CAAC,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,wCCnWrE,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,oBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAAa,CAC7D,OAAU,yBACV,KAAQ,sBACR,KAAQ,QAEV,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,aAAc,KAAQ,YAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAC/C,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,OAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,WACV,KAAQ,WACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CAAC,OAAU,gBAAiB,KAAQ,gBAAiB,KAAQ,QAAS,CACpE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,eACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,mBACV,KAAQ,gBACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,QAItB,CACE,SAAY,sBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,aAE9C,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAC7C,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,kBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,wBACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,oBACV,KAAQ,mBACR,KAAQ,WACR,aAAgB,IAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,6BACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAClD,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAClE,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,WACR,aAAgB,CAAC,EAAG,EAAG,EAAG,IAE5B,CACE,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,MAItB,CACE,SAAY,SACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,UAAW,CACtD,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,QAElB,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,cAGzD,CACE,SAAY,aACZ,SAAY,cACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,YACjD,CAAC,OAAU,QAAS,KAAQ,YAAa,KAAQ,YACjD,CAAC,OAAU,UAAW,KAAQ,MAAO,KAAQ,yCCpVnD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,WAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,WAEtC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,SAAU,aAAgB,GAClE,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,SAAU,aAAgB,IAErE,MAAS,CACP,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,WACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAEzD,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,IAEjE,MAAS,CAAC,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAExD,CACE,SAAY,kBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,OACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,SACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAAW,CACpD,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,IAElB,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,MAGnE,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,WAErD,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,WAE7C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,UAC3C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,UAC7C,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,SACzC,CAAC,OAAU,eAAgB,KAAQ,eAAgB,KAAQ,wCC3KjE,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,YAGnD,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,WAEjD,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,IAAO,CACvE,OAAU,cACV,KAAQ,YACR,KAAQ,QACR,aAAgB,IAElB,CACE,OAAU,yBACV,KAAQ,qBACR,KAAQ,UAId,CACE,SAAY,sBACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,iBAAkB,KAAQ,UAC/C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,YAGjD,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,WAE5C,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCChGtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAE3D,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAGtC,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,yCCvC3C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,yBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CACE,SAAY,cACZ,SAAY,QACZ,MAAS,CACP,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,SAC7C,CAAC,OAAU,QAAS,KAAQ,QAAS,KAAQ,WAGjD,CAAC,SAAY,QAAS,SAAY,SAAU,CAC1C,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,YACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,WACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,SACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,IAAK,KAAQ,aAEzD,CACE,SAAY,QACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAEvC,MAAS,CACP,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,UAAW,CAC1D,OAAU,UACV,KAAQ,SACR,KAAQ,SACR,aAAgB,IAElB,CACE,OAAU,YACV,KAAQ,YACR,KAAQ,SACR,aAAgB,KAItB,CAAC,SAAY,OAAQ,SAAY,QAAS,OAAU,IAAK,CACvD,SAAY,eACZ,SAAY,QACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,0BACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,UACzC,CAAC,OAAU,MAAO,KAAQ,MAAO,KAAQ,yCC3G/C,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,wBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CACP,CAAC,OAAU,gBAAiB,KAAQ,eAAgB,KAAQ,QAC5D,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,QACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,aAE3C,MAAS,CACP,CAAC,OAAU,SAAU,KAAQ,SAAU,KAAQ,UAAW,CACxD,OAAU,sBACV,KAAQ,qBACR,KAAQ,yCCvDhB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,QACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,UACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,eACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,OACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,aACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,WACZ,SAAY,UACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,YAAa,KAAQ,UAC1C,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCC9ItB,AAmBO,GAAM,IAAmB,CAC9B,CACE,SAAY,eACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,IAAK,EAAG,KAAQ,OAAQ,KAAQ,YAE/C,MAAS,CACP,CAAC,OAAU,WAAY,KAAQ,UAAW,KAAQ,UAAW,CAC3D,OAAU,YACV,KAAQ,WACR,KAAQ,WACR,aAAgB,IAElB,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,SACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,cACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,gBACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CACE,OAAU,QACV,KAAQ,aACR,KAAQ,OACR,aAAgB,IAElB,CAAC,OAAU,IAAK,KAAQ,QAAS,KAAQ,QAAS,aAAgB,MAGtE,CACE,SAAY,YACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CACR,OAAU,IACV,KAAQ,QACR,KAAQ,QACR,aAAgB,mCCtItB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,iBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,mBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,UACtC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,UACrC,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,WAE3C,MAAS,CACP,CACE,OAAU,UACV,KAAQ,UACR,KAAQ,SACR,aAAgB,MAElB,CACE,OAAU,cACV,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,MACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,eACV,KAAQ,SACR,KAAQ,SACR,aAAgB,GAElB,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GACrE,CACE,OAAU,QACV,KAAQ,QACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,OACV,KAAQ,OACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,aACZ,SAAY,gBACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,gBACZ,SAAY,gBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCCjJtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,MACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,OACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,aAEvC,MAAS,CAAC,CAAC,OAAU,YAAa,KAAQ,WAAY,KAAQ,UAEhE,CACE,SAAY,SACZ,SAAY,YACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CACP,CAAC,OAAU,YAAa,KAAQ,YAAa,KAAQ,QACrD,CAAC,OAAU,UAAW,KAAQ,UAAW,KAAQ,uCC1GvD,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,GAAI,KAAQ,UAAW,KAAQ,WACnD,CAAC,MAAS,GAAI,KAAQ,OAAQ,KAAQ,WAExC,MACI,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAEtE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,WAClD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,WAEvC,MAAS,CAAC,CAAC,OAAU,IAAK,KAAQ,IAAK,KAAQ,SAAU,aAAgB,KAG3E,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,WAE1C,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,MAItB,CACE,SAAY,UACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,OAAQ,aAAgB,MAGjE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,eACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,YACtC,CAAC,MAAS,EAAG,KAAQ,MAAO,KAAQ,YACpC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CACP,CACE,OAAU,aACV,KAAQ,YACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,WACV,KAAQ,UACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,cACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,gBACV,KAAQ,eACR,KAAQ,SACR,aAAgB,GAElB,CACE,OAAU,mBACV,KAAQ,iBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,IAAO,EAAG,KAAQ,UAAW,KAAQ,YAEpD,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGzE,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,WAEzC,MAAS,CACP,CAAC,OAAU,OAAQ,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAAI,CACvE,OAAU,MACV,KAAQ,MACR,KAAQ,SACR,aAAgB,EAChB,aAAgB,MAItB,CACE,SAAY,OACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,cAGzC,CACE,SAAY,QACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,GAC/D,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,YACV,KAAQ,kBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,SACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,kBAAmB,KAAQ,YAChD,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,SAAU,aAAgB,KAGnE,CACE,SAAY,YACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,UACxC,CAAC,MAAS,EAAG,KAAQ,SAAU,KAAQ,UACvC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,WACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAG5C,CACE,SAAY,gBACZ,SAAY,aACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,gBAAiB,KAAQ,UAC9C,CAAC,MAAS,EAAG,KAAQ,cAAe,KAAQ,YAC5C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,UAC7C,CAAC,MAAS,EAAG,KAAQ,eAAgB,KAAQ,WAE/C,MAAS,CAAC,CACR,OAAU,mBACV,KAAQ,kBACR,KAAQ,OACR,aAAgB,GAChB,aAAgB,mCC1NtB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,MACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CAAC,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,YAE/C,CACE,SAAY,OACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,MAItB,CACE,SAAY,QACZ,SAAY,WACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAAW,CAC3C,MAAS,EACT,KAAQ,aACR,KAAQ,SACR,aAAgB,mCChDxB,AAiBO,GAAM,IAAmB,CAC9B,CACE,SAAY,OACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CACE,OAAU,OACV,KAAQ,SACR,KAAQ,QACR,aAAgB,IAElB,CAAC,OAAU,OAAQ,KAAQ,QAAS,KAAQ,WAGhD,CACE,SAAY,aACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,OAAQ,KAAQ,YAGzC,CACE,SAAY,MACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,aAE1C,MAAS,CAAC,CACR,OAAU,iBACV,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAGpB,CACE,SAAY,QACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,UAAW,KAAQ,YAAa,CACnD,MAAS,EACT,KAAQ,gBACR,KAAQ,SACR,aAAgB,KAItB,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,UACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CAAC,CACR,OAAU,OACV,iBAAoB,eACpB,KAAQ,OACR,KAAQ,cAGZ,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,WAAY,KAAQ,cAG7C,CACE,SAAY,iBACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,aAAc,KAAQ,YAC3C,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,cAG1C,CACE,SAAY,eACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,WAEpC,MAAS,CACP,CAAC,OAAU,aAAc,KAAQ,YAAa,KAAQ,UACtD,CAAC,OAAU,cAAe,KAAQ,aAAc,KAAQ,YAG5D,CACE,SAAY,cACZ,SAAY,iBACZ,OAAU,CACR,CAAC,MAAS,EAAG,KAAQ,IAAK,KAAQ,UAClC,CAAC,MAAS,EAAG,KAAQ,QAAS,KAAQ,aAExC,MAAS,iCChIb,AAyCA,GAAA,IAAA,WAWE,aACE,GAAM,GAAM,CACV,GAAY,GAAW,GAAS,GAAa,GAAU,GACvD,GAAY,GAAS,GAAO,GAAO,GAAU,GAAe,GAC5D,GAAW,GAAU,IAEjB,EAA0B,GAAG,OAAM,MAAT,GAAa,EAAI,IAAI,SAAA,GAAM,MAAA,GAAG,QAE9D,KAAK,UAAY,EAAY,OACzB,SAAC,EAAK,GACJ,SAAI,EAAO,UAAY,EAChB,GAET,IAlBN,cAAA,eAAkB,EAAA,WAAQ,KAA1B,WACE,MAAO,MAAK,WAAc,MAAK,UAAY,GAAI,wCAsBjD,EAAA,UAAA,eAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAU,EAAM,KAChB,EAAuB,GACvB,EAAkB,GAClB,EAAoB,GACpB,EAAQ,EAAQ,OAA8B,SAAC,EAAK,GACxD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,GAAG,WAAW,eACrB,EAAa,KAAK,EAAI,EAAK,OACtB,AAAI,EAAK,KAAO,QACrB,EAAQ,KAAK,EAAI,EAAK,OACjB,AAAI,GAAK,OAAS,MAAQ,EAAK,MAAM,SAAW,IACrD,EAAU,KAAK,EAAI,EAAK,OAEnB,GACN,IAEC,EAAiB,GACf,EAAkB,GACpB,EAA8C,GAC9C,EAA+C,GACnD,AAAI,GAAa,MACf,GAAqB,KAAK,oBAAoB,EAAU,QACxD,EAAsB,KAAK,oBAAoB,EAAU,UAE3D,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAMlC,AAAI,OAAO,KAAK,GAAqB,SAAW,EAC9C,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,AAAI,EAAK,SAAS,SAAW,GAC3B,EAAQ,KAAK,KAIjB,OAAO,KAAK,GAAqB,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,aAAe,EAAoB,GACxC,EAAQ,KAAK,MAKnB,AAAI,OAAO,KAAK,GAAoB,OAAS,EAC3C,OAAO,KAAK,GAAoB,QAAQ,SAAA,GAC/B,GAAA,GAAA,GAAA,GAAA,GACD,EAAO,EAAM,GACnB,AAAI,GACF,GAAK,aAAe,EAAmB,GACvC,EAAO,KAAK,MAIhB,EAAS,EAGX,GAAI,GAAY,GAChB,AAAI,EAAM,SAAW,MAAQ,EAAM,QAAQ,UAAY,MACrD,GAAY,EAAM,QAAQ,SAAS,OAAO,SAAC,EAAW,GACpD,SAAU,EAAK,UAAU,MAAQ,EAAK,YAAY,GAC3C,GACN,KAGL,GAAM,GACF,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,EAAE,UAAS,GAExE,MAAI,GAAU,OAAS,GACrB,GAAO,UAAY,GAGd,GAGD,EAAA,UAAA,oBAAR,SAA4B,GAC1B,MAAO,QAAO,KAAK,GAAW,IACzB,OAAgC,SAAC,EAAM,GACtC,SAAK,EAAQ,GAAM,MAAQ,EACpB,GACN,KAGD,EAAA,UAAA,QAAR,SAAgB,GAGd,GAAM,GACF,GAAgB,EAAK,KAAO,KAAK,UAAU,EAAK,KAAO,GAC3D,AAAI,EAAK,MAAQ,MACf,GAAK,KAAO,IAGd,GAAM,GAAgB,CACpB,KAAM,EAAK,KACX,GAAI,EAAK,GACT,SAAU,EAAO,SACjB,WACK,GAAK,OACL,IAAI,IAAI,SAAA,GAAS,MAAA,GAAM,WAAW,KAAO,EAAM,OAAO,GAAK,IAChE,OAAQ,GACR,SAAU,GACV,YAAa,GACb,WAAY,GACZ,SAAU,EAAK,MAGjB,MAAI,GAAO,QAAU,MACnB,GAAQ,YACJ,EAAO,OAAO,OACV,SAAC,EAAK,GACJ,SAAI,EAAM,MAAQ,CAChB,KAAM,EAAM,KACZ,gBAAiB,EAAM,MACvB,cAAe,EAAM,KAEhB,GAET,KAEV,AAAI,EAAO,OAAS,MAClB,GAAQ,WACJ,EAAO,MAAM,OAAoC,SAAC,EAAK,GACrD,GAAM,GAAO,EAAM,KACf,EAAQ,OACZ,OAAQ,EAAM,UACP,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cAEnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAChB,EAAM,cAAgB,GAC3B,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,WACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,SACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,QACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,UACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,OACH,EAAQ,GACJ,EAAK,KAAM,EAAM,OAAQ,EAAM,cACnC,AAAI,IAAU,QAAa,CAAC,CAAC,EAAM,kBACjC,GAAQ,GACJ,EAAK,KAAM,EAAM,iBACjB,EAAM,eAEZ,UACG,aACA,UACH,cAEA,KAAM,IAAI,OACN,2BAA2B,EAAM,KAAI,YAAY,EAAK,IAE9D,SAAI,EAAM,MAAQ,CAAC,MAAK,EAAE,KAAI,GACvB,GACN,KAEF,GAID,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAU,EAAY,QACtB,EAAuB,GACvB,EAAkB,GACpB,EAA+B,GACnC,AAAI,GAAW,MACb,GAAQ,EAAQ,OAA8B,SAAC,EAAK,GAClD,SAAI,EAAK,MAAQ,EAAK,QAAQ,GAC9B,AAAI,EAAK,KAAO,SACd,EAAQ,KAAK,EAAI,EAAK,OAEjB,GACN,KAEL,GAAM,GAAiB,GACjB,EAAkB,GAExB,EAAY,UAAU,SAAS,QAAQ,SAAA,GAC9B,GAAA,GAAA,GAAA,EAAA,MAAA,GACD,EAAa,CACjB,KAAM,EACN,GAAI,cACJ,OAAQ,GACR,WAAY,GACZ,SAAU,QACV,YAAa,GACb,WAAY,CAAC,MAAO,CAAC,MAAO,GAAgB,EAAI,MAAO,KAAM,UAC7D,SAAU,IAEZ,EAAK,aAAe,EAAI,KACxB,EAAO,KAAK,GACZ,EAAM,GAAY,IAGpB,GAAM,GAAW,OAAO,KAAK,GAC7B,EAAS,QAAQ,SAAA,GACf,GAAM,GAAO,EAAM,GACnB,EAAK,WAAW,QAAQ,SAAA,GACf,GAAA,GAAA,GAAA,GAAA,GACP,EAAK,OAAO,KAAK,EAAM,IACvB,EAAM,GAAU,SAAS,KAAK,OAIlC,GAAM,GAAgB,EAAY,IAElC,EAAY,UAAU,UAAU,QAAQ,SAAA,GAChC,GAAA,GAAA,GAAA,EAAA,EAAA,OAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAO,EAAM,GACnB,AAAI,GAAQ,MACV,GAAK,cAAgB,EACrB,EAAQ,KAAK,MAIjB,GAAM,GAAY,KAAK,mBAAmB,GAC1C,MAAO,CAAC,MAAK,EAAE,OAAM,EAAE,QAAO,EAAE,QAAO,EAAE,aAAY,EAAE,UAAS,IAG1D,EAAA,UAAA,mBAAR,SAA2B,GAA3B,GAAA,GAAA,KAEE,MAAO,CACL,WAAY,EAAY,UAAU,KAClC,OAAQ,EAAY,UAAU,SAAS,OACnC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,GACjC,GAET,IACJ,QAAS,EAAY,UAAU,UAAU,OACrC,SAAC,EAAK,GACJ,SAAI,EAAI,MAAQ,EAAK,mBAAmB,EAAK,EAAY,KAClD,GAET,MAIA,EAAA,UAAA,mBAAR,SACI,EACA,GACF,GAAI,GAAO,EAAI,KACf,MAAI,IAAW,MACb,GAAO,EAAQ,IAEV,CAAC,KAAI,EAAE,MAAO,EAAI,OAE7B,iBAE6B,GAC3B,GAAM,GAAS,EAAA,MAAM,OACrB,GAAI,MAAO,GAAO,MAAS,YACzB,MAAO,GAAO,KAAK,GACd,GAAI,MAAO,SAAW,YAC3B,MAAO,IAAI,QAAO,EAAM,UAAU,WAElC,KAAM,IAAI,OACN,gGAKyB,EAAc,GAC7C,GAAM,GACF,MAAM,QAAQ,GAAK,OAAO,aAAa,MAAM,KAAM,GAAK,GAAa,GACzE,MAAO,GAAW,EAAQ,EAAM,0BAI9B,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,KACJ,GAAiB,EAAM,EAAG,GAE5B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAO,GAAQ,EAAM,EAAI,cAIvB,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,IAAS,GACvB,EACF,EAAM,GAAQ,KAAO,EAAM,EAAQ,EAAM,GAAQ,KAAO,EAAM,EAAO,EACzE,MAAQ,OAAO,IAAU,SAAY,EAAQ,SAAS,EAAO,gBAG/B,GAC9B,AAAI,MAAQ,IAAW,UAErB,GAAQ,GAAoB,IAE9B,OAAQ,OACD,IAAoB,SACvB,MAAO,cACJ,IAAoB,aACpB,IAAoB,aACpB,IAAoB,YACpB,IAAoB,SACvB,MAAO,YACJ,IAAoB,QACvB,MAAO,WACJ,IAAoB,UACvB,MAAO,cACJ,IAAoB,UACvB,MAAO,iBAIP,MAAO,mBAKT,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,EAAM,KAAK,KAEb,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,KACV,GAAgB,EAAM,MAExB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,KAC7B,EAAM,KAAK,KAAK,IAAI,SAAA,GAAK,MAAA,IAAgB,KAE3C,cAG6B,GAEpC,MAAI,GAAM,YACR,OAEE,EAAM,KAAO,KACR,EAAM,IAAI,IACb,SAAA,GACI,MAAC,OAAO,GAAI,MAAS,SAAY,EAAI,KAAO,SAAS,EAAI,KAAM,MAElE,eAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MACV,GAAsB,EAAM,OAE9B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,GACO,IAAM,KAAK,GAAK,EAAM,KAAK,EAAE,OAAS,EAAM,KAAK,EACX,EAAM,KAAK,IAClD,IACH,IAAI,SAAA,GAAK,MAAC,OAAO,IAAM,SAAY,EAAI,SAAS,EAAG,MAEnD,cAIL,EAA+C,EAAc,EAC7D,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAAE,IAAI,SAAC,GACvB,MAAO,IAAiB,EAAG,KAGxB,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,MAC7B,EAAM,KAAK,MAAM,IAAI,SAAC,GAC3B,MAAO,IAAsB,KAG1B,cAIL,EAA+C,EAC/C,GACF,GAAM,GAAQ,EAAM,GACpB,MAAI,IAAS,EAAM,MAAQ,EAAM,KAAK,EAC7B,EAAM,KAAK,EAEb,ECtkBT,AA4BA,GAAA,IAAA,WAGE,WACY,EAAoB,EACpB,GAFZ,GAAA,GAAA,KACY,KAAA,KAAA,EAAoB,KAAA,UAAA,EACpB,KAAA,QAAA,EAJI,KAAA,OAAmB,GACnB,KAAA,MAAoC,GAIlD,KAAK,OAAS,EAAK,WAAW,IAAI,SAAA,GAAQ,MAAA,GAAK,SAAS,KACxD,AAAI,EAAK,UAAY,MACnB,MAAK,MAAQ,OAAO,KAAK,EAAK,UACZ,OAAO,SAAC,EAAmC,GAC1C,SAAM,GAAO,EAAK,QAAQ,GACnB,GACN,KAQhB,SAAA,UAAA,SAAR,SAAiB,GACf,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,UAOtC,EAAA,UAAA,QAAR,SAAgB,EAAc,GAC5B,GAAM,GAAQ,KAAK,KAAK,SAAS,GACjC,GAAI,EAAM,QAAU,KAClB,MAAO,IAAU,EAAM,KAAK,UAAW,KAAK,SAE9C,GAAI,EAAM,GAAK,MAAQ,EAAM,GAAK,KAChC,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAe,KAAK,KAAK,SAAU,EAAM,GAElD,GAAI,EAAM,GAAK,KACb,MAAO,IAAa,KAAK,KAAK,SAAU,EAAM,GAEhD,GAAI,EAAM,OAAS,KACjB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,MAAQ,KAChB,MAAO,IAAc,KAAK,KAAK,SAAU,EAAM,GAEjD,GAAI,EAAM,MAAQ,MAChB,GAAI,EAAM,KAAK,GAAK,MAAQ,EAAM,KAAK,GAAK,KAC1C,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,OAAS,KACtB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,GAAK,KAClB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAEhC,GAAI,EAAM,KAAK,MAAQ,KACrB,MAAO,IACH,KAAK,KAAK,SAAU,EAAM,GAIlC,MAAO,IAEX,KCtGA,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,cACA,YACA,MACH,MAAO,CAAC,EAAA,IACH,EAAc,IAAK,EAAM,EAAW,GACrC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,UAAW,EAAM,EAAW,SAE3C,eACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,cACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,oBACH,MAAO,CAAC,EAAA,kBACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC7F9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACA,aACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,OAAQ,EAAM,EAAW,SACxC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SACrC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SACrC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SACrC,YACA,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,eAAgB,EAAM,EAAW,GAC/C,EAAc,eAAgB,EAAM,EAAW,SAEhD,QACH,MAAO,CAAC,EAAA,MACJ,GAAU,EAAK,WAAW,GAAI,EAAW,SAC1C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SACxC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SACzC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAE5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC5K9C,YAyBI,EAAkB,EAAkB,GAAA,AAAA,IAAA,QAAA,GAAA,IACtC,EAAA,KAAK,OACD,GAA8B,EAAQ,GACtC,WAAM,MAAA,GAAqB,YAAW,EAAM,QAAQ,EAAM,6BAGlB,EAAc,GAC1D,GAAI,EAAG,SAAW,EAAG,OACnB,MAAO,GAET,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAI,EAAG,KAAO,IAAM,EAAG,KAAO,IAAM,EAAG,KAAO,EAAG,GAC/C,MAAO,GAGX,MAAO,GCzCT,AA+BA,GAAA,IAAA,WAIE,WACa,EAAuB,EAAyB,EACjD,EAAiC,EAChC,EAA+B,GAF/B,KAAA,KAAA,EAAuB,KAAA,MAAA,EAAyB,KAAA,QAAA,EACjD,KAAA,aAAA,EAAiC,KAAA,uBAAA,EAChC,KAAA,YAAA,EAA+B,KAAA,eAAA,EANpC,KAAA,QAA6B,GAC7B,KAAA,QAAU,GAMhB,KAAK,SAAW,EAAA,OAAO,GACvB,EAAA,KAAK,KAAK,UAGZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,yCAMd,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,OAAO,MAChD,EAAO,OAAO,YAGlB,KAAK,QAAU,GACf,KAAK,QAAU,GACf,KAAK,SAAS,WAGhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAOtB,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,GAAS,KAAK,OAC7B,KAAM,IAAI,OAAM,4BAA4B,EAAK,wBAC7C,KAAK,QAGX,GAAM,GAAkB,KAAK,QAAQ,GACrC,GAAI,EAAgB,QAClB,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0BACpB,EAAK,wGAIf,MAAI,MAAK,gBACP,GAAgB,QAAU,IAG5B,EAAgB,KAAO,GAChB,EAAgB,QAMzB,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GAAS,MAAA,GAAK,KAAK,MAQxC,EAAA,UAAA,MAAA,SAAM,EAAe,GACnB,GAAI,KAAK,QACP,KAAM,IAAI,OAAM,eAAe,KAAK,KAAI,6BAG1C,GAAI,EAAQ,GAAK,CAAC,KAAK,aAAe,GAAS,KAAK,QAClD,KAAM,IAAI,OAAM,2BACZ,EAAK,8CAA8C,KAAK,SAG9D,GAAM,GAAI,KAAK,QAAQ,IAAU,GAEjC,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,eACZ,KAAK,KAAI,0CAA0C,EAAK;uCAExD,EAAO,MAAK,8BAA8B,KAAK,MAAK,KAc1D,GAVA,AAAI,KAAK,SAAW,GACf,MAAK,cAAgB,MAAQ,KAAK,aAAa,SAAW,IAC7D,MAAK,aAAe,EAAO,OAG7B,GACI,KAAK,aAAc,EAAO,MAC1B,eAAe,KAAK,KAAI,0CACpB,EAAK,KAET,EAAE,KACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,uCAGf,GAAI,EAAE,QACJ,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,0CACpB,EAAK,0CAGf,EAAE,OAAS,EACX,EAAA,KAAK,GACL,EAAE,QAAU,GAEZ,KAAK,QAAQ,GAAS,GAMxB,EAAA,UAAA,UAAA,SAAU,EAAmB,GAA7B,GAAA,GAAA,KACE,GAAI,EAAQ,SAAW,EAAQ,OAC7B,KAAM,IAAI,OACN,eAAe,KAAK,KAAI,sCACxB,4BACI,EAAQ,OAAM,qCACd,EAAQ,OAAM,MAGxB,EAAQ,QAAQ,SAAC,EAAG,GAAU,MAAA,GAAK,MAAM,EAAG,EAAQ,OAWtD,EAAA,UAAA,OAAA,SAAO,EAAoB,GACzB,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAK,EAMH,EAAU,EAAQ,MAAM,EAAG,KAAK,aALhC,EAAU,GACV,OAAS,GAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAMjB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAKpC,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAAO,gCAElC,EAAA,MAAM,EAAS,IAMxB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,CAAC,CAAC,GAAS,IAAU,KAAK,MAC5B,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,+BAA+B,GAG/C,GAAI,KAAK,SAAW,EAClB,MAAO,GAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAIpC,OADM,GAAU,GACP,EAAI,EAAG,EAAI,KAAK,OAAQ,IAC/B,EAAQ,KAAK,GAGf,GAAM,GAAU,KAAK,SAAS,GAE9B,UACI,KAAK,aAAc,EAAQ,GAAG,MAC9B,mDACI,KAAK,aAAY,4BAA4B,EAAQ,GAAG,MAAK,KAE9D,EAAA,OAAO,EAAS,IASzB,EAAA,UAAA,QAAA,SAAQ,EAAmB,GACzB,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAGhD,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,CAAC,KAAK,aAAe,GAAY,KAAK,QACxC,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,KAAK,QAAO,KAGtE,KAAK,UAAU,EAAS,EAAA,QAAQ,EAAQ,KAS1C,EAAA,UAAA,MAAA,SAAM,EAAkB,GAAxB,GAAA,GAAA,KACE,GAAI,EAAO,QAAU,KAAK,MACxB,KAAM,IAAI,OAAM,wBACZ,KAAK,MAAK,yBAAyB,EAAO,OAEhD,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEZ,EAAW,4BAA4B,EAAO,OAGpD,GAAI,CAAC,KAAK,aAAe,EAAO,SAAW,KAAK,QAC9C,KAAM,IAAI,OACN,2DACI,KAAK,QAAO,QAAQ,EAAO,OAAM,kEAI3C,GAAM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,GAC1B,EAAA,KAAK,WACH,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,EAAK,cAE3D,MAAO,KAGT,OADM,GAAU,GACP,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAQ,GAAK,EAEf,KAAK,UAAU,EAAS,IAE5B,KC3TA,AAoCA,GAAA,IAAA,WAeE,WACa,EAA4B,EAC5B,EAAwB,GAAA,AAAA,IAAA,QAAA,GAAA,IADxB,KAAA,QAAA,EAA4B,KAAA,aAAA,EAC5B,KAAA,aAAA,EACX,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,IAAiB,EAAO,MAC1B,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,EAAO,OAEhD,GACI,EAAc,EAAO,MAAO,+BAEhC,EAAA,KAAK,KAGT,KAAK,SAAW,EAAA,OAAO,GACvB,KAAK,eAAiB,EACtB,EAAA,KAAK,KAAK,UA5BZ,cAAA,eAAI,EAAA,UAAA,KAAE,KAAN,WACE,MAAO,MAAK,SAAS,oCAiCvB,EAAA,UAAA,KAAA,WACE,MAAO,IAAI,GACH,KAAK,QAAO,QAAG,KAAK,aAAc,KAAK,eAMjD,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,QAAQ,QAAQ,SAAA,GACnB,AAAI,IAAW,MAAQ,CAAC,EAAQ,IAAI,EAAO,MACzC,EAAO,YAGX,KAAK,QAAQ,OAAS,EACtB,KAAK,SAAS,WAKhB,EAAA,UAAA,KAAA,WACE,MAAO,MAAK,QAAQ,QAUtB,EAAA,UAAA,MAAA,SAAM,EAAwB,EAAwB,GAAtD,GAAA,GAAA,KAEE,GAFoD,AAAA,IAAA,QAAA,GAAA,IAEhD,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,IAAgB,IAAM,KAAK,QAAQ,SAAW,EAChD,KAAM,IAAI,OAAM,kCACZ,EAAW,iCACX,KAAK,QAAQ,OAAM,cAEzB,UACI,EAAc,KAAK,aAAc,+BAC9B,EAAA,KAAK,WACV,GAAM,GACF,EAAK,QAAQ,IAAI,SAAA,GAAU,MAAA,GAAA,QAAQ,EAAQ,KAC/C,MAAO,GAAA,MAAM,EAAiB,MASlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAC9B,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAG9C,GAAI,KAAK,SAAW,EAClB,KAAM,IAAI,OAAM,qCAGlB,GAAM,GAAS,KAAK,QAAQ,MAC5B,UACI,EAAO,MAAO,EAAc,+BACzB,EAAA,QAAQ,EAAQ,IAOzB,EAAA,UAAA,SAAA,SAAS,GACP,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAM9C,GAHA,GACI,EAAO,MAAO,KAAK,aAAc,+BAEjC,KAAK,iBAAmB,KAAK,OAC/B,KAAM,IAAI,OAAM,4CAElB,EAAA,KAAK,GACL,KAAK,QAAQ,KAAK,IAOpB,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,EAAO,EACT,KAAM,IAAI,OACN,0DAA0D,GAGhE,GAAI,KAAK,iBAAmB,IAAM,EAAO,KAAK,eAC5C,KAAM,IAAI,OAAM,+BACZ,EAAI,6BAA6B,KAAK,eAAc,KAE1D,KAAK,QAAQ,OAAS,GASxB,EAAA,UAAA,QAAA,SAAQ,EAAsB,EAAwB,GAEpD,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAE9C,GAAI,EAAe,GAAK,EAAe,KAAK,QAAQ,OAClD,KAAM,IAAI,OAAM,4BACZ,EAAY,mBAAmB,KAAK,QAAQ,OAAM,cAGxD,GAAI,KAAK,QAAQ,IAAiB,KAChC,KAAM,IAAI,OAAM,oBAAoB,EAAY,aAGlD,UACI,KAAK,QAAQ,GAAc,MAAO,EAClC,+BAEG,KAAK,QAAQ,IAQtB,EAAA,UAAA,QAAA,SAAQ,EAAsB,GAC5B,GAAI,EAAO,QAAU,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,KAAK,cAG9C,GAAI,EAAe,GACf,KAAK,iBAAmB,IAAM,GAAgB,KAAK,eACrD,KAAM,IAAI,OAAM,yBACZ,EAAY,uBAAuB,KAAK,eAAc,cAG5D,GACI,KAAK,aAAc,EAAO,MAAO,+BACrC,EAAA,KAAK,GACL,KAAK,QAAQ,GAAgB,GAU/B,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAwB,GAAlD,GAAA,GAAA,KAEE,GAAI,IAAiB,KAAK,aACxB,KAAM,IAAI,OAAM,mCACZ,EAAY,uBAAuB,KAAK,cAU9C,MAPA,IACI,KAAK,aAAc,EAAc,+BAIrC,EAAU,EAAQ,MAAM,EAAG,KAAK,QAE5B,EAAQ,SAAW,EACd,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAK,QAAQ,GAAI,KAC1D,MAAO,GAAA,MAAM,EAAS,MAS1B,EAAA,UAAA,OAAA,SAAO,EAAwB,GAA/B,GAAA,GAAA,KACE,GAAI,CAAC,CAAC,GAAgB,IAAiB,KAAK,aAC1C,KAAM,IAAI,OAAM,uBACZ,KAAK,aAAY,+BAA+B,GAMtD,MAHA,IACI,KAAK,aAAc,EAAc,+BAEjC,KAAK,SAAW,EACX,EAAA,OAAO,GAAI,CAAC,GAAG,OAAO,KAAK,eAG7B,EAAA,KAAK,WACV,GAAM,GAAU,EAAK,QAAQ,IAAI,SAAA,GAAK,MAAA,GAAA,QAAQ,EAAG,KACjD,MAAO,GAAA,OAAO,EAAS,MAG7B,iBAQI,EAAgB,EAAwB,GAC1C,GAAM,GAAQ,EAAO,MACrB,GAAI,EAAO,MAAM,OAAS,EACxB,KAAM,IAAI,OACN,oDAAoD,EAAO,OAEjE,GAAI,EAAO,QAAU,EACnB,KAAM,IAAI,OAAM,mCACZ,EAAO,MAAK,uBAAuB,GAEzC,GAAM,GAAc,EAAO,MAAM,MAAM,GACvC,GACI,EAAa,EAAc,+BAE/B,GAAM,GAAuB,EAAA,QAAQ,GACrC,MAAO,IAAI,IAAW,EAAY,EAAc,eAU9C,EAAwB,EAAwB,GAClD,MAAO,IAAI,IAAW,GAAI,EAAc,EAAc,eAWpD,EAAgB,EAAmB,EACnC,GACF,GAAI,EAAQ,SAAW,EAAO,MAAM,GAClC,KAAM,IAAI,OAAM,sDACZ,EAAQ,OAAM,QAAQ,EAAO,MAAM,IAGzC,GAAM,GAAW,KAAK,IAAG,MAAR,KAAY,GAE7B,GAAI,GAAe,MAAQ,IAAgB,IAAM,GAAY,EAC3D,KAAM,IAAI,OACN,mCAAmC,EAAQ,SAAS,EAAW,KAGrE,GAAM,GAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,GACtD,EAAU,EAAA,QAAQ,EAAQ,GAChC,SAAQ,QAAQ,SAAC,EAAO,GACtB,EAAK,QAAQ,EAAO,EAAQ,MAEvB,cAWL,EAAgB,EAAkB,GACpC,GAAI,GAAc,EACZ,EAAoB,EAAO,IAAI,SAAA,GACnC,UAAe,EACR,IAGT,GAAI,IAAgB,EAAO,MAAM,GAC/B,KAAM,IAAI,OAAM;;UAEV,EAAW,4BAA4B,EAAO,OAmBtD,OAhBM,GAAgB,IAAgB,EAAI,EAAI,EAAO,KAAO,EACtD,EAAoB,EAAA,KAAK,WAC7B,GAAM,GAAU,GAChB,EAAS,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAa,IAC1C,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAkB,IAAM,EAAK,EAAI,EAAkB,EAAI,GACvD,EAAU,CAAC,EAAG,EAAgB,GAC9B,EAAQ,CAAC,EAAG,EAAO,GAAI,GAC7B,EAAQ,GAAK,EAAA,QAAQ,EAAA,MAAM,EAAQ,EAAS,GAAQ,GAEtD,SAAO,UACA,IAGH,EAAO,GAAI,IAAW,GAAI,EAAc,EAAO,MAAO,EAAO,QAE1D,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAK,QAAQ,EAAG,EAAQ,IAE1B,MAAO,GC3YT,AAiBA,GAAA,IAAA,OAUa,GAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,kUACnB,EAAA,EAAK,iBACN,KAAA,MAAA,CAAA,EAAA,OACA,cAAA,MAAA,CAAA,EAAA,OAgBA,QAAA,MAAA,CAAA,EAAA,OACA,iBAAA,MAAA,CAAA,EAAA,OAsDA,WAAA,MAAA,CAAA,EAAA,OAIA,SAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QASA,QAAA,MAAA,CAAA,EAAA,QAOA,OAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAKA,gBAAA,MAAA,CAAA,EAAA,QAoBA,qBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAQA,sBAAA,MAAA,CAAA,EAAA,QAUA,uBAAA,MAAA,CAAA,EAAA,QAWA,sBAAA,MAAA,CAAA,EAAA,QAQA,qBAAA,MAAA,CAAA,EAAA,QAWA,oBAAA,MAAA,CAAA,EAAA,QAMA,qBAAA,MAAA,CAAA,EAAA,QAOA,oBAAA,MAAA,CAAA,EAAA,QAUA,oBAAA,MAAA,CAAA,EAAA,QAaA,sBAAA,MAAA,CAAA,EAAA,QACA,oBAAA,MAAA,CAAA,EAAA,QAcA,oBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAYA,kBAAA,MAAA,CAAA,EAAA,QAYA,uBAAA,MAAA,CAAA,EAAA,QAWA,mBAAA,MAAA,CAAA,EAAA,QAUA,qBAAA,MAAA,CAAA,EAAA,QASA,oBAAA,MAAA,CAAA,EAAA,QAUA,kBAAA,MAAA,CAAA,EAAA,wBAzTG,SACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClC,CAAA,EAAM,EAAK,eAC7B,MADM,GAAY,GAAA,OACd,EAAU,GACZ,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,gBAE1C,CAAA,EAAO,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAKtC,SACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAO,EAAc,OAAQ,EAAM,EAAW,GAI/C,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAM,EAAQ,eAAgB,EAAQ,uBAFxC,SACD,GAAA,OAEC,EAAS,EAAK,IAAI,SAAA,IAAU,MAAA,IAAO,KACzB,CAAA,EAAM,EAAW,GAAG,eAAhC,EAAY,GAAA,OAEhB,EAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAChD,GAAO,YAIP,EAAmB,gFAIf,UAAa,EAEV,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACzC,EAAQ,EAAQ,eAAgB,EAAQ,uBAD5C,SAAS,GAAA,OAEH,GAAY,EAAO,IAAI,SAAA,IAAU,MAAA,IAAO,KAI9C,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,YAMN,CAAA,EAAM,EAAQ,YAAY,GAAU,qBACjC,EAAQ,EAAQ,eAAgB,EAAQ,uBAF1C,UACD,GAAA,OAEO,CAAA,EAAM,GAAW,GAAG,eAAhC,SAAY,GAAA,OAEZ,GAAW,QAAQ,SAAA,IACjB,AAAI,CAAC,GAAO,MAAQ,EAAO,QAAQ,GAAO,MAAQ,IAC9C,GAAU,QAAQ,GAAO,MAAQ,IACnC,GAAO,4CA1BN,GAAU,WAAE,CAAA,EAAA,wCA8BnB,MAAA,CAAA,EAAO,UAGD,SAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAChD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAClD,AAAK,EAAK,MACR,GAAO,GAAY,IAGb,CAAA,EAAM,EAAK,gBAAnB,MAAA,CAAA,EAAQ,GAAA,OAAmB,GAAK,CAAC,OAAW,GAAQ,CAAC,EAAM,iBAK3D,MAFM,GAAY,EAAK,WAAW,KAC9B,SAAA,IAAQ,MAAA,IAAU,GAAM,EAAW,KAAa,SAChD,EACI,GAAO,GAAU,EAAW,EAAW,GAC7C,CAAA,EAAO,CAAC,GAAY,MAEtB,CAAA,EAAO,gBAGD,SACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,WAAW,GACnB,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,YACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,SAAU,EAAM,EAAW,GACtD,EAAQ,gBACR,CAAA,EAAO,CAAC,GAAY,aAGd,SAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,yBAA0B,EAAM,EAAW,GAEvD,EAAO,EAAc,OAAQ,EAAM,EAAW,GAC9C,EAAc,GAAI,IACpB,EAAM,EAAO,EAAM,EAAc,EAAwB,EACzD,GACJ,EAAQ,eAAe,GACvB,CAAA,EAAO,CAAC,EAAY,SAAU,EAAA,OAAO,aAG/B,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAmB,EAAQ,eAAe,EAAG,IACnD,EAAiB,MAAM,EAAO,GAC9B,CAAA,EAAO,CAAC,EAAiB,mBAGnB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAkB,EAAQ,eAAe,EAAO,IACtD,CAAA,EAAO,CAAC,EAAgB,KAAK,aAGvB,SACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EAAoB,EAAQ,eAAe,EAAS,IAC1D,CAAA,EAAO,CAAC,EAAkB,OAAO,EAAe,aAG1C,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAqB,EAAQ,eAAe,GAAU,IAC5D,GAAmB,QAAQ,GAAgB,IAC3C,CAAA,EAAO,CAAC,GAAmB,mBAGrB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAoB,EAAQ,eAAe,GAAS,IACpD,GACF,EAAc,QAAS,EAAM,EAAW,GAC5C,CAAA,EAAO,CAAC,GAAkB,OAAO,cAG3B,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,MAAM,GAAS,IAChC,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAkB,EAAQ,eAAe,GAAO,IACtD,CAAA,EAAO,CAAC,EAAA,OAAO,GAAgB,OAAQ,mBAGjC,UACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,GAAmB,EAAQ,eAAe,GAAQ,IACxD,GAAiB,gBACjB,CAAA,EAAO,CAAC,GAAiB,mBAGnB,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,QAAQ,EAAO,GAC1B,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GAE7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAW,EAAc,cAI9C,UACF,EAAc,UAAW,EAAM,EAAW,GACxC,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GACF,GAAQ,GAAe,GAAgB,EAAc,IACzD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,GAAQ,EAAc,GAAc,IACvD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,SACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,EAAS,IAClD,CAAA,EAAO,CAAC,GAAW,OAAO,EAAe,GAAc,aAGjD,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,cAAe,EAAM,EAAW,GAC5C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,MAAM,EAAc,GAAc,cAG/C,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,GAAW,GAAQ,EAAc,IACpD,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAC5C,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,CAAA,EAAO,CAAC,GAAW,OAAO,GAAa,aAGjC,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,GAAa,EAAQ,cAAc,GAAS,IAClD,GAAW,SAAS,GACpB,CAAA,EAAO,CAAC,GAAW,mBAGb,UACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GAAa,EAAQ,cAAc,GAAS,IAClD,CAAA,EAAO,CAAC,GAAW,QAAQ,EAAc,cAGnC,UACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,GACF,EAAc,UAAW,EAAM,EAAW,GAExC,GAAa,GAAM,GAAa,GAAS,GAC/C,EAAQ,cAAc,IACtB,CAAA,EAAO,CAAC,GAAW,mBAGnB,KAAM,WAAU,aAAa,EAAK,GAAE,6BCvW1C,AA2BA,YACI,EAAY,EAA4B,GACpC,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAC7B,EAAc,IAAY,iBAE1B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,yGAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,oFAIR,GAAI,EACF,KAAM,IAAI,OACN,yEAEN,GAAM,GAAS,EAAc,UAAW,EAAM,EAAW,GACnD,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGhB,MAAO,CACL,OAAM,EACN,IAAG,EACH,WAAU,EACV,UAAS,EACT,QAAO,EACP,SAAQ,EACR,eAAc,GAIX,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,UACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAQ,EAAyB,EACjC,QAED,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,gBACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAIvB,8BACG,GAAA,GAAA,GAAA,EAAA,EAAA,GACJ,EAAA,EAAA,OACA,EAAA,EAAA,IACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,eAGF,MAAO,CAAC,EAAA,MAAY,gBAAgB,CAClC,EAAG,EAAc,IAAK,EAAM,EAAW,GAEvC,OAAQ,EAAc,SAAU,EAAM,EAAW,GAEjD,QAAS,CAAC,EAAO,GAAI,EAAO,IAC5B,IAAK,EACL,WAAY,EACZ,UAAW,CAAC,EAAU,GAAI,EAAU,IACpC,KAAM,EACN,WAAY,EACZ,uBAAwB,SAGvB,0BACA,mBACH,GAAM,GAAQ,EACI,cAAe,EAAM,EACrB,GAEZ,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GACxC,MAAO,CAAC,EAAA,gBACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAO,CAAC,EAAO,GAAI,EAAO,IAAK,QAEhC,4BACA,mBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,GAAW,EAAM,EAAW,GAClC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cAET,MAAO,CAAC,EAAA,gBACJ,EAAc,QAAS,EAAM,EAAW,GAExC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAO,GAAI,EAAO,IAAK,EACxB,EAA+B,CAAC,EAAU,GAAI,EAAU,UAEzD,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACD,EAAc,aAAc,EAAM,EAAW,GACzC,cACH,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GAEzC,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,EACnC,EACA,CAAC,EAAU,GAAI,EAAU,GAAI,EAAU,UAExC,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,WACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,CAAC,EAAW,GAAI,EAAW,IAAK,CAAC,EAAO,GAAI,EAAO,IACnD,QAED,qBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,sBAAuB,EAAM,EAAW,GAEpD,EAAA,EAAA,kBAAA,EAAA,IAAA,EAAA,EAAA,GAAA,CAAA,EAAA,GAAA,EAAA,IAAA,CAAA,EAAA,GAAA,EAAA,IAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,QAIf,MAAO,CAAC,EAAQ,OAEb,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,aAAc,EAAM,EAAW,GAEjD,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,CAAC,EAAW,GAAI,EAAW,GAAI,EAAW,IAC1C,CAAC,EAAO,GAAI,EAAO,GAAI,EAAO,IAAK,QAGpC,cACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EAAM,EAAc,MAAO,EAAM,EAAW,GAC5C,EACF,EAAc,YAAa,EAAM,EAAW,GAG1C,EAAe,EAAQ,GACvB,EAAc,EAAQ,GAGtB,EAAiB,EAAU,GAC3B,EAAgB,EAAU,GAEhC,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,CAAC,EAAc,GAAc,EAC7B,CAAC,EAAgB,GAAgB,iBAIrC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC/S9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,KAAW,EAAO,EAAO,QAE9B,YACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAM,EAAc,MAAO,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,SAAe,EAAO,EAAM,QAEjC,eACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,YAAkB,EAAQ,EAAY,QAE3C,UACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,OAAa,EAAS,EAAO,EAAS,QAE3C,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cAEJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,SAEzC,SACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,MACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,SAGzC,mBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,gBACJ,EAAO,EAAM,EACb,EAAc,QAAS,EAAM,EAAW,GAExC,QAED,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1H9C,AAiBA,GAAA,IAAA,OAUA,YACI,EAAY,EAA4B,GAC1C,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAChD,EAAS,EAAc,SAAU,EAAM,EAAW,GAClD,EACF,EAAc,gBAAiB,EAAM,EAAW,GAC9C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAC/C,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CACL,MAAK,EACL,OAAM,EACN,cAAa,EACb,aAAY,EACZ,eAAc,EACd,aAAY,GAIT,GAAM,IAAqC,SAC9C,EAAY,EACZ,GAAyB,MAAA,IAAA,GAAA,OAAA,OAAA,mHACnB,EAAA,EAAK,iBACN,sBAAA,MAAA,CAAA,EAAA,OAgBA,sBAAA,MAAA,CAAA,EAAA,OAcA,sBAAA,MAAA,CAAA,EAAA,OACA,sBAAA,MAAA,CAAA,EAAA,OAQA,QAAA,MAAA,CAAA,EAAA,OAQA,WAAA,MAAA,CAAA,EAAA,uBA9CG,SAOF,GAAU,EAAM,EAAW,GAN7B,EAAK,EAAA,MACL,EAAM,EAAA,OACN,EAAa,EAAA,cACb,EAAY,EAAA,aACZ,EAAc,EAAA,eACd,EAAY,EAAA,aAGC,CAAA,EAAM,EAAA,MAAY,gCAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,wBAGjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAG3D,EACF,EAAc,qBAAsB,EAAM,EAAW,GAG1C,CAAA,EAAM,EAAA,MAAY,6BAC7B,EAAmB,EAAoB,EAAe,EACtD,EAAgB,WAFd,SAAS,EAAA,OAIf,CAAA,EAAO,CAAC,EAAO,gBAAiB,EAAO,sBAIjC,SACF,GAAU,EAAM,EAAW,GADxB,EAAK,EAAA,MAAE,EAAM,EAAA,OAAE,EAAa,EAAA,cAAE,EAAY,EAAA,aAAE,EAAc,EAAA,eAGzD,CAAA,EAAM,EAAA,MAAY,uBACtB,EAAmB,EAAoB,EAAe,EACtD,WAFJ,MAAA,CAAA,EAAA,CAAQ,EAAA,gBAKF,SAAY,EAAA,KACb,EAAc,YAAa,EAAM,EAAW,GAC7C,QACY,CAAA,EAAM,EAAA,WAAiB,WAAjC,SAAM,CAAI,EAAA,QAChB,EAAU,UACV,CAAA,EAAO,UAGP,MAAA,CAAA,EAAO,EAAA,eACH,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,6BC3G1C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EAAS,EAAA,KAAW,EAAG,EAAG,GAChC,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EAAS,EAAA,OAAa,GAC5B,MAAO,CAAC,EAAO,OAAQ,EAAO,aAE3B,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAS,EAAA,OAAa,EAAG,GAC/B,MAAO,CAAC,EAAO,OAAQ,EAAO,iBAG9B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCpDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,GAAU,EAAK,UAEnB,yBACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,IAAY,OACjD,cACH,MAAO,CAAC,GAAU,EAAK,KAAM,EAAW,QACrC,eACA,mBACA,2BACH,GAAM,GAAO,EAAc,IAAK,EAAM,EAAW,GACjD,MAAO,CAAC,GAAY,QAEjB,YACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,IAAY,SACjC,WACH,GAAM,GACD,EAAc,IAAK,EAAM,EAAW,GACzC,MAAO,CAAC,GAAY,QACjB,QACH,MAAO,CAAC,EAAA,SACH,EAAc,IAAK,EAAM,EAAW,GAAoB,MACzD,cACD,SACH,MAAQ,GAAc,IAAK,EAAM,EAAW,GACvC,IAAI,SAAC,GAAc,MAAA,GAAA,SAAe,EAAE,aACtC,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OACH,EAAc,IAAK,EAAM,EAAW,GAAoB,KACzD,cACD,OACH,MAAO,CAAC,EAAA,OAAa,QAClB,QACH,GAAM,GAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAChD,QAAQ,KACJ,kGAEJ,QAAQ,IAAI,GACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,QAAQ,IAAI,MAAM,UAAU,MAAM,KAAK,EAAK,GAAG,YAC9B,MAAM,EAAG,IAE5B,MAAO,CAAC,WAGR,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,eAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,yBACH,GAAM,GACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,eAAgB,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,MAAY,sBAChB,EAA+B,CAAC,EAAK,GAAI,EAAK,IAAK,QAEpD,iBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,WAAY,EAAM,EAAW,GACzC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,qBAAsB,EAAM,EAAW,GAEzD,MAAO,CAAC,EAAA,MAAY,cAChB,EAAmB,EAAmB,EACtC,EAA8B,EAC9B,YAGJ,KAAM,WAAU,aAAa,EAAK,GAAE,yBCzE9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,WACH,MAAO,CAAC,EAAA,SACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,eACH,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,SAErC,aACA,WACH,MAAO,CAAC,EAAA,MACJ,EAAc,YAAa,EAAM,EAAW,GAC5C,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,kBACA,oBACA,SACH,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,aAAc,EAAM,EAAW,GAC7C,EAAc,aAAc,EAAM,EAAW,SAG9C,YACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,SAExC,eACG,GAAA,GAAA,EAAA,WAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAGV,EAAY,IAAY,UACxB,EAAU,IAAmB,QAE7B,EACD,EAAc,UAAW,EAAM,EAAW,GAC/C,GAAI,GACF,GAAI,GAAW,IAAY,EACzB,KAAM,IAAI,OACN,sFAGN,GAAI,CAAC,GAAW,IAAY,EAC1B,KAAM,IAAI,OACN,iEAGF,GAAA,GAAA,EAAA,OAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEhB,MAAO,CAAC,EAAA,MAAY,OAAO,CACzB,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,EAAG,EAAc,IAAK,EAAM,EAAW,GACvC,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,WAAY,EAAc,aAAc,EAAM,EAAW,GAEzD,KAAM,EACN,WAAY,EACZ,uBAAwB,aAI1B,KAAM,WAAU,aAAa,EAAK,GAAE,yBCjF9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,qBACA,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,mBACH,MAAO,CAAC,EAAA,UACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,WAAY,EAAM,EAAW,GAC3C,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,UAAW,EAAM,EAAW,SAE3C,MACH,MAAO,CAAC,EAAA,2BACJ,EAAc,IAAK,EAAM,EAAW,GAEpC,EAAc,SAAU,EAAM,EAAW,GACzC,EAAc,OAAQ,EAAM,EAAW,GACvC,EAAc,QAAS,EAAM,EAAW,GACxC,EAAc,OAAQ,EAAM,EAAW,SAExC,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,SAErC,aACH,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,SAErC,gBACH,MAAO,CAAC,EAAA,cACJ,EAAc,gBAAiB,EAAM,EAAW,GAEhD,EAAc,cAAe,EAAM,EAAW,GAC9C,EAAc,eAAgB,EAAM,EAAW,GAE/C,EAAc,eAAgB,EAAM,EAAW,aAInD,KAAM,WAAU,aAAa,EAAK,GAAE,yBC9E9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,OACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,EAAW,YAGf,KAAM,WAAU,aAAa,EAAK,GAAE,yBCtH9C,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,eACA,UACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,OAAQ,EAAM,EAAW,GACvC,EACA,EAAc,UAAW,EAAM,EAAW,GAC9C,SAAS,EAAO,MAAM,EAAG,GAClB,CAAC,EAAA,OAAa,EAAQ,QAE1B,eACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAC5C,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,OAAa,EAAO,EAAA,KAAW,EAAS,SAAU,QAEvD,gBACA,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EAAQ,EAAc,IAAK,EAAM,EAAW,GAClD,MAAO,CAAC,EAAA,QAAc,EAAO,QAE1B,SAEH,GAAM,GAAQ,EAAc,QAAS,EAAM,EAAW,GAEhD,EAAO,EAAc,OAAQ,EAAM,EAAW,GACpD,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,EACxD,QAED,gBACH,GAAM,GACF,EAAc,QAAS,EAAM,EAAW,GACtC,EACF,EAAc,MAAO,EAAM,EAAW,GACpC,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,cAAe,EAAM,EAAW,GAC5C,EACF,EAAc,iBAAkB,EAAM,EAAW,GAE/C,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,CAAC,EAAA,aACJ,EAAQ,EAAO,EAAK,EAAS,EAAW,EAAS,EACjD,EAAa,QAEd,OACH,MAAO,GAAA,KAAK,WACV,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,UAAW,EAAM,EAAW,GAGxC,EAAQ,EAAQ,GAAG,MACnB,EAAgB,EAAA,QAAc,EAAQ,IAAI,MAC1C,EAAS,EAAQ,IAAI,SAAA,GACzB,GAAM,GAAY,EAAA,KAAK,YAAY,EAAO,MAAO,GACjD,GAAI,CAAC,GACD,CAAC,EAAA,KAAK,YACF,EAAA,QAAc,GAAQ,MAAO,GACnC,KAAM,IAAI,OAAM,0CAElB,MAAO,GAAY,EAAS,EAAA,QAAc,EAAQ,KAEpD,MAAO,CAAC,EAAA,MAAY,EAAQ,UAG3B,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,SAAU,EAAM,EAAW,GAC7C,MAAO,GAAA,QAAc,EAAQ,OAE1B,QACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,YACA,UACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GACrC,EACF,EAAc,kBAAmB,EAAM,EAAW,GAGhD,EAAS,EAAc,IAAK,EAAM,EAAW,GAEnD,MAAO,GAAA,MAAY,EAAQ,EAAiB,OAEzC,aACH,GAAM,GACF,EAAc,UAAW,EAAM,EAAW,GACxC,EACF,EAAc,SAAU,EAAM,EAAW,GACvC,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,UAAgB,EAAS,EAAQ,QAEtC,YACH,GAAM,GAAI,EAAc,IAAK,EAAM,EAAW,GACxC,EACF,EAAc,UAAW,EAAM,EAAW,GAC9C,MAAO,CAAC,EAAA,SAAe,EAAG,QAEvB,iBACH,GAAM,GACF,EAAc,gBAAiB,EAAM,EAAW,GAE9C,EACF,EAAc,cAAe,EAAM,EAAW,GAE5C,EACF,EAAc,eAAgB,EAAM,EAAW,GAC7C,EACF,EAAc,eAAgB,EAAM,EAAW,GACnD,MAAO,CAAC,EAAA,cACJ,EAAS,EAAc,EACvB,EAAa,QAAU,EAAa,MAChC,EACA,EAAA,KAAW,EAAc,EAAa,iBAG9C,KAAM,WAAU,aAAa,EAAK,GAAE,yBC1K9C,AA2BO,GAAM,IACT,SAAC,EAAY,EAA4B,GAEnC,OAAQ,EAAK,QACN,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,SAErC,QACH,MAAO,CAAC,EAAA,MACJ,EAAc,IAAK,EAAM,EAAW,aAGxC,KAAM,WAAU,aAAa,EAAK,GAAE,yBChDlD,AA2BO,GAAM,IACT,SAAC,EAAY,EACZ,GACC,OAAQ,EAAK,QACN,OACH,MAAO,CAAC,EAAA,KACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAGzC,cACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,WACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAEzD,WACH,GAAM,GACF,EAAc,OAAQ,EAAM,EAAW,GAC3C,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GAAoB,QAGzD,UACH,MAAO,CAAC,EAAA,QACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,SAEzC,YACA,MACH,MAAO,CAAC,EAAA,IACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,UAAW,EAAM,EAAW,GAE1C,EAAc,gBAAiB,EAAM,EAAW,SAGjD,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,WAAY,EAAM,EAAW,GAC/C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,kBACH,GAAM,GACF,EAAc,aAAc,EAAM,EAAW,GAC3C,EACF,EAAc,QAAS,EAAM,EAAW,GAC5C,MAAO,CAAC,EAAA,eACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAY,QAEb,gBACH,GAAM,GACF,EAAc,YAAa,EAAM,EAAW,GAC1C,EACD,EAAc,aAAc,EAAM,EAAW,GACrC,cAEb,MAAO,CAAC,EAAA,aACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAW,QAEZ,cACH,MAAO,CAAC,EAAA,YACJ,EAAc,IAAK,EAAM,EAAW,GACpC,EAAc,QAAS,EAAM,EAAW,aAG5C,KAAM,WAAU,aAAa,EAAK,GAAE,yBCnG9C,YAgDI,EAAY,EACZ,GACF,GAAM,GACD,SAAC,EAAY,EAA4B,GACxC,OAAQ,EAAK,cACN,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,cACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAsB,EAAM,EAAW,SAC9C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,UACH,MAAO,IAAkB,EAAM,EAAW,OACvC,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAqB,EAAM,EAAW,SAC7C,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,QACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAgB,EAAM,EAAW,SACpD,UACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAkB,EAAM,EAAW,SACtD,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,gBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAwB,EAAM,EAAW,SAChD,YACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,aACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAoB,EAAM,EAAW,SAC5C,WACH,MAAO,GAAA,KAAS,WAAM,MAAA,IAAmB,EAAM,EAAW,SACvD,iBACH,MAAO,GAAA,KACH,WAAM,MAAA,IAAyB,EAAM,EAAW,SACjD,SACH,GAAM,GAAW,GAAgB,EAAK,IACtC,GAAI,GAAY,EAAS,eACvB,MAAO,GAAS,eACZ,GAAI,IAAc,EAAM,EAAW,IAEvC,KAAM,WAAU,aAAa,EAAK,GAAE,+BAGtC,KAAM,WACF,eAAe,EAAK,GAAE,yIAI7B,EAAM,EAAW,GACxB,MAAI,aAAiB,SACZ,EAAM,KAAK,SAAC,GAAS,MAAA,GAAG,OAAO,KAEjC,GAAG,OAAO,GCvEnB,GAAA,IAAA,WAME,WACa,EACA,EACA,EACA,GAHA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,IAHA,KAAA,UAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,YAAA,EATL,KAAA,YAAc,CAAC,GAAI,EAAG,UAAW,GAAI,YAAa,GAClD,KAAA,SAAmC,CAAC,KAAK,aACzC,KAAA,OAAS,EAQf,KAAK,4BAGC,SAAA,UAAA,SAAR,SAAiB,EAAY,GAC3B,MAAO,CAAC,GAAE,EAAE,UAAS,EAAE,YAAa,IAQtC,OAAA,eAAI,EAAA,UAAA,iBAAc,KAOlB,WACE,MAAO,MAAK,cARd,SAAmB,GACjB,AAAI,KAAK,WAAa,GACpB,MAAK,SAAW,EAChB,KAAK,8DAWT,OAAA,eAAI,EAAA,UAAA,mBAAgB,KAApB,WACE,MAAO,MAAK,mBAAmB,oCAOjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGN,EAAA,UAAA,0BAAR,WAEE,OADM,GAAQ,GACL,EAAI,EAAG,EAAI,KAAK,SAAS,OAAS,EAAG,KAC5C,GAAM,GAAW,KAAK,SAAS,MAAM,EAAG,KAAK,SAAS,OAAS,GAC/D,EAAM,KAAK,KAAK,qBAAqB,IAEvC,EAAM,KAAK,IACX,KAAK,mBAAqB,GAGpB,EAAA,UAAA,qBAAR,SAA6B,GAC3B,MAAO,GACH,EACK,IACG,SAAA,GAAW,MAAC,GAAQ,KAAO,GAAK,EAAQ,cAAgB,EACpD,GACG,EAAQ,UAAS,IAAI,EAAQ,cACvC,KAAK,KACV,IAON,EAAA,UAAA,WAAA,SAAW,GACT,AAAI,KAAK,UACP,MAAK,SACL,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,KAAK,KAAK,SAAS,KAAK,OAAQ,IAC9C,KAAK,mBAAmB,QAAQ,KAAK,qBAAqB,KAAK,aAQnE,EAAA,UAAA,UAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,EAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SAAS,OAAO,IACrB,KAAK,kBAAkB,YAEvB,MAAM,IAAI,OAAM,4CAQpB,EAAA,UAAA,cAAA,WACE,GAAI,KAAK,UAAY,KAAK,SAAS,OAAS,GAC1C,KAAK,SAAW,KAAK,SAAS,QAC9B,KAAK,SACL,GAAM,GACF,OAAO,OAAO,GAAI,KAAK,SAAS,KAAK,SAAS,OAAS,IAC3D,EAAQ,aAAe,EACvB,EAAQ,GAAK,KAAK,OAClB,KAAK,SAAS,OAAO,GAAI,EAAG,GAC5B,KAAK,mBAAmB,OACpB,EAAG,EAAG,KAAK,qBAAqB,KAAK,eAEzC,MAAM,IAAI,OAAM,0DAIpB,EAAA,UAAA,UAAA,SAAU,GACR,MAAO,MAAK,UAAU,IAGxB,EAAA,UAAA,eAAA,SAAe,GACb,KAAK,eAAe,EAAY,IAAM,GAGxC,EAAA,UAAA,eAAA,SAAe,GACb,MAAO,MAAK,eAAe,IAG7B,EAAA,UAAA,cAAA,SAAc,GACZ,KAAK,cAAc,EAAW,IAAM,GAGtC,EAAA,UAAA,cAAA,SAAc,GACZ,MAAO,MAAK,cAAc,IAG5B,EAAA,UAAA,QAAA,SAAQ,GACN,OAAW,KAAO,MAAK,eACrB,KAAK,eAAe,GAAK,cAAc,GAGzC,OAAW,KAAO,MAAK,cACrB,KAAK,cAAc,GAAK,cAAc,IAG5C,KC1LA,YAyCI,EAAwB,EAAiB,EACzC,GACF,GAAM,GAAY,GAAI,KAChB,EAA0B,GAC5B,EAAoB,KACpB,EAAuB,KAIrB,EAAO,GAAI,KACX,EACF,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAEpD,EAA0B,GAC9B,AAAI,GAAa,MACf,GAAgB,EAAU,IAAI,SAAA,GAAQ,MAAA,IAAc,EAAK,MAAM,MAIjE,OADM,GAAe,EAAO,QACrB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MAWtB,GAVA,AAAI,IAAc,IAAS,GAAe,KACxC,CAAI,GAAe,MACjB,GAAc,EACd,EAAa,EAAY,SAAS,IAAI,SAAA,GAAS,MAAA,GAAM,OACnC,OAAO,SAAA,GAAQ,MAAA,GAAU,IAAI,OAGnD,EAAU,IAAI,EAAK,MAGf,EAAU,EAAK,OAAS,KAC1B,SAGF,GAAI,EAAe,QAAQ,EAAK,QAAU,GACxC,SAGF,GAAI,EAAc,QAAQ,EAAK,QAAU,GACvC,SAEF,GAAI,EAAK,OAAO,SAAW,GACzB,EAAc,KAAK,EAAK,MACxB,SAEF,EAAK,OAAO,QAAQ,SAAA,GAElB,GAAI,EAAK,IAAI,EAAM,MACjB,OAEF,EAAK,IAAI,EAAM,MACf,EAAS,KAAK,KAGlB,MAAO,CAAC,OAAM,EAAE,QAAO,EAAE,UAAS,EAAE,cAAa,EAAE,YAAW,EAAE,WAAU,eAQxE,EAAc,EACd,GACK,GAAA,GAAA,EAAA,UAAW,EAAA,EAAA,OACZ,EAAmB,GACnB,EAAa,OAAO,KAAK,GACP,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAChC,IAAI,SAAA,GAAQ,MAAA,GAAM,MAAM,KAC1C,EAAY,EAAM,UAExB,EAAW,QAAQ,SAAA,GACjB,AAAI,EAAU,IAAI,EAAM,OACtB,EAAS,KAAK,KAGlB,EAAM,QAAQ,QAAQ,SAAA,GACpB,AAAI,EAAU,IAAI,EAAO,OACvB,EAAS,KAAK,KAGlB,AAAI,GAAa,MACf,EAAU,QAAQ,SAAA,GAChB,AAAI,EAAU,IAAI,EAAK,OACrB,EAAS,KAAK,KAMpB,OAFM,GAAO,GAAI,KACX,EAAuB,GACtB,EAAS,OAAS,IACvB,GAAM,GAAO,EAAS,MACtB,EAAK,IAAI,EAAK,MACd,AAAK,EAAU,EAAK,OAClB,EAAa,KAAK,GAEpB,EAAK,SAAS,QAAQ,SAAA,GACpB,AAAI,CAAC,EAAK,IAAI,EAAM,OAAS,EAAU,IAAI,EAAM,OAC7C,EAAM,OAAO,MAAM,SAAA,GAAS,MAAA,GAAK,IAAI,EAAM,SAC7C,EAAS,KAAK,KAIpB,MAAO,GAGT,GAAM,IAAmB,CACvB,SAAU,QAAS,QAAS,OAAQ,gBAAiB,cACrD,iBAAkB,KAAM,SAEpB,GAAoB,CACxB,sBAAuB,sBAAuB,sBAAuB,qBAGzC,GAC5B,MAAO,IAAiB,QAAQ,EAAK,KAAO,cAGf,GAC7B,MAAO,IAAkB,QAAQ,EAAK,KAAO,ECjK/C,AAkCA,GAAA,IAAA,WAsFE,WAAoB,EAAsB,GAA1C,GAAA,GAAA,KAAoB,KAAA,MAAA,EAAsB,KAAA,OAAA,EArFlC,KAAA,YAAmC,GAAI,KACvC,KAAA,WAA8B,GAM9B,KAAA,UAAY,IACZ,KAAA,WAAqC,GACrC,KAAA,qBAA0D,GA6EhE,KAAK,SAAW,EAAM,QACtB,KAAK,QAAU,EAAM,OACrB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UACxB,KAAK,WAAa,EAAM,UAExB,AAAI,EAAM,WAAa,MACrB,OAAO,KAAK,EAAM,WAAW,QAAQ,SAAA,GACnC,EAAK,qBAAqB,GACtB,GAAI,GAAc,EAAM,UAAU,GAAO,KApFnD,cAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,4CAGpD,OAAA,eAAI,EAAA,UAAA,sBAAmB,KAAvB,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,oBACZ,KAAK,sDAG5B,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WACE,MAAO,MAAK,OAAS,KAAK,OAAO,UAAY,KAAK,gBAGpD,SAAc,GACZ,GAAM,GAAY,OAAO,KAAK,GAAW,IACrC,SAAA,GAAO,MAAA,GAAU,GAAK,IAAI,SAAA,GAAU,MAAA,GAAO,OAC/C,KAAK,WAAa,GAAG,OAAM,MAAT,GAAa,GAC/B,KAAK,WAAa,mCAGpB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GACtB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,IAAI,SAAA,GACvB,MAAO,CACL,KAAM,EAAK,KACX,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,OACJ,MAAO,EAAK,WAAW,MACnB,EAAK,WAAW,MAAS,MACzB,2CAKV,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,QAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,cAAgB,EAAK,wCAG5D,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,IAAI,SAAC,GACxB,GAAM,GAAO,EAAK,cAAgB,EAAK,KACvC,MAAO,GAAK,cAAoB,EAAI,IAAI,EAAK,cAAmB,qCAIpE,OAAA,eAAI,EAAA,UAAA,YAAS,KAAb,WAAA,GAAA,GAAA,KACE,MAAO,QAAO,KAAK,KAAK,YAAY,OAAO,SAAC,EAAK,GAC/C,SAAI,GAAO,EAAK,WAAW,GAAK,UACzB,GACN,qCA0BG,EAAA,UAAA,kBAAR,SAA0B,EAAgB,GACxC,GAAM,GAAe,EAAO,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OAC7C,EAAgB,EAAQ,IAAI,SAAA,GAAQ,MAAA,GAAK,OAAM,OACrD,MAAO,GAAa,KAAK,KAAK,WAAa,KACvC,EAAc,KAAK,KAAK,YAOtB,EAAA,UAAA,QAAR,SAAgB,EAAwB,GACtC,GAAM,GACF,GAAqB,EAAQ,EAAS,KAAK,UAAW,KAAK,YACxD,EAAA,EAAA,cAAe,EAAA,EAAA,YAAa,EAAA,EAAA,WACnC,GAAI,GAAe,KACjB,KAAM,IAAI,OACN,qCAAqC,EAAY,KAAI,gBACrD,oBAAmB,EAAY,GAAE,kBACjC,6DACA,qCAAoC,EAAU,MAGpD,GAAI,EAAc,OAAS,GACzB,GAAM,GAAW,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,OAC9B,EAAU,OAAO,KAAK,GAC5B,KAAM,IAAI,OACN,+BAA+B,EAAQ,8BACvC,KAAI,EAAO,qCAAqC,EAAa,MAGnE,MAAO,IACH,KAAK,MAAO,KAAK,UAAW,IAYlC,EAAA,UAAA,QAAA,SAAQ,EAAwB,GAAhC,GAAA,GAAA,KACE,EAAS,KAAK,UAAU,GACxB,GAAM,GAAQ,OAAO,KAAK,GAAQ,OAClC,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,GAClB,GAAM,GACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC5D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAG/D,AAAI,EAAY,SAAW,GACzB,GAAc,KAAK,UAGrB,GAAM,GAAiB,KAAK,kBAAkB,EAAY,GAGtD,EAAe,KAAK,YAAY,IAAI,GACxC,AAAI,GAAgB,MAClB,GAAe,KAAK,QAAQ,EAAQ,GACpC,KAAK,YAAY,IAAI,EAAgB,IAGvC,GAAM,GAAiC,GACjC,EAA+B,GACrC,MAAO,GAAA,KAAK,WACV,GAAM,GAAU,GAAI,IAChB,EAAK,UAAW,EAAgB,EAChC,EAAK,qBACH,EAAU,GAAA,GAAwB,EAAK,WAE7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAKzB,OAFM,GAAgB,EAAK,mBAAmB,GACxC,EAA2D,GACxD,EAAI,EAAG,EAAI,EAAa,OAAQ,KACvC,GAAM,GAAO,EAAa,GAC1B,GAAI,CAAC,EAAW,EAAK,OACnB,GAAM,GAAU,GAAU,EAAM,EAAY,GAC5C,GAAI,YAAmB,SACrB,KAAM,IAAI,OACN,4BAA4B,EAAK,GAAE,kEAGzC,EAAW,EAAK,MAAQ,EACxB,EAAK,uBACD,EAAK,KAAM,EAAM,EAAY,EAAS,EACtC,EAAiB,IAIzB,MAAI,GAAK,QAAU,MACjB,EAAQ,QAAQ,GAEX,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAY,QAInD,EAAA,UAAA,mBAAR,SAA2B,GACzB,GAAM,GAAM,GAAG,OAAO,MAClB,GACA,OAAO,KAAK,GACP,IAAI,SAAA,GAAO,MAAA,GAAU,KACrB,IAAI,SAAA,GAAW,MAAA,GAAQ,IAAI,SAAA,GAAU,MAAA,GAAO,QACrD,MAAO,IAAI,KAAI,IAET,EAAA,UAAA,uBAAR,SACI,EAAkB,EAAY,EAC9B,EAA2B,EAC3B,EACA,GAGF,GAAI,EAAK,WAAa,WAAa,EAAY,QAAQ,KAAc,GACnE,OAGF,EAAU,GAAU,QAAQ,SAAA,GAC1B,AAAI,GAAU,MACZ,GAAgC,EAAO,IAClC,GAAgC,EAAO,KAAO,GAC/C,EAAK,SAAS,UAGtB,EAAK,OAAO,QAAQ,SAAA,GAGlB,GAAI,EAAM,WAAa,WACrB,GAAM,GACF,GAA6B,EAAM,KAAM,EAAW,GACxD,AAAI,GAAW,MACb,EAAQ,QAAQ,SAAA,GACd,GAAI,GAAU,CAAC,EAAc,IAAI,EAAO,KACtC,GAAM,GAAQ,EAAgC,EAAO,IACrD,AAAI,IAAU,EACZ,GAAO,UACP,MAAO,GAAgC,EAAO,KACzC,AAAI,GAAS,MAGlB,EAAgC,EAAO,aAkB/C,EAAA,UAAA,aAAN,SAAmB,EAAwB,sEAEzC,MAAA,CAAA,EAAO,KAAK,cAAc,EAAQ,SAiBtB,EAAA,UAAA,cAAd,SACI,EAAwB,EAAmB,EAC3C,EACA,GAF2C,MAAA,KAAA,QAAA,GAAA,IAC3C,AAAA,IAAA,QAAA,GAAA,IACA,AAAA,IAAA,QAAA,GAAA,uGACF,MAAK,IACH,GAAS,KAAK,UAAU,GACxB,KAAK,YAAY,GACjB,KAAK,uBAAuB,GAC5B,EAAU,KAAK,WAAW,GAC1B,KAAK,aAAa,IAGd,EAAU,GAAI,IAChB,KAAK,UAAW,EAAgB,EAChC,KAAK,qBAKS,CAAA,EAAM,KAAK,uBACzB,EAAQ,EAAS,EAAS,WADxB,SAAY,EAAA,OAEZ,EAAU,EAAQ,IAAI,SAAA,GAAQ,MAAA,IAAU,EAAM,EAAW,KAGzD,EAAY,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAC/B,EAAW,OAAO,KAAK,GAAQ,IAAI,SAAA,GAAQ,MAAA,GAAO,GAAM,KACxD,EACF,GAAI,KAAgB,EAAS,OAAK,EAAa,KAAK,YACxD,OAAO,KAAK,GAAW,QAAQ,SAAA,GAC7B,GAAM,GAAc,EAAU,GAC9B,EAAY,QAAQ,SAAA,GAClB,AAAI,GAAU,CAAC,EAAO,YAAc,CAAC,EAAQ,IAAI,EAAO,KACtD,EAAO,cAKb,AAAI,KAAK,QAAU,MACjB,EAAQ,QAAQ,GAGlB,CAAA,EAAO,SAGH,EAAA,UAAA,qBAAN,SACI,EAAkB,EAClB,mFACI,SAAe,EAAO,OAAO,SAAC,EAAK,EAAQ,GAC/C,SAAI,EAAK,OAAO,GAAO,MAAQ,EACxB,GACN,IAEH,CAAA,EAAO,KAAK,cACR,EAAc,KAAK,YAAa,GAAM,EAAgB,SAS9C,EAAA,UAAA,uBAAd,SACI,EAAwB,EAA2B,EACnD,0IACI,EAAQ,OAAO,KAAK,GACpB,EACF,EAAM,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,GAAc,GAAM,MACrD,EAAkB,EAAY,IAAI,SAAA,GAAQ,MAAA,IAAc,GAAM,KAC9D,EAAc,EAAgB,IAAI,SAAA,GAAQ,MAAA,GAAK,MAAM,MAAM,KAC3D,EACF,GAAqB,EAAQ,EAAa,KAAK,WAD5C,EAAS,EAAA,UAAE,EAAa,EAAA,cAAE,EAAW,EAAA,YAAE,EAAU,EAAA,WAGlD,EACE,EAAU,OAAK,KAAK,MAAM,SAAS,IAAI,SAAA,GACzC,MAAO,CAAC,KAAI,EAAE,SAAU,EAAQ,kBAEhC,EAAU,GAAA,GAAwB,KAAK,WAC7C,OAAO,KAAK,GAAQ,QAAQ,SAAA,GACpB,GAAA,GAAA,GAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GACX,EAAoB,GAC1B,EAAQ,GAAS,EAAO,GACxB,EAAW,GAAY,IAEnB,EAA2D,GAC3D,EAAgB,KAAK,mBAAmB,GACxC,EAAkC,0BACjC,GAAM,OAAS,EACd,GAAW,KAAK,aAClB,EAAY,EAAO,EAAS,EAAY,EAAO,EAC/C,EAAiB,EAAiC,GACtD,CAAA,EAAM,QAAQ,IAAI,KAJG,CAAA,EAAA,UAIrB,SAAA,oBAaF,GAXA,AAAI,GAAe,MAAQ,CAAC,GAC1B,QAAQ,KACJ,mIAGA,EACF,EACK,OACG,SAAA,GAAQ,MAAA,CAAC,GAAc,IACnB,CAAC,GAAU,EAAK,KAAM,EAAY,KACzC,IAAI,SAAA,GAAQ,MAAA,GAAK,OACtB,EAAe,OAAS,EACtB,QAAiB,GACrB,AAAI,GAAe,MACjB,GACI,gEACA,4BAA2B,EAAU,MAErC,GAAI,OACN,+BAA+B,EAAc,uBAC7C,YAAW,EAAK,gDAChB,KAAI,EAAa,MAAM,IAE7B,MAAA,CAAA,EAAO,SAGD,EAAA,UAAA,aAAR,SACI,EAAoB,EAA2B,EAC/C,EAA4B,EAC5B,EAA4B,EAC5B,EACA,GAEF,OAPF,GAAA,KAMQ,EAAqC,gBAEzC,GAAM,GAAO,EAAM,MACnB,EAAQ,eAAiB,EAAK,SAC9B,GAAI,GAAW,GAUf,GANA,AAAI,EAAK,KAAK,KAAO,SACjB,EAAc,aAAc,EAAK,KAAM,EAAW,IACnD,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAIC,EAAW,QAAQ,EAAK,QAAU,IACpC,GAAM,GAAU,GAAU,EAAK,KAAM,EAAW,GAChD,AAAK,GACF,GAAA,GAAA,EAAA,KAAA,KAAA,GAAA,IAEH,GAAM,GAAiB,EAAQ,eAC/B,AAAI,YAAmB,SACrB,EAAS,KAAK,EAAQ,KAAK,SAAA,GACzB,SAAU,GAAY,EACtB,EAAQ,eAAiB,EACzB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,GAC1C,KAGT,GAAU,GAAY,EACtB,EAAK,uBACD,EAAU,EAAK,KAAM,EAAW,EAAS,EACzC,EAAa,GACjB,EAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,QAGnD,GAAK,kBACD,EAAK,KAAM,EAAO,EAAS,EAAW,EAAO,WAxC9C,EAAM,OAAS,OA2CtB,MAAO,IAGD,EAAA,UAAA,kBAAR,SACI,EAAY,EAA2B,EACvC,EAA4B,EAC5B,GACF,EAAK,SAAS,QAAQ,SAAC,GACd,GAAA,GAAA,GAAA,EAAA,KAAA,GAAA,GACP,GAAI,EAAM,IAAa,CAAC,EAAU,IAAI,EAAU,MAC9C,OAGF,AAAI,EAAU,KAAO,QACnB,AAAI,EAAU,WAAW,KAAK,SAAA,GACxB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAExC,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,KAGpD,AAAI,EAAU,WAAW,MAAM,SAAA,GACzB,MAAO,CAAC,CAAC,GAAU,EAAM,EAAW,MAE5C,GAAM,GAAY,GAClB,EAAM,KAAK,CAAC,SAAU,EAAQ,eAAgB,KAAM,QAQ1D,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,OAAO,KAAK,KAAK,WACZ,QACG,SAAA,GAAO,MAAA,GAAK,UAAU,GAAK,QAAQ,SAAA,GAAU,MAAA,GAAO,eAGtD,EAAA,UAAA,uBAAR,SAA+B,GAA/B,GAAA,GAAA,KACE,OAAO,KAAK,GAAQ,QAAQ,SAAA,GAC1B,GAAM,GAAQ,EAAO,GACd,EAAA,GAAA,GAAA,GACD,EAAO,EAAK,MAAM,MAAM,GAC9B,GAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,GAAM,GAAQ,EAAK,WAAW,MAAS,MACjC,EAAQ,EAAM,SAAW,EAAM,MAAM,QACvC,EAAM,MAAM,MACR,SAAC,EAAK,GAAU,MAAA,GAAM,KAAW,IAAM,EAAM,KAAW,IAChE,EAAA,KAAK,OACD,EACA,WAAM,MAAA,sBAAsB,EAAK,KAAI,kBACjC,iCAAgC,EAAK,eACrC,KAAI,EAAM,MAAK,OAEzB,AAAI,EAAK,WAAW,OAAY,EAAK,WAAW,MAAS,OACvD,EAAA,KAAK,OACD,EAAM,QAAU,EAAK,WAAW,MAAS,MACzC,WAAM,MAAA,sBAAsB,EAAK,KAAI,8CAE9B,GAAK,WAAW,MAAS,MAAK,aAAa,EAAM,YAK1D,EAAA,UAAA,UAAR,SAAkB,GAChB,GAAM,GAAyB,GAC/B,OAAW,KAAa,GACtB,GAAI,KAAK,YAAc,MAAQ,KAAK,WAAW,QAAU,MACrD,KAAK,WAAW,OAAO,IAAc,MACvC,GAAM,GAAS,KAAK,WAAW,OAAO,GACtC,EAAO,EAAO,MAAQ,EAAO,OAE7B,GAAO,GAAa,EAAO,GAG/B,MAAO,IAGD,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACQ,EAAa,OAAO,KAAK,GAAQ,OAAO,SAAA,GACrC,GAAA,GAAA,GAAA,GAAA,GACP,MAAO,GAAK,MAAM,MAAM,IAAa,OAEvC,GAAI,EAAW,OAAS,EACtB,KAAM,IAAI,OACN,gDACA,WAAU,EAAU,kCAIpB,EAAA,UAAA,WAAR,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,GAAQ,IAAI,SAAA,GACjB,GAAI,EAAK,YAAc,MAAQ,EAAK,WAAW,SAAW,MACtD,EAAK,WAAW,QAAQ,IAAS,MACnC,GAAM,GAAS,EAAK,WAAW,QAAQ,GACvC,MAAO,GAAO,KAEhB,MAAO,IACN,KAGG,EAAA,UAAA,aAAR,SAAqB,GAArB,GAAA,GAAA,KACE,EAAQ,QAAQ,SAAA,GACP,GAAA,GAAA,GAAA,GAAA,GACP,GAAI,CAAC,EAAK,MAAM,MAAM,GACpB,KAAM,IAAI,OAAM,eAAe,EAAI,kCAI3C,KC3lBA,AAyBO,GAAM,IAAqB,oBACrB,GAAqB,2BAoDhC,WACY,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IADA,KAAA,SAAA,EACA,KAAA,YAAA,EAzCJ,KAAA,QAAU,MA0ChB,AAAI,GAAe,MACjB,MAAK,YAAc,IArCvB,cAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,yCAGd,OAAA,eAAI,EAAA,UAAA,aAAU,KAAd,WACE,MAAO,MAAK,SAAS,4CAGvB,OAAA,eAAI,EAAA,UAAA,cAAW,KAAf,WACE,MAAO,MAAK,SAAS,6CAGvB,OAAA,eAAI,EAAA,UAAA,SAAM,KAAV,WACE,MAAO,MAAK,SAAS,wCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,yCAGvB,OAAA,eAAI,EAAA,UAAA,UAAO,KAAX,WACE,MAAO,MAAK,SAAS,2CAoBf,EAAA,UAAA,cAAR,WACE,GAAM,GAAO,KAAK,SAClB,GAAK,EAAsB,MAAQ,KAEjC,KAAK,QAAU,UACN,KAAK,YAAY,aAAe,KACzC,KAAK,QAAU,EAAA,GAAG,mBAAmB,EAAgB,KAAK,kBAE1D,GAAM,GAAW,EAAA,GAAG,gBAAgB,EAAgB,KAAK,aACzD,GAAI,EAAS,SAAW,EAGtB,EAAS,KAAK,EAAA,GAAG,mBAAmB,EAAgB,KAAK,sBAChD,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,CAAC,GAAK,MAEpB,KAAK,QAAU,EAAS,KAQtB,EAAA,UAAA,KAAN,2GAEE,GADA,KAAK,gBACD,KAAK,QAAQ,MAAQ,KACvB,KAAM,IAAI,OACN,iHAGY,MAAA,CAAA,EAAM,KAAK,QAAQ,eAA/B,SAAY,EAAA,OAElB,CAAA,EAAO,KAAK,SAAS,UASvB,EAAA,UAAA,SAAA,SAAS,GACP,KAAK,UAAY,EACjB,GAAM,GAAQ,KAAK,UAAU,cACzB,EAAY,GAChB,AAAI,KAAK,UAAU,qBAAuB,MACxC,GACK,KAAK,UAAU,oBAA4B,WAIlD,KAAK,QAAa,EAAM,SAAS,SAAQ,IAAI,EAAM,SAAS,YAC5D,GAAM,GACF,EAAA,GAAG,cAAc,KAAK,UAAU,WAAY,KAAK,UAAU,aAK/D,GAJA,KAAK,SAAW,GAAI,IAChB,GAAgB,SAAS,eAAe,EAAO,IACnD,KAAK,SAAS,UAAY,KAAK,6BAA6B,GAExD,EAAU,kBAAoB,MAChC,GAAM,GACF,GAAgB,SAAS,eAAe,EAAU,kBACtD,KAAK,YAAc,GAAI,IAAc,GACrC,KAAK,YAAY,UAAY,KAAK,SAAS,UAC3C,KAAK,YAAY,QAAQ,GAAI,IAG/B,MAAO,IA+CH,EAAA,UAAA,KAAN,SAAW,EAAmC,4EAE5C,GAAI,MAAO,IAAiB,UAE1B,GADM,EAAW,EAAA,GAAG,gBAAgB,GAChC,EAAS,SAAW,EACtB,KAAM,IAAI,OACN,0CAA0C,EAAY,KACrD,GAAI,EAAS,OAAS,EAC3B,KAAM,IAAI,OACN,wBAAwB,EAAS,OAAM,uBACvC,SAAQ,EAAY,MAE1B,EAAe,EAAS,GAE1B,GAAI,EAAa,MAAQ,KACvB,KAAM,IAAI,OACN,+GAIN,MAAA,CAAA,EAAO,EAAa,KAAK,KAAK,iBAyChC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,MAAO,MAAK,QAAQ,EAAQ,KAAK,cAG3B,EAAA,UAAA,gBAAR,SAAwB,GAEtB,GAAI,CAAE,aAAkB,GAAA,SAAW,CAAC,MAAM,QAAQ,GAEhD,MAAO,GAGT,GADA,EAAS,MAAM,QAAQ,GAAU,EAAS,CAAC,GACvC,EAAO,SAAW,KAAK,WAAW,OACpC,KAAM,IAAI,OACN,+BACA,wBAAuB,KAAK,WAAW,OAAM,mBAC7C,oBAAmB,EAAO,OAAM,oBAEtC,MAAO,MAAK,WAAW,OAAO,SAAC,EAAK,EAAW,GAC7C,SAAI,GAAc,EAAoB,GAC/B,GACN,KAGG,EAAA,UAAA,iBAAR,SAAyB,GACvB,SAAU,GAAW,KAAK,YACnB,AAAC,MAAM,QAAQ,GAAuB,EAAZ,CAAC,IAmBpC,EAAA,UAAA,QAAA,SAAQ,EAAwC,GAE9C,EAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GAChC,GAAM,GAAS,KAAK,SAAS,QAAQ,EAAQ,GAC7C,MAAO,GAAO,OAAS,EAAI,EAAS,EAAO,IAkBvC,EAAA,UAAA,aAAN,SACI,EACA,mGACF,SAAS,KAAK,gBAAgB,GAC9B,EAAU,KAAK,iBAAiB,GACjB,CAAA,EAAM,KAAK,SAAS,aAAa,EAAQ,WAAlD,SAAS,EAAA,OACf,CAAA,EAAO,EAAO,OAAS,EAAI,EAAS,EAAO,UAGrC,EAAA,UAAA,6BAAR,SAAqC,GACnC,MAAO,QAAO,KAAK,GAAK,OAAO,SAAC,EAAyB,GACvD,SAAO,GAAO,CAAC,EAAI,IACZ,GACN,KAQL,EAAA,UAAA,QAAA,WACE,KAAK,SAAS,UAEd,AAAI,KAAK,aACP,KAAK,YAAY,WAGvB,iBAiCI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,6FACF,GAAI,GAAY,KACd,KAAM,IAAI,OACN,0GAGN,MAAI,IAAW,MACb,GAAU,IAGZ,AAAI,EAAQ,WACV,CAAK,EAA0B,MAAQ,MACrC,CAAM,EAAoB,SAAS,MACjC,GAAY,EAAsB,KAEpC,EAAW,GAAG,EAAW,GAAqB,KAG5C,EAAQ,GAAI,IAAW,EAAU,GACvC,CAAA,EAAM,EAAM,eAAZ,SAAA,OACA,CAAA,EAAO,QC7ZT,GAGM,IAAU,4GCHhB,o8DC2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GAC+B,EAC9B,wBC3GH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC1EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC7EH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAEZ,GAAQ,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAC3B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GAC+B,EAC9B,wBCrEH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GAC+B,EAC9B,wBC5IH,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GAC+B,EAC9B,wBC3EH,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMa,EAAO,SACxC,EAAA,QAAiB,EAEjB,IACE,EAAa,WACN,QAOT,GACA,QCjMF,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAA,IAAiB,cC3DjB,YAkDwB,EAAY,GAElC,MAAO,IAAgB,EAAO,GAShC,YACI,EAAY,EACZ,EAAiC,GAEnC,GAFE,AAAA,IAAA,QAAA,GAAA,GAA0B,MAAO,AAAA,IAAA,QAAA,GAAA,GAA2B,MAE1D,GAAS,KACX,MAAO,MAET,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAI,EAAK,IAAI,GACX,MAAO,GAAK,IAAI,GAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAGL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,GAChB,OAAW,KAAK,IACd,GAAM,GAAQ,EAAM,GACd,EAAc,GAAgB,EAAO,EAAO,EAAM,GACxD,EAAe,GAAK,EAEtB,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAdzD,UAAK,IAAI,EAAO,EAAO,OAChB,EAAO,kBA2Cd,EAAe,GAAA,MAAA,KAAA,QAAA,GAAA,IACV,GAAgB,EAAQ,GAOjC,YACI,EAAe,EACf,GAAA,AAAA,IAAA,QAAA,GAAA,GAA2B,MAG7B,GAAM,GAAQ,EAAO,GACrB,GAAI,EAAY,IAAI,GAClB,KAAM,IAAI,OAAM,0CAElB,GAAM,GAAS,EAAM,GAErB,GAAI,EAAO,SAAW,EAAO,QAAU,KACrC,KAAM,IAAI,OACN,qEAGN,GAAK,EAAO,QAEL,GAAI,GAAW,IAEpB,GAAM,GAA4B,MAAM,QAAQ,GAAS,GAAK,GAC9D,EAAY,IAAI,kBACL,GACT,GAAM,GAAW,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,KAC7B,EAAc,GAAgB,EAAU,EAAO,GACrD,EAAe,GAAK,GAHtB,OAAW,KAAK,KAAL,GAKX,SAAY,OAAO,GACZ,MAEP,MAAM,IAAI,OAAM,yCAAyC,OAbzD,OAAO,GAAO,kBAkBQ,GACxB,MAAI,KAAM,KACD,KAIL,GAAW,EAAE,IACR,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAG,QAAS,gBAsC3B,EAAY,+GACR,EAA+B,GAAI,KAGzC,GAAgB,EAAO,EAAO,OAMZ,EAAA,MAAM,KAAK,EAAK,+BAAhB,GAAA,EAAA,OAAP,GAAG,EAAA,GACN,EAAQ,EAAK,IAAI,GACnB,YAAiB,SACC,CAAA,EAAM,GADxB,CAAA,EAAA,IAFmC,CAAA,EAAA,UAG/B,EAAc,EAAA,OACpB,EAAK,IAAI,EAAK,oBAJA,wBAWZ,SAAS,GAAgB,EAAO,EAAO,GAC7C,CAAA,EAAO,oBASkB,GACzB,MAAO,IAAO,MAAS,CAAC,YAAY,OAAO,IACtC,OAAM,QAAQ,IACb,MAAO,IAAQ,UAAY,CAAE,aAAe,IAAA,qBAYvB,GAC3B,MAAO,IAAO,MAAQ,GAAY,IAAQ,MAAM,QAAQ,IACnD,MAAO,IAAQ,UAAa,YAAe,IAAA,QAC5C,GAAA,KAAQ,aAAa,GAO3B,YAAqB,GACnB,MACI,KAAU,MACT,MAAO,IAAU,UAAY,MAAO,IAAU,WCnRrD,YAqB6B,GAC3B,MAAO,IAAQ,EAAW,IAI5B,YAAuB,GACrB,MAAI,aAAgB,IAAA,OACV,CAAC,MAAO,EAAK,QAAS,QAAS,IAC9B,GAAW,GACb,CAAC,MAAO,KAAM,QAAS,IAEvB,CAAC,MAAO,EAAM,QAAS,IChClC,AAqBA,GAAA,IAAA,WAcE,WAAmB,GACjB,GADiB,KAAA,SAAA,EAVT,KAAA,MAAQ,EACR,KAAA,IAAM,EAUV,GAAY,KACd,KAAM,IAAI,YAAW,mDAEvB,GAAI,EAAW,EACb,KAAM,IAAI,YAAW,6CAEvB,KAAK,KAAO,GAAI,OAAS,GACzB,KAAK,gBAAkB,EAAI,EAMnB,SAAA,UAAA,KAAV,SAAe,GAEb,KAAO,EAAQ,GACb,GAAS,KAAK,gBAEhB,MAAO,GAAQ,KAAK,iBAGZ,EAAA,UAAA,IAAV,SAAc,GACZ,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,MAAO,MAAK,KAAK,EAAQ,KAAK,WAGtB,EAAA,UAAA,IAAV,SAAc,EAAe,GAC3B,GAAI,EAAQ,EACV,KAAM,IAAI,YAAW,uCAEvB,KAAK,KAAK,EAAQ,KAAK,UAAY,GAMrC,EAAA,UAAA,OAAA,WACE,GAAI,GAAS,KAAK,IAAM,KAAK,MAC7B,MAAI,GAAS,GACX,GAAS,KAAK,gBAAkB,GAE3B,GAQT,EAAA,UAAA,OAAA,WACE,MAAO,MAAK,WAAa,KAAK,UAQhC,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,WAAa,GAM3B,EAAA,UAAA,KAAA,SAAK,GACH,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,IAAI,KAAK,IAAK,GACnB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,IAMlC,EAAA,UAAA,QAAA,SAAQ,GACN,OAAoB,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,OAAA,KAAf,GAAM,GAAK,EAAA,GACd,KAAK,KAAK,KAOd,EAAA,UAAA,IAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,KAAK,IAAM,KAAK,KAAK,KAAK,IAAM,GAChC,GAAM,GAAS,KAAK,IAAI,KAAK,KAC7B,YAAK,IAAI,KAAK,IAAK,QACZ,GAMT,EAAA,UAAA,QAAA,SAAQ,GACN,GAAI,KAAK,SACP,KAAM,IAAI,YAAW,wBAEvB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GACpC,KAAK,IAAI,KAAK,MAAO,IAMvB,EAAA,UAAA,MAAA,WACE,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAS,KAAK,IAAI,KAAK,OAC7B,YAAK,IAAI,KAAK,MAAO,QACrB,KAAK,MAAQ,KAAK,KAAK,KAAK,MAAQ,GAC7B,GAYT,EAAA,UAAA,cAAA,SAAc,GACZ,GAAI,KAAK,UACP,KAAM,IAAI,YAAW,yBAEvB,GAAM,GAAQ,KAAK,KAAK,KAAK,MAAQ,GAC/B,EAAS,KAAK,IAAI,GACxB,YAAK,IAAI,EAAO,KAAK,OACd,GAEX,KC/KA,AAoBA,GAAA,IAAA,SAAA,GAA0C,GAAA,EAAA,GAMxC,mBACE,GAAA,KAAA,KAAM,EAAkB,mBAAiB,KAG3C,SAAA,UAAA,OAAA,WACE,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,GACH,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,KAAI,KAAA,KAAC,IAGb,EAAA,UAAA,QAAA,SAAQ,GACN,AAAI,EAAA,UAAM,OAAM,KAAA,OACd,KAAK,SAEP,EAAA,UAAM,QAAO,KAAA,KAAC,IAMR,EAAA,UAAA,OAAR,WAOE,OANM,GAAc,KAAK,SAAW,EAC9B,EAAU,GAAI,OAAS,GACvB,EAAM,KAAK,SAIR,EAAI,EAAG,EAAI,EAAK,IACvB,EAAQ,GAAK,KAAK,IAAI,KAAK,KAAK,KAAK,MAAQ,IAG/C,KAAK,KAAO,EACZ,KAAK,SAAW,EAChB,KAAK,gBAAkB,EAAI,KAAK,SAChC,KAAK,MAAQ,EACb,KAAK,IAAM,GA7CE,EAAA,iBAAmB,GA+CpC,GAhD0C,ICpB1C,YAuCqC,GACnC,MAAO,IAAI,IAAc,eAyBvB,GAEF,MAAO,IAAI,IAAqB,eAgB9B,EACA,GACF,MAAO,IAAI,IAAgB,EAAe,eAmDxC,EACA,GAAA,MAAA,KAAA,QAAA,GAAgC,GAAgB,MAC3C,GAAI,IAAe,EAAW,GAUvC,GAAA,IAAA,WAAA,cAwBQ,SAAA,UAAA,QAAN,6GACQ,SAAc,GACZ,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,KAAK,gBAAf,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAcH,EAAA,UAAA,eAAN,+GACQ,SAAS,KAAK,SAAS,KACvB,EAAc,GACZ,CAAA,EAAM,EAAO,eAAjB,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACZ,GAAO,KAAK,EAAE,OACV,CAAA,EAAM,EAAO,gBAAjB,SAAI,EAAA,oBAEN,MAAA,CAAA,EAAO,SAUH,EAAA,UAAA,aAAN,2GACU,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,8BACA,GAAE,KAAI,CAAA,EAAA,GACR,CAAA,EAAM,KAAK,eAAf,SAAI,EAAA,oCAWF,EAAA,UAAA,aAAN,SAAmB,qGACT,MAAA,CAAA,EAAM,KAAK,eAAf,EAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,8BACzB,CAAC,EAAE,MAAS,EACd,CAAA,EAAM,KAAK,QADiB,CAAA,EAAA,UAChC,SAAI,EAAA,OACJ,EAAiB,EAAU,EAAE,oCAgBjC,EAAA,UAAA,aAAA,SAAa,GACX,MAAO,IAAI,IAA0B,KAAM,IAa7C,EAAA,UAAA,OAAA,SAAO,GACL,MAAO,IAAI,IAAe,KAAM,IAWlC,EAAA,UAAA,IAAA,SAAO,GACL,MAAO,IAAI,IAAY,KAAM,IAW/B,EAAA,UAAA,SAAA,SAAY,GACV,MAAO,IAAI,IAAiB,KAAM,IAWpC,EAAA,UAAA,eAAA,SAAkB,GAChB,MAAO,IAAI,IAAiB,KAAM,GAAW,UAW/C,EAAA,UAAA,QAAA,SAAW,GACT,MAAO,IAAI,IAAgB,KAAM,IAQ7B,EAAA,UAAA,aAAN,SAAmB,sEACjB,MAAA,CAAA,EAAO,KAAK,IAAI,GAAG,qBAUf,EAAA,UAAA,cAAN,SAAoB,sEAClB,MAAA,CAAA,EAAO,KAAK,eAAe,GAAG,aAAa,SAAA,GAAK,MAAC,KAAM,WAqBzD,EAAA,UAAA,cAAA,SAAc,EAAmB,GAAA,MAAA,KAAA,QAAA,GAAA,IACxB,GAAI,IAAsB,KAAM,EAAW,IAmCpD,EAAA,UAAA,iBAAA,SACI,EAAmB,EAEnB,GAFmB,AAAA,IAAA,QAAA,GAAA,IAEnB,AAAA,IAAA,QAAA,GAAA,IAGF,GAAM,GAAa,KAAK,cAAc,EAAW,GAGjD,MAAO,GAAW,IAAI,SAAA,GAAK,MAAA,IAAQ,EAAG,MAaxC,EAAA,UAAA,YAAA,SACI,EACA,GACF,MAAO,IAAI,IACP,GAAkB,CAAC,KAAM,IAAY,IAU3C,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAShC,EAAA,UAAA,KAAA,SAAK,GACH,MAAI,GAAQ,GAAK,GAAS,KACjB,KAEF,GAAI,IAAa,KAAM,IAYhC,EAAA,UAAA,SAAA,SAAS,GACP,MAAO,IAAI,IAAiB,KAAM,IAapC,EAAA,UAAA,QAAA,SAAQ,EAAoB,GAC1B,MAAO,IAAI,IAAgB,KAAM,EAAY,IAO/C,EAAA,UAAA,OAAA,WACE,MAAO,IAAI,IAAe,OAE9B,KAWA,GAAA,SAAA,GAA+B,GAAA,EAAA,GAE7B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,MAAA,EADd,EAAA,KAAO,IAKf,SAAA,UAAA,QAAA,WACE,MAAO,YAAY,KAAK,MAAM,OAAM,UAGhC,EAAA,UAAA,KAAN,oFACE,MAAI,MAAK,MAAQ,KAAK,MAAM,OAC1B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAO,KAAK,MAAM,KAAK,MAC7B,KAAK,OACL,CAAA,EAAO,CAAC,MAAO,GAAU,GAAO,KAAM,WAE1C,GAlB+B,IAoB/B,GAAA,SAAA,GAAsC,GAAA,EAAA,GACpC,WACc,GADd,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,OAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAO,iBAGH,EAAA,UAAA,KAAN,8EACE,IACE,MAAA,CAAA,EAAO,KAAK,gBACL,GAEP,QAAE,QACE,mDAAmD,EAAE,QACnD,iBAGZ,GApBsC,IAsBtC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,8EACE,MAAA,CAAA,EAAO,KAAK,SAAS,aAEzB,GA1BgC,IA4BhC,GAAA,SAAA,GAA8B,GAAA,EAAA,GAQ5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAF3D,EAAA,MAAQ,EAIN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,iHAKS,MAAK,QAAU,KAAK,SACT,CAAA,EAAM,KAAK,SAAS,QADH,CAAA,EAAA,UAGjC,MAFM,GAAU,EAAA,OAEZ,EAAQ,KACV,CAAA,EAAO,GAET,IAAA,QAAW,EAAQ,qBAErB,MAAA,CAAA,EAAO,KAAK,SAAS,cAEzB,GAzC8B,IA2C9B,GAAA,SAAA,GAA8B,GAAA,EAAA,GAE5B,WAAsB,EAAqC,GAA3D,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAAqC,EAAA,SAAA,EAD3D,EAAA,MAAQ,IAKR,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,8EACE,MAAI,MAAK,SAAW,KAAK,SACvB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,CAAA,EAAO,KAAK,SAAS,aAEzB,GAhB8B,IAqB9B,GAAA,SAAA,GAAuC,GAAA,EAAA,GAKrC,WACc,EAAqC,EACrC,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EAAqC,EAAA,UAAA,EACrC,EAAA,qBAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,qBAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,6GACQ,EAAa,0BACZ,GAAM,OAAS,KAAK,UACZ,CAAA,EAAM,KAAK,SAAS,QADC,CAAA,EAAA,UAElC,MADM,GAAO,EAAA,OACT,EAAK,KACH,KAAK,sBAAwB,EAAM,OAAS,EAC9C,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,KAE9B,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAM,KAAK,EAAK,qBAElB,MAAA,CAAA,EAAO,CAAC,MAAO,EAAO,KAAM,WAEhC,GAvCuC,IAyCvC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAK9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,EAEZ,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAG7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,WAAd,2GAEiB,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,MAAQ,KAAK,UAAU,EAAK,OACnC,CAAA,EAAO,GAET,IAAA,QAAW,EAAK,qCAGtB,GAlCgC,IAoChC,GAAA,SAAA,GAAgC,GAAA,EAAA,GAC9B,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,WAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAc7B,IAZM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAOzD,EAAS,KAAK,UAAU,EAAK,OAC7B,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCgC,IAqChC,GAAA,SAAA,GAA2C,GAAA,EAAA,GAEzC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,QAAA,EAHd,EAAA,MAAQ,EAKN,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGtD,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,oBAO7B,EAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGR,EAAA,UAAA,WAAN,yJAGa,CAAA,EAAM,KAAK,SAAS,eAA3B,MAAA,CAAA,EAAO,EAAA,eAEP,kBAAI,AAAC,KAAK,QAAQ,SAChB,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,+CAUrC,GA1C2C,IA4C3C,GAAA,SAAA,GAAqC,GAAA,EAAA,GACnC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,MADM,GAAO,EAAA,OACT,EAAK,KACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAe,GAAA,YAAe,sBAAsB,EAAK,OAOhD,CAAA,EAAM,KAAK,UAAU,EAAK,gBAKzC,IALM,EAAS,EAAA,OACT,EAAgB,GAAA,YAAe,sBAAsB,GAI3D,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAGN,MAAA,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAEjC,GAnCqC,IAgDrC,GAAA,SAAA,GAAmD,GAAA,EAAA,GAQjD,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KACP,SAAK,YAAc,GAAI,IACvB,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAiBR,EAAA,UAAA,WAAN,2GAIS,MAAK,YAAY,WAAa,EAE9B,CAAA,EAAM,KAAK,QAFoB,CAAA,EAAA,UAEpC,MAAK,GAAA,aACH,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,YAG/B,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,YAAY,QAAS,KAAM,WAEnD,GAjDmD,IAkDnD,GAAA,SAAA,GAAoC,GAAA,EAAA,GAClC,WACc,EACA,GAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,SAAA,EACA,EAAA,UAAA,IAId,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,eAG7B,EAAA,UAAA,KAAN,uHACe,MAAA,CAAA,EAAM,KAAK,SAAS,eACjC,GADM,EAAO,EAAA,OACT,EAAK,KACP,MAAA,CAAA,EAAO,IAeT,IAbM,EAAe,GAAA,YAAe,sBAAsB,EAAK,OAMzD,EAAc,KAAK,UAAU,EAAK,OAClC,EACF,GAAA,YAAe,sBAAsB,GACzC,KAAK,YAAY,QAAQ,GAIzB,EAAA,EAAgB,EAAA,EAAA,EAAA,EAAA,OAAA,IAAL,EAAC,EAAA,GACV,AAAK,GAAA,YAAe,eAAe,EAAG,IACpC,EAAE,UAIN,MAAA,CAAA,EAAO,UAEX,GArCoC,IAgDpC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAStC,WACI,EACiB,GAFrB,GAAA,GAGE,EAAA,KAAA,OAAO,KADY,SAAA,iBAAA,EARb,EAAA,SAAuC,KAGvC,EAAA,SAA4B,KAOlC,EAAK,cAAgB,IAGvB,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,8CAC1B,MAAU,GAAiB,eAGvB,EAAA,UAAA,KAAN,8EACE,YAAK,SAAW,KAAK,cAAc,KAAK,UACxC,CAAA,EAAO,KAAK,eAGA,EAAA,UAAA,cAAd,SAA4B,qGAO1B,MAAA,CAAA,EAAM,gBAAN,GAAA,OACI,KAAK,UAAY,KACI,CAAA,EAAM,KAAK,cAAc,QAD9C,CAAA,EAAA,UAEF,GADM,EAAiB,EAAA,OACnB,EAAe,KAEjB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,KAAK,SAAW,EAAe,MAC/B,AAAI,KAAK,kBAAoB,MAC3B,MAAK,SAAW,KAAK,SAAS,aAAa,KAAK,oCAGjC,MAAA,CAAA,EAAM,KAAK,SAAS,eACvC,MADM,GAAa,EAAA,OACf,EAAW,KACb,MAAK,SAAW,KAChB,CAAA,EAAO,KAAK,cAAc,KAE5B,CAAA,EAAO,SAEX,GApDwC,IAsD5B,GAAZ,AAAA,UAAY,GACV,EAAA,EAAA,KAAA,GAAA,OACA,EAAA,EAAA,SAAA,GAAA,WACA,EAAA,EAAA,QAAA,GAAA,YAHU,IAAA,IAAe,KAmC3B,GAAA,IAAA,SAAA,GAAwD,GAAA,EAAA,GAItD,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAgC,GAAgB,MAFvE,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,UAAA,EACA,EAAA,aAAA,EALf,EAAA,MAAQ,EACR,EAAA,eAA6C,OAQrD,SAAA,UAAA,QAAA,WACE,GAAM,GAAoB,0CAC1B,MAAO,IAAI,EAAiB,YAGhB,EAAA,UAAA,UAAd,SAAwB,2CAWtB,WAAiB,GACf,GAAI,YAAqB,KACvB,GAAM,GAAS,EAAU,OACzB,MAAO,CACL,MAAO,EAAO,KAAK,SAAA,GACjB,WACA,AAAI,EAAE,MACJ,IAEK,EAAE,QAEX,QAAS,QAGX,OAAO,CAAC,MAAO,KAAM,QAAS,gEArBlC,MAAA,CAAA,EAAM,UAAN,SAAA,OAII,EAAe,EACf,EAAgB,EAoBF,CAAA,EAAM,GAAmB,KAAK,UAAW,WAE3D,GAFM,EAAY,EAAA,OAEd,IAAiB,EAEnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAE7B,GAAI,EAAgB,EAClB,OAAQ,KAAK,kBACN,IAAgB,KACnB,KAAM,IAAI,OACN,+CACA,0BAAyB,KAAK,MAAK,UACpC,IAAgB,SACnB,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,SACxB,IAAgB,SAMzB,YAAK,QACL,CAAA,EAAO,CAAC,MAAO,EAAQ,KAAM,WAGzB,EAAA,UAAA,KAAN,8EACE,YAAK,eAAiB,KAAK,UAAU,KAAK,gBAC1C,CAAA,EAAO,KAAK,qBAEhB,GAxEwD,IAoFxD,GAAA,SAAA,GAAyC,GAAA,EAAA,GAGvC,WACc,EAAqC,GADnD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAAqC,EAAA,WAAA,EAEjD,EAAK,OAAS,GAAI,IAAuC,KAG3D,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,gBAOzB,EAAA,UAAA,OAAV,WACE,KAAO,CAAC,KAAK,OAAO,WAClB,GAAM,GAAI,KAAK,SAAS,OACxB,KAAK,OAAO,KAAK,KAIrB,EAAA,UAAA,KAAA,WACE,YAAK,SAIE,KAAK,OAAO,SAEvB,GA/ByC,IAuCzC,GAAA,SAAA,GAAwC,GAAA,EAAA,GAUtC,WACc,EAAqC,EAC/C,GAFJ,GAAA,GAGE,EAAA,KAAA,KAAM,EAAU,IAAW,KAFf,SAAA,SAAA,EAAqC,EAAA,WAAA,EAH3C,EAAA,kBAAoB,GAM1B,EAAK,OAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACpD,EAAK,SAAW,QAAQ,QAAQ,CAAC,MAAO,KAAM,KAAM,OAGhD,SAAA,UAAA,KAAN,yFAKE,YAAK,SAAW,KAAK,SAAS,KAAK,WAAM,MAAA,GAAK,eAC9C,CAAA,EAAO,KAAK,eAGN,EAAA,UAAA,UAAR,SAAkB,GAChB,MAAO,MAAK,MAAM,KAAK,SAAW,IAG1B,EAAA,UAAA,YAAV,WACE,MAAO,MAAK,UAAU,KAAK,OAAO,WAG9B,EAAA,UAAA,WAAN,6GAEE,AAAK,KAAK,mBACR,KAAK,gCAEC,MAAK,OAAO,UAAS,CAAA,EAAA,GACrB,GAAc,KAAK,cACV,CAAA,EAAM,KAAK,OAAO,cAAc,YAC/C,GADM,EAAS,EAAA,OACX,EAAO,KACT,KAAK,kBAAoB,OAEzB,aAAK,SACL,CAAA,EAAO,sBAGX,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,WAE/B,GApDwC,ICjnCxC,kBA6DA,aAWW,KAAA,KAAe,KA+DxB,SAAA,UAAA,MAAA,SAAM,EAAmB,GAAzB,GAAA,GAAA,KAAyB,AAAA,IAAA,QAAA,GAAA,IACvB,GAAM,GAAO,KACb,GAAA,KAAQ,OACJ,EAAY,EAAG,WAAM,MAAA;QACrB,IACJ,GAAI,GACJ,MAAI,MAAK,OAAS,UAAY,KAAK,MAAQ,KAGzC,EAAO,KAAK,KACP,AAAI,EAGT,EAAO,KAAK,KAAK,KAAK,KAAO,GAI7B,EAAO,KAAK,MAAM,KAAK,KAAO,GAEzB,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OACH,iBAAiB,EAAW,EAAgB,WAChD,IAkBL,EAAA,UAAA,YAAA,SAAY,GAAZ,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,UAAY,EAAQ,OAAS,SAG7C,EAAO,SACF,AAAI,KAAK,MAAQ,MAAQ,EAAQ,MAAQ,KAG9C,EAAO,KAAK,KAAO,EAAQ,KAI3B,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,EAAA,MAAA,IAAA,KAAA,SAAA,0BACK,MAAA,CAAA,EAAM,EAAK,mBAAZ,SAAA,GAAC,EAAA,QAAuB,YAAY,CAAA,EAAM,EAAQ,mBAAlD,MAAA,CAAA,EAAA,EAAA,MAAA,EAAA,CAAoC,EAAA,gBACxC,IAmBN,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,OAAS,SAEhB,EAAO,SAIP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,OAAO,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAClE,IAmBC,EAAA,UAAA,aAAN,SAAmB,6FACT,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,aAAa,UAkB9C,EAAA,UAAA,IAAA,SAAkC,GAAlC,GAAA,GAAA,KACQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,IAAI,SAAA,GAAK,MAAA,IAAA,KAAQ,WAAM,MAAA,GAAU,cAC/D,KAAK,OA0BV,EAAA,UAAA,SAAA,SAAuC,GAAvC,GAAA,GAAA,KAEQ,EAAO,KACb,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,6DACnB,MAAA,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,SAAS,UACvC,KAAK,OAYV,EAAA,UAAA,SAAA,SAAS,GAAT,GAAA,GAAA,KACE,GAAI,GAAc,KAChB,KAAM,IAAI,YACN,6DAGN,GAAM,GAAO,KACb,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,SAAS,UAAa,KAAK,OAqBrE,EAAA,UAAA,OAAA,SAAO,GAAP,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,EAAQ,EAI/B,EAAO,KAAK,KAAO,EACd,AAAI,IAAU,EAEnB,EAAO,EACF,AAAI,KAAK,MAAQ,MAAS,KAAU,QAAa,EAAQ,GAG9D,EAAO,SAGP,EAAO,KAEF,GAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mDACrB,SAAmB,GACrB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,GAAA,GAAA,MAAA,IAAA,KAAA,SAAA,sCAAqB,CAAA,EAAM,EAAK,mBAApB,MAAA,CAAA,EAAE,GAAA,MAAO,EAAA,OAAuB,EAAA,KAAM,GAAK,WAC3D,CAAA,EAAO,GAAyB,EAAiB,KAAK,UACrD,IAoBL,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,GAAS,GAAK,KAAK,MAAQ,EAIlD,EAAO,KAAK,KAAO,EACd,AACH,KAAK,MAAQ,MACZ,MAAK,KAAO,GAAS,IAAU,QAAa,EAAQ,GAGvD,EAAO,EAGP,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IA4BvD,EAAA,UAAA,QAAA,SAAQ,EAAoB,EAAe,GAA3C,GAAA,GAAA,KAEE,GAFyC,AAAA,IAAA,QAAA,GAAA,IAErC,GAAc,MAAQ,EAAa,EACrC,KAAI,MAAK,MAAQ,KACT,GAAI,YACN,4DAEE,GAAI,YACN,+KAGA,oCAAmC,KAAK,KAAI,eAGpD,GAAM,GAAO,KACP,EAAS,GAAgB,GAAQ,GAAA,KAAQ,MAAM,YACrD,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACvB,SAAQ,EAAO,QACnB,AAAI,GACF,IAAS,EAAO,SAEV,CAAA,EAAM,EAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,QAAQ,EAAY,EAAM,mBACxD,KAAK,OAoBV,EAAA,UAAA,KAAA,SAAK,GAAL,GAAA,GAAA,KACQ,EAAO,KACT,EACJ,MAAI,MAAK,MAAQ,MAAQ,KAAK,KAAO,EAGnC,EAAO,EACF,AAAI,KAAK,MAAQ,MAAQ,KAAK,MAAQ,EAG3C,EAAO,KAAK,KAGZ,EAAO,KAEF,GACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,0BAAa,MAAA,CAAA,EAAM,EAAK,mBAAZ,MAAA,CAAA,EAAC,EAAA,OAAuB,KAAK,UAAQ,IAmBjD,EAAA,UAAA,QAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,iBAc3B,EAAA,UAAA,eAAN,qGACE,GAAI,KAAK,OAAS,SAChB,KAAM,IAAI,OAAM,kDAEV,MAAA,CAAA,EAAM,KAAK,mBAAnB,MAAA,CAAA,EAAQ,EAAA,OAAuB,wBA1HjB,EAAA,gBAAkB,IA4HpC,iBAeI,EACA,GAAA,MAAA,KAAA,QAAA,GAAA,MACK,GAAA,UAAA,GAAkB,GAAA,EAAA,GAAd,aAAA,GAAA,GAAA,IAAA,MAAA,EAAA,MAAA,KAAA,YAAA,KACT,SAAA,KAAO,IAMD,SAAA,UAAA,SAAN,8EACE,MAAA,CAAA,EAAO,UAEX,GAVyB,iBAgCyB,GAApD,GAAA,GAAA,KACE,MAAO,IACH,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,GAAkB,SAAQ,EAAM,oBA4CA,GAAlD,GAAA,GAAA,KAGE,GAAI,CAAC,GAAW,GACd,KAAM,IAAI,OAAM,qDAElB,GAAI,GACJ,GAAI,MAAM,QAAQ,GAChB,OAAS,GAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAO,GAAQ,KAAQ,EAAS,GAAkB,KAC5B,KAAK,IAAI,EAAO,EAAS,GAAkB,cAE1D,YAAoB,QAC7B,OAAW,KAAM,GACf,EAAO,GAAQ,KAAQ,EAAS,GAAmB,KAC7B,KAAK,IAAI,EAAO,EAAS,GAAmB,MAGtE,MAAO,IAAyB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACd,MAAA,CAAA,EAAM,GAAmB,EAAU,SAAA,GACjD,GAAI,YAAa,IACf,MAAO,CAAC,MAAO,EAAE,WAAY,QAAS,IACjC,GAAI,GAAW,GACpB,MAAO,CAAC,MAAO,KAAM,QAAS,IAE9B,KAAM,IAAI,OACN,uFAPF,SAAU,EAAA,OAWhB,CAAA,EAAO,GAAsB,EAAS,GAAgB,iBACrD,GAWL,YAAyB,GACvB,GAAI,IAAS,KACX,MAAO,MAIT,GAAM,GAAa,EAAK,GAExB,GAAI,GAAa,IAEf,GAAM,GAAQ,GAAY,GAC1B,MAAO,CAAC,MAAK,EAAE,QAAS,IAI1B,MAAO,CAAC,MAAO,KAAM,QAAS,IAOhC,YAAwD,GAEtD,GAAI,EAAO,SAAW,EAEpB,KAAM,IAAI,OAAM,wCAGlB,MAAI,GAAO,YAAc,IAAA,OAEhB,GAAA,MAAS,GAGT,GAAA,OAAU,GC9rBrB,mBA2BqC,GAAA,EAAA,GAMnC,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,IAIzB,SAAA,UAAA,SAAN,+GACwB,MAAA,CAAA,EAAM,KAAK,MAAM,mBAAjC,SAAgB,EAAA,OAChB,EAAe,EAAc,aAC7B,EAAe,EAAa,MAAM;GAAM,IAAI,SAAA,GAEhD,MAAI,GAAK,SAAS,OAChB,GAAO,EAAK,MAAM,EAAG,KAEhB,IAET,CAAA,EAAO,SAEX,GAtBqC,IC3BrC,AAyBA,GAAM,IAAa,IACb,GAAY,OAAO,OACnB,GAAc,OAAO,SACrB,GAAc,OAAO,SACrB,GAA0B,OAAO,mBACjC,GAA8B,OAAO,+BAeX,GAAA,EAAA,GAgI9B,WAA+B,EAAmB,GAAlD,GAAA,GACE,EAAA,KAAA,OAAO,KADsB,SAAA,MAAA,EA9HvB,EAAA,UAAY,GACZ,EAAA,gBAA4B,KAC5B,EAAA,qBAAuB,GACvB,EAAA,cAA+C,KAC/C,EAAA,sBAAwB,GACxB,EAAA,UAAY,IACZ,EAAA,gBAAkB,GA0HxB,EAAK,KAAO,GAAI,IAAgB,GAChC,AAAK,GACH,GAAY,IAEd,EAAK,UAAY,IAAU,YAAc,IACzC,EAAK,gBAAkB,EAAU,YACjC,EAAK,cAAgB,EAAU,cAC/B,EAAK,sBAAwB,EAAU,sBACvC,AAAI,EAAU,gBACZ,IAAA,KAAK,OACD,EAAU,WAAa,KACvB,WACI,MAAA,mEACR,EAAK,gBAAkB,GACvB,EAAK,UAAY,KAEjB,EAAK,UAAY,EAAU,UAAY,EAAU,UAAY,MA9H3D,SAAA,UAAA,YAAN,2GACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEF,MAAA,CAAA,EAAO,KAAK,sBAAwB,OAAO,KAAK,KAAK,eACjB,KAAK,uBAW7B,EAAA,UAAA,eAAd,8HAC8B,MAAA,CAAA,EAAM,KAAK,8BACvC,GADM,EAAsB,EAAA,OACxB,CAAC,KAAK,iBAAmB,CAAC,EAE5B,KAAM,IAAI,OACN,6DA0BN,GAzBO,AAAI,KAAK,iBAAmB,GAEjC,GAAA,KAAK,OACD,EAAoB,SAAW,KAAK,gBAAgB,OACpD,WAAM,MAAA,uCACF,EAAK,gBAAgB,OAAO,WAC5B,kEACW,EAAoB,OAAO,WAAa,OAE7D,AAAK,KAAK,iBACR,MAAK,gBAAkB,GAGnB,EAAkC,KAAK,gBAAgB,OACzD,SAAC,EAAmC,GAClC,SAAS,GAAS,EAAS,GAAQ,GAAM,EAClC,GAET,IACE,EACF,OAAO,KAAK,GAAQ,OAAO,SAAC,GAAS,MAAC,GAAO,GAAQ,IACzD,GAAA,KAAK,OACD,EAAe,SAAW,EAC1B,WAAM,MAAA,iCAAmC,EAAe,aAExD,KAAK,eACP,IAAA,EAAA,EAAkB,EAAA,OAAO,KAAK,KAAK,eAAjB,EAAA,EAAA,OAAA,IAEhB,GAFS,EAAG,EAAA,GACN,EAAQ,KAAK,gBAAgB,QAAQ,GACvC,IAAU,GACZ,KAAM,IAAI,OACN,YAAc,EACd,uEACY,KAAK,gBAAgB,WAAa,MAIxD,YAAK,qBAAuB,aAGhB,EAAA,UAAA,oBAAd,uHACM,MAAK,UACM,CAAA,EAAM,KAAK,KAAK,YAD3B,CAAA,EAAA,UACI,SAAO,EAAA,OACQ,CAAA,EAAM,EAAK,eAChC,GADM,EAAe,EAAA,OACjB,EAAa,KACf,KAAM,IAAI,OAAM,sCAEZ,SAAoB,EAAa,MACjC,EAAU,KAAK,SAAS,EAAW,IACzC,CAAA,EAAO,UAEP,MAAA,CAAA,EAAO,YA0DL,EAAA,UAAA,SAAN,wHACO,MAAK,qBAAN,CAAA,EAAA,GACF,CAAA,EAAM,KAAK,yBAAX,EAAA,wBAEU,MAAA,CAAA,EAAM,KAAK,KAAK,mBAAxB,SAAQ,EAAA,OACZ,AAAI,KAAK,WAGP,GAAQ,EAAM,KAAK,IAErB,CAAA,EAAO,EAAM,IAAI,SAAA,GAAK,MAAA,GAAK,gBAAgB,YAG7C,EAAA,UAAA,gBAAA,SAAgB,GAKd,OAJM,GAAS,KAAK,SAAS,GACvB,EAA6C,GAC7C,EAA2C,GAExC,EAAI,EAAG,EAAI,KAAK,gBAAgB,OAAQ,KAC/C,GAAM,GAAM,KAAK,gBAAgB,GAC3B,EAAS,KAAK,cAAgB,KAAK,cAAc,GAAO,KAC9D,GAAI,KAAK,uBAAyB,CAAC,EAEjC,SAEA,GAAM,GAAQ,EAAO,GACjB,EAAc,KAClB,GAAI,IAAU,GAGZ,GAAI,GAAU,EAAO,UAAY,OAC/B,EAAc,EAAO,aAChB,GAAI,GAAW,GAAO,UAAY,EAAO,SAC9C,KAAM,IAAI,OACN,mBAAmB,EAAG,2BAA2B,GAErD,EAAc,YAIhB,GAAM,GAAa,OAAO,GAC1B,GAAI,MAAM,GAGR,AAAI,GAAU,EAAO,QAAU,OAC7B,EAAc,KAAK,WAAW,GAG9B,EAAc,UAEP,CAAC,GAAU,CAAC,EAAO,MAG5B,EAAc,MAId,QAAQ,EAAO,WACR,UACH,EAAc,EACd,UACG,QACH,EAAc,KAAK,MAAM,GACzB,UACG,OACH,EAAc,KAAK,WAAW,GAC9B,cAEA,EAAc,GAKtB,AAAC,GAAU,EAAO,QAAW,EAAO,GAAO,EACd,EAAS,GAAO,EAKjD,MAAI,QAAO,KAAK,GAAQ,SAAW,EAC1B,EAGA,CAAC,GAAI,EAAU,GAAI,IAItB,EAAA,UAAA,WAAR,SAAmB,GACjB,MAAI,KAAU,KAAO,EAAM,gBAAkB,OACpC,EAEA,GAKH,EAAA,UAAA,SAAR,SAAiB,EAAc,GAAA,AAAA,IAAA,QAAA,GAAA,IAM7B,OALM,GAAmB,GACrB,EAAa,EACX,EAAa,EAAK,OACpB,EAAe,GAEV,EAAI,EAAG,EAAI,EAAY,IAC9B,OAAQ,OAED,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAa,EAAI,EACjB,EAAe,GACf,UAEG,MAAK,UAIR,GAHA,EAAa,EAAI,EAGb,KAAK,YAAc,KAAO,KAAK,gBACjC,MAEF,EAAO,KAAK,IACZ,EAAe,GACf,cAGA,EAAe,GACf,EAAa,EACb,MAEJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,IACvC,EAAe,GACf,EAAa,EAAI,EACjB,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,UAEG,IACH,OAAQ,EAAK,OAAO,QAEb,MAAK,UACR,EAAO,KAAK,EAAK,UAAU,EAAY,EAAI,IAC3C,EAAe,GACf,EAAa,EAAI,EACjB,UAEG,IACH,EAAe,GACf,cAGA,EAAe,GACf,MAEJ,UACG,IACH,OAAQ,EAAK,OAAO,QAEb,IACH,EAAe,GACf,MAGJ,MAWN,GANA,AAAI,IAAiB,GACnB,EAAO,KAAK,EAAK,UAAU,EAAY,EAAa,IAEpD,EAAO,KAAK,EAAK,UAAU,IAGzB,GAAwB,EAAO,SAAW,KAAK,gBAAgB,OACjE,KAAM,IAAI,OAAM,wCACZ,KAAK,gBAAgB,OAAM,+BAA+B,GAEhE,MAAO,IAEX,GA1VgC,IC7ChC,AA4BA,GAAA,IAAA,SAAA,GAAwC,GAAA,EAAA,GAgBtC,WAAuC,GAAvC,GAAA,GACE,EAAA,KAAA,OAAO,KAD8B,EAAA,iBAAA,EAf/B,EAAA,SAAW,GAiBjB,EAAK,QAAU,EAAiB,SAAW,KAC3C,GAAM,GAAc,KAAK,KAAK,EAAK,SACnC,GAAI,EAAK,QAAU,GAAK,EAAc,GAAK,EAAc,IACrD,CAAC,OAAO,UAAU,GACpB,KAAM,IAAI,OACN,oDACA,gCAA+B,EAAK,UAc1C,GAXA,EAAK,UAAY,EAAiB,yBAA2B,GAC7D,EAAK,aAAe,EAAiB,aACrC,EAAK,qBACD,EAAiB,sBAAwB,EAAK,QAClD,EAAK,sBAAwB,EAAiB,sBAC9C,EAAK,sBAAwB,EAAiB,uBAAyB,EAEvE,EAAK,mBACD,IAAiB,qBAAuB,IAC5C,EAAK,gBACD,EAAiB,kBAAoB,GACrC,CAAC,EAAK,oBAAsB,CAAC,EAAK,gBACpC,KAAM,IAAI,OACN,iHAKR,SAAA,UAAA,QAAA,WACE,MAAO,cAII,EAAA,OAAb,SAAoB,GAAA,MAAA,KAAA,QAAA,GAAA,6FAClB,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGA,SAAqB,GAAI,GAAmB,GAGlD,CAAA,EAAM,EAAmB,gBAAzB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,8IAEI,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,KAAK,uBAAyB,KAAO,GACA,KAAK,sBACjD,MAAO,aAHT,SAAK,OAAS,EAAA,qCAMR,GAAI,OACN,iDAAiD,EAAE,gBAGzD,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,2CAQlB,GALM,EAED,OAAe,cAAiB,OAAe,mBACpD,KAAK,aAAe,GAAI,GAEpB,CAAC,KAAK,aAGR,KAAK,aAAe,KAAK,aAAa,mBAC7B,KAAK,aAAa,aAAe,KAAK,aAC/C,KAAM,IAAI,OACN,8BACA,cAAa,KAAK,aAAY,MAC9B,YAAW,KAAK,aAAa,aAG7B,SAAe,KAAK,aAAa,wBAAwB,KAAK,QACpE,KAAK,SAAW,KAAK,aAAa,iBAClC,KAAK,SAAS,QAAU,KAAK,QAAU,EACvC,KAAK,SAAS,sBAAwB,KAAK,sBAC3C,EAAa,QAAQ,KAAK,UAC1B,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,KAAK,SAAW,GAAI,cAAa,KAAK,SACtC,CAAA,SAGI,EAAA,UAAA,KAAN,mHACE,MAAI,MAAK,SACP,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAMN,CAAA,EAAM,KAAK,uBAA5B,SAAiB,EAAA,OACvB,AAAI,KAAK,oBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAoB,KAAK,4BACrB,EAAU,CAAC,KAAK,UAAW,KAAK,qBAAsB,KAE5D,AAAI,KAAK,iBACD,GAAW,KAAK,aAAa,EAAe,eAClD,EAAiB,KAAK,4BAClB,EAAU,CAAC,KAAK,UAAY,KAAK,QAAS,KAGhD,CAAA,EAAO,CACL,MAAO,CAAC,YAAe,EAAmB,SAAY,GACtD,KAAM,WAMJ,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAIf,EAAA,UAAA,aAAd,+FAEQ,SAAgC,GAChC,EAAgC,GAClC,EAAgB,EACpB,CAAA,EAAO,GAAI,SAAQ,SAAA,GACjB,GAAM,GAAa,YAAY,WAC7B,AAAI,EAAK,oBACP,GAAK,SAAS,sBAAsB,EAAK,UAEzC,AAAI,EAAK,SAAS,KAAO,WACvB,EAAQ,CAAC,cAAa,EAAE,cAAa,IAEvC,EAAc,KAAK,EAAK,SAAS,MAAM,EAAG,EAAK,wBAEjD,AAAI,EAAK,iBACP,GAAK,SAAS,uBAAuB,EAAK,UAC1C,EAAc,KAAK,EAAK,SAAS,UAInC,AAAI,EAAE,IAAkB,EAAK,WAC3B,eAAc,GACd,EAAQ,CAAC,cAAa,EAAE,cAAa,MAEtC,EAAK,QAAU,EAAK,aAAe,aAK1C,EAAA,UAAA,KAAA,WACE,AAAK,KAAK,UACR,MAAK,SAAW,GAChB,KAAK,SAAS,aACd,KAAK,aAAa,QAClB,AAAI,KAAK,QAAU,MAAQ,KAAK,OAAO,YAAY,OAAS,GAC1D,KAAK,OAAO,YAAY,GAAG,SAMjC,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAIlB,EAAA,UAAA,cAAA,WACE,MAAO,MAAK,cAGN,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAY,EAAM,GAAG,OACrB,EAAW,GAAI,cAAa,EAAM,OAAS,GACjD,SAAM,QAAQ,SAAC,EAAM,GAAM,MAAA,GAAS,IAAI,EAAM,EAAI,KAC3C,GAGD,EAAA,UAAA,4BAAR,SAAoC,EAAwB,GAE1D,GAAM,GAAO,GAAI,cAAa,GAAA,KAAK,cAAc,IAEjD,SAAK,IAAI,EAAU,EAAK,OAAS,EAAS,QACnC,GAAA,OAAO,EAAM,IAExB,GA7MwC,IC5BxC,AA0BA,GAAA,IAAA,SAAA,GAAoC,GAAA,EAAA,GAQlC,WACuB,EACA,GAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KACP,GAHqB,EAAA,mBAAA,EACA,EAAA,aAAA,EATf,EAAA,SAAW,GAEX,EAAA,OAAS,GASX,EAAK,eAKP,GAJA,EAAK,OAAS,GACd,EAAK,SACD,CAAC,EAAK,aAAa,aAAc,EAAK,aAAa,aACvD,EAAK,WAAa,GAAA,SAAS,CAAC,GAAI,SAC5B,EAAK,aAAa,YAEpB,GAAM,GACF,EAAK,aAAa,YAAc,EAAM,EAAK,mBAAmB,MAC5D,EAAsB,EAAK,aAAa,aAAe,EACzD,EAAK,mBAAmB,OACtB,EAAkB,GAAI,GAAsB,EAC5C,EAAmB,GAAI,GAAuB,EAC9C,EAAe,EAAiB,EAChC,EAAgB,EAAsB,EAC5C,EAAK,QAAU,GAAA,SACX,CAAC,EAAiB,EAAgB,EAAe,GACjD,CAAC,EAAG,QAER,GAAK,QAAU,GAAA,SAAS,CAAC,EAAG,EAAG,EAAG,GAAI,CAAC,EAAG,aAKhD,SAAA,UAAA,QAAA,WACE,MAAO,UAII,EAAA,OAAb,SACI,EAAuC,GAAA,MAAA,KAAA,QAAA,GAAA,6FACzC,GAAI,GAAA,MAAM,IAAI,WACZ,KAAM,IAAI,OACN,4DAGN,GAAI,CAAC,GAIH,GADA,EAAqB,SAAS,cAAc,SACxC,CAAC,EAAa,aAAe,CAAC,EAAa,aAC7C,KAAM,IAAI,OACN,0GAGN,EAAmB,MAAQ,EAAa,YACxC,EAAmB,OAAS,EAAa,aAErC,SAAiB,GAAI,GAAe,EAAoB,GAG9D,CAAA,EAAM,EAAe,gBAArB,SAAA,OAEA,CAAA,EAAO,SAIH,EAAA,UAAA,MAAN,oHACE,AAAI,KAAK,aAAa,YACpB,GAAA,KAAK,OACA,KAAK,aAAa,aAAe,QAC7B,KAAK,aAAa,aAAe,cACtC,WACI,MAAA,+BAA+B,EAAK,aAAa,WAAU,2FAKnE,EAAA,KAAc,CAAA,EAAM,UAAU,aAAa,aAAa,CACtD,MAAO,CACL,SAAU,KAAK,aAAa,SAC5B,WAAY,KAAK,aAAa,WAC1B,KAAK,aAAa,WAClB,OACJ,MAAO,KAAK,mBAAmB,MAC/B,OAAQ,KAAK,mBAAmB,kBAPpC,SAAK,OAAS,EAAA,qCAYd,EAAE,QAAU,iDAAiD,EAAE,QACzD,SAGR,GAAI,CAAC,KAAK,OACR,KAAM,IAAI,OAAM,uCAIlB,IACE,KAAK,mBAAmB,UAAY,KAAK,aAClC,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,OAAO,IAAI,gBAAgB,KAAK,QAGhE,YAAK,mBAAmB,OAExB,KAAK,SAAW,GAEhB,CAAA,EAAO,GAAI,SAAc,SAAA,GAEvB,EAAK,mBAAmB,iBAAmB,WACzC,cAKA,EAAA,UAAA,KAAN,oFACE,GAAI,KAAK,SACP,MAAA,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAI7B,IACE,EAAM,GAAA,QAAQ,WAAW,KAAK,0BACvB,GACP,KAAM,IAAI,OACN,4CAA4C,KAAK,UAAU,IAEjE,GAAI,KAAK,OACP,IACE,MAAA,CAAA,EAAO,CAAC,MAAO,KAAK,mBAAmB,GAAM,KAAM,WAC5C,GACP,KAAM,IAAI,OAAM,oCAAoC,EAAE,iBAEtD,EAAI,cAGN,OAAA,CAAA,EAAO,CAAC,MAAO,EAAK,KAAM,oBAItB,EAAA,UAAA,aAAR,WAIE,MAAI,QAAK,aAAa,aAAe,KAAK,aAAa,cAClD,MAAK,mBAAmB,QAAU,KAAK,aAAa,aACpD,KAAK,mBAAmB,SAAW,KAAK,aAAa,gBAO5D,EAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,IAAA,KAAK,WACV,GAAM,GAA0B,EAAI,UAAU,WAAW,GACrD,EACJ,EAAe,GAAA,MAAM,cACjB,EAAe,EAAK,QAAS,EAAK,WAAY,EAAK,SACnD,YAEJ,GAAM,GAAQ,EAAa,MAC3B,MAAO,GAAa,QAAQ,EAAM,MAAM,OAMtC,EAAA,UAAA,QAAN,qGACU,MAAA,CAAA,EAAM,KAAK,eAAnB,MAAA,CAAA,EAAQ,EAAA,OAAmB,aAI7B,EAAA,UAAA,KAAA,WACE,GAAM,GAAS,KAAK,OAAO,YAE3B,EAAO,QAAQ,SAAA,GAAS,MAAA,GAAM,SAE9B,IACE,KAAK,mBAAmB,UAAY,WAC7B,GACP,QAAQ,IAAI,GACZ,KAAK,mBAAmB,IAAM,KAEhC,KAAK,SAAW,IAIlB,EAAA,UAAA,QAAA,WACE,KAAM,IAAI,OAAM,oDAEpB,GApMoC,IC1BpC,AA2BA,GAAA,IAAA,WAAA,cAUA,MAAA,MCrCA,AAoBA,GAAA,IAAA,SAAA,GAA6C,GAAA,EAAA,GAA7C,4DAmBE,SAAA,UAAA,MAAA,SAAM,GACJ,MAAO,IAAI,IAAc,KAAM,IAEnC,GAtB6C,IAmC7C,GAAA,SAAA,GAA4B,GAAA,EAAA,GAG1B,WAAsB,EAAgC,GAAtD,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAkB,EAAU,KAG9C,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf4B,IAiB5B,GAAA,SAAA,GAAgC,GAAA,EAAA,GAI9B,WACc,EAA0C,GADxD,GAAA,GAEE,EAAA,KAAA,OAAO,KADK,SAAA,SAAA,EAA0C,EAAA,UAAA,EAHxD,EAAA,UAAY,KAOZ,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,cAAc,KAAK,UAAS,MAGzD,EAAA,UAAA,KAAN,mHACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eACxC,GADM,EAAc,EAAA,OAChB,EAAY,KACd,MAAI,MAAK,YAAc,GACrB,CAAA,EAAO,IAKT,MAAK,YAAY,KAAK,KAAK,WAC3B,KAAK,UAAY,GACjB,CAAA,EAAO,KAQT,IANM,EAAQ,EAAY,MAAM,MAAM,KAAK,WAK3C,EAAM,GAAK,KAAK,UAAY,EAAM,GAClC,EAAA,EAAmB,EAAA,EAAM,MAAM,EAAG,IAAf,EAAA,EAAA,OAAA,IAAR,EAAI,EAAA,GACb,KAAK,YAAY,KAAK,GAExB,YAAK,UAAY,EAAM,EAAM,OAAS,GAEtC,CAAA,EAAO,UAEX,GAvCgC,ICxEhC,AAsBA,GAAA,IAAA,SAAA,GAAgD,GAAA,EAAA,GAAhD,4DAUE,SAAA,UAAA,WAAA,WACE,MAAO,IAAI,IAAa,OAE5B,GAbgD,IA0BhD,GAAA,SAAA,GAA2B,GAAA,EAAA,GAGzB,WAAsB,GAAtB,GAAA,GACE,EAAA,KAAA,OAAO,KADa,SAAA,SAAA,EAEpB,EAAK,KAAO,GAAI,IAAiB,KAGnC,SAAA,UAAA,QAAA,WACE,MAAO,MAAK,KAAK,WAGb,EAAA,UAAA,KAAN,8EACE,MAAA,CAAA,EAAO,KAAK,KAAK,aAErB,GAf2B,IAuC3B,GAAA,SAAA,GAA+B,GAAA,EAAA,GAM7B,WAA+B,GAA/B,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GAF6B,EAAA,SAAA,EAEzB,GAAA,MAAM,IAAI,cACZ,EAAK,QAAU,GAAI,aAAY,cAGxB,GAAA,GAAA,KAAA,cACP,EAAK,QAAU,GAAI,GAAc,iBAGrC,SAAA,UAAA,QAAA,WACE,MAAU,MAAK,SAAS,UAAS,YAG7B,EAAA,UAAA,KAAN,+GACsB,MAAA,CAAA,EAAM,KAAK,SAAS,eAExC,MAFM,GAAc,EAAA,OAEhB,EAAY,KACd,CAAA,EAAO,IAEP,GAAQ,EAAY,MAItB,AAAI,GAAA,MAAM,IAAI,cACZ,EAAO,KAAK,QAAQ,OAAO,EAAO,CAAC,OAAQ,KAE3C,EAAO,KAAK,QAAQ,MAAM,OAAO,KAAK,EAAM,SAE9C,KAAK,YAAY,KAAK,GACtB,CAAA,EAAO,WAEX,GAtC+B,ICvF/B,AAqCA,GAAA,IAAA,SAAA,GAAuC,GAAA,EAAA,GAIrC,WACc,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFd,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,KAAA,EACA,EAAA,QAAA,EAEZ,GAAA,KAAK,OACA,YAAgB,aACZ,IAAA,MAAM,IAAI,cACL,YAAgB,OAAQ,YAAgB,MACzC,IACT,WAAM,MAAA,yEAEV,EAAK,OAAS,EAAQ,QAAU,EAEhC,EAAK,UAAY,EAAQ,WAAa,KAAO,OAG/C,SAAA,UAAA,QAAA,WACE,MAAO,cAAc,KAAK,MAGtB,EAAA,UAAA,KAAN,oHACE,MAAI,MAAK,QAAY,MAAK,eAAgB,YAClB,KAAK,KAAK,WACV,KAAK,KAAK,MAChC,CAAA,EAAO,CAAC,MAAO,KAAM,KAAM,KAEvB,GAAQ,GAAI,SAAoB,SAAC,EAAS,GAC9C,GAAM,GAAM,EAAK,OAAS,EAAK,UAC/B,GAAI,EAAK,eAAgB,YAGvB,EAAQ,GAAI,YAAW,EAAK,KAAK,MAAM,EAAK,OAAQ,UAMpD,GAAM,GAAa,GAAI,YACvB,EAAW,OAAS,SAAC,GACnB,GAAI,GAAsC,EAAW,OAOrD,GAHA,AAAI,YAAgB,cAClB,GAAO,GAAI,YAAW,IAEpB,CAAE,aAAgB,aACpB,MAAO,GAAO,GAAI,WAAU,sCAE9B,EAAQ,IAEV,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,aAE1B,EAAW,QAAU,SAAC,GACpB,MAAO,GAAO,GAAI,OAAM,EAAM,QAIhC,GAAM,GAAQ,EAAK,KAAK,MAAM,EAAK,OAAQ,GAG3C,EAAW,kBAAkB,GAE/B,EAAK,OAAS,SAEA,CAAA,EAAM,WAAtB,MAAA,CAAA,EAAQ,GAAA,MAAQ,EAAA,OAAc,EAAA,KAAM,GAAK,UAE7C,GAxEuC,ICrCvC,YA6BI,EAAkB,GAAA,MAAA,KAAA,QAAA,GAAA,qGAGpB,MAAK,OAAO,IAAS,SACnB,EAAY,EAEZ,GAAa,EAAgB,IAC7B,EAAc,GAA0B,IAEzB,CAAA,EAAM,GAAA,KAAK,MAAM,EAAW,iBAAvC,GAAW,EAAA,OACb,EAAS,MACY,WAAU,KAAC,CAAA,EAAM,EAAS,gBAD/C,CAAA,EAAA,UACI,SAAa,GAAA,GAAA,MAAI,WAAU,CAAA,OAAC,EAAA,UAClC,CAAA,EAAO,GAAI,IAAkB,EAAY,WAEzC,KAAM,IAAI,OAAM,EAAS,iBAK7B,GAAM,IAA4B,SAAC,GACjC,GAAM,GAAO,CACX,OAAQ,EAAQ,OAChB,QAAS,EAAQ,QACjB,KAAM,EAAQ,KACd,KAAM,EAAQ,KACd,YAAa,EAAQ,YACrB,MAAO,EAAQ,MACf,SAAU,EAAQ,SAClB,SAAU,EAAQ,SAClB,UAAW,EAAQ,WAErB,MAAO,IC5DT,YAqB4B,GAC1B,MAAQ,OAAO,IAAW,UAAa,EAAO,OAAO,EAAG,KAAO,UCtBjE,mBA6BoC,GAAA,EAAA,GASlC,WACc,EACS,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFK,SAAA,MAAA,EACS,EAAA,QAAA,IAIjB,SAAA,UAAA,SAAN,oFACE,MAAI,IAAY,KAAK,QAAU,GAAA,MAAM,IAAI,YAEjC,GAAK,cACX,KAAK,MAAQ,EAAG,aAAc,KAAK,MAAiB,OAAO,KAI7D,CAAA,EAAO,GAAI,IAAkB,KAAK,MAAsB,KAAK,eAEjE,GAzBoC,IC7BpC,mBA4BmC,GAAA,EAAA,GAQjC,WACuB,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IAFvB,GAAA,GAGE,EAAA,KAAA,OAAO,KAFc,SAAA,IAAA,EACA,EAAA,YAAA,IAQjB,SAAA,UAAA,SAAN,8EACE,MAAI,IAAY,KAAK,KACnB,CAAA,EAAQ,GAAI,IAAe,KAAK,IAAe,KAAK,aAC/C,YAEL,CAAA,EAAO,GAAiB,KAAK,IAAK,KAAK,mBAG7C,GA1BmC,IC5BnC,YA0GI,EAAqB,GAAA,MAAA,KAAA,QAAA,GAAA,IAChB,GAAI,IAAW,GAAI,IAAc,GAAS,eA4B/C,GADJ,GAAA,GAAA,KAEQ,EAAO,GAAqB,GAClC,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,WAAA,MAAA,IAAA,KAAA,SAAA,GAAY,MAAA,CAAA,EAAA,qBAgEvC,GADJ,GAAA,GAAA,KAEE,MAAO,IAAsB,WAAA,MAAA,IAAA,EAAA,OAAA,OAAA,mEACf,MAAA,CAAA,EAAM,YAAZ,SAAM,EAAA,OACZ,CAAA,EAAO,GAAqB,WAAM,MAAA,GAAI,6BAmCtC,EACA,sEACF,MAAA,CAAA,EAAO,GAAe,OAAO,EAAoB,oBAqClB,sEAE/B,MAAA,CAAA,EAAO,GAAmB,OAAO,QCxRnC,GAGM,IAAU,oNCHhB,oBA2BA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAc,GACZ,GAAI,GAAK,KAAM,EAAO,IAEtB,EAAG,KAAO,WACR,GAAI,GAAI,QAAU,EAAG,GAAK,EAAG,EAAI,sBACjC,SAAG,GAAK,EAAG,GACX,EAAG,GAAK,EAAG,GACJ,EAAG,GAAK,EAAK,GAAG,EAAI,EAAI,IAIjC,EAAG,EAAI,EACP,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,GAAK,EAAK,KACb,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAG,IAAM,EAAK,GACd,AAAI,EAAG,GAAK,GAAK,GAAG,IAAM,GAC1B,EAAO,KAGT,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACT,EAAE,GAAK,EAAE,GACF,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAK,GACd,EAAQ,GAAQ,EAAK,MACrB,EAAO,EAAG,KACd,SAAK,MAAQ,WAAa,MAAQ,GAAG,OAAS,WAAe,GAC7D,EAAK,OAAS,WACZ,MAAO,KAAU,KAAS,QAAW,GAAK,uBAE5C,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,aACE,GAAI,GAAI,WAEJ,EAAO,SAAS,GAClB,EAAO,EAAK,WACZ,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,KAC/B,GAAK,EAAK,WAAW,GACrB,GAAI,GAAI,mBAAsB,EAC9B,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EACL,EAAI,IAAM,EACV,GAAK,EACL,GAAK,EAAI,WAEX,MAAQ,KAAM,GAAK,uBAGrB,MAAO,GAIT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,KAAO,IAIZ,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9GnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAEzB,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGP,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAK,EAAG,GAAK,GACxB,SAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACV,EAAG,EAAI,EAAG,EACH,EAAG,GAAM,EAAG,IAAM,GAAM,EAAK,IAAM,GAG5C,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC7EnC,oBAGA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAK,EAAG,EAAK,EAAG,IAAM,EAC1B,SAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAI,EAAG,EACzC,GAAG,EAAK,EAAG,EAAI,OAAS,GAC5B,GAAG,EAAK,EAAG,EAAK,EAAG,GAAK,EAAO,GAAK,GAAK,IAAO,GAGtD,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAEP,AAAI,IAAU,GAAO,GAEnB,EAAG,EAAI,EAGP,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,AAAI,GAAK,EAAQ,QACf,GAAG,EAAI,EAAG,GAAK,GAAK,EAAG,IAAM,GAE/B,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UClFnC,oBAKA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WAER,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAAG,EAC9B,SAAI,EAAE,GAAI,GAAM,IAAM,EAAI,EAAI,EAAK,GAAK,GACxC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,GACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,IAAM,EACpC,EAAI,EAAG,EAAI,EAAK,GAAI,GAAK,EAAK,GAAK,EACnC,EAAI,EAAG,EAAI,EAAK,GAAI,EAAI,EAAK,GAAK,GAAK,GAAK,EAAK,GAAK,EACtD,EAAE,GAAK,EACP,EAAG,EAAK,EAAI,EAAK,EACV,GAGT,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAI,GAEd,GAAI,IAAU,GAAO,GAEnB,EAAI,EAAE,GAAK,MAIX,KADA,EAAO,GAAK,EACP,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC7B,EAAE,EAAI,GAAM,EAAE,EAAI,IAAM,GACnB,EAAK,WAAW,GAAK,EAAG,EAAI,EAAK,IAAM,GAIhD,KAAO,EAAE,OAAS,GAAG,EAAE,KAAK,GAC5B,IAAK,EAAI,EAAG,EAAI,GAAK,EAAE,KAAO,EAAG,EAAE,EAAE,CAOrC,IANA,AAAI,GAAK,EAAG,EAAI,EAAE,GAAK,GAAS,EAAI,EAAE,GAEtC,EAAG,EAAI,EACP,EAAG,EAAI,EAGF,EAAI,IAAK,EAAI,EAAG,EAAE,EACrB,EAAG,OAIP,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EAAE,QACV,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,UAAY,IAIjB,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UC9FnC,oBAyBA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAGT,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAG,EAE3B,SAAG,EAAI,EAAK,EAAI,WAAc,EAE9B,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GAEX,EAAI,EAAE,GAAK,EAAI,EACf,EAAG,EAAI,EAEC,EAAK,GAAK,IAAM,IAAQ,GAGlC,WAAc,EAAI,GAChB,GAAI,GAAG,EAAG,EAAG,EAAG,EAAG,EAAI,GAAI,EAAQ,IAYnC,IAXA,AAAI,IAAU,GAAO,GAEnB,GAAI,EACJ,EAAO,MAGP,GAAO,EAAO,KACd,EAAI,EACJ,EAAQ,KAAK,IAAI,EAAO,EAAK,SAG1B,EAAI,EAAG,EAAI,IAAK,EAAI,EAAO,EAAE,EAEhC,AAAI,GAAM,IAAK,EAAK,WAAY,GAAI,IAAM,EAAK,SAE/C,AAAI,IAAM,GAAG,GAAI,GACjB,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,GAAK,EACV,GAAK,IAAM,GACX,AAAI,GAAK,GACP,GAAK,EAAI,WAAc,EACvB,EAAK,EAAE,EAAI,MAAS,EAAI,EACxB,EAAK,AAAK,GAAL,EAAU,EAAI,EAAI,GAW3B,IAPA,AAAI,GAAK,KACP,GAAG,IAAQ,EAAK,QAAU,GAAK,KAAO,IAKxC,EAAI,IACC,EAAI,EAAI,IAAK,EAAI,EAAG,EAAE,EACzB,EAAI,EAAG,EAAI,GAAM,KACjB,EAAI,EAAE,EAAM,EAAI,EAAK,KACrB,GAAK,GAAK,GACV,GAAK,GAAK,GACV,GAAK,IAAM,GACX,GAAK,IAAM,GACX,EAAE,GAAK,EAAI,EAGb,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,EAGT,EAAK,EAAI,GAGX,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAGT,WAAc,EAAM,GAClB,AAAI,GAAQ,MAAM,GAAO,CAAE,GAAI,OAC/B,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,EAAM,GAAG,EAAK,EAAO,GACzB,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,QAAU,IAIf,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UChJnC,oBAIA,AAAC,UAAS,EAAQ,EAAQ,GAE1B,WAAgB,GACd,GAAI,GAAK,KAAM,EAAU,GAGzB,EAAG,KAAO,WACR,GAAI,GAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACzC,SAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAK,GAAK,GAAO,IAAM,EAAK,EAC5B,EAAK,EAAI,EAAK,EACd,EAAG,EAAI,EAAK,GAAK,GAAO,IAAM,GAAM,EACpC,EAAG,EAAI,EAAK,EAAI,EAAK,EACrB,EAAG,EAAK,GAAK,GAAO,IAAM,GAAM,EACzB,EAAG,EAAK,EAAI,EAAK,GAmB1B,EAAG,EAAI,EACP,EAAG,EAAI,EACP,EAAG,EAAI,WAAa,EACpB,EAAG,EAAI,WAEP,AAAI,IAAS,KAAK,MAAM,GAEtB,GAAG,EAAK,EAAO,WAAe,EAC9B,EAAG,EAAI,EAAO,GAGd,GAAW,EAIb,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAS,GAAI,IACvC,EAAG,GAAK,EAAQ,WAAW,GAAK,EAChC,EAAG,OAIP,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACD,EAGT,WAAc,EAAM,GAClB,GAAI,GAAK,GAAI,GAAO,GAChB,EAAQ,GAAQ,EAAK,MACrB,EAAO,WAAa,MAAQ,GAAG,SAAW,GAAK,YACnD,SAAK,OAAS,WACZ,EACE,IAAI,GAAM,EAAG,SAAW,GACpB,EAAO,GAAG,SAAW,GAAK,WAC1B,EAAU,GAAM,GAAQ,IAAK,UAC1B,IAAW,GACpB,MAAO,IAET,EAAK,MAAQ,EAAG,KAChB,EAAK,MAAQ,EACb,AAAI,GACF,CAAI,MAAO,IAAU,UAAU,EAAK,EAAO,GAC3C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAI,MAErC,EAGT,AAAI,GAAU,EAAO,QACnB,EAAO,QAAU,EACZ,AAAI,GAAU,EAAO,IAC1B,EAAO,WAAa,MAAO,KAE3B,KAAK,OAAS,IAId,GACC,MAAO,KAAW,UAAY,GAC9B,MAAO,SAAW,YAAc,UCnGnC,oBAwBA,AAAC,UAAU,EAAM,GAIjB,GAAI,GAAS,KACT,EAAQ,IACR,EAAS,EACT,EAAS,GACT,EAAU,SACV,EAAa,EAAK,IAAI,EAAO,GAC7B,EAAe,EAAK,IAAI,EAAG,GAC3B,EAAW,EAAe,EAC1B,EAAO,EAAQ,EACf,EAMJ,WAAoB,EAAM,EAAS,GACjC,GAAI,GAAM,GACV,EAAW,GAAW,GAAQ,CAAE,QAAS,IAAU,GAAW,GAG9D,GAAI,GAAY,EAAO,EACrB,EAAQ,QAAU,CAAC,EAAM,EAAS,IACjC,GAAgB,IAAmB,GAAI,GAGtC,EAAO,GAAI,GAAK,GAIhB,EAAO,WAIT,OAHI,GAAI,EAAK,EAAE,GACX,EAAI,EACJ,EAAI,EACD,EAAI,GACT,EAAK,GAAI,GAAK,EACd,GAAK,EACL,EAAI,EAAK,EAAE,GAEb,KAAO,GAAK,GACV,GAAK,EACL,GAAK,EACL,KAAO,EAET,MAAQ,GAAI,GAAK,GAGnB,SAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,GAC7C,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAE,GAAK,YAC7C,EAAK,OAAS,EAGd,EAAO,EAAS,EAAK,GAAI,GAGjB,GAAQ,MAAQ,GACpB,SAAS,EAAM,EAAM,EAAc,GAUjC,MATA,AAAI,IAEF,CAAI,EAAM,GAAK,EAAK,EAAO,GAE3B,EAAK,MAAQ,WAAa,MAAO,GAAK,EAAM,MAK1C,EAAgB,GAAK,GAAW,EAAa,GAIrC,IAElB,EACA,EACA,UAAY,GAAU,EAAQ,OAAU,MAAQ,EAChD,EAAQ,OAEV,EAAK,OAAS,GAAW,EAYzB,WAAc,GACZ,GAAI,GAAG,EAAS,EAAI,OAChB,EAAK,KAAM,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,EAAG,EAAI,GAMtD,IAHA,AAAK,GAAU,GAAM,CAAC,MAGf,EAAI,GACT,EAAE,GAAK,IAET,IAAK,EAAI,EAAG,EAAI,EAAO,IACrB,EAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,EAAI,EAAI,GAAW,GAAI,EAAE,KAClD,EAAE,GAAK,EAIT,AAAC,GAAG,EAAI,SAAS,GAIf,OAFI,GAAG,EAAI,EACP,EAAI,EAAG,EAAG,EAAI,EAAG,EAAG,EAAI,EAAG,EACxB,KACL,EAAI,EAAE,EAAI,EAAQ,EAAI,GACtB,EAAI,EAAI,EAAQ,EAAE,EAAS,GAAE,GAAK,EAAE,EAAI,EAAQ,EAAI,IAAQ,GAAE,GAAK,IAErE,SAAG,EAAI,EAAG,EAAG,EAAI,EACV,IAIN,GAOL,WAAc,EAAG,GACf,SAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EACR,EAAE,EAAI,EAAE,EAAE,QACH,EAOT,WAAiB,EAAK,GACpB,GAAI,GAAS,GAAI,EAAO,MAAO,GAAM,EACrC,GAAI,GAAS,GAAO,SAClB,IAAK,IAAQ,GACX,IAAM,EAAO,KAAK,EAAQ,EAAI,GAAO,EAAQ,UAAc,IAG/D,MAAQ,GAAO,OAAS,EAAS,GAAO,SAAW,EAAM,EAAM,KAQjE,WAAgB,EAAM,GAEpB,OADI,GAAa,EAAO,GAAI,EAAO,EAAI,EAChC,EAAI,EAAW,QACpB,EAAI,EAAO,GACT,EAAS,IAAS,EAAI,EAAO,GAAK,IAAM,EAAW,WAAW,KAElE,MAAO,GAAS,GAQlB,aACE,IACE,GAAI,GACJ,MAAI,IAAe,GAAM,EAAW,aAElC,EAAM,EAAI,GAEV,GAAM,GAAI,YAAW,GACpB,GAAO,QAAU,EAAO,UAAU,gBAAgB,IAE9C,EAAS,SACT,GACP,GAAI,GAAU,EAAO,UACjB,EAAU,GAAW,EAAQ,QACjC,MAAO,CAAC,CAAC,GAAI,MAAM,EAAQ,EAAS,EAAO,OAAQ,EAAS,KAQhE,WAAkB,GAChB,MAAO,QAAO,aAAa,MAAM,EAAG,GAgBtC,GANA,EAAO,EAAK,SAAU,GAMjB,MAAO,KAAW,UAAY,GAAO,SACxC,GAAO,QAAU,EAEjB,IACE,EAAa,WACN,SACJ,AAAK,OAAO,SAAW,YAAc,OAAO,KACjD,OAAO,WAAa,MAAO,OAK3B,GACA,QCrPF,oBAYA,GAAI,IAAO,KAKP,GAAS,KAKT,GAAS,KAQT,GAAY,KASZ,GAAU,KAOV,GAAS,KAIT,GAAK,KAET,GAAG,KAAO,GACV,GAAG,OAAS,GACZ,GAAG,OAAS,GACZ,GAAG,UAAY,GACf,GAAG,QAAU,GACb,GAAG,OAAS,GAEZ,GAAO,QAAU,6vDC3DjB,YAoBI,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MACF,GAAM,8DC7BpB,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBACvC,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAM/B,YACI,EAAyB,EAAW,EACpC,GACF,GAAI,IAAe,SACjB,MAAO,GAAQ,OAAO,GACjB,GAAI,IAAe,OACxB,MAAO,GAAQ,KAAK,GACf,GAAI,IAAe,MACxB,MAAO,GAAA,IAAO,GACT,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,GAChB,GAAI,IAAe,QACxB,MAAO,GAAQ,MAAM,EAAG,GAE1B,KAAM,IAAI,OACN,cAAc,EAAU,qEAeM,GAAA,EAAA,GAMlC,aAAA,GAAA,GACE,EAAA,KAAA,OAAO,KANF,SAAA,UAAY,GAGX,EAAA,SAAW,GAIjB,EAAK,KAAO,GAAI,GAAA,YAAY,EAAM,EAAA,YAGpC,SAAA,UAAA,MAAA,SAAM,EAAoC,EAAiB,GAEzD,AAAI,KAAK,UACP,MAAK,SAAW,GAChB,AAAI,EAAA,MAAM,IAAI,YACZ,EAAA,aAAa,KACT;;;gCAYR,GAAM,GAAS,GAEf,YAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,IAEzC,GAST,EAAA,UAAA,eAAA,SACI,EAAiB,EACjB,GACF,GAAM,GAAQ,KAAK,MAAM,EAAQ,EAAO,GAExC,MAAO,CAAC,OAAQ,EAAO,MAAK,EAAE,MAAK,IAIrC,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,YAIb,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GACjC,EAAW,aAIf,EAAA,UAAA,KAAA,SACI,EAAgB,EAAoC,EACpD,GACF,KAAK,KAAK,IAAI,EAAQ,CAAC,OAAM,EAAE,MAAK,EAAE,SAAU,KAGlD,EAAA,UAAA,WAAA,WACE,MAAO,MAAK,KAAK,cAGb,EAAA,UAAA,KAAN,SAAW,sEACT,MAAA,CAAA,EAAO,KAAK,SAAS,SAEvB,EAAA,UAAA,SAAA,SAAS,GACD,GAAA,GAAA,KAAA,KAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,mBAEd,GAAI,IAAU,aACZ,GAAM,GACF,KAAK,SAAS,EAAmB,KAAK,QACpC,EACF,KAAK,SAAS,EAAmB,KAAK,QAC1C,MAAO,GAAA,aAAa,uBAAuB,EAAY,GAGzD,MAAO,MAAK,KAAK,IAAI,GAAQ,QAGvB,EAAA,UAAA,WAAR,SAAmC,GACjC,GAAM,GAAO,KAAK,SAAS,EAAE,QACzB,EAAc,EAClB,GAAI,EAAE,QAAU,SACd,IAEE,EAAe,EAAsB,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,WAChE,GACA,KAAM,IAAI,OAAM,oDAGpB,MAAO,GAAA,OAAU,EAAE,MAAO,EAAE,MAAO,IAGrC,EAAA,UAAA,WAAA,SACI,EAAoC,EAAiB,GACvD,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAG7D,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,KAAK,IAAI,IACT,GAAA,GAAA,KAAA,KAAA,IAAA,GAAA,mBAEP,AAAI,GAAsB,MACxB,MAAK,YAAY,EAAmB,KAAK,QACzC,KAAK,YAAY,EAAmB,KAAK,SAG3C,KAAK,KAAK,OAAO,KAIrB,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,KAAK,IAAI,IAChB,GAAM,GAAa,KAAK,KAAK,IAAI,GAEjC,EAAW,WAEX,AAAI,EAAW,SAAW,GACxB,KAAK,YAAY,KAKjB,EAAA,UAAA,KAAN,SAAW,8EACH,SAAQ,EAAA,KAAK,MACnB,IACM,EAAW,EAAA,KAAK,MAAQ,EAC9B,CAAA,EAAO,CAAC,SAAQ,SAGlB,EAAA,UAAA,OAAA,WACE,MAAO,CAEL,WAAY,GACZ,QACI,CAAC,wHAKT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GACxC,GAAiB,EAAG,gBAEpB,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAU,GAAI,GAKvB,OAFM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GACpB,EAAI,EAAG,EAAI,EAAO,KAAM,KAI/B,OAHM,GAAM,EAAO,WAAW,GAExB,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,GAAK,EAAQ,GAAK,EAAM,GAE1C,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAM,OAAM,IAGrC,MAAO,GAAO,YAGhB,EAAA,UAAA,KAAA,SAAK,GAIH,OAHM,GAAQ,KAAK,SAAS,EAAE,QACxB,EAAS,EAAA,OAAU,CAAC,EAAE,KAAM,EAAE,MAAO,EAAE,OACvC,EAAO,EAAO,OACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAK,EAAI,EAAE,KAAO,GAAK,EAAM,GAE/B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,EAAA,MAAS,EAAG,EAAO,GAAM,QAAQ,GAE5C,MAAO,IAGT,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAiB,EAAG,WAKpB,OAHM,GAAS,EAAA,OAAU,EAAE,MAAO,EAAE,OAC9B,EAAO,KAAK,WAAW,cAEpB,GACP,GAAM,GAAS,EAAO,WAAW,GAC3B,EAAQ,EAAO,QACrB,EAAK,QAAQ,SAAA,GAAM,MAAA,GAAM,GAAM,EAAE,MAAM,GAAM,EAAI,EAAM,KACvD,EAAO,IAAG,MAAV,EAAM,CAAK,EAAK,IAAG,MAAR,EAAY,IAAK,OAAM,KAJ3B,EAAI,EAAG,EAAI,EAAO,KAAM,MAAxB,GAOT,MAAO,GAAO,YAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,UAAiB,EAAG,OAGb,EAAA,IAAO,EAAA,OAAU,IAAK,IAG/B,EAAA,UAAA,KAAA,SAAuB,GAAvB,GAAA,GAAA,KACE,GAAiB,EAAS,QAK1B,OAHM,GAAO,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAK,SAAS,EAAE,UACxC,EAAS,EAAA,OAAU,EAAQ,GAAG,MAAO,EAAQ,GAAG,OAChD,EAAa,EAAO,OACjB,EAAI,EAAG,EAAI,EAAQ,OAAQ,IAElC,OADM,GAAW,EAAK,GACb,EAAI,EAAG,EAAI,EAAW,OAAQ,IACrC,EAAW,IAAM,EAAS,GAG9B,MAAO,GAAO,YAGhB,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAGhD,EAAI,EAAA,IAAO,EAAQ,EAAS,QAAQ,IACpC,EAAI,EAAA,IAAO,GACX,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAO,EAAG,IAGnB,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,IAAI,EAAQ,MAIjE,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAiB,CAAC,EAAG,GAAI,UAqBzB,OAnBM,GAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAU,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5C,EAAW,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7C,EAAW,EAAE,MAAM,GAEnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,8DAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAAY,EAAA,EAAA,GAGrB,EAAA,8DAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAIzB,EAAO,EAAU,EACjB,EAAS,EAAA,OAAU,CAAC,EAAU,EAAS,GAAW,EAAE,OACpD,EAAU,EAAO,OACjB,EAAY,KAAK,UAEd,EAAI,EAAG,EAAI,EAAU,IAC5B,OAAS,GAAK,EAAG,EAAK,EAAS,GAAM,EACnC,OAAS,GAAK,EAAG,EAAK,EAAU,GAAM,EACpC,OAAS,GAAK,EAAG,EAAK,EAAW,GAAM,EAMrC,OAJM,GAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAClC,EAAS,KAAK,IAAI,EAAK,EAAW,GAE/B,EAAI,EAAI,EAAI,EAAQ,IAC3B,OAAS,GAAI,EAAI,EAAI,EAAQ,KAG3B,OAFI,GAAM,EAED,EAAI,EAAI,EAAI,EAAQ,IAC3B,GAAO,EAAQ,EAAI,EAAS,EAAI,EAAa,EAAI,GAC7C,EAAQ,EAAI,EAAa,EAAI,EAAa,EAAI,GAEpD,EAAQ,EAAI,EAAQ,GAAI,EAAW,KAAO,EAOtD,MAAO,GAAO,YAGhB,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE/C,EAAS,KAAK,YAAY,EAAG,EAAG,EAAY,GAChD,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAIvC,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAiB,CAAC,EAAG,GAAI,YAEzB,GAAM,GAAK,SAAC,EAAW,GAAc,MAAA,MAAK,MAAM,EAAI,IAC9C,EAAc,QACpB,MAAO,MAAK,oBAAoB,EAAG,EAAG,EAAa,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MASvD,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EACD,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAO,EAAM,EAAS,GAExB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,GAAiB,EAAG,OAUpB,OARM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAU,GAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAO,EACF,EAAI,EAAG,EAAI,EAAY,EAAE,EAChC,GAAQ,EAAM,EAAS,GAEzB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,EAAG,sBAOpB,OALM,GAAM,GAIN,EAAW,EAAE,KAAO,EAAW,KAC5B,EAAI,EAAG,EAAI,EAAU,EAAE,EAC9B,EAAa,EAAW,WAAW,EAAI,GAGzC,OAAS,GAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAY,EAAA,OAAU,EAAG,SACzB,EAAO,EAAA,MAAS,EAAW,GAAY,OAAO,WAC9C,EAAM,EAAK,IAAI,GAAG,IAAI,GAC5B,EAAI,KAAK,GAGX,MAAO,GAAA,MAAS,IAGlB,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,GAAiB,EAAG,UAEpB,GAAM,GAAO,CAAC,GACd,EAAA,aAAa,2BAA2B,SAAU,EAAM,EAAE,MAQ1D,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,SAC5B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAIjC,OAHM,GAAS,EAAI,EACf,EAAM,EAAM,GACZ,EAAW,EACN,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,EACN,EAAW,GAGf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAIlD,GAFA,GAAiB,EAAG,UAEhB,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,oDAAoD,GAAE,KAAO,GAAC,IAC9D,iBAAgB,IAWtB,OATM,GAAc,EAAA,WAAW,EAAE,MAAO,SAClC,EAAS,EAAA,MAAS,EAAE,MAAO,GAC3B,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAW,EAAE,MAAM,EAAE,KAAO,GAC5B,EAAgB,EAClB,SAAC,EAAW,GAAc,MAAA,GAAI,EAAW,EAAI,GAC7C,SAAC,EAAW,GAAc,MAAA,GAAI,GACzB,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACrC,OAAS,GAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAM,EAAc,EAAG,GAC7B,GAAI,IAAM,EACR,EAAK,GAAO,EAAY,EAAI,EAAM,QAElC,GAAM,GAAU,EAAc,EAAG,EAAI,GACrC,EAAK,GAAO,EAAY,EAAM,GAAW,EAAK,GACtB,EAAM,GAAO,EAAK,IAIhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,UAAiB,CAAC,EAAG,GAAI,YAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,KAAS,EAAQ,EAAI,KAIjC,EAAA,UAAA,KAAA,SAAK,EAAW,GACd,UAAiB,CAAC,EAAG,GAAI,QAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,GAAO,EAAQ,EAAI,KAI/B,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,UAAiB,CAAC,EAAG,GAAI,gBAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAQ,IAAQ,EAAQ,EAAI,KAIhC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,UAAiB,CAAC,EAAG,GAAI,cAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,UAAiB,CAAC,EAAG,GAAI,aAElB,KAAK,oBAAoB,EAAG,EAAG,OAAQ,SAAC,EAAM,GACnD,MAAO,IAAQ,KAInB,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAiB,CAAC,EAAW,EAAG,GAAI,UAYpC,OAVM,GAAS,KAAK,SAAS,EAAU,QACjC,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,MAAS,EAAE,MAAO,EAAA,WAAW,EAAE,MAAO,EAAE,QACjD,EAAY,KAAK,SAAS,EAAO,QACnC,EAAQ,EACN,EAAS,EAAU,OAAS,GAAK,EAAU,KAAO,GAAK,EAAE,OAAS,EACpE,EACA,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAE5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,OAAS,GAAI,EAAG,EAAI,EAAQ,IAC1B,AAAI,EAAO,KAAO,EAChB,EAAU,KAAW,EAAQ,GAE7B,EAAU,KAAW,EAAQ,GAKnC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,GACJ,GAAiB,CAAC,GAAY,SAE9B,GAAM,GAAW,KAAK,SAAS,EAAU,QACzC,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAiB,EAAG,QAEpB,GAAM,GAAQ,KAAK,SAAS,EAAE,QAC9B,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,UAAiB,CAAC,EAAG,GAAI,OAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAM,EAAO,EACnB,MAAK,GAAO,GAAK,EAAO,GAAO,GAAQ,GAAK,GAAQ,EAC3C,EAEC,GAAM,GAAQ,KAK5B,EAAA,UAAA,QAAA,SAAQ,EAAW,GACjB,UAAiB,CAAC,EAAG,GAAI,WAElB,KAAK,oBACR,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GAAS,MAAA,MAAK,IAAI,EAAM,MAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAM,GAAO,EAEf,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAiB,EAAG,OAEpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MAQvD,OAPM,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,MAAS,EAAU,EAAE,OAC9B,EAAa,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,SAAS,EAAO,QAE5B,EAAQ,KAAK,SAAS,EAAE,QACrB,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAS,EAAM,GACV,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,EAAS,GAAU,EAErB,EAAK,GAAK,EAEZ,MAAO,IAGT,EAAA,UAAA,kBAAA,SAAkB,EAAW,GAC3B,UAAiB,CAAC,EAAG,GAAI,qBAElB,KAAK,oBAAoB,EAAG,EAAG,EAAE,MAAO,SAAC,EAAM,GACpD,GAAM,GAAO,EAAO,EACpB,MAAO,GAAO,KAIlB,EAAA,UAAA,OAAA,SAAyB,GACvB,MAAO,IAGT,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,EAAG,EAAO,IAElC,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAiB,EAAG,QAKpB,OAHM,GAAM,EAAA,MAAS,EAAE,MAAO,EAAE,OAC1B,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAS,KAAK,SAAS,EAAE,QACtB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAQ,GAAK,KAAK,IAAI,KAAK,IAAI,EAAG,EAAO,IAAK,GAEhD,MAAO,IAGT,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MACR,SAAC,EAAQ,GAAW,MAAA,GAAS,EAAI,EAAS,EAAS,KAGhE,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAiB,CAAC,EAAI,GAAI,UAK1B,OAHM,GAAe,GAAI,cAAa,EAAE,MAClC,EAAS,KAAK,SAAS,EAAE,QACzB,EAAW,KAAK,SAAS,EAAG,QACzB,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAI,EAAO,GACjB,AAAI,GAAK,EACP,EAAa,GAAK,EAAS,GAE3B,EAAa,GAAK,EAAS,GAAM,GAAI,GAGzC,MAAO,MAAK,WAAW,EAAc,EAAE,MAAO,YAGhD,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,UAAiB,CAAC,EAAG,GAAI,SAElB,KAAK,oBACD,EAAG,EAAG,EAAE,MAAO,SAAC,EAAQ,GAAW,MAAA,MAAK,MAAM,EAAQ,MAInE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,OAAO,EAAO,EAAQ,GAExC,MAAI,IAEF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAiB,CAAC,EAAG,GAAS,UAyB9B,OAvBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAiB,EAAS,aAAe,eAEzC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OAEnC,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAChD,EAAe,EAAE,QAAQ,GACzB,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAAE,QAAQ,GACvD,EAAa,EAAiB,EAAE,QAAQ,GAAK,EAC7C,EAAiB,EAAiB,EAAI,EAAE,QAAQ,GAEhD,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EACf,EAAW,EAAI,EACZ,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EACxB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAC3B,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,IAAK,EAAW,EAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAC7B,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,GAAK,GAC1B,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,GAAK,IAClB,GAAO,EAAM,GAAW,IAE9B,IAAY,EAAS,cAOjC,MAAO,GAAE,YAGX,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAiBpC,OAfM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAI,EAAA,OAAmB,EAAS,SAAU,EAAE,OAE5C,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,KACjC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAKF,OAHM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAElC,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,KAClC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAS,YACpC,EAAW,EAAK,EAAS,YAAc,EACpC,GAAK,EAAG,GAAK,EAAa,MACjC,GAAM,IAAK,EAAW,GAAK,EAC3B,GAAI,GAAK,GAAK,IAAM,EAAS,QAC3B,SAKF,OAHM,IAAW,EAAW,GAAK,EAAO,QAAQ,GAC1C,GAAW,EAAW,GAAK,EAAS,WACtC,GAAW,GACN,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,IAE3C,OADM,IAAO,EAAM,GAAW,IACrB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,GAC5C,EAAM,EAAW,KAAO,GAAO,EAAM,GAAW,IAElD,IAAY,EAAS,eASrC,MAAO,GAAE,YAGX,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,kBAkC/B,OAhCM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAiB,IAAe,eAChC,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GACjD,EAAe,EAAG,QAAQ,GAC1B,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAAG,QAAQ,GACzD,EAAa,EAAiB,EAAG,QAAQ,GAAK,EAC9C,EAAiB,EAAiB,EAAI,EAAG,QAAQ,GAE9C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,IAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAOjC,OANM,IAAK,GAAK,EAAc,GACxB,GACF,EAAe,EAAI,EAAa,GAAK,EAAa,GAChD,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,EAAiB,IAC7C,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,GAAM,IAAW,EAAe,EAAI,EAAa,GAC7C,EAAa,GAAK,EAAiB,EACvC,EAAS,IAAY,GAK7B,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GA6BF,OA5BM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAE1B,EAAA,EAAA,UACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,GAAA,EAAA,YAEI,GAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,GAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,GAAU,EAAc,EAAI,EAAS,QAAQ,KAE1C,GAAI,EAAG,GAAI,EAAW,EAAE,GAC/B,OAAS,IAAK,EAAG,GAAK,EAAY,EAAE,GAElC,OAAS,IAAK,EAAG,GAAK,EAAS,EAAE,GAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAGzC,GAAK,EAAG,GAAK,EAAU,EAAE,GAMhC,OALM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,GAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,KACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,IAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAc,GAErB,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAQjC,OAPM,IAAK,GAAK,GAAc,GACxB,GACF,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GACxC,GAAY,EAAS,GAAc,EAAI,IACzC,EAAS,GAAe,EAAI,IAC5B,EAAS,GAAc,EAAI,IAAM,EAAQ,GAEpC,GAAK,EAAG,GAAK,EAAa,EAAE,IACnC,GAAM,IAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAK3B,EAAS,EAAO,GAAI,EAAO,GAAK,EAAO,GAAK,EAAO,GAAK,IACpD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAiB,CAAC,EAAG,GAAK,mBAa1B,OAXM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eACzC,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,EAC3C,OAAS,GAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAG5C,OADI,GAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,AAAI,EACF,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAEnD,GACI,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAK3D,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAKpC,MAAO,GAAG,YAGZ,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAqBzC,OAnBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAC9C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACnB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAAK,EAAA,EAAA,GAEhB,EAAW,EAAS,QAAQ,MAC5B,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAEvB,EAAK,EAAG,EAAK,EAAa,EAAE,EAMnC,OALM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAW,GAAM,IAChD,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAW,GAAM,GACtD,EAAW,EAAK,EAEb,EAAK,EAAG,EAAK,EAAc,EAAE,EAOpC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,GAAQ,KAAK,IACf,EAAS,UACR,GAAS,SAAW,EAAS,GAAM,GAClC,GAAW,EAAK,EAAO,EAEpB,GAAK,EAAG,GAAK,EAAa,EAAE,GAOnC,OANM,IAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,IAAM,IAC/C,GAAQ,KAAK,IACf,EAAS,SACR,GAAS,QAAU,EAAU,IAAM,GAClC,GAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,WAAY,EAAE,GAG3C,OAFM,IAAW,GAAK,EAAO,GAEpB,GAAK,EAAG,GAAK,EAAS,YAAa,EAAE,IAE5C,OADI,IAAU,EACL,GAAI,EAAG,GAAI,EAAS,UAAW,EAAE,GAIxC,OAHM,IAAW,GAAI,EACf,GAAW,GAAI,EAEZ,GAAK,EAAO,GAAK,EAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,GAKjC,OAJM,IAAK,EAAK,GAAK,EAAe,EAC9B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAEpB,GAAK,GAAO,GAAK,GAAO,EAAE,IACjC,GAAM,IAAK,GAAK,GAAK,EAAc,EAC7B,GAAW,GAAK,EAAM,GACtB,GAAW,GAAK,EAAO,GAE7B,IACI,EAAQ,GAAW,IAAM,EAAS,GAAW,IAKzD,EAAS,GAAW,IAAM,GAMpC,MAAO,GAAG,YAGZ,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE1C,EAAS,KAAK,gBAAgB,EAAO,EAAQ,GAEjD,MAAI,IAGF,GAAS,EAAA,IAAO,EAAQ,IAE1B,AAAI,GACF,GACI,GAAc,KAAM,EAAQ,EAAY,IAGvC,GAGT,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAiB,CAAC,EAAG,GAAS,mBAc9B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WACxC,EAAI,EAAA,OAAU,EAAS,SAAU,EAAE,OACnC,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAO,QAC7B,EAAQ,EAAE,OAEP,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAW,EAAI,EAAE,QAAQ,GACzB,EAAW,EAAI,EAAE,QAAQ,GACtB,EAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAG1C,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,aAAe,EACrC,EAAK,EAAG,EAAK,EAAc,EAAE,GACpC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,SAC3B,SAIF,OAFM,GAAW,EAAK,EAAO,QAAQ,GAC/B,EAAW,EAAW,EAAK,EAAE,QAAQ,GAClC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,EAGzC,OAFM,GAAW,EAAW,EAAK,EAAE,QAAQ,GACrC,EAAW,EAAK,EAAS,YAAc,EACpC,EAAK,EAAG,EAAK,EAAa,EAAE,GACnC,GAAM,GAAK,EAAW,EAAK,EAC3B,GAAI,EAAK,GAAK,GAAM,EAAS,QAC3B,SAMF,OAJM,GAAW,EAAW,EAAK,EAAO,QAAQ,GAC1C,EAAW,EAAW,EAAK,EAAS,WACtC,EAAW,EACX,EAAW,EACN,EAAK,EAAG,EAAK,EAAS,WAAY,EAAE,GAE3C,OADM,GAAO,EAAM,EAAW,GACrB,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAM,EAAW,IAAM,EAAO,EAAM,EAAW,GAEjD,GAAY,EACZ,GAAY,IAQxB,MAAO,GAAE,YAGX,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAiB,CAAC,EAAI,GAAS,2BA0B/B,OAxBM,GAAK,EAAA,OAAmB,EAAS,QAAS,WAC1C,EAAW,EAAG,OACd,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAA,EAAA,QAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAAM,EAAA,EAAA,GACb,EAAY,KAAK,SAAS,EAAO,QACjC,EAAA,EAAA,QAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAO,EAAA,EAAA,GAEnB,EAAA,EAAA,UACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,YACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,aACA,EAAA,EAAA,YAEI,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAQ,EAAc,EAEnB,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAK,EAAG,EAAK,EAAY,EAAE,EAClC,OAAS,GAAK,EAAG,EAAK,EAAU,EAAE,EAMhC,OALM,IAAW,EAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAY,GAAe,IAAY,GAE3C,GAAK,EAAG,GAAK,EAAS,EAAE,IAO/B,OANM,IAAW,GAAK,EAChB,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAK,GAAW,IACzC,GACF,KAAK,IAAI,EAAW,GAAc,IAAY,GAE9C,GAAU,EACL,GAAK,GAAO,GAAK,GAAO,EAAE,GAGjC,OAFM,IAAK,GAAK,EAAe,GAEtB,GAAK,GAAO,GAAK,GAAO,EAAE,GAMjC,OALM,IAAK,GAAK,EAAc,GACxB,GAAW,EAAO,EAAI,EAAO,GAAK,EAAO,GACzC,GAAY,EAAS,GAAe,EAAI,IAC1C,EAAS,GAAc,EAAI,IAAM,EAAQ,EAEpC,GAAK,EAAG,GAAK,EAAO,EAAE,IAC7B,GAAM,IAAK,EAAK,EAAQ,GAClB,GAAQ,EAAS,GAAW,IAC5B,GAAS,EAAU,GAAY,IACrC,IAAW,GAAQ,GAIzB,EAAS,EAAO,EAAI,EAAO,EAAK,EAAO,GAAK,GAAM,GAK1D,MAAO,GAAG,YAGZ,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAiB,CAAC,EAAG,GAAK,4BAc1B,OAZM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAK,EAAA,OAAmB,EAAS,YAAa,WAE9C,EAAU,EAAS,QAAQ,KAC3B,EAAS,EAAS,QAAQ,IAC1B,EAAQ,EAAS,YAAc,EAAS,WAExC,EAAO,KAAK,WAAW,GACvB,EAAQ,KAAK,WAAW,GACrB,EAAK,EAAG,EAAK,EAAc,EAAE,EAKpC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAS,GAAM,IAC9C,EAAQ,KAAK,IACf,EAAS,UAAY,GAAS,SAAW,EAAS,GAAM,GAEnD,EAAK,EAAG,EAAK,EAAa,EAAE,EAKnC,OAJM,GAAQ,KAAK,IAAI,EAAG,KAAK,KAAM,GAAU,GAAM,IAC/C,EAAQ,KAAK,IACf,EAAS,SAAW,GAAS,QAAU,EAAU,GAAM,GAElD,EAAK,EAAG,EAAK,EAAS,YAAa,EAAE,GAK5C,OAJM,GAAK,KAAK,MAAM,EAAK,GACrB,EAAK,EAAK,EAEZ,EAAU,EACL,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAK,EAAO,EAAK,EAAO,EAAE,EAEjC,OADM,GAAK,EAAK,EAAK,EAAe,EAC3B,EAAK,EAAO,EAAK,EAAO,EAAE,GACjC,GAAM,GAAK,EAAK,EAAK,EAAc,EACnC,GAAW,EAAK,IAAI,EAAG,EAAI,EAAI,GAAM,EAAM,IAAI,EAAG,EAAI,EAAI,GAIhE,EAAG,IAAI,EAAS,EAAI,EAAI,EAAI,GAIlC,MAAO,GAAG,YAGZ,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,UAAiB,EAAG,QACb,GAAK,KAAK,WAAW,GAAI,IAGlC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAChD,GAAiB,CAAC,EAAG,GAAU,UAE/B,GAAM,GAAqB,EAAE,MAAM,QAC7B,EAAgB,KAAK,SAAS,EAAQ,QAC5C,EAAS,GAAQ,EAAc,OAI/B,OAHM,GAAS,EAAA,OAAU,EAAU,EAAE,OAC/B,EAAO,KAAK,WAAW,GAEpB,EAAI,EAAG,EAAI,EAAO,KAAM,EAAE,GACjC,GAAM,GAAS,EAAO,WAAW,GAE3B,EAAwB,EAAO,QACrC,EAAY,GAAQ,EAAc,EAAO,IAEzC,GAAM,GAAgB,EAAK,WAAW,GACtC,EAAO,OAAO,GAAK,EAAK,OAAO,GAEjC,MAAO,GAAO,YAGhB,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAa,EAAE,QAAQ,GAAW,GAC7B,QAAQ,GACR,MAAM,EAAkB,IAG9B,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAiB,EAAG,UA8BpB,OA5BM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,EAAA,OAAU,EAAS,SAAU,EAAE,OACxC,EAAa,EAAO,OAEpB,EAAqB,EAAS,SAAS,GAAK,EAAS,SAAS,GAChE,EAAS,SAAS,GAAK,EAAS,SAAS,GACvC,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAGhD,OAFM,GAAoB,EAAQ,EAC5B,EAAmB,EAAQ,EAAE,QAAQ,GAClC,EAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAMf,OAJM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAChD,EACF,EAAoB,EAAS,EACxB,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAKb,OAHM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,GAAkB,EAAoB,EAAO,EAC1C,GAAO,EAAG,GAAO,EAAS,SAAU,EAAE,IAG7C,OAFM,IAAa,GAAO,EAAc,EACpC,GAAU,GACP,GAAU,GACf,IAAW,EASb,OAPM,IACF,KAAK,IAAI,EAAS,QAAS,EAAuB,IAEhD,GAAkB,GAAkB,GAAO,EAC7C,GAAc,EACd,GAAW,EACX,GAAQ,EACH,GAAS,EAAW,GAAS,EACjC,IAAU,GAEb,OADM,IAAe,EAAmB,GAAS,EAAE,QAAQ,GAClD,GAAO,EAAS,GAAO,EAC3B,IAAQ,GAEX,OADM,IAAa,GAAe,GAAO,EAAE,QAAQ,GAC1C,GAAO,GAAS,GAAO,GAC3B,IAAQ,GACX,GAAM,IAAa,GAAa,GAAO,EAAE,QAAQ,GAC3C,GAAQ,EAAQ,GAAa,GAOnC,GANA,AAAK,IAAa,OAAS,GAAQ,GACjC,GAAc,GACT,AAAI,IAAa,OACtB,KAAY,GACZ,MAEE,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAI,MAAM,IACR,MAGJ,GAAM,IAAe,GAAkB,EACvC,EAAW,IACP,IAAa,MAAQ,GAAW,GAAQ,KAMtD,MAAO,GAAO,YAGhB,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGzC,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,qBAuB1B,OArBM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAgB,EAAK,GAAc,EAAe,GAElD,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,KAIjB,EAAG,IACC,EAAU,EAAe,EAAO,EAAS,EAAO,EAChD,GAMd,MAAO,GAAG,YAGZ,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,UAAiB,EAAG,aAEb,KAAK,OAAO,EAAG,EAAU,OAAO,WAGjC,EAAA,UAAA,mBAAR,SAA2B,EAAa,GAiBtC,OAfM,GAAe,EAAA,OAAU,EAAS,SAAU,SAC5C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,KAAK,WAAW,GACpB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAS,EAAG,EAAS,EAAS,SAAU,EAAE,GAGjD,OAFM,GAAe,EAAS,EAAc,EACxC,EAAY,EACT,EAAY,GACjB,GAAa,EAIf,OAFM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAC7C,EAAO,EAAG,EAAO,EAAS,UAAW,EAAE,GAG9C,OAFM,GAAa,EAAO,EAAe,EACrC,EAAU,EACP,EAAU,GACf,GAAW,EAIb,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAO,EAAG,EAAO,EAAS,SAAU,EAAE,GAG7C,OAFM,GAAa,EAAO,EAAc,EACpC,EAAU,EACP,EAAU,GACf,GAAW,EASb,OAPM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAGlD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAS,EAAW,EAAS,EACjC,GAAU,EAEb,OADM,GAAS,EAAS,EACf,EAAO,EAAS,EAAO,EAC3B,GAAQ,EAEX,OADM,GAAO,EAAO,EACX,GAAO,EAAS,GAAO,EAC3B,IAAQ,GACX,GAAM,IAAO,GAAO,EACd,GAAQ,EAAK,IAAI,EAAO,EAAQ,EAAM,GAAM,GAClD,AAAI,IAAS,GACX,GAAW,GACX,EAAc,EAAS,EACf,EACJ,EAAO,EAAwB,IAM3C,EAAa,IAAI,EAAa,EAAO,EAAQ,EAAM,EAAM,KAMnE,MAAO,GAAa,YAGtB,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAiB,CAAC,EAAG,GAAI,qBAoBzB,OAlBM,GAAe,KAAK,mBAAmB,EAAG,GAC1C,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAK,EAAA,OAAmB,EAAE,MAAO,WAEjC,EAAY,KAAK,WAAW,GAC5B,EAAQ,KAAK,WAAW,GAErB,EAAQ,EAAG,EAAQ,EAAS,UAAW,EAAE,EAChD,OAAS,GAAU,EAAG,EAAU,EAAS,WAAY,EAAE,EACrD,OAAS,GAAU,EAAG,EAAU,EAAS,QAAS,EAAE,EAClD,OAAS,GAAQ,EAAG,EAAQ,EAAS,SAAU,EAAE,EAC/C,OAAS,GAAQ,EAAG,EAAQ,EAAS,QAAS,EAAE,GAM9C,OAJM,GAAgB,EAAU,EAC1B,EAAc,EAAQ,EACtB,EAAc,EAAQ,EACxB,EAAU,EACL,EAAS,EAAG,EAAS,EACzB,GAAU,GACb,GAAM,GAAW,GAAgB,GAAU,EAC3C,GAAI,EAAU,GAAK,GAAW,EAAS,UACnC,KAAK,MAAM,KAAa,EAC1B,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,WAC/B,KAAK,MAAM,KAAW,EACxB,SAEF,OAAS,GAAO,EAAG,EAAO,EACrB,GAAQ,GACX,GAAM,GAAS,GAAc,GAAQ,EACrC,GAAI,EAAQ,GAAK,GAAS,EAAS,UAC/B,KAAK,MAAM,KAAW,EACxB,SAGF,GAAM,GAAS,EACP,EAAwB,EAC5B,EACA,EAAU,IAAI,EAAO,EAAS,EAAO,EAAO,GAC1C,GACF,EAAS,EAAwB,EACjC,EAAO,EAAuB,EAE5B,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IACF,EAAM,IAAI,EAAO,EAAS,EAAO,EAAO,GAC5C,GAAW,GAAQ,KAIzB,EAAG,IAAI,EAAS,EAAO,EAAS,EAAO,EAAO,GAMxD,MAAO,GAAG,YAGZ,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,kBAqBpB,OAnBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cACf,EAAA,KAAK,cAAc,CAAC,EAAO,EAAW,EAAU,KAE9C,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAE9C,EAAY,EACV,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GACvC,EAAI,EAAG,EAAI,EAAO,IACzB,OAAS,GAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EAAgB,KAAK,IAAI,EAAY,EAAG,KAAK,KAAK,IAClD,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAiB,EAAE,QAAQ,GAC7D,EAAe,EAAI,EAAE,QAAQ,GAAK,EAAgB,EAAE,QAAQ,GACzD,EAAI,EAAG,EAAI,EAAU,IAU5B,OATM,GAAgB,EAAwB,EACxC,EAAiB,KAAK,MAAM,GAC5B,EAAU,EAAgB,EAC1B,EACF,KAAK,IAAI,EAAW,EAAG,KAAK,KAAK,IAC/B,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAgB,EAAe,EAAiB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GAC1D,EAAiB,EAAe,EAAgB,EAAE,QAAQ,GACvD,EAAI,EAAG,EAAI,EAAa,KAI/B,GAAM,GAAU,EAAQ,EAAgB,GAClC,EAAa,EAAQ,EAAgB,GACrC,GAAW,EAAQ,EAAiB,GACpC,GAAc,EAAQ,EAAiB,GAEvC,GAAM,EAAW,IAAW,GAAW,EACvC,GAAS,EAAc,IAAc,GAAc,EACnD,GAAW,GAAO,IAAS,IAAO,EAExC,EAAO,KAAe,GAK9B,MAAO,GAAA,OAAU,EAAQ,CAAC,EAAO,EAAW,EAAU,KAGxD,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAChD,GAAiB,CAAC,EAAI,GAAI,0BA+B1B,OA7BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GAOrD,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAMhD,EAAW,KAAK,SAAS,EAAG,QAC9B,EAAS,EACJ,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAU,EAAI,EAAE,QAAQ,GACrB,EAAI,EAAG,EAAI,EAAS,IAU3B,OATM,GAAM,EAAI,EACV,EAAc,KAAK,MAAM,GACzB,EAAiB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAU,GAEpD,EAAe,EAAU,EAAc,EAAE,QAAQ,GACjD,EAAkB,EAAU,EAAiB,EAAE,QAAQ,GAEvD,EAAU,EAAM,EAChB,EAAiB,EAAM,EACpB,EAAI,EAAG,EAAI,EAAQ,IAmB1B,OAlBM,GAAM,EAAI,EACV,EAAe,KAAK,MAAM,GAC1B,EAAgB,KAAK,IAAI,KAAK,KAAK,GAAM,EAAS,GAClD,EAAU,EAAM,EAChB,EAAiB,EAAM,EAEvB,EAAkB,EAAe,EAAe,EAAE,QAAQ,GAC1D,EAAmB,EAAe,EAAgB,EAAE,QAAQ,GAC5D,GACF,EAAkB,EAAe,EAAE,QAAQ,GACzC,GACF,EAAkB,EAAgB,EAAE,QAAQ,GAE1C,GACF,EAAiB,EACf,GAA6B,EAAiB,EAC9C,GAA6B,EAAU,EACvC,GAAsB,EAAU,EAC7B,GAAI,EAAG,GAAI,EAAO,MACzB,GAAM,IAAQ,EAAS,KACvB,EAAO,EAAkB,KACrB,GAAQ,GACZ,EAAO,EAAmB,KAAM,GAAQ,GACxC,EAAO,GAAqB,KACxB,GAAQ,GACZ,EAAO,GAAsB,KAAM,GAAQ,GAKnD,MAAO,GAAA,SAAY,EAAQ,CAAC,EAAO,EAAQ,EAAS,GAAQ,EAAE,QAGhE,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAiB,EAAG,yBAsBpB,OApBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GAC7B,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAS,GAAI,cAAa,EAAQ,EAAY,EAAW,GAEzD,EAAuC,CAC1C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAwC,CAC3C,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EACF,EAAmB,GAAK,EAAoB,GAC1C,EACF,EAAmB,GAAK,EAAoB,GAE5C,EAAe,EACV,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAW,IAO7B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAY,EACZ,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAc,EAAmB,EAAE,QAAQ,GACpD,EAAI,EAAG,EAAI,EAAU,IAO5B,OANM,GAAgB,EAAwB,EACxC,EAAmB,KAAK,IAC1B,EAAW,EACX,EAAe,KAAK,MAAM,GACX,KAAK,MAAM,IACxB,EAAY,EAAY,EAAmB,EAAE,QAAQ,GAClD,EAAI,EAAG,EAAI,EAAa,KAG/B,GAAM,GAAS,EAAQ,EAAY,GACnC,EAAO,KAAkB,EAKjC,MAAO,GAAA,OACH,EAAQ,CAAC,EAAO,EAAW,EAAU,GAAc,EAAE,QAG3D,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAiB,CAAC,EAAI,GAAI,iCAiC1B,OA/BM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAS,EAAA,EAAA,GAAQ,EAAA,EAAA,GACzB,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAEZ,EAAS,GAAI,cAAa,EAAQ,EAAU,EAAS,GACrD,EAAW,KAAK,SAAS,EAAG,QAK5B,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAGzC,EAAI,EAAG,EAAI,EAAO,IAEzB,OADM,GAAc,EAAI,EAAE,QAAQ,GACzB,EAAI,EAAG,EAAI,EAAS,IAM3B,OALM,GAAY,EAAc,EAAI,EAAE,QAAQ,GAGxC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAY,GAC7C,EAAI,EAAG,EAAI,EAAQ,IAO1B,OANM,GAAY,EAAY,EAAI,EAAE,QAAQ,GAGtC,EAAa,KAAK,MAAM,EAAI,GAC5B,EAAW,KAAK,MAAM,EAAc,EAAW,GAE5C,EAAI,EAAG,EAAI,EAAO,KAIzB,OAHI,GAAQ,EAGH,EAAW,EAAG,EAAW,EAAW,KAC3C,GAAM,GAAM,EAAW,EAEvB,GAAI,EAAM,GAAK,GAAO,EACpB,SAGF,GAAM,GAAY,EAAc,EAAM,EAAG,QAAQ,GAC3C,GAAgB,EAAM,EACtB,GAAmB,KAAK,IAC1B,EAAU,EACV,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAC9B,GAAI,IAAM,GACR,SAEF,OAAS,IAAW,EAAG,GAAW,EAAU,MAC1C,GAAM,IAAM,GAAW,EAEvB,GAAI,GAAM,GAAK,IAAO,EACpB,SAGF,GAAM,IAAY,EAAY,GAAM,EAAG,QAAQ,GACzC,GAAgB,GAAM,EACtB,GAAmB,KAAK,IAC1B,EAAS,EACT,EAAe,KAAK,MAAM,IACX,KAAK,MAAM,KAE9B,AAAI,IAAM,IACR,IAAS,EAAS,GAAY,KAIpC,EAAO,EAAY,GAAK,EAKhC,MAAO,GAAA,SAAY,EAAQ,EAAE,MAAO,EAAE,QAGxC,EAAA,UAAA,6BAAA,SACI,EAAa,EAAqB,EAAc,EAChD,GACF,GAAiB,EAAG,gCAEpB,GAAM,GAAW,EAAE,MAAM,GACnB,EAAO,EAAW,EAClB,EAAU,KAAK,SAAS,EAAE,QAC1B,EAAO,EAAE,KACT,EAAS,GAAI,cAAa,GAEhC,WAA2B,GAQzB,OAPM,GAAiB,EAAS,EAC5B,EACA,EAAS,EAAiB,KAAK,IAAI,EAAG,EAAiB,GACrD,EAAe,EAAS,EAC1B,KAAK,IAAI,EAAiB,EAAa,GAEvC,EAAM,EACH,GAAkB,EAAc,KACrC,GAAM,GAAI,EAAQ,GAClB,GAAO,EAAI,EAEb,MAAO,GAGT,OAAS,GAAS,EAAG,EAAS,EAAM,KAClC,GAAM,GAAM,EAAkB,GACxB,EAAM,EAAQ,GAAU,KAAK,IAAI,EAAO,EAAQ,EAAK,CAAC,GAC5D,EAAO,GAAU,EAGnB,MAAO,GAAA,SAAY,EAAQ,EAAE,QAG/B,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAiB,EAAI,WAQrB,OAPM,GAAW,EAAG,MAAM,GACpB,EAAW,KAAK,SAAS,EAAG,QAC5B,EAAmB,KAAK,SAAS,EAAW,QAC5C,EAAoB,KAAK,SAAS,EAAY,QAC9C,EAAS,GAAI,cAAa,EAAG,MAC7B,EAAO,EAAG,KAEP,EAAS,EAAG,EAAS,EAAM,KAQlC,OAPM,GAAiB,EAAS,EAC1B,EACD,EAAS,EAAkB,KAAK,IAAI,EAAG,EAAiB,GACvD,EAAY,EAAS,EACvB,KAAK,IAAI,EAAU,EAAiB,EAAc,GAElD,EAAO,EACF,EAAI,EAAY,EAAI,EAAU,IACrC,GAAQ,KAAK,IAAI,EAAiB,GAAI,GAExC,EAAO,EAAQ,EAAO,EAEtB,OAAS,GAAI,EAAY,EAAI,EAAU,KACrC,GAAI,GAAM,GAAK,EAAQ,EAAO,EAAiB,GAC3C,EAAkB,GAAU,EAChC,AAAI,IAAW,GACb,IAAO,KAAK,IAAI,EAAM,CAAC,IAEzB,GAAO,EAAS,GAChB,EAAO,IAAM,GAGjB,MAAO,GAAA,SAAY,EAAQ,EAAG,QAGhC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAiB,EAAQ,eASzB,OAPM,GAAgB,EAAa,EAAS,EAAA,QAAW,GACjD,EAAY,EAAc,MAAM,GAChC,EAAY,EAAc,MAAM,GAChC,EAAM,EAAA,MAAkB,CAAC,EAAW,GAAa,SACjD,EAAU,KAAK,SAAS,EAAI,QAC5B,EAAW,KAAK,SAAS,EAAc,QAEpC,EAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,GAAS,EAAI,EAGb,EAAM,GAAI,cAAa,EAAY,GACzC,EAAI,GAAK,EAAS,GAClB,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,EAAE,EACxC,EAAI,GAAS,EAAI,EAAQ,GAAK,EAAS,EAAS,GAKlD,OAFM,GAAS,GAAA,KAAgB,EAAK,YAC9B,EAAY,EAAI,EACb,EAAW,EAAG,EAAW,EAAY,EAAE,GAC9C,GAAM,GAAI,IAGV,EAAQ,EAAY,GAAY,EAAI,OAEpC,OAAS,GAAQ,EAAG,EAAQ,EAAI,OAAQ,IACtC,GAAI,EAAI,EAAI,IACV,EAAQ,EAAY,GAAY,EAChC,QAKR,MAAO,IAGT,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAiB,EAAS,UAE1B,GAAM,GAAM,GAAI,cAAa,EAAQ,KAAO,GAC5C,EAAI,KAAK,GAGT,OAFM,GAAa,KAAK,SAAS,EAAQ,QAEhC,EAAQ,EAAG,EAAQ,EAAQ,KAAM,EAAE,EAC1C,AAAI,EAAW,IAAU,GAAK,EAAW,GAAS,GAChD,GAAI,EAAQ,EAAQ,EAAW,IAAU,GAG7C,MAAO,GAAA,SAAY,EAAK,CAAC,EAAQ,KAAM,GAAQ,UAGjD,EAAA,UAAA,kBAAA,SACI,EAAiB,EAAkB,EACnC,EAAsB,GACxB,GAAiB,EAAO,qBAExB,GAAM,GAAY,KAAK,SAAS,EAAM,QAChC,EAAa,KAAK,SAAS,EAAO,QACxC,MAAO,IACH,EAAW,EAAY,EAAe,EAAc,IAG1D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,IAAe,OACf,WAAM,MAAA,+DACF,IACR,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAgB9D,OAdM,GAAY,EAAE,MAAM,GACpB,EAAc,EAAE,MAAM,GACtB,EAAa,EAAE,MAAM,GACrB,EAAa,EAAE,MAAM,GAErB,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAU,KAAK,SAAS,EAAE,QAC1B,EACF,GAAI,cAAa,EAAY,EAAe,EAAc,GAE1D,EAAY,EACP,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAI,EAAG,EAAI,EAAc,EAAE,EAGlC,OAFM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACZ,EAAI,EAAG,EAAI,EAAa,EAAE,EAIjC,OAHM,GAAM,KAAK,MAAM,EAAI,GACrB,EAAW,EAAI,EACf,EAAW,GAAU,EAAY,GAAW,EACzC,EAAI,EAAG,EAAI,EAAa,EAAE,GACjC,GAAM,GAAM,EAAI,EACV,EACF,EAAM,EAAc,GAAM,EAAc,GAAM,EAAc,IAChE,EAAO,KAAe,EAAQ,GAKtC,MAAO,GAAA,SACH,EAAQ,CAAC,EAAW,EAAc,EAAa,KAG7C,EAAA,UAAA,oBAAR,SACI,EAAW,EAAW,EACtB,GACF,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAE,MAAO,EAAE,OAC9D,EAAS,EAAA,OAAU,EAAU,GAC7B,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAQ,KAAK,SAAS,EAAE,QACxB,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GACxD,EAAiB,EAAA,aAAa,iBAAiB,EAAE,MAAO,GAExD,EAAU,EAAO,OACvB,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,EACpC,EAAQ,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAK3D,QAFM,GAAO,KAAK,WAAW,GACvB,EAAO,KAAK,WAAW,cACpB,GACP,GAAM,GAAM,EAAO,WAAW,GAExB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAEzB,EAAO,EAAI,MAAM,CAAC,EAAE,MAC1B,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAK,WAAW,GAE/B,EAAQ,GAAK,EAAG,EAAM,GAAS,EAAM,KAX9B,EAAI,EAAG,EAAI,EAAQ,OAAQ,EAAE,IAA7B,GAcX,MAAO,GAAO,YAGhB,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,QAAA,aAEA,EAAA,UAAA,eAAA,WACE,MAAO,KAIT,EAAA,UAAA,QAAA,WACE,MAAO,GAAA,UAAM,QAAO,KAAA,OAGtB,EAAA,UAAA,cAAA,SACI,EACA,EACA,EACA,EACA,EACA,GAmBF,OAjBM,GAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GACjC,EAAW,EAAM,MAAM,GAEtB,EAAA,EAAA,GAAY,EAAA,EAAA,GACb,EACF,EAAA,OAAU,CAAC,EAAU,EAAY,EAAW,GAAc,WAExD,EAAU,KAAK,SAAS,EAAM,QAC9B,EAAa,KAAK,SAAS,EAAS,QACpC,EAAY,KAAK,SAAS,EAAO,QAEjC,EAAW,EAAO,QAClB,EAAY,EAAO,QAKhB,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAW,EAAI,EACf,EAAK,EAAQ,GACb,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GACxB,EAAK,EAAQ,EAAW,GAExB,EAAe,EAAW,GAChC,GAAI,GAAQ,EACV,SASF,OANM,GAAe,EAAa,EAC7B,GAAK,GAAO,GAAc,GAAM,GAAa,GAC9C,EACE,EACD,EAAY,EAAM,GAAK,GAAO,GAAa,GAAM,GAAY,GAAK,EAE9D,EAAI,EAAG,EAAI,EAAY,KAC9B,GAAM,GAAgB,EAAa,EAC/B,EAAM,GAAc,GAAK,EAAK,EAC9B,GAAO,GAAK,GAAO,GAAc,GAErC,GAAI,EAAO,GAAK,EAAO,EAAc,GACnC,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAGzB,SAGF,GAAI,IAAW,WAKb,OAJM,GAAS,KAAK,MAAM,GACpB,EAAY,KAAK,KAAK,GACtB,GAAQ,EAAO,EAEZ,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAOF,OAJM,IAAU,KAAK,MAAM,IACrB,GAAW,KAAK,KAAK,IACrB,GAAQ,GAAO,GAEZ,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAI,GAAM,EAAI,GAAU,EAAS,GAAK,EAAS,EAAS,GACpD,EAAO,EAAS,GACd,GAAU,EAAU,GAE1B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAS,EAAS,GACjD,EAAO,EAAS,GACpB,GAAM,IAAW,EAAU,GAE3B,EAAM,EAAI,GAAU,EAAS,GAAK,EAAY,EAAS,GACnD,EAAO,EAAS,GACpB,GAAM,IAAa,EAAU,GAE7B,EAAM,EAAI,GAAW,EAAS,GAAK,EAAY,EAAS,GACpD,EAAO,EAAS,GACpB,GAAM,IAAc,EAAU,GAExB,GAAM,GAAW,IAAW,IAAW,GACvC,GAAS,GAAc,IAAc,IAAc,GAEzD,EAAM,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC9D,EAAO,OAAO,GAAO,GAAQ,IAAS,IAAO,QAIjD,QAAS,GAAI,EAAG,EAAI,EAAW,EAAE,GAC/B,GAAM,IAAQ,EAAY,EACtB,EAAM,GAAa,GAAK,EAAI,EAC5B,GAAO,GAAK,GAAO,GAAa,GAEpC,GAAI,GAAO,GAAK,GAAO,EAAa,GAClC,OAAS,GAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,GAAO,EAEvB,SAKF,OAFM,IAAW,KAAK,MAAM,IACtB,GAAW,KAAK,MAAM,GACnB,EAAI,EAAG,EAAI,EAAa,KAC/B,GAAM,IAAQ,EAAI,GAAW,EAAS,GAClC,GAAW,EAAS,GAAK,EAAO,EAAS,GACvC,GACF,EAAI,EAAI,EAAU,GAAK,EAAI,EAAU,GAAK,EAAI,EAAU,GAC5D,EAAO,OAAO,IAAU,EAAU,OAM5C,MAAO,GAAO,YAGhB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAiB,GACvB,MAAO,MAAK,QACR,EAAe,EAAc,EAAa,EAAY,EACtD,EAAY,EAAW,EAAS,EAAc,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAE1C,GAAI,IAAc,EAChB,MAAO,GAAA,OAAU,GAAI,EAAa,EAAE,OAOtC,OAJM,GAAS,GAAI,GAAA,aAAa,CAAC,EAAW,GAAY,EAAE,OACpD,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAQ,KAAK,SAAS,EAAE,QAErB,EAAI,EAAG,EAAI,EAAW,KAG7B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,GAAgB,EAAM,EAAQ,GAC9B,EAAM,KAAK,GAEb,GAAI,EAAe,GAAK,GAAgB,EAAE,KAAO,EAC/C,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,EAAE,OAGzD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,EAAO,OAAO,EAAI,EAAY,GAAK,EAAM,EAAe,EAAY,GAGxE,MAAO,GAAO,WAAW,QAAQ,IAGnC,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAE5C,EAAe,EAAA,OAAU,GACzB,EAAiB,GACvB,MAAO,MAAK,QACR,EAAS,EAAS,EAAO,EAAY,EAAW,EAAY,EAC5D,EAAS,EAAc,IAG7B,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAC5C,EAAQ,GAAS,EAAA,KAAK,WAAW,GACjC,GAAM,GACF,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IACrD,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,OAGnD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAEhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,GAAM,GAAS,EAAA,KAAK,kBACD,EAAE,MAAO,EAAA,KAAK,cAAc,EAAE,QACjD,MAAO,MAAK,WAAW,EAAQ,EAAE,MAAO,EAAE,QAG5C,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GACpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGxC,EAAA,UAAA,QAAR,SACI,EAAiB,EAAiB,EAAoB,EACtD,EAAmB,EAAoB,EACvC,EAAmB,EACnB,GACF,GAAM,GAAe,CAAC,EAAa,EAAW,GAExC,EAAc,KAAK,SAAS,EAAQ,QACpC,EAAc,KAAK,SAAS,EAAQ,QAE1C,GAAI,IAAe,EACjB,MAAO,GAAA,OAAU,GAAI,EAAO,EAAQ,OAGtC,GAAM,GAAS,GAAI,GAAA,aAAa,EAAc,EAAQ,OACtD,EAAO,OAAO,KAAM,KAAK,SAAS,EAAa,QAAuB,IAEtE,OAAS,GAAI,EAAG,EAAI,EAAY,KAG9B,OAFM,GAAQ,GACV,EAAe,EACV,EAAI,EAAG,EAAI,EAAW,KAC7B,GAAM,GAAM,EAAY,EAAI,EAAY,GACxC,EAAM,KAAK,GACX,GAAgB,EAAM,EAAQ,GAGhC,GAAI,EAAe,GAAK,GAAgB,EAAa,EACnD,KAAM,IAAI,OACN,oBAAoB,EAAK,wBAAwB,GAGvD,OAAS,GAAI,EAAG,EAAI,EAAW,IAC7B,AAAI,EACF,EAAO,OAAO,EAAe,EAAY,IACrC,EAAY,EAAI,EAAY,GAEhC,EAAO,OAAO,EAAe,EAAY,GAAK,EAAQ,OAAS,EAC3D,EAAY,GACZ,EAAY,EAAI,EAAY,GAItC,MAAO,GAAO,WAAW,QAAQ,IAErC,GAzjFoC,EAAA,eC5DpC,YAqB8B,GAE5B,OADM,GAAe,GAAI,cAAa,EAAK,QAClC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EACjC,EAAa,GAAK,KAAK,IAAI,EAAK,IAElC,MAAO,GAGF,GAAM,IACT,SAAC,GACQ,GAAA,GAAA,EAAA,OAAA,EACD,EAAa,EAAK,QACpB,EAAe,GAAI,cAAa,EAAA,KAAK,cAAc,EAAE,QACzD,GAAI,EAAE,QAAU,aACd,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC7C,EAAe,GAAc,OAS7B,QAPM,GAAc,EAAW,KAAK,IAAI,EAAE,QACpC,EAAO,EAAY,mBAAmB,KACtC,EAAO,EAAY,mBAAmB,KACtC,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC/B,EACF,EAAW,KAAK,IAAI,EAAK,QAAQ,OAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,KACnC,GAAM,GAAO,EAAS,GAChB,EAAO,EAAS,GACtB,EAAa,GAAK,KAAK,MAAM,EAAM,GAGvC,MAAO,GAAW,WAAW,EAAc,EAAE,MAAO,YAG7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICzDd,YAwB6C,GAE3C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAmB,GACzB,GAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAE3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAEhC,EACF,EAAA,KAAK,uBAAuB,EAA0B,GAEpD,EAAQ,EAAO,OACf,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAE7D,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAG1D,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAE5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,KAX7B,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,IAA5B,GAeX,MAAO,CAAC,EAAQ,IClEpB,YAqBwB,GAEf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,KAAM,EAAA,EAAA,KAEP,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACzC,EAAW,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAEzC,EAAc,EAAQ,eAAe,EAAK,MAAO,aAEjD,EAAU,EAAQ,KAAK,IAAI,EAAY,QAK7C,SAAQ,mBAAqB,CAC3B,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,GACpD,KAAM,EAAQ,eAAe,EAAK,MAAO,UAAW,IAG/C,EAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC/Cd,YAsBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IClCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAGP,GAAI,IAAU,aACZ,GAAI,EAAE,QAAU,YACd,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAIvC,GAAM,GAAc,EAAA,MAAS,EAAE,OACzB,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,EAAE,MAAO,CAAC,MAAO,aAEpD,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAQ,KAAM,GAAc,QAAO,IAE/D,SAAY,UACZ,EAAQ,8BAA8B,GAE/B,EAIT,GAAI,EAAE,QAAU,aACd,GAAM,GAAW,GAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,IAC5C,EAAS,GAAK,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAO,EAAE,MAAO,CAAC,MAAK,KAElE,SAAQ,8BAA8B,GAE/B,EAGT,GAAI,CAAC,EAAA,KAAK,gBAAgB,EAAE,MAAO,IAGjC,GAAM,GAAS,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAC7C,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAO,MAAO,MAAK,GAG3D,GAAI,IAAU,SACZ,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAe,WAAW,KAAK,GACrC,MAAO,GAAQ,eAAe,EAAE,MAAO,QAAS,GAGlD,GAAI,IAAU,QAIZ,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAO,EAAA,KAAK,aAAa,CAAC,GAAI,EAAE,OAEhC,EAAA,GAAA,SAAA,EAAA,GAAA,MAAA,KAAA,EAAA,EAAA,IAAA,EAAA,MAAA,GAAA,EAAA,EAAA,QAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAQ,eAAe,EAAa,OAAQ,GAGrD,KAAM,IAAI,OAAM,iCAAiC,EAAE,MAAK,OAAO,GAG1D,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC9Fd,YAsCI,EAAc,EACd,EAAuC,GACzC,MAAI,IAAe,KACV,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAiB,CAAC,EAAG,GAAI,GAEzB,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,IAInD,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAa,EAEnB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,aACzC,GAAM,GAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAY,GACd,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAS,EAAY,MAAO,CAAC,MAAO,eAEnD,EAAgB,EAAW,KAAK,IAAI,EAAU,QAE9C,EAAQ,EAAc,mBAAmB,KACzC,EAAQ,EAAc,mBAAmB,KAEzC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAChC,EACF,EAAW,KAAK,IAAI,EAAM,QAAQ,OAEhC,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAgB,EAAA,EAAA,GAAgB,EAAA,EAAA,GAGjC,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAY,KAAM,GAAa,QAAS,IAE5D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,OAEP,GAAM,GAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAEtC,EAAS,GAAS,EAAE,MAEpB,EAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAGnB,MAAO,GAAW,eAAe,EAAa,EAAQ,iBASd,GAE5C,MAAO,UAAC,EAAkB,EAAkB,EACpC,EAAyB,EACzB,GACN,GAAM,GAAc,EAAA,aAAa,2BAA2B,EAAQ,GAC9D,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAY,OACzB,EAAgB,EAAA,KAAK,eAAe,GAEpC,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GACxD,EAAiB,EAAA,KAAK,uBAAuB,UAAW,GAExD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAEvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GACvD,EAAQ,EAAA,aAAa,uBAAuB,EAAW,GAEvD,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAE/B,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,EAAe,OAAS,EAAe,SAAW,EACpD,OAAS,GAAI,EAAG,EAAI,EAAe,OAAQ,KACzC,GAAM,GAAO,EAAI,EAAM,OACjB,EAAO,EAAI,EAAM,OAEjB,EACF,EAAG,EAAM,EAAO,GAAI,EAAM,EAAO,EAAI,GAAI,EAAM,EAAO,GACnD,EAAM,EAAO,EAAI,IAExB,EAAe,GAAK,EAAO,KAC3B,EAAe,GAAK,EAAO,SAG7B,oBAAS,GACP,GAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GAErC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,SAAA,GAAK,MAAA,GAAK,GAAK,IACtC,GAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAEtC,EACF,EAAG,EAAM,EAAS,GAAI,EAAM,EAAS,EAAI,GAAI,EAAM,EAAS,GACzD,EAAM,EAAS,EAAI,IAE1B,EAAe,GAAK,EAAS,KAC7B,EAAe,GAAK,EAAS,MAhBtB,EAAI,EAAG,EAAI,EAAe,OAAQ,MAAlC,GAmBX,MAAO,CAAC,EAAgB,EAAgB,ICvL5C,AAsBO,GAAM,IAAU,GAA8B,SAAC,EAAG,GAAM,MAAA,GAAI,IACtD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAGpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAwBsC,GAEpC,MAAO,UAAC,EAAQ,EAAO,GAGrB,OAFM,GACF,EAAA,KAAK,uBAAuB,EAA0B,EAAO,QACxD,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,IChCX,YAiCI,EAAc,EAA0B,GAC1C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAQlB,OALM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,GAAS,EAAE,MACpB,EAAY,EAAA,KAAK,kBAAkB,EAAQ,GACxC,EAAI,EAAG,EAAI,EAAO,EAAE,EAC3B,EAAU,GAAK,EAAG,EAAO,GAAI,GAE/B,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,gBAclD,EAAc,EAA4B,GAC5C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EAEP,GADA,GAAiB,EAAG,GAChB,EAAE,QAAU,UAAY,IAAU,SACpC,KAAM,IAAI,OAAM,wDAGlB,GAAM,GAAa,EACb,EAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAS,GAAS,EAAE,MACpB,EAAY,EAAU,EAAQ,EAAQ,GAC5C,MAAO,GAAW,eAAe,EAAE,MAAO,EAAQ,IC3EtD,AAsBO,GAAM,IAAW,GAAsB,SAAC,GAAO,MAAA,MAAK,KAAK,KACnD,GAAiB,GAAwB,EAAA,KAAM,IAE/C,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,MAAK,MAAM,KACrD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,AAsBO,GAAM,IAAU,GAAsB,SAAC,GAAO,MAAA,MAAK,IAAI,KACjD,GAAgB,GAAwB,EAAA,IAAK,IAE7C,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Bd,YAoBI,EAAmB,EAAoB,EACvC,GAIF,OAHM,GAAO,EAAA,KAAK,uBACd,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAGjC,OAFM,GAAS,EAAI,EACf,EAAM,EAAM,GACP,EAAI,EAAG,EAAI,EAAY,EAAE,GAChC,GAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACV,GAAM,GAGV,EAAK,GAAK,EAEZ,MAAO,GCpCT,AAqBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CACL,KAAM,EAAQ,EAAQ,EAAQ,EAC9B,KAAM,EAAQ,EAAQ,EAAQ,KAIzB,GACT,GAAiB,EAAA,SAAU,GAAc,IAEhC,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICrCd,AAsBO,GAAM,IAAY,GAAsB,SAAC,GAAO,MAAA,GAAI,KAAK,KAAK,KACxD,GAAkB,GAAwB,EAAA,MAAO,IAEjD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC5Bd,YAuBI,EAAkB,EAAiB,EAAgB,EACnD,GACF,GAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GAErC,GAAI,GACF,GAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAIhD,OADM,GAAU,EAAA,KAAK,uBAAuB,EAA0B,GAC7D,EAAI,EAAG,EAAI,EAAQ,EAAE,GAC5B,GAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,SAAC,EAAa,GAAM,MAAA,GAAM,EAAM,KAC/C,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEpB,MAAO,eAIL,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAAO,EAAA,EAAA,KAEd,GAAiB,EAAG,SAEd,GAAA,GAAA,EAAA,WAAA,iBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,EAAA,WAAW,kBAAkB,EAAG,EAAQ,GAExC,GAAM,GAAO,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAClC,EAAU,GAAU,EAAM,EAAQ,EAAO,EAAE,MAAO,EAAE,OAC1D,MAAO,GAAQ,eAAe,EAAO,EAAE,MAAO,GAGzC,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IClEd,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAQ,GAAW,MAAA,GAAS,IAClD,GACT,GAA+B,SAAC,EAAO,EAAO,EAAO,GACnD,MAAO,CAAC,KAAM,EAAQ,EAAO,KAAM,EAAQ,KAEpC,GAAM,GAAiB,EAAA,IAAK,GAAS,IAErC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICjCd,YAqBI,EAAmB,EAAkB,EAAiB,EACtD,GASF,OARM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GAEjC,EAAS,EAAA,KAAK,uBAChB,EAA0B,EAAA,KAAK,cAAc,IAExC,EAAI,EAAG,EAAI,EAAO,EAAE,GAK3B,OAJM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAGhC,EAAmB,GAAI,OAAM,EAAI,QAC9B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAO,GAAK,EAAI,EAAK,IAGvB,GAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE3B,MAAO,GC3CT,YAoBI,EAAuB,EAAc,EAAiB,GAgExD,OA1DM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAyDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GACtB,EAAI,EAAG,EAAI,EAAO,IACzB,EAAS,IAAM,EAAM,GAEvB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACxC,EAAS,IAAM,EAAM,GAevB,OAVM,GAA0C,GAG1C,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAA0B,GAC1B,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EAC/C,EAAI,EAAG,EAAI,EAAM,GAAQ,KAEhC,GAAI,GAAO,OACX,GAAI,EAEF,EAAU,EAAO,GAAG,gBAGpB,OADM,GAAa,GACV,EAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG1C,EAAU,EAAW,KAAK,KAI5B,GAAI,EAAe,KAAa,OAC9B,EAAQ,GAAK,EAAe,QAE5B,GAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAOvB,GAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,GAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,SAAC,EAAoB,GACzC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC/B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAOxE,GAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAE7B,CACL,aAAc,EAAa,OAC3B,YAAW,EACX,QAAO,GCzJX,+MCAA,GAGM,IAAU,QCHhB,AA8BA,EAAA,gBAAgB,MAAO,WAAM,MAAA,IAAI,KAAkB,GC9BnD,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,YAoBI,EAAqB,EAAkB,EAAiB,EACxD,EACA,GAsBF,OArBM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EACD,IAAa,MAAQ,OAAO,kBACP,OAAO,kBAE3B,EAAS,EAAA,OAAO,EAAS,SAAU,GACnC,EAAa,EAAO,OAEpB,EACF,EAAS,SAAS,GAAK,EAAS,SAAS,GAAK,EAAS,SAAS,GAC9D,EAAmB,EAAS,SAAS,GAAK,EAAS,SAAS,GAC5D,EAAmB,EAAS,SAAS,GAElC,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EAGxC,OAFM,GAAoB,EAAI,EACxB,EAAmB,EAAI,EAAQ,GAC5B,EAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,EAM1C,OALM,GAAW,EAAK,EAAe,EAC/B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAClD,EAAkB,EAAoB,EAAK,EACxC,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAQzC,OAPM,GAAW,EAAK,EAAc,EAC9B,EAAQ,KAAK,IAAI,EAAG,GACpB,EACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAc,EACd,EAAW,EACX,EAAQ,EACH,EAAK,EAAO,EAAK,EAAO,GAAM,GAErC,OADM,GAAW,EAAmB,EAAK,EAAQ,GACxC,GAAK,EAAO,GAAK,EAAO,IAAM,GACrC,GAAM,IAAW,EAAW,GAAK,EAAQ,GACnC,GAAQ,EAAQ,GAAW,GACjC,AAAK,IAAa,OAAS,GAAQ,EACjC,EAAc,GACT,AAAI,IAAa,OACtB,IAAY,GACZ,KAGJ,GAAI,MAAM,GACR,MAGJ,GAAM,IAAe,EAAkB,EAAK,EAAmB,EAC/D,EAAW,IACP,IAAa,MAAQ,EAAW,EAAQ,EAKpD,MAAO,eAIL,EAAqB,EAAkB,EACvC,EAAmC,EACnC,GADmC,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,IAYF,OAXM,GAAe,EAAA,OAAO,EAAS,SAAU,SACzC,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAE3B,EAAO,EAAA,OAAO,EAAQ,EAAO,GAC1B,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAK,EAAG,EAAK,EAAS,UAAW,EAAE,GAG1C,OAFM,GAAW,EAAK,EAAe,EACjC,EAAQ,EACL,EAAQ,GACb,GAAS,EAKX,OAFM,GACF,KAAK,IAAI,EAAS,SAAU,EAAwB,GAC/C,EAAK,EAAG,EAAK,EAAS,SAAU,EAAE,GAGzC,OAFM,GAAW,EAAK,EAAc,EAChC,EAAQ,EACL,EAAQ,GACb,GAAS,EAOX,OALM,GACF,KAAK,IAAI,EAAS,QAAS,EAAuB,GAClD,EAAW,OAAO,kBAClB,EAAc,GAET,EAAK,EAAO,EAAK,EAAO,GAAM,EAErC,OADM,GAAK,EAAK,EACP,EAAK,EAAO,EAAK,EAAO,GAAM,GACrC,GAAM,GAAK,EAAK,EACV,EAAQ,EAAK,IAAI,EAAG,EAAI,EAAI,GAClC,AAAI,EAAQ,GACV,GAAW,EACX,AAAI,EACF,EAAc,EACR,IAAI,EAAS,SAAW,GAAM,EAAS,QAAU,GAC3C,EAAS,WACb,EACH,GAAK,EAAS,QAAU,GAAM,EAAS,WAAa,EAEzD,EAAc,EAAK,EAAuB,GAKlD,EAAa,IAAI,EAAa,EAAG,EAAI,EAAI,IAKjD,MAAO,GCrJT,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAqBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBAyB9B,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAgB,EAAK,GAAe,GAEpC,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAGF,GAAM,GAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAGf,EAAG,IAAI,EAAU,EAAe,EAAG,EAAK,EAAK,GAKrD,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,IC1Fd,YAsBoC,GAK3B,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EAAG,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,KAAM,EAAA,EAAA,SAE/B,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGV,GAAiB,CAAC,EAAG,EAAM,EAAU,EAAO,GAAS,aAEhD,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAsBpB,OAnBM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAQ,KAAK,IAAI,EAAK,QAAQ,OACtC,EAAU,EAAQ,KAAK,IAAI,EAAS,QAAQ,OAC5C,EAAQ,EAAQ,EAAQ,KAAK,IAAI,EAAM,QAAQ,OAC/B,GAAI,cAAa,CAAC,IAClC,EAAU,EACZ,EAAQ,KAAK,IAAI,EAAO,QAAQ,OAChC,GAAI,cAAa,CAAC,IAChB,EAAU,GAAI,cAAa,EAAM,QAEjC,EAAgB,EAAQ,OACxB,EAAc,EAAM,OACpB,EAAgB,EAAQ,OACxB,EAAc,EAAM,OAEtB,EAAO,EACP,EAAK,EACL,EAAK,EACL,EAAK,EACA,EAAI,EAAG,EAAI,EAAM,OAAQ,EAAE,EAClC,EAAQ,GAAK,EAAQ,KAChB,GAAM,GAAK,EAAM,MAAS,EAAM,KAC7B,KAAK,KAAK,EAAQ,KAAQ,GAClC,AAAI,GAAQ,GACV,GAAO,GAET,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAEP,AAAI,GAAM,GACR,GAAK,GAGT,MAAO,GAAQ,eAAe,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAM,IAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,IC5Fd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,YAAa,SAAC,EAAI,GAC9D,GAAM,GAAY,EAClB,MAAI,GAAK,EAAU,aACV,EAAU,aAEZ,EAAK,EAAU,aAAe,EAAU,aAAe,IAGnD,GAA2B,CACtC,WAAY,EAAA,YACZ,YAAa,MACb,WAAY,IChCd,YAqBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAO,EAAQ,KAAK,IAAI,EAAM,QAAQ,mBAAmB,KACzD,EAAU,EAAQ,KAAK,IAAI,EAAK,QAAQ,OAK9C,MAAO,GAAQ,eAAe,EAAK,MAAO,EAAK,MAAO,GAGjD,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtCd,YAsBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MAED,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,EAAQ,OAAO,EAAE,QAEjB,GAAM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAEjC,GAAI,EAAM,oBAAsB,MAC9B,GAAM,GAAO,EAAM,mBAAmB,KAChC,EAAO,EAAM,mBAAmB,KAEtC,EAAK,MAAQ,EACb,EAAK,MAAQ,EAGf,MAAO,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,OAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,ICzDd,YA2BI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,KAED,EAAQ,EAAA,KAAK,eAAe,EAAM,EAAO,GAAG,OAAO,GACrD,EAAW,EAAA,aAAa,gBAAgB,EAAO,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtE,GAAI,EAAA,KAAK,cAAc,KAAc,EACnC,MAAO,GAAQ,eAAe,EAAU,EAAO,GAAG,MAAO,IAI3D,GAAM,GAAU,EAAO,OAAO,SAAA,GAAK,MAAA,GAAA,KAAK,cAAc,EAAE,OAAS,IACjE,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAGjB,GAAM,GAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAGlC,GAFA,EAAA,aAAa,uBAAuB,EAAQ,GAExC,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAC5D,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,IAAK,CAAC,OAAQ,CAAC,MAAO,GAAI,QAAO,MAE5D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAC3D,EAAe,GAAO,CAAC,OAAQ,EAAO,QAAO,EAAE,MAAO,CAAC,KAAI,KAE3D,EACF,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAO,IAEtE,SAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAM,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KACzD,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAUT,GAAM,GAAW,EAAQ,IAAI,SAAA,GAC3B,GAAM,GAAY,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,IAC7C,EAAQ,CAAC,GAAI,GACnB,MAAO,IAAQ,CAAC,OAAQ,CAAC,EAAG,GAAI,QAAO,EAAE,MAAO,CAAC,MAAK,OAIxD,EACI,EAAA,aAAa,gBAAgB,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAE7D,GAAM,GAAU,EAAA,KAAK,uBACjB,EAAQ,GAAG,MAAoB,EAAA,KAAK,cAAc,IAEtD,GAAI,EAAS,GAAG,MAAM,KAAO,GAE3B,GAAI,GAAS,EACb,EAAS,QAAQ,SAAA,GACf,GAAM,GAAM,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACjC,EAAO,EAAA,KAAK,cAAc,EAAE,OAElC,EAAQ,IAAI,EAAK,GACjB,GAAU,SAGZ,GAAI,GAAY,EAEhB,EAAS,QAAQ,SAAA,GAKf,OAJM,GAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OAErC,EAAO,EAEF,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EAEpC,OADM,GAAS,EAAM,EAAS,GAAK,EAC1B,EAAM,EAAG,EAAM,EAAE,MAAM,GAAI,EAAE,EACpC,EAAQ,EAAS,GAAO,EAAM,KAIlC,GAAa,EAAE,MAAM,KAIzB,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GAEtD,EACF,EAAQ,eAAe,EAAe,EAAO,GAAG,MAAO,GAE3D,SAAS,QAAQ,SAAA,GAAK,MAAA,GAAQ,8BAA8B,KAErD,EAGF,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,ICjId,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,SAAC,GAwCX,OAxCY,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OACJ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OACtC,EAAQ,EAAE,MAAM,OAEhB,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAChD,EAAa,EAAO,MAAM,OAE1B,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOI,EAAU,EAAA,KAAK,cAAc,GAC7B,EAAU,EAAS,OACnB,EAAa,EAAA,KAAK,kBAAkB,EAAE,MAAO,GAM1C,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,IAAO,EAAO,EAAc,EAAQ,KACjC,GAAI,EAAG,GAAI,EAAY,EAAE,IAEhC,OADI,IAAS,OAAO,iBACX,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,GAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAS,EAAA,KAAK,WAChB,CAAC,EAAG,GAAK,GAAK,IAAI,EAAO,EAAA,KAAK,eAAe,EAAE,QAC7C,GAAc,EAAA,KAAK,WACrB,CAAC,GAAG,GAAG,IAAI,EACX,EAAA,KAAK,eAAe,EAAO,QACzB,GAAM,EAAM,IAAU,EAAW,IACvC,AAAI,GAAM,IACR,IAAS,MAMnB,GAAM,IAAc,EAAA,KAAK,WACrB,CAAC,EAAG,EAAM,EAAM,IAAI,EAAS,EAAA,KAAK,eAAe,IACrD,EAAW,IAAe,GAMlC,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAY,EAAE,OAAQ,EAAU,EAAE,OAExD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAU,MAAO,EAAE,SCrG9C,AAsBO,GAAM,IAA+C,CAC1D,WAAY,EAAA,yBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,yBAAwB,QACtC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EAAY,EAAA,KAAK,0BACD,EAAO,MAAO,EAAO,OAOlC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAO,EACP,GAAO,EACF,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAO,GACP,GAAO,MAMjB,EAAU,IAAM,IAAM,IAAM,EAAI,GAAG,GAAM,GAAM,GAMvD,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAO,MAAO,EAAO,OAEhE,MAAO,CAAC,OAAM,GAAE,MAAO,EAAO,MAAO,MAAO,EAAO,SCrHvD,AAsBO,GAAM,IAA8C,CACzD,WAAY,EAAA,wBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,OAAQ,EAAA,EAAA,GAEZ,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,UACf,EAAa,EAEb,EACF,EAAA,KAAK,cACD,EAAE,MAAO,EAAW,KAAK,IAAI,EAAE,QAAQ,QAGzC,EAAU,EAAA,KAAK,cACD,EAAO,MACP,EAAW,KAAK,IAAI,EAAO,QAAQ,QAGjD,EAAA,EAAA,aAAA,sBAAA,EAAA,MAAA,EAAA,MAAA,EAAA,EAAA,OAAA,GACJ,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,WACA,EAAA,EAAA,UACA,EAAA,EAAA,SACA,EAAA,EAAA,QACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,YACA,EAAA,EAAA,eACA,EAAA,EAAA,cACA,EAAA,EAAA,SAOF,EAAA,KAAK,OACD,EAAG,OAAS,EAAS,OACrB,WAAM,MAAA,YAAY,EAAA,wBAAuB,QACrC,sCAAqC,EAAS,OAAM,cACpD,IAAG,EAAG,QAiBd,OAfM,GACF,EAAA,KAAK,cACD,EAAU,EAAW,KAAK,IAAI,EAAG,QAAQ,QAK3C,EACF,EAAA,KAAK,0BAA0B,EAAE,MAAO,EAAE,OAOrC,EAAI,EAAG,EAAI,EAAW,EAAE,EAC/B,OAAS,GAAO,EAAG,EAAO,EAAW,EAAE,EAErC,OADM,GAAO,EAAO,EAAe,EAAQ,IAClC,EAAO,EAAG,EAAO,EAAU,EAAE,EAEpC,OADM,GAAO,EAAO,EAAc,EAAQ,KACjC,EAAI,EAAG,EAAI,EAAY,EAAE,GAIhC,OAHI,IAAS,OAAO,iBAChB,GAAU,EAAO,EAAK,EAAI,EAC1B,GAAU,EAAO,EAAK,EAAI,EACrB,GAAI,EAAG,GAAI,EAAc,EAAE,IAClC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,EACpB,OAAS,IAAI,EAAG,GAAI,EAAa,EAAE,IACjC,GAAM,IAAM,EAAO,GAAI,EACvB,GAAI,IAAO,GAAK,GAAM,GACpB,GAAM,IAAM,EAAG,GAAG,IAAK,IAAK,GAAK,EAAQ,IAAG,IAAG,GAC/C,AAAI,GAAM,IACR,IAAS,GACT,GAAS,GACT,GAAS,MAMnB,EAAU,GAAG,IAAQ,IAAQ,IAAM,EAAI,GAAG,GAAM,GAAM,GAM9D,GAAM,IAAS,EAAW,MACtB,EAAA,KAAK,aAAa,EAAW,EAAE,OAAQ,EAAE,MAAO,EAAE,OAEtD,MAAO,CAAC,OAAM,GAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrH7C,AAsBO,GAAM,IACT,GAA6B,SAAC,EAAW,GAAc,MAAA,GAAI,IAClD,GAAM,GAAiB,EAAA,IAAK,IAE5B,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC7Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,IAAM,EAAI,EAAM,KAAK,IAAI,GAAM,IAEnD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC3Bd,AAqBA,GAAM,IAAI,EAAA,aAAa,MACjB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAClB,GAAK,EAAA,aAAa,OAEX,GAAgB,GACzB,EAAA,IACA,SAAC,GACC,GAAM,GAAO,KAAK,KAAK,GACjB,EAAI,KAAK,IAAI,GACb,EAAI,EAAO,GAAM,GAAI,GAC3B,MAAO,GACF,GACK,OAAK,EAAI,IAAM,EAAK,IAAM,EAAI,IAAM,EAAI,IAAM,EAC/C,KAAK,IAAI,CAAC,EAAI,MAIhB,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC5Cd,YAmCI,EAAmB,EACnB,GAgBF,OAfM,GAAa,EAAM,MACnB,EAAQ,EAAW,GACnB,EAAW,EAAW,GAEtB,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EAAS,EAAU,mBAAmB,KACtC,EAAS,EAAU,mBAAmB,KAGtC,EAAc,CAAC,EAAO,GACtB,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAA,KAAK,uBAAuB,UAAW,GACpD,EAAa,EAAA,KAAK,uBAAuB,UAAW,GAEjD,EAAI,EAAG,EAAI,EAAO,KAmBzB,OAjBM,GAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAE7B,EAAI,GAAM,CACd,OAAQ,CAAC,EAAG,GACZ,QAAS,EACT,MAAO,CAAC,MAAO,CAAC,EAAG,GAAI,KAAM,CAAC,EAAG,MAG7B,EAAQ,GAAQ,CAAC,OAAQ,CAAC,KAAM,EAAG,KAAM,GAAI,QAAS,IAGtD,EAAA,GAAA,EAAA,EAAA,GAAC,EAAA,EAAA,KAAM,EAAA,EAAA,KACP,EAAM,EAAA,aAAa,uBAAuB,EAAM,GAE7C,EAAI,EAAG,EAAI,EAAU,KAC5B,GAAM,GAAI,EAAA,aAAa,oBAAoB,EAAK,GAChD,EAAW,EAAI,EAAW,GAAK,EAAE,KACjC,EAAW,EAAI,EAAW,GAAK,EAAE,KAGnC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAG3C,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,GAChD,EACF,EAAW,eAAe,EAAa,UAAW,GAEhD,EAAS,GACX,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,GAAY,QAAS,IAE1D,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,cAIL,EAAmB,EACnB,GACF,GAAM,GAAY,EAAA,KAAK,cAAc,EAAM,OAErC,EAAY,EAAW,KAAK,IAAI,EAAM,QAEtC,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAG5D,EACF,EAAW,KAAK,IAAI,EAAU,mBAAmB,KAAK,QAAQ,OAGlE,GAAI,GAAc,IAChB,GAAM,GACF,GAAU,EAAU,EAAU,EAAW,EAAS,GAEhD,EAAc,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAEjD,GAAI,GACF,GAAM,GACF,EAAW,eAAe,EAAa,UAAW,EAAO,MACvD,EACF,EAAW,eAAe,EAAa,UAAW,EAAO,MAEvD,EAAuB,EAAW,eACpC,GAAI,UACJ,EAAA,KAAK,kBAAkB,EAA8B,YACnD,EACF,GAAS,CAAC,OAAQ,CAAC,EAAG,GAAW,QAAS,IAExC,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAW,QAAS,IAEhD,EACF,GAAU,WACN,CAAC,OAAQ,CAAC,EAAG,EAAU,EAAG,GAAe,QAAS,IAGpD,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OACtC,EACF,EAAW,KAAK,IAAI,EAAY,QAAQ,OAE5C,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GAElC,CAAC,KAAM,EAAa,KAAM,GAGnC,MAAO,QAEP,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EACF,GAAyB,EAAM,EAAW,GAE9C,MAAO,GAAA,aAAa,uBAAuB,IAI/C,YAAuB,GACrB,MAAQ,GAAO,EAAO,KAAO,EAI/B,YACI,EAAwB,EAAwB,EAChD,EACA,GACF,GAAI,IAAS,EACX,MAAO,CAAC,KAAM,EAAU,KAAM,GAGhC,GAAM,GAAO,EAAA,aAAa,uBAAuB,EAAU,GAErD,EAAO,EAAO,EAEd,EAAc,EAAA,aAAa,qBAAqB,GAEhD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAa,EAAA,aAAa,oBAAoB,GAE9C,EAAc,EAAW,KACzB,EAAc,EAAW,KAEzB,EAAW,CAAC,EAAY,QAExB,EACF,EAAW,eAAe,EAAU,UAAW,GAC7C,EACF,EAAW,eAAe,EAAU,UAAW,GAE7C,EAAgB,GAClB,CAAC,OAAQ,CAAC,KAAM,EAAa,KAAM,GAAc,QAAS,IAGxD,EACF,GAAU,EAAc,EAAc,EAAM,EAAS,GAEnD,EAAgB,EAAa,KAC7B,EAAgB,EAAa,KAE7B,EAAa,CAAC,EAAc,QAE5B,EACF,EAAW,eAAe,EAAY,UAAW,GAC/C,EACF,EAAW,eAAe,EAAY,UAAW,GAE/C,EAAkB,GAAQ,CAC9B,OAAQ,CAAC,KAAM,EAAe,KAAM,GACpC,QAAS,IAGL,EACF,GAAU,EAAa,EAAa,EAAM,EAAS,GAEjD,EAAe,EAAY,KAC3B,EAAe,EAAY,KAE3B,EAAY,CAAC,EAAa,QAE1B,EACF,EAAW,eAAe,EAAW,UAAW,GAC9C,EACF,EAAW,eAAe,EAAW,UAAW,GAE9C,EAAiB,GACnB,CAAC,OAAQ,CAAC,KAAM,EAAc,KAAM,GAAe,QAAS,IAE1D,EAAI,EAAA,aAAa,UAAU,EAAM,GACjC,EAAS,CAAC,EAAE,KAAK,QAEjB,EAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAC3D,GAAY,EAAW,eAAe,EAAQ,UAAW,EAAE,MAE3D,GAAc,GAChB,CAAC,OAAQ,CAAC,KAAM,EAAW,KAAM,IAAY,QAAS,IAEpD,GACF,GACI,CAAC,OAAQ,CAAC,EAAG,GAAa,EAAG,GAAiB,QAAS,IAGzD,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAErB,GAAU,GAAI,CACF,OAAQ,CAAC,EAAG,EAAiB,EAAG,IAChC,QAAS,IAGrB,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IACvD,GAAc,GAAK,CAAC,OAAQ,CAAC,MAAO,IAAU,QAAS,IAEvD,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAEV,GAAQ,GAAO,CACnB,OAAQ,CAAC,GAAuB,IAChC,QAAS,EACT,MAAO,CAAC,KAAM,KAGV,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAC9C,GAAY,EAAW,KAAK,IAAI,GAAM,QAAQ,OAEpD,SAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,GACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IACzC,EAAW,8BAA8B,IAElC,CAAC,KAAM,GAAW,KAAM,IAIjC,YACI,EAAkB,EAAc,GAGlC,OAFM,GAAM,GAAI,cAAa,EAAO,GAE3B,EAAI,EAAG,EAAI,EAAM,KAGxB,OAFI,GAAO,EACP,EAAO,EACF,EAAI,EAAG,EAAI,EAAM,KACxB,GAAM,GAAI,EAAA,aAAa,SAAS,EAAI,EAAG,EAAM,GACvC,EAAO,EAAA,aAAa,oBAAoB,EAAsB,GACpE,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAC3C,GAAQ,EAAK,KAAO,EAAE,KAAO,EAAK,KAAO,EAAE,KAE7C,AAAI,GACF,IAAQ,EACR,GAAQ,GAEV,EAAA,aAAa,mBAAmB,EAAK,EAAM,EAAM,GAEnD,MAAO,GCjVT,YAuBoB,GAEX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAO,GAElC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,ICtDd,AAsBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,MACb,WAAY,SAAC,GAUX,OAVY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GAEX,EAAS,KAAK,MAAM,EAAa,GACjC,EAAS,EAAc,EAAY,EAAY,EAEjD,EAAc,EAAU,GAE5B,GAAI,GAAU,GAAK,EAAS,GAE1B,GAAM,GAAmB,EAAS,EAC5B,EACF,EAAc,EAAY,EAAmB,EACjD,EAAc,EAAU,GAE1B,EAAO,GAAU,EAMzB,GAAM,GAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,EAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SCpErD,YAuBqB,GAEZ,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,MAED,EAAY,EAAA,KAAK,cAAc,EAAM,OAGrC,EAAqB,EAAM,MAAM,EAAM,MAAM,OAAS,GACtD,EAAQ,EAAY,EAEpB,EAAU,GAAQ,CACtB,OAAQ,CAAC,EAAG,GACZ,QAAO,EACP,MAAO,CAAC,MAAO,CAAC,EAAO,MAGnB,EAAS,GAAS,EAAS,GAAM,GAEjC,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAS,QAAO,EAAE,MAAO,CAAC,MAAO,EAAM,SAEhE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICtDd,AAqBO,GAAM,IACT,GAAgB,EAAA,SAAU,SAAC,GAAO,MAAA,QAAO,SAAS,GAAM,EAAI,GAAG,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,IAAI,KAAQ,SAAW,EAAI,GAAG,QAEzD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,QAAO,MAAM,GAAM,EAAI,GAAG,QAEhD,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAAO,MAAA,MAAK,MAAM,KAE5D,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IACT,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAK,EAAI,GAAG,QAEvC,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC3Bd,AA2BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAa,EACf,EAAS,EAAE,MACT,EAAQ,EAAO,OAEf,EAAW,EAAA,KAAK,eAAe,EAAkB,GACnD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACvD,EAAQ,EAAW,KAAK,IAAI,EAAE,QAAQ,OAC1C,GAAI,GAAgB,MAElB,OADM,GAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAO,EAAa,IAGpC,EAAQ,GAAc,EAAO,EAAQ,EAAE,MAAO,EAAc,GAC5D,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAElD,EAAS,EAGX,GAAiB,EAAG,OACpB,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGd,EAAa,EAAA,KAAK,cAAc,GAEhC,EAAS,GAAQ,EAAO,EAAY,EAAa,EAAE,OACnD,EAAS,EAAW,MAAM,EAAQ,EAAa,EAAE,OAEnD,EAAW,EACf,GAAI,GAEF,GAAM,GAAW,EAAA,aAAa,qBAAqB,EAAa,GAChE,EAAW,EAGb,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,SCtE9C,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GAChB,EAEJ,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,EAAM,GAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,SAEpC,GAAM,GAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAU,EAAA,KAAK,eAAe,EAAE,OAChC,EAAS,GAAK,EAAS,EAAE,MAAO,EAAE,MAAO,EAAS,EAAU,OAClE,EAAM,EAAQ,eACV,EAAS,SAAU,EAAE,MAAO,EAAO,QAEzC,MAAO,GAGF,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Dd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAyBlC,OAxBO,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAU,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACrC,EAAY,EAAA,OACd,EAAS,SAAU,EAAE,MACrB,GAAiB,EAAS,EAAE,MAAO,EAAE,MAAO,GAAU,QACpD,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAChC,EAAU,EAAuB,EAAI,EAAS,QAAQ,KACtD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EACF,EAAA,OAAgB,EAAE,MAA2C,WAE3D,EAAS,EAAQ,KAAK,IAAI,EAAG,QAAQ,OACrC,EAAQ,EAAA,OACV,EAAG,MAA2C,UAAW,GAEpD,EAAI,EAAG,EAAI,EAAS,UAAW,EAAE,EACxC,OAAS,GAAI,EAAG,EAAI,EAAS,WAAY,EAAE,EACzC,OAAS,GAAM,EAAG,EAAM,EAAS,SAAU,EAAE,EAC3C,OAAS,GAAM,EAAG,EAAM,EAAS,QAAS,EAAE,GAK1C,OAHM,GAAY,EAAM,EAClB,EAAY,EAAM,EACpB,EAAU,EACL,EAAK,EAAG,EAAK,EAAuB,GAAM,GACjD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,WAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,OAAS,GAAK,EAAG,EAAK,EAAsB,GAAM,GAChD,GAAM,GAAO,GAAY,GAAM,EAC/B,GAAI,EAAM,GAAK,GAAO,EAAS,UAC3B,KAAK,MAAM,KAAS,EACtB,SAEF,GAAM,GAAS,EAAwB,EAAuB,EACzD,EAAU,IAAI,EAAG,EAAK,EAAK,GAC1B,GAAS,EAAK,EAAuB,EAErC,GAAO,IAAW,GAAS,EAAI,EACrC,GAAI,KAAS,EACX,SAGF,GAAM,IAAQ,EAAM,IAAI,EAAG,EAAK,EAAK,GACrC,GAAW,GAAQ,IAGvB,EAAG,IAAI,EAAS,EAAG,EAAK,EAAK,GAKrC,MAAO,GAAQ,eAAe,EAAG,MAAO,EAAG,MAAO,EAAG,QAGhD,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,MACb,WAAY,ICnGd,YAoBI,EAAqB,EAAkB,EACvC,EAA8B,GAChC,GAAM,GAAU,EAAA,KAAK,eAAe,GAC9B,EAAW,GAAK,EAAS,EAAQ,EAAO,EAAS,EAAU,OAC3D,EAAe,GACjB,EAAS,EAAQ,EAAO,EAAU,GAAM,GAE5C,MAAO,CAAC,EAAS,OAAQ,EAAa,QC3BxC,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAa,EACnB,GAAiB,EAAG,qBAEpB,GAAM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,CAAC,EAAG,GAAI,GACN,EAAA,GAAA,EAAA,EAAA,MAAA,EAAA,MAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAGT,EACF,EAAW,MAAM,EAAwB,EAAS,SAAU,EAAE,OAC5D,EACF,EAAW,MAAM,EAAuB,EAAS,SAAU,EAAE,OACjE,MAAO,CACL,CAAC,OAAQ,EAAc,MAAO,EAAS,SAAU,MAAO,EAAE,OAC1D,CAAC,OAAQ,EAAe,MAAO,EAAS,SAAU,MAAO,YC/C/D,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEnB,GAAiB,EAAO,2BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC3C7B,AAoBA,GAAM,IAA0B,EAAA,aAAa,wBAIhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACvB,EAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEnB,GAAiB,EAAO,8BAExB,GAAM,GAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAC9C,EAAa,EAAW,KAAK,IAAI,EAAO,QAAQ,OAEhD,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAsBO,GAAM,IACT,GAA8B,SAAC,EAAG,GAAM,MAAC,KAAM,EAAK,EAAI,IAC/C,GACT,GAAiB,EAAA,SAAU,GAAc,KAAsB,QAEtD,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,IC9Bd,YAuBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,SAAU,EAAA,EAAA,cAEjB,GAAiB,EAAG,OAEpB,GAAM,GAAW,EAAS,IACtB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAE,MAAM,GAAK,EAAE,KAE9C,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAE5B,EAAQ,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACnC,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAQ,EAAE,MAAM,OAChB,EAAW,EAAA,KAAK,eAAe,EAAE,OAEjC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EACF,EAAA,KAAK,uBAAuB,EAAE,MAA0B,GAE5D,AAAI,IAAkB,GACpB,EAAQ,KAAK,GAGf,OAAS,GAAI,EAAG,EAAI,EAAO,KACzB,GAAM,GAAS,EAAA,KAAK,WAAW,EAAG,EAAO,GACnC,EAAY,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAI,EAAM,KAC3C,EAAW,EAAA,KAAK,WAAW,EAAW,EAAY,GAExD,EAAQ,GAAY,EAAM,GAG5B,GAAM,GAAQ,EAAQ,MAAM,EAAS,EAAU,EAAE,OAEjD,MAAO,CAAC,OAAQ,EAAO,MAAO,EAAU,MAAO,EAAE,OAG5C,GAAM,IAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICnEd,AAqBO,GAAM,IAAuB,GAAgB,EAAA,WAAY,SAAC,GAAO,MAAA,GAAI,IAE/D,GAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,MACb,WAAY,SAAC,GAiBX,OAjBY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAa,EAEb,EAAS,EAAA,KAAK,uBAChB,EAAM,MAA0B,EAAA,KAAK,cAAc,EAAM,QACvD,EAAA,EAAA,MAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAEjC,EAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAmB,IAEnB,EAAY,KAAK,IAAI,GACrB,EAAY,KAAK,IAAI,GACrB,EAAY,EAAW,KAAK,IAAI,EAAM,QAAQ,OAE3C,EAAW,EAAG,EAAW,EAAO,IAGvC,OAFM,GAAc,EAAW,EAAa,EAAc,EAEjD,EAAM,EAAG,EAAM,EAAa,IAGnC,OAFM,GAAY,EAAO,GAAa,GAE7B,EAAM,EAAG,EAAM,EAAY,IAGlC,OAFM,GAAY,EAAM,EAEf,EAAU,EAAG,EAAU,EAAa,KAC3C,GAAM,GAAS,CAAC,EAAO,EAAK,EAAK,GAE3B,EAAI,EAAO,GACX,EAAI,EAAO,GAGb,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACrD,EAAU,GAAI,GAAW,EAAa,GAAI,GAAW,EACzD,EAAS,KAAK,MAAM,EAAS,GAC7B,EAAS,KAAK,MAAM,EAAS,GAE7B,GAAI,GAAc,EAUlB,GATA,AAAI,MAAO,IAAc,UACvB,CAAI,IAAY,EACd,EAAc,EAEd,EAAc,EAAU,IAKxB,GAAU,GAAK,EAAS,GAAc,GAAU,GAChD,EAAS,GAEX,GAAM,GAAmB,EAAU,GAAa,GAC1C,EAAmB,EAAS,EAC5B,GACF,EAAc,EAAmB,EAAmB,EACxD,EAAc,EAAU,IAG1B,GAAM,IAAS,EAAc,EAAY,EAAY,EACrD,EAAO,IAAU,EAMzB,GAAM,IAAS,EAAW,MAAM,EAAQ,EAAM,MAAO,EAAM,OAC3D,MAAO,CAAC,OAAM,GAAE,MAAO,EAAM,MAAO,MAAO,EAAM,SC3FrD,AAqBO,GAAM,IAAkB,GAAgB,EAAA,MAAO,SAAC,GAErD,GAAM,GAAO,KAAK,MAAM,GACxB,MAAI,GAAK,EAAO,GACP,KAAK,MAAM,GACT,EAAK,EAAO,GACd,KAAK,KAAK,GAEb,EAAO,IAAQ,EACV,EAEA,EAAO,IAKP,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,MACb,WAAY,ICxCd,AAqBA,GAAM,IAAa,EAAA,aAAa,gBAC1B,GAAQ,EAAA,aAAa,WAEd,GAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,IAAM,EACD,GAAQ,EAER,GAAc,MAAK,IAAI,GAAM,KAI3B,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICnCd,AAqBO,GAAM,IACT,GAAgB,EAAA,QAAS,SAAC,GAAO,MAAA,GAAK,GAAI,KAAK,IAAI,CAAC,MAE3C,GAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,MACb,WAAY,IC3Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GACnD,MAAI,GAAK,EACA,GACE,EAAK,EACP,EAEA,IAIE,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IClCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AA0BA,GAAM,IAAU,sBACV,GAAY,KAAK,IAAI,IAAW,EAEzB,GAAqB,GAAgB,EAAA,SAAU,SAAC,GAG3D,GAAM,GAAW,EAAK,CAAC,GAIjB,EAAW,EAAK,GAEhB,EAAO,KAAK,IAAI,GAClB,EAEJ,MAAI,GACF,EAAS,EACJ,AAAI,EACT,EAAS,EAET,EAAS,KAAK,IAAI,EAAM,GAEnB,IAGI,GAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,MACb,WAAY,ICtDd,YAwB0B,GAKjB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,EACA,EAAA,EAAA,KAEP,GAAiB,EAAG,aAKpB,OAHM,GAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAS,GAAc,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAEvD,EAAS,EAAQ,MAAM,EAAQ,EAAU,EAAE,OACjD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAU,MAAO,EAAE,OAGrC,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,MACb,WAAY,ICpDd,YA0B+B,GAKtB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,WAAY,EAAA,EAAA,SAEnB,GAAiB,CAAC,GAAI,kBAEtB,GAAM,GAAO,EAAA,KAAK,cAAc,GAE1B,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAA0B,GAE1B,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,GAAY,WAAW,CACrC,OAAQ,CAAC,EAAC,GACV,QAAO,EACP,MAAO,CAAC,SAAU,EAAkB,cAAe,KAG/C,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EACF,EAAA,aAAa,oBAAoB,EAAQ,MAAO,EAAY,EAAM,IAEhE,EAA+B,CAAC,EAAG,GACnC,EAA6B,CAAC,MAAO,GACrC,EACF,GAAQ,CAAC,OAAQ,EAAe,QAAO,EAAE,MAAO,IAE9C,EAAmC,CAAC,EAAG,GACvC,EACe,CAAC,KAAM,GACtB,EACF,GAAU,CAAC,OAAQ,EAAiB,QAAO,EAAE,MAAO,IAElD,EAAqC,CAAC,EAAG,GACzC,EAAmC,CAAC,MAAO,GAC3C,EAAS,GACX,CAAC,OAAQ,EAAqB,QAAO,EAAE,MAAO,IAElD,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,EAGF,GAAM,IAAqC,CAChD,WAAY,EAAA,eACZ,YAAa,MACb,WAAY,ICvFd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,AAsBO,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,EACD,EAAa,EACnB,GAAiB,EAAG,UAIpB,OAFM,GAAS,EAAW,KAAK,IAAI,EAAE,QAAQ,OACvC,EAAY,GAAI,cAAa,EAAO,QACjC,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACnC,GAAM,GAAQ,EAAO,GACrB,EAAU,GAAK,EAAQ,EAEzB,GAAM,GAAS,EAAW,MAAM,EAAW,EAAE,MAAO,EAAE,OACtD,MAAO,CAAC,OAAM,EAAE,MAAO,EAAE,MAAO,MAAO,EAAE,SCrC7C,AAsBO,GAAM,IAAwB,GAA8B,SAAC,EAAG,GACrE,GAAM,GAAO,EAAI,EACjB,MAAO,GAAO,IAEH,GACT,GAAiB,EAAA,kBAAmB,IAE3B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,MACb,WAAY,IChCd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,EAAI,GACvD,GAAM,GAAY,EAClB,MAAI,OAAM,GACD,IAEA,EAAK,EAAI,EAAI,EAAU,QAIrB,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,ICjCd,AAqBO,GAAM,IAAgB,GAAgB,EAAA,IAAK,SAAC,GAAO,MAAA,MAAK,IAAI,KAEtD,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,MACb,WAAY,IC1Bd,AAqBO,GAAM,IAAiB,GAAgB,EAAA,KAAM,SAAC,GAAO,MAAA,MAAK,KAAK,KAEzD,GAA2B,CACtC,WAAY,EAAA,KACZ,YAAa,MACb,WAAY,IC1Bd,YAyBI,GAEK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAEpB,GAAM,GAAS,EAAQ,KAAK,IAAI,EAAE,QAAQ,OACpC,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,MACb,WAAY,IC5Cd,AA6FA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,0yDCvKjB,AAiBA,GAAM,IAAmD,GAEnD,GAA2C,CAC/C,MAAO,GACP,UAAW,GACX,mBAAoB,GACpB,sBAAuB,GACvB,MAAO,GACP,QAAS,GACT,6BAA8B,gBAI5B,EAAsB,GACxB,GAAS,GAAgB,cAGK,GAC9B,GAAI,CAAE,KAAgB,MACpB,GAAM,GAAS,GAAyB,GACxC,GAAI,IAAW,KACb,GAAS,GAAgB,MAEzB,gBAAQ,IAAI,0CAA2C,GAChD,KAGX,GAAM,GAAK,GAAS,GACpB,MAAI,GAAG,gBACL,OAAO,IAAS,GACT,GAAgB,IAGzB,GAAG,QAAQ,EAAG,YACd,EAAG,QAAQ,EAAG,cACd,EAAG,QAAQ,EAAG,OACd,EAAG,QAAQ,EAAG,QACd,EAAG,QAAQ,EAAG,qBACd,EAAG,QAAQ,EAAG,iBACd,EAAG,OAAO,EAAG,cACb,EAAG,OAAO,EAAG,WACb,EAAG,SAAS,EAAG,MAER,GAAS,IAGlB,YAAsB,GACpB,GAAI,MAAO,kBAAoB,aAAe,IAAiB,EAC7D,MAAO,IAAI,iBAAgB,IAAK,KAC3B,GAAI,MAAO,WAAa,YAC7B,MAAO,UAAS,cAAc,UAE9B,KAAM,IAAI,OAAM,0CAIpB,YAAkC,GAChC,GAAI,IAAiB,GAAK,IAAiB,EACzC,KAAM,IAAI,OAAM,0DAElB,GAAM,GAAS,GAAa,GAM5B,MAJA,GAAO,iBAAiB,mBAAoB,SAAC,GAC3C,EAAG,iBACH,MAAO,IAAS,IACf,IACC,IAAiB,EACX,EAAO,WAAW,QAAS,KAC3B,EAAO,WAAW,qBAAsB,IAG3C,EAAO,WAAW,SAAU,ICxFrC,AAoBA,GAAY,IAAZ,AAAA,UAAY,GAgBV,EAAA,EAAA,MAAA,GAAA,QAkBA,EAAA,EAAA,aAAA,GAAA,iBAlCU,IAAA,IAAa,KAqCzB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,OAAA,GAAA,SACA,EAAA,EAAA,SAAA,GAAA,aAJU,IAAA,IAAY,KAOxB,GAAY,IAAZ,AAAA,UAAY,GACV,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,iBAAA,GAAA,mBACA,EAAA,EAAA,yBAAA,GAAA,2BACA,EAAA,EAAA,mBAAA,GAAA,qBACA,EAAA,EAAA,mBAAA,GAAA,uBALU,IAAA,IAAmB,iBAqC3B,EAAc,GAChB,MAAO,CAAC,EAAS,eAIf,EAAoB,GACtB,MAAO,GAAa,cAWW,GAC/B,GAAM,GAAO,EAAA,KAAK,cAAc,GAC1B,EAAe,KAAK,KAAK,EAAO,GACtC,MAAO,GAAA,KAAK,oBAAoB,eA6B9B,EAAc,GAChB,MAAO,CACL,KAAK,IAAI,EAAG,KAAK,KAAK,EAAU,IAAK,KAAK,IAAI,EAAG,KAAK,KAAK,EAAO,iBAKlE,EAAc,GACV,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GACV,MAAO,GAAI,EAAI,cAsBb,EAA2B,GAE7B,GAAM,GAAQ,EAEV,EACA,EACA,EACA,EACA,EAEA,EACA,EAEA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAsB,EAAM,KAC5B,EAA0B,EAAM,KAChC,EAAgC,EAAM,QACtC,EAA4B,EAAM,QAClC,EAAqB,EAAM,IAC3B,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,EAAM,WAC7B,EAAmB,EAAM,OAEzB,GAAsB,EAAG,KACzB,EAA0B,EAAG,KAC7B,EAAgC,EAAG,KACnC,EAA4B,EAAM,KAClC,EAAqB,EAAG,KACxB,EAA4B,EAC5B,EAAqB,EACrB,EAAuB,GAA6B,KAChD,EAA0B,eAC1B,KACJ,EAAmB,EAAG,OAExB,EAAwB,EAAG,KAEpB,CACL,oBAAmB,EACnB,wBAAuB,EACvB,8BAA6B,EAC7B,0BAAyB,EACzB,mBAAkB,EAClB,sBAAqB,EACrB,0BAAyB,EACzB,mBAAkB,EAClB,qBAAoB,EACpB,iBAAgB,GCzOpB,YAsBgC,EAA2B,GACzD,GAAM,GAAc,IACpB,MAAI,GAAA,MAAM,QAAQ,UAChB,GAAgB,GAEX,EAGT,YAAyB,GACvB,GAAM,GAAQ,EAAG,WACjB,GAAI,IAAU,EAAG,SACf,KAAM,IAAI,OAAM,gBAAkB,GAAqB,EAAI,IAK/D,GAAM,IAAc,QACd,GAAc,kBAEa,GAC/B,MAAI,KAAA,MAAM,QAAQ,iCAAmC,IAAQ,GACxD,GAAc,KAAK,IAAI,IAAQ,KAAK,IAAI,GAAO,gBAOlD,EAA2B,GAC7B,OAAQ,OACD,GAAG,SACN,MAAO,eACJ,GAAG,aACN,MAAO,mBACJ,GAAG,cACN,MAAO,oBACJ,GAAG,kBACN,MAAO,wBACJ,GAAG,8BACN,MAAO,oCACJ,GAAG,cACN,MAAO,oBACJ,GAAG,mBACN,MAAO,6BAEP,MAAO,sBAAsB,eAK/B,EAA2B,GAC7B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,aAAa,IAC1B,cAAgB,EAAgB,gDAIlC,EAA2B,GAC7B,GAAM,GAA4B,GAC9B,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,gBAC7B,wCAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAc,KACrD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAc,EAAG,kBAAoB,GAC7D,cAAQ,IAAI,EAAG,iBAAiB,IAC1B,GAAI,OAAM,oCAElB,MAAO,eAIL,EAA2B,GAC7B,GAAM,GAA8B,GAChC,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,kBAC7B,0CAGJ,GAFA,GAAa,EAAI,WAAM,MAAA,GAAG,aAAa,EAAgB,KACvD,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,KACpC,EAAG,mBAAmB,EAAgB,EAAG,kBAAoB,GAC/D,SACI,EAAsB,EAAG,iBAAiB,IACxC,GAAI,OAAM,sCAElB,MAAO,GAGT,GAAM,IAAkB,2BACxB,YACI,EAAsB,GACxB,GAAM,GAAwB,GAAgB,KAAK,GACnD,GAAI,GAAyB,MAC3B,QAAQ,IAAI,wCAAwC,GACpD,QAAQ,IAAI,GACZ,OAWF,OARM,GAAa,CAAC,EAAsB,GAEpC,EAAc,EAAa,MAAM;GACjC,EAAM,EAAY,OAAO,WAAW,OAAS,EAC7C,EAAuB,EAAY,IACrC,SAAC,EAAM,GACH,MAAA,GAAA,KAAK,SAAU,GAAa,GAAG,WAAY,GAAO,IACtD,EAAgB,EACX,EAAI,EAAG,EAAI,EAAqB,OAAQ,IAC/C,EAAgB,KAAK,IAAI,EAAqB,GAAG,OAAQ,GAG3D,GAAM,GAAmB,EAAqB,MAAM,EAAG,EAAa,GAC9D,EAAY,EAAqB,MAAM,EAAa,EAAG,GACvD,EAAkB,EAAqB,MAAM,GAEnD,QAAQ,IAAI,EAAiB,KAAK;IAClC,QAAQ,IAAI,EAAc,MAAM;GAAM,IACtC,QAAQ,IACJ,MAAM,EAAA,KAAK,SAAS,EAAU,GAAI,GAClC,iEACJ,QAAQ,IAAI,EAAgB,KAAK;gBAGL,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGR,EAA2B,GAErD,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,KAClC,EAAG,oBAAoB,EAAS,EAAG,eAAiB,GACtD,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,2DAKhB,EAA2B,GAE7B,GADA,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,KACtC,EAAG,oBAAoB,EAAS,EAAG,mBAAqB,GAC1D,cAAQ,IAAI,EAAG,kBAAkB,IAC3B,GAAI,OAAM,iDAKhB,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,EAAM,EAAG,eACxD,cAIL,EAA2B,GAC7B,GAAM,GAAsB,GACxB,EAAI,WAAM,MAAA,GAAG,gBAAgB,gCACjC,UAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,KAC9D,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,EAAM,EAAG,eACvD,gBAIP,MAAI,GAAA,MAAM,UAAU,mBAAqB,EAChC,EAEF,cAGqB,GAC5B,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,iBAAiB,8CAGA,EAAe,GACjD,GAAM,GAAiB,EAAA,MAAM,UAAU,0BACvC,GAAK,GAAS,GAAO,GAAU,GAC7B,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IACrC,KAAM,IAAI,OAAM,0BAA4B,EAAY,gBAE1D,GAAK,EAAQ,GAAoB,EAAS,GACxC,GAAM,GAAY,IAAI,EAAK,IAAI,EAAM,IAC/B,EAAM,IAAI,EAAc,IAAI,EAAc,IAChD,KAAM,IAAI,OACN,0BAA4B,EAC5B,qDAAuD,EAAM,kBAInC,GAChC,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,qBAAqB,kDAIpC,EAA2B,EAAuB,EAClD,EAAqB,EAA6B,EAClD,GACF,GAAM,GAAM,EAAG,kBAAkB,EAAS,GAC1C,MAAI,KAAQ,GAGH,GAET,IAAa,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KACtD,GACI,EACA,WAAM,MAAA,GAAG,oBACL,EAAK,EAAqB,EAAG,MAAO,GAAO,EAC3C,KACR,GAAa,EAAI,WAAM,MAAA,GAAG,wBAAwB,KAC3C,gBAIL,EAA2B,EAAuB,GACpD,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,iBAInD,EAA2B,GAC7B,GAAoB,EAAI,GACxB,GAAa,EAAI,WAAM,MAAA,GAAG,cAAc,EAAG,SAAW,KACtD,GAAa,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAInD,EAA2B,EAC3B,GACF,MAAO,IACH,EAAI,WAAM,MAAA,GAAG,mBAAmB,EAAS,IACzC,YAAc,EAAc,yCAI9B,EAA2B,EAC3B,GACF,MAAO,GAAG,mBAAmB,EAAS,eAIpC,EAA2B,EAC3B,EAA8C,GAChD,GAAa,EAAI,WAAM,MAAA,IAAgB,EAAI,EAAS,KACpD,GAAa,EAAI,WAAM,MAAA,GAAG,UAAU,EAAwB,iBAGtB,GACtC,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QAC1D,GAAa,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,UACpE,GAAa,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAG,OAAO,MAAO,EAAG,OAAO,sBAIjE,EAA2B,EAC3B,GACF,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,iBAIpE,EAA2B,GAC7B,GAAa,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,KAC1D,GACI,EACA,WAAM,MAAA,GAAG,qBACL,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,KAAM,iBAGjC,GAClC,GAAM,GAAS,EAAG,uBAAuB,EAAG,aAC5C,GAAI,IAAW,EAAG,qBAChB,KAAM,IAAI,OACN,8BAAgC,GAA2B,EAAI,gBAKnE,EAA2B,GAC7B,OAAQ,OACD,GAAG,kCACN,MAAO,wCACJ,GAAG,0CACN,MAAO,gDACJ,GAAG,kCACN,MAAO,wCACJ,GAAG,wBACN,MAAO,kCAEP,MAAO,iBAAiB,GAI9B,YACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAa,EAAI,WAAM,MAAA,OAC/C,GAAI,GAAW,KACb,KAAM,IAAI,OAAM,GAElB,MAAO,GAGT,YAA6B,EAA2B,GACtD,GAAM,GAAiB,EAAG,iCAAmC,EACvD,EAAgB,EAAc,EAAG,SACvC,GAAI,EAAgB,EAAG,UAAY,EAAgB,GACjD,GAAM,GAAmB,2BAA2B,EAAc,IAClE,KAAM,IAAI,OAAM,0BAA0B,EAAgB,kBAIlC,EAAiB,GAAA,MAAA,KAAA,QAAA,GAAA,GACpC,EAAA,KAAK,cAAc,EAAM,MAAM,EAAG,EAAM,OAAS,gBAG9B,GAC1B,GAAI,EAAM,SAAW,EACnB,KAAM,OAAM,wDAGd,MAAO,CACL,EAAM,OAAS,EAAI,EAAM,EAAM,OAAS,GAAK,EAAG,EAAM,EAAM,OAAS,gBAI5C,GAC3B,GAAI,GAAsC,CAAC,EAAG,EAAG,GAC3C,EAAW,EAAM,SAAW,GAAM,EAAM,SAAW,GAAK,EAAM,KAAO,EAC3E,MAAK,IACH,GACI,CAAC,GAAY,IAAM,OAAK,GAAY,KAEnC,cAIL,EAAoB,SAAA,AAAA,IAAA,QAAA,GAAA,IACtB,GAAI,GAAa,EAAA,MAAM,UAAU,0BAsBjC,GArBA,AAAI,GACF,GAAa,EAAa,EAO1B,EAAW,EAAS,IAChB,SAAC,EAAG,GAAM,MAAA,IAAK,EAAS,OAAS,EAC7B,EAAA,KAAK,kBAAkB,EAAS,IAChC,EAAS,KAIjB,AAAI,EAAS,SAAW,GACtB,GAAW,CAAC,EAAG,EAAS,MAKxB,EAAS,SAAW,GACtB,GAAM,GAAgB,EAAA,KAAK,aAAa,GACxC,EAAW,EAAc,SAG3B,GAAI,GAAO,EAAA,KAAK,cAAc,GAC9B,GAAI,EAAS,QAAU,GAAK,GAAQ,EAClC,MAAO,CAAC,EAAG,GACN,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,IAAM,EACjB,MAAO,GACF,GACH,EAAS,SAAW,GAAK,EAAS,GAAK,EAAS,IAAM,GACtD,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAI,EAAS,IACvC,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,IAAM,EAC/B,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,IACvC,GACH,EAAS,SAAW,GACpB,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,GAC3C,EAAS,IAAM,EACjB,MAAO,CAAC,EAAS,GAAK,EAAS,GAAK,EAAS,GAAI,EAAS,IACrD,GACH,EAAS,SAAW,GAAK,EAAS,IAAM,GACxC,EAAS,GAAK,EAAS,GAAK,EAAS,IAAM,EAC7C,MAAO,CAAC,EAAS,GAAI,EAAS,GAAK,EAAS,GAAK,EAAS,IAE1D,GAAI,GAOF,GAAM,GAAW,GAAY,GACzB,EAAO,EAAG,EAAO,EACrB,MAAI,GAAS,QACX,GAAA,GAAA,GAAC,EAAA,EAAA,GAAM,EAAA,EAAA,IAET,EAAO,EAAY,GAAO,GAAM,GAAO,GAChC,EAAA,KAAK,oBAAoB,GAAM,IAAI,SAAA,GAAK,MAAA,GAAI,IAErD,MAAO,GAAA,KAAK,oBAAoB,GAIpC,YAAgB,GACd,MAAO,GAAI,IAAM,cAOW,EAAkB,GAI9C,GAHA,EAAS,EAAO,MAAM,IACtB,EAAS,EAAO,MAAM,IAElB,EAAA,KAAK,YAAY,EAAQ,GAC3B,MAAO,GAGT,GAAI,CAAC,EAAO,QAAU,CAAC,EAAO,OAC5B,MAAO,GAGT,GAAI,EAAO,KAAO,GAAK,EAAO,KAAO,GAAK,EAAO,KAAO,GACpD,EAAO,KAAO,EAChB,MAAO,GAGT,GAAI,EAAO,SAAW,EAAO,QAC3B,GAAM,GAAa,EAAO,MAAM,IAAI,GAC9B,EAAa,EAAO,MAAM,IAAI,GACpC,GAAI,IAAe,EACjB,MAAO,GAGT,GAAI,GAAO,IAAe,GAAO,IAC5B,GAAO,KAAO,GAAK,EAAO,KAAO,GACpC,MAAO,GAGX,MAAO,GAAO,KAAO,EAAO,IAAM,GAAO,EAAO,KAAO,GAAO,EAAO,IAMvE,GAAI,IACA,eAEmC,GACrC,GAAI,IAAoB,MACtB,GAAM,GAAK,GAAgB,GAC3B,GAAmB,EAAG,aAAa,EAAG,kBAExC,MAAO,kBAIP,GAAmB,mBAGnB,GAAyB,iBAGY,GACrC,GAAI,IAA0B,MAC5B,GAAM,GAAK,GAAgB,GAC3B,GAAyB,EAAG,aAAa,EAAG,yBAG9C,MAAO,MAAK,IAAI,GAAI,gBAG4B,GAEhD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAI,GACE,EAAK,GAAgB,GAE3B,MAAI,IAAa,EAAI,oCACjB,IAAiB,EACnB,EAAoB,EACf,AAAI,GAAa,EAAI,4BAC1B,EAAoB,EAEpB,EAAoB,EAEf,cAGoB,EAA2B,GACtD,GAAM,GAAM,EAAG,aAAa,GAC5B,MAAO,IAAO,iBAGsB,GACpC,IACE,GAAM,GAAK,GAAgB,GAC3B,GAAI,GAAM,KACR,MAAO,SAEF,GACP,eAAQ,IAAI,qCAAsC,GAC3C,GAET,MAAO,eAG0C,GAEjD,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,WAGL,CAAC,GAAa,EAAI,0BACpB,MAAO,GAIX,GAAM,GAAwB,GAAuC,GACrE,MAAO,eAYqC,GAC5C,GAAI,IAAiB,EACnB,MAAO,GAGT,GAAM,GAAK,GAAgB,GAE3B,GAAI,IAAiB,GACnB,GAAI,CAAC,GAAa,EAAI,qBACpB,MAAO,GAET,GAAI,CAAC,GAAa,EAAI,4BACpB,MAAO,QAGT,GAAI,GAAa,EAAI,0BACnB,MAAO,IAAuC,GAGhD,GAAM,GAA0B,8BAChC,GAAI,GAAa,EAAI,IACnB,GAAM,GACF,EAAG,aAAa,GACpB,MAAO,IACH,EAAI,GAGV,MAAO,GAGT,GAAM,GAAwB,GAAuC,GACrE,MAAO,GAGT,YAAgD,GAE9C,GAAM,GAAY,GAAiB,GAE7B,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,oBAAqB,EAAO,EAAQ,EAChE,EAAU,mBAAoB,EAAU,iBAAkB,MAE9D,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,EAGT,YAEI,EAA2B,GAC7B,GAAM,GAAY,GAAiB,EAAI,GACjC,EAAU,EAAG,gBACnB,EAAG,YAAY,EAAG,WAAY,GAE9B,GAAM,GAAQ,EACR,EAAS,EACf,EAAG,WACC,EAAG,WAAY,EAAG,EAAU,wBAAyB,EAAO,EAAQ,EACpE,EAAU,mBAAoB,EAAU,qBAAsB,MAElE,GAAM,GAAc,EAAG,oBACvB,EAAG,gBAAgB,EAAG,YAAa,GACnC,EAAG,qBACC,EAAG,YAAa,EAAG,kBAAmB,EAAG,WAAY,EAAS,GAElE,GAAM,GACF,EAAG,uBAAuB,EAAG,eAAiB,EAAG,qBAErD,SAAG,YAAY,EAAG,WAAY,MAC9B,EAAG,gBAAgB,EAAG,YAAa,MACnC,EAAG,cAAc,GACjB,EAAG,kBAAkB,GAEd,cAG2B,GAClC,GAAI,IAAiB,EACnB,MAAO,GAET,GAAM,GAAK,GAAgB,GAGrB,EAAa,EAAW,WAAa,KAC3C,MAAO,eAIL,EAAiC,GACnC,AAAK,MAAM,QAAQ,IACjB,GAAS,CAAC,IAEZ,EAAO,QAAQ,SAAA,GACb,AAAI,GAAK,MACP,EAAA,KAAK,OACD,EAAE,QAAU,YACZ,WAAM,MAAG,GAAM,smCC7qBzB,AAqBA,GAAM,IAAM,EAAA,MASZ,GAAI,aAAa,YAAa,WAAM,MAAA,IAAI,UAAU,iBAAmB,IAGrE,GAAI,aAAa,gBAAiB,WAChC,MAAI,IAAsB,GACjB,EACE,GAAsB,GACxB,EAEF,IAIT,GAAI,aAAa,iCAAkC,WAAM,MAAA,KAEzD,GAAI,aACA,yBAA0B,WAAM,MAAA,IAAI,IAAI,mBAAqB,IAGjE,GAAI,aAAa,oBAAqB,WAAM,MAAA,KAG5C,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aAAa,aAAc,WAAM,MAAA,IAAI,QAAQ,eAGjD,GAAI,aAAa,2BAA4B,WAAM,MAAA,IAAI,QAAQ,gBAG/D,GAAI,aAAa,kBAAmB,WAAM,MAAA,IAAI,QAAQ,gBAItD,GAAI,aAAa,2BAA4B,WAAM,MAAA,KAGnD,GAAI,aACA,+BAAgC,WAAM,MAAA,IAAI,QAAQ,gBAGtD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aACA,8BAA+B,WAAM,MAAA,IAAI,QAAQ,gBAGrD,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aAAa,sBAAuB,WAAM,MAAA,IAAI,QAAQ,gBAG1D,GAAI,aAAa,oBAAqB,WAAM,MAAA,IAAI,QAAQ,gBAGxD,GAAI,aACA,yBACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAG/C,GAAI,aACA,+BACA,WAAM,MAAA,IAAuB,GAAI,UAAU,oBAU/C,GAAI,aAAa,+CAAgD,WAC/D,GAAM,GAAe,GAAI,UAAU,iBAEnC,MAAI,KAAiB,EACZ,EAEF,GAAkC,KAO3C,GAAI,aACA,gDACA,WAAM,MAAA,IAAI,UAAU,gDAAkD,GAClE,CAAC,EAAA,YAAY,aAKrB,GAAI,aACA,+BACA,WAAM,MAAA,IAAmC,GAAI,UAAU,oBAM3D,GAAI,aAAa,+BAAgC,WAC/C,MAAO,IAAI,QAAQ,4BACf,GACA,GAAI,QAAQ,kCAOlB,GAAI,aACA,+BACA,WAAM,MAAA,IAA8B,GAAI,UAAU,oBAGtD,GAAI,aACA,0BACA,WAAM,MAAA,IAAoB,GAAI,UAAU,oBAK5C,GAAI,aAAa,4BAA6B,WAK5C,GAAM,GAAc,GAAI,QAAQ,gCAChC,MAAO,GAAc,EAAI,IAU3B,GAAI,aACA,iCACA,WACE,MAAO,IAET,SAAA,GACE,GAAI,EAAY,GAAK,IAAc,GACjC,KAAM,IAAI,OACN,+DACA,mCAAkC,EAAS,QC1LvD,AAiBO,YAAuB,GAC1B,KAAM,GAAe,GAAI,cAAa,EAAK,QAC3C,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,EAC/B,EAAa,GAAK,KAAK,IAAI,EAAK,IAEpC,MAAO,GCtBX,AAoBO,YAAsC,GACzC,MAAO,CAAC,EAAQ,EAAQ,EAAO,EAAO,KAClC,KAAM,GAAW,EAAA,aAAa,2BAA2B,EAAQ,GAC3D,EAAa,EAAS,OACtB,EAAgB,EAAA,KAAK,eAAe,GACpC,EAAa,EAAA,KAAK,cAAc,GAChC,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAQ,EAAO,OACf,EAAQ,EAAO,OACf,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GACvD,EAAiB,EAAA,aAAa,iBAAiB,EAAQ,GAC7D,GAAI,EAAe,OAAS,EAAe,SAAW,EAClD,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAO,GAAK,EAAG,EAAM,EAAI,EAAM,QAAS,EAAM,EAAI,EAAM,aAI5D,QAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,GACjC,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAY,GACrC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GACtC,EAAO,EAAI,MAAM,CAAC,GACxB,EAAe,QAAQ,GAAK,EAAK,GAAK,GACtC,KAAM,GAAS,EAAA,KAAK,WAAW,EAAM,EAAO,GAC5C,EAAO,GAAK,EAAG,EAAM,GAAS,EAAM,IAG5C,MAAO,CAAC,EAAQ,IClDxB,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAG,IAAM,EAAI,GCnBnE,AAoBO,YAA+B,GAClC,MAAO,CAAC,EAAQ,EAAO,KACnB,KAAM,GAAY,EAAA,KAAK,uBAAuB,EAAO,EAAO,QAC5D,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACjC,EAAU,GAAK,EAAG,EAAO,GAAI,GAEjC,MAAO,IC1Bf,AAmBO,KAAM,IAAW,GAAsB,AAAC,GAAO,KAAK,KAAK,ICnBhE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,KAAK,MAAM,ICnBlE,AAmBO,KAAM,IAAU,GAAsB,AAAC,GAAO,KAAK,IAAI,ICnB9D,AAiBO,YAAiB,EAAO,EAAY,EAAU,GACjD,KAAM,GAAO,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACnE,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,EAAE,GAC/B,KAAM,GAAS,EAAI,EACnB,GAAI,GAAM,EAAM,GAChB,OAAS,GAAI,EAAG,EAAI,EAAY,EAAE,GAC9B,KAAM,GAAQ,EAAM,EAAS,GAC7B,AAAI,EAAQ,GACR,GAAM,GAGd,EAAK,GAAK,EAEd,MAAO,GC9BX,AAmBO,KAAM,IAAe,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBvF,AAmBO,KAAM,IAAY,GAAsB,AAAC,GAAO,EAAI,KAAK,KAAK,ICnBrE,AAkBO,YAAmB,EAAM,EAAO,EAAM,EAAO,GAChD,KAAM,GAAc,EAAA,WAAW,iBAAiB,EAAO,EAAO,GACxD,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAW,EAAA,KAAK,eAAe,GACrC,GAAI,GACA,KAAM,GAAa,EAAA,WAAW,kBAAkB,EAAO,GACvD,MAAO,GAAK,SAAS,EAAY,EAAa,GAElD,KAAM,GAAU,EAAA,KAAK,uBAAuB,EAAO,GACnD,OAAS,GAAI,EAAG,EAAI,EAAQ,EAAE,GAC1B,KAAM,GAAO,EAAK,OACZ,EAAU,EAAA,KAAK,eAAe,GAC9B,EAAM,EAAA,KAAK,WAAW,EAAG,EAAM,GAC/B,EAAO,EAAI,IAAI,CAAC,EAAK,IAAM,EAAM,EAAM,IACvC,EAAS,EAAA,KAAK,WAAW,EAAM,EAAM,OAAQ,GACnD,EAAQ,GAAK,EAAK,GAEtB,MAAO,GCnCX,AAmBO,KAAM,IAAU,GAA8B,CAAC,EAAQ,IAAW,EAAS,GCnBlF,AAiBO,YAAuB,EAAO,EAAQ,EAAO,EAAM,GACtD,KAAM,GAAQ,EAAO,OACf,EAAQ,EAAA,KAAK,cAAc,GAC3B,EAAW,EAAA,KAAK,eAAe,GAC/B,EAAa,EAAA,KAAK,eAAe,GACjC,EAAS,EAAA,KAAK,uBAAuB,EAAO,EAAA,KAAK,cAAc,IACrE,OAAS,GAAI,EAAG,EAAI,EAAO,EAAE,GACzB,KAAM,GAAM,EAAA,KAAK,WAAW,EAAG,EAAO,GAEhC,EAAS,GAAI,OAAM,EAAI,QAC7B,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAO,GAAK,EAAI,EAAK,IAEzB,KAAM,GAAW,EAAA,KAAK,WAAW,EAAQ,EAAO,GAChD,EAAO,GAAY,EAAM,GAE7B,MAAO,GCjCX,AAiBO,YAAoB,EAAQ,EAAM,EAAO,GAE5C,KAAM,GAAQ,EAAA,KAAK,eAAe,EAAM,GAAO,GAwDzC,EAAW,CAAC,EAAG,EAAM,GAAI,GAC/B,OAAS,GAAI,EAAG,EAAI,EAAO,IACvB,EAAS,IAAM,EAAM,GAEzB,EAAS,GAAK,EAAM,GACpB,OAAS,GAAI,EAAQ,EAAG,EAAI,EAAM,OAAQ,IACtC,EAAS,IAAM,EAAM,GAIzB,KAAM,GAAiB,GAGjB,EAAU,GAAI,YAAW,EAAM,IAE/B,EAAc,GAAI,GAAA,aAAa,EAAU,EAAO,GAGhD,EAAgB,GAChB,EAAa,EAAS,KAAO,GAAK,EAAS,KAAO,EACxD,OAAS,GAAI,EAAG,EAAI,EAAM,GAAQ,KAE9B,GAAI,GACJ,GAAI,EAEA,EAAU,EAAO,GAAG,gBAGpB,KAAM,GAAa,GACnB,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAW,KAAK,EAAY,IAAI,EAAG,EAAG,IAG9C,EAAU,EAAW,KAAK,KAG9B,GAAI,EAAe,KAAa,OAC5B,EAAQ,GAAK,EAAe,QAG5B,KAAM,GAAc,OAAO,KAAK,GAAgB,OAChD,EAAe,GAAW,EAC1B,EAAQ,GAAK,EACb,EAAc,KAAK,IAM3B,KAAM,GAAiB,EAAS,QAChC,EAAe,GAAK,OAAO,KAAK,GAAgB,OAChD,KAAM,GAAe,GAAI,GAAA,aAAa,EAAgB,GACtD,EAAc,QAAQ,CAAC,EAAoB,KACvC,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,OAAS,GAAI,EAAG,EAAI,EAAS,GAAI,IAC7B,EAAa,IAAI,EAAY,IAAI,EAAG,EAAoB,GAAI,EAAG,EAAG,KAM9E,KAAM,GAAc,EAAM,QAC1B,SAAY,GAAS,EAAe,GAC7B,CACH,aAAc,EAAa,OAC3B,cACA,WC9IR,AAuBE,GAAA,IAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GACA,GAAA,GCpCF,AAmBA,GAAA,IAAA,WAKE,WAAY,EAAuB,GAHnC,KAAA,YAAwB,GAItB,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,UAAU,EAAQ,SAAS,EAAQ,oBAInD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;yBAEZ,EAAS;;;MAKlC,MAAA,MClDA,AAmBA,GAAA,IAAA,WAOE,WAAY,EAAuB,GALnC,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAqB,GAE3B,KAAK,cAAc,QAAQ,SAAA,GACzB,EAAS,KAAK,SAAS,EAAQ,SAAS,EAAQ,oBAIlD,GAAM,GAAY,KAAK,cACA,IAAI,SAAA,GACH,MAAO,IAAI,IAEZ,KAAK,OAE5B,KAAK,SAAW;;UAEV,EAAS,KAAK;WAAa;;wBAEb,EAAS;;;MAKjC,MAAA,MCpDA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAqC,EACrC,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAC9B,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,KAAK,YAAc,CAAC,EAAW,GAC/B,GAAM,GAAU,IAAO,MAAS,IAAM,IAChC,EAAe,EACjB,gBACA,+CAEJ,KAAK,SAAW;;;;;kCAKc,EAAU;;;;;8BAKd,EAAU;wBAChB,EAAY;;0BAEV,EAAM;;;;;;;MAShC,MAAA,MC5DA,YAiB+B,EAAc,GAC3C,MAAO,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAG,GAAI,IAAI,gBAG/C,EAAc,GACxC,MAAI,KAAS,EACJ,CAAC,GAEH,GAAe,EAAM,eAGE,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,KAIT,OADI,GAAS,GACJ,EAAI,EAAG,EAAI,EAAM,IACxB,GAAU,EAAK,GACf,AAAI,EAAI,EAAO,GACb,IAAU,KAGd,MAAO,GCxCT,cAgCE,GAAI,GACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EAEJ,MAAI,GAAA,MAAM,UAAU,mBAAqB,EACvC,GAAU,kBACV,EAAY,KACZ,EAAY,MACZ,EAAY,KACZ,EAAY,UACZ,EAAS,cACT,EAAe,wBAMf,EAAmB;;;;;;;;;;;MAcnB,EAAmB,GACnB,EAAc;;;;;;;;;OAWd,GAAU,GACV,EAAY,YACZ,EAAY,UACZ,EAAY,UACZ,EAAY,YACZ,EAAS,eACT,EAAe,GAEf,EAAmB;;;;;;;;MASnB,EAAmB;;;;;;;;;MAUnB,EAAc;;;;;;;;OAWT,CACL,QAAO,EACP,UAAS,EACT,UAAS,EACT,UAAS,EACT,UAAS,EACT,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,iBAAgB,EAChB,YAAW,GClIf,YAyBI,EAAkB,EAAiB,GAAA,AAAA,IAAA,QAAA,GAAA,SACrC,GAAM,GAAU,EAAA,KAAK,eAAe,GACpC,MAAO,GACF,IAAI,SAAC,EAAQ,GACZ,GAAM,GAAQ,OAAO,EAAO,GAAE,MAAM,EAAK,MAAM,EACzC,EAAQ,IAAM,EAAQ,OAAS,EACjC,OAAO,EAAO,EAAI,GAAE,MAAM,EAAK,MAAM,EAAO,GAAE,MAAM,EACpD,YAAY,EAAO,GAAE,MAAM,EAC/B,MAAU,GAAK,KAAK,EAAK,MAE1B,KAAK,gBA+CuB,GACjC,GAAM,GAAU,EAAA,KAAK,eAAe,GAAO,IAAI,SAAA,GAAK,MAAA,GAAE,aAEtD,MAAO;;wBAEe,EAAQ,GAAE,iBAAiB,EAAQ,GAAE;;EAKtD,GAAM,IAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EC5FpC,AAkBO,GAAA,IAAA,EAAA,aAAA,6BAkBH,EAAyB,EAAwB,EACjD,GACF,GAAM,GAA2B,GACjC,EAAW,QAAQ,SAAA,GACjB,GAAM,GAAO,EAAA,KAAK,cAAc,EAAE,UAAU,cAG5C,AAAI,EAAE,UAAU,UACd,EAAe,KACX,iBAAiB,EAAE,KAAO,GAAO,EAAI,IAAI,EAAI,IAAM,IAAE,KAEzD,GAAe,KAAK,qBAAqB,EAAE,KAAI,KAC/C,EAAe,KAAK,qBAAqB,EAAE,KAAI,QAGnD,GAAM,GAAqB,EAAe,KAAK;GAEzC,EACF,EACK,IAAI,SAAA,GAAK,MAAA,IAAwB,EAAG,EAAa,KACjD,KAAK;GACR,EAAc,EAAY,SAC1B,EAAO,KACP,EAA4B,GAA6B,GAC3D,EACA,EACA,EAAe,GAAgB,GAEnC,AAAI,EAAY,SACd,GACI,GAA+B,EAAY,aAAc,GAC7D,EAA+B,GAA8B,IAE7D,GACI,GAAyB,EAAY,aAAc,GACvD,EAA+B,GAA2B,IAG5D,AAAI,GACF,IAAgB,IAGlB,GAAM,GAAS,CACb,EAAc,EAA2B,EACzC,EAAoB,EAAuB,EAAsB,GACjE,KAAK;GACP,MAAO,GAGT,YAA8B,GAC5B,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAiB,OACrB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,OACjB,GACH,MAAO,IAAa,WAEpB,KAAM,IAAI,OACH,EAAM,OAAM,2CAKzB,YAAoC,GAClC,GAAM,GAAQ,EAAO,UAAU,aAC/B,OAAQ,EAAM,YACP,GACH,MAAO,IAAuB,OAC3B,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,OACvB,GACH,MAAO,IAAmB,WAE1B,MAAO,IAAmB,IAIhC,YACI,EAAmB,EACnB,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAI,GAAM,GACV,AAAI,EACF,GAAO,GAA2B,GAElC,GAAO,GAAqB,GAG9B,GAAM,GAAU,EAAO,UAAU,aAC3B,EAAW,EAAa,aAC9B,MAAI,GAAQ,QAAU,EAAS,QAC7B,CAAI,EACF,GAAO,GAA+B,EAAQ,GAE9C,GAAO,GAAyB,EAAQ,IAGrC,EAGT,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAwB,EAAsB,OAClD,GACH,MAAO,IAAwB,EAA8B,OAC1D,GACH,MAAO,IACH,EAAsC,WAE1C,MAAO,IAAwB,EAAU,IAI/C,YACI,EAAoB,GACtB,OAAQ,EAAS,YACV,GACH,MAAO,UACJ,GACH,MAAO,IAAkB,EAAsB,OAC5C,GACH,MAAO,IAAkB,EAA8B,OACpD,GACH,MAAO,IACH,EAAsC,OACvC,GACH,MAAO,IACH,EAA8C,OAC/C,GACH,MAAO,IACH,EAAsD,OACvD,GACH,MAAO,IACH,EACA,WAEJ,KAAM,IAAI,OACH,EAAS,OAAM,4CAI5B,YAAsC,GACpC,MAAO;;eAEM,EAAK,UAAS;;IAK7B,YAAoC,GAClC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAuC,GACrC,MAAO;;QAED,EAAK,OAAM;;IAKnB,YAAyB,GACvB,GAAM,GAAmB,EAAK,QAAO;;;;MAIjC,EAAK,UAAS;MACd,EAAK,aAAY;;;;;;;;;;;;;;;;;;;;;;;MAuBjB,EAAK,iBAAgB;MACrB,EAAK,iBAAgB;MACrB,EAAK,YAAW;;;;;;;;;;;;;;;;;;;;;;;;;MAyBhB,GAAiB;MACjB,GAAiB;MACjB,GAAiB;IAGrB,MAAO,GAGT,GAAM,IAAoB;;;;;;;;;;;;EAcpB,GAAoB;;;;;;;;EAUpB,GAAoB;;;;;;;;;EAWpB,GAAuB;;;;;;;;;;;EAa7B,cACE,MAAO;;;;IAOT,YACI,EAAiB,GACnB,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,MAAI,GAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAKjD,EAAe,KAAO,EACjB;;sCAE2B,EAAe,GAAE;;MAK9C;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;IAKlD,YACI,EAAiB,GACnB,MAAI,GAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIvC,EAAS,KAAO,EACX;;kCAEuB,EAAS,GAAE;;MAIpC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;4BACnC,EAAS,GAAE;;IAKvC,YACI,EAAiC,GACnC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAqB,KAAK,KAAK,EAAM,GAAK,GAC1C,EAAgB,EAAqB,KAAK,KAAK,EAAM,GAAK,GAEhE,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;wBAE1B,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAiC,GACnC,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EAAiB,GAWnB,OAVM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEnD,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EACF,EAAqB,KAAK,KAAK,EAAM,EAAM,OAAS,GAAK,GACzD,EAAiB,EACjB,EAAU,GACV,EAAS,UAEJ,EAAI,EAAG,EAAI,EAAM,OAAS,EAAG,IACpC,GAAkB,EAAM,EAAM,OAAS,EAAI,GAC3C,EAAU;aACD,EAAC,cAAc,EAAc;kBACxB,EAAC,MAAM,EAAc;MAC/B,EACJ,EAAS,IAAI,EAAC,KAAO,EAGvB,MAAO;UACC,EAAM,OAAM;;oCAEc,EAAe,GAAE,KAAK,EAAe,GAAE;iCAC1C,EAAe,GAAE;;QAE1C,EAAO;;wBAES,EAAa;qBAChB,EAAa;;6BAEL,EAAkB;4BACnB,EAAkB;;mBAE3B,EAAM,OAAM,IAAI,EAAM;;IAKzC,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,MAAO,GAE3B,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;QACpC,EAAsB;;;IAM9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,MAAO,GAEjC,MAAO;;kDAEyC,EAAS,GAAE;+BAC9B,EAAS,GAAE;;iCAET,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EACA,GACF,GAAM,GAAyB,GAC3B,CAAC,IAAK,IAAK,IAAK,KAAM,KAAM,MAAO,GAEvC,MAAO;;;eAGM,EAAS,GAAE,KAAK,EAAS,GAAE;iCACT,EAAS,GAAE;;QAEpC,EAAsB;;;;;IAQ9B,YACI,EAAyB,GAC3B,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACzD,GAAI,EAAA,KAAK,YAAY,EAAO,GAC1B,MAAO;;8CAEmC,EAAe,GAAE,KACvD,EAAe,GAAE;;MAMvB,GAAM,GAAqB,KAAK,KAAK,EAAM,GAAK,GAWhD,MAAO;;;oCAG2B,EAAe,GAAE,KAAK,EAAe,GAAE;;iCAE1C,EAAe,GAAE;6BACrB,EAAkB;4BACnB,EAAkB;;;;IAO9C,YACI,EAAyB,GAC3B,MAAI,GAAA,KAAK,YAAY,EAAO,GACnB;;0CAE+B,EAAS,GAAE,KAAK,EAAS,GAAE;;MAI/D,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKxC,EAAM,KAAO,EACR;;;sCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;mCAC9B,EAAS,GAAE;;;MAKrC;;;oCAG2B,EAAS,GAAE,KAAK,EAAS,GAAE;iCAC9B,EAAS,GAAE;wBACpB,EAAM,GAAE;4BACJ,EAAM,GAAE;;;IAMpC,YAAkC,GAChC,MAAO,SAAS,EAGlB,YAAgC,GAC9B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAO,KACb,MAAO;WACE,EAAQ;eACJ,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAA0B,GACxB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACzE,GAAI,EAAU,UAAU,UACtB,MAAO,SAAS,EAAQ,cAAc,EAAO,KAEzC,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAChB,GAAI,IAAY,GAAK,IAAY,EAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAK9B,GAAA,GAAA,EAAA,UAAA,SAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GACR,EAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,KAAK,EAAM;6BAC1B,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAe,GAAE,KAAK,EAAe,GAAE;eAClC,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEzE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;UACZ,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAQ,EAAS,GACjB,EAAQ,EAAS,GAEvB,GAAI,IAAU,GAAK,IAAU,EAC3B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAI,KAAU,EACL;cACG,EAAQ;6CACuB,EAAM,cAAc,EAAK;+BACvC,EAAO;;MAIhC,IAAU,EACL;cACG,EAAQ;wCACkB,EAAM,cAAc,EAAK;+BAClC,EAAO;;MAI7B;YACG,EAAQ;6BACS,EAAK,KAAK,EAAK,aAAa,EAAM;6BAClC,EAAO;;IAKpC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAE/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAO,KACb,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,GAC9C,MAAO;aACE,EAAQ;qDACgC,EAAO,OAAO,EAAO;;iBAEzD,EAAK,UAAS,IAAI,EAAO;;MAKxC,GAAM,GACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAe,KAAK,KAAK,EAAM,GAAK,GAE1C,MAAO;WACE,EAAQ;iCACc,EAAY,KAAK,EAAe,GAAE,KAC7D,EAAe,GAAE;eACR,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAErC,GAAI,GAAY,MAAQ,EAAA,KAAK,YAAY,EAAO,IAC9C,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACzB,MAAO;YACC,EAAQ;mDAC+B,EAAO,OAAO,EAAO;6BAC3C,EAAO;;IAK5B,GAAA,GAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,OACvB,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;qDAC+B,EAAM,GAAE;UACnD,GAAkB,GAAU;;MAKpC,GAAM,GAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAS,GAAyB,GACxC,MAAI,KAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;4CACtB,EAAO;6BACtB,EAAO;;IAI9B,IAAY,EAEP;YACC,EAAQ;yCACqB,EAAM,WAAW,EAAM,GAAE;uCAC3B,EAAO;6BACjB,EAAO;;IAK3B;UACC,EAAQ;;wBAEM,EAAM,GAAE,YAAY,EAAM;2BACvB,EAAO,KAAK,EAAO;2BACnB,EAAO;;EAKlC,YAA4B,GAC1B,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IAEzD,GAAI,EAAM,KAAO,GACf,GAAM,GAAgB,EAAM,MAAM,GAC5B,EAAW,CAAC,EAAG,GACf,EAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,IAAK,MAAO,OAC5B,MAAO;UACD,GAA2B,GAAa;eACnC,EAAQ;mBACJ,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAM,GAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpC,EAAgB,EAAe,KAAK,KAAK,EAAM,GAAK,GACpD,EAAO,KAEb,MAAO;WACE,EAAQ;;UAET,EAAO,KAAK,EAAO,KAAK,EAAa,KAAK,EAAY;eACjD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAAK,EAAM,GAC3B,EAAU,EAAM,GAEhB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACX,EAAgB,EACtB,GAAI,EAAc,OAAS,EAAM,QAC/B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,SAC9B,MAAO;UACD,GAAqB,GAAa;gBAC5B,EAAQ;mBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;QAKhE,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO;UAC1C,GAAkB,GAAU;;MAKpC,GAAM,GAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACnB,EAAa,EAAU,UAAU,WACvC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;gBACK,EAAQ;;oDAE4B,EAAO;;4BAE/B,EAAO,OAAO,EAAO;iCAChB,EAAO;;QAKtC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;YACC,EAAQ;8CAC0B,EAAM,GAAE;;qDAED,EAAO,OAAO,EAAO;6BAC7C,EAAO;;IAKlC,GAAM,GAAS,GAAyB,GACxC,MAAO;cACK,EAAQ;;4BAEM,EAAO,YAAY,EAAO,cAAc,EAAM;+BAC3C,EAAO,KAAK,EAAO;+BACnB,EAAO;;IAKtC,YAA4B,GAe1B,OAdM,GAAQ,EAAU,UAAU,aAC5B,EAAO,EAAM,OACb,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAW,EAAU,UAAU,SAC/B,EACF,CAAC,KAAK,KAAK,EAAS,GAAK,GAAI,KAAK,KAAK,EAAS,GAAK,IACnD,EAAU,EAAe,GACzB,EAAU,EAAe,GAEzB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC7C,EAAgB,EAAe,KAAK,KAAK,EAAM,EAAO,GAAK,GAC3D,EAAS,0BACT,EAAQ,OAAO,EAAa,kBAAkB,EAAY,eACrD,EAAI,EAAG,EAAI,EAAO,EAAG,IAC5B,EAAS,QAAQ,EAAC,KAAO,EACzB,GAAiB,EAAM,EAAO,EAAI,GAClC,EAAQ,IAAI,EAAC,MAAM,EAAa,MAAQ,EAE1C,GAAM,GAAO,KACb,MAAO;WACE,EAAQ,IAAI,EAAM;oBACT,EAAK;2BACE,EAAO;kCACA,EAAO;qDACY,EAAO,KAAK,EAAO;eACzD,EAAK,UAAS,IAAI,EAAO;;IAKxC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,UACvC,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;iCAEW,EAAO,KAAK,EAAO,KAAK,EAAO;UACtD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;uBAIC,EAAO,KAAK,EAAO;;0BAEhB,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;gCAEU,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGvC,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO;oBAChC,EAAO;6BACE,EAAO,KAAK,EAAO,aAAa,EAAM;6BACtC,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACnE,EAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAErB,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,UACjD,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;iBACL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;iBAGL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;UAElD,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GAEzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;gCAGU,EAAO,KAAK,EAAO,KAAK,EAAO;;0BAErC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAKpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE,KAAK,EAAM,GAAE;;;yBAGxB,EAAO,OAAO,EAAO;+BACf,EAAO;;MAKpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;0BAEM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAM;6BACpB,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAAsB,GACpB,GAAM,GAAQ,EAAU,UAAU,aAC5B,EAAU,EAAU,KACpB,EAAW,MAAQ,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GAEnE,EAAA,EAAA,KAAA,aAAA,GAAC,EAAA,EAAA,SAAU,EAAA,EAAA,SACjB,GAAI,EAAS,OAAS,EAAM,QAC1B,GAAM,GAAe,GAAiB,EAAW,GAC3C,EAAS,CAAC,MAAO,MAAO,QAAS,SAAU,SAAU,UAC3D,MAAO;QACH,GAAqB,GAAa;cAC5B,EAAQ;;iBAEL,EAAQ,IAAI,GAAkB,EAAQ,GAAS;;MAK9D,GAAM,GAAU,EAAM,GAChB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EACrB,EAAU,EAAM,GAAK,EAE3B,GAAI,EAAU,UAAU,UAEtB,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;mBAGzC,EAAO;UAChB,GAAkB,GAAU;;MAKpC,GAAM,GAAa,EAAU,UAAU,WACjC,EAAW,EAAU,UAAU,SAC/B,EAAU,EAAS,GACnB,EAAU,EAAS,GACzB,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;;iBAIL,EAAO,KAAK,EAAO,KAAK,EAAO,KAAK,EAAO;;;0BAGlC,EAAO,OAAO,EAAO;+BAChB,EAAO;;MAIpC,GAAI,IAAY,GAAW,GAAc,KAEvC,MAAO;cACG,EAAQ;;;iBAGL,EAAM,GAAK,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBACzC,EAAM,GAAK,EAAM,GAAK,EAAM,GAAE;iBAC9B,EAAM,GAAK,EAAM,GAAE;iBACnB,EAAM,GAAE;;;yBAGA,EAAO,OAAO,EAAO;+BACf,EAAO;;MAIpC,GAAM,GAAS,GAAyB,GACxC,MAAO;YACG,EAAQ;;;0BAGM,EAAO,YAAY,EAAO,cAAc,EAAO;qBACpD,EAAO,eAAe,EAAO,eAAe,EAAM;6BAC1C,EAAO,KAAK,EAAO;6BACnB,EAAO;;IAKpC,YAA2B,GACzB,GAAM,GAAU,EAAU,KACpB,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAEtD,MAAI,GAAS,EACJ,UAAU,EAAO,IAEnB;0BACiB,EAAM;;iBAEf,EAAO;;;IAMxB,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAEpC,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAE7C,EAAO,GAAkB,GACzB,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,AAAI,EAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAGpC,GAAI,GAAS,sBACP,EAAS,EAAA,KAAK,cAAc,EAAU,UAAU,cAChD,EAAgB,IAAW,EAC3B,EAAU,EAAA,KAAK,cAAc,EAAa,cAC1C,EAAiB,IAAY,EAEnC,GAAI,IAAW,GAAK,CAAC,GAAiB,CAAC,EACrC,EAAS;;cAGA,GAAiB,CAAC,EAC3B,AAAI,IAAY,EACd,EAAS;;QAIT,EAAS;;gBAIF,EAAc,QACvB,GAAM,GAAO,EAAS,EAChB,EAAO,EAAS,EAEtB,AAAI,EAAc,QAAQ,GAAQ,IAAM,EAAc,QAAQ,GAAQ,GACpE,EAAS,8BACJ,AAAI,EAAc,QAAQ,GAAQ,GACvC,EAAS,2EAEJ,AAAI,EAAc,QAAQ,GAAQ,IACvC,GAAS,gDAIb,MAAO;WACE,EAAQ;QACX,EAAI;QACJ,EAAa;8BACS,EAAc,IAAI,EAAqB;QAC7D,EAAM;;IAKd,YACI,EAAsB,GACxB,GAAM,GAAU,EAAU,KACpB,EAAiB,EAAQ,OAAO,GAAG,cAAgB,EAAQ,MAAM,GACjE,EAAW,MAAQ,EAAiB,cACpC,EAAc,EAAa,SAC3B,EAAa,EAAU,UAAU,SACjC,EAAS,EAAU,UAAU,aAAa,OAC1C,EAAU,EAAa,aAAa,OAE1C,GAAI,CAAC,EAAU,UAAU,WAAa,IAAW,GAC7C,EAAU,UAAU,YAAc,MAClC,EAAA,KAAK,YAAY,EAAY,GAC/B,MAAO;cACG,EAAQ;+BACS,EAAO;;MAKpC,GAAM,GAAO,GAAkB,GACzB,EAAgB,GAClB,EAAU,UAAU,aAAc,EAAa,cAC7C,EAAW,EAAU,EACvB,EACE,EAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,AAAI,IAAW,EACb,EAAgB,GACX,AAAI,EAAU,GAAK,EAAc,QAAU,EAChD,EAAgB,cAEhB,EACI,EAAc,IAAI,SAAA,GAAK,MAAA,UAAU,EAAO,EAAI,GAAS,UAChD,KAAK;GAEhB,GAAI,GAAwB,GAC5B,MAAI,GAAU,GAAK,EAAS,EAC1B,EAAwB,SAExB,EAAwB,EAAU,UAAU,aACf,IAAI,SAAC,EAAG,GAAM,MAAA,UAAU,EAAO,EAAI,KACnC,KAAK,MAG7B;YACG,EAAQ;QACZ,EAAI;QACJ,EAAa;kBACH,EAAc,IAAI,EAAqB;;gBAKvB,GAChC,GAAI,GAAQ,EACV,MAAO,MACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QACF,GAAI,IAAS,EAClB,MAAO,QAEP,KAAM,OAAM,gBAAgB,EAAI,yBAKpC,YACI,EAAmB,GAErB,GAAM,GAA0B,KAAK,MAAM,KAAK,UAAU,IAC1D,SAAa,UAAU,aAAe,EAC/B,EAGT,YAA2B,EAAkB,GAC3C,MAAO,GAAS,IAAI,SAAA,GAAK,MAAA,GAAO,KAAI,KAAK,MCt2C3C,AAuBA,GAAA,IAAA,WAOE,WACI,EAAiB,EAAoB,EACrC,GARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,EAAA,KAAK,OACD,EAAM,OAAS,EACf,WAAM,MAAA,aACF,GAAG,OAAO,GAAG,cACb,EAAG,MAAM,IAAE,6CACnB,GAAM,GAAS,EAAM,EAAM,OAAS,GAC9B,EAAU,KAAK,KAAK,EAAS,GACnC,KAAK,YAAc,EAAM,MAAM,EAAG,IAClC,AAAI,EAAU,GACZ,KAAK,YAAY,KAAK,GAExB,AAAK,GACH,KAAK,cAAc,KAAK,gBAE1B,GAAM,GAAW,KAAK,YAChB,EAAO,EAAS,OAChB,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAEjC,EACA,EACJ,GAAI,IAAY,GACd,EAAa,EAAO,EACpB,GAAM,GAAiB,GAAkB,GACzC,EAAiB;UACb,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE;UAClB,EAAc,iBAAiB,EAAc,IAAI,EAAO,OAAM;YAC5D,EAAO,EAAO,GAAE,QAEtB,GAAa,EACb,EAAiB;UACb,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE;UAClB,EAAK;YACH,EAAO,EAAO,GAAE,IAExB,GAAM,GAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACnD,EAAY,IAAM,EAAS,EAAa,GACxC,EAAc,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IACzC,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAC/C,EACF,GAAY,aAAc,EAAa,GAAG,OAAO,WAE/C,EAAU,IAAO,MAAS,cAAgB,WAC1C,EAAoB,EAAY,GAAK;sDACO,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM;sDACjB,EAAW,OAAM,OAE7D,EAAa;0BACG,EAAW,OAAM;uCACJ,EAAW,OAAM;uCACjB,EAAW,OAAM;qDACH,EAAW,OAAM,UAE5D,EAAgC,EAAY,GAAK;qCACtB,EAAY,OAAM;4CACX,EAAS,OAAM;iDACV,EAAS,MAAM,IAAI,OAAM;SAGtE,KAAK,SAAW;0BACM,EAAY,OAAM;iCACX,EAAS,OAAM;sCACV,EAAS,MAAM,IAAI,OAAM;;QAEvD,EAA6B;;UAE3B,EAAK;4BACa,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;4BAC5C,EAAO,EAAO,GAAE,MAAM,GAAS,EAAO,GAAK,GAAC;UAC9D,EAAc;yCACiB,EAAS,eAAe,EAAS;sBACpD,EAAS,eAAe,EAAS,OAAO,EAAU;;;2BAG7C,EAAU;;8BAEP,EAAU;;YAE5B,EAAiB;6BACA,EAAU;;;mBAGpB,EAAM;;;;;;;;;;;MAazB,MAAA,MCjJA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAe,GAE1C,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;0CACT,EAAa;;;;;;;;;;;;;;gCAcvB,EAAqB;oBACjC,EAAc;gDACc,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;mBACnC,EAAa;kDACkB,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;;;;;;;MAevD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,MAKf,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAgB,EAAK,GAAc,EAAe,GAExD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAa;;;;;;;;;;;;;;;;;gCAiBvB,EAAoB;oBAChC,EAAa;gDACe,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;;;;;;;;MAgBzD,MAAA,MC3KA,AAwBO,GAAM,IAAmB,CAC9B,KAAM,wCACN,KAAM,yCAGR,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,QAAS,QAAS,QAAS,SAK1C,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GAEnE,KAAK,SAAW;;;UAGV,EAAE;;;;;;;;;;MAYZ,MAAA,MCpDA,AAqBA,GAAM,IAAoB;;;EAKb,GAAM,gBACN,GAAM,gBACN,GAAM,gBAMN,GAAU;;;;;;;;;;EAYV,GAAM;;;;;;;;;EAYN,GAAQ,wBAER,GAAY,wBAEZ,GAAO,uBAEP,GAAa,wBAEb,GAAU,uBAEV,GAAgB,wBAEhB,GAAc,sCAEd,GAAa,sCAEb,GAAM,GAAoB;;EAG1B,GAAM,GAAoB;;EAG1B,GAAM;qBAGN,GAAU,yCAEV,GAAQ,+BAErB,GAAA,WAKE,WAAY,EAAY,EAAkB,GAJ1C,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;MAUZ,MAAA,MC1GA,AAuBA,GAAM,IAAoB;;;;;EAOb,GAAU;;;;;;;;;;;;;;;;;;;;;EAuBV,GAAM;;;;;;;;;;;;;;IAef,GAAoB;;EAIX,GAAQ;;;EAKR,GAAU;;;EAKV,GAAQ;;EAIR,GAAY;;EAIZ,GAAO;;EAIP,GAAa;;EAIb,GAAU;;EAIV,GAAgB;;EAIhB,GAAc;;;;EAMd,GAAa;;;;;EAOb,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIX,GAAM;;;IAIf,GAAoB;;EAIxB,GAAA,WAQE,WACI,EAAY,EAAkB,EAC9B,GAAA,AAAA,IAAA,QAAA,GAAA,IATJ,KAAA,cAAgB,CAAC,IAAK,KAGtB,KAAA,qBAAuB,GACvB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EAAA,aAAa,2BAA2B,EAAQ,GACnE,GAAM,GAAO,KAAK,YAAY,OAC1B,EAAyB,GAC7B,GAAI,EACF,GAAI,IAAS,GAAK,EAAA,KAAK,cAAc,KAAK,eAAiB,EACzD,EAAyB;;;;eAMzB,GAAM,GAAQ,GAAkB,GAIhC,GAHA,EAAyB;YACrB,EAAK;UAEL,IAAS,EACX,GAA0B;yCACK,KAAK,YAAY,GAAE;;;iBAKlD,GAAM,GAAW,GAAY,SAAU,GACvC,GAA0B;;iBAEnB,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;iBAExD,EAAS,EAAO,GAAE,YAAY,KAAK,YAAY,EAAO,GAAE;;;;aASrE,KAAK,SAAW;;UAEV,EAAE;;;;;;;;UAQF,EAAsB;;;;MAMhC,MAAA,MC9MA,AAoBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,KASf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAgBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KCzDA,AAoBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;MAiBlB,SAAA,UAAA,mBAAA,SAAmB,EAAa,GAAhC,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,QAAU,MACjB,GAAK,OAAS,EAAM,0BAA0B,EAAc,UAC5D,EAAK,OAAS,EAAM,0BAA0B,EAAc,WAE9D,EAAM,GAAG,UAAU,EAAK,OAAQ,GAChC,EAAM,GAAG,UAAU,EAAK,OAAQ,KAGtC,KC5DA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,OAAQ,QAKvB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;MAepB,MAAA,MCzCA,AAoBA,GAAA,IAAA,WAME,WAAY,GAJZ,KAAA,YAAwB,GAKtB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAI1C,OADM,GAAW,CAAC,YAAY,EAAQ,GAAE,+BAC/B,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAC1B,EAAS,KACL,iBAAiB,EAAQ,GAAE,KAC3B,kBAAiB,EAAC,WAAW,EAAK,QAExC,GAAM,GAAY,EAAQ,OACpB,EAAY,EAAQ,EAAQ,OAAS,GAC3C,EAAS,KAAK,sBAAsB,EAAS,WAAW,EAAS,OAEjE,KAAK,SAAW;;;;;;UAMV,EAAS,KAAK;WAAa;;MAIrC,MAAA,MCzDA,AAuBA,GAAA,IAAA,WAOE,WAAY,EAAoB,GALhC,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAA,aAAa,gBAAgB,EAAQ,GACxD,GAAM,GAAQ,KAAK,YACb,EAAO,EAAM,OACb,EAAQ,GAAkB,GAC1B,EAAS,GAAY,SAAU,GAC/B,EAAW,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAAK,MAAM,EAAG,GACzD,KAAK,cAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,IAAI,IAE9C,GAAM,GAAoB,GAAI,OAAM,EAAO,OAAS,GACpD,EAAQ,GAAK,EAAO,GAAG,GACvB,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,IAClC,EAAQ,GAAK,EAAQ,EAAI,GAAK,EAAO,GAAG,GAW1C,OARM,GAAU,EAAS,GACnB,EAAe,EAAS,MAAM,IAC9B,EAAc,EAAS,OAEzB,EAAkB,OAAO,EAAO,MAAM,EAAQ,GAAE;;oBAEpC,EAAW,WAAW,EAAa,OAAM;WAEhD,EAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,GAAM,GAAQ,EAAQ,EAAI,GAK1B,GAAmB;cACX,EAAO,MAAM,EAAQ,GAAE,QAAQ,EAAO,OAAO,EAAQ,EAAI,GAAE;;kBAEvD,EAAC,IAAI,GAAgB,EAAU,EAAS,GAAM;mBAC7C,GAAgB,EAAc,EAAS,GAAM;WAG5D,GAAM,GAAY,EAAQ,OACpB,EAAQ,EAAQ,EAAQ,OAAS,GACvC,GAAmB;;gBAEP,EAAS,IAAI,GAAgB,EAAU,EAAS,GAAM;iBACrD,GAAgB,EAAc,EAAS,GAAM,MAE1D,KAAK,SAAW;uBACG,EAAS,IAAI,SAAA,GAAK,MAAA,OAAS,IAAE;UAC1C,EAAe;;;;UAIf,EAAK;sCACuB,EAAM;;UAElC,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;UAG5B,EAAO,EAAO,GAAE,MAAM,EAAO,EAAO,GAAE;cAClC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;cACrC,EAAO,EAAO,GAAE,MAAM,EAAM,EAAO,GAAE;gCACnB,EAAM;;;;MAMtC,MAAA,MAaA,YAAyB,EAAoB,EAAiB,GAC5D,GAAM,GAAa,EAAS,QAAQ,GAC9B,EAAM,EAAS,IAAI,SAAC,EAAG,GAC3B,MAAI,KAAQ,EACA,EAAC,MAAM,EAEV,IAGX,MAAO,GAAI,OC3Hb,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAiB,EAAS,aAAe,eAE/C,KAAK,SAAW;;;;;;;;;;;;8BAYU,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;oBAIhC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAE7C,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAExC,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;wCAEI,EAAM,aAAa,EAAM;;;;;;;gCAOjC,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;oCAEF,EAAS,YAAW;;oBAEpC,EAAc;;;;;;;;;;;;;;;MAiBlC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAEjC,KAAK,SAAW;;;;;;;;;;;8BAWU,EAAS,UAAS;kCACd,EAAS,SAAQ;iCAClB,EAAW,MAAM,EAAQ;;kCAExB,EAAS,QAAO;;;;oCAId,EAAS,UAAS;mCACnB,EAAY,MAAM,EAAM;;oCAEvB,EAAS,SAAQ;;;;sCAIf,EAAS,SAAQ;qCAClB,EAAW,MAAM,EAAO;;sCAEvB,EAAS,QAAO;;;;;;;;;;;;;MAetD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAW,EAAc,EAAI,EAAS,QAAQ,MAC9C,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAEnD,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAchC,EAAW;gDACK,EAAW;;oCAEvB,EAAS,SAAQ;;;;;yBAK5B,EAAW;;kCAEF,EAAY;kDACI,EAAY;;sCAExB,EAAS,UAAS;;;;;;2BAM7B,EAAY;;oCAEH,EAAW;oDACK,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;6BAM5B,EAAW;;sCAEF,EAAS,YAAW;;;;;;;;;;MAY1D,MAAA,MCzSA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,MAKpB,KAAK,YAAc,EAAS,YAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;;;;;;;wBAOI,EAAU;;;;;8BAKJ,EAAS,UAAS;kCACd,EAAS,UAAS;iCACnB,EAAY,MAAM,EAAM;;kCAEvB,EAAS,SAAQ;;;;oCAIf,EAAS,SAAQ;mCAClB,EAAW,MAAM,EAAO;;oCAEvB,EAAS,QAAO;;;;;;;;;;;;MAcpD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,KAKrB,KAAK,YAAc,EAAS,QAE5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAS,EAAe,EAAI,EAAS,QAAQ,IAC7C,EAAU,EAAc,EAAI,EAAS,QAAQ,KAC7C,EAAa,EAAS,YAAc,EAAS,WAEnD,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;gCAYnB,EAAY;gDACI,EAAY;;oCAExB,EAAS,UAAS;;;;;yBAK7B,EAAY;;kCAEH,EAAW;kDACK,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;2BAM5B,EAAW;;;oCAGF,EAAU;8BAChB,EAAU;;;;;;;;;MAWxC,MAAA,MCzIA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAChD,EAAiB,EAAS,aAAe,eAEzC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAC9B,EAAa,EAAiB,EAAI,EAEpC,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;0BAKzB,EAAU;;;2BAGT,EAAM,aAAa,EAAM;;;;;;;gCAOpB,EAAY;qCACP,EAAc;;gCAEnB,EAAS,SAAQ;;;;kCAIf,EAAW;uCACN,EAAa;;kCAElB,EAAS,QAAO;;;;oCAId,EAAqB;;;;;;;;oBAQrC,EAAc;;;;;;;;;;;;;;;;;;;kBAmBhB,KAA4B,GAAC;;oBAE3B,EAAc;;0CAEQ,EAAqB;mCAC5B,EAAqB;;;kCAGtB,EAAqB;mCACpB,EAAqB;;;yBAG/B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;;;;;yBAK5B,KAA4B,GAAC;;+BAEvB,EAAqB;+BACrB,EAAqB;+BACrB,EAAqB;;;oBAGhC,EAAc;;wCAEM,EAAqB;wCACrB,EAAqB;wCACrB,EAAqB;;;;;gCAK7B,EAAqB;gCACrB,EAAqB;gCACrB,EAAqB;;;;;;;;;;UAU3C,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,IAAK,KAKpB,KAAK,YAAc,EAAS,SAC5B,GAAM,GAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YAEvB,EAAwB,KAAK,MAAM,EAAS,WAAa,GAAK,EAC9D,EAA0B,EAAS,WAAa,EAEtD,KAAK,SAAW;oCACgB,EAAW,KAAK,EAAY,KACxD,EAAW;iCACc,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;gCAgBhC,EAAW;qCACN,EAAa;;gCAElB,EAAS,QAAO;;;;kCAId,EAAY;uCACP,EAAc;;kCAEnB,EAAS,SAAQ;;;;oCAIf,EAAW;yCACN,EAAa;;oCAElB,EAAS,QAAO;;;;sCAId,EAAqB;;;;;;;;;;;;;;;;;oBAiBvC,KAA4B,GAAC;;4CAEL,EAAqB;qCAC5B,EAAqB;2BAC/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;;;2BAG/B,KAA4B,GAAC;;4CAEZ,EAAqB;4CACrB,EAAqB;4CACrB,EAAqB;;;qCAG5B,EAAqB;qCACrB,EAAqB;qCACrB,EAAqB;;;;;;;;;MAW1D,MAAA,MC3TA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAN/B,KAAA,cAAgB,CAAC,IAAK,KAOpB,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAa,EAAS,YAAc,EAAS,WAE/C,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;;cAEd,EAAU;;UAKlB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;wBAO3B,EAAU;4BACN,EAAU;;;;;;;;;gCASN,EAAY;qCACP,EAAc;;gCAEnB,EAAQ;;;;kCAIN,EAAW;uCACN,EAAa;;kCAElB,EAAQ;;;;;;;;;;;UAWhC,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MCrHA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAmC,EACnC,EAA2B,GADQ,AAAA,IAAA,QAAA,GAAA,IACnC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAR/B,KAAA,cAAgB,CAAC,IAAK,KACtB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,SAgB5B,OAdM,GAAW,EAAS,SACpB,EAAW,EAAS,QACpB,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KAC3B,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAe,EAEjB,EAAW,gCAEN,EAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY;wBACI,EAAC,IAAI,EAAI,EAAC;mBACf,EAAC,IAAI,EAAC;mBACN,EAAC,IAAI,EAAC,eAYrB,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAS,EAAG,EAAS,EAAc,KAC1C,GAAM,GAAI,EAAS,EAOnB,GALA,GAAY;4BACQ,EAAI,EAAc;4BAClB,EAAI,EAAa;UAGjC,IAAgB,GAClB,GAAI,EAAI,GAEN,CAAI,EAAU,IAAM,EAUlB,GAAY;;qCAEW,EAAQ,mCAC3B,EAAQ;2BACC,EAAC,IAAI,EAAC;;;;uCAIM,EAAQ;6BAClB,EAAC,IAAI,EAAC;;;2BAGR,EAAC,IAAI,EAAC;;;;qCAII,EAAQ,mCAC3B,EAAQ;;;;;uCAKa,EAAQ;;;;sBAIzB,EAAC,IAAI,EAAC,+BAA+B,EAAC,IAAI,EAAC;;sBAE3C,EAAC,IAAI,EAAC,wBAAwB,EAAC,IAAI,EAAC;;gBAK5C,GAAY;qCACW,EAAQ,uBAAuB,EAAQ;2BACjD,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;oBAGb,EAAC,IAAI,EAAC,aAAa,EAAC,IAAI,EAAC;gBAI7B,EAAI,EAAI,IAOV,GAAM,GAAkB,EAAU,IAAM,EACpC,EAAA,KAAK,kBAAkB,GACvB,EAEJ,AAAK,EAAgB,IAAM,GAAK,EAAU,IAAM,GAC3C,EAAgB,IAAM,GAAK,EAAU,IAAM,EAC9C,IAAY;oCACQ,EAAU,EAAC,MAAM,EAAe;;uCAE7B,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;kBAMvB,AAAI,EAAgB,GAClB,IAAY;;yCAEW,EAAQ;oDACG,EAAQ;+BAC7B,EAAC,IAAI,EAAC;;+BAEN,EAAC,IAAI,EAAC;;qBAKrB,GAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;mBAG5C,GAAY;oCACQ,EAAe;;uCAEZ,EAAQ;kDACG,EAAQ;6BAC7B,EAAC,IAAI,GAAI,GAAC;;;sBAGjB,EAAC,IAAI,GAAI,GAAC,aAAa,EAAC,IAAI,GAAI,GAAC;uBAM7C,AAAI,GAAI,GACN,IAAY;mCACW,EAAQ;cAS/B,AAAI,EAAU,IAAM,EAClB,IAAY;sCACY,EAAW;iDACA,EAAQ;2BAC9B,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;6CAGY,EAAQ;2BAC1B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;;wCAEY,EAAW;mDACA,EAAQ;;;sBAGrC,EAAC,IAAI,GAAI,GAAC,kBAAkB,EAAC,IAAI,GAAI,GAAC;oBAI9C,IAAY;qCACW,EAAQ;2BAClB,EAAC,IAAI,EAAC;;2BAEN,EAAC,IAAI,EAAC;;;kCAGC,EAAW;iDACI,EAAQ;2BAC9B,EAAC,IAAI,GAAI,GAAC;;2BAEV,EAAC,IAAI,GAAI,GAAC;;;oBAGjB,EAAC,IAAI,EAAC;2BACC,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;gBAG5C,AAAI,EAAI,EAAI,GACV,IAAY;sBACN,EAAC,IAAI,GAAI,GAAC;6BACH,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,GAAI,GAAC;oBAKhD,GAAY,KAIhB,AAAI,EAAI,GACN,IAAY;0BACI,EAAC,IAAI,EAAC,WAAW,EAAC,KAAK,EAAC;gBAClC,EAAC,IAAI,EAAC,kBAAkB,EAAC,IAAI,EAAC,eAAe,EAAC,IAAI,EAAC;YAGzD,AAAI,EAAI,EAAI,GACV,IAAY;4BACI,EAAC,IAAI,GAAI,GAAC,WAAW,EAAC,KAAK,GAAI,GAAC;kBAC1C,EAAC,IAAI,GAAI,GAAC;8BACE,EAAC,IAAI,GAAI,GAAC,eAAe,EAAC,IAAI,GAAI,GAAC,UAM7D,OAAS,GAAI,EAAG,EAAI,EAAc,IAChC,OAAS,GAAI,EAAG,EAAI,EAAa,IAC/B,GAAY,gBAAgB,EAAC,IAAI,EAAC,QAAQ,EAAC,IAAI,EAAC,IAIpD,GAAI,GAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;oCAEW,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;UAezC,EAAQ;;;UAGR,EAAc;UACd,EAAsB;;;MAKhC,MAAA,MC3UA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAA4B,EAC5B,GAPJ,KAAA,cAAgB,CAAC,QAAS,QAAS,UACnC,KAAA,YAAwB,GAOf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAa,EAAA,EAAA,GAAY,EAAA,EAAA,GAChC,EAAA,EAAA,GACA,EAAA,EAAA,GAAY,EAAA,EAAA,GACnB,KAAK,YAAc,CAAC,EAAU,EAAY,EAAW,GACrD,GAAM,GAAW,IAAW,WAAa,EAAI,EAEvC,EAAA,CAAA,EAAA,EAAA,KAAA,EAAA,EAAA,MAAC,EAAA,EAAA,GAAkB,EAAA,EAAA,GAGnB,EAAA,EAAA,oHAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAa,EAAA,EAAA,GAW3B,EAAA,EAAA,kHAAC,EAAA,EAAA,GAAY,EAAA,EAAA,GAAY,EAAA,EAAA,GAe/B,KAAK,SAAW;yCACqB,EAAW;wCACZ,EAAU;;;;;;;;;;;;;;;;iCAgBjB,EAAK;;;;+BAIP,EAAW;8BACZ,EAAU;;uBAEjB,EAAG;mCACS,EAAgB;4BACvB,EAAkB;;;uBAGvB,EAAG;mCACS,EAAe;4BACtB,EAAkB;;;;;aAKjC,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;MA0BrB,MAAA,MC1GA,GAAA,WAQE,WAAY,EAAiB,EAAoB,GAPjD,KAAA,cAAgB,CAAC,KAQf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAM,OACb,EAAM,EAAY,MAAQ,QAAQ,GAAU,EAAM,UAAS,IAC3D,EAAS,EAAM,EAAM,OAAS,GAChC,EAAY,GACZ,EAAY,GAIhB,AAAI,EACF,GAAY,EAAU,UAAU,GAAS,GAAM,WAC/C,EAAY,EAAU,UAAY,WAElC,GAAY,EAAU,gBAAgB,EAAW,cACjD,EAAa,EAAU,aAAe,cAGxC,KAAK,SAAW;;;UAGV,GAAkB,GAAK;oBACb,GAAc,EAAM,UAAS;sBAC3B,EAAG;;cAEX,EAAS;sBACD,EAAS;YACnB,GAAc,EAAM,UAAS;wBACjB,GAAU,EAAM,UAAS;;;;MAO/C,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,OAAS,MAChB,GAAK,MAAQ,EAAM,mBAAmB,EAAc,UAEtD,EAAM,GAAG,UAAU,EAAK,MAAO,KAGrC,KAEA,YAAmB,EAAc,GAC/B,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,KACpB,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,KAC/B,GAAI,IAAS,EAClB,MAAU,GAAI,OAAO,EAAI,OAAO,EAAI,OAAO,EAAI,KAE/C,KAAM,OAAM,2BAA2B,EAAI,yBAI/C,YAAuB,EAAc,GACnC,GAAI,IAAS,EACX,MAAO,GAAG,EACL,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KACT,GAAI,IAAS,EAClB,MAAU,GAAI,KAEd,KAAM,OAAM,2BAA2B,EAAI,yBCjG/C,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAEjB,KAAA,aAAe,GACf,KAAA,aAAe,GAEf,KAAA,iBAAmB,GAAc,MAG/B,GAAM,GAAW,GAAiB,GAC5B,EAAO,KACb,KAAK,YAAc,EAEnB,KAAK,SAAW;;UAGZ,GACI,CAAC,IAAK,IAAK,KAAM,GAAY;;;;;;iBAMxB,EAAS,GAAE,KAAK,EAAS,GAAE;wCACJ,EAAS,GAAE;;;;;;;;;;UAUzC,EAAK,OAAM;;MAIrB,MAAA,MC5DA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAuB,EAAmB,GAP9C,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAOtB,KAAK,YAAc,EACnB,KAAK,UAAY,EACjB,KAAK,WAAa,EAClB,KAAK,SAAW;;;;gBAIJ,KAAK,uBAAsB;gBAC3B,KAAK,sBAAqB;gBAC1B,KAAK,sBAAqB;;uBAEnB,EAAS;+BACD,EAAS;uBACjB,EAAS;+BACD,EAAS;mCACL,EAAS;UAClC,KAAK,qBAAoB;;;uBAGZ,KAAK,yBAAwB;;;IAM1C,SAAA,UAAA,qBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,oBAAR,WACE,MAAI,MAAK,aAAe,OACf,YAEA,aAIH,EAAA,UAAA,mBAAR,WACE,MAAI,MAAK,aAAe,OACf,KAAK,YAAY,GAEjB,KAAK,YAAY,IAIpB,EAAA,UAAA,uBAAR,WACE,MAAI,MAAK,aAAe,OACf,4BAEA,6BAGb,KC5FA,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,CAAC,EAAM,GAC1B,KAAK,SAAW;;;;;;MAQpB,MAAA,MClCA,AAsBA,GAAA,IAAA,WAME,WAAY,GALZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;UAIlB,EAAK,OAAM;;MAIrB,MAAA,MCxCA,AAsBA,GAAA,IAAA,WAQE,WAAY,GAPZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAc,GAAa,SAGzB,GAAM,GAAO,KACb,KAAK,YAAc,EACnB,KAAK,SAAW;QACZ,GAAoB;;;;;UAKlB,EAAK,OAAM;;MAIrB,MAAA,MC3CA,AAqBA,GAAA,IAAA,WAKE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IANJ,KAAA,cAAgB,CAAC,KAOf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAS,SACb,AAAI,GACF,GAAS,8BAGX,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;8BAUrB,EAAK;kCACD,EAAK;iDACU,EAAK,OAAO,EAAM;wBAC3C,EAAK,UAAS;;;;;;;;;;;;;;UAc5B,EAAK,OAAM,WAAW,EAAM;;MAItC,MAAA,MCtEA,AAqCA,GAAA,IAAA,WAOE,WACI,EAAuC,EACvC,GAAA,AAAA,IAAA,QAAA,GAAA,IARJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EAEnB,GAAI,GAAW,GACX,EAAS,SACb,AAAI,GACF,GAAS,8BAGX,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAC1B,GAAM,GAAU,EAAM,EAAI,EAE1B,GAAY;;gCAEY,EAAG,MAAM,EAAY,GAAE;gCACvB,EAAG;kCACD,EAAG,MAAM,EAAY,GAAE;kCACvB,EAAG;;;;;;;gCAOL,EAAK;oCACD,EAAK;kDACS,EAAK,OAAO,EAAM;yBAC3C,EAAK,UAAS;;;yBAGd,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;yBAEP,EAAO;;;;UAQ5B,KAAK,SAAW;QACZ,GAA+B,GAAY;;;;;;;;;;;UAWzC,EAAQ;;UAER,EAAK,OAAM,MAAM,EAAM;;MAIjC,MAAA,MC/GA,AAmBO,GAAM,IAAc,CACzB,KAAM,oCACN,KAAM,qCAGR,GAAA,WAKE,WAAY,EAAY,EAA8B,GAJtD,KAAA,cAAgB,CAAC,OAAQ,QAKvB,GAAM,GAAW,EAAW,GAC5B,KAAK,YAAc,EAEnB,GAAM,GACF,EAAU,SAAS,KAAK,GAAO,UAAU,KAAK,GAC5C,EAAoB,EAAa,EAAQ,KAAO,MAEtD,KAAK,SAAW;yCACqB,EAAyB;;;UAGxD,EAAE;;;;kDAIsC,EAAQ;;;;;;8BAM5B,EAAQ;;;;;;;;;yDASmB,EAAiB;;;;;;;;;;MAY1E,MAAA,MCxEA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAiB,GAL7B,KAAA,YAAwB,GAMtB,KAAK,cAAgB,CAAC,KACtB,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;;MASlB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,UAEhE,EAAM,GAAG,UAAU,EAAK,SAAU,KAGxC,KChDA,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,EAAuB,GALrD,KAAA,cAAgB,CAAC,IAAK,WAMpB,GAAM,GAAwB,EAAO,QACrC,EAAY,GAAQ,EACpB,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,EAAQ,GAE7C,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,EAAkB,GACzC,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,mBAAmB,EAAI,yBAErC,GAAI,IAAS,EACX,MAAO,yBAMT,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAElD,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,AAAI,IAAM,EACR,EAAa,KAAK,kBAAkB,EAAc,GAAE,MAEpD,EAAa,KAAK,GAAG,EAAc,IAGvC,MAAO,GAAa,OC3CtB,GAAA,IAAA,WAIE,WACY,EAA0B,EAAmB,GAA7C,KAAA,SAAA,EAA0B,KAAA,QAAA,EAJtC,KAAA,cAAgB,CAAC,IAAK,WAKpB,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAChC,EAAe,KAAK,SAAW,EAAI,aAAe,UACxD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,KAAK,QAAO;;YAElD,EAAK;;gCAEe,KAAK,SAAQ;;sCAEP,EAAY;;;;QAMlD,MAAA,MC1CA,YAwBmC,GACjC,GAAM,GAAO,KACP,EAAwB,EAAK,QAAO;;MAEtC,EAAK,UAAS;MACd,EAAK,UAAS;MACd,EAAK,UAAS;;;;;OAMlB,MAAO,IAA8B,EAAI,eAGR,GAEjC,GAAM,GAAc,GAAI,cACpB,CAAC,GAAI,EAAG,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,GAAI,EAAG,EAAG,IAClE,MAAO,IAAoC,EAAI,eAGf,GAEhC,GAAM,GAAwB,GAAI,aAAY,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,IAC9D,MAAO,IAAmC,EAAI,GAGhD,YACI,EAA2B,EAAe,EAC1C,EAAwB,EACxB,GACF,GAA+B,EAAO,GACtC,GAAM,GAAU,GAAyB,GAEnC,EAAQ,EAAG,WACjB,UAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAO,KACxD,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,eAAgB,EAAG,iBAC5D,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EAAI,WAAM,MAAA,GAAG,cAAc,EAAO,EAAG,mBAAoB,EAAG,WAChE,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAO,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAC5C,EAAa,QACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,QACzD,cAIL,GACF,MAAO,GAAc,gCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAG,mBAIvC,GACF,MAAO,GAAc,oCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAAyC,GACzC,EAAc,mBAAoB,EAAc,kCAIlD,GACF,MAAO,GAAc,kCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAG,2BAIL,GACF,MAAO,GAAc,sCAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EAAQ,GAAwC,GAC3D,EAAG,KAAM,EAAG,mBAId,GACF,MAAO,GAAc,0CAInB,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,MAAO,IACH,EAAI,EAAO,EACX,GAA+C,GAAgB,EAAG,KAClE,EAAc,kCAIhB,EAA2B,EAC3B,GACF,GAAM,GAAY,EACZ,EAAW,EAAI,EACf,EAAU,EAAI,EAAM,EAAI,EAC9B,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,KAC7C,GAAM,GAAU,GACZ,EAAI,EAAS,eAAgB,EAAc,EAAG,EAAQ,GAC1D,MAAO,IACH,GACI,EAAI,EAAS,KAAM,EAAc,EAAG,EAAQ,eAIlD,EAA2B,EAAuB,EAClD,EAAgB,EAAkB,GACpC,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAEhE,GAAI,GAA2B,EAAuB,EACtD,AAAI,YAAgB,YAClB,GAAgB,GAAI,YAAW,EAAQ,EAAS,GAChD,EAAgB,EAAG,cACnB,EAAiB,EAAG,MAEpB,GAAgB,GAAI,cAAa,EAAQ,EAAS,GAClD,EAAgB,EAAG,MACnB,EAAiB,EAAc,2BAGjC,EAAc,IAAI,GAElB,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAgB,EAAO,EAAQ,EAAG,EAAG,KACvD,EAAe,KAEvB,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA2B,EAC3B,GAEF,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,KAChE,AAAK,EAAqB,eAAgB,YACxC,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAO,MAAO,EAAO,OAAQ,EAAG,EAAG,KAC9D,EAAG,cAAgB,EAAqB,QAEhD,GACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,WAAY,EAAG,EAAG,KAAM,EAAG,KAAM,EAAG,cACvC,KAIV,GAAwB,EAAI,WAAM,MAAA,GAAG,YAAY,EAAG,WAAY,oBAI9D,EAA6B,EAAc,EAC3C,GAEF,GAAM,GAAS,EAAI,eACnB,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,KAGrD,GAAM,GAAgB,EAChB,EAAiB,EACjB,EAAkB,EAAgB,EAAiB,EAAO,EAEhE,UACI,EACA,WAAM,MAAA,GAAI,WACN,EAAI,kBAAmB,EAAiB,EAAI,eAIpD,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAG,EAAG,EAAS,EAAM,EAAI,KAAM,EAAI,MAAO,KAExE,GACI,EAAK,WAAM,MAAA,GAAI,WAAW,EAAI,kBAAmB,QAE9C,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAM,EAEN,EAAiB,GAAI,cAAa,GAExC,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAAc,EACzC,GACI,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAG,EAAA,EAAA,GAGJ,EAAc,EACd,EAAiB,GAAI,YACvB,GAA4C,EAAO,EAAS,IAEhE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAG,EAAG,EAAc,sBAAuB,EAAG,cACpD,KAID,GAAI,cAAa,EAAe,oBAIrC,EAA2B,EAAqB,EAAe,EAC/D,EAAc,EAAsB,EACpC,GACF,GAAM,GAAM,EAEN,EACF,GAAI,cAAa,GACb,EAAc,IAEtB,SAAI,WAAW,EAAI,kBAAmB,GACtC,EAAI,iBAAiB,EAAI,kBAAmB,EAAG,GAC/C,EAAI,WAAW,EAAI,kBAAmB,MAE/B,cAIL,EAA2B,EAC3B,GACF,GAAM,GAAa,GAAI,cAAa,EAAe,EAAe,GAClE,UACI,EACA,WAAM,MAAA,GAAG,WACL,EAAG,EAAG,EAAc,EAAc,EAAG,KAAM,EAAG,MAAO,KAEtD,owBCxTT,kBAgDE,WAAY,GANZ,KAAA,cAAmC,KACnC,KAAA,QAA6B,KACrB,KAAA,SAAW,GAqOX,KAAA,oBAAsB,GAsPtB,KAAA,YAA0B,GAtdhC,GAAM,GAAY,EAAA,MAAM,UAAU,iBAClC,AAAI,GAAM,KACR,MAAK,GAAK,EACV,GAAgB,EAAW,IAE3B,KAAK,GAAK,GAAgB,GAG5B,GAAI,GAAqB,2BACnB,EAA0B,8BAChC,GAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAM,GAAgB,oBAChB,EAAqB,yBAI3B,GAFA,KAAK,sBACD,GAA+B,KAAK,GAAI,GACxC,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,sHAKN,GADA,KAAK,0BAA4B,KAAK,GAAG,aAAa,GAClD,GAAwB,KAAK,GAAI,GACnC,KAAK,8BACD,GAA+B,KAAK,GAAI,WACnC,EAAA,MAAM,IAAI,4BACnB,KAAM,IAAI,OACN,uIAIN,EAAqB,yBACjB,GAAwB,KAAK,GAAI,GACnC,KAAK,0BACD,KAAK,GAAG,aAAa,WAChB,GAAwB,KAAK,GAAI,GAC1C,KAAK,8BACD,KAAK,GAAG,aAAa,OAEzB,MAAM,IAAI,OAAM,uDAIpB,KAAK,aAAe,GAA8B,KAAK,IACvD,KAAK,YAAc,GAA6B,KAAK,IACrD,KAAK,YAAc,GAA6B,KAAK,IAErD,KAAK,cACD,GAA0B,KAAK,GAAI,KAAK,2BAG9C,cAAA,eAAY,EAAA,UAAA,QAAK,KAAjB,WACE,MAAO,GAAA,MAAM,QAAQ,0CAGhB,EAAA,UAAA,QAAP,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAEF,AAAI,KAAK,SAAW,MAClB,QAAQ,KACJ,wKAIN,AAAI,KAAK,eAAiB,MACxB,QAAQ,KACJ,sMAKN,GAAM,GAAK,KAAK,GAChB,GAAwB,EAAI,WAAM,MAAA,GAAG,WACrC,GAAwB,EAAI,WAAM,MAAA,GAAG,gBAAgB,EAAG,YAAa,QACrE,GAAwB,EAAI,WAAM,MAAA,GAAG,kBAAkB,EAAK,eAC5D,GAAwB,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,aAAc,QACjE,GACI,EAAI,WAAM,MAAA,GAAG,WAAW,EAAG,qBAAsB,QACrD,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAK,eACvD,KAAK,SAAW,IAGX,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,2BAAP,SAAkC,EAAc,GAE9C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,yBAAP,SACI,EACA,GACF,KAAK,kBACL,GAAoC,KAAK,GAAI,EAAS,IAGjD,EAAA,UAAA,2BAAP,SACI,EAAuB,EAAe,EAAgB,GACxD,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,EAAO,EAAQ,EAAM,KAAK,gBAG3C,EAAA,UAAA,iCAAP,SAAwC,EAAc,GAEpD,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,0BAAP,SAAiC,EAAc,GAE7C,YAAK,kBACE,GACH,KAAK,GAAI,EAAM,EAAS,KAAK,gBAG5B,EAAA,UAAA,oBAAP,SAA2B,GAA3B,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,KAAK,gBAAkB,GACzB,IAA6C,KAAK,GAAI,KAAK,aAC3D,KAAK,cAAgB,MAEvB,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAGxD,EAAA,UAAA,gDAAP,SACI,EAAuB,EAAc,GADzC,GAAA,GAAA,KAEE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAM,EAAS,EAAK,kBAGhC,EAAA,UAAA,+BAAP,SACI,EAAqB,EAAe,EAAc,EAClD,EAAsB,GACxB,MAAO,IACH,KAAK,GAAI,EAAQ,EAAO,EAAM,EAAS,EAAc,EACrD,KAAK,gBAGJ,EAAA,UAAA,gCAAP,SAAuC,EAAqB,GAE1D,MAAO,IAA2C,KAAK,GAAI,EAAQ,IAG9D,EAAA,UAAA,wBAAP,SACI,EAAuB,EAAc,GACvC,KAAK,yBAAyB,GAC9B,GAAM,GAAS,GACX,KAAK,GAA8B,EAAM,EAAS,KAAK,eAC3D,YAAK,6BACE,GAGF,EAAA,UAAA,sBAAP,WACE,GAAM,GAAe,KAAK,YAAY,KAAK,IAC3C,MAAO,MAAK,UAAU,IAGhB,EAAA,UAAA,YAAR,SAAoB,GAApB,GAAA,GAAA,KACM,EACA,EAEJ,GAAI,EAAA,MAAM,QAAQ,4BAChB,GAAM,GAAM,EAEN,EAAO,EAAI,UAAU,EAAI,2BAA4B,GAC3D,EAAG,QAEH,EAAgB,WACd,GAAM,GAAS,EAAI,eAAe,EAAM,EAAG,GAC3C,MAAO,KAAW,EAAI,kBAClB,IAAW,EAAI,qBAGrB,EAAQ,MACH,AACH,GAAA,MAAM,UAAU,gDAAkD,EACpE,GAAQ,KAAK,aACb,KAAK,WACL,EAAgB,WAAM,MAAA,GAAK,iBACvB,EACA,EAAA,MAAM,UAAU,mDAMpB,EAAgB,WAAM,MAAA,IAGxB,MAAO,CAAC,MAAK,EAAE,cAAa,IAGvB,EAAA,UAAA,gCAAP,SACI,EAAuB,EACvB,GAFJ,GAAA,GAAA,KAGE,MAAO,MAAK,qBACR,EACA,WAAM,MAAA,IACF,EAAK,GAAI,EAAc,MAK1B,EAAA,UAAA,cAAP,SAAqB,GACnB,KAAK,kBACL,GAAM,GAAK,KAAK,GACV,EACF,GAAgC,EAAI,GAClC,EAA4B,GAA8B,GAC1D,EAAwB,GAAyB,GACvD,UAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,aAAa,EAAS,KAC3D,GAAuB,EAAI,GAC3B,AAAI,KAAK,OACP,GAA2B,EAAI,GAEjC,AAAK,KAAK,qBACR,MAAK,WAAW,GAChB,KAAK,oBAAsB,GACvB,EAAI,KAAK,QAAS,KAAK,eAEtB,GAGF,EAAA,UAAA,cAAP,SAAqB,GAArB,GAAA,GAAA,KACE,KAAK,kBACL,AAAI,IAAY,KAAK,SACnB,MAAK,QAAU,MAEjB,AAAI,GAAW,MACb,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,cAAc,MAI1D,EAAA,UAAA,WAAP,SAAkB,GAAlB,GAAA,GAAA,KACE,KAAK,kBACL,KAAK,QAAU,EACf,AAAK,KAAK,SAAW,MAAS,KAAK,OACjC,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,WAAW,MAGrD,EAAA,UAAA,mBAAP,SACI,EAAuB,EACvB,GAEF,MAFE,AAAA,KAAA,QAAA,GAAA,IACF,KAAK,kBACD,EACK,GACH,KAAK,GAAI,EAAS,GAEf,GACH,KAAK,GAAI,EAAS,IAInB,EAAA,UAAA,qBAAP,SAA4B,EAAuB,GAAnD,GAAA,GAAA,KAEE,YAAK,kBACE,GACH,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,kBAAkB,EAAS,MAGjD,EAAA,UAAA,0BAAP,SAAiC,EAAuB,GAEtD,YAAK,kBACE,KAAK,GAAG,mBAAmB,EAAS,IAGtC,EAAA,UAAA,sBAAP,SACI,EAAkC,EAClC,GACF,KAAK,kBACL,KAAK,mBACL,GACI,KAAK,GAAI,EAAoB,EAAiB,IAG7C,EAAA,UAAA,uBAAP,SACI,EAAmC,EAAc,GACnD,KAAK,6BAA6B,EAAqB,EAAS,IAG3D,EAAA,UAAA,6BAAP,SACI,EAAyC,EAAc,GACzD,KAAK,kBACC,GAAA,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,KAAK,6BAA6B,EAA2B,EAAO,IAG/D,EAAA,UAAA,2BAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAK,iCACD,EAAa,EAAU,EAAY,IAGlC,EAAA,UAAA,iCAAP,SACI,EAAkB,EAAiB,EACnC,GACF,KAAM,IAAI,OAAM,sDAGX,EAAA,UAAA,cAAP,WACE,AAAI,KAAK,SAAW,MAClB,GAA2B,KAAK,GAAI,KAAK,SAE3C,GAA+B,KAAK,KAG/B,EAAA,UAAA,eAAP,WACE,KAAK,kBACL,KAAK,mBACL,GAAM,GAAK,KAAK,GAChB,AAAI,KAAK,OACP,KAAK,gBAEP,GACI,EAAI,WAAM,MAAA,GAAG,aAAa,EAAG,UAAW,EAAG,EAAG,eAAgB,MAG7D,EAAA,UAAA,+BAAP,WAAA,GAAA,GAAA,KACE,KAAK,kBACL,GAAwB,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,YAGzC,EAAA,UAAA,uBAAR,WAEE,MAAI,MAAK,6BAA+B,MACtC,MAAK,4BACD,GACI,KAAK,GACL,EAAA,MAAM,UACF,kDAAoD,EACpD,kCACA,6BAIP,KAAK,6BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGN,EAAA,UAAA,6BAAR,WACE,MAAO,MAAK,0BAGd,EAAA,UAAA,WAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EAAQ,EAAI,cAClB,SAAI,WAAW,EAAI,iBAAkB,GAC9B,EAET,GAAM,GAAM,KAAK,+BACX,EAAQ,EAAI,iBAClB,SAAI,cAAc,EAAI,iBAAkB,GACjC,GAGT,EAAA,UAAA,SAAA,WACE,GAAI,EAAA,MAAM,UAAU,kDAAoD,GACtE,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BACjB,EAAI,SAAS,EAAI,kBACjB,OAEF,GAAM,GAAM,KAAK,+BACjB,EAAI,YAAY,EAAI,mBAGT,EAAA,UAAA,uBAAb,SAAoC,wGAClC,MAAA,CAAA,EAAM,EAAA,KAAK,YACP,WAAM,MAAA,GAAK,UAGP,EAAK,iBACD,EACA,EAAA,MAAM,UACF,2DAPhB,SAAA,OAQA,CAAA,EAAO,KAAK,aACR,EAAO,EAAA,MAAM,UAAU,wDAGrB,EAAA,UAAA,aAAR,SAAqB,EAAmB,GACtC,GAAI,IAAsB,EACxB,MAAO,MAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GAEX,EAAmB,EAAI,kBAAkB,EAAO,EAAI,cAE1D,MAAO,GAAmB,SAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,kBAErC,MAAO,GAAmB,MAItB,EAAA,UAAA,iBAAR,SAAyB,EAAmB,GAE1C,GAAI,IAAsB,EACxB,MAAO,GAGT,GAAI,IAAsB,GACxB,GAAM,GAAM,KAAK,GACX,EAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,wBACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,cAE1B,GAAM,GAAM,KAAK,+BAEX,EACF,EAAI,kBAAkB,EAAO,EAAI,4BACrC,MAAI,MAAK,UAAY,MACnB,MAAK,SAAW,KAAK,GAAG,aAAa,EAAI,mBAGpC,GAAa,CAAC,KAAK,WAI9B,EAAA,UAAA,UAAA,SAAU,GAAV,GAAA,GAAA,KACE,MAAO,IAAI,SAAc,SAAA,GACvB,EAAK,cAAc,WAAM,MAAA,GAAa,iBAAiB,WAAM,MAAA,UAMjE,EAAA,UAAA,UAAA,WAGE,OADM,GAAQ,GAAqB,KAAK,YAAY,IAAI,SAAA,GAAK,MAAA,GAAE,YACtD,EAAI,EAAG,GAAK,EAAO,EAAE,GACrB,GAAA,GAAA,KAAA,YAAA,GAAA,UACP,IAEF,KAAK,YAAc,KAAK,YAAY,MAAM,EAAQ,IAG5C,EAAA,UAAA,cAAR,SAAsB,EAAyB,GAA/C,GAAA,GAAA,KAEE,GADA,KAAK,YAAY,KAAK,CAAC,SAAQ,EAAE,UAAS,IACtC,KAAK,YAAY,OAAS,EAE5B,OAGF,EAAA,KAAK,YAAY,WACf,SAAK,YAEE,EAAK,YAAY,SAAW,KAI/B,EAAA,UAAA,yBAAR,SAAiC,GAC/B,KAAK,kBACL,GACI,KAAK,GAAI,EAAS,KAAK,aAC3B,AAAI,KAAK,OACP,GAA+B,KAAK,KAIhC,EAAA,UAAA,2BAAR,WACE,AAAI,KAAK,eAAiB,KACxB,IACI,KAAK,GAAI,KAAK,cAAe,KAAK,aACtC,AAAI,KAAK,OACP,GAA+B,KAAK,KAGtC,GAA6C,KAAK,GAAI,KAAK,cAIvD,EAAA,UAAA,qBAAR,SACI,EACA,GACF,KAAK,yBAAyB,GAC9B,GAAM,GAAS,IACf,YAAK,6BAEE,GAGD,EAAA,UAAA,6BAAR,SACI,EAA8C,EAC9C,GACF,KAAK,kBACL,GAAM,GAAK,KAAK,GAChB,GACI,EAAI,EAAgC,KAAK,aAC7C,AAAI,KAAK,OACP,GAA+B,GAEjC,KAAK,cAAgB,EACrB,GAAwB,EAAI,WAAM,MAAA,GAAG,SAAS,EAAG,EAAG,EAAO,KAC3D,GAAwB,EAAI,WAAM,MAAA,GAAG,QAAQ,EAAG,EAAG,EAAO,MAGpD,EAAA,UAAA,iCAAR,SACI,EAAW,EAAW,EAAe,GADzC,GAAA,GAAA,KAEE,KAAK,kBACL,GACI,KAAK,GAAI,WAAM,MAAA,GAAK,GAAG,QAAQ,EAAG,EAAG,EAAO,MAG1C,EAAA,UAAA,gBAAR,WACE,GAAI,KAAK,SACP,KAAM,IAAI,OAAM,4CAIZ,EAAA,UAAA,iBAAR,WACE,GAAI,KAAK,SAAW,KAClB,KAAM,IAAI,OAAM,qCAGtB,iBAaqC,GAEnC,OADI,GAAI,EACD,EAAI,EAAI,OAAQ,EAAE,GACvB,GAAM,GAAS,EAAI,KACnB,GAAI,CAAC,EACH,MAGJ,MAAO,GAAI,ECtnBb,YAgEI,EAAqB,EAAuB,EAC5C,GACF,GAAM,GAAW,EAAQ,SACnB,EAA0B,EAAO,IAAI,SAAC,EAAO,GACjD,GAAM,GAAuB,CAC3B,aAAc,EAAM,MACpB,SAAU,EAAM,UAAY,KAAO,EAAM,QAAQ,SACjD,UAAW,EAAM,UACjB,SAAU,EAAM,UAAY,GAAQ,EAAM,QAAQ,SAClD,WAAY,MAEd,MAAI,GAAM,SAAW,MAAQ,EAAM,QAAQ,OAAS,MAChD,EAAM,QAAQ,MAAM,WAAa,GACnC,GAAU,WAAa,EAAM,QAAQ,MAAM,YAEtC,CAAC,KAAM,EAAQ,cAAc,GAAI,UAAS,KAE7C,EAAe,EAAW,IAAI,SAAA,GAAK,MAAA,GAAE,YACrC,EAA0B,CAC9B,aAAc,EAAO,MACrB,SAAU,EAAO,QAAQ,SACzB,UAAW,GACX,SAAU,EAAO,QAAQ,SACzB,WAAY,MAER,EAAS,GACX,EAAY,EAAc,EAAU,EAAQ,cAE1C,EAAe,EAAM,cAAc,GAGrC,EAA+B,KAC7B,EAAS,EAAM,mBAAmB,EAAc,MAAO,IAC7D,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,GAAS,EAAM,mBAAmB,EAAc,WAAY,KAK9D,OADM,GAA2D,GACxD,EAAI,EAAG,EAAI,EAAQ,cAAc,OAAQ,KAChD,GAAM,GAAU,EAAQ,cAAc,GAChC,EAAc,GACpB,EAAiB,GACb,EAAM,mBAAmB,EAAc,EAAS,GACpD,EAAiB,SAAS,GACtB,EAAM,mBAAmB,EAAc,SAAS,EAAW,GAGjE,MAAO,CACL,QAAO,EACP,OAAM,EACN,aAAY,EACZ,iBAAgB,EAChB,aAAY,EACZ,aAAY,EACZ,OAAM,EACN,OAAM,GAIV,YACI,EAAyB,GAC3B,GAAI,EAAW,SAAW,EAAO,OAC/B,KAAM,OACF,4BAA4B,EAAW,OAAM,gBAC7C,sBAAqB,EAAO,OAAM,YAGxC,EAAW,QAAQ,SAAC,EAAG,GACrB,GAAM,GAAS,EAAE,aACX,EAAQ,EAAO,GACf,EAAS,EAAM,MAErB,GAAI,CAAC,EAAA,KAAK,YAAY,EAAQ,GAC5B,KAAM,OACF,kDACA,6BAA4B,EAAM,QAAQ,EAAM,gBAGtD,GAAI,EAAE,WAAa,EAAM,UACvB,OAGF,GAAM,GAAY,EAAE,SACd,EAAY,EAAM,UAAY,KAAO,EAAM,QAAQ,SACzD,GAAI,CAAC,EAAA,KAAK,YAAY,EAAW,GAC/B,KAAM,OACF,6DACA,yBAAwB,EAAS,QAAQ,EAAS,8BAMxD,EAAqB,EAAqB,EAC1C,EACA,GAEF,GAAyB,EAAO,aAAc,GAC9C,GAAyB,CAAC,EAAO,cAAe,CAAC,IAEjD,GAAM,GAAS,EAAO,QAAQ,QACxB,EAAc,EAAO,QAAQ,SACnC,AAAI,EAAO,QAAQ,SACjB,EAAM,6BAA6B,EAAQ,EAAY,GAAI,EAAY,IAEvE,EAAM,uBAAuB,EAAQ,EAAY,GAAI,EAAY,IAEnE,EAAM,WAAW,EAAO,cAGxB,AAAI,EAAA,MAAM,UAAU,mBAAqB,GACvC,CAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,WAGtC,AAAI,EAAO,SAAW,MACpB,EAAM,GAAG,UAAU,EAAO,OAAQ,KAIpC,EAAO,QAAQ,SAAC,EAAO,GACrB,GAAM,GAAU,EAAO,QAAQ,cAAc,GACvC,EAAS,EAAO,iBAAiB,GACjC,EAAe,EAAO,iBAAiB,SAAS,GAEtD,GAAI,GAAU,KAEZ,OAGF,GAAI,EAAM,WAER,GAAI,EAAA,KAAK,cAAc,EAAM,OAAS,EACpC,EAAM,GAAG,UAAU,EAAQ,EAAM,cAAc,SAE/C,GAAI,GAAO,EAAM,cACjB,AAAM,YAAgB,eACpB,GAAO,GAAI,cAAa,IAE1B,EAAM,GAAG,WAAW,EAAQ,GAE9B,OAIF,AAAI,EAAM,QAAQ,OAAS,MAAQ,GAAgB,MACjD,EAAM,GAAG,UAAU,EAAc,EAAM,QAAQ,MAAM,YAGvD,EAAM,sBAAsB,EAAM,QAAQ,QAAS,EAAQ,KAG7D,AAAI,GAAe,MACjB,EAAY,EAAO,EAAO,cAE5B,EAAM,6BAIJ,EAAuB,EAAsB,GAC/C,GAAI,GAAY,GAChB,EAAO,OAAO,GAAQ,QAAQ,SAAA,GAC5B,GAAM,GAAY,EAAE,SAAW,MAAQ,EAAE,QAAQ,OAAS,MACtD,EAAE,QAAQ,MAAM,WAAa,EAC3B,EAAW,EAAE,UAAY,UAAY,EAAE,QAAQ,SACrD,GAAgB,EAAE,MAAK,IAAI,EAAQ,IAAI,IAEzC,GAAM,GAAc,EAAQ,SACxB,EAAM,EAAQ,YAAY,KAE9B,UAAO,IAAM,EAAY,IAAM,EACxB,EC5OT,AAqBA,GAAA,IAAA,WAOE,WACI,EAAuB,EACvB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAsBnB,OAnBE,GAAA,EAAA,YACA,EAAA,EAAA,WACA,EAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,QACA,EAAA,EAAA,SACA,EAAA,EAAA,cACA,EAAA,EAAA,eACA,EAAA,EAAA,WAEK,EAAA,EAAA,KAAM,EAAA,EAAA,IACP,EAAmB,EAAa,EAChC,EAAO,KACP,EAAiB,IAAe,eAChC,EAAS,EAAiB,EAAI,EAC9B,EAAS,EAAiB,EAAI,EAEhC,EAAW,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,IAC1B,GAAY;gCACY,EAAG;yBACV,EAAG;;4BAEA,EAAY,GAAE,aAAa,EAAY,GAAE;0CAC3B,EAAQ,QAAQ,EAAY,MAC1D,EAAG;6BACc,EAAc,aAAa,EAAgB;;sBAElD,EAAW,GAAO;;qDAEa,EAAQ,QACjD,EAAW,OAAO,EAAI;+BACH,EAAa,2BAChC,EAAgB,QAAQ,EAAU;;wBAEtB,EAAW,GAAO;;2CAEC,EAAU;;sBAE/B,EAAc;;2BAET,GAAM,EAAI,GAAG;;;;;2BAKb,GAAM,EAAI,GAAG;;;;;;;UAWpC,KAAK,SAAW;;;;;;;;;UASV,EAAQ;;UAER,EAAK,OAAM;;MAIrB,MAAA,MC5GA,AAmBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAgB,EAAc,EAChD,GANJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMtB,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;;wBASI,EAAG,UAAU,EAAG;;oCAEJ,EAAI;;;;;0BAKd,EAAW;;;MAKrC,MAAA,MClEA,AAmBA,GAAA,IAAA,WAUE,WACI,EAAsB,EAAqB,EAAc,EACzD,GAXJ,KAAA,cAAgB,CAAC,aAAc,cAAe,MAC9C,KAAA,YAAwB,GAWtB,KAAK,YAAc,EACnB,KAAK,MAAQ,EAAW,GACxB,KAAK,YAAc,EACnB,KAAK,KAAO,EACZ,KAAK,MAAQ,EACb,KAAK,KAAO,EACZ,KAAK,SAAW;;;;;;;;8BAQU,KAAK,MAAK;oDACY,EAAW;yCACtB,KAAK,MAAK;0BACzB,EAAW;;;sCAGC,KAAK,MAAK;;;;;;;;;;;;;;;yBAevB,EAAK,oBAAoB,EAAI;;;;;;;yCAOb,EAAK;0BACpB,EAAI;;;;0CAIY,EAAI;;;;;;;;;;;;;;MAgB9C,MAAA,MC/FA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAgB,EAAc,EAChD,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAExB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,GAAM,GAAM,EACN,EAAO,EAAO,GAAK,EACzB,KAAK,YAAc,EAMnB,GAAI,GACE,EAAQ,SAAS,EAAI,aAAa,EAAK,UAC7C,AAAI,IAAS,GACX,EAAc,eAAe,EAAK,IAC7B,AAAI,IAAS,EAClB,EAAc,QAAQ,EAAK,IAE3B,EAAc,WAAW,EAAK,cAAc,EAAI,MAGlD,KAAK,SAAW;;;;;;;;gCAQY,KAAK,YAAY,GAAE;gCACnB,KAAK,YAAY,GAAE;;;;;;;;;;;;;;;iCAelB,EAAG;;;;;;;;;;;yBAWX,EAAG,UAAU,EAAG;;;6DAGoB,EAAI;;;;;;;;;;;;;;;;;;;;;0CAqBvB,EAAW;;;MAKrD,MAAA,MC/GA,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EAAY,EAAwB,EAAuB,EACjE,KAAK,SAAW;iCACa,EAAM,KAAK,EAAO;;;;;;;;;;;;;;gCAcnB,EAAqB;kBACnC,EAAc;gDACgB,EAAY;;oCAExB,EAAS,UAAS;;;;;kCAKpB,EAAoB;kDACJ,EAAW;;sCAEvB,EAAS,SAAQ;;;;;;;gCAOvB,EAAS;;;;qCAIJ,EAAoB;;;;;;;;MAUzD,MAAA,MAEA,GAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAAM,UAKrB,KAAK,YAAc,EAAS,QAC5B,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAuB,EAAI,EAAS,QAAQ,MACvD,EAAS,EAAwB,EAAI,EAAS,QAAQ,IACtD,EAAU,EAAuB,EAAI,EAAS,QAAQ,KAEtD,EACF,EAAuB,EAAwB,EAAuB,EAC1E,KAAK,SAAW;iCACa,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;gCAiBhC,EAAoB;mBACjC,EAAa;gDACgB,EAAW;;oCAEvB,EAAS,SAAQ;;;;;kCAKnB,EAAqB;sBACjC,EAAc;kDACc,EAAY;;sCAExB,EAAS,UAAS;;;;;;oCAMpB,EAAoB;wBAChC,EAAa;oDACe,EAAW;;wCAEvB,EAAS,SAAQ;;;;;;;kCAOvB,EAAS;;;;;;yBAMlB,EAAqB,MAAM,EAAoB;yBAC/C,EAAoB;;;;;;;;;MAW7C,MAAA,MChLA,AAmBA,GAAA,IAAA,WAOE,WACI,EAAkC,EAClC,EAAoB,EAAoB,EACxC,EAA2B,GAD3B,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IAAoB,AAAA,IAAA,QAAA,GAAA,IACxC,AAAA,IAAA,QAAA,GAAA,MAA2B,AAAA,IAAA,QAAA,GAAA,IAT/B,KAAA,cAAgB,CAAC,UAAW,WAC5B,KAAA,aAAe,GACf,KAAA,aAAe,GAQb,KAAK,YAAc,EAEnB,GAAM,GAAY,EAAa,EAAO,GAAK,EAAO,GAC5C,EAAwB,KAAK,KAAK,EAAY,GAE9C,EAAU,EAAa,cAAgB,cACvC,EAAU,EAAa,cAAgB,cACvC,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAC1D,EAAW,EAAa,CAAC,SAAU,UAAY,CAAC,SAAU,UAE5D,EAAoB,GAAI,EAAyB,GACrD,AAAI,GACF,CAAI,EACF,EAAoB;;YAEhB,EAAU;WAGd,EAAoB;YAChB,EAAU;WAIhB,EAAyB,gCAG3B,GAAM,GAAiB,EAAU,kCAAoC,GACrE,AAAI,GACF,KAAK,cAAc,KAAK,QAG1B,AAAI,GACF,KAAK,cAAc,KAAK,0BAG1B,KAAK,SAAW;QACZ,EAAiB;;sCAEa,EAAqB;;;;8BAI7B,EAAqB;sCACb,EAAO;sCACP,EAAO;;;;uBAItB,EAAS,GAAE,MAAM,EAAS,GAAE;uBAC5B,EAAS,GAAE,MAAM,EAAS,GAAE;;;;;;;;;UASzC,EAAc;;UAEd,EAAsB;;;;MAMhC,MAAA,MChGA,AAoBA,GAAA,IAAA,WAQE,WAAY,EAAmB,EAAqB,GAPpD,KAAA,cAAgB,CAAC,SAQf,KAAK,YAAc,CAAC,EAAW,GAE/B,KAAK,SAAW;;;;;;;;;;8BAUU,GAAc,GAAC;;;;;;;;;;0BAUnB,GAAc,GAAC;;MAKvC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,MAAO,UAAC,EAAqB,GAC3B,AAAI,EAAK,SAAW,MAClB,GAAK,QAAU,EAAM,mBAAmB,EAAc,SAExD,EAAM,GAAG,UAAU,EAAK,QAAS,KAGvC,KChEA,AAmBA,GAAA,IAAA,WAQE,WACI,EAAoB,EAAe,EAAiB,GARxD,KAAA,cAAgB,CAAC,WASf,KAAK,YAAc,CAAC,EAAY,GAEhC,KAAK,SAAW;;;;8BAIU,EAAQ,YAAY,EAAO;;;MAKzD,MAAA,MCxCA,AAqBA,GAAA,IAAA,WAOE,WACI,GAPJ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEzB,GAAI,IAAS,EACX,KAAK,SAAW;;;;aAMhB,GAAM,GAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EACF,GAAwB,EAAM,EAAa,GACzC,EAAQ,GACV,EAAM,EAAY,EAAY,OAAS,GACvC,EAAY,EAAY,OAAS,GAAI,GACnC,EAAS,GAAU,EAAa,GAEtC,KAAK,SAAW;;YAEV,EAAK;;eAEF,EAAoB;;;cAGrB,EAAK;;6BAEU,EAAM;;;SAMnC,MAAA,MAEA,YAA4B,EAAc,GAGxC,OAFM,GAAS,GAEN,EAAM,EAAG,GAAO,EAAG,IAC1B,OAAS,GAAM,EAAG,GAAO,EAAG,KAG1B,OAFI,GAAW,KAAQ,EAAI,IAAM,OAAK,KAAK,KAAQ,EAAI,IAAM,OAEpD,EAAI,EAAG,EAAI,EAAM,IACxB,EAAW,EAAK,EAAK,OAAS,EAAI,GAAE,IAAM,EAG5C,EAAO,KAAK,GAGhB,MAAO,GAGT,YACI,EAAc,EAAiB,GACjC,GAAI,IAAS,EACX,MAAO,QAAQ,EAAM,GAIvB,OADI,GAAO,GACF,EAAI,EAAO,EAAG,EAAI,EAAM,IAC/B,GAAW,EAAK,GAAE,OAAO,EAAM,GAC/B,AAAI,EAAI,EAAO,GACb,IAAQ,MAIZ,MAAO,GAGT,YACI,EAAc,EAAc,EAAc,GAC5C,GAAI,IAAS,EACX,MAAO,GAGT,GAAM,GAAY,EAAK,MAAM,IAE7B,MAAO;cACK,EAAU,GAAE;cACZ,EAAU,GAAE;;;;0BAIA,EAAI;0BACJ,EAAI;IAI9B,YAAmB,EAAiB,GAClC,GAAM,GAAO,EAAM,OACb,EAAe,GAAmB,EAAM,GAC9C,MAAI,KAAS,EACJ;wBACa,EAAM,GAAE;kBAIvB,QAAQ,EAAa,GAAE;8BACF,EAAa,GAAE;8BACf,EAAa,GAAE;uCACN,EAAa,GAAE,ICrItD,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAClB,GANJ,KAAA,cAAgB,CAAC,KAOf,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KACnD,GAAM,GAAO,EAAO,OACd,EAAO,GAAkB,GAEzB,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EACF,CAAC,YAAa,YAAa,YAAa,aAAa,MAAM,EAAG,GAElE,GAAI,IAAS,GACX,KAAK,SAAW;sBACA,EAAK;oBACP,EAAG;;;;;8BAKO,EAAa;;;;;QAMrC,OAEF,KAAK,SAAW;QACZ,EAAI,YAAY,EAAI,IAAI,EAAK;QAC7B,EAAI,UAAU,EAAI,IAAI,EAAG;;;UAGvB,EAAI;;4BAEc,EAAa;;YAE7B,EAAI;2BACW,EAAc;;;MAKzC,MAAA,MCrEA,AAqBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAClB,GARJ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAAS,IACxB,SAAC,EAAG,GAAM,MAAA,GAAE,GAAqB,EAAO,GAAK,EAAE,KA4BnD,OA3BM,GAAO,EAAO,OACd,EAAQ,GAAkB,GAE1B,EAAQ,EAAS,IAAI,SAAA,GAAK,MAAA,GAAE,KAAI,KAAK,KACrC,EAAM,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,GAAE,GAAK,EAAO,KAAI,KAAK,KACpD,EAAS,GAAY,KAAM,GAC3B,EAAS,GAAY,SAAU,GAC/B,EAAY,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAC1D,EACF,IAAS,EAAI,SAAW,QAAQ,EAAO,MAAM,IAAI,OAAM,IAErD,EAAiB,CAClB,EAAK,mBAAuB,EAAO,EAAO,GAAE;YACzC,EAAM;QAEZ,IAAS,EAAI,GAAK;;SAEf,EAAO,EAAO,GAAE;YACb,EAAO,EAAO,GAAE,MAAM,KAAK,YAAY,EAAO,GAAE,MACtD,IAAS,EAAI,GAAK,KAAK,EAAO,EAAO,GAAE;cAC/B,EAAM,OAGV,EAAc,IAAS,EACzB,0BACA,6DACA,EAAW,GACN,EAAI,EAAG,EAAI,IAAS,EAAI,EAAI,EAAG,EAAI,EAAG,IAC7C,GAAY;UACR,EAAe,GAAE;cACb,EAAW;mBACN,EAAC,aAAa,EAAa;;YAElC,EAAK;mBACE,EAAC,uBAAuB,EAAO,OAAM,MAAM,EAAS;;QAInE,GAAa,IAAS,EAAI,KAAO,KAEjC,KAAK,SAAW;cACN,EAAK,YAAY,EAAK,IAAI,EAAK;cAC/B,EAAK,UAAU,EAAK,IAAI,EAAG;;;UAG/B,EAAK;;UAEL,EAAQ;;;MAKlB,MAAA,MCrFA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MACzB,EAA0B,cAAc,EAAS,SAAQ,YAC3D,EAAS,QAAO,YAAY,EAAS,WAAU,OAC7C,EACF,SAAS,EAAS,QAAO,YAAY,EAAS,WAAU,OAExD,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;sCACgB,EAAY,KAAK,EAAW;mCAC/B,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBnB,EAAqB;sBACjC,EAAc;;;kCAGF,EAAS,SAAQ;;;;oCAIf,EAAoB;wBAChC,EAAa;;;oCAGD,EAAS,QAAO;;;;;;;;;;0BAU1B,EAAS;;;mCAIzB,GAAoB,EAAsB,EACA,EACvB,QAAQ,EAAoB,SAAO;;;;;;QAO1D,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;oCACgB,EAAY,KAAK,EAAW;iCAC/B,EAAM,KAAK,EAAO;0CACT,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;kCAkBZ,EAAmB;;;;gCAIrB,EAAqB;oBACjC,EAAc;;;gCAGF,EAAS,SAAQ;;;;kCAIf,EAAsB;uCACjB,EAAa;;;;yCAIX,EAAa;6CACT,EAAa;6CACb,EAAa;;;cAG5C,EAAa;;;gCAGK,EAAsB;gBACtC,KAA6B,GAAC;;;;;;;;cAQhC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;;;;;cAKxC,EAAa;uBACJ,KAA6B,GAAC;;;yCAGZ,EAAa;6CACT,EAAa;;;;cAI5C,EAAa;;;oBAGP,EAAW;;MAI/B,MAAA,MAEA,GAAA,WAKE,WACI,EAAmC,EACnC,EAA2B,EAC3B,GACF,GAF6B,AAAA,IAAA,QAAA,GAAA,IAC3B,AAAA,IAAA,QAAA,GAAA,IAPJ,KAAA,cAAgB,CAAC,KAQX,IAAa,OAAS,EACxB,KAAM,IAAI,OAAM,8CAGlB,GAAM,GAAc,EAAS,YACvB,EAAc,EAAS,YACvB,EAAe,EAAS,aACxB,EAAc,EAAS,YACvB,EAAgB,EAAS,cACzB,EAAiB,EAAS,eAC1B,EAAgB,EAAS,cACzB,EAAuB,EAAS,qBAChC,EAAwB,EAAS,sBACjC,EAAuB,EAAS,qBAEhC,EAAW,EAAS,QAAQ,MAC5B,EAAS,EAAS,QAAQ,IAC1B,EAAU,EAAS,QAAQ,KACjC,KAAK,YAAc,EAAS,SAE5B,GAAM,GAAY,IAAa,MAE3B,EAAsB,MAM1B,GALA,AAAK,GAEH,GAAsB,gBAGpB,GACF,GAAM,GAAY,KAElB,KAAK,SAAW;;oBAEF,EAAW,KAAK,EAAY,KAAK,EAAW;mCAC7B,EAAQ,KAAK,EAAM,KAAK,EAAO;;;;;;;;;;;;;;;;;;kCAkBhC,EAAoB;sBAChC,EAAa;;;kCAGD,EAAS,QAAO;;;;oCAId,EAAqB;wBACjC,EAAc;;;oCAGF,EAAS,SAAQ;;;;sCAIf,EAAoB;0BAChC,EAAa;;;sCAGD,EAAS,QAAO;;;;;;;;;;4BAU1B,EAAS;;;qCAI3B,GACK,EACI,cAAc,EAAS,QAAO,YAC1B,EAAS,SAAQ,YAAY,EAAS,QAAO,YAC7C,EAAS,WAAU,QACvB,UAAU,EAAS,SAAQ,YACvB,EAAS,QAAO,YAAY,EAAS,WAAU,QACxD,QAAQ,EAAqB,MAAM,EAAoB;6BACxC,EAAoB,SAAO;;;;;;;QAQlD,OAGF,GAAM,GAAY,MAEd,EAAiB,EAAQ,IAAI,EAAQ,IAAI,EAAQ,qEAErD,AAAI,IAAa,OACf,GAAc,oBAGhB,GAAM,GAAyB,KAAK,MAAM,EAAc,GAAK,EACvD,EAA2B,EAAc,EAEzC,EAAgB;YACd,EAAS;;;wBAGG,EAAS;;MAI7B,KAAK,SAAW;;gBAEJ,EAAW,KAAK,EAAY,KAAK,EAAW;iCAC3B,EAAQ,KAAK,EAAM,KAAK,EAAO;0CACtB,EAAmB;;;;;;8BAM/B,EAAS,QAAO;;;;;;;;;;;;;;;;;;;kCAmBZ,EAAmB;;;;gCAIrB,EAAoB;oBAChC,EAAa;;;gCAGD,EAAS,QAAO;;;;kCAId,EAAqB;oBACnC,EAAc;;;kCAGA,EAAS,SAAQ;;;;oCAIf,EAAsB;yCACjB,EAAa;;;;+CAIP,EAAa;mDACT,EAAa;mDACb,EAAa;;;gBAGhD,EAAa;;;kCAGK,EAAsB;kBACtC,KAA6B,GAAC;;;;;;;;gBAQhC,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;;;;;gBAK5C,EAAa;yBACJ,KAA6B,GAAC;;;+CAGR,EAAa;mDACT,EAAa;;;;gBAIhD,EAAa;;;sBAGP,EAAW;;;MAKjC,MAAA,MCvcA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,KAOR,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,OAAQ,EAAA,EAAA,QACtC,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAI,GAAsB,MACtB,EAAY,GAEhB,AAAI,IAAe,OACjB,EAAsB,MACjB,AAAI,IAAe,MAExB,GAAsB,cACtB,EAAY,OACP,AAAI,IAAe,OAExB,GAAsB,eACtB,EAAY,OAGd,GAAI,GAAiB,EAAU,IAAI,EAAU,IAAI,EAAU,qEAG3D,AAAI,IAAe,MACjB,EAAc,WACT,AAAI,IAAe,OACxB,EAAc,YACT,AAAI,IAAe,MACxB,EAAc,WACT,AAAI,IAAe,OACxB,GAAc,YAGhB,GAAM,GAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEzC,EAAgB;YACZ,KAAe,OAAK;;mBAEb,KAAe,QAAM;;;;wBAIhB,EAAS;;MAIzB,EAAU,OAEd,AAAI,IAAe,MACjB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SACL,AAAI,IAAe,OACxB,GAAsB,MACtB,EAAgB;;;;QAKhB,EAAU,SAGZ,GAAI,GAAmB,GACvB,AAAI,EAAS,EAAa,GACxB,GAAmB;oCACW,EAAM;;;SAKtC,KAAK,SAAW;0CACsB,EAAmB;;;;UAInD,EAAgB;;;;;;;;kCAQQ,EAAU;;kCAEV,EAAmB;;;;;;8BAMvB,EAAqB;;YAEvC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;YAC/B,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;qBACJ,KAA4B,GAAC;YACtC,EAAO,aAAa,EAAO;;;;;;;YAO3B,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCvKA,AAoBA,GAAA,IAAA,WAOE,WAAY,EAAuC,GANnD,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAOb,KAAK,YAAc,EAGnB,OADI,GAAW,GACN,EAAI,EAAG,EAAI,EAAG,KACrB,GAAI,GAAS,eACb,AAAI,EAAI,IAAM,GACZ,IAAU,kBAEZ,AAAI,EAAI,GACN,IAAU,kBAGZ,GAAY;UACR,EAAM;UACN,GAAI,EAAI,0CAA4C,IAAE;;;;;;mBAM7C,EAAC;;UAEV,GAAI,EAAI,IAAM,IAAE;QAItB,KAAK,SAAW;QACZ,GAAuB,GAAW;QAClC,GAA+B,GAAY;;;;;;;;qBAQ9B,EAAY,GAAE;qBACd,EAAY,GAAE;;UAEzB,EAAQ;;;;MAMlB,MAAA,MAEA,YAAgC,GAC9B,GAAM,GACF,GAA+C,CAAC,IAAK,IAAK,KAAM,GAEpE,MAAO;;QAED,EAAsB;;;ICnF9B,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;;sDAMY,GAAU,GAAC;;;;;;qDAMZ,GAAS,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAgC/D,MAAA,MC1IA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MA+B/D,MAAA,MC3EA,AAmBA,GAAA,IAAA,WAOE,WACI,EAA8C,EAC9C,EAAkB,GARtB,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GACf,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAGlD,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;uCACxB,EAAQ;;;;;;;;;;;;;;;;;;;;;;gCAsBf,GAAQ,GAAC;uCACF,GAAW,GAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;MAkDnD,MAAA,MCzHA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAc,EAAa,GAJvC,KAAA,cAAgB,CAAC,MACjB,KAAA,YAAwB,GAItB,KAAK,YAAc,EAAE,MACf,GAAA,GAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GACZ,EAAA,EAAA,MAAG,EAAA,EAAA,GAAS,EAAA,EAAA,GAMZ,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAmC,CACtC,GAAgB,EAAU,EAAK,EAAU,EAAI,EAC7C,GAAgB,EAAS,EAAK,EAAS,EAAI,GAGxC,EAAc,EAAe,GAAK,EAAe,GACjD,EAAa,EAAe,GAAK,EAAe,GAEhD,EAAiB,EAAI,EACrB,EAAgB,EAAI,EAIpB,EAAa,KAAK,KAAK,GAAkB,EAAK,EAC9C,EAAY,KAAK,KAAK,GAAiB,EAAK,EAElD,KAAK,SAAW;;;;;;;;;;0CAUsB,EAAW;yCACZ,EAAU;;6CAEN,EAAc;4CACf,EAAa;;oCAErB,EAAS;mCACV,EAAQ;;;;;;;;;;;;;;kCAcT,EAAO;;;;;;;;oCAQL,EAAM;;;;;sBAKpB,EAAe,GAAE;sCACD,EAAe,GAAE;;;wBAG/B,EAAe,GAAE;wCACD,EAAe,GAAE;;;4BAG7B,EAAO;kBACjB,EAAY;;;;4BAIF,EAAM;kBAChB,EAAY;;;;;;;;;;;;MAc9B,MAAA,MC/HA,AAmBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EAAkB,GANtB,KAAA,cAAgB,CAAC,KACjB,KAAA,YAAwB,GAMf,GAAA,GAAA,EAAA,GAAO,EAAA,EAAA,GAAW,EAAA,EAAA,GAAU,EAAA,EAAA,GACnC,KAAK,YAAc,CAAC,EAAO,EAAW,EAAU,GAEhD,GAAM,GAAoC,CACvC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAG5C,EAAqC,CACxC,GAAgB,EAAY,EAAK,EAAY,EAAI,EACjD,GAAgB,EAAW,EAAK,EAAW,EAAI,GAI5C,EAAY,EAAe,MAAQ,MAEzC,KAAK,SAAW;;YAER,EAAgB,GAAK,EAAiB,GAAE;YACxC,EAAgB,GAAK,EAAiB,GAAE;uCACb,EAAS,OAAO,EAAQ;;;;;;;;;;;;;8DAaD,EAAS;;;;;;MAQvE,MAAA,MCpEA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAI5C,GAFA,KAAK,YAAc,EAEf,IAAS,GACX,KAAK,SAAW;;;2BAGK,EAAO,GAAE;;QAG9B,OAEF,GAAM,GAAa,SAAC,GAClB,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,aAAa,EAAC,QAE5B,UAAU,EAAC,KAEd,EAAW,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,KAAI,KAAK,KACpD,EAAO,GAAkB,GAE/B,KAAK,SAAW;;UAEV,EAAI;yBACW,EAAQ;;MAIjC,MAAA,MC1DA,AAqBA,GAAA,IAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,IAAI,OACN,kCAAkC,EAAI,gCAE5C,KAAK,YAAc,EACnB,GAAM,GAAW,GAAY,KAAM,GAC7B,EACC,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACrD,EAAa,EAAS,EAAO,GAAE,UAAU,KAAK,YAAY,EAAO,GACjE,EAAO,GAAkB,GAC/B,AAAI,IAAS,EACX,KAAK,SAAW;;;;uCAIiB,EAAO,GAAE;cAClC,EAAO,GAAE;eACR,EAAU;2CACkB,EAAO,GAAE;kBAClC,EAAO,GAAE;;;;QAMrB,KAAK,SAAW;;YAEV,EAAI;;uBAEO,EAAK,EAAS,SAAQ;eAC9B,EAAU;yBACA,EAAK,EAAS,SAAQ;;eAEhC,EAAO;yBACG,EAAK,EAAS,SAAQ;iBAC9B,EAAU;2BACA,EAAK,EAAS,SAAQ;;;;;MAQ7C,WAAc,GACZ,MAAO,GAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAc,GACZ,SAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QAChD,EAAS,EAAO,GAAK,IAAM,EAAS,EAAO,GAAK,QACzC,EAAW,GAGpB,WAAoB,GAClB,GAAM,GAAgB,EAAO,IAAI,SAAC,EAAG,GAAM,MAAA,GAAW,EAAG,KACnD,EAAW,EAAc,KAAK,KAC9B,EAAY,EAAc,MAAM,IAAI,KAAK,KAC/C,MAAO,mBAAmB,EAAQ,WAAW,EAAS,KAGxD,WAAoB,EAAW,GAC7B,MAAI,GAAK,QAAQ,KAAO,IAAM,EAAO,KAAO,EAChC,EAAO,GAAE,MAAM,EAAU,GAAE,OAE9B,GAAG,EAAU,IAI5B,MAAA,MC7GA,AAoBA,GAAA,IAAA,WAKE,WACI,EAAoB,EAAkB,EACtC,EAAqB,EAAmB,EACxC,GAPJ,KAAA,cAAgB,CAAC,UAAW,UAAW,gBAQrC,KAAK,YAAc,EACnB,GAAM,GAAc,GAAkB,EAAQ,QACxC,EAAQ,GAAkB,EAAM,QAClC,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,QAElB,GAAM,GAAiB,cAAc,EAAa,IAE9C,EAAgB,GACpB,AAAI,IAAgB,EAClB,EAAgB,IACX,AAAI,IAAgB,GACzB,GAAgB,gBAElB,GAAM,GAAiB,cAAc,EAAa,IAE5C,EAAe,EAAW,EAAI,aAAe,UACnD,KAAK,SAAW;UACV,EAAW,cAAc,EAAW,IAAI,EAAO;;;YAG7C,EAAK;;;gCAGe,EAAU;;kCAER,EAAQ;kCACR,EAAc;0CACN,EAAY;;;uBAG/B,EAAc;;;;;;QAQrC,MAAA,MCvEA,AAoBA,GAAA,IAAA,WAKE,WACI,EACA,GANJ,KAAA,cAAgB,CAAC,IAAK,cAOpB,GAAM,GAAa,EAAU,WACvB,EAAY,EAAU,UACtB,EAAS,EAAU,OACnB,EAAc,EAAU,YACxB,EAAU,EAAc,KAAK,KAAK,EAAS,GACjD,KAAK,YAAc,CAAC,EAAW,GAE/B,GAAM,GAAsB,MACtB,EAAc,WAEd,EAAwB,KAAK,MAAM,EAAa,GAAK,EACrD,EAA0B,EAAa,EAEvC,EAAgB;;MAIlB,EAAwB,GAC5B,AAAI,EAAS,EAAa,GACxB,GAAwB;oCACM,EAAM;;;SAMtC,GAAI,GAA4B,GAChC,AAAI,EAAS,EAAa,GACxB,GAA4B;oCACE,EAAM;;;SAMtC,KAAK,SAAW;0CACsB,EAAmB;;;UAGnD,EAAqB;;;;;UAKrB,EAAyB;;;;;;;;;YASvB,EAAW,cAAc,EAAU;wDACS,EAAW;;;;8BAIrC,EAAqB;;;;;;;;;;;;;;;;YAgBvC,EAAa;;;iCAGQ,EAAqB;cACxC,KAA4B,GAAC;;;;;;;;;;;;;;;;;YAiB/B,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;qBACJ,KAA4B,GAAC;;;;;;;;;;;;;;;YAetC,EAAa;;oBAEL,EAAW;;MAI/B,MAAA,MCjKA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAe,EAAiB,GAJ5C,KAAA,cAAgB,CAAC,IAAK,IAAK,KAKzB,KAAK,YAAc,EAEnB,GAAI,GACA,EACJ,GAAI,EAAO,EACT,KAAM,OAAM,kBAAkB,EAAI,yBAGpC,GAAI,IAAS,EACX,EAAW,QACX,EAAU,aAKV,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,WAClD,EAAa,GACb,EAAc,GACX,EAAI,EAAG,EAAI,EAAM,OAAQ,IAChC,EAAY,KAAK,GAAG,EAAc,IAClC,AAAI,EAAI,GACN,EAAW,KAAK,GAAG,EAAc,IAGrC,EAAU,EAAW,OACrB,EAAW,EAAY,OAGzB,GAAM,GAAQ,GAAkB,GAEhC,KAAK,SAAW;;UAEV,EAAK;4BACa,EAAO;;2BAER,EAAQ;;2BAER,EAAQ;;;MAKnC,MAAA,MCjEA,AAqBA,GAAA,IAAA,WASE,WAAY,GARZ,KAAA,cAAgB,CAAC,UASf,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAc,qBAAqB,KAAK,KAAI,KAC5C,EAAe,GAAU,KAAK,MAEhC,EACE,EAAW,EAAS,IAAI,SAAC,EAAG,GAChC,MAAO,aAAa,GAAO,GAAE,YAAY,EAAC,cAAc,GAAO,GAAE,MAEnE,EAAO;UACD,EAAK;UACL,EAAK;UACL,EAAS,KAAK;GAAK;QAEzB,KAAK,SAAW;QACZ,EAAW;;UAET,EAAI;8BACgB,EAAY;;MAKxC,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KAEM,GAAS,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAEzC,YAAmB,GACjB,GAAI,IAAS,EACX,MAAO,YACF,GAAI,GAAQ,EACjB,MAAO,IAAO,MAAM,EAAG,GAAM,IAAI,SAAA,GAAK,MAAA,aAAe,IAAG,KAAK,KAE7D,KAAM,OAAM,oBAAoB,EAAI,yBCpFxC,AAsBA,GAAA,IAAA,WAWE,WAAY,GAVZ,KAAA,cAAgB,CAAC,UACjB,KAAA,aAAe,GACf,KAAA,aAAe,GASb,KAAK,YAAc,EACnB,KAAK,KAAO,EAAS,OAErB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAS,GAAY,SAAU,KAAK,MACpC,EAAY,GAAY,YAAa,KAAK,MAE1C,EACF,KAAK,OAAS,EAAI,YAAc,QAAQ,EAAU,MAAM,IAAI,OAAM,IAChE,EACF,wBAAwB,EAAU,OAAM,MAAM,EAAS,IACrD,EAAW;mBACF,EAAU;cACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;YACnB,EAAU,KAAK,KAAO,GAAE;;MAG1B,EAAW,KAAK,OAAS,EAAI,GAAK;UAClC,EAAO,KAAK,KAAO,GAAE;cACjB,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;YACpD,EAAU,KAAK,KAAO,GAAE;qBACf,EAAU;gBACf,EAAO,KAAK,KAAO,GAAE,MAAM,EAAS,KAAK,KAAO,GAAE;cACpD,EAAU,KAAK,KAAO,GAAE;uBACf,EAAU;;;MAKvB,EAAiB,KAAK,MAAQ,EAChC;cACM,EAAK,IAAI,EAAS,IAAI,SAAC,EAAG,GAAM,MAAA,SAAS,EAAC,MAAK,OAAM,KAC3D,EAAS,IAAI,SAAC,EAAG,GAAM,MAAG,GAAU,GAAE,MAAM,EAAO,GAAE,YAAY,EAAC,OAC7D,KAAK;GACd,KAAK,SAAW;0BACM,KAAK,KAAI;;UAEzB,EAAK;UACL,EAAK;UACL,EAAc;;UAEd,EAAQ;UACR,EAAQ;;;MAMhB,SAAA,UAAA,mBAAA,SAAmB,GAAnB,GAAA,GAAA,KACE,GAAI,EAAM,SAAW,KAAK,KACxB,KAAM,OACF,aAAa,KAAK,KAAI,mCACtB,qBAAoB,EAAM,OAAM,MAEtC,MAAO,UAAC,EAAqB,GAC3B,GAAI,EAAK,UAAY,MACnB,GAAK,SAAW,EAAM,0BAA0B,EAAc,SAC1D,EAAK,UAAY,MAGnB,OAGJ,EAAM,GAAG,WAAW,EAAK,SAAU,KAGzC,KCtGA,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAiB,EAAmB,GAJhD,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,GAAM,GAAO,EAAK,OACZ,EAAa,GAAkB,EAAK,QACpC,EAAQ,GAAkB,EAAK,QAEjC,EAAY,GAChB,GAAI,IAAS,EACX,EAAY,gCAEZ,GAAI,GAAa,EACjB,EACI,EAAK,IAAI,SAAC,EAAG,GACP,WACO,EAAK,SAAW,EACnB,oBAAoB,EAAC,aAAa,EAAC,IACnC,UAAU,GAAa,GAAC,eAAe,EAAC,aAAa,EAAC,MAE3D,KAAK,KAGhB,KAAK,SAAW;QACZ,EAAU,YAAY,EAAU,IAAI,EAAK;QACzC,EAAU,cAAc,EAAU,IAAI,EAAO;;;UAG3C,EAAK;yBACU,EAAS;;MAIlC,MAAA,MCxDA,AAuBA,GAAA,IAAA,WAUE,WAAoB,GAAA,KAAA,MAAA,EATZ,KAAA,gBAAkB,EAClB,KAAA,gBAAkB,EAClB,KAAA,mBAAqB,EACrB,KAAA,cAAgB,EAEhB,KAAA,aAAkD,GAClD,KAAA,WAAa,GACb,KAAA,aAAkD,GAI1D,SAAA,UAAA,eAAA,SACI,EAA2B,EAC3B,GACF,GAAM,GAAkB,GAAkC,EAAO,GAE3D,EAAW,GAAuB,EAAS,EAAiB,GAClE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAEhC,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAS,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cACpD,GAEJ,GAAI,KAAK,aAAa,GAAU,OAAS,GACvC,KAAK,kBACL,KAAK,kBACL,KAAK,eAAiB,EACtB,KAAK,MACL,GAAM,GAAa,KAAK,aAAa,GAAU,QAC/C,YAAK,aAAa,GAAU,KAAK,GAC1B,EAGT,GAAI,GACJ,MAAI,KAAoB,GAAoB,mBAC1C,EAAa,KAAK,MAAM,0BAA0B,EAAQ,GAAI,EAAQ,IACjE,AAAI,IAAoB,GAAoB,mBACjD,EACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,IAC/D,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AAAI,IAAoB,GAAoB,iBACjD,EACI,KAAK,MAAM,2BAA2B,EAAQ,GAAI,EAAQ,IACzD,AACH,IAAoB,GAAoB,0BAC1C,GACI,KAAK,MAAM,iCAAiC,EAAQ,GAAI,EAAQ,KAEtE,KAAK,aAAa,GAAU,KAAK,GAEjC,KAAK,kBACL,KAAK,oBAAsB,EAC3B,KAAK,MAEE,GAGT,EAAA,UAAA,eAAA,SACI,EAAuB,EACvB,EAA8B,GAChC,GAAI,KAAK,cAAgB,KAEvB,OAEF,GAAM,GACF,GAAkC,EAAgB,GAChD,EAAW,GAAuB,EAAO,EAAiB,GAChE,AAAM,IAAY,MAAK,cACrB,MAAK,aAAa,GAAY,IAGhC,GAAM,GAAW,GACb,EAAO,EAAiB,KAAK,MAAM,GAAI,KAAK,MAAM,cAClD,GACE,EAAqB,EAAA,MAAM,IAAI,kCACrC,AAAI,IAAuB,IACvB,KAAK,mBAAqB,EAC5B,MAAK,MAAM,oBAAoB,GAC/B,KAAK,oBAAsB,GAE3B,MAAK,aAAa,GAAU,KAAK,GACjC,KAAK,kBACL,KAAK,eAAiB,GAGxB,KAAK,kBAEL,GAAM,GAAU,KAAK,aAAa,GAC5B,EAAW,EAAQ,QAAQ,GACjC,GAAI,EAAW,EACb,KAAM,IAAI,OACN,4EAGN,EAAQ,OAAO,EAAU,GACzB,KAAK,OAGC,EAAA,UAAA,IAAR,WACE,GAAI,CAAC,KAAK,WACR,OAEF,GAAM,GAAQ,KAAK,gBAAkB,KAAK,gBAC1C,QAAQ,IACJ,YAAgB,KAAK,gBAAe,MAAM,KAAK,gBAC/C,IAAI,EAAK,KACb,GAAM,GAAY,KAAK,cAAgB,KAAK,mBAC5C,QAAQ,IAAI,oBAAoB,KAAK,oBACrC,QAAQ,IAAI,iBAAiB,KAAK,cAAa,KAC3C,KAAK,MAAM,IAAM,GAAU,OAGjC,OAAA,eAAI,EAAA,UAAA,oBAAiB,KAArB,WACE,MAAO,MAAK,oDAGd,OAAA,eAAI,EAAA,UAAA,eAAY,KAAhB,WACE,MAAO,MAAK,+CAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,mBAAA,WACE,MAAO,MAAK,iBAGd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,cAAgB,KAEvB,OAEF,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,OAAW,KAAY,MAAK,aAC1B,KAAK,aAAa,GAAU,QAAQ,SAAA,GAClC,EAAK,MAAM,oBAAoB,KAGnC,KAAK,aAAe,KACpB,KAAK,aAAe,KACpB,KAAK,gBAAkB,EACvB,KAAK,gBAAkB,EACvB,KAAK,mBAAqB,EAC1B,KAAK,cAAgB,GAEzB,KAEA,YACI,EAA2B,GAE7B,GAAM,GAAQ,EACd,GAAI,IAAmB,EAAM,KAC3B,MAAO,GACF,GAAI,IAAmB,EAAM,KAClC,MAAO,GACF,GAAI,IAAmB,EAAM,QAClC,MAAO,IACF,GAAI,IAAmB,EAAG,KAC/B,MAAO,IACF,GAAI,IAAmB,EAAM,QAClC,MAAO,GAET,KAAM,IAAI,OAAM,2BAA2B,eAIzC,EAAyB,EACzB,EAA2B,EAC3B,GAMF,GAAM,GACF,GAAiC,EAAiB,GAElD,EACJ,GAAI,GACI,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAEpB,EAAc,EAAc,OAGtB,GAAA,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAEd,EAAc,EAAQ,EAGxB,GAAM,GAAkB,GAA0B,EAAI,GACtD,MAAO,GAAc,EAGvB,YACI,EACA,GACF,OAAQ,OACD,IAAoB,mBACvB,MAAO,IAAwC,OAC5C,IAAoB,mBACvB,MAAO,IAA+C,OACnD,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,iBACvB,MAAO,IAAyC,OAC7C,IAAoB,yBACvB,MAAO,IAA+C,WAEtD,KAAM,IAAI,OAAM,iCAAiC,IAIvD,YAAwC,GAEtC,MAAI,GAAA,MAAM,QAAQ,gCACZ,EACK,GAAoB,mBAEtB,GAAoB,iBAGzB,EACK,GAAoB,mBAEtB,GAAoB,iBAG7B,YACI,EAA8B,GAChC,GAAI,IAAmB,GAAa,OAClC,MAAO,IAAoB,mBACtB,GAAI,IAAmB,GAAa,QAAU,GAAkB,KACrE,MAAO,IAA+B,GACjC,GACH,IAAmB,GAAa,UAChC,IAAmB,GAAa,OAClC,MAAO,IAAoB,yBAE7B,KAAM,IAAI,OAAM,gCAAgC,GAGlD,YACI,EAAgC,EAChC,GACF,MAAU,GAAa,GAAE,IAAI,EAAa,GAAE,IAAI,EAAe,IAAI,ECxRrE,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,GAAK,EAAK,GAEpC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAe,GAAgB,GAErC,KAAK,SAAW;;UAEV,EAAK;yBACU,EAAY;;MAIrC,MAAA,MAEA,YAAyB,GACvB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,iBAAiB,EAAI,yBAEnC,GAAI,IAAS,EACX,MAAO,eAAe,EAAO,GAAE,IAMjC,OAHM,GAAgB,CAAC,UAAW,UAAW,UAAW,UAAW,WAE7D,EAAe,GACZ,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAa,KAAK,QAAQ,EAAc,GAAE,KAAK,EAAO,GAAE,KAE1D,MAAO,GAAa,OC5DtB,AAoBA,GAAA,IAAA,WAKE,WAAY,EAAkB,GAJ9B,KAAA,cAAgB,CAAC,KAKf,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MAEM,GAAoB,0BAEb,GAAS,YAET,GAAM,iBAEN,GAAO,GAAoB;;EAI3B,GAAQ,GAAoB;;EAI5B,GAAM,0CAEN,GAAO;;;uBAGG,EAAA,aAAa,gBAAe;kBACjC,EAAA,aAAa,WAAU;;cAIpB,GAAA,MAAA,KAAA,QAAA,GAAA,GACZ,GAAoB;mCACM,EAAK;KAIjC,GAAM,IAAM,aAEN,GAAO,kBAEP,GAAQ,mBAER,GAAO;;;EAKP,GAAS,0BAET,GAAS,0BAET,GAAY,wCAEZ,GAAQ;;;;;;;;;;;;;;;EAiBR,GAAM,iBAEN,GAAQ,uBAER,GAAM;kBAGN,GAAQ,uBAER,GAAO,kBAEP,GAAQ,yBAER,GAAU,sCAeV,GAAW;;;;;;;;;;;;;;;;;;;;EAsBX,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;;;;EAO3B,GAAO,GAAoB;;EAI3B,GAAO;;;EAKP,GAAO;;;EAKP,GAAO;;;EAKP,GAAQ,GAAoB,qCAE5B,GAAQ,GAAoB;;sCAI5B,GAAQ,GAAoB;;+CAI5B,GAAM;;;;cAIL,EAAA,aAAa,MAAK;eACjB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;eACnB,EAAA,aAAa,OAAM;;;;;;EAQrB,GAAa,kBAEb,GAAc,6BAEd,GAAS,wBAET,GAAQ,YC9NrB,AAmBO,GAAM,IAAS,YAET,GAAM;;;;;;;;;EAWN,GAAO;;;;;;;;;;EAYP,GAAQ;;;;;;;;;;EAYR,GAAM;;;;;;;;;EAWnB,GAAA,WAOE,WAAY,EAAkB,GAN9B,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,KAAK,YAAc,EACnB,KAAK,SAAW;;UAEV,EAAS;;;;;;;;;MAWnB,MAAA,MCzFA,AAqBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KACjB,KAAA,aAAe,GACf,KAAA,aAAe,GAKb,KAAK,YAAc,EACnB,GAAM,GAAO,EAAY,OAEnB,EAAW,GAAY,KAAM,GAC7B,EAAQ,GAAkB,GAC1B,EAAe,GAAgB,EAAM,GACrC,EAAY,EAAS,MAAM,IAC3B,EAAS,GAAQ,EAAI,KAAO,QAAQ,EAAU,KAAK,KAAI,IAE7D,KAAK,SAAW;;UAEV,EAAK;kCACmB,EAAY;;4CAEF,EAAM;;MAIlD,MAAA,MC/CA,AA2BO,GAAA,IAAA,EAAA,aAAA,aACD,GAAQ,EAAA,aAAa,MACrB,GAAO,EAAA,aAAa,KACpB,GAAW,EAAA,aAAa,SACxB,GAAY,EAAA,aAAa,UAiFlB,GAAkB,KAClB,GAAkB,KA4BzB,GAAuE,eAE9C,GAC7B,MAAI,KAAgB,KAGpB,IAAa,GAAgB,IACtB,GAAa,GAGtB,YACI,EAAqC,GACvC,GADuC,AAAA,IAAA,QAAA,GAAA,IACnC,IAAe,SACjB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,OACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,MACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GACF,GAAI,IAAe,QACxB,MAAI,GACK,GAEF,GAET,KAAM,IAAI,OAAM,cACZ,EAAU,oDAKhB,GAAM,IAA6B,IAK7B,GAAyB,IAC/B,cACE,MAAI,GAAA,MAAM,OAAO,QAAU,KAClB,KAED,EAAA,MAAM,OAAO,OAAO,OAAS,EAAA,MAAM,OAAO,OAAO,MACjD,OAAO,iBACX,GAAyB,KAAO,KAM/B,GAAM,IAA8B,mBAEL,GAAA,EAAA,GAkCpC,WAAY,GAAZ,GAAA,GACE,EAAA,KAAA,OAAO,KACP,GA/BM,EAAA,YAAc,GAAI,SAGlB,EAAA,gBAAkB,GAAI,SAGtB,EAAA,aAAe,GAAI,SACnB,EAAA,cAAgB,EAOhB,EAAA,aAAe,EAEf,EAAA,eAAiB,EAUjB,EAAA,kBAAoB,GACpB,EAAA,sBAAwB,GAgXxB,EAAA,eAAiB,EAq+DjB,EAAA,SAAW,GAj1Eb,CAAC,EAAA,MAAM,QAAQ,aACjB,KAAM,IAAI,OAAM,yCAGlB,GAAI,GAAS,MACX,GAAM,GAAK,GAAgB,EAAA,MAAM,UAAU,kBAC3C,EAAK,YAAc,GAAe,EAAA,MAAM,UAAU,kBAClD,EAAK,MAAQ,GAAI,IAAa,GAC9B,EAAK,OAAS,EAAG,OACjB,EAAK,oBAAsB,OAE3B,GAAK,MAAQ,EACb,EAAK,YAAc,GACnB,EAAK,oBAAsB,GAC3B,EAAK,OAAS,EAAM,GAAG,OAEzB,SAAK,eAAiB,GAAI,IAAe,EAAK,OAC9C,EAAK,mBAAqB,KAE1B,EAAK,QAAU,GAAI,GAAA,YAAY,EAAM,EAAA,YAGvC,SAAA,UAAA,WAAA,WACE,MAAO,MAAK,QAAQ,aACf,MAAK,WAAa,KAAK,WAAW,aAAe,GAClD,KAAK,gBAGX,EAAA,UAAA,MAAA,SAAM,EAAuB,EAAiB,GAK5C,GAJA,AAAI,GAAA,MAAM,QAAQ,mCACd,EAAA,MAAM,QAAQ,WAChB,KAAK,uBAAuB,GAE1B,IAAU,aAAe,GAAU,KACrC,KAAM,IAAI,OACN,yEAGN,GAAM,GAAS,GACf,YAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,IAC1D,GAIT,EAAA,UAAA,OAAA,SAAO,GACL,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,YAIV,EAAA,UAAA,OAAA,SAAO,GACL,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,aAIZ,EAAA,UAAA,KAAA,SAAK,EAAgB,EAAuB,EAAiB,GAK3D,GAHA,AAAI,EAAA,MAAM,QAAQ,UAChB,KAAK,uBAAuB,GAE1B,IAAU,YACZ,KAAM,IAAI,OACN,yEAGN,KAAK,QAAQ,IACT,EACA,CAAC,MAAK,EAAE,MAAK,EAAE,OAAM,EAAE,MAAO,GAAa,OAAQ,SAAU,KAGnE,EAAA,UAAA,8BAAA,SAA8B,GAC5B,GAAM,GAAS,EAAW,OAE1B,GAAI,KAAK,QAAQ,IAAI,IACnB,GAAM,GAAc,KAAK,QAAQ,IAAI,GAErC,EAAY,WAEZ,AAAI,EAAY,SAAW,GACzB,KAAK,YAAY,KAKvB,EAAA,UAAA,SAAA,SAAS,GACP,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,eAAgB,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SAKpD,GAAI,GAAS,MACX,GAAI,GAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEtC,GAAM,GACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,SAAS,EAAI,QAC/B,YAAK,8BAA8B,GAC5B,EAET,GAAI,GAAU,KACZ,MAAO,MAAK,qBAAqB,GAEnC,GAAI,IAAU,SACZ,MAAO,GAET,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GACJ,GAAI,IAAU,aACZ,GAAM,GAAa,EAAe,KAAK,WACjC,EAAa,EAAe,KAAK,WACvC,EAAS,EAAA,aAAa,uBAAuB,EAAY,OAEzD,GAAS,KAAK,qBAAqB,GAGrC,MAAI,IACF,MAAK,gBAAkB,EAAA,KAAK,MAAQ,GAE/B,KAAK,qBAAqB,EAAQ,IAGrC,EAAA,UAAA,KAAN,SAAW,6IACT,GAAI,KAAK,YAAY,IAAI,GACjB,SAAc,KAAK,YAAY,IAAI,GACzC,CAAA,EAAO,GAAI,SAAoB,SAAA,GAAW,MAAA,GAAY,KAAK,MAQ7D,GANM,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAyD,EAAO,OAAxD,EAAiD,EAAO,MAAjD,EAA0C,EAAO,MAA1C,EAAmC,EAAO,MAAnC,EAA4B,EAAO,eAAnB,EAAY,EAAO,SAKnE,GAAS,KACP,SAAO,OACX,AAAI,EACF,EAAU,GAAI,IAAqB,EAAO,IAE1C,EAAU,GAAI,IAAe,EAAO,IAEhC,EACF,KAAK,gBAAgB,EAAS,CAAC,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAAI,GACtD,EAAO,KAAK,KAAK,EAAI,QAC3B,KAAK,8BAA8B,GACnC,CAAA,EAAO,GAGT,GAAI,GAAU,KACZ,MAAA,CAAA,EAAO,KAAK,qBAAqB,IAGnC,GAAI,CAAC,EAAA,MAAM,QAAQ,iCACf,EAAA,MAAM,UAAU,mBAAqB,EACvC,KAAM,IAAI,OACN,sGAIF,GAAS,KAGb,AAAI,IAAU,aAAe,EAAA,MAAM,IAAI,2BAErC,GAAoB,KAAK,OAAO,GAC1B,EAAU,KAAK,QAAQ,IAAI,EAAkB,QAEnD,EAAS,GAAA,KAAK,OAAM,wBAAuB,MAAA,EAAA,CACvC,EAAQ,SAAO,OAAK,GAA0B,MAGpD,KAAK,YAAY,IAAI,EAAQ,IAEzB,IAAU,YAEZ,CAAA,EAAM,KAAK,MAAM,yBAFf,CAAA,EAAA,UAEF,EAAA,8BAKE,KAAU,YACD,CAAA,EAAM,QAAQ,IACrB,CAAC,EAAe,KAAK,OAAQ,EAAe,KAAK,UAFnD,CAAA,EAAA,UACI,SAAK,EAAA,OAEL,EAAa,EAAG,GAChB,EAAa,EAAG,GACtB,EAAO,EAAA,aAAa,uBAChB,EAA4B,gBAC3B,AAAI,GAAU,KACnB,EAAO,KAAK,qBAAqB,GAE3B,GAAO,EAAA,KAAK,cAAc,GAChC,EAAO,KAAK,MAAM,gCAAgC,EAAQ,qBAE5D,MAAI,IAAqB,MACvB,KAAK,8BAA8B,GAE/B,EAAY,KAAK,qBAAqB,EAAQ,GAE9C,EAAc,KAAK,YAAY,IAAI,GACzC,KAAK,YAAY,OAAO,GAGxB,EAAY,QAAQ,SAAA,GAAW,MAAA,GAAQ,KACvC,AAAI,KAAK,gBAAgB,IAAI,IAC3B,MAAK,gBAAgB,OAAO,GAC5B,KAAK,YAAY,GACjB,KAAK,kBAEP,CAAA,EAAO,SAGD,EAAA,UAAA,uBAAR,SAA+B,GAC7B,GAAI,GAAU,KACZ,OAEF,OAAS,GAAI,EAAG,EAAI,EAAO,OAAQ,KACjC,GAAM,GAAM,EAAO,GACnB,GAAI,CAAC,GAA4B,GAC/B,KAAI,GAAA,MAAM,QAAQ,gCACV,MACF,aAAa,EAAG,iJAIhB,MAAM,aAAa,EAAG,4CAK1B,EAAA,UAAA,qBAAR,SAA6B,SACrB,EAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,SACf,EAAO,EAAA,KAAK,cAAc,GAChC,GAAI,EAAA,MAAM,QAAQ,iCAChB,GAAM,GAAY,KAAK,OAAO,GACxB,EAAU,KAAK,QAAQ,IAAI,EAAU,QACrC,EAAO,GAAA,KAAK,OACA,gCAA+B,MAAA,EAAA,CAC5B,EAAQ,SAAO,OAAK,GAA0B,KACjD,SAAS,EAAG,GAE9B,YAAK,8BAA8B,GAE5B,EAGT,GAAM,GACF,EAAA,MAAM,QAAQ,eAAiB,IAAa,GAC1C,EACF,EAAyB,GAAwB,GAAS,EACxD,EAAU,EACZ,GAAI,IAAyB,GAC7B,GAAI,IAAmB,GACrB,EAAS,KAAK,gBAChB,EAAS,CAAC,CAAC,MAAO,EAAa,MAAK,EAAE,OAAM,IAAI,WAC9C,EAAU,KAAK,QAAQ,IAAI,EAAO,QAClC,EACF,KAAK,MACA,gDACG,EAAQ,QAAS,EAAQ,SAAS,GAAI,EAAQ,SAAS,IAC1D,SAAS,EAAG,GACrB,YAAK,8BAA8B,GAE5B,GAGH,EAAA,UAAA,KAAN,SAAW,qHACH,GAAkB,KAAK,aACvB,EAA+B,GAEjC,EAAgB,GACpB,AAAI,KAAK,oBAAsB,KAC7B,MAAK,mBAAqB,EAC1B,EAAgB,IAEhB,KAAK,aAAa,KAAK,GAEzB,KAAK,aAAe,EAEpB,IAGM,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,SACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OACpB,EACF,EAAA,KAAK,QAAQ,KAAK,aAAa,IAAI,SAAC,GAAkB,MAAA,GAAE,QACnD,OAAO,SAAA,GAAK,MAAA,IAAK,OAE1B,KAAK,aAAe,EAEpB,AAAI,GACF,MAAK,mBAAqB,MAGtB,EAAuB,CAC3B,aAAc,KAAK,aACnB,eAAgB,KAAK,eACrB,SAAU,KACV,OAAQ,MAGN,EAAA,MAAM,UAAU,iDAAmD,EACpD,CAAA,EAAM,QAAQ,IAAI,IADjC,CAAA,EAAA,UACI,SAAW,EAAA,OAEjB,EAAI,SAAc,EAAA,KAAK,IAAI,GAC3B,EAAI,oBAAyB,WACzB,MAAA,GAAS,IAAI,SAAC,EAAG,GAAM,MAAC,CAAC,KAAM,EAA0B,GAAI,GAAI,KAC5D,IAAI,SAAA,GAAK,MAAG,GAAE,KAAI,KAAK,EAAE,KACzB,KAAK,oBAEd,EAAI,SAAc,CAChB,MAAO,8EAIX,YAAK,aAAe,EACpB,KAAK,eAAiB,EACtB,CAAA,EAAO,SAET,EAAA,UAAA,OAAA,WACE,MAAO,CACL,WAAY,GACZ,cAAe,KAAK,cACpB,uBAAwB,KAAK,eAAe,kBAC5C,kBAAmB,KAAK,eAAe,eAInC,EAAA,UAAA,WAAR,WACE,MAAI,GAAA,MAAM,UAAU,iDAAmD,EAC9D,KAAK,MAAM,aAEb,CAAC,QAAS,EAAA,KAAK,MAAO,MAAO,OAG9B,EAAA,UAAA,SAAR,SAAiB,GACf,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,MAAK,MAAM,WACJ,GAER,GAAwB,MAAQ,EAAA,KAAK,MAC/B,IAGK,EAAA,UAAA,aAAd,SAA2B,4EACzB,MAAI,GAAA,MAAM,UAAU,iDAAmD,EACrE,CAAA,EAAO,KAAK,MAAM,uBAAuB,IAErC,GAAa,EACnB,CAAA,EAAO,EAAW,MAAQ,EAAW,eAKvC,EAAA,UAAA,YAAA,SAAY,GACV,GAAI,KAAK,gBAAgB,IAAI,GAC3B,OAEF,GAAI,KAAK,YAAY,IAAI,IACvB,KAAK,gBAAgB,IAAI,GACzB,KAAK,iBACL,OAGF,GAAI,CAAC,KAAK,QAAQ,IAAI,GACpB,OAGF,KAAK,eAAe,GACb,GAAA,GAAA,KAAA,QAAA,IAAA,GAAA,eACP,AAAI,GAAkB,MACpB,GAAe,KAAK,UACpB,EAAe,KAAK,WAEtB,KAAK,QAAQ,OAAO,IAGd,EAAA,UAAA,eAAR,SAAuB,GACf,GAAA,GAAA,KAAA,QAAA,IAAA,GAAC,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,SAAU,EAAA,EAAA,MAE5C,EAAM,GAAS,EAAM,YAAc,EACnC,EAAW,KAAK,aAAa,IAAI,GACvC,AAAI,EAAW,EACb,KAAK,aAAa,IAAI,EAAK,EAAW,GAEtC,MAAK,aAAa,OAAO,GACzB,AAAI,GAAW,MACb,MAAK,eAAiB,KAAK,aAAa,EAAU,GAClD,KAAK,eAAe,eAAe,EAAS,EAAU,EAAO,KAGjE,GAAM,GAAU,KAAK,QAAQ,IAAI,GACjC,EAAQ,QAAU,KAClB,EAAQ,SAAW,KACnB,EAAQ,SAAW,GACnB,EAAQ,MAAQ,MAGlB,EAAA,UAAA,WAAA,SAAW,GACT,YAAK,YAAY,GACV,KAAK,QAAQ,IAAI,GAAQ,SAOlC,EAAA,UAAA,YAAA,SAAY,GACV,MAAO,MAAK,QAAQ,IAAI,IAGlB,EAAA,UAAA,cAAR,WACE,MAAK,GAAA,MAAM,QAAQ,qBAInB,CAAI,KAAK,YAAc,MACrB,MAAK,WAAa,EAAA,SAAS,YAAY,QAGlC,KAAK,YAPH,MAiBX,EAAA,UAAA,mBAAA,SACI,EACA,GAFJ,GAAA,GAAA,KAEI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAa,KAAK,gBACxB,MAAI,CAAC,KAAK,uBAAyB,GAAc,MAC/C,SAAQ,KACJ,6NAKJ,KAAK,sBAAwB,IAGxB,GAAc,MACjB,EAAO,MACH,SAAA,GAAS,MAAA,GAAK,QAAQ,IAAI,EAAM,QAAQ,SAAW,MAC/C,EAAA,KAAK,cAAc,EAAM,OAAS,KAGhD,EAAA,UAAA,gBAAA,WACE,MAAO,MAAK,OAGd,EAAA,UAAA,QAAA,SAA0B,EAAS,GACjC,GAAM,GAAS,KAAK,WAAW,EAAK,MAAO,aACrC,EAAa,KAAK,QAAQ,IAAI,EAAO,QAI3C,SAAW,eAAiB,CAC1B,KAAM,EAAA,SAAS,KAAK,EAAK,SACzB,KAAM,EAAA,SAAS,KAAK,EAAK,UAGpB,GAET,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAExC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAa,KAAK,QAAQ,IAAI,EAAM,QAC1C,MAAO,GAAW,eAAe,KAAK,SAGxC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAiB,GAC7C,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAO,EAAM,EAAE,MAChE,EAAE,OACN,MAAO,MAAK,WAAW,EAAM,EAAE,MAAO,GAGxC,GAAI,EAAA,KAAK,cAAc,KAAU,EAC/B,MAAO,GAAA,OAAO,GAAI,EAAM,EAAE,OAErB,GAAA,GAAA,KAAA,QAAA,IAAA,EAAA,QAAA,SACD,EAAc,EAAA,WAAW,iBAAiB,EAAE,MAAO,EAAO,GAChE,GAAI,GAAY,CAAC,GACf,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAmB,GACvB,GAAI,IAAa,GACf,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,GAEhD,YAAK,YAAY,EAAE,QACZ,KAAK,aAAa,EAAG,EAAO,IAG7B,EAAA,UAAA,aAAR,SAAqB,EAAW,EAAiB,GAC/C,GAAM,GAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAI,KAAK,WAAW,EAAM,EAAE,OAC5B,EAAa,KAAK,QAAQ,IAAI,EAAE,QAEtC,OAAO,OAAO,EAAY,GAC1B,EAAW,MAAQ,EACnB,EAAW,MAAQ,EAAE,MACrB,GAAI,GAAa,EAAA,WAAW,kBAAkB,EAAO,EAAE,SACvD,AAAI,EAAS,OAGX,IAAc,EAAS,MAAM,YAE/B,EAAW,MAAQ,CACjB,WAAU,EAEV,WAAY,EAAS,OAAS,EAAS,MAAM,YAAc,EAAE,QAI/D,GAAM,GAAW,KAAK,aAAa,IAAI,EAAW,MAAM,aAAe,EACvE,YAAK,aAAa,IAAI,EAAW,MAAM,WAAY,EAAW,GAEvD,GAGT,EAAA,UAAA,aAAA,SACI,EAAM,EAAiB,EAAe,GAD1C,GAAA,GAAA,KAEQ,EAAS,KAAK,mBAChB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,aAAa,EAAG,EAAO,EAAK,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAW,EAAA,WAAW,gBAAgB,EAAO,EAAK,GAExD,GAAI,EAAS,KAAK,SAAA,GAAQ,MAAA,KAAS,IACjC,MAAO,GAAA,OAAO,GAAI,GAGpB,GAAM,GAAU,GAAI,IAAoB,EAAO,EAAS,GACxD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAM,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAqB,EAAE,MAAO,GAClC,GAAI,IAAe,EAAE,MAAO,GAChC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAO,EAAmB,GACxB,GAAI,EAAQ,GAAG,QAAU,aACvB,GAAM,GAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KAChC,EAAQ,EAAQ,IAAI,SAAC,GAAM,MAAA,GAAA,KAAK,KACtC,MAAO,GAAA,QAAQ,KAAK,OAAO,EAAO,GAAO,KAAK,OAAO,EAAO,IAG9D,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAEjB,GAAI,EAAQ,OAAS,EAAA,MAAM,UAAU,iCACnC,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,OAAO,EAAQ,MAAM,EAAG,GAAW,GACnD,EAAY,KAAK,OAAO,EAAQ,MAAM,GAAW,GACvD,MAAO,MAAK,OAAO,CAAC,EAAU,GAAY,GAE5C,GAAI,EAAA,MAAM,QAAQ,gCAAkC,EAAQ,GAAG,KAAO,GACpE,GAAM,GAAU,GAAI,IAAoB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACnE,MAAO,MAAK,cAAc,EAAS,GASrC,GAAM,GACF,EAAA,aAAa,gBAAgB,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAQ,GACtD,EACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,KAAK,GAAI,EAAA,KAAK,cAAc,EAAE,MAAM,MAAM,OAC3D,EAAU,GAAI,IAAc,EAAU,IAAI,SAAA,GAAK,MAAA,GAAE,SACjD,EAAc,KAAK,cAAc,EAAS,GAChD,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAsB,GAAtB,GAAA,GAAA,KACQ,EAAS,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,IAAI,KACtE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAE/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAa,EAAa,EAC1B,GACF,GAAM,GAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAY,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAC9C,EAAA,EAAA,MAAC,EAAA,EAAA,GAIP,GAAK,KAAgB,GAAK,IAAgB,IACtC,EAAY,IACd,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAE1B,AAAI,GACF,GAAI,EAAA,UAAU,EAAG,CAAC,EAAG,EAAG,KAG1B,GAAM,GAAM,IAAgB,EAAI,EAAI,EAAE,KAAK,EAAO,EAAW,GACvD,EAAO,IAAgB,EAAI,EAAI,EAC/B,EAAM,IAAgB,EAAI,EAAE,KAAK,EAAO,EAAG,GAAa,EAC9D,MAAO,MAAK,SAAS,EAAK,GAAK,IAAI,EAAM,IAG3C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,GAC5D,MAAO,MAAK,cAAwB,EAAS,CAAC,EAAG,GAAI,IAGvD,EAAA,UAAA,iBAAA,SACI,MAAC,GAAA,EAAA,EAAG,EAAA,EAAA,EAAG,EAAA,EAAA,WAAY,EAAA,EAAA,WAAY,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE7C,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAc,EAAa,EAAE,MAAM,GAAK,EAAE,MAAM,GAChD,EAAA,EAAA,MAAC,EAAA,EAAA,GAED,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OAE9B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAU,GAAI,IAChB,EAAE,MAAO,CAAC,EAAO,EAAa,GAAc,EAAY,EACxD,EAAS,EAAiB,GACxB,EAAuB,CAAC,EAAG,GACjC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAwB,EAAS,EAAQ,IAGvD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aACd,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OACrD,EAAc,GAAI,IACpB,GAAsC,KAAM,EAAE,MAAO,EAAE,OAErD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAExD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAE/C,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,EAGT,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,EAAE,QAG/C,EAAA,UAAA,6BAAA,SACI,EAAa,EAAgB,EAAc,EAC3C,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAQ,EAAM,EAAO,GACnD,GAAI,IAAW,EAAE,MAAO,EAAQ,EAAM,EAAO,GACjD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SACI,EAAc,EAAsB,EACpC,EAAqB,EAAc,EACnC,GACF,GAAM,GACF,GAAI,IAAe,EAAW,MAAO,EAAa,EAAM,EAAO,GACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAY,EAAa,KAG/D,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAI,EAAE,QAAU,UACd,GAAM,GAAO,KAAK,SAAS,EAAE,QACvB,EAAc,EAAK,IAAI,SAAA,GAAK,MAAA,GAAA,KAAK,aAAa,KAC9C,EAAM,EAAA,OAAO,EAAE,MAAO,EAAE,MAAO,GACrC,MAAO,IAAK,EAAK,GAEnB,GAAM,GAAU,GAAI,IAAY,EAAE,MAAO,GACzC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SACI,EAAM,EAAmC,GAC3C,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAiB,EAAE,MAAO,EAAU,GACxC,GAAI,IAAW,EAAE,MAAO,EAAU,GACtC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAM,EAAmB,GAAlD,GAAA,GAAA,KACQ,EAAS,KAAK,mBAChB,CAAC,EAAG,GAAU,WAAM,MAAA,GAAK,WAAW,OAAO,EAAG,EAAS,KAC3D,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAQ,KAAM,GACzD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAEV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAAW,EAAA,aAAa,YAAY,EAAE,MAAO,EAAY,GACzD,EACF,EAAA,aAAa,YAAY,EAAS,OAAQ,EAAW,QACnD,EACF,EAAA,aAAa,oBAAoB,EAAE,MAAO,EAAY,GACpD,EACF,EAAA,aAAa,oBAAoB,EAAO,EAAW,QACjD,EACF,EAAA,aAAa,aAAa,EAAkB,EAAO,EAAW,QAElE,MAAO,GAAA,UAAU,EAAE,QAAQ,GAAW,GAC1B,QAAQ,GACR,MAAM,EAAkB,IAGtC,EAAA,UAAA,eAAA,SACI,EAAM,EAAsB,GAC9B,EAAA,KAAK,OACD,EAAE,MAAQ,EACV,WAAM,MAAA,yEAGV,GAAM,GAAO,EAAW,OAAO,SAAC,EAAG,GAAM,MAAA,GAAI,IAEvC,EAA4C,CAAC,CAAC,EAAG,IACvD,EAAiB,KAAI,MAArB,EAAyB,GACzB,OAAS,GAAI,EAAI,EAAW,OAAQ,EAAI,EAAE,MAAM,OAAQ,EAAE,EACxD,EAAiB,KAAK,CAAC,EAAG,IAG5B,GAAM,GAAU,EAAE,IAAI,GAEhB,EACF,EAAA,aAAa,YAAY,EAAQ,MAAO,EAAY,EAAM,IAExD,EAAoC,EAAA,aAAa,YACnD,EAAoB,OAAQ,EAAW,OAAQ,IAE7C,EAAe,EAAA,aAAa,oBAC9B,EAAQ,MAAO,EAAY,EAAM,IAE/B,EAAW,EAAA,UACb,EAAQ,QAAQ,GAChB,GACJ,MAAO,GAAA,QAAQ,EAAU,IAGnB,EAAA,UAAA,OAAR,SACI,EAAa,EACb,GACF,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,KAAK,KAAK,EAAS,GAC7B,EAAa,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,QAAO,GACpD,EAAU,GAAI,IAAc,EAAY,GACxC,EAAS,KAAK,cAAwB,EAAS,CAAC,GAAI,GAE1D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,OAAO,EAAQ,EAAY,IAGjC,EAAA,UAAA,UAAR,SACI,EAAa,EACb,GAAA,AAAA,IAAA,QAAA,GAAA,MACF,GAAI,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACrB,AAAI,GAAgB,MAClB,GAAY,EAAa,MAAM,GAC/B,EAAS,EAAa,MAAM,IAE9B,GAAM,GAAa,EAAA,aAAa,yBAAyB,GACnD,EAAa,CACjB,WAAU,EACV,OAAM,EACN,UAAS,EACT,QAAS,KAAK,KAAK,EAAS,IAExB,EACF,GAAI,IAAiB,EAAY,EAAY,GAAgB,MAC3D,EAAS,CAAC,GAChB,AAAI,GAAgB,MAClB,EAAO,KAAK,GAEd,GAAM,GAAS,KAAK,cAAwB,EAAS,EAAQ,SAE7D,MAAI,GAAO,MAAM,KAAO,EACf,EAEF,KAAK,UAAU,EAAG,EAAY,IAG/B,EAAA,UAAA,gBAAR,SACI,EAAW,EAAyB,GAAA,AAAA,IAAA,QAAA,GAAA,MACtC,GAAM,GAAU,GAAgB,KAAO,EAAa,MAAQ,EAAE,MACxD,EAAS,EAAQ,EAAQ,OAAS,GAClC,EAAa,EAAA,aAAa,yBAAyB,GACnD,EAAU,GAAI,IAChB,EAAS,EAAY,EAAY,GAAgB,MAC/C,EAAS,GAAgB,KAAO,CAAC,GAAK,CAAC,EAAG,GAC1C,EAAS,KAAK,cAAsB,EAAS,EAAQ,SAC3D,MAAI,GAAO,OAAS,EAAE,KACb,KAAK,gBAAgB,EAAG,EAAY,GAEtC,GAGT,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,MAAO,GAAa,QAAQ,IAGtD,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAC/D,GAAI,EACF,MAAO,GAGH,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACjB,EAAc,EAAA,WAAc,EAAE,OACpC,MAAO,MAAK,OAAO,EAAK,OAAQ,GAAa,QAAQ,IAGvD,EAAA,UAAA,mBAAA,SACI,EAAM,EAAsB,GAC9B,GAAI,GAAO,EACL,EAAc,EAAA,aAAa,mBAAmB,CAAC,GAAO,EAAE,MAC1D,EAAY,EAChB,AAAI,GAAe,MACjB,GAAY,EAAA,UAAU,EAAG,GACzB,EAAO,EAAA,aAAa,iBAAiB,EAAG,EAAE,MAAM,IAGlD,GAAM,GACF,GAAa,gBAAgB,EAAU,MAAO,EAAM,GAClD,EAAS,EAAA,KAAK,cAAc,CAAC,EAAU,MAAM,KAC7C,EAAM,EAAU,KAAK,GAAI,GACzB,EAAc,EAAA,WAAc,EAAE,OAChC,EACA,KAAK,aACG,EAAK,qBAAsB,EAAY,EAAa,GACvD,QAAQ,GACjB,MAAI,IAAe,MACjB,GACI,EAAA,UAAU,EAAQ,EAAA,aAAa,uBAAuB,KAErD,GAGD,EAAA,UAAA,aAAR,SACI,EAAa,EAAiC,EAC9C,EAAiB,GACnB,GAAM,GAAY,EAAE,MAAM,GACpB,EAAS,EAAE,MAAM,GACjB,EACF,GAAa,8BAA8B,EAAQ,GACjD,EAAY,CAAC,WAAU,EAAE,OAAM,EAAE,UAAS,EAAE,YAAW,GACvD,EAAU,GAAI,IAAiB,EAAW,GAC1C,EACF,KAAK,cAAwB,EAAS,CAAC,EAAG,GAAa,GAE3D,MAAI,GAAO,MAAM,KAAO,EACf,EAET,GAAa,EAAA,MAAM,EAAG,GAAa,KAAK,CAAC,EAAS,IAC3C,KAAK,aAAa,EAAQ,EAAW,EAAY,EAAO,KAGzD,EAAA,UAAA,gBAAR,SAAwB,EAAW,EAAc,GAE/C,GAAM,GAAO,CAAC,GAId,GAHA,EAAA,aAAa,2BACT,MAAQ,EAAW,OAAO,GAAG,cAAgB,EAAW,MAAM,GAAI,EAClE,EAAE,MACF,CAAC,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAQ,GAC7C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,UAAU,EAAK,GAAY,QAAQ,GAEjD,MAAO,MAAK,gBAAgB,EAAG,IAGjC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,GAChB,MAAO,MAAK,gBAAgB,EAAG,EAAM,QAGvC,EAAA,UAAA,OAAA,SAAO,EAAW,EAAc,EAAoB,GAElD,GAAI,IAAS,EAAE,KAAO,EACpB,KAAM,IAAI,OACN,kDAAkD,GAAE,KAAO,GAAC,IAC5D,iBAAgB,IAMtB,OAJM,GAAO,EAAE,MAAM,GACjB,EAAS,EAGJ,EAAI,EAAG,GAAK,KAAK,KAAK,KAAK,KAAK,IAAS,EAAG,KACnD,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,GAAO,GAC5C,EAAc,EAAQ,mBAAmB,GACzC,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,GAAS,EAAO,MAAO,GAC7D,EAAW,UAIb,GAAI,GACF,GAAM,GAAU,GAAI,IAAc,EAAE,MAAO,EAAW,GAChD,EAAa,EACnB,EAAS,KAAK,cAAc,EAAS,CAAC,IACtC,EAAW,UAGb,MAAO,IAGT,EAAA,UAAA,MAAA,SAAM,EAAW,GACf,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA2B,QAE9D,GAAM,GAAU,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAE,OACnE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA+B,QAElE,GAAM,GACF,GAAI,IAAgB,GAAwB,EAAE,MAAO,EAAE,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,KAAA,SAAK,EAAW,GAAhB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,KAAK,EAAG,KAClE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA0B,QAG7D,GAAM,GAAU,GAAI,IAAgB,GAAmB,EAAE,MAAO,EAAE,OAClE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAA6B,QAGhE,GAAM,GAAU,GAAI,IAAgB,GAAsB,EAAE,MAAO,EAAE,OACrE,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,aAAA,SAAa,EAAW,GACtB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAAmC,QAE/C,GAAM,GACF,GAAI,IAAgB,GAA4B,EAAE,MAAO,EAAE,OAC/D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAAW,EAAW,GACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAiC,QAEpE,GAAM,GACF,GAAI,IAAgB,GAA0B,EAAE,MAAO,EAAE,OAC7D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,UAAA,SAAU,EAAW,GACnB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAgC,QAEnE,GAAM,GACF,GAAI,IAAgB,GAAyB,EAAE,MAAO,EAAE,OAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,GAAI,SAG7C,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAW,GACnC,GAAM,GAAU,GAAI,IAAc,EAAU,KAAM,EAAE,MAAO,EAAE,MAC7D,MAAO,MAAK,cACR,EAAS,CAAC,EAAW,EAAG,GAAI,EAAA,WAAW,EAAE,MAAO,EAAE,SAGxD,EAAA,UAAA,MAAA,SAAM,GACJ,EAAA,aAAa,KACT,yEAEJ,GAAM,GAAW,EAAU,WAC3B,MAAO,IAAU,EAAU,MAAO,IAGpC,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAW,GACtC,GAAM,GAAQ,EAAE,WAChB,MAAO,IAAS,EAAO,EAAE,MAAO,EAAE,MAA0B,EAAG,IAGjE,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,QAAA,SAAQ,EAAW,GAAnB,GAAA,GAAA,KACQ,EACF,KAAK,mBAAmB,CAAC,EAAG,GAAI,WAAM,MAAA,GAAK,WAAW,QAAQ,EAAG,KACrE,GAAI,EACF,MAAO,GAGT,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACrD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,EAAA,aAAa,2BAA2B,MAAO,EAAM,EAAE,MACjD,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAU,EAAA,EAAA,GAEX,EAAS,EAAA,KAAK,cAAc,GAC5B,EAAM,EAAE,KAAK,GAAI,GACvB,MAAO,MAAK,OAAO,EAAK,MAAO,EAAI,OAAO,QAAQ,IAGpD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAK,GACL,EAAc,QACpB,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eACR,EAAG,EAAG,GAA6B,GAEzC,GAAM,GAAU,GAAI,IAAgB,EAAI,EAAE,MAAO,EAAE,OACnD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAI,EAAW,GACb,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAG1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,GAErD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAG7C,EAAA,UAAA,cAAR,SAAsB,EAAe,EAAY,GAC/C,GAAM,GAAU,GAAI,IAAqB,EAAE,MAAO,GAClD,MAAO,MAAK,cAAsB,EAAS,CAAC,GAAI,IAG1C,EAAA,UAAA,eAAR,SACI,EAAe,EAAe,EAAY,EAC1C,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GACF,GAAI,IAAsB,EAAI,EAAE,MAAO,EAAE,MAAO,GACpD,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAO7C,EAAA,UAAA,yBAAR,SAAiC,EAAW,EAAW,GAAvD,GAAA,GAAA,KACQ,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAA,4SAAC,EAAA,EAAA,GAAM,EAAA,EAAA,GAcP,EAAU,KAAK,QAAQ,EAAM,GACnC,SAAK,UACL,EAAK,UACE,GAMD,EAAA,UAAA,+BAAR,SACI,EAAuB,GACzB,MAAO,CACL,OAAQ,EAAY,OACpB,MAAO,EAAY,MACnB,MAAO,EAAc,QAIzB,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,EAAQ,SAAW,EACrB,MAAO,GAAQ,GAIjB,GAAI,EAAQ,OAAS,EAAA,MAAM,IAAI,iCAC7B,GAAM,GAAW,KAAK,MAAM,EAAQ,OAAS,GACvC,EAAW,KAAK,KAAK,EAAQ,MAAM,EAAG,IACtC,EAAY,KAAK,KAAK,EAAQ,MAAM,IAC1C,MAAO,MAAK,KAAK,CAAC,EAAU,IAG9B,GAAM,GACF,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAAO,OAAO,SAAC,EAAI,GAAO,MAAA,GAAA,WAAW,EAAI,KAC1D,EAAS,EAAQ,IAAI,SAAA,GAAK,MAAA,GAAE,QAE5B,EAAc,EAAA,MAAM,QAAQ,cAC5B,EAAU,EACZ,GAAI,IAAkB,EAAQ,GAAG,MAAO,GACxC,GAAI,IAAY,EAAQ,GAAG,MAAO,GACtC,MAAO,MAAK,cAAiB,EAAS,EAAS,IAGjD,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAI,EAAE,QAAU,aAAe,EAAE,QAAU,YACzC,MAAO,MAAK,yBAAyB,EAAG,EAAG,IAG7C,GAAM,GAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,GAAI,KAAK,mBAAmB,CAAC,EAAG,KAC9B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAQ,KAAK,QAAQ,IAAI,EAAE,QAC3B,EAAA,GAAA,EAAA,MAAA,EAAA,MAAA,EAAA,OAAA,EAAA,OAAA,GAAC,EAAA,EAAA,GAAW,EAAA,EAAA,GAGlB,MAAO,MAAK,WAAW,EAAU,EAAO,GAE1C,GAAI,EAAA,MAAM,QAAQ,gCAChB,MAAO,MAAK,eAAe,EAAG,EAAG,GAAkB,EAAE,OAEvD,GAAM,GAAU,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OACjE,MAAO,MAAK,cAAsB,EAAS,CAAC,EAAG,GAAI,IAGrD,EAAA,UAAA,IAAA,SAAsB,EAAM,GAC1B,GAAM,GAAc,EAAA,MAAM,QAAQ,gCAC5B,EAAU,EACZ,GAAI,IAAsB,GAAyB,EAAE,MAAO,EAAE,OAC9D,GAAI,IAAgB,GAAkB,EAAE,MAAO,EAAE,OAC/C,EAAQ,EAAA,WAAW,EAAE,MAAO,EAAE,OACpC,MAAO,MAAK,cAAiB,EAAS,CAAC,EAAG,GAAI,IAGhD,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAY,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACnE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAe,EAAE,OAGhD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAE1C,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,SAG1C,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAgB,EAAE,OAGjD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,QAAA,SAA0B,EAAW,GACnC,GAAM,GAAO,EAAA,KAAK,eAAe,CAAC,GAAM,EAAO,OAGzC,EAAW,EAAA,IAAI,EAAQ,GACvB,EACF,EAAA,aAAa,qBAAqB,EAAS,MAAO,GAChD,EAAI,KAAK,SAAS,EAAQ,EAAS,QAAQ,IAC3C,EAAI,KAAK,IAAI,GACb,EAAS,KAAK,IAAI,EAAG,GAAM,QAAQ,GAIzC,MAAO,GAAA,IAAI,EAAG,IAGhB,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GACF,GAAW,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OAClE,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAGtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,KAAK,mBAAmB,CAAC,KAC3B,GAAM,GAAY,GACd,KAAK,QAAQ,IAAI,EAAE,QAAQ,OAAsB,EAAE,OACvD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAE3C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,cAChB,EAAU,GAAI,IAAqB,EAAE,MAAO,IAE5C,EAAU,GAAI,IAAe,EAAE,MAAO,IAEjC,KAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,GAC5B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA2B,EAAE,MAAO,EAAM,OAC9C,GAAI,IAAgB,GAAoB,EAAE,MAAO,EAAM,OAC3D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAqB,EAAE,OAEtD,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,OAAA,SAAyB,EAAO,GAC9B,GAAM,GAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,GAA6B,EAAG,MAAO,EAAE,OAC7C,GAAI,IAAgB,GAAsB,EAAG,MAAO,EAAE,OAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,UAG1C,EAAA,UAAA,KAAA,SAAuB,EAAM,EAAa,GACxC,GAAI,GACJ,AAAI,EAAA,MAAM,QAAQ,mBAChB,EAAU,GAAI,IAAkB,EAAE,OAElC,EAAU,GAAI,IAAY,EAAE,OAE9B,GAAM,GAAc,EAAQ,mBAAmB,EAAK,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,KAAM,IAGhD,EAAA,UAAA,IAAA,SAAsB,GAEpB,GAAI,KAAK,mBAAmB,CAAC,KAAO,EAAE,QAAU,aAC9C,GAAM,GACF,GAAiB,KAAK,QAAQ,IAAI,EAAE,QAAQ,QAChD,MAAO,MAAK,WAAW,EAAE,MAAO,EAAE,MAAO,GAG3C,GAAI,EAAA,MAAM,QAAQ,+BAChB,MAAO,MAAK,cAAc,EAAG,GAAc,EAAE,OAG/C,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,WAAA,SAA6B,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EAAU,GAAI,IAAkB,EAAE,OAClC,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAG9D,MAAO,MAAK,cAAsB,EAAS,IAG7C,EAAA,UAAA,QAAA,SAA0B,GACxB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,SAAA,SAA2B,GACzB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,GACrB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,GACtB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,IAAA,SAAsB,GACpB,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,IAC5C,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,GAAM,GAAU,GAAI,IAAe,EAAE,MAAO,GAAc,IAC1D,MAAO,MAAK,cAAc,EAAS,CAAC,KAG9B,EAAA,UAAA,eAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAGF,GAAM,GAAS,EAAE,MACX,EAAW,KAAK,QAAQ,IAAI,EAAE,QAC9B,EAAkB,EAAS,WAC3B,EAAc,EAAO,GAAK,EAAO,GAAK,EAAO,GAC7C,EAAmB,EAAS,YAC5B,EAAiB,EAAS,aAAe,eACzC,EAAa,GACb,EAAa,GAIb,EACD,KAAgB,GAAK,IAAqB,IAC3C,EAAkB,GAChB,EAAyB,EAAO,GAAK,IAAM,GAAK,CAAC,CAAC,EAAS,SAEjE,GAAI,GAA6B,CAAC,EAAA,MAAM,QAAQ,wBAC5C,CAAC,EAAA,MAAM,QAAQ,iCACf,CAAC,GACH,GAAM,GAAc,EAAiB,EAAO,GAAK,EAAO,GAAK,EAAO,GAC/B,EAAO,GAAK,EAAO,GAAK,EAAO,GAC9D,EAAY,EAAA,QAAQ,EAAG,CAAC,EAAG,EAAa,EAAS,aACjD,EACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAS,KAAK,iBAAiB,CACnC,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAExB,MAAO,GAAA,QAAQ,EAAQ,EAAS,UAWlC,GAAM,GAAc,EAChB,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACrC,EAAO,GAAK,EAAO,GAAM,GAAO,GAAK,GACnC,EAAwB,CAC5B,OAAQ,EAAE,OACV,MAAO,CAAC,EAAG,EAAa,EAAS,YACjC,MAAO,EAAE,OAUL,EAAwB,EAAS,MACvC,EAAS,MAAQ,EAAS,MAAM,QAChC,EAAS,MAAM,EAAS,MAAM,OAAS,KACvC,EAAA,KAAK,OACD,GAAyB,EAAS,MAAO,EAAU,OACnD,WAAM,MAAA,kBAAkB,EAAS,MAAK,OAClC,EAAU,MAAK,gBACvB,GAAM,GACF,EAAA,QAAQ,EAAQ,CAAC,EAAG,EAAS,WAAY,EAAS,cAEhD,EAAgB,KAAK,iBAAiB,CAC1C,EAAG,EACH,EAAG,EACH,WAAU,EACV,WAAU,EACV,KAAI,EACJ,WAAU,EACV,uBAAsB,IAElB,EAAuB,KAAK,QAAQ,IAAI,EAAc,QAC5D,SAAA,KAAK,OACD,EAAqB,SACrB,WAAM,MAAA,gDAEV,EAAS,MAAQ,EAGjB,EAAqB,MAAQ,EAAS,SAC/B,EAAA,SAAS,qBACL,EAAc,OAAQ,EAAS,SAAU,EAAc,QAI5D,EAAA,UAAA,iBAAR,SACI,EAAa,EAAkB,EAC/B,EAAe,EACf,GAQA,GAAA,GAAA,EAAA,YACA,EAAA,EAAA,aACA,EAAA,EAAA,WACA,EAAA,EAAA,SACA,EAAA,EAAA,UACA,EAAA,EAAA,WAGI,EAAiB,IAAe,eAEhC,EAAY,EAAc,EAAe,EACzC,EAAU,EAAY,EACtB,EAAa,CAAC,EAAW,GACzB,EAAa,GACb,EAAa,GAEb,EAAY,EAAE,QAAQ,CAAC,IACvB,EAAQ,EAAO,QAAQ,CAAC,EAAG,EAAW,KAEtC,EACF,GAAI,IAAoB,EAAY,EAAU,MAAO,GACnD,EACF,KAAK,cAAwB,EAAe,CAAC,IAAY,QAAQ,CAC/D,EAAG,EAAW,GAAI,EAAW,KAG7B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAQ,KAC5D,EAAgB,GAAI,IACtB,EAAO,MAAO,CAAC,EAAG,EAAS,EAAS,aAAc,EAClD,EAAY,EAAS,EAAiB,GACpC,EAAuB,CAAC,EAAQ,GACtC,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEd,GAAM,GAAU,KAAK,cAAwB,EAAe,GAE5D,MAAI,GACK,EAAQ,QAAQ,CAAC,EAAG,EAAW,EAAU,EAAS,cAElD,EAAQ,QAAQ,CAAC,EAAG,EAAS,YAAa,EAAW,KAIhE,EAAA,UAAA,YAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAE9C,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAEjD,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAM,MAAM,KAAO,EAC3D,MAAO,MAAK,iBACR,EAAO,EAAQ,EAAU,EAAM,EAAY,GAGjD,GAAM,GAAU,GAAQ,KAClB,EAA4B,GAA0B,KACtD,EACF,EAAa,GAA6B,EAAY,IAAS,KAC7D,EAAU,GAAI,IAChB,EAAU,EAAS,EAAiB,GAClC,EAAuB,CAAC,EAAO,GACrC,MAAI,IACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAEP,KAAK,cAAc,EAAS,IAGrC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAI,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACxD,EAAS,iBAAmB,GAAK,EAAS,gBAAkB,GAC5D,EAAS,eAAiB,GAAK,EAAS,cAAgB,GACvD,GAAS,QAAQ,OAAS,QAC1B,EAAS,QAAQ,OAAS,SAC7B,MAAO,MAAK,eAAe,EAAG,EAAQ,GAExC,GAAI,EAAA,MAAM,QAAQ,sBAAwB,EAAE,MAAM,KAAO,EACvD,MAAO,MAAK,iBAAiB,EAAG,EAAQ,GAE1C,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,qBAAA,SACI,MAAC,GAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,SAAU,EAAA,EAAA,KAAM,EAAA,EAAA,WAAY,EAAA,EAAA,uBAExC,EAA0B,EAAA,MAAM,QAAQ,6BAC1C,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EAC7C,EAAkB,EACpB,GAA6B,EAAY,GACzC,KACE,EAAmB,CAAC,EAAO,GAE3B,EAAU,GAAQ,KAClB,EAA4B,GAA0B,KAC5D,AAAI,GACF,EAAO,KAAK,GAEd,AAAI,GACF,EAAO,KAAK,GAGd,GAAI,GACJ,MAAI,GACF,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,IAGrC,GAAU,GAAI,IACV,EAAU,EAAS,EAAiB,GACjC,KAAK,cAAc,EAAS,KAGrC,EAAA,UAAA,gBAAA,SACI,EAAa,EACb,GACF,GAAI,GACJ,MAAI,GAAA,MAAM,QAAQ,6BACd,EAAS,aAAe,GACxB,EAAS,YAAc,EAAS,aAAe,EACjD,GAAU,GAAI,IAA6B,GACpC,KAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,GAAU,GAAI,IAAuB,GAC9B,KAAK,cAAc,EAAS,CAAC,EAAG,MAGzC,EAAA,UAAA,wBAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAA+B,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,yBAAA,SACI,EAAa,EAAc,GAC7B,GAAM,GAAU,GAAI,IAAgC,GACpD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,OAAA,SAAO,EAAa,EAAkB,GAEpC,GAAM,GAAU,GAAI,IAAc,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,eAAA,SACI,EAAc,EACd,GACF,GAAM,GAAU,GAAI,IAAsB,GAC1C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAI,KAG1C,EAAA,UAAA,gBAAA,SAAgB,EAAa,EAAc,GAEzC,GAAM,GAAU,GAAI,IAAuB,GAC3C,MAAO,MAAK,cAAc,EAAS,CAAC,EAAG,KAGzC,EAAA,UAAA,KAAA,SAAuB,EAAM,GAC3B,MAAO,GAAA,aAAa,WAAW,EAAG,EAAO,OAG3C,EAAA,UAAA,QAAA,SAAQ,EAAW,GAIjB,OAHM,GAAM,EAAE,MAAM,GACd,EAAqB,GAAI,OAAM,EAAE,KAAO,GAC1C,EAAW,EACN,EAAI,EAAG,EAAI,EAAE,KAAM,IAC1B,AAAI,IAAM,GACR,GAAS,KAAc,EAAE,MAAM,IAInC,GAAM,GAAQ,GAAI,OAAM,EAAE,MAAM,KAAK,GAC/B,EAAO,EAAE,MAAM,QACrB,EAAK,GAAQ,EAEb,OADM,GAAM,GAAI,OAAM,GACb,EAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAM,GAAQ,EACd,EAAI,GAAK,KAAK,MAAM,EAAG,EAAO,GAAM,QAAQ,GAE9C,MAAO,IAGT,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GAA2B,GAAI,IAAyB,GAC9D,MAAO,MAAK,cAAc,EAA0B,CAAC,GAAK,EAAE,QAG9D,EAAA,UAAA,UAAA,SAAU,EAAa,GACrB,GAAM,GAAU,GAAI,IAAc,EAAU,MAAO,IACnD,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,kBAAA,SACI,EAAc,EAAa,EAC3B,GACF,GAAM,GAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,KAAK,cAAc,EAA2B,CAAC,IAC7C,EAA2B,GAAI,IAAyB,GACxD,EAAS,KAAK,cAChB,EAA0B,CAAC,EAAI,GAAqB,EAAE,OAC1D,SAAmB,UACZ,GAGT,EAAA,UAAA,eAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IACA,EAAE,MAAO,EAAW,EAAU,GAClC,GAAI,IAAsB,EAAE,MAAO,EAAW,EAAU,GAC5D,MAAO,MAAK,cAAc,EAAS,CAAC,GAAI,YAG1C,EAAA,UAAA,uBAAA,SAAuB,EAAc,EAAa,GAEhD,GAAM,GAAU,GAAI,IAA8B,EAAI,EAAG,GAEzD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,sBAAA,SACI,EAAa,EAAmB,EAChC,GACF,GAAM,GAAU,GAAI,IAChB,EAAE,MAAO,EAAW,EAAU,GAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,8BAAA,SACI,EAAc,EAAa,GAC7B,GAAM,GACF,GAAI,IAAoC,EAAI,EAAG,GACnD,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,YAAA,SACI,EAAkB,EAAqB,EACvC,GACF,GAAM,GAAQ,EAAa,EAAS,EAAA,QAAQ,GACtC,EAAY,EAAM,MAAM,GACxB,EAAc,EAAM,MAAM,GAC1B,EAAU,GAAI,IAAmB,EAAW,EAAa,GACzD,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,CAAC,GAAQ,QAAS,IAGvD,EAAA,UAAA,OAAA,SAAO,EAAmB,EAAe,EAAiB,GAExD,GAAM,GAAU,GAAI,IAAc,EAAQ,KAAM,EAAO,EAAS,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAAI,IAAY,EAAE,MAClC,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,cAAA,SACI,EAAiB,EAAiB,EAClC,EAA4B,EAC5B,GACF,GAAM,GAAU,GAAI,IAChB,EAAM,MAAO,EAAM,MAAO,EAAU,EAAQ,GAChD,MAAO,MAAK,cAAc,EAAS,CAAC,EAAO,EAAO,GAAW,YAG/D,EAAA,UAAA,aAAA,SAAa,EAAa,EAAmB,GAE3C,EAAA,KAAK,OACD,EAAY,EACZ,WACI,MAAA,sDAAsD,IAE9D,GAAM,GAAY,EAAE,MAAM,GACpB,EAAe,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC7D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAC5D,EAAc,IAAe,OAAU,EAAE,MAAM,GAAK,EAAE,MAAM,GAE5D,EAAe,EAAc,EAC7B,EAAc,EAAa,EAC3B,EAAc,EAAc,GAAY,GAExC,EAAe,IAAe,OAChC,CAAC,EAAW,EAAc,EAAa,GACvC,CAAC,EAAW,EAAa,EAAc,GAErC,EAAU,GAAI,IAAoB,EAAa,EAAW,GAChE,MAAO,MAAK,cAAc,EAAS,CAAC,KAGtC,EAAA,UAAA,MAAA,SAAwB,EAAM,EAAsB,GAClD,MAAO,IAAM,EAAG,EAAY,IAG9B,EAAA,UAAA,UAAA,SACI,EAAiB,EAAiB,GAC9B,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,UAAW,EAAA,EAAA,QAAS,EAAA,EAAA,WAG5C,EAAe,CAAC,EAAa,EAAW,GACxC,EAAiB,EAAQ,QAAQ,CAAC,EAAY,IAC9C,EAAW,EAAQ,QAAQ,CAAC,EAAY,IAE9C,GAAI,IAAe,EACjB,MAAO,GAAA,aAAa,cAAc,EAAA,OAAO,IAAK,GAEhD,GAAM,GAAe,EAAA,OAAO,GACtB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAe,KAAM,EAAS,KAAM,EAC3D,GACE,EACF,KAAK,cAAc,EAAS,CAAC,EAAU,EAAgB,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,GACI,GAAA,GAAA,EAAA,aAAA,gBAAA,EAAA,EAAA,GAAC,EAAA,EAAA,UAAW,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,WAGjC,EAAiB,GACjB,EAAU,GAAI,IAChB,EAAY,EAAW,EAAc,KAAM,EAAa,KAAM,EAC9D,CAAC,EAAY,GAAI,GACf,EAAc,KAAK,cACrB,EAAS,CAAC,EAAc,EAAe,IAC3C,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,IAAA,SAAI,GACF,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGzB,EAAA,UAAA,KAAA,SAAK,GACH,GAAM,GAAU,GAChB,MAAO,MAAK,QAAQ,EAAG,IAGjB,EAAA,UAAA,QAAR,SAAgB,EAAa,GAC3B,GAAM,GAAQ,KAAK,QAAQ,IAAI,EAAE,QAE3B,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EACF,GAAI,IAAW,GAAoB,KAAM,EAAE,MAAO,GAChD,EAAS,CACb,KAAK,+BAA+B,EAAG,EAAM,eAAe,MAC5D,KAAK,+BAA+B,EAAG,EAAM,eAAe,OAGxD,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAO,KAAK,cAAsB,EAAa,GAC/C,EAAU,KAAK,QAAQ,EAAM,GAAM,KAAK,EAAE,MAAM,GAAI,EAAE,MAAM,IAClE,SAAK,UACL,EAAK,UACE,GAGT,EAAA,UAAA,SAAA,SAAS,EAAW,GAClB,GAAM,GAAe,EAAQ,MACvB,EAAY,EAAa,EAAa,OAAS,GAE/C,EAAA,EAAA,aAAA,mBAAA,EAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAAW,EAAA,EAAA,GAAW,EAAA,EAAA,GAGpC,EAAiB,EAAQ,QAAQ,CAAC,EAAW,IAC7C,EAAW,EAAE,QAAQ,CAAC,EAAE,KAAO,EAAW,IAC1C,EACF,GAAI,IAAgB,EAAW,EAAS,CAAC,EAAW,IAClD,EAAc,KAAK,cAAc,EAAS,CAAC,EAAU,IAC3D,MAAO,GAAI,QAAQ,IAGrB,EAAA,UAAA,KAAA,SACI,EAAoB,EAAsB,GAG5C,GAFA,EAAQ,GAAS,EAAA,KAAK,WAAW,GAE7B,IAAU,UAEZ,GAAM,GAAS,EAAA,KAAK,kBAAkB,EAAO,EAAA,KAAK,cAAc,IAChE,SAAO,KAAK,GACL,EAAA,SAAS,WAAW,EAAQ,EAAO,EAAO,WAEjD,GAAM,GAAU,GAAI,IAAY,EAAO,GACjC,EAAc,EAAQ,mBAAmB,GAC/C,MAAO,MAAK,cAAc,EAAS,GAAI,EAAO,KAIlD,EAAA,UAAA,SAAA,SAAyB,GACvB,GAAI,EAAE,QAAU,SACd,KAAM,IAAI,OAAM,gDAIhB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAG,EAAE,QAInC,EAAA,UAAA,UAAA,SAA0B,GACxB,MAAO,MAAK,KAAK,EAAE,MAAO,EAAE,QAAU,SAAW,GAAK,EAAG,EAAE,QAG7D,EAAA,UAAA,SAAA,SAAS,EAAe,EAAc,GAEpC,MAAO,GAAA,aAAa,aAAa,EAAO,EAAM,IAGhD,EAAA,UAAA,eAAA,SAAe,EAAiB,EAAiB,GAE/C,GAAM,GAAS,KAAK,MAAM,EAAQ,EAAO,GACzC,YAAK,QAAQ,IAAI,GAAQ,MAAQ,KAC1B,CAAC,OAAM,EAAE,MAAK,EAAE,MAAK,IAGtB,EAAA,UAAA,WAAR,SACI,EAAiB,EAAiB,GAC7B,GAAA,GAAA,KAAA,eAAA,EAAA,EAAA,GAAA,OACP,MAAO,GAAA,SAAS,qBAAqB,EAAQ,EAAO,EAAO,OAGrD,EAAA,UAAA,aAAR,SAAqB,GACnB,GAAM,GAAU,GAAI,IAAc,EAAM,OACxC,MAAO,MAAK,gBAAgB,EAAS,CAAC,GAAQ,EAAM,QAG9C,EAAA,UAAA,WAAR,SAAmB,GACjB,GAAM,GAAU,GAAI,IAAY,EAAM,OAChC,EAA8B,GACpC,MAAO,MAAK,gBACR,EAAS,CAAC,GAAQ,EAAM,MAAO,KAC/B,IAGE,EAAA,UAAA,cAAR,SAAsB,EAAmB,GACvC,GAAM,GAAe,CACnB,GAAuB,EAAM,eAC1B,GAAuB,EAAM,QAE5B,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EAAiB,CACrB,GAAuB,WAAgB,GAAuB,IAG1D,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,KAAK,gBAChB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,QAG1D,EAAA,UAAA,OAAR,SAAe,GACb,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,SAAU,EAAA,EAAA,MAAO,EAAA,EAAA,MAClB,EACF,GAAwB,GACxB,EACJ,AAAI,EACF,EAAU,GAAI,IAA0B,GAExC,EAAU,GAAI,IAAoB,GAEpC,GAAM,GAAgC,GAChC,EAAM,KAAK,gBACb,EAAS,CAAC,CAAC,MAAO,EAAW,MAAK,EAAE,OAAM,IAAI,EAC9C,KAAwB,GAC5B,MAAO,CAAC,MAAK,EAAE,MAAK,EAAE,OAAQ,EAAI,SAGpC,EAAA,UAAA,gBAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAHJ,GAAA,GAAA,KAGI,AAAA,IAAA,QAAA,GAAA,IACF,GAAM,GAAS,KAAK,eAAe,EAAQ,YAAa,GAClD,EAAU,KAAK,QAAQ,IAAI,EAAO,QAIxC,GAHA,AAAI,EAAQ,cACV,GAAQ,SAAW,IAEjB,EAAQ,mBAAqB,GAAuB,OACtD,GAAM,GAAa,GAA0B,EAAQ,aAKrD,EAAQ,SAAW,EAAW,IAAI,SAAA,GAAK,MAAA,GAAI,IAK7C,GAHA,AAAI,EAAQ,aAAe,MACzB,GAAQ,MAAQ,EAAQ,aAEtB,EAAA,KAAK,cAAc,EAAO,SAAW,EAGvC,SAAQ,OACJ,EAAA,KAAK,uBAAuB,EAAO,MAAoB,GACpD,EAGT,GAAM,GAA8B,GAC9B,EAA2B,EAAO,IAAI,SAAA,GAC1C,GAAI,EAAM,QAAU,YAClB,KAAM,IAAI,OACN,mIAKN,GAAI,GAAU,EAAK,QAAQ,IAAI,EAAM,QAErC,GAAI,EAAQ,SAAW,MACrB,GAAI,CAAC,EAAQ,cACT,EAAA,KAAK,cAAc,EAAM,QACrB,EAAA,MAAM,UAAU,6BAMtB,MAAO,CACL,MAAO,EAAM,MACb,QAAS,KACT,UAAW,GACX,cAAe,EAAQ,QAM3B,AAAI,EAAQ,cACV,GAAQ,SAAW,GACnB,EAAQ,MAAQ,EAAM,eAEf,CAAC,CAAC,EAAQ,WAAa,CAAC,CAAC,EAAQ,aAC1C,EAAQ,EAAQ,SAAW,EAAK,aAAa,GAClB,EAAK,WAAW,GAC3C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,gBAE/B,EAAQ,UACR,CAAC,GAAyB,EAAQ,MAAO,EAAM,QAQjD,GAAM,GAAa,EACb,EAAc,EAAM,MAE1B,EAAM,MAAQ,EAAQ,MACtB,EAAQ,EAAK,cAAc,EAAiB,GAC5C,EAAc,KAAK,GACnB,EAAU,EAAK,QAAQ,IAAI,EAAM,QAEjC,EAAW,MAAQ,EAGrB,SAAK,YAAY,EAAM,QAChB,CAAC,MAAO,EAAM,MAAO,QAAO,EAAE,UAAW,MAGlD,KAAK,YAAY,EAAO,QACxB,GAAM,GACW,CAAC,MAAO,EAAO,MAAO,QAAS,EAAS,UAAW,IAC9D,EAAM,GAAyB,EAAS,EAAY,GACpD,EAAS,KAAK,iBAAiB,EAAK,WACxC,MAAO,IACH,EAAK,MAAO,EAAS,EAAY,KAEjC,EAAoB,KAAK,cAAgB,KAC3C,EAgBJ,GAfA,AAAI,GACF,GAAQ,KAAK,cAGf,GACI,KAAK,MAAO,EAAQ,EAAY,EAAY,GAEhD,EAAc,QAAQ,SAAA,GAAQ,MAAA,GAAK,8BAA8B,KAEjE,AAAI,GACF,GAAQ,KAAK,SAAS,GACtB,KAAK,aAAa,KACd,CAAC,KAAM,EAAQ,YAAY,KAAM,MAAO,KAAK,aAAa,MAG5D,CAAC,EAAA,MAAM,QAAQ,wBAA0B,EAAQ,UACjD,IAAkC,IACpC,GAAM,GAAW,KAAK,aAAa,GACnC,YAAK,8BAA8B,GAC5B,EAET,MAAO,IAGT,EAAA,UAAA,cAAA,SACI,EAAuB,EAAsB,EAC7C,EACA,GAAA,AAAA,IAAA,QAAA,GAAA,IACF,EAAc,GAAe,EAAO,GAAG,MACvC,GAAM,GAAU,KAAK,gBACjB,EAAS,EAAQ,EAAa,EAC9B,GACJ,MAAO,GAAA,SAAS,qBACL,EAAQ,OAAQ,EAAQ,MAAO,EAAQ,QAG5C,EAAA,UAAA,iBAAR,SAAyB,EAAa,GAEpC,MAAM,KAAO,MAAK,aAChB,MAAK,YAAY,GAAO,KAEnB,KAAK,YAAY,IAG1B,EAAA,UAAA,kBAAA,WACE,MAAO,MAAK,gBAKd,EAAA,UAAA,QAAA,WAAA,GAAA,GAAA,KACE,GAAI,KAAK,SACP,OAIF,GAAI,CAAC,EAAA,MAAM,QAAQ,YACjB,GAAM,GAAU,OAAO,KAAK,KAAK,aACjC,EAAQ,QAAQ,SAAA,GACd,EAAK,MAAM,cAAc,EAAK,YAAY,GAAK,cAC/C,MAAO,GAAK,YAAY,KAG5B,KAAK,eAAe,UACpB,AAAI,KAAK,QAAU,MACd,MAAQ,oBAAuB,aAC/B,KAAK,iBAAkB,mBAC1B,KAAK,OAAO,SAEZ,KAAK,OAAS,KAEhB,AAAI,KAAK,qBACP,MAAK,MAAM,QAAU,KACrB,KAAK,MAAM,WAEb,KAAK,SAAW,IAGlB,EAAA,UAAA,eAAA,WAAA,GAAA,GAAA,KACE,MAAI,MAAK,qBAAuB,MAC9B,MAAK,oBAAsB,EAAA,KAAK,WAC9B,GAAI,CAAC,EAAA,MAAM,IAAI,iCAGb,GAAM,GAAY,EAAA,MAAM,QAAQ,SAChC,EAAA,MAAM,IAAI,QAAS,IACnB,GAAM,GAAsB,EAAK,IAAI,EAAA,OAAO,OAAO,WAAW,GAG9D,GAFA,EAAA,MAAM,IAAI,QAAS,GAEf,EAAsB,EACxB,MAAO,IAGX,MAAO,OAGJ,KAAK,qBAGd,EAAA,UAAA,QAAA,WACE,MAAO,MAAK,mBAAqB,GAAK,GAAkB,IAGlD,EAAA,UAAA,YAAR,SAAoB,SACZ,EAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAAO,EAAA,EAAA,MAAO,EAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAAO,EAAA,EAAA,SAE7C,GAAI,GAAW,KAEb,OAEF,GAAM,GAAoB,KAAK,cAAgB,KAC3C,EACJ,AAAI,GACF,GAAQ,EAAA,KAAK,OAGf,GAAI,GAAW,EAAQ,SAMvB,GALA,AAAI,GAAY,MACd,GAAW,GAA2C,EAAO,GAC7D,EAAQ,SAAW,GAGjB,GAAU,MACZ,GAAM,GAAY,GAAwB,GAEtC,EAAO,OACP,EAAQ,EAAS,GAAI,EAAS,EAAS,GACrC,EAAc,YAAkB,YAEtC,AAAI,EACF,GAAA,GAAA,EAAA,GAAA,EAAA,IAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAER,EAAU,GAAI,IACV,EAAW,CAAC,EAAQ,GAAQ,IAEhC,EACI,GAAI,IAAoB,EAAW,CAAC,EAAQ,GAAQ,GAG1D,GAAM,GAAuB,KAAK,eAAe,CAAC,EAAQ,GAAQ,GAClE,AAAI,EACF,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEjB,KAAK,QAAQ,IAAI,EAAqB,QAAQ,MAC1C,GAAa,OAEnB,KAAK,MAAM,2BACP,KAAK,WAAW,EAAqB,QAAS,EAAO,EACrD,GAIJ,GAAM,GAAwB,GACxB,EAAsB,KAAK,gBAC7B,EAAS,CAAC,GAAuB,EAAO,KAAM,GAG5C,EAAgB,KAAK,QAAQ,IAAI,EAAoB,QAC3D,EAAQ,QAAU,EAAc,QAChC,EAAQ,SAAW,EAAc,SACjC,EAAQ,SAAW,EAAc,SACjC,EAAQ,MAAQ,EAAc,MAE9B,KAAK,8BAA8B,GACnC,KAAK,QAAQ,OAAO,EAAoB,QAGxC,EAAQ,OAAS,KACjB,AAAI,GACF,MAAK,cAAgB,EAAA,KAAK,MAAQ,QAGpC,GAAM,GAAa,KAAK,eAAe,EAAU,EAAO,EAAO,GAC/D,EAAQ,QAAU,IAId,EAAA,UAAA,qBAAR,SAA6B,EAAgB,GAE3C,GAAM,GAAU,KAAK,QAAQ,IAAI,GAC1B,EAAA,EAAA,MAEP,YAAK,eAAe,GAEpB,AAAI,GAAiB,MACnB,GAAQ,OAAS,GAAoB,EAAe,IAE/C,EAAQ,QAGT,EAAA,UAAA,eAAR,SACI,EAA4B,EAAuB,EACnD,GAEF,GADA,KAAK,eAAiB,KAAK,aAAa,EAAU,GAC9C,CAAC,KAAK,mBACN,KAAK,cAAgB,KAAK,mBAAqB,KAAO,MACxD,GAAM,GAAM,MAAK,cAAgB,KAAO,MAAM,QAAQ,GACtD,KAAK,kBAAoB,GACzB,QAAQ,KACJ,6BAA6B,EAAE,yCAGrC,MAAO,MAAK,eAAe,eAAe,EAAU,EAAS,IAGvD,EAAA,UAAA,aAAR,SAAqB,EAAyB,GAC5C,MAAO,GAAM,GAAK,EAAM,GAAK,EAAA,KAAK,gBAAgB,IAG5C,EAAA,UAAA,mBAAR,SACI,EAAsB,GACxB,GAAI,KAAK,mBAAmB,GAC1B,IACE,MAAO,WACA,GACP,GAAI,EAAA,MAAM,QAAQ,WAChB,KAAM,IAAI,OAAM,yBAItB,MAAO,OAEX,GAriFsC,EAAA,eAuiFtC,YACI,EAAiB,GACnB,GAAI,IAAU,WAAa,IAAU,YACnC,MAAO,GACF,GAAI,IAAU,SAAW,IAAU,QAGxC,OAFM,GAAU,IAAU,QAAW,GAAI,YAAW,EAAE,QACjB,GAAI,YAAW,EAAE,QAC7C,EAAI,EAAG,EAAI,EAAO,OAAQ,EAAE,EACnC,EAAO,GAAK,KAAK,MAAM,EAAE,IAE3B,MAAO,OAEP,MAAM,IAAI,OAAM,iBAAiB,GChwFrC,GAGM,IAAU,QCHhB,cAmCE,EAAA,MAAM,IAAI,2BAA4B,ICnCxC,AAuBA,AAAI,EAAA,YAAY,aACd,EAAA,gBAAgB,QAAS,WAAM,MAAA,IAAI,KAAoB,MAQ5C,IAAQ,CAAC,eAAc,ICzBvB,GAA0B,0BAE1B,GAA2B;;;EAK3B,GAAkC;;;;;cAWf,GAC9B,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EACD,EAAe,EACf,EAAU,GAAI,IAAe,EAAE,MAAO,GAC5C,MAAO,GAAa,gBAAgB,EAAS,CAAC,GAAI,EAAE,oBAepD,EAAmB,EACnB,EAA4C,GAG9C,MAAO,UAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACT,EAAA,EAAC,EAAA,EAAA,EAAG,EAAA,EAAA,EACJ,EAAe,EACf,EAAU,EAAA,MAAM,QAAQ,gCAC1B,GAAI,IACA,EAAiB,EAAE,MAAO,EAAE,MAC5B,CAAC,CAAC,GACN,GAAI,IAAgB,EAAW,EAAE,MAAO,EAAE,OACxC,EAAS,GAAS,EAAE,MACpB,EAAS,EAAa,gBAAgB,EAAS,CAAC,EAAG,GAAI,GAC7D,MAAO,IC3DX,AAsBA,GAAM,IAAQ,GAA2B;;EAInC,GAAe;;;IAIjB,GAAkC;;EAIzB,GAAkB,GAAiB,GAAO,IAE1C,GAA4B,CACvC,WAAY,EAAA,MACZ,YAAa,QACb,WAAY,ICvCd,YAqBI,GACK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QACR,EAAA,EAAA,EAEP,SAAQ,OAAO,EAAE,QAEV,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAE,MAAO,MAAO,EAAE,OAG9C,GAAM,IAA+B,CAC1C,WAAY,EAAA,SACZ,YAAa,QACb,WAAY,ICjCd,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,WAG/C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAsBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MACL,EAAI,EACV,GAAiB,CAAC,EAAI,GAAQ,mBACvB,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAEtB,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,GACjB,EAAyB,GAAI,IAAyB,GAC5D,MAAO,GAAQ,gBAAgB,EAAwB,CAAC,GAAK,EAAE,OAG1D,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,IC3Cd,AAoBA,GAAA,IAAA,WAKE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,YAAwB,GAOtB,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,MACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,MACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;yBAKK,EAAa;wBACd,EAAY;2DACuB,EAAe;;;MAK1E,MAAA,MC5DA,AAoBA,GAAA,IAAA,WAOE,WACI,EAAkB,EAAqB,EACvC,EAA4B,EAC5B,GANJ,KAAA,aAAe,GACf,KAAA,aAAe,GAMb,KAAK,cAAgB,CAAC,IAAK,OAAQ,YACnC,EAAA,aAAa,2BAA2B,EAAQ,GAChD,EAAA,aAAa,2BAA2B,EAAQ,GAEhD,GAAI,GAAgB,YACpB,AAAI,GAAe,MACjB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,UACxB,EAAgB,0BAGlB,GAAI,GAAe,YACnB,AAAI,GAAc,MAChB,GAAA,aAAa,2BAA2B,EAAQ,GAChD,KAAK,cAAc,KAAK,SACxB,EAAe,yBAGjB,KAAK,YAAc,EACnB,KAAK,SAAW;;wBAEI,EAAa;uBACd,EAAY;;;;;;yDAMsB,EAAe;;;;MAMxE,MAAA,MChEA,AAuBO,GAAM,IAIM,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,EAAA,EAAG,EAAA,EAAA,KAAM,EAAA,EAAA,SAAU,EAAA,EAAA,OAAQ,EAAA,EAAA,MAElC,EAAA,KAAK,OACD,EAAK,MAAM,SAAW,EAAS,MAAM,OACrC,WAAM,MAAA,iFAEV,EAAA,KAAK,OACD,GAAU,MAAQ,EAAK,MAAM,SAAW,EAAO,MAAM,OACrD,WAAM,MAAA,+EAEV,EAAA,KAAK,OACD,GAAS,MAAQ,EAAK,MAAM,SAAW,EAAM,MAAM,OACnD,WAAM,MAAA,8EAGL,GAAA,GAAA,EAAA,gBACL,AAAI,GAAmB,MACrB,GAAkB,MAGpB,GAAM,GAAc,CAAC,EAAG,EAAM,GAE1B,EAAc,KAClB,AAAI,GAAU,MACZ,GAAc,EAAO,MACrB,EAAY,KAAK,IAGnB,GAAI,GAAa,KACjB,AAAI,GAAS,MACX,GAAa,EAAM,MACnB,EAAY,KAAK,IAGnB,GAAM,GAAU,EAAA,MAAM,QAAQ,4BAC1B,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACJ,GAAI,IACA,EAAE,MAAO,EAAK,MAAO,EAAS,MAAO,EAAa,EAClD,GACF,EACF,EAAQ,gBAAgB,EAAS,EAAa,EAAY,GAAG,OAEjE,MAAO,IAGI,GAAgC,CAC3C,WAAY,EAAA,eACZ,YAAa,QACb,WAAY,IC/Ed,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAuBA,GAAM,IAAM;;;;eAQN,GAAa;;;;;;;;;;;;;;;;;;EAoBN,GAAgB,GACzB,GAAK,GAAY,IAER,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,ICzDd,AAmBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAItB,GAAM,GAAa,EAAW,GAC9B,KAAK,YAAc,EAEnB,KAAK,SAAW;;;;;yBAKK,EAAU;;uCAEI,EAAU;;;;;;;MASjD,MAAA,MC5CA,AAuBO,GAAM,IAAoC,CAC/C,WAAY,EAAA,cACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QACb,EAAA,EAAA,MACD,EAAe,EAEf,EAAU,GAAI,IAAsB,EAAmB,OACvD,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KChCX,AAoBA,GAAA,IAAA,WAKE,WAAY,GAJZ,KAAA,cAAgB,CAAC,KAKf,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;uDAMmC,EAAK,OAAO,EAAM;;wBAEjD,EAAK,UAAS;;;;;;;;;;;;;;MAgBtC,MAAA,MCrDA,AAoBA,GAAA,IAAA,WAOE,WAAY,GANZ,KAAA,cAAgB,CAAC,KAGjB,KAAA,aAAe,GACf,KAAA,aAAe,GAGb,GAAM,GAAO,KACN,EAAA,EAAA,GAAQ,EAAA,EAAA,GACf,KAAK,YAAc,EACnB,KAAK,SAAW;;;;;;;;;;;;;;;8BAeU,EAAK,OAAO,EAAM;4BACpB,EAAK,UAAS;;;;;;;;;;;;;;;;UAgBhC,EAAK,OAAM;;MAIrB,MAAA,MCnEA,AA0BO,GAAM,IAAiC,CAC5C,WAAY,EAAA,WACZ,YAAa,QACb,WAAY,IAGV,GAEJ,YAAoB,GAKX,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACnB,EAAA,EAAA,OACE,EAAA,EAAA,YAED,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAU,MAAQ,mBAAsB,aAC1C,YAAkB,kBAChB,EAAA,kDAAC,EAAA,EAAA,GAAO,EAAA,EAAA,GAOR,EAA6B,CAAC,EAAQ,GACtC,EAAW,CAAC,EAAQ,EAAO,GAEjC,AAAI,IAAW,IACb,CAAI,IAAuB,MACzB,IAAsB,SAAS,cAAc,UAAU,WAAW,OAGpE,GAAoB,OAAO,MAAQ,EACnC,GAAoB,OAAO,OAAS,EACpC,GAAoB,UAChB,EAA+C,EAAG,EAAG,EAAO,GAChE,EAAS,GAAoB,QAG/B,GAAM,GAAkB,EAAQ,eAAe,EAAU,SAEzD,EAAQ,QAAQ,IAAI,EAAgB,QAAQ,MAAQ,GAAa,OACjE,EAAQ,MAAM,yBACV,EAAQ,WAAW,EAAgB,QAAS,GAChD,GAAM,GAAU,EAAA,MAAM,QAAQ,cAC1B,GAAI,IAAwB,GAC5B,GAAI,IAAkB,GACpB,EAAM,EAAQ,gBAAgB,EAAS,CAAC,GAAkB,SAChE,SAAQ,YAAY,EAAgB,QAC7B,EC/ET,AA0BA,YAA4B,GAI1B,OAFM,GAAS,GAER,EAAO,SAAW,GAAK,EAAO,EAAO,OAAS,GAAG,UAAY,IAClE,GAAM,GACF,EAAO,OAAS,EAAO,EAAO,OAAS,GAAG,QAAU,EAAQ,GAC1D,EAAa,EAAA,aAAa,yBAAyB,GACzD,EAAO,KAAK,CACV,OAAQ,EACR,WAAU,EACV,QAAS,KAAK,KAAK,EAAU,KAIjC,MAAO,eAIL,EAAe,EAAiB,EAChC,GAIF,OAHM,GAAkB,GAAmB,EAAE,OAEzC,EAAS,EACJ,EAAI,EAAG,EAAI,EAAgB,OAAQ,KACpC,GAAA,GAAA,EAAA,GAAC,EAAA,EAAA,OAAQ,EAAA,EAAA,WAAY,EAAA,EAAA,QAErB,EAAU,GAAI,IAChB,CAAC,WAAU,EAAE,OAAM,EAAE,UAAW,EAAE,MAAM,GAAI,QAAO,GAAG,GACpD,EAAiB,EACvB,EAAS,EAAQ,gBAAgB,EAAS,CAAC,GAAS,GAEpD,AAAI,EAAe,SAAW,EAAE,QAC9B,EAAQ,YAAY,EAAe,QAIvC,MAAO,GC/DT,YAwBI,EAAmB,EACnB,GACF,GAAM,GACF,CAAC,GAAY,EAAM,QAAM,OACrB,GAAY,EAAM,QACpB,EAAsB,CAC1B,MAAO,EAAM,MACb,MAAO,EACP,OAAQ,EAAM,QAEV,EACF,CAAC,GAAY,IAAW,OACpB,GAAY,IAEd,EAAU,GAAI,IAAqB,EAAgB,GACnD,EAAgC,GAChC,EAAS,EAAQ,gBACnB,EAAS,CAAC,GAAU,EAAM,MAAO,KACjC,GACJ,MAAO,CAAC,OAAQ,EAAO,OAAQ,MAAO,EAAY,MAAO,EAAO,OC3ClE,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACA,EAAA,EAAA,MACD,EAAe,EAEf,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAS,EAAA,KAAK,uBAAuB,EAAO,GAC5C,EAAS,EAAA,KAAK,cAAc,GAElC,EAAA,KAAK,OACD,IAAU,EACV,WAAM,MAAA,kBAAkB,EAAM,SAAS,EAAM,yBACzC,WAAU,EAAE,MAAK,SAAS,EAAK,qCAC/B,iDAER,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QAC5C,MAAI,GAAS,UAAY,CAAC,GAAc,EAAE,MAAO,IAC7C,CAAE,GAAS,UAAY,MAAQ,GAAc,EAAS,MAAO,IACxD,GAAc,EAAG,EAAQ,GAGlC,GAAa,OAAO,EAAE,QAEf,CAAC,OAAQ,EAAE,OAAQ,MAAO,EAAQ,MAAO,EAAE,QAG7C,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICzDd,YAwBI,EAAe,EAAuB,EACtC,GACF,GAAM,GAAS,EAAA,KAAK,cAAc,GAC5B,EAAQ,EAAA,KAAK,cAAc,EAAE,OAC7B,EAAY,EAAQ,EACpB,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAC,GAAG,MAAO,CAAC,MAAO,CAAC,EAAW,IAAU,QAAO,IAEhE,EAAU,GAAO,EAAe,EAAE,MAAO,MAAO,GAChD,EACF,GAAQ,CAAC,OAAQ,CAAC,EAAG,GAAU,MAAO,CAAC,MAAO,GAAW,QAAO,IAEpE,SAAQ,8BAA8B,GACtC,EAAQ,8BAA8B,GAE/B,ECvCT,AAoBA,GAAA,IAAA,WAME,WAAY,EAAkB,GAL9B,KAAA,cAAgB,CAAC,KAOf,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAEjC,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACxB,GAAM,GAAQ,GAAkB,KAAK,MAC/B,EAAW,GAAkB,GAEnC,KAAK,SAAW;;QAEZ,EAAK;uBACU,EAAQ;;MAI/B,MAAA,MAEA,YAA2B,GACzB,GAAM,GAAO,EAAO,OACpB,GAAI,EAAO,EACT,KAAM,OAAM,sBAAsB,EAAI,yBAKxC,OAHM,GACF,CAAC,UAAW,UAAW,UAAW,UAAW,UAAW,WACtD,EAAiB,GAAI,OAAM,GACxB,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAe,EAAO,IAAM,EAAc,GAE5C,MAAO,GAAe,OCxDxB,AAqBA,GAAA,IAAA,WAQE,WAAY,EAAkB,GAP9B,KAAA,cAAgB,CAAC,KAIjB,KAAA,aAAe,GACf,KAAA,aAAe,GAIb,OADM,GAAwB,GAAI,OAAM,EAAO,QACtC,EAAI,EAAG,EAAI,EAAY,OAAQ,IACtC,EAAY,GAAK,EAAO,EAAO,IAIjC,GAFA,KAAK,YAAc,EACnB,KAAK,KAAO,EAAY,OACpB,KAAK,KAAO,EACd,KAAM,OACF,6BAA6B,KAAK,KAAI,0BAM5C,OAJM,GAAQ,GAAkB,KAAK,MAE/B,EAAc,GAAe,KAAM,KAAK,MACxC,EAAgB,GAAI,OAAM,KAAK,MAC5B,EAAI,EAAG,EAAI,EAAO,OAAQ,IACjC,EAAc,EAAO,IAAM,EAAY,GAEzC,GAAM,GAAY,QAAQ,EAAc,MAAM,IAAI,OAAM,IAClD,EACF,KAAK,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAC3D,EAAO,mBAAmB,EAAc,OAAM,MAAM,EAAS,IAEnE,KAAK,SAAW;;QAEZ,EAAK;;oBAEO,EAAI;WACb,EAAU;sBACC,EAAI;;UAEhB,EAAY,KAAK,KAAO,GAAE;aACvB,EAAY,KAAK,KAAO,GAAE,MAAM,EAAY,KAAK,KAAO,GAAE;sBACjD,EAAI;aACb,EAAU;wBACC,EAAI;;;;;MAO5B,MAAA,MCvEA,YAyBI,EAAe,EAAgB,GACjC,GAAM,GAAU,EAAA,MAAM,QAAQ,+BAC1B,GAAI,IAAuB,EAAE,MAAO,GACpC,GAAI,IAAiB,EAAE,MAAO,GAClC,MAAO,GAAQ,gBAAgB,EAAS,CAAC,GAAI,EAAE,OC7BjD,AA0BO,GAAM,IAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,iBAAkB,EAAA,EAAA,SACnB,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAW,EAAA,KAAK,eAAe,EAAkB,EAAE,OACrD,EAAO,EACL,EAAe,EAAA,aAAa,mBAAmB,EAAM,GACrD,EAAuB,GAAgB,KACvC,EAAqB,EAAa,mBAAmB,CAAC,IAExD,EAAW,EACf,GAAI,GACF,GAAI,GAKF,OAJM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAa,IAErC,GAAM,GACF,GAAiB,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAc,GAE7D,EAAW,EAAa,eAAe,EAAU,EAAE,OACnD,GAAM,GAAe,EAAa,QAAQ,IAAI,EAAS,QACvD,EAAa,OAAS,MAEtB,GAAW,GAAc,EAAG,EAAc,GAG5C,EAAO,EAAA,aAAa,iBAAiB,EAAK,OAAQ,GAGpD,EAAA,aAAa,2BAA2B,MAAO,EAAM,GAC/C,GAAA,GAAA,EAAA,aAAA,0BAAA,EAAA,MAAA,GAAC,EAAA,EAAA,GAAa,EAAA,EAAA,GAGhB,EAAW,EACf,AAAI,GAEF,GAAW,EAAA,aAAa,qBAAqB,EAAa,IAG5D,GAAI,GACJ,GAAI,GACF,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAS,QAC7C,EAAS,EAAS,OAElB,EAAY,GACd,EAAQ,EAAA,KAAK,cAAc,GAAc,EAAU,EAAE,OAEzD,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAQ,EAAU,EAAa,EAAU,GAGjD,MAAI,IACF,EAAa,8BAA8B,GAGtC,IC9FX,YAuBwB,GAKf,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,EACP,GAAiB,EAAG,WACb,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAC3B,EAAY,EAElB,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,EAAK,GACpB,GAAI,EAAS,cAAgB,GAAK,EAAS,eAAiB,GACxD,EAAA,KAAK,YAAY,EAAS,QAAS,EAAS,UAC9C,MAAO,IAAS,CAAC,OAAQ,CAAC,EAAC,GAAG,QAAO,IAEvC,GAAM,GAAiB,GAAI,IAAc,EAAU,MAAO,IAC1D,MAAO,GAAQ,gBAAgB,EAAgB,CAAC,GAAI,EAAE,OAGjD,GAAM,IAA8B,CACzC,WAAY,EAAA,QACZ,YAAa,QACb,WAAY,ICrDd,YAuBgC,GAKvB,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MACjB,EAAA,EAAA,GAAI,EAAA,EAAA,MAAO,EAAA,EAAA,OACZ,EAAI,EACV,GAAiB,CAAC,EAAO,GAAS,mBAC3B,GAAA,GAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,gBAE3B,EAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAmB,EAAK,GACtB,EAAe,GACf,EACF,GAAI,IAAc,EAAU,MAAO,GACjC,EACF,EAAQ,gBAAgB,EAAyB,CAAC,GAAI,EAAE,OAEtD,EAAyB,GAAI,IAAyB,GACtD,EAAS,EAAQ,gBACnB,EAAwB,CAAC,EAAI,GAAmB,EAAE,OACtD,SAAQ,8BAA8B,GAC/B,EAGF,GAAM,IAAsC,CACjD,WAAY,EAAA,gBACZ,YAAa,QACb,WAAY,ICrDd,YAuBI,EAAe,EACf,EACA,GACF,GAAI,GAAU,GAAI,IAAc,EAAU,MAAO,IAC3C,EAAa,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAEzD,EAAU,GAAI,IAAc,EAAU,MAAO,GAAM,GAAM,GACzD,GAAM,GAAc,EAAQ,gBAAgB,EAAS,CAAC,GAAI,WAC1D,MAAO,CAAC,EAAY,GC/BtB,AAwBO,GAAM,IAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACD,EAAA,EAAC,EAAA,EAAA,WAAY,EAAA,EAAA,QAAS,EAAA,EAAA,IAAK,EAAA,EAAA,oBAE3B,EAAe,EAErB,EAAA,KAAK,OACD,EAAE,MAAM,SAAW,EACnB,WAAM,MAAA,uDACF,EAAE,MAAM,OAAM,MACtB,GAAM,GAA8B,CAAC,EAAG,GACxC,EAAA,KAAK,OACD,EAAA,aAAa,+BAA+B,EAAS,GACrD,WAAM,MAAA,4DACF,gBAAe,EAAO,mBAAmB,EAAS,OAE1D,GAAM,GAAW,EAAA,aAAa,kBAC1B,EAAE,MAA2C,EAAY,EACzD,EAAW,GAET,EAAA,GAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAQ,EAAA,EAAA,GAEf,MAAO,CAAC,EAAQ,KCjDpB,AAoBO,GAAM,IAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAG9B,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EAE1B,MAAO,GAAA,aAAa,wBAChB,EAAW,EAAY,EAAkB,EACzC,KC3CR,AAkBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,mBAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,aAIxB,MAAO,CAAC,EAAiB,KC1C7B,AAmBA,GAAM,IAA0B,EAAA,aAAa,wBAGhC,GAA0C,CACrD,WAAY,EAAA,oBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,QAAS,EAAA,EAAA,MAC7B,EAAA,aAAa,KACT,iGAGE,GAAA,GAAA,EAAC,EAAA,EAAA,MAAO,EAAA,EAAA,OACR,EAAA,EAAC,EAAA,EAAA,cAAe,EAAA,EAAA,aAAc,EAAA,EAAA,eAAgB,EAAA,EAAA,aAG9C,EAAa,EAEb,EAAY,EAAW,SAAS,EAAM,QACtC,EAAa,EAAW,SAAS,EAAO,QAExC,EAAmB,EACnB,EAAkB,EAClB,EAAoB,EACpB,EAAkB,EAElB,EAAA,GAAA,EAAA,EAAA,EAAA,EAAA,EAAA,GAAC,EAAA,EAAA,gBAAiB,EAAA,EAAA,eAIxB,MAAO,CAAC,EAAiB,KChD7B,AAqBA,GAAA,IAAA,WAKE,WACI,EAA8C,EAC9C,EACA,GAPJ,KAAA,cAAgB,CAAC,SACjB,KAAA,YAAwB,GAOtB,GAAM,GAAc,EAAW,GACzB,EAAa,EAAW,GACxB,EAAY,KAAK,IAAI,GAAS,QAAQ,GACtC,EAAY,KAAK,IAAI,GAAS,QAAQ,GAC5C,KAAK,YAAc,EAEb,GAAA,GAAA,EAAA,aAAA,eAAA,EAAA,EAAA,GAAC,EAAA,EAAA,GAAS,EAAA,EAAA,GAEV,EAAgB,EAAQ,QAAQ,GAChC,EAAgB,EAAQ,QAAQ,GAElC,EAAc,GAClB,AAAI,MAAO,IAAc,SACvB,EAAc,uBAAuB,EAAU,QAAQ,GAAE,IAEzD,EAAc;2BACO,EAAU,KAAK,KAAI;8CAI1C,KAAK,SAAW;;;;;4CAKwB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;4CACpB,EAAa,OACjD,EAAS,kBAAkB,EAAa,OAAO,EAAS;iDACf,EAAa;iDACb,EAAa;YAClD,EAAW;uCACgB,EAAU,+BACzC,EAAW;;;;;MAOnB,MAAA,MCtEA,AAuBO,GAAM,IAAuC,CAClD,WAAY,EAAA,iBACZ,YAAa,QACb,WAAY,SAAC,MAAC,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,MACD,EAAA,EAAC,EAAA,EAAA,QAAS,EAAA,EAAA,UAAW,EAAA,EAAA,OACrB,EAAe,EAEf,EAAU,GAAI,IACf,EAAmB,MAAO,EAAS,EAAW,GAC7C,EAAS,EAAa,gBAAgB,EAAS,CAAC,GAAQ,EAAM,OACpE,MAAO,KClCX,AAqBA,GAAM,IAAM,GAA0B;;EAIzB,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC9Bd,AAqBA,GAAM,IAAS,gBAEF,GAAmB,GAAgB,IAEnC,GAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IC5Bd,AAqBA,GAAM,IAAqB,4BAEd,GACT,GAAiB,GAAoB,IAE5B,GAAwC,CACnD,WAAY,EAAA,kBACZ,YAAa,QACb,WAAY,IC7Bd,AAqBA,GAAM,IAAM,iBAEC,GAAgB,GAAgB,IAEhC,GAA0B,CACrC,WAAY,EAAA,IACZ,YAAa,QACb,WAAY,IC5Bd,AAwBO,GAAM,IAAgC,CAC3C,WAAY,EAAA,UACZ,YAAa,QACb,WAAY,SAAC,GAQX,OARY,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACpB,EAAA,EAAA,EACA,EAAA,EAAA,KACD,EAAe,EAEf,EAAQ,EAAE,MAAM,OAEhB,EAAqB,GAAI,OAAM,GAC5B,EAAI,EAAG,EAAI,EAAS,OAAQ,IACnC,EAAS,GAAK,EAAE,MAAM,EAAK,IAG7B,GAAI,GACJ,GAAI,EAAa,mBAAmB,CAAC,KACnC,GAAM,GAAW,EAAa,QAAQ,IAAI,EAAE,QACtC,EAAS,EAAS,OAClB,EAAY,GAAa,EAAQ,EAAE,MAAO,EAAE,MAAO,EAAM,GAE/D,EAAM,EAAa,eAAe,EAAU,EAAE,OAC9C,GAAM,GAAU,EAAa,QAAQ,IAAI,EAAI,QAC7C,EAAQ,OAAS,MAEjB,GAAM,GAAc,EAAG,EAAM,GAE/B,MAAO,KCnDX,YAwBI,GAGK,GAAA,GAAA,EAAA,OAAQ,EAAA,EAAA,MAAO,EAAA,EAAA,QACf,EAAA,EAAA,KACA,EAAA,EAAA,EACP,GAAiB,EAAG,UAGpB,QAAQ,KACJ,YACA,8DACJ,GAAM,GAAS,EAAQ,SAAS,EAAE,QAC5B,EAAA,GAAA,EAAA,EAAA,EAAA,MAAA,EAAA,OAAC,EAAA,EAAA,aAAc,EAAA,EAAA,YAAa,EAAA,EAAA,QAElC,MAAO,CACL,EAAQ,eAAe,EAAa,EAAE,MAAO,GAC7C,EAAQ,eAAe,CAAC,EAAQ,QAAS,QAAS,IAI/C,GAAM,IAA6B,CACxC,WAAY,EAAA,OACZ,YAAa,QACb,WAAY,IChDd,AA4CA,GAAM,IAAgC,CACpC,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACA,IAGF,OAA2B,IAAA,EAAA,GAAA,GAAA,GAAA,GAAA,OAAA,MAAtB,GAAM,IAAY,GAAA,IACrB,EAAA,eAAe,+RCxEjB,AAGA,GAAM,IAAU,QCHhB,GAsCa,IAAU,CACrB,YAAa,GAAA,aACb,mBAAoB,GAAA,YACpB,qBAAsB,GAAA,cACtB,YAAa,GAAA,aACb,cAAe,GAAA,eACf,iBAAkB,GAAA,kBAClB,KAAQ,yZC7CV,oBAAM,IAAK,KAEL,GAAgB,EAEtB,YAAyB,GACvB,KAAM,GAAO,CAAE,QAAS,CAAC,EAAY,GAAI,EAAY,GAAI,QAAS,CAAC,EAAG,IAChE,EAAU,GAChB,OAAS,GAAI,EAAG,EAAI,EAAK,QAAQ,OAAQ,KACvC,KAAM,GAAS,EAAK,QAAQ,GACtB,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAW,KAAK,MAAO,GAAY,EAAS,GAAK,GACjD,EAAa,EAAK,QAAQ,GAChC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAQ,EAAG,EAAQ,EAAU,KACpC,KAAM,GAAU,EAAU,GAAQ,IAClC,OAAS,GAAI,EAAG,EAAI,EAAY,IAC9B,EAAQ,KAAK,CAAC,EAAS,MAK/B,MAAO,GAGT,KAAM,IAAa,AAAC,IAClB,EAAI,eAAe,UACnB,EAAI,WAAW,UACf,EAAI,SAAS,WAGT,GAAY,AAAC,GAAoB,EACrC,iBACA,WAAY,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,IAClD,SAAU,GAAG,MAAM,EAAgB,CAAC,EAAG,GAAI,CAAC,GAAI,MAG5C,GAAW,CAAC,EAAK,KACrB,KAAM,GAAS,GAAG,IAAI,EAAI,WAAY,GAChC,EAAO,GAAG,IAAI,EAAI,SAAU,GAC5B,EAAiB,GAAG,SAAS,CAAC,EAAQ,GAAO,GACnD,MAAO,IAAU,IAGnB,YAAsB,EAAY,EAAS,GACzC,KAAM,GAAY,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC9C,EAAU,GAAG,IAAI,EAAW,GAC5B,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAqB,GAAG,IAAI,EAAU,GACtC,EAAoB,GAAG,IAAI,EAAS,GACpC,EAAc,GAAG,IAAI,EAAoB,GACzC,EAAS,GAAG,IAAI,EAAmB,GACnC,EAAO,GAAG,IAAI,EAAmB,GACjC,EAAkB,GAAG,IAAI,EAAQ,GACjC,EAAgB,GAAG,IAAI,EAAM,GAC7B,EAAa,EACnB,MAAO,IAAG,SAAS,CAAC,EAAiB,GAAgB,GAGvD,YAAgC,EAAM,GACpC,MAAO,IAAG,KAAK,KACb,KAAM,GAAM,EAAK,IAAS,EAAK,IAAS,EACxC,MAAO,IAAS,EAAK,GAAa,eAAe,YAIrD,SACE,YAAY,EAAO,GACjB,KAAK,eAAiB,EACtB,KAAK,MAAQ,EAAO,SAAS,UAC7B,KAAK,OAAS,EAAO,SAAS,UAC9B,KAAK,SAAW,EAAO,SAAS,SAChC,KAAK,YAAc,GAAgB,EAAO,SAAS,WACnD,KAAK,QAAU,GAAG,SAAS,KAAK,aAChC,KAAK,UAAY,GAAG,SAAS,CAAC,KAAK,MAAO,KAAK,SAC/C,KAAK,aAAe,EAAO,SAAS,aACpC,KAAK,WAAa,GAClB,KAAK,eAAiB,EAAO,SAAS,oBAGlC,kBAAiB,GAErB,GAAK,CAAC,GAAgB,EAAW,oBAAwB,EAAW,MAAM,SAAW,GAAO,EAAW,MAAM,GAAK,GAAO,EAAW,MAAM,GAAK,EAAI,MAAO,MAC1J,KAAM,CAAC,EAAiB,EAAO,GAAU,GAAG,KAAK,KAC/C,KAAM,GAAe,EAAW,eAAe,CAAC,KAAK,MAAO,KAAK,SAC3D,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAa,IAAI,KAAM,IAAM,GAC7D,EAAoB,KAAK,eAAe,QAAQ,GACtD,GAAI,GAEJ,GAAI,MAAM,QAAQ,IAChB,KAAM,GAAS,EAAkB,KAAK,CAAC,EAAG,IAAM,EAAE,KAAO,EAAE,MACrD,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAY,GAAG,OAAO,CAAC,EAAO,GAAI,EAAO,IAAK,GAC9C,EAAS,GAAG,OAAO,CAAC,EAAW,GAAY,GACjD,EAAa,EAAO,QAAQ,OAE5B,GAAa,EAAkB,UAEjC,KAAM,GAAgB,GAAa,EAAY,KAAK,QAAS,KAAK,WAC5D,EAAS,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC3C,EAAY,GAAG,QAAQ,GAAQ,UACrC,MAAO,CAAC,EAAY,EAAe,KAG/B,EAAmB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBAC/G,EAAa,KAAM,GAAiB,QAC1C,EAAiB,UACjB,GAAI,GAAgB,EAAW,IAAI,AAAC,GAAa,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,MACpF,EAAgB,KAAM,SAAQ,IAAI,EAAc,IAAI,KAAO,KACzD,KAAM,GAAO,KAAM,GAAY,QAC/B,SAAY,UACL,KAET,KAAM,GAAiB,GACvB,OAAS,GAAI,EAAG,EAAI,EAAc,OAAQ,KACxC,KAAM,GAAc,EAAc,GAC5B,EAAe,GAAG,KAAK,KAC3B,KAAM,GAAM,GAAU,GAChB,EAAW,EAAW,GACtB,EAAS,KAAK,YAAY,GAC1B,EAAY,GACf,MAAM,EAAiB,CAAC,EAAU,GAAgB,GAAI,CAAC,EAAG,KAC1D,UACA,QAAQ,CAAC,GAAe,KACrB,EAAc,GAAG,MAAM,EAAQ,CAAC,GAAW,CAAC,IAClD,MAAO,CAAE,MAAK,YAAW,cAAa,YAExC,EAAe,KAAK,GAEtB,SAAM,UACN,EAAO,UACP,EAAgB,UACT,CACL,MAAO,EACP,YAAa,CAAC,EAAW,MAAM,GAAK,KAAK,MAAO,EAAW,MAAM,GAAK,KAAK,cAIzE,eAAc,GAClB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,CAAE,QAAO,eAAgB,KAAM,MAAK,iBAAiB,GAC3D,SAAM,UACC,QAAQ,IAAI,EAAM,IAAI,KAAO,KAClC,KAAM,GAAY,GAAuB,EAAM,GACzC,CAAC,EAAc,EAAS,GAAmB,KAAM,SAAQ,IAAI,CAAC,EAAK,UAAW,EAAW,EAAK,aAAa,IAAI,KAAO,IAAM,EAAE,UAC9H,EAAS,EAAK,OACd,CAAC,EAAc,GAAgB,EAC/B,EAAkB,EACrB,IAAI,AAAC,GAAc,CACjB,GAAS,GAAK,EAAO,IAAM,EAC3B,GAAS,GAAK,EAAO,IAAM,IAE1B,EAAiB,CACrB,QAAS,EAAQ,MAAM,EAAG,GAC1B,YAAa,EAAQ,MAAM,GAC3B,UAAW,EACX,YAAa,GAEf,UAAW,EAAK,KAChB,EAAK,UAAU,UACf,EAAK,YAAY,UACjB,EAAU,UACH,MAKb,kBAAoB,GAClB,KAAM,GAAY,KAAM,IAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC/G,EAAQ,GAAI,IAAe,EAAW,GAC5C,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,eAAiB,GACzB,GAAQ,WAAa,KCpLrB,kBAAQ,iBAAmB,CACzB,WAAY,CACV,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtD,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvD,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,KAEpD,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,KAC7D,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC3D,eAAgB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,GAAI,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,KAC9D,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,eAAgB,CAAC,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,KAC1C,eAAgB,CAAC,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,KACpD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC/C,eAAgB,CAAC,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,eAAgB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACzD,kBAAmB,CAAC,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,KACnD,kBAAmB,CAAC,GAAI,IAAK,GAAI,GAAI,GAAI,IACzC,aAAc,CAAC,IAAK,IAAK,IAAK,IAAK,KACnC,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KAC9C,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,cAAe,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACxD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,KACtD,iBAAkB,CAAC,IAAK,IAAK,IAAK,IAAK,IAAK,KAC5C,YAAa,CAAC,IAAK,IAAK,IAAK,IAAK,KAClC,kBAAmB,CAAC,KACpB,QAAS,CAAC,GACV,WAAY,CAAC,GACb,gBAAiB,CAAC,IAClB,eAAgB,CAAC,KACjB,WAAY,CAAC,KACb,UAAW,CAAC,MAEd,GAAQ,yBAA2B,CACjC,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,GAAI,GAAI,GAAI,GAAI,GAAI,KACrD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KACtD,CAAE,IAAK,YAAa,QAAS,CAAC,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,EAAG,IACtD,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,YAAa,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC9D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,KAC7D,CAAE,IAAK,eAAgB,QAAS,CAAC,GAAI,GAAI,GAAI,GAAI,GAAI,QC/CvD,oBAAM,IAAK,KAEX,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IACxE,MAAO,CAAE,aAAY,YAEvB,GAAQ,oBAAsB,GAC9B,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GACrB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GACvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GACnC,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,WAAa,GACrB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,UAAW,EAAI,WAEhD,GAAQ,YAAc,KClDtB,kBAAQ,gBAAkB,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAKxD,YAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAM3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAC1B,YAAsB,GACpB,MAAO,GAAM,IAAM,KAAK,GAE1B,GAAQ,aAAe,GACvB,YAAgC,EAAG,GACjC,MAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IAEvC,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GACd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAC7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAC9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAChC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,GACtB,YAAiC,EAAG,GAClC,MAAO,MAAK,KAAO,GAAE,GAAK,EAAE,KAAO,EAAO,GAAE,GAAK,EAAE,KAAO,GAE5D,GAAQ,wBAA0B,KCvFlC,eACA,KAAM,IAAK,KACL,GAAW,KACX,GAAY,KACZ,GAAO,KAEP,GAAkB,IAClB,GAA0C,IAC1C,GAAmB,GACnB,GAA0C,CAAC,GAAkB,GAAU,iBAAiB,kBAAqB,IAC7G,GAAwB,EACxB,GAAuB,EACvB,GAA+C,CAAC,GAAuB,IACvE,GAAmB,GAAU,iBAAiB,cAC9C,GAAkB,CAAC,GAAiB,GAAI,GAAiB,GAAiB,OAAS,IACnF,GAAoB,GAAU,iBAAiB,eAC/C,GAAmB,CAAC,GAAkB,GAAI,GAAkB,GAAkB,OAAS,IACvF,GAA0B,EAC1B,GAA0B,EAC1B,GAAkB,GAClB,GAAuB,GAG7B,YAA+B,EAAW,EAAW,EAAQ,GAC3D,OAAS,GAAI,EAAG,EAAI,GAAU,yBAAyB,OAAQ,KAC7D,KAAM,CAAE,MAAK,WAAY,GAAU,yBAAyB,GACtD,EAAkB,GAAU,iBAAiB,GAAG,IAAS,KACzD,EAAuB,GAAQ,KACrC,GAAI,GAAwB,EAAK,SAAS,GACxC,OAAS,GAAI,EAAG,EAAI,EAAQ,OAAQ,KAClC,KAAM,GAAQ,EAAQ,GACtB,EAAU,EAAgB,IAAM,CAC9B,EAAU,GAAO,GAAI,EAAU,GAAO,GACrC,GAAU,GAAO,GAAK,EAAU,EAAgB,IAAI,IAAM,KAOrE,SACE,YAAY,EAAqB,EAAc,EAAW,GAExD,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EACjB,KAAK,UAAY,EAAO,KAAK,UAC7B,KAAK,WAAa,EAAO,KAAK,UAC9B,KAAK,SAAW,EAAO,KAAK,UAC5B,KAAK,YAAc,EAAO,KAAK,cAGjC,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC1E,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAW,CAC7C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,GAAW,CAAC,GAAG,GAAK,YAAY,EAAO,GAAuB,EAAM,KACtG,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAAa,GAC/F,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAW,CACnC,EAAM,GAAK,EAAkB,GAC7B,EAAM,GAAK,EAAkB,GAAI,EAAM,KAI3C,iCAAiC,GAC/B,KAAM,GAAW,EAAU,GAAgB,IAAI,GACzC,EAAY,EAAU,GAAiB,IAAI,GACjD,MAAO,GAAW,EAIpB,UAAU,EAAW,EAAM,EAAqB,EAAqB,EAAO,IAC1E,KAAM,GAAM,GAAS,YAAY,GAAS,WAAW,KAAK,8BAA8B,CAAC,EAAU,GAAsB,EAAU,KAAwB,KAAK,cAC1J,EAAU,GAAS,WAAW,GACpC,GAAI,GAAO,GAAG,MAAM,cAAc,EAAM,CAAC,CACvC,EAAI,WAAW,GAAK,KAAK,WACzB,EAAI,WAAW,GAAK,KAAK,UAAW,EAAI,SAAS,GAAK,KAAK,WAC3D,EAAI,SAAS,GAAK,KAAK,YACrB,CAAC,GAAI,CAAC,KAAK,SAAU,KAAK,WAC9B,MAAI,IACF,GAAO,GAAG,MAAM,cAAc,IAEzB,CAAE,MAAK,UAAS,QAIzB,aAAa,EAAS,EAAQ,EAAY,EAAO,IAC/C,KAAM,GAAe,GACrB,OAAS,GAAI,EAAG,EAAI,GAAsB,KACxC,KAAM,GAAI,EAAQ,EAAI,GAChB,EAAI,EAAQ,EAAI,EAAI,GACpB,EAAI,EAAQ,EAAI,EAAI,GAC1B,EAAa,KAAK,CACf,GACI,EAAK,EAAI,KAAK,SACd,EAAI,KAAK,UAAa,EAAW,GAAK,EAAO,WAAW,GAC5D,EAAI,KAAK,SAAY,EAAW,GAAK,EAAO,WAAW,GAAI,IAGhE,MAAO,CAAE,UAAW,EAAc,KAAM,EAAa,MAAM,KAI7D,sBAAsB,EAAW,EAAY,GAC3C,KAAM,GAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAe,EAAU,GAAU,iBAAiB,GAAG,cAAsB,KAA0B,GACvG,EAAY,GAAe,GAAgB,EAEjD,MAAO,GAAW,IAAI,CAAC,EAAO,KAC5B,GAAI,GAAI,EACR,MAAI,KAAM,EACR,EAAI,EACC,AAAI,IAAM,GACf,GAAI,GAEC,CAAC,EAAM,GAAI,EAAM,GAAI,UAI1B,SAAQ,EAAO,GAGnB,GAFA,KAAK,WAAa,EAAO,SAAS,WAClC,KAAK,SAAW,EAAO,SAAS,SAC5B,KAAK,iCACP,KAAM,CAAE,QAAO,eAAgB,KAAM,MAAK,oBAAoB,iBAAiB,GAC/E,GAAI,EAAM,SAAW,EACnB,YAAK,kBAAoB,GAClB,KAET,KAAM,GAAc,EAAM,IAAI,AAAC,IAC7B,KAAM,GAAgB,CACpB,WAAY,EAAW,IAAI,WAAW,UAAU,YAChD,SAAU,EAAW,IAAI,SAAS,UAAU,aAE9C,EAAW,IAAI,WAAW,UAC1B,EAAW,IAAI,SAAS,UACxB,KAAM,GAAY,GAAS,oBAAoB,EAAe,GACxD,EAAc,GAAS,WAAW,GAClC,EAAY,EAAW,UAAU,YACvC,SAAW,UAAU,UACrB,EAAW,YAAY,UAChB,IAAK,EAAa,eAE3B,KAAK,wBAAwB,GAC7B,KAAK,wBAA0B,MAE/B,MAAK,0BAEP,KAAM,GAAU,GAAG,KAAK,IAAM,KAAK,kBAAkB,IAAI,CAAC,EAAK,KAC7D,GAAI,GAAQ,EAEZ,KAAM,GAA4B,EAAI,UAAU,QAAU,GAC1D,GAAI,CAAC,EAAc,GAAmB,GACtC,AAAI,IAA8B,IAChC,EAAC,EAAc,GAAmB,IAEpC,EAAQ,GAAK,gBAAgB,EAAI,UAAU,GAAe,EAAI,UAAU,IACxE,KAAM,GAAa,GAAS,aAAa,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,WAC/E,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IAC1F,GAAI,GAAe,EACf,EAAiB,GAAK,gBAC1B,AAAI,IAAU,GACZ,GAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,IAEpD,KAAM,GAAS,CAAE,WAAY,EAAI,WAAY,SAAU,EAAI,UACrD,EAAO,GAAS,yBAAyB,EAAQ,EAAc,CAAC,KAAK,WAAY,KAAK,YAAY,IAAI,KAEtG,CAAC,CAAE,EAAM,GAAU,KAAK,aAAa,QAAQ,GAC7C,EAAiB,GAAG,QAAQ,EAAQ,CAAC,GAAI,IAC/C,GAAI,GAAY,EAAe,YAC/B,GAAI,EAAO,KAAK,SACd,KAAM,CAAE,IAAK,EAAY,QAAS,EAAgB,KAAM,GAAgB,KAAK,UAAU,EAAW,EAAM,GAAgB,GAAI,GAAgB,GAAI,IAC1I,CAAE,IAAK,EAAa,QAAS,EAAiB,KAAM,GAAiB,KAAK,UAAU,EAAW,EAAM,GAAiB,GAAI,GAAiB,IAC3I,EAAkB,KAAK,UAAU,QAAQ,GAAG,OAAO,CAAC,EAAa,KACjE,EAAqB,EAAe,WAC1C,EAAe,UACf,KAAM,GAAc,EAAmB,MAAM,EAAG,GAAuB,GACjE,CAAE,UAAW,EAAkB,KAAM,GAAsB,KAAK,aAAa,EAAa,EAAY,EAAgB,IACtH,EAAe,EAAmB,MAAM,GAAuB,GAC/D,CAAE,UAAW,EAAmB,KAAM,GAAuB,KAAK,aAAa,EAAc,EAAa,GAC1G,EAAgC,KAAK,iCAAiC,GAC5E,AAAI,KAAK,IAAI,GAAiC,GAC5C,IAAsB,EAAW,EAAkB,QACnD,GAAsB,EAAW,EAAmB,UAE/C,AAAI,EAAgC,EACzC,GAAsB,EAAW,EAAkB,OAAQ,CAAC,YAAa,cAEzE,GAAsB,EAAW,EAAmB,QAAS,CAAC,YAAa,cAE7E,KAAM,GAAyB,KAAK,sBAAsB,EAAW,EAAmB,QAClF,GAA0B,KAAK,sBAAsB,EAAW,EAAoB,SAC1F,EAAY,EAAU,OAAO,GAAwB,OAAO,IAE9D,KAAM,GAAwB,KAAK,mBAAmB,EAAW,EAAK,EAAO,GAC7E,GAAG,QAAQ,GACX,KAAM,GAAe,GAAS,WAAW,KAAK,8BAA8B,IAC5E,GAAI,EAAO,KAAK,SACd,KAAM,GAAoB,GAAG,SAAS,GACtC,KAAK,kBAAkB,GAAK,IAAK,EAAc,UAAW,EAAkB,aAC5E,KAAM,GAAa,CACjB,OAAQ,EACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,GAET,KAAM,GAAa,CACjB,OAAQ,KACR,IAAK,EACL,WAAY,EAAK,UACjB,MAAO,GAET,MAAO,MAET,MAAO,GAIT,wBAAwB,GACtB,OAAS,GAAI,EAAG,EAAI,EAAM,OAAQ,KAChC,KAAM,GAAM,EAAM,GACZ,EAAc,KAAK,kBAAkB,GAC3C,GAAI,GAAM,EACV,GAAI,GAAe,EAAY,YAC7B,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,AAAI,EAAM,IACR,MAAK,kBAAkB,GAAK,GAGhC,KAAK,kBAAoB,KAAK,kBAAkB,MAAM,EAAG,EAAM,QAGjE,sBAAsB,GACpB,AAAI,KAAK,kBAAkB,IAAU,MACnC,MAAK,kBAAoB,CACvB,GAAG,KAAK,kBAAkB,MAAM,EAAG,GACnC,GAAG,KAAK,kBAAkB,MAAM,EAAQ,KAK9C,gCACE,KAAM,GAAY,KAAK,kBAAkB,OACnC,EAAS,IAAc,EAC7B,MAAI,MAAK,WAAa,GAAK,EAClB,EAEF,IAAc,KAAK,UAAY,KAAK,yBAA2B,KAAK,WAG7E,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,aAGzB,GAAQ,SAAW,KC3RnB,kBAAQ,UAAY,CAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,iBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,iBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,iBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,gBAAkB,kBACnB,CAAC,cAAgB,kBACjB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,gBAAkB,kBACnB,CAAC,eAAiB,kBAClB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,kBACpB,CAAC,iBAAmB,qBCpdtB,0CAAO,IAAQ,CACb,IAAK,GAAI,IAAK,GAAI,EAAG,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,EAC1E,IAAK,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,GACxE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IACpE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACtE,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACvE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,GACxE,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IACrE,GAAI,GAAI,GAAI,GAAI,GAAI,EAAG,EAAG,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,EAAG,IACrE,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IACxE,EAAG,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,EAAG,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,IAAK,IAAK,GAAI,EACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,EAC1E,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACzE,GAAI,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GACtE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,GACxE,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,GAAI,GAAI,EAAG,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,GAAI,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,IACvE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,EAAG,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACzE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GACxE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,EAAG,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,EAAG,IAAK,GAAI,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,GACvE,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,GAAI,GAAI,GAAI,EAAG,GAAI,EAAG,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACxE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IACpE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,GAAI,IAAK,EAAG,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACxE,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,GAAI,GAAI,EAAG,IAAK,GACrE,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACvE,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IACvE,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GACxE,GAAI,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,GACzE,GAAI,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,GACvE,IAAK,GAAI,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,GAAI,GAAI,GAAI,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GACrE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GACtE,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GACrE,IAAK,EAAG,IAAK,GAAI,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,EAC1E,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,GAAI,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,GAAI,GACvE,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GACzE,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,EACvE,IAAK,IAAK,GAAI,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,GAAI,EAC1E,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,GAAI,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IACrE,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IACpE,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,GAAI,GAAI,IAAK,IACtE,IAAK,IAAK,GAAI,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,GAAI,IAAK,IAAK,IAAK,GAAI,IAAK,EAAG,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,EAAG,GAAI,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IACrE,IAAK,GAAI,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACpE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,IAAK,IACxE,IAAK,IAAK,IAAK,GAAI,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACrE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,EAAG,IAAK,EAAG,IAAK,IAAK,IAAK,IAAK,IAAK,IACvE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,GAAI,EAAG,IAAK,IAAK,IAAK,IAAK,IACxE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IACtE,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,IAAK,OCxKnE,oBAAM,IAAK,KACL,GAAY,KACZ,GAAY,KACZ,GAAO,KACP,GAAY,KACZ,GAAgB,KAA2B,QAEjD,SACE,YAAY,EAAW,EAAgB,EAAW,GAChD,KAAK,SAAW,GAAI,IAAK,SAAS,EAAW,EAAgB,EAAW,GACxE,AAAI,GAAQ,MAAK,OAAS,QAGtB,eAAc,EAAO,GACzB,AAAI,GAAQ,MAAK,OAAS,GAC1B,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QAAS,GAAQ,GAAG,QAAQ,WAAW,IAC1D,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,QAAQ,EAAO,GACvD,GAAG,QAAQ,GACX,KAAM,GAAU,GAChB,SAAW,KAAe,IAAe,IACvC,KAAM,GAAa,EAAW,WAAW,YACzC,GAAI,GAAc,KAAK,OAAO,SAAS,eACrC,KAAM,GAAO,EAAW,OAAS,EAAW,OAAO,YAAc,KAC3D,EAAc,GACpB,GAAI,GAAQ,EAAK,OAAS,EACxB,SAAW,KAAO,IAAU,iBAC1B,AAAI,MAAK,OAAO,KAAK,SAAW,EAAI,SAAS,UAAY,KACvD,GAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAK,KAI7E,EAAQ,KAAK,CACX,WAAY,GAAc,EAC1B,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,GAAI,EAAW,IAAI,SAAS,GAAK,EAAW,IAAI,WAAW,IAAM,EAC3M,OACA,cACA,MAAO,EAAW,MAAQ,GAAG,MAAM,EAAW,OAAS,OAG3D,EAAW,WAAW,UACtB,EAAW,MAAM,UAEnB,MAAO,IAIX,kBAAoB,GAClB,KAAM,GAAS,KAAM,SAAQ,IAAI,CAC/B,GAAU,KAAK,GACf,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,eACrF,GAAG,eAAe,EAAO,KAAK,UAAW,CAAE,UAAW,EAAO,KAAK,UAAU,SAAS,iBAEjF,EAAW,GAAI,IAAkB,EAAO,GAAI,EAAO,GAAI,EAAO,GAAI,GACxE,MAAO,GAGT,GAAQ,KAAO,GACf,GAAQ,kBAAoB,GAC5B,GAAQ,UAAY,GACpB,GAAQ,cAAgB,KC9DxB,oBAAM,IAAK,KAEL,GAAS,GACf,GAAI,IAAO,CAAE,IAAK,EAAG,OAAQ,IACzB,GAAQ,EAEZ,kBAAwB,EAAO,GAC7B,KAAM,GAAS,GAAG,QAAQ,WAAW,GAC/B,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,GAGT,kBAAuB,GACrB,MAAK,IAAO,KAAK,IAAO,IAAM,KAAM,IAAG,eAAe,EAAO,KAAK,IAAI,YAC/D,GAAO,IAGhB,kBAA0B,GACxB,MAAK,IAAO,QAAQ,IAAO,OAAS,KAAM,IAAG,eAAe,EAAO,KAAK,OAAO,YACxE,GAAO,OAGhB,kBAAuB,EAAO,GAM5B,GALA,AAAI,GAAQ,EAAO,KAAK,IAAI,WAC1B,GAAQ,EAER,IAAS,EAEP,KAAU,EAAG,MAAO,IACxB,GAAI,GACJ,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,IAAI,UAAW,EAAO,KAAK,IAAI,WAAY,IACtG,EAAU,GAAG,IAAI,EAAQ,CAAC,MAC1B,GAAG,QAAQ,OAEX,GAAU,KAAM,IAAS,EAAO,EAAO,KAAK,IAAI,WAElD,KAAM,GAAM,GACZ,GAAI,EAAO,KAAK,IAAI,SAClB,KAAM,GAAO,KAAM,IAAO,IAAI,QAAQ,GAChC,EAAO,KAAM,GAAK,OACxB,EAAI,IAAM,KAAK,MAAM,GAAK,EAAK,IAAM,GACrC,GAAG,QAAQ,GAEb,GAAI,EAAO,KAAK,OAAO,SACrB,KAAM,GAAU,KAAM,IAAO,OAAO,QAAQ,GACtC,EAAO,KAAM,GAAQ,OACrB,EAAa,KAAK,MAAM,KAAK,IAAI,IAAM,IAAO,GAAK,GAAK,MAAS,IACvE,AAAI,EAAa,EAAO,KAAK,OAAO,eAClC,GAAI,OAAS,EAAK,IAAM,GAAM,SAAW,OACzC,EAAI,WAAa,GAEnB,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,QAAU,GAClB,GAAQ,WAAa,KC9DrB,oBAAM,IAAK,KAEL,GAAc,CAAC,QAAS,UAAW,OAAQ,QAAS,MAAO,UAAW,WACtE,GAAS,GACf,GAAI,IAAO,GACP,GAAQ,EACZ,KAAM,IAAa,IAEnB,YAAkB,EAAO,GACvB,KAAM,GAAS,GAAG,KAAK,KACrB,KAAM,GAAS,GAAG,QAAQ,WAAW,EAAO,GACtC,EAAS,GAAG,MAAM,eAAe,EAAQ,CAAC,EAAM,IAChD,EAAS,GAAG,KAAK,GAAG,WAAW,EAAQ,GAAI,WACjD,MAAO,KAET,MAAO,GAGT,kBAAoB,GAClB,MAAK,IAAO,SAAS,IAAO,QAAU,KAAM,IAAG,eAAe,EAAO,KAAK,QAAQ,YAC3E,GAAO,QAGhB,kBAAuB,EAAO,GAE5B,GADA,IAAS,EACL,IAAS,EAAO,KAAK,QAAQ,WAC/B,UAAQ,EACD,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,YAAiB,IAAG,QACtB,KAAM,GAAS,GAAG,MAAM,eAAe,EAAO,CAAC,EAAO,KAAK,QAAQ,UAAW,EAAO,KAAK,QAAQ,WAAY,IACxG,CAAC,EAAG,EAAG,GAAK,GAAG,MAAM,EAAQ,EAAG,GACtC,GAAI,EAAO,KAAK,QAAQ,cAEtB,KAAM,GAAK,GAAG,IAAI,EAAG,CAAC,QAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAK,GAAG,IAAI,EAAG,CAAC,OAChB,EAAY,GAAG,KAAK,CAAC,EAAI,EAAI,IACnC,MAAO,GAET,MAAO,GAET,MAAO,IAAS,EAAO,EAAO,KAAK,QAAQ,aAEvC,EAAM,GACZ,GAAI,EAAO,KAAK,QAAQ,SACtB,KAAM,GAAW,KAAM,IAAO,QAAQ,QAAQ,GACxC,EAAO,KAAM,GAAS,OAC5B,OAAS,GAAI,EAAG,EAAI,EAAK,OAAQ,IAC/B,AAAI,GAAa,EAAK,GAAK,EAAO,KAAK,QAAQ,eAAe,EAAI,KAAK,CAAE,MAAO,KAAK,IAAI,IAAM,KAAK,MAAM,IAAM,GAAa,EAAK,IAAM,KAAM,QAAS,GAAY,KAErK,EAAI,KAAK,CAAC,EAAG,IAAM,EAAE,MAAQ,EAAE,OAC/B,GAAG,QAAQ,GAEb,UAAG,QAAQ,GACX,GAAO,EACA,EAGT,GAAQ,QAAU,GAClB,GAAQ,KAAO,KC7Df,oBAAM,IAAK,KAEX,SACE,YAAY,EAAO,GACjB,KAAK,MAAQ,EACb,KAAK,aAAe,EACpB,KAAM,GAAa,KAAK,MAAM,OAAO,GAAG,MACxC,GAAG,KAAK,OAAQ,EAAW,KAAO,IAAQ,EAAW,KAAO,GAAK,IAAM,gBAAgB,EAAW,OAAO,EAAW,mCAgBtH,QAAQ,GACN,MAAO,IAAG,KAAK,KACb,KAAM,GAAU,KAAK,gBAAgB,EAAM,WACrC,EAAU,EAAQ,WAAW,GAC7B,EAAU,KAAK,MAAM,QAAQ,GAC7B,EAAY,EAAQ,IAAI,AAAC,GAAM,EAAE,QAAQ,CAAC,KAC1C,EAAe,KAAK,kBAAkB,GAC5C,MAAO,CACL,cAAe,EAAa,QAAQ,UACpC,QAAS,EAAa,QACtB,gBAAiB,EAAa,gBAC9B,gBAAiB,EAAa,mBAQpC,UACE,KAAK,MAAM,WAGf,GAAQ,UAAY,KC9CpB,oBAAM,IAAK,KACL,GAAY,KAElB,gBAAwB,IAAU,UAEhC,gBAAgB,GAEd,MAAO,IAAG,KAAK,IAAM,GAAG,IAAI,EAAO,OAAO,IAAI,IAIhD,kBAAkB,GAChB,KAAM,CAAC,EAAS,EAAS,EAAiB,GAAmB,EAC7D,MAAO,CAAE,UAAS,UAAS,kBAAiB,oBAGhD,GAAQ,UAAY,KChBpB,eACA,YAAc,GACZ,MAAO,MAAK,MAAM,EAAI,GAExB,SACE,YAAY,EAAS,GACnB,KAAK,cAAgB,GAAI,OAAM,GAC/B,KAAK,iBAAmB,GACxB,KAAK,gBAAkB,EAGzB,QAAQ,GACN,KAAK,cAAc,EAAE,KAAK,kBAAoB,EAC9C,KAAK,KAAK,KAAK,kBAGjB,UACE,KAAM,GAAM,KAAK,cAAc,GAC/B,YAAK,SAAS,EAAG,KAAK,oBACtB,KAAK,KAAK,GACV,KAAK,cAAc,KAAK,iBAAmB,GAAK,KACzC,EAGT,QACE,MAAO,MAAK,mBAAqB,GAGnC,OACE,MAAO,MAAK,iBAAmB,EAGjC,MACE,MAAO,MAAK,cAAc,MAAM,EAAG,KAAK,iBAAmB,GAG7D,MACE,MAAO,MAAK,cAAc,GAG5B,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,KAAK,GAAK,GAAI,IACjC,KAAK,SAAS,EAAG,GAAK,IACtB,EAAI,GAAK,GAIb,KAAK,GACH,KAAO,EAAI,GAAK,KAAK,mBACnB,GAAI,GAAI,EAAI,EAEZ,GADA,AAAI,EAAI,KAAK,kBAAoB,KAAK,KAAK,EAAG,EAAI,IAAI,IAClD,CAAC,KAAK,KAAK,EAAG,GAAI,MACtB,KAAK,SAAS,EAAG,GACjB,EAAI,GAIR,WAAW,GACT,MAAO,MAAK,gBAAgB,KAAK,cAAc,IAGjD,KAAK,EAAG,GACN,MAAO,MAAK,WAAW,GAAK,KAAK,WAAW,GAG9C,SAAS,EAAG,GACV,KAAM,GAAI,KAAK,cAAc,GAC7B,KAAK,cAAc,GAAK,KAAK,cAAc,GAC3C,KAAK,cAAc,GAAK,GAG5B,GAAQ,QAAU,KCvElB,oBAAM,IAAW,KAEjB,YAAqC,EAAY,EAAO,EAAU,EAAU,EAAoB,GAC9F,KAAM,CAAC,EAAQ,GAAS,EAAO,MAC/B,GAAI,GAAe,GACnB,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,GAC7C,KAAM,GAAS,KAAK,IAAI,EAAW,EAAoB,GACjD,EAAO,KAAK,IAAI,EAAW,EAAqB,EAAG,GACzD,OAAS,GAAW,EAAQ,EAAW,EAAM,EAAE,EAC7C,GAAI,EAAO,IAAI,EAAU,EAAU,GAAc,GAC/C,EAAe,GACf,MAGJ,GAAI,CAAC,EACH,MAGJ,MAAO,GAOT,YAAiC,EAAgB,EAAoB,GACnE,KAAM,CAAC,EAAQ,EAAO,GAAgB,EAAO,MACvC,EAAQ,GAAI,IAAS,QAAQ,EAAS,EAAQ,EAAc,CAAC,CAAE,WAAY,GACjF,OAAS,GAAW,EAAG,EAAW,EAAQ,EAAE,EAC1C,OAAS,GAAW,EAAG,EAAW,EAAO,EAAE,EACzC,OAAS,GAAa,EAAG,EAAa,EAAc,EAAE,GACpD,KAAM,GAAQ,EAAO,IAAI,EAAU,EAAU,GAE7C,GAAI,EAAQ,EAAgB,SAE5B,AAAI,GAA4B,EAAY,EAAO,EAAU,EAAU,EAAoB,IACzF,EAAM,QAAQ,CAAE,QAAO,KAAM,CAAE,WAAU,WAAU,GAAI,KAK/D,MAAO,GAET,GAAQ,wBAA0B,KC7ClC,kBAAQ,UAAY,CAClB,OAAQ,UAAW,WAAY,UAAW,WAAY,eACtD,gBAAiB,YAAa,aAAc,YAAa,aACzD,UAAW,WAAY,WAAY,YAAa,YAAa,cAE/D,GAAQ,cAAgB,GAAQ,UAAU,OAC1C,GAAQ,QAAU,GAAQ,UAAU,OAAO,CAAC,EAAQ,EAAW,IAC7D,GAAO,GAAa,EACb,GACN,IACH,KAAM,IAAqB,CACzB,CAAC,UAAW,gBAAiB,CAAC,YAAa,gBAC3C,CAAC,YAAa,aAAc,CAAC,UAAW,YACxC,CAAC,WAAY,aAAc,CAAC,WAAY,iBACxC,CAAC,aAAc,iBAAkB,CAAC,aAAc,cAChD,CAAC,WAAY,aAAc,CAAC,YAAa,cACzC,CAAC,eAAgB,iBAAkB,CAAC,UAAW,aAQjD,GAAQ,UAAY,CAClB,CAAC,OAAQ,WAAY,CAAC,UAAW,WAAY,CAAC,OAAQ,YACtD,CAAC,WAAY,YAAa,CAAC,OAAQ,gBACnC,CAAC,eAAgB,aAAc,CAAC,YAAa,aAC7C,CAAC,eAAgB,WAAY,CAAC,UAAW,YACzC,CAAC,WAAY,aAAc,CAAC,OAAQ,iBACpC,CAAC,gBAAiB,cAAe,CAAC,aAAc,cAChD,CAAC,gBAAiB,YAAa,CAAC,WAAY,aAC5C,CAAC,YAAa,eAEhB,GAAQ,qBAAuB,GAAmB,IAAI,CAAC,CAAC,EAAY,KAAiB,CAAC,GAAQ,QAAQ,GAAa,GAAQ,QAAQ,KACnI,GAAQ,aAAe,CACrB,YACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,YACA,cACA,aACA,wBACA,uBACA,uBACA,uBACA,uBACA,sBACA,sBACA,aACA,wBACA,eC3DF,oBAAM,IAAM,KAEZ,YAAwB,EAAG,EAAG,EAAU,GACtC,MAAO,CACL,EAAG,EAAQ,IAAI,EAAG,EAAG,GACrB,EAAG,EAAQ,IAAI,EAAG,EAAG,EAAW,GAAI,gBAGxC,GAAQ,eAAiB,GAEzB,YAAwB,EAAM,EAAc,GAC1C,KAAM,CAAE,WAAU,WAAU,GAAI,GAAa,EACvC,CAAE,IAAG,KAAM,GAAe,EAAU,EAAU,EAAU,GAC9D,MAAO,CACL,EAAG,EAAK,SAAW,EAAe,EAClC,EAAG,EAAK,SAAW,EAAe,GAGtC,GAAQ,eAAiB,GAEzB,YAAmB,EAAS,GAC1B,KAAM,GAAS,GAAI,OAAM,GACzB,OAAS,GAAI,EAAG,EAAI,EAAM,IACxB,EAAO,GAAK,EAEd,MAAO,GAET,GAAQ,UAAY,GAEpB,YAAe,EAAG,EAAK,GACrB,MAAI,GAAI,EAAY,EAChB,EAAI,EAAY,EACb,EAET,GAAQ,MAAQ,GAEhB,YAAyB,EAAI,EAAI,EAAI,GACnC,KAAM,GAAK,EAAK,EACV,EAAK,EAAK,EAChB,MAAO,GAAK,EAAK,EAAK,EAExB,GAAQ,gBAAkB,GAE1B,YAAoB,EAAG,GACrB,MAAO,CAAE,EAAG,EAAE,EAAI,EAAE,EAAG,EAAG,EAAE,EAAI,EAAE,GAEpC,GAAQ,WAAa,GAErB,YAAqB,EAAG,EAAK,GAC3B,MAAO,CAAE,EAAG,GAAM,EAAE,EAAG,EAAK,GAAM,EAAG,GAAM,EAAE,EAAG,EAAK,IAEvD,GAAQ,YAAc,KCnDtB,oBAAM,IAAY,KACZ,GAAU,KAEV,GAAuB,GAAU,UAAU,IAAI,CAAC,CAAC,EAAgB,KAAoB,CAAC,GAAU,QAAQ,GAAiB,GAAU,QAAQ,KAC3I,GAAqB,GAAqB,IAAI,CAAC,CAAC,CAAE,KAAkB,GACpE,GAAqB,GAAqB,IAAI,CAAC,CAAC,KAAmB,GACzE,YAAyB,EAAQ,EAAO,GACtC,KAAM,GAAW,EAAc,MAAM,GAAK,EAC1C,MAAO,CACL,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,GACvC,EAAG,EAAc,IAAI,EAAM,EAAG,EAAM,EAAG,EAAW,IAGtD,YAAkC,EAAO,EAAc,EAAQ,GAC7D,MAAO,CACL,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAS,GACjE,EAAG,GAAQ,MAAM,KAAK,MAAM,EAAM,EAAI,GAAe,EAAG,EAAQ,IAUpE,YAAkC,EAAQ,EAAgB,EAAkB,EAAc,EAAS,EAAc,EAAe,EAAmB,GACjJ,KAAM,CAAC,EAAQ,GAAS,EAAa,MAE/B,EAAwB,GAAyB,EAAe,SAAU,EAAc,EAAQ,GAChG,EAAe,GAAgB,EAAQ,EAAuB,GAC9D,EAAiB,GAAQ,WAAW,EAAe,SAAU,GACnE,GAAI,GAAiB,EACrB,OAAS,GAAI,EAAG,EAAI,EAAkB,KACpC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAc,GAAQ,eAAe,EAAsB,EAAG,EAAsB,EAAG,EAAkB,GAC/G,EAAiB,GAAQ,WAAW,CAClC,EAAG,EAAsB,EAAI,EAC7B,EAAG,EAAsB,EAAI,GAC5B,CAAE,EAAG,EAAY,EAAG,EAAG,EAAY,IAExC,KAAM,GAAwB,GAAyB,EAAgB,EAAc,EAAQ,GACvF,EAAQ,EAAa,IAAI,EAAsB,EAAG,EAAsB,EAAG,GACjF,MAAO,CAAE,SAAU,EAAgB,KAAM,GAAU,UAAU,GAAmB,SAQlF,YAAoB,EAAM,EAAQ,EAAS,EAAc,EAAkB,GACzE,KAAM,GAAW,EAAO,MAAM,GACxB,EAAW,GAAmB,OAC9B,EAAoB,GAAI,OAAM,GAE9B,CAAE,KAAM,EAAU,MAAO,GAAc,EACvC,EAAY,GAAQ,eAAe,EAAU,EAAc,GACjE,EAAkB,EAAS,IAAM,CAC/B,MAAO,EACP,KAAM,GAAU,UAAU,EAAS,IACnC,SAAU,GAIZ,OAAS,GAAO,EAAW,EAAG,GAAQ,EAAG,EAAE,GACzC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAK/J,OAAS,GAAO,EAAG,EAAO,EAAU,EAAE,GACpC,KAAM,GAAmB,GAAmB,GACtC,EAAmB,GAAmB,GAC5C,AAAI,EAAkB,IAAqB,CAAC,EAAkB,IAC5D,GAAkB,GAAoB,GAAyB,EAAM,EAAkB,GAAmB,EAAkB,EAAQ,EAAS,EAAc,IAG/J,MAAO,GAET,GAAQ,WAAa,KCnFrB,oBAAM,IAAa,KACb,GAAa,KACb,GAAU,KAEhB,YAA6C,EAAO,EAAkB,CAAE,IAAG,KAAK,GAC9E,MAAO,GAAM,KAAK,CAAC,CAAE,gBACnB,KAAM,GAAwB,EAAU,GAAY,SACpD,MAAO,IAAQ,gBAAgB,EAAG,EAAG,EAAsB,EAAG,EAAsB,IAAM,IAO9F,YAA0B,EAAe,EAAkB,GACzD,KAAM,GAA8B,EAAkB,OAAO,CAAC,EAAQ,CAAE,WAAU,SAAS,IACzF,CAAK,GAAoC,EAAe,EAAkB,EAAU,IAClF,IAAU,GAEL,GACN,GACH,MAAO,GAA8B,EAAkB,OAKzD,KAAM,IAAsB,EAwD5B,YAA6B,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAmB,EAAiB,GAAK,EAAY,IAC3K,KAAM,GAAQ,GACR,EAAQ,GAAW,wBAAwB,EAAgB,GAAqB,GAChF,EAAmB,EAAY,EAGrC,KAAO,EAAM,OAAS,GAAqB,CAAC,EAAM,UAEhD,KAAM,GAAO,EAAM,UAIb,EAAkB,GAAQ,eAAe,EAAK,KAAM,EAAc,GACxE,GAAI,GAAoC,EAAO,EAAkB,EAAiB,EAAK,KAAK,IAAK,SAEjG,KAAM,GAAY,GAAW,WAAW,EAAM,EAAc,EAAe,EAAc,EAAwB,GAC3G,EAAQ,GAAiB,EAAO,EAAkB,GACxD,EAAM,KAAK,CAAE,YAAW,UAE1B,MAAO,GAET,GAAQ,oBAAsB,KCvG9B,oBAAM,IAAK,KACL,GAAM,KAEZ,YAAyC,EAAG,EAAG,GAC7C,MAAQ,GAAI,GAAiB,EAAI,EAGnC,YAA8B,EAAW,GACvC,MAAO,IAAI,qBAAqB,OAAO,CAAC,EAAQ,CAAC,EAAW,KACtD,IAAgC,EAAU,GAAW,MAAO,EAAU,GAAY,MAAO,IAG7F,EAAO,KAAK,CAAC,EAAU,GAAY,EAAU,KACtC,GACN,IAEL,GAAQ,qBAAuB,GAE/B,KAAM,CAAE,qBAAmB,sBAAsB,OACjD,YAAwB,GACtB,MAAO,GAAU,OAAO,CAAC,CAAE,OAAM,OAAM,OAAM,QAAQ,CAAE,SAAU,CAAE,IAAG,QAAW,EAC/E,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,GACrB,KAAM,KAAK,IAAI,EAAM,KACnB,CACF,KAAM,GACN,KAAM,GACN,KAAM,GACN,KAAM,KAGV,GAAQ,eAAiB,GACzB,YAA8B,GAC5B,KAAM,CAAE,OAAM,OAAM,OAAM,QAAS,GAAe,GAClD,MAAO,CAAC,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,GAAQ,CAAE,EAAG,EAAM,EAAG,IAE1F,GAAQ,qBAAuB,GAC/B,kBAAiC,GAC/B,MAAO,SAAQ,IAAI,EAAQ,IAAI,AAAC,GAAW,EAAO,WAEpD,GAAQ,kBAAoB,GAE5B,YAAmB,EAAM,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAC9D,MAAO,CACL,MAAO,EAAK,MACZ,UAAW,EAAK,UAAU,IAAI,CAAC,CAAE,QAAO,OAAM,cAAgB,EAC5D,QACA,OACA,SAAU,CACR,EAAG,EAAS,EAAI,EAAS,EACzB,EAAG,EAAS,EAAI,EAAS,OAKjC,GAAQ,UAAY,GAEpB,YAAoB,EAAO,EAAQ,EAAQ,EAAU,EAAG,EAAU,GAChE,MAAI,KAAW,GAAK,IAAW,GAAK,IAAY,GAAK,IAAY,EACxD,EAEF,EAAM,IAAI,AAAC,GAAS,GAAU,EAAM,EAAQ,EAAQ,EAAS,IAEtE,GAAQ,WAAa,GAErB,YAAkC,GAChC,MAAO,aAAiB,IAAG,OAAS,CAAC,EAAM,MAAM,GAAI,EAAM,MAAM,IAAM,CAAC,EAAM,OAAQ,EAAM,OAE9F,GAAQ,yBAA2B,GAEnC,YAAuB,GACrB,MAAO,aAAiB,IAAG,OAAS,EAAQ,GAAG,QAAQ,WAAW,GAEpE,GAAQ,cAAgB,GAExB,YAA8B,EAAO,EAAc,GACjD,MAAO,IAAG,KAAK,KACb,KAAM,GAAc,GAAc,GAClC,MAAO,GAAY,eAAe,CAAC,EAAc,MAGrD,GAAQ,qBAAuB,GAE/B,YAAwB,EAAO,CAAC,EAAS,IACvC,KAAM,CAAC,EAAQ,GAAS,GAAyB,GAC3C,EAAe,EAAU,EACzB,EAAS,EAAQ,EACvB,GAAI,CAAC,EAAM,EAAM,EAAM,GAAQ,CAAC,EAAG,EAAG,EAAG,GACzC,AAAI,EAAS,EAEX,GAAO,EACP,EAAO,EACP,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,IACjD,EAAO,KAAK,MAAM,GAAO,GAAe,EAAS,KAGjD,GAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,KAAK,MAAM,GAAQ,GAAM,EAAgB,EAAQ,IACxD,EAAO,EACP,EAAO,GAET,KAAM,GAAU,GAAG,KAAK,KACtB,GAAI,GAAc,GAAc,GAChC,SAAc,GAAG,MAAM,EAAa,CAAC,CAAC,EAAM,GAAO,CAAC,EAAM,GAAO,CAAC,EAAG,KAC9D,EAAY,eAAe,CAAC,EAAS,MAE9C,MAAO,CAAE,UAAS,QAAS,CAAE,IAAK,EAAM,KAAM,EAAM,MAAO,EAAM,OAAQ,IAE3E,GAAQ,eAAiB,GAEzB,YAA2B,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAuB,GAAuB,GAChG,KAAM,GAAU,GAAS,EAAQ,IAAM,EAAQ,QAAW,EACpD,EAAU,GAAQ,EAAQ,KAAO,EAAQ,OAAU,EACnD,EAAc,GAAW,EAAO,EAAQ,EAAQ,CAAC,EAAQ,IAAK,CAAC,EAAQ,MAC7E,MAAO,GAET,GAAQ,kBAAoB,KCrH5B,oBAAM,IAAK,KACL,GAAiB,KACjB,GAAiB,KACjB,GAAO,KAEb,SACE,YAAY,GACV,KAAK,UAAY,OAuBb,eAAc,EAAO,GACzB,KAAM,GAAe,EAAO,aAEtB,CAAC,EAAQ,GAAS,GAAK,yBAAyB,GAChD,CAAE,UAAS,WAAY,GAAK,eAAe,EAAO,CAAC,EAAO,gBAAiB,EAAO,kBAClF,CAAE,gBAAe,UAAS,kBAAiB,mBAAoB,KAAK,UAAU,QAAQ,GACtF,EAAmB,KAAM,IAAK,kBAAkB,CAAC,EAAe,EAAS,EAAiB,IAC1F,EAAe,EAAiB,GAChC,EAAgB,EAAiB,GACjC,EAAyB,EAAiB,GAC1C,EAAyB,EAAiB,GAC1C,EAAQ,KAAM,IAAe,oBAAoB,EAAc,EAAe,EAAwB,EAAwB,EAAc,EAAO,cAAe,EAAO,eAAgB,EAAO,WAChM,EAAc,GAAK,kBAAkB,EAAO,CAAC,EAAQ,GAAQ,CAAC,EAAO,gBAAiB,EAAO,iBAAkB,GACrH,SAAc,UACd,EAAQ,UACR,EAAgB,UAChB,EAAgB,UAChB,EAAQ,UACD,EAGT,UACE,KAAK,UAAU,WAGnB,GAAQ,QAAU,GAClB,kBAA6B,GAC3B,KAAM,GAAa,KAAM,IAAG,eAAe,EAAO,WAC5C,EAAY,GAAI,IAAe,UAAU,EAAY,EAAO,cAClE,MAAO,IAAI,IAAQ,GAYrB,kBAAoB,GAClB,MAAO,IAAc,GAEvB,GAAQ,KAAO,KC1Ef,oBAAM,IAAiB,KACjB,GAAe,KACf,GAAiB,KACjB,GAAY,KACZ,GAAO,KAEb,GAAQ,KAAO,GAAa,KAC5B,GAAQ,QAAU,GAAa,QAE/B,GAAQ,UAAY,GAAe,UACnC,GAAQ,oBAAsB,GAAe,oBAC7C,GAAQ,aAAe,GAAU,aACjC,GAAQ,QAAU,GAAU,QAC5B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,UAAY,GAAU,UAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,eAAiB,GAAK,eAC9B,GAAQ,qBAAuB,GAAK,qBACpC,GAAQ,kBAAoB,GAAK,kBACjC,GAAQ,UAAY,GAAK,YCnBzB,oBAAM,IAAK,KAEX,YAAoB,GAClB,MAAO,CACL,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAC1C,KAAK,IAAI,EAAI,SAAS,GAAK,EAAI,WAAW,KAG9C,GAAQ,WAAa,GAErB,YAAsB,GACpB,MAAO,CACL,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,EAC5D,EAAI,WAAW,GAAM,GAAI,SAAS,GAAK,EAAI,WAAW,IAAM,GAGhE,GAAQ,aAAe,GAEvB,YAAkC,EAAK,EAAO,GAC5C,KAAM,GAAI,EAAM,MAAM,GAChB,EAAI,EAAM,MAAM,GAChB,EAAQ,CAAC,CACb,EAAI,WAAW,GAAK,EAAG,EAAI,WAAW,GAAK,EAAG,EAAI,SAAS,GAAK,EAChE,EAAI,SAAS,GAAK,IAEpB,MAAO,IAAG,MAAM,cAAc,EAAO,EAAO,CAAC,GAAI,GAEnD,GAAQ,yBAA2B,GAEnC,YAA6B,EAAK,GAChC,KAAM,GAAa,CAAC,EAAI,WAAW,GAAK,EAAO,GAAI,EAAI,WAAW,GAAK,EAAO,IACxE,EAAW,CAAC,EAAI,SAAS,GAAK,EAAO,GAAI,EAAI,SAAS,GAAK,EAAO,IAClE,EAAgB,EAAI,cAAc,IAAI,AAAC,IAC3C,KAAM,GAAc,CAAC,EAAM,GAAK,EAAO,GAAI,EAAM,GAAK,EAAO,IAC7D,MAAO,KAET,MAAO,CAAE,aAAY,WAAU,iBAEjC,GAAQ,oBAAsB,GAE9B,YAAoB,EAAK,EAAS,KAChC,KAAM,GAAS,GAAa,GACtB,EAAO,GAAW,GAClB,EAAc,CAAC,EAAS,EAAK,GAAK,EAAG,EAAS,EAAK,GAAK,GACxD,EAAa,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IAClE,EAAW,CAAC,EAAO,GAAK,EAAY,GAAI,EAAO,GAAK,EAAY,IACtE,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,WAAa,GAErB,YAAqB,GACnB,KAAM,GAAU,GAAa,GACvB,EAAO,GAAW,GAClB,EAAU,KAAK,IAAI,GAAG,GACtB,EAAW,EAAU,EACrB,EAAa,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GAClD,EAAW,CAAC,EAAQ,GAAK,EAAU,EAAQ,GAAK,GACtD,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,YAAc,GAEtB,YAAkB,EAAK,GACrB,KAAM,GAAU,CACd,EAAI,SAAS,GAAK,EAAI,WAAW,GAAI,EAAI,SAAS,GAAK,EAAI,WAAW,IAElE,EAAc,CAAC,EAAQ,GAAK,EAAY,GAAI,EAAQ,GAAK,EAAY,IACrE,EAAa,CAAC,EAAI,WAAW,GAAK,EAAY,GAAI,EAAI,WAAW,GAAK,EAAY,IAClF,EAAW,CAAC,EAAI,SAAS,GAAK,EAAY,GAAI,EAAI,SAAS,GAAK,EAAY,IAClF,MAAO,CAAE,aAAY,WAAU,cAAe,EAAI,eAEpD,GAAQ,SAAW,KCtEnB,oBAAM,IAAK,KACL,GAAW,KAEjB,SACE,YAAY,EAAO,EAAS,GAC1B,KAAK,MAAQ,EACb,KAAK,MAAQ,EAAO,UACpB,KAAK,OAAS,EAAO,UACrB,KAAK,QAAU,EAAQ,IAAI,AAAC,GAAW,CAAC,EAAO,SAAU,EAAO,WAChE,KAAK,cAAgB,GAAG,SAAS,KAAK,SACtC,KAAK,gBAAkB,GAAG,SAAS,CAAC,EAAO,UAAW,EAAO,YAC7D,KAAK,sBAAwB,GAAG,SAAS,CAAC,EAAO,UAAY,EAAG,EAAO,UAAY,IAGrF,eAAe,GACb,MAAO,IAAG,KAAK,KACb,KAAM,GAAa,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IAC1C,EAAW,GAAG,MAAM,EAAO,CAAC,EAAG,GAAI,CAAC,GAAI,IACxC,EAAkB,GAAG,IAAI,GAAG,IAAI,EAAY,KAAK,iBAAkB,KAAK,eACxE,EAAe,GAAG,IAAI,EAAU,KAAK,uBACrC,EAAc,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACjE,EAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,GAAe,KAAK,iBACrE,MAAO,IAAG,SAAS,CAAC,EAAa,GAAY,KAIjD,mBAAmB,EAAkB,GACnC,MAAO,IAAG,KAAK,KACb,KAAM,GAAY,GAAG,IAAI,GAAG,IAAI,EAAiB,QAAQ,CAAC,GAAI,EAAG,IAAK,KAAK,iBAAkB,KAAK,QAAQ,IAC1G,MAAO,IAAG,IAAI,EAAW,KAAK,wBAI5B,kBAAiB,GACrB,KAAM,GAAkB,GAAG,KAAK,IAAM,GAAG,IAAI,GAAG,IAAI,EAAO,IAAM,IAC3D,EAAoB,KAAK,MAAM,QAAQ,GACvC,EAAa,EAAkB,UAE/B,EAAS,GAAG,KAAK,IAAM,GAAG,QAAQ,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,KAAK,WAEzE,EAAW,GAAG,MAAM,EAAY,CAAC,EAAG,GAAI,CAAC,GAAI,IAC7C,EAAQ,KAAK,eAAe,GAC5B,EAAuB,KAAM,IAAG,MAAM,uBAAuB,EAAO,EAAQ,KAAK,SAAU,KAAK,aAAc,KAAK,gBACnH,EAAiB,KAAM,GAAqB,QAC5C,EAAY,CAChB,EAAiB,EAAmB,EAAsB,EAC1D,EAAO,EAAU,GAEnB,GAAI,EAAe,SAAW,EAC5B,SAAU,QAAQ,AAAC,GAAW,EAAO,WAC9B,KAET,KAAM,GAAgB,GAAG,KAAK,KAC5B,KAAM,GAAgB,GACtB,SAAW,KAAK,IACd,KAAM,GAAW,EAAe,GAC1B,EAAc,GAAG,MAAM,EAAO,CAAC,EAAU,GAAI,CAAC,EAAG,KACjD,EAAmB,GAAG,MAAM,EAAY,CAAC,EAAU,GAAI,CAAC,EAAG,KAC3D,EAAgB,GAAG,KAAK,IAAM,KAAK,mBAAmB,EAAkB,GAAU,QAAQ,CAAC,GAAI,KACrG,EAAc,KAAK,CAAE,MAAO,EAAa,kBAE3C,MAAO,KAET,MAAO,QASH,oBAAmB,EAAO,GAC9B,KAAM,GAAc,EAAM,MAAM,GAC1B,EAAa,EAAM,MAAM,GAC/B,KAAK,aAAe,EAAO,aAC3B,KAAK,eAAiB,EAAO,eAC7B,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IAAM,EAAM,eAAe,CAAC,KAAK,MAAO,KAAK,SAAS,IAAI,MAC1E,EAAc,KAAM,MAAK,iBAAiB,GAEhD,GADA,EAAM,UACF,CAAC,GAAgB,EAAY,SAAW,EAAI,MAAO,MACvD,KAAM,GAAQ,GACd,SAAW,KAAK,IACd,KAAM,GAAa,EAAY,GACzB,EAAgB,KAAM,GAAW,MAAM,QACvC,EAAa,EAAc,GAAG,MAAM,EAAG,GACvC,EAAW,EAAc,GAAG,MAAM,EAAG,GACrC,EAAgB,KAAM,GAAW,cAAc,QACrD,EAAW,MAAM,UACjB,EAAW,cAAc,UACzB,EAAM,KAAK,GAAS,oBAAoB,CAAE,aAAY,WAAU,iBAAiB,CAAC,EAAa,KAAK,MAAO,EAAc,KAAK,UAEhI,MAAO,IAGX,GAAQ,aAAe,KChGvB,kBAAQ,iBAAmB,CACzB,MAAO,CAAC,EAAG,EAAG,EAAG,GACjB,YAAa,CAAC,EAAG,EAAG,EAAG,GACvB,aAAc,CAAC,EAAG,GAAI,GAAI,IAC1B,WAAY,CAAC,GAAI,GAAI,GAAI,IACzB,MAAO,CAAC,GAAI,GAAI,GAAI,IACpB,SAAU,CAAC,MCNb,2BAA0B,GACxB,MAAO,GAAQ,EAAI,KAAK,GAAK,KAAK,MAAO,GAAQ,KAAK,IAAO,GAAI,KAAK,KAExE,GAAQ,iBAAmB,GAE3B,YAAyB,EAAQ,GAC/B,KAAM,GAAU,KAAK,GAAK,EAAI,KAAK,MAAM,CAAE,GAAO,GAAK,EAAO,IAAK,EAAO,GAAK,EAAO,IACtF,MAAO,IAAiB,GAE1B,GAAQ,gBAAkB,GAE1B,KAAM,IAAyB,CAAC,EAAG,IAAO,CAAC,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,GAAI,CAAC,EAAG,EAAG,IACxE,YAAa,EAAI,GACf,GAAI,GAAU,EACd,OAAS,GAAI,EAAG,EAAI,EAAG,OAAQ,IAC7B,GAAW,EAAG,GAAK,EAAG,GAExB,MAAO,GAET,GAAQ,IAAM,GAEd,YAA4B,EAAK,GAC/B,KAAM,GAAS,GACf,OAAS,GAAI,EAAG,EAAI,EAAI,OAAQ,IAC9B,EAAO,KAAK,EAAI,GAAG,IAErB,MAAO,GAET,GAAQ,mBAAqB,GAE7B,YAAmC,EAAM,GACvC,KAAM,GAAU,GACV,EAAO,EAAK,OAClB,OAAS,GAAM,EAAG,EAAM,EAAM,KAC5B,EAAQ,KAAK,IACb,OAAS,GAAM,EAAG,EAAM,EAAM,IAC5B,EAAQ,GAAK,KAAK,GAAI,EAAK,GAAM,GAAmB,EAAM,KAG9D,MAAO,GAET,YAA6B,EAAU,GACrC,KAAM,GAAO,KAAK,IAAI,GAChB,EAAO,KAAK,IAAI,GAChB,EAAiB,CAAC,CAAC,EAAM,CAAC,EAAM,GAAI,CAAC,EAAM,EAAM,GAAI,CAAC,EAAG,EAAG,IAC5D,EAAoB,GAAuB,EAAO,GAAI,EAAO,IAC7D,EAA2B,GAA0B,EAAmB,GACxE,EAA4B,GAAuB,CAAC,EAAO,GAAI,CAAC,EAAO,IAC7E,MAAO,IAA0B,EAA0B,GAE7D,GAAQ,oBAAsB,GAE9B,YAA+B,GAC7B,KAAM,GAAoB,CAAC,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAAK,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,KAC5E,EAAuB,CAAC,EAAO,GAAG,GAAI,EAAO,GAAG,IAChD,EAAsB,CAC1B,CAAC,GAAI,EAAkB,GAAI,GAC3B,CAAC,GAAI,EAAkB,GAAI,IAE7B,MAAO,CACL,EAAkB,GAAG,OAAO,EAAoB,IAChD,EAAkB,GAAG,OAAO,EAAoB,IAChD,CAAC,EAAG,EAAG,IAGX,GAAQ,sBAAwB,GAEhC,YAAqB,EAAuB,GAC1C,MAAO,CACL,GAAI,EAAuB,EAAe,IAC1C,GAAI,EAAuB,EAAe,KAG9C,GAAQ,YAAc,KCzEtB,oBAAM,IAAK,KACL,GAAW,KACX,GAAO,KAEP,GAA0C,GAC1C,GAAwB,CAAC,EAAG,KAC5B,GAAwB,CAAC,EAAG,KAC5B,GAA0B,KAC1B,GAAoB,CAAC,EAAG,EAAG,EAAG,GAAI,GAAI,EAAG,GACzC,GAAoC,EACpC,GAA6C,EAGnD,SACE,YAAY,EAAqB,EAAc,GAC7C,KAAK,kBAAoB,GACzB,KAAK,wBAA0B,EAC/B,KAAK,oBAAsB,EAC3B,KAAK,aAAe,EACpB,KAAK,UAAY,EAAO,UACxB,KAAK,WAAa,EAAO,UACzB,KAAK,cAAgB,EAAO,cAI9B,uBAAuB,EAAe,GACpC,KAAM,GAAuB,EAAc,IAAI,AAAC,IAC9C,KAAM,GAAwB,CAAC,GAAG,EAAO,GACzC,MAAO,IAAK,YAAY,EAAuB,KAE3C,EAAgB,KAAK,8BAA8B,GAGzD,MAAO,IAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAe,KAAyB,KAAK,eAIjH,uBAAuB,GAIrB,KAAM,GAAc,KAAK,8BAA8B,GACjD,EAAgB,GAAS,WAAW,GAAS,YAAY,GAAS,SAAS,EAAa,KAAyB,IACjH,EAAgB,GACtB,OAAS,GAAI,EAAG,EAAI,GAAkB,OAAQ,IAC5C,EAAc,KAAK,EAAU,GAAkB,IAAI,MAAM,EAAG,IAE9D,SAAc,cAAgB,EACvB,EAKT,mBAAmB,EAAW,EAAK,EAAO,GACxC,KAAM,GAAU,GAAS,WAAW,GAC9B,EAAc,CAAC,EAAQ,GAAK,KAAK,UAAW,EAAQ,GAAK,KAAK,YAC9D,EAAe,EAAU,IAAI,AAAC,GAAU,CAC5C,EAAY,GAAM,GAAM,GAAK,KAAK,UAAY,GAC9C,EAAY,GAAM,GAAM,GAAK,KAAK,WAAa,GAAI,EAAM,KAErD,EAAuB,GAAK,oBAAoB,EAAO,CAAC,EAAG,IAC3D,EAAgB,EAAa,IAAI,AAAC,IACtC,KAAM,GAAU,GAAK,YAAY,EAAO,GACxC,MAAO,CAAC,GAAG,EAAS,EAAM,MAEtB,EAAwB,GAAK,sBAAsB,GACnD,EAAY,CAAC,GAAG,GAAS,aAAa,GAAM,GAC5C,EAAoB,CACxB,GAAK,IAAI,EAAW,EAAsB,IAC1C,GAAK,IAAI,EAAW,EAAsB,KAE5C,MAAO,GAAc,IAAI,AAAC,GAAU,CAClC,EAAM,GAAK,EAAkB,GAAI,EAAM,GAAK,EAAkB,GAC9D,EAAM,UAIJ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAc,KAAK,gCACzB,GAAI,IAAgB,IAClB,KAAM,GAAyB,KAAM,MAAK,oBAAoB,mBAAmB,EAAO,GACxF,KAAK,kBAAoB,GACzB,SAAW,KAAK,GACd,KAAK,wBAAwB,EAAuB,GAAI,GAAyB,GAEnF,KAAK,wBAA0B,MAE/B,MAAK,0BAGP,KAAM,GAAQ,GACd,GAAI,CAAC,KAAK,kBAAmB,MAAO,GACpC,SAAW,KAAK,MAAK,mBACnB,KAAM,GAAa,KAAK,kBAAkB,GAAG,GAC7C,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAQ,GAAK,gBAAgB,EAAW,cAAc,IAAoC,EAAW,cAAc,KACnH,EAAa,GAAS,aAAa,GACnC,EAAuB,CAAC,EAAW,GAAK,EAAM,MAAM,GAAI,EAAW,GAAK,EAAM,MAAM,IACpF,EAAe,GAAG,MAAM,iBAAiB,EAAO,EAAO,EAAG,GAC1D,EAAiB,GAAK,oBAAoB,CAAC,EAAO,GAClD,EAAM,EAAc,KAAK,uBAAuB,EAAW,cAAe,GAAkB,EAC5F,EAAe,GAAS,yBAAyB,EAAK,EAAc,CAAC,KAAK,UAAW,KAAK,aAC1F,EAAY,EAAa,IAAI,KACnC,EAAa,UACb,EAAa,UACb,KAAM,GAAa,KAAK,aAAa,QAAQ,GACvC,CAAC,EAAM,GAAa,EAC1B,EAAU,UACV,KAAM,GAAY,EAAK,WAAW,GAElC,GADA,EAAK,UACD,EAAY,EAAO,cACrB,SAAU,UACV,KAAK,kBAAkB,GAAK,GACrB,EAET,KAAM,GAAoB,GAAG,QAAQ,EAAW,CAAC,GAAI,IAC/C,EAAY,KAAM,GAAkB,QAC1C,EAAU,UACV,EAAkB,UAClB,KAAM,GAAS,KAAK,mBAAmB,EAAW,EAAK,EAAO,GACxD,EAAkB,KAAK,uBAAuB,GACpD,KAAK,wBAAwB,EAAiB,GAA2B,GACzE,KAAM,GAAS,CACb,UAAW,EACX,WAAY,EACZ,IAAK,CACH,QAAS,EAAgB,WACzB,YAAa,EAAgB,WAGjC,EAAM,KAAK,GAEb,MAAO,GAIT,8BAA8B,GAC5B,KAAM,GAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAK,EAAU,IAAI,AAAC,GAAM,EAAE,IAC5B,EAAa,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC3C,EAAW,CAAC,KAAK,IAAI,GAAG,GAAK,KAAK,IAAI,GAAG,IAC/C,MAAO,CAAE,aAAY,YAKvB,wBAAwB,EAAK,EAAa,GACxC,GAAI,EACF,KAAK,kBAAkB,GAAS,CAAC,QAEjC,KAAM,GAAc,KAAK,kBAAkB,GAAO,GAClD,GAAI,GAAM,EACV,GAAI,GAAe,MAAQ,EAAY,YAAc,MACnD,KAAM,CAAC,EAAW,GAAa,EAAI,WAC7B,CAAC,EAAS,GAAW,EAAI,SACzB,CAAC,EAAmB,GAAqB,EAAY,WACrD,CAAC,EAAiB,GAAmB,EAAY,SACjD,EAAY,KAAK,IAAI,EAAW,GAChC,EAAY,KAAK,IAAI,EAAW,GAChC,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAU,KAAK,IAAI,EAAS,GAC5B,EAAgB,GAAU,GAAc,GAAU,GAClD,EAAW,GAAU,GAAc,GAAU,GAC7C,EAAmB,GAAkB,GAAsB,GAAkB,GACnF,EAAM,EAAgB,GAAU,EAAkB,GAEpD,KAAK,kBAAkB,GAAO,GAAK,EAAM,GAA0C,EAAc,GAIrG,gCACE,MAAO,CAAC,KAAK,mBAAsB,KAAK,kBAAkB,SAAW,GAAO,KAAK,yBAA2B,KAAK,qBAGrH,GAAQ,aAAe,KCjLvB,oBAAM,IAAK,KACL,GAAO,KACP,GAAY,KACZ,GAAO,KAEb,SACE,YAAY,GACV,KAAK,SAAW,OAGZ,eAAc,EAAO,GACzB,KAAK,oBAAsB,EAAO,WAClC,KAAK,oBAAsB,EAAO,cAClC,KAAK,SAAW,EAAO,SACvB,KAAM,GAAQ,GAAG,KAAK,IACpB,CAAM,YAAiB,IAAG,QACxB,GAAQ,GAAG,QAAQ,WAAW,IAEzB,EAAM,UAAU,WAAW,KAE9B,EAAc,KAAM,MAAK,SAAS,cAAc,EAAO,GAC7D,EAAM,UACN,KAAM,GAAQ,GACd,GAAI,CAAC,EAAa,MAAO,GACzB,SAAW,KAAc,IACvB,GAAI,CAAC,EAAY,MAAO,GACxB,KAAM,GAAc,GACpB,SAAW,KAAO,QAAO,KAAK,GAAU,kBACtC,EAAY,GAAO,GAAU,iBAAiB,GAAK,IAAI,AAAC,GAAU,EAAW,UAAU,IAEzF,EAAM,KAAK,CACT,WAAY,EAAW,YAAc,EACrC,IAAK,EAAW,IAAM,CAAC,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,GAAI,EAAW,IAAI,YAAY,GAAK,EAAW,IAAI,QAAQ,IAAM,EACrM,UAAW,EAAW,UACtB,gBAGJ,MAAO,IAGX,GAAQ,SAAW,GAEnB,kBAA2B,GACzB,GAAI,GAAG,MAAM,SAAS,SAEpB,KAAM,GAAK,cACL,EAAO,KAAM,GAAG,aAAa,EAAI,QAAQ,UAAW,KAC1D,MAAO,MAAK,MAAM,GAEpB,MAAO,IAAG,KAAK,MAAM,GAAK,KAAK,AAAC,GAAM,EAAE,QAG1C,kBAAoB,GAClB,KAAM,CAAC,EAAS,EAAmB,GAAiB,KAAM,SAAQ,IAAI,CACpE,GAAY,EAAO,SAAS,SAC5B,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,eAC7F,GAAG,eAAe,EAAO,SAAS,UAAW,CAAE,UAAW,EAAO,SAAS,UAAU,SAAS,iBAEzF,EAAW,GAAI,IAAK,aAAa,EAAmB,EAAS,GAC7D,EAAW,GAAI,IAAK,aAAa,EAAU,EAAe,GAC1D,EAAW,GAAI,IAAS,GAC9B,MAAO,GAET,GAAQ,KAAO,KC/Df,uCAGA,GAAO,IAAQ,CACb,QAAS,QACT,QAAS,GACT,KAAM,CACJ,QAAS,GAGT,SAAU,CACR,UAAW,sCAEX,UAAW,IACX,SAAU,GACV,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,IAElB,KAAM,CACJ,QAAS,GACT,UAAW,gCACX,UAAW,KAEb,KAAM,CACJ,QAAS,GACT,UAAW,4BACX,cAAe,IACf,UAAW,IAEb,IAAK,CACH,QAAS,GACT,UAAW,uCAEX,UAAW,GACX,WAAY,IAEd,OAAQ,CACN,QAAS,GACT,cAAe,GACf,UAAW,2CAEb,QAAS,CACP,QAAS,GACT,UAAW,GACX,cAAe,GACf,WAAY,GACZ,aAAc,GACd,UAAW,iCAGf,KAAM,CACJ,QAAS,GACT,UAAW,+BACX,gBAAiB,IACjB,aAAc,GACd,cAAe,GACf,eAAgB,GAChB,UAAW,IAEb,KAAM,CACJ,QAAS,GACT,UAAW,IACX,WAAY,GAGZ,cAAe,GACf,aAAc,GACd,eAAgB,GAChB,cAAe,KACf,SAAU,GACV,SAAU,CACR,QAAS,oCACT,UAAW,mCAEb,SAAU,CACR,UAAW,6vEC/EjB,oBAAM,IAAK,KACL,GAAW,KACX,GAAS,KACT,GAAU,KACV,GAAU,KACV,GAAW,KACX,GAAW,KAAuB,QAClC,GAAM,KAEZ,GAAI,IAGJ,KAAM,IAAS,CACb,SAAU,KACV,QAAS,KACT,SAAU,KACV,KAAM,KACN,IAAK,KACL,OAAQ,KACR,QAAS,MAEL,GAAM,IACN,MAAO,cAAgB,YAAoB,YAAY,MACpD,SAAS,OAAO,QAAQ,OAAO,UAAY,IAAO,KAGrD,GAAM,IAAI,KAEd,AAAI,GAAO,SAAS,QAAQ,IAAI,GAAG,IAIrC,eAAsB,GACpB,KAAM,GAAW,AAAC,GAAQ,GAAO,MAAO,IAAQ,SAChD,MAAO,GAAQ,OAAO,CAAC,EAAM,IAC3B,QAAO,KAAK,GAAO,IAAI,QAAQ,AAAC,IAC9B,KAAM,GAAO,EAAK,GACZ,EAAO,EAAI,GACjB,AAAI,MAAM,QAAQ,IAAS,MAAM,QAAQ,GACvC,EAAK,GAAO,EAAK,OAAO,GAAG,GACtB,AAAI,EAAS,IAAS,EAAS,GACpC,EAAK,GAAO,GAAU,EAAM,GAE5B,EAAK,GAAO,IAGT,GACN,IAGL,YAAgB,GACd,GAAI,CAAC,EAAO,MAAO,uBACnB,KAAM,GAAQ,EAAM,cAAgB,EAAM,YAAc,EAAM,OAAU,EAAM,OAAU,EAAM,MAAM,GAAK,EACzG,GAAI,CAAC,GAAU,IAAU,EAAI,MAAO,iBACpC,GAAI,EAAM,YAAe,EAAM,YAAc,EAAI,MAAO,qBACxD,IACE,GAAG,mBAEH,MAAO,qBAET,MAAO,MAGT,kBAAsB,EAAO,GAC3B,GAAS,GAAU,GAAU,GAG7B,KAAM,GAAQ,GAAO,GACrB,MAAI,GACF,IAAI,EAAO,GACJ,CAAE,UAIJ,GAAI,SAAQ,KAAO,KAExB,KAAM,GAAe,OAAO,OAAO,IAAQ,OAAO,AAAC,GAAM,GAAG,OAC5D,AAAI,IAAiB,GAAG,GAAI,0BAG5B,AAAI,GAAG,eAAiB,GAAO,SAC7B,IAAI,iCAAkC,GAAO,SAC7C,KAAM,IAAG,WAAW,GAAO,SAC3B,KAAM,IAAG,SAUX,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,GAAO,OACvF,AAAI,GAAO,KAAK,SAAW,CAAC,GAAO,UAAU,IAAO,SAAW,KAAM,IAAS,KAAK,GAAO,OAC1F,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,IAAI,SAAW,CAAC,GAAO,KAAK,IAAO,IAAM,KAAM,IAAO,QAAQ,KACrG,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,OAAO,SAAW,CAAC,GAAO,QAAQ,IAAO,OAAS,KAAM,IAAO,WAAW,KACjH,AAAI,GAAO,KAAK,SAAW,GAAO,KAAK,QAAQ,SAAW,CAAC,GAAO,SAAS,IAAO,QAAU,KAAM,IAAQ,KAAK,KAE/G,KAAM,GAAO,GACb,GAAI,GAGJ,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,QAAQ,cAAc,EAAO,GAAO,MAAQ,GAC/F,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAU,GAAO,KAAK,QAAU,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAAQ,GAChG,GAAG,SAAS,WACZ,EAAK,KAAO,KAAK,MAAM,KAAQ,GAG/B,KAAM,GAAU,GAChB,GAAI,GAAO,KAAK,SACd,EAAY,KACZ,GAAG,SAAS,aACZ,KAAM,GAAQ,KAAM,IAAO,SAAS,cAAc,EAAO,GAAO,MAChE,EAAK,KAAO,KAAK,MAAM,KAAQ,GAC/B,SAAW,KAAQ,IAEjB,GAAI,CAAC,EAAK,OAAS,EAAK,MAAM,oBAC5B,GAAI,2BAA4B,EAAK,OACrC,SAGF,EAAY,KACZ,KAAM,GAAW,GAAO,KAAK,IAAI,SAAW,GAAO,KAAK,OAAO,QAAW,KAAM,IAAO,QAAQ,EAAK,MAAO,IAAU,GACrH,EAAK,UAAY,KAAK,MAAM,KAAQ,GAEpC,EAAY,KACZ,KAAM,GAAc,GAAO,KAAK,QAAQ,QAAU,KAAM,IAAQ,QAAQ,EAAK,MAAO,IAAU,GAC9F,EAAK,QAAU,KAAK,MAAM,KAAQ,GAClC,EAAK,MAAM,UAGX,KAAM,GAAQ,EAAK,YAAY,aAAe,EAAK,YAAY,aAC3D,KAAK,IAAI,EAAK,YAAY,YAAY,GAAG,GAAK,EAAK,YAAY,YAAY,GAAG,GAAI,EAAK,YAAY,aAAa,GAAG,GAAK,EAAK,YAAY,aAAa,GAAG,IACzJ,EACJ,EAAQ,KAAK,CACX,WAAY,EAAK,WACjB,IAAK,EAAK,IACV,KAAM,EAAK,KACX,YAAa,EAAK,YAClB,IAAK,EAAQ,IACb,OAAQ,EAAQ,OAChB,aAAc,EAAQ,WACtB,QAAS,EACT,KAAO,IAAS,EAAK,KAAK,MAAM,IAAM,KAAmC,GAAQ,IAAM,IAG3F,GAAG,SAAS,WAOd,EAAK,MAAQ,OAAO,OAAO,GAAM,OAAO,CAAC,EAAG,IAAM,EAAI,GACtD,EAAQ,CAAE,KAAM,EAAS,KAAM,EAAS,KAAM,EAAS,YAAa,MAIxE,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,OAAS,GACjB,GAAQ,SAAW,GACnB,GAAQ,OAAS,GACjB,GAAQ,QAAU,GAClB,GAAQ,SAAW,GACnB,GAAQ,GAAK,GACb,GAAQ,QAAU,GAAI",
   "names": []
 }
diff --git a/src/config.js b/src/config.js
index 1b4b00db..39448db0 100644
--- a/src/config.js
+++ b/src/config.js
@@ -1,65 +1,77 @@
+/* eslint-disable indent */
+/* eslint-disable no-multi-spaces */
+
 export default {
-  backend: 'webgl',
-  console: true,
+  backend: 'webgl',          // select tfjs backend to use
+  console: true,             // enable debugging output to console
   face: {
-    enabled: true, // refers to detector, but since all other face modules rely on detector, it should be a global
+    enabled: true,           // controls if specified modul is enabled
+                             // face.enabled is required for all face models: detector, mesh, iris, age, gender, emotion
+                             // (note: module is not loaded until it is required)
     detector: {
-      modelPath: '../models/blazeface/back/model.json', // can be blazeface-front or blazeface-back
-      inputSize: 256, // fixed value: 128 for front and tfhub and 256 for back
-      maxFaces: 10, // maximum number of faces detected in the input, should be set to the minimum number for performance
-      skipFrames: 10, // how many frames to go without running the bounding box detector
-      minConfidence: 0.5, // threshold for discarding a prediction
-      iouThreshold: 0.3, // threshold for deciding whether boxes overlap too much in non-maximum suppression
-      scoreThreshold: 0.7, // threshold for deciding when to remove boxes based on score in non-maximum suppression
+      modelPath: '../models/blazeface/back/model.json', // can be 'tfhub', 'front' or 'back'.
+                                                        // 'front' is optimized for large faces such as front-facing camera and 'back' is optimized for distanct faces.
+      inputSize: 256,        // fixed value: 128 for front and 'tfhub' and 'front' and 256 for 'back'
+      maxFaces: 10,          // maximum number of faces detected in the input, should be set to the minimum number for performance
+      skipFrames: 10,        // how many frames to go without re-running the face bounding box detector
+                             // if model is running st 25 FPS, we can re-use existing bounding box for updated face mesh analysis
+                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)
+      minConfidence: 0.5,    // threshold for discarding a prediction
+      iouThreshold: 0.3,     // threshold for deciding whether boxes overlap too much in non-maximum suppression
+      scoreThreshold: 0.7,   // threshold for deciding when to remove boxes based on score in non-maximum suppression
     },
     mesh: {
       enabled: true,
       modelPath: '../models/facemesh/model.json',
-      inputSize: 192, // fixed value
+      inputSize: 192,        // fixed value
     },
     iris: {
       enabled: true,
       modelPath: '../models/iris/model.json',
-      enlargeFactor: 2.3, // empiric tuning
-      inputSize: 64, // fixed value
+      enlargeFactor: 2.3,    // empiric tuning
+      inputSize: 64,         // fixed value
     },
     age: {
       enabled: true,
-      modelPath: '../models/ssrnet-age/imdb/model.json',
-      inputSize: 64, // fixed value
-      skipFrames: 10,
+      modelPath: '../models/ssrnet-age/imdb/model.json', // can be 'imdb' or 'wiki'
+                                                         // which determines training set for model
+      inputSize: 64,         // fixed value
+      skipFrames: 10,        // how many frames to go without re-running the detector
     },
     gender: {
       enabled: true,
+      minConfidence: 0.8,    // threshold for discarding a prediction
       modelPath: '../models/ssrnet-gender/imdb/model.json',
     },
     emotion: {
       enabled: true,
-      inputSize: 64, // fixed value
-      minConfidence: 0.5,
-      skipFrames: 10,
-      useGrayscale: true,
+      inputSize: 64,         // fixed value
+      minConfidence: 0.5,    // threshold for discarding a prediction
+      skipFrames: 10,        // how many frames to go without re-running the detector
+      useGrayscale: true,    // convert image to grayscale before prediction or use highest channel
       modelPath: '../models/emotion/model.json',
     },
   },
   body: {
     enabled: true,
     modelPath: '../models/posenet/model.json',
-    inputResolution: 257, // fixed value
-    outputStride: 16, // fixed value
-    maxDetections: 5,
-    scoreThreshold: 0.7,
-    nmsRadius: 20,
+    inputResolution: 257,    // fixed value
+    outputStride: 16,        // fixed value
+    maxDetections: 10,       // maximum number of people detected in the input, should be set to the minimum number for performance
+    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression
+    nmsRadius: 20,           // radius for deciding points are too close in non-maximum suppression
   },
   hand: {
     enabled: true,
-    inputSize: 256, // fixed value
-    skipFrames: 10,
-    minConfidence: 0.5,
-    iouThreshold: 0.3,
-    scoreThreshold: 0.7,
-    enlargeFactor: 1.65, // empiric tuning
-    maxHands: 2,
+    inputSize: 256,          // fixed value
+    skipFrames: 10,          // how many frames to go without re-running the hand bounding box detector
+                             // if model is running st 25 FPS, we can re-use existing bounding box for updated hand skeleton analysis
+                             // as face probably hasn't moved much in short time (10 * 1/25 = 0.25 sec)
+    minConfidence: 0.5,      // threshold for discarding a prediction
+    iouThreshold: 0.3,       // threshold for deciding whether boxes overlap too much in non-maximum suppression
+    scoreThreshold: 0.7,     // threshold for deciding when to remove boxes based on score in non-maximum suppression
+    enlargeFactor: 1.65,     // empiric tuning as skeleton prediction prefers hand box with some whitespace
+    maxHands: 10,            // maximum number of hands detected in the input, should be set to the minimum number for performance
     detector: {
       anchors: '../models/handdetect/anchors.json',
       modelPath: '../models/handdetect/model.json',
diff --git a/src/index.js b/src/index.js
index 1ea656b2..2456104d 100644
--- a/src/index.js
+++ b/src/index.js
@@ -84,11 +84,11 @@ async function detect(input, userConfig) {
       await tf.ready();
     }
     // explictly enable depthwiseconv since it's diasabled by default due to issues with large shaders
-    let savedWebglPackDepthwiseConvFlag;
-    if (tf.getBackend() === 'webgl') {
-      savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');
-      tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);
-    }
+    // let savedWebglPackDepthwiseConvFlag;
+    // if (tf.getBackend() === 'webgl') {
+    //   savedWebglPackDepthwiseConvFlag = tf.env().get('WEBGL_PACK_DEPTHWISECONV');
+    //  tf.env().set('WEBGL_PACK_DEPTHWISECONV', true);
+    // }
 
     // load models if enabled
     if (config.face.enabled && !models.facemesh) models.facemesh = await facemesh.load(config.face);
@@ -149,6 +149,7 @@ async function detect(input, userConfig) {
           annotations: face.annotations,
           age: ssrData.age,
           gender: ssrData.gender,
+          agConfidence: ssrData.confidence,
           emotion: emotionData,
           iris: (iris !== 0) ? Math.trunc(100 * 11.7 /* human iris size in mm */ / iris) / 100 : 0,
         });
@@ -157,7 +158,7 @@ async function detect(input, userConfig) {
     }
 
     // set depthwiseconv to original value
-    tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);
+    // tf.env().set('WEBGL_PACK_DEPTHWISECONV', savedWebglPackDepthwiseConvFlag);
 
     // combine and return results
     perf.total = Object.values(perf).reduce((a, b) => a + b);
diff --git a/src/ssrnet/ssrnet.js b/src/ssrnet/ssrnet.js
index 36292e75..f6d9cebe 100644
--- a/src/ssrnet/ssrnet.js
+++ b/src/ssrnet/ssrnet.js
@@ -22,11 +22,12 @@ async function loadGender(config) {
 }
 
 async function predict(image, config) {
-  frame += 1;
-  if (frame >= config.face.age.skipFrames) {
+  if (frame > config.face.age.skipFrames) {
     frame = 0;
-    return last;
+  } else {
+    frame += 1;
   }
+  if (frame === 0) return last;
   let enhance;
   if (image instanceof tf.Tensor) {
     const resize = tf.image.resizeBilinear(image, [config.face.age.inputSize, config.face.age.inputSize], false);
@@ -45,7 +46,11 @@ async function predict(image, config) {
   if (config.face.gender.enabled) {
     const genderT = await models.gender.predict(enhance);
     const data = await genderT.data();
-    obj.gender = Math.trunc(100 * data[0]) < 50 ? 'female' : 'male';
+    const confidence = Math.trunc(Math.abs(1.9 * 100 * (data[0] - 0.5))) / 100;
+    if (confidence > config.face.gender.minConfidence) {
+      obj.gender = data[0] <= 0.5 ? 'female' : 'male';
+      obj.confidence = confidence;
+    }
     tf.dispose(genderT);
   }
   tf.dispose(enhance);